text
stringlengths 4
2.78M
| meta
dict |
|---|---|
---
abstract: 'We present a theory for the large suppression of the superfluid-density, $\rho_s$, in BaFe$_2$(As$_{1-x}$P$_x$)$_2$ in the vicinity of a putative spin-density wave quantum critical point at a P-doping, $x=x_{c}$. We argue that the transition becomes weakly first-order in the vicinity of $x_{c}$, and disorder induces puddles of superconducting and antiferromagnetic regions at short length-scales; thus the system becomes an electronic micro-emulsion. We propose that frustrated Josephson couplings between the superconducting grains suppress $\rho_s$. In addition, the presence of ‘normal’ quasiparticles at the interface of the frustrated Josephson junctions will give rise to a highly non-trivial feature in the low-frequency response in a narrow vicinity around $x_c$. We propose a number of experiments to test our theory.'
author:
- Debanjan Chowdhury
- 'J. Orenstein'
- Subir Sachdev
- 'T. Senthil'
title: 'Phase transition beneath the superconducting dome in BaFe$_2$(As$_{1-x}$P$_x$)$_2$'
---
[*Introduction.-*]{} An important focus of the study of high temperature superconductivity (SC) has been on the role of antiferromagnetism (AFM) and its relation to SC [@Taillefer10]. There is clear evidence across many different families of compounds that SC appears in close proximity to an AFM phase [@DS12]; these families include the iron-pnictides, the electron-doped cuprates and the heavy-fermion superconductors. Moreover, the optimal transition temperature ($T_c$) of the SC is often situated where the normal state AFM quantum critical point (QCP) would have been located, in the absence of superconductivity. The experimental detection of the QCP is often challenging in the normal state, and more so in the superconducting state.
Recently, a number of measurements were reported in a member of the pnictide family, BaFe$_2$(As$_{1-x}$P$_x$)$_2$, as a function of the isovalent P-doping, $x$. The experiments show a phase transition involving onset of spin-density wave (SDW) order in the normal state above $T_c$, which extrapolates to a $T=0$ SDW QCP (see [@Matsudareview] and references therein). These experiments include: ([*i*]{}) a sharp enhancement in the effective mass, $m^*$, upon approaching a critical doping from the overdoped side, as obtained from de Haas-van Alphen oscillations [@Matsuda10a] and from the jump in the specific-heat at $T_c$ [@walmsley13] , and, ([*ii*]{}) a vanishing Curie-Weiss temperature ($\theta_{CW}$), extracted from the $1/T_1T$ measurements using NMR.
As we will review below, a number of puzzling results have appeared from experiments investigating whether the SDW QCP actually survives “under the SC dome.” Here we propose a resolution of these puzzles by postulating a weakly first-order transition for the onset of SDW order in the presence of SC order (see Fig. \[ph\]a). Our results are independent of the specific microsopic mechanism responsible for rendering the transition weakly first-order [@RFAVC]. It is well known that ‘random bond’ disorder has a strong effect on symmetry-breaking first-order transitions [@dis], and ultimately replaces them with a disorder-induced second order transition in two dimensional systems. Our main claim is that the inhomogeneities associated with these highly relevant effects of disorder can resolve the experimental puzzles.
The possiblity of a QCP within the SC state was investigated by measurements [@Matsuda12] of the zero temperature London penetration depth, $\lambda_L(0)\propto1/\sqrt\rho_s$ ($\rho_s\equiv$ superfluid-density), as a function of $x$. A sharp peak in $\lambda_L^2(0)$ was observed at $x=x_c$ and interpreted as evidence for a QCP [@AVC13]. However, this interpretation is at odds with general theoretical considerations [@DCSS13] concerning a QCP associated with the onset of SDW order in the presence of a superconductor with gapped quasiparticle excitations [@comment1; @Nodes]. These considerations suggest that such systems will display a [*monotonic*]{} variation in $\lambda_L^2(0)$ across the QCP, rather than a sharp peak (see dashed-blue/solid-red curves in Fig. \[ph\]b) [@DCSS13].
As a first step toward resolving this discrepancy, it is useful to place measurements of $\rho_s$ in the context of what is known about the normal state conductivity of the BaFe$_2$(As$_{1-x}$P$_x$)$_2$ system, as these quantities are intimately related through a sum rule. The low temperature superfluid density of a spatially homogeneous superconductor can be estimated from the “missing area" relation, \_s\_0\^[2/]{} (z)dz, \[homese\] where $\Gamma$ is the elastic scattering rate and $z\equiv \omega/\Gamma$. In the dirty limit where $\Delta/\Gamma\ll 1$, the above relation yields Homes’ Law [@homes], $\rho_s\approx \sigma(0)\Delta$, whereas in the clean limit $\rho_s=\rho_n$ where $\rho_n$ is the conductivity spectral weight in the normal state. Eqn. \[homese\] is particularly useful when the normal state resistivity data can reasonably be extrapolated to $T=0$. By combining dc transport data as a function of $x$ [@Matsuda10b] and a measurement of 2$\Delta$ from optical conductivity [@Matsuda14], Eq. \[homese\] provides a lower bound on $\lambda_L^2(0)$ (with the assumption that $\Delta$ is independent of $x$). Fig.\[homes\] shows $\lambda_L^2(0)$ as a function of $x$ obtained under this assumption (details of the procedure are presented as Supplementary Information). The decrease of superfluid density on the underdoped side reflects the growth in residual resistivity that begins as $x$ drops below about 0.33.
The values of $\lambda_L^2(0)$ estimated from Eq. \[homese\] form a baseline for comparison with the experimental results presented in Ref. [@Matsuda12]. On the same graph in Fig. \[homes\], we show the experimentally measured $\lambda_L^2(0)$ [@Matsuda12]. The data generally reflect the trend expected from the variation in the residual resistivity, with the exception of the sample with $x=0.3$, in which the condensate spectral weight is suppressed by about 40% from the Homes’ Law estimate. Given the constraints imposed by the sum rule, there are two possible sources of this discrepancy: (*i*) the quasiparticle mass could be renormalized at this value of $x$, corresponding to an intrinsic decrease in $\rho_n$, or, ([*ii*]{}) a considerable fraction of the (unrenormalized) $\rho_n$ could fail to contribute to the low temperature superfluid density. The latter possibility is suggested within the scenario that we develop here.
We analyze the above experiments by assuming a weakly first-order transition [@RFAVC], and argue that the presence of quenched disorder leads to formation of a [*micro-emulsion*]{} at small scales [@dis]. The system consists of SC puddles, where some of the puddles additionally have SDW order (see Fig. \[ph\]a inset). The SDW(+SC) regions, which have a locally well-developed antiferromagnetic moment but no long-range orientational order, act as barriers between the different SC grains. Upon moving deeper into the ordered side of the transition, the SDW(+SC) regions start to percolate and crossover to a state with long-range SDW order; this is the regime with a microscopically coexistent SC+SDW. As a function of decreasing $x$, the micro-emulsion is therefore a transitional state (shown as grey region in Fig. \[ph\]a) between a pure SC and a coexistent SC+SDW. Recent experiments in the vicinity of optimal doping using neutron-scattering and NMR have found results broadly consistent with our proposed phase diagram [@dai15]. We note that the granular nature of superconductivity should have no effect on the bulk $T_c$ in the presence of percolating SC channels.
[*Model.-*]{} When the system is well described in the vicinity of $x_{c}$ by a micro-emulsion as explained above, the phase fluctuations associated with the SC grains (shown as purple regions in Fig. \[ph\]a inset), can be modeled by the following effective theory, H\_= - \_[a,b]{}J\_[ab]{}(\_a-\_b), where $J_{ab}$ represent the Josephson junction (JJ) couplings between grains ‘$a$’ and ‘$b$’. We have ignored the capacitive contributions.
The Josephson current across the junction will be given by $I_s=J_{ab}\sin(\theta_a-\theta_b)$, and $J_{ab}$ may therefore be interpreted as the lattice version of the local superfluid density, $\rho_s(\r)$, i.e. ${\boldsymbol{J}}_s({\boldsymbol{r}})=\rho_s({\boldsymbol{r}})~ {\boldsymbol{v}}_s(\r)$, with ${\boldsymbol{J}}_s({\boldsymbol{r}}),~{\boldsymbol{v}}_s(\r)$ representing the superfluid-current and velocity respectively. Having a frustrated JJ (also known as a $\pi-$junction) with a negative value of $J_{ab}$ leads to a local suppression in $\rho_s$. Similar ideas have been discussed in the past in a variety of contexts (see Refs. [@KivSpi91] for a specific example), though the mechanism considered here will be different. We shall now propose an explicit scenario under which a suppression in $\rho_s$ arises in the vicinity of putative magnetic QCPs, utilizing the SC gap structure in the material under question.
The basic idea is as follows: suppose that the tunneling of electrons between the two grains is mediated by the SDW moment in the intervening region [@PWA66], and is accompanied by a transfer of finite momentum that scatters them from a hole-like to an electron-like pocket. Because the SC gaps on the two pockets have a relative phase-difference of $\pi$, the JJ coupling will be frustrated [@Ambegaokar63].
Let us first focus on a single grain. In order to capture the multi-band nature of the SCs, we introduce two superconducting order parameters, $\D_i$ with $i=\pm$ to model the $s^\pm$ state on the two pockets. Microscopically, these belong to regions in the grain having different momenta, $\k_\Vert$, parallel to the junction. The gaps are related to the microscopic degrees of freedom [@EBNL] via the following relation, \_i(z)=\_[\_\_i]{} V\_[\_,’\_]{}\_[’\_]{}\_[-’\_]{}, where $\psi^\dagger_{\k_{\Vert}\sigma}$ creates an electron at position $z$ with momentum $\k_\Vert$ parallel to the junction and spin $\sigma$. $V_{\k_\Vert,\k'_\Vert}$ is the pairing interaction in the Cooper channel and $z$ is the coordinate perpendicular to the junction with area $A$. The regions $\R_i$ are defined as, $\R_+=\{\k_\Vert| k_0>|\k_\Vert|\}$ and $\R_-=\{\k_\Vert| k_0\leq|\k_\Vert|\}$, where $k_0$ is an arbitrary momentum scale chosen such that $\D_+>0,~\D_-<0 $ (see Fig. \[JJ\] for an illustration). We’ll assume that such a prescription is valid for each grain, with possibly different values of $k_0$.
Let us then write down a model for the two coupled SC grains with an intervening proximity coupled SDW that has a well developed moment, ${\boldsymbol{\n}}$. Our notation is as follows: we use $\alpha=a, b$ to denote the grain index and $i=\pm$ to denote the band index within each grain. From now on, we relabel $\k_\Vert$ as $\k$. We introduce the Nambu spinor, $\Psi^{\alpha\dagger}_{i,\k,\sigma}=(\psi_{i,\k,\sigma}^{\dagger\alpha}~\epsilon_{\sigma\sigma'}\psi^\alpha_{i,-\k,\sigma'})$, where now $\psi^{\dagger\alpha}_{i,\k,\sigma}$ creates an electron with momentum $\k$ parallel to the junction and at a position $z$ (label suppressed), which belongs to a region of band “$i$" within grain “$\alpha$". The effective Hamiltonian is given by, \[Heff\] H\_&=&H\_+H\_T,\
H\_&=&\_[,i,]{} \_[i,,]{}\^\_[i,,]{}\^,\
H\_T&=&g\_[k]{} (\^[a]{}\_[+,,]{}\[\_[’]{}\^0\]\_[-,,’]{}\^b\
&& + \^[a]{}\_[-,,]{}\[\_[’]{}\^0\]\_[+,,’]{}\^b ) + , where $g$ is the tunneling matrix element, $\hat\tau^i$ $(i=0,x,y,z)$ act in Nambu space and $\hat\sigma^i$ $(i=0,x,y,z)$ act in spin space.
In the above, $H_\D$ corresponds to the bare pairing Hamiltonian written for the $\pm$ bands within each of the two grains. $H_T$ represents the SDW moment mediated hopping of electrons from one grain to the other (represented by the $a, b$ superscripts) and simultaneously scattering from one band to the other (represented by the $\pm$ subscripts). Therefore, ${\boldsymbol{\n}}$ imparts a finite momentum (along the interface) to the electrons when it scatters them from the electron (hole) pocket on one grain to the hole (electron) pocket on the other grain (shown as the black arrows in Fig. \[JJ\]).
[*Results.-*]{} Using the Ambegaokar-Baratoff relation [@Ambegaokar63], we can write the Josephson coupling (at $T=0$) between the two grains as, J\_[ab]{}= where $E_\ell^2=\e_\ell^2+\D_\ell^2$ and $\ell,~\ell'$ represent the band indices on the different grains. Since $\D_\ell\D_{\ell'}<0$, the coupling $J_{ab}<0$. Note that the specific nature of the frustrated tunneling arises from the same spin-fluctuation mediated mechanism that is predominantly responsible for the $s^\pm-$ pairing symmetry [@DS12]. However, there will also be a direct tunneling term (not included in Eqn. \[Heff\]) in the Hamiltonian, which does not scatter the electrons from one pocket to the other, as they hop across the junction. The contribution to the JJ coupling from this term will be unfrustrated (i.e. $J_{ab}>0$).
The ratio of the tunneling amplitudes in the two different channels is non-universal and depends on various microscopic details. In particular, the emulsion is associated with a distribution of Josephson-couplings, ${\cal{P}}(J)$, with a mean coupling strength, $\langle J\rangle=\bar{J}$. If a substantial fraction of the JJ couplings become negative due to the mechanism proposed above, $\bar{J}$ will be small, and the superfluid density will be suppressed (see green curve in Fig.\[ph\]b).
We now propose a resolution as to the fate of the uncondensed spectral weight (highlighted in Fig. \[homes\]), which can potentially be tested by measurements of the low frequency optical conductivity. Frustrated $\pi-$junctions host gapless states at the interface between the two grains [@RJCR76; @hutanaka], giving rise to a finite density of states around zero energy (see Fig \[sigw\] inset). As a result of the gapless ‘normal’-fluid component at the interface, a fraction $f$ of the spectral weight will be displaced from the superfluid-density to non-zero frequencies (shaded region in Fig. \[sigw\]). Given that the weight of the condensate is proportional to $\bar{J}(1-f)$, the 40% suppression in $\rho_s$ for BaFe$_2$(As$_{1-x}$P$_x$)$_2$ in the vicinity of the putative QCP corresponds to $f\sim 0.6$.
Our proposed optical conductivity, $\sigma(\omega)$, in the vicinity of penetration depth anomaly is shown in in Fig \[sigw\]. The spectrum shows clearly that the connection between normal state conductivity and superfluid density implied by Eq. \[homes\] will break down. In particular, $\sigma_{\tn{dc}}$ (which is a property of the normal state), could vary monotonically with isovalent-doping across $x_c$, while the abundance of low-energy excitations in the immediate vicinity of $x_c$ would give rise to a non-monotonic variation in the superfluid density. This allows for an unusual way of rearranging spectral weight in the [*superconducting*]{} state below the gap, without violating optical sum-rules.
The above scenario will give rise to a number of interesting low temperature thermodynamic and transport properties, as we now discuss. First of all, there should be a striking enhancement in the low-temperature thermal conductivity and specific-heat, as a function of $x$ in the narrow vicinity of $x_c$, due to the ‘normal’-component. It is important to recall that this material has loop-like nodes on the electron-pockets [@Nodes]. However, the geometry of the electron-pockets and the magnitude of the gap do not change substantially in the vicinity of $x_c$, and therefore it is unlikely that the contribution to the above quantities from the nodal-quasiparticles will have a drastic modificiation. It should therefore be relatively straightforward to disentangle the contribution arising from the nodal versus the ‘normal’ quasiparticles. Studying the NMR-spectra as a function of decreasing temperature (across $T_c$) and down to sufficiently low temperatures in the vicinity of $x_c$ should also reveal the spatial inhomogeneity associated with the SDW regions. A large residual density of states in the superconducting state has been detected at a particular P-doping via the power-law temperature dependence of $1/T_1\sim T$ [@ishida10]. Within our scenario, there should be a striking enhancement in this quantity as a function of doping around $x_c$. Finally, we note that a promising direction for future studies would be to measure the magnetic-field distribution due to the propagating currents in the emulsion using NV-based magnetometers [@AY13].
[*Discussion.-*]{} The theoretical study in this paper was motivated by a number of remarkable experiments carried out in BaFe$_2$(As$_{1-x}$P$_x$)$_2$, as a function of $x$ in the normal and superconducting phases. Our primary objective was to provide an explanation for the striking enhancement of the London penetration depth in the vicinity of a putative SDW QCP in the SC state. We developed a scenario based on the idea that true SDW criticality is masked by a weak first-order phase transition in the superconducting state at $T=0$. In this picture, quenched disorder naturally gives rise to an [*emulsion*]{} at small length scales with puddles of SC and SDW(+SC). It is then, in principle, possible for SDW moments at the interface of the SC grains to generate frustrated Josephson couplings, which deplete the local superfluid-density. Our proposed scenario naturally calls for a number of experimental tests that should be carried out in the near future, which should directly look for both the spatial inhomogeneities associated with the emulsion [@curro15], and probe the gapless excitations using thermodynamic probes, as explained above.
In addition to experiments on BaFe$_2$(As$_{1-x}$P$_x$)$_2$, it should be important to further investigate the contrasting behavior of the electron-doped system, Ba(Fe$_{1-x}$Co$_x$)$_2$As$_2$, where $\lambda_L(0)$ behaves monotonically as a function of $x$ across the putative QCP [@Gordon10]. Electron-doping leads to significantly higher amounts of disorder compared to the isovalently-doped case, and would therefore lead to puddles with typically much smaller size [@Curro13]. Our proposed mechanism for the strong suppression of the superfluid-density in the isovalently-doped material relies on the existence of an emulsion with puddles of appreciable size, in the presence of an optimal amount of disorder. A comparison of the NMR spectra in the narrow vicinity of the putative QCP in the electron and isovalently doped materials would shed light on these microscopic differences between the two families.
Finally, though we have hypothesized that the SDW onset transition [*inside*]{} the SC is, in the absence of disorder, a weak first order transition, we emphasize that the normal state properties are consistent with the presence of a “hidden" QCP around optimal doping [@walmsley13; @Matsuda10a; @IF14]. It is plausible that in the normal state, different experimental techniques are probing the critical fluctuations associated with not one, but distinct QCPs as a function of $x$. For instance, $m^*$ extracted from high-field quantum oscillations is dominated by the vicinity of ‘hot-spots’, where quasiparticles are strongly damped due to coupling to the SDW fluctuations [@TS14]. On the other hand, strong critical fluctuations associated with the nematic order-parameter [@FCS14], that couple to the entire Fermi-surface, would dominate $m^*$ extracted at zero-field from the jump in the specific heat at $T_c$.
[*Acknowledgements.-*]{} We thank A. Carrington, A. Chubukov, N. Curro, J.C. Davis, R. Fernandes, K. Ishida, M.-H. Julien, S. Kivelson, Y. Matsuda, A. Millis and A. Vishwanath for useful discussions. We thank K. Hashimoto and Y. Matsuda for providing us with the data shown in Fig.\[homes\]. DC is supported by the Harvard-GSAS Merit Fellowship and acknowledges the “Boulder summer school for condensed matter physics - Modern aspects of Superconductivity", where some preliminary ideas for this work were formulated. DC and SS were supported by NSF under Grant DMR-1360789, the Templeton foundation, and MURI Grant W911NF-14-1-0003 from ARO. TS was supported by Department of Energy DESC-8739- ER46872, and partially by a Simons Investigator award from the Simons Foundation. JO acknowledges the Office of Basic Energy Sciences, Materials Sciences and Engineering Division, of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231 for support. Part of this work was completed when JO was visiting MIT as a Moore Visitor supported by grant GBMF4303. Research at Perimeter Institute is supported by the Government of Canada through Industry Canada and by the Province of Ontario through the Ministry of Research and Innovation.
[99]{} L. Taillefer, Ann. Rev. Cond. Mat. Phys. [**1**]{}, 51 (2010); S. Sachdev, Science [**336**]{}, 1510 (2012). D.J. Scalapino, Rev. Mod. Phys. [**84**]{}, 1383 (2012). T. Shibauchi, A. Carrington and Y. Matsuda, Ann. Rev. Cond. Mat. Phys. [**5**]{}, 113 (2014). H. Shishido et al., Phys. Rev. Lett. [**104**]{}, 057008 (2010). P. Walmsley et al., Phys. Rev. Lett. [**110**]{}, 257002 (2013). R. Fernandes, S. Maiti, P. Wolfle and A. Chubukov, Phys. Rev. Lett. [**111**]{}, 057001 (2013); J. Wu, Q. Si and E. Abrahams arXiv:1406.5136. Y. Imry and M. Wortis, Phys. Rev. B [**19**]{}, 7 (1979); K. Hui and A. Nihat Berker, Phys. Rev. Lett. [**62**]{}, 21 (1989). K. Hashimoto et al., Science [**336**]{}, 1554 (2012). A. Levchenko, M.G. Vavilov, M. Khodas and A.V. Chubukov, Phys. Rev. Lett. [**110**]{}, 177003 (2013). D. Chowdhury, B. Swingle, E. Berg and S. Sachdev, Phys. Rev. Lett. [**111**]{}, 157004, (2013). There is reason to believe that there are accidental (loop-like) nodes on the electron-pockets in the s$^\pm$ state in this particular material (see refs. [@Nodes]); this feature doesn’t affect most of the qualitative features of the computation as long as the nodes do not coincide with the SDW “hot-spots". K. Hashimoto [*et al.*]{}, Phys. Rev. B [**81**]{}, 220501(R) (2010); T. Shimojima et al., Science [**332**]{}, 564 (2011); Y. Zhang et al., Nat. Phys. [**8**]{}, 371 (2012); Y. Mizukami [*et al.*]{}, Nat. Comms. [**5**]{}, 5657 (2014). C.C. Homes et al., Nature [**430**]{}, 539 (2004); S.V. Dordevic, D.N. Basov and C.C. Homes, Sci. Rep. [**3**]{}, 1713 (2013). S. Kasahara et al., Phys. Rev. B [**81**]{}, 184519 (2010). S.J. Moon et al., Phys. Rev. B [**90**]{}, 014503 (2014). D. Hu et al., Phys. Rev. Lett. [**114**]{}, 157002 (2015) B.I. Spivak and S.A. Kivelson, Phys. Rev. B [**43**]{}, 3740 (1991); J.A. van Dam, Y.V. Nazarov, E.P.A.M. Bakkers, S.DeFranceschi and L.P. Kouwenhoven, Nature [**442**]{}, 667 (2006). P.W. Anderson, Phys. Rev. Lett. [**17**]{}, 95 (1966). V. Ambegaokar and A. Baratoff, Phys. Rev. Lett. [**10**]{}, 486 (1963). E. Berg, N.H. Lindner and T. Pereg-Barnea, Phys. Rev. Lett. [**106**]{}, 147003 (2011). R. Jackiw and C. Rebbi, Phys. Rev. D [**13**]{}, 12 (1976). C-R Hu, Phys. Rev. Lett. [**72**]{}, 10 (1994); Y. Tanaka and S. Kashiwaya, Phys. Rev. Lett. [**74**]{}, 17 (1995). Y. Nakai et al., Phys. Rev. B [**81**]{}, 020503(R), 2010. S. Hong, M.S. Grinolds, L.M. Pham, D.L. Sage, L. Luan, R.L. Walsworth and A. Yacoby, MRS Bulletin [**38**]{}, 155 (2013). N. Curro, private communication and <http://meetings.aps.org/Meeting/MAR15/Session/T5.7> R.T. Gordon [*et al.*]{}, Phys. Rev. B [**82**]{}, 054507 (2010). A. P. Dioguardi [*et al.*]{}, Phys. Rev. Lett. [**111**]{}, 207201 (2013). J. Analytis [*et al.*]{}, Nat. Phys. [**10**]{}, 194 (2014). T. Senthil, arXiv:1410.2096. R. Fernandes, A.V. Chubukov and J. Schmalian, Nat. Phys. [**10**]{}, 97 (2014).
Supplementary Material
----------------------
Estimate of $\lambda_L^2(0)$ from Homes’ law
============================================
In this supplementary material, we compare the value of the penetration depth obtained from experiments [@Matsuda12] with the prediction from Homes’ law; for the latter, we use a combination of the experimental data obtained from optical-conductivity and dc transport. For each value of the doping ($x$), we estimate the (approximate) dc resistivity ($\rho_{xx}$) by extrapolating the curves to $T=0$, from the transport data in fig.1(b) of Ref.[@Matsuda10b]. We estimate the value of $2\D$, where $\Delta$ is the superconducting gap, from the data for optical conductivity in the superconducting state, as shown in fig. 3(b) of Ref. [@Matsuda14]. Since $T_c$ remains relatively unchanged as a function of $x$ in the vicinity of optimal doping, we assume $\D$ to be independent of $x$ such that $2\D\approx150 $cm$^{-1}(=2.827\times10^{13}$s$^{-1})$. Then, in the dirty limit, \_s = \_.
In order to obtain the penetration depth, we need to restore various dimensionful constants such that, \_L\^2(0)=, where $c(=3\times10^{8}$ m/s) is the speed of light and $\ve_0(=8.85\times10^{-12}$ F/m; 1 F=1 $\Omega^{-1}$s) is the permitivity of free space. The values obtained are shown in the table below and have been presented in fig. [ 2]{} of the main text, along with a comparison to the experimental data [@Matsuda12].
------ ----------------------------------------- ------------------ --
$x$ $\rho_{xx}$ (from Ref. [@Matsuda10b]) $\lambda_L^2(0)$
($\mu\Omega$ cm) ($\mu$m$^2$)
0.23 130 0.057
0.27 120 0.053
0.33 35 0.015
0.41 20 0.009
0.56 15 0.007
0.64 10 0.004
------ ----------------------------------------- ------------------ --
\[data\]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Instead of the commonly used chemical doping, it can be more favorable to consider transforming graphene through proximity effects by carefully choosing its adjacent regions. While gate-tunable room-temperature spin-dependent properties could be induced in graphene by magnetic proximity effects from common metallic ferromagnets, this approach is complicated by chemical bonding between a metal and graphene suggesting the need for an intervening buffer layer. However, even with a buffer layer there is still a large energy shift of the Dirac cone in graphene away from the Fermi level. Compared to such a large negative shift and its resulting $n$-doping when graphene is separated from cobalt by a monolayer h-BN or another layer of graphene, we show that it can be favorable to instead separate graphene by a monolayer of gold or platinum. The resulting proximity induced magnetization is larger, energy shift is somewhat reduced and changes its sign, offering a path for proximity-induced spin polarization in graphene which can be tuned at smaller gate-controlled electric field than for the h-BN buffer layer.'
author:
- Mihovil Bosnar
- Ivor Lončarić
- 'P. Lazić'
- 'K. D. Belashchenko'
- Igor Žutić
bibliography:
- 'references.bib'
title: 'Proximity-induced magnetization and towards the efficient spin gating in graphene'
---
Introduction
============
Despite its attractive properties, there is a strong interest to modify graphene (Gr) by introducing superconductivity, magnetism, a sizeable energy gap, or strong spin-orbit coupling. A common approach to accomplish this by chemical doping or functionalization typically poses inherent difficulties, from introducing unwanted disorder to significantly reducing Gr’s high mobility [@Zutic2019:MT]. An alternative approach to transform graphene and overcome these difficulties is provided by proximity effects, whereby it acquires properties of its neighbors, for example, becoming superconducting, magnetic, topologically nontrivial, or with an enhanced spin-orbit coupling [@Heersche2007:N; @Wei2016:NM; @Gmitra2016:PRB; @Avsar2014:NC; @Rossi2020:AP; @Lee2016:PRB; @PhysRevMaterials.2.054002]. Being atomically thin, Gr and a growing class of two-dimensional (2D) materials are changed significantly even with short-range proximity effects [@Zutic2019:MT].
In this work we focus on magnetic-proximity effects in Co/Gr-based hetrostructures, as shown in Fig. 1, that could enhance spintronic applications. However, the interest in magnetic-proximity effects in 2D systems is much broader, as they are considerd for implementing magnetic skyrmions [@Yang2018:NM] and exotic properties of topological insulators [@Qi2011:RMP], as well as realizing Majorana bound states for topological quantum computing [@Sau2010:PRL; @Fatin2016:PRL; @Zhou2019:PRB; @Desjardins2019:NM].
With its high mobility and low spin-orbit coupling, Gr is expected to be a particularly suitable material for spin transport and spintronics [@Han2014:NN; @Lin2019:NE; @Avsar2020:RMP]. From the first demonstration of spin injection in Gr [@Tombros2007:N], there was a significant progress in extending characteristic timescales and lenghtscales over which the spin information can be sustained [@Drogeler2014:NL; @Zomer2012:PRB]. Gr-based spin-logic gates have been demonstrated at room temperature [@Wen2016:PRA], supporting proposals for specialized applications which could outperform CMOS-based counterparts [@Dery2012:IEEETED].
![Schematic view of the system.[]{data-label="fig:scheme"}](shema.png)
Many spintronic applications rely on ferromagnet(F)/Gr junctions as their building block where the functionalities are realized by changing the magnetization orientation in multiple ferromagnetic layers [@Han2014:NN; @Lin2019:NE; @Avsar2020:RMP; @Lazic2014:PRB; @Raes2017:PRB; @Ringer2018:APL]. An alternative approach, to those employing an external applied magnetic field, is to take advantage of the electrically-tunable magnetic proximity effect in Gr, which is potentially faster and more energy-efficient [@Lazic2016:PRB; @Zhao2019:2DM]. Both the magnitude and the sign of proximity-induced spin polarization in Gr can be controlled by electric gating [@Lazic2016:PRB].
Common expectations for tunable magnetic proximity effects imply that magnetic insulator is required to avoid a short-circuit effect of a metallic F [@Yang2013:PRL; @Wang2015:PRL; @Swartz2012:ACSN; @Behera2019:PCCP; @Takiguchi2019:NP; @Cortes2019:PRL]. This is further supported since Gr forms a strong chemical bonding on Co [@Lazic2016:PRB; @Zollner2016:PRB; @Giovannetti2008:PRL] which removes its spin degeneracy, but also essentially turns Gr into a metallic continuation of Co with a large density of states (DOS) near the Fermi level. Because of that, it is practically impossible to change the spin polarization of Gr on Co by applying the gate electric field [@Lazic2016:PRB].
Therefore, the idea to select heterostructures with a van der Waals (vdW) bonded layer of Gr was introduced [@Lazic2016:PRB]. A common metallic F with high Curie temperature could then be considered for tunable magnetic proximity effects. In particular, surface passivation through an addition of another Gr layer [@Lazic2016:PRB] or a layer of h-BN [@Lazic2016:PRB; @Zollner2016:PRB] as a buffer between Gr and Co, as shown in Fig. 1, was studied. In both cases the passivation results in the top layer of Gr being bound to the underlying structure by the vdW interaction [@Lazic2016:PRB].
A characteristic property of vdW bonding is that it preserves the features of the electronic structure of bound systems, such as the Dirac cone in Gr [@Lazic2016:PRB; @Giovannetti2008:PRL; @Asshoff2017:2DM]. However, a few major differences in comparison to a freestanding system are also introduced. In systems studied previously, the most relevant are the ($n$-type) doping [@Lazic2016:PRB; @Giovannetti2008:PRL] and removing the spin degeneracy due to the proximity effect [@Lazic2016:PRB; @Zollner2016:PRB]. The change in the electronic states near the Fermi level compared to the freestanding Gr can be described as a spin-dependent band shift [@Lazic2016:PRB]. This is supported by the recent transport experiments [@Asshoff2017:2DM]. The role of Gr was not viewed as a spin filter [@Karpan2007:PRL], but a source of spin-polarized carriers itself, arising from an interplay of doping by the metallic F and the proximity-induced spin splitting in graphene [@Asshoff2017:2DM].
Gr and the passivated surface can be thought of as two plates of a capacitor; gate voltage produces an electrostatic potential difference and charge transfer between them. This changes the spin polarization of Gr at the Fermi level in magnitude and sometimes even sign [@Lazic2016:PRB; @Zollner2016:PRB]. An electrostatic model based on this picture predicts that the induced potential difference between the slab and Gr is increased by lowering the DOS at the Fermi level in grahene [@Lazic2016:PRB]. It is, therefore, expected that the spin polarization should be more sensitive to gating if the Fermi level is close to the Dirac point in Gr.
With previously studied h-BN and Gr passivation layers, it was found that the pysisorbed Gr is n-doped and the apex of its Dirac cone is shifted approximately 0.5 eV below the Fermi level, resulting in relatively large DOS at the Fermi level [@Lazic2016:PRB]. First-principles calculations have shown that tuning the DOS spin polarization in Gr in this system requires large electric fields [@Lazic2016:PRB], achievable only by liquid gating [@Hebard1987:IEEETM; @Mannhart1993:APL] and unsuitable for applications. This is further corroborated experimentally in lateral Co/h-BN/Gr-based spin valves [@Xu2018:NC]. Within the 2D ferromagnetic contacts in a planar geometry, similar as that depicted in Fig. 1, an electrostatic gating failed to produce tunable magnetic proximity effects. Instead, to reduce a detrimental $n$-doping of Gr, 1D edge contacts were used to realize such gate-controlled tunability and the reversal of proximity-induced spin polarization in Gr [@Xu2018:NC].
In this work we explore a path to reduce the magnitude of electric fields and realize tunable spin polarization in common 2D geometries by recognizing that Gr on Au and Pt is predicted to have $p$-doping [@Giovannetti2008:PRL] and could partially counteract the $n$-doping of Co contacts. Our related first-principles calculations provide guidance towards optimized structures for proximity-induced magnetization and spin gating in Gr, which allows spin control of its electronics properties. While the two doping effects are not cancelled, compared to Co/h-BN/Gr heterostructures Au or Pt passivation yields more than an order of magnitude larger spin polarization in Gr at electric fields < 0.01 V/Å, within the range of electrostatic gating, reaching 15 % with Pt.
Computational details
=====================
Proximity-induced spin polarization of graphene was studied using density functional theory (DFT) as implemented in a real-space code GPAW [@gpaw1; @gpaw2; @ase]. The reason for using a real space approach instead of a plane wave expansion is to avoid the use of periodic boundary conditions in the non-periodic direction of the system to prevent electron tunneling in that direction [@Lazic2016:PRB], since such use of the periodic boundary conditions introduces spurious interactions with periodically repeated images of the system which need to be corrected [@Hakan2013:JPCM].
In all calculations the semilocal vdW exchange-correlation functional vdw-df-cx [@cx1; @cx2; @PhysRevMaterials.3.063602] from the libvdwxc library [@libvdwxc] was used. The cell was sampled by a grid of $0.1332\ {\textup{\AA}}$ spacing in the planes parallel to the heterostructure layers, and $0.166\ {\textup{\AA}}$ in the perpendicular direction. The GPAW PAW [@paw1; @paw2] PBE [@pbe] setups were used for all calculations.
Input structures were constructed in QuantumWise Virtual NanoLab [@vnl]. The slab of hexagonal close-packed ${\textup{Co}}$ terminated by a $(0001)$ surface was modeled by $7$ layers of atoms. The passivating ${\textup{Au}}$ and ${\textup{Pt}}$ monolayers were added in an extension of this slab, forming a $1\times1$ supercell. Since the lattice constant of Gr ($a=2.46\ {\textup{\AA}}$) is similar to that of the ${\textup{Co}}$ $(0001)$ surface ($a=2.51\ {\textup{\AA}}$), the slab was strained $2.3\ \%$ so that it forms a $1\times1$ supercell with Gr, which greatly reduces the computational burden compared to working with larger supercells. The cell contained $10\ {\textup{\AA}}$ of vacuum on both sides of the heterostructure.
Since Gr can be positioned differently relative to the surface atoms, the systems were first relaxed in three high-symmetry configurations (${\textup{Au}}$ or ${\textup{Pt}}$ atom located above the hollow site of Gr, over a bond between the two ${\textup{C}}$ atoms in the unit cell, or on top of one of them) to find the one with the lowest energy. The relaxations were done using the BFGS line search (Quasi Newton) algorithm as implemented in the ASE package [@ase], with GPAW as the DFT calculator. The relaxations were performed until the force between the atoms was smaller than $0.05\ {\textup{eV}}/{\textup{\AA}}$, with a $15\times15\times1$ Monkhorst-Pack sampling of the Brillouin zone, the Fermi-Dirac smearing of $200\ {\textup{m}}{\textup{eV}}$ and the bottom two layers of ${\textup{Co}}$ fixed.
The calculations show that the configuration in which the ${\textup{Au}}/{\textup{Pt}}$ atoms are located above the hollow sites of Gr, shown in Fig. \[system\], is optimal for both passivation layers. The obtained optimal ${\textup{Co}}/{\textup{Au}}/{\textup{Gr}}$ and ${\textup{Co}}/{\textup{Pt}}/{\textup{Gr}}$ structures were relaxed further with a homogeneous electric field included in the $\pm z$ direction. The field magnitudes of $0.01\ {\textup{V}}/{\textup{\AA}}$, $0.05\ \ {\textup{V}}/{\textup{\AA}}$, $0.1\ {\textup{V}}/{\textup{\AA}}$ and $0.2\ {\textup{V}}/{\textup{\AA}}$ were used. To improve the precision of the calculated magnetic moments, a self-consistent calculation was performed for each relaxed structure with a dense $63\times63\times1$ k-point mesh and the Fermi-Dirac smearing of $10\ {\textup{m}}{\textup{eV}}$. The magnetic moments are estimated as a projection of the spin density
$$\mathcal{M}(\textbf{r})=\rho_{\uparrow}(\textbf{r})-\rho_{ \downarrow}(\textbf{r}),
\label{density}$$
on the localized atomic setup functions.
A large number of $k$ points is also needed to resolve the Dirac cone of Gr in the calculations of the projected density of states (PDOS) on carbon atoms. The PDOS was calculated with the Gaussian broadening of $200\ {\textup{m}}{\textup{eV}}$.
The STM images were simulated following the Tersoff-Hamann theory [@tersoff_hamann], as implemented in the ASE package. The ASE tool was modified so that the images can be made separately for the two spin channels.
Results and discussion
======================
Zero-gate voltage
-----------------
The band structures of both systems without the applied field are shown in Fig. \[bandstructure\]. Clearly, in both cases the Dirac cone of Gr is preserved upon adsorption but shifted to higher energies by approximately $0.369\ {\textup{eV}}$ and $0.295\ {\textup{eV}}$ in the case of ${\textup{Pt}}$ and ${\textup{Au}}$ passivation layer, respectively. Along with the obtained distances of about $3.5\ {\textup{\AA}}$ between Gr and the Pt or Au layer (Fig. \[system\]), these results suggest vdW bonding between Gr and the metal. Gr is p-doped by an amount somewhat smaller than the $n$-doping of Gr in the system with a h-BN passivation layer [@Lazic2016:PRB]. As shown in the insets in Fig. \[bandstructure\], lifting of the spin degeneracy is also induced in Gr.
According to the model Hamiltonian of Gr in contact with passivated ${\textup{Co}}$ surfaces [@Zollner2016:PRB], the spin splitting can be separated into two contributions: Exchange splitting that is constant in energy and $k$ near the Fermi level and $K$ point, and the energy and $k$ dependent contribution due to hybridization. The exchange part can be estimated as a difference of the energies of spin-up and spin-down Gr bands far away from the avoided crossings, while the hybridization contribution is more difficult to quantify.
The spin splitting of the Gr bands in ${\textup{Co}}/{\textup{Au}}/{\textup{Gr}}$ is approximately constant near the Fermi level, as illustrated in the inset of Fig. \[bandstructure\](a), and averages to about $1.05\ {\textup{m}}{\textup{eV}}$. In contrast, the gaps at the avoided crossings, such as the one shown in Fig. \[bandstructure\](b), are so large in ${\textup{Co}}/{\textup{Pt}}/{\textup{Gr}}$ that the spin splitting is not uniform in any part of the band structure near the Fermi level.
The larger spin splitting on avoided crossings for ${\textup{Pt}}$ passivation layer can be understood from the orbital analysis of the flat band near the Fermi level, which shows that this band have dominantly contributions from the orbitals of ${\textup{Pt}}$ and the first ${\textup{Co}}$ layer under it. For the ${\textup{Au}}$ passivation layer the flat bands contributions does not come from ${\textup{Co}}$ atoms close to surface or ${\textup{Au}}$ orbitals, but from bulk Co atoms.
This is because the spin-down $d$ states of both ${\textup{Pt}}$ and ${\textup{Co}}$ are partially filled and can be found near the Fermi level, which in turn allows their strong hybridization in forming of the chemical bond between ${\textup{Pt}}$ and ${\textup{Co}}$. Additionally, ${\textup{Pt}}$ is close to fulfilling the Stoner criterion, meaning the electron correlations are strong enough to cause significant spin splitting of hybrid states. On the other hand, the $d$ states of ${\textup{Au}}$ are fully occupied so its $d$ states are located far from the Fermi level.
Therefore, polarized surface bands are present on ${\textup{Pt}}$ in the ${\textup{Co}}/{\textup{Pt}}/{\textup{Gr}}$ system, which means their overlap with Gr orbitals is larger than in the case of polarized bands of ${\textup{Co}}/{\textup{Au}}/{\textup{Gr}}$, and the larger spin degeneracy splitting is induced in Gr.
As a result of the spin splitting there is a finite spin density in Gr. The magnetization of graphene in ${\textup{Co}}/{\textup{Au}}/{\textup{Gr}}$ can be estimated from the Pauli susceptibility of pristine graphene. On one hand, it can be obtained by subtracting the integrals of spin-up and spin-down PDOS in graphene over the filled states, which gives $-9.27\times10^{-6}\ \mu_B$. On the other hand, calculating it from the Pauli susceptibility, with the exchange field playing role of magnetic field, gives $-8.61\times10^{-6}\ \mu_B$ (note that positive splitting as defined above yields spin-down polarization of graphene as spin-up band is raised in energy compared to spin-down). Similar prediction cannot be made for ${\textup{Co}}/{\textup{Pt}}/{\textup{Gr}}$.
However, the spin density of graphene is not homogeneous in either system because surface is not homogeneous. The plot of the spin density plane cut at $0.33\ {\textup{\AA}}$ above the graphene in Fig. \[polarization\_density\] shows that in the case of ${\textup{Co}}/{\textup{Pt}}/{\textup{Gr}}$ only the magnitude of the induced spin density in graphene varies, while in ${\textup{Co}}/{\textup{Au}}/{\textup{Gr}}$ the sign varies as well.
Spin-dependent transport effects are sensitive to the spin polarization $P(E)$ of the PDOS $N_s(E)$ as $$P(E)=\frac{N_{\uparrow}(E)-N_{\downarrow}(E)}{N_{\uparrow}(E)+N_{\downarrow}(E)},
\label{polarization}$$ which, near the the Fermi level, is smaller in ${\textup{Co}}/{\textup{Au}}/{\textup{Gr}}$ compared to ${\textup{Co}}/{\textup{Pt}}/{\textup{Gr}}$, as can be seen from Figs \[PDOS\](a) and (b). This is not caused by the variation of the sign of the spin density in space as seen in Fig. \[polarization\_density\]. By integrating the PDOS per atom $a$ in a selected spin channel in an interval from the Fermi level to a selected energy $E_0$ $$I_s^a(E_0)=\text{sgn}(E_F-E_0)\int_{E_0}^{E_F}N_{a,s}(E){\textup{d}}E,
\label{integral}$$ and subtracting the results for the spin channels, the contribution of the bands in the selected energy range to the atomic spin density is obtained. From the dependence of this difference on the selected $E_0$ for graphene, shown in Fig. \[integrated\_PDOS\], it is clear that near the Fermi level the $\pi$ bands give a “ferromagnetic” contribution to the net polarization in case of both ${\textup{Au}}$ and ${\textup{Pt}}$ passivation layer.
Therefore, the spin polarization of graphene $\pi$ bands near the Fermi level is larger for ${\textup{Co}}/{\textup{Pt}}/{\textup{Gr}}$ than for ${\textup{Co}}/{\textup{Au}}/{\textup{Gr}}$ only because of a much larger spin splitting in graphene near the Fermi level in former compared to latter, as discussed previously. Particularly, in the case of the ${\textup{Pt}}$ passivation layer, there is a surface band in the spin down channel around $12\ {\textup{m}}{\textup{eV}}$ above the Fermi level, which causes large splitting on the avoided crossing. Such a band is absent in the case of ${\textup{Co}}/{\textup{Au}}/{\textup{Gr}}$.
Furthermore, Fig. \[integrated\_PDOS\] also shows that regardless of the passivation layer, the contributions of graphene bands up to $8\ {\textup{eV}}$ below the Fermi level have to be included in order to obtain the spin density shown in Fig. \[polarization\_density\]. This means that not only the $\pi$ bands of graphene are spin polarized by the proximity effect, but also a part of its $\sigma$ bands. The energy range in which the graphene orbitals are spin polarized corresponds to the range of energies of the $d$ bands of the metals. Hence, we conclude that the hybridization of graphene bands with the $d$ bands of the metal gives rise to the proximity effect. We note that the total spin density differs significantly from the Fermi level spin density. This is because hybridization caused by avoided crossings far from the Fermi level also brings about additional charge transfer.
The energy-dependent spin-polarized DOS in Gr could be measured experimentally by a spin-polarized STM. According to Fig. \[integrated\_PDOS\], by selecting an appropriate bias a varying sign or a constant sign image could be seen in either system. In Fig. \[stm\_images\] we show an example of the difference of the simulated spin-resolved STM images, where the bias of the STM was selected so that the net spin polarization of the scanned orbitals in the system with the ${\textup{Au}}$ passivation layer varies in sign. For the same bias, the simulated difference in the system with the ${\textup{Pt}}$ passivation layer is constant in sign.
Effects of gating
-----------------
![Comparison of the atomic magnetic moments of ${\textup{C}}$ atoms of Gr for the system with ${\textup{Au}}$ (circles, lighter shade) and ${\textup{Pt}}$ (triangles, darker shade) passivation layer. Lines are guide to the eye.[]{data-label="atomic_polarizations"}](polarization_at_atoms.pdf)
With the gate electric field applied, charge is transferred between the metal and Gr. This produces a shift of the PDOS relative to the Fermi level, as shown in Figs. \[PDOS\](c) and (d). The shift can be traced through the position of the point where the PDOS has a minimum. This point moves away from the Fermi level for fields oriented along the positive $z$ direction because the electronic states of Gr are raised in energy compared to the metal and electrons flow from Gr to the metal, and vice versa.
For the same magnitude of the electric field, the PDOS of the system with the ${\textup{Au}}$ passivation layer shifts more than the PDOS of the system with the ${\textup{Pt}}$ passivation layer. This is in agreement with the electrostatic model [@Lazic2016:PRB], as the doping of Gr for the ${\textup{Au}}$ passivation layer is lower, and consequently the PDOS at the Fermi level is smaller compared to ${\textup{Pt}}$.
However, the change in the spin density does not directly reflect the amount of the charge transferred. Figure \[atomic\_polarizations\] shows the magnetic moments of ${\textup{C}}$ atoms of Gr as a function of the gate field magnitude. It is clear that the field-induced change in magnetic moments is greater in the case of the ${\textup{Pt}}$ passivation layer. Furthermore, the spin polarization of bands near the Fermi level as a function of the magnitude of the gate field is shown in Fig. \[PDOS\](a) and (b). Similarly to the magnetic moments, the energy-resolved spin polarization for the system with the ${\textup{Au}}$ passivation layer is less responsive to the gate field.
Stronger response in ${\textup{Co}}/{\textup{Pt}}/{\textup{Gr}}$ is due to the presence of spin-polarized bands with strong surface character slightly above the Fermi level. An electric field applied in the positive $z$ direction lifts the surface bands higher above the Fermi level, because they are localized on the side of the metal with lower electrostatic potential, and vice versa. Due to the strong dispersion near the avoided crossing in the spin-down channel, this shift results in a significant change in the polarization when the electric field is applied.
For ${\textup{Co}}/{\textup{Au}}/{\textup{Gr}}$ the surface character of $d$ bands is less pronounced and these bands shift significantly less with the applied field. On the other hand, the homogeneous splitting caused by interlayer exchange is unaffected by the field. Therefore, the total change in polarization of Gr in ${\textup{Co}}/{\textup{Au}}/{\textup{Gr}}$ is significantly smaller than in ${\textup{Co}}/{\textup{Pt}}/{\textup{Gr}}$.
The previously studied system with a h-BN passivation layer also exhibits a change in the sign of the spin polarization at the Fermi level with large gate field [@Lazic2016:PRB; @Zollner2016:PRB]. Due to different electronic configuration in ${\textup{Co}}/{\textup{Pt}}/{\textup{Gr}}$ or ${\textup{Co}}/{\textup{Au}}/{\textup{Gr}}$, the sign of the spin polarization at the Fermi level does not change with the tested gate fields.
Conclusions
===========
We have studied a system consisting of the ferromagnetic ${\textup{Co}}$ slab passivated by a layer of ${\textup{Au}}$ or ${\textup{Pt}}$ and a physisorbed layer of graphene which develops spin polarization through the proximity effect. This polarization can be tuned by the electric field, similarly to the previously studied systems where h-BN or graphene were used as passivation layers [@Lazic2016:PRB; @Zollner2016:PRB]. Another manifestation of the magnetic proximity is the induced spin density in the graphene layer, which has the same (opposite) sign on the two inequivalent carbon atoms in Co/Pt/Gr (Co/Au/Gr). The net polarization is produced by the polarization of both $\sigma$ and $\pi$ orbitals in the energy range where the $d$ bands of the metal are found. Inequivalence of induced polarization of carbon atoms manifests through different spin resolved local density of states on these carbon atoms which can be experimentally tested with a spin-polarized STM [@Tsymbal:2019].
For spintronic applications relying on transport [@Tsymbal:2019], the spin polarization near the Fermi level is important. Graphene in the ${\textup{Co}}/{\textup{Pt}}/{\textup{Gr}}$ system shows larger polarization near the Fermi level and stronger response to the applied field compared to ${\textup{Co}}/{\textup{Au}}/{\textup{Gr}}$. Without an applied electric field, the spin polarization for both passivation layers is larger compared to the previously studied Co/h-BN/Gr system [@Lazic2016:PRB], while the response to the electric field is only slightly stronger.
The large Gr spin polarization at the Fermi level in ${\textup{Co}}/{\textup{Pt}}/{\textup{Gr}}$ originates from the presence of a spin-down surface band just above the Fermi level. Stronger response to the electric field in this system is mediated by the field-induced shift of this band and is also facilitated by the lower doping of graphene compared to the systems with h-BN or Gr passivation layers. However, because the response to the field is weaker in ${\textup{Co}}/{\textup{Au}}/{\textup{Gr}}$ where the doping level is even lower compared to ${\textup{Co}}/{\textup{Pt}}/{\textup{Gr}}$, we conclude that the effect of the surface band is the most important.
While Co/Au/Gr and Co/Pt/Gr do not exhibit gate-controlled reversal of the proximity-induced spin polarization, which could be useful for implementation of Gr-based magnetologic gates and lateral spin-valves [@Han2014:NN; @Lin2019:NE; @Wen2016:PRA; @Xu2018:NC], we have obtained several important findings, discussed below, which could also stimulate future work. Co/Pt/Gr-based structures support gate-controlled modulation of spin polarization, which is the key element of spin-interconnects [@Zutic2019:MT]. Their constant-level charge current eliminates the crosstalk problems of unintended signals from electromagnetic induction in the neighboring wires based on the modulation of charge current in conventional interconnects [@Dery2011:APL].
Since Au and Pt layers are spatially closer to Gr than Co, their $p$-doping character dominates the position of the Dirac cone compared to the $n$-doping of Gr by Co known from Co/h-BN/Gr heterostructures [@Lazic2016:PRB; @Zollner2016:PRB; @Asshoff2017:2DM]. Further studies should concentrate on finding the surfaces where doping is lower, by changing Au and Pt layer concentration and their different spatial positions, as well as the surfaces that feature polarized surface bands on the Fermi level.
Elucidating magnetic proximity effects in graphene heterostructures with metallic ferromagnets remains an important issue as recent experiments on bias-dependent reversal of magnetoresistance in vertical Co/Gr/NiFe spin valves support an overlooked role of graphene [@Asshoff2017:2DM]. Instead of ideally lattice-matched single-crystalline ferromagnet/Gr structures required for effective spin filtering [@Karpan2007:PRL], they confirm the formation of van der Waals heterostructures without such lattice matching where the proximitized graphene itself is a source of spin-polarized carriers [@Lazic2016:PRB; @Asshoff2017:2DM]. While in these experiments Co and NiFe were responsible for effective $n$- and $p$-doping of graphene [@Asshoff2017:2DM], our studies show that even with a single ferromagnet, both $n$- and $p$-doping of graphene is possible, which could enable different approaches for designing bias-dependent magnetoresistive effects [@Asshoff2017:2DM; @Ringer2018:APL; @Zhu2018:PRB; @Gurram2017:NC].
M.B. and I.L. acknowledge support from Croatian Science Foundation and the European Union through the European Regional Development Fund within the Competitiveness and Cohesion Operational Programme (Grant No. KK.01.1.1.06). K.D.B. is supported by NSF through Grant No. DMR-1609776 and the Nebraska Materials Research Science and Engineering Center (MRSEC) Grant No. DMR-1420645. I.Z. acknowledges support form US ONR N000141712793 and NSF ECCS-1810266, US ONR N000141712793, and UB Center for Computational Research.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The progenitors of hydrogen-poor core-collapse supernovae (SNe) of types Ib, Ic and IIb are believed to have shed their outer hydrogen envelopes either by extremely strong stellar winds, characteristic of classical Wolf-Rayet stars, or by binary interaction with a close companion star. The exact nature of the progenitors and the relative importance of these processes are still open questions. One relatively unexplored method to constrain the progenitors is to search for high-mass X-ray binaries (HMXB) at SN locations in pre-explosion X-ray observations. In a HMXB, one star has already exploded as a core-collapse SN, producing a neutron star or a stellar-mass black hole. It is likely that the second star in the system will also explode as a supernova, which should cause a detectable long-term change in the system’s X-ray luminosity. In particular, a pre-explosion detection of a HMXB coincident with a SN could be informative about the progenitor’s nature. In this paper we analyze pre-explosion ACIS observations of 18 nearby type Ib, Ic and IIb supernovae from the *Chandra* X-ray observatory public archive. Two sources that could potentially be associated with the supernova are identified in the sample. Additionally we make similar post-explosion measurements for 46 SNe. Although our modelling indicates that progenitor systems with compact binary companions are probably quite rare, studies of this type can in the future provide more stringent constraints as the number of discovered nearby SNe and suitable pre-explosion X-ray data are both increasing.'
author:
- |
T. Heikkil[ä]{}$^{1}$, S. Tsygankov$^{1}$, S. Mattila$^{1, 2, 3}$, J.J. Eldridge$^{4}$, M. Fraser$^{3}$, J. Poutanen$^{1}$\
$^{1}$Tuorla Observatory, Department of Physics and Astronomy, University of Turku, Väisäläntie 20, FI-21500 Piikkiö, Finland\
$^{2}$Finnish Centre for Astronomy with ESO (FINCA), University of Turku, Väisäläntie 20, FI-21500 Piikkiö, Finland\
$^{3}$Institute of Astronomy, University of Cambridge, Madingley Road, Cambridge CB3 0HA, UK\
$^{4}$Department of Physics, University of Auckland, Private Bag 92019, Auckland, New Zealand
bibliography:
- 'bibtex\_uptd.bib'
title: 'Progenitor constraints for core-collapse supernovae from *Chandra* X-ray observations'
---
\[firstpage\]
supernovae: general – supernovae: individual: SN 2004gt – stars: Wolf-Rayet – binaries: close – X-rays: binaries – stars: evolution.
Introduction {#sect_intro}
============
Supernovae are broadly divided into two classes based on the presence (type II) or absence (type I) of strong hydrogen features in their spectrum [e.g. @ref22]. SNe other than type Ia are believed to be the result of core-collapse explosions of massive stars at the end of their lifespan. Among hydrogen-poor core-collapse supernovae (CCSN), the lack of H-features in their spectra is attributed to these stars having lost their hydrogen-envelope prior to the explosion, and hence are collectively referred to as stripped-envelope CCSNe. These are supernova types Ib, Ic and also IIb (which, while they show H-features initially, rapidly lose them and begin to display He-features instead). There are believed to be two primary channels to produce stripped-envelope CCSNe, both of which are likely to be responsible for producing some fraction of these SNe. Previously, the preferred model was that the progenitors were single Wolf-Rayet stars with initial masses above $\sim$25M$_{\odot}$, characterised by high mass-loss rates, resulting in their envelopes being lost through radiatively driven winds [see e.g. @crowther for review]. However, recently the alternative mechanism for producing a stripped-envelope SN progenitor, binary interaction, has become the more favored explanation. In an interacting binary scenario, a progenitor that is not sufficiently massive to enter a WR-stage by itself could nonetheless lose its hydrogen envelope either by Roche-lobe overflow to a close companion star or via common-envelope evolution [@ref5]. In fact, based on the studies by e.g. @ref5, @vanbeveren_1998, @ref9 and @ref4, the high relative rate[^1] of type Ibc SNe could not be reconciled with a single-star scenario where stellar wind dominates the removal of the envelope, and that classical WR-stars could not account for more than half of the observed type Ibc SNe. Instead, the interacting binary scenario was found to be the more likely mechanism for most type Ib and IIb (and large fraction of Ic) SN progenitors. Furthermore, @lyman analyzed the ejecta-masses of 36 stripped-envelope CCSNe and found that all but three of them were inconsistent with massive progenitor stars, which further supports the likelihood of less massive binary progenitors and suggests this might even be the dominant progenitor mechanism for stripped-envelope CCSNe. It should also be noted that there is a distinct lack of pre-explosion high-mass SN progenitor star detections (with initial mass above $\sim$18M$_{\odot}$) [@newsmartt]. It has been suggested that this lack is caused by failed SNe where the core of the star collapses directly into a black hole without detectable SN [see e.g. @gerke_dsn; @reynolds_dsn]. This would reduce the population of massive stars available as potential single WR-star SN progenitors. For a more detailed discussion on the various progenitor models, see eg. the recent review by @yoon.
While a number of SN type II-P progenitor stars have been directly identified in optical pre-explosion observations [see @ref3 for review], only three type IIb progenitors have been identified: SN 2008ax [@crockett], SN 1993J [@ref20; @maund_smartt] and SN 2011dh [@ref21; @vandyk_2011dh], and only one type Ibc progenitor with the recent discovery of the progenitor candidate for the type Ib SN iPTF13bvn by [@ref23]. The progenitor of SN 2008ax is the only one of these where a single WR-star progenitor cannot be ruled out, although a binary system seems to be the more likely possibility (see discussion in [@ref3] and also [@crockett] and [@roming]). The progenitor of SN 1993J has been characterised by direct pre-explosion imaging to have been a red supergiant in a binary system. The progenitor of SN 2011dh was likewise identified by direct observation as a yellow supergiant in a probable binary system [@maund_smartt; @ref21; @folatelli]. The type Ib SN iPTF13bvn is also now believed to have had a binary progenitor, the single massive WR-star scenario having been ruled out by @ref24, @bersten, @ref25 and @nebular_phase. Studies of the observed properties of type IIb SNe have also been used to constrain their progenitors. For example, @ref16 found the progenitor of the type IIb SN 2011hs to be consistent with a supergiant star with an initial mass less than 20M$_{\odot}$ as also found for SNe 1993J and 2011dh. Based on radio-observations, @ref17 suggest the progenitor to the type IIb SN 2001ig was a WR-star in a binary, and the likely companion star has been identified as a late-B through late-F type supergiant [@ref19].
In X-ray frequencies, there has been some effort to identify or constrain the properties of stripped-envelope CCSN progenitors. For example, @ref7 and more recently @ref8 studied the shock-breakout of SN 2008D and found its progenitor to be consistent with a WR-star. In a case study for the type Ib SN 2010O in Arp299, @ref2 used archival pre-explosion *Chandra* data and claimed the detection of a variable X-ray source coincident with the supernova by comparing two observations taken at different times in order to estimate the flux. They suggested this might be indicative of the SN having a high-mass X-ray binary (HMXB) as a progenitor. A HMXB is a binary system consisting of a massive star which is losing mass to a compact companion, with accompanying X-ray emission. Similarly, @ref26 noted the coincidence of the type Ib SN 2009jf with the position of the ultraluminous X-ray source (ULX) CXOU J230453.0+121959, suggesting the possibility that the ULX may be a HMXB progenitor for the SN although they also note that a chance superposition is not unlikely. In this paper, we aim to extend the search for possible HMXB-progenitors for stripped-envelope CCSNe in X-rays. Nielsen et al. recently carried out two studies in which they measured the X-ray luminosity at the sites of 13 type Ia SNe, establishing upper limits for their progenitors [see @ref1; @ref37]. Here we make use of the overall methods of these two studies, but using a sample of supernovae of types Ib, Ic and IIb instead of Ia.
The type IIb SNe with pre-explosion progenitor detections (SN 1993J, SN 2008ax and SN 2011dh, as well as the SNe with indirect progenitor constraints, SN 2001ig and SN 2011hs) were also included in this study for comparison. For iPTF13bvn no suitable *Chandra* observations were found and it was therefore excluded. We mainly focus on pre-explosion observations. Based on observed SN X-ray lightcurves the X-ray luminosity of a CCSN can be expected to remain relatively high for many years after the explosion due to the expansion of the SN ejecta into the circumstellar material [see e.g. @ref15], which is the principal source of X-rays during this phase. However, we have also included a large sample of post-explosion results in order to present a more complete picture. Although the main intent is to search for the progenitor of the second supernova in a system that already contains a compact object (neutron star or a stellar-mass black hole), it is quite possible that some of the SNe studied here could also have been the initial SN in the close binary system that resulted in a formation of a HMXB. Some of the SNe in our sample already had previously published luminosities or upper limits derived from the same post-explosion data. In particular, @ref34 reported X-ray luminosities or their upper limits from post-explosion observations of 100 core-collapse SNe of all types from *Chandra*, *XMM-Newton* and *Swift* to study a link between X-ray luminosity and the rotation speeds of pulsars at birth. However, because these previously published values are typically measured using a different selection of observations, energy range, assumed spectral model or aperture, they are not directly comparable to the results presented in this study.
In this paper we use the following structure: In section \[sect\_obs\], we cover our sample selection and the observations used in this study. In section \[sect\_data\_analysis\], we explain the processes we used for our data analysis, describe the spectral models used, and explain how the results were obtained. In section \[sect\_results\], we examine the results in detail, consider the theoretical fraction of X-ray bright progenitor systems and present a prediction for the upper limits that can be obtained for future nearby SNe. Finally in section \[sect\_discussion\] we discuss some of the implications these results have for future studies of this type, followed by brief conclusions in section \[sect\_conclusions\].
Observations {#sect_obs}
============
*Chandra X-ray observatory*, launched in 1999, carries two instruments (HRC and ACIS), and has a spatial resolution of $\sim$0.5 arcsec, although the PSF depends highly on photon energy and distance from the optical axis. Only ACIS observations were included in this study. ACIS is sensitive in the 0.3-8 keV energy range. The precision of the absolute astrometry of *Chandra* data is sufficient for our study, with coordinate offsets being $\sim$1 arcsec or less in most instances[^2].
We selected from the Asiago supernova catalog [@asiago] all core-collapse supernovae (as of 2015 May 27) of types Ib, Ic and and IIb discovered after 1980 to ensure reasonably accurate astrometry. We then cross-correlated the known supernova positions with *Chandra* archival data to find observations with ACIS-data from either before or after the supernova explosion. We removed from the sample all SNe with distance more than 100 Mpc. Some SNe were also removed from the sample due to data quality (SN near or outside detector edge, SN more than 10 arcmin from optical axis). To confirm that we were not missing a number of supernovae without an IAU SN designation we also checked the ”Bright Supernova Archive”[^3] for objects between 2013-2015 fulfilling the above criteria. A total of 31 such SNe were found, none of which had suitable ACIS-data. This way, we found that suitable archival ACIS-data was available for approximately 18 percent of SNe within our search criteria. In total, 57 SNe were selected for the sample, of which 11 had only pre-explosion data available, 7 had both pre- and post-explosion data and the remaining 39 post-explosion data only. The full list of all supernovae included in the study and the *Chandra* observations used for each target are presented in Table \[obstable\], along with supernova optical maximum or discovery dates (from Asiago) and the combined exposure times for both pre- and post-explosion categories.
All SN coordinates from Asiago were compared to both the Unified Supernova Catalog (USC) [@lennarz] and the Sternberg Supernova Catalog (SSC) [@sternberg]. Whenever there were significant differences in the SN positions between Asiago and the other two catalogs[^4], we attempted to find a different source (see Tables \[blocktable\_pre\] and \[blocktable\_post\]) for the position. In all cases, adjustments were made if more accurate astrometry was available, even if the catalogs were in agreement.
In cases where multiple *Chandra* observations were found for a particular SN, all of the available data were first aligned. The data were then divided into pre- and post-explosion datasets for each SN. If any observation was made within several months of the SN date listed in Asiago catalog[^5] and there was a possibility of incorrectly assigning an observation to pre- or post-explosion dataset, we ensured the categorization was correct by searching for (or narrowing down) the actual SN explosion date wherever possible. All pre- or post-explosion observations were then combined together to obtain the best possible signal-to-noise (S/N) ratio. Some observations were excluded however, either because their relatively short exposure times of $<$ 5 ks would not have made any significant contribution to the total exposure time of the combined observation, or any data that were otherwise unsuitable, such as observations taken in *Chandra’s* continuous-clocking mode. The observations, whether combined or single, were then used to obtain the flux and luminosity at the position of the SN before and/or after the explosion, or an upper limit where no source was detected (as was the case for most SNe).
----------- --------------------------------------------------------------------------------- --------------------------- -------------- ---------------
Name *Chandra* observation ID$^\mathrm{A}$ Epoch$^\mathrm{B}$ Exposure Exposure
(pre-SN)(ks) (post-SN)(ks)
SN 1983I 859, 2148, 10403, 10404, 10775, 10776, 10777, 10778, 10779, 10780, 10781, 10782 1983 Apr 25 - 444.6
10801, 10824
SN 1983N 793, 2064, 12992, 12993, 12994, 12995, 12996, 13202, 13241, 13248, 14332, 14342 1983 July 15 - 791.3
SN 1983V 3554, 6868, 6869, 6870, 6871, 6872, 6873, 13920, 13921 1983 Dec 4 - 302.2
SN 1984L 7861 1984 Aug 20 - 5.1
SN 1985F 7147 1984 June 12 - 9.4
SN 1990U 10120, 11230 1990 Aug 2 - 35.4
SN 1991N 2939 1991 Apr 2 - 48.2
SN 1993J 735, 5600, 5601, 5935, 5936, 5937, 5938, 5939, 5940, 4941, 5942, 5943, 5944 1993 Mar 30 - 752.3
5945, 5946, 5947, 5948, 5949, 6174, 6346, 6347, 6892, 6893, 6894, 6895, 6896
6897, 6898, 6899, 6900, 6901, 9122, 12301
SN 1994I 354, 1622, 3932, 12562, 12668, 13812, 13813, 13814, 13815, 13816, 15496, 15553 1994 Apr 9 - 856.1
SN 1996D 15050 1996 Feb 9 - 16.0
SN 1996N 16350 1996 Mar 12 - 48.7
SN 1996aq 11229 1996 Aug 17 - 7.0
SN 1997X 4061 1997 Feb 1 - 10.8
SN 1998T 1641, 6227, 15077, 15619 1998 Mar 3 - 125.6
SN 1998bo 11505 1998 Apr 22 - 20.9
SN 1998bw 1956 1998 May 10 - 48.3
SN 1999dn 4800 1999 Aug 27 - 60.1
SN 1999eh 9104 1999 Oct 12 - 18.1
SN 1999ex 10392 1999 Nov 15 - 12.3
SN 2000cr 10395 2000 June 25 - 16.1
SN 2000ds 9528 2000 Oct 10 - 65.5
SN 2001ci 2038 2001 Apr 25$^\mathrm{C}$ 26.9 -
SN 2001ig 3495, 3496 2001 Dec 10 - 47.9
SN 2002hf 2244 2002 Oct 29 9.2 -
SN 2003L 4417 2003 Jan 12 - 30.4
SN 2003bg 3870, 3871, 7605 2003 Mar 19 - 127.3
SN 2003is 16577 2003 Oct 14 - 9.8
SN 2004C 4659, 4660, 7607 2004 Jan 12 - 139.0
SN 2004dk 11226 2004 Aug 20 - 8.1
SN 2004gn 4017 2004 Dec 1 5.0 -
SN 2004gt 315, 3040, 3041, 3042, 3043, 3044, 3718 2004 Dec 12 425.2 -
SN 2005U 1641, 6227, 15077, 15619 2005 Jan 30 24.9 100.7
SN 2005at 15384 2005 Mar 5 - 52.8
SN 2005cz 6785 2005 July 17 - 15.2
SN 2006ep 7608, 9098 2006 Aug 30 - 8.0
SN 2007Y 15392, 16487 2007 Mar 3 - 49.7
SN 2007gr 387, 9579 2007 Aug 28 2.4 19.7
SN 2007ke 908, 11717, 12016, 12017, 12018 2007 Sept 24 48.5 135.5
SN 2007kj 6978, 8491 2007 Oct 2 46.3 -
SN 2008D 9104 2008 Jan 28$^\mathrm{D}$ - 18.1
SN 2008ax 1579, 4725, 4726 2008 Mar 22 98.9 -
SN 2008bo 9105, 10118 2008 Mar 31 - 19.4
SN 2009bb 10140 2009 Mar 21 - 10.0
SN 2009dt 10265 2009 Apr 28 5.1 -
SN 2009jf 10120, 11230 2009 Sept 27$^\mathrm{E}$ 25.2 10.2
SN 2009mk 11236 2009 Dec 15 - 10.0
SN 2010O 1641, 6227, 15077, 15619 2010 Jan 24 35.2 90.4
SN 2011dh 354, 1622, 3932, 12562, 12668, 13812, 13813, 13814, 13815, 13816, 15496, 15553 2011 June 1 90.8 765.3
SN 2011ei 12669 2011 July 25 - 10.0
SN 2011hs 3947 2011 Nov 12 55.7 -
SN 2012ap 13785 2012 Feb 18 - 9.9
SN 2013ak 14795 2013 Mar 9 - 9.9
SN 2013dk 315, 3040, 3041, 3042, 3043, 3044, 3718 2013 June 22 425.2 -
SN 2013ff 11776 2013 Aug 31 30.1 -
SN 2013ge 16556 2013 Nov 8 - 18.8
SN 2014C 2198, 17569, 17570 2014 Jan 5 30.1 19.8
SN 2014L 7863 2014 Jan 26 5.1 -
----------- --------------------------------------------------------------------------------- --------------------------- -------------- ---------------
\[A\] Only those *Chandra* OBSIDs are listed which were used for the measurements. \[B\] From Asiago Max. epoch: The date of optical maximum or discovery. \[C\] Explosion epoch narrowed down from @ref40 to ensure the observation is correctly categorized. \[D\] Explosion epoch verified from @ref6 to ensure the observation is correctly categorized. \[E\] Explosion epoch verified from @ref10 to ensure the observations are correctly categorized.
Ref.no Description Model Parameters $\langle E_{\gamma} \rangle$ (keV)$^\mathrm{A}$ $C_\mathrm{abs}$ $^\mathrm{B}$
-------- ----------------------------------- -------------------- --------------------------------------------------------- ------------------------------------------------- --------------------------------
1 Soft-state black-hole (low abs.) Absorbed blackbody $T=1$ keV, $N_\mathrm{H}=0.5\times10^{22}$ cm$^{-2}$ 2.48 1.15
2 Soft-state black-hole (high abs.) Absorbed blackbody $T=1$ keV, $N_\mathrm{H}=2.0\times10^{22}$ cm$^{-2}$ 3.10 1.47
3 Accreting pulsar (low abs.) Absorbed powerlaw $\Gamma=0.5$, $N_\mathrm{H}=0.5\times10^{22}$ cm$^{-2}$ 2.81 1.11
4 Accreting pulsar (high abs.) Absorbed powerlaw $\Gamma=0.5$, $N_\mathrm{H}=2.0\times10^{22}$ cm$^{-2}$ 3.64 1.30
5 Hard-state black-hole (low abs.) Absorbed powerlaw $\Gamma=1.5$, $N_\mathrm{H}=0.5\times10^{22}$ cm$^{-2}$ 2.00 1.35
6 Hard-state black-hole (high abs.) Absorbed powerlaw $\Gamma=1.5$, $N_\mathrm{H}=2.0\times10^{22}$ cm$^{-2}$ 3.02 1.80
\[A\] Average photon energy for the model. \[B\] Correction coefficient for absorption.
Data analysis {#sect_data_analysis}
=============
We calculated luminosity for three simple spectral models approximating common properties of high-mass X-ray binaries [@ref27; @ref11]: a blackbody at the temperature of 1 keV (a black hole in soft state; accretion disk), a powerlaw with photon index $\Gamma=$0.5 (accreting pulsar with high magnetic field), and powerlaw with photon index $\Gamma=$1.5 (a black hole in hard state). All three models were further combined with a model of interstellar absorption (XSPEC phabs[^6]) with two different absorption values (for a total of six models considered): $0.5\times10^{22}$ cm$^{-2}$ and $2.0\times10^{22}$ cm$^{-2}$. The models and their parameters are also listed in Table \[models\]. For these models, the main source of absorption is assumed to be the circumstellar material around the progenitor. Assuming a conservative average host galaxy extinction for CCSNe of $A_V\sim$1 [@dist11] and converting this into a hydrogen column density according to the formula $N_\mathrm{H}=1.79\times10^{21}A_V$ [@predehl] results in a column density of $0.2\times10^{22}$ cm$^{-2}$. This is significantly lower than the absorption values used in our models and we therefore do not consider the effects of host galaxy interstellar gas separately. The one exception to this is SN 2001ci, which has an unusually high $A_V$ of $\sim 5-6$ [@ref40] corresponding to $N_\mathrm{H}\sim1.0\times10^{22}$ cm$^{-2}$. We do not account for this higher extinction separately, but note that only the higher-absorption models are realistic for this particular SN.
The data were processed using *Chandra’s* CIAO 4.6 software package. In most cases, only a single observation existed of the given target. For these, prior to analysis, all of the data were reprocessed with the $chandra\_repro$ script to apply standard corrections (using CALDB version 4.6.1.1). In cases where multiple observations of the same area were available, these were first aligned with each other. In most cases, the relative astrometry of the original data was already sufficiently accurate that after reprocessing with $chandra\_repro$, the data could simply be aligned with the $reproject\_obs$ tool. However, in some cases, it was beneficial to apply a correction into the aspect solution of the individual observations to ensure their coordinate grids were better aligned prior to reprojecting their tangent planes. In such cases, a wavelet analysis (with CIAO’s $wavdetect$ tool) was first applied to detect a number of sources in each observation. These sources were then matched on the individual observations (using CIAO’s $reproject\_aspect$ script) to produce an aspect correction, which was then applied by reprocessing the data with $chandra\_repro$ before finishing the alignment process with $reproject\_obs$.
After alignment and reprocessing, all datasets were divided into pre-explosion and post-explosion data for each supernova. Combined photon-count maps and exposure maps were then made using CIAO’s $flux\_obs$ tool for each SN in both categories. The exposure maps are in units of cm$^2$ s and combine instrument quantum efficiency with the exposure time as the telescope is dithered across the sky. They were calculated using spectral weights (produced with $make\_instmap\_weights$) for each of the six models. The energy range used for all models and maps was 0.5-7.0 keV.
For most targets, the flux was measured using a circular aperture with a radius of 2.5 arcsec at the known position of the SN (which contains more than 90 percent of *Chandra’s* PSF for sources less than 2 arcmin from the optical axis, see http://cxc.harvard.edu/proposer/POG/html/chap4.html, Fig.4.13). In cases where the optical axis of one or more of the original individual observations was further than 2 arcmin from the SN position, the source aperture radius was increased accordingly (in increments of 0.5 arcsec) so that it contained at least 90 percent of the PSF. Background was measured using an annulus around the source aperture circle, the size of which varied according to target. Generally the outer radius of the annulus was increased either to maintain roughly 2/5 ratio between the source aperture radius and the background annulus outer radius, or increased even further to ensure that at least 10 photons were located within the background area. For a list of the used source aperture and background annulus sizes, see Tables \[blocktable\_pre\] and \[blocktable\_post\]. Any unrelated sources were excluded from both the source apertures and the background annuli. For the purposes of this study, we conservatively consider a source unrelated if it is located further than $\sim$1.5 arcsec from the SN position or is not a point-source. For most pre-explosion targets there were no point-sources closer than 5 arcsec to the SN position. In all but two of the cases where more nearby point-sources existed, the position of the SN was known from literature with sufficient accuracy to rule out any association between such a source and the SN. The two exceptions, SN 2009jf and SN 2004gt, will be discussed in more detail in section \[sect\_detected\].
Background corrected flux was measured for each source aperture. If no source was found (within $3\sigma$), then a $3\sigma$ upper limit was calculated instead using the number of detected counts inside the source aperture (equation \[flux\_ulim\] below). However, in many cases a short exposure time would result in the number of photons inside the aperture itself to be very low (or even zero), so in cases where there were no visible sources and the number of counts inside the source aperture was $<20$, we used Poissonian statistics to find the matching upper limit at a 99.87 percent confidence level [corresponding to Gaussian $3\sigma$, see e.g. @gehrels], and calculated the $3\sigma$ upper limit using equation \[flux\_poisson\] below instead for these cases. To obtain unabsorbed flux, we calculated an absorption correction coefficient for each model by comparing the absorbed models with the corresponding unabsorbed model in XSPEC. The coefficients are listed in table \[models\]. Following @ref1, for any $3\sigma$ detection we calculated the flux from
$$\label{flux}
F= \frac{(n-b) \langle E_{\gamma} \rangle}{\zeta}$$
where $n$ is the number of photons from the source aperture, $\langle E_{\gamma} \rangle$ is average photon energy for the given model (see table \[models\]), $\zeta$ is the average value of the exposure map within the source aperture and $b$ is scaled background, the number of background counts $n_{\mathrm{bg}}$ scaled to the size of the source aperture: $$\label{bgscaled}
b= \frac{n_{\mathrm{bg}} \times A_{\mathrm{src}}}{A_{\mathrm{bg}}},$$ where $A_{\mathrm{src}}$ and $A_\mathrm{bg}$ are the surface areas of the source and background regions respectively. The flux upper limit was calculated from $$\label{flux_ulim}
F_\mathrm{UL}= \frac{3\sqrt{n}\langle E_{\gamma} \rangle}{\zeta}$$ for high photon count cases. For the low photon count cases, the flux upper limit was instead calculated from $$\label{flux_poisson}
F_\mathrm{UL}=\frac{(\mu -b)\langle E_{\gamma} \rangle}{\zeta},$$ where $\mu$ is the 0.9987 confidence-level Poissonian upper limit corresponding to the number of photons within the source aperture, according to @gehrels. Finally, the unabsorbed luminosity for each $F$ and $F_\mathrm{UL}$ was calculated from $$\label{lum}
L= 4 \pi C_\mathrm{abs} F d^{2}$$
where $C_\mathrm{abs}$ is the correction coefficient for absorption and $d$ is the distance to the host galaxy. The distances used for calculating the luminosities were obtained via NED[^7] (see Tables \[blocktable\_pre\] and \[blocktable\_post\] for details), or calculated from the Hyperleda[^8] radial velocity corrected for infall on to the Virgo cluster (vvir) using $H_0=70$ km s$^{-1}$Mpc$^{-1}$.
Results {#sect_results}
=======
Basic information about each SN (host galaxy, SN type, position, distance), as well as non-model-specific results of our measurements are presented in Table \[blocktable\_pre\] for those SNe with pre-explosion data, and in Table \[blocktable\_post\] for those with post-explosion data. Targets with both types of data are present in both. The model-specific results – average exposure map values and luminosity upper limits – are presented in Table \[resultstable\_pre1\] and Tables \[resultstable\_post1\] and \[resultstable\_post2\] for pre- and post-explosion data respectively. Any measured luminosities for sources that have a flux above the 3$\sigma$ detection threshold and their errors are also included in these tables. The models are numbered according to Table \[models\].
----------- --------------- ------ ---------------------------------------- ----------------------- ----------------------- ------------ ------------
Name Host SN Position$^\mathrm{A}$ Distance$^\mathrm{B}$ Aperture$^\mathrm{C}$ Counts Scaled
Galaxy Type (Mpc) (Aperture) background
SN 2001ci NGC 3079 Ic 10:01:57.21 +55:41:14.0$^\mathrm{K}$ 18.1$^\mathrm{D}$ 2.5“/6” 4 4.1
SN 2002hf MCG-05-03-020 Ic 00:57:47.74 $-$27:30:21.5$^\mathrm{U}$ 77.5 3.5“/9” 10 9.3
SN 2004gn NGC 4527 Ic 12:34:12.10 +02:39:34.4$^\mathrm{U}$ 14.2$^\mathrm{E}$ 2.5“/25” 0 0.2
SN 2004gt NGC 4038 Ic 12:01:50.42 $-$18:52:13.5$^\mathrm{M}$ 20.4$^\mathrm{F}$ 2.5“/12” 694 206.4
SN 2005U Arp 299 IIb 11:28:33.13 +58:33:41.3$^\mathrm{L}$ 48.2 2.5“/6” 15 11.1
SN 2007gr NGC 1058 Ic 02:43:27.98 +37:20:44.7$^\mathrm{U}$ 9.3$^\mathrm{G}$ 2.5“/25” 0 0.1
SN 2007ke NGC 1129 Ib 02:54:23.90 +41:34:16.3$^\mathrm{U}$ 77.2 6.5“/16” 430 467.5
SN 2007kj NGC 7803 Ib 00:01:19.58 +13:06:30.6$^\mathrm{U}$ 77.0 2.5“/8” 3 1.5
SN 2008ax NGC 4490 IIb 12:30:40.80 +41:38:14.5$^\mathrm{U}$ 9.6$^\mathrm{H}$ 2.5“/7” 8 8.2
SN 2009dt IC 5169 Ic 22:10:09.27 $-$36:05:42.6$^\mathrm{U}$ 42.0 2.5“/20” 0 0.2
SN 2009jf NGC 7479 Ib 23:04:52.98 +12:19:59.5$^\mathrm{U}$ 34.9 3“/8” 105 1.7
SN 2010O Arp 299 Ib 11:28:33.86 +58:33:51.6$^\mathrm{U}$ 48.2 2.5“/6” 239 67.3
SN 2011dh NGC 5194 IIb 13:30:05.11 +47:10:10.9 7.8$^\mathrm{I}$ 3“/8” 3 8.7
SN 2011hs IC 5267 IIb 22:57:11.77 $-$43:23:04.8 21.8 2.5“/10” 1 1.4
SN 2013dk NGC 4038 Ic 12:01:52.72 $-$18:52:18.3 20.4$^\mathrm{F}$ 2.5“/6” 202 172.7
SN 2013ff NGC 2748 Ic 09:13:38.88 +76:28:10.8 24.8 2.5“/8” 1 2.6
SN 2014C NGC 7331 Ib 22:37:05.60 +34:24:31.9 15.1$^\mathrm{E}$ 2.5“/6” 5 5.8
SN 2014L NGC 4254 Ic 12:18:48.68 +14:24:43.5 17.3$^\mathrm{J}$ 2.5“/10” 1 1.1
----------- --------------- ------ ---------------------------------------- ----------------------- ----------------------- ------------ ------------
\[A\] From Asiago if not indicated otherwise. \[B\] Distances that do not have a reference have been calculated from Hyperleda radial velocity corrected for the infall to the Virgo cluster (vvir) with $H_0 = 70$ km s$^{-1}$Mpc$^{-1}$. \[C\] Radius of source aperture/outer radius of background annulus. \[D\] Average of values from @dist7 [@dist16] \[E\] @dist4 \[F\] @dist14 \[G\] @dist3 \[H\] @dist6 \[I\] @dist2 \[J\] Average of values from @dist15 [@dist16; @dist10] \[K\] Position confirmed from @vandykpos. \[L\] Position from @2005U_pos_Li (Li position) \[M\] Position derived in this paper, see section \[sect\_detected\]. \[U\] Position confirmed from Unified Supernova Catalog [@lennarz]
----------- -------------- ------ ---------------------------------------- ----------------------- ----------------------- ------------ ------------
Name Host SN Position$^\mathrm{A}$ Distance$^\mathrm{B}$ Aperture$^\mathrm{C}$ Counts Scaled
Galaxy Type (Mpc) (Aperture) background
SN 1983I NGC 4051 Ic 12:03:11.77 +44:31:00.6 13.7$^\mathrm{D}$ 2.5“/8” 17 17.5
SN 1983N NGC 5236 Ib 13:36:51.24 $-$29:54:02.7$^\mathrm{U}$ 4.6$^\mathrm{E}$ 5“/13” 301 264.7
SN 1983V NGC 1365 Ic 03:33:31.63 $-$36:08:55.0$^\mathrm{U}$ 19.6$^\mathrm{E}$ 2.5“/6” 16 12.3
SN 1984L NGC 991 Ib 02:35:30.52 $-$07:09:30.5 20.4 2.5“/20” 0 0.2
SN 1985F NGC 4618 Ib 12:41:33.01 +41:09:05.9$^\mathrm{U}$ 7.9$^\mathrm{F}$ 2.5“/15” 1 0.4
SN 1990U NGC 7479 Ib 23:04:54.92 +12:18:20.1$^\mathrm{U}$ 34.9 2.5“/8” 3 3.2
SN 1991N NGC 3310 Ic 10:38:46.37 +53:30:04.7$^\mathrm{U}$ 17.5$^\mathrm{G}$ 2.5“/6” 162 117.5
SN 1993J NGC 3031 IIb 09:55:24.77 +69:01:13.7$^\mathrm{O}$ 3.6$^\mathrm{H}$ 3“/8” 9498 60.2
SN 1994I NGC 5194 Ic 13:29:54.12 +47:11:30.4$^\mathrm{P}$ 7.8$^\mathrm{I}$ 3“/8” 362 200.9
SN 1996D NGC 1614 Ic 04:34:00.30 $-$08:34:44.0 66.4 2.5“/6” 27 15.3
SN 1996N NGC 1398 Ib 03:38:55.31 $-$26:20:04.1$^\mathrm{U}$ 17.2 2.5“/10” 0 1.0
SN 1996aq NGC 5584 Ic 14:22:22.72 $-$00:23:23.8$^\mathrm{U}$ 22.0$^\mathrm{J}$ 2.5“/20” 1 0.3
SN 1997X NGC 4691 Ib 12:48:14.28 $-$03:19:58.5 16.4 2.5“/6” 1 7.1
SN 1998T Arp 299 Ib 11:28:33.16 +58:33:43.7$^\mathrm{U}$ 48.2 2.5“/6” 603 216.1
SN 1998bo ESO185-031 Ic 19:57:22.55 $-$55:08:18.4$^\mathrm{U}$ 67.4 10“/30” 12 9.8
SN 1998bw E184-G82 Ic 19:35:03.31 $-$52:50:44.6$^\mathrm{Q}$ 33.8 2.5“/6” 83 4.3
SN 1999dn NGC 7714 Ib 23:36:14.81 +02:09:08.4$^\mathrm{R}$ 40.0 2.5“/6” 8 7.7
SN 1999eh NGC 2770 Ib 09:09:32.67 +33:07:16.9$^\mathrm{U}$ 29.5 2.5“/10” 1 0.7
SN 1999ex IC 5179 Ib 22:16:07.27 $-$36:50:53.7$^\mathrm{U}$ 46.3$^\mathrm{K}$ 2.5“/12” 3 1.0
SN 2000cr NGC 5395 Ic 13:58:38.37 +37:26:12.9$^\mathrm{U}$ 52.7 2.5“/8” 0 1.7
SN 2000ds NGC 2768 Ib 09:11:36.28 +60:01:43.3$^\mathrm{V}$ 23.2 2.5“/6” 8 4.1
SN 2001ig NGC 7424 IIb 22:57:30.69 $-$41:02:25.9$^\mathrm{U}$ 10.8$^\mathrm{L}$ 2.5“/7” 38 1.1
SN 2003L NGC 3506 Ic 11:03:12.33 +11:04:38.3$^\mathrm{U}$ 92.2 2.5“/6” 43 6.2
SN 2003bg M-05-10-15 Ic 04:10:59.42 $-$31:24:50.3$^\mathrm{U}$ 16.3 2.5“/7” 632 7.1
SN 2003is MCG+07-40-03 Ic 19:21:08.00 +43:19:35.4$^\mathrm{U}$ 81.9 5“/22” 0 0.9
SN 2004C NGC 3683 Ic 11:27:29.72 +56:52:48.2$^\mathrm{U}$ 27.8 2.5“/6” 136 30.3
SN 2004dk NGC 6118 Ib 16:21:48.93 $-$02:16:17.3$^\mathrm{U}$ 23.5 2.5“/12” 7 0.6
SN 2005U Arp 299 IIb 11:28:33.13 +58:33:41.3$^\mathrm{S}$ 48.2 2.5“/6” 90 53.4
SN 2005at NGC 6744 Ic 19:09:53.57 $-$63:49:22.8$^\mathrm{U}$ 9.0 2.5“/10” 1 0.8
SN 2005cz NGC 4589 Ib 12:37:27.85 +74:11:24.5$^\mathrm{U}$ 32.6 2.5“/8” 0 1.6
SN 2006ep NGC 0214 Ib 00:41:24.88 +25:29:46.7$^\mathrm{U}$ 66.1 2.5“/15” 0 0.3
SN 2007Y NGC 1187 Ib 03:02:35.92 $-$22:53:50.1$^\mathrm{U}$ 17.3 3.5“/10” 4 1.7
SN 2007gr NGC 1058 Ic 02:43:27.98 +37:20:44.7$^\mathrm{U}$ 9.3$^\mathrm{M}$ 2.5“/25” 3 0.5
SN 2007ke NGC 1129 Ib 02:54:23.90 +41:34:16.3$^\mathrm{U}$ 77.2 6.5“/16” 1182 1215.6
SN 2008D NGC 2770 Ib 09:09:30.65 +33:08:20.3$^\mathrm{U}$ 29.5 2.5“/10” 11 0.7
SN 2008bo NGC 6643 IIb 18:19:54.41 +74:34:21.0 25.4 2.5“/8” 1 0.8
SN 2009bb NGC 3278 Ic 10:31:33.88 $-$39:57:30.0$^\mathrm{T}$ 40.0 2.5“/8” 10 1.7
SN 2009jf NGC 7479 Ib 23:04:52.98 +12:19:59.5$^\mathrm{U}$ 34.9 3“/8” 76 1.5
SN 2009mk E293-G34 IIb 00:06:21.37 $-$41:28:59.8$^\mathrm{U}$ 18.7 2.5“/15” 2 0.5
SN 2010O Arp 299 Ib 11:28:33.86 +58:33:51.6$^\mathrm{U}$ 48.2 2.5“/6” 759 244.3
SN 2011dh NGC 5194 IIb 13:30:05.11 +47:10:10.9 7.8$^\mathrm{I}$ 3“/8” 1193 49.1
SN 2011ei NGC 6925 IIb 20:34:22.62 $-$31:58:23.6 38.3 2.5“/8” 1 1.3
SN 2012ap NGC 1729 Ic 05:00:13.72 $-$03:20:51.2 50.5 2.5“/20” 2 0.3
SN 2013ak E430-G20 IIb 08:07:06.69 $-$28:03:10.1 11.6$^\mathrm{N}$ 2.5“/15” 30 0.4
SN 2013ge NGC 3287 Ic 10:34:48.46 +21:39:41.9 19.7 2.5“/12” 0 0.5
SN 2014C NGC 7331 Ib 22:37:05.60 +34:24:31.9 15.1 2.5“/6” 515 6.0
----------- -------------- ------ ---------------------------------------- ----------------------- ----------------------- ------------ ------------
\[A\] From Asiago if not indicated otherwise. \[B\] Distances that do not have a reference have been calculated from Hyperleda radial velocity corrected for the infall to the Virgo cluster (vvir) with $H_0 = 70$ km s$^{-1}$Mpc$^{-1}$. \[C\] Radius of source aperture/outer radius of background annulus. \[D\] Average of values from @dist7 [@dist8; @dist9; @dist10]. \[E\] @dist4 \[F\] @dist5 \[G\] @dist12 \[H\] @dist1 \[I\] @dist2 \[J\] @dist14 \[K\] Average of values from @dist17 [@dist18; @dist19; @dist20; @dist21; @dist22; @dist23; @dist24; @dist25; @dist26; @dist27]. \[L\] @dist11 \[M\] @dist3 \[N\] Average value from @dist13. \[O\] Position from @pos_1993J. \[P\] Position from @pos_1994I. \[Q\] Position from @pos_1998bw. \[R\] Position confirmed from @vandykpos. \[S\] Position from @2005U_pos_Li \[T\] Position confirmed from @pos_2009bb. \[U\] Position confirmed from Unified Supernova Catalog [@lennarz]. \[V\] Position obtained from Unified Supernova Catalog [@lennarz].
--------------------- ------------------ --------------------------------- ------------------ --------------------------------- ------------------ --------------------------------- -- -- -- -- -- --
Name Exp. map av. Luminosity Exp. map av. Luminosity Exp. map av. Luminosity
value (cm$^2$s) (erg s$^{-1}$) value (cm$^2$s) (erg s$^{-1}$) value (cm$^2$s) (erg s$^{-1}$)
2001ci$^\mathrm{A}$ 1.17$\times10^7$ $<$ 1.6$\times10^{38}$ 1.06$\times10^7$ $<$ 2.8$\times10^{38}$ 1.04$\times10^7$ $<$ 1.9$\times10^{38 }$
2002hf 3.83$\times10^6$ $<$ 1.2$\times10^{40}$ 3.48$\times10^6$ $<$ 2.2$\times10^{40}$ 3.41$\times10^6$ $<$ 1.5$\times10^{40}$
2004gn 2.12$\times10^6$ $<$ 3.3$\times10^{38}$ 1.93$\times10^6$ $<$ 5.8$\times10^{38}$ 1.89$\times10^6$ $<$ 4.1$\times10^{38}$
2004gt$^\mathrm{B}$ 1.80$\times10^8$ $($6.1$\pm$0.3$)\times10^{38}$ 1.64$\times10^8$ $($10.8$\pm$0.6$)\times10^{38}$ 1.61$\times10^8$ $($7.6$\pm$0.4$)\times10^{38}$
2005U 7.25$\times10^6$ $<$ 3.4$\times10^{39}$ 6.86$\times10^6$ $<$ 5.8$\times10^{39}$ 6.72$\times10^6$ $<$ 4.1$\times10^{39}$
2007gr 9.29$\times10^5$ $<$ 3.3$\times10^{38}$ 8.32$\times10^5$ $<$ 5.9$\times10^{38}$ 8.24$\times10^5$ $<$ 4.1$\times10^{38}$
2007ke 1.57$\times10^7$ $<$ 1.3$\times10^{40}$ 1.48$\times10^7$ $<$ 2.2$\times10^{40}$ 1.44$\times10^7$ $<$ 1.5$\times10^{40}$
2007kj 1.37$\times10^7$ $<$ 2.6$\times10^{39}$ 1.31$\times10^7$ $<$ 4.4$\times10^{39}$ 1.27$\times10^7$ $<$ 3.1$\times10^{39}$
2008ax 4.15$\times10^7$ $<$ 1.5$\times10^{37}$ 3.78$\times10^7$ $<$ 2.7$\times10^{37}$ 3.69$\times10^7$ $<$ 1.9$\times10^{37}$
2009dt 1.76$\times10^6$ $<$ 3.5$\times10^{39}$ 1.69$\times10^6$ $<$ 5.9$\times10^{39}$ 1.63$\times10^6$ $<$ 4.2$\times10^{39}$
2009jf 1.03$\times10^7$ $($6.7$\pm$0.7$)\times10^{39}$ 9.40$\times10^6$ $($1.2$\pm$0.1$)\times10^{40}$ 9.12$\times10^6$ $($8.3$\pm$0.8$)\times10^{39}$
2010O$^\mathrm{C}$ 1.09$\times10^7$ $($2.0$\pm$0.2$)\times10^{40}$ 1.01$\times10^7$ $($3.4$\pm$0.3$)\times10^{40}$ 9.90$\times10^6$ $($2.4$\pm$0.2$)\times10^{40}$
2011dh 3.78$\times10^7$ $<$ 3.5$\times10^{36}$ 3.44$\times10^7$ $<$ 6.1$\times10^{36}$ 3.37$\times10^7$ $<$ 4.3$\times10^{36}$
2011hs 2.27$\times10^7$ $<$ 8.6$\times10^{37}$ 2.07$\times10^7$ $<$ 1.5$\times10^{38}$ 2.02$\times10^7$ $<$ 1.1$\times10^{38}$
2013dk 1.79$\times10^8$ $<$ 5.4$\times10^{37}$ 1.62$\times10^8$ $<$ 9.6$\times10^{37}$ 1.59$\times10^8$ $<$ 6.7$\times10^{37}$
2013ff 1.24$\times10^7$ $<$ 1.7$\times10^{38}$ 1.14$\times10^7$ $<$ 3.0$\times10^{38}$ 1.11$\times10^7$ $<$ 2.1$\times10^{38}$
2014C 1.30$\times10^7$ $<$ 9.8$\times10^{37}$ 1.18$\times10^7$ $<$ 1.7$\times10^{38}$ 1.16$\times10^7$ $<$ 1.2$\times10^{38}$
2014L 2.06$\times10^6$ $<$ 6.2$\times10^{38}$ 1.89$\times10^6$ $<$ 1.1$\times10^{39}$ 1.84$\times10^6$ $<$ 7.6$\times10^{38}$
Name Exp. map av. Luminosity Exp. map av. Luminosity Exp. map av. Luminosity
value (cm$^2$s) (erg s$^{-1}$) value (cm$^2$s) (erg s$^{-1}$) value (cm$^2$s) (erg s$^{-1}$)
2001ci$^\mathrm{A}$ 9.23$\times10^6$ $<$ 3.3$\times10^{38}$ 1.23$\times10^7$ $<$ 1.4$\times10^{38}$ 1.05$\times10^7$ $<$ 3.4$\times10^{38}$
2002hf 3.04$\times10^6$ $<$ 2.6$\times10^{40}$ 3.99$\times10^6$ $<$ 1.1$\times10^{40}$ 3.44$\times10^6$ $<$ 2.6$\times10^{40}$
2004gn 1.69$\times10^6$ $<$ 6.9$\times10^{38}$ 2.20$\times10^6$ $<$ 3.0$\times10^{38}$ 1.91$\times10^6$ $<$ 7.0$\times10^{38}$
2004gt$^\mathrm{B}$ 1.43$\times10^8$ $($12.9$\pm$0.7$)\times10^{38}$ 1.89$\times10^8$ $($5.6$\pm$0.3$)\times10^{38}$ 1.62$\times10^8$ $($13.1$\pm$0.7$)\times10^{38}$
2005U 6.30$\times10^6$ $<$ 6.6$\times10^{39}$ 7.32$\times10^6$ $<$ 3.2$\times10^{39}$ 6.78$\times10^6$ $<$ 7.0$\times10^{39}$
2007gr 7.24$\times10^5$ $<$ 7.0$\times10^{38}$ 9.86$\times10^5$ $<$ 2.9$\times10^{38}$ 8.25$\times10^5$ $<$ 7.1$\times10^{38}$
2007ke 1.34$\times10^7$ $<$ 2.5$\times10^{40}$ 1.59$\times10^7$ $<$ 1.2$\times10^{40}$ 1.46$\times10^7$ $<$ 2.7$\times10^{40}$
2007kj 1.21$\times10^7$ $<$ 5.0$\times10^{39}$ 1.36$\times10^7$ $<$ 2.5$\times10^{39}$ 1.29$\times10^7$ $<$ 5.3$\times10^{39}$
2008ax 3.30$\times10^7$ $<$ 3.2$\times10^{37}$ 4.30$\times10^7$ $<$ 1.4$\times10^{37}$ 3.73$\times10^7$ $<$ 3.2$\times10^{37}$
2009dt 1.55$\times10^6$ $<$ 6.6$\times10^{39}$ 1.74$\times10^6$ $<$ 3.4$\times10^{39}$ 1.66$\times10^6$ $<$ 7.1$\times10^{39}$
2009jf 8.20$\times10^6$ $($1.4$\pm$0.1$)\times10^{40}$ 1.05$\times10^7$ $($6.2$\pm$0.6$)\times10^{39}$ 9.27$\times10^6$ $($1.4$\pm$0.1$)\times10^{40}$
2010O$^\mathrm{C}$ 9.13$\times10^6$ $($4.0$\pm$0.4$)\times10^{40}$ 1.11$\times10^7$ $($1.9$\pm$0.2$)\times10^{40}$ 1.00$\times10^7$ $($4.2$\pm$0.4$)\times10^{40}$
2011dh 3.01$\times10^7$ $<$ 7.2$\times10^{36}$ 3.93$\times10^7$ $<$ 3.2$\times10^{36}$ 3.40$\times10^7$ $<$ 7.4$\times10^{36}$
2011hs 1.81$\times10^7$ $<$ 1.8$\times10^{38}$ 2.35$\times10^7$ $<$ 7.9$\times10^{37}$ 2.04$\times10^7$ $<$ 1.8$\times10^{38}$
2013dk 1.42$\times10^8$ $<$ 1.1$\times10^{38}$ 1.87$\times10^8$ $<$ 4.9$\times10^{37}$ 1.60$\times10^8$ $<$ 1.2$\times10^{38}$
2013ff 1.00$\times10^7$ $<$ 3.5$\times10^{38}$ 1.27$\times10^7$ $<$ 1.6$\times10^{38}$ 1.13$\times10^7$ $<$ 3.6$\times10^{38}$
2014C 1.03$\times10^7$ $<$ 2.1$\times10^{38}$ 1.37$\times10^7$ $<$ 8.9$\times10^{37}$ 1.17$\times10^7$ $<$ 2.1$\times10^{38}$
2014L 1.66$\times10^6$ $<$ 1.3$\times10^{39}$ 2.12$\times10^6$ $<$ 5.7$\times10^{38}$ 1.87$\times10^6$ $<$ 1.3$\times10^{39}$
--------------------- ------------------ --------------------------------- ------------------ --------------------------------- ------------------ --------------------------------- -- -- -- -- -- --
\[A\] SN 2001ci has an unusually high host galaxy absorption which has not been corrected for here. Therefore, only the higher-absorption models (2, 4 and 6) should be considered realistic. \[B\] The luminosities presented here for SN 2004gt are the source aperture luminosities for the original 0.5-7.0 keV energy range and do not reflect the properties of the source discussed in section \[sect\_detected\]. \[C\] The luminosities presented here for SN 2010O are a combination of multiple different sources found inside the source aperture and do not represent any particular single source, see section \[sect\_detected\].
-------- ------------------ -------------------------------------------- ------------------ ------------------------------------------- ------------------ --------------------------------------------- -- -- -- -- -- --
Name Exp. map av. Luminosity Exp. map av. Luminosity Exp. map av. Luminosity
value (cm$^2$s) (erg s$^{-1}$) value (cm$^2$s) (erg s$^{-1}$) value (cm$^2$s) (erg s$^{-1}$)
1983I 5.69$\times10^7$ $<$ 2.9$\times10^{37}$ 5.91$\times10^7$ $<$ 4.5$\times10^{37}$ 5.67$\times10^7$ $<$ 3.2$\times10^{37}$
1983N 3.08$\times10^8$ $<$ 2.0$\times10^{36}$ 2.83$\times10^8$ $<$ 3.4$\times10^{36}$ 2.73$\times10^8$ $<$ 2.4$\times10^{36}$
1983V 6.22$\times10^7$ $<$ 6.7$\times10^{37}$ 6.08$\times10^7$ $<$ 1.1$\times10^{38}$ 5.86$\times10^7$ $<$ 7.7$\times10^{37}$
1984L 2.15$\times10^6$ $<$ 6.8$\times10^{38}$ 1.97$\times10^6$ $<$ 1.2$\times10^{39}$ 1.91$\times10^6$ $<$ 8.4$\times10^{38}$
1985F 3.97$\times10^6$ $<$ 7.3$\times10^{37}$ 3.64$\times10^6$ $<$ 1.3$\times10^{38}$ 3.54$\times10^6$ $<$ 9.0$\times10^{37}$
1990U 1.43$\times10^7$ $<$ 4.4$\times10^{38}$ 1.31$\times10^7$ $<$ 7.7$\times10^{38}$ 1.27$\times10^7$ $<$ 5.4$\times10^{38}$
1991N 2.04$\times10^7$ $($3.7$\pm$1.0$)\times10^{38}$ 1.85$\times10^7$ $($6.4$\pm$1.8$)\times10^{38}$ 1.81$\times10^7$ $($4.5$\pm$1.3$)\times10^{38}$
1993J 1.50$\times10^8$ $($45.3$\pm$0.5$)\times10^{37}$ 1.46$\times10^8$ $($74.5$\pm$0.8$)\times10^{37}$ 1.41$\times10^8$ $($52.7$\pm$0.5$)\times10^{37}$
1994I 3.46$\times10^8$ $($1.6$\pm$0.2$)\times10^{37}$ 3.20$\times10^8$ $($2.7$\pm$0.3$)\times10^{37}$ 3.08$\times10^8$ $($1.9$\pm$0.2$)\times10^{37}$
1996D 6.45$\times10^6$ $<$ 5.8$\times10^{39}$ 5.99$\times10^6$ $<$ 1.0$\times10^{40}$ 5.75$\times10^6$ $<$ 7.1$\times10^{39}$
1996N 1.64$\times10^7$ $<$ 5.5$\times10^{37}$ 1.60$\times10^7$ $<$ 9.0$\times10^{37}$ 1.53$\times10^7$ $<$ 6.5$\times10^{37}$
1996aq 2.95$\times10^6$ $<$ 7.7$\times10^{38}$ 2.71$\times10^6$ $<$ 1.3$\times10^{39}$ 2.63$\times10^6$ $<$ 9.4$\times10^{38}$
1997X 4.12$\times10^6$ $<$ 6.3$\times10^{37}$ 3.75$\times10^6$ $<$ 1.1$\times10^{38}$ 3.67$\times10^6$ $<$ 7.8$\times10^{37}$
1998T 4.89$\times10^7$ $($10.1$\pm$0.6$)\times10^{39}$ 4.55$\times10^7$ $($1.7$\pm$0.1$)\times10^{40}$ 4.38$\times10^7$ $($12.3$\pm$0.8$)\times10^{39}$
1998bo 5.93$\times10^6$ $<$ 7.0$\times10^{39}$ 5.64$\times10^6$ $<$ 1.2$\times10^{40}$ 5.41$\times10^6$ $<$ 8.4$\times10^{39}$
1998bw 2.09$\times10^7$ $($2.4$\pm$0.3$)\times10^{39}$ 1.89$\times10^7$ $($4.1$\pm$0.5$)\times10^{39}$ 1.86$\times10^7$ $($2.9$\pm$0.3$)\times10^{39}$
1999dn 2.40$\times10^7$ $<$ 4.7$\times10^{38}$ 2.19$\times10^7$ $<$ 8.3$\times10^{38}$ 2.14$\times10^7$ $<$ 5.8$\times10^{38}$
1999eh 7.38$\times10^6$ $<$ 5.3$\times10^{38}$ 6.76$\times10^6$ $<$ 9.2$\times10^{38}$ 6.56$\times10^6$ $<$ 6.5$\times10^{38}$
1999ex 4.83$\times10^6$ $<$ 2.8$\times10^{39}$ 4.44$\times10^6$ $<$ 4.9$\times10^{39}$ 4.30$\times10^6$ $<$ 3.5$\times10^{39}$
2000cr 6.63$\times10^6$ $<$ 1.1$\times10^{39}$ 6.09$\times10^6$ $<$ 2.0$\times10^{39}$ 5.90$\times10^6$ $<$ 1.4$\times10^{39}$
2000ds 2.71$\times10^7$ $<$ 1.8$\times10^{38}$ 2.48$\times10^7$ $<$ 3.1$\times10^{38}$ 2.41$\times10^7$ $<$ 2.2$\times10^{38}$
2001ig 2.06$\times10^7$ $($1.1$\pm$0.2$)\times10^{38}$ 1.87$\times10^7$ $($2.0$\pm$0.3$)\times10^{38}$ 1.83$\times10^7$ $($1.4$\pm$0.2$)\times10^{38}$
2003L 1.30$\times10^7$ $($1.3$\pm$0.2$)\times10^{40}$ 1.18$\times10^7$ $($2.3$\pm$0.4$)\times10^{40}$ 1.16$\times10^7$ $($1.6$\pm$0.3$)\times10^{40}$
2003bg 5.39$\times10^7$ $($16.9$\pm$0.7$)\times10^{38}$ 4.92$\times10^7$ $($3.0$\pm$0.1$)\times10^{39}$ 4.80$\times10^7$ $($20.7$\pm$0.8$)\times10^{38}$
2003is 2.96$\times10^6$ $<$ 7.0$\times10^{39}$ 2.86$\times10^6$ $<$ 1.2$\times10^{40}$ 2.65$\times10^6$ $<$ 8.6$\times10^{39}$
2004C 5.81$\times10^7$ $($7.7$\pm$0.8$)\times10^{38}$ 5.32$\times10^7$ $($1.3$\pm$0.1$)\times10^{39}$ 5.18$\times10^7$ $($9.4$\pm$1.0$)\times10^{38}$
2004dk 3.41$\times10^6$ $($5.7$^{+ 3.3}_{- 2.3}$ $)\times10^{38}$ 3.14$\times10^6$ $($9.9$^{+ 5.8}_{- 4.0}$ $)\times10^{38}$ 3.04$\times10^6$ $($7.0$^{+ 4.1}_{- 2.8}$ $)\times10^{38}$
2005U 4.15$\times10^7$ $($1.1$\pm$0.3$)\times10^{39}$ 3.85$\times10^7$ $($1.9$\pm$0.5$)\times10^{39}$ 3.70$\times10^7$ $($1.4$\pm$0.4$)\times10^{39}$
2005at 1.76$\times10^7$ $<$ 2.0$\times10^{37}$ 1.72$\times10^7$ $<$ 3.3$\times10^{37}$ 1.65$\times10^7$ $<$ 2.4$\times10^{37}$
2005cz 5.86$\times10^6$ $<$ 5.0$\times10^{38}$ 5.37$\times10^6$ $<$ 8.7$\times10^{38}$ 5.22$\times10^6$ $<$ 6.1$\times10^{38}$
2006ep 3.40$\times10^6$ $<$ 4.4$\times10^{39}$ 3.12$\times10^6$ $<$ 7.7$\times10^{39}$ 3.03$\times10^6$ $<$ 5.5$\times10^{39}$
2007Y 1.65$\times10^7$ $<$ 1.3$\times10^{38}$ 1.61$\times10^7$ $<$ 2.1$\times10^{38}$ 1.54$\times10^7$ $<$ 1.5$\times10^{38}$
2007gr 6.78$\times10^6$ $<$ 8.5$\times10^{37}$ 6.50$\times10^6$ $<$ 1.4$\times10^{38}$ 6.31$\times10^6$ $<$ 1.0$\times10^{38}$
2007ke 4.51$\times10^7$ $<$ 7.5$\times10^{39}$ 4.32$\times10^7$ $<$ 1.2$\times10^{40}$ 4.18$\times10^7$ $<$ 8.8$\times10^{39}$
2008D 7.52$\times10^6$ $($6.5$^{+ 2.8 }_{- 2.1}$ $)\times10^{38}$ 6.90$\times10^6$ $($1.1$^{+ 0.5}_{- 0.4}$ $)\times10^{39}$ 6.69$\times10^6$ $($8.0$^{+ 3.4 }_{- 2.5}$ $)\times10^{38}$
2008bo 8.16$\times10^6$ $<$ 3.5$\times10^{38}$ 7.50$\times10^6$ $<$ 6.1$\times10^{38}$ 7.27$\times10^6$ $<$ 4.3$\times10^{38}$
2009bb 4.21$\times10^6$ $($1.7$^{+ 0.9 }_{- 0.6}$ $)\times10^{39}$ 3.87$\times10^6$ $($3.0$^{+ 1.5}_{- 1.1}$ $)\times10^{39}$ 3.76$\times10^6$ $($2.1$^{+ 1.1 }_{- 0.8 }$ $)\times10^{39}$
2009jf 4.26$\times10^6$ $($1.2$\pm$0.1$)\times10^{40}$ 3.92$\times10^6$ $($2.0$\pm$0.2$)\times10^{40}$ 3.80$\times10^6$ $($1.4$\pm$0.2$)\times10^{40}$
2009mk 4.20$\times10^6$ $<$ 4.7$\times10^{38}$ 3.87$\times10^6$ $<$ 8.1$\times10^{38}$ 3.75$\times10^6$ $<$ 5.8$\times10^{38}$
2010O 3.72$\times10^7$ $($17.6$\pm$0.9$)\times10^{39}$ 3.46$\times10^7$ $($3.0$\pm$0.2$)\times10^{40}$ 3.32$\times10^7$ $($2.2$\pm$0.1$)\times10^{40}$
2011dh 3.14$\times10^8$ $($12.1$\pm$0.4$)\times10^{37}$ 2.91$\times10^8$ $($20.1$\pm$0.6$)\times10^{37}$ 2.80$\times10^8$ $($14.9$\pm$0.4$)\times10^{37}$
2011ei 4.12$\times10^6$ $<$ 1.5$\times10^{39}$ 3.81$\times10^6$ $<$ 2.6$\times10^{39}$ 3.67$\times10^6$ $<$ 1.8$\times10^{39}$
2012ap 4.11$\times10^6$ $<$ 3.6$\times10^{39}$ 3.80$\times10^6$ $<$ 6.2$\times10^{39}$ 3.66$\times10^6$ $<$ 4.4$\times10^{39}$
2013ak 4.08$\times10^6$ $($5.4$\pm$1.0$)\times10^{38}$ 3.80$\times10^6$ $($9.2$\pm$1.7$)\times10^{38}$ 3.64$\times10^6$ $($6.6$\pm$1.2$)\times10^{38}$
2013ge 7.66$\times10^6$ $<$ 1.7$\times10^{38}$ 7.13$\times10^6$ $<$ 2.9$\times10^{38}$ 6.83$\times10^6$ $<$ 2.1$\times10^{38}$
2014C 7.95$\times10^6$ $($8.0$\pm$0.4$)\times10^{39}$ 7.48$\times10^6$ $($13.5$\pm$0.6$)\times10^{39}$ 7.11$\times10^6$ $($9.8$\pm$0.4$)\times10^{39}$
-------- ------------------ -------------------------------------------- ------------------ ------------------------------------------- ------------------ --------------------------------------------- -- -- -- -- -- --
(Note that the luminosities presented here do not accurately reflect the luminosity of supernovae due to chosen spectral models or multi-epoch stacking of observations.)
-------- ------------------- ------------------------------------------- ------------------ -------------------------------------------- ------------------ ------------------------------------------- -- -- -- -- -- --
Name Exp. map av. Luminosity Exp. map av. Luminosity Exp. map av. Luminosity
value (cm$^2$s) (erg s$^{-1}$) value (cm$^2$s) (erg s$^{-1}$) value (cm$^2$s) (erg s$^{-1}$)
1983I 5.84$\times10^7$ $<$ 4.7$\times10^{37}$ 5.38$\times10^7$ $<$ 2.9$\times10^{37}$ 5.79$\times10^7$ $<$ 5.4$\times10^{37}$
1983N 2.46$\times10^8$ $<$ 4.1$\times10^{36}$ 3.14$\times10^8$ $<$ 1.8$\times10^{36}$ 2.78$\times10^8$ $<$ 4.1$\times10^{36}$
1983V 5.66$\times10^7$ $<$ 1.2$\times10^{38}$ 6.13$\times10^7$ $<$ 6.4$\times10^{37}$ 5.97$\times10^7$ $<$ 1.3$\times10^{38}$
1984L 1.73$\times10^6$ $<$ 1.4$\times10^{39}$ 2.20$\times10^6$ $<$ 6.3$\times10^{38}$ 1.94$\times10^6$ $<$ 1.4$\times10^{39}$
1985F 3.19$\times10^6$ $<$ 1.5$\times10^{38}$ 4.08$\times10^6$ $<$ 6.8$\times10^{37}$ 3.59$\times10^6$ $<$ 1.5$\times10^{38}$
1990U 1.15$\times10^7$ $<$ 9.1$\times10^{38}$ 1.46$\times10^7$ $<$ 4.1$\times10^{38}$ 1.30$\times10^7$ $<$ 9.3$\times10^{38}$
1991N 1.62$\times10^7$ $($7.6$\pm$2.2$)\times10^{38}$ 2.11$\times10^7$ $($3.3$\pm$1.0$)\times10^{38}$ 1.83$\times10^7$ $($7.8$\pm$2.2$)\times10^{38}$
1993J 1.36$\times10^8$ $($83.0$\pm$0.9$)\times10^{37}$ 1.49$\times10^8$ $($43.2$\pm$0.5$)\times10^{37}$ 1.44$\times10^8$ $($90.3$\pm$0.9$)\times10^{37}$
1994I 2.79$\times10^8$ $($3.2$\pm$0.4$)\times10^{37}$ 3.52$\times10^8$ $($1.4$\pm$0.2$)\times10^{37}$ 3.15$\times10^8$ $($3.3$\pm$0.4$)\times10^{37}$
1996D 5.25$\times10^6$ $<$ 1.2$\times10^{40}$ 6.51$\times10^6$ $<$ 5.5$\times10^{39}$ 5.89$\times10^6$ $<$ 1.2$\times10^{40}$
1996N 1.48$\times10^7$ $<$ 1.0$\times10^{38}$ 1.60$\times10^7$ $<$ 5.4$\times10^{37}$ 1.57$\times10^7$ $<$ 1.1$\times10^{38}$
1996aq 2.38$\times10^6$ $<$ 1.6$\times10^{39}$ 3.01$\times10^6$ $<$ 7.1$\times10^{38}$ 2.67$\times10^6$ $<$ 1.6$\times10^{39}$
1997X 3.28$\times10^6$ $<$ 1.3$\times10^{38}$ 4.28$\times10^6$ $<$ 5.8$\times10^{37}$ 3.70$\times10^6$ $<$ 1.3$\times10^{38}$
1998T 4.01$\times10^7$ $($2.0$\pm$0.1$)\times10^{40}$ 4.93$\times10^7$ $($9.4$\pm$0.6$)\times10^{39}$ 4.47$\times10^7$ $($2.1$\pm$0.1$)\times10^{40}$
1998bo 5.08$\times10^6$ $<$ 1.4$\times10^{40}$ 5.87$\times10^6$ $<$ 6.7$\times10^{39}$ 5.53$\times10^6$ $<$ 1.4$\times10^{40}$
1998bw 1.66$\times10^7$ $($4.9$\pm$0.6$)\times10^{39}$ 2.18$\times10^7$ $($2.1$\pm$0.2$)\times10^{39}$ 1.87$\times10^7$ $($5.0$\pm$0.6$)\times10^{39}$
1999dn 1.92$\times10^7$ $<$ 9.8$\times10^{38}$ 2.48$\times10^7$ $<$ 4.3$\times10^{38}$ 2.16$\times10^7$ $<$ 1.0$\times10^{39}$
1999eh 5.90$\times10^6$ $<$ 1.1$\times10^{39}$ 7.57$\times10^6$ $<$ 4.9$\times10^{38}$ 6.67$\times10^6$ $<$ 1.1$\times10^{39}$
1999ex 3.89$\times10^6$ $<$ 5.8$\times10^{39}$ 4.95$\times10^6$ $<$ 2.6$\times10^{39}$ 4.38$\times10^6$ $<$ 6.0$\times10^{39}$
2000cr 5.33$\times10^6$ $<$ 2.3$\times10^{39}$ 6.79$\times10^6$ $<$ 1.0$\times10^{39}$ 6.00$\times10^6$ $<$ 2.4$\times10^{39}$
2000ds 2.17$\times10^7$ $<$ 3.7$\times10^{38}$ 2.78$\times10^7$ $<$ 1.7$\times10^{38}$ 2.44$\times10^7$ $<$ 3.8$\times10^{38}$
2001ig 1.63$\times10^7$ $($2.4$\pm$0.4$)\times10^{38}$ 2.14$\times10^7$ $($1.0$\pm$0.2$)\times10^{38}$ 1.85$\times10^7$ $($2.4$\pm$0.4$)\times10^{38}$
2003L 1.03$\times10^7$ $($2.7$\pm$0.5$)\times10^{40}$ 1.35$\times10^7$ $($1.2$\pm$0.2$)\times10^{40}$ 1.17$\times10^7$ $($2.8$\pm$0.5$)\times10^{40}$
2003bg 4.30$\times10^7$ $($3.5$\pm$0.1$)\times10^{39}$ 5.57$\times10^7$ $($15.4$\pm$0.6$)\times10^{38}$ 4.85$\times10^7$ $($3.6$\pm$0.1$)\times10^{39}$
2003is 2.52$\times10^6$ $<$ 1.4$\times10^{40}$ 2.82$\times10^6$ $<$ 7.0$\times10^{39}$ 2.78$\times10^6$ $<$ 1.4$\times10^{40}$
2004C 4.66$\times10^7 $ $($1.6$\pm$0.2$)\times10^{39}$ 5.99$\times10^7$ $($7.0$\pm$0.8$)\times10^{38}$ 5.25$\times10^7$ $($1.6$\pm$0.2$)\times10^{39}$
2004dk 2.75$\times10^6$ $($1.2$^{+ 0.7}_{- 0.5}$ $)\times10^{39}$ 3.48$\times10^6$ $($5.3$^{+ 3.1}_{- 2.1}$ $)\times10^{38}$ 3.09$\times10^6$ $($1.2$^{+ 0.7}_{- 0.5}$ $)\times10^{39}$
2005U 3.38$\times10^7$ $($2.3$\pm$0.6$)\times10^{39}$ 4.19$\times10^7$ $($1.1$\pm$0.3$)\times10^{39}$ 3.79$\times10^7$ $($2.3$\pm$0.6$)\times10^{39}$
2005at 1.59$\times10^7$ $<$ 3.7$\times10^{37}$ 1.71$\times10^7$ $<$ 2.0$\times10^{37}$ 1.69$\times10^7$ $<$ 4.0$\times10^{37}$
2005cz 4.70$\times10^6$ $<$ 1.0$\times10^{39}$ 6.02$\times10^6$ $<$ 4.6$\times10^{38}$ 5.29$\times10^6$ $<$ 1.0$\times10^{39}$
2006ep 2.73$\times10^6$ $<$ 9.2$\times10^{39}$ 3.48$\times10^6$ $<$ 4.1$\times10^{39}$ 3.08$\times10^6$ $<$ 9.4$\times10^{39}$
2007Y 1.48$\times10^7$ $<$ 2.3$\times10^{38}$ 1.61$\times10^7$ $<$ 1.2$\times10^{38}$ 1.58$\times10^7$ $<$ 2.5$\times10^{38}$
2007gr 5.99$\times10^6$ $<$ 1.6$\times10^{38}$ 6.73$\times10^6$ $<$ 8.1$\times10^{37}$ 6.41$\times10^6$ $<$ 1.7$\times10^{38}$
2007ke 3.98$\times10^7$ $<$ 1.4$\times10^{40}$ 4.46$\times10^7$ $<$ 7.1$\times10^{39}$ 4.26$\times10^7$ $<$ 1.5$\times10^{40}$
2008D 6.03$\times10^6$ $($1.3$^{+ 0.6}_{- 0.4}$ $)\times10^{39}$ 7.70$\times10^6$ $($6.0$^{+ 2.6 }_{- 1.9}$ $)\times10^{38}$ 6.80$\times10^6$ $($1.4$^{+ 0.6}_{- 0.4}$ $)\times10^{39}$
2008bo 6.57$\times10^6$ $<$ 7.2$\times10^{38}$ 8.35$\times10^6$ $<$ 3.2$\times10^{38}$ 7.39$\times10^6$ $<$ 7.4$\times10^{38}$
2009bb 3.40$\times10^6$ $($3.5$^{+ 1.8}_{- 1.3}$ $)\times10^{39}$ 4.32$\times10^6$ $($1.6$^{+ 0.8 }_{- 0.6}$ $)\times10^{39}$ 3.82$\times10^6$ $($3.6$^{+ 1.9}_{- 1.4}$ $)\times10^{39}$
2009jf 3.44$\times10^6$ $($2.4$\pm$0.3$)\times10^{40}$ 4.36$\times10^6$ $($1.1$\pm$0.1$)\times10^{40}$ 3.86$\times10^6$ $($2.4$\pm$0.3$)\times10^{40}$
2009mk 3.39$\times10^6$ $<$ 9.6$\times10^{38}$ 4.29$\times10^6$ $<$ 4.3$\times10^{38}$ 3.82$\times10^6$ $<$ 9.8$\times10^{38}$
2010O 3.03$\times10^7$ $($3.6$\pm$0.2$)\times10^{40}$ 3.75$\times10^7$ $($16.5$\pm$0.9$)\times10^{39}$ 3.40$\times10^7$ $($3.7$\pm$0.2$)\times10^{40}$
2011dh 2.55$\times10^8$ $($24.8$\pm$0.7$)\times10^{37}$ 3.17$\times10^8$ $($11.4$\pm$0.3$)\times10^{37}$ 2.86$\times10^8$ $($25.3$\pm$0.8$)\times10^{37}$
2011ei 3.34$\times10^6$ $<$ 3.0$\times10^{39}$ 4.18$\times10^6$ $<$ 1.4$\times10^{39}$ 3.75$\times10^6$ $<$ 3.1$\times10^{39}$
2012ap 3.33$\times10^6$ $<$ 7.4$\times10^{39}$ 4.16$\times10^6$ $<$ 3.4$\times10^{39}$ 3.74$\times10^6$ $<$ 7.5$\times10^{39}$
2013ak 3.33$\times10^6$ $($1.1$\pm$0.2$)\times10^{39}$ 4.11$\times10^6$ $($5.0$\pm$0.9$)\times10^{38}$ 3.73$\times10^6$ $($1.1$\pm$0.2$)\times10^{39}$
2013ge 6.25$\times10^6$ $<$ 3.5$\times10^{38}$ 7.68$\times10^6$ $<$ 1.6$\times10^{38}$ 7.00$\times10^6$ $<$ 3.5$\times10^{38}$
2014C 6.58$\times10^6$ $($16.0$\pm$0.7$)\times10^{39}$ 7.86$\times10^6$ $($7.6$\pm$0.3$)\times10^{39}$ 7.34$\times10^6$ $($16.5$\pm$0.7$)\times10^{39}$
-------- ------------------- ------------------------------------------- ------------------ -------------------------------------------- ------------------ ------------------------------------------- -- -- -- -- -- --
(Note that the luminosities presented here do not accurately reflect the luminosity of supernovae due to chosen spectral models or multi-epoch stacking of observations.)
Detected pre-explosion sources {#sect_detected}
------------------------------
Although for most pre-explosion targets no source flux above the 3$\sigma$ detection threshold was found, there are three targets that do display a clear signal: SN 2004gt, SN 2009jf and SN 2010O.
In the case of SN 2004gt, a complex emission area permeates the entire source region, confusing any signal from a potential progenitor (see Fig. \[2004gt\_fig\]). Because this emission is primarily from low-energy photons, we re-examined this dataset using a higher, 1.5-8.0 keV energy range. First, we corrected the aspect solution of the seven original *Chandra* observations using CIAO’s *reproject\_aspect* tool by matching sources found with *wavdetect* with the USNO A2.0 catalog [@USNO], then reprocessing the data the same way as discussed before. We then remade the photon count and exposure maps for SN 2004gt position using the higher 1.5-8.0 keV energy range. The corrected count map was convolved with a Gaussian kernel with $\sigma$ = 1.27 pixels (FWHM = 1.5 arcsec) to enhance the detection of faint point sources. In the convolved image a point-like source becomes visible close to the position of SN 2004gt. We identify this source as the known X-ray source CXOU J120150.4-185212 [@2004gt_catalog_source]. To determine whether this pre-explosion source is coincident with the supernova position we determined its centre coordinates with 1-dimensional Gaussian fitting using the IRAF XIMTOOL package and with centroiding and Gaussian fitting as implemented in the IRAF APPHOT package. These three methods gave similar coordinates with a standard deviation of 0.4 pixels. However, the uncertainty in the centre coordinates is probably larger than this given the low S/N of the object in the original unconvolved data and we therefore adopt 1 pixel (corresponding to 0.5 arcsec) as the uncertainty in its (x,y) position. Combining this error with the uncertainty in the absolute astrometric calibration yields a total error of $\pm$0.6 arcsec for both the RA and Dec of the pre-explosion source (at the distance of SN 2004gt, 1 arcsec corresponds to a projected distance of $\sim$100 pc). We therefore adopt the position RA = 12:01:50.40, Dec = -18:52:12.7 for the source.
We also derived new absolute astrometry for SN 2004gt making use of near-infrared Ks-band imaging of NGC 4038/9 obtained as a part of a search for SNe in the nuclear regions of starburst galaxies [see @mattila_starburst] with the William Herschel Telescope (WHT) LIRIS instrument on 2005 January 30 and INGRID instrument on 2002 January 3. The images were reduced using standard procedures in IRAF [see @kangas_2013]. For this the World Coordinate System (WCS) of the LIRIS image was calibrated making use of the centroid positions of 26 point-like sources (incl. both foreground stars and star clusters in NGC 4038/9) together with their coordinates from the 2MASS catalog. The SN is located in proximity of a bright star cluster making the precise determination of its location in ground-based seeing limited images more difficult. To obtain accurate coordinates for the SN we therefore aligned the pre-explosion INGRID image to the LIRIS image and performed image subtraction using the ISIS 2.2 package [@alard_1998; @alard_2000]. The centroid position of the SN measured from the subtracted image corresponds to RA = 12:01:50.42, Dec = -18:52:13.5, with uncertainty of 0.16 arcsec in RA and 0.19 arcsec in Dec dominated by the uncertainty in the WCS transformation. The SN is therefore coincident with the derived position of the source within $\sim1\sigma$.
We measured the flux of this source in the higher energy range, and found 50 photons inside a source aperture with a 2.5 arcsec radius. A scaled background of 25 photons was obtained from an annulus with outer radius of 12 arcsec centred on the source (excluding two interfering sources from the background, see Fig. \[2004gt\_source\_fig\]). The number of photons was too low to obtain a spectrum, and we therefore measured the luminosity using the same models we used for the other SNe[^9]. Rescaling from 1.5-8.0 keV to the 0.5-7.0 keV energy range, we obtained the unabsorbed luminosity of $(3.7\pm1.1)\times10^{37}$ erg s$^{-1}$ for the low-absorption 1 keV temperature blackbody (model 1), $(5.8\pm1.6)\times10^{37}$ erg s$^{-1}$ for the high-absorption 0.5 photon-index powerlaw (model 4), with the luminosities for other models falling between these two values. The average luminosity from all models is $4.6\times10^{37}$ erg s$^{-1}$. We note that this constitutes a detection with an approximately 3$\sigma$ significance. Note that the values presented in Table \[resultstable\_pre1\] are the measurements for the original 0.5-7.0 keV energy range at the SN position using the extraction-regions depicted in Fig. \[2004gt\_fig\]. Because the emission in this energy range most likely originates from the diffuse emission field permeating the region, these measurements do not accurately reflect the nature of the source discussed above, although it is contained within the source aperture. We also note the proximity of a bright star cluster approximately 1 arcsec North and 0.6 arcsec West from the SN position [see Fig. 1 in @maund_2005], possibly contributing some X-ray sources/emission to the vicinity of the SN location [see eg. @juri_ulx for possible associations between star clusters and HMXBs].
In the case of SN 2009jf the flux is from the ultraluminous X-ray source CXOU J230453.0+121959, which overlaps with the SN position and could potentially be associated with the SN, although this cannot be confirmed [@ref26]. The post-explosion observation, taken $\sim$ 30 days after the SN explosion (2009 September 23, from @ref10), shows the ULX with a substantially increased luminosity ($L_\mathrm{diff} \sim 4.3\times10^{39}$ erg s$^{-1}$). As @ref26 also note, this would be consistent with the increase in the flux being caused by the SN.
![Top: SN 2013dk and SN 2004gt (pre-explosion), at the 0.5-7.0 keV energy range. The inner circles are the source apertures, surrounded by the background annuli. The radius of the source aperture is 2.5 arcsec for both cases, and the outer radii of the background annuli are 6 and 12 arcsec, respectively. The apertures are centred on the SN positions. Bottom: The source (CXOU J120150.4-185212) and our measured position of SN 2004gt (pre-explosion) at 1.5-8.0 keV energy range. The central circle is the source aperture (radius 2.5 arcsec) centred on the centroid of the source, and the annulus (outer radius 12 arcsec) around it is the background extraction area. The circles with red bars contain unrelated sources which were excluded from the background extraction area. Each image pixel corresponds to $\sim$0.5 arcsec. FWHM=3-pixel gaussian smoothing has been applied to the bottom image to make the source more visible.[]{data-label="2013dk_fig"}](2004gt_2013dk_NRM_EN.eps "fig:"){width="50.00000%"} ![Top: SN 2013dk and SN 2004gt (pre-explosion), at the 0.5-7.0 keV energy range. The inner circles are the source apertures, surrounded by the background annuli. The radius of the source aperture is 2.5 arcsec for both cases, and the outer radii of the background annuli are 6 and 12 arcsec, respectively. The apertures are centred on the SN positions. Bottom: The source (CXOU J120150.4-185212) and our measured position of SN 2004gt (pre-explosion) at 1.5-8.0 keV energy range. The central circle is the source aperture (radius 2.5 arcsec) centred on the centroid of the source, and the annulus (outer radius 12 arcsec) around it is the background extraction area. The circles with red bars contain unrelated sources which were excluded from the background extraction area. Each image pixel corresponds to $\sim$0.5 arcsec. FWHM=3-pixel gaussian smoothing has been applied to the bottom image to make the source more visible.[]{data-label="2013dk_fig"}](2004gt_H_EN_source.eps "fig:"){width="50.00000%"}
\[2004gt\_fig\] \[2004gt\_source\_fig\]
SN 2010O in the galaxy Arp 299 has an extremely complex background, which makes any measurements taken for this source difficult. @ref2 suggested the detection of a progenitor for SN 2010O by measuring the X-ray flux variability between the two *Chandra* pre-explosion datasets (OBSIDs 1641 and 6227). We had, in addition to these, two more recent post-explosion datasets (OBSIDs 15077 and 15619). We subtracted the combined (exposure-corrected) pre- and post-explosion observations from each other to see how much variability there was in the source aperture area (see Fig. \[2010O\_diffmap\] for the exposure-corrected difference map). While there is variability within the source aperture, it is widely distributed spatially and could not be attributed to any single point source (notably, SN 2010O cannot be clearly identified from the post-explosion dataset). This suggests the area contains multiple overlapping variable sources within a very complex X-ray background, and identifying any particular sources inside the source aperture could not be made with any confidence. We have included the measured luminosities for this SN in our sample for completeness but note that the results presented here cannot be associated with a progenitor but rather reflect the complex nature of the extraction area. The post-explosion measurements of SN 2010O likewise reflect the complexity of the region, rather than any clear source that could be associated with the SN.
![Exposure-corrected difference-map of SN 2010O position in Arp 299 (cm$^{-2}$s$^{-1}$) (Spectral model: 1 keV blackbody, $N_\mathrm{H}=0.5\times10^{22}$ cm$^{-2}$, darker areas indicate reduction in flux in the post-explosion data). FWHM=3-pixel gaussian smoothing has been applied to the image. The concentric circles depict the source aperture and the background region used for extraction. The circle with a red bar contains numerous unrelated bright sources which were excluded from the extraction area. The cross marks the position of the transient source found by @ref2.[]{data-label="2010O_diffmap"}](2010O_diffmap.eps){width="50.00000%"}
Detected post-explosion sources {#sect_detected_post}
-------------------------------
In the post-explosion observations, in addition to SN 2009jf and SN 2010O, there are 16 other sources that have flux above the 3$\sigma$ detection threshold. Of these, in the cases of SN 1993J [@ref30], SN 1998bw [@ref28], SN 2001ig [@ref31], SN 2003L [@dist28_b], 2003bg [@2003bg_disc], SN 2004C [@ref34], SN 2008D [@ref6], SN 2011dh [@ref32] and SN 2013ak [@ref33] the source is the supernova itself. A notable example is the case of SN 2008D, where the observation was taken only 10 days after the explosion, which is well recorded as it was serendipitously discovered by @ref6 while observing a different SN in the same galaxy using the *Swift* X-ray telescope. The sources at the positions of SN 2004dk, SN 2009bb and 2014C (see fig. \[unconfirmed\_SNe\] are most likely the SNe themselves as well given the timing of the observations, though we could not confirm these from literature. Note that SN 2009bb is partially blended with a nearby source. Because characterising the properties of these sources is outside the scope of this paper, we did not attempt to disentangle them. Similarly, in the case of SN 1998bw, the source aperture actually contains two sources at close proximity (Fig. \[1998bw\_fig\]), of which the source closer to the centre of the source aperture is SN 1998bw [for details, see @ref28]. Again, because the sources partially overlap, we did not attempt to disentangle them.
SN 1998T and SN 2005U are in the same galaxy as SN 2010O, Arp 299, and in close enough proximity to each other ($\sim$2.5 arcsec) that their source apertures partially overlap. Similar to SN 2010O, the region is very complex and it is difficult to identify particularly faint sources from the area. The position of SN 1998T overlaps with a known ULX, \[ZWM2003\] 12 [@2005U_zezas], which we identify as the source. SN 2005U had both pre- and post-explosion data available, and for both measurements we excluded the ULX and other interfering sources (see Fig. \[2005U\_fig\]) from both the background and the source aperture. Although this produces an apparent detection at the position of SN 2005U post-explosion (the flux from within the defined source aperture exceeds the background by 3$\sigma$), we associate this with the inhomogeneous distribution of diffuse emission in the region, rather than any point-sources within the aperture (the reason why this signal is not apparent in the pre-explosion observation is likely due to the shorter exposure time – the luminosity of this apparent source is lower than the upper limit established for the pre-explosion case). The measured luminosities presented in tables \[resultstable\_post1\] and \[resultstable\_post2\] for this SN should therefore be mainly treated as upper limits.
![The post-explosion sources at the positions of SN 2004dk, SN 2009bb and SN 2014C. FWHM=3-pixel gaussian smoothing has been applied to the images of SN 2004 dk and SN 2009bb to make the sources more visible. The circles with red bars contains sources which were excluded from the extraction areas. The exclusion-circle near SN 2014C is at the position of a variable source visible in the pre-explosion data. Note that the source at SN 2009bb position partially blends with a nearby excluded source.[]{data-label="unconfirmed_SNe"}](2004dk_post.eps "fig:"){width="50.00000%"} ![The post-explosion sources at the positions of SN 2004dk, SN 2009bb and SN 2014C. FWHM=3-pixel gaussian smoothing has been applied to the images of SN 2004 dk and SN 2009bb to make the sources more visible. The circles with red bars contains sources which were excluded from the extraction areas. The exclusion-circle near SN 2014C is at the position of a variable source visible in the pre-explosion data. Note that the source at SN 2009bb position partially blends with a nearby excluded source.[]{data-label="unconfirmed_SNe"}](2009bb_post_smooth.eps "fig:"){width="50.00000%"} ![The post-explosion sources at the positions of SN 2004dk, SN 2009bb and SN 2014C. FWHM=3-pixel gaussian smoothing has been applied to the images of SN 2004 dk and SN 2009bb to make the sources more visible. The circles with red bars contains sources which were excluded from the extraction areas. The exclusion-circle near SN 2014C is at the position of a variable source visible in the pre-explosion data. Note that the source at SN 2009bb position partially blends with a nearby excluded source.[]{data-label="unconfirmed_SNe"}](2014C_post.eps "fig:"){width="50.00000%"}
![Top: SN 1998bw position (post-explosion). Bottom: SN 2005U position (post-explosion). The central circle is the source extraction area and the annulus around it is the background extraction area. The circles with red bars contain unrelated sources which were excluded from the extraction area. The relative location of SN 1998T is also marked.[]{data-label="1998bw_fig"}](1998bw_post.eps "fig:"){width="50.00000%"} ![Top: SN 1998bw position (post-explosion). Bottom: SN 2005U position (post-explosion). The central circle is the source extraction area and the annulus around it is the background extraction area. The circles with red bars contain unrelated sources which were excluded from the extraction area. The relative location of SN 1998T is also marked.[]{data-label="1998bw_fig"}](2005U_post_1998T_post_FINAL.eps "fig:"){width="50.00000%"}
\[2005U\_fig\]
In the case of SN 1994I we see three faint point sources inside the source aperture, one of which appears to be very close to the SN position, but overall all three are relatively faint and within *Chandra’s* 90 percent encircled energy radius for the PSF which makes associating any of them with a particular source within this source aperture unreliable, although the central one most likely is the SN [also see @ref36]. For the purposes of this study and unlike in the case of SN 1998bw, because the sources do not overlap as clearly, we have excluded the other sources and measured the central source, assuming it to be the SN. Similarly, with SN 1991N, the source region is extremely crowded, with high background emission and multiple nearby bright point sources. There are also several possible sources inside the source aperture which cannot be directly associated with the SN position or identified with confidence to exclude, which together produce enough flux to rate as a false detection.
Finally, in the case of SN 1983I, we note that there is some uncertainty to the coordinates of this supernova (Asiago catalog only reports the coordinates of the host galaxy while USC and SSC both report different coordinates $\sim$2 arcsec apart[^10]). The coordinates used in this study are those reported by USC. There is a clear X-ray point-source $\sim$3 arcsec from the USC position which we were unable to conclusively identify and therefore cannot rule out that this is actually SN 1983I. Note that the luminosities for post-explosion sources presented in this paper are not directly representative of the SNe, as our models represent HMXBs and were not fitted to the individual spectra of the SNe. Similarly, some of the measurements (such as those of SN 1993J) involved stacking observations across multiple epochs over a potentially long time-period, and therefore would include any variability in the source during that timeframe.
SN luminosity upper limits {#sect_ulim}
--------------------------
Figures \[pre\_sn\_plot\] and \[post\_sn\_plot\] show the measured upper limits plotted for the six spectral models used. Targets with a detection above 3$\sigma$ are not included. Because Wolf-Rayet stars as a part of HMXB systems will result in stripped-envelope SNe, in both plots we also compare our results against the luminosities of the three Wolf-Rayet X-ray binaries that are known to exist. These are NGC 300 X-1 with average unabsorbed luminosity $4.1\times10^{38}$ erg s$^{-1}$ [@ref12], IC 10 X-1 with unabsorbed luminosity $1.8\times10^{38}$ erg s$^{-1}$ [@ref13] and Cygnus X-3, presented in two modes: quiescent with $L=1.4\times10^{37}$ erg s$^{-1}$ and hypersoft with $L=5.7\times10^{37}$ erg s$^{-1}$ [both from @ref14 representing the states with the lowest and highest X-ray fluxes, respectively]. A fourth candidate WR-binary CXOU J004732.0-251722.1 in the galaxy NGC 253 identified by @ref11 is also included (with unabsorbed luminosity $L=9.9\times10^{37}$ erg s$^{-1}$). The aforementioned luminosities have been re-scaled to our 0.5-7.0 keV energy range from their original values. Finally, we also include a comparison to ultraluminous X-ray sources (ULX), which are defined as sources with X-ray luminosity above $1\times10^{39}$ erg s$^{-1}$.
{width="90.00000%"}
{width="90.00000%"}
Fraction of X-ray bright progenitor systems {#sect_eldridge}
-------------------------------------------
There are two requirements for a progenitor to be a luminous X-ray source in a pre-explosion observation. The first is for the progenitor star to have a compact object as a binary companion. The second is that the binary separation must be small enough for the progenitor to supply mass to the companion, either by Roche-lobe overflow or stellar wind, to produce the X-ray luminosity.
We have attempted to gain an estimate of the number of binary systems that might be luminous in X-rays when the system explodes. We have used the latest version (2.0) of the Binary Population and Spectral Synthesis, BPASS, stellar population models [@new_bpass; @bpass Eldridge et al., in prep; http://bpass.auckland.ac.nz]. The stellar population included stars at solar metallicity with initial masses from 0.1 to 300M$_{\odot}$. We determined the mass of the primary using a Salpeter initial mass function (IMF) with a slope of $-$2.35. We then use a binary period distribution that is flat in log of the period and flat in mass ratio distribution to determine the mass of the secondary. These are consistent with the distributions observed by @sana. The orbits are assumed to be circular. While observed binaries have eccentric orbits, we note that if the stars are close enough to interact, circularization will most likely have occurred and, as found by @hurley, a full treatment of eccentricity is not generally required in population synthesis.
We assume that a star explodes if it has a final mass greater than 2M$_{\odot}$ and a CO core mass greater than 1.38M$_{\odot}$. We count the stellar models which will explode as type Ib, Ic or IIb SNe being those which have less than 0.5M$_{\odot}$ hydrogen at the point of core-collapse. We then also record each event that occurs within a binary with a compact object where the size of the progenitor model is comparable to the size of the orbit and therefore is likely to be emitting X-rays via either Roche-lobe overflow or by wind accretion. This is a simple way to estimate the number of progenitors that might be in X-ray binary systems. We also separately record SNe that are the first ones to occur in the binary system, although these are less likely to be detectable in pre-explosion X-ray observations, as wind-wind collision systems have much weaker X-ray emission than HMXBs.
We show these estimates in Table \[eldridgetable\], where the first column is the lower limit for the ratio of the SN progenitor’s radius to the binary separation. The second column shows the percentage of all stripped-envelope SNe (in binary systems) that have the corresponding or higher ratio of progenitor radius to binary separation. The third column is the same as the second one, except only for systems where the progenitor’s companion has already exploded as a SN, producing a compact object. From this, we can see that the fraction of stripped envelope progenitor systems that are likely to be luminous X-ray sources is only up to a few percent. This number could be significantly higher if close binaries without a compact object, where stellar winds interact and produce X-rays (with significantly lower luminosities), were also included. However, the values provided by our modelling are uncertain and depend on many factors. They could be increased if, for example, mass-transfer is more efficient than assumed or supernova kicks are weaker than assumed. We therefore stress that these values are only rough estimates, but they do at least indicate the order of magnitude for the expected rate.
----------------------- -------------------- -------------------- -- -- -- -- -- -- --
Ratio of progenitor Fraction of binary Fraction of binary
radius to separation, systems with a systems with a
$(R_* / a)$ normal companion compact companion
$>$ 1 0.07 0.004
$>$ 0.5 0.18 0.009
$>$ 0.1 0.39 0.018
$>$ 0.01 0.72 0.026
----------------------- -------------------- -------------------- -- -- -- -- -- -- --
: Stripped envelope SN progenitors as a function mm of binary separation.[]{data-label="eldridgetable"}
Upper limits for future nearby SNe {#sect_future_SN}
----------------------------------
In order to investigate the potential of the presented method for constraining the progenitor system of a future SN in a nearby galaxy, we measured the upper limit of unabsorbed luminosity for an empty field in the host galaxy of SN 1993J, NGC 3031, 10.2 arcsec northeast of the SN and using the same set of observations. We chose this galaxy not only because it is the host to the closest SN in our sample (at distance 3.63 Mpc), but also because it has one of the longest combined exposure times available (752.3 ks). The extraction area was a circular aperture with 2.5 arcsec radius, which was found to contain 24 photons for the combined observation. For the spectral models used in this study, the range of upper limits of luminosity (corrected for absorption) obtained were $L_\mathrm{UL}=(0.7-1.4)\times10^{36}$ erg s$^{-1}$ for models 5 and 6 respectively, with other models falling between these values. This gives us a reasonable estimate for the faintest progenitor source that could be detected for a future extragalactic SN. It is notable that this is almost an order of magnitude fainter than the quiescent state of Cygnus X-3, the closest and faintest known WR X-ray binary at the distance of $\sim$9 kpc [@ref14].
Discussion {#sect_discussion}
==========
Of the 18 SNe with pre-explosion data available, three were found to have flux above the 3$\sigma$ detection threshold at their positions. Two of these are identifiable point sources and possible candidates for a HMXB SN progenitor, both previously known sources. The possible association between the ULX CXOU J230453.0+121959 and SN 2009jf was already suggested by @ref26. We also consider it possible that CXOU J120150.4-185212 might be associated with the progenitor of SN 2004gt. @mineo categorized this source as a HMXB based on its relative position within the host galaxy, and although we note that this categorization is purely statistical and has not been confirmed observationally, it does suggest that the source can plausibly be considered a HMXB progenitor candidate for the SN. The lack of post-explosion data for SN 2004gt prevents the determination of any variability in this source, however. Given that over a decade has passed since the explosion of SN 2004gt, the SN may have faded enough that it might be possible in the near-future to measure this source again for possible variability, although given the relative faintness of the source in the pre-explosion data (with luminosity $\sim5\times10^{37}$ erg s$^{-1}$) and the overall complexity of the region, this may be difficult. The third detection is at the location of SN 2010O, which contains a complex system of multiple variable sources. There is a possibility that SN 2010O has an X-ray bright progenitor as suggested by @ref2, although as noted previously, the complexity of the region makes associating any particular source with the SN difficult. All three of the aforementioned potential associations between X-ray sources and their corresponding supernovae remain uncertain, however.
One SN in our sample, SN 2008ax is known to not have a HMXB progenitor from optical pre-explosion data. However, for the 14 other SNe, X-ray luminosity upper limits were established, constraining the properties of any progenitor systems. Although all possible HMXB progenitor systems cannot be ruled out based on these limits, some of the most luminous cases can be excluded. Specifically, a ULX progenitor could be ruled out for 9 out of 14 SNe in the sample. For one of these SNe, SN 2011dh, optical pre-explosion observations had previously identified a yellow supergiant progenitor in a binary system. However, the recent observations by [@maund_2015] suggest that the companion to the progenitor is either fainter than previously believed, or may not exist at all. As the nature of the companion, if one exists, is uncertain, SN 2011dh makes an interesting target for our study. The relatively low upper limit of X-ray luminosity evident from our results could be taken as an indication that this SN did not have a HMXB progenitor.
We also compared the upper limits to the luminosities of the three known (and one candidate) Wolf-Rayet X-ray binaries, a type of HMXB likely to produce a stripped-envelope CCSN. The known Wolf-Rayet HMXBs tend to be more X-ray luminous than HMXBs on average: NGC 300 X-1, IC 10 X-1 and CXOU J004732 all have X-ray luminosities that are higher than $10^{38}$ erg s$^{-1}$, whereas the majority of HMXBs appear to be less luminous [see e.g. @mineo]. However, the case of Cygnus X-3, the only known WR X-ray binary in our own galaxy, demonstrates both that much fainter systems of this nature can exist and that their luminosity can be greatly variable depending on the state of the system. In general, given the relative rarity of WR-HMXBs, it is quite likely that there is an observational bias favoring the detection of only these brighter systems. The upper limits obtained in this study are mostly comparable to the average luminosity of the known WR-HMXBs, suggesting that such systems could in principle have been detected.
One problem inherent in interpreting pre-explosion X-ray observations is that it is not known whether or not accretion from the progenitor to the compact companion continues all the way to the explosion, and if not, how long before the core-collapse it would shut down. Because this accretion is the source of X-ray flux from the assumed progenitor system, it is possible that the system has ceased to be a luminous X-ray source before the SN on a timescale which would make pre-explosion detection impractical or impossible. Our BPASS-modelling discussed earlier indicates that, while a significant number of stripped-envelope SN progenitors can be expected to be found in binary systems, only a small fraction of these would contain a compact companion at a binary separation where substantial mass-transfer could be expected to occur. HMXB-type progenitors would therefore be quite rare. The possibility of observing X-rays from non-HMXB binary progenitors remains, but such objects would most likely be relatively faint (comparable to a WR-star), and well below the threshold found in this study.
One way to constrain the timeframe in which a progenitor system might become X-ray bright is to observe HMXBs that are associated with supernova remnants (SNR). SNRs are relatively short-lived, so this would help establish how soon after the supernova accretion might be initiated in a newly formed HMXB. Several such systems have been identified, for example SXP 1062 [@SNR_brunet] and DEM L241 [@SNR_seward], both of which have an estimated age of several times $10^{4}$ years. Another possible case is Circinus X-1, for which @SNR_heinz derive an age upper-limit of $\sim 4600$ years. While none of these systems are X-ray luminous enough to allow detection by our study, these examples indicate that accretion in a newly-formed HMXB might begin on a timescale that is less than a few thousand years after the SN. Another example to consider is the association between SNR W50 and SS 433, although it is estimated to be older than the aforementioned cases [see eg. @SNR_goodall].
X-ray binaries are not the only type of pre-explosion X-ray source that might be associated with a SN. Colliding wind binaries (CWB), such as the luminous blue variable (LBV) binary Eta Carinae, or the Wolf-Rayet binary WR 48a, are known to have X-ray luminosities up to $\sim 10^{35}$ erg s$^{-1}$ [see e.g. @etacar; @lbv_sur; @zhekov]. Compared to the upper limits we derived, such objects are an order of magnitude fainter than what could reasonably be detected with *Chandra*, unless they were significantly closer. Also, given that eg. Eta Carinae is known to have a fairly hard X-ray spectrum [eg. @hamaguchi], they would likely be difficult to distinguish from HMXBs based on their spectrum alone. Future studies are therefore unlikely to find such SN progenitors using this method.
Conclusions {#sect_conclusions}
===========
The aim of this study was to examine the possibility of high-mass X-ray binaries being the progenitors to some stripped-envelope CCSNe with a direct observational test, or to constrain the possibility of such progenitors in the sample. If a SN progenitor was part of a HMXB before the explosion, the system could be an X-ray source due to accretion from the progenitor star to the compact object. In this study we note the existence of two pre-explosion sources that could potentially be associated with supernovae. The first of these is the ULX CXOU J230453.0+121959 at the position of the type Ib SN 2009jf [@ref26], with luminosity $\sim1\times10^{40}$ erg s$^{-1}$. The second candidate is CXOU J120150.4-185212, at the position of the type Ic SN 2004gt, with luminosity $\sim5\times10^{37}$ erg s$^{-1}$. For both of these, however, it is important to note that chance co-alignment is possible. We also measure the unabsorbed pre-explosion upper limits of X-ray luminosity for the positions of 14 stripped-envelope CCSNe. Most of the upper limits obtained are comparable to the luminosities of the two known and one candidate extragalactic WR-HMXBs, showing that at least more luminous examples of such systems could in principle have been detected. Therefore, while detecting a candidate progenitor in this way is possible, the existence of one cannot be completely ruled out without using significantly deeper exposure times or observing much more nearby SNe. The main limiting factor to future studies using this method is the scarcity of deep pre-explosion observations. We found that such *Chandra* data exists only for $\sim6$ percent of SNe within $\sim$100 Mpc distance. Furthermore, our modelling indicates that HMXB-progenitor systems would likely be quite rare. However, these numbers could be increased if mass-transfer between the binary components is more efficient or supernova kicks are weaker than assumed. The long X-ray luminosity decline time of most SNe of these types [see e.g. @ref15] limits the usefulness of post-explosion observations for recent SNe but could potentially allow the confirmation of candidate HMXB progenitors, once the SN has faded sufficiently.
Acknowledgements {#sect_acnowledgements .unnumbered}
================
We thank the referee for their helpful comments and feedback. This research has made use of data obtained from the Chandra Data Archive and the Chandra Source Catalog, and software provided by the Chandra X-ray Center (CXC) in the application packages CIAO, ChIPS, and Sherpa. This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. This research has made use of the VizieR catalogue access tool, CDS, Strasbourg, France. [@vizier] We also acknowledge the use of the Hyperleda database (http://leda.univ-lyon1.fr). The plots in this paper were made using matplotlib [@matplotlib], http://matplotlib.org. This research has made use of NASA’s Astrophysics Data System. This research was partly supported by the European Union FP7 programme through ERC grant number 320360.
\[lastpage\]
[^1]: According to a study by @ref4 (based on a sample of 100 nearby SNe) 26 percent of CCSNe are of type Ibc and 12 percent type IIb in a volume limited sample (the remaining 62 percent are non-stripped-envelope CCSNe such as type IIP).
[^2]: For details, see http://cxc.harvard.edu/cal/ASPECT/celmon/
[^3]: http://www.rochesterastronomy.org/snimages/
[^4]: Differences less than $\sim$1 arcsec between the catalogs were not considered significant. Overall, the more recent USC was considered more accurate, and in case of discrepancies, if more accurate astrometry was not available we adopted the USC position instead.
[^5]: Asiago catalog max. epoch: The date of optical maximum or discovery.
[^6]: The default values for cross-sections (xsect) and abundances (abund) were used, these being bcmc and angr, respectively.
[^7]: http://ned.ipac.caltech.edu/
[^8]: @refleda: See http://atlas.obs-hp.fr/hyperleda/
[^9]: The average photon energies and absorption correction factors were scaled accordingly for the 1.5-8.0 keV energy range, with other parameters being the same as in Table \[models\].
[^10]: Both calculated from the same nuclear offsets, possibly based on @1983I_disc, which may have a high uncertainty.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We calculate the branching ratio and angular distributions of hydrogen of the bound–state $\beta^-$–decay of the neutron, induced by left–handed and right–handed hadronic and leptonic currents. The branching ratio of the bound–state $\beta^-$–decay is calculated by taking into account radiative corrections. We show that the bound-state $\beta^-$–decay can be hardly a good laboratory for experimental investigations of electroweak models with left–right symmetries.'
author:
- 'A. N. Ivanov'
- 'M. Pitschmann'
- 'N. I. Troitskaya'
- 'Ya. A. Berdnikov'
bibliography:
- 'RevND2mp.bib'
title: 'The Bound-State $\beta^-$–Decay of the Neutron Revisited'
---
Introduction {#sec:introduction}
============
In Ref. [@Faber:2009ts] the bound-state $\beta^-$–decay of the neutron $n \to {\rm H} + \bar{\nu}_e$, where ${\rm H}$ is hydrogen, has been revised by taking into account the new value of the axial coupling constant $\lambda = - 1.2750(9)$ [@Abele:2008zz; @Ivanov:2012qe] as well as new effective scalar and tensor weak lepton–nucleon interactions. The amplitude of the continuum-state $\beta^-$–decay of the neutron, calculated in the rest frame of the neutron in the non–relativistic approximation for the proton, including radiative corrections by virtual one–$\gamma$, $W$ and $Z$–boson exchanges as well as QCD corrections [@Sirlin:1967zza; @Sirlin:1974ni; @Marciano:1974tv; @Sirlin:1977sv; @Sirlin:1981ie; @Marciano:1985pd; @Czarnecki:2004cw; @Marciano:2005ec; @Czarnecki:2007th], is given by [@Ivanov:2012qe] $$\begin{aligned}
\label{label1}
&& \hspace{-5mm}M(n \to p e^- \bar{\nu}_e) =
-\,2m_n\,\frac{G_F}{\sqrt{2}}\,V_{ud}\,\Big\{[\varphi^{\dagger}_p\varphi_n][\bar{u}_e
\gamma^0(C_V + \bar{C}_V \gamma^5) v_{\bar{\nu}_e}]\Big(1 +
\frac{\alpha}{2\pi}\,f_{\beta^-_c}(E_e, \mu)\Big) \nonumber\\ &&-
[\varphi^{\dagger}_p\vec{\sigma}\,\varphi_n]\cdot [\bar{u}_e
\vec{\gamma}\,(\bar{C}_A + C_A
\gamma^5)v_{\bar{\nu}_e}]\Big(1 +
\frac{\alpha}{2\pi}\,f_{\beta^-_c}(E_e, \mu)\Big) +
[\varphi^{\dagger}_p\varphi_n][\bar{u}_e (C_S + \bar{C}_S
\gamma^5) v_{\bar{\nu}_e}]\nonumber\\ && +
[\varphi^{\dagger}_p\vec{\sigma}\,\varphi_n]\cdot
[\bar{u}_e \gamma^0 \vec{\gamma}\,(\bar{C}_T + C_T
\gamma^5) v_{\bar{\nu}_e}] -
\frac{\alpha}{2\pi}\,g_F(E_e)\,[\varphi^{\dagger}_p
\varphi_n][\bar{u}_e\,(1 -
\gamma^5)v_{\bar{\nu}_e}]\nonumber\\ && +
\frac{\alpha}{2\pi}\,\lambda\, g_F(E_e)
[\varphi^{\dagger}_p \vec{\sigma}\,\varphi_n]\cdot
[\bar{u}_e \gamma^0\vec{\gamma}\,(1 -
\gamma^5)v_{\bar{\nu}_e}]\Big\},\end{aligned}$$ where the coupling constants $C_j$ and $\bar{C}_j$ for $j = V$, $A$, $S$ and $T$ describe effective weak interactions [@Jackson:1957zz; @Jackson1957206; @Severijns:2006dr], which may be induced by left–handed and right–handed hadronic and leptonic currents (see also [@Ivanov:2012qe]) caused by interactions beyond the Standard model (SM) such as supersymmetric interactions [@RamseyMusolf:2006vr] and so on. They are related to the coupling constants, analogous to those which were introduced by Herczeg [@Herczeg:2001vk], as follows (see also Eq. (2) in Appendix G of Ref. [@Ivanov:2012qe]) $$\begin{aligned}
\label{label2}
C_V &=&1 + a^h_{LL} + a^h_{LR} + a^h_{RR} + a^h_{RL},\nonumber\\
\bar{C}_V &=& - 1 - a^h_{LL} -
a^h_{LR} + a^h_{RR} + a^h_{RL},\nonumber\\
C_A &=& -\lambda + a^h_{LL} - a^h_{LR} + a^h_{RR} -
a^h_{RL},\nonumber\\
\bar{C}_A &=&\lambda - a^h_{LL} + a^h_{LR} + a^h_{RR} -
a^h_{RL},\nonumber\\
C_S &=& A^h_{LL} + A^h_{LR} + A^h_{RR} +
A^h_{RL},\nonumber\\
\bar{C}_S &=& - A^h_{LL} -
A^h_{LR} + A^h_{RR} + A^h_{RL},\nonumber\\
C_T &=& 2( \alpha^h_{LL} + \alpha^h_{RR}),\nonumber\\
\bar{C}_T &=& 2( -
\alpha^h_{LL} + \alpha^h_{RR}),\end{aligned}$$ where the index $h$ means that the coupling constants are introduced at the [*hadronic*]{} level [@Ivanov:2012qe] but not at the [ *quark*]{} level as it was done by Herczeg [@Herczeg:2001vk]. In addition, in comparison with Herczeg [@Herczeg:2001vk] we have taken away the common factor $G_F V_{ud}/\sqrt2$ and defined the coupling constants $a_{LL}^h$ and $a_{LR}^h$ as deviations from the coupling constants of the SM.[^1] The SM weak interactions are defined by coupling constants $C_V = - \bar{C}_V = 1$, $C_A = - \bar{C}_A = -
\lambda$ and $C_S = \bar{C}_S = C_T = \bar{C}_T = 0$. The functions $f_{\beta^-_c}(E_e, \mu)$ and $g_F(E_e)$ in Eq. (\[label1\]) are equal to [@Ivanov:2012qe] $$\begin{aligned}
\label{label3}
f_{\beta^-_c}(E_e,\mu) &=& \frac{3}{2}\,{\ell
n}\Big(\frac{m_p}{m_e}\Big) - \frac{11}{8} + {\ell
n}\Big(\frac{\mu}{ m_e}\Big)\,\Big[\frac{1}{\beta}\,{\ell
n}\Big(\frac{1 + \beta}{1 - \beta}\Big) - 2 \Big] +
\frac{1}{\beta}\,L\Big(\frac{2\beta}{1 + \beta}\Big) \nonumber\\ &&-
\frac{1}{4\beta}\,{\ell n}^2\Big(\frac{1 + \beta}{1 - \beta}\Big) +
\frac{1}{2\beta}\,{\ell n}\Big(\frac{1 + \beta}{1 - \beta}\Big) +
C_{WZ},\nonumber\\ g_F(E_e) &=& \frac{\sqrt{1 - \beta^2}}{2
\beta}\,{\ell n}\Big(\frac{1 + \beta}{1 - \beta}\Big),\end{aligned}$$ where $m_p$ and $m_e$ are the proton and electron masses, $\mu$ is a finite-photon mass regularisation parameter for the regularisation of infrared divergences in virtual one–photon exchanges [@Sirlin:1967zza; @Sirlin:1974ni; @Marciano:1974tv; @Sirlin:1977sv; @Sirlin:1981ie; @Marciano:1985pd; @Czarnecki:2004cw; @Marciano:2005ec; @Czarnecki:2007th] (see also [@Ivanov:2012qe]), $\alpha = 1/137.036$ is the fine–structure constant [@Beringer:1900zz], $\beta = \sqrt{E^2_e - m^2_e}/E_e$ is the electron velocity and $L(2\beta/(1+\beta))$ is the Spence function [@abramowitz+stegun] (see also [@Ivanov:2012qe]). The constant $C_{WZ} = 10.249$ is caused by electroweak boson exchanges and QCD corrections [@Sirlin:1967zza; @Sirlin:1974ni; @Marciano:1974tv; @Sirlin:1977sv; @Sirlin:1981ie; @Marciano:1985pd; @Czarnecki:2004cw; @Marciano:2005ec; @Czarnecki:2007th] (see also Appendix D in Ref. [@Ivanov:2012qe]).
As has been pointed out in [@Bhattacharya:2011qm; @Cirigliano:2009wk; @Cirigliano:2012ab] (see also [@Ivanov:2012qe]), the contributions of interactions beyond the SM from the coupling constants $a^h_{LL}$ and $a^h_{LR}$ can be absorbed by a redefined axial coupling constant $\lambda_{\rm eff}$ and CKM matrix element $(V_{ud})_{\rm eff}$ [@Ivanov:2012qe], i.e. $$\begin{aligned}
\label{label4}
&&\lambda \to \lambda_{\rm eff} = \frac{\lambda -
a^h_{LL} + a^h_{LR}}{1 + a^h_{LL} + a^h_{LR}},\nonumber\\
&&V_{ud} \to (V_{ud})_{\rm eff} = V_{ud}\,(1 + a^h_{LL} +
a^h_{LR}).\end{aligned}$$ As has been shown in [@Ivanov:2012qe] the axial coupling constant $\lambda_{\rm eff}$ is real at the level of order $10^{-4}$. After such a redefinition the phenomenological coupling constants $C_j$ and $\bar{C}_j$ for $j = V$, $A$, $S$ and $T$ become $$\begin{aligned}
\label{label5}
C_{V,{\rm eff}} &=&1 + \frac{a^h_{RR} + a^h_{RL}}{1 + a^h_{LL} +
a^h_{LR}}= 1 + \bar{a}^h_{RR} +
\bar{a}^h_{RL},\nonumber\\ \bar{C}_{V,{\rm eff}} &=& - 1 +
\frac{a^h_{RR} + a^h_{RL}}{1 + a^h_{LL} + a^h_{LR}} = -1 +
\bar{a}^h_{RR} + \bar{a}^h_{RL},\nonumber\\ C_{A,{\rm eff}} &=&
-\lambda_{\rm eff} + \frac{a^h_{RR} - a^h_{RL}}{1 + a^h_{LL} +
a^h_{LR}} = -\lambda_{\rm eff} + \bar{a}^h_{RR} -
\bar{a}^h_{RL},\nonumber\\ \bar{C}_{A,{\rm eff}} &=& \lambda_{\rm eff}
+ \frac{a^h_{RR} - a^h_{RL}}{1 + a^h_{LL} + a^h_{LR}} = \lambda_{\rm
eff} + \bar{a}^h_{RR} - \bar{a}^h_{RL},\nonumber\\ C_{S,{\rm eff}}
&=& \frac{A^h_{LL} + A^h_{LR} + A^h_{RR} + A^h_{RL}}{1 + a^h_{LL} +
a^h_{LR}} = \bar{A}^h_{LL} + \bar{A}^h_{LR} + \bar{A}^h_{RR} +
\bar{A}^h_{RL},\nonumber\\ \bar{C}_{S,{\rm eff}} &=& \frac{-
A^h_{LL} - A^h_{LR} + A^h_{RR} + A^h_{RL}}{1 + a^h_{LL} +
a^h_{LR}} = -\bar{A}^h_{LL} - \bar{A}^h_{LR} + \bar{A}^h_{RR} +
\bar{A}^h_{RL},\nonumber\\ C_{T,{\rm eff}} &=&
2\,\frac{\alpha^h_{LL} + \alpha^h_{RR}}{1 + a^h_{LL} + a^h_{LR}} =
2\,(\bar{\alpha}^h_{LL} +
\bar{\alpha}^h_{RR}),\nonumber\\ \bar{C}_{T,{\rm eff}} &=& 2\,\frac{
- \alpha^h_{LL} + \alpha^h_{RR}}{1 + a^h_{LL} + a^h_{LR}}=
2\,(-\bar{\alpha}^h_{LL} + \bar{\alpha}^h_{RR}).\end{aligned}$$ Since the velocity of the electron in the hydrogen bound state with principal number $n$ is equal to $\beta = \alpha/n$, we take the non-relativistic limit for the electron Dirac spinor in the calculation of the bound-state $\beta^-$–decay of the neutron.
Bound–state $\beta^-$–decay of neutron and left–handed neutrinos {#sec:lefthand}
================================================================
For the calculation of the amplitude of the bound–state $\beta^-$–decay we use the following Dirac wave function for the antineutrino $$\begin{aligned}
\label{label6}
v_{\bar{\nu}_e} = \sqrt{E}\left(\begin{array}{c}
(\vec{\sigma}\cdot \vec{n}\,)\chi_{\bar{\nu}_e}
\\ \chi_{\bar{\nu}_e}
\end{array}\right),\end{aligned}$$ where $\vec{n} = \vec{k}/E$ and normalisation equals $v^{\dagger}_{\bar{\nu}_e}v_{\bar{\nu}_e} = 2 E$. For the right–handed polarisation states of antineutrinos, corresponding to the left–handed polarisation states of neutrinos, the Pauli wave function $\chi_{\bar{\nu}_e}$ obeys the equation $(\vec{\sigma}\cdot
\vec{n}\,)\chi_{\bar{\nu}_e} = - \chi_{\bar{\nu}_e}$[^2]. If the axis of the antineutrino–spin quantisation is inclined relative to the axis of the neutron–spin quantisation with a polar angle $\vartheta$, the Pauli wave function $\chi_{\bar{\nu}_e}$ is given by [@Faber:2009ts] $$\begin{aligned}
\label{label7}
\chi_{\bar{\nu}_e} = \left(\begin{array}{c} {\displaystyle - e^{\,- i
\varphi}\sin\frac{\vartheta}{2}} \\ {\displaystyle
\cos\frac{\vartheta}{2}}
\end{array}\right),\end{aligned}$$ where $\varphi$ is an azimuthal angle. Keeping the leading order contributions in the $\alpha$–expansion of the amplitude Eq. (\[label1\]) and following [@Faber:2009ts], we obtain the amplitude of the transition $n \to {\rm H} + \bar{\nu}_e$, where hydrogen is in the hyperfine $(ns)_F$ state with hyperfine spin $F$, in the form $$\begin{aligned}
\label{label8}
&&M(n\to {\rm H} + \bar{\nu}_e)_{\rm rhps} = G_F (V_{ud})_{\rm eff}\sqrt{2
m_n 2 E_{\rm H} 2 E}\,\Big(1 + \frac{\alpha}{2\pi}\,(f_{\beta^-_c}
- 1)\Big) \nonumber\\
&&\quad\times\Big\{(1 + g_S)\, [\varphi^{\dagger}_{p}
\varphi_n]\,[\varphi^{\dagger}_e \chi_{_{\bar{\nu}_e}}] + (\lambda_{\rm eff} + g_T )\,
[\varphi^{\dagger}_p\vec{\sigma}\,\varphi_n] \cdot
[\varphi^{\dagger}_e\vec{\sigma}\,\chi_{_{\bar{\nu}_e}}]
\Big\}\,\psi^*_{(ns)_F}(0),\end{aligned}$$ where the abbreviation “rhps” means the [*right–handed polarisation state*]{}. Then, $\varphi_j$ for $j = p, n, e$ and $\chi_{\bar{\nu}_e}$ are Pauli spinorial functions of the proton, neutron, electron and antineutrino, respectively. The term $(-\alpha/2\pi)$ is the contribution of those terms in Eq. (\[label1\]), which are proportional to $(\alpha/2\pi)\,g_F(E_e)$. Furthermore, $(f_{\beta^-_c} - 1)$ and the effective coupling constants $g_S$ and $g_T$ are given by $$\begin{aligned}
\label{label9}
f_{\beta^-_c} - 1 = \frac{3}{2}\,{\ell
n}\Big(\frac{m_p}{m_e}\Big) - \frac{27}{8} + C_{WZ},\end{aligned}$$ and $$\begin{aligned}
\label{label10}
g_S &=& \frac{1}{2}\Big((C_{S,{\rm eff}} - \bar{C}_{S,{\rm eff}}) +
(C_{V,{\rm eff}} - 1) - (\bar{C}_{V,{\rm eff}} + 1)\Big) =
\bar{A}^h_{LL} + \bar{A}^h_{LR},\nonumber\\ g_T &=& \frac{1}{2}\Big(
(C_{T,{\rm eff}} - \bar{C}_{T,{\rm eff}}) - (C_{A,{\rm eff}} +
\lambda_{\rm eff}) + (\bar{C}_{A,{\rm eff}} - \lambda_{\rm eff})\Big)
= 2 \bar{\alpha}^h_{LL}.\end{aligned}$$ Note that the right–handed leptonic currents with the left- and right–handed hadronic currents do not contribute to the effective coupling constants $g_S$ and $g_T$.
As has been shown in [@Ivanov:2012qe], the effective coupling constant $\lambda_{\rm eff}$ is real with an accuracy better than $10^{-4}$. Following [@Faber:2009ts] and keeping only the linear terms in the expansion in powers of $g_S$ and $g_T$ we obtain the branching ratio of the bound-state $\beta^-$–decay of the neutron $$\begin{aligned}
\label{label11}
R_{\beta^-_b} = \Big(1 + \frac{2}{1 + 3 \lambda^2_{\rm eff}}\,{\rm Re}(g_S + 3
\lambda_{\rm eff} g_T) - \Big\langle \frac{m_e}{E_e}\Big\rangle_{\rm
SM}\,b_F\Big)\,R_{\rm SM},\end{aligned}$$ where $\langle m_e/E_e\rangle_{\rm SM} = 0.6556$, averaged over the electron–energy density spectrum of the neutron $\beta^-$–decay [@Ivanov:2012qe]. The Fierz term $b_F$, defined to linear approximation with respect to the Herczeg coupling constants [@Ivanov:2012qe], is given by $$\begin{aligned}
\label{label12}
b_F = \frac{1}{1 + 3 \lambda^2_{\rm eff}}\,{\rm Re}\Big((C_{S,{\rm
eff}} -\bar{C}_{S,{\rm eff}}) + 3\lambda_{\rm eff} (C_{T,{\rm
eff}} - \bar{C}_{T,{\rm eff}})\Big) = \frac{2}{1 + 3
\lambda^2_{\rm eff}}\,{\rm Re}\big(g_S + 3\lambda_{\rm eff}
g_T\big).\end{aligned}$$ Hence, we may write the branching ratio $R$ in the form $$\begin{aligned}
\label{label13}
R_{\beta^-_b} = \Big(1 + \Big(1 -\Big\langle
\frac{m_e}{E_e}\Big\rangle_{\rm SM}\Big)\,b_F\Big)\,R_{\rm SM}.\end{aligned}$$ As has been pointed out in [@Ivanov:2012qe], the Fierz term can be measured from the experimental data on the electron asymmetry $A_{\exp}(E_e)$ of correlations between the neutron spin and the electron 3–momentum and the proton–energy spectrum $a(T_p)$ (see also [@Ivanov:2013fca]), related to correlations between the 3–momenta of the proton and electron. The branching ratio $R_{\rm
SM}$, calculated in the SM, is equal to $$\begin{aligned}
\label{label14}
R_{\rm SM} = 2\pi\alpha^3\zeta(3)\,\Big(1 +
\frac{\alpha}{\pi}\,(f_{\beta^-_c} - 1)\Big)\,\frac{m_p +
m_e}{m_n}\,\frac{E^2}{m^2_e f_n}\,\sqrt{1 +
\frac{E^2}{(m_p + m_e)^2}} = 3.905\times 10^{-6}.\end{aligned}$$ where $\zeta(3) = 1.202$, $E = (m^2_n - (m_p + m_e)^2)/2m_n =
0.782\,{\rm MeV}$ and $f_n= 1.755$ are the Riemann zeta function [@abramowitz+stegun], the antineutrino energy and the phase–space factor of the neutron $\beta^-$–decay rate, including the contributions of the corrections, caused by the “weak magnetism” and the proton recoil as well as radiative corrections [@Ivanov:2012qe], respectively.
The contributions of different spinorial states to the helicity amplitudes of the bound-state $\beta^-$–decay as functions of the angles $\vartheta$ and $\varphi$ are given in Table I.
---------------------------------------------------------------------------------------------------------------------------------------------------
$\sigma_n$ $\sigma_p$ $\sigma_e$ $\sigma_{\bar{\nu}_e}$ $f$
----------------- ----------------- ----------------- ------------------------ --------------------------------------------------------------------
$+\frac{1}{2} $ $+\frac{1}{2} $-\frac{1}{2} $ $+\frac{1}{2} $ $(1 + g_S - \lambda_{\rm eff} -
$ g_T)\,\cos\frac{\vartheta}{2}$
$+\frac{1}{2} $ $+\frac{1}{2} $ $+\frac{1}{2} $ $+\frac{1}{2} $ $-(1 + g_S +
\lambda_{\rm eff} + g_T)\,e^{\,- i\varphi}\sin\frac{\vartheta}{2}$
$+\frac{1}{2} $ $-\frac{1}{2} $ $-\frac{1}{2} $ $+\frac{1}{2} $ $0$
$+\frac{1}{2} $ $-\frac{1}{2} $ $+\frac{1}{2} $ $+\frac{1}{2} $ $ 2(\lambda_{\rm eff} + g_T)\,
\cos\frac{\vartheta}{2}$
$-\frac{1}{2} $ $+\frac{1}{2} $ $-\frac{1}{2} $ $+\frac{1}{2} $ $ - 2 (\lambda_{\rm eff} +
g_T)\,e^{\,- i\varphi}\sin\frac{\vartheta}{2}$
$-\frac{1}{2} $+\frac{1}{2} $ $+\frac{1}{2} $ $+\frac{1}{2} $ $0$
$
$-\frac{1}{2} $ $-\frac{1}{2} $ $-\frac{1}{2} $ $+\frac{1}{2} $ $(1 + g_S + \lambda_{\rm eff} + g_T)\,\cos\frac{\vartheta}{2}$
$-\frac{1}{2} $ $-\frac{1}{2} $ $+\frac{1}{2} $ $+\frac{1}{2} $ $-(1 + g_S - \lambda_{\rm eff} - g_T)\,e^{\,-
i\varphi}\sin\frac{\vartheta}{2}$
---------------------------------------------------------------------------------------------------------------------------------------------------
: The contributions of different spinorial states of the interacting particles to the amplitudes of the bound-state $\beta^-$–decay of the neutron and the antineutrino in the state with the wave function Eq. (\[label7\]); $f$ is defined by $f = (1
+
g_S)[\varphi^{\dagger}_e\chi_{_{\bar{\nu}_e}}][\varphi^{\dagger}_p
\varphi_n] + (\lambda_{\rm eff} + g_T)
[\varphi^{\dagger}_e\vec{\sigma}\,\chi_{_{\bar{\nu}_e}}]\cdot [
\varphi^{\dagger}_p\vec{\sigma}\,\varphi_n]$.
Following [@Faber:2009ts] we define the angular distributions of the production of hydrogen in the hyperfine states with $F = 0$ and $F =
1$ in the bound-state $\beta^-$–decay of the polarised neutron $$\begin{aligned}
\label{label15}
&&4\pi \frac{d W^{(\pm)}_{F = 0}(\theta)}{d \Omega} =
\Big(\frac{1}{8}\,\frac{(1 - 3 \lambda_{\rm eff})^2}{1 + 3
\lambda^2_{\rm eff}} + \frac{3}{4}\frac{(1 + \lambda_{\rm eff})(1
- 3\lambda_{\rm eff})}{(1 + 3 \lambda^2_{\rm eff})^2}\,{\rm
Re}(\lambda_{\rm eff}\,g_S - g_T)\Big)\,(1 \mp \cos\theta),
\nonumber\\ &&4\pi \frac{d W^{(\pm)}_{F = 1}(\theta)}{d \Omega} =
\Big(\frac{3}{8}\, \frac{(1 + \lambda_{\rm eff})^2}{1 + 3
\lambda^2_{\rm eff}}- \frac{3}{4}\frac{(1 + \lambda_{\rm eff})(1 -
3\lambda_{\rm eff})}{(1 + 3\lambda^2_{\rm eff})^2}\,{\rm
Re}(\lambda_{\rm eff}\,g_S - g_T)\Big)\,\Big(1 \pm
\frac{1}{3}\,\cos\theta\Big). \nonumber\\\end{aligned}$$ The upper indices $(\pm)$ in $W^{(\pm)}_F$ correspond to the neutron polarisation, $\theta = \pi - \vartheta$ is an angle between the neutron polarisation and the 3–momentum of hydrogen.
From Table I we define the angular distributions of the probabilities $W^{(\pm)}_1(\theta)$, $W^{(\pm)}_2(\theta)$, $W^{(\pm)}_3(\theta)$ and $W^{(\pm)}_4(\theta)$ of the neutron decay into the bound $(p
e^-)$ spinorial states $|+1/2\rangle_p |+ 1/2\rangle_e$, $|+
1/2\rangle_p |- 1/2\rangle_e$, $|- 1/2\rangle_p |+ 1/2\rangle_e$ and $|- 1/2\rangle_p |- 1/2\rangle_e$ (see [@Byrne:2001sj]), respectively. For the neutron spin polarisation $\sigma_n = + 1/2$ the angular distributions of the probabilities $W^{(+)}_1(\theta)$, $W^{(+)}_2(\theta)$, $W^{(+)}_3(\theta)$ and $W^{(+)}_4(\theta)$ are equal to $$\begin{aligned}
\label{label16}
&&4\pi \frac{d W_1^{(+)}(\theta)}{d \Omega} =
\Big(\frac{1}{2}\,\frac{(1 + \lambda_{\rm eff})^2}{1 + 3
\lambda^2_{\rm eff}} - \frac{(1 +
\lambda_{\rm eff})(1 - 3\lambda_{\rm eff})}{(1 + 3 \lambda^2_{\rm eff})^2}\,{\rm Re}(\lambda_{\rm eff}\,g_S -
g_T)\Big)\,(1 + \cos\theta), \nonumber\\
&&4\pi \frac{d W_2^{(+)}(\theta)}{d \Omega} =
\Big(\frac{1}{2}\, \frac{(1 - \lambda_{\rm eff})^2}{1 + 3
\lambda^2_{\rm eff}} + \frac{(1 - \lambda_{\rm eff})(1 +
3\lambda_{\rm eff})}{(1 + 3\lambda^2_{\rm eff})^2}\,{\rm Re}(\lambda_{\rm eff}\,g_S -
g_T)\Big)\,(1 - \cos\theta),\nonumber\\
&&4\pi \frac{d W_3^{(+)}(\theta)}{d \Omega} = \Big(
\frac{2\lambda^2_{\rm eff}}{1 + 3 \lambda^2_{\rm eff}}- \frac{4
\lambda_{\rm eff}}{(1 + 3\lambda^2_{\rm eff})^2}\,{\rm
Re}(\lambda_{\rm eff}\,g_S - g_T)\Big)\,(1 -
\cos\theta),\nonumber\\
&&4\pi \frac{d W_4^{(+)}(\theta)}{d \Omega} =0.\end{aligned}$$ For the neutron spin polarisation $\sigma_n = -1/2$ the angular distributions of the probabilities $W^{(-)}_1(\theta)$, $W^{(-)}_2(\theta)$, $W^{(-)}_3(\theta)$ and $W^{(-)}_4(\theta)$ are $$\begin{aligned}
\label{label17}
\frac{d W^{(-)}_1(\theta)}{d \Omega} &=& \frac{d W^{(+)}_4(\pi -
\theta)}{d \Omega},\nonumber\\ \frac{d W^{(-)}_2(\theta)}{d \Omega}
&=& \frac{d W^{(+)}_3(\pi - \theta)}{d \Omega},\nonumber\\ \frac{d
W^{(-)}_3(\theta)}{d \Omega} &=& \frac{d W^{(+)}_2(\pi - \theta)}{d
\Omega},\nonumber\\ \frac{d W^{(-)}_4(\theta)}{d \Omega} &=& \frac{d
W^{(+)}_1(\pi - \theta)}{d \Omega}.\end{aligned}$$ In comparison with [@Faber:2009ts] we have expanded the angular distributions in powers of the Fierz term $b_F$ and have kept only the linear contributions. As result, we have obtained that the bound–state $\beta^-$–decay of a free neutron is sensitive to the effective coupling constant ${\rm Re}(\lambda_{\rm eff}\,g_S - g_T)$ only, which contains no information about the right–handed leptonic currents.
Bound–state $\beta^-$–decay of neutron and right–handed neutrinos {#sec:righthand}
=================================================================
The absence of contributions of right–handed leptonic currents in the probabilities of the bound–state $\beta^-$–decay, calculated in section \[sec:lefthand\], is not a surprise, because we have used there the wave functions of antineutrinos in the right–handed polarisation state, corresponding to the left–handed polarisation state of neutrinos. Being multiplied by the projection operator $P_R =
(1 + \gamma^5)/2$, appearing in the right–handed leptonic currents, the wave function of antineutrinos $v_{\bar{\nu}_e}$ in the right–handed polarisation state with $(\vec{\sigma}\cdot \vec{n}\,)
\chi_{\bar{\nu}_e} = - \chi_{\bar{\nu}_e}$ gives a vanishing contribution, i.e. $P_R v_{\bar{\nu}_e} = 0$.
In this section, we assume that antineutrinos can have also left–handed polarisation state that is possible if antineutrinos (neutrinos) are massive. According to [@Beringer:1900zz], a mass of the electron neutrino (antineutrino) should not exceed a few ${\rm
eV}$. Since in the bound–state $\beta^-$–decay $E = Q_{\beta^-_c}
= 0.782\,{\rm MeV}$ [@Faber:2009ts], one can neglect the antineutrino mass with respect to the antineutrino energy $E$ and use the Dirac wave function Eq. (\[label6\]). However, the Pauli wave function $\chi_{\bar{\nu}_e}$ of antineutrinos in the left–handed polarisation state should obey the equation $(\vec{\sigma}\cdot
\vec{n}\,)\chi_{\bar{\nu}_e} = + \chi_{\bar{\nu}_e}$[^3]. If the axis of the antineutrino–spin quantisation is inclined relative to the axis of the neutron–spin quantisation with a polar angle $\vartheta$, the Pauli wave function $\chi_{\bar{\nu}_e}$ is given by $$\begin{aligned}
\label{label18}
\chi_{\bar{\nu}_e} = \left(\begin{array}{c} {\displaystyle
\cos\frac{\vartheta}{2}} \\ {\displaystyle e^{\,+ i
\varphi}\sin\frac{\vartheta}{2}}
\end{array}\right).\end{aligned}$$ The amplitude of the bound–state $\beta^-$–decay of the neutron with antineutrinos in the left–handed polarisation state is given by $$\begin{aligned}
\label{label19}
&&M(n\to {\rm H} + \bar{\nu}_e)_{\rm lhps} = G_F (V_{ud})_{\rm
eff}\sqrt{2 m_n 2 E_{\rm H} 2 E}\,\Big(1 +
\frac{\alpha}{2\pi}\,(f_{\beta^-_c} - 1)\Big)
\nonumber\\ &&\quad\times\Big\{\bar{g}_S\, [\varphi^{\dagger}_{p}
\varphi_n]\,[\varphi^{\dagger}_e \chi_{_{\bar{\nu}_e}}] +
\bar{g}_T\, [\varphi^{\dagger}_p\vec{\sigma}\,\varphi_n] \cdot
[\varphi^{\dagger}_e\vec{\sigma}\,\chi_{_{\bar{\nu}_e}}]
\Big\}\,\psi^*_{(ns)_F}(0),\end{aligned}$$ where the abbreviation “lhps” means the [*left–handed polarisation state*]{}. The coupling constants $\bar{g}_S$ and $\bar{g}_T$ are equal to $$\begin{aligned}
\label{label20}
\bar{g}_S &=& \frac{1}{2}\Big((C_{S,{\rm eff}} + \bar{C}_{S,{\rm
eff}}) + (C_{V,{\rm eff}} + \bar{C}_{V,{\rm eff}})\Big) =
\bar{A}^h_{RR} + \bar{A}^h_{RL} + \bar{a}^h_{RR} +
\bar{a}^h_{RL},\nonumber\\ \bar{g}_T &=& \frac{1}{2}\Big( (C_{T,{\rm
eff}} + \bar{C}_{T,{\rm eff}}) - (C_{A,{\rm eff}} +
\bar{C}_{A,{\rm eff}})\Big) = 2 \bar{\alpha}^h_{RR} - \bar{a}^h_{RR} +
\bar{a}^h_{RL}.\end{aligned}$$ One may see that the coupling constants $\bar{g}_S$ and $\bar{g}_T$ are defined in terms of the contributions of the right–handed leptonic currents and left(right)–handed hadronic currents only. For the calculation of the amplitude Eq. (\[label20\]) we have neglected contributions of order $(\alpha/2\pi)^2 \sim 10^{-6}$, $(\alpha/2\pi)\bar{g}_S$ and $(\alpha/2\pi)\bar{g}_T$. The coupling constants $\bar{a}^h_{RR}$ and $\bar{a}^h_{RL}$ can in principle be induced by exchanges of electroweak $W^{\pm}_R$–bosons, causing effective low–energy current–current interactions $(V + A)_{\rm
leptonic}(V + A)_{\rm hadronic}$ and $(V + A)_{\rm leptonic} (V -
A)_{\rm hadronic}$ [@Beg:1977ti; @Holstein:1977qn; @Carnoy:1988uv]. Of course, the contributions of these interactions can be screened by scalar and tensor interactions with coupling constants $\bar{A}^h_{RR}$, $\bar{A}^h_{RL}$ and $2 \bar{\alpha}^h_{RR}$.
The contributions of different spinorial states to the helicity amplitudes of the bound–state $\beta^-$–decay of the neutron with the antineutrino in the left–handed polarisation state as functions of the angles $\vartheta$ and $\varphi$ are given in Table II.
---------------------------------------------------------------------------------------------------------------------------------------------------
$\sigma_n$ $\sigma_p$ $\sigma_e$ $\sigma_{\bar{\nu}_e}$ $f$
----------------- ----------------- ----------------- ------------------------ --------------------------------------------------------------------
$+\frac{1}{2} $ $+\frac{1}{2} $-\frac{1}{2} $ $-\frac{1}{2} $ $(\bar{g}_S - \bar{g}_T)
$ \sin\frac{\vartheta}{2}\,e^{\, i \varphi}$
$+\frac{1}{2} $ $+\frac{1}{2} $ $+\frac{1}{2} $ $-\frac{1}{2} $ $(\bar{g}_S +
\bar{g}_T) \cos\frac{\vartheta}{2}$
$+\frac{1}{2} $ $-\frac{1}{2} $ $-\frac{1}{2} $ $-\frac{1}{2} $ $0$
$+\frac{1}{2} $ $-\frac{1}{2} $ $+\frac{1}{2} $ $-\frac{1}{2} $ $ 2 \bar{g}_T \sin\frac{\vartheta}{2}\, e^{i\varphi}$
$-\frac{1}{2} $ $+\frac{1}{2} $ $-\frac{1}{2} $ $-\frac{1}{2} $ $2 \bar{g}_T \cos\frac{\vartheta}{2}$
$-\frac{1}{2} $ $+\frac{1}{2} $ $+\frac{1}{2} $ $-\frac{1}{2} $ $0$
$-\frac{1}{2} $ $-\frac{1}{2} $ $-\frac{1}{2} $ $-\frac{1}{2} $ $(\bar{g}_S + \bar{g}_T) \sin\frac{\vartheta}{2}\,e^{\,i \varphi}$
$-\frac{1}{2} $ $-\frac{1}{2} $ $+\frac{1}{2} $ $-\frac{1}{2} $ $(\bar{g}_S - \bar{g}_T) \cos\frac{\vartheta}{2}$
---------------------------------------------------------------------------------------------------------------------------------------------------
: The contributions of different spinorial states of the interacting particles to the amplitudes of the bound-state $\beta^-$–decay of the neutron and the antineutrino in the state with the wave function Eq. (\[label19\]); $f$ is defined by $f =
\bar{g}_S\,[\varphi^{\dagger}_e\chi_{_{\bar{\nu}_e}}][\varphi^{\dagger}_p
\varphi_n] + \bar{g}_T\,
[\varphi^{\dagger}_e\vec{\sigma}\,\chi_{_{\bar{\nu}_e}}]\cdot
[ \varphi^{\dagger}_p\vec{\sigma}\,\varphi_n]$.
Since the contributions of the left–handed polarisation state of antineutrinos to the angular distributions $dW^{(\pm)}_F(\theta)/d\Omega$ for $F = 0,1$ are of order $\bar{g}^2_S$, $\bar{g}^2_T$, $\bar{g}_S\bar{g}_T$ and can be screened by the contributions of the right–handed polarisation state of antineutrinos, we may neglect them. However, the angular distributions of the probabilities $W^{(\pm)}_1(\theta)$, $W^{(\pm)}_2(\theta)$, $W^{(\pm)}_3(\theta)$ and $W^{(\pm)}_4(\theta)$ may be meaningful if the corresponding angular distributions, caused by the right–handed polarisation state of antineutrinos, vanish. From Table II we obtain $$\begin{aligned}
\label{label21}
4\pi \frac{d W^{(+)}_1(\theta)}{d \Omega} &=&
\frac{1}{2}\,\frac{|\bar{g}_S + \bar{g}_T|^2}{1 + 3 \lambda^2_{\rm
eff}} \,(1 - \cos\theta),\nonumber\\ 4\pi \frac{d
W^{(+)}_2(\theta)}{d \Omega} &=& \frac{1}{2}\,\frac{|\bar{g}_S -
\bar{g}_T|^2}{1 + 3 \lambda^2_{\rm eff}} \,(1 +
\cos\theta),\nonumber\\ 4\pi \frac{d W^{(+)}_3(\theta)}{d \Omega} &=&
\frac{2|\bar{g}_T|^2}{1 + 3 \lambda^2_{\rm eff}}\,(1 +
\cos\theta),\nonumber\\ 4\pi \frac{d W^{(+)}_4(\theta)}{d \Omega} &=&
0,\end{aligned}$$ and $$\begin{aligned}
\label{label22}
\frac{d W^{(-)}_1(\theta)}{d \Omega} &=& \frac{d W^{(+)}_4(\pi -
\theta)}{d \Omega},\nonumber\\\frac{d W^{(-)}_2(\theta)}{d \Omega}
&=& \frac{d W^{(+)}_3(\pi - \theta)}{d \Omega}, \nonumber\\ \frac{d
W^{(-)}_3(\theta)}{d \Omega} &=& \frac{d W^{(+)}_2(\pi - \theta)}{d
\Omega},\nonumber\\\frac{d W^{(-)}_4(\theta)}{d \Omega} &=& \frac{d
W^{(+)}_1(\pi - \theta)}{d \Omega}.\end{aligned}$$ The contributions of the left–handed polarisation state of the electron antineutrino are of the second order in comparison with the contributions of the right–handed polarisation state. This implies that the contributions of the left–handed polarisation state can be fully screened by the contributions of the right–handed one.
Conclusion {#sec:conclusion}
==========
We have revised the bound-state $\beta^-$–decay of the free neutron. For the calculation of the branching ratio we have taken into account the contribution of radiative corrections and analysed the dependence of the probabilities of the bound-state $\beta^-$–decay on phenomenological coupling constants [@Jackson:1957zz; @Jackson1957206; @Severijns:2006dr; @Herczeg:2001vk], describing the most general weak effective lepton–nucleon interactions. We have found that to linear approximation with respect to Herczeg’s phenomenological coupling constants, introduced at the hadronic level [@Ivanov:2012qe], these probabilities depend only on the effective coupling constant ${\rm Re}(\lambda_{\rm eff}g_S -
g_T)$, which carries no information above weak interactions, caused by exchanges of electroweak bosons coupled to right–handed leptonic and hadronic currents. This implies that the bound-state $\beta^-$–decay is not a good laboratory for experimental investigations of electroweak models with left–right symmetries, as has previously been argued [@Byrne:2001sj].
Acknowledgement
===============
This work was supported by the Austrian “Fonds zur Förderung der Wissenschaftlichen Forschung” (FWF) under the contracts I689-N16, I534-N20 PERC and I862-N20 and by the Russian Foundation for Basic Research under the contract No. 11-02-91000 -ANF$_-$a.
[^1]: With respect to the definition of the coupling constants in the original paper [@Herczeg:2001vk] we have extracted the common factor $G_FV_{ud}/\sqrt{2}$ and replaced $a^h_{LL}$ and $a^h_{LR}$ by $a^h_{LL} \to a^h_{LL} + (1 - \lambda)/2$ and $a^h_{LR} \to a^h_{LR}
+ (1 + \lambda)/2$, respectively, where the coupling constants $a^h_{LL}$ and $a^h_{LR}$ describe deviations from the coupling constants of the SM.
[^2]: The Dirac wave function of antineutrinos in the right–handed polarisation state is equal to $u_{\bar{\nu}_e} = C \bar{v}^T_{\bar{\nu}_e}$ or $v_{\bar{\nu}_e} = C \bar{u}^T_{\bar{\nu}_e}$, where $C = i
\gamma^0\gamma^2$ and $T$ is a transposition [@Itzykson:1980rh]. The wave function $u_{\bar{\nu}_e} = C
\bar{v}^T_{\bar{\nu}_e}$ is a column matrix function with elements $\sqrt{E}\,(\varphi_{\bar{\nu}_e}, (\vec{\sigma}\cdot
\vec{n}\,)\varphi_{\bar{\nu}_e})$, where the Pauli spinor wave function $\varphi_{\bar{\nu}_e}$ is equal to $\varphi_{\bar{\nu}_e}
= - i\sigma^2 \chi^*_{\bar{\nu}_e}$ and obeys the equation $(\vec{\sigma}\cdot \vec{n}\,)\varphi_{\bar{\nu}_e} = +
\varphi_{\bar{\nu}_e}$ [@Itzykson:1980rh].
[^3]: The Dirac wave function of antineutrinos in the left–handed polarisation state is equal to $u_{\bar{\nu}_e} = C
\bar{v}^T_{\bar{\nu}_e}$, where the Pauli spinor wave function $\varphi_{\bar{\nu}_e} = - i\sigma^2 \chi^*_{\bar{\nu}_e}$ obeys the equation $(\vec{\sigma}\cdot \vec{n}\,)\varphi_{\bar{\nu}_e} = -
\varphi_{\bar{\nu}_e}$ [@Itzykson:1980rh].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We find a Lax pair equation corresponding to the Connes-Kreimer Birkhoff factorization of the character group of a Hopf algebra. This flow preserves the locality of counterterms. In particular, we obtain a flow for the character given by Feynman rules, and relate this flow to the Renormalization Group Flow.'
address:
- ' Institute of Mathematics of the Romanian Academy, PO Box 1-764, 014700 Bucharest, Romania and Max-Planck-Institut f" ur Mathematik, P.O. Box 7280, D-53072 Bonn, Germany. `[email protected]`'
- 'Department of Mathematics and Statistics, Boston University, Boston, MA 02215, USA. `[email protected]`'
author:
- Gabriel Bădiţoiu
- Steven Rosenberg
bibliography:
- 'paper-new2.bib'
date: '. Erwin Schrödinger Institut preprint number 2144. AMS classification: 81T15,17B80. Keywords: Renormalization, Lax pair equations, Hopf algebras. '
title: Feynman diagrams and Lax pair equations
---
Introduction
============
In the theory of integrable systems, many classical mechanical systems are described by a Lax pair equation associated to a coadjoint orbit of a semisimple Lie group, for example via the Adler-Kostant-Symes theorem [@adler]. Solutions are given by a Birkhoff factorization on the group, and in some cases, this technique extends to loop group formulations of physically interesting systems such as the Toda lattice [@guest; @sts]. By the work of Connes-Kreimer [@ck1], there is a Birkhoff factorization of characters on general Hopf algebras, in particular on the Kreimer Hopf algebra of 1PI Feynman diagrams. In this paper, we reverse the usual procedure in integrable systems: we construct a Lax pair equation $\frac{d L}{dt}=[L,M]$ on the Lie algebra of infinitesimal characters of the Hopf algebra whose solution is given precisely by the Connes-Kreimer Birkhoff factorization (Theorem \[t:8.2\]). The Lax pair equation is nontrivial in the sense that it is not an infinitesimal inner automorphism. The main technical issue, that the Lie algebra of infinitesimal characters is not semisimple, is overcome by passing to the double Lie algebra with the simplest possible Lie algebra structure. In particular, the Lax pair equation induces a flow for the character given by Feynman rules in dimensional regularization. This flow has the physical significance that it preserves locality, the independence of the character’s counterterm on the mass parameter.
In §§1-4, we introduce a method to produce a Lax pair on any Lie algebra from equations of motion on the double Lie algebra. In §\[hopf\], we apply this method to the particular case of the Lie algebra of infinitesimal characters of a Hopf algebra, and prove Theorem \[t:8.2\].
The Renormalization Group Flow (RGF) usually considered in quantum field theory is a flow on the character group $G_{\mathcal A}$, while the Lax pair flow is on the corresponding Lie algebra ${\mathfrak g}_{\mathcal A}$ of infinitesimal characters. There are various bijections from ${\mathfrak g}_{\mathcal A}$ to $G_{\mathcal A}$, and via these bijections we can compare the Lax pair flow to the RGF. These flows are not the same, so we study how physically significant quantities behave under the Lax pair flow. In §6, we derive an equation for the flow of the $\beta$-function of characters $\varphi_t\in G_{\mathcal A}$ associated to the Lax pair flow via the exponential map $\exp: \mathfrak
g_{\mathcal A}\to G_{\mathcal A}$ (Corollary \[cor:beta\]). In §7, we first show that the Lax pair flow is trivial on primitives in the Hopf algebra. We then use Manchon’s bijection [@man] $\tilde R^{-1}:\mathfrak
g_{\mathcal A}\to G_{\mathcal A}$ to prove various locality results (Theorems \[t:7.9\], \[t:7.14\]). The $\beta$-function flow defined via $\tilde R^{-1}$ itself satisfies a Lax pair equation (Theorem \[t:corr\]). Thus $\tilde R^{-1}$ is much better behaved than the exponential map. In §8, we work out several examples of this theory, and in particular keep track of the leading log terms.
An alternative algebraic geometric approach to Lax pair equations is to apply spectral curve techniques to linearize the flow on the Jacobian of the spectral curve. Unfortunately, in the worked example of §\[worked-example\], the spectral curve is reducible, and the only invariants we find are trivial. We hope to find examples with nontrivial invariants in the future.
We would like to thank Dirk Kreimer for suggesting we investigate the connection between the Connes-Kreimer factorization and integrable systems, and Dominique Manchon for helpful conversations.
The double Lie algebra and its associated Lie Group
===================================================
There is a well known method to associate a Lax pair equation to a Casimir element on the dual $\mathfrak g^*$ of a semisimple Lie algebra $\mathfrak g$ [@sts]. The semisimplicity is used to produce an $\mathrm{Ad}$-invariant, symmetric, non-degenerate bilinear form on $\mathfrak g$, allowing an identification of $\mathfrak g$ with $\mathfrak g^*$. For a general Lie algebra $\mathfrak g$, there may be no such bilinear form. To produce a Lax pair, we need to extend $\mathfrak g$ to a larger Lie algebra with the desired bilinear form. We do this by constructing a Lie bialgebra structure on $\mathfrak g$, whose definition we now recall (see e.g. [@ksch]).
\[bialg\] A Lie bialgebra is a Lie algebra $(\mathfrak g, [\cdot ,\cdot ])$ with a linear map $\gamma:\mathfrak g\to\mathfrak g\otimes\mathfrak g$ such that
- $^t\gamma:\mathfrak g^*\otimes\mathfrak g^*\to\mathfrak g^*$ defines a Lie bracket on $\mathfrak g^*$,
- $\gamma$ is a $1$-cocycle of $\mathfrak g$, i.e. $$\mathrm{ad}^{(2)}_x(\gamma(y))-\mathrm{ad}^{(2)}_y(\gamma(x))-\gamma([x,y])=0,$$ where $\mathrm{ad}^{(2)}_x:\mathfrak g\otimes\mathfrak
g\to\mathfrak g\otimes\mathfrak g$ is given by $\mathrm{ad}^{(2)}_x(y\otimes z)=\mathrm{ad}_x(y)\otimes z+y\otimes
\mathrm{ad}_x(z)
= [x,y]\otimes z + y \otimes [x,z]$.
A Lie bialgebra $(\mathfrak g,[\cdot,\cdot ],\gamma)$ induces an Lie algebra structure on the [*double Lie algebra*]{} $\mathfrak g\oplus\mathfrak g^*$ by $$[X,Y]_{\mathfrak g\oplus\mathfrak g^*}=[X,Y],$$ $$[X^*,Y^*]_{\mathfrak g\oplus\mathfrak g^*}= {}^t\gamma(X\otimes Y),$$ $$[X,Y^*]=\mathrm{ad}^*_X(Y^*),$$ for $X$, $Y\in\mathfrak g$ and $X^*$, $Y^*\in\mathfrak g^*$, where $\mathrm{ad}^*$ is the coadjoint representation given by $\mathrm{ad}^*_X(Y^*)(Z)=-Y^*(\mathrm{ad}_X(Z))$ for $Z\in\mathfrak g$.
Since it is difficult to construct explicitly the Lie group associated to the Lie algebra $\mathfrak g\oplus\mathfrak g^*$, we will choose the trivial Lie bialgebra given by the cocycle $\gamma=0$ and denote by $\delta=\mathfrak g\oplus \mathfrak g^*$ the associated Lie algebra. Let $\{Y_i,
i= 1,\ldots ,l\}$ be a basis of $\mathfrak g$, with dual basis $\{Y^*_i\}$. The Lie bracket $[ \cdot,\cdot ]_\delta$ on $\delta$ is given by $$[Y_i,Y_j]_\delta=[Y_i,Y_j],\ [Y_i^*,Y_j^*]_\delta=0,\
[Y_i,Y_j^*]_\delta=-\sum_k c^j_{ik}Y^*_k,$$ where the $c^j_{ik}$ are the structure constants: $[Y_i,Y_j]=\sum_kc^k_{ij}Y_k$. The Lie group naturally associated to $\delta$ is given by the following proposition.
\[prop22\] Let $G$ be the simply connected Lie group with Lie algebra $\mathfrak{g}$ and let $\theta:G\times \mathfrak g^*\to \mathfrak g^*$ be the coadjoint representation $\theta(g,X)=\mathrm{Ad}^*_{G}(g)(X)$. Then the Lie algebra of the semi-direct product $\tilde G =
G\ltimes_\theta \mathfrak g^*$ is the double Lie algebra $\delta$.
The Lie group law on the semi-direct product $\tilde
G$ is given by $$(g,X)\cdot (g',X')=(gg',X+\theta(g,X')).$$ Let $\tilde{\mathfrak g}$ be the Lie algebra of $\tilde G$. Then the bracket on $\tilde{\mathfrak g}$ is given by $$[X,Y^*]_{\tilde{ \mathfrak g}}
=d\theta(X,Y^*), \ \ [X,Y]_{\tilde {\mathfrak
g}}=[X,Y],\ \ [X^*,Y^*]_{\tilde {\mathfrak g}}=0,$$ for left-invariant vector fields $X$, $Y$ of $G$ and $X^*, Y^*\in\mathfrak g^*$. We have $d\theta(X,Y^*)=d\mathrm{Ad}^*_{G}(X)(Y^*)=[X,Y^*]_\delta$ since $d\mathrm{Ad}_{G}=\mathrm{ad}_{\mathfrak g}$.
The main point of this construction is existence of a good bilinear form on the double.
The natural pairing $\langle\cdot, \cdot\rangle:\delta\otimes\delta\to{\mathbb C}$ given by $$\langle (a, b^*), (c,d^*)\rangle = d^*(a) + b^*(c), \ \ \ a,c\in\mathfrak{g},\ \
b^*, d^*\in\mathfrak{g^*},$$ is an $\mathrm{Ad}$-invariant symmetric non-degenerate bilinear form on the Lie algebra $\delta$.
By [@ksch], this bilinear form is ad-invariant. Since $\tilde G$ is simply connected, the Ad-invariance follows. As an explicit example, we have $$\mathrm{Ad}_{\tilde G}((g,0))(Y_i,0) = (\mathrm{Ad}_G(g)(Y_i),0),\ \
\text{and}\ \ \mathrm{Ad}_{\tilde
G}((g,0))(0,Y_j^*) = (0, \mathrm{Ad}_{G}^*(g)(Y_j^*)),$$ from which the invariance under $\Ad_{\tilde G}(g,0)$ follows.
The loop algebra of a Lie algebra
=================================
Following [@adler], we consider the loop algebra $$L\delta=\{L(\lambda)=\sum\limits_{j=M}^N \lambda^jL_j \ |
\ M,N\in \mathbb Z, L_j\in\delta\}.$$ The natural Lie bracket on $L\delta $ is given by $$\left[\sum \lambda^iL_i,\sum \lambda^j L_j'\right]=
\sum\limits_k \lambda^k\sum\limits_{i+j=k}[L_i,L_j'].$$ Set $$\begin{aligned}
L\delta _+ &=&
\{L(\lambda)= \sum\limits_{j=0}^N
\lambda^jL_j \ | \ N\in\mathbb Z^+\cup \{0\}, L_j\in\delta\}\\
L\delta _-&=&
\{L(\lambda)=\sum\limits_{j=-M}^{-1} \lambda^jL_j \ | \ M\in\mathbb Z^+,
L_j\in\delta\}.\end{aligned}$$ Let $P_+:L\delta \to L\delta _+$ and $P_-:L\delta \to L\delta _-$ be the natural projections and set $R=P_+-P_-$.
The natural pairing $\langle\cdot,\cdot\rangle$ on $\delta$ yields an $\mathrm{Ad}$-invariant, symmetric, non-degenerate pairing on $L\delta $ by setting $$\left\langle\sum\limits_{i=M}^N \lambda^iL_i ,\sum\limits_{j=M'}^{N'}
\lambda^jL'_j \right\rangle=
\sum\limits_{i+j=-1}\langle L_i,L_j'\rangle.$$
For our choice of basis $\{Y_i\}$ of $\mathfrak g$, we get an isomorphism $$\label{I}I: L(\delta ^*)\to L\delta$$ with $$I\left(\sum L^j_iY_j\lambda^i\right)=\sum
L^j_iY^*_j\lambda^{-1-i}.$$ We will need the following lemmas.
[@adler] We have the following natural identifications: $$L\delta _+=L(\delta^*)_- \ \mathrm{and\ } L\delta _-=L(\delta^*)_+.$$
[@sts Lem. 4.1]\[lem1\] Let $\varphi$ be an $\mathrm{Ad}$-invariant polynomial on $\delta$. Then $$\varphi_{m,n}[L(\lambda)]=
\mathrm{Res}_{\lambda=0}(\lambda^{-n}\varphi(\lambda^mL(\lambda)))$$ is an $\mathrm{Ad}$-invariant polynomial on $L\delta $ for $m, n\in\mathbb
Z.$
As a double Lie algebra, $\delta$ has an Ad-invariant polynomial, the quadratic polynomial $$\psi(Y)=\langle Y,Y\rangle$$ associated to the natural pairing. Let $Y_{l+i}=Y^*_i$ for $i\in\{1,\ldots ,l =\mathrm{dim}({\mathfrak g})
\}$, so elements of $L\delta$ can be written $L(\lambda)=\sum\limits_{j=1}^{2l} \sum\limits_{i=-M}^N
L_i^jY_j\lambda^i$. Then the Ad-invariant polynomials $$\label{psimn1}
\psi_{m,n}(L(\lambda))=\mathrm{Res}_{\lambda=0}(\lambda^{-n}
\psi(\lambda^mL(\lambda))),$$ defined as in Lemma \[lem1\] are given by $$\label{psimn2}
\psi_{m,n}(L(\lambda))=2\sum
\limits_{j=1}^l\sum\limits_{i+k-n+2m=-1} L_i^jL_k^{j+l}.$$ Note that powers of $\psi$ are also $\mathrm{Ad}$-invariant polynomials on $\delta $, so $$\label{psimnk}
\psi^k_{m,n}(L(\lambda))=\mathrm{Res}_{\lambda=0}(\lambda^{-n}\psi^k(\lambda^mL(\lambda)))$$ are $\mathrm{Ad}$-invariant polynomials on $L\delta $. It would be interesting to classify all Ad-invariant polynomials on $L\delta$ in general.
The Lax pair equation {#s:7}
=====================
Let $P_+$, $P_-$ be endomorphisms of a Lie algebra $\mathfrak h$ and set $R =
P_+-P_-.$ Assume that $$[X,Y]_R=[P_+X,P_+Y]-[P_-X,P_-Y]$$ is a Lie bracket on $\mathfrak h$. From [@sts Theorem 2.1], the equations of motion induced by a Casimir (i.e. Ad-invariant) function $\varphi$ on ${\mathfrak h}^*$ are given by $$\label{e:5.1}
\frac{dL}{dt}=-\mathrm{ad}^*_{\mathfrak h}M\cdot L,$$ for $L\in\mathfrak h^*,$ where $ M=\frac{1}{2}R(d\varphi(L))\in\mathfrak h.$
Now we take $\mathfrak h=(L\delta)^*=L(\delta^*)$, with $\delta$ a finite dimensional Lie algebra and with the understanding that $(L\delta)^*$ is the graded dual with respect to the standard $\Z$-grading on $L\delta.$ Let $P_\pm$ be the projections of $L\delta^*$ onto $L\delta^*_\pm$. After identifying $L\delta ^*=L\delta $ and $\mathrm{ad}^*=-\mathrm{ad}$ via the map $I$ in (\[I\]), the equations of motion (\[e:5.1\]) can be written in Lax pair form $$\label{e:5.2}
\frac{{d}L}{{d}t}=[M,L],$$ where $ M=\frac{1}{2}R(I(d\varphi(L(\lambda))))\in L\delta,$ and $\varphi$ is a Casimir function on $L\delta^* = L\delta$ [@sts Theorem 2.1]. Finding a solution for (\[e:5.2\]) reduces to the Riemann-Hilbert (or Birkhoff) factorization problem. The following theorem is a corollary of [@adler Theorem 4.37] [@sts Theorem 2.2].
\[t:7.4\] Let $\varphi$ be a Casimir function on $L\delta$ and set $X=I(d\varphi(L(\lambda)))\in L\delta$, for $L(\lambda) = L(0)(\lambda)\in L\delta$. Let $g_{\pm}(t)$ be the smooth curves in $L\tilde G$ which solve the factorization problem $$\exp(-tX)=g_-(t)^{-1}g_+(t),$$ with $g_{\pm}(0)=e$, and with $g_+(t) = g_+(t)(\lambda)$ holomorphic in $\lambda\in {\mathbb C}$ and $g_-(t)$ a polynomial in $1/\lambda$ with no constant term. Let $M=\frac{1}{2}R(I(d\varphi(L(\lambda))))\in L\delta$. Then the integral curve $L(t)$ of the Lax pair equation $$\frac{d L}{d t}=[L, M]$$ is given by $$\label{quick}
L(t)=\mathrm{Ad}_{L\tilde G}g_{\pm}(t)\cdot L(0).$$
This Lax pair equation projects to a Lax pair equation on the loop algebra of the original Lie algebra $\mathfrak g.$ Let $\pi_1$ be either the projection of $\tilde G$ onto $G$ or its differential from $\delta$ onto $\mathfrak g$. This extends to a projection of $L\delta$ onto $L\mathfrak g$. The projection of (\[e:5.2\]) onto $L\mathfrak g$ is $$\label{e:pi1}
\frac{d(\pi_1( L(t)))}{dt}=[\pi_1(L),\pi_1(M)],$$ since $\pi_1 = d\pi_1 $ commutes with the bracket. Thus the equations of motion (\[e:5.2\]) induce a Lax pair equation on $L\mathfrak g$, although this is not the equations of motion for a Casimir on $L \mathfrak g.$
\[t:4.2\] The Lax pair equation of Theorem \[t:7.4\] projects to a Lax pair equation on $L\mathfrak g.$
The content of this theorem is that a Lax pair equation on the Lie algebra of a semi-direct product $G\ltimes G'$ evolves on an adjoint orbit, and the projection onto $\mathfrak g$ evolves on an adjoint orbit and is still in Lax pair form. Lax pair equations often appear as equations of motion for some Hamiltonian, but the projection may not be the equations of motion for any function on the smaller Lie algebra. We thank B. Khesin for this observation.
When $\psi_{m,n}$ is the Casimir function on $L\delta$ given by (\[psimn1\]), $X$ can be written nicely in terms of $L(\lambda)$.
\[p:5.3\] Let $X=I(d \psi_{m,n}(L(\lambda)))$. Then $$\label{e:5.3}
X=2\lambda^{-n+2m} L(\lambda).$$
Write $
L(\lambda) =\sum\limits_{i,j}L_i^j\lambda^{i}Y_j.$ By formula (\[psimn2\]), we have $$\label{e:partialpsimn}
\frac{\partial \psi_{m,n}}{\partial L_p^t}= \left\{
\begin{array}{cc}
2L_{n-1-2m-p}^{t+l}, \ \ \ \text{if} \ \ t\leq l\\
2L_{n-1-2m-p}^{t-l}, \ \ \ \text{if} \ \ t>l.\\
\end{array}
\right.$$ Therefore $$\begin{aligned}
\nonumber X&=&I(d\psi_{m,n}(L(\lambda)))
=\sum\limits_{p,t}\frac{\partial \psi_{m,n}}{\partial L_p^t}\lambda^{-1-p}Y^*_t\\
\nonumber &=&
2\lambda^{-n+2m}\sum\limits_p\left(
\sum\limits_{t=1}^{l}L^{t+l}_{n-1-2m-p}Y_{t+l}\lambda^{n-1-2m-p}
+\sum\limits_{t=l+1}^{2l}
L^{t-l}_{n-1-2m-p}Y_{t-l}\lambda^{n-1-2m-p}\right)\\
\nonumber &=&2 \lambda^{-n+2m}L(\lambda).
\end{aligned}$$
The main theorem for Hopf algebras {#hopf}
==================================
In this section we give formulas for the Birkhoff decomposition of a loop in the Lie group of characters of a Hopf algebra and produce the Lax pair equations associated to the Birkhoff decomposition. We present two approaches, both motivated by the Connes-Kreimer Hopf algebra of 1PI Feynman graphs. First, in analogy to truncating Feynman integral calculations at a certain loop level, we truncate a (possibly infinitely generated) Hopf algebra to a finitely generated Hopf algebra, and solve Lax pair equations on the finite dimensional piece (Theorem \[t:8.2fg\]). We also discuss the compatibility of solutions related to different truncations. Second, we solve a Lax pair equation associated to the full Hopf algebra, but for a restricted family of Casimirs (Theorem \[t:8.2\]).
Let ${\mathcal H}=({\mathcal H}, 1, \mu,\De,\ep,S)$ be a graded connected Hopf algebra over ${\mathbb C}$. Let $\mathcal A$ be a unital commutative algebra with unit $1_\mathcal A$. Unless stated otherwise, $\mathcal{A}$ will be the algebra of Laurent series; the only other occurrence in this paper is ${\mathcal A} = \CC.$
The [**character group**]{} $G_\mathcal A$ of the Hopf algebra ${\mathcal H}$ is the set of algebra morphisms $\phi:{\mathcal H}\to\mathcal{A}$ with $\phi(1)=1_\mathcal{A}.$ The group law is given by the convolution product $$(\psi_1\star\psi_2)(h)=\langle \psi_1\otimes\psi_2,\De h\rangle;$$ the unit element is $\ep$.
An $\mathcal A$-valued [**infinitesimal character**]{} of a Hopf algebra ${\mathcal H}$ is a ${\mathbb C}$-linear map $Z:{\mathcal H}\to\mathcal{A}$ satisfying $$\langle Z,hk\rangle =\langle Z, h\rangle \varepsilon(k)+\varepsilon(h)
\langle Z,k\rangle.$$ The set of infinitesimal characters is denoted by $\mathfrak{g}_\mathcal{A}$ and is endowed with a Lie algebra bracket: $$[Z,Z']=Z\star Z'-Z'\star Z,\ \ \mathrm{for\ }Z,\ Z'\in\mathfrak{g}_\mathcal{A},$$ where $\langle Z\star Z',h\rangle=\langle Z\otimes Z',\Delta(h)\rangle$. Notice that $Z(1)=0$.
For a finitely generated Hopf algebra, $G_\CC$ is a Lie group with Lie algebra ${\mathfrak g}_\CC$, and for any Hopf algebra and any ${\mathcal A}$, the same is true at least formally.
We recall that $\delta=\mathfrak{g}_{\mathcal{{\mathbb
C}}}\oplus\mathfrak{g}_{\mathcal{{\mathbb C}}}^{*}$ is the double of $\mathfrak{g}_{\mathbb C}$ and the $\mathfrak{g}_{\mathbb C}^{*}$ is the graded dual of $\mathfrak{g}_{\mathbb C}$. We consider the algebra $\Omega\delta = \delta\otimes\mathcal{A}$ of formal Laurent series with values in $\delta$ $$\Omega\delta=\{L(\lambda)=\sum\limits_{j=-N}^\infty \lambda^jL_j \ |
\ L_j\in\delta, N\in \Z\}.$$ The natural Lie bracket on $\Omega\delta $ is $$\left[\sum \lambda^iL_i,\sum \lambda^j L_j'\right]=
\sum\limits_k \lambda^k\sum\limits_{i+j=k}[L_i,L_j'].$$ Set $$\begin{aligned}
\Omega\delta _+ &=&
\{L(\lambda)= \sum\limits_{j=0}^\infty
\lambda^jL_j \ | \ L_j\in\delta\}\\
\Omega\delta _- &=& \{L(\lambda)=\sum\limits_{j=-N}^{-1}
\lambda^jL_j \ | \ L_j\in\delta, N\in \Z^+\}.\end{aligned}$$
Recall that for any Lie group $K$, a loop $L(\lambda)$ with values in $K$ has a Birkhoff decomposition if $L(\lambda) = L(\lambda)
_-^{-1}L(\lambda)_+$ with $L(\lambda)_-^{-1}$ holomorphic in $\lambda^{-1} \in \mathbb P^1 -\{0\}$ and $L(\lambda)_+$ holomorphic in $\lambda \in\mathbb P^1 - \{\infty\}.$ In the next lemma, $\tilde G$ refers to $G\ltimes_\theta \mathfrak g^*$ as in Prop. \[prop22\].
We prove the existence of a Birkhoff decomposition for any element $(g,\alpha)\in \Omega\tilde G$.
\[t:omegatildeg\] Every $(g,\alpha)\in\Omega\tilde
G=G_\mathcal{A}\ltimes_{Ad^*_{G_{\mathcal A}}}
\mathfrak{g}_{\mathcal{A}}^{*}
$ has a Birkhoff decomposition $(g,\alpha)=(g_-,\alpha_-)^{-1}(g_+,\alpha_+)$ with $(g_+,\alpha_+)$ holomorphic in $\lambda$ and $(g_-,\alpha_-)$ a polynomial in $\lambda^{-1}$ without constant term.
We recall that $(g_1,\alpha_1)(g_2,\alpha_2)=
(g_1g_2,\alpha_1+\mathrm{Ad}^*(g_1)(\alpha_2)).$ Thus\
$
(g,\alpha)=(g_-,\alpha_-)^{-1}(g_+,\alpha_+)
\text{ if and only if }
g=g_-^{-1}g_+
\text{ and }
\alpha=\mathrm{Ad}^*(g_-^{-1})(-\alpha_-+\alpha_+).
$ Let $g=g_-^{-1}g_+$ be the Birkhoff decomposition of $g$ in $ G_\mathcal{A}$ given in [@ck1; @egk; @man]. Set $\alpha_+=P_+(\mathrm{Ad}^*(g_-)(\alpha))$ and $\alpha_-=-P_-(\mathrm{Ad}^*(g_-)(\alpha))$, where $P_+$ and $P_-$ are the holomorphic and pole part, respectively. Then for this choice of $\alpha_+$ and $\alpha_-$, we have $(g,\alpha)=(g_-,\alpha_-)^{-1}(g_+,\alpha_+)$. Note that the Birkhoff decomposition is unique.
For a finitely generated Hopf algebra, we can apply Theorems \[t:7.4\], \[t:4.2\] to produce a Lax pair equation on $L\delta$ and on the loop space of infinitesimal characters $L\mathfrak{g}$. However, the common Hopf algebras of 1PI Feynman diagrams and rooted trees are not finitely generated.
As we now explain, we can truncate the Hopf algebra to a finitely generated Hopf algebra, and use the Birkhoff decomposition to solve a Lax pair equation on the infinitesimal character group of the truncation. A graded Hopf algebra $\mathcal H=\oplus_{n\in\mathbb N} \mathcal H_n$ is said to be of [**finite type**]{} if each homogeneous component $\mathcal H_n$ is a finite dimensional vector space. Let $\mathcal B=\{T_i\}_{i\in\mathbb N}$ be a minimal set of homogeneous generators of the Hopf algebra $H$ such that $\deg(T_i)\leq\deg(T_j)$ if $i<j$ and such that $T_0=1$. For $i>0$, we define the $\mathbb{C}$-valued infinitesimal character $Z_i$ on generators by $Z_i(T_j)=\delta_{ij}.$ The Lie algebra of infinitesimal characters $\mathfrak g$ is a graded Lie algebra generated by $\{Z_i\}_{i>0}$. Let $\mathfrak{g}^{(k)}$ be the vector space generated by $\{Z_i\ | \ \deg(T_i)\leq k\}$. We define $\deg(Z_i)=\deg(T_i)$ and set $$[Z_i,Z_j]_{\mathfrak{g}^{(k)}}=\left\{
\begin{array}{cc}
[Z_i,Z_j] & \text{if }\deg(Z_i)+\deg(Z_j)\leq k \\
0 & \text{if }\deg(Z_i)+\deg(Z_j)>k
\end{array}\right.$$ We identify $\varphi\in G_{\mathbb C}$ with $\{\varphi(T_i)\}\in\mathbb C^{\mathbb N}$ and on $\mathbb
C^{\mathbb N}$ we set a group law given by $\{\varphi_1(T_i)\}\oplus\{\varphi_2(T_i)\}=\{(\varphi_1\star\varphi_2)(T_i)\}$. $G^{(k)}=\{ \{\varphi(T_i)\}_{\{i\, |\, \deg(T_i)\leq
k\}}\ | \ \varphi\in G_{\mathbb C}\}$ is a finite dimensional Lie subgroup of $G_{\mathbb C}=(\mathbb C^{\mathbb N},\oplus)$ and the Lie algebra of $G^{(k)}$ is $\mathfrak{g}^{(k)}$. There is no loss of information under this identification, as $\varphi(T_iT_j)=\varphi(T_i)\varphi(T_j)$.
Let $\delta^{(k)}$ be the double Lie algebra of $\mathfrak{g}^{(k)}$ and let $\tilde G^{(k)}$ be the simply connected Lie group with $\mathrm{Lie}(\tilde G^{(k)})=\delta^{(k)}$ as in Proposition \[prop22\]. The following theorem is a restatement of Theorem \[t:7.4\] in our new stage.
\[t:8.2fg\] Let $\mathcal H=\oplus_{n}\mathcal
H_n$ be a graded connected Hopf algebra of finite type, and let $\psi:L\delta^{(k)}\to{\mathbb C}$ be a Casimir function (e.g. $\psi(L ) = \psi_{m,n}(L(\lambda))=\mathrm{Res}_{\lambda=0}
(\lambda^m\psi(\lambda^n L(\lambda)))$ with $\psi:\delta^{(k)}\times\delta^{(k)}\to{\mathbb C}$ the natural paring of $\delta^{(k)}$). Set $
X=I(d\psi(L_0 ))$ for $L_0 \in L\delta^{(k)}$. Then the solution in $L\delta^{(k)}$ of $$\label{8:1fg}
\frac{dL}{dt}=[L,M]_{L\delta^{(k)}}, \ \ \ M=\frac{1}{2}
R(I(d\psi(L)))$$ with initial condition $L(0)=L_0$ is given by $$\label{e:8.3fg}
L(t)=\mathrm{Ad}_{L\tilde G^{(k)}}g_{\pm}(t)\cdot L_0,$$ where $\exp(-tX)$ has the Connes-Kreimer Birkhoff factorization\
$\exp(-tX)=g_-(t)^{-1}g_+(t)$.
\(i) If $L_0\in L\delta$, there exists $k\in\mathbb N$ such that $L_0\in L\delta^{(k)}$. Indeed $L_0\in L\delta$ is generated over $\mathbb C[\lambda,\lambda^{-1}]$ by a finite number of $\{Z_i\}$, and we can choose $k\geq\max\{\deg(Z_i)\}$.
\(ii) While the Hopf algebra of rooted trees and the Connes-Kreimer Hopf algebra of 1PI Feynman diagrams satisfy the hypothesis of Theorem \[t:8.2fg\], the Feynman rules character does not lie in $L\tilde G$, as explained below.
In the next sections, we will investigate the relationship between the Lax pair flow $L(t)$ and the Renormalization Group Equation. In preparation, we project from $L\delta^{(k)}$ to $L\mathfrak{g}^{(k)}$ via $\pi_1$ as in §4.
\[tc:8.2fg\] Let $\psi $ be a Casimir function on $L\delta^{(k)}$. Set $L_0\in L\mathfrak{g}^{(k)}\subset L\delta^{(k)}$, $X=\pi_1(I(d\psi(L_0)))$. Then the solution of the following equation in $L\mathfrak{g}^{(k)}$ $$\label{c:e:8.1}
\frac{dL}{dt}=[L,M_1]_{L\mathfrak{g}^{(k)}}, \ \ \ M_1=\pi_1(\frac{1}{2}
R(I(d\psi (L ))))$$ with initial condition $L(0)=L_0$ is given by $$\label{c:e:8.3}
L(t)=\mathrm{Ad}_{L G^{(k)}}g_{\pm}(t)\cdot L_0,$$ where $\exp(-tX)$ has the Connes-Kreimer Birkhoff factorization in $L\mathfrak{g}^{(k)}$ $$\exp(-t X)=g_-(t)^{-1}g_+(t).$$
\[lastrem\] (i) For Feynman graphs, this truncation corresponds to halting calculations after a certain loop level. From our point of view, this truncation is somewhat crude. $\mathfrak g^{(k)}$ is not a subalgebra of $\mathfrak g$, and if $k < \ell$, $\mathfrak g^{(k)}$ is not a subalgebra of $\mathfrak g^{(\ell)}$. Although the Casimirs $\psi_{m,n}$ and the exponential map restrict well from $\mathfrak g$ to $\mathfrak g^{(k)}$, the Birkhoff decomposition $\exp(-tX)$ of $X\in
L{\mathfrak g}^{(k)}$ is very different from the Birkhoff decompositions in $L\mathfrak g, L\mathfrak g^{(\ell)}$. In fact, if $g\in G^{(k)}$ has Birkhoff decomposition $g = g_-^{-1}g_+$ in $G$, there does not seem to be $f(k)\in \mathbb N$ such that $g_\pm\in G^{(f(k))}.$ Nevertheless, in the last section we will follow standard procedure and present calculations of truncated Hopf algebras.
(ii)It would interesting to know, especially for the Hopf algebras of Feynman graphs or rooted trees, whether there exists a larger connected graded Hopf algebra ${\mathcal H}'$ containing ${\mathcal H}$ such that the associated infinitesimal Lie algebra $\mathrm{Lie}(G'_\mathbb{C})$ is the double $\delta$. This would provide a Lax pair equation associated to an equation of motion on the infinitesimal Lie algebra of ${\mathcal H}'$. The most natural candidate, the Drinfeld double $\mathcal D({\mathcal H})$ of ${\mathcal H}$, does not work since the dimension of the Lie algebra associated to $\mathcal D({\mathcal H})$ is larger than the dimension of $\delta$.
In [@ck1], Connes and Kreimer give a Birkhoff decomposition for the character group of the Hopf algebra of 1PI graphs, and in particular for the Feynman rules character $\varphi(\lambda)$ given by minimal subtraction and dimensional regularization. The truncation process treated above does not handle the Feynman rules character, as the Feynman rules character and the toy model character of the Hopf algebra of rooted trees considered in §8 are not polynomials in $\lambda,\lambda^{-1}$, but Laurent series in $\lambda$. Thus Corollary \[tc:8.2fg\] does not apply, as in our notation $\log(\varphi(\lambda)) \in
\Omega\mathfrak{g} \setminus L\mathfrak{g}$. This and Remark \[lastrem\](i) force us to consider a direct approach in $\Omega\mathfrak{g}$ as in next theorem. However, we cannot expect that the Lax pair equation is associated to any Hamiltonian equation, and we replace Casimirs with Ad-covariant functions.
[@suris] Let $G$ be a Lie group with Lie algebra $\mathfrak{g}$. A map $f:\mathfrak{g}\to \mathfrak{g}$ is $\mathrm{Ad}$-[*covariant*]{} if $\mathrm{Ad}(g)(f(L))=f(\mathrm{Ad}(g)(L))$ for all $g\in G$, $L\in\mathfrak{g}$.
\[t:8.2\] Let ${\mathcal H}$ be a connected graded commutative Hopf algebra with $\mathfrak{g}_{\mathcal A}$ the associated Lie algebra of infinitesimal characters with values in Laurent series. Let $f:\mathfrak{g}_{\mathcal A}\to\mathfrak{g}_{\mathcal A}$ be an $\mathrm{Ad}$-covariant map. Let $L_0\in\mathfrak{g}_{\mathcal A}$ satisfy $[f(L_0),L_0]=0$. Set $X=f(L_0)$. Then the solution of $$\label{8:1}
\frac{dL}{dt}=[L,M], \ \ \ M=\frac{1}{2}
R(f(L))$$ with initial condition $L(0)=L_0$ is given by $$\label{e:8.3}
L(t)=\mathrm{Ad}_{G}g_{\pm}(t)\cdot L_0,$$ where $\exp(-tX)$ has the Connes-Kreimer Birkhoff factorization\
$\exp(-tX)=g_-(t)^{-1}g_+(t)$.
The proof is similar to [@sts Theorem 2.2]. First notice that $$\begin{aligned}
\frac{d}{dt}\left({\Ad}(g_-(t)^{-1}g_+(t))\cdot
L_0\right) &=&
\frac{d}{dt}(\exp(-tX)L_0\exp(tX))\\
&=& -\exp(-tX)XL_0\exp(tX)+\exp(-tX)L_0X\exp(tX)\\
&=& \exp(-tX)[X,L_0]\exp(tX)=0,\end{aligned}$$ which implies ${\Ad}(g_-(t)^{-1}g_+(t))\cdot
L_0=L_0$ and ${\Ad}(g_-(t) )\cdot L_0={\Ad}(g_+(t))\cdot L_0$. Set $L(t)={\Ad}(g_\pm(t) )\cdot L_0 = g_\pm(t)L_0g_\pm(t)^{-1}$. As usual, $$\frac{dL}{dt}
= \left[\frac{d g_\pm(t)}{dt}g_\pm(t)^{-1},L(t)\right],$$ so $$\frac{dL}{dt}=\frac{1}{2}\left[\frac{d g_+(t)}{dt}g_+(t)^{-1}+\frac{d
g_-(t)}{dt}g_-(t)^{-1},L(t)\right].$$ The Birkhoff factorization $g_+(t)=g_-(t)\exp(-tX)$ gives $$\frac{dg_+(t)}{dt}
=\frac{dg_-(t)}{dt}\exp(-tX)+g_-(t)(-X)\exp(-tX),$$ and so $$\frac{dg_+(t)}{dt}g_+(t)^{-1}
=\frac{dg_-(t)}{dt}g_-(t)^{-1}+g_-(t)(-X)g_-(t)^{-1}.$$ Thus $$\begin{aligned}
2M &=& R(f(L(t)))=R(f({\Ad}( g_-(t))\cdot L_0))=R({\Ad}( g_-(t))\cdot f(L_0))\\
&=& R({\Ad}( g_-(t))\cdot X)) =
-R(\frac{dg_+(t)}{dt}g_+(t)^{-1})
+R(\frac{dg_-(t)}{dt}g_-(t)^{-1})\\
&= & -\frac{dg_+(t)}{dt}g_+(t)^{-1}
-\frac{dg_-(t)}{dt}g_-(t)^{-1}.\end{aligned}$$ Here we use $(\frac{dg_\pm(t)}{dt}g_\pm(t)^{-1})(x)\in\mathcal
A_\pm$ for $x\in {\mathcal H}$. Thus $
\frac{dL}{dt}= [L,M].$
If $f:\mathfrak{g}_\mathcal{A}\to\mathfrak{g}_\mathcal{A}$ is given by $f(L)=2\lambda^{-n+2m}L$, then $f$ is $\mathrm{Ad}$-covariant and $[f(L_0),L_0]=[2\lambda^{-n+2m}L_0,L_0]=0$.
\[tc:8.2\] Let ${\mathcal H}$ be a connected graded commutative Hopf algebra with $\mathfrak{g}_{\mathcal A}$ the Lie algebra of infinitesimal characters with values in Laurent series. Pick $L_0\in\mathfrak{g}_{\mathcal A}$ and set $X=2\lambda^{-n+2m}L_0$. Then the solution of $$\label{8:1gen}
\frac{dL}{dt}=[L,M], \ \ \ M=
R(\lambda^{-n+2m} L)$$ with initial condition $L(0)=L_0$ is given by $$\label{e:8.3gen}
L(t)=\mathrm{Ad}_{ G_\mathcal{A}}g_{\pm}(t)\cdot L_0,$$ where $\exp(-tX)$ has the Connes-Kreimer Birkhoff factorization\
$\exp(-tX)=g_-(t)^{-1}g_+(t)$.
\[FRrmk\] Let $\varphi$ be the Feynman rules character. We can find the Birkhoff factorization of $\varphi$ itself within this framework by adjusting the initial condition. Namely, set $L_0(\lambda)=\frac{1}{2}\lambda^{n-2m}\exp^{-1}(\varphi(\lambda)).$ Then $\exp(X) = \varphi$ by Prop. \[p:5.3\], so the solution of involves the Birkhoff factorization $ \varphi =
g_-(-1)^{-1}g_+(-1)$. Namely, we have $$L (-1 )= \frac{\lambda^{n-2m}}{2}
\mathrm{Ad}_{ G_\mathcal{A}}g_\pm (-1 )
\exp^{-1}( \varphi).$$
The Connes-Kreimer $\beta$-function
===================================
The flow of characters usually considered in quantum field theory is the renormalization group flow (RGF). In contrast, the Lax pair flow lives on the Lie algebra of the character group. Since the $\beta$-function of the RGF is an element of the Lie algebra of the $\CC$-valued characters, it is natural to examine the relationship between the Lax pair equations and the $\beta$-function. In this section, we continue to work in the general setup of Hopf algebras and character groups.
Here we consider two flows for the $\beta$-function. First, we extend the (scalar) beta function of a local character $\varphi$ (see (\[locality\])) to an infinitesimal character $\tilde \beta_\varphi$ (Lemma \[l:holo\]). This “beta character” has already appeared in the literature: $\tilde\beta_\varphi = \lambda\tilde R(\varphi)$, in the language of [@man] explained below (Lemma \[lemma:6.6\]), but it seems worth highlighting. For certain Casimirs, we show that the beta character is a fixed point of the Lax pair flow (Theorem \[t:4.5.3\]).
It is more important and more difficult to consider the flow of the $\beta$-function itself. Namely, given a character $\varphi$, we can set $L_0 =
\log(\varphi)$ and study the $\beta$-functions of the characters $\varphi(s)
= \exp(L(s)).$ In Theorem \[t:flow\], we give a differential equation for $\beta_{\varphi(s)}.$
To define the beta character, we recall material from [@ck2; @ef-manchon; @man]. Throughout this section, $\mathcal A$ denotes the algebra of Laurent series.
Let $\mathcal H=\bigoplus\limits_{n}\mathcal H_n$ be a connected graded Hopf algebra. Let $Y$ be the biderivation on $\mathcal H$ given on homogeneous elements by $$Y:\mathcal H_n\to \mathcal H_n,\ \ \ \ \ Y(x)=nx\ \ \ \text{for }
x \in\mathcal H_n.$$
\[def:6.1\] [@man] We define the bijection $\tilde R:G_{\mathcal A}\to\mathfrak g_{\mathcal A}$ by $$\tilde R(\varphi)=\varphi^{-1}\star(\varphi\circ Y).$$
Consider the semidirect product Lie algebra $\tilde{\mathfrak g}_{\mathcal A}
=\mathfrak g_{\mathcal A}\rtimes{\mathbb C}\cdot Z_0,$ where $Z_0$ acts via $[Z_0,X]=X\circ Y$ for $X\in\mathfrak g_{\mathcal A}$. Let $\{\theta_t\}_{t\in\mathbb C}$ be the one-parameter group of automorphisms of $\mathcal
H$ given by $$\theta_t(x)=e^{nt}x, \text{ for }x\in\mathcal H_n.$$ Then $\varphi\ \
\mapsto
\varphi\circ\theta_t$ is an automorphism of $G_{\mathcal A}$. Let $\tilde{G}_{\mathcal A}$ be the semidirect product $$\tilde{G}_{\mathcal A}=G_{\mathcal A}\rtimes\mathbb C ,$$ with the action of $\mathbb C $ on $G_{\mathcal A}$ given by $\varphi\cdot t=\varphi\circ\theta_t$. $G_{\mathcal A}$ has Lie algebra $\mathfrak g_{\mathcal A}$.
We now define a second action of $\mathbb C$ on $G_{\mathcal A}$. For $t\in\mathbb C$ and $\varphi\in G_{\mathcal A}$ we define $\varphi^t(x)$ on an homogeneous element $x\in \mathcal H$ by $$\varphi^t(x)(\lambda)=e^{t\lambda |x|} \varphi(x)(\lambda),$$ for any $\lambda\in\mathbb C$, where $|x|$ is the degree of $x$.
Let $$\label{locality}
G^{\Phi}_{\mathcal A}=\{\varphi\in G_{\mathcal A}\ \big|\ \frac{ \ \ d}{dt}
(\varphi^t)_-=0\},$$ be the set of characters with the negative part of the Birkhoff decomposition independent of $t$. Elements of $ G^{\Phi}_{\mathcal A}$ are called [*local characters*]{}.
The dimensional regularized Feynman rule character $\varphi $ is local. Referring to [@ck2; @ef-manchon], the physical meaning of locality is that the counterterm $\varphi_-$ does not depend on the mass parameter $\mu$: $\frac{\partial\varphi_- }{\partial\mu}=0$.
Let $\varphi\in G_{\mathcal A}^{\Phi}$. Then the limit $$F_\varphi(t)=\lim\limits_{\lambda\to 0}\varphi ^{-1}(\lambda)\star
\varphi^t (\lambda)$$ exists and is a one-parameter subgroup in $G_{\mathcal A}\cap G_{\mathbb C}$ of scalar valued characters of $\mathcal H$.
Notice that $(\varphi ^{-1}(\lambda)\star \varphi^t (\lambda))(\Gamma)\in\mathcal A_+$ as $$\varphi ^{-1}(\lambda)\star \varphi^t
(\lambda)=\varphi_+^{-1}\star\varphi_-\star
(\varphi^t)_-^{-1}\star(\varphi^t)_+=
\varphi_+^{-1}\star(\varphi^t)_+.$$
For $\varphi\in G^\Phi_{\mathcal A}$, the $\beta$-function of $\varphi$ is defined to be $\beta_\varphi=-(\mathrm{Res}(\varphi_-))\circ Y)$.
We have [@ck2] $$\beta_\varphi={d\over dt}\Big|_{t=0} F_{\varphi_-^{-1}}(t),$$ where $F_{\varphi_-^{-1}}$, the one-parameter subgroup associated to $\varphi_-^{-1}$, also belongs to $G^\Phi_\mathcal{A}$.
To relate the $\beta$-function $\beta_\varphi\in {\mathfrak g}_\CC$ to our Lax pair equations, which live on ${\mathfrak g}_{\mathcal A}$, we can either consider ${\mathfrak g}_\CC$ as a subset of ${\mathfrak g}_{\mathcal A}$, or we can extend $\beta_\varphi$ to an element of ${\mathfrak g}_{\mathcal A}.$ Since ${\mathfrak g}_\CC$ is not preserved under the Lax pair flow, we take the second approach.
For $\varphi\in G_{\mathcal A}^{\Phi}$, $x\in H$, set $$\tilde{\beta}_\varphi(x)(\lambda) ={ d\over dt}\Big|_{t=0}
(\varphi^{-1}\star\varphi^t)(x)(\lambda).$$
The following lemma establishes that $\tilde\beta$ is an infinitesimal character.
\[l:holo\] Let $\varphi\in G_{\mathcal A}^{\Phi}$.
i\) $\tilde{\beta}_\varphi $ is an infinitesimal character in $\mathfrak g_{\mathcal A}$.\
ii) $\tilde{\beta}_\varphi $ is holomorphic (i.e. $\tilde{\beta}_\varphi(x)\in\mathcal A_+ $ for any $x$).
i\) For two homogeneous elements $x,y\in\mathcal H$, we have: $$\varphi^t(xy)=e^{t|xy|\lambda}\varphi(xy)=
e^{t|x|\lambda}\varphi(x)e^{t|y|\lambda}\varphi(y)=\varphi^t(x)\varphi^t(y).$$ Therefore $\varphi\star\varphi^t\in G_{\mathcal A}$. Since $\varphi^{-1}\star\varphi^0=e$ we get $${d\over dt}\Big|_{t=0} \varphi^{-1}\star\varphi^t\in\mathfrak
g_{\mathcal A}.$$
ii\) Since ${d\over dt}(\varphi^t)_-=0$, we get $$\tilde{\beta}_\varphi=(\varphi_+)^{-1}\star\varphi_-\star((\varphi^t)_-)^{-1}\star(\varphi^t)_+
=(\varphi_+)^{-1}\star(\varphi^t)_+.$$ Then $$\tilde{\beta}_\varphi(x)=(\varphi_+)^{-1}(x')(\varphi^t)_+(x'')
=(\varphi_+)(S(x'))(\varphi^t)_+(x'')$$ Therefore $\tilde{\beta}_\varphi(x)\in\mathcal A_+$.
\[lemma:6.6\] If $\varphi\in G^\Phi_{\mathcal{A}}$ then
\(i) $\tilde\beta_\varphi=\lambda\tilde R(\varphi)$,
\(ii) $\beta_\varphi=\mathrm{Ad}(\varphi_+(0))(\tilde\beta_\varphi\big|_{\lambda=0})$,
\(iii) $\tilde{\beta}_{\varphi_-}(x)(\lambda = 0) =-\beta_\varphi(x).$
\(i) For $\Delta(x)=x'\otimes x''$, we have $$\tilde{\beta}_\varphi(x)(\lambda) ={ d\over dt}\Big|_{t=0}
(\varphi^{-1}\star\varphi^t)(x)(\lambda)=
\varphi^{-1}(x'){ d\over dt}\Big|_{t=0}(\varphi^t)(x'')$$ $$=
\varphi^{-1}(x')\lambda\cdot
\mathrm{deg}(x'')\varphi(x'')=\lambda\varphi^{-1}(x')\varphi\circ
Y(x'')=\lambda(\varphi^{-1}\star(\varphi\circ Y))(x)=\lambda\tilde
R(\varphi)(x)
.$$
\(ii) The cocycle property of $\tilde R$ [@ef-manchon], $\tilde
R(\phi_1\star\phi_2)=\tilde R(\phi_2)+\phi_2^{-1}\star\tilde
R(\phi_1)\star\phi_2$, implies that $$\begin{aligned}
\label{e:e:6.6}
\lambda\tilde R(\varphi)=\lambda\tilde
R(\varphi_-^{-1}\star\varphi_+)=\lambda\tilde
R(\varphi_+)+\varphi_+^{-1}\star\lambda\tilde
R(\varphi_-^{-1})\star\varphi_+.\end{aligned}$$ Since $\tilde
R(\varphi_+)=\varphi_+^{-1}\star(\varphi_+\circ Y)$ is always holomorphic and since $\lambda\tilde
R(\varphi_-^{-1})=\mathrm{Res}(\varphi_-^{-1})\circ
Y=-\mathrm{Res}(\varphi_-)\circ
Y=\beta$ by [@man Theorem IV.4.4], when we evaluate (\[e:e:6.6\]) at $\lambda=0$ we get $\tilde\beta(\varphi)\big|_{\lambda=0}=\mathrm{Ad}(\varphi_+^{-1}(0))\beta$.
\(iii) The Birkhoff decomposition of $\varphi_-=(\varphi_-)_-^{-1}\star(\varphi_-)_+$ is given by $(\varphi_-)_-=\varphi_-^{-1}$ and $(\varphi_-)_+=\varepsilon$. By definition, $\beta_{\varphi_-}=-\mathrm{Res}((\varphi_-)_-)\circ Y
=-\mathrm{Res}(\varphi_-^{-1})\circ Y
=\mathrm{Res}(\varphi_- )\circ Y=-\beta_\varphi
$. Applying (ii) to $\varphi_-$, we get $$-\beta_{\varphi} = \beta_{\varphi_-}
=\mathrm{Ad}(\varepsilon\big|_{\lambda=0})
(\tilde\beta_{\varphi_-}\big|_{\lambda=0})
=\tilde\beta_{\varphi_-}\big|_{\lambda=0}.$$
If $\varphi\in G_{\mathcal A}^\Phi$, the Lax pair equation in Corollary \[tc:8.2\] for $L_0=\tilde{\beta}_\varphi $ is $$\begin{aligned}
\label{p:beta-flow-h}
{d\over ds}\tilde{\beta}_\varphi(s)=[\tilde{\beta}_\varphi(s),M],\end{aligned}$$ where $M=R(\lambda^{-n+2m}\tilde{\beta}_\varphi(s))$ and the solution is given by $$\begin{aligned}
\label{p:beta-flow}
\tilde{\beta}_\varphi(s)=\mathrm{Ad}(g_+(s))\tilde{\beta}_\varphi(0)\end{aligned}$$ for $g_\pm(s)$ given by the Birkhoff decomposition $\exp(s\lambda^{-n+2m}\tilde{\beta}_\varphi)=g^{-1}_-(s)\star
g_+(s).$
The next theorem shows that the $\beta$-function is a fixed point of the Lax pair flow for certain Casimirs. Of course this is not the same as having the $\beta$-function a fixed point of the RGF.
\[t:4.5.3\] $\tilde{\beta}_\varphi(s)$ and therefore $\beta_\varphi(s)=\mathrm{Ad}(\varphi_+(s)\Bigl|_{\lambda=0}
)(\tilde{\beta}_\varphi(s)\Big|_{\lambda=0})$ are constant under the Lax flow if $-n+2m\geq 0$.
We drop $s$ from the notation. If $-n+2m\geq 0$ then $$M=R( \lambda^{-n+2m}\tilde{\beta}_\varphi)=
\lambda^{-n+2m}\tilde{\beta}_\varphi,$$ since $\tilde{\beta}_\varphi$ is holomorphic by Lemma \[l:holo\] and Theorem \[p:holo\]. (The proof of this Theorem is independent of this section.) So the Lax pair equation becomes $${d\over ds}\tilde{\beta}_\varphi
=[ \tilde\beta_\varphi, \lambda^{-n+2m}\tilde{\beta}_\varphi]
=
\lambda^{-n+2m}[\tilde{\beta}_\varphi, \tilde{\beta}_\varphi]=0.$$
Now we consider the more interesting case of the flow $\beta_{\psi(s)}$ of the $\beta$-function of exponentiated infinitesimal characters. We first establish some simple properties of $\varphi^t$.
Let $\varphi\in G_{\mathcal A}$.
\(i) $(\varphi\star\psi)^t=\varphi^t\star\psi^t$,
\(ii) $(\varphi^{-1})^t=(\varphi^t)^{-1}$,
We have $$\begin{aligned}
(\varphi\star\psi)^t(x)&=&e^{t|x|\lambda}(\varphi\star\psi)
(x)=\sum\limits_{(x)}e^{t|x|\lambda}\varphi(x')\psi(x'') \\
&=&\sum\limits_{(x)}e^{t(|x'|+|x''|)\lambda}\varphi(x')\psi(x'')
=e^{t|x'|\lambda}\varphi(x')\ e^{t|x''|\lambda}\psi(x'')
=\varphi^t(x')\psi^t(x'')\\
&=&(\varphi^t\star\psi^t)(x).
\end{aligned}$$ Therefore $$\varphi^t\star(\varphi^{-1})^t
=(\varphi\star\varphi^{-1})^t=\varepsilon^t=\varepsilon=\varphi^t\star(\varphi^t)^{-1},$$so $(\varphi^{-1})^t=(\varphi^t)^{-1}$.
The exponential map $\exp:{\mathfrak g}_{\mathcal A}\to G_{\mathcal A}$ is a bijection. Therefore, we can transfer the Lax pair flow on ${\mathfrak g}_{\mathcal
A}$ to a flow on $G_{\mathcal A}$, and study the associated flow of beta characters.
\[t:flow\] Let $\varphi\in G^\Phi_\mathcal A$. Let $$\dot\psi(s)=[\psi(s),M]$$ be the Lax pair from Theorem \[t:8.2\] with $\psi(0)=\psi = \log(\varphi)$. Let $\varphi(s)=\exp(\psi(s))$. For $$\tilde{\beta}_{\varphi(s)}=
{d\over dt}\Big|_{t=0}\varphi(s)^{-1}\star(\varphi(s))^t,$$ we have $$\begin{aligned}
\label{e:t:phi}
{d\over ds}
\tilde{\beta}_{\varphi(s)}&=&
[\tilde{\beta}_{\varphi(s)},\varphi^{-1}(s)\star
d\exp[\log\varphi(s),M]]\\
&&\qquad +\lambda(\varphi^{-1}(s)\star d\exp[\log\varphi(s),M])\circ Y.\nonumber\end{aligned}$$
Omitting some stars, we have $$\begin{aligned}
\label{e:beta1}
{d\over ds}
\tilde{\beta}_{\varphi(s)}(x)
\nonumber &=&
{d\over ds} {d\over dt}\Big|_{t=0}(\varphi^{-1}(s)\star\varphi^t(s))(x)\\
\nonumber &=&
{d\over dt}\Big|_{t=0}( -\varphi^{-1}(s) {d\over ds}\exp\psi(s)\varphi^{-1}(s)\varphi^t(s)
+\varphi^{-1}(s) ({d\over ds}\exp\psi(s))^t)(x)\\
\nonumber&=&
{d\over dt}\Big|_{t=0}( -\varphi^{-1}(s)
d\exp\dot\psi(s)\varphi^{-1}(s)\varphi^t(s)
+\varphi^{-1}(s) (d\exp\dot\psi(s))^t)(x)\\
\nonumber &=&
( -\varphi^{-1}(s)
d\exp[\psi(s),M]\tilde{\beta}_{\varphi(s)}
+ {d\over dt}\Big|_{t=0}\varphi^{-1}(s)(d\exp[\psi(s),M])^t)(x).\\\end{aligned}$$ The last term in is $$\begin{aligned}
\label{e:beta2}
\nonumber
\lefteqn{\frac{d}{dt}\Big|_{t=0}\left(\varphi^{-1}(s)(d\exp[\psi(s),M])^t\right)(x)}\\
\nonumber &=&
{d\over
dt}\Big|_{t=0}\left(\varphi^{-1}(s)\varphi(s)^t
(\varphi(s)^t)^{-1}(d\exp[\psi(s),M])^t\right)(x)\\
\nonumber &=&
{d\over dt}\Big|_{t=0}\left(\varphi^{-1}(s)\varphi(s)^t\right)\
\left((\varphi(s)^t)^{-1}(d\exp[\psi(s),M])^t\right)\Big|_{t=0}(x)\\
\nonumber && +\left(\varphi^{-1}(s)\varphi(s)^t\right)\Big|_{t=0}\ \
{d\over dt}\Big|_{t=0}\left((\varphi(s)^{-1})^{t}(d\exp[\psi(s),M])^t\right)(x)\\
\nonumber &=&
\left(\tilde{\beta}_{
\varphi(s)}\star(\varphi(s)^{-1}d\exp[\psi(s),M])\right)(x)
+{d\over dt}\Big|_{t=0}\left((\varphi(s)^{-1}(d\exp[\psi(s),M]))^t\right)(x)\\
\nonumber &=&
\left(\tilde{\beta}_{
\varphi(s)}\star(\varphi(s)^{-1}d\exp[\psi(s),M])\right)(x)
+|x|\lambda\left(\varphi(s)^{-1}(d\exp[\psi(s),M])\right)(x)\\
\nonumber &=&
\left(\tilde{\beta}_{
\varphi(s)}\star(\varphi(s)^{-1}d\exp[\psi(s),M])\right)(x)
+\left(\lambda(\varphi(s)^{-1}(d\exp[\psi(s),M]))\circ Y\right)(x)\nonumber\end{aligned}$$ Substituting back into we get .
\[cor:beta\] Let $\varphi\in G^\Phi_\mathcal A$. Let $\dot\psi(s)=[M,\psi(s)]$ be the Lax pair from Theorem \[t:8.2\] with $\psi(0)=\psi = \log (\varphi)$. Let $\varphi(s)=\exp(\psi(s))$ and assume that $\varphi(s)\in
G^\Phi_{\mathcal A}$ for all $s$. Then $$\begin{aligned}
{d\over ds}
\beta_{\varphi(s)} &=&
\mathrm{Ad}(\varphi(s)_+(0))\left([\tilde{\beta}_{\varphi(s)},\varphi^{-1}(s)\star
d\exp[\log(\varphi(s)),M]]_+ \Big|_{\lambda=0}\right.\\
&&\qquad \left. +
\mathrm{Res}\;\left((\varphi^{-1}(s)\star
d\exp[\log(\varphi(s)),M])\circ Y\right)\right).\end{aligned}$$
This follows from the previous Theorem and Lemma \[lemma:6.6\].
In general, $\varphi\in G^\Phi_{\mathcal A}$ does not imply $\varphi(s)= \exp(\psi(s))\in G^\Phi_{\mathcal
A}$ for all $s$ (see Theorem \[t:h3\]). A simple example with $\varphi(s)\in
G^\Phi_{\mathcal A}$ is given by a holomorphic $\varphi$ (i.e $\varphi(x)\in\mathcal A_+$) with $-n+2m=0$. Indeed $(\varphi^t)_-=\varepsilon$ as $\varphi^t$ is holomorphic, so $(\varphi^t)_-$ does not depend on $t$. From the Taylor series of the exponential, $\exp(-s\log(\varphi))$ has only a holomorphic part so $g_-(s)=\varepsilon$. Therefore the solutions $\psi(s)$ of the Lax pair equation are constant, so $\varphi(s)=\varphi(0)\in
G^\Phi_{\mathcal A}$.
The Lax pair flow and the renormalization group flow {#s:rge}
====================================================
The Lax pair flow lives on the Lie algebra $\mathfrak g_{\mathcal A}$ of infinitesimal characters, while the beta character flow is on the Lie group $G_{\mathcal A}$ of characters. Theorem \[t:flow\] and Corollary \[cor:beta\] show that under the exponential map $\exp:{\mathfrak
g}_{\mathcal A}\to G_{\mathcal A}$, the corresponding flow of beta characters and $\beta$-functions are not in Lax pair form. The main point of this section is that the bijection $\tilde R^{-1}
:{\mathfrak g}_{\mathcal A}\to G_{\mathcal A}$ of [@man] is much better behaved: under $\tilde R^{-1}$, local characters remain local under the Lax pair flow (Theorem \[t:7.9\]), and the beta characters and the $\beta$-functions satisfy Lax pair equations (Theorems \[t:corr\]). In contrast, we give a rooted trees example of the nonlocality of the Lax pair flow of characters using the exponential map.
The pole order under the Lax pair flow
--------------------------------------
To begin, we investigate the dependence of the pole order of the Lax pair flow $L(t)$ on the pole order of the initial condition $L_0$ and the Casimir function (e.g. the functions $\psi_{m,n}$). In the rooted trees case, the computations are considerably simplified using the normal coordinates of [@chr], which we refer to for details.
Let $H$ be the Hopf algebra of rooted trees and $\mathcal{T}$ the set of trees. We choose a order on $\mathcal{T}=\{t_i\}_{i\in\mathbb{N}}$, such that $deg(t_i)\leq
deg(t_j)$ for any $i<j$ and such that $h(t_i)\geq h(t_j)$ for any trees $t_i,t_j$ with $deg(t_i)=deg(t_j)$ and $i<j$. Here $deg(t)$ is the number of vertices of $t$, and $h(t)$ is the height of the tree $t$, the length of the path from the root to the deepest node in the tree. For example, we can choose $$t_0=1_\mathcal{T}, \;\;\;\;
t_1= \ta1, \;\;\;\;
t_2= \tb2, \;\;\;\;
t_3= \tc3, \;\;\;\;
t_4=\td31, \;\;\;\;
t_5=\te4, \;\;\;\;
t_6=\tf41, \;\;\;\;
t_7=\thj44, \;\;\;\;
t_8=\th43$$
We recall that the $\exp:\mathfrak g_{\mathcal A}\to G_\mathcal{A}$ is bijective with inverse $\log:G_\mathcal{A}\to\mathfrak g_{\mathcal A}$ given by $$\log(\varphi)=\sum_{k=1}^\infty
(-1)^{k-1}\frac{(\varphi-\varepsilon)^{k}}{k}.$$ Set $f_0= 1_\mathcal{T}$ and let $\{f_i\}_{i\in\mathbb{N^*}}$ be the normal coordinates, i.e. $f_i$ is the forest in $H$ satisfying $$\log(\varphi)(t_i)=(\varphi-\varepsilon)(f_i),$$ for every character $\varphi$. For example, $$f_1=\ta1\; , \;\;\;\; f_2=\tb2-\frac{1}{2}\ta1\ta1 \; ,\;\;\;\;
f_3=\tc3-\ta1\tb2+\frac{1}{3}\ta1^3\; ,\;\;\;\;
f_4=\td31-\ta1\tb2+\frac{1}{6} \ta1^3\;.$$ $$f_5=\te4-\ta1\tc3-\frac{1}{2}\tb2^2+\ta1^2\tb2-\frac{1}{4}\ta1^4\; ,\;\;\;\;
f_8=\th43-\frac{3}{2}\ta1\td31+\frac{1}{2}\ta1^2\tb2
\;.$$
For a ladder tree $t$, the forest $f$ given by $\log(\varphi)(t )=(\varphi-\varepsilon)(f )$ for every character $\varphi$, is a primitive element of the Hopf algebra $H$.
We identify $\varphi\in G_\mathcal{A}$ with $\{\varphi(f_i)\}_{i\in\mathbb{N}}\in\mathcal{A}^\mathbb{N}$ and call the $\varphi(f_i)$ the $i$-component of $\varphi$. Since $\varphi(f_0)=1$ for all $\varphi$, we drop the $0$-component. We use Sweedler’s notation for the reduced coproduct $\tilde\Delta(x)=x'\otimes x''$, where $\tilde\Delta(x)=\Delta(x)-x\otimes 1_{\mathcal{T}}- 1_{\mathcal{T}}\otimes
x$. Notice that $\deg(x')+\deg(x'')=\deg(x)$ and $1\leq\deg(x'),\ \deg(x'')<\deg(x).$ For $x\neq 1_{\mathcal{T}}$ and $\tilde\Delta(x')=(x')'\otimes (x')''$, we have $$\begin{aligned}
((\varphi_1\varphi_2)\varphi_1^{-1})(x)
&=& \langle
(\varphi_1\varphi_2)\otimes\varphi_1^{-1},
x\otimes1_{\mathcal{T}}
+1_{\mathcal{T}}\otimes x
+ x'\otimes x''\rangle\\
&=& (\varphi_1\varphi_2)(x)+\varphi_1^{-1}(x)
+(\varphi_1\varphi_2)(x')\varphi_1^{-1}(x'')\\
&=& \varphi_1(x)+\varphi_2(x)
+\varphi_1(x')\varphi_2(x'')
+\varphi_1^{-1} (x)+\big(\varphi_1(x')+\varphi_2(x')\\
&&\qquad
+\varphi_1((x')')\varphi_2((x')'')\big)\varphi_1^{-1}(x'')\\
&=& \varphi_2(x)
+\varphi_1(x')\varphi_2(x'')
+ \big(\varphi_1(x)+\varphi_1^{-1}(x)
+\varphi_1(x')\varphi_1^{-1}(x'')\big)\\
&&\qquad
+\varphi_2(x' )\varphi_1^{-1}(x'')
+\varphi_1((x')')\varphi_2((x')'') \varphi_1^{-1}(x'')\\
&=& \varphi_2(x)
+\varphi_1(x')\varphi_2(x'')
+\varphi_2(x' )\varphi_1^{-1}(x'')
+\varphi_1((x')')\varphi_2((x')'') \varphi_1^{-1}(x'')\end{aligned}$$ Differentiating with respect to $\varphi_2$ and setting $L=\dot\varphi_2$ gives the adjoint representation: $$\begin{aligned}
\label{e:adjoint-repres}
\mathrm{Ad}(\varphi_1)(L)(x)&=&L(x)
+\varphi_1(x')L(x'')
+L(x' )\varphi_1(S x'')\\
&&\qquad +\varphi_1((x')')L((x')'') \varphi_1(Sx''),\nonumber\end{aligned}$$ where $S$ is the antipode of the Hopf algebra.
\[p:holo\] i) If the initial condition $L_0\in\mathfrak{g}_{\mathcal{A}}$ is holomorphic in $\lambda$, then the solution $L(t)={\Ad}(g_+(t))L_0$ of the Lax pair equation is holomorphic in $\lambda$.\
ii) If $L_0\in\mathfrak{g}_\mathcal{A}$ has a pole of order $n$, then $L(t)={\Ad}(g_+(t))L_0$ has a pole of order at most $n$.
By , we have $$\begin{aligned}
\label{e:adjoint-repres2}
\mathrm{Ad}(g_+(t))(L_0)(x)&=& L_0(x)
+g_+(t)(x')L_0(x'')
+L_0(x')g_+(t)(Sx'')\\
&&\qquad +g_+(t)((x')')L_0((x')'') g_+(t)(Sx'').\nonumber\end{aligned}$$ Notice that $g_+(t)(x)$ is holomorphic for $x\in H$. If $L_0$ is holomorphic, then every term of the right hand side of is holomorphic, so $\mathrm{Ad}(g_+(t))(L_0)$ is holomorphic. Since multiplication with a holomorphic series cannot increase the pole order, $L(t)$ cannot have a pole order greater than the pole order of $L_0$.
We can also use normal coordinates to measure the nontriviality of the Lax pair flow.
\[t:ladder\] If $f_i$ is a primitive element of $H$ (e.g. $f_i$ corresponds to a ladder tree), then the $i$-component $L(t)(f_i)$ of the Lax pair flow is does not depend on $t$.
It is shown in [@chr] that $\varphi^{-1}(f_i)=-\varphi(f_i),$ for every character $\varphi$ and $i\geq 1$. Since $f_i$ is a primitive element, the inner automorphism $C_g:G_{\mathcal{A}}\to G_{\mathcal{A}}$, $C_g(h)=ghg^{-1}$, satisfies $C_g(h)(f_i)=g(f_i)+h(f_i)-g(f_i)=h(f_i).$ Therefore $(Ad(g)L_0)(f_i)=(L_0)_i,$ where $(L_0)_i$ is the i-component of $L_0$.
Thus everything of interest in the Lax pair flow occurs off the primitives, e.g. the normal coordinate $f_4$ corresponding to $\td31$ is the first component in the Hopf algebra on which the Lax pair is nonconstant.
The Lax pair flow, the RGE flow, and locality
---------------------------------------------
We now investigate whether the Lax pair flow can ever be identified with the RGF. Some identification is necessary: the RGF $(\varphi^{t})_+(\lambda=0)$ lives in the Lie group of characters $G_{\mathbb C}$, while the Lax pair flow $L(t)$ lives in a Lie algebra $\mathfrak g_\mathcal{A}$ To match these flows, we can transfer the Lax pair flow to the Lie group level using either of the maps $\tilde
R^{-1}$ and $\exp$, namely by defining $$\label{phi-chi}
\varphi_t=\tilde R^{-1}(L(t))\ \ \ \text{and} \ \ \ \chi_t=\exp(L(t))$$ and then setting $\lambda =0.$
The most naive hope would be that $\varphi_t$ or $\chi_t$ coincide with the RGF $\varphi^t$, perhaps after a rescaling of the parameter $t$. We shall see that this fails even in the trivial case. As usual, we take $\mathcal A$ to be the algebra of Laurent series.
On a commutative, cocommutative, graded connected Hopf algebra $\mathcal H$, $\varphi_t\not=\varphi^t$ and $\chi_t\not=\varphi^t$.
If $\tilde R(\varphi^t)=L(t)$ then by Definition \[def:6.1\] $$(\varphi^t\circ Y)(\Gamma)=(\varphi^t\star L(t))(\Gamma)$$ for every $\Gamma\in \mathcal H$. For a primitive homogenous element $\Gamma\in {\mathcal H}_n$ we get $$\label{7.4a}
|\Gamma|e^{|\Gamma|t\lambda}\varphi(\Gamma)
=e^{|\Gamma|t\lambda}\varphi(\Gamma)L(t)(1)
+e^{0|\Gamma|\lambda}\varphi(\Gamma)L(t)(\Gamma).$$ Therefore $$L(t)(\Gamma)=|\Gamma|e^{|\Gamma|t\lambda}\varphi(\Gamma).$$ Since ${\mathcal H}$ is cocommutative, its Lie bracket is abelian. Thus the left hand side of (\[7.4a\]) is constant in $t$, while the right hand side is not.
The same argument works for $\chi_t$ on $\mathcal H$.
In a positive direction, we will show that locality of characters is preserved under the Lax pair flow, using the identification given by $\tilde R.$ This indicates that $\tilde R$ is more useful than the exponential map.
Recall from [@man Theorem IV.4.1] that $\lambda\tilde R:G_\mathcal{A}\to\mathfrak{g}_{\mathcal{A}}$ restricts to a bijection from $G^\Phi_\mathcal{A}$ to $\mathfrak{g}_{\mathcal{A}+}$, where $\mathfrak{g}_{\mathcal{A}+}$ is the set infinitesimal characters on $\mathcal H$ with values in $\mathcal
A_+$. In this sense, $\lambda \tilde R$ is better behaved than $\tilde R$, as the following locality result shows.
\[t:7.9tau\] For a local character $\varphi\in G^\Phi_\mathcal{A}$, let $L(t)$ be the solution of the Lax pair equation (\[8:1\]) with initial condition $L_0=\lambda\tilde R(\varphi)$ and any $\mathrm{Ad}$-covariant function $f$. Let $\tau_t$ be the flow of characters given by $$\tau_t=(\lambda\tilde R)^{-1}(L(t)).$$ Then $\tau_t$ is a local character for all $t$.
By [@ef-manchon], $$\tilde R(\varphi\star\xi)=\tilde R(\xi)+\xi^{*-1}\star\tilde R(\varphi)\star
\xi.$$ Taking $\xi=g_+(t)^{-1}$ and multiplying by $\lambda$, we get $$\lambda\tilde R(\varphi\star g_+(t)^{-1})=\lambda\tilde R(g_+(t)^{-1})+
g_+(t)\star\lambda\tilde R(\varphi)\star
g_+(t)^{-1}.$$ Since $\varphi\in G^\Phi_\mathcal{A}$ and $g(t)_+^{-1}$ is an element in $G_\mathcal{A}$ without polar part, by [@man Lemma IV.4.3.], $\varphi\star g(t)_+^{*-1}$ is local. Thus $\lambda\tilde R(\varphi\star g_+(t)^{-1})\in{\mathcal{A}_+}$. We have $g(t)_+^{-1}\in G^\Phi_\mathcal{A}$, simply because $g(t)_+^{-1}$ does not have a polar part, so $\lambda\tilde R (g(t)_+^{-1})$ is holomorphic. It follows that $$\tau_t=(\lambda\tilde R)^{-1}({\Ad}(g_+(t))L_0)=(\lambda\tilde R)^{-1}(
g_+(t)\star L_0\star g_+(t)^{-1})\in G^\Phi_\mathcal{A}$$
We can now show that locality of characters is preserved under the Lax pair flow via the $\tilde R$ identification.
\[t:7.9\] For a local character $\varphi\in G^\Phi_\mathcal{A}$, let $L(t)$ be the solution of the Lax pair equation for any $\mathrm{Ad}$-covariant function $f$, with the initial condition $L_0=\tilde R(\varphi)$. Let $\varphi_t$ be the flow given by $$\varphi_t=\tilde R^{-1}(L(t)).$$ Then $\varphi_t$ is a local character for all $t$.
We show that the flow $\tau_t$ constructed in the previous Proposition with the initial condition $L_0=\lambda\tilde
R(\varphi)$, for the $\mathrm{Ad}$-covariant function $h:\mathfrak{g}_\mathcal{A}\to\mathfrak{g}_\mathcal{A}$ given by $h(L)=f(\lambda^{-1} L)$, is equal to the flow $\varphi_t$ constructed for the $\mathrm{Ad}$-covariant function $f$, with the initial condition $L_0=\tilde
R(\varphi)$. We have $$\tau_t=(\lambda\tilde R)^{-1}\left(g_+(t)\star\lambda\tilde
R(\varphi)\star g_+^{-1}(t)\right)=\tilde R^{-1}\left(g_+(t)\star \tilde
R(\varphi)\star g_+^{-1}(t)\right),$$ where $g_+(t)$ is given by the Birkhoff decomposition of $$\exp(-t f( \tilde R(\varphi)))= \exp(-t h(\lambda\tilde
R(\varphi))).$$ Therefore the two $g_+(t)$ involved in the definitions of $\varphi_t$ and $\tau_t$ coincide, so $\varphi_t=\tau_t$.
In contrast to Theorem \[t:flow\], it is immediate that the flow of beta characters associated to $\tilde R$ is in Lax pair form.
\[l:rminusoneversion\] For a local character $\varphi\in G^\Phi_\mathcal{A}$, let $\varphi_t$ be the flow from Theorem 7.5. Then $$\begin{aligned}
\label{e:rminusoneversion}
\frac{d\tilde\beta_{\varphi_t}}{dt}=[\tilde\beta_{\varphi_t},M],
$$
By Lemma 6.7, we get $\tilde\beta_{\varphi_t}=\lambda\tilde R(\varphi_t)=\lambda L(t)$. Then $$\frac{d\tilde\beta_{\varphi_t}}{dt}= \frac{d(\lambda\tilde
R(\varphi_t))}{dt}=
\frac{\lambda d
L(t)}{dt}=\lambda[L(t),M]=[\tilde\beta_{\varphi_t},M].$$
The corresponding $\beta$-functions also satisfy a Lax pair equation.
\[t:corr\] For a local character $\varphi\in G^\Phi_\mathcal{A}$, let $L(t)$ be the Lax pair flow of Corollary 5.10 with initial condition $L_0=\tilde R(\varphi)$. Let $\varphi_t=\tilde R^{-1}(L(t))$. Then
- for $-n+2m\geq 1$, $\varphi_t = \varphi$ and hence $\beta_{\varphi_t}=\beta_{\varphi}$ for all $t$.
- for $-n+2m\leq 0$, $\beta_{\varphi_t}\in
\mathfrak{g}_{\mathbb C}$ satisfies $$\frac{d\beta_{\varphi_t}}{dt}=\big[\beta_{\varphi_t},
-\frac{d((\varphi_t)_+(0))}{dt}((\varphi_t)_+(0))^{-1} +
2\mathrm{Ad}((\varphi_t)_+(0))\big(\mathrm{Res}(\lambda^{-n+2m-2}\tilde\beta_{\varphi_t})\big)
\big].$$
By Theorem 7.5, $\varphi_t$ are local characters, so by [@man Theorem IV.4.], $\tilde\beta_{\varphi_t}=\lambda
L(t)=\lambda \tilde R (\varphi_t)$ is holomorphic.
\(i) If $-n+2m\geq 1$, then $\lambda^{-n+2m}L(t)$ is holomorphic, which implies $$M=R(\lambda^{-n+2m}L(t))=\lambda^{-n+2m}L(t)=
\lambda^{-n+2m-1}\tilde\beta_{\varphi_t}.$$ $L(t)$ satisfies the Lax pair equation $$\frac{dL}{dt}=[L,M]=[L,\lambda^{-n+2m}L]=\lambda^{-n+2m}[L,L]=0.$$ Thus $L(t)=L_0$ for all $t$, which gives $\varphi_t=\tilde
R^{-1}(L(t))=\tilde R^{-1}(L_0)=\varphi$ for all $t$.
\(ii) For $-n+2m\leq 0$, we have $$\begin{aligned}
M &=&
R(\lambda^{-n+2m}L(t))=\lambda^{-n+2m}L(t)-2P_-(\lambda^{-n+2m}L(t))\\
&=&
\lambda^{-n+2m-1}\tilde\beta_{\varphi_t}-2P_-(\lambda^{-n+2m-1}\tilde\beta_{\varphi_t}).\end{aligned}$$ (\[e:rminusoneversion\]) becomes $$\frac{d\tilde\beta_{\varphi_t}}{dt}=
-2[\tilde\beta_{\varphi_t},P_-(\lambda^{-n+2m-1}\tilde\beta_{\varphi_t})]
=-2[P_+(\lambda^{-n+2m-1}\tilde\beta_{\varphi_t}),\lambda^{n-2m+1}
P_-(\lambda^{-n+2m-1}\tilde\beta_{\varphi_t})]
.$$ Expand $\tilde\beta_{\varphi_t}$ as $$\tilde\beta_{\varphi_t}=\sum_{k=0}^{\infty} \tilde\beta_k(t)\lambda^k.$$ Then $$\frac{d\tilde\beta_{\varphi_t}}{dt}=-2\left[\sum_{k=n-2m+1}^\infty
\tilde\beta_{k}(t)\lambda^{k-n+2m-1},\sum_{j=0}^{n-2m}\tilde\beta_j(t)\lambda^j\right],$$ and evaluating at $\lambda=0$ gives $$\begin{aligned}
\frac{d\tilde\beta_0(t)}{dt}=-2[\tilde\beta_{n-2m+1}(t),\tilde\beta_0(t)]
=2[\tilde\beta_0(t),\tilde\beta_{n-2m+1}(t)].\end{aligned}$$ Using the facts that Ad$(g)$ is a Lie algebra homomorphism and $(d/dt){\rm Ad}(g(t))X$\
$ = [(dg(t)/dt)g^{-1}(t), {\rm Ad}(g(t))X]$ (see the proof of Theorem \[t:7.9tau\]), we get $$\begin{aligned}
\frac{d\beta_{\varphi_t}}{dt}&=&
\frac{d}{dt}\left(\mathrm{Ad}(\varphi_t)( \tilde\beta_0(t)) \right)\nonumber\\
& =&
\left[\frac{d((\varphi_t)_+(0))}{dt}((\varphi_t)_+(0))^{-1},
\mathrm{Ad}(\varphi_t)(\tilde\beta_0(t))\right]
+
\mathrm{Ad}((\varphi_t)_+(0))\left(\frac{d\tilde\beta_0(t)}{dt}\right)\\
&=&
\left[\frac{d((\varphi_t)_+(0))}{dt}((\varphi_t)_+(0))^{-1},\beta_{\varphi_t}\right]
+2\left[\mathrm{Ad}((\varphi_t)_+(0))\tilde\beta_0(t),
\mathrm{Ad}((\varphi_t)_+(0))\tilde\beta_{n-2m+1}(t)\right]\nonumber\\
&=&\left[\beta_{\varphi_t},-\frac{d((\varphi_t)_+(0))}{dt}((\varphi_t)_+(0))^{-1
}+2\mathrm{Ad}((\varphi_t)_+(0))\left(\mathrm{Res}(\lambda^{-n+2m-2}\tilde\beta_{\varphi_t})
\right)\right],\nonumber \end{aligned}$$ since $\tilde\beta_{n-2m+1}(t)=\mathrm{Res}(\lambda^{-n+2m-2}\tilde\beta_{\varphi_t})$.
Local characters satisfy the abstract Renormalized Group Equation [@0609035], which we now recall. For a local character $\varphi\in G^\Phi_{\mathcal A}$, the renormalized character is defined by $\varphi_{\mathrm{ren}}(t)=(\varphi^t)_+(\lambda=0)$.
\[aRGE\] For $\varphi\in G^\Phi_{\mathcal A}$, the renormalized character $\varphi_{\mathrm{ren}}$ satisfies the abstract Renormalized Group Equation: $$\frac{\partial}{\partial t}
\varphi_{\mathrm{ren}}(t)=\beta_\varphi\star\varphi_{\mathrm{ren}}(t).$$
Here our parameter $t$ corresponds to $e^t$ in [@0609035].
In light of Theorem \[t:7.9\], we can ask for the relation between $(\varphi_t)_{\mathrm{ren}}(s)$ and $\varphi_{\mathrm{ren}}(s)$ corresponding to $\varphi_t$ and $\varphi$. In §9, we consider a toy model character on a Hopf algebra of rooted trees and show that these renormalized characters differ.
We can also show that for certain initial conditions, the flow $\tau_t$ is constant.
If $\varphi\in G^\Phi_\mathcal{A}$ and $\varphi_+=\varepsilon$ (i.e. $\varphi$ has only a pole part), then the flow $\tau_t$ of Proposition \[t:7.9tau\] for the $\mathrm{Ad}$-covariant function $f(L)=\lambda^{-n+2m}L$ has $\tau_t=\varphi$ for all $t$.
If we show that either $g_\pm(t)=\varepsilon$, then $$\tau_t=(\lambda\tilde R)^{-1}\big(g_\pm(t)\star\lambda\tilde R
(\varphi)\star g_\pm(t)^{-1}\big)=(\lambda\tilde R)^{-1}\big(\varepsilon\star\lambda\tilde R
(\varphi)\star\varepsilon^{-1}\big)=\varphi.$$ $g_\pm(t)$ are given by the Birkhoff decomposition of $$\begin{aligned}
\label{e:expforX}
g(t)=\exp(-2t\lambda^{-n+2m}L_0)=\sum_{k=0}^\infty
\frac{(-2t\lambda^{-n+2m})^k L_0^{k}}{k!},\end{aligned}$$ where $L_0=\lambda\tilde R(\varphi)\in\mathfrak{g}_{\mathbb C}$ [@man Theorem IV.4.4]. We split the problem into two cases depending on the sign of $-m+2n$. If $-m+2n\geq 0$, then $g(t)(x)\in\mathcal A_+$ for any $x$, which implies $g_-(t)=\varepsilon$. Similarly, if $-m+2n< 0$, then $g(t)(x)\in\mathcal A_-$ for any $x$, which implies $g_+(t)=\varepsilon$. Notice that the right hand side of (\[e:expforX\]) is a finite sum, namely up to $k=\mathrm{deg}(x)$ when evaluated on $x\in \mathcal H$.
We next study the locality of the flow $\chi_t$ defined in (\[phi-chi\]) for the usual Lax pair flow $L(t).$ . Thus for an initial character $\varphi$ and $L_0 = \log(\varphi)$, $$\chi_t =\exp(g_+(t)\star L_0\star g_+(t)^{-1})=g_+(t)\star
\exp(L_0)\star g_+(t)^{-1}=g_+(t)\star \varphi\star g_+(t)^{-1},$$ with $\chi_0=\varphi$. As before, in normal coordinates $\chi_t$ is trivial on primitives.
\[l:constant\] If $\varphi\in G^\Phi_{\mathcal A} $ and $f_i$ is a primitive element, then $ \chi_t(f_i)$ does not depend on $t$.
$\chi_t(f_i)=g_+(t)(f_i)+\varphi(f_i)-g_+(t)(f_i)=\varphi(f_i)$.
We now present some calculations showing the interplay between the Lax pair flow and locality.
For the first example, we construct a nontrivial Hopf subalgebra on which $\chi_t$ is local. Let $\mathcal H^{2}$ be the Hopf subalgebra generated by the following trees $$t_0=1_\mathcal{T}, \;\;\;\;
t_1= \ta1, \;\;\;\;
t_2= \tb2, \;\;\;\;
t_4=\td31,$$ together with any set of ladder trees. The first normal coordinate of $\chi_t$ to depend on $t$ is $f_4$, corresponding to $\td31$. Let $G^2_\mathcal{A}$ be the group of characters associated to the $\mathcal H^2$.
\[t:7.9a\] For $\varphi\in G^{2\, \Phi}_{\mathcal A}$, let $\chi_t$ be the flow of characters on $\mathcal H^2$ given by $$\chi_t=\exp(L(t)),$$ where $L(t)$ is the solution of the Lax pair equation (\[8:1\]) for any $\mathrm{Ad}$-covariant function with the initial condition $L_0=\log(\varphi)$. Then $\chi_t$ is local for all $t$.
Let $\pi$ denote the projection to the pole part of a Laurent series. By [@chr], $\tilde\Delta(f_4)=f_1\otimes
f_2-f_2\otimes f_1$, so $$(\chi_t^s)_-(f_4)=-\pi\big(\chi_t^s(f_4)+(\chi_t^s)_-(f_1)\chi_t^s(f_2)
-(\chi_t^s)_-(f_2)\chi_t^s(f_1)\big).$$ Subtracting from this equation the corresponding equation for $t=0$, and remembering that $\chi_t(f_1)$ and $\chi_t(f_2)$ do not depend on $t$ by Lemma \[l:constant\], we get $$(\chi_t^s)_-(f_4)=(\varphi^s)_-(f_4)-\pi(\chi_t^s(f_4)-\varphi^s(f_4)).$$ We have $$\begin{aligned}
\label{e:pole}
\pi(\chi_t^s(f_4)-\chi^s(f_4))
&=& \pi
\big(
e^{3\lambda s}
(
-2g_+(t)(f_2)\varphi(f_1)
+2g_+(t)(f_1)\varphi(f_2)
)
\big)\\
&=& \pi
\big(
- 2g_+(t)(f_2)\pi(e^{3\lambda s}\varphi(f_1))
+ 2g_+(t)(f_1)\pi(e^{3\lambda s}\varphi(f_2))
\big).\end{aligned}$$ Since $\varphi\in G^{2\, \Phi}_\mathcal{A}$, both $\pi(e^{s\lambda}\varphi(f_1))=-(\varphi^s)_-(f_1)$ and $\pi(e^{2s\lambda}\varphi(f_2))=-(\varphi^s)_-(f_2)$ are independent of $s$. By rescaling $s$, $\pi(e^{3s\lambda}\varphi(f_1))$ and $\pi(e^{3s\lambda}\varphi(f_2))$ are independent of $s$. Therefore $\pi(\chi_t^s(f_4)-\varphi^s(f_4))$ is independent of $s$, which finishes the proof.
We can apply the previous proposition to the Hopf subalgebra of Feynman diagram generated by the empty graph and the graphs $$\begin{aligned}
A_1= \p\ ,\ A_2= \pdp\ ,\ A_3= \pdpdp\ ,\
A_4= \pddpp\ ,\ A_5= \pdpdpdp\end{aligned}$$ and with $\varphi$ the Feynman rules character. The characters $\chi_t$ restricted to this Hopf algebra are all local.
To investigate how $\chi_t$ fails to be local on a non-ladder tree with a larger number of vertices, we consider the Hopf subalgebra $\mathcal H^3$ generated by $$t_0=1_\mathcal{T}, \;\;\;\;
t_1= \ta1, \;\;\;\;
t_2= \tb2, \;\;\;\;
t_3=\tc3, \;\;\;\;
t_4=\td31, \;\;\;\;
t_6=\tf41, \;\;\;\;
t_7=\thj44, \;\;\;\;
t_8=\th43$$ together with any set of ladder trees. Let $f_i$ be the corresponding normal coordinates.
The next lemma gives the pole order of a local character $\varphi\in G^\Phi_{\mathcal A}$ on primitives.
\[lemma:7.2\] If $\varphi$ is a local character and $f_i$ is primitive, then both $\varphi(f_i)$ and $L_0(f_i)=\log(\varphi)(f_i)$ have a pole of order at most one.
If ${\rm deg}(f_i)=d$ and $\varphi=\sum_{k=-n}^\infty \varphi_k\lambda^k$ is the Laurent expansion of $\varphi$, then $$(\varphi^s)_-(f_i)=-\pi(e^{sd\lambda}\varphi(f_i))=
-\pi(
\varphi_{-n}(f_i)\lambda^{-n}+(\varphi_{-n}(f_i)sd+\varphi_{-n+1}(f_i))\lambda^{-n+1}
+o(\lambda^{-n+2}))$$ If $\varphi$ has a pole, then $\varphi^s_- = \varphi_-$ implies $-n+1=0$.
\[t:h3\] Let $\varphi$ be a local character on $\mathcal H^3$ and let $\chi_t$ be the flow of characters given by $$\chi_t=\exp(L(t)),$$ where $L(t)$ is the solution of the Lax pair equation (\[8:1\]) with the initial condition $L_0=\log(\varphi)$. Then $\chi_t$ is local on $\mathcal H^3$ for all $t$ if and only if either $$\begin{aligned}
\label{e:condition:chi}
\varphi_-(f_1)=0\ \ \text{or} \ \
3\varphi_-(f_1)(\varphi_+(f_2)\big|_{\lambda=0})=\varphi_-(f_2)(\varphi_+(f_1)\big|_{\lambda=0}).\end{aligned}$$
The point is that (\[e:condition:chi\]) is unlikely to hold.
By [@chr], $$\tilde\Delta(f_8)=\frac{3}{2}f_1\otimes f_4-\frac{3}{2}f_4\otimes f_1
-\frac{1}{2}f_1\otimes f_1 f_2-\frac{1}{2}f_1f_2\otimes f_1
+\frac{1}{2}f_1f_1\otimes f_2+\frac{1}{2}f_2\otimes f_1f_1.$$
Since $(\chi^s_t)_-(f_4)$ does not depend on $s$ (Prop. \[t:7.9a\]) and since $\chi_t(f_1)$ and $\chi_t(f_2)$ do not depend on $t$, after cancelations of terms involving only the primitives $f_1$ and $f_2$, we get $$\begin{aligned}
\label{e:tf8}
\lefteqn{
(\chi^s_t)_-(f_8)-(\varphi^s)_-(f_8)}\\
&=&
-\pi
\Big(\chi^s_t(f_8)-\varphi^s(f_8)\\
&&\qquad +\frac{3}{2}\varphi_-(f_1)(\chi^s_t(f_4)-\varphi^s(f_4))
-\frac{3}{2}\varphi^s(f_1)((\chi_t)_-(f_4)-\varphi_-(f_4))
\Big)\nonumber\end{aligned}$$
By Lemma \[lemma:7.2\], $\varphi(f_1)$ and $\varphi(f_2)$ have poles of order at most one. Set $$\varphi(f_1)=\sum_{k=-1}^\infty a_k\lambda^k\ \ \ \text{and}\ \ \
\varphi(f_2)=\sum_{k=-1}^\infty b_k\lambda^k.$$
From the proof of Proposition \[t:7.9a\], $$\begin{aligned}
\label{e:chi1}
\chi_t^s(f_4)-\varphi^s(f_4))
=
e^{3\lambda s}
(
-2g_\pm(t)(f_2)\varphi(f_1)
+2g_\pm(t)(f_1)\varphi(f_2)
)\end{aligned}$$ We have $g_-(t)(f_1)=\exp(-2t\lambda^{-n+2m}L_0)_-(f_1)=-\pi(-2t\varphi(f_1))=2ta_{-1}\lambda^{-1}$, and\
$g_+(t)(f_1)=-2t(a_0+a_1\lambda+o(\lambda^2))$. Similarly $g_-(t)(f_2)=2tb_{-1}\lambda^{-1}$. (\[e:chi1\]) becomes $\chi_t^s(f_4)-\varphi^s(f_4))=4te^{3\lambda
s}\big((-b_{-1}a_0+a_{-1}b_0)\lambda^{-1}+o(\lambda^0)\big)$, which implies $$\begin{aligned}
\lefteqn{
\pi\big(\frac{3}{2}\varphi_-(f_1)(\chi^s_t(f_4)-\varphi^s(f_4))
-\frac{3}{2}\varphi^s(f_1)((\chi_t)_-(f_4)-\varphi_-(f_4))\big)}\\
&=&
\frac{3}{2}\pi\big((-a_{-1}\lambda^{-1}e^{3s\lambda}(4t((-b_{-1}a_0+a_{-1}b_0)
\lambda^{-1}+o(\lambda^0)))\big)\\
&&\qquad
-\frac{3}{2}\pi\big(e^{s\lambda}(a_{-1}\lambda^{-1}+o(\lambda^0))(4t)(-1)(-b_{-1}a_0
+a_{-1}b_0)\lambda^{-1}\big)\\
&=& -12sta_{-1}(-b_{-1}a_0+a_{-1}b_0)\lambda^{-1}+P(\lambda^{-1}),\end{aligned}$$ where $P(\lambda^{-1})$ is some polynomial in $\lambda^{-1}$ independent of $s$. We get $$\begin{aligned}
\label{e:tf8s}
\lefteqn{ \pi
\Big(\chi^s_t(f_8)-\varphi^s(f_8)
\Big)}\\
& =&
\pi
\big(
e^{4s\lambda}
(
( -3g_+(t)(f_1)g_+(t)(f_2)-3g_+(t)(f_3)
)\varphi(f_1)
)
\big)\\
&&\qquad +3\pi
\big(e^{4s\lambda}
( g_+(t)(f_1) )^2\varphi(f_2)
\big)
+\pi
\big(3e^{4s\lambda}
g_+(t)(f_1)\varphi(f_4)
\big)\nonumber\end{aligned}$$ Since $\varphi(f_1)$ and $\varphi(f_2)$ have poles of order at most one, the first two terms of the right hand side of (\[e:tf8s\]) do not depend on $s$. Since $\varphi\in G^{3\, \Phi}_\mathcal{A}$, $$\varphi^s_-(f_4)=-\pi\big(e^{3s\lambda}\varphi(f_4)+\varphi_-(f_1)e^{2s\lambda}\varphi(f_2)
-\varphi_-(f_2)e^{s\lambda}\varphi(f_1))\big)$$ is independent of $s$, and so $\pi(e^{3s\lambda}\varphi(f_4)-a_{-1}b_{-1}s\lambda^{-1})$ is also independent of $s$. By rescaling $s$, $\pi(3e^{4s\lambda}\varphi(f_4)-4a_{-1}b_{-1}s\lambda^{-1})$ does not depend on $s$. In conclusion, the terms independent of $s$ in $-(\chi^s_t)_-(f_8)+(\varphi^s)_-(f_8)$ are $$-12sta_{-1}(-b_{-1}a_0+a_{-1}b_0)\lambda^{-1} +
4a_{-1}b_{-1}s\lambda^{-1}(-2ta_0).$$ Therefore $\chi^s_t(f_8)$ is independent of $s$ if and only if either $a_{-1}=0$ or $3a_{-1}b_0=a_0b_{-1}$. Similar computations hold for the normal coordinates $f_6$ and $f_7$.
The choice $-n+2m=0$ in the Proposition is just for the sake of concreteness. A more detailed analysis reveals the following:
- If $-n+2m\geq 1$ then $\chi_t$ is local on $\mathcal H^3$ without any additional conditions. Indeed, in this case $L_0(f_1)$ and $L_0(f_2)$ are holomorphic and thus $g_-(t)(f_1)=g_-(f_2)=0$. which implies that $\chi^s_t(f_4)-L_0(f_4)=0$ for every $t$. By (\[e:tf8\]) we get that $(\chi_t^s)_-(f_8)$ does not depend on $s$. Similar statements hold for $f_6$ and $f_7$.
- If $-n+2m=-1$, the situation is similar to Proposition \[t:h3\], namely $\chi_t$ is local on $\mathcal H^3$ if and only if $$\varphi_-(f_1)(\varphi_+(f_2)\big|_{\lambda=0})=\varphi_-(f_2)(\varphi_+(f_1)\big|_{\lambda=0})$$ and either $$\varphi_-(f_1)=0\ \ \text{or} \ \
3\varphi_-(f_1)(\frac{\partial\varphi_+(f_2)}{\partial\lambda}\big|_{\lambda=0})
=\varphi_-(f_2)(\frac{\partial\varphi_+(f_1)}{\partial\lambda}\big|_{\lambda=0}).$$
For $-n+2m\in\Z^+$, the flows $\chi_t$ and $L(t)$ gain locality, in the sense that they become constant on larger Hopf subalgebras as $-n+2m$ increases. Indeed, $\chi_t$ and $L(t)$ are constant on the Hopf algebra generated by the primitives (e.g. the normal coordinates associated to ladder trees. In contrast, if we decrease $-n+2m<0$, we preserve locality only when an increasing number of conditions are fulfilled.
The Lax pair flow of the $\beta$-function
-----------------------------------------
Recall from §6 that the beta characters for the exponentiated Lax pair flow $\exp(L(t))$ do not themselves satisfy a Lax pair equation. In the next theorem, we reverse this procedure by taking a Lax pair flow $L(t)$ starting at the $\beta$-function of a character, and then producing characters $\xi_t$ whose $\beta$-functions are $L(t)\big|_{\lambda = 0}.$
\[t:7.14\] Let $\varphi\in G^\Phi_{\mathcal A}$ and let $L(t)$ be the flow given by Theorem \[t:8.2\] with the initial condition $L_0=\beta_\varphi.$ Let $\xi_t=(\lambda \tilde{R})^{-1}(L(t)\big|_{\lambda=0})$. Then $\xi_t$ is local for all $t$. The $\beta$-function of $\xi_t$ satisfies $$\beta_{\xi_t} = L(t)\big|_{\lambda =0}.$$ Moreover, $\xi(0)= \varphi_-$.
Since $L_0=\beta_\varphi$ is scalar valued, by Theorem \[p:holo\] $L(t)=\mathrm{Ad}(g_+(t))(L_0)$ is holomorphic. Therefore. $L(t)$ can be evaluated at $\lambda=0$. $L(t)\big|_{\lambda=0}$ is also scalar valued, so by [@man Theorem IV.4.4], $\xi_t\in G^\Phi_{\mathcal A-}$, the set of local characters taking values in $\lambda^{-1}\mathbb C[\lambda^{-1}]$. In particular, $\xi_t$ is local. By Lemma \[lemma:6.6\], $\tilde
\beta_{\xi_t} = (\lambda \tilde R)(\xi_t) = L_t\big|_{\lambda = 0}$, so $\tilde\beta_{\xi_t}$ must be constant in $\lambda.$ This implies $$\beta_{\xi_t} = \tilde\beta_{\xi_t}\big|_{\lambda =0} = L(t)\big|_{\lambda =0}.$$ It follows from the Connes-Kreimer scattering formula [@ck2; @man] that $\xi(0)=\varphi_-$.
Worked examples {#worked-example}
================
In this section we give some explicit computations on two Hopf algebras which illustrate results in previous sections. We first consider the Hopf algebra $\mathcal H^1$ generated by the following trees: $$t_0=1_\mathcal{T}, \;\;\;\;
t_1= \ta1, \;\;\;\;
t_2= \tb2, \;\;\;\;
t_4=\td31, \;\;\;\;
t_8=\th43,$$ and the regularized toy model character $\varphi=\varphi(q,\mu,\lambda)$ (see [@krei99; @krde99]) given on trees by $$\varphi(T)(q,\mu,\lambda)=(q/\mu)^{-\lambda\mathrm{deg(T)}}\prod_{v}
B_{w(T_v)}(\lambda).$$ Here the product is taken over all vertices $v$ of the tree $T$, $w(T_v)$ is the number of vertices of the subtree $T_v$ of $T$ which has $v$ as a root, and $B_j(\lambda)=B(j\lambda,1-j\lambda)$ for $j\in\mathbb N^*$, with $B$ the Euler beta function. Referring to [@ef-manchon], $q$ is interpreted as a dimensional external parameter, and $\mu$ is the ’t Hooft mass. $\varphi$ has enough similarity of realistic QFT calculations to be worth considering [@bk99; @krde99; @krei99; @krei00]. Set $b=q/\mu$ and $a=\log(b)$. Thus terms in $a$ (or $\log(q^2/\mu^2) = 2a$ as in e.g. [@ky]) are the leading log terms in the various expansions. We have $\varphi(\ta1)=b^{-\lambda}B_1(\lambda)$, $\varphi(\tb2)=b^{-2\lambda}B_2(\lambda)B_1(\lambda)$, $\varphi(\td31)=b^{-3\lambda}B_3(\lambda)B_1(\lambda)^2$, $\varphi(\th43)=b^{-4\lambda}B_4(\lambda)B_1(\lambda)^3$, etc. In the normal coordinates $f_i$, the Laurent series of $\varphi(f_i)$ (cf. [@chr]) are given by $$\varphi(f_1)=\frac{1}{\lambda}-a+o(\lambda),\ \ \ \ \
\varphi(f_2)=\frac{\pi^2}{4}+o(\lambda),
$$ $$\varphi(f_4)=\frac{7\pi^2}{36 \lambda} -\frac{7\pi^2a}{12} +
o(\lambda), \ \ \
\varphi(f_8)=\frac{\pi^2}{12 \lambda^2}
-\frac{\pi^2a}{3\lambda} + o(\lambda^0).$$ The character $\varphi$ is local, and the Lax pair flow on $\mathcal H^1$ $\varphi_t=\tilde R^{-1}(L(t))$ as in Corollary \[tc:8.2\], with $L(t)=Ad(g_\pm(t))L_0$, $-m+2n=0$, and initial condition $L_0=\tilde R(\varphi)$ is given by $$\varphi_t(f_1)=\varphi(f_1),\ \ \ \ \varphi_t(f_2)=\varphi(f_2),\ \
\ \varphi_t(f_4)=\frac{\pi^2 (7 + 24 t)}{36 \lambda}
-\frac{\pi^2}{12} (7 + 16 t)a + o(\lambda),$$ $$\begin{aligned}
\varphi_t(f_8)&=&\frac{\pi^2 (1 + 6 t)}{12 \lambda^2} - \frac{\pi^2(2 + 15 t + 9
t^2) a}{6 \lambda}\\
&&\qquad + \frac{\pi^2}{144} (\pi^2 (83 + 288 t + 126 t^2)
+ 12 (8 + 51 t + 27 t^2) a^2)+ o(\lambda).\end{aligned}$$ The flow $L(t)$ has poles of order at most one: $$L(t)(f_1)=\frac{1}{\lambda}-a+o(\lambda),\ \ \
L(t)(f_2)=\frac{\pi^2}{2}+o(\lambda),$$ $$L(t)(f_4)=
\frac{\frac{\pi^2}{3} + 2 \pi^2 t}
{\lambda}
+ (-\pi^2 - 4 \pi^2 t) a +o(\lambda),$$ $$L(t)(f_8)=\frac{-2 (\pi^2 t (2 + 3 t)a)}{\lambda}
+\frac{\pi^2}{6}
(\pi^2 (7 + 37 t + 21 t^2) + 6 t (8 + 9 t) a^2)+o(\lambda).$$ This confirms that $\lambda L(t)=(\lambda\tilde
R)(\varphi_t)=\tilde\beta_{\varphi_t}$ is holomorphic, which implies that $\varphi_t$ is local on $\mathcal H^1$. It can be explicitly checked that $(\varphi_t^s)_-$ does not depend on $s$: $$(\varphi_t^s)_-(f_1)=-\frac{1}{z}, \ \ \ \ \ \
(\varphi_t^s)_-(f_2)=0, \ \ \ \ \ \ \
(\varphi_t^s)_-(f_4)=\frac{\pi^2(1-12t)}{18\lambda},$$ $$(\varphi_t^s)_-(f_8)=-\frac{\pi^2(1-12t)}{24 \lambda^2}+\frac{3\pi^2t(1+t)a}{2\lambda}.$$ The Connes-Kreimer $\beta$-functions $\beta_{\varphi_t}=-{\rm Res}(\varphi_t)_-\circ Y$ are: $$\beta_{\varphi_t}(f_1)=1,\ \ \ \beta_{\varphi_t}(f_2)=0,\ \ \
\beta_{\varphi_t}(f_4)=-\frac{\pi^2}{6} + 2 \pi^2 t, \ \ \ \
\beta_{\varphi_t}(f_8)=-6 \pi^2 t (1 + t)a.$$ The associated RGFs $(\varphi_t)_{\mathrm{ren}}(s)=(\varphi_t^s)_+\big|_{\lambda=0}$, which all satisfy the abstract RGE, are $$(\varphi_t)_{ren}(s)(f_1)=(s - a), \ \ \ \ \ (\varphi_t)_{ren}(s)(f_2)=\frac{\pi^2}{4},$$ $$(\varphi_t)_{ren}(s)(f_4)=
\frac{\pi^2}{12}(s + 24 s t - (1 + 16 t)a),$$ $$(\varphi_t)_{ren}(s)(f_8)=\frac{\pi^2}{96}
(12 s^2 + \pi^2 (11 + 136 t + 84 t^2) - 24 s (1 + 20 t + 24 t^2)a + 12 (1 + 22 t + 18 t^2)
a^2).$$
Thus in this model there is a polynomial dependence in $t$ of the leading log terms in each of $L(t), \beta_{\varphi_t}$ and $(\varphi_t)_{\rm ren}$, although the first diagram with a nonzero leading log term differs. By the recursion formula in [@ky (26)], the next to ... leading log terms in the Green’s functions will then also depend polynomially on $t$. We conjecture that this polynomial dependence extends to the Feynman rules character on the full Hopf algebra.
We emphasize that the Renormalized Group Flows $(\varphi_t)_{\mathrm{ren}}$ and $\varphi_{\mathrm{ren}}(s)=
(\varphi_t)_{\mathrm{ren}}(s)\big|_{t=0}$ of the characters $\varphi_t$ and $\varphi$ are different. While $\varphi_t(f_1),
\varphi_t(f_2)$ are independent of $t$, we have $$\label{8:1new}
\varphi_t(f_3)=\varphi(f_3)-4t(\varphi_-(f_1)\varphi(f_2)-\varphi_-(f_2)\varphi(f_1).$$ We have used $\varphi_t=\tilde
R^{-1}(\mathrm{Ad}(g_-(t))\tilde R(\varphi))$ for this calculation, since it is easier to extract the pole part of a Laurent series than the holomorphic part, but we could also use $\varphi_t=\tilde
R^{-1}(\mathrm{Ad}(g_+(t))\tilde R(\varphi))$. In this case, we get $$\label{8:2}
\varphi_t(f_3)=\varphi(f_3)-4t(\varphi_+(f_1)\varphi(f_2)-\varphi_+(f_2)\varphi(f_1).$$ As a check, we verify that (\[8:1new\]) and (\[8:2\]) are equal. Let $\pi$ denote the projection onto the pole part of a Laurent series. Then $$\begin{aligned}
\varphi_+(f_1)\varphi(f_2)-\varphi_+(f_2)\varphi(f_1)
&=& (\varphi(f_1)-\pi(\varphi(f_1))\varphi(f_2)
(\varphi(f_2)-\pi(\varphi(f_2))\varphi(f_1)\\
&=& -\pi(\varphi(f_1)\varphi(f_2) -(-\pi(\varphi(f_2)))\varphi(f_1)\\
&=&\varphi_-(f_1)\varphi(f_2)-\varphi_-(f_2)\varphi(f_1).\end{aligned}$$
The computations for the character $\chi_t$ in (\[phi-chi\]) associated to the toy model character $\varphi$ with $-n+2m=0$ give $$(\chi_t^s)_-(f_1)=-\frac{1}{\lambda},\ \ \ \ \ (\chi_t^s)_-(f_1)=0, \
\ \ \ \
(\chi_t^s)_-(f_4)=-\frac{-\frac{\pi^2}{18}+\pi^2t}{\lambda},$$ $$(\chi_t^s)_-(f_8)=\frac{\pi^2}{24\lambda^2}+\frac{\pi^2t(18s+(5+18t)a)}{6\lambda}.$$ In agreement with Theorem \[t:7.9a\], $(\chi_t^s)_-$ is independent of $s$ when evaluated on $f_1$, $f_2$ and $f_4$. However, $(\chi_t^s)_-(f_8)$ depends on $s$, so $\chi_t$ is not local. We confirm that the necessary condition (\[e:condition:chi\]) for locality in Theorem \[t:h3\] does not hold. Indeed, $\varphi_-(f_1)=1/z\not= 0$ and $3(-1/z)(\pi^2)\not=0\cdot a$. The $\beta$-function on $f_1$, $f_2$, $f_4$ is given by $$\beta_{\chi_t}(f_1)=1, \ \ \ \ \ \beta_{\chi_t}(f_2)=0,\ \ \
\ \ \
\beta_{\chi_t}(f_4)=-\frac{\pi^2}{6}+3\pi^2t.$$ The renormalized character $(\chi_t)_{\mathrm{ren}}(s)$ is given by $$(\chi_t)_{\mathrm{ren}}(s)(f_1)=s-a, \ \ \
(\chi_t)_{\mathrm{ren}}(s)(f_2)=\pi^2/4, \ \ \
(\chi_t)_{\mathrm{ren}}(s)(f_4)=\frac{\pi^2}{12}(s+36st-(1+24t)a),$$ and satisfies the abstract RGE.
Let $\mathcal H^2$ be the Hopf subalgebra generated by the trees $$t_0=1_\mathcal{T}, \;\;\;\;
t_1= \ta1, \;\;\;\;
t_2= \tb2, \;\;\;\;
t_3= \tc3, \;\;\;\;
t_4=\td31, \;\;\;\;
t_5=\te4
.
$$ For $T\in \{t_1,\ldots , t_{5}\}$, let $Z_T$ be the corresponding infinitesimal character. The Lie algebra $\mathfrak g_2$ of scalar valued infinitesimal characters of $\mathcal H^2$ is generated by $Z_{t_1},\ldots,\ Z_{t_{5}}$. Let $G_1$ be the scalar valued character group of $\mathcal H^2$, and let $G_0$ be the semi-direct product $G_1\rtimes{\mathbb C}$ given by $$(g,t)\cdot (g',t')=(g\cdot\theta_t( g'),t+t'),$$ where $\theta_t(g)(T)=e^{t\mathrm{deg}(T)}g(T)$ homogenous $T$. Define a new variable $Z_0$ with $[Z_0,Z_{t_i}] = \mathrm{deg}(t_i)Z_{t_i},$ so formally $Z_0=\frac{d}{d\theta}$. The Lie algebra $\mathfrak g_0$ of $G_0$ is generated by $Z_0,Z_{t_1},\ldots,\ Z_{t_{5}}$.
The conditions a) and b) in Definition \[bialg\] of a Lie bialgebra can be written in a basis as a system of quadratic equations. We can solve this system explicitly, e.g. via Mathematica. It turns out that there are $43$ families of Lie bialgebra structures $\gamma$ on $\mathfrak g_0$. In more detail, the system of quadratic equations involves 90 variables. Mathematica gives 1 solution with 82 linear relations (and so 8 degrees of freedom), 7 solutions with 83 linear relations, 16 solutions with 84 linear relations, 13 solutions with 85 linear relations, 5 solutions with 86 linear relations, and 1 solution with 87 linear relations.
To any Lax equation with a spectral parameter, one can associate a spectral curve and study its algebro-geometric properties (see [@sts]). In our case, we consider the adjoint representation $\mathrm{ad}:\delta\to
{\mathfrak gl}(\delta)$ and the induced adjoint representation of the loop algebra. The spectral curve is given by the characteristic equation of $\mathrm{ad}(L\lambda)$: $\Gamma_0=\{ (\lambda,\nu)\in{\mathbb C}-\{0\}\times{\mathbb C}\ | \
\det(\mathrm{ad}(L(\lambda)-\nu {\mathrm{ Id} }))=0\}$.
The theory of the spectral curve and its Jacobian usually assumes that the spectral curve is irreducible. For all 43 families of Lie bialgebra structures on $\delta$, on the associated Lie algebra $\mathrm{ad}(\delta)$ all eigenvalues of the characteristic equation are zero, and the zero eigenspace is nine dimensional. The spectral curve itself is the union of degree one curves. Thus each irreducible component has a trivial Jacobian, and the spectral curve theory breaks down. The integrability of these Lax pair equation remains open for future investigations.
Gabriel Baditoiu would like to thank the Max-Planck-Institute for Mathematics, Bonn and the Erwin Schrödinger International Institute for Mathematical Physics for the hospitality. Steven Rosenberg would also like to thank ESI and the Australian National University.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
The strongest evidence for dark energy comes presently from geometric techniques such as the supernova distance-redshift relation. By combining the measured expansion history with the Friedmann equation one determines the energy density and its time evolution, hence the equation of state of dark energy. Because these methods rely on the Friedmann equation which has not been independently tested it is desirable to find alternative methods that work for both general relativity and other theories of gravity.
Assuming that sufficiently large patches of a perturbed Robertson-Walker spacetime evolve like separate Robertson-Walker universes, that shear stress is unimportant on large scales and that energy and momentum are locally conserved, we derive several relations between long-wavelength metric and matter perturbations. These relations include generalizations of the initial-value constraints of general relativity. For a class of theories including general relativity we reduce the long-wavelength metric, density, and velocity potential perturbations to quadratures including curvature perturbations, entropy perturbations, and the effects of nonzero background curvature. When combined with the expansion history measured geometrically, the long-wavelength solution provide a test that may distinguish modified gravity from other explanations of dark energy.
author:
- Edmund Bertschinger
title: On the Growth of Perturbations as a Test of Dark Energy and Gravity
---
Introduction
============
Current evidence for dark energy is based on two key assumptions:
1. The Cosmological Principle holds, i.e. on large scales the matter distribution and its expansion are homogeneous and isotropic and the spacetime geometry is Robertson-Walker; and
2. The cosmic expansion scale factor $a(t)$ obeys the Friedmann equation, which may be written $$\label{friedmann}
{\cal H}^2\equiv\left(\frac{1}{a}\frac{da}{d\tau}\right)^2=a^2H^2=
\frac{8\pi Ga^2}{3}\rho(a)-K\ ,$$
where $K$ is the spatial curvature with units of inverse length squared (we set $c=1$) and $\tau$ is conformal time, related to cosmic proper time $t$ by $dt=a(\tau)d\tau$. Usually the Friedmann equation is applied indirectly through an expression for the angular-diameter or luminosity distance; these formulae depend crucially on the expression for $\tau(a)$ obtained by integrating equation (\[friedmann\]).
The Cosmological Principle is more general than general relativity (GR) and it is amenable to direct observational test through measurements of distant objects such as galaxies, Type Ia supernovae, and the microwave background radiation. The Friedmann equation, however, is equivalent to one of the Einstein field equations of GR applied to the Robertson-Walker metric and, so far at least, it has not been independently tested. Instead, the Friedmann equation is used by astronomers in effect to deduce $\rho(a)$ including dark energy through measurements of the Hubble expansion rate $H(a)$.
If the evidence for dark energy is secure, there are four possible explanations:
1. The dark energy is a cosmological constant or, equivalently, the energy density and negative pressure of the vacuum (Gliner 1966; Zel’dovich 1967).
2. The dark energy is some other source of stress-energy, for example a scalar field with large negative pressure (Ratra & Peebles 1988; Steinhardt et al. 1999).
3. General relativity needs to be modified and we cannot use it to deduce the existence of either a cosmological constant or an exotic form of energy (Dvali et al. 2000; Lue et al. 2004).
4. General relativity is correct but our understanding of it is not; e.g. long-wavelength density perturbations modify the Friedmann equation (Kolb et al. 2005a,b).
It would be interesting to find observational tests that can distinguish among these possibilities. The first case can be tested by measuring the equation of state parameter $w\equiv p/\rho$ on large scales; a cosmological constant has $w=-1$. This test can be made using methods such as the supernova distance-redshift relation (Riess et al. 1998; Perlmutter et al. 1999) and baryon acoustic oscillations (Eisenstein et al. 2005) whose interpretation relies on the Friedmann equation. Such methods are called geometric (they measure the large-scale geometry of spacetime) or kinematic (they rely on energy conservation and on the first time derivative of the expansion scale factor).
It is more difficult to test the third possibility, namely that dark energy represents a modification of GR rather than (or in addition to) a new form of mass-energy. Redshift-distance tests invoke the Friedmann equation of GR or its equivalent in other theories in order to determine the abundance and equation of state of dark energy. Such tests do not work without a formula relating $a(\tau)$ to the cosmic energy density and pressure. While specific alternative models can be tested, it would be desirable to have general cosmological tests independent of the Friedmann equation.
The evolution of density perturbations has been proposed as an independent test of dark energy and general relativity (Linder 2005; Ishak et al. 2005). For example, weak gravitational lensing measurements, the evolution of galaxy clustering on large scales, and the abundance of rich galaxy clusters all have some sensitivity to the gravitational effects of dark energy at redshifts $z<1$. These methods may be called dynamic because the evolution equations for the perturbations are at least second order in time.
This paper examines what the linear growth of metric, density, and velocity perturbations can tell us about gravity. By generalizing previous work on evolution of separate universes (Wands et al.2000; Gordon 2005 and references therein), we will show that in GR, on length scales larger than the Jeans length (more generally, on scales large enough so that spatial gradients may be neglected in the equations of motion), the evolution of the metric and density perturbations of a background Robertson-Walker universe can be determined from the Friedmann equation and local energy-momentum conservation. Generalizing this to arbitrary theories of gravity, under certain conditions explained in this paper, the evolution of Robertson-Walker spacetimes (assuming they are solutions of the gravitational field equations) combined with local energy-momentum conservation is sufficient to determine the evolution of long-wavelength perturbations of the metric and matter variables.
Because GR is so fully integrated into most treatments of cosmological perturbation theory, and we wish to test gravitation more generally, it is worth recalling its basic elements:
1. Spacetime is describable as a classical 4-dimensional manifold with a metric locally equivalent to Minkowski. This is generalizable to higher dimensions where matter fields reside on a three-dimensional spatial brane.
2. Special relativity holds locally. In particular, energy-momentum is locally conserved. This is more general than GR.
3. The weak equivalence principle holds, i.e. freely-falling bodies follow spacetime geodesics. This is more general then GR.
4. The metric is the solution to the Einstein field equations subject to appropriate initial and boundary conditions. This is uniquely true in GR.
The first three ingredients are assumed by most viable theories of gravitation applied on cosmological scales. They are assumed to be correct throughout this paper. The Einstein field equations are not assumed to hold except where explicitly stated below. We assume throughout that the universe is approximately (or, for some calculations, exactly) Robertson-Walker.
Robertson-Walker Spacetimes and their Perturbations
===================================================
The general Robertson-Walker spacetime is specified by giving a spatial curvature constant $K$ (with units of inverse length squared) and a dimensionless scale factor $a(\tau)$ (normalized so that $a=1$ today). Let us assume that the evolution of $a(\tau)$ depends on $K$ and on the properties of the matter and energy filling the universe. Applying the Cosmological Principle, the matter must behave as a perfect fluid at rest in the comoving frame. The pressure of a perfect fluid may be written $p(\rho,S)$ where $\rho$ is the proper energy density and $S$ is the comoving entropy density in the fluid rest frame. The line element may thus be written $$\label{rw}
ds^2=a^2(\tau,K,S)\left[-d\tau^2+d\chi^2+r^2(\chi,K)d\Omega^2\right]$$ where $r(\chi,K)=K^{-1/2}\sin(K^{1/2}\,\chi)$ for $K>0$ (and is analytically continued for $K\le0$) and $d\Omega^2\equiv
d\theta^2+\sin^2\theta\,d\phi^2$. Initially we make no assumption about the dynamics except that the evolution of the scale factor depends only on the geometry ($K$) and composition ($S$) of the (3+1)-dimensional universe and not, for example, on parameters describing extra dimensions.[^1] Only later will the Friedmann equation be assumed to determine the exact form of $a(\tau,K,S)$.
Curvature and Coordinate Perturbations {#sec:cpert}
--------------------------------------
Metric perturbations are obtained by comparing two slightly different spacetimes. We consider two homogeneous and isotropic Robertson-Walker spacetimes differing only by their (spatially homogeneous) spatial curvature. The first spacetime has spatial curvature $K$; the second one has spatial curvature $K(1+\delta_K)$ where $\delta_K$ is a small constant. We write the metric of the second spacetime as a perturbation of the first, as follows. First, the angular radius may be Taylor-expanded to first order in $\delta_K$ to give $$\label{modr2}
r^2(\chi,K+K\delta_K)=
r^2(\chi,K)(1-\delta_K)+\chi r(\chi,4K)\delta_K\ ,$$ while $a(\tau,K+K\delta_K,S)=a(\tau,K,S)+\delta_K(\partial a/
\partial\ln K)$. To simplify the appearance of the line element we change variables $\tau\to\tau+\alpha(\tau)$ and $\chi\to\chi(1-\kappa)$ where $\alpha$ and $\kappa$ are assumed to be first order in $\delta_K$. In these new coordinates, to first order in $\delta_K$ the second spacetime has line element $$\begin{aligned}
\label{rw2}
ds^2&=&a^2(\tau,K,S)\left(1+2{\cal H}\alpha
+2\delta_K\frac{\partial\ln a}{\partial\ln K}\right)\\
&&\times\left\{-(1+2\dot\alpha)d\tau^2
+(1-2\kappa)d\chi^2+\left[(1-\delta_K)r^2(\chi,K)
+(\delta_K-2\kappa)\chi r(\chi,4K)\right]d\Omega^2\right\}\ ,
\nonumber\end{aligned}$$ where ${\cal H}(\tau,K,S)\equiv\partial\ln a/\partial\tau$. This line element describes a perfectly homogeneous and isotropic Robertson-Walker spacetime with spatial curvature $K(1+\delta_K)$. However, for appropriate choices of $\alpha$ and $\kappa$ (i.e., the appropriate coordinate transformation), it takes precisely the same form as a perturbed Robertson-Walker spacetime with background spatial curvature $K$ and perturbation $\Psi$, $$\label{rw3}
ds^2=a^2(\tau,K,S)\left\{-(1+2\Psi)d\tau^2+(1-2\Psi)\left[d\chi^2
+r^2(\chi,K)d\Omega^2\right]\right\}\ ,$$ provided that the following three conditions hold: $$\label{rwmatch1}
\kappa=\frac{1}{2}\delta_K\ ,\ \ 2\Psi=\dot\alpha+\kappa\ ,\ \
\Psi=\kappa\left(1-2\frac{\partial\ln a}{\partial\ln K}\right)
-{\cal H}\alpha\ .$$ In other words, a perturbed Robertson-Walker spacetime whose metric perturbation $\Psi(\tau)$ is spatially homogeneous is identical to an unperturbed Robertson-Walker spacetime represented by a perturbed coordinate system.[^2]
This equivalence is significant because it suggests that long-wavelength curvature perturbations (for which $\Psi$ is effectively independent of spatial position) should evolve like patches of a Robertson-Walker spacetime (Wands et al. 2000; Gordon 2005 and references therein), whose dynamics is more general than general relativity. So far we have assumed that the Cosmological Principle holds but we have not assumed the validity of the Einstein field equations of general relativity. However, if we know how a perfect Robertson-Walker spacetime evolves, the evolution of long-wavelength curvature perturbations follows.
To make further progress let us assume the form of $a(\tau,K,S)$, which requires specifying a theory of gravity. The simplest choice is to assume the validity of the Friedmann equation, which gives $$\label{friedsol}
\tau(a,K,S)=\int_0^a\left[\frac{8\pi}{3}G\tilde a^4\rho(\tilde a,S)
-K\tilde a^2\right]^{-1/2}\,d\tilde a\ .$$ Using this, one finds $$\label{aderivK}
\left(\frac{\partial\ln a}{\partial\ln K}\right)_{\tau,S}
=-\frac{(\partial\tau/\partial\ln K)_{a,S}}{(\partial\tau/
\partial\ln a)_{K,S}}=-\frac{1}{2}K{\cal H}
\int^\tau\frac{d\tau'}{{\cal H}^2(\tau',K,S)}\ .$$ We also require conservation of energy, which may be written $$\label{econs}
\frac{\partial}{\partial\ln a}\rho(a,S)=-3[\rho+p(\rho,S)]\ .$$ When combined with the Friedmann equation, this implies $$\label{Hdot}
\gamma\equiv4\pi Ga^2(\rho+p)={\cal H}^2+K-\dot{\cal H}
=\frac{3}{2}(1+w)({\cal H}^2+K)\ ,$$ where $\dot{\cal H}=\partial{\cal H}(\tau,K,S)/\partial\tau$. Combining equations (\[rwmatch1\]), (\[aderivK\]), and (\[Hdot\]) yields a relation between the long-wavelength potential $\Psi$ and the curvature perturbation[^3] $\kappa$: $$\label{psikappa}
\kappa=\frac{{\cal H}^2}{\gamma a^2}\frac{\partial}{\partial\tau}
\left(\frac{a^2\Psi}{{\cal H}}\right)\ .$$ For long wavelengths $\kappa$ may depend on the wavenumber $k$ but cannot depend on $\tau$, so $\partial\kappa/\partial\tau=0$, which implying $$\label{psiddotfrw}
\frac{\gamma}{{\cal H}}\frac{\partial}{\partial\tau}\left[\frac{{\cal H}^2}
{\gamma a^2}\frac{\partial}{\partial\tau}\left(\frac{a^2\Psi}{{\cal H}}
\right)\right]=\ddot\Psi+3(1+c_w^2){\cal H}\dot\Psi+3(c_w^2-w){\cal H}^2
\Psi-(2+3w+3c_w^2)K\Psi=0\ ,$$ where $$\label{eos}
w\equiv\frac{p(\rho,S)}{\rho}\ ,\ \ c_w^2\equiv\left(\frac{\partial p}
{\partial\rho}\right)_{\!S}\ .
%=w-\frac{1}{3}\frac{d\ln(1+w)}{d\ln a}\ .$$
By comparing two different Friedmann-Robertson-Walker models we have arrived at a second-order differential equation for the metrc perturbation $\Psi$. The only dynamical equations assumed have been the Friedmann equation and energy conservation in a homogeneous and isotropic universe. We have not made use of the perturbed Einstein or fluid equations. Nevertheless, as we show next, equation (\[psiddotfrw\]) is identical with the dynamical evolution equation for long-wavelength curvature perturbations obtained using the perturbed fluid and Einstein equations of general relativity.
Linear Cosmological Perturbations in General Relativity
-------------------------------------------------------
Cosmological perturbation theory has been well studied (e.g.Lifshitz 1946; Bardeen 1980; Kodama & Sasaki 1984; Hwang & Noh 2002). Nonetheless, the curvature and entropy variables relevant for long-wavelength density perturbations differ from those given previously in the literature, so a brief summary is presented here.
In the conformal Newtonian gauge (Mukhanov et al. 1992; Ma & Bertschinger 1995), the metric of a perturbed Robertson-Walker spacetime with scalar perturbations may be written as a generalization of equation (\[rw3\]), $$\label{pertrw}
ds^2=a^2(\tau)\left\{-(1+2\Phi)d\tau^2+(1-2\Psi)
[d\chi^2+r^2(\chi,K)d\Omega^2]\right\}\ .$$ Here, $\Phi(x^i,\tau)$ and $\Psi(x^i,\tau)$ are small-amplitude gravitational potentials. The dependence of the scale factor on $K$ and $S$ is suppressed because we are considering now a single universe with unique values of these parameters and are introducing perturbations only through the potentials.
For scalar perturbations the stress-energy tensor components may be written in terms of spatial scalar fields $\delta\rho$, $u$, and $\pi$, as follows,
\[emtensor\] $$\begin{aligned}
T^0_{\ \,0}&=&-(\bar\rho+\delta\rho)\ ,\ \label{t00}\\
T^0_{\ \,i}&=&-(\bar\rho+\bar p)\nabla_i u\ ,\label{t0i}\\
T^i_{\ \,j}&=&\delta^i_{\ \,j}(\bar p+\delta p)+\frac{3}{2}
(\bar\rho+\bar p)\left(\nabla^i\nabla_j-\frac{1}{3}\delta^i_{\ \,j}
\Delta\right)\pi\ ,\label{tij}\end{aligned}$$
where $\nabla_i$ and $\nabla^i$ are the three-dimensional covariant derivative for the spatial line element $d\chi^2+r^2d\Omega^2$ while $\Delta=\nabla^i\nabla_i$. Unless stated otherwise, all variables refer to the total stress-energy summed over all components. The unperturbed energy density and pressure are $\bar\rho(\tau)$ and $\bar p(\tau)$, respectively, while $\delta\rho$ and $\delta p$ are the corresponding perturbations measured in the coordinate frame.
In the scalar mode, all perturbations to the metric and the stress-energy tensor arise from spatial scalar fields and their spatial gradients. For example, the energy flux is a potential field with velocity potential $u(x^i,\tau)$. Similarly, the shear stress follows from a shear stress (viscosity) potential $\pi(x^i,\tau)$ (defined as in Bashinsky & Seljak 2004). For an ideal gas, $\pi=0$. Equations (\[emtensor\]) are completely general for the scalar mode. Stress-energy perturbations that arise from divergenceless (transverse) vectors contribute only to the vector mode, while divergenceless, trace-free tensors contribute only to the tensor mode. Vector and tensor modes are ignored in this paper.
Combining the perturbed Einstein field equations (Bertschinger 1996) with equations (\[Hdot\]) and (\[eos\]) gives the following second-order partial differential equation for the linear evolution of the potential $\Psi$: $$\begin{aligned}
\label{psieom}
\frac{\gamma}{{\cal H}}\frac{\partial}{\partial\tau}\left[
\frac{{\cal H}^2}{\gamma a^2}\frac{\partial}{\partial\tau}
\left(\frac{a^2}{{\cal H}}\Psi\right)\right]-c_w^2\Delta
\Psi
=\gamma\left[\frac{\delta p-c_w^2\delta\rho}{\bar\rho
+\bar p}+\frac{3}{{\cal H}}\frac{\partial}{\partial
\tau}\left({\cal H}^2\pi\right)+\Delta\pi\right]\ .\end{aligned}$$ Equation (\[psieom\]) is exact in linear perturbation theory of general relativity and is fully general for scalar perturbations. It was given in another form by Hwang & Noh (2002). Aside from the right-hand side and the sound speed term $c_w^2\Delta\Psi$, it agrees exactly with equation (\[psiddotfrw\]), whose derivation did not assume the validity of the perturbed Einstein field equations. The right-hand side of equation (\[psieom\]) includes entropy perturbations proportional to $\delta p-c_w^2\delta\rho$ and shear stress perturbations, both of which were absent in the Robertson-Walker models of the previous subsection.
In a perfectly homogeneous and isotropic universe the shear stress must vanish, and a correct treatment of shear stress requires going beyond the Friedmann and energy conservation equations. In the universe at low redshift, the shear stress due to photons, neutrinos, and gravitationally bound structures is orders of magnitude smaller than the mass density perturbations. Unless the dark energy is a peculiar substance with large shear stress, we may neglect $\pi$ in the equation of motion for $\Psi$.
The sound wave term represents the effect of pressure forces in resisting gravitational instability. For a perturbation of comoving wavenumber $k$, $-c_w^2\Delta\Psi=k^2c_w^2\Psi$; for comparison the time derivative terms in equation (\[psieom\]) are $\sim {\cal
H}^2\Psi$. Thus, for wavelengths much longer than the comoving Jeans length $\lambda_J$ defined by $\pi/\lambda_J={\cal H}/c_w$, the sound wave term may be neglected. In the standard cosmology, the Jeans length at $z<100$ is less than about 20 Mpc.
The entropy source term in equation (\[psieom\]) can also be obtained by comparing separate Robertson-Walker spacetimes with slightly different entropies, allowing us to reduce the long-wavelength evolution entirely to quadratures, as we show next.
Entropy Perturbations {#sec:epert}
---------------------
Consider two Robertson-Walker spacetimes with identical spatial curvature $K$ but with entropies $S$ and $S+\delta S$ respectively, where $\delta S$ is a small constant. At a given expansion factor $a$ ($a^3$ plays the role of volume), the density $\rho(a,S)$ and pressure $p(\rho,S)$ will differ slightly in the two spacetimes. Writing the pressure as $p(\rho(a,S),S)$, from equations (\[econs\]) and (\[eos\]) we obtain $$\label{pderivs}
\left(\frac{\partial p}{\partial\ln a}\right)_S=-3c_w^2(\rho+p)
\ ,\ \
\left(\frac{\partial p}{\partial S}\right)_{\!a}
=\left(\frac{\partial p}{\partial S}\right)_{\!\rho}
+c_w^2\left(\frac{\partial\rho}{\partial S}\right)_{\!a}\ .$$ Using equations (\[econs\]) and (\[pderivs\]), we find $$\label{psderiv}
\left(\frac{\partial p}{\partial S}\right)_{\!\rho}
=-\left(\frac{\rho+p}{3}\right)\frac{\partial}{\partial\ln a}
\left[\frac{1}{\rho+p}\left(\frac{\partial\rho}{\partial S}
\right)_{\!a}\,\right]\ .$$ Using this result, we define a fractional entropy perturbation variable $$\label{sigdef}
\sigma(a,S)\equiv\frac{\delta p-c_w^2\delta\rho}{\rho+p}
=\frac{\delta S}{\rho+p}\left(\frac{\partial p}{\partial S}
\right)_{\!\rho}=-\frac{\delta S}{3}\frac{\partial}{\partial
\ln a}\left[\frac{1}{\rho+p}\left(\frac{\partial\rho}{\partial S}
\right)_{\!a}\,\right]\ .$$ This is a formal result because for a multicomponent imperfect fluid one would not evaluate $\rho(a,S)$ but would instead characterize density and pressure perturbations for the individual components, as we shown below in §\[sec:multi\]. For now, we assume that $\sigma$ can be determined and we use it to derive a quadrature for the isocurvature modes.
The scale factor in the second Robertson-Walker spacetime is $a(\tau,K,S+\delta S)=a(\tau,K,S)+(\partial a/\partial S)\delta S$. Changing time variable $\tau\to\tau+\alpha(\tau)$ (with no change in spatial coordinates), the line element for the second spacetime becomes $$\label{rw4}
ds^2=a^2(\tau,K,S)\left(1+2{\cal H}\alpha+2\delta S\frac
{\partial\ln a}{\partial S}\right)\left[-(1+2\dot\alpha)
d\tau^2+d\chi^2+r^2(\chi,K)d\Omega^2\right]\ .$$ This line element describes a perfectly homogeneous and isotropic Robertson-Walker spacetime with entropy $S+\delta S$. However, for appropriate choice of $\alpha$ (i.e., the appropriate coordinate transformation), it takes precisely the same form as a perturbed Robertson-Walker spacetime with background entropy $S$, given by equation (\[rw3\]), provided that the following two conditions hold: $$\label{rwmatch2}
2\Psi=\dot\alpha\ ,\ \
\Psi=-\delta S\frac{\partial\ln a}{\partial S}-{\cal H}\alpha\ .$$
So far we have assumed nothing about gravity. To proceed further we assume that equation (\[friedsol\]) is valid, giving $$\label{aderivS}
\left(\frac{\partial\ln a}{\partial S}\right)_{\tau,K}
=-\frac{(\partial\tau/\partial S)_{a,K}}{(\partial\tau/
\partial\ln a)_{K,S}}=\frac{4\pi G{\cal H}}{3}\int^\tau
\frac{a^2(\tau',K,S)}{{\cal H}^2(\tau',K,S)}\left(
\frac{\partial\rho}{\partial S}\right)_{\!a'}\,d\tau'\ .$$ Combining equations (\[Hdot\]), (\[sigdef\]), (\[rwmatch2\]) and (\[aderivS\]) gives $$\begin{aligned}
\label{psidotfrw}
\frac{1}{a^2}\frac{\partial}{\partial\tau}\left(\frac{a^2\Psi}
{{\cal H}}\right)&=&-\delta S\frac{\partial}{\partial\tau}
\left[\frac{1}{{\cal H}}\left(\frac{\partial\ln a}{\partial S}
\right)_{\!\tau,K}\,\right]
=-\frac{4\pi Ga^2\delta S}{3{\cal H}^2}\left(\frac
{\partial\rho}{\partial S}\right)_{\!a}\nonumber\\
&=&\frac{\gamma}{{\cal H}^2}
\int^\tau\sigma(a(\tau',K,S),S){\cal H}(\tau',K,S)\,d\tau'\ .\end{aligned}$$ This result agrees exactly with equation (\[psieom\]) for long-wavelength entropy perturbations in GR with vanishing shear stress potential $\pi$. Therefore, in GR the dynamical evolution of long-wavelength entropy perturbations in a Robertson-Walker spacetime follows directly from the Friedmann and energy conservation equations without requiring the perturbed Einstein field equations. Equation (\[psidotfrw\]) implies equation (\[psiddotfrw\]) when the entropy perturbation vanishes.
General Solution for Long-Wavelength Perturbations in GR {#sec:quadGR}
--------------------------------------------------------
Equations (\[psiddotfrw\]) and (\[psidotfrw\]) have a simple exact solution. The homogeneous solution with $\sigma=0$ (curvature perturbations) is (suppressing the dependencies on $K$ and $S$, which are held fixed) $$\label{curv}
\Psi(\tau)=\kappa\Psi_+(\tau)+C\Psi_-(\tau)\ ,\ \
\Psi_+(\tau)=\frac{{\cal H}}{a^2}\int^\tau\frac{\gamma(\tau')
a^2(\tau')}{{\cal H}^2(\tau')}\,d\tau'\ ,\ \
\Psi_-(\tau)=\frac{{\cal H}}{a^2}\ .$$ Here $\kappa$ is the curvature perturbation of equation (\[psikappa\]) and it gives the amplitude of the “growing mode” of density perturbations; because the lower limit of integration for $\Psi_+(\tau)$ is unspecified one may add any constant multiple $C$ of the “decaying mode” solution $\Psi_-(\tau)={\cal H}a^{-2}$. The decaying mode is a gauge mode which can be eliminated using the coordinate transformation $\tau\to\tau+Ca^{-2}$. The particular solution with $\Psi=\dot\Psi=0$ at $\tau=0$ (isocurvature perturbations) and $\sigma\ne0$ (again suppressing the dependencies on $K$ and $S$) is $$\label{isocurv}
\Psi(\tau)=\int_0^\tau\left[\Psi_+(\tau)\Psi_-(\tau')-\Psi_+(\tau')
\Psi_-(\tau)\right]\sigma(\tau')a^2(\tau')\,d\tau'\ .$$
The general solution of equation (\[psieom\]) with $\pi=\Delta
\Psi=0$ is given by adding equations (\[curv\]) and (\[isocurv\]). We see that it is equivalent to integration of $$\label{rwmatch3}
2\Psi=\dot\alpha+\kappa\ ,\ \
\Psi=-{\cal H}\alpha+\kappa\left(1-2\frac{\partial\ln a}
{\partial\ln K}\right)-\delta S\frac{\partial\ln a}{\partial S}$$ when the Friedmann equation governs the background expansion. By comparing the evolution of separate Robertson-Walker universes we have solved equations (\[rwmatch3\]) by the quadratures of equations (\[curv\]) and (\[isocurv\]). The solution requires only that the background evolution obeys the Friedmann and energy conservation equations, and that spatial gradient terms are negligible. Later we will drop the assumption of the Friedmann equation to obtain quadratures for any theory of gravity.
Although we have set $k^2=-\Delta=0$, the solution is valid for all wavelengths much greater than the Jeans length. One simply allows $\kappa$ and $\delta S$ to depend on wavevector $\vec k$ as determined by initial conditions. At short wavelengths (shorter than the Jeans length, for example) $\kappa$ and $\delta S$, as obtained using equations (\[psikappa\]) and (\[sigdef\]), depend on time.
Density and Velocity Perturbations and Einstein Constraints
===========================================================
In addition to the metric perturbations, the density and velocity perturbations of the matter can also be reduced to quadratures in the long-wavelength limit assuming local energy-momentum conservation. The density and velocity perturbations then obey initial-value constraint equations which relate them to the metric perturbations.
Under the coordinate transformations given in the preceding sections, the value of $T^0_{\ \,0}$ does not change to first order in the perturbations. However, comparing the density field of two different Robertson-Walker spacetimes at the same coordinate values $(\tau,x^i)$ gives a density perturbation:[^4] $$\begin{aligned}
\label{denpert}
\delta(\tau)\equiv\frac{\delta\rho}{\bar\rho+\bar p}&=&-3\left[{\cal H}
\alpha+2\kappa\left(\frac{\partial\ln a}{\partial\ln K}\right)_{\tau,S}
+\delta S\left(\frac{\partial\ln a}{\partial S}\right)_{\tau,K}
\right]+\frac{\delta S}{\bar\rho+\bar p}\left(
\frac{\partial\rho}{\partial S}\right)_{\!a}\nonumber\\
&=&3(\Psi-\kappa)+\frac{\delta S}{\bar\rho+\bar p}\left(
\frac{\partial\rho}{\partial S}\right)_{\!a}
=3(\Psi-\kappa)-3\int_0^\tau\sigma(\tau'){\cal H}(\tau')\,
d\tau'+A\ ,\qquad\end{aligned}$$ where $A$ is an integration constant. Equation (\[denpert\]) also follows from energy conservation in a perturbed Robertson-Walker spacetime, which gives the perturbed continuity equation $$\label{pertecons}
\dot\delta+3{\cal H}\sigma=3\dot\Psi+\Delta u$$ where $u$ is the velocity potential. For long wavelength perturbations of wavenumber $k\to0$, $\Delta u=-k^2u$ is generally negligible compared with the other terms in the equation.
Determining the velocity perturbations requires considering spatial variations of the velocity potential, which do not exist in a perfectly homogenous Robertson-Walker spacetime. Thus we consider a perturbed Robertson-Walker spacetime with metric (\[pertrw\]) and energy-momentum tensor (\[emtensor\]). Evaluating $\nabla_\mu
T^\mu_{\ \,i}=0$ gives the first-order result $$\label{pertpcons}
\dot u+(1-3c_w^2){\cal H}u=c_w^2\delta+\sigma+\Phi+(\Delta+3K)\pi\ .$$ If the shear stress can be neglected, we can integrate to obtain $$\label{upert}
u(\tau)=\frac{1}{(\rho+p)a^4}\int_0^\tau\left[c_w^2(\tau')
\delta(\tau')+\sigma(\tau')+\Phi(\tau')\right]
[\rho(\tau')+p(\tau')]a^4(\tau')\,d\tau'+\frac{B}{(\rho+p)a^4}\ ,$$ where $B$ is an integration constant.
Equations (\[denpert\])–(\[upert\]) do not assume the Einstein field equations – they hold for any theory of gravity provided consistent with local energy-momentum conservation. However, the integral forms of these equations are rarely used. Instead, given the potential $\Psi$ on long wavelengths, the usual procedure in general relativity is to evaluate the density and velocity potential using initial-value constraints, of which the Poisson equation is one. That these constraint equations are more general will be shown next.
We define the following combinations of metric and energy-momentum perturbations:
\[constraints\] $$\begin{aligned}
C_1&=&(\Delta+3K)\Psi-\gamma(\delta+3{\cal H}u)\ ,\ \label{C1}\\
C_2&=&\dot\Psi+{\cal H}\Phi-\gamma u\ ,\label{C2}\\
C_3&=&\Psi-\Phi-3\gamma\pi\ .\label{C3}\end{aligned}$$
Here, $\delta+3{\cal H}u$ is the number density perturbation on a hypersurface in which $T^0_{\ \,i}$ vanishes, i.e. number the density perturbation in the local fluid rest frame.
In general relativity each of the constraints vanishes ($C_1=C_2=C_3=0$), however we will not assume this to be automatically true. Differentiating the constraints and using equations (\[Hdot\]), (\[pertecons\]) and (\[pertpcons\]) gives
\[dconstr\] $$\begin{aligned}
&&\frac{1}{a}\frac{\partial}{\partial\tau}(aC_1)-(\Delta+3K
-3\gamma)C_2-{\cal H}(\Delta+3K)C_3=3\gamma({\cal H}^2+K
-\dot{\cal H}-\gamma)u\ ,\label{cons1}\\
&&\frac{\gamma}{a}\frac{\partial}{\partial\tau}\left(
\frac{a}{\gamma}C_2\right)-c_w^2C_1+\frac{\gamma}{{\cal H}}
\frac{\partial}{\partial\tau}\left(\frac{{\cal H}^2}{\gamma}
C_3\right)-KC_3\nonumber\\
&&\qquad=\frac{\gamma}{{\cal H}}\frac{\partial}{\partial\tau}
\left[\frac{{\cal H}^2}{\gamma a^2}\frac{\partial}{\partial\tau}
\left(\frac{a^2}{{\cal H}}\Psi\right)\right]-c_w^2\Delta
\Psi-\gamma\left[\sigma+\frac{3}{{\cal H}}\frac{\partial}
{\partial\tau}\left({\cal H}^2\pi\right)+\Delta\pi\right]
\nonumber\\
&&\qquad+({\cal H}^2+K-\dot{\cal H}-\gamma)(\Phi-2\Psi)
-[2\gamma-3(\eta^2+K)(1+w)]\Psi\ .
\label{cons2}\end{aligned}$$
Here we set $\gamma\equiv4\pi Ga^2(\bar\rho+\bar p)$ and have assumed nothing about gravity. If we assume that the Einstein field equations hold, the right-hand sides of equations (\[dconstr\]) vanish. If we make the weaker assumption that the background is governed by the Friedmann equation and that $\Delta\Psi$ and $\pi$ (shear stress) can be neglected on large scales so that equation (\[psidotfrw\]) holds, the right-hand sides still vanish. In keeping with the previous treatment we also assume $\Psi=\Phi$ so that $C_3=0$. With these assumptions we can integrate equations (\[dconstr\]) to obtain $$\label{constrsol}
C_1=3{\cal H}C_2-A\gamma\ ,\ \
C_2=-\frac{A}{a^2}\int_0^\tau c_w^2(\tau')\gamma(\tau')a^2(\tau')\,
d\tau'-\frac{4\pi GB}{a^2}\ ,$$ where $A$ and $B$ are integration constants. Comparison with equations (\[denpert\]), (\[upert\]), and (\[constraints\]) shows that these are exactly the same initial-value constants obtained from integrating the equations of energy-momentum conservation. In other words, they correspond to changing $\delta$ and $u$ without changing $\Psi$. The third initial-value constant $C$ appearing in equation (\[curv\]) contributes nothing new to the constraints because the pure decaying mode has $\delta=3\Psi=-3{\cal H}u$; the decaying mode can be eliminated by a change of the time coordinate.
In general relativity, the Einstein field equations give $C_1=C_2=C_3=0$ implying $A=B=0$. However, in other theories of gravity the constraints may be nonzero without contradicting equations (\[dconstr\]). The initial-value constants follow from $$\label{AB}
A=\lim_{a\to0}\frac{T\delta S}{a^3(\bar\rho+\bar p)}\ ,\ \
B=\lim_{a\to0}a^4(\bar\rho+\bar p)u\ .$$ While nonzero values $A$ and $B$ are testable, in principle, by comparing the density and velocity fields of galaxies with the gravitational potential implied by gravitational lensing or hydrostatic equilibrium in clusters of galaxies, much more stringent tests obtain at high redshift, where the $A$ and $B$ terms contribute relatively more to the energy-momentum tensor. It would be interesting to place limits on these constants using measurements of the cosmic microwave background anisotropy.
Assuming that $A$ and $B$ are small compared with $\delta$ and $a^4(\bar\rho+\bar p)u$, respectively, at low redshift, and assuming furthermore that the Friedmann equation and energy-momentum conservation are valid, then the density and velocity fields in a perturbed Robertson-Walker spacetime whose metric takes the form of equation (\[rw3\]) must on large scales obey the same constraint equations as in general relativity, namely $C_1=C_2=C_3=0$.
Although the linearized Einstein field equations give three constraints, the first two ($C_1=C_2=0$) are initial-value constraints with no dynamical content. If $C_1=C_2=0$ on an initial timeslice, energy-momentum conservation combined with the other Einstein equations is enough to ensure $C_1=C_2=0$ for all times. The third constraint, $C_3=0$, is a true dynamical constraint because its time derivative is not forced to vanish as a result of the Einstein field equations and energy-momentum conservation.
We will see later that initial-value constraints exist not only in GR but in any gravity theory yielding a long-wavelength perturbed Robertson-Walker solution with local energy-momentum conservation. The key distinguishing features of general relativity are then seen to be the Friedmann equation plus the dynamical constraint $C_3=0$.
Multicomponent Fluids {#sec:multi}
---------------------
Equations (\[isocurv\]), (\[denpert\]) and (\[upert\]) are true quadratures only if $\sigma(\tau)$ is known. While equation (\[sigdef\]) gives $\sigma$ from $p(\rho,S)$ or $\rho(a,S)$, the entropy depends on the internal degrees of freedom of a fluid.
Consider a multicomponent imperfect fluid whose density and pressure are described by a set of parameters $\{X\}$ which can vary with position at fixed expansion factor $a$. We replace the single entropy $S$ by as many parameters as are necessary to characterize spatial variations in the equation of state of the fluid. For example, a system of non-interacting fluids has $$\label{multifluid}
\rho(a,X)=\sum_i\tilde\rho_ir_i(a)\ ,\ \
p(a,X)=\sum_iw_i\tilde\rho_ir_i(a)\ ,\ \
r_i(a)=\exp\left[3\int_a^1(1+w_i)\,
d\ln a\right]\ ,$$ where the $\tilde\rho_i$ are independent of $a$. The $w_i$ generally are not spatially varying, for example $w=0$ for cold dark matter (“dust”) and $w=\frac{1}{3}$ for a relativistic ideal gas (“radiation”). In this case the parameters are the abundances of each fluid component, $\{X\}=\{\tilde\rho_1,\tilde\rho_2,\ldots\}$. The same procedure will work regardless of what parameters characterize the multicomponent fluid but the set $\{X\}$ includes only those that can vary with position at a fixed value of $a$ ($a$ playing the role of volume).
The net equation of state and sound speed parameters are $w\equiv
p(a,X)/\rho(a,X)$ and $c_w^2\equiv(\partial p/\partial a)/(\partial
\rho/\partial a)$. The entropy perturbation follows similarly to equation (\[sigdef\]): $$\label{sigmulti}
\sigma=\sum_X\frac{\delta X}{\rho+p}\left(\frac{\partial p}
{\partial X}-c_w^2\frac{\partial\rho}{\partial X}\right)
=-\frac{1}{3}\frac{\partial\epsilon}{\partial\ln a}\ ,\ \
\epsilon(a,X)\equiv\sum_X\frac{\delta X}{\rho+p}\left(\frac{\partial\rho}
{\partial X}\right)_{\!a}\ .$$ The symbol $\epsilon$, equalling the number density perturbation at fixed expansion factor $a$, is introduced here to avoid confusion with the number density perturbation $\delta$ measured at fixed $\tau$. For the example of equation (\[multifluid\]), $$\label{multisig0}
\sigma=\sum_if_i(w_i-c_w^2)\epsilon_i=\sum_if_iw_i
(\epsilon_i-\epsilon)\ ,$$ where $$\label{multidefs}
f_i(a)\equiv\frac{(\tilde\rho_i+\tilde p_i)r_i(a)}{\rho+p}\ ,\ \
\epsilon_i\equiv\frac{\delta\tilde\rho_i}{\tilde\rho_i+\tilde p_i}\ ,\ \
\epsilon=\sum_if_i\epsilon_i\ .$$ Here $\tilde p_i=w_i\tilde \rho_i$ and $\epsilon_i$ are both independent of time and $f_i$ is the enthalpy fraction for component $i$, normalized so that $\sum_if_i=1$. We see that the entropy perturbation of a superposition of ideal gases arises entirely from differences in the particle number density of the different species measured at fixed expansion $a$. For example, for cold dark matter (subscript $m$) and radiation (subscript $r$) with $y(a)\equiv\rho_m/\rho_r$, $f_m=3y/(3y+4)$, $f_r=4/(3y+4)$, and $\sigma=\frac{1}{3}f_rf_m(\epsilon_r-\epsilon_m)$. For a superposition of $N$ fluids there are $N-1$ independent entropy modes corresponding to the $N-1$ independent differences of the number density perturbations $\epsilon_i$.
In the general case where the individual components have a time-varying $w_i$ and may also have entropy perturbations, $$\label{multisig}
\sigma=\sum_if_i\left[\sigma_i+(c_i^2-c_w^2)\epsilon_i\right]
=\sum_if_i\left[\sigma_i+c_i^2(\epsilon_i-\epsilon)\right]\ ,$$ where $c_i^2\equiv[\partial(w_ir_i)/\partial a]/(\partial r_i/
\partial a)$ and $\sigma_i$ are the sound speed squared and entropy perturbation for each component, respectively. As an example, consider a tightly-coupled plasma of photons and nonrelativistic baryons with $w=[3(1+y_b)]^{-1}$ where $y_b(a)\equiv\rho_b/\rho_r$ (subscript $b$ denotes baryons with $r$ for the photons). In this case, $w=f_r/(4-f_r)$ and $c_w^2=\frac{1}{3}f_r$ compared with $w=f_r/(4-f_r)$ where $f_r=4/(3y_b+4)$. Although this plasma behaves like a single fluid, it can have a nonzero entropy perturbation arising from initial variations in the baryon to photon ratio, $\sigma=\frac{1}{3}f_rf_b(\epsilon_r-\epsilon_b)$ where $\epsilon_r-\epsilon_b=\delta\ln(n_r/n_b)$, the fractional perturbation in the photon to baryon number density ratio, is constant in time. Thus, by examining the composition of each fluid component, we can write the time-dependent total entropy perturbation $\sigma$ in terms of a set of time-independent constants $\epsilon_i$. By doing so, we are able to fully reduce to quadratures the metric perturbation as shown in §\[sec:quadGR\].
Using the same methods we obtain exact results for the density of individual fluid components. Here one must be careful to evaluate the perturbations at fixed $(\tau,\chi)$ after the coordinate transformations $\tau\to\tau+\alpha(\tau)$ and $\chi\to\chi(1-\kappa)$ discussed in §\[sec:cpert\]. The density perturbation of component $i$ at fixed $\tau$ is $$\label{dpert}
\delta_i(\tau)=\frac{\rho_i(a(\tau+\alpha,K+\delta K,X+\delta X),
X+\delta X)-\rho_i(a,X)}{(\rho+p)_i}=3(\Psi-\kappa)+\epsilon_i\ .$$ The $\epsilon_i$ comes from the initial number density perturbation at fixed $a$ while $3(\kappa-\Psi)$ is the fractional change in volume introduced by shifting the initial hypersurface of constant $\tau$. Averaging over all components with weights $f_i$ gives the net density perturbation $$\label{dpertnet}
\delta(\tau)=3(\Psi-\kappa)+\epsilon(\tau)\ ,$$ where $\epsilon(\tau)\equiv\epsilon(a(\tau,K,X),X)$ is given by equation (\[sigmulti\]) and we are holding $K$ and $X$ fixed to lowest order in perturbation theory. By comparing equations (\[denpert\]) and (\[dpertnet\]) and using $\partial\epsilon/\partial\ln a=-3\sigma$ we see that the constant $A$ has been determined.
The velocity potential of individual components follows from equation (\[upert\]) with subscript $i$ applying to all quantities except $a$, $\tau$, and $\Phi$. For $N$ uncoupled fluids there are $N$ constants $B_i$. The net velocity potential is the average over the fluids, $u=\sum_if_iu_u$.
We have succeeded in reducing the long-wavelength density, velocity, and entropy perturbations of all fluid components to quadratures. This presentation works for both general relativity and alternative theories of gravity. In the former case, equations (\[curv\]) and (\[isocurv\]) reduce the problem entirely to quadratures specified by a set of constants $(\kappa,\epsilon_1,\ldots\epsilon_N,B_1,\ldots,B_N)$. We explore next the reduction of the metric perturbation to quadratures for alternative theories of gravity.
Quadratures for Alternative Theories of Gravity {#sec:alt}
===============================================
Much of the treatment given so far assumes the validity of the Friedmann equation or GR. It is straightforward to redo the calculations of equations (\[rw2\])–(\[psikappa\]) and (\[aderivS\])–(\[psidotfrw\]) without making these assumptions. The expansion rate ${\cal H}$ can depend only on the scale factor $a$ and curvature $K$ of the background Robertson-Walker spacetime and the parameters $\{X\}$ describing the equation of state as discussed in §\[sec:multi\]: ${\cal H}= {\cal H}(a,K,X)$. In this case equation (\[friedsol\]) is replaced by some function $\tau(a,K,X)=\int_0^a[a{\cal H}(a,K,X)]^{-1}\,da$. In addition, for an arbitrary theory of gravity the scalar mode has two distinct gravitational potentials, equation (\[pertrw\]). Repeating the derivations of §§\[sec:cpert\] and \[sec:epert\] with the metric of equation (\[pertrw\]) with $\Phi\ne\Psi$ and with an arbitrary background expansion rate ${\cal H}(a,K,X)$ gives the following result for long-wavelength metric perturbations with $\pi=0$: $$\begin{aligned}
\label{pside}
\frac{1}{a^2}\frac{\partial}{\partial\tau}\left(\frac{a^2\Psi}
{{\cal H}}\right)+(\Phi-\Psi)&=&\frac{\kappa}{a}\frac{\partial}{\partial
\tau}\left(\frac{a}{{\cal H}}\right)+{\cal H}\frac{\partial}
{\partial\ln a}\left[2\kappa\left(\frac{\partial\tau}
{\partial\ln K}\right)_{\!a,X}+\sum_X\delta X\left(\frac{\partial\tau}
{\partial X}\right)_{\!a,K}\,\right]\nonumber\\
&=&\frac{\kappa}{a}\frac{\partial}{\partial
\tau}\left(\frac{a}{{\cal H}}\right)-2\kappa\left(\frac{\partial
\ln{\cal H}}{\partial\ln K}\right)_{\!a,X}-\sum_X\delta X\left(
\frac{\partial\ln{\cal H}}{\partial X}\right)_{\!a,K}\ .\end{aligned}$$ The quantities $\kappa$ and $\delta X$ are (for long wavelengths) independent of time.
Equation (\[pside\]) provides one differential equation for two functions $\Phi$ and $\Psi$. Without a second relation, provided by a law of gravity, we can determine neither $\Phi$ nor $\Psi$.
Even without a full theory of gravity, however, we can still obtain interesting and useful results if we assume energy-momentum conservation. In this case we have the quadratures (\[upert\]) and (\[dpertnet\]). By combining these equations with equation (\[pside\]) one can derive the following results which generalize the energy-momentum constraints $C_1$ and $C_2$ of general relativity in the limit of vanishing shear stress ($\pi=0$) and long wavelengths (the spatial Laplacian $\Delta=0$ when applied to all perturbations):
\[genconstraints\] $$\begin{aligned}
\delta+3{\cal H}u&=&\frac{3{\cal H}B'}{a^4(\rho+p)}\ ,\ \label{C1gen}\\
\dot\Psi+{\cal H}\Phi&=&-a\frac{\partial}{\partial\tau}\left(
\frac{{\cal H}}{a}\right)u+\frac{{\cal H}}{a^2(\rho+p)}\frac{\partial}
{\partial a}\left(\frac{{\cal H}B'}{a}\right)\ ,\label{C2gen}\end{aligned}$$
where $$\label{Bdef}
B'(a)\equiv\int_0^a a^2(\rho+p)
\left[-2\kappa\left(\frac{\partial\ln{\cal H}}{\partial\ln K}
\right)_{\!a,X}-\sum_X\delta X\left(\frac{\partial\ln{\cal H}}
{\partial X}\right)_{\!a,K}+\frac{\epsilon}{3a}
\frac{\partial}{\partial\tau}\left(\frac{a}{{\cal H}}\right)
_{\!K,X}\,\right]\,\frac{ada}{{\cal H}}+B\ .$$ The constant $B$ is essentially the same integration constant as in equation (\[upert\]). Without loss of generality we may set $B=0$ in equation (\[Bdef\]).
Equations (\[dpertnet\])–(\[genconstraints\]) are not all independent. Any three of the four imply the remaining one. Thus, we may describe the perturbations by equations (\[dpertnet\]) and (\[genconstraints\]). To reduce the system fully to quadratures requires one more equation equivalent to a relation between $\Phi$ and $\Psi$.
It is worth emphasizing that the reduction to quadratures is based on only a few assumptions:
1. The spacetime is nearly Robertson-Walker with perturbation amplitude small enough for linear perturbation theory to apply.
2. Local energy-momentum conservation holds: $\nabla_\mu
T^{\mu\nu}=0$.
3. In the conformal Newtonian gauge, spatial gradients of fields are small enough to be neglected in all equations of motion.
4. Shear stress perturbations are negligible.
5. A theory of gravity provides some relation between $\Phi$ and $\Psi$ and possibly the matter perturbations. This assumption is needed only for a complete reduction to quadratures.
There are no restrictions on the equation of state of matter and radiation fields. There are no restrictions on the geometry of perturbations; in particular, no assumption of spherical symmetry. The neglect of spatial gradients becomes invalid on scales over which non-gravitational forces act so as to modify local energy-momentum conservation. By definition, spatial gradients of pressure are important and modify our results on scales less than the Jeans length.
The constraints (\[genconstraints\]) are particularly useful when $B'=0$. For any theory having a flat Robertson-Walker solution (including GR), $\lim_{K\to0}(\partial{\ln{\cal H}}/\partial\ln
K)=0$. When there are no entropy perturbations, $\epsilon=0$. Thus, in a flat universe with initially isentropic perturbations, the only possible contribution to $B'$ comes from the $\sum_X\delta
X(\partial\ln{\cal H}/\partial\ln X)$ term. If the only dependence of ${\cal H}$ on equation of state parameters $X$ is through the density $\rho$, then $$\label{Hrho}
\sum_X\delta X\left(\frac{\partial\ln{\cal H}}{\partial X}\right)_{\!a,K}
=\epsilon(\rho+p)\frac{\partial\ln{\cal H}}{\partial\rho}\ ,$$ which vanishes for isentropic initial conditions $\epsilon=0$. Thus, for a broad class of theories, $B'=0$ for curvature perturbations in a flat universe implying the long-wavelength constraints $\delta+3{\cal H}u=0$, $\dot\Psi+{\cal H}\Phi=\gamma u$ where $\gamma=-a\partial({\cal H}/a)/\partial\tau$ for $K=0$. From equation (\[dpertnet\]) we also have $\delta=3(\Psi-\kappa)$ for long wavelengths and isentropic initial conditions. In this case we have three relations between the four time-dependent functions $(\Phi,\Psi,\delta,u)$. In the general case we have an additional function, $B'$, whose specification requires a theory for the homogeneous expansion.
Thus, without knowing anything about the underlying gravity theory except that it is consistent with local energy-momentum conservation and that it admits a Robertson-Walker solution, we have deduced initial-value constraints similar to those of GR under the conditions of long wavelengths and negligible shear stress. One is not free to specify arbitrary metric and matter perturbations on long wavelengths, at least under the assumption that long-wavelength perturbations evolve like separate universes.
However, the initial-value constraints (\[genconstraints\]) combined with equations (\[sigmulti\]) and (\[dpertnet\]) are insufficient to fully determine the growth of perturbations. We need a third constraint, a relation between $\Phi$ and $\Psi$ given by some theory of gravity. Such a dynamical constraint would allow equation (\[pside\]) to be integrated and, with the initial-value constraints, reduce $(\Phi,\Psi,\delta,u)$ fully to quadratures for long-wavelength perturbations with vanishing shear stress.
The simplest case of a dynamical constraint on the gravitational fields is $\Phi=\Psi$ for $\pi=0$ as in GR (although this is more general than GR since we do not assume the Friedmann equation). Under this assumption, equation (\[pside\]) may be integrated completely giving (suppressing the dependencies on $K$ and $X$ and using $a$ as the time variable) $$\label{general}
\Psi(a)=\kappa\Psi_+(a)+\sum_X(\delta X)\Psi_X(a)+C\Psi_-(a)
\ \ \hbox{if}\ \ \Phi=\Psi\ ,$$ where $$\label{curvgen}
\Psi_+(a)=\frac{{\cal H}}{a^2}\int_0^a\left[1-\left(\frac{\partial
\ln{\cal H}}{\partial\ln a}\right)_{\!K,X}-2\left(\frac{\partial
\ln{\cal H}}{\partial\ln K}\right)_{\!a,X}\,\right]\,\frac{ada}{{\cal H}}
\ ,\ \
\Psi_-(a)={\cal H}a^{-2}\ ,$$ and $$\label{entgen}
\Psi_X(a)=-\frac{{\cal H}}{a^2}\int_0^a\left(\frac{\partial\ln{\cal H}}
{\partial X}\right)_{\!a,K}\,\frac{ada}{{\cal H}}.$$ The potential has been reduced entirely to quadratures depending only on the expansion rate ${\cal H}$ as a function of expansion, curvature, and equation of state parameters.
Considering again a flat universe with isentropic initial conditions and assuming equation (\[Hrho\]) holds, but now with the added condition $\Phi=\Psi$, we see that $\Psi=\kappa\Psi_+$ is determined completely by the expansion rate ${\cal H}(a,0,X)$ with constant values of the equation of state parameters $X$. The density and velocity potential perturbation follow from $\delta=3(\Psi-\kappa)=
-3{\cal H}u$. In this case the growth of long-wavelength perturbations is completely determined by the expansion history ${\cal H}(a)$ with fixed curvature ($K=0$) and composition ($\delta
X=0$).
For example, in the DGP brane-world theory of Dvali et al. (2000), the Friedmann equation is modified by replacing $\rho$ by $(\sqrt{\rho+\rho_{r_c}}+\sqrt{\rho_{r_c}})^2$ with $\rho_{r_c}$ being a constant. If $\Phi=\Psi$, equation (\[psikappa\]) remains valid for the long-wavelength curvature mode provided that $\gamma$ changes to $$\label{dgpgamma}
\gamma'\equiv{\cal H}^2+K-\dot{\cal H}=\frac{3}{2}(1+w)
({\cal H}^2+K)\left(\frac{\sqrt{\rho+\rho_{r_c}}}
{\sqrt{\rho+\rho_{r_c}}+\sqrt{\rho_{r_c}}}\right)\ .$$ The isocurvature mode is also modified slightly, with $\gamma\to\gamma'\rho/(\rho+\rho_{r_c})$ in the last line of equation (\[psidotfrw\]). Equations (\[denpert\]) and (\[upert\]) are unchanged.
The solutions implied here are only valid as $k\to0$ so that spatial gradient terms may be neglected in the equations of motion. The DGP model has a length scale $r_c=(32\pi G\rho_{r_c}/3)^{-1/2}$ comparable to $H_0^{-1}$. The long-wavelength limit becomes $kr_c\ll1$ suggesting that the formal solution given here is not applicable for wavelengths shorter than the present Hubble length. Other theories of gravity will give different ranges of applicability. In general, any long-range departures from general relativity will modify the evolution of patches of a Robertson-Walker spacetime.
The results of this section highlight the importance of testing the dynamical constraint $C_3=\Psi-\Phi-3\gamma\pi=0$ in general relativity. The relation between $\Phi$ and $\Psi$, combined with the background expansion rate ${\cal H}(a,K,X)$, is the key to long-wavelength perturbations of a Robertson-Walker cosmology.
Comparing Geometric and Dynamic Methods for Testing Gravity and Dark Energy
===========================================================================
As we have seen, the evolution of long-wavelength perturbations requires two ingredients: (1) a relation between the two gravitational potentials $\Phi$ and $\Psi$ (possibly involving auxiliary fields), and (2) a generalization of the Friedmann equation for the background expansion rate. Without the first ingredient, the long-wavelength perturbations cannot be predicted. Let us assume that a dynamical constraint exists relating $\Phi$ and $\Psi$ so that equation (\[pside\]) can be integrated. Then the evolution of perturbations is determined by the expansion rate of the universe, the very same information that determines the redshift-distance relation used by geometric methods. Does this mean that dynamic and geometric methods measure the same thing?
Not necessarily. First, the dependence of perturbation (dynamic) methods on the expansion rate necessarily depends on the value of $\Phi-\Psi$ which may be nonzero for theories of gravity other than GR. Second, dynamic methods depend not only on ${\cal H}(a,K,X)$ and its variation with $a$ but also on $(\partial\ln{\cal
H}/\partial\ln K)_{a,X}$ (for curvature perturbations) or $(\partial\ln{\cal H}/\partial X)_{a,X}$ (for entropy perturbations; here $X$ are parameters describing the fluid composition or equation of state). Geometric methods, by comparison, depend only on ${\cal
H}(a,K,X)$ with fixed $K$ and $X$. If the universe is nonflat or there exist entropy perturbations, dynamic methods have the potential to reveal information about the dependence of ${\cal H}$ on $K$ and $X$ which is not present in the expansion history for one universe with fixed $K$ and $X$. Finally, the quadrature results for perturbations are valid only in the limit of long wavelengths; long-range forces different from Einstein gravity may lead to spatial gradient terms that modify the quadratures on presently observable scales.
Geometric tests are generally cast in terms of the luminosity distance or angular-diameter distance but in fact depend strictly on the redshift-distance relation for radial null geodesics, $$\label{lightcone}
\chi(z)=\tau_0-\tau(a,K,X)=\int_0^z\frac{dz'}{H(z')}\ ,$$ where $\tau_0=\tau(1,K,X)$ and $a=(1+z)^{-1}$. The redshift-dependence of the long-wavelength curvature perturbation $\Psi_+(z)$ requires specifying a theory of gravity. If $\Phi=\Psi$, for both general relativity and DGP brane worlds the same result holds on very long wavelengths:[^5] $$\label{zint}
\Psi_+(z)=(1+z)H(z)\int_z^\infty\left[\frac{K}{H^2(z')}
+\frac{1}{(1+z')H(z')}\frac{dH}{dz'}\right]\,
\frac{dz'}{H(z')}\ .$$
More generally, let us make no assumptions about $\Phi-\Psi$ but consider a flat universe with negligible shear stress and initially isentropic curvature fluctuations as predicted by the simplest inflationary universe models. In this case, equation (\[pside\]) reduces to the following relation between $H(z)$, $\Psi(z)$, and $\Phi(z)$ for long wavelengths: $$\label{HPsiPhi}
\frac{\Phi(z)}{1+z}=\frac{d\Psi}{dz}+\frac{(\kappa-\Psi)}
{H}\frac{dH}{dz}\ .$$ Here $\kappa$ is independent of redshift on large scales although $\kappa$, $\Psi$, and $\Phi$ will vary with position or spatial wavenumber. In principle, measurements of $H(z)$ from geometric methods and $\Psi(z)$ from perturbations could (for isentropic perturbations of a flat universe with negligible shear stress) determine $\Phi(z)$ up to an additive term proportional to $d\ln
H/d\ln(1+z)$ (since the relation between $\kappa$ and $\Psi$ is unknown if the theory of gravity is unspecified). In particular, such measurements could enable GR or alternative theories of gravity to be tested without any assumptions about the density and pressure of mass-energy in the universe.
This kind of test is difficult to imagine carrying out because of the difficulty of measuring $\Psi(z)$.[^6] Alternatively, if one assumes that GR is valid, dynamic methods can be used to provide additional constraints on dark energy because the geometric and dynamic methods have a different dependence on the equation of state of dark energy.
![Sensitivity of geometric methods (left panel, characterized by the comoving distance $\chi$ to redshift $z$) and dynamic methods (right panel, characterized by the curvature perturbation $\Psi_+$ at redshift $z$) to the equation of state parameter $w$. A flat model with $\Omega_m=0.3$ was assumed. Curves are shown for $w=-0.6,-0.7,\ldots,-1.1$. Dynamic (perturbation) methods are insensitive to dark energy at high redshift but are more sensitive than geometric methods at low redshift.[]{data-label="fig:fig1"}](f1a.eps "fig:") ![Sensitivity of geometric methods (left panel, characterized by the comoving distance $\chi$ to redshift $z$) and dynamic methods (right panel, characterized by the curvature perturbation $\Psi_+$ at redshift $z$) to the equation of state parameter $w$. A flat model with $\Omega_m=0.3$ was assumed. Curves are shown for $w=-0.6,-0.7,\ldots,-1.1$. Dynamic (perturbation) methods are insensitive to dark energy at high redshift but are more sensitive than geometric methods at low redshift.[]{data-label="fig:fig1"}](f1b.eps "fig:")
Geometric measurements of $\chi(z)$ at a given redshift depend on the expansion history at smaller redshifts; dynamic measurements of $\Psi(z)$ depend on $H(z)$ at higher redshifts. Because the effects of dark energy typically decline with increasing redshift, perturbation methods are at a disadvantage at high redshift but may be superior to geometric methods at low redshift.
To assess their relative merits for measuring dark energy (assuming GR is the correct gravity theory) the methods were compared using a flat model with $\Omega_m=0.3$ and a dark energy component with constant $w$ independent of $z$. While unphysical, this model is commonly used to compare theories with data. The logarithmic derivatives of equations (\[lightcone\]) and (\[zint\]) with respect to $w$ at fixed $z$ were evaluated numerically to produce the results shown in Figure \[fig:fig1\]. A large value of $|\partial\ln\chi/\partial w|$ indicates that the geometric method is relatively powerful — a given measurement error in $\ln\chi$ translates into a smaller uncertainty in $w$ for larger values of $|\partial\ln\chi/\partial w|$. The same is true for dynamic methods using $\ln\Psi_+$.
As expected, perturbation methods are insensitive to the dark energy parameter $w$ at high redshift. At low redshift, dark energy dominates more rapidly for smaller $w$ leading to a greater suppression of linear growth hence a larger variation with $w$. Geometric methods, on the other hand, are insensitive to $w$ at small redshift where the lookback time is much less than the Hubble time. For the standard Friedmann equation and $w\approx-1$, dynamic methods are more sensitive than geometric methods for $z<0.2$. At high redshift the comoving distance remains sensitive to $w$ despite the declining importance of dark energy because $\chi(z)$ is an integral over the past lightcone. The expansion history at low redshift affects the redshift-distance relation at high redshift.
The theoretical sensitivities shown in Figure \[fig:fig1\] must be combined with the measurement uncertainties of the two methods before a reliable estimate can be made of their relative merits. The perturbation methods face a severe challenge — to achieve a discriminating power of $0.1$ in $w$ requires 2% accuracy in measurement of $\Psi_+$ at low redshift.
Summary and Outlook for Cosmological Tests of Gravity
=====================================================
Long-wavelength cosmological perturbations involve at least two metric perturbations $(\Phi,\Psi)$ and two matter perturbations $(\delta,u)$ (density and velocity potential, all in conformal Newtonian gauge). One might think that, without a theory of gravity, no predictions can be made about relations between these variables.
In fact, we have shown that, with minimal assumptions, on large scales there are automatically three independent relations between $(\Phi,\Psi,\delta,u)$. For an arbitrary theory of gravity these are any three of the four equations (\[dpertnet\]), (\[pside\]) and (\[genconstraints\]a,b). Equations (\[genconstraints\]) enforce local energy-momentum conservation. They generalize the initial-value constraints of general relativity. GR also has a dynamical constraint on $\Phi-\Psi$ enabling a complete reduction to quadratures. These relations depend on the expansion rate of the background spacetime and on the dependence of the expansion rate on spatial curvature (if $K\ne0$) and on the composition or entropy of the matter filling space (if there are composition or entropy perturbations).
Thus, the key ingredients needed to specify long-wavelength perturbations are (1) a relation between the two gravitational potentials $\Phi$ and $\Psi$, and (2) a relation between the expansion rate, density, and pressure (e.g. the Friedmann equation). Given these two ingredients, we have shown how to reduce $(\Phi,\Psi,\delta,u)$ to quadratures by introducing conserved curvature and entropy variables. Explicit expressions for the time-dependence of the metric perturbations were given for Einstein gravity, equations (\[curv\]) and (\[isocurv\]) or (\[general\])–(\[entgen\]).
In some circumstances (e.g. isentropic fluctuations in a flat universe with $\Phi=\Psi$) the long-wavelength growth of perturbations is determined completely by the expansion history $H(z)$. In such cases measurements of perturbation growth cannot be combined with geometric measurements (e.g. supernova distances) to discern whether dark energy is a new form of mass-energy or a failure of the Friedmann equation. A loophole exists in this argument if $\Phi\ne\Psi$ or there are significant long-range non-gravitational forces. For a flat universe with curvature fluctuations, the argument can be inverted to yield information about $\Phi(z)$ from measurements of $H(z)$ and $\Psi(z)$ thereby providing a test of gravity theories independently of assumptions about the energy density and pressure.
The long-wavelength limit corresponds to wavelengths larger than any relevant spatial scales so that spatial gradients may be neglected in the equations of motion. Pressure forces modify the dynamics on scales smaller than the Jeans length; the dark energy may plausibly be a scalar field with a Jeans length comparable to the Hubble length. By measuring the wavelength-dependence of the linear growth rate on scales greater than 1 Gpc one might measure the time-dependence of the dark energy Jeans length and thereby constrain its intrinsic properties. This measurement is exceedingly difficult because the amplitude of density perturbations in the dark energy is expected to be less than about $10^{-4}$.
In addition to providing tests of general relativity, perturbation measurements can provide constraints on dark energy under the assumption that general relativity is valid. By comparing the dependence of curvature perturbations and the redshift-distance relation on the dark energy equation of state parameter we verified quantitatively the expected conclusion that perturbation methods are most useful at small redshift when the accelerated expansion begins to suppress curvature perturbations. A changing gravitational potential generates cosmic microwave background anisotropy through the integrated Sachs-Wolfe effect. Measurement of this effect (Padmanabhan et al. 2005) has the potential to further constrain dark energy (Pogosian et al. 2005).
The analysis in this paper shows the importance of testing the equality of the two Newtonian gauge gravitational potentials, $\Phi=\Psi$ in equation (\[pertrw\]). While this equality holds in general relativity (in the absence of large shear stress), it may not be true for other theories of gravity. In addition to tests combining geometric and perturbation methods using equation (\[HPsiPhi\]), one can in principle measure $\Phi-\Psi$ by comparing the deflection of light by gravitational lenses (an effect proportional to $\Psi+\Phi$) with the non-relativistic motion of galaxies (an effect proportional to $\Phi$). A similar test exists on solar-system scales (or in binary pulsars) where the deflection (or Shapiro delay) of light is compared with Newtonian dynamics. Thus, for testing modified gravity as an alternative to GR it is important to extend tests of the post-Newtonian parameter $\gamma_{PPN}\equiv\Psi/\Phi$ (Will 2006). Stringent limits on $|\gamma_{PPN}-1|$ apply on solar system scales. It would be worthwhile to improve the limits on Mpc and larger scales by combining weak gravitational lensing and galaxy peculiar velocity measurements or by adding $\gamma_{PPN}$ to the parameters used in analyzing cosmic microwave background anisotropy.
I thank Eric Linder for helpful comments. This work was supported by the Kavli Foundation and by NSF grant AST-0407050.
[99]{} Bardeen, J. M. 1980, , 22, 1882 Bardeen, J. M., Steinhardt, P. J., & Turner, M. S. 1983, , 28, 679 Bashinsky, S. & Seljak, U. 2004, , 69, 083002 Bertschinger, E. 1996, in Cosmology and Large Scale Structure, proc. Les Houches Summer School, Session LX, ed. R. Schaeffer, J. Silk, M. Spiro, and J. Zinn-Justin (Amsterdam: Elsevier Science), 273 Dvali, G., Gabadadze,G., & Porrati, M. 2000, Phys. Lett. B, 485, 208 Eisenstein, D.J. 2005, , 633, 560 Gliner, E. B. 1966, Sov. Phys. JETP, 22, 378 Gordon, C. 2005, Nucl. Phys. Proc. Suppl. 148, 51 Hwang, J. & Noh, H. 2002, Class. Quant. Grav. 19, 526 Ishak, M., Upadhye, A., & Spergel, D. N. 2005, astro-ph/0507184 Kolb, E. W., Matarrese, S., Notari, A., & Riotto, A. 2005a, , 71, 023524 Kolb, E. W., Matarrese, S., & Riotto, A. 2005b, astro-ph/0506534 Kodama, H. & Sasaki, M. 1984, Prog. Theor. Phys. Suppl., 78, 1 Liftshitz, E. M. 1946, J. Phys. (USSR), 19, 116 Linder, E. V. 2005, , 72, 043529 Lue, A., Scoccimarro, R., & Starkman, G. D. 2004, , 69, 124015 Ma, C.-P. & Bertschinger, E. 1995, , 455, 7 Mukhanov, V. F., Feldman, H. A., & Brandenberger, R. H. 1992, Phys. Rep., 215, 1 Padmanabhan, N., Hirata, C. M., Seljak, U., Schlegel, D. J., Brinkmann, J., & Schneider, D. P. 2005, , 72, 043525 Perlmutter, S. et al. 1999, , 517, 565 Pogosian, L., Corasaniti, S., Stephan-Otto, C., Crittenden, R., & Nichol, R. 2005, , 72, 103519 Ratra, B. & Peebles, P. J. E. 1988, , 37, 3406 Riess, A. et al. 1998, , 116, 1009 Steinhardt, P. J., Wang, L., & Zlatev, I. 1999, , 59, 123504 Wands, D., Malik, K. A., Lyth, D. H., & Liddle, A. R. 2000, , 62, 043527 Will, C. M. 2006, Living Rev. Relativity 9, 3, URL http://relativity.livingreviews.org/Articles/lrr-2006-3/ Zel’dovich, Ya. B. 1967, JETP Lett., 6, 316
[^1]: It would be straightforward to add such parameters to the argument list of $a(\tau,K,S)$ and then perturb them.
[^2]: In §\[sec:alt\] we will generalize equation (\[rw3\]) to the case of two distinct potentials for the time- and space- parts of the metric.
[^3]: For spatially uniform, isentropic perturbations with vanishing shear stress, $\kappa$ reduces to the $\zeta$ variable of Bardeen et al. (1983). In other cases the two variables differ, with $\kappa$ being simpler in both its dynamics and interpretation.
[^4]: Note the unusual definition of $\delta$; for a gas of particles it gives the relative number density perturbation rather than the relative energy density perturbation. The equations simplify with this choice.
[^5]: This does not mean that the two theories predict identical perturbations on scales larger than the Jeans length. The DGP model has a much longer length scale, $r_c$, below which equation (\[zint\]) may be invalid.
[^6]: Observations of density perturbations, e.g. from galaxy redshift surveys, are easier than observations of the gravitational potential perturbations. The relation between density perturbations and $\Psi$ depends on the theory of gravity and may require specifying more than just $H(z)$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper we present a sequential hypothesis test for the detection of general jump size distrubution. Infinitesimal generators for the corresponding log-likelihood ratios are presented and analyzed. Bounds for infinitesimal generators in terms of super-solutions and sub-solutions are computed. This is shown to be implementable in relation to various classification problems for a crude oil price data set. Machine and deep learning algorithms are implemented to extract a specific deterministic component from the crude oil data set, and the deterministic component is implemented to improve the Barndorff-Nielsen and Shephard model, a commonly used stochastic model for derivative and commodity market analysis.'
author:
- |
Michael Roberts[^1], Indranil SenGupta[^2]\
Department of Mathematics\
North Dakota State University\
Fargo, North Dakota, USA.
title: '<span style="font-variant:small-caps;">Sequential hypothesis testing in machine learning driven crude oil jump detection</span> '
---
<span style="font-variant:small-caps;">Key Words:</span> Lévy processes, Hypothesis test, Machine learning, Crude oil price, Barndorff-Nielsen and Shephard model.
Introduction {#ch:newintro}
============
Various existing hedging algorithms and insurance risks depend on the underlying statistical model of the commodity market. Consequently, an improvement in the underlying model directly improves the hedging strategies and the understanding of insurance risks. In this paper, we develop a novel statistical methodology for the the refinement of stochastic models using various machine and deep learning algorithms.
As availability of information to the public through alternative data sources increases, machine learning is necessary for adequate analysis. Currently, 97% of North American businesses are using machine learning capabilities to analyze and apply data sources to their trading platforms and analytic focused activities (see [@refinitiv]). The advent of these technologies allows participants to train, test, and project models using data that have historically been inaccessible. “Any innovation that makes better use of data, and enables data scientists to combine disparate sources of data in a meaningful fashion, offers the potential to gain competitive advantage" (see [@refinitiv]). Trading capabilities, scale, scope, and speeds have increased exponentially with advancements and applications of Artificial Intelligence and Algorithmic trading.
A commonly used stochastic model for derivative and commodity market analysis is the Barndorff-Nielsen and Shephard (BN-S) model (see [@BN1; @BN-S1; @BN-S2; @BJS; @Semere; @Issaka]). In [@SWW], the BN-S model is implemented to find an optimal hedging strategy for the oil commodity from the Bakken, a new region of oil extraction that is benefiting from fracking technology. In [@Wil2], the BN-S model is used such, in the presence of quantity risk for oil produced in that region. In the recent paper [@recent], a machine learning-based improvement of the BN-S model is proposed. It is shown that this refined BN-S model is more efficient and has fewer parameters than other models which are used in practice as improvements of the BN-S model. Machine learning based techniques are implemented for extracting a *deterministic component* ($\theta$) out of processes that are usually considered to be completely stochastic. Equipped with the aforementioned $\theta$, the obtained refined BN-S stochastic model can be implemented to incorporate *long range dependence* without actually changing the model.
It is clear that the real challenge is to obtain an estimation of the the value of the *deterministic component* for an empirical data set. In [@recent], a naive way to find this value for crude oil price is proposed. The method proposed in that paper provides an algorithm to form a classification problem for the data set. After that, various machine and deep learning techniques are implemented for that classification problem.
In this paper, we investigate the problem from the perspective of sequential hypothesis testing. As described in [@Weld], a sequential test of a hypothesis means any statistical test that gives a specific rule, at any stage of the experiment, for making one of the three decisions: (1) to accept the null hypothesis $H_0$, (2) to reject $H_0$, (3) to continue the experiment by making additional observation. Sequential hypothesis testing has many applications (see [@Baum; @Brodsky; @Dayanik; @Golubev]). In the paper [@O], the problem of testing four hypotheses on two streams of observations is examined. A minimization result is obtained for the sampling time subject to error probabilities for distinguishing sequentially a standard versus a drifted two-dimensional Brownian motions. This result is further generalized in [@Roberts], where the testing of four hypotheses on two streams of observations that are driven by Lévy processes is presented. Consequently, the results in [@Roberts] are applicable for sequential decision making on the state of two-sensor systems. In one case, each sensor receives or does not receive a signal obstructed by noise. For the other case, each sensor receives data driven by Lévy processes with large or small jumps. In this paper we show that a sequential test of a hypothesis can be implemented in relation to various classification problems for an empirical data set. Subsequently, machine and deep learning algorithms can be implemented to extract a *deterministic component* from a financial data set.
The organization of the paper is as follows. In Section \[sec12\], a refined Barndorff-Nielsen and Shephard model with some of its properties is presented. In Section \[ch:generalization\], we provide a general jump-size detection analysis based on the sequential testing of hypotheses. In Section \[dataanalysissetup\], an overview of the data set is provided, and then three procedures in the predictive classification problem are introduced. Numerical results are shown in Section \[numericalresults\], and finally, a brief conclusion is provided in Section \[conclusion\].
A refined Barndorff-Nielsen and Shephard model {#sec12}
==============================================
Many models in recent literature try to capture the stochastic behavior of time series. As an example, for the Barndorff-Nielsen and Shephard model (BN-S model), the stock or commodity the price $S= (S_t)_{t \geq 0}$ on some filtered probability space $(\Omega, \mathcal{G}, (\mathcal{G}_t)_{0 \leq t \leq T}, \mathbb{P})$ is modeled by
$$\label{1}
S_t= S_0 \exp (X_t),$$
$$\label{2}
dX_t = (\mu + \beta \sigma_t ^2 )\,dt + \sigma_t\, dW_t + \rho \,dZ_{\lambda t},$$
$$\label{3}
d\sigma_t ^2 = -\lambda \sigma_t^2 \,dt + dZ_{\lambda t}, \quad \sigma_0^2 >0,$$
where the parameters $\mu, \beta, \rho, \lambda \in \mathbb{R}$ with $\lambda >0$ and $\rho \leq 0$ and $r$ is the riskfree interest rate where a stock or commodity is traded up to a fixed horizon date $T$. In the above model $W_t$ is a Brownian motion and the process $Z_{\lambda t}$ is a subordinator. Also $W$ and $Z$ are assumed to be independent, and $(\mathcal{G}_t)$ is assumed to be the usual augmentation of the filtration generated by the pair $(W, Z)$.
However, the results and theoretical framework are far from being satisfactory. The BN-S model does not incorporate the *long-range dependence* property. As such, the model fails significantly for a longer ranges of time. To incorporate *long-range dependence*, a class of superpositions of Ornstein-Uhlenbeck (OU)-type processes is constructed in literature in terms of integrals with respect to independently scattered random measures (see [@BN1; @Sem2]). With appropriate conditions, the resulting processes are incorporated with *long-range dependence*. A limiting procedure results in processes that are second-order self-similar with stationary increments. Other resulting limiting processes are stable and self-similar with stationary increments. However, it is statistically unappealing to fit a large number of OU processes, at least by any formal likelihood-based method. To address this issue, in [@recent] a new method is developed.
As proposed in [@recent], $S= (S_t)_{t \geq 0}$ on some filtered probability space $(\Omega, \mathcal{F}, (\mathcal{F}_t)_{0 \leq t \leq T}, \mathbb{P})$, is given by , where the dynamics of $X_t$ in is given by $$\label{2new}
dX_t = (\mu + \beta \sigma_t ^2 )\,dt + \sigma_t\, dW_t + \rho\left( (1-\theta) \,dZ_{\lambda t}+ \theta dZ^{(b)}_{\lambda t}\right),$$ where $Z$ and $Z^{(b)}$ are two independent subordinators, and $\theta \in [0,1]$ is a deterministic parameter. Machine learning algorithms can be implemented to determine the value of $\theta$. The process $Z^{(b)}$ in is a subordinator that has greater intensity than the subordinator $Z$. Also, $W$, $Z$ and $Z^{(b)}$ are assumed to be independent, and $(\mathcal{F}_t)$ is assumed to be the usual augmentation of the filtration generated by $(W, Z, Z^{(b)} )$.
In this case will be given by $$\label{4new}
d\sigma_t ^2 = -\lambda \sigma_t^2 \,dt + (1- \theta') dZ_{\lambda t} + \theta' dZ^{(b)}_{\lambda t} , \quad \sigma_0^2 >0,$$ where, as before, $\theta' \in [0,1]$ is deterministic.
We conclude this section with some properties of this new model. Note that $(1-\mu) \,dZ_{\lambda t}+ \mu dZ^{(b)}_{\lambda t}$, where $\mu \in [0,1]$, is also a Lévy subordinator that is positively correlated with both $Z$ and $Z^{(b)}$. Note that the solution of can be explicitly written as $$\label{5}
\sigma_t^2= e^{-\lambda t}\sigma_0^2 + \int_0^t e^{-\lambda (t-s)}\, \left( (1-\theta') dZ_{\lambda t} + \theta' dZ^{(b)}_{\lambda t}\right).$$ The *integrated variance* over the time period $[t, T]$ is given by $\sigma_{I}^{2}= \int_t^T \sigma_s^2\, ds$, and a straight-forward calculation shows $$\label{6}
\sigma_{I}^{2}= \epsilon(t,T) \sigma_t^2 + \int_t^T \epsilon(s,T)\, \left((1-\theta') dZ_{\lambda t} + \theta' dZ^{(b)}_{\lambda t}\right),$$ where $$\label{95}
\epsilon(s,T)= \left(1- \exp(-\lambda(T-s))\right)/\lambda, \quad t \leq s \leq T.$$ We derive a general expression for the characteristic function of the conditional distribution of the log-asset price process appearing in the BN-S model given by equations , and . For simplicity, we assume $$\theta=\theta'.$$
As shown in [@recent], the advantages of the dynamics given by , , and over the existing models are significant. The following theorem is proved in [@recent]. From this result, it is clear that as $\theta$ is constantly adjusted, for a fixed $s$, the value of $t$ always has an upper limit. Consequently, $\text{Corr}(X_t, X_s)$ never becomes very small and thus *long-range dependence* is incorporated in the model.
\[big12222\] If the jump measures associated with the subordinators $Z$ and $Z^{(b)}$ are $J_Z$ and $J^{(b)}_Z$ respectively, and $J(s)= \int_0^s \int_{\mathbb{R}^+} J_Z(\lambda d\tau, dy)$, $J^{(b)}(s)= \int_0^s \int_{\mathbb{R}^+} J^{(b)}_Z(\lambda d\tau, dy)$; then for the log-return of the improved BN-S model given by , , and , $$\begin{aligned}
\label{corrBNSimproved}
\text{Corr}(X_t, X_s)= \frac{\int_0^s \sigma_{\tau}^2 d\tau + \rho^2 (1-\theta)^2 J(s) + \rho^2 \theta^2 J^{(b)}(s)}{ \sqrt{\alpha(t) \alpha(s)}},\end{aligned}$$ for $t>s$, where $\alpha(\nu) = \int_0^{\nu} \sigma_{\tau}^2 d\tau + \nu\rho^2 \lambda ((1-\theta)^2 \text{Var}(Z_1)+ \theta^2 \text{Var}(Z^{(b)}_1)) $.
We denote $Z^{(e)}= (1- \theta) Z + \theta Z^{(b)}$. Note that $Z^{(e)}$ is also a subordinator. We call this the *effective* subordinator. We denote the cumulant transforms as $\kappa^{(e)}(\theta)= \log E^{\mathbb{P}}[e^{\theta Z_1^{(e)}}] $. In this work, we make the following assumption similar to [@NV; @ijtaf].
\[a2\] Assume that $\hat{\theta}^{(e)}= \sup\{ \theta \in \mathbb{R} : \kappa^{(e)}(\theta) < + \infty \}>0$.
We state the following well-known result from [@NV; @ijtaf] and denote the real part and imaginary part of $z \in \mathbb{C}$ as $\Re(z)$ and $\Im(z)$, respectively.
Let $Z$ be a subordinator with cumulant transform $\kappa$, and let $f: \mathbb{R}_{+} \to \mathbb{C}$ be a complex-valued, left continuous function such that $\Re(f) \leq 0$. Then $$\label{8}
E\left[ \exp\left( \int_0^t f(s)\, dZ_{\lambda s}\right) \right]= \exp \left( \lambda \int_0^t \kappa (f(s))\, ds \right).$$ The above formula still holds if $Z= Z^{(e)}$ satisfies Assumption \[a2\] and $f$ is such that $\Re(f) \leq \frac{\hat{\theta}^{(e)}}{(1+\epsilon)}$, for $\epsilon >0$. \[le1\]
The *Laplace transform* of $X_{T|t}$, the conditional distribution of $X_T$ given the information up to time $t \leq T$, is given by $\phi(z)= E^{\mathbb{P}}[\exp(zX_T)| \mathcal{F}_t]$, for $z \in \mathbb{C}$ such that the expectation is well-defined.
\[th1\] In the case of the general BN-S model described in equations , and , the Laplace transform $\phi(z)= E[\exp(zX_T)| \mathcal{F}_t]$ of $X_{T|t}$ is given by $$\label{91}
\phi(z)= \exp \left(z(X_t+ \mu(T-t))+ \frac{1}{2}(z^2+ 2\beta z) \epsilon(t, T) \sigma_t^2+ \lambda \int_t^T G(s,z)\, ds \right),$$ where $G(s,z)= \kappa^{(e)}\left(\rho z + \frac{1}{2}(z^2+ 2 \beta z) \epsilon(s, T) \right)$.
The transform $\phi(z)$ is well defined in the open strip $\mathcal{S}= \{ z\in \mathbb{C} : \Re (z) \in (\theta_{-}, \theta_{+})\}$, where $$\theta_{-} = \sup_{t \leq s \leq T} \{-\beta - \frac{\rho}{\epsilon(s, T)}- \sqrt{\Delta_1}\},$$ and $$\theta_{+} = \inf_{t \leq s \leq T} \{-\beta - \frac{\rho}{\epsilon(s, T)} + \sqrt{\Delta_1}\},$$ where $\Delta_1= (\beta + \frac{\rho}{\epsilon(s, T)})^2 + 2 \frac{\hat{\theta}^{(e)}}{\epsilon(s,T)}$.
We obtain from $$X_T= \zeta + \beta \sigma_{I}^{2}+ \int_t^T \sigma_s\, dW_s + \rho \int_t^T dZ^{(e)}_{\lambda s},$$ where $\zeta= X_t + \mu(T-t)$. Let $\mathcal{G}$ denote the $\sigma$-algebra generated by $Z^{(e)}$ up to time $T$ and by $\mathcal{F}_t$. Then, proceeding by iterated conditional expectations, we obtain $$\begin{aligned}
\phi(z) & = E^{\mathbb{P}}[\exp(zX_T)| \mathcal{F}_t] \\
& = E^{\mathbb{P}}\left[ E^{\mathbb{P}}\left[\exp(z(\zeta + \beta \sigma_{I}^{2}+ \int_t^T \sigma_s\, dW_s + \rho \int_t^T dZ^{(e)})) | \mathcal{G}\right] | \mathcal{F}_t \right] \\
& = E^{\mathbb{P}} \left[ \exp (z(\zeta + \beta \sigma_{I}^{2} + \rho \int_t^Td Z^{(e)}_{\lambda s})) E^{\mathbb{P}}\left[\exp(z \int_t^T \sigma_s\, dW_s)| \mathcal{G} \right] | \mathcal{F}_t\right] \\
& = E^{\mathbb{P}}\left[ \exp\left(z(\zeta + \beta \sigma_{I}^{2} + \rho \int_t^T dZ^{(e)}_{\lambda s}) + \frac{1}{2}\sigma_{I}^{2} z^2 \right) | \mathcal{F}_t \right].\end{aligned}$$ Using we obtain $$\phi(z)= \exp\left( \zeta z + \frac{1}{2}\epsilon(t, T) \sigma_t^2(z^2+ 2\beta z) \right) E^{\mathbb{P}} \left[\exp\left( \int_t^T \left(\rho z + \frac{1}{2}(z^2+ 2 \beta z) \epsilon(s, T) \right)\, dZ^{(e)}_{\lambda s} \right) \right].$$
Clearly if $z \in \mathcal{S}$, then $\Re (\rho z + \frac{1}{2}(z^2+ 2 \beta z) < \hat{\theta}$. Thus the result follows from .
Jump-size detection based on sequential hypothesis tests {#ch:generalization}
========================================================
In Section \[sec12\], it is observed that the refined BN-S model can be successfully implementable only when $\theta$ can be successfully computed for and (with $\theta= \theta'$). To find $\theta$, in [@recent], a machine learning based empirical analysis is implemented. In this section, we provide a general jump-size detection analysis based on the sequential test of a hypothesis. We consider a Lévy process $Z$ defined by Lévy triplet $(\mu, \sigma^2, \nu^*)$, where $\mu$ is the drift, $\sigma$ is the diffusion, and $\nu^*(dx)=(1+\alpha x)\nu(dx)$ for some Lévy measure $\nu$ defined on $\mathbb{R}^+$.
We are interested in detecting a *significant jump* in the process. Consequently, we wish to test the hypotheses $$\begin{aligned}
\label{jumphyp}
H_{0}: \alpha=0, \quad \quad H_{1}: \alpha=a> 0, \end{aligned}$$ which clearly address the size of the jumps in the Lévy process.
The Lévy process generates a filtration, which will be denoted $\mathcal{F}_t$, $i=0,1$. Further, the hypotheses induce probability measures $P_{i}$, $i=0,1$. We seek to create a decision rule $(\tau, \delta_\tau)$, where $\tau$ is a stopping rule with respect to $\mathcal{F}_t$, and $\delta_\tau$ is a random variable taking values in the index set $\{0, 1\}$.
Let the log-likelihood ratio of the marginal density be given by $u^{(i)}_t$, $i=0,1$. Then, $$\label{logratio}
u^{(i)}_t = \log \frac{dP_{i}}{dP_{1-i}}.$$ and consider an interval $[l,r] \subset \mathbb{R}$. We define the decision rules to be $$\begin{aligned}
\label{onedec}
\nonumber \tau&=\inf \{ t\geq 0: u^{(i)}_t \notin [l,r] \},\\
\nonumber \delta_{\tau}^{(i)} & = 1-i, \hbox{ if } u^{(i)}_{\tau}\leq l, \\
\delta_{\tau}^{(i)} & = i, \hbox{ if } u^{(i)}_{\tau}\geq r.\end{aligned}$$
It is known that if $(X_t)_{t \geq 0}$ is a Lévy process then there exists a unique cádlág process $(Z_t)_{t \geq 0}$ such that $$dZ_t= Z_{t-}\, dX_t, \quad Z_0=1.$$ $Z$ is called the stochastic exponential or Doléans-Dade exponential of $X$ and is denoted by $Z= \mathcal{E}(X)$. We now derive the infinitesimal generators. The results are motivated by [@Roberts].
\[jumpgen\] With the process $u^{(i)}_t$ defined as in , we have infinitesimal generators, given by $$\mathcal{L}_{i}\xi(x):= (-1)^{i+1}\gamma \xi'(x) +\frac{1}{2} \beta^2 \xi''(x) +(-1)^{i}\int_{\mathbb{R}_+} \left(\xi(x+y)-\xi(x)-\frac{y \xi'(x)}{1+|y|}\right)K(dy),$$ for any suitable $\xi$, where $$\begin{aligned}
\beta&= -a\int_{x>0} (1\wedge x) \sigma^{-1} x\nu(dx)\label{beta},\\
m&=a\int_{x>1} x\nu(dx)\label{m},\\
\gamma&=m-\frac{\beta^2}{2}+\int_0^1 (\log(1+x)^2-x)a\nu(dx),\label{gamma}\\
K&=a\log(1+x)\nu\label{K}. \end{aligned}$$
Since $z$ is a Lévy process with characteristics $(\mu, \sigma^2, \nu)$ under $P_0$ and characteristics $(\mu, \sigma^2, (1+ax)\nu)$ under $P_1$, we apply the generalized Girasanov’s Theorem. Using $\beta$ as in , we obtain $$\frac{dP_i}{dP_{1-i}}=\mathcal{E}\left((-1)^{i+1}N.\right)_t,$$ where $$N_t=\beta W_t +\int_0^t\int_{x>0} ax (J-\nu)(ds,dx),$$ $aJ$ is the jump measure for $N$, $W$ is a standard Brownian motion, and $\mathcal{E}$ is the Doléans-Dade exponential. This gives that $N_t$ is a Lévy process with characteristics $$( (-1)^{i}m, \beta^2,(-1)^{i+1} a\nu).$$ Then, by [@Tank] (Proposition 8), we obtain characteristics $$((-1)^{i}\gamma,\beta^2,(-1)^{i+1}K),$$ for $u^{(i)}_t$. Finally, by [@Tank; @AS], the process has the stated generator.
Assign $\xi_{i}$ to be the probability of a correct decision in world $i$. Then we have the partial integro-differential equations $\mathcal{L}_{i}\xi_{i}=0$ with boundary conditions $$\begin{aligned}
\xi_{0}(l)=1, \quad \quad \quad & \xi_{1}(l)=0, \nonumber \\
\xi_{0}(r)=0, \quad \quad \quad & \xi_{1}(r)=1. \label{bcs}\end{aligned}$$ Further, we have $ \xi_{i}>0$ inside $R=(l,r)$.
Before proving the existence of a solution to the stated boundary value problem, we need a few more definitions and a theorem from [@EU] that will be used:
An upper semicontinuous function $l: {\mathbb{R}}\to {\mathbb{R}}$ is a *subsolution* of $$F(0,\xi,D\xi,D\xi^2,\mathcal{I}[\xi](x))=0$$ subject to boundary conditions if for any test function $\phi\in C^2({\mathbb{R}})$, at each maximum point $x_0 \in \bar{R}$ of $l-\phi$ in $B_\delta(x_0)$, we have $$E(l,\phi,x_0):=F(x_0,l(x_0),D\phi(x_0),D^2\phi(x_0), I_\delta^1[\phi](x_0)+I_\delta^2[l](x_0))\leq 0 \hbox{ if $x_0\in R$}$$ or $$\min(E(l,\phi,x_0);u(x_0)-g(x_0))\leq 0 \hbox{ if $x_0\in \partial R$},$$ where $$\begin{aligned}
I_\delta^1[\phi](x_0)=\int_{|z|<\delta} \left( \phi(x_0+z)-\phi(x_0)-(D\phi(x_0) \cdot z)\textbf{1} _B (z)\right) d\mu_{x_0}(z), \\
I_\delta^2[u](x_0)=\int_{|z|\geq\delta} \left( u(x_0+z)-u(x_0)-(D\phi(x_0) \cdot z)\textbf{1}_B (z)\right) d\mu_{x_0}(z) .\end{aligned}$$
Similarly, a lower semicontinuous function $u: {\mathbb{R}}\to {\mathbb{R}}$ is a *supersolution* of the same boundary value problem if for any test function $\phi\in C^2({\mathbb{R}})$, at each minimum point $x_0 \in \bar{R}$ of $u-\phi$ in $B_\delta(x_0)$, we have $$E(u,\phi,x_0)\geq 0\hbox{ if $x_0\in R$}$$ or $$\max(E(l,\phi,x_0);u(x_0)-g(x_0))\leq 0 \hbox{ if $x_0\in \partial R$}.$$ Finally, a *viscosity solution* is a function whose upper and lower semicontinuous envelopes are respectively a *sub-solution* and a *super-solution*.
\[EUtheorem\] If $F:{\mathbb{R}}^5\to {\mathbb{R}}$, and
1. $F(x,u,p,X,i_1)\leq F(x,u,p,Y,i_2)$ if $X\geq Y$ and $i_1\geq i_2$,
2. there exists $\gamma>0$ such that for any $x, u, v, p, X, i \in {\mathbb{R}},$ $$F(x,u,p,X,i)-F(x,v,p,X,i)\geq \gamma (u-v) \hbox{ if $u \geq v$},$$ for some $\epsilon>0$ and $r(\beta)\to 0$ as $\beta \to 0$, we have $$F(y,v,\epsilon^{-1}(x-y),Y,i)-F(x,v,\epsilon^{-1}(x-y),X,i)\leq \omega_R (\epsilon^{-1}|x-y|^2+|x-y|+r(\beta)),$$
3. $F$ is uniformly continuous with respect to all arguments,\
4. $\sup_{x\in {\mathbb{R}}} |F(x,0,0,0,0)|<\infty$,\
5. $K$ is a Lévy-Itô measure,
6. the inequalities in are strict,
7. for any $R>0$, there exists a modulus of continuity $\omega_R$ such that, for any $x,y\in{\mathbb{R}}$, $|v|\leq R$, $i\in {\mathbb{R}}$, and for any $X,Y\in {\mathbb{R}}$ satisfying $$\left[\begin{array}{cc} X & 0 \\ 0 & Y\end{array}\right] \leq \frac{1}{\epsilon}\left[\begin{array}{cc} 1 & -1 \\ -1 & 1 \end{array}\right]+r(\beta)\left[\begin{array}{cc} 1 & 0 \\ 0 & 1 \end{array}\right],$$
then there is a unique solution to $F(0,\xi,D \xi, D^2 \xi,\mathcal{I}[\xi](x))=0$ between any pair of super-solution and sub-solutions, defined below, where $$\mathcal{I}[\xi](x):=\int_{\mathbb{R}_+} \left(\xi(x+y)-\frac{y \xi'(x)}{1+|y|}\right)K(dy).$$
In particular, the function $$F(x,u,p,X,i):=Mu+ \gamma p -\frac{\beta}{2}X - i$$ satisfies (A1)-(A4) and our measure $K$ satisfies (A5) in , where $$M=\int_{\mathbb{R}_+}K(dy).$$
First, consider (A1): $$F(x,u,p,X,i_1)- F(x,u,p,Y,i_2) = \frac{\beta}{2}(Y-X) +i_2- i_1 \geq0$$ if $i_2\leq i_1$ and $Y\leq X$.
Next, $F(x,u,p,X,i)- F(x,v,p,X,i)=M(u-v)$, so choosing $\gamma=M >0$, we have property (A2).
Property (A3) is satisfied because $F$ is linear in each argument, and (A4) is satisfied because $F$ does not depend on its first argument explicitly. Last, $K$ is a Lévy-Itô measure by the assumptions of the underlying Lévy process.
Note that the $F$ above corresponds to case $i=0$. The other case can be similarly satisfied through manipulation of the signs in $F$. Before proceeding, we present another formal definition:
We write that a function $f(x)=O(g(x))$ if we have some $M,\epsilon \in {\mathbb{R}}$ satisfying $|f(x)|\leq Mg(x)$ for all $x>\epsilon$. Similarly, we write that a function $f(x)=o(g(x))$ if we have some $M,\epsilon \in {\mathbb{R}}$ satisfying $|f(x)|< Mg(x)$ for all $x>\epsilon$.
The norm $\| f \|_\infty$ is defined as the essential supremum of the absolute value of $f$ over $\Omega$. It is the smallest number so that $\{ x : |f(x)|\geq \|f\|_\infty\}$ has measure zero.
We state the additional limit assumptions on $F$ from [@EU]: $$\begin{aligned}
\liminf_{y\to x, y\in \bar{\Omega}, \eta \downarrow 0, d(y)\eta^{-1} \to 0} \left[ \sup_{0<\delta \in [d(y),r)} \inf_{s\in [-R,R]} F(y,s,p_\eta(y),M_\eta(y),I_{\eta,\delta,r}(y))\right]&<0, \nonumber \\
\limsup_{y\to x, y\in \bar{\Omega}, \eta \downarrow 0, d(y)\eta^{-1} \to 0} \left[ \inf_{0<\delta \in [d(y),r)} \sup_{s\in [-R,R]} F(y,s,-p_\eta(y),-M_\eta(y),-I_{\eta,\delta,r}(y))\right]&<0, \label{limsupinf}\end{aligned}$$ where $$p_\eta(y) =O(\epsilon^{-1})+\frac{k_1+o(1)}{\eta} Dd(y),$$ $$M_\eta(y)= O(\epsilon^{-1})+\frac{k_1+o(1)}{\eta} D^2d(y)-\frac{k_2+o(1)}{\eta^2} Dd(y) \otimes Dd(y),$$ $$\begin{aligned}
I_{\eta,\delta,r}(y)&=-\nu I_{\delta,r}^{\hbox{ext},1}(y)+2\|u\|_\infty I_{\beta(\nu),r}^{\hbox{int},1}(y)\\
&-\frac{k_1+o(1)}{\eta} \left(I^{\hbox{tr}}(y)+I_{\beta(\eta),r}^{\hbox{int},2}(y)+I_{\delta,r}^{\hbox{ext},2}(y)-\|D^2d\|_\infty I^4_{\delta,\beta(\eta),r}(y)\right)\\
&+O(\epsilon^{-1})\left( 1+ o(1)I_{\beta(\eta),r}^{\hbox{int},3}(y)+o(1)I_{\delta,r}^{\hbox{ext},3}(y) \right),\end{aligned}$$ with $O(\epsilon^{-1})$ not depending on $k_1$ nor $k_2$, and $$\mathcal{A}_{\delta,\beta,r}(x) :=\{z\in B_r: -\delta\leq d(x+z)-d(x)\leq \beta\},$$ $$\mathcal{A}^{\hbox{ext}}_{\delta,r}(x):=\{z\in B_r: d(x+z)-d(x)< -\delta\},$$ $$\mathcal{A}^{\hbox{int}}_{\beta,r}(x):=\{z\in B_r: d(x+z)-d(x)>\beta\},$$ $$I^{\hbox{ext},1}_{\delta,r}(x) := \int_{\mathcal{A}^{\hbox{ext}}_{\delta,r}(x)} d\mu_x(z),$$ $$I^{\hbox{ext},2}_{\delta,r}(x) := \int_{\mathcal{A}^{\hbox{ext}}_{\delta,r}(x)} Dd(x)\cdot zd\mu_x(z),$$ $$I^{\hbox{ext},3}_{\delta,r}(x) := \int_{\mathcal{A}^{\hbox{ext}}_{\delta,r}(x)}|z| d\mu_x(z),$$ $$I^{\hbox{int},1}_{\beta,r}(x) := \int_{\mathcal{A}^{\hbox{ext}}_{\beta,r}(x)} d\mu_x(z),$$ $$I^{\hbox{int},2}_{\beta,r}(x) := \int_{\mathcal{A}^{\hbox{ext}}_{\beta,r}(x)} Dd(x)\cdot zd\mu_x(z),$$ $$I^{\hbox{int},3}_{\beta,r}(x) := \int_{\mathcal{A}^{\hbox{ext}}_{\beta,r}(x)}|z| d\mu_x(z),$$ $$I^{4}_{\delta,\beta,r}(x) := \frac{1}{2}\int_{\mathcal{A}_{\delta,\beta,r}(x)}|z|^2 d\mu_x(z),$$ $$I^{\hbox{tr}}(x) := \int_{r<|z|<1} Dd(x)\cdot zd\mu_x(z).$$
Using all of the previous, we can finally state the existence theorem.
\[solution\] Assuming $\xi$ is monotonic, the partial integro-differential equation $\mathcal{L}_{i}\xi_{i}=0$, subject to boundary conditions and $\xi_{i}>0$ has a viscosity solution between sub-solutions and super-solutions (irrespective of ordering) $$\begin{aligned}
f_i(x)&= \frac{\exp((-1)^i2Br)-\exp((-1)^i2Bx)}{\exp((-1)^i2Br)-\exp((-1)^i2Bl)},\nonumber \\
g_i(x)&= \exp((-1)^iB(x-l))\frac{\sinh \left(\frac{r-x}{\beta} \sqrt{ 2M +B^2}\right)}{\sinh \left(\frac{r-l}{\beta} \sqrt{ 2M +B^2}\right)},\end{aligned}$$ where $$\begin{aligned}
C&=\int_0^\infty \frac{y}{1+|y|}K(dy),\nonumber\\
B&= \frac{2 (C+\gamma) }{\beta^2},\nonumber\\
M&=\int_0^\infty K(dy).\end{aligned}$$
We define $$\begin{aligned}
H(x)&=\int_{0}^{\infty} \xi(x+y) K(dy), \nonumber \\
M&=\int_0^\infty K(dy).\end{aligned}$$ Consequently, $$\begin{aligned}
0=\mathcal{L}_{i}\xi(x)&= (-1)^{i+1}\gamma \xi'(x) +\frac{1}{2} \beta^2 \xi''(x) +(-1)^{i}\int_{\mathbb{R}_+} \left(\xi(x+y)-\xi(x)-\frac{y \xi'(x)}{1+|y|}\right)K(dy)\nonumber\end{aligned}$$ can be rewritten as $$\begin{aligned}
0=& (-1)^{i+1}\gamma \xi'(x) +\frac{1}{2} \beta^2 \xi''(x) +(-1)^{i}H(x)+(-1)^{i+1}M \xi(x)+(-1)^{i+1}C \xi'(x). \nonumber\end{aligned}$$
When $i=0$, the sign on $H$ is positive; therefore, we have sub-solution equation $$0= \frac{1}{2} \beta^2 \xi''(x) -(C+\gamma) \xi'(x) -M \xi(x).$$ Alternatively, when $i= 1$, we have as a super-solution equation $$0= \frac{1}{2} \beta^2 \xi''(x) +(C+\gamma) \xi'(x) +M \xi(x).$$
On the other hand, since $\xi >0$ inside $R$, there exists some $K_0\geq 0$ so that $$\xi(x+y)-\xi(x)\leq K_0\xi(x) \iff H(x)-M\xi(x)\leq K_0M\xi(x).$$ Using this, in the case $i=0$, we have super-solution equation $$0= -\gamma \xi'(x) +\frac{1}{2} \beta^2 \xi''(x) +K_0M\xi(x)-C \xi'(x).$$ When $i=1$, we have as a sub-solution equation instead $$0= \gamma \xi'(x) +\frac{1}{2} \beta^2 \xi''(x) -K_0M\xi(x)+C \xi'(x).$$
Now, because $\xi_i$ is monotonic, we can choose $K_0=0$, and so, the boundary-value problem gives the super-solution and sub-solutions claimed. Finally, applying the previous theorem \[EUtheorem\], we have the existence of a viscosity solution.
The existence of a more general viscosity solution to a higher dimensional problem is shown in the paper [@Roberts] .
We will use these super- and sub-solutions as envelopes to approximate a crucial parameter in the following algorithm to classify Lévy processes as having small or large jumps.
Given oil price close values in length-$n$ work day periods, we do the following:
1. An inverse Gaussian density, that is $\nu^*(dx)=c_1 x^{-3/2}e^{-c_2x}dx$, is fit to the distribution of negative percent daily jumps for the entire (training) data set.
2. We then fit the density of the Lévy measure from \[jumpgen\], $\nu^*(dx)=(1+ax)\nu(dx)$, to the distribution of the negative percent daily jumps for the $n$-length period. This gives a test statistic $a$ for the parameter in the hypothesis test.
3. A new list of close prices is produced by subtracting out any crashes above a certain threshold, and from this new list, we calculate the standard deviation $\sigma$. That is, if a crash occurs above a certain threshold, the new list of close prices proceeds as though there were no change from the previous day. This seeks to separate the jumps from volatility.
4. Using the density $\nu^*$ and standard deviation $\sigma$, we calculate $\gamma$, $\beta$, and $C$ from \[jumpgen\] and \[solution\].
5. The left side of the interval is chosen to be $-1$, then using $a$, $\sigma$, $\beta$, $\gamma$, and $C$ in the super- and sub-solution equations in \[solution\], we can solve for the right sides of the interval using $f_0(0)=1-\alpha_0$ and $g_0(0)=1-\alpha_0$, and take the average of the two. The parameter $\alpha_0$ is chosen to be the maximum desired probability of a Type I error.
6. Simulations of the log-likelihood process with drift $\gamma$, volatility $\beta$, and jumps represented by an inverse Gaussian process with expected value $-t \int_0^\infty x K(dx)$ at time $t$, are run. We record the frequency of exits out of the right-side of the interval to get a number that represents, relatively, the size of the jumps. We call this number the right-exit frequency.
Data analysis set-up {#dataanalysissetup}
====================
In this section, we present an overview of the data set in its entirety, and then develop three procedures used in the predictive classification problem. Motivated by the data analysis in [@recent], we consider the West Texas Intermediate (WTI or NYMEX) crude oil prices data set for the period June 1, 2009 to May 30, 2019 (Figure 1). We index the available dates from 0 (for June 1, 2009) to 2529 (for May 30, 2019). The following table (Table 1) summarizes various estimates for the data set.
Daily Price Change Daily Price Change %
--------- -------------------- ----------------------
Mean -0.0047 0.01370 %
Median 0.04399 0.06521 %
Maximum 7.62 12.32 %
Minimum -8.90 -10.53 %
: Properties of the empirical data set.
In Figure 1, the data, i.e., the West Texas Intermediate (WTI or NYMEX) crude oil prices data set for the period June 1, 2009 to May 30, 2019 (crude oil close price), is shown. In Figure 2 the distribution plot for close oil price is provided. Histograms for daily change in close oil price and daily change percentage in close oil price are provided in Figure 3 and Figure 4, respectively.
In the following subsections, three procedures are described for constructing the related classification problem. The procedures differ in the features used for the analysis: close prices, percent daily changes, and right-exit frequencies.
Close prices as features {#closeatt}
------------------------
We implement the following procedure to create a machine learning classification problem:
1. We consider the close price for the historical oil price data and create a new dataframe from the old where the columns will be $n$ consecutive close prices. For example, if the close prices are $$a_1,a_2,a_3,...,$$ then the first row of the data set will be $$a_1,a_2,...,a_n,$$ and the second row will be $$a_2,a_3,...,a_{n+1},$$ and so forth. \[newdf\]
2. We create a target column that is $0$ if the right-exit frequency of the next disjoint $n$ days is less than some threshold $p^*$, and is $1$ otherwise. For example, if the time period $$a_{1+n},a_{2+n},...,a_{2n-1}$$ has a significant frequency of right-exits, then the time period $$a_1,a_2,...,a_n$$ is given a target value $1$.
3. We run various classification algorithms where the input is a list of $n$ consecutive close prices, and the output is a $1$ to represent large jumps or $0$ to represent small jumps of the next $n$ consecutive close prices. Classification reports and confusion matrices are evaluated for each algorithm.
Percent daily changes as features {#percentatt}
---------------------------------
We implement the following procedure to create a machine learning classification problem:
1. We consider the percent daily changes for the historical oil price data and create a new data-frame from the old where the columns will be $n$ consecutive daily change percents. For example, if the changes are $$a_1,a_2,a_3,...,$$ then the first row of the data set will be $$a_1,a_2,...,a_n,$$ and the second row will be $$a_2,a_3,...,a_{n+1},$$ and so forth. \[newdf\]
2. We create a target column that is $0$ if the right-exit frequency of the next disjoint $n$ days is less than some threshold $p^*$, and is $1$ otherwise. For example, if the time period $$a_{1+n},a_{2+n},...,a_{2n-1}$$ has a significant frequency of right-exits, then the time period $$a_1,a_2,...,a_n$$ is given a target value $1$.
3. We run various classification algorithms where the input is a list of $n$ consecutive close prices, and the output is a $1$ to represent large jumps or $0$ to represent small jumps of the next $n$ consecutive close prices. Classification reports and confusion matrices are evaluated for each algorithm.
Right-exit frequencies as features {#rightatt}
----------------------------------
We implement the following procedure to create a machine learning classification problem:
1. Similar to the previous, we consider close prices for the historical oil price data and create a new data-frame exactly as before.
2. A new column is created that holds the right-exit frequencies for each consecutive set of $n$ days, say, $$b_1,b_2,b_3,....$$ These represent how large the jumps in close prices are for the previous $n$ days.
3. From this column, a new dataframe is created similarly to \[newdf\].
4. Finally, a target column is created: if the row is $$b_{30},b_{31},...,b_{30+n-1},$$ then the entry in the target column will be $b_{30+2n-1}$. This is the right-exit frequency of the proceeding $n$ days.
5. We run various classification algorithms where the input is a list of $n$ consecutive right-exit frequencies, and the output is $1$ to represent large jumps or $0$ to represent small jumps of the next $n$ consecutive close prices. Classification reports and confusion matrices are evaluated for each algorithm.
Numerical results {#numericalresults}
=================
Now we apply the procedures described in the last section to specific cases. For this section, the period length $n=10$. Further, $\alpha$, the parameter representing an approximation for the Type-I Error of the test is chosen to be $\alpha=0.9$. Three different time periods are used for training, and three are used for testing. The time periods are
- $T_1$: *training date(index)*: October 19, 2010 (350) to October 14, 2011 (600); and *testing date(index)*: August 1, 2012 (800) to December 21, 2012 (900), yielding constants $c_1=7.10 \times 10^{-7}$ and $c_2=4.80 \times 10^3$ in the training inverse Gaussian density;
- $T_2$: *training date(index)*: March 3, 2011 (450) to October 9, 2013 (1100); and *testing date(index)*: July 13, 2014 (1270) to May 13, 2015 (1500), yielding constants $c_1=6.77 \times 10^{-7}$ and $c_2=3.34 \times 10^3$ in the training inverse Gaussian density;
- $T_3$: *training date(index)*: March 17, 2010 (200) to October 9, 2013 (1100); and *testing date(index)*: December 19, 2013 (1150) to March 3, 2015 (1450), yielding constants $c_1=7.07 \times 10^{-7}$ and $c_2=3.54 \times 10^3$ in the training inverse Gaussian density.
Because the data is significantly imbalanced in favor of small-jump time periods, random small-jump periods from the training data are removed after performing algorithms \[closeatt\], \[percentatt\], and \[rightatt\]. The results of the machine learning algorithms using the time periods above are recorded in the following tables (Tables 2-10). Those used are linear regression (LR), decision trees (DT), random forests (RF), and three different types of neural nets, (A) a standard net, (B) a long-short term memory net, and (C) a LSTM net with a batch normalizer.
Many of the machine learning algorithms perform as one might expect from guessing uniformly whether the next time period would have big jumps. Some performed notably poorly, particularly the neural nets without a batch normalizer. However, the neural nets with a batch normalizer fairly consistently performed better than simply guessing uniformly. Probably the best method here is using a moderately small amount of data to train to avoid training too generally, using right-exit proportions as features, and then using a long-short term memory neural net with a batch normalizer. Figure 5 provides a histogram showing the distribution of right-exit frequencies for period lengths of 10 business days, for indices 91 (October 8, 2009) to 2460 (February 12, 2019). For each 10 consecutive days, 500 simulations are run, and the frequency of right-exits is recorded. The $x$-axis in the figure is the number of simulated processes that exit to the right of the testing interval for a given period, while the $y$-axis is the number of 10 day periods with that frequency of right-exits. The cut off for significant right-exit frequencies is chosen to be $p^*=3$.
Once the value of $\theta$ is estimated, this can be implemented in the refined BN-S model (and, , with $\theta=\theta'$). Equipped with $\theta$, as described in [@recent] and as shown in Theorem \[big12222\], the refined BN-S stochastic model can be used to incorporates *long range dependence* without actually changing the model. In addition, this shows a real-time application of data science for extracting a *deterministic component* out of processes that are thus far considered to be completely stochastic. For the computational effectiveness of $\theta$, the results in Tables 2-10 show better estimation compared to the benchmark study in [@recent].
LR DT RF Neural Network (A) LSTM (B) BN (C)
---------------------- ------ ------ ------ -------------------- ---------- --------
precision $\theta=0$ 0.62 0.76 0.87 0.00 0.00 0.66
recall $\theta=0$ 0.46 0.28 0.39 0.00 0.00 0.58
f1-score $\theta=0$ 0.53 0.41 0.54 0.00 0.00 0.62
support $\theta=0$ 67 67 67 67 67 67
precision $\theta=1$ 0.29 0.37 0.42 0.34 0.34 0.33
recall $\theta=1$ 0.44 0.82 0.88 1.00 1.00 0.41
f1-score $\theta=1$ 0.35 0.51 0.57 0.50 0.50 0.37
support $\theta=1$ 34 34 34 34 34 34
: Various estimations for $T_1$, using close prices as features.
LR DT RF Neural Network (A) LSTM (B) BN (C)
---------------------- ------ ------ ------ -------------------- ---------- --------
precision $\theta=0$ 0.66 0.67 0.56 0.64 0.60 0.64
recall $\theta=0$ 0.46 0.58 0.43 0.42 0.22 0.48
f1-score $\theta=0$ 0.54 0.62 0.49 0.50 0.33 0.55
support $\theta=0$ 67 67 67 67 67 67
precision $\theta=1$ 0.33 0.35 0.22 0.32 0.32 0.31
recall $\theta=1$ 0.53 0.44 0.32 0.53 0.71 0.47
f1-score $\theta=1$ 0.54 0.39 0.27 0.40 0.44 0.38
support $\theta=1$ 34 34 34 34 34 34
: Various estimations for $T_1$, using daily percent changes as features.
LR DT RF Neural Network (A) LSTM (B) BN (C)
---------------------- ------ ------ ------ -------------------- ---------- --------
precision $\theta=0$ 0.60 0.57 0.67 0.65 0.00 0.76
recall $\theta=0$ 0.43 0.46 0.57 0.48 0.00 0.55
f1-score $\theta=0$ 0.50 0.51 0.61 0.55 0.00 0.62
support $\theta=0$ 67 67 67 67 67 67
precision $\theta=1$ 0.28 0.23 0.34 0.33 0.34 0.42
recall $\theta=1$ 0.44 0.32 0.44 0.50 1.00 0.65
f1-score $\theta=1$ 0.34 0.27 0.38 0.40 0.50 0.51
support $\theta=1$ 34 34 34 34 34 34
: Various estimations for $T_1$, using right-exit frequencies as features.
LR DT RF Neural Network (A) LSTM (B) BN (C)
---------------------- ------ ------ ------ -------------------- ---------- --------
precision $\theta=0$ 0.71 0.71 0.70 0.00 0.71 0.71
recall $\theta=0$ 0.36 0.78 0.29 0.00 1.00 0.88
f1-score $\theta=0$ 0.47 0.74 0.41 0.00 0.83 0.79
support $\theta=0$ 163 163 163 163 163 163
precision $\theta=1$ 0.30 0.32 0.29 0.34 0.00 0.33
recall $\theta=1$ 0.65 0.25 0.69 1.00 0.00 0.15
f1-score $\theta=1$ 0.41 0.28 0.41 0.50 0.00 0.20
support $\theta=1$ 68 68 68 68 68 68
: Various estimations for $T_2$, using close prices as features.
LR DT RF Neural Network (A) LSTM (B) BN (C)
---------------------- ------ ------ ------ -------------------- ---------- --------
precision $\theta=0$ 0.70 0.67 0.72 0.75 0.71 0.69
recall $\theta=0$ 0.63 0.45 0.53 0.60 1.00 0.50
f1-score $\theta=0$ 0.66 0.54 0.61 0.66 0.83 0.58
support $\theta=0$ 163 163 163 163 163 163
precision $\theta=1$ 0.29 0.26 0.31 0.35 0.00 0.27
recall $\theta=1$ 0.35 0.47 0.51 0.53 0.00 0.46
f1-score $\theta=1$ 0.66 0.34 0.39 0.42 0.00 0.34
support $\theta=1$ 68 68 68 68 68 68
: Various estimations for $T_2$, using daily percent changes as features.
LR DT RF Neural Network (A) LSTM (B) BN (C)
---------------------- ------ ------ ------ -------------------- ---------- --------
precision $\theta=0$ 0.68 0.71 0.63 0.72 0.71 0.71
recall $\theta=0$ 0.45 0.55 0.52 0.52 1.00 0.48
f1-score $\theta=0$ 0.54 0.62 0.57 0.60 0.83 0.57
support $\theta=0$ 163 163 163 163 163 163
precision $\theta=1$ 0.27 0.30 0.19 0.31 0.00 0.30
recall $\theta=1$ 0.49 0.47 0.28 0.51 0.00 0.53
f1-score $\theta=1$ 0.35 0.37 0.23 0.39 0.00 0.38
support $\theta=1$ 68 68 68 68 68 68
: Various estimations for $T_2$, using right-exit frequencies as features.
LR DT RF Neural Network (A) LSTM (B) BN (C)
---------------------- ------ ------ ------ -------------------- ---------- --------
precision $\theta=0$ 0.72 0.70 0.69 0.71 0.71 0.70
recall $\theta=0$ 0.51 0.65 0.65 1.00 1.00 0.45
f1-score $\theta=0$ 0.60 0.67 0.67 0.83 0.83 0.55
support $\theta=0$ 214 214 214 214 214 214
precision $\theta=1$ 0.30 0.27 0.26 0.0 0.00 0.28
recall $\theta=1$ 0.51 0.32 0.30 0.00 0.00 0.52
f1-score $\theta=1$ 0.37 0.29 0.28 0.00 0.00 0.36
support $\theta=1$ 87 87 87 87 87 87
: Various estimations for $T_3$, using close prices as features.
LR DT RF Neural Network (A) LSTM (B) BN (C)
---------------------- ------ ------ ------ -------------------- ---------- --------
precision $\theta=0$ 0.74 0.80 0.73 0.69 0.00 0.77
recall $\theta=0$ 0.52 0.44 0.58 0.47 0.00 0.48
f1-score $\theta=0$ 0.61 0.57 0.65 0.56 0.00 0.59
support $\theta=0$ 214 214 214 214 214 214
precision $\theta=1$ 0.32 0.35 0.31 0.27 0.29 0.34
recall $\theta=1$ 0.55 0.74 0.47 0.49 1.00 0.66
f1-score $\theta=1$ 0.40 0.47 0.38 0.35 0.45 0.45
support $\theta=1$ 87 87 87 87 87 87
: Various estimations for $T_3$, using daily percent changes as features.
LR DT RF Neural Network (A) LSTM (B) BN (C)
---------------------- ------ ------ ------ -------------------- ---------- --------
precision $\theta=0$ 0.73 0.70 0.67 0.71 0.00 0.73
recall $\theta=0$ 0.37 0.52 0.50 0.46 0.00 0.53
f1-score $\theta=0$ 0.49 0.60 0.57 0.56 0.00 0.61
support $\theta=0$ 214 214 214 214 214 214
precision $\theta=1$ 0.30 0.28 0.24 0.28 0.29 0.31
recall $\theta=1$ 0.66 0.46 0.40 0.53 1.00 0.52
f1-score $\theta=1$ 0.41 0.35 0.30 0.37 0.45 0.39
support $\theta=1$ 87 87 87 87 87 87
: Various estimations for $T_3$, using right-exit frequencies as features.
Conclusion
==========
Mathematical modeling of agriculture and energy trading data is directly inspired by various stochastic models. Understanding and theoretical development of appropriate stochastic models contribute to a better understanding of the risk-management problem of various commodities, and various existing algorithms in a financial market depend on the underlying statistical model. Consequently, an improvement in the underlying model directly improves the existing algorithms. In this paper, a sequential decision making problem in connection to the Lévy process is studied to analyze the jump-size distribution. This is coupled with various machine and deep learning techniques to improve the existing stochastic models. Consequently, the analysis presented in this paper provides a necessary mathematical framework for an appropriate generalization of various stochastic models.
Future works related to this topic should definitely include seeking to find a more adequate approximation for the right side of the decision rule interval. This would greatly increase the sensitivity of the algorithms in classifying large-jump time periods. Applications to other data sets more independent of exogenous forces, and even across multiple streams of data using [@Roberts], should also be explored. Finally, constructing decision rules for hypothesis tests on other parameters in the underlying processes could open up this type of analysis to more generalized scenarios.
[1]{}
G. Barles, E. Chasseigne, & C. Imbert (2008), *On the Dirichlet Problem for Second-Order Elliptic Integro-Differential Equations*, Indiana Univ. Math. J., **57** (1), 213-246.
O. E. Barndorff-Nielsen (2001), Superposition of Ornstein-Uhlenbeck Type Processes, *Theory Probab. Appl.*, **45**, 175-194.
O. E. Barndorff-Nielsen & N. Shephard (2001), Non-Gaussian Ornstein-Uhlenbeck-based models and some of their uses in financial economics, *J. R. Stat. Soc. Ser. B Stat. Methodol.*, **63**, 167-241.
O. E. Barndorff-Nielsen & N. Shephard (2001), Modelling by Lévy Processes for Financial Econometrics, In *Lévy Processes : Theory and Applications* (eds O. E. Barndorff-Nielsen, T. Mikosch & S. Resnick), 283-318, Birkhäuser.
O. E. Barndorff-Nielsen, J. L. Jensen & M. S$\o$rensen (1998), Some stationary processes in discrete and continuous time, *Adv. in Appl. Probab.*, **30**, 989-1007.
C. Baum & V. Veeravalli (1994), *A sequential procedure for multihypothesis testing*, IEEE Transactions on Information Theory, **40** (6), 1994-1997.
B. Brodsky & B. Darkhovsky (2008), *Minimax methods for multihypothesis sequential testing and change-point detection problems*, Sequential Analysis, **27** (2), 141-173.
M. Carlisle & O. Hadjiliadis (2013), *Sequential Decision Making in Two-Dimensional Hypothesis Testing*, 52nd IEEE Conference on Decision and Control, <https://ieeexplore.ieee.org/document/6760919>.
Y. S. Chow, H. Robbins, & D. Siegmund (1991), *The theory of optimal stopping*, Dover Publications.
R. Cont & P. Tankov (2003), *Financial Modelling with Jump Processes*, hapman and Hall/CRC Financial Mathematics Series.
S. Dayanik, V. Poor, & S. Sezer (2008), *Sequential multi-hypothetis testing for compound Poisson processes*, Stochastics, **80** (1), 19-50.
G. K. Golubev & R.Z. Khas’minski (1983), *Sequential testing for several signals in Gaussian white noise*, Theory of Probability and its applications, **28**, 573-584.
S. Habtemicael & I. SenGupta (2016), Pricing variance and volatility swaps for Barndorff-Nielsen and Shephard process driven financial markets, *International Journal of Financial Engineering*, **03** (04), 1650027 (35 pages).
S. Habtemicael, M. Ghebremichael & I. SenGupta (2019), Volatility and variance swap using superposition of the Barndorff-Nielsen and Shephard type Lévy processes, *Sankhya B*, **81**, 75-92.
A. Irle (1981), Transitivity in problems of optimal stopping, *Annals of Probability*, **9**, 642-647.
A. Issaka & I. SenGupta (2017), Analysis of variance based instruments for Ornstein–Uhlenbeck type models: swap and price index, *Annals of Finance*, **13**(4), 401-434.
G. Lowther (2010), *Lévy Processes*, Stochastic Calculus Notes , <https://almostsure.wordpress.com/2010/11/23/levy-processes/>.
E. Nicolato & E. Venardos (2003), Option Pricing in Stochastic Volatility Models of the Ornstein-Uhlenbeck type, *Math. Finance*, **13**, 445-466.
M. Roberts & I. SenGupta (2020), Infinitesimal generators for two-dimensional Lévy process-driven hypothesis testing, *Annals of Finance*, **16** (1), 121-139.
I. SenGupta (2016), Generalized BN-S stochastic volatility model for option pricing, *International Journal of Theoretical and Applied Finance*, **19**(02), 1650014 (23 pages).
I. SenGupta, W. Nganje & E. Hanson (2020), Refinements of Barndorff-Nielsen and Shephard model: an analysis of crude oil price with machine learning, To appear in *Annals of Data Science*, Accepted March, 2020.
I. SenGupta, W. Wilson, & W. Nganje (2019), Barndorff-Nielsen and Shephard model: oil hedging with variance swap and option, *Mathematics and Financial Economics*, **13**(2), 209-226.
A. N. Shiryayev (1978), *Optimal Stopping Rules*, Springer-Verlag, New York.
A. Wald (1947), *Sequential Analysis*, Wiley, New York.
J. Wannenwetsch (2005), *Lévy Processes in Finance: The Change of Measure and Non-Linear Dependence*, Inaugural-Dissertation zur Erlangung des Grades eines Doktors der Wirtschafts- und Gesellschaftswissenschaften durch die Rechts- und Staatswissenschaftliche Fakultät der Rheinischen Friedrich-Wilhelms-Universität, Bonn. <http://hss.ulb.uni-bonn.de/2005/0545/0545.pdf>
W. Wilson, W. Nganje, S. Gebresilasie & I. SenGupta (2019), Barndorff-Nielsen and Shephard model for hedging energy with quantity risk, *High Frequency*, **2** (3-4), 202-214.
Refinitiv (2018), <https://www.refinitiv.com/en/resources/special-report/refinitiv-2019-artificial-intelligence-machine-learning-global-study>.
[^1]: Email: [email protected]
[^2]: Email: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We prove that the Baire Category Theorem is equivalent to the following: Let $G$ be a topological groupoid such that the unit space is a complete metric space, and
there is a countable cover of $G$ by neighbourhood bisections. If $G$ is effective, then $G$ is topologically principal.
address:
- |
Lisa Orloff Clark\
Department of Mathematics and Statistics\
University of Otago\
PO Box 56\
Dunedin Dunedin 9054\
New Zealand.
- |
Jonathan Brown\
Mathematics Department\
Kansas State University\
138 Cardwell Hall\
Manhattan, KS 66506-2602\
USA.
author:
- Jonathan Brown
- Lisa Orloff Clark
date: 'September 12, 2012'
title: A groupoid formulation of the Baire Category Theorem
---
Introduction
============
(See, for example, [@M Theorem 7.7.2].) \[BCT\] Suppose $X$ is a complete metric space. If $\{C_n\}$ is a countable collection of closed subsets of $X$, each with empty interior, then $\bigcup_n C_n$ has empty interior.
The proof of the Baire Category Theorem, originally formulated by Baire in the 1890’s, requires a variant of the Axiom of Choice [@D Chapter 13]. In fact, [@Blair] and [@Goldblatt] show the Baire Category Theorem is equivalent to the *the Principle of Dependant Choices* which says:
> Suppose $X$ is a set and $R \subseteq X \times X$ is a relation such that for each $x \in X$, there exists $y \in X$ such that $(x,y) \in R$. Then there is a sequence $\{x_n\} \subseteq X$ such that $(x_n,x_{n+1}) \in R$ for all $n$.
The Principle of Dependent Choices falls strictly between the Countable Axiom of Choice and the Axiom of Choice; see [@jech] for more details.
In this note, we show the Baire Category Theorem is equivalent to Theorem \[GET\] below, which is a surprising result about *effective groupoids*. We discovered a version of Theorem \[GET\] in our study of simple groupoid $C^*$-algebras [@BCFS Lemma 3.3]. A version also appears in [@Renault:IMSB08 Proposition 3.6] in the context of maximal abelian subalgebras of $C^*$-algebras. That Theorem \[GET\] implies the Baire Category Theorem is entirely new; but even our proof that the Baire Category Theorem implies Theorem \[GET\] is different from those in [@BCFS] and [@Renault:IMSB08].
Preliminaries
=============
A groupoid $G$ is a small category in which every morphism is invertible. We identify the set of objects of $G$ with the set of identity morphisms and denote this set ${G^{(0)}}$. For $\gamma\in G$, we denote the range and source (domain) of $\gamma$ by $r(\gamma)$ and $s(\gamma)$ respectively. Thus $r,s: G\to {G^{(0)}}\subseteq G$. We define $G^{(2)}:=\{(\gamma,\eta)\in G\times G: r(\eta)=s(\gamma)\}$; $G^{(2)}$ consists of precisely those pairs of morphisms that can be composed in $G$. For any $x\in {G^{(0)}}$, the *isotropy group* at $x$ is the group $$xGx:=\{\gamma\in G: r(\gamma)=s(\gamma)=x\}.$$ The *isotropy subgroupoid* of $G$ is $$\operatorname{Iso}(G):=\bigcup_{x\in {G^{(0)}}} xGx$$ which is itself a groupoid. If $B \subseteq G$, then we also write ${\operatorname{Iso}}(B) := {\operatorname{Iso}}(G) \cap B$. We say $G$ is a *group bundle* if $\operatorname{Iso}(G)=G$.
A groupoid $G$ is a *topological groupoid* if $G$ is equipped with a topology so that composition and inversion are continuous. In this case, $r$ and $s$ are continuous maps. If ${G^{(0)}}$ is Hausdorff, then the continuity of $r$ and $s$ implies that ${\operatorname{Iso}}(G)$ is a closed subset of $G$.
An open set $A\subseteq G$ is called an *open bisection* if $r(A)$ and $s(A)$ are open in $G$ and $r$ and $s$ restricted to $A$ are homeomorphisms onto their image; in particular $r$ and $s$ are injective on $A$.
We say a groupoid $G$ is *topologically principal* if the set $\{x\in {G^{(0)}}: xGx\ \neq \{x\}\}$ has empty interior in ${G^{(0)}}$. A groupoid $G$ is *effective* if $\operatorname{Iso}(G) - {G^{(0)}}$ has empty interior.
When does effective imply topologically principal?
==================================================
In [@Renault:IMSB08 Proposition 3.6(ii)] Renault considers effective groupoids whose unit spaces are ‘Baire’. We can interpret Renault’s result as saying that Theorem \[BCT\] implies the following:
[@Renault:IMSB08 Proposition 3.6(ii)] \[thm: effect ren\] Suppose $G$ is a topological groupoid such that:
1. the unit space is a complete metric space;
2. \[it:2ren\] $G$ has a countable cover consisting of open bisections.
If $G$ is effective, then $G$ is topologically principal.
Our original intention was to show that Theorem \[thm: effect ren\] is equivalent to Theorem \[BCT\]. However, we eventually realised that such a result will only hold if we weaken the hypotheses of Theorem \[thm: effect ren\]. To see why, consider the class of effective groupoids constructed in Example \[ex: main\] below. Each has the property that Theorem \[BCT\] implies it is topologically principal. At the same time, groupoids in this class may not satisfy the hypotheses of Theorem \[thm: effect ren\]. (We will also use this class of examples later in the proof of our main result.)
\[ex: main\] Let $X$ be a complete metric space and $\{C_n\}$ be a countable collection of closed subsets of $X$, each with empty interior. Define $C:=\bigcup_n C_n$.
Let $G$ be the group bundle with unit space $X$ and isotropy groups $$xGx :=
\begin{cases} \mathbb{Z}_2, & \text{ if $x \in C$;}\\
\{ 0 \}, & \text{otherwise.}
\end{cases}$$ We identify the identity element $0 \in xGx$ with $x$. For each $x \in C$, we write $\gamma_x$ for the nontrivial element of $xGx$. Notice that $$G^{(2)} = \{(x,x) : x\in X\} \cup \{(\gamma_x, \gamma_x): x \in C\} \cup \{(x, \gamma_x): x \in C\} \cup \{(\gamma_x, x): x \in C\}.$$
To make $G$ into a topological groupoid, first let ${\mathcal T}$ be the topology for $X$. Define the collection $${\mathcal B}:= {\mathcal T}\cup \{V \subseteq G : V = (W - \{ x \}) \cup \{ \gamma_x\} \text{ for some } W \in {\mathcal T}\text{ and } x \in C\cap W\}.$$ The collection ${\mathcal B}$ forms a basis for a topology on $G$. To see this, note that ${\mathcal B}$ covers $G$ and since ${\mathcal T}$ is the topology for $X$, it is easy to see that $U, V\in {\mathcal B}$ implies $U\cap V\in {\mathcal B}$. We claim that $G$ endowed with the topology generated by ${\mathcal B}$ is a topological groupoid. Indeed, inversion is given by the identity and is thus continuous. Now let $m: G^{(2)}\to G$ be the composition map. Fix $V \in {\mathcal B}$. If $V \in {\mathcal T}$, then $$m^{-1}(V) = \{(x,x):x \in V\} \cup \{(\gamma_x, \gamma_x): x \in V\}.$$ If $V = (W - \{ y \}) \cup \{ \gamma_y\}$ for some $W \in {\mathcal T}$ and $y \in C \cap W$, then $$m^{-1}(V) = \{(x,x):x \in W - \{y\} \} \cup \{(\gamma_y, y), (y, \gamma_y)\}\cup \{(\gamma_x,\gamma_x): x\in (W-\{y\})\cap C\} .$$ In both cases, it is straightforward to show that $m^{-1}(V)$ is open in $G^{(2)}$, hence composition is continuous as claimed.
Since every element of ${\mathcal B}$ intersects the unit space, the set $\operatorname{Iso}(G) - {G^{(0)}}= G - {G^{(0)}}$ contains no open sets, so $G$ is effective. By construction, ${G^{(0)}}= X$ is a complete metric space and $$C = \{x\in {G^{(0)}}: xGx\ \neq \{x\}\},$$ so Theorem \[BCT\] implies that $G$ is also topologically principal. Notice that $G$ need not satisfy item of Renault’s Theorem \[thm: effect ren\]. Indeed if $X=[0,1]$ and $C_n=C$ is the Cantor set for all $n$, there is no countable cover of $G$ consisting of open bisections. To see this suppose $\{U_i\}$ is any countable open cover of $G$. Since the Cantor set is uncountable there exists an $i_0$ such that $A_{i_0}:=\{x\in C: \gamma_x\in U_{i_0}\}$ is uncountable. For each $x\in A_{i_0}$ pick a basis element $(V_x-\{x\})\cup \{\gamma_x\}$ contained in $U_{i_0}$. Since the standard basis for $[0,1]$ is given by connected intervals, we can assume that $V_x$ is connected. For each $n \in {{\mathbb{Z}}}^+$ define $D_n:=\{x\in A_{i_0}: \text{~diameter of~} V_x\text{~is greater than~} 2/n\}$. Since $A_{i_0}$ is uncountable, there exists $n_0$ such that $D_{n_0}$ is uncountable. Now consider the partition $\{P_m:=[m/2n_0, (m+1)/2n_0]\}$ of $[0,1]$ where $0 \leq m \leq 2n_0-1$. Since $D_{n_0}$ is uncountable, there exists an $m_0$ such that $D_{n_0}\cap P_{m_0}$ is uncountable. By the definition of $D_{n_0}$ this implies that for every $x\in D_{n_0}\cap P_{m_0}$ both $x$ and $\gamma_x$ are in $U_{i_0}$ and so $U_{i_0}$ is *not* an open bisection.
While the groupoids considered above need not have a countable cover of open bisections, they do have a countable cover consisting of ‘well-behaved sets’. We call these sets *neighbourhood bisections*. (We denote the interior of a set $D$ by ${\operatorname{Int}}(D)$.)
\[def:nb\] A set $B \subseteq G$ is called a *neighbourhood bisection* if the following hold:
1. \[it:nb\_1\] $B \subseteq \overline{{\operatorname{Int}}(B)}$;
2. \[it:nb\_2\]$r|_B$ and $s|_B$ are injective;
3. \[it:nb\_3\]$r(B)$ and $s(B)$ are open in $G$;
4. \[it:nb\_4\]${\operatorname{Int}}(B)$ is an open bisection;
5. \[it:nb\_5\]$B - {\operatorname{Int}}(B) \subseteq {\operatorname{Iso}}(B) - {G^{(0)}}$.
In the next section we prove the following theorem is equivalent to Theorem \[BCT\]. One part of our proof involves showing the class of groupoids constructed in Example \[ex: main\] do indeed have a countable cover consisting of neighbourhood bisections.
\[GET\] Suppose $G$ is a topological groupoid such that:
1. \[it:GET\_1\] the unit space is a complete metric space;
2. \[it:GET\_2\] $G$ has a countable cover consisting of neighbourhood bisections.
If $G$ is effective, then $G$ is topologically principal.
Suppose $G$ is a groupoid satisfying the hypotheses of Theorem \[GET\], then ${G^{(0)}}$ is open in $G$. To see this, let $\{B_n\}$ be a countable cover of $G$ by neighbourhood bisections, then ${G^{(0)}}=\bigcup r(B_n)$ which is open.
An *étale* groupoid is a topological groupoid that has a cover consisting of open bisections. When studying $C^*$-algebras associated to groupoids, one often considers second-countable, locally compact, Hausdorff groupoids that are étale. These groupoids satisfy the hypothesis of Theorem \[GET\].
Suppose $G$ is a topological groupoid. If $r$ is an open map, then $G$ topologically principal implies $G$ is effective. See [@BCFS Examples 6.3 and 6.4] for examples of groupoids (that do not satisfy the hypothesis of Theorem \[GET\]) that are effective but not topologically principal.
Main Result
===========
\[main\] Theorem \[BCT\] is equivalent to Theorem \[GET\].
Before we prove Theorem \[main\], we establish the following two lemmas. The first lemma is used to prove the second (Lemma \[lem:main\]); Lemma \[lem:main\] is a key step in our proof of Theorem \[main\].
\[lem:1\] Suppose $G$ is a topological groupoid such that ${G^{(0)}}$ is open in $G$, $B \subseteq G$ is a neighbourhood bisection and $D \subseteq B$ is closed in $B$ where $B$ is endowed with the subspace topology. Suppose that $B - {\operatorname{Int}}(B) \subseteq D$. Then $r(D)$ is closed in $r(B)$ where $r(B)$ is endowed with the subspace topology.
Let $G$, $B$ and $D$ be as stated. Then $$D = (D \cap {\operatorname{Int}}(B) ) \cup (B - {\operatorname{Int}}(B))$$ which means $$r(D) = r(D \cap {\operatorname{Int}}(B) ) \cup r(B - {\operatorname{Int}}(B)).$$ Since $r|_B$ is a bijection onto its image, $r(B - {\operatorname{Int}}(B)) = r(B) - r({\operatorname{Int}}(B))$ which is closed in $r(B)$ as $r({\operatorname{Int}}(B))$ is open. Further $r(D \cap {\operatorname{Int}}(B))$ is closed in $r({\operatorname{Int}}(B))$ because $r$ restricted to ${\operatorname{Int}}(B)$ is a homeomorphism. Thus there exists a closed set $C$ such that $r({\operatorname{Int}}(B)) \cap C = r({\operatorname{Int}}(B) \cap D)$. Therefore $$\begin{aligned}
r(D) &= r({\operatorname{Int}}(B) \cap D) \cup r(B - {\operatorname{Int}}(B))\\
&= (r({\operatorname{Int}}(B)) \cap C) \cup (r(B - {\operatorname{Int}}(B)) \cap C) \cup r(B - {\operatorname{Int}}(B))\\
&= (r(B) \cap C) \cup r(B - {\operatorname{Int}}(B))\end{aligned}$$ which is closed in $r(B)$.
\[lem:main\] Suppose $G$ is an effective groupoid such that ${G^{(0)}}$ is open in $G$ and $B$ is a neighbourhood bisection. Then
1. \[it: empty nbhd\] $r({\operatorname{Iso}}(B) - {G^{(0)}})$ has empty interior, and
2. \[it: empty cls\] $\overline{r({\operatorname{Iso}}(B) - {G^{(0)}})}$ has empty interior.
For , by way of contradiction, suppose there exists a nonempty open set $W \subseteq r({\operatorname{Iso}}(B) - {G^{(0)}})$. Thus $W \cap r(B) \neq \emptyset$, and since $ B \subseteq \overline{{\operatorname{Int}}(B)}$, we have $
W \cap r\left(\overline{{\operatorname{Int}}(B)}\right) \neq \emptyset$. Therefore $$W \cap \overline{r({\operatorname{Int}}(B))} \neq \emptyset
\text{ because $r\left(\overline{{\operatorname{Int}}(B)}\right) \subseteq \overline{r({\operatorname{Int}}(B))}$.}$$ Hence $W \cap r({\operatorname{Int}}(B))$ is a nonempty open set contained in ${G^{(0)}}$. Since $$\phi:=r|_{{\operatorname{Int}}(B)}$$ is a homeomorphism, $$\phi^{-1}(W \cap r({\operatorname{Int}}(B)))$$ is a nonempty open subset of ${\operatorname{Int}}(B)$ and thus is open in $G$. Since $r$ is injective on $B$ and $W \subseteq r({\operatorname{Iso}}(B) - {G^{(0)}})$, $$(\phi^{-1}(W \cap r({\operatorname{Int}}(B))) \subseteq {\operatorname{Iso}}(B) - {G^{(0)}}\subseteq {\operatorname{Iso}}(G) - {G^{(0)}}.$$ This is a contradiction because $G$ is effective.
For , by way of contradiction, assume there exists a nonempty open subset $$V \subseteq \overline{r({\operatorname{Iso}}(B) - {G^{(0)}})}.$$ Notice that $V \cap r(B)$ is a nonempty open subset of ${G^{(0)}}$. Further, $$V \cap r(B) \subseteq \overline{r({\operatorname{Iso}}(B) - {G^{(0)}})} \cap r(B).$$
We show $\overline{r({\operatorname{Iso}}(B) - {G^{(0)}})} \cap r(B)=r({\operatorname{Iso}}(B) - {G^{(0)}}).$ Since ${\operatorname{Iso}}(B)$ is closed in $B$ and ${G^{(0)}}$ is open, ${\operatorname{Iso}}(B) - {G^{(0)}}$ is also closed in $B$. Also, $B - {\operatorname{Int}}(B) \subseteq {\operatorname{Iso}}(B) - {G^{(0)}}$ by assumption. Therefore we apply Lemma \[lem:1\] to see that $r({\operatorname{Iso}}(B) - {G^{(0)}})$ is closed in $r(B)$. Thus $$\begin{aligned}
r({\operatorname{Iso}}(B) - {G^{(0)}}) &= \overline{r({\operatorname{Iso}}(B) - {G^{(0)}})} \cap r(B), \text{ and so }\\
V \cap r(B) &\subseteq r({\operatorname{Iso}}(B) - {G^{(0)}})\end{aligned}$$ which contradicts item .
Suppose Theorem \[BCT\] holds. Let $G$ be a topological groupoid with a countable cover of neighbourhood bisections $\{ B_n\}$ such that $G^{(0)}$ is a complete metric space. Suppose also that $G$ is effective.
By Lemma \[lem:main\], the set $\overline{r({\operatorname{Iso}}(B_n)-{G^{(0)}})}$ has empty interior for every $n$. Define $C_n:= \overline{r({\operatorname{Iso}}(B_n)-{G^{(0)}})} \cap {G^{(0)}}$ for each $n$. Notice that each $C_n$ is a closed subset of ${G^{(0)}}$. Because ${G^{(0)}}$ is open in $G$, each $C_n$ also has empty interior in ${G^{(0)}}$. Applying Theorem \[BCT\] (Baire Category Theorem) to the collection $\{C_n\}$ we see that $$C:= \bigcup_n C_n$$ has empty interior. By construction, $C$ contains the units with nontrivial isotropy. Therefore, $G$ is topologically principal.
Conversely, suppose that Theorem \[GET\] holds. Let $X$ be a complete metric space with topology ${\mathcal T}$ and $\{C_n\}$ be a countable collection of closed subsets of $X$, each with empty interior. With out loss of generality we can assume $C_0=\emptyset$. Let $C=\bigcup_n C_n$. Define $G$ as in Example \[ex: main\]. Since ${G^{(0)}}=X$ as a topological space, $G$ satisfies of Theorem \[GET\].
For each $n$, define $$B_n:=\left(X-C_n\right) \cup \{\gamma_x : x\in C_n\}.$$ We claim that each $B_n$ is a neighbourhood bisection. To prove this, we must check each of the items in Definition \[def:nb\]. To see , first note that $X- C_n$ is open in $G$ and contained in $B_n$. Thus $X- C_n\subseteq {\operatorname{Int}}(B_n)$. We show that $B_n\subseteq \overline{X- C_n}\subseteq \overline{{\operatorname{Int}}(B_n)}$. Consider $\gamma_x$ for some $x\in C_n$. For every $V\in {\mathcal B}$ with $\gamma_x\in V$ we have $V=W-\{x\}\cup \{\gamma_x\}$ where $W\in {\mathcal T}$ and $x\in W$. Now $V\cap (X- C_n)=W\cap (X- C_n)$ is nonempty because $C_n$ has empty interior. Therefore $\gamma_x\in \overline{X- C_n}$. Since $B_n=X- C_n\cup \{\gamma_x: x\in C_n\}$, we have $B_n\subseteq \overline{X- C_n}$. That $B_n$ satisfies item is clear and $r(B_n)=X=s(B_n)$ is open in $G$ giving us item . Since $r(V)$ is in ${\mathcal T}$ for every $V \in {\mathcal B}$, $r=s$ is an open map and hence $r|_{{\operatorname{Int}}(B_n)}=s|_{{\operatorname{Int}}(B_n)}$ is a homeomorphism with open image giving item . Lastly, since $B_n \cap {G^{(0)}}= X- C_n \subseteq {\operatorname{Int}}(B_n)$ and $G = {\operatorname{Iso}}(G)$, we get item . Thus $\{B_n\}_n$ is a countable cover of $G$ by neighbourhood bisections and $G$ satisfies item (\[it:GET\_2\]) of Theorem \[GET\].
We showed in Example \[ex: main\] that $G$ is effective. Therefore $G$ is topologically principal by Theorem \[GET\]. Thus $$C = \{x\in {G^{(0)}}: xGx\ \neq \{x\}\}$$ has empty interior proving Theorem \[BCT\].
Thanks to François Dorais for the very helpful email correspondence.
[99]{}
C.E. Blair, *The Baire Category Theorem implies the Principle of Dependent Choices*, Bull. Acad. Polon. Sci. Sér. Sci. Math. Astronom. Phys. **25** (1977), 933–934.
J.H. Brown, L.O. Clark, C. Farthing and A. Sims, *Simplicity of algebras associated to étale groupoids*, submitted. arXiv:1204.3127v1 \[math.OA\]
W. Dunham, *The calculus gallery: masterpieces from Newton to Lebesgue*, Princeton University Press, Princeton, NJ, 2005.
R. Goldblatt, *On the role of the Baire Category Theorem and Dependent Choice in the foundations of logic*, J. Symbolic Logic **50** (1985), 412–422.
T.J. Jech, *The Axiom of Choice*, reprint, Dover Publications, Inc., Mineola, NY, 2008.
J.R. Munkres, *Topology*, second edition, Prentice Hall, Upper Saddle River, 2000.
A. Paterson, *Groupoids, inverse semigroups, and their operator algebras,* Birkhäuser Boston, Inc., Boston, MA, 1999.
J. Renault, *Cartan subalgebras in [$C^*$]{}-algebras*, Irish Math. Soc. Bulletin **61** (2008), 29–63.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We show that a sizeable azimuthal asymmetry, characterized by a coefficient $v_2$, is to be expected for large $p_T$ direct photons produced in non-central high energy nuclear collisions. This signal is generated by photons radiated by jets interacting with the surrounding hot plasma. The anisotropy is out of phase by an angle $\pi/2$ with respect to that associated with the elliptic anisotropy of hadrons, leading to negative values of $v_2$. Such an asymmetry, if observed, could be a signature for the presence of a quark gluon plasma and would establish the importance of jet-plasma interactions as a source of electromagnetic radiation.'
author:
- 'S. Turbide'
- 'C. Gale'
- 'R. J. Fries'
title: Azimuthal Asymmetry of Direct Photons in High Energy Nuclear Collisions
---
At the Relativistic Heavy Ion Collider (RHIC) and soon at the Large Hadron Collider (LHC), nuclei are collided at ultrarelativistic energies in order to create a new state of matter: the quark gluon plasma (QGP) [@Harris:1996zx]. Emitted particles that only interact through the electroweak interactions are very unlikely to interact again despite the dense medium which is created. Thus they are able to carry to the detectors information about the state of the system at the time they were created [@Feinb:76]. Photons and leptons thus constitute a unique class of penetrating probes.
The two most interesting sources of photons are those where the plasma is directly involved in the emission. These are the thermal radiation from the hot QGP [@KaLiSei:91] and the radiation induced by the passage of high energy jets through the plasma [@FMS:02; @Zakharov:04; @TGJM:05]. The thermal radiation is emitted predominantly with low transverse momentum $p_T$ and has to compete with photon emission from the hot hadronic gas at later times [@XSB:92; @Turb:04]. Photons from jets are an important source at intermediate $p_T$, where they compete with photons from primary hard scatterings between partons of the nuclei [@Owens:86]. They probe the thickness of the medium: the longer the path of the jet, the more photons are emitted.
Obviously, measuring photons from either of the sources involving the quark gluon plasma would be an important step toward establishing its existence and would provide a stringent test of the reaction dynamics. We refer the reader to [@TGJM:05] for a discussion of the different photon sources. Recently, the PHENIX collaboration at RHIC published their first results on direct photons measured in Au+Au collisions at $\sqrt{s_{\rm{NN}}}=200$ GeV [@phenix:05gammaauau]. Note that direct photons here are defined as the total inclusive photon yield minus the photons originating from the decay of hadrons like $\pi^0$ and $\eta$. The sources discussed above will contribute to the direct photon signal. The data covers the intermediate and high-$p_T$ range where thermal photons are not a leading source. This makes it particularly attractive to look for photons from jet-plasma interactions.
Nuclear collisions at finite impact parameter $b>0$ start out in an initial state which is not azimuthally symmetric around the beam axis. Instead, the initial overlap zone of the two nuclei has an “almond” shape. Therefore particle spectra measured in the final state are not necessarily isotropic around the beam axis. It has been argued that the translation of the original space-time asymmetry into a momentum space anisotropy can reveal important information about the system [@Olli:92]. Two different mechanisms are important here: hydrodynamic pressure for the bulk of the matter, at low- to intermediate-$p_T$, and a simple optical-depth argument for intermediate- to high-$p_T$ particles. We introduce a novel third mechanism in this Letter.
Let us define the reaction plane as the plane spanned by the beam axis and the impact parameter of the colliding nuclei. For the bulk of the matter, the initial space-time asymmetry leads to an anisotropic pressure gradient which is larger where the material is thinner, i.e. in the event plane. This translates into a larger flow of matter in this direction. The anisotropy is analyzed in terms of Fourier coefficients $v_k$ defined from the particle yield $dN/p_T dp_Td\phi$ as [@VoloZhang:94] $$\frac{dN}{p_Tdp_Td\phi} = \frac{dN}{2\pi p_Tdp_T} \left[ 1+
\sum_k 2 v_k(p_T) \cos( k\phi) \right]$$ where the angle $\phi$ is defined with respect to the event plane. At midrapidity all odd coefficients vanish for symmetry reasons, leaving the coefficient $v_2$ to be the most important one. Its size determines the ellipsoidal shape of the anisotropy. It is clear that the elliptic asymmetry coefficient $v_2$ is always positive for hadrons at low and intermediate $p_T$ due to the hydrodynamic flow. On the other hand, jets lose more energy when they are born into a direction where the medium is thicker, i.e. out of the reaction plane. The stronger jet quenching leads to fewer hadrons at intermediate and high $p_T$ emitted into this direction. This “optical $v_2$” is not associated with flow but with absorption and implies positive $v_2$ for hadrons from jets. Measurements at RHIC for several hadron species confirm this behavior [@star:02v2].
In this Letter, we discuss $v_2$ of direct photons. We concentrate on intermediate and high $p_T$ and the $v_2$ from all the relevant processes. We define a mechanism that works by absorption of particles or jets going through the medium as optical. It turns out that in some cases a new inverse-optical mechanism is in place for photons: there are more of them emitted into the direction where the nuclear overlap zone is thicker, thus leading to a situation where the anisotropy is shifted by a phase $\pi/2$. Correspondingly $v_2$ is negative in this case.
Let us now discuss the different contributions to the direct photon spectrum. Direct photons from primary hard Compton and annihilation processes $a+b\to \gamma+c$ are produced symmetrically with $$\label{N-N}
\frac{dN^{\rm N-N}}{p_T dp_Td\phi} = T_{AB} f_{a/A} \otimes
\sigma_{a+b\to \gamma+c} \otimes f_{b/B}.$$ Here $\sigma_{a+b\to \gamma+c}$ is the cross section between partons, $f_{a/A}$, $f_{b/B}$ are parton distribution functions in the nuclei $A$ and $B$ and $T_{AB}$ is the overlap factor of the nuclei. The primary hard direct photons do not suffer any final state effect and do not exhibit any elliptic asymmetry.
Jets from processes ($a+b \to c+d$) are also produced symmetrically, however they are quenched once they start to propagate through the plasma. This is the optical mechanism that leads to positive $v_2$ for hadrons fragmenting from jets. We expect photons fragmenting from such jets in the vacuum ($c\to c+\gamma$, after $c$ propagated through the medium) to exhibit the same anisotropy. Their yield at midrapidity is given by $$\frac{dN^{\rm jet-frag}}{p_T dp_Td\phi} = \sum_f \left.
\frac{dN^f(\phi)}{dq} \right|_{q=p_T/z} \otimes D_{f/\gamma}(z,p_T)$$ where $dN^f(\phi)/dq$ is the distribution of jet partons $f$ with momentum $q$ traveling into the direction given by the angle $\phi$, and $D_{f/\gamma}$ is the photon fragmentation function.
The interaction of jets with the medium can also produce photons in different ways: (i) scattering off plasma components can induce photon bremsstrahlung, (ii) hard leading partons may annihilate with thermal ones ($q+\bar q \to \gamma+g)$, or they can participate in Compton scattering ($q (\bar{q}) +g \to q (\bar{q})
+\gamma$). The latter case is also called jet-photon conversion, because the cross section is dominated by transfer of the entire jet momentum to the photon, $\mathbf{p}_\gamma\approx \mathbf{p}_{\rm jet}$. The jet-photon conversion yield at midrapidity is given by [@wong] $$\begin{gathered}
\frac{dN^{\rm jet-th}}{p_T dp_T d\phi dy} = \int d^4 x
\frac{\alpha\alpha_s T^2 }{8\pi^2} \sum_q \left( \frac{e_q}{e} \right)^2
\\ \times f_q(x;p_T,\phi,y) \left[ 2 \ln\frac{4E_\gamma T}{m^2} -C \right]\end{gathered}$$ with $C=2.332$ and $m^2=4\pi\alpha_s T^2/3$. The distribution of jet partons $f_q(x;p_T,\phi,y)$ at a space-time point $x$ is determined from the time dependent spectrum of jet partons propagating in the plasma, $dN^q/dq(\tau)$, as discussed in [@FMS:02; @TGJM:05]. The time dependence is governed by the energy loss through induced gluon radiation, obtained with the complete leading order description by Arnold, Moore and Yaffe (AMY) [@AMY]. It is clear that an anisotropy in $\phi$ is introduced by the different path lengths for jets traveling in and out of the event plane, leading to an increased probability for a jet-photon conversion in the direction where the medium is thicker. Such an inverse optical effect has not been observed before.
Medium induced bremsstrahlung ($q\to\gamma+q$) is implemented directly in the AMY formalism through splitting functions $d\Gamma^{q\to q\gamma}/dkdt$. The photon yield from this process is obtained by $$\label{jet-brem}
\frac{dN^{\rm jet-br}}{dp_Td\phi} = \int d^2 r_\perp
\mathcal{P}(\mathbf{r}_\perp)\int^{d}_0 dt dk\frac{dN^q}{dq}\frac{d\Gamma^{q\to q\gamma}(q,k)}{dkdt}$$ Here $\mathcal{P}(\mathbf{r}_\perp)$ is the spatial distribution of hard processes creating jets in the transverse plane and $d=d(\mathbf{r}_\perp,\phi)$ is the distance the jet has to travel from $\mathbf{r}_\perp$ into the direction of the angle $\phi$ to leave the fireball. Again it is obvious that the probability for induced bremsstrahlung to occur increases with the path length $d$ of the jet. Hence these photons are preferentially emitted into the direction where the medium is thicker, leading to negative $v_2$.
Finally, thermal photon emission constitutes another contribution to the direct photon spectrum. Thermal photons emitted by the medium are not prone to any optical effects, but the emitting matter experiences the anisotropic hydrodynamic push. However, the emission of intermediate and large $p_T$ photons peaks strongly at very early times where the temperature is highest [@kolb]. We verified that at the time the system generates significant radial and elliptic flow the rate of thermal photons at intermediate and high $p_T$ is negligible. Let us summarize what we have so far. We identified two processes, induced bremsstrahlung from jets and jet-photon conversion, that we expect to exhibit an inverse optical anisotropy ($v_2 <0$). Photons from fragmentation show the regular optical anisotropy ($v_2>0$), while primary hard and thermal photons do not contribute to $v_2$ at intermediate and high $p_T$.
To quantify our arguments, we carry out a numerical calculation for Au+Au collisions at RHIC ($\sqrt{s}=200$ GeV). Photon spectra at midrapidity with their dependence on the azimuthal angle $\phi$ are calculated as described above for three different centrality classes. Our modeling of the nuclear collision is introduced in [@TGJM:05]. The initial conditions are constrained by $\tau_iT_i^3\sim dN(b)/dy/A_\perp(b)$, where the charged particle pseudorapidity densities $dN/dy$ can be found in [@phobos] and $A_\perp$ is the overlap area of the nuclei in the transverse plane. With fixed initial time $\tau_i=0.26$ fm/c, the initial temperatures are $T_i$=370, 360 and 310 MeV for centrality classes $0-20\%, 20-40\%$ and $40-60\%$ respectively. Comparing with measured photon spectra, a good agreement is obtained, for all centrality classes. Details will appear elsewhere [@santa_fe]. The coefficients $v_2$ can then be calculated by using $$\label{eq:v2}
v_2 (p_T) = \frac{\int d\phi \cos(2\phi) dN/dp_Td\phi}{
dN/dp_T}.$$
![\[fig:1\] (Color online) Photon $v_2$ as a function of $p_T$ for Au+Au collisions at RHIC. Three different centrality bins are given. The dotted lines show $v_2$ for primary hard photons and jet fragmentation only, and the solid lines show all direct photons. Energy loss is included in both cases. The dashed line is the same as the solid line but without energy loss of jets taken into account.](v2_ph_alph34_side.eps){width="\columnwidth"}
Fig. \[fig:1\] shows the coefficient $v_2$ as a function of $p_T$ for Au+Au collisions at RHIC and for the centrality classes 0-20%, 20-40% and 40-60%. The dotted lines give the results for primary hard direct photons and photon fragmentation. As expected photons from fragmentation lead to a positive $v_2$ which is diluted by adding primary hard photons The solid lines are the results when bremsstrahlung, jet-photon conversion and thermal photons are included. They meet our expectations for $v_2$ of direct photons including all source discussed above. The $v_2$ for induced bremsstrahlung and jet-photon conversion is indeed negative. Together they are able to overcome the positive $v_2$ from fragmentation, leading to an overall negative elliptic asymmetry for direct photons at moderate $p_T$. Only above 8 GeV/$c$ the direct photon $v_2$ is again positive, because the yield of photons from fragmentation is dominating over medium induced bremsstrahlung [@TGJM:05] in this range. The dashed lines in Fig. \[fig:1\] show the $v_2$ for direct photons with no jet energy loss included. In this case, fragmentation photons do not exhibit an anisotropy and the elliptic asymmetry is only due to jet-photon conversion. Measurements of $v_2$ with sufficient accuracy could therefore constrain models for jet energy loss. The absolute size of $v_2$ is not large, about 2-3% for the 20-40% centrality bin around $p_T=4$ GeV/$c$ and up to 5% for the more peripheral bin. The reason is that the signal is diluted by isotropic photons (primary hard and thermal) and partially cancelled between the optical and inverse optical mechanisms. We also checked the dependence of $v_2$ on the temperature of the medium, by varying the initial temperature $T_i$ with $\tau_i T_i^3$ kept constant. The resulting changes are small: a change in $T_i$ of 40$\%$ generates a shift in $v_2$ of less than 20$\%$.
![\[fig:2\] (Color online) $v_2$ as a function of $p_T$ for Au+Au collisions at RHIC. The dashed line shows jet-fragmentation and induced bremsstrahlung only while the solid lines give jet-photon conversion, primary hard and thermal photons. The dotted line shows direct photons and the background from decay of neutral mesons coming from jets. The dot-dashed line adds photons from decay of recombined pions as well and can be compared to the inclusive photon $v_2$ measured by PHENIX [@Adler:2005rg].](v2_phot_alph34_RF_side.eps){width="\columnwidth"}
In Fig. \[fig:2\] we show some $v_2$ signals that should be detectable at RHIC in the near future. The dotted lines show $v_2$ for inclusive photons before background subtraction. In this case the $v_2$ signal is dominated by contributions from decaying $\pi^0$ and $\eta$ hadrons. The resulting $v_2$ is positive and larger in magnitude. Only hadrons from fragmentation have been included. However, it has been shown that hadron production up to a $p_T$ of 4 to 6 GeV/$c$ receives significant contributions from recombination of quarks [@FMNB:03]. The dot-dashed lines show the $v_2$ of inclusive photons if decays of $\pi^0$ and $\eta$ from recombination are included, using [@FMNB:03] with parameters consistent with our jet fragmentation calculation. Data on $v_2$ for inclusive photons have been measured by the PHENIX collaboration [@Adler:2005rg]. Our calculations for the total inclusive photons including decays from recombined hadrons agrees well with their results.
A very exciting option for the future is the possibility to experimentally distinguish between direct photons associated with jets and isolated direct photons. Photons from jet fragmentation and induced bremsstrahlung are in the former category, while the latter includes primary hard and thermal photons, and photons from jet-photon conversion. The dashed line in Fig. \[fig:2\] shows the result for the fragmentation and bremsstrahlung processes only. They contribute with different signs and one notes a characteristic change of sign from negative values at low $p_T$ to larger positive values, up to 5%, at large $p_T$, where fragmentation dominates. The solid line shows the $v_2$ of all isolated direct photon processes, including primary hard direct photon, thermal and jet-photon conversion. Jet-photon conversion is the only source of an anisotropy, so that the resulting $v_2$ is relatively large and negative.
We briefly mention here that the inverse optical mechanism for photon $v_2$ could be also at work for jets going through hadronic matter. Possible examples are hadronic processes after hadronization of the jet, e.g. $\rho+\pi \to \gamma+\pi$, as well as Compton, annihilation and bremsstrahlung processes with partons in surrounding hadrons before hadronization. In the first case the $p_T$ of the interacting jet hadron will be smaller than the original $p_T$ of the jet due to energy loss and hadronization, shifting the emitted photons to smaller $p_T$. In the second case the yield will ultimately depend on the unknown parton content of the hot hadron phase. However the Compton and annihilation yields have a $T^2$ dependence at intermediate $p_T$, hence any reasonable assumption of the parton content and the temperature of the hadronic phase will lead to small yields at intermediate $p_T$. The admixture of photon $v_2$ from jet-hadron interactions is therefore suppressed.
To summarize, we present the first calculation of the lowest order azimuthal asymmetry coefficient $v_2$ for direct photons in high energy nuclear collisions. Jets interacting with a deconfined quark gluon plasma provide photons exhibiting an inverse optical anisotropy with characteristic negative values of $v_2$. An experimental confirmation would emphasize the existence of a quark gluon plasma and confirm jet-medium interactions as important sources of photons at intermediate $p_T$. The $v_2$ signal is generally of order 3-5% and should be experimentally accessible at RHIC. Even more promising would be a separation of direct photons emitted in a jet from isolated photons. Both sources carry their own characteristic $p_T$ dependence for $v_2$. The arguments presented here for direct photons immediately apply to production of lepton pairs as well. Dileptons from annihilation of jets in the medium and from medium-induced virtual photon bremsstrahlung [@SGF:02] should also exhibit negative $v_2$.
We thank B. Cole and S. Jeon for stimulating discussions, and G. Moore for his help with details of the AMY energy loss. R. J. F. thanks the nuclear theory group at McGill University for their hospitality while part of this work was carried out. This work was supported in part by the Natural Sciences and Engineering Research Council of Canada, by le Fonds Nature et Technologies du Québec, and by DOE grant DE-FG02-87ER40328.
[99]{}
J. W. Harris and B. Müller, Ann. Rev. Nucl. Part. Sci. [**46**]{}, 71 (1996). E. L. Feinberg, Nuovo Cim. A [**34**]{}, 391 (1976). J. I. Kapusta, P. Lichard and D. Seibert, Phys. Rev. D [**44**]{}, 2774 (1991); [*Erratum ibid.*]{} D [**47**]{}, 4171 (1993); R. Baier, H. Nakkagawa, A. Niegawa and K. Redlich, Z. Phys. C [**53**]{}, 433 (1992); P. K. Roy, D. Pal, S. Sarkar, D. K. Srivastava and B. Sinha, Phys. Rev. C [**53**]{}, 2364 (1996); P. Aurenche, F. Gelis, R. Kobes and H. Zaraket, Phys. Rev. D [**58**]{}, 085003 (1998); P. Arnold, G. D. Moore and L. G. Yaffe, JHEP [**0112**]{}, 009 (2001). R. J. Fries, B. Müller and D. K. Srivastava, Phys. Rev. Lett. [**90**]{}, 132301 (2003). B. G. Zakharov, JETP Lett. [**80**]{}, 1 (2004); R. J. Fries, B. Müller and D. K. Srivastava, Phys. Rev. C [**72**]{}, 041902 (2005). C. Gale, T. C. Awes, R. J. Fries and D. K. Srivastava, J. Phys. G [**30**]{}, S1013 (2004).
S. Turbide, C. Gale, S. Jeon and G. D. Moore, Phys. Rev. C [**72**]{}, 014906 (2005).
L. Xiong, E. V. Shuryak and G. E. Brown, Phys. Rev. D [**46**]{}, 3798 (1992); V. V. Goloviznin and K. Redlich, Phys. Lett. B [**319**]{}, 520 (1993); C. Song, Phys. Rev. C [**47**]{}, 2861 (1993). S. Turbide, R. Rapp, and C. Gale, Phys. Rev. C [**69**]{}, 014903 (2004).
J. F. Owens, Rev. Mod. Phys. [**59**]{}, 465 (1987). S. S. Adler [*et al.*]{} (PHENIX Collaboration), Phys. Rev. Lett. [**94**]{}, 23201 (2005). J. Y. Ollitrault, Phys. Rev. D [**46**]{}, 229 (1992); S. Voloshin and Y. Zhang, Z. Phys. C [**70**]{}, 665 (1996). R. J. Fries, B. Müller, C. Nonaka and S. A. Bass, Phys. Rev. C [**68**]{}, 044902 (2003); C. Adler [*et al.*]{} (STAR Collaboration), Phys. Rev. Lett. [**90**]{}, 032301 (2003). S. S. Adler [*et al.*]{} (PHENIX Collaboration), Phys. Rev. Lett. [**91**]{}, 182301 (2003).
P. F. Kolb and U. Heinz, [*Preprint*]{} nucl-th/0305084.
C.Y. Wong, [*Introduction to High-Energy Heavy Ion Collisions*]{} (World Scientific, Singapore, 1994).
P. Arnold, G. D. Moore and L. Yaffe, JHEP [**0111**]{}, 057 (2001); JHEP [**0206**]{}, 030 (2002).
B.B. Back [*et al.*]{} (PHOBOS Collaboration), Phys. Rev. C [**65**]{}, 061901 (2002).
S. Turbide and C. Gale, proceeding for the Particles and Nuclei International Conference (PANIC05), Santa Fe, U.S, 24-28 October 2005, arXiv:hep-ph/0512200.
S. S. Adler [*et al.*]{} \[PHENIX Collaboration\], arXiv:nucl-ex/0508019.
D. K. Srivastava, C. Gale and R. J. Fries, Phys. Rev. C [**67**]{}, 034903 (2003); S. Turbide, C. Gale, D. K. Srivastava and R. J. Fries, arXiv:hep-ph/0601042.
S. S. Adler [*et al.*]{} (PHENIX Collaboration), Phys. Rev. D [**71**]{}, 071102 (2005).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
I analyze the density perturbations in a cosmological model with a scalar field coupled to ordinary matter, such as one obtains in string theory and in conformally transformed scalar-tensor theories. The spectrum of multipoles on the last scattering surface and the power spectrum at the present are compared with observations to derive bounds on the coupling constant and on the exponential potential slope. It is found that the acoustic peaks and the power spectrum are strongly sensitive to the model parameters. The models that best fit the galaxy spectrum and satisfy the cluster abundance normalization have field energy density $\Omega _{\phi
}\simeq 0.05-0.15$ and a scale factor expansion law $a\sim t^{p}$, $p\simeq
0.67-0.69.$
address: |
Osservatorio Astronomico di Roma,\
Viale Parco Mellini 84,\
00136 Roma, Italy\
[*[email protected]*]{}
author:
- Luca Amendola
title: Perturbations in a coupled scalar field cosmology
---
epsf.tex -.5in
Introduction
============
Perhaps the most important concept in modern cosmology is that fundamental physics, along with gravity, shapes the distribution of matter on very large scales. Fundamental physics enters in at least two distinct ways: through the potential of the inflationary field, which sets the initial conditions of the fluctuation field, and through the properties of the dark matter, which govern the evolution of the fluctuations up to the present. As a consequence, the imprint of the density fluctuations on the microwave background and on the galaxy distribution allows tests of basic laws of physics that, in many cases, could not be realized with any other mean.
An impressive array of different proposals have been formulated for as concerns the inflationary side of the story, that is, the initial conditions. So far, there is not an overwhelming reason to modify the simplest inflationary prescription of a flat spectrum, although several variations on the theme, like a small tilting (Lucchin & Matarrese 1985, Cen et al. 1992) or some break in the scale invariance (Gottloeber et al. 1994, Amendola et al. 1995) or the contribute of primordial voids (Amendola et al. 1996) cannot be excluded either.
Similarly, many theories have been proposed for as concerns the evolution of the fluctuations, trying to elucidate the nature and properties of the dark matter component. A partial list of the dark matter recipes includes the standard CDM and variations such as CDM plus a hot component (MDM), or plus a cosmological constant ($\Lambda $CDM), or plus a scalar field ($\phi $ CDM). The latter class of models, in particular, has been explored greatly in recent times, to various purposes. First, a light scalar field is predicted by many fundamental theories (string theory, pseudo-Nambu-Goldstone model, Brans-Dicke theory etc), so that it is natural to look at its cosmological consequences (Wetterich 1995, Frieman et al. 1995, Ferreira & Joyce 1998). Second, a scalar field may produce an effective cosmological constant, with the benefit that its dynamics can be linked to some underlying theory, or can help escape the strong constraints on a true cosmological constant (Coble, Dodelson & Frieman 1997, Waga & Miceli 1999). In turn, this effective cosmological constant can be tuned to explain the observation of an accelerated expansion (Perlmutter et al. 1998, Riess et al. 1998) and to fix the standard CDM spectrum as well (Zlatev et al. 1998, Caldwell et al. 1998, Perrotta & Baccigalupi 1999, Viana & Liddle 1998). Finally, even a small amount of scalar field density may give a detectable contribution to the standard CDM scenario, similar to what one has in the MDM model (Ferreira & Joyce 1998, hereinafter FJ).
In this paper we pursue the investigation of the effects of a scalar field in cosmology by adding a coupling between the field and ordinary matter. Such a coupling has been proposed and studied several times in the past (e.g. Ellis et al. 1989, Wetterich 1995) but, as far as we know, its consequences on the microwave background and on the power spectrum have not been determined. Scope of this paper is to solve the fluctuation equations for a coupled scalar field theory, and to compare with the already available data in the microwave sky and in galaxy surveys. We refer to this model as coupled $\phi $CDM. Up to a conformal transformation, the model we study is equivalent to a non-minimal coupling theory in which the scalar field couples to gravity like in a Brans-Dicke Lagrangian; the perturbations in such models have been investigated by Chen & Kamionkowsky (1999) and Baccigalupi et al. (1999) in a background in which the scalar field acts like a dynamical cosmological constant (see also Uzan 1999). The model we present here is in fact more general, since a wide class of non-minimal coupling models can be recast in the form we study below (Amendola et al. 1993, Wetterich 1995, Amendola 1999).
There are several models of cosmological scalar field in the literature, essentially characterized by the scalar field potential and by the initial conditions. We can divide the models in two broad class: in the first one, the field potential energy dominates at the present, so that it resembles a cosmological constant. In the second one, the field kinetic energy is not negligible, and the field adds to the ordinary matter as an additional component, like in MDM models. To this second class belongs the model of FJ. They adopt an exponential potential for the scalar field, able to drive an attractor scaling solution which self-adjusts to the dominant matter component. In such a model, the density fraction of the field does not depends on the initial conditions, but is determined by the potential parameters. Therefore, the coincidence that the energy density in the field and in the matter components are comparable can be explained by the underlying physics (the field potential) rather than by the initial conditions. Although the coupling we will introduce can be applied to any scalar field model, we focus our attention in this paper on the exponential potential model of FJ. Beside being particularly simple, because of its attractor properties (Wetterich 1988, Ratra & Peebles 1988), such a model is also easily falsifiable, because the effect of the scalar field is important at all times (not just at the present as when the field acts as a cosmological constant), and therefore induces a strong effect on the cosmological sky. As a consequence, the constraints we derive on the model parameters are rather strong.
The same exponential potential also allows solutions which belong to the first class mentioned above, in which the field acts much like a cosmological constant, and drives at the present an accelerated expansion. These solutions, and their linear perturbations, have been studied by Viana & Liddle (1998) and Caldwell et al. (1998). The effect of adding a coupling to these models will be analyzed in a subsequent work.
Coupled scalar field model
==========================
Consider two components, a scalar field $\phi $ and ordinary matter (e.g., baryons plus CDM) described by the energy-momentum tensors $T_{\mu \nu (\phi
)}$ and $T_{\mu \nu (m)}$. General covariance requires the conservation of their sum, so that it is possible to consider a coupling such that, for instance, $$\begin{aligned}
T_{\nu (\phi );\mu }^{\mu } &=&CT_{(m)}\phi _{;\mu }, \nonumber \\
T_{\nu (m);\mu }^{\mu } &=&-CT_{(m)}\phi _{;\mu }. \label{coup1}\end{aligned}$$ Such a coupling arises for instance in string theory, or after a conformal transformation of Brans-Dicke theory (Wetterich 1995, Amendola 1999). It has also been proposed to explain ’fifth-force’ experiments, since it corresponds to a new interaction that can compete with gravity and be material-dependent. The coupling arises from Lagrangian terms of the form (Wetterich 1995) $$-m_{\psi }^{2}\exp (-C\kappa \phi )\psi _{,\mu }\psi ^{,\mu },
\label{string}$$ where $\kappa ^{2}=8\pi G$ and $\psi $ is the ordinary matter field of mass $
m_{\psi }$, e.g., the nucleon field.
The specific coupling (\[coup1\]) is only one of the possible form. Non-linear couplings as $CT_{(m)}F(\phi )\phi _{;\mu }$ or more complicate functions are also possible. Also, one can think of different coupling to different matter species, for instance coupling the scalar field only to dark matter and not to baryons. Such a species-dependent coupling has been proposed by Damour, Gibbons & Gundlach (1990), and shown to be observationally viable. Notice that the coupling to radiation (subscript $
\gamma $) vanishes, since $T_{(\gamma )}=0.$ Here we restrict ourselves to the simplest possibility (\[coup1\]), which is also the same investigated earlier by Wetterich (1995) and is the kind of coupling that arises from Brans-Dicke models. For instance, a field with coupling to gravity $\frac{1}{
2}\xi \phi ^{2}R$ in the Lagrangian acquires, after conformal transformation, a coupling to matter of the form (\[coup1\]) with $$C=\frac{\kappa }{\left( 6+1/\xi \right) ^{1/2}};$$ in the limit of small positive coupling this becomes $$C=\kappa \sqrt{\xi }.$$
There are several constraints on the coupling constant $C$ along with constraints on the mass of the scalar field particles, reviewed by Ellis et al. (1989). The constraints arise from a variety of observations, ranging from Cavendish-type experiments, to primordial nucleosynthesis bounds, to stellar structure, etc etc. Most of them, however, apply only if the scalar couples to baryons, which is not necessarily the case, and/or involve the mass of the scalar field particles, which is unknown. The most stringent bound, quoted by Wetterich (1995) amounts to $$|C|<0.1M_{P}^{-1} \label{wetlimit}$$ but again holds only for a coupling to baryons. Moreover, these constraints are local both in space and time, and could be easily escaped by a time-dependent coupling constant. In the following we leave therefore $C$ as a free parameter.
The constraints from nucleosynthesis refer to the energy density in the scalar component. This has to be small enough not to perturb element production, so that at the epoch of nucleosynthesis (Wetterich 1995, Sarkar 1996, FJ) $$\Omega _{\phi }<0.1-0.2$$ We will see that this bound is satisfied by all the interesting models.
There is an immediate consequence of the coupling for as concerns cosmology. The coupling modifies the conservation equation for the ordinary matter, leading to a different effective equation of state for the matter. This alters the scale factor expansion law in matter dominated era (MDE) from $
a\sim t^{2/3}$ to $t^{p}$, $p\neq 2/3$ . In turn, this has three effects. First, the sound horizon at decoupling (when decoupling occurs in MDE) is modified with respect to the uncoupled case, being larger for $p>2/3$ and smaller in the opposite case, as we will show. This modifies the peak structure of the microwave background multipoles. Second, the epoch of matter/radiation equivalence moves to a later epoch if $p>2/3$ and to an earlier epoch in the opposite case. This shifts the range of scales for which there is the growth suppression of the sub-horizon modes in radiation dominated era (RDE), leading to a turnaround of the power spectrum on larger scales if $p>2/3$ (smaller if $p<2/3$). Finally, the different scale factor law modifies the fluctuation growth for sub-horizon modes in the MDE, generally reducing the growth for all values of $C$. Similar effects have been observed by Chen & Kamionkowsky (1999) and Baccigalupi et al. (1999) in Brans-Dicke models. The next sections investigate these effects in detail.
Background
==========
Here we derive the background equations in the conformal FRW metric $$ds^{2}=a^{2}(-d\tau ^{2}+\delta _{ij}dx^{i}dx^{j}).$$ The scalar field equation is $$\ddot{\phi}+2H\dot{\phi}+a^{2}U_{,\phi }=C\rho _{m}a^{2},$$ where $H=\dot{a}/a$ , and we adopt the exponential potential $$U(\phi )=Ae^{s\phi }.$$ The matter (subscript $m$) and the radiation (subscript $\gamma $) equations are $$\begin{aligned}
\dot{\rho}_{m}+3H\rho _{m} &=&-C\rho _{m}\dot{\phi} \\
\dot{\rho}_{\gamma }+4H\rho _{\gamma } &=&0.\end{aligned}$$ Denoting with $\tau _{0}$ the conformal time today, let us put $$a(\tau _{0})=1,\quad \rho _{m}(\tau _{0})=\frac{3H_{0}^{2}}{8\pi }\Omega
_{m}=\rho _{m0},\quad \rho _{\gamma }(\tau _{0})=\rho _{\gamma 0},\quad \phi
(\tau _{0})=\phi _{0}.$$ Without loss of generality, the scalar field can be rescaled by a constant quantity, by a suitable redefinition of the potential constant $A$. We put then $\phi _{0}=0$. This gives $$\begin{aligned}
\rho _{m} &=&\rho _{m0}a^{-3}e^{-C\phi }, \nonumber \\
\rho _{\gamma } &=&\rho _{\gamma 0}a^{-4}.\end{aligned}$$ The (0,0) Einstein equation can be written $$H^{2}=\frac{\kappa ^{2}}{3}\left( \frac{\rho _{m0}}{a}e^{-C\phi }+\frac{\rho
_{\gamma 0}}{a^{2}}+\frac{1}{2}\dot{\phi}^{2}+Ua^{2}\right) .$$
The dynamics of the model is very simple to study in the regime in which either matter or radiation dominates. Assume that the dominant component has equation of state $$p=(w-1)\rho .$$ Then, following Copeland et al. (1997) we define $$x=\frac{\kappa \dot{\phi}}{\sqrt{6}H},\quad y=\frac{\kappa \sqrt{U}}{\sqrt{3}
H},$$ and introduce the independent variable $\alpha =\log a$. Notice that $x^{2}$ and $y^{2}$ give the fraction of total energy density carried by the scalar field kinetic and potential energy, respectively . Then, we can rewrite the equations as (Amendola 1999) $$\begin{aligned}
x^{\prime } &=&-3x+3x\left[ x^{2}+\frac{1}{2}w(1-x^{2}-y^{2})\right] -\mu
y^{2}+\beta (1-x^{2}-y^{2}), \nonumber \\
y^{\prime } &=&\mu xy+3y\left[ x^{2}+\frac{1}{2}w(1-x^{2}-y^{2})\right] .
\label{syst2}\end{aligned}$$ where the prime denotes here $d/d\alpha $ and where we introduce the adimensional constants $$\beta =\sqrt{\frac{3}{2}}\frac{C}{\kappa },\quad \mu =\sqrt{\frac{3}{2}}
\frac{s}{\kappa },\quad$$ (in Amendola 1999 we defined $\beta $ one half of the definition above). Notice that in this simplified system with a single component, plus the scalar field, the constant $\beta $ is the coupling constant for the dominant component only, so that we are implicitly assuming $\beta =0$ during RDE. The system is invariant under the change of sign of $y$ and of $
\alpha $. Since it is also limited by the condition $\rho >0$ to the circle $
x^{2}+y^{2}\leq 1$, we limit the analysis only to the unitary semicircle of positive $y$. The critical points, i.e. the points that verify $x^{\prime
}=y^{\prime }=0$, are scaling solutions, on which the scalar field equation of state is $$w_{\phi }=\frac{2x^{2}}{x^{2}+y^{2}}=const,$$ the scalar field total energy density is $\Omega _{\phi }=x^{2}+y^{2}$, and the scale factor is $$a\sim \tau ^{\frac{p}{1-p}}=t^{p},\quad p=\frac{2}{3}\left[ \frac{1}{
w+\Omega _{\phi }(w_{\phi }-w)}\right]$$ ($t$ being the time defined by $dt=a(\tau )d\tau $).
The system (\[syst2\]) with an exponential potential has up to five critical points, that can be classified according to the dominant energy density: one dominated by the scalar field total energy density, one in which the fractions of energy density in the matter and in the field are both non-zero, one in which the matter field and the field kinetic energy are both non-zero, while the field potential energy vanishes, and finally two dominated by the kinetic energy of the scalar field (at $x=\pm 1$) . The critical points are listed in Tab. I, where we put $g(\beta ,w,\mu
)\equiv 4\beta ^{2}+4\beta \mu +18w.$ For any value of the parameters there is one and only one stable critical point (attractor). More details on the phase space dynamics in Amendola (1999) and, for $\beta=0$, in Copeland et al. (1997). $$\begin{tabular}{|c|c|c|c|c|c|}
\hline
& $x$ & $y$ & $\Omega _{\phi }$ & $p$ & $w_{\phi }$ \\ \hline
$a$ & $-\mu /3$ & $\left( 1-x_{a}^{2}\right) ^{1/2}$ & $1$ & $3/\mu ^{2}$ & $
2\mu ^{2}/9$ \\ \hline
$b$ & $-\frac{3w}{2\left( \mu +\beta \right) }$ & $-x_{b}\left( \frac{g}{
9w^{2}}-1\right) ^{1/2}$ & $\frac{g}{4\left( \beta +\mu \right) ^{2}}$ & $
\frac{2}{3w}\left( 1+\frac{\beta }{\mu }\right) $ & $\frac{18w^{2}}{g}$ \\
\hline
$c$ & $\frac{2}{3}\frac{\beta }{2-w}$ & $0$ & $\frac{4}{9}\left( \frac{\beta
}{2-w}\right) ^{2}$ & $\frac{6(2-w)}{4\beta ^{2}+9(2-w)w}$ & $2$ \\ \hline
$d$ & $-1$ & $0$ & $1$ & $1/3$ & $2$ \\ \hline
$e$ & $+1$ & $0$ & $1$ & $1/3$ & 2 \\ \hline
\multicolumn{6}{|c|}{Tab. I} \\ \hline
\end{tabular}$$
The perturbations on solutions converging toward the attractor $a$ have been studied in Viana & Liddle (1998) and in Caldwell et al. (1999) for zero coupling. In this case the scalar field is starting to dominate today, and mimics a cosmological constant. The case of interest here is instead the solution $b$ in Tab. I, since this is the only critical point which allow a partition of the energy between the scalar field and the matter and (contrary to $c$) is stable also in the RDE, when $\beta =0$. The solution $
b $ is compatible with a $p$ larger or smaller than $2/3$. It exists and is stable (that is, is an attractor) in the region delimited by $\mu <\mu _{-}$ and $\mu >\mu _{+}$ and the two branches of the curve $$\mu _{0}=-\frac{1}{4\beta }\left( 4\beta ^{2}+18w-9w^{2}\right) .$$ The scale factor slope on the attractor is (Wetterich 1995) $$p=\frac{2}{3w}\left( 1+\frac{\beta }{\mu }\right) .$$ and, if $w=1$, is inflationary for $$2\beta >\mu .$$ The parametric space region in which the attractor exists is shown in Fig. 1. For any value of the parameters $\beta ,\mu $ there is a pair of observables $\Omega _{\phi },p$. When radiation dominates, $\beta =0$, and the scale factor is the usual RDE one, $p=1/2$. The mapping from $\beta ,\mu
$ to $\Omega _{\phi },p$ is shown in the same Fig. 1: as one can see, to get a large $p$ a large $\Omega _{\phi }$ is also needed. In Fig. 2 we show the phase space of the system for $\Omega _{\phi }=0.1,p=0.7$ assuming matter domination. Notice that only for $\beta \neq 0$ there is the possibility to get an inflationary attractor with $\Omega _{\phi }<1$, as some observations suggest. It can be easily demonstrated that the coupled exponential potential with $2\beta >\mu $ is the only model that allows inflationary attractors with a non-vanishing matter component. Although such a possibility is intriguing, it is hardly realistic, since an inflationary expansion that lasted for most of the MDE would not allow any fluctuation growth via gravitational instability.
When both radiation and matter are present, the system goes rapidly from the radiation attractor, for which $$\quad \Omega _{\phi (R)}=6/\mu ^{2},\quad p_{R}=1/2,$$ to the matter attractor $$\quad \Omega _{\phi }=\frac{g}{4(\beta +\mu )^{2}},\quad p=\frac{2}{3}\left(
1+\frac{\beta }{\mu }\right) .$$ It is convenient to note that $\beta /\mu $ is a measure of the deviation from the uncoupled $p_{0}=2/3$ law in MDE: $$\frac{\beta }{\mu }=\frac{C}{s}=\frac{3p}{2}-1=\frac{p}{p_{0}}-1=\frac{
\delta \rho }{p}.$$ We give also the relation between the parameters $(\beta ,\mu )$ and the observables $\left( \Omega _{\phi },p\right) :$ $$\begin{aligned}
g &=&\frac{18p\Omega _{\phi }}{p_{0}-p(1-\Omega _{\phi })}, \nonumber \\
\beta &=&\frac{(g-18)}{2}\left( \frac{\Omega _{\phi }}{g}\right)
^{1/2},\quad \mu =\frac{1}{2}\left[ \left( \frac{g}{\Omega _{\phi }}\right)
^{1/2}-2\beta \right] .\end{aligned}$$ For small $\frac{\delta \rho }{p}$ we have $$C\simeq \kappa \sqrt{3}\Omega _{\phi }^{-1/2}\frac{\delta \rho }{p}.
\label{bound}$$
Since the slope and the matter content in the MDE depend on the model parameters, the equivalence epoch (subscript $e$) also depends on them. It is easy to see that the following relation holds $$a_{e}^{4-3p_{0}/p}=\frac{\rho _{\gamma 0}}{\rho _{m0}}.$$ Clearly, the equivalence occurs earlier with respect to the uncoupled case if $p<2/3$ (that is, $C/s<0$), later if $p>2/3$ (that is, $C/s>0$).
We will make often use of the fact that on the attractor in the RDE (subscript R) and in the MDE (subscript M) we have $$\phi =\alpha _{M,R}\log a,$$ where $$\alpha _{R}=-\frac{4}{s},\quad \alpha _{M}=-\frac{3}{s+C}.$$ Finally, it is useful to note that $$C\alpha _{M}=-\frac{3C}{s+C}\simeq -3\frac{\delta p}{p}$$ (the latter is valid for $\frac{\delta p}{p}\ll 1$).
Perturbations
=============
We now proceed to study the evolution of the perturbations in the coupled $
\phi $CDM theory. This involves the following tasks: 1) calculate the linear perturbation equations (we choose the synchronous gauge for the perturbed metric) for the coupled system of baryons (subscript $b$), CDM ($c$), radiation ($\gamma $), scalar field ($\phi $), massless neutrinos ($\nu $); 2) establish initial conditions (we adopt adiabatic initial conditions); 3) evolve the equations from deep into the radiation era and outside the horizon down to the present; 4) calculate the radiation fluctuations on the microwave background and the matter power spectrum at the present; 5) compare with observations.
Let us identify the effects of adding a scalar field to standard CDM. The field component clearly induces two main consequences for as concerns the perturbation equations: delays the epoch of equivalence, because the matter density at the present is smaller than without scalar field, and changes the perturbation equations. The first effect induces a turn-over of the power spectrum at larger scales, just as in the case of an open universe, or a model with a large cosmological constant, so that the power spectrum normalized to COBE has less power on small scales, as observed. The modification to the perturbation equations goes in the same direction: the evolution in the MDE for sub-horizon modes is suppressed with respect to standard CDM, as we will see below. The evolution equations in the other cases (super-horizon modes, RDE) give the same behavior as for the pure CDM . The net result is that FJ find that $\Omega _{\phi }\simeq 0.1$ gives a good fit to observations, comparable or superior to MDM or $\Lambda $CDM$.$
When we insert the coupling, the two effects above mentioned are again the dominant ones. But now, the consequences of the coupling can be in either directions, that is, the equivalence epoch can be delayed or anticipated, and the perturbations can be either suppressed or enhanced with respect to the uncoupled case, although not by a large factor. To understand this effects we first discuss analytically the perturbation equations. Following the discussion in FJ, we simplify the problem by reducing the system to three components, CDM, scalar field, and radiation. The notation is
$$\delta =\delta \rho /\rho ,\quad \varphi =\delta \phi ,\quad v_{i}=\delta
u_{i},\quad ik^{i}v_{i}=\theta .$$
where $u_{i}$ is the comoving velocity. The perturbation equations in synchronous gauge are:
Scalar field equation: $$\ddot{\varphi}+2H\dot{\varphi}+k^{2}\varphi +a^{2}U_{,\phi \phi }\varphi +
\frac{1}{2}\dot{h}\dot{\phi}=Ca^{2}\rho _{m}\Omega _{c}\delta _{c},$$ CDM: $$\begin{aligned}
\dot{\delta}_{c} &=&-\theta _{c}-\frac{1}{2}\dot{h}-C\dot{\varphi},
\label{decdm} \\
\dot{\theta}_{c} &=&-H\theta _{c}+C(k^{2}\varphi +\dot{\phi}\theta _{c}).\end{aligned}$$ Radiation: $$\begin{aligned}
\dot{\delta}_{\gamma } &=&-\frac{4}{3}\theta _{\gamma }-\frac{2}{3}\dot{h},
\\
\dot{\theta}_{\gamma } &=&\frac{k^{2}}{4}\delta _{\gamma }.\end{aligned}$$ Energy-momentum tensor: $$\begin{aligned}
a^{2}\delta T_{0}^{0} &=&a^{2}\left( \delta _{c}\rho _{c}+\delta _{\gamma
}\rho _{\gamma }\right) +\dot{\phi}\dot{\varphi}+a^{2}U_{,}\varphi \\
\frac{a^{2}}{k^{2}}ik^{i}\delta T_{i}^{0} &=&\frac{a^{2}}{k^{2}}w_{\gamma
}\theta _{\gamma }\rho _{\gamma }+\dot{\phi}\varphi \\
a^{2}\delta T_{i}^{i} &=&-a^{2}\delta _{\gamma }\rho _{\gamma }-3\left( \dot{
\phi}\dot{\varphi}-a^{2}U_{,}\varphi \right) .\end{aligned}$$ Metric: $$\begin{aligned}
H\dot{h} &=&2k^{2}\eta +8\pi a^{2}\delta T_{0}^{0} \\
\dot{\eta} &=&4\pi \frac{a^{2}}{k^{2}}ik^{i}\delta T_{i}^{0} \\
\ddot{h} &=&-H\dot{h}-8\pi a^{2}(\delta T_{0}^{0}-\delta T_{i}^{i}).
\label{hdotdot}\end{aligned}$$
Deriving Eq. (\[decdm\]) and inserting equation (\[hdotdot\]) we get $$\ddot{\delta}_{c}+H\dot{\delta}_{c}-\frac{3}{2}H^{2}(\Omega _{c}\delta
_{c}+2\Omega _{\gamma }\delta _{\gamma })-8\pi \left( 2\dot{\phi}\dot{\varphi
}-sa^{2}U\varphi \right) +C(H\dot{\varphi}+k^{2}\varphi +\ddot{\varphi}
-4H^{2}\Omega _{\gamma }\varphi )=0. \label{peq}$$ The equation for the scalar field becomes (putting $\theta _{c}=0$) $$\ddot{\varphi}+2H\dot{\varphi}+k^{2}\varphi +s^{2}a^{2}U\varphi -\dot{\delta}
_{c}\dot{\phi}=C\left( \frac{3H^{2}}{8\pi }\Omega _{c}\delta _{c}+\dot{\phi}
\dot{\varphi}\right) .$$ Finally, the radiation equation is $$\ddot{\delta}_{\gamma }+\frac{k^{2}}{3}\delta _{\gamma }-\frac{4}{3}\left(
\ddot{\delta}_{c}+\dot{\theta}_{c}+C\ddot{\varphi}\right) =0.$$ The adiabatic initial condition gives now, putting for the initial value of the scalar field $\varphi =\chi \delta _{c}$ ($\chi $ is determined below), $$\delta _{\gamma }=\frac{4}{3}\delta _{c}\left( 1+C\chi \right) .$$ In the large scale limit, $k^{2}\rightarrow 0$, and in RDE, where $H=\tau
^{-1}$ and $\Omega _{c}\rightarrow 0,$ and assuming the adiabatic condition, the system reduces to $$\begin{aligned}
\ddot{\delta}_{c}+\tau ^{-1}\dot{\delta}_{c}-4\tau ^{-2}\delta _{c}\Omega
_{\gamma }-8\pi \left( 2\dot{\phi}\dot{\varphi}-sa^{2}U\varphi \right)
+C(\tau ^{-1}\dot{\varphi}+\ddot{\varphi}-4\tau ^{-2}\Omega _{\gamma
}\varphi ) &=&0, \nonumber \\
\ddot{\varphi}+2\tau ^{-1}\dot{\varphi}+s^{2}a^{2}U\varphi -\dot{\delta}_{c}
\dot{\phi}-C\dot{\phi}\dot{\varphi} &=&0.\end{aligned}$$ Inserting the RDE attractor solution for $\phi $, we obtain that the growing mode both for $\delta $ and $\varphi $ goes as $\tau ^{2}.$ Therefore, the super-horizon perturbations in RDE grow similarly in CDM, in $\phi $CDM$,$ and in coupled $\phi $CDM. Moreover, we have that, initially, $$\varphi =-\frac{4}{5s}\delta _{c}\left( 1+\frac{4C}{5s}\right) ^{-1}\equiv
\chi \delta _{c}.$$ Therefore, the initial condition for the CDM density fluctuations on the attractor in the RDE is $$\delta _{c}=-\frac{1}{2}\frac{h}{1+C\chi }.$$
Now we consider the super-horizon modes in MDE. The equations are now $$\begin{aligned}
\ddot{\delta}_{c}+H\dot{\delta}_{c}-\frac{3}{2}H^{2}\Omega _{c}\delta
_{c}-8\pi \left( 2\dot{\phi}\dot{\varphi}-sa^{2}U\varphi \right) +C(H\dot{
\varphi}+\ddot{\varphi}) &=&0, \nonumber \\
\ddot{\varphi}+2H\dot{\varphi}+s^{2}a^{2}U\varphi -\dot{\delta}_{c}\dot{\phi}
-C\left( \frac{3H^{2}}{8\pi }\Omega _{c}\delta _{c}+\dot{\phi}\dot{\varphi}
\right) &=&0.\end{aligned}$$ The growing mode is again $\tau ^{2}$, that is, there is no difference with respect to the standard case.
In the sub-horizon regime, neglecting the gravitational feed-back, we have in RDE
$$\begin{aligned}
\ddot{\delta}_{c}+H\dot{\delta}_{c}-4H^{2}\Omega _{\gamma }\delta _{\gamma
}-8\pi \left( 2\dot{\phi}\dot{\varphi}-sa^{2}U\varphi \right) +C(H\dot{
\varphi}+k^{2}\varphi +\ddot{\varphi}-4H^{2}\Omega _{\gamma }\varphi ) &=&0
\label{peq2} \\
\ddot{\varphi}+H(2-C\alpha )\dot{\varphi}+k^{2}\varphi &=&0 \\
\ddot{\delta}_{\gamma }+\frac{k^{2}}{3}\delta _{\gamma } &=&0.\end{aligned}$$
The oscillating behavior of $\varphi $ and of $\delta _{\gamma }$ gives a negligible influence on $\delta _{c}$, so that $$\ddot{\delta}_{c}+H\dot{\delta}_{c}=0.$$ which gives $\delta _{c}=const,\log \tau $, once again with no difference with respect to standard CDM.
We finally come to the regime where the new physics makes the difference. In the sub-horizon MDE regime, neglecting again the gravitational feed-back, we have $$\begin{aligned}
\ddot{\delta}_{c}+H\dot{\delta}_{c}-\frac{3}{2}H^{2}\Omega _{c}\delta
_{c}-8\pi \left( 2\dot{\phi}\dot{\varphi}-sa^{2}U\varphi \right) +C\left[ -H
\dot{\varphi}+C\left( \frac{3H^{2}}{8\pi }\Omega _{c}\delta _{c}+\dot{\phi}
\dot{\varphi}\right) \right] &=&0 \\
\ddot{\varphi}+2H\dot{\varphi}+k^{2}\varphi -C\left( \frac{3H^{2}}{8\pi }
\Omega _{c}\delta _{c}+\dot{\phi}\dot{\varphi}\right) &=&0.\end{aligned}$$ Neglecting the oscillating behavior of $\varphi $ we obtain $$\ddot{\delta}_{c}+H(1+C\alpha _{M})\dot{\delta}_{c}-\frac{3}{2}H^{2}\Omega
_{c}\delta _{c}(1-\frac{C^{2}}{4\pi })=0. \label{basic}$$ Inserting the trial solution $\delta _{c}=Ba^{m}$ we obtain two solutions for $m$: [ ]{} $$m_{\pm }=\frac{1-p}{2p}\left\{ -1\pm \left[ 1+F(\Omega _{\phi },p)\right]
^{1/2}\right\} ,$$ where[ ]{} [ ]{} $$F(\Omega _{\phi },p)=\frac{6p(1-\Omega _{\phi })\left[ -8+26p+3(\Omega
_{\phi }-7)p^{2}\right] }{(p-1)^{2}\left[ 2+3p(\Omega _{\phi }-1)\right] }.
\label{mpm}$$ For[ ]{}$p=2/3$[ ]{}this reduces to the form found in FJ[ ]{} $$m_{\pm }=\frac{1}{4}\left( -1\pm \sqrt{25-24\Omega _{\phi }}\right) .$$ In Fig. 3 we show the contour plot of $m_{+}(\Omega _{\phi },p)$. This figure is crucial for the understanding of the perturbation evolution, so we discuss it at some length. First, we observe that for [*all*]{} values of $\Omega _{\phi },p$ there is suppression with respect to CDM: the slope is in fact always less than 1 and, for $p\simeq 2/3$, the slope is smaller for larger $\Omega
_{\phi }$. Second, we notice the unexpected fact that the value $p=2/3$ is close to the maximum for all values of $\Omega _{\phi }$, and closest for small $\Omega _{\phi }$. For $\Omega _{\phi }=0.1$, for instance, the maximum is at $p=0.672$, while for $\Omega _{\phi }=0.6$ it is at $p=0.710$. This implies immediately that the coupling does not enhance much the fluctuation growth with respect to the uncoupled case, while it can sensibly reduce it further as long as $p$ is far from $2/3.$ Third, there is only a finite range of $p$, almost centered around $2/3$, for which real values of $
m_{\pm }$ exist. Beyond that range, the power-law solutions of Eq. (\[basic\]) are replaced by oscillating solutions $\cos \left( \log a\right) $, in which the restoring force is the coupling interaction.
Let us then summarize the asymptotic evolution of the fluctuations in the coupled model. There are two relevant cases. If $p>2/3$, the equivalence epoch occurs later than in the uncoupled case. Then, smaller wavenumbers reenter during the RDE than in the uncoupled case, and therefore there is extra suppression at these scales. Then, in the subsequent MDE regime, the modes are further suppressed with respect to the uncoupled case, unless $p$ is close to $2/3$. The transfer function will be then more steeply declining with respect to the uncoupled case. If $p<2/3$, on the other hand, the equivalence occurs earlier, and the scales smaller than $2\pi \tau _{e}/a_{e}
$ are less suppressed. At the same time, the MDE regime induces again a slower fluctuation growth, so that there is an intermediate region of wavenumbers with a depleted transfer function, and a large wavenumber region with an enhanced transfer function. Fig. 4 displays some of these features.
The only important difference that arises when the baryons are added is in the tight coupling approximation. Referring to the notation used in Ma & Bertschinger (1995), we have the two equations $$\begin{aligned}
\dot{\theta}_{\gamma } &=&k^{2}\left( \frac{1}{4}\delta _{\gamma }-\sigma
_{\gamma }\right) +\frac{1}{\tau _{c}}\left( \theta _{b}-\theta _{\gamma
}\right) \label{the1} \\
\dot{\theta}_{b} &=&-H\theta _{b}+c_{s}^{2}k^{2}\delta _{b}-\frac{R}{\tau
_{c}}\left( \theta _{b}-\theta _{\gamma }\right) +C(k^{2}\varphi +\dot{\phi}
\theta _{b}). \label{the2}\end{aligned}$$ The slip equation $\dot{\theta}_{b}-\dot{\theta}_{\gamma }$ in the tight coupling approximation can be derived exactly as in Ma & Bertschinger (1995), taking into account that now (here $n_{e}$ is the electron density and $\sigma _{T}$ the Thomson cross section) $$\begin{aligned}
\tau _{c} &=&(an_{e}\sigma _{T})^{-1}\sim a^{-2}e^{C\phi },\quad \dot{\tau}
_{c}=\left( 2H+C\dot{\phi}\right) \tau _{c} \nonumber \\
R &=&\frac{4\rho _{\gamma }}{3\rho _{b}},\quad \dot{R}=\left( H-C\dot{\phi}
\right) R.\end{aligned}$$ To second order in $\tau _{c}$ we obtain that the slip between baryons and photons is $$\begin{aligned}
\dot{\theta}_{b}-\dot{\theta}_{\gamma } &=&\frac{2(H-C\dot{\phi})R}{1+R}
\left( \theta _{b}-\theta _{\gamma }\right) \nonumber \\
&&+\frac{\tau _{c}}{1+R}[-\frac{\ddot{a}}{a}\theta _{b}-\frac{1}{2}\left( H+
\frac{1}{2}C\dot{\phi}\right) k^{2}\delta _{\gamma }+k^{2}\left( c_{s}^{2}
\dot{\delta}_{b}-\frac{1}{4}\dot{\delta}_{\gamma }\right) + \nonumber \\
&&C\left( Hk^{2}\varphi +H\dot{\phi}\theta _{b}+k^{2}\dot{\varphi}+\ddot{\phi
}\theta _{b}\right) ].\end{aligned}$$ The equation for the photons is $$\dot{\theta}_{\gamma }=-R^{-1}\left[ \dot{\theta}_{b}+H\theta
_{b}-k^{2}c_{s}^{2}\delta _{b}-C(k^{2}\varphi +\dot{\phi}\theta _{b})\right]
+k^{2}\left( \frac{1}{4}\delta _{\gamma }-\sigma _{\gamma }\right) .$$
This concludes the analysis of the asymptotic regimes in the coupled $\phi $ CDM model. The results that will be presented in the next Sections make use of the full machinery of the Boltzmann code, as implemented in the CMBFAST code of Seljak and Zaldarriaga (1996), opportunely modified to take into account the coupled scalar field (including the transient from the RDE attractor to the MDE one). The equations are essentially the same as in FJ, with the new terms due to the coupling as detailed above. We tested the code with the results of FJ when $p=2/3$, and we also checked our results with the asymptotics found above.
Comparison with observations: cosmic microwave background
=========================================================
The main effect of the coupling on the cosmic microwave background is on the location and amplitude of the acoustic peaks. The location of the peak is related to the size of the sound horizon at decoupling (subscript $d$). Since the photon-baryon fluid has sound velocity $$c_{spb}^{2}=\frac{1}{3}r,\quad r\equiv \frac{R}{w_{c}^{\prime }+R},$$ where $w_{c}^{\prime }=1+C\alpha /3$, the sound horizon is $$r_{s}=\int_{0}^{\tau _{d}}\frac{d\tau }{\left( 3r\right) ^{1/2}}
=\int_{0}^{a_{d}}\frac{da}{\left( 3H^{2}a^{2}r\right) ^{1/2}}.$$ This expression can be simplified as follows. First, we put ourselves in the case $a_{e}\ll a_{d}\simeq 10^{-3}$ and neglect the RDE stage altogether. In MDE we have $$H^{2}a^{2}\simeq H_{0}^{2}ae^{-C\phi }.$$ Then we can write, remembering that on the attractor $e^{-C\phi
}=a^{-C\alpha },$ and defining the standard sound horizon $
r_{s0}=2a_{d}^{1/2}H_{0}^{-1}/\sqrt{3}$ $$r_{s}=\frac{r_{s0}}{a_{\ast }^{1/2}}\int_{0}^{a_{d}}\frac{da}{2\left(
a^{1-C\alpha }r\right) ^{1/2}}.$$ We can further simplify, for $r\simeq 1$, i.e. $R\gg 1$ (which is true at decoupling) $$r_{s}=r_{s0}\frac{a_{d}^{C\alpha /2}}{1+C\alpha },$$ and the corresponding peak multipole is, for $\delta p/p\ll 1$ $$\ell _{peak}\simeq \frac{2\pi }{r_{s}H_{0}}=\ell _{0}\left( 1+C\alpha
\right) a_{d}^{-C\alpha /2}=\ell _{0}\left( 1-3\frac{\delta p}{p}\right)
a_{d}^{1.5\delta p/p},$$ where the standard peak multipole is $$\ell _{0}=\frac{2\pi }{r_{s0}H_{0}}\simeq 200.$$ The qualitative behavior is clear: for $p<2/3$ there is a larger $\ell
_{peak}$ than in the uncoupled model, for $p>2/3$ a smaller $\ell _{peak}.$ For instance, for $p=0.65$ we expect $\ell _{peak}\simeq 250,$ in agreement with the numerical results.
We calculated the $C_{\ell }$ spectrum for several coupled $\phi $CDM models, parametrized by the two observables $\Omega _{\phi },p$. The range of values we explore, in this and in the next Section, is $$\Omega _{\phi }=0.05-0.2,\quad p=0.65-0.70.$$ The values of the other relevant parameters are fixed as follows $$h=0.7,\quad \Omega _{b}=0.04,\quad \Omega _{\Lambda }=0,\quad n=1.$$ In Fig. 5 we display the multipole spectra. As anticipated, the acoustic peaks move to larger multipoles as $p$ decreases.
There are two other effects worth discussing: the amplitude of the acoustic peaks and the slope of the multipole spectrum at small $\ell $. The amplitude of the peak is depressed as $\delta p/p$ increases, save for values close to 2/3, since the matter fluctuations that drive the radiation peaks are suppressed, as shown above. The small $\ell $ region is dominated by the Sachs-Wolfe (SW) effect. As well known, the integrated SW (ISW) effect in flat space vanishes only if the fluctuations grow as $a$, which is not the case here. The ISW then adds at small multipoles and tilts the $
C_{\ell }$ spectrum. Moreover, the overall normalization now takes into account the ISW power, and as a consequence the normalization for the perturbation at decoupling time is reduced. This effect shows also in the final amplitude of the power spectrum.
Deriving precise constraints from the whole set of observations on the CMB requires considerable detail in the statistical procedure, beyond the scope of this paper. Here we content ourselves to derive rough limits on the parameters. It is probably safe to state that current observations rule out values $p<0.63$ or $p>0.72$, although the present level of errors does not permit to attach a strong significance to such bounds. Future precision observations around the first peaks are likely to constrain $p$ to two decimal digits. As already found by FJ, on the other hand, the microwave sky does not impose strong constraints on $\Omega _{\phi }$, since this parameter influences mainly the fluctuation growth, and thus the absolute normalization. To constrain it, we have to evaluate the present power spectrum of the fluctuations.
Comparison with observations: power spectrum
============================================
The analytical expression (\[mpm\]) for the fluctuation growth exponent found in Section 3 is a clear guide to the results of this Section. As anticipated, the coupling introduces an extra suppression for the scales that enter the horizon in the MDE. The suppression is larger for models with high $\Omega _{\phi }$ and high $\mid \delta p/p\mid $ . A small suppression factor, as well known, helps to bring the standard CDM model into agreement with observations. FJ found that the best uncoupled models have $\Omega _{\phi }=0.1$; here we see that the coupling allows also models with smaller $\Omega _{\phi }$, but $\delta p/p\neq 0,$ to meet the observations. This can be helpful to reduce the constraints from nucleosynthesis, which, in some restrictive analysis, require $\Omega _{\phi
}<0.1$.
In Fig. 6 we report the power spectra $\Delta ^{2}(k)=k^{3}P(k)/(2\pi ^{2})$ normalized to COBE, compared to the data as compiled and corrected for redshift and non-linear distortions by Peacock & Dodds (1994). For a quantitative comparison, we plot in Fig. 7 the contour plot of $\sigma _{8}(\Omega _{\phi },p)$, the number density variance in 8 $Mpc$/h spheres. The models with a value of $\sigma _{8}$ larger than 0.5, as required by cluster abundance (White, Efstathiou & Frenk 1993, Viana & Liddle 1996, Girardi et al. 1998), have $\Omega _{\phi }<0.15$ and at most a small deviations from $
p=2/3$. The suppression of $\sigma _{8}$ with respect to COBE-normalized standard CDM is due both to the growth suppression in MDE and to the fact that now the COBE normalization includes the ISW effect.
For as concerns the shape of the spectrum, the comparison with the galaxy data is uncertain due to biasing. Assuming a scale-independent bias between matter and galaxies, we can quantify the agreement with the data by evaluating the $\chi ^{2}$ of the ratio between the theoretical and the galaxy spectrum, $R_{i}=P_{G}(k_{i})/P_{T}(k_{i}),$ that is by evaluating $$\chi ^{2}=\sum_{i}\left[ R_{i}-\hat{R}\right] ^{2}/\sigma ^{2}(k_{i}),$$ where $\hat{R},\sigma ^{2}(k)$ are the average and variance of $R_{i}$, neglecting cosmic variance. $\ $The contour plots of $\chi ^{2}(\Omega
_{\phi },p)$ are in Fig. 8. They show, as anticipated, that the models with $
p>2/3$ follow better the real data because are more bent at small scales. The best models among those studied here have $\chi ^{2}\simeq 13$ for $N=$ 15 degrees of freedom (16 real data, minus the average $\hat{R}$ estimated from the data themselves). For instance the model with $\Omega _{\phi
}=0.15,p=0.685$ gives a very good fit, and has $\sigma _{8}\simeq 0.6$ as required. Notice that we performed the fits without varying all the other cosmological parameters, which, at least in principle, can be determined by other observations.
In Fig. 9 we summarize the constraints obtained in this Section, considering the models which have $\sigma _{8}\in (0.45,0.75)$ and $\chi ^{2}/N<2$. The cluster abundance normalization usually quoted for $\Omega _{m}\simeq 1$ is $\sigma _{8}\simeq 0.60\pm 0.05$, but this is calculated for standard models, so that conservatively a larger region has been adopted. Only a stripe around $p=0.67-0.69$ and $\Omega _{\phi }=0.05-0.2$ appears to be viable, if the bias is indeed scale-independent. From Eq. (\[bound\]) we deduce a limit $$0<C\lesssim 1M_{p}^{-1}$$ which, although still far from the limit 0.1$M_{P}^{-1}$ quoted in Wetterich (1995) from local measures, is global and applies even if the scalar field is not coupled to baryons, as proposed in Damour, Gibbons & Gundlach (1990). Future data will certainly tighten the constraint even further.
Conclusion
==========
In this paper we discussed the perturbations of a coupled scalar field with exponential potential on the CMB and on the present large scale structure along an attractor solution. For as concerns the CMB, we found that the coupling induces a strong effect on the location and amplitude of the acoustic peaks, due to the variation of the scale factor expansion law. Future precision measures in the region $\ell >100$ have the potential of constraining the coupling to two orders of magnitudes better than at the present (see for instance the discussion in Chen & Kamionkowsky 1999).
We found that subhorizon perturbations are always more suppressed in MDE with respect to standard CDM, no matter what the parameters $\Omega _{\phi }$ and $p$ are. Moreover, the suppression increases for $p$ far from the standard value. The amplitude $\sigma _{8}$ at the present is between 0.5 and 1 only for $\Omega _{\phi }\in (0.05-0.2)$, being smaller for larger $
\Omega _{\phi }$, as already found in the uncoupled case by FJ. Adding the request to fit the galaxy power spectrum shape, the parametric space is reduced as in Fig. 9. A positive coupling has the advantage to warp the spectrum to a closer agreement with the data.
The background solution we adopted here is only one of the possible solutions. An equally interesting one is to consider a solution heading toward the inflationary attractor $a$, but still short of it. This would provide closure density to a $\Omega _{m}\simeq 0.3$ universe, and an acceleration as recently claimed, although at the price of sensibility to the initial conditions. Such a model will be investigated in a future work.
Acknowledgments {#acknowledgments .unnumbered}
===============
I am indebted to Carlo Baccigalupi, Francesca Perrotta and Michael Joyce for useful discussions on the topic.
L. Amendola, astro-ph/9904120 (1999), to be published in Phys. Rev. D60
L. Amendola, C. Baccigalupi, R. Konoplich, F. Occhionero & S. Rubin, Phys. Rev. D54, 7199 (1996)
L. Amendola, M. Litterio & F. Occhionero, Int. J. Mod. Phys. A5, 3861 (1990)
L. Amendola, S. Gottloeber, J. Mucket, V. Muller, Ap.J., 451, 444 (1995)
L. Amendola, D. Bellisai & F. Occhionero, Phys. Rev. D47, 4267 (1993)
C. Baccigalupi, F. Perrotta & S. Matarrese, astro-ph/9906066 (1999)
R.R. Caldwell, R. Dave, & P.J. Steinhardt, Phys. Rev. Lett. 80, 1582 (1998)
R. Cen, N. Gnedin, L. Kofman, J.P. Ostriker, Ap.J., 399, L11 (1992)
X. Chen & M. Kamionkowski, astro-ph/9905368 (1999)
K. Coble, S. Dodelson, J. Frieman, Phys. Rev. D55, 1851 (1997)
E. J. Copeland, A.R. Liddle & D. Wands, Phys. Rev. D57, 4686 (1997)
T. Damour, G.W. Gibbons, C. Gundlach, Phys. Rev. Lett., 64, 123 (1990)
J. Ellis, S. Kalara, K.A. Olive & C. Wetterich, Phys. Lett. B228, 264 (1989)
P. G. Ferreira & M. Joyce, Phys. Rev. D58, 2350 (1998)
J. Frieman, C. T. Hill, A. Stebbins, & I. Waga, Phys. Rev. Lett. 75, 2077 (1995)
M. Girardi, S. Borgani, G. Giuricin, F. Mardirossian, M. Mezzetti, Ap.J., 506, 45 (1998)
S. Gottloeber, J. Mucket & A. Starobinsky, Ap.J., 434, 417 (1994)
F. Lucchin & S. Matarrese, Phys. Rev. D32, 1316 (1985)
C.P. Ma & E. Bertschinger, Ap.J. 455, 7 (1995)
J. Peacock & S. Dodds, MNRAS, 267, 1020 (1994)
S. Perlmutter et al. Nature 391, 51 (1998)
F. Perrotta & C. Baccigalupi, astro-ph/9811156, to be published in Phys. Rev. D59
B. Ratra, & P.J.E. Peebles, Phys. Rev. D37, 3406 (1988)
A. G. Riess et al. astro-ph/9805201 (1998)
S. Sarkar, Rep. Prog. Phys. 59, 1493 (1996)
U. Seljak & M. Zaldarriaga, Ap.J., 469, 437 (1996)
J.-P. Uzan, astro-ph/9903004, to be published in Phys. Rev. D59
P. Viana & A. Liddle, Phys. Rev. D57, 674 (1998)
P. Viana & A. Liddle, MNRAS 281, 323 (1996)
I. Waga & A. Miceli, astro-ph/9811460
C. Wetterich, Nucl. Phys. B252, 309 (1985)
C. Wetterich, Nucl. Phys. B302, 668 (1988)
C. Wetterich, A&A, 301, 321 (1995)
S.D.M. White, G. Efstathiou, C.S. Frenk, MNRAS 262, 1023 (1993)
I. Zlatev, L. Wang & P. J. Steinhardt, astro-ph/9807002 (1998)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Opinion dynamics –the research field dealing with how people’s opinions form and evolve in a social context– traditionally uses agent-based models to validate the implications of sociological theories. These models encode the causal mechanism that drives the opinion formation process, and have the advantage of being easy to interpret. However, as they do not exploit the availability of data, their predictive power is limited. Moreover, parameter calibration and model selection are manual and difficult tasks.
In this work we propose an inference mechanism for fitting a generative, agent-like model of opinion dynamics to real-world social traces. Given a set of observables (e.g., actions and interactions between agents), our model can recover the most-likely latent opinion trajectories that are compatible with the assumptions about the process dynamics. This type of model retains the benefits of agent-based ones (i.e., causal interpretation), while adding the ability to perform model selection and hypothesis testing on real data.
We showcase our proposal by translating a classical agent-based model of opinion dynamics into its generative counterpart. We then design an inference algorithm based on online expectation maximization to learn the latent parameters of the model. Such algorithm can recover the latent opinion trajectories from traces generated by the classical agent-based model. In addition, it can identify the most likely set of macro parameters used to generate a data trace, thus allowing testing of sociological hypotheses. Finally, we apply our model to real-world data from Reddit to explore the long-standing question about the impact of the *backfire effect*. Our results suggest a low prominence of the effect in Reddit’s political conversation.
author:
- Corrado Monti
- Gianmarco De Francisci Morales
- Francesco Bonchi
bibliography:
- 'references.bib'
title: Learning Opinion Dynamics From Social Traces
---
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The one-dimensional random trap model with a power-law distribution of mean sojourn times exhibits a phenomenon of dynamical localization in the case where diffusion is anomalous: The probability to find two independent walkers at the same site, as given by the participation ratio, stays constant and high in a broad domain of intermediate times. This phenomenon is absent in dimensions two and higher. In finite lattices of all dimensions the participation ratio finally equilibrates to a different final value. We numerically investigate two-particle properties in a random trap model in one and in three dimensions, using a method based on spectral decomposition of the transition rate matrix. The method delivers a very effective computational scheme producing numerically exact results for the averages over thermal histories and initial conditions in a given landscape realization. Only a single averaging procedure over disorder realizations is necessary. The behavior of the participation ratio is compared to other measures of localization, as for example to the states’ gyration radius, according to which the dynamically localized states are extended. This means that although the particles are found at the same site with a high probability, the typical distance between them grows. Moreover the final equilibrium state is extended both with respect to its gyration radius and to its Lyapunov exponent. In addition, we show that the phenomenon of dynamical localization is only marginally connected with the spectrum of the transition rate matrix, and is dominated by the properties of its eigenfunctions which differ significantly in dimensions one and three.'
author:
- 'Franziska Flegel and Igor M. Sokolov'
title: Dynamical localization and eigenstate localization in trap models
---
Introduction
============
The random trap model together with its close relative, the barrier model, can be applied to a variety of physical problems [@Haus1987] which are related to random walks in disordered media, like properties of a photocurrent in amorphous solids [@Scher1975], or have an equivalent mathematical representation, like the behavior of resistor and capacitance networks [@Bouchaud1990] or lattice vibrations of harmonic chains [@Alexander1981; @Dyson1953]. It has also become prominent as a toy model for the phase space dynamics of spin glasses [@Bouchaud1992].
The random trap model on a regular lattice consists of a trapping landscape $\{ E_k \}$ of energy wells located on the lattice’s sites $k$. The $E_k$ are iid random variables which in our case have the common density $$\begin{aligned}
\rho_E (\epsilon) &= \frac{1}{kT_g} ~\exp\left(-\frac{\epsilon}{kT_g}\right),
\end{aligned}$$ with $T_g$ being the characteristic temperature. According to the Van’t Hoff-Arrhenius law [@Haenggi1990], at a temperature $T$ these trapping potentials correspond to *mean waiting times* $\tau_k = \tau_0 \exp\left( \frac{E_k}{kT} \right)$, where we put $\tau_0 = 1$ in what follows. The density of the mean waiting times $\tau_k$ is $$\begin{aligned}
\rho_\tau (t) &= \alpha t^{-1-\alpha}\qquad (t \geq 1),
\label{equ:rhotau}
\end{aligned}$$ with $\alpha = T/T_g$. In what follows, we consider lattices in dimensions $D = 1$ and $D = 3$. In 1D the particle’s dynamics on the lattice is described by a continuous-time Markov chain with transition rates $$\begin{aligned}
w_{k\rightarrow l} &= \begin{cases}
-1/\tau_k, &\text{if~} k = l,\\
1/(2D\tau_k), &\text{if~} k,l \text{~neighbors},\\
0, &\text{else},
\end{cases}
\label{equ:GeneralRates}
\end{aligned}$$ with $2D$ being the number of neighbors of site $k$ and $\tau_k$ denoting the mean waiting time associated with the site. If the first moment of the distribution, Eq.(\[equ:rhotau\]), diverges, which is the case for $0 < \alpha < 1$, the dynamics of the system starts to show highly nontrivial behavior (as exemplified by anomalous diffusion in such models). Of course, one could also define a mixture of the trap and barrier model, or generalize the model as in [@BenArous2005].
The model above describes the behavior of random walkers who, while exploring the lattice, get trapped in the potential wells. Once trapped at site $k$, they have to wait for a time $\tau$ (actual waiting time) with pdf $$\begin{aligned}
\psi_k (\tau) &= \frac{1}{\tau_k} \exp\left(-\frac{\tau}{\tau_k}\right),
\label{equ:exponentialpdf}
\end{aligned}$$ containing the mean waiting time $\tau_k$ at a site $k$ as parameter. Note that these local means $\tau_k$ are distributed according to Eq. (\[equ:rhotau\]) and the exponential waiting time density of Eq. (\[equ:exponentialpdf\]) is already conditioned on the fact that a random walker resides on a site with mean waiting time $\tau_k$. After the random walkers have waited long enough, they jump with equal probability to any of the recent site’s neighbors. Thus, the spatial properties of each single random walker’s trajectory are the same as in a simple random walk on the lattice. What is interesting, is the temporal evolution. The single-particle (one-time) quantities such as the mean squared displacement are well-understood, and can in high dimensions be described by the continuous-time random walk scheme; the behavior in 1D, which shares many properties with the CTRW behavior, is also well-investigated [@Bouchaud1990]. The properties of the motion of two or more random walkers exploring the same energy landscape are much less understood.
In what follows we concentrate on some of these multiparticle quantities, from which the (inverse) participation ratio $Y_2(t)$, which represents the probability that two random walkers who started at $t_0 = 0$ at the same site $k_0$ meet again at time $t$, was considered the most interesting. If the lattice is finite, the disorder-averaged long-time limit $\lim_{t \rightarrow \infty} \langle Y_2 (t)\rangle_\tau$ converges to an equilibrium value $\langle Y_2 \rangle_\tau^{(eq)} = 1-\alpha$ [@Bertin2002; @Derrida1997], independent of the lattice’s dimension $D$. The question about the non-equilibrium behavior of $\langle Y_2 \rangle_\tau$, however, is much more involved. It seems especially fascinating that in dimension one even for infinitely large lattices there exist a finite disorder-averaged long-time limit $\langle Y_2 \rangle_\tau^{(dyn)} >0$ with $\langle Y_2 \rangle_\tau^{(dyn)} \neq \langle Y_2 \rangle_\tau^{(eq)}$, the phenomenon called *dynamical localization*. This manifests itself also in finite but large one-dimensional lattices as a pronounced plateau in the temporal evolution of $\langle Y_2\rangle_\tau$ (see Fig. \[fig:Y2p0av\]). The existence of this regime has been mathematically proven by Fontes *et al.* [@Fontes2002] and was further investigated by Bertin and Bouchaud [@Bertin2002]. For lattices with dimension $D \geq 3$, Fontes *et al.* [@Fontes1999] proved that the dynamical localization is absent; it is conjectured that it is also absent in $D=2$. The phenomenon of dynamical localization is thus a property of one-dimensional lattices, i.e.: $$\begin{aligned}
\lim_{t\rightarrow \infty} \left( \lim_{N \rightarrow \infty} \langle Y_2\rangle_{p_0, \tau} \right) &= \begin{cases}
0, &\text{if~} D \geq 2,\\
\langle Y_2\rangle^{(dyn)}_\tau>0, &\text{if~} D = 1.
\end{cases}
\end{aligned}$$ The order of the limits is crucial since otherwise the expression converges to the equilibrium value $1-\alpha$.
The goal of this paper is to study the dynamical localization in somewhat more detail, and to connect it with the properties of the spectrum and of the eigenfunctions of the Laplacians defined by Eq. (\[equ:GeneralRates\]) in the one- and three-dimensional lattices with periodic boundary conditions. This is done by extensive numerical simulations relying on the spectral representation of the dynamics. The method allows for a numerically exact calculation of the system’s properties averaged over the initial position of the walkers and their thermal histories (trajectories) for a given realization of traps, and for considering longer observation times than the direct Monte-Carlo simulations. Apart from investigating the behavior of $Y_2$, we consider some other measures of localization, as well as the properties of the eigenfunctions of the corresponding Laplacians, trying to hint onto the main properties of the system responsible for dynamical localization.
![Temporal evolution of sample $\langle Y_2\rangle_{p_0}$ (dashed lines) and $\langle Y_2\rangle_{p_0, \tau}$ (dots) for $\alpha = 0.37$, $N = 11^3$ and the disorder average over 100 realizations. The light gray area gives the whole range of $\langle Y_2\rangle_{p_0}$ of all 100 realizations and the dark gray area depicts $\langle Y_2\rangle_{p_0,\tau} \pm \sigma$ with $\sigma$ being the standard deviation.[]{data-label="fig:Y2p0av"}](N1331a37real100Y2p0avWithSinglePaper){width="\linewidth"}
The Laplacian matrix
====================
Let us consider a 1D ring of $N$ sites and a realization $\{ \tau_k \}$ of waiting times according to the prescribed waiting time distribution $\rho_\tau$ of Eq. (\[equ:rhotau\]). Let ${
\mathbf{p}
}(t) \in [0,1]^N$ be the vector whose components $\pi_k$ describe the probability to find the random walker at site $k$ at time $t$. Then the participation ratio $Y_2$ is simply ${
\mathbf{p}
}^2 = {
\mathbf{p}
}^T {
\mathbf{p}
}$.
The temporal evolution of ${
\mathbf{p}
}$ is given by the matrix $W$ of the transition rates $w_{k\rightarrow l}$ which are explicitly defined by Eq. (\[equ:GeneralRates\]). All eigenvalues of $W$ are nonpositive. In what follows we will refer to the Laplacian matrix (Laplacian) $L = -W$, with $(L)_{lk}=-w_{k\rightarrow l}$ and nonnegative eigenvalues.
The master equation thus reads $\dot{{
\mathbf{p}
}} = -L{
\mathbf{p}
}$ and it leads to the temporal evolution of ${
\mathbf{p}
}$ in the form $$\begin{aligned}
{
\mathbf{p}
}(t) = \exp(-Lt) {
\mathbf{p}
}_0,\end{aligned}$$ with ${
\mathbf{p}
}_0$ being the initial condition. In our case, the initial distribution ${
\mathbf{p}
}_0$ will always be completely localized, i.e. ${
\mathbf{p}
}_0^2 = 1$.
If we average over all these localized initial conditions, then $Y_2 = {
\mathbf{p}
}_0^T \exp(-L^T t) \exp(-L t) {
\mathbf{p}
}_0$ becomes the trace $$\begin{aligned}
\langle Y_2\rangle_{p_0} &= \frac{1}{N} ~{
\mathrm{trace}\left\{ e^{-L^T t} e^{-L t} \right\}
}\label{equ:pTpSingularValue}\\
&= \frac{1}{N} \sum_{j=1}^N \sigma_j(t),\nonumber \end{aligned}$$ with $\sigma_j(t)$ being the $j$-th eigenvalue of the time-dependent operator $\mathcal{O}(t) = e^{-L^T t} e^{-L t}$. In what follows we will however use a more convenient representation and describe the process in the basis of the eigenfunctions of the Laplacian $L$ where the spectral decomposition is time-independent.
If $\mathcal{L}$ is the Laplacian for a simple random walk on a ring of size $N$ (i.e. a Toeplitz matrix with $1$ on the main diagonal, minus one half on the two adjacent diagonals and in the upper right and lower left corners, and all other elements equal to zero) and $S = \text{diag}\left(\frac{1}{\sqrt{\tau_1}}, \ldots, \frac{1}{\sqrt{\tau_N}}\right)$, the Laplacian $L$ is given by $$\begin{aligned}
L = \mathcal{L}S^2.\end{aligned}$$ The transpose of $L$ is $$\begin{aligned}
L^T &= S^2 \mathcal{L} = S^2 L S^{-2},\end{aligned}$$ and the matrix $$\begin{aligned}
A &= SLS^{-1}
\label{equ:A}\end{aligned}$$ is symmetric and hence diagonalizable over $\mathbb{R}$. This diagonalizability is inherited to $L$ and $L^T$. We will denote the eigenvectors of $L$ by ${
\mathbf{X}
}_i$, of $A$ by ${
\mathbf{Q}
}_i$ and of $L^T$ by ${
\mathbf{Z}
}_i$. Since $A$ is symmetric, we can choose the ${
\mathbf{Q}
}_i$ such that they form an orthonormal basis of $\mathbb{R}^N$ and define the eigenvectors of $L$ and $L^T$ such that for eigenvectors to the same eigenvalue $\lambda_i$ the following relation holds: $$\begin{aligned}
{
\mathbf{Z}
}_i &= S {
\mathbf{Q}
}_i = S^2 {
\mathbf{X}
}_i.\end{aligned}$$ From this follows: $$\begin{aligned}
\delta_{ij} &= {
\mathbf{Q}
}_i^T {
\mathbf{Q}
}_j = (S {
\mathbf{X}
}_i)^T (S^{-1} {
\mathbf{Z}
}_j) = {
\mathbf{X}
}_i^T {
\mathbf{Z}
}_j,
\label{equ:Orthogonality}\end{aligned}$$ where we have used the fact that $S$ is diagonal. Furthermore, there is an equilibrium state of $L$, namely the eigenvector ${
\mathbf{X}
}_1 = (\tau_1, \tau_2, \ldots , \tau_N)^T/\sqrt{\sum_k\tau_k}$ to eigenvalue zero, corresponding to the eigenvector ${
\mathbf{Z}
}_1 = (1, \ldots , 1)^T/\sqrt{\sum_k\tau_k}$ of $L^T$. All other eigenvalues are strictly positive since $A = S^T\mathcal{L}S$ with $S$ being invertible and $\mathcal{L}$ being (symmetric) positive-semidefinite with exactly one vanishing eigenvalue. The eigenvalues can be ordered such that $$\begin{aligned}
0 = \lambda_1 < \lambda_2 \leq \lambda_3 \ldots.\end{aligned}$$
Participation ratio {#sec:Y2Eig}
===================
We want to express the temporal evolution of $Y_2 = {
\mathbf{p}
}^2$ in terms of eigenvectors and eigenvalues of $L$ and $L^T$. It is easily verified that $$\begin{aligned}
{
\mathbf{p}
} = \sum_i \left( {
\mathbf{p}
}_0^T {
\mathbf{Z}
}_i \right) {
\mathbf{X}
}_i e^{-\lambda_i t},\end{aligned}$$ and thus $$\begin{aligned}
{
\mathbf{p}
}^T {
\mathbf{p}
} &= \sum_{ij} \left( {
\mathbf{Z}
}_i^T {
\mathbf{p}
}_0 {
\mathbf{p}
}_0^T {
\mathbf{Z}
}_j \right) \left( {
\mathbf{X}
}_i^T {
\mathbf{X}
}_j\right) e^{-(\lambda_i + \lambda_j)t}.
\label{equ:pTpInitial}\end{aligned}$$ If we average over all initial conditions ${
\mathbf{p}
}_0$ fulfilling ${
\mathbf{p}
}_0^2 = 1$, then $$\begin{aligned}
\langle {
\mathbf{p}
}_0 {
\mathbf{p}
}_0^T\rangle_{p_0} &= \frac{\mathbb{I}_N}{N},
\label{equ:p0average}\end{aligned}$$ with $\mathbb{I}_N$ the $N \times N$ identity matrix. Note, that we have not yet applied any disorder average. Averaging Eq. (\[equ:pTpInitial\]) over the initial conditions and applying Eq. (\[equ:p0average\]) we obtain $$\begin{aligned}
\langle Y_2\rangle_{p_0} &= \frac{1}{N} \sum_{ij} G_{ij} e^{-(\lambda_i + \lambda_j)t}.
\label{equ:pTpAveraged}\end{aligned}$$ where the elements $$\begin{aligned}
G_{ij} &= ({
\mathbf{Z}
}_i^T {
\mathbf{Z}
}_j)({
\mathbf{X}
}_i^T {
\mathbf{X}
}_j)\label{equ:DefG}{
\parbox[0pt][2em][c]{0cm}{}
}\\
&= ({
\mathbf{Q}
}_i^T S^2 {
\mathbf{Q}
}_j)({
\mathbf{Q}
}_i^T S^{-2} {
\mathbf{Q}
}_j){
\parbox[0pt][2em][c]{0cm}{}
}\nonumber, \end{aligned}$$ define a new matrix $G$ which is the element-wise product of the Gramian matrices corresponding to the vector sets $\{ {
\mathbf{Z}
}_1, {
\mathbf{Z}
}_2, \ldots, {
\mathbf{Z}
}_N \}$ and $\{ {
\mathbf{X}
}_1, {
\mathbf{X}
}_2, \ldots, {
\mathbf{X}
}_N \}$. Thus, the temporal development of $\langle Y_2\rangle_{p_0}$ depends on $G_{ij}$ in relation to the characteristic decay times $1/(\lambda_i + \lambda_j)$. As we proceed to show, these are the properties of $G$ which dominate the dynamical localization.
Taking the temporal limit of Eq. (\[equ:pTpAveraged\]) in a finite system and averaging over disorder, we can also reproduce the finite-size equilibrium value $\langle Y_2 \rangle^{(eq)}_\tau$ as it was given in [@Bertin2002; @Derrida1997]: $$\begin{aligned}
\left\langle \lim_{t\rightarrow \infty} \langle Y_2\rangle_{p_0} \right\rangle_\tau &= \frac{1}{N} \left\langle\left( {
\mathbf{Z}
}_1^T {
\mathbf{Z}
}_1 \right)
\left( {
\mathbf{X}
}_1^T {
\mathbf{X}
}_1\right)\right\rangle_\tau\nonumber\\
&= \left\langle\frac{1}{\sum_k \tau_k} \left( {
\mathbf{X}
}_1^T {
\mathbf{X}
}_1\right)\right\rangle_\tau\nonumber\\
&= \left\langle\frac{\sum_l \tau_l^2}{\left(\sum_k \tau_k\right)^2}\right\rangle_\tau\simeq 1-\alpha,\end{aligned}$$ see [@Sokolov2010].
Eq. (\[equ:pTpAveraged\]) comes in very handy because it enables us to determine $\langle Y_2\rangle_{p_0}$ at any time $t$ without previously computing everything that happened before $t$. Additionally, the temporal behavior of $\langle Y_2\rangle_{p_0}$ (which is still dependent on the disorder) can be simulated quite efficiently, i.e. without the statistical error attached to the average over initial conditions and over single trajectories. Thus the matrix approach gives a useful numerical tool of investigating the problem of dynamical localization, the tool which will be continuously used in the present paper. A nice consequence is that we are able to examine the temporal behavior of $\langle Y_2\rangle_{p_0}$ as averaged over initial conditions and trajectories in a fixed disorder realization and compare it to the disorder-averaged behavior to evaluate the inter-sample differences, as is done in Fig. \[fig:Y2p0av\].
Fig. \[fig:Y2p0av\] shows the behavior of $\langle Y_2\rangle_{p_0}$ in three different realizations of the random potential together with the result $\langle Y_2 \rangle_{p_0, \tau}$ as following from averaging over 100 such realizations. The pronounced plateau of $\langle Y_2 \rangle_{p_0, \tau}$ between $t \simeq 10^2$ and $10^9$ corresponds to the dynamical localization and up to now was only considered after applying the disorder average. We see that up to the crossover at $t \approx 10^8$ the $\langle Y_2 \rangle_{p_0}$ do not differ strongly from each other, so that the hints onto the dynamical localization are present in each single realization of the random traps. Especially at the onset of the plateau at $t \approx 10^2$ all different $\langle Y_2 \rangle_{p_0}$ appear to follow the same pattern, independent on what equilibrium value of $\langle Y_2\rangle_\tau$ would be attained. Fig. \[fig:Y2p0av\] shows that, even though the temporal evolution of the sample $\langle Y_2\rangle_{p_0}$ fluctuates around the disorder average $\langle Y_2\rangle_{p_0, \tau}$, the effect of the dynamical localization is not an exclusive property of $\langle Y_2\rangle_{p_0, \tau}$ but does already appear in the average over initial conditions and thermal histories.
The analytical calculation of $Y_2$ in the non-equilibrium case pertinent to dynamical localization turned out to be quite difficult. Bertin and Bouchaud [@Bertin2002] conjectured $$\begin{aligned}
\langle Y_2 \rangle_\tau^{(dyn)} &= \frac{2}{3}~\langle Y_2 \rangle_\tau^{(eq)},
\label{equ:BBconjecture}\end{aligned}$$ using a simple approximation and presented simulations with infinite-time extrapolation that confirmed their conjecture. We find it worth mentioning that according to their simulations [@Bertin2002 Fig. 5] as well as in our attempts to reproduce their figure, the simulated data tend to be smaller than Eq. (\[equ:BBconjecture\]) predicts, whereas it is more likely that the fitting procedures overestimate the infinite-time value $\langle Y_2\rangle^{(dyn)}_{p_0, \tau}$. Nevertheless, Eq. (\[equ:BBconjecture\]) gives a simple and elegant approximation.
Participation ratio and other localization measures {#sec:dynloc}
===================================================
What does dynamical localization (not) mean? {#subsec:locmeasures}
--------------------------------------------
The localization of states was extensively studied for vibrational excitations and in quantum models, where different localization criteria were applied mostly to the eigenstates of the corresponding Hamiltonians; in a typical case all of them lead to the same kind of conclusions on whether the corresponding state is localized or not. Thus, the participation ratio quantifies to what extent the localized state is *concentrated* on a small subset of all sites. The standard criterion for localization which is both theoretically transparent and easy for numerical implementation, especially in one dimension, is based on the evaluation of the Lyapunov exponent $\gamma$ which corresponds to the rate of exponential growth of the amplitude with the coordinate [@Ishii1973; @Comtet2013; @Thouless1972], and determines the behavior of the state outside of its localization region. We will consider the behavior of the Lyapunov exponent later on in this paper.
One can also quantify the localization by the radius of gyration $R_G$ of the corresponding state, which indicates the spatial *spread* of the state, the size of the spatial domain which the particle or excitation is confined to.
![Equilibrium value $\langle R_G^2\rangle_\tau^{(eq)}$ of the radius of gyration in dependence of the system size. In agreement with Eq. (\[equ:RG2eq\]), the simulation results (dots) follow $\langle R_G^2\rangle_\tau^{(eq)} \simeq \alpha N^2/12$ (dashed line).[]{data-label="fig:RG2ofN"}](RadiusofGyration2){width="\linewidth"}
It is defined as $$\begin{aligned}
R_G^2 &= \sum_{kl} d(k,l)^2 p_k p_l,\label{equ:RG}\end{aligned}$$ with $d(k,l)$ denoting the distance between sites $k$ and $l$. Thus, the gyration radius gives the mean squared distance between the two diffusion particles. In a linear lattice with closed boundary conditions $d(k,l) = |k-l|$. In the case of a ring of size $N$ the distance is given by $$\begin{aligned}
d(k,l) &= \begin{cases}
|k-l|, &\text{if~} |k-l| \leq N/2\\
N-|k-l|, &\text{else.}
\end{cases}\end{aligned}$$
We say a state is localized if the radius of gyration is small. With the definition of the matrix $R \in \mathbb{R}^{N \times N}$ with the components $R_{kl} = d(k,l)^2$ we find: $$\begin{aligned}
R_G^2 &= {
\mathbf{p}
}^T R {
\mathbf{p}
}.\end{aligned}$$
![Temporal evolution of $\langle Y_2\rangle_{p_0, \tau}$ () in comparison to $1/\langle R_G^2\rangle_{p_0, \tau}$ ([**+**]{}), calculated as average over 100 landscape realizations of size $N = 11^3$ and $\alpha = 0.37$. []{data-label="fig:RG2temporal"}](N1331Y2p0avGyration2){width="\linewidth"}
This has the same structure as $Y_2 = {
\mathbf{p}
}^T \mathbb{I} ~{
\mathbf{p}
}$ with $\mathbb{I}$ being the identity matrix. But, whereas the bilinear form $\mathbb{I}$ is a diagonal matrix, $R$ has vanishing diagonal entries and is dominated by non-diagonal ones. In analogy to the calculations that led to Eq. (\[equ:pTpAveraged\]), we find: $$\begin{aligned}
\langle R_G^2 \rangle_{p_0} &= \frac{1}{N} \sum_{ij} \left( {
\mathbf{Z}
}_i^T {
\mathbf{Z}
}_j \right) \left( {
\mathbf{X}
}_i^T R {
\mathbf{X}
}_j\right) e^{-(\lambda_i + \lambda_j)t}.
\label{equ:RG2Averaged}\end{aligned}$$ The equilibrium value in disorder average can be computed easily: $$\begin{aligned}
\langle R_G^2 \rangle_\tau^{(eq)} &= \sum_{kl} d(k,l)^2 \langle p_k p_l \rangle_\tau^{(eq)}\\
&= \langle p_k p_l \rangle_{\tau,k \neq l}^{(eq)} \sum_{kl} d(k,l)^2\end{aligned}$$ The two factors can now be evaluated independently: $$\begin{aligned}
&\left\langle p_k p_l \right\rangle_{\tau, k\neq l}^{(eq)} = \frac{1}{2} \left\langle \left( p_k + p_l\right)^2 - p_k^2 - p_l^2 \right\rangle_{\tau, k\neq l}^{(eq)}\\
&= \frac{1}{2}\left\langle \left( p_k + p_l\right)^2\right\rangle_{\tau, k\neq l}^{(eq)} - \left\langle p^2 \right\rangle_{\tau}^{(eq)}
~=~ \frac{\alpha}{N^2},\end{aligned}$$ where we have used the results of [@Sokolov2010]. The evaluation of $\sum_{k,l=1}^N d(k,l)^2$ yields $$\begin{aligned}
\sum_{k,l=1}^N d(k,l)^2 \simeq \frac{N^4}{12},\end{aligned}$$ so that $$\begin{aligned}
\langle R_G^2 \rangle_\tau^{(eq)} &\simeq \frac{\alpha N^2}{12}
\label{equ:RG2eq}\end{aligned}$$ Fig. \[fig:RG2ofN\] shows the results of simulations of Eq. (\[equ:RG2Averaged\]) in the limit $t \rightarrow \infty$ for different $\alpha < 1$ and system sizes $N$ and the comparison to Eq. (\[equ:RG2eq\]).
The important information is that the equilibrium value of the radius of gyration grows with the system size whereas the equilibrium value of $\langle Y_2 \rangle_\tau$ becomes independent of $N$ (for $N$ large enough), i.e. the states localized according to the $Y_2$ criterion are delocalized with respect to their gyration radius.
The reason is quite clear. If a site attracts a considerable amount of probability in the equilibrium state ${
\mathbf{p}
}^{(eq)} = (\tau_1, \tau_2, \ldots , \tau_N)^T/\sum_k\tau_k$, let us call it a “deep trap”. If there is only one deep trap in the landscape, $Y_2$ is large, and $R_G$ is small. Many landscape realizations do contain *more than one* deep trap, in which case $Y_2$ is still large, but $R_G$ might be large as well, being of the size of the distance between the deep traps. The expectation of the distance between the deep traps grows with the system size and thus the radius of gyration does also grow, whereas the participation ratio is independent of the deep traps’ positions.
Apart from the equilibrium value, we can also compare the temporal evolution of $1/\langle R_G^2\rangle_{p_0, \tau}$ to the one of $\langle Y_2\rangle_{p_0, \tau}$. Fig. \[fig:RG2temporal\] depicts such a comparison with numerical averages over 100 landscape realizations.
Altogether, the phenomenon of dynamical localization is pertinent to the behavior of one, very specific localization measure. In the range of times when the dynamical localization is observed, the probability that the two independent random walkers starting at the same site appear to be at the same site at a later instant of time $t$ (as represented by $Y_2$) stagnates, but the mean squared distance between them (given by $R_G$) grows continuously. $Y_2$ is thus a measure of the concentration of the state on some small subset of sites; $R_G$, on the other hand, quantifies the size of the spatial domain in which the particles are confined. Both the dynamically localized state and the equilibrium state of the system are essentially delocalized with respect to this measure.
Localization of the eigenstates
-------------------------------
![The Lyapunov exponent $\gamma$ vs. inverse radius of gyration $\sqrt{\sigma^{-2}}$ (see Eq. (\[equ:sigma-2\])) on double logarithmic scales. The shaded area gives the density of single data points, whereas the dots represent averaged values. The larger circles define the ensemble for a linear fit (solid line), yielding $\log \gamma = 0.440 + 0.434 \cdot \log \sqrt{\sigma^{-2}}$. The dashed line represents the numerical zero $\log(2N)/N$.[]{data-label="fig:gammaOfsigma"}](GammaOfSigma){width="\linewidth"}
The equilibrium state is essentially an eigenvector of $L$ to its zero eigenvalue; therefore it is interesting to look at the eigenvalue localization properties of $L$ also for other eigenvalues $\lambda$. The corresponding measure of localization is given by the Lyapunov exponent $\gamma$ which is defined as follows: Let $\pi_k$ be the components of an eigenvector of $L$ to eigenvalue $\lambda$. Then the following relation holds: $$\begin{aligned}
\begin{pmatrix}
\frac{\pi_k}{\tau_k}{
\parbox[0pt][2em][c]{0cm}{}
}\\ \frac{\pi_{k+1}}{\tau_{k+1}}{
\parbox[0pt][2em][c]{0cm}{}
}
\end{pmatrix}
&= \underbrace{\begin{pmatrix}
0 & 1\\ -1 & 2 - \lambda \tau_k
\end{pmatrix}}_{=~T_k}
\begin{pmatrix}
\frac{\pi_{k-1}}{\tau_{k-1}}{
\parbox[0pt][2em][c]{0cm}{}
}\\ \frac{\pi_k}{\tau_k}{
\parbox[0pt][2em][c]{0cm}{}
}
\end{pmatrix}.
\label{equ:Tk2}\end{aligned}$$ The matrix $T_k$ which transfers between the vector on the r.h.s. and the vector on the l.h.s. is called the $k$th transfer matrix. The Lyapunov exponent is now defined as $$\begin{aligned}
\gamma = \lim_{N \rightarrow \infty} \frac{1}{N} \left\langle \log \| T_N \ldots T_2 T_1 \|_1\right\rangle_\tau,
\label{equ:Defgamma}\end{aligned}$$ with $\| M \|_1 = \max_{j = 1,2} \left( |m_{1j}| + |m_{2j}| \right)$ denoting the maximum absolute column sum of the matrix $M$. Eq. (\[equ:Defgamma\]) together with Eq. (\[equ:Tk2\]) defines $\gamma(\lambda)$ regardless of the fact whether $\lambda$ is an eigenvalue of $L$ or not. (Note that the vectors in Eq. (\[equ:Tk2\]) are actually built from the components of the eigenvectors of $L^T$. In the limit $N \rightarrow \infty$, however, it can be shown that the Lyapunov exponent $\gamma(\lambda)$ of both $L$ and $L^T$ is the same.)
As the Lyapunov exponent gives the rate of exponential growth in an (eigen-)vector, it is a standard measure for localization: $\gamma = 0$ means that the state is delocalized whereas a high Lyapunov exponent means strong localization. Considering this, it seems peculiar to us that the equilibrium state of the trap model should have a high value of participation ratio but a vanishing Lyapunov exponent $\gamma$:
![Participation ratio as a function of time for 1D and 3D systems. The parameters are: $N = 11^3$, $\alpha = 0.37$, the number of realizations is 100.[]{data-label="fig:Y21D3D"}](N1331a37real100Y2p0av1D3D){width="\linewidth"}
$$\begin{aligned}
\gamma(0) &= \lim_{N \rightarrow \infty} \frac{1}{N} \left\langle \log \left\| \begin{pmatrix}
0 & 1\\ -1 & 2
\end{pmatrix}^N \right\|_1\right\rangle_\tau\\
&= \lim_{N \rightarrow \infty} \frac{1}{N} \log \left\| \begin{pmatrix}
1-N & N\\ -N & N+1
\end{pmatrix}\right\|_1\\
&= 0.\end{aligned}$$
Thus, the equilibrium state is essentially delocalized with respect to the its Lyapunov exponent. Since in all our simulations we consider finite systems, the measured Lyapunov exponent at $\lambda = 0$ is of order $\log(2N)/N$.
A double-logarithmic plot of the Lyapunov exponent for $N = 1331$ and $\alpha \in \{ 0.5, 0.75, 1.25, 2 \}$ is shown in Fig. \[fig:gammaOfECloseUp\]. For all four values of $\alpha$ the Lyapunov exponent $\gamma(0)$ is of order $\log(2N)/N$ (dashed line) which corresponds to zero in the infinite system size limit.
{width=".48\linewidth"} {width=".48\linewidth"}
The Lyapunov criterion of localization gives information which is strongly correlated with the one delivered by the gyration radius of single eigenstates. The inverse squared gyration radius for an eigenstate can be defined via $$\begin{aligned}
\sigma_k^{-2} = \left( \frac{\sum_{i,j=1}^N d(i,j)^2 X_{ki}^2 X_{kj}^2}{\sqrt{\sum_{j=1}^N X_{kj}^4}}\right)^{-1}
\label{equ:sigma-2},\end{aligned}$$ and $X_{kj}$ being the $j$th entry of the $k$th eigenvector. In Fig. \[fig:gammaOfsigma\] we plot $\gamma_k$ vs. $\sqrt{\sigma_k^{-2}}$ on double logarithmic scales. Strong correlation (linear on double logarithmic scales, i.e. a power-law dependence) between both is seen. A linear fit shows that $\log \gamma = 0.440 + 0.434 \cdot \log \sqrt{\sigma^{-2}}$. This means that the Lyapunov exponent and the gyration radius provide related information on the eigenstate localization: the eigenstates confined to a smaller region typically decay faster outside of it, and vice versa.
Spectral properties and dynamical localization {#subsec:spectrum}
----------------------------------------------
Now we turn to another important question, namely where exactly the information on the dynamical localization is hidden: Is it determined by the spectral properties of the Laplacian matrix, or is it mostly coded in the entries of the matrix $G$ composed from the eigenvectors of $L$? To understand this we compare the behavior of a one-dimensional system of $N$ sites with the one of its three-dimensional analogue, where dynamical localization is absent [@Fontes1999]. The total number of sites shall also be $N$, i.e. the side length of the cube is $N^{1/3}$. The rates for the 3D case are as follows: $$\begin{aligned}
w_{k\rightarrow l}^{(3)} &= \begin{cases}
-1/\tau_k, &\text{if~} k = l,\\
1/(6\tau_k), &\text{if~} k,l \text{~neighbors},\\
0, &\text{else}.
\end{cases}\end{aligned}$$ A comparison between the disorder averaged evolution of $Y_2$ on the one- and three-dimensional case is shown in Fig. \[fig:Y21D3D\]. As Fontes *et al.* [@Fontes1999] proved, there is no dynamical localization in the 3D case, i.e. $\lim_{t\rightarrow \infty} \lim_{N \rightarrow \infty} \langle Y_2\rangle_\tau = 0$, which, for finite lattices with site number $N$, can be understood as $\lim_{N \rightarrow \infty} \min_t \langle Y_2\rangle_\tau = 0$. In Fig. \[fig:Y21D3D\] this can be seen by the fact that $\min_t\langle Y_2^{3D}\rangle_\tau$ is of the order of $N^{-1}$ and $\min_t\langle Y_2^{1D}\rangle_\tau$ is considerably larger although in both cases the number of sites is $N = 11^3$.
In Fig. \[fig:spectrum\] we plotted the spectrum $\rho_\lambda$ for both Laplacians $L$ and $L^{(3D)}$ on linear and on double-logarithmic scales. Note that the cusp at $\lambda = 1$ is due to the fact that we use pure Pareto-distributed waiting times and it does not show up for analytic distributions as for example a one-sided Lévy-distribution with corresponding asymptotics. In any case, the behavior of $\rho_\lambda$ for large values of $\lambda$ does not seem interesting to us because we are looking for hints for the long-time behavior of the system which means we are concerned about small $\lambda$. The double-logarithmic plot of Fig. \[fig:spectrum\] reveals that for small values of $\lambda$, the spectrum $\rho_\lambda$ exhibits a power-law behavior. The exponent in the 3D case is in fact equal to $\alpha - 1$, as the dashed line in the double-logarithmic plot indicates. This is the exponent Bovier and Faggionato [@Bovier2005] determined for infinite-dimensional models. Note, that the dashed line in Fig. \[fig:spectrum\] is not a fit. The absolute value of the exponent for the 1D case is slightly larger than in 3D.
![Approximation of Eqs. (\[equ:Factorized\]) and (\[equ:1Dwith3Deigenvalues\]) for $N = 11^3$, $\alpha = 0.37$ and 100 landscape realizations. The exact result for the 1D case is given by the solid line and the approximation of Eq. (\[equ:Factorized\]) is depicted by circles, the 3D case by plus signs. The approximation of Eq. (\[equ:1Dwith3Deigenvalues\]) is represented by squares.[]{data-label="fig:Factorized"}](N1331a37real100Y2p0avFactorized){width="\linewidth"}
Nevertheless, the qualitative behavior of both the spectrum of $L$ and $L^{(3D)}$ does not differ strongly.
The question is, whether the different exponents of $\rho_\lambda$ do contribute to the effect of dynamical localization, or whether they are only an expression of the different time scales of the corresponding random walks. In order to investigate this, as well as the interplay between the eigenvectors and eigenvalues, we test the following decoupling approximation of Eq. (\[equ:pTpAveraged\]): $$\begin{aligned}
\langle Y_2\rangle_{p_0, \tau} &= \sum_{ij} \left\langle G_{ij} e^{-(\lambda_i + \lambda_j)t} \right\rangle_\tau\nonumber\\
&\simeq \sum_{ij} \left\langle G_{ij}\right\rangle_\tau \left(e^{-\tilde \lambda_i}\right)^t \left(e^{-\tilde \lambda_j}\right)^t = \langle \tilde{Y}_2\rangle_{p_0, \tau}.
\label{equ:Factorized}\end{aligned}$$ with $e^{-\tilde \lambda_i} = \langle e^{-\lambda_i} \rangle_\tau$. The result of this approximation is depicted in Fig. \[fig:Factorized\]. The figure shows that the decoupling approximation adequately reproduces the results of direct simulations and that the correlations between eigenvalues and the entries of $G$ do not actually play a role.
![First $20 \times 20$ entries of the matrix $G$ defined by Eq. (\[equ:DefG\]). $N = 1331$, $\alpha = 0.37$, 100 realizations. Gray bars represent positive entries and white bars negative ones.[]{data-label="fig:GAveraged"}](G1 "fig:"){width=".9\linewidth"} ![First $20 \times 20$ entries of the matrix $G$ defined by Eq. (\[equ:DefG\]). $N = 1331$, $\alpha = 0.37$, 100 realizations. Gray bars represent positive entries and white bars negative ones.[]{data-label="fig:GAveraged"}](G3 "fig:"){width=".9\linewidth"}
To elucidate the the influence of the different spectra in the 1D and 3D case, we simply substitute the eigenvalues of the 1D case by the ones of the 3D case (i.e. use 1D $G$-matrix together with the corresponding 3D eigenvalues): $$\begin{aligned}
\langle \tilde{Y}_2^{(1D)}\rangle_{p_0, \tau} &\simeq \sum_{ij} \left\langle G^{(1)}_{ij}\right\rangle_\tau \left( e^{-\tilde{\lambda}^{(3)}_i}\right)^t \left( e^{-\tilde{\lambda}^{(3)}_j}\right)^t,
\label{equ:1Dwith3Deigenvalues}
\intertext{but still take}
\langle \tilde{Y}_2^{(3D)}\rangle_{p_0, \tau} &\simeq \sum_{ij} \left\langle G^{(3)}_{ij}\right\rangle_\tau \left( e^{-\tilde{\lambda}^{(3)}_i}\right)^t \left( e^{-\tilde{\lambda}^{(3)}_j}\right)^t.\nonumber\end{aligned}$$ Thus, in this approximation, the only difference between the 1D and the 3D case is the shape of the matrix $G$ defined by Eq. (\[equ:DefG\]). The result of the approximation Eq. (\[equ:1Dwith3Deigenvalues\]) is also presented in Fig. \[fig:Factorized\]. Although the plateau of the case Eq. (\[equ:1Dwith3Deigenvalues\]) lies lower than the one of Eq. (\[equ:Factorized\]), it is still there and pronounced. Thus, the exact form of the spectrum is only responsible for the height and the duration of the plateau, not for the fact of its presence or absence. We thus conclude that the difference in the spectrum $\rho_\lambda$ is primarily responsible for the different time scales but *not* for the fact of dynamical localization.
Since we have shown that it is the matrix $G$ built from the eigenvectors of $L$ which codes for the presence or absence of dynamical localization, the question arises, what properties of its elements are mainly responsible for it. The elements of $G$ differ strongly in their magnitude. Is the plateau visible in Fig. \[fig:Factorized\] the result of a complex interplay between positive and negative small components of $G$ or is its presence dominated by the behavior of large components of $G$ other than $G_{11}$ (coding for the equilibrium)?
To attack this question, the first $20 \times 20$ entries of $\langle G\rangle_\tau$ in the 1D and 3D case are depicted in Fig. \[fig:GAveraged\]. In order to make the important information visible, the $z$-axis represents the absolute values of the elements, $|\left\langle G_{jk}\right\rangle_\tau|$, whereas the sign of the entries is indicated by the color of the bars: Gray bars represent positive entries and white bars negative ones. Since the equilibrium state is the same in 1D and 3D, $\langle G_{11} \rangle = 1-\alpha$ in both cases. In the 3D case however, the values of the other diagonal entries decay more rapidly than in the 1D case. Moreover, while in the 3D case, for $k \neq j$ and $j,k > 1$, the entries $\left\langle G_{jk}\right\rangle_\tau$ are close to zero and of fluctuating sign, in 1D case all shown off-diagonal entries are negative (although for larger indices $j,k$ they also fluctuate).
![Temporal evolution of the summands $A$ (filled symbols) and $B$ (empty symbols) of Eq. (\[equ:SplittedSum\]) for the 1D (circles) and 3D (squares) case. Simulations with system size $N = 1331$, $\alpha = 0.37$, and 100 landscape realizations.[]{data-label="fig:Factorization"}](Factorization){width="1.1\linewidth"}
Let us split the diagonal and the non-diagonal contributions to the overall sum over eigenstates: $$\begin{aligned}
\langle Y_2 \rangle_{p_0, \tau} &= \underbrace{\sum_{j} \left\langle G_{jj}\right\rangle_\tau \left\langle e^{-\lambda_j}\right\rangle_\tau^{2t}}_{= A}\nonumber\\
&\quad + \underbrace{\sum_{j\neq k} \left\langle G_{jk}\right\rangle_\tau \left\langle e^{-\lambda_j}\right\rangle_\tau^t \left\langle e^{-\lambda_k}\right\rangle_\tau^t}_{= -B}
\label{equ:SplittedSum}\end{aligned}$$ The diagonal elements of $G$ are always positive, and so is $A$. According to the simulations, the term $-B$ is negative such that $B$ is positive. Fig. \[fig:Factorization\] shows the temporal evolution of $A$ and $B$ in both 1D and 3D case, where in both cases we have used again the spectrum of the 3D system. It seems that the plateau in Fig. \[fig:Factorized\] is indeed the result of an interplay of the positive diagonal entries in $A$ and the mostly negative ones in $-B$. In 1D both terms decay logarithmically but the sum $A-B$ stays constant over a long time. In three dimensions the terms behave completely different. This shows that the presence or absence of dynamical localization is fully coded in the matrix $G$, i.e. depends on the properties of eigenvectors of $L$, and not so much on its spectrum. It is strongly dominated by the scalar products of eigenvectors to different eigenvalues (which are not orthogonal since $L$ is not symmetric) entering the non-diagonal terms, and in this sense indeed has to do with eigenvector localization, although the direct relation is not clear yet.
Conclusions
===========
In the present work we applied an algebraic approach to investigate the phenomenon of dynamical localization in the random trap model with power-law distributed mean waiting times. Apart from its formal elegance, the approach works extremely well as a computational scheme, since it allows for obtaining numerically exact results for the participation ratio $Y_2$ averaged over initial conditions and thermal histories in a given system’s realization. Only a single averaging procedure over the samples is necessary. Because this approach enables us to compute the relevant properties at any given time, it allows for considering longer observation times than in the direct Monte-Carlo simulations.
The phenomenon of dynamical localization, as observed in the participation ratio, is a very peculiar property of one-dimensional trap models. Its physical interpretation is that the probability, that two particles with the same starting position are found together, is surprisingly high for intermediate time scales. In this context, *intermediate time scales* means that full equilibration is not yet established. The fact, that the particles often meet at the same site, does not mean that they stay close to each other: the typical distance between the particles continuously grows with time until the terminal equilibration is reached. Moreover, the equilibrium state, which is the strongest localized one with respect to $Y_2$, is essentially delocalized with respect to its radius of gyration and its Lyapunov exponent. Numerical simulations show that the later two are well-correlating localization measures for the eigenstates of the system.
We have moreover shown, that the phenomenon of dynamical localization is only marginally connected with the spectral properties of the Laplacian operator governing the system’s dynamics, and is dominated by the properties of the eigenfunctions of $L$, which outlines the direction of further investigations.
Acknowledgements
================
The financial support of DFG within the IRTG 1740 “Dynamical Phenomena in Complex Networks: Fundamentals and Applications” is gratefully acknowledged.
Furthermore, the authors thank Prof. D. H. U. Marchetti for useful discussions.
[10]{}
J. Haus, K. Kehr, Phys. Rep. [**150**]{}, 263 (1987).
H. Scher, E. W. Montroll, Phys. Rev. B [**12**]{}, 2455 (1975).
J.-P. Bouchaud and A. Georges, Phys. Rep. [**195**]{}, 127 (1990).
S. Alexander, J. Bernasconi, W. R. Schneider, R. Orbach, Rev. Mod. Phys. [**53**]{}, 175 (1981).
F. J. Dyson, Phys. Rev. [**92**]{}, 1331 (1953).
J. P. [Bouchaud]{}, J. Phys. I [**2**]{}, 1705 (1992).
P. Hänggi, P. Talkner, M. Borkovec, Rev. Mod. Phys. [**62**]{}, 251 (1990).
G. Ben Arous, J. Černý, Ann. Appl. Probab. [**15**]{}, 1161 (2005).
E. M. Bertin, J.-P. Bouchaud, Phys. Rev. E [**67**]{}, 026128 (2003).
B. Derrida, Physica D: Nonlinear Phenomena [**107**]{}, 186 (1997).
L. [Fontes]{}, M. [Isopi]{}, C. [Newman]{}, Ann. Probab. [**30**]{}, 579 (2002).
L. R. G. Fontes, M. Isopi, C. M. Newman, Probab. Theory Related Fields [**115**]{}, 417 (1999).
I. M. Sokolov, I. I. Eliazar, Phys. Rev. E [**81**]{}, 026107 (2010).
K. Ishii, Progress of Theoretical Physics Supplement [**53**]{}, 77 (1973).
A. Comtet, C. Texier, Y. Tourigny, Journal of Physics A: Mathematical and Theoretical [**46**]{} 254003 (2013).
D. J. Thouless, Journal of Physics C: Solid State Physics [**5**]{}, 77 (1972).
A. Bovier, A. Faggionato, Ann. Appl. Probab. [**15**]{}, 1997 (2005).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Coherence arises from the superposition principle and plays a key role in quantum mechanics. Recently, Baumgratz et al. \[T. Baumgratz, M. Cramer, and M. B. Plenio, Phys. Rev. Lett. 113, 140401 (2014)\] established a rigorous framework for quantifying the coherence of finite dimensional quantum states. In this work we provide a framework for quantifying the coherence of Gaussian states and explicitly give a coherence measure based on the relative entropy.'
author:
- Jianwei Xu
title: Quantifying coherence of Gaussian states
---
Introduction
============
Coherence is a basic feature in quantum mechanics, it is a common necessary condition for both entanglement and other types of quantum correlations. Many works have been undertaken to theoretically formulate quantum coherence [@Glauber1963; @Sudarshan1963; @Luo2005; @Aberg2006; @Monras2014; @Aberg2014; @Girolami2014], but up to now there has been no well-accepted efficient method for quantifying coherence. Recently, Baumgratz et al. established a comprehensive framework of coherence quantification [@Baumgratz2014], by which coherence is considered to be a resource that can be characterized, quantified, and manipulated in a manner similar to quantum entanglement [@Bennett1996; @Vedral1998; @Plenio2007; @Horodecki2009]. This seminal work has triggered the community’s interest of other proper measures for coherence [@Streltsov2015; @Shao2015], the freezing phenomenon [@Bromley2015], and some further developments [@Xi2015; @Pires2015; @Bera2015; @Singh2015; @Yadin2015; @Yuan2015; @Yao2015]. Ref.[@Du2015] revealed the condition of coherence transformations for pure states, and then built a universal method for quantifying the coherence of mixed states via the convex roof scheme.
All above results for quantifying quantum coherence are implicitly assumed the finite dimensional setting, which is neither necessary nor desirable. In relevant physical situations such as quantum optics, it must require the quantum states in infinite dimensional systems, especially the Gaussian states [@Braunstein2005; @Wang2007; @Weedbrook2012]. In this work we investigate the quantification of coherence for Gaussian states.
This work is organized as follows. In section II, we discuss the necessary conditions any measure of coherence for Gaussian states should satisfy. In section III, we prove that the incoherent one-mode Gaussian states are just thermal states. In section IV, we determine the structure of one-mode incoherent Gaussian operations. In section V, we explicitly provide a coherence measure for one-mode Gaussian states based on the relative entropy. In section VI we consider the multi-mode case. Section VII is a brief summary.
How to quantify the coherence of Gaussian states
================================================
A state $\rho $ (finite or infinite dimensional) is said to be incoherent if it is diagonal when expressed it in a fixed orthonormal basis. We denote the set of all incoherent states by $\mathcal{I}$. A quantum map is called incoherent operation (ICPTP) if it is completely positive, trace-preserving, and maps any incoherent states into incoherent states. Ref.[@Baumgratz2014] presented the postulates that any proper measure of the coherence $C(\rho )$ for finite-dimensional state $\rho $ must satisfy as follows.
$(C1)$ $C(\rho )\geq 0$ and $C(\rho )=0$ iff $\rho \in \mathcal{I}$.
$(C2a)$ Monotonicity under all incoherent completely positive and trace-preserving (ICPTP) maps: $C(\rho )\geq C(ICPTP(\rho )).$
$(C2b)$ Monotonicity for average coherence under subselection based on measurement outcomes: $C(\rho )\geq \sum_{n}p_{n}C(\rho _{n})$, where $\rho _{n}=K_{n}\rho K_{n}^{+}/p_{n}$ and $p_{n}=tr(K_{n}\rho
K_{n}^{+})$ for all $n$, $\sum_{n}K_{n}^{+}K_{n}=I$, $K_{n}\mathcal{I}%
K_{n}^{+}\subset \mathcal{I}$, with $+$ the adjoint and $I$ identity operator.
$(C3)$ Nonincreasing under the mixing of quantum states: $\sum_{n}p_{n}C(\rho _{n})\geq C(\sum_{n}p_{n}\rho _{n}).$
Note that $(C2b)$ and $(C3)$ together imply $C(2a)$.
For the case of Gaussian states, we adopt $(C1)$ and $C(2a)$ as necessary conditions that any coherence measure should satisfy, while give up $C(2b)$ and $C(3)$. Gaussian states do not form a convex set, then it seems hard to establish the counterparts of $C(2b)$ and $C(3)$.
Incoherent states of one-mode Gaussian states
=============================================
In this section, we find out the incoherent states of one-mode Gaussian states. We first note that, coherence is basis dependent, so wherever we talk about coherence we must be clear which basis is presupposed.
***Theorem 1.*** For fixed orthonormal basis $\{|n\rangle \}_{n=0}^{\infty }$, a one-mode Gaussian state is diagonal iff it is a thermal state.
**Proof.** A state $\rho $ is called Gaussian if its characteristic function $%
\chi (\rho ,\lambda )=tr[\rho D(\lambda )]$ is of the form $$\begin{aligned}
\chi (\rho ,\lambda )=\exp \{-\frac{1}{2}(x_{\lambda },y_{\lambda })\Omega
V\Omega ^{t}\binom{x_{\lambda }}{y_{\lambda }} \nonumber \\
-i[\Omega \binom{d_{1}}{d_{2}}%
]^{t}\binom{x_{\lambda }}{y_{\lambda }}\},\end{aligned}$$ where $D(\lambda )=e^{\lambda a^{+}-\lambda ^{\ast }a}$ is the displacement operator, $a$, $a^{+}$ are creator operator and annihilation operator, $%
x_{\lambda }$ and $y_{\lambda }$ are the real and imaginary parts of $%
\lambda $, $\Omega =\left(%
\begin{array}{cc}
0 & 1 \\
-1 & 0%
\end{array}\right)
,d=(d_{1},d_{2})^{t}$ with $d_{1}$ and $d_{2}$ real numbers, $t$ denotes transpose, $V=\left(
\begin{array}{cc}
V_{11} & V_{12} \\
V_{21} & V_{22}%
\end{array}\right)
$ is real symmetric positive and satisfies $V+i\Omega \geq 0$. A Gaussian state $\rho $ is fully described by the covariance matrix $V$ and the displacement vector $d$. $detV\geq 1$ and $detV=1$ iff $\rho $ is pure. We write a Gaussian state as $\rho (V,d).$
A state is called thermal if it has the form $$\begin{aligned}
\rho _{th}(\overline{n})=\sum_{n=0}^{\infty }\frac{\overline{n}^{n}}{(%
\overline{n}+1)^{n+1}}|n\rangle \langle n|,\end{aligned}$$ where $\overline{n}=tr[a^{+}a\rho _{th}(\overline{n})]\geq 0$ is the mean number.
It is easy to check that the characteristic function of the thermal state $%
\rho _{th}(\overline{n})$ is Gaussian with covariance matrix $(2\overline{n}%
+1)I$ and zero displacement vector. Hence we only need to prove that the diagonal Gaussian states must be thermal states. To this aim, we calculate the elements $\rho _{mn}=$ $\langle m|\rho |n\rangle $ from Eq.(1) and its inverse relation $$\begin{aligned}
\rho =\int \frac{d^{2}\lambda }{\pi }\chi (\rho ,\lambda )D(-\lambda ),\end{aligned}$$ where $d^{2}\lambda =dx_{\lambda }dy_{\lambda }$ and $\int =\int_{-\infty
}^{\infty }$. We thus have $$\begin{aligned}
\langle m|\rho |n\rangle =\int \frac{d^{2}\lambda }{\pi }\chi (\rho
,\lambda )\langle m|D(-\lambda )|n\rangle ,
\\
\langle m|D(-\lambda )|n\rangle \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \nonumber \\
=\int \int \frac{d^{2}\alpha }{\pi }\frac{%
d^{2}\beta }{\pi }\langle m|\alpha \rangle \langle \alpha |D(-\lambda
)|\beta \rangle \langle \beta |n\rangle ,
\\
\langle \alpha |D(-\lambda )|\beta \rangle =\langle 0|D(-\alpha )D(-\lambda
)D(\beta )|0\rangle ,\end{aligned}$$ where $|\alpha \rangle =e^{-\frac{|\alpha |^{2}}{2}}\sum_{n=0}^{\infty }%
\frac{\alpha ^{n}}{\sqrt{n!}}|n\rangle $, $|\beta \rangle =e^{-\frac{|\beta
|^{2}}{2}}\sum_{n=0}^{\infty }\frac{\beta ^{n}}{\sqrt{n!}}|n\rangle $ are coherent states. Using the formula $$\begin{aligned}
D(\alpha )D(\beta )=e^{\frac{\alpha \beta ^{\ast }-\alpha ^{\ast }\beta }{2}%
}D(\alpha +\beta ),\end{aligned}$$ and after direct algebras, we get $$\begin{aligned}
\langle m|\rho |n\rangle =\int \int \int \frac{d^{2}\lambda }{\pi }\frac{%
d^{2}\alpha }{\pi }\frac{d^{2}\beta }{\pi }\frac{\alpha ^{m}\beta ^{\ast n}}{%
\sqrt{m!n!}}\exp b, \ \ \ \
\\
b=-\alpha ^{\ast }\lambda +\alpha ^{\ast }\beta +\lambda ^{\ast }\beta
-|\alpha |^{2}-|\beta |^{2}-\frac{1}{2}|\lambda |^{2}-\chi (\rho ,\lambda ).\end{aligned}$$ Eq.(8) is somewhat similar to the results of Refs.[@Isserlis1918; @Wick1950; @Withers1985], but in fact not the same thing. To calculate Eq.(8), we introduce the integration $$\begin{aligned}
J=\int \int \int d^{2}\lambda d^{2}\alpha d^{2}\beta \exp \{b+u\alpha
+v\beta ^{\ast }\},\end{aligned}$$ where $u$, $v$ are real numbers. As a result, $$\begin{aligned}
\rho _{mn}=\frac{1}{\pi ^{3}}(\frac{\partial ^{m}}{\partial u^{m}}\frac{%
\partial ^{n}}{\partial v^{n}}J)_{u=v=0}.\end{aligned}$$ $J$ is a Gaussian integration we can calculate as follows.
Write $b+u\alpha +v\beta ^{\ast }$ as $$\begin{aligned}
b+u\alpha +v\beta ^{\ast }=-\frac{1}{2}(x_{\alpha },y_{\alpha },x_{\beta
},y_{\beta },x_{\lambda },y_{\lambda })A \ \ \ \ \ \ \ \ \nonumber \\
\cdot(x_{\alpha },y_{\alpha },x_{\beta
},y_{\beta },x_{\lambda },y_{\lambda })^{t}+B(x_{\alpha },y_{\alpha
},x_{\beta },y_{\beta },x_{\lambda },y_{\lambda })^{t},\end{aligned}$$ where $A$ is a $6\times 6$ symmetric complex matrix, $B$ is a $6\times 1$ row complex vector, $$\begin{aligned}
A=\left(
\begin{array}{cccccc}
2 & 0 & -1 & -i & 1 & i \\
0 & 2 & i & -1 & -i & 1 \\
-1 & i & 2 & 0 & -1 & i \\
-i & -1 & 0 & 2 & -i & -1 \\
1 & -i & -1 & -i & 1+V_{22} & -V_{12} \\
i & 1 & i & -1 & -V_{12} & 1+V_{11}%
\end{array} \right),
\\
detA=16(\det V+V_{11}+V_{22}+1)>0, \ \ \ \ \ \ \ \
\\
B=(u,iu,v,-iv,-id_{2},id_{1}). \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \\end{aligned}$$ Apply the Gaussian integration formula we get $$\begin{aligned}
J=\frac{(2\pi )^{3}}{\sqrt{\det A}}\exp [\frac{1}{2}BA^{-1}B^{t}].\end{aligned}$$ Let $$\begin{aligned}
\xi =\frac{1}{2}BA^{-1}B^{t}=\frac{1}{2}%
[(u,v)B_{2}(u,v)^{t}+B_{1}(u,v)^{t}+B_{0}], \ \\end{aligned}$$ where $$\begin{aligned}
B_{2}=\left(
\begin{array}{cc}
\frac{V_{11}-V_{22}+2iV_{12}}{1+V_{11}+V_{22}+V_{11}V_{22}-V_{12}^{2}} &
\frac{V_{11}V_{22}-V_{12}^{2}-1}{1+V_{11}+V_{22}+V_{11}V_{22}-V_{12}^{2}} \\
\frac{V_{11}V_{22}-V_{12}^{2}-1}{1+V_{11}+V_{22}+V_{11}V_{22}-V_{12}^{2}} &
\frac{V_{11}-V_{22}-2iV_{12}}{1+V_{11}+V_{22}+V_{11}V_{22}-V_{12}^{2}}%
\end{array}\right), \nonumber
\\
B_{1}=2(\frac{(1-iV_{12}+V_{22})d_{1}+i(1+V_{11}+iV_{12})d_{2}}{%
1+V_{11}+V_{22}+V_{11}V_{22}-V_{12}^{2}}, \ \ \ \ \ \ \ \ \ \nonumber \\
\frac{%
(1+iV_{12}+V_{22})d_{1}-i(1+V_{11}-iV_{12})d_{2}}{%
1+V_{11}+V_{22}+V_{11}V_{22}-V_{12}^{2}}) \ \ \ \ \ \ \ \ \ \nonumber
\\
B_{0}=-\frac{(1+V_{22})d_{1}^{2}-2V_{12}d_{1}d_{2}+(1+V_{11})d_{2}^{2}}{%
1+V_{11}+V_{22}+V_{11}V_{22}-V_{12}^{2}}. \ \ \ \ \ \ \ \ \ \ \ \ \ \nonumber\end{aligned}$$ Introduce the symbols $$\begin{aligned}
J_{k_{1}k_{2}...k_{m}}=\frac{\partial ^{m}J}{\partial k_{1}\partial
k_{2}...\partial k_{m}},
\\
\xi _{k_{1}k_{2}...k_{m}}=\frac{\partial ^{m}\xi }{\partial k_{1}\partial
k_{2}...\partial k_{m}},\end{aligned}$$ where $k_{1},k_{2},...k_{m}\in \{u,v\}$, $m\in \{1,2,3,...\},$and let $J(0)$, $J_{k_{1}k_{2}...k_{m}}(0)$, $\xi (0)$, $\xi _{k_{1}k_{2}...k_{m}}(0)$ represent the corresponding values when $u=v=0$. Direct calculations show that $$\begin{aligned}
\xi _{k_{1}}(0)=\frac{1}{2}(B_{1})_{k_{1}}, \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\\
\xi _{k_{1}k_{2}}(0)=(B_{2})_{k_{1}k_{2}}, \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\\
\xi _{k_{1}k_{2}...k_{m}}(0)=0 \text{\ when \ } m\geq 3, \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\\
J_{k_{1}}(0)=J(0)\xi _{k_{1}}(0), \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\\
J_{k_{1}k_{2}}(0)=J(0)[\xi _{k_{1}k_{2}}(0)+\xi _{k_{1}}(0)\xi
_{k_{2}}(0)], \ \ \ \ \ \ \ \
\\
J_{k_{1}k_{2}k_{3}}(0)=J(0)[\xi _{k_{1}}(0)\xi _{k_{2}k_{3}}(0)+\xi
_{k_{2}}(0)\xi _{k_{1}k_{3}}(0) \nonumber \\
+\xi _{k_{3}}(0)\xi _{k_{1}k_{2}}(0)+\xi
_{k_{1}}(0)\xi _{k_{2}}(0)\xi _{k_{3}}(0)],...,
\\
J_{k_{1}k_{2}...k_{m}}(0)=J(0)\sum_{\sigma
(k_{1}k_{2}...k_{m})}\sum_{r=0}^{[\frac{m}{2}]}\xi _{k_{1}k_{2}}(0)\xi
_{k_{3}k_{4}}(0) \nonumber \\
...\xi _{k_{2r-1}k_{2r}}(0)\xi _{k_{2r+1}}(0)...\xi
_{k_{m}}(0),\end{aligned}$$ where $[\frac{m}{2}]=\frac{m}{2}$ when $m$ is even and $[\frac{m}{2}]=\frac{%
m-1}{2}$ when $m$ is odd, $\sigma (k_{1}k_{2}...k_{m})$ is any permutation of $k_{1}k_{2}...k_{m}$, $\sum_{\sigma (k_{1}k_{2}...k_{m})}$ sums all permutations of $k_{1}k_{2}...k_{m}$.
From Eqs.(11,26) we can calculate any $\rho _{mn}$ in principle.
If the Gaussian state $\rho $ is diagonal thus $\rho _{01}=\rho _{02}=0,$ $$\begin{aligned}
\rho _{01}=\frac{2^{3}}{\sqrt{\det A}}J(0)\frac{1}{2}(B_{1})_{2}=0%
\Rightarrow (B_{1})_{2}=0.
\\
\rho _{02}=\frac{2^{3}}{\sqrt{\det A}}J(0)[(B_{2})_{22}+\frac{1}{2}%
(B_{1})_{2}\frac{1}{2}(B_{1})_{2}]=0 \nonumber \\
\Rightarrow (B_{2})_{22}=0.\end{aligned}$$ Similarly $\rho _{10}=\rho _{20}=0$ yield $$\begin{aligned}
(B_{1})_{1}=0, \\
(B_{2})_{11}=0.\end{aligned}$$ Taking Eqs.(27-30) into $B_{2},B_{1},$ we get $$\begin{aligned}
V_{11}-V_{22}=V_{12}=d_{1}=d_{2}=0,\end{aligned}$$ hence $\rho $ is a thermal state. We then complete this proof.
Incoherent operations of one-mode Gaussian states
=================================================
A one-mode Gaussian operation is described by $(T,N,\overline{d})$, it performs on the Gaussian state $\rho (V,d)$ and get the Gaussian state with the covariance matrix and displacement vector as [@Holevo2001] $$\begin{aligned}
d\rightarrow Td+\overline{d},V\rightarrow TVT^{t}+N,\end{aligned}$$ where $N$, $T$ are real matrices satisfying $N=N^{t}\geq 0,\det N\geq (\det
T-1)^{2}$.
We give the definition of incoherent operation for one-mode Gaussian states. A one-mode Gaussian operation is called incoherent if it maps any incoherent Gaussian state into incoherent Gaussian states. We denote the set of all incoherent operations by $\mathcal{I}$. We now study the structure of one-mode Gaussian incoherent operation.
Suppose one-mode Gaussian operation $(T,N,\overline{d})$ is incoherent, for any incoherent state $\rho (rI,0)$ with $r\geq 1,$ after the action of $(T,N,%
\overline{d})$, $\rho (rI,0)$ becomes the incoherent state $\rho (sI,0)$ with $s\geq 1.$ From Eq.(32), we have $$\begin{aligned}
\overline{d}=0, \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\\
rTT^{t}+N=sI, \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\\
N=sI-rTT^{t}=\left(
\begin{array}{cc}
s-r(TT^{t})_{11} & -r(TT^{t})_{12} \\
-r(TT^{t})_{21} & s-r(TT^{t})_{22}%
\end{array}\right). \\end{aligned}$$ For any $r\geq 1$, there exists $s\geq 1$ such that Eq.(35) holds, then we get $(TT^{t})_{11}=(TT^{t})_{22}$, $(TT^{t})_{12}=(TT^{t})_{21}=0$. Let $$\begin{aligned}
(TT^{t})_{11}=(TT^{t})_{22}=t^{2},
\\
T=tO \text{ \ with \ } OO^{t}=I,\end{aligned}$$ where $t$, $\alpha $, $\beta $ are all real numbers, $O$ is a real orthogonal matrix. Taking Eqs.(36,37) into Eq.(35) we get $$\begin{aligned}
N=\left(
\begin{array}{cc}
w & 0 \\
0 & w%
\end{array} \right),\end{aligned}$$ where $w=s-rt^{2}$. The conditions $N\geq 0,\det N\geq (\det T-1)^{2}$ yield $$\begin{aligned}
w\geq |t^{2}\det O-1|.\end{aligned}$$ We conclude this section as Theorem 2 below.
***Theorem 2.*** A one-mode Gaussian operation is called incoherent if it maps any incoherent Gaussian state into incoherent Gaussian states. Any one-mode incoherent Gaussian operation can be expressed by $(T,N)$ in Eqs.(37-39).
A coherence measure of one-mode Gaussian states based on relative entropy
=========================================================================
For any one-mode Gaussian state $\rho (V,d)$, we define a coherence measure as $$\begin{aligned}
C(\rho )=\inf_{\delta }\{S(\rho ||\delta ):\delta \text{ \ is \ an \ incoherent \ state\}},\end{aligned}$$ where $S(\rho ||\delta )=tr(\rho \log _{2}\rho )-tr(\rho \log _{2}\delta )$ is the relative entropy, $inf$ runs over all incoherent Gaussian states. The entropy of $\rho $, $S(\rho )=-tr(\rho \log
_{2}\rho )$ is [@Holevo1999] $$\begin{aligned}
S(\rho )=g(\nu )=\frac{\nu +1}{2}\log _{2}\frac{\nu +1}{2}-\frac{\nu -1}{2}%
\log _{2}\frac{\nu -1}{2},\end{aligned}$$ where $\nu =\sqrt{\det V}$. We now calculate $\sup_{\delta }tr(\rho \log
_{2}\delta )$. Suppose $$\begin{aligned}
\delta (\overline{n})=\sum_{n=0}^{\infty }\frac{\overline{n}^{n}}{(%
\overline{n}+1)^{n+1}}|n\rangle \langle n|,\end{aligned}$$ then $$\begin{aligned}
tr[\rho \log _{2}\delta ]=tr[\rho _{diag}\log _{2}\delta
]=\sum_{n=0}^{\infty }\rho _{nn}\log _{2}\frac{\overline{n}^{n}}{(\overline{n%
}+1)^{n+1}}
\nonumber \\
=(\sum_{n=0}^{\infty }n\rho _{nn})\log \overline{n}%
-(\sum_{n=0}^{\infty }n\rho _{nn}+1)\log (\overline{n}+1), \ \\end{aligned}$$ where $\rho _{diag}=\sum_{n=0}^{\infty }\rho _{nn}|n\rangle \langle n|$. It follows that $$\begin{aligned}
\frac{\partial }{\partial \overline{n}}tr[\rho \log _{2}\delta ]=\frac{1}{%
\ln 2}\frac{1}{\overline{n}+1}[\frac{\sum_{n=0}^{\infty }\rho _{nn}n}{%
\overline{n}}-1].\end{aligned}$$ Let $\frac{\partial }{\partial \overline{n}}tr[\rho \log _{2}\delta ]=0,$ we get $$\begin{aligned}
\overline{n}=\sum_{n=0}^{\infty }\rho _{nn}n.\end{aligned}$$ The remaining is how to calculate $\overline{n}=\sum_{n=0}^{\infty }\rho
_{nn}n.$ $$\begin{aligned}
\overline{n}=\sum_{n=0}^{\infty }\langle n|\rho |n\rangle
n=\sum_{n=0}^{\infty }\langle n|\rho a^{+}a|n\rangle =tr(\rho a^{+}a)
\nonumber \\
=\int
\frac{d^{2}\alpha }{\pi }\langle \alpha |\rho a^{+}a|\alpha \rangle =\int
\frac{d^{2}\alpha }{\pi }\alpha \langle \alpha |\rho a^{+}|\alpha \rangle \ \
\nonumber \\
=\int \frac{d^{2}\alpha }{\pi }\alpha \int \frac{d^{2}\lambda }{\pi }\chi
(\rho ,\lambda )\langle \alpha |D(-\lambda )a^{+}|\alpha \rangle , \ \ \\end{aligned}$$ where we have used $a^{+}a|n\rangle =n|n\rangle $, the coherent state $%
|\alpha \rangle $, $a|\alpha \rangle =\alpha |\alpha \rangle $, and Eq.(3).
It is easy to check that $$\begin{aligned}
a^{+}|\alpha \rangle =e^{-\frac{|\alpha |^{2}}{2}}\sum_{n=1}^{\infty }\frac{%
n\alpha ^{n-1}}{\sqrt{n!}}|n\rangle ,
\\
\langle \alpha |D(-\lambda )=e^{\frac{\alpha \lambda ^{\ast }-\alpha ^{\ast
}\lambda }{2}}\langle \alpha +\lambda |,\end{aligned}$$ thus direct algebras show that $$\begin{aligned}
\langle \alpha |D(-\lambda )a^{+}|\alpha \rangle =(\alpha ^{\ast }+\lambda
^{\ast })e^{\frac{3\alpha \lambda ^{\ast }-\alpha ^{\ast }\lambda +|\alpha
|^{2}-|\alpha +\lambda |^{2}}{2}}. \ \
\\
\overline{n}=\frac{1}{\pi ^{2}}\int \int dx_{\alpha }dy_{\alpha
}dx_{\lambda }dy_{\lambda }[x_{\alpha }^{2}+y_{\alpha }^{2}+x_{\alpha
}x_{\lambda }+y_{\alpha }y_{\lambda }
\nonumber \\
+i(y_{\alpha }x_{\lambda }-x_{\alpha
}y_{\lambda })]\exp [-\frac{1}{2}\sum_{ij}Q_{ij}x_{i}x_{j}+%
\sum_{i}a_{i}x_{i}], \ \\end{aligned}$$ where $\alpha =x_{\alpha }+iy_{\alpha },$ $\lambda =x_{\lambda }+iy_{\lambda
},$ $a=(0,0,-id_{2},id_{1}),$ $(x_{1},x_{2},x_{3},x_{4})=(x_{\alpha },y_{\alpha
},x_{\lambda },y_{\lambda }).$ $$\begin{aligned}
Q=\left(
\begin{array}{cccc}
0 & 0 & 0 & 2i \\
0 & 0 & -2i & 0 \\
0 & -2i & 1+V_{22} & -V_{12} \\
2i & 0 & -V_{12} & 1+V_{11}%
\end{array}\right),
\\
Q^{-1}=\left(
\begin{array}{cccc}
\frac{1+V_{11}}{4} & \frac{V_{12}}{4} & 0 & -\frac{i}{2} \\
\frac{V_{12}}{4} & \frac{1+V_{22}}{4} & \frac{i}{2} & 0 \\
0 & \frac{i}{2} & 0 & 0 \\
-\frac{i}{2} & 0 & 0 & 0%
\end{array}\right),
\\
Q^{-1}a^{t}=(\frac{d_{1}}{2},\frac{d_{2}}{2},0,0)^{t},aQ^{-1}a^{t}=0.\end{aligned}$$ Using the result of Ref.[@Withers1985] with some algebras we get $$\begin{aligned}
\overline{n}=\frac{1}{4}(V_{11}+V_{22}+d_{1}^{2}+d_{2}^{2}-2).\end{aligned}$$ In conclusion, we get
$$\begin{aligned}
C[\rho (V,d)] =\frac{\nu -1}{2}\log \frac{\nu -1}{2}-\frac{\nu +1}{2}\log
\frac{\nu +1}{2}
\nonumber \\
+(\overline{n}+1)\log (\overline{n}+1)-\overline{n}\log
\overline{n}, \\
\nu =\sqrt{\det V}=\sqrt{V_{11}V_{22}-V_{12}^{2}}, \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\\
\overline{n}=\frac{1}{4}%
(V_{11}+V_{22}+d_{1}^{2}+d_{2}^{2}-2). \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \\end{aligned}$$
We next prove $C[\rho ]$ is nondecreasing under any incoherent operation. For any incoherent operation $O$, suppose $C[\rho ]=S(\rho ||\overline{\rho }%
)$ with the thermal state $\overline{\rho }=\overline{\rho }(\overline{n})$ and $\overline{n}=\sum_{n=0}^{\infty }\rho _{nn}n$ as specified in Eq.(57), then we have $$\begin{aligned}
C[O(\rho )]\leq S[O(\rho )||O(\overline{\rho })]\leq S[\rho ||\overline{%
\rho }]=C[\rho ].\end{aligned}$$ In Eq.(58) above, the first inequality comes from the definition of $C[O(\rho
)]$ and the fact that $O(\overline{\rho })$ is a thermal state, the second inequality comes from the monotonicity of relative entropy under completely positive and trace preserving mapping [@Lindblad1975].
From Eqs.(55-57), we see that the coherence measure $C[\rho (V,d)]$ is strictly monotonically decreasing in $\nu $ while strictly monotonically increasing in $\overline{n}$. For pure Gaussian states $\nu =1$ reaches the minimum of $\nu $. In this sense, we say that the maximally coherent states are pure.
Multi-mode Gaussian states
==========================
We extend the results of one-mode Gaussian states into multi-mode Gaussian states. For the positive integer $m\geq 2$, an $m$-mode Gaussian state $\rho(V,d)$ is described by [@Weedbrook2012] its covariance matrix $V$, a $2m\times2m$ real symmetric positive matrix, and its displacement vector $d$, a $2m$ dimensional real vector. $V$ satisfies $V+i\Omega\geq0$ with $\Omega =\left(
\begin{array}{cc}
0 & 1 \\
-1 & 0%
\end{array}\right)
^{\otimes m}.$
Incoherent states
-----------------
For fixed orthonormal basis $(\{|n\rangle \}_{n=0}^{\infty })^{\otimes m}$ with positive integer $m\geq 2$, the diagonal states is of the form $$\begin{aligned}
\rho ^{A_{1}A_{2}...A_{m}} \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\nonumber \\
=\sum_{n_{1},...,n_{m}}\rho
_{n_{1}n_{1},...,n_{m}n_{m}}|n_{1}\rangle \langle n_{1}|\otimes ...\otimes
|n_{m}\rangle \langle n_{m}|,\end{aligned}$$ where $A_{i}$ denotes the $i$th mode, and $$\begin{aligned}
\rho _{n_{1}n_{1},...,n_{m}n_{m}} \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\nonumber \\
=\langle n_{1}|\langle n_{2}|...\langle
n_{m}|\rho ^{A_{1}A_{2}...A_{m}}|n_{m}\rangle ...|n_{2}\rangle |n_{1}\rangle. \ \ \ \ \ \\end{aligned}$$ It is easy to check that $$\begin{aligned}
[\rho ^{A_{1}A_{2}},\rho ^{A_{1}}]=0,\end{aligned}$$ where $\rho ^{A_{1}}$ is the reduced state with respect to the $A_{1}$ mode, $\rho ^{A_{1}A_{2}}$ is the reduced state with respect to the $A_{1}A_{2}$ modes, and $[\cdot ]$ denotes commutator. Recall that for two-mode Gaussian states [@Xu2015] $$\begin{aligned}
[\rho ^{A_{1}A_{2}},\rho ^{A_{1}}]=0\Leftrightarrow \rho ^{A_{1}A_{2}}=\rho
^{A_{1}}\otimes \rho ^{A_{2}}.\end{aligned}$$ Together with Theorem 1 above, we get that the incoherent two-mode Gaussian states are of the form $$\begin{aligned}
\rho ^{A_{1}A_{2}}=\rho _{th}^{A_{1}}(\overline{n_{1}})\otimes \rho
_{th}^{A_{2}}(\overline{n_{2}}),\end{aligned}$$ where $\rho _{th}^{A_{1}}(\overline{n_{1}})$ and $\rho _{th}^{A_{2}}(%
\overline{n_{2}})$ are all thermal states with mean numbers $\overline{n_{1}}
$ and $\overline{n_{2}}.$
For $m$-mode incoherent Gaussian state $\rho ^{A_{1}A_{2}...A_{m}}$, we consider its covariance matrix $V$. Since $\rho ^{A_{i}A_{j}}$ for any $%
1\leq i\leq j$ is of the form $\rho ^{A_{i}A_{j}}=\rho ^{A_{i}}\otimes \rho
^{A_{j}}$, hence the covariance matrix $V$ must be of diagonal form, that is $$\begin{aligned}
\rho ^{A_{1}A_{2}...A_{m}}=\otimes _{i=1}^{m}\rho _{th}^{A_{i}}(\overline{%
n_{i}}).\end{aligned}$$
Incoherent operation
--------------------
An $m$-mode Gaussian operation is described by $(T,N,\overline{d})$, it performs on the Gaussian state $\rho (V,d)$ and get the Gaussian state with the covariance matrix and displacement vector as [@Holevo2001] $$\begin{aligned}
d\rightarrow Td+\overline{d},V\rightarrow TVT^{t}+N,\end{aligned}$$ where $\overline{d}\in R^{2m}$, $N$, $T$ are $2m\times 2m$ real matrices satisfying $$\begin{aligned}
N+i\Omega -iT\Omega T^{t}\geq 0.\end{aligned}$$ Similar to the one-mode case, it is easy to determine the incoherent operation is of the form as follows. $T$ consists of $\{t_{i}O_{i}%
\}_{i=1}^{m}$ with $t_{i}$ real number, $O_{i}$ $2\times 2$ real matrix satisfying $O_{i}O_{i}^{t}=I,$ each $(2i-1,2i)$ row has just $t_{i}O_{i}$, each $(2j-1,2j)$ column has just one of $\{t_{i}O_{i}\}_{i=1}^{m}$, and other elements are all zero. $N=diag\{w_{1}I,w_{2}I,...,w_{m}I\}$ with $%
w_{i}\geq 0$ and $I$ being $2\times 2$ identity$.$The condition Eq.(65) then reads $$\begin{aligned}
w_{i}\geq |t_{i}^{2}\det O_{i}-1| \text{ \ for \ all \ } i.\end{aligned}$$
A coherence measure
-------------------
Similar to the one-mode case, we can generalize Eqs.(55-57) into $m$-mode case, that is $$\begin{aligned}
C[\rho (V,d)] =-S(\rho ) \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\nonumber \\
+\sum_{i=1}^{m}[(\overline{n_{i}}+1)\log _{2}(%
\overline{n_{i}}+1)-\overline{n_{i}}\log _{2}\overline{n_{i}}],\ \\
S(\rho ) =-\sum_{i=1}^{m}[\frac{\nu _{i}-1}{2}\log _{2}\frac{\nu _{i}-1}{2}%
-\frac{\nu _{i}+1}{2}\log _{2}\frac{\nu _{i}+1}{2}], \ \\
\overline{n_{i}} =\frac{1}{4}%
\{V_{11}^{(i)}+V_{22}^{(i)}+[d^{(i)}]_{1}^{2}+[d^{(i)}]_{2}^{2}-2\},\ \ \ \ \ \ \ \ \ \ \ \ \\end{aligned}$$ where $S(\rho )$ is the entropy of $\rho $ [@Holevo1999], $\{\nu _{i}\}_{i=1}^{m}$ are symplectic eigenvalues of $V
$ [@Weedbrook2012], $\overline{n_{i}}$ is determined by the $i$th-mode covariance matrix $V^{(i)}$ and displacement vector $d^{(i)}$. We can prove Eq.(68) fulfills (C2a) in the similar way as the one-mode case.
Summary
=======
In summary, along the line of quantifying coherence of finite-dimensional quantum states, we provided a measure for Gaussian states. To this aim, we proved that the incoherent Gaussian states are just thermal states. We defined the Gaussian incoherent operations as the Gaussian operations which maps incoherent states into incoherent states and found out the structure of Gaussian incoherent operations. The central result is that we provided a coherence measure for Gaussian states based on the relative entropy, it satisfies $(C1)$ and $(C2a)$.
There remain many questions for future investigations. Firstly, whether or not we can establish the counterparts of $(C2b)$ and $(C3)$ for Gaussian states, in present work we only adopt $(C1)$ and $(C2a)$ as necessary conditions for any coherence measure. Secondly, how about the behaviors of coherence in Gaussian dynamical systems, such as frozen coherence [@Bromley2015], sudden change etc.
This work was supported by the Chinese Universities Scientific Fund (Grant No.2014YB029) and the National Natural Science Foundation of China (Grant No.11347213).
[99]{} R. Glauber, Phys. Rev. 131, 2766 (1963).
E. Sudarshan, Phys. Rev. Lett. 10, 277 (1963).
S. Luo, Theor. Math. Phys. 143, 681 (2005).
J. Åberg, arXiv:quant-ph/0612146.
A. Monras, A. Chȩcińska, and A. Ekert, New J. Phys. 16, 063041 (2014).
J. Åberg, Phys. Rev. Lett. 113, 150402 (2014).
D. Girolami, Phys. Rev. Lett. 113, 170401 (2014).
T. Baumgratz, M. Cramer, and M. B. Plenio, Phys. Rev. Lett. 113, 140401 (2014).
C. H. Bennett, H. J. Bernstein, S. Popescu, and B. Schumacher, Phys. Rev. A 53, 2046 (1996).
V. Vedral and M. B. Plenio, Phys. Rev. A 57, 1619 (1998).
M. B. Plenio and S. Virmani, Quantum Info. Comput. 7, 1 (2007).
R. Horodecki, P. Horodecki, M. Horodecki, and K. Horodecki, Rev. Mod. Phys. 81, 865 (2009).
A. Streltsov, U. Singh, H. S. Dhar, M. N. Bera, and G. Adesso, Phys. Rev. Lett. 115, 020403 (2015).
L.-H. Shao, Z. Xi, H. Fan, and Y. Li, Phys. Rev. A 91, 042120 (2015).
T. R. Bromley, M. Cianciaruso, and G. Adesso, Phys. Rev. Lett. 114, 210401 (2015).
Z. Xi, Y. Li, and H. Fan, Sci. Rep. 5, 10922 (2015).
D. P. Pires, L. C. Céleri, and D. O. Soares-Pinto, Phys. Rev. A 91, 042330 (2015).
M. N. Bera, T. Qureshi, M. A. Siddiqui, and A. K. Pati, Phys. Rev. A 92, 012118 (2015).
U. Singh, M. N. Bera, H. S. Dhar, and A. K. Pati, Phys. Rev. A 91, 052115 (2015).
B. Yadin and V. Vedral, arXiv:1505.03792.
X. Yuan, H. Zhou, Z. Cao, and X. Ma, Phys. Rev. A 92, 022124 (2015).
Y. Yao, X. Xiao, L. Ge, and C. P. Sun, Phys. Rev. A 92, 022112 (2015).
S. Du, Z. Bai, and Y. Guo, Phys. Rev. A 91, 052120 (2015).
S. L. Braunstein and P. van Loock, Rev. Mod. Phys. 77, 513 (2005).
X. B. Wang, T. Hiroshimab, A. Tomitab, and M. Hayashi, Phys. Rep. 448, 1 (2007).
C. Weedbrook, S. Pirandola, R. García-Patrón, N. J. Cerf, T. C. Ralph, J. H. Shapiro, and S. Lloyd, Rev. Mod. Phys. 84, 621 (2012).
L. Isserlis, Biometrika 12, 134 (1918).
G. C. Wick, Physical Review 80 (2), 2689 (1950).
C. S. Withers Bull. Austral. Math. Soc. 32, 103 (1985).
A. S. Holevo, and R. F. Werner, 2001, Phys. Rev. A 63, 032312(2001).
A. S. Holevo, M. Sohma, and O. Hirota, Phys. Rev. A 59, 1820(1999).
G. Lindblad, Commun. Math. Phys. 40, 147 (1975).
J. Xu. Int. J. Theor. Phys. 54, 860 (2015).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We study quantum causal inference in a set-up proposed by Ried et al. \[Nat. Phys. 11, 414 (2015)\] in which a common-cause scenario can be mixed with a cause-effect scenario, and for which it was found that quantum mechanics can bring an advantage in distinguishing the two scenarios: Whereas in classical statistics, interventions such as randomized trials are needed, a quantum observational scheme can be enough to detect the causal structure if the common cause results from a maximally entangled state.\
We analyze this setup in terms of the geometry of unital positive but not completely positive qubit-maps, arising from the mixture of qubit-channels and steering maps. We find the range of mixing parameters that can generate given correlations, and prove a quantum advantage in a more general setup, allowing arbitrary unital channels and initial states with fully mixed reduced states. This is achieved by establishing new bounds on signed singular values of sums of matrices. Based on the geometry, we quantify and identify the origin of the quantum advantage depending on the observed correlations, and discuss how additional constraints can lead to a unique solution of the problem.
author:
- Jonas Kübler and Daniel Braun
bibliography:
- 'library.bib'
title: 'Two-qubit causal structures and the geometry of positive qubit-maps'
---
Introduction
============
Imagine a scenario where two experimenters, Alice and Bob, sit in two distinct laboratories. At one point Alice opens the door of her laboratory, obtains a coin, checks whether it shows heads or tails and puts it back out of the laboratory. Some time later also Bob obtains a coin and also he checks whether it shows heads or tails. This experiment is repeated many times (ideally: infinitely many times) and after this they meet and analyze their joint outcomes. Assuming their joint probability distribution entails correlations, there must be some underlying causal mechanism which causally connects their coins [@Reichenbach1971]. This could be an unobserved confounder (acting as a common-cause), and they actually measured two distinct coins influenced by the confounder. Or it could be that Alice’s coin was propagated by some mechanism to Bob’s laboratory, and hence they actually measured the same coin, with the consequence that manipulations of the coin by Alice can directly influence Bob’s result (cause-effect scenario). The task of Alice and Bob is to determine the underlying causal structure, i.e. to distinguish the two scenarios. This would be rather easy if Alice could prepare her coin after the observation by her choice and then check whether this influences the joint probability (so-called “interventionist scheme”). In the present scenario, however, we assume that this is not allowed (so-called “observational scheme”). All that Alice and Bob have are therefore the given correlations, and from those alone, in general they cannot solve this task without additional assumptions. Ried et al. [@Ried2015] showed that in a similar quantum scenario involving qubits the above task [*can*]{} actually be accomplished in certain cases even in an observational scheme (see below for a discussion of how the idea of an observational scheme can be generalized to quantum mechanics).\
In the present work we consider the same setup as in [@Ried2015], and allow arbitrary convex combinations of the two scenarios: The common-cause scenario is realized with probability $p$, the cause-effect scenario with probability $1-p$. Our main result are statements about the ranges of the parameter $p$ for which observed correlations can be explained with either one of the scenarios, or both. For this, we cast the problem in the language of affine representations of *unital positive qubit maps* [@Bengtsson2006] in which all the information is encoded in a $3\times 3$ real matrix, as is standard in quantum information theory for *completely positive unital qubit maps* [@Nielsen2009].\
The paper is structured as follows: In section \[sec:Causal\] we introduce causal models for classical random variables and for quantum systems. Therein we define what we consider a *quantum observational scheme*. Section \[sec:positiveMaps\] introduces the mathematical framework of ellipsoidal representations of qubit quantum-channels and qubit steering-maps. In section \[sec:Results\] we define our problem mathematically and prove the main results, which we then comment in the last section \[sec:discussion\].
Causal inference: classical versus quantum {#sec:Causal}
==========================================
Classical causal inference
--------------------------
At the heart of a *classical causal model* is a set of random variables $X_1, X_2, ..., X_N$. The observation of a specific value of a variable, $X_i = x_i$, is associated with an *event*. Correlations between events hint at some kind of causal mechanism that links the events [@Reichenbach1971] . Such a mechanism can be a deterministic law as for example $x_i = f(x_j)$ or can be a probabilistic process described by conditional probabilities $P(x_i|x_j)$, i.e. the probability to find $X_i=x_i$ given $X_j=x_j$ was observed. The causal mechanism may not be merely a direct causal influence from one observed event on the other, but may be due to common causes that lead with a certain probability to both events — or a mixture between both scenarios. Hence, by merely analysing correlations $P(x_1,x_2,\ldots,x_n)$, i.e. the joint probability distribution of all events, one can, in general, without prior knowledge of the *data generating process*, not uniquely determine the causal mechanism that leads to the observed correlations (purely *observational* scheme). To remedy this, an intervention is often necessary, where the value of a variable $X_i$ whose causal influence one wants to investigate, is set by an experimentalist to different values, trying to see whether this changes the statistics of the remaining events (*interventionist* scheme). One strategy for reducing the influence of other, unknown factors, is to randomize the samples. This is for example a typical approach in clinical studies, where one group of randomly selected probands receives a treatment whose efficiency one wants to investigate, and a randomly selected control group receives a placebo. If the percentage of cured people in the first group is significantly larger than in the second group, one can believe in a positive causal effect of the treatment. The probabilities obtained in this interventionist scheme are so-called “do-probabilities” (or “causal conditional probabilities”) [@Pearl2009a]: $P\left(x_i|\text{do}(x_j)\right)$ is the probability to find $X_i=x_i$ if an experimentalist intervened and set the value of $X_j$ to the value $x_j$. This is different from $P\left(x_i|x_j\right)$, as a possible causal influence from some other unknown event on $X_j=x_j$ is cut, i.e. one deliberately modifies the underlying causal structure for better understanding a part of it. If $X_j=x_j$ was the only direct cause of $X_i=x_i$ then $P\left(x_i|x_j\right) = P\left(x_i|\text{do}(x_j)\right)$. If instead the event $X_i=x_i$ was a cause of $X_j=x_j$, then intervening on $X_j$ cannot change $X_i$: $P(x_i)=
P\left(x_i|\text{do}(x_j)\right)=P\left(x_i|\text{do}(\bar{x_j})\right)$, where $\bar{x_j}$ is a value different from $x_j$. If the correlation between $X_i=x_i$ and $X_j=x_j$ is purely because of a common cause, then no intervenion on $X_i$ or $X_j$ will change the probability to find a given value of the other: $P(x_i) = P\left(x_i|\text{do}(x_j)\right)$ for all $x_j$, and $P(x_j)
= P\left(x_j|\text{do}(x_i)\right)$ for all $x_i$. Observing these do-probabilities one can hence draw conclusions about the causal influences behind the correlations observed in the occurence of $X_i=x_i$ and $X_j=x_j$.\
In practice, direct causation in one direction is often excluded by time-ordering and need not to be investigated. For example, when doubting that one can conclude that smoking causes lung cancer from the observed correlations between these two events, it does not make sense to claim that having lung cancer causes smoking, as usually smoking comes before developing lung cancer. But even dividing a large number of people randomly into two groups and forcing one of them to smoke and the other not to smoke in order to find out if there is a common cause for both would be ethically inacceptable. The needed do-probabilities can therefore not always be obtained by experiment. Interestingly, the causal-probability calculus allows one in certain cases, depending notably on the graph structure, to [*calculate*]{} do-probabilities from observed correlations without having to do the intervention. Inversely, apart from only predicting the conditional probabilities for a random variable say $X_i$ given the observation of $X_j=x_j$, denoted as $P(x_i|x_j)$, a causal model can also predict the do-probabilities, i.e. the distribution of $X_i$ if one *would* [intervene]{} on the variable $X_j$ and set its value to $x_j$. This is crucial for deriving informed recommendations for actions targeted at modifying certain probabilities, e.g. recommending not to smoke in order to reduce the risk for cancer.\
The structure of a causal model can be depicted by a graph. Each random variable is represented by a vertex of the graph. Causal connections are represented by directed arrows and imply that signaling along the direction of the arrow is possible. In a classical causal model it is assumed that events happen at specific points in space and time, therefore bidirectional signaling is not possible as it would imply signaling backward in time. Hence the graph cannot contain cycles and is therefore a *directed acyclic graph* (DAG) [@Pearl2009a], see FIG.\[Fig:4-DAG\]. The set of parents $PA_j$ of the random variable $X_j$ is defined as the set of all variables that have an immediate arrow pointing towards $X_j$, and $pa_j$ denotes a possible value of $PA_j$. The causal model is then defined through its graph with random variables $X_i$ at its vertices and the weights $P(x_j|pa_j)$ of each edge, i.e. the probabilities that $X_j=x_j$ happens under the condition that $Pa_j=pa_j$ occurred. The model generates the entire correlation function according to $$\label{eq:Pcm}
P(x_1,\ldots,x_n)=\prod_{j=1}^nP(x_j|pa_j)\,,$$ which is referred to as *causal Markov condition* [@Pearl2009a]. When all $P(x_1,\ldots,x_n)$ are given, then all conditional probabilities follow, hence all $P(x_j|pa_j)$ that appear in a given graph, but in general not all correlations nor all $P(x_j|pa_j)$ are known (see below). The causal inference probleme consists in finding a graph structure that allows one to satisfy eq. for given data $P(x_1,\ldots,x_n)$ and all known $P(x_j|pa_j)$, where the unknown $P(x_j|pa_j)$ can be considered fit-parameters in case of incomplete data. With access to the full joint probability distribution, the causal inference only needs to determine the graph. In practice, however, one often has only incomplete data: as long as a common cause has not been determined yet, one will not have data involving correlations of the corresponding variable. For example, one may have strong correlations between getting lung cancer (random variable $X_2\in\{0,1\}$) and smoking (random variable $X_1\in\{0,1\}$), but if there is a unknown common cause $X_0$ for both, one typically has no information about $P(x_0,x_1,x_2)$: One will only start collecting data about correlations between the presence of a certain gene, say, and the habit of smoking or developing lung cancer once one suspects that gene to be a cause for at least one of these. In this case $P(x_1|x_0)$ and $P(x_2|x_0)$ are fit parameters to the model as well. The possibility of extending a causal model through inclusion of unknown random variables is one reason why in general there is no unique solution to the causal inference problem based on correlations alone. Interventions on $X_i$ make it possible, on the other hand, to cut $X_i$ from its parents and hence eliminate unknown causes one by one for all random variables.\
Once a causal model is known, one can calculate all distributions $$\begin{aligned}
P(x_1,...,x_n \vert i_1,..., i_n) &= \prod_{j=1}^n P(x_j \vert pa_j, i_j), \label{eq:causalMarkov}\end{aligned}$$ for all possible combinations of interventions and observations, where the $i_j$ are the values of the intervention variable $I_j$ for the event $X_j$, $i_j = \text{idle}$ or $i_j = \text{do}(x_j)$. Here, $P\left(x_j| pa_j,i_j =\text{do}(\tilde{x}_j) \right)=
\delta_{x_j,\tilde{x}_j}$ reflects that an intervention on $X_j$ deterministically sets its value, independently of the observed values of its causal parents. If $I_j = \text{idle}$ then the value of $X_j$ only depends on its causal parents $PA_j$, i.e. $P(x_j| \{x_i\}_{i\neq j},i_j =
\text{idle}) = P(x_j| pa_j,i_j = \text{idle})$.
![Simple DAG in a four party scenario. The parental structure is $\text{PA}_A = \{\},\; \text{PA}_B = \{A\},\; \text{PA}_C = \{A,B\},\;\text{PA}_D = \{C\}$. According to the causal Markov condition, eq. , the probability distribution then factorizes as ${P(a,b,c,d|i_A, i_B, i_C, i_D) = P(d|c, i_D) P(c|a,b, i_C) P(b|a, i_B) P(a|i_A)}$.[]{data-label="Fig:4-DAG"}](4-DAG.pdf)
The field of *causal discovery* or *causal inference* aims at providing methods to determine the causal model, that is the DAG and the joint-probability distributions entering for a given scenario. Different combinations of the $I_j$ correspond to different strategies. If all the interventions are set to idle, and hence all the outcomes are determined by the causal parents, one has the purely [observational]{} approach. In multivariate scenarios, where more than two random variables are involved, the observation of the joint probability distribution alone can still contain hints of the causal structure based on conditional independencies [@Pearl2009a]. Nevertheless, in the bivariate scenario, i.e. when only two random variables are involved, classical correlations obtained by observations do not comprise any causal information. Only if assumptions for example on the noise distribution are taken a priori, information on the causal model can be obtained from observational data [@Mooij2016].
Quantum causal inference
------------------------
The notion of causal models does not easily translate to quantum mechanics. The main problem is that in quantum systems not all observables can have predefined values independent of observation. Similiar to an operational formulation of quantum mechanics [@Chiribella2011], the *process matrix formalism* was introduced [@Oreshkov] and a quantum version of an *event* defined. In [@Costa2016] this is reviewed for the purpose of causal models. In place of the random variables in the classical case there are local *laboratories*. Within a process each laboratory obtains a quantum system as input and produces a quantum system as output. A quantum event corresponds to information which is obtained within a laboratory and is associated with a *completely positive* (CP) map mapping the input Hilbert space to the output Hilbert space of the laboratory. The possible events depend on the choice of *instrument*. An instrument is a set of CP maps that sum to a *completely positive trace preserving* (CPTP) map. For example an instrument can be a projective measurement in a specific basis, with the events the possible outcomes. The possibility to choose different instruments mirrors the possibility of interventions in the classical case [@Costa2016 3.3]. The whole information about mechanisms, which are represented as CPTP maps, and the causal connections is contained in a so-called *process matrix*. Besides its analogy for a classical causal model, the process framework goes beyond classical causal structures as it does not assume such a fixed causal structure [@Oreshkov]. This recently stirred a lot of research [@Oreshkov2016; @Procopio2015; @Chiribella2012; @Guerin]. For a more detailed introduction we refer the reader especially to reference [@Costa2016] where a comprehensive description is provided.\
The analogue of causal inference in the classical case is the reconstruction of a process matrix. This can be done using informationally complete sets of instruments, theoretically described in [@Costa2016 4.1] and experimentally implemented in [@Ried2015]. Defining a *quantum observational scheme* in analogy to the classical one is not straight forward. In general a quantum measurement destroys much of the states’ character and hence can almost never be considered a passive observation. For example if the system was initially in a pure state $\ket{\psi}$ but one measures in a basis such that $\ket{\psi}$ is not an eigenstate of the projectors onto the basis states, then the measurement truly changes the state of the system and the original state is not reproduced in the statistical average. In [@Costa2016 sect. 5] an observational scheme is simply defined as projective measurements in a fixed basis, in particular without assumptions about the incoming state of a laboratory and thus without assumptions about the underlying process. Another possibility to define an observational scheme is based on the idea that in the classical world observations reveal pre-existing properties of physical systems and that quantum observations should reproduce this. As a consequence, if one mixes the post-measurement states with the probabilities of the corresponding measurement outcomes, one should obtain the same state as before the measurement. That is ensured if and only if operations that do not destroy the quantum character of the state are allowed, as coherences cannot be restored by averaging. Ried et al. [@Ried2015] formalized this notion as “informational symmetry”, but considered only preservation of local states. For the special case of locally completely mixed states, they showed that projective measurements in arbitrary bases possess informational symmetry. This definition of a quantum observational scheme is problematic due to two reasons: Firstly, the allowed class of instruments depends on the incoming state, i.e. one can only apply projective measurements that are diagonal in the same basis as the state itself. This is at variance with the typical motivation for an observational scheme, namely that the instruments are restricted a-priori due to practical reasons. Moreoever, having measurements depend on the state requires prior knowledge about the state of the system, but finding out the state of the system is part of the causal inference (e.g.: are the correlations based on a state shared by Alice and Bob?). Hence, in general one cannot assume sufficient knowledge of the state for restricting the measurements such that they do not destroy coherences.\
Secondly, the definition is unnaturally restrictive as it only considers the local state and not the global state. For example if Alice and Bob share a singlet state $\ket{\psi} = \frac{\ket{01}-\ket{10}}{\sqrt{2}}$, then both local states are completely mixed. Hence according to the informational symmetry, they are allowed to perform projective measurements in arbitrary bases. If Alice and Bob now both measure in the computational basis, they will each obtain both outcomes with probability $1/2$ and their local states will remain invariant in the statistical average $\rho_A' =
\rho_A = \frac{\mathbb{1}}{2}=\rho_B' =\rho_B$. However, the global state does not remain intact. The post-measurement state is given as $\rho_{AB}' = \frac{1}{2} \left( \ket{01}\bra{01} +
\ket{10}\bra{10}\right)$ which is not even entangled anymore. But even defining a “global informational symmetry”, i.e. requiring the global state to remain invariant, does not settle the issue in a convenient way, as this would not allow any local measurements of Alice and Bob.\
Here we propose three different schemes ranging from full quantum interventions over a quantum-observational scheme with the possibility of an active choice of measurements, to a passive quantum observational scheme in a fixed basis that comes closest to the classical observational scheme.
[[|c||>p[2.5cm]{}|>p[2.5cm]{}|>p[2.5cm]{}||>p[2.5cm]{}|>p[2.5cm]{}|]{}]{} & arbitrary & arbitrary projections & fixed basis projection &signaling& causal inference\
Q-interventionist & $\surd$ & $\surd$ &$\surd$ &$\surd$ & $\surd$\
Active Q-observational & X & $\surd$ & $\surd$ &$\surd^1$ & $(\surd)^2$\
Passive Q-observational & X&X& $\surd$&X&X$^3$\
The definitions are based on restricting the allowed set of instruments. An instrument is to be understood in the process-matrix context. In all three schemes the set of allowed instruments is independent of the actual underlying processes, which is a reasonable assumption, since the motivation for causal inference comes from the fact that states or processes are not known in the first place.
- **Quantum interventionist scheme:** Arbitrary instruments can be applied in local laboratories. These include for example deterministic operations such as state preparations or simply projective measurements. An appropriate choice of the instruments enables one to detect causal structure in arbitrary scenarios, i.e. to reconstruct the process matrix [@Costa2016]. This scheme resembles most closely an interventionist scheme in a classical scenario but offers additional quantum-mechanical possibilities of intervention.
- **Active quantum-observational scheme:** Only projective measurements in arbitrary orthogonal bases are allowed, but no post-processing of the state after the measurement. The latter request translates the idea of not intervening in the quantum realm, as it is not possible to deterministically change the state by the experimenters choice. Depending on the state and the instrument, the state may change during the measurement, hence the scheme is invasive, but the difference to the classical observational scheme arises solely from the possible destruction of quantum coherences. This is a quantum effect without classical correspondence and hence opens up a new possibility of defining an observational scheme that has no classical analogue. Repetitive application of the same measurement within a single run always gives the same output. Furthermore, we allow projective measurements in different bases in different runs of the experiment. This freedom allows one to completely characterize the incoming state.\
This scheme allows for signaling, i.e. there exist processes for which Alice’s choice of instrument changes the statistics that Bob observes. As an example consider the process, where Alice always obtains a qubit in the state $\ket{1}$. She applies her instrument on it, and then the outcome is propagated to Bob by the identity channel. Bob measures in the basis where $\ket{1}$ is an eigenstate. If Alice measured in the same basis as Bob, then both of them deterministically obtain 1 as result. If Alice instead measures in the basis $\left\{\ket{\pm} = \frac{1}{\sqrt{2}}
(\ket{0}\pm\ket{1})\right\}$, then Bob would obtain 1 only with probability $\frac{1}{2}$. This is considered as signaling according to the definition in [@Costa2016]. Clearly, signaling presents a direct quantum advantage for causal inference compared to a classical observational scheme, and motivates the attribute “active” of the scheme. In the present work we focus on this scheme, but exclude such a direct quantum advantage by considering exclusively unital channels and a completely mixed incoming state for Alice, as was done also in [@Ried2015]. It is then impossible for Alice to send a signal to Bob if her instruments are restricted to quantum observations, even if she is allowed to actively set her measurement basis. One might wonder whether the quantum-observational scheme can be generalized to POVM measurements. However, these do not fit into the framework of instruments that transmit an input state to an output state, as POVM measurements do not specify the post-measurement state.
- **Passive quantum-observational scheme:** For the whole setup a fixed basis is selected. Only projective measurements with respect to this basis are permitted, and it is forbidden to change the basis in different runs of the experiment. This is also what is used in [@Costa2016] to obtain classical causal models as a limit of quantum causal models. Since the basis is fixed independently of the underlying process, the measurement can still be invasive in the sense that it can destroy coherences, and hence it is still not a pure observational scheme in the classical sense. Nevertheless, Alice cannot signal to Bob here as she has no possibility of actively encoding information in the quantum state, regardless of the nature of the state, which motivates the name “passive quantum-observational scheme”. As without any change of basis it is impossible to exploit stronger-than-classical quantum correlations, this scheme comes closest to a classical observational scheme. And due to the restriction to observing at most classical correlations, it is not possible to infer anything more about the causal structure than classically possible.
Affine representation of quantum channels and steering maps {#sec:positiveMaps}
===========================================================
In this section we introduce the tools of quantum information theory that we need to analyze the problem of causal inference in section \[sec:Results\].
Bloch-sphere representation of qubits
-------------------------------------
A qubit is a quantum system with a two-dimensional Hilbert space with basis states denoted as $\ket{0}$ and $\ket{1}$. An arbitrary state of the qubit is described by a density operator $\rho$, a positive linear operator with unit trace, $\rho \geq 0, \; {\ensuremath{\text{tr}\left[\rho\right]}} =
1$. Every single-qubit state can be represented geometrically by its *Bloch-vector* $\boldsymbol{r} = {\ensuremath{\text{tr}\left[\rho
\boldsymbol{\sigma}\right]}}$, with $|\boldsymbol{r}| \leq 1$ as $$\begin{aligned}
\rho = \frac{\mathbb{1}+\boldsymbol{r}\cdot \boldsymbol{\sigma}}{2},\end{aligned}$$ where $\boldsymbol{\sigma} = (\sigma_1, \sigma_2, \sigma_3)^T$ denotes the vector of Pauli matrices.
Channels
--------
A quantum channel $\mathcal{E}$ is a *completely positive trace preserving map* (CPTP map). A quantum channel maps a density operator in the space of linear operators $\rho \in \mathcal{L}(\mathcal{H})$ on the Hilbert space $\mathcal{H}$ to a density operator in the space of linear operators $\rho' \in \mathcal{L}(\mathcal{H}')$ on a (potentially different) Hilbert space $\mathcal{H}'$. $$\begin{aligned}
\mathcal{E} : \rho \rightarrow \mathcal{E}(\rho) \equiv \rho', \qquad \rho, \rho' \geq 0, \; {\ensuremath{\text{tr}\left[\rho\right]}}= {\ensuremath{\text{tr}\left[\rho'\right]}} = 1.\end{aligned}$$ This formalism describes any physical dynamics of a quantum system. Every quantum channel can be understood as the unitary evolution of the system coupled to an environment [@Nielsen2009]. The constraint of complete positivity can be understood the following way. If we extend the map $\mathcal{E}$ with the identity operation of arbitrary dimension, the composed map $\mathcal{E} \otimes
\mathbb{1}$, which acts on a larger system, should still be positive. An example of a map that is positive but not completely positive is the transposition map, that, if extended to a larger system, maps entangled states to non-positive-semi-definite operators [@Bengtsson2006 chapter 11.1].\
*Geometrical representation of qubit maps*\
Every qubit channel (a quantum channel mapping a qubit state onto a qubit state) $\mathcal{E}$ can be described completely by its action on the Bloch sphere, see [@Fujiwara1999; @Braun2014; @BethRuskai2002] and is completely described by the matrix $\Theta_\mathcal{E}$ mapping the 4D Bloch vector $(1,\boldsymbol{r})$, $$\begin{aligned}
\Theta_\mathcal{E} = \begin{pmatrix}
1 & 0 \\
\boldsymbol{t_\mathcal{E}} & T_\mathcal{E}
\end{pmatrix},\end{aligned}$$ where the upper left 1 ensures trace preservation. A state $\rho$ described by its Bloch vector $\boldsymbol{r}$ is then mapped by the quantum channel $\mathcal{E}$ to the new state $\rho'$ with Bloch vector $$\begin{aligned}
\boldsymbol{r}' = T_\mathcal{E} \boldsymbol{r} +\boldsymbol{t_\mathcal{E}}.\end{aligned}$$ A qubit channel is called *unital* if it leaves the completely mixed state invariant: $\mathcal{E}(\rho_\text{mixed}) = \rho_\text{mixed}$, with $\rho_\text{mixed} = \frac{\mathbb{1}}{2}$, i.e. $\boldsymbol{r}_\text{mixed} = \boldsymbol{0}$. For unital channels $\boldsymbol{t}_\mathcal{E}$ vanishes. The whole information is then contained in the 3x3 real matrix $T_\mathcal{E}$, which we refer to as *correlation matrix* of the channel. The matrix $T$ (from now on we drop the index $\mathcal{E}$) can be expressed by writing it in its signed singular value decomposition [@Braun2014 eq. (9)], [@Bengtsson2006 eq. (10.78)] (see also the appendix around equation ), $$\begin{aligned}
T =R_1 \eta R_2. \label{eq:signed_singulars}\end{aligned}$$ Here, $R_1$ and $R_2$ are proper rotations (elements of the $SO(3)$ group), corresponding to unitary channels, that is $R_iR_i^T =
\mathbb{1}$ with $\det(R_i)=1$, and $\eta= {\text{diag}}(\eta_1,\eta_2,
\eta_3)$ is a real diagonal matrix. This can be interpreted rather easily. A unital qubit channel maps the Bloch sphere onto an ellipsoid, centered around the origin, that fits inside the Bloch sphere. First the Bloch sphere is rotated by $R_2$ than it is compressed along the coordinate axis by factors $\eta_i$. The resulting ellipsoid is then again rotated. Hence, apart from unitary freedom in the input and output, the unital quantum channel is completely characterized by its *signed singular values* (SSV) $\bm \eta$ [@Braun2014 II.B]. The CPTP property gives restrictions to the allowed values of $\boldsymbol{\eta} \equiv (\eta_1,\eta_2, \eta_3)^T$. These are commonly known as the *Fujiwara-Algoet* conditions [@Fujiwara1999; @Braun2014] $$\begin{aligned}
\begin{aligned}
1+\eta_3 \geq |\eta_1+\eta_2|,\\
1-\eta_3 \geq |\eta_1-\eta_2|.
\end{aligned}\label{eq:Fujiwara}\end{aligned}$$ The allowed values for $\boldsymbol{\eta}$ lie inside a tetrahedron $\mathcal{T}_\text{CP}$ (the index CP stands for completely positive), $$\begin{aligned}
\mathcal{T}_\text{CP} \equiv \text{Conv}\left(\left\lbrace\boldsymbol{v}^\text{CP}_i \right\rbrace_i\right), \label{def:T_CP}\end{aligned}$$ where $\text{Conv}\left(\left\lbrace x_i\right\rbrace_i\right) \equiv \left\lbrace \sum_i p_i x_i | p_i\geq 0, \sum_i p_i =1\right\rbrace$ denotes the convex hull of the set $\left\{x_i\right\}_i$ and the vertices are defined as, $$\begin{aligned}
\begin{aligned}\label{eq:CP_vertices}
\boldsymbol{v}^\text{CP}_1 &= (1,1,1)^T,\\
\boldsymbol{v}^\text{CP}_2 &= (-1,-1,1)^T,\\
\boldsymbol{v}^\text{CP}_3 &= (-1,1,-1)^T,\\
\boldsymbol{v}^\text{CP}_4 &= (1,-1,-1)^T.
\end{aligned}\end{aligned}$$
For a more detailed discussion of qubit maps we refer the reader to chapter 10.7 of [@Bengtsson2006].
Steering
--------
In quantum mechanics, measurement outcomes on two spatially separated partitions of a composed quantum system can be highly correlated [@Bell1964], and further the choice of measurement operator on one side can strongly influence or even determine the outcome on the other side [@Schrodinger1935], a phenomenon known as “steering”. Suppose Alice and Bob share the two qubit state $\rho_{AB}$. If Alice performs a measurement on it, leaving her qubit in the state $\rho_A$ then Bob’s qubit is steered to the state $\rho_B$ proportional to the (unnormalized) state $ {\ensuremath{\text{tr}_{A}\left[\rho_{AB} (\rho_A
\otimes \mathbb{1})\right]}}$ [@Jevtic p.2]. This defines a positive linear trace preserving map $\mathcal{S}: \rho_A \rightarrow
\mathcal{S}(\rho_A) = \rho_B$, called *steering map*, that depends on the state $\rho_{AB}$.\
Steering maps have been intensely studied especially in terms of entanglement characterization [@Jevtic; @Milne2014]. In analogy to the treatment of qubit channels, we can associate an unique ellipsoid inside the Bloch sphere with a two-qubit state, known as steering ellipsoid, that encodes all the information about the bipartite state [@Jevtic].\
Every bipartite two qubit state can be expanded in the Pauli basis as $$\begin{aligned}
\rho_{AB} = \frac{1}{4} \sum_{\mu, \nu =0}^3 \Theta_{\mu \nu} \sigma_\nu \otimes \sigma_\mu,\end{aligned}$$ where $$\begin{aligned}
\Theta_{\mu\nu} = {\ensuremath{\text{tr}\left[\rho_{AB} \sigma_\nu \otimes \sigma_\mu\right]}}.\end{aligned}$$ Note that we defined $\Theta$ to be the transposed of the one defined in [@Jevtic], since we want to treat steering from Alice to Bob. The matrix contains all the information about the bipartite state and can be written as $$\begin{aligned}
\Theta = \begin{pmatrix}
1 & \boldsymbol{a}^T \\
\boldsymbol{b} & T_\mathcal{S}
\end{pmatrix},\end{aligned}$$ where $\boldsymbol{a}$ ($\boldsymbol{b}$) denotes the Bloch vector of Alice’s (Bob’s) reduced state. $T_\mathcal{S}$ is a 3x3 real orthogonal matrix and encodes all the information about the correlations, and we will refer to it as *correlation matrix* of the steering map.\
In this work we only consider bipartite qubit states which have completely mixed reduced states ${\ensuremath{\text{tr}_{A}\left[\rho_{AB}\right]}} = {\ensuremath{\text{tr}_{B}\left[\rho_{AB}\right]}} =
\mathbb{1}/2$ or equivalently $\boldsymbol{a}=
\boldsymbol{b}=\boldsymbol{0}$. In analogy to unital channels we call such states *unital two-qubit states* and the corresponding maps *unital steering maps*. Up to local unitary operations on the two partitions, the correlation matrix $T_\mathcal{S}$ is characterized by its signed singular values $\eta_1,\eta_2,\eta_3$. The allowed values of these are given through the positivity constraint on the density operator $\rho_{AB}$ defined up to local unitaries as (cf. equation (6) in [@Milne2014]) $$\begin{aligned}
\rho_{AB} = \frac{1}{4}\left(\mathbb{1}\otimes \mathbb{1} + \sum_{i=1}^3 \eta_i \sigma_i\otimes \sigma_i\right). \end{aligned}$$ The positivity of $\rho_{AB}$ implies the conditions (the derivation is analogue to the derivation of (10)-(15) in [@Braun2014]) $$\begin{aligned}
\begin{aligned}
1+\eta_3 \geq |\eta_1-\eta_2|,\\
1-\eta_3 \geq |\eta_1+\eta_2|.
\end{aligned}\end{aligned}$$ These are the same as for unital qubit channels (eq. ) up to a sign flip, and define the tetrahedron $\mathcal{T}_\text{CcP}$ of unital *completely co-positive trace preserving maps* (CcPTP) [@Bengtsson2006; @Braun2014], $$\begin{aligned}
\mathcal{T}_\text{CcP} \equiv \text{Conv}\left(\left\lbrace\boldsymbol{v}^\text{CcP}_i \right\rbrace_i\right), \label{def:T_CcP}\end{aligned}$$ with the vertices $$\begin{aligned}
\label{eq:CcP_vertices}
\begin{aligned}
\boldsymbol{v}^\text{CcP}_1 &= (-1,-1,-1)^T,\\
\boldsymbol{v}^\text{CcP}_2 &= (-1,1,1)^T,\\
\boldsymbol{v}^\text{CcP}_3 &= (1,-1,1)^T,\\
\boldsymbol{v}^\text{CcP}_4 &= (1,1,-1)^T.
\end{aligned}\end{aligned}$$
CcPTP maps are exactly CPTP maps with a preceding transposition map, i.e. for every steering map $\mathcal{S}$ there exists a quantum channel $\mathcal{E}$ such that $\mathcal{S} = \mathcal{E}\circ
\mathcal{T}$, where $\mathcal{T}$ is the transposition map with respect to an arbitrary but fixed basis (see e.g. [@Bengtsson2006]).
[0.3]{} ![**Geometry of positive maps**: For positive trace preserving single-qubit maps, the allowed signed singular values lie within a cube $\mathcal{C}$ defined in . Quantum channels corresponding to CPTP maps lie within the blue tetrahedron $\mathcal{T}_\text{CP}$ defined in , steering maps corresponding to CcPTP maps lie within the yellow tetrahedron $\mathcal{T}_\text{CcP}$ defined in . The maps with SSV inside the intersection of $\mathcal{T}_\text{CP}$ and $\mathcal{T}_\text{CcP}$ (green octahedron) are called superpositive. These maps only produce classical correlations corresponding to separable states or entanglement breaking channels, but can also be generated by mixtures of quantum correlations. []{data-label="Fig:SSV"}](completelypos2.pdf "fig:")
[0.3]{} ![**Geometry of positive maps**: For positive trace preserving single-qubit maps, the allowed signed singular values lie within a cube $\mathcal{C}$ defined in . Quantum channels corresponding to CPTP maps lie within the blue tetrahedron $\mathcal{T}_\text{CP}$ defined in , steering maps corresponding to CcPTP maps lie within the yellow tetrahedron $\mathcal{T}_\text{CcP}$ defined in . The maps with SSV inside the intersection of $\mathcal{T}_\text{CP}$ and $\mathcal{T}_\text{CcP}$ (green octahedron) are called superpositive. These maps only produce classical correlations corresponding to separable states or entanglement breaking channels, but can also be generated by mixtures of quantum correlations. []{data-label="Fig:SSV"}](ccpos2.pdf "fig:")
[0.3]{} ![**Geometry of positive maps**: For positive trace preserving single-qubit maps, the allowed signed singular values lie within a cube $\mathcal{C}$ defined in . Quantum channels corresponding to CPTP maps lie within the blue tetrahedron $\mathcal{T}_\text{CP}$ defined in , steering maps corresponding to CcPTP maps lie within the yellow tetrahedron $\mathcal{T}_\text{CcP}$ defined in . The maps with SSV inside the intersection of $\mathcal{T}_\text{CP}$ and $\mathcal{T}_\text{CcP}$ (green octahedron) are called superpositive. These maps only produce classical correlations corresponding to separable states or entanglement breaking channels, but can also be generated by mixtures of quantum correlations. []{data-label="Fig:SSV"}](superpos2.pdf "fig:")
Positive maps
-------------
We have seen that a quantum channel is a CPTP map and that a steering map is a CcPTP map. Both of them are necessarily positive maps. But are there positive maps that are neither CcP nor CP? Or are there maps that are even both? This issue is nicely worked out in [@Bengtsson2006 chapter 11]. We shortly review this for unital qubit maps. Since we still deal with linear maps, it is straightforward that also every unital positive one-qubit map can be described by a $3\times 3$ correlation matrix. Hence we can also analyze its SSV. The allowed SSV are inside the cube $\mathcal{C}$ defined by [@Bengtsson2006 FIG.11.3] $$\begin{aligned}
\mathcal{C} \equiv \left\lbrace \boldsymbol{x}| -1\leq x_i \leq 1 \text{ for } i= 1,2,3\right\rbrace. \label{def:C}\end{aligned}$$ This is illustrated in FIG.\[Fig:SSV\]. Note again that we only treat unital maps.\
We see that there are positive maps which are neither CP nor CcP. According to the *St[ø]{}rmer-Woronowicz theorem* (see e.g. [@Bengtsson2006 p. 258]) every positive qubit map is *decomposable*, i.e. it can be written as a convex combination of a CP and a CcP map. Maps that are both CP and CcP are called *super positive* (SP). The set of allowed SSV of the correlation matrices of these maps forms an octahedron (green region in FIG.\[Fig:SSV\]) given as $$\begin{aligned}
\mathcal{O}_\text{SP} = \text{Conv}\left(\{\pm \hat{e}_i | i \in {x,y,z}\}\right), \label{eq:octahedron}\end{aligned}$$ where $\hat{e}_i$ denotes the unit vector along the $i$-axis. These correlations are generated by entanglement breaking quantum channels [@Ruskai2003] and steering maps based on separable states [@Jevtic]. When such *classical* correlations are observed one cannot infer anything about the causal structure [@Ried2015 p.10 of supplementary information].\
For higher dimensional systems things change. Already for three dimensional maps, i.e. qutrit maps, there exist positive maps, that cannot be represented as a convex combination of a CP and a CcP map [@Bengtsson2006 chapter 11.1]. In the next section we discuss how much information about causal influences we can obtain by looking only at the SSV related to the correlations Alice and Bob can observe in a bipartite experiment.
Causal explanation of unital positive maps {#sec:Results}
==========================================
Setting
-------
We now tackle the problem of causal inference in the two-qubit scenario [@Ried2015]. The setting is as follows. An experimenter, Alice, sits in her laboratory. She opens her door just long enough to obtain a qubit in a (locally) completely mixed state and closes the door again. She performs an projective measurement in any of the Pauli-states, records her outcome, opens her door again and puts the qubit in the now collapsed state outside. Apart from the qubit she has no way of interacting with the environment. Some time later another experimenter, Bob, opens the door of his laboratory and obtains a qubit. Also he measures in the eigenbasis of one of the Pauli matrices and records the outcome. They repeat this procedure a large (ideally: an infinite) number of times. Then they meet and analyze their joint measurement outcomes. These define the probabilities $P(a, b| j, i)$ for the outcomes $a\in\{-1,1\}$ and $b\in\{-1,1\}$ of Alice’s and Bob’s measurements, given they measured in the eigenbasis of the $j$th and $i$th Pauli matrix, respectively. For the marginals we assume $P(a|j,i) = \sum_b P(a,b|j,i) = 1/2 \,\, \forall a\in \{-1,1\}$ and accordingly for Bob. They are thus able to define a correlation matrix $M$ with elements $$\begin{aligned}
M_{ij} = 2 P(b=1|j,i, a=1)-1 = \braket{\sigma_j \sigma_i},\end{aligned}$$ where $P(b=1|j,i, a=1)$ is the probability that Bob obtains outcome $1$ when measuring the observable $\sigma_i$, conditioned on Alice’s measurement of $\sigma_j$ with outcome $1$, and $\braket{\sigma_j\sigma_i}$ denotes the expectation value of the product of Alice’s $\sigma_j$ and Bob’s $\sigma_i$ measurement outcomes.\
The correlation matrix defines a unique positive trace preserving unital map $\mathcal{M}: \rho_A \mapsto \rho_B$. They are guaranteed one of the following three possibilities: either they measured the same qubit, which was propagated in terms of a unital quantum channel $\mathcal{E}$ from Alice to Bob; or that they each measured one of the two qubits in a unital bipartite state $\rho_{AB}$ acting as a common cause, and hence the correlations where caused by the corresponding steering map $\mathcal{S}$; or that the map from $\rho_A$ to $\rho_B$ is a probabilistic mixture where with probability $p$ the steering map $\mathcal{S}$ was realized and with probability $(1-p)$ the quantum channel $\mathcal{E}$, that is $$\begin{aligned}
\mathcal{M} = (1-p) \mathcal{E} + p \mathcal{S}, \label{eq:decomposition}\end{aligned}$$ with the “causality parameter” $p \in [0,1]$.
The task of Alice and Bob is now to find the true value of $p$ and possibly also the nature of $\mathcal{S}$ and $\mathcal{E}$. In general there does not exist a unique solution and in this case they want to find the values of $p$ for which maps of the form explain the observed correlations.\
As we mentioned in the previous section, every positive one qubit map is decomposable, so a possible explanation always exists. The decomposition can be given a causal interpretation, where $\mathcal{E}$ is considered to be a cause-effect explanation of the correlations and $\mathcal{S}$ a common-cause.\
In the following subsections we give bounds on the causality parameter $p$ and then consider some extremal cases. In subsection \[sec:AdditionalConst\] we generalize a part of the work of Ried et al. [@Ried2015] and see how additional assumptions on the nature of $\mathcal{E}$ and $\mathcal{S}$ can lead to a unique solution.
Possible causal explanations
----------------------------
**$p$-causality/$p$-decomposability:** A single qubit unital positive trace preserving map $\mathcal{M}$ is called $p$-causal/$p$-decomposable with $p \in [0,1]$, if it can be written as $$\begin{aligned}
\mathcal{M} = (1-p) \mathcal{E} + p \mathcal{S}, \label{def.pdec}\end{aligned}$$ with $\mathcal{E}$ ($\mathcal{S}$) being a CPTP (CcPTP) unital qubit map. Eq. is called a $p$-decomposition of $\mathcal{M}$.
In the following let $M, E, S$ denote the correlation matrices of $\mathcal{M}, \mathcal{E}, \mathcal{S}$, and $\boldsymbol{\eta}^\mathcal{M}, \boldsymbol{\eta}^\mathcal{E}, \boldsymbol{\eta}^\mathcal{S}$ the SSV of $M,E,S$, respectively. We first investigate for a fixed $p$ what the possible SSV of the correlation matrix of a map $\mathcal{M}$ are, such that $\mathcal{M}$ is $p$-causal. This leads to the following theorem:
**Signed singular values of $p$-causal maps**\[theo:p-causal\]\
Let $\mathcal{M}$ be a positive unital trace preserving qubit map with associated SSV given by $\boldsymbol{\eta}^\mathcal{M}$. Let $p \in
[0,1]$ be fixed. Then the following statement holds: $$\begin{aligned}
\mathcal{M}\text{ is $p$-causal} \Leftrightarrow \boldsymbol{\eta}^\mathcal{M} \in \mathcal{C}_p,\end{aligned}$$ where $$\begin{aligned}
\mathcal{C}_p = \text{Conv}\left( \left\{ (1-p) \boldsymbol{v}_i^\text{CP} + p \boldsymbol{v}_j^\text{CcP}| i,j \in\{1,2,3,4\}\right\} \right), \label{def:C_p}\end{aligned}$$ where the vertices $\boldsymbol{v}_i^\text{CP}$ of CP maps are given in , and the vertices $\boldsymbol{v}_j^\text{CcP}$ of CcP maps in .
“$\Leftarrow$”: From we see that $$\begin{aligned}
\boldsymbol{\eta}^\mathcal{M} \in \mathcal{C}_p \Leftrightarrow \exists \left(p_{ij} \geq 0, \;\sum_{i,j=1}^4 p_{ij} =1\right): \boldsymbol{\eta}^\mathcal{M}=\sum_{i,j=1}^4 p_{ij} \left((1-p) \boldsymbol{v}_i^\text{CP} + p \boldsymbol{v}_j^\text{CcP}\right).\end{aligned}$$ Now define $q_i \equiv \sum_j p_{ij}$ and $r_j \equiv \sum_i p_{ij}$. Clearly $q_i, r_j \geq 0$ and $\sum_i q_i = \sum_j r_j = 1$. We can then write $$\begin{aligned}
\boldsymbol{\eta}^\mathcal{M} = (1-p) \sum_i q_i \boldsymbol{v}_i^\text{CP} + p \sum_j r_j \boldsymbol{v}_j^\text{CcP} = (1-p)\, \boldsymbol{\eta}^\mathcal{E} + p \,\boldsymbol{\eta}^\mathcal{S},\end{aligned}$$ with $\boldsymbol{\eta}^\mathcal{E} \equiv \sum_i q_i
\boldsymbol{v}_i^\text{CP} \in \mathcal{T}_\text{CP} $ and $
\boldsymbol{\eta}^\mathcal{S} \equiv \sum_j r_j \boldsymbol{v}_j^\text{CcP} \in \mathcal{T}_\text{CcP}$. We herewith explicitly constructed a $p$-decomposition of $\mathcal{M}$ where the correlation matrices of $\mathcal{E}$ and $\mathcal{S}$ have their SSV-decomposition involving the same rotations as the SSV-decomposition of the correlation matrix of $\mathcal{M}$.\
“$\Rightarrow$”: Let $p$ be fixed. Suppose that $\mathcal{E}$ and $\mathcal{S}$ are both extremal maps, i.e. $\boldsymbol{\eta}^\mathcal{E}$ and $\boldsymbol{\eta}^\mathcal{S}$ are given by one of the vertices defined in and , respectively, and without loss of generality we assume that these are $\boldsymbol{v}_1^\text{CP}$ and $\boldsymbol{v}_1^\text{CcP}$ (this is justified as taking another vertex leads to the same result). Define $A = (1-p) E$ and $B= p S$, where $A$ has SSV $(1-p,1-p,1-p)$ and B has SSV $(-p,-p,-p)$. In the Appendix we prove theorem \[theo:SSV\] that restricts the possible SSV of $A+B$. For our case it gives $$\begin{aligned}
SSV(M) \in \mathcal{C}_p.\end{aligned}$$ Now suppose $\mathcal{E}$ and $\mathcal{S}$ are not extremal maps. Since the SSV of those are simply convex combinations of the SSV of the extremal maps, it follows that also for such maps the signed singular values of $M$ lie within $\mathcal{C}_p$.
\
[0.4]{} ![**Signed singular values of $p$-causal maps**: Set of attainable vectors of signed singular values associated with $\mathcal{M}$ in for different values of $p$. By theorem \[theo:p-causal\], for fixed $p$ there exists a CPTP map $\mathcal{E}$ and a CcPTP map $\mathcal{S}$ such that $\mathcal{M}$ is given by if and only if the vector of signed singular values $\boldsymbol{\eta}^\mathcal{M}$ of the correlation matrix of $\mathcal{M}$ is in $\mathcal{C}_p$ defined in .[]{data-label="Fig:SSV_mixture"}](mixed_0.pdf "fig:")
[0.4]{} ![**Signed singular values of $p$-causal maps**: Set of attainable vectors of signed singular values associated with $\mathcal{M}$ in for different values of $p$. By theorem \[theo:p-causal\], for fixed $p$ there exists a CPTP map $\mathcal{E}$ and a CcPTP map $\mathcal{S}$ such that $\mathcal{M}$ is given by if and only if the vector of signed singular values $\boldsymbol{\eta}^\mathcal{M}$ of the correlation matrix of $\mathcal{M}$ is in $\mathcal{C}_p$ defined in .[]{data-label="Fig:SSV_mixture"}](mixed_0,25.pdf "fig:")
[0.4]{} ![**Signed singular values of $p$-causal maps**: Set of attainable vectors of signed singular values associated with $\mathcal{M}$ in for different values of $p$. By theorem \[theo:p-causal\], for fixed $p$ there exists a CPTP map $\mathcal{E}$ and a CcPTP map $\mathcal{S}$ such that $\mathcal{M}$ is given by if and only if the vector of signed singular values $\boldsymbol{\eta}^\mathcal{M}$ of the correlation matrix of $\mathcal{M}$ is in $\mathcal{C}_p$ defined in .[]{data-label="Fig:SSV_mixture"}](mixed_0,75.pdf "fig:")
[0.4]{} ![**Signed singular values of $p$-causal maps**: Set of attainable vectors of signed singular values associated with $\mathcal{M}$ in for different values of $p$. By theorem \[theo:p-causal\], for fixed $p$ there exists a CPTP map $\mathcal{E}$ and a CcPTP map $\mathcal{S}$ such that $\mathcal{M}$ is given by if and only if the vector of signed singular values $\boldsymbol{\eta}^\mathcal{M}$ of the correlation matrix of $\mathcal{M}$ is in $\mathcal{C}_p$ defined in .[]{data-label="Fig:SSV_mixture"}](mixed_1.pdf "fig:")
We have seen that for a given value of $p$ the allowed SSV associated with a positive map $\mathcal{M}$ that is $p$-causal lie within $\mathcal{C}_p$ given in . We now turn the task around and go back to the causal inference scenario. Given a positive map $\mathcal{M}$ we want to tell if we can bound the causality parameter $p$. We will do this based on the following definition:
**Causal interval $I_\mathcal{M}$**:\
For a given positive unital qubit map $\mathcal{M}$ we define the interval of possible causal explanations (for short: the causal interval) $I_\mathcal{M}$, such that $\mathcal{M}$ is $p$-causal if and only if $p\in I_\mathcal{M}$.
Since every qubit map is decomposable [@Bengtsson2006 p.258] the causal interval is always non empty, $I_\mathcal{M} \neq \emptyset$.
\[theo:causal\_interval\] Let $\mathcal{M}$ be a positive unital qubit map, with associated signed singular values $\boldsymbol{\eta}^\mathcal{M}$ (we assume $\eta^\mathcal{M}_i \geq 0$ for $i=1,2$). Then the causal interval of $\mathcal{M}$ is given by $$\begin{aligned}
p_\text{max} &= \min\left(\frac{3 - \boldsymbol{\eta}^\mathcal{M}\cdot \boldsymbol{v}_1^\text{CP}}{2}, 1\right) \label{eq:pmax},\\
p_\text{min} &= \max\left(\frac{\boldsymbol{\eta}^\mathcal{M}\cdot \boldsymbol{v}_4^\text{CcP}-1}{2},0\right),\label{eq:pmin}\end{aligned}$$ with $\boldsymbol{v}_1^\text{CP} = (1,1,1)^T$ ($\boldsymbol{v}_4^\text{CcP} = (1,1,-1)^T$) defining a vertex of the CPTP (CcPTP) tetrahedron $\mathcal{T}_{CP}$ ($\mathcal{T}_{CcP}$).
Note that the assumption $\eta^\mathcal{M}_i \geq 0$ for $i=1,2$ can always be met, using the unitary freedom in the decomposition in the right way.\
We show the theorem for $p_\text{max}$, the determination of $p_\text{min}$ can be treated in an analogue way.\
First we check if $\mathcal{M}$ is a CcPTP map, by checking if $\boldsymbol{\eta}^\mathcal{M} \in \mathcal{T}_\text{CcP}$. If it is CcPTP then $p_\text{max}=1$, trivially.\
![**Sketch for proof of theorem \[theo:causal\_interval\]**: The value of $p_\text{max}$ is determined through the projection of $\boldsymbol{\eta}^\mathcal{M}$ onto $\boldsymbol{v}_1^\text{CP}$, which is given by $\boldsymbol{u}$. The red triangle is one of the facets of $\mathcal{C}_{p_\text{max}}$.[]{data-label="Fig:Sketch"}](sketchCausalInterval.eps)
Now suppose it is not CcPTP. $p_\text{max}$ is then given such that $\boldsymbol{\eta}^\mathcal{M} \in \mathcal{C}_{p_\text{max}}$ but $\boldsymbol{\eta}^\mathcal{M} \notin \mathcal{C}_{p'}$ with $p' \in
(p_\text{max},1]$. This implies that $\boldsymbol{\eta}^\mathcal{M}$ lies on the surface of $\mathcal{C}_{p_\text{max}}$. Since we assumed $\eta^\mathcal{M}_i \geq 0$ for $i=1,2$, the critical facet of $\mathcal{C}_{p_\text{max}}$ is the one which is perpendicular to $\boldsymbol{v}^\text{CP}_1$ and has the vertices $(1,1,1-2p_\text{max})^T,(1,1-2p_\text{max},1)^T,
(1-2p_\text{max},1,1)^T$ (see FIG.\[Fig:Sketch\]). Since this facet is perpendicular to $\boldsymbol{v}^\text{CP}_1$, $\boldsymbol{\eta}^\mathcal{M}$ lies on this facet if its projection onto $\boldsymbol{v}_1^\text{CP}$ equals the vector pointing from the origin to the intersection of the facet and $\boldsymbol{v}^\text{CP}_1$, given as $\boldsymbol{u}\equiv(1-(2/3)\,p_\text{max})\boldsymbol{v}^\text{CP}_1$, see Fig.\[Fig:Sketch\]. Hence we get the following equation $$\begin{aligned}
&\frac{1}{3}\boldsymbol{v}^\text{CP}_1 (\boldsymbol{v}^\text{CP}_1
\cdot \boldsymbol{\eta}^\mathcal{M}) \overset{!}{=} \boldsymbol{u}
\\
&\Leftrightarrow \boldsymbol{v}^\text{CP}_1 \cdot \boldsymbol{\eta}^\mathcal{M} = 3 - 2p_\text{max}\\
&\Leftrightarrow p_\text{max} = \frac{3 - \boldsymbol{\eta}^\mathcal{M}\cdot \boldsymbol{v}^\text{CP}_1}{2}. \end{aligned}$$
Extremal cases {#sec:extremal}
--------------
In the previous section we found the general form of the causal interval $I_\mathcal{M}$ for an observed map $\mathcal{M}$. We now analyze the extremal cases where the interval reduces to a single value or on the other hand the interval is given as $I_\mathcal{M}=[0,1]$.\
As already noted in [@Ried2015 Table 1.] there are extremal cases that allow for a complete solution of the problem even without any additional constraints. This is the case if $\boldsymbol{\eta}^\mathcal{M}$ equals one of the vertices of the cube of positive maps, see Fig. \[Fig:SSV\]. The solution is then either $p=0$ (pure cause-effect) if the SSV are all positive or exactly two are negative or $p=1$ (pure common-cause) if the SSV are all negative or exactly one positive. The exact reconstruction of $\mathcal{E}$ or $\mathcal{S}$ in this cases is trivial.\
Interestingly, with theorem \[theo:causal\_interval\] we can show that *every* point on the edges of the cube $\mathcal{C}$ defined in gives us a unique solution without additional constraints:\
Let $\mathcal{M}$ be a positive map and $M$ be the corresponding correlation matrix with $M = R_1 \eta^\mathcal{M} R_2$ where $\eta^\mathcal{M} = {\text{diag}}(\boldsymbol{\eta}^\mathcal{M})$ with the signed singular values $\boldsymbol{\eta}^\mathcal{M} = (1,1,1-2p)^T,
\; p \in [0,1]$, and two rotations $R_1,R_2 \in SO(3)$. Due to the freedom in $R_1$ and $R_2$ this describes all maps with corresponding vector of SSV on one of the edges of the cube $\mathcal{C}$ defined in . According to theorem \[theo:causal\_interval\] we find $$\begin{aligned}
p_\text{max} &= \min\left(\frac{3 - \boldsymbol{\eta}^\mathcal{M}\cdot
\boldsymbol{v}_1}{2}, 1\right) = \frac{3- \left(2+(1-2p)\right)}{2}
= p,\\
p_\text{min} &= \max\left(\frac{\boldsymbol{\eta}^\mathcal{M}\cdot
\boldsymbol{v}_4^\text{CcP}-1}{2},0\right)=
\frac{2-(1-2p)-1}{2} = p. \end{aligned}$$ By theorem \[theo:SSV\] it follows, that the maps $\mathcal{E}$ and $\mathcal{S}$ in the decomposition necessarily correspond to extremal points in $\mathcal{T}_\text{CP}$ and $\mathcal{T}_\text{CcP}$ defined in and (unitary channel and maximally entangled state). It is then obvious that $$\begin{aligned}
M = R_1\left((1-p){\text{diag}}(\boldsymbol{v}^\text{CP}_1) + p\,{\text{diag}}(
\boldsymbol{v}^\text{CcP}_4)\right) R_2 \end{aligned}$$ is the only possible solution.
In the other extreme case, if the map $\mathcal{M}$ is superpositive, i.e. CP and CcP (see Figure \[Fig:SSV\]), it could be explained by a pure CPTP, a pure CcPTP map, or any convex combination of those two. Therefore one cannot give any restrictions of possible values of $p$ [@Ried2015 III.E of supplementary information].\
Let $\mathcal{M}$ be a superpositive map. There exists a SSV decomposition of its correlation matrix for which $\boldsymbol{\eta}^\mathcal{M} \in
\mathcal{O}_\text{SP}$, defined in , and for which $\eta^\mathcal{M}_i \geq 0$ for $i=1,2$. Hence we can write $\boldsymbol{\eta}^\mathcal{M} = p_1 \hat{e}_x + p_2 \hat{e}_y + p_3
\hat{e}_z + p_4 (-\hat{e}_z)$, with $\sum_i p_i =1$. The scalar product of each component of $\boldsymbol{\eta}^\mathcal{M} $ with $\boldsymbol{v}^\text{CP}_1 = (1,1,1)^T$ is upper bounded by 1. Hence we have $\boldsymbol{\eta}^\mathcal{M} \cdot
\boldsymbol{v}^\text{CP}_1 \leq 1$ and with that eq. evaluates to $p_\text{max} = 1$. Analogously one finds $p_\text{min}
=0$.
Additional assumptions / Causal inference with constrained classical correlations {#sec:AdditionalConst}
---------------------------------------------------------------------------------
So far we only assumed that our data is generated by a unital channel and a unital state (a state whose local partitions are completely mixed). We have seen that in some extreme cases a unique solution to the problem can be found. Ried et al. showed that one can always find a unique solution for $p$ if one restricts the channel to unitary channels and the bipartite states to maximally entangled pure states [@Ried2015]. Furthermore, it is then possible to reconstruct the channel and the state up to binary ambiguity, meaning there are two explanations leading to the same observed correlations. The ellipsoids associated with unitary channels and maximally entangled states are spheres with unit radius and the SSV of their correlation matrices correspond to the vertices of $\mathcal{T}_\text{CP}$ and $\mathcal{T}_\text{CcP}$ respectively .\
In the following we investigate this scenario again, but add a known amount of noise in the channel or in the bipartite state. For the channel this is done by mixing the unitary evolution with a completely depolarizing channel [@Nielsen2009]. The completely depolarizing channel maps every Bloch-vector to the origin, $\rho \mapsto \frac{\mathbb{1}}{2}$ and hence is represented by the zero matrix. The ellipsoid associated with the mixture of a completely depolarizing channel with a unitary channel thus results in a shrinked sphere. For strong enough noise the result eventually becomes an entanglement breaking channel, which only produces “classical” correlations [@Ruskai2003]. Due to the unitary freedom compared to standard depolarizing channels, we call these channels *generalized depolarizing channel*. For the state we mix a pure maximally entangled state with the completely mixed state, whose correlation matrix is given by the zero-matrix. We call the state a *generalized Werner state*, in the sense that instead of a convex combination of a singlet and a completely mixed state [@Werner1989] we allow the convex combination of an arbitrary maximally entangled state with the completely mixed state. States at a certain threshold of noise become separable and the correlations become “classical” [@Jevtic]. We will then see that even when confronted with purely classical correlations, if we have enough a-priori-knowledge about the data generation, *i.e.* we know the amount of noise, we can still find a solution analogous to [@Ried2015], in the sense of determining uniquely the parameter $p$, and the channel and the state up to binary ambiguity[^1]. We will first keep the unitary channel and start with a generalized Werner-state and show how one can recreate the scenario of Ried et al. Then we will add the noise in the channel.
### Solution of the causal inference problem using generalized Werner states
The analysis follows closely in spirit section III.D in the supplementary information in [@Ried2015]. We start again with equation and assume that the steering map $\mathcal{S}$ is generated by a shared generalized Werner state $\rho_{AB} = \epsilon \frac{\mathbb{1}}{4} + (1-\epsilon) \ket{\psi}\bra{\psi}$, where the parameter $\epsilon \in [0,1)$ is known and fixed in advance and $\ket{\psi}$ is an unknown maximally entangled pure state. The map $\mathcal{E}$ is generated by an unknown unitary channel $U$.\
Since $\epsilon$ is fixed, the class of allowed explanations is completely defined up to unitary freedom in the channel and in the state. Hence the number of free parameters is the same as in the case considered in [@Ried2015], which coincides with the case $\epsilon=0$. For $\epsilon > 2/3$ the state $\rho_{AB}$ becomes separable, i.e. is not entangled anymore, see [@Werner1989] and Fig.5 in the supplementary information of [@Jevtic]. But the reconstruction works independently of $\epsilon$. Hence, we see here that the possibility of reconstruction hinges not on the entanglement in $\rho_{AB}$ but on the prior knowledge we have about $\rho_{AB}$.\
The correlation matrix corresponding to the generalized Werner-state is simply the one of a maximally entangled state shrinked by a factor $1-\epsilon$ and will thus be denoted $(1-\epsilon) S$, where $S$ is the correlation matrix corresponding to a maximally entangled state. Thus in our scenario the information Alice and Bob obtain characterizes the matrix $$\begin{aligned}
M = p(1-\epsilon) S +(1-p) E. \label{eq:Mwerner}\end{aligned}$$ The ellipsoid is described by the eigenvalues and -vectors of $MM^T$. The eigenvectors correspond to the direction of the semi axes and the squareroots of the eigenvalues are their lengths. There is one degenerate pair and another single one. The eigenvector corresponding to the non-degenerate semi axis is parallel to $\boldsymbol{n}$ which is defined as the axis on which the images of $S$ and $E$ are diametrically opposed. Hence the length of this semi axis is $l_1 =|1-p -
p(1-\epsilon)|$. Furthermore we have $$\begin{aligned}
&\text{sign}(\det M) =\text{sign} (1-p - p(1-\epsilon)),$$ if $l_1>0$ and $\det M = 0$ if $l_1=0$. Thus if we calculate the length of this semi axis we can already determine the causality parameter $p$ as $$\begin{aligned}
l_1 = |1-2p +p\epsilon| \Leftrightarrow p = \frac{1\mp l_1}{2-\epsilon}, \label{eq:p2}\end{aligned}$$ where the ambiguity is solved by considering the sign of $\det M$.\
Now that we have $p$ and $\epsilon$ at hand we can define a new map with correlation matrix $$\begin{aligned}
M' \equiv \frac{1}{1-p\epsilon} M = \frac{p(1-\epsilon)}{1-p\epsilon}S + \frac{1-p}{1-p\epsilon} E \equiv p' S + (1-p') E, \label{eq:T'}\end{aligned}$$ where we defined $$\begin{aligned}
&p' \equiv \frac{p(1-\epsilon)}{1-p\epsilon},\\
& 1- p' = 1- \frac{p(1-\epsilon)}{1-p\epsilon} = \frac{1-p\epsilon - p(1-\epsilon}{1-p\epsilon} = \frac{1-p}{1-p\epsilon}.\end{aligned}$$ The properties of the ellipsoid can also be found in the SSV decomposition of the correlation matrix $$\begin{aligned}
M = R_1 DR_2, \qquad \text{with } D={\text{diag}}(\boldsymbol{\eta}^\mathcal{M}) \text{ and } \; R_1,R_2 \in SO(3).\end{aligned}$$ The absolute values of the entries of $\boldsymbol{\eta}^\mathcal{M}$ equal the lengths of the semi axes of the ellipsoid and we choose $R_1$ and $R_2$ such that $\eta^\mathcal{M}_1= \eta^\mathcal{M}_2$. The axis on which the images of $S$ and $E$ are diametrically opposed is then given by the last column of $R_1$, i.e. $\hat{n} = R_1 \hat{e}_3$. The length of this axis is $l_1 = |\eta^\mathcal{M}_3|$.\
In the promise is given that $S$ is the correlation matrix of a maximally entangled state and that $E$ is the correlation matrix of a unitary channel. The reconstruction of those is extensively studied in the supplementary information of [@Ried2015]. With the method presented there we find the value of $p'$ and can restore the correlation matrices corresponding to $U$ and $\ket{\psi}$ up to a binary ambiguity, and hence solve the causal inference problem. We review this in terms of SSV and discuss where the binary ambiguity arises.\
Starting from the l.h.s. of the goal is to determine $p', S,$ and $M$ on the r.h.s. Consider the SSV decomposition of the correlation matrix $$\begin{aligned}
M' = R'_1 D'R'_2, \qquad \text{with } D'={\text{diag}}(\boldsymbol{\eta}^\mathcal{M'}) \text{ and } R_1',R_2'\in SO(3).\end{aligned}$$ The absolute values of the entries of $\boldsymbol{\eta}^\mathcal{M'}$ equal the lengths of the semi axes of the ellipsoid and we choose $R'_1,R'_2$ s.t. $\eta_1^\mathcal{M'}=\eta_2^\mathcal{M'} $. The axis on which the images of $S$ and $E$ are diametrically opposed is then given by the last column of $R'_1$, i.e. $\hat{n}' = R'_1 \hat{e}_3$. The length of this axis is $l'_1 = |\eta_3^\mathcal{M'}|$. However, the direction of $\hat{n}'$, depending on the choice of $R'_1$ and $R'_2$, cannot be determined uniquely and allows two possible solutions $\pm\hat{n}'$. The parameter $p'$ is determined by the length $l_1'$ and can be calculated as $$\begin{aligned}
p' = \frac{1 - \left(\text{sign}\det(M') \right)l_1'}{2},\end{aligned}$$ and if $\det(M') = 0$ we have $p' = 1/2$. If $p'=0$ or $p'=1$ the reconstruction is trivial (of course in these cases one cannot reconstruct $S$ or $E$, respectively). If $p'\in (0,1)$, we can define [@Ried2015] $$\begin{aligned}
r' &= |\eta_1^\mathcal{M'}|,\\
\gamma' &= 2 \arcsin\left( \sqrt{\frac{1-r'^2}{4(p'-{p'}^2)}}\right),\\
\gamma_2' &= \arccos\left(\frac{1+r'^2- \left[2p'\sin\frac{\gamma'}{2}\right]}{2r'}\right). \end{aligned}$$ The reconstruction of the correlation matrices $S$ and $E$ can then be done, c.f. eq. (58) and (59) in the supplementary information of [@Ried2015]: $$\begin{aligned}
E &= R_{\hat{n}', \gamma'_2} S_{\perp\hat{n}', 1/r'} S_{\hat{n}', 1/(1-2p')} M', \label{eq:ERec}\\
S &= R_{\hat{n}', -\gamma'+\gamma'_2} S_{\perp\hat{n}', 1/r'} S_{\hat{n}', 1/(2p'-1)} M', \label{eq:SRec}\end{aligned}$$ where $R_{\hat{n}, \alpha}$ indicates a rotation about axis $\hat{n}$ with rotation angle $\alpha$, $ S_{\hat{n}', 1/(1-2p')}$ a scaling along $\hat{n}'$ by a factor $1/(1-2p')$ and $S_{\perp\hat{n}', 1/r'}
$ a scaling of the plane perpendicular to $\hat{n}'$ by a factor $1/r'$. From and we see that a reconstruction of $E$ and $S$ is not possible if $p'=1/2$.\
Let us summarize what we can infer about the causation of $M$ given in :
- The causality parameter $p$ can be determined uniquely in all cases, see eq..
- If $r'=0$ or $p'=1/2$ then $S$ and $E$ cannot be determined,
- else we can determine two sets of solutions for $E$ and $S$ given by and , distinguished by the choice of direction of $\hat{n}'$.
On the other hand, if we do not have prior knowledge of $\epsilon$, then in general we cannot determine $p$ with . This ambiguity can easily be illustrated by looking at an example:\
Take $U = \sigma_x$ and $\ket{\psi} = \frac{\ket{00} -\ket{11}}{\sqrt{2}}$. We then have: $$\begin{aligned}
E = \text{diag}(1,-1,-1), \qquad S = \text {diag}(-1,1,1).\end{aligned}$$ Combining this for arbitrary $\epsilon$ and $p$ gives $$\begin{aligned}
M = \text{diag}(1-p\epsilon, -(1-p\epsilon), -(1-p\epsilon)).\end{aligned}$$ Hence for all values of the parameters where $p\epsilon = \text{cons.}$, the measurement statistics for Alice and Bob are exactly the same and there is no way to distinguish different pairs of values.\
Analogously to using a generalized Werner state for the steering map, we can also use a generalized depolarizing channel. Then, with prior knowledge of the amount of noise, we can still find a complete solution even though the resulting channel might be entanglement breaking.
### Generalized depolarizing channel and generalized Werner state
We shall now consider the case where both the channel as well as the state are mixed with a known amount of noise. Therefore we take $S'
= (1-\epsilon_c) S$ for a generalized Werner state (thus $S$ corresponds again to a rotated and inverted Bloch-sphere) and $E' =
(1-\epsilon_e) E$ for a generalized depolarizing channel. We again assume $\epsilon_e\,
\in (0,1)$ and $\epsilon_c\, \in (0,1)$ to be known. We then have $$\begin{aligned}
M = (1-p)(1-\epsilon_e) E + p (1-\epsilon_c) S.\end{aligned}$$ The reconstruction works as follows. Without loss of generality we assume $\epsilon_e \leq \epsilon_c$ (in the other case we just have to make the reconstruction discussed in the previous subsection for the entanglement breaking channel and not for the Werner-state). The only thing we have to do is to divide by $(1-\epsilon_e)$ to restore the problem of the previous section $$\begin{aligned}
M' = \frac{M}{1-\epsilon_e} = (1-p) E + p \frac{1-\epsilon_c}{1-\epsilon_e} S \equiv (1-p) E + p (1-\epsilon) S,\end{aligned}$$ with $1-\epsilon \equiv \frac{1-\epsilon_c}{1-\epsilon_e}$. The rest can then be solved as in the previous subsection.\
Again we remark that nothing changes if we have $\epsilon_c > 2/3$ and $\epsilon_e > 2/3$ even though at that transition the states become separable and the channels entanglement-breaking, respectively.
Discussion {#sec:discussion}
==========
In this work we extended the results initially found by Ried et al. [@Ried2015]. We introduced an active and a passive quantum-observational scheme as analogies to the classical observational scheme. The passive quantum-observational scheme does not allow for an advantage over classical casual inference. In the active quantum observational scheme Alice and Bob can freely choose their measurement bases, which in principle allows for signaling. However, we investigated the quantum advantage over classical causal inference in a scenario where signaling is not possible in the active quantum observation scheme, as Alice’ incoming state is completely mixed.\
We showed how the geometry of the set of signed singular values (SSV) of correlation matrices representing positive maps of the density operator $\rho_A\mapsto\rho_B$ determines the possibility to reconstruct the causal structure linking $\rho_A$ and $\rho_B$. We showed that there are more cases than previously known for which a complete solution of the causal inference problem can be found without additional constraints, namely all correlations created by maps whose signed singular values of the correlation matrix lie on the edges of the cube of positive maps $\mathcal{C}$ defined in . A necessary and sufficient condition for this is that the state is maximally entangled, that the channel is unitary, and that the corresponding correlation matrices have a SSV decomposition involving the same rotations.\
For correlations guaranteed to be produced by a mixture of a unital channel and a unital bipartite state, we quantified the quantum advantage by giving the intervals for possible values of the causality parameter $p$. Here, in order to constrain $p$, and hence have an advantage over classical causal inference, it is necessary that the correlations were caused by an entangled state and/or an entanglement preserving channel. This is because correlations caused by any mixture of a separable state and an entanglement breaking quantum channel always describe super-positive maps. According to theorem \[theo:causal\_interval\] the causal interval for any super-positive map $\mathcal{M}$ is $I_\mathcal{M} = [0,1]$. Hence, super-positive maps do not allow any causal inference.\
Things change when we further strengthen the assumptions on the data generating processes and allow only unitary freedom in the state, corresponding to a generalized Werner state with given degree of noise $\epsilon_c$, or unitary freedom in the channel, corresponding to a generalized depolarizing channel with given degree of noise $\epsilon_e$. We showed that in this scenario the causality parameter $p$ can always be uniquely determined and in most cases the state and the channel can be reconstructed up to binary ambiguity. For $\epsilon_c >
2/3$ the state becomes separable and for $\epsilon_e > 2/3$ the channel entanglement breaking but still causal inference is feasible. Therefore entanglement and entanglement preservation are not a necessary condition in this scenario. The assumptions on the data generating processes, i.e. a-priori knowledge of $\epsilon_c$ and $\epsilon_e$, are strong enough, such that even correlations corresponding to super-positive maps reveal the underlying causal structure.\
Appendix {#sec:Appendix}
========
Signed singular values of sums of matrices {#signed-singular-values-of-sums-of-matrices .unnumbered}
------------------------------------------
Let $A$ be a $n\times n$ real matrix. A possible *singular value decomposition* (SVD) of $A$ is given as $$\begin{aligned}
A = O_1 D O_2, \label{def: App_ASV}\end{aligned}$$ where $O_{1,2}$ are orthogonal matrices ($O_i O_i^T = \mathbb{1}$) and $D$ is a positive semi-definite diagonal matrix $D = {\text{diag}}(\sigma_1^A,...,\sigma_n^A)$, with $\sigma_i^A$ called the (absolute) *singular values* (SV) of $A$. The matrices in are not uniquely defined and all possible permutations of the singular values on the diagonal of $D$ are possible for different orthogonal matrices $O_1$ and $O_2$. We use this freedom to write the SV in *canonical* order, $\sigma_1^A \geq \sigma_2^A \geq ...\geq \sigma_n^A$.\
*Example:* We give two different SVDs of a $3\times3$ matrix $B$ $$\begin{aligned}
B \equiv \begin{pmatrix}
-1 & 0&0\\ 0&0&-3\\0&2&0
\end{pmatrix} &=\begin{pmatrix}
1&0&0\\0&0&-1\\0&1&0
\end{pmatrix} \begin{pmatrix}
1&0&0\\0&2&0\\0&0&3
\end{pmatrix}
\begin{pmatrix}
-1&0&0\\0&1&0\\0&0&1
\end{pmatrix} \\
&= \begin{pmatrix}
0&-1&0\\ 0&0&1\\-1&0&0
\end{pmatrix} \begin{pmatrix}
3&0&0\\0&2&0\\0&0&1
\end{pmatrix}\begin{pmatrix}
0&0&1\\0&1&0\\1&0&0
\end{pmatrix}.
\end{aligned}$$ The last decomposition gives the singular values of $B$ in canonical order $\sigma_1^B =3,\sigma_2^B = 2$ and $\sigma_3^B = 1$.\
Next we call $$\begin{aligned}
A = R_1 D' R_2 \label{def:App_SSV}\end{aligned}$$ the *signed singular value decomposition* (also called real singular values [@Amir-Moez1958]) of $A$, where $R_i \in SO(n)$ are orthogonal matrices with determinant equal to one. In the $3\times 3$ scenario these correspond to proper rotations in $\mathbb{R}^3$. The diagonal matrix $D'$ contains the *signed singular values* (SSV) of A. The SSV have the same absolute values as the SV but additionally can have negative signs. Concretely, the freedom in choosing $R_1$ and $R_2$ allows one to get any permutations of the SV on the diagonal of $D'$ together with an even or odd number of minus signs, depending on whether $A$ has positive or negative determinant, respectively. If at least one singular value equals 0, the number of signs becomes completely arbitrary. Using the same matrix $B$ as above we give two different signed singular value decompositions as an *example:* $$\begin{aligned}
B \equiv \begin{pmatrix}
-1 & 0&0\\ 0&0&-3\\0&2&0
\end{pmatrix} &=\begin{pmatrix}
1&0&0\\0&0&-1\\0&1&0
\end{pmatrix} \begin{pmatrix}
-1&0&0\\0&2&0\\0&0&3
\end{pmatrix}
\begin{pmatrix}
1&0&0\\0&1&0\\0&0&1
\end{pmatrix} \\
&= \begin{pmatrix}
0&-1&0\\ 0&0&1\\-1&0&0
\end{pmatrix} \begin{pmatrix}
3&0&0\\0&2&0\\0&0&-1
\end{pmatrix}\begin{pmatrix}
0&0&1\\0&1&0\\-1&0&0
\end{pmatrix}. \label{eq:canonicalSSV}
\end{aligned}$$ For the SSV decomposition we define a *canonical* order with the absolute values of the singular values sorted in decreasing order and only a negative sign on the last entry if the matrix has negative determinant, as in . The rotational freedom in allows for arbitrary permutations of the order of singular values and addition of any even number of minus signs.\
Confusion may arise since for example an $\mathbb{R}^3$ permutation matrix corresponding to a permutation of exactly two coordinates has determinant -1, so why would it be allowed? The point is, that we not only want to permute elements of a vector, but the diagonal elements of a matrix. We illustrate that by permuting two components of *i)* a vector and *ii)* a diagonal matrix. $$\begin{aligned}
&P_{yz} \equiv \begin{pmatrix} 1&0&0\\0&0&1\\0&1&0 \end{pmatrix}, \qquad \det P_{yz} = -1,\\
& \textit{i) }P_{yz}\cdot \begin{pmatrix}
a\\b\\c \end{pmatrix} = \begin{pmatrix}
a\\c\\b
\end{pmatrix},\\
&\textit{ii) }P_{yz} \cdot\text{diag}(a,b,c) \cdot P_{yz} = \text{diag}(a,c,b) = (-P_{yz}) \, \text{diag}(a,b,c)\, (-P_{yz}).
\end{aligned}$$ I.e. as $-P_{yz} = R_{\hat{\boldsymbol{x}}}(\pi/2)\cdot
R_{\hat{\boldsymbol{y}}}(\pi) $ the effect of permuting the second and third diagonal entry of a diagonal matrix can also be obtained by proper rotations, and correspondingly for other permutations of the SSV. Hence all permutations of the SSV are allowed.\
Fan [@Fan1951] gave bounds on the SV of $A+B$ given the SV of two real matrices $A$ and $B$, derived from the corresponding results for eigenvalues of hermitian matrices and using that the matrix $\tilde{A} \equiv \begin{pmatrix}
0_{n\times n}& A\\A^T & 0_{n\times n}
\end{pmatrix} $ has the singular values of $A$ and their negatives as eigenvalues [@Marshall1979 p.243 for review]. In the main part of this work we need a more constraining statement using the SSV, and thus taking the determinant of $A$, $B$, and $A+B$ into account as well. This leads to theorem \[theo:SSV\]. In the following we will denote with $\tilde{\boldsymbol{\sigma}}(A)$ the vector of canonical SSV of the $n\times n$ real matrix A. Since the product of two rotations is again a rotation it follows directly from that $$\begin{aligned}
\tilde{\boldsymbol{\sigma}}(Q_1A Q_2) = \tilde{\boldsymbol{\sigma}}(A), \; \forall Q_1,Q_2 \in SO(n). \label{eq:invariance}\end{aligned}$$ Let $\boldsymbol{w}$ be a $n$-dimensional vector. We define $$\begin{aligned}
\Delta_{\boldsymbol{w}}\equiv \text{Conv}\left( \left\{\left.\left( s_1 w_{\pi(1)}, ..., s_n w_{\pi(n)}\right)^T\right|s_\nu \in \{-1,1\}: \prod_\nu s_\nu =1, \pi\in S_n\right\}\right) \label{def:Delta}\end{aligned}$$ as the convex hull of all possible permutations $\pi\in S_n$ of the components of $\boldsymbol{w}$ multiplied with an even number of minus signs. Let now $\boldsymbol{w_1}$ and $\boldsymbol{w_2}$ be two $n$-dimensional vectors. We define $$\begin{aligned}
\Sigma_{\boldsymbol{w_1}, \boldsymbol{w_2}} \equiv \left\{ a+b| a\in \Delta_{\boldsymbol{w_1}}, b\in \Delta_{\boldsymbol{w_2}}\right\}. \label{def:Sigma}\end{aligned}$$ Figure \[Fig:sets\] presents an illustration of the case $n=2$.
![**Illustration of theorem \[theo:SSV\]:** Suppose we have two $2\times 2$ matrices $A$ and $B$ with SSV $\boldsymbol{w}_1$ and $\boldsymbol{w}_2$ respectively. The red and the yellow sets correspond to $\Sigma_{\boldsymbol{w}_1}$ and $\Sigma_{\boldsymbol{w}_2}$ defined by . By theorem \[theo:SSV\] the vector of SSV of $A+B$ then lies within the blue set, defined by .[]{data-label="Fig:sets"}](TheoremV1.eps)
\[theo:SSV\] Let $A$ and $B$ be two $n\times n$ real matrices whose SSV are known. Then $$\begin{aligned}
\tilde{\boldsymbol{\sigma}}(A+B) \in \Sigma_{\tilde{\boldsymbol{\sigma}}(A),\tilde{\boldsymbol{\sigma}}(B)}.\end{aligned}$$
Let $A$ be a $n\times n$ real matrix and let $\boldsymbol{d}(A)$ denote the vector of diagonal entries of $A$. Thompson showed the following two statements about the diagonal elements of $A$ [@Thompson1977 theorems 7 and 8] $$\begin{aligned}
i)& \boldsymbol{d}(A) \in \Delta_{\tilde{\boldsymbol{\sigma}}(A)},\label{eq:diag1}\\
ii)& \forall \boldsymbol{d} \in \Delta_{\tilde{\boldsymbol{\sigma}}(A)}\; \exists R_1,R_2 \in SO(n): \boldsymbol{d}= \boldsymbol{d}(R_1AR_2). \label{eq:diag2}\end{aligned}$$ Now let $A$ and $B$ be two $n\times n$ real matrices. Let $R_1,R_2 \in SO(n)$ such that $\boldsymbol{d}(R_1(A+B)R_2) = \tilde{\boldsymbol{\sigma}}(A+B)$. We then have $$\begin{aligned}
\tilde{\boldsymbol{\sigma}}(A+B) = \boldsymbol{d}(R_1(A+B)R_2) = \boldsymbol{d}(R_1AR_2) + \boldsymbol{d}(R_1BR_2) \in \Sigma_{\tilde{\boldsymbol{\sigma}}(R_1AR_2) ,\tilde{\boldsymbol{\sigma}}(R_1BR_2) } = \Sigma_{\tilde{\boldsymbol{\sigma}}(A) ,\tilde{\boldsymbol{\sigma}}(B) },\end{aligned}$$ where the second equation follows from the linearity of matrix addition in every element and the last equality from .
\
As mentioned above, results for the absolute singular values of $A+B$ have been known before. To complete, we show that the above proof works analogously for the corresponding statement on absolute singular values: Let $\boldsymbol{\sigma}(A)$ denote the vector of canonical absolute singular values of an $n\times n$ real matrix $A$, $\sigma_1(A) \geq \sigma_2(A)
\geq ... \geq \sigma_n(A)$. Let $B$ be another $n\times n$ real matrix. Then [@Marshall1979 chapter 9 G.1.d.] $$\begin{aligned}
\boldsymbol{\sigma}(A+B) \prec_w \boldsymbol{\sigma}(A) + \boldsymbol{\sigma}(B), \label{eq:majorizaiton}\end{aligned}$$ i.e. the vector of canonical singular values of $A+B$ is *weakly majorized* by the sum of the vectors of canonical singular values of $A$ and $B$. *Weak majorization* for two vectors $\boldsymbol{x}$ and $\boldsymbol{y}$ with $x_1\geq x_2 \geq ...\geq x_n$ and $y_1 \geq y_2\geq...\geq y_n$ is defined as $$\begin{aligned}
\boldsymbol{x} \prec_w \boldsymbol{y} \Leftrightarrow \sum_{i=1}^k x_i \leq \sum_{i=1}^k y_k\; \forall k \in \left\{1,2,...,n\right\}.\end{aligned}$$ To see define $\Delta'_{\boldsymbol{w}}$ analogously to but without the constraint $ \prod_\nu s_\nu =1$, i.e. allowing arbitrary sign flips. The analogue statements of and hold if we exchange the SSV with the absolute singular values, proper rotations (elements of $SO(n)$) with orthogonal matrices (elements of $O(n)$), and $\Delta_{\boldsymbol{w}}$ with $\Delta'_{\boldsymbol{w}}$. We then find, that $\boldsymbol{\sigma}(A+B) \in
\Sigma'_{\boldsymbol{\sigma}(A),\boldsymbol{\sigma}(B)}$, with $\Sigma'_{\boldsymbol{w_1}, \boldsymbol{w_2}} \equiv \left\{ a+b| a\in \Delta'_{\boldsymbol{w_1}}, b\in \Delta'_{\boldsymbol{w_2}}\right\}$. Since per definition the absolute singular values are non-negative, we can further restrict $\Sigma'$ to the first hyperoctant. On the other hand, for two vectors $\boldsymbol{x}$, $\boldsymbol{y} \in
\mathbb{R}^n_+$ we have (proposition C.2. of chapter 4 in [@Marshall1979]) $$\begin{aligned}
\boldsymbol{x} \prec_w \boldsymbol{y} \Leftrightarrow \boldsymbol{x} \in \text{Conv}\left(\left\{s_1 y_{\pi(1)},..., s_n y_{\pi(n)}| s_\nu \in \{0,1\}, \pi\in S_n\right\}\right).\end{aligned}$$ The set on the r.h.s. coincides with the restriction of $\Sigma'$ to the first hyperoctant if we take $\boldsymbol{y} = \boldsymbol{\sigma}(A) + \boldsymbol{\sigma}(B)$. Taking $\boldsymbol{x} = \boldsymbol{\sigma}(A+B)$, eq. follows.
[^1]: Strictly speaking, only for $p\neq 1/2$ one can always determine the unitary and the state. For $p=1/2$ there is an infinite number of channels and states (all those where every point is diametrically opposed for the unitary channel and the state.), for which the ellipsoid reduces to a single point, and hence the correlation matrix is the zero matrix. The parameter $p=1/2$ can then be restored but not the unitary and the state.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'A.-L. Maire, T. Stolker, S. Messina, A. Müller, B. A. Biller, T. Currie, C. Dominik, C. A. Grady, A. Boccaletti, M. Bonnefoy, G. Chauvin, R. Galicher, M. Millward, A. Pohl, W. Brandner, T. Henning, A.-M. Lagrange, M. Langlois, M. R. Meyer, S. P. Quanz, A. Vigan, A. Zurlo, R. van Boekel, E. Buenzli, T. Buey, S. Desidera, M. Feldt, T. Fusco, C. Ginski, E. Giro, R. Gratton, N. Hubin, J. Lannier, D. Le Mignant, D. Mesa, S. Peretti, C. Perrot, J. R. Ramos, G. Salter, M. Samland, E. Sissa, E. Stadler, C. Thalmann, S. Udry, and L. Weber'
bibliography:
- 'biblio.bib'
date: 'Received 14/10/2016; accepted 14/02/2017'
title: 'Testing giant planet formation in the transitional disk of SAO 206462 using deep VLT/SPHERE imaging[^1]'
---
[The SAO 206462 (HD 135344B) disk is one of the few known transitional disks showing asymmetric features in scattered light and thermal emission. Near-infrared scattered-light images revealed two bright outer spiral arms and an inner cavity depleted in dust. Giant protoplanets have been proposed [to account for]{} the disk morphology.]{} [We aim to search for giant planets responsible for the disk features and, in [the]{} case of non-detection, to constrain recent planet predictions using the data detection limits.]{} [We obtained new high-contrast and high-resolution [total intensity images of the target spanning the $Y$ to the $K$ bands (0.95–2.3 $\muup$m) using the VLT/SPHERE near-infrared camera and integral field spectrometer]{}.]{} [The spiral arms and the outer cavity edge are revealed at high resolutions and sensitivities without the need for aggressive image post-processing techniques, which introduce photometric biases. We do not detect any close-in companions. For the derivation of the detection limits on putative giant planets embedded in the disk, we show that the knowledge of the disk aspect ratio and viscosity is critical for the estimation of the attenuation of a planet signal by the protoplanetary dust because of the gaps that these putative planets may open. Given assumptions on these parameters, the mass limits can vary from $\sim$2–5 to $\sim$4–7 Jupiter masses at separations beyond the disk spiral arms. The SPHERE detection limits are more stringent than those derived from archival NaCo/$L^{\prime}$ data and provide new constraints on a few recent predictions of massive planets (4–15 $M_{\rm{J}}$) based on the spiral density wave theory. The SPHERE and ALMA data do not favor the hypotheses on massive giant planets in the outer disk (beyond 0.6$''$). There could still be low-mass planets in the outer disk and/or planets inside the cavity.]{}
Introduction
============
SAO206462 (HD135344B) is a rapidly rotating ($v$ sin $i$ $\sim$ 83 kms$^{-1}$) Herbig Ae/Be F4Ve star of age 9$\pm$2 Myr and mass 1.7$^{+0.2}_{-0.1}$ $M_{\odot}$ [@Mueller2011] [located at 156$\pm$11 pc [@GaiaCollaboration2016]]{} in the Upper Centaurus Lupus star-forming region, known to harbor a transitional disk resolved in scattered light [@Grady2009; @Muto2012; @Garufi2013; @Wahhaj2015; @Stolker2016] [and]{} in thermal emission, both at mid-IR [@Doucet2006; @Maaskant2013] and (sub-)mm wavelengths [@Brown2009; @Andrews2011; @Lyo2011; @Perez2014; @Pinilla2015; @vanderMarel2016a; @vanderMarel2016b]. It is part of a [widely separated ($\sim$21$''$, i.e. $\sim$3300 au) binary]{} with SAO 206463 [@Coulson1995]. Transitional disks are associated with an intermediate stage of disk evolution, where the dust opacity has been reduced (perhaps tracing [a]{} dip in the gas surface density) in the near- and mid-IR [e.g., @Strom1989; @Espaillat2014]. [Several scenarios have been proposed [that can explain]{} the gas and dust depletion [such as]{} photoevaporation, dust grain growth, [or ongoing]{} planet formation [@Hollenbach1994; @Dullemond2005; @Lin1993]]{}.
UT date Seeing ($''$) $\tau_0$ (ms) AM start/end Mode Bands DIT(s)$\times$Nfr $\Delta$PA ($^{\circ}$) SR
------------ --------------- --------------- -------------- ------------- -------------- ------------------- ------------------------- -----------
2015/05/15 0.45–0.57 7–10 1.04–1.03 IRDIFS\_EXT $YJH$+$K1K2$ 64$\times$64 63.6 0.78–0.85
Measurements of the CO line profiles [@Dent2005; @Pontoppidan2008; @Lyo2011] and of the stellar rotation [@Mueller2011] for SAO 206462 are consistent with an almost face-on geometry for the system ($i$$\sim$11$^{\circ}$). The star is actively accreting [$\sim$10$^{-11}$–10$^{-8}$ $M_{\odot}$/yr, @GarciaLopez2006; @Grady2009]. The disk is composed of a [massive outer component [up to $\sim$2$''$ i.e. $\sim$300 au, @Lyo2011] with masses for the dust $M_{\mathrm{dust}}$$=$1.3$\times$10$^{-4}$ $M_{\odot}$ and gas $M_{\mathrm{gas}}$$=$1.5$\times$10$^{-2}$ $M_{\odot}$ [@vanderMarel2016a]]{}, scattered-light spiral features [within $\sim$80 au]{} indicative of dynamical processes [@Muto2012], a large inner sub-mm dust cavity [[$\sim$50 au]{}, @Andrews2011], and a sub-au dust accretion disk [@Fedele2008; @Carmona2014; @Menu2015]. Recently, @Stolker2016 presented SPHERE optical and near-IR polarimetric images showing shadowing of the outer disk by warping/perturbation of the inner disk and the (marginal) detection of scattered light [up to $\sim$160 au]{}.
[One plausible explanation]{} for the spiral features seen in the SAO 206462 system is that they might be driven by planets. Two approaches have been used [to investigate]{} this hypothesis in the literature. The first approach consists in fitting analytical formulae derived for linear perturbations, i.e. one planet driving one spiral, to the observations. @Muto2012 proposed two planets located beyond $\sim$50 au with masses of $\sim$0.5 $M_{\rm{J}}$. Recently, @Stolker2016 used a similar approach assuming two planets located interior or exterior to the spirals and showed that planets in the latter configuration with separations $\gtrsim$100 au provided a better match to their shape. The second approach exploits hydrodynamical simulations, which can account for non-linear perturbations, i.e. one planet can drive more than one spiral [e.g., @Dong2015; @Zhu2015; @Juhasz2015; @Pohl2015]. In particular, [some studies have predicted one massive ($\sim$4–15 $M_{\rm{J}}$) planet located exterior to the SAO 206462 spirals [@Fung2015; @Bae2016; @Dong2017]]{}.
@Garufi2013 suggested that one planet of mass from 13 to 5 $M_{\rm{J}}$ located at [17.5–20 au]{} could explain the increase [in]{} the cavity size measured for the small dust grains (28$\pm$6 au) and for the large dust grains [39 to 50 au, @Brown2009; @Andrews2011; @Lyo2011][^2]. This planet could also explain the spiral features. They could not rule out several massive planets inside the cavity, although this hypothesis seems at odds with the presence of diffuse gas. They also rejected several mechanisms not involving perturbing planet(s) [photoevaporation, dust grain growth, magnetorotational instability, @Hollenbach1994; @Dullemond2005; @Chiang2007] [to explain]{} the disk cavity. They finally showed that the disk is globally highly stable [see also @Stolker2016]. Nevertheless, they could not rule out gravitational interactions between SAO 206462 and SAO 206463 [as exciting mechanism for the disk spirals,]{} if [the components]{} have highly eccentric orbits.
Recently, @vanderMarel2016b presented ALMA data at spatial resolutions of 0.16$''$, which show that the previously detected asymmetric millimeter dust ring [@Perez2014; @Pinilla2015] consists [of]{} an inner ring and an outer asymmetric feature. They also proposed a different hypothesis [that can explain]{} the disk features detected in both scattered light and thermal emission, where a vortex creates one spiral arm and a planet inside the cavity produces the other spiral arm.
Studies tried to detect stellar and [substellar]{} companions in the disk of SAO 206462. @Pontoppidan2008 ruled out a stellar companion inside the cavity from the detection of CO gas. VLT/NaCo imaging at 1.75 and 2.12 $\muup$m [discarded]{} low-mass stellar companions within the gap region ($>$0.22 $M_{\odot}$ [beyond 14 au]{}) and [brown dwarf]{} companions more massive than [$>$19 $M_{\rm{J}}$ beyond 70 au]{} [@Vicente2011][^3]. However, these limits are not sensitive to planetary-mass companions.
We present in this paper new high-contrast images of SAO 206462 covering the spectral range 0.95–2.32 $\muup$m obtained with the instrument VLT/SPHERE [@Beuzit2008] as part of the SpHere INfrared survey for Exoplanets (SHINE; Langlois et al., in prep.). We describe the observations and the data reduction [(Sects. \[sec:obs\] and \[sec:datareduc\])]{}. Then, we discuss the detection of point sources (Sect. \[sec:pointsources\]), the observed disk features (Sect. \[sec:disk\]) and the detection limits on putative giant protoplanets with respect to predictions from the literature (Sect. \[sec:detectionlimits\]). Finally, we analyze the sensitivity of the detection limits to the protoplanetary dust opacity considering the presence of gaps opened by putative giant embedded planets (Sect. \[sec:dustopacity\]).
Observations {#sec:obs}
============
SPHERE [@Beuzit2008] is an extreme adaptive optics (AO) instrument dedicated to high-contrast and high-resolution imaging of young giant exoplanets and circumstellar disks. The AO system [@Fusco2006] includes a fast 41$\times$41-actuators wavefront control, pupil stabilization, differential tip tilt [control,]{} and toric mirrors [@Hugot2012] for beam transportation to the coronagraphs [@Boccaletti2008c] and science instruments. [In this paper, we used the near-infrared science instruments, the infrared dual-band imager and spectrograph IRDIS [@Dohlen2008a] and the integral field spectrometer IFS [@Claudi2008; @Antichi2009]. The focal plane masks of the near-infrared coronagraphs are located in the common path and infrastructure of SPHERE, whereas the Lyot stops are located in IFS and IRDIS and are optimized for each instrument.]{}
We observed SAO 206462 on 2015 May 15 UT in the IRDIFS\_EXT mode (Table \[tab:obs\]). In this mode, [IRDIS and IFS]{} are operated in parallel, with IRDIS observing in the $K12$ filter pair [$\lambda_{K1}$=2.110 $\muup$m and $\lambda_{K2}$=2.251 $\muup$m, $R$$\sim$20, @Vigan2010] and IFS in the $YJH$ bands [0.95–1.65 $\muup$m, $R$$\sim$33, @Claudi2008]. The observing conditions were good and stable (wind speed $\sim$6–10 m/s, see Table \[tab:obs\]). The star was imaged with an apodized pupil Lyot coronagraph [@Soummer2005] of diameter 185 mas [@Boccaletti2008c]. At the beginning and the end of the sequence, [we acquired two dedicated calibrations. First, we recorded]{} for flux calibration purposes unsaturated images of the star out of the coronagraphic mask and inserting a neutral density filter of average transmission $\sim$1/100[^4] in the optical path. [Then, we]{} obtained coronagraphic images with four crosswise faint replicas of the star artificially generated using the deformable mirror [@Langlois2013] to measure the star location [for frame registering (Sect. \[sec:datareduc\]).]{} After the observation, we measured sky backgrounds. However, because of a hardware issue, the second unsaturated PSF image and the sky background images for IRDIS were not recorded properly. As a consequence, we used for the sky subtraction the closest [sky backgrounds (taken the previous night)]{} obtained with the same filter and coronagraph configuration [as]{} our observations and scaled them to the DIT and the background measured at large separation in the science images. [This non-optimal sky subtraction can affect the accuracy of the background subtraction to values below $\sim$8% for the IRDIS RDI images, which are used [to measure]{} the disk photometry]{} (Sect. \[sec:disk\]). All the other calibration data (darks, detector flats, wavelength calibration, IFU flats) were obtained during the following day.
Data reduction and analysis {#sec:datareduc}
===========================
The data were reduced with the SPHERE Data Center pipeline, which uses the Data Reduction and Handling software [v0.15.0, @Pavlov2008] and dedicated IDL routines for the IFS data reduction [@Mesa2015]. The pipeline subtracts the sky background, corrects for the detector flat field, removes bad pixels, derives the IFS wavelength calibration, corrects for the IFU [flat, corrects]{} for anamorphism [0.60$\pm$0.02%, @Maire2016a], and registers the frames using the coronagraphic images taken with the satellite spots. We measured the full width at half maximum of the point-spread function (PSF) for the IRDIS images to be $\sim$56 mas ($K1$) and $\sim$59 mas ($K2$) and for the IFS images to be $\sim$37 mas ($J$) and $\sim$42 mas ($H$)[^5]. Then, the data were analyzed with a consortium image processing pipeline (R. Galicher, private comm.). This pipeline allows for several imaging post-processing techniques: classical angular differential imaging [ADI, @Marois2006a], subtraction of [a radial profile from the data]{}, Locally Optimized Combination of Images [LOCI, @Lafreniere2007a], Template-LOCI [TLOCI, @Marois2014], and principal component analysis [PCA, see @Soummer2012; @Amara2012]. [We also considered reference differential imaging (RDI) in order to obtain unbiased views of the disk morphology. [To build]{} the RDI reference, we [scaled in intensity the images of another star]{} without any disk features obtained in the same observing mode during the previous night. The quasi-static speckle pattern did not vary significantly between the two sequences, allowing for efficient rejection of the stellar residuals.]{} We show in Figs. \[fig:ifsim\] and \[fig:irdisim\] the median-collapsed IFS and IRDIS images obtained with this method and with classical ADI and in Fig. \[fig:irdisimfov\] the full field of view of the IRDIS classical-ADI image. [Regions inside the magenta circles in Figs. \[fig:ifsim\] and \[fig:irdisim\] were found to be dominated by the stellar residuals using the chromaticity of the stellar residuals.]{}
{width="11.5cm"}
Three point sources are detected in the IRDIS field (Fig. \[fig:irdisimfov\]). Their photometry and astrometry were measured [using the TLOCI algorithm]{} applied to each spectral band separately. We divided each science frame into annuli of 1.5 full width at half maximum. Then, for each science frame and annulus, we computed a reference frame of the stellar residuals using the best linear combination of the 80 most correlated frames for which the [self-subtraction of mock point sources, [modeled]{} using the observed PSF]{}, was at maximum 20%. These parameters were selected from internal tests for the analysis of the consortium data. Negative synthetic companions [@Marois2010b; @Bonnefoy2011] [modeled]{} from the observed unsaturated PSF were inserted in the pre-processed data at the location of the detected point sources. We then processed the data assuming the TLOCI coefficients computed for the analysis without the synthetic companions. The [subpixel]{} position and the flux of the modeled images were optimized to minimize the image residuals within a disk of radius 1.5 full width at half maximum (FWHM) centered on the measured companions [@Galicher2011a]. The error bars include the variations in the stellar flux during the sequence (estimated from the fluctuations of the stellar residuals, 0.009 mag for the $K1$ band) and the accuracy of the fitting procedure. The error term related to the variations in the PSF could not be estimated because no IRDIS PSF was recorded after the sequence (Sect. \[sec:obs\]). The astrometry of the three detected companions (Table \[tab:photometryastrometry\]) was calibrated using pixel scales of 12.267$\pm$0.009 and 12.263$\pm$0.009 mas/pix for the $K1$ and $K2$ filters respectively, and a [north]{} angle offset of $-1.712$$\pm$0.063$^{\circ}$[^6] [@Maire2016b].
We finally used TLOCI [to derive]{} the S/N maps for point sources. Each pixel value was divided by the standard deviation of the flux measured in [an annulus of 1 FWHM]{} at the same angular separation. The TLOCI throughput was assessed using synthetic companions inserted at regular separations between 0.15$''$ and 6$''$. We iterated the TLOCI analysis on several position angles to average the effects of random speckle residuals. The azimuthally averaged contrast curves discussed in Sect. \[sec:detectionlimits\] were estimated from these S/N maps. [We note that although the disk signal is strongly attenuated by the TLOCI algorithm (which was optimized for point-source detection) in the SPHERE and NaCo data (the latter are described in Sect. \[sec:nacodata\]), the disk residuals increase somewhat the measured noise level. This effect is stronger in the SPHERE data, because of the higher S/N detection of the disk. However, these effects are intrinsic to the data and do not have to be corrected for.]{} [Finally, we corrected the SPHERE detection limits for the coronagraph transmission radial profile (A. Boccaletti, priv. comm.) and the small sample statistics following the prescription in @Mawet2014.]{}
Filter $\rho$ (mas) $\theta$ (deg) $\Delta$mag
---------- -------- -------------- ----------------- ----------------
Star 1 $K1$ 5723$\pm$6 126.13$\pm$0.14 9.13$\pm$0.04
$K2$ 5723$\pm$6 126.14$\pm$0.14 9.26$\pm$0.03
Star 2 $K1$ 5499$\pm$6 123.70$\pm$0.14 10.91$\pm$0.03
$K2$ 5499$\pm$6 123.70$\pm$0.14 10.98$\pm$0.03
[Bkgd]{} $K1$ 5035$\pm$10 319.84$\pm$0.17 14.55$\pm$0.10
[star]{} $K2$ 5038$\pm$19 319.88$\pm$0.25 14.55$\pm$0.23
: [Astrometry and photometry relative to the star of the point sources detected in the IRDIS field of view (Fig \[fig:irdisimfov\]).]{}[]{data-label="tab:photometryastrometry"}
We note in the IFS ADI image (bottom panel of Fig. \[fig:ifsim\]) a blob feature interior to the S2 spiral south of the star at $\sim$0.20$''$ at the $\sim$3 $\sigma$ level[, excluding the pixels at the same separation dominated by the disk features]{}. The feature is visible at the same location in all the individual IFS images and also in the images obtained after subtraction of a radial profile[, but]{} not in aggressive PCA and TLOCI reduced data (results not shown). We conclude that this feature is likely a residual from a disk feature.
Detection of point sources {#sec:pointsources}
==========================
The two first point sources listed in Table \[tab:photometryastrometry\] ([Stars 1 and 2]{}) were classified as a background stellar binary by @Grady2009. For the last point source ([denoted]{} “bkgd star” for background star), we derived a mass range of $\sim$2–4 $M_{\rm{J}}$ according to the [atmospheric and evolutionary models of @Baraffe2015 and @Baraffe2003]{} considering the uncertainties on its measured magnitudes and the system distance and age [@Mueller2011]. Using consortium tools developed for the classification and the ranking of the companion candidates discovered in SPHERE/SHINE, we also computed a background probability for this point source of $\sim$13.5% assuming the Besançon model predictions [@Robin2003]. With the same suite of tools, we derived the color-magnitude diagram of the [point source]{} shown in Fig. \[fig:cmdiagram\] to compare its $K1-K2$ color to the colors of field and young dwarfs covering the M, L, T, and Y types. Further details about the derivation of the color-magnitude diagram are provided in [@Mesa2016. Its $K1-K2$ color]{} is similar to a late-L or early-T [dwarf,]{} but its $K1$ magnitude is slightly fainter with respect to what is expected for objects of these spectral types. The point source is also seen in an HST/STIS image published in @Grady2009 (program GO-8674, P.I. Lagrange). We retrieved [the reduced image (corrected for cosmetics and distortion) from the HST archive,]{} and measured the astrometry of the [point source]{} ($r$=4932$\pm$28 mas, $\theta$=314.5$\pm$0.33$^{\circ}$) to compare it with the SPHERE astrometry. The results are shown in Fig. \[fig:cpmtest\]. The motion of the [point source]{} between the two epochs reflects the s[tellar proper motion [$\mu_{\mathrm{\alpha}}$=$-$20.159$\pm$0.603 mas/yr, $\mu_{\mathrm{\delta}}$=$-$22.481$\pm$0.719 mas/yr, @GaiaCollaboration2016]]{}[; it]{} is consistent with an object not gravitationally bound to SAO 206462. Therefore, we conclude that the [point source]{} is a background star.
Disk imaging and photometry {#sec:disk}
===========================
We analyze [the]{} morphology and photometry of the detected disk features using the spatial and spectral information provided by the SPHERE data. A full [modeling]{} of the total intensity images of the disk will allow [a more quantitative analysis of]{} the physical [constraints,]{} but is beyond the scope of this paper.
Disk features
-------------
The spiral arms and the outer edge of the inner cavity of the disk are clearly detected in the SPHERE data [with non-aggressive differential imaging techniques]{} (top panels of Figs. \[fig:ifsim\] and \[fig:irdisim\], and Fig. \[fig:spiralfit\]), providing a nice illustration of the instrument capabilities for high-contrast imaging of faint extended sources such as circumstellar disks. [The disk is also [clearly]{} seen in individual RDI IFS images]{} (Appendix \[sec:ifsimchannels\]). Except for an increase [in]{} the disk bulk contrast with respect to the star, we do not see any clear evidence [of]{} morphological spectral variations. The IRDIS and IFS images confirm that the disk spirals are launched axisymmetrically at the same corotation radius, as also found by @Stolker2016. We also detect in both IFS and IRDIS images the S3 spiral feature previously noted by @Garufi2013 and @Stolker2016 in the east part of the S2 spiral arm, especially at the longest wavelengths. We finally note that the outer rim looks brighter in the IRDIS $K$-band image (top panel of Fig. \[fig:irdisim\]).
We do not recover in the SPHERE images the radially extended spots to the northeast and the southwest seen in GPI total intensity $J$-band images by @Wahhaj2015. The latter suggested that they could be accretion streams from the disk into the star. Nevertheless, they mentioned that they could also be ADI and/or GPI artifacts because of the small field of view rotation of the data and [the]{} presence of similar features in other GPI data [e.g., @Perrin2015]. Our dataset covers a field of view rotation 1.8 times larger [than]{} the GPI data, which suggests that the features detected in the GPI data are probably artifacts.
@Stolker2016 detected shadow lanes in SPHERE polarized light images covering the $R$ to the $J$ bands (0.6–1.25 $\muup$m) obtained in late March and early May 2015. [Figure]{} \[fig:polarproj\] shows the polar projections of the $r^2$-scaled RDI IFS and IRDIS images with the positions of the shadow lanes in the IRDIS $J$-band polarized image obtained on 2015 May 2, i.e. about two weeks prior to these observations. Features A, B, and D [[notations from @Stolker2016]]{} are recovered in both polar-projected images, but feature C is only seen in the IFS image. Feature C was not seen in the SPHERE data obtained in late March 2015. The minima in the total intensity images are not as pronounced as in the polarized light images. This might be caused by sensitivity effects.
Photometry and morphology of the spiral arms
--------------------------------------------
We [used the RDI IFS and IRDIS images]{} (top panels of Figs. \[fig:ifsim\] and \[fig:irdisim\]) to derive the radial surface brightness profiles of the disk along its major and minor axes (Fig. \[fig:brightnessprof\]), assuming that the major axis has a position angle of 62$^{\circ}$ [@Perez2014]. For IFS, we analyzed [the $Y$, $J$, and $H$ spectral bands separately]{}. We used radial cuts through the disk of [1 pixel in width]{}. The conversion of the intensity into mag/arcsec$^2$ was [performed]{} using the 2MASS stellar magnitudes [$J$=7.279, $H$=6.587, and $K$=5.843, @Cutri2003] and the ratio of the maximum to the total flux of the measured [unsaturated non-coronagraphic PSF]{}. With this normalization choice, [the profiles provide information on the scattering efficiency of the dust grains with the wavelength [see @Quanz2012]]{}. We did not include the error on the photometric calibration. The noise levels (horizontal dashed lines) were estimated at large separation from the star.
[We note]{} [the similar overall shapes of the profiles of]{} the spectral bands with bumps associated to the two spiral arms at the same locations. Since the spiral pattern is not centrosymmetric (the opening angle of S1 decreases abruptly by $\sim$15$^{\circ}$ at the location of [its bright blob feature]{}, cf. Figs. \[fig:ifsim\] and \[fig:irdisim\]), the locations of the bumps are not centrosymmetric. The S/N of the disk detection drops quickly exterior to the spiral arms. The IFS radial profiles also show a sharp increase [($\sim$2.5 mag)]{} from the $Y$ to the $H$ bands, which suggests an increasing efficiency of the scattering with the wavelength. An increasing trend (although less significant) was also seen in the polarized-light radial profiles from the $R$ to the $J$ bands presented in @Stolker2016. The IRDIS $K$-band profile does not seem to confirm the increasing trend observed for the IFS data. As discussed in @Stolker2016, the red colors of the dust grains could be explained if they are composed of $\muup$m-sized aggregates [@Min2016].
Then, we registered the spine, photometry, and width of each spiral arm in the [collapsed RDI IFS and IRDIS images]{}. The images were first deprojected assuming for the disk an inclination of 11$^{\circ}$ [@Dent2005; @Pontoppidan2008; @Lyo2011] and a position angle of 62$^{\circ}$ [@Perez2014] and scaled by the square of the distance to the star. Then, [we rotated the images by steps of 2$^{\circ}$ in]{} order to align the spiral parallel to the horizontal direction and [to fit it with a 1D]{} Gaussian function using boxes of $\sim$0.15$''$ width (i.e., 20 pixels for IFS and 12 pixels for IRDIS) width approximately centered on the spiral spine to extract the location of the spine, the intensity at the spine location, and the width of the spiral. The error bars on the data points were assessed using their standard deviation in a [three-bin]{} sliding box.
We [show the width of the spirals as a function of the position angle in Fig. \[fig:widthspirals\]]{}. Each spiral arm exhibits [a]{} similar overall shape [in]{} the IFS and IRDIS images. As the position angle increases (i.e., going closer to the star), the IFS profile for S1 shows [alternating]{} maxima and minima. [The peak-to-peak variations are $\sim$25–40 mas]{}. The IFS profile for S2 has a constant width of $\sim$60 mas in its external part then exhibits a slightly increasing slope for position angles larger than $\sim$140$^{\circ}$. We note some [local differences in the IRDIS profiles with respect to the IFS data that can be attributed to a poorer spatial resolution or sensitivity effects]{} (shallower minima depths).
[Figure]{} \[fig:spiralpeak\] shows the surface brightness profiles along the spiral spines for the IFS and IRDIS bands. [We retrieve the]{} increasing trend of the disk surface brightness with the wavelength seen for the radial surface brightness profiles (Fig. \[fig:brightnessprof\]), especially for the IFS bands.
Finally, we extracted from the deprojected and $r^2$-scaled RDI images the reflectance spectra integrated [on the S1 bright blob feature, the S1 regions outside its blob feature, and S2]{}. Figure \[fig:reflectancespectrum\] shows the spectra normalized to the [stellar spectrum measured in the unsaturated non-coronagraphic data]{}. The error bars on the measurements account for the errors in the photometric calibration [and uncertainties in the RDI subtraction]{}. We removed the spectral parts at $\sim$1.34–1.43 $\muup$m because of a strong water telluric absorption band which affects significantly the S/N of the corresponding images. [The S1 region is brighter than S2,]{} as previously noted [@Garufi2013], because of the contribution from [its bright blob feature]{}. [The spectra of the S1 blob feature and of S2 show a red slope in the $H$ band, whereas the spectrum of S1 outside its blob feature looks rather flat. The spectrum of S2 [looks]{} flat in the $K$ band, where S1 appears slightly red. We also note spectral differences at the shortest [wavelengths;]{} S2 and the S1 blob [look red,]{} while the S1 regions outside its blob feature have a flatter spectrum. However, we stress that these [S1]{} regions are poorly detected at wavelengths shorter than 1.13 $\muup$m because of the overlapping of remaining residuals from the AO correction radius ($R_{\rm{corr}}$=$N_{\rm{act}}$$\times$$\lambda/2D$, with $N_{\rm{act}}$ the number of actuators on one side of the deformable mirror). The scattering efficiency [is higher]{} at longer wavelengths for S2, while for S1 it increases [across]{} the $YJH$ bands [and]{} then flattens in the $K$ band. This feature was also seen for the surface brightness profiles (Figs. \[fig:brightnessprof\] and \[fig:spiralpeak\]). The spectral differences between S1 and S2 and within S1 could indicate that the surface dust grains might have different optical properties according to their location, which can be [a]{} sign of different composition or size.]{} The SPHERE reflectance spectra of SAO 206462 look somewhat different from the SPHERE spectra of HR 4796A [@Milli2017], which show an increasing slope at $\sim$1–1.1 $\muup$m [and]{} then a plateau in the $J$ band.
Summary
-------
The analysis of the SPHERE total intensity images confirm the red colors for the dust grains and the presence of the shadow lanes [(though at lower significance)]{} found by @Stolker2016 in SPHERE polarized-light images. They also indicate [sharp variations [($\sim$40%)]{} in the S1 width]{}, while the S2 width does not show strong variations. Finally, the reflectance spectra suggest [local spectral differences between S1 and S2 and within S1]{}, which might be a hint [of]{} different composition or sizes of the dust grains.
Detection limits on putative giant protoplanets {#sec:detectionlimits}
===============================================
Archival NaCo/$L^{\prime}$ imaging data {#sec:nacodata}
---------------------------------------
We reduced and analyzed archival NaCo/$L^{\prime}$ imaging data (ESO program 090.C-0443, P.I. Currie) using a custom pipeline for the reduction steps (cosmetics, frame registering) and the SPHERE consortium image processing tools (Sect. \[sec:datareduc\]). SAO 206462 [$L$=4.89, @Coulson1995] was observed on 2013 March 24 UT around the meridian passage (airmass start/end 1.09–1.14) for $\sim$3.33h in pupil-tracking and saturated imaging modes. The star was regularly dithered between the bottom quadrants of the detector for sky background measurements. The observing conditions were good at the beginning of the sequence (DIMM seeing 0.5–0.7$''$ and coherence time $\sim$3ms)[, but deteriorated]{} with time (DIMM seeing 1.1–1.4$''$ and coherence time $\sim$1.5ms at the end of the observations). [Two hundred]{} data cubes of 160 co-adds of 0.25s were recorded covering a field of view rotation of 152.2$^{\circ}$. After binning the frames by groups of 100 and removing poor-quality frames based on the statistics of the total flux measured in an annulus of inner and outer radii 0.25$''$ and 0.5$''$, we were left with 226 images ($\sim$71% of the complete sequence). The inner and outer radii of the annulus were chosen to exclude the regions dominated by the bright stellar PSF wings and the background noise. The unsaturated PSF was recorded several times at regular intervals during the sequence using the same individual integration time and number of co-adds with the ND\_LONG neutral density filter (transmission $\sim$1.78%) for a total integration time of 480s. No point source was detected close to the star. We computed the TLOCI 5$\sigma$ detection limits using the same parameters as in Sect. \[sec:datareduc\] [and corrected them for the small sample statistics [@Mawet2014].]{}
[The SPHERE and NaCo]{} contrast limits are shown in the left panel of Fig. \[fig:detlimits\]. The thermal background increases with the wavelength, hence the poorer contrast performance observed for IRDIS in the $K2$ band and NaCo in the $L^{\prime}$ band at large separations. The SPHERE contrasts are deeper than the NaCo contrasts by factors [greater]{} than $\sim$10 in the speckle-limited regions ($\lesssim$0.6$''$), illustrating the improved performances of new dedicated high-contrast instruments like SPHERE and GPI.
Comparison of the detection limits to planet predictions
--------------------------------------------------------
Several predictions for planet(s) shaping the disk of SAO 206462 have been proposed [@Muto2012; @Garufi2013; @Fung2015; @Bae2016; @vanderMarel2016b]. Using linear equations from the spiral density wave theory, @Muto2012 suggested two planets with separations beyond $\sim$50 au by fitting independently the two spiral arms seen in Subaru/HiCIAO data and with masses of $\sim$0.5 $M_{\rm{J}}$ by using the amplitude of the spiral wave. @Garufi2013 proposed that one planet of mass [5–13 $M_{\rm{J}}$]{} located inside the cavity [at a separation of 17.5–20 au]{} could be responsible for the different cavity sizes measured for the small and large dust grains. @Fung2015 presented scaling relations between the azimuthal separation of the primary and secondary arms and the planet-to-star mass ratio for a single companion on a circular orbit with a mass between Neptune mass and 16 $M_{\rm{J}}$ around a 1 $M_{\odot}$ star. They predicted [with 30% accuracy that a single putative planet responsible for both spiral features of SAO 206462 would have a mass of $\sim$6 $M_{\rm{J}}$]{}. @Bae2016 presented dedicated hydrodynamical simulations of the SAO 206462 disk and proposed that both the bright scattered-light feature [@Garufi2013] and the dust emission peak [@Perez2014] seen for the southwestern spiral arm result from the interaction of the spiral arm with a vortex, although a vortex alone can account for the [S1 brightness peak]{}. They suggested that a 10–15 $M_{\rm{J}}$ planet may orbit at 100–120 au from the star. However, ALMA observations at two different frequencies seem to contradict a dust particle trapping scenario by a vortex [@Pinilla2015]. @Stolker2016 performed new fitting of the spiral arms observed in SPHERE data and found a best-fit solution with two protoplanets located exterior to the spirals: $r_1$$\sim$168 au, $\theta_1$$\sim$52$^{\circ}$ and $r_2$$\sim$99 au, $\theta_2$$\sim$355$^{\circ}$. @vanderMarel2016b proposed that the features seen in thermal emission in ALMA data and the scattered-light spiral arms are produced by a single massive giant planet located inside the cavity at a separation of $\sim$30 au. [Recently, @Dong2017 used the contrast of the spiral arms to predict a giant planet of $\sim$5–10 $M_{\rm{J}}$ at $\sim$100 au.]{}
We present [the SPHERE and NaCo mass limits in the right panel of Fig. \[fig:detlimits\]]{}. We assumed a system age of 9 Myr [@Mueller2011] and the predictions of the [atmospheric and evolutionary “hot-start” models of @Baraffe2015 and @Baraffe2003 for the SPHERE limits and of @2012RSPTA.370.2765A and @Baraffe2003 for the NaCo limits]{}, respectively[^7]. The detection limits are deeper with respect to the work of @Vicente2011, with the [exclusion of companions more massive than $\sim$14–12 $M_{\rm{J}}$ [in the range 0.12–0.2$''$ (19–31 au)]{}, $\sim$8 $M_{\rm{J}}$ beyond 0.5$''$ ($\sim$80 au) and $\sim$4 $M_{\rm{J}}$ beyond 0.6$''$ ($\sim$90 au)]{}. They do not allow us to strongly constrain the predictions of @Muto2012, @Garufi2013, and @vanderMarel2016b. They seem at odds with the predictions of a massive ($\sim$4–15 $M_{\rm{J}}$) giant planet in the outer disk by @Fung2015, @Bae2016[, and @Dong2017]{}. However, such models are expected to be degenerate. A less massive planet closer to the spiral arms should produce perturbations of similar amplitude [to]{} a more massive planet farther out. Nevertheless, the pitch angle of the produced spiral arms changes with the location of the planet (which also corresponds to the launching point of the spiral wave) because of its dependency on the sound speed and disk temperature. For SAO 206462, the planet should still be more massive than $\sim$2 $M_{\rm{J}}$ according to the predictions of @Fung2015 to account for the large pitch angle observed between the spiral arms ($\sim$130$^{\circ}$).
Estimating detection limits from high-contrast images in terms of planet mass depends on the assumptions for the system age and the mass-luminosity relation [e.g., @Marley2007; @Spiegel2012; @Mordasini2012a]. We note that for the upper age limit for the system of 16 Myr [@vanBoekel2005] the SPHERE [$K1$-band mass limits are $>$11–7 $M_{\rm{J}}$ beyond 0.5–0.7$''$]{} according to the [models of @Baraffe2015 [@Baraffe2003]]{}. Hence, a massive giant planet formed following a “hot-start” formation mechanism would have been detected in the SPHERE data if located exterior to the spiral features. A massive giant planet in [a]{} wide orbit such as predicted by @Fung2015, [@Bae2016, and @Dong2017]{} would still be compatible with the SPHERE constraints if formed according to a “warm-start” scenario with low initial entropy [for e.g., a 10-$M_{\rm J}$ planet with age 16 Myr and initial entropy [below 10 $k_B$/baryon]{} could not be excluded assuming the models of @Spiegel2012].
[Another possible planet scenario for SAO 206462 would be a low-mass protoplanet embedded in the disk and surrounded by a circumplanetary disk. This scenario has been proposed for planet companions/candidates in the transitional disks of HD 100546 [@Quanz2013a; @Quanz2015; @Currie2015] and HD 169142 [@Biller2014; @Reggiani2014]. The mass-luminosity relations of standard evolutionary models [e.g., @Baraffe2003] assume giant planets without circumplanetary disks. A giant planet surrounded by a circumplanetary disk could be significantly brighter than the same planet without a disk, hence the mass-luminosity relations of standard evolutionary models could be pessimistic. Therefore, the mass limits shown in Fig. \[fig:detlimits\] could be upper mass limits for a given system age. However, a lower-mass giant planet with a circumplanetary disk would exert weaker gravitational perturbations on the disk than a more massive giant planet without a disk and produce a different spiral shape (e.g., smaller opening angle, smaller spiral contrast). [Modeling]{} studies of disk spirals suggested that lower planet mass limits could be estimated using an observed disk spiral pattern [@./figs/fung2015 see previous paragraph].]{}
Summary
-------
The SPHERE near-IR detection limits improve significantly the constraints on putative planets in the SAO 206462 disk with respect to the study of @Vicente2011 and the detection limits measured in archival NaCo thermal IR images. The SPHERE data exclude massive giant planets ($>$3 $M_{\rm J}$ assuming a “hot-start” scenario and a system age of 9 Myr) exterior to the spiral arms, which may rule out a few recent [predictions based on spiral modelling [@Fung2015; @Bae2016; @Dong2017]]{}. However, we cannot exclude low-mass giant planets in the outer disk and/or giant planets inside the scattered-light cavity, as predicted in other studies [@Muto2012; @Garufi2013; @vanderMarel2016b]. These planets could still account for the morphology of the SAO 206462 spirals.
[Effect]{} of protoplanetary dust on the detectability of embedded planets {#sec:dustopacity}
==========================================================================
The detection of planets embedded in protoplanetary disks is expected to be hampered by the dust grains residing between the disk midplane and surface [e.g., @Quanz2015], hence Fig. \[fig:detlimits\] likely provides limits of detection [that are too optimistic]{}. However, a massive planet will open a (partial) gap [e.g., @Crida2006; @Malik2015] leading to a smaller attenuation of a planet’s thermal emission. Since the dust opacity decreases with the wavelength, the SPHERE/$YJHK$ detection limits are expected to be more affected than the NaCo/$L^{\prime}$ detection limits.
Methods and assumptions
-----------------------
We determined the minimum masses of embedded planets that would be detectable given the derived IRDIS/$K1$ and NaCo/$L^{\prime}$ contrast limits (Fig. \[fig:detlimits\]) by assuming a simple disk structure and considering [theoretical mass-luminosity relations for giant planets without circumplanetary disk from @Baraffe2015 [@Baraffe2003]]{} at the age of the SAO 206462 system. We used DIANA standard dust opacities [@Woitke2016] in the $K1$ and $L^{\prime}$ bands of 330 and 250 cm$^2$/g, respectively, and a surface density profile inversely proportional to the distance to the star. We assumed a total dust mass of $2\times 10^{-4}$ $M_{\odot}$, an [inner and outer disk radius of 28 and 300 au]{} [@Carmona2014] and a zero surface density within the [scattered light cavity (28 au)]{}. Since the disk is observed close to face-on [@vanderMarel2016a], we have neglected projection effects.
![Depth of gaps opened by putative giant planets in the disk of SAO 206462 as a function of the planet-to-star mass ratio computed from [the empirical relations in @Fung2014]{} for the two cases considered in this paper (see text). [For the pessimistic case (dashed curve) and a planet-to-star mass ratio below $\sim$3.5$\times$10$^{-3}$ (i.e., planet mass below $\sim$6 $M_{\rm J}$), a planet cannot open a gap in the disk, hence the dust column density above the planet is not reduced.]{}[]{data-label="fig:planetgapdepths"}](./figs/f13){width="44.00000%"}
Then, we take a planet gap depth into account by following the empirical relations of [@Fung2014]{} $$\begin{aligned}
\frac{\Sigma_{\rm p}}{\Sigma_0} & = & 0.14\left( \frac{q}{10^{-3}} \right)^{-2.16} \left(\frac{\alpha}{10^{-2}} \right)^{1.41} \left( \frac{H/r}{0.05} \right)^{6.61} \label{eq:fung1}\\
& & \mathrm{for}~10^{-4} \leq q \leq 5\times10^{-3}, \nonumber \\
\frac{\Sigma_{\rm p}}{\Sigma_0} & = & 4\times10^{-3}\left( \frac{q}{5\times10^{-3}} \right)^{-1.00} \left(\frac{\alpha}{10^{-2}} \right)^{1.26} \left( \frac{H/r}{0.05} \right)^{6.12} \label{eq:fung2} \\
& & \mathrm{for}~5\times10^{-3} \leq q \leq 10^{-2} \nonumber,\end{aligned}$$ where $\Sigma_{\rm p}$ is the surface density at the orbital radius of the planet, $\Sigma_0$ the initial surface density before a gap was formed, $q$ the planet-to-star mass ratio, $\alpha$ the dimensionless Shakura–Sunyaev viscosity parameter and $H/r$ the disk aspect ratio. Eqs. (\[eq:fung1\]) and (\[eq:fung2\]) are strongly dependent on the assumed properties for the disk and the dust grains, which are not well constrained by observations and/or [modeling]{} of the disk of SAO 206462. We had to make assumptions on several of these parameters for our analysis in the remainder of this section. We note that our quantitative results are highly sensitive to our parameter choice. @Fung2014 investigated the morphology of planet gaps up to mass ratios of 10$^{-2}$ (i.e., planet masses below 17 $M_{\rm J}$ in the case of SAO 206462), while the [NaCo and IRDIS detection limits]{} probe planet-to-star mass ratios above this limit at close-in separations beyond the scattered-light cavity (Fig. \[fig:detlimits\]). [For these cases, we simply used Eq. (\[eq:fung2\])]{}. For the disk aspect ratio and viscosity, we considered two extreme cases shown in Fig. \[fig:planetgapdepths\]: (1) deep planet gaps by assuming $H/r=0.05$ and $\alpha=10^{-2}$ ([for]{} this value of $H/r$, the assumed value for $\alpha$ has little effect on the derived results) and (2) shallow planet gaps corresponding to $H/r=0.1$ and $\alpha=10^{-2}$. The vertical optical depth to the disk midplane is now given by $\tau = \kappa \Sigma/2$ , with $\kappa$ the dust opacity, and the attenuation of a planet’s thermal emission is a factor $e^{-\tau}$.
Mass limits vs. disk aspect ratio and viscosity
-----------------------------------------------
Figure \[fig:planetmassesopacity\] shows the IRDIS/$K1$ and NaCo/$L^{\prime}$ planet mass limits for the two cases of protoplanetary dust attenuation compared to the limits for which the attenuation has been neglected. We note that for a disk aspect ratio of 0.05, the attenuation by the dust is negligible even for a [high]{} viscosity because the disk properties are more favorable to the formation of deep gaps (Fig. \[fig:planetgapdepths\]). For a disk aspect ratio of 0.1 combined with a [high]{} viscosity, the effect is significant at all separations beyond the scattered-light cavity (for $\alpha=10^{-3}$, the mass limits are degraded by less than $\sim$0.5 $M_{\rm J}$). This is expected because the planet gaps are shallower (Fig. \[fig:planetgapdepths\]). In particular, a planet less massive than $\sim$6 $M_{\rm J}$ cannot open a gap so that the dust column density above the planet is not reduced for this mass range. We also note that the IRDIS/$K1$ mass limits are deeper than the NaCo/$L^{\prime}$ limits. The decreased sensitivity to the protoplanetary dust opacity in the $L^{\prime}$ band does not compensate for the poorer contrasts provided by the NaCo observations. Higher contrast observations in the $L^{\prime}$ band by $\sim$1 mag could compete with our SPHERE data.
![[Planet mass detection limits (5$\sigma$) for the IRDIS $K1$-band and NaCo $L^{\prime}$-band images, both with the effect of attenuation of a planet emitted flux by protoplanetary dust (dashed and dash-dotted curves, see text) compared to the case where the attenuation by protoplanetary dust is neglected (solid curves, see Fig. \[fig:detlimits\]).]{} []{data-label="fig:planetmassesopacity"}](./figs/f14){width="46.00000%"}
Detectability of gaps opened by giant embedded planets
------------------------------------------------------
[In this section we test]{} the detectability of the gaps opened by putative embedded planets with the masses probed by SPHERE in Fig. \[fig:planetmassesopacity\]. In particular, we want to investigate the detectability of a planet gap beyond the spiral arms because several authors [have]{} suggested that a single massive ($\sim$4–15 $M_{\rm J}$) planet located in these regions might be responsible for both spiral features [@Dong2015; @Zhu2015; @Fung2015; @Bae2016]. Our SPHERE/IRDIS observations seem to reject the high-mass tail of these predictions, since the mass limits beyond the spiral arms are $\sim$5 $M_{\rm J}$ (see red dashed curve for $H/r$=0.05 and $\alpha$=0.01 in Fig. \[fig:planetmassesopacity\]). However, a giant planet of $\sim$4 $M_{\rm J}$ [close]{} to the spiral arms ($\sim$0.6–0.65$''$ if $H/r$=0.05 and $\alpha$=0.01) would still be compatible with the detection limits [and]{} could account for the spiral morphology.
For this analysis, we considered [the case $H/r=0.05$ and $\alpha=10^{-2}$]{}. This case allows for the detection of lower-mass giant planets in the disk, hence the widths of the gaps produced by these planets may be considered as lower limits. We show in Fig. \[fig:planetgaps\] the width of the planet gaps as a function of the orbital radius[^8]. The Hill radius $R_{\rm H}$ is a first-order estimate of the gap width in the gas and can be applied to gaps in the small dust grains (traced in scattered light) assuming that they are coupled to the gas. However, this is a lower limit since the gas gap width can be as large as 5 $R_{\rm H}$ [@Dodson2011]. Therefore, we also used the [empirical formula from @Kanagawa2016, derived from 2D hydrodynamical simulations with planets of mass ratios 10$^{-4}$–2$\times$10$^{-3}$ (i.e., $\sim$0.2–4 $M_{\rm J}$ in the case of SAO 206462)]{}. When comparing the latter gap widths to the SPHERE resolution, we find that the gaps are significantly larger, meaning that a massive planet that is located beyond the spiral arms might have opened a detectable gap. Interestingly, the disk was marginally detected in scattered light beyond the spiral arms up to 1$''$ by @Stolker2016 without any indication of a gap in that region. [Given that the disk is not exactly seen face-on, we finally note that we]{} cannot rule out shadowing of the gaps by their outer/inner wall because of the uncertainties in the disk scale height.
However, scattered-light observations probe the disk surface so the detectability of a planet gap depends on the disk aspect ratio which, for a flaring disk, increases toward larger disk radii [@Crida2006]. Sub-mm data, like those provided by ALMA, probe the dust grains located in the disk midplane, [and]{} hence are more relevant for such an analysis. [Figure]{} \[fig:planetgaps\] shows the approximate gap width (7 $R_{\rm H}$) in the large, mm-sized grains as a result of dust trapping by a gas giant planet of [mass ratio 10$^{-3}$–3$\times$10$^{-3}$ in a pressure maximum]{} [@Pinilla2012]. The gap width is also larger than the angular resolution of recent ALMA observations [@vanderMarel2016b]. However, the disk appears to be compact in both $^{13}$CO and dust continuum [e.g., @vanderMarel2016a] with only a marginal detection of the disk beyond 0.5$''$. This feature might indicate that the disk is truncated by a companion located beyond the spiral arms, which would support the interpretation by @Dong2015. On the other hand, the low S/N of the sub-mm detection beyond 0.5$''$ might have been limited by the sensitivity of the ALMA observations.
Summary
-------
[Since the disk dust may attenuate the signal of an embedded planet and lead to an underestimation of the mass detection limits, we estimated the amplitude of this potential effect for the SAO 206462 disk.]{} The analysis is in fact strongly dependent on the assumptions on the disk properties, especially its aspect ratio and viscosity. By considering two extreme cases for the planet gap depths, [we showed that the mass limits in Fig. \[fig:detlimits\] might be underestimated by up to $\sim$2 $M_{\rm J}$ exterior to the spiral arms]{}. Then, we analyzed the detectability of gaps opened by planets with masses compatible with the detection limits in the [case of deep planet gaps]{}. The comparison of the gap widths for small dust grains to the SPHERE resolution suggests that such gaps might have been resolved outside the scattered-light cavity. Nevertheless, the sensitivity of the scattered-light data to planet gaps is potentially hampered by the disk scale height, which is poorly constrained for SAO 206462. Therefore, we considered ALMA data and found that planet gaps in the mm-sized grains could also have been resolved exterior to the cavity measured at these wavelengths. While current ALMA data exclude such gaps at separations [up to 80 au]{}, they cannot exclude them at larger separations because of their limited sensitivity. [We note that very little dust/gas is detected outside of the spiral arms [where]{} putative planets are predicted to reside, so both ALMA and SPHERE data do not strongly constrain the presence of putative planet gaps.]{}
Conclusions
===========
We presented new high-resolution and high-contrast observations with the instrument VLT/SPHERE of the transitional disk of SAO 206462 to search for giant embedded planets responsible for its spiral features and, in [the]{} case of non-detection, constrain recent predictions for such planets. We also exploited the spatial and spectral resolutions offered by SPHERE to analyze the colors of the disk spiral arms and to search for spatial variations in the spectral properties of the dust grains at the surface of the disk. The main results of the analysis are [as follows]{}:\
*Search for giant protoplanets and constraints on predictions*
1. The IRDIS data [shows]{} a point source at a separation of $\sim$5$''$ in addition to the known background stellar binary at the same separation range. From the IRDIS colors and the IRDIS and HST/STIS astrometry of the point source, we [classify]{} it as a background star.
2. [Considering that the disk dust may attenuate the signal of a planet embedded within a disk, we evaluated that the mass detection limits computed from the contrast achieved in the SPHERE data may be underestimated by up to $\sim$2 $M_{\rm J}$ at separations exterior to the spiral arms]{}.
3. The SPHERE detection limits allowed us to exclude the presence of planets more massive than $\sim$5–7 $M_{\rm J}$ beyond the spiral arms, which provides new constraints for the hydrodynamical [modeling]{} of the disk spiral arms. The non-detection of planet gaps up to [separations of $\sim$80 au]{} in ALMA data do not favor the presence of massive giant planets in the outer disk close to the spiral arms, although we cannot exclude that the non-detection might be related to a lack of resolution of the instrument. [However, very little disk signal is detected in the ALMA data outside the spiral arms, so the observational constraints on the presence of planet gaps are loose.]{} A giant planet of $\sim$4–6 $M_{\rm J}$ in the outer disk would still be compatible with the SPHERE data [and]{} can account for the spiral morphology. Another possibility would be that the spiral arms are driven by one or more giant planets located inside the scattered-light cavity.
4. Although NaCo/$L^{\prime}$-band data are less sensitive to protoplanetary dust opacity than SPHERE near-IR observations, the poorer contrasts achieved in archival NaCo data in this band make the mass limits shallower than SPHERE/IRDIS limits in the $K1$ band.
*Imaging and spectrophotometry of the disk features*
1. We confirmed the shadow lanes detected in SPHERE polarized data [@Stolker2016], although at lower significance.
2. We found sharp [variations of $\sim$40% for the width of spiral arm S1]{}, while the width of spiral arm S2 does not exhibit strong variations.
3. We found [a factor of $\sim$10 increase]{} for the disk surface brightness from the $Y$ to the $K$ bands, hinting at an increasing scattering efficiency of the dust grains at the surface of the disk. This could be explained by the presence of $\muup$m-sized dust aggregates.
4. The reflectance spectra suggest small spectral variations for the dust properties between [spiral arms S1 and S2 and between the bright blob feature and the other parts of S1]{}.
[In order to perform further searches]{} for protoplanets in the SAO 206462 disk, dedicated SPHERE observations might [increase]{} the contrast/mass constraints beyond the disk spirals. Another method of interest would be to use H$\alpha$ imaging using SPHERE or MagAO given the system young age and that young protoplanets may still accrete. In addition, sparse aperture masking observations with 8–10 m ground-based telescopes will allow [the scattered-light cavity for substellar companions to be probed]{} as close as $\sim$50–200 mas [($\sim$8–31 au)]{}, well within the angular resolution provided by high-contrast coronagraphic imagers on the same class of telescopes, but at limited contrasts of $\sim$10$^{2}$–10$^{3}$ [[$\sim$30–300 $M_{\rm{J}}$ in the $L^{\prime}$ band for the age of SAO 206462 according to the predictions of @2012RSPTA.370.2765A; @Baraffe2003]]{}. We note that putative embedded planets surrounded by circumplanetary disks could be detected with these instruments down to significantly lower masses. In the near future, JWST imaging will probe for very low-mass planets outside the spiral arms, while high-contrast imagers on 30–40 m telescopes will be required to search the scattered-light cavity for planetary-mass companions.
New promising possibilities for disk analyses are now offered by the high angular resolution combined with the spectral and/or polarimetric capabilities of the high-contrast imaging instruments SPHERE, GPI, and CHARIS. In particular, the spectra of distinct parts of a disk can be compared to search for differences indicating the presence of dust grain populations with different properties (e.g., scattering efficiency, size).
Finally, we outline the complementarity of scattered-light and thermal emission information[, such]{} as those brought by SPHERE and ALMA to constrain the presence of giant protoplanets in transitional disks. More multiwavelength analyses of such targets will help to better understand the relations between some features seen in these disks (spirals, gaps, rings) and putative giant embedded planets. These analyses will require further multiwavelength efforts.
The authors thank the ESO Paranal Staff for support in conducting the observations and Philippe Delorme and Eric Lagadec (SPHERE Data Center) for their help with the data reduction. We thank Myriam Benisty, Philippe Thébault, and an anonymous referee for helpful comments. A.-L.M., S.M., S.D., R.G., and D.M. acknowledge support from the “Progetti Premiali” funding scheme of the Italian Ministry of Education, University, and Research. We acknowledge financial support from the Programme National de Planétologie (PNP) and the Programme National de Physique Stellaire (PNPS) of CNRS-INSU. This work has also been supported by a grant from the French Labex OSUG@2020 (Investissements d’avenir – ANR10 LABX56). M.R.M and S.P.Q. acknowledge the financial support of the SNSF. The project is supported by CNRS, by the Agence Nationale de la Recherche (ANR-14-CE33-0018). This work has made use of the SPHERE Data Centre, jointly operated by OSUG/IPAG (Grenoble), PYTHEAS/LAM/CeSAM (Marseille), OCA/Lagrange (Nice) and Observatoire de Paris/LESIA (Paris). This research made use of the SIMBAD database and the VizieR Catalogue access tool, both operated at the CDS, Strasbourg, France. The original description of the VizieR service was published in Ochsenbein et al. (2000, A&AS 143, 23). SPHERE is an instrument designed and built by a consortium consisting of IPAG (Grenoble, France), MPIA (Heidelberg, Germany), LAM (Marseille, France), LESIA (Paris, France), Laboratoire Lagrange (Nice, France), INAF – Osservatorio di Padova (Italy), Observatoire astronomique de l’Université de Genève (Switzerland), ETH Zurich (Switzerland), NOVA (Netherlands), ONERA (France), and ASTRON (Netherlands), in collaboration with ESO. SPHERE was funded by ESO, with additional contributions from the CNRS (France), MPIA (Germany), INAF (Italy), FINES (Switzerland), and NOVA (Netherlands). SPHERE also received funding from the European Commission Sixth and Seventh Framework Programs as part of the Optical Infrared Coordination Network for Astronomy (OPTICON) under grant number RII3-Ct-2004-001566 for FP6 (2004–2008), grant number 226604 for FP7 (2009–2012), and grant number 312430 for FP7 (2013–2016).
IFS individual images {#sec:ifsimchannels}
=====================
Figure \[fig:ifsimchannels\] shows the IFS individual images in the $YJH$ bands. We note that the disk does not exhibit morphological spectral variations, except for a global increase in contrast with the wavelength.
Photometric variability of the host star
========================================
{width="85.00000%"}
As supporting observations for the SPHERE observations reported in this paper, we performed a photometric monitoring of the host star in order to analyze its photometric variability. In high-contrast observations, the photometric measurements are relative to the host star, so their accuracy is dependent on the stellar variability. SAO 206462 is known to show high photometric variability [$\sim$23% at 1.7 $\muup$m, @Grady2009].
The star was observed from August 14 to 31, 2015 for a total of four nights at the York Creek Observatory (41$^{\circ}$06$'$06.4$''$S, 146$^{\circ}$50$'$33$''$E, Georgetown, Tasmania) using a f/10 25cm Takahashi Mewlon reflector, equipped with a QSI 683ws-8 camera, and $B$, $V$, and $R$ standard Johnson-Cousins filters. The telescope has a field of view of 24.5$'$$\times$18.5$'$. The pixel scale is 0.44$''$/pix. A total of 51 frames were collected in the $V$ filter using an integration time of 20s. Aperture photometry was used to extract the magnitudes of SAO 206462 and other stars in the field to be used as comparison stars. All reduction steps were performed using the tasks within IRAF[^9]. The achieved photometric accuracy was $\sigma_V$=0.01 mag.
We could identify three stars whose light curves were very stable and therefore suitable as comparison stars. Their differential magnitudes during our observing run were found to be constant within our photometric precision. We performed the Lomb-Scargle periodogram analysis of differential $V$ light curve. In Fig. \[fig:photometricvar\] we summarize the results for the $V$-band observations. On the basis of stellar radius $R$=1.4 $R_{\odot}$ and projected rotational velocity $v$sin$i$=82.5 kms$^{-1}$ [@Mueller2011], the expected rotation period must be shorter than about 1 day. We searched for photometric periodicities in the 0.1–1.0-day interval. In the periodogram we found a number of significant power peaks that are related to the observation window function. However, after filtering the major power peak, the pre-whitened time series showed a significant periodicity at $P$=0.137 d that we assume to be the stellar rotation period. In a previous analysis, @Mueller2011 measured a rotation period $P$=0.160 d using a generalized Lomb-Scargle analysis of a series of FEROS radial velocity measurements covering a time period of 151 days. However, they also found in their periodogram two other peaks of comparable power at $P$=0.138 and $P$=0.191 d, the first peak being in good agreement with our estimate.
[^1]: Based on observations collected at the European Organisation for Astronomical Research in the Southern Hemisphere under ESO programmes 095.C-0298 and 090.C-0443.
[^2]: [All these studies assumed a system distance of 140 pc]{}.
[^3]: [They assumed for the star a distance of 140 pc and an age of 8 Myr, as well as the BT-DUSTY evolutionary tracks of @Allard2011. Using the contrasts measured at the corresponding angular separations and assuming the evolutionary models, distance and age used in this paper, we derive mass limits of $>$0.34 $M_{\odot}$ beyond 16 au and $>$28 $M_{\rm{J}}$ beyond 80 au.]{}
[^4]: The transmission curves of the neutral density filters can be found in the User Manual available at [www.eso.org/sci/facilities/paranal/instruments/sphere/doc.html](www.eso.org/sci/facilities/paranal/instruments/sphere/doc.html).
[^5]: The temporal variations are below 0.2 mas. For IRDIS, we measured the variations on the individual frames recorded before the sequence. [The diffraction limits for SPHERE data are $\sim$31, $\sim$42, $\sim$54, and $\sim$58 mas in the $J$, $H$, $K1$, and $K2$ bands, respectively. Our data are thus diffraction-limited [in]{} the $H$ and $K$ bands, but not in the $J$ band.]{}
[^6]: We also accounted for offsets of $-$135.99$\pm$0.11$^{\circ}$ for the derotator zeropoint and $+$100.48$\pm$0.10$^{\circ}$ for the IFS field orientation with respect to the IRDIS field [[@Maire2016b]]{}.
[^7]: [The [@Baraffe2015 atmospheric models]{} do not provide planet luminosities in the NaCo filters]{}.
[^8]: [The curves were cut to separations larger than the scattered-light cavity radius because we assumed zero surface density for the dust inside the cavity.]{}
[^9]: IRAF is distributed by the National Optical Astronomy Observatory, which is operated by the Association of the Universities for Research in Astronomy, Inc. (AURA), under cooperative agreement with the National Science Foundation.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Multilingual machine translation, which translates multiple languages with a single model, has attracted much attention due to its efficiency of offline training and online serving. However, traditional multilingual translation usually yields inferior accuracy compared with the counterpart using individual models for each language pair, due to language diversity and model capacity limitations. In this paper, we propose a distillation-based approach to boost the accuracy of multilingual machine translation. Specifically, individual models are first trained and regarded as teachers, and then the multilingual model is trained to fit the training data and match the outputs of individual models simultaneously through knowledge distillation. Experiments on IWSLT, WMT and Ted talk translation datasets demonstrate the effectiveness of our method. Particularly, we show that one model is enough to handle multiple languages (up to 44 languages in our experiment), with comparable or even better accuracy than individual models.'
author:
- |
Xu Tan$^{1}$[^1], Yi Ren$^{2}$, Di He$^{3}$, Tao Qin$^{1}$, Zhou Zhao$^{2}$ & Tie-Yan Liu$^{1}$\
\
$^{1}$Microsoft Research Asia\
`{xuta,taoqin,tyliu}@microsoft.com`\
\
$^{2}$Zhejiang University\
`rayeren,[email protected]`\
\
$^{3}$Key Laboratory of Machine Perception, MOE, School of EECS, Peking University\
`[email protected]`\
bibliography:
- 'iclr2019\_conference.bib'
title: Multilingual Neural Machine Translation with Knowledge Distillation
---
Introduction
============
Neural Machine Translation (NMT) has witnessed rapid development in recent years [@DBLP:journals/corr/BahdanauCB14; @DBLP:conf/emnlp/LuongPM15; @DBLP:journals/corr/WuSCLNMKCGMKSJL16; @DBLP:conf/icml/GehringAGYD17; @DBLP:conf/nips/VaswaniSPUJGKP17; @guo2018non; @shen2018dense], including advanced model structures [@DBLP:conf/icml/GehringAGYD17; @DBLP:conf/nips/VaswaniSPUJGKP17; @he2018layer; @xia2008tied] and human parity achievements [@DBLP:journals/corr/abs-1803-05567]. While conventional NMT can well handle single pair translation, training a separate model for each language pair is resource consuming, considering there are thousands of languages in the world[^2]. Therefore, multilingual NMT [@DBLP:journals/tacl/JohnsonSLKWCTVW17; @DBLP:conf/naacl/FiratCB16; @DBLP:journals/corr/HaNW16; @DBLP:journals/corr/abs-1804-08198] is developed which handles multiple language pairs in one model, greatly reducing the offline training and online serving cost.
Previous works on multilingual NMT mainly focus on model architecture design through parameter sharing, e.g., sharing encoder, decoder or attention module [@DBLP:conf/naacl/FiratCB16; @DBLP:journals/corr/abs-1804-08198] or sharing the entire models [@DBLP:journals/tacl/JohnsonSLKWCTVW17; @DBLP:journals/corr/HaNW16]. They achieve comparable accuracy with individual models (each language pair with a separate model) when the languages are similar to each other and the number of language pairs is small (e.g., two or three). However, when handling more language pairs (dozens or even hundreds), the translation accuracy of multilingual model is usually inferior to individual models, due to language diversity.
It is challenging to train a multilingual translation model supporting dozens of language pairs while achieving comparable accuracy as individual models. Observing that individual models are usually of higher accuracy than the multilingual model in conventional model training, we propose to transfer the knowledge from individual models to the multilingual model with *knowledge distillation*, which has been studied for model compression and knowledge transfer and well matches our setting of multilingual translation. It usually starts by training a big/deep teacher model (or ensemble of multiple models), and then train a small/shallow student model to *mimic* the behaviors of the teacher model, such as its hidden representation [@yim2017gift; @romero2014fitnets], its output probabilities [@hinton2015distilling; @freitag2017ensemble] or directly training on the sentences generated by the teacher model in neural machine translation [@DBLP:conf/emnlp/KimR16]. The student model can (nearly) match the accuracy of the cumbersome teacher model (or the ensemble of multiple models) with knowledge distillation.
In this paper, we propose a new method based on knowledge distillation for multilingual translation to eliminate the accuracy gap between the multilingual model and individual models. In our method, multiple individual models serve as teachers, each handling a separate language pair, while the student handles all the language pairs in a single model, which is different from the conventional knowledge distillation where the teacher and student models usually handle the same task. We first train the individual models for each translation pair and then we train the multilingual model by matching with the outputs of all the individual models and the ground-truth translation simultaneously. After some iterations of training, the multilingual model may get higher translation accuracy than the individual models on some language pairs. Then we remove the distillation loss and keep training the multilingual model on these languages pairs with the original log-likelihood loss of the ground-truth translation. We conduct experiments on three translation datasets: IWSLT with 12 language pairs, WMT with 6 language pairs and Ted talk with 44 language pairs. Our proposed method boosts the translation accuracy of the baseline multilingual model and achieve similar (or even better) accuracy as individual models for most language pairs. Specifically, the multilingual model with only $1/44$ parameters can match or surpass the accuracy of individual models on the Ted talk datasets.
Background
==========
Neural Machine Translation
--------------------------
Given a set of bilingual sentence pairs $D=\{(x,y) \in \mathcal{X}\times \mathcal{Y}\}$, an NMT model learns the parameter $\theta$ by minimizing the negative log-likelihood $-\sum_{(x,y) \in D} \log P(y|x;\theta)$. $P({y}|{x};\theta)$ is calculated based on the chain rule $\prod_{t=1}^{T_y} P(y_t|y_{<t}, x;\theta)$, where $y_{<t}$ represents the tokens preceding position $t$, and $T_y$ is the length of sentence $y$.
The encoder-decoder framework [@DBLP:journals/corr/BahdanauCB14; @DBLP:conf/emnlp/LuongPM15; @DBLP:conf/nips/SutskeverVL14; @DBLP:journals/corr/WuSCLNMKCGMKSJL16; @DBLP:conf/icml/GehringAGYD17; @DBLP:conf/nips/VaswaniSPUJGKP17] is usually adopted to model the conditional probability $P(y|x; \theta)$, where the encoder maps the input to a set of hidden representations $h$ and the decoder generates each target token $y_t$ using the previous generated tokens $y_{<t}$ as well as the representations $h$.
Multilingual NMT
----------------
While how to further improve the translation of a single language pair is a hot research topic [@DBLP:conf/nips/VaswaniSPUJGKP17; @DBLP:conf/icml/GehringAGYD17; @wu2018beyond; @song2018double; @gong2018sentence], a new trend is multilingual neural machine translation [@dong2015multi; @DBLP:journals/corr/LuongLSVK15; @DBLP:conf/naacl/FiratCB16; @DBLP:journals/corr/abs-1804-08198; @DBLP:journals/tacl/JohnsonSLKWCTVW17; @DBLP:journals/corr/HaNW16], considering the large amount of languages pairs in the world. Some of these works focus on how to share the components of the NMT model among multiple language pairs. @dong2015multi use a shared encoder but different decoders to translate the same source language to multiple target languages. @DBLP:journals/corr/LuongLSVK15 use the combination of multiple encoders and decoders, with one encoder for each source language and one decoder for each target language respectively, to translate multiple source languages to multiple target languages. @DBLP:conf/naacl/FiratCB16 share the attention mechanism but use different encoders and decoders for multilingual translation. Similarly, @DBLP:journals/corr/abs-1804-08198 design the neural interlingua, which is an attentional LSTM encoder to bridge multiple encoders and decoders for different language pairs. In @DBLP:journals/tacl/JohnsonSLKWCTVW17 and @DBLP:journals/corr/HaNW16, multiple source and target languages are handled with a universal model (one encoder and decoder), with a special tag in the encoder to determine which target language to translate. In @DBLP:conf/naacl/GuHDL18 [@gu2018meta] and @neubig2018rapid, multilingual translation is leveraged to boost the accuracy of low-resource language pairs with better model structure or training mechanism.
It is observed that when there are dozens of language pairs, multilingual NMT usually achieves inferior accuracy compared with its counterpart which trains an individual model for each language pair. In this work we propose the multilingual distillation framework to boost the accuracy of multilingual NMT, so as to match or even surpass the accuracy of individual models.
Knowledge Distillation
----------------------
The early adoption of knowledge distillation is for model compression [@buciluǎ2006model], where the goal is to deliver a compact student model that matches the accuracy of a large teacher model or the ensemble of multiple models. Knowledge distillation has soon been applied to a variety of tasks, including image classification [@hinton2015distilling; @furlanello2018born; @yang2018knowledge; @anil2018large; @li2017learning], speech recognition [@hinton2015distilling] and natural language processing [@DBLP:conf/emnlp/KimR16; @freitag2017ensemble]. Recent works [@furlanello2018born; @yang2018knowledge] even demonstrate that student model can surpass the accuracy of the teacher model, even if the teacher model is of the same capacity as the student model. @zhang2017deep propose the mutual learning to enable multiple student models to learn collaboratively and teach each other by knowledge distillation, which can improve the accuracy of those individual models. @anil2018large propose online distillation to improve the scalability of distributed model training and the training accuracy.
In this paper, we develop the multilingual distillation framework for multilingual NMT. Our work differs from @zhang2017deep and @anil2018large in that they collaboratively train multiple student models with codistillation, while we use multiple teacher models to train a single student model, the multilingual NMT model.
Method
======
As mentioned, when there are many language pairs and each pair has enough training data, the accuracy of individual models for those language pairs is usually higher than that of the multilingual model, given that the multilingual model has limited capacity comparing with the sum of all the individual models. Therefore, we propose to teach the multilingual model using the individual models as teachers. Here we first describe the idea of knowledge distillation in neural machine translation for the case of one teacher and one student, and then introduce our method in the multilingual setting with multiple teachers (the individual models) and one student (the multilingual model).
One Teacher and One Student
---------------------------
Denote $D=\{(x,y) \in \mathcal{X}\times \mathcal{Y}\}$ as the bilingual corpus of a language pair. The log-likelihood loss (cross-entropy with one-hot label) on corpus $D$ with regard to an NMT model $\theta$ can be formulated as follows: $$\begin{aligned}
\mathcal{L}_{\text{NLL}} (D; \theta) =& -\sum_{(x, y) \in D} \log P(y|x;\theta), \\
\log P(y|x;\theta)= & \sum_{t=1}^{T_y} \sum_{k=1}^{|\mathcal{V}|} \mathbbm{1}\{y_t=k\}\log P(y_t=k|y_{<t}, x;\theta),
\end{aligned}
\label{nll_loss}$$ where $T_y$ is the length of the target sentence, $|V|$ is the vocabulary size of the target language, $y_t$ is the $t$-th target token, $\mathbbm{1}\{\cdot\} $ is the indicator function that represents the one-hot label, and $P(\cdot|\cdot)$ is the conditional probability with model $\theta$.
In knowledge distillation, the student (with model parameter $\theta$) not only matches the outputs of the ground-truth one-hot label, but also to the probability outputs of the teacher model (with parameter $\theta_T$). Denote the output distribution of the teacher model for token $y_t$ as $Q(y_t|y_{<t}, x;\theta_{T})$. The cross entropy between two distributions serves as the distillation loss: $$\begin{aligned}
\mathcal{L}_{\text{KD}} (D; \theta, \theta_{T}) = -\sum_{(x, y) \in D} \sum_{t=1}^{T_y} \sum_{k=1}^{|\mathcal{V}|} Q\{y_t=k |y_{<t}, x;\theta_{T}\}\log P(y_t=k|y_{<t}, x;\theta).
\end{aligned}$$ The difference between $\mathcal{L}_{\text{NLL}} (D; \theta)$ and $\mathcal{L}_{\text{KD}} (D; \theta, \theta_{T})$ is that the target distribution of $\mathcal{L}_{\text{KD}} (D; \theta, \theta_{T})$ is no longer the original one-hot label, but teacher’s output distribution which is more smooth by assigning non-zero probabilities to more than one word and yields smaller variance in gradients [@hinton2015distilling]. Then the total loss function becomes $$\begin{aligned}
\mathcal{L}_{\text{ALL}} (D; \theta, \theta_{T}) = (1-\lambda) \mathcal{L}_{\text{NLL}} (D; \theta) + \lambda \mathcal{L}_{\text{KD}} (D; \theta, \theta_{T}),
\end{aligned}
\label{eq_loss}$$ where $\lambda$ is the coefficient to trade off the two loss terms.
Multilingual Distillation with Multiple Teachers and One Student
----------------------------------------------------------------
Let $L$ denote the total number of language pairs in our setting, superscript $l \in [L]$ denote the index of language pair, $D^{l}$ denote the bilingual corpus for the $l$-th language pair, $\theta_M$ denote the parameters of the (student) multilingual model, and $\theta^{l}_I$ denote the parameters of the (teacher) individual model for $l$-th language pair. Therefore, $\mathcal{L}_{\text{NLL}} (D; \theta_M)$ denotes the log-likelihood loss on training data $D$, and $\mathcal{L}_{\text{ALL}} (D^l; \theta_M, \theta^{l}_{I})$ denotes the total loss on training data $D^l$, which consists of the original log-likelihood loss and the distillation loss by matching to the outputs from the teacher model $\theta^{l}_I$.
The multilingual distillation process is summarized in Algorithm \[alg\]. As can be seen in Line 1, our algorithm takes pretrained individual models for each language pair as inputs. Note that those models can be pretrained using the same datasets $\{D^{l}\}^{L}_{l=1}$ or different datasets, and they can share the same network structure as the multilingual model or use different architectures. For simplification, in our experiments, we use the same datasets to pretrain the individual models and they share the same architecture as the multilingual model. In Line 8-9, the multilingual model learns from both the ground-truth data and the individual models with loss $\mathcal{L}_{\text{ALL}}$ when its accuracy has not surpassed the individual model for a certain threshold $\tau$ (which is checked in Line 15-19 every $\mathcal{T}_{\text{check}}$ steps according to the accuracy in validation set); otherwise, the multilingual model only learns from the ground-truth data using the original log-likelihood loss $\mathcal{L}_{\text{NLL}}$ (in Line 10-11).
**Input**: Training corpus $\{D^{l}\}^{L}_{l=1}$ and pretrained individual models $\{\theta_I^{l}\}^{L}_{l=1}$ for $L$ language pairs, learning rate $\eta$, total training steps $\mathcal{T}$, distillation check step $\mathcal{T}_{\text{check}}$, threshold $\tau$ of distillation accuracy. **Initialize**: Randomly initialize multilingual model $\theta_M$. Set current training step $T$ = $0$, accumulated gradient $g$ = $\mathbf{0} $, distillation flag $f^{l}$ = $\textit{True}$ for $l \in [L]$. **while** $T < \mathcal{T}$ **do** $T$ = $T$+$1 $ $g$ = $\mathbf{0} $ **for** $l \in [L]$ **do** Randomly sample a mini-batch of sentence pairs $(\textbf{x}^{l}, \textbf{y}^{l})$ from $D^{l}$. **if** $f^{l}$ == $\textit{True} $ **do** Compute and accumulate the gradient on loss $\mathcal{L}_{\text{ALL}} ((\textbf{x}^{l}, \textbf{y}^{l}); \theta_M, \theta^{l}_{I})$: $g$ += $\partial \mathcal{L}_{\text{ALL}} / \partial \theta_M$. **else** Compute and accumulate the gradient on loss $\mathcal{L}_{\text{NLL}} ((\textbf{x}^{l}, \textbf{y}^{l}); \theta_M)$: $g$ += $\partial \mathcal{L}_{\text{NLL}} / \partial \theta_M$. **end if** **end for** Update $\theta_M$: $\theta_M$ = $\theta_M$ - $\eta * g$ **if** $T$ % $\mathcal{T}_{\text{check}}$ == $0$ **do** **for** $l \in [L]$ **do** **if** $\text{Accuracy}(\theta_M) < \text{Accuracy}(\theta^{l}_{I})$ + $\tau $ **do** $f^{l}$ = $\textit{True} $ **else** $f^{l}$ = $\textit{False}$ **end if** **end for** **end if**
**end while**
Discussion {#method_discuss}
----------
#### Selective Distillation
Considering that distillation from a bad teacher model is likely to hurt the student model and thus result in inferior accuracy, we selectively use distillation in the training process, as shown in Line 15-19 in Algorithm \[alg\]. When the accuracy of multilingual model surpasses the individual model for the accuracy threshold $\tau$ on a certain language pair, we remove the distillation loss and just train the model with original negative log-likelihood loss for this pair. Note that in one iteration, one language may not use the distillation loss; it is very likely in later iterations that this language will be distilled again since the multilingual model may become worse than the teacher model for this language. Therefore, we call this mechanism as selective distillation. We also verify the effectiveness of the selective distillation in experiment part (Section \[sec\_analysis\]).
#### Top-K Distillation
It is burdensome to load all the teacher models in the GPU memory for distillation considering there are dozens or even hundreds of language pairs in the multilingual setting. Alternatively, we first generate the output probability distribution of each teacher model for the sentence pairs offline, and then just load the top-K probabilities of the distribution into memory and normalize them so that they sum to 1 for distillation. This can reduce the memory cost again from the scale of $|V|$ (the vocabulary size) to K. We also study in Section \[sec\_analysis\] that top-K distribution can result in comparable or better distillation accuracy than the full distribution.
Experiments
===========
We test our proposed method on three public datasets: IWSLT, WMT, and Ted talk translation tasks. We first describe experimental settings, report results, and conduct some analyses on our method.
Settings
--------
#### Datasets
We use three datasets in our experiment. *IWSLT*: We collect 12 languages$\leftrightarrow$English translation pairs from IWSLT evaluation campaign[^3] from year 2014 to 2016. *WMT*: We collect 6 languages$\leftrightarrow$English translation pairs from WMT translation task[^4]. *Ted Talk*: We use the common corpus of TED talk which contains translations between multiple languages [@Ye2018WordEmbeddings]. We select 44 languages in this corpus that has sufficient data for our experiments. More descriptions about the three datasets can be found in Appendix (Section \[appendix\_dataset\]). We also list the language code according to ISO-639-1 standard[^5] for the languages used in our experiments in Appendix (Section \[appendix\_lancode\]). All the sentences are first tokenized with moses tokenizer[^6] and then segmented into subword symbols using Byte Pair Encoding (BPE) [@DBLP:conf/acl/SennrichHB16a]. We learn the BPE merge operations across all the languages and keep the output vocabulary of the teacher and student model the same, to ensure knowledge distillation.
#### Model Configurations
We use the Transformer [@DBLP:conf/nips/VaswaniSPUJGKP17] as the basic NMT model structure since it achieves state-of-the-art accuracy and becomes a popular choice for recent NMT researches. We use the same model configuration for individual models and the multilingual model. For IWSLT and Ted talk tasks, the model hidden size $d_{\text{model}}$, feed-forward hidden size $d_{\text{ff}}$, number of layer are 256, 1024 and 2, while for WMT task, the three parameters are 512, 2048 and 6 respectively considering its large scale of training data.
#### Training and Inference
For the multilingual model training, we up sample the data of each language to make all languages have the same size of data. The mini batch size is set to roughly 8192 tokens. We train the individual models with 4 NVIDIA Tesla V100 GPU cards and multilingual models with 8 of them. We follow the default parameters of Adam optimizer [@kingma2014adam] and learning rate schedule in @DBLP:conf/nips/VaswaniSPUJGKP17. For the individual models, we use 0.2 dropout, while for multilingual models, we use 0.1 dropout according to the validation performance. For knowledge distillation, we set $\mathcal{T}_{\text{check}} =3000$ steps (nearly two training epochs), the accuracy threshold $\tau = 1$ BLEU score, the distillation coefficient $\lambda=0.5$ and the number of teacher’s outputs $K=8$ according to the validation performance. During inference, we decode with beam search and set beam size to 4 and length penalty $\alpha=1.0$ for all the languages. We evaluate the translation quality by tokenized case sensitive BLEU [@DBLP:conf/acl/PapineniRWZ02] with multi-bleu.pl[^7]. Our codes are implemented based on fairseq[^8] and we will release the codes once the paper is published.
Language *Individual* *Multi-Baseline* *Multi-Distillation* $\Delta$
----------- -- -------------- -- ------------------ -- ---------------------- -- ---------- -- --
Ar$\to$En 31.19 29.24 (-1.95) 31.25 (+0.06) +2.01
Cs$\to$En 28.04 26.09 (-1.95) 27.09 (-0.95) +1.00
De$\to$En 33.07 32.74 (-0.33) 34.02 (+0.95) +1.28
He$\to$En 37.42 35.18 (-2.24) 37.33 (-0.09) +2.15
Nl$\to$En 35.94 36.54 (+0.60) 37.69 (+1.75) +1.15
Pt$\to$En 44.30 43.49 (-0.81) 44.69 (+0.39) +1.20
Ro$\to$En 36.92 36.41 (-0.51) 38.01 (+1.09) +1.60
Ru$\to$En 23.04 23.12 (+0.08) 23.76 (+0.72) +0.64
Th$\to$En 18.24 19.33 (+1.09) 19.90 (+1.66) +0.57
Tr$\to$En 22.74 22.42 (-0.32) 23.75 (+1.01) +1.33
Vi$\to$En 26.06 26.37 (+0.31) 27.04 (+0.98) +0.67
Zh$\to$En 19.44 18.82 (-0.62) 19.52 (+0.08) +0.70
: BLEU scores of 12 languages$\to$English on the IWLST dataset. The BLEU scores in $( )$ represent the difference between the multilingual model and individual models. $\Delta$ represents the improvements of our multi-distillation method over the multi-baseline.[]{data-label="iwslt_12_bleu"}
Language *Individual* *Multi-Baseline* *Multi-Distillation* $\Delta$
----------- -- -------------- -- ------------------ -- ---------------------- -- ---------- -- --
En$\to$Ar 13.67 12.73 (-0.94) 13.80 (+0.13) +1.07
En$\to$Cs 17.81 17.33 (-0.48) 18.69 (+0.88) +1.37
En$\to$De 26.13 25.16 (-0.97) 26.76 (+0.63) +1.60
En$\to$He 24.15 22.73 (-1.42) 24.42 (+0.27) +1.69
En$\to$Nl 30.88 29.51 (-1.37) 30.52 (-0.36) +1.01
En$\to$Pt 37.63 35.93 (-1.70) 37.23 (-0.40) +1.30
En$\to$Ro 27.23 25.68 (-1.55) 27.11 (-0.12) +1.42
En$\to$Ru 17.40 16.26 (-1.14) 17.42 (+0.02) +1.16
En$\to$Th 26.45 27.18 (+0.73) 27.62 (+1.17) +0.45
En$\to$Tr 12.47 11.63 (-0.84) 12.84 (+0.37) +1.21
En$\to$Vi 27.88 28.04 (+0.16) 28.69 (+0.81) +0.65
En$\to$Zh 10.95 10.12 (-0.83) 10.41 (-0.54) +0.29
: BLEU scores of English$\to$12 languages on the IWLST dataset. The BLEU scores in $( )$ represent the difference between the multilingual model and individual models. $\Delta$ represents the improvements of our multi-distillation method over the multi-baseline.[]{data-label="iwslt_12_bleu_enxx"}
Results
-------
#### Results on IWSLT
Multilingual NMT usually consists of three settings: many-to-one, one-to-many and many-to-many. As many-many translation can be bridged though many-to-one and one-to-many setting, we just conduct the experiments on many-to-one and one-to-many settings. We first show the results of 12 languages$\to$English translations on the IWLST dataset are shown in Table \[iwslt\_12\_bleu\]. There are 3 methods for comparison: 1) *Individual*, each language pair with a separate model; 2) *Multi-Baseline*, the baseline multilingual model, simply training all the language pairs in one model; 3) *Multi-Distillation*, our multilingual model with knowledge distillation. We have several observations. First, the multilingual baseline performs worse than individual models on most languages. The only exception is the languages with small training data, which benefit from data augmentation in multilingual training. Second, our method outperforms the multilingual baseline for all the languages, demonstrating the effectiveness of our framework for multilingual NMT. More importantly, compared with the individual models, our method achieves similar or even better accuracy (better on 10 out of 12 languages), with only $1/12$ model parameters of the sum of all individual models.
One-to-many setting is usually considered as more difficult than many-to-one setting, as it contains different target languages which is hard to handle. Here we show how our method performs in one-to-many setting in Table \[iwslt\_12\_bleu\_enxx\]. It can be seen that our method can maintain the accuracy (even better on most languages) compared with the individual models. We still improve over the multilingual baseline by nearly 1 BLEU score, which demonstrates the effectiveness of our method.
Language *Individual* *Multi-Baseline* *Multi-Distillation* $\Delta$
---------- -- -------------- -- ------------------ -- ---------------------- -- ----------
Cs-En 25.29 23.82 (-1.47) 25.37 (+0.08) +1.55
De-En 34.44 34.21 (-0.23) 36.22 (+1.78) +2.01
Fi-En 21.23 22.99 (+1.76) 24.32 (+3.09) +1.33
Lv-En 16.26 16.25 (-0.01) 18.43 (+2.17) +2.18
Ro-En 35.81 35.04 (-0.77) 36.51 (+0.70) +1.47
Ru-En 29.39 28.92 (-0.47) 30.82 (+1.43) +1.90
: BLEU scores of 6 languages$\to$English on the WMT dataset. The BLEU scores in $( )$ represent the difference between the multilingual model and individual models. $\Delta$ represents the improvements of our multi-distillation method over the multi-baseline.[]{data-label="wmt_6_bleu"}
Language *Individual* *Multi-Baseline* *Multi-Distillation* $\Delta$
---------- -- -------------- -- ------------------ -- ---------------------- -- ----------
En-Cs 22.58 21.39 (-1.19) 23.10 (+0.62) +1.81
En-De 31.40 30.08 (-1.32) 31.42 (+0.02) +1.34
En-Fi 22.08 19.52 (-2.56) 21.56 (-0.52) +2.04
En-Lv 14.92 14.51 (-0.41) 15.32 (+0.40) +0.81
En-Ro 31.67 29.88 (-1.79) 31.39 (-0.28) +1.51
En-Ru 24.36 22.96 (-1.40) 24.02 (-0.34) +1.06
: BLEU scores of English$\to$ 6 languages on the WMT dataset.[]{data-label="wmt_6_bleu_enxx"}
#### Results on WMT
The results of 6 languages$\to$English translations on the WMT dataset are reported in Table \[wmt\_6\_bleu\]. It can be seen that the multi-baseline model performs worse than the individual models on 5 out of 6 languages, while in contrast, our method performs better on all the 6 languages. Particularly, our method improves the accuracy of some languages with more than 2 BLEU scores over individual models. The results of one-to-many setting on WMT dataset are reported in Table \[wmt\_6\_bleu\_enxx\]. It can be seen that our method outperforms the multilingual baseline by more than 1 BLEU score on nearly all the languages.
Language Ar Bg Cs Da De El Es Et Fa Fi Frca
-------------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- -------
$\Delta_{1}$ -1.50 -9.46 1.88 4.02 -0.10 0.80 0.23 8.20 0.09 6.44 15.8
$\Delta_{2}$ 1.73 1.42 1.13 1.82 1.68 1.45 1.63 0.77 1.83 1.10 1.24
Language Fr Gl He Hi Hr Hu Hy Id It Ja Ka
$\Delta_{1}$ 0.13 19.26 -1.59 10.16 1.46 -0.11 8.87 1.36 -0.56 -0.03 11.20
$\Delta_{2}$ 1.48 1.58 2.26 1.07 1.21 1.80 0.92 1.48 1.48 0.95 1.55
Language Ko Ku Lt Mk My Nb Nl Pl Ptbr Pt Ro
$\Delta_{1}$ -0.42 7.75 4.46 10.72 7.63 14.07 -0.20 1.32 0.13 8.76 0.66
$\Delta_{2}$ 1.43 1.55 1.69 0.80 1.31 1.47 1.68 0.80 1.45 1.98 1.70
Language Ru Sk Sl Sq Sr Sv Th Tr Uk Vi Zh
$\Delta_{1}$ 0.65 4.23 11.87 5.03 1.58 2.39 1.17 -0.79 2.04 0.15 6.83
$\Delta_{2}$ 0.99 0.93 1.15 1.68 1.44 1.00 0.62 1.88 0.98 0.77 0.58
: BLEU scores improvements of our method over the individual models ($\Delta_{1}$) and multi-baseline model ($\Delta_{2}$) on the 44 languages$\to$English in the Ted talk dataset. []{data-label="ted_xx_bleu_table"}
#### Results on Ted Talk
Now we study the effectiveness of our method on a large number of languages. The experiments are conducted on the 44 languages$\to$English on the Ted talk dataset. Due to the large number of languages and space limitations, we just show the BLEU score improvements of our method over individual models and the multi-baseline for each language in Table \[ted\_xx\_bleu\_table\], and leave the detailed experiment results to Appendix (Section \[appendix\_ted\]). It can be seen that our method can improve over the multi-baseline for all the languages, mostly with more than 1 BLEU score improvements. Our method can also match or even surpass individual models for most languages, not to mention that the number of parameters of our method is only $1/44$ of that of the sum of 44 individual models. Our method achieves larger improvements on some languages, such as Da, Et, Fi, Hi and Hy, than others. We find this is correlated with the data size of the languages, which are listed in Appendix (Table \[append\_ted\_data\]). When a language is of smaller data size, it may get more improvement due to the benefit of multilingual training.
Analysis {#sec_analysis}
--------
In this section, we conduct thorough analyses on our proposed method for multilingual NMT.
#### Selective Distillation
We study the effectiveness of the selective distillation (discussed in Section \[method\_discuss\]) on the Ted talk dataset, as shown in Table \[ted\_early\_stop\]. We list the 16 languages on which the two methods (selective distillation, and distillation all the time) that have difference bigger than 0.5 in terms of BLEU score. It can be seen that selective distillation performs better on 13 out of 16 languages, with large BLEU score improvements, which demonstrates the effectiveness of the selective distillation.
Bg Et Fi Fr Gl Hi Hy Ka
--------------------------- -- ------- ------- ------- ------- ------- ------- ------- ------- -- -- -- -- --
distillation all the time 28.07 12.64 15.13 33.69 30.28 18.86 19.88 14.04
selective distillation 29.18 15.63 17.23 34.32 31.90 21.00 21.17 18.27
$\Delta$ +1.11 +2.99 +2.10 +0.63 +1.62 +2.14 +1.29 +4.23
Ku Mk My Sl Zh Pl Sk Sv
distillation all the time 8.50 32.10 14.02 22.10 17.22 25.05 30.45 37.88
selective distillation 13.38 32.65 15.17 23.68 19.39 24.30 29.91 36.92
$\Delta$ +4.88 +0.55 +1.15 +1.58 +2.17 -0.75 -0.54 -0.96
: BLEU scores of selective distillation (our method) and distillation all the time during the training process on the Ted talk dataset.[]{data-label="ted_early_stop"}
#### Top-K Distillation
In our experiments, the student model just matches the top-K output distribution of the teacher model, instead of the full distribution, in order to reduce the memory cost. We analyze whether there is accuracy difference between the top-K distribution and the full distribution. We conduct experiments on IWSLT dataset with varying $K$ (from 1 to $|V|$, where $|V|$ is the vocabulary size), and just show the BLEU scores on the validation set of De-En translation due to space limitation, as illustrated in Table \[table\_topk\]. It can be seen that increasing $K$ from 1 to 8 will improve the accuracy, while bigger $K$ will bring no gains, even with the full distribution ($K=|V|$).
Top-K 1 2 4 8 16 32 64 128 $|V|$
------- ------- ------- ------- ------- ------- ------- ------- ------- ------- -- -- -- -- --
BLEU 33.45 33.86 34.47 34.76 34.66 34.68 34.54 34.47 34.49
: BLEU scores on De-En translation with varying Top-K distillation on the IWSLT dataset.[]{data-label="table_topk"}
Language Ar Cs De Nl Ro Ru Th Tr Vi
-------------------- ------- ------- ------- ------- ------- ------- ------- ------- ------- --
Individual 31.19 28.04 33.07 35.94 36.92 23.04 18.24 22.74 26.06
+Back Distillation 31.39 29.44 33.71 36.86 37.28 23.36 19.42 23.58 27.17
$\Delta$ +0.20 +1.40 +0.64 +0.92 +0.36 +0.32 +1.18 +0.84 +1.11
: BLEU score improvements of the individual models with back distillation on the IWSLT dataset. []{data-label="back_distill"}
#### Back Distillation
In our current distillation algorithm, we fix the individual models and use them to teach and improve the multilingual model. After such a distillation process, the multilingual model outperforms the individual models on most of the languages. Then naturally, we may wonder whether this improved multilingual model can further be used to teach and improve individual models through knowledge distillation. We call such a process back distillation. We conduct the experiments on the IWSLT dataset, and find that the accuracy of 9 out of 12 languages gets improved, as shown in Table \[back\_distill\]. The other 3 languages (He, Pt, Zh) cannot get improvements because the improved multilingual model performs very close to individual models, as shown in Table \[iwslt\_12\_bleu\].
#### Comparison with Sequence-Level Knowledge Distillation
We conduct experiments to compare the word-level knowledge distillation (the exact method used in our paper) with sequence-level knowledge distillation[@kim2016sequence] on IWSLT dataset. As shown in Table \[seq\_vs\_word\], sequence-level knowledge distillation results in consistently inferior accuracy on all languages compared with word-level knowledge distillation used in our work.
Language Sequence-level Word-level (Our Method) $\Delta$
---------- ---------------- ------------------------- ---------- --
En-Ar 12.79 13.80 1.01
En-Cs 17.01 18.69 1.68
En-De 25.89 26.76 0.87
En-He 22.92 24.42 1.50
En-Nl 29.99 30.52 0.53
En-Pt 36.12 37.23 1.10
En-Ro 25.75 27.11 1.36
En-Ru 16.38 17.42 1.04
En-Th 27.52 27.62 0.10
En-Tr 11.11 12.84 1.73
En-Vi 28.08 28.69 0.61
En-Zh 10.25 10.41 0.16
: BLEU scores of sequence-level knowledge distillation and word-level knowledge distillation on the IWSLT dataset.[]{data-label="seq_vs_word"}
#### Generalization Analysis
Previous works [@yang2018knowledge; @lan2018knowledge] have shown that knowledge distillation can help a model generalize well to unseen data, and thus yield better performance. We analyze how distillation in multilingual setting helps the model generalization. Previous studies [@keskar2016large; @chaudhari2016entropy] demonstrate the relationship between model generalization and the width of local minima in loss surface. Wider local minima can make the model more robust to small perturbations in testing. Therefore, we compare the generalization capability of the two multilingual models (our method and the baseline) by perturbing their parameters.
Specifically, we perturb a model $\theta$ as $\theta_i(\sigma) = \theta_i + \Bar{\theta}* \mathcal{N}(0, \sigma^2)$, where $\theta_i$ is the $i$-th parameter of the model, $\Bar{\theta}$ is the average of all the parameters in $\theta$. We sample from the normal distribution $\mathcal{N}$ with standard variance $\sigma$ and larger $\sigma$ represents bigger perturbation on the parameter. We conduct the analyses on the IWSLT dataset and vary $\sigma \in [0.05,0.1,0.15,0.2,0.25,0.3]$. Figure \[fig\_generalization\_a\] shows the loss curve in the test set with varying $\sigma$. As can be seen, while both the two losses increase with the increase of $\sigma$, the loss of the baseline model increases quicker than our method. We also show three test BLEU curves on three translation pairs (Figure \[fig\_generalization\_b\]: Ar-En, Figure \[fig\_generalization\_c\]: Cs-En, Figure \[fig\_generalization\_d\]: De-En, which are randomly picked from the 12 languages pairs on the IWSLT dataset). We observe that the BLEU score of the multilingual baseline drops quicker than our method, which demonstrates that our method helps the model find wider local minima and thus generalize better.
[0.22]{} ![The loss (Figure a) and BLEU score (Figure b: Ar-En, Figure c: Cs-En, Figure d: De-En) changes on the test set of the IWSLT dataset, with varying perturbation parameter $\sigma$.[]{data-label="fig_generalization_analysis"}](fig/perturb.png "fig:"){width="\textwidth"}
[0.23]{} ![The loss (Figure a) and BLEU score (Figure b: Ar-En, Figure c: Cs-En, Figure d: De-En) changes on the test set of the IWSLT dataset, with varying perturbation parameter $\sigma$.[]{data-label="fig_generalization_analysis"}](fig/perturb_bleu_ar.png "fig:"){width="\textwidth"}
[0.23]{} ![The loss (Figure a) and BLEU score (Figure b: Ar-En, Figure c: Cs-En, Figure d: De-En) changes on the test set of the IWSLT dataset, with varying perturbation parameter $\sigma$.[]{data-label="fig_generalization_analysis"}](fig/perturb_bleu_cs.png "fig:"){width="\textwidth"}
[0.23]{} ![The loss (Figure a) and BLEU score (Figure b: Ar-En, Figure c: Cs-En, Figure d: De-En) changes on the test set of the IWSLT dataset, with varying perturbation parameter $\sigma$.[]{data-label="fig_generalization_analysis"}](fig/perturb_bleu_de.png "fig:"){width="\textwidth"}
Conclusion
==========
In this work, we have proposed a distillation-based approach to boost the accuracy of multilingual NMT, which is usually of lower accuracy than the individual models in previous works. Experiments on three translation datasets with up to 44 languages demonstrate the multilingual model based on our proposed method can nearly match or even outperform the individual models, with just $1/N$ model parameters (N is up to 44 in our experiments).
In the future, we will conduct more deep analyses about how distillation helps the multilingual model training. We will apply our method to larger datasets and more languages pairs (hundreds or even thousands), to study the upper limit of our proposed method.
Appendix {#appendix .unnumbered}
========
Dataset Description {#appendix_dataset}
===================
We give a detailed description about the *IWSLT*,*WMT* and *Ted Talk* datasets used in experiments.
*IWSLT*: We collect 12 languages$\leftrightarrow$English translation pairs from IWSLT evaluation campaign[^9] from year 2014 to 2016. Each language pair contains roughly 80K to 200K sentence pairs. We use the official validation and test sets for each language pair. The data sizes of the training set for each language$\leftrightarrow$English pair are listed in Table \[append\_iwslt\_data\].
Language Ar Cs De He Nl Pt
--------------- ------ ------ ------ ------ ------ ------
Training Data 174K 114K 167K 180K 174K 167K
Language Ro Ru Th Tr Vi Zh
Training Data 177K 173K 83K 150K 131K 209K
: The training data size on the 12 languages$\leftrightarrow$ English on the IWSLT dataset.[]{data-label="append_iwslt_data"}
*WMT*: We collect 6 languages$\leftrightarrow$English translation pairs from WMT translation task[^10]. We use 5 language$\leftrightarrow$English translation pairs from WMT 2016 dataset: Cs-En, De-En, Fi-En, Ro-En, Ru-En and one other translation pair from WMT 2017 dataset: Lv-En. We use the official released validation and test sets for each language pair. The training data sizes of each language$\leftrightarrow$English pair are shown in the Table \[append\_wmt\_data\].
Language Cs De Fi Lv Ro Ru
--------------- ------ ------ ------ ------ ------ ------
Training Data 1.0M 4.5M 2.5M 4.5M 2.2M 2.1M
: The training data size on the 6 languages$\leftrightarrow$English on the WMT dataset.[]{data-label="append_wmt_data"}
*Ted Talk*: We use the common corpus of TED talk which contains translations between multiple languages [@Ye2018WordEmbeddings][^11]. We select 44 languages in this corpus that has sufficient data for our experiments. We use the official validation and test sets for each language pair. The data sizes of the training set for each language$\leftrightarrow$English pair are listed in Table \[append\_ted\_data\].
Language Ar Bg Cs Da De El Es Et Fa Fi Frca
--------------- ------ ------ ------ ----- ------ ------ ------ ------ ------ ------ ------
Training Data 214K 174k 103k 45k 168k 134k 196k 11k 151k 24k 20k
Language Fr Gl He Hi Hr Hu Hy Id It Ja Ka
Training Data 192K 10K 212K 19K 122K 147K 21K 87K 205K 204K 13K
Language Ko Ku Lt Mk My Nb Nl Pl Ptbr Pt Ro
Training Data 206K 10K 42K 25K 21K 16K 184K 176K 185K 52K 180K
Language Ru Sk Sl Sq Sr Sv Th Tr Uk Vi Zh
Training Data 208K 61K 20K 45K 137K 57K 98K 182K 108K 172K 200K
: The training data size on the 44 languages$\leftrightarrow$ English on the Ted talk dataset.[]{data-label="append_ted_data"}
Language Name and Code {#appendix_lancode}
======================
The language names and their corresponding language codes according to ISO 639-1 standard[^12] are listed in Table \[language\_code\].
Language Code Language Code Language Code Language Code
----------- ------ ------------ ------ ------------ ------ ------------ ------
Arabic Ar Bulgarian Bg Czech Cs Danish Da
German De Greek El English En Spanish Es
Persian Fa Finnish Fi French Fr Galician Gl
Hebrew He Hindi Hi Croatian Hr Hungarian Hu
Armenian Hy Indonesian Id Italian It Japanese Ja
Georgian Ka Korean Ko Kurdish Ku Lithuanian Lt
Latvian Lv Macedonian Mk Burmese My Norwegian Nb
Dutch Nl Polish Pl Portuguese Pt Romanian Ro
Russian Ru Slovak Sk Slovenian Sl Albanian Sq
Serbian Sr Swedish Sv Thai Th Turkish Tr
Ukrainian Uk Vietnamese Vi Chinese Zh
: The ISO 639-1 code of each language in our experiments. There are two extra language codes in our datasets: Ptbr represents Portuguese spoken in Brazil, Frca represents French spoken in Canada.[]{data-label="language_code"}
Results on Ted Talk Dataset {#appendix_ted}
===========================
The detailed results of the 44 languages$\to$English on the Ted talk dataset are listed in Table \[append\_ted\_xx\_bleu\]. It can be seen that while multilingual baseline performs worse than the individual model, multilingual model based on our method nearly matches and even outperforms the individual model. Note that the multilingual model handles 44 languages in total, which means our method can reduce the model parameters size to $1/44$ without loss of accuracy.
Language Ar Bg Cs Da De El Es Et Fa
----------------------------- ------- ------- ------- ------- ------- ------- ------- ------- -------
*Individual* 31.07 38.64 26.42 38.21 34.63 36.69 41.20 7.43 26.67
*Multilingual (Baseline)* 27.84 27.76 27.17 40.41 32.85 36.04 39.80 14.86 24.93
*Multilingual (Our method)* 29.57 29.18 28.30 42.23 34.53 37.49 41.43 15.63 26.76
Language Fi Frca Fr Gl He Hi Hr Hu Hy
*Individual* 10.78 18.52 39.62 12.64 36.81 10.84 34.14 24.67 12.30
*Multilingual (Baseline)* 16.12 33.08 38.27 30.32 32.96 19.93 34.39 22.76 20.25
*Multilingual (Our method)* 17.22 34.32 39.75 31.9 35.22 21.00 35.6 24.56 21.17
Language Id It Ja Ka Ko Ku Lt Mk My
*Individual* 29.20 38.06 13.31 7.06 18.54 5.63 18.19 21.93 7.53
*Multilingual (Baseline)* 29.08 36.02 12.33 16.71 16.71 11.83 20.96 31.85 13.85
*Multilingual (Our method)* 30.56 37.50 13.28 18.26 18.14 13.38 22.65 32.65 15.16
Language Nb Nl Pl Ptbr Pt Ro Ru Sk Sl
*Individual* 27.28 35.85 22.98 44.28 33.81 34.07 24.36 25.67 11.80
*Multilingual (Baseline)* 39.88 33.97 23.50 42.96 40.59 33.03 24.02 28.97 22.52
*Multilingual (Our method)* 41.35 35.65 24.30 44.41 42.57 34.73 25.01 29.90 23.67
Language Sq Sr Sv Th Tr Uk Vi Zh
*Individual* 29.70 32.13 34.53 20.95 24.46 25.76 26.38 12.56
*Multilingual (Baseline)* 33.05 32.27 35.92 21.50 21.79 26.82 25.76 18.81
*Multilingual (Our method)* 34.73 33.71 36.92 22.12 23.67 27.80 26.53 19.39
: BLEU scores of the individual and multilingual models on the 44 languages$\to$English on the Ted talk dataset.[]{data-label="append_ted_xx_bleu"}
[^1]: Authors contribute equally to this work.
[^2]: https://www.ethnologue.com/browse
[^3]: https://wit3.fbk.eu/
[^4]: http://www.statmt.org/wmt16/translation-task.html, http://www.statmt.org/wmt17/translation-task.html
[^5]: https://www.loc.gov/standards/iso639-2/php/code\_list.php
[^6]: https://github.com/moses-smt/mosesdecoder/blob/master/scripts/tokenizer/tokenizer.perl
[^7]: https://github.com/moses-smt/mosesdecoder/blob/master/scripts/generic/multi-bleu.perl
[^8]: https://github.com/pytorch/fairseq
[^9]: https://wit3.fbk.eu/
[^10]: http://www.statmt.org/wmt16/translation-task.html, http://www.statmt.org/wmt17/translation-task.html
[^11]: https://github.com/neulab/word-embeddings-for-nmt
[^12]: https://www.loc.gov/standards/iso639-2/php/code\_list.php
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present - an open-source code for rapid spectral modelling of supernovae (SNe). Our goal is to develop a tool that is sufficiently fast to allow exploration of the complex parameter spaces of models for SN ejecta. This can be used to analyse the growing number of high-quality SN spectra being obtained by transient surveys. The code uses *Monte Carlo* methods to obtain a self-consistent description of the plasma state and to compute a synthetic spectrum. It has a modular design to facilitate the implementation of a range of physical approximations that can be compared to asses both accuracy and computational expediency. This will allow users to choose a level of sophistication appropriate for their application. Here, we describe the operation of the code and make comparisons with alternative radiative transfer codes of differing levels of complexity (<span style="font-variant:small-caps;">syn++</span>, <span style="font-variant:small-caps;">python</span>, and <span style="font-variant:small-caps;">artis</span>). We then explore the consequence of adopting simple prescriptions for the calculation of atomic excitation, focussing on four species of relevance to Type Ia supernova spectra – , , , and . We also investigate the influence of three methods for treating line interactions on our synthetic spectra and the need for accurate radiative rate estimates in our scheme.'
author:
- |
Wolfgang E. Kerzendorf$^1$, Stuart A. Sim$^2$\
$^1$Department of Astronomy and Astrophysics, University of Toronto, 50 Saint George Street, Toronto, ON M5S 3H4, Canada\
$^2$Astrophysics Research Centre, School of Mathematics and Physics, Queen’s University Belfast, Belfast BT7 1NN, UK
bibliography:
- 'tardis.bib'
nocite: '[@2002ApJ...579..725L]'
title: A spectral synthesis code for rapid modelling of supernovae
---
radiative transfer – methods: numerical – supernovae: general
Introduction {#sec:intro}
============
The goal of modelling is typically to infer the composition, mass and explosion energy of an event with the aim of understanding the explosion mechanisms and progenitor systems. Studies often rely on radiative transfer / spectrum synthesis codes to interpret the complex spectra originating in rapidly expanding and often metal-rich ejecta. A variety of different codes exist and are used for different purposes: these range from very simple approaches designed purely for line identification \[run times of second e.g. \] via computationally inexpensive one-dimensional codes \[runtimes of minutes, e.g. the \] to vastly more complex codes that implement sophisticated micro-physics, time-dependence and/or multi-dimensionality . Such sophisticated approaches, however, come at the expense of execution times running to many hours (even days) on massively parallel machines, making them impractical for efficient exploration of parameter space or rapid characterization of new data sets. Thus, for the foreseeable future, the need for the simpler approaches will remain. Nevertheless, given the quality of modern data sets, it is important to test and understand the implications of any simplifying assumptions that are made.
Our goal is to provide analysis tools that strike an optimal balance between accuracy and computational expense, ideally in a manner that can be adjusted to suit the demands of (and resources available to) a variety of studies. To this end, we have undertaken the development of a modular radiative transfer code (). As input, the code takes a one-dimensional model for the supernova eject (which can include arbitrary density and/or abundance stratification, if required) together with a luminosity and time since explosion. From these, ionization and excitation states are iteratively estimated and a synthetic spectrum is calculated. The code serves two main purposes. First, it is designed to have an execution time of minutes, which allows for rapid exploration of parameter space with the possibility of fitting and interpretation of observational data. Secondly, it serves as a platform to trial different approximations for important microphysics (e.g. line interaction, level populations, etc.). allows for swift implementation and testing of new physics that – if successful – might be used in the more complete but complex codes like and , which are founded on the same principles as . Even if the newly implemented physics is computationally too costly to use in large parameter-space studies, it can be still be used to quantify the uncertainty associated with using faster, but less accurate, approaches (i.e. Uncertainty Quantification).
In Section \[sec:method\], we present the numerical methods and a sub-set of the modes of operation currently implemented in the code. Sources of atomic data are described in Section \[sec:datasources\]. Sections \[sec:numerics\] and \[sec:code\_comparisons\] present results of numerical convergence tests and detailed comparisons of our synthetic spectra with those from a selection of existing radiative transfer codes of varying degrees of complexity.
In this paper we explore the effect of approximations used by two different physics sub-modules, focussing on their importance in the modelling of .
Studies of SNe Ia using 1D codes that combine realistic treatments of ionization and excitation have already shown that the effects of departures from can be rather important. Consequently, for our goal of developing a code that is both reasonably fast and accurate, we must attempt to identify suitable approaches and quantify the impact of approximations made in the interests of computational expediency. Thus, we first (in Section \[sec:results\_nlte\]) examine the influence of adopting a simple analytic treatment for the populations of excited atomic/ionic levels by comparing to results using level populations obtained by numerically solving a set of equations of statistical equilibrium.
Secondly, in Section \[sec:results\_line\_interaction\], we compare results obtained using different levels of sophistication in the treatment of line scattering and fluorescence. It has been clearly established that non-resonance line scattering has a key role in shaping the spectra of metal-rich SN ejecta . In this work, we will compare two versions of the “macro atom” scheme to a resonance scattering scheme and quantify their influence on synthetic spectra.
In Section \[sec:conclusions\], we summarise our findings and outline future plans for the further development of this project.
Method {#sec:method}
======
The code is a new implementation of indivisible energy-packet methods (@1985ApJ...288..679A [@1993ApJ...405..738L], ML93, ) that have previously been used to model radiation transport in Type Ia supernovae (SNe Ia; e.g. ML93; @2006ApJ...651..366K [@2007MNRAS.375..154S; @2009MNRAS.398.1809K]). The algorithm operates by seeding a set of quanta representing photon bundles and following their propagation through a model for the ejecta, as described below.
is based on the same methods used by , but the scope of is much more limited so that runtimes are orders of magnitude shorter. In particular, having in mind the goal of rapid fitting of observations (i.e. using to compute synthetic spectra that can be fed into a fitting algorithm and explore model parameter space), currently neglects multi-dimensionality and time-dependence. Thus, a calculation provides a single snapshot spectrum while an run provides a time series of spectra from which synthetic light curves can be constructed.[^1] In addition, adopts an inner boundary approximation, which greatly simplifies the calculations by avoiding the need to describe the diffusion of radiation at very high optical depths.
already includes a number of different options and modules that make it possible to compare the effects of particular assumptions on both the outputs and the computational expense. The current options will each be introduced in the sections below but, for reference, we provide a summary in Table \[tab:modes\]. So far, the development of has focused on applications. However, we stress that our goal is to continue development and that the code has been structured to make it easy to implement additional physics and improve, alter or lift many of the existing approximations (see below). Further modules extending the code for application to are being developed (Klauser, Kromer et al., in prep.).
is written in Python with C extensions and is available as open source (BSD 3 clause license) at <http://pypi.python.org/pypi/tardis-sn>. For all random numbers uses the Mersenne Twister random number generator [@Matsumoto:1998:MTE:272991.272995]. In Appendix A we show an example input file and provide basic operational details for using . However, we refer potential users to the technical manual (available at <http://tardis.rtfd.org>) for further information on running the code.
[ll]{} Mode name & Mode of operation\
\
`lte`& Ionization fractions from the Saha-Boltzmann equation and ${\ensuremath{T_\textrm{R}}\xspace}$ (Equation \[eq:lte\_ionization\])\
`nebular` & Ionization fractions from the nebular approximation (Equation \[eq:modified\_nebular\])\
\
\
`lte` & Level populations from Boltzmann equation and ${\ensuremath{T_\textrm{R}}\xspace}$ (Equation \[eq:lte\_excitation\])\
`dilute-lte` & Excited level populations reduced by $W$ (Equation \[eq:dilute\_lte\_excitation\])\
`nlte` & Population ratios determined from statistical equilibrium (Section \[sec:plasma\_state\])\
\
\
`dilute-blackbody` & Radiative rates obtained using $J_{lu}^b = W B_{\nu_{lu}}({\ensuremath{T_\textrm{R}}\xspace})$\
`detailed` & Radiative rates obtained using estimators for $J_{lu}^b$ (Section \[sec:montecarlo\_estimators\])\
\
\
`scatter` & All line interactions treated as resonance scattering events\
`downbranch` & Macro atom scheme with all internal transition probabilities set to zero (Section \[sec:rad\_matter\_interaction\])\
`macroatom` & Full macro atom scheme (Section \[sec:rad\_matter\_interaction\])\
\[tab:modes\]
Model Setup
-----------
In the interest of computational expedience, approximates the ejecta as spherically symmetric such that all physical properties depend only on the radial coordinate $r$. The computational domain is defined by inner and outer radial boundaries. The code currently assumes that material inside the domain is in radiative equilibrium and neglects any non-radiative energy sources (i.e. it is assumed that energy injection by within the computation volume is negligible). This approximation limits the applicability of to epochs at which the effective photosphere is external to the volume in which the majority of the luminosity is generated (for , this is most valid at early times and becomes an increasingly poor approximation at later epochs).
The computational domain is discretised into multiple cells (spherical shells). For each of these, the density and elemental abundances must be specified (the “input model”). The input model is setup at runtime and can involve any density profile and set of stratified abundances. Simple models with uniform abundances and standard density profiles are directly created by the code (see Appendix \[sec:using\_tardis\]) while models with stratified abundances or arbitrary density profiles are setup by providing simple input files, as described in the manual. During the simulation, various quantities are computed on a cell-by-cell basis, most importantly the parameters of a simple radiation-field model (see Section \[sec:rad\_model\]), which are used to estimate ionization fractions and level populations (see Section \[sec:plasma\_state\]).
We assume that the ejecta are in homologous expansion, which becomes an adequate approximation for within $\sim~100$ seconds after explosion .
Radiation Field Model {#sec:rad_model}
---------------------
In general, we assume that both ionization and excitation are primarily controlled by the radiation field and we follow @1985ApJ...288..679A, @1993ApJ...405..738L and ML93 in adopting approximations based on a simple model for the frequency ($\nu$) dependent mean intensity
$$J_{\nu} = W B_{\nu} ({\ensuremath{T_\textrm{R}}\xspace})$$
where, $B_{\nu}$ is the Planck function. The radiation temperature (${\ensuremath{T_\textrm{R}}\xspace}$) and dilution factor ($W$) are parameters that are iteratively derived from estimators in each grid cell (see Section \[sec:montecarlo\_estimators\]).
Plasma State {#sec:plasma_state}
------------
To compute opacities and handle radiation-matter interactions, it is necessary to specify the number densities for the states of the atomic/ionic species included. In principal, these could be obtained by a full solution of the complete system of equations of statistical equilibrium with rate coefficients for radiative processes derived from properties of the radiation field . This will be implemented in future versions of . However, since the computational cost of such an approach will be significant, it is also valuable to consider simpler approximations that can be used for rapid calculations to explore parameter space. We pursue the approximate approach here, closely following ML93 who have demonstrated the utility of such a philosophy .
Currently, determines the ionization balance for each of the model cells based on the density, elemental abundance, radiation temperature (${\ensuremath{T_\textrm{R}}\xspace}$), electron temperature (${\ensuremath{T_\textrm{e}}\xspace}$) and the dilution factor ($W$). In the calculations presented in this work, we will adopt either a standard Saha-Boltzmann equation (`lte` ionization mode), $$\frac{N_{i, j+1} n_e}{N_{i, j}} = \Phi_{i,j,{\ensuremath{T_\textrm{R}}\xspace}} \equiv \frac{2 Z_{i,j+1}({\ensuremath{T_\textrm{R}}\xspace})}{Z_{i,{j}}({\ensuremath{T_\textrm{R}}\xspace})} \left(\frac{2\pi m_e k {\ensuremath{T_\textrm{R}}\xspace}}{h^2} \right) ^{3/2} e^{-\frac{\chi_{i,j}}{k\,{\ensuremath{T_\textrm{R}}\xspace}}}
\label{eq:lte_ionization}$$ or a modified nebular approximation (ML93), referred to as `nebular` ionization mode $$\frac{N_{i, j+1} n_e}{N_{i, j}} = W [\delta\zeta_{i,j} + W(1-\zeta_{i,j})] \left( \frac{{\ensuremath{T_\textrm{e}}\xspace}}{{\ensuremath{T_\textrm{R}}\xspace}} \right)^{1/2} \Phi_{i,j,{\ensuremath{T_\textrm{R}}\xspace}}
\label{eq:modified_nebular}$$ Here, $n_e$ is the free electron number density, $N_{i,j}$ is the ion number density, $Z_{i,j}$ the partition function and $\chi_{i,j}$ the ionization potential for ion $j$ of element $i$. $\zeta_{i,j}$ is the fraction of recombinations that go directly to the ground state in recombination to ion $j$ of element $i$. $\delta$ is a correction factor introduced to approximately account for the dominance of locally created radiation at short wavelengths (see ). We follow ML93 in setting $T_{e} = 0.9 T_{R}$.
With ionization ratios determined using one of these approximations, the code solves for a complete set of ion number densities (and the electron density) by enforcing the appropriate total elemental number densities and insisting on charge conservation in the usual manner.
can use a variety of approximations for calculating the level populations within each ion. As a simplest case, we adopt the Boltzmann excitation formula, $$n_{i, j, k} = \frac{g_{i,j,k}}{Z_{i, j}} N_{i, j} \exp (-\epsilon_{i,j,k} / k {\ensuremath{T_\textrm{R}}\xspace}),
\label{eq:lte_excitation}$$ where $n_{i, j, k}$ is the number density of level $k$ of ionization state $j$ of element $i$, which has excitation energy $\epsilon_{i,j,k}$ relative to the ion ground state. We will refer to this as `lte` excitation mode.
As a slightly more sophisticated approach, includes a crude NLTE approximation for level populations in which we continue to apply Equation \[eq:lte\_excitation\] to all metastable levels but adopt
$$n_{i, j, k} = W \frac{g_{i,j,k}}{Z_{i, j}} N_{i, j} \exp{(-\epsilon_{i,j,k}/k {\ensuremath{T_\textrm{R}}\xspace})}
\label{eq:dilute_lte_excitation}$$
for all other excited state . We refer to this approach as the `dilute-lte` excitation mode in .
Finally, is also able to obtain NLTE level populations by solving a set of statistical equilibrium equations for chosen species (`nlte` excitation mode). Here we formulate rates between pairs of levels (upper level $u$ and lower level $l$)[^2] in an ion as
$$\begin{aligned}
{\cal R}_{ul} =& \beta_{lu} A_{ul} n_l + \beta_{lu} B_{ul} n_l J_{lu}^b + C_{ul} n_u n_e\label{eq:nlte_rul}\\
{\cal R}_{lu} =& \beta_{lu} B_{lu} n_l J_{lu}^b + C_{lu} n_l n_e\label{eq:nlte_rlu},\end{aligned}$$
where $A, B$ are the usual Einstein coefficients for radiative transitions, $C$ is the rate coefficient for electron collisions and $\beta_{lu}=\frac{1}{\tau_{lu}}[1-\exp(-\tau_{lu})]$ is the Sobolev escape probability . $J_{lu}^b$ is the mean intensity at the extreme violet wing of the bound-bound transition between levels $l$ and $u$ . can currently estimate [$J_{lu}^b$]{}in one of two ways. The first option (`dilute-blackbody`) sets ${\ensuremath{J_{lu}^b}\xspace}= W B_{\nu_{lu}}({\ensuremath{T_\textrm{R}}\xspace})$ (i.e. imposes the radiation-field model of Section \[sec:rad\_model\]). Alternatively, we have also implemented a `detailed` setting that calculated [$J_{lu}^b$]{}-values using estimators (see Section \[sec:montecarlo\_estimators\]).
The net rate of change for a level population is $$\frac{dn_k}{dt} = \sum_{i \ne k} R_{ik} - \sum_{i \ne k} R_{ki} \; .$$ In `nlte` excitation mode, solves for level populations satisfying $\frac{d n_{k}}{dt} = 0$ for all $k$ (with $\sum_k n_{k} = N$, the total ion population). In practise, the code is initialized with LTE level populations and initial guesses for $\beta_{lu}$ and ${\ensuremath{J_{lu}^b}\xspace}$. It then solves the statistical equilibrium equations using LU decomposition [LAPACK; @laug]. After each iteration (see Section \[sec:montecarlo\_iteration\]) the statistical equilibrium equations are re-solved using the last level populations to compute values for $\beta_{lu}$ and results from the step to estimate ${\ensuremath{J_{lu}^b}\xspace}$. This process eventually converges to a stable set of level populations (see Section \[sec:montecarlo\_iteration\]).
Note that our current implementation includes only bound-bound rates in the set of statistical equilibrium equations – bound-free processes are neglected (but will be implemented as part of future ionization modules). Consequently, our `nlte` excitation mode will not yield accurate populations for levels that are significantly (de)-populated via bound-free processes. Nevertheless, our `nlte` excitation mode is certainly an improvement over LTE (or dilute-LTE) populations and provides a convenient means by which to quantify the plausible scale of errors introduced by the simple analytic formulae (Section \[sec:results\_nlte\]).
Initialization of Monte Carlo quanta
------------------------------------
For each simulation of the radiation field, we begin by initializing a population of $r$-packets at the inner boundary of the computational domain. Each $r$-packet is an indivisible quantum of radiative energy with an associated photon frequency ($\nu$). We choose to initialize all our $r$-packets with equal co-moving frame energy $E$.
Currently it is assumed that the radiation field injected through the inner boundary has a black body frequency distribution at temperature ${\ensuremath{T_\textrm{i}}\xspace}$ such that the luminosity launched at the inner boundary is
$$L_{i} = 4 \pi {\ensuremath{r_\textrm{i}}\xspace}^2 \sigma {\ensuremath{T_\textrm{i}}\xspace}^4$$
where ${\ensuremath{r_\textrm{i}}\xspace}$ is the radius of the inner boundary. Consequently, the co-moving frame frequencies of the packets are selected by randomly sampling the Planck function (for ${\ensuremath{T_\textrm{i}}\xspace}$) and the packet energies are chosen to match $L_{i}$: i.e., in a simulation with $N$ $r$-packets, the co-moving frame energy of each packet is initialised to
$$E = \frac{4 \pi {\ensuremath{r_\textrm{i}}\xspace}^{2} \sigma {\ensuremath{T_\textrm{i}}\xspace}^4}{N} \Delta t$$
where $\Delta t$ is the (arbitrarily chosen) time interval represented by the simulation. The procedure adopted to choose ${\ensuremath{T_\textrm{i}}\xspace}$ is described in Section \[sec:montecarlo\_iteration\]. The $r$-packets are assigned initial directions of propagation specified by $\mu = \sqrt{z}, z \in (0, 1]$, as appropriate for zero limb darkening. [^3]
Radiation-matter interactions {#sec:rad_matter_interaction}
-----------------------------
In our indivisible energy packet scheme, opacity does not destroy quanta but can change both the photon frequency and direction of propagation associated with a packet. The co-moving frame energy of the packet is always conserved during interaction. This ensures that radiative equilibrium is strictly enforced throughout the simulation.
Currently, treats only two classes of radiation-matter interaction – Thomson scattering by free electrons and bound-bound interactions with atoms/ions. For Thomson scattering the opacity encountered by an $r$-packet in a path length $s$ in direction $\mu$ is given by
$${\ensuremath{\tau_\textrm{e}}\xspace}=\kappa_e s$$
where $\kappa_e = \sigma_\textrm{T} {\ensuremath{n_\textrm{e}}\xspace}D_{\mu}$ is the observer frame opacity coefficient. Here, $\sigma_\textrm{T}$ is the Thomson cross section and ${\ensuremath{n_\textrm{e}}\xspace}$ is the number density of free electrons. We include the first order Doppler factor $D_{\mu} = (1 - \mu v / c)$ to preserve $O(v/c)$ accuracy in transforming between observer and co-moving frames . treats Thomson scattering as a coherent scattering process: it causes packets to change their direction but not their co-moving frame frequency or energy. Following a Thomson scattering event, the new direction is drawn from an isotropic distribution, $\mu = -1 + 2z, z \in [0, 1]$.
Bound-bound transitions are the dominant (and also most complex) opacity source in ejecta [@2000ApJ...530..757P]. treats bound-bound opacity in the Sobolev approximation [see e.g. @1999isw..book.....L], which is appropriate for media with large velocity gradients. In the case of homologous expansion, the Sobolev optical depth of the transition between lower state $l$ and upper state $u$ is given by
$${\ensuremath{\tau_\textrm{S}}\xspace}= \frac{\pi e^2}{m_e c}\, f \lambda_{lu} t_{\rm exp} n_l\, \left(1 - \frac{g_l n_u}{g_u n_l}\right) \, ,$$
where $f$ is the absorption oscillator strength of the transition, $\lambda_{lu}$ is the wavelength and $t_\textrm{exp}$ the time since explosion. The last term (in brackets) corrects for stimulated emission.
During the simulations (see Section \[sec:montecarlo\_propagation\]), $r$-packets can be absorbed by bound-bound transitions. Currently, three approaches are implements in to described the re-emission of an absorbed $r$-packet. In the simplest approach (mode `scatter`), all bound-bound interactions are treated as resonance scattering events \[cf. @1985ApJ...294..619B [@1993ApJ...405..738L]; Long & Knigge (2002, LK02)\]. Consequently, absorbed $r$-packets are simply re-emitted with the same co-moving frame frequency but a new direction of propagation (determined by randomly sampling an isotropic distribution, as above).
Our most sophisticated approach to line interactions is an implementation of the macro atom scheme (mode `macroatom`) devised by and used by @2009MNRAS.398.1809K. In the `macroatom` mode, bound-bound absorption of an $r$-packet activates a macro atom to the upper level of the absorbing transition. The macro atom algorithm is then used to simulate the re-emission of the absorbed energy, in accordance with the assumptions of radiative and statistical equilibrium. Currently, our macro atom implementation includes only bound-bound radiative transitions although we stress that the method can be readily extended to account for collisional and/or bound-free processes if suitable atomic data are provided. When a macro atom is activated, it can undergo a sequence of internal state transitions until ultimately deactivating and returning an $r$-packet to the simulation .
Our third method of handling bound-bound interaction is a simplified version of macro atom, referred to as `downbranch`. This version is effectively the same as the `macroatom` scheme except that the probabilities for all internal transitions are set to zero. Thus macro atoms are forced to deactivate directly from the state originally activated. This approach avoids some of the computational cost of the full macro atom machinery while retaining a simplified treatment of fluorescent processes, akin to that introduced by .
Propagation of Monte Carlo quanta {#sec:montecarlo_propagation}
---------------------------------
In the current implementation, there are three processes that can act to terminate the flight path of an $r$-packet: reaching a grid zone boundary, undergoing electron scattering or being absorbed by a bound-bound transition. Choosing which class of event occurs is determined via a simple experiment. First, an optical depth (${\ensuremath{\tau_\textrm{n}}\xspace}$) that the $r$-packet may propagate without being absorbed is randomly selected in accordance with the $\exp(-{\ensuremath{\tau_\textrm{n}}\xspace})$ attenuation law: ${\ensuremath{\tau_\textrm{n}}\xspace}= \ln{z}, z \in (0, 1]$. Next we identify the closest redward line transition to the co-moving frame frequency of the $r$-packet. Then we calculate the distance ([$d_\textrm{l}$]{}) the packet must travel to Doppler-shift into Sobolev resonance with that line. We also calculate [$d_\textrm{e}$]{}, defined by ${\ensuremath{\tau_\textrm{n}}\xspace}= \kappa_e {\ensuremath{d_\textrm{e}}\xspace}$. Finally, we calculate the distance [$d_\textrm{s}$]{}the $r$-packet would need to propagate to reach the boundary of the current shell.
To determine which class of event will terminate the flight path, we compare the three distances [$d_\textrm{l}$]{}, [$d_\textrm{e}$]{}and [$d_\textrm{s}$]{}. If [$d_\textrm{l}$]{}is the shortest, then bound-bound absorption is possible. To test whether line absorption occurs, we compare ${\ensuremath{\tau_\textrm{tot}}\xspace}= {\ensuremath{\tau_\textrm{S}}\xspace}+ \sigma_\textrm{T} {\ensuremath{n_\textrm{e}}\xspace}{\ensuremath{d_\textrm{l}}\xspace}$ to ${\ensuremath{\tau_\textrm{n}}\xspace}$. If ${\ensuremath{\tau_\textrm{tot}}\xspace}> {\ensuremath{\tau_\textrm{n}}\xspace}$ an interaction with the line will occur: we propagate the $r$-packet to the position of Sobolev resonance and then process the line interaction event in accordance with the procedures described in Section \[sec:rad\_matter\_interaction\] for bound-bound absorption. Alternatively, if ${\ensuremath{\tau_\textrm{tot}}\xspace}< {\ensuremath{\tau_\textrm{n}}\xspace}$, we reduce ${\ensuremath{\tau_\textrm{n}}\xspace}\rightarrow {\ensuremath{\tau_\textrm{n}}\xspace}- {\ensuremath{\tau_\textrm{S}}\xspace}$ and recompute ${\ensuremath{d_\textrm{e}}\xspace}$. We also recompute ${\ensuremath{d_\textrm{l}}\xspace}$, now considering the next redward line transition. With new values of ${\ensuremath{d_\textrm{e}}\xspace}$ and ${\ensuremath{d_\textrm{l}}\xspace}$ we again compare the three distances ([$d_\textrm{l}$]{}, [$d_\textrm{e}$]{}, [$d_\textrm{s}$]{}) and continue this process until either a line absorption event is triggered or until [$d_\textrm{l}$]{}is no longer the shortest distance.
If [$d_\textrm{e}$]{}is shortest, an electron scattering event terminates the $r$-packet flight path. The packet is propagated through distance ${\ensuremath{d_\textrm{e}}\xspace}$ and sent to the algorithm dealing with electron scattering events (see Section \[sec:rad\_matter\_interaction\]).
Finally, if [$d_\textrm{s}$]{}is shortest, the $r$-packet is propagated to the appropriate grid cell boundary and flagged as having successfully crossed that boundary. If the cell boundary is internal to the computational domain, the propagation of the packet through the new shell continues as before (noting that all material properties relevant to computing the opacity may have changed). When an $r$-packet reaches either of the boundaries of the computational domain (outer or inner), its flight path terminates and its final properties are recorded. $r$-packets reaching the outer boundary are assumed to the escape freely (such that they can contribute to the observable spectrum). Packets crossing the inner boundary are assumed to be reabsorbed by the inner ejecta and thus lost.
Monte Carlo estimators {#sec:montecarlo_estimators}
----------------------
During simulations, uses the $r$-packet trajectories to collect estimators for radiation field properties, as required for calculation of the ionization and excitation conditions. Specifically, we record a pair of estimators (following ML93; LK02), $${\ensuremath{J_\textrm{est}}\xspace}= \frac{1}{4\pi \Delta t V} \sum{E l D_{\mu}}$$
and $${\ensuremath{\bar{\nu}}\xspace}= \frac{1}{4\pi \Delta t V} \sum{E \nu l D_{\mu}}$$ for each grid cell. The summation is made over all $r$-packet trajectories inside the cell. For each trajectory, $E$ and $\nu$ are the packet energy and frequency (in the co-moving frame) and $l$ is the length of the trajectory (measured in the observer frame). After each calculation, we use these estimators to obtain new values for the parameters of our radiation field model (see Section \[sec:rad\_model\]):
$${\ensuremath{T_\textrm{R}}\xspace}= \frac{h}{k_{B}} \frac{\pi^4}{360 \zeta(5)} \frac{{\ensuremath{\bar{\nu}}\xspace}}{{\ensuremath{J_\textrm{est}}\xspace}}$$
and
$$W = \frac{\pi {\ensuremath{J_\textrm{est}}\xspace}}{\sigma_{SB} {\ensuremath{T_\textrm{R}}\xspace}^4}\:\: .$$
In [`detailed`]{} mode, we also record estimators for the mean intensity at the violet wing of each line transition $J_{lu}^b$ in each cell :
$$J_{lu}^b = \frac{1}{4\pi \Delta t V} \frac{t_{exp}}{c} \sum \frac{E}{\nu} D_{\mu}
\label{eq:jblue}$$
where the summation is now over all $r$-packets that pass through Sobolev resonance with the $l \rightarrow u$ transition in the cell.
Iteration cycle {#sec:montecarlo_iteration}
---------------
performs a sequence of simulation during which the values of ${\ensuremath{T_\textrm{i}}\xspace}$, ${\ensuremath{T_\textrm{R}}\xspace}$ and $W$ (and, in `detailed` mode, $J_{lu}^b$) are iteratively improved until convergence is reached.
To initialize a calculation, we set ${\ensuremath{T_\textrm{R}}\xspace}$ (in all shells) to a sensible guess, typically $10000$ K. The dilution factor is initialized to follow geometric dilution, $W=\frac{1}{2}[1 - (1- ({\ensuremath{r_\textrm{i}}\xspace}/r)^2)^{1/2}]$, where $r$ is the radius at the centre of each shell and ${\ensuremath{r_\textrm{i}}\xspace}= {\ensuremath{v_\textrm{i}}\xspace}t_{exp}$ is the radius of the inner boundary. In `detailed` mode, we initialize ${\ensuremath{J_{lu}^b}\xspace}= W B_{\nu_{lu}}({\ensuremath{T_\textrm{R}}\xspace})$. Although ${\ensuremath{T_\textrm{i}}\xspace}$ can also be treated as a simple input parameter, we anticipate the goal of fitting observations and so generally adopt the luminosity at the [*outer*]{} boundary ($L_o$) as a simulation parameter. As a first guess we then adopt
$${\ensuremath{T_\textrm{i}}\xspace}= \left( {\frac{{\ensuremath{L_\textrm{o}}\xspace}}{ 4 \pi {\ensuremath{r_\textrm{i}}\xspace}^2 \sigma}} \right)^{1/4} \; .$$
A simulation is then performed (see Section \[sec:montecarlo\_propagation\]) and the resulting estimators (see Section \[sec:montecarlo\_estimators\]) are used to update the plasma properties (see Section \[sec:plasma\_state\]) in each shell. We compare the total energy emerging through the outer boundary to the requested value of ${\ensuremath{L_\textrm{o}}\xspace}$ and modify ${\ensuremath{T_\textrm{i}}\xspace}$ (and therefore the luminosity at the inner boundary) to obtain better agreement (i.e., if the emergent luminosity is lower than requested, ${\ensuremath{T_\textrm{i}}\xspace}$ is increased and vice versa). Using these updated quantities, a new calculation is carried out and used to update the model properties again. This process is repeated until the plasma state has converged to sufficient accuracy (or a chosen maximum number of iterations is reached); convergence properties are described for example calculations in Section \[sec:numerics\].
Synthetic spectrum {#sec:synthetic_spectrum}
------------------
Synthetic spectra can be obtained directly from radiative transfer calculations by binning the quanta that emerge through the outer boundary in frequency-space. However, and , showed that higher-quality synthetic spectra can be extracted from the simulations using only slightly more sophisticated algorithms and various such approaches have been used in codes (ML93; LK02; @2008MNRAS.385.1681S).
In , we have implemented a simple approach to compute the synthetic spectrum, very similar to that adopted by LK02 and [@2010MNRAS.404.1369S]. This method is fully consistent with the macro atom scheme . The spectrum is calculated during one final simulation that is performed once the iterative sequence of simulations discussed in Section \[sec:montecarlo\_iteration\] has converged. During this last simulation, whenever an $r$-packet is about to be launched on a new trajectory (either because it is about to be injected through the inner boundary, or because an interaction occurred inside the computational domain), the main simulation is suspended and a set of $N_{v}$ test packets, which we refer to as [*virtual*]{} ($v$-) packets, are created. These $v$-packets have properties identical to those of the $r$-packet that was about to be launched, except that they each have a different propagation direction and their energies are assigned based on the probability distribution associated with the creation of the original $r$-packet (see below). Since the only purpose of $v$-packets is to estimate a contribution to the emergent spectrum, we do not spawn $v$-packets on trajectories that ultimately intersect the inner boundary. Thus, for $v$-packets created at radius $r$, we assign directions ($\mu_v$) in the interval $\mu_{min} < \mu_v < 1$, where $\mu_{min} = -\sqrt{(1 - ({\ensuremath{r_\textrm{i}}\xspace}/r)^2)}$. For $v$-packets generated following a physical interaction inside the domain, the co-moving frame energy assigned ($E_{v}$) is simply
$$E_{v} = E \frac{1 - \mu_{min}}{2 N_{v}}$$
where $E$ is the co-moving frame energy of the parent $r$-packet. For $v$-packets created when an $r$-packet is about to be injected through the inner boundary, the energies assigned are modified to account for the adopted angular distribution of the incoming radiation field $$E_{v} = E \frac{2 \mu_v}{N_{v}}$$ Once created, each $v$-packet is propagated though the simulation domain. Unlike $r$-packets, $v$-packets do not undergo physical interactions with the medium (they are never scattered), but the total optical depth along the trajectory of the$v$-packet ($\tau_v$) is recorded. Each $v$-packet makes a contribution to the photon frequency-bin of the synthetic luminosity (ergs s$^{-1}$ Hz$^{-1}$) spectrum in which its observer frame frequency lies
$$L_{v} = \frac{E_v}{\Delta t \Delta \nu} \exp({- \tau_v})$$
where $\Delta \nu$ is the width of the spectral frequency bin. The spectrum computed from the $v$-packets agrees with that obtained by direct binning of emerging $r$-packets but typically has considerably less noise. The reduction in noise and the computational overhead introduced by the $v$-packet step are determined by the choice of $N_{v}$. For the calculations in this paper we typically use $N_{v} = 3$ which reduced the noise by a factor of $\approx 3$.
Datasources {#sec:datasources}
===========
Our radiative transfer calculations require input atomic data for the calculation of the ionization/excitation state of the plasma (ionization potentials and atomic models), bound-bound opacities and radiative rate coefficients (lines lists and oscillator strengths) and, in our `nlte` excitation mode, electron collision rate coefficients (thermally averaged collision strengths).
For the simulations presented below we use an atomic database that includes all elements with $Z\le30$. Atomic masses were taken from @wieser2011atomic and ionization threshold energies from . The modified nebular approximation (see Section \[sec:plasma\_state\] equation \[eq:modified\_nebular\]) requires that the fraction of recombinations to the ground state ($\zeta$-factor) be specified. For this, values were extracted from the data compiled by LK02 for a fixed set of temperatures (2000 – 50000 K; 2000 K grid spacing). During code execution, linear interpolation between these values is used to obtain $\zeta$ as a function of temperature.
We have drawn atomic/ionic energy levels and bound-bound (line) transition data from two sources and constructed two atomic databases. In the first case, we populate the lines and levels database from the omitting levels above the ionization threshold and lines with an oscillator strength $\log{gf} < -3$. This dataset contains no electron collision rate data. For ions not present in the , we use atomic models consisting of only the ground state (statistical weights from ) and no line transitions. We use this dataset for several of the test calculations presented in Section \[sec:code\_comparisons\].
Secondly, we constructed a data set in which the data are replaced with data from for select species of relevance to SNe Ia: , and . provides atomic models and data for both radiative and electron collision bound-bound transitions. For ions taken from , no cuts were applied to $\log{gf}$ in selecting radiative data. Electron collision rates were calculated using the atomic dataset for temperatures between 2000 K and 48000 K (in steps of 2000 K) and interpolated during code execution. This data set is used for most of the the calculations presented in Sections \[sec:numerics\], \[sec:code\_comparisons\] and \[sec:results\].
The two atomic data sets used here have been constructed with the goals of this study in mind (code testing and simple differential tests of modelling assumptions). However, is capable of handling much larger atomic datasets and can easily be extended to include additional atomic processes.
Convergence tests {#sec:numerics}
=================
As described in Section \[sec:montecarlo\_iteration\], performs an iterative sequence of simulations. Here we describe tests verifying the convergence properties of this iteration procedure. For these tests, we have chosen to use a model that is based on the output of a 1D explosion simulation – this allows us to study the operation of the code applied in the most complex regime available (model with arbitrary density profile and stratified abundances including a large number of elements). [^4] Specifically, we adopt the model for a detonation of a 1.06 M$_{\odot}$ white dwarf (WD) [@2010ApJ...714L..52S] that was computed for a uniform initial composition of $^{12}$C and $^{16}$O (50:50 mix with no $^{22}$Ne). The density and composition for this model during the homologous phase are illustrated (at $t = 11.1$ days) in Fig. [\[fig:artis\_model\]]{}. The model has strongly layered ejecta, with dominating the composition above $\sim 9000$ km s$^{-1}$ and iron-group elements dominating below. The complete model from Sim et al. (2010) consists of 70 shells. However, since inner and outer boundaries are imposed by , typically only around half of the shells are included in the calculations. For the test calculations we adopt $t_{exp} = 11.1$ days, $v_{i} = 11000$ km s$^{-1}$, ${\ensuremath{v_\textrm{o}}\xspace}= 22000$ km s$^{-1}$ and $\log_{10} {\ensuremath{L_\textrm{o}}\xspace}/ L_{\odot} = 9.34$ (the luminosity obtained at this epoch from simulations – see Section \[sec:artis\_compare\]).
![Densities for the 1.06 M$_{\odot}$ CO WD detonation model of @2010ApJ...714L..52S at 11.1 days after explosion. The black line shows the total mass density and the coloured lines show the densities of important elements in the ejecta.[]{data-label="fig:artis_model"}](\plotdir /artis_model.pdf){width="\columnwidth"}
We have tested convergence by carrying out a sequence of 30 iterations with $2\times 10^5$ quanta per iteration. To test all aspects of the current implementation, we show results from calculations with the most sophisticated set of modes of operation: `macroatom` mode for line interactions, `detailed` mode for calculation of radiative rates and `nebular` ionization mode. We use `nlte` excitation mode for and `dilute-lte` excitation for all other ions.
As described in Section \[sec:montecarlo\_iteration\], the code iteratively tries to match the requested SN luminosity (${\ensuremath{L_\textrm{o}}\xspace}$) by changing the temperature of the black-body spectrum emitted from the inner boundary. This in turn changes the radiation field in the cells above (parametrized by [$T_\textrm{R}$]{}and $W$), which subsequently influences the ionization and excitation of the plasma. To speed convergence, we found it useful to adopt a scheme in which [$T_\textrm{i}$]{}is only changed after every third iteration: this allows [$T_\textrm{R}$]{}and $W$ to respond to changes in [$T_\textrm{i}$]{}, helping the code make a more informed choice when modifying [$T_\textrm{i}$]{}in an attempt to match ${\ensuremath{L_\textrm{o}}\xspace}$.
Fig. \[fig:tardis\_convergence\_nlte\] (upper panels) shows the rapid and uniform convergence for both [$T_\textrm{R}$]{}and $W$ for our test calculation. By iteration 20 both quantities have ceased to evolve with iteration number. This is driven by the quick convergence of [$T_\textrm{i}$]{}(see Fig. \[fig:tardis\_convergence\_mcquanta\]).
We have tested the convergence properties of level populations for species treated with our `nlte` excitation mode by examining the evolution with iteration number of departure coefficients defined by
$$b_\textrm{LTE}=\frac{n_i/n_0}{n_i^\textrm{LTE}/n_0^{\textrm{LTE}}},$$
where $n_i$ is the level population of the selected level and $n_0$ is that of the species ground state. This is illustrated for the $3\textrm{s}^2 4\textrm{s}\ ^2\textrm{S}_{1/2}$ state of (see Fig. \[fig:grotrian\_si2\]), in Fig. \[fig:tardis\_convergence\_nlte\]. Since the `nlte` excitation level populations are strongly affected by the radiative rates, we also show the convergence of [$J_{lu}^b$]{}for $\lambda 6347$ in Fig. \[fig:tardis\_convergence\_nlte\]. Although considerably more affected by noise than [$T_\textrm{R}$]{}or $W$, both [$J_{lu}^b$]{}and the departure coefficient have converged to a stable pattern after roughly 20 iterations. We explore the influence of noise in the estimators in Section \[sec:results\_jblue\].
We have tested the sensitivity of our results to the number of quanta by repeating the calculation described above using $10^6$ and $2 \times 10^6$ quanta. The convergence of these runs was essentially identical to those of our calculation with $2 \times 10^5$ quanta and led to no changes in the output spectra (see Fig. \[fig:tardis\_convergence\_mcquanta\]). In addition, we show that the spectra extracted using *virtual* packets (see Section \[sec:synthetic\_spectrum\]) are much less noisy than those obtained by directly extracting a spectrum by binning emergent quanta.
Finally, to test the influence of grid resolution, we carried out additional runs in which we increased the number of shells used by factors of two, five and ten (using $2 \times 10^5$ quanta in each case). We interpolated the velocities linearly (the entire model follows a strictly homologous velocity law), but did not interpolate the densities or abundances (we aim to test the influence of resolution on our computed radiation field properties, not the input model). Despite the finer sampling, the converged properties and the emergent spectrum for these tests were unaffected indicating that the resolution adopted in our standard calculation was adequate.
Code Comparisons {#sec:code_comparisons}
================
We have made comparisons between calculations and those of several other codes. We first focus on the calculation of the emergent spectrum (for fixed plasma conditions) by comparison with (Section \[sec:compare\_synpp\]). We then use the package to test our solver for NLTE level populations in Section \[sec:compare\_chianti\].
In Sections \[sec:python\_compare\] and \[sec:artis\_compare\] we make more sophisticated tests by comparing synthetic spectra to those of two alternative radiative transfer codes, [python]{} and . Like , these codes use methods and a Sobolev treatment of line opacity. However, they have different treatments of NLTE ionization, bound-free opacity and radiation sources (as detailed below). Comparing their results allows us both to identify potential issues (with any of the codes) and to quantify the effects of the different assumptions made by the codes.
Comparison to [syn++]{} {#sec:compare_synpp}
-----------------------
Similar to , adopts a spherically symmetric SN model in homologous expansion. handles line opacity in the Sobolev approximation and assumes all lines can be treated in the resonance scattering limit. In contrast to , does not accept density and abundances as input parameters, but rather the opacity for a reference line associated with each ion species. then calculates the opacities of other lines using the Boltzmann excitation formula at a user-specified temperature. Thus, comparison with does not test the calculation of the plasma state (ionization/excitation) in . However, it does provide a direct test of the accuracy to which the spectrum is computed for fixed plasma conditions.
For the test, we setup with a pure silicon one-zone model. Line interactions were treated in `scatter` mode and electron scattering was disabled for ease of comparison with . Ionization and excitation were both treated in `lte` modes and we used the atomic data set drawn only from the . The temperature of the inner boundary and of the temperature of the radiation field inside the only cell were set to 10000 K and $10^6$ quanta were used to generate the spectrum. The output [$\tau_\textrm{S}$]{} for the reference lines of species -<span style="font-variant:small-caps;">iv</span> were taken from and used to calculate the spectrum with . In Fig. \[fig:compare\_synpp\_tardis\], we illustrate the excellent agreement between both codes, validating the method of spectrum formation used by .
Testing of NLTE level populations {#sec:compare_chianti}
---------------------------------
To test the accuracy of our NLTE level population scheme, we compare our results to those obtained using . For this test, we calculate the NLTE level populations for a plasma element of fixed electron density ${\ensuremath{n_\textrm{e}}\xspace}$ and kinetic temperature ${\ensuremath{T_\textrm{e}}\xspace}$ irradiated by a diluted black body ($W=0.5$) with radiation temperature ${\ensuremath{T_\textrm{R}}\xspace}$. Note that, as in , the current version of includes only bound-bound (radiative and electron collision) rates when solving for level populations (i.e. the influence of photo-ionization and recombination are not included).
In general we found excellent agreement between the calculation performed by and . Fig. \[fig:tardis\_chianti\_classical\_nebular\] shows the departure coefficient $b_\textrm{LTE}$ for computed for ${\ensuremath{T_\textrm{e}}\xspace}=9000$ K, ${\ensuremath{T_\textrm{R}}\xspace}=10000$ K and ${\ensuremath{n_\textrm{e}}\xspace}=4.15\times10^8 \textrm{cm}^{-3}$. We find a bias of 0.006 and a standard deviation of 0.003 in the difference between the sets of departure coefficients obtained in the two calculations.
Comparison to [python]{} {#sec:python_compare}
------------------------
is a 2D MC radiative transfer code, which has been used to model synthetic spectra for accretion disk winds in a variety of astrophysical systems: cataclysmic variables (LK02; @2012MNRAS.425.1430N), massive young stellar objects [@2005MNRAS.363..615S] and active galactic nuclei [@2013MNRAS.436.1390H]. As in , an input radiation field is specified and synthetic spectra are computed by tracking the propagation of radiation through a computational domain. includes bound-bound, bound-free, free-free and electron scattering opacities. Although a `macroatom` treatment is implemented in , to date this has only been used for modelling recombination lines [@2005MNRAS.363..615S]: for metal lines, a two-level atom is used.
For our comparison calculation, we adopt a simple SN Ia ejecta model, following . Specifically, we choose a density profile based on a simple fit [$\rho \propto v^{-7}$; cf. @1985ApJ...294..619B] to the W7 model of @1984ApJ...286..644N. We carry out the comparison for an epoch of 13 days post explosion, choosing an emergent bolometric luminosity of $\log_{10} {\ensuremath{L_\textrm{o}}\xspace}/L_{\odot} = 9.44$ and setting the inner boundary of the computational domain at ${\ensuremath{v_\textrm{i}}\xspace}= 11000$ km s$^{-1}$. We set the outer boundary at ${\ensuremath{v_\textrm{o}}\xspace}= 20000$ . Since cannot currently handle position-dependent compositions, we adopt a uniform composition for the ejecta. At these epochs, the observed optical spectra of normal are dominated by features associated with . Therefore, we adopted an -rich composition (see Table \[tab:python\_tardis\_abund\]) with relative abundances based on those found in modern SN Ia explosion models [specifically @2013MNRAS.429.1156S]. The calculation was carried out using `nebular` ionization and `dilute-lte` excitation modes (see Section \[sec:plasma\_state\]). For ease of comparison, we used our `scatter` mode for line opacity, which is our closest equivalent to the two-level treatment of metal lines used in , and our atomic data set drawn from (see Section \[sec:datasources\]).
Element Mass fraction Element Mass fraction
--------- --------------- --------- --------------- -- --
O 0.19 Mg 0.03
Si 0.52 S 0.19
Ar 0.04 Ca 0.03
: Elemental mass fractions adopted for the test model used in Section \[sec:python\_compare\].[]{data-label="tab:python_tardis_abund"}
For the comparison run we set up the same model, using spherical geometry and imposing an homologous velocity law. The input radiation field for the calculation was forced to be the same as in the calculation (i.e. we inject a black-body radiation field through the inner boundary with temperature set to be the same as that used by ). The calculations were also carried out using an atomic data from the .
![Synthetic spectra computed with and for the test model described in Section \[sec:python\_compare\]. The calculations were carried out using our `scatter` and `nebular` ionization modes. In addition to our standard case, we also plot the spectrum obtained if we enforce $\delta = 1$ in (see Equation \[eq:modified\_nebular\]).[]{data-label="fig:python_compare"}](\plotdir /omgsisarca.pdf){width="\columnwidth"}
The synthetic spectra are compared in Fig. \[fig:python\_compare\]. In general, the agreement is very good: the strengths and shapes of the line feature are well-matched, as is the total emergent intensity (in these test calculations, roughly 30 per cent of the radiation launched through the inner boundary is backscattered by the ejecta such that the total amplitude of the emergent spectrum is not trivially reproduced by imposing the same radiative flux at the inner boundary).
{width="\columnwidth"} {width="\columnwidth"}\
{width="\columnwidth"} {width="\columnwidth"}
There are, however, some discrepancies. For example, the strength of the feature ($\sim 5300$ Å) is noticeably different and there are small wavelength offsets: for example, $\lambda$ 6355, 5050 and the H+K / IR triplet features have larger blueshifts in the spectrum. Many of these discrepancies can be attributed to subtle differences in the treatment of ionization. In particular, the modified nebular approximation used by does not incorporate the correction factor $\delta$ proposed by ML93. If we do not include this correction (i.e. we force $\delta = 1$ in Equation 3), we find an even closer match between the two codes (see Fig. \[fig:python\_compare\]). Although still imperfect, this level of agreement between two independent codes gives confidence that both are performing well. Our comparisons also confirm that, although not qualitatively critical, even relatively minor changes to the ionization approximation are relevant to high-precision modelling.
Comparison to [artis]{} {#sec:artis_compare}
-----------------------
The code is a purpose-build, multi-dimensional SN radiative transfer code. In contrast to and , carries out time-dependent calculations that produce time sequences of synthetic spectra for an input model. calculations do not involve an inner boundary through which radiation is injected – instead, the energy injection is followed in detail by simulating the emission of $\gamma$-rays origination from radioactive decays . Avoiding the need for an artificial inner boundary is the greatest advantage of codes such as , but this comes at considerable computational expense: a single calculation typically takes 10 – 100 kCPU hours, which is prohibitive for studies in which large parameter spaces are to be searched.
Like , includes bound-bound, bound-free, free-free and electron scattering opacities. also includes an implementation of the macro atom scheme for all ions and an ionization approximation that is based on detailed photo-ionization rate estimators [see @2009MNRAS.398.1809K for details]. As in , the ejecta are assumed to be in homologous expansion.
For our comparisons between and , we use the same detonation model described in Section \[sec:numerics\]. The simulations for the model yield a sequence of synthetic spectra covering the ultraviolet to infrared wavelength regions. We shall focus on comparisons at two epochs, 11.1 and 18 days after explosion (i.e. roughly at, and one week before, the time of maximum light in the optical bands). For the comparison calculations, we use the luminosities calculated by ($\log_{10} {\ensuremath{L_\textrm{o}}\xspace}/ _{\odot} = 9.34$, 9.50 for $t = 11.1, 18.0$ days, respectively) and adopt ${\ensuremath{v_\textrm{o}}\xspace}= 22000$ km s$^{-1}$. We must also choose a location for the inner boundary of the computational domain. Based on the model structure (Fig. [\[fig:artis\_model\]]{}), it seems inappropriate to consider $v_{i} {\,\hbox{\lower0.6ex\hbox{$\sim$}\llap{\raise0.6ex\hbox{$<$}}}\,}8000$ km s$^{-1}$, since this would place the inner boundary within the $^{56}$Ni-rich layers. Consequently, we have carried out several runs for both epochs that explore the effect of choosing a range of values for $v_{i} \ge 9000$ km s$^{-1}$.
Synthetic optical and ultraviolet spectra from our calculations are compared to spectra in Fig. \[fig:artis\_model\_spec\]. First, we note that the choice of $v_{i}$ does not significantly affect the optical spectrum, even when varied over a fairly wide range ($9000 - 11000$ km s$^{-1}$). This is reassuring since significant sensitivity to the choice of $v_{i}$ would pose a challenge for the rapid exploration of model parameter space for which is intended. If $v_{i}$ is made very large ($v_{i} {\,\hbox{\lower0.6ex\hbox{$\sim$}\llap{\raise0.6ex\hbox{$>$}}}\,}12000$ km s$^{-1}$) consequences do appear, primarily in the ultraviolet and blue regions of the spectra, due to the lack of line-blocking by iron-group elements when only the high velocity ejecta are included (see Fig. [\[fig:artis\_model\]]{}).
Second, for a reasonable choice of $v_i$, the shapes and strengths of the optical spectral features agree fairly well in the two codes, suggesting that the simplifications made in do not severely limit its ability to model optical lines, particularly for early epochs. Our comparison do, however, highlight important differences: in the calculations there is more ultraviolet emission (blueward of $\sim 3000$ Å) and correspondingly less optical emission. This discrepancy is significantly stronger at the later epoch considered – at $18$ days, the optical pseudo continuum in is clearly brighter than predicted by . The origin of these differences is likely a combination of factors arising from the different approximations used. In Fig. \[fig:artis\_model\_spec\], we show that both the radiation temperature ([$T_\textrm{R}$]{}) and the mean intensity[described in terms of $T_{J} = W^{1/4} T_{r}$; see @2009MNRAS.398.1809K] and fairly similar in both calculations. However, there is a distinct offset in the computed ionization state. For example, ionization fractions for Fe are shown in Fig. \[fig:artis\_model\_spec\]: while the overall pattern of the ionization is similar in both calculations, the ionization approximation used by leads to systematically lower mean ionization. Since higher abundances of the singly ionized iron-group elements will lead to more effective line-blocking at blue wavelengths, this likely is responsible for much of the difference between the ultraviolet flux level in the two calculations.
Clearly, further study and improvement of the ionization approximations used is needed for quantitative modelling, and the limitations of these approximations must be borne in mind when comparing synthetic spectra from any of the codes discussed here to observations. Nevertheless, we conclude that, despite a considerable reduction in complexity, the code is already capable of providing a reasonable approximation to the calculations (for suitably chosen luminosity), lending credence to its use for initial attempts to fit spectra and explore model parameter spaces.
Results {#sec:results}
=======
Thanks to its modularity, is well suited to study the influences of different physical assumptions on the spectrum. Here we present results of simple calculations that investigate the differential effects of two important approximations that are sometimes adopted in the modelling of SN Ia spectra. For these tests we again use the sub-Chandrasekhar mass detonation model introduced in Section \[sec:numerics\]. We will show results computed for two epochs for which we adopt [$L_\textrm{o}$]{}and [$v_\textrm{i}$]{}as given in Table \[tab:artis\_epoch\_models\].
NLTE excitation {#sec:results_nlte}
---------------
As noted in Section \[sec:intro\], several of the existing radiative transfer codes (including , and ) make use of simple approximations for the treatment of excited level populations. This has advantages for computational expediency and, particularly for multi-dimensional simulations, for reducing memory requirements but it comes at the cost of reduced accuracy. In contrast, studies by , @1996MNRAS.283..297B and @2013MNRAS.429.2127B have made use of the full equations of statistical equilibrium incorporating both ionization and excitation and shown that NLTE effects can be important. Here, we focus on exploring approximations for excitation (rather than ionization) with a view to quantifying the systematic errors introduced by the simple (and computationally cheap) `dilute-lte` formula to estimate excitation states.
We focus on four important ions, namely , , and . Each of these exhibit prominent features in the spectra of and understanding potential systematic uncertainties in their modelling is an important ingredient in attempts to make synthetic fits to observed data sets. For our comparisons, we adopted the `nebular` ionization, `macroatom` line interaction and `detailed` radiative rates modes (i.e. the most sophisticated set of assumptions currently implemented) and then compared results obtained with `dilute-lte` and `nlte` excitation modes. We considered each of our selected ions in turn: i.e. for both epochs considered, we have computed a total of five synthetic spectra – one in which each of the four selected ions is treated in `nlte` excitation mode (all other ions in `dilute-lte` excitation mode) and one comparison calculation in which all ions are handled in `dilute-lte` excitation mode. This approach is convenient to assess the direct consequences of the level populations in key ions but we note that it does not capture any complex NLTE effects where one ion can influence the radiation field and therefore the excitation state of other ions. In all calculations, $10^7$ quanta were used and 30 iterations were carried out. Synthetic spectra from these calculations are compared in Fig. \[fig:tardis\_result\_nlte\].
To help quantify the effects of our different excitation treatment, we define a departure coefficient for multiplets
$$b_\textrm{LTE}^{m}=\frac{\sum_i n_i/n_0}{\sum_i n_i^\textrm{LTE}/n_0^{\textrm{LTE}}},$$
where $i$ runs over the $J$ sub-states of the term.
### Silicon
The $\lambda 6355$ feature ($3\textrm{s}^2 4\textrm{s}~^2\textrm{S} \rightarrow 3\textrm{s}^2 4\textrm{p}~^2\textrm{P}$; see Fig. \[fig:grotrian\_si2\]) is characteristic of spectra and one of the main identifiers for this class. We find that the population of the $3\textrm{s}^2 4\textrm{s}~^2\textrm{S}$ state is relatively unaffected by our choice of excitation mode, resulting in a very small difference of departure coefficients between the two excitation treatments (see Table \[tab:departure\_coefficients\]) and little change in the strength or shape of the spectral feature (Fig. \[fig:tardis\_result\_nlte\]).
In contrast, both the $\lambda 5050$ ($3\textrm{s}^2 4\textrm{p}~^2\textrm{P} \rightarrow 3\textrm{s}^2 4\textrm{d}~^2\textrm{D}$) and the $\lambda 5972$ ($3\textrm{s}^2 4\textrm{p}~^2\textrm{P} \rightarrow 3\textrm{s}^2 5\textrm{s}~^2\textrm{S}$) features are affected by the excitation mode: departure coefficients for $3\textrm{s}^2 4\textrm{p}~^2\textrm{P}$ are illustrated in Fig. \[fig:tardis\_result\_departure\_evolution\] (and also listed in Table \[tab:departure\_coefficients\]). Clearly our `nlte` excitation mode leads to significantly reduced populations for this state and correspondingly weaker spectral features compared to the `dilute-lte` approximation. E.g., in our $11.1$ d (18.0 d) spectra, the equivalent width (EW) of $\lambda 5972$ drops from 48 (38) to 24 (18) Å.
The different sensitivities of the departure coefficients for $3\textrm{s}^2 4\textrm{s}~^2\textrm{S}$ and $3\textrm{s}^2 4\textrm{p}~^2\textrm{P}$ mean that the choice of excitation treatment will systematically affect quantitative modelling of the equivalent width ratio $EW(\lambda 6355) / EW(\lambda 5972)$, which has been shown to act as a luminosity indicator [@1995ApJ...455L.147N; @2008MNRAS.389.1087H]. Thus our calculations verify that, as in [@2013MNRAS.429.2127B], NLTE effects must be included if quantitative analysis (factor $\sim 2$ accuracy or better) of this line ratio is to be used to test models.
### Sulphur
is identified by iconic features around $\lambda 5449$ and $\lambda5623$ in the spectra of SNe Ia (see Fig. \[fig:tardis\_result\_nlte\]): these are dominated by the $3\textrm{s}^2 3\textrm{p}^2 (3\textrm{P}) 4\textrm{s}~^4\textrm{P} \rightarrow 3\textrm{s}^2 3\textrm{p}^2 (3\textrm{P}) 4\textrm{p}~^4\textrm{D}$ and $3\textrm{s}^2 3\textrm{p}^2 (3\textrm{P}) 3\textrm{d}~^4\textrm{F} \rightarrow 3\textrm{s}^2 3\textrm{p}^2 (3\textrm{P}) 4\textrm{p}~^4\textrm{D}$ multiplets. As for the $3\textrm{s}^2 4\textrm{p}~^2\textrm{P}$ states in , we find that our `nlte` excitation mode results in smaller level populations than `dilute-lte`, particularly for $3\textrm{d}~^4\textrm{F}$, which is classified as meta-stable in the simpler scheme. This has quite noticeable ramifications for the shape of the S [ii]{} feature (Fig. \[fig:tardis\_result\_nlte\]), again highlighting the sensitivity to excitation treatment in high-precision modelling.
### Magnesium
Two particular features associated with Mg [ii]{} could be of note in the modelling of optical SN Ia spectra: the strong $\lambda 4481$ line ($3\textrm{d}~^2\textrm{D} \rightarrow 4 \textrm{f}~^2\textrm{F}$) and the $4 \textrm{p}~^2\textrm{P} \rightarrow 4 \textrm{d}~^2\textrm{D}$ ($\lambda 7896$) transition, which is a potential contaminant of the oxygen feature at $\lambda 7774$. Once again we find that our `nlte` excitation treatment noticeably depopulates the lower levels of these transitions (compared to `dilute-lte`). This causes the $\lambda 4481$ to become weaker (particularly at the later epoch, when it is less saturated) and the contribution of to the blend with $\lambda 7774$ to nearly vanish. This lends confidence to the use of $\lambda 7774$ as an oxygen abundance tracer [@2013MNRAS.429.2127B].
### Calcium
Ca [ii]{} imprints distinct spectral features at both ends of the optical spectrum: in the blue, the H and K lines ($\lambda 3950$ arising from $4\textrm{s}~^2\textrm{S} \rightarrow 4\textrm{p}~^2\textrm{P}$) and in the near-infrared at $\lambda 8498, 8542, 8662$ ($3\textrm{d}~^2\textrm{D} \rightarrow 4\textrm{p}~^2\textrm{P}$). In both epochs considered, we find only a modest difference between departure coefficients in `nlte` and `dilute-lte` excitation modes for the $3\textrm{d}~^2\textrm{D}$ state (see Table \[tab:departure\_coefficients\]), but this does not lead to a significant change in the shape or strength of the near-infrared feature (the absorption is already saturated). Thus, of those considered, the features are the least sensitive to the choice of excitation mode and are well-represented by the simple excitation formula.
Line interaction schemes {#sec:results_line_interaction}
------------------------
As described in Section \[sec:rad\_matter\_interaction\], allows for three different bound-bound interaction schemes: `scatter`, `downbranch` and `macroatom` modes. In this section, we explore the influence of the interaction scheme on the spectrum.
showed that the ultraviolet flux is significantly affected if line branching is taken into account. When compared to calculations that include only resonance scattering, inverse fluorescence by iron group elements leads to enhanced emission in the ultraviolet. Our calculations also show this behaviour quite clearly – spectra calculated using `scatter` and `downbranch` modes are compared in Fig. \[fig:tardis\_result\_line\_interaction\].
With we can also test the extent to which the macro atom scheme (see Section \[sec:rad\_matter\_interaction\]) alters the spectrum. In our calculations, the difference between `macroatom` (here combined with `detailed` mode for radiative rates) and `downbranch` mode is most apparent near maximum light (see Fig. \[fig:tardis\_result\_last\_line\_interaction\]) in the region around the features, $\sim$5500Å. The differences arise from the more complex pattern of frequency redistribution afforded by the macro atom scheme – this is illustrated in Fig. \[fig:tardis\_result\_last\_line\_interaction\] where we show the increasing complexity of the frequency redistribution associated with the last line-interaction events that occurred for quanta in simulations with each of our line interaction modes.
To quantify the effects of the different line interaction methods we have measured equivalent widths of the feature ($\lambda 5449$ and $\lambda5623$) and the $\lambda 5987$ line (see Table \[tab:line\_interaction\_eqw\]). We find that the former is sensitive to the line interaction method on the level of $\sim 30$ Å but the later by only $\sim 3$ Å.
Radiative rates estimation {#sec:results_jblue}
--------------------------
Amongst the options currently implemented in , the treatment of the bound-bound radiative rates has the most significant implications for the computational cost of a calculation. In `detailed` mode, it is necessary to have a sufficiently large number of quanta (many millions) to provide adequate statistics for the individual [$J_{lu}^b$]{}estimators (see Section \[sec:montecarlo\_estimators\]) in every grid cell. Although more accurate, this is considerably more demanding than `dilute-blackbody` mode (Section \[sec:plasma\_state\]) for which good convergence can be obtained with numbers of packets that are one to two orders of magnitude smaller. In this section we compare results from our two modes for bound-bound radiative rates to investigate whether the extra computation cost of `detailed` mode is warranted.
For the comparison, we repeat the calculation from Section \[sec:results\_nlte\] adopting `dilute-blackbody` mode for the radiative rates. In Fig. \[fig:tardis\_result\_jblue\_evolution\], we compare values of [$J_{lu}^b$]{}for a selection of line transitions between the calculations with `detailed` and `dilute-blackbody` mode. We find that, although rather simplistic, the `dilute-blackbody` assumption for [$J_{lu}^b$]{}performs rather well – although there are modest deviations, the overall shape of the `detailed` calculation is well-matched, particularly in the inner regions where most of the spectrum formation occurs.
Given that [$J_{lu}^b$]{}is well-represented by the `dilute-blackbody` assumption, it is to expected that the important quantities that depend on the bound-bound radiative rates will not be very adversely affected by this assumption. This is confirmed in Fig. \[fig:tardis\_result\_departure\_evolution\], which compares departure coefficients for one example, and Fig. \[fig:tardis\_result\_j\_blue\_spectral\_comparison\], in which the complete synthetic spectra computed using our two treatments of the radiative rates are shown. Consequently, we conclude that the `dilute-blackbody` assumption is generally acceptable and is therefore recommended for rapid modelling owing the considerable reduction in computational cost that it brings.
Conclusion & Future work {#sec:conclusions}
========================
We have presented a new 1D radiative code () for modelling of spectra, which is based on the indivisible-packet methods developed by Lucy (@1985ApJ...288..679A [@1993ApJ...405..738L]; ML93; ). The purpose of the code is to allow rapid but accurate synthesis of spectra with only a few input parameters (see Appendix \[sec:using\_tardis\]). It is built in a modular fashion, making it possible for the user to activate or deactivate different physics and approximations so that it can be tuned for use in different applications.
Our successful comparisons with other codes (, , ) verify that is operating as expected and can accurately reproduce the shapes and strengths of complex spectral features obtained from independent calculations. Our results also highlight the importance of different ionization treatments in reconciling the results obtained with different codes and the ongoing need to pursue a good balance between accuracy and computational expediency in the treatment of ionization and excitation.
The modularity built into allows for the exploration of different treatments of microphysics. In this work, we first explored the effect of simple excitation approximations (applied to four ions of relevance to , namely , , , ). We conclude that, for all of these except , the excitation treatment has a significant effect on the strength of line features which should be considered when attempting quantitative modelling of observations. We defer to a future study an equivalent investigation of approximate NLTE ionization treatments, which are likely to have more significant ramifications for the overall spectral shape .
Secondly, we investigated the effect of different line interaction schemes on synthetic spectra. Similar to previous studies we found that the assumption of pure resonance scattering underestimates the ultraviolet luminosity of . Comparing the full macro atom approach to a simple down branching scheme shows that there is a minor but noticeable difference. As the computational overhead of using the macro atom scheme is only 10% above the simpler down branching, we advocate for the use of the macro atom scheme in future studies.
We also explored the sensitivity of results to the different schemes for obtaining radiative rates for bound-bound transitions. Comparing results obtained using radiative rates recorded with line-by-line estimators to a simplified dilute-blackbody model (as adopted for the radiative rates in ), we see only minute changes in the synthetic spectrum. Given that the detailed estimators lead to an increase in computational time by orders of magnitude, our findings support the use of the simpler treatment.
is now publicly available for use. For approximate modelling of observations, we would generally recommend combining the `nebular` ionization / `dilute-lte` excitation plasma modes with the `dilute-blackbody` and `macroatom` interaction modes as a viable (computationally manageable) mode of operation. To fit a particular observation, a user would need to supply the luminosity and then develop a model by choosing a density profile \[which could be empirical or based on an explosion model such as W7 [@1984ApJ...286..644N]\] and a set of ejecta abundances (which can be uniform or stratified). The density and abundances can then be modified to attempt to improve the fit (and therefore constrain the SN properties, as in e.g. @2005MNRAS.360.1231S [@2008MNRAS.386.1897M; @2009MNRAS.399.1238H; @2011MNRAS.410.1725T]). However, we stress that this is only one potential use/mode of operation for and we encourage potential users to refer to the manual <http://tardis.rtfd.org> for further details.
In the near future, we plan to focus on two distinct projects. First and foremost, the implementation of additional physics (bound-free/thermalization processes) with the goal of adding modules that include more sophisticated ionization approximations and allow for spectral synthesis for SNe II (Klauser & Kromer et al., in prep.). Secondly, since was mainly developed to provide a means to fit with an approach similar to @2007Sci...315..825M, we aim to couple with a suitable algorithm for automatic fitting of observations. We have explored this problem already, using the ML93 spectral synthesis code and genetic algorithms as the optimization algorithm and find that this is a promising approach [see @2011PhDT.......324K priv. comm. S. Hachinger, P. Mazzali].
Acknowledgements
================
We gratefully acknowledge Markus Kromer and Michael Klauser for many useful discussions and suggestions during all stages in the development of . We thank Paolo Mazzali and Stephan Hachinger for helpful discussions of ionization/excitation treatments and formulation of the automatic spectral fitting project from which this work arose. We thank Knox Long for making available the code, the LK02 $\zeta$-values and for advice in making our comparisons with . We are grateful to Ken Dere for help with [ChiantiPy]{} and to Brian Schmidt for suggestions and support throughout the development of this project. We thank Adam Suban-Loewen for help in developing the user interface and Aoife Boyle for participating in the testing of the code. We also would like to thank Erik Bray for helping with the installation frameworks implemented in . We acknowledge the referee R. C. Thomas for very helpful suggestions and critical assessment of the paper. This research made use of Astropy, a community-developed core Python package for Astronomy , and of the <span style="font-variant:small-caps;">randomkit</span>[^5] library.
Using [tardis]{} {#sec:using_tardis}
================
At runtime requires (1) an atomic database and (2) a user specified configuration file.
The atomic database is supplied in HDF5[^6] format and consists of tables containing the data described in Section \[sec:datasources\]. A relatively simple atomic database is downloadable from the online manual (<http://tardis.rtfd.org> - in the “Running ” section). For custom versions of the atomic database users are currently encouraged to contact the authors.
The configuration file is supplied in YAML[^7] markup language and defines both the physical and numerical parameters of the calculation and selects which of the available modes of operation are to be used. A simple example of the input file is shown and explained in Section \[sec:example\_input\] (further details in the code manual - <http://tardis.rtfd.org>).
In the standard mode of operation, the major physical parameters that can be varied are the output luminosity, the time since explosion and the density/composition profile of the model (including the locations of the inner and outer boundaries). The density profile can be chosen from preset standard options \[e.g. constant density or a W7-like [@1984ApJ...286..644N] density, as in the example given below\] or it can be supplied via an addition ASCII input table (density tabulated as a function of velocity; see manual). The abundance distribution can be uniform (see example given below) or stratified (again using an ASCII input table; see manual).
Example input file {#sec:example_input}
------------------
 \[fig:tardis\_yaml\_input\]
Fig. \[fig:tardis\_yaml\_input\] shows a configuration file for . This file initiates a calculation equivalent to that used in Section \[sec:python\_compare\] (i.e. see Fig. \[fig:python\_compare\]) and runs successfully with version 0.9 of . The sections of the file are described in detail below (with reference to line numbers). We stress that this is only one simple example and refer users to the manual (<http://tardis.rtfd.org>) for further information.
All dimensional quantities in the configuration file require a unit \[these are parsed using the `Quantity` framework\].
The first line of the configuration file identifies the version. Currently only `v1.0` is supported but future versions of may involve a restructure of the configuration file and will recognise the layout according to the version number.
The section (`supernova`) defines the requested emergent luminosity ([$L_\textrm{o}$]{}; see Section \[sec:montecarlo\_iteration\]) and the time since explosion (). The standard units are available.
The atomic data file (HDF5 format) is specified at line 7 (absolute or relative path can be given).
The `model` section controls the setup of the radial density and abundance profiles. In the example, we specify 20 shells (line 17) equally spaced in the velocity range 11000 to 20000 (lines 15–16). We adopt a W7-like density profile (specifically, a polynomial fit to the W7 structure similar to @1985ApJ...294..619B; lines 19–20). The example calculation adopts uniform (line 23) composition of all zones by specifying mass fractions for six elements in lines 24–29. Abundances of elements not specified are set to zero. More complex radial stratification of density and/or composition can be specified (see user manual).
The settings to be used in determining the ionization/excitation/line-interactions modes are provided in the `plasma` section (lines 31–35) using the keywords described in Table \[tab:modes\].
In the `montecarlo` section, is supplied with the seed value for the random number generator (line 38), the requested number of packets to be used in each MC simulation (line 39) and the number of iterations to be performed in determining the plasma state (line 40). In addition (line 41), one can specify a different number of packets for the last MC simulation (making it possible to increase the signal-to-noise of the final spectrum). If the keyword `no_of_virtual_packets` is greater than zero (line 42), the virtual packet mode will be activated in the last MC simulation with $N_{v}$ set to the value given (see Section \[sec:synthetic\_spectrum\]).
In the last section (`spectrum`), the spectral range for the synthetic spectrum is set by specifying start and end wavelengths (lines 45 and 46). The spectral resolution is determined by choosing the total number of wavelength bins (line 47).
Running the model defined by the example configuration file on a mid 2012 MacBook Air (Intel Core i7 2.00 GHz, 8 GB RAM) takes approximately $\sim 3$ minutes to complete all 30 iterations. The runtime scales roughly linearly with the number of MC packets.
[^1]: We note, however, that can be used to compute a time series of synthetic spectra for a model via a sequence of calculations in which the input parameters specifying luminosity and time since explosion are varied.
[^2]: For clarity, we drop the explicit reference to atomic number and ionization number here.
[^3]: We specify direction of propagation by $\mu = \cos \theta$, where $\theta$ is the angle between the packet flight path and the radial direction.
[^4]: We note, however, that does not require such a complex input model (see Section \[sec:method\] and Appendix A). In many applications, simpler models will be more efficient for exploration of parameter space and attempting to fit observations.
[^5]:
[^6]:
[^7]:
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This article classifies the real forms of Lie Superalgebra by Vogan diagrams, developing Borel de Seibenthal theorem of semisimple Lie algebras for Lie superalgebras. A Vogan diagram is a Dynkin diagram of triplet $(\mathfrak{g}_{C},\mathfrak{h_{\overline{0}}},\triangle^{+})$, where $\mathfrak{g}_{C}$ is a real Lie superalgebra, $\mathfrak{h_{\overline{0}}}$ cartan subalgebra, $\triangle^{+}$ positive root system. Although the classification of real forms of contragradient Lie superalgebras already furnished but this article’s method is a quicker one to classify.'
author:
- 'B Ransingh$^{*}$ and K C Pati'
title: A quick proof of classification of real forms of Basic Lie superalgebras by Vogan diagrams
---
Department of Mathematics\
National Institute of Technology\
Rourkela (India)
Email$^{*}$- [email protected]
2010 AMS Subject Classification : 17B05, 17B22, 17B40\
Keywords : Lie superalgebras, Vogan diagrams
Introduction
============
For a complex semi-simple Lie algebra $\mathfrak{g}$, it is well known that the conjugacy classes of real forms of $\mathfrak{g}$ are in one to one correspondence with the conjugacy classes of involutions of $\mathfrak{g}$, if we associate to a real form $\mathfrak{g}_{\mathbb{R}}$ one of its Cartan involutions $\theta$. Using a suited pair $(\mathfrak{h}, \prod)$ of a Cartan subalgebra $\mathfrak{h}$ and a basis $\prod$ of the associated root system, an involution is described by a “Vogan diagram”.
Knapp [@Knapp:Lie; @groups] brought Vogan diagrams of simple Lie algebras into the light to represent the real forms of the complex simple Lie algebras. Batra [@batra:affine; @batra:vogan] developed a corresponding theory of Vogan diagrams for almost compact real forms of indecomposable nontwisted affine Kac-Moody Lie algebras. Tanushree [@tanushree:twisted] developed the theory of Vogan diagrams for almost compact real forms of indecomposable twisted affine Kac-Moody Lie algebras. Similar theory is also developed to find out the Vogan diagrams of hyperbolic Kac-Moody algebras[@ransingh:pati].
A Vogan diagram is a Dynkin diagram with some additional information as follows the 2-elements orbits under $\theta$ (Cartan involution) are exhibited by joining the corresponding simple roots by a double arrow and the 1-element orbit is painted in black (respectively, not painted), if the corresponding imaginary simple root is noncompact (respectively compact).
The real form is defined as a real Lie superalgebra such that its complexification is the original complex Lie superalgebra. It can be seen easily that every standard real form is naturally associated to an antilinear involutive automorphism of the complex Lie superalgebra.
The classification of real semisimple Lie algebras will use maximally compact and split Cartan subalgebras, The Vogan diagram is based on the classification of maximally compact Cartan subalgebras.
Recently similar work has been done using Vogan superdiagrams to classify the real forms of contragradient Lie superalgebras[@chuah:vsuper], where the extended Dynkin diagrams of Lie superalgebra is used. But our method uses the ordinary Dynkin diagram as done by Knapp[@Knapp:Lie; @groups]. In this article we construct all the real forms of Lie superalgebras by Vogan diagrams.
Preliminary
===========
The general and special linear Lie superalgebras.
-------------------------------------------------
Let $V=V_{\overline{0}}\oplus V_{\overline{1}}$ be a vector superspace, so that End$(V)$ is an associative superalgebra. The End$(V)$ with the supercummutator forms a Lie superalgebra, called the general linear Lie superalgebra and is denoted by $\mathfrak{gl}(m|n)$, where $V=\mathbb{C}^{m|n}$. With respect to an suitable ordered basis End$(V)$ and $\mathfrak{gl}(m|n)$ can be realized as $(m+n)\times(m+n)$ complex matrices of the block form.
$\left(\begin{array}{cc}
a & b\\
c & d
\end{array}\right)$ where $a$ , $b$, $c$ and $d$ are respectivily $m\times m$, $m\times n$, $n\times m$ and $n\times n$ matrices. The even subalgebra of $\mathfrak{gl}(m|n)$ is $\mathfrak{gl}(m)\oplus\mathfrak{gl}(n)$ , which consists of matrices of the form $\left(\begin{array}{cc}
a & 0\\
0 & d
\end{array}\right)$, While the odd subspace consists of $\left(\begin{array}{cc}
0 & b\\
c & 0
\end{array}\right)$
A Lie superalgebras $\mathfrak{g}$ is an algebra graded over $\mathbb{Z}_{2}$ , i.e., $\mathfrak{g}$ is a direct sum of vector spaces $\mathfrak{g}=\mathfrak{g}_{\overline{0}}\oplus \mathfrak{g}_{\overline{1}}$, and such that the bracket satisfies
1. $[\mathfrak{g}_{i},
\mathfrak{g}_{j}]\subset\mathfrak{g} _{i+j(mod2)}$,
2. $[x,y]=-(-1)^{|x||y|}[y,x]$, (Skew supersymmetry) $\forall$ homogenous $x,y,z\in \mathfrak{g}$ (Super Jacobi identity)
3. $[x,[y,z]]=[[x,y],z]+\left(-1\right)^{|x||y|}[y,[x,z]]\forall z\in \mathfrak{g}$
A bilinear form $(.,.):\mathfrak{g}\times\mathfrak{g}\rightarrow\mathbb{C}$ on a Lie superalgebra is called **invariant** if $([x,y],z)=(x,[y,z])$, for all $x,y,z\in \mathfrak{g}$
The Lie superalgebra $\mathfrak{g}$ has a root space decomposition with respect to $\mathfrak{h}$ $$\mathfrak{g}=\mathfrak{h}\oplus\bigoplus_{\alpha\in\triangle}\mathfrak{g}_\alpha$$
A root $\alpha$ is even if $\mathfrak{g}_{\alpha}\subset\mathfrak{g}_{\overline{0}}$ and it is odd if $\mathfrak{g}_{\alpha}\subset\mathfrak{g}_{\overline{1}}$
A *Cartan subalgebra* $\mathfrak{h}$ of diagonal matrices of $\mathfrak{g}$ is defined to be a Cartan subalgebra of the even subalgebra $\mathfrak{g}_{\overline{0}}$. Since every inner automorphism of $\mathfrak{g}_{\overline{0}}$ extends to one of Lie superalgebra $\mathfrak{g}$ and Cartan subalgebras of $\mathfrak{g}_{\overline{0}}$ are conjugate under inner automorphisms. So the Cartan subalgebras of $\mathfrak{g}$ are conjugate under inner automorphism.
Real forms of Basic Lie superalgebras
--------------------------------------
\[[@Parker:classification] Proposition 1.4\] Let $\mathcal{\mathfrak{g}}$ be a complex classical Lie superalgebra and let $C$ be an involutive semimorphism of $\mathcal{\mathfrak{g}}$. Then $\mathcal{\mathfrak{g}}_{C}=\left\{ x+Cx|x\in\mathfrak{g}\right\} $ is a real classical Lie superalgebra.
\[[@Parker:classification] Proposition 1.5\] If $\mathcal{\mathfrak{g}_C}$ is a real classical Lie superalgebra, its complexification $\mathfrak{g}=\mathfrak{g}_{C}\otimes\mathbb{C}$ is a Lie superalgebra which is either classical or direct sum of two isomorphic ideals which are classical
\[[@Parker:classification] Theorem 4\] Up to isomorphism, the real forms of the classical Lie superalgebras are uniquely determined by the real form $\mathcal{\mathfrak{g}}_{\overline{0}C}$ of the Lie subalgebra $\mathcal{\mathfrak{g}}_{\overline{0}}$.
The real form is said to standard (graded) when the real structure is standard (graded). Let $\mathfrak{g}_{C}$ be a real form of $\mathfrak{g}$ and let $\omega$ be the corresponding complex conjugation. Then $\omega|_{\mathfrak{g}_{\overline{0}}}$ is an antilinear involution of the Lie algebra $\mathfrak{g}_{\overline{0}}$. Hence there is a corresponding Cartan decomposition $\mathfrak{g}_{\overline{0}}=t_{\overline{0}}\oplus p_{\overline{0}}$, with Cartan involution $C$.
The following table gives a list of all the real forms associated with basic classical Lie superalgebras.
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
$\mathfrak{g}$ $\mathfrak{g}_{\overline{0}}$ $\mathfrak{g}_C$ $\mathfrak{g}_{\overline{0}C}$
------------------- ------------------------------------------------------------------ ------------------------------------------ ---------------------------------------------------------------------------------------------------
$A(m,n)$ $\mathfrak{sl}(m)\oplus \mathfrak{sl}(n)\oplus U(1)$ $\begin{array}{l} $\begin{array}{l}
\mathfrak{sl}(m|n;\mathbb{R})\\ \mathfrak{sl}(m,\mathbb{R})\oplus \mathfrak{sl}(n,\mathbb{R})\oplus\mathbb{R}\\
\mathfrak{sl}(m|n;\mathbb{H})\\ \mathfrak{su}^{*}(m)\oplus \mathfrak{su}^{*}(n)\oplus\mathbb{R}\\
\mathfrak{su}(p,m-p|q,n-q) \mathfrak{su}(p,m-p)\oplus \mathfrak{su}(q,n-q)\oplus i\mathbb{R}
\end{array}$ \end{array}$
$A(n,n)$ $\mathfrak{sl}(n)\oplus \mathfrak{sl}(n)$ $\begin{array}{l} $\begin{array}{l}
\mathfrak{psl}(n|n;\mathbb{R})\\ \mathfrak{sl}(n,\mathbb{R})\oplus \mathfrak{sl}(n,\mathbb{R})\\
\mathfrak{psl}(n|n;\mathbb{H})\\ \mathfrak{su}^{*}(n)\oplus \mathfrak{su}^{*}(n)\\
\mathfrak{su}(p.n-p|q,n-q) \mathfrak{su}(p,n-p)\oplus \mathfrak{su}(q,n-q)
\end{array}$ \end{array}$
$\begin{array}{l} $\begin{array}{l} $\begin{array}{l} $\begin{array}{l}
B(m,n)\\ \mathfrak{so}(2m+1)\oplus \mathfrak{sp}(2n)\\ \mathfrak{osp}(p,2m+1-p|2n;\mathbb{R})\\ \mathfrak{so}(p,2m+1-p)\oplus \mathfrak{sp}(2n;\mathbb{R})\\
B(0,n) \mathfrak{sp}(2n) \mathfrak{osp}(1|2n;\mathbb{R}) \mathfrak{sp}(2n,\mathbb{R})
\end{array}$ \end{array}$ \end{array}$ \end{array}$
$C(n+1)$ $\mathfrak{so}(2)\oplus \mathfrak{sp}(2n)$ $\begin{array}{l} $\begin{array}{l}
\mathfrak{osp}(p|2n;\mathbb{R})\\ \mathfrak{so}^{*}(2)\oplus \mathfrak{sp}(2n;\mathbb{R})\\
\mathfrak{osp}(2|2q,2n-2q;\mathbb{H}) \mathfrak{so}^{*}(2)\oplus \mathfrak{sp}(2q,2n-2q)
\end{array}$ \end{array}$
$D(m,n)$ $\mathfrak{so}(2m)\oplus \mathfrak{sp}(2n)$ $\begin{array}{l} $\begin{array}{l}
\mathfrak{osp}(p,2m-p|2n;\mathbb{R})\\ \mathfrak{so}(p,2m-p)\oplus \mathfrak{sp}(2n;\mathbb{R})\\
\mathfrak{osp}(2m|2q,2n-2q;\mathbb{H}) \mathfrak{so}^{*}(2m)\oplus \mathfrak{sp}(2q,2n-2q)\\
\end{array}$ \mathfrak{sp}(n)\oplus \mathfrak{so}^{*}(2m)\\
\mathfrak{so}(2m)\oplus \mathfrak{so}^{*}(2n,\mathbb{R})
\end{array}$
$F(4)$ $\mathfrak{sl}(2)\oplus \mathfrak{so}(7)$ $\begin{array}{l} $\begin{array}{l}
F(4;0)\\ \mathfrak{sl}(2,\mathbb{R})\oplus \mathfrak{so}(7)\\
F(4;3)\\ \mathfrak{sl}(2,\mathbb{R})\oplus \mathfrak{so}(1,6)\\
F(4;2)\\ \mathfrak{sl}(2,\mathbb{R})\oplus \mathfrak{so}(2,5)\\
F(4;1) \mathfrak{sl}(2,\mathbb{R})\oplus \mathfrak{so}(3,4)
\end{array}$ \end{array}$
$G(3)$ $\mathfrak{sl}(2)\oplus G_{2}$ $\begin{array}{l} $\begin{array}{l}
G(3,0)\\ \mathfrak{sl}(2,\mathbb{R})\oplus G_{2,0}\\
G(3,1) \mathfrak{sl}(2,\mathbb{R})\oplus G_{2,2}
\end{array}$ \end{array}$
$D(2,1;\alpha)$ $\mathfrak{sl}(2)\oplus \mathfrak{sl}(2)\oplus \mathfrak{sl}(2)$ $\begin{array}{l} $\begin{array}{l}
D(2,1;\alpha;0)\\ \mathfrak{sl}(2,\mathbb{R})\oplus \mathfrak{sl}(2,\mathbb{R})\oplus \mathfrak{sl}(2,\mathbb{R})\\
D(2,1;\alpha;1)\\ \mathfrak{sl}(u)\oplus \mathfrak{sl}(u)\oplus \mathfrak{sl}(u)\\
D(2,1;\alpha;2) \mathfrak{sl}(2,\mathbb{C})\oplus \mathfrak{sl}(2,\mathbb{R})
\end{array}$ \end{array}$
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Vogan diagrams of Basic Lie Superalgebras
=========================================
The following Lemma shows the existence of
\[Lemma 4.7 [@kac:denominator]\]There exists $C\in Aut(\mathfrak{g})$ such that $C|_{g_{0}}$ is a nontrivial automorphism if and only if $\mathfrak{g}$ is of type $A(m,n)$, $D(m,n)$, $D(2,1;\alpha)$ ,$\alpha\in\{1,(-2)^{\pm 1}\}$ and $C|_{g_{0}}$is as follows.
1. If $\mathfrak{g}$ is of type $A(m,n)$ with $n\ne m$, then $C|_{\mathfrak{g}_0}$ restricts to the nontrivial diagram automorphism of both $A_n$ and $A_m$.
2. If $\mathfrak{g}$ is of type $A(n,n)$, then $C|_{\mathfrak{g}_{0}}$ is either the nontrivial diagram automorphism of both $A_n$ components, or it is the flip automorphism between the two $A_n$ components, or the composition of these two automorphisms.
3. If $\mathfrak{g}$ is of type $D(m,n),\,m>2$, then $C|_{\mathfrak{g}_0}$ is the unique diagram automorphism of $\mathfrak{g}_0$.
4. If $\mathfrak{g}$ is of type $D(2,1)\cong D(2,1,a)$, $a\in{1, (-2)^{\pm1} }$, then $C$ is the unique diagram automorphism of the diagram.
The Vogan diagram of Lie superalgebras is the Vogan diagram of the even part of Lie superalgerbas. In addition to that
- The vertices fixed by the Cartan involution of the even part is painted (or unpainted) depending whether the the root is noncompact (or compact).
- Label the 2- elements orbit by the diagram automorphism indicated with two sided arrow.
- The odd root remain unchanged.
An abstract Vogan diagram is an abstract Dynkin diagram with two pieces of additional structure , one is an automorphism of order 1 or 2 of the diagram, which is to be indicated by labeling the 2 element orbits. The other is the subset of the 1 element orbits which is to be indicated by painting the vertices corresponding to the members of the subset of noncompact roots. Every Vogan diagram is of course an abstract Vogan diagram of Lie superalgebra.
If an abstract Vogan diagram is given, then there exist a real Lie superalgebra $\mathfrak{g}_{C}$, a Cartan involution $\theta$, a maximally compact $\theta$ stable Cartan subalgebra and a positive system $\triangle_{\overline{0}}^{+}$ for $\triangle=\triangle(\mathfrak{g,h})$ that takes $it_{\overline{0}}$ before $a_{\overline{0}}$ such that the given diagram is the Vogan diagram of $(\mathfrak{g}_{C},\mathfrak{h}_{\overline{0}},\triangle_{0}^{+})$.
Briefly the theorem says that any abstract Vogan diagram comes from some $\mathcal{\mathfrak{g}}_{\overline{0}}$. Thus the theorem is an analog for real semisimple Lie algebras of the existence theorem for complex semisimple Lie algebras.
We will modify the Borel and de Siebenthal Theorem for Lie superalgebra.
Let $\mathfrak{g}_{C}$ be a non complex real Lie superalgebra and Let the Vogan diagram of $\mathfrak{g}_{C}$ be given that correponding to the triple $(\mathfrak{g}_{C},\mathfrak{h}_{0},\triangle^{+})$. Then $\exists$ a simple system $\prod'$ for $\triangle=\triangle(\mathcal{\mathfrak{g}},\mathfrak{h})$, with corresponding positive system $\triangle^{+}$, such that $(\mathcal{\mathfrak{g}}_
{C},\mathfrak{h}_{\overline{0}},\triangle^{+})$ is a triple and there is at most two painted simple root in its Vogan diagrams of $A(m,n),D(m,n)$ and at most three painted vertices in $D(2,1;\alpha)$. Furthermore suppose the automorphism associated with the Vogan diagram is the identity,that $\prod'={\alpha_{1},\cdots,\alpha_{l}}$ and that ${\omega_{1},\cdots,\omega_{l}}$ is the dual basis for each even part such that $\left\langle\omega_{j},\alpha_{k}\right\rangle=\delta_{jk}/\epsilon_{kk} $, where $\epsilon_{kk}$ is the diagonal entries to make cartan matrix symmetric. The the double painted simple root of even parts may be chosen so that there is no $i'$ with $\left\langle \omega_{i}-\omega_{i'},\omega_{i'}\right\rangle>0$ for each even part.
We know $\mathfrak{g}=\mathcal{\mathfrak{g}}_{\bar{0}}\oplus\mathcal{\mathfrak{g}}_{\bar{1}}$. The positive even root system $\triangle_{0}^{+}$ can be written as $$\triangle_{0}^{+}=\triangle_{01}^{+}\cup\triangle_{02}^{+}$$ where $\triangle_{01}^{+}$ are the even positive root system for simple root system formed by $e_{i}$ basis and $\triangle_{02}^{+}$ are for $\delta_{j}$ basis. For the even part, we take $<\omega_{i},\alpha_{j}>=\delta_{ij}/\epsilon_{kk}$ . This makes the Cartan matrix symmetric and so that we can get the inverse of cartan matrix of $A_{m}$ and $A_n$ for $A(m,n)$ Lie superalgebra. Similarly construction will follows for other Lie superalgebras. Each inverse for even part is associted with the dual basis $\omega$. For the odd part the condition is $<\omega_{i},\alpha_{j}>=\delta_{ij}$ and donot get any painted vertices. The Symmetrizable condition of Kac-Cartan matrix gives $S=\epsilon_{kk} A$, where $S$ is the symmetric cartan matrix. The below table gives the values of $\epsilon_{kk}$ for different superalgebras.
Lie superalgebra $\epsilon_{kk}$
------------------ -------------------------------------
$A(m,n)$ $(1,\cdots,1,-1,\cdots,-1)$
$B(m,n)$ $(1,\cdots,1,-1\cdots,-1,-2)$
$B(0,n)$ $1,\cdots,1,2$
$C(n)$ $(-1,1,\cdots,1,\frac{1}{2})$
$D(m,n)$ $(1,\cdots,1,-1-1,\cdots,-1,-1,-1)$
$D(2,1;\alpha)$ $(1,-1,\frac{1}{a})$
$F(4)$ $ (-1,1,\frac{1}{2}) $
$G(3)$ $(-\frac{1}{2},1,\frac{1}{3},)$
Taking suitable normalization condition for each type of Lie superalgebras and from the two Lemmas 6.97 and 6.98 [@Knapp:Lie; @groups] we get redudancy test for each even part. So now the Vogan diagram of Lie superalgebras becomes two painted vertices Vogan diagram.
Let $\bigtriangleup$ be an irreducible abstract reduced root system in a real vector superspace $V$, let $\Pi_{01}$ and $\Pi_{02}$ be the two simple simple root system for even parts $e_{i}$ and $\delta_{j}$ basis respectivily and let $\omega$ and $\omega'$ be nonzero members of $V$ that are domiant relative to $\Pi_{i}'$s. Then $\left\langle \omega,\omega'\right\rangle >0$.
Using the suitable normalisations of $e_i$ and $\delta_j$ we get the proof of the Lemma.
Let $\mathcal{\mathfrak{g}}_{0C}$ be a noncomplex simple real Lie superalgebra and let the Vogan diagram of $\mathcal{\mathfrak{g}}_{0C}$ be given that corresponding to the triple $(\mathfrak{g}_{0},\mathfrak{h}_{0},\triangle^{+})$. Write $\mathfrak{h}_{01}=\mathfrak{t}_{01}\oplus\mathfrak{a}_{01}$ and $\mathfrak{h}_{02}=\mathfrak{t}_{02}\oplus\mathfrak{a}_{02}$ for two even parts. . Let $V$ be the span of simple roots that are imaginary, let $\triangle_{0}$ be the root system $\triangle\cap V$, let $\mathcal{H}$ be the subset of $it_{0}$ paired with $V$ and let $\Lambda$ be the subset of $\mathcal{H}$ where all roots of $\triangle_{0}$ take integer values and all noncompact roots of $\triangle_{0}$ take odd integer values. Then $\Lambda$ is nonempty. In fact if $\alpha_{1},\cdots,\alpha_{m}$ is any simple system for $\triangle_{0}$ and if $\omega_{1},\cdots,\omega_{m}$ in $V$ are defined by $\left\langle \omega,\alpha_{k}\right\rangle =\delta_{jk}$, then the element
$$\omega=\underset{i\mbox{ with }\alpha_{i}\mbox{ noncompact}}{\sum}\omega_{i}$$
Fix a simple system $\alpha_{1},\cdots,\alpha_{m}$ for $\bigtriangleup_{\overline{0}}$ and let $\bigtriangleup^{+}_{\overline{0}}$ be the set of positive roots of $\bigtriangleup_{\overline{0}}$. Define $\omega_{1},\cdots,\omega_{m}$ by $\left\langle \omega,\alpha_{k}\right\rangle =\delta_{jk}$. If $\alpha=\sum_{i=1}^{m}
n_{i}\alpha_{i}$ is a positive root of $\bigtriangleup_{\overline{0}}$, then $\left\langle \omega,\alpha\right\rangle $ is the sum of the $n_{i}$ for which $\alpha_{i}$ is noncompact.
Using induction of the Lemma 6.98 [@Knapp:Lie; @groups] for even part of root system the above Lemma will be proved and each even roots satisfy
compact root + compact root = compact root
compact root+ noncompact root = noncompact root
noncompact root + noncompact root = noncompact root
1. $A(m,n)$
The Vogan diagrams and real foms of Lie superalgebras $A(m,n)$ are as follows.
$$\begin{picture}(60,10) \thicklines
\put(-125,0){\circle{9}} \put(-97,0){\circle{9}}\put(-70,0){\circle{9}} \put(-43,0){\circle{9}} \put(-20,-2.5){$\bigotimes$}\put(12,0){\circle{9}}
\put(39,0){\circle{9}}\put(66,0){\circle{9}}\put(93,0){\circle{9}}
\put(-120,0){\line(1,0){19}} \put(-92,0){\dottedline{4}(1,0)(17,0)}\put(-65,0){\line(1,0){17}}\put(-38,0){\line(1,0){19}}
\put(-10,0){\line(1,0){17}}
\put(16,0){\line(1,0){18}} \put(43,0){\dottedline{4}(1,0)(18,0)}\put(70,0){\line(1,0){19}}
\put(-15,-25){\makebox(0,0){$\mathfrak{sl}(n,\mathbb{C})$}}
\end{picture}$$
$$\begin{picture}(60,10) \thicklines
\put(-125,0){\circle{9}} \put(-97,0){\circle*{9}}\put(-70,0){\circle{9}} \put(-43,0){\circle{9}} \put(-20,-2.5){$\bigotimes$}\put(12,0){\circle{9}}
\put(39,0){\circle{9}}\put(66,0){\circle*{9}}\put(93,0){\circle{9}}
\put(-120,0){\dottedline{4}(1,0)(19,0)} \put(-92,0){\dottedline{4}(1,0)(17,0)}\put(-65,0){\line(1,0){17}}\put(-38,0){\line(1,0){19}}
\put(-10,0){\line(1,0){17}}
\put(16,0){\line(1,0){18}} \put(43,0){\dottedline{4}(1,0)(18,0)}\put(70,0){\dottedline{4}(1,0)(19,0)}
\put(-65,-25){\makebox(0,0){$\mathfrak{su}(p,m-p)$}}\put(65,-25){\makebox(0,0){$\mathfrak{su}(r,n-r)$}}
\end{picture}$$
$$\begin{picture}(60,10) \thicklines
\put(-125,0){\circle{9}} \put(-97,0){\circle{9}}\put(-70,0){\circle{9}} \put(-43,0){\circle{9}}
\put(-20,-2.5){$\bigotimes$}\put(12,0){\circle{9}}
\put(39,0){\circle{9}}\put(66,0){\circle{9}}\put(93,0){\circle{9}}
\put(-120,0){\line(1,0){19}} \put(-92,0){\dottedline{4}(1,0)(17,0)}\put(-65,0){\line(1,0){17}}\put(-38,0){\line(1,0){19}}
\put(-10,0){\line(1,0){17}}
\put(16,0){\line(1,0){18}} \put(43,0){\dottedline{4}(1,0)(18,0)}\put(70,0){\line(1,0){19}}
\qbezier(-97,-6)(-81.5,-14)(-69, -6)
\put(-97,-6){\vector( -2, 1){0}}\put(-69,-6){\vector( 2, 1){0}}
\qbezier(-125,-6)(-81,-24)(-43, -6)
\put(-125,-6){\vector( -2, 1){0}}\put(-43,-6){\vector( 2, 1){0}}
\qbezier(39,-6)(52.5,-14)(66, -6)
\put(39,-6){\vector( -2, 1){0}}\put(66,-6){\vector( 2, 1){0}}
\qbezier(12,-6)(52.5,-24)(93, -6)
\put(12,-6){\vector( -2, 1){0}}\put(93,-6){\vector( 2, 1){0}}
\put(-75,-25){\makebox(0,0){$\mathfrak{sl}(m,\mathbb{R})$}}\put(55,-25){\makebox(0,0){$\mathfrak{sl}(n,\mathbb{R})$}}
\end{picture}$$
$$\begin{picture}(60,10) \thicklines
\put(-152,0){\circle{9}}\put(-125,0){\circle{9}} \put(-97,0){\circle*{9}}\put(-70,0){\circle{9}} \put(-43,0){\circle{9}}
\put(-20,-2.5){$\bigotimes$}\put(12,0){\circle{9}}
\put(39,0){\circle{9}}\put(66,0){\circle*{9}}\put(93,0){\circle{9}}\put(120,0){\circle{9}}
\put(-147,0){\line(1,0){18}}\put(-120,0){\dottedline{4}(1,0)(19,0)} \put(-92,0){\dottedline{4}(1,0)(17,0)}
\put(-65,0){\line(1,0){17}}\put(-38,0){\line(1,0){19}}
\put(-10,0){\line(1,0){17}}
\put(16,0){\line(1,0){18}} \put(43,0){\dottedline{4}(1,0)(18,0)}\put(70,0){\dottedline{4}(1,0)(19,0)} \put(98,0){\line(1,0){18}}
\qbezier(-125,-6)(-97.5,-16)(-70, -6)
\put(-125,-6){\vector( -3, 1){0}}\put(-70,-6){\vector( 3, 1){0}}
\qbezier(-152,-6)(-97.5,-32)(-43, -6)
\put(-152,-6){\vector( -2, 1){0}}\put(-43,-6){\vector(2, 1){0}}
\qbezier(39,-6)(66,-16)(93, -6)
\put(39,-6){\vector( -3, 1){0}}\put(93,-6){\vector( 3, 1){0}}
\qbezier(12,-6)(66,-32)(120, -6)
\put(12,-6){\vector( -2, 1){0}}\put(120,-6){\vector(2, 1){0}}
\put(-95,-28){\makebox(0,0){$\mathfrak{su}^{*}(m)$}}\put(65,-28){\makebox(0,0){$\mathfrak{su}^{*}(n)$}}
\end{picture}$$
2. $B(m,n)$ The Lie subalgebra of $\mathfrak{g}_{0}$ is $C_{m}\oplus B_{n}$ The only trivial automophism of even part of Vogan diagram of $B(m,n)$ is shown below and the real form is $\mathfrak{sp}(2n,\mathbb{R})\oplus \mathfrak{so}(p,2m+1-p)$
$$\begin{picture}(60,10) \thicklines
\put(-97,0){\circle{9}}\put(-70,0){\circle{9}} \put(-43,0){\circle{9}} \put(-20,-2.5){$\bigotimes$}\put(12,0){\circle*{9}}
\put(39,0){\circle{9}}\put(66,0){\circle{9}}
\put(-92,0){\line(1,0){19}}\put(-65,0){\dottedline{4}(1,0)(17,0)}\put(-38,0){\line(1,0){19}}\put(-11,0){\dottedline{4}(1,0)(18,0)}
\put(16,0){\dottedline{4}(1,0)(18,0)}\put(37,0){{\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(20,0){\line(-1,1){10}}\put(20,0){\line(-1,-1){10}}
\end{picture}}}
\put(45,-25){\makebox(0,0){$so(p,2m+1-p)$}}
\end{picture}$$
Because of missing of real form of the first even part, we need and additional $C_{n}$ Dynkin diagram superimposed Vogan diagrams below. $$\begin{picture}(60,10) \thicklines
\put(-97,0){\circle{9}}\put(-70,0){\circle{9}} \put(-43,0){\circle{9}} \put(-20,-2.5){$\bigotimes$}\put(12,0){\circle*{9}}
\put(39,0){\circle{9}}\put(66,0){\circle{9}}
\put(-65,0){\dottedline{4}(1,0)(17,0)}\put(-38,0){\line(1,0){19}}\put(-11,0){\dottedline{4}(1,0)(18,0)}
\put(16,0){\dottedline{4}(1,0)(18,0)}\put(37,0){{\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(20,0){\line(-1,1){10}}\put(20,0){\line(-1,-1){10}}
\end{picture}}}
\put(-65,-25){\makebox(0,0){$sp(2n,\mathbb{R})$}}\put(45,-25){\makebox(0,0){$so(p,2m+1-p)$}}
\put(-100,0){\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(20,0){\line(-1,1){10}}\put(20,0){\line(-1,-1){10}}
\end{picture}}\end{picture}$$
The another real form of Lie superalgebra $B(m,n)$ is $sp(2n,\mathbb{R})\oplus so(2m+1)$ and the corresponding Vogan diagram is drawn below. $$\begin{picture}(60,10) \thicklines
\put(-97,0){\circle{9}}\put(-70,0){\circle{9}} \put(-43,0){\circle{9}} \put(-20,-2.5){$\bigotimes$}\put(12,0){\circle{9}}
\put(39,0){\circle{9}}\put(66,0){\circle{9}}
\put(-65,0){\dottedline{4}(1,0)(17,0)}\put(-38,0){\line(1,0){19}}\put(-11,0){\dottedline{4}(1,0)(18,0)}
\put(16,0){\dottedline{4}(1,0)(18,0)}\put(37,0){{\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(20,0){\line(-1,1){10}}\put(20,0){\line(-1,-1){10}}
\end{picture}}}
\put(-65,-25){\makebox(0,0){$sp(2n,\mathbb{R})$}}\put(45,-25){\makebox(0,0){$so(2m+1)$}}
\put(-100,0){\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(20,0){\line(-1,1){10}}\put(20,0){\line(-1,-1){10}}
\end{picture}}\end{picture}$$
3. Case $B(0,n)$
The Vogan diagram below is a unpainted diagram but it consists of its own painted vertices on the extreme right.\
$$\begin{picture}(60,20) \thicklines
\put(-43,0){\circle{9}} \put(-15,0){\circle{9}}\put(12,0){\circle{9}} \put(39,0){\circle{9}}\put(66,0){\circle*{9}}
\put(-38,0){\line(1,0){19}}\put(-12,0){\dottedline{4}(1,0)(19,0)}\put(16,0){\line(1,0){19}}\put(37,0){{\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(20,0){\line(-1,1){10}}\put(20,0){\line(-1,-1){10}}
\end{picture}}}
\put(25,-25){\makebox(0,0){$sp(2n,\mathbb{R})$}}
\end{picture}$$
4. Case $C(n+1)$
The unpainted Vogan diagram of $C(n+1)$ creats the real form $so^{*}(2)\oplus sp(2n,\mathbb{R})$\
$$\begin{picture}(60,20) \thicklines
\put(-48,-3){$\bigotimes$} \put(-15,0){\circle{9}}\put(12,0){\circle{9}} \put(39,0){\circle{9}}\put(66,0){\circle{9}}
\put(-38,0){\line(1,0){19}}\put(-12,0){\dottedline{4}(1,0)(19,0)}\put(16,0){\dottedline{4}(1,0)(19,0)}\put(37,0){{\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(15,0){\line(1,-1){10}}\put(15,0){\line(1,1){10}}
\end{picture}}}
\end{picture}$$
The trivial automophism of the even part of $C(n+1)$ makes the Vogan diagram below and the real form is $so^{*}(2)\oplus sp(2q,2n-2q)$\
$$\begin{picture}(60,20) \thicklines
\put(-48,-3){$\bigotimes$} \put(-15,0){\circle{9}}\put(12,0){\circle*{9}} \put(39,0){\circle{9}}\put(66,0){\circle{9}}
\put(-38,0){\line(1,0){19}}\put(-12,0){\dottedline{4}(1,0)(19,0)}\put(16,0){\dottedline{4}(1,0)(19,0)}\put(37,0){{\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(15,0){\line(1,-1){10}}\put(15,0){\line(1,1){10}}
\end{picture}}}
\end{picture}$$
5. Case $D(m,n)$. The Lie subalgebra of $\mathfrak{g}_{0}$ is $C_{m}\oplus D_{n}$ . The compact real form of $C_{m}$ is $\mathfrak{sp}(m)$. The real form $sp(2n,\mathbb{R})\oplus so(2m)$ of the abstract Vogan diagram of $D(m,n)$ for the above subalgebras is followed below.
$$\begin{picture}(-20,20)\thicklines
\put(-137,0){\circle{9}}\put(-109,0){\circle{9}}\put(-81,0){\circle{9}}\put(-56,-2.5){$\bigotimes$} \put(-22,0){\circle{9}}
\put(5,0){\circle{9}}\put(30,16){\circle{9}} \put(30,-16){\circle{9}}
\put(-132,0){\dottedline{4}(1,0)(19,0)} \put(-105,0){\dottedline{4}(1,0)(19,0)}
\put(-76,0){\line(1,0){20}} \put(-46,0){\dottedline{4}(1,0)(19,0)}\put(-18,0){\dottedline{4}(1,0)(19,0)} \put(10,2){\line(3,2){16}}
\put(10,-2){\line(3,-2){16}}
\put(5,-25){\makebox(0,0){$\mathfrak{so}(2m)$}}
\end{picture}$$
Since from the diagram we get only $\mathfrak{so}(2m)$ part of real form, so for the $\mathfrak{sp}(2n)$ part we need the addition $C_{n}$ diagram in the diagram above. Subsequently the Vogan diagram becomes
$$\begin{picture}(-20,20)\thicklines
\put(-137,0){\circle{9}}\put(-109,0){\circle{9}}\put(-81,0){\circle{9}}\put(-56,-2.5){$\bigotimes$} \put(-22,0){\circle{9}}
\put(5,0){\circle{9}}\put(30,16){\circle{9}} \put(30,-16){\circle{9}}
\put(-105,0){\dottedline{4}(1,0)(19,0)}
\put(-76,0){\line(1,0){20}} \put(-46,0){\dottedline{4}(1,0)(19,0)}\put(-18,0){\dottedline{4}(1,0)(19,0)}
\put(10,2){\line(3,2){16}}\put(10,-2){\line(3,-2){16}}
\put(-105,-25){\makebox(0,0){$\mathfrak{sp}(2n,\mathbb{R})$}}\put(5,-25){\makebox(0,0){$\mathfrak{so}(2m)$}}
\put(-139,0){\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(20,0){\line(-1,1){10}}\put(20,0){\line(-1,-1){10}}
\end{picture}}\end{picture}$$
The first trivial involution for one of the even part for the Vogan diagram of $D(m,n)$ is given below and the real form of this diagram is $\mathfrak{sp}(2n,\mathbb{R})\oplus \mathfrak{so}(p,2m-p)$.
$$\begin{picture}(-20,20)\thicklines
\put(-137,0){\circle{9}}\put(-109,0){\circle{9}}\put(-81,0){\circle{9}}\put(-56,-2.5){$\bigotimes$} \put(-22,0){\circle*{9}}
\put(5,0){\circle{9}}\put(30,16){\circle{9}} \put(30,-16){\circle{9}}
\put(-105,0){\dottedline{4}(1,0)(19,0)}
\put(-76,0){\line(1,0){20}} \put(-46,0){\dottedline{4}(1,0)(19,0)}\put(-18,0){\dottedline{4}(1,0)(19,0)}
\put(10,2){\line(3,2){16}}\put(10,-2){\line(3,-2){16}}
\put(-105,-25){\makebox(0,0){$\mathfrak{sp}(2n,\mathbb{R})$}}\put(5,-25){\makebox(0,0){$\mathfrak{so}(p,2m-p)$}}
\put(-139,0){\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(20,0){\line(-1,1){10}}\put(20,0){\line(-1,-1){10}}
\end{picture}}\end{picture}$$
The nontrivial involution for the Vogan diagram of $D(m,n)$ is given below and the real form of this diagram is $sp(2q,2n-2q)\oplus so^{*}(2m)$.
$$\begin{picture}(-20,20)\thicklines
\put(-165,0){\circle{9}}
\put(-137,0){\circle{9}} \put(-137,0){\circle{9}}\put(-109,0){\circle*{9}}\put(-81,0){\circle{9}}\put(-56,-2.5){$\bigotimes$} \put(-22,0){\circle{9}}
\put(5,0){\circle{9}}\put(30,16){\circle{9}} \put(30,-16){\circle{9}}
\put(-132,0){\dottedline{4}(1,0)(19,0)} \put(-105,0){\dottedline{4}(1,0)(19,0)}
\put(-76,0){\line(1,0){20}} \put(-46,0){\dottedline{4}(1,0)(19,0)}\put(-18,0){\dottedline{4}(1,0)(19,0)} \put(10,2){\line(3,2){16}}\put(10,-2){\line(3,-2){16}}
\qbezier(35,16)(51,1)(35, -16)
\put(35,16){\vector( -1, 1){0}}\put(35,-16){\vector( -1, -1){0}}
\put(-105,-25){\makebox(0,0){$\mathfrak{sp}(2q,2n-2q)$}}\put(5,-25){\makebox(0,0){$\mathfrak{so}^{*}(2m)$}}
\put(-167,0){\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(20,0){\line(-1,1){10}}\put(20,0){\line(-1,-1){10}}
\end{picture}}\end{picture}$$
The below Vogan diagram is formed by the nontrivial involution . The real form for this superlagebra is $\mathfrak{sp}(n)\oplus \mathfrak{so}^{*}(2m)$.
$$\begin{picture}(-20,20)\thicklines
\put(-137,0){\circle{9}}\put(-109,0){\circle{9}}\put(-81,0){\circle{9}}\put(-56,-2.5){$\bigotimes$} \put(-22,0){\circle{9}}
\put(5,0){\circle{9}}\put(30,16){\circle{9}} \put(30,-16){\circle{9}}
\put(-105,0){\dottedline{4}(1,0)(19,0)}
\put(-76,0){\line(1,0){20}} \put(-46,0){\dottedline{4}(1,0)(19,0)}\put(-18,0){\dottedline{4}(1,0)(19,0)} \put(10,2){\line(3,2){16}}\put(10,-2){\line(3,-2){16}}
\qbezier(35,16)(51,1)(35, -16)
\put(35,16){\vector( -1, 1){0}}\put(35,-16){\vector( -1, -1){0}}
\put(-105,-25){\makebox(0,0){$\mathfrak{sp}(n)$}}\put(5,-25){\makebox(0,0){$\mathfrak{so}^{*}(2m)$}}
\put(-139,0){\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(20,0){\line(-1,1){10}}\put(20,0){\line(-1,-1){10}}
\end{picture}}\end{picture}$$
6. Case $D(2,1;\alpha)$
The unpainted and no two element orbit Vogan diagram is given below with the real form. Since we can get only the real form $\mathfrak{su}(2)\oplus \mathfrak{su}(2)$ from ordinary Vogan diagram $$\begin{picture}(40,20)\thicklines
\put(0,-2.5){$\bigotimes$}\put(30,16){\circle{9}} \put(30,-16){\circle{9}}
\put(10,2){\line(3,2){16}}\put(10,-2){\line(3,-2){16}}
\put(100,0){\makebox(0,0){$\mathfrak{su}(2)\oplus \mathfrak{su}(2)$}}
\end{picture}$$
So our requisite Vogan diagrams for the suitable real forms are $$\begin{picture}(40,20)\thicklines
\put(-22,0){\circle{9}} \put(0,-2.5){$\bigotimes$}\put(30,16){\circle{9}} \put(30,-16){\circle{9}}
\put(-18,0){\line(1,0){19}} \put(10,2){\line(3,2){16}}\put(10,-2){\line(3,-2){16}}
\put(100,0){\makebox(0,0){$\mathfrak{su}(2)\oplus \mathfrak{su}(2)\oplus \mathfrak{su}(2)$}}
\end{picture}$$ $$\begin{picture}(40,20)\thicklines
\put(-22,0){\circle*{9}} \put(0,-2.5){$\bigotimes$}\put(30,16){\circle*{9}} \put(30,-16){\circle*{9}}
\put(-18,0){\line(1,0){19}} \put(10,2){\line(3,2){16}}\put(10,-2){\line(3,-2){16}}
\put(105,0){\makebox(0,0){$\mathfrak{sl}(2,\mathbb {R})\oplus\mathfrak{ sl}(2,\mathbb {R})\oplus
\mathfrak{sl}(2,\mathbb {R})$}}
\end{picture}$$
The nontrivial involution of the Dynkin diagram of $D(2,1;\alpha)$ makes the following Vogan diagram as shown below.
$$\begin{picture}(40,20)\thicklines
\put(-22,0){\circle*{9}} \put(0,-2.5){$\bigotimes$}\put(30,16){\circle{9}} \put(30,-16){\circle{9}}
\put(-18,0){\line(1,0){19}} \put(10,2){\line(3,2){16}}\put(10,-2){\line(3,-2){16}}
\qbezier(35,16)(51,1)(35, -16)
\put(35,16){\vector( -1, 1){0}}\put(35,-16){\vector( -1, -1){0}}
\put(120,0){\makebox(0,0){$\mathfrak{sl}(2,\mathbb{C})\oplus \mathfrak{sl}(2,\mathbb{R})$}}
\end{picture}$$
7. Case F(4) From the Dynkin diagram we can get only the real form $so(7)$ and the Vogan diagram
$$\begin{picture}(60,20) \thicklines
\put(-48,-3){$\bigotimes$} \put(-15,0){\circle{9}}\put(12,0){\circle{9}} \put(39,0){\circle{9}}
\put(-38,0){\line(1,0){19}}\put(-18,0){{\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(15,0){\line(1,-1){10}}\put(15,0){\line(1,1){10}}
\end{picture}}}\put(16,0){\line(1,0){19}}
\put(0,-32){\makebox(0,0){$\mathfrak{sl}(2,\mathbb{R})\oplus \mathfrak{so}(7)$}}
\end{picture}$$
we add the extra even part root to get the desired real forms and Vogan diagrams. $$\begin{picture}(60,20) \thicklines
\put(-75,0){\circle*{9}}\put(-48,-3){$\bigotimes$} \put(-15,0){\circle{9}}\put(12,0){\circle{9}} \put(39,0){\circle{9}}
\put(-71,0){\line(1,0){24}} \put(-38,0){\line(1,0){19}}\put(-18,0){{\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(15,0){\line(1,-1){10}}\put(15,0){\line(1,1){10}}
\end{picture}}}\put(16,0){\line(1,0){19}}
\put(0,-32){\makebox(0,0){$\mathfrak{sl}(2,\mathbb{R})\oplus \mathfrak{so}(7)$}}
\end{picture}$$ $$\begin{picture}(60,20) \thicklines
\put(-75,0){\circle{9}}\put(-48,-2.5){$\bigotimes$} \put(-15,0){\circle*{9}}\put(12,0){\circle{9}} \put(39,0){\circle{9}}
\put(-71,0){\line(1,0){24}} \put(-38,0){\line(1,0){19}}\put(-18,0){{\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(15,0){\line(1,-1){10}}\put(15,0){\line(1,1){10}}
\end{picture}}}\put(16,0){\line(1,0){19}}
\put(0,-32){\makebox(0,0){$\mathfrak{su}(2)\oplus \mathfrak{so}(1,6)$}}
\end{picture}$$ $$\begin{picture}(60,20) \thicklines
\put(-75,0){\circle{9}}\put(-48,-2.5){$\bigotimes$} \put(-15,0){\circle{9}}\put(12,0){\circle*{9}} \put(39,0){\circle{9}}
\put(-71,0){\line(1,0){24}}\put(-38,0){\line(1,0){19}}\put(-18,0){{\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(15,0){\line(1,-1){10}}\put(15,0){\line(1,1){10}}
\end{picture}}}\put(16,0){\line(1,0){19}}
\put(0,-32){\makebox(0,0){$\mathfrak{su}(2)\oplus \mathfrak{so}(2,5)$}}
\end{picture}$$ $$\begin{picture}(60,20) \thicklines
\put(-75,0){\circle{9}} \put(-48,-2.5){$\bigotimes$} \put(-15,0){\circle{9}}\put(12,0){\circle{9}} \put(39,0){\circle*{9}}
\put(-71,0){\line(1,0){24}} \put(-38,0){\line(1,0){19}}\put(-18,0){{\begin{picture}(20,20)
\put(6,-2){\line(1,0){20}}\put(6,2){\line(1,0){20}}
\put(15,0){\line(1,-1){10}}\put(15,0){\line(1,1){10}}
\end{picture}}}\put(16,0){\line(1,0){19}}
\put(0,-32){\makebox(0,0){$\mathfrak{su}(2)\oplus \mathfrak{so}(3,4)$}}
\end{picture}$$
8. Case $G(3)$
The Vogan diagram of $G(3)$ with real form $\mathfrak{sl}(2,\mathbb{R})\oplus G_{2,0}$ and $\mathfrak{sl}(2,\mathbb{R})\oplus G_{2,2}$ are
$$\begin{picture}(60,20) \thicklines
\put(-34,0){\circle*{9}}\put(-7,-2.5){$\bigotimes$}\put(26,0){\circle{9}}\put(54,0){\circle{9}}
\put(-30,0){\line(1,0){24}}\put(3,0){\line(1,0){19}}\put(23,0){{\begin{picture}(20,20)
\put(7,0){\line(1,0){20}}
\put(6,-3){\line(1,0){22}}\put(6,3){\line(1,0){22}}
\put(15,0){\line(1,1){10}}\put(15,0){\line(1,-1){10}}
\end{picture}}}
\end{picture}$$
$$\begin{picture}(60,20) \thicklines
\put(-34,0){\circle*{9}}\put(-7,-2.5){$\bigotimes$}\put(26,0){\circle{9}}\put(54,0){\circle*{9}}
\put(-30,0){\line(1,0){24}}\put(3,0){\line(1,0){19}}\put(23,0){{\begin{picture}(20,20)
\put(7,0){\line(1,0){20}}
\put(6,-3){\line(1,0){22}}\put(6,3){\line(1,0){22}}
\put(15,0){\line(1,1){10}}\put(15,0){\line(1,-1){10}}
\end{picture}}}
\end{picture}$$
#### Acknowledgement:
The authors thank National Board of Higher Mathematics, India (Project Grant No. 48/3/2008-R&DII/196-R) for financial support.
[9]{} Batra P, [*Invariant of Real forms of Affine Kac-Moody Lie algebras*]{}, Journal of Algebra 223, 208-236 (2000). Batra P, [*Vogan diagrams of affine Kac-Moody algebras*]{}, Journal of Algebra 251, 80-97 (2002).
Frappat L., Sciarrino A. and Sorba P. [*Dictionary on Lie algebras and Superalgebras*]{}, academic press (2000). Kac V.G. , Frajria P.M. and Papi P., [*Denominator Identities for finite dimensional Lie superalgebras and Howe duality for compact dual pairs*]{}, arxiv:1102.3785v1 \[math.RT\] 18 february (2011). Knapp A.W., [*Lie groups beyond an Introduction*]{}, Second Edition. Meng-Kiat Chuah, [*Cartan automorphisms and Vogan superdiagrams*]{}, Math.Z.DOI 10.1007/s00209-012-1030-z Pati, K.C., Parashar, D. [*Satake superdiagrams, real forms and Iwasawa decomposition of classical Lie superalgebras*]{}, J. Phys. A 31, 767–778 (1998) Parker M, [*Classification of real simple Lie superalgebras of classical type*]{}, J. Math.Phys. 21(4), April (1980). Ransingh B, Behera A and Pati K.C., [*Vogan diagrams of some hyperbolic Kac-Moody algebras and Iwasawa decomposition*]{}, arXiv:1205.3724v2 \[math.RT\] 16 May 2012. Paul Tanushree, [*Vogan diagram of twisted Affine Kac-Moody lie algebras*]{}, Pacific Jouranl of mathematics, 239 (1), January (2009). Sergeev Alexander N. and Veselov Alexander P., [*Grothendieck rings of basic classical Lie superalgebras*]{}, Annals of Mathematics 173 (2) 663 - 703 (2011).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this work we report on the results obtained in a detailed and systematical study of the possibility to measure the parameters appearing in the electroweak chiral lagrangian. The main novelty of our approach is that we do not use the Equivalence Theorem and therefore we work explicitly with all the gauge boson degrees of freedom.'
---
21.2cm 16.0cm -1.0in 5 pt -42pt
[**DETERMINATION OF THE ELECTROWEAK CHIRAL-LAGRANGIAN PARAMETERS AT THE LHC**]{} [^1]\
[Departamento de Física Teórica\
Universidad Complutense de Madrid\
28040 Madrid, Spain\
and\
]{} [**M. T. Urdiales**]{} [^2]\
[Departamento de Física Teórica\
Universidad Autónoma de Madrid\
28049 Madrid, Spain\
]{} February 1995\
1.0cm hep-ph/9502255\
20 true cm
Introduction
============
Today it is clear for many physicists that one of the main goals of the future CERN Large Hadron Collider ($LHC$) is to find as much information as possible about the nature of the Standard Model ($SM$) Electroweak Symmetry Breaking ($ESB$) [@M.E.]. In spite of the huge amount of data obtained in the last years at the Large Electron-Positron Collider ($LEP$) it is very few what we really know about the $ESB$. The proposed mechanisms include ideas such as supersymmetry (see [@Susy] and references therein), technicolor [@Techni] and many others. Therefore it would be very interesting to have some model independent framework to make a phenomenological description of the $ESB$ physics.
In fact such a framework exists, at least for the strongly interacting case i.e., when no light modes are present in the $ESB$. It has been developed in the last years and used to describe the scattering of the longitudinal components ($LC$) of the electroweak gauge bosons [@HeE1] as well as the precision tests of the Standard Model coming from $LEP$ [@HeE2]. It is based on the application of the chiral Lagrangians or Chiral Perturbation Theory ($\chi PT$ )[@Wleff] (previously invented for the description of the low-energy hadron interactions) to the dynamics of the $ESB$ Goldstone bosons ($GB$).
One assumes that there is a physical system with a global symmetry group $G$ which is spontaneously broken to some other group $H$. This global symmetry breaking drives the gauge electroweak $SU(2)_L \times U(1)_Y$ symmetry breaking down to $U(1)_{em}$ through the well known Higgs mechanism. The only election for the $G$ and $H$ groups compatible with the presence of the $SU(2)_{L+R}$ custodial symmetry [@cus] (to have a $\rho$ parameter naturally close to one) and the existence of three massive gauge bosons after the $ESB$ i.e., the $W^+$, $W^-$ and $Z^0$, is $G=SU(2)_L \times SU(2)_R$, $H=SU(2)_{L+R}$.
Then the low energy dynamics of the $ESB$ is described by a $SU(2)_L \times U(1)_Y$ gauged non-linear sigma model ($GNLSM$) including an arbitrary large number of terms in the action with different number of derivatives of the $GB$ fields and electroweak gauge bosons. The corresponding couplings (parameters) encode the dynamics of the $ESB$ sector of the $SM$ and must be renormalized to absorb divergences. However, at low enough energies only a small number of terms (and couplings) are needed to eliminate all the divergences. In principle the values of these couplings or parameters could be obtained from the underlying theory or directly by fitting them from future experiments.
In this work we will study the possibilities of the $LHC$ for measuring these parameters and the expected corresponding errors. In particular we will concentrate in events producing $Z^0Z^0$ or $W^{\pm}Z^0$ pairs. As it was mentioned above, the application of the chiral Lagrangian technique to the production of electroweak gauge bosons is not new and has already been considered at the literature [@si]. However, all the applications worked out until now are based in the so called Equivalence Theorem ($ET$) [@ET]. This theorem relates the $S$ matrix elements of processes containing electroweak gauge bosons $LC$ with the corresponding processes with $GB$. However, in a recent work [@ETCL] concerning the formulation of the $ET$ in the context of $\chi PT$, the Equivalence Theorem is severely restricted to a narrow energy applicability window. It can also be applied in the high energy domain together with the $\chi PT$ but using some non-perturbative technique, like dispersion relations [@Uni] or the large $N$ limit [@largeN], in order to have an appropriate unitarity behaviour of the amplitudes.
For this reason we consider prioritary to apply directly the chiral Lagrangian description of the $ESB$ without using the $ET$. The main problems of this approach are two: First one has to include explicitly the gauge degrees of freedom in the model which makes the computations extremely more difficult. Second, one has to restrict the results to the low energy region where standard $\chi PT$ can safely be applied thus losing many higher energy events. The advantage is that the values of the fitted parameters will be more reliable since one is not using the $ET$ complemented with some non-perturbative method.
The plan of this work goes as follows: In section 2 we introduce the chiral effective lagrangian to be used with its parameters and the corresponding Feynman rules. In section 3 we compute the cross-section of the $LHC$ subprocesses that are relevant for measuring the parameters. In section 4 we consider the signatures and the possible backgrounds. In section 5 we show how we compute the total number of expected events at the $LHC$ from the subprocess cross-sections for the signals and backgrounds. In section 6 we discuss the sensitivity of the machine to the different parameters, we define the optimal cuts and we compute the statistical significance of the different parameter measures and estimate the statistical errors. In section 7 we consider other sources of systematical errors such as the uncertainty on the proton structure functions and in section 8 the effect of the running of the parameters. Finally, in section 9 we review the main conclussions of our work.
The effective Lagrangian and the Feynman rules
==============================================
In this section we consider the most general effective Lagrangian for the $ESB$ compatible with the $SU(2)_L \times U(1)_Y$ local symmetry of the $SM$ and the breaking pattern $SU(2)_L \times U(1)_Y
\rightarrow U(1)_{\rm em}$. The well known chiral Lagrangian describing the $ESB$ of the $SM$, can be written as an infinite expansion with terms of increasing number of gauge fields and derivatives of the Goldstone bosons ($GB$), with an infinite number of arbitrary parameters. This chiral Lagrangian can be seen as a low momentum expansion for the corresponding Green functions. At some given order in the number of $GB$ derivatives one can work only with a finite number of terms and parameters. In this case the model can only be applied to much smaller energies than $4\pi v$ which is the parameter controlling this expansion ($4\pi v
\simeq 3 TeV$ since $v\simeq 250 GeV$).
The first term of this effective Lagrangian ($O(p^2)$) is given by:
$${\cal L}^{(2)}=\frac{v^2}{4}tr[D_{\mu} U(D^{\mu}U)^{\dagger}]
\label{eq:lag1}$$
and by the Yang-Mills Lagrangian:
$${\cal L}_{YM}=-\frac{1}{4} B_{\mu\nu} B^{\mu\nu} - \frac{1}{2}
Tr[F_{\mu\nu}F^{\mu\nu}]
\label{eq:YM}$$
We choose the GB parametrization as an unitary matrix $U$ belonging to the quotient space $SU(2)_L \times SU(2)_R / SU(2)_{L+R}$.
$$U = {\rm exp} (i \frac{\vec{\tau}.\vec{\pi}}{v})$$
where $\vec{\tau} \equiv ({\tau}_1, {\tau}_2,{\tau}_3)$ are the Pauli matrices and $\vec{\pi} \equiv ({\pi}_1(x), {\pi}_2(x), {\pi}_3(x))$ represents the triplet of $GB$.
The covariant derivative $D_{\mu}U$ is defined as:
$$D_{\mu}U = {\partial}_{\mu} U + ig W_{\mu} U - ig' U Y_{\mu}$$
Here, $W_{\mu}$ and $Y_{\mu}$ are the $SU(2)_L$ and the $U(1)_Y$ gauge fields given by:
$$W_{\mu} = \frac{{\vec{W}}_{\mu} {\vec{\tau}}}{2}$$
$$Y_{\mu} = \frac{B_{\mu} {\tau}_3}{2}$$
where $W_{\mu}=(W^1_{\mu},W^2_{\mu},W^3_{\mu})$ represents the triplet of $SU(2)_L$ gauge fields.
As usual, the covariant field strength tensors are defined as
$$\begin{aligned}
F_{\mu\nu}(x) & = & {\partial}_{\mu} W_{\nu}(x) - {\partial}_{\nu}
W_{\mu}(x) + ig [W_{\mu}(x),W_{\nu}(x)] \\
B_{\mu\nu}(x) & = & {\partial}_{\mu} B_{\nu}(x) - {\partial}_{\nu}
B_{\mu}(x)\end{aligned}$$
The transformation properties under the $SU(2)_L \times U(1)_Y$ gauge transformations are
$$\begin{aligned}
ig W'_{\mu} & = & g_L (x) ig W_{\mu} g_L^{\dagger} (x) + g_L (x)
{\partial}_{\mu} g_L^{\dagger} (x) \\
ig' Y'_{\mu} & = & ig' Y_{\mu} + g_Y (x) {\partial}_{\mu}
g_Y^{\dagger} (x) \\
U' & = & g_L (x) U {g_Y}^{\dagger} (x)\end{aligned}$$
where
$$\begin{aligned}
g_L(x) & = & e^{i {\theta}_k (x) {\tau}^{k} /2} \\
g_Y(x) & = & e^{i \beta(x).{\tau}^3 /2}\end{aligned}$$
Note that, as expected, the $GB$ fields, ${\pi}_i$, transform non-linearly .
In order to construct the chiral lagrangian to order $O(p^4)$ we define, following Longhitano in [@Long], the quantities $T$, $V_{\mu}$ and ${\cal D}_{\mu} O(x)$:
$$\begin{aligned}
T & = & U {\tau}^3 U^{\dagger} \\
V_{\mu} & = & D_{\mu}U U^{\dagger} \\
{\cal D}_{\mu} O(x) & = & {\partial}_{\mu} O(x) + ig [W_{\mu}(x), O(x)]\end{aligned}$$
Then the complete electroweak chiral Lagrangian with the whole set of $SU(2)
\times U(1)_Y$, Lorentz, $C$, $P$ and $T$ invariant operators up to dimension four has the form:
$${\cal L}={\cal L}^{(2)}+{\cal L}_{YM}+{\cal L'}_1+ \sum_{i=1}^{13} {\cal
L}_i
\label{eq:lag3}$$
In this equation, the different terms ${\cal L}_i$ are the $SU(2)_L
\times U(1)_Y$ invariant functions of the gauge vector bosons and the $GB$ fields, containing four derivatives, whereas ${\cal L}'_1$ has dimension two. They have the following expressions:
$$\begin{aligned}
{\cal L'}_1 & = & \frac{1}{4} g^2 {\alpha}_0 v^2 [Tr(TV_{\mu})]^2
\nonumber \\
{\cal L}_1 & = & \frac{1}{2} g^2 {\alpha}_1 B_{\mu\nu} Tr(TF^{\mu\nu})
\nonumber \\
{\cal L}_2 & = & \frac{1}{2} i g {\alpha}_2 B_{\mu\nu}
Tr(T[V^{\mu},V^{\nu}]) \nonumber \\
{\cal L}_3 & = & i g {\alpha}_3 Tr(F_{\mu\nu}[V^{\mu},V^{\nu}])
\nonumber \\
{\cal L}_4 & = & {\alpha}_4 [Tr(V_{\mu}V_{\nu})]^2 \nonumber \\
{\cal L}_5 & = & {\alpha}_5 [Tr(V_{\mu}V^{\mu})]^2 \nonumber \\
{\cal L}_6 & = & {\alpha}_6 Tr[(V_{\mu}V_{\nu})]
Tr(TV^{\mu}) Tr(TV^{\nu}) \nonumber \\
{\cal L}_7 & = & {\alpha}_7 Tr[(V_{\mu}V^{\mu})] [Tr(TV^{\nu})]^2
\nonumber \\
{\cal L}_8 & = & \frac{1}{4} g^2 {\alpha}_8 [Tr(TF_{\mu\nu})]^2
\nonumber \\
{\cal L}_9 & = & \frac{1}{2} i g {\alpha}_9 Tr(TF_{\mu\nu})
Tr(T[V^{\mu},V^{\nu}]) \nonumber \\
{\cal L}_{10} & = & \frac{1}{2} {\alpha}_{10}
[Tr(TV_{\mu})Tr(TV_{\nu})]^2 \nonumber \\
{\cal L}_{11} & = & {\alpha}_{11} Tr[({\cal D}_{\mu} V^{\mu})^2]
\nonumber \\
{\cal L}_{12} & = & \frac{1}{2} {\alpha}_{12} Tr(T{\cal D}_{\mu}{\cal
D}_{\nu}V^{\nu}) Tr(TV^{\mu}) \nonumber \\
{\cal L}_{13} & = & \frac{1}{2} {\alpha}_{13} [Tr(T{\cal D}_{\mu}
V^{\nu})]^2
\label{eq:lag2}\end{aligned}$$
As it is well known there is some arbitrariness in the choice of a particular base of invariants. Using the equations of motion it would be possible to eliminate the operators ${\cal L}_{11}$, ${\cal L}_{12}$ and ${\cal L}_{13}$ redefining the rest of the terms (see for example [@Fer]).
So, a set of $\alpha_k$ ($k=0,..13$) parameters have appeared in the definition of this effective Lagrangian. We display in Table \[tab:equiv\] the relations between some different sets of the chiral Lagrangian parameters used in the literature, i.e. by Longhitano [@Long], Feruglio [@Fer] and Dobado et.al. [@HeE2].
The theory defined by the Lagrangian above is non-renormalizable in the strict sense, but the divergences appearing when we calculate at one-loop with the chiral Lagrangian at lowest order eq.(\[eq:lag1\]) have the same form that some of the terms obtained at tree level in the following order in the chiral expansion. The counterterms needed to reabsorb the divergences generated to the one-loop level with ${\cal
L}^{(2)}$ in the Landau gauge were obtained by the authors in [@Long]. They have the same form that ${\cal L}'_1$, ${\cal L}_1$, ${\cal L}_2$, ${\cal L}_3$, ${\cal L}_4$ and ${\cal L}_5$, whereas the rest of the terms in eq.(\[eq:lag2\]) are not needed for renormalization.
We will describe the interactions between gauge bosons and fermions by the same Lagrangian than that of the $SM$, which can also describe the couplings between fermions and scalars by means of the usual Yukawa terms. However the latter are not used in our calculations since we work all the time with massless fermions.
Once we have set the Lagrangian to be used, the next step is to define the quantum theory. This can be done in a standard way using the Faddeev-Popov method and choosing some appropriate $R_{\xi}$ covariant gauge. The gauge fixing functions are constructed by requiring that the terms of the form $W^{i}_{\mu}
{\partial}^{\mu} {\pi}_i(x)$ and $B_{\mu} {\partial}^{\mu} {\pi}_3$ cancel out; thus yielding:
$$\begin{aligned}
f_i(W_{i\mu}) & = & {\partial}^{\mu} W_{i\mu} - \frac{v}{2} {\xi}_W
g{\pi}_i,
\;\;\;\;\; i=1,2,3 \nonumber \\
f_0(B_{\mu}) & = & {\partial}^{\mu} B_{\mu} + \frac{v}{2} {\xi}_B g'
{\pi}_3\end{aligned}$$
Now we can write the gauge-fixing and Faddeev-Popov terms as:
$${\cal L}_{GF} = - \frac{1}{2 {\xi}_W} \sum_{i=1}^{3} [f_i(W_{i\mu})]^2 -
\frac{1}{2 {\xi}_B} [f_0(B_{\mu})]^2$$
$$\begin{aligned}
{\cal L}_{FPG} & = & \int d^4 y {c^{\dag}}_m(x) \frac{\delta
f_m(x)}{\delta {\theta}_n (y)} c_n(y) \;\;\;\;\;\; m,n= 0,1,2,3\end{aligned}$$
Therefore, the effective action and the effective Lagrangian (non-linear $SM$ Lagrangian, ${\cal L}_{NLSM}$) take the form:
$$S_{eff} = \int d^4x [ {\cal L}(x) + {\cal L}_{GF} + {\cal L}_{FPG}]$$
$${\cal L}_{NLSM} = {\cal L}^{(2)}+{\cal L'}_1+{\cal
L}^{(4)}+{\cal L}_{YM}+{\cal L}_{GF}
+{\cal L}_{FPG}+{\cal L}_{\Psi}
\label{eq:MENL}$$
In the following we will to work in the Landau gauge (${\xi}_W={\xi}_B=0$). In this gauge the perturbative $\pi$ propagator remains massless and the Faddeev-Popov ghosts only couple to the gauge field. There is not direct coupling of the $\pi$ field to the ghosts. Moreover, the counterterms necessary to cancel divergences with ${\cal L}^{(2)}$ at one loop are gauge invariant functions of the gauge fields and the $GB$ alone (the ${\cal L}'_1$, ${\cal L}_1$, ${\cal L}_2$, ${\cal L}_3$, ${\cal L}_4$ and ${\cal L}_5$ terms mentioned above). In other gauges, or other parametrizations of the coset space, the counterterms could also be functions of the ghost fields, and their structure would be determined using the Becchi-Rouet-Stora ($BRS$) invariance.
Thus, the gauge fixing and the Faddeev-Popov term in the Landau gauge read:
$$\begin{aligned}
{\cal L}_{GF} & = & - \frac{1}{2 {\xi}} \sum_{i=1}^{3}
[{\partial}^{\mu}(W_{i\mu})]^2 - \frac{1}{2 {\xi}}
[{\partial}^{\mu}(B_{\mu})]^2 \nonumber \\
{\cal L}_{FPG} & = & {\partial}_{\mu} {c_i}^{\dag} {\partial}^{\mu} c^i
- g{\epsilon}_{ijk} {c^i}^{\dag} c^j {\partial}^{\mu} {W_{\mu}}^k +
{\partial}_{\mu} c_0^{\dag} {\partial}^{\mu} c^0
\label{eq:quant}\end{aligned}$$
In the Appendix we show the Feynman rules derived from the Lagrangian ${\cal L}_{NLSM}$ eq.(\[eq:MENL\]) that we use for our calculations. They correspond to Figures \[fig:GF2\], \[fig:GF3\] and \[fig:GF4\]. As usual it has been useful to redefine the gauge fields as:
$$\begin{aligned}
W^{\pm}_{\mu} & = & \frac{{W_{\mu}}^1 \mp i{W_{\mu}}^2}{\sqrt{2}} \\
Z_{\mu} & = & {\rm cos}{\theta}_W {W_{\mu}}^3 -{\rm sin}{\theta}_W
B_{\mu}
\\ A_{\mu} & = & {\rm sin}{\theta}_W {W_{\mu}}^3 + {\rm cos}{\theta}_W
B_{\mu}\end{aligned}$$
where the weak angle is defined by
$${\rm tg}{\theta}_W = \frac{g'}{g}$$
and the $GB$ as:
$$\begin{aligned}
{\pi}^{\pm} & = & \frac{{\pi}^1 \mp i{\pi}^2}{\sqrt{2}} \\
{\pi}^0 & = & {\pi}^3\end{aligned}$$
The subprocesses cross-sections
===============================
In this work we are interested in studying of the different cross sections that contribute to the final states $Z^0Z^0$ and $W^{\pm}Z^0$ at $pp$ colliders such as $LHC$ since they are the most promising from the experimental point of view. In both cases we focus our attention on the so-called gold-plated events, where the produced bosons decay to the leptonic final states $l=e,\mu,
\nu$. In spite of the very small branching ratios ($BR$),
$$\begin{aligned}
BR(Z^0Z^0 \rightarrow l \bar{l} l'\bar{l'}) & = & 0.0044 \\
BR(W^{\pm}Z^0 \rightarrow l {\nu}_{l} l'\bar{l'}) & = & 0.013\end{aligned}$$
these are the most interesting events since they are much easier to detect than the hadronic channels.
On the other hand, we will assume that it is not possible to detect experimentally the different polarizations of the gauge bosons in the final state. Therefore we have to include in our computations all the $Z^0$ or $W^{\pm}$ gauge bosons, either transversally or longitudinally polarized. As it was mentioned in the introduction, instead of using the $ET$ to calculate the collisions of the weak bosons, as it is usually done, we have computed the amplitudes with gauge bosons as initial and final states and then projected them in all their polarizations. Thus, as we are not going to apply the $ET$ we only consider the maximal bound on the energy required for the applicability of the chiral Lagrangian formalism, but we do not have any low energy bound. In this way we can take into account the influence of the chiral lagrangian parameters in all the gauge bosons polarization channels and not only in the longitudinal ones as it is usually the case. The main disadvantage in our procedure is that the calculations become much more complicated. This fact leads us to work only at tree level to make the computation more accessible.
In principle, according to the spirit of $\chi PT$, the one-loop corrections coming from the lowest order lagrangian should also be included since they are $O(p^4)$. However, the computation of these corrections is extremely involved. In addition, they have not dependence on the chiral parameters. As it will become clear later, our main interest is to study the dependence of the number of events on the chiral parameters in order to see which of them could be measured at the $LHC$. For this reason we do not expect that our results concerning the measurable parameters would change too much when the one-loop corrections are included but, of course, the precise number of events will do.
Later we will also study the effect of the running of the parameters on our results which in some way takes into account part of the one-loop contribution, but, in any case, our approach is the simplest one including the right dependence on the chiral lagrangian parameters.
Now we describe how we have calculated the cross sections of the above mentioned processes. From the quantized chiral Lagrangian that appears in eq.(\[eq:MENL\]) we obtain the corresponding Feynman rules (see Appendix). Then we compute, at tree level, the amplitudes for the different subprocesses contributing to the $Z^0Z^0$ or $W^{\pm}Z^0$ final state, using a REDUCE code, and then the corresponding cross-sections. We calculate them in the center of mass frame, with a total energy $\sqrt{\hat{s}}$. There, the four momenta of the interacting particles are:
$$\begin{aligned}
p_1 & = & (E_1, 0,0,p) \\
p_2 & = & (E_2, 0,0,p) \\
p_3 & = & (E_3, p'{\rm sin}\theta , 0, p'{\rm cos}\theta) \\
p_4 & = & (E_4, -p'{\rm sin}\theta , 0, -p'{\rm cos}\theta)\end{aligned}$$
Here, $E_i$ are the particle energies, and $p$ and $p'$ are, respectively, the magnitudes of the three-momenta of the initial and the final states.
$$\begin{aligned}
p & = & \frac{1}{2\sqrt{\hat{s}}}{[{\hat{s}}^2
-2{\hat{s}}({m_1}^2+{m_2}^2)+({m_2}^2-{m_1}^2)^2]}^{1/2} \\
p' & = & \frac{1}{2\sqrt{\hat{s}}}{[{\hat{s}}^2
-2{\hat{s}}({m_3}^2+{m_4}^2)+({m_4}^2-{m_3}^2)^2]}^{1/2} \\
E_i & = & \sqrt{{m_i}^2+{p_i}^2} \;\;\;\;\; i=1,..4 \\
{\hat{s}} & = & (p_1+p_2)^2 \\
{\hat{t}} & = & (p_1-p_3)^2 \\
{\hat{u}} & = & (p_1-p_4)^2\end{aligned}$$
where $\hat{s}$, $\hat{t}$ and $\hat{u}$ are the Mandelstam variables. The differential cross section is given by this expression:
$$d {\hat{\sigma}} = \frac{1}{32 \pi {\hat{s}}} \frac{p'}{p} \sum {\mid M
\mid}^2 d{\rm cos} {\theta}
\label{eq:subpro}$$
where $M$ is the helicity amplitude, and the $\Sigma$ symbol refers to the sum of all the final polarizations and the average on the initial ones. In the case of electroweak gauge bosons scattering, where we separate the transversal and longitudinal polarizations in the initial state (as we will see in section 5), we take into account the different helicities contributing to $TT$, $TL$, $LT$ and $LL$ polarizations of the initial bosons, to do this average. We also include the corresponding branching ratios to the gold-plated events.
Let us consider now the processes corresponding to the elastic collision of gauge bosons (fusion processes). Symbolically this reaction can be written as:
$$V_1 (p_1,{\lambda}_1) V_2 (p_2,{\lambda}_2) \rightarrow
V_3 (p_3,{\lambda}_3) V_4 (p_4,{\lambda}_4)$$
We choose the following polarization vectors in the helicity base (${\lambda}_i=\pm 1,0$):
$$\begin{aligned}
{\epsilon}^{\pm}_1 & = & \frac{1}{\sqrt{2}} (0,{\mp}1, -{\rm i},0) \\
{\epsilon}^0_1 & = & \frac{1}{m_{V_1}} (p,0,0,E_1) \\
{\epsilon}^{\pm}_2 & = & \frac{1}{\sqrt{2}} (0,{\mp}1, {\rm i},0) \\
{\epsilon}^0_2 & = & \frac{1}{m_{V_2}} (p,0,0,-E_2) \\
{\epsilon}^{\pm}_3 & = & \frac{1}{\sqrt{2}} (0, \mp {\rm cos} \theta,
-{\rm i}, \pm {\rm sin}\theta) \\
{\epsilon}^0_3 & = & \frac{1}{m_{V_3}} (p', E_3 {\rm sin} \theta,
0, E_3 {\rm cos} \theta) \\
{\epsilon}^{\pm}_4 & = & \frac{1}{\sqrt{2}} (0, \mp {\rm cos} \theta,
{\rm i}, \pm {\rm sin} \theta) \\
{\epsilon}^0_4 & = & \frac{1}{m_{V_4}} (p', -E_4 {\rm sin} \theta,
0, -E_4 {\rm cos} \theta)\end{aligned}$$
The number of independent helicity amplitudes in fusion processes of massive gauge bosons, such as $Z^0Z^0 \rightarrow
Z^0Z^0$, $W^+W^- \rightarrow Z^0Z^0$, $W^+Z^0 \rightarrow
W^+Z^0$ or $W^-Z^0 \rightarrow W^-Z^0$, is $3^4=81$. The photon has only two polarized transverse states and therefore, in the processes $W^+\gamma
\rightarrow W^+Z^0$ or $W^-\gamma \rightarrow W^-Z^0$ the number of helicity amplitudes is $2 \times 3^3 = 54$. As our chiral Lagrangian affecting the boson scattering is invariant under $C$, $P$ and $T$ transformations, we can derive many relations between different helicity amplitudes. However we have not used those relations to save computations. Instead we have calculated all the helicity amplitudes in each process and the relations between them have been only used to check our results.
$Z^0Z^0$ final state
--------------------
Now we concentrate in the different processes that contribute to the final state $Z^0Z^0$ [@si]:
$$\begin{aligned}
gg & \rightarrow & Z^0Z^0 \\
q\bar{q} & \rightarrow & Z^0Z^0 \\
Z^0_TZ^0_T & \rightarrow & Z^0Z^0 \\
Z^0_TZ^0_L & \rightarrow & Z^0Z^0 \\
Z^0_LZ^0_T & \rightarrow & Z^0Z^0 \\
Z^0_LZ^0_L & \rightarrow & Z^0Z^0 \\
W^+_TW^-_T & \rightarrow & Z^0Z^0 \\
W^+_TW^-_L & \rightarrow & Z^0Z^0 \\
W^+_LW^-_T & \rightarrow & Z^0Z^0 \\
W^+_LW^-_L & \rightarrow & Z^0Z^0\end{aligned}$$
Note that we separate the contribution of the different polarizations channels in the initial state, since different polarizations will have different luminosities in $pp$ collisions. In particular we use the Weizsaker-Williams [@fegamm] and the effective $W$ approximation [@Daw] to compute the $V_1V_2$ pair luminosity in the $LHC$ beams. Therefore, we have to divide these cross-sections into the contributions coming from the different gauge boson initial polarizations.
On the other hand, it is well known that at supercollider energies the one-loop process $gg \rightarrow ZZ$ is not negligible. The gluon-gluon fusion cross section was calculated in the Minimal Standard Model, $MSM$ (with just one Higgs doublet) via one-loop of quarks by Glover and Van deer Bij [@Glo]. The corresponding diagrams are shown in Figure \[fig:gg\]. As we are using a chiral Lagrangian description of the ESB sector we do not include the contribution coming from diagrams with the Higgs boson in our computations.
The quark-antiquark annihilation represents the main source of $Z^0Z^0$ pairs in $pp$ colliders like the $LHC$. As in the case of gluon fusion, there is no dependence on the chiral parameters in the calculation, at our level of approximation. The cross section at tree level, which only receives contribution from the $t$ and $u$ channels (Figure \[fig:qq\]), has the same well-known expression as in the $MSM$.
The $Z^0Z^0$ fusion calculated using the chiral Lagrangian, only receives contribution from the vertex $Z^0Z^0Z^0Z^0$ (Figures \[fig:GF4\] and \[fig:ZZ\]). The only dependence on the effective Lagrangian parameters is through the following combination eq.(\[eq:GF4\]):
$${\alpha}_4 + {\alpha}_5 + 2({\alpha}_6 + {\alpha}_7 + {\alpha}_{10})$$
Again we can use the relations derived from the $C$, $P$ and $T$ to relate the helicity amplitudes and thus to check our results. These relations are:
$$\begin{aligned}
|M_{{{\lambda}_1}{{\lambda}_2}{{\lambda}_3}{{\lambda}_4}}| &
\stackrel{C}{=} &
|M_{{{\lambda}_2}{{\lambda}_1}{{\lambda}_4}{{\lambda}_3}}| \\
|M_{{{\lambda}_1}{{\lambda}_2}{{\lambda}_3}{{\lambda}_4}}| &
\stackrel{P}{=} &
|M_{{-{\lambda}_1}{-{\lambda}_2}{-{\lambda}_3}{-{\lambda}_4}}| \\
|M_{{{\lambda}_1}{{\lambda}_2}{{\lambda}_3}{{\lambda}_4}}| &
\stackrel{T}{=} &
|M_{{{\lambda}_3}{{\lambda}_4}{{\lambda}_1}{{\lambda}_2}}|\end{aligned}$$
In order to relate some helicity amplitudes we have also taken into account whether the scattering particles are identical or not., There are still $15$ independent amplitudes remaining after applying these symmetry relations.
The $W^+W^- \rightarrow Z^0Z^0$ reaction represents another source of $Z^0Z^0$ pairs from $pp$ beams. Every helicity amplitude in this process receives contributions from the $t$ and $u$ channels, by exchanging a $W$ or a $\pi$, and directly from the vertex $W^+W^-Z^0Z^0$ (Figure \[fig:WW\]). Therefore we can write:
$$M_{{{\lambda}_1}{{\lambda}_2}{{\lambda}_3}{{\lambda}_4}} = [M_{t_1} +
M_{t_2} + M_{u_1} + M_{u_2} + M_4 ]_{{{\lambda}_1}{{\lambda}_2}
{{\lambda}_3}{{\lambda}_4}}$$
where the subscript $1$($2$) is referred to the exchange of a gauge boson (Goldstone boson) through the corresponding channel. There are $81$ helicity amplitudes, but the number of independent $M_{{{\lambda}_1}
{{\lambda}_2} {{\lambda}_3} {{\lambda}_4}}$ is reduced to $25$ by means of symmetry relations derived from $P$ and $C$ invariance.
As we expected, the amplitudes that we have obtained satisfy these equalities. On the other hand, all the chiral parameters, ${\alpha}_k$ (but ${\alpha}_{10}$), affect this process as we can deduce from the Feynman rules eqs.(\[eq:GF3\] and \[eq:GF4\]) and from the different diagrams that contribute to the collision $W^+W^- \rightarrow Z^0Z^0$.
$W^{\pm}Z^0$ final state
------------------------
As it was previously said, we have also studied the $W^{\pm}Z^0$ final state. The different sources for $W^{\pm}Z^0$ pairs in $pp$ colliders at tree level order are the following:
$$\begin{aligned}
q\bar{q'} & \rightarrow & W^+Z^0 \\
q'\bar{q} & \rightarrow & W^-Z^0 \\
W^+_T Z^0_T & \rightarrow & W^+ Z^0 \\
W^+_T Z^0_L & \rightarrow & W^+ Z^0 \\
W^+_L Z^0_T & \rightarrow & W^+ Z^0 \\
W^+_L Z^0_L & \rightarrow & W^+ Z^0 \\
W^-_T Z^0_T & \rightarrow & W^- Z^0 \\
W^-_T Z^0_L & \rightarrow & W^- Z^0 \\
W^-_L Z^0_T & \rightarrow & W^- Z^0 \\
W^-_L Z^0_L & \rightarrow & W^- Z^0 \\
W^+_T \gamma & \rightarrow & W^+ Z^0 \\
W^+_L \gamma & \rightarrow & W^+ Z^0 \\
W^-_T \gamma & \rightarrow & W^- Z^0 \\
W^-_L \gamma & \rightarrow & W^- Z^0\end{aligned}$$
where we call $q=u,c$ and $q'=d,s,b$.
Most of the boson pairs $W^{\pm}Z^0$ are obtained in $pp$ colliders via quark-antiquark annihilations. Using the chiral Lagrangian framework we calculate the corresponding cross section at tree level order. As we can see in Figure \[fig:qq’\], three standard $s$, $t$ and $u$ diagrams contribute, but the new physics coming from the non-linear lagrangian is isolated in the three boson vertex, which in this process affects only the $s$ channel.
$$M = M_s + M_t + M_u$$
If we look at the Feynman rules (Figure \[fig:GF3\]) we can deduce that the set of ${\alpha}_k$ parameters contributing to this process are ${\alpha}_1$, ${\alpha}_2$, ${\alpha}_3$, ${\alpha}_8$, ${\alpha}_9$, ${\alpha}_{11}$, ${\alpha}_{12}$ and ${\alpha}_{13}$ eq.(\[eq:GF3\]). In the hadronic collider case, the three boson vertex has been studied in this process with chiral Lagrangians, in [@Luk] at tree level, using the $ET$ (by J.Bagger, et.al. in [@si]), or in our previous study [@AnMa] for the case $g'=0$ and including the running of the couplings. In this last case we analyzed the sensitivity to the ${\alpha}_3$ parameter. Here we work firstly at tree level order, but including the dependence on the $\alpha_k$ parameters in all the channels to produce $W^{\pm}Z^0$ pairs in $pp$ colliders. Finally we will include the running parameters effect in section 8.
Other mechanism to obtain $W^{\pm}Z^0$ pairs is through $W^{\pm}Z^0$ or $W^{\pm} \gamma$ collisions transversally or longitudinally polarized at the initial state. In both cases the different polarization amplitudes are obtained by adding the contribution of the $s$ and $u$ channels and the four gauge boson vertex in (Figures \[fig:WZ\] and \[fig:Wf\]). The exchanged particles in the $s$ and $u$ diagrams are a gauge boson ($s_1$, $u_1$) and a $GB$ ($s_2$, $u_2$), respectively.
$$M_{{{\lambda}_1}{{\lambda}_2}{{\lambda}_3}{{\lambda}_4}} = [M_{s_1} +
M_{s_2} + M_{u_1} + M_{u_2} + M_4 ]_{{{\lambda}_1}{{\lambda}_2}
{{\lambda}_3}{{\lambda}_4}}$$
In the $W^{\pm}Z^0 \rightarrow W^{\pm}Z^0$ collisions there are initially $2 \times 3^4$ helicity amplitudes. This $2$ factor can be dropped using $C$ invariance that relates the magnitude of the helicity amplitudes $M_{{\lambda}_1{\lambda}_2{\lambda}_3{\lambda}_4}$ in the process with $W^+$ and with $W^-$. If we apply the relations between amplitudes derived from $P$ and $T$ invariance, there will be only $25$ remaining amplitudes. These subprocesses cross sections are affected by all the ${\alpha}_k$, but ${\alpha}_{10}$, as we can deduce from the Feynman rules eqs.(\[eq:GF3\] and \[eq:GF4\]).
The number of helicity amplitudes corresponding to the processes $W^{\pm}\gamma \rightarrow W^{\pm}Z^0$ is $2 \times (2
\times 3^3)$. The $2$ factor disappears when we take into account $C$ invariance as in the previous case. The number of remaining amplitudes is reduced again, by another $2$ factor after applying $P$ invariance. Therefore in these processes we have $3^3=27$ independent helicity amplitudes that we calculate using the Feynman rules shown in eqs.(\[eq:GF3\] and \[eq:GF4\]). Here we can observe that the set of the chiral lagrangian parameters affecting the $W^{\pm}\gamma \rightarrow
W^{\pm}Z^0$ cross sections is the following: ${\alpha}_0$, ${\alpha}_1$, ${\alpha}_2$, ${\alpha}_3$, ${\alpha}_8$, ${\alpha}_9$, ${\alpha}_{11}$, ${\alpha}_{12}$ and ${\alpha}_{13}$.
In order to see the relative importance of the different channels we have evaluated the contribution to the total cross section to obtain $Z^0Z^0$ and $W^{\pm}Z^0$ pairs in $pp$ colliders from the different considered subprocesses in a typical example. We have chosen the ${\alpha}_k$ parameters that mimic the $MSM$ with a heavy scalar Higgs whose mass is $m_H=1 TeV$ . The rates for the contributions of the different subprocesses to the total $Z^0Z^0$ final state, $q\bar{q}$, $gg$, $Z^0Z^0$ and $W^+W^-$ are, respectively: $45$, $17$, $3$, $35\%$ and to the total $W^{\pm}Z^0$ the rates are $61$, $22$, $17\%$ coming from $q\bar{q'}$, $W^{\pm}Z^0$ and $W^{\pm} \gamma$ fusion.
As it has been discussed above, the essential point in our calculations is that we have not used the $ET$ [@ET; @ETCL] to obtain the scattering amplitudes of the gauge bosons as it is customary. We have calculated the corresponding tree level amplitudes for the gauge bosons and projected them into their transversal or longitudinal components. Therefore we do not have the limitations that appear when the $ET$ and the chiral lagrangian formulation of the $ESB$ are used together [@ET]. The chiral lagrangian approach provides a low-energy description of the $GB$ dynamics as an expansion on the momenta over $4 \pi v$ and the $ET$ refers to the large energy relation between $GB$ and the longitudinally polarized gauge bosons $S$ matrix elements. In our case we only need to fix an upper energy bound (as it was said $E_{\rm
max}=1.5 TeV$) so that we can safely apply $\chi PT$.
However, the $ET$ can be used as a helpful tool to test our calculated longitudinally polarized amplitudes. In order to check our results we have compared our tree level expressions with those calculated applying the $ET$ to the corresponding Goldstone boson amplitudes. We have found both amplitudes to agree at lowest order in $g$ and $g'$ for the cases $Z^0_LZ^0_L \rightarrow Z^0_LZ^0_L$, $W^+_LW^-_L \rightarrow Z^0_LZ^0_L$ and $W^{\pm}_LZ^0_L \rightarrow
W^{\pm}_LZ^0_L$ as it was expected according to the results of [@ETCL]. However, the details of this comparison will be described in detail elsewhere [@GLASGOW] since they concern the applicability of the $ET$ which has not been used in the present computation. Here we only wanted to quote that our results are compatible with the information that one could have obtained from the $ET$ on the above mentioned processes.
Signatures and background
=========================
As it was stressed in the previous section we have calculated the tree level scattering amplitudes, up to order $p^4$, corresponding to all the helicity states of the gauge bosons, without using the $ET$. This fact allows us to include the dependence on the ${\alpha}_k$ parameters in all the helicity amplitudes. In contrast, when the $ET$ is used, the only channel where the ${\alpha}_k$ coefficients are taken into account is the following:
$$V_LV_L \rightarrow V_LV_L \;\;\; (V=Z^0,W^{\pm})$$
The definition of this channel as the signal was considered appropriate because the $V_LV_L \rightarrow V_LV_L$ channels are expected to be strongly interacting at high energies, if the $GB$ are, due to the $ET$. However, our computation makes it possible for the first time to study the effect of a strongly interacting ESB in other gauge boson polarization states by means of the ${\alpha}_k$ parameters. In fact, the aim of this work is to see how measurable will these parameters be at the $LHC$. In order to make this point more precise we must define the statistical significance corresponding to some given value of the chiral parameters ${\alpha}_k$. With this goal in mind we define our signal and our background as:
$$\begin{aligned}
({\rm Signal}) \;\;\;\;\;\; n_S & = & |N(\{ {\alpha} \}) -
N(\{ {\alpha}^0 \})|
\label{eq:senal}\end{aligned}$$
$$\begin{aligned}
({\rm Background}) \;\;\; n_B & = & N( \{ {\alpha}^0 \})
\label{eq:ruido}\end{aligned}$$
where $N\{ {\alpha} \}$, $N\{ {\alpha}^0 \}$ are the total number of $Z^0Z^0$ or $W^{\pm}Z^0$ pairs obtained for some given experimental cuts when the chiral parameters have been set to the values $\{ {\alpha} \}$ or $\{
{\alpha}^0 \}$. The background is defined in terms of some reference model $\{
{\alpha}^0 \}$. For simplicity this model has been taken as the one with all the parameters set to zero i.e. $ \alpha^0_k =0$. We call it [*Zero Model*]{} and, incidentally, it corresponds to the $MSM$ with an infinite Higgs mass.
For the final state $Z^0Z^0$ we have considered the following processes that contribute to the signal and the background:
$$\begin{aligned}
Z^0Z^0 \rightarrow Z^0Z^0 \\
W^+W^- \rightarrow Z^0Z^0\end{aligned}$$
In addition we include the $Z^0Z^0$ production via gluon fusion and $q\bar{q}$ annihilation which do not depend on ${\alpha}_k$ and therefore contribute only to the background. The experimental signature for this process consists of four leptons as the result of the $Z^0Z^0$ decays. This gives a clean and distinct signal because the $Z^0Z^0$ pairs can be fully reconstructed. The disadvantage is the rather small leptonic branching ratio ($0.44\%$).
In the $W^{\pm}Z^0$ case we have considered the channels:
$$\begin{aligned}
W^{\pm}Z^0 \rightarrow W^{\pm}Z^0 \\
W^{\pm}\gamma \rightarrow W^{\pm}Z^0 \\
q\bar{q'} \rightarrow W^{\pm}Z^0\end{aligned}$$
We focus our attention on the gold-plated events where the $W^{\pm}$ and $Z^0$ decay to the charged leptonic final states ($l=e,\mu,\nu$). The corresponding branching ratio is $1.3\%$. All these cross sections depend on the chiral ${\alpha}_k$ parameters so that they are taken into account in the signature and background calculations.
The main source of $Z^0Z^0$ or $W^{\pm}Z^0$ pairs in $pp$ colliders is via quark-antiquark annihilation. The total production rate of $gg \rightarrow Z^0Z^0$, in the studied cases, is $20-50\%$ than that from $q{\bar{q}} \rightarrow Z^0Z^0$ depending on the top quark mass (we have chosen $m_t=170\,GeV$). On the other hand, the production rates of $Z^0Z^0$ or $W^{\pm}Z^0$, via gauge boson fusion are suppressed by powers of (${\alpha}/ {\rm sin}^2 {\theta}_w$) due to the application of the $W$ effective approximation [@fegamm; @Daw] to obtain the initial bosons from $pp$ beams, as we will see in next section.
The proton and gauge boson structure functions
==============================================
Here we describe how to compute the total cross sections of the different processes studied to obtain $Z^0Z^0$ or $W^{\pm}Z^0$ pairs in $pp$ colliders.
$$\begin{aligned}
pp & \rightarrow & (q\bar{q} \rightarrow V_3V_4) + X \\
pp & \rightarrow & (gg \rightarrow Z^0Z^0) + X \\
pp & \rightarrow & (V_1V_2 \rightarrow V_3V_4) + X\end{aligned}$$
We have to integrate the differential cross section for the subprocess, $(d{\sigma}/d {\rm cos} {\theta})$, with the distribution functions of the quark, antiquark and gluon (given by $f_i$, $f_j$ and $g$) inside the proton:
$$\begin{aligned}
\sigma (pp \rightarrow (q\bar{q'} \rightarrow V_3V_4) + X) & = &
\sum_{i,j} \int\int\int dx_1dx_2 dcos\theta f_i(x_1,Q^2)f_j(x_2,Q^2)
\nonumber \\
& &
\frac{d \hat{\sigma}}{d cos{\theta}}(q\bar{q'} \rightarrow V_3V_4)
\nonumber \\
\sigma (pp \rightarrow (gg \rightarrow Z^0Z^0) + X) & = &
\int\int\int dx_1dx_2 dcos\theta g(x_1,Q^2)g(x_2,Q^2)
\nonumber \\
& &
\frac{d \hat{\sigma}}{d cos{\theta}} (gg \rightarrow Z^0Z^0)
\label{eq:secc}\end{aligned}$$
These formulae are used to compute the processes:
$$\begin{aligned}
q\bar{q} & \rightarrow & Z^0Z^0 \\
q\bar{q'} & \rightarrow & W^{\pm}Z^0 \\
gg & \rightarrow & Z^0Z^0\end{aligned}$$
In order to compute the number of events of $Z^0Z^0$ and $W^{\pm}Z^0$ produced in $pp$ collisions via gauge boson fusion we apply the effective $W$ approximation [@fegamm; @Daw] and we use the formula:
$$\begin{aligned}
\sigma (pp \rightarrow (V_1V_2 \rightarrow V_3V_4) + X) & = &
\sum_{i,j} \int\int dx_1dx_2 dcos\theta f_i(x_1,Q^2)f_j(x_2,Q^2)
\nonumber \\
& & \int\int d\hat{\tau} d\hat{\eta} \frac{{\partial}^2
L}{{\partial}\hat{\tau}{\partial}\hat{\eta}}
\frac{d \hat{\sigma}}{d cos{\theta}} (V_1V_2 \rightarrow V_3V_4)
\label{eq:secc2}\end{aligned}$$
Thus the total cross section in $pp$ colliders (like $LHC$) can be written as the result of the convolution of the subprocess cross section with the $V_1V_2$ pair luminosity in $pp$ beams. This luminosity is calculated from the convolution of the double bremsstrahlung of the $(V_1V_2)$ from the quark structure functions. Thus, $\partial^2 L/\partial\hat{\tau}\partial\hat{\eta}$ is the luminosity function for the gauge boson pair $V_1^h V_2^h$ to be radiated from the quark pair $q_iq_j$. It depends on the helicity state, transversal or longitudinal, of the initial bosons $V_1$ and $V_2$. Therefore we have to separate in our computations the contribution of the different polarization channels of the initial bosons.
$$\begin{aligned}
Z^0Z^0 & \rightarrow & Z^0Z^0 \\
W^+W^- & \rightarrow & Z^0Z^0 \\
W^{\pm}Z^0 & \rightarrow & W^{\pm}Z^0 \\
W^{\pm}\gamma & \rightarrow & W^{\pm}Z^0\end{aligned}$$
The amplitudes and differential cross sections for all these processes have been obtained as it was described in the previous sections. According to the effective $W$ approach we take the following functions corresponding to the probability of a gauge boson $V_T$ or $V_L$ ($V=W^{\pm}, Z^0$) or a photon, to be radiated from the quark $q$, with a momentum fraction of the quark, $x$.
$$\begin{aligned}
f_{q/V^T}(x) & = & f_V \frac{x^2+2(1-x)}{2x} {\rm ln}
\left( \frac{E^2}{M^2_V} \right) \;\;\;\;\; V=W,Z \nonumber \\
f_{q/V^L}(x) & = & f_V \frac{1-x}{x} \nonumber \\
f_{q/ \gamma}(x) & = & \frac{\alpha}{2\pi} {\epsilon}^2_q
\frac{1+(1-x)^2}{x} {\rm ln} \left( \frac{E^2}{m^2_q} \right)
\label{eq:Wef}\end{aligned}$$
where the values of the $f_{{V_i}'s}$ depend on the particular gauge boson as well as on the type of quark that it comes from
$$\begin{aligned}
f_W(x) & = & \frac{\alpha}{4 \pi x_W} \\
f_{Z_{u\bar{u}}}(x) & = & \frac{\alpha}{16 \pi x_W (1-x_W)}
\left[ 1+(1-\frac{8}{3} x_W)^2 \right] \\
f_{Z_{d\bar{d}}}(x) & = & \frac{\alpha}{16 \pi x_W (1-x_W)}
\left[ 1+(1-\frac{4}{3} x_W)^2 \right]\end{aligned}$$
where $x_W={\rm sin}^2 {\theta}_W$.
The variables $\tau$ and $\eta$ are related to the momentum fractions of the quarks by $x_{1,2}=\sqrt{\tau} e^{{\pm}\eta}$. The connection between the variables $\hat{\tau}$, $\hat{\eta}$ and the momentum fractions of $V_1$, $V_2$ respect to $q_i$, $q_j$, $\hat{x_1}$ and $\hat{x_2}$ is given by $\hat{x_{1,2}}=\sqrt{\hat{\tau}}e^{\pm {\hat{\eta}}}$. Note that the rate of transversally polarized gauge bosons obtained is enhanced by the logarithmic factors eq.(\[eq:Wef\]) with respect to the longitudinal gauge boson production.
The structure functions of quarks and gluons we use are those of $EHLQ$ [@EHLQ], set $II$ with $\Lambda = 290 MeV$ and we neglect the contribution of the top quark to the proton sea. However we have also studied the effect of changing the structure function on our results (see below). The assignment for the $Q^2$ appearing in the distribution functions is $Q^2 =
\hat{s}$ for $q\bar{q'}$ and $gg$ processes, and $Q^2=M^2_W$ for elastic gauge bosons scattering processes. The resulting integrals are computed using the $VEGAS$ Monte Carlo program [@Vegas].
With the described machinery we have built up a big code which computes the total number of expected gold-plated events (for some given $LHC$ integrated luminosity) in terms of the chiral parameters $\alpha_k$ and the final state cuts .
Initially we impose the following cuts on the invariant mass of the weak boson pair, the transverse momentum of the final $Z^0$ boson and the rapidities of both bosons:
$$\begin{array}{lclll}
200 GeV & \leq & \sqrt{\hat{s}} & \leq &
1500 GeV \nonumber \\
{p_T}_Z & \geq & 10 GeV & &
\nonumber \\
| y_{1_{\rm max}} | & = & | y_{2_{\rm max}} | & = & 2.5
\end{array}$$
From now on, we will call them the minimal cuts. The upper limit in the invariant mass ensures the validity of $\chi PT$ approach. For the rest of the paper we compute the number of $Z^0Z^0$ and $W^{\pm}Z^0$ obtained in the $LHC$ with an integrated luminosity of $3 \times 10^5
pb^{-1}$. This approximately corresponds to a total $LHC$ working time of 1 year ($3\times 10^7 sec$) assuming a luminosity ${\cal L} = 10^{34} cm^{-2}s^{-1}$.
Parameter sensitivity, optimal cuts and measurable parameters
=============================================================
In previous sections we have presented in detail how we have computed the number of $Z^0Z^0$ or $W^{\pm}Z^0$ pairs obtained at the $LHC$. If we fix the integrated luminosity, the $pp$ center of mass frame energy for the $LHC$ ($\sqrt{s}=16TeV$), and the upper bound on the invariant mass at the subprocesses ($\sqrt{\hat{s}_{{\rm max}}}=1.5
TeV$), the total cross sections will depend on the chiral ${\alpha}_k$ parameters and on some kinematical cuts. For the sake of simplicity we will only study the subset of the ${\alpha}_k$ coefficients which is needed to reabsorb the one-loop divergences obtained calculating with the chiral Lagrangian to lowest order. They are ${\alpha}_0$, ${\alpha}_1$, ${\alpha}_2$, ${\alpha}_3$, ${\alpha}_4$ and ${\alpha}_5$ [@Long] (however, our code includes as well the contribution of the other chiral parameters). There is also the possibility to modify our results choosing different kinematical cuts, $c_l$ ($c_1 = \sqrt{\hat{s}_{\rm min}}$, $c_2 = {p_{TZ}}_{\rm
min}$ and $c_3 = y_{1_{\rm max}}=y_{2_{\rm max}}$). Thus we can define two vectors: ${\alpha}$ and $c$, as follows: $$\begin{aligned}
{\alpha} & = & ({\alpha}_0,{\alpha}_1,{\alpha}_2,{\alpha}_3,
{\alpha}_4,{\alpha}_5) \nonumber \\
c & = & (c_1,c_2,c_3)
\label{eq:vectors}\end{aligned}$$ which are inputs of our code; whereas the number of $Z^0Z^0$ ($i=1$) or $W^{\pm}Z^0$ ($i=2$) events is the output. Therefore we write the result of our computations as $N^{(i)}({\alpha};c)$. In principle one could expand these functions around the [*Zero Model*]{} (we defined it previously as $\{ \alpha^0_k=0 , \forall k \}$) and the minimal cuts defined above so that one could write:
$$\begin{aligned}
N^{(i)} ({\alpha}; c) & = & N^{(i)} ({\alpha}^0; c^{\rm min}) +
\sum^{5}_{k=0} \frac{\partial N^{(i)}}{\partial {\alpha}_k}
{\mid}_{{\alpha}_k=0} {\alpha}_k + \sum^{3}_{l=1}
\frac{\partial N^{(i)}}{\partial c_l}{\mid}_{c_l=c^{\rm min}_l}
(c_l - c^{\rm min}_l) \nonumber \\
& & + O({\alpha}^2) + O({(c-c^{\rm min})}^2)\end{aligned}$$
If the dependence of the number of events on the parameters and the cuts were approximately linear, this formula could be used to compute $N^{(i)}({\alpha};c)$. However, we will see that this is not always the case. Indeed, $N^{(i)}({\alpha};c)$ is a polynomial in $\alpha_k$ and nonlinear terms can become important even for moderate values of these parameters.
In order to see how the number of events changes when one of the ${\alpha}_k$ parameters varies by some amount $\Delta {\alpha}_k$, we fix the kinematical bounds $c_l$ to the minimal cuts and set the other chiral parameters to zero. We choose the following set of values for $\Delta {\alpha}_k$:
$$\Delta {\alpha}_k = \pm 10^{-3}, \pm 5 \times 10^{-3}, \pm 10^{-2}$$
The result of our computations can be found in Figures \[fig:sena0\], to \[fig:sena5\]. We can also define the sensitivity function $s^{(i)}_k(c)$ associated to the ${\alpha}_k$ parameter, with the kinematical cuts $c$, and for the final state $i$, as follows:
$$s^{(i)}_k (c) \equiv \frac{\partial N^{(i)}
({\alpha};c)}{\partial {\alpha}_k} {\mid}_{{\alpha}_k=0}
\label{eq:sens}$$
The different sensitivity functions for the minimal cuts that we have obtained are displayed in Table \[tab:line\]. Obviously, the $|s^{(i)}_k
(c)|$ values are a measure of the variations of the number of events with ${\alpha}_k$. However, they are not a direct measure of the statistical significance of the corresponding parameter variation.
We can also study the number of $Z^0Z^0$ and $W^{\pm}Z^0$ event distributions with respect to the cuts given in the minimal invariant mass ($c_1$), the minimal transverse momentum of $Z^0$ in the final state ($c_2$) and in the maximal rapidity of the final bosons ($c_3$). In these computations we have fixed the ${\alpha}$ parameters at their values in the [*Zero Model*]{}, and the other two kinematical cuts have been set to their minimal values. The results obtained in this way are shown in Figures \[fig:mvcut\] to \[fig:yicut\].
In order to perform a statistical analysis, we define the $r_k$ function as:
$$\begin{array}{lclcl}
r^{(i)}_k (c) & = & \frac{| N^{(i)}({\alpha}_k; c)
- N^{(i)}({\alpha}^0; c) |}{\sqrt{N^{(i)}({\alpha}^0;c)}}
& = & \frac{n_S}{\sqrt{n_B}}
\end{array}
\label{eq:rk}$$
The $r^{(i)}_k$ function is a measure of the statistical significance of the signal (corresponding to the increments $\Delta {\alpha}_k$) relative to the background, provided $N^{(i)}({\alpha}_k; c)$ is large enough to apply the Central Limit Theorem. In this case $r_k$ is an estimate of the number of sigmas, and therefore it defines the confidence level for the hypothesis that ${\alpha}_k$ is different from zero. In order to evaluate $r^{(i)}_k$ we use the number of events that was previously obtained. The resulting $r^{(i)}_k$ functions are shown in Figures \[fig:siga0\] to \[fig:siga5\].
From Figures \[fig:mvcut\] and \[fig:ptcut\], it is clear that there is no linear dependence of $N^{(i)}({\alpha}; c)$ on $c_1$ and $c_2$. Moreover, looking at Figures \[fig:sena0\] at \[fig:sena5\] and from Figures \[fig:siga0\] at \[fig:siga5\], nonlinear behaviour is observed with respect to ${\alpha}_1$, ${\alpha}_2$ and ${\alpha}_3$ (for $W^{\pm}Z^0$ final state) nor when $\mid
{\alpha}_4 \mid$ or $\mid {\alpha}_5 \mid$ are bigger than $0.001$ in both channels, $Z^0Z^0$ and $W^{\pm}Z^0$.
By looking at Figures \[fig:siga0\] to \[fig:siga5\], we can see that the slope corresponding to the different curves $r^{(i)}_k(c^{\rm min})$ varies according to the different values of ${\alpha}_k$ and the channel ($Z^0Z^0$ or $W^{\pm}Z^0$). Indeed, the greater is the slope, the higher is the statistical significance. Moreover it is already possible to observe that some of the ${\alpha}$ parameters have no chances to be measured at the $LHC$, at least in the way described here. For this reason, in the following we will concentrate our statistical analysis on the set of ${\alpha}$ parameters which can be considered as [*potentially measurable*]{}. Being more precise, we define a parameter $\alpha_k$ as [*potentially measurable*]{} in a given channel when the $r^{(i)}_k(c^{\rm min})$ corresponding to the maximal considered variation ($\mid \Delta {\alpha}_k \mid = 0.01$), is bigger than $0.5$. According to this we find that the [*potentially measurable*]{} ${\alpha}_k$ are the following:
- ${\alpha}_3<0$ and ${\alpha}_3>0$ (both in $Z^0Z^0$ and $W^{\pm}Z^0$ channels)
- ${\alpha}_4<0$ ($W^{\pm}Z^0$) and ${\alpha}_4>0$ ($Z^0Z^0$)
- ${\alpha}_5<0$ ($W^{\pm}Z^0$) and ${\alpha}_5>0$ ($Z^0Z^0$)
In the following we carry out an statistical analysis with all these [*potentially measurable*]{} parameters in the corresponding final state. To do so, the next step is to look for the optimal cuts in the minimal invariant mass and transverse momentum for the detection of a certain $\Delta
{\alpha}_k = ({\alpha}_k-{\alpha}^0_k)$ different from zero. Initially we impose the minimal cuts on the kinematical variables. The optimization procedure we use is to search for the pair of cuts ($\sqrt{\hat{s}}$, $p^c_{TZ}$) in the kinematical region
$$\begin{aligned}
200 GeV \;\; \leq & \sqrt{\hat{s}} & \leq \;\; 1500 GeV \nonumber \\
10 GeV \;\; \leq & p_{TZ} & \leq \;\; \sqrt{\hat{s}_{\rm max}/4-m^2_z}
\label{eq:lig}\end{aligned}$$
that maximize the function $r^{(i)}_k({\alpha}_k; \sqrt{\hat{s}^c}, p^c_{TZ})$ defined as follows:
$$r^{(i)}_k ({\alpha}_k; \sqrt{\hat{s}}^c, p^c_{TZ}) = \frac{ | N^{(i)}
({\alpha}_k; \sqrt{\hat{s}} > \sqrt{\hat{s}^c}, p_{TZ}>p^c_{TZ}) -
N^{(i)} ({\alpha}^0_k; \sqrt{\hat{s}} > \sqrt{\hat{s}^c},
p_{TZ}>p^c_{TZ}) |}
{\sqrt{N^{(i)} ({\alpha}^0_k; \sqrt{\hat{s}} > \sqrt{\hat{s}^c},
p_T>p^c_T)}}
\label{eq:opti}$$
To find the optimal cuts for the different parameters and channels we build a bidimensional grid on the plane ($\sqrt{\hat{s}},p_{TZ}$) so that the different points are separated from each other by the increments ($\Delta
\sqrt{\hat{s}} =
50 GeV$, $\Delta p_T = 50 GeV$). For each pair of points in this grid ($\sqrt{\hat{s}^c}$, $p^c_{TZ}$) we compute the total number of events in the [*Zero Model*]{} and in a $ESB$ scenario corresponding to a certain positive or negative $\Delta{\alpha}_k$. Finally, we evaluate the $r^{(i)}_k ({\alpha}_k; \sqrt{\hat{s}^c}, p^c_{TZ})$ function. The optimal cuts, $c^{\rm op}$, are those which maximize eq.(\[eq:opti\]), and therefore, the confidence level. Of course, these $c^{\rm op}$ which have been found in this way depend slightly on the particular choice of $\Delta{\alpha}_k$ for each ${\alpha}_k$ parameter. In our computation we have taken the typical values:
$$\Delta {\alpha}_k = {\alpha}_k - {\alpha}^0_k = {\alpha}_k = \pm 0.005$$
The results are collected in Tables \[tab:opti1\] and \[tab:opti2\]. There it is displayed, for each $\Delta{\alpha}_k$ in the corresponding channel, the optimal cuts, the number of events that satisfies these cuts and the statistical significance function $r_k$ obtained with the minimal and the optimal cuts. In all cases we have fixed a maximal rapidity of $2.5$.
The $c^{\rm op}$ we have obtained by maximizing the $r_k^{(i)}$ function eq.(\[eq:opti\]) will be considered from now on as the optimal cuts to detect either a $\Delta{\alpha}_k < 0$ or a $\Delta{\alpha}_k > 0$. For example, if we wanted to observe the signature corresponding to a heavy Higgs $SM$-like $ESB$ scenario, the best choice would be the $Z^0Z^0$ final state with the optimal cuts obtained for $\Delta{\alpha}_5>0$ ($\sqrt{\hat{s}_{\rm min}} = 1150 GeV$, $p_{TZ{\rm min}} = 400 GeV$). This is so because, at the tree level order, the only ${\alpha}_k$ different from zero needed to mimic this scenario is ${\alpha}_5$ (${\alpha}_5=v^2/(8m^2_H)$).
As it can be seen in Table \[tab:opti2\], the optimization procedure has clearly improved the statistical significance $r_k$ in most of the studied cases. Only when we tried to optimize the signature corresponding to a ${\alpha}_3>0$ and ${\alpha}_3<0$, both in the $Z^0Z^0$ final state, we obtained no significant improvement with respect to the minimal cuts. Therefore we find it impossible to detect a signature of a negative or positive ${\alpha}_3$ in $Z^0Z^0$ events.
Finally, we have carried out a similar optimization procedure for $\alpha_4$ and $\alpha_5$ in which we searched for a slightly different pair of optimal cuts: ($\sqrt{\hat{s}_{\rm max}}$, ${p_{TZ}}_{\rm min}$). We obtained $c^{\rm op}$ = ($1500 GeV$, $500 GeV$) for ${\alpha}_4=-0.005$ ($W^{\pm}Z^0$), ${\alpha}_4=0.005$ ($Z^0Z^0$) and ${\alpha}_5=0.005$ ($Z^0Z^0$) and the $r_k$ function was respectively $6.02$, $1.56$ and $3.82$. On the other hand, we also found the optimal cuts to detect ${\alpha}_5=-0.005$ in the $W^{\pm}Z^0$ final state ($c^{\rm op}=(1500 GeV, 400 GeV)$) and the reached statistical significance was $2.97$. These results are at most equal to those of Table \[tab:opti2\].
Moreover, we have repeated our whole study to find other sets of optimal cuts belonging to the parameter space, such as ($\sqrt{\hat{s}_{\rm max}}$, ${p_{TZ}}_{\rm min}$), ($\sqrt{\hat{s}_{\rm max}}$, ${p_{TZ}}_{\rm
max}$), ($\sqrt{\hat{s}_{\rm min}}$, ${p_{TZ}}_{\rm max}$). In all these cases, the statistical significances obtained are smaller than those corresponding to the first choice of $c^{\rm op}$ ($c^{\rm
op}=(\sqrt{\hat{s}_{\rm min}}, {p_{TZ}}_{\rm min}$)). Therefore, in the following we will apply this type of optimal cuts.
As it was mentioned in the introduction the main aim of this work is to determine which chiral Lagrangian parameters will be more easily measurable at the $LHC$ and the size of their corresponding statistical errors. Initially, we made a criterion to choose a set of [*potentially measurable*]{} ${\alpha}_k$ parameters, on which we have concentrated our analysis. After carrying out the described optimization procedure, we conclude that the statistical significance to measure most of these chiral Lagrangian coefficients, has clearly increased. In the following we will treat only these [*measurable*]{} ${\alpha}_k$. The set of the ${\alpha}_k$ parameters obtained as the [*measurable*]{} ones are contained in Tables \[tab:opti1\] and \[tab:opti2\]. Here we exclude the study of ${\alpha}_3$ in the $Z^0Z^0$ final state, as it was previously argued.
In order to obtain an estimation of the statistical errors ${(\Delta
{\alpha}_k)}_{\rm stat.}$ we will work under some hypothesis. We will assume a linear behaviour in ${\alpha}_k$ and that the number of events is large enough to consider $N^{(i)}(\alpha;c)$ following a gaussian distribution.
At this point, we define the statistical error ${(\Delta
{\alpha}_k)}_{\rm stat.}$ associated to a certain ${\alpha}_k$ as the minimal value of this parameter that could be detected in the $LHC$ with a statistical significance, eq.(\[eq:rk\]), equal to one sigma, after applying its corresponding optimal cuts. Using the results displayed in Table \[tab:opti2\] which satisfy the optimal cuts, we obtain the ${(\Delta {\alpha}_k)}_{\rm stat.}$ in this way, and we show the results in Table \[tab:error\]. As it could be expected, the smallest ${(\Delta {\alpha}_k)}_{\rm stat.}$ value corresponds to the biggest statistical significance, which is ${(\Delta {\alpha}_4)}_{\rm stat.}=8.31 \times 10^{-4}$, in the $W^{\pm}Z^0$ final state.
The effect of the structure function indetermination
====================================================
In all the previous calculations we have used the $EHLQ$ structure functions [@EHLQ] (set $II$ with $\Lambda=290 MeV$) to obtain the total number of $Z^0Z^0$ and $W^{\pm}Z^0$ events. Here we wonder how our results could be affected by the indetermination in the parton distribution functions. In order to estimate this effect we have selected two new parametrizations of the structure function: the set $MRSD-$ (Martin, Roberts and Stirling [@MRS]) and the $GRVHO$ (Glück, Reya and Vogt [@GRV]). Both them exhibit a similar behaviour of sea quarks and gluon distributions at low $x$, that grow when $x \rightarrow 0$. The values of $\Lambda$ considered in these two sets are the following: ${\Lambda}_{\rm MRS}=230 MeV$, ${\Lambda}_{\rm GRV}=200 MeV$. With these new structure functions we evaluate the total number of $Z^0Z^0$ or $W^{\pm}Z^0$ events, in the [*Zero Model*]{} with the different optimal cuts. The way we proceed is to compare our estimation of the number of standard deviations obtained by changing these parton density function sets (we will call it $r^{(i)}_{\rm struc.}$), with the statistical significance corresponding to a certain signal, after having applied the optimal cuts (the $r^{(i)}_k(c^{\rm op}$) functions shown in Table \[tab:opti2\]). We define $r^{(i)}_{\rm struc.}$ similarly as in eq.(\[eq:rk\]):
$$r^{(i)}_{\rm struc.}(c) = \frac{| N_{set}^{(i)}({\alpha}^0; c)
- N_{set'}^{(i)}({\alpha}^0; c) |}
{\sqrt{N_{set}^{(i)}({\alpha}^0;c)}}
\label{eq:rstruc}$$
Now, to obtain $r^{(i)}_{\rm struc.}$ we need to evaluate the total number of events in the [*Zero Model*]{}, with certain kinematical cuts, $c$, but using two different parametrizations called [*set*]{} and [*set’*]{} for the structure functions.
In Table \[tab:festru\] we show the results of our calculations. We represent the total number of $Z^0Z^0$ and $W^{\pm}Z^0$ events obtained for the optimal cuts, when the $EHLQ$ (set II), $MRSD-$ and $GRVHO$ sets are used, as well as the number of sigmas $r_{\rm struc.}$ eq.(\[eq:rstruc\]) corresponding to change one of these sets with respect to the others:
- $r_{\rm struc.} \equiv r_{12}$ ($EHLQ$ with respect to $MRSD-$)
- $r_{\rm struc.} \equiv r_{13}$ ($EHLQ$ with respect to $GRVHO$)
- $r_{\rm struc.} \equiv r_{23}$ ($MRSD-$ with respect to $GRVHO$).
Now, we pay attention to the $r_k$ functions in Table \[tab:opti2\] and $r_{\rm struc.}$ in Table \[tab:festru\] and we compair them. We will say that the typical chosen value of ${\alpha}_k$ (${\alpha}_k=-0.005$ or ${\alpha}_k=0.005$) can be measured in the $LHC$ with a statistical significance given by $r_k$, if the following inequality is verified:
$$r_k >> r_{\rm struc.}
\label{eq:ineq}$$
When we make $r_{\rm struc.}=r_{12}$ or $r_{\rm struc.}=r_{13}$, we must reject the value ${\alpha}_3=0.005$ in $Z^0Z^0$ and $W^{\pm}Z^0$ channels, and ${\alpha}_3=-0.005$ in $Z^0Z^0$ final state, because relation (\[eq:ineq\]) is violated. However, if we take $r_{\rm struc.}=r_{23}$ we can measure positive and negative values of ${\alpha}_3$, ${\alpha}_4$ and ${\alpha}_5$ parameters in some $Z^0Z^0$ or $W^{\pm}Z^0$ final state, with a statistical significance given by $r_k$. On the opposite, for ${\alpha}_3=\pm 0.005$ in $Z^0Z^0$ final state, eq.(\[eq:ineq\]) is not fulfilled, so we can only measure these values of ${\alpha}_3$ through the $W^{\pm}Z^0$ channel. This last choice ($r_{\rm struc.}=r_{23}$) is most appropriate since $MRSD-$ and $GRVHO$ agree much better with some recent experimental results [@Terron] obtained at $HERA$. In any case, it can be expected that future new experimental data coming from $HERA$ and even from the $LHC$ itself, will allow us to reduce the size of these uncertainties $r_{\rm struc.}$.
Renormalization effects
=======================
The other effect we are going to take into account is the dependence of the ${\alpha}_k$ parameters and the couplings $g$ and $g'$, on the energy. Our whole previous study was carried out at tree level order. Now we want to include some quantum effects by means of the Renormalization Group Equations ($RGE$). By that we will understand the one loop contributions coming from ${\cal L}^{(2)}$ and ${\cal L}_{\rm YM}$ plus the tree level contributions coming from ${\cal L}'_1 + {\cal L}^{(4)}$ to the corresponding beta functions.
As it is well known from the $RGE$, the renormalized Green function with the renormalized parameters ${\lambda}$ and $m$ at the renormalization scale $\mu$ and the same Green function with the parameters at another scale $\mu e^t$, are related by the equation:
$$G^{(n)}_R(p_i,\lambda,\mu) = G^{(n)}_R(p_i,\bar{\lambda}(t),
\mu e^t) {\rm exp} [-n \int^t_0 \gamma(\bar{\lambda}(t')) dt']
\label{eq:EGR}$$
Where we include in ${\lambda}$ all the ${\alpha}_k$ parameters and the electroweak coupling constants, $g$ and $g'$. Thus the renormalized parameters are scale dependent. The $\bar{\lambda}$ stands for all the running coupling constants that depend on the renormalization scale. In order to obtain these $\bar{\lambda}$ we need to calculate the $\beta({\lambda})$ and $\gamma({\lambda})$ functions, following the well known $RGE$ techniques:
$$\begin{aligned}
\beta (\bar{\lambda}) & = & \frac{\partial \bar{\lambda}(\lambda,t)}
{\partial t} \nonumber \\
\gamma (\bar{\lambda}(t)) & = & \frac{1}{2} \frac{\partial {\rm log}
Z_{3V}}{\partial t}
\label{eq:begam}\end{aligned}$$
At present, when using the effective theory to describe the $ESB$ sector we only know these functions to one loop-order. The exponential factor in eq.(\[eq:EGR\]) is the anomalous dimension term, that depends on the wave function renormalization. We have estimated the order of magnitude of this effect to calculate the statistical significance $r_k$, in a typical case, with a subprocess energy of $1TeV$. We obtained changes on $r_k$ of $0.1\%$. Therefore they are completely irrelevant. Thus we neglect this contribution and we will only consider the dependence on the renormalization scale, $\mu$, of ${\alpha}_k$, $g$ and $g'$, in eq.(\[eq:EGR\]). We refer to this approach as the tree level approximation improved by the $RGE$.
Now, we have to calculate the running coupling constants ${\alpha}_k(\mu)$, $g(\mu)$ and $g'(\mu)$, and replace their tree level values in our cross section formulae by the corresponding energy dependent running coupling constants evaluated at the subprocess center of mass energy (we have taken $\mu=\sqrt{\hat{s}}$).
In order to do so we begin with the ${\alpha}_k$ parameters and we use the dimensional regularization scheme since it is the most appropriate for gauge theories as well as for non-linear sigma models. In [@Long], Longhitano obtained all the divergences that appear in one-loop diagrams calculated with the effective lagrangian at lowest order (${\cal L}^{(2)}+{\cal L}_{\rm YM}$). They could be absorbed by redefinitions of a subset of the ${\alpha}_k$ parameters ($k=0,1,2,3,4,5$). By means of the dependence of ${\alpha}_k$ on the $\epsilon$ parameter [@Long], it is easy to obtain the following ${\alpha}_k$ running expressions:
$$\begin{aligned}
{\alpha}^R_0(\mu) & = & {\alpha}^R_0({\mu}_0) - \frac{1}{{16 \pi}^2}
\frac{3}{4} {\rm tg}^2 {\theta}_{\rm w} {\rm
log} \left( \frac{{\mu}_0}{\mu} \right)
\nonumber \\
{\alpha}^R_1(\mu) & = & {\alpha}^R_1({\mu}_0) - \frac{1}{{16 \pi}^2}
\frac{1}{6} {\rm tg}{\theta}_{\rm w} {\rm log}
\left( \frac{{\mu}_0}{\mu} \right)
\nonumber \\
{\alpha}^R_2(\mu) & = & {\alpha}^R_2({\mu}_0) - \frac{1}{{16 \pi}^2}
\frac{1}{12} {\rm tg}{\theta}_{\rm w} {\rm log}
\left( \frac{{\mu}_0}{\mu} \right)
\nonumber \\
{\alpha}^R_3(\mu) & = & {\alpha}^R_3({\mu}_0) - \frac{1}{{16 \pi}^2}
\frac{1}{12} {\rm log} \left( \frac{{\mu}_0}{\mu} \right) \nonumber
\\
{\alpha}^R_4(\mu) & = & {\alpha}^R_4({\mu}_0) + \frac{1}{{16 \pi}^2}
\frac{1}{6} {\rm log} \left( \frac{{\mu}_0}{\mu} \right) \nonumber
\\
{\alpha}^R_5(\mu) & = & {\alpha}^R_5({\mu}_0) + \frac{1}{{16 \pi}^2}
\frac{1}{12} {\rm log} \left( \frac{{\mu}_0}{\mu} \right)
\label{eq:arun}\end{aligned}$$
Apart from these equations, we need also to calculate the running of the gauge couplings $g_R({\mu})$ and $g'_R({\mu})$. As usual we have
$$\begin{aligned}
g_R & = & Z_3^{3/2} Z_1^{-1} g_0 \\
g'_R & = & {Z'}_3^{1/2} g'_0\end{aligned}$$
The only differences with respect to the linear model, given by $\delta
Z_3$ and $\delta Z'_3$ functions, come from the scalar sector contribution. In our case, they are
$$\begin{aligned}
\delta Z_3 & = & \frac{g^2}{16 {\pi}^2} \frac{1}{6 \epsilon} \\
\delta Z'_3 & = & \frac{g'^2}{16 {\pi}^2} \frac{1}{6 \epsilon}\end{aligned}$$
Taking into account loops of scalars, fermions, ghosts and gauge bosons coming from the rest of the $SM$ particles we arrive to the following values of the $\beta$ functions eq.(\[eq:begam\]), ${\beta}_g$ and ${\beta}_{g'}$:
$$\begin{aligned}
{\beta}_g & = & - C_g g^3 \nonumber \\
{\beta}_{g'} & = & - C_{g'} {g'}^3
\label{eq:betas}\end{aligned}$$
where
$$\begin{aligned}
C_g & = & \frac{1}{16 {\pi}^2} \frac{13}{4} \\
C_{g'} & = & - \frac{1}{16 {\pi}^2} \frac{27}{4}\end{aligned}$$
The eqs.(\[eq:betas\]) are standard corresponding for gauge theories. The first one leads to asymptotic freedom since $C_g>0$, whereas the second corresponds to an abelian gauge theory. Therefore one has:
$$\begin{aligned}
g_R^2(\mu) & = & \frac { g_R^2(\mu_0) } {1 + \frac{1}{16 \pi^2}
\frac{13}{2} g^2_R(\mu_0) {\rm log} \frac{\mu}{\mu_0} } \nonumber \\
{g'}_R^2(\mu) & = & \frac{ {g'}_R^2(\mu_0) } {1 - \frac{1}{16
\pi^2}
\frac{27}{2} {g'}^2_R(\mu_0) {\rm log} \frac{\mu}{\mu_0} }
\label{eq:gsR}\end{aligned}$$
Once we have obtained the expression giving the dependence of the ${\alpha}_k$, $g$ and $g'$ on the renormalization scale eq.(\[eq:gsR\]), we can study cuantitatively this effect. The way we proceed is the following:
- First we factorize the $g$ and $g'$ coupling constants so that the cross sections appear with the same power in the electroweak couplings than in the $MSM$. We are referring only to the dominant terms. Moreover, as we can see in the Appendix, eqs.(\[eq:GF2\], \[eq:GF3\] and \[eq:GF4\]), higher powers on these coupling constants appear as factors of the ${\alpha}_k$ parameters.
- In all our calculations, for the sake of simplicity, we are considering the physical masses [@datos]: $m_W=80.6 GeV$, $m_Z= 91.1GeV$ and $m_q=0$ (except to in $gg \rightarrow
Z^0Z^0$ process where we have taken a top quark in the loop with a mass of $170 GeV$).
- We replace in the cross section the constant values of ${\alpha}_k$, $g$ and $g'$ by those which are renormalized and scale dependent, ${\alpha}^R_k(\mu)$, $g_R(\mu)$ and $g'_R(\mu)$ given by eqs.(\[eq:arun\] and \[eq:gsR\]). We take $\mu=\sqrt{\hat{s}}$, where $\sqrt{\hat{s}}$ is the invariant mass (the center of mass energy in the considered subprocess) and ${\mu}_0=m_Z$. Thus we have substituted the tree level values of ${\alpha}_k$, $g$ and $g'$ by the running coupling constants depending on the energy scale of the subprocess.
The assignment for the numerical values of $g_R({\mu}_0)$, $g'_R({\mu}_0)$ and ${\alpha}^R_k({\mu}_0)$ parameters appearing in our formulas has been done as follows:
- The quantities $g_R$ and $g'_R$, at the chosen scale ${\mu}_0=m_Z$, have been taken from recent $LEP$ data [@datos].
- The symmetry breaking pattern that we have considered corresponds to neglecting all the ${\alpha}^R_k$ parameters at the ${\mu}_0$ scale. We refer to this scenario as the [*Running Zero Model*]{}:
$$\begin{array}{lccl}
{\alpha}_k ({\mu}_0) & = & 0 & \;\; (k=0,1,2,3,4,5) \\
{\alpha}_k & = & 0 & \;\; (k=6,7,8,9,10,11,12,13)
\end{array}$$
After all these changes in the program, we apply the different sets of optimal cuts previously obtained, (Table 3) and calculate the total number of events to produce $Z^0Z^0$ or $W^{\pm}Z^0$. The corresponding results are displayed in Table \[tab:run\]. We present for each pair of optimal cuts, the total number of $Z^0Z^0$ or $W^{\pm}Z^0$ events, in the [*Zero Model*]{} and with $g$ and $g'$ constant ($N^{(i)}({\alpha}^0; g,g',c)$), and in the [*Running Zero Model*]{} with $g_R(\sqrt{\hat{s}})$ and $g'_R(\sqrt{\hat{s}})$ ($N^{(i)}({\alpha}^0(\sqrt{\hat{s}});
g_R(\sqrt{\hat{s}}), g'_R(\sqrt{\hat{s}}),
c)$). We also give an estimation of the statistical significance or the number of sigmas, $r^{(i)}_{\rm run.}$, corresponding to include the dependence on the energy of the parameters, with respect to the tree level results. Here, we calculate $r^{(i)}_{\rm run.}$ function analogously to eqs.(\[eq:rk\]) and (\[eq:rstruc\])
$$r^{(i)}_{\rm run.}(c) = \frac{| N^{(i)}({\alpha}^0; g,g',c)
- N^{(i)}({\alpha}^0(\sqrt{\hat{s}}); g(\sqrt{\hat{s}}),
g'(\sqrt{\hat{s}}),c) |}
{\sqrt{N^{(i)}({\alpha}^0; g,g',c)}}
\label{eq:rrun}$$
As it can be observed in Table \[tab:run\], the $r^{(i)}_{\rm run.}$ function, that reflects the relevance of this effect, is very high when we use the minimal cuts. Besides, with all the optimal cuts, the statistical significance function $r^{(i)}_{\rm run.}$ whose values go from $1$ to $4$ standard deviations is also important. Thus, the effect of the running of the couplings is important and it should be taken into account if one wants to compute the total number of $Z^0Z^0$ or $W^{\pm}Z^0$ events.
Now one could ask how the $r_k$ that were previously obtained (Table \[tab:opti2\]) would change if we used the improved tree level approximation with the running parameters, instead of working only at tree level order. In sight of the results contained in Table \[tab:run\] and if we supposed linear behaviour we would obtain fluctuations in the statistical significance functions, $r_k$, varying between $2$ to $15\%$ with respect to their values calculated without including the dependence on the energy of the parameters.
We can summarize these results saying that the inclusion of the running coupling constants in our calculations leads to important variations in the total number of events, given by $r^{(i)}_{\rm run.}$ functions. However, this effect is not so relevant when we want to determine the $r_k$ functions corresponding to each ${\alpha}_k$, since the differences obtained with respect to the tree level results are small. Therefore, the running of the couplings does not change significantly our previous discussion about which couplings will be [*measurable*]{} at the $LHC$.
Conclusions
===========
Using $\chi PT$ to describe the $ESB$ sector, we have studied the different processes contributing to $Z^0Z^0$ and $W^{\pm}Z^0$ final states at the $LHC$, considering only gold-plated events. The main novelty in our analysis is that we do not use the $ET$ but we have worked explicitly with all the the gauge boson polarization states. Thus we can study the low energy region, after having imposed a maximal bound on the subprocess energy of $1.5TeV$. We have elaborated a $FORTRAN$ code that generates the subprocess cross sections at tree level order, producing $Z^0Z^0$ or $W^{\pm}Z^0$ pairs.
With this code we have carried out a systematical study of the possibilities for measuring the chiral parameters ${\alpha}_k$, at the $LHC$. This analysis includes an optimization procedure, that makes possible to obtain the greatest statistical significance for measuring certain values of these parameters. We have also computed their corresponding statistical errors, $(\Delta\alpha_k)_{\rm stat.}$, and set the minimum values of these ${\alpha}_k$ to be unambiguously detected with a certain statistical significance at the $LHC$. The results of our studies are displayed in Tables \[tab:opti2\] and \[tab:error\]. From our results it is clear that the ${\alpha}_0$, $\alpha_1$ and $\alpha_2$ parameters cannot be probed at the $LHC$. On the contrary, as it can be seen in Tables \[tab:opti2\] and \[tab:error\], the $r_k$ function obtained for $\alpha_4$ and $\alpha_5$ can be, for some channels, greater than $3$ sigmas ($2$ sigmas for $\alpha_3$). In fact, the parameters that could be more easily measured in the $LHC$ are $\alpha_4 < 0$ in $W^{\pm}Z^0$ and $\alpha_5 > 0$ in $Z^0Z^0$ channel. As it can be expected their corresponding statistical errors are the smallest. In all our calculations, we have fixed a running time corresponding to one full $LHC$ year according to the nominal luminosity $L=10^{34}
cm^{-2} sec^{-1}$ ($L_{\rm int.}=3 \times 10^5 pb^{-1}$). The results corresponding to different integrated luminosities can be obtained just rescaling the statistical significance by the square root of the running time.
On the other hand, we have also estimated the size of the imprecisions in our calculations due to the indetermination in the structure functions by choosing three different distribution functions. The results (which can be found in Table \[tab:festru\]) show that the errors coming from this effect are smaller than the statistical errors obtained for the $\alpha$ parameters. Thus our present ignorance about the parton distribution functions does not restrict the measurement of the chiral parameters.
We have also tried to improve our tree level results considering the dependence on the subprocess energy of the weak couplings and the chiral parameters. The size of these effects is shown in Table \[tab:run\]. From our results it is clear that the total number of events changes when this effect is taken into account. However, our previous conclusions about which parameters can be probed at the $LHC$, do not change at all.
Moreover one could ask about the possibility of increasing the number of measurable chiral parameters by relaxing some of the experimental assumptions of this work. For instance we could also take into account the gauge boson hadronic decays, separate the final polarizations, etc... In that case we could certainly enhance the statistical significance and more chiral parameters could be probed. However it is not possible today to have any idea about how well this new information will be obtained at the $LHC$ and for this reason we have not considered such possibilities.
Finally we have also assumed that the $Z^0Z^0$ or $W^{\pm}Z^0$ pairs produced via electroweak gauge boson fusion cannot be separated experimentally from those coming from another sources such as $gg$ fusion (in $Z^0Z^0$ case) or quark-antiquark annihilation. Nevertheless some forward calorimeters could, presumably, be incorporated to the $LHC$ detectors. This fact could allow for a jet tagging at a certain level of efficiency. This experimental improvement is a realistic way to increase the statistical significance functions $r_k$ to measure the chiral Lagrangian parameters at $LHC$. Work is in progress to treat this possibility and to analyze how much the sensibility to the $\alpha_k$ parameters could be enhanced.
Acknowledgements
================
We thank M. J. Herrero for her patient help along the more than two years that this work took to be finished, as well as D. Denegri, T. Rodrigo, J. Terrón for some interesting discussions and J. R. Peláez for reading the manuscript. We also thank support by the Ministerio de Educación y Ciencia (Spain) (CICYT AEN90-0034, AEN93-0776 and AEN93-0673). A. D. thanks to the CERN Theory Division, where the final part of this work was done, for its kind hospitality.
Appendix {#appendix .unnumbered}
========
In this Appendix, we write some of the Feynman rules obtained from the quantized chiral electroweak Lagrangian ${\cal L}_{\rm NLSM}$ eq.(\[eq:MENL\]), described in section 2.
The different contributions to ${\cal L}_{\rm NLSM}$ are given in eqs.(\[eq:lag1\], \[eq:YM\], \[eq:lag2\] and \[eq:quant\]), where the Landau gauge has been chosen.
We show the Feynman rules corresponding to the propagators and vertices used in our calculations:
i\) The $\pi$ and $W^{\pm}$ propagators in the Landau gauge have the following expressions (see Figure \[fig:GF2\]):
$$\begin{aligned}
-i\Delta({\pi}^{\pm}) & = &
\frac{i}{k^2} \nonumber \\
%\frac{i}{k^2-\frac{4{\alpha}_{11}k^2}{v^2}} \nonumber \\
i\Delta_{\mu\nu}(W^{\pm}) & = & \frac{-i}{k^2-m_W^2+i\epsilon}
[g_{\mu\nu}-\frac{k_{\mu}k_{\nu}}{k^2}]
\label{eq:GF2}\end{aligned}$$
ii\) In the following, we write the Feynman rules corresponding to the vertices with either 3 gauge electroweak bosons (${\rm I}_{V_1V_2V_3}$) or to 2 gauge bosons and a $GB$ (${\rm
I}_{V_1V_2\pi}$) displayed in Figure \[fig:GF3\].
$$\begin{aligned}
{\rm I}_{\pi^+W^-_{\mu}Z^0_{\nu}} & = &
\frac{g^2 v}{2} {\rm sec}{\theta}_W g_{\mu\nu}({\rm sin}^2 {\theta}_W -
2g^2{\beta}_1) + \frac{2 g^2}{v} [(q.r)g_{\mu\nu} - q_{\mu}r_{\nu}]
({\alpha_1}{\rm sin}{\theta}_W
- {\alpha_8}{\rm cos}{\theta}_W + \nonumber \\
& & {\alpha}_{13}{\rm sec}{\theta}_W) + \frac{2g^2}{v}
{\alpha_3}{\rm sec}{\theta}_W
[(k.r)g_{\mu\nu} - k_{\mu}r_{\nu}] + \frac{2g^2}{v}
[(k.q)g_{\mu\nu}-q_{\mu}k_{\nu}]({\alpha}_2 {\rm sin}{\theta}_W -
\nonumber \\
& & ({\alpha}_3+{\alpha}_9){\rm cos}{\theta}_W) + \frac{2 g^2}{v}
{\alpha}_{11}
{\rm cos}{\theta}_W [k_{\mu}q_{\nu} - r_{\mu}k_{\nu}] -
\frac{2g^2}{v}{\alpha}_{11} {\rm sin}^2 {\theta}_W {\rm sec} {\theta}_W
[k^2g_{\mu\nu} - \nonumber \\
& & k_{\mu}q_{\nu}+r_{\mu}r_{\nu}] +
\frac{g^2}{v}{\alpha}_{12}{\rm sec}{\theta}_W
[k^2g_{\mu\nu}-k_{\mu}q_{\nu}+r_{\mu}r_{\nu}+q_{\mu}q_{\nu}] -
\frac{2g^2}{v} {\alpha}_{13}{\rm sec}{\theta}_W q_{\mu}q_{\nu}
\nonumber \\
{\rm I}_{\pi^+W^-_{\mu}A_{\nu}} & = &
-\frac{g^2 v}{2}{\rm sin}^2 {\theta}_W g_{\mu\nu} - \frac{2 g^2}{v}
[(q.r)g_{\mu\nu}-q_{\mu}r_{\nu}]({\alpha_1}{\rm cos}{\theta}_W
+ {\alpha_8}{\rm sin}{\theta}_W) - \frac{2
g^2}{v}[(k.q)g_{\mu\nu} \nonumber \\
& & -q_{\mu}k_{\nu}]({\alpha}_2 {\rm cos}{\theta}_W +
({\alpha}_3+{\alpha}_9){\rm sin}{\theta}_W) + \frac{2 g^2}{v}
{\alpha}_{11}
{\rm sin}{\theta}_W [k^2g_{\mu\nu}-k_{\nu}r_{\mu}+r_{\mu}r_{\nu}]
\nonumber \\
{\rm I}_{W^-_{\mu}W^+_{\nu}Z^0_{\lambda}} & = &
ig{\rm cos}{\theta}_W (1 + {\alpha_3} g^2
{\rm sec}^2{\theta}_W) [(k-r)_{\lambda}
g_{\mu\nu} + (r-q)_{\mu}g_{\nu\lambda} + (q-k)_{\nu} g_{\mu\lambda}] +
ig^2 \{ g{\alpha}_3 \nonumber \\
& & {\rm sin} {\theta}_W {\rm tg}{\theta}_W + g
({\alpha}_2-{\alpha}_1){\rm sin}{\theta}_W + g({\alpha}_8 -
{\alpha}_9){\rm cos}{\theta}_W - {\alpha}_{13}g {\rm sec}{\theta}_W \}
[q_{\mu}g_{\nu\lambda} - q_{\nu} \nonumber \\
& & g_{\mu\lambda}]
+ i g^2 \{ -g{\alpha}_{11} {\rm sin}{\theta}_W {\rm tg}{\theta}_W
+ \frac{{\alpha}_{12}}{2}g
{\rm sec}{\theta}_W \} [k_{\mu}g_{\nu\lambda} - r_{\nu}g_{\mu\lambda}]
\nonumber \\
{\rm I}_{W^-_{\mu}W^+_{\nu}A_{\lambda}} & = &
ig{\rm sin}{\theta}_W [(k-r)_{\lambda}
g_{\mu\nu} + (r-q)_{\mu}g_{\nu\lambda} + (q-k)_{\nu} g_{\mu\lambda}] +
ig^3 \{ -{\alpha}_3{\rm sin}{\theta}_W +
({\alpha}_1- \nonumber \\
& & {\alpha}_2){\rm cos}{\theta}_W + ({\alpha}_8 -
{\alpha}_9){\rm sin}{\theta}_W \}
[q_{\mu}g_{\nu\lambda} - q_{\nu}g_{\mu\lambda}] + i g^3
{\alpha}_{11} {\rm sin}{\theta}_W
[k_{\mu}g_{\nu\lambda} \nonumber \\
& & - r_{\nu}g_{\mu\lambda}]
\label{eq:GF3}\end{aligned}$$
iii\) Here, we write the Feynman rules corresponding to the different vertices with 4 gauge electroweak bosons (${\rm I}_{V_1V_2V_3V_4}$), shown in Figure \[fig:GF4\].
$$\begin{aligned}
{\rm I}_{W^-_{\mu}W^-_{\nu}W^+_{\lambda}W^+_{\rho}} & = &
i g^2 \{ 1 + g^2({\alpha}_4 - {\alpha}_8) + 2 g^2 ({\alpha}_3 +
{\alpha}_9 + {\alpha}_{13}) [2g_{\mu\nu}g_{\lambda\rho} -
g_{\mu\lambda}g_{\nu\rho} - g_{\mu\rho}g_{\nu\lambda}] + 2ig^4
\nonumber \\
& & ({\alpha}_4 + {\alpha}_5) [g_{\mu\lambda}g_{\nu\rho} +
g_{\mu\rho}g_{\nu\lambda}] \nonumber \\
{\rm I}_{Z^0_{\mu}Z^0_{\nu}W^-_{\lambda}W^+_{\rho}} & = &
i g^2 \{ -{\rm cos}^2{\theta}_W - 2g^2{\alpha}_3 + g^2{\rm sec}^2{\theta}_W
({\alpha}_5 + {\alpha}_7) \}[2g_{\mu\nu}g_{\lambda\rho} -
g_{\mu\lambda}g_{\nu\rho} - g_{\mu\rho}g_{\nu\lambda}] + ig^4 \nonumber \\
& & \{ ({\alpha}_4
+ {\alpha}_5 + {\alpha}_6 + {\alpha}_7) {\rm sec}^2{\theta}_W +
{\alpha}_{12} {\rm tg}^2{\theta}_W -
{\alpha}_{11}{\rm sin}^2{\theta}_W{\rm tg}^2{\theta}_W \}
[g_{\mu\lambda}g_{\nu\rho}+ g_{\mu\rho}g_{\nu\lambda}] \nonumber \\
{\rm I}_{A_{\mu}Z^0_{\nu}W^-_{\lambda}W^+_{\rho}} & = &
-i g^2 {\rm sin}{\theta}_W {\rm cos}{\theta}_W [1 + g^2{\rm sec}^2{\theta}_W
{\alpha}_3] [2g_{\mu\nu}g_{\lambda\rho} -
g_{\mu\lambda}g_{\nu\rho} - g_{\mu\rho}g_{\nu\lambda}] + ig^4
\{ {\rm sin}^2{\theta}_W \nonumber \\
& & {\rm tg}{\theta}_W{\alpha}_{11} -
\frac{1}{2} {\rm tg}{\theta}_W
{\alpha}_{12} \} [g_{\mu\lambda}g_{\nu\rho}+ g_{\mu\rho}g_{\nu\lambda}]
\nonumber \\
{\rm I}_{A_{\mu}A_{\nu}W^-_{\lambda}W^+_{\rho}} & = &
-i g^2 \{ {\rm sin}^2{\theta}_W [2 g_{\mu\nu}g_{\rho\lambda} -
g_{\mu\lambda}g_{\nu\rho} - g_{\mu\rho}g_{\nu\lambda}] + g^2
{\rm sin}^2{\theta}_W
{\alpha}_{11} [ g_{\mu\lambda}g_{\nu\rho} +
g_{\mu\rho}g_{\nu\lambda} ] \} \nonumber \\
{\rm I}_{Z^0_{\mu}Z^0_{\nu}Z^0_{\lambda}Z^0_{\rho}} & = &
2i{\rm sec}^4{\theta}_W g^4 [({\alpha}_4 + {\alpha}_5) + 2 ({\alpha}_6 +
{\alpha}_7 + {\alpha}_{10})] [g_{\mu\nu}g_{\lambda\rho} +
g_{\mu\lambda}g_{\nu\rho} + g_{\mu\rho}g_{\nu\lambda}]
\label{eq:GF4}\end{aligned}$$
[ ]{} S. L. Glashow, [*Nucl. Phys.*]{} [**22**]{} (1961) 579\
S. Weinberg, [*Phys. Rev. Lett.*]{} [**19**]{} (1967) 1264\
A. Salam, [*Proc. 8th Nobel Symp.,*]{} ed. N. Svartholm, p. 367, Stockholm, Almqvist y Wiksells (1968)
H.Haber and G. Kane, [*Phys. Rep.*]{} [**117**]{} (1985) 75
S.Weinberg, [*Phys. Rev.*]{} [**D19**]{} (1979) 1277\
S.Dimopoulos and L.Susskind, [*Nucl. Phys.*]{} [**B155**]{} (1979) 237\
E.Farhi and L.Susskind, [*Phys. Rep.*]{} [**74**]{} (1981) 277
A. Dobado and M.J. Herrero, [*Phys. Lett.*]{} [**B228**]{} (1989) 495 and [**B233**]{} (1989) 505\
J.Donoghue and C. Ramírez, [*Phys. Lett.*]{} [**B234**]{} (1990) 361
B.Holdom and J. Terning, [*Phys.Lett.*]{} [**B247**]{} (1990) 88\
A. Dobado, D. Espriu and M.J. Herrero, [*Phys.Lett.*]{} [**B255**]{} (1991) 405\
M. Golden and L. Randall, [*Nucl. Phys.*]{} [**B361**]{} (1991) 3
S Weinberg, [*Physica*]{} [**A 96**]{} (1979) 327\
J.Gasser and H.Leutwyler, [*Ann. of Phys.*]{} [**158**]{} (1984) 142, [*Nucl. Phys.*]{} [**B250**]{} (1985) 465 y 517
P. Sikivie et al., [*Nucl. Phys.*]{} [**B173**]{} (1980) 189\
M. S. Chanowitz, M. Golden and H. Georgi, [ *Phys.Rev.*]{} [**D36**]{} (1987)1490
A. Dobado, M.J. Herrero and J. Terrón, [*Z. Phys.*]{} [**C50**]{} (1991) 205; [*Z. Phys.*]{} [**C50**]{} (1991) 465 and Proceedings of the Large Hadron Collider Workshop, Aachen, 1990, Vol II, (G. Jarlskog and D. Rein, Geneva, 1990)\
J.Bagger, S.Dawson y G.Valencia, [*Nucl.Phys.*]{} [**B399**]{} (1993) 364\
J. Bagger, V. Barger, K. Cheung, T. Han, G.A. Ladinsky, R. Rosenfeld y C.P. Yuan, [*Phys. Rev.*]{} [**D49**]{} (1994) 1246\
I.Jossa, F.Pauss and T.Rodrigo, Large Hadron Collider Workshop Aachen (1990), (1990), Vol II (Geneva 1990) 796.
J.M. Cornwall, D.N. Levin and G. Tiktopoulos, [*Phys. Rev.*]{} [**D10**]{} (1974) 1145\
C.E. Vayonakis, [*Lett. Nuovo Cim.*]{}[**17**]{}(1976) 383\
B.W. Lee, C. Quigg and H. Thacker, [*Phys. Rev.*]{} [**D16**]{} (1977) 1519\
M.S. Chanowitz and M.K. Gaillard, [*Nucl. Phys.*]{} [**261**]{} (1985)379\
A. Dobado J. R. Peláez [*Nucl. Phys.*]{} [**B425**]{} (1994) 110; [*Phys. Lett.*]{}[**B329**]{}(1994)469 (Addendum, ibid, [**B335**]{} (1994) 554)
Tran N. Truong, [*Phys. Rev.*]{} [**D61**]{} (1988)2526\
A. Dobado, M.J. Herrero and J.N. Truong, [*Phys. Lett.*]{} [**B235**]{} (1990) 134\
T.N.Truong, [*Phys. Rev. Lett.*]{} [**67**]{} (1991)2260; A. Dobado and J.R. Peláez, [*Phys. Rev.*]{} [**D47**]{}(1992)4883\
C.J.C. Im, [*Phys. Lett.*]{} [**B281**]{} (1992)357; A. Dobado and J.R. Peláez, [*Phys. Lett.*]{} [**B286**]{} (1992)136
T.Appelquist C.Bernard, [*Phys. Rev.*]{} [**D22**]{} (1980) 200\
A.Longhitano [*Phys. Rev.*]{} [**D22**]{} (1980) 1166; [*Nucl. Phys.*]{} [**B188**]{} (1981) 118
F. Feruglio [*in*]{} the $2^{nd}$ Nat. Seminar of Th. Physics, Parma, Sept. 1992. DFPD92/TH/50
C.Weizsaker and E.J.Williams, [*Z. Phys.* ]{} [**88**]{} (1934) 612
S. Dawson, [*Nucl. Phys.*]{} [**B249**]{} (1985) 42
E.W.N.Glover and J.J.Van deer Bij, [*Nucl. Phys.*]{} [**B321**]{} (1989) 561
A.Falk, M.Luke and E.Simmons, [*Nucl. Phys.*]{} [**B365**]{} (1991) 523
A.Dobado and M.Urdiales, [*Phys. Lett.*]{} [**B292**]{} (1992) 128
M.Herrero and E.Ruiz Morales, [*Nucl. Phys.*]{} [**B418**]{} (1994) 431; FTUAM 94/11, hep-ph/9411207 to appear in [*Nucl. Phys.*]{} [**B**]{} (1995)
A.Dobado, J.R.Peláez and M.T.Urdiales, [*Contributed paper to the 27th International Conference in High Energy Physics*]{} (Glasgow , July 1994) hep-ph / 9407384, and work in preparation.
E.Eitchen et.al., [*Rev. Mod. Phys.*]{} [**56**]{} (1984) 579
G.P. Lepage, [*J. Comput. Phys.*]{} [**27**]{} (1978) 192
A.D. Martin, R.G Roberts and W.J. Stirling, [*Phys. Lett.*]{} [**306B**]{} (1993) 145 and [*Phys. Lett.*]{} [bf 309B]{} (1993) 492
M. Glück, E. Reya and A. Vogt, [*Z. Phys.*]{} [**C48**]{} (1990) 471, [*Z. Phys.*]{} [**C53**]{} (1992) 127 and [*Phys. Lett.*]{} [**306B**]{} (1993) 391
J. Terrón, private communication
Particle Data Group, J.J.Hernández et.al., Review of particle properties, [*Phys. Lett.*]{} [**B239**]{} (1990) 1
Table Captions {#table-captions .unnumbered}
==============
Table \[tab:equiv\]. In this Table we display the relations between different sets of chiral Lagrangian parameters. They correspond to those used by Longhitano [@Long], Feruglio [@Fer] and Dobado et.al. [@HeE2], in columns $1$, $2$ and $3$, respectively.
Table \[tab:line\]. In this table we show the sensitivity functions $s^{(ZZ)}_k$ and $s^{(WZ)}_k$ associated to each ${\alpha}_k$ ($k=0,1,2,3,4,5$) parameter. They were calculated with the minimal cuts and under the hypothesis of linear behaviour of $N^{(i)}({\alpha};c)$ with respect to each ${\alpha}_k$.
Table \[tab:opti1\]. Optimal cuts to calculate the statistical significance $r^{(i)}_k(c)$ for typical positive and negative ${\alpha}_k$ values (${\alpha}_k = \pm
0.005$) in the $Z^0Z^0$ or $W^{\pm}Z^0$ final states. The first entry in column 3 refers to the minimum invariant mass and the second to the minimum allowed $p_{TZ}$, both them given in $GeV$. The total number of events in the [*Zero Model*]{} and with the typical chosen ${\alpha}_k$ obtained after having applied the optimal cuts, are represented in columns $4$ and $5$. They correspond to $1$ year of running for the $LHC$ and a maximal rapidity of $2.5$ for the final bosons.
Table \[tab:opti2\]. Here, we shown a comparison between the statistical significance function $r^{(i)}_k(c)$ given by the typical ${\alpha}_k$ in $Z^0Z^0$ or $W^{\pm}Z^0$ final state, for the minimal and the optimal cuts. In both cases, we have fixed the maximal rapidity cut of $2.5$ for two gauge final bosons, and $1$ year of working time for the $LHC$.
Table \[tab:error\]. In column 3, we display the statistical errors, $(\Delta \alpha_k)_{\rm stat.}$, corresponding to each ${\alpha}_k$ parameter in the $Z^0Z^0$ or $W^{\pm}Z^0$ channels. They were obtained with the optimal cuts (column 1) and under the assumption of linear dependence of the total number of events on ${\alpha}_k$.
Table \[tab:festru\]. Comparison of the total number of $Z^0Z^0$ or $W^{\pm}Z^0$ events obtained in the [*Zero Model*]{} applying the optimal cuts and choosing different structure functions: $EHLQ$, $MRSD-$ and $GRVHO$. In the last three columns, we represent the statistical significance to estimate these imprecisions, $r^{(i)}_{\rm struc.}(c)$ eq.(\[eq:rstruc\]), between $EHLQ$ and $MRSD-$ (column $6$), $EHLQ$ and $GRVHO$ ($7$ column) and $MRSD-$ and $GRVHO$ ($8$ column). These uncertainties are called, respectively, $r^{(i)}_{12}$, $r^{(i)}_{13}$ and $r^{(i)}_{23}$. In all these computations we have taken $y_{1 {\rm max}}=y_{2 {\rm
max}}=2.5$ and an integrated luminosity for the $LHC$ of $3 \times
10^5 pb^{-1}$.
Table \[tab:run\]. In this Table we represent the total number of $Z^0Z^0$ or $W^{\pm}Z^0$ pairs obtained at $LHC$ using the optimal cuts and setting $y_{1 {\rm max}}=y_{2 {\rm
max}}=2.5$ and an integrated luminosity for the $LHC$ of $3 \times 10^5 pb^{-1}$. In columns 2 and 3 we display the results in the [*Zero Model*]{} and with constant $g$ and $g'$ couplings. Again in columns 4 and 5 we show the total number of $Z^0Z^0$ or $W^{\pm}Z^0$ events but in the [*Running Zero Model*]{} and with $g(\sqrt{\hat{s}})$ and $g'(\sqrt{\hat{s}})$, eq.(\[eq:gsR\]). In the last two columns we display a measure of the statistical significance that corresponds to include the dependence on the energy of the parameters and couplings, given by the $r^{(ZZ)}_{\rm run.}(c)$ or $r^{(WZ)}_{\rm run.}(c)$ functions, eq.(\[eq:rrun\]).
Figure Captions {#figure-captions .unnumbered}
===============
Figure \[fig:GF2\]. Propagators of the $\pi^{\pm}$ and $W^{\pm}$.
Figure \[fig:GF3\]. Vertices with three gauge bosons or two gauge bosons and one $GB$, obtained from ${\cal L}_{\rm NLSM}$ eq.(\[eq:MENL\]).
Figure \[fig:GF4\]. Different four gauge boson vertices that appear in the theory described by the lagrangian ${\cal L}_{\rm
NLSM}$ eq.(\[eq:MENL\]).
Figure \[fig:gg\]. Different diagrams to one loop order contributing to $gg \rightarrow Z^0Z^0$ cross section, in the $SM$ without a Higgs boson.
Figure \[fig:qq\]. We display the diagrams at tree level order that contribute to the cross section $q\bar{q} \rightarrow Z^0Z^0$ in the framework of chiral perturbation theory (with ${\cal L}_{\rm NLSM}$). In this case we obtain the same result as in the $SM$ without a Higgs boson.
Figure \[fig:ZZ\]. We show the only diagram contributing to the different $Z^0Z^0 \rightarrow Z^0Z^0$ helicity amplitudes obtained with ${\cal L}_{\rm NLSM}$ eq.(\[eq:MENL\]), at tree level order.
Figure \[fig:WW\]. The scattering amplitudes for the process $W^+W^- \rightarrow Z^0Z^0$, calculated with ${\cal L}_{\rm
NLSM}$ eq.(\[eq:MENL\]) at tree level order in the Landau gauge, receive contribution from all these channels.
Figure \[fig:qq’\]. The cross sections $q\bar{q'} \rightarrow
W^{\pm}Z^0$ are calculated adding the contribution of the diagrams displayed in this Figure. They are obtained with ${\cal L_{\rm NLSM}}$ eq.(\[eq:MENL\]) at tree level order and in the Landau gauge. The initial quarks are supposed to be massless.
Figure \[fig:WZ\]. Different diagrams contributing to the $W^{\pm}Z^0 \rightarrow W^{\pm}Z^0$ scattering amplitudes. They are calculated with ${\cal L}_{\rm NLSM}$ eq.(\[eq:MENL\]) at tree level order and in the Landau gauge.
Figure \[fig:Wf\]. The scattering amplitudes $W^{\pm}\gamma \rightarrow W^{\pm}Z^0$ are obtained with the Lagrangian given in eq.(\[eq:MENL\]) adding the contribution of the diagrams displayed in this Figure. The calculation is made at tree level order and fixing the Landau gauge.
Figure \[fig:sena0\]. This graphic represents the sensitivity of the $LHC$ to the ${\alpha}_0$ parameter in $Z^0Z^0$ and $W^{\pm}Z^0$ channels. We display the total number of $Z^0Z^0$ (solid line) and $W^{\pm}Z^0$ (dashed line) events obtained in the $LHC$, for different values of the ${\alpha}_0$ parameter. The kinematical cuts are the minimal ones. The other ${\alpha}_k$ parameters, apart from ${\alpha}_0$, have been set to zero.
Figure \[fig:sena1\]. This Figure represents the sensitivity function to ${\alpha}_1$ parameter for the $LHC$, using the same kinematical cuts as those in Figure \[fig:sena0\]. We display the total $N^{(ZZ)}$ (solid line) and $N^{(WZ)}$ (dashed line) events obtained versus ${\alpha}_1$ values.
Figure \[fig:sena2\]. This Figure represents the sensitivity function to ${\alpha}_2$ parameter for the $LHC$. As in Figures \[fig:sena0\] and \[fig:sena1\] the total number of $Z^0Z^0$ and $W^{\pm}Z^0$ events versus ${\alpha}_2$ values.
Figure \[fig:sena3\]. This Figure represents the sensitivity function to ${\alpha}_3$ parameter for the $LHC$, using the same kinematical cuts as in previous Figures. We show the total $N^{(ZZ)}$ (solid line) and $N^{(WZ)}$ (dashed line) events versus ${\alpha}_3$.
Figure \[fig:sena4\]. This graphic represents the sensitivity of the $LHC$ to the ${\alpha}_4$ parameter in the $Z^0Z^0$ and $W^{\pm}Z^0$ channels. We display, like in Figures \[fig:sena0\], \[fig:sena1\], \[fig:sena2\] and \[fig:sena3\], the total number of $Z^0Z^0$ (solid line) and $W^{\pm}Z^0$ (dashed line) events obtained at $LHC$, for different values of the ${\alpha}_4$ parameter, using the same kinematical cuts as in previous Figures.
Figure \[fig:sena5\]. Sensitivity function to ${\alpha}_5$ parameter for the $LHC$. We have used the same kinematical cuts as in previous Figures. We display the total $N^{(ZZ)}$ (solid line) and $N^{(WZ)}$ (dashed line) events obtained versus ${\alpha}_5$ values, as we did in figures \[fig:sena0\] to \[fig:sena4\].
Figure \[fig:mvcut\]. In this Figure we study the variation of the total number of $Z^0Z^0$ (solid line) and $W^{\pm}Z^0$ (dashed line) events with the cut on the minimal invariant mass, $\sqrt{\hat{s}_{\rm min}}$. We have set all the ${\alpha}_k$ parameters to their values in the [*Zero Model*]{} and the cuts ${p_{TZ}}_{\rm min}$, ${y_1}_{\rm max}$ and ${y_2}_{\rm max}$ to their minimal values.
Figure \[fig:ptcut\]. As in Figure \[fig:mvcut\], we display the behaviour of the total number of $Z^0Z^0$ (solid line) and $W^{\pm}Z^0$ (dashed line) events in the [*Zero Model*]{} versus the cut in the minimal transversal momentum of the $Z^0$ boson. We have taken the other kinematical bounds as their minimal values.
Figure \[fig:yicut\]. Here we reflect the dependence of the total number of $Z^0Z^0$ and $W^{\pm}Z^0$ events on the maximal rapidity bounds of final gauge bosons, $y_{1 {\rm max}} = y_{2 {\rm max}}$. We work in the [*Zero Model*]{} and fix the minimal cuts in invariant mass and transversal moment. The solid line corresponds to the $Z^0Z^0$ final state and the dashed one to the produced $W^{\pm}Z^0$ pairs.
Figure \[fig:siga0\]. Statistical significance corresponding to the ${\alpha}_0$ parameter. We display the $r_0$ function eq.(\[eq:rk\]) for different values of ${\alpha}_0$ with respect to the [*Zero Model*]{}. The solid line represents the $Z^0Z^0$ final state and the dashed line the $W^{\pm}Z^0$. We use the same kinematical cuts as in Figures \[fig:sena0\] to \[fig:sena5\].
Figure \[fig:siga1\]. Statistical significance corresponding to the ${\alpha}_1$ parameter. We show as in Figure \[fig:siga0\], the $r^{(ZZ)}_1$ (solid) eq.(\[eq:rk\]) and $r^{(WZ)}_1$ (dashed) functions with respect to the [*Zero Model*]{} versus these ${\alpha}_1$ values. We use the same kinematical cuts as in Figures \[fig:sena0\] to \[fig:sena5\].
Figure \[fig:siga2\]. This Figure represents the statistical significance for some values of ${\alpha}_2$ with respect to the [*Zero Model*]{}. We display as in Figures \[fig:siga0\] and \[fig:siga1\], and for the same kinematical cuts, the $r^{(ZZ)}_2$ and $r^{(WZ)}_2$ eq.(\[eq:rk\]) functions.
Figure \[fig:siga3\]. This Figure reflects the statistical significance of ${\alpha}_3$. We show the $r_k$ eq.(\[eq:rk\]) functions for $Z^0Z^0$ (solid) and $W^{\pm}Z^0$ (dashed) with respect to the [*Zero Model*]{} versus ${\alpha}_3$ values, using the same kinematical cuts as in previous Figures.
Figure \[fig:siga4\]. Statistical significance corresponding to the ${\alpha}_4$ parameter. As in Figures \[fig:siga0\] to \[fig:siga3\] we show the $r^{(ZZ)}_4$ (solid) (eq. \[eq:rk\]) and $r^{(WZ)}_4$ (dashed) with respect to the [*Zero Model*]{} versus the ${\alpha}_4$ values. We have used the same kinematical cuts as in previous figures.
Figure \[fig:siga5\]. The same as in Figures \[fig:siga0\] to \[fig:siga4\] for the ${\alpha}_5$ parameter.
${\alpha}_0$ $a_0/g^2$ $\delta\beta/g^2$
----------------- ------------ ---------------------------
${\alpha}_1$ $g'/g a_1$ $L_{10}$tg${\theta}_W$
${\alpha}_2$ $g'/g a_2$ $-L_{9R}/2$tg${\theta}_W$
${\alpha}_3$ $-a_3$ $-L_{9L}/2$
${\alpha}_4$ $a_4$ $L_2$
${\alpha}_5$ $a_5$ $L_1$
${\alpha}_6$ $a_6$ $\;-\;$
${\alpha}_7$ $a_7$ $\;-\;$
${\alpha}_8$ $-a_8$ $\;-\;$
${\alpha}_9$ $-a_9$ $\;-\;$
${\alpha}_{10}$ $2a_{10}$ $\;-\;$
${\alpha}_{11}$ $a_{11}$ $\;-\;$
${\alpha}_{12}$ $2a_{12}$ $\;-\;$
${\alpha}_{13}$ $a_{13}$ $\;-\;$
\[tab:equiv\]
${\alpha}_k$ $s^{(ZZ)}_k(c^{\rm min})$ $s^{(WZ)}_k(c^{\rm min})$
-------------- --------------------------- ---------------------------
${\alpha}_0$ $-899.54$ $2779.89$
${\alpha}_1$ $1241.27$ $-5408.20$
${\alpha}_2$ $-1481.65$ $6150.98$
${\alpha}_3$ $6751.24$ $-48366.79$
${\alpha}_4$ $2445.43$ $-18981.46$
${\alpha}_5$ $5924.90$ $-20142.33$
\[tab:line\]
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
${\alpha}_k$ value Channel ${c}^{\rm op} = $N^{(i)} ({\alpha}^0; {c}^{\rm op})$ $N^{(i)} ({\alpha}; {c}^{\rm op})$
(\sqrt{\hat{s}^{\rm op}_{\rm min}}, p^{\rm op}_{T {\rm min}})$
----------------------- -------------- ---------------------------------------------------------------- -------------------------------------- ------------------------------------
${\alpha}_3 = -0.005$ $Z^0Z^0$ $(200,10)$ 16152.55 16119.28
${\alpha}_3 = -0.005$ $W^{\pm}Z^0$ $(200,300)$ 522.59 581.41
${\alpha}_3 = 0.005$ $Z^0Z^0$ $(200,10)$ 16152.55 16186.47
${\alpha}_3 = 0.005$ $W^{\pm}Z^0$ $(200,200)$ 2268.62 2160.74
${\alpha}_4 = -0.005$ $W^{\pm}Z^0$ $(1150,500)$ $32.29$ $66.48$
${\alpha}_4 = 0.005$ $Z^0Z^0$ $(1150,400)$ $13.50$ $20.01$
${\alpha}_5 = -0.005$ $W^{\pm}Z^0$ $(1150,400)$ $61.56$ $85.78$
${\alpha}_5 = 0.005$ $Z^0Z^0$ $(1150,400)$ $13.50$ $28.70$
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
\[tab:opti1\]
${\alpha}_k$ value Channel $r_k({\alpha}; {c}^{\rm min})$ $r_k({\alpha}; {c}^{\rm op})$
----------------------- -------------- -------------------------------- -------------------------------
${\alpha}_3 = -0.005$ $Z^0Z^0$ $0.26$ $0.26$
${\alpha}_3 = -0.005$ $W^{\pm}Z^0$ $1.01$ $2.57$
${\alpha}_3 = 0.005$ $Z^0Z^0$ $0.27$ $0.27$
${\alpha}_3 = 0.005$ $W^{\pm}Z^0$ $0.51$ $2.27$
${\alpha}_4 = -0.005$ $W^{\pm}Z^0$ $0.45$ $6.02$
${\alpha}_4 = 0.005$ $Z^0Z^0$ $0.18$ $1.77$
${\alpha}_5 = -0.005$ $W^{\pm}Z^0$ $0.39$ $3.09$
${\alpha}_5 = 0.005$ $Z^0Z^0$ $0.37$ $4.14$
\[tab:opti2\]
-------------------------------------------------------------------------------------------------------------------------
${c}^{\rm op} = Channel $(\Delta {\alpha}_k)_{\rm stat.}$
(\sqrt{\hat{s}^{\rm op}_{\rm min}}, p^{\rm op}_{T {\rm min}})$
---------------------------------------------------------------- -------------- -----------------------------------------
$(200,300)$ $W^{\pm}Z^0$ $(\Delta {\alpha}_3)_{\rm stat.} = 1.95
\times 10^{-3}$
$(200,200)$ $W^{\pm}Z^0$ $(\Delta {\alpha}_3)_{\rm stat.} = 2.20
\times 10^{-3}$
$(1150,500)$ $W^{\pm}Z^0$ $(\Delta {\alpha}_4)_{\rm stat.} = 8.31
\times 10^{-4}$
$(1150,400)$ $Z^0Z^0$ $(\Delta {\alpha}_4)_{\rm stat.} = 2.82
\times 10^{-3}$
$(1150,400)$ $W^{\pm}Z^0$ $(\Delta {\alpha}_5)_{\rm stat.} = 1.62
\times 10^{-3}$
$(1150,400)$ $Z^0Z^0$ $(\Delta {\alpha}_5)_{\rm stat.} = 1.21
\times 10^{-3}$
-------------------------------------------------------------------------------------------------------------------------
\[tab:error\]
--------- ------------------------------------------------------------------- ----------------------------- ----------------------------------- ----------------------------------- ---------------- ---------------- ----------------
Channel Optimal cuts $N^{(i)}({\alpha}^0; c^{\rm $N^{(i)}({\alpha}^0; c^{\rm op})$ $N^{(i)}({\alpha}^0; c^{\rm op})$ $r^{(i)}_{12}$ $r^{(i)}_{13}$ $r^{(i)}_{23}$
op})$
$({\sqrt{\hat{s}_{\rm min}}}^{\rm op}$, $p^{\rm op}_{T \rm min})$ $EHLQ$([*set*]{} $II$) $MRS -D'$ $GRV HO$
$ZZ$ ($200,10$) 16113.31 17242.93 16617.27 8.90 3.97 4.76
$WZ$ ($200,200$) 2298.57 2526.75 2452.56 4.76 3.21 1.48
$WZ$ ($200,300$) 535.80 594.34 578.75 2.53 1.86 0.64
$WZ$ ($1150,500$) 36.02 40.79 39.97 0.79 0.66 0.13
$ZZ$ ($1150,400$) 14.92 15.88 15.39 0.25 0.12 0.12
$WZ$ ($1150,400$) 68.02 76.72 74.78 1.05 0.82 0.22
--------- ------------------------------------------------------------------- ----------------------------- ----------------------------------- ----------------------------------- ---------------- ---------------- ----------------
\[tab:festru\]
[|r|r|r|c|c|c|c|]{} Optimal cuts & & &\
$(\sqrt{\hat{s}_{\rm min}})$ & $N^{(ZZ)}(\alpha^0;c^{\rm op})$ & $N^{(WZ)}(\alpha^0;c^{\rm op})$ & $N^{(ZZ)}(\alpha^0;c^{\rm op})$ & $N^{(WZ)}(\alpha^0;c^{\rm op})$ & $r^{(ZZ)}_{\rm run.}$ & $r^{(WZ)}_{\rm run.}$\
$(200,10)$ & 16149.13 & 91141.49 & 15355.82 & 86149.13 & 6.24 & 16.54\
$(200,200)$ & $-\;\;\;\;$ & 2281.63 & $\;\;-\;\;$ & 2093.83 & $\;\;-\;\;$ & 3.93\
$(200,300)$ & $-\;\;\;\;$ & 532.33 & $\;\;-\;\;$ & 505.18 & $\;\;-\;\;$ & 1.18\
$(1150,500)$ & $-\;\;\;\;$ & 36.01 & $\;\;-\;\;$ & 45.75 & $\;\;-\;\;$ & 1.62\
$(1150,400)$ & 14.91 & 68.01 & 11.28 & 78.07 & 0.94 & 1.22\
\[tab:run\]
[^1]: E-mail: [email protected]
[^2]: E-mail: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
address:
- 'Department of Mathematics & Computer Science, TU Eindhoven, The Netherlands'
- 'Department of Mathematics & Statistics, Dalhousie University, Halifax, Canada'
- 'Mathematics Institute, University of Warwick, United Kingdom'
author:
- Iason Zisis
- 'Joep H.M. Evers'
- Bas van der Linden
- Manh Hong Duong
bibliography:
- 'SPHbib.bib'
title: Recent results in the systematic derivation and convergence of SPH
---
Introduction
============
In the literature of the Smoothed Particle Hydrodynamics (SPH) computational method, it is established that the standard SPH scheme can be derived by applying the principle of least action to a particle system, where the SPH density estimate acts as a constraint [@Bonet1999; @Monaghan2005; @Price2012]. Nonetheless, a subtlety lies in the fact that in the derivation of the SPH equations *the action of the particle system is minimized rather than the action of the continuum*. The procedure of starting from the continuum, minimizing its action and then discretizing the equations reveals the mathematical formalities necessary to convince oneself that SPH indeed comes from principles of continuum mechanics.
The theory of *measure-valued evolution equations* was first shown by Di Lisio *et al.* [@DiLisio1998] to provide the adequate framework for the study of both the SPH particle system and the limiting continuum setting in a single context. Following them, in Evers *et al.* [@Evers2015] we adopt *the Wasserstein distance in the space of probability measures*, which determines how “close" two measures are by computing the (optimal) cost of transforming one into the other. Its definition allows to conveniently derive suitable upper bounds on the distance between measures, which are necessary ingredients to prove convergence. After constructing a discrete approximation of the initial measure, we prove the convergence of measure-valued solutions [@Evers2015]. There are two major differences with Di Lisio *et al.* [@DiLisio1998] (and accordingly with a recent review [@Colagrossi2014]). First, their scheme is not the classic SPH scheme, but rather one that is known not to conserve momentum, whenever applied to relevant physical processes. Our proof applies to both the traditional SPH scheme and the former one. Second, we allow for a much more general class of force fields, including external and internal conservative forces, as well as friction and non-local interactions. The overall aim of the present paper is to report recent developments in the formal derivation of SPH —focusing on the authors’ work [@Evers2015]— and eventually bridge the gap between the SPH literature and the few publications dealing with the rigorous mathematical formalism of SPH. The structure of the paper is the following: Section \[sec:SPH from cm\] summarizes the systematic procedure followed for the derivation of measure-valued and particle formulations of continuum mechanics equations. Section \[sec: Conv\] states the theoretical convergence result. Section \[sec: add\] discusses on the limitations of the proof. Section \[sec: numerics\] presents numerical paradigms which exhibit the applicability of the convergence with respect to the Wasserstein distance. Finally, the related open problems are stated in Section \[sec: concl\].
SPH from continuum mechanics {#sec:SPH from cm}
============================
Denote the particle trajectories $x = x(x_0,t) \in \Omega_t \subset {\mathbb R}^d$, taken with respect to an initial configuration of the medium $x_0 \in \Omega_0 \subset {\mathbb R}^d$, with $d$ being the spatial dimension. In the following, on grounds of concise notation, we suppress the dependence on $t$ and it is only implied. For a medium found in the domain $\Omega$, its mass: $$\label{eq: mass_measure}
\mu(\Omega) = \int_{\Omega} {\textup d}\mu(x) = \int_{\Omega_0} {\textup d}\mu_0(x_0) = \mu_0(\Omega_0),$$ is always conserved in the absence of sinks or sources. The above equality is guaranteed by the one-to-one correspondence between the set of particles in the medium’s reference configuration $\Omega_0$ and any later configuration $\Omega$. Thus, for the integral of some bounded and measurable $f$ with respect to the measure $\mu$, it is possible to perform a coordinate transform from $\Omega$ to $\Omega_0$ and obtain $
\int_\Omega f(x) \, {\textup d}\mu(x) = \int_{\Omega_0} f(x) \, {\textup d}\mu_0(x_0),
$ where we recall that $x=x(x_0,t)$. In the context of measures the *mass density* function is defined as the (*Radon-Nikodym*) derivative of $\mu$ with respect to the *Lebesgue measure* $\lambda$: $$\label{eq: density}
\rho = \frac{{\textup d}\mu}{{\textup d}\lambda}.$$ Consequently, (\[eq: mass\_measure\]) attains the typical form: $
\int_{\Omega} \rho(x) \, {\textup d}\lambda(x) = \int_{\Omega_0} \rho_0(x_0) \, {\textup d}\lambda(x_0).
$ The measure-valued definition (\[eq: mass\_measure\]) of mass conservation is more general, since it does not require $\rho$ to be well-defined. Thus, it offers the basis for a generalized approach for both continuous and discretized media. Within this context, we may construct a discrete approximation $\mu_0^N$ of the measure of mass $\mu_0$, which then can be used for numerical analysis. The one-to-one correspondence between particles in $\Omega_0$ and $\Omega$ and the definition of the medium’s density function (\[eq: density\]) deliver $
\rho(x) \, {\textup d}\lambda(x) = \rho_0(x_0) \, {\textup d}\lambda(x_0),
$ and consequently the medium’s density is given by: $$\label{eq: density total}
\rho(x) = \frac{\rho_0(x_0)}{J(x)},$$ where $J$ is the determinant of the Jacobian matrix [@Seliger1968 Eq. (5)].
The action of the medium, $
\mathcal S = \int_{0}^{T} L \, {\textup d}t,
$ involves the Lagrangian: $$L = \int_\Omega \Big ( \frac{1}{2} \|\dot x\|^2 - e(\rho(x)) \Big ) \rho(x) \, {\textup d}\lambda(x),$$ where $e$ is the internal energy of the medium prescribed by the thermodynamic relation ${\textup d}e/{\textup d}\rho = P(\rho)/\rho^2$ for ideal processes, with $P$ the mean volumetric stress — *pressure*. According to the principle of least action, the equations of motion follow from the minimization problem $
\updelta \mathcal S |_{0}^{T} = \int_{0}^{T} \updelta L \, {\textup d}t = 0,
$ where the differential $\updelta$ denotes a variation of the particle trajectories $x$. Therefore, in order to obtain the variation and eventually the equations of motion, the spatial domain of integration (in the Lagrangian) should be independent of the variation. This is achieved either by writing the Lagrangian with respect to the initial configuration or in the context of measures.
In particular, three steps are necessary to pass from the action of the continuous system to the motion equations of the resulting particle system:
- [A:]{} Introduce the measure-valued formulation by replacing $\rho \, {\textup d}\lambda$ with ${\textup d}\mu$ and, wherever necessary, approximate $\rho$ by some $\tilde \rho$, which depends on the measure $\mu$. Typically, this is: $$\label{eq: mass_conservation}
\tilde \rho(x) = \int_\Omega W_h(x-y) {\textup d}\mu(y),$$ where $W_h$ is a symmetric positive mollifier, see e.g., [@Monaghan2005].
- [B:]{} Substitute for $\mu$ the discrete measure: $$\label{eq: discrete}
\mu^N = \sum_{i=1}^N m_i \, \delta_{x_i}.$$ where $\delta_{x_i} = 1$ if $x = x_i$ and zero otherwise, is the Dirac measure at $x_i$.
- [C:]{} Derive the Euler-Lagrange equations.
These three steps have been introduced in more details in [@Evers2015]. Step A takes us to a regularized version of the problem, which is a problem different from the original one. Step B cannot happen before A, but we have the freedom to choose the further ordering. This gives rise to three different derivations:
- [ABC:]{} Write the measure-valued Lagrangian: $$\tilde L = \int_\Omega \Big ( \frac{1}{2} \|\dot x\|^2 - e(\tilde \rho(x)) \Big ) {\textup d}\mu(x),$$ discretize it: $$\tilde L^N= \sum_i \Big ( \frac{1}{2} \|\dot x_i^2\| - e(\tilde \rho_i) \Big ) m_i,$$ and derive the corresponding equations of motion afterwards: $$\frac{d}{dt}\Big ( \frac{\partial \tilde L^N}{\partial \dot x_i} \Big ) - \frac{\partial \tilde L^N}{\partial x_i} = 0,$$ $$\label{eq: sph_motion1}
\ddot x_i = -\sum_j \Big ( \frac{P(\tilde \rho_i)}{\tilde \rho_i^2} + \frac{P(\tilde \rho_j)}{\tilde \rho_j^2} \Big ) \, \nabla W_h(x_i-x_j) \, m_j.$$ It is the common technique encountered in the SPH literature [@Bonet1999; @Monaghan2005; @Price2012] and its importance was recognized already in early articles about SPH (e.g. [@Monaghan1978]).
- [ACB:]{} From the measure-valued Lagrangian: $$\tilde L = \int_\Omega \Big ( \frac{1}{2} \|\dot x\|^2 - e(\tilde \rho(x)) \Big ) {\textup d}\mu(x),$$ derive the equations of motion: $$\frac{d}{dt}\Big ( \frac{\partial \tilde L}{\partial \dot x} \Big ) - \frac{\partial \tilde L}{\partial x} = 0,$$ $$\label{eq: continuous}
\ddot x = -\int_\Omega \Big ( \frac{P(\tilde \rho)}{\tilde \rho^2} \Big |_{(x)} + \frac{P(\tilde \rho)}{\tilde \rho^2} \Big |_{(y)} \Big ) \, \nabla W_h(x-y) \, {\textup d}\mu(y),$$ and discretize these equations afterwards: $$\label{eq: sph_motion2}
\ddot x_i = -\sum_j \Big ( \frac{P(\tilde \rho_i)}{\tilde \rho_i^2} + \frac{P(\tilde \rho_j)}{\tilde \rho_j^2} \Big ) \, \nabla W_h(x_i-x_j) \, m_j.$$ It is the technique introduced in the authors’ work [@Evers2015].
- [CAB:]{} From the Lagrangian of the continuous system in the reference configuration: $$L = \int_{\Omega_0} \Big ( \frac{1}{2} \|\dot x\|^2 - e(\rho_0(x_0)/J(x)) \Big ) \rho_0(x_0) \, {\textup d}\lambda(x_0),$$ derive the equations of motion taking into account the functional relation $\rho(x) = \rho_0(x_0)/J(x)$ as in [@Seliger1968]: $$\frac{d}{dt}\Big ( \frac{\partial L}{\partial \dot x} \Big ) - \frac{\partial L}{\partial x} = 0,$$ $$\ddot x = -\frac{1}{\rho} \frac{{\textup d}P(\rho)}{{\textup d}\rho} \nabla \rho,$$ write them in the measure-valued form: $$\ddot x = -\frac{1}{\tilde \rho(x)} \frac{{\textup d}P(\tilde \rho)}{{\textup d}\tilde \rho} \Big |_{(x)} \int_\Omega \nabla W_h(x-y) {\textup d}\mu(y),$$ and finally discretize them: $$\ddot x_i = -\frac{1}{\tilde \rho_i} \frac{{\textup d}P(\tilde \rho)}{{\textup d}\tilde \rho} \Big |_i \sum_j \nabla W_h(x_i-x_j) m_j.$$ This strategy is implied by [@DiLisio1998; @Colagrossi2014].
It thus turns out that the order in which these steps are executed determines what the resulting equation is. To be more precise, the classical SPH scheme [@Price2012] is obtained, whenever the regularization of the density takes place before applying the principle of least action (A-B-C and A-C-B). If we apply the principle of least action to the action at the continuum level before regularizing the density (C-A-B), then we obtain the scheme appearing in [@DiLisio1998; @Colagrossi2014], which is rarely ever employed for SPH computations. We emphasize that although both schemes arrive from the principle of least action, the latter can also be derived directly from Newtonian mechanics and introduction of the density regularization. Finally, a parameter $\theta$ may be used to put the two equations in the common formulation: $$\begin{aligned}
\label{eq: system}
\ddot x = - F_\theta(\tilde \rho(x)) \int_\Omega \nabla W_h(x-y)\, {\textup d}\mu(y) - \theta \int_\Omega F_\theta(\tilde \rho(y)) \nabla W_h(x-y)\, {\textup d}\mu(y),\end{aligned}$$ for $\theta \in \{0,1\}$, where $F_0(\tilde \rho) = 1/\tilde \rho \, d(P(\tilde \rho))/{\textup d}\tilde \rho$ and $F_1(\tilde \rho) = P(\tilde \rho)/\tilde \rho^2$.
Convergence result {#sec: Conv}
==================
The *Wasserstein distance* between two probability measures $\mu_1$ and $\mu_2$ is defined as: $$\mathcal W(\mu_1, \mu_2) = \inf_{\pi \in \Pi(\mu_1,\mu_2)} \int_{{\mathbb R}^d \times {\mathbb R}^d} |\chi-\psi| \, \pi({\textup d}\chi,{\textup d}\psi),$$ where $\Pi(\mu_1,\mu_2)$ is the set of all *joint representations* of $\mu_1$ and $\mu_2$. Joint representations are also called *couplings* and are defined such that for each $i=1,2$, $
\int_{{\mathbb R}^d\times {\mathbb R}^d} f(\chi_i) \, \pi({\textup d}\chi_1,{\textup d}\chi_2) = \int_{{\mathbb R}^d} f(\chi) \, {\textup d}\mu_i(\chi),
$ for all measurable, bounded functions $f$ on ${\mathbb R}^d$. As said in the introduction, effectively, the Wasserstein distance computes the (optimal) cost of transforming one probability measure into another. An exposition on the Wasserstein distance and the related concept of *optimal transport*, can be found in [@Villani2009]. Considering the system of (\[eq: system\]) and (\[eq: mass\_conservation\]), written with respect to the absolutely continuous measure $\mu_t$, and the corresponding system written for the discrete measure $\mu_t^N$ from (\[eq: discrete\]), the proof in Evers *et al.* [@Evers2015] establishes that: $$\sup_{t\in [0,T]}\mathcal{W}(\mu^N_t,\mu_t) \rightarrow 0, \,\,\, \textup{as} \,\,\, N\rightarrow \infty,$$ provided that we can approximate the initial measure arbitrarily well. This holds for both, the classical SPH scheme and the non-conservative one of Di Lisio *et al.* [@DiLisio1998] as well; $\theta = 1$ and $\theta=0$ respectively in (\[eq: system\]). For clarity, here we explicitly write the dependence of the measures on time.
The paradigms of the numerical illustration of Section \[sec: numerics\] are performed for a series of increasing particle numbers $N_k$. For each computation, after normalizing the total mass of the system, at each time instance $t\in I:= \{\varphi T/10$, $\varphi=\{0,...,10\}\subset {\mathbb N}\}$, we solve a linear programming problem to calculate: $$\label{eqn: approx sup wass}
M_{k,k+1} := \max_{t\in I} \mathcal{W}(\mu^{N_k}_t,\mu^{N_{k+1}}_t) \approx \sup_{t\in [0,T]}\mathcal{W}(\mu^{N_k}_t,\mu^{N_{k+1}}_t), \hspace{0.1in} C_{k+1}^{(d)} := \log_{\frac{N_{k+1}}{N_k}} \Big ( \frac{M_{k+1,k+2}}{M_{k,k+1}} \Big ).$$ The theoretically predicted convergence rate is the same as for the initial measure, i.e. $O(N^{-1/d})$, whence we expect that $C^{(d)}_{k+1}$ tends to the value $-1/d$.
A critical point of the theoretical result of Evers et al. [@Evers2015] (and Di Lisio *et al.* [@DiLisio1998]) is that it makes no conclusion on the smoothing length $h$. The convergence proof is achieved for $h$ fixed with the number of particles, and the dependence of $h$ on $N$ is not investigated. It is known that in order for the regularized equations of hydrodynamics to approximate the real physics well, $h$ should be sufficiently small. In the SPH literature (e.g. [@Monaghan2005]), it is common practice to achieve this by taking $h = \eta \, N^{-1/d}$, with parameter $1.2\leq\eta\leq1.5$, for Gaussian-like kernels. By extension, cases of spatially and temporally varying $h$, like those used in shock problems [@Monaghan2005], are not covered by the theoretical result.
Additional processes and limitations of the theoretical result {#sec: add}
==============================================================
The theoretical proof of convergence [@Evers2015] covers cases broader than the one discussed in Section \[sec:SPH from cm\]. First and foremost, the form of the potential energy covered by the theoretical proof of convergence is $
e^* = e^*(\rho(x),x) = e(\rho(x)) + u(x),
$ with $e$ the internal energy of the medium given by $\partial e/\partial \rho = P(\rho)/\rho^2$, as discussed in Section \[sec:SPH from cm\], and $u$ an external field, such as gravity. A limitation of the theoretical proof is that the admissible equations of state are of the form $
P(\rho) = \mathcal{K}\rho^\gamma,
$ where $\mathcal{K}$ is a parameter and $\gamma$ is the so-called *polytropic exponent*, which needs to satisfy $\gamma > 1$. It should be underlined that the theoretical proof is not conclusive about equations of state in the form $P(\rho) = B \cdot((\rho/\rho_0)^\gamma - 1)$, which are typically employed in SPH computations for the modeling of water [@Monaghan2005].
Processes involving $e^*$ are conservative, and therefore, the related equation of motion follows the procedures of Section \[sec:SPH from cm\]. Apart from them, the proof of convergence covers processes described by the equation: $
\updelta \mathcal S |_{0}^{T} = -\updelta \mathcal Q |_{0}^{T},
$ where: $$\updelta \mathcal Q |_{0}^{T} = \int_{0}^{T} \int_\Omega \Big ( - \nu(x) \, \dot x + \int_\Omega K(x-y) \, {\textup d}\mu(y) \Big ) \, \updelta x \,{\textup d}\mu(x) \, {\textup d}t.$$ In the general case $K$ can be an anisotropic kernel, $K(x-y) \neq K(\|x-y\|)$, describing non-local interactions within the system. On the other hand, $\nu = \nu(x)$ is a dissipative term, without non-local characteristics. It should be stressed that this is in contrast to the non-locality of the dissipative term: $$\label{eq: nonlocal_dissipation}
\sim \int_\Omega (\dot x - \dot y) \nabla^2 W_h(x-y) \, \frac{{\textup d}\mu}{\tilde \rho} \Big |_{(y)},$$ typically constructed (via approximation of $\nabla^2 W_h$) in SPH to model viscosity (e.g. [@Monaghan2005]). Note that this construction further assumes the approximation: ${\textup d}\lambda = {\textup d}\mu/\rho \approx {\textup d}\mu/\tilde \rho$, which is necessary if in the SPH-approximation of the continuity equation [@Monaghan2005; @Monaghan2013] the latter approximation is preferred to describe mass conservation (see Section \[sec: numerics\]) over the temporal evolution of (\[eq: mass\_conservation\]).
Numerical paradigms {#sec: numerics}
===================
We construct the initial measure $\mu^N_0$, corresponding to the $N$-particle approximation of $\mu_0$, according to a partitioning of the initial domain in $N$ subdomains of incremental volume $V_i$. Masses are assigned as $m_i = \rho_0(x_i)\,V_i$ for each $i=1,\ldots,N$. In Evers *et al.* [@Evers2015 Section 3.5] we show two formal ways of constructing the sequence $\mu^N_0$ such that it converges to $\mu_0$ at rate $\mathcal{O}(N^{-1/d})$; they correspond to particle initialization strategies typically used in the SPH literature. The theoretical convergence result [@Evers2015] establishes that the corresponding solutions $\mu^N$ converge at the same rate. In Evers *et al.* [@Evers2015], test cases which conform to the assumptions of the proof are examined. The cases that follow here, suggest that the same theoretical results may be expected to hold also for cases that are typically used for benchmarking SPH algorithms, but do not satisfy all the assumptions of the convergence theorem in [@Evers2015]. The evolution of an elliptical drop is a benchmark problem for weakly compressible flows, which admits analytical solution [@Monaghan1994]. It refers to an initially circular water drop which attains an elliptical shape under a shearing velocity field. The problem involves a conservative part, with $P(\rho) = B \cdot ((\rho/\rho_0)^7 - 1)$, and a dissipative part. The numerical recipe is considered standard in the current SPH literature. For the motion due to the hydrodynamic conservative part we use (\[eq: sph\_motion1\]) and dissipation is modeled with the analogous term of Monaghan and Raffie [@Monaghan2013], which pertains to (\[eq: nonlocal\_dissipation\]). Additionally, the Wendland kernel [@Monaghan2013], and a leapfrog time integrator —preferred for its symplectic nature— are used, with $h = 1.5N^{-1/2}$ assigned to all particles. Furthermore, we employ the artificial mass-flux term of Zisis *et al.* [@Zisis2015a] with the corresponding parameters $\alpha = 0.5$ and $\beta = 0$, to counteract oscillations in the density profile. We examined two different equations for mass conservation: 1) the discretized temporal evolution of (\[eq: mass\_conservation\]); 2) the discrete SPH-approximation of the continuity equation [@Monaghan2013]. Results are practically indistinguishable. The left plot of Figure \[fig: droplet\] shows the upper half plane of the problem for $N=7232$ at normalized time $t=0.0076$, when Monaghan [@Monaghan1994] records the height of the semi-major axis. He finds the latter height $1.91$, compared to $1.95$ of the analytical result (black dashed horizontal line in Figure \[fig: droplet\]) and our $1.93$. We follow the process until normalized final time $t=0.01$, achieved with a time step $\Delta t= 10^{-6}$. The right plot of Figure \[fig: droplet\] shows the convergence rates $C^{(2)}_{k+1}$ of the initial measure $\mu_0^N$ and the final one $\mu_T^N$, with respect to the Wasserstein distance between particle systems of successive particle numbers $N_k\in\{2,12,32,52,112,208,448,812,1804,3228,7232\}$. The convergence rates oscillate around the theoretically predicted value $-1/2$, and they tend to become identical. Computing the Wasserstein distance for higher $N_k$ becomes computationally too expensive for our brute-force algorithm. In order to fill the the initial circle (red dashed circle in Figure \[fig: droplet\]) with $N_k$ particles, we use $\ell_k=\{2,4,6,8,12,16,24,32,48,64,96\}$ particles per unit length to pack particles within a larger square and then disregard all particles falling outside. Recall that the theoretical result (Sections 3 and 4) does not support the current form of dissipation, the equation of state, or the specific functional dependence of $h$ on $N$ —in the proof $h$ is assumed to be a fixed parameter for all examined $N_k$. Moreover, we obtained indistinguishable results using the SPH-approximation of the continuity equation [@Monaghan2013], for which the theoretical result is not applicable. Yet, our numerical results provide evidence that typically used weakly compressible SPH schemes converge with respect to the Wasserstein distance. Rigorous proofs are left for future work. Our conjecture that the scaling $h\sim N^{-1/d}$ is the correct one, may serve as a guideline.
{width="100.00000%"}
{width="100.00000%"}
\[fig: droplet\]
{width="75.00000%"}
{width="100.00000%"}
\[fig: shocktube\]
The *shock tube test* is a classic one-dimensional test, frequently employed for the validation of fully compressible SPH schemes [@Monaghan2005; @Price2012; @Zisis2015a]. In this test, there is a discontinuity in the density profile of the medium, with $\rho_0(x_0<0.5) = 1$ and $\rho_0(x_0\geq0.5) = 0.125$. We construct the initial density profile with particles of equal masses and solve the SPH system using the differential mass conservation coming from (\[eq: mass\_conservation\]) and the equation of motion (\[eq: sph\_motion1\]), with varying smoothing length $h_i:=1.2 \, m_i/\tilde \rho_i$. The complete solution strategy can be found in Zisis *et al.* [@Zisis2015a] and falls within the standard framework [@Monaghan2005; @Price2012]. The typical resolution is $450$ particles in total [@Price2012; @Zisis2015a] and therefore, we examine convergence with respect to the Wasserstein distance for $N_k \in \{18, 45, 90, 225, 450, 900, 1800\}$, to the theoretical value $C^{(1)}_{k+1} = -1$. The distinct characteristic of this case is that $h$ varies spatially and temporally. This is not supported by the theoretical result of convergence [@Evers2015], neither are the —necessary for the solution— artificial dissipative terms in all variables. Nonetheless, in Figure \[fig: shocktube\] the system is shown to converge in a manner similar to the prediction of the theoretical result. The density profile for the highest resolution is also presented in Figure \[fig: shocktube\], against the analytical solution (red solid line) at final normalized time $t=0.2$.
Conclusions {#sec: concl}
===========
The present paper summarizes the authors’ results [@Evers2015], regarding the derivation and convergence of SPH. It focuses on describing the three ways to obtain SPH from continuum mechanics via a formulation based on measures. The theoretical convergence is established with respect to the Wasserstein distance, as the number of particles increases, similarly to Di Lisio *et al.* [@DiLisio1998]. In fact, the older result is extended by including external fields, local dissipation and non-local interaction forces of the system.
Perhaps the most important limitation of the theoretical proof is that it is not conclusive regarding the convergence of the SPH system as the number of particles grows to infinity and the smoothing length goes to zero at the same time. It rather holds for examining a fixed value of the smoothing length as the number of particles goes to infinity. Additionally, open problems are the inclusion of the following features: dissipation with a non-local character (similar to the one typically used in SPH to mimic viscosity of fluids); equation of states for liquids and solids; spatially varying smoothing length.
Last but not least, the present paper introduces an innovation, by supporting the theoretical proof with numerical evidence. Calculations of the Wasserstein distance reveal that the theoretically predicted convergence rate is observed in the SPH solutions of widely used weakly and fully compressible tests.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We present a group-theoretic criterion under which one may verify the Artin conjecture for some (non-monomial) Galois representations, up to finite height in the complex plane. In particular, the criterion applies to $S_5$ and $A_5$ representations. Under more general conditions, the technique allows for the possibility of verifying the Riemann hypothesis for Dedekind zeta functions of non-abelian extensions of ${\mathbb{Q}}$.
In addition, we discuss two methods for locating zeros of arbitrary $L$-functions. The first uses the explicit formula and techniques developed in [@bs] for computing with trace formulae. The second method generalizes that of Turing for verifying the Riemann hypothesis. In order to apply it we develop a rigorous algorithm for computing general $L$-functions on the critical line via the Fast Fourier Transform.
Finally, we present some numerical results testing Artin’s conjecture for $S_5$ representations, and the Riemann hypothesis for Dedekind zeta functions of $S_5$ and $A_5$ fields.
address: |
Mathematics Department\
530 Church Street\
University of Michigan\
Ann Arbor, MI 48109
author:
- 'Andrew R. Booker'
bibliography:
- 'turing.bib'
title: 'Artin’s conjecture, Turing’s method and the Riemann hypothesis'
---
[^1]
Introduction
============
Artin’s conjecture
------------------
Let $K/{\mathbb{Q}}$ be a Galois extension and $\rho:{{\rm Gal}}(K/{\mathbb{Q}})\to{{\rm GL}}_n({\mathbb{C}})$ a non-trivial, irreducible representation of its Galois group. In [@artin], Artin associated to this data an $L$-function $L(s,\rho)$, defined initially for $\Re(s)>1$, which he conjectured to continue to an entire function and satisfy a functional equation. By a theorem of Brauer [@brauer], one now knows the [*meromorphic*]{} continuation and functional equation of Artin’s $L$-functions. The question remains whether they can have poles in the critical strip $0<\Re(s)<1$.
Artin established his conjecture for the [*monomial*]{} representations, those induced from a $1$-dimensional representation of a subgroup; this of course includes all $1$-dimensional $\rho$, in which case $L(s,\rho)=L(s,\chi)$ for a Dirichlet character $\chi$. Although the conjecture has not been decided in any dimension $\ge 2$, more evidence is provided in dimension $2$ by the Langlands-Tunnell theorem [@langlands; @tunnell], which affirms the conjecture for those representations whose image in ${{\rm PGL}}_2({\mathbb{C}})$ is isomorphic to $A_4$ (tetrahedral) or $S_4$ (octahedral); only the $A_5$ (icosahedral) case remains. When $\rho$ is an [*odd*]{} icosahedral representation, meaning $\det\rho$ determines an odd Dirichlet character, infinitely many examples of Artin’s conjecture are known from the work of Taylor et al. [@taylor1; @taylor2].
Moreover, in the odd $2$-dimensional case, there is an algorithm for verifying the conjecture, as follows. By a construction of Deligne and Serre [@deligne-serre], given a holomorphic modular form $f$ of weight $1$, one may associate an odd 2-dimensional representation $\rho$ such that $L(s,f)=L(s,\rho)$. Conversely, every odd $2$-dimensional $\rho$ such that $L(s,\rho)$ is entire arises from the Deligne-Serre construction. For any particular $\rho$, one can search for the associated form; once found, comparing the representation constructed by Deligne-Serre to $\rho$ via an effective version of the Cebotarev density theorem allows one to deduce the conjecture for $\rho$. This and other related techniques have been carried out in a number of cases; see [@buhler; @kiming; @jehanne; @stein].
On the other hand, if one considers [*even*]{} $2$-dimensional representations, the situation is somewhat different. There as well the conjecture has been established for all but the icosahedral cases. However, the correspondence is not with holomorphic forms, but rather Maass forms of eigenvalue $\frac14$. Unfortunately, no analogue of the result of Deligne and Serre is known in that setting. Moreover, computation of the associated forms remains elusive; existing techniques (see e.g. [@bsv]) only allow one to calculate Maass forms to within a prescribed precision, never exactly. Thus, at present this approach does not yield an algorithm for verifying Artin’s conjecture.
The apparent difference between these two cases leads naturally to the following question: Given a Galois representation $\rho$, is there an algorithm that will decide in finite time, with proof, whether $L(s,\rho)$ is entire? Note that like the Riemann hypothesis, Artin’s conjecture is falsifiable, i.e. it may be disproven by observing a counterexample, in this case a pole. The challenge is thus to find a way of demonstrating the conjecture when true.
Although we are unable to provide a definitive answer to this question, one approach, at least for $2$-dimensional representations, is suggested by a theorem from [@booker]: If a given $2$-dimensional $\rho$ is not associated to a holomorphic or Maass form as above, then $L(s,\rho)$ has infinitely many poles. In particular, once $L(s,\rho)$ has at least one pole, it must have infinitely many. Unfortunately, the result is ineffective, in the sense that it does not predict where the first pole must occur. A natural question, therefore, is whether an effective version of this theorem exists. First, however, we must consider exactly what that would mean; since the only handle that we have on an Artin $L$-function in the critical strip is as the ratio of entire functions given by Brauer’s theorem, it is not immediately clear that we can check its holomorphy at a zero of the denominator without [*a priori*]{} knowing a lower bound on the residue of any poles.
In this paper we address precisely this issue, in Section \[sec:artin\]. There we present a criterion which, when satisfied, yields an algorithm for verifying the holomorphy of an Artin $L$-function up to a given height in the critical strip. In particular, we give the first direct evidence (as far as we are aware) of holomorphy in the critical strip of an $L$-function for which the conjecture cannot be established through the methods mentioned above. Although our criterion is not always satisfied, we are in general able to deduce partial information, such as a bound on the multiplicities and residues of possible poles. Moreover, the limitations of the information that we obtain give an idea of the hypotheses that one would have to impose in any effective version of the converse theorem in order to make the above approach work.
Turing’s method and the Riemann hypothesis
------------------------------------------
One application of our criterion is to the Riemann hypothesis for Dedekind zeta functions. Turing [@turing] devised a method for checking the hypothesis in a bounded region for the Riemann $\zeta$ function.[^2] The method depends on the simplicity of the zeros of $\zeta$. Because of that, it is only directly extendable to Dedekind zeta functions of non-normal extensions of small degree (see [@tollis]) or abelian extensions, for which it is more natural to verify the hypothesis for the associated Dirichlet $L$-functions instead (see [@rumely]).
Similarly, for a non-abelian extension one can factorize the zeta function into Artin $L$-functions of irreducible representations. As these are also expected to have simple zeros, Turing’s method applies, provided one assumes the Artin conjecture. However, combining our criterion with Turing’s method, we will in some cases be able to deduce the Riemann hypothesis and holomorphy of the relevant Artin $L$-functions simultaneously. In fact, as we will see, there are even cases where we may check the Riemann hypothesis without being able to verify Artin’s conjecture. We carry out the necessary generalization of Turing’s method in Section \[turing\].
Rigorous zero computations
--------------------------
In order to implement these ideas, we develop, in Sections \[sec:explicit\] and \[sec:rigorous\], two methods of locating zeros of $L$-functions. The first uses the explicit formula and techniques developed for the Selberg trace formula in [@bs]. If one assumes the Riemann hypothesis, this method may be used with our criterion, in place of Turing’s method, for verifying the Artin conjecture. More importantly, the explicit formula is clean to implement and yields estimates for low zeros quickly. It can thus serve as a check for later computations, or to fine tune the parameters of Turing’s method for greater speed.
The second method is a technique for fast, rigorous computations of $L$-functions on the critical line. This is a hard problem in general, basically because of the difficulty of providing uniform, effective bounds for the relevant Mellin transforms. By making use of the Fast Fourier Transform, our technique allows one to compute many values of the same $L$-function simultaneously, which is particularly appropriate for Turing’s method. In doing so, we need only consider a single Mellin transform, making rigorous computation more practical. In addition, the method has complexity comparable to that of computing a [*single*]{} value by the approximate functional equation.
Although our primary interest is in Artin $L$-functions, we carry out the details of Sections \[sec:explicit\], \[turing\] and \[sec:rigorous\] for [*arbitrary*]{} $L$-functions $L(s)$, in the hope that the results may be useful outside of the present context. More precisely, we make the following assumptions, notations and conventions throughout:
- $L(s)$ is given by an Euler product of degree $r$: $$L(s)=\prod_{p\,\mbox{\scriptsize prime}}
\frac1{(1-\alpha_{p,1}p^{-s})\cdots(1-\alpha_{p,r}p^{-s})},
\label{eulerprod}$$ where the $\alpha_{p,j}$ are complex parameters satisfying the individual bound $|\alpha_{p,j}|\le p^{1/2}$, and the product is absolutely convergent for $\Re(s)>1$. Further, for all but finitely many $p$, there is a pairing $\alpha\mapsto\alpha'$ such that $|\alpha_{p,j}\alpha_{p,j}'|=1$. For the exceptional $p$, such a pairing exists for a subset of the $\alpha_{p,j}$, and those not in the subset satisfy $|\alpha_{p,j}|\le 1$.
- Define $$\Gamma_{{\mathbb{R}}}(s) := \pi^{-s/2}\Gamma\!\left(\frac{s}2\right),
\qquad
\gamma(s) := \epsilon N^{\frac12(s-\frac12)}\prod_{j=1}^r\Gamma_{{\mathbb{R}}}(s+\mu_j),
\qquad \Lambda(s) := \gamma(s)L(s),$$ where $|\epsilon|=1$, $N$ is a positive integer and $\Re(\mu_j)\ge -\frac12$. For a certain choice of these parameters, $\Lambda(s)$ has meromorphic continuation to ${\mathbb{C}}$, is a ratio of entire functions of order $1$, and satisfies the functional equation $$\Lambda(s) = {\overline{\Lambda}}(1-s),
\label{funceq}$$ where for a complex function $f$ we denote by $\overline{f}(s)$ the function $\overline{f(\bar{s})}$. Note that $\epsilon$ here is the square root of the usual root number, and is only defined up to multiplication by $\pm1$; we choose the value with argument in $\bigl(-\frac{\pi}2,\frac{\pi}2\bigr]$. Including $\epsilon$ as part of the $\gamma$ factor makes $\Lambda(s)$ real for $\Re(s)=\frac12$, as can be seen from .
Let $Q(s)$ be the [*analytic conductor*]{}: $$Q(s):=N\prod_{j=1}^r\frac{s+\mu_j}{2\pi}.$$ Note that $\gamma(s)$ satisfies the recurrence $\gamma(s+2)=Q(s)\gamma(s)$. Further, we define $$\chi(s):=\frac{{\overline{\gamma}}(1-s)}{\gamma(s)},$$ so that $L(s)=\chi(s){\overline{L}}(1-s)$.
- $L(s)$ may have at most finitely many poles, which we assume to lie along the line $\Re(s)=1$. We label them $1+\lambda_k$ with $\lambda_k\in i{\mathbb{R}}$, $k=1,\ldots,m$, repeating with the appropriate multiplicity. Further, from the functional equation , each $\lambda_k$ will equal $-\mu_j$ for some $j$, counting multiplicity; in particular, $m\le r$. We set $$P(s) := \prod_{k=1}^m(s-\lambda_k),$$ so that $P(s)P(s-1)\Lambda(s)$ is entire.
- Some progress is known toward the Ramanujan conjecture for $L$; that is, there exists $\theta<\frac12$ such that $$|\alpha_{p,j}| \le p^{\theta}
\qquad\mbox{and}\qquad
\Re(\mu_j)\ge -\theta
\label{ramanujan}$$ for all $p,j$. This assumption is not strictly necessary, as we could instead use average bounds of Rankin-Selberg type. However, bounds of the form are now known in the cases of greatest interest (automorphic $L$-functions [@lrs]), and the results are easier to state and use assuming it.
Numerical results
-----------------
Finally, in Section \[numerics\] we describe the implementation of the above ideas and give some numerical results of tests of the Riemann hypothesis for a few $S_5$ and $A_5$ extensions in the region $|\Im(s)|\le 100$. For the $S_5$ cases, this includes a verification of Artin’s conjecture in the same region for the $L$-functions of all representations of the group.
Acknowledgements {#acknowledgements .unnumbered}
================
The idea for this paper arose from a conversation with Laurent Clozel, and it is a pleasure to acknowledge him here. I thank Harold Stark, who has considered problems of this nature in the past and whose interest motivated me to finish the project. Thanks also to Bob Griess, Martin Isaacs, Peter Sarnak and Kannan Soundararajan for helpful discussions.
A criterion for verifying Artin’s conjecture {#sec:artin}
============================================
Let $\rho:{{\rm Gal}}(K/{\mathbb{Q}})\to{{\rm GL}}_n({\mathbb{C}})$ be a Galois representation, as in the introduction. Brauer’s theorem expresses the $L$-function $L(s,\rho)$ as a ratio $N(s)/D(s)$, where $N(s)$ and $D(s)$ are Artin $L$-functions associated to sums of monomial representations. If $f(s)$ is any holomorphic Artin $L$-function, we have a formula for the number $N_f(t_1,t_2)$ of zeros of $f$ between heights $t_1$ and $t_2$, from the argument principle: $$N_f(t_1,t_2)=\frac1{2\pi i}\int_C\frac{f'}{f}(s)\,ds,
\label{eq:argp1}$$ where $C$ is the rectangular contour with vertices at $2+it_1$, $2+it_2$, $-1+it_2$, $-1+it_1$ and counter-clockwise orientation. We also have available in this case algorithms to compute $f$ and $f'$ at an arbitrary point in the complex plane; see Section \[sec:rigorous\]. Thus, in principle we could compute exactly by numerical integration. Although has the advantage of applying in great generality, to do so would be inefficient and difficult to implement rigorously. In the special case that the zeros of $f$ are simple, a much more efficient algorithm was given by Turing; see Section \[turing\].
No matter how we arrive at the numbers $N_f(t_1,t_2)$, there is always some uncertainty in the locations of the zeros of $f$. In this is due to the fact that as $t_i$ approaches the ordinate of a zero, higher and higher precision is needed in order to compute $f'/f$ accurately. This is in line with the expectation that the general zero is transcendental, meaning that one can never know it exactly.
For $L(s,\rho)$, we can recover the [*net*]{} number of zeros (i.e. zeros minus poles) between heights $t_1$ and $t_2$ as $N_N(t_1,t_2)-N_D(t_1,t_2)$. If Artin’s conjecture is true then for every zero of $D(s)$ there is a zero of $N(s)$ at the same point. However, because of the uncertainty in the locations of the zeros of $N(s)$ and $D(s)$, from this computation alone we cannot rule out the possibility that $L(s,\rho)$ has a pole with a zero very close by in the neighborhood of a zero of $D(s)$. In other words, we can only observe the counts of net zeros in these small neighborhoods.
Fortunately, there are some restrictions on potential poles. For instance, the Dedekind zeta function of the extension, $\zeta_K(s)$, factors into Artin $L$-functions: $$\zeta_K(s) = \prod_{\rho}L(s,\rho)^{\dim\rho},$$ where the product is over all irreducible representations of ${{\rm Gal}}(K/{\mathbb{Q}})$. Since $\zeta_K(s)$ is holomorphic (except for a simple pole at $s=1$), we see that any pole of $L(s,\rho)$ in the critical strip must be located at the zero of another function. More generally, if $\sigma$ is any representation, we have $$L(s,\sigma) = \prod_{\rho}L(s,\rho)^{\langle\sigma,\rho\rangle},$$ where $\langle\cdot,\cdot\rangle$ is the inner product on the space of characters, and by abuse of notation we write $\langle\sigma,\rho\rangle$ for $\langle{{\rm Tr}\,}\sigma,{{\rm Tr}\,}\rho\rangle$. When $\sigma$ is monomial, we again have $L(s,\sigma)$ holomorphic with the possible exception of a pole at $s=1$.
This information is described most concisely by use of the [*Heilbronn (virtual) character*]{}: For $s_0\in{\mathbb{C}}\setminus\{1\}$, define $$\theta_{s_0} = \sum_{\rho}{{\rm ord}}_{s=s_0}L(s,\rho)\cdot{{\rm Tr}\,}\rho,$$ where ${{\rm ord}}_{s=s_0}L(s,\rho):={{\rm Res}}_{s=s_0}\frac{L'}{L}(s,\rho)$. Thus, $${{\rm ord}}_{s=s_0}L(s,\sigma)=\langle\theta_{s_0},\sigma\rangle
\ge 0\quad\mbox{for all monomial }\sigma.
\label{eq:monoholo}$$ The study of Heilbronn characters leads to many useful results. For example, in [@foote-murty] it is shown that $$\sum_{\rho}\bigl({{\rm ord}}_{s=s_0}L(s,\rho)\bigr)^2
\le \bigl({{\rm ord}}_{s=s_0}\zeta_K(s)\bigr)^2.
\label{eq:zetaKbound}$$ In particular, the zeros and poles of each $L(s,\rho)$ are among the zeros of $\zeta_K(s)$.
The idea now is to combine with observations of net zeros. If we look in a small enough neighborhood of a zero of $\zeta_K(s)$, we expect to find one net zero for a single $L(s,\rho)$ and no net zeros for the others. This is based on the assumption that the zeros of different irreducible Artin $L$-functions are distinct and simple. While such a statement is likely impossible to prove, we may use it as a working hypothesis to be tested at run time. This is analogous to assuming the simplicity of the zeros of $\zeta$ in order to check the Riemann hypothesis. (Note that if there were a multiple zero of $\zeta$, it is doubtful that one could distinguish it from a counterexample.)
In other words, if the working hypothesis is true, then our net zero observations correspond to the character ${{\rm Tr}\,}\rho$ for some $\rho$. Thus, we have ${{\rm Tr}\,}\rho=\theta_{s_1}+\ldots+\theta_{s_n}$, where $s_1,\ldots,s_n$ are the distinct zeros of $\zeta_K(s)$ in the neighborhood that we examine. We would like to conclude that there is just one such point, meaning that the actual zero counts agree with our observations. Since the Heilbronn characters satisfy , it is enough to show that $${{\rm Tr}\,}\rho\ne\chi_1+\chi_2\mbox{ for virtual characters }
\chi_i\ne 0\mbox{ with }\langle\chi_i,\sigma\rangle\ge 0
\mbox{ for all monomial }\sigma.
\label{eq:monohyp}$$
The one notable exception to this philosophy is at the central point $\frac12$, where there can be forced vanishing if $\rho$ is self-dual (an example of which is given in [@armitage]). In that case, we expect one zero for each self-dual $\rho$ with an odd functional equation, and no zeros for the rest. However, we can only determine the parity of the order of vanishing at $\frac12$. This leads to the following replacement for condition at $\frac12$: $$\sum_{\substack{\rho\;{\rm self-dual}\\ \Lambda(1-s,\rho)=-\Lambda(s,\rho)}}
{{\rm Tr}\,}\rho\ne\chi_1+2\chi_2\quad\mbox{with }\chi_i\ne 0\mbox{ and }
\langle\chi_i,\sigma\rangle\ge 0\mbox{ for all monomial }\sigma.
\label{eq:monohyp2}$$
When is satisfied for all irreducible representations $\rho$, we may check the holomorphy of all $L(s,\rho)$ at any point at which the working hypothesis turns out to be true. We give a name to describe this situation:
A finite group $G$ is [*almost monomial* ]{}if, for each irreducible representation $\rho$, if ${{\rm Tr}\,}\rho = \chi_1+\chi_2$ for virtual characters $\chi_i$ such that $\langle\chi_i,\sigma\rangle\ge 0$ for all monomial $\sigma$, then either $\chi_1 = 0$ or $\chi_2$ = 0.
The terminology is explained with aid of Figure \[lattice\]. The plane represents the lattice of virtual characters, with the first quadrant being the monoid of characters, and the shaded cone the monoid generated by the monomial characters. We consider all virtual characters within $90$ degrees of the cone, which in the figure is everything within the dashed lines. The group is almost monomial if this set is not much larger than the character monoid, in the precise sense that the irreducible representations, which are the coordinate axis vectors represented by thick arrows, remain indecomposable in this set. Equivalently, the monoid generated by the monomial characters should be close to the full character monoid. From the picture it is easy to see that any monomial group is almost monomial.
One could argue that we should include condition in our definition as well. We prefer to keep it separate, taking the view that it is more important to be able to demonstrate holomorphy at a generic zero of the denominator. Indeed, we have already seen that the $L$-function of a $2$-dimensional representation cannot have a finite number of poles, so we do not lose much generality by excluding a single point. It is plausible that such a result holds for higher dimensions as well. Moreover, condition seems usually to be weaker than almost monomiality; cf. Proposition \[prop:monoex\] below.
A potentially more serious issue is that $N(s)$ and $D(s)$ may have high order zeros at $\frac12$, in which case Turing’s method does not apply. This could be remedied by computing the contour integral around $\frac12$, but we would like to avoid doing so. Fortunately, if the order of vanishing at $\frac12$ is at most $3$, we can still conclude that we have the correct count by sign changes alone; that is because for a self-dual representation, if we miss a zero away from $\frac12$ then we must miss at least four such zeros. Fortunately again, in all cases that we consider, $N(s)$ has at most three irreducible factors with a potential zero at $\frac12$.
Like monomiality, the notion of almost monomiality behaves well under some common group operations. In particular, we have the following.
If $G$ is almost monomial then so are quotients of $G$ and products $G\times H$ for any monomial group $H$. \[prop:monofunc\]
1\. Let $K$ be a normal subgroup of $G$ and $\tilde{\pi}$ an irreducible representation of $G/K$. Suppose that ${{\rm Tr}\,}\tilde{\pi}=\tilde{\chi}_1+\tilde{\chi}_2$, with $\langle\tilde{\chi}_i,\tilde{\sigma}\rangle\ge 0$ for all monomial $\tilde{\sigma}$. Let $\pi$, $\chi_i$ be the lifts of $\tilde{\pi}$, $\tilde{\chi}_i$ to $G$ obtained by composition with the natural projection. Then $\pi$ is irreducible and ${{\rm Tr}\,}\pi=\chi_1+\chi_2$. Further, if $\rho$ is an irreducible representation of $G$ then $\langle\chi_i,\rho\rangle=0$ unless $\rho$ factors through $G/K$. If that is the case, let $\tilde\rho$ denote the induced map on $G/K$.
Now, if $\sigma={{\rm Ind}}_H^G\lambda$ is a monomial representation then, by Frobenius reciprocity, we have $\langle\sigma,\rho\rangle=
\bigl\langle{{\rm Res}}_H^G\rho,\lambda\bigr\rangle$ for all $\rho$ factoring through $G/K$. If at least one of these is non-zero, i.e. $\lambda$ occurs in ${{\rm Res}}_H^G\rho$, then since $\rho$ factors through $G/K$, $\lambda$ must factor through $H/H\cap K
\cong HK/K$. Let $\tilde{\lambda}$ denote the induced map on $HK/K$. Then $\bigl\langle{{\rm Res}}_{HK/K}^{G/K}\tilde{\rho},\tilde{\lambda}\bigr\rangle=
\bigl\langle{{\rm Res}}_H^G\rho,\lambda\bigr\rangle$. Thus, $\tilde{\sigma}={{\rm Ind}}_{HK/K}^{G/K}\tilde{\lambda}$ satisfies $\langle\sigma,\rho\rangle=
\langle\tilde{\sigma},\tilde{\rho}\rangle$. Therefore, $\langle\chi_i,\sigma\rangle
=\langle\tilde{\chi}_i,\tilde{\sigma}\rangle\ge 0$. The conclusion follows by almost monomiality of $G$.
2\. Let $\rho_G$ and $\rho_H$ be irreducible representations of $G$ and $H$, respectively, and suppose that ${{\rm Tr}\,}\rho_G\otimes\rho_H
=({{\rm Tr}\,}\rho_G)({{\rm Tr}\,}\rho_H)=\chi_1+\chi_2$ with $\langle\chi_i,\sigma\rangle_{G\times H}\ge 0$ for all monomial $\sigma$. Taking the inner product over $H$ with $\rho_H$, we get ${{\rm Tr}\,}\rho_G=\langle\chi_1,\rho_H\rangle_H+\langle\chi_2,\rho_H\rangle_H$.
Next, if $\sigma_G$ is any monomial representation of $G$, we have $\bigl\langle\langle\chi_i,\rho_H\rangle_H,\sigma_G\bigr\rangle_G
=\langle\chi_i,\sigma_G\otimes\rho_H\rangle_{G\times H}
\ge 0$, since $\sigma_G\otimes\rho_H$ is monomial. Thus, since $G$ is almost monomial, we have $\langle\chi_i,\rho_H\rangle=0$ for some $i$.
Similarly, if $\rho_H'$ is any other irreducible representation of $H$, we find $0=\langle\chi_1,\rho_H'\rangle_H+\langle\chi_2,\rho_H'\rangle_H$. Thus, $\langle\chi_1,\rho_H'\rangle=\langle\chi_2,\rho_H'\rangle=0$. Therefore $\chi_i=0$ for some $i$.
The next proposition shows that the class of almost monomial groups is strictly larger than that of monomial groups.
The groups ${{\rm SL}}_2({\mathbb{F}}_3)$, $A_5$ and $S_5$ are almost monomial and satisfy . \[prop:monoex\]
These are shown with the aid of the computer algebra system GAP [@GAP]. We illustrate the general procedure for checking almost monomiality for a given group with the example $A_5$. Note first that $A_5$ has five irreducible representations, of dimensions $1$, $3$, $3$, $4$ and $5$. We use GAP to determine all monomial representations. In this case they are non-negative linear combinations of the vectors $(1,0,0,0,0)$, $(0,0,0,0,1)$, $(1,0,0,1,0)$, $(0,1,1,0,0)$, $(0,1,0,1,1)$, $(0,0,1,1,1)$, and $(0,1,1,1,0)$, where the components indicate the multiplicities of the irreducible representations. We label the monomial representations associated to these vectors $\sigma_1,\ldots,\sigma_7$. The first five form a ${\mathbb{Z}}$-basis for the lattice of virtual characters, i.e. any virtual character $\chi$ may be written uniquely as an integral linear combination $\chi=\sum_{i=1}^5x_i{{\rm Tr}\,}\sigma_i$.
Now, almost monomiality is equivalent to the assertion that for each irreducible representation $\rho$, whenever $0\le\langle\chi,\sigma\rangle\le\langle\rho,\sigma\rangle$ for all monomial $\sigma$, we have either $\chi=0$ or $\chi={{\rm Tr}\,}\rho$. Using our integral basis, we investigate the solutions to $$0\le\sum_{i=1}^5x_i\langle\sigma_i,\sigma_j\rangle
\le\langle\rho,\sigma_j\rangle
\label{eq:A5proof}$$ for $j=1,\ldots,7$. Restricting to $j=1,\ldots,5$, we get an invertible system, i.e. the matrix $A=\bigl(\langle\sigma_i,\sigma_j\rangle\bigr)_{1\le i,j\le5}$ lies in ${{\rm SL}}_5({\mathbb{Z}})$. We consider the vectors ${\bf x}=A^{-1}{\bf y}$ for all ${\bf y}=(y_1,\ldots,y_5)$ satisfying $0\le y_j\le\langle\rho,\sigma_j\rangle$. By construction, these satisfy for $j=1,\ldots,5$. We check that the only ${\bf x}$ satisfying for $j=6,7$ are $0$ and $A^{-1}\bigl(\langle\rho,\sigma_j\rangle\bigr)$, corresponding to $\chi=0$ and $\chi={{\rm Tr}\,}\rho$, respectively.
Similarly, for we try all possible combinations of $\rho$ having odd functional equation. We may exclude those whose $L$-functions may be expressed in terms of Dedekind zeta functions, for which the root number is always $1$. For $A_5$, the only non-trivial possibility is that the two $3$-dimensional representations have odd functional equation.
With the evidence provided by Propositions \[prop:monofunc\] and \[prop:monoex\], one might hope that all groups are almost monomial. That is not the case, as the counterexamples ${{\rm GL}}_2({\mathbb{F}}_3)$ and ${{\rm SL}}_2({\mathbb{F}}_5)$ show. ${{\rm SL}}_2({\mathbb{F}}_5)$ has irreducible representations of dimensions $1$, $2$, $2$, $3$, $3$, $4$, $4$, $5$ and $6$, and it is the smallest group supporting an icosahedral representation (since $A_5$ has no $2$-dimensional representations), meaning that our criterion unfortunately does not apply to checking the icosahedral case. In fact, one knows Artin’s conjecture for [*all*]{} induced representations of this group; while they are not all monomial, the only exceptions come from a pair of tetrahedral representations, for which we have the Langlands-Tunnell theorem. Even with this added information, we cannot rule out the possibility of a simple pole with undetectably small residue at a zero of the $L$-function of the $6$-dimensional representation. More precisely, we find with GAP that the induced representations are spanned by the twelve vectors $$\begin{aligned}
&(1,0,0,0,0,0,0,0,0)\qquad
&(0,0,0,0,0,0,0,1,0)\qquad
&(1,0,0,0,0,1,0,0,0)\\
&(0,0,0,1,1,0,0,0,0)\qquad
&(0,0,0,1,0,1,0,1,0)\qquad
&(0,0,0,0,1,1,0,1,0)\\
&(0,0,0,1,1,1,0,0,0)\qquad
&(0,0,0,0,0,0,0,0,1)\qquad
&(0,0,0,0,0,0,1,0,1)\\
&(0,1,0,0,0,0,1,0,1)\qquad
&(0,0,1,0,0,0,1,0,1)\qquad
&(0,1,1,0,0,0,0,0,1).
\end{aligned}$$ The first seven of these are the monomial representations lifted from ${{\rm SL}}_2({\mathbb{F}}_5)/\{\pm I\}\cong A_5$, while the others give “new” information. One easily checks that for $\rho$ the $6$-dimensional representation, fails with $\chi_1$ corresponding to any of the vectors $(0,-1,0,0,0,0,0,0,1)$, $(0,0,-1,0,0,0,0,0,1)$ and $(0,0,0,0,0,0,-1,0,1)$, i.e. the representations of dimension $2$ and one of dimension $4$ can hide a pole at a zero of $L(s,\rho)$. This shows in a strong sense that information from induced representations is in general insufficient to show Artin’s conjecture.
However, all is not lost concerning icosahedral representations. For a given icosahedral $\rho$, the [*adjoint square*]{} ${{\rm Ad}}(\rho)$ is a $3$-dimensional representation with image isomorphic to $A_5$. A result of Flicker [@flicker] implies that modularity of $\rho$ is equivalent to that of ${{\rm Ad}}(\rho)$. (In fact, modularity of all representations of the underlying group follows from that of ${{\rm Ad}}(\rho)$ and its Galois conjugate, by known cases of functoriality; see [@wang].) Combining this fact with the ${{\rm GL}}(3)$ converse theorem, one could give a converse theorem for ${{\rm GL}}(2)$ using analytic properties of $L(s,{{\rm Ad}}(\rho)\otimes\chi)$ for Dirichlet characters $\chi$. Weissman, in his undergraduate thesis [@weissman], used this idea to give indirect evidence for the modularity of an even icosahedral representation. By Propositions \[prop:monofunc\] and \[prop:monoex\] we see that in principle we may directly verify the holomorphy of these $L$-functions up to finite height. Moreover, an “effective” version of the ${{\rm GL}}(3)$ converse theorem (requiring, say, meromorphy of all twists and holomorphy of a finite number in a bounded region) would suffice to give an algorithm for verifying the conjecture in the icosahedral case.
Unfortunately, there is the more practical problem that totally real $A_5$ fields (those that give rise to even icosahedral representations) are very rare; the smallest known discriminant is far too large to test with current computers. Thus, for the $A_5$ examples that we consider in Section \[numerics\], the Artin conjecture is already known. To test our criterion, we consider instead some examples of $S_5$ extensions, which exist in much greater abundance.
Finally, we note that in the course of verifying Artin’s conjecture, the information that we collect implies that the zeros of each $L(s,\rho)$ are simple and lie on the line $\Re(s)=\frac12$. Thus, in the process we also verify the Riemann hypothesis for $\zeta_K(s)$. Interestingly, we do not need to establish the holomorphy of all $L(s,\rho)$ in order to do this; it is enough, for example, that they have at most simple poles. More precisely, in order to check the Riemann hypothesis around a generic zero of $\zeta_K(s)$ we need to have $${{\rm Tr}\,}\rho\ne\chi_1+2\chi_2\mbox{ for virtual characters }
\chi_i\ne 0\mbox{ with }\langle\chi_i,\sigma\rangle\ge 0
\mbox{ for all monomial }\sigma,$$ which is a weaker condition than almost monomiality. In particular, we may still check the Riemann hypothesis for ${{\rm SL}}_2({\mathbb{F}}_5)$ extensions.
Locating zeros via the explicit formula {#sec:explicit}
=======================================
Let notation be as in the introduction, and define numbers $c_n$ by $-\frac{L'}{L}(s) = \sum_{n=1}^{\infty}c_nn^{-s}$, i.e. $c_n=(\log p)\sum_{j=1}^r\alpha_{p,j}^k$ for $n=p^k$ a prime power, and $c_n=0$ otherwise. Further, we enumerate the zeros of $\Lambda(s)$ as $\rho_n=\frac12+i\gamma_n$ for $n\in{\mathbb{Z}}$, repeated with multiplicity. Weil’s explicit formula relates the sequences $\{c_n\}$ and $\{\gamma_n\}$. Precisely, let $g\in C^1_c({\mathbb{R}})$ be a differentiable function of compact support such that its Fourier transform $h(t):=\int_{-\infty}^{\infty}g(x)e^{ixt}\,dx$ is real for $t\in{\mathbb{R}}$. Then $$\begin{aligned}
\sum_{n\in{\mathbb{Z}}}&h(\gamma_n)
-2\Re\sum_{k=1}^m h\!\left(-i\!\left(\frac12+\lambda_k\right)\!\right)\\
&=g(0)\log N
+2\Re\!\left[\sum_{j=1}^r\frac1{2\pi}\int_{-\infty}^{\infty}
\frac{\Gamma_{{\mathbb{R}}}'}{\Gamma_{{\mathbb{R}}}}\!\left(\frac12+\mu_j+it\right)\!h(t)\,dt
-\sum_{n=1}^{\infty}\frac{c_n}{\sqrt n}g(\log n)\right],
\end{aligned}$$ This follows from the Cauchy integral formula and the functional equation; see [@rudnick-sarnak]. Note that all terms of the formula may be put in terms of $g$; in particular, $$\frac1{2\pi}\int_{-\infty}^{\infty}
\frac{\Gamma_{{\mathbb{R}}}'}{\Gamma_{{\mathbb{R}}}}\!\left(\frac12+\mu+it\right)\!h(t)\,dt
=\frac12\int_0^{\infty}\log\bigl(\pi e^{\gamma}(e^{2x}-1)\bigr)
\,d\bigl(g(x)e^{-(1/2+\mu)x}\bigr).$$ This form is convenient for computation, since $g$ has compact support.
The important thing to note is that given a list of the $c_n$ for $n\le e^X$, the explicit formula gives a method for evaluating $\sum_n
h(\gamma_n)$ for essentially any function $h$ whose Fourier transform is supported in $[-X,X]$. When $X$ is large, we may choose $h$ to be narrowly concentrated around any particular point, and thus resolve features of the spectrum in places where the density of zeros is not too large compared to $X$; a variant of this technique, with explicit test functions (not of compact support), was worked out by Omar [@omar] to estimate the lowest zero of some Dedekind zeta functions. For a fixed support $[-X,X]$, there is a canonical way of choosing a “best” test function, by a method developed for the Selberg trace formula in [@bs]. In order to use the method, which depends crucially on a positivity argument, it is necessary to assume the Riemann hypothesis for our given $L$-function. With that caveat, we recall briefly the construction from [@bs].
For $t_0\in{\mathbb{R}}$, let ${\mathcal C}(X,t_0)$ be the class of functions $h$ as above, with the corresponding $g$ supported in $[-X,X]$, and the additional restrictions $h(t)\ge 0$ for $t\in{\mathbb{R}}$ and $h(t_0)=1$. Define $$F_X(t_0) := \inf_{h\in{\mathcal C}(X,t_0)}\sum_{n\in{\mathbb{Z}}}h(\gamma_n).
\label{eq:FXdef}$$ Then as $X\to\infty$, $F_X$ tends pointwise to the characteristic function of the zeros. Moreover, if $F_X(t_0)<1$ for any value of $X$ then $t_0$ cannot be the ordinate of a zero. Thus, by evaluating $F_X$ we can find provable intervals in which the zeros must lie.
Although the definition of $F_X$ is abstract, it is easy to construct concrete families of functions that closely approximate any desired function. For instance, let $M$ be a large integer, $\delta=X/2M$ and set $$h(t) = \left(\frac{\sin\delta t/2}{\delta t/2}\right)^4
\left(a_0+\sum_{n=1}^{M-1}
\bigl(a_n\cos\delta nt+b_n\sin\delta nt\bigr)\right)^2,$$ for arbitrary real numbers $a_n,b_n$. (For self-dual $L$-functions, we restrict to even test functions, i.e. all $b_n=0$, and divide the final formula by $2$.) On the other side of the Fourier transform, this corresponds to taking $g=f*f$, where $f$ linearly interpolates arbitrary values at multiples of $\delta$.
The sum over zeros in is then a positive definite quadratic form in the numbers $a_n$ and $b_n$. To compute the matrix of the form essentially involves computing the explicit formula for functions $g$ that are translates of a fixed function of small compact support. That requires almost no extra work, since we may compute the formula for all localized test functions simultaneously. Once the matrix is known, the infemum in over this restricted class of test functions is easily found as the minimum of the quadratic form subject to the linear constraint $h(t_0)=1$. This involves inverting the matrix, after which the minimum may be found quickly for many different values of $t_0$.
For an $L$-function of degree $r$ and conductor $N$, the density of zeros at height $T$ is roughly $\frac1{2\pi}\log
N\!\left(\frac{T}{2\pi}\right)^r$. Therefore, in order to resolve features around height $T$, the uncertainty principle says we should know the numbers $c_n$ for $n$ up to about $N\!\left(\frac{T}{2\pi}\right)^r$. In the self-dual case, the extra division by $2$ replaces this by its square root; thus, the complexity is on par with that of the approximate functional equation or the algorithm of Section \[sec:rigorous\], although it is much more sensitive to the local spacing of zeros. (Heuristic arguments based on experiments and random matrix theory [@odlyzko] indicate that the minimum gap between zeros can be arbitrarily small relative to the mean value; although such small gaps are expected to be very rare, we could in principle need many more coefficients than for the “typical” zero at height $T$.) In practice, the explicit formula is clean and easy to implement since there are no error terms to estimate with functions of compact support. It is particularly well-suited to finding low zeros or to situations where the numbers $c_n$ may be computed quickly, as is the case for Artin $L$-functions; cf. Section \[numerics:coeff\].
As mentioned above, the minimization procedure requires assuming the Riemann hypothesis. If one is willing to do so, the method may be made completely rigorous, and may even be used in place of Turing’s method for verifying Artin’s conjecture. However, it is more natural to use it as a quick check in order to fine tune and validate the subsequent rigorous methods. In fact, it is helpful to assume Artin’s conjecture and apply the method to the irreducible Artin $L$-functions directly. That thins out the spectrum, making it easier to isolate individual zeros. We have carried out this procedure for a few examples in Section \[numerics:explicit\].
Turing’s method {#turing}
===============
Turing’s method for verifying the Riemann hypothesis is described well in his paper [@turing], although there are some errors in the details that were later corrected by Lehman [@lehman]. The method has subsequently been extended to Dirichlet $L$-functions by Rumely [@rumely] and Dedekind zeta functions by Tollis [@tollis][^3]. Our contribution is to work out the details necessary to apply it to an arbitrary $L$-function with simple zeros.
Our argument essentially follows that of Turing. We begin by setting some notation to be used only in this section. For $t$ not the ordinate of a zero or pole of $\Lambda$, let $$S(t):=\frac1{\pi}\Im\int_{\infty}^{1/2}\frac{L'}{L}(\sigma+it)\,d\sigma.$$ By convention, we make $S(t)$ upper semi-continuous, i.e. when $t$ is the ordinate of zero or pole, we define $S(t)=\lim_{\varepsilon\to0^+}S(t+\varepsilon)$.
Next, for $t_1<t_2$ let $N(t_1,t_2)$ denote the net number of zeros with imaginary part in $(t_1,t_2]$, counting multiplicity. When neither $t_1$ nor $t_2$ is the ordinate of a zero or pole, we may calculate $N(t_1,t_2)$ using the argument principle, as in . Let $C$ be the rectangle with corners at $2+it_1$, $2+it_2$, $-1+it_2$, $-1+it_1$, with counter-clockwise orientation, $H$ the half plane $\bigl\{s\in{\mathbb{C}}:\Re(s)\ge\frac12\bigr\}$, and $H^c$ its complement. Note that by the functional equation, we have $\overline{\frac{\Lambda'}{\Lambda}(s)}=
-\frac{\Lambda'}{\Lambda}(1-\bar{s})$. Hence, $$\begin{aligned}
N(t_1,t_2)
&=\frac1{2\pi}\Im\int_C\frac{\Lambda'}{\Lambda}(s)\,ds
=\frac1{2\pi}\Im\!\left(
\int_{C\cap H}\frac{\Lambda'}{\Lambda}(s)\,ds
-\overline{\int_{C\cap H^c}\frac{\Lambda'}{\Lambda}(s)\,ds}\right)\\
&=\frac1{2\pi}\Im\!\left(
\int_{C\cap H}\frac{\Lambda'}{\Lambda}(s)\,ds
+\int_{C\cap H^c}\frac{\Lambda'}
{\Lambda}(1-\bar{s})\,d\bar{s}\right)\\
&=\frac1{\pi}\Im\int_{C\cap H}\frac{\Lambda'}{\Lambda}(s)\,ds
=\frac1{\pi}\Im\int_{C\cap H}\frac{\gamma'}{\gamma}(s)\,ds
+\frac1{\pi}\Im\int_{C\cap H}\frac{L'}{L}(s)\,ds.
\label{Ntargprinc}
\end{aligned}$$ Now for the integral of $L'/L$ we move the right edge of the contour out to $\infty$, where the integrand vanishes. We thus obtain $$N(t_1,t_2) = \frac1{\pi}\Im\log\gamma(s)\bigr|_{\frac12+it_1}^{\frac12+it_2}
+S(t_2)-S(t_1).$$
We select a particular branch of $\log\gamma\bigl(s)$ by using the principal branch of $\log\Gamma$. With this choice, set $$\Phi(t) := \frac1{\pi}\left[
\arg\epsilon+\frac{\log N}2t
-\frac{\log\pi}2\!\left(rt+\Im\sum_{j=1}^r\mu_j\right)
+\Im\sum_{j=1}^r
\log\Gamma\!\left(\frac{1/2+it+\mu_j}2\right)\right]
\label{gammabranch}$$ and $$N(t) := \Phi(t)+S(t).$$ Then $N(t_1,t_2)=N(t_2)-N(t_1)$. Note that if $L$ is self-dual and vanishes to order $\le 1$ at $\frac12$ then $N(t)=N(0,t)$. In the general case, although we still have $N(t)\in{\mathbb{Z}}$, there is no standard reference point, so only changes in $N(t)$ are meaningful. (Put another way, the branch of $\log\gamma$ chosen in is non-canonical.) For large $t$, $\Phi(t)$ may be evaluated quickly by an effective version of Stirling’s formula: $$\Im\log\Gamma(z)=\Im\!\left[\left(z-\frac12\right)\log\frac{z}{e}\right]
+\Theta\!\left(\frac1{8|\Im(z)|}\right)
\qquad\mbox{for }z\in{\mathbb{C}}\setminus{\mathbb{R}},$$ where the notation $f=\Theta(g)$ means $|f|\le g$.
Turing’s method is as follows. Recall that $\Lambda\bigl(\frac12+it\bigr)$ is real valued. Thus, if we have an accurate procedure to compute $\Lambda(s)$ then we may locate all simple zeros on the line $\Re(s)=\frac12$ by observing its sign changes. If it turns out that all of the zeros between ordinates $t_1$ and $t_2$ are simple and on the line, then we can deduce the Riemann hypothesis in that interval by computing $N(t_1,t_2)$ (minus the contribution from any poles between $t_1$ and $t_2$) and finding the same number of sign changes over the interval.
To compute $N(t_1,t_2)$, we could evaluate numerically. However, this would require many evaluations of $\Lambda(s)$ and would be difficult to carry out rigorously. Fortunately, Turing devised a simpler method, based on the fact (first due to Littlewood for $\zeta(s)$) that $S(t)$ has mean value $0$. Thus, the graph of $N(t_0,t)-\Phi(t)$ for any fixed $t_0$ oscillates around a constant value; if we were to plot the same function using the [*measured*]{} number of zeros in $(t_0,t]$, then any zeros that we had missed would be obvious as jumps in the graph.
This can be made rigorous as follows. Let $t_0$ be a large number that is not the ordinate of a zero or pole, and assume that between ordinates $t_0-h$ and $t_0+h$ (for some $h>0$), we have located several zeros of $\Lambda(s)$, i.e. we have found small intervals $(a_n,b_n)$ such that $\Lambda\bigl(\frac12+ia_n\bigr)$ and $\Lambda\bigl(\frac12+ib_n\bigr)$ have opposite sign. Let ${N_{\rm left}}(t_0,t)$ (resp. ${N_{\rm right}}(t_0,t)$) be the step function which is upper semi-continuous, increases by $1$ at each $a_n$ (resp. $b_n$) and vanishes at $t=t_0$. We then have $$N(t)\le N(t_0)+{N_{\rm left}}(t_0,t)\mbox{ for }t\le t_0
\qquad\mbox{and}\qquad
N(t)\ge N(t_0)+{N_{\rm right}}(t_0,t)\mbox{ for }t\ge t_0.
\label{Ntbounds}$$ From these, we can deduce upper and lower bounds for $N(t_0)$; integrating , we get $$N(t_0)h+\int_{t_0}^{t_0+h}{N_{\rm right}}(t_0,t)\,dt
\le \int_{t_0}^{t_0+h}N(t)\,dt
=\int_{t_0}^{t_0+h}\Phi(t)\,dt
+\int_{t_0}^{t_0+h}S(t)\,dt
\label{Nupper}$$ and $$N(t_0)h+\int_{t_0-h}^{t_0}{N_{\rm left}}(t_0,t)\,dt
\ge \int_{t_0-h}^{t_0}N(t)\,dt
=\int_{t_0-h}^{t_0}\Phi(t)\,dt
+\int_{t_0-h}^{t_0}S(t)\,dt.
\label{Nlower}$$
If we have in fact located all zeros in the interval $(t_0-h,t_0+h)$ with some amount of precision (as measured by the size of the intervals $(a_n,b_n)$), then we can expect these bounds to be close to the truth. Moreover, if we have effective upper and lower bounds for the integral of $S(t)$, then for $h$ large enough, and will bound a single integer, i.e. we can unambiguously determine $N(t_0)$. Doing this for two different values $t_0=t_1, t_2$, we obtain $N(t_1,t_2)$.
One nice feature of Turing’s method is that precise knowledge of the zeros is only required in the short intervals around $t_1$ and $t_2$, and even there one can make a trade-off between the precision of the zeros and the length $h$ of the interval. For the bulk of the zeros between $t_1$ and $t_2$ it suffices to observe the sign changes.
The remainder of this section is devoted to bounding $\int S(t)\,dt$, cf. Theorem \[Stbound\] below. Our starting point is the following formula, obtained by Littlewood’s box principle (see [@titchmarsh §9.9]): $$\pi\int_{t_1}^{t_2}S(t)\,dt
=\int_{1/2}^{\infty}\log|L(\sigma+it_2)|\,d\sigma
-\int_{1/2}^{\infty}\log|L(\sigma+it_1)|\,d\sigma.
\label{Stformula}$$
\[Lupperbound\] Let notation be as above, and set $B:=\sup_{\Re(s)=3/2}|L(s)|^2$. Then, for $s$ in the strip $\bigl\{s\in{\mathbb{C}}:-\frac12\le \Re(s)\le \frac32\bigr\}$, $$|L(s)|^2\le B|\chi(s)Q(s)|\left|\frac{P(s+1)^2P(s-2)}{P(s)^2P(s-1)}\right|.$$
The power of $|Q(s)|$ in the above is not optimal; for $\Re(s)=\frac12$, the “convexity bound” says that we can put instead $|Q(s)|^{1/2+\varepsilon}$, with a constant depending on $\varepsilon$ (see [@iwaniec-sarnak]), while the Lindelöf hypothesis would have $|Q(s)|^{\varepsilon}$. Our present choice permits us to avoid Stirling’s formula in the proof, and thus obtain a clean bound that is uniform in all parameters.
We consider first the case when $L(s)$ is entire. Set $$F(s):=L(s){\overline{L}}(1-s) = \chi(s)^{-1}L(s)^2.$$ Plugging in the definition of $\chi(s)$, $$|F(\sigma+it)|=|L(\sigma+it)|^2
\left|\frac{\gamma(\sigma+it)}{{\overline{\gamma}}(1-\sigma-it)}\right|
=|L(\sigma+it)|^2
\left|\frac{\gamma(\sigma+it)}{\gamma(1-\sigma+it)}\right|.$$ Note that when $\sigma=\frac12+\mbox{a positive integer}$, the ratio of $\gamma$ factors reduces to a polynomial; in particular, $$|F(3/2+it)|=|L(3/2+it)|^2|Q(-1/2+it)| \le B|Q(3/2+it)|.
\label{F32bound}$$ The inequality holds since $\Re(\mu_j)\ge -\frac12$ for all $j$. Next, from the functional equation we have $F(s)={\overline{F}}(1-s)$, so that $$|F(\sigma+it)|=|F(1-\sigma+it)|.$$ Hence, by , $$|F(-1/2+it)|\le B|Q(-1/2+it)|.$$ Thus, the function $F(s)/Q(s)$ is bounded by $B$ on the lines $\Re(s)=-\frac12$ and $\Re(s)=\frac32$. Note that although $Q(s)$ has zeros, $F(s)$ has trivial zeros at the same points; in fact $$\frac{F(s)}{Q(s)}=\frac{\Lambda(s){\overline{L}}(1-s)}{\gamma(s)Q(s)}
=\frac{\Lambda(s){\overline{L}}(1-s)}{\gamma(s+2)}.
\label{FPholo}$$ Since $F$ has finite order, it follows from the Phragmen-Lindelöf theorem that $|F(s)|\le B|Q(s)|$ for all $s$ in the strip.
If $L(s)$ has poles then the above argument breaks down since $F(s)/Q(s)$ is not holomorphic in the strip. In fact, for each $k$ we get three poles, one at $1+\lambda_k$ and two at $\lambda_k$, as shows. To compensate for this, we consider $F(s)P(s)^2P(s-1)$ in the above, in place of $F(s)$. One checks that $|s^2(s-1)|\le |(s+1)^2(s-2)|$ on the lines $\Re(s)=-1/2$ and $\Re(s)=3/2$, so that $$|F(s)P(s)^2P(s-1)|\le B|Q(s)P(s)^2P(s-1)| \le B|Q(s)P(s+1)^2P(s-2)|.$$ Further, the ratio $\frac{F(s)P(s)^2P(s-1)}{Q(s)P(s+1)^2P(s-2)}$ is holomorphic in the strip, so we may proceed as above. The lemma follows.
\[technical\] Suppose that $$(t+\Im(\mu_j))^2\ge (5/2+\Re(\mu_j))^2+X^2
\quad\mbox{for some }X>5\mbox{ and all }j=1,\ldots,r.
\label{tmuhyp}$$ Then
1. For $\sigma\in [1/2,5/2]$, $$-r\left(\frac1{2\sqrt2X}+\frac{4/\pi^2+1/4}{X^2}\right) \le
\Re\frac{\gamma'}{\gamma}(\sigma+it)
-\frac12\log\left|Q\!\left(\frac32+it\right)\right|
\le \frac{4r}{\pi^2X^2}.
\label{gammabound}$$
2. For $\sigma\in [-1/2,3/2]$, $$\Re\frac{Q'}{Q}(\sigma+it)\le \frac{r}{\sqrt2 X}.$$
3. For all $\sigma$, $$\Re\frac{P'}{P}(\sigma+it)\le\frac{\max(\sigma m,0)}{X^2}.$$
1\. We have $$\frac{\gamma'}{\gamma}(\sigma+it)
=\frac12\log\frac{N}{\pi^r}+\frac12\sum_{j=1}^r
\frac{\Gamma'}{\Gamma}\!\left(\frac{\sigma+it+\mu_j}2\right).$$ We apply the Stirling-type estimate [@lehman] $$\frac{\Gamma'}{\Gamma}(z) = \log z-\frac1{2z}
+\Theta\!\left(\frac{2/\pi^2}{|\Im(z)^2-\Re(z)^2|}\right)
\qquad\mbox{for }\Re(z)\ge 0.
\label{Gammalogderiv}$$ This yields $$\begin{aligned}
&\Re\frac{\gamma'}{\gamma}(\sigma+it)
-\frac12\log\left|Q\!\left(\frac32+it\right)\right|\\
&=-\frac12\sum_{j=1}^r\left[
\log\left|\frac{3/2+it+\mu_j}{\sigma+it+\mu_j}\right|
+\Re\frac1{\sigma+it+\mu_j}
+\Theta\!\left(\frac{8/\pi^2}{|(t+\Im(\mu_j))^2-(\sigma+\Re(\mu_j))^2|}\right)
\right].
\end{aligned}$$
For the lower bound in we need an upper bound for the expression in brackets. By hypothesis, the $\Theta$ term is bounded by $\frac8{\pi^2 X^2}$. For the others, put $\sigma+it+\mu_j=x+iy$, $\beta=3/2-\sigma$, so that $|\beta|\le 1$ and $x$ and $y$ are constrained by $y^2\ge x^2+X^2$, $x\ge 0$. Then, using the inequality $\log(1+u)\le u$, we have $$\begin{aligned}
\log\left|\frac{x+\beta+iy}{x+iy}\right|+\Re\frac1{x+iy}
&=\frac12\log\!\left(1+\frac{2\beta x+\beta^2}{x^2+y^2}\right)+
\frac{x}{x^2+y^2}\\
&\le \frac{(\beta+1)x+\beta^2/2}{x^2+y^2}
\le\frac{2x+1/2}{2x^2+X^2} \le \frac1{\sqrt{2}X}+\frac1{2X^2}.
\end{aligned}$$ There are $r$ such terms, and the lower bound follows after multiplying by $-\frac12$.
The upper bound is similar, but uses the second order inequality $\log(1+u)\ge u-u^2/2$. We omit the details.
2\. Similarly, $$\Re\frac{Q'}{Q}(\sigma+it)
=\sum_{j=1}^r\Re\frac1{\sigma+\mu_j+it}
\le\sum_{j=1}^r\frac{\sigma+\Re(\mu_j)}
{2(\sigma+\Re(\mu_j))^2+X^2}\le \frac{r}{\sqrt2X}.$$
3\. $$\Re\frac{P'}{P}(\sigma+it)
=\sum_{k=1}^m\Re\frac1{\sigma+it-\lambda_k}
\le\sum_{k=1}^m\frac{\max(\sigma,0)}
{\sigma^2+X^2}\le \frac{\max(\sigma m,0)}{X^2}.$$ (Note that this bound is of faster decay than estimates 1 and 2. That is because we have control over the real parts of the poles, while nothing prevents $\Re(\mu_j)$ from being of comparable size to $X$. One could also obtain an $O\bigl(X^{-2}\bigr)$ bound in 1 and 2, with a constant depending on the $\mu_j$.)
\[hardlemma\] Let $w$ be a complex number with $|\Re(w)|\le \frac12$. Then $$\int_0^1\log\left|
\frac{(x+1+w)(x+1-\overline{w})}{(x+w)(x-\overline{w})}\right|dx
\le (\log 4)\Re\!\left(\frac1{1+w}+\frac1{1-\overline{w}}\right).
\label{hardineq}$$
Note first that equality is attained at $w=0$. Using the principal branch of the logarithm, set $$f(w)=\int_0^1\log\!\left(
\frac{(x+1+w)(x+1-w)}{(x+w)(x-w)}\right)dx
\quad\mbox{and}\quad
g(w)=\frac1{1+w}+\frac1{1-w}.$$ These define analytic functions on ${\mathbb{C}}\setminus{\mathbb{R}}$, with real parts extending continuously to ${\mathbb{R}}$. Further, is equivalent to the assertion $$\Re(f(w)-(\log 4)g(w)) \le 0 \quad\mbox{for }|\Re(w)|\le\frac12.
\label{equivassertion}$$
Note that $f(w)$ and $g(w)$ are each asymptotic to $-2/w^2$ as $|w|\to\infty$. Thus, holds for $\Im(w)$ sufficiently large; we check that in fact $|\Im(w)|\ge 2$ is enough. By symmetry and the maximum modulus principle applied to the function $e^{f(w)-(\log 4)g(w)}$ on the rectangle with corners at $\pm\frac12$ and $\pm\frac12+2i$, it suffices to check on the real axis and for $\Re(w)=\frac12$. On the real axis we calculate the integral explicitly and verify the inequality using calculus. For $\Re(w)=\frac12$ the inequality is strict, so we may verify it computationally for $0\le \Im(w) \le 2$.
\[zetalemma\] For $\sigma > \theta+1$, define $$z_{\theta}(\sigma):=
\left(\frac{\zeta(2\sigma+2\theta)\zeta(2\sigma-2\theta)}
{\zeta(\sigma+\theta)\zeta(\sigma-\theta)}\right)^{1/2}
\quad\mbox{and}\quad
Z_{\theta}(\sigma):=
\bigl(\zeta(\sigma+\theta)\zeta(\sigma-\theta)\bigr)^{1/2}.$$ Then $$r\frac{z_{\theta}'}{z_{\theta}}(\sigma)\ge
\Re\frac{L'}{L}(\sigma+it)\ge
r\frac{Z_{\theta}'}{Z_{\theta}}(\sigma)
\label{zetapart1}$$ and $$z_{\theta}(\sigma)^r\le |L(\sigma+it)|\le Z_{\theta}(\sigma)^r.
\label{zetapart2}$$
From we have $$r\frac{Z_{\theta}'}{Z_{\theta}}(\sigma)-\Re\frac{L'}{L}(\sigma+it)
=\sum_p\sum_{k=1}^{\infty}p^{-k\sigma}\log{p}
\left(\Re\sum_{j=1}^r\bigl(\alpha_{p,j}p^{-it}\bigr)^k
-\frac r2\bigl(p^{k\theta}+p^{-k\theta}\bigr)\right).$$ Pairing the terms for $\alpha_{p,j}$ and $\alpha_{p,j}'$, we see that each summand is $\le 0$, from which the second inequality of follows. The first inequality is similar. For , integrate from $\sigma$ to $\infty$.
\[Stbound\] Suppose $t_1$ and $t_2$ satisfy , and set $$c_{\theta}:=\log Z_{\theta}\!\left(\frac32\right)
+\int_{3/2}^{\infty}
\log\frac{Z_{\theta}(\sigma)}{z_{\theta}(\sigma)}\,d\sigma
-\int_{3/2}^{5/2}\log z_{\theta}(\sigma)\,d\sigma
+(\log 4)\frac{z_{\theta}'}{z_{\theta}}\!\left(\frac32\right),$$ where $z_{\theta}$ and $Z_{\theta}$ are as in Lemma \[zetalemma\]. (In particular, $c_0 \lessapprox 5.65055$.) Then $$\pi\int_{t_1}^{t_2}S(t)\,dt \le
\frac14\log\left|Q\!\left(\frac32+it_2\right)\right|
+\left(\log2-\frac12\right)
\log\left|Q\!\left(\frac32+it_1\right)\right|
+c_{\theta}r+\frac{r}{\sqrt2(X-5)}.$$
Note that there is no assumption on the order of $t_1$ and $t_2$, so one obtains a lower bound as well by reversing their roles.
By , we need upper and lower bounds for $\int_{1/2}^{\infty}\log|L(\sigma+it)|\,d\sigma$. For the upper bound, we use Lemma \[Lupperbound\]: $$\begin{aligned}
\int_{1/2}^{\infty}\log|L(\sigma+it)|&\,d\sigma
=\int_{1/2}^{3/2}\log|L(\sigma+it)|\,d\sigma
+\int_{3/2}^{\infty}\log|L(\sigma+it)|\,d\sigma\\
&\le\frac12\log B
+\frac12\int_{1/2}^{3/2}\log\left|
\frac{\gamma(1-\sigma+it)Q(\sigma+it)}{\gamma(\sigma+it)}
\right|d\sigma\\
&+\frac12\int_{1/2}^{3/2}\log\left|
\frac{P(\sigma+1+it)^2P(\sigma-2+it)}{P(\sigma+it)^2P(\sigma-1+it)}
\right|d\sigma
+\int_{3/2}^{\infty}\log|L(\sigma+it)|\,d\sigma.
\end{aligned}$$ We bound the first and last terms with Lemma \[zetalemma\]: $$\sup_{\Re(s)=3/2}|L(s)|
+\int_{3/2}^{\infty}\log|L(\sigma+it)|\,d\sigma
\le \log Z_{\theta}\!\left(\frac32\right)
+\int_{3/2}^{\infty}\log Z_{\theta}(\sigma)\,d\sigma.$$ For the second term, we replace $\sigma$ by $2-\sigma$ in the top $\gamma$ factor, and use the recurrence for $\gamma$ to get $$\frac12\int_{1/2}^{3/2}\log\left|
\frac{\gamma(\sigma+1+it)}{\gamma(\sigma+it)}
\frac{Q(\sigma+it)}{Q(\sigma-1+it)}\right|d\sigma.
\label{uppersecond}$$ By the mean value theorem, the integral equals $\Re\bigl[\frac{\gamma'}{\gamma}(\sigma^*+it)
+\frac{Q'}{Q}(\sigma^*-1+it)\bigr]$ for some $\sigma^* \in [1/2,5/2]$. Thus, by Lemma \[technical\], is at most $$\frac14\log\left|Q\!\left(\frac32+it\right)\right|
+r\left(\frac2{\pi^2X^2}+\frac1{2\sqrt 2X}\right).$$ Similarly, we see that the third term is bounded by $\frac{5m}{2X^2}$.
We now turn to the lower bound. This part is more delicate since we must take into account the contribution of zeros near $\frac12+it$. We use Turing’s idea of comparing $\log|L(s)|$ to $\log|L(s+1)|$; the difference between these looks like a value of the logarithmic derivative, which we can make precise with the help of Lemma \[hardlemma\].
Proceeding, we first clear the poles of $\Lambda$ by writing $F(s)=\Lambda(s)P(s)P(s-1)$. This function then has the Weierstrass-Hadamard product $$F(s) = e^{as+b}\prod_{\rho}\left(1-\frac{s}{\rho}\right)
e^{s/\rho},
\label{hadamard}$$ where $\rho$ runs over the zeros of $\Lambda$ and $\Re(a) = -\sum_{\rho}\Re\bigl(\frac1{\rho}\bigr)$. Next, we split the integral as follows: $$\begin{aligned}
\int_{1/2}^{\infty}&\log|L(\sigma+it)|\,d\sigma
=\int_{1/2}^{3/2}\log\left|\frac{F(\sigma+it)}{F(\sigma+1+it)}\right|d\sigma
+\int_{1/2}^{3/2}\log\left|\frac{\gamma(\sigma+1+it)}
{\gamma(\sigma+it)}\right|d\sigma\\
&+\int_{1/2}^{3/2}\log\left|\frac{P(\sigma+1+it)}
{P(\sigma-1+it)}\right|d\sigma
+ \int_{3/2}^{5/2}\log|L(\sigma+it)|\,d\sigma
+\int_{3/2}^{\infty}\log|L(\sigma+it)|\,d\sigma.
\end{aligned}$$ The second term may be estimated, as above, by Lemma \[technical\] and the mean value theorem: $$\int_{1/2}^{3/2}\log\left|\frac{\gamma(\sigma+1+it)}
{\gamma(\sigma+it)}\right|d\sigma
\ge \frac12\log\left|Q\!\left(\frac32+it\right)\right|
-r\left(\frac1{2\sqrt2X}+\frac{4/\pi^2+1/4}{X^2}\right).$$ The third term is positive since $\bigl|\frac{P(\sigma+1+it)}{P(\sigma-1+it)}\bigr|\ge 1$ for $\sigma\ge 0$. The fourth and fifth terms are handled by Lemma \[zetalemma\].
As for the first term, from we have $$\log\left|\frac{F(s)}{F(s+1)}\right|
=\sum_{\rho}\log\left|\frac{1-\frac{s}{\rho}}{1-\frac{s+1}{\rho}}\right|
=-\sum_{\rho}\log\left|\frac{s+1-\rho}{s-\rho}\right|.$$ Thus, $$\int_{1/2}^{3/2}\log\left|\frac{F(\sigma+it)}
{F(\sigma+it+1)}\right|d\sigma
=-\sum_{\rho}\int_{1/2}^{3/2}
\log\left|\frac{\sigma+it+1-\rho}{\sigma+it-\rho}\right|d\sigma.
\label{zerosum}$$ Now, by the functional equation, the zeros of $\Lambda$ either lie on the line $\Re(s)=\frac12$ or come in pairs $\rho, 1-\overline{\rho}$. Applying Lemma \[hardlemma\] with $w=\frac12+it-\rho$, we see that is bounded below by $$-(\log 4)\sum_{\rho}\Re\frac1{3/2+it-\rho}.$$ Again by , this equals $$\begin{aligned}
-(\log 4)\Re\frac{F'}{F}\!&\left(\frac32+it\right)\\
&=-(\log 4)\Re\left[
\frac{\gamma'}{\gamma}\!\left(\frac32+it\right)
+\frac{L'}{L}\!\left(\frac32+it\right)
+\frac{P'}{P}\!\left(\frac32+it\right)
+\frac{P'}{P}\!\left(\frac12+it\right)\right]\\
&\ge -(\log 4)\left[
\frac12\log\left|Q\!\left(\frac32+it\right)\right|
+\frac{4r}{\pi^2X^2}
+r\frac{z_{\theta}'}{z_{\theta}}\!\left(\frac32\right)
+\frac{2m}{X^2}\right].
\end{aligned}$$
Altogether, we get $$\begin{aligned}
\int_{1/2}^{\infty}\log|L(\sigma+it)|\,d\sigma
&\ge \left(\frac12-\log2\right)
\log\left|Q\!\left(\frac32+it\right)\right|\\
&+ r\left[\int_{3/2}^{5/2}\log z_{\theta}(\sigma)\,d\sigma
+\int_{3/2}^{\infty}\log z_{\theta}(\sigma)\,d\sigma
-(\log 4)\frac{z_{\theta}'}{z_{\theta}}\!\left(\frac32\right)\right]\\
&- r\left(\frac1{2\sqrt2X}+\frac{\frac4{\pi^2}\log(4e)+\frac14}{X^2}\right)
-\frac{4m\log{2}}{X^2}.
\end{aligned}$$
Finally, we combine the upper bound for $t=t_2$ and lower bound for $t=t_1$. We get the stated main term plus error $$r\left(\frac1{\sqrt2X}+\frac{\frac2{\pi^2}\log(16e^3)+\frac14}{X^2}\right)
+m\frac{4(\log2)+5/2}{X^2} < \frac{r}{\sqrt2(X-5)}.$$
Rigorous Computation of $L$-functions {#sec:rigorous}
=====================================
The methods of Section \[turing\] depend on a fast, rigorous algorithm for evaluating $\Lambda(s)$. We describe one such algorithm, based on the Fast Fourier Transform, in this section. We note that in the case of the Riemann zeta function, a similar technique was developed and used by Odlyzko and Schönhage [@odlyzko-schonhage].
Some algorithms for computing general $L$-functions were described by Dokchitser [@dokchitser] and Rubinstein [@rubinstein]. They ultimately boil down to the Cauchy integral formula: $$\Lambda(s_0) = \frac1{2\pi i}\int\frac{\gamma(s)L(s)}{s-s_0}\,ds,$$ where the contour consists of two vertical lines enclosing $s_0$. Writing $L(s)$ as a Dirichlet series and using the functional equation, one is lead to study integrals of the form $$\frac1{2\pi i}\int\frac{\gamma(s)n^{-s}}{s-s_0}\,ds,
\label{twoparam}$$ taken along a vertical line far to the right. Rubinstein, following an idea of Lagarias and Odlyzko [@lagarias-odlyzko], inserts a factor designed to cancel the decay of the $\gamma$ factor, e.g. $e^{i\frac{\pi r}4\eta s}$ for some $\eta$ close to $\pm 1$. Without this factor, very high precision is required to calculate $\Lambda(s)$ when $\Im(s)$ is large.
These algorithms are good when one is interested in computing $\Lambda(s)$ at specific points, e.g. for locating zeros of $L(s)$ precisely. They suffer from the disadvantage of being difficult to carry out rigorously, basically because is a two parameter family (indexed by $s_0$ and $n$) of integrals, for which uniform asymptotics are hard to obtain in certain transition ranges.
For Turing’s method, we need an algorithm for [*rigorously*]{} computing $\Lambda(s)$ for many values of $s$, not necessarily with high precision. For that we consider instead the one parameter integrals $$\frac1{2\pi i}\int\Lambda(s)e^{-zs}\,ds
\quad\mbox{and}\quad
\frac1{2\pi i}\int\gamma(s)e^{-zs}\,ds.$$ These are essentially Fourier transforms, and they contain enough information for evaluating $\Lambda(s)$ quickly, if one is interested in many points. They also involve only a single Mellin transform, making rigorous computation more accessible.
Precisely, let $\eta\in(-1,1)$ and set $F(t) := \Lambda\!\left(\frac12+it\right)
e^{\frac{\pi r}4\eta t}$. Then the (inverse) Fourier transform of $F$ is $$\begin{aligned}
\hat{F}(x)&:=\frac1{2\pi}\int_{-\infty}^{\infty}
F(t)e^{-ixt}\,dt
=\frac1{2\pi i}\int_{\Re(s)=\frac12}
\Lambda(s)e^{(x+i\frac{\pi r}4\eta)(1/2-s)}\,ds\\
&=\frac1{2\pi i}\int_{\Re(s)=2}
\Lambda(s)e^{(x+i\frac{\pi r}4\eta)(1/2-s)}\,ds
-\sum_{\rho\in\{1+\lambda_k:1\le k\le m\}}
{{\rm Res}}_{s=\rho}\Lambda(s)e^{(x+i\frac{\pi r}4\eta)(1/2-s)}.
\label{residuesum}
\end{aligned}$$ The residue sum is straightforward to evaluate assuming we have complete information on any poles of $L(s)$. We multiply the Euler product out to a Dirichlet series, writing $L(s)=\sum_{n=1}^{\infty}a_nn^{-s}$. Then the first term of is $$\sum_{n=1}^{\infty}a_n\frac1{2\pi i}\int_{\Re(s)=2}
\gamma(s)e^{(x+i\frac{\pi r}4\eta)(1/2-s)}n^{-s}\,ds
=\epsilon\sum_{n=1}^{\infty}\frac{a_n}{\sqrt n}
G\!\left(x+\log\frac{n}{\sqrt N};\eta,\{\mu_j\}\right),
\label{Gseries}$$ where $$G(u;\eta,\{\mu_j\}) := \frac1{2\pi i}\int_{\Re(s)=2}
e^{(u+i\frac{\pi r}4\eta)(1/2-s)}
\prod_{j=1}^r\Gamma_{{\mathbb{R}}}(s+\mu_j)
\,ds.
\label{Gdef}$$
Let us assume for now that we have a procedure to compute $G(u;\eta,\{\mu_j\})$, and thereby $\hat{F}(x)$, to prescribed precision; we return to this point in Section \[sec:Gcompute\] below. In order to use the FFT to compute $F$ from $\hat{F}$, we first need to discretize the problem. To that end, let $A,B>0$ be parameters such that $q=AB$ is an integer. By the Poisson summation formula, $$\sum_{k\in{\mathbb{Z}}}F\!\left(\frac{m}A+kB\right)
=\frac{2\pi}{B}\sum_{k\in{\mathbb{Z}}}
\hat{F}\!\left(\frac{2\pi k}B\right)\!e\!\left(\frac{km}{AB}\right)
=\frac{2\pi}{B}
\sum_{n\,(\mbox{\scriptsize mod }q)}
e\!\left(\frac{mn}q\right)\!
\sum_{k\in{\mathbb{Z}}}\hat{F}\!\left(\frac{2\pi n}B+2\pi Ak\right)\!.$$ Thus, the functions $\widetilde{F}(m):=\sum_{k\in{\mathbb{Z}}}F\!\left(\frac{m}A+kB\right)$ and $\widetilde{\hat{F}}(n):=
\sum_{k\in{\mathbb{Z}}}\hat{F}\!\left(\frac{2\pi n}B+2\pi Ak\right)$, which are periodic in $m,n$ with period $q$, form a discrete Fourier transform pair.
Note that since $F$ is real-valued, $\hat{F}(-x) = \overline{\hat{F}(x)}$. Thus, for $|n|\le q/2$ we have $$\widetilde{\hat{F}}(n)=\hat{F}\!\left(\frac{2\pi n}B\right)
+\sum_{k=1}^{\infty}
\hat{F}\!\left(\frac{2\pi n}B+2\pi kA\right)
+\sum_{k=1}^{\infty}
\overline{\hat{F}\!\left(-\frac{2\pi n}B+2\pi kA\right)}.
\label{hatFsmall}$$ For $A$ even moderately large, the terms for $k\ge 1$ fall within the asymptotic range. Precise bounds are given in Section \[sec:asymp\] below; in particular, we may apply Lemma \[lem:Fasymp\] with $x=2\pi(A\pm\frac{n}{B})$ to compute the sums over $k$. Hence, it suffices to calculate $\hat{F}(2\pi n/B)$ for $0\le n\le q/2$. On the other hand, to compute $F(m/A)$, we need to bound the terms of $\widetilde{F}(m)$ for $k\ne 0$. We have already obtained a suitable bound for the $L$-function in Lemma \[Lupperbound\]. The sum of this bound over $k\ne 0$ is the content of Lemma \[lem:Lambdasum\].
Computing $G(u;\eta,\{\mu_j\})$ {#sec:Gcompute}
-------------------------------
For brevity, some of the results of this section are only sketched. Our emphasis is on the details necessary for rigorous computation. For more general background information we refer the reader to [@booker2; @dokchitser; @rubinstein].
One simple method for calculating integrals such as that is easy to make rigorous is the power and log series, obtained by shifting the contour of to the left: $$\begin{aligned}
G(u;\eta,\{\mu_j\}) &= \sum_{\rho\in{\mathbb{C}}}
{{\rm Res}}_{s=\rho}\left(
e^{(u+i\frac{\pi r}4\eta)(1/2-s)}
\prod_{j=1}^r\Gamma_{{\mathbb{R}}}(s+\mu_j)
\right)\\
&= \sum_{\mathrm{poles}\,\rho}
P(u;\rho,\eta,\{\mu_j\})e^{(1/2-\rho)u},
\end{aligned}
\label{powerlog}$$ where $P(u;\rho,\eta,\{\mu_j\})$ is a polynomial of degree one less than the order of the pole at $\rho$.
For example, in the case of Galois representations, the $\mu_j$ are all either $0$ or $1$, and the residues in may be evaluated by the following: $$\begin{aligned}
&\begin{aligned}
\Gamma_{{\mathbb{R}}}(s)=
\frac2{s+2k}\frac{(-2\pi)^k}{(2k)!!}
\exp&\!\left[
\!\left(\sum_{n=1}^k\frac1{2n}-
\frac12\log(\pi e^{\gamma})\right)\!(s+2k)\right.\\
&+\left.\sum_{j=2}^{\infty}\frac{1}j
\!\left((-1)^j2^{-j}\zeta(j)+\sum_{n=1}^k\frac1{(2n)^j}\right)\!
(s+2k)^j\right]
\end{aligned}\\
&\hspace{11mm}\begin{aligned}
=\frac{(-2\pi)^k}{(2k-1)!!}
\exp&\!\left[
\!\left(\sum_{n=1}^k\frac1{2n-1}-
\frac12\log(4\pi e^{\gamma})\right)\!(s+2k-1)\right.\\
&+\left.\sum_{j=2}^{\infty}\frac{1}j
\!\left((-1)^j(1-2^{-j})\zeta(j)+\sum_{n=1}^k\frac1{(2n-1)^j}\right)\!
(s+2k-1)^j\right],
\end{aligned}
\end{aligned}$$ for any integer $k\ge 0$. For general $\mu_j$, we need an algorithm to calculate the values of $\Gamma$ and its derivatives at an arbitrary point in the complex plane; we assume without further comment that this is available when necessary.
We say that $\mu_j$ and $\mu_k$ are equivalent if $\mu_j-\mu_k\in 2{\mathbb{Z}}$. For $\mu$ ranging over an equivalence class, the functions $\Gamma_{{\mathbb{R}}}(s+\mu)$ share all but finitely many poles. Thus, may be broken naturally into parts corresponding to each class. We can bound the tail of each part as follows.
Let $\rho$ be a pole of $g(s)=e^{(u+i\frac{\pi r}4\eta)(1/2-s)}
\prod_{j=1}^r\Gamma_{{\mathbb{R}}}(s+\mu_j)$ of order $n$, with $\Re(\rho+\mu_j)\le 0$ for $j=1,\ldots,r$ and $(2\pi)^re^{2u}<\frac12\prod_{j=1}^r(|2-\rho-\mu_j|-1)$. Let $c_j$ be the coefficients of the polar part of $g$ around $\rho$, i.e. such that $g(s+\rho)-\sum_{j=1}^nc_js^{-j}$ is holomorphic at $s=0$. Then $$\left|\sum_{k=1}^{\infty}{{\rm Res}}_{s=\rho-2k}g(s)\right|
< \max|c_j|.$$
First note that $$\begin{aligned}
g(s+\rho-2)&=(2\pi)^re^{2u+i\frac{\pi r}2\eta}g(s+\rho)
\prod_{j=1}^r(s+\rho-2+\mu_j)^{-1}\\
&=\frac{(-2\pi)^re^{2u+i\frac{\pi r}2\eta}}
{\prod_{j=1}^r(2-\rho-\mu_j)}
\cdot\frac{g(s+\rho)}
{\prod_{j=1}^r\bigl(1-\frac{s}{2-\rho-\mu_j}\bigr)}.
\end{aligned}
\label{eq:gminus2}$$
Next, let $f(s)$ be a meromorphic function with polar part $a_1s^{-1}+\ldots+a_ns^{-n}$ at $0$. If $x$ is a complex number with $|x|<1$ then the function $\frac{f(s)}{1-xs}$ has polar part $a_1's^{-1}+\ldots+a_n's^{-n}$, where $a_j'=\sum_{k=0}^{n-j}a_{j+k}x^k$. Thus, $$\max|a_j'| \le \frac{\max|a_j|}{1-|x|}.
\label{eq:xsmax}$$
Let $c_1's^{-1}+\ldots+c_n's^{-n}$ be the polar part of $g(s+\rho-2)$. Applying $r$ times, we see from that $$\begin{aligned}
\max|c_j'|& \le
\frac{(2\pi)^re^{2u}}{\prod_{j=1}^r|2-\rho-\mu_j|}
\cdot\frac{\max|c_j|}{\prod_{j=1}^r\bigl(1-\frac1{|2-\rho-\mu_j|}\bigr)}\\
&=\frac{(2\pi)^re^{2u}\max|c_j|}{\prod_{j=1}^r(|2-\rho-\mu_j|-1)}
< \frac12\max|c_j|.
\end{aligned}$$ Repeating this procedure, we see that the coefficients of the polar part of $g(s+\rho-2k)$ are $< 2^{-k}\max|c_j|$. The conclusion follows.
The lemma says roughly that if we compute the residue sum for all poles with real part down to $\Re(\rho)$, the tail of the series (from poles at $\rho-2k$) may be bounded by the data from the last term added. Moreover, gives an algorithm for computing the data at $\rho-2$ from that at $\rho$, and shows that the terms eventually decrease factorially. Thus, we may use this to compute $G(u;\eta,\{\mu_j\})$ to any desired precision for a given $u$.
Since the $\mu_j$ are arbitrary, this procedure is general enough to compute derivatives of $G(u;\eta,\{\mu_j\})$ as well. For instance, for any $k$ we have $$G'(u;\eta,\{\mu_j\})
=(\mu_k+1/2)G(u;\eta,\{\mu_j\})-2\pi G(u;\eta,\{\mu_j'\}),$$ where $\mu_j' = \mu_j$ if $j\ne k$ and $\mu_k' = \mu_k+2$. Higher derivatives may be computed in a similar fashion. (In fact $G$ satisfies an $r$th order differential equation, due to the recurrence for $\Gamma$; thus, the derivatives of all orders are determined from the first $r$.)
Note that for $u$ large this method requires high precision due to cancellation, and is therefore inefficient. The essential point that makes it worthwhile is that for a given $\gamma$ factor the calculations need only be performed once, as one can develop local approximations to $G(u;\eta,\{\mu_j\})$ for later rapid evaluation. The computation may then be recycled and used for any $L$-function with the same $\mu_j$; this is useful for functions in an arithmetic family, such as Artin $L$-functions.
More precisely, suppose we wish to calculate $G(u;\eta,\{\mu_j\})$ for $u$ in an interval $I$. Choose $\varepsilon>0$ and sample points $u_m$ such that each $u\in I$ is contained in a unique interval $[u_m-\varepsilon,u_m+\varepsilon)$. For $u$ in the $m$th interval, we have by Taylor’s theorem $$G(u;\eta,\{\mu_j\}) = \sum_{k=0}^{K-1}\frac{G^{(k)}(u_m;\eta,\{\mu_j\})}{k!}
(u-u_m)^k +
\Theta\!\left(\max_{|u^*-u_m|\le\varepsilon}
\frac{\bigl|G^{(K)}(u^*;\eta,\{\mu_j\})\bigr|}{K!}\varepsilon^K\right).
\label{localapprox}$$ We may evaluate the derivatives precisely using . As for the $K$th derivative, a uniform bound is obtained by shifting the contour of to $\Re(s)=\frac12$: $$\frac{\bigl|G^{(K)}(u;\eta,\{\mu_j\})\bigr|}{K!} \le
\frac1{2\pi}\int_{-\infty}^{\infty}\frac{|t|^K}{K!}
e^{\frac{\pi r}4\eta t}
\prod_{j=1}^r\left|\Gamma_{{\mathbb{R}}}\!\left(\frac12+it+\mu_j\right)\right|
\,dt.$$ For large $K$ this is of size $\left(\frac4{\pi r(1-\eta)}\right)^K$; thus as long as $\varepsilon$ is small compared to $\frac{\pi
r}4(1-\eta)$, we may compute and store the coefficients of , yielding a fast method to calculate $G(u;\eta,\{\mu_j\})$ for any $u\in I$.
Moreover, we can improve the efficiency of our algorithm if the sample points of coincide with multiplies of $2\varepsilon$, i.e. if $\frac{\pi}{\varepsilon B}\in{\mathbb{Z}}$. For any given sample point $x$, we approximate ${\hat F}(x)$ via a truncated series (keeping track of the error terms from and Lemma \[lem:Ftail\] below): $$\begin{aligned}
&\hspace{-1cm}\sum_{n=1}^M\frac{a_n}{\sqrt n}
G\!\left(x+\log\frac{n}{\sqrt N};\eta,\{\mu_j\}\right)\\
&\approx
\sum_m\sum_{\log\frac{n}{\sqrt N}\in [u_m-\varepsilon,u_m+\varepsilon)}
\frac{a_n}{\sqrt n}\sum_{k=0}^{K-1}
\frac{G^{(k)}(x+u_m;\eta,\{\mu_j\})}{k!}
\left(\log\frac{n}{\sqrt N}-u_m\right)^k\\
&=
\sum_{k=0}^{K-1}\sum_m
\frac{G^{(k)}(x+u_m;\eta,\{\mu_j\})}{k!}
S^{(k)}_m,
\end{aligned}
\label{eq:Fxlocal}$$ where $$S^{(k)}_m :=
\sum_{\log\frac{n}{\sqrt N}\in [u_m-\varepsilon,u_m+\varepsilon)}
\frac{a_n}{\sqrt n}
\left(\log\frac{n}{\sqrt N}-u_m\right)^k.$$ Since $x+u_m$ is another sample point $u_{m'}$, the $k$th term of is a convolution of the sequences (indexed by $m$) $\frac{G^{(k)}(u_m;\eta,\{\mu_j\})}{k!}$ and $S^{(k)}_m$; thus, we may evaluate it efficiently for all $x$ simultaneously by appealing again to the FFT.
Complexity
----------
We may now consider the complexity of the algorithm. Note that by Stirling’s formula, $F(t)$ decays roughly like $e^{-(1-\eta)\frac{\pi
r}4t}$ for $t>0$. Ideally we should choose $1-\eta$ of size $T^{-1}$ in order to compute values up to height $T$. Adjusting the constant of proportionality (i.e. choosing $\eta$ relatively close to or far from $1$) allows us to trade off the computational precision and number of coefficients needed to overcome the error terms below. Finding a good compromise between these two is best done by trial and error; cf. Section \[numerics\].
Since $\delta\asymp T^{-1}$, Lemma \[lem:Ftail\] shows that in order to compute we need on the order of $\sqrt{NT^r}$ terms of , or roughly the square root of the analytic conductor. Note that the values of $A$ and $B$ enter only in the Fourier transforms, and do not significantly affect the computation of . We set $B$ equal to a multiple of $T$, depending on the chosen value of $\eta$. As for $A$, as mentioned in Section \[sec:explicit\], the density of zeros of $F(t)$ around height $T$ is $\frac1{2\pi}\log N\!\left(\frac{T}{2\pi}\right)^r$; one can expect to take $A$ equal to a multiple of this. Thus, this method has complexity consistent with computing a [*single*]{} value by the approximate functional equation, after which we get many values in mean time $O_{\varepsilon}\bigl((NT)^{\varepsilon}\bigr)$, which is essentially best possible. The gain comes from the fact, as emphasized above, that only a single $G$-function is involved.
Asymptotics {#sec:asymp}
-----------
To complete our understanding of $G(u;\eta,\{\mu_j\})$, in order to accurately calculate , we need an asymptotic bound for large $u$. If we write $\mu=-\frac12+\frac1r\bigl(
1+\sum_{j=1}^r\mu_j\bigr)$ then by the method of stationary phase, we have $$G(u;\eta,\{\mu_j\})=
\sqrt{\frac{2^{r+1}}r}e^{\mu(u+i\frac{\pi r}4\eta)}
\exp\!\left(-e^{\frac2r(u+i\frac{\pi r}4\eta)}\right)\!
\bigl(1+O\bigl(e^{-2u/r}\bigr)\bigr),
\label{Gasymp}$$ where the implied constant depends on the $\mu_j$. For $r=1$ the formula is exact, i.e. the $O$ term is $0$. For $r>1$, one can work out explicit constants case by case, which is preferable if sharp error terms are desired. Otherwise, we get a bound that is close to simply by shifting the contour of to the right.
\[lem:Gasymp\] Let $\delta=\frac{\pi}2(1-|\eta|)$, $\nu_j=\frac{\Re(\mu_j)-1}2+\frac1{2r}$, $\mu=-\frac12+\frac1r\bigl(1+\sum_{j=1}^r\mu_j\bigr)$, $K=2\sqrt{\frac{2^{r+1}}r\frac{e^{\delta(r-1)}}{\delta}}
e^{-\frac{\pi r\eta\Im(\mu)}4}$, and $X=\pi r\delta e^{-\delta}e^{2u/r}$. Then for $X\ge r$, $$\bigl|G(u;\eta,\{\mu_j\})\bigr| \le
Ke^{\Re(\mu)u}e^{-X}
\prod_{j=1}^r\left(1+\frac{r\nu_j}X\right)^{\nu_j}.
\label{Gbound}$$
This is within a factor $O\bigl(\delta^{-1/2}\bigr)$ of the correct asymptotic if $\delta\ll re^{-u/r}$.
We write $s=2\sigma+2it$ in to get $$|G(u;\eta,\{\mu_j\})|\le
\pi^{-r(\sigma+\frac{\Re(\mu)}2+\frac14)-\frac12}
e^{u(\frac12-2\sigma)-\frac{\pi r\eta\Im(\mu)}4}
\int_{-\infty}^{\infty}\prod_{j=1}^r
\left|\Gamma\!\left(\sigma+it+\frac{\mu_j}2\right)
e^{\frac{\pi \eta}4(2t+\Im(\mu_j))}
\right|dt.
\label{Gbound1}$$
Applying Hölder’s inequality, we get integrals of the form $$\int_{-\infty}^{\infty}|\Gamma(a+it)|^re^{\frac{\pi r\eta}2 t}\,dt,$$ where $a=\sigma+\Re(\mu_j)/2$. Assuming $\sigma\ge 1$, we may apply the inequality $$|\Gamma(a+it)|\le \sqrt{2\pi}(a+|t|)^{a-1/2}e^{-\pi|t|/2}
\quad\mbox{for }a\ge\frac12.$$ (To see this, note that $\bigl|\Gamma\bigl(\frac12+it\bigr)\bigr|=\sqrt{\pi {\rm sech}\,\pi t}$, use the recurrence for $\Gamma$ and to reduce to the region $\frac12\le a\le \frac32$, $0\le t\le 2$, where the inequality may be checked computationally.) Thus, we have $$\begin{aligned}
\int_{-\infty}^{\infty}|\Gamma(a+it)|^re^{\frac{\pi r\eta}2 t}\,dt
&\le 2(2\pi)^{r/2}\int_0^{\infty}(a+t)^{r(a-1/2)}e^{-r\delta t}\,dt\\
&\le 2(2\pi)^{r/2}e^{\delta ra}
\frac{\Gamma(r(a-1/2)+1)}{(\delta r)^{r(a-1/2)+1}} \\
&\le 2(2\pi)^{\frac{r+1}2}\sqrt{\frac{e^{\delta(r-1)}}{\delta r}}
\left(\frac{e^{\delta}}{\delta}\frac{a-1/2+1/2r}e\right)^{r(a-1/2)+1/2}.
\end{aligned}$$ Substituting this bound into and collecting terms we obtain $$Ke^{\Re(\mu)u}\prod_{j=1}^r
\left(\frac{\sigma+\nu_j}{eX/r}\right)^{\sigma+\nu_j}
\le
Ke^{\Re(\mu)u}e^{-\sigma r}\prod_{j=1}^r
\left(\frac{r\sigma}{X}\right)^{\sigma+\nu_j}
\left(1+\frac{\nu_j}{\sigma}\right)^{\nu_j}.$$ The result follows upon taking $\sigma=X/r$.
With this bound in hand, we can estimate the error in truncating the series .
\[lem:Ftail\] Let $M$ be a positive integer, $x\in{\mathbb{R}}$. Let $\delta,\nu_j,\mu,K$ be as in Lemma \[lem:Gasymp\] and set $X=\pi r\delta e^{-\delta}\bigl(e^x/\sqrt N\bigr)^{2/r}$. Let $C,\alpha\ge0$ be such that $|a_n|\le Cn^{\alpha}$ for all $n$, and put $c=\Re(\mu)+\frac12+\alpha$, $c'=\max(cr/2-1,0)$. Then for $XM^{2/r}>\max(c',r)$, $$\begin{aligned}
&\hspace{-1cm}\left|\sum_{n>M}\frac{a_n}{\sqrt n}
G\!\left(x+\log\frac{n}{\sqrt N};\eta,\{\mu_j\}\right)\right|\\
&\le \frac{Kr}2\left(\frac{e^x}{\sqrt N}\right)^{\Re(\mu)}
\frac{CM^ce^{-XM^{2/r}}}{XM^{2/r}-c'}
\prod_{j=1}^r
\left(1+\frac{r\nu_j}{XM^{2/r}}\right)^{\nu_j}.
\end{aligned}$$
Different values of $C$ and $\alpha$ are appropriate for different ranges. For small $M$, one can take $C=1,\alpha=\log_2r+\theta$, while for larger $M$ it is better to choose a smaller value of $\alpha$ and compute $C$ from the coefficients.
Using Lemma \[lem:Gasymp\], we have $$\sum_{n>M}\frac{|a_n|}{\sqrt n}
\left|G\!\left(x+\log\frac{n}{\sqrt N};\eta,\{\mu_j\}\right)\right|
\le K'\sum_{n>M}n^{c-1}e^{-Xn^{2/r}},
\label{eq:Ftail1}$$ where $$K'=CK\left(\frac{e^x}{\sqrt N}\right)^{\Re(\mu)}
\prod_{j=1}^r\left(1+\frac{r\nu_j}{XM^{2/r}}\right)^{\nu_j}.$$ The condition on $X$ ensures that the terms of are monotonically decreasing. Thus, we can estimate by the integral $$K'\int_M^{\infty}t^{c-1}e^{-Xt^{2/r}}\,dt
=\frac{K'r}2\bigl(X^{-r/2}\bigr)^c\int_{XM^{2/r}}^{\infty}
y^{cr/2-1}e^{-y}\,dy
\le\frac{K'r}2\frac{M^ce^{-XM^{2/r}}}{XM^{2/r}-c'}.$$
The next two lemmas bound the error introduced in discretization.
\[lem:Fasymp\] Let $x\in{\mathbb{R}}$, $A\ge \frac1{2\pi}$, and let notation be as in Lemma \[lem:Ftail\]. Then for $X>\max(c',r)$, $$\begin{aligned}
\sum_{k=0}^{\infty}\hat{F}(x+2\pi kA)&=
-\sum_{\rho\in\{1+\lambda_k:1\le k\le m\}}
{{\rm Res}}_{s=\rho}\frac{\Lambda(s)e^{(x+i\frac{\pi r}4\eta)(1/2-s)}}
{1-e^{2\pi A(1/2-s)}}\\
&+\Theta\!\left[
\frac{K}{1-e^{-\pi A}}
\!\left(\frac{e^x}{\sqrt N}\right)^{\Re(\mu)}e^{-X}
\!\left(1+\frac{Cr/2}{X-c'}\right)
\prod_{j=1}^r\left(1+\frac{r\nu_j}X\right)^{\nu_j}\right]\!.
\end{aligned}$$
The residue sum comes from summing the polar part of with $x+kA$ in place of $x$. For the rest, we apply Lemmas \[lem:Gasymp\] and \[lem:Ftail\] (with $M=1$) to get the bound $$K\!\left(\frac{e^x}{\sqrt N}\right)^{\Re(\mu)}\!e^{-X}
\!\left(1+\frac{Cr/2}{X-c'}\right)
\prod_{j=1}^r
\!\left(1+\frac{r\nu_j}X\right)^{\nu_j}$$ for the $k=0$ term. To pass from this to the $k$th term, we multiply by a factor not exceeding $$\begin{aligned}
e^{2\pi kA\Re(\mu)}e^{-X(\exp(4\pi kA/r)-1)}
&=\exp\!\left(2\pi kA\left[\Re(\mu)-\frac{2X}r-
\frac{X}{2\pi kA}\left(e^{4\pi kA/r}-1-\frac{4\pi kA}r\right)\right]\right)\\
&\le \exp\!\left(-2\pi kA\left[\alpha+\frac12-\frac2r
+\frac{4\pi kA}r\right]\right)
\le e^{-\pi kA}.
\end{aligned}$$ The result follows on summing the geometric series.
\[lem:Lambdasum\] Let $t\in{\mathbb{R}}$ and put $s=\frac12+it$, $$E=Z_{\theta}(3/2)^r
|\gamma(s)|e^{\frac{\pi r}4\eta t}
\left|Q(s)\frac{P(s+1)^2P(s-2)}{P(s)^2P(s-1)}\right|^{1/2},$$ and $$\beta = \frac{\pi r}4
-\frac12\sum_{j=1}^r\arctan\frac{\Re(s+\mu_j)}{|\Im(s+\mu_j)|}
-\frac4{\pi^2}\sum_{j=1}^r
\frac1{|\Im(s+\mu_j)^2-\Re(s+\mu_j)^2|}.$$
1. If $\Im(s+\mu_j) > 0$ for all $j=1,\ldots,r$ and $\beta-\frac{\pi r}4\eta > 0$ then $$\left|\sum_{k=0}^{\infty}F\bigl(t+kB\bigr)\right|
\le \frac{E}{1-e^{-(\beta-\frac{\pi r}4\eta)B}}.$$
2. If $\Im(s+\mu_j) < 0$ for all $j=1,\ldots,r$ and $\beta+\frac{\pi r}4\eta > 0$ then $$\left|\sum_{k=0}^{\infty}F\bigl(t-kB\bigr)\right|
\le \frac{E}{1-e^{-(\beta+\frac{\pi r}4\eta)B}}.$$
We treat only the first case, the second being similar. Lemmas \[Lupperbound\] and \[zetalemma\] imply the bound $|F(t)|\le E$. We consider the same bound with $t$ replaced by $t+kB$. Note that if $|\Im(s+\mu_j)|$ increases for all $j$ then the factor involving $P$ is non-increasing. For the $\gamma$ and $Q$ factors, by the mean value theorem we have $$\log\!\left(\left|\frac{\gamma(s+ikB)}{\gamma(s)}\right|
\left|\frac{Q(s+ikB)}{Q(s)}\right|^{1/2}\right)
=-kB\Im\!\left(
\frac{\gamma'}{\gamma}(s^*)+\frac12\frac{Q'}{Q}(s^*)\right)$$ for some $s^*$ on the line between $s$ and $s+ikB$. Using , this is $$\begin{aligned}
-kB&\Im
\sum_{j=1}^r\left(
\frac12\log\frac{s^*+\mu_j}2
+\Theta\!\left(\frac{4/\pi^2}{|\Im(s^*+\mu_j)^2-\Re(s^*+\mu_j)^2|}
\right)\right)\\
&\le
-kB\!\left(
\frac{\pi r}4
-\frac12\sum_{j=1}^r\arctan\frac{\Re(s^*+\mu_j)}{\Im(s^*+\mu_j)}
-\frac4{\pi^2}\sum_{j=1}^r
\frac1{|\Im(s^*+\mu_j)^2-\Re(s^*+\mu_j)^2|}\right)\\
&\le -\beta kB.
\end{aligned}$$ Thus, $|F(t+kB)|\le Ee^{-(\beta-\frac{\pi r}4\eta)kB}$. The conclusion follows.
Numerical results {#numerics}
=================
We have applied the methods described in the previous sections to a few examples of splitting fields of polynomials with Galois group $S_5$ and $A_5$, as listed in Table \[tab:examples\]. For the $A_5$ cases, the Artin conjecture is true for all representations by known cases of functoriality [@kiming; @jehanne2]. That speeds up the process, since we may apply Turing’s method to the Artin $L$-functions directly. For the $S_5$ examples we verify both conjectures. As expected, we found no counterexamples to either conjecture in the tested range $|t|\le 100$.
polynomial group splitting field discriminant
--------------------------- ------- ------------------------------
$x^5-68x-68$ $S_5$ $2^{96}3^{60}17^{96}$
$x^5-x^4-8x^3+10x^2-x-5$ $S_5$ $2^{160}3^{96}7^{96}$
$x^5-x^4+3x^3-11x^2-8x-8$ $S_5$ $2^{220}13^{96}$
$x^5+2x^3-4x^2-2x+4$ $A_5$ $2^{90}73^{30}$
$x^5+20x+16$ $A_5$ $2^{90}5^{78}$
$x^5-x^4+8x^3-6x^2+14x-6$ $A_5$ $2^{90}193^{30}$
$x^5-7x^3-17x^2+18x+73$ $A_5$ $2^{40}487^{30}$
$x^5+8x^3+7x^2+172x+53$ $A_5$ $2083^{30}$
: Tested polynomials[]{data-label="tab:examples"}
To illustrate the methods, we discuss in detail the $S_5$ field of discriminant $2^{96}3^{60}17^{96}$ given by the polynomial $f(x)=x^5-68(x+1)$. Recall that $S_5$ has seven irreducible representations. We label them $1$, $\chi$, $\rho_4$, $\rho_4'=\rho_4\otimes\chi$, $\rho_5$, $\rho_5'=\rho_5\otimes\chi$ and $\rho_6$, where $\chi$ is the sign character and the subscripts indicate the dimensions. As it will turn out, the limiting factor in our computations is the conductor of $\rho_6$, which in our example is $36081072=2^4 3^3 17^4$. This is the smallest among the table of $S_5$ polynomials given in [@nfdatabase]; since that table is ordered by the conductor of $\rho_4$, it is likely that smaller examples exist. (We note, however, that if one is interested only in verifying some instances of Artin’s conjecture and not the Riemann hypothesis, the holomorphy of $L(s,\rho_5')$ may be checked much more easily; there the limiting factor is the conductor of $\rho_4$, of which [@nfdatabase] yields examples as small as $1609$. We have not pursued this possibility.)
Note that $1$, $\chi$ and $\rho_6$ are monomial representations, so Artin’s conjecture is true for those. Equation below shows that $L(s,\rho_4)$ and $L(s,\rho_5)$ are holomorphic except possibly at the zeros of $\zeta(s)$. Twisting by $\chi$, we see similarly that $L(s,\rho_4')$ and $L(s,\rho_5')$ are holomorphic away from the zeros of $L(s,\chi)$. Moreover, we learn from GAP that the representations $\rho_4\oplus\rho_6$, $\rho_4'\oplus\rho_6$, $\rho_5\oplus\rho_4'\oplus\chi$ and $\rho_5'\oplus\rho_4\oplus1$ are all monomial. Thus, in order to verify the holomorphy of $L(s,\rho_4)$, $L(s,\rho_4')$, $L(s,\rho_5)$ and $L(s,\rho_5')$ it is enough to check that $L(s,\rho_6)$ and $L(s,\rho_4'\oplus\chi)$ are non-vanishing at zeros of $\zeta(s)$ and, similarly, that $L(s,\rho_6)$ and $L(s,\rho_4\oplus1)$ do not vanish at the zeros of $L(s,\chi)$. Applying Turing’s method to these functions as well as $L(s,\rho_5\oplus1)$ and $L(s,\rho_5'\oplus\chi)$, we can deduce both Artin’s conjecture and the Riemann hypothesis (up to the tested height) for all representations.
In what follows we describe the numerical procedure in detail for $L(s,\rho_6)$. First we must choose a value of $\eta$ to use for the computation of $G(u;\eta,\{\mu_j\})$. One can aim to limit either the number of Dirichlet coefficients $a_n$ or the precision required in the computation. Since the coefficients are relatively easy to compute in our case (we have $2^{32}$ of them), we try for the latter. The largest error comes from Lemma \[lem:Ftail\] with $x=0$, and is of size roughly $M^c\exp\bigl(-\pi r\delta e^{-\delta}(M/\sqrt{N})^{2/r}\bigr)$, where $\delta=\frac{\pi}2(1-|\eta|)$, $r=6$, $N=36081072$ and $M=2^{32}$. Examining the local factors at small primes we determine that $|a_n|\le 1.26n^{\log_{4243}6}$, yielding $c=\frac23+\log_{4243}6$. (The $\mu_j$ in this case are $0,0,0,1,1,1$.) This error term should be compared to the size of the function being evaluated, which is roughly $e^{-\delta rt/2}$. From Theorem \[Stbound\] we find that to apply Turing’s method up to height $t=100$ we need to be able to compute the $L$-function up to about $t=115$. Trying a few values of $\eta$, we find that with $\eta=0.98$ the error terms are of size $10^{-14}$, compared to $10^{-5}$ for the size of the function. Thus, with this choice we should use a precision of at least $14$ digits; in fact we carry out most computations to $30$ digits.
All computations were performed on a 3GHz PC running Linux. They were divided into several steps:
1. Computing the Dirichlet coefficients $a_n$;
2. Estimating zeros by the explicit formula;
3. Computing $G^{(k)}(u_m;\eta,\{\mu_j\})$;
4. Computing $S^{(k)}_m$;
5. Computing and $L(s,\rho_6)$ by FFT;
6. Turing’s method.
To ensure correct results, we used the arbitrary precision interval arithmetic package MPFI [@MPFI] for steps 3 through 7. We discuss the steps in more detail below.
Computing $a_n$ {#numerics:coeff}
---------------
First we consider methods of computing the coefficients of the $L$-functions $L(s,\rho)$ for all irreducible representations $\rho$. One way is to express them as ratios of products of Hecke $L$-functions, as given by Brauer’s theorem; in fact for $S_5$ we may express each in terms of Dedekind zeta functions of intermediate fields. Precisely, let $k$ be the quadratic extension of ${\mathbb{Q}}$ associated to $\chi$, $F={\mathbb{Q}}(x_1)\subset M={\mathbb{Q}}(x_1,x_2)$ where $x_1$ and $x_2$ are distinct roots of $f$, and $E={\mathbb{Q}}(y)\subset K$ where $y$ is a root of the sextic resolvent (a formula for which is given in [@dummit]); then we have $$\begin{aligned}
L(s,\chi) &= \frac{\zeta_k(s)}{\zeta(s)},&\quad
L(s,\rho_4) &= \frac{\zeta_F(s)}{\zeta(s)},&\quad
L(s,\rho_5) &= \frac{\zeta_E(s)}{\zeta(s)},&\\
L(s,\rho_6)&=
\frac{\zeta_k(s)\zeta_E(s)\zeta_M(s)}{\zeta_{kE}(s)\zeta_F(s)^2},&\quad
L(s,\rho_4')&=
\frac{\zeta(s)\zeta_{kF}(s)}{\zeta_k(s)\zeta_F(s)},&\quad
L(s,\rho_5')&=
\frac{\zeta(s)\zeta_{kE}(s)}{\zeta_k(s)\zeta_E(s)}.&
\end{aligned}
\label{eq:zetaratio}$$ In turn, we may compute each of the Dedekind zeta functions using the ideal factorization functions built in to PARI [@PARI]. This facilitates the computation of local factors at primes dividing the discriminant, allowing us to avoid a detailed study of the possible types of ramification. However, it is not well-suited to working out many coefficients.
Fortunately, there is a faster method that works for all but finitely many primes. Table \[tab:S5factors\] shows the unramified local factors for each representation and conjugacy class (labelled by the order of elements in the class), where we write $x$ for $p^{-s}$. For $S_5$ it turns out that the Frobenius conjugacy class at $p$ is determined by the number of linear and quadratic factors of the reduction $\bar{f}$ of $f$ modulo $p$, which may be computed from the degrees of $\gcd\bigl(x^{p^n}-x,\bar{f}(x)\bigr)$ for $n=1,2$. That computation requires $O(\log p)$ multiplications and additions mod $p$. Thus, by the prime number theorem, for each $L(s,\rho)$ we may determine the Dirichlet coefficients $a_n$ for $n\le X$ in time $O(X)$ (assuming mod $p$ multiplications and additions take bounded time, which is appropriate for numbers of the size that we consider). Up to the implied constant, that is best possible. Moreover, the technique is very fast in practice; we found that it takes approximately seven hours to compute the local factors for all $p<2^{32}$.
$\rho$ 1 2a 2b 3
----------- ------------------ ------------------- -------------------- --------------------
$1$ $1-x$ $1-x$ $1-x$ $1-x$
$\chi$ $1-x$ $1+x$ $1-x$ $1-x$
$\rho_4$ $(1-x)^4$ $(1-x)^2(1-x^2)$ $(1-x^2)^2$ $(1-x)(1-x^3)$
$\rho_4'$ $(1-x)^4$ $(1+x)^2(1-x^2)$ $(1-x^2)^2$ $(1-x)(1-x^3)$
$\rho_5$ $(1-x)^5$ $(1+x)(1-x^2)^2$ $(1-x)(1-x^2)^2$ $(1+x+x^2)(1-x^3)$
$\rho_5'$ $(1-x)^5$ $(1-x)(1-x^2)^2$ $(1-x)(1-x^2)^2$ $(1+x+x^2)(1-x^3)$
$\rho_6$ $(1-x)^6$ $(1-x^2)^3$ $(1+x)^2(1-x^2)^2$ $(1-x^3)^2$
$\rho$ 4 5 6
$1$ $1-x$ $1-x$ $1-x$
$\chi$ $1+x$ $1-x$ $1+x$
$\rho_4$ $1-x^4$ $1+x+x^2+x^3+x^4$ $(1+x)(1-x^3)$
$\rho_4'$ $1-x^4$ $1+x+x^2+x^3+x^4$ $(1-x)(1+x^3)$
$\rho_5$ $(1-x)(1-x^4)$ $1-x^5$ $(1+x+x^2)(1+x^3)$
$\rho_5'$ $(1+x)(1-x^4)$ $1-x^5$ $(1-x+x^2)(1-x^3)$
$\rho_6$ $(1+x^2)(1-x^4)$ $(1-x)(1-x^5)$ $1-x^6$
: $S_5$ unramified local factors[]{data-label="tab:S5factors"}
Estimating zeros {#numerics:explicit}
----------------
With our computed coefficients, we readily obtain estimates for the low zeros by the method of Section \[sec:explicit\]. Figure \[fig:FXt\] shows graphs of $F_X(t)$, with $X=\log(2^{32})$, for each irreducible $L$-function. The spikes correspond to zeros, from which we get the estimates for the ordinate of the lowest zero of each function shown in Table \[tab:zeros\]; note that for $\zeta$ the estimate agrees with the known value $14.1347251417\ldots$ to within the precision of the computation. The increase in density of zeros with the conductor and degree is apparent in the graphs. Moreover, as the explicit formula is very sensitive to errors in the coefficients, the fact that we see spikes of height $1$ for the low zeros indicates that our coefficients were computed correctly. Each graph took a few minutes to generate.
=6.4in
$\rho$ conductor lowest zero $\rho$ conductor lowest zero
---------- ------------ ---------------- ----------- ------------ ----------------
$1$ $1$ $14.134725142$ $\chi$ $3$ $8.039737156$
$\rho_4$ $4009008$ $1.108937765$ $\rho_4'$ $36081072$ $0.5717508665$
$\rho_5$ $36081072$ $1.062064850$ $\rho_5'$ $12027024$ $0.8132800720$
$\rho_6$ $36081072$ $1.376872200$
: Conductor and ordinate of the lowest zero of each irreducible $L$-function[]{data-label="tab:zeros"}
Computing $G^{(k)}(u_m;\eta,\{\mu_j\})$
---------------------------------------
Next we compute local approximations of $G(u;\eta,\{\mu_j\})$ for $u$ in the interval $\Bigl[\log\frac1{\sqrt N},\log\frac{2^{32}}{\sqrt N}\Bigr]$. We evaluate $2^{13}$ Taylor series of $16$ terms using ; with these choices, the error term in is less than $10^{-28}$. This calculation is the most delicate, due to high precision and catastrophic cancellation. Nevertheless, the computation time for this stage was only a few hours. The graph of $|G(u;\eta,\{\mu_j\})|$ is shown in Figure \[fig:Ggraph\].
=6.1in
Computing $S^{(k)}_m$, and $L(s,\rho_6)$
----------------------------------------
Now we come to the main part of the computation, . Most of the time, approximately twelve hours, was spent computing $S^{(k)}_m$. Note that if we had not adjusted $\eta$ to reduce the precision, this calculation could have taken substantially longer.
Once we have $S^{(k)}_m$, the computation of and $L(s,\rho_6)$ is very fast. We choose $B=\frac{2\pi\cdot2^{12}}{\log(2^{32})}\approx1160$. Since this is much larger than $t=115$, the errors terms from Lemma \[lem:Lambdasum\] are negligible. We choose $A=2^{20}/B\approx900$, which is about $160$ times the expected density $\frac1{2\pi}\log N\!\left(\frac{t}{2\pi}\right)^r$ of zeros around $t=115$. Thus, the main Fourier transform is of $2^{20}$ points, which takes only a few minutes to compute.
Figure \[fig:Lgraph1\] shows the graph of $Z(t):=\Lambda\bigl(\frac12+it\bigr)/\bigl|\gamma\bigl(\frac12+it\bigr)\bigr|$, which is the analogue of Riemann’s $Z$ function. We have superimposed the graph of $25F_X(t)$ over the same range; note the good agreement in location of zeros between the two, which gives evidence that our computations are correct. Figure \[fig:Lgraph2\] shows $Z(t)$ over the higher range $t\in [90,100]$.
=6.1in
=6.1in
Turing’s method {#turings-method}
---------------
Finally, we apply Turing’s method to the computed $L$-functions. At the same time, we verify the “working hypothesis” that we can isolate the zeros of the irreducible $L$-functions. This verification takes only a few seconds.
[^1]: The author was supported by an NSF postdoctoral fellowship
[^2]: Reading Turing’s paper on the subject, which was his last, one marvels at what he accomplished with the limited computational resources of the day. His method was truly ahead of its time.
[^3]: Tollis applied his method to cubic and quartic fields. In these cases, there is a slight advantage in passing to the normal closure and separating into irreducible Artin $L$-functions, as we have done for the $A_5$ cases in Section \[numerics\].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Motivated by the conjecture of Hartsfield and Ringel on antimagic labelings of undirected graphs, Hefetz, Mütze, and Schwartz initiated the study of antimagic labelings of digraphs in 2010. Very recently, it has been conjectured in \[Antimagic orientation of even regular graphs, J. Graph Theory, 90 (2019), 46-53.\] that every graph admits an antimagtic orientation, which is a strengthening of an earlier conjecture of Hefetz, Mütze and Schwartz. In this paper, we prove that every $2d$-regular graph (not necessarily connected) admits an antimagic orientation, where $d\ge2$. Together with known results, our main result implies that the above-mentioned conjecture is true for all regular graphs.'
author:
- |
Donglei Yang[^1]\
[ Department of Mathematics]{}\
[ Shandong University, Jinan, China]{}\
title: 'A note on antimagic orientations of even regular graphs\'
---
**Keywords**: Antimagic labeling; Antimagic orientation; Euler tour
Introduction
============
All graphs in this paper are finite and simple. For a graph $G$, we use $e(G)$ to denote the number of edges of $G$ and $d_G(u)$ to denote the degree of a vertex $u\in V(G)$. For a positive integer $k$, we denote $[k]:=\{1,2,\dots, k\}$. An *antimagic labeling* of a graph $G$ is a bijection from $E(G)$ to $[e(G)]$ such that for any distinct vertices $u$ and $v$, the sum of labels on edges incident to $u$ differs from that for edges incident to $v$. A graph $G$ is antimagic if it has an antimagic labeling. Hartsfield and Ringel [@NG] introduced antimagic labelings in 1990 and conjectured that every connected graph other than $K_2$ is antimagic. The first significant progress on this problem is a result of Alon, Kaplan, Lev, Roditty, and Yuster [@Alon], which states that there exists an absolute constant $c$ such that every graph on $n$ vertices with minimum degree at least $c\log n$ is antimagic. Later, Eccles [@E] improved this by showing that there exists an absolute constant $c_0$ such that if $G$ is a graph with average degree at least $c_0$, and $G$ contains no isolated edge and at most one isolated vertex, then $G$ is antimagic. For some special classes of graphs, Hefetz [@DH] proved that any graph on $n=3^k$ vertices that admits a triangle factor is antimagic. Later, Hefetz, Saluz and Tran [@DHA] generalized this by proving that any graph on $n=p^k$ vertices that admits a $C_p$-factor is antimagic, where $p$ is an odd prime. For regular graphs, Cranston, Liang, and Zhu [@DYZ] proved that every odd regular graph is antimagic. Later, Chang, Liang, Pan, and Zhu [@CLPZ] proved the even case. For more result, we recommend a survey [@JAG].
Motivated by antimagic labelings of graphs, Hefetz, Mütze, and Schwartz [@DTJ] initiated the study of antimagic labelings of digraphs. Let $D$ be a digraph and $A(D)$ and $V(D)$ be the set of arcs and vertices of $D$, respectively. An *antimagic labeling* of $D$ is bijection from $A(D)$ to $\{1,2,\ldots,|A(D)|\}$ such that no two vertices in $D$ have the same vertex-sum, where the vertex-sum of a vertex $u\in V(D)$ is the sum of labels of all arcs entering $u$ minus the sum of labels of all arcs leaving $u$. A digraph is *antimagic* if it has an antimagic labeling. A graph $G$ has an *antimagic orientation* if an orientation of $G$ is antimagic.
Hefetz, Mütze, and Schwartz [@DTJ] raised the questions “Is every orientation of any connected graph antimagic?” and “Does every graph admit an antimagic orientation?”. They proved an analogous result of Alon, Kaplan, Lev, Roditty, and Yuster [@Alon] that there exists an absolute constant $c$ such that every orientation of any graph on $n$ vertices with minimum degree at least $c\log n$ is antimagic. Except for $K_{1,2}$ and $K_3$, no other counterexample to the first question is known. For the second question, the same authors proposed the following conjecture.
*[@DTJ]*\[conj1\] Every connected graph admits an antimagic orientation.
They proved the following result on certain classes of regular graphs.
*[@DTJ]*\[odd\] For any integer $d\ge1$,
(a)
: every $(2d-1)$-regular graph admits an antimagic orientation;
(b)
: every connected, $2d$-regular graph $G$ admits an antimagic orientation if $G$ has a matching that covers all but at most one vertex of $G$.
They [@DTJ] also pointed out that “It seems hard to discard any of the two conditions in Theorem \[odd\] (b), that is connectedness and having a matching that covers all but at most one vertex. In fact, we do not even know if every disjoint union of cycles admits an antimagic orientation.”
Recently, Li et al. [@lt] confirmed the case when $G$ is a disjoint union of cycles and also proved a stronger result than Theorem \[odd\] (b).
*[@lt]*\[li\] For any integer $d\ge2$,
(a)
: every $2$-regular graph admits an antimagic orientation;
(b)
: every $2d$-regular graph with at most two odd components admits an antimagic orientation.
In the same paper [@lt], the authors proposed a stronger conjecture as follows.
*[@lt]*\[conj2\] Every graph admits an antimagic orientation.
In this paper, we prove the following main result.
\[main\] Every $2d$-regular graph admits an antimagic orientation, where $d\geq2$.
Together with Theorem \[odd\] (a) and Theorem \[li\] (a), Theorem \[main\] implies that Conjecture \[conj2\] holds for all regular graphs.
\[reg\] Every regular graph admits an antimagic orientation.
Recently, Shan and Yu [@SY] affirmed Conjecture \[conj2\] for biregular bipartite graphs, where we only require that any two vertices in the same part have the same degree.
It is worth noting that the proof of Thereom \[main\] is similar to the proof of Theorem \[li\] given in [@lt]. We label the edges of odd components judiciously so that we obtain a desired antimagic orientation.
We need to introduce one more notation. A closed walk in a graph $G$ is an *Euler tour* if it traverses every edge of the graph exactly once. The following is a result of Euler.
\[euler\] A connected graph admits an Euler tour if and only if every vertex has even degree.
Proof of Theorem \[main\]
=========================
Let $G$ be a $2d$-regular graph on $n$ vertices with $d\geq2$, then by Theorem \[li\] (b), we assume that $G$ contains more than two odd components. Let $G_1,G_2,\ldots,G_s,G_{s+1},\ldots,G_t$ be components of $G$ with $|V(G_i)|=n_i$ and $e(G_i)=m_i$ for each $i\in[t]$, where $G_1,G_2,\ldots,G_s$ are all the odd components of $G$ with $n_1\leq n_2\leq\ldots\leq n_s$ and $s\geq3$. By Theorem \[euler\], each $G_i$ has an Euler tour $T_{i}$ for $i\in[t]$. We can regard each $T_i$ as a simple cycle of size $e(G_i)$, say $C^*_i$, and each vertex $u\in V(G)$ has $d$ copies in $C^*_i$. For each vertex $u\in V(G_i)$, we can arbitrarily pick one of the $d$ copies of $u$ as a real vertex and the remaining $d-1$ copies as imaginary vertices. Therefore, $C^*_i$ has $n_i$ real vertices and $(d-1)n_i$ imaginary vertices. Moreover, we can pick the real vertices as follows.
\[c1\] For any $i\in [t]$, we can pick $n_i$ real vertices in $C^*_i$ such that $C^*_i$ contain four consecutive vertices, say, $a,b,c,d$ such that $a,b,d$ are real vertices and $c$ is a imaginary vertex.
Since $G_i$ is $2d$-regular with $d\geq2$, we have $n_i\geq5$ and let $x_1,x_2,x_3,x_4,x_5$ be five consecutive vertices of $C^*_i$. If $x_1$ and $x_4$ are distinct vertices, then treat $x_3$ as an imaginary vertex (this is possible when $d\geq2$) and $a=x_1, b=x_2, c=x_3,d=x_4$ will suffice. If $x_1=x_4$, i.e., $x_1x_2x_3$ forms a triangle in $G_i$, then treat $x_4$ as an imaginary vertex and $a=x_2,b=x_3,c=x_4,d=x_5$ will suffice.
For simplicity, let $V^i_R$ be the set of $n_i$ real vertices in $C^*_i$, and label them with $v^i_{1},v^i_{2},\ldots,v^i_{n_i}$ in order. By renaming them if necessary, together with Claim \[c1\], we may further assume that $a=v^i_{n_i}, b=v^i_{1},d=v^i_{2}$ for each $i\in[t]$. For each $i\in [t]$ and each $1\leq l\leq n_i-1$, let $P^i_l$ be the path between $v^i_{l}$ and $v^i_{l+1}$ in $C^*_i$ such that either $v^i_{l}v^i_{l+1}\in E(C^*_i)$ or all the internal vertices of $P^i_l$ are imaginary vertices, and we call such paths $P^i_l$ *good*.
We observe that each orientation (labelling) of $C^*_i$ corresponds to an unique orientation (labelling) of $G_i$, so in the following proof we only consider orientations (labellings) on $C^*_i$. Next, we will finish the proof by orienting and labelling the edge set of all $C^*_i$ such that the corresponding orientation of $G$ has an antimagic labelling.
We orient each $C^*_i$ as follows. For each $C^*_i$ with $i\in[s]$, firstly we orient the edge $v^i_{n_i}v^i_{1}$ from $v^i_{n_i}$ to $v^i_{1}$, and then orient the path $P^i_1$ to be a directed path from $v^i_{1}$ to $v^i_{2}$. Next, for all $l\in\{2,3,\ldots,n_i-1\}$, we orient the path $P^i_l$ to be a directed path between $v_l^i$ and $v^i_{l+1}$ such that each real vertex in $\{v^i_{2},\ldots,v^i_{n_i}\}$ has outdegree either $2$ or $0$ (see Figure \[f1\]). For each $C^*_j$ with $j>s$, firstly we orient the edge $v^j_{n_j}v^j_{1}$ from $v^j_{n_j}$ to $v^j_{1}$. Next, for all $l\in\{1,2,3,\ldots,n_j-1\}$, we orient the path $P^j_l$ to be a directed path between $v_l^j$ and $v^j_{l+1}$ such that any real vertex in $\{v^j_{1},v^j_{2},\ldots,v^j_{n_j}\}$ has outdegree either $2$ or $0$ (this is possible since $n_j$ is even).
Now we have the desired orientation of all $C^*_i$, say $D$. Let $N_i=\sum\limits_{j=1}^im_j$ for each $i\in [t]$, we label $C^*_{1},C^*_{2},\ldots,C^*_{s}$ with $[N_s]$ and then $C^*_{s+1},\ldots,C^*_{t}$ with $\{N_s+1,N_s+2,\ldots,N_t\}$ as follows.
\[label1\]
Denote by $L$ the set of edges which have been labelled as above
Note that in **Algorithm 1**, **steps** $1$-$10$ are devoted to labelling $C^*_1,C^*_2,\ldots,C^*_s$ with integers in $[N_s]$. Since $D$ is a family of oriented cycles, we say a real vertex $u$ *sees* $(x,y)$ in the resulting labelling $c$ if the two arcs incident with $u$ have labels $x$ and $y$ with $x<y$. For simplicity, we say each real vertex $v^i_l$ in $C^*_i$ sees $(x^i_l,y^i_l)$ for any $i\in [s]$ and $l\in [n_i]$.
By running **steps** $7$-$8$, intuitively, we have the following claim. Using induction on $l$, we can easily prove the claim, here we omit it.
\[c3\] For any two $ C^*_i, C^*_j$ with $i<j\leq s$ and any integer $l$ with $2\leq l\leq\lceil n_i/2\rceil$, we have $x^i_{n_i-l+2}<x^j_{n_i-l+2}<x^i_l<x^j_l$.
![$C^*_i$ and $C^*_j$ in Claim \[c3\]. []{data-label="f1"}](f11.pdf "fig:"){width="465pt"}\
Since each $G_i$ is $2d$-regular, every vertex $u$ has $d-1$ imaginary copies in $D$ and each imaginary copy contributes $-1$ to the vertex sum of $u$ in the resulting labelling $c$. So in order to make sure that no two vertices in $G$ have the same vertex sum, it suffices to prove that no two real vertices in $D$ have the same vertex sum. For any vertex $u\in V(D)$, denote by $S(u)$ the vertex sum of $u$ in $D$. We complete our proof with the following claim.
\[c2\] No two real vertices in $D$ have the same vertex sum.
Recall that $V_R^i$ is the set of real vertices in $C^*_i$ and $|V_R^i|=n_i$ for each $i\in [t]$. Let $X=\{v^i_1| i\in [s]\}$ and $Y=\bigcup\limits^s_{i=1}V_R^i-X$. Firstly, we have the following intuitive observations.
1. For each $v^i_1\in X$, $S(v^i_1)=-i-s+1$.
2. For each $u\in Y$, $|S(u)|\geq 3s+1$.
3. For any real vertices $u\in V_R^i, v\in V_R^j$ and $w\in V_R^k$ with $i<s+1\leq j<k\leq t$, we have $|S(u)|<|S(v)|<|S(w)|$.
Actually, it is not difficult to prove that for any real vertices $u, v\in V_R^k$ with $k>s$, we have $S(u)\neq S(v)$, and the proof can be found in [@lt]. Hence, it remains to prove that no two real vertices in $Y$ have the same vertex sum.
Suppose to the contrary there exist two real vertices $u,v\in Y$ with the same vertex sum and assume that $u$ sees $(a,b)$, $v$ sees $(c,d)$ with $a+b=c+d$. Remember that $u,v$ must have the same outdegree $2$ or $0$.
If $u,v\in V_R^i$ for some $i\in [s]$, then $u,v$ cannot be two consecutive real vertices in $C^*_i$. By symmetry, we assume $a<c$, then by **steps** $6-7$, the number $b$ will be used prior to $d$, thus $b<d$, a contradiction since $a+b=c+d$.
Suppose $u\in V_R^i, v\in V_R^j$ with $i<j\leq s$, we consider the following cases depending on the locations of $u,v$.
If $u\neq v^i_{\lceil n_i/2\rceil}$ and $v\neq v^j_{\lceil n_j/2\rceil}$, then $a<c$ (or $c<a$) implies that $b$ ($d$) is used prior to $d$ ($b$), thus $b<d$ ($d<b$), a contradiction.
If $u=v^i_{\lceil n_i/2\rceil}$ and $v=v^j_{\lceil n_j/2\rceil}$, then we claim that $n_i=n_j$, in fact, $n_i<n_j$ implies $\lceil n_i/2\rceil<\lceil n_j/2\rceil$ and it follows that $a,b$ are used prior to $b,d$, i.e., $a<b<c<d$, a contradiction. Since $i<j$ and $n_i=n_j$, we have $a<c$ and $x^i_{\lceil n_i/2\rceil+1}<x^j_{\lceil n_j/2\rceil+1}$ by Claim \[c3\]. Thus $b$ is used in **step** $7$ prior to $d$, i.e., $b<d$, a contradiction.
If $u=v^i_{\lceil n_i/2\rceil}$ and $v=v^j_{l}$ with $l\neq\lceil n_j/2\rceil$, then we claim that $l=\lceil n_i/2\rceil$. To see this, by Claim \[c3\], we observe that either $l=\lceil n_i/2\rceil$ or $l=n_j-\lceil n_i/2\rceil+2$ and since $v^j_{n_j-\lceil n_i/2\rceil+2}$ and $v^i_{\lceil n_i/2\rceil}$ do not have the same outdegree in $D$, the case $l=n_j-\lceil n_i/2\rceil+2$ cannot happen. So we have $x^i_{\lceil n_i/2\rceil+1}<x^i_{\lceil n_i/2\rceil}=a<c$, and it follows that $b$ is used in **step** $7$ prior to $d$, i.e., $b<d$, a contradiction.
If $u=v^i_{l}$ with $l\neq\lceil n_i/2\rceil$ and $v=v^j_{\lceil n_j/2\rceil}$, then by Claim \[c3\], we must have $n_i=n_j$ and $l=\lceil n_i/2\rceil+1$. By the orientation $D$, we can easily observe that $u,v$ do not have the same out-degree, a contradiction. This completes the proof of Claim \[c2\].
**Remark.** As we present an orientation and an antimagic labelling explicitly, the proof of Theorem \[main\] is elementary. Claim \[c1\] turns out to be very helpful, that is, we could pick three real vertices and control the length of path between any two consecutive real vertices to some extent. Applying the same arguments in the proof of Theorem \[main\], we can prove a more general result.
Let $d\ge2$ be an integer. If every vertex of a graph $G$ has degree $2d$ or $2d+2$, then $G$ admits an antimagic orientation.
**Acknowledgement.** The author would like to thank Tong Li, Zi-Xia Song and Guanghui Wang for helpful discussion. This work was supported by the National Natural Science Foundation of China (11871311).
[99]{}
N. Alon, G. Kaplan, A. Lev, Y. Roditty, R. Yuster, Dense graphs are antimagic, J. Graph Theory, 47 (2004), 297–309. F. Chang, Y-C. Liang, Z. Pan, X. Zhu, Antimagic labeling of regular graphs, J. Graph Theory, 82 (2016), 339–349. D. W. Cranston, Y-C. Liang, X. Zhu, Regular graphs of odd degree are antimagic, J. Graph Theory, 80 (2015), 28-33. T. Eccles, Graphs of large linear size are antimagic, J. Graph Theory, 81 (2016), 236-261.
J. A. Gallian, A dynamic survey of graph labeling, Electron J. Combin., DS6 (2016).
N. Hartsfield, G. Ringel, Pearls in Graph Theory, Academic Press, Boston, (1990), 108-109 (revised version, 1994).
D. Hefetz, Anti-magic graphs via the Combinatorial Nullstellensatz, J. Graph Theory, 50 (2005), 263–272. D. Hefetz, A. Saluz, H. T. Tran, An application of the Combinatorial Nullstellensatz to a graph labeling problem, J. Graph Theory, 65 (2010), 70-82. D. Hefetz, T. Mütze, J. Schwartz, On antimagic directed graphs, J. Graph Theory, 64 (2010), 219-232. T. Li, Z-X. Song, G. Wang, D. Yang, C-Q. Zhang, Antimagic orientations of even regular graphs, J. Graph Theory, 90 (2019), 46-53. S. Shan, X. Yu, Antimagic orientation of biregular bipartite graphs, Electron J. Combin., 24 (4)(2017), P4.31
[^1]: Email address: [email protected] (D. Yang)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We introduce an embedding of the Torelli group of a compact connected oriented surface with non-empty connected boundary into the completed Kauffman bracket skein algebra of the surface, which gives a new construction of the first Johnson homomorphism.'
author:
- Shunsuke Tsuji
title: The Torelli group and the Kauffman bracket skein module
---
Introduction
============
In [@Turaev], Turaev draw an analogy between the Goldman Lie algebra and some skein algebra. On the other hand, in [@Kawazumi] [@KK] [@MT], Kawazumi, Kuno, Massuyeau and Turaev found that the Goldman Lie algebra on a compact connected oriented surface $\Sigma$ plays an important role in study of the mapping class group of the surface. In our preceding paper [@TsujiCSAI], we establish an explicit relationship between the Kauffman bracket skein algebra and the mapping class group on the surface. In particular, we obtain a formula for the action of the Dehn twist along a simple closed curve $c$ on the completed Kauffman bracket skein module with base point set an finite set $J \in \partial \Sigma$, in terms of the inverse function of the hyperbolic cosine function $$\exp(\sigma((L(c)))(\cdot)
=t_c (\cdot): \widehat{{\mathcal{S}( \Sigma,J )}} \to \widehat{{\mathcal{S}( \Sigma,J )}}$$ [@TsujiCSAI] Theorem 4.1 where $$L(c) {\stackrel{\mathrm{def.}}{=}}\frac{-A+{ A^{-1}}}{4 \log (-A)} ({\mathrm{arccosh}}(-\frac{c}{2}))^2
-(-A+{ A^{-1}})\log (-A).$$ This formula is an analogy of the formula for the action of the Dehn twist along a simple closed curve $c$ on the completion of the group ring of the fundamental group of the surface [@Kawazumi] [@KK] [@MT]. The aim of this paper is to clarify the relationship between mapping class groups and Kauffman bracket skein algebras based on the Dehn twists formula stated above.
Let $\Sigma$ be a compact connected oriented surface with non-empty connected boundary and genus $g>1$, and $\widehat{{\mathcal{S}( \Sigma )}}$ the completion of the Kauffman bracket skein algebra on $\Sigma$ in an $\ker \epsilon$-adic topology where $\ker \epsilon$ is an augmentation ideal. We define a filtration ${\{ F^n \widehat{{\mathcal{S}( \Sigma )}} \}_{n \geq 0}}$ in [@Tsujipurebraid] satisfying $$[F^3 \widehat{{\mathcal{S}( \Sigma )}}, F^n \widehat{{\mathcal{S}( \Sigma )}}]
\subset F^{n+1} \widehat{{\mathcal{S}( \Sigma )}}$$ where Lie bracket $[ \ \ , \ \ ]$ is defined by $$[x,y] {\stackrel{\mathrm{def.}}{=}}\frac{1}{-A+{ A^{-1}}} (xy-yx)$$ for $x$ and $y \in \widehat{{\mathcal{S}( \Sigma )}}$. We remark this filtration also induces the topology induced by the augmentation ideal. By this filtration, we can consider the logarithm of any element of the Torelli group of $\Sigma$.
By the above condition, we can define the Baker-Campbell-Hausdorff series ${\mathrm{bch}}(\cdot, \cdot)$ on $F^3 \widehat{{\mathcal{S}( \Sigma )}}$. We define $C(c_1,c_2)$ by ${\mathrm{bch}}(L(c_1),L(c_2),-L(c_1),-L(c_2))$ for a pair ${\{ c_1,c_2 \}}$ of simple closed curves whose algebraic intersection is $0$. In view of Putman’s result [@Pu2008], we study the three subsets $\mathcal{L}_{\mathrm{comm}}(\Sigma)$, $\mathcal{L}_{\mathrm{bp}}(\Sigma)$ and $\mathcal{L}_{\mathrm{sep}}(\Sigma)$ of $F^3 \widehat{{\mathcal{S}( \Sigma )}}$ where we denote by $\mathcal{L}_{\mathrm{comm}}(\Sigma)$, $\mathcal{L}_{\mathrm{bp}}(\Sigma)$ and $\mathcal{L}_{\mathrm{sep}}(\Sigma)$ the set of all $C(c_1,c_2)$ for a pair ${\{ c_1,c_2 \}}$ of two simple closed curves whose intersection number is $0$, the set of all $L(c_1)-L(c_2)$ for a bounding pair ${\{ c_1,c_2 \}}$ and the set of all $L(c)$ for a separating simple closed curve $c$, respectively. The subsets $\mathcal{L}_{\mathrm{comm}}(\Sigma)$, $\mathcal{L}_{\mathrm{bp}}(\Sigma)$ and $\mathcal{L}_{\mathrm{sep}}(\Sigma)$ correspond to the subsets $\mathcal{I}_{\mathrm{comm}} (\Sigma)$, $\mathcal{I}_{\mathrm{bp}}(\Sigma)$ and $\mathcal{I}_{\mathrm{sep}}(\Sigma)$ of the Torelli group of $\mathcal{I} (\Sigma)$ where we denote by $\mathcal{I}_{\mathrm{comm}}(\Sigma)$, $\mathcal{I}_{\mathrm{bp}}(\Sigma)$ and $\mathcal{I}_{\mathrm{sep}}(\Sigma)$ the set of all $C_{c_1,c_2}=t_{c_1}t_{c_2}{t_{c_1}}^{-1}
{t_{c_2}}^{-1}$ for a pair ${\{ c_1,c_2 \}}$ of two simple closed curves whose intersection number is $0$, the set of all $t_{c_1}{t_{c_2}}^{-1}$ for a bounding pair ${\{ c_1,c_2 \}}$ and the set of all $t_c$ for a separating simple closed curve $c$, respectively. In fact, in subsection \[subsection\_injectivity\_of\_zeta\], we construct a surjective homomorphism $\theta : (I {\mathcal{S}( \Sigma )}, {\mathrm{bch}}) \to \mathcal{I}(\Sigma)$ defined by $C(c_1,c_2) \in \mathcal{L}_{\mathrm{comm}}(\Sigma) \mapsto C_{c_1,c_2}
\in \mathrm{I}_{\mathrm{comm}}(\Sigma)$, $L(c_1)-L(c_2) \in \mathcal{L}_{\mathrm{bp}}(\Sigma) \mapsto
t_{c_1}{t_{c_2}}^{-1} \in \mathcal{I}_{\mathrm{bp}}(\Sigma)$ and $L(c) \in \mathcal{L}_{\mathrm{sep}}(\Sigma)
\mapsto t_{c} \in \mathcal{I}_{\mathrm{sep}}(\Sigma$ where we denote by $I{\mathcal{S}( \Sigma )}$ the subgroup of $(F^3 \widehat{{\mathcal{S}( \Sigma )}},{\mathrm{bch}})$ generated by $\mathcal{L}_{\mathrm{comm}}(\Sigma)$, $\mathcal{L}_{\mathrm{bp}}(\Sigma)$ and $\mathcal{L}_{\mathrm{sep}}(\Sigma)$. This homomorphism satisfies $$\exp (\sigma (x)) (\cdot) =\theta (x) (\cdot) \in {\mathrm{Aut}}(\widehat{{\mathcal{S}( \Sigma,J )}})$$ for any $x \in I {\mathcal{S}( \Sigma )}$ and any finite subset $J$ of $\partial \Sigma$. Using the Putman’s infinite presentation [@Pu2008], we prove that $\theta$ is injective in subsection \[subsection\_well\_defined\_zeta\].
This embedding gives us a new way of studying the mapping class group. In fact, the embedding gives a new construction of the first Johnson homomorphism by $$\mathcal{I}(\Sigma) \to I {\mathcal{S}( \Sigma )} \hookrightarrow
F^3 \widehat{{\mathcal{S}( \Sigma )}} \twoheadrightarrow
F^3 \widehat{{\mathcal{S}( \Sigma )}}/F^4 \widehat{{\mathcal{S}( \Sigma )}}
\stackrel{\lambda^{-1}}{\simeq}
\wedge^3 H_1 (\Sigma,{\mathbb{Q}}).$$ where $\zeta {\stackrel{\mathrm{def.}}{=}}\theta^{-1}$. Here $\wedge^n H_1 (\Sigma,{\mathbb{Q}})$ the $n$-th exterior power of $H_1 (\Sigma,{\mathbb{Q}})$. These are analogies of [@KK] 6.3. In a subsequent paper, we study the Johnson kernel $\mathcal{K} (\Sigma)$, which is defined to be the kernel of the first Johnson homomorphism. The isomorphism $\zeta_{|\mathcal{K} (\Sigma)} :\mathcal{K} (\Sigma) \to
F^4 \widehat{{\mathcal{S}( \Sigma )}} \cap I {\mathcal{S}( \Sigma )}$ induces the natural map $\tau_{\zeta 2}: \mathcal{K} (\Sigma) \to
F^4 \widehat{{\mathcal{S}( \Sigma )}}/F^5 \widehat{{\mathcal{S}( \Sigma )}}
\simeq S^2 (S^2 (H_1 (\Sigma,{\mathbb{Q}}))) \oplus S^2 (H_1 (\Sigma, {\mathbb{Q}})) \oplus
{\mathbb{Q}}$ where we denote by $S^2 (V)$ the second symmetric tensor of a ${\mathbb{Q}}$ vector space $V$. Then we have $\tau_{\zeta 2} =(\iota \circ \tau_2) \oplus 0 \oplus d_{\mathrm{Casson}}$. Here $\iota: S^2 (\wedge^2 (H_1 (\Sigma, {\mathbb{Q}})))
\to S^2 (S^2 (H_1(\Sigma,{\mathbb{Q}})))$ is the ${\mathbb{Q}}$-linear map defined by $(a \wedge b) \cdot (c \wedge d) \to (a \cdot c) \cdot(b\cdot d)-
(a \cdot d) \cdot (b \cdot c)$, $\tau_2 $ the second Johnson homomorpshim and, $d_{\mathrm{Casson}}$ the core of the Casson invariant defined in [@Morita_Casson_core].
Furthermore, we expect that the embedding brings us some information about integral homology $3$-spheres including the Casson invariant. In a subsequent paper [@Tsujihom3], we construct an invariant $z(M)$ for an integral homology $3$-sphere $M$ which is an element of ${\mathbb{Q}}[[A+1]]$ using the embedding $\zeta$. We remark the coefficient of $(A+1)$ in $z(M)$ is the Casson invariant.
In section 2, we review some facts about the skein algebra and the mapping class group of a compact connected surface. In section 3, we construct an embedding from $\mathcal{I} (\Sigma) \to F^3 \widehat{{\mathcal{S}( \Sigma )}}$.
In section 3, we assume that the genus of a surface $\Sigma$ is larger than $2$ in order to use Putman’s theorem. Since $\mathcal{I} (\Sigma_{1,1})$ is generated by only the Dehn twist along the simple slosed curve which is parallel to the boundary of $\Sigma_{1,1,}$, the map $\zeta$ is aloso isomorphism.
Acknowledgment {#acknowledgment .unnumbered}
==============
The author would like to thank his adviser, Nariya Kawazumi, for helpful discussion and encouragement. This work was supported by JSPS KAKENHI Grant Number 15J05288 and the Leading Graduate Course for Frontiers of Mathematical Sciences and Phsyics.
Definition and Review {#section_definition_review}
=====================
In the section, we review some definitions and facts about the filtered Kauffman bracket skein algebra and module of a surface, for details, see our papers [@TsujiCSAI] and [@Tsujipurebraid].
Through this section, let $\Sigma$ be a compact connected surface with non-empty boundary and $I$ the closed interval $[0,1]$.
Kauffman bracket skein algebras and modules
-------------------------------------------
Let $J$ be a finite subset of $\partial \Sigma$. We denote by $\mathcal{T}(\Sigma,J)$ the set of unoriented framed tangles in $\Sigma \times I$ with base point set $J$ and by $T(d)$ the tangle presented by a tangle diagram $d$, for details, see [@TsujiCSAI] section 2 and [@Tsujipurebraid] section 2. Let ${\mathcal{S}( \Sigma,J )}$ be the Kauffman bracket skein module of $\Sigma$ with base point set $J \times {\{ \frac{1}{2} \}}$, which is the quotient of ${\mathbb{Q}}[A.{ A^{-1}}] \mathcal{T}(\Sigma,J)$ by the skein relation and the trivial knot relation [@TsujiCSAI] Definition 3.2. The skein relation is $$T(d_1) -AT(d_\infty)-{ A^{-1}} T(d_0)$$ where $d_1$, $d_\infty$ and $d_0$ are differ only in an open disk shown in Figure \[fig\_K2\], Figure \[fig\_Kinfi\] and Figure \[fig\_K0\], respectively. The trivial knot relation is $$T(d) -(-A^2-A^{-2})T(d')$$ where $d$ and $d'$ are differ only in a open disk shown a boundary of a disk and empty, respectively. We denote that we don’t assume “the boundary skein relation” and “the value of a contractible arc” in Muller [@Mu2012]. We write simply ${\mathcal{S}( \Sigma )} {\stackrel{\mathrm{def.}}{=}}{\mathcal{S}( \Sigma, \emptyset )}$. The element of ${\mathcal{S}( \Sigma,J )}$ represented by $T \in \mathcal{T}(\Sigma,J)$ is denoted by $[T]$.
[rrrr]{}
![$d_0$[]{data-label="fig_K0"}](K2_PNG.pdf){width="2cm"}
![$d_0$[]{data-label="fig_K0"}](Kinfi_PNG.pdf){width="2cm"}
![$d_0$[]{data-label="fig_K0"}](K0_PNG.pdf){width="2cm"}
There is a natural action of $\mathcal{M}(\Sigma)$ on ${\mathcal{S}( \Sigma,J )}$ [@TsujiCSAI] section 2, where $\mathcal{M}(\Sigma)$ is the mapping class group of $\Sigma$ fixing the boundary pointwise. The product of ${\mathcal{S}( \Sigma )}$ and the right and left actions of ${\mathcal{S}( \Sigma )}$ on ${\mathcal{S}( \Sigma,J )}$ are defined by Figure \[fig\_product\_action\], for details, see [@TsujiCSAI] 3.1. The Lie bracket $[,]:{\mathcal{S}( \Sigma )} \times {\mathcal{S}( \Sigma )} \to {\mathcal{S}( \Sigma )}$ is defined by $[x,y]{\stackrel{\mathrm{def.}}{=}}\frac{1}{-A+{ A^{-1}}} (xy-yx)$. and the action $\sigma()(): {\mathcal{S}( \Sigma )} \times {\mathcal{S}( \Sigma,J )}
\to {\mathcal{S}( \Sigma,J )}$ by $\sigma(x)(z) {\stackrel{\mathrm{def.}}{=}}\frac{1}{-A+{ A^{-1}}}(xz-zx)$. The action $\sigma$ makes ${\mathcal{S}( \Sigma,J )} $ a $({\mathcal{S}( \Sigma )},[,])$-module with $\sigma$. For details, see, [@TsujiCSAI] 3.2.
(300,83) (0,-20)[![The product and the actions[]{data-label="fig_product_action"}](sikaku_2_PNG.pdf "fig:"){width="90pt"}]{} (100,-20)[![The product and the actions[]{data-label="fig_product_action"}](sikaku_3_PNG.pdf "fig:"){width="90pt"}]{} (200,-20)[![The product and the actions[]{data-label="fig_product_action"}](sikaku_3_PNG.pdf "fig:"){width="90pt"}]{} (10,80)[$xy {\stackrel{\mathrm{def.}}{=}}$]{} (40,50)[$x$]{} (40,20)[$y$]{} (10,2)[$\mathrm{for} \ \ x,y \in {\mathcal{S}( \Sigma )}$]{} (0,10)[$0$]{} (0,55)[$1$]{} (0,32.5)[$I$]{} (40,65)[$\Sigma$]{}
(110,80)[$xz {\stackrel{\mathrm{def.}}{=}}$]{} (140,52)[$x$]{} (140,34)[$z$]{} (110,2)[$\mathrm{for} \ \ x\in {\mathcal{S}( \Sigma )}$]{} (110,-10)[$\ \ \mathrm{and} \ \ z \in {\mathcal{S}( \Sigma,J )}$]{} (100,10)[$0$]{} (100,55)[$1$]{} (100,32.5)[$I$]{} (140,65)[$\Sigma$]{}
(210,80)[$zx {\stackrel{\mathrm{def.}}{=}}$]{} (240,34)[$z$]{} (240,15)[$x$]{} (210,2)[$\mathrm{for} \ \ x \in {\mathcal{S}( \Sigma )}$]{} (210,-10)[$\ \ \mathrm{and} \ \ z \in {\mathcal{S}( \Sigma,J )}$]{} (200,10)[$0$]{} (200,55)[$1$]{} (200,32.5)[$I$]{} (240,65)[$\Sigma$]{}
Filtration and Lie bracket
--------------------------
The augmentation map $ \epsilon:{\mathcal{S}( \Sigma )} \to {\mathbb{Q}}$ is defined by $A+1 \mapsto 0$ and $[L]- (-2)^{{\lvert L \rvert}}\mapsto 0$ for $L \in \mathcal{T}(\Sigma)$, where ${\lvert L \rvert}$ is the number of components of $L$. The well-definedness of $\epsilon $ is proved in [@TsujiCSAI] Proposition 3.10.
We denote $\pi {\stackrel{\mathrm{def.}}{=}}\pi_1 (\Sigma, *)$ and by ${\mathbb{Q}}\pi$ the group ring of $\pi$ over ${\mathbb{Q}}$, where $*$ is a point of $\partial \Sigma$. For $x \in \pi_1(\Sigma)$, we define ${\langle x \rangle} \in (\ker \epsilon)/(\ker \epsilon)^2 $ by $[L_x]+2-3 w(L_x) (A-{ A^{-1}})$ using $L_x \in \mathcal{T}(\Sigma)$ with $p_1 (L_x)$ the conjugation class of $x$, where the writhe $w(L_x)$ is the sum of the signs of the crossing of a diagram presenting $L_x$. The well-definedness of ${\langle \cdot \rangle}$ is proved in [@Tsujipurebraid] Lemma 3.1. We also denote by ${\langle \cdot \rangle} :{\mathbb{Q}}\pi
\ker \epsilon / ( \ker \epsilon)^2$ its ${\mathbb{Q}}$-linear extension. By [@Tsujipurebraid] Proposition 3.3. we have ${\langle xy \rangle}+{\langle x{ y^{-1}} \rangle} =2 {\langle x \rangle}+2 {\langle y \rangle}$ for $x$ and $y \in \pi$. Using this formula, we have $$\begin{aligned}
&{\langle (a-1)(b-1)(c-1) \rangle} = {\langle -(b-1)(a-1)(c-1) \rangle}, \label{equation_calc_1}\\
&{\langle (a-1)(b-1)(c-1)(d-1) \rangle} =0, \label{equation_calc_2}\\
&{\langle (a-1)(b-1)(b-1) \rangle} = 0, \label{equation_calc_3}\\
&{\langle [a,b]c-c \rangle} = 2{\langle (a-1)(b-1)(c-1) \rangle}, \label{equation_calc_4}\\
&{\langle (a-1)(a-1) \rangle}=2{\langle a \rangle}, \label{equation_calc_5}\end{aligned}$$ for $a$, $b$, $c$ and $d \in \pi$. For details, see [@Tsujipurebraid] Lemma 3.4. By these equations, the ${\mathbb{Q}}$-linear map $$\lambda :H \wedge H \wedge H
\to \ker \epsilon / (\ker \epsilon)^2, [a]\wedge [b] \wedge [c]
\mapsto {\langle (a-1)(b-1)(c-1) \rangle}$$ is well-defined where $H {\stackrel{\mathrm{def.}}{=}}H_1 (\Sigma ,{\mathbb{Q}}) ={\mathbb{Q}}\otimes
\pi / [\pi , \pi]$. We remark that $\lambda$ is injective Corollary 4.6. Let $\varpi$ be the quotient map $\ker \epsilon \to \ker \epsilon /{\mathrm{im}}\lambda$. We define the filtration ${\{ F^n {\mathcal{S}( \Sigma )} \}_{n \geq 0}}$ by $$\begin{aligned}
&F^0 {\stackrel{\mathrm{def.}}{=}}{\mathcal{S}( \Sigma )}, \\
&F^1 {\mathcal{S}( \Sigma )} = F^2 {\mathcal{S}( \Sigma )} {\stackrel{\mathrm{def.}}{=}}\ker \epsilon, \\
&F^3 {\mathcal{S}( \Sigma )} {\stackrel{\mathrm{def.}}{=}}\ker \varpi, \\
&F^n {\mathcal{S}( \Sigma )} {\stackrel{\mathrm{def.}}{=}}\ker \epsilon F^{n-2} {\mathcal{S}( \Sigma )} \ \ (\mathrm{for} \ \ 4 \leq n).\end{aligned}$$ By [@Tsujipurebraid] Proposition 5.7. and Proposition 5.11, we have $$\begin{aligned}
&F^m {\mathcal{S}( \Sigma )} F^n {\mathcal{S}( \Sigma )} \subset F^{n+m} {\mathcal{S}( \Sigma )}, \\
&[F^m {\mathcal{S}( \Sigma )}, F^n {\mathcal{S}( \Sigma )}] \subset F^{n+m-2} {\mathcal{S}( \Sigma )}, \end{aligned}$$ for any $n$ and $m$. Let $\rho$ be the ${\mathbb{Q}}$-linear map $$\begin{aligned}
\rho :H \bullet H \to F^2 {\mathcal{S}( \Sigma )}/ F^3 {\mathcal{S}( \Sigma )}, \ \ [a] \bullet [b] \mapsto
{\langle (a-1)(b-1) \rangle},\end{aligned}$$ where $H \bullet H$ is the symmetric tensor of $H$. We remark $\rho$ is a ${\mathbb{Q}}$-module incjection by [@Tsujipurebraid] Theorem 4.1.
Completion and Dehn twist
-------------------------
We denote $$\begin{aligned}
&\widehat{{\mathcal{S}( \Sigma )}} {\stackrel{\mathrm{def.}}{=}}{\underleftarrow{\lim}_{i \rightarrow \infty}} {\mathcal{S}( \Sigma )}/(\ker \epsilon)^i, \\
&\widehat{{\mathcal{S}( \Sigma,J )}} {\stackrel{\mathrm{def.}}{=}}{\underleftarrow{\lim}_{i \rightarrow \infty}} {\mathcal{S}( \Sigma,J )}/
(\ker \epsilon)^i {\mathcal{S}( \Sigma,J )},\end{aligned}$$ for a finite subset $J \subset \partial \Sigma$. By [@TsujiCSAI] Theorem 5.5, the natural homomorphisms ${\mathcal{S}( \Sigma )} \to \widehat{{\mathcal{S}( \Sigma )}} $ and ${\mathcal{S}( \Sigma,J )} \to \widehat{{\mathcal{S}( \Sigma,J )}}$ is injective.
\[prop\_map\_inj\] Let $J$ be a finite subset of $\partial \Sigma$. If $\Sigma \neq \emptyset$ and $\partial_i \cap J \neq \emptyset$ for each $i \in {\{ 1,2, \cdots,b \}}$, then $\mathcal{M}(\Sigma) \to {\mathrm{Aut}}(\widehat{{\mathcal{S}( \Sigma,J )}})$ is injective, where $\partial_1, \cdots, \partial_b$ are the connected components of $
\partial \Sigma$.
We denote $$\label{equation_L_c}
L(c) {\stackrel{\mathrm{def.}}{=}}\frac{-A+{ A^{-1}}}{4 \log(-A)} ({\mathrm{arccosh}}(-\frac{c}{2}))^2
-(-A+{ A^{-1}})\log (-A)$$ where $c$ is also denoted by the element of ${\mathcal{S}( \Sigma )}$ represented by the knot presented by a simple closed curve $c$.
\[thm\_Dehn\_twist\] Let $c$ be a simple closed curve and $t_c$ the Dehn twist along $c$. Then we have $$t_c(\cdot) = \exp (\sigma(L(c))) {\stackrel{\mathrm{def.}}{=}}\sum_{i=0}^\infty \frac{1}{i!} (\sigma(L(c)))^i
\in {\mathrm{Aut}}(\widehat{{\mathcal{S}( \Sigma,J )}}).$$ for any finite subset $J \subset \partial \Sigma$.
The Baker-Campbell-Hausdorff series {#subsection_bch}
-----------------------------------
In this subsection, we will explain the Baker-Campbell-Hausdorff series. We choose $S \subset \widehat{{\mathcal{S}( \Sigma )}}$ such that, for any $i \in {\mathbb{Z}}_{\geq 0}$, there exists $j_i\in {\mathbb{Z}}_{\geq 0}$ satisfying $$\label{equation_jouken_bch}
\sigma(a_{1})\circ
\sigma(a_{2}) \circ \cdots \sigma ( a_{j_i})(F^i
{\mathcal{S}( \Sigma )})\subset F^{i+1} {\mathcal{S}( \Sigma )}$$ for $a_1, a_2, \cdots, a_{j_i} \in
S$. In this paper, the Baker-Campbell-Hausdorff series ${\mathrm{bch}}$ is defined by $$\begin{aligned}
&{\mathrm{bch}}(\epsilon_1 a_1, \epsilon_2 a_2, \cdots,\epsilon_m a_m) \\
&{\stackrel{\mathrm{def.}}{=}}(-A+{ A^{-1}}) \log (\exp(\frac{\epsilon_1 a_1}{-A+{ A^{-1}}})
\exp(\frac{\epsilon_2 a_2}{-A+{ A^{-1}}}) \cdots
\exp(\frac{\epsilon_m a_m}{-A+{ A^{-1}}}))\end{aligned}$$ for $a_1, a_2, \cdots, a_{m} \in
S$ and $\epsilon_1, \epsilon_2, \cdots, \epsilon_m$. We remark, as elements of the associated Lie algebra $(\widehat{{\mathcal{S}( \Sigma )}},[ \ \ , \ \ ])$, it has a usual expression. For example, $${\mathrm{bch}}(x,y) = x+y+\frac{1}{2}[x,y]+\frac{1}{12}([x,[x,y]]+[y,[y,x]])+ \cdots.$$ We denote $${\lvert S \rvert} {\stackrel{\mathrm{def.}}{=}}{\{ {\mathrm{bch}}(\epsilon_1 a_1, \cdots, \epsilon_m
a_m)|m \in {\mathbb{Z}}_{\geq 1}, a_1, \cdots, a_{m}
\in S, \epsilon_1, \cdots, \epsilon_m {\{ \pm1 \}} \}}.$$ For $a_1, a_2, \cdots, a_{j_1} \in {\lvert S \rvert}$, they satisfies the equation (\[equation\_jouken\_bch\]). The Baker-Campbell-Hausdorff series satisfies $$\begin{aligned}
& {\mathrm{bch}}(a,-a) =0, \\
& {\mathrm{bch}}(0,a)={\mathrm{bch}}(a,0) =a, \\
& {\mathrm{bch}}(a,{\mathrm{bch}}(b,c))={\mathrm{bch}}({\mathrm{bch}}(a,b),c), \\
& {\mathrm{bch}}(a, b, -a) =\exp(\sigma(a))(b).\end{aligned}$$ Hence $({\lvert S \rvert}, {\mathrm{bch}}) $ is a group whose identity is $0$. Furthermore, if there exists $j' \in {\mathbb{Z}}_{>0}$ satisfying $$\label{equation_bch_jouken_aut}
\sigma(a_{1})\circ
\sigma(a_{2}) \circ \cdots \sigma (a_{j'})(
{\mathcal{S}( \Sigma,J )})\subset \ker{\epsilon} {\mathcal{S}( \Sigma,J )}$$ for $a_1, a_2, \cdots, a_{j'} \in
S$ and any finite set $J \subset \partial{\Sigma}$, $\exp :({\lvert S \rvert},{\mathrm{bch}}) \to {\mathrm{Aut}}(\widehat{{\mathcal{S}( \Sigma,J )}})$ is a group homomorphism, i.e, $\exp(\sigma({\mathrm{bch}}(a,b))) =\exp(\sigma(a)) \circ \exp
(\sigma(b))$ for $a,b \in {\lvert S \rvert}$. For example, $F^3 \widehat{{\mathcal{S}( \Sigma )}}$ satisfies the condition. Furthermore, if the genus of $\Sigma$ is $0$, $\widehat{{\mathcal{S}( \Sigma )}}$ satisfies the condition and there exists an embedding $\mathcal{M} (\Sigma) \hookrightarrow
(\widehat{{\mathcal{S}( \Sigma )}}, {\mathrm{bch}}).$ For details, see [@Tsujipurebraid].
\[lemm\_bch\_jouken\] Let $V_1$ and $V_2 \subset V_1$ be ${\mathbb{Q}}$-linear subspaces of $H$ satisfying $\mu (v,v') =0$ for any $v \in V_2 $ and $v' \in V_1$. We denote $$S = {\{ x \in \widehat{{\mathcal{S}( \Sigma )}}|
\rho^{-1} (x \mathrm{ \ mod} F^3 \widehat{{\mathcal{S}( \Sigma )}}) \in
V_1 \bullet V_2 \}}.$$ Then $S$ satisfies the above conditions (\[equation\_jouken\_bch\]) and (\[equation\_bch\_jouken\_aut\]).
Torelli groups and Johonson homomorphisms {#section_Torelli}
=========================================
Through this section, let $\Sigma$ be a compact surface with a non-empty connected boundary and genus $g>1$.
The completed Kauffman bracket $\widehat{{\mathcal{S}( \Sigma )}}$ has a filtration ${\{ F^n \widehat{{\mathcal{S}( \Sigma )}} \}_{n \geq 0}}$ such that ${\mathcal{S}( \Sigma )}/F^n {\mathcal{S}( \Sigma )}\simeq \widehat{
{\mathcal{S}( \Sigma )}}/F^n \widehat{{\mathcal{S}( \Sigma )}}$ for $n \in {\mathbb{Z}}_{\geq 0}$.
The definition of $I {\mathcal{S}( \Sigma )}$ {#subsection_definition_I}
---------------------------------------------
The aim of this subsection is to define $I {\mathcal{S}( \Sigma )}$.
Let ${\{ c_1, c_2 \}}$ be a pair of curves whose algebraic intersection number is $0$. Then the set ${\{ L(c_1),L(c_2) \}}$ satisfies the conditions (\[equation\_jouken\_bch\]) and (\[equation\_bch\_jouken\_aut\]).
Let $\gamma_1$ and $\gamma_2$ be elements of $H_1$ such that $\pm [c_1] =\gamma_1$ and $\pm [c_2] =\gamma_2$. We remark that $$L(c_i) =-\frac{1}{2}(c_i+2) =-\frac{1}{4}\rho(\gamma_i \bullet \gamma_i) \mod F^3
\widehat{{\mathcal{S}( \Sigma )}}$$ for $i =1,2$. Since $\mu(\gamma_1,\gamma_2)=0$ and Lemma \[lemm\_bch\_jouken\], we have the claim of the Lemma. This finishes the proof.
By this lemma, we can define $C(c_1,c_2) ={\mathrm{bch}}(L(c_1),L(c_2),-L(c_1),-L(c_2))$ for a pair ${\{ c_1, c_2 \}}$ of curves whose algebraic intersection number is $0$. We denote by $\mathcal{L}_{\mathrm{comm}}(\Sigma)$ the set of all $C(c_1,c_2) $. Since $[L(c_1),L(c_2)] \in F^3 \widehat{{\mathcal{S}( \Sigma )}}$ for a pair ${\{ c_1,c_2 \}}$ of curves whose algebraic intersection number is $0$, we have $\mathcal{L}_{\mathrm{comm}} (\Sigma) \subset F^3
\widehat{{\mathcal{S}( \Sigma )}}$.
\[lemm\_bounding\_pair\] Let ${\{ c_1,c_2 \}}$ be a pair of non-isotopic disjoint homologous curves which is presented by $(r[a_1,b_1]\cdots[a_m,b_m])_\square$ and $(r)_\square$, respectively, for $r, a_1, b_1, \cdots, a_m, b_m \in \pi_1(\Sigma)$. Then we have $L(c_1)-L(c_2) =-\sum_{i=1}^m \lambda ([r]\wedge
[a_i]\wedge [b_i]) \mod F^4 \widehat{
{\mathcal{S}( \Sigma )} } = ( \ker \epsilon)^2$.
By the equation (\[equation\_calc\_4\]), we have $$\begin{aligned}
&L(c_1)-L(c_2)=-\frac{1}{2}(c_1-c_2) =-\frac{1}{2}{\langle r \prod_{i=1}^m [a_i,b_i]-r \rangle} \\
&=-\frac{1}{2}\sum_{j=1}^m {\langle r\prod_{i=1}^j[a_i,b_i]-r\prod_{i=1}^{j-1}[a_i,b_i] \rangle}
=-\sum_{j=1}^m{\langle (r\prod_{i=1}^{j-1}[a_i,b_i]-1)(a_j-1)(b_j-1) \rangle} \\
&=-\sum_{i=1}^m \lambda ([r]\wedge[a_i]\wedge [b_i]). \mod (\ker \epsilon)^2\\\end{aligned}$$ This finishes the proof.
For a separating simple closed curve $c$, we have $L(c) =0 \mod (\ker \epsilon)^2.$
We denote by $\mathcal{L}_{\mathrm{bp}}(\Sigma)$ and $\mathcal{L}_{\mathrm{sep}}(\Sigma)$ the set of all $L(c_1)-L(c_2)$ for a pair ${\{ c_1,c_2 \}}$ of non-isotopic disjoint homologous curves, i.e. which is a bounding pair (BP) and the set of all $L(c)$ for a separating simple closed curve $c$, i.e. for separating simple closed curve (sep. s.c.c.) $c$, respectively. By the above lemma and corollary, we have $\mathcal{L}_{\mathrm{bp}}(\Sigma) \cup
\mathcal{L}_{\mathrm{sep}}(\Sigma) \subset F^3 \widehat{{\mathcal{S}( \Sigma )}}$. We define $$I {\mathcal{S}( \Sigma )} {\stackrel{\mathrm{def.}}{=}}{\lvert \mathcal{L}_{\mathrm{gen}}(\Sigma) \rvert}$$ where we denote $\mathcal{L}_{\mathrm{gen}}(\Sigma) {\stackrel{\mathrm{def.}}{=}}\mathcal{L}_{\mathrm{comm}} (\Sigma)
\cup \mathcal{L}_{\mathrm{bp}}(\Sigma) \cup
\mathcal{L}_{\mathrm{sep}}(\Sigma)$.
We denote by $$\begin{aligned}
&\mathcal{I}_{\mathrm{comm}}(\Sigma) {\stackrel{\mathrm{def.}}{=}}{\{ C_{c_1c_2} {\stackrel{\mathrm{def.}}{=}}[t_{c_1}, t_{c_2}]|\mu(c_1,c_2)=0 \}}, \\
&\mathcal{I}_{\mathrm{sep}}(\Sigma) {\stackrel{\mathrm{def.}}{=}}{\{ t_{c_1c_2} {\stackrel{\mathrm{def.}}{=}}t_{c_1}{t_{c_2}}^{-1}|{\{ c_1,c_2 \}}: \mathrm{BP} \}}, \\
&\mathcal{I}_{\mathrm{sep}}(\Sigma) {\stackrel{\mathrm{def.}}{=}}{\{ t_{c}|c: \mathrm{sep. \ \ s.c.c.} \}}.\end{aligned}$$ The set $\mathcal{I}_{\mathrm{gen}}(\Sigma) {\stackrel{\mathrm{def.}}{=}}\mathcal{I}_{\mathrm{comm}} (\Sigma)
\cup \mathcal{I}_{\mathrm{bp}}(\Sigma) \cup
\mathcal{I}_{\mathrm{sep}}(\Sigma)$ is the generator set of the infinite presentation of $\mathcal{I} (\Sigma)$ in [@Pu2008].
The map $\theta :\mathcal{L}_{\mathrm{gen}}(\Sigma) \to \mathcal{I}_{\mathrm{gen}}(\Sigma)$ is defined by the following.
- For a pair ${\{ c_1,c_2 \}}$ of curves whose algebraic intersection number is $0$, $C(c_1,c_2) \mapsto C_{c_1c_2}$.
- For a pair ${\{ c_1,c_2 \}}$ of non-isotopic disjoint homologous curves, $L(c_1)-L(c_2) \mapsto t_{c_1c_2}$.
- For a separating simple closed curve $c$, $\theta (L(c)) =t_c$.
In subsection \[subsection\_injectivity\_of\_zeta\] Proposition \[prop\_theta\], we prove $\theta$ is well-defined.
Well-definedness of $\theta$ {#subsection_injectivity_of_zeta}
----------------------------
The aim of this subsection is to prove the following.
\[lemm\_zeta\_inv\_inj\] The map $\theta:\mathcal{L}_{\mathrm{gen}}(\Sigma) \to
\mathcal{I}_{\mathrm{gen}}(\Sigma)$ induces $\theta:I {\mathcal{S}( \Sigma )} \to \mathcal{I}(\Sigma)$.
\[prop\_theta\] The map $\theta$ is well-defined.
By Theorem \[thm\_Dehn\_twist\], we have the followings.
- For a pair ${\{ c_1,c_2 \}}$ of curves whose algebraic intersection number is $0$, $\exp(\sigma(C(c_1,c_2)))(\cdot)= C_{c_1c_2}(\cdot)\in {\mathrm{Aut}}(\widehat{{\mathcal{S}( \Sigma,J )}})$ for any finite set $J \in \partial{\Sigma}$.
- For a pair ${\{ c_1,c_2 \}}$ of non-isotopic disjoint homologous curves, $\exp(\sigma(L(c_1)-L(c_2))) (\cdot)= t_{c_1c_2}(\cdot)\in {\mathrm{Aut}}(\widehat{{\mathcal{S}( \Sigma,J )}})$ for any finite set $J \in \partial{\Sigma}$.
- For a separating simple closed curve $c$, $\exp(\sigma(L(c)))(\cdot) =t_c(\cdot) \in {\mathrm{Aut}}(\widehat{{\mathcal{S}( \Sigma,J )}})$ for any finite set $J \in \partial{\Sigma}$.
By Proposition \[prop\_map\_inj\], $\theta$ is well-defined. This finishes the proof.
By Theorem \[thm\_Dehn\_twist\], we have the following.
\[lemm\_theta\] For $x_1,x_2, \cdots, x_j \in {\{ \epsilon x|\epsilon \in {\{ \pm1 \}},
x \in \mathcal{L}_{\mathrm{gen}}(\Sigma) \}}$, we have $$\prod_{i=1}^j(\theta ( x_i)) (\cdot) =
\exp({\mathrm{bch}}( x_1, x_2, \cdots, x_j))(\cdot) \in {\mathrm{Aut}}(\widehat{{\mathcal{S}( \Sigma,J )}})$$ for any finite set $J \subset \partial \Sigma$.
For $x_1,x_2, \cdots, x_j ,
\in {\{ \epsilon x|\epsilon \in {\{ \pm1 \}},
x \in \mathcal{L}_{\mathrm{gem}} (\Sigma) \}}$, we assume $${\mathrm{bch}}(x_1, \cdots, x_j) = 0.$$ By Lemma \[lemm\_theta\], we have $$\prod_{i=1}^j\theta (x_i) (\cdot) =\exp (\sigma ({\mathrm{bch}}(x_1, \cdots, x_j)))( \cdot) ={\mathrm{id}}(\cdot)
\in {\mathrm{Aut}}(\widehat{{\mathcal{S}( \Sigma,J )}})$$ for any finite set $J \subset \partial \Sigma$. By Proposition \[prop\_map\_inj\], we have $
\prod_{i=1}^j\theta (x_i) ={\mathrm{id}}.
$ This finishes the proof.
Well-definedness of $\zeta$ {#subsection_well_defined_zeta}
---------------------------
In this section, we prove $\theta$ is injective. In order to check Putman’s relation [@Pu2008], we need the following.
\[lemm\_relation\_key\] Let $\Sigma$ be a compact surface with non-empty boundary. We choose an element $x \in \widehat{{\mathcal{S}( \Sigma )}}$ satisfying $\sigma (x)(\widehat{{\mathcal{S}( \Sigma,J )}}) =
{\{ 0 \}}$ for any finite set $J \subset \partial \Sigma$. Then we have followings.
1. For any embedding $e' :\Sigma \to \tilde{\Sigma}$ and any element $\xi \in
\mathcal{M}(\tilde{\Sigma})$, we have $\xi (e'(x)) =e'(x) \in
\widehat{{\mathcal{S}( \tilde{\Sigma} )}}$ where we also denote by $e'$ the homomorphism $\widehat{{\mathcal{S}( \Sigma )}}
\to \widehat{{\mathcal{S}( \tilde{\Sigma} )}}$ induced by $e'$.
2. For any embedding $e'':\Sigma \times I \to I^3 \to \Sigma \times I$, we have $x =e''(x) \in {\mathbb{Q}}[[A+1]] [\emptyset] \subset \widehat{{\mathcal{S}( \Sigma )}}$ where we also denote by $e''$ the homomorphism $\widehat{{\mathcal{S}( \Sigma )}}
\to \widehat{{\mathcal{S}( \Sigma )}}$ induced by $e''$.
Since $\mathcal{M} (\tilde{\Sigma})$ is generated by Dehn twists, it is enough to check $t_c (x) =x $ for any simple closed curve $c$. By assumption of $x$ and Theorem \[thm\_Dehn\_twist\], we have $$\begin{aligned}
t_c(x) =\exp(\sigma(L(c))(x) =x.\end{aligned}$$ This finishes the proof.
It is enough to show there exists an embedding $e'':\Sigma \times I \to I^3 \to \Sigma \times I$ such that $x = e''(x)$. We choose compact connected surfaces $\tilde{\Sigma}$ with non-empty connected boundary and an embedding $e''': \tilde{\Sigma} \times I \to \Sigma \times I$ such that there exists submanifolds $\Sigma'$, $\Sigma'' \subset \tilde{\Sigma}$ satisfying the followings.
- The intersection $\Sigma' \cap \Sigma''$ is empty.
- There exists diffeomorphisms $\chi' : \Sigma \to \Sigma'$, $\chi'' : \Sigma \to \Sigma''$.
- There exists en embedding $e_{I^3}: I^3 \to \Sigma \times I$ satisfying $$e'''\circ (\chi'' \times {\mathrm{id}}_I) (\Sigma \times I) \subset e''(I^3).$$
- The embedding $e''' \circ (\chi' \times {\mathrm{id}}_I)$ induces the identity map of $\widehat{{\mathcal{S}( \Sigma )}}$.
(300,540) (0,240)[![$e''': \tilde{\Sigma}\times I \to \Sigma \times I$[]{data-label="figure_embedding_Sigma_tilde"}](surface_zeta_4_PNG.pdf "fig:"){width="310pt"}]{} (0,60)[![$e''': \tilde{\Sigma}\times I \to \Sigma \times I$[]{data-label="figure_embedding_Sigma_tilde"}](surface_zeta_7_PNG.pdf "fig:"){width="310pt"}]{} (0,-110)[![$e''': \tilde{\Sigma}\times I \to \Sigma \times I$[]{data-label="figure_embedding_Sigma_tilde"}](surface_zeta_6_PNG.pdf "fig:"){width="310pt"}]{} (90,370)[[$\Sigma$]{}]{} (120,205)[$\tilde{\Sigma}$]{} (240,90)[[$\Sigma'$]{}]{} (90,35)[$\Sigma''$]{}
For example, see Figure \[figure\_embedding\_Sigma\_tilde\]. By (1), we have $\chi'(x) =\chi''(x)$. Hence, we obtain $x =e''' \circ (\chi' \times {\mathrm{id}}_I) (x) =e''' \circ
(\chi'' \times {\mathrm{id}}_I)(x)$. The embedding $e''' \circ
(\chi'' \times {\mathrm{id}}_I)$ satisfies the claim $e''$. This finishes the proof.
To check the relations (F.1), (F.2), ... (F.7) and (F.8) in [@Pu2008], we need the following. We obtain the lemma by a straightforward calculation and definition of the Baker-Campbell-Hausdorff series. Lemma \[lemm\_relation\_F\] (k) corresponds (F.k) for $k=1,2, \cdots ,8$.
\[lemm\_relation\_F\]
1. For a pair ${\{ c_1,c_2 \}}$ of non-isotopic disjoint homologous curves, we have $$L(c_1)-L(c_2) =-(L(c_2)-L(c_1)).$$
2. For a pair ${\{ c_1,c_2 \}}$ of curves whose algebraic intersection number is $0$, we have $${\mathrm{bch}}({\mathrm{bch}}(L(c_1),L(c_2),-L(c_1),-L(c_2)),{\mathrm{bch}}(L(c_2),L(c_1),-L(c_2),-L(c_1)))=0.$$
3. For a triple ${\{ c_1,c_2,c_3 \}}$ of non-isotopic disjoint homologous curves, we have $${\mathrm{bch}}(L(c_1)-L(c_2), L(c_2)-L(c_3)) = L(c_1)-L(c_3).$$
4. If ${\{ c_1,c_2 \}}$ is a pair of non-isotopic disjoint homologous curves such that $c_1$ and $c_2$ are separating curves, we have $$L(c_1)-L(c_2) = {\mathrm{bch}}(L(c_1),-L(c_2)).$$
5. If ${\{ c_1,c_2 \}}$ is a pair of non-isotopic disjoint homologous curves and ${\{ c_3,c_2 \}}$ is a pair of curves whose algebraic intersection number is $0$ such that $c_1$ and $c_3$ are disjoint, we have $$L(t_{c_3}(c_2)) -L(c_{1}) ={\mathrm{bch}}({\mathrm{bch}}(L(c_3),L(c_2),-L(c_3),-L(c_2)),L(c_2)-L(c_1)).$$
6. For a pair ${\{ c_1,c_2 \}}$ of curves whose algebraic intersection number is $0$ and $x \in \mathcal{L}_{\mathrm{gen}}(\Sigma)$, we have $${\mathrm{bch}}(x,C(c_1,c_2),-x) =C(\theta(x)(c_1),\theta(x)(c_2)).$$
7. For a pair ${\{ c_1,c_2 \}}$ of non-isotopic disjoint homologous curves and $x \in \mathcal{L}_{\mathrm{gen}}(\Sigma)$, we have $${\mathrm{bch}}(x,L(c_1)-L(c_2),-x) =L(\theta(x)(c_1))-L(\theta(x)(c_2)).$$
8. For a separating simple closed curve $c$ and $x \in \mathcal{L}_{\mathrm{gen}}(\Sigma)$, we have $${\mathrm{bch}}(x,L(c),-x) =L(\theta(x)(c))$$
In order to check lantern relation in [@Pu2008], we need the following.
(300,450) (0,300)[![The simple closed curves in $\Sigma_{04}$[]{data-label="figure_Sigma_0_4"}](Sigma_0_4_1_2_3_PNG.pdf "fig:"){width="310pt"}]{} (0,150)[![The simple closed curves in $\Sigma_{04}$[]{data-label="figure_Sigma_0_4"}](Sigma_0_4_12_23_PNG.pdf "fig:"){width="310pt"}]{} (0,0)[![The simple closed curves in $\Sigma_{04}$[]{data-label="figure_Sigma_0_4"}](Sigma_0_4_13_123_PNG.pdf "fig:"){width="310pt"}]{} (70,337)[$c_1$]{} (150,337)[$c_2$]{} (229,337)[$c_3$]{} (70,187)[$c_{12}$]{} (229,187)[$c_{23}$]{} (150,35)[$c_{123}$]{} (150,110)[$c_{13}$]{}
Let $\Sigma_{0,4}$ be a compact connected oriented surface of genus $0$ and $4$ boundary components and $c_1,c_2,c_3,c_{12},c_{23},c_{13}$ and $c_{123}$ the simple closed curves as Figure \[figure\_Sigma\_0\_4\]. We have ${\mathrm{bch}}(L(c_{123},-L(c_{12}),-L(c_{23}),-L(c_{13}),L(c_1),L(c_2),L(c_3)))=0\in
\widehat{{\mathcal{S}( \Sigma_{0,4} )}}$.
Let $e$ be an embedding $\Sigma_{0,4} \times I \to I^3$ as Figure \[figure\_Sigma\_0\_4\]. Since $$\begin{aligned}
&\exp ({\mathrm{bch}}(L(c_{123},-L(c_{12}),-L(c_{23}),-L(c_{13}),L(c_1),L(c_2),L(c_3))))(\cdot)
= \\
&t_{c_{123}}{t_{c_{12}}}^{-1}{t_{c_{23}}}^{-1}{t_{c_{31}}}^{-1}t_{c_{1}}t_{c_{2}}t_{c_3}(\cdot) = {\mathrm{id}}(\cdot)
\in {\mathrm{Aut}}(\widehat{{\mathcal{S}( \Sigma_{0.4},J )}})\end{aligned}$$ for all finite set $J \subset \partial \Sigma$, we have $$\sigma({\mathrm{bch}}(L(c_{123},-L(c_{12}),-L(c_{23}),-L(c_{13}),L(c_1),L(c_2),L(c_3))))
(\widehat{{\mathcal{S}( \Sigma_{0,4},J )}}) = {\{ 0 \}}.$$ By Lemma \[lemm\_relation\_key\], it is enough to show $e({\mathrm{bch}}(L(c_{123},-L(c_{12}),-L(c_{23}),-L(c_{13}),L(c_1),L(c_2),L(c_3))))
=0$. We remark $e([\widehat{{\mathcal{S}( \Sigma_{0,4} )}}, \widehat{{\mathcal{S}( \Sigma_{0,4} )}}])
={\{ 0 \}}$. We have $$\begin{aligned}
&e({\mathrm{bch}}(L(c_{123},-L(c_{12}),-L(c_{23}),-L(c_{13}),L(c_1),L(c_2),L(c_3)))) \\
&=(1-3+3)(\frac{-A+{ A^{-1}}}{4\log(-A)} ({\mathrm{arccosh}}(\frac{A^2+A^{-2}}{2}))^2-
(-A+{ A^{-1}})\log (-A))=0.\end{aligned}$$ This finishes the proof.
We check the crossed lantern relation in [@Pu2008]. Let $\Sigma_{1,2}$ be a connected compact surface of genus $g=1$ with two boundary components. We denote simple closed curves in $\Sigma_{1,2}$ as in Figure \[figure\_CL\]. We have $H_1(\Sigma_{1,2},{\mathbb{Q}}) ={\mathbb{Q}}a\oplus {\mathbb{Q}}b \oplus {\mathbb{Q}}v$ where $a {\stackrel{\mathrm{def.}}{=}}[c'_a], b {\stackrel{\mathrm{def.}}{=}}[c_b], v {\stackrel{\mathrm{def.}}{=}}[c_v] \in H_1 (\Sigma_{1,2},{\mathbb{Q}})$. In order to check the crossed lantern relation, we need the following.
(300,140) (0,-160)[![$\Sigma_{1,2}$[]{data-label="figure_CL"}](surface_CL_PNG.pdf "fig:"){width="340pt"}]{} (47,70)[$c_b$]{} (120,86)[$c'_b$]{} (80,115)[$c'_a$]{} (80,10)[$c_a$]{} (190,50)[$c_v$]{} (240,30)[$c'_{ab}$]{} (247,115)[$c_{ab}$]{}
1. The set $ {\{ \pm(L(c_a)-L(c'_a))
,\pm (L(c_b)-L(c'_b)), \pm (L(c_{ab})-L(c'_{ab})) \}}$ satisfies the conditions (\[equation\_jouken\_bch\]) and (\[equation\_bch\_jouken\_aut\]).
2. We have ${\mathrm{bch}}(L(c_b)-L(c'_b),-L(c_a)+L(c'_a),-L(c_{ab})+L(c'_{ab}))=0$.
We have $$\begin{aligned}
L(c_a)-L(c'_a) &=\frac{1}{4}\rho(-(a-v)\cdot(a-v)+a\cdot a ) \mod F^3
\widehat{{\mathcal{S}( \Sigma_{0,4} )}} \\
&=\frac{1}{4}\rho(-v\cdot v+2v \cdot a ) \mod F^3
\widehat{{\mathcal{S}( \Sigma_{0,4} )}}, \\
L(c_b)-L(c'_b)&=\frac{1}{4}\rho(-b\cdot b+(b-v)\cdot (b-v)) \mod F^3
\widehat{{\mathcal{S}( \Sigma_{0,4} )}} \\
&=\frac{1}{4}\rho(v\cdot v-2v \cdot b ) \mod F^3
\widehat{{\mathcal{S}( \Sigma_{0,4} )}}, \\
L(c_{ab})-L(c'_{ab}) &=\frac{1}{4}\rho(-(a+b) \cdot (a+b) +(a+b-v)\cdot(a+b-v)) \mod F^3
\widehat{{\mathcal{S}( \Sigma_{0,4} )}} \\
&=\frac{1}{4}\rho(v\cdot v-2v \cdot (a+b) ) \mod F^3
\widehat{{\mathcal{S}( \Sigma_{0,4} )}}. \\\end{aligned}$$ By Lemma \[lemm\_bch\_jouken\], this finishes the proof.
Let $e$ be an embedding $\Sigma_{0,4} \times I \to I^3$ as Figure \[figure\_CL\]. Since $$\begin{aligned}
&\exp (\sigma({\mathrm{bch}}(L(c_b)-L(c'_b),-L(c_a)+L(c'_a),-L(c_{ab})+L(c'_{ab}))))(\cdot) \\
&=t_{c_b c'_b} \circ t_{c'_a c_a}\circ {t_{c_{ab} c'_{ab}}}^{-1}(\cdot) = {\mathrm{id}}(\cdot)
\in {\mathrm{Aut}}(\widehat{{\mathcal{S}( \Sigma_{1,2},J )}})\end{aligned}$$ for all finite set $J \subset \partial \Sigma$, we have $$\sigma({\mathrm{bch}}(L(c_b)-L(c'_b),-L(c_a)+L(c'_a),-L(c_{ab})+L(c'_{ab})))
(\widehat{{\mathcal{S}( \Sigma_{1,2},J )}}) = {\{ 0 \}}.$$ By Lemma \[lemm\_relation\_key\], it is enough to show $e({\mathrm{bch}}(L(c_b)-L(c'_b),-L(c_a)+L(c'_a),-L(c_{ab})+L(c'_{ab})))=0$. Since $e(x (L(c_i)-L(c'_i))) =e((L(c_i)-L(c_i)) x) =0 $ for $i \in {\{ a, b,ab \}}$ and any $x \in \widehat{{\mathcal{S}( \Sigma_{1,2} )}}$, we have $e({\mathrm{bch}}(L(c_b)-L(c'_b),-L(c_a)+L(c'_a),-L(c_{ab})+L(c'_{ab})))=0$. This finishes the proof.
In order to check Witt-Hall relation and commutator shuffle relation in [@Pu2008], it is enough to show the following lemma.
Let $\Sigma'$ be a compact surface, $D$ a open disk in $\Sigma'$ and $\Sigma''$ the surface $\Sigma' \backslash D$. We fix the points $*_1$ and $*_2$ and the paths $\gamma_1$ and $\gamma_2$ as in Figure \[figure\_WH\_CS\]. We denote by $\mathcal{S}_{\mathrm{push}}(\Sigma'', \partial D)$ the set of all pair ${\{ c_1,c_2 \}}$ of simple closed curves satisfying the followings.
- There exists a path $\gamma$ from $*_1$ to $*_2$ such that $c_1 =\gamma_1 \cup \gamma$ and $c_2 =\gamma_2 \cup \gamma$.
- We have $[c_1] =[c_2] \in H_1 (\Sigma').$
For ${\{ c_{11},c_{21} \}} \cdots {\{ c_{1k},c_{2k} \}} \in
\mathcal{S}_{\mathrm{push}}(\Sigma'', \partial D)$ and $\epsilon_1 \cdots \epsilon_k \in {\{ \pm1 \}}$, if $ \prod_{i=1}^k (t_{c_{1i}c_{2i}})^{\epsilon_i} = {\mathrm{id}}\in \mathcal{I}(\Sigma'')$, then we have $${\mathrm{bch}}(\epsilon_1 (L(c_{11})-L(c_{21})), \cdots, \epsilon_k(L(c_{1k})-L(c_{2k}))) =0.$$
(300,140) (0,-160)[![$\Sigma'$[]{data-label="figure_WH_CS"}](surface_WH_CS_PNG.pdf "fig:"){width="340pt"}]{} (60,90)[$D$]{} (100,86)[$*_1$]{} (20,86)[$*_2$]{} (90,100)[$\gamma_1$]{} (55,70)[$\gamma_2$]{}
Since $$\exp (\sigma({\mathrm{bch}}(\epsilon_1 (L(c_{11})-L(c_{21})), \cdots, \epsilon_k(L(c_{1k})-L(c_{2k})))
))(\cdot)
=\prod_{i=1}^k (t_{c_{1i}c_{2i}})^{\epsilon_i}(\cdot) = {\mathrm{id}}(\cdot)
\in {\mathrm{Aut}}(\widehat{{\mathcal{S}( \Sigma,J )}})$$ for all finite set $J \subset \partial \Sigma$, we have $$\sigma({\mathrm{bch}}(\epsilon_1 (L(c_{11})-L(c_{21})), \cdots, \epsilon_k(L(c_{1k})-L(c_{2k}))))
(\widehat{{\mathcal{S}( \Sigma,J )}}) = {\{ 0 \}}.$$ We denote by $e'$ the embedding $\Sigma'' \to \Sigma'$. We choose an embedding $e:\Sigma' \times I \to I^3$. By Lemma \[lemm\_relation\_key\], it is enough to show $e\circ e'({\mathrm{bch}}(\epsilon_1 (L(c_{11})-L(c_{21})), \cdots,
\epsilon_k((c_{1k})-L(c_{2k}))))=0$. Since $e'(L(c_{1i})-L(c_{2i}))$ for $i \in {\{ 1,2, \cdots,k \}}$, we have $e'({\mathrm{bch}}(\epsilon_1 (L(c_{11})-L(c_{21})), \cdots,
\epsilon_k(L(c_{1k})-L(c_{2k}))))=0$. This finishes the proof.
By above lemmas, we have the main theorem.
The group homomorphism $\theta :
I {\mathcal{S}( \Sigma )} \to \mathcal{I}(\Sigma)$ is an isomorphism. In other words, $\zeta {\stackrel{\mathrm{def.}}{=}}\theta^{-1} :
\mathcal{I} (\Sigma) \to \widehat{{\mathcal{S}( \Sigma )}}$ is embedding.
Since $\mathcal{I} (\Sigma) $ is generated by ${\{ t_{c_1c_2}|c_1,c_2: \mathrm{BP} \}}$, we have the following.
We have $I {\mathcal{S}( \Sigma )} ={\lvert \mathcal{L}_{bp}(\Sigma) \rvert}$. Furthermore $\zeta$ can be defined by $\zeta(t_{c_1c_2})=L(c_1)-L(c_2)$.
The 1st Johnson homomorphism and $\zeta$ {#subsection_johnson_zeta}
----------------------------------------
In this subsection, we prove the following.
\[thm\_johnson\_zeta\] The isomorphism $\zeta :\mathcal{I} (\Sigma) \to I {\mathcal{S}( \Sigma )}$ induces $\tau_\zeta :\mathcal{I} (\Sigma) \to
F^3 \widehat{{\mathcal{S}( \Sigma )}}/
F^4 \widehat{{\mathcal{S}( \Sigma )}} \stackrel{\lambda^{-1}}{\simeq}
H_1 \wedge H_1 \wedge H_1$. Then we have $\tau_\zeta =\tau$ where $\tau$ is the 1st Johnson homomorphism.
By [@Johnson80] Lemma 4B, it is enough to show $\tau_\zeta (t_{c_1c_2}) =\sum_{i=1}^k
[a_i] \wedge [b_i] \wedge [b_{k+1}]$ where $c_1$ and $c_2$ are disjoint simple closed curves which is presented by $(b_{k+1})_\square$ and $( \prod_{i=1}^k[a_i,b_i] b_{k+1})_\square$, respectively. By Lemma \[lemm\_bounding\_pair\], we have $L(c_1)-L(c_2) =\lambda (\sum_{i=1}^k
[a_i] \wedge [b_i] \wedge [b_{k+1}]) \mod F^4 \widehat{{\mathcal{S}( \Sigma )}}$. This finishes the proof.
Since $\mathcal{K} (\Sigma) {\stackrel{\mathrm{def.}}{=}}\ker \tau$ is generated by ${\{ t_c|c:\mathrm{sep. s.c.c.} \}}$, we have the following.
The subgroup $F^4 \widehat{{\mathcal{S}( \Sigma )}} \cap I{\mathcal{S}( \Sigma )}$ is generated by $\mathcal{L}_{\mathrm{sep}}(\Sigma)$. Furthermore, the restriction of $\zeta$ to $\mathcal{K} (\Sigma)$ $\zeta_{|\mathcal{K}(\Sigma)}:\mathcal{K}(\Sigma) \to
F^4 \widehat{{\mathcal{S}( \Sigma )}} \cap I{\mathcal{S}( \Sigma )}$ is an isomorphism defined by $\zeta (t_c) =L(c)$ for a separating simple closed curve $c$.
[99]{} D. Johnson, *An abelian quotient of the mapping class group $\mathcal{I}_g$*, Math. Ann. 249, 225-242(1980). N. Kawazumi and Y. Kuno, *The logarithms of Dehn twists*, Quantum Topology, Vol. 5, Issue 3, pp. 347–423, 2014 N. Kawazumi and Y. Kuno, *Groupoid-theoretical methods in the mapping class groups of surfaces*, arXiv: 1109.6479 (2011), UTMS preprint: 2011–28 G. Massuyeau and V. Turaev, *Fox pairings and generalized Dehn twists*, Ann. Inst. Fourier 63 (2013) 2403-2456. S. Morita, *On the structure of the Torelli group and the Casson invariant*, Topology, Volume 30, Issue 1991, 603-621. G. Muller, *Skein algebra and cluster algebras of marked surfaces*, arXiv: 1104.0020(2012). A. Putman, *An infinite presentation of the Torelli group*, Geom. Funct. Anal. 19 (2009), no. 2, 591-643. S. Tsuji, *Dehn twists on Kauffman bracket skein algebras*, preprint, arXiv:1510.05139(2015). S. Tsuji, *The quotient of a Kauffman bracket skein algebra by the square of an augmentation ideal*, in preparation. S. Tsuji, in preparation. Turaev, V. G., *Skein quantization of Poisson algebras of loops on surfaces*, Ann. Sci. Ecole Norm. Sup. (4) 24 (1991), no. 6
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
Joseph Harrington,$^{1,2}$ Brad M. Hansen,$^{3\ast}$ Statia H. Luszcz,$^{2,4}$ Sara Seager,$^{5}$\
Drake Deming,$^{6}$ Kristen Menou,$^{7}$ James Cho,$^{8}$ L. Jeremy Richardson$^{9}$\
\
\
\
\
,\
\
\
\
\
\
\
\
\
\
\
\
\
title: |
The Phase-Dependent Infra-Red\
Brightness of the Extrasolar\
Planet $\upsilon$ Andromeda b
---
The star $\upsilon$ Andromeda is orbited by three known planets[@But2], the innermost of which has an orbital period of 4.617 days and a mass at least 0.69 that of Jupiter. This planet is close enough to its host star that the radiation it absorbs overwhelms its internal heat losses. Here we present the 24- light curve of this system, obtained with the Spitzer Space Telescope. It shows a clear variation in phase with the orbital motion of the innermost planet. This is the first demonstration that such planets possess distinct hot substellar (day) and cold antistellar (night) faces.
Last year, two independent groups[@Dem1; @Charb1] reported the first measurements of the infrared light emitted by extrasolar planets orbiting close to their parent stars. These ‘hot Jupiter’[@Vulcan] planets have small enough orbits that the energy they absorb from their hosts dominates their own internal energy losses. How they absorb and reradiate this energy is fundamental to understanding the behavior of their atmospheres. One way to address this question is to monitor the emitted flux over the course of an orbit, to see whether the heat is distributed asymmetrically about the surface of the planet.
We have observed the $\upsilon$ Andromeda system with the 24- channel of the Multiband Imaging Photometer for Spitzer (MIPS)[@Rieke] aboard the Spitzer Space Telescope[@Werner]. We took 168 three-second images at each of five epochs spread over 4.46 days (97% of the 4.617-day orbital period of $\upsilon$ Andromeda b) beginning on 18 February 2006 at 12:52 UTC (Coordinated Universal Time). After rejecting frames with bad pixels near the star and those with Spitzer’s ‘first frame effect’[@Dem1] (2% – 8% of the data, depending on epoch), we measure the flux of the system and that of the surrounding sky using both sub-pixel, interpolated aperture photometry and optimal photometry[@Horne; @Rich1] on each frame.
The detection of eclipses[@Sec] from the “hot Jupiter” planetary systems HD 209458b[@Dem1], TrES-1[@Charb1] and HD 189733b[@Dem2] demonstrate that a small fraction ($\sim 0.1\%$) of the total infrared light we observe from these systems is actually emitted from the planet rather than the star. Thus, if we can measure the flux of a system at a signal-to-noise ratio (S/N) $>1000$, temperature differences between the day and night faces of the planet will appear as an orbital modulation of the total system flux. With a star as bright as $\upsilon$ Andromeda, our three-second exposures each have S/N $\sim 500$, so that our SNR expectation is $\sim \sqrt{160} \times 500 \approx 6300$ at each epoch.
The MIPS instrument acquires data by placing the stellar image in a sequence of 14 positions on the detector. The detector’s response varies with position at the $\sim 1\%$ level. This variation is stable and reproducible, so we calculate correction factors as follows: At each epoch we compute the mean measured system flux at each position and take the ratio with the mean in the first position. We then average this ratio over all epochs for each position. This results in corrections $< 2\%$ between positions, with uncertainties $\sim 6\times 10^{-4}$. Bringing the photometry to a common normalization allows us to average over all the frames in each epoch to achieve S/N $\approx 4350$ at each epoch.
As with most infrared instruments, MIPS’s sensitivity varies in time. We correct for such drifts by dividing the system flux value by the measured background in each frame. The background at 24 is thermal emission from the zodiacal dust. This dust pervades the inner solar system, absorbing light from the sun and reradiating it at infrared wavelengths. At 24 , its emission is strong enough for use as a flux standard, a technique used successfully in measuring the eclipse of HD 209458b[@Dem1]. However, the present work requires one additional correction. The zodiacal background is the integrated emission by dust along the line of sight between the telescope and the object. The observed value thus undergoes an annual modulation as that line of sight varies with the telescope’s orbit about the sun. The best available model[@DIRBE] predicts a linear drift over the brief interval of our observations. However, we cannot use the Spitzer model directly, since it is calculated for a line of sight from the Earth to the object in question. The difference in position between the Earth-trailing telescope and the Earth itself is large enough that the slope of the variation may be slightly different. Thus, we fit for the linear drift directly, simultaneously with any model lightcurve fits.
The phase curve for the $\upsilon$ Andromeda system shows a variation (Fig \[Panel\]) in absolute photometry, even before any corrections for instrumental or zodiacal drifts are made. After the calibration with respect to the zodiacal background is applied, this variation is revealed to be in phase with the known orbit of the innermost planet of the system, our principal result. A simple model can be fit to the phase curve (Fig \[Fit\]), assuming local, instantaneous thermal reradiation of the absorbed stellar flux. In the simplest model, the phase of the variation is not a free parameter, but is rather set by the measured radial velocity curve[@But1]. Phase offsets are possible for models in which the energy is absorbed deep within the atmosphere and redistributed about the surface[@Cho; @CS]. There is weak (2.5$\sigma$) evidence for a small phase offset in this data (Fig \[Fit\]) but the large offsets predicted from some models are excluded at high significance. Fitting the peak-to-trough amplitude to the observations yields a best-fit value for the planet-star flux ratio $2.9 \pm 0.7\times 10^{-3}$ times the star’s brightness. This is very similar to the result at this wavelength for HD 209458b[@Dem1]. However, the latter is a measure of the absolute flux from the planet divided by that from the host star, while the present result is a measure of the flux difference between the projected day and night sides, divided by the flux of the (different) host star. Another difference between the cases of $\upsilon$ Andromeda b and HD 209458b is that we do not have a strong constraint on the orbital inclination in this system, so we must include the unknown inclination in the model fit (Fig \[IC\]). At higher inclinations, parts of both the night side and the day side are always visible, so the true contrast between the day and night sides must be larger than the amplitude of the observed variation. This contrast is ultimately driven by the light absorbed from the star, which therefore provides an upper limit. We know the distance of the planet from the star and the stellar properties, so we can estimate the contrast that would result if all of the observed flux were reradiated from the day side and nothing from the night side. If we assume the planet’s radius is $<$1.4 Jupiter radii (as observed for other planets of this class), then we can constrain the expected amplitude to be $<3.4\times 10^{-3}$ (2$\sigma$) for a simple black-body, no-redistribution model with zero albedo. Thus, a consistent picture of the atmospheric energetics emerges as long as the orbital inclination is $>$30.
A natural question to ask is whether there are any plausible alternative models for the observed variation. The estimated rotation period of the star is too long to explain our phase curve as the result of a normal starspot (which is darker than other parts of the stellar surface). One could posit a feature on the stellar surface similar to a starspot but induced by a magnetic interaction between the star and the planet, and therefore moving synchronously with the planet. However, [@Bali] place an upper limit of $1.6 \times 10^{-4}$ on the amplitude of optical variation with the planetary orbital period, so infrared variability from the star should be even weaker than this. Some evidence for such magnetospheric interactions is found in observations of chromospheric calcium H and K lines[@Shkol] and has even been seen in the $\upsilon$ Andromeda system. However, the energy input needed to explain the Ca lines is $\sim10^{27}\rm ergs \, s^{-1}$, much less than the minimum planetary luminosity we infer here ($\sim 4\times 10^{29} \rm ergs\, s^{-1}$). Indeed, one can make a quite general argument that our observations cannot be powered by the same mechanism, since any heating of the star due to magnetic interaction with the planet ultimately extracts energy from the planetary orbit. Thus, one may calculate an orbital decay time $$\tau = \frac{G M_* M_p}{2 a \dot{E}} = 5\tttt{6} {\rm yr} \left( \frac{M_p}{M_J} \right)
\left( \frac{a}{12 R_{\odot}} \right)^{-1} \left( \frac{\dot{E}}{\rm 10^{30} ergs\; s^{-1}} \right)^{-1},$$ where $M_*$ and $M_p$ are the stellar and planetary masses, $a$ is the semi-major axis and $\dot{E}$ is the observed heating rate. Heating at the level necessary to explain our observations would result in the decay of the planetary orbit on timescales $<$ years, while the estimated age of the system is 3 Gyr. As such, the chromospheric heating of the star is unlikely to be related to the effect seen at 24 .
This observation reveals the presence of a temperature asymmetry on the surface of an extrasolar planet. The first measurements of eclipses[@Dem1; @Charb1] yielded measurements of the absolute flux levels emerging from the day sides of two extrasolar planets. When compared with models of radiative transfer in such atmospheres[@Seag; @Bar; @Fort; @Burr], those observations are consistent with a situation intermediate between no redistribution and full redistribution. A similar comparison is possible in this case (Fig \[Spec2\]). Our observed day-night flux difference is comparable to the flux emerging at full phase in the models of [@Seag], which indicates that there is little evidence for redistribution of energy to the night side.
In conclusion, the observation of the phase curve of $\upsilon$ Andromeda b indicates that significant temperature differences exist between the day and night faces of the planet, consistent with a model in which very little horizontal energy transport occurs in the planetary atmosphere. Furthermore, it indicates that the opportunities for direct extrasolar planetary observations are better than previously thought, since useful data can be obtained even in cases where the planetary orbit is not so fortuitously aligned that the system exhibits transits or eclipses.
[99]{} Butler, R. P., et al., [*Ap.J.*]{}, [**526**]{}, 916 (1999) Butler, R. P., et al., [*Ap.J.*]{}, [**474**]{}, L115 (1997) There is no official terminology for planets that orbit close to their parent stars. The term “Hot Jupiter” is the most common, although some authors have adopted other terms, such as “Pegasids”,“Pegasi planets”, or “Roasters”. Deming, D., Seager, S., Richardson, L. J., & Harrington, J., [*Nature*]{}, [**434**]{}, 740 (2005) Charbonneau, D. et al., [*Ap.J.*]{}, [**626**]{}, 523 (2005) Rieke, G. H., et al., in [*Proc. SPIE. Optical, Infrared, and Millimeter Space Telescopes*]{} (ed. Mather, J. C.) vol. 5487, 50-61 (SPIE, Bellingham, Washington, 2004) Werner, M. W., et al., [*Ap. J. Suppl.*]{}, [**154**]{}, 1 (2004) Horne, K., [*Publ. Astron. Soc. Pacif.*]{}. [**98**]{}, 609 (1986) Richardson, L. J., Harrington, J., Seager, S. & Deming, D., [*astro-ph/0606096*]{}, (2006) The passage of an extrasolar planet behind its star — an eclipse — results in a drop in the total system flux. This allows a measurement of the absolute flux emerging from the part of the planet that faces the star (the day side). This is possible when the orbit is aligned almost edge-on to the line of sight. Deming, D., Harrington, J., Seager, S. & Richardson, L. J., [*Ap.J.*]{}, [**644**]{}, 560 (2006) We use the estimator provided by the Spitzer Science Center, which is based on the DIRBE model described by [@Kels]. Further information on the specific implementation is given at [http://ssc.Spitzer.caltech.edu/documents/background]{}. Cho, J., Menou, K., Hansen, B. & Seager, S., [*Ap. J.*]{}, [**587**]{}, L117 (2003) Cooper, C. S. & Showman, A. P., [*Ap. J.*]{}, [**629**]{}, L45 (2005) Henry, G. W., Baliunas, S. L., Donahue, R. A., Fekel, F. C. & Soon, W., [*Ap. J.*]{}, [**531**]{}, 415 (2000). Shkolnik, E., Walker, G. A. H., Bohlender, D. A., Gu, P.-G. & Kürster, M., [*Ap. J.*]{}, [**622**]{}, 1075 (2005) Seager, S., Richardson, L. J., Hansen, B., Menou, K., Cho, J. & Deming, D., [*Ap. J.*]{}, [**632**]{}, 1122 (2005) Barman, T., Hauschildt, P. & Allard, F., [*Ap. J.*]{}, [**632**]{}, 1132 (2005) Fortney, J., Marley, M., Lodders, K., Saumon, D. & Freedman, R., [*Ap. J.*]{}, [**627**]{}, L69 (2005) Burrows, A., Hubeny, I. & Sudarsky, D., [*Ap. J.*]{}, [**625**]{}, L135 (2005) Fischer, D. A. & Valenti, J., [*Ap. J.*]{}, [**622**]{}, 1102 (2005) Kelsall, T., et al., [*Ap. J.*]{}, [**508**]{}, 44 (1998). The usual estimate given for planetary temperatures is equilibrium temperature $T_{\rm eq}$, defined as the effective temperature of a uniformly bright planet radiating energy at a rate that balances the irradiation received from the star. $T_{\rm eq}$ is thus determined by the stellar effective temperature $T_{\rm eff}$, stellar radius $R$ and distance of the planet from the star $a$; $T_{\rm eq} = 1840~{\rm
K} (T_{\rm eff}/ 6212~{\rm K}) (R/ 1.57 R_{\odot})^{1/2} (a/0.059
{\rm AU})^{-1/2}$, in the case of $\upsilon$ And b with zero albedo. However, in a proper no-redistribution model, the temperature distribution is not uniform but rather hottest at the substellar point and coolest at the limb, and the full-phase temperature average over the planetary surface is better approximated by $(4/3)^{1/4} T_{\rm eq}$. This is the temperature we adopt, which is $1977$ K in this case.
This work is based on observations made with the Spitzer Space Telescope, which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract to NASA. Support for this work was provided directly by NASA, its Origins of Solar Systems and Astrophysical Theory Programs as well as the Astrobiology Institute and Spitzer. We thank the personnel of the Spitzer Science Center and its MIPS instrument, who ultimately made these measurements possible.
=6truein
=6truein
=6truein
=6truein
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We show that single-atoms can be trapped on the surface of a subwavelength-diameter silica-fiber, an optical nanofiber, without any external field, and that single photons spontaneously emitted from the atoms can be readily detected through the single guided-mode of the nanofiber. A key point of the work is our finding that atom trapping sites are created on the nanofiber surface by irradiating the atom cloud around the nanofiber with a violet laser radiation.'
author:
- 'K. P. Nayak and K. Hakuta'
title: Single Atoms on an Optical Nanofiber
---
In recent years significant progress has been achieved in manipulating single atoms. Many kinds of single-atom-localizing and trapping schemes have been demonstrated so far using various external fields. Examples would be magneto-optical traps with high magnetic field gradient [@one; @two; @three], ion traps [@four; @five], far-off-resonance optical traps [six,seven]{} or high-finesse optical cavities [@eight; @nine]. However, in order to further extend these technologies, development of a simpler atom-trapping method without external fields will prove promising. Also in this context a major problem arises of how to detect the small number of atoms. Quite often, these experiments rely on complicated design of high numerical aperture optics or high-finesse optical cavities that may add to the technical difficulties. Hence development of a more realistic detection system will also accelerate the technological advancements.
![(a) Conceptual diagram of the experiment. The nanofiber locates at the waist of a tapered optical fiber. Fluorescence photons coupled to the guided mode of the nanofiber are detected at one end of the single mode optical fiber using an avalanche photodiode (APD). (b) Fluorescence excitation spectra measured through the nanofiber for the closed cycle transition, 6S$_{1/2}$ F = 4 $\leftrightarrow$ 6P$_{3/2}$ F’ = 5. Detuning is measured with respect to the atomic resonance. Traces A and B correspond to without and with the effect of violet laser irradiation, respectively. The spectra are displayed in two vertical scales so that the change of the spectrum is readily seen.[]{data-label="fig1"}](fig1.eps){width="8cm"}
Here we show that single atoms can be trapped on a subwavelength-diameter silica-fiber, an optical nanofiber [@ten; @eleven], and that single photons spontaneously emitted from the atoms can be readily guided into a single-mode optical fiber. Experiments are performed by overlapping cold Cs-atoms in a magneto-optical trap (MOT) with an optical nanofiber and observing fluorescence photons through the nanofiber after switching off the MOT laser beams. A key point of the work is our finding that atom trapping sites are created on the nanofiber surface by irradiating the atom cloud around the nanofiber with a violet laser radiation.
Figure 1(a) shows the conceptual diagram of the work. The nanofiber is located at the waist of a tapered optical fiber which is produced by heating and pulling commercial single-mode optical fibers. The fibers are adiabatically tapered so that the single mode propagation condition can be maintained for the whole fiber length. Note that atoms around the nanofiber emit an appreciable fraction of the fluorescence photons into the guided mode, since the mode distribution around the nanofiber is strongly confined to the guided mode [@eleven; @twelve]. In the present experiments, we use nanofibers [@eleven] with a diameter of 400 nm and a length of 2 mm. A MOT equipped with a resistively heated alkali dispenser source is used to produce cold Cs-atoms. The MOT position is controlled to overlap with the nanofiber [@eleven]. Atom number density in MOT and MOT size are controlled by adjusting the dispenser current. The maximum number density around the nanofiber is 7 $\times $ 10$^{9}$ cm$^{-3}$ with a MOT diameter of 2 mm and the temperature of atoms is around 100 $\mu $K. The MOT laser beams are switched off for 10 $\mu $s periodically at an interval of 200 $\mu $s. During the switched-off periods, atoms around the nanofiber are excited by a probe laser beam irradiated perpendicular to the nanofiber in a standing-wave configuration with a polarization perpendicular to the fiber. Fluorescence photons are observed at one end of the fiber using an avalanche photodiode and a single-photon counting system, and the photon counts are accumulated for many cycles.
Figure 1(b) shows the excitation spectra measured for a closed cycle transition, 6S$_{1/2}$ F = 4 $\leftrightarrow $ 6P$_{3/2}$ F’ = 5, by scanning the probe laser frequency around the atomic resonance. The MOT is set to the maximum density condition and the probe laser diameter is set to 2 mm to irradiate the whole atom cloud. Intensity of the probe laser is set to 3.3 mW/cm$^{2}$. The trace A denotes the observed spectrum without the effect of violet laser. As reported previously, the observed line shape is quite different from the usual atomic line shape [@eleven; @thirteen; @fourteen]. The spectrum consists of a sharp peak near the resonance and a long tail on the red detuned side. These observations are attributed to the van der Waals (vdW) interaction between the Cs-atom and the nanofiber surface which would be dominant for distances closer than $\lambda /2\pi $ from the surface. The observed spectrum is well understood through a process in which atoms close to the nanofiber fall into a deep vdW potential. During the experiments, we have found that the excitation spectrum changes drastically when we have irradiated the nanofiber with violet laser radiation of wavelength 407 nm. The conditions are the following: The nanofiber region is irradiated with the violet laser in the presence of the MOT for several minutes. Irradiating power is 5 mW with a beam diameter of 2 mm (the irradiating intensity is 150 mW/cm$^{2}$). After switching off the violet laser the fluorescence excitation spectrum is measured. The observed spectrum is denoted by trace B in Fig. 1(b) and the inset. As seen readily, the spectrum is very different from the trace A. The spectrum becomes very much squeezed towards the atomic resonance, and the peak fluorescence count increases almost 30 times. The width of the spectrum is 15 MHz FWHM. We should note that the irradiation in the absence of the MOT has no effect on the spectrum. We have found that once the nanofiber is irradiated, the observed effect lasts for several days. The observations suggest that the violet laser irradiation has modified the nanofiber surface in such a way that most of the atoms are kept from falling into the vdW potential, and dwell near the surface at some specific distance.
In order to clarify the single-atom characteristics around the nanofiber, we perform Hanbury-Brown and Twiss (HBT) experiments by reducing the atom number. The fluorescence light through fiber are split into two using a 3 dB fiber coupler, and are detected by two separate avalanche photodiodes. The photon correlations between the two channels are measured using a time-correlated single-photon-counter with a time resolution of 1 ns. In the measurements, the dispenser current is decreased to reduce both atom number and MOT size. The minimum MOT size is 80 $\mu $m in diameter with an atom density of $\sim $* *$0.7\times 10^{9}$ cm$^{-3}$ for dispenser current I$_{D}$ = 3.8 A. The probe laser is tuned close to the atomic resonance, and is line focused down to 100 $\mu $m using a cylindrical lens to spatially restrict the observation region to the atom cloud. The focused probe laser intensity is 60 mW/cm$^{2}$. Average atom number in the observation region is estimated to be much less than one for I$_{D}$ = 3.8 A, assuming an observation volume around the nanofiber of 200 nm in thickness [@eleven] and 100 $\mu $m in length. Fluorescence photon count is $\sim $* *$1\times 10^{4}$ counts/s under this condition.
![Photon correlations with Hanbury-Brown and Twiss arrangement for atoms around the nanofiber. $\protect\tau $ and I$_{D}$ denote the delay time between the two channels and the dispenser current, respectively. Dashed curves denote the theoretically calculated photon coincidences for N = 1 and 2, respectively, where N denotes the number of atoms.[]{data-label="fig2"}](fig2.eps){width="8cm"}
Figure 2 displays the observed coincidences for different delay time between the two channels. Each curve is obtained after an integration time of 3 minutes which requires a measurement time of 1 hour. The coincidences for I$_{D}$ = 0 A correspond to the background for the uncorrelated scattered light from the nanofiber. The coincidences for I$_{D}$ = 3.8 A and 4 A show the correlations for the fluorescence photons. Both curves show clear antibunching of fluorescence photons at zero time-delay and Rabi oscillation behaviors in the wings. The curves are fitted using a formula for photon coincidences from N-atoms [@eighteen], $Ng^{(2)}(\tau )+N(N-1)$, where $g^{(2)}(\tau )$ is the correlation function for a single atom and is calculated assuming spontaneous emission time of 30 ns and Rabi frequency of 13 MHz. Observed and fitted curves are in good agreement. Thus, observed coincidences for I$_{D}$ = 3.8 A and 4 A are ascribed to photon correlations from one atom and two atoms, respectively. Note that the antibunching is observable for atoms only after the violet laser irradiation. We should note that the assumed Rabi frequency is 1.3 times smaller in value than that simply estimated for the F = 4 $\leftrightarrow $ 5 transition.
![Temporal behavior of fluorescence signal measured through the nanofiber under single-atom condition. The black dots denote the observed fluorescence counts at various delay time after switching off the MOT beams. The red curve is the exponential fit to the data giving a decay time of 180 $\protect\mu $s.[]{data-label="fig3"}](fig3.eps){width="7cm"}
Next, the temporal behavior of the fluorescence signals is measured under the single-atom condition. Measurements are performed by extending the switched-off (on) periods to 1 ms (20 ms). Signals are measured for a gate time of 50 $\mu $s in the switched-off periods, at various delay time after switching off the MOT beams. The signals are accumulated for many cycles. The results are exhibited in Fig. 3. The temporal behavior shows an exponential decay with a lifetime of 180 $\mu $s. This lifetime corresponds to the atom dwelling time in the observation volume. If atoms are free around the nanofiber, the dwelling time may be determined by the transit time across the observation volume. Assuming a mean velocity of atoms as 10 cm/s and an atom transit-length of 1 $\mu $m, one can estimate the transit time to be $\sim $ 10 $\mu $s, which is about twenty-times shorter than the observed lifetime. It means that atoms around the nanofiber are not completely free; atoms are localized in the close vicinity of the nanofiber surface.
![Solid curves in (a), (b), and (c) exhibit the fluorescence excitation spectra measured through the nanofiber under one single-atom condition for three different probe laser intensities 0.7, 3.5 and 7 mW/cm$^{2}$, respectively. Detuning is measured with respect to the atomic resonance. Dashed curves are theoretically calculated spectra assuming a V-type three level scheme.[]{data-label="fig4"}](fig4.eps){width="7cm"}
We measure the excitation spectra under the single atom condition to further clarify the behaviors of atoms. Observed results for three different probe laser intensities are displayed in Figs. 4(a-c) by solid curves. Measurements are performed by using the similar procedures as those for Fig. 1(b). Peculiar features are readily seen. For the lowest laser intensity, the spectrum exhibits almost a Lorentzian shape with 8 MHz FWHM, slightly broader than the spontaneous-emission lifetime broadening, but a sharp, small dip is seen at the peak. With increasing laser intensity, the signal becomes stronger and the sharp dip becomes more apparent, but the dip width is narrower than the lifetime broadening. Such spectral characteristics with a sharp dip cannot be explained by simply overlapping two Lorentzian profiles. Such characteristics can be explained by incorporating a quantum interference effect. The observed spectral characteristics may be explained with a V-type three-level scheme discussed in Refs. [@ninteen; @twenty]. The scheme consists of two upper levels closely spaced within the radiative broadening and one lower ground level, and the upper levels spontaneously decay to the lower level with the same rate. By solving density matrix equations under stationary condition, the excitation spectrum can be calculated for various probe laser intensities (Rabi frequencies). Calculated spectra for three Rabi frequencies are overlaid on the observed ones in Figs. 4(a-c) with dashed lines. Spacing between the upper levels is assumed to be 1.5 MHz. The sharp dips are reproduced, and the relative spectral intensities are also well reproduced. Regarding the Rabi frequencies, we have used 1.4 times smaller values than those for the F = 4 $\leftrightarrow $ 5 two-level transition to keep the total decay rate of the upper state equal to the value for the F = 5 level.
Observed spectra clearly reveal that the upper atomic state of the transition is split into two levels. The observed splitting may be understood as a consequence of the localization of atoms into a tiny potential in the close vicinity of the nanofiber surface; that is, the atom is trapped in the potential and, moreover, the atomic motion is quantized resulting in the two motional sub-levels for the excited electronic state. We suspect that two sub-levels are also created for the ground electronic state, because the depths for the center dip can be better reproduced by introducing two sub-levels for the ground state.
The spectral measurements clarify the discrepancy found in the HBT-experiments that the fitted Rabi frequencies are much smaller than the two level values, because the localized atom is not a simple two-level atom, and the two-level estimation is not appropriate anymore. Regarding the spectrum B in Fig. 1(b), we have not observed any sharp spectral dip. It may be understood to be due to some inhomogeneous broadening for many localized atoms in many localizing sites which might have washed out the quantum interference effects. Number of localized atoms in the spectrum B may be estimated to be $\sim $ 200 by comparing the integrated intensities for the spectrum B and that for one single-atom.
{width="6cm"}
We have not yet identified the mechanism of the atom trapping definitively, but we speculate that the following scenario may be occurring. Regarding the effect of the violet laser irradiation, it is not due to the light induced atom desorption (LIAD) well known for alkalis [@sixteen]. If it were due to the LIAD process, the effect would also appear in the absence of the MOT. By irradiating the MOT atoms with a 407 nm laser, the atoms are photo-ionized [@seventeen] from the excited electronic state 6P$_{3/2}$. Generated Cs-ions then stick to the cusps of some prominences on the nanofiber surface, and the stuck charges form atom-trapping orbits around them due to the mutual balance between the induced-dipole force, the centrifugal force, and the vdW force, as schematically illustrated in Fig. 5. Since the observed resonances are close to the free atomic resonance and the width of the trace B in Fig. 1 (b) is 15 MHz, height of the prominences may distribute from 50 to 100 nm assuming the vdW shift. Similar atom trapping with orbiting trajectory is discussed in Ref. [@orbiting]. Based on their results and assuming an inverse proportionality of the orbiting frequency to the orbit radius, one can expect an orbit radius to be around 30 nm for the observed splitting of 1.5 MHz.
In conclusion, we have found that single-atom localizing sites are created on a nanofiber surface by irradiating the MOT atoms around the fiber with a violet laser radiation so that single atoms are trapped without any external field, and that single photons spontaneously emitted from the atoms are readily guided into a single-mode optical fiber. The present finding may be extended to various surfaces and various atomic and molecular species.
We are thankful to Fam Le Kien, Shinichi Watanabe, Makoto Morinaga, Mark Sadgrove, and Manoj Das for useful discussions and technical assistance. This work was carried out under the 21st Century COE program on Innovation in Coherent Optical Science.
[99]{} Z. Hu and H. J. Kimble, Opt. Lett. **19**, 1888 (1994).
F. Ruschewitz *et al.*, Europhys. Lett. **34**, 651 (1996).
D. Haubrich *et al.*, Europhys. Lett. **34**, 663 (1996).
W. Neuhauser *et al.*, Phys. Rev. A **22**, 1137 (1980).
D. J. Wineland and W. M. Itano, Phys. Lett. A **82**, 75 (1981).
D. Frese *et al.*, Phys. Rev. Lett. **85**, 3777 (2000).
B. Darquié *et al.*, Science **309**, 454 (2005).
J. Ye, D. W. Vernooy, and H. J. Kimble, Phys. Rev. Lett. **83**, 4987 (1999).
P. W. H. Pinkse *et al.*, Nature **404**, 365 (2000).
L. Tong *et al.*, Nature **426**, 816 (2003).
K. P. Nayak *et al.*, Opt. Express **15**, 5431 (2007); quant-ph/0610136 (2006).
F. L. Kien *et al.*, Phys. Rev. A **72**, 032509 (2005).
F. L. Kien and K. Hakuta, Phys. Rev. A **75**, 013423 (2007).
F. L. Kien, S. Dutta Gupta, and K. Hakuta, Phys. Rev. A **75**, 032508 (2007).
V. Gomer *et al.*, Appl. Phys. B **67**, 689 (1998).
S. E. Harris, Phys. Rev. Lett. **62**, 1033 (1989).
P. Zhou and S. Swain, Phys. Rev. Lett. **78**, 832 (1997).
A. Hatakeyama, M. Wilde, and K. Fukutani, e-J. Surf. Sci. Nanotech. **4**, 63 (2006).
O. Marago *et al.*, Phys. Rev. A **57**, R4110 (1998).
V. I. Balykin *et al.*, Phys. Rev. A **70**, 011401(R) (2004).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In recent years, the discovery of complex dynamic systems in various fields through data-driven methods has attracted widespread attention. This method has played the role of data and has become an advantageous tool for us to study complex phenomena. In this work, we propose a framework for detecting the dynamic behavior, such as the maximum likelihood transition path, of stochastic dynamic systems from data. For the stochastic dynamic system, we need to use the Kramers-Moyal formula to convert it into a deterministic form for processing, then use the extended SINDy method to obtain the parameters of stochastic dynamic systems, and finally calculate the maximum likelihood transition path. We give two examples of stochastic dynamical systems driven by additive and multiplicative Gaussian noise, and demonstrate the validity of the method by reproducing the known dynamical system behavior.'
address:
- |
School of Mathematics and Statistics, & Center for Mathematical Science,\
Huazhong University of Science and Technology, Wuhan, 430074, China.
- 'Twitter, 1335 Market St \#900, San Francisco, CA 94103, USA.'
- |
Wuhan National Laboratory for Optoelectronics,\
Huazhong University of Science and Technology, Wuhan, 430074,China.
- |
Department of Applied Mathematics, College of Computing,\
Illinois Institute of Technology, Chicago, IL 60616, USA.
author:
- Min Dai
- Ting Gao
- Yubin Lu
- Yayun Zheng
- Jinqiao Duan
title: Detecting the maximum likelihood transition path from data of stochastic dynamic systems
---
data-driven ,maximum likelihood transition path,Stochastic dynamical systems,Kramers-Moyal formula
Introduction
============
Stochastic differential equations are widely used to describe random phenomena in disciplines such as physics, biology, chemistry, and geophysics. For such a stochastic problem, we usually build an appropriate mathematical model based on the basic laws, and then analyze or simulate the model to obtain a characterization of the nonlinear phenomena of the problem. However, for some phenomena that are too complicated, we lack sufficient understand for them, and it is difficult to establish complete mathematical models, or the models corresponding to some phenomena are too complicated to analyze. Fortunately, with the improvement of observation technology and computing power, although there is not enough understanding of the problem, there are still many valuable observation or simulation data that can be used. Therefore, it is necessary to directly discover the dynamic system indicators from data and obtain a characterization of stochastic phenomena.
The analysis of dynamical complex behavior based on data has received extensive attention in recent years. Many authors have come up with insightful methods based on areas they are familiar with. For example, the sparse identification of nonlinear dynamics (SINDy) was used by Brunton, Kutz and et al. [@Bru; @Rudy; @Alla], to discover the governing equation from data. Furthermore, SINDy was applied in the learning of biological networks, which effectively dealt with the problem of rational functions [@Man]. Zhang and Lin [@Zhang] used threshold sparse Bayesian regression to discover the governing physical laws from data. Moreover, there were many other methods, such as stochastic parametrization [@Cho], learning informed observation geometries [@Yai], the Koopman Operator [@Wil; @Sch], Gaussian processes [@Arc; @Rut; @Batz], extended the SINDy methods [@Bon] and so on. However, most of authors focused on system identification [@Gar; @Sch] and lacked sufficient attention to extract dynamic system indicators [@Wu] from data.
The maximum likelihood transition path is a significant indicator for describing the behavior of stochastic dynamic systems [@Zheng; @Duan; @Gao], which provides information about the transition phenomena under the interaction of nonlinearity and uncertainty. We know that it is difficult to extract the accurate the maximum likelihood transition path from data under the influence of nonlinearity and uncertainty. The purpose of this paper is to devise a method to extract the maximum likelihood transition path of the stochastic dynamical system from data. We first have used the Kramers-Moyal formula [@Schuss] to convert the stochastic problem into the deterministic problem, then estimated the coefficients of stochastic differential equations [@Oksendal] by extending the SINDy method [@Bon] from data to calculate the maximum likelihood transition path.
The work is arranged as follows. In section 2, we introduce the theory of the article, including the Kramers-Moyal formula, extending the SINDy method and the maximum likelihood transition path. In section 3, the results of the numerical experiment by stochastic differential equations with additive and multiplicative noise are presented. In section 4, we give some conclusions and discussions.
Theory
======
With the advancement of technology, the combination of stochastic differential equations and data has become a powerful tool for our research. In this work, we consider the stochastic differential equation in $\mathbb{R}$ as follows $$\label{eq:1}
dX_t=f(X_t)dt+\sigma(X_t){dW_t},~~~ X_0=x,$$ where $f,\sigma$ are called the drift and the diffusion of the process in $\mathbb{R}$, respectively. The vector $X_t\in{\mathbb{R}}$ denotes the state of a system at time $t$ and $W_t$ is a standard Brownian motion. For the stochastic differential equation , the drift $f$ and the diffusion $\sigma$ are two important parts. And fortunately, the Kramers-Moyal formula [@Schuss] provides us with a way to detect them from data $$\label{eq:2}
{f(x)}=\lim_{\Delta{t}\rightarrow{0}}{\mathbb{E}{\Bigg[\frac{(X_{{\Delta{t}}}-X_0)}{\Delta{t}}|X_0=x\Bigg]}},$$ $$\label{eq:3}
{\sigma^2(x)}=\lim_{\Delta{t}\rightarrow{0}}{\mathbb{E}{\Bigg[\frac{(X_{{\Delta{t}}}-X_0)^2}{\Delta{t}}|X_0=x\Bigg]}}.$$ The above equations calculate the expectation and variance of the process $X(t)$ under the condition of position $x$ at time $t$ respectively. If the expectation and variance can be calculated from data, we can easily obtain the parameters of the stochastic differential equation.
**[Data-Driven Parameterization of Stochastic Differential Equation]{}**
-------------------------------------------------------------------------
We now apply the extend the SINDy algorithm [@Bon] by considering conditional expectation and with model $Y={\Theta\Xi}$, where $Y^1=\lim_{\Delta{t}\rightarrow{0}}{\mathbb{E}{[\frac{(X_{{\Delta{t}}}-X_0)}{\Delta{t}}|X_0=x]}}$ or $Y^2=\lim_{\Delta{t}\rightarrow{0}}{\mathbb{E}{[\frac{(X_{{\Delta{t}}}-X_0)^2}{\Delta{t}}|X_0=x]}}$. For example, suppose that the basis function is polynomial $\{1,X,X^2,X^3,\cdots\}$, then using the basis function $$\Theta(X)=
\left[
\begin{array}{ccccc}
1 & x_1 & x_1^2 & x_1^3 & \cdots\\
1 & x_2 & x_2^2 & x_2^3 & \cdots\\
\vdots & \vdots & \vdots & \vdots & \vdots \\
1 & x_N & x_N^2 & x_N^3 & \cdots\\
\end{array}
\right] ,$$ we can change Eqs. and into the following form $$\label{eq:4}
\begin{split}
Y^1=\lim_{\Delta{t}\rightarrow{0}}{\mathbb{E}{\Bigg[\frac{(X_{{\Delta{t}}}-X_0)}{\Delta{t}}|X_0=x\Bigg]}}={\Theta(X)\Xi^1},\\
Y^2=\lim_{\Delta{t}\rightarrow{0}}{\mathbb{E}{\Bigg[\frac{(X_{{\Delta{t}}}-X_0)^2}{\Delta{t}}|X_0=x\Bigg]}}={\Theta(X)\Xi^2},
\end{split}$$ where $\Xi^i=[\xi_1^i,\xi_2^i,\cdots,\xi_N^i]^T,~i=1,2$ are the coefficients of the sparse vectors.
Vectors $\Xi^1$ and $\Xi^2$ are sparse vector of coefficients. They can be obtained by the following formula $$\Xi^i=\Theta(X)^{\dag}Y^i,~~~i=1,2.$$ Brunton et al.[@Bru] considered such a learned model, however, their only focused on the deterministic dynamical system. In our work, we consider the stochastic dynamical system. Due to the influence of random term, the original optimization method can not meet our requirements, so we introduce a new optimization form [@Bon].
### [**1. Binning**]{} {#binning .unnumbered}
The effect of random noise is a difficult problem in dealing with stochastic problems, so we need to preprocess the data to reduce noise. The coordinate binning is an effective method to reduce the influence of noise. To solve Eq., we need to use the data $X$ from the stochastic differential equation . Therefore, we first place the time series data $X$ into $G$ bins, $$\{X(t_i)\}_{i=1,\cdots,N}\mapsto\{\overline{x}_j\}_{j=1,\cdots,G},$$ with $\overline{x}_j$ representing the center of the $j$th bin. Next, we have $$\mathbb{X}\in{\mathbb{R}^{N\times{M}}}\mapsto{\mathbb{X}_G\in{\mathbb{R}^{G\times{M}}}},$$ $$\mathbb{Y}\in{\mathbb{R}^{N}}\mapsto{\mathbb{Y}_G\in{\mathbb{R}^{G}}},$$ with the element of ${Y}_G$ representing the center of each bin.
### [**2. Stepwise Sparse Regressor**]{} {#stepwise-sparse-regressor .unnumbered}
Inspired by Brunton et al.[@Bru], we will use the threshold approach to find the sparse coefficients. We first use a standard linear regression which is unconstrained to compute Eq. to obtain a non-sparse solution $c$. Then, let the element in the coefficient $c$ that is less than the preset threshold value $\lambda$ be $0$. Next, the remaining coefficients are continued to repeat the linear regression of the first step until no element in the coefficient $c$ is found to be less than the threshold. However, a significant drawback to this approach is need to adjust the threshold $\lambda$ appropriately. Therefore, we introduce the Cross Validation instead of threshold $\lambda$ to avoid this problem. This algorithm is called *Stepwise Sparse Regressor*$(SSR)$, and the following is a specific description£º
Step $1$: The initial solution $\widetilde{\textbf{b}}$ is obtained by solving the following least squares regression equation, which may not be a sparse, $$\widetilde{\textbf{b}}=\arg\min_{\textbf{b}\in{\mathbb{R}^M}}{\|\mathbb{Y}-\mathbb{X}\textbf{b}\|_2^2}.$$
Step $2$: Taking the absolute value of all coefficients of the initial solution $\widetilde{\textbf{b}}$ and making the smallest one be $0$, $$\widetilde{b}_i=0,~~~ i=\min_m{|\widetilde{b}_m|}.$$ This approach is similar to the threshold method, and the advantage is that the number of iterations required to run is equal to the sparsity of the solution, allowing us to program accurately.
Step $3$: Least square regression calculation for the rest again $$\mathbb{Y}=\mathbb{X}[:,\widehat{j}]\widetilde{b}[\widehat{j}],$$ with $\widehat{j}$ denoting all sets except $j$ and $\mathbb{X}[:,\widehat{j}]$ representing the remaining matrix except column $j$.
Step $4$: Repeating the above steps continuously until the optimal solution $\widetilde{c}$ indicated by Cross Validation is reached.
Note that we simply write SSR as $$\emph{SSR}(\mathbb{X},\mathbb{Y})_q,$$ indicating that running algorithm $q$ times can obtain the solution $\textbf{b}$. This means that the solution has $q$ zeros, called $q$-sparse.
### [**3. Cross Validation**]{} {#cross-validation .unnumbered}
Cross Validation (CV) is a traditional statistical verification technique that is applied in interdisciplinary fields such as model selection and hyperparameter selection. In this work, we find the optimal solution through cross validation techniques, which has the advantage that the number of iterations required for Stepwise Sparse Regression can be determined.
There are many parametric models for a given data set $\mathcal{D}$, we however want to find the most suitable mathematical model. For Cross Validation, the data set is usually divided into training set and test set. Then we obtain the cross-validation (CV) score by computing the deviation $\delta$ of the test set on the training model, which measures how to balance accuracy and predictability in the model. We finally select the parameter model corresponding to the minimum value of the CV score as the optimal model.
In addition, we know that each iteration of the algorithm will produce the solution with different degrees of sparsity $$\{SSR_q\}_{q=1,\cdots,M},$$ we then can obtain different models, and calculate the CV scores $\delta[SSR_q]$.
There are many types of cross validation methods. Here, we will partition the data set into $k$ parts and then use each part as a test set in turn to train the parameter model. Firstly, we randomly divide the data set $\mathcal{D}$ into $k$ parts with the same size, that do not intersect each other, i.e., $\bigcup_i{B_i}=\mathcal{D}$, $B_i\cap{B_j}=\emptyset$. Moreover,$$\mathbb{X}_{B_i}=\mathbb{X}[p_{B_i},:], ~~~p_{B_i}=\bigcup_{p\in{B_i}}p,$$ with $\mathbb{X}[p_{B_i},:]$ denoting the $p_{B_i}$th row of the matrix. We can next define the CV score as follows $$\label{eq:5}
\delta^2[SSR_q]=\frac{1}{k}{\sum_{i=1}^{k}}{\|\mathbb{Y}_{B_i}-\mathbb{X}_{B_i}\cdot{SSR(\mathbb{X}_{C_i},\mathbb{Y}_{C_i})_q}}\|_2^2,$$ $$\label{eq:6}
C_p=\bigcup_{i\neq{p}}{B_i},$$ with $SSR(\mathbb{X}_{C_i},\mathbb{Y}_{C_i})_q$ representing the parameter coefficients which are $q$-sparse obtained by running $SSR$ algorithm on training set $C_i$. The cross validation score has a good measure of the accuracy, predictability and sparseness of the model. We hope that the model results corresponding to low cross validation score with variable sparsity are better. The drift and diffusion terms can be learned in the above way, thus we can extract stochastic model and some dynamic system behaviors.
We use the extended SINDy method in this work. On the one hand, the coordinate binning effectively reduces the impact of noise. In addition, we cleverly use the CV method to avoid the problem of adjusting the appropriate threshold, and find the number of iterations precisely, which improves the accuracy of the results.
**The maximum likelihood transition path**
------------------------------------------
A good tool for describing the dynamic behavior of stochastic dynamic systems is the maximum likelihood transition path [@Zheng]. Therefore, in this work, we will use it to characterize the information we extract from the data and determine whether it is accurate or not. The maximum likelihood transition path is a manifestation of transition of a dynamic system from one state to another. In other words, for $t\in[0,t_f]$, $x,x_0,x_{f}\in\mathbb{R}$, when conditions $X(0)=x_0$ and $X(t_f)=x_{f}$ are given, we then suppose that the function $\mathcal{P}_{A}(x,t)$ which is conditional probability density exists (where $A$ indicates these two-point conditions) and expresses as follows: $$\label{eq:7}
\begin{split}
\mathcal{P}_{A}(x,t)&=p(X(t)=x|X(0)=x_0;X(t_f)=x_{f})\\
&=\frac{Q(x_{f},t_f|x, t)Q(x,t|x_0,0)}{Q(x_{f},t_f|x_0,0)},
\end{split}$$ with transition probability density $Q$, which is the solution of Fokker-Planck equation.
Next, we will give a brief introduction to $Q$. First of all, we will introduced two expressions of probability density. One is $p(X(t)=u)$ ,which we use to express the probability density of SDE solution $X(t)$ at $X(t)=u$ in $\mathbb{R}$, the other is what we call the transition density $Q(u;t|\xi,s)$, which is defined on $\mathbb{R}\times[0,t_f]\times\mathbb{R}\times[0,t_f]$. Furthermore, there is a relationship between $p$ and $Q$. For instance, for $0\leqslant s<t \leqslant t_f$, given $X(s)=v$, then the density of $X(t)$ can be expressed by $Q(u;t|\xi;s)$ at $X(t)=u$. That is to say, we have the following expression $$Q(u;t|\xi;s)=p(x(t)=u|x(s)=v).$$ For every $x_0\in\mathbb{R}$, we assume that the transition density $Q(u;t|\xi,s)$ for Eq. meets the following Fokker-Planck equation [@Duan] $$\label{eq:8}
\begin{split}
\frac{\partial}{\partial t}Q(x;t|\xi,s)&=-\sum^{d}_{i=1}\frac{\partial}{\partial x_i}(f_i(x)Q(x;t|\xi,s))+\frac{1}{2}\sum^{d}_{i,j=1} \frac{\partial^2}{\partial x_i x_j}(\eta^{i,j}(x)Q(x;t|\xi,s)),
\end{split}$$ where $\eta(x)=\sigma(x)\sigma(x)^T$.
Finally, we will further explain the significance of the maximum likelihood transition path. In Eq., we can see that the density function $\mathcal{P}_{A}(x,t)$ is determined by the transition probability density $Q(u;t|\xi;s)$ of the state at three different moments, and there is a peak of $\mathcal{P}_{A}(x,t)$ at $t\in[0,t_f]$ according to condition $A$. However, the stochastic orbit $X(t)$ will reach a position state $x_{m}(t)$ of maximum likelihood when the peak value is reached. Therefore, we can maximize the density function $\mathcal{P}_{A}(x,t)$ to get the state $x_m(t)$, $$\label{eq:9}
x_{m}(t)=\arg\max_{x}\mathcal{P}_{A}(x, t).$$ We can obtain the state $x_{m}(t)$ through global numerical optimization method. In addition, it is worth mentioning that the most probable state depends on the choice of $t_f$, that is, different values of $t_f$, the maximum likelihood transition path will change.
Numerical experiments
=====================
A few examples verifying the feasibility of our method for detecting the maximum likelihood transition path from data will be presented. Here, we consider two stochastic dynamic systems driven by additive and multiplicative Gaussian noise, respectively. Basic functions are selected as polynomials.
**Example 1**: The double-well system driven by additive Gaussian noise $$\label{eq:10}
dX_{t}=(4X_t-X_t^3)dt+{dB_{t}},
$$ where $X_t$ is a $\mathbb{R}$-valued stochastic process and $\sigma=1$. Firstly, we select the basis function $\Theta(X)$, which has different sizes $n$, and then obtain the cross-validation score $\delta_{\Theta}$ by calculating Eq.. Figure $1$ shows the relationship between $n$ and the cross validation score: as $n$ increases, the cross validation decreases gradually, and tends to constant.
According to the cross validation score, we can determine that the size of the dictionary $\Theta(X)$ of the drift term and the diffusion term are $n=2$ and $n=1$, respectively. And the results for the Eq. are as follows,\
captype[table]{}
------- --------- -- --
basis $f(x)$
1 0
$x$ 4.0362
$x^2$ 0
$x^3$ -1.0225
$x^4$ 0
$x^5$ 0
------- --------- -- --
captype[table]{}
------- ------------ -- --
basis $\sigma^2$
1 1.0094
$x$ 0
$x^2$ 0
$x^3$ 0
$x^4$ 0
$x^5$ 0
------- ------------ -- --
So, the results corresponding to the original equation are $f(x)=4.0362x-1.0225x^3$ and $\sigma^2=1.0094$.
Next, we study the dynamic behavior of the double-well system with additive Gaussian noise and draw the maximum likelihood transition path diagram. In Figure $2$, (a) gives the maximum likelihood transition path diagram of original equation under what conditions, and (b) gives the maximum likelihood transition path of double-well system obtained from learning data under the same conditions.
**Example 2**: The double-well system driven by multiplicative Gaussian noise $$\label{eq:11}
dX_{t}=(4X_t-X_t^3)dt+(X_t+1){dB_{t}},
$$ where $X_t$ is a $\mathbb{R}$-valued stochastic process and $\sigma(x)=x+1$. Firstly, we select the basis function $\Theta(X)$, which has different sizes $n$, and then obtain the cross-validation score $\delta_{\Theta}$ by calculating Eq.. Figure $3$ shows the relationship between $n$ and the cross validation score: as $n$ increases, the cross validation decreases gradually, and tends to constant.
According to the cross validation score, we can determine that the size of the dictionary $\Theta(X)$ of the drift term and the diffusion term are $n=2$ and $n=1$, respectively. And the results for the Eq. are as follows,\
captype[table]{}
------- --------- -- --
basis $f(x)$
1 0
$x$ 3.9464
$x^2$ 0
$x^3$ -0.9998
$x^4$ 0
$x^5$ 0
------- --------- -- --
captype[table]{}
------- --------------- -- --
basis $\sigma^2(x)$
1 1.1450
$x$ 1.9535
$x^2$ 0.9135
$x^3$ 0
$x^4$ 0
$x^5$ 0
------- --------------- -- --
So, the results corresponding to the original equation are $f(x)=3.9464x-0.9998x^3$ and $\sigma^2(x)=1.1450+1.9535x+0.9135x^2$.
Next, we study the dynamic behavior of the double-well model with multiplicative Gaussian noise and draw the maximum likelihood transition path diagram. In Figure $4$, (a) gives the maximum likelihood transition path diagram of original equation under what conditions, and (b) gives the maximum likelihood transition path of double-well equation obtained from learning data under the same conditions.
Conclusion and discussions
==========================
In this work, we proposed a way to obtain dynamical quantities of stochastic dynamical systems from data. Especially, we demonstrated how to extract the maximum likelihood transition path from data. Another contribution is that our method adopted the Kramers-Moyal formula and extended SINDy to deal with stochastic differencial equations with multiplicative Gaussian noise. The advantage of our method is that it might be used to analyze complex phenomena directly from data and obtain quantitative characterization of stochastic systems, which reduces our dependence on models. Furthermore, our method has the potential to be used to analyze actual data such as genetic data, climate data, etc.
In addition, there are some interesting extensions for this study. First, we only consider the one-dimensional case. For the high-dimensional case, the coordinate binning method is not good for removing noise. How to deal with the high-dimensional stochastic dynamic system situation will become our challenge. On the other hand, for non-Gaussian noise, such as Lévy noise, it is also a subject worthy of our consideration. Currently, we are conducting research on Lévy noise.
Acknowledgements {#acknowledgements .unnumbered}
================
We would like to thank Prof. Xiangjun Wang, Dr. Xiujun Cheng, Dr. Xiaoli Chen, Dr. Wei Wei, Dr. Zibo Wang and Dr. Li Lv for helpful discussions. This work was supported by the NSFC grants 11531006, 11801192, 11771449 and NSF grant 1620449.
Appendix {#SI .unnumbered}
========
**S1. The Kramers-Moyal expansion**
We know that the solution of SDE is an It$\mathrm{\hat{o}}$ diffusion, which has an identity of the form $$\label{S1:1}
\begin{split}
p(x,t+\Delta{t})=\int{p(x^{'},t)p(x,t+\Delta{t}|x^{'},t)dx^{'}}
\end{split}$$ with the probability density $p(x,t+\Delta{t})$ and $p(x^{'},t)$ at time $t+\Delta{t}$ and $t$, respectively, and transition density $p(x,t+\Delta{t}|x^{'},t)$. Firstly, suppose that all the conditional moments $M^{(n)}(x^{'},t,\Delta{t})$ exist and have the following form $$\label{S1:2}
\begin{split}
M^{(n)}(x^{'},t,\Delta{t})&=\mathbb{E}{[{(x(t+\Delta{t})-x(t))^{n}}|x(t)=x]}\\
&=\int{(x-x^{'})^{n}{p(x,t+\Delta{t}|x^{'},t)dx}}.
\end{split}$$ In addition, the transition density $p(x,t+\Delta{t}|x^{'},t)$ can be written as $$\label{S1:3}
\begin{split}
p(x,t+\Delta{t}|x^{'},t)&=\int{\delta(y-x)p(y,t+\Delta{t}|x^{'},t)dy}\\
&=\sum_{n=0}^{\infty}{\frac{1}{n!}(-\frac{\partial}{\partial{x}})^{n}}\int{(y-x^{'})^{n}{p(y,t+\Delta{t}|x^{'},t)dy}\delta(x^{'}-x)}\\
&=\Bigg[1+\sum_{n=1}^{\infty}{\frac{1}{n!}(-\frac{\partial}{\partial{x}})^{n}}M^{(n)}(x^{'},t,\Delta{t})\Bigg]\delta(x^{'}-x),
\end{split}$$ where $$\begin{split}
\delta(y-x)&=\delta(y-x^{'}+x^{'}-x)\\
&=\sum_{n=0}^{\infty}{\frac{(y-x^{'})^{n}}{n!}(-\frac{\partial}{\partial{x^{'}}})^{n}}\delta(x^{'}-x)\\
&=\sum_{n=0}^{\infty}{\frac{(y-x^{'})^{n}}{n!}(-\frac{\partial}{\partial{x}})^{n}}\delta(x^{'}-x).
\end{split}$$ Substituting into , we then have $$\label{S1:4}
\begin{split}
p(x,t+\Delta{t})-p(x,t)&=\frac{\partial{p(x,t)}}{\partial{t}}\Delta{t}+o(\Delta{t}^2)\\
&=\int{p(x,t+\Delta{t}|x^{'},t)p(x^{'},t)dx^{'}}-p(x,t)\\
&=\int\Bigg[1+\sum_{n=1}^{\infty}{\frac{1}{n!}\Bigg(-\frac{\partial}{\partial{x}}\Bigg)^{n}}M^{(n)}(x^{'},t,\Delta{t})\Bigg]\delta(x^{'}-x)p(x^{'},t)dx^{'}-p(x,t)\\
&=\sum_{n=1}^{\infty}\Bigg(-\frac{\partial}{\partial{x}}\Bigg)^{n}\Bigg[\frac{M^{(n)}(x^{'},t,\Delta{t})}{n!}\Bigg]p(x,t).
\end{split}$$ Assume that the Taylor series of $M^{(n)}(x^{'},t,\Delta{t})$ is $n!(T^{(n)}(x,t)\Delta{t}+o(\Delta{t}^2))$. Next, dividing both sides of formula by $\Delta{t}$ at the same time and let $\Delta{t}\rightarrow{0}$, $$\label{S1:5}
\begin{split}
\frac{\partial{p(x,t)}}{\partial{t}}=\sum_{n=1}^{\infty}\Bigg(-\frac{\partial}{\partial{x}}\Bigg)^{n}[T^{(n)}(x,t)p(x,t)].\end{split}$$ So far, we have got the Kramers-Moyal expansion and then the integral notation of the SDE is $$\label{S1:6}
\begin{split}
X(t)=x+\int_0^{t}f(X(s))ds+\int_0^t\sigma(X(s))dW_s.
\end{split}$$ Suppose that $f,\sigma$ have the expansion $$\label{S1:7}
\begin{split}
&f(X(t))=f(x)+f^{'}(x)(X(t)-x)+\cdots\\
&\sigma(X(t))=\sigma(x)+\sigma^{'}(x)(X(t)-x)+\cdots.
\end{split}$$ Substituting into , $$\label{S1:8}
\begin{split}
X(t)-x=&\int_0^{t}{f(x)}ds+\int_0^{t}f^{'}(x)(X(s)-x)ds+\cdots\\
&+\int_0^{t}{\sigma(x){dW_s}}+\int_0^{t}{\sigma^{'}(x)(X(s)-x){dW_s}}+\cdots\\
=&\int_0^{t}{f(x)}ds+\int_0^{t}f^{'}(x)\int_0^s{f(x_{u})d{u}}ds+\int_0^{t}f^{'}(x)\int_0^s{\sigma(x_{u})dW_{u}}ds+\cdots\\
&+\int_0^{t}{\sigma(x){dW_s}}+\int_0^{t}{\sigma^{'}(x)\int_0^s{f(x_{u})d{u}}{dW_s}}+\int_0^{t}{\sigma^{'}(x)\int_0^s{\sigma(x_{u})dW_{u}}{dW_s}}+\cdots,
\end{split}$$ we then obtain $$\label{S1:9}
\begin{split}
\mathbb{E}[X(t)-x]&=tf(x)+\frac{t^2}{2}f^{'}(x)f(x)+\mathbb{E}\Bigg[\int_0^{t}{\sigma^{'}(x)\int_0^s{\sigma(x_{u})dW_{u}}dW_s}\Bigg]\\
&=tf(x)+\frac{t^2}{2}f^{'}(x)f(x)+\sigma^{'}(x)\sigma(x)\mathbb{E}\Bigg[\int_0^{t}{W_sdW_s}\Bigg].
\end{split}$$ In the sense of It$\mathrm{\hat{o}}$, the expectation $\mathbb{E}[\int_0^{t}{W_sdW_s}]=0$. Therefore, $$\label{S1:10}
\begin{split}
T^{(1)}=\lim_{t\rightarrow{0}}\frac{1}{t}\mathbb{E}[X(t)-x]=f(x).
\end{split}$$ In addition, $$\label{S1:11}
\begin{split}
T^{(2)}=&\frac{1}{2}\lim_{t\rightarrow{0}}\frac{1}{t}{\mathbb{E}[(X(t)-x)^2]}\\
=&\frac{1}{2}\lim_{t\rightarrow{0}}\frac{1}{t}\mathbb{E}\Bigg[{\Bigg(\int_0^{t}{f(x)}ds+\int_0^{t}f^{'}(x)\int_0^s{f(x_{u})d{u}}ds+\int_0^{t}f^{'}(x)\int_0^s{\sigma(x_{u})dW_{u}}ds+\cdots}\\
&{+\int_0^{t}{\sigma(x){dW_s}}+\int_0^{t}{\sigma^{'}(x)\int_0^s{f(x_{u})d{u}}{dW_s}}+\int_0^{t}{\sigma^{'}(x)\int_0^s{\sigma(x_{u})dW_{u}}{dW_s}}+\cdots}\Bigg)^2\Bigg]\\
=&\frac{1}{2}\lim_{t\rightarrow{0}}\frac{1}{t}\Bigg(t^2f^2(x)+\frac{t^4}{4}(f^{'}(x))^{2}f^2(x)+(f^{'}(x))^{2}\sigma^2(x)\mathbb{E}\Bigg(\int_0^tW_sds\Bigg)^2+\sigma^2(x)\mathbb{E}(W_t^2)\\
&+t^2(\sigma^{'}(x))^2f^2(x)\mathbb{E}(W_t^2)+(\sigma^{'}(x))^2\sigma^2(x)\mathbb{E}\Bigg(\int_0^tW_sdW_s\Bigg)^2+\cdots\\
&+\frac{t^3}{2}f^{2}(x)f^{'}(x)+tf(x)f^{'}(x)\sigma(x)\mathbb{E}\Bigg(\int_0^tW_sds\Bigg)+f^{'}(x)\sigma^2(x)\mathbb{E}\Bigg(W_t\int_0^tW_sds\Bigg)\\
&+tf^{'}(x)\sigma(x)\sigma^{'}(x)f(x)\mathbb{E}\Bigg(W_t\int_0^tW_sds\Bigg)+t\sigma(x)\sigma^{'}(x)f(x)\mathbb{E}(W_t^2)+\cdots\Bigg)\\
=&\frac{1}{2}\sigma^{2}(x).
\end{split}$$ Similarly, $T^{(n)}=\frac{1}{n!}\lim_{t\rightarrow{0}}\frac{1}{t}{\mathbb{E}[(X(t)-x)^n]}=0,~n\geq{3}$. So, the Kramers-Moyal formula is obtained.
**S2. The conditional density function**
Suppose that SDE has a unique strong solution that has a strictly positive probability density, and the conditional density function of the solution also exists. In fact, we know that the value of the conditional density function is only related to the immediately preceding moment through the Markov property of SDE , therefore, for $0<t<t_f$, we have $$\label{p8}
\begin{split}
p(X(t_f)=x_{f}|X(0)=x_0;X(t)=x)&=p(X(t_f)=x_{f}|X(t)=x)\\
&=Q(x_{f},t_f|x,t).
\end{split}$$ In addition, if conditions $X(0)=x_0$ and $X(t_f)=x_{f}$ are given, according to Bayesian formula, we can calculate the conditional density function of the solution $$\label{p9}
\begin{split}
p(X(t)&=x|X(0)=x_0;X(t_f)=x_{f})\\
&= \frac{p(X(t)=x;X(0)=x_0; X(t_f)=x_{f})}{p(X(0)=x_0;X(t_f)=x_{f})}\\
&= \frac{p(X(t)=x|X(0)=x_0)p(X(t_f)=x_{f}|X(t)=x)}{p(X(t_f)=x_{f}|X(0)=x_0)}.
\end{split}$$ We then substitute \[p8\] into \[p9\], and the formal expression of conditional density function $\mathcal{P}_{A}(x, t)$ will be obtained $$\label{p10}
\begin{split}
\mathcal{P}_{A}(x,t)&=p(X(t)=x|X(0)=x_0;X(t_f)=x_{f})\\
&=\frac{Q(x_{f},t_f|x,t)Q(x,t|x_0,0)}{Q(x_{f},t_f|x_0,0)} .
\end{split}$$
References {#references .unnumbered}
==========
[99]{} S. L. Brunton, J. L. Proctor, J. N. Kutz. Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proc. Natl. Acad. Sci. U.S.A. 113:3932-3937, 2016. S. H. Rudy, S. L. Brunton, J. L. Proctor, J. N. Kutz. Data-driven discovery of partial differential equations. Sci. Adv. 3: e1602614, 2017. S. Rudy, A. Alla, S. L. Brunton, J. N. Kutz. Data-Driven Identification of Parametric Partial Differential Equations. SIAM J. Applied Dynamical Systems, 18(2): 643-660, 2019. N. M. Mangan, S. L. Brunton, J. L. Proctor, J. N. Kutz. Inferring Biological Networks by Sparse Identification of Nonlinear Dynamics. IEEE Tansactions on molecular, biological, and multi-scale communications, 2(1): June 2016. S. Zhang, G. Lin. Robustdata-driven discovery of governing physical laws with error bars. Proc. R. Soc. A 474: 20180305, 2018. A. J. Chorin, F. Lu. Discrete approach to stochastic parametrization and dimension reduction in nonlinear dynamics. Proc. Natl. Acad. Sci. U.S.A. 112(32): 9804-9809, 2015. O. Yaira, R. Talmona, R. R. Coifmanb, I. G. Kevrekidisc. Reconstruction of normal forms by learning informed observation geometries from data. Proc. Natl. Acad. Sci. U.S.A. 114(38): E7865-E7874, 2017. M. O. Williams, I. G. Kevrekidis, C.W. Rowley. A Data-Driven Approximation of the Koopman Operator: Extending Dynamic Mode Decomposition. Journal of Nonlinear Science, 25(6):1307-1346, 2015. P. J. Schmid. Dynamic Mode Decomposition of numerical and experimental data. Journal of Fluid Mechanics, 656(10):5-28, 2008. C. Archambeau, D. Cornford, M. Opper, J. S-Taylor. Gaussian Process Approximations of Stochastic Differential Equations. JMLR: Workshop and Conference Proceedings 1: 1-16, 2007. A. Ruttor, P. Batz, M. Opper. Approximate Gaussian process inference for the drift of stochastic differential equations. International Conference on Neural Information Processing Systems Curran Associates Inc, 2013. P. Batz, A. Ruttor, M. Opper. Approximate Bayes learning of stochastic differential equations. arXiv preprint arXiv: 1702.05390v1, 2017. L. Boninsegna, F. N$\ddot{u}$ske, C. Clementi. Sparse learning of stochastic dynamical equations. J. Chem. Phys. 148, 241723, 2018. R. G. Garciaa, R. R. Martinez, I. G. Kevrekidis. Identification of distributed parameter systems: A neural net based approach. Computers and Chemical Engineering, 22(1): 965-968, 1998. H. Schaeffera, R. Caflischa, C. D. Hauckb, S. Oshera. Sparse dynamics for partial differential equations. Proc. Natl. Acad. Sci. U.S.A. 110(17): 6634-6639, 2013. D. Wu, M. Fu and J.Duan. Discovering mean residence time and escape probability from data of stochastic dynamical systems. Chaos, 29, 093122, 2019. Y. Zheng, X. Sun. Governing equations for probability densities of stochastic differential equations with discrete time delays. Discrete and Continuous Dynamical Systems-B, 22(9):3615, 2017. J. Duan. An Introduction to Stochastic Dynamics, volume 51. Cambridge University Press, 2015. T. Gao, J. Duan, X. Kan, Z. Cheng. Dynamical inference for transitions in stochastic systems with $\alpha$-stable L$\mathrm{\acute{e}}$vy noise. J. Phys. A: Math. Theor, 49, 294002, 2016. Z. Schuss. Theory and Applications of Stochastic Processes. volume 170. Applied Mathematical Sciences, 2010. B. Oksendal. Stochastic Differential Equation (6nd edition). New York: Springer-Verlag, 2005.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The perturbative QCD expansion for $\J$ photoproduction appears to be unstable: the NLO correction is large (and of opposite sign) to the LO contribution. Moreover, the predictions are very sensitive to the choice of factorization and renormalization scales. Here we show that perturbative stability is greatly improved by imposing a $`Q_0$ cut’ on the NLO coefficient functions; a cut which is required to avoid double counting. $Q_0$ is the input scale used in the parton DGLAP evolution. This result opens the possibility of high precision exclusive $\J$ data in the forward direction at the LHC being able to determine the low $x$ gluon distribution at low scales.'
---
MPP-2016-303\
IPPP/16/87\
LTH 1102\
\
**The exclusive $J/\psi$ process at the LHC\
**
tamed to probe the low $x$ gluon
$^{a}$, [A.D. Martin]{}$^b$, [ M.G. Ryskin]{}$^{b, c}$ and [T. Teubner]{}$^{d}$\
$^a$ [*Max-Planck-Institute for Physics, Föhringer Ring 6, 80805 München, Germany*]{}\
$^b$ [*Institute for Particle Physics Phenomenology, Durham University, Durham DH1 3LE, U.K.*]{}\
$^c$ [*Petersburg Nuclear Physics Institute, NRC Kurchatov Institute, Gatchina, St. Petersburg, 188300, Russia*]{}\
$^d$ [*Department of Mathematical Sciences, University of Liverpool, Liverpool L69 3BX, U.K.*]{}\
Introduction
============
It would be valuable to be able to constrain the gluon parton distribution function (PDF) at low $x$ using $\J$ photoproduction data measured at HERA and at the LHC, via exclusive $pp \to p+\J+p$ events, especially events in the forward region measured by the LHCb collaboration. Indeed, for LHCb kinematics at 13 TeV we can reach down to $x\simeq 3\times 10^{-6}$. Exclusive $\J$ production is driven by the subprocess $\gamma^*p\to \J +p$, see Fig. \[fig:subproc\].
\[h\]
![ $d\sigma(pp\to p+\J+p)/dy$ driven by the subprocess $\gamma p\to \J+p$ at two different $\gamma p$ centre-of-mass energies, $W_\pm$. []{data-label="fig:subproc"}](wplus "fig:"){width="40.00000%"} ![ $d\sigma(pp\to p+\J+p)/dy$ driven by the subprocess $\gamma p\to \J+p$ at two different $\gamma p$ centre-of-mass energies, $W_\pm$. []{data-label="fig:subproc"}](wminus "fig:"){width="40.00000%"}
Unfortunately, it turns out that the NLO corrections calculated in the conventional $\MSbar$ collinear approach are found to be very large and to depend strongly on the choice of factorization and renormalization scales [@ISSK; @IPSW; @JMRT]. Indeed, for an ‘optimum’ choice of scales it is found that the NLO correction has the opposite sign to the LO contribution and even changes the sign of the whole amplitude, see the continuous curves in Fig. \[fig:f1\]. Thus one may doubt the convergence of the whole perturbation series.
Optimum scale \[sec:opt\]
--------------------------
What do we mean by the ‘optimum’ scale? It was shown in Ref. [@JMRT] that it is possible to find a scale (namely $\mu_F=m_c$) which resums all the double logarithmic corrections enhanced by large values of ${\rm ln}(1/\xi)$ into the gluon and quark PDFs, where $\xi$ is the skewedness parameter of the Generalised Parton Distributions (GPDs) describing the proton-gluon (and proton-quark) vertices. That is, it is possible to take the $(\alpha_S{\rm ln}(1/\xi){\rm ln}(\mu_F^2))$ term from the NLO gluon (and quark) coefficient functions and to move it to the LO GPDs. This allows a resummation of all the double log $(\alpha_S{\rm ln}(1/\xi){\rm ln}(\mu_F^2))^n$ terms in the LO contribution by choosing the factorization scale to be $\mu_F=m_c$. The details are given in Ref. [@JMRT], see also Ref. [@DY].
The result is that the $\gamma p\to \J+p$ amplitudes are schematically of the form A(\_f) = C\^[LO]{} (\_F) + C\^[NLO]{}\_[rem]{}(\_F)(\_f), where the GPD can be related to the conventional PDF via the Shuvaev transform for $\xi<|x|\ll 1$ [@Shuv]. With the choice $\mu_F=m_c$ there is a smaller remaining term in the NLO coefficent funcions, and so the residual dependence on the scale $\mu_f$ is reduced.
Unfortunately, even after this, the NLO corrections, and their variations with scale, although reduced, are still unacceptably large, as shown in Fig. \[fig:f1\]. The dashed and dot-dashed curves correspond to NLO predictions for two different values of the residual scale $\mu_f$: namely $\mu_f^2=$ 4.8 and 1.7 GeV$^2$ respectively, while the continuous curves correspond to the ‘optimum’ scale choice $\mu_F^2=\mu^2_R=m_c^2=M_\psi^2/4 =2.4 $ GeV$^2$. [^1] The choice $\mu_R=\mu_F$ is justified in subsection \[sec:scales\].
\[t\]
Double counting
---------------
So for the QCD prediction to be useful we should search for some other sizeable physical contribution to the NLO correction. Here we consider a power correction which may further reduce the NLO correction and, moreover, may reduce the sensitivity to the choice of scale. The correction is ${\cal O}(Q_0^2/M^2_\psi)$ where $Q_0$ denotes the input scale in the parton evolution. It turns out to be important for the relatively light charm quark, $m_c \simeq M_\psi /2$. Let us explain the origin of this ‘$Q_0$ correction’. We begin with the collinear factorization approach at LO. Here, we never consider parton distributions at low virtualities, that is for $Q^2<Q_0^2$. We start the PDF evolution from some phenomenological PDF input at $Q^2=Q_0^2$. In other words, the contribution from $|l^2|<Q^2_0$ of Fig. \[fig:f2\](b) (which can be considered as the LO diagram, Fig. \[fig:f2\](a), supplemented by one step of DGLAP evolution from quark to gluon, $P_{gq}$) is already included in the input gluon GPD at $Q_0$. That is, to avoid double counting, we must exclude from the NLO diagram the contribution coming from virtualities less than $Q_0^2$. At large scales, $Q^2\gg Q^2_0$ this double-counting correction will give small power suppressed terms of ${\cal O}(Q_0^2/Q^2)$, since there is no infrared divergence in the corresponding integrals. On the other hand, with $Q_0 \sim 1$ GeV and $\mu_F=m_c~ (\sim M_\psi /2$) a correction of ${\cal O}(Q^2_0/m^2_c)$ may be crucial.
In the present paper we re-calculate the NLO contribution for $\J$ photoproduction excluding the contribution coming from the low virtuality domain $(<Q^2_0)$. We find that for $\J$ this procedure substantially reduces the resulting NLO contribution and, moreover, reduces the scale dependence of the predictions. It indicates the convergence of the perturbative series.
An outline of the procedure is given in [@Diff2016], where also the NLO description of exclusive $\J$ production in the $k_T$ factorization and collinear factorization schemes are compared.
\[t\]
![(a) LO contribution to $\gamma p \to V +p$. (b) NLO quark contribution. For these graphs all permutations of the parton lines and couplings of the gluon lines to the heavy-quark pair are to be understood. Here $P\equiv (p+p^\prime)/2$ and $l$ is the loop momentum. []{data-label="fig:f2"}](gluonlo "fig:"){width="40.00000%"} ![(a) LO contribution to $\gamma p \to V +p$. (b) NLO quark contribution. For these graphs all permutations of the parton lines and couplings of the gluon lines to the heavy-quark pair are to be understood. Here $P\equiv (p+p^\prime)/2$ and $l$ is the loop momentum. []{data-label="fig:f2"}](quarknlo "fig:"){width="40.00000%"}
Avoiding double counting of the low $Q^2$ contribution
======================================================
The NLO quark contribution
--------------------------
We start with the NLO quark contribution to the $\gamma p \to \J +p$ process. The corresponding Feynman diagrams are that of Fig. \[fig:f2\](b) together with the diagram where both gluons couple to the same heavy quark line. Here we will use the non-relativistic approximation for the $\J$ wave function. Since the momentum fractions $(x+\xi)$ and $(x-\xi)$ carried by the left and right quarks are different we have to use the skewed (generalized) parton distribution (GPD), $F_q(x,\xi,Q^2)$. The skewedness parameter $\xi=M_\psi^2/(2W^2-M_\psi^2)$, where $W$ is the $\gamma p$ energy. We see that the upper part of diagram Fig. \[fig:f2\](b) is the same as the diagram for the LO gluon Fig. \[fig:f2\](a) contribution. For the LO contribution the integral over the gluon virtuality $|l^2|$ starts from the input scale $Q^2_0$, while all the contributions from low virtualities $|l^2|<Q^2_0$ are collected in the input gluon GPD, $F_g(x,\xi,Q_0^2)$. Note that this input distribution already includes that part of the quark contribution of Fig. \[fig:f2\](b) coming from $|l^2|<Q^2_0$. Thus to avoid double counting when computing the NLO quark coefficient function, $C_q^{\rm NLO}$, of Fig. \[fig:f2\](b) we have to include the theta function $\Theta(|l^2|>Q^2_0)$ in the integration over $l^2$. Depending on the ratio $Q^2_0/m^2_c=4Q^2_0/M^2_\psi$ this can be a significant correction. The corresponding integral has no infrared or ultraviolet divergence and can be calculated in $D=4$ dimensions.
Actually, the calculation is performed in the physical scheme (with $D=4)$. On the other hand, parton distributions are usually presented in the $\MSbar$ factorization scheme where dimensional regularization is used. The problem is that when we calculate the coefficient function in $D=4+2\epsilon$ we have finite contributions of $\epsilon/\epsilon$ origin. Formally these $\epsilon/\epsilon$ terms come from unphysically large distances $\propto {\cal O}(1/\epsilon)$. In fact, these $\epsilon/\epsilon$ terms are compensated by a corresponding re-definition of the PDFs. In order to retain the $\epsilon/\epsilon$ terms and to use the $\MSbar$ scheme we do not calculate diagram \[fig:f2\](b) in $D=4$ dimensions for $|l^2|>Q^2_0$, but instead calculate the part corresponding to small $|l^2|<Q^2_0$. We consider this part as the correction which should be subtracted from the known NLO $\MSbar$ coefficient function [@ISSK; @SJonesThesis]. Recall that after the subtraction of the contribution generated by the last step of the LO evolution, $P^{\rm LO}\otimes C^{\rm LO}$, there is no infrared divergence and the subtracted part of $C^{\rm NLO}$ coming from $|l^2|<Q_0^2$ does not contain $\epsilon/\epsilon$ terms.
The NLO gluon contribution
--------------------------
The NLO ‘$Q_0$ corrections’ for the gluon coefficient function are more complicated. Besides the ladder-type diagrams analogous to Fig. \[fig:f2\](b), but with the light quark line replaced by a gluon line, there are other diagrams which have a structure similar to vertex corrections, see [@ISSK; @SJonesThesis]. However the ‘dangerous’ contribution is again from the ladder-type diagrams, where to avoid double counting we have to exclude the $|l^2|<Q^2_0$ domain whose contribution is already included in the LO term using the input gluon GPD, $F_g(x,\xi,Q^2_0)$. Qualitatively this is exactly the same calculation as that for the NLO quark. The only difference is that the lower line in the diagrams of Fig. \[fig:f4\] is now replaced by a gluon line and the lower part of the diagram is now given by the product of two three-gluon vertices averaged over the incoming gluon transverse polarizations. The notation is identical to that for the quark contribution. Both the quark- and the gluon-induced contributions are determined as described in the Appendix. They involve the calculation of the diagrams of Fig. \[fig:f4\] (given in the Appendix), and the analogous diagrams for the gluon-induced contribution.
Results
=======
\[t\]
Fig. \[fig:f5\] shows the LO and NLO contributions to the imaginary part of the $\J$ photoproduction amplitude when the $Q_0$ cut in the NLO contribution is taken into account. It should be compared to Fig. \[fig:f1\] which had exactly the same scale choices, but without the $Q_0$ cut imposed. The improvement in going from Fig. \[fig:f1\] to Fig. \[fig:f5\] is dramatic. First, the NLO contribution is now much smaller than the LO contribution. Second, the scale variation is much smaller. The continuous curves in Figs. \[fig:f1\] and \[fig:f5\] show the LO and NLO comparison for the choice of scales $\mu_F=\mu_R=m_c\equiv M_\psi/2$, which we had previously argued was optimal [@JMRT]. The stability achieved by imposing the $Q_0$ cut means that $J/\psi$ photoproduction ($\gamma p \to J/\psi ~p$) data and LHC exclusive $J/\psi$ ($pp \to p+J/\psi+p$) data can now be included in the global parton analyses.
The choice of scales \[sec:scales\]
-----------------------------------
Let us discuss the above scale choices in more detail. By choosing the ‘optimal’ factorization scale $\mu_F=m_c$ we resum all the higher-order double-logarithmic corrections $(\alpha_s\ln(1/\xi)\ln\mu^2_F)^n$ (enhanced at high energies by the large value of $\ln(1/\xi)$) into the gluon generalized parton distribution (gluon GPD) [@JMRT].
The renormalization scale is taken to be $\mu_R=\mu_F$. The arguments are as follows. First, this corresponds to the BLM prescription [@BLM]; such a choice eliminates from the NLO terms the contribution proportional to $\beta_0$ (i.e. the term $\beta_0\ln(\mu^2_R/\mu^2_F)$ in eq. (3.95) of [@ISSK]). Second, following the discussion in [@LHKR] for the analogous QED case, we note that the new quark loop insertion into the gluon propagator appears twice in the calculation. The part with scales $\mu <\mu_F$ is generated by the virtual component ($\propto \delta(1-z)$) of the LO splitting during DGLAP evolution, while the part with scales $\mu>\mu_R$ accounts for the running $\alpha_s$ behaviour obtained after the regularization of the ultraviolet divergence. In order not to miss some contribution and/or to avoid double counting we take the renormalization scale equal to the factorization scale, $\mu_R=\mu_F$.
Discussion of the results
-------------------------
Note that in the present paper we have calculated the imaginary part of the $\gamma p \to \J~p$ amplitude. The real part of the amplitude can be restored via dispersion relations assuming positive signature, as in eq. (5) of Ref. [@JMRT1311]. Recall that we obtain the necessary GPDs from the CTEQ6.6 parton set [@Nadolsky:2008zw] using the Shuvaev transform [@Shuv]. We use a relatively old parton set [@Nadolsky:2008zw] in which the low $x$ gluons are forced to be positive so as to make a meaningful comparison with our earlier work. The goal of this paper is not to make a quantitative description of the data, but to demonstrate that we can achieve stability of the perturbative QCD description of relatively low scale $\J$ production by imposing the $Q_0$ cut. We have shown this is a power correction – a correction which is needed to avoid double counting. This will allow future high precision exclusive $\J$ production data obtained at the LHC to be incorporated in global parton analyses.
The general procedure to include the HERA $\gamma p \to \J~p$ data and, in particular, the LHCb data for exclusive $\J$ production, $pp \to p+\J+p$, in a global analysis follows that used to produce Fig. 4 of Ref. [@JMRT1311]. These processes are driven by the gluon PDF and the LHCb data probe the gluon at very low values of $x$. However, in Ref. [@JMRT1311] we [*approximated*]{} the NLO corrections to the coefficient functions by accounting for the explicit $l_\perp$ integration in the last step of the interaction. Moreover, we just fitted the $\J$ data and used a parametric form for the gluon which approximated its $x$ and $Q^2$ dependence. So the analysis of Ref. [@JMRT1311] was quite simplified, although very informative; see, for example, Fig. 5 of [@JMRT1311] which compared the resulting gluon PDF with those of different global analyses[^2].
The present paper, on the other hand, retains collinear factorization and calculates the complete NLO contribution. We may expect the high $\gamma p$ energy, $W$, data points in the updated version of Fig. 4 of Ref. [@JMRT1311] to require a larger gluon distribution in the region from $x \lapproxeq 10^{-3}$ down to $10^{-5}$, at low scales, than coming from [*extrapolations*]{} of the NLO gluon PDFs from global fits to data not including the $\J$ data. An indication in favour of a larger gluon PDF in this domain comes also from the recent LHCb data on open charm (and beauty) [@LHCbg].
Finally, it is useful to compare our approach with that of [@IPSW16], where it was demonstrated that the re-summation of the BFKL-induced $(\alpha_S\ln(1/\xi))^n$ terms in the coefficient functions additionally reduces the factorization scale dependence. Recall that our choice of $\mu_F=M_\psi/2$ resums only the [*double logarithmic*]{}, $(\alpha_S\ln(1/\xi)\ln\mu_F)^n$ contributions[^3]. The remaining part, which does not contain $\ln\mu_F$, should be considered, in the collinear factorization approach, as higher-order, NNLO, N$^3$LO, ... corrections. Of course, it would be good to account for these corrections as well. However, to properly calculate these corrections one has to exclude the low ($<Q^2_0$) virtuality contribution. Otherwise we will face the problem of double counting again. The present paper shows these (power) corrections (necessary to avoid double counting) are crucial to achieve perturbative stability.
\[t\]
![Two diagrams (a,b) computed for the NLO quark coefficient function. Note that $p$ and $p'$ refer to the incoming and outgoing quark lines. In the corresponding diagrams computed for the NLO gluon coefficient function the light quark line is replaced by a gluon. The other two diagrams of the different coupling of the two $t$-channel gluons to the heavy quarks are implicitly included.[]{data-label="fig:f4"}](quarkcoeffnlo1 "fig:"){width="35.00000%"} ![Two diagrams (a,b) computed for the NLO quark coefficient function. Note that $p$ and $p'$ refer to the incoming and outgoing quark lines. In the corresponding diagrams computed for the NLO gluon coefficient function the light quark line is replaced by a gluon. The other two diagrams of the different coupling of the two $t$-channel gluons to the heavy quarks are implicitly included.[]{data-label="fig:f4"}](quarkcoeffnlo2 "fig:"){width="35.00000%"}
Appendix {#appendix .unnumbered}
========
Here we describe the calculation of the piece that we subtract from the full result. Only the imaginary part of the ladder-type cut diagrams shown in Fig. \[fig:f4\] and the corresponding diagrams where the light-quark line is replaced by gluons is computed.
All momenta appearing in the calculation may be decomposed in terms of light-like momenta $p, n$ and a transverse four-momentum $l_\perp$, $$\begin{aligned}
& l^\mu = \beta p^\mu + \alpha n^\mu + l_\perp^\mu, &
&h_1^\mu = h_2^\mu = \beta_h p^\mu + \alpha_h n^\mu, &\end{aligned}$$ where $l$ is the loop momentum and $h_1, h_2$ are the momenta of the outgoing heavy quark and heavy anti-quark, respectively. Here $p$ can be chosen as the momentum of the incoming light parton and $n$ the momentum of the incoming on-shell photon. With this convention we have $$\begin{aligned}
& p \cdot p = n \cdot n = 0, & & p \cdot n = \hat{s}/2, & & p \cdot l_\perp = n \cdot l_\perp = 0, &\end{aligned}$$ where $\hat{s}$ is the photon-parton centre-of-mass energy squared. The four momenta of the incoming and the outgoing light partons are proportional. We may write $p_\mu$ and $ p'_\mu=Xp_\mu$ with X===, [where]{} y==-1. \[eq:X\]
To leading order in the heavy quark relative velocity, the S-wave spin-triplet component of $\J$ can be computed using the projection [@Petrelli:1997ge; @Bodwin:2002hg; @Braaten:2002fi] $$v_\alpha(h_2) \bar{u}_\beta(h_1) \rightarrow
N_{\J}
\left[ (\slashed{h}_2 - m_c) \slashed{\epsilon}_{\J}^* (\slashed{K} + M_\psi) (\slashed{h}_1 +m_c) \right]_{\alpha \beta}.$$ Here $\bar{u},v$ are the spinors of the outgoing heavy quark and anti-quark which form the $\J$. The indices $\alpha$ and $\beta$ label their spin. $N_{\J}$ is an overall factor which contains the non-perturbative NRQCD matrix element describing the $\J$ formation. The vector $\epsilon_{\J}$ describes the polarisation of the $\J$ with momentum $K=h_1+h_2$ and mass $M_\psi=2 m_c$.
The projections onto the quark and gluon GPDs are given by [@Ji:1996ek; @Ji:1998xh; @Radyushkin:1996ru; @Radyushkin:1997ki], $$\begin{aligned}
&u_\alpha(p) \bar{u}_\beta(p^\prime) \rightarrow N_q \slashed{p}_{\alpha \beta}~,&
&\epsilon_1^{\mu} {\epsilon^*_2}^{\nu} \rightarrow N_g g_\perp^{\mu \nu} = N_g \left( g^{\mu \nu} - \frac{2}{\hat{s}} p^\mu n^\nu - \frac{2}{\hat{s}} n^\mu p^\nu \right),&\end{aligned}$$ respectively. Here $u, \bar{u}$ are the spinors of the light quarks connected to the quark GPD and $\epsilon_1$, $\epsilon^*_2$ are the polarisation vectors of gluons connected to the gluon GPD. $N_q$, $N_g$ are overall factors containing the quark and gluon GPDs.
The on-shell conditions $h_1^2 - m_c^2 = 0$ and $h_2^2 - m_c^2 = 0$ for outgoing heavy quarks and the cut-constraints, $(p-l)^2 = 0$ and $(n-h_2+l)^2 -m_c^2 = 0$ for Fig. \[fig:f4\] diagram (a), $(p-l)^2=0$ and $(h_1-l-n)^2 - m_c^2 = 0$ for Fig. \[fig:f4\] diagram (b), allow us to choose $\alpha_h = 1/2, \beta_h = 2 m^2/\hat{s}$ and fix $\alpha, \beta$ in terms of $l_\perp^2, \hat{s}, m_c$. Specifically, $$\begin{aligned}
&\beta=4(1+l^2/\hat{s})m_c^2/\hat{s}-2l^2/\hat{s},&
&\alpha=l^2/\hat{s},& \label{eq:beta} \\
&l^2=l^2_\perp/(1-\beta),&
&l^{'2}=l^2(1-4m_c^2/\hat{s}).&\end{aligned}$$ Additionally, we obtain $q^2=-m_c^2$ for diagram \[fig:f4\](a) and q\^2=l\^2\_-(-l\^2)(-) = 3m\_c\^2- for diagram \[fig:f4\](b).
In our calculation we split each diagram of Fig. \[fig:f4\] into two parts. An “upper” part which contains a trace over the heavy quark fermion line and a “lower” part which in the quark channel contains a trace over the light quark line and in the gluon case consists of two triple gluon vertices contracted with $g_\perp^{\mu \nu}$.
First we discuss the “upper” part which is different for the diagrams (a) and (b) of Fig. \[fig:f4\] but identical for the quark and gluon channels. Where it appears, we replace the contraction of $l_\perp$ with the polarisation vectors using $$(l_\perp \cdot \epsilon^*_{\J}) (l_\perp \cdot \epsilon_\gamma) = (\epsilon_\gamma \cdot \epsilon^*_{\J}) l_\perp^2/2$$ which follows from tensor decomposing the $l_\perp$ integral after the integration over the $\vec{l}$ azimuthal angle. We can simplify the calculation by noting that the sum of the “upper” parts of diagrams (a) and (b) obey the gauge condition $$\mbox{T(h.loop)}^{\mu\nu}l_\mu=\mbox{T(h.loop)}^{\mu\nu}l'_\nu=0 , \label{eq:gaugecondition}$$ where \^ = \_a\^ + \_b\^. Here T(h.loop) is the upper part of the amplitude, which besides the trace over the quark loop, includes the heavy quark propagator $1/(q^2-m^2_c)$.
Using the gauge condition the only contractions of the “upper” part that appear in the sum of diagrams are )\^\_ag\_ &=& N\_ 4m\_c(\_\^\*\_[J/]{})(6m\_c\^2-),\
)\^\_ap\_p\_&=& N\_ 4m\_c(\_\^\*\_[J/]{})\^2(1/2+)/2,\
)\^\_ap\_l\_ &=& N\_ 4m\_c(\_\^\*\_[J/]{})l\^2\_/2,\
)\^\_al\_ p\_&=& N\_ 4m\_c(\_\^\*\_[J/]{})l\^2\_/2,\
)\^\_bg\_ &=& N\_ 4m\_c(\_\^\*\_[J/]{})2(-m\_c\^2(2+1)),\
)\^\_bp\_p\_&=& - N\_ 4m\_c(\_\^\*\_[J/]{})\^2/4,\
)\^\_bp\_l\_ &=& - N\_ 4m\_c(\_\^\*\_[J/]{})l\^2\_(-l\^2)/2,\
)\^\_bl\_ p\_&=& N\_ 4m\_c(\_\^\*\_[J/]{})l\^2\_(+l\^2)/2. The contractions involving $p^\mu n^\nu, n^\mu p^\nu, n^\mu n^\nu, n^\mu l_\perp^\nu, l_\perp^\mu n^\nu, l_\perp^\mu l_\perp^\nu$ appear in the computation of individual diagrams but cancel for the sum of diagrams.
Quark-induced NLO correction {#quark-induced-nlo-correction .unnumbered}
----------------------------
For an unpolarized light quark the trace over the “lower” light quark line gives A\^q\_= 4 N\_q \[p\_(p-l)\_+(p-l)\_p\_+g\_(pl)\] , \[a1\] where the normalization factor $$N_q= C_F F_q(x,\xi,\mu_F)$$ includes the colour factor $C_F$ and the quark GPD, $F_q$.
This light quark part should be contracted with the trace, ${\rm Tr(h.loop)}^{\mu \nu}$, given by the heavy quark (upper) loop. Due to the gauge condition [(\[eq:gaugecondition\])]{} we have that $(p-l)_\mu$ acts as $p_\mu$, while $(p-l)_\nu$ acts as $p'_\nu=Xp_\nu$ giving $$\begin{aligned}
M^q_a &= \frac{4 N_q}{(-2m_c^2) \, l^2 {l^\prime}^2} \left[ \mbox{Tr(h.loop})^{\mu\nu}_ag_{\mu\nu} \left( \frac{\alpha \hat{s}}{2} \right) + \mbox{Tr(h.loop})^{\mu\nu}_ap_\mu p_\nu (1 + X) \right] + \overline{M}^q \\
&= \frac{4 N_q N_{\J} (2m_c) (\epsilon_\gamma \cdot \epsilon^*_{J/\psi}) }{(-2m_c^2) \, l^2 {l^\prime}^2} \left[(6m_c^2-\hat{s}\beta) \alpha \hat{s} + \hat{s}^2(1/2+\alpha)(1+X) \right] + \overline{M}^q, \label{a2}\end{aligned}$$ for diagram (a) and $$\begin{aligned}
M^q_b &= \frac{4 N_q}{(2 m_c^2 - \beta \hat{s}) \, l^2 {l^\prime}^2} \left[ \mbox{Tr(h.loop})^{\mu\nu}_bg_{\mu\nu} \left( \frac{\alpha \hat{s}}{2} \right) + \mbox{Tr(h.loop})^{\mu\nu}_bp_\mu p_\nu (1 + X) \right] - \overline{M}^q \\
&= \frac{4 N_q N_{\J} m_c (\epsilon_\gamma \cdot \epsilon^*_{J/\psi})}{(2 m_c^2 - \beta \hat{s}) \, l^2 {l^\prime}^2}\ [4 (\hat{s}\alpha\beta-m_c^2(2\alpha+1)) \alpha \hat{s} - (1+X) \hat{s}^2] - \overline{M}^q, \label{b2}\end{aligned}$$ for diagram (b). The term $\overline{M}^q$ accounts for terms which cancel between the two diagrams. The denominators come from the uncut propagators: $1/l^2$ for the left, and $1/l^{'2}$ for the right gluon and $1/(q^2-m_c^2)$ for the uncut heavy quark propagator.
The result is to be integrated over the gluon transverse momentum ($dl^2_\perp$) while the longitudinal components are fixed by the quark on-mass-shell conditions. It is easy to perform this integral numerically accounting for the condition which was introduced in Section 2 in order to avoid double counting. Recall, however, that we are not going to calculate the whole NLO contribution, but just the correction to the known $\overline{\rm MS}$ coefficient function. So, in order to compute the [*correction*]{}, in the integration over the $l_\perp$ we only consider the region of $|l^2|<Q^2_0$. Actually, we integrate over $dl^2$ directly; the factor $(1-\beta)$ coming from the relation $l^2=l^2_\perp/(1-\beta)$ is exactly cancelled by the residue from the light quark on-mass-shell pole. So we obtain the correction to the quark-induced part of the $\gamma p\to \J+p$ amplitude \^q = \^1\_dx(F\_q(x,,m\_c)-F\_q(-x,,m\_c))(\^[Q\_0\^2]{}\_0 (M\^q\_a+M\^q\_b) dl\^2 ) where the ‘hard matrix elements’ $M^q_{a,b}$ are given by (\[a2\]) and (\[b2\]). The factor $1/\hat{s}^2$ comes from the delta functions needed to put the lower light quark and the heavy quark coupled to the right gluon in Fig. \[fig:f4\] on-mass-shell. The factor $m_c^4$ accounts for the normalization $N_{\J}$, defined to be consistent with the normalization of eqs. (3.93) and (3.95) of [@ISSK] for which the correction was calculated; actually the last factor (...) is the correction to $f_q$ of (3.93) of [@ISSK].[^4] For the gluon correction $\Delta{\cal M}^g$ there is an additional factor $\hat{s}/2m_c^2=1/\xi$ due to the definition of the gluon GPD, $F_g$; see the extra factor of $\xi$ in eq. (3.94) of [@ISSK], see also [@SJonesThesis].
Note that we have explicitly calculated the NLO diagrams (a) and (b) of Fig. \[fig:f4\] which contain both LO [^5] and NLO contributions. To identify the NLO part we therefore have to subtract the contribution generated by the LO evolution equation, which is of the form of the convolution $P^{\rm LO}\otimes C^{\rm LO}$, before we integrate over $l^2_\perp$. This subtraction completely cancels the logarithmic infrared divergence $dl^2/l^2$. Note that the subtraction must be done only in the region of $|l^2|<\mu^2_F$ since at the factorization scale $\mu_F$ the DGLAP evolution stops.[^6] Also note that in the LO approximation the convolution $P^{\rm LO}\otimes C^{\rm LO}$ is larger than the value of the matrix element given by explicit calculation of the diagrams shown in Fig. \[fig:f4\]. Thus the final result has the sign opposite to that for the LO amplitude.
In this way we obtain the quark NLO coefficient function. Since we are looking for the power correction needed to avoid double counting of the low $|l^2|<Q^2_0$ contribution[^7], we actually have to integrate the matrix element $M^q$ over $|l^2|<Q^2_0$ only (as explained above) and to subtract the result from the known NLO coefficient function given in the $\overline{\rm MS}$ scheme.
In the notation of Ref. [@ISSK] this should be considered as the new form of Im$f_q(y)$ of their eq. (3.93), after allowing for the changes made by our introduction of the ‘$Q_0$ cut’.
Gluon NLO correction {#gluon-nlo-correction .unnumbered}
--------------------
In the gluon case the tensor $A^g_{\mu\nu}$ corresponding to the lower part of Fig. \[fig:f4\] diagrams (with the lower quark line replaced by a gluon line) was calculated explicitly. It can be written in the form A\^g\_= N\^g(ag\_+b\_[11]{}p\_p\_+b\_[22]{}h\_h\_+b\_[12]{}p\_h\_+b\_[21]{}h\_p\_+c\_1p\_l\_+d\_1l\_ p\_+c\_2h\_l\_+d\_2l\_ h\_) , \[ag\] where $h_\mu=p_\mu-l_\mu\,$ and a=l\^2(1+X+4(1-)), b\_[11]{}=X(4-2)-4(1-) , $$b_{22}=2,\ b_{12}=2X+4,\ b_{21}=2+4X,\ c_1=3-2X,\ d_1=3X-2,\ c_2=3,\ d_2=3\ .$$ Here the normalization factor is [^8] N\_g = . Note that $X$ is defined in (\[eq:X\]) and $\beta$ is given by (\[eq:beta\]). Recall that we are looking for the imaginary part of the amplitude (i.e. $s$-channel discontinuity).
This expression should be convoluted with the “upper” part of the diagram. The result for the sum of diagrams can again be simplified using the gauge conditions [(\[eq:gaugecondition\])]{}. That is vector $h_\mu=(p-l)_\mu$ acts as $p_\mu$, while $h_\nu$ acts as $p'_\nu=Xp_\nu$.
As before, the result is multiplied by the terms $1/l^2$ and $1/l^{'2}$ from the $t$-channel gluon propagators and by the term $1/(q^2-m_c^2)$ from the corresponding heavy quark propagator. Then we have to subtract the part generated by the LO evolution equation which is given by the convolution $P^{\rm LO}\otimes C^{\rm LO}$. Finally we integrate over $l^2_\perp$, accounting for the condition $|l^2|<Q^2_0$ (the longitudinal components are fixed by the heavy quark and gluon $(p-l)^2=0$ on mass-shell conditions). In this way we obtain the power correction which should be subtracted from the known NLO gluon coefficient function ${\rm Im}f_g(y)$ given by eq. (3.95) of [@ISSK] (see also [@SJonesThesis]), which we then use to obtain the $Q_0$ subtracted NLO gluon contribution.
Acknowledgements {#acknowledgements .unnumbered}
================
We thank Ronan McNulty for interesting discussions and for encouraging us to make these predictions. MGR thanks the IPPP at the University of Durham for hospitality. MGR is supported by the RSCF grant 14-02-00281. SPJ is supported by the Research Executive Agency (REA) of the European Union under the Grant Agreement PITN-GA2012316704 (HiggsTools), and TT is supported by STFC under the consolidated grant ST/L000431/1.
[x]{} D.Yu. Ivanov, A. Schäfer, L. Szymanowski and G. Krasnikov, Eur. Phys. J. [**C34**]{} (2004) 3, 297, [*Erratum ibid.*]{} [**C75**]{} (2015) 2, 75, [*Erratum*]{} [arXiv:hep-ph/0401131v3]{}. D.Yu. Ivanov, B. Pire, L. Szymanowski and J. Wagner, AIP Conf. Proc. [**1654**]{} (2015) 090003.
S.P. Jones, A.D. Martin, M.G. Ryskin and T. Teubner, J. Phys. G [**43**]{} (2016) 035002.
Q. Cao, J. Huston, H. Lai, P.M. Nadolsky, J. Pumplin, W. Tung, D. Stump and C.-P. Yuan, Phys. Rev. [**D78**]{} (2008) 013004. E.G. de Oliveira, A.D. Martin and M.G. Ryskin, Eur. Phys. J. [**C72**]{} (2012) 2069.
A. Shuvaev, K.J. Golec-Biernat, A.D. Martin and M.G. Ryskin, Phys. Rev. D [**60**]{} (1999) 01415.
P. Hoodbhoy, Phys. Rev. [**D56**]{} (1997) 388.
A.D. Martin, C. Nockles, M.G. Ryskin and T. Teubner, Phys. Lett. [**B682**]{} (2008) 252.
S.P. Jones, A.D. Martin, M.G. Ryskin and T. Teubner, [arXiv:1609.09738 \[hep-ph\]]{}.
S.P. Jones, [*A Study of Exclusive Processes to NLO and Small-$x$ PDFs from LHC Data*]{}, PhD thesis, University of Liverpool, September 2014 (unpublished).
S.J. Brodsky, G.P. Lepage and P.B. Mackenzie, Phys. Rev. D 28 (1983) 228.
L.A. Harland-Lang, M.G. Ryskin and V.A. Khoze, Phys. Lett. [**B761**]{} (2016) 20.
S.P. Jones, A.D. Martin, M.G. Ryskin and T. Teubner, JHEP [**11**]{} (2013) 085,\
[arXiv:1307.7099 \[hep-ph\]]{}.
LHCb Collaboration (Roel Aaij et al.), JHEP [**03**]{} (2016) 159, [arXiv:1510.01707 \[hep-ex\]]{}.
D.Yu. Ivanov, B. Pire, L. Szymanowski and J. Wagner, Eur. Phys. J. Web Conf. [**112**]{} (2016) 01020, [arXiv:1601.07338 \[hep-ph\]]{}.
A. Petrelli, M. Cacciari, M. Greco, F. Maltoni and M. L. Mangano, Nucl. Phys. B [**514**]{} (1998) 245, [arXiv:hep-ph/9707223]{}.
G. T. Bodwin and A. Petrelli, Phys. Rev. D [**66**]{} (2002) 094011, Erratum: Phys. Rev. D [**87**]{} (2013) 039902.
E. Braaten and J. Lee, Phys. Rev. D [**67**]{} (2003) 054007, Erratum: Phys. Rev. D [**72**]{} (2005) 099901, [arXiv:hep-ph/0211085]{}.
X. D. Ji, Phys. Rev. Lett. [**78**]{} (1997) 610, [arXiv:hep-ph/9603249]{}.
X. D. Ji and J. Osborne, Phys. Rev. D [**58**]{} (1998) 094018, [arXiv:hep-ph/9801260]{}.
A. V. Radyushkin, Phys. Lett. B [**385**]{} (1996) 333, [arXiv:hep-ph/9605431]{}.
A. V. Radyushkin, Phys. Rev. D [**56**]{} (1997) 5524, [arXiv:hep-ph/9704207]{}.
[^1]: Recall that the choice $m_c=M_\psi/2$ effectively accounts for the relativistic corrections to the $\J$ wave function, see [@Hoodbhoy; @Noc].
[^2]: Recall, however, that strictly speaking the global analyses use the $\overline{\rm MS}$ collinear factorization scheme whereas $k_T$ factorization uses the physicsl scheme, see, for example, [@Diff2016].
[^3]: This result for the optimal scale (see Section \[sec:opt\]) is confirmed by the formula after eq. (8) in [@IPSW16]. Note that, in [@IPSW16], $L(L-\ln 16)+\ln^24=(L-\ln 4)^2=\ln^2(M^2_\psi/4\mu^2_F)$, since $L\equiv {\rm ln}(M^2_\psi/\mu^2_F)$.
[^4]: The overall normalization has been checked against [@ISSK] and correctly reproduces the leading log term $\propto{\rm ln}(4m_c^2/\mu^2_F)$.
[^5]: The integration of the pure logarithmic form $dl^2/l^2$ up to $\mu_F$ actually reproduces the LO contribution already included in Fig. \[fig:f2\](a). On the other hand some non-logarithmic corrections originating from higher powers of $l^2$, together with the whole contribution above $\mu_F$, are NLO $\alpha_s$ corrections which are not enhanced by the large collinear $(l^2)$ logarithms.
[^6]: This is the origin of the $\ln(4m^2/\mu^2_F)$ factor in the first term of $f_q(y)$ of eq. (3.93) of [@ISSK]. Since now we integrate over the $|l^2|<Q^2_0<\mu^2_F$ the correction does not depend on $\mu_F$.
[^7]: This contribution is already included in the input value GPD($Q_0)$.
[^8]: Here the denominator $(x+\xi-i\epsilon)(x-\xi+i\epsilon) $ arises from the particular definition of the gluon GPD.
| {
"pile_set_name": "ArXiv"
} |
---
abstract:
- |
Let $T$ be a self-adjoint operator in a Hilbert space $H$ with domain $\mathcal D(T)$. Assume that the spectrum of $T$ is confined in the union of disjoint intervals $\Delta_k =[\alpha_{2k-1},\,
\alpha_{2k}]$, $k\in \mathbb{Z}$, and $$\alpha_{2k+1}-\alpha_{2k} \geqslant b
|\alpha_{2k+1}+\alpha_{2k}|^p\quad \text{ for some }\, b>0,\,
p\in[0,1).$$Suppose that a linear operator $B$ in $H$ is $p$-subordinated to $T$, i.e. $\mathcal D(B) \supset\mathcal D(T)$ and $\|Bx\| \leqslant
b'\,\|Tx\|^p\|x\|^{1-p} +M\|x\| \text{\, for all } x\in \mathcal
D(T)$, with some $b'>0$ and $M\geq 0$. Then the spectrum of the perturbed operator $A=T+B$ lies in the union of a rectangle in $\mathbb{C}$ and double parabola $P_{p,h} = \bigl\{\lambda \in
\mathbb{C}\, \bigl|\ \ |\mathop{\rm Im} \lambda| \leqslant
h|\mathop{\rm Re} \lambda|^p\bigr\}$, provided that $h>b'$. The vertical strips $\Omega_k =\{\lambda\in\mathbb C\,\vert \ \,
|r_k-{\mathop{\rm Re}}\,\lambda|\leqslant \delta r_k^p\}$, $\ r_k =
(\alpha_{2k}+\alpha_{2k+1})/2$, belong to the resolvent set of $T$, provided that $\delta <b -b'$ and $ |k|\geq N$ for $N$ large enough. For $|k|\geqslant N+1$, denote by $\Pi_k$ the curvilinear trapezoid formed by the lines ${\mathop{\rm Re}}\,\lambda = r_{k-1}$, ${\mathop{\rm Re}}\,\lambda = r_{k}$, and the boundary of the parabola $P_{p,h}$. Assume that $Q_0$ is the Riesz projection corresponding to the (bounded) part of the spectrum of $T$ that lies outside $\bigcup_{|k|\geqslant N+1}\Pi_k$. And let $Q_k$, $|k|\geqslant
N+1$, be the Riesz projection for the part of the spectrum of $T$ confined within the curvilinear trapezoid $\Pi_k$. Main result of the work consists in proving that the system of the invariant subspaces $Q_k(H)$, $|k|\geqslant N+1$, together with the invariant subspace $Q_0(H)$ forms an unconditional basis of subspaces in the space $H$. We also prove a generalization of this theorem to the case where any gap $(\alpha_{2k},\,\alpha_{2k+1})$, $k\in\mathbb{Z}$, may contain a finite number of eigenvalues of $T$ with their total multiplicity bounded by a number $m\in\mathbb{N}$ independent of $k$.
- |
Пусть $T$ — самосопряженный оператор в гильбертовом пространстве $H$ с областью определения $\mathcal D(T)$. Будем считать, что спектр $T$ лежит в объединении непересекающихся интервалов $\Delta_k
=[\alpha_{2k-1},\, \alpha_{2k}]$, $k\in \mathbb{Z}$, длины которых подчиняются неравенствам $$\alpha_{2k+1}-\alpha_{2k} \geqslant b
|\alpha_{2k+1}+\alpha_{2k}|^p\quad \text{ при некоторых }\, b>0,\,
p\in[0,1).$$Предположим, что линейный оператор $B$, действующий в $H$, является $p$-подчиненным оператору $T$, т.е. $\mathcal D(B)
\supset\mathcal D(T)$ и $\|Bx\| \leqslant b'\,\|Tx\|^p\|x\|^{1-p}
+M\|x\|$ с некоторыми $b'>0$ и $M\geq 0$ при всех $x\in \mathcal
D(T)$. Тогда при всяком $h>b'$ спектр возмущенного оператора $A=T+B$ лежит в объединении некоторого конечного прямоугольника в $\mathbb{C}$ и двойной параболы $P_{p,h} = \bigl\{\lambda \in
\mathbb{C}\, \bigl|\ \ |\mathop{\rm Im} \lambda| \leqslant
h|\mathop{\rm Re} \lambda|^p\bigr\}$. Более того, если $\delta <b
-b'$, то можно указать такое $N\in\mathbb{N}$, что при всех $|k|>N$ вертикальные полосы $\Omega_k =\{\lambda\in\mathbb C\,\vert \ \,
|r_k-{\mathop{\rm Re}}\,\lambda|\leqslant \delta r_k^p\}$, $\ r_k =
(\alpha_{2k}+\alpha_{2k+1})/2$, лежат в резольвентном множестве $T$. При $|k|\geqslant N+1$ обозначим через $\Pi_k$ криволинейные трапеции, образованные прямыми ${\mathop{\rm Re}}\,\lambda = r_{k-1}$, ${\mathop{\rm Re}}\,\lambda = r_{k}$ и границей параболы $P_{p,h}$. Пусть $Q_0$ — проектор Рисса, отвечающий той (ограниченной) части спектра оператора $T$, которая лежит вне $\bigcup_{|k|\geqslant N+1}\Pi_k$. И пусть $Q_k$, $|k|\geqslant N+1$, — проектор Рисса для той части спектра $T$, которая лежит в криволинейной трапеции $\Pi_k$. Основным результатом работы является доказательство теоремы о том, что система инвариантных подпространств $Q_k(H)$, $|k|\geqslant
N+1$, вместе с инвариантным подпространством $Q_0(H)$ образует безусловный базис из подпространств в гильбертовом пространстве $H$. Мы доказываем также обобщение этой теоремы на случай, когда в любой из лакун $(\alpha_{2k},\,\alpha_{2k+1})$, $k\in\mathbb{Z}$, может присутствовать конечный набор собственных значений $T$, суммарная кратность которых не превышает некого числа $m\in\mathbb{N}$, не зависящего от $k$.
author:
- 'A.K. Motovilov, A.A. Shkalikov'
title: 'Preserving of the unconditional basis property under non-self-adjoint perturbations of self-adjoint operators[^1]'
---
[english]{}
[**Key words:**]{} Riesz basis, unconditional basis of subspaces, non-self-adjoint perturbations.
0.2cm
[**AMS Mathematics Subject Classification:**]{} 47A55, 47A15.
0.2cm
Сохранение безусловной базисности при несамосопряженных возмущениях самосопряженных операторов$^*$
А.К. Мотовилов, А.А. Шкаликов
------------------------------------------------------------------------
[$^*$Работа выполнена при финансовой поддержке Немецкого научно-исследовательского общества (DFG) и Российского фонда фундаментальных исследований (РФФИ)]{}
$\S 1.$ [**Введение**]{}
Задача о спектральных свойствах операторов, которые представимы в виде возмущений самосопряженных операторов, имеет давнюю историю. Изложение разных аспектов этой теории для операторов с дискретным спектром имеется в [@Sh1]. В настоящей работе нас будет интересовать вопрос о сохранении свойств базисности для подчиненных возмущений самосопряженных операторов в общем случае без предположения дискретности спектра невозмущенного оператора. По-видимому, первым результатом на эту тему была работа авторов [@MSh]. Здесь мы получим существенное обобщение основного результата работы [@MSh].
Одним из глубоких результатов, который можно рассматривать как итог развития в 1951-1985 годах тематики, посвященной задаче о базисности корневых векторов несамосопряженных операторов, является теорема Маркуса-Мацаева [@Ma Гл. 1, Теорема 6.12] (см. также [@MaM1]). Приведем формулировку этой теоремы (для простоты вместо нормальных операторов со спектром на конечном числе лучей мы рассматриваем самосопряженные операторы).
[**Теорема А.** ]{} [*Пусть $T$ — самосопряженный оператор с дискретным спектром в гильбертовом пространстве $H$, а его собственные значения $\{\mu_k\}$, занумерованные с учетом кратности, подчинены оценке $$\label{1a}
|\mu_k| \geqslant C |k|^\beta, \qquad C =\text{const}, \quad \beta
>0$$ (здесь предполагается, что отрицательные собственные значения нумеруются в порядке их убывания отрицательными целыми числами, а неотрицательные, в порядке их возрастания — неотрицательными индексами). Пусть линейный (не обязательно замыкаемый) оператор $B$ задан на на области определения $\mathcal D(T)$ оператора $T$ и при некоторых $b>0$, $0\leqslant p<1$ выполняется оценка $$\label{2}
\|Bx\| \leqslant b\|Tx\|^p\|x\|^{1-p}, \quad \ x\in \mathcal D(T).$$ Тогда оператор $A=T+B$, определенный на $\mathcal D(T)$, имеет только дискретный спектр. При дополнительном условии $\beta^{-1}
\leqslant 1-p$ система его корневых векторов образует безусловный базис со скобками в исходном гильбертовом пространстве $H$.* ]{}
Отметим, что условие вместе с неравенством $\beta^{-1}
\leqslant 1-p$ эквивалентно условию $$\label{0}
\overline{\lim}_{t\to\infty} \frac {n(t, T)}{t^{1-p}} <\infty,
\qquad\text {где }\ \, n(t, T) = \sum_{|\mu_k|\leqslant t} 1$$ функция распределения собственных значений оператора $T$. Здесь же отметим, что предложенное Маркусом и Мацаевым доказательство по существу без изменений переносится на случай, когда верхний предел в последнем соотношении меняется на нижний.
В работе [@Sh1] было получено обобщение этой теоремы в следующей форме.
[**Теорема B.** ]{} [*Пусть $T$ — самосопряженный оператор с дискретным спектром в гильбертовом пространстве $H$, а оператор $B$ является $p$-подчиненным оператору $T$, то есть при некоторых $p\in[0,1),$ выполняется оценка $$\label{4}
\|Bx\| \leqslant b\|Tx\|^p\|x\|^{1-p} +M\|x\|, \quad \ x\in
\mathcal D(T), \quad b, M = \text{const}.$$ Пусть $b'$ — точная нижняя грань, при которых выполняется оценка с произвольными постоянными $M= M(b)$ (но при $p=0$ считаем, что постоянная $M$ в равна нулю!) и при некотором $b_1 > b'$ выполняется условие $$\label{m}
\underline{\lim}_{\ r\to\infty}\, n_\pm (r +b_1r^p) -n_\pm
(r-b_1r^p) = m <\infty,$$ где $n_\pm (r)$ — функции распределения собственных значений оператора $T$ на положительной и отрицательной осях соответственно. Тогда система корневых векторов оператора $A=T+B$ образует безусловный базис со скобками.* ]{}
Отметим, что фигурирующее в Теореме B условие существенно слабее условия или условия .
Предположение о дискретности спектра оператора $T$ в Теореме B существенно использовалось в ее доказательстве. В частном, но важном случае, когда порядок подчиненности $p=0$, число $M$ в условии равно нулю, а вместо условия требуется существование лакун $\Lambda_k = (\alpha_{2k},\alpha_{2k+1}) $ в спектре $T$ c равномерно ограниченной снизу длиной, авторы [@MSh] показали, что утверждение Теоремы B допускает обобщение на операторы с непустым существенным и, в частности, с непрерывным спектром. А именно, был получен следующий результат.
[**Теорема С.**]{} [*Пусть $T$ — самосопряженный оператор, спектр которого лежит внутри отрезков $\Delta_k = [\alpha_{2k-1},\,
\alpha_{2k}] \subset \mathbb R$, $k\in \mathbb Z$, причем $$\inf_{k\in\mathbb Z} (\alpha_{2k+1} - \alpha_{2k}) = d >0$$ (т.е. длины лакун в спектре между отрезками $\Delta_k$ всегда $\geqslant d$). Пусть $B$ — ограниченный (и, в общем, несамосопряженный) оператор, причем $\|B\|=b^*<d/2$. Тогда при любом $\varepsilon \in (b^*,\, d/2)$ спектр возмущенного оператора $A=T+B$ лежит в объединении $\bigcup_k U_\varepsilon
\left(\Delta_k\right)$, где $U_\varepsilon \left(\Delta_k\right)$ — непересекающиеся между собой $\varepsilon$-окрестности отрезков $\Delta_k$. Если $Q_k$ — проекторы Рисса на инвариантные подпространства оператора $A=T+B$, отвечающие частям спектра в областях $U_\varepsilon \left(\Delta_k\right)$, т.е. $$\label{Riesz}
Q_k=-\frac{1}{2\pi i} \int_{\Gamma_k}
\left(A-\lambda\right)^{-1}\,d\lambda, \qquad \Gamma_k=\partial
U_\varepsilon \left(\Delta_k\right),$$ то подпространства $\mathcal{L}_k=Q_k(H)$, $k\in\mathbb Z$, образуют безусловный базис в исходном пространстве $H$.* ]{}
В этой работе мы получим сначала обобщение Теоремы С на случай $p$-подчиненных возмущений при $p\in (0,1)$. А затем получим еще более общий результат, который является аналогом Теоремы B для операторов, которые могут иметь существенный спектр, но в лакунах между отрезками существенного спектра имеется $\leqslant m$ собственных значений, где $m$ — фиксированное число.
Перед формулировкой основных теорем введем следующие обозначения. Обозначим через $$\label{par}
P_{p,h} = \{\lambda \in \mathbb C\, \vert\ \ |{\mathop{\rm Im}}\lambda| < h|{\mathop{\rm Re}}\lambda|^p\}$$ — область, ограниченную двойной параболой. Часть этой области в правой (левой) полуплоскости будем обозначать через $P^+_{p,h}\
(P^-_{p,h})$. Пусть $\Delta = [s,t]$ — отрезок на положительной или отрицательной полуоси, $t>s$. Множество $$\label{Updel} \mathcal U_{p\delta}(\Delta) = \{\lambda \in P_{p,
b'+\delta}\, \vert \ \ s -\delta |s|^p < {\mathop{\rm Re}}\,\lambda < t+\delta
|t|^p\}$$ назовем $p\delta$-окрестностью отрезка $[s,t]$. Отметим, что в определении $p\delta$-окрестности помимо чисел $p$ и $\delta$ участвует еще число $b'$, которое предполагается равным нижней грани чисел $b$ в условии $p$-подчиненности . Конечно, таким образом определенная окрестность завит от этого числа $b'$, но для удобства обозначений мы будем считать это число фиксированным и указывать эту зависимость не будем. Всюду далее считаем, что в случае полуограниченного оператора $T$ нумерация отрезков его спектра проводится индексами $k\geqslant 1$, а в общем случае — целыми индексами $k\in \mathbb Z$, причем для этого случая принимается $\alpha_0 \leqslant 0, \alpha_1 >0$.
Сформулируем основные результаты работы.
[**Tеорема 1.**]{} [*Пусть $T$ — самосопряженный оператор в гильбертовом пространстве $H$ с областью определения $\mathcal
D(T)$, а линейный оператор $B$ определен на области $\mathcal
D(T)$ и является $p$-подчиненным оператору $T$ (в смысле неравенства ) с нижней гранью $p$-подчиненности $b'$ (как и ранее при $p=0$ в неравенстве полагаем $M=0$). Предположим, что спектр оператора $T$ заключен внутри непересекающихся отрезков $\Delta_k =[\alpha_{2k-1},\,
\alpha_{2k}]$, а длины лакун $\Lambda_k = (\alpha_{2k},\,
\alpha_{2k+1})$ между отрезками $\Delta_k$ таковы, что[^2] $$\label{gaps}
\alpha_{2k+1}-\alpha_{2k} \geqslant 2^{1-p}\,\, b_1
|\alpha_{2k+1}+\alpha_{2k}|^p\quad \text{при некоторых}\ \, b_1>0,\,
p\in[0,1),$$ причем $b_1 > b'$. Тогда при любом $\delta \in (0, b_1 -b')$ найдется число $N\in \mathbb N$, такое, что спектр оператора $A=T+B$ лежит в объединении некоторого прямоугольника $\mathcal
U_0$ и взаимно непересекающихся $p\delta$-окрестностей $ \mathcal
U_k =\mathcal U_{p\delta}(\Delta_k)$ отрезков $\Delta_k$ при $|k| \geqslant N$. Если $$ Q_k=-\frac{1}{2\pi i} \int_{\partial\mathcal U_k}
\left(A-\lambda\right)^{-1}\,d\lambda,\qquad k=0, \pm N, \pm (N+1),
\dots,$$ — проекторы Рисса, отвечающие частям спектра оператора $A$, расположенным внутри $\mathcal U_k$, то инвариантные подпространства $\{Q_k(H)\}$, $|k|\geqslant N$, вместе с $Q_0(H)$ образуют безусловный базис из подпространств в гильбертовом пространстве $H$.* ]{}
[**Tеорема 2.**]{} [*Пусть выполнены условия Теоремы 1, но в ослабленном виде: допускается, что вне отрезков $\Delta_k$, содержащих существенный спектр оператора $T$, имеется (только) дискретный спектр, причем в каждой лакуне $\Lambda_k$ содержится $\leqslant m$ собственных значений оператора $T$ с учетом кратности и число $m$ не зависит от $k$. Тогда спектр оператора $A=T+B$ асимптотически лежит внутри двойной параболы $ P_{p,h}$, $h>b'$, и при любом $\delta\in (0, b_1-b')$ вертикальные полосы $$\Omega_k =\{\lambda\in \mathbb C\, \vert\ \, r_k -\delta r_k^p
\leqslant {\mathop{\rm Re}}\, \lambda \leqslant r_k+\delta r^p_k\}, \qquad r_k =
(\alpha_{2k}+ \alpha_{2k+1})/2,$$ не могут содержать никаких иных точек спектра оператора $A$ кроме собственных значений c суммарной алгебраической кратностью $\leqslant Cm$, где $C$ не зависит от $k$. Найдутся такие числа $r'_k\in \Omega_k \bigcap \mathbb R$, что границы криволинейных трапеций $\Pi_k$, образованных прямыми ${\mathop{\rm Re}}\lambda = r'_k$, ${\mathop{\rm Re}}\lambda = r'_{k+1}$ и параболой $P_{p,h}$, не содержат точек спектра оператора $A=T+B$, а проекторы Рисса $$ Q_k=-\frac{1}{2\pi i} \int_{\partial\Pi_k}
\left(A-\lambda\right)^{-1}\,d\lambda,\qquad k=\pm N, \pm (N+1),
\dots,$$ таковы, что подпространства $Q_k(H)$, $|k|\geqslant N$, вместе с $Q_0(H)$ образуют безусловный базис из подпространств в гильбертовом пространстве $H$. Здесь $Q_0$ — проектор Рисса, отвечающий той части спектра оператора $A$, которая не входит в криволинейные трапеции $\Pi_k$, $|k| \geqslant N$.* ]{}
$\S 2.$ [**Определения и вспомогательные предложения.**]{}
Напомним известные понятия базиса и безусловного базиса (см. например, [@GK гл. 6]. Последовательность векторов $\{\psi_k\}_{k=1}^\infty$ пространства $H$ называется [*базисом*]{} этого пространства, если каждый вектор $x\in H$ разлагается единственным образом в сходящийся по норме $H$ ряд $$\label{ser}
x=\sum_{k=1}^\infty c_k\psi_k,$$ где $c_k$ — числовые коэффициенты. Базис называется [*безусловным*]{}, если он остается базисом после любой перестановки векторов последовательности. Система $\{\psi_k\}_{k=1}^\infty$ называется [*базисом со скобками*]{}, если для любого $x\in H$ ряд сходится по норме $\mathcal H$ после некоторой расстановки скобок, не зависящей от $x$. Если система остается базисом после любой перестановки наборов ее векторов, которые отвечают членам ряда, заключаемым в скобки, то такая система называется [*безусловным базисом со скобками*]{} или [*базисом Рисса со скобками*]{}.
Далее будет идти речь о базисности и безусловной базисности последовательности подпространств, о чем удобно говорить на языке проекторов (см. [@Sh1 $\S 6$]). Используются следующие определения:
1. Система ограниченных проекторов $ \{Q_j\}_1^\infty$ полна, если равенства $$(Q_jx, y) =0 \quad \forall x\in H, \ \ j=1,2,\dots,\ \
\text{влекут}\ \, y=0.$$
2. Система ограниченных проекторов $ \{Q_j\}_1^\infty$ минимальна, если $ Q_jQ_k =\delta_{kj} Q_j$.
3. Пусть $ \{Q_j\}_1^\infty$ — минимальная система ограниченых проекторов в $H$. Система подпространств $\mathcal L_j = Q_j(H),
j=1,2,\dots,$ называется базисом (безусловным базисом) в $H$, если ряд $\sum_j Q_j$ сходится (сходится после любой перестановки индексов в сумме) в сильной операторной топологии к единичному оператору.
Далее будем использовать следующий результат [@Sh1 $\S 6$].
[**Предложение 1.**]{} [*Пусть $\{Q_j\}_{j=1}^\infty$ — система ограниченных проекторов в $H$. Система подпространств $\mathcal L_j
=Q_j(H), \ j=1,2, \dots,$ является безусловным базисом из подпространств в $H$ тогда и только тогда, когда система проекторов $\{Q_j\}_{j=1}^\infty$ полна, минимальна и $$\label{converge}
\sum_{j=1}^\infty |(Q_jx,x)| < \infty \qquad \forall\, x\in H.$$* ]{}
Приведем утверждение о локализации спектра оператора оператора $A=
T+B$ при $p$-подчиненных возмущениях.
[**Предложение 2.**]{} [*Пусть $T$ — самосопряженный положительный (или полуограниченный) оператор, а возмущение $B$ является $p$-подчиненным оператору $T$ с точной верхней гранью $b'$. Тогда спектр оператора $A=T+B$ асимптотически локализован в параболической области (при $p=0$ полуполосе) $$\label{Pi}
P^+_{p,\, h} =\{\lambda\in\mathbb{C}\, | \ \, {\mathop{\rm Re}}\lambda \geqslant
0,\ \, |{\mathop{\rm Im}}\lambda| \leqslant h({\mathop{\rm Re}}\lambda)^p\}, \quad h>b'.$$ При этом в правой полупоскости вне параболы $P^+_{p,\, h}$ при любом $\varepsilon >0$ асимптотически выполняются оценки (при любом $\varepsilon\in (0, h-b')$) $$\label{resS}
\|B(T-\lambda)^{-1}\| \leqslant \frac{b'+\varepsilon}{h} \qquad |\lambda| \geqslant R = R(\varepsilon),$$ и $$\label{resA}
\|(A-\lambda)^{-1}\| \leqslant
\left(\frac{b'+\varepsilon}{h-b'}\right)\, \frac 1{|{\mathop{\rm Im}}\lambda|},
\qquad |\lambda| \geqslant R,$$ а в левой полуплоскости вне круга радиуса $\gg 1$ выполняется оценка оценка и оценка $\|(A-\lambda)^{-1}\| \leqslant
C|\lambda|^{-1}$. Если $T$ самосопряжен, но не полуограничен, то обе оценки и выполняются вне области, ограниченной двойной параболой $P_{p,\, h}$.*]{}
При выполнении условия это утверждение доказано в работе [@MaM2]. При выполнении более общего условия $p$-подчиненности доказательство Предложения 1 имеется в [@Sh1 $\S 3$].
[**Предложение 3.**]{} [*Пусть $\{s_j\}_1^\infty$ и $\{\lambda_j\}_1^\infty$ — $s$-числа и собственные значения ядерного оператора $K$, занумерованные с учетом алгебраической кратности. Тогда $$\label{K}
(1-\mu K)^{-1} = \frac{F(\mu)}{D(\mu)}, \quad \text{где}\ \, D(\mu)
= \prod_{j=1}^\infty \left(1-\mu\,\lambda_j\right).$$ Произведение $D(\mu)$ сходится и является целой функцией переменной $\mu\in\mathbb{C}$ порядка $\leqslant 1$ и минимального типа при порядке 1. Оператор-функция $F(\mu)$ также целая и допускает оценку $$\|F(\mu)\| \leqslant \prod_{j=1}^\infty \left( 1+ |\mu| s_j\right).$$ Тем самым $F(\mu)$ также порядка $\leqslant 1$ и минимального типа при порядке 1.*]{}
[*Доказательство*]{} этого важного утверждения имеется в [@GK гл. 3].
Нам понадобится также одно предложение из теории функций.
[**Предложение 4.**]{} [*Пусть $f(\lambda)$ голоморфная ограниченная функция в $\lambda$ прямоугольнике $$\label{Pi}
\Pi\ = \{\lambda: \ \, |{\mathop{\rm Re}}\lambda| <a, \ \, |{\mathop{\rm Im}}\lambda| <c\}.$$ Фиксируем $\tau \in (0,1)$, положим $a' =a(1-\tau),\ \,
c'=c(1-\tau)$ и обозначим через $\Pi'$ прямоугольник, который определен равенством с заменой $a$ и $c$ на $a'$ и $c'$, соответственно. Обозначим $${M}= \sup_{\lambda\in \Pi} f(\lambda),\quad {M}' = \sup_{\lambda\in \Pi'} f(\lambda).$$ Тогда для числа нулей $n_f(\Pi')$ функции $f(\lambda)$ в прямоугольнике $\Pi'$ справедлива оценка $$\label{n_f}
n_f(\Pi') \leqslant C(\ln M-\ln M').$$ Для любого $\varepsilon\in (0,1)$ найдутся круги с общей суммой радиусов $\leqslant \varepsilon$, такие, что вне объединения $\mathcal E$ этих кругов выполняется оценка снизу $$\label{below1}
\ln|f(\lambda)| \geqslant С \ln\varepsilon\ (\ln M -\ln M') + \ln
M', \quad \lambda\in\Pi' \setminus \mathcal E.$$ Постоянная $C$ в обеих оценках зависят только от $\tau$ и отношения $a/c$ и не зависит от $f$. При одновременном увеличении или уменьшении сторон прямоугольников $\Pi$ и $\Pi'$ в $\eta$ раз оценка сохраняется с той же постоянной $C$, но исключительное множество кругов нужно брать с общей суммой радиусов $\eta\varepsilon$.* ]{} Варианты сформулированных в этом предложении утверждений в другой форме можно найти, например, в книге [@Lev]. В сформулированном виде первое утверждение имеется в работе [@MaM2 Лемма 1.1 ], а первое и второе утверждение в работах [@Sh4] и [@Sh6 Лемма 2.1].
$\S 3.$ [**Доказательство основных теорем.**]{}
Поскольку Теорема 2 является обобщением Теоремы 1, мы не будем проводить их доказательства отдельно. Сначала проведем основную часть доказательства в условиях Теоремы 1, затем укажем на изменения, которые нужно сделать в общем случае, а на заключительном этапе докажем полноту проекторов Рисса в условиях Теоремы 2. Доказательство разобьем на несколько этапов. На некоторых этапах доказательство проводится почти так же, как в случае Теореме 6.6 из работы [@Sh1]. Однако для удобства читателя мы приводим доказательство полностью. Всюду далее через ${\sigma}(Z)$ обозначается спектр $Z$.
[*Шаг 1.*]{} Упростим максимально задачу и будем считать, что оператор $T=T^*$ положителен, а лакуны $\Lambda_k:=(\alpha_{2k},\alpha_{2k+1})$ в спектре невозмущенного оператора $T$ не содержат собственных значений. Позже мы избавимся от этих предположений. Доказательство проведем на основе следующих представлений, справедливых при $\lambda \notin
\sigma(T) \cup \sigma(A)$: $$\label{65}
(A-\lambda)^{-1} = (T-\lambda)^{-1} - G(\lambda), \qquad G(\lambda)
= (A-\lambda)^{-1}B(T-\lambda)^{-1}$$ и $$\label{66}
G(\lambda) =(T-\lambda)^{-1} M(\lambda)B(T-\lambda)^{-1}, \quad
M(\lambda) = (1+S(\lambda))^{-1}, \ \ S(\lambda) =
B(T-\lambda)^{-1}.$$ Согласно Предложению 2, спектр оператора $A= T+B$ лежит в объединении круга достаточно большого радиуса и параболической области $P_{p, h}$, если $h>b'$, а вне этого объединения выполнена оценка . Напоминаем, что $b'$ есть точная нижняя грань постоянных $b$, при которых выполняется оценка $p$-подчиненности . Далее считаем $h > 2b'$, тогда в силу Предложения 2 существует такое $r_0$, что $$\|S(\lambda)\|\leqslant 1/2, \ \ \ \|M(\lambda)\| \leqslant 2,
\qquad\text{когда \,\,} \lambda\notin P_{p, h}, \ \, |\lambda| >r_0.$$
Ниже на шаге 4 мы покажем, что если интервалы $\Lambda_j=
(\alpha_{2j},\, \alpha_{2j+1})$ не содержат собственных значений оператора $T$, то при $|j|\geqslant N$ (где $N$ — достаточно большое число) на вертикальных прямых ${\mathop{\rm Re}}\, \lambda =r_j=
(\alpha_{2j+1}+ \alpha_{2j})/2 $ справедлива оценка $\|S(\lambda)\|\leqslant 1-\varepsilon$ при некотором $\varepsilon
>0$, а потому оператор-функция $M(\lambda)$ равномерно ограничена на этих прямых. В частности, эта оператор-функция равномерно ограничена на вертикальных отрезках $\gamma_j$, которые проходят через точки $r_j$ и соединяют кривые $$\Gamma_\pm = \{\lambda \in \mathbb C \, \vert \ {\mathop{\rm Im}}\, \lambda =
\pm h|{\mathop{\rm Re}}\,\lambda|^p\}$$ — границы области $P_{p, h}$ в верхней и нижней полуплоскостях.
Обозначим через $Q_j$ и $P_j$ проекторы Рисса на части спектра операторов $A$ и $T$, которые заключены внутри криволинейных трапеций $\Pi_j \subset P_{p, h}$, ограниченных вертикальными отрезками $\gamma_{j-1}$ и $\gamma_j$. Сумма ортогональных проекторов $P_j$ безусловно сходится к единичному оператору. Система проекторов $\{Q_j\}$, отвечающих за непересекающиеся части спектра, минимальна в $H$ для любого оператора. Поэтому в силу представления для доказательства теоремы достаточно показать, что система подпространств $Q_j(H)$, инвариантных относительно оператора $A$, полна в $H$ и при всех $x\in H$ сходится ряд $$\label{67}
\sum_{j=N}^\infty |(Q_j x, x)|, \qquad Q_j = -\frac 1{2\pi i}
\int_{\Gamma_j} (A-\lambda)^{-1}\, d\lambda,$$ где $\Gamma_j$ — ориентированные границы криволинейных трапеций $\Pi_j$.
Задача о полноте инвариантных подпространств оператора $A$ будет положительно решена на заключительном шаге 6. Если равномерная ограниченность оператор-функции $M(\lambda)$ на вертикальных отрезках $\gamma_j$ уже доказана, то из представления получаем, что для завершения доказательства теоремы достаточно доказать сходимость интегралов $$\label{Gam}
\int_{\Gamma_\pm} \|(T-\lambda)^{-1}x\|\ \| S(\lambda)x\|\
|d\lambda| \ < \infty, \qquad x\in H,$$ и ряда $$\label{Ser1}
\sum_{j=1}^\infty \int_{\gamma_j} \|(T-\lambda)^{-1}x\|\ \|
S(\lambda)x\|\ |d\lambda| \ < \infty.$$ Итак, наш план состоит в следующем. На шаге 2 мы оценим оператор-функцию $M(\lambda)$ на вертикальных прямых ${\mathop{\rm Re}}\,\lambda =
r_j$ при условии, что лакуны $\Lambda_j = (\alpha_{2j},
\alpha_{2j+1})$ не содержат собственных значений оператора $T$. Более того, равномерная оценка для $M(\lambda)$ будет сохраняться в вертикальных полосах $$\label{Om}
\Omega_j = \{\lambda \in \mathbb C\, \vert \ \, r_j-\delta r_j^p <
{\mathop{\rm Re}}\,\lambda < r_j+\delta r_j^p\}$$ при любом фиксированном $\delta\in(0, b_1- b')$, где число $b_1$ берется из условия . Далее мы докажем сходимость интегралов и сходимость ряда , где $\gamma_j$ произвольные вертикальные отрезки в полосах $\Omega_j$, соединяющие кривые $\Gamma_{\pm}$. На шаге 5 с помощью приема искусственной лакуны мы покажем, что при выполнении условия с произвольным числом $m\in \mathbb N$ в полосе $\Omega_j$ можно выбрать вертикальные отрезки $\gamma_j'$, на которых нужные оценки сохраняются. Тем самым, доказательство будет завершено для случая $T=T^* >0$. Изменения, которые нужно провести для доказательства в случае, когда оператор $T$ не является полуограниченным, будут отражены на шаге 7.
[*Шаг 2.*]{} Оценим оператор-функцию $S(\lambda)$ на прямой ${\mathop{\rm Re}}\,\lambda =r_j$ при условии, что интервал $\Theta_j: = (r_j- b_1
r_j^p,\, r_j + b_1 r_j^p)$ не содержит собственных значений оператора $T$. Далее всюду, где это удобно, опускаем индекс $j$ и вместо $r_j$ пишем $r$.
Пусть $\lambda = \eta +i\tau$, $\eta,\tau\in\mathbb{R}$. Согласно спектральной теореме для самосопряженного оператора $T$ имеем $$\begin{gathered}
\label{TT}
\|T(T-\lambda)^{-1}\| = \sup_{t\in \sigma(T)} \left|\frac t{t-\lambda}\right| =
\left( \inf_{t\in \sigma(T)} \left| 1-\frac\lambda{t}\right|\right)^{-1}
\leqslant \\
\left( \inf_{t\in \sigma(T)} \left|
1-\frac\eta{t}\right|\right)^{-1}, \quad \lambda\not\in{\sigma}(T).\end{gathered}$$ Так как $\sigma(T)\cap \Theta_j = \emptyset$, а функция $|1-\eta
t^{-1}|$ убывает при $t\in (0, \eta)$ и возрастает при $t\in
(\eta, \infty)$, то правая часть на прямой ${\mathop{\rm Re}}\,\lambda = r =r_j$ не превосходит максимума значений этой функции на концах интервала $\Theta_j$, то есть не превосходит величины $$\max \left(\left| 1- \frac{r}{r-b_1r^p}\right|^{-1}, \ \left( 1
-\frac{r}{r+b_1r^p}\right)^{-1}\right)= \frac{r+b_1r^p}{b_1r^p}.$$ В силу условия середина лакуны $\Lambda_j$ удалена от концов отрезков, содержащих спектр оператора $T$, на расстояние $\geqslant b_1 r^p$. Тогда, используя очевидную оценку $\|(T-\lambda)^{-1}\|\leqslant (b_1 r^p)^{-1}$ на прямой ${\mathop{\rm Re}}\,\lambda =r= r_j$ и оценку получаем (далее $o(1)$ есть бесконечно малая величина при $r =r_j\to\infty$) $$\begin{gathered}
\|S(\lambda)\| \leqslant b \left(\frac{r+ b_1r^p}{b_1r^p}\right)^p
\left( \frac 1 {b_1r^p}\right)^{1-p} + \frac M{b_1r^p} = b
\left(\frac{r+ b_1r^p}{b_1r^p}\right)^p \left( \frac 1
{b_1r^p}\right)^{1-p} \left(1+o(1)\right)\\
=\frac b{b_1}\left(1+o(1)\right) <1,\end{gathered}$$ если $b\in (b', b_1)$. Очевидно, последняя оценка сохраняется для $\lambda$ в полосе $\Omega_r = \Omega_j(\delta)$, определяемой , если $\delta\in (0, b_1-b)$ и $r_j$ достаточно велико. Следовательно, спектр возмущенного оператора $A= T+B$ асимптотически лежит в $p\delta$-окрестностях $\mathcal U_{p\delta}(\Delta_j)$ (см. определение в $\S$ 1). В частности, при $\delta =
(b_1 -b')/2$ и $b= b'+ \delta/4$ получаем $$\label{Sl}
\|S(\lambda)\| \leqslant \frac b{(b_1-\delta)} = \frac {2b}{b_1+
b'}, \quad \| M(\lambda)\|\leqslant \frac{b_1+ b'}{b_1+b' -2b}=
\frac{2(b_1+ b')}{(b_1-b')}, \qquad \lambda\in \Omega_j, \ \lambda
\gg 1.$$
[*Шаг 3.*]{} Оценим интеграл по кривой $\Gamma_+$. Оценка по $\Gamma_-$ аналогична. Из условия $p$-подчиненности получаем $$\label{Slam}
\| S(\lambda)x\| \leqslant b\| T(T-\lambda)^{-1}x\|^p\
\|(T-\lambda)^{-1}x\|^{1-p} + M\|(T-\lambda)^{-1}x\|.$$ При $p=0$ ситуация сильно упрощается, так как второе слагаемое равно нулю, а в первом слагаемом нужно работать только со вторым множителем (соответствующие оценки при $p=0$ имеются в работе [@MSh]). При $p>0$ интеграл в мажорируется величиной $$b\,I_1+M\,I_2,$$ где $$\begin{aligned}
I_1 &=\int_{\Gamma_+}\, \left(\|T(T-\lambda)^{-1}x\|^p\
|\lambda|^{\frac{(p-2)p}2}\right)\
\left( |\lambda|^{\frac{(2-p)p}2}\|(T-\lambda)^{-1}x\|^{2-p}\right)\ |d\lambda|,\\
I_2 &=\int_{\Gamma_+}\, \|(T-\lambda)^{-1}x\|^{2}\, |d\lambda|.\end{aligned}$$ Легко видеть, что при $\lambda\to\infty$, $\lambda\in\Gamma_+$, подынтегральная функция в $I_2$ являетcя бесконечно малой более высокого порядка нежели подынтегральная функция в $I_1$. Поэтому достаточно убедиться в сходимости интеграла $I_1$. Мы оценим этот интеграл по неравенству Гёльдера (в качестве гёльдеровских сопряженных чисел берем $q= 2/p$ и $q' = 2/(2-p)$: $$\label{Gel1}
I_1\leqslant \left( \int_{\Gamma_+} |\lambda|^{p-2}\
\|T(T-\lambda)^{-1}x\|^2\ |d\lambda|\right)^{p/2} \left(
\int_{\Gamma_+} |\lambda|^p\ \|(T-\lambda)^{-1}x\|^2\, |d\lambda|
\right)^{(2-p)/2}.$$ =1000 Обозначим через $E(t)$ спектральную функцию оператора $T$ и положим $e(t) =(E(t)x,x).$ Пусть $\xi\geqslant 1$ и $\lambda =\xi +ih\xi^p\,\,(\in \Gamma_+)$. Очевидно, при $\xi\to+\infty$ имеем $$|\lambda|^p = \xi^p \left(1+ o(1)\right), \ \, |d\lambda| =
d\xi\left(1+o(1)\right),\ \, |t-\lambda|^2 = |t-\xi|^2
+ih^2\xi^{2p}.$$ Поэтому сходимость интегралов в эквивалентна сходимости следующих интегралов $$\int_1^\infty\hspace*{-0.5em} d \xi \int_0^\infty
\hspace*{-0.5em}d\, e(t)\frac{\xi^p\,}{(t-\xi)^2 + h^2\xi^{2p}}
=\int_1^\infty F_1(t)\, d\, e(t),\, \text{ где }\, F_1(t)
=\int_1^\infty \frac{\xi^p}{(t-\xi)^2 + h^2\xi^{2p}} \, d \xi,$$ $$\int_1^\infty\hspace*{-0.5em} d\xi \int_0^\infty\hspace*{-0.5em} d\,
e(t) \frac{t^2\, \xi^{p-2}}{(t-\xi)^2 + h^2\xi^{2p}} =\int_1^\infty
F_2(t)\, d\, e(t),\, \text{ где }\, F_2(t) =\int_1^\infty
\frac{t^2\,\xi^{p-2}}{(t-\xi)^2 + h^2\xi^{2p}} \, d \xi.$$ Поскольку функция $e(t)$ монотонна и $0\leqslant e(t) \leqslant
\|x\|^2$, выписанные интегралы сходятся, если функции $F_1$ и $F_2$ ограничены. Заметим, что $$F_1(t) \geqslant F_2(t), \ \text{если}\, t\geqslant \xi, \
\text{и}\ F_1(t) \leqslant F_2(t), \ \text{если}\, 0\leqslant
t\leqslant \xi.$$ Отсюда заключаем, что, в частности, при $t\geqslant 1$ $$F_1(t) + F_2(t) \leqslant 2 \int_1^t \frac{t^2\,
\xi^{p-2}}{(t-\xi)^2 + h^2\xi^{2p}}\, d\xi +2 \int_t^\infty
\frac{\xi^{p}}{(t-\xi)^2 + h^2\xi^{2p}}\, d\xi =:2F_3(t) +2F_4(t).$$ Оценка последних интегралов очевидна. Например, функция $F_3(t)$ при $t^p<t/2$ (т.е. при $t>2^{1/(1-p)}$) оценивается следующим образом: $$\begin{gathered}
F_3(t) \leqslant \left(\int_1^{t/2} + \int_{t/2}^{t-t^p} +
\int_{t-t^p}^t\right)
\frac{t^2 \, \xi^{p-2}}{(t-\xi)^2 +h^2\xi^{2p}}\, d\xi \\
\leqslant \int_1^{t/2}\, \frac{t^2\,\xi^{p-2}}{(t-\xi)^2}\, d\xi +
\int_{t/2}^{t-t^p} \,\frac{t^p}{(t-\xi)^2}\, d\xi+\frac{1}{h^2}
\int_{t-t^p}^t\, \frac{t^p}{\xi^{2p}}\,d\xi < \frac 4{1-p}
+1+\frac{4^p}{h^2}.\end{gathered}$$ Аналогично оценивается интеграл $F_4(t)$. Тем самым оценка доказана.
[*Шаг 4.*]{} Докажем сходимость ряда . Так как интервалы $\Delta_n$ и $\Delta_{n+1}$ не пересекаются, то $r_{n+1}\geqslant r_n +b_1(r_n^p+ r^p_{n+1})$. Поэтому при некотором $n_0\in\mathbb{N}$ $$r_{n+1}^{1-p} \geqslant r_n^{1-p}\left(1+ 2b_1\,
r_n^{p-1}\right)^{1-p} > r_n^{1-p}+ b_1(1-p), \quad \text{если}\
n\geqslant n_0.$$ Следовательно, $r_{n+1}^{1-p} - r_n^{1-p}> b_1(1-p)$, если $n\geqslant n_0$. Тогда по индукции при $j\geqslant n$ получаем $$\label{1-p}
r_j^{1-p} - r_n^{1-p}> c(j-n), \qquad c= b_1(1-p), \ \, j\geqslant
n\geqslant n_0.$$ Поэтому $$r_j-r_n \geqslant r_j^p (r_j^{1-p} -r_n^{1-p}) \geqslant c\, r_j^p\,
(j-n), \quad j\geqslant n\geqslant n_0.$$ Меняя $j$ и $n$ местами, получаем $$\label{r}
|r_j-r_n| \geqslant c\max (r_j^p, \, r_n^p)\, |j-n|.$$
Подынтегральная функция в с учетом неравенства оценивается величиной $$\|T(T-\lambda)^{-1} x\|^p\, \|(T-\lambda)^{-1} x\|^{2-p} \, + \,
M\|(T-\lambda)^{-1} x\|^2.$$ Здесь главную роль играет первое слагаемое, поскольку второе слагаемое при $\lambda\to\infty$ является бесконечно малой более высокого порядка. Поэтому интегралы в оцениваются сверху через интегралы $$\label{Intj} \int_{\gamma_j}\, \left(\|T(T-\lambda)^{-1} x\|^p\,
r_j^{\frac{p(p-2)}2}\right)\,\left( r_j^{\frac{p(2-p)}2} \,
\|(T-\lambda)^{-1} x\|^{2-p}\right)\, |d\lambda|, \quad {\mathop{\rm Re}}\,\lambda
= r_j.$$ Теперь для оценки ряда применим сначала неравенство Гёльдера для интегралов, а затем неравенство Гёльдера для сумм. В результате для ряда придем к оценке сверху величиной $$\begin{aligned}
\nonumber &\qquad\sum_{j=1}^\infty\ \left(\int_{\gamma_j}\,
r_j^{p-2}\| T(T-\lambda)^{-1}x\|^2\, |d\lambda|\right)^{p/2}
\left(\int_{\gamma_j}\, r_j^p\|(T-\lambda)^{-1}x\|^2\, |d\lambda|
\right)^{(2-p)/2}\\
\label{Prod} &\quad\leqslant \left(\sum_{j=1}^\infty
\int_{\gamma_j}\, r_j^p\|(T-\lambda)^{-1}x\|^2\,
|d\lambda|\right)^{(2-p)/2} \left(\sum_{j=1}^\infty
\int_{\gamma_j}\, r_j^{p-2}\| T(T-\lambda)^{-1}x\|^2\,
|d\lambda|\right)^{p/2}.\end{aligned}$$ Как и прежде, будем использовать обозначение $e(t) = (E(t)x,\, x)$, где $E(t)$ — спектральная функция оператора $T$. При $\lambda \in
\gamma_j$ имеем $$\|(T-\lambda)^{-1}x\|^2 = \int_0^\infty \frac{de(t)}{(t- r_j)^2 +
\tau^2} \leqslant \int_0^\infty \frac{de(t)}{(t- r_j)^2}, \quad
\lambda = r_j+i\tau \in \gamma_j.$$ Длина отрезка отрезка интегрирования вдоль $\gamma_j$ равна $2hr_j^p$, поэтому первый множитель в оценивается величиной $$\int_0^\infty\, de(t) \left(\sup_{t\in \sigma(T)}
\sum_{j=1}^\infty\, \frac{r_j^{2p}}
{(t-r_j)^2} \right).$$ Достаточно показать, что присутствующий здесь супремум ограничен (так как $e(t)$ — ограниченная монотонная функция). Фиксируем число $t\in \sigma(T)$. Найдется число $n\in \mathbb N$, такое, что $$t\in \Delta_n = [\alpha_{2n-1},\,\alpha_{2n}], \quad\text{а тогда}\
\, r_{n-1} + b_1 r_{n-1}^p\leqslant t \leqslant r_n - b_1 r_n^p.$$ Учитывая оценку , получаем $$\begin{aligned}
\nonumber \sum_{j=1}^\infty\, \frac{r_j^{2p}} {(t-r_j)^2} &=
\frac{r_n^{2p}} {(t-r_n)^2} + \left( \sum_{j=1}^{n-1} +
\sum_{j=n+1}^\infty \right)\frac{r_j^{2p}}
{(t-r_j)^2}\\
\nonumber &\quad \leqslant \frac{r_n^{2p}} {(r_n - b_1r^p - r_n)^2}
+ \left( \sum_{j=1}^{n-1}
+ \sum_{j=n+1}^\infty\right)\, \frac{r_j^{2p}}{(r_n-r_j)^2} \\
\label{r_n} &\quad\leqslant
\frac 1{b_1^2} + \frac 1{c^2} \left( \sum_{j=1}^{n-1}
+ \sum_{j=n+1}^\infty\right)\, \frac 1{(n-j)^2} < C.
$$ Здесь мы учли, что конечное число слагаемых в сумме при $n\leqslant
n_0$, для которых выполняется с другой константой, на оценку не влияет.
Остается оценить второй множитель в . С учетом того, что $\lambda\in \gamma_j$, а длина отрезка $\gamma_j$ равна $2hr_j^p$, для этого множителя находим мажоранту (с точностью до константы) $$\sum_{j=1}^\infty \, r_j^{2p-2} \int_0^\infty \, \frac{\ t^2 \,
dt}{(t- r_j)^2} = \int_0^\infty \, de(t) \left(\sup_{t\in \sigma(T)}
\sum_{j=1}^\infty\, \frac{r_j^{2p-2}\, t^2}{(t - r_j)^2}\right).$$ Нужно лишь показать ограниченность супремума. Фиксируем точку $t\in
\sigma(T)$. Разобьем сумму под знаком супремума на две части и оценим каждую из частей: $$\left(\sum_{2r_j > t} + \sum_{2r_j \leqslant t}\right)\,
\frac{r_j^{2p-2}\, t^2}{(t - r_j)^2} \leqslant 4\sum_j \,
\frac{r_j^{2p}}{(t - r_j)^2} + 4\sum_{2r_j \leqslant t} \,
r_j^{2(p-1)}.$$ Первая сумма уже была оценена в . Вторая сумма конечна, но число слагаемых в ней растет при $t \to \infty$. Для оценки этой суммы используем неравенства . Получаем $$r_j^{(1-p)} \geqslant c(j-j_0) - r_0^{1-p}, \quad \text{для всех} \
\, j\geqslant j_0,$$ где $j_0$ — достаточно большое фиксированное число. Эта оценка влечет неравенство $r_j^{(1-p)} \geqslant (c/2)j$ для всех достаточно больших $j\geqslant j_1$, что обеспечивает сходимость второго ряда. Тем самым доказана сходимость ряда .
[*Шаг 5.*]{} Теперь усложним задачу. Предположим, что выполнено условие , то есть каждый интервал $\Theta_j = (r_j-b_1
r_j^p,\, r_j+ b_1 r_j^p)$ содержит не более $m$ собственных значений $\mu_k$ оператора $T$, пронумерованных с учетом кратности. Тогда интегралы по кривым $\Gamma_\pm$ оцениваются так же, как и прежде. Однако оценку интегралов по вертикальным отрезкам $\gamma_j$ нужно проводить по-другому, поскольку ограниченность оператор-функции $M(\lambda)$ уже нельзя гарантировать. Идея этой оценки состоит в следующем. Мы покажем, что в полосе $\Omega_r$ ширины $2\delta r^p$, определяемой , найдется полоса $\Omega'_r$ ширины $2\delta'r^p$ при некотором $\delta' < \delta$, не зависящем от $ r_j$, в которой выполняется оценка $$\label{Fl}
\| F(\lambda)\| \leqslant C_1, \quad \text{где}\ \, F(\lambda) =
B(A-\lambda)^{-1}, \ \ \lambda \in \Omega'_j \subset \Omega_j.$$ Более того, мы покажем, что полосу $\Omega'_j$ можно выбрать так, что отрезок $\Theta'_j = \Omega'_j \bigcap \mathbb R$ не содержит собственных значений $\mu_k$ оператора $T$. А пока предположим, что это уже доказано. Воспользуемся равенством $$(A-\lambda)^{-1} = (T-\lambda)^{-1} (1- F(\lambda)).$$ Тогда $$\begin{gathered}
\label{1+C}
|(G(\lambda)x,\, x)| = |((T-\lambda)^{-1}\, (1-F(\lambda))\,
S(\lambda)x, \, x)| \\
\leqslant(1+C_1)\|(T-\overline\lambda)^{-1} x\|\,
\|S(\lambda)x\|\leqslant (1+C_1) \|T(T-\lambda)^{-1} x\|^p\,
\|(T-\lambda)^{-1} x\|^{2-p}.\end{gathered}$$ Здесь мы применили оценку и равенство $\|(T-\lambda)^{-1} x\| =\|(T-\overline{\lambda})^{-1}x\|,$ которое справедливо для самосопряженных операторов. Пусть $\gamma'_j$ — вертикальные отрезки, проходящие через точки $r'_j$, являющиеся серединами отрезков $\Theta'_j$ длины $2\delta' r_j^p$. Тогда интегралы по $\gamma'_j$ от функции оцениваются точно так же, как на шаге 4. Разница лишь в том, что вместо слагаемого $1/b^2_1$ в появится константа $
1/(\delta')^2$.
Итак, надо доказать, что существуют полосы $\Omega'_j \subset
\Omega_j$, в которых нет собственных значений оператора $T$ и в которых выполняется оценка . Воспользуемся приемом Мацаева создания искусственной лакуны. Рассмотрим оператор $$\label{K_r}
K_j: = \sum_{\mu_k < r_j} (\mu_k - r_j^-) (\cdot,
\varphi_k)\varphi_k - \sum_{\mu_k \geqslant r_j} (\mu_k - r_j^+)
(\cdot,
\varphi_k)\varphi_k, \quad r_j^\pm := r_j \pm b_1\, r_j^p,$$ где $\varphi_k$ — ортонормированные собственные векторы оператора $T$, отвечающие собственным значениям $\mu_k$, занумерованным с учетом кратностей, а в сумме участвуют только те индексы $k$, для которых собственные значения $\mu_k$ принадлежат отрезку $$\hat{\Lambda}_j =[r^+,\, r^-] \subset \Lambda_j = (\alpha_{2j},\,
\alpha_{2j+1}).$$ Ранг этого оператора не превышает $m$. Положим $T_j:\, = T-K_j$. Оператор $T_j$ остается самосопряженным, сохраняет систему собственных векторов $\{\varphi_k\}_1^\infty$, но меняет собственные значения, лежащие в интервале $\Theta_j$, сдвигая их из этого интервала в ближайший из его концов $r_j^+$ или $r_j^-$.
Воспользуемся равенством $$\label{BA}
B(A-\lambda)^{-1} = B(T_j +B -\lambda)^{-1} \left(1+
L_j(\lambda)\right)^{-1}, \quad L_j(\lambda): =K_j(T_j + B
-\lambda)^{-1}.$$ Далее, $$(T_j +B -\lambda)^{-1} = (T_j -\lambda)^{-1}\left(1 +
S_j(\lambda)\right)^{-1}, \quad S_j(\lambda): = B(T_j-\lambda)^{-1}.$$ Заметим, что при $\lambda \in\Omega_j$ (см. оценку на шаге 2 для оператора $T$, не имеющего собственных значений внутри лакун $\Lambda_j$) $$\begin{gathered}
\label{Sla}
\|S_j(\lambda)\| \leqslant b \|T(T_j -\lambda)^{-1}\|^p\, \|(T_j
-\lambda)^{-1}\|^{1-p} +M\|(T_j-\lambda)^{-1}\| \leqslant \\
b\|TT_j^{-1}\| \|T(T_j -\lambda)^{-1}\|^p\, \|(T_j
-\lambda)^{-1}\|^{1-p} +M\|(T_j-\lambda)^{-1}\| \leqslant\\
\frac b{b_1}\|TT_j^{-1}\|(1+o(1)) \leqslant \frac b{b_1}(1+o(1)),
\quad \lambda \in \Omega_j,\end{gathered}$$ так как $$\|TT_j^{-1}\| \leqslant \frac{r_j}{r_j^-} = \frac{r_j}{r_j +
b_1r_j^p} = 1+o(1) \quad \text{при} \ r_j\to\infty.$$ Но тогда ) получаем $$\label{-1}
\|(1+ S_j(\lambda))^{-1}\| \leqslant
\frac{2(b_1 +b')}{b_1 -b')} < C, \lambda \in \Omega_j, |\lambda| \gg 1.$$ Следовательно, в полосе $\Omega_j$ справедлива оценка $$\label{T_r}
\|(T_j+B-\lambda)^{-1}\| \leqslant \|(T_j - \lambda)^{-1}\|\
\|(1+S_j(\lambda))^{-1}\| \leqslant Cr_j^{-p}, \ \ \lambda\in
\Omega_j.$$ =1000 Здесь мы учли простую оценку для резольвенты самосопряженного оператора $\|(T_j(\lambda))^{-1}\| \leqslant 1/
{\operatorname{dist}}(\lambda,\, \sigma(T_j))$. Из оценок и получаем $$\|B(T_j+B-\lambda)^{-1}\| = \|S_j(\lambda)(1+S_j(\lambda))^{-1}\|
\leqslant C_2, \quad \lambda \in \Omega_j.$$
Итак, учитывая представление , для доказательства неравенств остается оценить оператор-функцию $$\label{finite}
(1+L_j^{-1}(\lambda))^{-1}, \quad \ \, L_j(\lambda) = K_j(T_j +B
-\lambda)^{-1}.$$ Из определения операторов $K_j$ следует, что $\|K_j\| \leqslant
b_1 r^p$. Из получаем $\|L_j(\lambda)\| \leqslant C$. Теперь для оценки оператор-функции воспользуемся Предложением 3, полагая $\mu = -1, K= L_j(\lambda)$. Заметим, что в рассматриваемом случае числитель в правой части оценивается величиной $ (1+C)^m$, так как $s$-числа оператора $K =
L_j(\lambda)$ не превышают его нормы. Знаменатель есть скалярная функция $ D(\lambda):= \det(1+L_j(\lambda))$ — произведение $\leqslant m$ чисел $1+\lambda_k(L_j(\lambda))$, где $\lambda_k(L_j(\lambda))$ — собственные значения оператора$L_j(\lambda)$. Собственные значения также оцениваются постоянной $C$ (нормой оператора $L_j(\lambda)$), поэтому во всей полосе $\Omega_j$ справедлива оценка сверху $|D(\lambda)| \leqslant (1+C)^m$. В силу Предложения 2 имеем $\|(T_r+B-\lambda)^{-1}\| <C\tau^{-1}, \ \tau = {\mathop{\rm Im}}\,
\lambda$, если $\lambda$ лежит на сторонах параболы $P_{p,h},\ h>
b'$ или вне этой параболы. Возьмем $ h= 2Cb_1>b'.$ Тогда $\|L_j(r_j+ihr_j^p)\|\leqslant 1/2$ и, значит, модули всех собственных значений оператора $L_j(r_j+ihr_j^p)$ оцениваются сверху $1/2$. Следовательно, имеется оценка снизу $|D(\lambda)| >
(1-1/2)^m =2^{-m}$ в точке $\lambda = r_j^p +ihr_j^p$.
Далее воспользуемся Предложением 4 для оценки снизу функции $D$. В качестве прямоугольника $\Pi$ возьмем прямоугольник с центром в точке $r_j$, высотой $4hr_j^p$ и шириной $2\delta r_j^p$, а в качестве $\Pi'$ — вдвое меньший прямоугольник с тем же центром. Функция $D$ в прямоугольнике $\Pi$ ограничена постоянной $M=(1+c)^m$, а для постоянной $M'$ в прямоугольнике $\Pi'$ мы получили оценку снизу $M'\geqslant 2^{-m}$. Согласно Предложению 4 функция $D$ оценивается снизу во всем прямоугольнике $\Pi'$ некой постоянной вне исключительного множества кружков с общей суммой радиусов $\varepsilon r_j^p$. Кроме того, число нулей функции $D$ в прямоугольнике $\Pi'$ с учетом кратности не превышает $Cm$. Поэтому $\varepsilon$ можно выбрать столь малым, чтобы полоса $\Pi''$ ширины $c r^p$ в в полосе $\Omega'$ при некотором достаточно малом $c>0$ не пересекала исключительное множество. Тогда в этой полосе $\Pi''$ оператор-функция $(1+L_j(\lambda))^{-1}$, а потому и оператор-функция $B(A-\lambda)^{-1}$ равномерно ограничены постоянной, не зависящей от $r_j \to \infty$. Этим заканчивается доказательство нужных оценок.
[*Шаг 6.*]{} Нужно еще доказать полноту системы проекторов $\{Q_j\}$. Выберем произвольные точки $r'_n \in \mathbb R,
n\geqslant N,$ в полосах $\Pi''=\Pi''_n$, построенных на предыдущем шаге. На прямых ${\mathop{\rm Re}}\,\lambda = r'_n$ оператор-функции $F(\lambda)
= B(A-\lambda)^{-1}$ и $(T-\lambda)^{-1}$ равномерно ограничены. Обозначим через $R_n$ прямоугольник, вертикальные стороны которого проходят через точки $r'_{n-1}$ и $r'_n$, а горизонтальные стороны таковы, что $R_n$ содержит весь спектр обоих операторов $T$ и $A$ между прямыми ${\mathop{\rm Re}}\,\lambda = r'_{n-1}$ и ${\mathop{\rm Re}}\,\lambda = r'_n$. Через $\widehat R_n$ обозначим квадрат с центром в нуле и длиной стороны $2r'_n$. Из представления получаем $$\sum_{j=N}^n Q_jx=-\frac{1}{2\pi i}\sum_{j=N}^n \int_{\partial
R_n}\left(A-\lambda \right)^{-1}x\,\,d\lambda=\sum_{j=N}^n
P_jx+\frac{1}{2\pi i}\sum_{j=N}^n \int_{\partial R_n}G(\lambda)x\,\,
d\lambda,$$ где $P_j$ — ортогональные проекторы Рисса на инвариантные подпространства оператора $T$, такие, что спектр сужения оператора $T$ на эти инвариантные подпространства содержится в отрезках $[r'_{j-1},\, r'_j]$. Через $P_0$ и $Q_0$ обозначим проекторы Рисса операторов $T$ и $A$, отвечающие за спектры, лежащие в полуплоскости ${\mathop{\rm Re}}\,\lambda < r'_N$. Далее для упрощения записи считаем $N=1.$ Положим $$\widehat P_n = \sum_{j=0}^n P_j, \qquad \widehat Q_n = \sum_{j=0}^n Q_j,
\qquad H_n = \hat{P_n}(H).$$ Тогда $$\label{Gn}
\widehat Q_nx = \widehat P_n x + \widehat G_n x, \qquad \text{где}\
\widehat G_n = \frac{1}{2\pi i}\int_{\partial \widehat R_n}\,
G(\lambda)\, d\lambda.$$ Покажем, что $$\label{Gla}
\widehat G_nx: =\int_{\partial \widehat R_n}\, G(\lambda)x\,
d\lambda \to 0 \quad \text{при}\, n\to \infty \ \,\forall\, x\in H.$$ Тогда $$\sum_{j=0}^n Q_jx \to x \ \, \text{при}\ n\to\infty, \ \,
\text{поскольку, очевидно,}\ \sum_{j=0}^n P_jx\to x\ \text{при}\ n\to\infty.$$ Первое соотношение влечет полноту системы проекторов $\{Q_j\}$, поэтому достаточно доказать соотношение .
Пусть $e(t) = (E(t)x,x)$, где $E(t)$ — спектральная функция оператора $T$. Для всех $x\in H_n$ имеем $$\left\|\left(T-\lambda \right)^{-1}x_n\right\|^2=
\int\limits_{-r'_n}^{r'_n}\frac{de(t)}{(\xi -t)^2+\tau^2}, \quad
\lambda=\xi+i\tau,\quad \xi,\tau\in\mathbb{R},$$ откуда получаем (при фиксированном $n$) $$\label{small}
\left\|\left(T-\lambda \right)^{-1}x_n\right\|^2=
O\left(|\lambda|^{-1}\right)\ \ \ \text{при}\ \, |\lambda|\to
\infty.$$ Не ограничивая общности, будем считать оператор $T$ обратимым (иначе можно провести сдвиг спектрального параметра). Так как $H_n$ инвариантно относительно $T$, то оператор $T$ изоморфно отображает $H_n$ на $H_n$. Поэтому для всех $x_n\in H_n$ $$\label{infty}
G(\lambda)x_n = (T-\lambda)^{-1} M(\lambda)BT^{-1}(T-\lambda)^{-1}y_n,
\quad \text{где } \ y_n = Tx_n.$$ Так как оператор-функция $M(\lambda)BT^{-1}$ ограничена на сторонах квадрата $\widehat{R}_n$, то в силу и получаем (при фиксированном $n$) $$\label{j}
\int_{\partial \widehat R_j}\, G(\lambda) x_n\, d\lambda \to
0,\quad \text{если} \ j\to \infty,$$ так как на горизонтальных и левой сторонах квадрата $\widehat R_j$ имеем $\|G(\lambda)x_n\| = O(|\lambda|^{-2})$, а на правой стороне $\|G(\lambda)x_n\| = O(|\lambda|^{-1-p})$ (при $p=0$ соотношение также выполняется, если учесть оценку $\|(T-\lambda)^{-1}\| \leqslant |{\mathop{\rm Im}}\,\lambda|^{-1}$).
Таким образом соотношение имеет место на плотном множестве в $H$ (поскольку множество векторов $x_n\in H_n$ при всех $n$ плотно в $H$). Поэтому для доказательства соотношения достаточно показать, что нормы $\|\widehat G_n\|$ равномерно ограничены. Из сходимости ряда следует равномерная ограниченность $|(\widehat Q_n x,\, x)|$ при всех $x\in
H$. Квадратичная форма $|(\widehat Q_n x,\, x)|$ в гильбертовом пространстве определяет билинейную форму, поэтому $|(\widehat Q_n
x,\, y)| <\infty$ для всех $x,y\in H$. Дважды применяя теорему Банаха-Штейнгауза, получим сначала равномерную ограниченность норм $\|\widehat Q_n x\|$, а затем равномерную ограниченность $\|\widehat
Q_n\|$. Тогда в силу равенства нормы $\|\widehat G_n\|$ также равномерно ограничены. Этим заканчивается доказательство полноты проекторов $\{Q_n\}$.
[*Шаг 7.*]{} Мы рассмотрели случай $T=T^* >0$. Это условие эквивалентно тому, что $T$ — нормальный оператор со спектром на одном луче. Случай, когда оператор $T$ не является полуограниченным, не вносит трудностей. В этом случае, согласно Предложению 2, спектр возмущенного оператора $T+B$ лежит внутри объединения двойной параболы $P_{p, h}$ и круга достаточно большого радиуса. Интегралы по границам этих двойных парабол оцениваются без изменений. Конструкция отрезков $\gamma'_j$, соединяющих ветви левой параболы, и оценки интегралов по этим отрезкам проводятся аналогично. Все остальные рассуждения остаются прежними, только индексацию нужно проводить не по натуральным, а по целым числам.
[99]{}
Gohberg I.C. and M.G.Krein, *Introduction to the Theory of Linear Nonselfadjoint Operators*, Amer. Math. Soc., Providence, RI, 1969. B.Ya.Levin, *Lectures on Entire Functions*, Amer. Math. Soc., Providence, RI, 1996.
А.С.Маркус, *Введение в спектральную теорию полиномиальных операторных пучков*, Изд-во Штиница Кишинев, 1986. English translation: *Introduction to the Spectral Theory of Polynomial Operator Pencils*, Amer. Math. Soc., Providence, RI, 1988.
А.С.Маркус, В.И.Мацаев, “О сходимости разложений по собственным векторам оператора, близкого к самосопряженному”, *Математические исследования* **61** (1981), 104–129. А.С.Маркус, В.И.Мацаев, “Теоремы сравнения спектров линейных операторов и спектральные асимптотики”, *Труды Моск. матем. общества* **45** (1982), 133–181. English translation in: Transactions of Moscow Math. Soc. **1** (1984), 139–187.
A.K.Motovilov and A.A.Shkalikov, “Unconditional bases of subspaces related to non-self-adjoint perturbations of self-adjoint operators”, *Eurasian Mathematical Journal* **8**:1 (2017), 119–127.
А.А.Шкаликов, “Возмущения самосопряженных и нормальных операторов с дискретным спектром”, *Успехи матем. наук* **71**:5(431) (2016), 113–174. English translation in: *Russian Math. Surveys* **71**:5 (2016), 907–964.
A.A.Shkalikov, “Estimates of meromorphic functions and summability theorems”, *Pacific J. Math.* **103**:2 (1982), 569–582.
А.А.Шкаликов, “Теоремы тауберова типа о распределении нулей голоморфных функций”, *Матем. сборник* **123**(165):3 (1984), 317 –347. English translation in: *Math. USSR Sb.* **51**:2 (1985), 315–344.
Alexander Konstantinovich Motovilov\
Bogoliubov Laboratory of Theoretical Physics\
Joint Institute for Nuclear Research\
Joliot-Curie 6\
141980 Dubna, Moscow Region, Russia\
E-mail: [email protected]
and
Faculty of Natural and Engineering Sciences\
Dubna State University\
Universitetskaya 19\
141980 Dubna, Moscow Region, Russia
Andrei Andreevich Shkalikov\
Faculty of Mathematics and Mechanics\
Lomonosov Moscow State University\
Leninskiye Gory 1\
119991 Moscow GSP-1, Russia\
E-mail: [email protected]
[^1]: This work was supported by the Deutsche Forschungsgemeinschaft (DFG) and by the Russian Foundation for Basic Research (RFBR).
[^2]: Условие в Теореме 1 эквивалентно тому, что лакуны $\Lambda_k$ содержат интервалы $(r_k -b_1 |r_k|^p,\, r_k
+b_1 |r_k|^p)$, где $r_k =(\alpha_{2k}+\alpha_{2k+1})/2$ — середины лакун $\Lambda_k$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The SPIRE Photometer Simulator reproduces the entire Herschel-SPIRE system in a modular IDL program. Almost every aspect of the operation of SPIRE can be investigated in a systematic way to ensure that observations are performed in the most efficient way possible when Herschel flies. This paper describes some of the work done with the Simulator to help prepare for large observing programs such as deep extra-galactic, high-redshift surveys.'
author:
- 'T. J. Waskett, B. Sibthorpe, M. J. Griffin'
title: 'Simulations of Deep Extra-galactic Surveys with Herschel-SPIRE'
---
Introduction
============
Expensive space missions with limited operating lifetimes, such as Herschel [@pilbratt], require efficient operation in order to obtain the best possible value for money in terms of the quantity and quality of astronomical data obtained. Careful preparation of every aspect of the mission needs to be undertaken, including how the various instruments on board the spacecraft are operated to achieve optimum performance. SPIRE [@matt1] is one of three scientific instruments on board the Herschel spacecraft and covers the longer wavelength range from $\sim~200 - 600~\mu$m with both a photometer and an imaging Fourier Transform Spectrometer (FTS). This paper describes how the operation of the photometer half of the SPIRE instrument has been optimised through the use of an instrument Simulator, and how the same Simulator can be used to prepare for large observing programmes.
The SPIRE Photometer Simulator
------------------------------
The Simulator is an IDL coded virtual version of the photometer half of the Herschel-SPIRE system. It incorporates as many of the physical instrumental and telescope characteristics as it is possible to include in a computationally practical and user-friendly program. Full details of the individual modules and their interaction with each other are given in @bruce1.
Briefly, the user creates a suitable input sky – as realistic or fantastic as desired – for each of the three SPIRE bands. These are fed into the Simulator where they are convolved with a representative beam profile and then ’observed’ with the bolometer detector arrays. Parameters for the observation are predefined by the user in the same way that a real observation would be planned. The astronomical power from the sky and the background radiation from the telescope and internal instrument components are all passed into a module containing a model of the individual detectors, which calculates their response to the incident radiation. This bolometer model also calculates and superimposes realistic noise on the output detector time-lines. The detector time-lines are then low-pass filtered and sampled, in the same way as done by the on-board electronics, to produce output voltage time-lines. Additionally, a pointing time-line is generated based on the observation parameters.
Optimisation of Observing Modes
===============================
SPIRE has several modes of operation [@matt1], but the primary mapping modes are scan and jiggle. The SPIRE arrays are hexagonally close-packed, feedhorn coupled bolometers, with a detector spacing of twice the beam full width half maximum (FWHM), so they do not fully sample the sky in a single pointing. The two mapping modes account for this in different ways. Jiggle map mode is used to observe individual $4\arcmin \times 4\arcmin$ fields by pointing the telescope in one direction and ’jiggling’ the instrument’s internal beam steering mirror to fill the whole field area with data. While large maps can be created by tiling together individual jiggle map fields it is more observationally efficient to use scan map mode, which involves scanning the telescope across the sky at a constant rate to create a strip of data. By decelerating the telescope at the end of one strip, slewing perpendicular to the scan direction and accelerating back up to scanning speed in the opposite direction, a large map can be built up with successive, adjacent scan strips. The detector arrays are rotated slightly, relative to the scan direction, so that the gaps between adjacent detectors are filled by other detectors following on behind.
Because of the nature of the bolometer detectors scan map data time-lines are subject to $1/f$ noise – excess noise at low frequencies, or drifts on long time-scales – that cause ’stripes’ in the resultant maps (see left hand panel of Figure \[pic\]. Jiggle map mode is immune to $1/f$ noise because the signal is ’chopped’, so the map is the difference between on- and off-source signals. @bruce2 describe how the Simulator has been used to determine the optimum parameters for operating SPIRE in scan map mode, principally how the competing effects of the low-pass electronics filter and $1/f$ noise lead to an optimum scan speed around $30\arcsec$s$^{-1}$. Additionally, the angle of rotation between the scan direction and the array orientation, and the spacing between successive scan strips have been optimised to produce highly uniform sky coverage.
Preparation for Deep Extra-Galactic Surveys
===========================================
With the scan map mode parameters optimised we can investigate how instrumental characteristics will affect actual observations. Part of the SPIRE Guaranteed Time (GT) programme will involve deep extra-galactic, high-redshift surveys that cover large areas of sky down to, and even below, the confusion limit. To see how $1/f$ noise is likely to affect the sensitivity of such surveys to faint point sources we designed a series of ’observations’ to be carried out with the Simulator. We used the GALICS model of galaxy evolution [@galics] to produce input skies of a suitably realistic extra-galactic cosmological field covering 1 sq. deg. The field was populated with 58590 sources down to a $250~\mu$m band flux limit of 2 mJy (a confusion limit of 40 beams per source is $\sim 1000$ sources per sq. deg.) Since the GALICS model is based on a hierarchical galaxy formation model the clustering of the sources is more realistic than a simple Poisson distribution, and so the confusion noise should be closer to what will be experienced by SPIRE in flight.
Two sets of simulated observations were performed, one with the $1/f$ noise switched off, leaving simply white noise in the detector time-lines, and the other with the $1/f$ noise switched on and at a level that matches the design requirement of the SPIRE bolometers (a knee frequency of 100 mHz.) This is a worst case scenario for the real instrument as the majority of the detectors exhibit less $1/f$ noise than the requirement. The final sensitivity of the combined simulations (for each noise test) was designed to approximate a confusion limited observation.
A source detection routine was run on naive maps created from each noise test to compare the relative sensitivity of the observations with and without $1/f$ noise. The extracted source list was then cross-referenced with the input source catalogue. As expected, the map with the $1/f$ noise showed a significantly higher detection threshold ($\sim 30\%$) and with a higher fraction of spurious source detections. Therefore, if left untreated, $1/f$ noise could seriously affect such a survey. Integration times would need to be $\sim 70\%$ longer to reach the desired detection threshold than expected from the sensitivity estimates, which currently assume no contribution from $1/f$ noise.
However, there is much that can be done to alleviate $1/f$ noise, for example iterative map making methods borrowed from CMB analysis, or filtering schemes. High-pass filtering involves removing all the low frequency modes from a Fourier transform of the time-line data from each detector. This eliminates the long period drifts associated with $1/f$ noise, effectively removing the stripes in scan maps. Some point source flux is also lost in this process however, so the cut-off frequency must be a compromise between this and removing as much $1/f$ noise as possible.
We devised the following scheme to filter the data, based on common cleaning routines used in radio astronomy: first construct a map from the unfiltered data; detect the brightest sources and subtract a point source model for each source from the time-series data; Fourier transform and high-pass filter the data, using a cut-off frequency of half the $1/f$ knee frequency (50 mHz in this case); inverse Fourier transform the data back into time-series and add the bright point sources back on; finally, re-create a map from this filtered data. Removing the bright point sources before filtering prevents negative dips appearing in the map either side of the sources in the scan direction. This reduces the likelihood of the filtering process adversely affecting fainter source flux recovery.
Having made a new map we ran the source detection routine again. The situation was much improved with the detection threshold now only slightly higher than the white noise case. The number of spurious source detections was also much reduced, although still higher than the white noise case. Based on this analysis we believe that the level of $1/f$ noise expected for SPIRE will not cause a significant problem for these types of faint point source surveys. Simple filtering schemes can go a long way towards reducing $1/f$ noise effects while more sophisticated methods may eliminate it entirely. See figure \[pic\] for a comparison of pre- and post-filtered images.
Summary and Future Work
=======================
The SPIRE photometer Simulator is a very powerful tool for investigating the consequences of instrumental effects on the quality of astronomical data to be obtained with SPIRE. Not only has it helped in the optimisation of the observing modes but it is also providing a head start in understanding the data and how it will be best processed when the real thing becomes available.
As well as simulating cosmological surveys the Simulator is also being used to simulate galactic (Sibthorpe, in preparation) and local Universe observations. Together, these simulations are central to the selection and development of the map making code to be delivered as part of the SPIRE pipeline processing package. This process is currently underway and as launch approaches the Simulator will be used to make further simulations to help refine the performance of the map making algorithm as well as helping to develop optimised source detection routines capable of fully exploiting the SPIRE data.
Finally, being a modular program the Simulator is fully customisable and can be modified to represent any other facility of a similar design to SPIRE, or even to help design the next generation of far-IR observatories.
Griffin, M. J., & Swinyard, B. M., et al., 2006, SPIE, 6265 Hatton, S., Devriendt, J. E. G., Ninin, S., Bouchet, F. R., Guiderdoni, B., & Vibert, D, 2003, MNRAS, 343, 75 Pilbratt, G., L., 2006, SPIE, 5487, 401 Sibthorpe, B., Woodcraft, A., Griffin, M., & Watkin, S. 2004, SPIE, 5487, 491 Sibthorpe, B., Waskett, T., & Griffin, M. 2006a, SPIE, 6270, 41
| {
"pile_set_name": "ArXiv"
} |
by -by -
8.9in
0.5in
-.6in ${\left(}
\def$[)]{} ł Ł ø Ø
[*Elso Drigo Filho $^a$ and Regina Maria Ricotta $^b$*]{}\
$^a$ Instituto de Biociências, Letras e Ciências Exatas, IBILCE-UNESP\
Rua Cristovão Colombo, 2265 - 15054-000 São José do Rio Preto - SP\
$^b$ Faculdade de Tecnologia de São Paulo, FATEC/SP-CEETPS-UNESP\
Praça Fernando Prestes, 30 - 01124-060 São Paulo-SP\
Brazil\
\
The formalism of Supersymmetric Quantum Mechanics supplies a trial wave function to be used in the Variational Method. The screened Coulomb potential is analysed within this approach. Numerical and exact results for energy eigenvalues are compared.\
[**I. Introduction**]{}\
The screened Coulomb potential has been used in several branches of Physics, for instance, in nuclear physics (as the name of Yukawa potential), in plasma and in the study of electrolytic solution properties (Debye-Huckel potential). The Schr$\ddot {o}$dinger equation for this potential is not exactly solvable and exact numerical, [@Rogers], [@Roussel] and approximative, [@Imbo]-[@Dutt] methods have been applied to obtain the energy eigenvalues, including variational calculations, [@Greene]-[@Fessatidis].
More recently, a new methodology based within the variational method associated to supersymmetric quantum mechanics formalism has been introduced, [@Gozzi]-[@Drigo]. References [@Gozzi] and [@Cooper1] introduce a scheme based in the hierarchy of Hamiltonians; it permits the evaluation of excited states for one-dimensional systems. In reference [@Drigo] an [*ansatz*]{} for the superpotential which is related to the trial wave function is proposed. The new approach showed to be useful to get answers when applied to atomic systems, [@Drigo]-[@Drigo1].
In this letter, the variational energy eigenvalues for the static screened Coulomb potential in three dimensions are determined using the variational method using a trial wave function induced by Supersymmetric Quantum Mechanics, SQM.
In the approach followed here the first step taken is to look for an effective potential similar to the original screened Coulomb potential. Inspired by SQM, an [*ansatz*]{} is made to the superpotential which determines the variational (trial) wavefunction through the algebraic approach of SQM.
Our system is three dimensional and in this case it is possible to determine the variational eigenfunctions for each value of angular momentum $\it l$. The first eigenfunction, obtained from direct factorization of the effective Hamiltonian, corresponds to the minimum energy for each $\it l$.
This new methodology has been successfully applied to other atomic systems such as the Hulthén, [@Drigo], and the Morse, [@Drigo1], potentials. Here it is applied to the screened Coulomb potential.
In what follows, we briefly introduce, for completeness, the SQM scheme, then introduce the variational method and show our results.\
[**II. Supersymmetric Quantum Mechanics**]{}\
In SQM, [@Cooper1]-[@Drigo], for $N=2$ we have two nilpotent operators, $Q$ and $Q^+$, that satisfying the algebra { Q, Q\^+} = H\_[SS]{} ; Q\^2 = [Q\^+]{}\^2 = 0, where $H_{SS}$ is the supersymmetric Hamiltonian. This algebra can be realized as Q = (
[cc]{} 0 & 0\
A\^- & 0
) , Q\^+ = (
[cc]{} 0 & A\^+\
0 & 0
) where $A^{\pm}$ are bosonic operators. With this realization the supersymmetric Hamiltonian $H_{SS}$ is then given by H\_[SS]{} = (
[cc]{} A\^+A\^- & 0\
0 & A\^-A\^+
) = (
[cc]{} H\^+ & 0\
0 & H\^-
). where $H^{\pm}$ are called supersymmetric partner Hamiltonians and share the same spectra, apart from the nondegenerate ground state. Using the super-algebra, a given Hamiltonian can be factorized in terms of the bosonic operators. In $\hbar = c = 1$ units, it is given by H\_1 = -[1ø2]{}[d\^2 ød r\^2]{} + V\_1(r) = A\_1\^+A\_1\^- + E\_0\^[(1)]{} where $ E_0^{(1)}$ is the lowest eigenvalue and the function $V_1(r)$ includes the barrier potential term. The bosonic operators are defined by A\_1\^ = [1ø2]{}( + W\_1(r) ) where the superpotential $W_1(r)$ satisfies the Riccati equation \[Riccati\] W\_1\^2 - W\_1’= 2V\_1(r) - 2E\_0\^[(1)]{} which is a consequence of the factorization of the Hamiltonian $H_1$.
The eigenfunction for the lowest state is related to the superpotential $W_1$ by \[eigenfunction\] \_0\^[(1)]{} (r) = N exp( -\_0\^r W\_1(|r) d|r). Now it is possible to construct the supersymmetric partner Hamiltonian H\_2 = A\_1\^-A\_1\^+ + E\_0\^[(1)]{} = -[1ø2]{}[d\^2 ød r\^2]{} + [1ø2]{}(W\_1\^2 + W\_1’)+ E\_0\^[(1)]{} . If one factorizes $H_2$ in terms of a new pair of bosonic operators, $A_2^{\pm}$ one gets, H\_2 = A\_2\^+A\_2\^- + E\_0\^[(2)]{} = -[1ø2]{}[d\^2 ød r\^2]{} + [1ø2]{}(W\_2\^2 - W\_2’)+ E\_0\^[(2)]{} where $E_0^{(2)} $ is the lowest eigenvalue of $H_2$ and $W_2$ satisfy the Riccati equation, W\_2\^2 - W\_2’= 2V\_2(r) - 2E\_0\^[(2)]{} . Thus a whole hierarchy of Hamiltonians can be constructed, with simple relations connecting the eigenvalues and eigenfunctions of the $n$-members, [@Sukumar1], [@Cooper2].
Thus, the formalism of SQM allows us to evaluate the ground state eigenfunction from the knowledge of the superpotential $W(r)$, satisfying the Riccati equation, eq.(\[Riccati\]). However, since the potential is not exactly solvable, the purpose is to propose an [*ansatz*]{} for the superpotential and, based in the superalgebra information, we evaluate a trial wavefunction that minimizes the expectation value of the energy. The energy eigenvalues pursued are evaluated by minimizing the energy expectation value with respect to a free parameter introduced by the [*ansatz*]{}.\
[**III. Trial wavefunction for the variational calculation**]{}\
The screened Coulomb potential is given, in atomic units, by \[Coulomb\] V\_[SC]{} = - [e\^[-r]{} ør]{} where $\delta$ is the screened length. The associated radial Schr$\ddot
{o}$dinger equation includes the potential barrier term and it is given by \[Hamiltonian\] (-[1ø2]{}[d\^2 ød r\^2]{} - [e\^[-r]{} ør]{} + [ l(l + 1) ø2r\^2]{}) = E where the unit length is $\hbar^2 /me^2 $ and the energy unit is $\epsilon_0 = - me^4/\hbar^2$.
In order to determine an effective potential similar to the potential in the Hamiltonian (\[Hamiltonian\]), that is the screened Coulomb potential plus the potential barrier term, the following [*ansatz*]{} to the superpotential is suggested \[superpotential\] W(r) = - (l+1)[e\^[-r]{} ø1 - e\^[-r]{} ]{}+ [1ø(l+1)]{} - [ø2]{}. Substituting it into (\[eigenfunction\]), one gets \[Psi\] \_0 (r) = (1-e\^[-r]{})\^[l+1]{} e\^[- ([1ø(l+1)]{} - [ø2]{})r]{}. Assuming that the radial trial wave function is given by (\[Psi\]), replacing $\d $ by the variational parameter $\mu $, i.e., \[Psimu\] \_ (r) = (1-e\^[-r]{})\^[l+1]{} e\^[ -([1ø(l+1)]{} - [ø2]{})r]{}, the variational energy is given by \[energymu\] E\_ = [\_0\^ \_(r) \[ - [ e\^[-r]{}ør]{} + [l(l+1)ø2r\^2]{}\] \_(r) dr ø\_0\^ \_(r)\^2 dr]{}. Thus, minimizing this energy with respect to the variational parameter $\mu$ one obtains the best estimate for the energy of the screened Coulomb potential.
As our potential is not exactly solvable, the superpotential given by eq.(\[superpotential\]) does not satisfy the Riccati equation (\[Riccati\]) but it does satisfy it for an effective potential instead, $V_{eff}$ V\_[eff]{}(r) = [|W\_1\^2 - |W\_1’ ø2]{}+ E(|) where $ \bar W_1 = W_1(\d=\bar\mu)$ is given by eq.(\[superpotential\]) and $\bar\mu$ is the parameter that minimises the energy expectation value, (\[energymu\]). It is given by \[effective\] V\_[eff]{}(r) = - [e\^[-r]{}ø1-e\^[-r]{}]{} + [l(l+1)ø2]{}[\^2 e\^[-2r]{}ø(1-e\^[-r]{})\^2]{} + [1ø2]{}([1øl+1]{} - [ø2]{})\^2 + E(), where $\d = \bar\mu$ that minimises energy expectation value. One observes that for small values of $\d$ the first term is similar to the potential (\[Coulomb\]) and the last is approximately the potential barrier term. This observation allows us to conclude that the superpotential (\[superpotential\]) can be used to analyse the three dimensional screened Coulomb potential variationally through the trial wavefunction (\[Psi\]).\
[**IV. Results**]{}\
For $l=0$ the effective potential (\[effective\]) becomes identical to the Hulthén potential. Thus, the results presented in Table 1 coincide with those of ref.[@Lam], where the Hulthén potential eigenfunctions are directly used in the variational calculation. The deviation on the fifth decimal algarism can be attributed to the accuracy of the numerical calculation.
[|c|c|c|c|]{}\
& SQM Variational & Variational (Ref. 7) & Exact Numerical\
0.001 & -0.49902 & -0.49900 & -\
0.002 & 0.49802 & -0.49800 & -0.4980\
0.005 & -0.49504 & -0.49502 & -0.4950\
0.010 & -0.49009 & -0.49007 & -0.4901\
0.02 & -0.48031 & -0.48030 & -0.4803\
0.025 & -047548 & -0.47546 & -0.4755\
0.03 & -0.47068 & -0.47066 & -\
0.04 & -0.46119 &-0.46117 & -\
0.05& -0.45180 &-0.45182 & -0.4518\
0.06 & -0.44259 & -0.44260 & -\
0.07 & -0.43351 & -0.43352 & -\
0.08 & -0.42456 & -0.42457 & -\
0.09 & -0.41574 & -0.41575 & -\
0.10 & -0.40705 & -0.40706 & -0.4071\
0.20 & -0.32681 & -0.32681 & -0.3268\
0.25 & -0.29092 & -0.29092 & -0.2909\
0.30 & -0.25764& -0.25763 & -\
0.40 & -0.19842 & -0.19836 & -\
0.50 & -0.14806 & -0.14808 & -0.1481\
0.60 & -0106077 & -0.10608 & -\
0.70 & -0.07175 & -0.07174 & -\
0.80 & -0.04459 & -0.04459 & -\
0.90 & -0.02420 & -0.02418 & -\
1.00 & -0.01026 & -0.01016 & -0.01029\
1.05 & -0.00568 & -0.00544 & -\
\
Energy eigenvalues as function of the screening parameters $\d$ for $1 s$ state ($l=0$), in rydberg units of energy. Comparison is make with variational and exact numerical results from [@Rogers] and [@Lam].\
The results become more interesting for $l\not=0$. In this case the effective potential differs from the Hulthén potential. Table 2 shows the results for $2p$ ($l=1$), $3d $ ($l=2$) and $4f$ ($l=3$) energy levels. Also given in this table are the correponding numerical results [@Rogers].\
\
.5cm
------------------------ ----------------- --------------- ----------------- --------------- ----------------- ---------------
[$ \;\;\delta \;\;$]{} [Variational]{} [Numerical]{} [Variational]{} [Numerical]{} [Variational]{} [Numerical]{}
0.001 -0.2480 -0.2480 -0.10910 -0.10910 -0.06051 -0.06052
0.005 - - - - -0.52930 -0.05294
0.010 -0.2305 -0.2305 -0.09212 -0.09212 -0.04419 -0.04420
0.020 -0.2119 -0.2119 -0.07503 -0.07503 -0.02897 -0.02898
0.025 -0.2030 -0.2030 -0.06714 -0.06715 - -
0.050 -0.1615 -0.1615 -0.03374 -0.03383 - -
0.100 -0.09289 -0.09307 - - - -
------------------------ ----------------- --------------- ----------------- --------------- ----------------- ---------------
\
Energy eigenvalues as function of the screening parameters $\d$ for $2p$ ($l=1$), $3d $ ($l=2$) and $4f$ ($l=3$) states, in rydberg units of energy. Variational values obtained by the trial function (\[Psimu\]) are compared with exact numerical results obtained from reference [@Rogers], (see also [@Greene]).\
[**V. Conclusions**]{}\
We have proposed trial wavefunctions to be used in the variational calculation in order to determine the energy eigenvalues of the screened Coulomb potential. These functions were induced from supersymmetric quantum mechanics formalism. The approach consists of making an [*ansatz*]{} in the superpotential which satifies the Riccati equation by an effective potential. The trial wavefunctions were then determined from this superpotential through the superalgebra.
For $l=0$ the effective potential obtained is identical to the Hulthén potential. However for $l\not=0$ the potential has a new structure. The trial wavefunctions suggested for this case are different from those proposed in references [@Greene]-[@Fessatidis]. Within our framework the energy eigenvalue for each value of $l$ is obtained using the same function (\[Psi\]).
In terms of the hierarchy of Hamiltonians, we obtained the first member for each value of $l$. Other members can be determined from the usual approach in supersymmetric quantum mechanics, [@Cooper2].
One observes that the results obtained are in very good agreement to those found in the literature. The results are better for small values of the parameter $\d$. This observation is justified by the fact that for small values of $\d $ the effective potential is more similar to the original potential than for higher values of $\d$.
We stress that even though the problem has been attacked by different methods our new methodology is very simple to supply accurate results. We believe that other applications to atomic physics problems can be made by this new method.\
[99]{} K. M. Rogers, H. C. Graboske Jr. and D. J. Harwood, Phys. Rev. [**A1**]{} (1970) 1577 K. M. Roussel and R. F. O’Connell, Phys. Rev. [**A4**]{} (1971) 52 T. Imbo, A. Pagnamenta and U. P. Sukhatme, Phys. Lett. [**A105**]{} (1984) 183 P. Chaudhury, S. P. Bhattacharyya, Chem. Phys. Lett. [**296**]{} (1998) 51 R. Dutt, M. Bag and Y. P. Varshni, J. Phys. B: At. Mol. Opt. Phys. [**21**]{} (1988) 927 R. L. Greene and C. Aldrich, Phys. Rev. [**A14**]{} (1976) 2363 C. S. Lam and Y. P. Varshni, Phys. Rev. [**A4**]{} (1971) 1875 V. Fessatidis, J. Mancini, Q. Haider, Y. Zhou and L. Greco, Phys. Lett. [**A242**]{} (1998) 74 E. Gozzi, M. Reuter and W. D. Thacker, Phys. Lett. [**A183**]{} (1993) 29 F. Cooper, J. Dawson and H. Shepard, Phys. Lett. [**187A**]{} (1994) 140 E. Drigo Filho and R. M. Ricotta, Mod. Phys. Lett. [**A10**]{} (1995) 1613 E. Drigo Filho and R. M. Ricotta, hep-th 9910254 C. V. Sukumar, J. Phys. A: Math. Gen. [**18**]{} (1985) L57 E. Drigo Filho and R. M. Ricotta, Mod. Phys. Lett. [**A14**]{} (1989) 2283 F. Cooper, A. Khare and U. P. Sukhatme, Phys. Rep. [**251**]{} (1995) 267
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Victor P. Utrobin'
date: 'Received 21 July 2006 / accepted 22 August 2006'
title: An optimal hydrodynamic model for the normal Type IIP supernova 1999em
---
Introduction
============
The supernova (SN) 1999em was discovered by the Lick Observatory Supernova Search on October 29.44 UT in the nearly face-on SBc galaxy NGC 1637 (Li [@li99]). Detected shortly after the explosion at an unfiltered magnitude of $\sim 13.5$, SN 1999em was bright enough to be observed well both photometrically and spectroscopically for more than 500 days (Hamuy et al. [@hpm01]; Leonard et al. [@lfg02]; Elmhamdi et al. [@edc03]). SN 1999em was the first Type II-plateau supernova (SN IIP) detected at both X-ray and radio wavelengths, being the least radio luminous and one of the least X-ray luminous SNe (Pooley et al. [@plf02]). The X-ray data indicated the interaction between SN ejecta and a pre-SN wind with a low mass-loss rate of $\sim 2 \times 10^{-6}
M_{\sun}\,yr^{-1}$. Leonard et al. ([@lfab01]) presented the first spectropolarimetry of SN IIP based on the optical observations of SN 1999em during $\sim 160$ days after SN discovery. The weak continuum polarization increasing from $p \approx 0.2\%$ on day 7 to $p \approx 0.5\%$ in the final observations was detected with an unchanging polarization angle $\theta \approx 160^{\circ}$. To date this event has become the most studied normal SN IIP.
In order to study properly any individual object, an accurate distance is required in addition to high quality observations. It is this point that is not clear for SN 1999em. There was a reasonable agreement in determining the distance to SN 1999em using the expanding photosphere method (EPM; Kirshner & Kwan [@kk74]): $7.5 \pm 0.5$ Mpc (Hamuy et al. [@hpm01]), $8.2 \pm 0.6$ Mpc (Leonard et al. [@lfg02]), and $7.838 \pm 0.331$ Mpc (Elmhamdi et al. [@edc03]). However, Leonard et al. ([@lknt03]) identified 41 Cepheid variable stars in the galaxy NGC 1637 and found that the Cepheid distance to the host galaxy was $11.7 \pm 1.0$ Mpc. This distance to SN 1999em is $\sim 50\%$ larger than the earlier EPM distance estimates. Assuming a correlation between the total $^{56}$Ni mass and the explosion energy for SNe IIP, Nadyozhin ([@n03]) evaluated the distance to the galaxy NGC 1637 of 11.08 Mpc close to the Cepheid distance. Using the spectral-fitting expanding atmosphere model (SEAM), Baron et al. ([@bnbh04]) obtained the distance to SN 1999em of $12.5 \pm 1.8$ Mpc in a good agreement with the Cepheid distance scale. Finally, studying various ingredients entering the original EPM, Dessart & Hillier ([@dh06]) improved this method and also achieved better agreement to the Cepheid distance with an estimate of $11.5 \pm 1.0$ Mpc.
Starting the investigation of SN 1999em, a normal SN IIP, it is impossible to pass by the well-known and studied SN 1987A in the Large Magellanic Cloud (LMC), a peculiar SN IIP. Recent progress in reproducing the bolometric light curve observed in SN 1987A with a modern hydrodynamic model (Utrobin [@utr04]) and in modelling the H$\alpha$ profile and the Ba II 6142 Å line in SN 1987A at the photospheric phase with a time-dependent approach for the SN atmosphere (Utrobin & Chugai [@uc02], [@uc05]) makes this experience very instructive for other SNe IIP. First, it turned out that a good agreement between the hydrodynamic models and the photometric observations of SN 1987A did not guarantee a correct description of this phenomenon as a whole. Second, the strength of the H$\alpha$ line and its profile provided hard constraints on the hydrodynamic models. The most important lesson from this study of SN 1987A is that we have to take *both* the photometric *and* spectroscopic observations into account to obtain an adequate hydrodynamic model (Utrobin [@utr05]). Now such an approach should be applied to SN 1999em.
Here we present the comprehensive hydrodynamic study of SN 1999em complemented by the atmosphere model with the time-dependent kinetics and energy balance. A brief description of the hydrodynamic model and the atmosphere model based on it is given in Sect. \[sec:snmodel\]. The study of SN 1999em we begin with a construction of the optimal hydrodynamic model (Sect. \[sec:optmod\]). Then we continue with a question: at what distance to SN 1999em, short or long, the hydrodynamic model and the proper atmosphere model are consistent with the photometric and H$\alpha$ line observations (Sect. \[sec:evidis\]). The time development of the optimal hydrodynamic model is presented in Sect. \[sec:devmod\], and its general regularities in Sect. \[sec:genpro\]. The basic relationships between the physical and observed parameters for SNe IIP similar to the SN 1999em event are obtained in Sect. \[sec:phyobs\], while in Sect. \[sec:sn87a\] we address the comparison of SN 1999em with SN 1987A. In Sect. \[sec:disc\] we discuss our results from the theoretical and observational points of view. Finally in Sect. \[sec:concl\] we summarize the results obtained.
We adopt here a distance of 11.7 Mpc (Leonard et al. [@lknt03]), a recession velocity to SN 1999em of 800 kms$^{-1}$ (Leonard et al. [@lfg02]), an explosion date of JD 2451476.77, and a total extinction $A_V=0.31$ (Baron et al. [@bbh00]; Hamuy et al. [@hpm01]; Elmhamdi et al. [@edc03]).
Supernova modelling and input physics {#sec:snmodel}
=====================================
Keeping in mind the importance of a hydrodynamic study and of atmosphere modelling, including the time-dependent kinetics and energy balance for the interpretation of the SN 1987A phenomenon, we use this approach to investigate SN 1999em.
A hydrodynamic model is computed in terms of radiation hydrodynamics in the one-group approximation taking into account non-LTE effects on the average opacities and the thermal emissivity, effects of nonthermal ionization, and a contribution of lines to opacity as in the case of SN 1987A (Utrobin [@utr04]). Note that the bolometric luminosity of an SN is calculated by including the retardation and limb-darkening effects.
The atmosphere model includes the time-dependent ionization and excitation kinetics of hydrogen and other elements, the time-dependent kinetics of molecular hydrogen, and the time-dependent energy balance (Utrobin & Chugai [@uc05]). The density distribution, chemical composition, radius of the photosphere, and effective temperature are provided by the corresponding hydrodynamic model. The obtained time-dependent structure of the atmosphere is then used to calculate synthetic spectra at selected epochs. The spectra are modelled by the Monte Carlo technique suggesting that the photosphere diffusively reflects the incident photons and that the line scattering is generally non-conservative and is described in terms of the line scattering albedo. The Thomson scattering on free electrons and Rayleigh scattering on neutral hydrogen are also taken into account.
Optimal hydrodynamic model {#sec:optmod}
==========================
Elmhamdi et al. ([@edc03]) have constructed the “UBVRI” bolometric light curve of SN 1999em from the corresponding photometric data. To account for a possible contribution of the missing infrared bands, we add a value of 0.19 dex, taken from Elmhamdi et al. ([@edc03]), to the “UBVRI” luminosity and adopt the resultant light curve as the bolometric light curve of SN 1999em. Our aim is to find an adequate hydrodynamic model that reproduces the photometric and spectroscopic observations of SN 1999em. To fit the bolometric light curve of SN 1999em, various hydrodynamic models were explored. The bolometric light curve of SN 1999em is fitted by adjusting the pre-SN radius $R_0$, the ejecta mass $M_{env}$, and the explosion energy $E$, along with the density distribution in the pre-SN model and its chemical composition in the transition region between the hydrogen-rich envelope and metal/helium core. The radioactive $^{56}$Ni mass is reliably measured by the light curve tail after the plateau phase. As the pre-SN model of SN 1999em, we consider non-evolutionary models similar to those of SN 1987A (Utrobin [@utr04]), but for the outer layers assume the standard solar composition with the mass fraction of hydrogen $X=0.735$, helium $Y=0.248$, and the metallicity $Z=0.017$ (Grevesse & Sauval [@gs98]) taking a normal spiral nature of the host galaxy NGC 1637 into account. The best version of such a fitting was obtained with the optimal model that is characterized by the basic parameters: the pre-SN radius of 500 $R_{\sun}$, the ejecta mass of 19 $M_{\sun}$, and the explosion energy of $1.3\times10^{51}$ erg (model D11 in Table \[tab:hydmods\]).
The density profile of the pre-SN model consisting of a central white dwarf like core and an outer envelope with the size of the red supergiant is shown in Fig. \[fig:rhomr\]. In the calculations, the 1.58 $M_{\sun}$ central core is removed from the computational mass grid and assumed to collapse to a neutron star, while the rest of the star is ejected by the SN explosion, which is modelled by a piston action near the edge of the central core. The pre-SN model has a heterogeneous chemical composition containing a 5.6 $M_{\sun}$ helium core and a 11.9 $M_{\sun}$ outer shell of the solar chemical composition (Fig. \[fig:chcom\]). Note that the helium cores up to about 8 $M_{\sun}$ are consistent with the observations (Sect. \[sec:grg-chcom\]). There is no sharp boundary between the hydrogen-rich and helium-rich layers in the ejecta of the optimal model. Hydrogen-rich material is mixed into the central region, and helium-rich material, in turn, is mixed outwards. It is evident that such a distribution of hydrogen and helium implies a strong mixing at the helium/hydrogen composition interface. The fact that the radioactive $^{56}$Ni is confined to the innermost ejected layers suggests its weak mixing during the SN explosion.
In Fig. \[fig:lmbol\] we show the very good match between the bolometric light curve calculated for the optimal hydrodynamic model and the one observed for SN 1999em (Elmhamdi et al. [@edc03]). Note that hereafter $t_{obs}$ is the time in the observer’s frame of reference. The model agrees well with the observed tail of the bolometric light curve for the total $^{56}$Ni mass of 0.036 $M_{\sun}$, the bulk of the radioactive $^{56}$Ni being mixed in the velocity range $\le 660$ kms$^{-1}$ (Fig. \[fig:denicl\]) in order to reproduce the observed transition from the plateau to the tail.
H[$\alpha$]{} profile: evidence against short distance {#sec:evidis}
======================================================
It is quite clear that any well-observed SN should be described by a unique hydrodynamic model in combination with the atmosphere model based on it. The experience during the study of SN 1987A showed that such a combination of the models had to fit not only the photometric, but also the spectroscopic observations. We believe that the adequate hydrodynamic and atmosphere models of SN 1999em are able to distinguish between the short and long distances. A distance of 7.85 Mpc, the average value of the EPM distance estimates, is taken as the short distance, and the Cepheid distance of 11.7 Mpc is taken as the long distance.
[c c c @[ ]{} c @[ ]{} c c c]{} Model & $R_0$ & $M_{env}$ & $E$ & $M_{\mathrm{Ni}}$ & $X$ & $Z$\
& $(R_{\sun})$ & $(M_{\sun})$ & ($10^{51}$ erg) & $(10^{-2} M_{\sun})$ & &\
D07 & 375 & 16 & 0.686 & 1.62 & 0.735 & 0.017\
D11 & 500 & 19 & 1.300 & 3.60 & 0.735 & 0.017\
As demonstrated above, we constructed the optimal hydrodynamic model for the long distance (model D11 in Table \[tab:hydmods\]). The model reproduces the observed bolometric light curve of SN 1999em very well (Fig. \[fig:hmd117\]a). In Fig. \[fig:hmd117\]b the calculated expansion velocity at the photosphere level, the photospheric velocity, is compared with the radial velocities at maximum absorption of the different spectral lines measured by Hamuy et al. ([@hpm01]) and Leonard et al. ([@lfg02]). The photospheric velocity of model D11 is consistent with the observed points, at least for the first 60 days. To verify the hydrodynamic model by matching the constraints from the spectral observations of SN 1999em, we examined the H$\alpha$ profile on days 26.24 and 52.14 computed in the time-dependent approach.
In the time-dependent atmosphere model for SN 1987A, we considered two extreme cases to allow for the uncertainty of our approximation in a description of the ultraviolet radiation field: (i) the photospheric brightness is black-body with the effective temperature (model A); (ii) the photospheric brightness corresponds to the observed spectrum (model B) (Utrobin & Chugai [@uc05]). For SN 1999em the photospheric brightness in model B is black-body with the effective temperature and the corresponding brightness reduction taken from the SN 1987A observations.
The time-dependent approach with model B satisfactorily reproduces the strength of the H$\alpha$ emission component on day 26.24 with some emission deficit near the maximum in comparison to what was observed (Fig. \[fig:hmd117\]c). This deficit is not significant because model A, calculated for this phase and not plotted on Fig. \[fig:hmd117\]c for the sake of clarity, gives the emission component with a relative flux of 3.1 at the maximum, which is much higher than the observed one, so the real situation is somewhere between these two cases and is closer to model B. Note that in SN 1987A the H$\alpha$ emission component demonstrated the same behavior in the early phase (Utrobin & Chugai [@uc05]). In contrast, the H$\alpha$ absorption component calculated in model B is stronger than that observed on day 26.24 in SN 1999em. This discrepancy is presumably related to a poor description of the ultraviolet radiation at frequencies between the Balmer and Lyman edges. This radiation interacts with numerous metal lines and controls the populations of hydrogen levels. It is very important that both the emission and absorption components of the H$\alpha$ line calculated in model B extend over the whole range of the observed radial velocities from $-15\,000$ kms$^{-1}$ to $15\,000$ kms$^{-1}$ (Fig. \[fig:hmd117\]c). Unfortunately, the calculated absorption component runs above the observed one in a radial velocity range between $-15\,000$ kms$^{-1}$ and $-12\,500$ kms$^{-1}$. On day 52.14 the emission component of the H$\alpha$ line computed with the time-dependent approach with model B fits the observed one fairly well, while the absorption component is still stronger than the observed one (Fig. \[fig:hmd117\]d). Thus, we may state that the above hydrodynamic and atmosphere models are in a good agreement with the photometric and spectroscopic observations of SN 1999em.
Now let us pay attention to the time-dependent effects in hydrogen lines of SN 1999em, a normal SN IIP. The time-dependent approach with model B reproduces the strength of the H$\alpha$ line at day 26.24 and day 52.14 fairly well (Figs. \[fig:hmd117\]c and \[fig:hmd117\]d). In contrast, a steady-state model B demonstrates an extremely weak H$\alpha$ line on days 26.24 and 52.14. This reflects the fact that the steady-state ionization is significantly lower than in the time-dependent model, while the electron temperature is too low for the collisional excitation of hydrogen. The radioactive $^{56}$Ni is mixed too weakly to affect the ionization and excitation of hydrogen and other elements in the atmosphere at the plateau phase. Thus, it is possible to conclude that the hydrogen recombination in the atmosphere of SN 1999em during the whole plateau phase is essentially a time-dependent phenomenon.
In turn, the short distance results in the hydrodynamic model with the following parameters: the pre-SN radius of 375 $R_{\sun}$, the ejecta mass of 16 $M_{\sun}$, and the explosion energy of $6.86\times10^{50}$ erg (model D07 in Table \[tab:hydmods\]). In this model the central core of 1.4 $M_{\sun}$ is assumed to collapse to a neutron star. The hydrodynamic model fits the observed bolometric light curve of SN 1999em but for the $^{56}$Ni mass of 0.0162 $M_{\sun}$, the most of $^{56}$Ni being mixed in the velocity range $\le 580$ kms$^{-1}$ (Fig. \[fig:hmd785\]a). The photospheric velocity curve runs well below the observed points (Fig. \[fig:hmd785\]b), and it might be expected that the spectral lines in this model would be narrower than those observed. Indeed, on days 26.24 and 52.14 both the emission and absorption components of the H$\alpha$ line computed with the time-dependent approach with model B are significantly narrower than those observed in SN 1999em (Figs. \[fig:hmd785\]c and \[fig:hmd785\]d). The hydrodynamic model D07 agrees fairly well with the observed bolometric light curve, but the relevant atmosphere model fails to reproduce the H$\alpha$ profile observed in SN 1999em. We thus conclude that the short distance of 7.85 Mpc should be discarded.
Development of the optimal model {#sec:devmod}
================================
Although the major issues of light curve theory were recognized a long time ago (Grassberg et al. [@gin71]; Falk & Arnett [@fa77]), it is useful to consider an SN outburst in a more detailed approach. This study of SN 1999em provides a good opportunity to examine the time development of a normal SN IIP. Figures \[fig:snphs\] and \[fig:swbrck\]a show the following stages in the observer time scale: a shock breakout ($t_{obs} \le t_1$), an adiabatic cooling phase ($t_1 < t_{obs} \le t_2$), a phase of cooling and recombination wave (CRW) ($t_2 < t_{obs} \le t_3$), a phase of radiative diffusion cooling ($t_3 < t_{obs} \le t_4$), an exhaustion of radiation energy ($t_4 < t_{obs} \le t_5$), a plateau tail ($t_5 < t_{obs} \le t_6$), and a radioactive tail ($t_{obs} > t_6$). In addition to the above stages there are two specific points: a transition from acceleration of the envelope matter to a homologous expansion and a moment of complete hydrogen recombination. In the optimal model D11, the characteristic moments are $t_1 \approx 0.93$ days, $t_2 \approx 18$ days, $t_3 \approx 94$ days, $t_4 \approx 116$ days, $t_5 \approx 124$ days, and $t_6 \approx 150$ days, and the complete hydrogen recombination occurs at $t_\mathrm{H} \approx 111.3$ days. Below we consider the basic stages in the time development of the optimal model.
Shock breakout {#sec:dmd-shbrk}
--------------
The explosion of the star is assumed to be triggered by a piston action near the edge of the central core immediately after the epoch zero, $t=0$. From here on, $t$ is the time in the comoving frame of reference. This energy release generates a strong shock wave that propagates towards the stellar surface. In moving out of the center, the shock wave heats matter and accelerates it to velocities increasing outward and exceeding the local escape velocity. From $t=0.0065$ days to $t=0.2416$ days, the shock wave, propagating outward from the compact dense core of the pre-SN (Fig. \[fig:rhomr\]), is attenuated slightly due to the spherical divergence (Fig. \[fig:velvsm\]). It then reaches the outermost layers with a sharp decline in density and after $t=0.6234$ days gains strength and accelerates due to the effect of hydrodynamic cumulation. Only a small portion of the star undergoes this acceleration and acquires a high velocity. For example, the layers of velocities exceeding 5000 km s$^{-1}$ have a mass of $\approx 0.478 M_{\sun}$.
By day 0.8651 the shock wave reaches the stellar surface and then begins to heat the external layers, so that the color temperature jumps to $3.84 \times 10^{5}$ K at day 0.8823, the effective temperature increases up to $1.76 \times 10^{5}$ K at day 0.8943, and the bolometric luminosity rises accordingly up to $6.46 \times 10^{44}$ ergs$^{-1}$ at day 0.8943 (Fig. \[fig:swbrck\]). Note that the maximum of the color temperature coincides closely with the moment of 0.8803 days when a velocity at the stellar surface reaches the escape velocity.
A very rapid rise in the bolometric luminosity to maximum, starting at day 0.8651, instantly changes a growth of the photospheric radius into its reduction because of the intense radiative losses of energy in the outermost layers (Fig. \[fig:swbrck\]b). At day 0.8691 these layers begin to move outward, and the additional cooling by adiabatic expansion makes the reduction of the photospheric radius more noticeable. At the same time the envelope expansion is favorable to a photon diffusion, decreasing the characteristic diffusion time. The photon diffusion eventually dominates, stops this reduction at day 0.8802, and then, along with the envelope expansion, blows the photospheric radius outward.
When a velocity at the stellar surface exceeds the escape velocity, the outside layers of the star begin to cool rapidly, the color temperature begins to decrease from its maximum value, and both the effective temperature and the luminosity decrease somewhat later. A narrow peak in the bolometric luminosity forms as a result (Fig. \[fig:swbrck\]a). The peak has a width of about $0.02$ days at a half level of the luminosity maximum. Most of its radiation is emitted in an ultraviolet flash. The total number of ionizing photons above 13.598 eV for the whole outburst is $2.768 \times 10^{58}$. During the first 1.234 days the number of ionizing photons is $90\%$ of the total number, and the radiated energy adds up to $1.59 \times 10^{48}$ ergs. We conditionally define the transition time between the shock breakout and the adiabatic cooling phase, $t_1$, as the time at which the bolometric luminosity drops by one dex from its maximum value, with the understanding that the adiabatic cooling becomes essential soon after the onset of the envelope expansion. In the optimal model this transition time is nearly 0.93 days.
Scattering processes are fundamentally nonlocal and result in exceeding the color temperature over the effective one when they are dominant (Mihalas [@m78]; Sobolev [@s80]). During the shock breakout, the adiabatic cooling phase and the CRW phase scattering processes dominate those of true absorption near the photospheric level in the ejecta of the optimal model and, as a consequence, lead to a color temperature that is considerably higher than the effective temperature. Figure \[fig:swbrck\]b shows that the maximum of the color temperature coincides very closely in time with the local minimum of the photospheric radius at day 0.8802 and comes before the maximum of the effective temperature and the bolometric luminosity.
Adiabatic cooling phase {#sec:dmd-adbcln}
-----------------------
During the shock passage throughout the star, the gas is heated up to about $10^{5}$ K and, as a consequence, is totally ionized. Both the internal gas energy and the radiation energy increase in the SN envelope. After the shock breakout, a dominant process in the subphotospheric, optically thick layers of the expelled envelope is the cooling by adiabatic expansion. The adiabatic losses drastically reduce the stored energy and completely determine the evolution of the bolometric luminosity during the adiabatic cooling phase (Figs. \[fig:snphs\] and \[fig:swbrck\]a). Such behavior of the luminosity lasts till the gas and radiation temperatures drop to the critical value in the subphotospheric layers, and a recombination of hydrogen, the most abundant element in the ejected envelope, starts. In these layers hydrogen becomes partially ionized by around day 18. For the adiabatic cooling phase, the color and effective temperatures drop rapidly from $1.35 \times 10^{5}$ K and $9.32 \times 10^{4}$ K, respectively, at day 0.93 to 6650 K and 6560 K at day 18. Their ratio also reduces as the contribution of scattering processes to the opacity decreases.
In an optically thick medium, the strong shock wave propagates almost adiabatically throughout the stellar matter. When the shock wave emerges on the pre-SN surface, the adiabatic regime is broken and transforms into the isothermal regime. This transformation takes place during the adiabatic cooling phase and gives rise to a thin, dense shell at the outer edge of the ejected envelope. The dense shell is arising from day 0.945 to day 1.022, reaching the density contrast of $\sim$ 210 at a velocity of $\sim$ 12300 kmc$^{-1}$. The formation of the dense shell starts in the optically thick medium at the optical depth of $\sim$ 8, and ends in semi-transparent medium at the optical depth of $\sim$ 1. By day 18 the shell accelerates to a velocity of $\sim$ 13400 kmc$^{-1}$ reducing the density contrast to a value of $\sim$ 2.5. The matter in the dense shell is subject to the Rayleigh-Taylor instability, which may result in the strong mixing of matter (Falk & Arnett [@fa77]). The latter, in turn, may prevent the thin shell-like structure from forming in the outer layers of the envelope.
Homologous expansion {#sec:dmd-free}
--------------------
The SN explosion causes an acceleration of the envelope matter. At late times, when the acceleration of the ejecta becomes negligible, the envelope matter expands homologously. Evidently, there is a transition from the acceleration to the homologous expansion. This transition for different mass shells occurs at different times: the deeper the layer, the later the transition (Fig. \[fig:freexp\]). Thus, the transition time of the whole expelled envelope is determined by the deepest layer. If we consider physical processes in an SN atmosphere, the relevant transition time is given by the photospheric location. It is clear that the transition time for the SN atmosphere is much shorter than for the whole envelope. For instance, the atmosphere layers exceed $99\%$ and $99.9\%$ of their terminal velocities starting from nearly day 2.8 and day 11, respectively (Fig. \[fig:freexp\]).
It is interesting that a monotonic increase in the transition time with embedding inward the ejecta is broken at the $99.9\%$ level of the terminal velocity. There is a prominent feature at a velocity of $\sim$ 14500 kmc$^{-1}$ marked in Fig. \[fig:freexp\]. As we show later in Sect. \[sec:grg-lop\], it is the result of an additional acceleration due to the resonance scattering in numerous metal lines.
Homologous expansion is characterized by a density distribution frozen in velocity space and scaled in time as $\propto t^{-3}$. Such a density profile for the optimal model is shown at $t=50$ days in Fig. \[fig:denicl\]. In the inner region of the ejected envelope there is a dense shell with a density contrast of $\sim$ 20 produced by the $^{56}$Ni bubble at a velocity of $\sim$ 660 kmc$^{-1}$. The density distribution above the $^{56}$Ni bubble shell is nearly uniform up to $\sim$ 2000 kmc$^{-1}$. The outer layers with velocities in the ranges 3000–6900 km s$^{-1}$ and 6900–13000 km s$^{-1}$ may be fitted by an effective index $n=-\partial \ln \rho / \partial \ln r$ of 7 and 9.5, respectively, as seen from the density distribution with respect to the expansion velocity (Fig. \[fig:denicl\]). At a velocity of $\sim$ 13000 kmc$^{-1}$, there is another dense shell with a density contrast of $\sim$ 3 originated during the adiabatic cooling phase. Note that both dense shells slightly change their density profiles with time. For example, the evolution of the $^{56}$Ni bubble shell is clearly demonstrated by Fig. \[fig:crwdnt\]a.
Cooling and recombination wave {#sec:dmd-crw}
------------------------------
As the envelope expands, cooling by radiation – a cooling and recombination wave – occurs and completely dominates the luminosity of the SN by about day 18. From this time to nearly day 94 the bolometric luminosity is mainly determined by properties of the CRW (for details see Grassberg & Nadyozhin [@gn76]; Imshennik & Nadyozhin [@in89]). A main property of the CRW consists in generating virtually the entire energy flux carried away by radiation within its front. This radiation flux exhausts the thermal and recombination energy of cooling matter by means of recombination. At the inner edge of the CRW, the radiation flux is negligibly small and matter is ionized, but at the outer edge the flux is equal to the luminosity of the star, and matter completely recombines.
During the CRW phase the evolution of the density in the ejecta illustrates the homologous expansion (Fig. \[fig:crwdnt\]a). The behavior of the gas and radiation temperatures reflects two important facts (Fig. \[fig:crwdnt\]b). First, in the subphotospheric layers in an optically thick medium, under conditions very close to LTE, the radiation temperature is virtually equal to the gas temperature. Second, in the transparent layers, beyond the photosphere, the gas temperature turns out to be lower than the radiation temperature because of the weak interaction between matter and radiation and of the dominant role of adiabatic losses. From an inspection of the interior luminosity and the fraction of ionized hydrogen from $t=20$ days to $t=100$ days, the CRW characteristics described above are evident (Figs. \[fig:crwlxh\]a and \[fig:crwlxh\]b). Hence in this period the CRW occurs.
From $t=20$ days to $t=100$ days, the CRW propagates through a mass of $\approx 15.96 M_{\sun}$, while the outermost layers, those responsible for the luminosity during the first 18 days, contribute only about $0.13 M_{\sun}$. It is interesting that for this period the gas and radiation temperatures at the photosphere position are nearly constant in time and equal to about 5600 K (Fig. \[fig:crwdnt\]b). And from $t=40$ days to $t=100$ days, the photosphere is located at the outer edge of the hydrogen recombination front (Fig.\[fig:crwlxh\]b).
After $t=100$ days the radiation flux at the inner edge of the CRW becomes comparable to the luminosity of star (Fig. \[fig:crwlxh\]a). This effect is caused by radiative diffusion from the central envelope layers where the energy of radioactive decays is being released. Later on the luminosity is entirely determined by radiative diffusion and not by the CRW.
Radiative diffusion cooling {#sec:dmd-difcln}
---------------------------
A cooling by radiative diffusion starts in the optically thick expelled envelope at day 94 and ends in the semi-transparent medium at day 116 (Fig. \[fig:snphs\]). Shortly before the end of the phase of radiative diffusion cooling, the complete hydrogen recombination occurs at $t_\mathrm{H} \approx 111.3$ days. In the optimal model the radiative diffusion takes place in the inner layers with the nearly uniform density distribution and results in the characteristic shoulder of the bolometric luminosity. It should be noted that this behavior of the light curve was discussed by Falk & Arnett ([@fa77]). To describe photon diffusion, we may use a simple one-zone approximation (Arnett [@arn79]), ignoring the contribution of radioactive decays to the bolometric light curve: $${d \ln L_{bol} \over d t} = - {1 \over t_{dif}} \; , \,
t_{dif} = {\pi \over 3}{R_d \over c} \tau_d \; ,
\label{eq:lboldif}$$ where $t_{dif}$ is the diffusion time, $R_d$ the characteristic radius of the diffusion region, and $\tau_d$ the optical depth of this region.
Evidently, it is possible to estimate the diffusion time from both the decline of the bolometric light curve and the physical properties of the diffusion region. The decline of the luminosity shoulder between day 94 and day 116 measures a diffusion time of $\approx 35$ days (Fig. \[fig:snphs\]). On the other hand, the radius of the diffusion region of $\approx 1.2 \times 10^{15}$ cm and its optical depth of $\approx 63$ evaluated at day 105 give a diffusion time of $\approx 30$ days. These estimates are consistent with the above description of photon diffusion (\[eq:lboldif\]). Thus, the luminosity shoulder at the phase of radiative diffusion cooling is approximated well by the simple model with the constant diffusion time. Note that a transition from the constant diffusion time to the variable diffusion time, which is getting shorter, transforms the linear shoulder of the luminosity in logarithmic scale into a convex light curve at the end of the plateau.
Exhaustion of radiation energy {#sec:dmd-exrden}
------------------------------
During the entire outburst, the total radiation energy in the ejecta is much greater than its total internal gas energy. In the inner optically thick layers of the envelope, in which radiation is in equilibrium with matter, the radiative diffusion reduces both the radiation and internal gas energies. By day $116$ the total optical depth of the ejecta falls to a value of $\sim 4$, and a low opacity of matter significantly weakens the interaction between radiation and matter. As a consequence, for the next 9 days the photon diffusion mainly reduces the stored radiation energy and eventually exhausts it, and the luminosity decreases abruptly and becomes close to the gamma-ray luminosity due to the decay of the radioactive $^{56}$Co, which is calculated by taking account of the retardation effect in the envelope transparent for gamma-rays (Fig. \[fig:scndpl\]a).
Plateau tail phase {#sec:dmd-secnd}
------------------
By the end of the previous phase, the bolometric luminosity does not fall directly to the gamma-ray luminosity (Fig. \[fig:scndpl\]a), and the total radiation energy is not exhausted completely. Nearly at the same time the photosphere disappears, and the SN envelope becomes optically thin for radiation. For example, the total optical depth of the ejecta is 2.96 at the moment $t_5$, while the optical depth above the $^{56}$Ni bubble shell is only 0.78. Because during the phase of the radiation energy exhaustion the photon gas in the outer layers cools more intensively than in the inner part of the envelope ($t=126$ and 130 days in Fig. \[fig:scndpl\]b), the radiation flow is generated in the warm inner layers ($t=130$ and 134 days in Fig. \[fig:scndpl\]b), then it propagates throughout the transparent layers, and results in some luminosity excess in the light curve just after the steep decline in the luminosity compared to the radioactive tail (Fig. \[fig:scndpl\]a). We call this behavior of the light curve after the main plateau at the beginning of the radioactive tail “a plateau tail”.
The radiation flow generated near the center of the envelope exists from $t \sim 130$ days to $t \sim 150$ days and then disappears (Fig. \[fig:scndpl\]b). The characteristic duration of the plateau tail in the optimal model is 26 days (Fig. \[fig:scndpl\]a). Note that the energy excess radiated during the plateau tail over the luminosity of radioactive decays is $2.5 \times 10^{46}$ erg and $11\%$ of the total radiation energy within the ejecta at the moment $t_5$.
Radioactive tail of the light curve {#sec:dmd-rdtail}
-----------------------------------
After the phase of the plateau tail, the bolometric luminosity of the optimal model decreases to the gamma-ray luminosity due to the decay of the radioactive $^{56}$Co (Fig. \[fig:scndpl\]a) and is solely powered by the radioactive energy source. This is how the stage of the radioactive tail starts, and the calculated light curve is in good agreement with the observations of SN 1999em after day 150 (Fig. \[fig:lmbol\]). Note that the envelope remains optically thick for the gamma rays during a few hundred days (the total optical depth is 30.6 at $t=150$ days), and they deposit the bulk of their energy locally, at least at the beginning of the radioactive tail.
General properties of the optimal model {#sec:genpro}
=======================================
[c l]{} Model & Remarks\
Dcc & no dense central core (Fig. \[fig:denstr\]a)\
Ldn & less dense outer layers (Fig. \[fig:denstr\]c)\
Mdn & more dense outer layers (Fig. \[fig:denstr\]c)\
Hol & helium-rich composition in the outer layers:\
& $X=0.368$, $Y=0.615$, and $Z=0.017$ (Fig. \[fig:sfchcm\]a)\
Zol & low metallicity in the outer layers:\
& $X=0.735$, $Y=0.259$, and $Z=0.006$ (Fig. \[fig:sfchcm\]c)\
Sci & sharp metals/He/H composition interface (Fig. \[fig:crchcm\]a)\
Hec & helium core of about 8 $M_{\sun}$ (Fig. \[fig:crchcm\]c)\
Nhf & a half of $^{56}$Ni amount: $M_{\mathrm{Ni}}=0.018 M_{\sun}$\
Nno & no radioactive $^{56}$Ni\
Nvm & $^{56}$Ni mixed up to a velocity of 830 kms$^{-1}$\
Nvh & $^{56}$Ni mixed up to a velocity of 1095 kms$^{-1}$\
Lop & no contribution of the expansion opacity\
Ldk & without the limb-darkening effect\
It is well known that the photometric characteristics of the SN IIP outburst are determined mainly by the basic parameters: the pre-SN radius $R_0$, the ejecta mass $M_{env}$, and the explosion energy $E$. Moreover, to get better agreement with the observations of SN 1999em, we have to vary the density distribution in the pre-SN model, its chemical composition, the $^{56}$Ni mass, and its mixing. A dependence of the bolometric light curve on the expansion opacity and the effect of limb darkening is of particular interest. These general properties of the optimal model are studied by means of the models listed in Table \[tab:modsD11\].
Presupernova structure {#sec:grg-presn}
----------------------
At first, we study the dependence of the bolometric luminosity on the inner layers’ density in the pre-SN model and consider the extreme case of an auxiliary model similar to the optimal model but without a dense central core (Fig. \[fig:denstr\]a). In such a model the inner layers do less work moving outwards in the gravitational field than do those in the optimal model with a dense central core and hence acquire higher velocities than in the optimal model. In contrast, the outer layers of the optimal model expand faster than those of the auxiliary model. As a consequence, the photospheric radius of model D11 is greater than in the auxiliary model up to about day 35 and is smaller later on. At the comparable effective temperatures of the models, this results in completely different behavior by the bolometric light curves (Fig. \[fig:denstr\]b). This implies that there is a dense central core inside the real pre-SN of SN 1999em.
In terms of the CRW properties the growth in the bolometric luminosity during the CRW phase in the auxiliary model (Fig. \[fig:denstr\]b) is controlled by effective index $n$ at the photosphere position that is greater by $\sim 3.5-0.5$ than in the optimal model. A greater effective index corresponds to a greater increase in the total flux of mass through the surface of the CRW front with time and, as a consequence, to faster growth in the bolometric luminosity with time (Grassberg & Nadyozhin [@gn76]).
The influence of the density distribution in the outer layers of the pre-SN on the bolometric luminosity is illustrated by model D11 and the same models but with both less dense and denser outer layers (Figs. \[fig:denstr\]c and \[fig:denstr\]d). A roughly similar behavior of the effective index $n$ at the photosphere position in these models reflects the approximately equal slopes of the corresponding bolometric light curves during the CRW phase. A transition from the model with the less dense outer layers to that of the denser layers is characterized by a reduction in the cooling by the adiabatic expansion and, consequently, by an increase in the bolometric luminosity in the whole SN outburst except for phases of the shock breakout and the radioactive tail. As a result, for these models the energies radiated during the first 180 days are $1.251 \times 10^{49}$ erg, $1.355 \times 10^{49}$ erg, and $1.430 \times 10^{49}$ erg, respectively. Thus, the bolometric light curve is fairly sensitive to the initial structure of the outer layers.
Chemical composition {#sec:grg-chcom}
--------------------
In addition to the pre-SN structure, the chemical composition of the ejecta also affects the SN luminosity. Let us consider a dependence on the chemical composition in the outer layers of the ejecta beyond the helium core. An enhancement of helium abundance at the expense of hydrogen abundance in model Hol (Fig. \[fig:sfchcm\]a and Table \[tab:modsD11\]) results in the higher luminosity during the adiabatic cooling phase than that of model D11, nearly the same luminosity in the CRW phase, a small bump at the end of the plateau and then the shortened duration of the plateau (Fig. \[fig:sfchcm\]b). This behavior of the luminosity is due to the lower opacity in the outer layers of model Hol. On the other hand, lower metallicity for model Zol (Fig. \[fig:sfchcm\]c and Table \[tab:modsD11\]) than for model D11 (Table \[tab:hydmods\]) increases the bolometric luminosity slightly during the CRW phase and then reduces it somewhat at the phase of radiative diffusion cooling (Fig. \[fig:sfchcm\]d). This is explained by a smaller contribution of heavy elements to opacity in the outer layers of the ejecta in model Zol. Thus, the bolometric light curve depends weakly on the chemical composition in the outer layers of the ejecta beyond the helium core, while hydrogen is abundant there and controls opacity of matter.
A good fit of the calculated bolometric light curve for the optimal model to what is observed (Fig. \[fig:lmbol\]) is obtained with mutual mixing of the hydrogen-rich and helium-rich matter in the inner layers of the ejecta (Fig. \[fig:chcom\]). It is appropriate to ask here in what way the sharp boundary between the hydrogen-rich and helium-rich layers at the edge of the helium core — a characteristic of the evolutionary models of pre-SNe — changes the light curve. In the auxiliary model Sci without the mutual mixing at the edge of the helium core (Table \[tab:modsD11\]) this boundary is located at the interior mass of 5.4 $M_{\sun}$ (Fig. \[fig:crchcm\]a). It is known that a transition from the hydrogen-rich to helium-rich layers, i.e. from the low to high ionization potential matter, causes an increase in the effective temperature and, as a consequence, a growth in the bolometric luminosity (Grassberg & Nadyozhin [@gn76]; Utrobin [@utr89]). In fact, when the CRW front reaches the sharp boundary between the hydrogen-rich and helium-rich layers, crosses it, and then enters the helium core, a distinct bump in the light curve at the end of the plateau appears (Fig. \[fig:crchcm\]b). Evidently, this local maximum of the luminosity is inconsistent with the observations of SN 1999em and other SNe IIP.
It is well known that at the time of an SN event the luminosity of a massive progenitor is determined mainly by the mass of the helium core, since a contribution of the hydrogen burning shell to the total energy generation is negligible. Hence the mass of the helium core is a critical quantity in evaluating the evolutionary state of a massive star. In this context it is intriguing that the increase in the helium core mass from 5.6 $M_{\sun}$ in model D11 to 8.1 $M_{\sun}$ in model Hec (Fig. \[fig:crchcm\]c), under a deep hydrogen mixing downward to $\approx 700$ kms$^{-1}$, leads to some decrease in the opacity of matter in the mixing region at the helium/hydrogen composition interface and, as a consequence, to the increase in the bolometric luminosity at the plateau (Fig. \[fig:crchcm\]d). The resulting bolometric light curve is still consistent with the observations of SN 1999em. Thus, the mass of the helium core between 5.6 $M_{\sun}$ and 8.1 $M_{\sun}$ is acceptable for the optimal model.
Mass of [$^{56}$]{}Ni and its mixing {#sec:grg-nicl}
------------------------------------
After the CRW stage, when the radiative diffusion takes place, the radioactive decay of the $^{56}$Ni and $^{56}$Co nuclides begins to dominate in powering the luminosity. This fact is clearly demonstrated by model Nno without $^{56}$Ni in the envelope (Fig. \[fig:nimsmx\]a and Table \[tab:modsD11\]). A lack of $^{56}$Ni shortens the duration of the bolometric light curve to about 100 days. It is evident that at later times the bolometric light curve depends on both the total mass of $^{56}$Ni and its distribution over the ejecta. Of course, the total mass of $^{56}$Ni is measured by the radioactive tail of the observed light curve and is 0.036 $M_{\sun}$ in the optimal model for the Cepheid distance of 11.7 Mpc. A reduction in the total mass of $^{56}$Ni from 0.036 $M_{\sun}$ (model D11 in Table \[tab:hydmods\]) to 0.018 $M_{\sun}$ (model Nhf in Table \[tab:modsD11\]) results mainly in a shorter duration of the light curve (Fig. \[fig:nimsmx\]a). Thus, the duration of the bolometric light curve has a minimum value of about 100 days and is a function of the $^{56}$Ni amount.
To investigate the influence of the $^{56}$Ni mixing on the light curve, hydrodynamic models Nvm and Nvh were calculated with the $^{56}$Ni distribution different from that of model D11, but containing the same total mass of $^{56}$Ni (Table \[tab:modsD11\]). In these models the radioactive nickel is spread over a larger mass range of the ejected envelope or, equivalently, over a wider velocity range than in model D11. Hence, more favorable conditions for the radiative diffusion and an additional heating due to the radioactive decays should appear and therefore should increase the luminosity at the phase of radiative diffusion cooling that is clearly seen in Fig. \[fig:nimsmx\]b. In addition, this increase in the luminosity causes its subsequently earlier fall to the radioactive tail. As a result, with increasing the $^{56}$Ni mixing in velocity space the linear luminosity shoulder at the phase of radiative diffusion cooling transforms into the convex light curve at the end of the plateau. Thus, a linear shoulder of the luminosity in logarithmic scale at the phase of radiative diffusion cooling implies a weak $^{56}$Ni mixing. Because a discrepancy between the light curve of model D11 (which is in agreement with the observations of SN 1999em) and those of models Nvm and Nvh is rather significant, one can conclude that the bulk of the radioactive $^{56}$Ni should be confined to the innermost layers of the ejected envelope expanding with velocities less than 660 kms$^{-1}$.
Plateau tail {#sec:grg-secnd}
------------
The general properties of the plateau tail on the radioactive tail are illustrated by the light curves shown in Fig. \[fig:gpsdpl\] and the relevant parameters of the models listed in Table \[tab:bp2plt\] where $t_5$ is the time of the onset of the plateau tail as defined in Fig. \[fig:snphs\], $\Delta t^{2p}$ the characteristic duration of the plateau tail, $\Delta t_R = R_{ext}/c$ is the characteristic time of radiation flow for the external radius of the expelled envelope $R_{ext}$ in the middle of the plateau tail, $\tau_{tot}$ the total optical depth of the ejecta at the moment $t_5$, $\Delta E_r^{2p}$ the energy excess radiated during the plateau tail over the luminosity of radioactive decays, and $\Delta E_r^{2p}/E_r^{tot}$ is a ratio of this energy to the total radiation energy within the ejecta at the moment $t_5$.
[c @[ ]{} c @[ ]{} c @[ ]{} c @[ ]{} c @[ ]{} c @[ ]{} c @[ ]{} c]{} Model & $t_5$ & $\Delta t^{2p}$ & $\Delta t_R$ & $\Delta t^{2p}/\Delta t_R$ & $\tau_{tot}$ & $\Delta E_r^{2p}$ & $\Delta E_r^{2p}/E_r^{tot}$\
& (days) & (days) & (days) & & & ($10^{46}$ erg) &\
D11 & 124 & 26 & 8.76 & 2.97 & 2.96 & 2.5 & 0.110\
Lop & 110 & 25 & 7.83 & 3.19 & 1.71 & 3.8 & 0.136\
Nhf & 115 & 22 & 7.94 & 2.77 & 2.99 & 1.5 & 0.100\
Mps & 127 & 22 & 7.89 & 2.79 & 2.64 & 2.3 & 0.089\
Eps & 119 & 28 & 9.25 & 3.03 & 3.01 & 3.1 & 0.107\
It is clear that there is no correlation between the characteristic duration of the plateau tail and the total optical depth of the ejecta (Table \[tab:bp2plt\]), and the duration depends very weakly on the opacity of matter (models D11 and Lop in Table \[tab:bp2plt\]). These regularities reflect the fact that the radiation responsible for the plateau tail flows throughout an optically thin medium. On the other hand, the characteristic duration is proportional to the characteristic time of radiation flow independent of the parameters of the listed models (Table \[tab:bp2plt\]), and their ratio is 2.95 on average. It is worth noting that the energy excess of the plateau tail is roughly $10 \%$ of the total radiation energy at the moment $t_5$ for all but model Lop (Table \[tab:bp2plt\]).
The plateau tail of the bolometric light curve demonstrates the following important properties. A decrease in the $^{56}$Ni mass (models D11 and Nhf) shortens the characteristic duration of the plateau tail and reduces its energy excess (Fig. \[fig:gpsdpl\]a and Table \[tab:bp2plt\]). This influence of the $^{56}$Ni amount is related to its large role in powering the light curve at the end of the main plateau after day 100 (Fig. \[fig:nimsmx\]a). The higher the average expansion velocity of the ejecta (models Mps, D11, and Eps), the longer the duration of the plateau tail and the higher the energy excess (Fig. \[fig:gpsdpl\]b and Table \[tab:bp2plt\]). An increase in the duration of the plateau tail with the average expansion velocity results from its proportionality to the characteristic time of radiation flow. The higher average expansion velocity causes the onset of the plateau tail at an earlier time, when the total radiation energy is higher and, as a consequence, the energy excess, radiated over the luminosity of radioactive decays during the plateau tail, becomes greater.
Expansion opacity {#sec:grg-lop}
-----------------
A contribution of numerous metal lines to opacity plays a fundamental role in reproducing the observed light curve of SN 1999em, a normal SN IIP, as shown by the optimal model D11 and model Lop calculated by neglecting the expansion opacity (Fig. \[fig:lopldk\]). The decrease in opacity due to disregarding the line contribution in model Lop speeds up the radiation diffusion and causes the bolometric luminosity to increase significantly compared to that of model D11 during nearly the whole outburst and, as a consequence, the characteristic duration of the plateau to shorten by about 10 days. It is equivalent in action to an increase in the explosion energy that will be demonstrated in Sect. \[sec:phyobs\] (Fig. \[fig:baspar\]c). This fact leads to the firm conclusion that neglecting the expansion opacity underestimates the explosion energy and introduces an error of nearly $20\%$ into its value.
It is worth noting that the slope of the light curve shoulder at the phase of radiative diffusion cooling in model D11 is flatter than in model Lop without the expansion opacity (Fig. \[fig:lopldk\]). This behavior of the light curve is consistent with the dependence of the characteristic diffusion time on the optical depth of the diffusion region (\[eq:lboldif\]) which, in turn, is proportional to the opacity of matter.
The resonance scattering of radiation in numerous metal lines is also important in forming the gas flow in the outer layers of the expanding SN envelope (Fig. \[fig:lopvel\]). After the shock emerges on the pre-SN surface at $t=0.8937$ days, an expansion of the envelope begins, and it becomes homologous during the next 10 days. An additional acceleration induced by the expansion opacity occurs between $t \approx 1$ days and $t \approx 1.5$ days, lasts up to $t \approx 7.5$ days, and takes place in the outer layers with a mass of about $10^{-4} M_{\sun}$. A comparison of models D11 and Lop shows that these layers gain a velocity excess as large as 600 kms$^{-1}$, except for the outermost layers with a mass of about $10^{-8} M_{\sun}$. Note that this additional acceleration is responsible for a feature in the gas flow as seen in Fig. \[fig:freexp\] where the velocities of different mass shells in the ejected envelope as a function of time are overplotted on the photospheric velocity.
Effect of limb darkening {#sec:grg-lmbdkg}
------------------------
During the first 116 days, when a well-defined photosphere exists, particularly at the CRW phase, the emitted radiation is nearly isotropic, and the bolometric light curves of models D11 and Ldk, calculated without the effect of the limb darkening, almost coincide (Fig. \[fig:lopldk\]). As the envelope expands and becomes optically thin, the continuum formation region gradually becomes more extended, and the degree of anisotropy of the emergent radiation increases. The increase in the radiation anisotropy with time clearly results in the difference between the light curves of models D11 and Ldk during the phase of the radiation energy exhaustion: the bolometric luminosity calculated by taking the limb-darkening law into account is lower than for isotropic radiation (Fig. \[fig:lopldk\]).
Utrobin ([@utr04]) shows that the difference between the bolometric luminosity calculated by taking into account the emergent anisotropic radiation and the luminosity for the isotropic radiation depends on both the limb-darkening law and the retardation effect. Interestingly, this difference increases with growing the degree of the anisotropy of the emergent radiation, while its sign is determined solely by the time derivative of the bolometric luminosity. The latter property explains the behavior of the bolometric luminosity for models D11 and Ldk discussed above. It seems that in SN 1999em, a normal SN IIP, the effect of limb darkening is more of purely academic interest than of practical value.
Physical and observed parameters {#sec:phyobs}
================================
Generally, our main goal is to construct the adequate hydrodynamic model by fitting the photometric and spectroscopic observations of the object under study. Sometimes it is tempting to evaluate the relevant physical parameters for other SN with close observed properties. In this case, it is reasonable to use the results already obtained for the similar object. For this purpose, we have to find the relationships between the basic physical and observed parameters. Evidently, the important physical parameters are the initial radius $R_0$, the ejecta mass $M_{env}$, and the explosion energy $E$. The large impact of $^{56}$Ni on the light curve near the end of the plateau, which is shown and discussed in Sect. \[sec:grg-nicl\], makes us take the total $^{56}$Ni mass into account. Because of the temporal character of the energy deposition from radioactive decays, it is impossible to treat this energy input as an addition to the explosion energy, so we have to introduce the total $^{56}$Ni mass $M_{\mathrm{Ni}}$ as the fourth basic physical parameter.
To describe the observed properties of SN IIP, Litvinova & Nadyozhin ([@ln83], [@ln85]) used the three parameters: the plateau duration, the absolute V magnitude, and the photospheric velocity in the middle of the plateau. However, in the course of time it became evident that this simple description needed to be corrected for the influence of $^{56}$Ni on the light curve via an additional parameter (Nadyozhin [@n03]). Unfortunately, the bounds of the plateau are unclear because they are no specific points on the light curve. Moreover, the interval of a plateau duration misses the bolometric luminosity jump at the shock breakout and the subsequent adiabatic cooling phase together with the relevant dependence on the initial radius, the ejecta mass, and the explosion energy (Fig. \[fig:baspar\]). To overcome this problem, instead of the plateau duration we define a characteristic duration $\Delta t$ of the light curve as a whole, measuring it from the bolometric luminosity jump at the shock breakout to the end of the plateau for an $e$-fold reduction of the luminosity in the middle of plateau (Fig. \[fig:bscshm\]). Of course, in this way the characteristic duration is easy to determine for a theoretical light curve, as shown in a schematic bolometric light curve of SN IIP, but very difficult for an observed one. Fortunately, this situation is not hopeless. First, it is possible to apply the theoretical light curve as a template for measuring the specific moments $t_o$ and $t_e$ (Fig. \[fig:bscshm\]) and, consequently, for calculating the characteristic duration. Then, as the observational data become more and more extensive, the relevant templates may be constructed from the observed light curves of the well-studied SNe IIP. The second modification we made is to use the bolometric luminosity $L_{bol}^{p}$ in the middle of the plateau because it is more adequate than the absolute V magnitude for estimating the explosion energy. The photospheric velocity $v_{ph}^{p}$ in the middle of the plateau is the third observed parameter in accordance with Litvinova & Nadyozhin ([@ln83], [@ln85]).
[c c c @[ ]{} c @[ ]{} c c c]{} Model & $R_0$ & $M_{env}$ & $E$ & $M_{\mathrm{Ni}}$ & $X$ & $Z$\
& ($R_{\sun}$) & ($M_{\sun}$) & ($10^{51}$ erg) & $(10^{-2} M_{\sun})$ & &\
D11 & 500 & 19 & 1.30 & 3.60 & 0.735 & 0.017\
Rms & 425 & 19 & 1.30 & 3.60 & 0.735 & 0.017\
Rps & 575 & 19 & 1.30 & 3.60 & 0.735 & 0.017\
Mms & 500 & 16 & 1.30 & 3.60 & 0.735 & 0.017\
Mps & 500 & 22 & 1.30 & 3.60 & 0.735 & 0.017\
Ems & 500 & 19 & 1.10 & 3.60 & 0.735 & 0.017\
Eps & 500 & 19 & 1.50 & 3.60 & 0.735 & 0.017\
Nms & 500 & 19 & 1.30 & 3.06 & 0.735 & 0.017\
Nps & 500 & 19 & 1.30 & 4.14 & 0.735 & 0.017\
To establish the relationships among the basic physical and observed parameters accepted above, we carried out a parameter study of hydrodynamic models varying the physical parameters and measuring the observed ones. Evidently, a four-parameter approximation is a crude description of the hydrodynamic model properties. The sensitivity of the bolometric light curve to the pre-SN structure and its chemical composition makes this approximation valid only for the vicinity of the optimal model D11, not for a whole region of the basic parameters. We restricted ourselves to a local parameter study in the vicinity of the optimal model and explored a $15\%$ limited range of the basic physical parameters (Table \[tab:auxmods\]). The resulting bolometric light curves of auxiliary hydrodynamic models are plotted in Fig. \[fig:baspar\] and compared to the optimal model. The basic observed parameters of these models are evaluated according to the definition, given in Fig. \[fig:bscshm\], and listed in Table \[tab:pobpar\].
The influence of the basic physical parameters is illustrated by the bolometric light curves of the auxiliary models (Fig. \[fig:baspar\]). Increasing the initial radius causes an increase in the characteristic time of the envelope expansion and, as a consequence, an increase in the width of the narrow intense peak on the bolometric light curve produced by the heating of the outer layers at the shock breakout phase. In addition, an increase in the initial radius leads to a reduction in the cooling by the adiabatic expansion and, accordingly, to an increase in the bolometric luminosity during the whole SN outburst except for the phase of the radioactive tail (Fig. \[fig:baspar\]a). It is interesting that this behavior by the bolometric light curve with initial radius is similar to its dependence on the density structure in the outer layers (Fig. \[fig:denstr\]c), which is discussed in Sect. \[sec:grg-presn\]. The latter dependence may be considered as a dependence on some effective radius of the pre-SN. In other words, an increase in the density in the outer layers of the pre-SN mimics an increase in the initial radius. Both decreasing the ejecta mass and increasing the explosion energy enlarge the average expansion velocity of the envelope. The increase in the average velocity of the envelope makes the shock wave propagating through the pre-SN matter stronger and, as a result, it heats the matter to a higher temperature, which increases the luminosity in the narrow peak of the light curve, at the subsequent phase of adiabatic cooling and at the beginning of the CRW phase (Figs. \[fig:baspar\]b and \[fig:baspar\]c). By the end of the CRW phase, decreasing the ejecta mass reduces the bolometric luminosity and shortens the characteristic duration of the light curve (Fig. \[fig:baspar\]b). In contrast, increasing the explosion energy results in the luminosity growing and, accordingly, in shortening the characteristic duration as well (Fig. \[fig:baspar\]c). Up to the phase of the radiative diffusion cooling, the radioactive $^{56}$Ni does not affect the light curve; but during this phase and later on, the energy deposition from radioactive decays powers the luminosity (Fig. \[fig:baspar\]d). The higher the total $^{56}$Ni mass, the greater the energy deposition and the later the luminosity descending to the radioactive tail. Note that a slope of the luminosity shoulder at the phase of the radiative diffusion cooling clearly depends on the initial radius (Fig. \[fig:baspar\]a) and on the explosion energy (Fig. \[fig:baspar\]c) in a qualitative agreement with the simple description of photon diffusion (\[eq:lboldif\]).
[c c c c c c]{} Model & $\Delta t$ & $\log L_{bol}^{p}$ & $v_{ph}^{p}$ & $q_{kin}$ & $q_{rad}$\
& (days) & (ergs$^{-1}$) & (kms$^{-1}$) & &\
D11 & 117.57 & 42.0350 & 2374.4 & 0.8541 & 0.8258\
Rms & 114.88 & 41.9927 & 2335.3 & 0.8088 & 0.8429\
Rps & 118.25 & 42.0521 & 2388.3 & 0.8354 & 0.7826\
Mms & 111.70 & 42.0542 & 2481.5 & 0.7767 & 0.7881\
Mps & 121.64 & 42.0170 & 2291.9 & 0.9141 & 0.8252\
Ems & 122.50 & 41.9629 & 2134.7 & 0.8134 & 0.8040\
Eps & 111.93 & 42.0979 & 2640.5 & 0.8890 & 0.8424\
Nms & 113.98 & 42.0372 & 2452.5 & 0.8943 & 0.8303\
Nps & 119.39 & 42.0264 & 2320.3 & 0.7968 & 0.8102\
To evaluate the three basic physical parameters (the initial radius, the ejecta mass, and the explosion energy) from the four values measured from observations (the characteristic duration of the light curve, the bolometric luminosity and the photospheric velocity in the middle of the plateau, and the total $^{56}$Ni mass), we use the following approximate relations: $$\begin{aligned}
\log R_0 & = & + 3.481 \log L_{bol}^{p} - 5.937 \log v_{ph}^{p}
- 1.999 \log \Delta t \nonumber \\
& & - 0.499 \log M_{\mathrm{Ni}} - 120.174 \; ,
\label{eq:logr}\end{aligned}$$ $$\begin{aligned}
\log M_{env} & = & - 2.942 \log L_{bol}^{p} + 7.606 \log v_{ph}^{p}
+ 7.807 \log \Delta t \nonumber \\
& & - 0.042 \log M_{\mathrm{Ni}} + 83.045 \; ,
\label{eq:logm}\end{aligned}$$ $$\begin{aligned}
\log E & = & - 1.476 \log L_{bol}^{p} + 4.899 \log v_{ph}^{p}
+ 3.040 \log \Delta t \nonumber \\
& & + 0.312 \log M_{\mathrm{Ni}} + 90.765 \; ,
\label{eq:loge}\end{aligned}$$ where $R_0$, $M_{env}$, and $M_{\mathrm{Ni}}$ are in solar units, $E$ in units of erg, $L_{bol}^{p}$ in ergs$^{-1}$, $v_{ph}^{p}$ in kms$^{-1}$, and $\Delta t$ in days. The above approximate formulae (\[eq:logr\]–\[eq:loge\]) are found by means of minimizing the errors between the values calculated with these relations and the corresponding physical parameters from Table \[tab:auxmods\]. For this purpose we constructed a variational functional in a quadratic form depending on the functional values, i.e. coefficients in the approximate formulae (\[eq:logr\]–\[eq:loge\]), and then used the direct search method of Powell ([@p64]). Note that the obtained approximate formulae give the estimates of the physical parameters for the auxiliary models, the accurate values of which are listed in Table \[tab:auxmods\], with an error as great as $\approx 2.5\%$.
In addition to the observed parameters of the auxiliary models, the important energetic ratios $q_{kin} = 0.5 M_{env} {v_{ph}^{p}}^2 / E_{kin}$ and $q_{rad} = L_{bol}^{p} \Delta t / E_{rad}$, where $E_{kin}$ is the kinetic energy of the ejecta and $E_{rad}$ is the total radiation energy emitted during the first 180 days, are given in Table \[tab:pobpar\]. The mean values of the ratios $q_{kin}$ and $q_{rad}$ are 0.842 and 0.816 with the standard deviations of $4.4\%$ and $2.3\%$, respectively. A quite low value of the standard deviation in the energetic ratios shows that the kinetic energy of the envelope and the total radiation energy are approximated more or less accurately by the observed parameters: $L_{bol}^{p}$, $v_{ph}^{p}$, and $\Delta t$.
A functional dependence of the observed parameters on the physical ones is, in turn, derived directly from the approximate formulae (\[eq:logr\]–\[eq:loge\]) and can be written as $$\begin{aligned}
\log L_{bol}^{p} & = & + 0.488 \log R_0 - 0.267 \log M_{env}
+ 1.006 \log E \nonumber \\
& & - 0.082 \log M_{\mathrm{Ni}} - 10.469 \; ,
\label{eq:logl}\end{aligned}$$ $$\begin{aligned}
\log v_{ph}^{p} & = & + 0.083 \log R_0 - 0.247 \log M_{env}
+ 0.688 \log E \nonumber \\
& & - 0.184 \log M_{\mathrm{Ni}} - 31.951 \; ,
\label{eq:logv}\end{aligned}$$ $$\begin{aligned}
\log \Delta t & = & + 0.103 \log R_0 + 0.268 \log M_{env}
- 0.291 \log E \nonumber \\
& & + 0.154 \log M_{\mathrm{Ni}} + 16.548 \; .
\label{eq:logdt}\end{aligned}$$ The estimates of the observed parameters for the auxiliary models obtained with the above relations (\[eq:logl\]–\[eq:logdt\]) differ from the accurate values given in Table \[tab:pobpar\] by an error less than $\approx 2.2\%$.
Finally, it should be emphasized that all of the above formulae approximate the physical parameters and the observable properties of hydrodynamic models but only in the vicinity of the optimal model D11. To cover a wider range of the parameters, we have to construct a global approximation, which should be based, in our opinion, on a set of hydrodynamic models of *many real* SNe IIP, but not on that of arbitrary hydrodynamic models.
Comparison with SN 1987A {#sec:sn87a}
========================
[@[ ]{} c @[ ]{} c @[ ]{} c @[ ]{} c @[ ]{} c @[ ]{} c @[ ]{} c @[ ]{} c @[ ]{}]{} SN & $R_0$ & $M_{env}$ & $E$ & $M_{\mathrm{Ni}}$ & $Z$ & $v_{\mathrm{Ni}}^{max}$ & $v_{\mathrm{H}}^{min}$\
& $(R_{\sun})$ & $(M_{\sun})$ & ($10^{51}$ erg) & $(10^{-2} M_{\sun})$ & & (kms$^{-1}$) & (kms$^{-1}$)\
99em & 500 & 19 & 1.3 & 3.60 & 0.017 & 660 & 700\
87A & 35 & 18 & 1.5 & 7.65 & 0.006 & 3000 & 600\
Now we have got a confidence in reproducing the observed bolometric light curve and the H$\alpha$ line of SN 1999em, a normal SN IIP, we can compare this object with the well-studied SN 1987A, a peculiar SN IIP. The important parameters of the optimal model D11 for SN 1999em and model M18 for SN 1987A (Utrobin [@utr05]) are given in Table \[tab:hmsn99sn87\]. It is remarkable that the ejecta mass $M_{env}$, the explosion energy $E$, and the minimum velocity of the hydrogen-rich envelope $v_{\mathrm{H}}^{min}$ are comparable, while the pre-SN radius $R_0$, the mass fraction of heavy elements in the outer layers of the pre-SN $Z$, the total mass of $^{56}$Ni $M_{\mathrm{Ni}}$, and its maximum velocity $v_{\mathrm{Ni}}^{max}$ are quite different.
The mass of a star is a fundamental parameter that determines the properties of the star and the course of its evolution. The pre-SN masses of SN 1999em and SN 1987A, estimated by the mass of the neutron star and the ejecta mass, are nearly 20.6 $M_{\sun}$ and 19.6 $M_{\sun}$, respectively. The masses of helium cores in the pre-SNe of SN 1999em and SN 1987A are 5.6–8.1 $M_{\sun}$ and 6.0 $M_{\sun}$ (Woosley [@w88]), respectively. These masses are close enough to suppose that, in the final stages of stellar evolution, nearly the same iron cores form within the pre-SNe. This fact and roughly the same explosion energies of SN 1999em and SN 1987A imply a unique explosion mechanism for these core collapse SNe.
It is evident that the amount of radioactive $^{56}$Ni and its distribution in the SN envelope are a clear trace left by the explosion mechanism. Both SN 1999em and SN 1987A differ in the total mass of radioactive $^{56}$Ni and in its distribution throughout the envelope (Table \[tab:hmsn99sn87\]). Note that for these SNe the total $^{56}$Ni mass is correlated to the explosion energy. This dependence is consistent with the empirical correlation between the total $^{56}$Ni mass and the explosion energy for SNe IIP found by Nadyozhin ([@n03]) and Hamuy ([@ham03]). The optimal model for SN 1999em is characterized by a weak $^{56}$Ni mixing in velocity space up to $\approx 660$ kms$^{-1}$ and a deep hydrogen mixing downward to $\approx 700$ kms$^{-1}$ (Fig. \[fig:chcom\] and Table \[tab:hmsn99sn87\]). In turn, the hydrodynamic model for SN 1987A is characterized by a moderate $^{56}$Ni mixing up to $\sim 3000$ kms$^{-1}$ and a deep hydrogen mixing downward to $\sim 600$ kms$^{-1}$ (Utrobin [@utr05]).
The structure of the pre-SN and the chemical composition of its outer layers, which are unaffected by explosive nucleosynthesis and which are the end result of the entire evolution of the star, determine the pattern of SN outburst in many respects. The pre-SN model for SN 1999em has the structure of a red supergiant with a radius of 500 $R_{\sun}$, and the chemical composition of its outer layers is the standard solar composition. It is quite definite that the progenitor of SN 1987A was the star Sanduleak $-69^{\circ}202$, a B3 Ia blue supergiant. The pre-SN model for SN 1987A has a radius of 35 $R_{\sun}$, and the chemical composition of its outer layers is typical of the LMC chemical composition, $X=0.743$, $Y=0.251$, and $Z=0.006$ (Dufour [@dufour84]). It is worth noting that a deficit of heavy elements in the LMC matter compared to the standard solar composition favors the formation of blue supergiants (Arnett [@arn87]; Hillebrandt et al. [@hhtw87]). In the case of SN 1987A, a relative compactness of the pre-SN is a major factor in understanding the peculiar properties of this phenomenon (Grassberg et al. [@ginu87]).
[@[ ]{} c @[ ]{} c @[ ]{} c @[ ]{} c @[ ]{} c @[ ]{} c @[ ]{} c @[ ]{} c @[ ]{} c @[ ]{}]{} SN & $t_{sh}$ & $T_c^{max}$ & $T_{eff}^{max}$ & $\log L_{bol}^{max}$ & $\Delta t_L$ & $N^{tot}$ & $t^N$ & $E_{rad}^N$\
& (days) & ($10^5$K) & ($10^5$K) & (ergs$^{-1}$) & (days) & ($10^{56}$) & (days) & ($10^{46}$erg)\
99em & 0.865 & 3.84 & 1.76 & 44.81 & 0.0200 & 276.8 & 1.234 & 159\
87A & 0.049 & 13.2 & 5.68 & 44.63 & 0.0007 & 7.615 & 0.266 & 6.24\
The basic characteristics of the shock breakout and the UV flash in SN 1999em and SN 1987A are summarized in Table \[tab:sbsn99sn87\] where $t_{sh}$ is the time when the shock wave reaches the stellar surface, $T_c^{max}$ the maximum of the color temperature, $T_{eff}^{max}$ the maximum of the effective temperature, $L_{bol}^{max}$ the maximum of the bolometric luminosity, $\Delta t_L$ the width of the luminosity peak at a half level of its maximum, $N^{tot}$ the total number of ionizing photons above 13.598 eV for the whole outburst, $t^N$ the time when the number of ionizing photons measures up $90\%$ of the total number, and $E_{rad}^N$ the energy radiated by the SN during this time.
Under the comparable ejecta masses and the explosion energies, the smaller the pre-SN radius, the higher the velocity of matter in the outermost layers with a sharp decline of density where the matter accelerates due to the effect of hydrodynamic cumulation. At the same time the characteristic time of the envelope expansion becomes shorter and the adiabatic losses of energy greater. Clearly, in SN 1999em the shock wave reaches the stellar surface much later than in SN 1987A, and the color and effective temperatures jump to lower values, while the bolometric luminosity rises to a higher value and its peak is much wider (Table \[tab:sbsn99sn87\]). Note that in SN 1999em the bolometric luminosity peak coincides with the maximum of the effective temperature and occurs after the shock wave reaches the stellar surface in contrast to SN 1987A in which the maximum of the effective temperature precedes the peak, and the shock wave emerges between them on the pre-SN surface. The UV flash in SN 1999em produces a greater total number of ionizing photons by a factor of 36.3, and it lasts longer than in SN 1987A by a factor of 4.6, the radiated energy being greater by a factor of 25.5 (Table \[tab:sbsn99sn87\]).
The smaller the pre-SN radius, the greater the adiabatic losses of energy and the shorter the CRW phase — a characteristic feature of SNe IIP. A difference between the explosions of the red and blue supergiants is radical in the total radiation and gas energies as a function of time (Fig. \[fig:engsn99sn87\]) and, especially, in the light curves (Fig. \[fig:lumsn99sn87\]). To clarify the physical processes in SN 1999em and SN 1987A, we compare the optimal models of these SNe with the special hydrodynamic models containing no $^{56}$Ni in the envelope. Note that during the entire outburst the total radiation energy is much greater than the total gas energy of the envelope. A lack of $^{56}$Ni makes the energy losses due to the adiabatic expansion a dominant process controlling the energy balance of an SN. In this case, the blue supergiant structure of the pre-SN for SN 1987A results in such huge energy losses that its total radiation energy is much less than that of SN 1999em the whole time and is completely exhausted in about 50 days, while for SN 1999em this happens later by around 50 days. Accordingly, the luminosity at the CRW phase is maintained by the stored energy during only 22 days for SN 1987A and about 75 days for SN 1999em.
The subsequent run of the bolometric light curves in the optimal models of these SNe is entirely determined by the energy deposition of gamma rays from the radioactive decays and shows the key role played by $^{56}$Ni and $^{56}$Co in powering the light curve (Fig. \[fig:lumsn99sn87\]). The luminosity of SN 1999em exceeds the gamma-ray luminosity from the peak to the radioactive tail. On the contrary, the luminosity of SN 1987A drops abruptly well below the gamma-ray luminosity during the adiabatic cooling phase, grows to the maximum after the CRW phase in about 65 days exceeding the latter at day 60, and forms a wide dome on the light curve, declining to the radioactive tail and subsequently diving under it. The wide dome on the light curve of the SN 1987A is physically equivalent to the phases of the radiative diffusion cooling, the exhaustion of radiation energy, and the plateau tail in the evolution of SN 1999em in the sense that it contains the basic features of all these phases.
At the stage of the radioactive tail, there is a striking difference between SN 1999em and SN 1987A: the internal gas energy of the envelope decreases with time in SN 1999em and increases in SN 1987A (Fig. \[fig:engsn99sn87\]); the bolometric luminosity of SN 1999em coincides with the corresponding gamma-ray luminosity, while the bolometric luminosity of SN 1987A runs below it (Fig. \[fig:lumsn99sn87\]). The gas energy is controlled by the energy rate of gamma-ray deposition, the energy losses due to the adiabatic expansion, and the net rate of absorption-emission processes. At a given phase the last is dominated by the emission processes. In the envelope of SN 1999em, the adiabatic energy losses are much greater than those in the envelope of SN 1987A where the thermal gas energy is much less than that of SN 1999em. This fact explains both the decrease in the total gas energy with time in SN 1999em and its increase in SN 1987A.
The radiation energy is ruled mainly by the energy rate of gamma-ray deposition and the rate of work done by radiation pressure. In the envelope of SN 1999em the total radiation energy is much less than that in the envelope of SN 1987A, and, consequently, the rate of work done by radiation pressure in SN 1999em is much less than that in SN 1987A as well. It leads to a negligible difference between the bolometric and gamma-ray light curves in the optimal model of SN 1999em and to the bolometric light curve running below the gamma-ray light curve in the optimal model of SN 1987A. Note that the higher a rate of work done by radiation pressure in an SN envelope, the larger the excess of the gamma-ray luminosity over the bolometric one at the radioactive tail when the SN envelope still remains optically thick for gamma rays. In the optimal model of SN 1987A, the excess is nearly $7\%$ (Fig. \[fig:lumsn99sn87\]), which corresponds to an underestimation of the $^{56}$Ni amount of the same value in equating the observed bolometric luminosity to the gamma-ray one. Such an excess of the gamma-ray luminosity over the bolometric one is of great interest and should be taken into account in measuring the $^{56}$Ni amount from the observed bolometric luminosity at the radioactive tail.
It is very important that the approximation of homologous expansion may be used in the atmosphere models starting from nearly day 2.8 (Sect. \[sec:dmd-free\]) for SN 1999em, a normal SN IIP, and from day 1 (Utrobin [@utr04]) for SN 1987A, a peculiar SN IIP. Because of higher expansion velocities in the ejecta, the effects of expansion opacity and limb darkening are more prominent in SN 1987A than in SN 1999em. For both SNe IIP, the crucial role of the time-dependent approach in atmosphere models at the photospheric epoch is evident from a comparison of spectral lines computed in this approach with those in the steady-state approximation.
Discussion {#sec:disc}
==========
The aim of this paper is to explore the SN 1999em event, a normal SN IIP, by comparing the hydrodynamic models and the time-dependent atmosphere models with *both* the photometric *and* spectroscopic observations. The resulting optimal model of SN 1999em succeeds in reproducing the observed features. The bolometric light curve and the spectral evolution of the H$\alpha$ line are consistent with a radius of the pre-SN of $500 R_{\sun}$, a mass of the ejected envelope of $19 M_{\sun}$, an explosion energy of $1.3\times10^{51}$ erg, and a mass of radioactive $^{56}$Ni of $0.036 M_{\sun}$, the bulk of which is confined to layers ejected with velocities less than $\approx 660$ kms$^{-1}$.
The observational data of an SN are measured with some errors that yield the uncertainties in values inferred from them. In the case of SN 1999em, we can translate these measurement errors of the observed parameters into uncertainties of the physical parameters using the approximate formulae (\[eq:logr\]–\[eq:loge\]). To assign an uncertainty to the bolometric luminosity in the middle of the plateau, we adopt a relative error of $7\%$ for measuring the bolometric luminosity itself (Elmhamdi et al. [@edc03]) and an uncertainty of $\pm1$ Mpc for the Cepheid distance to the host galaxy (Leonard et al. [@lknt03]). The resulting relative error in the bolometric luminosity in the middle of the plateau is $26\%$, as well as the error in the mass of radioactive $^{56}$Ni. The relative errors in the photospheric velocities do not exceed $5\%$, inferred from the Fe II 4924, 5018, and 5169 Å absorption lines in the middle of the plateau (Hamuy et al. [@hpm01]; Leonard et al. [@lfg02]). The scatter in the explosion date yielded by the EPM varies within $\pm2$ days (Hamuy et al. [@hpm01]; Leonard et al. [@lfg02]; Elmhamdi et al. [@edc03]). To estimate the measurement errors in the characteristic duration of the light curve, we consequently adopted an uncertainty of $\pm4$ days. All the accepted errors in the observed values yield the following uncertainties in physical parameters: $\pm200 R_{\sun}$ in the pre-SN radius, $\pm1.2 M_{\sun}$ in the ejecta mass, $\pm0.1\times10^{51}$ erg in the explosion energy, and $\pm0.009 M_{\sun}$ in the mass of radioactive $^{56}$Ni.
Elmhamdi et al. ([@edc03]) note a clear flattening in the BVRI light curves of SN 1999em after the main plateau at the beginning of the radioactive tail and call it “the second plateau”. The “UBVRI” bolometric light curve of SN 1999em consequently reflects the second plateau. The observed second plateau coincides in time with the plateau tail of the bolometric light curve for the optimal model (Sect. \[sec:dmd-secnd\]), but runs slightly below a line of the radioactive tail in contrast to the latter. It seems that the real bolometric luminosity of SN 1999em is greater than the “UBVRI” luminosity during the second plateau. Elmhamdi et al. ([@edc03]) also report on the second plateau on the tail for SN 1991G and SN 1997D, and find that its duration is correlated with the amount of ejected $^{56}$Ni, as well as the duration of the plateau tail does. Of course, a second plateau nature that is different from that of the plateau tail cannot be excluded at the present time.
In the same study, Elmhamdi et al. explore the width and position of the He I 10830 Å line, which is sensitive to the nonthermal ionization and excitation produced by the radioactive decays, and show that in the ejecta of SN 1999em the bulk of $^{56}$Ni was distributed inside a region with a velocity less than 1100 kms$^{-1}$ in the close hemisphere. Obtained in the optimal model, a weak $^{56}$Ni mixing in velocity space up to $\approx 660$ kms$^{-1}$ agrees well with this $^{56}$Ni distribution. By interpreting the temporal evolution of the \[O I\] 6300, 6364 Å doublet profile in terms of the dust formation, they conclude that the dust occupies a sphere with a velocity of $\approx 800$ kms$^{-1}$. Dust grains are formed by heavy elements, so we should estimate the volume containing them. The characteristic velocity of the outer edge of the metal-rich region is nearly $850$ kms$^{-1}$ in the optimal model. Moreover, the $^{56}$Ni bubble shell containing heavy elements is unstable to the Rayleigh-Taylor instability and, consequently, should be broken by it (Basko [@basko94]) and mixed, thereby producing dense clumps in which dust can form. Thus, inside the sphere of a velocity of roughly $850$ kms$^{-1}$ in the optimal model of SN 1999em, there are conditions that favor the dust formation in the ejecta at late times. In addition, from the observed \[O I\] 6300, 6364 Å doublet luminosity, they infer the oxygen mass of $\sim 0.3-0.4 M_{\sun}$, which is consistent with the oxygen mass of $0.53 M_{\sun}$ estimated for the optimal model.
Baklanov et al. ([@bbp05]) carry out the hydrodynamic study of SN 1999em in multi-group approximation, compare the calculated hydrodynamic model to both the observed UBVRI light curves and the photospheric velocity, and advocate the hydrodynamic model for the distance of 7.5 Mpc with the pre-SN radius of 450 $R_{\sun}$, the ejecta mass of 15 $M_{\sun}$, and the explosion energy of $0.7\times10^{51}$ erg. However, the low metallicity of $Z=0.004$ for a red giant pre-SN and the uniform chemical composition with $X=0.7$ and $Z=0.004$ from the surface to its center are questionable. It is worth noting that their model for the distance of 12 Mpc is not far from our optimal model in the basic parameters: the pre-SN radius of 1000 $R_{\sun}$, the ejecta mass of 18 $M_{\sun}$, and the explosion energy of $10^{51}$ erg.
The ejected envelope of $19 M_{\sun}$ and a neutron star of $1.58 M_{\sun}$ in the optimal model of SN 1999em correspond to a $20.58 M_{\sun}$ pre-SN star. Assuming an extremely low mass loss, we may now conclude that the lower mass limit of the progenitor of SN 1999em on the main sequence is nearly $21 M_{\sun}$. Heger et al. ([@hlw00]) report the results of stellar evolution calculations of non-rotating main-sequence stars in the mass range of $10-25 M_{\sun}$. It turns out that a main-sequence star of $25 M_{\sun}$ evolves to the pre-SN star of $18.72 M_{\sun}$ with the final helium core mass of $7.86 M_{\sun}$. A linear extrapolation for the pre-SN mass of $20.58 M_{\sun}$ yields the stellar mass of $28.9 M_{\sun}$ on the main sequence and the final helium core mass of $9.6 M_{\sun}$. Thus, the pre-SN of SN 1999em most likely developed from a main-sequence star in the mass range of $21-29 M_{\sun}$.
Generally speaking, there is an opportunity to put constraints on the masses and the evolutionary states of the progenitors of core collapse SNe from direct observations. But it is a difficult task to identify the progenitors of SNe, and it is impossible to predict the time of explosion of a massive star precisely. For these reasons Smartt ([@s02]) suggests estimating the initial masses of the progenitors by directly identifying SNe progenitors in the archive images of outburst sites taken prior to the explosions, by estimating the bolometric luminosity limits of progenitors as a function of stellar effective temperature, and by comparing them with stellar evolutionary tracks on the Hertzsprung-Russell diagram. Applying this method to SN 1999em, Leonard et al. ([@lknt03]) have derived the upper mass limit for the progenitor of $20\pm5 M_{\sun}$ for the Cepheid distance to NGC 1637, and Smartt et al. ([@smg03]) obtained the upper limit of $15 M_{\sun}$ for the distance of 11 Mpc. Evidently, our estimate of the progenitor mass is in good agreement with the first upper mass limit.
It should be emphasized that the estimates of the bolometric luminosity for progenitors are hardly affected by the circumstellar dust, and a comparison with stellar evolutionary tracks is far from being unambiguous as the well-known case of SN 1987A demonstrated. There is growing observational evidence of the crucial role played by circumstellar dust in evaluating the bolometric luminosity of a progenitor star. For example, Barlow et al. ([@bsf05]) show that a large fraction of the 5 mag of extinction within the host galaxy NGC 6946 toward SN 2002hh might be due to circumstellar gas dust, which might condense within a stellar wind from an earlier M supergiant or luminous blue variable phase of the evolution of the progenitor star. And Massey et al. ([@mpl05]) find that a significant fraction of red supergiants in Galactic OB associations and clusters show up to several magnitudes of excess visual extinction compared to OB stars in the same regions and argue that this is very likely due to circumstellar dust around the red supergiants. Note that the existence of circumstellar dust before an SN outburst does not contradict the lack of influence from dust on the photometric properties of an SN. Dust grains are most likely evaporated by the intense radiation flash during the shock breakout inside a sphere of radius of $\sim 0.1$ pc (Dwek [@dwek83]).
A good fit of the calculated bolometric light curve for the optimal model to that observed for SN 1999em is achieved with mutual mixing of the hydrogen-rich and helium-rich matter at the helium/hydrogen composition interface. The sharp boundary between the hydrogen-rich and helium-rich layers (Fig. \[fig:crchcm\]a) at the edge of the helium core — a characteristic of the evolutionary models of pre-SNe — changes the light curve producing a distinct bump at the end of the plateau (Fig. \[fig:crchcm\]b). Note that the same feature was obtained by Chieffi et al. ([@cdhls03]) in modelling the theoretical light curves of SNe IIP with the evolutionary models of pre-SNe. In addition to a deep hydrogen mixing in velocity space downward to $\approx 700$ kms$^{-1}$, the optimal model of SN 1999em is characterized by a weak $^{56}$Ni mixing up to $\approx 660$ kms$^{-1}$ (Fig. \[fig:chcom\] and Table \[tab:hmsn99sn87\]).
The hydrodynamic model for SN 1987A is, in turn, characterized by a moderate $^{56}$Ni mixing in velocity space up to $\sim 3000$ kms$^{-1}$ and a deep hydrogen mixing downward to $\sim 600$ kms$^{-1}$ (Utrobin [@utr05]). Moreover, the observations of SN 1987A provide clear evidence of moderate mixing of the bulk of radioactive $^{56}$Ni up to velocity $\sim$ 3000 kms$^{-1}$ and deep hydrogen mixing down to $\sim$ 500 kms$^{-1}$. For example, the \[Ni II\] 6.64 $\mu$m profile at day 640 gives a velocity $v_{\mathrm{FWHM}}=3100$ kms$^{-1}$ and its modelling results in a maximum velocity of 2600 kms$^{-1}$ (Colgan et al. [@chelh94]). The fact that the H$\alpha$ profile on day 498 (Phillips et al. [@phhsk90]) is not flat-topped implies that there is no large cavity free of hydrogen at the center of the envelope and that hydrogen is mixed downward to $\sim$ 500 kms$^{-1}$.
Within the framework of the neutrino-driven explosion mechanism for core collapse SNe, Kifonidis et al. ([@kpsjm03], [@kpsjm06]) carried out two-dimensional simulations of explosion models with high-order mode perturbations and those with low-order ($l=2$ and $l=1$) unstable modes induced by the initial global deformation of the shock. The low-mode explosion models exhibit final iron-group velocities of $\sim$ 3300 kms$^{-1}$ and deep hydrogen mixing downward to a velocity of $\sim$ 500 kms$^{-1}$, which are consistent with the observations of SN 1987A. The moderate $^{56}$Ni mixing up to a velocity of $\sim$ 3300 kms$^{-1}$ is the result of the larger initial maximum velocities of metal-rich clumps, compared to the high-mode models, which protect the fastest clumps from the strong interaction with the reverse shock that forms below the helium/hydrogen composition interface. The initial global deformation of the shock leads to the growth of the Richtmyer-Meshkov instability that in turn results in the strong inward mixing of hydrogen at the helium/hydrogen composition interface.
It is likely that the strongly anisotropic explosions are able to explain both weak and moderate $^{56}$Ni mixing, depending on the pre-SN structure and the character of the interaction between metal-rich clumps and the reverse shock. Thus, the weak $^{56}$Ni mixing and the deep hydrogen mixing, which are the characteristic features of the optimal model for SN 1999em, cannot be excluded within the framework of the neutrino-driven explosion mechanism (Kifonidis [@k06]). These promising capabilities make our interpretation of the SN 1999em event more physically grounded and confirm the suggestion of Chieffi et al. ([@cdhls03]) that in reality the inner layers of the expelled envelope are strongly mixed during the explosion.
Based on the hydrodynamic models of SNe IIP, Litvinova & Nadyozhin ([@ln83], [@ln85]) constructed simple approximate formulae for evaluating the initial radius, the ejecta mass, and the explosion energy — three basic physical parameters — from the observed properties of individual SN IIP. Using these approximate formulae for SN 1999em and assuming the distance to the host galaxy to be 11.08 Mpc, Nadyozhin ([@n03]) derived the initial radius of 414 $R_{\sun}$, the ejecta mass of 15.0 $M_{\sun}$, and the explosion energy of $0.68\times10^{51}$ erg. The difference in the ejecta mass and the explosion energy, compared to those of the optimal model, results mainly from neglecting the influence of radioactive $^{56}$Ni on the light curve and using the approximation of equilibrium radiative diffusion in their hydrodynamic models.
We go one step further to construct the approximate formulae (\[eq:logr\]–\[eq:loge\]), which include the influence of radioactive $^{56}$Ni on the light curve and are based on the hydrodynamic models computed in terms of radiation hydrodynamics in the one-group approximation with non-LTE effects in the average opacities and the thermal emissivity, with nonthermal ionization, and with the contribution of lines to opacity. The resulting approximate formulae, obtained in space of the physical parameters, are valid only for the vicinity of the optimal model because of the sensitivity of the bolometric light curve to the pre-SN structure and to its chemical composition. To construct the global, approximate relations covering a wider range of the parameters, we have to use a set of hydrodynamic models computed for *many real* SNe IIP. The optimal model for SN 1999em is the first element in this set.
Conclusions {#sec:concl}
===========
We have presented a comprehensive study of a normal Type IIP SN 1999em. Our main aim in this paper was to show the absolute necessity of the simultaneous interpretation of *both* the photometric *and* spectroscopic observations. To do this, we calculated the hydrodynamic models of SN 1999em complemented by the atmosphere models with the time-dependent kinetics and energy balance. We focused mainly on the optimal hydrodynamic model of SN 1999em and subsequent comparison to a peculiar Type IIP SN 1987A. Our results can be summarized in the following conclusions.
- The bolometric light curve of SN 1999em and its spectral evolution of the H$\alpha$ line are consistent with a radius of the pre-SN of $500\pm200 R_{\sun}$, a mass of the ejected envelope of $19.0\pm1.2 M_{\sun}$, an explosion energy of $(1.3\pm0.1)\times10^{51}$ erg, and a mass of radioactive $^{56}$Ni of $0.036\pm0.009 M_{\sun}$, the bulk of which is confined to layers ejected with velocities less than $\approx 660$ kms$^{-1}$. The optimal hydrodynamic model with the helium core mass in the range of 5.6–8.1 $M_{\sun}$ matches the observed bolometric light curve.
- The adequate hydrodynamic and atmosphere models of SN 1999em distinguish between the short distance of 7.85 Mpc, the average value of the EPM distance estimates to the host galaxy, and the Cepheid distance of 11.7 Mpc. They are inconsistent with the short distance of 7.85 Mpc, which should be discarded.
- It is shown that the hydrogen recombination in the atmosphere of a normal Type IIP SN 1999em, as well as most likely other SNe IIP, at the photospheric epoch is essentially a time-dependent phenomenon in accordance with the analysis of Utrobin & Chugai ([@uc02], [@uc05]).
- This study reveals the time development of a normal SN IIP with the following stages: a shock breakout, an adiabatic cooling phase, a phase of cooling and recombination wave, a phase of radiative diffusion cooling, an exhaustion of radiation energy, a plateau tail, and a radioactive tail.
- The plateau tail of the theoretical bolometric light curve coinciding in time with the second plateau observed in SN 1999em (Elmhamdi et al. [@edc03]) runs slightly above a line of the radioactive tail in contrast to the latter. It seems that the real bolometric luminosity of SN 1999em is greater than the “UBVRI” luminosity during the second plateau. The duration of the plateau tail, as well as the second plateau, is correlated with the amount of ejected $^{56}$Ni.
- The bolometric light curve depends weakly on the chemical composition in the outer layers of the ejecta beyond the helium core, while hydrogen is abundant there and controls the opacity of matter. A mutual mixing of hydrogen-rich and helium-rich matter in the inner layers of the ejecta guarantees a good fit of the calculated bolometric light curve to what is observed. Note that the sharp boundary between the hydrogen-rich and helium-rich layers at the edge of the helium core — a characteristic of the evolutionary models of pre-SNe — changes the light curve producing an unobserved bump at the end of the plateau.
- A contribution of numerous metal lines to opacity plays a fundamental role in reproducing the observed light curve of SN 1999em. The decrease in opacity due to disregarding the line contribution speeds up the radiation diffusion and causes the bolometric luminosity to increase significantly compared to what is observed during nearly the whole outburst and, as a consequence, causes the characteristic duration of the light curve to shorten by about 10 days.
- The comparison of a normal Type IIP SN 1999em with a peculiar Type IIP SN 1987A reveals two very important results for SN theory. First, the masses of helium cores in the pre-SNe of SN 1999em and SN 1987A are close enough to suppose that in the final stages of stellar evolution nearly the same iron cores form within the pre-SNe. This fact and roughly the same explosion energies of SN 1999em and SN 1987A together imply a unique explosion mechanism for these core collapse SNe. Second, the optimal model for SN 1999em is characterized by a weaker $^{56}$Ni mixing in velocity space up to $\approx 660$ kms$^{-1}$ compared to a moderate $^{56}$Ni mixing up to $\sim 3000$ kms$^{-1}$ in the case of SN 1987A, hydrogen being mixed deeply downward to $\approx 700$ kms$^{-1}$ and $\sim 600$ kms$^{-1}$.
- A significant excess of the gamma-ray luminosity over the bolometric one, shown by the optimal model of SN 1987A at the radioactive tail, is of great interest and should be taken into account in measuring the $^{56}$Ni amount from the observed bolometric luminosity at this stage when the SN envelope still remains optically thick for gamma-rays. The higher the rate of work done by radiation pressure in an SN envelope, the larger an excess of the gamma-ray luminosity over the bolometric one at the radioactive tail.
- It is shown that the approximation of homologous expansion may be used in the EPM and SEAM for determining the distances to normal SNe IIP starting from nearly day 3 after the SN explosion.
- Based on the hydrodynamic models in the vicinity of the optimal model of SN 1999em, we derive the approximate relationships between the basic physical and observed parameters. The basic physical parameters are the pre-SN radius, the ejecta mass, the explosion energy, and the total $^{56}$Ni mass. The observed properties of SNe IIP are described by the three parameters: the characteristic duration of the light curve, the bolometric luminosity and the photospheric velocity in the middle of the plateau. To cover a wider range of the parameters, beyond the vicinity of the optimal model, we have to construct a global approximation that should be based on a set of hydrodynamic models of *many real* SNe IIP.
Finally, we emphasize that the simultaneous analysis of the photometric and spectroscopic observations is in fact crucial for correctly interpreting core collapse SNe and express the hope that the obtained basic parameters and properties of SN 1999em determine its evolutionary state and will be a reliable reference for the proper mechanism of SN explosions.
The author is grateful to Wolfgang Hillebrandt and Ewald Müller for hospitality during his stay at the MPA. The author would also like to thank Nikolai Chugai, Konstantinos Kifonidis, and Dmitrij Nadyozhin for many discussions, and the referee David Branch for helpful comments. This work was supported in part by the Russian Foundation for Fundamental Research (04-01-17255).
Arnett, W. D. 1979, ApJ, 230, L37
Arnett, W. D. 1987, ApJ, 319, 136
Baklanov, P. V., Blinnikov, S. I., & Pavlyuk, N. N. 2005, Astron. Lett., 31, 429
Barlow, M. J., Sugerman, B. E. K., Fabbri, J., et al. 2005, ApJ, 627, L113
Baron, E., Branch, D., Hauschildt, P. H., et al. 2000, ApJ, 545, 444
Baron, E., Nugent, P. E., Branch, D., & Hauschildt, P. H. 2004, ApJ, 616, L91
Basko, M. 1994, ApJ, 425, 264
Chieffi, A., Domínguez, I., Höflich, P., Limongi, M., & Straniero, O. 2003, MNRAS, 345, 111
Colgan, S. W. J., Haas, M. R., Erickson, E. F., Lord, S. D., & Hollenbach, D. J. 1994, ApJ, 427, 874
Dessart, L., & Hillier, D. J. 2006, A&A, 447, 691
Dufour, R. J. 1984, in IAU Symp. 108: Structure and Evolution of the Magellanic Clouds, ed. S. van den Bergh & K. S. de Boer (Dordrecht: Reidel), 353
Dwek, E. 1983, ApJ, 274, 175
Elmhamdi, A., Danziger, I. J., Chugai, N., et al. 2003, MNRAS, 338, 939
Falk, S. W., & Arnett, W. D. 1977, ApJS, 33, 515
Grassberg, E. K., & Nadyozhin, D. K. 1976, Ap&SS, 44, 409
Grassberg, E. K., Imshennik, V. S., & Nadyozhin, D. K. 1971, Ap&SS, 10, 28
Grassberg, E. K., Imshennik, V. S., Nadyozhin, D. K., & Utrobin, V. P. 1987, Astron. Lett., 13, 227
Grevesse, N., & Sauval, A. J. 1998, Space Sci. Rev., 85, 161
Hamuy, M. 2003, ApJ, 582, 905
Hamuy, M., Pinto, P. A., Maza, J., et al. 2001, ApJ, 558, 615
Heger, A., Langer, N., & Woosley, S. E. 2000, ApJ, 528, 368
Hillebrandt, W., Höflich, P., Truran, J. W., & Weiss, A. 1987, Nature, 327, 597
Imshennik, V. S., & Nadyozhin, D. K. 1989, Astrophysics and Space Physics Reviews, 8, 1
Kifonidis, K. 2006, privite communication
Kifonidis, K., Plewa T., Scheck L., Janka H.-Th., & Müller, E. 2003, A&A, 408, 621
Kifonidis, K., Plewa T., Scheck L., Janka H.-Th., & Müller, E. 2006, A&A, 453, 661
Kirshner, R. P., & Kwan, J. 1974, ApJ, 193, 27
Leonard, D. C., Filippenko, A. V., Ardila, D. R., & Brotherton, M. S. 2001, ApJ, 553, 861
Leonard, D. C., Filippenko, A. V., Gates, E. L., et al. 2002, PASP, 114, 35
Leonard, D. C., Kanbur, S. M., Ngeow, C. C., & Tanvir, N. R. 2003, ApJ, 594, 247
Li, W. D. 1999, IAU Circ. 7294
Litvinova, I. Yu., & Nadyozhin, D. K. 1983, Astrophys. Space Sci., 89, 89
Litvinova, I. Yu., & Nadyozhin, D. K. 1985, Sov. Astron. Lett., 11, 145
Massey, P., Plez, B., Levesque, E. M., et al. 2005, ApJ, 634, 1286
Mihalas, D. 1978, Stellar Atmospheres (Freeman, San-Francisco)
Nadyozhin, D. K. 2003, MNRAS, 346, 97
Phillips, M. M., Hamuy, M., Heathcote, S. R., Suntzeff, N. B., & Kirhakos, S. 1990, AJ, 99, 1133
Pooley, D., Lewin, W. H. G., Fox, D. W., et al. 2002, ApJ, 572, 932
Powell, M. J. D. 1964, Computer Journal, 7, 155
Smartt, S. J. 2002, Astrophys. Space Sci., 281, 187
Smartt, S. J., Maund, J. R., Gilmore, G. F., et al. 2003, MNRAS, 343, 735
Sobolev, V. V. 1980, Afz, 16, 695
Utrobin, V. P. 1989, Astron. Lett., 15, 42
Utrobin, V. P. 2004, Astron. Lett., 30, 293
Utrobin, V. P. 2005, Astron. Lett., 31, 806
Utrobin, V. P., & Chugai, N. N. 2002, Astron. Lett., 28, 386
Utrobin, V. P., & Chugai, N. N. 2005, A&A, 441, 271
Woosley, S. E. 1988, ApJ, 330, 218
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Rough sets were proposed to deal with the vagueness and incompleteness of knowledge in information systems. There are many optimization issues in this field such as attribute reduction. Matroids generalized from matrices are widely used in optimization. Therefore, it is necessary to connect matroids with rough sets. In this paper, we take field into consideration and introduce matrix to study rough sets through vector matroids. First, a matrix representation of an equivalence relation is proposed, and then a matroidal structure of rough sets over a field is presented by the matrix. Second, the properties of the matroidal structure including circuits, bases and so on are studied through two special matrix solution spaces, especially null space. Third, over a binary field, we construct an equivalence relation from matrix null space, and establish an algebra isomorphism from the collection of equivalence relations to the collection of sets, which any member is a family of the minimal non-empty sets that are supports of members of null space of a binary dependence matrix. In a word, matrix provides a new viewpoint to study rough sets.
**Keywords.** Rough set, Matroid, Field, Matrix, Null space, Vector matroid.
author:
- 'Aiping Huang, William Zhu [^1]'
title: Matrix approach to rough sets through vector matroids over a field
---
Introduction
============
The vagueness and incompleteness of knowledge are commom phenomena in information systems. Rough set theory [@Pawlak82Rough], based on equivalence relations (resp. partitions), was proposed by Pawlak in hybrid approaches to improve the performance of data analysis tools. This technique has led to many practical applications in various areas, such as attribute reduction [@HeWuChenZhao11Fuzzy; @FanZhu12Attribute; @MinHeQianZhu11Test; @YaoZhao08Attribute], feature selection [@DashLiu03Consistency-based; @HuYuLiuWu08Neighborhood; @TsengHuang07Rough], rule extraction [@Baesens2003; @Cruz-Cano2012; @DuHuZhuMa11Rule; @WangTsangZhaoChenYeung07Learning], and so on. In order to generalize the rough set theory’s applications, some scholars have extended rough sets to generalized rough sets based on tolerance relation [@SkowronStepaniuk96tolerance], similarity relation [@SlowinskiVanderpooten00AGeneralized] and arbitrary binary relation [@LiuZhu08TheAlgebraic; @ZhuWang06ANew; @Zhu09RelationshipBetween]. Through extending a partition to a covering, rough sets have been extended to covering-based rough sets [@WangZhu12Quantitative; @ZhuWang07OnThree; @Zhu07Topological; @ZhuWang03Reduction]. Matroid theory also has been promoted further to study rough set theory and its applications [@Edmonds71Matroids].
Matroid theory [@Lai01Matroid; @Oxley93Matroid] borrows extensively from linear algebra theory and graph theory. There are dozens of equivalent ways to define a matroid. Significant definitions of a matroid include those in terms of independent sets, bases, circuits, closed sets (resp. flats) and rank functions, which provides well-established platforms to connect with other theories. In applications, matroids have been widely used in many fields such as combinatorial optimization, network flows and algorithm design, especially greedy algorithm design [@Edmonds71Matroids; @Lawler01Combinatorialoptimization]. In recent years, there are many fruitful achievements about the connection between matroids and rough sets [@BarnabeiNicolettiPezzoli98Matroids; @Huang2012; @LiLiu12Matroidal; @WangZhu11Matroidal; @WangZhuZhuMin12Matroidal; @WangZhu12Quantitative; @ZhuWang11Matroidal].
Matrix, which is a good computational tool and easy to represent, compute and accelerate, finds many applications in most scientific fields. In physics, it is used to study physical phenomena, such as the motion of rigid bodies. In computer graphics, it is used to project a 3-dimensional image onto a 2-dimensional screen. In probability theory and statistics, stochastic matrices are used to describe sets of probabilities; for instance, they are used within the PageRank algorithm that ranks the pages in a Google search. As an approach to study rough sets, matrix has existed in many papers [@HuangZhu12On; @Liu10Closures; @Liu06TheTransitive; @Liu10Rough].
In this paper, through another way, namely vector matroids, we introduce matrix to study rough sets. First, an approach to construct a matroid is introduced from the viewpoint of set theory, and a matrix representation of an equivalence relation was proposed. Over a field, though proving the matroid is the same as the one induced by the matrix through vector matroids, we construct a matroidal structure of rough sets over a field from matrix. Second, we introduce two special matrix solution spaces, especially null space, to study the characteristics of the matroid. Over a field, one matrix with entries can induce a vector matroid. It is interesting that the circuits of the matroid are the minimal non-empty sets that are supports of members of the null space of the matrix over the field. Third, we take binary field into consideration and construct an equivalence relation from a matrix null space. Moreover, we find that a collection of equivalence relations and a collection of sets, which any member is a collection of the minimal non-empty sets that are supports of members of null space of a binary dependence matrix, are algebra isomorphism. In a word, this work indicates that we can study rough sets from the viewpoint of matrix.
The rest of this paper is arranged as follows. Section \[S:Basicdefinitions\] reviews some fundamental concepts related to rough set theory, matroid theory and linear algebra theory. In section \[S:Representationofmatroidinducedbyanequivalencerelationoverafiled\], we study a matroidal structure of rough sets over a field through matrix. Section \[S:Nullspaceapproachtoroughsetthroughmatroid\] introduces two special matrix solution spaces, especially null space, to study the matroidal structure. In section \[Equivalencerelationinducedbymatrix\], we construct an equivalence relation from the matrix null space over binary field and obtain an algebra isomorphism system between a collection of equivalence relations and a collection of sets which any member is a family of the minimal non-empty sets that are supports of members of null space of a binary dependence matrix. Section \[S:Conclusions\] concludes this paper.
Basic definitions {#S:Basicdefinitions}
=================
In this section, we present some fundamental concepts about rough set theory, matroid theory and linear algebra theory. First of all, we review some basic concepts of rough set theory.
Rough set theory
----------------
In Pawlak’s rough set theory, the lower and upper approximation operations are two key concepts. An equivalence relation, that is, a partition, is the simplest formulation of the lower and upper approximation operations.
Let $U$ be a finite set and $R$ an equivalence relation on $U$. $R$ will generate a partition $U/R=\{P_{1}, P_{2}, \cdots, P_{s}\}$ on $U$, where $P_{1}, P_{2}, \cdots, P_{s}$ are the equivalence classes generated by $R$. $\forall X\subseteq U$, the lower and upper approximations of $X$ are defined as follows, respectively.\
$R_{\ast}(X)=\bigcup\{P_{i}\in U/R:P_{i}\subseteq X\}$,\
$R^{\ast}(X)=\bigcup\{P_{i}\in U/R:P_{i}\bigcap X\neq \emptyset\}$.
Linear algebra theory
---------------------
In this subsection, we introduce some basic concepts of linear algebra theory used in this paper. Field plays an important role in linear algebra theory, we introduce the concept firstly.
(Field)[@Xiong84Modern] \[field\] A field is defined as a set together with two operations, usually called addition and multiplication, and denoted by $+$ and $\cdot$, respectively, such that the following axioms hold (subtraction and division are defined implicitly in terms of the inverse operations of addition and multiplication):\
(1): For all $a, b \in F$, $a+b \in F$ and $a \cdot b \in F$.\
(2): For all $a, b, c \in F$, $a + (b + c) = (a + b) + c$ and $(a \cdot b) \cdot c = a \cdot (b \cdot c )$.\
(3): For all $a, b \in F$, $a + b = b + a$ and $a \cdot b =b \cdot a$.\
(4): There exists an element of $F$, called the additive identity element and denoted by $0$, such that for all $a \in F$, $a + 0 = a$. Likewise, there is an element, called the multiplicative identity element and denoted by $1$, such that for all $a \in F$, $a \cdot 1 = a$.\
(5): For every $a \in F$, there exists an element $-a \in F$ such that $a + (-a) = 0$. Similarly, for any $a \in F$ other than $0$, there exists an element $a^{-1} \in F$ such that $a \cdot a^{-1} = 1$.\
(6) For all $a, b, c \in F$, the following equalities holds: $ a \cdot (b + c) = (a \cdot b) + (a \cdot c)$ and $(b + c) \cdot a = (b \cdot a) + (c \cdot a)$.
A field is therefore an algebraic structure $<F, +, \cdot, -, ^{-1}, 0, 1>$. Generally, for a field $F$ and positive integer $n$, $V(n, F)$ denotes the $n-$dimensional vector space over $F$. Any element of $V(n, F)$ is denoted as $\mathbf{v}=(v_{1},v_{2}, \cdots, v_{n})^{T}$ where $v_{i} \in F$ ($ 1\leq i \leq n$). The operations on $V(n, F)$ are established as follows. For all $\mathbf{v}=(v_{1}, v_{2}, \cdots, v_{n})^{T} \in V(n, F)$, $\mathbf{v}^{'}=(v_{1}^{'}, v_{2}^{'}, \cdots, v_{n}^{'})^{T} \in V(n, F)$ and $k \in F$, $k \mathbf{v}=(kv_{1}, kv_{2}, \cdots, kv_{n})^{T}$ and $\mathbf{v}+\mathbf{v}^{'}=(v_{1}+v_{1}^{'}, v_{2}+v_{2}^{'}, \cdots, v_{n}+v_{n}^{'})^{T}$.
Null space, as an important concept in linear algebra theory, provides us a method to study rough sets in this paper.
[@Lay2010] Let $F$ be a field and $A$ an $m \times n$ matrix over $F$. The null space of an $m \times n$ matrix $A$, written as $\mathcal{N}_{F}(A)$, is the set of all solutions to the homogeneous equation $A\mathbf{x}=\mathbf{0}$. In set notion, $\mathcal{N}_{F}(A)=\{\mathbf{x}\in V(n, F): A \mathbf{x}=\mathbf{0}\}$.
Vectors $\mathbf{v}_{1}, \mathbf{v}_{2}, \cdots, \mathbf{v}_{n} \in V(n, F)$ are said to be linearly independent over $F$ if there exist $x_{1}, x_{2}, \cdots, x_{n} \in F$ such that the vector equation $x_{1}\mathbf{v}_{1} + x_{2}\mathbf{v}_{2} + \cdots +x_{n}\mathbf{v}_{n}=\mathbf{0}$ has only the trivial solution, and are said to be linearly dependent over $F$ if there exist $c_{1}, c_{2}, \cdots, c_{p} \in F$, not all zero, such that $c_{1}\mathbf{v}_{1} + c_{2}\mathbf{v}_{2} + \cdots +c_{n}\mathbf{v}_{n}= \mathbf{0}$. The rank of matrix $A$ over $F$ is the maximum number of linearly independent columns in $A$ and the maximum number of linearly independent columns in $A^{T}$ (rows in $A$), and we denote it by $r_{F}(A)$.
(Algebra isomorphism) [@Xiong84Modern] Let $(\mathbb{A}, \cdot )$ and $(\mathbb{B}, \circ)$ be two closed algebraic systems. If there exists a bijection $f$ from $A$ to $B$ such that $f(A_{1}\cdot A_{2}) = f(A_{1}) \circ f(A_{2})$ for all $A_{1}, A_{2} \in A$, then we say $f$ is an isomorphism, and $A, B$ are isomorphic, denoted by $A \cong B$.
Matroid theory
--------------
Matroid theory borrows extensively from the terminology of linear algebra theory and graph theory, largely because it is the abstraction of various notions of central importance in these fields, such as independent set, base, rank function. For convenience, we introduce some symbols firstly.
[@Lai01Matroid; @Oxley93Matroid] Let $\mathcal{A}$ be a family of subsets of $U$. One can denote\
$Low (\mathcal{A})=\{X \subseteq U: \exists A\in \mathcal{A}~such~that~A \subseteq X\}$;\
$Min(\mathcal{A})=\{X \subseteq U: \forall A \in \mathcal{A}, if~Y \subseteq X, then~X=Y\}$;\
$Max(\mathcal{A})=\{X \subseteq U: \forall A\in \mathcal{A}, if X \subseteq X, then~X=Y\}$.
(Matroid) [@Lai01Matroid; @Oxley93Matroid] A matroid is an ordered pair $(U,\mathcal{I})$ consisting of a finite set $U$ and a collection $\mathcal{I}$ of subsets of $U$ satisfying the following three conditions:\
(I1) $\emptyset \in \mathcal{I}$;\
(I2) If $I\in \mathcal{I}$ and $I^{'} \subseteq I$, then $I^{'}\in \mathcal{I}$;\
(I3) If $I_{1},I_{2}\in \mathcal{I}$ and $|I_{1}|<|I_{2}|$, then there is an element $e\in I_{2}-I_{1}$ such that $I_{1}\bigcup e\in \mathcal{I}$, where $|X|$ denotes the cardinality of $X$.
Let $M(U,\mathcal{I})$ be a matroid. The members of $\mathcal{I}$ are the independent sets of $M$. A set in $\mathcal{I}$ is maximal, in the sense of inclusion, is called a base of the matroid $M$. If $A\notin \mathcal{I}$, $A$ is called dependent set. In the sense of inclusion, a minimal dependent subset of $U$ is called a circuit of the matroid $M$. The collections of the bases, the dependent sets and the circuits of a matroid $M$ are denoted by $\mathcal{B}(M)$, $\mathcal{D}(M)$ and $\mathcal{C}(M)$, respectively.
Matroids can be defined in many different but equivalent ways. The following definition defines a matroid from the viewpoint of circuit.
(Circuit axiom)[@Lai01Matroid; @Oxley93Matroid] \[P:circuitaxiom\] Let $\mathcal{C}$ be a family of subset of $U$. There exists a matroid $M$ such that $\mathcal{C}=\mathcal{C}(M)$ if and only if $\mathcal{C}$ satisfies the following conditions:\
(C1) $\emptyset \in \mathcal{C}$;\
(C2) for all $C_{1}, C_{2}\in \mathcal{C}$, if $C_{1}\subseteq C_{2}$, then $C_{1}=C_{2}$;\
(C3) for all $C_{1}, C_{2}\in \mathcal{C}$, if $C_{1}\neq C_{2}$ and $x\in C_{1} \bigcap C_{2}$, then there exists $C_{3}\in \mathcal{C}$ such that $C_{3}\subseteq C_{1}\bigcup C_{2}-\{x\}$.
The name “matroid” was coined by Whitney because a class of fundamental examples of such objects arises from matrices in the following way.
(Vector matroid)[@Lai01Matroid; @Oxley93Matroid] Let $U$ be the set of column labels of an $m \times n$ matrix $A$ over a field $F$, and $\mathcal{I}$ the set of subsets $X$ of $U$ for which the columns labeled by $X$ is linearly independent in the vector space $V(m, F)$. Then $(E,\mathcal{I})$ is a matroid. It is called the vector matroid of $A$, which denoted by $M_{F}[A]$.
\[example1\] Let $\mathbf{R}$ be a real number field and $A$ a matrix over $\mathbf{R}$. $$A=\bordermatrix[{[]}]{
& 1 & 2 & 3 & 4 & 5\cr
& 1 & 0 & 1 & 0 &0 \cr
& 0 & 1 & 0 & 1 &1 \cr
}$$ Then $U=\{1, 2, 3, 4, 5\}$ and $M_{\mathbf{R}}[A]=(E, \mathcal{I})$, where $\mathcal{I}=\{\emptyset, \{1\}, \{2\}, \{3\}, \{4\},$ $ \{5\}, \{1,2\},$ $\{1,4\}, \{1,5\}, $ $\{2,3\}, \{3,4\}, \{3,5\}\}$.
(Isomorphism)[@Lai01Matroid; @Oxley93Matroid] Let $M_{1}=M(U_{1}, \mathcal{I}_{1})$ and $M_{2}=M(U_{2}, \mathcal{I}_{2})$ be two matroids. $M_{1}$ and $M_{2}$ are isomorphic, denoted as $M_{1}\cong M_{2}$, if there is a bijection $\varphi:U_{1}\rightarrow U_{2}$ such that $I\in \mathcal{I}_{1}$ if and only if $\varphi(I)\in \mathcal{I}_{2}$.
Matrix to matroidal structure of rough sets over a field {#S:Representationofmatroidinducedbyanequivalencerelationoverafiled}
========================================================
Matrix, which is a good computational tool and easy to represent, compute and accelerate, finds many applications in most scientific fields. As an important branch of matroid theory, vector matroid, which is defined on the set of columns of matrix, provides good tool to study rough sets. In this section, we will construct matroidal structures of rough sets over a field from matrices through vector matroid. First, an existing matroidal structure of rough sets is provided, and a matrix representation of an equivalence relation is established. Given a field, through proving the existing matroidal structure is the same as the one induced by the matrix representation through vector matroids, we construct a matroidal structure of rough sets over a field from matrix. First of all, an approach to induce a matroidal structure from an equivalence relation is provided.
[@WangZhuZhuMin12Matroidal] \[matroidinducedbyenquivalencerelation\] Let $R$ be an equivalence relation on $U$ and $U/R=\{P_{1}, P_{2}, \cdots, P_{s}\}$.
$\mathcal{C}(R)=\{\{x,y\} \subseteq U|\{x,y\} \subseteq P_{i}, \forall i \in \{1, 2, \cdots, s\}\}$
satisfies circuit axiom $(C1), (C2)$ and $(C3)$. Moreover, there exists a matroid $M$ such that $\mathcal{C}(M)=\mathcal{C}(R)$, and we denote this matroid as $M(R)$.
The above proposition proposes an approach to induce a matroidal structure of rough sets from the viewpoint of set theory. Matrix as a research tool has existed in most scientific fields. We also want to use matrix to study rough sets. Therefore, we define a matrix representation of an equivalence relation firstly.
\[matrixinducedbyequivalencerelation\] Let $R$ an equivalence relation on $U=\{x_{1}, x_{2}, \cdots, x_{n}\}$ and $U/ R =\{P_{1}, P_{2}, \cdots, P_{s}\}$. We denote a matrix $B(R)=(b_{ij})_{s\times n}$ as follows:\
$$\begin{aligned}
b_{ij}=\left\{\begin{aligned}%
&1 && \mbox x_{j} \in P_{i},\\
&0 && \mbox x_{j} \notin P_{i}.\\
\end{aligned}\right.\end{aligned}$$
$B(R)$ does not contain zero rows and zero columns, and any column of it has only one non-zero component. If we denote $1$ and $0$ as the multiplicative and additive identity elements of $F$, respectively, then $B(R)$ can be regarded as a matrix over $F$.
For any element of $U$, we can denote it by $x_{i}(i\in \mathbf{N}_{+})$. For any element of $U/R$, we can denote it by $P_{i}(i \in \mathbf{N}_{+})$. In above definition, we label the columns of $B(R)$ by the elements of $U$ (in the sequential order of $\Gamma_{U}$) and the rows of $B(R)$ by the elements of $U/R$ (again, in the sequential order of $\Gamma_{U/R}$), where $\Gamma_{U}$ indicates the index set of all elements of $U$ and $\Gamma_{U/R}$ indicates the index set of all elements of $U/R$. If the order of $\Gamma_{U}$(resp. $\Gamma_{U/R}$) changes, then $B(R)$ changes accordingly. An example is provided to illustrate the statements.
\[example2\] Suppose $R$ is an equivalence relation on $U=\{x_{1}, x_{2}, x_{3}, x_{4}, x_{5}\}$ and $U/R=\{P_{1}, P_{2}\}$, where $P_{1} = \{x_{1},x_{3}\}$ and $P_{2} = \{x_{2},x_{4},x_{5}\}$. Then $\Gamma_{U} = \{1,2,3,4,$ $5\}$(sequential order) and $\Gamma_{U/R} = \{1,2\}$. We also can obtain one matrix representation of $R$ as follows: $$B(R)=\bordermatrix[{[]}]{
& x_{1} & x_{2} & x_{3} & x_{4} & x_{5}\cr
P_{1}& 1 & 0 & 1 & 0 &0 \cr
P_{2}& 0 & 1 & 0 & 1 &1 \cr
}.$$ If $U = \{x_{1}, x_{3}, x_{2}, x_{4}, x_{5} \}$ and $U/R=\{P_{1}, P_{2}\}$, then $\Gamma_{U} = \{1, 3, 2, 4, 5\}$ and $\Gamma_{U/R} = \{1,2\}$. We also can obtain the other matrix representation of $R$ as follows: $$B(R)=\bordermatrix[{[]}]{
& x_{1} & x_{3} & x_{2} & x_{4} & x_{5}\cr
P_{1}& 1 & 1 & 0 & 0 &0 \cr
P_{2}& 0 & 0 & 1 & 1 &1 \cr
}.$$ If $U = \{x_{1}, x_{3}, x_{2}, x_{4}, x_{5} \}$ and $U/R=\{P_{2}, P_{1}\}$, then $\Gamma_{U} = \{1,2,3,4,$ $5\}$ and $\Gamma_{U/R} = \{2,1\}$. We also can obtain another matrix representation of $R$ as follows: $$B(R)=\bordermatrix[{[]}]{
& x_{1} & x_{3} & x_{2} & x_{4} & x_{5}\cr
P_{1}& 0 & 0 & 1 & 1 &1 \cr
P_{2}& 1 & 1 & 0 & 0 &0 \cr
}.$$ Therefore, different orders of $\Gamma_{U}$ (resp. $\Gamma_{U/R}$) determine different matrix representations of $R$.
As we know, a matrix with entries in a field gives rise to a matroid on its set of columns through vector matroids. The dependent sets of the matroid are those columns of the matrix that are linearly dependent as vectors over the field. Let $M=M_{F}[A]$. In general, $M$ does not uniquely determine $A$. One can obtain a matrix from $A$ by using some row elementary transformations which root in matroid theory. It is not difficult to check that $M$ remains unchanged through these transformations. In addition to that, if we interchange any two columns of $A$ with the labels of them, then $M$ remains unchanged. Once the labels change but the columns labeled by them dose not change accordingly, then these two vector matroids may not be the same one. In order to understand the above standpoints better, we take the following matrix for example.
\[example3\] Let us revisit Example \[example2\]. Suppose $U=\{x_{1}, x_{2}, x_{3}, x_{4}, x_{5}\}$ and $U/R=\{P_{1}, P_{2}\}$. We obtain a matrix $B(R)$ (the first one in Example \[example2\]). If we interchange the $2th$ column and the $3th$ column of $B(R)$ but the labels of them remain unchanged, then we obtain the other matrix as follows: $$B^{'}(R)=\bordermatrix[{[]}]{
& x_{1} & x_{2} & x_{3} & x_{4} & x_{5}\cr
& 1 & 1 & 0 & 0 &0 \cr
& 0 & 0 & 1 & 1 &1 \cr
}.$$ If we interchange the $2th$ column and the $3th$ column of $B(R)$ with the labels of them, then we obtain another matrix (the second one in Example \[example2\]) and we denote it by $B^{''}(R)$. It is clear that $M_{\mathbf{R}}[B^{'}(R)] = (U, \mathcal{I}^{'})$, where $\mathcal{I}^{'}=\{\emptyset, \{x_{1}\}, \{x_{2}\}, \{x_{3}\}, \{x_{4}\},$ $ \{x_{5}\},
\{x_{1},x_{3}\},$ $\{x_{1},x_{4}\}, \{x_{1},x_{5}\}, $ $\{x_{2},x_{3}\}, \{x_{2},x_{4}\}, \{x_{2},x_{5}\}\}$ and $M_{\mathbf{R}}[B^{''}(R)]$ $= (U, \mathcal{I}^{''})$, where $\mathcal{I}^{''}=\{\emptyset, \{x_{1}\}, \{x_{2}\},$ $ \{x_{3}\},
\{x_{4}\},$ $ \{x_{5}\}, \{x_{1}, x_{2}\},$ $\{x_{1}, x_{4}\}, \{x_{1}, x_{5}\}, $ $\{x_{2}, x_{3}\}, \{x_{3}, x_{4}\},$ $ \{x_{3},x_{5}\}\}$. One can define a mapping $\varphi:U \rightarrow U$ as follows: $\varphi(x_{2})=x_{3}$, $\varphi(x_{3})=x_{2}$ and $\varphi(x_{i})= x_{i}(i=1,4,5)$. It is obvious that the mapping $\varphi$ is bijection. Thus $M_{\mathbf{R}}[B(R)] =M_{\mathbf{R}}[B^{''}(R)] \cong M_{\mathbf{R}}[B^{'}(R)]$.
As the order of row labels of a matrix dose not change the vector matroid induced by the matrix, we will not take them into consideration in the following discussion. For a ground set $U$, we denote the $ith$ element of $U$ by $x_{i}$ ($i\in \mathbf{N}_{+}$ and $ 1 \leq i \leq |U|$) and we obtain $\Gamma_{U}$. For an order of $\Gamma_{U}$, we obtain one matrix representation of an equivalence relation. Changing the order of $\Gamma_{U}$, we obtain the other matrix. Essentially, the latter matrix is obtained from the former matrix through exchanging some columns with the labels of them. Therefore, for arbitrary order of $\Gamma_{U}$, we obtain different $B(R)$. However, these matrices induce the same matroid over the same field. In a word, in order to study the relation between the matroid and $M(R)$, we just need to study the relation between $M(R)$ and the vector matroid induced by $B(R)$ which defined in Definition \[matrixinducedbyequivalencerelation\]. The following theorem indicates that the matroid is the same as $M(R)$ over any field.
\[T: circuitsbetweenmatroidandvectormatroid\] Let $R$ be an equivalence relation on $U$ and $B(R)$ a matrix representation of $R$ over $F$. $M(R) = M_{F}[B(R)]$.
Let $U=\{x_{1}, x_{2}, \cdots, x_{n}\}$. We can obtain $B(R) =[\beta_{1}, \beta_{2}, \cdots, $ $\beta_{n}]$ and we know the columns of $B(R)$ are labeled, in order, by $x_{1}, x_{2}, \cdots, x_{n}$. Since a set of circuits decides only one matroid, we need to prove only $\mathcal{C}(R)=\mathcal{C}(M_{F}[B(R)])$, i.e., $\mathcal{C}(R)=Min (\{X \subseteq U:$ The columns of $B(R)$ labeled by $X$ are linearly dependent in $V(n, F)\})$. Suppose $U/R=\{P_{1}, P_{2}, \cdots, $ $P_{s}\}$. $\forall \{x_{i}, x_{j}\} \in \mathcal{C}(R)$, there exists $P_{k}\in U/R$ such that $\{x_{i}, x_{j}\} \subseteq P_{k}$. According to the definition of $B(R)$, we know that $\beta_{i} = \beta_{j}$. Thus $\beta_{i}$ and $\beta_{j}$ are linearly dependent in $V(n, F)$. Because $B(R)$ dose not contain zero columns, the column labeled by $x_{i}$ or $x_{j}$ is linearly dependent in $V(n, F)$. Hence $\mathcal{C}(R)\subseteq Min(\{X \subseteq U:$ The columns of $B(R)$ labeled by $X$ are linearly dependent in $V(n, F)\})$. For all $X\in $ $Min(\{$ $X \subseteq U:$ The columns of $B(R)$ labeled by $X$ are linearly dependent in $V(n, F)$ $\})$, then $X$ is a dependent set in $M(R)$; otherwise, $X$ dose not contain circuits. According to the definition of $B(R)$, we know the columns of $B(R)$ labeled by $X$ are different and these columns form a $|X| \times |X|$ identity matrix, where $|X|$ denotes the cardinality of $X$. Hence these columns are linearly independent in $V(n,F)$, which implies contradictory. Now we need to prove $X$ is a circuit of $M(R)$. From above proof, we know $X$ is circuit of $M(R)$; otherwise, there exists $C \in \mathcal{C}(M(R))$ such that $C \subset X$, i.e., $C \in \{X \subseteq U:$ The columns of $B(R)$ labeled by $X$ are linearly dependent in $V(n, F)\}$ such that $C \subset X$ because $\mathcal{C}(R)\subseteq Min(\{X \subseteq U:$ The columns of $B(R)$ labeled by $X$ are linearly dependent in $V(n, F)\})$. That contradicts the fact that $ X\in Min(\{$ $X \subseteq U:$ The columns of $B(R)$ labeled by $X$ are linearly dependent in $V(n, F)$ $\})$. Hence, $Min \{X \subseteq U:$ The columns of $B(R)$ labeled by $X$ are linearly dependent in $V(n, F)\} \subseteq \mathcal{C}(R)$.
As we know, a common field is finite field or Galois field which contains a finite number of elements. Now we introduce the simplest field, i.e., binary field.
(Binary field)[@Xiong84Modern] Let $GF=\{0,1\}$. If the addition and multiplication of $GF$ are defined in Table 1, then $(GF,+,\cdot)$ is called binary field and we denote it as $GF(2)$.
$GF(2)$ is a special field. Based on Theorem \[T: circuitsbetweenmatroidandvectormatroid\], it is not difficult for us to obtain the following corollary.
Let $R$ be an equivalence relation on $U$. $M(R) = M_{GF(2)}[B(R)]$.
For an order of $\Gamma_{U}$, we can obtain a matrix $B(R)$. From above theorem, we find that the matrix induces the same matroid over different fields, which is determined by the particularity of the matrix. However, in many cases, the vector matroids induced by the same matrix over different fields may not be the same one. The example below illustrates this viewpoint.
Suppose
$A = \left[\begin{array}{cccccc}
1 ~&~ 0 ~&~ 0 ~&~1 ~&~ 1 ~&~ 0 \\
0 ~&~ 1 ~&~ 0 ~&~1 ~&~ 0 ~&~ 1 \\
0 ~&~ 0 ~&~ 1 ~&~0 ~&~ 1 ~&~ 1
\end{array}\right]$.
We may as well suppose $A=[\alpha_{1}, \alpha_{2}, \alpha_{3}, \alpha_{4}, \alpha_{5}, \alpha_{6}]$ which are labeled, in order, by $x_{1}, x_{2}, x_{3}, x_{4}, x_{5}, x_{6}$. Then $\alpha_{4}, \alpha_{5}, \alpha_{6}$ are linearly independent over real number field because $det([\alpha_{4}, \alpha_{5}, \alpha_{6}])=-2 \neq 0$. However, $det([\alpha_{4}, \alpha_{5}, \alpha_{6}])=0$ over $GF(2)$, i.e., $\alpha_{4}, \alpha_{5}, \alpha_{6}$ are linearly dependent over binary field. In addition to that, any two columns of $\alpha_{4}, \alpha_{5}$ and $\alpha_{6}$ are linearly independent over $GF(2)$ because the determinant of them is not zero over $GF(2)$. Therefore, $\{x_{4},x_{5},x_{6}\} \in \mathcal{C}(M_{GF(2)}[A])$ but $\{x_{4},x_{5},x_{6}\} \notin \mathcal{C}(M_{\mathbf{R}}[A])$, i.e., $M_{GF(2)}[A] \neq M_{\mathbf{R}}[A]$.
Matroidal structure of rough sets over a field to two special matrix solution spaces {#S:Nullspaceapproachtoroughsetthroughmatroid}
=====================================================================================
In section \[S:Representationofmatroidinducedbyanequivalencerelationoverafiled\], we has obtained a matroidal structure of rough sets over a field from matrix. In this section, we will study some characteristics of the matroidal structure through two matrix solution spaces, especially matrix null space. First of all, an operator is proposed to connect vector space with set theory.
[@Lai01Matroid; @Oxley93Matroid] Let $F$ be a field and $U=\{x_{1}, x_{2}, \cdots, x_{n}\}$. We define a mapping $\theta: V(n, F) \rightarrow 2^{U}$ as follows: for all $\mathbf{v}=(v_{1}, v_{2}, \cdots, v_{n})^{T} \in V(n, F)$, $\theta(\mathbf{v})=\{x_{i} \in U: v_{i} \neq 0, 1 \leq i \leq n\}$, where $0$ is the additive identity element of $F$. we call $\theta(\mathbf{v})$ the support of $\mathbf{v}$.
Let $F$ be a field and $U = \{x_{1}, x_{2}, x_{3}, x_{4}, x_{5}\}$. If $\mathbf{v}_{1}=(1, -1, 3, 0, 4 )^{T} \in V(5, F)$ and $\mathbf{v}_{2}=(1, 1, 0, 0, 1)^{T} \in V(5, F)$, then $\theta(\mathbf{v}_{1}) = \{x_{1}, x_{2}, x_{3}, x_{5}\}$ and $\theta(\mathbf{v}_{2}) = \{x_{1}, x_{2}, x_{5}\}$.
Matrix null space is an important concept in linear algebra theory. According to the characteristic of it and the definition of vector matroid, it is natural for us to combine them with each other.
\[P:nullspaceanddependentset\] Let $F$ be a field and $A$ an $m\times n$ matrix over $F$. If $M=M_{F}[A]$, then $\{\theta(\mathcal{N}_{F}(A))-\emptyset\} \subseteq \mathcal{D}(M)$.
It is well known that, by row elementary transformations and some column elementary transformations which root in matroid theory, one can reduce any matrix $G$ to the form $[I_{r}| D]$, where $r = r_{F}(G)$, $I_{r}$ is the $r \times r$ identity matrix and $D$ is a $r \times (n-r)$ matrix over $F$. Then we may as well suppose $A=[I_{r}| D]=[\alpha_{1}, \alpha_{2}, \cdots, \alpha_{n}]$. And the columns of $A$ are labeled, in order, by $x_{1}, x_{2}, \cdots, x_{n}$. If $r=n$, then $A=I_{n}$. Since $r_{F}(A) = n$. Thus $D(M)=\emptyset$, and $A \mathbf{v}=\mathbf{0}$ has only trivial solution, i.e, $\{\theta(\mathcal{N}_{F}(A))-\emptyset\} = \emptyset$. Hence we obtain the result. If $r < n$, then $A \mathbf{v}=\mathbf{0}$ has nontrivial solution, i.e., $\{\theta(\mathcal{N}_{F}(A))-\emptyset\} \neq \emptyset$. For all $D = \{x_{i_{1}}, x_{i_{2}}, \cdots, x_{i_{s}}\} \in \{\theta(\mathcal{N}_{F}(A))-\emptyset\}$, there exists $\mathbf{v}_{D} = (v_{1}, v_{2}, \cdots, v_{n})^{T} \in V(n, F)$ such that $A \mathbf{v}_{D} = \mathbf{0}$ and $\theta(\mathbf{v}_{D})=D$. Then $v_{i_{1}}, v_{i_{2}}, \cdots, v_{i_{s}}(1 \leq s \leq n)$ are non-zero components of vector $\mathbf{v}_{D}$. Thus $\mathbf{0} = A \mathbf{v}_{D} = \sum_{i=1}^{n} v_{i} \alpha_{i}= \sum_{j=1}^{s} v_{i_{j}} \alpha_{i_{j}}$, that is, the columns of $A$ labeled by $D$ are linearly dependent over $F$. Hence, $\{\theta(\mathcal{N}_{F}(A))-\emptyset\} \subseteq \mathcal{D}(M)$.
Conversely, the collection of dependent sets of $M_{F}[A]$ may not be contained in $\{\theta(\mathcal{N}_{F}(A))$ $-\emptyset\}$. The following example illustrates that viewpoint.
Let us revisit Example \[example2\]. We may as well suppose $B(R)=[\mathbf{\beta}_{1}, \mathbf{\beta}_{2}, \mathbf{\beta}_{3}, \mathbf{\beta}_{4},$ $\mathbf{\beta}_{5}]$ (the first one in Example \[example2\]). We know that the columns of $B(R)$ are labeled, in order, by $x_{1}, x_{2}, x_{3}, x_{4}, x_{5}$. It is easy to check that $\mathbf{\beta}_{1}, \mathbf{\beta}_{2}, \mathbf{\beta}_{4}$ are linearly dependent in $V(n, GF(2))$, that is, $\{x_{1}, x_{2}, x_{4}\}$ $ \in \mathcal{D}(M_{GF(2)}[B(R)])$. Suppose $\mathbf{x}=(1, 1, 0, 1, 0)^{T}\in V(5, GF(2))$. We know that $\{x_{1}, x_{2}, x_{4}\}=\theta(\mathbf{x})$, but $B(R)\mathbf{x}=(1,0)^{T} \neq \mathbf{0}$ over binary field. Thus $\{x_{1}, x_{2}, x_{4}\} \notin \{\theta(\mathcal{N}_{GF(2)}(B(R)))-\emptyset\}$.
What about the relation between the matrix null space and the circuits of a vector matroid? In order to solve this problem, we present the following proposition firstly.
\[theminimalityofset\] Let $\mathcal{F}$ and $\mathcal{S}$ are two families of subsets of $U$. If $\mathcal{F} \subseteq \mathcal{S}$ and $Min(\mathcal{S})\subseteq Min(\mathcal{F})$, then $Min(\mathcal{S}) = Min(\mathcal{F})$.
We need to prove $Min(\mathcal{F}) \subseteq Min(\mathcal{S})$. If $Min(\mathcal{F}) \nsubseteq Min(\mathcal{S})$, then there exists $F \in Min(\mathcal{F}) - Min(\mathcal{S})$. Since $\mathcal{F} \subseteq \mathcal{S}$, $Min(\mathcal{F}) \subseteq \mathcal{S}$, i.e, $F \in Min(\mathcal{F}) - Min(\mathcal{S}) \subseteq \mathcal{S} - Min(\mathcal{S})$. Thus there exists $S \in \mathcal{S}$ such that $S \subset F$. Denote $\mathcal{W} =Min \{W \in \mathcal{S}: W \subset F\}$. For all $W \in \mathcal{W}$, we know $W \in Min (\mathcal{S})$; otherwise, there exists $W_{1} \in \mathcal{S}$ such that $W_{1} \subset W$. Since $W \subseteq F$, $W_{1} \subset F$. Thus there exists $W_{1} \in \{W \in \mathcal{S} : W \subset F\}$ such that $W_{1} \subset W$, which contradicts $W \in \mathcal{W}$. Therefore we have $W \in Min(\mathcal{F})$ according to $Min(\mathcal{S}) \subseteq Min(\mathcal{F})$, which contradict the fact that $F \in Min(\mathcal{F})$. Hence we prove $Min(\mathcal{F}) \subseteq \mathcal{A}$, that is, $\mathcal{A} = Min(\mathcal{F})$.
For any two family of sets $\mathcal{F}$ and $\mathcal{S}$, if $\mathcal{F} \subseteq \mathcal{S}$, then we may not have the result $Min(\mathcal{F}) \subseteq Min(\mathcal{S})$.
Let $\mathcal{F} = \{\{2,3\}\}$ and $\mathcal{S} = \{\{2\}, \{3\}, \{2,3\}\}$. It is obvious that $\mathcal{F} \subseteq \mathcal{S}$, but $Min{F} = \{\{2,3\}\} \nsubseteq Min(S) = \{\{2\}, \{3\}\}$.
The following theorem indicates that, over a field, the collection of circuits of the vector matroid induced by a matrix is just the collection of the minimal non-empty sets that are supports of members of null space of the matrix.
\[circuitandresolutionspace\] Let $F$ be a field and $A$ an $m\times n$ matrix over $F$. If $M=M_{F}[A]$, then $\mathcal{C}(M)=Min(\{\theta(\mathcal{N}_{F}(A))-\emptyset\})$.
It is well known that, by row elementary transformations and some column elementary transformations which root in matroid theory, one can reduce matrix $G$ to the form $[I_{r}| D]$, where $r = r_{F}(G)$, $I_{r}$ is the $r \times r$ identity matrix and $D$ is a $r \times (n-r)$ matrix over $F$. Then we may as well suppose $A=[I_{r}| D]=[\mathbf{\alpha}_{1}, \mathbf{\alpha}_{2}, \cdots, \mathbf{\alpha}_{n}]$ of which the columns are labeled, in order, by $x_{1}, x_{2}, \cdots, x_{n}$. If $r=n$, then $\mathcal{C}(M)=\emptyset$ and $A \mathbf{v}=\mathbf{0}$ has only trivial solution, i.e, $\{\theta(\mathcal{N}_{F}(A))-\emptyset\} = \emptyset$. Thus we obtain the result. If $r < n$, then $A \mathbf{v}=\mathbf{0}$ has nontrivial solution, i.e., $\{\theta(\mathcal{N}_{F}(A))-\emptyset\} \neq \emptyset$. We may as well suppose $C=\{x_{i_{1}}, x_{i_{2}}, \cdots, x_{i_{s}}\}$ is a circuit of $M$, where $1 \leq s \leq n$. According to the definition of vector matroid, we know the columns of $A$ which are labeled by the elements of $C$ are linearly dependent, that is, there exist some elements $k_{i_{1}}, k_{i_{2}}, \cdots, k_{i_{s}}$ of $F$ which all are not equal to zero such that $\mathbf{0} = k_{i_{1}} \alpha_{i_{1}} + k_{i_{2}} \alpha_{i_{2}} + \cdots + k_{i_{s}} \alpha_{i_{s}}=
\sum_{j=1}^{s} k_{i_{j}}\alpha_{i_{j}} + \sum_{t\neq i_{j}, 1 \leq j \leq s} 0 \cdot \alpha_{t}=A \mathbf{v}$, where $\mathbf{v}=(v_{1}, v_{2}, \cdots, v_{n})^{T}\in V(n,F)$. If $p \neq i_{j}(1 \leq j \leq s)$, then $v_{p}=0$. Of course, if $p = i_{j}(1 \leq j \leq s)$, then $v_{p}$ may equal to $0$. Next, we want to prove that $k_{i_{j}} \neq 0$ for all $j \in \{1, 2, \cdots, s\}$; otherwise, we may as well suppose $k_{i_{1}}, k_{i_{2}}, \cdots, k_{i_{t}} \neq 0 (t < s)$, then $\theta(\mathbf{v}^{'}) \subset C$, where $\mathbf{v}^{'} = (v_{1}^{'}, v_{2}^{'}, \cdots, v_{n}^{'})^{T}$, and $ v_{p}^{'} \neq 0$ if and only if $p\in \{i_{1}, i_{2}, \cdots, i_{t}\}$. Moreover, $\mathbf{0} = k_{i_{1}} \alpha_{i_{1}} + k_{i_{2}} \alpha_{i_{2}} + \cdots + k_{i_{s}} \alpha_{i_{s}}=k_{i_{1}}\alpha_{i_{1}} + k_{i_{2}}\alpha_{i_{2}} + \cdots + k_{i_{t}}\alpha_{i_{t}}$, then the columns $\alpha_{i_{1}}, \alpha_{i_{2}}, \cdots, \alpha_{i_{t}}$ are linearly dependent in $V(n,F)$, hence $\theta(\mathbf{v}^{'})$ is a dependent set of $M$. According to the definition of dependent set of matroid, there exists $C_{1} \in \mathcal{C}(M)$ such that $C_{1} \subseteq \theta(\mathbf{v}^{'}) \subset C$. According to the (2) of circuit axiom, we can obtain the contradictory. Hence, $\theta(\mathbf{v})=C$, that is, $C \in \{\theta(\mathcal{N}_{F}(A))-\emptyset\}$. Moreover, $C \in Min(\{\theta(\mathcal{N}_{F}(A))-\emptyset\})$; otherwise, there exists $D \in \{\theta(\mathcal{N}_{F}(A))-\emptyset\}$ such that $D \subset C$. According to Proposition \[P:nullspaceanddependentset\], we have $\{\theta(\mathcal{N}_{F}(A))-\emptyset\} \subseteq \mathcal{D}(M)$. Thus $D\in \mathcal{D}(M)$, that is, there exists $C_{2} \in \mathcal{C}(M)$ such that $C_{2} \subseteq D \subset C$ which contradicts the (2) of circuit axiom. Therefore $Min (\mathcal{D}(M)) = \mathcal{C}(M) \subseteq Min \{\theta(\mathcal{N}_{F}(A))-\emptyset\}$. Combing with $\{\theta(\mathcal{N}_{F}(A))-\emptyset\} \subseteq \mathcal{D}(M)$ and Proposition \[theminimalityofset\], we can obtain $Min (\mathcal{D}(M))= Min (\{\theta(\mathcal{N}_{F}(A))-\emptyset\})=\mathcal{C}(M))$.
As we know, over the same field, any two matrices which have the same number of columns may generate the same vector matroid. Theorem \[circuitandresolutionspace\] indicates that the families of the minimal non-empty sets that are supports of members of null space of these matrices are unique. Combining with Theorem \[T: circuitsbetweenmatroidandvectormatroid\] and \[circuitandresolutionspace\], we obtain the following two corollaries.
\[circuitandresolutionspaceofequivalencerelation\] $\mathcal{C}(R)=Min(\{\theta(\mathcal{N}_{F}(B(R)))-\emptyset\})$.
For a field, the addition and multiplication of it are uncertain, we can not compute the circuits of a matroid easily. Binary field is the simplest field and the operations of it are clear. Hence we can obtain the circuits of $M(R)$ easily through calculating null space of $B(R)$ over the field.
\[nullspaceandcirciuitoverbinaryfield\] $\mathcal{C}(R)=Min(\{\theta(\mathcal{N}_{GF(2)}(B(R)))-\emptyset\})$.
According to the particularity of binary field, we may consider whether we can characterize $M(R)$ by using other solution spaces. Inspired by the null space, we introduce the other matrix solution space to study $M(R)$.
Let $F$ be a field and $A$ an $m \times n$ matrix over $F$. We denote the the set of all solutions to the non-homogeneous equation $A\mathbf{x}=\mathbf{1}$ by $I_{F}(A)$. In set notion, $I_{F}(A)=\{\mathbf{x}\in V(n, F): A \mathbf{x}=\mathbf{1}\}$.
So how do you use the space to characterize matroid $M(R)$ over binary field? Firstly, we establish the relation between $\{\theta(I_{GF(2)}(B(R)))\}$ and the family of sets of which the upper approximations are equal to ground set.
\[anotherspaceandroughset\] Let $R$ be an equivalence relation on $U$ and $B(R)$ a matrix representation of $R$. $\{\theta(I_{GF(2)}(B(R)))\} \subseteq \{X \subseteq U: R^{\ast}(X)=U\}$.
Suppose $U=\{x_{1}, x_{2}, \cdots, x_{n}\}$ and $U/R=\{P_{1}, P_{2}, \cdots, P_{s}\}$. We can obtain $B(R)=[\beta_{1}, \beta_{2}, \cdots, \beta_{n}]$ of which the columns are labeled, in order, by $x_{1}, x_{2}, \cdots, x_{n}$. For all $X \in \{\theta(I_{GF(2)}(B(R)))\}$, we may as well suppose $X=\{x_{i_{1}}, x_{i_{2}}, \cdots, x_{i_{t}}\}$ and $\mathbf{v} \in V(n, GF(2))$ such that $\theta(\mathbf{v})=X$. Then we can obtain $\mathbf{\beta}_{i_{1}} + \mathbf{\beta}_{i_{2}} + \cdots + \mathbf{\beta}_{i_{t}}= \mathbf{1}$. Assume $A = [\mathbf{\beta}_{i_{1}}, \mathbf{\beta}_{i_{2}}, \cdots, \mathbf{\beta}_{i_{t}}]$. In order to satisfy the equality $\mathbf{\beta}_{i_{1}} + \mathbf{\beta}_{i_{2}} + \cdots + \mathbf{\beta}_{i_{t}}= \mathbf{1}$ over binary field, the $jth(j \in \{1, 2, \cdots, s\})$ row of $A$ has odd number of $1$. Combining with the definition of $B(R)$, we have $X \bigcap P_{i} \neq \emptyset$ for all $i\in \{1, 2, \cdots, s\}$, that is, $R^{\ast}(X)=U$. Thus $X \subseteq \{X \subseteq U: R^{\ast}(X)=U\}$, that is, $\{\theta(I_{GF(2)}(B(R)))\} \subseteq \{X: R^{\ast}(X)=U\}$.
However, there exists a subset $X$ of $U$ satisfies $R^{\ast}(X) = U$ but $B(R)\mathbf{v} \neq \mathbf{1}$ over binary field, where $\mathbf{v} \in V(n, GF(2))$ and $\theta(\mathbf{v}) = X$.
Let us revisit Example 2. We take the first one matrix $B(R)$ for example. Suppose $X=\{x_{1}, x_{2}, x_{3}\}$. Then $\mathbf{v}=(1,1,1,0,0)^{T}$ $ \in V(5,GF(2))$ and $\theta(\mathbf{v})=X$. It is clear that $R^{\ast}(X)=U$, but $B(R)\mathbf{v}= \mathbf{\beta}_{1}+ \mathbf{\beta}_{2} + \mathbf{\beta}_{3}= (0,1)^{T}$, i.e., $X \notin \{\theta(I_{GF(2)}(B(R)))\}$ over binary field. Hence $\{X: R^{\ast}(X)=U\} \nsubseteq \{\theta(I_{GF(2)}(B(R)))\}$.
In order to characterize $M(R)$ by using $I_{GF(2)}(B(R))$, we need to search a certain characteristic of the matroid which has close relation with the set $\{X \subseteq U: R^{\ast}(X)=U\}$. According to the peculiarity of $B(R)$, the following proposition establishes the bases of $M(R)$.
\[baseofM(R)\] Let $R$ be an equivalence relation on $U$ and $U/R = \{P_{1}, P_{2}, \cdots, P_{s}\}$. $\mathcal{B}(M(R)) = \{X \subseteq U: |X \bigcap P_{i}| = 1, \forall i\in \{1, 2, \cdots, s\}\}$.
Let $U=\{x_{1}, x_{2}, \cdots, x_{n}\}$. we obtain the $B(R)=[\beta_{1}, \beta_{2},$ $ \cdots, \beta_{n}]$ of which the columns are labeled, in order, by $x_{1}, x_{2}, \cdots, x_{n}$. According to the definition of matrix $B(R)$, we know $r_{F}(B(R))=s$. Since $M(R) = M_{F}([B(R)])$, $B$ is a base of $M(R)$ if and only if the columns of $B(R)$ labeled by the elements of $B$ is a maximal independent subset of the matrix. Since $r_{F}(B(R))=s$, any linearly independent columns of $B(R)$ with the cardinality $s$ form a maximal independent subset of $B(R)$. For all $B\in \mathcal{B}(M(R))$, we may as well suppose $B=\{x_{i_{1}}, x_{i_{2}}, \cdots, $ $x_{i_{s}}\}$. Then $x_{i_{j}} \in P_{j}$ for all $j\in \{1, 2, \cdots, s\}$; otherwise, we may as well suppose there exist $x_{i_{1}}$ and $x_{i_{2}}$ such that $x_{i_{1}}, x_{i_{2}}\in P_{1}$, then $\beta_{i_{1}}=\beta_{i_{2}}$. Thus $\beta_{i_{1}}$ and $\beta_{i_{2}}$ are linearly dependent in $V(n,F)$, which makes the columns labeled by $B$ are linearly dependent in $V(n,F)$. That contradicts $B \in \mathcal{B}(M([B(R)])$. Hence $|B\bigcap P_{i}| = 1$ for all $i\in \{1,2,\cdots, s\}$, i.e., $\mathcal{B}(M(R)) \subseteq \{X \subseteq U: |X \bigcap P_{i}| = 1, \forall i\in \{1, 2, \cdots, s\}\}$. Conversely, $\forall X\in \{X \subseteq U: |X \bigcap P_{i}| = 1, \forall i\in \{1, 2, \cdots, s\}\}$, then the cardinality of $X$ is $s$ and the columns of $B(R)$ labeled by $X$ are linearly independent over $F$, hence $X \in \mathcal{B}(M_{F}[B(R)])$. According to Theorem \[T: circuitsbetweenmatroidandvectormatroid\], We have $\mathcal{B}(M_{F}[B(R)]) = \mathcal{B}(M(R))$. Thus we obtain the result.
According to the relation between the independent sets and the bases of a matroid, we obtain the following corollary.
Let $R$ be an equivalence relation on $U$ and $U/R = \{P_{1}, P_{2}, \cdots, P_{s}\}$. $\mathcal{I}(M(R)) = \{X \subseteq U: |X \bigcap P_{i}| \leq 1, \forall i\in \{1, 2, \cdots, s\}\}$.
Combing Proposition \[baseofM(R)\] with the definition of upper approximation of any subset of ground set, we can obtain the following proposition.
\[anotherformofbaseofM(R)\] Let $R$ be an equivalence relation on $U$. $\mathcal{B}(M(R))= Min (\{X \subseteq U: R^{\ast}(X)$ $=U\})$.
Based on the above three propositions, the following theorem connects the bases of $M(R)$ with the solution space $I_{GF(2)}(B(R))$. It is interesting to find that, over binary field, the collection of bases of the vector matroid induced by a matrix is just the collection of the minimal non-empty sets that are supports of members of $I_{GF(2)}(B(R))$.
\[baseandanotherspace\] Let $R$ be an equivalence relation on $U$ and $B(R)$ a matrix representation of $R$. $\mathcal{B}(M(R))=Min(\{\theta(I_{GF(2)}(B(R)))\})$.
Let $U=\{x_{1}, x_{2}, \cdots, x_{n}\}$ and $U/R=\{P_{1}, P_{2}, \cdots, P_{s}\}(s \leq n)$. we obtain the $B(R)=[\beta_{1}, \beta_{2},$ $ \cdots, \beta_{n}]$ of which the columns are labeled, in order, by $x_{1}, x_{2}, \cdots, x_{n}$. For all $B\in \mathcal{B}(M(R))$, we may as well suppose $B=\{x_{i_{1}}, x_{i_{2}}, \cdots, $ $x_{i_{s}}\}$. According to Proposition \[baseofM(R)\], we know that $x_{i_{j}} \in P_{j}$ for all $j\in \{1, 2, \cdots, s\}$. Let $\mathbf{v}=(v_{1}, v_{2}, \cdots, v_{n})^{T} \in V(n, GF(2))$, where $v_{i}=1$ if and only if $x_{i} \in B$. Then $\theta(\mathbf{v})=B$. According to the definition of $B(R)$, we know $B(R)\mathbf{v}=\mathbf{1}$. Thus $\mathcal{B}(M(R)) \subseteq \{\theta(I_{GF(2)}(B(R)))\}$. Next, we prove the minimality of $B$. If $B \notin Min(\{\theta(I_{GF(2)}(B(R)))\})$, then there exists $B_{1}\in \{\theta(I_{GF(2)}(B(R)))\}$ such that $B_{1} \subset B$. However, for all $x_{i_{j}}\in B$, $B-\{x_{i_{j}}\} \notin \{\theta(I_{GF(2)}(B(R)))\}$ which implies contradiction. Hence, we have $\mathcal{B}(M(R)) \subseteq Min(\{\theta(I_{GF(2)}(B(R)))\})$. Combining with Proposition \[theminimalityofset\], \[anotherspaceandroughset\] and \[anotherformofbaseofM(R)\], we have $\mathcal{B}(M(R))=Min(\{\theta(I_{GF(2)}(B(R)))\})$.
As we know, any independent set of matroid $M(R)$ are those columns of $B(R)$ that are linearly independent as vectors over a field. The following proposition establishes another representation of the independent sets of $M(R)$ by using the solution space $I_{GF(2)}(B(R)))\})$.
Let $R$ be an equivalence relation on $U$ and $B(R)$ a matrix representation of $R$. $\mathcal{I}(M(R))= Low(Min(\{\theta(I_{GF(2)}(B(R)))\}))$.
Characterizing the circuits, the bases and the independent sets of the matroid induced by an equivalence relation by matrix approaches lays the sound foundation for us to study the characteristics of the matroidal structure of rough sets from the viewpoint of matrix.
Matrix null space over binary field to rough sets {#Equivalencerelationinducedbymatrix}
==================================================
In section \[S:Representationofmatroidinducedbyanequivalencerelationoverafiled\], we has obtained a matroidal structure of rough sets from matrix. Over any field, the matroid induced by an equivalence relation is the one induced by a matrix representation of the equivalence relation. Section \[S:Nullspaceapproachtoroughsetthroughmatroid\] has studied certain characteristics of the matroid through matrix solution space such as null space. In this section, we study how to construct an equivalence relation from matrix null space over binary field, and establish an isomorphism from a family of equivalence relations to a family of sets which any member is a collection of the minimal non-empty sets that are supports of members of null space of a binary dependence matrix.
In the following discussion, for any $m \times n$ matrix $A$, we suppose the columns of it are, in order, labeled by $x_{1}, x_{2}, \cdots, x_{n}$, and we denote the collection of the column labels as $U$. Over binary field, we define a relation on $U$ by the null space of matrix as follows.
\[relationinducedbymatroid\] Let $F$ be a field and $A$ be an $m \times n$ matrix over $F$. One can define a relation $R_{F}(A)$ on $U$ as follows: for all $x_{i}, x_{j} \in U$,
$(x_{i}, x_{j}) \in R_{F}(A) \Leftrightarrow x_{i}=x_{j}$ or $\mathbf{e}_{i}+\mathbf{e}_{j} \in \mathcal{N}_{F}(A)$,
where $U$ is a collection of column labels of $A$ and $\mathbf{e}_{i}, \mathbf{e}_{j}\in V(n, F)$ satisfy $\theta(\mathbf{e}_{i})= \{x_{i}\}$ and $\theta(\mathbf{e}_{j})=\{x_{j}\}$.
If $F$ is a binary field, then $\forall x_{i}, x_{j}\in U$ and $x_{i}R_{F}(A)x_{j}$ implies $x_{i}=x_{j}$ or the columns of $A$ labeled by $x_{i}$ and $x_{j}$, respectively, are equivalent.
Suppose $$A=\bordermatrix[{[]}]{
& x_{1} & x_{2} & x_{3} \cr
&1 & -1 & 1 \cr
&1 & -1 & 1 \cr
}.$$ According to the above definition, we know $U = \{x_{1}, x_{2}, x_{3}\}$ and $(x,x)\in R$ for all $x \in U$. Since $\mathbf{e}_{1} = (1, 0, 0)^{T}$, $\mathbf{e}_{2}=(0, 1, 0)^{T}$ and $\mathbf{e}_{3}=(0, 0, 1)^{T}$, $\theta(\mathbf{e}_{i})=\{x_{i}\}$ for all $x_{i}\in U$. We can find that $\mathbf{e}_{1} + \mathbf{e}_{2} \in \mathcal{N}_{\mathbf{R}}(A)$ and $\mathbf{e}_{2} + \mathbf{e}_{3} \in \mathcal{N}_{\mathbf{R}}(A)$. Thus we know $(x_{1},x_{2})\in R_{\mathbf{R}}(A), (x_{2}, x_{3})\in R_{\mathbf{R}}(A), (x_{2},x_{1})\in R_{\mathbf{R}}(A)$ and $(x_{3}, x_{2})\in R_{\mathbf{R}}(A)$. Therefore $R_{\mathbf{R}}(A)= \{(x_{1},x_{1}), (x_{2},x_{2}), (x_{3},x_{3}),$ $ (x_{1},x_{2}), (x_{2},x_{1}),$ $ (x_{2},x_{3}), (x_{3},x_{2})\}$. It is clear that $R_{\mathbf{R}}(A)$ is not an equivalence relation on $U$ because $(x_{1},x_{2})\in R_{\mathbf{R}}(A)$ and $(x_{2},x_{3})\in R_{\mathbf{R}}(A)$ but $(x_{1},x_{3})\notin R_{\mathbf{R}}(A)$. Similarly, if $F$ is binary field, then $R_{GF(2)}(A)=\{(x_{1},x_{1}), (x_{2},x_{2}), (x_{3},x_{3}), (x_{1},x_{3}), (x_{3},x_{1})\}$. It is clear that $R_{GF(2)}(A)$ is an equivalence relation on $U$.
Form above example, we find that, over a field, the relation defined in Definition \[relationinducedbymatroid\] may not be an equivalence relation. However, it inspires us to consider whether the relation is an equivalence relation over binary field or not. For convenience, we take $R(A)$ instead of $R_{GF(2)}(A)$ in the following section.
\[equivalencerelationoverbinaryfield\] Let $A=(a_{ij})_{m \times n}=[\mathbf{\alpha}_{1}, \mathbf{\alpha}_{2}, \cdots,$ $ \mathbf{\alpha}_{n}]$ be a matrix over binary field and $U$ the collection of the column labels of $A$. $R(A)$ is an equivalence relation on $U$.
The reflexivity and symmetry of $R(A)$ are obvious. Now we prove the transitivity of $R(M)$. $\forall x_{i}, x_{j}, x_{k} \in U$, there exist identity vectors $\mathbf{e}_{i}, \mathbf{e}_{j}, \mathbf{e}_{k} \in V(n, GF(2))$ such that $\theta(\mathbf{e}_{i})=\{x_{i}\}$, $\theta(\mathbf{e}_{j})=\{x_{j}\}$ and $\theta(\mathbf{e}_{k})=\{x_{k}\}$. $x_{i}Rx_{j}$ and $x_{j}Rx_{k}$, if $x_{i}=x_{j}$ and $x_{j}=x_{k}$, then we obtain the result. If $x_{i}=x_{j}$ and $A(\mathbf{e}_{j} + \mathbf{e}_{k})=0$, then $\mathbf{e}_{i}=\mathbf{e}_{j}$ and $A(\mathbf{e}_{j} + \mathbf{e}_{k})=0$, thus we have $A(\mathbf{e}_{i}+\mathbf{e}_{k})=0$, hence we obtain the result. If $A(\mathbf{e}_{i} + \mathbf{e}_{j})=0$ and $A(\mathbf{e}_{j} + \mathbf{e}_{k})=0$, then $\mathbf{0}=\mathbf{0}-\mathbf{0}=A(\mathbf{e}_{i} + \mathbf{e}_{j})-
A(\mathbf{e}_{j} + \mathbf{e}_{k})=A\mathbf{e}_{i} + A\mathbf{e}_{j}-A\mathbf{e}_{j} - A\mathbf{e}_{k}=A\mathbf{e}_{i} - A\mathbf{e}_{k}=
A\mathbf{e}_{i} + A(-\mathbf{e}_{k})=A\mathbf{e}_{i} + A\mathbf{e}_{k}= A(\mathbf{e}_{i} + \mathbf{e}_{k})$, then $\mathbf{e}_{i} + \mathbf{e}_{k} \in \mathcal{N}_{GF(2)}(A)$, thus we obtain the result that if $x_{i} R x_{j}$ and $x_{j} R x_{k}$, then $x_{i} R x_{j}$. Therefore, $R(A)$ is an equivalence relation on $U$.
As we know, over a field, any two matrices which has the same number of columns may generate the same vector matroid. However, the following proposition indicates that the equivalence relation induced by these two matrices are the same over binary field.
\[therelationbetweentwoequivalencerelationinducedbytworepresentablematrix\] Let $A_{1}$ and $A_{2}$ be two matrices which have not zero columns over binary field. If $M_{GF(2)}[A_{1}]=M_{GF(2)}[A_{2}]$, then $R(A_{1})=R(A_{2})$.
Suppose the columns of $A_{1}=[\mathbf{\alpha}_{1}, \mathbf{\alpha}_{2}, \cdots,$ $ \mathbf{\alpha}_{n}]$ are, in order, labeled by $x_{1}, x_{2},$ $\cdots, x_{n}$, so does $A_{2}=[\mathbf{\alpha}^{'}_{1}, \mathbf{\alpha}^{'}_{2}, \cdots, \mathbf{\alpha}^{'}_{n}]$. We know $U=\{x_{1}, x_{2}, \cdots, x_{n}\}$. First, we prove $R(A_{1}) \subseteq R(A_{2})$. For all $(x_{i},x_{j})\in R(A_{1})$, then $x_{i}=x_{j}$ or $\mathbf{e}_{i}+\mathbf{e}_{j}\in \mathcal{N}_{GF(2)}(A_{1})$, where $\mathbf{e}_{i}, \mathbf{e}_{j} \in V(n, GF(2))$ satify $\theta(\mathbf{e}_{i})=\{x_{i}\}$ and $\theta(\mathbf{e}_{j})= \{x_{j}\}$. If $x_{i} = x_{j}$, then we obtain the result. If $\mathbf{e}_{i}+\mathbf{e}_{j}\in \mathcal{N}_{GF(2)}(A_{1})$, then $\theta(\mathbf{e}_{i}+\mathbf{e}_{j})=\{x_{i}, x_{j}\} \in \mathcal{D}(M_{GF(2)}[A_{1}])=\mathcal{D}(M_{GF(2)}[A_{2}])$ according to Proposition \[P:nullspaceanddependentset\]. Thus the columns of $A_{2}$ labeled by $x_{i}$ and $x_{j}$, respectively, are linearly dependent in $V(n, GF(2))$. Since $A_{2}$ has not zero columns, $\mathbf{\alpha}^{'}_{i} + \mathbf{\alpha}^{'}_{j} = \mathbf{0}$, that is, $A_{2}(\mathbf{e}_{i}+\mathbf{e}_{j}) = \mathbf{0}$. Thus $\mathbf{e}_{i}+\mathbf{e}_{j} \in \mathcal{N}_{GF(2)}(A_{2})$ which implies $(x_{i}, x_{j})\in R(A_{2})$. Hence $R(A_{1})\subseteq R(A_{2})$. Similarly, we can prove $R(A_{2})\subseteq R(A_{1})$.
Over binary field, for an equivalence relation, one can obtain a matrix through Theorem \[T: circuitsbetweenmatroidandvectormatroid\], and then obtain the other equivalence relation through Definition \[relationinducedbymatroid\]. What about the relationship between these two equivalence relations?
\[relationbetweenRandA\] Let $R$ be an equivalence relation on $U$. If $M(R)=M_{GF(2)}[A(R)]$, then $R(A(R))=R$.
We may as well suppose the columns of $A(R)$ are, in order, labeled by $x_{1}, x_{2}, $ $\cdots, x_{n}$. Thus $U=\{x_{1}, x_{2}, \cdots, x_{n}\}$. Since for all $i\in \{1, 2, \cdots, n\}$, $\{x_{i}\}$ is an independent set of $M(R)$. Thus $A(R)$ dose not contain zero columns. For all $(x_{i}, x_{j})\in R(A(R))$, then $x_{i}=x_{j}$ or $\mathbf{e}_{i}+\mathbf{e}_{j}\in \mathcal{N}_{GF(2)}(A(R))$, where $\mathbf{e}_{i}, \mathbf{e}_{j} \in V(n, GF(2))$ satisfy $\theta(\mathbf{e}_{i})=\{x_{i}\}$ and $\theta(\mathbf{e}_{j})=\{x_{j}\}$. If $x_{i}=x_{j}$, then $(x_{i}, x_{j})\in R$ for $R$ is an equivalence relation. If $x_{i} \neq x_{j}$, then $\mathbf{e}_{i}+\mathbf{e}_{j} \in \mathcal{N}_{GF(2)}(A(R))$. According to Proposition \[P:nullspaceanddependentset\], then $\theta(\mathbf{e}_{i}+\mathbf{e}_{j})=\{x_{i},x_{j}\}\in \mathcal{D}(M(R))$. But $\{x_{i}\}$ or $\{x_{j}\}$ is an independent set of $M(R)$. Thus $\{x_{i}, x_{j}\} \in \mathcal{C}(M(R))$, that is, there exists $P\in U/R$ such that $\{x_{i}, x_{j}\} \in P$ which implies $(x_{i}, x_{j})\in R$. Conversely, $\forall (x_{i}, x_{j})\in R$, then there exists $P\in U/R$ such that $\{x_{i}, x_{j}\} \in P$, that is, $\{x_{i},x_{j}\}\in \mathcal{C}(M(R))$. According to Theorem \[circuitandresolutionspace\], then $\mathbf{e}_{i} + \mathbf{e}_{j}\in \mathcal{N}_{GF(2)}(A(R))$, that is, $(x_{i}, x_{j}) \in R(A(R))$. Thus $R(A(R))=R$.
Since $B(R)$ is a matrix which can generate matroid $M(R)$ over binary field, we can obtain the following corollary.
Let $R$ be an equivalence relation on $U$ and $B(R)$ a matrix representation of $R$. $R(B(R))=R$.
Next, we define a special type of matrix. Over binary field, the null space of this type of matrix has close relation with equivalence relation.
\[specialmatrix\] Let $F$ be a field and $A=(a_{ij})_{m \times n}=[\mathbf{\alpha}_{1}, \mathbf{\alpha}_{2}, \cdots, \mathbf{\alpha}_{n}]$ a matrix over $F$. If $A$ satisfies the following conditions:\
(1) for all $i \in \{1, 2, \cdots, n\}$, $\mathbf{\alpha}_{i} \neq 0$,\
(2) for all $k \in \{2, \cdots, n\}$, if $r_{F}[\mathbf{\alpha}_{i_{1}}, \mathbf{\alpha}_{i_{2}}, \cdots, \mathbf{\alpha}_{i_{k}}]<k$, then there exists $ \{\mathbf{\alpha}_{i_{p}}, \mathbf{\alpha}_{i_{q}}\} \subseteq
\{\mathbf{\alpha}_{i_{1}}, \mathbf{\alpha}_{i_{2}}, \cdots, \mathbf{\alpha}_{i_{k}}\}$ such that $r_{F}[\mathbf{\alpha}_{i_{p}}, \mathbf{\alpha}_{i_{q}}]< 2$,\
then $A$ is called a binary dependence matrix and we denote the set of this type of matrices as $\mathcal{A}$.
The following proposition shows the relation between the matroid $M(R)$ induced by an equivalence relation and the collection of binary dependence matrices $\mathcal{A}$.
\[A(R)isabinarydependentmatrix\] Let $R$ be an equivalence relation on $U$. If $M(R)=M_{F}[A(R)]$, then $A(R)\in \mathcal{A}$.
Suppose $U=\{x_{1}, x_{2}, \cdots, x_{n}\}$, $U/R=\{P_{1}, P_{2}, \cdots, P_{s}\}$ and the columns of $A(R)=[\mathbf{\alpha}_{1}, \mathbf{\alpha}_{2}, \cdots, \mathbf{\alpha}_{n}]$ are labeled, in order, by $x_{1}, x_{2}, \cdots, x_{n}$. According to Proposition \[matroidinducedbyenquivalencerelation\], we know that $\mathbf{\alpha}_{i} \neq \mathbf{0}$ for all $i \in \{1, 2, \cdots, n \}$; otherwise, $M(R)$ has single-point sets as its circuits which implies contradictory. Since $M(R)=M_{F}[A(R)]$, for all $k \geq 2$, if $\mathbf{\beta}_{i_{1}}, \mathbf{\beta}_{i_{2}}, \cdots, $ $\mathbf{\beta}_{i_{k}}$ are linearly dependent over $F$, then $\{x_{i_{1}}, x_{i_{2}}, \cdots, x_{i_{k}}\} \in D(M(R))$. Thus there exists $C \in C(M(R))$ such that $C \subseteq \{x_{i_{1}}, x_{i_{2}}, \cdots, x_{i_{k}}\}$. According to Proposition \[matroidinducedbyenquivalencerelation\], we may as well suppose $C=\{x_{i_{p}}, x_{i_{q}}\}$. Then the columns labeled by the elements of $C$ are linearly dependent over $F$, i.e., $\mathbf{\beta}_{i_{p}},\mathbf{\beta}_{i_{q}}$ are linearly dependent over $F$. Hence $A(R)\in \mathcal{A}$.
Over a field, matrix $B(R)$ can induce a matroid and the matroid is $M(R)$. Based on the above proposition, we can obtain the following result.
Let $R$ be an equivalence relation on $U$ and $B(R)$ a matrix representation of $R$. $B(R) \in \mathcal{A}$.
If we first convert a binary dependence matrix into an equivalence relation, then covert the equivalence relation into a matrix. The matroid induced by the second conversion over binary field is the one induced by the first conversion.
\[relationbetweenA(R(A))andA\] Let $A \in \mathcal{A}$. $M_{GF(2)}[A(R(A))] = M_{GF(2)}[A]$.
Suppose the columns of $A$ are labeled, in order, by $x_{1}, x_{2}, \cdots, x_{n}$, so dose $A(R(A))$. Then $U=\{x_{1}, x_{2}, \cdots, x_{n}\}$. For all $C \in \mathcal{C}(M_{GF(2)}[A(R(A))])$, according to Proposition \[matroidinducedbyenquivalencerelation\], we may as well suppose $C=\{x_{i}, x_{j}\}$ and there exists $P_{i} \in U/R(A)$ such that $\{x_{i}, x_{j}\}\in P_{i}$, that is, $(x_{i}, x_{j}) \in R(A)$. Then for $\mathbf{e}_{i}, \mathbf{e}_{j}\in V(n,GF(2))$ satisfying $\theta(\mathbf{e}_{i})=\{x_{i}\}$ and $\theta(\mathbf{e}_{j})=\{x_{j}\}$, we have $A(\mathbf{e}_{i} + \mathbf{e}_{j})=0$, that is, the columns of $A$ labeled by $x_{i}$ and $x_{j}$ are linearly dependent in $V(n, GF(2))$. Since $A \in \mathcal{A}$, $A$ dose not contain zero columns, thus the column of $A$ labeled by $x_{i}$ or $x_{j}$ is linearly independent in $V(n, GF(2))$. Hence $C\in \mathcal{C}(M_{GF(2)}[A])$, that is, $\mathcal{C}(M_{GF(2)}[A(R(A))]) \subseteq \mathcal{C}(M_{GF(2)}[A])$. Conversely, $\forall C\in \mathcal{C}(M_{GF(2)}[A])$, the columns of $A$ labeled by the elements of $C$ are linearly dependent in $V(n, GF(2))$. Since $A \in \mathcal{A}$, then there exists $C_{1}$ which has only two elements such that $C_{1} \subseteq C$. We may as well suppose $C_{1}=\{x_{i}, x_{j}\}$. Since $A \in \mathcal{A}$, the column of $A$ labeled by $x_{i}$ or $x_{j}$ is linearly independent in $V(n, GF(2))$. Thus $C_{1} \in \mathcal{C}(M_{GF(2)}[A])$. Based on circuit axiom and Theorem \[circuitandresolutionspace\], we can obtain $C=C_{1}=\{x_{1}, x_{j}\} \in Min\{\theta(\mathcal{N}_{GF(2)}(A))-\emptyset\}$, that is, $\mathbf{e}_{i} + \mathbf{e}_{j} \in \mathcal{N}_{GF(2)}(A)$. Hence, $(x_{i}, x_{j}) \in R(A)$, that is, there exists $P_{i}\in U/R(A)$ such that $\{x_{i}, x_{j}\} \subseteq P_{i}$, thus $C=\{x_{i}, x_{j}\} \in \mathcal{C}(M_{GF(2)}[A(R(A))])$. Therefore, we obtain $\mathcal{C}(M_{GF(2)}[A]) \subseteq \mathcal{C}(M_{GF(2)}[A(R(A))])$, that is, $M_{GF(2)}[A(R(A))] = M_{GF(2)}[A]$.
The following result is the combination of Theorem \[circuitandresolutionspace\] and Proposition \[relationbetweenA(R(A))andA\].
\[therelationbetweenA(R(A)andA\] Let $A \in \mathcal{A}$. $Min (\theta(\mathcal{N}_{GF(2)}(A(R(A))))-\emptyset) = Min (\theta(\mathcal{N}_{GF(2)}(A))-\emptyset)$.
Suppose $NS = \{Min (\theta(\mathcal{N}_{GF(2)}(A))-\emptyset): A \in \mathcal{A}\}$ and $\mathbb{R} = \{R: R$ is an equivalence relation on $U\}$. Proposition \[relationbetweenRandA\], \[A(R)isabinarydependentmatrix\] and \[therelationbetweenA(R(A)andA\] indicate that there is a one-to-one correspondence between $\mathbb{R}$ and $NS$. The following theorem shows a deeper relation between them. In fact, $(\mathbb{R}, \bigcap)$ and $(NS, \bigcap)$ are algebra isomorphism.
$(\mathbb{R}, \bigcap) \cong (NS, \bigcap)$.
For any equivalence relation, we can obtain a matrix $A(R)$ through Theorem \[T: circuitsbetweenmatroidandvectormatroid\]. Based on Proposition \[A(R)isabinarydependentmatrix\], we know $A(R)$ is a binary dependence matrix, thus we define an operator $f:\mathbb{R}\rightarrow NS$ as follows: $f(R)=Min(\theta(\mathcal{N}_{GF(2)}(A(R))-\emptyset)$. First we need to prove $f$ is bijection. According to Theorem \[circuitandresolutionspace\], we know $Min (\theta(\mathcal{N}_{GF(2)}(\mathcal{A}))-\emptyset) = C(R)$. According to Proposition \[matroidinducedbyenquivalencerelation\], we know $f$ is injection. $\forall Min(\theta(\mathcal{N}_{GF(2)}(A)-\emptyset) \in NS$, let $R=R(A)$. According to Proposition \[therelationbetweenA(R(A)andA\], $f(R(A))=Min(\theta(\mathcal{N}_{GF(2)}(A)$ $-\emptyset)$, that is, $f$ is a surjection. In the following, we need to prove $f(R_{1} \bigcap R_{2}) = f(R_{1}) \bigcap f(R_{2})$, that is, $Min(\theta(\mathcal{N}_{GF(2)}(A(R_{1} \bigcap R_{2}))-\emptyset) = Min (\theta$ $(\mathcal{N}_{GF(2)}(A(R_{1}))$ $-\emptyset) \bigcap Min(\theta(\mathcal{N}_{GF(2)}(A(R_{2}))-\emptyset)$. For all $\{x_{i}, x_{j}\} \in Min(\theta(\mathcal{N}_{GF(2)}(A(R_{1} \bigcap R_{2}))$ $-\emptyset)$, then $\{x_{i}, x_{j}\} \in \mathcal{C}(R_{1}\bigcap R_{2})$, i.e., there exists $P_{k} \in U/(R_{1}\bigcap R_{2})$ such that $\{x_{i}, x_{j}\}\in P_{k}$. Thus there exist $P^{1}_{s}\in U/R_{1}$ and $P^{2}_{t}\in U/R_{2}$ such that $\{x_{i}, x_{j}\}\in P^{1}_{s}$ and $\{x_{i}, x_{j}\}\in P^{2}_{t}$. Hence, $\{x_{i}, x_{j}\}\in \mathcal{C}(R_{1}) \bigcap \mathcal{C}(R_{2})= Min(\theta(\mathcal{N}_{GF(2)}(A(R_{1}))-\emptyset) \bigcap Min(\theta(\mathcal{N}_{GF(2)}(A(R_{2})-\emptyset)$. Therefore, $Min(\theta(\mathcal{N}_{GF(2)}(A(R_{1} \bigcap R_{2}))-\emptyset) \subseteq Min(\theta(\mathcal{N}_{GF(2)}(A(R_{1}))-\emptyset) \bigcap Min(\theta(\mathcal{N}_{GF(2)}$ $(A(R_{2}))-\emptyset)$. For all $\{x_{i}, x_{j}\} \in Min(\theta(\mathcal{N}_{GF(2)}(A(R_{1}))-\emptyset) \bigcap Min(\theta(\mathcal{N}_{GF(2)}(A(R_{2})-\emptyset)$, that is, $\{x_{i}, x_{j}\} \in \mathcal{C}(R_{1})$ $ \bigcap \mathcal{C}(R_{2})$, then there exist $P^{1}_{s}\in U/R_{1}$ and $P^{2}_{t}\in U/R_{2}$ such that $\{x_{i}, x_{j}\}\in P^{1}_{s}$ and $\{x_{i}, x_{j}\}\in P^{2}_{t}$, i.e., $(x_{i}, x_{j})\in R_{1}\bigcap R_{2}$. Hence there exists $P_{k}\in U/(R_{1}\bigcap R_{2})$ such that $\{x_{i}, x_{j}\}\in P_{k}$, that is, $\{x_{i}, x_{j}\}\in \mathcal{C}(R_{1}\bigcap R_{2})$. Therefore, $Min(\theta(\mathcal{N}_{GF(2)}(A(R_{1}))$ $-\emptyset) \bigcap Min(\theta(\mathcal{N}_{GF(2)}(A(R_{2})-\emptyset) \subseteq Min(\theta(\mathcal{N}_{GF(2)}(A(R_{1} \bigcap R_{2}))-\emptyset) $. In a word, $(\mathbb{R}, \bigcap) \cong (Min (\theta(\mathcal{N}_{GF(2)}(\mathcal{A}))$ $-\emptyset), \bigcap)$.
Isomorphisms are studied in mathematics in order to extend insights from one phenomenon to others. If two algebra systems are isomorphic, then they can be regarded as similarity. Hence, the study of the rough sets is equal to the study of the family of sets which any member is a collection of the minimal non-empty sets that are supports of members of null space of a binary dependence matrix.
Conclusions {#S:Conclusions}
===========
In this paper, we employed the matrix approach to study rough sets over a field. A matroidal structure of rough sets was constructed through matrix, and the matrix approaches such as null space were employed to study the characteristics of the matroidal structure. We also found that a family of equivalence relations and a family of sets, which any member is a collection of the minimal non-empty sets that are supports of members of null space of a binary dependence matrix, are algebra isomorphism. Though some works have been studied in this paper, there are also many interesting topics deserving further investigation. In the future, we will study rough sets from the following two aspects. On one hand, nullity which an important concept in matroid theory will be introduced to study rough sets. On the other hand, matrix will be promoted to study covering-based rough sets and relation-based rough sets.
Acknowledgments
===============
This work is supported in part by the National Natural Science Foundation of China under Grant No. 61170128, the Natural Science Foundation of Fujian Province, China, under Grant Nos. 2011J01374 and 2012J01294, and the Science and Technology Key Project of Fujian Province, China, under Grant No. 2012H0043.
[10]{}
Baesens, B., Setiono, R., Mues, C., Vanthienen, J.: Using neural network rule extraction and decision tables for credit-risk evaluation. Management Science March **49** (2003) 312–329
Barnabei, M., Nicoletti, G., Pezzoli, L.: Matroids on partially ordered sets. Advances in Applied Mathematics **21** (1998) 78–112
Cruz-Cano, R., Lee, M.L.T., Leung, M.Y.: Logic minimization and rule extraction for identification of functional sites in molecular sequences. BioData Mining **5** (2012)
Dash, M., Liu, H.: Consistency-based search in feature selection. Artificial Intelligence **151** (2003) 155–176
Du, Y., Hu, Q., Zhu, P., Ma, P.: Rule learning for classification based on neighborhood covering reduction. Information Sciences **181** (2011) 5457–5467
Edmonds, J.: Matroids and the greedy algorithm. Mathematical Programming **1** (1971) 127–136
He, Q., Wu, C., Chen, D., Zhao, S.: Fuzzy rough set based attribute reduction for information systems with fuzzy decisions. Knowledge-Based Systems **24** (2011) 689–696
Hu, Q., Yu, D., Liu, J., Wu, C.: Neighborhood rough set based heterogeneous feature subset selection. Information Sciences **178** (2008) 3577–3594
Huang, A., Zhu, W.: Geometric lattice structure of covering-based rough sets through matroids. Journal of Applied Mathematics **2012**(2012), Article ID 236307, 25 pages
Huang, A., Zhu, W.: On matrix representation of three types of covering-based rough sets. In: Granular Computing. (2012) 215–220
Lai, H.: Matroid theory. Higher Education Press, Beijing (2001)
Lawler, E.: Combinatorial optimization: networks and matroids. Dover Publications (2001)
Lay, D.C.: Linear Algebra and Its Applications. Publishing House of Electronics Industry, Beijing (2010)
Li, X., Liu, S.: Matroidal approaches to rough set theory via closure operators. International Journal of Approximate Reasoning **53** (2012) 513–527
Liu, G.: Closures and topological closures in quasi-discrete closure. Applied Mathematics Letters **23** (2010) 772–776
Liu, G.: The transitive closures of matrices over distributive lattices. In: Granular Computing. (2006) 63–66
Liu, G.: Rough set theory based on two universal sets and its applications. Knowledge-Based Systems **23** (2010) 110–115
Liu, G., Zhu, W.: The algebraic structures of generalized rough set theory. Information Sciences **178** (2008) 4105–4113
Min, F., Zhu, W.: Attribute reduction of data with error ranges and test costs. Information Sciences **211** (2012) 48–67
Min, F., He, H., Qian, Y., Zhu, W.: Test-cost-sensitive attribute reduction. Information Sciences **181** (2011) 4928–4942
Oxley, J.G.: Matroid theory. Oxford University Press, New York (1993)
Pawlak, Z.: Rough sets. International Journal of Computer and Information Sciences **11** (1982) 341–356
Skowron, A., Stepaniuk, J.: Tolerance approximation spaces. Fundamenta Informaticae **27** (1996) 245–253
Slowinski, R., Vanderpooten, D.: A generalized definition of rough approximations based on similarity. IEEE Transactions on Knowledge and Data Engineering **12** (2000) 331–336
Tseng, T.L.B., Huang, C.C.: Rough set-based approach to feature selection in customer relationship management. Omega **35** (2007) 365–383
Wang, X., Tsang, E.C., Zhao, S., Chen, D., Yeung, D.S.: Learning fuzzy rules from fuzzy samples based on rough set technique. Information Sciences **177** (2007) 4493–4514
Wang, S., Zhu, W.: Matroidal structure of covering-based rough sets through the upper approximation number. International Journal of Granular Computing, Rough Sets and Intelligent Systems **2** (2011) 141–148
Wang, S., Zhu, Q., Zhu, W., Min, F.: Matroidal structure of rough sets and its characterization to attribute reduction. Knowledge-Based Systems **36** (2012) 155–161
Wang, S., Zhu, Q., Zhu, W., Min, F.: Quantitative analysis for covering-based rough sets through the upper approximation number. Information Sciences **220** (2013) 483–491
Xiong, Q., ed.: Modern Algebra. Wuhan University Press (1984)
Yao, Y., Zhao, Y.: Attribute reduction in decision-theoretic rough set models. Information Sciences **178** (2008) 3356–3373
Zhu, W., Wang, S.: Matroidal approaches to generalized rough sets based on relations. International Journal of Machine Learning and Cybernetics **2** (2011) 273–279
Zhu, W., Wang, F.: A new type of covering rough sets. In: IEEE International Conference on Intelligent Systems 2006, London, 4-6 September. (2006) 444–449
Zhu, W., Wang, F.: On three types of covering rough sets. IEEE Transactions on Knowledge and Data Engineering **19** (2007) 1131–1144
Zhu, W.: Topological approaches to covering rough sets. Information Sciences **177** (2007) 1499–1508
Zhu, W., Wang, F.: Reduction and axiomization of covering generalized rough sets. Information Sciences **152** (2003) 217–230
Zhu, W.: Relationship between generalized rough sets based on binary relation and covering. Information Sciences **179** (2009) 210–225
Zhu, W., Wang, S.: Rough matroid based on relation. to appear in Information Sciences. 2013
[^1]: Corresponding author. E-mail: [email protected] (William Zhu)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Let $s,t,m,n$ be positive integers such that $sm=tn$. Let $M(m,s;n,t)$ be the number of $m\times n$ matrices over $\{0,1,2,\dots\}$ with each row summing to $s$ and each column summing to $t$. Equivalently, $M(m,s;n,t)$ counts 2-way contingency tables of order $m\times n$ such that the row marginal sums are all $s$ and the column marginal sums are all $t$. A third equivalent description is that $M(m,s;n,t)$ is the number of semiregular labelled bipartite multigraphs with $m$ vertices of degree $s$ and $n$ vertices of degree $t$. When $m=n$ and $s=t$ such matrices are also referred to as $n \times n$ magic or semimagic squares with line sums equal to $t$. We prove a precise asymptotic formula for $M(m,s;n,t)$ which is valid over a range of $(m,s;n,t)$ in which $m,n\rightarrow\infty$ while remaining approximately equal and the average entry is not too small. This range includes the case where $m/n$, $n/m$, $s/n$ and $t/m$ are bounded from below.'
author:
- |
E. Rodney Canfield\
Department of Computer Science\
University of Georgia\
Athens, GA 30602, USA\
`[email protected]`
- |
Brendan D. McKay\
Department of Computer Science\
Australian National University\
Canberra ACT 0200, Australia\
`[email protected]`
date: ': Mar 13, 2009'
title: |
Asymptotic Enumeration of Integer Matrices\
with Constant Row and Column Sums
---
Introduction {#section:intro}
============
Let $m,s,n,t$ be positive integers such that $ms=nt$. Let $M(m,s;n,t)$ be the number of $m\times n$ matrices over $\{0,1,2,\dots\}$ with each row summing to $s$ and each column summing to $t$. Figure \[figure\] shows an example.
---- ---- ---- ---- ----
5 4 0 11 20
2 8 9 1 20
8 3 6 3 20
15 15 15 15
---- ---- ---- ---- ----
The matrices counted by $M(m,s;n,t)$ arise frequently in many areas of mathematics, for example enumeration of permutations with respect to descents, symmetric function theory, and statistics. The last field in particular has an extensive literature in which such matrices are studied as *contingency tables* or *frequency tables*; see [@gail] as an example of enumerative work, [@Greselin] for a lengthy bibliography, and [@DG] for a survey of applications. The matrices counted by $M(m,s;n,t)$ have also regularly been the topic of papers whose primary focus is algorithmic, both with respect to the problem of generating contingency tables with prescribed margins at random, and with respect to approximate counting. For random generation see, for example [@DKM; @morris]; for approximate counting see the recent [@bsy]. Other important recent studies of both random sampling and approximate counting include [@CDHL; @HJ].
A historically early study of integer matrices with specified row and column sums appeared in MacMahon [@macmahon], where the numbers $M(m,s;n,t)$ are identified as coefficients when certain functions are expanded in terms of standard bases of symmetric functions. In a later paper [@macmahon2], MacMahon studied “general magic squares”—$n\times n$ integer matrices with all row and column sums equal to a prescribed value $t$. Stanley [@stanleyA Chap.4] refers to these as magic squares with line sums equal to $t$, and Stanley’s [@stanley Chap.1] provides further history of this topic. For fixed $n$, $M(n,t;n,t)$ is a polynomial in $t$ known as the *Ehrhart polynomial* of the Birkhoff polytope. This has been computed in closed form for $n\le 9$ [@BP2003]. (The Birkhoff polytope is the set of all doubly stochastic real matrices.) The results in the present paper enable us to give the first asymptotic formula for the volume of this famous polytope [@CMComing].
Our focus in this paper is the asymptotic value of $M(m,s;n,t)$. The first significant result on asymptotics was that of Read [@read], who obtained the asymptotic behavior for $s=t=3$. In the four year period from 1971 to 1974 three published papers [@bbk; @bender; @everett] gave the asymptotic value for the case $s,t$ bounded. To our knowledge, this was the extent of published, rigorously proven formulas until the present work. In a paper currently in preparation [@GMsparse], Greenhill and McKay obtain a formula valid in the sparse range $st=o\((mn)^{1/2}\)$.
Define the [*density*]{} $\l=s/n=t/m$ to be the average entry in the matrix. As early as [@good50 p.100], Good implicitly gave the estimate $M(m,s;n,t) \approx G(m,s;n,t)$, where$$G(m,s;n,t) =
\frac{\displaystyle\binom{n+s-1}{s}^{\!m}\binom{m+t-1}{t}^{\!n}}
{\displaystyle\binom{mn+\l mn-1}{\l mn}}\,.$$ Good spelt this out explicitly in [@good76] and again with Crook in [@good]. In the first paper, Good gave a heuristic argument for this approximation based on steepest-descent, “leaving aside finer points of rigor.” His calculation also yielded another approximation [@good76 Eqn.B2.22] that is very similar when $m\approx n$ but larger otherwise. We will show in this paper that $G(m,s;n,t)$ is a remarkably accurate approximation of $M(m,s;n,t)$, being out by a constant factor over a wide range and perhaps always. Another important estimate of $M(m,s;n,t)$ was developed by Diaconis and Efron [@DE]. It aims for accuracy when the sum of the matrix entries $\lambda mn$ is very large, and is quite poor when that condition is not met.
Our main result is as follows.
\[thm:main\] Let $s=s(m,n),t=t(m,n)$ be positive integers satisfying $ms=nt$. Define $\l = s/n = t/m$. Let $a,b>0$ be constants such that $a+b<\frac12$. Suppose that $m,n \rightarrow \infty$ in such a way that $$\label{Hyp}
\frac{(1+2\l)^2}{4\l(1+\l)}
\biggl( 1 + \frac{5m}{6n} + \frac{5n}{6m} \biggr)
\le a \log n.$$ Then $$\begin{aligned}
M(m,s;n,t) &=
G(m,s;n,t) \exp\(\tfrac12 + O(n^{-b})\)\label{mainformula} \\
&= \frac{\(\l^{-\l}(1+\l)^{1+\l}\)^{mn}}
{(4\pi A)^{(m+n-1)/2} m^{(n-1)/2} n^{(m-1)/2}}\notag\\
&{\kern 10mm}\times
\exp\biggl(\frac12 - \frac{1+2A}{24A}
\Bigl(\frac mn + \frac nm\Bigr) + O(n^{-b})\biggl).
\label{mainformula2}\end{aligned}$$
\[biglambda\] Under the conditions of Theorem \[thm:main\], if in addition $mn/\l^2\to 0$, $$M(m,s;n,t) = \(\l+\tfrac12\)^{(m-1)(n-1)}
\frac{(mn)!}{m!^n n!^m} \exp\(\tfrac12+O(mn/\l^2+n^{-b})\).$$
\[RueDeRemarques\] The following is inspired by a similar observation made by Good [@good76]. The number of $k$-tuples of nonnegative integers $n_i$ satisfying $n_1+\cdots+n_k=K$ is well known to be $\binom{K+k-1}{k-1}=\binom{K+k-1}{K}$. This allows an instructive interpretation of as $M(m,s;n,t) = NP_1P_2E$, where $$\begin{aligned}
N &= \binom{mn+\l mn-1}{\l mn},~
P_1 = N^{-1}\binom{n+s-1}{s}^{\!m}\negthickspace,~
P_2 = N^{-1}\binom{m+t-1}{t}^{\!n}\negthickspace,\\[0.6ex]
E &= E(m,s;n,t) = \exp\(\tfrac12 + O(n^{-b})\).\end{aligned}$$ Clearly $N$ is the number of tables whose entries sum to $\l mn = sm = tn$. In the uniform probability space on these $N$ tables, $P_1$ is the probability of the event that all the row sums are equal to $s$, and $P_2$ is the probability of the event that all the column sums are equal to $t$. The final quantity $E$ is thus a correction to account for the non-independence of these two events.
In an earlier paper [@CM] we showed how $B(m,s;n,t)$, the number of such matrices whose entries are taken from $\{0,1\}$, can be computed exactly for small $m,n$. That algorithm is easily adaptable for efficient computation of exact values of $M(m,s;n,t)$, and the values so computed suggest the following conjecture.
\[conjecture\] Consider a 4-tuple of positive integers $m,s,n,t$ such that $ms=nt$. Define $\Delta(m,s;n,t)$ by $$\label{conjformula}
\begin{split}
M(m,s;n,t) &= G(m,s;n,t)\,
\Bigl(\frac{m+1}{m}\Bigr)^{(m-1)/2}
\Bigl(\frac{n+1}{n}\Bigr)^{(n-1)/2} \\
&{\kern 40mm}\times
\exp\Bigl(-\dfrac12 + \frac{\Delta(m,s;n,t)}{m+n}\Bigr).
\end{split}$$ Then $0<\Delta(m,s;n,t)<2$.
The factor $(1+1/m)^{(m-1)/2}$, which approaches $e^{1/2}$ as $m\to\infty$, and the similar factor $(1+1/n)^{(n-1)/2}$, appear naturally in the analysis of $M(m,s;n,t)$ when one of $m,n$ goes to $\infty$ much faster than the other. This can be seen in [@gail eqn.(3.5)] and will be extended rigorously in a forthcoming paper [@oblong]. Conjecture \[conjecture\] has been proved in several cases: (a) for $m=n\le 9$, using the exact values from [@BP2003]; (b) for sufficiently large $m,n$ when $st=o\((mn)^{1/5}\)$, using the asymptotics derived in [@GMsparse]; (c) for several thousand values of $(m,s;n,t)$ for $m,n\le 30$. It has also been established to a high degree of confidence for many larger sizes using a simulation similar to the one described in [@CM], which was a variation on a method of Chen, Diaconis, Holmes and Liu [@CDHL].
An interesting point of contrast between the integer and 0-1 cases is suggested by the above. Let $E_0(m,s;n,t)$ be the quantity corresponding to $E(m,s;n,t)$ in Remark \[RueDeRemarques\] when the number of binary tables is decomposed in the same manner (see [@CGM]). It was proved by Ordentlich and Roth [@Ord] that $E_0(m,s;n,t)\le 1$. A corollary of Corollary \[conjecture\] would be that the opposite bound $E(m,s;n,t)\ge 1$ holds in the integer case. (In fact the smallest value we have found for $s,t>0$ is $E(2,3;3,2)=\tfrac{539}{450}$.)
Some numerical examples comparing our estimates to other estimates and the correct values appear in Table \[numeric\].
\#1e\#2 [$#1{\times}10^{#2}$]{}
----------------------------------------------- ---------------------- ------------------- --------------------------- -------------
$m,s,n,t$ $G(m,s;n,t)$ [@good76 (B2.22)] [@DE]
[@GMsparse Cor. 5.1] Conjecture \[conjecture\] Exact
[height2.8ex width0pt depth0pt]{} 3,100,3,100 1.019e7 1.192e7 1.262e7
1.022e7 1.680e7 (1.3160.217)e7 13268976
\[0.7ex\] 3,98,49,6 7.594e67 3.716e68 1.278e68
9.630e67 1.252e68 (1.0170.020)e68 1.01100e68
\[0.7ex\] 3,99,9,33 2.116e21 2.788e21 2.864e21
2.506e21 3.488e21 (2.8440.236)e21 2.79207e21
\[0.7ex\] 10,20,10,20 7.434e58 9.021e58 1.511e59
1.059e59 1.226e59 (1.1190.056)e59 1.09747e59
\[0.7ex\] 18,13,18,13 5.157e127 6.962e127 5.109e130
7.850e127 8.502e127 (8.0650.224)e127 7.94500e127
\[0.7ex\] 30,3,30,3 1.404e92 7.496e92 1.086e138
2.223e92 2.315e92 (2.2420.037)e92 2.22931e92
----------------------------------------------- ---------------------- ------------------- --------------------------- -------------
: Comparison of six estimates for $M(m,s;n,t)$.[]{data-label="numeric"}
Throughout the paper, the asymptotic notation $O(f(m,n))$ refers to the passage of $m$ and $n$ to $\infty$. Generally, the constant implied by the $O$ may depend on $a$, $b$, and the sufficiently small and fixed constant $\eps$ introduced later in the definition (\[RDef\]) of region $\R$. We also use a modified notation $\tO(f(m,n))$. A function $g(m,n)$ belongs to this class provided that for some constant $c$, which is independent of $a$, $b$, and $\eps$, we have $$g(m,n) = O(f(m,n)n^{c\eps}).$$ Under the hypothesis (\[Hyp\]) of Theorem \[thm:main\], we have $\log m\sim\log n$, $n = O(m\log m)$, $m=O(n\log n)$, and $\l^{-1}=O(\log n)$. Consequently, in the newly introduced notation, $n=\tO(m)$, $m=\tO(n)$, and $\l^{-1}
=\tO(1)$. In general, if $c_1,c_2,c_3,c_4$ are constants, then $m^{c_1+c_2\eps}n^{c_3+c_4\eps}=\tO(m^{c_1}n^{c_3})
=\tO(n^{c_1+c_3})=\tO(m^{c_1+c_3})$.
An integral for $M(m,s;n,t)$ {#section:integral}
============================
We express $M(m,s;n,t)$ as an integral in $(m{+}n)$-dimensional complex space then estimate its value by the saddle-point method.
It is clear that $M=M(m,s;n,t)$ is the coefficient of $x_1^s\cdots x_m^s\, y_1^t\cdots y_n^t$ in $$\prod_{j=1}^m \,\prod_{k=1}^n \, \(1 - x_jy_k\)^{\!-1}.$$ Applying Cauchy’s Theorem we have
$$\label{cauchy1}
M = \frac{1}{(2\pi i)^{m+n}} \oint\cdots\oint
\frac{\prod_{j,k}(1-x_j y_k)^{-1}}{
x_1^{s+1}\cdots x_m^{s+1} y_1^{t+1}\cdots y_n^{t+1}}
\, dx_1\cdots dx_m\,dy_1\cdots dy_n,$$
where each contour circles the origin once in the anticlockwise direction.
It will suffice to take the contours to be circles; specifically, we will put $x_j=re^{i\t_j}$ and $y_k=re^{i\p_k}$ for each $j,k$, where $$r = \sqrt{\frac{\l}{1+\l}}\,.$$ This gives $$\label{Idef1}
M = \frac{1}{(2\pi)^{m+n}} \(\l^{-\l}(1+\l)^{1+\l}\)^{mn}\, I(m,n),$$ where $${\label{Idef2}}
I(m,n) = \int_{-\pi}^\pi\cdots \int_{-\pi}^\pi
\frac{\prod_{j,k} \( 1 - \l(e^{i(\t_j+\p_k)}-1)\)^{-1}}{
e^{is\sum_j\t_j + it\sum_k\p_k}}
\, d\bit\, d\bip,$$ where $\bit=(\t_1,\dots,\t_m)$ and $\bip=(\p_1,\dots,\p_n)$. Let $F(\bit,\bip)$ denote the integrand in equation (\[Idef2\]).
Evaluating the integral {#section:evalint}
=======================
This section follows closely the corresponding section in [@CGM]. However, all proofs are intended to be complete, and not require the reader to refer to the latter work, except for the following lemma which we restate here without proof.
\[MW3\] Let $\eps', \eps'', \eps''', \bar\eps,\Delta$ be constants such that $0<\eps'<\eps''<\eps'''$, $\bar\eps\ge 0$, and $0<\Delta<1$. The following is true if $\eps'''$ and $\bar\eps$ are sufficiently small.Let $\mwA=\mwA(N)$ be a real-valued function such that $\mwA(N)=\Omega(N^{-\eps'})$. Let $\mwa_j=\mwa_j(N)$, $\mwB_j=\mwB_j(N)$, $\mwC_{jk}=\mwC_{jk}(N)$, $\mwE_j=\mwE_j(N)$, $\mwF_{jk}=\mwF_{jk}(N)$ and $\mwJ_j=\mwJ_j(N)$ be complex-valued functions $(1 \le j,k \le N)$ such that $\mwB_j,\mwC_{jk},
\mwE_j,\mwF_{jk}=O(N^{\bar\eps})$, $\mwa_j=O(N^{1/2+\bar\eps})$, and $\mwJ_j=O(N^{-1/2+\bar\eps})$, uniformly over $1\le j,k\le N$. Suppose that $$\begin{aligned}
f(\zvec) &= \exp\biggl(
-\mwA N\sum_{j=1}^N z_j^2
+ \sum_{j=1}^{N} \mwa_j z_j^2 + N \sum_{j=1}^{N} \mwB_j z_j^3
+ \sum_{j,k=1}^N \mwC_{jk}z_j z_k^2
\\
&\kern14mm{}
+ N \sum_{j=1}^N \mwE_j z_j^4
+ \sum_{j,k=1}^N \mwF_{jk} z_j^2 z_k^2
+ \sum_{j=1}^{N} \mwJ_j z_j + \delta(\zvec) \biggr)\end{aligned}$$ is integrable for $\zvec=(z_1,z_2,\ldots,z_N)\in U_N$ and $\delta(N)=\max_{\zvec\in U_N} \abs{\delta(\zvec)} = o(1)$, where $$U_N = \bigl\{ \zvec \bigm| \abs{z_j} \le N^{-1/2+\hat\eps}
\text{ for\/ $1\le j\le N$}\bigr\},$$ where $\hat\eps=\hat\eps(N)$ satisfies $\eps''\le2\hat\eps\le\eps'''$. Then, provided the $O(\,)$ term in the following converges to zero, $$\int_{U_N}f(\zvec)\,d\zvec
= \biggl(\frac{\pi}{\mwA N}\biggr)^{\!N/2}
\!\exp\(
\Theta_1+\Theta_2 + O\((N^{-\Delta}+\delta(N)) \mwZ\)
\),$$ where $$\begin{aligned}
\Theta_1 &= \frac{1}{2\mwA N}\sum_{j=1}^{N}\mwa_j
+ \frac{1}{4\mwA^2N^2}\sum_{j=1}^{N}\mwa_j^2
+ \frac{15}{16\mwA^3N}\sum_{j=1}^{N}\mwB_j^2
+ \frac{3}{8\mwA^3N^2}\sum_{j,k=1}^{N}\mwB_j\mwC_{jk} \\
&\quad{} + \frac{1}{16\mwA^3N^3}\sum_{j,k,\ell=1}^N \mwC_{jk}\mwC_{j\ell}
+ \frac{3}{4\mwA^2N}\sum_{j=1}^N \mwE_j
+ \frac{1}{4\mwA^2N^2}\sum_{j,k=1}^N \mwF_{jk}
\displaybreak[0]\\
\Theta_2 &= \frac{1}{6\mwA^3N^3}\sum_{j=1}^N \mwa_j^3
+ \frac{3}{2\mwA^3N^2}\sum_{j=1}^N \mwa_j\mwE_j
+ \frac{45}{16\mwA^4N^2}\sum_{j=1}^N \mwa_j\mwB_j^2 \\
&\quad{} + \frac{1}{4\mwA^3N^3}\sum_{j,k=1}^N (\mwa_j+\mwa_k) \mwF_{jk}
+ \frac{3}{4\mwA^2N}\sum_{j=1}^N \mwB_j\mwJ_j
+ \frac{1}{4\mwA^2N^2}\sum_{j,k=1}^N \mwC_{jk}\mwJ_j \\
&\quad{} + \frac{1}{16\mwA^4N^4}\sum_{j,k,\ell=1}^N
(\mwa_j+2\mwa_k ) \mwC_{jk}\mwC_{j\ell}
+ \frac{3}{8\mwA^4N^3}\sum_{j,k=1}^N
(2\mwa_j+\mwa_k ) \mwB_j\mwC_{jk}
\displaybreak[0]\\
\mwZ &= \exp\biggl(
\frac{1}{4\mwA^2N^2}\sum_{j=1}^{N}\Im(\mwa_j)^2
+ \frac{15}{16\mwA^3N}\sum_{j=1}^{N}\Im(\mwB_j)^2 \\
&\kern7mm{}
+ \frac{3}{8\mwA^3N^2}\sum_{j,k=1}^{N}\Im(\mwB_j)\Im(\mwC_{jk})
+ \frac{1}{16\mwA^3N^3}\sum_{j,k,\ell=1}^N \Im(\mwC_{jk})\Im(\mwC_{j\ell})
\biggr).
\quad\qedsymbol\end{aligned}$$
We use the notation $\R^c$ for the complement of a region $\R$. To evaluate the integral $I(m,n)$ defined in (\[Idef2\]), we proceed as follows: $$I(m,n) = \int_{\R}F + \int_{\R^c}F = \int_{\R'}F + O(1)\int_{\R^c}|F|,
~~~~ \R' \supseteq \R.$$ The region $\R\subseteq [-\pi,+\pi]^{m+n}$ is defined below in (\[RDef\]). The larger containing region $\R'$ is defined in (\[RprimeDef\]). The asymptotic value of the integral $\int_{\R'}F$ is obtained by substituting $\R'$ for the variable $\R''$ in equation (\[IvsJ\]), and applying the main result of this section, Theorem \[Jintegral\]. In the next section we show that for a certain constant $c_5>0$ $$\int_{\R^c}|F| = O(e^{-c_5\min(m^{2\eps},n^{2\eps})/\log n})\int_{\R'}F.$$ (Again, the quantity $\eps$ is a small positive constant arising in the definition (\[RDef\]) of the region $\R$.) This is the complete summary of how we shall evaluate $I(m,n)$, and now we may proceed to the technical details.
For any region $\R''$ we set $$I_{\R''}(m,n) = \int_{\R''}F.$$ In the integral (\[Idef2\]), it is convenient sometimes to think of $\theta_j$, $\phi_k$ as points on the unit circle. We wish to define “averages” of the angles $\theta_j$, $\phi_k$. To do this cleanly we make the following definitions, as in [@CM]. Let $C$ be the ring of real numbers modulo $2\pi$, which we can interpret as points on a circle in the usual way. Let $z$ be the canonical mapping from $C$ to the real interval $(-\pi,\pi]$. An [*open half circle*]{} is $C_t=(t-\pi/2,t+\pi/2)\subseteq C$ for some $t$. Now define $$\Chat^N = \{\bix=(x_1,\dots,x_N) \in C^N | x_1,\dots,x_N \in C_t
{\rm ~for~some~} t \in \Reals\}.$$ If $\bix=(x_1,\dots,x_N) \in C_0^N$ then define $$\xav = z^{-1}\left( \frac{1}{N} \sum_{j=1}^N z(x_j)\right).$$ More generally, if $\bix\in C_t^N$ then define $\xav =
t+\overline{(x_1-t,\dots,x_N-t)}$. The function $\bix\mapsto\xav$ is well-defined and continuous for $\bix\in\Chat^N$.
For some sufficiently small $\eps>0$, let $\R$ denote the set of vector pairs $\bit,\bip\in \Chat^m\times\Chat^n$ such that $$\label{RDef}
\begin{split}
|\tb+\pb| &\le \tbd (mn)^{-1/2+2\eps} \\
{\rm region~} \R: ~~~~~~~~~~~~~~
|\th_j| &\le \tbd n^{-1/2+\eps}, 1 \le j \le m \\
|\ph_k| &\le \tbd m^{-1/2+\eps}, 1 \le k \le n,
\end{split}$$ where $\th_j = \t_j-\tb$ and $\ph_k = \p_k-\pb$. In this definition, values are considered in $C$.
Let $\bithat=(\th_1,\dots,\th_{m-1})$, $\biphat=(\ph_1,\dots,\ph_{n-1})$, and define $T_1$ to be the transformation $T_1(\bithat,\biphat,\mu,\delta)=(\bit,\bip)$ given by $$\mu = \tav + \pav,
\qquad \delta = \tav -
\pav,$$ together with $\that_j=\t_j-\tav$ ($1\leq j\leq m-1$) and $\phat_k=\p_k-\pav$ ($1\leq k\leq n-1$). We also define the 1-many transformation $T_1^*$ by $$T_1^*(\bithat,\biphat,\mu) =
\bigcup_\delta \, T_1(\bithat,\biphat,\mu,\delta).$$
After applying the transformation $T_1$ to $I_{\R}(m,n)$, the new integrand is easily seen to be independent of $\delta$, so we can multiply by the range of $\delta$ and remove it as an independent variable. Therefore, we can continue with an $(m{+}n{-}1)$-dimensional integral over the region $\S$ defined by $\R=T_1^*(\S)$. More generally, if $\S''\subseteq [-\tfrac12\pi,\tfrac12\pi]^{m+n-2}\times[-2\pi,2\pi]$ and $\R''=T_1^*(\S'')$, we have $$\label{IvsJ}
I_{\R''}(m,n) = 2\pi m n \int_{\S''} G(\bithat,\biphat,\mu)
\, d\bithat d\biphat d\mu,$$ where $G(\bithat,\biphat,\mu)
=F\(T_1(\bithat,\biphat,\mu,0)\)$. The factor $2\pi mn$ combines the range of $\delta$, which is $4\pi$, and the Jacobian of $T_1$, which is $mn/2$.
Note that $\S$ is defined by virtually the same inequalities as define $\R$. The first inequality is now $\abs{\mu}\leq \tbd (mn)^{-1/2 + 2\eps}$ and the bounds on $$\that_m = - \sum_{j=1}^{m-1} \that_j~\text{ and }~
\phat_n = - \sum_{k=1}^{n-1} \phat_k$$ still apply even though these are no longer variables of integration.
Our main result in this section is the following.
\[Jintegral\] Define $A = \frac12\l(1+\l)$. Under the conditions of Theorem \[thm:main\], there is a region $\S'\supseteq\S$ such that $$\begin{aligned}
\int_{\S'} G(\bithat,\biphat,\mu)
\, d\bithat d\biphat d\mu
&= (mn)^{-1/2} \Bigl(\frac{\pi}{Amn }\Bigr)^{\!1/2}
\Bigl(\frac{\pi}{An }\Bigr)^{(m-1)/2}
\Bigl(\frac{\pi}{Am }\Bigr)^{(n-1)/2} \\
&\kern-18mm {}\times \exp\biggl(
\frac12
- \frac{1+2A}{24A}\Bigl(\frac mn+\frac nm\Bigr)
+ O(n^{-b}) \biggr).\end{aligned}$$
For $x$ real and $|\l x|$ small, $$(1-\l(e^{ix}-1))^{-1} = \exp\(
\l ix - Ax^2 - iA_3x^3 + A_4 x^4 +\Oh\((1+\l)^5|x|^5\) \)$$ with $$A=\tfrac{1}{2}\l(1+\l),\; A_3=\tfrac{1}{6}\l(1+\l)(1+2\l),\;
A_4=\tfrac{1}{24}\l(1+\l)(1+6\l+6\l^2).$$ We are about to state an estimate for $G$, and some further estimates (\[EBounds\]) in a moment, all of which hold uniformly in the region $\S$. Let us alert the reader that we shall be defining a larger region, see (\[LargeR\]) below, which contains $\S$ and in which all of these estimates continue to hold true. Uniformly in $\S$, where all $(1+\l)|\mu+\that_j+\phat_k|$ are small, $$\begin{aligned}
G &= \exp\Bigl\{ -A\sumsum(\mu+\that_j+\phat_k)^2
- iA_3\sumsum(\mu+\that_j+\phat_k)^3
\\
& \qquad\qquad{}
+ A_4\sumsum(\mu+\that_j+\phat_k)^4
+ \Oh\((1+\l)^5\,\sumsum|\mu+\that_j+\phat_k|^5\,\)
\Bigr\}.\end{aligned}$$ Here and below, the undelimited summation over $j,k$ runs over $1\le j\le m$, $1\le k\le n$, and we continue to use the abbreviations $\th_m=-\sum_{j=1}^{m-1}\th_j$ and $\ph_n=-\sum_{k=1}^{n-1}\ph_k$.
We now proceed to a second change of variables, $(\bithat,\biphat,\mu)=T_2(\bis,\bitau,\mu)$ given by $$\that_j=\sigma_j+c\mu_1, ~~ \phat_k=\tau_k+d\nu_1,$$ where, for $1\le h \le 4$, $\mu_h$ and $\nu_h$ denote the power sums $\sum_{j=1}^{m-1}\sigma_j^h$ and $\sum_{k=1}^{n-1}\tau_k^h$, respectively. The scalars $c$ and $d$ are chosen to eliminate the second-degree cross-terms $\sigma_{j_1}\sigma_{j_2}$ and $\tau_{k_1}\tau_{k_2}$, and thus diagonalize the quadratic in $\bis=(\sigma_1,\dots,\sigma_{m-1})$ and $\bitau=(\tau_1,\dots,\tau_{n-1})$. Suitable choices for $c,d$ are $$c = -\frac{1}{m+m^{1/2}}, ~~ d = -\frac{1}{n+n^{1/2}},$$ and we find the following: $$\begin{aligned}
\sumsum(\mu+\that_j+\phat_k)^2 &=
mn\mu^2
+ n\mu_2
+ m\nu_2
\notag
\\[0.5ex]
\sumsum(\mu+\that_j+\phat_k)^3 &=
3\mu(n\mu_2+m\nu_2)
+ n(\mu_3+3c\mu_2\mu_1)
+ m(\nu_3+3d\nu_2\nu_1)
\notag
\\
& ~~~ + \tO((1+\l)^{-3}n^{-1/2})
\label{EBounds}
\\[0.5ex]
\sumsum(\mu+\that_j+\phat_k)^4 &=
6\mu_2\nu_2
+ n\mu_4
+ m\nu_4
+ \tO((1+\l)^{-4}n^{-1/2})
\notag
\\[0.5ex]
\sumsum|\mu+\that_j+\phat_k|^5
&= \tO((1+\l)^{-5}n^{-1/2})\,.
\notag\end{aligned}$$ The Jacobian of the matrix $T_2$ is $(mn)^{-1/2}$, and so $$\int_{\S}G = (mn)^{-1/2} \int_{T_2^{-1}(\S)} E_1,
\label{JEOne}$$ where $E_1=\exp(L_2+\tO(n^{-1/2}))$, and $$\label{L2value}
\begin{split}
L_2 =
- A n \mu_2 - A m \nu_2
- A m n \mu^2
- i A_3 n \mu_3 - i A_3 m \nu_3
- 3i A_3 n \mu \mu_2 - 3i A_3 m \mu \nu_2\\
- 3i A_3 c n \mu_2 \mu_1
- 3i A_3 d m \nu_2 \nu_1
+ A_4 n \mu_4 + A_4 m \nu_4
+ 6 A_4 \mu_2 \nu_2.
\end{split}$$
Define the regions $\Q$, $\M$, $\S'$, and $\R'$ by $$\begin{aligned}
\Q &= \bigl\{\,|\sigma_j| \le 2(1+\l)^{-1}n^{-1/2+\eps},
\;j=1,\ldots,m{-}1\,\bigr\} \\
&\qquad\qquad{}\cap
\bigl\{\,|\tau_k| \le 2(1+\l)^{-1}m^{-1/2+\eps},
\;k=1,\ldots,n{-}1\,\bigr\},\\
&\qquad\qquad{}\cap
\bigl\{\,|\mu|\le 2(1+\l)^{-1}(mn)^{-1/2+2\eps}\,\bigr\}
\\[0.3ex]
\M &= \bigl\{\,|\mu_1| \le (1+\l)^{-1}m^{1/2}n^{-1/2+2\eps}\,\bigr\}
\cap \bigl\{\,|\nu_1| \le (1+\l)^{-1}n^{1/2}m^{-1/2+2\eps}\,\bigr\},\end{aligned}$$ $$\label{SprimeDef}
\S' = T_2(\Q\cap\M),$$ and $$\label{RprimeDef}
\R' = T_1^{*}(\S').$$ Summing for $1 \le j \le m-1$ the equation $\that_j=\sigma_j+c\mu_1$, and inserting the value of $c$, we find $$m^{-1/2} \, \mu_1 = \sum_{j=1}^{m-1} \that_j.$$ In the region $\S$ we have $|\sum_{j=1}^{m-1} \that_j| \le
(1+\l)^{-1}n^{-1/2+\eps}$, and so in $T_2^{-1}(\S)$ we have $$|\mu_1| \le (1+\l)^{-1}m^{1/2}n^{-1/2+\eps}.$$ Using this, and the dual inequality for $|\nu_1|$, the reader can check that $$T_2^{-1}(\S) \subseteq \Q \cap \M.$$
It will be convenient if we can apply the bounds throughout the expanded region $\S'=T_2(\Q\cap\M)$, rather than only in $\S$. To see that this is valid, note that the calculations leading to these bounds hold equally well if the coefficient of $\eps$ in an exponent of $m$ or $n$ is increased; or, if an assumption such as $|\mu| \le \tbd (mn)^{1/2+2\eps}$ is made more permissive by a multiplicative constant: $|\mu| \le 2 \tbd (mn)^{1/2+2\eps}$. Therefore, it suffices to note that $S'$ lies inside the region $$\begin{aligned}
|\mu|&\le 2 \tbd (mn)^{-1/2+2\eps} \notag \\
|\th_j| &\le 3 \tbd n ^{-1/2+ \eps}, ~ 1 \le j \le m-1 \notag \\
|\th_m| &\le \tbd n ^{-1/2+2\eps}, & \label{LargeR} \\
|\ph_k| &\le 3 \tbd m ^{-1/2+ \eps}, ~ 1 \le k \le n-1 \notag \\
|\ph_n| &\le \tbd m ^{-1/2+2\eps}. \notag\end{aligned}$$
Define $E_2=\exp(L_2)$. We have shown that in the region $\Q\cap\M$ the integrand $E_1$ satisfies $E_1=E_2\(1+\tO(n^{-1/2})\)$. We can approximate our integral as follows: $$\begin{aligned}
\int_{\Q \cap \M} E_1
&= \int_{\Q \cap \M} E_2 + \tO(n^{-1/2})\int_{\Q \cap \M} \abs{E_2} \notag\\[0.4ex]
&= \int_{\Q \cap \M} E_2 + \tO(n^{-1/2})\int_{\Q} \,\abs{E_2} \notag\\[0.4ex]
&= \int_{\Q} E_2
+ O(1)\int_{\Q\cap \M^c} \abs{E_2} + \tO(n^{-1/2})\int_{\Q} \,\abs{E_2}.
\label{recipe}\end{aligned}$$ It suffices to estimate each of the three integrals in the last line of (\[recipe\]).
We first compute the integral of $E_2$ over $\Q$. We proceed in three stages, starting with integration with respect to $\mu$. For the latter, we can use the formula $$\int\limits_{-(mn)^{-1/2+2\eps}}^{(mn)^{-1/2+2\eps}}
\kern-2mm\exp\(-Amn\mu^2 - i\beta\mu\) d\mu
= \Bigl(\frac{\pi}{Amn}\Bigr)^{1/2}
\exp\biggl( -\frac{\beta^2}{4Amn} +O(n^{-1}) \biggr),$$ provided $\beta=o(A(mn)^{1/2+2\eps})$. In our case, $\beta=3A_3(n\mu_2+m\nu_2)$, which is small enough because $m=O(n\log n)$ and $n=O(m\log m)$. Integration over $\mu$ contributes $$\label{muint}
\Bigl(\frac{\pi}{Amn}\Bigr)^{1/2}
\exp\biggl(\frac{-9A_3^2(n\mu_2+m\nu_2)^2}{4Amn}
+O(n^{-1}) \biggr).$$
The second step is to integrate with respect to $\bis$ the integrand $$\begin{aligned}
\label{sigbits}
\begin{split}
\exp\biggl(\! &- An\mu_2
- \frac{9A_3^2n}{4Am}\mu_2^2
- i A_3 n\mu_3
- 3i A_3 cn \mu_1\mu_2 \\
&~{}+ \Bigl(6A_4-\frac{9A_3^2}{2A}\Bigr)\mu_2\nu_2 + A_4n\mu_4
+ O(n^{-1})\biggr).
\end{split}\end{aligned}$$ This is accomplished by an appeal to Lemma \[MW3\]. We must take the $N$ of that lemma equal to $m-1$, the number of variables. This dictates that the limits of integration be $\pm (m-1)^{1/2+\hat{\eps}}$, but our limits, based on the definition of $\Q$, are $\pm 2(1+\l)^{-1}n^{-1/2+\eps}$. Thus, we make a change of scale, $$\sigma_j \leftarrow \frac{\sigma_j}{(1+\lambda)(n/m)^{1/2}}$$ before integrating. This change of scale will introduce a multiplicative factor $(1+\lambda)^{-(m-1)}(m/n)^{(m-1)/2}$ into our evaluation of the integral. In the terminology of that lemma, we have $N = m-1$, $\delta(N)=O(n^{-1})$, $\eps'=\tfrac32\eps$, $\eps''=\tfrac53\eps$, $\eps'''=3\eps$, $\bar\eps=6\eps$, and $\hat\eps(N)=\eps+o(1)$ is defined by $2m^{-1/2+\eps}=(m-1)^{-1/2+\hat\eps}$. Furthermore, taking account of the scale, we find $$\begin{aligned}
\mwA &= \frac{A}{(1+\l)^2}\frac{m}{m-1}, &
\mwa_j &= \frac{12AA_4-9A_3^2}{2A(1+\l)^2} \frac{m}{n}\nu_2, &&\\
\mwB_j &= -\frac{iA_3}{(1+\l)^3}\frac{m^{1/2}}{n^{1/2}}\frac{m}{m-1}, &
\mwC_{jj'} &= -\frac{3iA_3}{(1+\l)^3} cn \frac{m^{3/2}}{n^{3/2}}, && \\
\mwE_j &= \frac{A_4}{(1+\l)^4} \frac{m}{n} \frac{m}{m-1}, &
\mwF_{jj'} &= -\frac{9A_3^2}{4A(1+\l)^4} \frac{m}{n}, & \mwJ_j &= 0.\end{aligned}$$ We can take $\Delta=\tfrac45$, and calculate that $$\begin{gathered}
\frac{3}{4\mwA^2N}\sum_{j=1}^N \mwE_j
+ \frac{1}{4\mwA^2N^2}\sum_{j,j'=1}^N \mwF_{jj'} = \frac{m}{n}
\biggl(\frac{3A_4}{4A^2}-\frac{9A_3^2}{16A^3}\biggr) + \tO(n^{-1}) \notag\\
\frac{15}{16\mwA^3N}\sum_{j=1}^{N}\mwB_j^2
+ \frac{3}{8\mwA^3N^2}\sum_{j,j'=1}^{N}\mwB_j\mwC_{jj'}
+ \frac{1}{16\mwA^3N^3}\sum_{j,j'\!,j''=1}^N \mwC_{jj'}\mwC_{jj''}
= -\frac{3A_3^2m}{8A^3n} + \tO(n^{-1}) \notag\\
\frac{1}{2\mwA N}\sum_{j=1}^{N}\mwa_j
+ \frac{1}{4\mwA^2N^2}\sum_{j=1}^{N}\mwa_j^2
=
\frac{m}{n}\biggl(\frac{3A_4}{A}-\frac{9A_3^2}{4A^2}\biggr)\nu_2
+\tO(n^{-1})
\label{taubit}\\
\mwZ = Z_1 = \exp\biggl( \frac{3A_3^2m}{8A^3n} + \tO(n^{-1}) \biggr)
= O(1) \exp\biggl(\frac {(1+2\lambda)^2 m}{24 An}\biggr).\notag\end{gathered}$$
After checking that $\Theta_2 = \tO(n^{-1})=o(m^{-4/5}Z_1)$, we conclude that integration with respect to $\bis$ contributes a $\bitau$-free factor $$\label{contrib1}
\Bigl(\frac{\pi}{An}\Bigr)^{\!(m-1)/2}
\exp\biggl(
\Bigl(\frac{3A_4}{4A^2}-\frac{15A_3^2}{16A^3}\Bigr) \frac{m}{n}
+ O(m^{-4/5}Z_1)
\biggr).$$ By the conditions of Theorem \[thm:main\], $Z_1\le n^{1/5}$, so $m^{-4/5}Z_1=o(1)$ as required by Lemma \[MW3\].
Finally, we need to integrate over $\bitau$. Collecting the remaining terms from , and the terms involving $\bitau$ from (\[muint\]) and (\[taubit\]), we have an integrand equal to $$\begin{aligned}
&
\exp\biggl(
-Am\nu_2
+\Bigl(\frac{3A_4m}{An}-\frac{9A_3^2m}{4A^2n}\Bigr)\nu_2
-\frac{9A_3^2m}{4An}\nu_2^2
\\
& \kern11mm{}
+A_4m\nu_4 -iA_3m\nu_3 -3iA_3dm\nu_2\nu_1
+ O(m^{-4/5}Z_1)
\biggr).\end{aligned}$$
Again we use Lemma \[MW3\]. This time, the factor due to scaling is $(1+\l)^{-(n-1)}(n/m)^{(n-1)/2}$, and $N = n-1$, $\delta(N)=O(m^{-4/5}Z_1)$, $\eps'=\tfrac32\eps$, $\eps''=\tfrac53\eps$, $\eps'''=3\eps$, $\bar\eps=4\eps$, and $\hat\eps(N)=\eps+o(1)$, as defined by $2n^{-1/2+\eps}=(n-1)^{-1/2+\hat\eps}$. The substitution table this time reads $$\begin{aligned}
\mwA &= \frac{A}{(1+\l)^2}\frac{n}{n-1}, &
\mwa_k &= \frac{12AA_4-9A_3^2}{4A^2(1+\l)^2}, &&\\
\mwB_k &= -\frac{iA_3}{(1+\l)^3}\frac{n^{1/2}}{m^{1/2}}\frac{n}{n-1}, &
\mwC_{kk'} &= -\frac{3iA_3}{(1+\l)^3} dm \frac{n^{3/2}}{m^{3/2}}, && \\
\mwE_k &= \frac{A_4}{(1+\l)^4} \frac{n}{m} \frac{n}{n-1}, &
\mwF_{kk'} &= -\frac{9A_3^2}{4A(1+\l)^4} \frac{n}{m}, & \mwJ_k &= 0.\end{aligned}$$ This time we take $\Delta_2>4/5$ so that $n^{-\Delta_2}=o(\delta(N))$; calculations similar to the previous case lead to $$\begin{gathered}
\frac{3}{4\mwA^2N}\sum_{k=1}^N \mwE_k
+ \frac{1}{4\mwA^2N^2}\sum_{k,k'=1}^N \mwF_{kk'} = \frac{n}{m}
\biggl(\frac{3A_4}{4A^2}-\frac{9A_3^2}{16A^3}\biggr) + \tO(n^{-1})
\\
\frac{15}{16\mwA^3N}\sum_{k=1}^{N}\mwB_k^2
+ \frac{3}{8\mwA^3N^2}\sum_{k,k'=1}^{N}\mwB_j\mwC_{kk'}
+ \frac{1}{16\mwA^3N^3}\sum_{k,k'\!,k''=1}^N \mwC_{kk'}\mwC_{kk''}\\
{\kern 8cm} = -\frac{3A_3^2n}{8A^3m} + \tO(n^{-1}) \\
\frac{1}{2\mwA N}\sum_{k=1}^{N}\mwa_k
+ \frac{1}{4\mwA^2N^2}\sum_{k=1}^{N}\mwa_k^2
=
- \frac{9A_3^2}{8A^3} + \frac{3A_4}{2A^2} + \tO(n^{-1}) \\
\mwZ = Z_2 = \exp\biggl( \frac{3A_3^2n}{8A^3m} + \tO(n^{-1}) \biggr)
= O(1) \exp\biggl(\frac {(1+2\lambda)^2 n}{24 Am}\biggr).\end{gathered}$$
Again $\Theta_2=\tO(n^{-1})$, implying that $\Theta_2=o(m^{-4/5}Z_1Z_2)$. Including the contributions from and , we obtain $$\label{QE2a}
\begin{split}
\int_\Q E_2
&= \Bigl(\frac{\pi}{Amn}\Bigr)^{1/2}
\Bigl(\frac{\pi}{An}\Bigr)^{\!(m-1)/2}
\Bigl(\frac{\pi}{Am}\Bigr)^{\!(n-1)/2} \\
&{\kern-5mm}\times\exp\biggl( - \frac{9A_3^2}{8A^3} + \frac{3A_4}{2A^2}
+\Bigl(\frac mn+\frac nm\Bigr)
\Bigl( \frac{3A_4}{4A^2}-\frac{15A_3^2}{16A^3}\Bigr) + O(m^{-4/5}Z_1Z_2) \biggr).
\end{split}$$
Since $$Z_1Z_2=O(1) \exp\left( \frac{3A_3^2}{8A^3} \left(
\frac{m}{n} + \frac{n}{m}
\right)
\right),$$ and since $$\frac{3A_3^2}{8A^3} = \frac{(1+2\l)^2}{24A},$$ it follows from the main hypothesis (\[Hyp\]) of Theorem \[thm:main\] that $Z_1Z_2 = O(n^{6a/5})$. The condition $a+b<1/2$ implies $-4/5+6a/5<-b-1/5$; hence, substituting the values of $A, A_3, A_4$, we conclude that $$\label{QE2}
\begin{split}
\int_\Q E_2
&= \Bigl(\frac{\pi}{Amn}\Bigr)^{1/2}
\Bigl(\frac{\pi}{An}\Bigr)^{\!(m-1)/2}
\Bigl(\frac{\pi}{Am}\Bigr)^{\!(n-1)/2} \\
&\qquad{}\times\exp\biggl( \frac12
- \frac{1+2A}{24A}\Bigl(\frac mn+\frac nm\Bigr)
+ O(n^{-b-1/5}) \biggr).
\end{split}$$
We next infer a estimate of $\int_\Q\, \abs{E_2}$. The calculation that lead to remains valid if we set $A_3$ to zero, which is the same as replacing $L_2$ by its real part. Since $\abs{E_2} = \exp\(\Re(L_2)\)$, this gives $$\begin{aligned}
\int_\Q\, \abs{E_2}
&= \exp\biggl(\frac{(1+2\lambda)^2}{8A}\Bigl( 1 +
\frac{5n}{6m} + \frac{5m}{6n}\Bigr)+ o(1)\biggr)\int_Q E_2 \notag\\
&= O(n^a)\int_Q E_2 \label{absE2}\end{aligned}$$ under the assumptions of Theorem \[thm:main\]. The third term of (\[recipe\]) can now be identified: $$\label{OEbit}
\tO(n^{-1/2}) \int_\Q\, \abs{E_2} = \tO(n^{-1/2+a}) \int_Q E_2
= O(n^{-b}) \int_Q E_2\,.$$
Finally, we consider the second term of , namely $$\int_{\Q\cap \M^c} \abs{E_2},$$ which we will bound as a fraction of $\int_\Q\, \abs{E_2}$ using a statistical technique. The following is a well-known result of Hoeffding [@hoeffding].
\[hoeffding\] Let $X_1,X_2,\ldots,X_N$ be independent random variables such that $EX_i=0$ and $\abs{X_i}\le M$ for all $i$. Then, for any $t\ge 0$, $$\Prob\biggl(\,\sum_{i=1}^N X_i \ge t\biggr)
\le \exp\biggl(-\frac{t^2}{2NM^2}\biggr).$$
Now consider $\abs{E_2} = \exp\(\Re(L_2)\)$. Write $\M=\M_1\cap\M_2$, where $\M_1=\{\, \abs{\mu_1}\le m^{1/2} n^{-1/2+2\eps}\,\}$ and $\M_2=\{\, \abs{\nu_1}\le n^{1/2} m^{-1/2+2\eps}\}$. For fixed values of $\mu$ and $\bis$, $\Re(L_2)$ separates over $\tau_1,\tau_2,\ldots,\tau_{n-1}$ and therefore, apart from normalization, it is the joint density of independent random variables $X_1,X_2,\ldots,X_{n-1}$ which satisfy $EX_k=0$ (by symmetry) and $\abs{X_k}\le 2(1+\l)^{-1}m^{-1/2+\eps}$ (by the definition of $\Q$). By Lemma \[hoeffding\], the fraction of the integral over $\bitau$ (for fixed $\mu,\bis$) that has $\nu_1\ge (1+\l)^{-1}n^{1/2}m^{-1/2+2\eps}$ is at most $\exp(-m^{2\eps}/8)$. By symmetry, the same bound holds for $\nu_1\le -(1+\l)^{-1}n^{1/2}m^{-1/2+2\eps}$. Since these bounds are independent of $\mu$ and $\bis$, we have $$\int_{\Q\cap\M_2^c} \,\abs{E_2}
\le 2\exp(-m^{2\eps}/8)\int_\Q\,\abs{E_2}.$$ By the same argument, $$\int_{\Q\cap\M_1^c} \,\abs{E_2}
\le 2\exp(-n^{2\eps}/8)\int_\Q\,\abs{E_2}.$$ Therefore we have in total that $$\begin{aligned}
\label{corners}
\int_{\Q\cap\M^c} \,\abs{E_2}
&\le 2\(\exp(-m^{2\eps}/8)+\exp(-n^{2\eps}/8)\) \int_\Q\,\abs{E_2}
\notag \\
&\le O(n^{-b}) \int_Q E_2,\end{aligned}$$ using again (\[QE2\]). Applying with , and , we find that $\int_{\Q\cap\M} E_1$ is given by with the error term replaced by $O(n^{-b})$. Multiplying by the Jacobian of the transformation $T_2$, we find that Theorem \[Jintegral\] is proved for $\S'$ given by (\[SprimeDef\]).
Concentration of the integral {#section:Boxing}
=============================
Recall that $F(\bit,\bip)$ is the integrand in equation (\[Idef2\]) defining $I(m,n)$, and that $\R$ is the region given by (\[RDef\]). In the previous section we estimated the integral of $F(\bit,\bip)$ over a particular superset $\R'\supseteq\R$. In this section we show that the integral of $F(\bit,\bip)$ outside $\R$ is negligible in comparison if $\lambda$ is polynomially bounded. Larger values of $\lambda$ will be handled in the following section.
\[boxing\] Suppose that $m,n \rightarrow \infty$ in such a way that holds and $\lambda=n^{O(1)}$. Define $I_0$ by $$\label{ThmAssumption}
I_0 = (mn)^{1/2}
\Bigl(\frac{\pi}{Amn }\Bigr)^{ 1/2}
\Bigl(\frac{\pi}{An }\Bigr)^{ (m-1)/2}
\Bigl(\frac{\pi}{Am }\Bigr)^{ (n-1)/2}
\! \exp\biggl(
- \frac{1+2A}{24A}\Bigl(\frac mn+ \frac nm\Bigr)
\biggr).$$ Then, for sufficiently small $\eps>0$, $$\int_{\R^c} |F| = O(e^{-n^{\eps}})I_0.$$
We begin with two technical lemmas whose proofs are omitted.
\[fbnd\] The absolute value of the integrand of $I(m,n)$ is $$|F(\bit,\bip)| = \prod_{j,k} f(\t_j + \p_k),$$ where $$f(z) = \(1+4A(1-\cos z))^{-1/2}.$$ Moreover, for all real $z$ with $\abs z\le \tfrac1{10}(1+\l)^{-1}$, $$0\le f(z) \le \exp\( -A z^2 + (\tfrac1{12} A+A^2) z^4\).
\quad\qedsymbol$$
\[ibnd\] Define $N = \lceil 6000 (1+\l)\rceil$, $\d=2\pi/N$ and $g(x)=-Ax^2+(\frac94A+27A^2)x^4$. Then, uniformly for $\l>0$, $K\ge 1$, $$\int_{-30\d}^{30\d} \exp\( K g(x) \)\,dx
\le \sqrt{\pi/(AK)}\, \exp\( O(K^{-1}+(AK)^{-1})\).
\quad\qedsymbol$$
Let $N$ and $\delta$ be as given in Lemma 4. Define the region $\A$ to be the set of those $(\bit,\bip)$ such that $$\cos(\theta_j+\phi_k) \le \cos \delta$$ for at least $\tfrac13\min(mn^{\eps},m^{\eps}n)$ pairs $(j,k)$. Define $x_0,x_1,\ldots,x_{N-1}$ by $x_\ell=\ell\delta$.
If $X\subseteq (-\pi,\pi]$, we denote by $\Ntheta(X)$ the number of values of $j$ such that $\theta_j\in X$, and similarly define $\Nphi(X)$. Define region $\R_1(\ell)$ to be the set of those $(\bit,\bip)$ such that $\Ntheta([x_\ell-4\delta,x_\ell+4\delta])\ge m-m^\eps$ and $\Nphi([-x_\ell-4\delta,-x_\ell+4\delta])\ge n-n^\eps$.
Let $\U=\bigcup_{\ell=0}^{N-1}\R_1(\ell)$. The proof of the theorem consists in proving these three relations: $$\begin{aligned}
\A\cup\U &= [-\pi,\pi]^{m+n}\label{R1}\\[0.8ex]
\int_{\A}|F| &= O(e^{-n})I_0\label{R2}\\[0.5ex]
\int_{\U\cap\R^c}|F|
&= O(e^{-n^\eps})I_0.\label{R3}\end{aligned}$$
To begin the proof of , we show that any point $(\bit,\bip)$ for which $\Ntheta([x_\ell-\delta,x_\ell+\delta]\ge m^\eps$ belongs to $\A\cup\U$. Indeed, if such a point does not belong to $\A$, then it must have $\Nphi([-x_\ell-2\delta,-x_\ell+2\delta])\ge \tfrac23 n$. This in turn forces $\Ntheta([x_\ell-3\delta,x_\ell+3\delta])\ge m-m^\eps$, which forces $\Nphi([-x_\ell-4\delta,-x_\ell+4\delta])\ge n-n^\eps$. In particular, $(\bit,\bip)\in\R_1(\ell)$.
To complete the proof of , we show that $(\bit,\bip)$ belongs to $\A$ if $\Ntheta([x_\ell-\delta,x_\ell+\delta])\le m^\eps$ for all $\ell$. Let $a$ be a minimum-length interval $[x_\ell,x_{\ell'}]$ such that $\Ntheta(a)\ge m^\eps$. Then $\Ntheta(a)\le 2m^\eps$ and the complementary interval $\overline{a}$ has $\Ntheta(\overline a)\ge m-2m^\eps$. If $b=[x_{\ell'+2},x_{\ell-2}]$ (a subinterval of $\overline a$), then $\Ntheta(b)\ge m-4m^\eps$. Thus, there are at least $m^\eps$ disjoint pairs $(j,j')$ with $\theta_j\in a$ and $\theta_{j'}\in b$. (By disjoint we mean there are $2m^\eps$ distinct indices $j$ involved.) Because of the $2\delta$ spaces between arcs $a$ and $b$, for each $k$ and each pair $(j,j')$ at least one of $\cos(\theta_j+\phi_k)$ or $\cos(\theta_{j'}+\phi_k)$ is bounded above by $\cos\delta$. This implies that $(\bit,\bip)\in\A$, as claimed, and completes the proof of .
We turn next to . Since $A\delta^2=\Theta\(\l(1+\l)^{-1}\)$, Lemma 3 implies $$\abs{F(\bit,\bip)} \le \exp\(-c_1\l(1+\l)^{-1}\min(m^{\eps}n,mn^{\eps})\)$$ for all $(\bit,\bip)\in\A$ and some $c_1>0$. The volume of $\A$ is no more than $(2\pi)^{m+n}$, and $\l(1+\l)^{-1}>(\log n)^{-1}$ by , so $$\int_{\A}|F|
\le (2\pi)^{m+n}\exp\(-c_1(\log n)^{-1}\min(m^{\eps}n,mn^{\eps})\).$$ From , which implies that $A=O(\log n)$, and the assumption that $\lambda=n^{O(1)}$, we have that $I_0=\exp\(O(m\log n+n\log m)\)$. Relation follows.
Finally we come to . For $(\bit,\bip)\in\R_1(\ell)$ we define $S_0=S_0(\bit)$, $S_1=S_1(\bit)$ and $S_2=S_2(\bit)$ to be the set of indices $j$ such that $|\t_j-x_{\ell}|$ is: less than or equal to $4\d$, in the interval $(4\d,5\d]$, and larger than $5\d$, respectively. The index sets $T_0=T_0(\bip)$, $T_1=T_1(\bip)$ and $T_2=T_2(\bip)$ are defined similarly. Define $\R_1(\ell;m_2,n_2)$ to be that subregion of $\R_1(\ell)$ for which $|S_2|=m_2$ and $|T_2|=n_2$. Define $\U(m_2,n_2)$ by $$\U(m_2,n_2) = \bigcup_{\ell=0}^{N-1}\R_1(\ell;m_2,n_2).$$ We note that $m_2$ and $n_2$ vary over the ranges $[0,m^{\eps}]$ and $[0,n^{\eps}]$, respectively. Define $\U_0=\U(0,0)$ and $\U_*=\U\setminus\U_0$.
Suppose $(\bit,\bip)\in\R_1(0;m_2,n_2)$. If $|\t_j|$ and $|\p_k|$ are both less than or equal to $5\d$, then Lemma 3 is applicable to $f(\t_j+\p_k)$. If one of the two is less than or equal to $4\d$ and the other exceeds $5\d$, then $\cos(\t_j+\p_k)\le\cos\d$. Thus, there exists $c_2>0$ such that for $(\bit,\bip)\in\R_1(0;m_2,n_2)$ $$\label{R4}
f(\t_j+\p_k) \le
\begin{cases}
\,\exp\(-A(\t_j+\p_k)^2+(\tfrac1{12}A+A^2)(\t_j+\p_k)^4\), \kern -32mm& \\
& \textrm{if }(j,k) \in (S_0\cup S_1)\times(T_0\cup T_1),\\[0.8ex]
\,\exp\( -2c_2\l(1{+}\l)^{-1} \),
& \textrm{if }(j,k) \in (S_0\times T_2)\cup (S_2\times T_0),\\[0.3ex]
\,1, & \textrm{otherwise}.
\end{cases}$$ Since $(\bit,\bip)\in\U$, we have $|S_0|\ge m-m^{\eps}$ and $|T_0|\ge n-n^{\eps}$; thus, the size of $(S_0\times T_2)\cup (S_2\times T_0)$ exceeds $\tfrac12(mn_2+nm_2)$. Integrating the upper bound on $|F|$ implied by (\[R4\]), we find $$\label{R22}
\int_{\R_1(0;m_2,n_2)}\negthickspace\negthickspace \abs{F}
\le (2\pi)^{m_2+n_2}
\binom{m}{m_2}\binom{n}{n_2}
\exp\(-c_2\l(1{+}\l)^{-1}(mn_2+m_2n)\)
\, I'_2(m_2,n_2)$$ with $$I'_2(m_2,n_2) = \int_{-5\d}^{5\d}\!\!\!\cdots \int_{-5\d}^{5\d}
\exp\Bigl(- A\sumpp(\t_j+\p_k)^2 +
(\tfrac1{12}A+A^2)\sumpp(\t_j+\p_k)^4\Bigr)
\, d\bit'' d\bip'',\notag$$ where the double-primes denote restriction to $j\in S_0\cup S_1$ and $k\in T_0\cup T_1$. The factor $(2\pi)^{m_2+n_2}$ comes from integrating over $\t_j$ for $j\in S_2$ and $\p_k$ for $k\in T_2$, while the binomial coefficients account for the choices of $S_2$ and $T_2$.
Set $m'=m-m_2,n'=n-n_2$. As implied by the notation, the value of $I'_2(m_2,n_2)$ is independent of which specific variables constitute the sets $S_0\cup S_1$ and $T_0\cup T_1$.
Summing on $m_2+n_2\ge 1$ yields an upper bound for the integral of $|F|$ over the region $\bigcup_{m_2+n_2\ge 1}\R_1(0;m_2,n_2)$. Notice, however, that the transformation $\t_j\mapsto\t_j-x_\ell$, $\p_k\mapsto\p_k+x_\ell$ is a volume-preserving bijection of $\R_1(\ell;m_2,n_2)$ with $\R_1(0;m_2,n_2)$ which leaves $F(\bit,\bip)$ invariant. Thus, introducing an additional factor of $N$ to account for all values of $\ell$, we have $$\begin{aligned}
\int_{\U_*}|F|
& \,\le\, N\negthickspace \sum_{m_2+n_2\ge 1}
\binom{m}{m_2}\binom{n}{n_2}(2\pi)^{m_2+n_2}
&\notag \\
& \qquad{}\times
\exp\( -c_2\l(1+\l)^{-1}(mn_2+nm_2) \)
\, I'_2(m_2,n_2). \label{R6}\end{aligned}$$
We now analyze $I_2'(m_2,n_2)$ more closely. Define $\tav'=(m')^{-1}\sumpp\t_j$, $\breve\t_j=\t_j-\tav'$ for $j\in S_0\cup S_1$, $\pav'=(n')^{-1}\sumpp\p_k$, $\breve\p_k=\p_k-\pav'$ for $k\in T_0\cup T_1$, $\mu'=\pav'+\tav'$ and $\nu'=\pav'-\tav'$. In terms of $\mu',\nu', \breve\t_j, \breve\p_k$ we have $$\sumpp(\t_j+\p_k)^2 =
m'n'{\mu'}^2 + n'\sumpp_j\breve\t_j^2
+ m'\sumpp_k \breve\p_k^2$$ and $$\sumpp(\t_j+\p_k)^4 \le
27m'n'{\mu'}^4 + 27n'\sumpp_j\breve\t_j^4 + 27m'\sumpp_k \breve\p_k^4.$$ The latter follows from the inequality $(x+y+z)^4 \le 27(x^4+y^4+z^4)$ valid for all $x,y,z$. It follows that $$\label{R5}
I'_2(m_2,n_2) \le
\int_{-5\d}^{5\d}\!\!\!\cdots \int_{-5\d}^{5\d}
\exp\Bigl(
m'n' g(\mu')
+ n'\sumpp_j g(\breve\t_j) + m'\sumpp_k g(\breve\p_k)
\Bigr) \, d\bit'' d\bip''.$$ For the moment we are thinking of $\mu'$, $\nu'$, $\breve\t_j$, and $\breve\p_k$ ($1\le j\le m', 1\le k\le n'$) as functions of the variables of integration. Arbitrarily choose one element $j_*\in S_0\cup S_1$ and one element $k_*\in T_0\cup T_1$. Define $S_3=(S_0\cup S_1)-\{j_*\}$ and $T_3=(T_0\cup T_1)-\{k_*\}$. We claim that the inequality (\[R5\]) remains valid when $\sumpp_j \equiv \sum_{j\in S_0\cup S_1}$ is replaced by $\sumpp_{j\neq j_*} \equiv \sum_{j\in S_3}$, and similarly for $k_*$. To verify this assertion, note that since $|\t_j|\le 5\d$ for all $j\in S_0\cup S_1$ it must be the case that $|\tav'|\le 5\d$, too, and so $|\breve\t_{j_*}|\le 10\d$. The claim now follows because, as is readily checked, $g(x)\le 0$ for $|x|\le 10\d$.
As in the previous section, the next step is to make a change of variables to $\mu'$, $\nu'$, $\breve\t_j$ ($j\in S_3$), and $\breve\p_k$ ($k\in T_3$). The region corresponding under this change of variables to $[-5\d,5\d]^{m'+n'}$ is not exactly a product, but it is contained in the product $[-10\d,10\d]^{m'+n'}$. (The argument given a few lines earlier to show $|\breve\t_{j_*}|\le 10\d$ applies equally well to $|\breve\t_j|$, $j\in S_3$.) Hence, $$\begin{aligned}
I'_2(m_2,n_2) \le O(\d)m'n'
\int_{-10\d}^{10\d}\!\!\cdots \int_{-10\d}^{10\d}
\exp\Bigl(& m'n' g(\mu') + n' \sum\nolimits_{j\in S_3} g(\breve\t_j) \\
&{}+ m' \sum\nolimits_{k\in T_3} g(\breve\p_k) \Bigr)\,
d\breve\t_{j\in S_3}d\breve\p_{k\in T_3}\, d\mu'.\end{aligned}$$ Here, the factor $O(\d)m'n'$ comes from the integration of $\nu'$ over $[-10\d,10\d]$ and the Jacobian $m'n'/2$ of the transformation. Now the integral splits into a product of $m'+n'-1$ one-dimensional integrals. Each of these factors can be bounded by Lemma \[ibnd\], and we find that $$\begin{aligned}
I'_2(m_2,n_2) &\le O(\d) m' n'
\Bigl(\frac{\pi}{Am'n'}\Bigr)^{1/2}
\, \Bigl(\frac{\pi}{Am'}\Bigr)^{(n'-1)/2}
\, \Bigl(\frac{\pi}{An'}\Bigr)^{(m'-1)/2} \\
&\qquad{}\times\exp\( o(1) + O(1+A^{-1})\(m'/n' + n'/m'\) \).\end{aligned}$$ By $m'\sim m,n'\sim n$ and (\[Hyp\]), the $\exp(\cdots\)$ term above equals $n^{O(1)}$. Also, $$\Bigl(
\frac{\pi}{Am'}
\Bigr)^{(n'-1)/2}
=
\Bigl(
\frac{\pi}{Am}
\Bigr)^{(n-1)/2}
n^{O(m_2+n_2)},$$ and similarly for the other terms. Noting that $\d N=O(1)$, we find from that $$\int_{\U_*}|F| \le
\sum_{m_2+n_2\ge 1} n^{O(m_2+n_2)}
\exp\( -c_3\l(1+\l)^{-1}(mn_2+nm_2)\) I_0.$$ Since $\l=\Omega\((\log n)^{-1}\)$, the sum is dominated by the terms with $m_2+n_2=1$. We conclude that $$\label{R7}
\int_{\U_*}|F| = O(e^{-n^{1-\eps}}) I_0.$$
It remains to consider $\U_0\cap\R^c$. As was argued before in obtaining (\[R22\]) we have in the $m_2=n_2=0$ case $$\int_{\R_1(0;0,0)\cap\R^c}|F|
\le
\int_{[-5\d,5\d]^{m+n}\cap\R^c} \negthickspace
\exp\Bigl(- A\sum(\t_j+\p_k)^2 +
(\tfrac1{12}A+A^2)\sum(\t_j+\p_k)^4\Bigr)
\, d\bit d\bip,$$ where now the sums are over all $j,k$. Observe that the transformation $\t_j\mapsto\t_j-x_{\ell},\p_k\mapsto\p_k+x_{\ell}$ takes $\R_1(\ell;0,0)\cap\R^c$ bijectively to $\R_1(0;0,0)\cap\R^c$, since $\R,\R^c$ are invariant under this mapping. Hence, the integral over $\U_0\cap\R^c$ of $|F|$ is no larger than $N$ times the integral on the right side of the last displayed relation.
The region of integration, $[-5\d,5\d]^{m+n}\cap\R^c$, is contained in the union $\bigcup_{h=1}^{m+n+1}\P_h$, where $\P_h$ equals the product $[-5\d,5\d]^{m+n}$ intersected with that part of $\R^c$ where the $h$th inequality of definition (\[RDef\]) fails. As was done in the $\U_*$ case, we may bound $\int_{\P_h}\cdots$ by a product of $m+n-1$ one-dimensional integrals. However, in the present situation, when we make the change of variables to $(\mu',\nu',\breve\t_j,\breve\p_k)$, the transformed region of integration, albeit still contained in the product $[-10\d,10\d]^{m+n-1}$ (we are omitting $\nu'$ from the discussion), has the additional property that one of the latter intervals, the one corresponding to the index $h$, is missing a neighborhood of size $(1+\l)^{-1}(mn)^{-1/2+2\eps}$, $(1+\l)^{-1}n^{-1/2+\eps}$, or $(1+\l)^{-1}m^{-1/2+\eps}$ about 0.
Throughout $[-10\d,10\d]$ we have $g(x)\le-Ax^2/2$. Because of the inequality $$\int_{(1+\l)^{-1}K^{-1/2+\alpha}}^{\infty}e^{-KAx^2/2}dx
\le
\frac{\pi}{(KA)^{1/2}}
\, \exp\(
-\dfrac15\l(1+\l)^{-1} K^{2\alpha}
\),$$ ($K$ being one of $mn$, $m$, or $n$; and $\alpha$ being, respectively, $2\eps$, $\eps$, or $\eps$), one of the $m+n-1$ factors is smaller than $(\pi/Amn)^{1/2}$, $(\pi/Am)^{1/2}$, or $(\pi/An)^{1/2}$ (respectively, depending on $K$) times $\exp\(-\tfrac15\l(1+\l)^{-1}\min(m^{2\eps},n^{2\eps})\)$. (One may assure that neither of the indices $j_*$, $k_*$ chosen for omission in the change of variables coincides with the index connected to the $h$th inequality of the negation of definition (\[RDef\]).) The factor $\exp(-\tfrac15\cdots)$ is, of course, independent of $h$. Allowing $N$ values of $\ell$ and $m+n+1$ values of $h$, we have $$\begin{aligned}
\int_{\U_0\cap\R^c} |F| &\le (m+n+1) N O(\delta)
n^{O(1)} \,
\left(\frac{\pi}{Amn}\right)^{1/2}
\, \left(\frac{\pi}{Am}\right)^{(n-1)/2}
\, \left(\frac{\pi}{An}\right)^{(m-1)/2} \\
& ~~~~ \times \exp\(
-\dfrac15\l(1+\l)^{-1}\min(m^{2\eps},n^{2\eps})
\) \\
&\le e^{-n^\eps} I_0.\end{aligned}$$ This inequality and (\[R7\]) together imply (\[R3\]), thus completing the proof of Theorem \[Jintegral\].
Proof of Theorem \[thm:main\] {#section:ProofThm1}
=============================
Once again we remind the reader that $F(\bit,\bip)$ denotes the integrand in equation (\[Idef2\]) defining $I(m,n)$. We continue to use $I_0$ for the quantity defined in equation (\[ThmAssumption\]).
Let $a,b$ be given positive numbers satisfying $a+b<\tfrac12$, and let $(m,n,s,t)$ be a sequence of 4-tuples such that $m,n\rightarrow\infty$ in such a way that hypothesis is satisfied.
We first note that both and Corollary \[biglambda\] are equivalent to for the respective ranges of $\l$ indicated. Consequently, to prove Theorem \[thm:main\] and Corollary \[biglambda\] it suffices to prove for $\lambda=O(n^5)$ and Corollary \[biglambda\] for $\lambda \ge n^5$.
We will divide the proof into two ranges of the parameter $\l$. First we assume that $\l=O(n^5)$, where Theorem \[boxing\] can be applied, then we use a different method for $\l\ge n^5$.
First suppose that $\l=O(n^5)$. As explained in Section \[section:evalint\], $$\label{FiveOne}
M(m,s;n,t) = \frac{1}{(2\pi)^{m+n}} \(\l^{-\l}(1+\l)^{1+\l}\)^{mn} \,
\biggl(\,
\int_{\R'}F + O(1)\int_{\R^c}|F|
\biggr),$$ where $\R$ is defined by and $\R' \supseteq \R$ is defined by . By and Theorem \[Jintegral\], the first integral on the right side of equals $2\pi \exp\(\frac12+O(n^{-b})\) I_0$, and by Theorem \[boxing\], the second equals $O(n^{-1}) I_0$. These values yield Theorem \[thm:main\] for this case in the form given by .
For the remainder of the proof, we assume that $\lambda\ge n^5$. By the first part of this proof, Corollary \[biglambda\] holds for $\l=O(n^5)$; hence, $$\label{ratio1}
\frac{M(m,2n^6; n,2mn^5)}{M(m,n^6; n,mn^5)}
= 2^{(m-1)(n-1)}
\(1+O(n^{-b})\).$$ Let $s_0=\lcm(m,n)/m$, $t_0=\lcm(m,n)/n$, and let $\P$ be the convex polytope of $m\times n$ real, nonnegative matrices whose rows sum to $s_0$ and whose columns sum to $t_0$. For every pair $(s,t)$ such that $ms=nt$ there is an integer $q$ such that $(s,t)=(qs_0,qt_0)$, and $M(m,s;n,t)$ equals the number of integer lattice points in the dilated polytope $q\P$. The latter count, as a function of the positive integer $q$, is called the [*polytope enumerator*]{} for $\P$, and is denoted $L_{\P}(q)$, [@BeckRobins]. Thus, we have $$M(m,s;n,t)=L_{\P}(q).$$ The polytope $\P$, an example of a [*transportation polytope*]{}, has integral vertices [@Brualdi], and so $L_{\P}(q)$ is a polynomial in $q$, the [*Ehrhart polynomial*]{}. The degree of $L_{\P}$ is $d$, the dimension of the polytope $\P$; in our case, $d=(m{-}1)(n{-}1)$. By a theorem of Stanley (an algebraic proof first appeared as Proposition 4.5 in [@stanleyC]; a more geometric proof is given as Theorem 2.1 in [@stanleyB]) the representation of $L_{\P}(q)$ in a particular basis, $$L_{\P}(q) = \sum_{i=0}^d h_{d-i} \binom{q+i}{d},$$ has all its coefficients $h_0,\ldots,h_d$ nonnegative. For $q\ge d$, we have the expansion $$\binom{q+i}{d}
= \binom{q}{d} \prod_{j=0}^{d-1}\biggl(1+\frac{i}{q-j}\biggr)
= \binom{q}{d} \negthinspace
\sum_{X\subseteq[0,d-1]} \frac{i^{|X|}}{\prod_{j\in X} (q-j)},$$ and so $$L_{\P}(q) = \binom{q}{d} \negthinspace
\sum_{X\subseteq[0,d-1]} \frac{g(|X|)}{\prod_{j\in X} (q-j)}$$ where $g(k)=\sum_{i=0}^d h_{d-i}i^k\ge 0$. Note that $\P$, and therefore $g(k)$ for each $k$, depend only on $m$ and $n$. Also note that $q/\prod_{j\in X} (q-j)$ is decreasing as a function of $q$ for $q\ge d$ and any non-empty $X\subseteq[0,d{-}1]$.
Since $q=\l\gcd(m,n)$, we conclude that there is a function $\alpha(m,n,\lambda)$ such that $$\begin{aligned}
M(m,\l n; n,\l m)& =
\binom{\l \gcd(m,n)}{(m{-}1)(n{-}1)} g(0)
\(1 + \alpha(m,n,\l)/\l\) \label{bit1}\\
\alpha(m,n,\l)&\ge 0
\text{~~for $q\ge d$}\label{bit2}\\[0.3ex]
\alpha(m,n,\l)&
\text{~~is decreasing in $\l$ for fixed $m,n$ and $q\ge d$}. \label{bit3}\end{aligned}$$ For $\l\ge n^5$, since $d=\tO(n^2)$, $$\binom{q}{d} = \frac{q^d}{d!}\(1+\tO(n^{-1})\).$$ Hence, by $$\label{ratio2}
\frac{M(m,2n^6; n,2mn^5)}{M(m,n^6; n,mn^5)}
= 2^{(m-1)(n-1)} \, \(1+\tO(n^{-1})\) \,
\frac{1+\tfrac12\alpha(m,n,2n^5)/n^5}{1+\alpha(m,n,n^5)/n^5}.$$ Comparing this to and noting from (\[bit2\],\[bit3\]) that $0\le \alpha(m,n,2n^5)\le\alpha(m,n,n^5)$, we conclude that $\alpha(m,n,n^5)=O(n^{5-b})$. This implies by that $\alpha(m,n,\lambda)=O(n^{5-b})$ for $\lambda\ge n^5$. Using this information about $\alpha(m,n,\l)$ with gives $$\frac{M(m,\lambda n; n,\lambda m)}{M(m,n^6; n,mn^5)}
= \biggl(\frac{\l}{n^5}\biggr)^{\!(m-1)(n-1)} \(1+O(n^{-b})\).$$ Combining this with Corollary \[biglambda\] for $\l=n^5$ shows that Corollary \[biglambda\] holds for all $\lambda\ge n^5$.
[99]{}
M. Beck and S. Robins, [*Computing the Continuous Discretely: Integer-Point Enumeration in Polyhedra*]{}, Springer, 2007.
M. Beck and D. Pixton, The Ehrhart polynomial of the Birkhoff polytope, *Discrete Comput. Geom.*, [**30**]{} (2003) 623–Ð637.
A. Békéssy, P. Békéssy and J. Komlós, Asymptotic enumeration of regular matrices, [*Studia Sci. Math. Hungar.*]{}, [**7**]{} (1972) 343–353.
E.A. Bender, The asymptotic number of nonnegative integer matrices with given row and column sums, [*Discrete Math.*]{}, [**10**]{} (1974) 345–353.
A. Barvinok, A. Samorodnitsky, and A. Yong, Counting magic squares in quasi-polynomial time, [arXiv]{}, math.CO/0703227.
R. Brualdi, [*Combinatorial Matrix Classes*]{}, Cambridge University Press, 2006.
E.R. Canfield, C. Greenhill, and B.D. McKay, Asymptotic enumeration of dense 0-1 matrices with specified line sums, [*J. Combin. Theory Ser. A*]{}, [**115**]{} (2008) 32–66.
E.R. Canfield and B.D. McKay, Asymptotic enumeration of dense 0-1 matrices with equal row sums and equal column sums, [*Electron. J. Combin.*]{}, [**12**]{} (2005) R29.
E.R. Canfield and B.D. McKay, The asymptotic volume of the Birkhoff polytope, [*Online J. Anal. Comb.*]{}, to appear.
E.R. Canfield and B.D. McKay, Asymptotic enumeration of highly oblong integer matrices with given row and column sums, in preparation.
Y. Chen, P. Diaconis, S.P. Holmes and J.S. Liu, Sequential Monte Carlo methods for statistical analysis of tables, [*J. Amer. Statist. Assoc.*]{}, [**100**]{} (2005) 109–120.
S.X. Chen and J.S. Liu, Statistical Applications of the Poisson-Binomial and conditional Bernoulli distributions, [*Statist. Sinica*]{}, [**7**]{} (1997) 875–892.
P. Diaconis and B. Efron, Testing for independence in a two-way table: new interpretations of the chi-square statistic (with discussion), [*Ann. Statist.*]{}, [**13**]{} (1985) 845–913.
P. Diaconis and A. Gangolli, Rectangular arrays with fixed margins, pages 15–41 of volume [**72**]{}, [*IMA Volumes on Mathematics and its Applications*]{}. (Proceedings of the conference on Discrete Probability and Algorithms, Minneapolis, MN, 1993.)
M. Dyer, R. Kannan, and J. Mount, Sampling contingency tables, [*Random Structures Algorithms*]{}, [**10**]{} (1997) 487–506.
C.J. Everett, Jr., and P.R. Stein, The asymptotic number of integer stochastic matrices, [*Discrete Math.*]{}, [**1**]{} (1971) 33–72.
M. Gail and N. Mantel, Counting the number of $r\times c$ contingency tables with fixed margins, [*J. Amer. Statist. Assoc.*]{}, [**72**]{} (1977) 859–862.
I.J. Good, *Probability and the Weighing of Evidence*, Charles Griffin, London, 1950.
I.J. Good, On the application of symmetric Dirichlet distributions and their mixtures to contingency tables, *Ann. Statist.*, [**4**]{} (1976) 1159–1189.
I.J. Good and J.F. Crook, The enumeration of arrays and a generalization related to contingency tables, [*Discrete Math.*]{}, [**19**]{} (1977) 23–45.
C.S. Greenhill and B.D. McKay, Asymptotic enumeration of sparse non-negative integer matrices with equal specified row and column sums, in preparation.
F. Greselin, Counting and enumerating frequency tables with given margins, [*Statistica & Applicazioni (Univ. Milano, Bicocca)*]{} [**1**]{} (2003) 87–104. Preprint available at\
`http://www.dimequant.unimib.it/pubblicazioni_prof/pubbl_2001.htm`.
W. Hoeffding, Probability inequalities for sums of bounded random variables, [*J. Amer. Statist. Assoc.*]{}, [**58**]{} (1963) 13–30.
R.B. Holmes and L.K. Jones, On uniform generation of two-way tables with fixed margins and the conditional volume Test of Diaconis and Efron, *Ann. Statist.*, [**24**]{} (1996) 64–68.
P. A. MacMahon, Combinatorial analysis. The foundation of a new theory, [*Philos. Trans. Roy. Soc. London Ser. A*]{}, [**194**]{} (1900) 361–386. (Paper 57 of Volume I of the [*Collected Papers*]{})
P. A. MacMahon, Combinations derived from $m$ identical sets of $n$ different letters, [*Proc. London Math. Soc. (2)*]{}, [**17**]{} (1918) 25–41. (Paper 89 of Volume I of the [*Collected Papers*]{}.)
B.D. McKay, Applications of a technique for labelled enumeration, [*Congressus Numerantium*]{}, [**40**]{} (1983) 207–221.
B.D. McKay, Asymptotics for 0-1 matrices with prescribed line sums, [*in* ]{} Enumeration and Design, (Academic Press, 1984) 225–238.
B.D. McKay and X. Wang, Asymptotic enumeration of 0-1 matrices with equal row sums and equal column sums, [*Linear Algebra Appl.*]{}, [**373**]{} (2003) 273–288.
B.D. McKay and N.C. Wormald, Asymptotic enumeration by degree sequence of graphs of high degree. [*European J. Combin.*]{}, [**11**]{} (1990) 565–580.
B. Morris, Improved bounds for sampling contingency tables, [*Random Structures and Algorithms*]{}, [**21**]{} (2002) 135–146.
E. Ordentlich and R.M. Roth, Two-dimensional weight-constrained codes through enumeration bounds, [*IEEE Trans. Inform. Theory*]{}, [**46**]{} (2000) 1292–1301.
R.C. Read, Some enumeration problems in graph theory, [*Doctoral Thesis*]{}, [*University of London*]{}, (1958).
R.P. Stanley, [*Enumerative Combinatorics. Vol. 1*]{}, corrected reprint of the 1986 original, Cambridge Studies in Advanced Mathematics, vol. 49, Cambridge University Press, Cambridge, 1997.
R.P. Stanley, [*Combinatorics and Commutative Algebra*]{}, volume 41 of the Progress in Mathematics series, Birkhäuser, 1983.
R.P. Stanley, Decompositions of rational convex polyhedra, [*Ann. Discrete Math.*]{}, [**6**]{} (1980) 333–342.
R.P. Stanley, Magic labelings of graphs, symmetric magic squares, systems of parameters, and Cohen-Macaulay rings, [*Duke Math. J.*]{}, [**43**]{} (1976), 511–531.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Person re-identification aims to identify the same pedestrian across non-overlapping camera views. Deep learning techniques have been applied for person re-identification recently, towards learning representation of pedestrian appearance. This paper presents a novel Contextual-Attentional Attribute-Appearance Network ($\rm CA^3Net$) for person re-identification. The $\rm CA^3Net$ simultaneously exploits the complementarity between semantic attributes and visual appearance, the semantic context among attributes, visual attention on attributes as well as spatial dependencies among body parts, leading to discriminative and robust pedestrian representation. Specifically, an attribute network within $\rm CA^3Net$ is designed with an Attention-LSTM module. It concentrates the network on latent image regions related to each attribute as well as exploits the semantic context among attributes by a LSTM module. An appearance network is developed to learn appearance features from the full body, horizontal and vertical body parts of pedestrians with spatial dependencies among body parts. The $\rm CA^3Net$ jointly learns the attribute and appearance features in a multi-task learning manner, generating comprehensive representation of pedestrians. Extensive experiments on two challenging benchmarks, i.e., Market-1501 and DukeMTMC-reID datasets, have demonstrated the effectiveness of the proposed approach.'
author:
- 'Jiawei Liu, Zheng-Jun Zha, Hongtao Xie, Zhiwei Xiong, Yongdong Zhang'
bibliography:
- 'sample-bibliography.bib'
title: |
CA$\bf{^3}$Net: Contextual-Attentional Attribute-Appearance Network\
for Person Re-Identification
---
[^1]
<ccs2012> <concept> <concept\_id>10002951.10003317.10003338.10003344</concept\_id> <concept\_desc>Information systems Combination, fusion and federated search</concept\_desc> <concept\_significance>500</concept\_significance> </concept> <concept> <concept\_id>10002951.10003317.10003338.10003346</concept\_id> <concept\_desc>Information systems Top-k retrieval in databases</concept\_desc> <concept\_significance>500</concept\_significance> </concept> <concept> <concept\_id>10002951.10003317.10003371.10003386.10003387</concept\_id> <concept\_desc>Information systems Image search</concept\_desc> <concept\_significance>300</concept\_significance> </concept> </ccs2012>
[^1]: Corresponding author: Zheng-Jun Zha ([email protected])
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study bounds on Higgs boson masses from perturbative unitarity in the Georgi-Machacek model, whose Higgs sector is composed of a scalar isospin doublet, a real and a complex isospin triplet fields. This model can be compatible with the electroweak precision data without fine tuning because of the imposed global $SU(2)_R$ symmetry in the Higgs potential, by which the electroweak rho parameter is unity at the tree level. All possible two-body elastic-scattering channels are taken into account to evaluate the S-wave amplitude matrix, and then the condition of perturbative unitarity is imposed on the eigenvalues to obtain constraint on the Higgs parameters. Masses of all scalar bosons turn out to be bounded from above, some of which receive more strict upper bounds as compared to that in the standard model (712 GeV). In particular, the upper bound of the lightest scalar boson, whatever it would be, is about 270 GeV.'
date:
-
-
-
title: GMM
---
-1.5cm -0.8cm -0.8cm
\#1\#2\#3\#4[[\#1]{} [**\#2**]{} (\#4) \#3]{}
=10000
[ ]{}
0.8cm
[**.3cm Mayumi Aoki$^1$[^1] and Shinya Kanemura$^2$[^2]** ]{}
[***1: ICRR, University of Tokyo, Kashiwa 277-8582, Japan*** ]{}
[***2: Department of Physics, University of Toyama,\
3190 Gofuku, Toyama 930-8555, Japan*** ]{}
2.0cm
1.0cm [**PACS index : 12.60.Fr, 14.80.Cp** ]{} 1.0cm [**Keywords : Non-standard model, Partial-wave unitarity, Higgs boson mass bounds**]{}
Introduction
============
The nature of electroweak symmetry breaking remains unknown at the present status of our knowledge for high energy physics. In the standard model (SM), a scalar isospin doublet field, the Higgs field, is introduced to be responsible for spontaneous breakdown of electroweak gauge symmetry. Its vacuum expectation value (VEV) triggers the symmetry breaking, so that it provides origins of masses of weak bosons via the Higgs mechanism, and also does those of quarks and charged leptons via Yukawa interaction. Although the SM Higgs sector is simple, the Higgs sector could have a more complicated structure in the actual world. In particular, when the Higgs sector would play an additional role to explain phenomena which the SM cannot, it should necessarily be an extended form from the SM one. Therefore, experimental detection of the Higgs boson and precision measurements of its properties are extremely important not only to confirm our basic idea of electroweak symmetry breaking but also to determine details of the Higgs sector and further to outline the structure of new physics.
In constructing an extended Higgs sector, there are two important requirements from current experimental data. First of all, the data indicate that the electroweak rho parameter ($\rho$) is very close to unity. Second, flavor of quarks and charged leptons is (approximately) conserved in the neutral current. In the SM, these two conditions are satisfied respectively by the custodial symmetry which ensures $\rho=1$ at the tree level, and by the Glashow-Iliopoulos-Maiani (GIM) mechanism which prohibits the tree-level flavor changing neutral current (FCNC). Needless to say that these experimental requirements must be respected in extended Higgs models which would appear in the low energy effective theory of a more fundamental theory beyond the SM.
Extension of the SM Higgs sector can be considered by including additional scalar isospin singlets, doublets and higher multiplets. It is known that additional singlets and doublets keep $\rho=1$ at the tree level [@Gildener:1976ih]. Radiative corrections can slightly deviate the rho parameter from unity, corresponding to explicit violation of the custodial symmetry in the dynamics in the loop. On the other hand, extension with higher multiplets such as triplets is usually problematic, predicting the rho parameter to be explicitly different from unity already at the tree level [@Gunion:1992hs]. One way to avoid this problem is to make a fine-tuning on the size of vacuum expectation values of the triplet fields; i.e., to set tiny values on them. Another possibility is to impose the custodial symmetry to the Higgs sector, so that the rho parameter is predicted to be unity at the tree level. In 1985 Georgi and Machacek proposed such a model with one real triplet ($Y$=0) and one complex triplet ($Y$=2) in addition to the Higgs doublet [@Georgi:1985nv]. Chanowitz and Golden have explicitly constructed the Higgs potential of this model [@Chanowitz:1985ug]; i.e., imposing the custodial $SU(2)_{\rm V}$ symmetry to the potential, VEVs of all the isospin triplets become common, and then the tree-level value of the rho parameter is unity. They also have shown that the quantum correction from the scalar sector is stabilized by such a global symmetry, so that the rho parameter is corrected at the loop level only due to explicit $SU(2)_{\rm V}$ violation in the other sectors such as hypercharge interaction and Yukawa interaction, just like in the SM. Several phenomenological studies have been done on this model in Refs. [@Gunion:1989ci; @Vega:1989tt; @Gunion:1990dt; @Godbole:1994np; @Cheung:1994rp; @Akeroyd:1998zr; @Haber:1999zh; @Cheung:2002gd].
Generally in extended Higgs models, there are many free parameters in the Higgs potential, which spoil predictive power of the model. Hence, it is important to clarify allowed regions in the parameter space not only by using current experimental data but also by investigating theoretical consistencies such as perturbative unitarity [@Lee:1977yc; @Dicus:1992vj], vacuum stability and triviality [@Lindner:1985uk]. This kind of study has been often developed to constrain parameters of the Higgs sector in the context of the two-Higgs-doublet model [@Kanemura:1993hm; @Ginzburg:2005dt; @Komatsu:1981xh; @Nie:1998yn], and in a specific triplet model [@Forshaw:2003kh].
In this paper, we study bounds on Higgs boson masses from perturbative unitarity in the Georgi-Machacek (GM) model. The Higgs potential respects the global $SU(2)_R$ symmetry, so that the custodial $SU(2)_{\rm V}$ symmetry remains after the electroweak symmetry breaking ($SU(2)_{L} \otimes SU(2)_R \to SU(2)_{\rm V}$). There are ten physical scalar states, which can be expressed by a $SU(2)_{\rm V}$ 5-plet $(H_5^{++}, H_5^+, H_5^0, H_5^-, H_5^{--})$, a 3-plet $(H_3^+, H_3^0, H_3^-)$ and two singlets $\tilde{H}_1^0$ and $\tilde{H}^{0'}_{1}$ [@Georgi:1985nv]. The scalar components in the same multiplet are degenerate in mass at the tree level. In the Higgs potential of the GM model, explicit $Z_2$ violation can only appear in the trilinear scalar interaction, but they must be forbidden to avoid excessive magnitudes for masses of neutrinos. Neglecting such terms by imposing the $Z_2$ symmetry, all Higgs boson masses in this model are described in terms of the VEV, the mixing angles and the dimension-less coupling constants $\lambda_i$ in the Higgs potential. This situation is somewhat similar to the two-Higgs-doublet model with the discrete $Z_2$ symmetry [@Glashow:1976nt], in which perturbative unitarity gives upper bounds on all the Higgs boson masses [@Kanemura:1993hm].
In our analysis, all possible two-body elastic scattering channels (91-channels) are taken into account to evaluate the S-wave amplitude matrix in the GM model. Constraints on the Higgs parameters are obtained by imposing the condition of partial wave unitarity on the eigenmatrix of the S-wave amplitude. Masses of all scalar bosons turn out to be bounded from above, some of which receive much stronger bounds as compared to that in the SM (712 GeV). In particular, the mass of at least one of the charged Higgs bosons should be less than about 400 GeV. At least one of the neutral Higgs boson is lighter than 322 GeV. Furthermore, the upper bound of the lightest scalar boson, whatever it would be, can be about 269 GeV. We also find that by using the experimental constraints from $Zb\bar b$ results [@Haber:1999zh], the combined upper bound for the lightest Higgs boson is lower than 269 GeV, depending on what the lighest is. Therefore, the model can be well testable at current and future collider experiments.
In Sec. 2, a brief review of the GM model is given. The transition matrix for two-body elastic scatterings is calculated in the high-energy limit, and its eigenmatrix is obtained in Sec. 3. In Sec. 4, the condition of S-wave unitarity is imposed for the eigenmatrix of the transition matrix, and bounds on the Higgs boson masses are evaluated. Conclusions are presented in Sec. 5.
Georgi-Machacek Model
=====================
The GM model contains a complex $SU(2)_L$ doublet field $\phi$ ($Y$=1), a real $SU(2)_L$ triplet field $\xi$ ($Y$=0) and a complex $SU(2)_L$ triplet field $\chi$ ($Y$=2) [@Georgi:1985nv], and respects the global $SU(2)_R$ symmetry in the Higgs potential [@Chanowitz:1985ug] . They can be described by the form of $SU(2)_L\otimes SU(2)_R$ multiplets $\Phi$ and $\Delta$ in the potential; =(
[cc]{} \^[0]{}&\^[+]{}\
\^- & \^0
), =(
[ccc]{} \^[0]{}&\^+&\^[++]{}\
\^- & \^0 &\^+\
\^[–]{} &\^- &\^0
) , where $\phi=(\phi^+, \phi^0)^T$, $\xi=(\xi^+, \xi^0, \xi^-)^T$ and $\chi=(\chi^{++}, \chi^+, \chi^0)^T$, and $\phi^-=-(\phi^+)^\ast$, $\xi^-=-(\xi^+)^\ast$ and $\chi^-=-(\chi^+)^\ast$ [@Gunion:1989ci]. The most general Higgs potential is given by V&=&m\_1\^2[Tr]{}(\^)+m\_2\^2[Tr]{}(\^)+\_1[Tr]{}(\^)\^2+\_2[Tr]{}(\^)\^2+\_3[Tr]{}(\^)[Tr]{}(\^)\
&&+\_4[Tr]{}(\^\^)+\_5[Tr]{}(\^)[Tr]{}(\^T\_iT\_j)\
&&+\_1[Tr]{}(\^)\_P\^[ij]{}+\_2[Tr]{}(\^T\_i T\_j)\_P\^[ij]{} , \[potential\] where $\tau_i$ are the $2\times 2$ Pauli matrices and \_P=P\^P, P=(
[ccc]{}-1/&i/&0\
0&0&1\
1/&i/&0
) .
The neutral components of the doublet and the real and the complex triplets have the VEVs, $v_\phi$, $v_\xi$, and $v_\chi$, respectively, which are defined as \^0&=& ,\
\^0&=&v\_+\_r\^0 ,\
\^0&=&v\_+ . After electroweak symmetry breaking, the custodial $SU(2)_{\rm V}$ symmetry remains in the Higgs sector $(SU(2)_L\times SU(2)_R\to SU(2)_{\rm V})$, by which the real and complex triplets have the same VEV, $v_\Delta\equiv v_\xi=v_\chi$. Consequently this leads to $\rho=1$ at the tree level [@Chanowitz:1985ug]. In this case the VEVs are constrained as $v^2=v_\phi^2 + 8v_\Delta^2$, where $v=(\sqrt{2}
G_F)^{-\frac{1}{2}} \simeq 246$ GeV. Therefore, differently from usual triplet models, $v_\Delta$ can be of order 100 GeV in this model without explicit inconsistency with the experimental value of the rho parameter. It is convenient to introduce the doublet-triplet mixing angle $\theta_H$, \_H= . The experimental constraint on $\theta_H$ is discussed in Ref. [@Haber:1999zh]. In the potential Eq.(\[potential\]), the last two terms with the coupling constants $\mu_1$ and $\mu_2$ explicitly violate the discrete $Z_2$ symmetry under the transformation of $\Phi\to \Phi$ and $\Delta \to -\Delta$. Without the $Z_2$ symmetry, the model is allowed to have the mass terms for the neutrinos by assigning of lepton number $-2$ to the complex triplet field, i(h\_)\_[ab]{}\_[La]{}\^TC\_2\_[Lb]{}+[h.c.]{}, where $\hat \chi=\frac{\tau^i}{2}(P^\dag\chi)^i$. In order to generate the tiny neutrino masses the Yukawa coupling $h_\nu$ should be fine-tuned to be very small as $h_\nu\sim {\cal O}(10^{-12})$ for the triplet VEV of order 100 GeV. Since we would like to avoid such fine tuning with respect to the neutrino masses, we require the discrete $Z_2$ symmetry in the Higgs potential and prohibit the last two terms in Eq.(\[potential\])[^3]. Therefore, quarks and leptons couple to the $SU(2)_L$ doublet field $\Phi$ in the same way as the SM Yukawa coupling, but do not to the triplet $\Delta$ at the tree level. Because all the masses of quarks and leptons are obtained from the VEV in $\Phi$, we do not have to worry about FCNC, and it is expected to appear at most at the same level as that in the SM. This property of the coupling to fermions would give an additional constraint on the value of the doublet-triplet mixing angle $\theta_H^{}$ by $\tan\theta_H^{}< {\cal O}(1)$, since large values of $\tan\theta_H^{}$ $(\gg 1)$ imply that the top-Yukawa coupling is much greater than ${\cal O}(1)$.
In the GM model, there are ten physical states in the Higgs sector, which are classified as a 5-plet ($H_5^{++}, H_5^{+}, H_5^{0}, H_5^{-}, H_5^{--}$), a 3-plet ($H_3^{+}, H_3^{0}, H_3^{-}$), and two singlets $H_1^0$ and $H_1^{0'}$ under the custodial $SU(2)_{\rm V}$ symmetry. These are given in terms of the original component fields and the doublet-triplet mixing angle $\theta_H$ as [@Gunion:1989ci] &&H\_5\^[++]{}=\^[++]{} ,\
&&H\_5\^[+]{}=(\^[+]{}-\^[+]{})/ ,\
&&H\_5\^[0]{}=(2\_r\^0-\^[0]{}\_r)/ ,\
&&H\_3\^[+]{}=(\^++\^+)/-\^+,\
&&H\_3\^[0]{}=i(-\^0\_i+\^0\_i) ,\
&&H\_1\^[0]{}=\_r\^0,\
&&H\_1\^[0’]{}=(\^0\_r+\_r\^0)/. The 5-plet components do not include the component fields from the isospin doublet field $\Phi$, so that the states of the 5-plet do not couple to the fermions at the tree level. On the other hand, the 3-plet fields can couple to the fermions. Because of invariance under the custodial $SU(2)_{\rm V}$ symmetry, states in the different multiplet cannot mix each other.
All members in the same $SU(2)_{\rm V}$ multiplet are degenerate in mass at the tree level. The masses of the 5-plet and the 3-plet are respectively given by m\_[H\_5]{}\^2&=&(\_4\^2\_H-\_5\^2\_H) v\^2 , \[m5\]\
m\_[H\_3]{}\^2&=&-v\^2 . On the other hand, two $SU(2)_{\rm V}$ singlets can mix, and the mass matrix \^2\_[H\_1\^0,H\_1\^[0’]{}]{}=(
[cc]{}8\^2\_H\_1 &\_H\_H (2\_3+\_5)\
\_H\_H (2\_3+\_5) &\^2\_H(3\_2+\_4)
)v\^2 \[masses\] is diagonalized by introducing the mixing angle $\alpha$. The eigenvalues correspond to the masses $m_{\tilde H_1^0}$ and $m_{\tilde H_1^{0'}}$ for the mass eigenstates ${\tilde H_1^0}$ and ${\tilde H_1^{0'}}$.
From Eqs.(\[m5\]) - (\[masses\]), the quartic couplings $\lambda_i$ are expressed in terms of the masses and the mixing angles as \_1&=&(m\_[H\_1\^0]{}\^2\^2+m\_[H\_1\^[0’]{}]{}\^2\^2)/(8v\^2\^2\_H) , \[lambda1\]\
\_2&=&(m\_[H\_1\^0]{}\^2\^2+m\_[H\_1\^[0’]{}]{}\^2\^2-m\_[H\_5]{}\^2+3m\_[H\_3]{}\^2\^2\_H)/(3v\^2\^2\_H) , \[lambda2\]\
\_3&=&(m\_[H\_1\^[0’]{}]{}\^2-m\_[H\_1\^0]{}\^2)/(v\^2\_H\_H)+m\_[H\_3]{}\^2/v\^2 , \[lambda3\]\
\_4&=&(m\_[H\_5]{}\^2-3m\_[H\_3]{}\^2\^2\_H)/(v\^2\^2\_H) ,\
\_5&=&-2m\_[H\_3]{}\^2/v\^2 . \[lambda5\] \[lambda\]
The $SU(2)_{\rm V}$ 3-plet fields receive the constraints from the current data of $Z\to b\bar b$, $B_0-\bar B_0$ and $K_0-\bar K_0$ mixings [@Kundu:1995qb; @Chakraverty:1995zw]. These data give bounds on the mass $m_{H_3}^{}$ with the mixing angle $\theta_H$. The most stringent experimental constraint comes from $Z \to b \bar b$. The mass $m_{H_3}^{}$ is constrained to be smaller than 1 (0.5) TeV for $\tan\theta_H~<$ 2 (1).
Although the 5-plet fields do not couple to the fermions, the singly-charged state in the 5-plet has a characteristic coupling of $H_5^\pm W^\mp Z$, which only appears beyond the tree level in multi-Higgs-doublet models [@Grifols:1980uq]. Experimental confirmation of a sizable coupling of $H_5^\pm W^\mp Z$ with $\rho \simeq 1$ should be a strong indication for the GM model[@Mukho:1990]. This coupling is testable via the process $p\bar p\to W^\pm H^\mp$ at the Fermilab Tevatron [@Cheung:2002gd], also via $pp\to W^{\pm\ast} Z^\ast X \to H^\pm X$ [@Asakawa:2005gv] and the decay process $H^\pm\to W^\mp Z$ [@Kanemura:1997ej] at the CERN LHC, and further via the processes $e^+e^-\to W^\mp
H^\pm$ [@Godbole:1994np; @Cheung:1994rp; @Kanemura:1999tg] and $e^+e^-\to \nu \bar \nu W^{\pm\ast} Z^\ast \to \nu \bar \nu
H^\pm$ [@Kanemura:2000cw] at the ILC. Another striking feature of models with complex isospin-triplets, such as the GM model, the left-right symmetric model, the littlest Higgs model, and some models motivated by neutrino masses, is the appearance of doubly-charged states $H^{\pm\pm}$. At hadron colliders, such doubly-charged Higgs bosons are studied via the pair production mechanism [@Akeroyd:1998zr; @Gunion:1989in; @Dion:1998pw] as well as the single production mechanism [@Dion:1998pw; @Akeroyd:2005gt] and the $W$-boson fusion mechanism [@Vega:1989tt; @Huitu:2000ut]. They can also be investigated at the ILC and its $e^-e^-$, $e^-\gamma$ and $\gamma\gamma$ option in various scenarios [@Gunion:1998ii].
The S-matrix for two-body elastic scatterings
=============================================
In this section, we calculate the transition matrix of elastic scatterings of two scalar-boson states in the GM model. The transition matrix $T(\varphi_1 \varphi_2 \to \varphi_3 \varphi_4)$ is equivalent to the S-wave amplitude $\langle \varphi_3 \varphi_4 | a^0 | \varphi_1 \varphi_2
\rangle$ at high energies ($\sqrt{s}\gg m_W^{2}$), where $\varphi_i$ represent longitudinally-polarized weak bosons or physical Higgs bosons of the model. The condition of partial wave unitarity is given for the S-wave amplitude matrix by [@Gunion:1992hs; @Lee:1977yc] | \_3 \_4 | a\^0 |\_1 \_2 | < . \[PU\] We employ this condition in the high energy limit to constrain the model parameters in the next section. Thanks to the equivalence theorem [@Cornwall:1974km], the S-matrix elements in which longitudinally-polarized weak bosons are in initial and final states are equivalent to those in which these weak bosons are replaced by the corresponding Nambu-Goldstone bosons in the high energy limit [@Lee:1977yc]. In addition, in this limit, only quartic couplings (scalar contact interactions) of the Higgs-Goldstone couplings are relevant to the unitarity conditions, which can be translated into the bounds on the related Higgs-boson masses after Eq.(\[PU\]) is imposed. Therefore, we here evaluate the matrix $\langle \varphi_3 \varphi_4 | a^0 |\varphi_1 \varphi_2 \rangle$ taking into account all possible two-body scalar channels in the high energy limit, and obtain all the eigenstates and the eigenvalues.
Under $O(4) (\simeq SU(2)_L \otimes SU(2)_R)$, the field components of $\phi$, $\chi$ and $\xi$ are expressed by a and a representations as \_D &=&( \_1, \_2, \_r\^0, \_i\^0 ),\
\_T &=&( \_1, \_2, \_3, \_4, \_1, \_2, \_r\^0, \_i\^0, \_r\^0 ), where $\phi^+ = (\omega_1 + i \omega_2)/\sqrt{2}$, $\phi^{0}=(\phi_r^0 + i \phi_i^0)/\sqrt{2}$, $\chi^{++}=(\chi_1 +
i \chi_2)/\sqrt{2}$, $\chi^{+}=(\chi_3 + i \chi_4)/\sqrt{2}$, $\chi^{0}=(\chi_r^0 + i \chi_i^0)/\sqrt{2}$ and $\xi^{+}=(\xi_1 + i \xi_2)/\sqrt{2}$. We consider all possible two-body scattering channels ($\Psi_{a} \Psi_{b}
\to \Psi_{c} \Psi_{d}$) not only for the neutral two-body states as initial and final states but also for the singly-, the doubly-, the triply- and the quadruply-charged two-body states. There are totally 91 initial (or final) two-body states, in which 25 are the neutral, 36 are singly-charged, 22 are doubly-charged, 6 are triply-charged, and the last 2 are the quadruply-charged states. We construct the 91 $\times$ 91 transition matrix of high-energy S-wave amplitudes, and then calculate their eigenvalues.
The initial (final) two-body states can be treated separately as $\Psi_D
\Psi_D$, $\Psi_T \Psi_T$ and $\Psi_D \Psi_T$. The high-energy S-wave amplitudes are block-diagonalized by the electric charge and also the discrete $Z_2$ symmetry ($\Phi \to \Phi$ and $\Delta\to -\Delta$). Each submatrix with respect to the $\Psi_D\Psi_D$ or $\Psi_T\Psi_T$ states can also be classified by irreducible decomposition of direct products of the representations for $O(4)$ as &=&(1)\_D\^ (9) (6),\
&=&(1)\_T\^ (44)(36). The only singlet and symmetric representations, (1)\_[a]{}&=& \_[k=1]{} (\^k\_a)\^2,\
(s)\^[ij]{} &=& \^i\_[a]{} \^j\_a - \_[k=1]{} (\^k\_a)\^2, contribute to the scatterings of our interests, where $a=D\; {\rm or}\;
T\, ;\, i,j=1{\rm -}4\; {\rm and}\; s=9 \;{\rm for}\; a=D\;, {\rm or} \;
i,j=1{\rm -}9\; {\rm and} \; s=44\;{\rm
for}\; a=T$, and $N_D^{}=4$ and $N_T^{}=9$. For the $\Psi_D \Psi_T$ states which should be of the $\underline{4} \otimes \underline{9}$ representation, there is no singlet representation so that $O(4)$ cannot help for the classification. Furthermore, several additional discrete transformations can be used to further classify the states, which will be defined below.\
\
We outline further classification of the decomposed irreducible states for the case of the neutral 25 two-body channels (4 for $\Psi_D\Psi_D$, 11 for $\Psi_T\Psi_T$ and 10 for $\Psi_D\Psi_T$). For $\Psi_D\Psi_D$ states, we have the singlet state $(1)_D^{}$ and the three neutral elements of $(9)^{ij}$ ($(9)^{33}$, $(9)^{44}$ and $(9)^{34}$), in which the $C$ parity separates $(9)^{34}$ from the other states. After taking appropriate linear combination, we obtain two separate states under the transformation of $\phi_r \to \phi_i$ and $\phi_i \to
-\phi_r$ as $((9)^{33}\pm (9)^{44})/\sqrt{2}$. Thus four eigenstates of the transition matrix for the neutral $\Psi_D\Psi_D \to \Psi_D\Psi_D$ channels are obtained [@Lee:1977yc].
Next, we consider $\Psi_T\Psi_T\to\Psi_T\Psi_T$ scatterings in which both the initial and final states are electrically neutral. In addition to the singlet state $(1)_T$, we have 10 neutral states from $(44)^{ij}$, in which { $(44)^{11}+(44)^{22}$, $(44)^{33}+(44)^{44}$, $(44)^{55}+(44)^{66}$, $(44)^{77}$, $(44)^{88}$ } are the diagonal element states ($i=j$), and {$(44)^{35}+(44)^{46}$, $(44)^{36}-(44)^{45}$, $(44)^{78}$, $(44)^{79}$, $(44)^{89}$} are the off-diagonal element states ($i\neq j$). Among the diagonal element states, the linear combination $(44)^{77}-(44)^{88}$ has different property under the transformation of $\chi_r \to \chi_i$ and $\chi_i \to - \chi_r$. Then, linear combinations { $(44)^{11}+(44)^{22}+(44)^{77}+(44)^{88}$, $(44)^{33}+(44)^{44}+(44)^{55}+(44)^{66}$ } and { $(44)^{11}+(44)^{22}-(44)^{77}-(44)^{88}$, $(44)^{33}+(44)^{44}-(44)^{55}-(44)^{66}$} show different property under the transformation of $\chi^{++}\chi^{--} \leftrightarrow \chi^0\chi^0$, and $\chi^{+}\chi^{-} \leftrightarrow \xi^+\xi^-$. The first two states have completely the same property as that of the singlet state $(1)_T$, so that the appropriate linear combination of these three states give the three eigenstates. For the off-diagonal element states, we can separate them by using the $C$ parity and the transformation of $\xi \to -\xi$, so that these states are block-diagonalized to two $2 \times 2$ submatrices and one singlet. By diagonalizing remained $2 \times 2$ matrices, we obtain all the eigenstates for the $\Psi_T\Psi_T \to \Psi_T\Psi_T$ channels.
In order to diagonalize all the $\Psi_D\Psi_D$ and $\Psi_T\Psi_T$ states, we take linear combinations of the eigenstates of $\Psi_D\Psi_D$ and $\Psi_T\Psi_T$ that have similar transformation properties. Consequently, all the $\Psi_D\Psi_D$ and $\Psi_T\Psi_T$ states are completely separated, and the eigenvalues of the transition matrix for these channels are obtained.
Finally, classifying the ten $\Psi_D\Psi_T$ states by using the $C$ transformation as well as some discrete transformations in a similar way to above, we completely diagonalized $25 \times 25$ scattering matrix for the electrically neutral two-body states.
The neutral states, $|A^0_i\rangle (i=1\sim 25)$, which give (at most $2\times 2$) block-diagonal transition matrices are found as below: |A\^0\_[1]{}&=&(2\^+\^-+\_i\^0\_i\^0+\_r\^0\_r\^0) /2 ,\
|A\^0\_2&=&(2\^[++]{}\^[–]{}+2\^[+]{}\^[-]{}+2\^[+]{}\^[-]{}+\_i\^0\_i\^0+\_r\^0\_r\^0+\_r\^0\_r\^0)/3 ,\
|A\^0\_3&=&(2\^+\^–\_i\^0\_i\^0-\_r\^0\_r\^0) /2 ,\
|A\^0\_4&=&(- 2\^[++]{}\^[–]{}+\_i\^0\_i\^0+ \_r\^0\_r\^0)/2 ,\
|A\^0\_5&=&(\_i\^0\_i\^0-\_r\^0\_r\^0)/2 ,\
|A\^0\_6&=&(\^+\^-+ \^-\^++\_r\^0\_r\^0)/2 ,\
|A\^0\_7&=&\_r\^0\_i\^0 ,\
|A\^0\_8&=&(\^+\^– \^-\^+-i\_r\^0\_i\^0)/(2i) ,\
|A\^0\_[9]{}&=&(\_i\^0\_i\^0- \_r\^0\_r\^0)/2 ,\
|A\^0\_[10]{}&=&(2\^[++]{}\^[–]{}-\^+\^–\^+\^-+\_i\^0\_i\^0+ \_r\^0\_r\^0-2\_r\^[0]{}\_r\^[0]{})/3 ,\
|A\^0\_[11]{}&=&(2\^[++]{}\^[–]{} -4\^[+]{}\^[-]{}-4\^[+]{}\^[-]{}+\_r\^0\_r\^0+\_i\^0\_i\^0+4\_r\^0\_r\^0)/6 ,\
|A\^0\_[12]{}&=&(\_r\^0\_r\^0-\_r\^0\_i\^0)/ ,\
|A\^0\_[13]{}&=&(\^+\^-+ \^-\^+-\_r\^0\_r\^0)/2 ,\
|A\^0\_[14]{}&=&(\^+\^– \^-\^++ i\_r\^0\_i\^0)/(2i) ,\
|A\^0\_[15]{}&=& \_r\^0\_i\^0 ,\
|A\^0\_[16]{}&=&{ 2(\^+\^- +\^-\^++ \^+\^-+\^-\^+)-(\_r\^0\_r\^0+\_i\^0\_i\^0) -2 \_r\^0\_r\^0}/3 ,\
|A\^0\_[17]{}&=&( \_r\^0\_r\^0-\_i\^0\_i\^0)/ ,\
|A\^0\_[18]{}&=&( \_i\^0\_r\^0+\_r\^0\_i\^0)/ ,\
|A\^0\_[19]{}&=&{2(\^+\^- - \^-\^+- \^+\^-+\^-\^+) +i(\_i\^0\_r\^0-\_r\^0\_i\^0)-4i\_i\^0\_r\^0}/(6i) ,\
|A\^0\_[20]{}&=&{ \^+\^- + \^-\^++\^+\^-+\^-\^++(\_r\^0\_r\^0+\_i\^0\_i\^0)+\_r\^0\^0}/3 ,\
|A\^0\_[21]{}&=&{\^+\^- - \^-\^+- \^+\^-+\^-\^++i(\_r\^0\_i\^0-\_i\^0\_r\^0)+i\_i\^0\_r\^0}/(3i) ,\
|A\^0\_[22]{}&=&{\^+\^- - \^-\^+-\^+\^-+\^-\^++2i( \_i\^0\_r\^0-\_r\^0\_i\^0)+4i\_i\^0\_r\^0}/(6i) ,\
|A\^0\_[23]{}&=&{\^+\^- + \^-\^++\^+\^-+\^-\^+-2( \_r\^0\_r\^0+\_i\^0\_i\^0)+4 \_r\^0\_r\^0)}/6 ,\
|A\^0\_[24]{}&=&(\^+\^- + \^-\^+- \^+\^–\^-\^+)/2 ,\
|A\^0\_[25]{}&=&(\^+\^- - \^-\^++\^+\^–\^-\^+)/(2i) . The state $(1)_D$ and the three neutral linear-combined states from $(9)^{ij}$ respectively correspond to $|A^0_{1}\rangle$ and {$|A^0_{3}\rangle$, $|A^0_{5}\rangle$ and $|A^0_{7}\rangle$}. The state $(1)_T$ and the linear combined states from $(44)^{ij}$ correspond to $|A^0_{2}\rangle$ and $|A^0_{i}\rangle$ $(i=4,6,8,11-15)$, respectively. The first eight states block-diagonalize the transition matrix to four $2 \times 2$ submatrices, and the other seventeen states give eigenstates.\
\
There are eighteen singly-charged states with the electric charge $+1$ among all the two-body states. The charge conservation ensures that these states are composed of a subset among all the states with various electric charges. The corresponding high-energy transition matrix is consequently (block-) diagonalized by the following states: |A\_1\^+&=&(\^+\_r\^0+\^+\_i\^0)/ ,\
|A\_2\^+&=&(\^[++]{}\^-+ \^+\_r\^0+\^+\_i\^0)/2 ,\
|A\_3\^+&=&(\^+\_r\^0-\^+\_i\^0)/ ,\
|A\_4\^+&=&(\^[++]{}\^-+ \^+\_r\^0-\^+\_i\^0)/2 ,\
|A\_5\^+&=&(\^[++]{}\^- -\^+\_r\^0-\^+\_i\^0+2\^+ \_r\^0)/2 ,\
|A\_6\^+&=&(\^[++]{}\^- - \^+\_r\^0+\^+\_i\^0+2\^+ \_r\^0)/2 ,\
|A\_7\^+&=&(\^[++]{}\^–\^+\_r\^0-\^+\_i\^0-2\^+\_r\^0)/2 ,\
|A\_8\^+&=&(\^[++]{}\^- -\^+\_r\^0+\^+\_i\^0-2\^+\_r\^0) /2 ,\
|A\_[9]{}\^+&=&(\^+\_r\^0+\^+\_i\^0)/ ,\
|A\_[10]{}\^+&=&(\^+\_r\^0-\^+\_i\^0)/ ,\
|A\_[11]{}\^+&=& (\^[++]{}\^–2\^+\_r\^0-\^+\_r\^0-\^+\_i\^0-\^+\_r\^0+\^+\_i\^0)/3 ,\
|A\_[12]{}\^+&=& (\^+\_r\^0+\^+\_i\^0+\^+\_r\^0+\^+\_i\^0)/ ,\
|A\_[13]{}\^+&=& (\^+\_r\^0-\^+\_i\^0+ \^+\_r\^0 -\^+\_i\^0)/ ,\
|A\_[14]{}\^+&=& (4\^[++]{}\^-+4\^+\_r\^0-\^+\_r\^0 -\^+\_i\^0-\^+\_r\^0+\^+\_i\^0)/6 ,\
|A\_[15]{}\^+&=& (\^+\_r\^0+\^+\_i\^0-\^+\_r\^0+\^+\_i\^0)/2 ,\
|A\_[16]{}\^+&=& (\^+\_r\^0+\^+\_i\^0- \^+\_r\^0-\^+\_i\^0)/ ,\
|A\_[17]{}\^+&=& (\^+\_r\^0-\^+\_i\^0- \^+\_r\^0 +\^+\_i\^0)/ ,\
|A\_[18]{}\^+&=& (2\^[++]{}\^–\^+\_r\^0+\^+\_r\^0 +\^+\_i\^0+\^+\_r\^0-\^+\_i\^0)/3 . The eighteen singly-charged states with the electric charge $-1$ can be obtained by the $C$ transformation for the above states with the charge $+1$.\
\
There are eleven doubly-charged two-body states with the electric charge $+2$. We can decompose the subset of the transition matrix for these states to at most $2 \times 2$ matrices by the following linear combination; |A\_1\^[++]{}&=&\^+\^+,\
|A\_2\^[++]{}&=&(\^+\^+ -\^[++]{}\_r\^0)/ ,\
|A\_3\^[++]{}&=&\^+\^+,\
|A\_4\^[++]{}&=&(\^[++]{}\_r\^0 -\^[++]{}\_i\^0)/ ,\
|A\_5\^[++]{}&=&\^+\^+,\
|A\_6\^[++]{}&=&(\^[++]{}\_r\^0 +\^[++]{}\_i\^0)/ ,\
|A\_7\^[++]{}&=&(\^+\^+ +\^[++]{}\_r\^0)/ ,\
|A\_8\^[++]{}&=&(\^+\^+ +\^+\^+ +\^[++]{}\_r\^0)/ ,\
|A\_9\^[++]{}&=&(\^+\^+ -\^+\^+ -\^[++]{}\_i\^0)/ ,\
|A\_[10]{}\^[++]{}&=&(\^+\^+ +\^+\^+ -2\^[++]{}\_r\^0)/ ,\
|A\_[11]{}\^[++]{}&=&(\^+\^+ - \^+\^+ +2\^[++]{}\_i\^0)/ . The corresponding doubly-charged two-body states with the charge $-2$ can be obtained by $C$ transformation in the above states with the charge $+2$.\
\
There are three triply-charged states with the electric charge $+3$, and the subset of the transition matrix for the initial and final states can be diagonalized by the following eigenstates as |A\_1\^[+++]{}&=&\^[++]{}\^+,\
|A\_2\^[+++]{}&=&\^[++]{}\^+,\
|A\_3\^[+++]{}&=&\^[++]{}\^+. All the eigenstates with the opposite electric charge can be obtained by the $C$ transformation of these eigenstates with the charge $+3$.\
\
Finally, we have only one quadruply-charged state for each electric charge of $+4$ and $-4$, |A\_1\^[++++]{}&=&\^[++]{}\^[++]{},\
|A\_1\^[—-]{}&=&\^[–]{}\^[–]{}.
\
In summary, the transition matrix $T$ has been block-diagonalized as T = , \[Tmatrix\] where block-diagonal transition submatrices for the neutral, the singly-charged, the doubly-charged, the triply-charged, and the quadruply-charged two-body states, $T^0$, $T^\pm$, $T^{\pm\pm}$, $T^{\pm\pm\pm}$, and $T^{\pm\pm\pm\pm}$, respectively, are given by T\^0&=&[diag]{}([**X\_1**]{}, [**X\_2**]{}, [**X\_2**]{}, [**X\_2**]{}, y\_1, y\_1, y\_2, y\_2, y\_2, y\_2, y\_2, y\_3, y\_3, y\_3, y\_3,y\_3, y\_4, y\_4, y\_4, y\_5, y\_5) , \
T\^&=&[diag]{}([**X\_3**]{}, [**X\_3**]{}, y\_6,y\_6, y\_7,y\_7, y\_2, y\_2, y\_3, y\_3, y\_3, y\_4, y\_4, y\_4, y\_4, y\_5) ,\
T\^&=&[diag]{}([**X\_4**]{}, [**X\_5**]{}, [**X\_5**]{}, y\_2, y\_3, y\_3, y\_4, y\_4) ,\
T\^&=&[diag]{}( y\_3, y\_2,y\_2) ,\
T\^&=&2 y\_2 . Here ${\bf X_i}$ are the $2 \times 2$ matrices whose eigenvalues $x_i^\pm$ are given by x\_1\^&=& 12\_1+22\_2+14\_4 , \[x1\]\
x\_2\^&=& 4\_1+4\_2-2\_4 ,\
x\_3\^&=& 4\_2+4\_1 ,\
x\_4\^&=& 8\_1+4\_2-2\_4 ,\
x\_5\^&=& 12\_2+14\_42 . The eigenvalues $y_i$ are obtained as y\_1&=&8\_2+16\_4 ,\
y\_2&=&8\_2+4\_4 ,\
\[b2\] y\_3&=&4\_3+\_5 ,\
y\_4&=&4\_3-2\_5 ,\
y\_5&=&4(\_3+\_5) ,\
y\_6&=& 8\_2+4( 2+ )\_4 ,\
y\_7&=& 8\_2+4( 2- )\_4 . \[y5\] Although the transition matrix between initial and final two-body states is originally $91 \times 91$, the number of independent eigenvalues turns out to be only seventeen.
Unitarity bounds on the masses
==============================
In this section we analyze mass bounds on the Higgs bosons in the GM model, imposing the condition of perturbative unitarity in Eq. (\[PU\]) to the transition matrix given in the previous section. Consequently we obtain seventeen inequations with respect to all the independent eigenvalues of the transition matrix $T$ in Eq.(\[Tmatrix\]) as |x\_1\^|, |x\_2\^|, |x\_3\^|, |x\_4\^|, |x\_5\^|, |y\_1|, |y\_2|, |y\_3|, |y\_4|, |y\_5|, |y\_6|, |y\_7| < 8. \[PUconstraint\] These eigenvalues are respectively given in Eqs. (\[x1\]) - (\[y5\]) as a combination of the dimensionless coupling constants $\lambda_i$ ($i=1\sim5$) in the Higgs potential, and $\lambda_i$ are related to the Higgs boson masses through Eqs.(\[lambda1\])-(\[lambda5\]), these constraints can be translated into the bounds on the masses $m_{\tilde H_1^0}$, $m_{\tilde H_1^{0'}}, m_{H_3}^{}$ and $m_{H_5}^{}$ and on the mixing angles $\theta_H$ and $\alpha$.
We here show the numerical results on the Higgs mass bounds. Fig. 1 shows the allowed regions of the masses in the $m_{H_3}^{}$ - $m_{H_5}^{}$ plane (a), in the $m_{H_3}^{}$ - $m_{\tilde H_1^0}$ plane (b) and in the $m_{\tilde H_1^0}$ - $m_{\tilde H_1^{0'}}$ plane (c). We vary the Higgs boson masses in the range $m_{H_3}^{}, m_{H_5}^{}, m_{\tilde H_1^0}, m_{\tilde H_1^{0'}} <$ 1 TeV and the mixing angles for $0 < \theta_H \le\pi/2$ and $-\pi/2 < \alpha \le \pi/2$. In each figure, the conditions of perturbative unitarity in Eq.(\[PU\]) are satisfied inside the regions. In Fig. 1(a), light shadowed region is excluded by the $Z\to b\bar b$ result.
![Allowed regions of the masses of the Higgs bosons in the $m_{H_3}^{}$ - $m_{H_5}^{}$ plane (a), in the $m_{H_3}^{}$ - $m_{\tilde H_1^0}$ plane (b) and in the $m_{\tilde H_1^0}$ - $m_{\tilde H_1^{0'}}$ plane (c). In Fig. 1(a) the light shadowed regions are excluded by the $Zb\bar b$ results. []{data-label="figure"}](fig1a.eps "fig:"){width="4.9cm"} ![Allowed regions of the masses of the Higgs bosons in the $m_{H_3}^{}$ - $m_{H_5}^{}$ plane (a), in the $m_{H_3}^{}$ - $m_{\tilde H_1^0}$ plane (b) and in the $m_{\tilde H_1^0}$ - $m_{\tilde H_1^{0'}}$ plane (c). In Fig. 1(a) the light shadowed regions are excluded by the $Zb\bar b$ results. []{data-label="figure"}](fig1b.eps "fig:"){width="4.9cm"} ![Allowed regions of the masses of the Higgs bosons in the $m_{H_3}^{}$ - $m_{H_5}^{}$ plane (a), in the $m_{H_3}^{}$ - $m_{\tilde H_1^0}$ plane (b) and in the $m_{\tilde H_1^0}$ - $m_{\tilde H_1^{0'}}$ plane (c). In Fig. 1(a) the light shadowed regions are excluded by the $Zb\bar b$ results. []{data-label="figure"}](fig1c.eps "fig:"){width="4.9cm"}
Fig. 1(a) shows that $m_{H_3}^{}$ and $m_{H_5}^{}$ are bounded from above respectively by about 400 GeV and about 700 GeV. These upper bound come from $|x_1^\pm| < 8\pi$, which give the most stringent constraint among the inequations in Eq.(\[PUconstraint\]). For $m_{H_3}^{} \gsim 170$ GeV, $m_{H_5}$ is bounded from above whose border is approximately corresponding to $\lambda_4 \simeq 0$ or $m_{H_5}^{}\simeq \sqrt{3} m_{H_3}^{}$ with $\theta_H \simeq 0$. Owing to the factor $\sqrt{3}$, the more strict constraint is given on $m_{H_3}^{}$ than on $m_{H_5}^{}$. For $m_{H_3}^{} \lsim 170$ GeV, on the other hand, the upper bound on $m_{H_5}^{}$ (290 GeV) is realized at $\theta_H\simeq \pi/2$. When all masses other than $m_{H_3}^{}$ are zero, $m_{H_3}^{}$ is bounded by $m_{H_3}<\sqrt{2\pi/3}\, v$ ($\simeq 356$ GeV) from $|x_1^\pm| < 8\pi$. However numerical analysis shows that the actual upper bound is a few decade GeV greater. This excess comes from some delicate cancellation in Eq.(\[x1\]). When we impose the experimental data from $Z\to b\bar b$ which give the constraint on the combination of $m_{H_3}^{}$ and $\theta_H$ [@Haber:1999zh], the allowed region is further limited in the dark shadowed regions. The upper bounds on $m_{H_3}^{}$ and $m_{H_5}^{}$ do not change, but the remained allowed region is in the vicinity of $m_{H_5}^{}\simeq \sqrt{3} m_{H_3}^{}$.
In Fig. 1(b) we can see that the upper bound on $m_{\tilde H_1^0}$ is about 710 GeV, which is almost the same as that on the mass of the SM Higgs boson [@Lee:1977yc]. Larger values of $m_{\tilde H_1^0}$ are allowed for smaller values of $\alpha$ and $\theta$ as well as smaller $m_{H_5}^{}$ and $m_{\tilde
H_1^{0'}}$ values. For instance, taking the limit $\alpha \to 0,~ \theta_H \to 0,~m_{H_5}^{}\to 0$ and $m_{\tilde H_1^{0'}}\to 0$, we obtain $m_{\tilde H_1^0} < \sqrt{8\pi/3}\, v$ ($\simeq 712$ GeV) in the condition $|x_1^+|<8\pi$. The mass bound for another singlet $\tilde H_1^{0'}$ can be obtained by replacing $m_{\tilde H_1^0}$ with $m_{\tilde H_1^{0'}}$ and $\alpha$ with $\alpha + \pi/2$, which can be seen from Eqs.(\[lambda1\])-(\[lambda3\]). Consequently the allowed regions in the $m_{H_3}^{}$ - $m_{\tilde H_1^{0'}}$ plane are given by the same as in Fig. 1(b). We find that contrary to the result in Fig. 1(a) there is only few difference in the case where we include the $Z\to b\bar b$ data.
The allowed region in the $m_{\tilde H_1^{0}}$ - $m_{\tilde H_1^{0'}}$ plane in Fig. 1(c) is symmetrical about the line of $m_{\tilde H_1^{0}} = m_{\tilde H_1^{0'}}(\equiv m)$. It is notable that at least one singlet receives very strict constraint from perturbative unitarity. The mass of lighter singlet, either $\tilde H_1^0$ or $\tilde H_1^{0'}$, is bounded from above by 322 GeV. In analytic calculation, this upper bound is obtained as $m < \sqrt{6\pi/11}\, v$ ($\simeq 322$ GeV) from $|x_1^\pm| < 8\pi$.
In the following, we evaluate the upper bound on $m_{\rm lightest}^{}$ $\left[\equiv {\rm Min}\left(m_{H_3}^{}, m_{H_5}^{}, m_{\tilde H_1^0}, m_{\tilde H_1^{0'}} \right)\right]$. We start from the case in which the constraints from the $Zb\bar b$ results are switched off. When the 3-plet is the lightest, we obtain the upper bound on $m_{\rm lightest} ( = m_{H_3}^{})$ as m\_[lightest]{} < 269 [ GeV]{}, \[lightest\] which is considerably lower than that of the SM Higgs boson, 712 GeV. This condition comes from the constraints $|x_1^+|<8\pi$ and $|x_5^-|<8\pi$. Similar analysis has been done for $m_{\rm lightest}^{} = m_{H_5}^{},
m_{\tilde H_1^0}$ and $m_{\tilde H_1^{0'}}$ in order, and the same bound as in Eq.(\[lightest\]) is derived for each case. When $m_{\rm lightest}^{} \simeq 269$ GeV, all the masses are degenerate in mass ($m_{H_3}^{} = m_{H_5}^{} = m_{\tilde H_1^0} = m_{\tilde H_1^{0'}}$). The situation turns out to be quite similar to the situation of the two-Higgs-doublet model with the discrete symmetry, where the lightest of all Higgs masses are bounded at 410 GeV [@Kanemura:1993hm]. In the case of the GM model the number of the two-body states is greater than that in the two-Higgs-doublet model. (The neutral two body states are 14 channels in the two-Higgs-doublet model and 25 channels in the GM model.) Thereby we have obtained the stronger bounds than the two-Higgs-doublet model. Finally, when we take into account the $Zb\bar b$ results [@Haber:1999zh], the angle $\theta_H$ is more limited for smaller values of $m_{H_3}^{}$. Consequently, the combined upper bound on $m_{\rm lightest}^{}$ becomes lower than 269 GeV. Depending on what the lightest is, the combined upper bound turns out to be about 249 GeV (176 GeV) when $m_{H_3}^{}$, $m_{\tilde H_1^0}$ or $m_{\tilde H_1^{0'}}$ ($m_{H_5}^{}$) is the lightest.
We have not included the LEP direct search results, which give the lower bound $m_{H_{\rm SM}}^{}>114$ GeV in the SM [@Yao:2006px]. In the GM model similar lower mass bounds can be obtained for neutral Higgs bosons but depending on the mixing angles, which would slightly affect the upper bounds by using the results in Figs. 1(a), (b) and (c). We have taken into account only the $Z\to b\bar b$ result as the experimental constraint [@Haber:1999zh], because this constraint drastically changes the bound in the $m_{H_3}^{}$-$m_{H_5}^{}$ plane and also that on $m_{\rm lightest}$.
Conclusions
===========
In this paper, we have analyzed unitarity constraints on the Higgs boson masses in the GM model, which includes a real and a complex isospin triplet fields but predicts $\rho=1$ at tree level. All possible two-body elastic-scattering channels (91 channels) have been taken into account to construct the S-wave amplitude matrix in the high energy limit. The condition of S-wave unitarity in Eq.(\[PUconstraint\]) has been applied to the eigenmatrix.
We have found that all the Higgs bosons receive their masses from the VEV under the discrete $Z_2$ symmetry, so that all the masses can be bounded from above by the condition of parturbative unitarity. In particular, the upper bound on the mass of the $SU(2)_{\rm V}$ 3-plet is about $1/\sqrt{3}$ lower than that on the SM Higgs boson mass (712 GeV). Hence at least one of the singly-charged Higgs boson masses is bounded from above at about 400 GeV. The mass of the lighter $SU(2)_{\rm V}$ singlet scalar state, either $\tilde{H}_0$ or $\tilde{H}_0'$, turns out to be bounded from above by about 300 GeV. Furthermore, the mass of the lightest Higgs boson among the 5-plet, the 3-plet and the two singlet states, whatever it would be, receives very strong constraint from above; i.e., $m_{\rm lightest}^{} \simeq 270$ GeV. The point of the parameter space at which $m_{\rm lightest}^{}$ takes its maximum value corresponds to that where all the mass parameters are degenerate. The combined upper bound with the $Zb\bar b$ results becomes about 150 GeV (95% C.L.). Therefore, the model turns out to be well testable at collider experiments. The $SU(2)_{\rm V}$ 5-plet and 3-plet have the doubly- and singly-charged states, so that the distinctive phenomenological features of this model should also appear in physics of charged Higgs bosons. Detailed phenomenological features will be discussed elsewhere.
In the analysis above, we have considered the Higgs potential with the $Z_2$ symmetry, neglecting the trilinear scalar terms of $\mu_1$ and $\mu_2$. The imposition of the $Z_2$ symmetry in our analysis would be justified to avoid large excess of the neutrino masses. When we do not respect the $Z_2$ symmetry, the upper bounds in above results become relaxed according to the scales of $\mu_1$ and $\mu_2$ which have linear mass dimension. Unless $\mu_1$ and $\mu_2$ are substantially larger than ${\cal O}$(100) GeV, our results above can sufficiently be applied by small relaxation.
Finally, in this paper, we have employed partial wave unitarity to constrain parameters of the GM model at the tree level. A more detailed study with the radiative effects such as vacuum stability or triviality might give more strict bounds on the Higgs boson masses in this model.
Acknowledgments {#acknowledgments .unnumbered}
===============
The work of M. A. was supported, in part, by Japan Society for the Promotion of Science. The work of S. K. was supported, in part, by Grant-in-Aid of the Ministry of Education, Culture, Sports, Science and Technology, Government of Japan, No. 18034004, and by Grant-in-Aid for Scientific Research, Japan Society for the Promotion of Science, No. 19540277.
[99]{}
E. Gildener and S. Weinberg, Phys. Rev. D [**13**]{}, 3333 (1976). J. F. Gunion, H. E. Haber, G. L. Kane and S. Dawson, [*The Higgs Hunter’s Guide*]{}, (Addison-Wesley, New York, 1990), arXiv:hep-ph/9302272.
H. Georgi and M. Machacek, Nucl. Phys. B [**262**]{}, 463 (1985). M. S. Chanowitz and M. Golden, Phys. Lett. B [**165**]{}, 105 (1985).
J. F. Gunion, R. Vega and J. Wudka, Phys. Rev. D [**42**]{}, 1673 (1990). R. Vega and D. A. Dicus, Nucl. Phys. B [**329**]{}, 533 (1990). J. F. Gunion, R. Vega and J. Wudka, Phys. Rev. D [**43**]{}, 2322 (1991). R. Godbole, B. Mukhopadhyaya and M. Nowakowski, Phys. Lett. B [**352**]{}, 388 (1995) \[arXiv:hep-ph/9411324\].
K. Cheung, R. J. N. Phillips and A. Pilaftsis, Phys. Rev. D [**51**]{}, 4731 (1995) \[arXiv:hep-ph/9411333\]. A. G. Akeroyd, Phys. Lett. B [**442**]{}, 335 (1998) \[arXiv:hep-ph/9807409\].
H. E. Haber and H. E. Logan, Phys. Rev. D [**62**]{}, 015011 (2000) \[arXiv:hep-ph/9909335\].
K. Cheung and D. K. Ghosh, JHEP [**0211**]{}, 048 (2002) \[arXiv:hep-ph/0208254\].
B. W. Lee, C. Quigg and H. B. Thacker, Phys. Rev. Lett. [**38**]{}, 883 (1977); Phys. Rev. D [**16**]{}, 1519 (1977). D. A. Dicus and V. S. Mathur, Phys. Rev. D [**7**]{}, 3111 (1973).
M. Lindner, Z. Phys. C [**31**]{}, 295 (1986); N. Cabibbo, L. Maiani, G. Parisi and R. Petronzio, Nucl. Phys. B [**158**]{}, 295 (1979). S. Kanemura, T. Kubota and E. Takasugi, Phys. Lett. B [**313**]{}, 155 (1993) \[arXiv:hep-ph/9303263\]. H. Hüffel and G. Pocsik, Z. Phys. C [**8**]{} (1981) 13; J. Maalampi, J. Sirkka and I. Vilja, ; A. Akeroyd, A. Arhrib, E.-M. Naimi, ; I. F. Ginzburg, I. P. Ivanov, hep-ph/0312374. I. F. Ginzburg and I. P. Ivanov, Phys. Rev. D [**72**]{}, 115010 (2005) \[arXiv:hep-ph/0508020\]; H. Komatsu, Prog. Theor. Phys. [**67**]{}, 1177 (1982); R.A. Flores and M. Sher, Ann. Phys. (NY), 148 (1983) 295; M. Sher, Phys. Rept. [**179**]{}, 273 (1989); D. Kominis and R. S. Chivukula, Phys. Lett. B [**304**]{}, 152 (1993) \[arXiv:hep-ph/9301222\]; S. Nie and M. Sher, Phys. Lett. B [**449**]{}, 89 (1999) \[arXiv:hep-ph/9811234\]. S. Kanemura, T. Kasai and Y. Okada, Phys. Lett. B [**471**]{}, 182 (1999) \[arXiv:hep-ph/9903289\]. J. R. Forshaw, A. Sabio Vera and B. E. White, JHEP [**0306**]{}, 059 (2003) \[arXiv:hep-ph/0302256\].
S. L. Glashow and S. Weinberg, Phys. Rev. D [**15**]{}, 1958 (1977). E. J. Chun, K. Y. Lee and S. C. Park, Phys. Lett. B [**566**]{}, 142 (2003) \[arXiv:hep-ph/0304069\]; P. Q. Hung, Phys. Lett. B [**649**]{}, 275 (2007) \[arXiv:hep-ph/0612004\].
M. Aoki and S. Kanemura, Work in progress.
A. Kundu and B. Mukhopadhyaya, Int. J. Mod. Phys. A [**11**]{}, 5221 (1996) \[arXiv:hep-ph/9507305\]. D. Chakraverty and A. Kundu, Mod. Phys. Lett. A [**11**]{}, 675 (1996) \[arXiv:hep-ph/9508234\]. J. A. Grifols and A. Mendez, Phys. Rev. D [**22**]{}, 1725 (1980).
B. Mukhopadhyaya, Phys. Lett. B [**252**]{}, 123 (1990). E. Asakawa and S. Kanemura, Phys. Lett. B [**626**]{}, 111 (2005) \[arXiv:hep-ph/0506310\]; E. Asakawa, S. Kanemura and J. Kanzaki, Phys. Rev. D [**75**]{}, 075022 (2007) \[arXiv:hep-ph/0612271\]. M. C. Peyranere, H. E. Haber and P. Irulegui, Phys. Rev. D [**44**]{}, 191 (1991); A. Mendez and A. Pomarol, Nucl. Phys. B [**349**]{}, 369 (1991); S. Kanemura, Phys. Rev. D [**61**]{}, 095001 (2000) \[arXiv:hep-ph/9710237\]; A. Arhrib, R. Benbrik and M. Chabab, J. Phys. G [**34**]{}, 907 (2007) \[arXiv:hep-ph/0607182\]; A. Arhrib, R. Benbrik and M. Chabab, Phys. Lett. B [**644**]{}, 248 (2007) \[arXiv:hep-ph/0701126\]. S. Kanemura, Eur. Phys. J. C [**17**]{}, 473 (2000) \[arXiv:hep-ph/9911541\]; A. Arhrib, M. Capdequi Peyranere, W. Hollik and G. Moultaka, Nucl. Phys. B [**581**]{}, 34 (2000) \[Erratum-ibid. [**2004**]{}, 400 (2004)\] \[arXiv:hep-ph/9912527\]; H. E. Logan and S. Su, Phys. Rev. D [**66**]{}, 035001 (2002) \[arXiv:hep-ph/0203270\]; O. Brein, arXiv:hep-ph/0209124; O. Brein and T. Hahn, Eur. Phys. J. [**52**]{}, 397 (2007) \[arXiv:hep-ph/0610079\]. S. Kanemura, S. Moretti and K. Odagiri, JHEP [**0102**]{}, 011 (2001) \[arXiv:hep-ph/0012030\]. J. F. Gunion, J. Grifols, A. Mendez, B. Kayser and F. I. Olness, Phys. Rev. D [**40**]{}, 1546 (1989); J. F. Gunion, C. Loomis and K. T. Pitts, arXiv:hep-ph/9610237; G. Azuelos, K. Benslama and J. Ferland, J. Phys. G [**32**]{}, 73 (2006) \[arXiv:hep-ph/0503096\]; A. Hektor, M. Kadastik, M. Muntel, M. Raidal and L. Rebane, Nucl. Phys. B [**787**]{}, 198 (2007) \[arXiv:0705.1495 \[hep-ph\]\]; T. Han, B. Mukhopadhyaya, Z. Si and K. Wang, Phys. Rev. D [**76**]{}, 075013 (2007) \[arXiv:0706.0441 \[hep-ph\]\].
B. Dion, T. Gregoire, D. London, L. Marleau and H. Nadeau, Phys. Rev. D [**59**]{}, 075006 (1999) \[arXiv:hep-ph/9810534\]. A. G. Akeroyd and M. Aoki, Phys. Rev. D [**72**]{}, 035011 (2005) \[arXiv:hep-ph/0506176\]. K. Huitu, J. Laitinen, J. Maalampi and N. Romanenko, Nucl. Phys. B [**598**]{}, 13 (2001) \[arXiv:hep-ph/0006261\]; J. Maalampi and N. Romanenko, Phys. Lett. B [**532**]{}, 202 (2002) \[arXiv:hep-ph/0201196\].
J. F. Gunion, Int. J. Mod. Phys. A [**13**]{}, 2277 (1998) \[arXiv:hep-ph/9803222\]; S. Chakrabarti, D. Choudhury, R. M. Godbole and B. Mukhopadhyaya, Phys. Lett. B [**434**]{}, 347 (1998) \[arXiv:hep-ph/9804297\]; J. E. Cieza Montalvo, N. V. . Cortez, J. Sa Borges and M. D. Tonasse, Nucl. Phys. A [**790**]{}, 554 (2007) \[arXiv:hep-ph/0612039\]; C. X. Yue, S. Zhao and W. Ma, Nucl. Phys. B [**784**]{}, 36 (2007) \[arXiv:0706.0232 \[hep-ph\]\].
J. M. Cornwall, D. N. Levin and G. Tiktopoulos, Phys. Rev. D [**10**]{}, 1145 (1974) \[Erratum-ibid. D [**11**]{}, 972 (1975)\]. W. M. Yao [*et al.*]{} \[Particle Data Group\], J. Phys. G [**33**]{}, 1 (2006).
[^1]: [email protected]
[^2]: [email protected]
[^3]: The neutrino masses might be generated by any other mechanism (e.g. [@Chun:2003ej]). We shall discuss it elsewhere [@Aoki-Kanemura:neutrino].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We seize the opportunity of the publication of selected papers from the *Logic, categories, semantics* workshop in the *Journal of Applied Logic* to survey some current trends in logic, namely intuitionistic and linear type theories, that interweave categorical, geometrical and computational considerations. We thereafter present how these rich logical frameworks can model the way language conveys meaning.
<span style="font-variant:small-caps;">Keywords:</span> Logic ,Type theory ,Category theory ,Formal linguistics
<span style="font-variant:small-caps;">AMS classification:</span> 18A15 ,03B65 ,03B15 ,03B40 ,68T50
address: 'IMB, LaBRI — Université de Bordeaux'
author:
- 'Jean Gillibert, Christian Retor[é]{}'
bibliography:
- 'bigbiblio.bib'
title: |
Category theory, logic and formal linguistics:\
some connections, old and new
---
A seminar and workshop on category theory, logic and linguistic applications
============================================================================
The present issue of the *Journal of Applied Logic* gathers a selection of papers presented at a workshop *Logic, categories, semantics* held in Bordeaux in November 2010. This workshop was organised as a fitting conclusion to the activities of a weekly reading group a weekly reading group called *Sheaves in logic and in geometry* in 2009/2010 and *Logic, categories, geometry* in 2010/2011.
Those activities are common to the maths and computing departments of the University of Bordeaux (IMB-CNRS and LaBRI-CNRS). Although he did not contribute to this introduction, we must thank Boas Erez (IMB, Bordeaux) who was an enthusiastic participant and speaker in our reading group. We would like to thank for their scientific and financial support: our tow departments, (Institut de Mathématiques de Bordeaux, Laboratoire Bordelais de Recherche en Informatique), INRIA Bordeaux Sud-Ouest, the ANR project LOCI, and the project ITIPY (Aquitaine Region) and the French mathematical society (SMF).
Unfortunately, the material in this issue of the *Journal of Applied Logic* does not cover all the great talks we had the privilege to hear, but one can find published material that more or less cover these talks (and these references themselves include further references):
- Pierre Cartier (IHES, Bures) *Sur les origines de la connexion entre logique intuitionniste, catégories et faisceaux*, cf. [@Cartier1979bourbaki];
- Jean-Yves Girard (CNRS, IML, Marseille) *Interdire ou réfuter? Le statut ambigu de la normativité*, cf. [@Girard2012normativity];
- Paul-André Melliès (CNRS, PPS, Paris) *Logical proofs understood as topological knots*, cf. [@Mellies2012lics];
- Michael Moortgat (Universiteit Utrecht) *Continuation semantics for generalized Lambek calculi*, cf. [@MoortgatMoot2013grishin];
- Carl Pollard (Ohio State University, Columbus) *Remarks on categorical semantics of natural language*, cf. [@Pollard2011lacl].
Intuitionistic logic {#intui}
====================
Type theory and categorical logic, which are at the hear of this issue, all rely on intuitionistic logic. Even linear logic, which is also discussed by two articles of this issue, may be viewed as a refinement of intuitionistic logic.
It is quite a challenge to explain briefly what intuitionistic logic is and how it relates to other logical trends. Regarding the historical development of logic, we refer to the standard [@KK86] and regarding a standard logic master course we refer the reader to the standard textbook [@vanDalen2013] which includes a chapter on intuitionistic logic.
Logic is concerned with“truth", and let us vaguely say that “truth" holds or not of formulae that are formalisations of common language sentences, or of mathematical statements, and as such logic has at least two facets. A formula $F$ can be true because it is correctly derived from formulae that are already established or assumed to be true, and this is the proof theoretical view of truth that started with Aristotle. The same formula $F$ can also be true in all or in some situation once the symbols in the formula are properly interpreted: that’s the model theoretical view of truth, which is much more recent. Besides these two well established viewpoints, there exists a slightly different view of “truth" with Ancient origins, dialectics which emerge from the interaction between proofs and refutations. Nowadays this view culminates in game theoretical semantics, and ludics, and the presentations by J.-Y. Girard and P.-A. Melliès during our workshop developed this interactive view. [@Girard2012normativity; @Mellies2012lics]
An important theorem, due to Gödel as an impressive number of the fundamental results logic, is that for first order classical logic, the two notions of truth coincide: a formula is true in any model if and only if it can be proved. This result admits a a more striking formulation as: a formula $C$ can be derived from formulae $H_1,\ldots,H_n$ if and only if any model satisfying $H_1,\ldots,H_n$ satisfies $C$ as well. Such a result holds for classical propositional logic and for classical first-order logic, also known as predicate calculus.
Intuitionistic logic comes from doubts that appeared during the crisis of the foundations of mathematics, between the XIX and XX centuries. These doubts concerned existential statements in infinite sets. For instance it is admittedly a bit strange, especially in an infinite universe, that $\exists x (B(x) \implies \forall y B(y))$, although it is absolutely correct in classical logic using principles like $U\implies V\equiv \lnot U \lor V$, $\lnot\lnot A\equiv A$, $\lnot \forall x A(x)\equiv \exists x \lnot A(x)$. According to intuitionistic logic, *tertium non datur*, or any of the principles that are equivalent to it, like *reductio ad absurdum* or *Pierce’s law*, should be left out from deduction principles.
When restricting ourselves to intuitionistic logic both facets of truth are modified:
- In order to maintain the completeness results, models need to be more complex: the usual models ought to be replaced with structured families of classical models, like Kripke models or (pre)sheaves, cf. section \[sheaves\].
- Intuitionistic proofs are more interesting: as opposed to classical proofs, they can be viewed as algorithms (as programs of a typed functional programming language) or as morphisms in a category (which is not a poset, i.e. with many morphisms from an object to another).
Each of these two aspects brings connections with category theory. The first one yields models of higher order intuitionistic logic, e.g. with (pre)sheaves of $\mathcal{L}$-structures (classical models, roughly speaking). The second one yields models that interpret proofs as morphisms in a cartesian closed category.
A historical connection between logic and geometry:\
(pre)sheaves, toposes and categorical models of intuitionistic logic {#sheaves}
====================================================================
Pierre Cartier reminded us in the inaugural lecture of the workshop that the connection between intuitionistic logic and topos theory appeared in the sixties. We refer the reader to [@Cartier1979bourbaki] for a survey, and to [@MacLaneMoerdijk1992] for a thorough treatment, and to [@Mac71] for an excellent textbook on category theory.
Inspired by the work of Eilenberg and Mac Lane on categories, Lawvere started in 1964 to introduce a categorical approach to set theory, and Lambek was looking at categories as deductive systems, the 1972 book edited by Lawvere *Toposes, Algebraic Geometry, and Logic* gives a pretty good idea of the landscape at the end of the 60s. [@Lawvere72]
At the same time, Grothendieck was introducing groundbreaking concepts in order to define a topology on an algebraic variety that would play a similar role to the topology of the complex points, namely the étale topology. The first step in Grothendieck’s construction is the notion of site, that is, a category endowed with a notion of covering that is reminiscent of open coverings in topology. The concept of site does not exactly capture the‘topology" of a situation, since different sites can give rise to the same category of sheaves. Thus Grothendieck chose to take a category of sheaves of sets over a given site as a notion of generalized topological space, and he called such a category a topos But another important idea of Grothendieck was somehow more surprising: the étale site is just a step in the construction, the final notion to study is the category of sheaves of sets on the site, that he calls a topos. This category has many properties in common with the category of sets, and possesses a good notion of localisation. These radically new ideas were first developed by Grothendieck and his group in [@SGA4].
Taking into account this new approach, Lawvere and Tierney in [@Lawvere1970congres] gave before 1970 a new definition of topos (elementary toposes), which does not refer to a site. One of the avantages of this theory is that each topos contains an object $\Omega$, that plays the role of all possible truth values, and has the (internal) structure of a Heyting algebra. Hence intuitionist logic naturally appears as being the internal logic of the topos.
This connection with intuitionistic logic was formulated by Lawvere at the same time [@Lawvere1970congres]. Shortly afterwards, Lawvere and Tierney in [@Lawvere72] discovered that models of set theory corresponding to proofs of the independence of the axiom of choice and continuum hypothesis by Cohen’s method of forcing [@cohen1966set] were in fact elementary toposes see e.g. [@MacLaneMoerdijk1992; @Borceux1994hdbk3].
On the other hand, the internal logic of the topos shed a new light on categories of sheaves. Namely, one can handle sheaves as if they were sets with elements but with respect to the internal language, but when reasoning about these elements one should follow should follow the rules of intuitionistic logic. In order to fix ideas, let us recall that a *category* $\mathcal{C}$ is defined by the following data: (a) a class of objects; (b) a class of arrows (or morphisms) whose domain and codomain are objects of $\mathcal{C}$; (c) a partial binary operation on arrows called composition, required to be associative; (d) every object is endowed with an identity arrow. If for any objects $A$ and $B$ of $\mathcal{C}$, the class ${\mathop{\mathrm{Hom}}}(A,B)$ of all morphisms from $A$ to $B$ is a set, the category is said to be *locally small*, and most of the useful categories are locally small. If the class of objects and of morphisms of $\mathcal{C}$ are sets, $\mathcal{C}$ is said to be *small*.
It is easy to define morphisms of categories, which for historical reasons are called *functors*.
A category can be seen as a directed graph with a composition operation on arrows. The most basic examples of categories are the category of sets, the category of groups, the category of $K$-vector spaces where $K$ is a field, the category of topological spaces, etc. Roughly speaking, categories allow to handle most kind of structural questions in mathematics. We will now focus our attention on sheaves, which are central in this story. Let $X$ be a topological space. A presheaf of sets is a contravariant functor from the category of open subsets of $X$ (a poset, actually) to the category of sets[^1].
A morphism of presheaves $\phi:A\to B$ is given by a family of maps $\phi_U:A(U)\to B(U)$ (where $U$ runs through all open subsets of $X$) that is compatible with restriction maps in the following sense: for any inclusion $V\subseteq U$ between open subsets of $X$, the following diagram commutes $$\begin{CD}
A(U) @>\phi_U>> B(U) \\
@VVV @VVV \\
A(V) @>\phi_V>> B(V) \\
\end{CD}$$ where vertical maps are the restrictions induced by the inclusion $V\subseteq U$.
It is worth noticing that (pre)sheaves of $\mathcal{L}$-structures of some theory $\mathcal{T}$ over a topological space — whose open sets can be viewed as a poset category that is a Heyting algebra — define a notion of model that is complete for intuitionistic logic. An $\mathcal{L}$-structure is a set with an interpretation of the first order language $\mathcal{L}$, and the morphisms between $\mathcal{L}$-structures, in particular the restrictions morphisms of the sheaf, are functions that preserves function symbols and the truth of atomic formulae — hence they are more general than the usual elementary morphisms of model theory which preserve the truth of all formulae. This construction due to Joyal is related to forcing and is a particular case of Kripke models [@Kripke59jsl] and therefore is known as Kripke-Joyal forcing. [@BoileauJoyal1981] The proof of completeness of (pre)sheaf models can be inferred from results in [@TD88v2]; recently, Ciardelli obtained a short and direct proof of this completeness result [@CIardelli2011tacl].
A sheaf of sets is a presheaf $\mathcal{F}$ which satisfies the following condition: for any open subset $U$ of $X$ and any family $(U_i)$ of open subsets of $X$ such that $U=\cup U_i$, there is an equaliser diagram $$\mathcal{F}(U)\rightarrow \prod_i \mathcal{F}(U_i) \rightrightarrows \prod_{j,k} \mathcal{F}(U_j\cap U_k)$$ (this means that the first map is injective and its image is exactly the set where the two other maps coincide[^2]).
In other words, a “compatible” family of local sections of $\mathcal{F}$ can be glued—in a unique way—into a global section of $\mathcal{F}$ i.e. a $U$-section of $\mathcal{F}$. This is exactly what one needs in order to perform “local to global” yoga.
By definition, a morphism of sheaves is just the same as a morphism of presheaves. Hence the category of sheaves on $X$, that we denote by ${\mathbf{Sh}}(X)$, is a full subcategory of the category of presheaves. In order to describe its properties, we introduce the notion of cartesian closed category ([CCC]{}).
\[ccc\] A category is said to be cartesian closed whenever:
1. \[terminal\] It has a terminal object here noted $\mathbf{1}$ (it can be noted $T$);
2. \[product\] Any two objects $A$ and $B$ have a product written $A\times B$;
3. \[exponential\] Any two objects $A$ and $B$ have an exponential noted $B^A$ or $A{\rightarrow}B$.
In more detail, \[terminal\] a terminal object is an object $\mathbf{1}$ such that for any object $A$ there exists a unique arrow $A\to \mathbf{1}$; \[product\] if $A$ and $B$ are two objects, the product $A\times B$ (if it exists) is an object such that, for all $Y$, there is a bijection ${\mathop{\mathrm{Hom}}}(Y,A)\times{\mathop{\mathrm{Hom}}}(Y,B)\simeq{\mathop{\mathrm{Hom}}}(Y,A\times B)$ which is natural in $Y$ ; \[exponential\] if $A$ and $B$ are two objects, the exponential noted $B^A$ (if it exists) is an object such that, for all $Y$, there is a bijection ${\mathop{\mathrm{Hom}}}(Y\times A,B)\simeq{\mathop{\mathrm{Hom}}}(Y,B^A)$ which is natural in $Y$. Another limit is useful, *equaliser*: given tow morphisms $f$ and $g$ from $A$ to $B$, an equaliser of $f$ and $g$ is a morphism $e$ from $C$ to $A$ such that $f\circ e= g\circ e$ and for any morphism $d$ from $D$ to $A$ with $f\circ d= g\circ d$ there is a unique $k$ from $D$ to $C$ such that $e \circ k= d$.
Let us check that ${\mathbf{Sh}}(X)$ is cartesian closed. The constant sheaf $\mathbf{1}$ defined by $\mathbf{1}(U)=\{*\}$ for all $U$ is clearly a terminal object in ${\mathbf{Sh}}(X)$. The product of two sheaves, as well as the equaliser of sheaf morphisms with the same domain and codomain exist, hence all finite limits exist in the category of sheaves. Finally, exponential objects exist: for any two sheaves $A$ and $B$, we let $B^A$ be the sheaf $U\mapsto {\mathop{\mathrm{Hom}}}(A|_U,B|_U)$, where $\mathcal{F}|_U$ denotes the restriction of $\mathcal{F}$ to $U$[^3].
More generally, the category ${\mathbf{Sh}}(X)$ inherits many properties from the category of sets, which is not surprising if we consider a sheaf as being a “continuous family of sets”. This leads us to the next topic, namely elementary toposes.
\[topos\] According to Lawvere and Tierney, an *elementary topos* is a category $E$ satisfying the following properties:
1. $E$ has finite limits;
2. $E$ has exponentials;
3. $E$ has a subobject classifier.
The first two conditions imply that a topos is a cartesian closed category. In fact, the main difference between [CCC]{} and toposes is the existence of a subobject classifier.
Let us recall that $\Omega$ is a subobject classifier if, for any object $A$ of $E$, there is a natural isomorphism between subobjects of $A$ and the set of maps $A\to \Omega$. More precisely, $\Omega$ comes with a canonical map $\mathbf{1}\to\Omega$ (where $\mathbf{1}$ is the terminal object of $E$) which satisfies the following universal property: for any monomorphism $f:B\to A$ there exists a unique morphism $g:A\to \Omega$ such that the square $$\begin{CD}
B @>>> \mathbf{1}\\
@VfVV @VVV \\
A @>g>> \Omega\\
\end{CD}$$ is a fibered product or pullback (roughly speaking, $B$ is the inverse image of $\mathbf{1}$ by $g$).
In the category of sets, $\Omega=\{0,1\}$ (Boolean truth-values).
In the category of sheaves, the subobject classifier is defined as follows: for any open subset $U$ of $X$, we let $\Omega(U)$ to be the set of all open subsets of $U$. Roughly speaking, an assertion inside the topos ${\mathbf{Sh}}(X)$ may have an intermediate truth value, which from the viewpoint of an open subset $U$ is the open subset of $U$ where the assertion is true.
*In this issue, the paper *Continuity and geometric logic* by Steve Vickers emphasises the connection with topology by exploring geometric logic, that is many sorted logic with finite conjunctions and arbitrary disjunctions (also existential quantification). Given that classical models of a geometric theory $\mathbf{T}$ are not complete, one has to look models of $\mathbf{T}$ inside a topos, and there exists a classifying topos whose category of T models is universal. This is a point-free manifestation of the “space of models of $\mathbf{T}$ ".*
Intuitionistic proofs, typed lambda calculus and cartesian closed categories {#lambda}
============================================================================
A second link between intutionistic logic and category theory is the following. Proofs in (propositional) intuitionistic logic can be viewed as typed lambda terms, that are functional programs. The main idea is that a proof of $A{\rightarrow}B$ maps proofs of $A$ to proofs of $B$, hence it is rather natural to interpret a formula (with $\land, {\rightarrow}$) by the sets of its proofs. Functions can be applied to arguments of the right type, and composed. To compute such a term, one substitutes the occurrences of the bound variable used to define the function with the argument of the function: $$(\lambda x^A.\ t^B)(u^A)\stackrel{\beta}{\rightsquigarrow}t[x^A{:}=u^A]$$ Values are normal terms i.e. terms that cannot be reduced any further.
In figure \[curry\] we give the proof rules and the corresponding proof terms, that are simply typed lambda terms, a good reference being [@GLT88] and [@LS86] on the categorical aspects.
\[curry\]
axiom x[:]{}Ax[:]{}A
[z\_1[:]{}Z\_1,...,z\_p[:]{}Z\_p]{}t[:]{}B [z\_[(1)]{}[:]{}Z\_[(1)]{},...,z\_[(p)]{}[:]{}Z\_[(p)]{}]{}t[:]{}B
[c]{}exchange\
(: permutation)
[z\_1[:]{}Z\_1,...,z\_p[:]{}Z\_p]{}, x[:]{}A, x[:]{}A t[:]{}B [z\_1[:]{}Z\_1,...,z\_p[:]{}Z\_p]{}, x[:]{}A t[:]{}B contraction
[z\_1[:]{}Z\_1,...,z\_p[:]{}Z\_p]{}, t[:]{}B [z\_1[:]{}Z\_1,...,z\_p[:]{}Z\_p]{}, x[:]{}A t[:]{}B weakening
[z\_1[:]{}Z\_1,...,z\_p[:]{}Z\_p]{}, x[:]{}A t[:]{}B , x[:]{}A (x\^A. t)[:]{}AB abstraction / hypothetical reasoning
[z\_1[:]{}Z\_1,...,z\_p[:]{}Z\_p]{}, f [:]{}AB [y\_1[:]{}Y\_1,...,y\_k[:]{}Y\_k]{}u[:]{}A [z\_1[:]{}Z\_1,...,z\_p[:]{}Z\_p]{},[y\_1[:]{}Y\_1,...,y\_k[:]{}Y\_k]{}f(u)[:]{}B application / modus ponens
Functions in this setting are algorithms (that always converge) that represent total recursive functions, the ones whose totality can be proved within the underlying logical system. They are defined intentionally, how could one view them as usual functions? Functions can be viewed as morphisms in a cartesian closed category, cf. definition \[ccc\] above — in this logical setting, $B^A$ is rather written $A{\rightarrow}B$. Observe that the logical rules match categorical isomorphisms, for instance, $$A {\rightarrow}(B {\rightarrow}C) \cong (A \times B) {\rightarrow}C$$ Closed terms (i.e. terms without free variables) of type $U$ also can be viewed as maps from the terminal object $\mathbf{1}$ to $U$ — this view is supported by the fact, that $\mathbf{1}=\{*\}$ in the category of set.
A function in the model interprets a function as a static notion (a “set" of pairs $\langle x,f(x)\rangle$) and not as a (functional) algorithm: indeed whenever a term/proof $t$ reduces to $t^\circ$ its interpretation as a morphism in the category is the same. Given that functions from $A$ to $B$ are definable by terms/proofs, the set ${\mathop{\mathrm{Hom}}}(A,B)$ should be countable. Therefore, “elements" in $B^A$, that are arrows from in $\textbf{1}$, should be countable as well, and this imposes a drastic restriction on the morphisms we consider in the category. The study of such models of intuitionistic proofs lead to linear logic as we shall see in section \[linear\].
A result by Joyal (unpublished see e.g. [@LS86; @Girard2011blindspot]) shows that all proofs of a given formula are equivalent in classical logic. Although there also exists a logical, syntactical analogous result by Girard and Lafont (see figure \[lafont\]), Joyal proved a category theoretical formulation of this statement:
In a cartesian closed category interpreting classical logic between any two objects (formulae) $A$ and $B$ there is at most one morphism from $A$ to $B$ (at most one interpretation of proofs of $A{\vdash}B$).
The interpretation of classical logic needs an initial object $\mathbf{0}$ (false) and a negation $\lnot A=(A{\rightarrow}\mathbf{0})$ which is involutive i.e. $A\cong ((A{\rightarrow}\mathbf{0}) {\rightarrow}\mathbf{0})$. To see that in such a situation there exists at most one arrow between any two objects, let us first see that in a [CCC]{} with an initial object $\mathbf{0}$ when there is a map from $Z$ some object to $\mathbf{0}$, then $Z$ is initial as well:
1. \[Zx0\] *If $\mathbf{0}$ is initial, then is $\mathbf{0}\times Z$ is initial as well.* Indeed, in a [CCC]{} , maps from $\mathbf{0}\times Z$ to any object $U$ are in bijection with maps from $\mathbf{0}$ to $U^Z=[Z{\rightarrow}U]$.
2. \[Zinit\] If there exists an arrow $f$ from $Z$ to $\mathbf{0}$ then $Z$ is isomorphic to $\mathbf{0}\times Z$ and therefore $Z$ is initial. Indeed, $\pi_2\circ \langle f, Id_Z\rangle=Id_Z$ (by definition of $\pi_2$) and $\langle f, Id_Z\rangle\circ \pi_2=Id_{\mathbf{0}\times Z}$ because it goes from $\mathbf{0}\times Z$ to itself, which is initial because of \[Zx0\] above.
Maps from $A$ to $B$ correspond to maps from $A$ to $\neg\neg B \cong
(B \to \mathbf{0}) \to \mathbf{0}$ and thus to maps from $A \times (B \to \mathbf{0})$ to $\mathbf{0}$. But from \[Zinit\] it follows that such a map is an isomorphism and there is precisely one such isomorphism since $\mathbf{0}$ is initial.
\_1 F
F,K weakening
\_2 F
F,K weakening
F, F cut
F contr.
This result, simple enough to be presented in this introduction, clearly shows that intuitionistic logic is the natural logic of the [CCC]{} interpretation of proofs — in the previous section we already saw that intuitionistic also is the logic of toposes which generalise set theory.
Martin-Löf type theory and identity types {#mtt}
=========================================
One may further refine the operations (exponential, product) on the simple types of Church [@Church1940types] of the previous section \[lambda\] as linear logic defined in next section \[linear\] does, or consider a wider set of types as Martin-Löf type theory does.
Actually, the most prominent feature of Martin-Löf type theory is that there are types which *depend* on terms as e.g. $\mathrm{Vect}(n)$ for $n:N$, the type of vectors of length $n$. Propositions are considered as types and, accordingly, predicates may be considered as dependent types. The most important example of dependent types will be identity types as discussed subsequently.
Martin-Löf type theory provides the necessary technology to deal with the standard mathematical concept of a family of sets. Moreover a dependent type can be considered as as predicate; for example if we are given a dependent type $B(x)$ with $x:A$ (a family of types indexed by the type $A$), we can think of $A$ as a set and $B(x)$ as a family of sets, but in addition, if we consider $B(x)$ as a predicate, given a hypothetical element $a$ of $A$ and a hypothetical $b$ in $A(a)$ we can think of $b$ as a *proof witness* that shows the truth of $B(a)$.
Type theory as formalized by Martin-Löf gives us access to the construction of
- the type $(\Pi x:A) B(x)$ whose intuitive elements should be seen as functions $f$ that map a term $a:A$ to a term $f(a) : B(a)$. In case $B$ does not depend on $x$, the type $(\Pi x:A) B(x)$ corresponds to the type $A{\rightarrow}B$ of simply typed lambda calculus. When we see dependent types as predicates, we now have access to the universal quantifier.
- the type $(\Sigma x:A) B(x)$ whose elements should be seen are pairs $(a,b)$ with $a : A$ and $b:B(a)$. If $B(x)$ does not depend on $x$, the type $(\Sigma x:A) B$ corresponds to the type $A\times B$ of simply types lambda calculus, and we also have access to the existential quantifier when we think of a type as a predicate.
The presence of dependent types forces us to consider the following judgements
------------------------ ---------------------------------------------
$\Gamma{\vdash}Valid$ The context $\Gamma$ is valid
$\Gamma{\vdash}A$ in the context $\Gamma$, $A$ is a type.
$\Gamma{\vdash}A=B$ the types $A$ and $B$ are equal
$\Gamma{\vdash}t:A$ in the context $\Gamma$, $t$ is of type $A$
$\Gamma{\vdash}t=u: A$ $t$ and $u$ are equal terms of type $A$.
------------------------ ---------------------------------------------
This notion of equality is called *judgemental* and can be decided by normalisation. But it cannot be combined into propositions for which reason one introduces the notion of *propositional* equality. If $t,s :A$ one may form the types $Id_A(t,s)$ whose elements will be proofs that $t$ and $s$ are equal. Reflexivity is given by the constructor $r_A(t) : Id_A(t,t)$ for $t:A$. As usual in type theory a constructor is accompanied by an *eliminator* which in case of identity types happens to be called $J$ and can be described as follows. Given a family of types $x,y:A, z:Id_A(x,y)
{\vdash}C$ and a term $x,y:A, z:Id_A(x,y) {\vdash}d : C(x,y,z)$ one may consider the term $x,y:A, z:Id_A(x,y) {\vdash}J((x) d,z) : C(x,y,z)$ whose meaning is provided by the reduction rule $ J((x) d,r_A(t)) = d[t/x]$. This is in accordance with functional programming where functions are defined by pattern matching *aka* structural recursion.
This is how equality is represented in *intensional* type theory of [@ML73]. In [@ML84] judgemental and propositional equality get identified via a so-called *equality reflection* rule allowing one to conclude $t =s : A$ from $p : Id_A(t,s)$. This, however, renders type checking undecidable and, therefore, has been abandoned. Current proof checkers as e.g. the widely used system $\mathtt{Coq}$ [@BC2004coq] are based on intensional type theory. This is sometimes inconvenient but allows one to avoid the manipulation and storing of derivation trees for typing judgements as in the much older $\mathtt{NuPrL}$ system.
As said above in section \[sheaves\], toposes admit an internal logic in form of intuitionistic higher order arithmetic. In this logic propositions are not represented as types but as subobjects of the terminal object $1$. Accordingly, predicates on a type $A$ are represented as monomorphisms into $A$. On the other hand, a good interpretation of Martin-Löf type theory needs a predicate on a type $A$ to be represented by a morphism $B\to A$ which is more general than just a monomorphism, since we want the ability to distinguish between proof witnesses. Thus we need a a distinguished class of morphisms (often called *display maps*), which are subject to a few simple conditions, the most important of which being that the pullback of a display map with codomain $A$ by *any* morphism with same codomain is again a display map. This corresponds to being able to substitute a term for a variable in a predicate. It is then possible to interpret the $\Pi$-types when a right adjoint always exists to that pullback functor. When this happens we have a *display category with (dependent) products*, and when all maps of the category can be used as display maps, we have a *locally cartesian closed* categories. There are many display categories with dependent products, but it turned out to be harder to find examples that have a good, nontrivial (i.e., non collapsed) versions of the intensional identity type. For example locally cartesian closed categories are guaranteed to give trivial identity types. The first good example was given by M. Hofmann and Th. Streicher in [@gpdtt], where types were interpreted as groupoids, display maps as *fibrations of groupoids* (very easy to define), and $Id_A(x,y)$ as the set of morphisms (necessarily isomorphisms) from $x$ to $y$. This model was only nontrivial in dimension one, in the sense that iterating the identy type would land us on the old, trivial, nonintensional interpretation where identity is just the diagonal.
It was realized that if a model of dependent types has an intensional identity type, then we are very close to being able to do topology (especially, homotopy theory) in it. This is because in such an identity type, we can view an intuitive element of $ID_{A}(x,y)$ (a proof that $x$ is equal to $y$) as a *path from $x$ to $y$,* and we are able to do many constructions of homotopy theory, with the additional bonus of dependent products.
Thus inspirations for good models of identity types were bound to come from algebraic topology, and this is the case for the first completely non-degenerate model, where paths in higher dimensions (higher homotopies) do not collapse, and which was found independently by Streicher and Voevodsky around 2006 (see e.g. [@Voevodsky2006lambda; @Streicher2006uppsala]. The base category is the presheaf topos of *simplicial sets*, the most studied of all models in combinatorial topology, and families of types (i.e., display maps) are interpreted as *Kan fibrations*, which forces ordinary, non-dependent types to be *Kan complexes*. As expected, the identity type $Id_A(x,y)$ is the standard construction for the *path object* in the sense of the usual Quillen model structure on simplicial sets.[@quillen1967homotopical]
Another inspiration from topology, namely in the theory of fiber bundles, has led V. Voevodsky [@ACV2013] to propose the *Univalence Axiom* in what is now called *Homotopy Type Theory* (HoTT), which applies to universes, like those found in $\mathtt{Coq}$ and those defined by Martin-Löf, and can be viewed as the statement that in a “good” universe *isomorphic types are equal*.
The present issue contains two papers about categorical models of identity types, both having links to homotopy type theory:
- *François Lamarche in his paper *Modelling Martin-Löf Type Theory in Categories* proposes a simple model of Martin-Löf type theory that includes both dependent products and the identity types.*
- *Thomas Streicher in the next paper describes *A Model of Type Theory in Simplicial Sets* in which the Voevodsky Univalence axiom holds.*
Although little work has been done in this direction, categorical models of type theory should be very appealing from the viewpoint of natural language semantics, to be evoked in section \[applications\].
Indeed, natural language semantics, philosophy of language, analytic philosophy has always been interested in models (classical models, possibles worlds, situations) and nowadays, natural language semantics is often formalised and computed within type theory — as exemplified in the present issue, see section \[applications\].
From a particular categorical interpretation to linear logic {#linear}
============================================================
While Martin-Löf type theory discussed in previous section \[mtt\] extends the expressive power of simply typed theory with dependent types, one can also have a closer look at the intuitionistic logical connectives $\&$ and ${\rightarrow}$ from a computational perspective bearing in mind that an involutive negation like the one of classical logic brings many convenient and elegant properties. One thus discompose the connective of intuitonistic logic and finds a classical but constructive logic called *linear logic*.[@Gir87; @Girard2011blindspot].
Linear logic, present in two papers of the present issue, arose from the study of a particular cartesian closed category that interprets proofs of intuitionistic logic, namely coherence spaces and deserve some extra information: *coherence spaces*, which are discussed in the present issue regarding their relation to formal ontologies, see section \[applications\].
A coherence space is basically a simple graph (possibly infinite, but without loops nor multiple edges) but entities that interpret terms or proofs in a coherence space are not its vertices but the cliques of this graph, i.e. its complete subgraphs. A stable map from $A$ to $B$ maps cliques of $A$ to cliques of $B$, and it enjoys certain properties: monotony ($a\subset b \Rightarrow F(a)\subset F(b)$), preservation of directed unions ($F(\cup\uparrow a_i)=\cup F(a_i)$) and of intersections inside a clique (when $a\cup a'$ is a clique of $A$ one has $F(a\cap b)=F(a)\cap F(b)$) The important point is that a stable function from a coherence space $A$ to another one $B$ can be viewed as a clique of a larger coherence space $A{\rightarrow}B$, whose vertices are $(a,\beta)$ where $\beta$ is a vertex of $B$ and $a$ a *finite* clique of $A$, such that $\beta\in F(a)$ and $a$ is minimal for this property — $F$ being stable there always exists such a finite $a$, and one of them $a$ is minimal and in fact the minimum. Two vertices $(a,\beta)$ and $(a',\beta')$ of $A{\rightarrow}B$ are said to be coherent whenever $a\cup a'$ is a clique of $A$ implies that $\beta$ and $\beta'$ are coherent in $B$ and when furthermore $a$ and $a'$ are distinct so are $\beta$ and $\beta'$.
One then has a bijection between stable functions from $A$ to $B$ and the cliques of the coherence space $A{\rightarrow}B$. There is also a product of coherence spaces, which is the disjoint union of the two graphs, plus edges between any vertex of $A$ and any vertex of $B$.
The categorical product corresponds to the rule from $\Gamma{\vdash}A$ and $\Gamma{\vdash}B$ infer $\Gamma{\vdash}A \& B$, while the other product $\otimes$ corresponds to the rule from $\Gamma{\vdash}A$ and $\Delta{\vdash}B$ infer $\Gamma,\Delta{\vdash}A \otimes B$. These two conjunctions of linear logic cannot be distinguished in intuitionistic logic, because of the weakening and contraction rules given in figure \[curry\].
Interpreting proofs in coherence spaces, where operation on coherence spaces correspond to logical connectives lead to linear logic. Indeed, besides the product described above there is another “product", which is not a categorical product, whose vertices are pairs of vertices, with an edge between $(a,b)$ and $(a',b')$ whenever there is an edge $\{a,a'\}$ and an edge $\{b,b'\}$ in $B$ — $a=a'$ or $b=b'$ is allowed as well, but not both since there are no loops in simple graphs.
Besides these pairs of conjunction, one discovers an involutive negation $A^\perp$ whose simple graph is the complement of the simple graph (simple graphs contain no loop). Using De Morgan laws it maps the two aforementioned conjunctions to two disjunctions respectively written $\oplus$ (additive disjunction) and $\wp$ (multiplicative disjunction).
There also exists a unary operator $!$ which relates linear logic (resource sensitive) to intuitionistic logic. The coherence space $!A$ is defined as follows: its vertices are the finite cliques of $A$, and two vertices $a$ and $a'$ of $!A$, that is to say two finite cliques of $A$, are coherent in $!A$ whenever the union $a\cup a'$ is a clique of $A$. This operator turns the additive conjunction $\&$ into the multiplicative conjunction $\otimes$: $!(A\&B)\equiv !A\otimes !B$.
A kind of classical implication can be defined from the $\wp$ disjunction: $A^\perp\wp B$, it is written as $A\multimap B$, and the intuitionistic implication can be defined as: $$A{\rightarrow}B = !A\multimap B$$
Linear logic was provided with a syntax, both with sequent calculus and with proofnets that identify proofs up to rule permutations. Propositional linear logic is also endowed with a truth value semantics *phase semantics*.
*In this issue, the paper entitled *Relational semantics for full linear logic* by Dion Coumans, Mai Gehrke, Lorijn van Rooijen proposes a Kripke style semantics for full propositional linear logic, i.e. with, additives, multiplicatives and exponentials that allow the addition of proper axioms.*
Linguistic applications: compositional semantics, lexical semantics, and knowledge representation {#applications}
=================================================================================================
Natural language semantics is usually expressed by first order (some times higher order) logic or predicate calculus, single sorted as Frege wished to, and computed according to Frege’s compositionally principle: the meaning of the compound is a function of the meaning of its parts. [@vBtM2010; @partee02; @DavisGillon2004intro].
This is easily implemented in lambda calculus where anything is a function, as noticed by Montague in 1970 [@montague:formal]. Typed $\lambda$-terms for representing logical formulae are usually defined out of two base types, ${\mathbf{e}}$ for individuals (also known as **e**ntities) and ${\mathbf{t}}$ for propositions (which have a **t**ruth value). Logical formulae can be defined in this typed $\lambda$-calculus as first observed by Church in 1940 [@Church1940types], but the combination of word meaning (partial formulae) according to the syntactic structure of the sentence is due to Montague.
This early use of lambda calculus, where formulae are viewed as typed lambda terms, can not be merged with the more familiar view of typed lambda terms as proofs evoked in the section \[lambda\]. The proof which such a typed lambda term corresponds to is simply the proof that the formula is well formed, e.g. that a two-place predicate is properly applied to *two* individual terms of type ${\mathbf{e}}$ and not to more or less arguements, nor to arguments of a different type etc. This initial vision of lambda calculus was designed for a proper handling of substitution in deductive systems à la Hilbert. One needs constants for the logical quantifiers and connectives:
$
\begin{array}[t]{r|l}
\multicolumn{2}{c}{\mbox{Logical\ connectives\ and\ quantifiers}}\\
\mbox{Constant} & \mbox{Type}\\ \hline
\textrm{\&, and} & {\mathbf{t}}{\rightarrow}({\mathbf{t}}{\rightarrow}{\mathbf{t}}) \\
\textrm{$\lor$, or} & {\mathbf{t}}{\rightarrow}({\mathbf{t}}{\rightarrow}{\mathbf{t}}) \\
\textrm{$\Rightarrow$, implies} & {\mathbf{t}}{\rightarrow}({\mathbf{t}}{\rightarrow}{\mathbf{t}})\\ \hline
\exists & ({\mathbf{e}}{\rightarrow}{\mathbf{t}}) {\rightarrow}{\mathbf{t}}\\
\forall & ({\mathbf{e}}{\rightarrow}{\mathbf{t}}) {\rightarrow}{\mathbf{t}}\\
\end{array}
$ $
\begin{array}[t]{r|l}
\multicolumn{2}{c}{\mbox{Predicates\ and\ constants}}\\
\mbox{Constant} & \mbox{Type}\\ \hline
\mathit{defeated,\ldots} & {\mathbf{e}}{\rightarrow}({\mathbf{e}}{\rightarrow}{\mathbf{t}}) \\
\mathit{won, voted,\ldots} & ({\mathbf{e}}{\rightarrow}{\mathbf{t}}) \\
\mathit{Liverpool, Leeds,\ldots} & {\mathbf{e}}\\
\mathit{\ldots} & \ldots
\end{array}
$
as well as predicates for the precise language to be described — a binary predicate like $won$ has the type ${\mathbf{e}}{\rightarrow}{\mathbf{e}}{\rightarrow}{\mathbf{t}}$.
A small example goes as follows. Assume that the syntactical analysis of the sentence “*Some club defeated Leeds.*" is
(some (club)) (defeated Leeds)
where the function is always the term on the left. If the semantic terms are as in the lexicon in figure \[semanticlexicon\], placing the semantical terms in place of the words yields a large $\lambda$-term that can be reduced:
------------ --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
**word** ***semantic type $u^*$***
***semantics : $\lambda$-term of type $u^*$***
[*$x{^{v}}$ the variable or constant $x$ is of type $v$*]{}
*some* $({\mathbf{e}}{\rightarrow}{\mathbf{t}}){\rightarrow}(({\mathbf{e}}{\rightarrow}{\mathbf{t}}) {\rightarrow}{\mathbf{t}})$
$\lambda P{^{{\mathbf{e}}{\rightarrow}{\mathbf{t}}}}\ \lambda Q{^{{\mathbf{e}}{\rightarrow}{\mathbf{t}}}}\
(\exists{^{({\mathbf{e}}{\rightarrow}{\mathbf{t}}){\rightarrow}{\mathbf{t}}}}\ (\lambda x{^{{\mathbf{e}}}} ({\mathop{\&}}{^{{\mathbf{t}}{\rightarrow}({\mathbf{t}}{\rightarrow}{\mathbf{t}})}} (P\ x) (Q\ x))))$
*club* ${\mathbf{e}}{\rightarrow}{\mathbf{t}}$
$\lambda x{^{{\mathbf{e}}}} (\texttt{club}{^{{\mathbf{e}}{\rightarrow}{\mathbf{t}}}}\ x)$
*defeated* ${\mathbf{e}}{\rightarrow}({\mathbf{e}}{\rightarrow}{\mathbf{t}})$
$\lambda y{^{{\mathbf{e}}}}\ \lambda x{^{{\mathbf{e}}}}\ ((\texttt{defeated}{^{{\mathbf{e}}{\rightarrow}({\mathbf{e}}{\rightarrow}{\mathbf{t}})}}\ x) y)$
*Leeds* ${\mathbf{e}}$
Leeds
------------ --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
$$\begin{array}{c}
\Big(\big(\lambda P{^{{\mathbf{e}}{\rightarrow}{\mathbf{t}}}}\ \lambda Q{^{{\mathbf{e}}{\rightarrow}{\mathbf{t}}}}\ (\exists{^{({\mathbf{e}}{\rightarrow}{\mathbf{t}}){\rightarrow}{\mathbf{t}}}}\ (\lambda x{^{{\mathbf{e}}}} ({\mathop{\&}}(P\ x) (Q\ x))))\big)
\big(\lambda x{^{{\mathbf{e}}}} (\texttt{club}{^{{\mathbf{e}}{\rightarrow}{\mathbf{t}}}}\ x)\big)\Big) \\
\Big(
\big(\lambda y{^{{\mathbf{e}}}}\ \lambda x{^{{\mathbf{e}}}}\ ((\texttt{defeated}{^{{\mathbf{e}}{\rightarrow}({\mathbf{e}}{\rightarrow}{\mathbf{t}})}}\ x) y)\big)\ Leeds{^{{\mathbf{e}}}}\Big)\\
\multicolumn{1}{c}{\downarrow \beta}\\
\big(\lambda Q{^{{\mathbf{e}}{\rightarrow}{\mathbf{t}}}}\ (\exists{^{({\mathbf{e}}{\rightarrow}{\mathbf{t}}){\rightarrow}{\mathbf{t}}}}\ (\lambda x{^{{\mathbf{e}}}} ({\mathop{\&}}{^{{\mathbf{t}}{\rightarrow}({\mathbf{t}}{\rightarrow}{\mathbf{t}})}}
(\texttt{club}{^{{\mathbf{e}}{\rightarrow}{\mathbf{t}}}}\ x) (Q\ x))))\big)\\
\big(\lambda x{^{{\mathbf{e}}}} \ ((\texttt{defeated}{^{{\mathbf{e}}{\rightarrow}({\mathbf{e}}{\rightarrow}{\mathbf{t}})}}\ x) Leeds{^{{\mathbf{e}}}})\big)\\
\multicolumn{1}{c}{\downarrow \beta}\\
\big(\exists{^{({\mathbf{e}}{\rightarrow}{\mathbf{t}}){\rightarrow}{\mathbf{t}}}}\ (\lambda x{^{{\mathbf{e}}}} ({\mathop{\&}}(\texttt{club}{^{{\mathbf{e}}{\rightarrow}{\mathbf{t}}}}\ x) ((\texttt{defeated}{^{{\mathbf{e}}{\rightarrow}({\mathbf{e}}{\rightarrow}{\mathbf{t}})}}\ x) Leeds{^{{\mathbf{e}}}})))\big)
\end{array}$$
This $\lambda$-term of type ${\mathbf{t}}$ that can be called the *logical form* of the sentence, represents the following formula of predicate calculus (admittedly more pleasant to read):
$$\exists x:{\mathbf{e}}\ (\texttt{club}(x)\ {\mathop{\&}}\ \mathit{defeated}(x,Leeds))$$
Thus defined, computational semantics looks a miracle. But this miracle does not account for selectional restriction (a [“*book*"]{} cannot [“*bark*"]{}), does not tell anything of the relations between predicates (a [“*book*"]{} can be [“*read*"]{}), nor of the correspondence of predicates with concepts (a book is both an informational and a physical entity), nor of the relation between the logical language for semantics and world knowledge and ontologies. Issues such as selectional restriction have been much discussed in type theoretical framework e.g. in [@Cooper2006dag; @asher-typedriven; @BMRjolli; @Asher2011wow; @Luo2011lacl].
*In his paper *Selectional Restrictions, Types and Categories*, Nicholas Asher discusses the introduction of types as objects of a category (instead of a single ${\mathbf{e}}$) endowed with categorical operations that go beyond the logical ones: he is thus able to correctly account for selectional restriction, felicitous and infelicitous copredications.*
*This framework in which one composes functions to compute meanings and ultimately the truth values in a given model is given a categorical formalisation by Anne Preller in her paper *Natural Language Semantics in Biproduct Dagger Categories*. Starting with pregroup grammars defined in an algebraic structure that can be viewed as non commutative linear logic in which $\wp=\otimes$, she computes the meaning categories with extra operations, as finite dimensional vector spaces or two sorted functions over finite sets. She can then account of meaning with the usual set theoretic interpretations or with bags of words vectors used for text classification.*
An underlying question, in relation to the interpretation of common nouns and to the bags of words approach is what a concept or a base type is, and the interrelations between base types. For instance one can say [“*a cat sleeps*"]{}, because [“*animals*"]{} can [“*sleep*"]{} and [“*cats*"]{} are [“*animals*"]{}. In the kind of model used in semantics ontological relations are quite common, but here is no specific structure to represent these ontological relations.
*The paper entitled *Formal ontologies and coherent spaces* by Michel Abrusci, Christophe Fouqueré and Marco Romano exploits the structure of coherent spaces that we discussed in section \[linear\] to give a formal account of the relation between entities and concepts, encoding properties of formal ontologies.*
#### Thanks:
We are indebted to Mai Gehrke, Alain Lecomte, Anne Preller, Steve Vickers, Gilles Zémor, and especially to François Lamarche and Thomas Streicher for a quick but efficient rereading of this introduction, which they all contributed to improve. Remaining errors and confusions are ours.
[^1]: Explicitely, a presheaf $A$ is a family of sets $A(U)$ (where $U$ runs through all open subsets of $X$) together with restriction maps $A(U)\to A(V)$ for any inclusion $V\subseteq U$ between open subsets of $X$. Of course, these restriction maps can be composed, that is, if $W\subseteq V \subseteq U$ then the restriction map $A(U)\to A(W)$ is equal to the composition $A(U)\to A(V)\to A(W)$.
[^2]: The first map is induced by inclusions $U_i\subseteq U$, and the two other maps are respectively induced by inclusions $U_j\cap U_k\subseteq U_j$ and $U_j\cap U_k\subseteq U_k$.
[^3]: This sheaf $U\mapsto {\mathop{\mathrm{Hom}}}(A|_U,B|_U)$ is often called “internal hom”.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A filtration of a representation whose successive quotients are isomorphic to Demazure modules is called an excellent filtration. In this paper we study graded multiplicities in excellent filtrations of fusion products for the current algebra $\mathfrak{sl}_2[t]$. We give a combinatorial formula for the polynomials encoding these multiplicities in terms of two dimensional lattice paths. Corollaries to our main theorem include a combinatorial interpretation of various objects such as the coeffficients of Ramanujan’s fifth order mock theta functions $\phi_0, \phi_1, \psi_0, \psi_1$, Kostka polynomials for hook partitions and quotients of Chebyshev polynomials. We also get a combinatorial interpretation of the graded multiplicities in a level one flag of a local Weyl module associated to the simple Lie algebras of type $B_n \text{ and } G_2$.'
address:
- 'Université Laval, Départment de mathématiques et de Statistique, Québec, QC, Canada'
- 'University of Bochum, Faculty of Mathematics, Universit[ä]{}tsstr. 150, 44801 Bochum, Germany'
author:
- Rekha Biswal
- Deniz Kus
bibliography:
- 'bibfile.bib'
title: A combinatorial formula for graded multiplicities in excellent filtrations
---
secnumfontstartsection[section]{}[1]{} @[.7plus]{}[.5]{} [<span style="font-variant:small-caps;"></span>]{}
===========================================================================================================
secnumfont[****]{}startsection[subsection]{}[2]{} [.5plus.7]{}[-.5em]{} [****]{}
--------------------------------------------------------------------------------
[^1]
[^2]
Introduction
============
In this paper, we are interested in objects in the category of finite-dimensional ${\mathbb Z}$–graded modules for the current algebra $\mathfrak{sl}_2[t]$ which admit a filtration whose successive quotients are isomorphic to stable Demazure modules. These types of filtrations are called excellent filtrations [@Jo03; @J06] but are also known in the literature as Demazure flags [@BCSV15; @CSSW14]. Demazure modules occur in highest weight integrable modules of the affine Lie algebra $\widehat{\mathfrak{sl}_2}$ and are called stable if they admit an action of the maximal parabolic subalgebra $\mathfrak{sl}_2[t]$; the integer given by the action of the canonical central element is called its level. The great interest in Demazure modules is due to several reasons. They appear as classical limits of a family of irreducible representations of the quantum affine algebra [@BCMo15; @CP01], their graded characters are specializations of Macdonald polynomials [@I03] and satisfy certain functional relations which are known as $Q$–systems [@CV13; @KV14], just to name a few.
An important result due to Naoi states that if $m \geq m' \geq 1$, then a stable Demazure module of level $m'$ admits a filtration such that the successive quotients are isomorphic to level $m$–Demazure modules. In fact Naoi proves this result for an affine Lie algebra associated to a simply laced simple Lie algebra. Later, this result was greatly extended for a wider class of modules for $\mathfrak{sl}_2[t]$, which are indexed by partitions and are known in the literature as fusion products [@FL99]. They generalize many known families of representations for $\mathfrak{sl}_2[t]$, e.g. stable Demazure modules [@CV13 Theorem 2], local Weyl modules [@CP01] and tuncated Weyl modules [@KL15 Theorem 4.3]. Under a suitable condition on the partition, fusion products admit a level $m$–Demazure flag [@CSSW14 Theorem 3.3] and recurrence relations were established in the same paper. A closed form solution of these recurrences was however, only obtained in very special cases (see Remark \[rem21\]). It turned out that in these special cases there is a beautiful link between these modules and number theory and combinatorics: numerical multiplicities are closely related to Chebyshev polynomials [@BCSV15 Corollary 1.3], several specializations of the generting series associated to the graded multiplicities specialize to Ramanujan’s fifth order mock theta functions [@BCSV15 Theorem 1.6.], certain weighted versions of the generating series give Carlitz q–Fibonacci polynomials [@BCK16 Proposition 2.5.3] and are limits of hypergeometric series [@BCK16 Section 2.5.4]. One of the motivations of the present paper is to complete the results of [@BCSV15; @CSSW14] by finding a formula for the graded multiplicities in a level $m$–Demazure flag of an arbitrary fusion product and to further understand the deep and unexpected link between the theory of Demazure flags and combinatorics. The main ingredients in our study are two dimensional lattice paths.
A lattice path is a path in a lattice in some euclidean vector space. They have a long history and have entered many fields of mathematic, computer science and physics; for a survey of results in the enumeration of lattice paths we refer the reader to [@Kra15]. A diagonal lattice path which consist only of up–steps $(1,1)$ and down–steps $(1,-1)$ and which do not pass below the $x$–axis is called a Dyck path. There are several statistics on the set of Dyck paths and we are interested in the comajor statistics (see Section \[section35\]) which was studied first by MacMahon [@MacM60] in his interpretation of the $q$–Catalan numbers. Our strategy for determining graded multiplicities in excellent filtrations of fusion products is as follows.
Let $\xi$ a partition and $V(\xi)$ the fusion product associated to $\xi$. We fix a level $m$–Demazure module $\tau_p^* D(m,n)$ of heighest weight $n$ whose degree is $p$ (see Section \[section2\] for the precise definitions). Further, set $$\mathcal{V}_{n}^{\xi\rightarrow m}(q)=\sum_{p\ge 0}\ [V(\xi):\tau_p^* D(m,n)] \hspace{0,04cm} q^p,$$ where $[V(\xi):\tau_p^* D(m,n)]$ is the multiplicity of $\tau_p^* D(m,n)$ in a level $m$–Demazure flag of $V(\xi)$. In the first step, we reduce the computation of graded multiplicities in excellent filtrations of fusion products to the computation of graded multiplicities in local Weyl modules. To be more precise, we show that there exists a polynomial in finitely many variables such that a suitable evaluation of that polynomial at elements of the form $\mathcal{V}_{\bullet}(q)^{(1,1,\dots,1)\rightarrow \bullet}$ gives the desired polynomial $\mathcal{V}_n(q)^{\xi\rightarrow m}$ (see Poposition \[reductionpr\]). In the second step we give a combinatorial formula for $\mathcal{V}_{n}^{\xi\rightarrow m}(q)$ in terms of Dyck paths when $\xi$ is a hook partition. Let $m,m'\in{\mathbb N}$, $s\in{\mathbb Z}_+$ such that $m\geq m'$ and $\xi=(m',1^s)$ a hook partition. Then we have (see Theorem \[mainthm1\]) $$\mathcal{V}^{\xi\rightarrow m}_{n}(q)=\sum_{P} q^{\text{comaj(P)}},$$ where the sum runs over the set of admissible Dyck paths (see Definition\[maindef\]) from the origin $(0,0)$ to $(s+m',n)$ which start with $m'$ up–steps and never cross the line $y=\max\{m-1,n\}$.
The appearance of Dyck paths in this set up is quite unexpected and intriguing. A number of consequences flow from our new combinatorial formula. We briefly summarize the consequences below:
- We find new combinatorial interpretations of Ramanujan’s fifth order mock theta functions $\phi_1, \phi_0, \psi_1, \psi_0$ in terms of Dyck paths and their comajor index (see Corollary \[mockcoef\]).
- We find a combinatorial interpretation of the quotients of powers of two Chebyshev polynomials in terms of Dyck paths of bounded height generalizing the results of Gessel and Xin [@GX05] (see Proposition \[numerical\]).
- The powers of $q$ which appear with non–zero coefficients in the polynomial encoding the graded multiplicities in Demazure flags of local Weyl modules form an interval of consecutive integers (see Corollary \[conint\]).
- The graded multiplicities of irreducible modules in local Weyl modules are principal specializations of Schur functions (see Corollary \[princp\]).
- We give formulae for the graded multiplicities of an irreducible module in a Demazure flag of a local Weyl module (see Corollary \[KrM\]).
Our paper is organized as follows. In Section 2 we recall the notion of excellent filtrations. In Section 3 we define our combinatorial model which is the crucial object in our main theorem. In Section 4 we reduce the computation of graded multiplicities in excellent filtrations of fusion products to the computation of graded multiplicities in local Weyl modules and state our main theorem. We also discuss several corollaries. In Section 5 we give a proof of our main theorem by using a different recursion which is obtained from a representation theoretical result proved in Section 6.
*Acknowledgements: R.B. thanks Francois Bergeron, Ira Gessel and Dennis Stanton for many fruitful discussions about Chebyshev polynomials which was the starting point of this work. D.K. thanks Christian Krattenthaler for many helpful discussions on the combinatorics of Dyck paths and for pointing him to formula . R.B. also gratefully acknowledges the funding received from NSERC discovery grant of her Postdoc supervisor Michael Lau and Université Laval for the hospitality.*
Excellent filtrations and graded multiplicities {#section2}
===============================================
In this section we recall the notion of excellent filtrations (Demazure flags) [@Jo03; @J06] and set up the notation needed in the rest of the paper.
Throughout this paper we denote by $\mathbb{C}$ the field of complex numbers and by $\mathbb{Z}$ (resp. $\mathbb{Z}_{+}$, $\mathbb{N}$) the subset of integers (resp. non-negative, positive integers). Given $n,m\in{\mathbb Z}$, set $$\begin{gathered}
{\genfrac[]{0pt}0{n}{m}}_q=\frac{(1-q^n)\cdots(1-q^{n-m+1})}{(1-q)\cdots (1-q^m)}, \ \ \ n\geq m>0 ,\\ \\ {\genfrac[]{0pt}0{n}{0}}_q=1,\ \ n\geq 0,\ \ {\genfrac[]{0pt}0{n}{m}}_q= 0,\ \ \text{otherwise}.\end{gathered}$$ Further, we introduce the $q$–Pochammer symbol $(a;q)_{n}=\prod_{i=0}^{n-1}(1-aq^{i}).$
Let ${\mathfrak{sl}}_2[t]\cong{\mathfrak{sl}}_2\otimes {\mathbb C}[t]$ the Lie algebra of two by two matrices of trace zero with entries in the algebra ${\mathbb C}[t]$ of polynomials with complex coefficients. The degree grading of ${\mathbb C}[t]$ defines a natural grading on ${\mathfrak{sl}}_2[t]$. A finite-dimensional ${\mathbb Z}$-graded ${\mathfrak{sl}}_2[t]$–module is a ${\mathbb Z}$–graded vector space admitting a compatible graded action of ${\mathfrak{sl}}_2[t]$: $$V=\bigoplus_{k\in{\mathbb Z}} V[k],\qquad (a\otimes t^r)V[k]\subset V[k+r]\ \ a\in {\mathfrak{sl}}_2,\ \ r\in{\mathbb Z}_+.$$ Given a ${\mathbb Z}$–graded space $V$ let $\tau_p^* V$ the graded vector space whose $r$–th graded piece is $V[r+p]$.
{#section23}
The category of finite-dimensional ${\mathbb Z}$–graded ${\mathfrak{sl}}_2[t]$–modules was the central subject of many recent papers (see, for example [@BC15; @BCM12; @BCSV15; @CSSW14; @CV13; @FL99; @KL15]). There is a well–known family of objects in that category which is of particular interest, namely the subclass of fusion products. We recall their description in terms of generators and relations from [@CV13 Section 6]; for a more traditional definition we refer the reader to [@FL99] (see also Section \[551\]). Let $x,h,y$ be the standard basis of ${\mathfrak{sl}}_2$ and set $u^{(r)}:=\frac{1}{r!} u^r$ for $u\in {\mathfrak{sl}}_2[t],\hspace{0,03cm} r\in{\mathbb Z}_+$. For a partition $\xi=(\xi_1\geq \xi_2\geq \cdots \geq \xi_\ell)$ we set $$|\xi|_i:=\sum_{j=i}^{\ell}\xi_j,\ 1\leq i \leq \ell, \ \ |\xi|:=|\xi|_1.$$ The fusion product associated to $\xi$ is the ${\mathfrak{sl}}_2[t]$–module $V(\xi)$ generated by an element $v_{\xi}$ with defining relations: $$\begin{gathered}
\label{demreldes}(x\otimes {\mathbb C}[t])v_{\xi}=0,\ \ (h\otimes f) v_{\xi}= |\xi| f(0)v_{\xi},\ \ \ (y\otimes 1)^{|\xi|+1}v_{\xi}=0,\\ \label{demrel2a}
(x\otimes t)^{(p)}(y\otimes 1)^{(r+p)}v_{\xi}=0,\ \ r,p\in {\mathbb N},\ r+p\geq 1+rk+\sum_{j\geq k+1}\xi_j\ \ \text{for some $k\in {\mathbb N}$.}\end{gathered}$$ It turns out that many other well–known families of representations belong to the class of fusion products. For example, Demazure modules occur in irreducible integrable representations of the affine Lie algebra $\widehat{\mathfrak{sl}_2}$ and are parametrized by tuples $(m,n)\in {\mathbb N}\times {\mathbb Z}_+$, where the integer $m$ is called the level. We denote such a module by $D(m,n)$. If $n_0,n_1\in{\mathbb Z}_+$ are such that $n_0<m$ and $n=n_1m+n_0$, then the fusion product $V(\xi(m,n))$ associated to the partition $\xi(m,n):=(m^{n_1},n_0)$ is isomorphic to $D(m,n)$ (see [@CV13 Theorem 2]). Hence Demazure modules can be categorized into the family of fusion products, but the class of fusion products is generically much bigger. Nevertheless, there is a beautiful result saying that fusion products admit a filtration by Demazure modules under a suitable condition on the partition. The following proposition was proved in [@CSSW14 Theorem 3.3].
\[existflag\] Let $\xi=(\xi_1\geq \xi_2\geq \cdots \geq \xi_{\ell})$ a partition and $m\in{\mathbb N}$. The module $V(\xi)$ admits a filtration of level $m$, i.e., there exists a decreasing sequence of graded submodules $$0=V_0\subset V_1\subset \cdots V_{k-1}\subset V_k=V(\xi)$$ such that $$V_i/V_{i-1}\cong \tau_{p_i}^*D(m, n_i),\ \ (p_i,n_i)\in {\mathbb Z}_+\times {\mathbb Z}_+,\ \ 1\leq i\leq k$$ if and only if $m\geq \xi_1$.
These types of filtrations are called excellent filtrations [@Jo03; @J06] but are also known in the literature as level $m$–Demazure flags [@BCSV15; @CSSW14]. The aim of the present paper is to give a combinatorial formula for the graded multiplicities in excellent filtrations.
\[rem23\] We discuss two further specializations of the partition $\xi$.
1. The specialization $\xi_i=1$, $1\leq i\leq \ell$ leads to a representation which is isomorphic to the local Weyl module $W_{\text{loc}}(|\xi|)$ which is the module generated by an element $w_{|\xi|}$ subject to the relations . The interest in local Weyl modules is its connections with quantum affine algebras [@CP01], the theory of Macdonald polynomials [@I03], $q$–Whittaker functions [@BF14a], $Q$–systems [@CV13; @KV14] and more recently with hypergeometric series [@BCK16].
2. Let $N\in {\mathbb N}$. The truncated Weyl module $W_{\text{loc}}(N,n)$ is a quotient of $W_{\text{loc}}(n)$ by the additional relation $(\mathfrak{sl}_2\otimes t^N{\mathbb C}[t])w_{n}=0$. The special choice $\xi=((d+1)^j,d^{N-j})$ yields an isomorphism $V(\xi)\cong W_{\text{loc}}(N,n)$, where $d\in {\mathbb Z}_+$ and $j<N$ are such that $n=dN+j$ (see [@KL15 Theorem 4.3]).
We will secretly assume in the rest of the paper that $m\geq \xi_1$ whenever we talk about level $m$–Demazure flags of $V(\xi)$.
{#section-2}
The number of times a particular level $m$–Demazure module appears as a quotient in a level $m$–flag is independent of the choice of the flag. We encode these multiplicities in a polynomial $$\label{poly}
\mathcal{V}_{n}^{\xi\rightarrow m}(q):=\sum_{p\ge 0}\ [V(\xi):\tau_p^* D(m,n)] \hspace{0,04cm} q^p,$$ where $[V(\xi):\tau_p^* D(m,n)]=\sharp\{1\leq i\leq k : V_i/V_{i-1}\cong \tau_p^* D(m,n)\}$. It is known that $$\label{initial12}
\mathcal{V}_{s}^{\xi(m',s)\rightarrow m}(q)=1,\quad \mathcal{V}_{n}^{\xi(m,s)\rightarrow m}(q)=\delta_{s,n},\ \ \mathcal{V}_{n}^{\xi\rightarrow m}(q)=0,\ \ \text{if }\ |\xi|-n\notin2{\mathbb Z}_+.$$ Moreover, for $m\geq m'\geq \xi_1$ we have $$\label{mat}\mathcal{V}_{n}^{\xi\rightarrow m}(q)=\sum_{p\geq 0} \mathcal{V}_{p}^{\xi\rightarrow m'}(q)\ \mathcal{V}_{n}^{\xi(m',p)\rightarrow m}(q).$$ For convinience, set $$\label{initial1}
\mathcal{V}_{n}^{\xi\rightarrow m}(q)=0,\ \ {\rm{if}} \ \ |\xi|<0\ \ {\rm{or}}\ n<0.$$ The following lemma will be needed later; for a proof see [@CSSW14 Lemma 3.8].
\[ausc\]Let $\xi=(\xi_1\geq \xi_2\geq \cdots \geq \xi_{\ell})$ a partition and set $\xi'=(\xi_2\geq \xi_3\geq \cdots \geq \xi_{\ell})$. We have $$\mathcal{V}_{n}^{\xi\rightarrow \xi_1}(q)=q^{(|\xi|-n)/2}\hspace{0,04cm} \mathcal{V}_{n-\xi_1}^{\xi'\rightarrow \xi_1}(q).$$
We will freuquently deal with the polynomials when $\xi=\xi(m',s)$, i.e. $V(\xi)$ itself is a level Demazure module of level $m'$. So we abbreviate $$\mathcal{V}_{s,n}^{m'\rightarrow m}(q):=\mathcal{V}_{n}^{\xi(m',s)\rightarrow m}(q)$$ and define the associated generating series by $$\mathcal{A}^{m'\rightarrow m}_n(x,q)=\sum_{s\geq 0}\mathcal{V}_{s,n}^{m'\rightarrow m}(q)\,x^{(s-n)/2},\ \ \ n\ge 0.$$
\[rem21\] The proof of Proposition \[existflag\] uses the short exact sequence of [@CV13 Theorem 5]. It has the advantage that one can derive recursive formulas for $\mathcal{V}_{s,n}^{m'\rightarrow m}(q)$ (see for example [@CSVW14 Theorem 2.2]) but closed formulas were established only in the case of $m'=1,m=2$ [@CSSW14 Theorem 3.3] and $m'=1,m=3$ [@CSVW14 Section 1.6]. Our approach uses a different short exact sequence (see Section \[section6\]) which enables us to give a combinatorial formula in the most general setting.
Combinatorics of Dyck paths {#section3}
===========================
In this section we introduce our combinatorial model and certain combinatorial statistics which will be needed to describe the polynomials .
{#section31}
A Dyck path is a diagonal lattice path from the origin $(0,0)$ to $(s,n)$ for some non–negative integrs $s,n\in {\mathbb Z}_+$, such that the path never goes below the x–axis. We encode such a path by a 01–word, where $1$ encodes the up–steps and $0$ the down–steps. For a path $P$ we set $$\text{supp}(P):=\{z\in {\mathbb Z}_+^2 : z \text{ is a point on $P$}\}.$$ The length of a point $z$ on a path $P$ is defined to be its $x$–coordinate and its $y$–coordinate is called the height; we will frequently use the notation $z=(\ell_z(P),\text{ht}_z(P))$. We denote by $\mathcal{D}_{n}$ the set of Dyck paths that end at height $n$ and by $\mathcal{D}_n(s)$ the subset of paths $P$ with $\ell(P)=s$, where $\ell(P)$ denotes the length of the endpoint of $P$. Obviously (compare with ), $$\mathcal{D}_{n}=\bigcup_{s\geq 0} \mathcal{D}_n(s),\ \ |\mathcal{D}_s(s)|=1, \ \ |\mathcal{D}_n(0)|=\delta_{n,0}, \ \ \mathcal{D}_n(s)=\emptyset, \ \text{ if } s-n\notin 2{\mathbb Z}_+.$$ Further, let $\mathcal{D}^{m}_n$ the subset of paths which do not cross the line $y=m$, i.e. $$\mathcal{D}^{m}_n=\{P\in \mathcal{D}_{n}: \text{ht}_z(P)\leq m \text{ for all } z\in \text{supp}(P)\},\ \ \mathcal{D}_n^{m}(s):=\mathcal{D}_n^m\cap \mathcal{D}_n(s).$$
(0,0) grid (20,6); (0,0) – (3,3) – (5,1) – (8,4)–(12,0)–(17,5)–(20,2);
In analogy to we set $$\label{initial2}
\mathcal{D}_n(s)=0,\ \ {\rm{if}} \ \ s<0\ \ {\rm{or}}\ n<0.$$
In the literature a Dyck path mostly ends at height $0$ and a diagonal lattice path that ends at an arbitrary height is often called a generalized Dyck path. It is well–known that the cardinality of $\mathcal{D}_{0}(2k)$ is given by the $k$–th Catalan number; for further combinatorial interpretations of the Catalan numbers we refer to the book of Stanley [@S99 pg. 219-229].
{#section-3}
The generating function $\mathcal{G}_n^m(x)=\sum_{s\geq 0}|\mathcal{D}^m_{n}(s)| x^{(s-n)/2}$ has been intensively studied in [@GX05]. It turns out that $\mathcal{G}_n^m(x)$ is a rational function which can be expressed in terms of Chebyshev polynomials. We shall explain this connection in more detail. The Chebyshev polynomials of the second kind are defined by the recurrence relation $$U_n(x)=2x U_{n-1}(x)-U_{n-2}(x),\ \ n\geq 2,$$ with initial data $U_0(x)=1$ and $U_1(x)=2x$. Define polynomials $$p_n(x):=x^{n/2}U_n((2\sqrt{x})^{-1})=\sum_{s=0}^{\lfloor\frac{k}{2}\rfloor} (-1)^s {\genfrac[]{0pt}0{k-s}{s}}.$$ The first equality of the next lemma follows from [@GX05 Lemma 4.2] and the second equality has been proved in [@BCSV15 Corollary 1.3].
\[Ges\] We have $\mathcal{G}^m_n(x)=0$ if $m<n$ and otherwise $$\mathcal{G}^m_n(x)=\frac{p_{m-n}(x)}{p_{m+1}(x)}=\mathcal{A}_{n}^{1\rightarrow m+1}(x,1).$$ In particular, $$\mathcal{V}_{s,n}^{1\rightarrow m+1}(1)=|\mathcal{D}^m_n(s)|,\ \ \forall n\leq m.$$
This result suggests a deeper connection between graded multiplicities in Demazure flags and certain Dyck path statistics and the motivation of the present paper is to determine this connection. As a byproduct we will generalize the above lemma and find a combinatorial model (in terms of Dyck paths) whose generating series is $\mathcal{A}^{1\rightarrow m}_n(x,1)$ (even for $n\geq m$) (see Proposition \[numerical\]).
{#admissibledef}
A point $(z_1,z_2)\in \text{supp}(P)$ is called a peak (resp. valley) of the path if $(z_1\pm 1,z_2-1)\in \text{supp}(P)$ (resp. $(z_1\pm 1,z_2+1)\in \text{supp}(P)$). By convention, we will also call the endpoint of a path which is immediately preceded by a down–step a valley. For example, the path in Figure \[fig1\] has three peaks and three valleys. Given a pair of non-negative integers $(a,b)\in {\mathbb Z}^2_+$, we say that $P\in \mathcal{D}_n$ is $(a,b)$–admissible if and only if $P$ satisfies the following property $$\label{admissible} \text{$P$ has a peak $z$ at height $b$} \Rightarrow \text{$\text{ht}_{z'}(P)>a$ for all $z'\in \text{supp}(P)$ with $\ell_{z'}(P)\geq \ell_{z}(P)$}.$$ With other words, once the path has a peak at height $b$, the remaining subsequent path is strictly above the line $y=a$ (see Figure \[fig55\] for a pictorial illustration).
(0,0) grid (16,6); (0,0) – (3,3); (3,3) – (4,4) – (5,3); (4,1) – (16,1); at (0,1) [$y=a$]{}; at (0,4) [$y=b$]{};
{#section-4}
For $n,m\in {\mathbb Z}_+$ let $n_0,n_1\in{\mathbb Z}_+$ such that $n_0<m$ and $n=m n_1+n_0$. *In what follows we set $N=\max\{m-1,n\}$.* We are interested in a suitable subset of $\mathcal{D}^N_n$ which we will call admissible Dyck path. If $n<m$ we set $A(m,n)=\emptyset$ and otherwise define $$A(m,n):=\{(i_1,m),(i_2,m+1),\dots,(i_{n-m+1},n)\}\subset {\mathbb Z}_+^2$$ where $i_1<\cdots<i_{n-m+1}$ is the natural ordering of the set $$\{0,\dots,n\}\backslash \{pn_1+n_0+\min\{0,(p-1)-n_0\},\ 1\leq p\leq m\}.$$ Equivalently, $$\begin{aligned}
A(m,n)=\notag &\{(p(n_1+1)+r,m+pn_1+r),\ 0\leq r<n_1,\ 0\leq p\leq n_0\}\ \cup &\\& \notag \{(pn_1+(n_0+1)+r,m+p(n_1-1)+(n_0+1)+r),\ 0\leq r<n_1-1,\ n_0< p<m\}.\end{aligned}$$ The following is straightforward: $$\label{1}
(a,b)\in A(m,n)\backslash\{(0,m)\} \Rightarrow \exists \ \tilde a<a: \ \ (\tilde a,b-1)\in A(m,n).$$
\[maindef\] We call a path $P\in \mathcal{D}_n^N$ admissible iff $P$ is $(a,b)$–admissible for all $(a,b)\in A(m,n)$. We denote by $\mathcal{D}_{m,n}$ the set of admissible Dyck paths and set $\mathcal{D}_{m,n}(s):=\mathcal{D}_n^N(s)\cap \mathcal{D}_{m,n}$.
Both sets will play a major role in the description of the generating series and graded multiplicities respectively, see Theorem \[mainthm1\].
We have $A(n,n)=\{(0,n)\}$ and hence $\mathcal{D}_{n,n}$ consists of all Dyck paths $P\in \mathcal{D}^n_{n}$ which do not return to the $x$–axis after a peak of height $n$. The green path in Figure \[fig3\] is admissible and the red path violates the condition and is not admissible.
(0,0) grid (16,5); (0,0) – (4,4) – (6,2) – (7,3) – (8,2) – (10,4) – (13,1) – (16,4);
(0,0) grid (16,5); (0,0) – (4,4) – (6,2) – (7,3) – (10,0) – (12,2) – (13,1) – (16,4);
{#section35}
The major statistics of a Dyck path was studied first by MacMahon [@MacM60] in his interpretation of the $q$–Catalan numbers. Let $P=a_1\cdots a_{s}$, $a_i\in\{0,1\}$ a Dyck path of length $s$. The major and comajor index are defined by $$\text{maj}(P)=\sum_{\substack{1\leq i < s,\\ a_i>a_{i+1}}} i,\ \ \ \text{comaj}(P)=\sum_{\substack{1\leq i < s,\\ a_i>a_{i+1}}} (s-i).$$
\[rem1\] The comajor index can also be defined for standard Young tableaux and play an important role in the expression of the character of the space of harmonics (the vector space spanned by the Vandermonde determinant and its partial derivatives of all orders; see the work of Haiman [@Hai94]). The definition is essentially the same in the sense that there is a bijection from $\mathcal{D}_n(n+2k)$ to the set of standard Young tableaux of shape $\lambda=(n+k,k)$, which preserves the comajor statistics (see Section \[section33\]). However, the tranlation of the admissibility conditions seem to be quite technical and hence we prefer to work with the notion of Dyck paths instead of two row partitions.
The main results {#section4}
================
In this section we summarize the main results of the paper. We have arranged it so that this section can be read essentially independently of the representation theory of $\mathfrak{sl}_2[t]$.
{#section-5}
We will determine the polynomial in two steps. In the first step we will reduce the computation of graded multiplicities in Demazure flags of fusion products to the computation of graded multiplicities in local Weyl modules only. To be more precise, we will show that there exists a polynomial in finitely many variables such that a suitable evaluation of that polynomial at elements of the form $\mathcal{V}^{1\rightarrow \xi_i}_{\bullet,\bullet}(q)$ for $1\leq i\leq \ell$ gives the desired polynomial $\mathcal{V}^{\xi\rightarrow m}_{n}(q)$. In the second step we give a combinatorial formula for graded multiplicities in Demazure flags of local Weyl modules in terms of Dyck paths.
\[reductionpr\]
1. Let $\xi_0\in {\mathbb N}$ and $\xi=(\xi_1\geq \xi_2\geq \cdots \geq \xi_{\ell})$, $\ell\geq 2$, a partition such that $\xi_0\geq \xi_1$. Then $$\mathcal{V}_{|\xi|-2k}^{\xi\rightarrow \xi_0}(q)=\sum_{ 0=p_0\leq p_1\leq \cdots \leq p_{\ell-2}\leq p_{\ell-1}=k} q^{p_1+\cdots+p_{\ell-2}}\ \prod_{i=1}^{\ell-1} \mathcal{V}_{|\xi|_i-2p_{\ell-i-1},|\xi|_i-2p_{\ell-i}}^{\xi_{i}\rightarrow \xi_{i-1}}(q).$$
2. If $m,m'\in {\mathbb N}$ and $s,n\in{\mathbb Z}_+$ are such that $m \geq m'$, then $$\mathcal{V}^{m'\rightarrow m}_{s,n}(q)=\sum_{j=0}^{s}\left( \sum_{\ell=0}^j(-1)^{\ell+1}\hspace{-0,5cm} \sum_{0=p_0<p_1<\cdots<p_{\ell}<p_{\ell+1}=j}\ \prod_{i=0}^{\ell}\ \mathcal{V}^{1\rightarrow m'}_{s-2p_i,s-2p_{i+1}}(q)\right)\ \mathcal{V}^{1\rightarrow m}_{s-2j,n}(q).$$
We prove the first part by an induction on $\ell$. If $\ell=2$, the discussion in Section \[section23\] implies that $V(\xi)\cong D(\xi_1,|\xi|)$. Hence $\mathcal{V}_{|\xi|-2k}^{\xi\rightarrow \xi_0}(q)=\mathcal{V}_{|\xi|,|\xi|-2k}^{\xi_{1}\rightarrow \xi_0}(q),$ and the induction begins. If $\ell>2$, we have $$\begin{aligned}
&& \mathcal{V}_{|\xi|-2k}^{\xi\rightarrow \xi_0}(q)&=\sum_{p=0}^k \mathcal{V}_{|\xi|-2p}^{\xi\rightarrow \xi_1}(q)\ \mathcal{V}_{|\xi|-2p,|\xi|-2k}^{\xi_1\rightarrow \xi_0}(q)&& \text{by \eqref{mat}}\\
&& &=\sum_{p=0}^k q^p\hspace{0.05cm}\mathcal{V}_{|\xi|_2-2p}^{\xi'\rightarrow \xi_1}(q)\ \mathcal{V}_{|\xi|-2p,|\xi|-2k}^{\xi_1\rightarrow \xi_0}(q)&& \text{by Lemma~\ref{ausc}}.\end{aligned}$$ The claim now follows by a trivial application of the induction hypothesis.
Now we prove the second part of the proposition. If $s-n\notin 2{\mathbb Z}_+$ both sides of the equation are zero and the statment is trivial. Otherwise assume that $s=n+2k$ for some $k\in{\mathbb Z}_+$. We can rewrite as a system of linear equations $Uv=b$, where $U=(\mathcal{V}_{i,j})_{0\leq i,j\leq k}$ is an upper triangular unipotent matrix of size $(k+1)\times (k+1)$ and $$\mathcal{V}_{i,j}=\mathcal{V}^{1\rightarrow m'}_{s-2i,s-2j}(q),\ \ b_i=\mathcal{V}^{1\rightarrow m}_{s-2i,n}(q),\ \ v_i=\mathcal{V}^{m'\rightarrow m}_{s-2i,n}(q).$$ Now it is straightforward to check that the inverse of $U$ is again upper triangular unipotent whose entries are given by $f_{i,j}(\mathcal{V}),\ 0\leq i,j\leq k,$ where $f_{i,j}(\mathcal{V})=\delta_{i,j}$ if $i\geq j$ and otherwise $$f_{i,j}(\mathcal{V})=\sum_{r\geq 0}(-1)^{r+1} \sum_{i=p_0<p_1<\cdots<p_{r}<p_{r+1}=j}\mathcal{V}_{p_0,p_1}\mathcal{V}_{p_1,p_2}\cdots \mathcal{V}_{p_r,p_{r+1}}.$$ Now the first entry of the vector $v=U^{-1}b$ gives the desired result.
Let $V(\xi)$ be a truncated Wey module, i.e. $\xi=((d+1)^j,d^{N-j})$ where $N,d,j\in {\mathbb Z}_+$ are such that $0<j<N$ and $|\xi|=dN+j$. Applying Proposition \[reductionpr\] to this setting gives $$\mathcal{V}_{|\xi|-2k}^{\xi\rightarrow \xi_0}(q)=\sum_{p=0}^k q^p\ \mathcal{V}_{(N-j)d,(N-j)d-2p}^{d \rightarrow d+1}(q)\ \mathcal{V}_{|\xi|-2p,|\xi|-2k}^{d+1 \rightarrow \xi_0}(q).$$
{#section-6}
After the reduction in Proposition \[reductionpr\], our second result focussed on a combinatorial formula for the graded multiplicities in a level $m$–Demazure flag of a local Weyl module. Recall the definition of admissible Dyck paths from Definition \[maindef\].
\[mainthm1\] Let $m\in{\mathbb N}$, $n\in{\mathbb Z}_+$. We have, $$\mathcal{A}^{1\rightarrow m}_n(x,q)=\sum_{P\in \mathcal{D}_{m,n}}q^{\text{comaj}(P)}\ x^{d(P)},$$ where $d(P)$ denotes the number of down–steps of $P$. In particular, for any $s\in{\mathbb Z}_+$ we get $$\mathcal{V}^{1\rightarrow m}_{s,n}(q)=\sum_{P\in\mathcal{D}_{m,n}(s)}q^{\text{comaj}(P)}.$$
The proof of the above theorem will be postponed to Section \[section5\], but to avoid technical difficulties in the rest of the paper we will handle the case $m=1$ seperately. If $m=1$ we have $A(1,n)=\{(0,1),(1,2),\dots,(n-1,n)\}$ and hence $$\mathcal{D}_{1,n}=\{\underbrace{11\cdots 1}_{n}\}=\mathcal{D}_{1,n}(n).$$ This yields, $$\sum_{P\in \mathcal{D}_{1,n}(s)}q^{\text{comaj}(P)}=\delta_{s,n}$$ and the theorem follows immediately from . *So we will assume from now on that $m\geq 2.$*
If $n=3$, $m=2$ and $s=5$ we have $A(2,3)=\{(0,2),(2,3)\}$ and hence $\mathcal{D}_{2,3}(5)=\{10111, 11011\}$. We get $\mathcal{V}_{5,3}^{1\rightarrow 2}(q)=q^3+q^4$.
(0,0) grid (5,3); (0,0) – (1,1)–(2,0) –(5,3);
(0,0) grid (5,3); (0,0) – (2,2) – (3,1)–(5,3);
\[fig3\]
We shall prove in Section \[section5\] a slightly more general version of Theorem \[mainthm1\], namely we will give a combinatorial formula for the polynomials in terms of Dyck paths for every hook partition $\xi$ (see Theorem \[hook\]).
{#section-7}
Before we proceed to the proof of Theorem \[mainthm1\] we will discuss several consequences. For simple finite-dimensional Lie algebras of non–simply laced type local Weyl modules are in general not isomorphic to level one Demazure modules. Nevertheless, Naoi proved in [@Na11 Sections 4 and 9] that a local Weyl module (see for example [@Na11 Definition 3.1] for a precise definition) admits a level one Demazure flag. If ${\mathfrak{g}}$ is of type $B_n$ or $G_2$ one can use the $\mathfrak{sl_2}$–theory (in particular, the combinatorial formula in Theorem \[mainthm1\]) to give combinatorial formulas for $\sum_p [W(\lambda): \tau_p^* D(1,\mu)]q^p$, where $\lambda,\mu$ are dominant integral ${\mathfrak{g}}$–weights. This fact follows from an inspection of the proof given in [@Na11] and a precise statement can be found in [@CSSW14 Proposition 2.5].
Let ${\mathfrak{g}}$ of type $B_n$ and $m=2$ or $G_2$ and $m=3$. Further denote by $\alpha$ the unique simple short root with coroot $h_{\alpha}$. Let $\lambda,\mu$ two dominant integral weights of ${\mathfrak{g}}$. We have $$[W(\lambda): \tau_p^* D(1,\mu)]= \begin{cases}|\{P\in \mathcal{D}_{m,\mu(h_{\alpha})}(\lambda(h_{\alpha})): \text{comaj}(P)=p\}|,& \lambda-\mu\in {\mathbb Z}_+\alpha\\
0,& \text{otherwise.}
\end{cases}$$
{#section-8}
In the special case when $m=2$, we can derive another combinatorial description of the graded multiplicities in terms of bounded partitions which we record in the next lemma. This result can also be obtained by a straightforward calculation combining [@CSSW14 Theorem 3.3] and [@And98 Chapter 3] as pointed out in [@MJ16 Section 4.3]. Our proof is different and uses only the combinatorial formula stated in Theorem \[mainthm1\]. For integers $a,b,c\in{\mathbb Z}$, denote by $\rho^{a}_{b}(c)$ the set of partitions of $c$ with at most $a$ parts such that each part is bounded by $b$. Recall the integers $n_0,n_1$ from Section \[section23\]
\[lpart\] Let $k,n\in {\mathbb Z}_+$. Then we have $$\mathcal{V}^{1\rightarrow 2}_{n+2k,n}(q)=\sum^{k(k+n)}_{\ell=0} |\rho^{k}_{n_1}(k(k+n)-\ell)|\hspace{0,04cm} q^{\ell}.$$
If $0\leq n\leq 1$ we have $n_1=0$ and hence the right hand side is equal to $q^{k^2}$. Since each path in $\mathcal{D}_{n}^{1}(n+2k)$ has height $\leq 1$ the same is true for the left hand side and the lemma is immediate; so assume that $n>1$. Let $P\in \mathcal{D}_{2,n}(n+2k)$ such that $\text{comaj}(P)=\ell$. Since $$A(2,n)=\{(i,i+2),(p,p+1): 0\leq i\leq n_1-1,\ n_1<p\leq n-1\}$$ we see that any admissible path has no two adjacent down–steps and the last $n_1$ steps are up–steps. Hence $P$ is uniquely determined by the $x$–coordinates of all its peaks and the number of down–steps is equal to the number of peaks (= $k$). Assume that $i_1,\dots,i_{k}$ are these $x$–coordinates, then we must have $i_{p+1}\geq i_p+2$ for all $1\leq p<k$ and $$1\leq i_1<i_2<\cdots<i_{k}<2k+n_1,\ \ \sum_{p=1}^{k} i_p=(n+2k)k-\ell.$$ Now substituting $i_p'=i_p-2p+1$ we get a partition $$0\leq i'_1\leq i'_2\leq \cdots\leq i'_{k}\leq n_1, \ \ \sum_{p=1}^{k} i'_p=k(k+n)-\ell,$$ which obviously gives a bijective correspondence $$\{P\in \mathcal{D}_{2,n}(n+2k) : \text{comaj}(P)=\ell\}\cong \rho^{k}_{n_1}(k(k+n)-\ell).$$
{#section-9}
Our next corollary answers the following question. What is the set of all powers $q$ which appear with nonzero coefficients in the polynomial $\mathcal{V}^{1\rightarrow m}_{s,n}(q)$? We first determine the degree of $\mathcal{V}^{1\rightarrow m}_{s,n}(q)$ as we did in Lemma \[lpart\] for $m=2$. By we can assume that $s=n+2k$ for some $k\in {\mathbb Z}_+$. The path $$\label{4}P=\underbrace{1010\cdots 1010}_{2k}\underbrace{111\cdots 11}_{n}$$ with exactly $k$ up–down steps at the beginning and $n$ up–steps at the end is contained in $\mathcal{D}_n(s)$. Note that all peaks of $P$ are at height $1$ and $(0,1)\notin A(m,n)$, which implies that $P$ is admissible. Moreover, $$\label{2}\text{comaj}(P)=\sum_{i=1}^k(s-2i+1)=k(n+k).$$ If $P'\in \mathcal{D}_n(s)$ is another Dyck path with peaks at length $s_1,\dots,s_p$, we have $s_i\geq 2i-1,\ i=1,\dots,p$ and $p\leq k$. Thus $$\label{3}\text{comaj}(P)=\sum_{i=1}^k(s-2i+1)\geq \sum_{i=1}^p(s-s_i)+\sum_{i=p+1}^k(s-2i+1)\geq\text{comaj}(P')$$ with equality if and only if $P=P'$.
\[conint\]We have that $\mathcal{V}^{1\rightarrow m}_{s,n}(q)=0$ if $s-n\notin 2{\mathbb Z}_+$ and otherwise it is a monic polynomial of degree $(s-n)(s+n)/4$. Moreover, the powers of $q$ which appear in the polynomial $\mathcal{V}^{1\rightarrow m}_{s,n}(q)$ form an interval of consecutive integers iff one of the following conditions hold (i) $n\neq 0$, (ii) $m=2$, (iii) $s=n$.
The first part of the corollary follows from the discussion preceding the corollary, see and . We start by proving the backward direction of the second part. Let $n\neq 0$ and $\tilde{P}\neq P$ an arbitrary path in $\mathcal{D}_{m,n}(s)$, where $P$ is as in . We shall show that $q^{\text{comaj}(\tilde{P})+1}$ appears with a nonzero coefficient in the polynomial $\mathcal{V}^{1\rightarrow m}_{s,n}(q)$. Since $\tilde{P}\neq P$, we must have two consecutive up-steps in $\tilde{P}$ followed by at least one down-step. Thus $\tilde{P}$ is of the form $$\tilde{P}=\cdots \underbrace{11\cdots 11}_{\ell}\underbrace{00\cdots 00}_{r}1\cdots,\ \ell\geq 2,\ \ r\geq 1,$$ and we can assume without loss of generality that the path preceding the $\ell$–up steps is contained in $\mathcal{D}_0$, i.e. it ends at the $x$–axis. *Case 1:* Suppose that $r<\ell$. We set $$\bar{P}=\cdots \underbrace{11\cdots 11}_{\ell-1}\underbrace{00\cdots 00}_{r}11\cdots$$ and obtain $\bar{P}\in \mathcal{D}_n^N(s)$ and $\text{comaj}(\bar{P})=\text{comaj}(\tilde{P})+1.$
(0,0) grid (19,7); (0,0) – (3,3) – (6,0) – (10,4)–(13,1)–(17,5)–(19,7); (9,3) – (12,0) – (13,1); at (8,2) [$\ell$]{}; at (12,2) [$r$]{};
To complete the proof in this case, we have to show that $\bar{P}$ is admissible. Again, there is nothing to show if $n<m$; so suppose that $n\geq m$. We denote by $z$ the peak of $\tilde{P}$ after the $\ell$ up–steps and set for simplicity $b:=\text{ht}_{z}(\tilde{P})$. It is clear that $\bar{P}$ has a peak at height $b-1$. If $b\leq m$, there is nothing to show. Otherwise we must show that $\text{ht}_{\tilde z}(\bar{P})>\tilde a$ for all points $\tilde z\in \text{supp}(\bar{P})$ with $\ell_{\tilde z}(\bar{P})\geq \ell_{z}(\tilde{P})-1$, where $\tilde a$ is the unique non–negative integer such that $(\tilde a,b-1)\in A(m,n)$. For a point $\tilde z\in \text{supp}(\bar{P})$ we have that $\tilde z\in \text{supp}(\tilde{P})$ or $(\ell_{\tilde z}(\bar{P})+1,\text{ht}_{\tilde z}(\bar{P})+1)\in \text{supp}(\tilde{P})$. In either case, $$\text{ht}_{\tilde z}(\bar{P})> a-1\geq \tilde a,$$ where first inequality follows by our assumption $\tilde{P}\in \mathcal{D}_{m,n}(s)$ and the second inequality follows from .
*Case 2:* In this case we can suppose that each peak in $\tilde{P}$ returns to the $x$–axis. Since $\tilde{P}\neq P$, we have that $\tilde{P}$ contains at least two down–steps. In particular $m\geq 3$, since $(0,m)\in A(m,n)$ or when $n<m$ the height of $\tilde{P}$ is at most $m-1$. Hence $$\tilde{P}=\cdots 1\underbrace{00\cdots 00}_{r}\underbrace{1010\cdots 1010}_{t}\underbrace{11\cdots 11}_{n},\ \ r\geq 2,\ \ t\in 2{\mathbb Z}_+.$$ If $t\neq 0$, we can set $$\tilde{P}=\cdots 1\underbrace{00\cdots 00}_{r-1}100\underbrace{10\cdots 1010}_{t-2}\underbrace{11\cdots 11}_{n},$$ which is admissible and has the desired property. If $t=0$, we set $$\bar{P}=\cdots 1\underbrace{11\cdots 11}_{N_{0,1}}\underbrace{00\cdots 00}_{r-1}\underbrace{11\cdots 11}_{n-2N_{0,1}}0\underbrace{11\cdots 11}_{N_{0,1}},$$ where $N_{0,1}=n_1+\delta_{n_0,m-1}$. We get $$\text{comaj}(\bar{P})=\text{comaj}(\tilde{P})-(n+r)+(n+r-N_{0,1})+(N_{0,1}+1)=\text{comaj}(\tilde{P})+1.$$
(0,0) grid (19,7); (0,0) – (3,3) – (6,0) – (9,3)–(12,0)–(19,7); (9,3) – (11,5) – (13,3)–(16,6)–(17,5); at (8,2) [$\ell$]{}; at (12,2) [$r$]{}; at (10,4) [$a$]{}; at (15,5) [$b$]{};
Again, we must prove that $\bar{P}$ is admissible. Recall that $r<m$, since $N=m-1$ or $(0,m)\in A(n,m)$. There are two new peaks $z_1$ and $z_2$ in the path $\bar{P}$, where $$z_1=(s-n-r+N_{0,1},r+N_{0,1}),\ \ z_2=(n-N_{0,1}-1,n-N_{0,1}+1).$$ We know that $$(n-N_{0,1}-1,n-N_{0,1}+1)\in A(n,m)$$ and hence the peak $z_2$ does not violate property . If $r+N_{0,1}<m$, there is nothing to check for $z_1$ and otherwise we have $(r+N_{0,1}+\delta_{n_0,m-1}-m,r+N_{0,1})\in A(m,n)$ and the claim follows with $$\text{ht}_{z'}(\bar{P})\geq N_{0,1}+1>r+N_{0,1}+\delta_{n_0,m-1}-m,\ \ \text{ $\forall z'$ with }\ell_{z'}(\bar{P})\geq s-n-r+N_{0,1}.$$ Hence in the case when $n\neq 0$ we have that $\mathcal{V}^{1\rightarrow m}_{s,n}(q)$ forms an interval of consecutive integers. If $n=0$ and $m=2$, the statement is clear since $\mathcal{D}_{2,0}(s)=\{P\}$. Similarly there is nothing to prove if $n=0=s$ and hence the backward direction is established.
Now we show that $\mathcal{V}^{1\rightarrow m}_{s,n}(q)$ does not form an interval of consecutive integers provided that $n=0$, $m\geq 3$ and $s\neq n$. For this we consider a path of minimal degree in $\mathcal{D}_{m,n}(s)$ given by $$Q=\underbrace{11\cdots 11}_{m-1}\underbrace{00\cdots 00}_{m-1}\cdots\underbrace{11\cdots 11}_{m-1}\underbrace{00\cdots 00}_{m-1}\underbrace{11\cdots 11}_{p}\underbrace{00\cdots 00}_{p},$$ where $(s-n)/2=\ell (m-1)+p$, $0\leq p<m-1$. Then it is easy to show that there is no path in $\mathcal{D}_{m,n}(s)$ whose comajor is given by $\text{comaj}(Q)+1$. The representation theoretic meaning of the above corollary is quite surprising.
The maximal degree in which a level $m$–Demazure module appears in a level $m$–Demazure flag of a local Weyl module is independent of $m$ and depends only on its highest weight. Moreover, given non–negative integers $s,p_1,p_2\in {\mathbb Z}_+$ with $p_1<p_2$ and $m,n\in {\mathbb N}$, we have an equivalence $$[W_{\text{loc}}(s):\tau_p^{*}D(m,n)]\neq 0\ \ \forall p\in [p_1,p_2]\cap {\mathbb Z}\iff [W_{\text{loc}}(s):\tau_{p}^{*}D(m,n)]\neq 0\ \ \forall p\in \{p_1,p_2\}.$$
{#section33}
Under the restriction $m\geq s$ the module $D(m,s)$ is irreducible as an $\mathfrak{sl}_2$–representation (see for example [@CSSW14]). Hence the level $m$–Demazure flag becomes the usual Jordan–Hölder series. In this situation graded multiplicities in fusion products were studied in [@Ke04] and are described by co–charge Kostka–Foulkes polynomials. Let $\langle\cdot,\cdot\rangle$ be the Hall inner product on $\Lambda$, the ring of symmetric functions. The modified Hall–Littlewood polynomials $Q'_{\mu}=Q'_{\mu}(x;q)$ are defined as the basis of $\Lambda[q]$ dual to the ordinary Hall–Littlewood polynomials $P_{\lambda}=P_{\lambda}(x;q)$ with respect to $\langle\cdot,\cdot\rangle$: $$\langle P_{\lambda},Q'_{\mu}\rangle=\delta_{\lambda,\mu}.$$ The Kostka–Foulkes polynomials $K_{\lambda,\mu}(q)$ are then defined as $$Q'_{\mu}(x;q)=\sum_{\lambda} K_{\lambda,\mu}(q)s_{\lambda}(x),$$ where the $s_{\lambda}(x)$ are the Schur functions. The closely related co–charge Kostka polynomials $\tilde{K}_{\lambda,\mu}(q)$ are given by $$\tilde{K}_{\lambda,\mu}(q)=q^{n(\mu)} K_{\lambda,\mu}(1/q),$$ where $n(\mu):=\sum_{i\geq 1} (i-1)\mu_i$. The following result goes back to [@Ke04] and an alternative proof can be given by using $q$–Bernstein operators (see [@Jin91]) to show that the co–charge Kostka polynomial satisfies a recursion similar to [@CV13 Theorem 5]. We thank Ole Warnaar for helping us with this observation.
\[ked\]Let $\xi$ a partition and $m\in {\mathbb N}$ sucht that $m\geq |\xi|$. Set $\mu=(\frac{|\xi|+n}{2}, \frac{|\xi|-n}{2} )$. We have $$\mathcal{V}_{n}^{\xi\rightarrow m}(q)=\tilde{K}_{\mu,\xi}(q).$$
In the rest of this section we discuss an interesting corollary. Let $s=n+2k$ for some $k\in{\mathbb Z}_+$. We claim that $$\label{6}\mathcal{D}_{m,n}(s)=\mathcal{D}_n(s),\ \text{ if $m\geq s$}.$$ If $k=0$, the claim is immediate. Otherwise, note that any path $P\in \mathcal{D}_n(s)$ has height at most $n+k$ and $n+k\leq m-1$. Hence $\mathcal{D}_n(s)=\mathcal{D}^{m-1}_n(s)= \mathcal{D}_{m,n}(s)$, since $N=m-1$ and $A(m,n)=\emptyset$. This shows . There is a bijection between $\mathcal{D}_n(s)$ and the set of standard Young tableaux $SYT(\lambda)$ of shape $\lambda=(n+k,k)$ as follows: $$\label{bijtoSYT}P=a_1\cdots a_{s}\mapsto T(P),$$ where we put $i$ into the first row if $a_i=1$ and otherwise into the second row. See Figure \[fig2\] for an example.
(0,0) grid (7,2); (0,0) – (2,2) – (3,1) – (4,2)–(6,0)–(7,1);
$\mapsto \young(1247,356)$
As mentioned in Remark \[rem1\] there is also a comajor statistics on standard Young tableaux which is often useful. Given a tableaux $T$ of shape $\lambda$, a descent of $T$ is a value $i$, $1\leq i\leq s$, for which $i+1$ occurs in one of the rows below $i$. Define $\text{comaj}(T)=\sum_i s-i$, where the sum runs over the descents of $T$. Stanley proved [@S99 pg. 363], $$\label{stan}s_{\lambda}(1,q,q^2,\dots)=\frac{1}{(q;q)_{s}}\sum_{T\in SYT(\lambda)}q^{\text{comaj}(T)}.$$
\[princp\]Let $\lambda=(n+k,k)$, $\xi=(1^{n+2k})$ and $\mu=(\frac{s+n}{2}, \frac{s-n}{2} )$. Then $$\tilde{K}_{\mu,\xi}(q)=(q;q)_{n+2k} \ s_{\lambda}(1,q,q^2,\dots).$$ Let $V_{\mathfrak{sl}_2}(n)$ the $(n+1)$–dimensional irreducible $\mathfrak{sl}_2$–representation. Theorem \[mainthm1\] together with imply $$\label{7}\sum_{p\geq 0}[W_{\text{loc}}(n+2k): \tau_p^* V_{\mathfrak{sl}_2}(n)]\hspace{0,03cm} q^p=\sum_{P\in \mathcal{D}_n(s)}q^{\text{comaj}(P)}.$$ Since preserves the comajor statistics we get that is equal to $\sum_{T\in SYT(\lambda)}q^{\text{comaj}(T)}$ and the claim follows with and Lemma \[ked\].
{#section-10}
In his last letter to G.H. Hardy, S. Ramanujan listed 17 functions which he called mock theta functions [@Ram88]. There are some number theoretic interpretations of some of these functions in the literature. For example, one of the third order mock theta functions has been interpreted as the generating function for partitions into odd parts without gaps [@Fi88]. Agarwal gave an interpretation of some fifth order mock theta functions in terms of $n$–color partitions [@Ag04] and later in terms of lattice paths [@Ag05]. Thanks to [@BCSV15 Theorem 1.6] we have a relationship between certain specializations of the series $\mathcal{A}_n^{1\rightarrow 3}(x,q)$, $n\in\{0,1,2\}$ and the fifth order mock theta functions $\phi_0(q),\phi_1(q),\psi_0(q),\psi_0(q)$, where $$\begin{aligned}
\phi_0(q) = \sum_{n=0}^\infty q^{n^2} (-q;q^2)_n,\ \
\phi_1(q) = \sum_{n=0}^\infty q^{(n+1)^2} (-q;q^2)_n,\\
\psi_0(q) = \sum_{n=0}^\infty q^{\frac{(n+1)(n+2)}{2}} (-q;q)_n,\ \
\psi_1(q) = \sum_{n=0}^\infty q^{\frac{n(n+1)}{2}}(-q;q)_n.\end{aligned}$$ As an immediate consequence (combining [@BCSV15 Theorem 1.6] and Theorem \[mainthm1\]) we get a new interpretation of their coefficients. We emphasize that our interpretation is different from [@Ag05].
\[mockcoef\]For $n\in{\mathbb Z}_+$ let $res_2(n)$ the remainder of $n$ modulo 2. We have
1. $\psi_1(q)=\sum_{n=0}^{\infty}|\{P\in \mathcal{D}^2_{1}: \text{comaj}(P)=n\}|q^n$,
2. $\psi_0(q)=\sum_{n=0}^{\infty}|\{P\in \mathcal{D}^2_{1}: \text{comaj}(P)+\lceil\frac{\ell(P)}{2}\rceil=n\}|q^n,$
3. $\phi_0(q)=\sum_{n=0}^{\infty}|\{P\in \mathcal{D}^2_{2\text{res}_2(n)}: \text{comaj}(P)=\lfloor n \rfloor\}|q^n$
4. $\phi_1(q)=\sum_{n=0}^{\infty}|\{P\in \mathcal{D}^2_{2(1-\text{res}_2(n))}: \text{comaj}(P)+\lceil\frac{\ell(P)}{2}\rceil+(1-\text{res}_2(n))=\lfloor n \rfloor\}|q^n.$
{#section-11}
The connection to mock theta functions [@BCSV15 Theorem 1.6] after specializing the generating series $\mathcal{A}_n^{1\rightarrow 3}(x,q)$, $n\in\{0,1,2\}$ is quite surprising. We emphasize that the relationship could only be made for one reason: the series $\mathcal{A}_n^{1\rightarrow 3}(x,q)$, $n\in\{0,1,2\}$ can be expressed in terms of $q$–binomial coefficients [@BCSV15 Sections 1.4 and 1.6]. In this section we want to generalize the situation and give a formulae for $\mathcal{A}_n^{1\rightarrow m}(x,q)$, $n\in\{0,\dots,m-1\}$ in terms of $q$–binomials, but the connection to number theory will be discussed elsewhere. The key ingredient in the proof of the next corollary is Theorem \[mainthm1\] and a result of Krattenthaler and Mohanty [@KM93 Theorem 1].
\[KrM\]Let $m\in {\mathbb N}$ and $n\in {\mathbb Z}_+$ such that $n<m$. We have that $\mathcal{V}_{n+2k,n}^{1\rightarrow m}(q)$ equals $$\label{b}\sum_{\ell=0}^k\sum_{s=-\ell}^{\ell} q^{\ell^2+s(sm+n+1)} \text{det} \begin{pmatrix}
{\genfrac[]{0pt}0{k-s(m-1)}{\ell+s}}_q &
{\genfrac[]{0pt}0{k+s(m-1)-1}{\ell-s}}_q \\
q^{-2s(n+1)}{\genfrac[]{0pt}0{n+1+k-s(m-1)}{\ell+s}}_q & {\genfrac[]{0pt}0{n+k+s(m-1)}{\ell-s}}_q \\
\end{pmatrix}.$$
Since $n<m$ we have $\mathcal{D}_{m,n}(n+2k)=\mathcal{D}^{m-1}_{n}(n+2k)$ and hence Theorem \[mainthm1\] implies $$\mathcal{V}_{n+2k,n}^{1\rightarrow m}(q)=\sum_{P\in \mathcal{D}^{m-1}_{n}(n+2k)} q^{comaj(P)}.$$ It means that we have to count the set of Dyck paths from $(0,0)$ to $(n+2k,n)$ of height $< m$ by its comajor index. A special case of Theorem 1 of [@KM93] counts the set of Dyck paths from $(0,0)$ to $(n+2k,n)$ of height $<m$ with respect to a different statistics, namely by its major and descent index. The descent index $\text{des}(P)$ counts the number of peaks of $P$ and hence $$\label{desmaj}\text{maj}(P)=\text{des}(P)(n+2k)-\text{comaj}(P).$$ In view of we only have to substitute the formula in Theorem \[mainthm1\] by $q\mapsto 1/q$, $x\mapsto x^{n+2k}$ and finally $x=q$.
The advantage of Corollary \[KrM\] is that it gives some hints to compute graded multiplicities for higher rank Lie algebras. As remarked in the introduction level $m'$–Demazure modules for $\mathfrak{sl}_n$ also admit a level $m$–Demazure flag if $m\geq m'$. So the polynomials can be defined in a similar manner; also the corresponding generating series. We do not know yet whether the size of the matrix in Corollary \[KrM\] has something to do with the rank of the underlying Lie algebra or the fact that our model is based on certain two dimensional lattice paths. One possibility for higher rank Lie algebras would be to study multidimensional lattice paths (depending on the rank of the Lie algebra), see for example [@HM79]. We will address this in a future publication.
Proof of Theorem \[mainthm1\] {#section5}
=============================
In this section we prove our main theorem.
{#section-12}
We first show that our combinatorial model gives the correct numerical multiplicity and a proof of that can be given purely combinatorial, which we will demonstrate in this subsection. The generating function $\mathcal{G}_{m,n}^N(x):=\sum_{s\geq 0}|\mathcal{D}_{m,n}(s)| x^{(s-n)/2}$ is again a rational function which can be expressed in terms of Chebyshev polynomials. The next lemma can be viewed as a generalization of Lemma \[Ges\].
\[numerical\] We have $$\mathcal{G}_{m,n}^N(x)=\frac{p_{m-n_0-1}(x)}{p_m(x)^{n_1+1}}=\mathcal{A}_n^m(x,1).$$ In particular, $$\mathcal{V}_{s,n}^{1\rightarrow m}(1)=|\mathcal{D}_{m,n}(s)|.$$
If $n<m$, the statement follows from Lemma \[Ges\]; so assume that $n\geq m$. Let $(i_1,\dots,i_{n-m+1})$ be the integers from Definition \[maindef\] and set $i_0=0$, $i_{n-m+2}=n$ and $\tilde{\mathcal{G}}_{m,n}^N(x)=x^n\mathcal{G}_{m,n}^N(x)$. Let $P$ be a path contained in $\mathcal{D}_{m,n}$. We want to factor $P$ as $$\label{factorpath}P=P_11P_21\cdots P_{n-m+1}1 P_{n-m+2},$$ where $P_{\ell}$ is a path of height at most $m-i_{\ell-1}+(\ell-2)+(\delta_{\ell,1}-1)$ that ends at height $i_{\ell}-i_{\ell-1}+(\delta_{\ell,1}-1)$. This would give, together with Lemma \[Ges\], $$\tilde{\mathcal{G}}_{m,n}^N(x)=\prod^{n-m+2}_{\ell=1}x^{i_{\ell}-i_{\ell-1}}\frac{p_{m-i_{\ell}+(\ell-2)}(x)}{p_{m-i_{\ell-1}+(\ell-2)+\delta_{\ell,1}}(x)}=\frac{p_{m-n_0-1}(x)}{p_m(x)^{n_1+1}}.$$ The rest of the claim follows from [@BCSV15 Corollary 1.3]. Now we make precise how we factor the path as in . The path $P$ starts at the origin and we follow the path until a point $z\in \text{supp}(P)$ with $z=(\ell_z(P),i_1)$. Since $P$ ends at height $n$ and $i_1<n$, the existence of $z$ is clear. Among all points we choose $z$ such that $\ell_z(P)$ is maximal, i.e. we take the right most point of height $i_1$. We define this part of the path by $P_1$ and note that the height of $P_1$ is at most $m-1$, since $(i_1,m)\in A(m,n)$. By the maximality of $z$, we must have that $P_1$ is immediately followed by an up–step. Hence $P=P_11Q$, where $Q$ is the remaining part of $P$. We continue the process with $Q$, but note that this path is shifted up by $(i_1+1)$–units. Again we follow the path $Q$ until a point $z$ on the path with $z=(\ell_z(P),i_2)$ and $\ell_z(P)$ is maximal with this property (the right most point with height $i_2$.) This part of the path we define as $P_2$ and again the height of $P_2$ is at most $m-(i_1+1)$, since $(m+1,i_2)\in A(m,n)$ (remember the shift by $(i_1+1)$). Continuing this way gives the desired factorization.
(0,0) grid (21,5); (0,0) – (3,3) – (5,1) – (7,3)–(10,0); (10,0) – (11,1); (11,1)– (14,4)–(17,1)–(18,2); (18,2) – (19,3); (19,3) – (21,5);
{#section-13}
It is clear that each Dyck path in $\mathcal{D}_n(s)$ is uniquely determined by the coordinates of all its peaks. Let $P$ a path with $d$ peaks whose coordinates are $$B_1=(x_1,y_1),\dots,B_d=(x_{d},y_{d}),\ \ x_1<x_2<\cdots< x_d.$$ We define inductively a subset $\eta(P)\subseteq\{1,\dots,d\}$ as follows. Set $i_1=1$ and define $$i_{t+1}:=\min\{i_t<u\leq d : y_u\geq y_{i_t}\},\ \ 1\leq t\leq \ell-1,$$ where $\ell$ denotes the maximal non–negative integer such that $i_{\ell}$ exists. Define $$\label{incsubsetofpeaks}
\eta(P):=\{i_1,\dots,i_{\ell}\}.$$ The meaning of the sequence $\eta(P)$ in the language of Dyck path is as follows: given a peak $B_{i_{t}}$ let $B_{i_{t+1}}$ the unique peak determined by the property that $B_{i_{t+1}}$ is the left most peak to the right of $B_{i_{t}}$ which is weakly above $B_{i_{t}}$. This sequence will be needed in the proof of Proposition \[eigenschaft1\].
Let $P\in \mathcal{D}_7(20)$ the unique path with peaks $$B_1=(3,3), B_2=(6,2), B_3=(10,4), B_4=(15,5), B_5=(17,5).$$ By definition we have $i_1=1$. Since $B_{2}$ is not weakly above $B_{i_1}$ but $B_{3}$ is weakly above $B_{i_1}$, we get $i_2=3$. Continuing in this way we get $\eta(P)=\{1,3,4,5\}$.
(0,0) grid (20,7); (0,0) – (3,3) – (5,1) – (6,2)–(7,1)–(10,4)–(12,2)– (15,5) – (16,4)– (17,5)–(18,4)–(19,5)–(20,6);
Similarly, a Dyck path $Q\in \mathcal{D}_n(s)$ is also uniquely determined by the coordinates of all its valleys. So assume that $Q$ has $d$ valleys whose coordinates are $$C_1=(x_1,y_1),\dots, C_d=(x_{d},y_{d}),\ \ x_1<x_2<\cdots< x_d.$$ This time we set $j_1=d$ and define inductively a subset $\zeta(Q)\subseteq\{1,\dots, d\}$ as follows. Set $$j_{t+1}:=\max\{1\leq u< j_t : y_u\leq y_{j_t}\},\ \ v-1\leq t\leq d,$$ where $v$ denotes the maximal non–negative integer such that $j_{v}$ exists. Define $$\label{incsubsetofvalleys}
\zeta(P):=\{j_1,\dots,j_{v}\}.$$ The meaning of the sequence $\zeta(P)$ in the language of Dyck paths is as follows: given a valley $C_{i_{t}}$ let $C_{i_{t+1}}$ the unique valley determined by the property that $C_{i_{t+1}}$ is the right most valley to the left of $C_{i_{t}}$ which is weakly below $C_{i_{t}}$.
{#section-14}
Let $_{j}\mathcal{D}_{m,n}(s)\subset \mathcal{D}_{m,n}(s)$ be the subset of Dyck paths which start with $j\in {\mathbb N}$ number of up–steps, i.e. $$_{j}\mathcal{D}_{m,n}(s)=\{a_1\cdots a_{s}\in\mathcal{D}_{m,n}(s): a_1=a_2=\cdots=a_{j}=1\}.$$ Furthermore, set $$_{j}\mathcal{E}_{s,n}^{1\rightarrow m}(q)=\sum_{P\in \hspace{0,02cm} _{j}\mathcal{D}_{m,n}(s)} q^{\text{comaj(P)}}.$$
\[eigenschaft1\]Let $n,s\geq m$ and $s-n\in 2{\mathbb Z}_+$. Then, there exists a bijection $$\Psi : \mbox{$_m\mathcal{D}_{m,n}(s)$} \rightarrow \mathcal{D}_{m,n-m}(s-m)$$ such that $\text{comaj}(P)=\text{comaj}(\Psi(P))+(s-n)/2$. Set $s=n+2k$ for some $k\in {\mathbb Z}_+$. Let $P\in \mbox{$_m\mathcal{D}_{m,n}(s)$}$ and assume that $P$ has $d$ peaks whose coordinates are $$B_1=(x_1,y_1),\dots,B_d=(x_{d},y_{d}),\ \ x_1<x_2<\cdots< x_d.$$ Furthermore, let $\eta(P)=\{i_1,i_2,\dots,i_{\ell}\}$ as in . Let $\Psi(P)$ the unique path in $\mathcal{D}_{m,n-m}(s-m)$ with $d$ peaks whose coordinates are given by $$B^{\Psi}_i=(x_i+\ell_i-m,2y_{i_t}-y_i+\ell_i-m),\ \ 1\leq t\leq \ell,\ \ i_t\leq i < i_{t+1},$$ where $\ell_i$ denotes the number of down–steps right after peak $B_i$ and $i_{\ell+1}:=d+1$.
This map has a pictorial interpretation, namely we reflect the path $P$ piecewise. To be more precise, we draw $\ell$ horizontal lines connecting the points $(x_{i_t},y_{i_t})$ with $(x_{i_t}+s_{i_t},y_{i_t})$, $1\leq t\leq \ell$, where $s_{i_t}=\min\{s\in {\mathbb N}: (x_{i_t}+s,y_{i_t})\in \text{supp}(P)\}$. Then we reflect the path locally at these lines and ignore the first $m$ up-steps (see Figure \[fig6\] for an example).
(0,0) grid (12,9); (0,0) – (5,5) – (6,4) – (8,6)–(10,4)–(12,6); (5,5) – (7,5); (8,6) – (12,6); (5,5) – (6,6) – (7,5)–(10,8)–(12,6); at (7,5) [$\ell_1$]{}; at (10.5,5.5) [$\ell_2$]{}; at (1.5,7.5) [$P$]{};
(0,0) grid (7,5); (0,0) – (1,1) – (2,0) – (5,3)–(7,1); at (3,3.5) [$\bold{\Psi}(P)$]{};
Since $\ell_1+\cdots+\ell_d=k$, we immediately get $\text{comaj}(\Psi(P))=\text{comaj}(P)-k$. In order to show that $\Psi$ is well–defined we note that $(x_i+\ell_i,y_i-\ell_i)\in \text{supp}(P).$ Fix $t\in\{1,\dots,\ell\}$ and assume that $y_{i_t}=m+pn_1+r$ for some $0\leq r<n_1$ and $0\leq p\leq n_0$. The case when $y_{i_t}=m+p(n_1-1)+(n_0+1)+r$ for some $0\leq r<n_1-1$ and $n_0<p< m$ is similar and will be omitted. Since $P\in \mbox{$_m\mathcal{D}_{m,n}(s)$}$ (hence $P$ is admissible) we must have $$y_i-\ell_i>y_{i_t}-m+p,\ \ \text{$\forall i\in\{ i_t,\dots,i_{t+1}-1\}$}.$$ Equivalently, $$y_{i_t}-p=m+p(n_1-1)+r>2y_{i_t}-y_i+\ell_i-m.$$ Hence, if we write $2y_{i_t}-y_i+\ell_i-m=m+p'(n_1-1)+r'$ for some $0\leq r'<n_1-1$ and $0\leq p'\leq n_0$ we must have $p\geq p'$ which yields $$\label{admgenug}y_{i_t}-m>2y_{i_t}-y_i+\ell_i-2m+p\geq 2y_{i_t}-y_i+\ell_i-2m+p'.$$ We claim that already implies that $\Psi(P)$ is admissible for all pairs in $A(m,n-m)$ and hence $\Psi(P)\in \mathcal{D}_{m,n-m}(s)$. To see this, let $z\in \text{supp}(\Psi(P))$ such that $\ell_z(\Psi(P))\geq x_i+\ell_i-m$. We know that the path $\Psi(P)$ never goes below the line $y_{i_t}-m$ once it passes peak $B^{\Psi}_i$ and hence $$\label{admgenug2}\text{ht}_z(\Psi(P))\geq y_{i_t}-m.$$ Equation together with proves that $\Psi$ is well–defined. It remains to verify that $\Psi$ is bijective. We define $\Psi^{-1}: \mathcal{D}_{m,n-m}(s-m) \rightarrow \mbox{$_m\mathcal{D}_{m,n}(s)$}$ as follows. Let $Q\in \mathcal{D}_{m,n-m}(s-m)$ a path with $d$ valleys whose coordinates are $$C_1=(x_1,y_1),\dots, C_d=(x_{d},y_{d}),\ \ x_1<x_2<\cdots< x_d.$$ Furthermore, let $\zeta(Q)$ be the sequence from . We draw $v$ horizontal lines connecting the points $(x_{i_t},y_{i_t})$ with $(x_{i_t}-s_{i_t},y_{i_t})$, $1\leq t\leq v$, where $s_{i_t}=\min\{s\in {\mathbb N}: (x_{i_t}-s,y_{i_t}) \in\text{supp}(Q)\}$ and reflect the path locally at these lines. Putting $m$ number of $1's$ in front of the reflected path yields $\Psi^{-1}(Q)$. The fact that $\Psi$ and $\Psi^{-1}$ are inverse to each other is clear and the fact that it is well–defined is similarly proven and we omit the details.
\[eigenschaft2\] We have $$_{m}\mathcal{E}_{s,n}^{1\rightarrow m}(q)=q^{(s-n)/2}\ _{1}\mathcal{E}_{s-m,n-m}^{1\rightarrow m}(q).$$ If $(i)\ (s-n)\notin 2{\mathbb Z}_+$ or $(ii)\ n<m$ or $(iii)\ m>s$, we have $_m\mathcal{D}_{m,n}(s)=\emptyset=\mathcal{D}_{m,n-m}(s-m)$ and the corollary is immediate. Otherwise the corollary follows from Proposition \[eigenschaft1\].
{#section-15}
The following lemma will be needed.
\[neededlemma\]Let $m,n,j,s\in {\mathbb N}$ with $j<m$ and $s\geq j$. Then we have, $$_{j}\mathcal{E}_{s,n}^{1\rightarrow m}(q)=\sum^{j}_{r=0} q^{(1-\delta_{r,0})(s-j)}\ \mbox{$_{(j+1-r-\delta_{r,s-j})}\mathcal{E}_{s-2r,n}^{1\rightarrow m}$}(q).$$ Each path $P\in \mbox{$_j\mathcal{D}_{m,n}(s)$}$ is of the form $$P=\underbrace{1\cdots 1}_{j}\underbrace{0\cdots 0}_{r}Q,$$ for some $r\in \{0,1\dots,j\}$ and a $01$–word $Q$, which is empty or starts with an up–step. Moreover, the path $Q$ is empty if and only if $r=s-j$. Define $$\tau(P):=\underbrace{1\cdots 1}_{j-r}Q\in\mbox{$_{(j+1-r-\delta_{r,s-j})}\mathcal{D}_{m,n}(s-2r)$}.$$ With other words we get a map $$\tau:\mbox{$_j\mathcal{D}_{m,n}(s)$}\rightarrow \bigcup_{r=0}^{j} \mbox{$_{(j+1-r-\delta_{r,s-j})}\mathcal{D}_{m,n}(s-2r)$},$$ which is obviously well–defined and injective.
(0,0) grid (11,6); (0,0) – (3,3) – (4,2)– (5,3); (5,3)–(6,4)–(8,2)–(11,5); at (1.5,5.5) [$P$]{};
(0,0) grid (9,6); (0,0) – (3,3)–(4,4)–(6,2)–(9,5); at (2.8,5.3) [$\tau(P)$]{};
We show that $\tau$ is also surjective. If $r=s-j$ we have $|_{(2j-s)}\mathcal{D}_{m,n}(2j-s)|=\delta_{n,2j-s}$ and the preimage of the unique element in $_{(2j-s)}\mathcal{D}_{m,2js}(2j-s)$ is the path which consists of $j$ up–steps and $(s-j)$ down–steps. If $r\neq s-j$ choose $P'=\underbrace{1\cdots 1}_{j+1-r}Q'\in \mbox{$_{(j+1-r)}\mathcal{D}_{m,n}(s-2r)$}$ arbitrary, where $Q'$ is a $01$–word and assume without loss of generality that $s\geq 2r$ (c.f. ). Obviously the only candidate for the preimage is the path $$\tau^{-1}(P'):=\underbrace{1\cdots 1}_{j}\underbrace{0\cdots 0}_{r}Q,\ \ Q=1Q'.$$ We only have to show that $\tau^{-1}(P')$ is admissible. Note that the peaks of $P'$ and $\tau^{-1}(P)$ are identical if $r=0$ and else there is one additional peak at height $j$. Since $j<m$ there is no condition on the additional peak and $\tau^{-1}(P)$ is admissibe. Finally, it is clear that $\text{comaj}(P)=\text{comaj}(\tau(P))+(1-\delta_{r,0})(s-j)$, which finishes the proof.
{#551}
The proof of the main theorem requires one further representation theoretical result. For the rest of this section we consider only hook partitions $$\xi=(m,1^s),\ m\in{\mathbb N},\ s\in {\mathbb Z}_+$$ and recall from Remark \[rem23\] that $$\label{fuseqloc}
V\left((1^s)\right)\cong W_{\text{loc}}(s).$$ The proof of the next theorem is postponed to Section \[section6\].
\[sesnot\] Let $s,m\in {\mathbb N}$. Then, we have a short exact sequence of $\mathfrak{sl}_2[t]$–modules $$0\rightarrow ker(\varphi)\rightarrow V\left((m,1^s)\right)\rightarrow V\left((m+1,1^{s-1})\right)\longrightarrow 0,$$ where $ker(\varphi)$ can be filtered by $$\bigoplus_{r=\max\{m+2-s,1\}}^{m} \tau_s^* V\left((r,1^{s-2-m+r})\right) \oplus \tau_s^* M$$ and $M={\operatorname{ev}}_0^* V_{\mathfrak{sl}_2}(m-s)$, if $s\leq m$ and $M=0$ otherwise.
Another family of short exact sequences among fusion products has been constructed in [@CV13 Theorem 5(i)] which coincides with the one given in Theorem \[sesnot\] only if $m=1$. Generically they are different as all kernels given in [@CV13 Theorem 5(i)] are proper fusion products.
Proof of Theorem \[mainthm1\] {#proof-of-theoremmainthm1}
-----------------------------
From Proposition \[existflag\] we know that $V\left((m',1^s)\right)$ has a level $m$–Demazure flag if and only if $m\geq m'$. It is immediate from that the main theorem is a consequence of the following stronger statement, which gives a combinatorial model for graded multiplicities in fusion products associated to hook partitions.
\[hook\]Let $m,m'\in{\mathbb N}$, $s\in{\mathbb Z}_+$ such that $m\geq m'$. Let $\xi=(m',1^s)$. We have $$\mathcal{V}^{\xi\rightarrow m}_{n}(q)=\mbox{$_{m^{'}}\mathcal{E}_{s+m',n}^{1\rightarrow m}$}(q)=\sum_{P\in \hspace{0,03cm} _{m'}\mathcal{D}_{m,n}(s+m')} q^{\text{comaj(P)}}.$$
We prove the claim by induction on $s$. If $s=0$ we get $$\mathcal{V}^{\xi\rightarrow m}_{n}(q)=\delta_{m',n}=\mbox{$_{m^{'}}\mathcal{E}_{\xi,n}^{1\rightarrow m}$}(q)$$ and the induction begins. Assume that $s>0$. If $m>m'+1$ we can use Theorem \[sesnot\] and our induction hypothesis to get $$\begin{aligned}
\mathcal{V}^{\xi\rightarrow m}_{n}(q)&=\notag \sum_{r=\max\{m'+2-s,1\}}^{m'+1}q^{(1-\delta_{r,m'+1})s}\hspace{0,03cm} \mathcal{V}^{(r,1^{s-2-m'+r})\rightarrow m}_{n}(q)+q^s\delta_{m,s+n}&\\&
=\label{needed2}\sum^{m'+1}_{r=1} q^{(1-\delta_{r,m'+1})s}\mbox{$_r\mathcal{E}_{s-2-m'+2r,n}^{1\rightarrow m}$}(q)+\mbox{$_{(m'-s)}\mathcal{E}_{m'-s,n}^{1\rightarrow m}$}(q).\end{aligned}$$ Now Lemma \[neededlemma\] implies that is equal to and the theorem is established in this case. So let $m=m'$ and note that Lemma \[ausc\] implies $$\mathcal{V}^{\xi\rightarrow m}_{n}(q)=q^{(s+m-n)/2}\hspace{0,03cm}\mathcal{V}^{(1,1^{s-1})\rightarrow m}_{n-m}(q).$$ We apply the induction hypothesis to $\mathcal{V}^{(1,1^{s-1})\rightarrow m}_{n-m}(q)$ and conclude that $$\mathcal{V}^{\xi\rightarrow m}_{n}(q)=q^{(s+m-n)/2}\mbox{$_1\mathcal{E}(s,m,n-m)$}.$$ The theorem is now immediate with Corollary \[eigenschaft2\].
Proof of Theorem \[sesnot\] {#section6}
===========================
{#section-16}
We first recall the more traditional definition of fusion products from [@FL99]. Let $n\in {\mathbb N}$ and denote by ${\operatorname{ev}}_z^{*}V_{\mathfrak{sl}_2}(n)$, $z\in{\mathbb C}$ the $\mathfrak{sl}_2[t]$–representation whose action is given by $$(w\otimes f(t)).v=f(z) w.v,\ \ v\in V(n),\ f(t)\in {\mathbb C}[t],\ x\in \mathfrak{sl}_2.$$ It is standard to show for pairwise distinct complex numbers $(z_1,\dots,z_{\ell})$ and a partition $\xi=(\xi_1\geq \cdots\geq \xi_{\ell})$ that ${\operatorname{ev}}_{z_1}^{*}V_{\mathfrak{sl}_2}(\xi_1)\otimes \cdots \otimes {\operatorname{ev}}_{z_{\ell}}^{*}V_{\mathfrak{sl}_2}(\xi_{\ell})$ is a cyclic representation for $\mathbf{U}(\mathfrak{sl}_2[t])$ with cyclic generator $v_{\xi}=v_{\xi_1}\otimes \cdots \otimes v_{\xi_{\ell}}$. Set $$\mathbf{U}(\mathfrak{sl}_2[t])_i=\text{span}\{(x_{i_1}\otimes f_1(t))\cdots(x_{i_{j}}\otimes f_{j}(t)): j\in {\mathbb Z}_+,\ \ \sum_{p=1}^{j}\text{deg }f_p(t)\leq i\}.$$ The associated graded space with respect to the filtration $$0\subset \mathbf{U}(\mathfrak{sl}_2[t])_1v_{\xi}\subset \mathbf{U}(\mathfrak{sl}_2[t])_2v_{\xi}\subset \cdots$$ is isomorphic to $V(\xi)$ (see [@CV13 Theorem 5]). This construction justifies why fusion products are sometimes called graded tensor products. Especially, we emphasize that the construction is independent of the chosen parameters and $$\label{dimfor1}\text{dim } V(\xi)=\prod_{i=1}^{\ell} (\xi_i+1).$$ From the defining relations (see Section\[section23\]) we immediately get the existence of a surjective homomorphism $$\label{fusionsurjection}
\varphi: V\left((m,1^s)\right)\rightarrow V\left((m+1,1^{s-1})\right).$$ The proof of the next lemma is a straightforward calculation using .
\[relaa\]Let $\xi=(m,1^s)$. Then we have $(y\otimes t^{s+1})v_{\xi}=0$ and $$\text{dim } ker \varphi=\sum _{r=\max\{m+2-s,1\}}^m \text{dim } V((r,1^{s-2-m+r}))+\max\{0,(m-s+1)\}.$$
{#section-17}
We need some more notation before we can prove Theorem \[sesnot\]. Let $$_{\ell}\mathbf S(r,p)=\Big\{(b_k)_{k\ge 0}:b_k\in\mathbb Z_+,\ b_k=0\ \forall k<\ell \ \sum_{k\ge 0} b_k=r,\ \ \sum_{k\ge 0} kb_k=p\Big\},\ \ \mathbf S(r,p):=\mbox{$_0$}\mathbf S(r,p)$$ and define $$\begin{aligned}
_{\ell}\mathbf{y}(r,p)=&\sum_{{\bold b}\in _{\ell}\mathbf S(r,p)}\ \prod_{i=0}^{p}(y\otimes t^i)^{(b_i)},\ \ \ \mathbf{y}(r,p):=\mbox{$_0$}\mathbf{y}(r,p).\end{aligned}$$ We collect several immediate consequences.
\[simt1\] Let $V$ any representation of $\mathfrak{sl}_2[t]$ and $v\in V$, such that for all $a\in {\mathbb Z}_+$ we have $(h\otimes t^{a+1})v=(x\otimes t^a)v=0$. Then for all $r,p\in{\mathbb N}$ we have
1. $(x\otimes t)^{p}(y\otimes 1)^{r+p}v=\mathbf{y}(r,p)v$
2. $\mathbf{y}(r,p)-\mbox{$_{1}\mathbf{y}(r,p)$} \in \sum_{r'<r}\mathbf{U}(\mathfrak{sl}_2[t]) \mathbf{y}(r',p).$
3. Assume in addition $(y\otimes t^{p+1})v=0$. Then $(h\otimes t^k)\mathbf{y}(r,p)v$ is contained in the $\mathbf{U}(\mathfrak{sl}_2[t])$–span generated by $$\label{moduloset}
\{\mathbf{y}(r,p+k)v,\ \mathbf{y}(r-1,p')v: \ p'\geq p+1\}.$$
The first claim of the lemma is a consequence of [@G78 Lemma 7.1]. To prove part (2) we do induction on $r$. If $r=1$, we have $$\mathbf{y}(1,p)-\mbox{$_{1}\mathbf{y}(1,p)$}=(y\otimes t^p)-(y\otimes t^p)=0.$$ If $r>0$, we use $$\mathbf{y}(r,p)- \mbox{$_{1}\mathbf{y}(r,p)$} \in \sum_{r'<r}\mathbf{U}(\mathfrak{sl}_2[t])\mbox{$_{1}\mathbf{y}(r',p)$}$$ and apply our induction hypothesis to each $\mbox{$_{1}\mathbf{y}(r',p)$}$, which gives the desired result. The last claim is immediate if $k>p$; so assume that $k\leq p$. The claim follows from the following calculation which we do modulo the $\mathbf{U}(\mathfrak{sl}_2[t])$–span of : $$\begin{aligned}
(h\otimes t^k)\mathbf{y}(r,p)v&=\sum_{{\bold b}\in \mathbf S(r,p+k)}(b_k+\cdots+b_p)\prod^{p}_{i=1}(y\otimes t^i)^{(b_i)}v&\\&=\sum_{\ell=0}^{k-1}\ \sum_{{\bold b}\in \mathbf S(r,p+k)}(-b_{\ell})\prod^{p}_{i=1}(y\otimes t^i)^{(b_i)}v,\ \ \ \ \ \ (\text{subtract $r\mathbf{y}(r,p+k)v$})&\\&=\sum_{\ell=0}^{k-1}-(y\otimes t^{\ell})\sum_{{\bold b}\in \mathbf S(r,p+k), b_{\ell}>0}(y\otimes 1)^{(b_0)}\cdots (y\otimes t^{\ell})^{(b_{\ell}-1)}\cdots (y\otimes t^{p})^{(b_{p})}v&\\&=\sum_{\ell=0}^{k-1}-(y\otimes t^{\ell})\ \mathbf{y}(r-1,p+k-\ell)v&\\&=0.\end{aligned}$$
{#section-18}
We are now able to determine generators of the kernel of the map .
\[bestkern\]Let $\xi=(m,1^s)$. The kernel of the homomorhism is generated by the set $$\label{genkernel}S=\{\mbox{$\mathbf{y}(r,s)$}v_{\xi}: 1\leq r\leq \min\{s,m\}\}.$$ Clearly Lemma \[simt1\](1) implies that $ker(\varphi)$ is generated by $$S':=\{\mbox{$\mathbf{y}(r,p)$}v_{\xi}: \ r,p\in {\mathbb N},\ r+p\geq 1+rk+\max\{0,(s-k)\}\ \ \text{for some $k\in {\mathbb N}$}\}.$$ The claim follows if the prove that any element in $S'$ is contained in the $\mathbf{U}(\mathfrak{sl}_2[t])$–span of the elements in $S$. Let $\mathbf{y}(r,p)v_{\xi}\in S'$. If $r+p>1+rk+\max\{0,(s-k)\}$ we have $$r+p\geq 1+rk+\max\{0,(s-k)\}+1\geq 1+rk+\max\{0,(s+1-k)\}$$ and hence $\mbox{$\mathbf{y}(r,p)$}v_{\xi}=0$. So we can assume from now on that $r+p=1+rk+\max\{0,(s-k)\}$. If $k>1$ and $r\geq 2$ we get $$r+p=1+r(k-1)+\max\{0,(s-k)\}+r\geq 1+r(k-1)+\max\{0,(s+1-(k-1))\},$$ and again $\mbox{$\mathbf{y}(r,p)$}v_{\xi}=0$. Otherwise if $k>1$ and $r=1$ we obtain $$r+p=1+k+\max\{0,(s-k)\}\geq s+1\Rightarrow p\geq s.$$ This implies in the case when $p=s$ that $\mbox{$\mathbf{y}(1,p)$}v_{\xi}\in S$ and in the case when $p>s$ that $\mbox{$\mathbf{y}(1,p)$}v_{\xi}=0$ since $$1+p\geq s+2\geq 1+k+\max\{0,(s+1-k)\}.$$ So if there exists an element $\mathbf{y}(r,p)v_{\xi}\in S'$ which is not contained in the $\mathbf{U}(\mathfrak{sl}_2[t])$–span of the elements in $S$ we must have $k=1$ and hence $p=s$. Moreover, if $r>m$, then using Lemma \[simt1\](1) we get $$r+s\geq m+s+1\Rightarrow (x\otimes t)^{(s)}(x\otimes t)^{(r+s)}v_{\xi}=\mbox{$\mathbf{y}(r,s)$}v_{\xi}=0.$$ So it remains to show that each element in $$\{\mbox{$\mathbf{y}(r,s)$}v_{\xi}: 1\leq r\leq m\}$$ is in the $\mathbf{U}(\mathfrak{sl}_2[t])$–span of the elements in $S$. If $s>m$ there is nothing to show; so assume from now on $s\leq m$. If $r\leq s$, we already have that $\mathbf{y}(r,s)v_{\xi}\in S$, so we can assume in addition that $r>s$. Let $(b_i)_{i\geq 0}$ a tuple of non–negative integers such that $$b_0+\cdots+b_s=r,\ \ b_1+2b_2+\cdots+sb_s=s.$$ We get $$\begin{aligned}
s&=b_1+\cdots+b_s+(b_2+2b_3+\cdots+(s-1)b_s)&\\&
=(r-b_0)+(b_2+2b_3+\cdots+(s-1)b_s)&\\&
\geq s+(r-s-b_0) \end{aligned}$$ and hence $b_0\geq r-s>0$. This proves $$\mbox{$\mathbf{y}(r,s)$}v_{\xi}\in \sum_{b_0=r-s}^r \mathbf{U}(\mathfrak{sl}_2[t])\mbox{$_1\mathbf{y}(r-b_0,s)$}v_{\xi}.$$ The claim now follows from Lemma\[simt1\] (2).
{#section-19}
An easy induction on $r$ shows that if $v$ is a weight vector of weight $N$ of a $\mathbf{U}(\mathfrak{sl}_2[t])$–representation $V$, then $$[(x\otimes 1),(y\otimes 1)^{(r+1)}]v=(N-r)(y\otimes 1)^{(r)}v.$$ This implies together with Lemma \[moduloset\](1) that $(x\otimes 1)\mathbf{y}(r,s)v_{\xi}=A \mathbf{y}(r-1,s)v_{\xi}$ for some non–zero scalar $A\in{\mathbb C}^*$. Hence using Proposition \[bestkern\] we get a filtration $$0=V_0\subset V_1\subset V_2\subset \cdots \subset V_{\min\{s,m\}}=ker(\varphi),\ \ V_r:=\mathbf{U}(\mathfrak{sl}_2[t])\mathbf{y}(r,s)v_{\xi}.$$ Moreover, we know from Lemma \[relaa\] that $(y\otimes t^{s+1})v_{\xi}=0$ and hence we obtain Lemma \[moduloset\](3) that $(h\otimes t^k)\mathbf{y}(r,s)v_{\xi}$ is in the $\mathbf{U}(\mathfrak{sl}_2[t])$–span of the elements $\mathbf{y}(r,s+k)v_\xi,\mathbf{y}(r-1,s+k)v_\xi,\dots,\mathbf{y}(r-1,s+1)v_\xi$. Clearly the defining relations of $V(\xi)$ imply that all these elements are zero. Hence $$(x\otimes 1)\mathbf{y}(r,s)v_{\xi}\in V_{r-1},\ \ (h\otimes t^k)\mathbf{y}(r,s)v_{\xi}\in V_{r-1},\ \ k\in {\mathbb N}.$$ Since $(x\otimes {\mathbb C}[t])\oplus (h\otimes t{\mathbb C}[t])$ is generated by $\{(x\otimes 1), (h\otimes t^k): k\in {\mathbb N}\}$ as a Lie algebra we obtain a well–defined map $$\label{brauchnur}W_{\text{loc}}(s+m-2r)\rightarrow V_r/V_{r-1},\ 1\leq r \leq \min\{s,m\}.$$ The following proposition along with Lemmn\[relaa\] finishes the proof of Theorem \[sesnot\].
For $1\leq r< \min\{s,m\}$ there exists a surjective homomorphism $$V\left((m+1-r,1^{s-1-r})\right)\rightarrow V_r/V_{r-1}.$$ Moreover, we have a surjective map $W_{\text{loc}}(s-m)\rightarrow V_{m}/V_{m-1}$ if $s\geq m$ and otherwise $$V_{s}/V_{s-1}\cong
{\operatorname{ev}}_0^* V_{\mathfrak{sl}_2}(m-s).$$ Assume that $1\leq r < \min\{s,m\}$. From and the defining relations of fusion products it will be enough to show that the following holds in $V_r/V_{r-1}$: $$(x\otimes t)^{(p')}(y\otimes 1)^{(r'+p')}\mbox{$_1\mathbf{y}(r,s)$}v_{\xi}=0,\ \forall r',p',k'\in {\mathbb N}: r'+p'\geq 1+r'k'+\max\{0,(s-r-k')\}.$$ Recall the traditional construction of fusion products from Section \[551\] and the independence of the choice of the parameters. Without loss of generality we will identify $v_{\xi}$ with $v_m\otimes v_1\otimes \cdots \otimes v_1$ and choose the first evaluation parameter to be zero, i.e. $z_1=0$. We get $$\begin{aligned}
(x\otimes t)^{(p')}(y\otimes 1)^{(r'+p')}\mbox{$_1\mathbf{y}(r,s)$}v_{\xi}&=(x\otimes t)^{(p')}(y\otimes 1)^{(r'+p')}\left(v_{m}\otimes \mbox{$_1\mathbf{y}(r,s)$}(v_1\otimes \cdots \otimes v_1)\right)&\\&=v_{m}\otimes \left((x\otimes t)^{(p')}(y\otimes 1)^{(r'+p')}\mbox{$_1\mathbf{y}(r,s)$}(v_1\otimes \cdots \otimes v_1)\right).\end{aligned}$$ Since $r'+p'+r>s$ and the lowest weight of $V_{\mathfrak{sl}_2}(1)^{\otimes s}$ is $(-s)$ we must have $$\left((x\otimes t)^{(p')}(y\otimes 1)^{(r'+p')}\mbox{$_1\mathbf{y}(r,s)$}(v_1\otimes \cdots \otimes v_1)\right)=0.$$ It remains to consider the case $r=\min\{s,m\}$. If $m\geq s$ we obtain with Lemma \[moduloset\](2) that the following holds in $V_s/V_{s-1}$: $$(y\otimes t^k)\mathbf{y}(s,s)v_{\xi}=(y\otimes t^k)\mbox{$_1\mathbf{y}(s,s)$}v_{\xi}=(y\otimes t^k)(y\otimes t)^{(s)}v_{\xi}=0,\ \forall k\in {\mathbb N}.$$ Hence $V_s/V_{s-1}\cong {\operatorname{ev}}_0^* V_{\mathfrak{sl}_2}(m-s)$. If $s\geq m$, the claim follows already from the discussion preceding the proposition, see .
[^1]:
[^2]:
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In these notes we study the duality between sigma-models and Toda QFT’s. We claim that $\mathfrak{gl}(n|n)$ affine Toda field theory behaves in the strong coupling limit as $\eta-$deformed $\mathbb{CP}(n-1)$ sigma-model plus a free field.'
author:
- 'A. V. Litvinov\'
bibliography:
- 'MyBib.bib'
title: |
**Integrable $\mathfrak{gl}(n|n)$ Toda field theory\
and its sigma-model dual**
---
Duality is an interesting concept of modern theoretical physics [@Polchinski:2014mva]. Especially the weak/strong coupling one. It replaces strongly interacting regime of one theory with perturbative regime of the other, and vise versa. In these notes we study the duality between integrable Toda QFT’s based on supergroups and $\eta$-deformed sigma-models along the lines suggested in [@Fateev:2018yos; @Litvinov:2018bou]. We consider $\mathfrak{gl}(n|n)$ affine Toda field theory [@Litvinov:2016mgi; @2015arXiv151208779B] $$\label{action}
\mathcal{A}_{n}=\int\left(\frac{1}{8\pi}\left(\partial_{a}\Phi\cdot\partial_{a}\Phi\right)+\frac{1}{8\pi}\left(\partial_{a}\phi\cdot\partial_{a}\phi\right)+
\Lambda\sum_{k=1}^{n}\left(e^{b\Phi_{k}-i\beta\phi_{k}}+e^{i\beta\phi_{k}-b\Phi_{k+1}}\right)\right)\,d^{2}\xi,\quad\beta=\sqrt{1+b^{2}},$$ where $\phi=(\phi_{1},\dots,\phi_{n})$, $\Phi=(\Phi_{1},\dots,\Phi_{n})$ are two $n-$component bosonic fields and $\Phi_{n+1}\overset{\text{def}}{=}\Phi_{1}$. The parameter $b$ plays the role of the coupling constant of the theory. Each exponent in has fermionic scaling dimensions $\Delta=\bar{\Delta}=\frac{1}{2}$ and hence $\Lambda$ has a dimension of mass. The theory contains the $U(1)$ part $\chi=\beta\sum\Phi_{k}-ib\sum\phi_{k}$, which is not interacting. In order to make it manifest, we perform the following transformation of the fields (here $Z=\sum_{k=1}^{n}(\Phi_{k}+i\phi_{k})$) $$\label{Z-Z}
Z\rightarrow (\beta+b)Z,\quad\bar{Z}\rightarrow(\beta-b)\bar{Z}.$$ After this transformation the action will have the form $$\label{action-modified}
\mathcal{A}_{n}'=\int\left(\frac{1}{8\pi}\left(\partial_{a}\Phi,\partial_{a}\Phi\right)+\frac{1}{8\pi}\left(\partial_{a}\phi,\partial_{a}\phi\right)+
\Lambda\sum_{k=1}^{n}\left(e^{b(h_{k},\Phi)-i\beta(\mathfrak{h}_{k},\phi)}+e^{i\beta(\mathfrak{h}_{k},\phi)-b(h_{k+1},\Phi)}\right)\right)\,d^{2}z,$$ where $h_{k}=e_{k}-\frac{1}{n}(e_{1}+\dots+e_{n})$ and $\mathfrak{h}_{k}=e_{k}-\frac{\beta-1}{n\beta}(e_{1}+\dots+e_{n})$ and $e_{k}$ is the orthonormal basis in $\mathbb{R}^{n}$. From it is clear that the “center of mass” field $\sum_{k}\Phi_{k}$ decouples.
In order to define QFT correctly, one has to specify the domain of the coupling constant $b$. One distinguishes between the weak $b\rightarrow0$ and the strong-coupling $b\rightarrow\infty$ regimes. At $b\rightarrow0$ one has to add an additional UV regularization term $$\frac{\pi\Lambda^{2}a^{2}}{b^{2}}\int\sum_{k=1}^{n}e^{b(\Phi_{k}-\Phi_{k+1})},$$ where $a$ is the UV cut-off. The model defined by the action possesses perturbative integrability in the mass parameter $\Lambda$. In the limit $\Lambda\rightarrow0$ one can construct an infinite tower of local integrals of motion of all spins $(\mathbf{I}_{s}^{(0)},\bar{\mathbf{I}}_{s}^{(0)})$, $s=1,2,\dots$ $$\mathbf{I}^{(0)}_{s}=\oint\left(\sum_{k=1}^{n}\left(b^{s-2}(\partial\Phi_{k})^{s}+(i\beta)^{s-2}(\partial\phi_{k})^{s}\right)+\dots\right)d\xi,$$ and similar expression for $\bar{\mathbf{I}}^{(0)}_{s}$. One can write explicit formulae for lowest integrals, but they are quite cumbersome. We note that there exists an analog of quantum Miura transformation for this theory, which was found in [@Prochazka:2018tlo]. Using it, one can in principle find a convenient expression for IM’s. For us it is important that the system of IM’s $\mathbf{I}^{(0)}_{s}$ can be defined as a commutant of screening charges $$\label{Scr-charges}
\mathcal{S}_{k}=\oint e^{b\Phi_{k}-i\beta\phi_{k}}d\xi,\quad \tilde{\mathcal{S}}_{k}=\oint e^{i\beta\phi_{k}-b\Phi_{k+1}}d\xi,\quad k=1,\dots,n,$$ which correspond to the exponential terms in the action. We stress that $\mathbf{I}^{(0)}_{s}$ are defined in the theory of free massless bosonic fields. They are images of total IM’s at the limit $\Lambda\rightarrow0$. Going beyond the leading order in $\Lambda$ is a quite complicated task, but it is believed that the full theory is integrable.
Another test of integrability comes from the perturbative analysis at $b\rightarrow0$ while keeping $\Lambda$ fixed. It is convenient to use Coleman-Mandelstam duality and replace bosonic fields $\phi_{k}$ by Dirac fermions $\psi_{k}$. This replacement leads to the theory of $n$ Dirac fermions and $n$ bosonic fields $\Phi_{k}$ with hidden $\mathfrak{gl}(n)$ symmetry. It can be shown perturbatively in the parameter $b$ that the scattering of fundamental particles in this theory shares the properties of factorized scattering [@Fateev-cpn]. In particular, one finds an absence of particle production – a remarkable property of integrable QFT’s. The exact $S-$matrix for this theory has been conjectured recently in [@Fateev-cpn].
In the strong coupling regime $b\rightarrow\infty$ the action is useless. However, one can use the following observation. Each pair of fermionic screening charges defines the conformal algebra of the coset CFT $SU(2)_{\kappa}/U(1)$ with $\kappa=-2-b^{2}$. It is well known that this algebra commutes with a third screening charge $$\mathcal{W}=\oint\left(b\partial\Phi_{k}-i\beta\partial\phi_{k}\right)e^{b^{-1}(\Phi_{k}-\Phi_{k+1})}d\xi,$$ known also as Wakimoto screening charge. It means that the theory $$\label{action-dual}
\tilde{\mathcal{A}}_{n}=\int\left(\frac{1}{8\pi}\left(\partial_{a}\Phi\cdot\partial_{a}\Phi\right)+\frac{1}{8\pi}\left(\partial_{a}\phi\cdot\partial_{a}\phi\right)+
\tilde{\Lambda}\sum_{k=1}^{n}\left(b\partial\Phi_{k}-i\beta\partial\phi_{k}\right)\left(b\bar{\partial}\Phi_{k}-i\beta\bar{\partial}\phi_{k}\right)e^{b^{-1}(\Phi_{k}-\Phi_{k+1})}\right)\,d^{2}\xi,$$ shares the same integrable structure in the limit $\tilde{\Lambda}\rightarrow0$ as the original one in the limit $\Lambda\rightarrow0$. Actually, the theory makes sense only in the region $b\rightarrow\infty$. In order to regularize its UV behavior one has to add counterterms. Total renormalized action will have a form of a sigma-model $$\label{action-SM}
\tilde{\mathcal{A}}_{n}=\frac{1}{8\pi}\int G_{\mu\nu}(\boldsymbol{X}|\tilde{\Lambda},b^{2})\partial_{a}X^{\mu}\partial_{a}X^{\nu}d^{2}\xi,\quad\text{where}\quad
\boldsymbol{X}=(\Phi_{1},\dots,\Phi_{n},\phi_{1},\dots,\phi_{n}).$$ The precise form of the metric $G_{\mu\nu}(\boldsymbol{X}|\tilde{\Lambda},b^{2})$ might be very complicated. Moreover, it depends on a chosen regularization scheme. The only reasonable thing is to find it in the semiclassical approximation $b\rightarrow\infty$. We define $$(x_{1},\dots,x_{n},y_{1},\dots,y_{n})=b^{-1}(\Phi_{1},\dots,\Phi_{n},\phi_{1},\dots,\phi_{n}), \quad
\tilde{\Lambda}b^{2}=e^{t},$$ and make the following anzatz for the classical metric (here $z_{k}=x_{k}-iy_{k}$) $$\label{metric}
ds^{2}=\sum_{k=1}^{n}dz_{k}d\bar{z}_{k}+\mu(t)\left(\sum_{k=1}^{n}dz_{k}\right)^{2}+
2\sum_{k=1}^{n}dz_{k}^{2}\sum_{l=1}^{n}\lambda_{l}(t)e^{(\boldsymbol{\alpha}_{k}+\dots+\boldsymbol{\alpha}_{k+l-1},\boldsymbol{x})},$$ where $\boldsymbol{\alpha}_{k}=\boldsymbol{\alpha}_{n+k}$ are the roots of $\mathfrak{sl}(n)$: $(\boldsymbol{\alpha}_{k},\boldsymbol{x})=x_{k}-x_{k+1}$, $x_{k+n}=x_{k}$. This anzatz is consistent with the symmetry of the problem, but of course it is a matter of a guesswork to find it. The metric should flow with the RG time $-\infty<t<t_{0}$ according to the Ricci flow equation [@Friedan:1980jm] $$\label{Ricci-flow}
R_{\mu\nu}+\nabla_{\mu}V_{\nu}+\nabla_{\nu}V_{\mu}=-\dot{G}_{\mu\nu},\qquad V_{\mu}=\nabla_{\mu}\Phi,$$ where the vector field $V_{\mu}$ describes the effect of renormalization of the fields. We assume, for simplicity, that this vector field is a gradient of a scalar function: $V_{\mu}=\nabla_{\mu}\Phi$. Then there exists a solution to with the desired UV asymptotic $$\mu(t)=-\frac{2}{n}\frac{e^{nt}}{e^{nt}-1},\quad
\lambda_{k}(t)=\frac{e^{kt}}{e^{nt}-1},\qquad\Phi=\sum_{k=1}^{n}x_{k}.$$
We note that we can perform the boost transformation $$\sum_{k=1}^{n}z_{k}\rightarrow\nu\sum_{k=1}^{n}z_{k},\quad
\sum_{k=1}^{n}\bar{z}_{k}\rightarrow\nu^{-1}\sum_{k=1}^{n}\bar{z}_{k},$$ and decouple the “center of mass” field $\sum_{k}z_{k}$ by sending $\nu\rightarrow0$. This is equivalent to the transformation , which changes the behavior of the fields in the semiclassical limit $b\rightarrow\infty$. The decoupled metric has the form $$\label{CPN-metric}
ds^{2}=|d\boldsymbol{z}|^{2}+\frac{2}{e^{nt}-1}\sum_{k=1}^{n}(h_{k},d\boldsymbol{z})^{2}f_{k}(x),\qquad
f_{k}(x)=\sum_{l=1}^{n}e^{lt}e^{(\boldsymbol{\alpha}_{k}+\dots+\boldsymbol{\alpha}_{k+l-1},\boldsymbol{x})}.$$ This metric is in $2(n-1)$ dimensional space with complex coordinates $z_{k}=x_{k}-iy_{k}$, $k=1,\dots, n-1$, $\boldsymbol{\alpha}_{k}=\boldsymbol{\alpha}_{n+k}$ are the roots of $\mathfrak{sl}(n)$ and $\boldsymbol{h}_{k}$ are the weights of the first fundamental representation. For example, for $n=2$ the metric reads $$ds^{2}=dzd\bar{z}+\frac{1}{e^{2t}-1}\left((e^{t+x}+e^{2t})dz^{2}+(e^{t-x}+e^{2t})d\bar{z}^{2}\right).$$ It can be transformed to the $T-$dual of the sausage metric [@Fateev:1992tk] $$ds^{2}=\frac{\kappa d\zeta^{2}}{4(1-\zeta^{2})(1-\kappa^{2}\zeta^{2})}+\frac{4(1-\kappa^{2}\zeta^{2})d\varphi^{2}}{\kappa(1-\zeta^{2})},\qquad
\kappa=-\tanh t,$$ by simple change of variables $$\cosh x=\frac{1+\zeta^{2}}{1-\zeta^{2}}\qquad
y=\frac{\varphi}{4}-i\log\left(\frac{(1-\zeta)(1+\kappa\zeta)}{(1+\zeta)(1-\kappa\zeta)}\right).$$ It is well known that the sausage model coincides with the $\eta-$deformed $\mathbb{CP}(1)$ sigma-model. We conjecture that our general metric coincides with the metric of the $\eta-$deformed $\mathbb{CP}(n-1)=SU(n)/SU(n-1)U(1)$ sigma-model after $T-$dualities in all isometry directions.
The action of general $\eta$-deformed (we take $\eta=i\kappa$) $G/H$ coset sigma model has the form [@Delduc:2013fga] $$\label{Coset-action-deformed}
\mathcal{S}=\frac{\kappa}{2}\int\textrm{Tr}\left(
\left(\mathbf{g}\partial_{+}\mathbf{g}^{-1}\right)^{(\textrm{c})}\,\frac{1}{1-i\kappa\mathcal{R}_{\mathbf{g}}\circ\mathrm{P}_{\textrm{c}}}\,
\left(\mathbf{g}\partial_{-}\mathbf{g}^{-1}\right)^{(\textrm{c})}\right) d^{2}x,$$ where $\boldsymbol{g}\in G$, $\mathcal{R}_{\mathbf{g}}=\textrm{Ad}\, \mathbf{g}\circ\mathcal{R}\circ\textrm{Ad}\,\mathbf{g}^{-1}$ and $\mathrm{P}_{\textrm{c}}$ is the projection on the coset space. In our case we take $G=SU(n)$ and $H=U(n-1)=U(1)\otimes SU(n-1)$. The operator $\mathcal{R}$ acts in the Lie algebra $\mathfrak{g}=\mathfrak{c}\oplus_{\alpha>0}\mathfrak{g}_{\alpha}\oplus_{\alpha>0}\mathfrak{g}_{-\alpha}$ as $$\mathcal{R}\Bigl|_{\mathfrak{c}}=0,\qquad
\mathcal{R}\Bigl|_{\mathfrak{g}_{\alpha}}=i,\qquad
\mathcal{R}\Bigl|_{\mathfrak{g}_{-\alpha}}=-i,$$ while $\mathcal{R}_{\mathbf{g}}=A^{-1}\mathcal{R}A$, $A_{ab}=\langle t_{a}\,\mathbf{g}\,t_{b}\,\mathbf{g}^{-1}\rangle$. Consider for example $G=SU(3)$. We take the basis in the Lie algebra $\mathfrak{sl}(3)$ as $$t_{1}=\begin{pmatrix}
0 & 1 & 0 \\
1 & 0 & 0 \\
0 & 0 & 0 \\
\end{pmatrix},\quad
t_{2}=\begin{pmatrix}
0 & -i & 0 \\
i & 0 & 0 \\
0 & 0 & 0 \\
\end{pmatrix},\quad
t_{3}=\begin{pmatrix}
1 & 0 & 0 \\
0 & -1 & 0 \\
0 & 0 & 0 \\
\end{pmatrix},\quad
t_{4}=\frac{1}{\sqrt{3}}\begin{pmatrix}
1 & 0 & 0 \\
0 & 1 & 0 \\
0 & 0 & -2 \\
\end{pmatrix}$$ and $$t_{5}=\begin{pmatrix}
0 & 0 & 1 \\
0 & 0 & 0 \\
1 & 0 & 0 \\
\end{pmatrix},\quad
t_{6}=\begin{pmatrix}
0 & 0 & -i \\
0 & 0 & 0 \\
i & 0 & 0 \\
\end{pmatrix},\quad
t_{7}=\begin{pmatrix}
0 & 0 & 0 \\
0 & 0 & 1 \\
0 & 1 & 0 \\
\end{pmatrix},\quad
t_{8}=\begin{pmatrix}
0 & 0 & 0 \\
0 & 0 & -i \\
0 & i & 0 \\
\end{pmatrix}.$$ The generators $\{t_{1},t_{2},t_{3},t_{4}\}$ form a subalgebra $\mathfrak{h}=\mathfrak{su}(2)\oplus\mathfrak{u}(1)$. It is convenient to take a coset representative as $$\mathbf{g}^{-1}=e^{\frac{i(\psi-\phi)}{4}t_{3}-\frac{i(3\phi+\psi)}{4\sqrt{3}}t_{4}}e^{\frac{i\theta}{2}t_{5}}e^{i(\chi+\frac{\pi}{2})t_{7}}.$$ For this choice $\phi$ and $\psi$ obviously correspond to isometry directions. Then, computing the metric and the $B-$field and performing $T-$dualities in $\phi$ and $\psi$ isometry directions, we find that the $B-$field vanishes while the metric takes the form $$\begin{gathered}
\label{CP2-sausage-metric}
ds^{2}=\kappa\Bigl(d\chi^{2}+\frac{\sin^2\chi}{4}d\theta^2-2i\sin\theta d\theta\left(d\psi\left(\sin^2\chi-\csc^2\theta\right)+d\phi\cot\theta\csc\theta\right)-\\-
4i\tan\chi d\chi\left(d\phi\cot^2\chi-d\psi\cos\theta\right)+
\frac{4d\phi^2 \left((1-\kappa^2)\csc^2\theta\csc^2\chi+\kappa^2\right)}{\kappa^2}+\frac{8d\psi d\phi(\kappa^2-1)\cot\theta\csc\theta\csc^2\chi}{\kappa^2}+\\+
\frac{d\psi^2\left(4\left(1-\kappa^2\right)\csc^2\theta\csc^2\chi+2\kappa^2\sin^2\theta\cos2\chi-2\sec^2\chi\left(\kappa^2 \cos2\theta+\kappa^2-2\right)+3\kappa^2 (\cos2\theta +3)\right)}{\kappa^2}\Bigr)\end{gathered}$$ This metric satisfies Ricci flow equation with $\Phi=-\log(\sin2\chi\sin\chi\sin\theta)-8i\phi$ and $\kappa=-\tanh6 t$. It is straightforward to find how the metric is related to the metric (here $n=3$, $q=e^{8 t}$) $$\begin{gathered}
e^{-(\boldsymbol{\alpha}_{1},\boldsymbol{x})}=q^{-1}\sin^{2}\frac{\theta}{2}\tan^{2}\chi\left(q^{\frac{3}{2}}+(1-q^{\frac{3}{2}})\cos^{2}\frac{\theta}{2}\sin^{2}\chi\right),\\
e^{(\boldsymbol{\alpha}_{2},\boldsymbol{x})}=q^{-\frac{1}{2}}\cos^{2}\frac{\theta}{2}\tan^{2}\chi
\left(1-(1-q^{\frac{3}{2}})\sin^{2}\frac{\theta}{2}\sin^{2}\chi\right),\\
(\boldsymbol{\alpha}_{1},y)=8\varphi-2i\log\left(\tan\frac{\theta}{2}\left(q^{\frac{3}{2}}+(1-q^{\frac{3}{2}})\cos^{2}\frac{\theta}{2}\sin^{2}\chi\right)\right),\\
(\boldsymbol{\alpha}_{1}+\boldsymbol{\alpha}_{2},y)=4\varphi-12\psi-2i\log\left(\sin\frac{\theta}{2}\tan\chi\left(1-(1-q^{\frac{3}{2}})\sin^{2}\frac{\theta}{2}\sin^{2}\chi\right)\right).
\end{gathered}$$ We have checked similar statement for $\mathbb{CP}(3)$ sigma-model, thus confirming our general conjecture. The formulae in that case are too long to be presented here.
Our results verify the conjecture that $\mathfrak{gl}(n|n)$ Toda QFT approaches in the strong coupling limit the $T-$dual of the $\eta$-deformed $\mathbb{CP}(n-1)$ sigma-model. At first sight this statement looks contradictory. It is well known that $\mathbb{CP}(n-1)$ sigma-model fails to have integrability at the quantum level. Taking a look at the action we note that in the limit $b\rightarrow\infty$ another “center of mass” field $\sum_{k}\phi_{k}$ decouples, thus exactly at $b=\infty$ the theory coincides with the deformed $\mathbb{CP}(n-1)$ sigma-model plus a free field. Taking into account loop corrections the interaction between the two parts will appear and the theory presumably will restore the integrability. The precise mechanism of this restoration is an interesting question to be addressed. Of course our conjecture has to be further checked by other methods in order to be fully justified. A lot such checks will be done in [@Fateev-cpn].
Acknowledgments {#acknowledgments .unnumbered}
===============
Some of the results of this paper were independently derived by Vladimir Fateev. The author thanks him for sharing his insights and for his kind scientific advisement. This work is supported by the RFBR under grant 18-02-01131.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present a formulation of a nonequilibrium Green’s function method for thermal current in nanojunction atomic systems with nonlinear interactions. This first-principle approach is applied to the calculation of the thermal conductance in carbon nanotube junctions. It is shown that nonlinearity already becomes important at low temperatures. Nonlinear interactions greatly suppress phonon transmission at room temperature. The peak of thermal conductance is found to be around 400K, in good agreement with experiments. High-order phonon scattering processes are important for diffusive heat transport.'
author:
- 'Jian-Sheng Wang'
- Jian Wang
- Nan Zeng
bibliography:
- 'basename of .bib'
date: 1 May 2006
title: 'Nonequilibrium Green’s function approach to mesoscopic thermal transport'
---
Thermal transport in materials has been studied for a long time, beginning with Joseph Fourier’s heat conduction law. However, a microscopic theory is possible only after the advent of quantum mechanics [@peierls]. The earlier treatments are mostly for bulk systems [@rmp-review]. In recent years, motivated by the shrinkage of sizes of electronic devices, researchers have paid more attention to the heat transport phenomena in meso- and nano-scales [@cahill]. Under such circumstances, some of the concepts have to be modified. For example, it has been found that Fourier’s law is no longer valid for many one-dimensional systems [@lepri]. What to replace it is both interesting theoretically and relevant experimentally.
A number of approaches have been used to study heat transport, such as classical molecular dynamics (MD) and the Boltzmann-Peierls equation [@peierls]. MD can handle nonlinearity, but it is not correct at low temperatures. The Boltzmann-Peierls method relies on the concept of a phonon distribution function in space which is not any more meaningful in nanojunctions where translational invariance is broken. The Landauer formula takes care of the low-temperature limit of ballistic heat transport. Some attempts have been made to cover both limits, such as a phenomenological investigation using the concept of phonon mean free path [@wang-jian]. Recent works in refs. [@michel; @haanggi] are efforts from more fundamental points of views, starting from quantum principles. However, these attempts rely on specific models and approximations. Clearly, a unified approach valid for the whole temperature range is still lacking.
In this paper we give a theory for heat transport in nanojunction using nonequilibrium Green’s functions. Our approach is an exact, first-principle formulism for general models with nonlinear interactions, provided that a self-energy can be computed. This technique is used extensively in electronic transport [@meir-wingreen]. Our theory goes beyond linear elastic regime [@ciraci; @mingo] by taking nonlinearity perturbatively or through mean-field approximations. A comparison of several approximations to the self-energy on a one-dimensional (1D) chain suggests that mean-field is reliable up to room temperature. We then apply the method to short carbon nanotube junctions and compare with experimental results.
We consider an insulating solid where only the vibrational degrees of freedom are important. The system is composed of a left lead and a right lead with an arbitrary junction region. Let the displacement from some equilibrium position for the $j$-th degree of freedom in the region $\alpha$ be $u_j^\alpha$, $\alpha = L, C, R$. The quantum Hamiltonian is given by $${\cal H} = \!\!\!\!\!\sum_{\alpha=L,C,R}\!\!\!\!\!H_\alpha + (u^L)^T V^{LC} u^C + (u^C)^TV^{CR} u^R + H_n,$$ where $T$ denotes matrix transpose, $H_{\alpha} = \frac{1}{2}
{(\dot{u}^\alpha)}^T \dot{u}^\alpha + \frac{1}{2} {(u^\alpha)}^T
K^\alpha u^\alpha$, $u^\alpha$ is a column vector consisting of all the displacement variables in region $\alpha$, and $\dot{u}^\alpha$ is the corresponding conjugate momentum. $K^\alpha$ is the spring constant matrix and $V^{LC}=(V^{CL})^T$ is the coupling matrix of the left lead to the central region; similarly for $V^{CR}$. For brevity, we have set all the atomic masses to 1, but the formulas are equally applicable to variable masses with a transformation $u_j \to
x_j\sqrt{m_j}$. Also, we’ll set the Planck constant $\hbar$ and Boltzmann constant $k_B$ to 1. The nonlinear part of the interaction is $$H_n = \frac{1}{3} \sum_{ijk} T_{ijk}\, u_i^C u_j^C u_k^C.$$ Quartic interaction can also be handled.
A great simplification is possible due to the linear nature of the leads $H_L$, $H_R$, and the interaction $V^{C\alpha}$ with the central region. Only the central region has nonlinear interactions. The leads are assumed semi-infinitely long which produce dissipation. A traditional approach for many-body problems is to work in second quantization framework with the phonon creation and annihilation operators. Yet, for junction systems without translational invariance, we find that the notation and the Green’s functions will be simpler if we stay in the first quantization and in the coordinate representation [@doniach].
The present proposal is parallel to the ideas of the nonequilibrium Green’s function method for electronic transport. Imagine that at time $-\infty$ the system is in three separate subsystems in respective thermal equilibrium at the inverse temperature $\beta_\alpha$, $\alpha = L, C, R$. The couplings are switched on adiabatically, so that at time $t=0$, a steady state is established. A key starting point is an expression for the heat current. We begin with the definition, $$I_L = - \langle \dot{H}_L(t) \rangle,$$ where the decrease in energy of the left lead gives the heat flow to the central region. The average is taken with respect to an unknown density matrix and will be clarified later. By the Heisenberg equation of motion, we obtain, at $t=0$, $I_L = \langle (\dot{u}^L)^T V^{LC} u^C
\rangle$. The expectation value can be expressed in terms of a Green’s function $G^{<}_{CL}(t,t') = - i \langle u^L(t') u^C(t)^T
\rangle^T$. Using the fact that operators $u$ and $\dot{u}$ are related in Fourier space as $\dot{u}[\omega] = (-i\omega) u[\omega]$, we get, $$I_L = - \frac{1}{2\pi} \int_{-\infty}^\infty \!\!\!\!{\rm Tr}\left(
V^{LC} G^{<}_{CL}[\omega]\right) \omega\, d\omega.$$ The next step is to eliminate the reference to the lead Green’s functions in favor of the Green’s functions of the central region. We use the contour ordered Green’s function, defined on a Keldysh contour [@keldysh] from $-\infty$ to $+\infty$ and back. The contour ordered Green’s function can be mapped onto four different normal Green’s functions by $G^{\sigma\sigma'}(t,t') = \lim_{\epsilon \to
0^+} G(t\! +\! i \epsilon \sigma, t'\!+\! i\epsilon \sigma')$, where $\sigma = \pm (1)$, and $G^{++} = G^{t}$ is the time ordered Green’s function, $G^{--} = G^{\bar{t}}$ is the anti-time ordered Green’s function, $G^{+-} = G^{<}$, and $G^{-+} = G^{>}$. The retarded Green’s function is given by $G^r = G^t - G^{<}$, and the advanced by $G^a = G^{<} - G^{\bar{t}}$. These relations also hold for the self energy discussed below. In terms of the contour ordered Green’s function, it can be shown from an equation of motion method or direct verification by definition, for our model, that $G_{CL}(\tau, \tau') =
\int d\tau'' G_{CC}(\tau, \tau'') V^{CL} g_L(\tau'', \tau')$, where the integral is along the contour. The function $g_L$ is the contour ordered Green’s function for the semi-infinite free left lead in equilibrium at $\beta_L$, e.g., the retarded Green’s function in frequency domain is obtained by the solution of $\bigl[(\omega + i
\eta)^2 - K^L\bigr]g^r_L[\omega] = I$, $\eta \to 0^+$, where $I$ is an identity matrix. Using the Langreth theorem and transforming to Fourier space, the above relation gives us $G^{<}_{CL}[\omega] =
G^r_{CC}[\omega] V^{CL} g^{<}_L[\omega]
+G^{<}_{CC}[\omega]V^{CL}g^{a}_L[\omega]$. The final expression for the energy current is $$I_L = - \frac{1}{2\pi}\!\! \int_{-\infty}^{+\infty}\!\!\!\!\!\!\!\! d\omega\, \omega
\, {\rm Tr}\left( G^r[\omega] \Sigma^{<}_L[\omega] +
G^{<}[\omega] \Sigma^a_L[\omega] \right),
\label{heat-current}$$ where the self-energy due to the interaction with the lead is $\Sigma_L = V^{CL} g_{L} V^{LC}$. For notational simplicity, we have omitted the subscript $C$ on the Green’s functions denoting the central region.
Next, we need a method to compute the full Green’s functions. The perturbative/diagrammatic expansion is used to derive Dyson equations. We can treat both the coupling $V^{C\alpha}$ and the nonlinear term $H_n$ as perturbations, or consider only the nonlinearity as a perturbation. The latter is simpler in terms of organization. The contour ordered Green’s function is expressed in interaction picture: $$\begin{aligned}
G_{jk}(\tau, \tau') & = & - i \langle T_{\tau} u_j^H(\tau) u_k^H(\tau') \rangle
\nonumber\\
& = & - i \langle T_{\tau}
u_j^I(\tau) u_k^I(\tau')
e^{-i \int H_n^I(\tau'')d\tau''}
\rangle_0,\end{aligned}$$ where the displacements refer to the central region, the operators in the top line are in Heisenberg picture; $T_\tau$ is the contour order operator. The average $\langle \cdots \rangle_0$ is with respect to the density matrix of the nonequilibrium steady state when $H_n=0$. Its explicit form is not known, but the Wick theorem is still valid. The Green’s function $G_0$ of the linear system can be computed from that of the free subsystems: $$G_0(\tau,\tau')\! = \! g_C(\tau,\tau')\! +\!
\int\!\! d\tau_1 d\tau_2 g_C(\tau,\tau_1) \Sigma(\tau_1,\tau_2) G_0(\tau_2,\tau'),
\label{dyson0}$$ where $\Sigma = \Sigma_L + \Sigma_R$. The full nonlinear Green’s function has three types of diagrams in a perturbation expansion. Diagrams with loops disconnected from the two terminals are zero and can be dropped. There is another class of diagrams where the two terminals are not connected. Such diagrams are not zero, but are constants in $\tau$. They represent a thermal expansion effect, and do not contribute to the heat current in Eq. (\[heat-current\]). Finally, the connected part of the Green’s function satisfies a similar contour ordered Dyson equation relating $G_c$ to $G_0$ using nonlinear self-energy $\Sigma_n$. In ordinary Green’s functions and in frequency domain ($\omega$ argument suppressed), the Dyson equations have solutions [@haug]: $$\begin{aligned}
G_0^r &\!=\!& {G_0^a}^\dagger = \left( (\omega+i\eta)^2 \!-\! K^C
\!-\! \Sigma^r) \right)^{-1}, \label{G1} \\
G_0^{<} &\!=\!& G_0^r \Sigma^{<} G_0^a, \\
G_c^r &\!=\!& \left( {G_0^r}^{-1} - \Sigma^r_n \right)^{-1}, \\
G_c^{<} &\!=\!& G_c^r \Sigma^{<}_n G_c^a +
\left( I + G_c^r \Sigma_n^r \right) G_0^{<} \left( I + \Sigma_n^a G_c^a \right).
\label{G4}\end{aligned}$$ We note that when there is no nonlinearity, $\Sigma_n = 0$, $G =
G_0=G_c$, the heat current formula, Eq. (\[heat-current\]), can be simplified to the Landauer formula, $$I_L = - I_R = \frac{1}{2\pi} \int_0^\infty \!\!\!d\omega\, \omega\, \tilde{T}[\omega] \left(
f_L - f_R \right),$$ with the energy transmission coefficient given by the Caroli formula [@caroli; @mingo], $\tilde T[\omega] = {\rm Tr}\left(G^r \Gamma_L
G^a \Gamma_R \right)$, $\Gamma_\alpha = i(\Sigma_\alpha^r -
\Sigma_\alpha^a)$, where $f_\alpha$ is the Bose-Einstein distribution function at $\beta_\alpha$. In order to facilitate comparison with linear results, we define an effective energy transmission by $\tilde{T}_{{\rm eff}}[\omega] = \frac{1}{2} {\rm
Tr}\left\{(G^r-G^a)(\Sigma_R^{<}-\Sigma_L^{<}) + i G^{<}
(\Gamma_R-\Gamma_L) \right\}/(f_L -f_R)$. The function $\tilde{T}_{{\rm eff}}[\omega]$ is real and even in $\omega$. Such effective transmission is temperature-dependent.
![\[fig1\]Feynman diagrams for the interaction self-energy $\Sigma_n$. Each long line corresponds to a propagator $G_0(\tau,\tau')$; each vertex is associated with the interaction strength $T_{ijk}$. All internal site indices are summed and contour time variables integrated. The number in front of a graph is the factor multiplying the graph value.](feynman){width="\columnwidth"}
So far the equations above are all exact. Several ways of approximating the nonlinear contribution to self-energy are possible. We can simply truncate the diagrams [@valle] in the perturbative expansion for the self-energy. The diagrams for $\Sigma_n$ to $O(T_{ijk}^4)$ are shown in Fig. \[fig1\]. These diagrams are still in the contour variable $\tau$. For practical calculation, they have to be changed to real time $t$ in terms of $G^{\sigma \sigma'}$ and Fourier transformed to the frequency domain. For example, the leading order contribution (first two diagrams) to the nonlinear self-energy is: $$\begin{aligned}
&&\!\!\!\!\!\!\Sigma^{\sigma\sigma'}_{n,jk}[\omega] \approx 2i
\!\!\sum_{lmrs}\!\! T_{jlm} T_{rsk} \!\! \int_{-\infty}^{+\infty}
\!\!\!\!\! G_{0,lr}^{\sigma\sigma'}[\omega']
G_{0,ms}^{\sigma\sigma'}[\omega\! -\! \omega']
\frac{d\omega'}{2\pi} + \nonumber\\
&&\;\;\; 2i\sigma \delta_{\sigma,\sigma'} \!\!\!\!\!\!\!\!\!\sum_{lmrs, \sigma''=\pm 1}\!\!\!\!\!\!\!\!\!
\sigma'' T_{jkl} T_{mrs}\!\!
\int_{-\infty}^{+\infty}
\!\!\!\!\!\!\! G_{0,lm}^{\sigma\sigma''}[0]
G_{0,rs}^{\sigma''\sigma''}[\omega'] \frac{d\omega'}{2\pi}.
\label{Sigma-n}\end{aligned}$$ Mean-field theory can be obtained by replacing $G_0$ by $G$, and the equations are solved iteratively.
A general program is implemented based on Eq. (\[heat-current\]), Eq. (\[G1\]) to (\[G4\]), and Eq. (\[Sigma-n\]). The surface Green’s function $g^r_L$ is computed using an efficient recursive method [@surface-green]. In numerical calculation of the Green’s functions, it is important to keep a small but finite $\eta$, as the functions are singular in the limit $\eta \to 0^{+}$. In addition, on-site potentials are applied to the leads to make the system stable.
![\[fig2\]Thermal conductance as a function of temperature for a 1D junction with three atoms. The harmonic spring constants are $k_L = 1.56$, $k_R = 1.44$, $k_C=1.38$ (eV/(Å$^2$amu)). The nonlinear strength is $t=1.8$ eV/(Å$^3$amu$^{3/2}$). Small onsite quadratic potentials are applied to the leads with spring constants $k_L^{\rm onsite} = 0.01$, and $k_{R}^{\rm onsite} = 0.02$ (eV/(Å$^2$amu)).](kappa1D){width="\columnwidth"}
We first test various approximations on a 1D junction with parameters comparable to that of a carbon chain. The system consists of harmonic leads and a junction part with harmonic interactions plus cubic interactions of the form $(1/3) t \sum (u_{j} - u_{j+1})^3$ of Fermi-Pasta-Ulam type. Figure \[fig2\] presents the results for a 3-atom junction system. We discuss the effect of nonlinearity to thermal transport. As we can see, adding nonlinear contributions suppresses the thermal conductance at high temperatures. The deviation from ballistic transport starts around 200 Kelvin. As the temperature rises further, we expect that the higher order graphs become important. The high order calculations are rather expensive, with computational complexity scaled as $O(N^4 M^2)$ where $N$ is system size and $M$ is sampling points in frequency. To partially take into account the higher order contributions but still keep the computation within reasonable limit for large systems, we find a mean-field theory is most satisfactory. In this version of mean-field treatment, we consider the leading diagrams of $O(T^2)$, and replace $G_0$ by $G$ only for the first diagram. The equations are then solved iteratively. Good agreement with $O(T^4)$ result is found for temperatures up to 400 K for the 1D chain. Thus we expect that the mean-field theory can give excellent results for moderately high temperatures. However, both the perturbative results and mean-field one break down at sufficiently high temperature, as the cubic nonlinearity is only metastable. To predict correctly diffusive behavior at high temperatures, the quartic interaction and higher order graphs are essential.
![\[fig3\] The ballistic transmission and the effective transmission at 300K for an (8,0) one-unit-cell carbon nanotube junction.](trans){width="\columnwidth"}
We now turn to the carbon nanotube junctions. The leads and the junction are of the same diameter (8,0) nanotube. The only difference is that the junction part has cubic nonlinear interactions while the leads are perfectly harmonic. The values of $T_{ijk}$ are derived from the Brenner potential by finite differences. For computational efficiency, small values of $T_{ijk}$ are truncated to zero. Figure \[fig3\] shows the ballistic transmission coefficient when the nonlinearity is set to zero, and is compared with the effective transmission due to the leading nonlinear effects (the first two graphs). It can be seen that the nonlinearity greatly suppresses the thermal transmission, particularly at low frequencies. The discontinuity is smeared out to more smooth curve due to thermal effect.
![\[fig4\] Thermal conductance of (8,0) carbon nanotube junction with one unit cell (0.426 nm). The inset shows the thermal conductivity as a function of tube length. ](conductance){width="\columnwidth"}
The temperature and length dependence of the nanotube thermal conductance (conductivity) is shown in Fig. \[fig4\]. The mean-field theory result gives a peak in the conductance around 400K. This behavior agrees with experiments [@exp-papers] and is in contrast with MD results which tend to give peaks at much lower temperatures [@md-papers]. The inset shows the thermal conductivity calculated from the conductance. The cross-section is defined as $\pi d^2/4$, where $d$ is the diameter of the tube. If we assume that the transport up to 1 $\mu$m is still quasi-ballistic, then we can estimate that the thermal conductivity at the experimental accessible length is about 2000 W/(mK). This is qualitatively in agreement with experimental values [@exp-papers].
We proposed a fully quantum mechanical approach for computing heat current of solid junctions with nonlinear interactions. We have demonstrated the method with 1D and molecular junction systems. The perturbation expansion for self-energy works well up to room temperatures. However, it is still a challenge to find efficient and good approximation for the self-energy at high temperatures.
This work is supported in part by a Faculty Research Grant of the National University of Singapore.
[01]{} R. E. Peierls, [*Quantum Theory of Solids*]{}, Chap. 2, (Oxford University Press, 1955).
P. Carruthers, Rev. Mod. Phys. **33**, 92 (1961).
D. G. Cahill, W. K. Ford, K. E. Goodson, G. D. Mahan, A. Majumdar, H. J. Maris, R. Merlin, and S. R. Phillpot, J. Appl. Phys. **93**, 793 (2003).
S. Lepri, R. Livi, and A. Politi, Phys. Rep. **377**, 1 (2003).
J. Wang and J.-S. Wang, Appl. Phys. Lett, [**88**]{}, 111909 (2006).
M. Michel, G. Mahler, and J. Gemmer, Phys. Rev. Lett. **95**, 180602 (2005).
D. Segal, A. Nitzan, and P. Hänggi, J. Chem. Phys. **119**, 6840 (2003); D. Segal and A. Nitzan, Phys. Rev. Lett. **94**, 034301 (2005).
Y. Meir and N. S. Wingreen, Phys. Rev. Lett. **68**, 2512 (1992); A.-P. Jauho, N. S. Wingreen, and Y. Meir, Phys. Rev. B **50**, 5528 (1994).
A. Ozpineci and S. Ciraci, Phys. Rev. B **63**, 125415 (2001).
N. Mingo and L. Yang, Phys. Rev. B **68**, 245406 (2003).
S. Doniach and E. H. Sondheimer, *Green’s Functions for Solid State Physicists*, Chap. 1, (W. A. Benjamin, Reading, 1974).
L. V. Keldysh, Sov. Phys. JETP, **20**, 1018 (1965).
H. Haug and A.-P. Jauho, *Quantum Kinetics in Transport and Optics of Semiconductors*, (Springer, Berlin, 1996).
C. Caroli, R. Combescot, P. Nozieres, and D. Saint-James, J. Phys. C **4**, 916 (1971).
R. G. D. Valle and P. Procacci, Phys. Rev. B **46**, 6141 (1992).
M. P. López Sancho, J. M. López Sancho, and J. Rubio, J. Phys. F: Met. Phys. **15**, 851 (1985).
P. Kim, L. Shi, A. Majumdar, and P. L. McEuen, Phys. Rev. Lett. **87**, 215502, (2001); E. Pop, D. Mann, Q. Wang, K. Goodson, and H. Dai, Nano Letters, **6**, 96 (2006).
S. Berber, Y.-K. Kwon, and D. Tománek, Phys. Rev. Lett. **84**, 4613 (2000).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study, by using tools of the dynamical system theory, a fermionic string streched from a non-commutative D2-brane (stack of D0-branes in the BFSS model) to a probe D0-brane as a quantum system driven by a chaotic system, the (classical and quantum) chaos being induced in the D2-brane dynamics by quantum fluctuations. We show that this dynamics with fluctuations induces a decoherence phenomenon on the reduced density matrix of the fermionic string which is characteristic of the chaotic behaviour since it presents an horizon of coherence. Moreover we show that this dynamics is associated with an invariant torus which involves extradimensions emerging from the fluctuations for the viewpoint of the fermionic string, extending a three dimensional space by six compact dimensions. The situation studied can be considered as a model of qubit (supported by the fermionic string) in interaction with a quantum black hole (modelized by the non-commutative D2-brane).'
address: 'Institut UTINAM (CNRS UMR 6213, Université de Bourgogne-Franche-Comté, Observatoire de Besançon), 41bis Avenue de l’Observatoire, BP1615, 25010 Besançon cedex, France.'
author:
- 'David Viennot & Lucile Aubourg'
title: 'Chaos, decoherence and emergent extradimensions in D-brane dynamics with fluctuations'
---
[*Keywords*]{}: Brane mechanics, matrix model, quantum fluctuations, decoherence, quantum chaos, extradimensions, black hole.
Introduction
============
In a recent paper [@Viennot], we have studied the decoherence phenomenon induced on a qubit (supported by the spin of a fermion) falling into a black hole. This study has needed some strong semi-classical approximations in quantum field theory on curved space-time (only first quantization of the fermion, localization of the qubit (WKB approximation), adiabatic approximation of the dynamics). In the present paper, we want to consider this problem in a “quantum gravity context”. An interesting model of quantum black holes arises in the matrix model with D-branes [@Asplund; @Berenstein; @Aoki]. In the BFSS (Banks-Fischler-Shenker-Susskind) matrix theory [@BFSS], a stack of D0-branes forms a noncommutative D2-brane represented by Hermitian matrices (the noncommutative coordinates on the D2-brane). The initial conditions on these matrices fix the topology of the D2-brane (as for example a fuzzy sphere) and adding small initial fluctuations induces thermalization of the system in a state corresponding to a black hole horizon. By adding a fermionic string stretched from the D2-brane to a probe D0-brane, as for example in [@Berenstein], we have a qubit model supported by the spin inner degree of freedom of the fermionic string. It has been proved [@Asano; @Gur; @Hanada] that under these conditions the matrix brane dynamics involves chaotic motions. The fermionic string is then driven by a chaotic system. In previous papers [@Viennot2; @Aubourg; @Aubourg2] we have shown that quantum systems driven by chaotic flows are distinguished from quantum systems driven by random noises by a phenomenon called horizon of coherence, i.e. a duration at the beginning of the dynamics in which the quantum coherence is sustained before to fall with an increase of the entropy. In the present paper, we want to study this phenomenon for a fermionic string in contact with a D2-brane. Moreover we have proposed in [@Viennot3] a natural mathematical approach to study quantum systems driven by chaotic flows, based on the Schrödinger-Koopman formalism [@Sapin; @Jauslin; @Gay]. This formalism is a generalization of the Floquet theory for which we have shown [@Viennot4] that it has similarities with the classical string theory. In particular, a closed extradimension must be added to the control manifold (playing the role of the space) in the context of this Floquet theory. In this paper, we want to show that by a similar mechanism, compact extradimensions emerge from the quantum chaotic fluctuations from the viewpoint of the fermionic string treated by the Schrödinger-Koopman approach. For this reason, at the starting point we consider a matrix model in a three dimensional space, the other compact six dimensions of the BFSS model emerging from the fluctuations in our approach. Note that the reduction to three dimensions, can also be interpreted as an orbifold reduction as in [@Berenstein].\
This paper is organized as follows. Section 2 presents elementary ingredients of the BFSS matrix theory. The different models used for the numerical simulations are also presented. Section 3 studies the dynamics of the quantum fluctuations, in particular we show the existence of an invariant torus in the phase space of the matrices modelizing the noncommutative D2-brane. Section 4 presents the main results of this paper concerning the dynamics of the fermionic string driven by the D2-brane: a quantum chaos signature in its spectrum, an horizon of coherence in the evolution of its reduced density matrix and the emergence of the compact extradimensions from the fluctuations. Throughout this paper, our results are enlighten by numerical simulations. In particular, we drawn the geometry of the compact extradimension manifold emerging in our approach.\
*Throughout this paper, we consider the unit system such that $\hbar=c=G=1$ ($\ell_P=t_P=m_P=1$ for the Planck units)*.
The BFSS model
==============
In this section we present the elementary ingredients of the BFSS theory, to more details the reader can see [@BFSS; @Sochichiu; @Zarembo].\
We consider a stack of $N$ D0-branes in a 3D-space, represented by Hermitian matrices $X_i \in \mathfrak M_{N \times N}(\mathbb C)$ for $i=1,2,3$ (noncommutative coordinates for the D2-brane formed by the stack). Intuitively, $X^i = \left(\begin{array}{cc} x^i_1 & s^i_{12} \\ \overline s^i_{12} & x^i_2 \end{array} \right)$ represents a stack of two D0-branes with coordinates in the target space $\mathbb R^3$ $\{x^i_1\}_i$ and $\{x^i_2\}_i$, linked by a bosonic string of oscillation radii $\{|s^i_{12}|\}_i$ ($\arg s^i_{12}$ is the phase for the oscillation in the $i$-direction).\
We consider also a probe D0-brane of coordinates $x_i \in \mathbb R$ and a fermionic string linking the D2-brane to the probe brane described by a state:\
$|\psi {\rangle \hspace{-0.2em} \rangle}= |\uparrow \rangle \otimes \left(\begin{array}{c} \psi^\uparrow_1 \\ \vdots \\ \psi^\uparrow_N \end{array} \right)+ |\downarrow \rangle \otimes \left(\begin{array}{c} \psi^\downarrow_1 \\ \vdots \\ \psi^\downarrow_N \end{array} \right) \in \mathbb C^2 \otimes \mathbb C^N$. We can interpret $\psi^\alpha_a$ as $\psi^\alpha(x_a)$ ($x_a$ being the pseudo-position of the $a$-th D0-brane of the stack): the spatial delocalisation of quantum particles is replaced by the quantum superposition of $N$ attachment points for the string. The inner degree of freedom modelized by the Hilbert space $\mathbb C^2$ (with canonical basis $(|\uparrow \rangle, |\downarrow \rangle)$ is the spin of the fermionic string.\
The total system is represented by the matrices: $$\mathbf X_i = \left(\begin{array}{cc} X_i & |\delta x_i\rangle \\ \langle \delta x_i| & x_i \end{array} \right) ; \qquad \Psi = \left(\begin{array}{cc} 0 & |\psi{\rangle \hspace{-0.2em} \rangle}\\ 0 & 0 \end{array} \right)$$ where $|\delta x_i \rangle \in \mathbb C^N$ represents quantum fluctuations of the vacuum creating and annihilating bosonic strings linking the stack (the D2-brane) to the probe D0-brane. $|\delta x_i(t=0)\rangle$ is randomly choosen following a gaussian law.\
The dynamics of the system is governed by the Lagrangian densities: $$\mathscr L_{boson} = \frac{1}{2} \dot {\mathbf X}^i \dot {\mathbf X}_i + \frac{1}{4} [\mathbf X_j,\mathbf X^i][\mathbf X^j,\mathbf X_i]$$ $$\mathscr L_{fermion} = -\imath \Psi^\dagger \dot \Psi + \Psi^\dagger \sigma^i \otimes [\mathbf X_i,\Psi]$$ involving the following equation of motion: $$\label{equX}
\ddot {\mathbf X}_i - [\mathbf X_j,[\mathbf X_i,\mathbf X^j]] = 0$$ with the Gauss constraint $[\dot {\mathbf X}_i,\mathbf X^i] = 0$, $$\label{equSchro}
\imath |\dot \psi {\rangle \hspace{-0.2em} \rangle}= \sigma^i \otimes(X_i-x_i) |\psi {\rangle \hspace{-0.2em} \rangle}$$ $\{\mathbf X_i\}_i$ obeys then to a matrix classical wave equation ($[\mathbf X_j,[\mathbf X_i,\mathbf X^j]]$ taking the role of a noncommutative Laplacian). The string state $|\psi {\rangle \hspace{-0.2em} \rangle}$ obeys to a Schrödinger equation with quantum Hamiltonian $H^{eff}(t) = \sigma^i \otimes(X_i(t)-x_i(t))$ [@Berenstein; @Badyn].\
The initial conditions concerning $X_i(t=0)$ fix an initial topology for the noncommutative D2-brane. Usually, the works concerning such a model use a fuzzy sphere at $t=0$. In order to test the independance of our analysis from the initial topology we consider also a fuzzy ellipsoid, a fuzzy cylinder, a fuzzy torus and two noncompact noncommutative manifolds, a noncommutative plane and a fuzzy hyperboloid. The characteristics of these models are presented table \[NCmanifold\].
+-----------------------+-----------------------+-----------------------+
| [*manifold*]{} | [*algebra*]{} | [*coordinates*]{} |
+:======================+:======================+:======================+
| Fuzzy sphere | $j$-representation of | $X_i = \frac{1}{j} J_ |
| | $\mathfrak{su}(2)$ | i$ |
| | ($N=2j+1$)\ | |
| | $[J_i,J_j] = \imath \ | |
| | epsilon_{ijk} J_k$ | |
+-----------------------+-----------------------+-----------------------+
| Fuzzy ellipsoid | | $X_1=\frac{2}{j} J_1$ |
| | | , |
| | | $X_2=\frac{2}{j} J_2$ |
| | | , |
| | | $X_3=\frac{1}{j} J_3$ |
+-----------------------+-----------------------+-----------------------+
| Fuzzy cylinder | $j$-representation of | $X_1 = \frac{1}{j} \f |
| | $\mathfrak{iso}(2)$ | rac{V+V^\dagger}{2}$, |
| | ($N=2j+1$)\ | $X_2 = \frac{1}{j} \f |
| | $[C,V]=V$ | rac{V-V^\dagger}{2\im |
| | | ath}$, |
| | | $X_3 = \frac{1}{j} C$ |
| | | . |
+-----------------------+-----------------------+-----------------------+
| Fuzzy torus | $C = \sum_{n=0}^{N-1} | $X_1 = \frac{V+V^\dag |
| | C_n |n\rangle \langl | ger}{2}$, |
| | e n|$\ | $X_2 = \frac{V-V^\dag |
| | $V=\sum_{n=0}^{N-1} V | ger}{2\imath}$, |
| | _n |n\rangle \langle | $X_3 = C$. |
| | n+1|$\ | |
| | with | |
| | $|N\rangle = |0\rangl | |
| | e$, | |
| | $C_n = \frac{1}{N\alp | |
| | ha} \sin(\frac{2\pi ( | |
| | n+\delta)}{N})$ | |
| | and | |
| | $V_n = \sqrt{\frac{\c | |
| | os(\frac{2\pi (n+\del | |
| | ta+0.5)}{N})}{N\alpha | |
| | \cos(\pi/N)}+\frac{1} | |
| | {N}}$ | |
| | ($\alpha \in \mathbb | |
| | R^+$, | |
| | $\delta \in [0,1]$). | |
+-----------------------+-----------------------+-----------------------+
| NC plane | $\mathfrak{ccr}$ | $X_1 = \frac{a+a^+}{2 |
| | algebra | }$, |
| | ($N=+\infty$)\ | $X_2 = \frac{a-a^+}{2 |
| | $[a,a^+]={\mathrm{id} | \imath}$, |
| | }$ | $X_3=0$ |
+-----------------------+-----------------------+-----------------------+
| Fuzzy hyperboloid | $k$-representation of | $X_i = \frac{1}{k} K_ |
| | $\mathfrak{su}(1,1)$ | i$ |
| | ($N=+\infty$)\ | |
| | $[K_1,K_2]=-\imath K_ | |
| | 3$,\ | |
| | $[K_3,K_1]=\imath K_2 | |
| | $,\ | |
| | $[K_2,K_3]=\imath K_1 | |
| | $ | |
+-----------------------+-----------------------+-----------------------+
: \[NCmanifold\] Noncommutative manifolds used as initial conditions for the D2-brane (see also [@Sykora]). For the numerical simulations, the choosen parameters of the models are $\alpha=1.5$, $\delta=0.4$ and $k=12.5$.
Numerical simulations for the integration of eq. \[equX\] are realized by using a leap-frog integrator as in [@Asplund]. For the non-compact manifolds, a cutoff in the basis is realized which needs a numerical correction to avoid numerical artefacts (see \[AppA\]). The integration of the Schrödinger equation \[equSchro\] is realised by a split operator method for the evolution operator: $U(t_n,0) \simeq e^{-\imath H^{eff}(t_n) \Delta t}...e^{-\imath H^{eff}(0) \Delta t}$ for a partition $0<t_1<...<t_n$ of step $\Delta t$ equal to the one used in the leap-frog integrator.
Dynamics of the quantum fluctuations
====================================
D-brane equation with fluctuations
----------------------------------
The equation $\ddot{\mathbf X}_i - [\mathbf X_j,[\mathbf X_i,\mathbf X^j]]=0$ corresponds to the set of coupled equations: $$\begin{aligned}
\ddot X_i & = & [X_j,[X_i,X^j]] \nonumber \\
& & + [X_j,|\delta x_i\rangle \langle \delta x^j|-|\delta x^j\rangle \langle \delta x_i|] - \{|\delta x_j\rangle \langle \delta x^j|,X_i-x_i\} \nonumber \\
& & + |\delta x_j\rangle \langle \delta x_i|(X^j-x^j) - (X^j-x^j)|\delta x_i\rangle \langle \delta x_j|\end{aligned}$$ $$\ddot x_i = 2\langle \delta x_j|(X_i-x_i)|\delta x^j\rangle -2\Re\mathrm{e}(\langle \delta x_j|(X^j-x^j)|\delta x_i\rangle)$$ $$\begin{aligned}
\label{equfluct}
|\delta \ddot x_i \rangle & = & - |X-x|^2|\delta x_i\rangle -([X_i,X^j]-(X^j-x^j)(X_i-x_i))|\delta x_j \rangle \nonumber \\
& & -\langle \delta x^j|\delta x_j\rangle |\delta x_i\rangle +(2\langle \delta x_i|\delta x^j\rangle - \langle \delta x^j|\delta x_i\rangle)|\delta x_j\rangle\end{aligned}$$ where $|A|^2 = A_i A^i$ for a set of matrices $\{A_i\}_{i=1,2,3}$ and $\{\cdot,\cdot\}$ denoting the anticommutator. We are interested by the last equation which governs the evolution of the fluctuations. Some examples are drawn figure \[fluctuations\].
![\[fluctuations\] Evolution of the amplitudes of the fluctuations $|\delta x_i(t) \rangle$, for the fuzzy sphere, ellipsoid, torus and the noncommutative plane. Initial conditions for $|\delta x_i(t=0) \rangle$ are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=201$ (for the noncommutative plane the cutoff is $N=11$) (simulations show that the thermalization duration grows with $N$).](fluctSphere201.eps "fig:"){width="6.4cm"} ![\[fluctuations\] Evolution of the amplitudes of the fluctuations $|\delta x_i(t) \rangle$, for the fuzzy sphere, ellipsoid, torus and the noncommutative plane. Initial conditions for $|\delta x_i(t=0) \rangle$ are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=201$ (for the noncommutative plane the cutoff is $N=11$) (simulations show that the thermalization duration grows with $N$).](fluctEllipsoid201.eps "fig:"){width="6.4cm"}\
![\[fluctuations\] Evolution of the amplitudes of the fluctuations $|\delta x_i(t) \rangle$, for the fuzzy sphere, ellipsoid, torus and the noncommutative plane. Initial conditions for $|\delta x_i(t=0) \rangle$ are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=201$ (for the noncommutative plane the cutoff is $N=11$) (simulations show that the thermalization duration grows with $N$).](fluctTorus201.eps "fig:"){width="6.4cm"} ![\[fluctuations\] Evolution of the amplitudes of the fluctuations $|\delta x_i(t) \rangle$, for the fuzzy sphere, ellipsoid, torus and the noncommutative plane. Initial conditions for $|\delta x_i(t=0) \rangle$ are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=201$ (for the noncommutative plane the cutoff is $N=11$) (simulations show that the thermalization duration grows with $N$).](fluctPlane.eps "fig:"){width="6.4cm"}
We see that the fluctuations evolve until to reach a plateau corresponding to the thermalization of the system (the system with the fluctuations has reached a steady state). The existence of this long term stability suggests that the dynamical system described by $(|\delta x_i\rangle)_i$ presents an invariant manifold.
Existence of an invariant torus {#invtorus}
-------------------------------
We consider the nonlinear equation \[equfluct\] where we treat $\{X_i\}_i$ and $\{x_i\}_i$ as being constant (the variations of these quantities are supposed very slow compared to the variations of $\{|\delta x_i \rangle\}_i$ and we consider them as being frozen in a first time).\
We restrict our attention to the case $N=1$ which is instructive for our discussion, $|\delta x_i \rangle$ is then reduced to be a complex scalar $\delta x_i \in \mathbb C$ obeing to the nonlinear equation: $$\label{eqsd}
\delta \ddot x_i = -\|X-x\|^2 \delta x_i + (X^j-x^j)(X_i-x_i)\delta x_j - 2 \overline{\delta x^j} \delta x_j \delta x_i + 2 \overline{\delta x_i} \delta x^j \delta x_j$$ We can rewritte this equation as $$\left\{ \begin{array}{rcl} \delta \dot x_i & = & F_i(\delta p) = \delta p_i \\ \delta \dot p_i & = & F_{\hat i}(\delta x) \end{array} \right.$$ where $\delta p_i \in \mathbb C$ is the momentum associated with $\delta x_i$, $F_{\hat i}(\delta x)$ being the right hand side of the equation \[eqsd\]. We adopt the notations such that the indices $i$, $\hat i$, $\bar i$ and $\tilde i$ respectively stand for $\delta x_i$, $\delta p_i$, $\overline{\delta x_i}$ and $\overline{\delta p_i}$. These equations define a dynamical system in the phase space $\Gamma = \mathbb C^6 \simeq \mathbb R^{12}$. It is clear that $\delta x_i = 0$ and $\delta p_i = 0$ ($\forall i$) is a fixed point of the dynamical system. Since the quantum fluctuations are small, the starting point of the dynamical system is in the neighbourhood of $0$. The Jacobian matrix of the dynamical system is $$\begin{aligned}
{\partial F_{\hat i}}^j = \frac{\partial F_{\hat i}}{\partial \delta x_j} & = & -\|X-x\|^2\delta^i_j + (X_i-x_i)(X_j-x_j) \nonumber \\
& & \quad - 2 \overline{\delta x^j} \delta x_i - 2 \overline{\delta x^k}\delta x_k \delta^i_j + 4 \overline{\delta x_i} \delta x^j \\
{\partial F_{\hat i}}^{\bar j} = \frac{\partial F_{\hat i}}{\partial \overline{\delta x_j}} & = & -2\delta x_j\delta x_i + 2 \delta x^k \delta x_k \delta^i_j \\
{\partial F_{i}}^{\hat j} = \frac{\partial F_{i}}{\partial \delta p_j} & = & \delta^i_j\end{aligned}$$ and ${\partial F_{i}}^{j} = {\partial F_{\hat i}}^{\hat j} = {\partial F_{i}}^{\tilde j} = {\partial F_{i}}^{\bar j} = {\partial F_{\hat i}}^{\tilde j} = 0$. It follows that the Jacobian matrix at $0$ (in the representation $(i,\hat i,\bar i, \tilde i)$) is $$\partial F_{|(\delta x,\delta p)=0} = \left(\begin{array}{cc} S & 0 \\ 0 & S \end{array} \right)$$ with $$S = \left(\begin{array}{cc} 0 & -I \\ {\mathrm{id}}_3 & 0 \end{array} \right)$$ where $I$ is the following “inertia matrix”: $$I = \left(\begin{array}{ccc} Y_2^2+Y_3^2 & -Y_1Y_2 & -Y_1Y_3 \\ -Y_1Y_2 & Y_1^2+Y_3^2 & -Y_2Y_3 \\ -Y_1 Y_3 & -Y_2 Y_3 & Y_1^2+Y_2^2 \end{array} \right)$$ with $Y_i = X_i-x_i$. ${\mathrm{Sp}}(S) \subset \imath \mathbb R$, indeed let $\lambda$ be an eigenvalue of $S$: $Se = \lambda e$. $e = (\mathring e,\hat e)$ with $\mathring e, \hat e \in \mathbb C^3$. The eigenequation is equivalent to $\left\{ \begin{array}{rcl} - I \hat e & = & \lambda \mathring e \\ \mathring e & = & \lambda \hat e \end{array} \right.$, and then $-I \hat e = \lambda^2 \hat e$. $\lambda^2 \in {\mathrm{Sp}}(-I)$, but $I \geq 0$ and then $\lambda^2 \leq 0 \iff \lambda \in \imath \mathbb R$. $S$ being real, its spectrum contains three couples of conjugated purely imaginary eigenvalues.\
The 12 local Lyapunov values of the dynamical system in the neighbourhood of $0$ being purely imaginary, the fixed point is simply stable, the evolution of $\delta x$ is then quasi-periodic with 3 fundamental frequencies (the eigenvalues of $\sqrt I$) and the phase trajectory in $\Gamma$ is wrapped around an invariant 6-torus $\mathbb T^6$. By the KAM (Kolmogorov-Arnold-Moser) theorem [@Goldstein], the torus is stable under the perturbations induced by the fluctuations of $X$ and $x$, and it is slowly deformed with the global evolutions of $X(t)$ and of $x(t)$.\
Now we consider the case $N>1$. By a similar approach, the Jacobian matrix at $0$ of the dynamical system governing the quantum fluctuations is $$\partial F_{|(|\delta x\rangle,|\delta p \rangle) = 0} = \left(\begin{array}{cc} S & 0 \\ 0 & S \end{array} \right), \qquad S = \left(\begin{array}{cc} 0 & -I_N \\ {\mathrm{id}}_{3N} & 0 \end{array} \right)$$ with the $\mathfrak M_{N \times N}(\mathbb C)$-valued “inertia matrix” $$I_N = \left(\begin{array}{ccc} Y_2^2+Y_3^2 & -Y_2Y_1+[Y_1,Y_2] & -Y_3Y_1 + [Y_1,Y_3] \\ -Y_1Y_2 + [Y_2,Y_1] & Y_1^2+Y_3^2 & -Y_3Y_2 + [Y_2,Y_3] \\ -Y_1 Y_3 + [Y_3,Y_1] & -Y_2 Y_3 + [Y_3,Y_2] & Y_1^2+Y_2^2 \end{array} \right)$$ Let $|\delta x_i \rangle = \delta x_i |u_i \rangle$ be a decomposition of the fluctuation vectors with $\delta x_i$ being the complex amplitude of the quantum fluctuations and $|u_i \rangle$ being the “polarisation” of the fluctuations (with $\|u_i(t=0)\|=1$). We set that $\delta x_i$ obeys to the same equation that in the case $N=1$ (with an abelianization of the brane coordinates): $$\begin{aligned}
\label{eqsd2}
\delta \ddot x_i & = & - \tr |X-x|^2 \delta x_i + \tr \left((X^j-x^j)(X_i-x_i)\right)\delta x_j \nonumber \\
& & \quad - 2 \overline{\delta x^j} \delta x_j \delta x_i + 2 \overline{\delta x_i} \delta x^j \delta x_j\end{aligned}$$ It follows that the Jacobian matrix of this equation is the same than for $N=1$ with the inertia matrix: $$I = \left(\begin{array}{ccc} \tr(Y_2^2+Y_3^2) & -\tr(Y_1Y_2) & -\tr(Y_1Y_3) \\ -\tr(Y_1Y_2) & \tr(Y_1^2+Y_3^2) & -\tr(Y_2Y_3) \\ -\tr(Y_1 Y_3) & -\tr(Y_2 Y_3) & \tr(Y_1^2+Y_2^2) \end{array} \right)$$ The phase trajectory of $\delta x$ is then wrapped around a 6-torus $\mathbb T^6$.\
Concerning the polarisation, it obeys to $$\begin{aligned}
\label{pola}
|\ddot u_i \rangle & = & -\mathrm{trm}\,|X-x|^2 |u_i \rangle - 2 \frac{d\ln \delta x_i}{dt} |\dot u_i \rangle \nonumber \\
& & \quad -\mathrm{trm} \left([X_i,X^j]-(X^j-x^j)(X_i-x_i) \right) \frac{\delta x_j}{\delta x_i} |u_j\rangle \nonumber \\
& & \quad +\left(\overline{\delta x^j} \delta x_j - 2 \frac{\overline{\delta x_i}}{\delta x_i} \delta x^j \delta x_j \right) |u_i \rangle \nonumber \\
& & \quad +\left(2 \frac{\overline{\delta x_i}}{\delta x_i} \delta x^j \delta x_j \langle u_i|u^j \rangle - \overline{\delta x^j}\delta x_j \langle u^j|u_i \rangle\right) |u_j \rangle\end{aligned}$$ where $\mathrm{trm}\, A = A - \tr A$.\
In order to illustrate the deformation of the invariant torus with respect to the evolution of $X(t)$ and $x(t)$, we can draw the Lyapunov frequencies of the dynamical system with respect to $t$, see fig. \[loclyap\].
![\[loclyap\] Evolution of the local Lyapunov frequencies of the equation \[eqsd2\] (${\mathrm{Sp}}(\sqrt{I})$) for fuzzy sphere, ellipsoid, torus and cylinder. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=201$.](locLyapSphere201.eps "fig:"){width="6.4cm"} ![\[loclyap\] Evolution of the local Lyapunov frequencies of the equation \[eqsd2\] (${\mathrm{Sp}}(\sqrt{I})$) for fuzzy sphere, ellipsoid, torus and cylinder. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=201$.](locLyapEllipsoid201.eps "fig:"){width="6.4cm"}\
![\[loclyap\] Evolution of the local Lyapunov frequencies of the equation \[eqsd2\] (${\mathrm{Sp}}(\sqrt{I})$) for fuzzy sphere, ellipsoid, torus and cylinder. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=201$.](locLyapTorus201.eps "fig:"){width="6.4cm"} ![\[loclyap\] Evolution of the local Lyapunov frequencies of the equation \[eqsd2\] (${\mathrm{Sp}}(\sqrt{I})$) for fuzzy sphere, ellipsoid, torus and cylinder. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=201$.](locLyapCylinder201.eps "fig:"){width="6.4cm"}
The effect of the thermalization can be also viewed with the evolution of the Lyapunov frequencies.
Dynamics onto the invariant torus
---------------------------------
Now we want describe the dynamics of the fluctuations onto the invariant torus $\mathbb T^6$. The linearized dynamics is defined by $$\begin{aligned}
\delta x(t) & \simeq & \sum_{\alpha=1}^3 (c_\alpha e^{\lambda_\alpha t} \mathring e_\alpha + d_\alpha e^{-\lambda_\alpha t} \mathring e_\alpha^*) \\
\delta p(t) & \simeq & \sum_{\alpha=1}^3 (c_\alpha e^{\lambda_\alpha t} \hat e_\alpha + d_\alpha e^{-\lambda_\alpha t} \hat e_\alpha^*)\end{aligned}$$ where $\lambda_\alpha \in \imath \mathbb R$ are the Lyapunov values and $-I \hat e_\alpha = \lambda_\alpha^2 e_\alpha$, $\mathring e_\alpha = \lambda_\alpha \hat e_\alpha$ ($S e_\alpha = \lambda_\alpha e_\alpha$ and $S e_\alpha^* = - \lambda_\alpha e_\alpha^*$, the star denoting the complex conjugation). $$\begin{aligned}
c_\alpha & = & \mathring e_\alpha^* \cdot \delta x(0) + \hat e_\alpha^* \cdot \delta p(0) \\
d_\alpha & = & \mathring e_\alpha \cdot \overline{\delta x(0)} + \hat e_\alpha \cdot \overline{\delta p(0)}\end{aligned}$$ Let $\{\theta_\alpha\}_{\alpha \in \{1,...,6\}}$ be the local coordinates onto $\mathbb T^6$. Since the previous equations represent a phase trajectory wrapped around $\mathbb T^6$, the immersion equations of $\mathbb T^6$ into $\mathbb C^6$ are $$\begin{aligned}
\delta x & \simeq & \sum_{\alpha=1}^3 (c_\alpha e^{\imath \theta_\alpha} \mathring e_\alpha + d_\alpha e^{\imath \theta_{\alpha+3}} \mathring e_\alpha^*) \\
\delta p & \simeq & \sum_{\alpha=1}^3 (c_\alpha e^{\imath \theta_\alpha} \hat e_\alpha + d_\alpha e^{\imath \theta_{\alpha+3}} \hat e_\alpha^*)\end{aligned}$$
We rewrite these equations as $$\begin{aligned}
\delta x_i & = & \sum_{\alpha=1}^6 \mathring r_{i\alpha} e^{\imath \theta_\alpha} \\
\delta p_i & = & \sum_{\alpha=1}^6 \hat r_{i\alpha} e^{\imath \theta_\alpha}\end{aligned}$$ with $\mathring r_{i\alpha} = c_\alpha \mathring e_{\alpha i}$ if $\alpha \leq 3$, $\mathring r_{i\alpha} = d_{\alpha-3} \mathring e_{\alpha-3, i}^*$ if $\alpha > 3$. It follows that $$\delta \dot p_i = \imath \sum_{\alpha=1}^6 \hat r_{i\alpha} e^{\imath \theta_\alpha} \dot \theta_\alpha$$ Finally we write $\delta \dot p_i = \imath \hat R \dot \theta$ where the matrix $\hat R$ is defined by the elements $\hat r_{i\alpha} e^{\imath \theta_\alpha}$. We have then $\dot \theta = - \imath \hat R^{-1} \delta \dot p$. The dynamics onto $\mathbb T^6$ is then defined by the equation $$\dot \theta = {\Im\mathrm{m}}\hat R(\theta)^{-1} \hat F(\chi(\theta))$$ where $\chi : \mathbb T^6 \to \mathbb C^3$ is defined by $\chi(\theta) = \sum_{\alpha=1}^3 (c_\alpha e^{\imath \theta_\alpha} \mathring e_\alpha + d_\alpha e^{\imath \theta_{\alpha+3}} \mathring e_\alpha^*)$, and $\hat F:\mathbb C^3 \to \mathbb C^3$ is defined by $\hat F_i(\delta x) = - \tr |X-x|^2 \delta x_i + \tr \left((X^j-x^j)(X_i-x_i)\right)\delta x_j - 2 \overline{\delta x^j} \delta x_j \delta x_i + 2 \overline{\delta x_i} \delta x^j \delta x_j$.\
Finally we have $\dot \theta = F(\theta)$ with $$\label{eqF}
F = {\Im\mathrm{m}}\hat R^{-1} \hat F \circ \chi$$
Remark: note that at the linear approximation, $\theta_\alpha(t) \simeq \lambda_\alpha t$ and $\theta_{\alpha+3}(t) \simeq -\lambda_\alpha t$, it follows then $\theta_\alpha(t) \simeq -\theta_{\alpha+3}(t)$.
Dynamics of the fermionic string
================================
Chaotic behaviour
-----------------
We consider now the dynamics of the fermionic string governed by eq. \[equSchro\] with the Hamiltonian $H^{eff}=\sigma^i \otimes (X_i-x_i)$. The dynamics of the string is indirectly affected by the fluctuations included in the dynamics of $X$ and $x$. Following ref. [@Asano; @Gur; @Hanada], these dynamics are chaotic. We can verify this fact by studying the dominant asymptotic Lyapunov exponent $\underline \lambda$ of the dynamics of $X$, fig. \[lyap\].
![\[lyap\] $\frac{1}{t} \ln \frac{\|X(t)-X_0(t)\|}{\|X(0)-X_0(0)\|}$ (computed with the Sprott algorithm [@Gur]) with $X(t)$ and $X_0(t)$ the dynamics of the D2-brane respectively with and without fluctuations in the initial conditions, with $\|X\| = \sqrt{\|X_i\|_2 \|X^i\|_2}$ ($\|\cdot \|_2$ being the usual matrix norm). The system is a fuzzy sphere with the initial fluctuations choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=201$. The curve tends to the dominant Lyapunov exponent $\underline \lambda=0.75$.](LyapSphere201.eps){width="6.4cm"}
$\underline \lambda>0$ confirming its chaotic behaviour. The numerical simulations show that the value of $\underline \lambda$ is highly dependent from the model and from the initial fluctuations. The spin of the fermionic string is then driven by a classical chaotic system.\
The classical chaos in the dynamics of $X(t)$ is established, but we want to know if the quantum Hamiltonian $H^{eff}$ presents quantum chaos. A commonly used criterion of quantum chaos is the level spacing distribution (LSD) of the spectum [@Haake]. A regular system presents a LSD as Dirac peaks, a (pseudo)-random system presents a LSD as a Poisson distribution and a chaotic system presents a LSD as a Wigner-Dyson distribution. The LSD for $H^{eff}$ for a fuzzy sphere at a time after thermalization is represented fig. \[LSD\].
![\[LSD\] Level spacing distribution of $H^{eff} = \sigma^i \otimes (X_i-x_i)$ at a time after the thermalization for a fuzzy sphere with the initial fluctuations choosen following a gaussian law with $\sigma=0.01$, compared with the Poisson distribution (green) and with the Wigner-Dyson distribution (red). The number of D0-branes is $N=201$.](LSDsphere.eps){width="6.4cm"}
We see clearly that the LSD of $H^{eff}$ (after thermalization) follows a Wigner-Dyson distribution characteristic of the quantum chaos (we have the same results with the other models). It is interesting to note that the thermalization does not involve a pseudo-random behaviour but well a quantum chaotic behaviour.\
We have then both classical and quantum chaos: in the classical evolution of $X$ and in the quantum dynamics governed by $H^{eff}$.
Horizon of coherence of the reduced density matrix
--------------------------------------------------
Let $|\psi {\rangle \hspace{-0.2em} \rangle}$ be the state of the fermionic string solution of eq. \[equSchro\]. We consider the reduced density matrix: $$\rho = \tr_{\mathbb C^N} |\psi {\rangle \hspace{-0.2em} \rangle}{\langle \hspace{-0.2em} \langle}\psi|$$ corresponding to the mixed state of the spin of the string. This spin is subject to the entanglement with the string attachment degree of freedom, to the classical chaotic fluctuations of $X$ and $x$, and to the quantum chaos induced by $H^{eff}$ after the thermalization. In previous works [@Viennot2; @Aubourg; @Aubourg2], we have shown that a quantum system submitted to a chaotic flow presents an horizon of coherence, an initial duration where the quantum system is preserved from the decoherence phenomenon induced by the chaos. This horizon of coherence is a quantum version of the horizon of predictability of the classical chaotic flows (see discussion in [@Viennot2]). In order to study this point for the spin of the fermionic string, we consider the coherence $|\langle \uparrow|\overline \rho|\downarrow \rangle|$ of the averaged density matrix $\overline \rho$ (the average being realized on the initial conditions of the fluctuations) and its von Neumann entropy $S(\overline \rho) = -\tr(\overline \rho \ln \overline \rho)$. The results are compared with the coherence and the von Neumann entropy of the reduced density matrix $\rho_0$ for $X_0$ without fluctuation, in order to distinguish the effects due to the quantum entanglement (present in the two cases) from the effects of the chaos (present only with the fluctuations). Some results are drawn figure \[cohe\].
![\[cohe\] Coherence and von Neumann entropy of the spin mixed state of the fermionic string, for $X_0$ without fluctuation ($\rho_0$) and for $X$ with fluctuations ($\overline \rho$), with fuzzy sphere, torus and hyperboloid. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=11$ (for the fuzzy hyperboloid the cutoff is $N=11$). The vertical line indicates the horizon of coherence. The initial condition for the string is $|\psi(t=0) {\rangle \hspace{-0.2em} \rangle}= \frac{1}{\sqrt 2}(|\uparrow \rangle+|\downarrow \rangle) \otimes (1,0,...,0)$.](coheSphere.eps "fig:"){width="6.4cm"} ![\[cohe\] Coherence and von Neumann entropy of the spin mixed state of the fermionic string, for $X_0$ without fluctuation ($\rho_0$) and for $X$ with fluctuations ($\overline \rho$), with fuzzy sphere, torus and hyperboloid. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=11$ (for the fuzzy hyperboloid the cutoff is $N=11$). The vertical line indicates the horizon of coherence. The initial condition for the string is $|\psi(t=0) {\rangle \hspace{-0.2em} \rangle}= \frac{1}{\sqrt 2}(|\uparrow \rangle+|\downarrow \rangle) \otimes (1,0,...,0)$.](entropySphere.eps "fig:"){width="6.4cm"}\
![\[cohe\] Coherence and von Neumann entropy of the spin mixed state of the fermionic string, for $X_0$ without fluctuation ($\rho_0$) and for $X$ with fluctuations ($\overline \rho$), with fuzzy sphere, torus and hyperboloid. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=11$ (for the fuzzy hyperboloid the cutoff is $N=11$). The vertical line indicates the horizon of coherence. The initial condition for the string is $|\psi(t=0) {\rangle \hspace{-0.2em} \rangle}= \frac{1}{\sqrt 2}(|\uparrow \rangle+|\downarrow \rangle) \otimes (1,0,...,0)$.](coheTorus.eps "fig:"){width="6.4cm"} ![\[cohe\] Coherence and von Neumann entropy of the spin mixed state of the fermionic string, for $X_0$ without fluctuation ($\rho_0$) and for $X$ with fluctuations ($\overline \rho$), with fuzzy sphere, torus and hyperboloid. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=11$ (for the fuzzy hyperboloid the cutoff is $N=11$). The vertical line indicates the horizon of coherence. The initial condition for the string is $|\psi(t=0) {\rangle \hspace{-0.2em} \rangle}= \frac{1}{\sqrt 2}(|\uparrow \rangle+|\downarrow \rangle) \otimes (1,0,...,0)$.](entropyTorus.eps "fig:"){width="6.4cm"}\
![\[cohe\] Coherence and von Neumann entropy of the spin mixed state of the fermionic string, for $X_0$ without fluctuation ($\rho_0$) and for $X$ with fluctuations ($\overline \rho$), with fuzzy sphere, torus and hyperboloid. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=11$ (for the fuzzy hyperboloid the cutoff is $N=11$). The vertical line indicates the horizon of coherence. The initial condition for the string is $|\psi(t=0) {\rangle \hspace{-0.2em} \rangle}= \frac{1}{\sqrt 2}(|\uparrow \rangle+|\downarrow \rangle) \otimes (1,0,...,0)$.](coheHyperboloid.eps "fig:"){width="6.4cm"} ![\[cohe\] Coherence and von Neumann entropy of the spin mixed state of the fermionic string, for $X_0$ without fluctuation ($\rho_0$) and for $X$ with fluctuations ($\overline \rho$), with fuzzy sphere, torus and hyperboloid. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=11$ (for the fuzzy hyperboloid the cutoff is $N=11$). The vertical line indicates the horizon of coherence. The initial condition for the string is $|\psi(t=0) {\rangle \hspace{-0.2em} \rangle}= \frac{1}{\sqrt 2}(|\uparrow \rangle+|\downarrow \rangle) \otimes (1,0,...,0)$.](entropyHyperboloid.eps "fig:"){width="6.4cm"}
We see the existence of the horizon of coherence before the fall of the coherence and the increase of the entropy, in accordance with the chaotic behaviour. Note that we have the same thing for $\rho$ (without average on the initial fluctuations) but with strongly noisy curves. In contrast with the quantum systems studied in ref. [@Viennot2], we do not see any correlation between the horizon of coherence and the dominant asymptotic Lyapunov exponent. This fact is maybe due to the possibility that the decoherence is dominated by the quantum chaotically behaviour of $H^{eff}$ rather than by the classical chaotically behaviour of $X(t)$, even if the horizon of coherence is shorter than the thermalization duration. In fact, after a lot of numerical simulations with each model of D2-brane and for different values of $\sigma$ (initial dispersion of the fluctuations), we have noted that for the small values of $\sigma$, the horizon of coherence follows a law of the form $t_H = - \alpha \ln \sigma - \beta$ (with $\alpha,\beta >0$ depending on the D2-brane model), and for the large values of $\sigma$, it follows a law of the form $t_H = \frac{\gamma}{\sqrt \sigma} + \delta$ (with $\gamma>0$, $\delta \in \mathbb R$ depending on the D2-brane model). These laws are empirical, a more theoretical analysis needs a better understanding of the quantum chaos or of the intertwining between classical and quantum chaos, which is not the subjet of this paper.
Schrödinger-Koopman approach and emergent extradimensions
---------------------------------------------------------
The natural mathematical structure to study a quantum system driven by a chaotic flow is the Schrödinger-Koopman approach [@Viennot3]. It consists to enlarge the Hilbert space of the quantum system by $L^2(\Gamma,d\mu(q))$ the space of square functions on the stable phase space $\Gamma$ of the chaotic flow ($\mu$ is a probability measure on $\Gamma$ preserved by the flow). In this enlarged Hilbert space, the dynamics is governed by a Koopman Hamiltonian $H_K = -\imath F^\mu(q) \frac{\partial}{\partial q^\mu} \otimes {\mathrm{id}}+ H(q)$ where $H$ is the usual quantum Hamiltonian of the driven system and $F:\Gamma \to \Gamma$ is the generator of the flow. The wave function solution of the Schrödinger-Koopman equation $\Psi(t,q)$ and the wave function solution of the usual Schrödinger equation $\psi_{q_0}(t)$ are related by $\psi_{q_0}(t) = \Psi(t,q(t))$ with $t\mapsto q(t) \in \Gamma$ the phase trajectory of the chaotic system with $q(0)=q_0$. The interest of the approach is that the mixed state associated with the average onto the initial conditions, $\overline \rho(t) = \int_\Gamma |\psi_{q}(t) \rangle \langle \psi_q(t)| d\mu(q)$, is obtained as a reduced density matrix $\overline \rho(t)= \tr_{L^2(\Gamma,d\mu)} |\Psi(t){\rangle \hspace{-0.2em} \rangle}{\langle \hspace{-0.2em} \langle}\Psi(t)|$. The decoherence induced by the chaotic flow can be then interpreted as an entanglement between the quantum system and the classical chaotic flow.\
In this section, we want to show how it is possible to adapt the Schrödinger-Koopman approach to the fermionic string driven by the D2-brane. Because of section \[invtorus\] we know that the stable phase space for the chaotic flow can be chosen as being a 6-torus $\mathbb T^6$. The polarization part of the fluctuations is directly driven by the flow onto $\mathbb T^6$ via eq. \[pola\]. Let $X_i(t,\theta)$ be the D2-brane coordinates for fixed values of $\theta$ ($X(t) = X(t,\theta(t))$ with $\dot \theta = F(\theta)$, $F: \mathbb T^6 \to \mathbb T^6$ being defined eq. \[eqF\]). Since at the linear limit we have $\theta_\alpha(t) \simeq - \theta_{\alpha+3}(t)$ we have $X(t,..\theta_\alpha..\theta_{\alpha+3}..) \simeq X(t,..-\theta_{\alpha+3}..-\theta_\alpha..)$. Due to this exchange symmetry, we can write that the fermionic state is of the form: $$\psi(t) = \psi_{+}(t,\theta_1(t)..\theta_{4}(t)..) + \psi_{-}(t,\theta_{4}(t)..\theta_1(t)..)$$ It follows that $$\begin{aligned}
\imath \frac{d}{dt} |\psi(t) {\rangle \hspace{-0.2em} \rangle}& = & \imath \partial_t |\psi_+{\rangle \hspace{-0.2em} \rangle}+ \imath F^1 \partial_1 |\psi_+ {\rangle \hspace{-0.2em} \rangle}+ \imath F^4 \partial_4 |\psi_+ {\rangle \hspace{-0.2em} \rangle}\nonumber \\
& & + \imath \partial_t |\psi_-{\rangle \hspace{-0.2em} \rangle}+ \imath F^4 \partial_1 |\psi_- {\rangle \hspace{-0.2em} \rangle}+ \imath F^1 \partial_4 |\psi_- {\rangle \hspace{-0.2em} \rangle}\end{aligned}$$ The dynamics can be rewritten as the following Schrödinger-Koopman equation: $$\begin{aligned}
&& \imath \frac{\partial}{\partial t} \left(\begin{array}{c} |\psi_+{\rangle \hspace{-0.2em} \rangle \hspace{-0.2em} \rangle}\\ |\psi_-{\rangle \hspace{-0.2em} \rangle \hspace{-0.2em} \rangle}\end{array}\right) = \left(- \imath \left(\begin{array}{cc} F^1 & 0 \\ 0 & F^4 \end{array} \right) \partial_1 - \imath \left(\begin{array}{cc} 0 & F^1 \\ F^4 & 0 \end{array} \right) \partial_4 \right. \nonumber \\
& & \left. + \left(\begin{array}{cc} 0 & \sigma^i \otimes (X_i(t,\theta)-x_i(t,\theta)) \\ \sigma^i \otimes (X_i(t,\theta)-x_i(t,\theta)) \end{array} \right) \right) \nonumber \\
& & \qquad \qquad \qquad \qquad \times \left(\begin{array}{c} |\psi_+{\rangle \hspace{-0.2em} \rangle \hspace{-0.2em} \rangle}\\ |\psi_-{\rangle \hspace{-0.2em} \rangle \hspace{-0.2em} \rangle}\end{array}\right)\end{aligned}$$ with $|\psi_\pm(t,\theta_{1/4},\theta_2(t),\theta_3(t),\theta_{4/1},\theta_5(t),\theta_6(t)) {\rangle \hspace{-0.2em} \rangle \hspace{-0.2em} \rangle}\in \mathbb C^2 \otimes \mathbb C^N \otimes L^2(\mathbb T^2_{1,4},d\mu)$ (where $d\mu$ is the Haar measure on $\mathbb T^2_{1,4}$ the torus generated by $\theta_1$ and $\theta_4$).\
$|\psi(t) {\rangle \hspace{-0.2em} \rangle}= |\psi_+(t,\theta_1(t),\theta_2(t),\theta_3(t),\theta_4(t),\theta_5(t),\theta_6(t)) \rangle + |\psi_-(t,\theta_4(t),\theta_2(t),\theta_3(t),\theta_1(t),\theta_5(t),\theta_6(t)) \rangle$ is solution of the fermionic string Schrödinger equation.\
Let the deformed Dirac matrices be $$\begin{aligned}
\gamma^i & = & \left(\begin{array}{cc} 0 & \sigma_i \\ \sigma_i & 0 \end{array} \right) \\
\kappa^4 & = & \left(\begin{array}{cc} F^1 & 0 \\ 0 & F^4 \end{array} \right) \simeq F^1 \left(\begin{array}{cc} 1 & 0 \\ 0 & -1 \end{array} \right) \\
\kappa^7 & = & \left(\begin{array}{cc} 0 & F^1 \\ F^4 & 0 \end{array} \right) \simeq F^1 \left(\begin{array}{cc} 0 & 1 \\ -1 & 0 \end{array} \right)\end{aligned}$$ where “$\simeq$” denotes the linear approximation where $F^4=-F^1$ ($\dot \theta^4 \simeq - \dot \theta^1$). Let $\gamma^4 = \left(\begin{array}{cc} 1 & 0 \\ 0 & -1 \end{array} \right)$ and $\gamma^7 = \left(\begin{array}{cc} 0 & 1 \\ -1 & 0 \end{array} \right)$, it is easy to verify that $\{\gamma^A,\gamma^B\} = \pm 2 \delta^{AB}$ ($\forall A,B=1,...,4,7$). We can then consider that $\gamma^i$, $\kappa^4/F^1$ and $\kappa^7/F^1$ constitute a deformed Clifford algebra (the deformation being associated with the nonlinear part of the dynamics onto the torus).\
We can now reiterate this argumentation with the couple $(\theta_2,\theta_5)$ by writting that $\psi_\pm(t) = \psi_{\pm +}(t,..\theta_2(t)..\theta_5(t)..) + \psi_{\pm -}(t,..\theta_5(t)..\theta_2(t)..)$ and with $(\theta_3,\theta_6)$ by writting that $\psi_{\pm \pm}(t) = \psi_{\pm \pm +}(t,..\theta_3(t)..\theta_6(t)) + \psi_{\pm \pm -}(t,..\theta_6(t)..\theta_3(t))$.\
Finally the dynamics is described by the following Schrödinger-Koopman equation: $$\imath \frac{\partial}{\partial t} |\Psi(t) {\rangle \hspace{-0.2em} \rangle \hspace{-0.2em} \rangle}= \left(-\imath \kappa^{\alpha+3}(t,\theta) \frac{\partial}{\partial \theta^\alpha} + \gamma^i \otimes (X_i(t,\theta)-x_i(t,\theta)) \right) |\Psi(t) {\rangle \hspace{-0.2em} \rangle \hspace{-0.2em} \rangle}$$ where $\theta \in \mathbb T^6$, $|\Psi {\rangle \hspace{-0.2em} \rangle \hspace{-0.2em} \rangle}= \mathbb C^{16} \otimes \mathbb C^N \otimes L^2(\mathbb T^6,d\mu)$ (where $d\mu$ is the Haar measure on $\mathbb T^6$). The spinor is defined by $$\Psi = \left(\begin{array}{c} \psi_{+++} \\ \psi_{++-} \\ \psi_{+-+} \\ \psi_{+--} \\ \psi_{-++} \\ \psi_{-+-} \\ \psi_{--+} \\ \psi_{---} \end{array} \right)$$ with $\psi_{\pm\pm\pm} \in \mathbb C^2 \otimes \mathbb C^N \otimes L^2(\mathbb T^6,d\mu)$. The deformed Dirac matrix (generating a deformed Clifford algebra) are defined by $$\gamma^i = \left(\begin{array}{ccc} & & \sigma^i \\ & \rotatebox{90}{$\ddots$} & \\ \sigma^i & & \end{array} \right)$$ $$\kappa^4 = \left(\begin{array}{cccccccc} & & & & & & F^1 & \\ & & & & & & & F^4 \\ & & & & F^1 & & & \\ & & & & & F^4 & & \\ & & F^1 & & & & & \\ & & & F^4 & & & & \\ F^1 & & & & & & & \\ & F^4 & & & & & & \end{array} \right)$$ $$\kappa^7 = \left(\begin{array}{cccccccc} & & & & & & & F^1 \\ & & & & & & F^4 & \\ & & & & & F^1 & & \\ & & & & F^4 & & & \\ & & & F^1 & & & & \\ & & F^4 & & & & & \\ & F^1 & & & & & & \\ F^4 & & & & & & & \end{array} \right)$$ $$\kappa^5 = \left(\begin{array}{cccccccc} & & & & F^2 & & & \\ & & & & & F^2 & & \\ & & & & & & F^5 & \\ & & & & & & & F^5 \\ F^2 & & & & & & & \\ & F^2 & & & & & & \\ & & F^5 & & & & & \\ & & & F^5 & & & & \end{array} \right)$$ $$\kappa^8 = \left(\begin{array}{cccccccc} & & & & & & & F^2 \\ & & & & & & F^2 & \\ & & & & & F^5 & & \\ & & & & F^5 & & & \\ & & & F^2 & & & & \\ & & F^2 & & & & & \\ & F^5 & & & & & & \\ F^5 & & & & & & & \end{array} \right)$$ $$\kappa^6 = \left(\begin{array}{cccccccc} F^3 & & & & & & & \\ & F^3 & & & & & & \\ & & F^3 & & & & & \\ & & & F^3 & & & &
\\ & & & & F^6 & & & \\ & & & & & F^6 & & \\ & & & & & & F^6 & \\ & & & & & & & F^6 \end{array} \right)$$ $$\kappa^9 = \left(\begin{array}{cccccccc} & & & & & & & F^3 \\ & & & & & & F^3 & \\ & & & & & F^3 & & \\ & & & & F^3 & & & \\ & & & F^6 & & & & \\ & & F^6 & & & & & \\ & F^6 & & & & & & \\ F^6 & & & & & & & \end{array} \right)$$ We recover the solution of the usual Schrödinger equation by $|\psi(t) {\rangle \hspace{-0.2em} \rangle}= \sum_{i=1}^8 \langle \theta(t)|\Psi_i(t) {\rangle \hspace{-0.2em} \rangle \hspace{-0.2em} \rangle}$ where $\theta(t)$ is the trajectory on $\mathbb T^6$.\
We can rewrite the Schrödinger-Koopman equation as: $$\label{eqSKD}
\imath |\dot \Psi {\rangle \hspace{-0.2em} \rangle \hspace{-0.2em} \rangle}= (\gamma^{\alpha+3} \otimes Y_\alpha + \gamma^i \otimes(X_i-x_i)) |\Psi {\rangle \hspace{-0.2em} \rangle \hspace{-0.2em} \rangle}+ Z |\Psi {\rangle \hspace{-0.2em} \rangle \hspace{-0.2em} \rangle}$$ where $Y_\alpha = - \imath \Delta F_{\alpha {\ \mathrm{mod}\ }3} \partial_\alpha$ with $\Delta F_\alpha = \frac{F_\alpha - F_{\alpha+3}}{2}$, and where $Z = - \imath \Sigma^{\alpha {\ \mathrm{mod}\ }3} \beta^\alpha \partial_\alpha$ modelizes the interaction with the nonlinear fluctuation field, $\Sigma^\alpha = \frac{F^\alpha+F^{\alpha+3}}{2}$ and $\beta^1 = \sigma^1 \otimes \sigma^1 \otimes \sigma^0$, $\beta^2 = \sigma^1 \otimes \sigma^0 \otimes \sigma^0$, $\beta^3 = \sigma^0 \otimes \sigma^0 \otimes \sigma^0$ and $\beta^4=\beta^5=\beta^6=\sigma^1 \otimes \sigma^1 \otimes \sigma^1$.\
Equation \[eqSKD\] (without $Z$) is the equation for a fermionic string in a 9-dimensional space $\mathbb R^3 \times \mathbb T^6$ with a D2-brane wrapped in the six compactified dimensions, $\{X_i\}_{i=1,2,3}$ being its noncommutative coordinates in the non-compactified dimensions and $\{Y_\alpha\}_{\alpha = 1,...,6}$ being its noncommutative coordinates in the compactified dimensions (see [@Taylor; @Brace; @Konechny]). The compactified dimension radii are $\frac{F^\alpha(t,\theta)}{2\pi}$ (the geometry is a dynamical variable). It is interesting to note that with the Schrödinger-Koopman representation, the six supplementary compactified dimensions can be considered as emerging from the quantum fluctuations in 3-dimensional space without adding any assumption or consideration. Since the space-time geometry is revealed only by test particle (probe fermionic string), it is not possible to distinguish the 4-dimensional space-time with quantum fluctuation in the Schrödinger-Koopman picture from a 10-dimensional space-time in the Schrödinger picture having the same operator algebra {$X^i, Y^\alpha \}_{i=1,2,3; \alpha=1,...,6}$.\
Since the dimension radii are known we can compute the geometry of the extradimensions by the numerical simulations. Some examples are drawn fig. \[extradim\].
![\[extradim\] The compactified dimensions as a non-regular 6-torus of radii $F^\alpha(t,\theta)$ (for $t$ after the thermalization) by section $\theta^\alpha=0$ for four values of $\alpha$. The section 2-torus $T[\alpha,\beta]$ (with $\alpha$ and $\beta$ the indices of the non-fixed angles) is represented by the classical embedding of a torus into $\mathbb R^3$. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=11$. The evolution of the geometry is low after the thermalization.](T14Sphere.eps "fig:"){width="6.4cm"} ![\[extradim\] The compactified dimensions as a non-regular 6-torus of radii $F^\alpha(t,\theta)$ (for $t$ after the thermalization) by section $\theta^\alpha=0$ for four values of $\alpha$. The section 2-torus $T[\alpha,\beta]$ (with $\alpha$ and $\beta$ the indices of the non-fixed angles) is represented by the classical embedding of a torus into $\mathbb R^3$. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=11$. The evolution of the geometry is low after the thermalization.](T16Sphere.eps "fig:"){width="6.4cm"}\
![\[extradim\] The compactified dimensions as a non-regular 6-torus of radii $F^\alpha(t,\theta)$ (for $t$ after the thermalization) by section $\theta^\alpha=0$ for four values of $\alpha$. The section 2-torus $T[\alpha,\beta]$ (with $\alpha$ and $\beta$ the indices of the non-fixed angles) is represented by the classical embedding of a torus into $\mathbb R^3$. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=11$. The evolution of the geometry is low after the thermalization.](T23Sphere.eps "fig:"){width="6.4cm"} ![\[extradim\] The compactified dimensions as a non-regular 6-torus of radii $F^\alpha(t,\theta)$ (for $t$ after the thermalization) by section $\theta^\alpha=0$ for four values of $\alpha$. The section 2-torus $T[\alpha,\beta]$ (with $\alpha$ and $\beta$ the indices of the non-fixed angles) is represented by the classical embedding of a torus into $\mathbb R^3$. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=11$. The evolution of the geometry is low after the thermalization.](T12Torus.eps "fig:"){width="6.4cm"}\
![\[extradim\] The compactified dimensions as a non-regular 6-torus of radii $F^\alpha(t,\theta)$ (for $t$ after the thermalization) by section $\theta^\alpha=0$ for four values of $\alpha$. The section 2-torus $T[\alpha,\beta]$ (with $\alpha$ and $\beta$ the indices of the non-fixed angles) is represented by the classical embedding of a torus into $\mathbb R^3$. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=11$. The evolution of the geometry is low after the thermalization.](T31Torus.eps "fig:"){width="6.4cm"} ![\[extradim\] The compactified dimensions as a non-regular 6-torus of radii $F^\alpha(t,\theta)$ (for $t$ after the thermalization) by section $\theta^\alpha=0$ for four values of $\alpha$. The section 2-torus $T[\alpha,\beta]$ (with $\alpha$ and $\beta$ the indices of the non-fixed angles) is represented by the classical embedding of a torus into $\mathbb R^3$. The initial fluctuations are choosen following a gaussian law with $\sigma=0.01$. The number of D0-branes is $N=11$. The evolution of the geometry is low after the thermalization.](T32Torus.eps "fig:"){width="6.4cm"}
Conclusion
==========
The classical and quantum chaotic behaviours of the fluctuations in the D2-brane dynamics involves decoherence in the spin mixed state of a linked fermionic string. This system can be viewed as a model of qubit interacting with a quantum (micro) black hole (the noncommutative D2-brane representing its quantum horizon). Except the duration of the thermalization, the results seems weakly dependent on the number $N$ of D0-branes in the stack, but for a better discussion concerning (large) black holes, it needs to study the thermodynamic limit $N \to +\infty$. But this needs different numerical approaches that the ones used in this paper. The chaotic behaviour induces the existence of an horizon of coherence in the evolution of the spin mixed state (its values seems depend only on the D2-brane model and on the initial dispersion of the fluctuations). The application of the Schrödinger-Koopman approach to treat the effects of the fluctuations onto the fermionic string state, makes appear six compactified extradimensions in the modelization. A 4-dimensional space-time with quantum fluctuations in the Schrödinger picture is then equivalent to a 10-dimensional space-time in the Schrödinger-Koopman picture with fluctuations corresponding to the nonlinear part of the evolution ($Z$ operator in equation \[eqSKD\]). It would be interesting to generalize the present study to the other matrix theories (BMN and IKKT models [@Sochichiu; @Zarembo]).
The authors acknowledge support from I-SITE Bourgogne-Franche-Comté under grants from the I-QUINS project, and support from OSU THETA under grants from the SRO projects. Simulations have been executed on computers of the Utinam Institute supported by the Région de Bourgogne-Franche-Comté and the Institut des Sciences de l’Univers (INSU).
About the numerical simulations {#AppA}
===============================
For the CCR algebra, the numerical simulations need to restrict the description to finite dimensional Hilbert space. Let $N \in \mathbb N^*$ be the cutoff in description and $P_N = \sum_{n=0}^N |n\rangle \langle n|$ be the projector onto the subspace used in the simulations. Let $a_N = P_N a P_N$, $a_N^+ = P_N a^+ P_N$ and $1_N = P_N$ be the numerical representations of the generators of the CCR algebra. The problem is that these operators do not satisfy the definition of the CCR, since $$a_N a_N^+ - a_N^+ a_N = 1_{N-1} - N |N \rangle \langle N|$$ We see that the error is not small, it is equal to the cutoff value. If it is limited to the last state at this stage, during the propagation of the brane equation, it quickly contaminates all states. The solution is a redefinition of the commutator in the numerical representation: $$\forall A,B \in {\mathrm{Env}}(\mathfrak{ccr}), \quad [A_N,B_N]_N = R_N^{CCR} \circledast (A_N B_N - B_N A_N)$$ where $A_N = P_N A P_N$, $\circledast$ is the term to term multiplication : $(A \circledast B)_{ij} = A_{ij} B_{ij}$, and the renormalisation matrix being $$(R_N^{CCR})_{ij} = \left\{ \begin{array}{cl} - \frac{1}{N} & \text{if } i=j=N \\ 1 & \text{everywhere} \end{array} \right.$$ We have then $$[a_N,a_N^+]_N = 1_N$$
The same thing occurs for the $\mathfrak{su}(1,1)$ algebra: let $P_N = \sum_{m=0}^N |k,m\rangle \langle k,m|$ $$K_{N}^+ K_{N}^- - K_{N}^- K_{N}^+ = -2 K_{N-1}^3 + N(N+2k-1)|k,N\rangle \langle k,N|$$ and $K_{N}^3 K_{N}^\pm - K_{N}^\pm K_{N}^3 = \pm K_{N}^\pm$. We set then $$\begin{aligned}
& & \forall A,B \in {\mathrm{Env}}(\mathfrak{su}(1,1)) \nonumber \\
& & \qquad [A_N,B_N]_N = R_N^{SU(1,1)} \circledast (A_N B_N - B_N A_N)\end{aligned}$$ with the renormalisation matrix $$(R_N^{SU(1,1)})_{ij} = \left\{ \begin{array}{cl} - \frac{2(N+k)}{N(N+2k-1)} & \text{if } i=j=N \\ 1 & \text{everywhere} \end{array} \right.$$ We have then $$[K_N^+,K_N^-]_N = -2 K_N^3 \qquad [K_N^3,K_N^\pm]_N = \pm K_N^\pm$$
References {#references .unnumbered}
==========
[0]{} Viennot D and Moro O 2017 [*Class. Quant. Gravity*]{} [**34**]{}, 055005. Asplund C T, Berenstein D and Trancanelli D 2011 [*Phys. Rev. Lett.*]{} [**107**]{}, 17602. Berenstein D and Dzienkowski E 2012 [*Phys. Rev. D*]{} [**86**]{}, 086001. Aoki S, Hanada M and Iizuka N 2015 [*JHEP*]{} 2015, 029. Banks T, Fischler W, Shenker S H and Susskind L 1997 [*Phys. Rev. D*]{} [**55**]{}, 5112. Asano Y, kawai D and Yoshida K 2015 [*JHEP*]{} 2015, 191. Gur-Ari G, Hanada M and Shenker S H 2016 [*JHEP*]{} 2016, 091. Hanada M, Shimada H and Tezuka M 2017 [*preprint*]{} arXiv:1702.06935. Viennot D and Aubourg L 2013 [*Phys. Rev. E*]{} [**87**]{}, 062903. Aubourg L and Viennot D 2015 [*Quant. Inf. Process.*]{} [**14**]{}, 1117. Aubourg L and Viennot D 2016 [*J. Phys. B*]{} [**49**]{}, 115501. Viennot D and Aubourg L 2018 [*preprint*]{} arXiv:1802.08186 Sapin O, Jauslin H R and Weigert S 2007 [*J. Stat. Phys.*]{} [**127**]{}, 699. Jauslin H R and Sugny D 2010 in [*Mathematical horizons for quantum physics”*]{} (Singapore: World Scientific). Gay-Balmaz F and Tronci C 2018 [*preprint*]{} arXiv:1802.04787. Viennot D 2009 [*J. Phys. A*]{} [**42**]{} 395302. Sochichiu C 2006 [*Lect. Notes Phys.*]{} [**698**]{}, 189. Zarembo K L and Makeenko Y M 1998 [*Uspekhi Fizicheskikh Nauk*]{} [**168**]{}, 3. Hudoba de Badyn M, Karczmarek J L, Sabella-Garnier P and Huai-Che Yeh K 2015 [*JGEP*]{} 2015, 89. Sykora A 2016 [*preprint*]{} arXiv:1610.015041 Goldstein H, Poole C, and Safko J 2000 [*Classical Mechanics*]{} (New York: Addison Wesley). Haake F 1991 [*Quantum signature of chaos*]{} (Berlin: Springer-Verlag). Taylor W 1998 [*preprint*]{} arXiv:hep-th/9801182. Brace D, Morariu B and Zumino B 1999 [*Nucl. Phys. B*]{} [**545**]{}, 192. Konechny A and Schwarz A 2002 [*Phys. Rept.*]{} [**360**]{}, 353.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'A.R. Beresnyak'
- 'Ya.N. Istomin'
- 'V.I. Pariev'
date: 'Received ; accepted '
title: 'Relativistic parsec–scale jets: . Particle acceleration'
---
Introduction
============
The nature of the Active Galactic Nuclei (AGN) is still unclear. The most common point of view is that there is a supermassive black hole in the centre of the active galaxy with a mass approximately $10^8$ to $10^9 \, M_{\sun}$ (Rees, 1984). Accretion onto the black hole drives powerful turbulent processes in accretion discs which effectively heat the plasma and generate a magnetic field of the order of $10^4\,\mbox{G}$. Ejection of plasma from discs with frozen magnetic field and radiation from the disc lead to the formation of collimated streams which transfer energy to large distances, up to $10^5$ pc. In addition, the processes of generation of electron–positron plasma may occur in the magnetospheres of black holes as well as in the vicinity of accretion discs. Particles are created in collisions of high–energy photons produced by inverse Compton scattering of ultraviolet photons emitted from the disc with fast particles accelerated by magnetic reconnection within the disc or within the magnetosphere of the rotating black hole, where the equilibrium charge density changes its sign (Beskin et al. 1992; Hirotani & Okamoto, 1998). Electron–positron plasma forms pinched streams (jets) which have radii of the order of one parsec. The electromagnetic energy density in these jets greatly exceeds the total energy density of electrons and positrons, including the rest energy. The angular momentum from the black hole and the accretion disc is transmitted to the jet via magnetic coupling, where a strong radial electric field $E_r$ arises. Drift in this field and in the longitudinal magnetic field, $B_z$, frozen into plasma leads to differential rotation. The jet transfers an electric current producing an azimuthal magnetic field, $B_\phi$, and magnetic field lines twist into spirals. The energy in the jet is transferred by the Pointing vector which is proportional to the product of the radial electric field and the azimuthal magnetic field. Therefore, there is no problem of energy transfer along the jet to large distances compared to the case where the main energy is in the hydrodynamic motion.
The amazing facts about the jets from AGN are their high collimation (the ratio of the length to the diameter is a few orders of magnitude), the stability, and apparent superluminal velocities of distinct knots (Begelman et al. 1984). The hydrodynamic and MHD stability of jets was investigated in many studies (Blandford & Pringle, 1976; Torricelli–Ciamponi & Petrini, 1990; Appl & Camenzind, 1992). The stability of the electron–positron jets, rotating and moving with relativistic speed and surrounded by a dense medium was studied by Istomin & Pariev (1994) and Istomin & Pariev (1996). Such jets were shown to be stable with respect to axisymmetric as well as spiral perturbations. The physical reason for the stability is the shear of the magnetic field, which in the case of low hydrodynamic pressure stabilizes small oscillations. The density of the media surrounding the jet, $\rho$, should satisfy the condition $\rho
\gg B^2/(4\pi c^2)$ for the inertia of surrounding gas to stabilize the relativistic jet. On parsec scales, a typical value of $B \sim 10^{-2}\,\mbox{G}$ gives a proton density $n \gg
0.05\,\mbox{cm}^{-3}$. The characteristic density of gas in galactic nuclei is $n \sim 10\,\mbox{cm}^{-3}$. Therefore, the approximation of stationary jet walls (Istomin & Pariev, 1994; 1996) is well justified. A relativistic force–free jet with infinite radial extent or with vacuum outside can develop weak instabilities (Lyubarskii, 1999; Li, 2000). In a stable jet, perturbations do not increase with time ($\mbox{Im}\,\omega\equiv0$) or have a small decay rate ($\mbox{Im}\,\omega\approx10^{-2} \mbox{Re}\,\omega$) because of the resonance with Alfvén waves $\omega^\prime=k_\|^\prime c$. When the specific energy density and pressure of the plasma are much less than the energy density of the magnetic field, the Alfvén velocity is equal to the speed of light in vacuum $c$ ($\omega^\prime$ and $k^\prime$ are the frequency and the wave vector in the plasma rest frame). The resonance condition is fulfilled on the specific magnetic surface and, in the case of cylindrical jet, at a definite distance from the axis. In the vicinity of that surface the magnetic and electric fields of the wave are large. Particles in the jet are accelerated by the electric field there, absorbing the energy of the perturbation. Thus, the stability of the jet is directly related to the production of energetic particles in the jet. This resolves the problem of [*in situ*]{} acceleration of electrons and positrons producing synchrotron emission from knots observed along jets. In fact, as is known (see Begelman at al., 1984), the energetic particles accelerated in the central region cannot survive far along the jet; they must lose their transverse momentum by synchrotron radiation already at the basement of the jet. The time for an electron to lose its perpendicular momentum due to synchrotron radiation is given by $$\tau_{\rm s}=\frac34\frac{c\gamma_\|}{r_{\rm e}\omega_{B0}^2}.$$ Here $\gamma_\|$ is the longitudinal Lorentz–factor of the particle, $\gamma_\|=(1-v_\|^2/c^2)^{-1/2}$; $r_{\rm e}$ is the classical radius of the electron $r_{\rm e}=e^2/m_{\rm e} c^2$; $\omega_{B0}=eB/m_{\rm e} c$ is the electron cyclotron frequency. Given $\gamma_\|=10$, $B=10^2\,\mbox{G}$, $\tau_{\rm s}$ is approximately $3\cdot10^5\, \mbox{sec}$, whereas the typical time of the flyby through the region of strong fields $\tau_{\rm
f}\approx \ell/c\approx 10^6\, \mbox{sec}$. Here $\ell$ is the diameter of the jet, $\ell\approx 3\cdot10^{16}\,\mbox{cm}$. Qualitatively the possibility of particle acceleration by electromagnetic disturbances propagating along a jet was proposed by Bisnovatyi–Kogan & Lovelace (1995). However, the authors of this work needed an ad hoc vacuum gap between conducting jet and surrounding medium, where the acceleration is possible due to the presence of the component of electric field parallel to the magnetic field.
In this paper we assume that the energy transferred along the jet is mainly the electromagnetic energy which can be transmitted for long distances along the jets as if along wires. In this case it is very possible that part of that energy is in the waves propagating along the jet, which are eigenmodes of a cylindrical beam. The source of the wave motion consists of non-stationary processes in the magnetospheres of the black hole and the accretion disc. Short time variability on scales from days to months is actually observed in AGN (Mushotzky et al. 1993; Witzel et al., 1993). Such a picture also gives a natural explanation of the superluminal motions but requires neither special orientation of the jet nor a relativistic jet with high speed ($\beta=u/c>\sqrt2/2$). As was shown in papers by Istomin & Pariev (1994) and Istomin & Pariev (1996) there exist so called standing eigenmodes in jets ($v_{\rm group}\equiv0$) which do not propagate with finite velocity along the jet but are only subject to diffuse spreading and, therefore, their amplitude is maximal. The phase velocity of these modes is greater than $c$. Wave crests move along the jet with superluminal velocity causing the acceleration of the particles on the Alfvén surface. These regions might be the observed bright knots with typical sizes of the order of the wavelength of the standing wave, which is about the radius of the jet.
There is some observational evidence in the case of 3C273, M87, and 3C345 that their jets are electron-positron rather than electron-proton (Morrison & Sadun, 1992; Reynolds et al., 1996; Hirotani et al., 2000). The recent observation of circular polarization of radio emission from 3C279, 3C273, 3C84 and PKS0528+134 (Wardle et al., 1998 ) supports the same conclusion. However, there are also indications that jets in Optically Violently Variable quasars cannot consist solely of e$^{+}$e$^{-}$ pairs because they would produce much larger soft X-ray luminosities than are observed. On the other hand, models with jets consisting solely of proton-electron plasma are excluded, since they predict much weaker nonthermal X-radiation than observed in Optically Violently Variable quasars (Sikora & Madejski, 2000).
We consider the acceleration processes near the Alfvén resonant surface and calculate the spectrum of accelerated particles. In an accompanying paper (Pariev et al. 2002), we calculate the polarization of synchrotron radiation generated by these particles.
Acceleration of particles
=========================
The equations of motion near the Alfvén surface are rather complicated. Nevertheless, we can use drift equations (Sivukhin, 1965) because the Larmor radius of relativistic electrons and positrons $r_{\rm c}= {\cal E}/eB\approx10^6\mbox{cm}$ (if we take $B\approx 10^{-2}$ and $\gamma\approx 10$) is much smaller than radius of the jet and the width of the resonant surface $r_0$ (see expression (28)): $$\frac{d{\bf r}}{dt}=v_\|\frac{{\bf B}}B+\frac c{B^2}[{\bf E\times B}]+
\frac{\e v_\|^2}{ecB^4}[{\bf B\times(B\nabla)B}]+
\frac{\e v_\perp^2}{2ecB^3}[{\bf B\times\nabla}B];$$ $$\left(\frac{d\e}{dt}\right)_{\rm e.m.}=e{\bf E}\frac{d{\bf r}}{dt}+
\frac{\e v_\perp^2}{2c^2B}\frac{\partial B}{\partial t}.\eqno (1)$$ Here the electric field, ${\bf E}$, is assumed to be small compared with the magnetic field, ${\bf B}$, and $\e$ is the energy of the particle. The index $\mbox{e.m.}$ means the energy change due to the action of the electromagnetic fields, ${\bf E}$ and ${\bf B}$ on the particle. We did not write the third equation here, the conservation of the adiabatic invariant $J_{\perp}=p_{\perp}^2/B$, which is usually used as one of the drift equations. The adiabatic invariant is not conserved because of synchrotron losses. But the characteristic time of the change of the adiabatic invariant is much larger than the period of the cyclotron rotation and we can use the drift approximation of the motion of the particle as a result of the separation of the motion into fast cyclotron rotation and slow motion of the guiding centre with subsequent averaging over the fast rotation. The conditions for using the drift approximation are a small change of the cyclotron frequency $\omega_B$ during cyclotron period $2\pi/\omega_B$ and also a small change of the particle cyclotron radius on the scale of the cyclotron radius (Alfvén & Fälthammar, 1963): $$\frac{2\pi}{\omega_B}\frac{\partial \omega_B}{\partial t} \ll
\omega_B; \quad r_B |\nabla r_B| \ll r_B. \eqno (1a)$$ A rigorous derivation of equations (1) under conditions (1a) was given by Sivukhin (1965) and Morozov & Solov’ev (1966). Let us demonstrate that the second equation (1) does not require the conservation of the adiabatic invariant $J_{\perp}$. The particle velocity is split into two parts ${\bf v} = d{\bf r}/dt + {\bf
v}_{\rm c}$. Here $d{\bf r}/dt$ is the velocity of the guiding centre and ${\bf v}_{\rm c}$ is the velocity of cyclotron rotation, $v_{\rm c} = v_\perp$. $$\left(\frac{d\e}{dt}\right)_{\rm e.m.} = e{\bf v}{\bf E} =
e\frac{d{\bf r}}{dt}{\bf E} + e{\bf v}_{\rm c}{\bf E} =
e\frac{d{\bf r}}{dt}{\bf E} + \frac{e\omega_B}{2\pi}\oint{\bf E}d{\bf l} = \\
e\frac{d{\bf r}}{dt}{\bf E} + \frac{e\omega_B}{2\pi}\int \>
\mbox{curl}\, {\bf E}\,d{\bf s}.$$ Using the Maxwell equation, $\mbox{curl}\, {\bf E} =
-(\partial{\bf B}/\partial t)/c$, and the relation ${\bf
s}=-\pi(v_\perp/\omega_B)^2{\bf B}/B$ we obtain the second equation of the drift approximation (1). The first equation of system (1) is the equation of the motion of the guiding centre $d{\bf r}/dt$ and contains the motion along the magnetic field line, the electric drift, gradient and centrifugal drifts. All drifts and the equation of energy change are independent of the conservation of the adiabatic invariant $J_{\perp}$. The value of $J_{\perp}$ is conserved in the slowly varying electromagnetic fields ${\bf B}({\bf r},t)$, ${\bf E}({\bf r},t)$, but is not conserved under the action of additional forces including radiative forces. The small change of the adiabatic invariant during the period of the cyclotron rotation does not violate the condition of applicability of the drift approximation (1a).
The electromagnetic fields are equal to the sum of the stationary fields (without subscript) and the wave fields (subscript 1). For a cylindrical jet the stationary configuration of fields is (Istomin & Pariev, 1994) (c=1): $${\bf B}=B_z{\bf e}_z+B_\phi{\bf e}_\phi;\quad B_\phi=\omr B_z;$$ $${\bf u}=K{\bf B}+\omr{\bf e}_\phi;\eqno (2)$$ $${\bf E}=-\omr[{\bf e}_\phi{\times\bf B}],$$ where $r,\phi,z$ are cylindrical coordinates. In components $${\bf B}=(0,\omr,1)B_z,$$ $${\bf E}=(-\omr,0,0)B_z.$$ Here $\Omega^{\rm F}$ and $K$ are functions of $r$, and $B_z$ does not depend on $r$. The stationary electric field, which has an absolute value of $\omr B_z$, is not small compared to the magnetic field when $\omr\sim 1$; therefore we must consider our equations in the frame moving with the plasma where ${\bf E}\equiv 0$. There are many such reference frames because there is an arbitrary parameter $K(r)$ in the expression for the plasma velocity which determines the radial profile of longitudinal velocity. $K(r)B_z$ is the velocity along the magnetic field which is not related to the rotation. We choose, however, the velocity ${\bf u}$ which minimizes the kinetic energy of the plasma in the stationary reference frame. $${\bf u}=\frac{(0,\omr,-(\omr)^2)}{1+(\omr)^2}.\eqno (3)$$ It is probable that the plasma moves with that velocity in real jets (Frolov & Novikov 1998). THE fields in the wave calculated in terms of the Lagrangian radial displacement $\xi$ are (Istomin & Pariev 1996) (here we drop out the phase coefficient $\exp
(-i\omega t+ikz+im\phi)$): $$B_{r1}=iB_zF\xi;$$ $$B_{\phi 1}=-B_z(\omr\frac{d\xi}{dr}+\xi\frac d{dr}(\omr)+\frac k S D);$$ $$B_{z1}=B_z(-\frac{d\xi}{dr}-\frac\xi r+\frac m{rS} D);\eqno (4)$$ $$E_{r1}=B_z(\omr\frac{d\xi}{dr}+\xi\frac d{dr}(\omr)-\frac\omega S D);$$ $$E_{\phi 1}=-iB_z(\omega-m\Omega^{\rm F})\xi;$$ $$E_{z1}=iB_z\omr(\omega+k)\xi;$$ where $$D=r\frac{d\xi}{dr}\left(\Omega^{\rm F}(\omega+k)-\frac m{r^2}\right)-
\xi\left(\Omega^{\rm F}(\omega+k)+\frac m{r^2}\right),$$ $$F=k+m\Omega^{\rm F},$$ $$S=\omega^2-k^2-m^2/r^2\mbox{.}$$ For Lagrangian displacement $\xi$ and dimensionless pressure disturbance $p_*=4\pi P_1/B_z^2$ we have a system of differential equations obtained by Istomin & Pariev (1996): \#1\#2 $$\left\{\begin{array}{rcl}
A\fracb1r\fracb d{dr}(r\xi)&=&C_1\xi-C_2p_*,\\
A\fracb{dp_*}{dr}&=&C_3\xi-C_1p_*.\\
\end{array} \right.\eqno (5)$$ Here $$C_1=\frac2r\left(m\Omega^{\rm F}-(\omr)^2(\omega+k)\right);$$ $$C_2=-\frac{\omega^2-k^2-m^2/r^2}{\omega+k};$$ $$C_3=-(\omega+k)(A^2-4{\Omega^{\rm F}}^2);$$ $$A=k-\omega+2m\Omega^{\rm F}-(\omr)^2(\omega+k).$$ Let us expand Eqs. (5) near the Alfvén surface, i.e. near the point $r_{\rm A}$ where $A(r_{\rm A})=0$. Denoting $x=r-r_{\rm
A}$ we get $$\left\{
\begin{array}{l}
\fracb{d\xi}{dx}=\fracb{1}{A'x}(C_1\xi-C_2p_*);\\
\fracb{dp_*}{dx}=\fracb{1}{A'x}(C_3\xi-C_1p_*),
\end{array}
\right.\eqno (6)$$ where $\left.A'=\fracb{dA}{dr}\right|_{r=r_{\rm A}} $. The general solution of these equations is $$\left(\xi\atop p_*\right)=\alpha_1\left(C_2\ln x\atop C_1\ln x-A'\right)+
\alpha_2\left( C_2 \atop C_1\right), \eqno (7)$$ where $\alpha_1$ and $\alpha_2$ are constants. They are not arbitrary; rather they are fixed by the boundary conditions on $\xi$ and $p_*$ obtained from solving general equation (5). We see that $\xi$ and $p_*$ have a logarithmic singularity at the Alfvén point. Later we adhere to our expansion over $r-r_{\rm A}$, assuming $r=r_{\rm
A}$, $A=0$ whenever the quantities under consideration have no singularities at $r=r_{\rm A}$. Let us calculate $d{\bf r}/dt$ and $(d\e/dt)_{\rm e.m.}$ to first order in the fields of the wave. Remembering that ${\bf E}=0$ in our reference frame and knowing that terms with the energy $\e$ contain a small parameter, either the ratio of the Larmor radius to the jet radius or to the characteristic length of variations of zero order fields, we can simplify the drift equations as follows, $$\left(\frac{{d\bf r}}{dt}\right)_0=v_\|{\bf e}_z,$$ $$\left(\frac{{d\bf r}}{dt}\right)_1=v_\|\frac{{\bf B}_1}{B}-
v_\|\frac{{\bf B(B\cdot B}_1)}{B^3}+
\frac{[{\bf E}_1\times{\bf B}]}{B^2}\mbox{.} \eqno (8)$$ The subscript 0 denotes the zero-order term and 1 denotes the first order term. The force of inertia is also present in the expression of ${d\bf r}/{dt}$ (see Sivukhin, 1965). The term with inertia force contains the electron mass and has the same order as terms with energy and is omitted. It is clear that in the force-free approximation the impact of the inertia force to the trajectory is negligible. When we substitute ${\bf B}_1$ and ${\bf
E}_1$ into Eq. (8) we get $$\left(\frac{{dr}}{dt}\right)_1=(v_\|-1)\frac{i\xi F}{\gamma}, \eqno(8*)$$ where $\gamma=\sqrt{1+(\omr)^2}$ is the Lorentz factor corresponding to the velocity of the moving frame, $\xi$ depends on $z,\phi,t$ by $\exp\{-i\omega t+ikz+im\phi\}$. Using the Lorentz transformation, the phase of the wave $\Psi$ ”seen” by the particle is found to be $\displaystyle \Psi=-\frac F\gamma
t'+\frac F\gamma z'$. Assuming $ dz/dt=v_\| $ we get $\displaystyle \Psi=(v_\|-1)\frac F\gamma t'$. As we see, the coefficient for the phase coincides with the coefficient in Eq. (8\*). This confirms our calculations because $\xi$ is the Lagrangian displacement. The question of the trajectory of the particle will be considered later.
Now we proceed with calculating $({d\e}/{dt})_{\rm e.m.}$: $$\left(\frac{d\e}{dt}\right)_{\rm e.m.}=
e{\bf E}_1\cdot\left(\frac{{d\bf r}}{dt}\right)_0+ \frac{\e
v_\perp^2}{2B}\frac{\partial B_1}{\partial t}.\eqno (9)$$ After substitution of expressions for fields into (9) we get $$\left(\frac{d\e}{dt}\right)_{\rm e.m.}=\e v_\perp^2
\left(-\frac{iF}\gamma\right)\frac\xi{rS}(\omega+k)[F-\omega],\eqno
(10)$$ or, combining with (8\*) $$\left(\frac{d\e}{dr}\right)_{\rm e.m.}=
\frac{\e
v_\perp^2}{1-v_\|}\left(\frac{(\omega+k)[F-\omega]}{rS}\right).\eqno
(11)$$ The inertia force gives the term without the coefficient $(\omega+k)[F-\omega]/S$. This coefficient is large if the fast magnetosonic resonance surface $S=0$ happens to be close to the Alfvén resonance $r_{\rm A}$. Our calculations (Istomin & Pariev 1996) show that $S$ is indeed small at $r=r_{\rm A}$ when the damping rate of the eigenmodes is maximal. The maximal damping rate is still small compared to the frequency $\omega$. Rapid damping of energy in these modes leads to a high efficiency of particle acceleration, which depends, as we will see later, on the smallness of $S$ ($Q\propto 1/S$ in Eq. (23) must be large in order to get high particle acceleration). The inertia force is of the order of $\e/r$, which may not give large acceleration because $\Delta r<<r_{\rm A}$. As we see, $\e$ is proportional to $r$ if we do not take into account the dependence of the right hand side on $\e$. (Note again that we consider all values at the Alfvén resonance, $r=r_{\rm A}$.)
Therefore it makes sense to consider the motion of the particle averaged over many periods. Taking the $r$-component of (8\*), we have $$\frac{dx}{dt}=i\omega^*\xi(x)e^{i\omega^*t},\quad \omega^*=(v_\|-1)\frac F\gamma. \eqno (12)$$ Near the Alfvén resonance $\xi(x)$ has a logarithmic singularity (Eq. (7)); therefore, Eq. (12) can be written as $$\frac{dx}{dt}=\omega^*[A_1\ln x\sin(\omega^*t)+B_1\cos(\omega^*t)], \eqno(13)$$ where $\omega^*A_1$ and $\omega^*B_1$ are the dimensionless amplitudes of the perturbation \[ampl\](remember that $c=1$). Let us average Eq. (13), expanding it in small amplitude of oscillations of a particle in the field of the wave. We denote quantities of zero, first, second order as $x_0$, $x_1$, $x_2$. $$\frac{dx_1}{dt}=\omega^*[A_1\ln x_0\sin(\omega^*t)+B_1\cos(\omega^*t)],$$ $$x_1=-A_1\ln x_0\cos(\omega^*t)+B_1\sin(\omega^*t),$$ $$\frac{dx_2}{dt}=\omega^*[A_1\ln|x_0-A\ln x_0\cos(\omega^*t)+B_1
\sin(\omega^*t)|\sin(\omega^*t)+B_1\cos(\omega^*t)],$$ $$\overline{\frac{dx_2}{dt}}=\omega^*A_1B_1\frac 1{2x_0}.$$ Now let us replace $x_0$ by ${\bar x}$ and $\displaystyle
\overline{\frac{dx_2}{dt}}$ by $\displaystyle \frac{d{\bar
x}}{dt}$ to obtain a smoothed equation of motion $$\frac{d\bar x}{dt}=\omega^*\frac {A_1 B_1}{2\bar x},$$ which has the solution $$\bar x=\sqrt{\omega^*A_1 B_1 t}. \eqno(14)$$ The result for the smoothed particle motion (14) was checked by the direct numerical integration of Eq. (13).
From Eqs. (11), (14) it follows that the particle gains energy drifting along the $r$ axis. The rate of acceleration is proportional to the transverse momentum. Because of synchrotron losses, $p_\perp$ and $J_{\perp}$ are monotonically decreasing according to ($\displaystyle\omega_B=\omega_{B0} m_{\rm e} c^2/\e$) $$\frac{dJ_\perp}{dt}=-\frac 43\omega_B^2\frac {r_{\rm e}}c\left(
\frac\e{m_{\rm e} c^2}\right)^3
\left(1-\frac{v_\|^2}{c^2}\right)J_\perp, \eqno (15)$$ this leads to $p_\|>p_\perp$. Such an anisotropic distribution can become unstable to the excitation of electromagnetic waves which results in isotropization of the distribution function in momentum space. The synchrotron losses and anisotropic instabilities lead to the violation of the adiabatic invariant conservation $J_\perp = \mbox{constant}$.
First of all we consider instabilities in a magnetically dominated plasma with parameter $\beta=4\pi nT/B^2<<1$ where T is the mean plasma energy, because the jet remains force–free when particles gain small energies in the process of acceleration. Parail & Pogutse (1986) show that instabilities may arise from the anomalous Doppler resonance of the waves with the accelerated tail of the distribution function. The expression they find for the increment of instability is $$\gamma_{\bf k}=\frac\pi 2\omk\frac{\omega_{\rm p}^2}{k^2}m_{\rm e}^2\int d{\bf p}\,
\delta(\omk-k_{\parallel}v_{\parallel})\frac{\partial f}{\partial
p_{\parallel}}k_{\parallel}+
\frac\pi 8\omk\frac{\omega_{\rm p}^2}{k^2}\times$$ $$\int d{\bf p}\,
\delta(\omk+\omega_{B}-k_{\parallel}v_{\parallel})
\frac{k_\perp^2p_\perp^2}{\omega_{B0}^2}
\left(k_{\parallel}\frac{\partial f}{\partial p_{\parallel}}-
\frac{m_{\rm e}\omega_{B0}^2}{p_\perp}
\frac{\partial f}{\partial p_\perp}\right). \eqno (16)$$ Here $\omega_{B}=\omega_{B0}(1-v^2/c^2)^{-1/2}$ and $f=f(p_{\parallel}, p_{\perp})$ is the distribution function of particles in drift variables $p_{\parallel}$, $p_{\perp}$. The first term corresponds to the Cherenkov resonance and the second corresponds to the anomalous Doppler resonance. This expression was written with the assumption that $k_\perp p_\perp/m_{\rm
e}\omega_{B0}<1$. Parail & Pogutse (1986) demonstrate the instability for the distribution function of electrons accelerated in an electric field and call it the “fan” instability. This instability is also present for all distribution functions with small $p_\perp$ and large $p_{\parallel}$. We consider a particular distribution function, Gaussian in $p_\perp$ and power-law with AN index of $\alpha$ in $p_{\parallel}$ with the minimal longitudinal momentum $p_{\rm min}$. The resulting $\gamma_{\bf k}$ is expressed as: $$\gamma_{\bf k}=\frac\pi
2\frac{\omk}{\omega_{B0}}\frac{\omega_{\rm p}^2}{k^2}
\frac{\alpha-1}{p_{\rm min}}m_{\rm e} \left(\frac{p_{\rm
min}k_{\parallel}}{\omega_{B0}m_{\rm e}}\right)^\alpha
k_{\parallel}\left[-\alpha\left(\frac{\omega_{B0}}{\omk}\right)^{\alpha+1}
\gamma_1^{2-\alpha}+
\frac12\frac{k_\perp^2}{k_{\parallel}^2}\gamma_2^3\right], \eqno
(17)$$ where $\gamma_1$, $\gamma_2$ are the $\gamma$-factors of the particles in Cherenkov resonance and anomalous Doppler resonance correspondingly. If $\gamma_1\approx\gamma_2\approx1$ (the particles in resonance are non-relativistic) we see that instability arises when $\displaystyle
2\alpha\left(\frac{\omega_{B0}}{\omk}\right)^{\alpha+1}<
\frac{k_\perp^2}{k_{\parallel}^2}<\frac{m_{\rm e}^2\omega_{B0}^2}
{k_{\parallel}^2p_\perp^2}$, or $\displaystyle
p_\perp^2<\frac{m_{\rm e}^2}{2\alpha}\frac{\omega_{B0}^2}
{k_{\parallel}^2}
\left(\frac{\omega_{B0}}{\omk}\right)^{\alpha+1}$ i.e. if $p_\perp$ is small enough. Quasilinear analysis of the influence of the excited waves on the distribution function shows (Parail & Pogutse, 1986) that the developing instability leads to the isotropization of the distribution function. This may easily be seen from the resonance condition $\hbar\omk=n\hbar\omega_{B}+\hbar
k_{\parallel}v_{\parallel}$ ($n$ is an integer), which may be treated as the energy conservation law for one interaction of a particle with the wave. Here $\hbar k_{\parallel} v_{\parallel}
\approx-[(m_{\rm e} v_{\parallel}-\hbar k_{\parallel})^2- m_{\rm
e} ^2v_{\parallel}^2]/(2m_{\rm e})$ is the decrement of the longitudinal energy, which is positive in the case of anomalous Doppler resonance ($n<0$). So the longitudinal energy decreases, and the transverse energy increases ($n<0$). In the case of $\omk<<\omega_{B}$ the energy transmitted to the waves is small. Due to this instability the distribution function of electrons becomes isotropic with deviations from isotropy of the order of $\displaystyle \simeq \frac{\omega_{B0}}{\omega_{\rm p}^2\tau}$, where $\tau$ is the smallest characteristic time of non-stationary processes increasing the anisotropy of the distribution function, and the plasma frequency is $\omega_{\rm p}^2= 4\pi n e^2/m_{\rm
e}$. Given typical values of $B_0 \approx 10^{-2}\,\mbox{G}$, $n\approx 0.1\,\mbox{cm}^{-3}$ we have $\omega_{B0}/ \omega_{\rm
p}^2\approx 10^{-3}\,\mbox{sec}$. This time is many orders of magnitude smaller than any possible $\tau$ either due to synchrotron losses $\tau_{\rm s}$ or $\approx 1/\omega$, or $\approx 1/\Omega^{\rm F}$. Therefore, the distribution of accelerated particles must be highly isotropic in momentum space.
Although the case with $\beta>1$ (strong accelerated particles) is apparently contradictory with our assumption of a force-free jet, the process of acceleration takes place in the close vicinity of the Alfvén resonance and does not affect the overall structure of the fields in the whole jet. As we show later the process of acceleration leads to the power-law spectrum of particles with high characteristic energy (see Eq.(22)) and the minimum energy a few orders of magnitude lower. So the case with $\beta>1$ can occur.
In the case of $\beta>1$ there exist many types of instabilities, some of which are hydrodynamic. We will treat plasma instabilities according to Mikhaylovskii (1975), who showed that in the case $p_\|>p_\perp$, perturbations with $k_{\parallel}=0$ have maximal growth rates. The frequencies of these perturbations with ${\bf k}\perp{\bf B}$ are the solutions of the dispersion relation $\oureps-N^2=0$, where $\oureps$ is the component of the dielectric tensor along the magnetic field, $N=ck/\omega$. The dispersion relation of the small oscillations has the form \#1\#2 $$\frac{c^2k^2}{4\pi e^2}=\frac{\om^2}{4\pi e^2}+
\int\,d{\bf p}\left\{ v_{\parallel}
\left[\frac{\pa f}{\pa p_{\parallel}}-\left(1-J_0^2
(kv_{\perp}/\omega_B)\right)
\frac{v_{\parallel}}{v_\perp}\frac{\pa f}{\pa p_\perp}\right]\right.$$ $$\left.+2\om^2\sum_{n=1}^\infty\frac{\frac{v_{\parallel}^2}{v_\perp}
\frac{\pa f}{\pa p_\perp}}{\om^2-n^2\om_B^2}J_n^2
(kv_{\perp}/\omega_B)\right\}.\eqno (18)$$ The anisotropic instabilities result in the excitation of potential waves which change the direction of the particle momentum $\bf{p}$ but do not change the energy $\e$ of the particle. Indeed, momentum conservation in the process of emitting or absorbing a wave reads $$\delta{\bf p} = \hbar{\bf k} =
\hbar\omega\left(\frac{N}{c}\frac{{\bf k}}{k}\right),$$ where $N$ is the wave refractive index. For the potential waves, $N>>1$. If $\delta{\bf p}$ is not orthogonal to ${\bf p}$, then $\delta\e \approx|\delta{\bf p}|\cdot c = \hbar\omega\cdot N
>>\hbar\omega$. This is in contradiction to the energy conservation $\delta\e = \hbar\omega$. So, $\delta{\bf p}\perp{\bf
p}$ with the accuracy $1/N << 1$.
We see that the effect of anisotropic instability is to change the particle distribution function in the direction of momentum, and not in energy. Because the instability is rapid, the quasi-stationary distribution function is determined from the condition $\gamma_{\bf k} = 0$. This condition is the equation which should be used instead of Eq. (15) describing the change of $J_\perp$ only due to synchrotron losses. The condition $\gamma_{\bf k} = 0$ is the integral equation for the particle distribution function $f(p_\parallel , p_\perp)$, which follows from Eq. (18). It is difficult to solve this integral equation without knowledge about the shape of $f(p_{\parallel},p_{\perp})$. We estimate the integral in the dispersion relation (18) assuming the cold distribution function $f$, although in reality it is smooth in momentum space. Also, we assume that $k v_\perp$ is of the order of $\omb$ and estimate expression (18) taking into account only the first terms of The expansion in Bessel functions over its arguments. Then, the threshold of stability ($\gamma_{\bf k}=0$) when particles have equal $p_\perp$ and $p_\|$ in the ultrarelativistic case is found to be $$\frac{p_{\parallel}^2}{p_\perp^2}=\frac{4-\beta+
\sqrt{\beta(17\beta-8)}}{4(\beta-1)}, \eqno (19)$$ where $\beta=4\pi nm_{\rm e} c^2\gamma/B^2$. We denote the threshold of stability $p_{\parallel}/p_\perp=\alpha$, $\alpha>1$. Expression (19) is not quite valid for very large $\beta(>8)$ but large $\beta$ are not achieved here. $\gamma_{\bf
k}>0$ arises when $p_{\parallel}>\alpha p_{\perp}$. The instability is fast because its growth rate is proportional to the cyclotron frequency; still the growth rate is much less than the cyclotron frequency. We will assume that the process of acceleration and the fast instability considered above will form a nonequilibrium distribution function with anisotropy of order unity. However, after particles leave the acceleration region ($\beta>1$), various instabilities such as the “fan" instability considered in this section lead to fast full isotropization of the distribution function.
Now we understand that ${dJ_\perp}/{dt}\ne 0$ not only because of the synchrotron losses but also because of the anisotropic plasma instability. We use the relation $p_{\parallel}/p_\perp\simeq\alpha$ as an approximate closure equation for the system of two equations (1), and express the longitudinal and perpendicular components of the particle momentum through the energy by the relations $$p_{\perp}^2=\frac{\e^2-m_{\rm e}^2 c^4}{(1+\alpha^2)c^2}; \quad
p_{\parallel}^2=\frac{\alpha^2(\e^2-m_{\rm
e}^2c^4)}{(1+\alpha^2)c^2}.$$ Since the dependence of $\alpha$ on $\beta$ (equation (19)) is weak for $\beta>2$, the value of $\alpha$ is approximated as a numerical constant independent of $r$ in further expressions for particles trajectories and particles distribution function. For the derivative of the energy we write equation (11) adding the synchrotron losses. $$\frac{d\e}{dt}=\frac{\e v_\perp^2}{1-v_\|}\frac{(\omega+k)[F-\omega]}{rS}
\frac{d{\bar x}}{dt}-\frac23 \frac{p_\perp^2}{m_{\rm
e}}\om_{B0}^2r_{\rm e} .\eqno (20)$$ Substituting $d{\bar x}/dt$ from (14) we obtain $$\frac{d\e}{dt}=\frac\e{1+\alpha^2-\alpha\sqrt{1+\alpha^2}}
\frac{(\omega+k)[F-\omega]}{rS}\left(\frac{\om^*A_1
B_1}{4t}\right)^{1/2}
-\frac23\frac{\e^2}{1+\alpha^2}\om_{B0}^2\frac{r_{\rm e}}{m_{\rm
e}} .\eqno (21)$$ Let us introduce some notations for convenience: $$Q=\frac1{1+\alpha^2-\alpha\sqrt{1+\alpha^2}}
\frac{(\omega+k)[F-\omega]}{S};$$ $$\e_1=\frac34\frac{m_{\rm e}}{1+2\alpha^2-2\alpha\sqrt{1+\alpha^2}}
\left(\frac{(\omega+k)[F-\omega]}S\right)^2 \frac{{\om^*}^2A_1
B_1}{\om_{B0}^2r_{\rm e}r^2\om^*}. \eqno (22)$$ Here $(\omega+k)[F-\omega]/S $ is fixed for given wavenumbers of the perturbation, $m$ and $k$. It is large near the fast magnetosonic resonance surface $S=0$. For an AGN jet where $B\approx10^{-2}\,\mbox{G}$ (Begelman at al., 1984), $r\approx 1\,\mbox{pc}$, the coefficient $1/(\om_{B0}^2r_{\rm
e}r^2\om^*) \approx 10^4$, which is a large number. The value of ${\om^*}^2A_1 B_1$ is the dimensionless amplitude of the perturbation squared as we noted after Eq. (13). Now taking into account our notations, the solution for $\e$ has the form $$\e=\left(\frac1{\e_1}\left(Q\frac xr-1\right)+\left(\frac1{\e_1}+
\frac1{\e_0}\right)\exp\left\{-Q\frac xr\right\} \right)^{-1}.
\eqno (23)$$ Here $\e_0=\e(x=0) $. Let us denote $Qx/r=x'$. If $x'$ is of order unity, then $\e \simeq \e_0$ and particle acceleration is not effective. We assume that $x'$ can be large because the quantity $Q$ is large in the case of maximal damping rate of eigenmodes. If $x'$ is large, then $\e \gg \e_0$ and the particle acceleration is effective. It is seen that first $\e$ increases exponentially and then decreases as $1/x'$. We consider the asymptotic behaviour of $\e$ in the case of different $x'$ and $\e_0$. In the case of $\e_0<\e_1$ $$\e=\left\{\begin{array}{rrr} \label{asym}
\e_1/x',&\quad&x'e^{x'}>\e_1/\e_0\\
\e_0e^{x'},&\quad&x'e^{x'}<\e_1/\e_0\\
\end{array}\right. \mbox{.} \eqno (24)$$ In the case of $\e_0>\e_1$ $$\e=\left\{\begin{array}{rrr}
\e_1/x',&\quad&x'>1\\
2\e_1/{x'}^2,&\quad&x'<1\\
\end{array}\right. \mbox{.}$$ We ignore the case when $\e_0>\e_1$ since the typical initial energy is much less then $\e_1$. Knowing the particle trajectories (Eqs. (14) and (23)), we can find the distribution function $F(\e,x)$ of accelerated particles in the phase space of $(\e,x)$ using the fact that the trajectories of the particles are the integrals of the collisionless kinetic equation.
Formation of the spectrum of accelerated particles
==================================================
Our goal is to calculate the distribution function averaged over $x$ knowing the distribution function at $x=0$. We will consider the stationary case, when $\pa f/\pa t=0$. Let the distribution function be given at point $x'_0$ as a function of $\e'=\e/\e_1$. Then the number of particles $$dN=F_0(\e'_0,x'_0)d\e'_0dx'_0=F_0\Bigl(\e'_0(\e',x'),x'_0(\e',x')\Bigr)
\frac{D(\e'_0,x'_0)}{D(\e',x')}d\e'dx'.$$ Below we scale $x$ with $r/Q$ and $\e$ with $\e_1$ and omit primes in the dimensionless $x'$ and $\e'$. Thus if we know the trajectory $\e_0(\e,x),x_0(\e,x)$ we may obtain the distribution function at the point $\e,x$. In our case the Jacobian $D(\e_0,x_0)/D(\e,x)$ is not equal to unity because the variables $\e,x$ are not canonical. So, $$F(\e,x)=F_0\Bigl(\e _0(\e ,x ),x _0(\e ,x )\Bigr)
\frac{D(\e _0,x _0)}{D(\e ,x )}. \eqno (25)$$ If we repeat the derivation of Eq. (23) with an arbitrary reference point $x_0\neq 0$ ($\e(x_0)=\e_0$) we obtain $$\e=\{x-1+\left(\frac{1}{\e(x_0)}-x_0+1\right)\exp(-x+x_0)\}^{-1}. \eqno(26)$$ Substituting $\e_0$ expressed from (26) into (25) we obtain \#1\#2 $$F(\e,x)=F_0\left(\left\{-1+x_0+\left(\frac1\e-x+1\right)\exp(x-x_0)\right\}^{-1},x_0\right)
\left(\der{\e_0}\e\der{x_0}x-\der{\e_0}x\der{x_0}\e\right)= \eqno(27)$$ $$F_0\left(\left\{-1+x_0+\left(\frac1\e-x+1\right)\exp(x-x_0)\right\}^{-1},x_0\right)
\frac{\e_0^2}{\e^2}\exp(x-x_0)\frac x{x_0}.$$ We know that initial fast particles are injected near the point $x=0$ \[init\]. Therefore in the last formula we let $x_0=0$ everywhere except in the denominator. Physically $F_0(\e_0,0)$ is the particle density at zero point and $x_0$ is the distance below which our formulae, for instance (14), are not valid. Because of the small imaginary part of $\omega$ which is an attenuation increment of the wave (Istomin & Pariev, 1996), the divergence $\ln x$ at the resonance is cut at the point $$|r_0-r_{\rm A}|\approx\left|\frac{A(r=r_{\rm A})}{dA/dr(r=r_{\rm A})}\right|
\approx \left|\mbox{Im}\,\omega \frac{r_{\rm
A}}{\mbox{Re}\,\omega}\right| ,\eqno (28)$$ where $\mbox{Im}\,\omega \ll \mbox{Re}\,\omega$. Here $A(r=r_{\rm A})$ is not equal to zero as we assumed earlier, but determined by the small value of the imaginary part of $\omega$. Thus, $x_0$ is the width of the resonance $ x_0=Q(r_0-r_{\rm A})/r_{\rm A}$. Eventually, Eq. (27) is $$F(\e,x)= F_0\left(\left\{-1+\left(\frac1\e-x+1\right)\exp(x)\right\}^{-1}
,x_0\right)
\frac{\e_0^2}{\e^2}\exp(x)\frac x{x_0}. \eqno (29)$$ Because the width of the resonance $x_0$ and the width of the acceleration region $\Delta r$ \[width\], which is some function of $x_0$, are small compared with the jet radius we average the distribution function over the jet radius near the resonance surface $$\bar F(\e)=\frac1{\Delta x}\int\limits_0^\infty F(\e,x)\,dx,\quad \Delta x=
\frac Q{r_{\rm A}}\Delta r. \eqno (30)$$ Denoting $N(\e)=\bar
F(\e)\Delta x$ we obtain $$N(\e)=\int\limits_0^\infty F_0
\left(\left\{-1+\left(\frac1\e-x+1\right)\exp(x)\right\}^{-1},
x_0\right) \frac{\e_0^2}{\e^2}\exp(x)\frac x{x_0}\,dx. \eqno(31)$$ Let us change the variable of integration to $\e_0$. The dependence of $\e_0$ versus $x$ under fixed $\e$ is presented in Fig. 1.
![The dependence of $\e_0$ on $x$ (the energy $\e$ and the distance $x$ are in units of $\e_1$ and $r/Q$ (22) correspondingly) []{data-label="fig1"}](fscan1.eps){width="\textwidth"}
The value of $\e_{\rm min}$ is $$\e_{\rm min}=1/(-1+\exp(1/\e)). \eqno (32)$$ Because the function $x=x(\e_0)$ has two branches, the equation (31) is transformed into $$N(\e)=\frac1{\e^2x_0}\int\limits_{\e_{\rm min}}^\e F_0(\e_0,x_0)
\frac x{1/\e-x}\,d\e_0+$$ $$\frac1{\e^2x_0}\int\limits_{\e_{\rm min}}^\infty F_0(\e_0,x_0)
\frac x{1/\e-x}\,d\e_0,\eqno(33)$$ where $x$ is taken on the first branch of $x(\e_0)$ in the first integral and on the second branch in the second integral. Let us make some natural assumptions. The initial distribution function is cut off at some energy of ${T_0}$, provided that $T_0 \ll 1$ (bearing in mind that $T_0$ as well as $\e$ are in the units of energy $\e_1$ (22)). Let us calculate $N(\e)$ in the region, where $T_0\ll\e\ll 1$, because it is seen from Eqs. (32) and (33) that in the case $\e\sim1$ the integral is small. Let us evaluate the contribution of the singularity $x=1/\e$ in integrals (33) $$\e_0-\e_{\rm min}=\frac{d^2\e_0}{dx^2}\left(x-\frac1{\e}\right)^2\mbox{.}$$ The integral near the singularity is $$\frac1{\e^2} \int\limits_{\e_{\rm min}} F_0(\e_{\rm min},x_0)\frac{\frac1{\e}\sqrt{
\frac{d^2\e_0}{dx^2}}}{x_0\sqrt{\e_0-\e_{min}}}\,d\e_0\sim\sqrt{\e_0-\e_{\rm
min}}.$$ Thus, its contribution is small. Because the main contribution to Eq. (26) is when $x$ is far from $1/\e$, we may use asymptotic values of $x$ (see Eq. (24)): $
x\approx\ln(\e/\e_0) $ on the first branch; $ x\approx1/\e+1 $ on the second branch and obtain $$N(\e)=\frac1{\e^2x_0}\int\limits_{\e_{\rm min}}^\e F_0(\e_0)
\frac{\ln(\e/\e_0)}{1/\e-\ln(\e_0/\e)}\,d\e_0
+\frac1{\e^2x_0}\int\limits_{\e_{\rm min}}^\infty F_0(\e_0)\frac
1\e\,d\e_0.\eqno(34)$$ Let us consider only $\e_{\rm min}\ll T_0$ i.e. $\e\ll-1/\ln T_0$ because in the reverse case the integral is small. Now we can replace $\e_{\rm min}$ by $0$ and also $1/\e\gg\ln(\e/\e_0)$. Eq. (34) becomes $$N(\e)=\frac1{\e x_0}\int\limits_0^\e F_0(\e_0)(-\ln \e_0)\,d\e_0+\frac n{\e^3 x_0}, \eqno(35)$$ where $n=\int_0^\infty F_0(\e_0)\,d\e_0$. Rewriting Eq. (35) as averaging $<\ldots >$ over the initial distribution function $F_0$ $$N(\e)=\frac n{\e x_0}<\! -\ln \e_0 \! >+ \frac n{\e^3 x_0}, \eqno (36)$$ we note that the first term is approximately equal to $$\frac n{\e x_0} (-\ln T_0),$$ and is small compared to the second term because the inequality $1/\e^2 \gg -\ln T_0$ is the consequence of inequality $1/\e \gg
-\ln T_0$. Therefore $N(\e)\propto 1/\e^3 $, and we obtain a power-law distribution with the index $-3$. The validity region of this formula $T_0<\e<-1/\ln T_0$ is quite large.
We performed the integration in Eq. (33) numerically, with Gaussian initial distributions at different temperatures. The results are presented in Figs. 2 and 3.
![Distribution function calculated with a Gaussian initial distribution with temperature $2\cdot10^{-5}\e_1$ (the energy $\e$ is in units of $\e_1$ (22)) []{data-label="fig2"}](fscan2.eps){width="\textwidth"}
![Distribution function calculated with Gaussian initial distribution with temperature $10^{-3}$ (the energy $\e$ is in units of $\e_1$ (22)) []{data-label="fig3"}](fscan3.eps){width="\textwidth"}
The value of the power-law index is $-3$ for small energies and rises slightly before the cutoff of the distribution function. However, the $-3$ power-law index is not applicable all the way down to $\e=0$. The natural boundary is the width of the acceleration region, $\Delta r$ (see Eq. (33)). Because the main contribution to the integral (28) is from the region near the point $x=1/\e+1$, the lower limit for power law $1/\e^3$ is $\e_{\rm l}=1/\Delta x$. We also found the mean power law index for the distribution of particles between energies $\e_{\rm l}$ and $-1/\ln T_0$. The results are presented in Fig. 4.
![Averaged power law index (the initial particle energy $T_0$ is in units of $\e_1$ (22)) []{data-label="fig4"}](fscan4.eps){width="\textwidth"}
Recall that in the acceleration region the distribution function of the accelerated particles has an anisotropy of order unity, but the distribution of particles escaping the acceleration region becomes isotropic in momentum space very quickly.
One remaining question in the formation of the spectrum of energetic particles is the origin \[origin\]of the initial particles, since to first order in the perturbed fields there is no acceleration of the cold plasma in Eqs. (1). We have already mentioned that the particles to be accelerated originate in the region close to the Alfvén resonance. It can be seen from Eqs. (1) that to second order in the perturbed fields, acceleration always takes place and the power acquired corresponds to the term ${\bf j\cdot E}$ in the energy conservation equation for the electromagnetic field. The damping of the wave occurs due to the heating of the plasma by current ${\bf j}_1$ which is described by second-order terms. If all of the wave energy is transmitted to the particles, the condition $\beta>1$ would imply that the energy density of the wave is greater than the energy density of the stationary field. Let us assume this at least on the surface of the resonance where the field is maximal. Then the energy acquired by one particle can be evaluated as $T_0\approx
B_0^2/8 \pi n$. Given typical values $B_0\approx 10^{-2}\,
\mbox{G}$ and $n\approx 0.1\, \mbox{cm}^{-3}$ we obtain $T_0\approx 10\,\mbox{MeV}$.
Summary
=======
In the present work we considered possible acceleration of electrons and positrons inside relativistic rotating electron-positron force–free cylindrical jets with spiral magnetic fields. It is very plausible that inner parsec–scale jets in active galaxies can be described in the frame of this physical model. The observations of jets in 3C273, M87, 3C279, 3C84 and PKS0528+134 show that extragalactic jets are likely to be electron-positron ( Morrison & Sadun, 1992; Reynolds et al., 1996; Wardle et al., 1998 ). It also seems that such jets are not in steady state and can contain waves excited by environmental processes, of which the most powerful would be the variability of the accretion rate onto the central black hole and accretion disc instabilities (magnetorotational instability). Both strongly perturb the magnetic field at the base of the jet. We considered the behaviour of such excitations inside cylindrical force–free jets embedded in a dense medium in our previous studies (Istomin & Pariev, 1994; Istomin & Pariev, 1996). We estimate that the density of the gas commonly found in the nuclei of galaxies is large enough that it should be taken into account when considering the evolution of perturbations inside the force–free magnetized jet. We found that in the wide range of parameters determining the equilibrium structure of electromagnetic fields and perturbations inside the jet there exist resonant surfaces on which the phase speed of eigenmodes is equal to the local Alfvén velocity. Those eigenmodes for which there is an Alfvén resonance have a small damping rate, whilst those having no Alfvén resonances are neutrally stable. This shows that the existence of an Alfvén resonance leads to losses of energy of corresponding eigenmodes.
Indeed, the amplitude of the wave is large in the vicinity of the Alfvén resonant surface and decreases away from the Alfvén resonant surface. Both electrons and positrons are subject to drifting out of this region of strong electric field of the wave due to the non-uniformity of the electric field. When moving to a region of weak electric field, particles gain energy. This energy is taken from the energy of the wave, which is an eigenmode in the jet, and the eigenmode decays. In this work we propose a novel mechanism for acceleration of particles [*in situ*]{} in strong electromagnetic waves of varying amplitude inside the jet. Unlike the acceleration on shock fronts proposed to operate inside bright knots (e.g., Blandford & Eichler 1987), the acceleration by electromagnetic waves is regular, i.e. it does not require turbulence or inhomogeneities to be present in the jet. Also, the net gain of energy by particles in our mechanism occurs slowly on a time scale greatly exceeding the Larmour orbital time and also exceeding the period of the wave. In this aspect, the regular electromagnetic acceleration by a nonuniform wave differs from acceleration in magnetic reconnection layers (Romanova & Lovelace 1992; Blackman 1996), where strong departures from ideal MHD occur and a particle gains its total energy during one passage through the region of strong electric field. Our acceleration mechanism does not rely upon violation of ${\bf E}\cdot {\bf B}=0$ in any region inside the jet as the gain of energy during the drift is a result of a kinetic treatment of plasma possible only when one takes into account a finite Larmour radius.
We found how particles are accelerated in the strong fields near the Alfvén surface. We considered the process of particle acceleration using equations of motion in drift approximation with addition of synchrotron losses and isotropization of particle distribution by plasma instabilities. Such instabilities are excited by accelerated particles if the distribution function is anisotropic in momentum space, and act to conceal any anisotropy of the distribution of particles. Acceleration process and synchrotron losses taken together form a power law energy spectrum of ultra-relativistic electrons and positrons with index between 2 and 3 depending upon the initial energy of the injected particles. This is consistent with the typical spectral indices of radio emission observed in parsec–scale jets. The power law spectrum extends up to the energy $\e_{\rm max}$, where a sharp cutoff occurs (see Figs. 2,3). The magnitude of $\e_{\rm max}$ depends on the initial particle temperature $T_0$ as well as on the characteristic acceleration energy $\e_1$ (26): $\e_{\rm max}\approx \e_1/\ln(\e_1/T_0)$. The quantity $T_0$ evaluated from the equipartition condition in the acceleration region is equal to $10\,\mbox{MeV}$ for the magnetic field $10^{-2}\,\mbox{G}$ and density $0.1\,\mbox{cm}^{-3}$ correspondingly. The quantity $\e_1$ is, according to (26), of the order of $10^4(\delta B/B_0)^2\,
\mbox{MeV}$ for the value of $B_0=10^{-2}\,\mbox{G}$, where $\delta B/B_0$ is the dimensionless amplitude of the perturbation. For large perturbations ($\delta B\approx B_0$) $\e_1\sim 10^4\,\mbox{MeV}$. Thus particles accelerated near Alfvén resonance are in the energy range of $10\,\mbox{MeV}<\e<10^3\,\mbox{MeV}$. These particles emit synchrotron radiation in the range of frequencies from approximately $100\,\mbox{MHz}$ to $1000\,\mbox{GHz}$ which covers the frequencies of modern radio observations. We present results of the calculations of synchrotron emission and polarization produced by the accelerated particles with power law distribution over energy in a separate article (Pariev et al. 2002).
We thank R.I. Selkowitz for checking the English in the final version of the manuscript. This work was done under support of the Russian Foundation for Fundamental Research (grant number 02-02-16762). VP acknowledges partial support from DOE grant DE-FG02-00ER54600.
[References]{}
Alfvén H., Fälthammar C.-G., 1963, Cosmical Electrodynamics, Fundamental Principles. Second edition, Clarendon Press, Oxford, p.28
Appl S., Camenzind M., 1992, A&A, 256, 354
Begelman M.C., Blanford R.D., Rees M.J. 1984, Rev. Mod. Phys., 56, 255
Beskin V.S., Istomin Ya.N., Pariev V.I., 1992, AZh, 69, 1258
Bisnovatyi–Kogan G.S., Lovelace R.V.E., 1995, A&A, 296, L17
Blackman E.G., 1996, ApJ, 456, L87
Blandford R.D., Eichler D., 1987, Physics Reports, 154, 1
Blanford R.D., Pringle J.E., 1976, MNRAS, 176, 443
Frolov V.P., Novikov I.D., 1998, Black hole physics: basic concepts and new developments. Kluwer Academic, Dordrecht
Hirotani K., Okamoto I., 1998, ApJ, 497, 563
Hirotani K., Iguchi S., Kimura M., Wajima K., 2000, ApJ, 545, 100
Istomin Ya.N., Pariev V.I., 1994, MNRAS, 267, 629
Istomin Ya.N., Pariev V.I., 1996, MNRAS, 281, 1
Li L.-X., 2000, ApJ, 531, L111
Lyubarskii Yu.E., 1999, MNRAS, 308, 1006
Mikhaylovskii A.B., 1975, in Reviews of Plasma Physics, ed. by Leontovich M.A., Consultants Bureau, New York, Vol. 6, p. 77
Morozov A.I., Solov’ev L.S., 1966, in Reviews of Plasma Physics, ed. by Leontovich M.A., Consultants Bureau, New York, Vol. 2, p. 201
Morrison P., Sadun A., 1992, MNRAS, 254, 488.
Mushotzky R.F., Done C., Pound K., 1993, Ann. Rev. Astron. Astrophys., 717
Parail V.V., Pogutse O.P., 1986, in Reviews of Plasma Physics, ed. by Leontovich M.A., Consultants Bureau, New York, Vol. 11, p. 1
Pariev V.I., Istomin Ya.N., Beresnyak A.R., 2002, A&A, accepted
Rees M.J., 1984, Ann. Rev. Astron. Astrophys. ,471
Reynolds C.S., Fabian A.C., Celotti A., Rees M.J., 1996, MNRAS, 283, 873.
Romanova M.M., Lovelace R.V.E., 1992, A&A, 262, 26
Sikora M., Madejski G., 2000, ApJ, 534, 109
Sivukhin D.V., 1965, in Reviews of Plasma Physics, ed. by Leontovich M.A., Consultants Bureau, New York, Vol.1, p.1
Torricelli-Ciamponi G., Petrini P., 1990, ApJ., 361, 32
Wardle J.F.C., Homan D.C., Ojha R., Roberts D.H., 1998, Nature, 395, 457
Witzel A., Wagner S., Wegner R., Steffen W., Kirchbaum T., 1993, in Davis R.J. & Booth R.S., eds Sub-arcsecond Radio Astronomy. Cambridge Univ. Press, Cambridge, p. 159
| {
"pile_set_name": "ArXiv"
} |
epsf
\#1\#2\#3\#4[[\#1]{} [**\#2**]{}, \#3 (\#4)]{} §[Science]{}
Introduction
============
Spin-$1/2$ Heisenberg ladder compounds have been the subject of extensive investigations over the last few years,[@Matsud] first and foremost due to their magnetic peculiarities. These are caused by the magnetic interactions within the legs and the rungs which, in principle, may be very different for the various sublattice units. Assuming, e. g., a coupling across the rungs that is much larger than that along the legs leads to the formation of a spin gap for compounds with even numbers of legs and to a tendency for “hole pairing”. Both features turn out to persist even for fairly similar coupling constants as in fact found for many actual ladder compounds. Early on, superconductivity was predicted for even-leg ladders[@rice] and was eventually found for Sr$_{0.4}$Ca$_{13.6}$Cu$_{24}$O$_{41.8}$.[@uehara] This compound is a member of the more complex family (Sr,Ca)$_{14}$(CuO$_2$)$_{10}$(Cu$_2$O$_3$)$_7$, which contains layers of Cu$_2$O$_3$ two-leg ladders extending along the crystallographic $c$ axis (see Fig. \[structure\]); the ladder layers are stacked along the $b$ axis and alternate with layers of edge-sharing CuO$_2$ chains.[@carron; @siegrist] The observation of superconductivity is highly remarkable as this compound is a cuprate but does not contain the CuO$_2$ planes so characteristic for high-temperature superconductors (HTSCs). It is, therefore, not surprising that these findings initiated a flurry of experimental and theoretical investigations on these compounds, their electronic structure, and its relationship to the magnetic properties.
Due to its composition, Sr$_{14}$Cu$_{24}$O$_{41}$ is doped by 6 holes, which are generally believed to reside essentially in the chain. Perhaps the simplest indication for this is
provided by bond-valence-sum (BVS) calculations,[@katobvs] giving higher BVS values for holes in the chains. Similar to the HTSCs, the holes are expected to be mainly of O $2p$ character as the large Hubbard $U$ for Cu $3d$ states prevents them from having much Cu character. On the experimental side, resistivity and magnetic measurements on “undoped” La$_6$Ca$_8$Cu$_{24}$O$_{41}$, with purely a Cu$^{2+}$ valence and thus no holes, as well as on compounds with increasing Sr content for La and a corresponding number of doped holes led to the conclusion that [*all*]{} doped holes reside on the chains of Sr$_{14}$Cu$_{24}$O$_{41}$, giving a Cu$^{2.6+}$ valence there.[@carter] Neutron scattering,[@Matsuda; @Regnault] however, points to a spin dimerization along the chain, which in turn was interpreted as a signature for a CDW-like arrangement of one hole for every other Cu(c) site of the chain. This means that in total only 5 out of the 6 doped holes reside in the (CuO$_2$)$_{10}$ chain subunit of Sr$_{14}$Cu$_{24}$O$_{41}$, leaving one hole for the (Cu$_2$O$_3$)$_7$ ladder unit. Optical investigations[@Uchida; @ruzicka] are also consistent with a hole count of one in the ladder unit of Sr$_{14}$Cu$_{24}$O$_{41}$; upon Ca substitution this was found to increase to up to 2.8 holes. A recent Cu NMR study[@magishi] relates measured spin correlation lenghts to the average distance between holes and determine a hole occupation on the ladder sites up to 1.75 holes for Sr$_{2.5}$Ca$_{11.5}$Cu$_{24}$O$_{41}$. The formation of dimers was supported by further NMR[@Takigawa] and NQR[@ohsugi] experiments, and the underlying charge ordering was observed by x-ray diffraction.[@Cox] Also, NMR measurements[@tsuji] suggest that the spin gap of the ladder unit decreases continuously over the whole substitutional range from zero holes to the fully doped and heavily Ca-substituted compounds until it almost reaches zero, while the spin gap of the chain remains essentially constant, lending support to the notion that it is indeed the ladders that carry the
superconductivity. On the other hand, other NMR experiments[@magishi] and neutron scattering[@katano5] find a constant spin gap upon Ca substitution for the ladders as well. Chemical pressure by Ca substitution for Sr as well as physical pressure reduces the interlayer distances[@Pachot] and increases the twisting of the CuO$_2$ chain.[@Isobe] As a consequence, about 30 - 50% of the O(c) atoms reduce their distance to the nearest Cu(l) atom to a value comparable with the typical Cu-O$_{\rm apex}$ distance for HTSCs.[@Isobe] A transfer of holes from the chains to the ladder sites concurrent with Ca substitution was predicted theoretically[@Theorie] and was also inferred from various experiments.[@Uchida; @ruzicka; @Experiment]
Despite all this wealth of information on the ladder compounds, however, there remains a considerable amount of controversy, and a more [*direct*]{} determination of the site-specific hole distribution is still missing. Both the evolution of hole distributions with increasing number of holes and with increasing Ca content with its structural implications is important to know. In the present study, polarization-dependent x-ray absorption spectroscopy is applied, which is a proven method to obtain such information. The results, in turn, may help to better understand the electronic structure and possibly the onset of the insulator-metal transition as well as the appearance of superconductivity in these systems.
Experiment and results
======================
Single crystalline samples (typically about 60 mm$^3$ in size) were cut from several cm long crystals grown by the traveling solvent floating zone (TSFZ) method;[@growth; @Ammerahl] their compositions are shown in Table I. The formal valence of Cu in the samples was determined by a precision volumetric method,[@Conder] where an overall accuracy of $< 5 \times 10^{-4}$ can be achieved. An average Cu valence +2+$\varepsilon$ determined in this way gives a total hole count $n_{\rm Cu}$=24$\varepsilon$, and any deviation from the nominal number of holes expected from stoichiometry, $n_h$, is accounted for by simply taking the O content, which could not be determined indepen-
dently, as 41+$\delta$ atoms per formula unit with $\delta$=($n_{\rm Cu}$-$n_h$)/2. The largest $\delta$ observed in the present study occurred for the La$_6$Ca$_8$Cu$_{24}$O$_{41.43}$ samples and corresponds to 0.86 extra holes. An investigation by scanning electron microscopy (SEM) and microspot Auger analysis revealed that a few percent of La in these samples is segregated as La-rich precipitates, illustrating the fact that a high La content leads to phase instability in these compounds.[@Ammerahl] The $\delta$=0.14 found for La$_5$Ca$_9$Cu$_{24}$O$_{41.14}$ may result from the same effect. For none of the other compositions traces of such phases could be detected by SEM or x-ray diffraction; also, much smaller values for $\delta$ were observed (Table I).
Polarization-dependent near-edge x-ray absorption fine structure (NEXAFS) measurements were performed at the National Synchrotron Light Source (NSLS) at the NRL/NSLS beamline U4B. The energy resolution in the range of the O 1$s$ NEXAFS was set to 210 meV. To ensure a flat and shiny surface, several slices ($\approx$200 nm each) were cut off from the samples using the diamond blade of an ultramicrotome, and the samples were then transferred to ultrahigh vacuum. The data were recorded at room temperature and in fluorescence detection mode probing a depth of about 100 nm. The spectra were corrected for incident flux variations and for self-absorption effects,[@troeger] and were normalized to tabulated cross sections[@yeh] at about 70 eV above the edge where the absorption is atomic-like and structureless. For energy calibration of the O $1s$ spectra, total electron yield data of NiO were taken simultaneously and referred to a NiO standard[@eels] from electron energy-loss spectroscopy with a reproducibility significantly better than the experimental energy resolution.
The aim of this experiment is to investigate the hole distribution in the three inequivalent O sites of (Sr,Ca,Y,La)$_{14}$Cu$_{24}$O$_{41}$. With each of the inequivalent sites, a different Cu-O bond configuration is associated ([*cf*]{}. Fig. \[structure\]): a Cu-O-Cu interaction geometry with about 90$^{\circ}$ for the O(c) sites of the chains, similar to the situation in Li$_2$CuO$_2$;[@Sapina; @neudert] straight bonds (180$^{\circ}$) connecting the Cu(l) sites in the two legs of a single ladder [*via*]{} the rung O(2) sites; and the situation for the O(1) sites of the legs, which form bonds to each of the three neighboring Cu(l) atoms. Assuming that, analogous to the situation in HTSCs, only $\sigma$ bonds formed between Cu $3d$ and O $2p$ states contribute spectral weight near the Fermi level, $E_F$, there are only five different O $2p$ orbitals in the $a$,$c$ plane (see Fig. \[structure\]) which may contribute spectral weight in O $1s$ NEXAFS. The excitation process in NEXAFS involves the highly localized O $1s$ core level; moreover, dipole selection rules apply. As a consequence, orienting the electric field vector of the incident radiation, [**E**]{}, parallel to the crystallographic axes allows one to investigate [*specifically*]{} the O $2p$ orbitals oriented likewise, and thus gain site-specific information on the hole distribution.
Fig. \[fullrange\] depicts the O $1s$ absorption and its polarization dependence over the full measured energy range for Sr$_{14}$Cu$_{24}$O$_{41}$ as a representative example for the whole set of compositions. The two large peaks observed near threshold for the in-plane, [**E**]{}$\|$$a,c$ spectra are due to Cu-O hybridization and are strongly suppressed for the out-of-plane, [**E**]{}$\|$$b$ spectrum. The latter exhibits right at the main absorption edge a strong feature which is most likely associated with Sr-O hybrids connecting the chain and ladder layers. In the extended (EXAFS) range the variation of the absorption coefficient $\sigma$ is very different from that for the in-plane geometry, illustrating that the spatial environment of O in the direction perpendicular to the planes is very different from that within the planes.
O $1s$ NEXAFS spectra for (Sr,Ca,Y,La)$_{14}$Cu$_{24}$O$_{41}$ are shown in Fig. \[alldata\] for all three orientations [**E**]{}$\|$$a$, [**E**]{}$\|$$b$, and [**E**]{}$\|$$c$ in the energy range of interest below the main edge. Concentrating first on the “planar” [**E**]{}$\|$$a$,$c$ spectra one observes considerable spectral weight near threshold. The first feature above threshold, around 528 eV and called H in the spectra, appears near the O 1$s$ binding energy as determined by x-ray photoemission spectroscopy (not shown). Feature H, therefore, represents hole states; its spectral weight is related to the number of holes per unit cell. Analogous to the situation for HTSC materials, a second maximum is observed in the spectra which represents the upper Hubbard band (UHB) and is denoted by U. It appears in the O $1s$ spectra due to the Cu $3d$ – O $2p$ hybridization the strength of which is described by the hopping integral $t_{pd}$. The spectral weight of feature H is smallest for La$_6$Ca$_8$Cu$_{24}$O$_{41.43}$ and grows with the Sr and/or Ca content, i. e., with the number of doped holes. At the same time, the spectral weight of feature U is seen to decrease. Particularly noteworthy is that the NEXAFS spectra of Y$_3$Sr$_{11}$Cu$_{24}$O$_{41}$ and La$_3$Sr$_3$Ca$_8$Cu$_{24}$O$_{41}$, both doped with three holes per formula unit, are identical within the experimental precision despite their very different Ca content; furthermore, they show equivalent spectral weight in H for [**E**]{}$\|$$a$ and [**E**]{}$\|$$c$. While for these two compounds, moreover, feature H is almost symmet-
ric in energy (and nearly Gaussian-shaped), one finds for the fully doped samples Sr$_{14-x}$Ca$_x$Cu$_{24}$O$_{41}$ a pronounced, shoulder-like asymmetry of H towards higher energy. An immediate conclusion that could be drawn from observing that these symmetries of H are present at low doping levels but absent at full doping is that for low doping (up to 4 holes per f.u.) all holes reside on the chain sites: only there spectral weight for [**E**]{}$\|$$a$ and [**E**]{}$\|$$c$ is naturally the same as both orientations measure equal projections of the [*same*]{} O $2p$ orbitals ([*cf*]{}. Fig. \[structure\]). This is in agreement with previous work ([*cf*]{}. Refs. and ). For [**E**]{}$\|$$b$, a similar double-peak structure in the energy range shown is observed; the corresponding spectral weights are, however, smaller by an order of magnitude compared to the total in-plane weight for [**E**]{}$\|$$a$,$c$. This is a nice illustration of the predominantly planar character of the electronic structure near $E_F$ for both the chain and the ladder units.
These qualitative observations can be put on a somewhat firmer footing by a more refined data analysis based on the following ideas: Systematic changes in H and U upon hole doping and upon Ca substitution can clearly be observed in Fig. \[alldata\], like the developing asymmetries in H described above, although individual O $2p$ contributions are not well-separated from each other. With the present study, on the other hand, a large number of spectra for different compositions is available, and a promising way to extract as much site-specific information as possible is to perform a least-square fit to [*all*]{} measured spectra [*simultaneously*]{}. With the assumption that for all compositions the individual, site-specific spectral distributions are identical in shape, and that only their amplitudes and energy positions may change with composition, one can even construct these spectral distributions corresponding to various O orbitals in the fitting procedure, and thus avoid introducing artifacts through a possibly imperfect choice of spectral shapes for the fitting functions. Of course, the spectral weight and energy shifts for each composition are also obtained in the same fit. The spectra for La$_6$Ca$_8$Cu$_{24}$O$_{41.43}$ and La$_5$Ca$_9$Cu$_{24}$O$_{41.14}$ were not included in the fitting procedure for construction of the spectral distributions since their strong off-stoichiometry might possibly lead to artifacts in the results. Their partial weight, however, was calculated as well, using the distributions determined from all the other spectra.
In practice one is, of course, limited in the number of spectral distributions that can reasonably be constructed in the fitting procedure; it must be much smaller than the number of independent spectra measured (generally the two in-plane spectra per sample are used). With two distributions each to fit H and U for each polarization the resulting spectral distributions for [**E**]{}$\|$$a$ and [**E**]{}$\|$$c$ turn out to be almost identical, allowing the use of the same distributions for both polarizations. Unlike the situation for H, where two distributions are [*required*]{} to properly approximate the experimental curve[@distance] and its doping dependence, a decomposition of U into two parts is far less justified: the improvement in rms deviation is only marginal, and the variation of relative intensities does not exhibit much connection with doping. It is thus possible to determine only three spectral weights: H1, H2, and U. The energies of peaks H1 and H2 associated with the doped holes remain constant to within 30 meV and are well within the limits of our energy calibration ($<100$ meV); such slight energy shifts could be due to changes in the O $1s$ binding energy. In the case of U, a single fitting function had to be used (as mentioned above) despite U being a compos-
ed feature, and the slightly larger shifts ($<200$ meV) observed for this fitting function may result from this simplification. The spectral weights fitted are discussed in the following.
In Fig. \[fits\] NEXAFS spectra with these three fitted partial distributions are shown for Y$_3$Sr$_{11}$Cu$_{24}$O$_{41}$, Sr$_{14}$Cu$_{24}$O$_{41}$, and Sr$_{2}$Ca$_{12}$Cu$_{24}$O$_{41}$ for [**E**]{}$\|$$a$ and [**E**]{}$\|$$c$ as representative examples for the full data set. The partial weights extracted from such fits, corresponding to hole counts and the weight in the UHB, as well as their evolution as a function of the hole count as derived from volumetry and of the Ca substitution level are shown in Fig. \[weight\]: part (a) depicts the individual contributions H1 and H2 for [**E**]{}$\|$$a$ and [**E**]{}$\|$$c$, while part (b) plots the total weight of H (i. e., H1+H2 summed over both in-plane polarizations) and the analogously determined total weight for U. The left-hand side of both panels in Fig. \[weight\] corresponds to increasing total doping levels, while the right-hand side depicts fully doped (6 holes) samples Sr$_{14-x}$Ca$_x$Cu$_{24}$O$_{41}$.
Concentrating first on part (b) in Fig. \[weight\], we observe the total weight H to increase linearly with doping level as derived by the chemical analysis up to the fully doped compound Sr$_{14}$Cu$_{24}$O$_{41}$. At the same time, the weight U of the UHB decreases. This trend in U is seen to continue smoothly into the range of the Ca-substituted compounds – unlike the behavior of H, which stays constant for Ca contents greater than 4. Turning now to the polarization-dependent decomposition of H into H1 and H2 (Fig. \[weight\] (a)), it is evident that up to and including the compound La$_2$Ca$_{12}$Cu$_{24}$O$_{41}$ with its four doped holes, almost all the weight is found in the lower-energy part H1, and that it is almost the same for [**E**]{}$\|$$a$ and [**E**]{}$\|$$c$ absorption. The straight line shown describes this behavior for H1 well. (H2 remains below 2 Mb$\cdot$eV per formula unit, and this value may be taken as the “threshold of significance” for the spectral decomposition procedure described above; the absence of significant spectral weight other than that for H1 again illustrates that the latter represents doped holes in the chain.) For Sr$_{14}$Cu$_{24}$O$_{41}$, the situation changes considerably: H1 for [**E**]{}$\|$$c$ lies substantially below its [**E**]{}$\|$$a$ counterpart (which itself remains close to the straight line extrapolated from the lower doping levels); and for the first time, there is a significant contribution from H2, amounting to almost 1/3 of H1. The corresponding H2 for [**E**]{}$\|$$a$, on the other hand, is still almost zero. In the total weight (Fig. \[weight\] (b)) the decrease of H1 is thus compensated. For the partially Ca-substituted compound Sr$_{9}$Ca$_{5}$Cu$_{24}$O$_{41}$, H1 for [**E**]{}$\|$$c$ is substantially larger than for Sr$_{14}$Cu$_{24}$O$_{41}$; further increasing the Ca content then leaves both H1 parts almost constant, similar to the behavior of the total H. H2 for [**E**]{}$\|$$c$, on the other hand, remains constant for all Ca-substitution levels, while H2 for [**E**]{}$\|$$a$ exhibits a slight increase over the whole range of Ca fractions,[@sr5] until it reaches, for Sr$_2$Ca$_{12}$Cu$_{24}$O$_{41}$, about 1/6 of the value for H1.
Interpretation and Discussion
=============================
The observation that for La$_3$Sr$_{3}$Ca$_8$Cu$_{24}$O$_{41}$,\
Y$_3$Sr$_{11}$Cu$_{24}$O$_{41}$, and La$_2$Ca$_{12}$Cu$_{24}$O$_{41}$, H consists of only one component, H1, which is furthermore almost symmetric within the $a$$c$ plane, is readily understood if the holes at these doping levels almost exclusively occupy sites in the chains. In the fully doped compound Sr$_{14}$Cu$_{24}$O$_{41}$, however, this symmetry is obviously broken (see above). Most significant is the appearance of H2, which for [**E**]{}$\|$$c$ exhibits considerable weight but for [**E**]{}$\|$$a$ still barely exceeds the “threshold of significance”.
These observations unambiguously show that H2 must be spectral weight corresponding to holes on the O(1) sites of the legs: only there, O $2p_{\sigma}$ orbitals exist which can contribute strongly to [**E**]{}$\|$$c$ weight while not at the same time leading to similarly strong [**E**]{}$\|$$a$ weight, [*cf.*]{} Fig. \[structure\]. Energetically, H2 is observed at $\approx 0.6$ eV above H1, consistent with the interpretation that the corresponding holes reside on O sites in different atomic environments. On the other hand, the [**E**]{}$\|$$a$ from the O(1) leg site should appear at an energy very similar to that for the [**E**]{}$\|$$c$ contributions, as both involve the same core level energy. While possible holes on the rung O(2) sites would appear for [**E**]{}$\|$$a$ only it is not so clear, however, where in energy this contribution would show up in the spectrum.
In the following, we present a more detailed picture of the hole distribution and the upper Hubbard bands, trying to account for all observations made so far, and check it against further experimental results as we proceed:
\(i) As discussed above, H1 is ascribed to the spectral weight of holes residing on the chain O(c) sites, giving a highly symmetric in-plane contribution in NEXAFS. The asymmetry between [**E**]{}$\|$$a$ and [**E**]{}$\|$$c$ at high doping levels, most prominently displayed by Sr$_{14}$Cu$_{24}$O$_{41}$, may have its origin in certain lattice distortions of the chains along $c$, as observed by x-ray diffraction[@Cox] for temperatures below room temperature. These distortions appear simultaneously with spin dimerisation and lead to deviations from the otherwise 90$^{\circ}$ Cu-O-Cu bond symmetry. NEXAFS as a local probe is still sensitive to the asymmetry[@gap] induced in this way to the hole distribution of the chain, even though at room temperature long-range order is already lost.
\(ii) H2 is observed experimentally for [**E**]{}$\|$$c$ and to a smaller extent for [**E**]{}$\|$$a$ as well and is thus ascribed mainly to the spectral weight of holes residing on the leg O(1) sites of the ladder. Increasing the Ca content $x$ for the fully doped compounds Sr$_{14-x}$Ca$_x$Cu$_{24}$O$_{41}$ (right-hand side of Fig. \[weight\] (b)) does not change H2 appreciably for [**E**]{}$\|$$c$, indicating that no more holes are transferred to the O(1) $2p_z$ orbitals; the increase of H2 observed for [**E**]{}$\|$$a$ points to a small but increasing number of holes that are transferred to ladder orbitals oriented along the $a$ axis; the data does not allow to unambiguously decide if to leg O(1) sites and/or rung O(2) sites.
\(iii) U consists of the spectral weight of both the chain UHB and the ladder UHB, and the fact that the fitting procedure described above was not able to separate these two contributions may indicate that the energetic positions and spectral distributions are too similar for all compositions. The in-plane Cu-O coordination and the Cu-O bond lengths in both the chain and the ladder units are almost identical and hence $t_{pd}$ will be to a good approximation the same for chains and ladders. In the undoped compound La$_6$Ca$_8$Cu$_{24}$O$_{41.4}$ there are 10 Cu $3d$ holes per f.u. in the chain and 14 Cu $3d$ holes per f.u. in the ladder unit. Hence one can expect the respective spectral weights in the UHB as observed in O $1s$ NEXAFS to be distributed accordingly: about 47:65 Mb$\cdot$eV for the chain vs. the ladder UHB. Hole doping of the chain by about 5 holes for Sr$_{14}$Cu$_{24}$O$_{41}$ (see above) results in a formal Cu(c) valence of +2.5, which in most cuprates would result in an almost complete transfer of spectral weight from the UHB to the hole states H. Experimentally, we observe a reduction in the total U by $< 20$ Mb$\cdot$eV when going from $n_h$=0 to $n_h$=6 in the left-hand part of Fig. \[weight\] (b). Doping of just a single hole per f.u. into the ladder unit of Sr$_{14}$Cu$_{24}$O$_{41}$ will raise the formal Cu(l) valence only to +2.07, and the associated reduction of the ladder UHB will be far smaller than for the chain unit with Cu(c)$^{+2.50}$. We may thus assign the full reduction observed for U to the chain UHB and find that for Sr$_{14}$Cu$_{24}$O$_{41}$, at most 45% of the spectral weight for the fully developed chain UHB is transferred to H. This is considerably less than observed for doped CuO$_2$ planes in high-$T_c$ cuprates, and even for the closely related doped chain structure of Sr$_{0.73}$CuO$_2$ with a formal Cu valence of 2.54 the UHB has almost vanished.[@dresden5] One does find for edge-sharing chains like the ones in the present study a Cu-O-Cu hopping that is strongly hampered as it is not only governed by $t_{pd}$ but is mediated by two different in-plane O $2p$ orbitals that are mutually orthogonal,[@UHB; @footnote2] which should result in a relatively slow transfer of spectral weight from the UHB to H upon hole doping.[@meinders] The difference to Sr$_{0.73}$CuO$_2$ which consists of edge-sharing chains as well may be that there the Cu-O-Cu bond angles are in fact not as close to 90$^{\circ}$ as in
the chains of the Sr$_{14}$Cu$_{24}$O$_{41}$ family.
\(iv) The spectral weight for [**E**]{}$\|$$b$ is too small to be interpreted individually, especially against the backdrop of the large contributions for [**E**]{}$\|$$a$,$c$. Thus, even small factors like the only 95% degree of linear polarization or possible small misalignments, both mixing small fractions of the planar spectral weight to the [**E**]{}$\|$$b$ spectra, may affect the results for [**E**]{}$\|$$b$ and their relationship to the hole counts for the ladder sites which are also small. It may still be allowed to note that the spectral weight in the hole peak for [**E**]{}$\|$$b$ tracks, in general, the nominal hole doping level; yet one also notes that for Sr$_{14}$Cu$_{24}$O$_{41}$ and Sr$_2$Ca$_{12}$Cu$_{24}$O$_{41}$, the [**E**]{}$\|$$b$ component is unusually high.
\(v) The observation made in the previous section that the total spectral weight H, which in turn is dominated by the chain H1, is significantly larger for the Ca-substituted compound Sr$_{9}$Ca$_{5}$Cu$_{24}$O$_{41}$ than for Sr$_{14}$Cu$_{24}$O$_{41}$ is highly intriguing: stoichiometry requires the total number of holes to be the [*same*]{}! A possible solution for this puzzle is suggested by looking at the Cu $2p_{3/2}$ NEXAFS spectra for Sr$_{14}$Cu$_{24}$O$_{41}$ and Sr$_{2.5}$Ca$_{11.5}$Cu$_{24}$O$_{41}$ shown in Fig. \[copper\]. They are recorded in fluorescence yield mode but as self-absorption effects are exceedingly strong at this edge, corrections like the ones applied to the O $1s$ edge (see above) are problematic here and thus were not performed. Total electron yield spectra, on the other hand, would not be reliable as the sample surfaces could not be cleaved [*in situ*]{}. As the large Hubbard $U$ enforces a Cu $3d^9$ ground state with one intrinsic hole per Cu site, it is still possible to at least qualitatively compare different samples by normalizing the amplitudes of the excitonic peak (or “white line”) around 932.5 eV. Observing a significantly [*lower*]{} ligand-hole peak at about 934 eV for the highly Ca-substituted compound – while the corresponding weight H1 for O $1s$ NEXAFS is clearly [*higher*]{} – may thus be interpreted as a signature for a reduced $t_{pd}$ hopping parameter in the chains, with the immediate consequence that the doped holes have increased O $2p$ character for heavy Ca substitution. This apparent decrease in the chain $t_{pd}$ is likely to be induced by strong distortions of the chain[@siegrist] on incorporation of much smaller Ca atoms, leading to highly inequivalent O sites in the chain with Cu(c)-O(c) distances now varying from 1.71 [Å]{} to 2.02 [Å]{},[@siegrist] and presumably to a quite inhomogeneous redistribution of holes. Thus the observation of a reduced $t_{pd}$ in NEXAFS does not stand in contrast to the structural observation of a reduced average Cu(c)-O(c) bond length but may be taken as an indication that the doped holes in the chain may be located preferably on O sites associated with a smaller effective $t_{pd}$.
\(vi) The fact that the spectral weight of U for the fully doped compounds Sr$_{14-x}$Ca$_x$Cu$_{24}$O$_{41}$ seems to decrease with increasing $x$ is qualitatively consistent with the slight increase observed for H2 in this substitutional regime: if holes are redistributed from chain sites to ladder sites, the “slow” transfer of spectral weight between H and the UHB for the chains (see (iii) above) would lead to a concurrent increase in the chain UHB that is smaller than and thus overcompensated by the decrease in the ladder UHB. The slope of U appears larger than that of H2 and thus at first glance suggests possible inconsistencies. It should be noted, however, that upon Ca substitution the chain $t_{pd}$ is reduced (see (v) above) and lattice distortions increase, presumably resulting in deviations from the 90$^{\circ}$ Cu-O-Cu bond geometry. Both effects directly lead to a reduction of the spectral weight of the chain UHB in O $1s$ NEXAFS and help to explain the decrease in U observed in the right-hand part of Fig. \[weight\] (b).
It is interesting to estimate the absolute hole counts $n_{i,\bf E}$ related to the various O orbitals, where $i$ denotes spectral weight corresponding to H1 or H2, and [**E**]{} denotes the polarization. For each sample, the total spectral weight of H, summed over the in-plane polarizations, is directly equated to the corresponding total hole count as determined by the precision chemical analysis described above. The individual contributions $n_{i,\bf E}$ directly follow from this “normalization”. In this way, the problems associated to possible changes in parameters like $t_{pd}$, for which an indication has been presented in (v) above, are reduced in their effect on the individual hole occupation numbers determined. The results are shown in Fig. \[holes\]. The hole count most obviously associated with the ladders, $n_{2,a}$+$n_{2,c}$, increases slightly from 0.8 holes for Sr$_{14}$Cu$_{24}$O$_{41}$ to 1.1 holes for Sr$_2$Ca$_{12}$Cu$_{24}$O$_{41}$; considering the individual contributions one observes that the increase is due to $n_{2,a}$ only while $n_{2,c}$ stays virtually constant, in effect rendering the hole distribution in ladders more two-dimensional for the heavily Ca-substituted compounds.
These values for fully doped compounds with high Ca content remain considerably smaller than the up to 2.8 holes found in optical measurements.[@Uchida] The interpretation of the optical data rests on assuming zero mobility
for holes on the chains, allowing one to distinguish between holes on the chains (localized) and holes on the ladders (mobile). This may be less valid if heavy Ca substitution leads to mobility in the chain as well, and the hole count for the ladders might thus be overestimated to some extent; but even in this case the difference between the optical and NEXAFS results appears too large. A point raised in Ref. concerns the role of an effective mass greater than one which would reduce the apparent ladder hole count deduced from the optical data; however, the assumption of $n_{\small chain}$=5 and $n_{\small ladder}$=1 made in Ref. to “normalize” their data already takes care of this. On the other hand, NEXAFS suffers to some extent from changes in parameters like $t_{pd}$ as pointed out above. A further effect could be that the holes on the rung O(2) sites might appear in NEXAFS at about the same energy as holes on the chain O(c) sites and thus would, in the evaluation used here, be counted not as a ladder contribution but instead as a chain contribution. If present at all, however, this effect is also expected to be small. The differing results of optical and x-ray determinations of the hole count are thus noted but remain unexplained at this time.
A further remark should be made on the fact that in the original work reporting on superconductivity in these ladder compounds,[@uehara] the very compound studied (Sr$_{0.4}$Ca$_{13.6}$Cu$_{24}$O$_{41.84}$) exhibited an excess oxygen content $\delta$ amounting to as much as 0.84. Although it is difficult to estimate the effect of up to 1.7 extra holes, it seems clear that such additional O atoms may lead to further substantial deformations in the crystal structure, and one could thus speculate that the hole concentration on the ladder sites may be increased considerably by the extra oxygen. Excess oxygen, on the other hand, may not even be necessary for superconductivity, as later reports[@Isobe; @nagata] are much less specific on excess oxygen content, and we note that crystals from the same growth rods as our Sr$_{2.5}$Ca$_{11.5}$Cu$_{24}$O$_{40.93}$ and Sr$_2$Ca$_{12}$Cu$_{24}$O$_{41}$ samples do exhibit superconductivity under high pressure. It may thus well be just the increased “2D” character of the electronic structure of the ladders, detected in NEXAFS as the increasingly similar H2 weight along $a$ and $c$, as well as the increased number of “pseudo-apical” O sites induced by the distortions, that in the end work together to support superconductivity in the highly Ca-substituted compounds under high pressure.
Conclusion
==========
Polarization-dependent NEXAFS has been performed on a large number of single-crystalline members of the Sr$_{14}$Cu$_{24}$O$_{41}$ family of layered compounds containing two-leg Cu$_2$O$_3$ ladders and CuO$_2$ chains. Studying different substitution levels of Y or La for Sr to change the total number of holes has shown in a direct and straightforward manner that for low and intermediate hole counts, all holes reside on the chain sites, and that this is independent of whether a further fraction of Sr is replaced by isoelectronic Ca as well. For the fully doped compound Sr$_{14}$Cu$_{24}$O$_{41}$, on the other hand, approximately one hole per formula unit is observed on the ladder unit, specifically in O $2p$ orbitals oriented along the $c$ axis on the legs. Studying the fully doped substitution series Sr$_{14-x}$Ca$_x$Cu$_{24}$O$_{41}$, we find the total hole count in the ladders to increase only marginally; however, a larger fraction of O $2p$ orbitals oriented along the $a$ axis (on the legs and possibly on the rungs of the ladder) is now contributing to the spectral weight.
Acknowledgements {#acknowledgements .unnumbered}
----------------
We greatly appreciate generous and valuable experimental help by J.-H. Park, S. L. Hulbert, P. Schweiss, T. Ivanova, E. Sohmen, D.-H. Lu and C. S. Gopinath. For illuminating discussions we are grateful to S.-L. Drechsler and M. Knupfer. Research was carried out in part at the National Synchrotron Light Source, Brookhaven National Laboratory, which is supported by the U. S. Department of Energy, Division of Material Sciences and Division of Chemical Sciences, under contract number DE-AC02-98CH10886. V. C. was supported by the Office of Naval Research.
Present address: Robert Bosch GmbH, Center of Research and Development, P. O. Box 1342, D - 72762 Reutlingen, Germany.
Present address: Experimental Facilities Division, Argonne National Laboratory, Argonne, IL 60439. K. Totsuka and M. Suzuki, ; N. Katoh and M. Imada, ; E. Dagotto and T. M. Rice, ; C. Kim, Z.-X. Shen, N. Motoyama, H. Eisaki, S. Uchida, and T. Tohyama, . T. M. Rice, S. Gopalan, and M. Sigrist, . M. Uehara, T. Nagata, J. Akimitsu, H. Takahashi, N. M[ô]{}ri, and K. Kinoshita, . E. M. McCarron III, M. A. Subramanian, J. R. Calabrese, and R. L. Harlow, ; K. Kato, ; K. Ukei, T. Shishido, T. Fukuda, ; K. Kato, E. Takyama-Muromachi, K. Kosuda, and Y. Uchida, . T. Siegrist, L. F. Schneemeyer, S. A. Sunshine, J. V. Wasczak, and R. S. Roth, ; T. Ohta, F. Izumi, M. Onoda, M. Isobe, E. Takayama-Muromachi, and A. W. Hewat, . M. Kato, K. Shiota, and Y. Koike, ; M. Isobe and E. Takayama-Muromachi, . S. A. Carter, B. Batlogg, R. J. Cava, J. J. Krajewski, W. F. Peck Jr, and T. M. Rice, . The energy shift between H1 and H2 was treated as a fit parameter as well but was assumed to be the same for all spectra. M. Matsuda, K. Katsumata, H. Eisaki, N. Motoyama, S. Uchida, S. M. Shapiro, and G. Shirane, . R. S. Eccleston, M. Uehara, J. Akimitsu, H. Eisaki, N. Motoyama, and S. Uchida, ; L. P. Regnault, J.P. Boucher, H. Moudden, J. E. Lorenzo, A. Hiess, U. Ammerahl, G. Dhalenne, and A. Revcolevschi, ; M. Matsuda, T. Yoshihama, K. Kakurai, and G. Shirane, . T. Osafune, N. Motoyama, H. Eisaki, and S. Uchida, . B. Ruzicka, L. Degiorgi, U. Ammerahl, G. Dhalenne, and A. Revcolevschi, ; B. Ruzicka, L. Degiorgi, G. I. Meijer, J. Karpinski, U. Ammerahl, G. Dhalenne, and A. Revcolevschi, . K. Magishi, S. Matsumoto, Y. Kitaoka, K. Ishida, K. Asayama, M. Uehara, T. Nagata, and J. Akimitsu, . M. Takigawa, N. Motoyama, H. Eisaki, and S. Uchida, . S. Ohsugi, K. Magishi, S. Matsumoto, Y. Kitaoka, T. Nagata, and J. Akimitsu, . D. E. Cox, T. Iglesias, K. Hirota, G. Shirane, M. Matsuda, N. Motoyama, H. Eisaki, and S. Uchida, . S. Tsuji, K. Kumagai, M. Kato, and Y. Koike, . S. Katano, T. Nagata, J. Akimitsu, M. Nishi, and K. Kakurai, . S. Pachot, P. Bordet, R. J. Cava, C. Chaillout, C. Darie, M. Hanfland, M. Marezio, and H. Takagi, . M. Isobe, T. Ohta, M. Onoda, F. Izumi, S. Nakano, J. Q. Li, Y. Matsui, E. Takayama-Muromachi, T. Matsumoto, and H. Hayakawa, . T. F. A. M[ü]{}ller, V. Anisimov, T. M. Rice, I. Dasgupta, and T. Saha-Dasgupta, . Y. Mizuno, T. Tohyama, and S. Maekawa, ; K. Magishi, S. Matsumoto, Y. Kitaoka, K. Ishida, K. Asayama, M. Uehara, T. Nagata, and J. Akimitsu, ; M. V. Abrashev, C. Thomsen, and M. Surtchev, . N. Motoyama, T. Osafune, T. Kakeshita, H. Eisaki, and S. Uchida, ; U. Ammerahl, G. Dhalenne, A. Revcolevschi, J. Berthon, H. Moudden, . U. Ammerahl and A. Revcolevschi, . K. Conder, S. Rusiecki, and E. Kaldis, . L. Tr[ö]{}ger, D. Arvanitis, K. Baberschke, H. Michaelis, U. Grimm, and E. Zschech, ; S. Eisebitt, T. B[ö]{}ske, J.-E. Rubensson, and W. Eberhardt, . J. J. Yeh and I. Lindau, At. Data Nucl. Data Tables [**32**]{}, 1 (1985). M. Knupfer, private communication. F. Sapi[ñ]{}a, J. Rodr[í]{}guez-Carvajal, M. J. Sanchis, R. Ib[á]{}[ñ]{}ez, A. Beltr[á]{}n, and D. Beltr[á]{}n, . R. Neudert, H. Rosner, S.-L. Drechsler, M. Kielwein, M. Sing, Z. Hu, M. Knupfer, M. S. Golden, J. Fink, N. N[ü]{}cker, M. Merz, S. Schuppler, N. Motoyama, H. Eisaki, S. Uchida, M. Domke, and G. Kaindl, . R. Neudert, M. Sing, M. Kielwein, Z. Hu, M. Knupfer, M. S. Golden, J. Fink, J. Karpinski, N. Motoyama, H. Eisaki, S. Uchida, G. Kaindl, C. Hellwig, and Ch. Jung, unpublished. To obtain spectra for the fitting procedure that consist solely of H and U, the higher-energy part of each spectrum was removed by approximating it, for the region of U and up to 534 eV, by a Gaussian which was then subtracted. For Sr$_5$Ca$_9$Cu$_{24}$O$_{41}$, H2 for [**E**]{}$\|$$a$ is less significant due to experimental difficulties: possible inhomogeneities and a slight misorientation will affect this contribution the most as it is the smallest. In Fig.s \[weight\] (b) and \[holes\] the respective data point is thus depicted in a gray shade. The asymmetric NEXAFS results for Sr$_{14}$Cu$_{24}$O$_{41}$ could also be viewed as a signature for a (SDW or CDW derived) gap, and a very rough estimate of its size could be obtained like this: a peak [*area*]{} in NEXAFS near threshold corresponds to the number of states (doped holes), and thus the peak [*amplitude*]{} right above threshold gives an approximation for the density of states near $E_F$. From Fig. \[alldata\] one reads off a total amplitude of $\approx 85$ Mb for H1, and thus the $a$,$c$ asymmetry of $\approx 12$ Mb$\cdot$eV (Fig. \[weight\]) would correspond to a shift of 12/85$\approx$0.14 eV between the $a$ and $c$ components of the spectral weight distributions. This value far exceeds, however, any spin gap reported for the chain ($\le 15$ meV). Y. Mizuno, T. Tohyama, S. Maekawa, T. Osafune, N. Motoyama, H. Eisaki, and S. Uchida, ; S. Tornow, O. Entin-Wohlman, and A. Aharony, . In Ref. , transfer integrals in the range 30 $\dots$ 80 meV along the chains of Li$_2$CuO$_2$ are presented, which are smaller by a factor of at least 7 compared to the value (0.55 eV) for the corner-sharing chain compound Sr$_2$CuO$_3$ where Cu-O-Cu hopping is not hampered. M. B. J. Meinders, H. Eskes, and G. A. Sawatzky, . T. Nagata, M. Uehara, J. Goto, N. Komiya, J. Akimitsu, N. Motoyama, H. Eisaki, S. Uchida, H. Takahashi, T. Nakanishi, and N. Mori, .
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We study the alignments between the angular momentum of individual objects and the large-scale structure in cosmological numerical simulations and real data from the Sloan Digital Sky Survey, Data Release 6. To this end we measure anisotropies in the two point cross-correlation function around simulated halos and observed galaxies, studying separately the 1- and 2-halo regimes. The alignment of the angular momentum of dark-matter haloes in $\Lambda$CDM simulations is found to be dependent on scale and halo mass. At large distances (2-halo regime), the spins of high mass haloes are preferentially oriented in the direction perpendicular to the distribution of matter; lower mass systems show a weaker trend that may even reverse to show an angular momentum in the plane of the matter distribution. In the 1-halo term regime, the angular momentum is aligned in the direction perpendicular to the matter distribution; the effect is stronger than for the 1-halo term and increases for higher mass systems.
On the observational side, we focus our study on galaxies in the Sloan Digital Sky Survey, Data Release 6 (SDSS-DR6) with elongated apparent shapes, and study alignments with respect to the major semi-axis. We study five samples of edge-on galaxies, the full SDSS-DR6 edge-on sample, bright galaxies, faint galaxies, red galaxies, and blue galaxies (the latter two consisting mainly of ellipticals and spirals respectively). Using the 2-halo term of the projected correlation function, we find an excess of structure in the direction of the major semi-axis for all samples; the red sample shows the highest alignment ($2.7\pm0.8\%$) and indicates that the angular momentum of flattened spheroidals tends to be perpendicular to the large-scale structure. These results are in qualitative agreement with the numerical simulation results indicating that the angular momentum of galaxies could be built up as in the Tidal Torque scenario. The 1-halo term only shows a significant alignment for blue spirals ($1.0\pm0.4\%$), consistent with the 1-halo results from the simulation but with a lower amplitude. This could indicate that even though the structure traced by galaxies is adequate to study large-scale structure alignments, this would not be the case for the inner structure of low mass haloes, $M\leq 10^{13}h^{-1}M_{\odot}$, an effect apparently more important around red $g-r>0.7$ galaxies.
author:
- |
Dante J. Paz$^1$, Federico Stasyszyn$^{1,2}$, Nelson D. Padilla $^{3}$\
$^1$ Instituto de Astronomía Teórica y Experimental IATE (UNC-CONICET),\
Observatorio Astronómico Córdoba, Francisco N. Laprida 922, Córdoba, Argentina.\
$^2$ Max-Planck-Institut fuer Astrophysik MPA, Karl-Schwarzschild-Str. 1, Garching, Germany.\
$^3$ Departamento de Astronomía y Astrofísica, Pontificia Universidad Católica de Chile,\
Vicuña Mackenna 4860, Santiago 22, Chile.\
date: 'Accepted 25 June 2008. Received 24 June 2008; in original form 25 April 2008'
title: 'Angular momentum$-$Large-scale structure alignments in $\Lambda$CDM models and the SDSS'
---
Cosmology: Galaxies – Angular Momentum – Large Scale Structure of the Universe
Introduction
============
In the current paradigm of large-scale structure formation in the Universe it is well established that the angular momentum of dark-matter (DM) halos is determined early in the history of the Universe by the gravitational interaction between the quadrupole of the collapsing proto-halo region and the surrounding matter. This was first formulated for the hierarchical theory of structure formation by @White1, @Doro1, and @Peebles1, and it is claimed that this is a natural consequence of the perturbative treatment of the gravitational instability scenario (@Porciani1). Several inherent properties to the tidal torque process (Tidal Torque Theory, TTT) have been noticed and studied in the literature. For instance it is expected from TTT, and also supported to some degree by N-body simulations, that the angular momentum starts growing linearly with time [@Catelan2; @Peebles1; @White1], and that later on this process looses efficiency after the halo turn-around. This is due to the reduction of halo inertia by collapse, and to the continued expansion of the neighboring matter responsible for the tidal torque.
There are several analytic studies of the TTT scenario [@White1; @Doro1; @Peebles1; @Catelan3], as well as semi-analytic approaches [@Catelan1], and fully non-linear work using cosmological simulations [@Barnes1; @Suger1; @Porciani1; @Porciani2; @Peirani1; @Nagashima1]. The agreement is that the resulting halo population is poorly rotationally supported; the velocity dispersion at virialisation is found to be the most important mechanism for halo equilibrium. As a result, the distribution of spin parameters[^1] is log-normal which peaks at quite low values, $\lambda_{med} \simeq 0.035$ [@Bullock1; @Gardner1; @Onghia1; @Maccio1; @Bett1]. Unfortunately it is not possible to obtain a direct measurement of spin parameters in actual galaxies. Despite this, @Hernandez1 used indirect methods to derive distributions of galaxy spin parameters consistent with theoretical predictions.
From an observational point of view, it can be easier in some cases to determine the direction of the angular momentum of a given galaxy rather than its amplitude. There has been a recent increase in the interest concerning the possibility of galaxy alignments with the large scale structure. Such studies are important to test the already well developed TTT, and to asses the possibility that such correlations affect the results of cosmic shear measurements. Regarding the former, TTT produces noisy predictions due to the non-linear nature of the accretion processes of clumpy matter [@Vitvitska1; @Porciani1], on scales below the coherent scale of flows predicted by linear theory. The second tests for the effects on cosmic shear studies used to analyse weak lensing, may answer whether there is an intrinsic alignment in the structure traced by galaxies that could result in a spurious contribution to the shear power spectrum [@Hirata1; @Bridle1; @Takada1; @King1; @King2; @Heymans1; @Heymans2; @Brown1; @Catelan4; @Crittenden1; @Croft1; @Jing1]. Several authors have searched for galaxy alignments in the super-galactic plane [@Vauco1], extending out to several tens of megaparsecs from our Galaxy, and more recently on large galaxy surveys [@Hirata1; @Hirata2; @Lee1; @Lee2; @Mandel1]. @Lee1 claim a direct evidence of an alignment between galaxy position angle in 2MRs and the tidal field deduced from galaxy positions.
An interesting approach to the analysis of galaxy alignments is to use the radial direction to the centres of voids as characteristic directions in the large scale structure. It is well known that as they evolve, large voids tend to be rounder, a behaviour which is opposite to that shown by bound structure. Negative density perturbations in the initial fluctuation field that will later become voids, are characterised by a decreasing asphericity as the expansion of the Universe proceeds, as is shown by @Sheth1 (and references therein). Taking advantage of this particular characteristic of voids, it is possible to study the whether galaxy or halo angular momenta are aligned with the void shells just by considering the angle with respect to the direction to the void centre. This is the approach used by @Trujillo1 applied to the SDSS and 2dFGRS galaxy redshift surveys, and by @Brunino1, @Cuesta1 and @Patiri1 to numerical simulations. @Trujillo1 find with a high level of confidence that spiral galaxies located on the shells of the largest cosmic voids have rotation axes that lie preferentially on the void surface, in agreement with the simulation results of @Brunino1 [@Cuesta1; @Patiri1]. Other more general methods that also take into account alignments with large scale filaments in numerical simulations show similar results [@Aragon1; @Hahn1; @Hahn2].
In this paper we present results from a study of alignments between individual objects and their surrounding structure in quantitative concordance with TTT predictions and previous observational results. We use a novel approach to study alignments with the large-scale structure using as a reference the galaxy/dark matter halo angular momentum. Using numerical simulations we study alignments predictions in a Cold Dark Matter model with a cosmological constant ($\Lambda$CDM), and compare such predictions with the results obtained from the Sloan Digital Sky Survey, Data Release 6 (SDSS-DR6, Adelman-McCarthy et al., 2007). To this end we use the galaxy two point correlation function applied both, in its three-dimensional definition to the full numerical simulation and in its projected version to mimic the observational limitations. The latter allows us to determine that we should be able to detect the expected low signal alignments, even in small galaxy samples. This allows us to measure the alignment of galaxies and halos over wide ranges in mass and separation, and also to quantify the dependence of this alignment on halo mass, galaxy colour and luminosity.
This paper is organised as follows. Section 2 presents the numerical simulation and the statistical tools we will use in this work; it also presents an analysis of the impact of the observational biases on the detection of alignments. Section 3 contains the analysis of observational samples of galaxies from the Sloan Digital Sky Survey. The comparison between models and observations as well as the discussion of the results is presented in Section 4, and Section 5 contains the final conclusions from this work.
Anisotropies in the 3D Correlation function
===========================================
In this section we analyse the three-dimensional two-point correlation function, using cluster-particle pairs, measured using $\Lambda$CDM numerical simulations, taking into account the orientation with respect to the centre halo angular momentum. To this end, we have run two periodic simulations of $500$ and $60$ $h^{-1 }$Mpc computational box sizes, for a flat low-density Universe, with a matter density $\Omega_{\rm m}=1-\Omega_{\Lambda}=0.3$, Hubble constant $H_{\circ}=70$ km s$^{-1}$ Mpc$^{-1}$, and normalisation parameter $\sigma_{8}=0.8$. The particle resolutions are $m_{\rm p}\ge
7.2\times 10^{10}\,h^{-1}\,M_{\odot}$ for the large simulation, and $m_{\rm p}\ge 1.2\times 10^{8}\,h^{-1}\,M_{\odot}$ for the small one. The identification of particle clumps has been carried out by means of a standard friends-of-friends algorithm with a percolation length given by $l=0.17$ $\bar{\nu}^{-1/3}$, where $\bar{\nu}$ is the mean number density of DM particles. Both simulations have been performed using the first version of the GADGET code developed by Springel et al. (2001). We consider halos with at least $30$ and $100$ particles in the large and small simulations, respectively, in order to ensure an accurate measurement of the halo angular momentum.
(430,250) (-40,0) (210,0)
The spatial halo-particle cross-correlation function $\xi(r)$ which we will use in this section, measures the excess probability with respect to a random distribution, that a DM particle will reside at a distance $r$ away from a given halo centre, in the volume element $dV$, $$dP=\bar{\nu}\left[ 1+\xi(r)\right] dV. \label{definicion}$$ A standard method to measure $\xi(r)$ in a simulation box is to count pairs at a given distance bin, and then normalise this number by the expected number of pairs for a constant number density of objects. In our measurements we perform these counts stacking pairs depending on the angle between their position relative to the centre halo and its angular momentum. We consider tree cases, (i) using all pairs regardless of this angle (spherical shell volumes, isotropic case), (ii) using all pairs subtending an angle from the angular momentum less than a given threshold $\theta_1$ (parallel case), and (iii) adding pairs at lower inclinations than a limit angle $\theta_2$ from the perpendicular plane to the angular momentum (perpendicular case). We choose the threshold angles, for the parallel and perpendicular cases mentioned above ($\theta_1$ and $\theta_2$, respectively), so that the volumes in each case are the same. This can be achieved by setting, $\sin(\theta_2)=1-\cos(\theta_1)=\chi$, and choosing a value for the threshold parameter $\chi$. Selecting $\chi\leq0.5$ implies angles $\theta_1 \leq 60^o$ and $\theta_2 \leq 30^o$. Throughout this paper we use $\chi=0.5$.
Errors in the correlation function are estimated using the jacknife technique, with a total of $50$ subsamples for the numerical simulation ($20$ jacknifes for the SDSS data analysis). We find that our errors are stable for this number of jacknife subsamples (we tested from $20$ to $100$) and does not produce underestimations. In the case of quotients between correlation functions, we apply the jacknife technique to the actual measurement of the ratio. We adopt the jacknife technique since it has been reported to provide similar error estimates to those obtained from independent samples (see for instance, Padilla et al., 2004).
Simulation results {#simresult}
------------------
The analysis of alignments of structure with the angular momentum of DM haloes presents a first difficulty in that in order to obtain a reliable and stable direction for the halo spin, it is often necessary to use a large number of particles per halo. We find that using $30$ particles provides reasonable results, but this imposes a rather high lower limit to halo masses of $2.2\times
10^{12}\,h^{-1}\,M_{\odot}$ for the large simulation, and $3.6\times 10^{9}\,h^{-1}\,M_{\odot}$ for the smaller one. We analised the distribution of angular momentum parameters and find that the reported high $\lambda$ tail (Bett et al., 2007) due to low number of particles per halo, is not present in any of our samples of DM haloes. We notice that the clustering signal in the small simulation on large scales may be affected by the small box size.
From this point on, we will centre most of our analysis on samples of DM haloes with different masses. From the small simulation we draw four DM halo subsamples separating haloes of different mass starting at a minimum of $100$ particles or $M\geq 1.2 \times 10^{10}\,h^{-1}\,M_{\odot}$ (the lowest mass sample considered includes haloes of up to 300 particles each; we do not use haloes of lower masses even though we are still above the minimum number of particles we consider safe to consider). Nine samples are drawn from the large simulation and contain DM haloes with more than $30$ particles. The subsample with the most massive haloes starts at $1000$ particle members corresponding to $M\geq 7.2\times 10^{13}\,h^{-1}\,M_{\odot}$.
=0.5
The left panels of Figure \[fig:3D\] show the cross-correlation function between low mass DM haloes and DM particles (top-left) in the directions parallel and perpendicular to their angular momenta, and in all directions (diamonds, triangles and squares, respectively). The lower panel shows the ratios between these three estimates and the correlation function considering all neighbors; their relative amplitudes help identify alignments in either direction. As can be seen, the relative amplitudes of the correlation function when using neighbors in the directions parallel and perpendicular to the angular momentum of the DM halo are different; for small separations $R<1$h$^{-1}$Mpc, there is an excess of correlation in the direction perpendicular to the angular momentum of up to a $20\%$ effect (between correlations parallel and perpendicular to the angular momentum, lower panel), whereas for larger separations the alignment diminishes and shows a tendency to reverse (although, as can be seen in the figure, errors are larger at these separations); we come back to this later in this section where we use a higher signal estimator for alignments.
When using higher mass DM haloes, the effect is also present with an even higher difference and statistical significance between the structure in the direction of the angular momentum and perpendicular to it (Right panels of Figure \[fig:3D\]). In this case, neighbors in the direction perpendicular to the angular momentum show more correlation, and there is a clear tendency for this alignment to decrease with separation, which goes from almost a $50\%$ effect on relative amplitudes (lower panel) at small separations to about $<5\%$ at large separations. Note that the alignment does not change at large separations as seemed to be the case for low mass haloes, even at $R=20$h$^{-1}$Mpc compared to $R=5$h$^{-1}$Mpc analised in the low mass halo sample.
Notice that the halo-particle cross-correlation functions in Figure \[fig:3D\] show the characteristic 1- and 2-halo terms reported in several previous measurements of both observational (e.g. Zehavi et al., 2004, Cooray, 2005) and numerical simulations (Zheng et al., 2004); these two regimes are separated in the figure by the vertical dashed lines. This separation is obtained by measuring the scale at which the correlation function is seen to shift between these two regimes (there is a local minimum in the first derivative of the correlation function). It can be seen that the alignment signal shows a clear transition from the 1- to the 2-halo regime; in particular, low mass haloes show indications of a different angular momentum$-$large-scale alignment behaviour in the 2-halo regime, suggesting an important relationship between alignment and halo term. This can be interpreted as different physical mechanisms producing the alignment in the 1- and 2-halo terms.
We make quantitative estimates of the alignments for the 1- and 2-halo terms separately from the cross-correlation functions shown in Figure \[fig:3D\] as well as from the other samples of DM haloes with intermediate masses (for a total of 13 samples of varying halo masses). We only consider pairs that are separated by a distance at least as large as the transition between regimes for the 2-halo term, and at shorter separations for the 1-halo calculations.
In order to avoid effects of covariance between the correlation function bins to some extent, we make global 1- and 2-halo term estimates of the ratios between the correlation functions in the directions perpendicular and parallel to the angular momentum of haloes, using all the halo-particle pairs separated by distances within the 1- and 2-halo ranges (i.e. we do not use the narrow bins in $\log_{10}(r)$ used to calculate the cross-correlation functions). We acknowledge that traces of covariance will still be present between the 1- and 2-halo term ratios. We present these results in Figure \[fig:rat3d\], where triangles show the 1-halo term ratio between correlation functions for neighbors in the directions perpendicular and parallel to the halo angular momentum; Diamonds correspond to the 2-halo term. The horizontal long dashed line at $\xi_\parallel/\xi_\perp = 1$ corresponds to the isotropic case. The data-points at masses lower than $10^{12}h^{-1}M_{\odot}$ correspond to the small numerical simulation, and show larger errors due to both, the smaller simulated volume, and the lower amplitude of clustering characterising low mass haloes.
As can be seen, the 1-halo term always shows more structure in the direction perpendicular to the angular momentum, and more so for larger masses, whereas the 2-halo term shows a similar behaviour with the possible exception of low mass haloes, whose angular momenta tend to point in the direction of the large-scale structure (although with a low statistical significance). The alignments are detected at about a $5-\sigma$ level for the higher mass halo subsamples drawn from the large simulation. As in the 2-halo regime, low mass haloes show considerably higher errors.
In the case of low mass halos ($M\simeq 10^{10}h^{-1}M_{\odot}$) in the 2-halo term regime (diamonds in Figure \[fig:rat3d\]), the alignment signal is the opposite to that of higher mass haloes (a $10\pm5\%$ effect). This anti-alignment signal fits in well with the overall tendency for this quotient $\xi_\parallel/\xi_\perp$ to increase with mass, which can be fit by a log-linear relation which crosses the unit value at $M\simeq10^{12}h^{-1}M_{\odot}$. This mass limit suggest a change in the alignment signal in qualitative agreement (regarding the alignment direction) with results on simulations by @Hahn1, with masses above and below the typical mass scale for collapse $M_*=5.5\times10^{12.5}h^{-1}M_{\odot}$ at $z=0$. The best-fit log-linear relation for the 2-halo term results is shown in Figure \[fig:rat3d\] (dash dotted line), and corresponds to, $$\xi_\parallel/\xi_\perp = (0.95\pm0.01) + (0.082\pm0.008) *[log_{10}(M)-11.5]\,\,$$ For the 1-halo term, the dependence of alignment on mass is well adjusted by (dashed line on Figure \[fig:rat3d\]), $$\xi_\parallel/\xi_\perp = (0.95\pm0.01) + (0.082\pm0.008) *[log_{10}(M)-11.5]\,\,$$
The 2-halo term effect can be understood within the framework of TTT, specially for the high mass haloes, which readily show that their angular momentum is perpendicular to the present-day mass distribution. It seems that for high enough mass haloes, the local large-scale structure still describes the original tidal field. In the case of low mass haloes, the alignment seems to reverse. This result can be understood by an intuitive non-linear effect beyond the abilities of TTT, for example if these haloes have formed recently in filaments, from the merger of material arriving from voids surrounding these elongated structures.
The 1-halo term results indicate that regardless of halo mass, the structure within dark-matter haloes is preferentially aligned with the plane perpendicular to the angular momentum, with a higher statistical significance and signal than the 2-halo term.
(430,250) (-35,0) (215,0)
(350,420) (15,240) (0,0)
Simulating observational biases in the numerical simulation
-----------------------------------------------------------
It can be difficult to estimate the angular momentum of real galaxies. In the case of the SDSS-DR6, which we use in this work, the only information that can allow us to obtain an estimate of the direction of the angular momentum is the photometry of galaxy images, specially for spiral galaxies but also for ellipticals. We assume that spiral discs or flattened spheroids are perpendicular to the angular momentum of galaxies and therefore, edge-on galaxies (either disks or spheroidals) will be more likely to have their angular momentum on the plane of the sky. Then, one can calculate the cross-correlation function for neighbors in the projected direction parallel and perpendicular to the direction of the angular momentum of the centre galaxies, in redshift slices around them, assuming all these neighbors to be at the same distance than the centre galaxy. This slice needs to be defined so that it avoids projecting large amounts of foreground and background structures while still allowing enough numbers of neighbors to obtain good statistics.
In order to assess whether the alignments found in the previous subsection are identifiable in observational catalogues analised this way, we reproduce the procedure mentioned in the previous paragraph in our analysis of the numerical simulation which consists on projecting the structure onto the plane of the sky, assuming all objects within $\Delta v=750$km/s from the centre haloes to be exactly at the same distance from the observer.
Therefore we only use information from two cartesian coordinates in the simulation, $x$ and $y$, and transform the third, $z$, into a velocity component by multiplying $z$ by the Hubble Constant and adding the $z$ component of the peculiar velocity. In order to complete the emulation of redshift space effects suffered by observational surveys, we apply this transformation to the simulation particles as well as to the halo catalogue. We then select “edge-on” haloes from the simulation, by requiring that their angular momentum is at an angle less than $60^o$ from the sky, represented by the $x-y$ plane [^2]. We can then study two projected correlation functions centred in these objects using DM particles, (i) a correlation function calculated using pairs that subtend an angle with the projected angular momentum of less than $30^o$, which we call the parallel projected correlation function ($\omega_{||})$, and (ii) a perpendicular projected correlation function ($\omega_{\bot}$) obtained by counting pairs that subtend an angle greater than $60^o$ from the projected angular momentum.
Figure \[fig:projsim\] shows the results from this procedure. The left panels show the results for low mass haloes in the small simulation (projected correlation functions in the top-left, and ratios in the bottom-left), whereas the right panels show the results for high mass haloes in the large simulation. As can be seen, these results are qualitatively consistent with the alignment signal found in the previous subsection using the full three-dimensional correlation function, which indicates that projected correlation functions are a suitable tool to detect this effect. The overall effect has a lower amplitude; high mass haloes show a ratio between structure in the directions perpendicular and parallel to their angular momenta of about a $30\%$ excess (for the 3D case, this was above $50\%$) and low mass haloes show a $10\%$ effect (compared to $20-30\%$), which makes this detection a slightly more difficult task.
Observations
============
Galaxies with spiral and flattened spheroidal shapes projected onto the sky give a handle on their orientation with respect to the line of sight and, therefore, it is possible to estimate in a statistical sense the direction of their angular momentum. Spiral galaxies with round projected shapes are likely discs seen face-on and therefore their angular momenta point in the direction perpendicular to the plane of the sky; elongated projected shapes indicate discs seen edge-on, and therefore their angular momenta point in the direction perpendicular to the long axis of the ellipsoid. On the other hand, elliptical galaxies with round projected shapes can either be nearly intrinsically spherical or flattened spheroids seen face-on; however, elongated ellipticals are more likely flattened spheroids seen edge-on since recent results on intrinsic elliptical shapes indicate that these are more commonly oblate rather than prolate spheroids (e.g. Padilla & Strauss, 2008). Therefore, elongated ellipticals are likely to be also characterised by angular momenta pointing in the direction of the plane of the sky roughly perpendicular to the major axis of the ellipsoid, as is the case for edge-on spiral galaxies. These last two types of galaxies will be the centre of the following analysis. We notice that the amplitude of the angular momentum of flattened spheroidals is likely to have a low amplitude relative to that of spiral galaxies.
The sample we selected to carry out our analysis is the spectroscopic SDSS-DR6, which contains a total of $\simeq 580,000$ galaxies with spectroscopic redshifts and photometry in five bands, $u,g,r,i,z$, as well as parameters determining the projected galaxy shapes consisting of the two semi-axes and the position angle of the ellipse that once convolved with the PSF (seeing) provides the best match to the photometric image of each individual galaxy. In the remainder of this paper, we only consider SDSS-DR6 galaxies within the redshift range $0.02<z<0.09$.
Sample Members $n/10^{-3}$h$^{-3}$Mpc$^3$ Log$_{10}(M_{\rm host}/$h$^{-1}M_{\odot})$
-------- --------- ---------------------------- --------------------------------------------
O1 132000 $10.4\pm0.8$ $12.50^{+0.30}_{-0.54}$
O2 59300 $6.50\pm0.12$ $13.06^{+0.19}_{-0.27}$
O3 72700 $5.50\pm0.12$ $11.40^{+0.56}_{-0.80}$
O4 10900 $7.15\pm0.17$ $12.19^{+0.45}_{-1.17}$
O5 1400 $1.10\pm0.20$ $12.67^{+0.28}_{-0.48}$
: Observational samples from the SDSS-DR6 spectroscopic main galaxy sample. The first and second column indicate the sample name and number of galaxies in them, the third column indicates the number density, and the fourth column the host-halo mass.[]{data-label="table"}
We study the alignment signal in the observational data for different subsamples defined by galaxy colour and luminosity in order to detect trends that can be associated to the results from the numerical simulation. The subsamples we study are sample O1, consisting of all the edge-on galaxies with axial ratio $b/a<0.7$ in the SDSS-DR6; O2, red galaxies with $g-r>0.7$; O3, blue galaxies with $g-r<0.7$; O4, faint galaxies with $M_r>-19.5$; and O5 composed by bright galaxies with $M_r<-19.5$. The selection of subsamples is done so as to ensure a number of target galaxies to allow good statistics ($> 1000$); the number of galaxies in each sample is shown in Table \[table\], where we also show the galaxy number density calculated out to the redshift of completeness of each sample.
In order to characterise in a quantitative way the subsamples selected from the SDSS-DR6, we calculate the median mass of $\Lambda$CDM haloes predicted to show the same 2-halo regime clustering amplitude as the galaxies in each subsample. We follow the procedure used in several previous works (cf. Croft et al. 1999, Padilla et al., 2001) consisting on calculating the auto-correlation function of galaxies in two coordinates, one parallel ($\sigma$) and one perpendicular ($\pi$) to the line of sight, $\xi(\sigma,\pi)$, and integrating over the $\pi$ direction to avoid the effect of redshift space distortions. The result from this integration is the projected correlation function (see for instance, Croft et al., 2001), $\Xi(\sigma)$, $$\Xi(\sigma)=2\int_{\pi_{min}}^{\pi_{max}}\xi(\sigma,\pi) d\pi,
\label{eq:Xi}$$ which in turn can be inverted to obtain the real-space correlation function via $$\begin{aligned}
&\xi(r)=\frac{-1}{\pi}\sum_{j\geq i}\frac{\Xi(\sigma_{j+1})-\Xi(\sigma_{j})}
{\sigma_{j+1}-\sigma_{j}}
\ln{\left(\frac{\sigma_{j+1}+\sqrt{\sigma_{j+1}^{2}-
\sigma_{i}^{2}}}{\sigma_{j}+
\sqrt{\sigma_{j}^{2}-\sigma_{i}^{2}}}\right)},&\end{aligned}$$ where the sum is performed over the bins in $\sigma$ where the projected correlation function has been calculated. The masses of host haloes can then be obtained by finding the bias between the resulting real-space correlation functions and the matter correlation functions in a $\Lambda$CDM cosmology (both in the 2-halo regime), which we obtain by Fourier transforming the non-linear matter power spectrum (from Peacock & Smith, 2000) for the same cosmological parameters used in the numerical simulations. With an estimate of the bias, the mass of the host haloes can be found by using the Sheth, Mo & Tormen (2001) formalism. Note that this procedure is equivalent to measure the real-space halo-halo correlation function for haloes of different masses (Sheth, Mo & Tormen, 2001, Padilla & Baugh, 2002), with the advantage of avoiding large uncertainties from small samples of haloes restricted to narrow ranges of halo mass in the simulation. We show one example of this procedure in Figure \[fig:xir\], where we find the mass of haloes with an equivalent clustering amplitude to the sample of red galaxies in the SDSS-DR6 (O2). In this figure, the top panel shows the auto-correlation function, $\xi(\sigma,\pi)$, of O2 galaxies, and the lower panel the resulting real-space correlation function (symbols) and the best fit $\Lambda$CDM curve. From this analysis we find that red galaxies in the SDSS-DR6 (with $b/a<0.7$) are consistent with haloes of a median mass of $\log_{10}(M/h^{-1}M_{\odot})=13.06^{+0.19}_{-0.27}$. This provides us with a quantitative way to compare the SDSS-DR6 subsamples to the halo samples in the numerical simulations. Also, in order to infer this scale for our observational samples, we will use the dependence of the 1- to 2-halo term transition on DM halo mass measured in our the numerical simulations.
The other SDSS-DR6 galaxy samples are fitted by the correlation function of DM haloes of masses indicated in Table \[table\]. Notice that our study covers over one and a half decades in the host-halo mass of SDSS-DR6 galaxies.
(130,498) (0,0) (7,173) (7,340)
As we demonstrated in the previous section both, the large scale structure and alignments surrounding galaxies, can be studied using the projected correlation function, $\omega(\sigma)$, measured in terms of the projected separation, $\sigma$. We now apply this method to our samples of edge-on galaxies, and compare the outcome from using tracers in the direction parallel and perpendicular to the inferred angular momentum, to detect differences between the structure along these two directions.
The projected correlation functions are calculated using galaxies in a given subsample, cross-correlated to the full spectroscopic SDSS-DR6 catalogue. As in the 2-dimensional analysis carried out in the numerical simulations, we calculate the cross-correlation for edge-on galaxies selected so that $b/a<0.7$. Note that in the numerical simulation we used an angle between the halo angular momentum and the line of sight of $60^o$, which in the case of a perfect thin disk is equivalent to $b/a=0.5$, but is larger for a thick disk as is the case of the intrinsic shapes of spiral galaxies or flattened ellipsoids (Padilla & Strauss, 2008).
Figure \[fig:obsxi\] shows the measured correlation functions for Edge-on galaxies for the three subsamples that show significantly different signatures of alignments with the angular momentum. These are the full sample of edge-on spirals, O1 (top panel), the red galaxy sample, O2 (middle panel), and the blue galaxies, sample O3 (bottom panel). As in the results for the numerical simulation, triangles show the correlation with galaxies in the direction of the angular momentum, diamonds in the perpendicular direction, and squares, when using all neighbors. A first glance at the shapes of these correlation functions indicates that there is not as clear a transition between the 1- and 2-halo terms as in the results from the numerical simulations (see Fig. \[fig:projsim\]); in these samples the main effect is a transition to a roughly constant slope for the $\log(\xi)$ vs. $\log \sigma$ relation. This may be an indication that the internal structure of haloes is not traced in the same way by galaxies in the SDSS-DR6 and DM particles in the $\Lambda$CDM model. Regarding the alignments, it can be seen that the three subsamples show the large-scale structure preferentially aligned with the direction perpendicular to the angular momentum, specially for the 2-halo regime, as was found in the cosmological simulation. The difference between the two correlation functions is roughly a $2-\sigma$ detection over the range of scales defined by $2<r/$h$^{-1}$Mpc$<20$ and corresponds to about a $2-4\%$ effect.
The analysis of the faint and bright samples (O4 and O5, respectively) provide alignment results in quantitative agreement with the full galaxy sample, O1, with results within a $1-\sigma$ difference. Therefore, we will not show results from these two samples in the remainder of this work.
A more quantitative comparison between observational and simulation results is presented in the following section.
Discussion
==========
In order to compare quantitatively the alignments found in the observations and in the numerical simulation, we use the transition scale from the 1- to the 2-halo regimes inferred in the previous section to determine the degree of alignment found in these two regimes separately. As in section \[simresult\], we estimate ratios between the correlation functions in the directions perpendicular and parallel to the estimated galaxy angular momentum direction, using all pairs within the 1- and 2-halo regimes. This can then be directly compared to the results from the numerical simulations for projected correlations.
Figure \[fig:2drat\] shows the ratios between projected correlation functions in the directions parallel and perpendicular to the angular momentum as a function of halo mass. The results from the numerical simulations are shown as diamonds, and the results from the observational samples at the corresponding host-halo masses are shown as filled symbols. We show results for samples O1 (full SDSS-DR6, filled squares), O2 (red galaxies, filled upward pointing triangles), and O3 (blue galaxies, filled downward pointing triangles). The top panel shows the results from the analysis of the 1-halo term, the lower panel from the 2-halo term. For the 2-halo term regime the resulting alignment signals for observational samples are $1.8\pm0.5\%$ for the full sample O1, $2.7\pm0.7\%$ for the red galaxy sample 02, and $0.8\pm0.4\%$ for the blue galaxy sample 03. The 1-halo term regime shows alignments of $0.1\pm0.5\%$, $0.1\pm0.5\%$, and $1.0\pm0.4\%$ for samples O1, O2, and O3, respectively. As can be seen, the alignment found in the 2-halo term is consistent between observational samples and the numerical simulation (differences are at a maximum a $1-\sigma$ effect).
(180,440) (0,0) (0,220)
This agreement indicates that TTT in combination with hierarchical clustering are a viable theory since not only the angular momentum of galaxies points in the direction perpendicular to the surrounding large-scale structure, but also the amplitude of the effect is similar to what is found in the numerical simulations. It should be noticed that this would also indicate that the direction of the angular momentum in galaxies in these samples is tracing the angular momentum of the host halo. The results for the 1-halo term, on the other hand, show that galaxies follow a different behaviour than DM haloes as was suspected from the lack of a clear transition to the 1-halo term in the shape of the SDSS projected correlation functions; at the range of masses corresponding to the observational samples, the simulated haloes show very significant alignments, whereas the only sample of galaxies that a shows non-zero alignment is that composed by blue galaxies (although still lower than the simulation data, at a regime of low expected alignments). The combination of the 1- and 2-halo term results could indicate that the general galaxy population in the SDSS-DR6 which our samples are cross-correlated to, can be used as good tracers of the large-scale structure (i.e. have not suffered important changes to deviate from the angular momentum of their host haloes) but would fail to serve as indicators of the internal distribution of matter in their host DM haloes; different astrophysical processes could produce this effect, specially for such low $M\simeq 10^{13}h^{-1}M_{\odot}$ halo masses[^3]. This result is particularly interesting since blue galaxies do show traces of an alignment in the same direction as the numerical simulation, which has a number of possible explanations. For instance, the angular momentum of red galaxies could be subject to a reshuffle that adds enough noise to loose its alignment with the internal halo structure while leaving the large-scale signal almost intact, an effect could also involve a positional reshuffle within the halo structure; another possibility is that red galaxies are able to rearrange their companion galaxies inside DM haloes. However, a more quantitative analysis is required in order to find the true underlying cause for this behaviour. We acknowledge the possibility that observational biases not included in our analysis of the numerical simulation are still affecting these measurements.
Conclusions
===========
We study the alignments between the angular momentum of individual objects and the large-scale structure in cosmological numerical simulations and in the SDSS-DR6. The angular momentum of DM haloes in $\Lambda$CDM simulations is found to be preferentially oriented in the direction perpendicular to the distribution of matter in the range of separations corresponding to the 1- and 2-halo terms. These results are in agreement with the Tidal Torque Theory.
We find that more massive haloes show a higher degree of alignment, and that the 1-halo term shows a much higher alignment than the 2-halo term. We find a maximum 1-halo alignment of $\sim40\%$ for haloes of $\simeq 10^{14}h^{-1}M_{\odot}$; at low masses, the 1-halo signal seems to tend to a constant value of $20\%$. The 2-halo term alignment shows a maximum signal of $\simeq 15\%$ at high masses, and possible inversion at low masses where the angular momentum might even preferentially point in the direction of the distribution of mass on large scales. The latter would be consistent with low mass groups formed not too long ago from mass distributed in the surrounding areas of the already-formed filaments or walls.
The alignment of the angular momentum with the large-scale structure found for low mass systems is in agreement with previous works by [@Trujillo1] on redshift surveys, and by @Brunino1 [@Patiri1; @Cuesta1; @Aragon1; @Hahn1; @Hahn2] on simulations. The main difference with these previous studies relies in that we quantify the strength of the alignment in terms of the two-point correlation function, as a function of halo mass. In particular, we find that a log-linear relation fits the dependence of alignment on mass, for both, the 1- and 2-halo term regimes. For the latter we observe a turning point between an angular momentum pointing in the direction perpendicular to the structure, to no-alignment, and then to a regime where the angular momentum is contained by the plane defined by the surrounding structure. This occurs at a mass $\simeq M_*$, in qualitative agreement with previous studies [@Bailin1; @Hahn1; @Hahn2].
Combining our results for the 1- and 2-halo terms for haloes in the numerical simulation, namely that the angular momentum of haloes is perpendicular to the bulk of the matter within haloes (1-halo term), and also to the surrounding large-scale structure (for masses $>M^*$), our findings are in agreement with previous works who find that the major axis of haloes is usually aligned with the large-scale structure (Gottl[ö]{}ber & Turchaninov 2006, Bett et al. 2007, Faltenbacher et al. 2002; Kasun & Evrard 2005; Bailin & Steinmetz 2005; Colberg, Krughoff & Connolly 2005; Altay, Coldberg & Croft 2006; Basilakos et al. 2006; Ragone-Figueroa & Plionis 2007, and references therein). Regarding the 1-halo term signal, our results are also in qualitative agreement with recent results by @Knebe1 on the alignments of substructure within dark matter halos.
We use numerical simulations to reproduce the procedure that can be applied to observational samples using projected correlation functions. We find that the alignment signal diminishes considerably but that it can still be measured.
Therefore, we study edge-on galaxies in the SDSS-DR6, and assign the direction perpendicular to the major semi-axis as the direction of the angular momentum. Since our samples are restricted to objects with flattened apparent shapes ($b/a<0.7$), these include both edge-on spiral galaxies and flattened spheroids. We are able to detect alignments in all our galaxy samples. These include the full SDSS-DR6 edge-on galaxies, a sample composed by blue ($g-r<0.7$) galaxies, the sample of red ($g-r>0.7$) galaxies, and a sample of faint and another composed by bright galaxies. The latter two samples show a quantitative agreement with the results from the full sample of edge-on galaxies in the SDSS.
In all cases we find a significant excess of structure in the direction perpendicular to the angular momentum for the 2-halo term, in good agreement with the numerical simulation results. The 1-halo term results show either a null or lower alignment signal than expected from the analysis of projected correlations in the numerical simulation. This would suggest the effects of astrophysical processes acting so that galaxies do not follow the DM structure inside low mass haloes of $\leq 10^{13}h^{-1}M_{\odot}$ or lower masses, a result that complements previous studies of high mass groups in the SDSS with $M\gtrsim 10^{13}h^{-1}M_{\odot}$, where galaxies are found to follow the internal DM structure (e.g. Paz et al., 2006). This effect is apparently more important for red, $g-r>0.7$ galaxies, indicating that there may be interesting astrophysical effects at work at low mass dark-matter haloes which could produce changes in the orientation of galaxy angular momenta, or the spatial re-distribution of galaxies within them.
Acknowledgments {#acknowledgments .unnumbered}
===============
DJP acknowledges receipt of a fellowship of Consejo Nacional de Investigaciones Cientifico Tecnicas and the support of the European Union’s ALFA-II programme, through LENAC, the Latin American European Network for Astrophysics and Cosmology. FAS acknowledges the receipt of an International Max-Planck Research School on Astrophysics fellowship. NDP was supported by a Proyecto FONDECYT Regular 1071006. This work was supported in part by Fondap “Center for Astrophysics” at Universidad Católica de Chile. The authors are very grateful to Manuel Merchán and Cintia Ragone, who kindly provided the numerical simulations and the “Friends of Friends” Code, respectively, used in this work. We acknowledge helpful discussions with Diego Garcia Lambas, Manuel Merchán and Ariel Sanchez. We are also grateful to the anonymous Referee, and to Manolis Plionis, Bernardo Cervantes Sodi, Oliver Hahn and Sebastien Peirani, whose feedback helped us improve this work in a significant way. This research has made use of the NASA Astrophysics Data System.
[99]{}
Adelman-McCarthy, J., et al. 2007, in press (arXiV:0707.3413). Altay G., Colberg J. M., Croft R. A. C., 2006, MNRAS, 370, 1422 Arag[ó]{}n-Calvo M. A., van de Weygaert R., Jones B. J. T., van der Hulst J. M., 2007, ApJ, 655, L5 Aryal B., Saurer W., 2005, A&A, 425, 431 Bailin J., Steinmetz M., 2005, ApJ, 627, 647 Barnes J. & Efstathiou G., 1987, ApJ, 319, 575 Basilakos S., Plionis M., Yepes G., Gottl[ö]{}ber S., Turchaninov V., 2006, MNRAS, 365, 539 Bett P., Eke V., Frenk C. S., Jenkins A., Helly J., Navarro J. 2007,MNRAS,376,215 Bridle S., Abdalla F. B., 2007, ApJ, 655, 1 Brown M. L., Taylor A. N., Hambly N. C., Dye S., 2002 MNRAS, 333, 501 Brunino R., Trujillo I., Pearce F.R., Thomas P.A., 2007, MNRAS, 375, 184 Bullock J. S., Dekel A., Kolatt T. S., Kravtsov A. V., Klypin A. A., Porciani C., Primack J. R. 2001, ApJ, 555, 240 Catelan P., Kamionkowski M., Blandford R. D., 2001, MNRAS, 320, 7 Catelan P., Porciani C., 2001, MNRAS, 323, 713 Catelan P., Theuns T., 1996a, MNRAS, 282, 436 Catelan P., Theuns T., 1996b, MNRAS, 282, 455 Colberg J. M., Krughoff K. S., Connolly A. J., 2005, MNRAS, 359, 272 Crittenden R. G., Natarajan P., Pen U., Theuns T., 2001, ApJ, 559, 552 Croft R., Dalton G., Efstathiou G. 1999, MNRAS, 305, 547. Croft R. A. C., Metzler C. A., 2000, ApJ, 545, 561 Cuesta A. J., Betancort-Rijo J. E., Gottl[ö]{}ber S., Patiri S. G., Yepes G., Prada F., 2008, MNRAS, 385, 867 D’Onghia E., Burkert A., 2004,ApJ, 612, 13 Doroshkevich A. G., 1970, Astrofisika, 6, 581 Eisenstein D.J. & Loeb A., 1995, ApJ, 439, 520 Faltenbacher A., Gottl[ö]{}ber S., Kerscher M., M[ü]{}ller V., 2002, A&A, 395, 1 Flin P., Godlowski W., 1989, SvAL, 15, 374 Gardner J. P., 2001,ApJ, 557, 616 Garrido J. L., Battaner E., Sanchez-Saavedra M. L.,Florido E., 1993, A&A, 271, 84 Gottl[ö]{}ber S., Turchaninov V., 2006, EAS, 20, 25 Hahn O., Porciani C., Carollo C. M., Dekel A., 2007, MNRAS, 375, 489 Hahn O., Carollo C. M., Porciani C., Dekel A., 2007, MNRAS, 381, 41 Han C., Gould A., Sackett P. D., 1995, ApJ, 445, 46 Heavens A. & Peacock J., 1988, MNRAS, 232, 339 Hernandez X., Park C., Cervantes-Sodi B., Choi Y.-Y., 2007, MNRAS, 375, 163 Heymans C., Heavens A., 2003, MNRAS, 339, 711 Heymans C., Brown M., Heavens A., Meisenheimer K., Taylor A., Wolf C., 2004, MNRAS, 347, 895 Hirata C. M., Mandelbaum R., Seljak U., Guzik J., Padmanabhan N., Blake C., Brinkmann J., Budávari T., Connolly A., Csabai I., Scranton R., Szalay A. S., 2004, MNRAS, 353, 529 Hirata C. M., Mandelbaum R., Ishak M., Seljak U., Nichol R., Pimbblet K. A., Ross N. P., Wake D.,2007, MNRAS, 381, 1197 Hoffman Y., 1986,ApJ, 301, 65 Hoffman Y., 1988,ApJ, 329, 8 Hu F. X., Yuan Q. R., Su H. J., Wu G. X., Liu Y. Z., 1998, ApJ, 495, 179 Hu F. X., Wu G. X., Song G. X., Yuan Q. R., Okamura S., 2006, Ap&SS, 302, 43 Jing Y. P., 2002, MNRAS, 335, 89 Kasun S. F., Evrard A. E., 2005, ApJ, 629, 781 King L.,2005, A&A, 441, 47 King L., Schneider P., 2003, A&A, 398, 23 Knebe A., Yahagi H., Kase H., Lewis G., Gibson B. K., 2008, arXiv, 805, arXiv:0805.1823, MNRAS Letter, in press Lee J., Erdogdu P., 2007,ApJ, 671,1248 Lee J., Pen U., 2007, ApJ, 670,1 Maccio’ A., Dutton A., van den Bosch F., Moore B., Potter D., Stadel J., 2006, MNRAS, 378, 55 Mackey J., White M., Kamionkowski M., 2002, MNRAS, 332, 788 Mandelbaum R., Hirata C. M., Ishak M., Seljak U., Brinkmann J., 2006, MNRAS, 367, 611 Nagashima M., Gouda N., 1998, MNRAS, 301, 849 Navarro J. F., Abadi M. G., Steinmetz M., 2004, ApJ, 613, 41 Padilla N., & Baugh C.M. 2002, MNRAS, 329,431. Padilla N., & Strauss M. 2008, MNRAS, submitted, arXiV:0802.0877. Padilla N., Merchan M., Valotto C., Lambas D.G., Maia M. 2001, ApJ, 554, 873. Padilla N., et al. (The 2dFGRS Team) 2004, MNRAS, 352, 211 Patiri S. G., Cuesta A. J., Prada F., Betancort-Rijo J., Klypin A., 2006, ApJ, 652, L75
Paz D., Lambas D.G., Padilla N., Merchan M. 2006, MNRAS, 366, 1503. Peacock J., & Smith R. 2000, MNRAS, 318, 1144. Peebles P. J. E., 1969, ApJ, 155, 393 Peirani S., Mohayaee R., de Freitas Pacheco J. A., 2004, MNRAS, 348, 921 Porciani C., Dekel A., Hoffman Y., 2002a, MNRAS, 322, 325 Porciani C., Dekel A., Hoffman Y., 2002b, MNRAS, 332,339 Quinn T. & Binney J., 1992, MNRAS, 255, 729 Ragone-Figueroa C., Plionis M., 2007, MNRAS, 377, 1785 Ryden B.S., 1988,ApJ, 329, 589 Sheth R. K., van de Weygaert R., 2004, MNRAS, 350, 517 Sheth R., Mo H.J., & Tormen G. 2001, MNRAS, 323, 1. Sugai H., Iye M., 1995, MNRAS, 276, 327 Sugerman B., Summers F. J., Kamionkowski M., 2000, MNRAS, 311,762 Takada M., White M., 2004, ApJ, 601, 1 Trujillo I., Carretero C., Patiri S.G., 2006, Vitvitska M., Klypin A. A.,Kravtsov A. V., Wechsler R. H., Primack J. R., Bullock J. S.,2002,ApJ,581,799 White S. M. D., 1984, ApJ,286, 38 de Vaucouleurs G., 1953, AJ, 58, 30
[^1]: The dimensionless spin parameter as defined by @Peebles1 is $\lambda=L\sqrt{|E|}/GM^{5/2}$, where $E$ is the total internal energy of the halo and $L$ and $M$ are its angular momentum and mass respectively. This number can be understood as a measure of the importance of rotational velocity vs. velocity dispersion as the dominant support mechanism of a halo
[^2]: Such angle threshold is equivalent to a ratio of $0.5$ of the ellipsoidal image semi-axes for a perfect thin disc.
[^3]: There is observational evidence that groups of galaxies with masses above this limit are characterised by intrinsic shapes consistent to those found in numerical simulations [@Paz]; however it is difficult to directly measure the shapes of lower mass systems due to discreteness effects.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This paper is the first work to perform spatio-temporal mapping of human activity using the visual content of geo-tagged videos. We utilize a recent deep-learning based video analysis framework, termed hidden two-stream networks, to recognize a range of activities in YouTube videos. This framework is efficient and can run in real time or faster which is important for recognizing events as they occur in streaming video or for reducing latency in analyzing already captured video. This is, in turn, important for using video in smart-city applications. We perform a series of experiments to show our approach is able to accurately map activities both spatially and temporally. We also demonstrate the advantages of using the visual content over the tags/titles.'
author:
- Yi Zhu
- Sen Liu
- Shawn Newsam
bibliography:
- 'sigproc.bib'
title: 'Large-Scale Mapping of Human Activity using Geo-Tagged Videos'
---
<ccs2012> <concept> <concept\_id>10002951.10003227.10003236</concept\_id> <concept\_desc>Information systems Spatial-temporal systems</concept\_desc> <concept\_significance>500</concept\_significance> </concept> <concept> <concept\_id>10002951.10003317.10003371.10003386.10003388</concept\_id> <concept\_desc>Information systems Video search</concept\_desc> <concept\_significance>300</concept\_significance> </concept> <concept> <concept\_id>10003120.10003145.10003147.10010887</concept\_id> <concept\_desc>Human-centered computing Geographic visualization</concept\_desc> <concept\_significance>500</concept\_significance> </concept> <concept> <concept\_id>10003120.10003145.10011769</concept\_id> <concept\_desc>Human-centered computing Empirical studies in visualization</concept\_desc> <concept\_significance>500</concept\_significance> </concept> <concept> <concept\_id>10010147.10010178.10010224.10010225.10010228</concept\_id> <concept\_desc>Computing methodologies Activity recognition and understanding</concept\_desc> <concept\_significance>500</concept\_significance> </concept> <concept> <concept\_id>10010147.10010178.10010224.10010225.10010230</concept\_id> <concept\_desc>Computing methodologies Video summarization</concept\_desc> <concept\_significance>500</concept\_significance> </concept> <concept> <concept\_id>10010147.10010257.10010258.10010259.10010263</concept\_id> <concept\_desc>Computing methodologies Supervised learning by classification</concept\_desc> <concept\_significance>500</concept\_significance> </concept> <concept> <concept\_id>10010147.10010257.10010293.10010294</concept\_id> <concept\_desc>Computing methodologies Neural networks</concept\_desc> <concept\_significance>500</concept\_significance> </concept> </ccs2012>
Introduction {#sec:introduction}
============
Mapping human activity on a large scale in real time or near real time is a fundamental yet challenging task in the geographic and social sciences. It is an essential component for making cities smart, particularly with regard to resource allocation, disease control, social interaction, traffic management, etc.
Researchers have exploited technological advances to map human mobility by using (GPS) trajectory data [@Density_Demsar_2010; @Trajectory_Qi_2013; @LULC_Activity_Li_2017] or mobile phone records [@Activity_Aware_Phithakkitnukoon_2010; @Mobile_Sagl_2014; @Phone_NE_2015; @Mobility_Yang_2016; @Generative_Yin_2017]. These approaches do not provide information about specific activities though.
Recently, researchers have utilized large-scale geo-referenced multimedia from social networks, like Twitter, Facebook, Instagram and YouTube, to map activity. Our work falls into this category. The public willingly shares their daily activities in real time or near real time by posting words, pictures and videos to these sites. The content of these large collections of multimedia along with the associated metadata such as geo-coordinates, time stamps, tags/titles, popularity rating, etc. thus represents a promising opportunity to map activity.
There has been work on using Twitter to geo-visualize human activity on maps [@City_Story_Kling_2012; @Twitter_Hawelka_2014]. Since the vast majority of Twitter feeds are only text, and all Tweets are very short, such approaches are not effective for precise activity mapping. Indeed, even though [@Twitter_Hawelka_2014] collects millions of geo-tagged Tweets, only global mobility patterns are detected and mapped.
There has also been work on using geo-tagged images to analyze human activity [@Flickr_Kisilevich_2010; @VGI_Sun_2013; @Magnetism_Paldino_2015; @Emotions_Hauthal_2016; @Travel_Yuan_2016]. While this is closer to our work, it is still limited since most activities have a temporal component which is not captured by an image.
We therefore propose using geo-tagged videos to map human activity. We consider both the appearance and temporal (dynamic) aspects of the videos. This allows more effective activity detection than using tags/titles or the visual content of images.
Performing activity recognition in video is a challenging problem. Video data is large which makes real-time or near real-time analysis difficult. And, video data is very complex. Activity recognition is a high-level vision task which is very difficult to do using the pixel information alone.
Fortunately, the field of computer vision has made great progress recently in high-level video understanding thanks to deep learning. Large-scale labeled video datasets [@ucf101; @activityNet; @KarpathyCVPR14; @YouTube_8M_2016] have been created, allowing deep convolutional neural networks (CNN) to be trained and achieve impressive performance on activity recognition. We take advantage of this recent progress to perform, for the first time, spatio-temporal mapping of human activity using geo-referenced videos.
This paper bridges activity recognition in video with geographic knowledge discovery. The salient aspects of the work include:
- Our work is the first to perform spatio-temporal mapping of human activity using the visual content of geo-tagged videos.
- We utilize a recent, efficient video analysis framework termed hidden two-stream networks. The framework performs activity recognition at $130$ frames per second (fps) which allows it to run in real time. Efficiency is important for analyzing live video streams and reducing latency in analyzing recently captured video.
- The video analysis framework is effective, achieving 90 percent accuracy on a 10 class activity classification problem.
- We show our approach is able to spatially map a diverse set of sports activities.
- We show our approach is able to detect the impact of weather, such as temperature and precipitation, on activity.
- We show our approach is able to spatio-temporally map specific events such as parades and street fights.
- We show that using video content is more accurate than using tags/titles.
- Our framework is flexible. It could easily be adapted to use geo-referenced videos to map a range of activities that are important for smart cities such as monitoring public safety, monitoring traffic, monitoring public health, etc. Further, the framework is robust enough that it does not require traditional surveillance videos but can use crowd-sourced, YouTube-like videos captured by mobile devices.
Related Work {#sec:related_work}
============
Our work is related to several lines of research.
**Large-Scale Geo-Tagged Multimedia** The exponential growth of publicly available geo-referenced multimedia has created a range of interesting opportunities to learn about our world. At the intersection of geographic information science and computer vision, large collections of geotagged photos/videos have been used to map world phenomenon [@CrandallMapWorld09], classify land use [@landuse_yi_sigspatial_2015], perform geo-location [@Haysim2gps08; @geolocalization_Gupta_2016], model landmarks [@3dModelSnavely08], conduct urban planning [@Magnetism_Paldino_2015], and detect sentiment hotspots [@emotion_yi_sigspatial_2016].
Although such open-access multimedia represents a wealth of information, analyzing it is challenging due to how noisy and diverse it is. Challenges specific to using this data for geographic discovery include inaccurate location information, uneven spatial distribution and varying photographer intent. We are mindful of these challenges and recognize they likely temper our results.
Our work is novel in that it uses a large collection of geo-tagged videos to map human activity as conveyed through the videos that ordinary people take. We specifically focus on spatial and spatio-temporal activity analysis in an urban area.
**Visual Geo-localization** Geo-localization is the problem of determining where something is. There exists an extensive body of literature on the large-scale visual geo-localization of images. Video geo-localization by comparison is relatively less studied [@Overhead_Hammoud_2013; @Structure_Bodensteiner_2015; @event_More_2016; @geolocalization_Gupta_2016]. Note that our goal is not to perform geo-localization. Our videos are already geo-tagged. Our goal is to perform geographic knowledge discovery by analyzing the geo-tagged videos.
**Video Activity Recognition** The field of human action recognition in video has evolved significantly over the past few years. Traditional handcrafted features such as Improved Dense Trajectories (IDT) [@idtfWang2013] dominated the field of video analysis for many years. However, despite their excellent performance, IDT and its improvements [@peng2014action; @MIFS2015; @dovf_lan_2017] are too computationally restrictive to be used for real time applications. CNNs [@KarpathyCVPR14; @c3d2015], which are often several orders of magnitude faster than IDTs, initially performed much worse than IDTs. This inferior performance was due mostly to the difficulty appearance-based CNNs have in capturing the movement between frames. Subsequent two-stream CNNs [@twostream2014; @wanggoodpractice2015] addressed this problem by pre-computing optical flow using traditional optical flow estimation methods [@tvl1realTime] and training a separate CNN to encode the movement captured by the optical flow. This additional stream (also known as the temporal stream) significantly improved the accuracy of CNNs and finally allowed them to outperform IDTs on several benchmark action recognition datasets [@wanggoodpractice2015; @convTwoStream2016; @depth2action; @TSN2016; @res_two_stream_nips16; @diba_tle_2016]. These accuracy improvements indicate the importance of temporal motion information for action recognition. Pre-computing optical flow is computational and storage intensive and prevents traditional two-stream networks from running in real time.
In this work, we utilize the recent hidden two-stream networks [@hidden_zhu_17] for activity recognition. Our framework is extremely efficient yet maintains competitive accuracy with slower approaches which cannot operate in real time. We compare it for activity recognition with another state-of-the-art real-time activity model named C3D [@c3d2015]. The results show the superiority of our method.
Methodology {#sec:methodology}
===========
In this section, we first formulate our problem in Section \[sec:problem\_formulation\]. We then describe our approach towards recognizing human activities in videos in Section \[sec:hidden\_two\_stream\_networks\].
Problem Formulation {#sec:problem_formulation}
-------------------
Recall from Section \[sec:introduction\] that work exists on using volunteered geographic information such as Twitter/Facebook texts and Instagram/Flickr images to perform geographic knowledge discovery. There is little work on using geo-tagged videos and, that which does, does not exploit the visual content. The challenge is in developing an effective and efficient video analysis framework. In this paper, we utilize hidden two-stream networks [@hidden_zhu_17] to overcome these challenges. The “hidden” part of the model addresses the efficiency while the “two-stream” aspect effectively handles both static appearance and dynamic motion. Details about the framework are provided in Section \[sec:hidden\_two\_stream\_networks\].
The overarching goal of our work is to show that geo-referenced videos, such as at YouTube, can be used to spatio-temporally map human activity on a large scale. We select $8$ popular sports, **baseball, basketball, football, golf, racquetball, soccer, swimming and tennis**, as common human activities to map. We also include the class **parade** to demonstrate how our approach can trace an event and the class **street fight** to show direct application to public safety. We thus consider $10$ human activities in total but this could easily be extended to others. The fundamental technical problem we now face is human activity recognition in video. The next few sections describe our solution to this problem.
![Illustration of the hidden two-stream networks that performs activity recognition using the visual content of a video. The MotionNet CNN takes consecutive video frames as input and outputs the estimated optical flow. The temporal stream CNN then uses this flow to predict activity labels. Late fusion is performed to combine the stacked temporal stream with a spatial stream. Significantly, both streams are end-to-end trainable.[]{data-label="fig:framework_hidden"}](framework_hidden.pdf){width="1.0\linewidth"}
Hidden Two-Stream Networks {#sec:hidden_two_stream_networks}
--------------------------
In this section, we describe the hidden two-stream networks [@hidden_zhu_17] we use for activity recognition in detail. We first recall the baseline two-stream networks [@twostream2014] and state its limitation for real world applications in Section \[sec:two\_stream\_networks\]. We then introduce the unsupervised network MotionNet for optical flow estimation in Section \[sec:motionnet\]. In Section \[sec:stacked\_temporal\_stream\], we stack the temporal stream network on top of MotionNet to allow end-to-end training for action recognition. Finally, we combine the stacked temporal stream CNN with the standard spatial stream CNN using late fusion to further improve the activity recognition accuracy.
### **Two-Stream Networks** {#sec:two_stream_networks}
As mentioned in Section \[sec:related\_work\], state-of-the-art activity recognition approaches [@wanggoodpractice2015; @convTwoStream2016; @depth2action; @TSN2016; @res_two_stream_nips16; @diba_tle_2016] build upon the popular two-stream networks framework [@twostream2014]. This framework includes a spatial stream CNN which takes RGB video frames (images) as input, and a temporal stream CNN which takes optical flow images as input. This allows the network to model both static appearance and dynamic motion.
However, computing optical flow is significantly more expensive than applying the CNNs. For example, it takes $0.065s$ to estimate a single optical flow image from consecutive $320\times240$ pixel video frames using a GPU accelerated TV-L1 [@tvl1realTime] algorithm. This is prohibitively slow for training a temporal stream CNN on-the-fly. The optical flow images also need to be cached before input to the temporal CNN which incurs significant storage costs. The need to pre-compute optical flow is the major speed and storage bottleneck that prevents traditional two-stream approaches from being used in real time applications.
### **MotionNet** {#sec:motionnet}
In order to achieve real time activity recognition, we use the MotionNet [@hidden_zhu_17] CNN instead of slower, handcrafted methods to compute optical flow. The key to using a CNN is to pose optical flow computation as a learning problem. MotionNet treats motion estimation as an image reconstruction problem [@jasonUnsup2016; @guided_flow_17; @densenet_denseflow_icip17] where we seek to learn the optimal optical flow that allows the current video frame to be constructed from the previous one. Formally, given a pair of adjacent video frames $I_{1}$ and $I_{2}$ as input, MotionNet generates a motion field $V$. $V$ and $I_{2}$ are then used to produce the estimate $I_{1}^{\prime}$ using inverse warping, i.e., $I_{1}^{\prime} = \mathcal{T}[I_{2}, V]$, where $\mathcal{T}$ is the inverse warping function. The goal is to minimize the photometric (pixelwise) error between $I_{1}$ and $I_{1}^{\prime}$.
The architectural details of MotionNet can be seen in Table \[tab:stacked\_model\]. The top section of the table corresponds to MotionNet and the bottom section is the traditional temporal stream CNN upon which MotionNet is stacked. Training MotionNet to learn optimal optical flow involves minimizing the following three objective functions:
- A standard pixelwise reconstruction error function $$L_{\text{pixel}} = \frac{1}{N} \sum_{i, j}^{N} \rho ( I_{1}(i, j) - I_{2}(i+V_{i,j}^{x}, j+V_{i,j}^{y}) )
\label{eq:pixel_loss}$$ where $i$ and $j$ are the frame numbers and $V^x$ and $V^y$ are the estimated optical flows in the horizontal and vertical directions. The inverse warping is performed using a spatial transformer module [@stn_nips15]. We use a robust convex error function, the generalized Charbonnier penalty $\rho(x) = (x^{2} + \epsilon^{2})^{\alpha}$, to reduce the influence of outliers. $N$ denotes the total number of pixels.
- A smoothness loss to address the ambiguity of estimating motion in non-textured regions (the aperture problem) $$L_{\text{smooth}} = \rho (\nabla V_{x}^{x} ) + \rho ( \nabla V_{y}^{x}) + \rho ( \nabla V_{x}^{y}) + \rho ( \nabla V_{y}^{y})
\label{eq:smoothness_loss}$$ where $\nabla V_{x}^{x}$ and $\nabla V_{y}^{x}$ are the gradients of the estimated flow field $V^{x}$ in the horizontal and vertical directions. Similarly, $\nabla V_{x}^{y}$ and $\nabla V_{y}^{y}$ are the gradients of $V^{y}$. A generalized Charbonnier penalty $\rho(x)$ is also used.
- A structural similarity (SSIM) loss [@SSIM_2004] that helps MotionNet learn the structures of frames. It is calculated as
$$L_{\text{ssim}} = \frac{1}{N} \sum ( 1 - \text{SSIM} (I_{1}, I_{1}^{\prime}) )
\label{eq:ssim_loss}$$
where SSIM($\cdot$) is a standard structural similarity function. This forces MotionNet to produce flow fields with clear motion boundaries.
The overall loss is a weighted sum of the pixelwise reconstruction loss, the pixelwise smoothness loss and the region-based SSIM loss $$L = \lambda_{1} \cdot L_{\text{pixel}} + \lambda_{2} \cdot L_{\text{smooth}} + \lambda_{3} \cdot L_{\text{ssim}}
\label{eq:total_unsup_loss}$$ where $\lambda_{1}$, $\lambda_{2}$ and $\lambda_{3}$ weight the relative importance of the different metrics during training. $\lambda_{1}$ and $\lambda_{3}$ are set to $1$. $\lambda_{2}$ is set as suggested in [@flownet].
### **Stacked Temporal Stream** {#sec:stacked_temporal_stream}
Since MotionNet and the temporal stream are both CNNs, they can be stacked on top of each other and trained in an end-to-end manner. The stacked temporal stream CNN is then combined with a standard spatial stream CNN as shown in Figure \[fig:framework\_hidden\]. Following previous literature, the two streams are combined through weighted average late fusion using a spatial to temporal ratio of $1$:$1.5$ as in [@wanggoodpractice2015; @TSN2016; @diba_tle_2016].
Experiments {#sec:experiments}
===========
This section first introduces our video dataset. We train and validate our activity recognition model using a large collection of videos with known classes, built from existing activity recognition datasets as well as self-crawled YouTube videos. This section also describes our implementation of the hidden two-stream networks, reports its performance and compares it with another real-time framework called C3D. This section finally presents the results of a number of application scenarios including the spatio-temporal mapping of sports activities, parade events, violence and the effects of weather.
Dataset {#sec:dataset}
-------
The dataset we use to train and validate our activity recognition model contains $10$ activity classes: baseball, basketball, football, golf, racquetball, soccer, swimming, tennis, parade and street fight. It includes videos from existing activity recognition datasets as well as self-crawled YouTube videos. We only need the activity labels of these videos–we do not need geo-tags. We first leverage existing datasets including Sports-1M [@KarpathyCVPR14], UCF101 [@ucf101] and FCVID [@FCVID] to create an initial dataset. This initial dataset is too small and unbalanced though for fine-tuning deep CNNs and so we also download YouTube videos[^1] using the activity labels as keywords. We perform a visual sanity check to remove irrelevant videos from these downloads. Our final dataset contains $10,000$ videos in total, $1,000$ for each activity class. This size is of similar order to the UCF101 [@ucf101] and ActivityNet 1.3 [@activityNet] datasets which have been shown to be large enough to fine-tune deep networks. We divide this dataset into training and validation components using a split ratio of $0.8$:$0.2$.
To perform our spatio-temporal mapping, we download all geo-tagged YouTube videos using the same keywords within the city of San Francisco for the year $2016$. This results in $265,477$ geo-tagged videos. Note that these videos are disjoint from the ones used to train and validate the activity recognition model above.
Implementation {#sec:implementation}
--------------
We use the Caffe toolbox [@jia2014caffe] to implement the CNNs. All timing results correspond to a workstation with an Intel Core I7 (4.00GHz) and an Nvidia Titan X GPU.
**MotionNet:** MotionNet is trained from scratch using the three unsupervised objectives: the pixelwise reconstruction loss $L_{\text{pixel}}$, the piecewise smoothness loss $L_{\text{smooth}}$ and the region-based SSIM loss $L_{\text{ssim}}$. The generalized Charbonnier parameter $\alpha$ is set to $0.4$ in the pixelwise reconstruction loss, and $0.3$ in the smoothness loss.
The models are trained using Adam optimization with the default parameter values $\beta_{1}=0.9$ and $\beta_{2}=0.999$. The batch size is $16$. The initial learning rate is set to $3.2\times10^{-5}$ and is divided in half every $100$k iterations. We end our training at $400$k iterations.
**Hidden two-stream networks:** The hidden two-stream networks includes the spatial stream and the stacked temporal stream. The MotionNet is pretrained as above. The spatial stream CNN is a VGG16 CNN pretained on the ImageNet challenge [@imagenet_cvpr09], and the stream temporal CNN is initialized with the snapshot provided by Wang [@wanggoodpractice2015]. We use stochastic gradient descent to train the networks with a batch size of $128$ and momentum of $0.9$. We also use horizontal flipping, corner cropping and multi-scale cropping as data augmentation to prevent over fitting.
For the spatial stream CNN, the initial learning rate is set to $0.001$, and is divided by $10$ every $4$K iterations. We stop the training at $10$K iterations. For the stacked temporal stream CNN, we set different initial learning rates for MotionNet and the temporal stream: $10^{-6}$ and $10^{-3}$. We divide the learning rates by $10$ after $5$K and $10$K iterations. The maximum iteration is set to $16$K.
**C3D:** C3D [@c3d2015] is a generic video analysis framework and consists of 3D convolutions instead of 2D convolutions as in most deep networks. The input to the network is sets of contiguous video frames organized as clips. The model has eight convolutional and five pooling layers which are followed by three fully connected layers. All 3D convolution filters are $3 \times 3 \times 3$ with stride 1, and all 3D poling layers are $2\times2\times2$ with stride 1, except for pool1. In order to preserve temporal information in the earlier layers, pool1 has size $1 \times 2 \times 2$. See [@c3d2015] for more details.
C3D is pre-trained on Sports-1M, which has $487$ classes. We fine tune it on our training data. During fine tuning, the network weights are learned using mini-batch (30 video clips) stochastic gradient descent with momentum (set to 0.9). The learning rate is initialized to 0.0001, except for the last fully connected layer with 0.01. Both learning rates decrease to 1/10 of their values whenever the performance saturates, and training is stopped when the learning rate is smaller than $10^{-6}$. Dropout is applied with a ratio of $0.7$ after the first two fully connected layers.
Activity Recognition Evaluation {#sec:activity_recognition_evaluation}
-------------------------------
Table \[tab:comparison\] compares the accuracy and efficiency of the hidden two-stream networks with the popular C3D network. The hidden two-stream networks achieves just over $90\%$ accuracy on the 10 class validation dataset. It is about $6\%$ more accurate than C3D. C3D is seen to be more efficient but both can run much faster than real time (30 fps).
Table \[tab:break\_down\] provides the per-class accuracy. Both approaches obtain the highest accuracy on swimming and racquetball. This is likely due to context since swimming is the only water-related sport and racquetball has the largest proportion of indoor scenes. Interestingly, the hidden two-stream networks achieves much higher accuracy than C3D on street fight. This is likely due to the increased capacity of the MotionNet CNN to learn optimal motion representations. Street fight contains significant motion (both subject and camera) which might not be captured by the 3D convolutional filtering performed by C3D.
The remainder of the experiments are performed with the hidden two-stream networks.
![Temporal analysis of user uploaded parade videos in the city of San Francisco in year 2016. The y axis indicates the number of geo-tagged parade videos for each day. The peaks correspond to the major parades.[]{data-label="fig:parade_temporal"}](parade_temporal.pdf){width="1.0\linewidth"}
![Spatial analysis of the 46th San Francisco Pride parade in 2016. Left: official parade route. Right: map of classified videos. Note the correlation. []{data-label="fig:parade_spatial"}](parade_spatial-min.pdf){width="1.0\linewidth"}
Spatial Sports Mapping {#sec:spatial_sports_mapping}
----------------------
We now apply our framework to the geo-tagged YouTube videos from San Francisco for 2016. During inference, we sample frames/clips every one second to reduce computational cost. Figure \[fig:spatialActivityMap\] shows the locations of videos classified as the six most popular sports: baseball, basketball, football, golf, soccer and tennis. Also shown are four detections for each sport. We show a sample frame from the video that resulted in the detection as well as a satellite image of the location of the video. These results demonstrate that our approach a) is able to correctly classify the YouTube videos, and b) can use this classification to map where the activities take place.
We make the following four observations based on the results in Figure \[fig:spatialActivityMap\].
**Observation 1**: Our approach is able to locate sports fields and complexes using the visual content of the geo-tagged videos. Figure \[fig:baseball\] contains a concentration of points in the area of AT$\&$T park, the home of the SF Giants baseball team. We also locate the San Francisco State University basketball court, George Washington High School football field, TPC Harding Park golf course, Crocker Amazon soccer fields, John McLaren Park tennis courts, etc.
We are also able to locate where sports are played in a more informally place. For example, there are a lot of videos labeled as basketball in high density residential neighborhoods. We checked these and, indeed, found that they are located in backyards and other places where basketball can be played.
**Observation 2**: There are relatively few videos classified as golf and football. This makes sense for golf since it requires a large, open area and good weather. The reason for so few football videos is likely that the training videos for this class from the Sports1M dataset are captured during actual games in stadiums with full teams of players wearing helmets, etc. This is different from the more informal types of football activities found in an urban area such as touch football or simply throw and catch. A more representative training dataset is needed for this class.
**Observation 3**: The video frames and satellite images are in agreement with the predicted sports and their locations. There is, however, one interesting exception (the top right example in Figure \[fig:golf\]) of a golf video located in downtown San Francisco. Upon further investigation, we found this makes sense since there is an indoor driving range inside the building named Eagle Club indoor golf. The club provides customized scenes and projects virtual driving ranges on large monitors for players to practice. The classified video is an advertisement. This example demonstrates a distinct advantage that ground-level images and videos have over satellite or aerial images–they can be used to perform geographic discovery indoors.
**Observation 4**: Our approach is able to use context to detect where a sport is played even if it is not occurring at the time the video was captured or the activity is difficult to discern. For example, in the top left example in Figure \[fig:football\], the video snippet is an oblique view of just the football field. And, in the bottom right example in Figure \[fig:soccer\], the players are very far from the camera. The ability of our approach to do this can be attributed to the spatial stream CNN’s capacity to learn the static appearance of where sports are played.
Spatio-Temporal Parade Mapping {#sec:spatio_temporal_parade_mapping}
------------------------------
The goal here is to locate specific events, such as a parade, both spatially and temporally. We first detect all parade videos and temporally group them by date. We then map the videos in a group to identify the parade route.
**Temporal Analysis**: We detect a total of $15,645$ parade videos in San Francisco in 2016. The daily distribution is shown in Figure \[fig:parade\_temporal\]. The peaks correlate with known parades including the Chinese New Year parade (February 20), the St. Patrick’s Day parade (March 12), the Carnaval Grand parade (May 28), the Pride parade (June 25) and the Italian Heritage parade (October 9).
Closer analysis shows that the videos of a parade tend to be uploaded after the event, sometimes days later. This is different from texts or images which tend to be shared during the event. This is likely because video requires better network connectivity. Also, users often first edit their videos before uploading them.
**Spatial Analysis**: We now map the videos of the most popular parade in San Francisco in 2016 (based on our detections), the 46th Pride parade. As shown in Figure \[fig:parade\_spatial\], our mapping results (right) are strongly correlated with the official parade route (left), from Market/Beale to Market/8th Street in downtown San Francisco.
![Weather impact on different activities. Left: Temperature. Right: Precipitation. []{data-label="fig:weather"}](weather.pdf){width="1.0\linewidth"}
Weather Impact on Activities {#sec:weather_impact_on_human_activities}
----------------------------
We here investigate if we can use our approach to observe the impact that weather has on activities. We separately consider temperature and precipitation[^2]. We use the number of videos uploaded on a month-to-month basis to indicate the prevalence of particular activity. We plot this versus temperature or precipitation. All values are normalized from $0\sim1$ based on their maximum and minimum values.
**Temperature**: As shown in Figure \[fig:weather\] left, we observe a clear positive correlation between temperature and both golf and swimming. This makes sense because people are less likely to play golf or swim outdoors when it is cold.
**Precipitation**: As shown in Figure \[fig:weather\] right, precipitation has a great impact on outdoor activities like basketball. The curve for precipitation (blue) and basketball (red) are negatively correlated. By comparison, there is little correlation, positive or negative, for indoor activities such as racquetball.
![Violence detection. Left: our predicted street fight mapping. Right: the official police record of Assault mapping. []{data-label="fig:street_fight"}](street_fight-min.pdf){width="1.0\linewidth"}
Crime Detection {#sec:violence_detection}
---------------
Detecting criminal activities is important for public safety. We here demonstrate how our framework can be used to map violence using YouTube videos. This shows how our framework can generalize to a range of applications related to smart cities given suitable training data.
We apply our framework to the San Francisco YouTube videos and detect $7,784$ instances of street fight. The locations of the videos are shown in Figure \[fig:street\_fight\] left. We notice concentrations of violence in downtown San Francisco, the Mission District, Hunters Point, etc. These are known to be high-crime areas. For comparison, we show the locations of Assault from a San Francisco crime map in Figure \[fig:street\_fight\] right[^3] derived from official police records. Our predicted locations are shown to be correlated with the official records.
We would like to point out how our framework is different and complementary to using traditional surveillance cameras to monitor crime. We use geo-tagged videos from YouTube. The challenge is that these videos are not taken from the same viewpoint, with the same camera, with controlled lighting conditions, etc. This makes our problem much more difficult. However, we are able to leverage the scale and embedded perspective of the crowd to detect incidents that might not be captured using surveillance cameras.
Discussion {#sec:discussion}
==========
**Advantage over Tag/Title-Based Mapping**
There are many advantages of using the visual content of geo-tagged videos instead of the tags/titles to map activities. Visual data is rich and can convey much more about what is occurring at a location than tags/titles. The visual content is also not subject to the ambiguity or imprecision of language. We demonstrate this here by mapping football and soccer using the San Francisco YouTube videos.
Most of the world uses football to refer to what is called soccer in the United States. This can cause problems when using tags/titles to map these two sports. Figure \[fig:content\_advantage\] shows a satellite image of several soccer fields at Ulloa elementary school in San Francisco’s Sunset District. The image on the left shows the locations of videos with “football” in the tag/title such as “football campaign” and “kids playing football”. Closer inspection shows these are not really videos of football but are of soccer. By comparison, the image on the right shows the results of our football activity detector. Using the visual content allows us to avoid the false positive detections in this case.
Conclusion {#sec:conclusion}
==========
We performed the first investigation into using the visual content of geo-tagged videos to map human activity. We utilized the recent hidden two-stream networks to detect 10 different activities in a large collection of YouTube videos of San Francisco. Our approach can run in real time which is important for real time applications. We performed a series of experiments to show our framework can map a diverse set of activities, can map specific events such as parades and street fights, can observe the impact that weather has on activities and is more accurate than using the tags/titles of the videos.
In the future, we plan to investigate whether our framework can be adapted to detect a range of suspicious activities in surveillance video such as theft, vandalism, etc. A challenge to using CNNs for surveillance video is the relative lack of training data. We will explore whether YouTube videos can be used to at least pre-train the models. The challenge will then be to generalize the models to the different viewpoint, etc. of the surveillance videos.
Additional directions include scaling the mapping to country or continental regions as well as to more activity classes. Finally, we will investigate reducing the sizes of our CNN models so they can be deployed in mobile devices or at endpoint equipment such as networked cameras.
Acknowledgments
===============
We gratefully acknowledge the support of NVIDIA Corporation through the donation of the Titan X GPU used in this work. This work was funded in part by a National Science Foundation CAREER grant, \#IIS-1150115, and a seed grant from the Center for Information Technology in the Interest of Society (CITRIS).
[^1]: We limit the duration of downloaded videos to be shorter than ten minutes and to be of high quality.
[^2]: All weather data are obtained from https://www.wunderground.com/.
[^3]: Data source is from: https://spotcrime.com/ca/san+francisco
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'For a function algebra $A$ we investigate relations between the following three topics: isomorphisms of singly generated $A$-modules, Morita equivalence bimodules, and ‘real harmonic functions’ with respect to $A$. We also consider certain groups which are naturally associated with a uniform algebra $A$. We illustrate the notions considered with several examples.'
address:
- |
Department of Mathematics\
University of Houston\
Houston, TX 77204-3476
- |
Department of Mathematics and Statistics\
Southern Illinois University\
Edwardsville, IL 62026-1653
author:
- David Blecher
- Krzysztof Jarosz
date: 'Sept 4, 1999'
title:
-
- |
Isomorphisms of function modules,\
and generalized approximation in modulus
---
Introduction.
=============
By a [*uniform algebra*]{} or [*function algebra*]{} on a compact Hausdorff space $\Omega$, we shall mean a subalgebra $A$ of $C(\Omega)$ (the continuous complex valued functions on $\Omega$) which contains constants and separates points. In most of this paper we are concerned with closed submodules of $C(\Omega)$ of the form $Af$, where $f$ is a strictly positive and continuous function on $\Omega$. In Part C we will allow $f$ to be nonnegative.
Before we proceed any further with this introduction, we shall take a paragraph to explain why these modules are more general than they appear to be at first. We need a little notation. By a [*concrete function $A$-module*]{} we shall mean a closed linear subspace of $C(K)$, for a compact Hausdorff space $K$, which is closed under multiplication by $\pi(A)$, where $\pi : A \rightarrow
C(K)$ is a unital homomorphism. By an [ *(abstract) function $A$-module*]{} we shall mean a Banach $A$-module $X$ which is isometrically $A$-module isomorphic to (in future we shall simply say ‘$A$-isometric to’ for short) a concrete function module. Although we shall not particularly use this here, this class of modules was given several equivalent abstract characterizations in [@BLM]; for example it coincides with the class of Banach $A$-modules whose module action is contractive with respect to the injective tensor product. It is proved in [@Bsh], that any algebraically singly generated faithful[^1] function $A$-module is $A$-isometric to one of the form $Af$ described in the first paragraph, for some strictly positive continuous function $f$ on some (possibly different) compact $\Omega$.
In this paper we investigate the relations between the following three topics: isometries and almost isometries between modules of the type discussed above, real ‘harmonic’ functions on $\Omega $ with respect to $A$, and Morita equivalence bimodules over $A$.
In Part A we provide a necessary and sufficient condition on functions $f_{1},f_{2}$ for the modules $Af_{1},Af_{2}$ to be isometric (as Banach spaces). The result generalizes the classical description of isometries of uniform algebras. We then extend the result to almost isometric modules; where modules $Af_{1},Af_{2}$ are called ‘almost isometric’ if for any $\varepsilon >0$ there is a surjective linear isomorphism $T:Af_{1}\rightarrow Af_{2}$ such that $\left\| T\right\| \left\|
T^{-1}\right\| <1+\varepsilon $.
One of the main results in Part A, is a characterization of modules of the type $Af$ which are almost isometric to $A$. If $\Omega = \partial A$, the Shilov boundary of $A$, we will see that “$Af \cong A$ almost isometrically” is equivalent to $f$ being uniformly approximable by the moduli of invertible elements of $A$. That is, $f$ is in the uniform closure $\bar{Q}$ of $Q$, where $Q = \{ |a| : a \in A^{-1} \}$. Indeed, up to $A$-isometric isomorphism, for any $\Omega$ on which $A$ is a function algebra, the Banach $A$-modules which are almost $A$-isometric to $A$, are exactly the submodules $Af \subset C(\Omega)$, for some strictly positive $f \in \bar{Q}$. The set $H_A(\Omega) = \{ \log f : 0 < f \in \bar{Q} \}$ is well known to those familiar with the theory of uniform algebras. In particular, $A$ is known as ‘logmodular’ if $H_A(\Omega) =
C_{{\mbox{$\mathbb{R}$}}}(\Omega)$. For any function algebra $A$, the class $H_A(\Omega)$ deserves to be called a ‘harmonic class’ of functions with respect to $A$. In this paper, to be more specific, by a ‘harmonic class’ we shall mean a class $\mathcal{B}(\Omega )$ of real continuous functions on $\Omega ,$ which have at least the following properties:
- If $f\in \mathcal{B}(\Omega )$ then $f_{|_{\partial A}}\in
\mathcal{B}(\partial A)$.
- If $f\in \mathcal{B}(\Omega )$ then there exists a unique $\tilde{f}\in \mathcal{B}(M_{A})$ such that $\tilde{f}_{|_{\Omega }}=f$. ($M_A$ is the maximal ideal space of $A$).
- Every $f\in \mathcal{B}(\Omega )$ achieves its maximum and minimum value on $\partial A$.
- $f_{1},f_{2}\in \mathcal{B}(\Omega )$ implies $f_{1}+f_{2}\in
\mathcal{B}(\Omega )$ (Indeed most of the classes we study in this paper are additive groups).
- If $A$ consists of functions which are analytic on a region $R$ in $\mathbb{C}^{n}$, and which separate points of $R$, then (via the obvious homeomorphic embedding $R\rightarrow M_{A}$) the functions in $\mathcal{B}(M_{A})$ are genuinely harmonic on $R$.
This leads us conveniently into a description of Part B of the paper. Here is a natural idea to attempt to generalize the class $H_A(\Omega)$, or equivalently, the strictly positive functions which are uniform limits $$f = \lim_n |k_n| \; ; \; \; \text{where} \; k_n \; , \;
h_n \; \in A \;
\; \text{with} \;
\; k_n h_n = 1 , \label{ac1}$$ on $\Omega$. We will write $A^{(n)}$ for the space of $n$-tuples with entries in $A$. An element of $A^{(n)}$ will be called an $A$-tuple, and will often be regarded as a function $\Omega \rightarrow \mathbb{C}^{n}$. For two $A$-tuples $H = (h_i), K = (k_i)$ of the same ‘length’ we define $H.K = \sum_i h_i k_i \in A$. Consider the set of strictly positive functions which are uniform limits $$f(w) = \lim_n \Vert K_n(w) \Vert_2 \; ;
\; \; \text{where} \; K_n \; , \; H_n \; \in A^{(m_n)}
\; \; \text{with} \; \; K_n . H_n = 1 ,
\label{ac2}$$ on $\Omega$, where the $m_n \in {\mbox{$\mathbb{N}$}}$. This looks like a natural generalization of $(\ref{ac1})$, however one quickly sees that there is a hidden condition in $(\ref{ac1})$ which is not in $(\ref{ac2})$, which results in $(\ref{ac2})$ not corresponding to a ‘harmonic class’. Namely, in $(\ref{ac1})$, because $h_n = k_n^{-1}$, we automatically have $|h_n| \rightarrow f^{-1}$ uniformly. We therefore define ${\mbox{${\mathcal M}$}}_A(\Omega)$ to be the set of strictly positive functions $f$ on $\Omega$ which satisfy $(\ref{ac2})$ and also: $$f(w)^{-1} = \lim_n \Vert H_n(w) \Vert_2 \; \; \label{ac3}$$ uniformly on $\Omega$. This might loosely be called a ‘tight convex approximation in modulus’. One finds that now $\log {\mbox{${\mathcal M}$}}_A(\Omega)$ is a ‘harmonic class’, which contains $H_A(\Omega)$. This is shown in Part C.
We show that $f \in {\mbox{${\mathcal M}$}}_A(\Omega)$ if and only if $Af$ is a strong Morita equivalence $A-A$-bimodule, with ‘inverse bimodule’ $Af^{-1}$. The notion of strong Morita equivalence was defined and studied in [@BMP], but for function algebras and singly generated $A$-modules $X$, this notion may be viewed as a generalization of the notion $X \cong A$. Indeed $X$ is a ‘rank one’ strong Morita equivalence bimodule if and only if $X \cong A$ almost $A$-isometrically. We display certain groups that are naturally associated with a uniform algebra $A$, such as the Picard group. We also illustrate the notions considered with some interesting examples.
In Part C, we generalize still further. We now allow topologically singly generated function modules. These correspond to submodules $(Af)^{\bar{}}$, the closure taken in $C(\Omega)$, where $f$ is now allowed to be nonnegative continuous function on $\Omega$. We define a larger harmonic class $\log {\mbox{${\mathcal R}$}}(\Omega)$, which contains the harmonic classes mentioned earlier. Just as the ${\mbox{${\mathcal M}$}}$ class corresponds to Morita equivalence, the ${\mbox{${\mathcal R}$}}$ class corresponds to the more general notion of ‘rigged module’. Rigged modules were intended to be a generalization of the notion of ‘Hilbert C$^*-$module’ and were studied in [@Bhmo; @BMP]; but in our (singly generated) situation these are the modules $X$ for which the identity map $X \rightarrow X$ factors asymptotically, via contractive $A$-module maps, through the ‘free’ $A$-modules $A^{(n)}$. This asymptotic factorization may be viewed as another generalization of the statement $X \cong A$ $A$-isometrically.
We show that for $A = A({\mbox{$\mathbb{D}$}})$, the disk algebra, the topologically singly generated $A$-rigged modules are exactly (up to $A$-isometric isomorphism) the modules of the form $(Af)^{\bar{}}$, where $f$ is a continuous function on $\bar{{\mbox{$\mathbb{D}$}}}$ such that $f = |\phi|$ for some outer function $\phi \in H^\infty$. This corresponds to the continuous nonnegative functions on ${\mbox{$\mathbb{T}$}}$ whose logarithm is integrable.
We admit that one of the purposes of Parts B and C, was to begin to illuminate the function algebra case of the theory of Morita equivalence and rigged modules, which the first author and various coauthors have developed over the years (see [@Bnat] for a leisurely survey), and in particular to see the connections with some problems concerning function algebras. Conversely, what we do here may lead to progress in the noncommutative situation.
Notations and definitions
=========================
For a compact set $\Omega $ we denote by $C\left( \Omega \right) $ (resp. $C_{\mathbb{R}}\left( \Omega \right) $) the space of all complex valued (resp. real valued) continuous functions on $\Omega $. For a function algebra $A$ on $\Omega$ we will write $M_{A}$ for the maximal ideal space of $A$, and $\partial A$ for the Shilov boundary of $A$. Then $C(\partial A),C(M_{A}),$ and $C(\Omega )$, may be regarded, respectively, as the minimal, maximal commutative, and generic commutative, C$^{\ast }-$algebra generated by $A$. $A$ may be viewed as a closed subalgebra of continuous functions on either of these three compact spaces. For a set of functions $\mathcal{E}$ we will write $\mathcal{E}_{+}$ for the nonnegative functions in $\mathcal{E}$, and $\mathcal{E}^{+}$ for the strictly positive functions in $\mathcal{E}$. We will refer very often to the following important subsets of $C(\Omega )$ which may be associated with $A$:
- $A^{-1}$ - the set of invertible elements of $A$,
- $P=\{f\in C(\Omega ):f=|g|,g\in A\}$,
- $Q=\{f\in C(\Omega ):f=|g|,$ where $g\in A^{-1}\}$,
- $\mathcal{F}=\{f=\sum_{i=1}^{n}|g_{i}|^{2}\in C(\Omega ):n\in \mathbb{N},g_{1},\cdots g_{n}\in A\}$,
- $\mathcal{G}=\bar{\mathcal{F}}$, the uniform closure of $\mathcal{F}$ in $C(\Omega )$,
- $G_{A}(\Omega )$ is the closure of $ReA$ in $C_{{\mbox{$\mathbb{R}$}}}(\Omega )$,
- $H_{A}(\Omega )$ is the closure of $\{\log |a|:a\in A^{-1}\}$ in $C_{{\mbox{$\mathbb{R}$}}}(\Omega )$.
If we wish to specify the dependence on $\Omega$, we will write, for example, $Q(\Omega)$. It is well known that $G_{A}(\Omega ) \subset H_{A}(\Omega )$, and they are both harmonic classes in the sense of the introduction (a proof is also contained in Part C). We recall that $A$ is called a ‘Dirichlet algebra’ (resp. ‘logmodular algebra’) if $G_{A}(\partial A)=C_{\mathbb{R}}(\partial A)$ (resp. $H_{A}(\partial
A)=C_{\mathbb{R}}(\partial A)$). The disk algebra $A({\mbox{$\mathbb{D}$}})$ is Dirichlet (and consequently logmodular). Indeed this is exactly saying that the ordinary Dirichlet problem (of harmonic extension from the boundary) can be solved on the circle ${\mbox{$\mathbb{T}$}}= \partial A$. This of course, was Gleason’s original reason for the name ‘Dirichlet algebra’ (see [@Gl]).
We write $\hat{}:A\rightarrow C(M_{A})$ for the Gelfand transform, and we often use the same symbol to denote an element of an algebra and the corresponding Gelfand transform. There is a surprising equivalence relation on $M_{A}$: $\phi ,\psi \in M_{A}$ are equivalent if and only if $\Vert \phi -\psi \Vert <2$. The distinct equivalence classes are called the (Gleason) parts [@Gl]. They are $\sigma -$compact subsets of $M_{A}$. We refer the reader to [@sto] or [@Gam] as general references on function algebras.
If $X_{1}$ and $X_{2}$ are Banach $A$-modules then we write $X_{1}\cong
X_{2} $ $A$-isometrically (resp. almost $A$-isometrically), if they are isometrically $A$-isomorphic (resp. almost isometrically $A$-isomorphic). As we said earlier, the latter term means that for any $\epsilon > 0$, there exists an $A$-module isomorphism $T : X_1 \rightarrow X_2$ with $\Vert T \Vert \Vert T^{-1} \Vert < 1 + \epsilon$. If, in the above, we replace the requirement that $T$ be an $A$-module map, with it being linear, then we simply say that $X_1 \cong X_2$ almost isometrically.
Whenever we use the words ‘singly generated’, it will be assumed, unless otherwise qualified, to have the topological connotation. Thus a ‘singly generated’ Banach module $X$ over $A$ has an element $x$ such that the closure of $Ax$ is $X$. An [*algebraically*]{} singly generated module $X$ has $Ax = X$ for some $x \in X$.
We will write $A^{(n)}$ for the space of $n$-tuples with entries in $A$. An element of $A^{(n)}$ will be called an $A$-tuple, and will often be regarded as a function $\Omega \rightarrow \mathbb{C}^{n}$.
For a Banach space $X$ we let $X_1$ be the unit ball of $X$, and we denote by $ext(X_1)$ the set of extreme points of the unit ball of $X$. If $A$ is a subspace of $C\left( \Omega \right) $, and $x$ is a point in $\Omega $ we use the same symbol $x$ to denote the corresponding functional on $A$ - namely, evaluation at $x$. There are several definitions of the Choquet boundary $Ch(A)$ of a linear subspace $A$ of $C\left( \Omega \right) $, here we adopt the following one: $$Ch(A)=\left\{ x\in \Omega :x\in ext(A^{\ast}_1)\right\} . \label{ChA}$$ For general information, elementary Choquet theory shows that this is the same as the set $$\{ x \in \Omega : \delta_x \; \text{is the
only probability measure on} \; \Omega \; \text{extending} \;
(\delta_x)_{|_A} \; \}$$ where $\delta_x$ is the Dirac mass considered as a functional on $C(\Omega)$. We shall not use this explicitly here. The Shilov boundary $\partial A$ is the closure of $Ch(A)$. As it may be necessary to distinguish between functions from $A$ and its restrictions to $\partial A$, we will denote by $A_{|\partial A}$ the set of these restrictions.
In the latter part of the paper, we will be working with operator spaces. However, usually, issues of ‘complete boundedness’ do not arise. This is because for a linear operator $T$ mapping into a subspace of a commutative C$^*-$algebras we have $\Vert T \Vert = \Vert T \Vert_{cb}$.
Part A.
Module isomorphisms
===================
It is well known [@N] that two uniform algebras are linearly isometric, that is isometric as Banach spaces, if and only they are isomorphic as algebras. In this section we show that similar results hold for function modules of the form $Af$. To explain the main idea, let $A$ be a uniform algebra on $\Omega$, and let $f_{1},f_{2}$ be strictly positive continuous functions on $\Omega$. Suppose that there is an invertible function $g\in A$ such that $\left| g\right| =\frac{f_{1}}{f_{2}},$ then $$Af_{1}\ni af_{1}\longmapsto agf_{2}\in Af_{2}$$ is an $A$-module isometry between $Af_{1}$ and $Af_{2}$. Suppose now that we also have a homeomorphism $\varphi$ of $\Omega$ onto itself such that $\left\{ a\circ\varphi:a\in A\right\} =A$, and $\frac{f_{1}\circ\varphi }{f_{2}}=\left| h\right| $, for some invertible element $h$ of $A$. Then $$Af_{1}\ni af_{1}\longmapsto (a\circ\varphi) hf_{2}\in Af_{2}, \label{can}$$ is a linear isometry between $Af_{1}$ and $Af_{2}$, but this time it is not an $A$-module isometry unless $\varphi$ is the identity map. We shall show that any linear isometry between modules $Af_{1}$ and $Af_{2}$ is essentially of the form (\[can\]).
We first need to introduce some technical results.
\[ext\]Suppose that $A$ is a subspace of $C\left( \Omega\right) $ separating the points of $\Omega$. Then
1. the original topology $\tau$ of $\Omega$ is identical with the weak$^\ast$ topology $\sigma^{\ast}$ on $\Omega$ considered as a subset of the unit ball of the dual space $A^{\ast}$,
2. if $F$ is an extreme point of the unit ball of the dual space $A^{\ast}$ then there is a point $x$ in $Ch\left( A\right) $ and a scalar $\alpha$ of absolute value one such that $F=\alpha x$,
3. for any $a\in A$ we have $$\left\| a\right\| =\sup\left\{ \left| a\left( x\right) \right| :x\in
Ch(A)\right\} . \label{KJ1}$$
**(i)** It is clear that the identity map $I:\left( \Omega,\tau\right)
\rightarrow\left( \Omega,\sigma^{\ast}\right) $ is continuous. Since $\left(
\Omega,\tau\right) $ is compact it follows that the topologies are identical.
**(ii)** Assume that $F$ is an extreme point of the unit ball of the dual space $A^{\ast}$. By the Krein-Milman Theorem, the set of norm one extensions of $F$ to $C\left( \Omega\right) $ has an extreme point $\mu$. It is easy to check that $\mu$ is also an extreme point of $C\left( \Omega\right)^{\ast}_1$ . The argument is concluded by appealing to the very well-known fact that any extreme point of the last space is of the form $\alpha x$ with $x\in \Omega,\left|
\alpha\right| =1.$
**(iii)** Let $a$ be a norm one element of $A$. By the Krein-Milman Theorem there is an extreme point $F$ of the unit ball of $A^{\ast}$ such that $F\left( a\right) =1$. Thus (\[KJ1\]) follows from the previous part.
\[ga\] Suppose that $A$ is a uniform algebra on $\Omega,$ that $x_{0}$ is in the Choquet boundary of $A,$ and that $f$ and $p$ are strictly positive continuous functions on $\Omega$. Then there is $a\in A$ such that $(af)\left( x_{0}\right) =p\left( x_{0}\right) $ and $\left|
af \right| \leq p.$
Moreover only points from the Choquet boundary have this property.
Follows from [@Gam] II.12.
\[Choquet\] Suppose that $A$ is a uniform algebra on $\Omega$ and that $f $ is a strictly positive continuous functions on $\Omega$. Then $Ch\left(
A\right) \subset Ch(Af)\subset\Omega.$
To prove the first inclusion assume that $x\in Ch\left( A\right)$, and assume that $x=\frac{1}{2}F_{1}+\frac{1}{2}F_{2}$ where $F_{1},F_{2}$ are norm one functionals on $Af.$ Let $\mu _{1},\mu _{2}$ be norm one extensions of $F_{1},F_{2}$ to functionals on $C\left( \Omega \right) $. By Lemma \[ga\], there is a net $a_{\gamma }$ in $A$ such that $\left\| a_{\gamma
}\right\| =a_{\gamma }\left( x\right) =1$ and $a_{\gamma }\rightarrow 0$ uniformly on compact subsets of $\Omega \backslash \left\{ x\right\} $. We have $$f\left( x\right) = (a_{\gamma }f)\left( x\right) =\frac{1}{2}\int_{\Omega
}a_{\gamma }fd\mu _{1}+\frac{1}{2}\int_{\Omega }a_{\gamma }fd\mu
_{2}\rightarrow \frac{f\left( x\right) }{2}\left( \mu _{1}+\mu _{2}\right)
\left( \left\{ x\right\} \right) ,$$ hence $\mu _{1}\left( \left\{ x\right\} \right) =1=\mu _{2}\left( \left\{
x\right\} \right) $. Since $\left\| \mu _{i}\right\| \leq 1$ we get $\mu
_{1}=x=\mu _{2}$.
Notice that, in general, $Ch(A)$ may be a proper subset of $Ch(Af)$. If, for example, $A$ is equal to the disk algebra, if $\Omega$ is the closed unit disk $\bar{{\mbox{$\mathbb{D}$}}}$, and if $f\left( z\right) = 2-\left| z\right| $, then $Ch\left( A\right) =\partial A=\partial\mathbb{D}$, while $0\in Ch(Af)$.
\[notAA\] Assume that $A$ and $B$ are uniform algebras on compact sets $\Omega_1,
\Omega_2$ respectively, and that $f_{1},f_{2}$ are strictly positive continuous functions on $\Omega_1,
\Omega_2$ respectively. Suppose that there is a surjective linear isometry $T:Af_{1}\rightarrow Bf_{2}$. Then there is an invertible element $h$ of $B,$ and a homeomorphism $\varphi $ of $\Omega_2$ onto a subset $\varphi \left( \Omega_2 \right)$ of $M_{A}$, such that
- the map $a \mapsto a \circ \varphi$ is an isometric isomorphism of $A$ onto $B$,
- $\varphi \left( \partial B\right) =\partial A$,
- $\frac{f_{1}\circ \varphi }{f_{2}}_{|\partial
(Bf_{2})}= \left| h\right| _{|\partial (Bf_{2})}$, and
$$T\left( af_{1}\right) =(a\circ \varphi) hf_{2}
\; \; \; \text{on} \; \Omega_2 \; , \; \text{for } \;
a\in A . \label{th1}$$
Moreover, if $A = B$, then $T$ is an $A$-module isometry if and only if $\varphi $ is equal to the identity map.
Before we prove this theorem, we give some consequences.
Assume that $A$ is a uniform algebra on a compact set $\Omega$, and that $f$ is a strictly positive continuous function on $\Omega$. Then $A$ and $Af$ are linearly isometric if and only if they are $A$-isometric.
Assume that $T:A\rightarrow Af$ is a linear isometry. Set $f_{1}=1 , f_{2}=f$ in the last Theorem, and let $h$ be as in that Theorem. We can define a module map $S:A\rightarrow Af$ by $$S\left( a\right) \left( x\right) =ahf\left( x\right) \text{, }\qquad\text{for }a\in A,\quad x\in\Omega.$$ That $S$ is an isometry follows from the last Theorem and Lemma \[ext\] (iii).
In general it is not true that modules $Af_{1}$ and $Af_{2}$ are linearly isometric if and only if they are $A$-module isometric. Even if $f_1,f_2 \in C(\partial A)^+$, this is not true. Assume, for example, that $A$ is equal to the product of the disk algebra and the two dimensional algebra $C\left( \left\{ -1,1\right\} \right)$. Set $\Omega=\partial A=\mathbb{T}\times\left\{ -1,1\right\} $, and define a map $\varphi : \Omega \rightarrow \Omega$ by $\varphi\left( z,j\right) =\left( z,-j\right)
$, let $f\in C\left( \mathbb{T}\right) ^{+}\backslash Q(A({\mbox{$\mathbb{D}$}}))$, and put $$f_{1}\left( z,j\right) =\left\{
\begin{array}{ccc}
1 & \text{for} & j=-1, \\
f\left( z\right) & \text{for} & j=1,
\end{array}
\right.$$ and $f_{2}\left( z,j\right) =f_{1}\left( z,-j\right). $ Define $T\left(
af_{1}\right) \left( z,j\right) = a\left( z,-j\right)
f_{2}\left( z,j\right) $. Then $T$ is a linear isometry from $Af_{1}$ onto $Af_{2}$. Assume now that there is an $A$-module isometry $S:Af_{1}\rightarrow Af_{2}$. By Theorem \[notAA\] there is an invertible element $h$ of $A$ such that $$S\left( af_{1}\right) =ahf_{2},\quad\text{for }a\in A.$$ Fix $\left( z,j\right) \in\mathbb{T}\times\left\{ -1,1\right\} $, and let $a_{n}$ be a sequence of norm one elements of $A$ convergent to zero almost uniformly on $\mathbb{T}\times\left\{ -1,1\right\} \backslash\left\{
(z,j)\right\} $. The norm of $af_{1}$ is convergent to $\left| f_{1}\left(
z,j\right) \right| $, while the norm of $ahf_{2}$ is convergent to $\left|
h\left( z,j\right) f_{2}\left( z,j\right) \right| $ hence $$\left| h\left( z,j\right) f_{2}\left( z,j\right) \right| =\left| f_{1}\left(
z,j\right) \right| \text{, for }\left( z,j\right) \in\mathbb{T}\times\left\{
-1,1\right\} \text{.}$$ Hence $\left| f\left( z\right) \right| =\left| f_{1}\left( z,1\right)
\right| =\left| h\left( z,1\right) f_{2}\left( z,1\right) \right| =\left|
h\left( z,1\right) \right| $. However $h\left( \cdot,1\right) $ is an invertible element of the disk algebra, contrary to our assumption about $f$.
\[NEED\] Assume that $A$ is a uniform algebra on a compact set $\Omega$, and that $f$ is a strictly positive continuous function on $\Omega$. Then the following conditions are equivalent:
1. $f\in Q$,
2. $A,Af$, and $Af^{-1}$ are $A$-module isometric,
3. $A,Af$, and $Af^{-1}$ are linearly isometric.
The only thing that is still not clear here is that ($2 \Rightarrow 1$). Note that if $T_1 : A \rightarrow Af$ is an $A$-isometric isomorphism, and if $T_1(1) = hf$, then $h \in A^{-1}$, $\Vert hf\Vert = 1$, and $|h|f \leq 1$ on $\Omega$. Since $\Vert a \Vert = \Vert ahf \Vert$ for all $a \in A$, it follows by a simple Choquet point argument, that $|h|f = 1$ on $Ch(A)$. (Since if $x \in Ch(A)$ with $|h(x)|f(x) < \alpha < 1$, let $V = \{w \in \Omega :
|h(w)|f(w) > \alpha \}$. By Lemma \[ga\], there exists $a \in A_1,
a(x) > \alpha,$ and $|a| < \frac{\alpha}{\Vert hf \Vert}$ on $V$. Hence $\alpha < \Vert a \Vert = \Vert ahf \Vert \leq \alpha$.)
A similar argument shows that there exists $k \in A^{-1}$ with $|k| f^{-1} \leq 1$ on $\Omega$, and $|k| f^{-1} = 1$ on $Ch(A)$. Hence $|hk| = 1$ on $Ch(A)$, and consequently on all of $\Omega$. Thus $$f \leq |h^{-1}| = |k| \leq f \;$$ everywhere, so that $f = |k| \in Q$.
\[wasla\] Assume that $\Omega$ is a Shilov boundary of a uniform algebra $A$, and that $f$ is a strictly positive continuous functions on $\Omega$. Then the following conditions are equivalent:
1. $f\in Q$,
2. $A$ and $Af$ are $A$-module isometric,
3. $A$ and $Af$ are linearly isometric.
Indeed, if $f_1, f_2 \in C(\partial A)^+$, then $Af_1 \cong Af_2$ $A$-isometrically, if and only if $f_1 f_2^{-1} \in Q$.
By Lemma \[Choquet\], $\Omega = \partial A = \partial(Af)$. Now it is clear that Theorem \[notAA\] gives ($3 \Rightarrow 1$). A similar argument proves the last statement. The rest of the numbered equivalences are clear.
The last Corollary is not valid without the assumption that $\Omega$ is equal to the Shilov boundary of $A.$ Indeed if we put $A=A\left( \mathbb{D}\right) $, $\Omega=\bar{{\mbox{$\mathbb{D}$}}}$, and if we define $f\left( z\right) = \frac{1+\left| z\right| }{2}$, then the map of multiplication by $f$ is an $A$-module isometry from $A$ onto $Af$. On the other hand, $f$ is not equal to an absolute value of a function from $A$.
The first part of the last corollary also follows from:
\[new2\] Suppose that $A$ is a uniform algebra on $\Omega$, and $f \in C(\Omega)^+$. Then $Af \cong A$ $A$-isometrically if and only if $Af \cong (Af)_{|(\partial A)}$ isometrically via the restriction map, and $f_{|(\partial A)} \in Q(\partial A)$.
The $(\Leftarrow)$ direction is trivial. The $(\Rightarrow)$ direction follows easily from the Theorem, but we will give a different proof, which will generalize later to the ‘almost isometric’ case. Assuming that $Af \cong A$ $A$-isometrically, as in the proof of Corollary \[NEED\], it follows that there exists a $h \in A^{-1}$ such that $\Vert a h f \Vert = \Vert a \Vert$, and that the latter statement implies that $|(hf)_{|(\partial A)}| = 1$ on $\partial A$, and $|hf| \leq 1$ on $\Omega$. Moreover for $w \in \Omega$, we have $$|(af)(w)| \leq |(a h^{-1})(w)| \leq
\Vert a h^{-1} \Vert_{\partial A} \leq
\Vert (af)_{|(\partial A)} \Vert \; .$$
We will assume that $A = B$ for simplicity, although the same argument works in general. Assume that $T:Af_{1}\rightarrow Af_{2}$ is a linear surjective isometry. Let $K_{i}$, for $i=1,2$, be the set of extreme points of unit ball in the dual space $\left( Af_{i}\right) ^{\ast}$. Let $T_{|K_{2}}^{\ast}$ be a restriction of the dual map $T^{\ast}$ to $K_{2}$. Since $T^{\ast}$ is a homeomorphism, in the weak\* topology, as well as being an isometry, $T_{|K_{2}}^{\ast}$ is a homeomorphism of $K_{2}$ onto $K_{1}$. Hence, for $\left| \alpha\right| =1$, and $x\in Ch\left( Af_{2}\right) $ we have $$T^{\ast}\left( \alpha x\right) =\chi\left( \alpha,x\right) \cdot
\varphi\left( \alpha,x\right) \text{, where }\left| \chi\left(
\alpha,x\right) \right| =1\text{, and }\varphi\left( \alpha,x\right) \in
Ch\left( Af_{1}\right) \text{.}$$ Since $T^{\ast}$ is linear, the functions $\varphi$ and $\chi$ depend only on $x,$ and we get the following representation of $T$: $$T\left( af_{1}\right) \left( x\right) =\chi\left( x\right) \cdot\left(
af_{1}\right) (\varphi\left( x\right)) \text{, \qquad for }a\in A,x\in
Ch\left( Af_{2}\right) ,$$ where $\varphi:Ch\left( Af_{2}\right) \rightarrow Ch(Af_{1})$ is a surjective homeomorphism and $\chi$ is a unimodular function.
It is easy to see that this implies that $\varphi$ is continuous on $Ch(Af_2)$. (Since if $x_i \rightarrow x_0$ but $\varphi(x_i) \rightarrow
x_1 \neq \varphi(x_0)$, then choose $a \in A$ with $a(x_1) = 0 \neq
a(\varphi(x_0))$. Then $T(af_1)(x_i) \rightarrow T(af_1)(x_0)$. However $|T(af_1)(x_i)| \leq D |a(\varphi(x_i))| \rightarrow 0$, but $T(af_1)(x_0) \neq 0$.)
For $a=1$ we get that $Tf_{1}\left( x\right) =\chi\left( x\right) \cdot
f_{1}(\varphi\left( x\right)) =f_{2}(x) \frac{\chi\left( x\right) \cdot
f_{1}(\varphi\left( x\right)) }{f_{2}(x)}$. Thus the restriction of $\frac{\chi \cdot
(f_{1}\circ\varphi)}{f_{2}}$ to $Ch(Af_2)$ is the restriction of a function in $A$. Also $\chi=\frac{Tf_{1}}{f_{1}\circ\varphi }.$ Hence it follows that $\chi$ is also continuous on $Ch(Af_2)$.
Since $T $ is surjective there is $a_{0}\in A$ such that $T\left( a_{0}f_{1}\right)
=f_{2}\cdot\left( \frac{\chi \cdot (f_{1}\circ\varphi)
}{f_{2}}\right) ^{2}$, as functions on $Ch\left( Af_{2}\right) .$ Hence $$f_{2}\cdot\left( \frac{\chi\cdot
(f_{1}\circ\varphi)}{f_{2}}\right)^{2}=
T\left( a_{0}f_{1}\right) =f_{2}\cdot\frac{\chi\cdot(\left(
a_{0}f_{1}\right) \circ\varphi)}{f_{2}},$$ as functions on $Ch\left( Af_{2}\right) $, so $$\frac{\chi}{f_{2}}= (\frac{a_{0}}{f_{1}})\circ\varphi.$$ Consequently we get $$T\left( af_{1}\right) \left( x\right) =f_{2}\left( x\right) \cdot (\left(
a_{0}a\right)( \varphi\left( x\right)))
\text{, \qquad for }x\in Ch\left(
Af_{2}\right) \text{, }a\in A. \label{brzeg}$$ Let $b_{0}\in A$ be such that $T\left( b_{0}f_{1}\right) =f_{2}$. We have $f_{2}=T\left( b_{0}f_{1}\right) =f_{2}\cdot (\left( a_{0}b_{0}\right)
\circ\varphi)$, so $a_{0}b_{0}=1$ on $Ch\left( Af_{1}\right) $. By Lemma \[Choquet\] the closure of $Ch\left( Af_{1}\right) $ contains the Shilov boundary of $A$ so $a_{0}b_{0}=1$ on $\partial A$, and consequently on $\Omega_1$. This proves that $a_{0}$ is an invertible element of $A$. Hence $\left\{ a_{0}\cdot a:a\in A\right\} =A$.
Note that the map $a \longmapsto a\circ\varphi$ is linear, multiplicative, one-to-one, and indeed is isometric since $Ch(A) \subset Ch(Af_1)$. As functions on $Ch(Af_2)$ we have: $$\left\{ a\circ\varphi:a\in A\right\} =\left\{ \left( a_{0}\cdot a\right)
\circ\varphi:a\in A\right\} =T\left( Af_{1}\right) /f_{2}=A. \label{brzeg2}$$ Thus $a\longmapsto a\circ\varphi$ may be viewed as an isometric automorphism $A \rightarrow A$. In particular $a_{0}\circ\varphi$ is an invertible element of $A$ and we have $$\frac{f_{1}\circ\varphi}{f_{2}}=\left| \frac{a_{0}\circ\varphi}{\chi}\right|
=\left| a_{0}\circ\varphi\right| \in Q$$ as required.
To finish the proof we need to show that $\varphi$ can be extended to a homeomorphism of $\Omega_2$ onto a subset of $M_A$, and that the formula (\[brzeg\]) remains valid on the entire set $\Omega_2$. Since $a\longmapsto a\circ\varphi$ is an automorphism of $A$, it is given by a homeomorphism of the maximal ideal space of the algebra. That is, $\varphi$ can be extended to a homeomorphism of $M_A$ onto itself mapping $\partial A$ onto $\partial A$. For a given $a$ in $A$, $T\left( af_{1}\right) $ and $f_{2}\cdot(\left( a_{0}a\right) \circ\varphi)$ are elements of $Af_{2}$ which, by (\[brzeg\]), are identical on the Choquet boundary of $A$. By dividing by $f_2$ if necessary, we see that these elements are identical at any point of $\Omega_2$ and we get (\[th1\]).
Another proof of this result, using the function multiplier algebra may be found in [@Bsh]. However it is the proof above which extends to the ‘almost isometric’ case.
Almost isometries.
==================
Recall that Banach spaces $X,Y$ are almost isometric if the Banach-Mazur distance between $X$ and $Y$ defined as $$d_{B-M}\left( X,Y\right) =\log \inf \left\{ \left\| T\right\| \left\|
T^{-1}\right\| :T:X\rightarrow Y\right\} ,$$ is equal to zero; two $A$-modules $X,Y$ are almost $A$-isometric if $$\log \inf \left\{ \left\| T\right\| \left\| T^{-1}\right\| :T:X\rightarrow
Y\right\} =0,$$ where this time the infimum is taken over the set of all $A$-isomorphisms. Of course two isometric Banach spaces are almost isometric, but even for the class of separable uniform algebras defined on subsets of a plane, almost isometric spaces need not be isometric [@J1]. Small bound isomorphisms between various classes of Banach spaces, primarily function spaces, have been investigated in a large number of papers, see for example [@BC; @J2; @J3; @RR].
\[into\]Assume that $A$ is a uniform algebra on a compact set $\Omega$, and that $f$ is a strictly positive continuous function on $\Omega$. Suppose that $0 < \epsilon < \frac{1}{3}$ is given, and suppose that $T$ is a surjective linear map from $A$ onto $Af$ such that $\left\| T\right\| \leq 1$, and $\left\| T^{-1}\right\| \leq1+\varepsilon.$ Then there is a subset $\Omega_{0}$ of the Shilov boundary of $Af$ and a surjective continuous map $\varphi:\Omega
_{0}\rightarrow Ch(A)$ such that $$\left| T\left( a\right) \left( x\right) -T\left( \mathbf{1}\right) \left(
x\right) \cdot a\circ\varphi\left( x\right) \right| \leq 4\varepsilon\frac{1+\varepsilon}{1-\varepsilon}\left\| a\right\| \text{, for }a\in A,\quad
x\in\Omega_{0} , \label{6.1}$$
and such that $$1-10\varepsilon\leq\left| T\left( \mathbf{1}\right) \left( x\right) \right|
\leq1\text{, \quad for }x\in\Omega_{0} . \label{6.10}$$ It follows that $$\sup\left\{ \left| (af)\left( x\right) \right| :x\in\Omega_{0}\right\}
\geq\left( 1- 15\varepsilon\right)
\left\| af\right\| \text{,\quad\ for }af\in
Af, \label{6.11}$$ and it also follows that the closure of $\Omega_{0}$ contains the Shilov boundary of $A$.
Let $\widetilde{T}:A\rightarrow Af_{|\partial(Af)}\subset C\left(
\partial(Af)\right) $ be defined by $\widetilde{T}\left( a\right) = (1+\epsilon)
T\left( a\right)
_{|\partial(Af)}$. We know from Lemma \[ext\] (iii), that the restriction map from $Af$ to $Af_{|\partial(Af)}$ is an isometry. By Theorem 6.1 of [@J2] applied to $\widetilde{T}$, there is a subset $\Omega_{0}$ of $\partial(Af)$, and a continuous function $\varphi$ from $\Omega_{0}$ onto $Ch(A)$ such that (\[6.1\]) holds .
Assume now that there is an $x_{0}\in\Omega_{0}$ with $\left| T\left( \mathbf{1}\right) \left( x_{0}\right) \right| <1-10\varepsilon$. At the beginning of the proof of Theorem 6.1 in [@J2], the set $\Omega_{0}$ is defined specifically as a subset of $\left\{ x\in\Omega:\left\| x\right\|
_{Af}> M\right\} $, where we denote by $\left\| x\right\| _{Af}$ the norm of the “evaluation at the point $x$” functional on $Af$, and where $M$ can be chosen to be any number satisfying $\frac{1-\varepsilon}{1+\varepsilon}>M>\frac
{1-\varepsilon}{1+\varepsilon}-\varepsilon^{2}$. Since $1-3\varepsilon
<\frac{1-\varepsilon}{1+\varepsilon}-\varepsilon^{2}$, there is a norm one element $a_{0}f$ of $Af$ such that $(a_{0}f)\left(
x_{0}\right) \geq1-3\varepsilon$. Put $a_{1}=T^{-1}\left( a_{0}f\right) $. We have $\left\| a_{1}\right\| \leq\left\| T^{-1}\right\| \leq 1+\varepsilon$ and $T\left( a_{1}\right) \left( x_{0}\right) \geq1-3\varepsilon$, while $$\left| T\left( \mathbf{1}\right) \left( x_0\right) \cdot a_{1}\circ
\varphi\left( x_{0}\right) \right| \leq\left| T\left( \mathbf{1}\right)
\left( x_0\right) \right| \left\| a_{1}\right\| \leq\left( 1-10\varepsilon
\right) \left( 1+\varepsilon\right) \text{.}$$ This contradicts (\[6.1\]) and shows (\[6.10\]).
To prove (\[6.11\]) assume that there is a norm one element $af$ of $Af$ such that $$\sup \left\{ \left| af\left( x\right) \right| :x\in \Omega _{0}\right\}
<\left( 1-15\varepsilon \right) .$$ Put $b=T^{-1}\left( af\right) $, let $\widetilde{x}\in Ch(A)$ be such that $\left| b\left( \widetilde{x}\right) \right| =\left\| b\right\| \geq 1$, and let $x_{1}\in \Omega _{0}$ be such that $\varphi \left( x_{1}\right) =\widetilde{x}$. By (\[6.10\]) we have that $$\left| T\left( \mathbf{1}\right) \left( x_1 \right) \cdot b\circ
\varphi \left( x_{1}\right) \right| \geq 1-10\varepsilon ,\text{ while }\left| Tb\left( x_{1}\right) \right| =\left| af\left( x_{1}\right) \right|
\leq 1- 15\varepsilon ,$$ which contradicts (\[6.1\]) and shows (\[6.11\]).
To finish the proof we need to show that (\[6.11\]) implies $\partial
A\subset\overline{\Omega}_{0}$. To this end, choose a point $x_{0}\in Ch(A)$ which is not in $\overline{\Omega}_{0}$. W.l.o.g. we may assume that $\Vert f \Vert \leq 1$. By Lemma \[ga\] there is an $a\in A$ such that $$\left\| a\right\| =1=a\left( x_{0}\right) \text{ and }\left| a\left(
x\right) \right| <\frac{1}{2}\min f\text{, for }x\in\overline{\Omega}_{0}.$$ By (\[6.11\]) we have $$\begin{aligned}
\min f & \leq\left| (af)\left( x_{0}\right) \right| \leq\left\|
af\right\| \\
& \leq\frac{1}{1-15\varepsilon}\sup\left\{ \left| (af)
\left( x\right) \right|
:x\in\Omega_{0}\right\} \\
& \leq\frac{1}{1-15\varepsilon}\frac{1}{2}\min f .
\end{aligned}$$ This contradiction proves that $\partial A\subset\overline{\Omega}_{0}$.
\[got\] Assume that $A$ is a uniform algebra on a compact set $\Omega$, and that $f$ is a strictly positive continuous function on $\Omega$. The following conditions are equivalent:
1. $A$ and $Af$ are almost $A$-isometric,
2. $A$ and $Af$ are almost isometric.
Moreover if $\Omega$ is equal to the Shilov boundary the above conditions are also equivalent to:
1. $f\in\overline{Q}^{+}.$
That (3) implies (1) implies (2) is left to the reader.
Assume that $A$ and $Af$ are almost isometric, and let $T:A\rightarrow Af$ be such that $\left\| T\right\| \leq 1$ and $\left\| T^{-1}\right\| \leq
1+\varepsilon $. Since $T$ is surjective and $\frac{T\left( \mathbf{1}\right) }{f}\in A$, there is an $a_{0}\in A$ such that $Ta_{0}=\left( \frac{T\left( \mathbf{1}\right) }{f}\right) ^{2}f$. By (\[6.1\]) we get $$\begin{aligned}
\left\| \left( \frac{T\left( \mathbf{1}\right) }{f}\right) ^{2}f-T\left(
\mathbf{1}\right) \cdot a_{0}\circ \varphi \right\|_{\Omega_0}
& \leq 4\varepsilon
\frac{1+\varepsilon }{1-\varepsilon }\left\| a_{0}\right\| \\
& \leq 4\varepsilon \frac{1+\varepsilon }{\left( 1-\varepsilon \right) ^{2}}\left\| \left( \frac{T\left( \mathbf{1}\right) }{f}\right) ^{2}f\right\| \\
& =4\varepsilon \frac{1+\varepsilon }{\left( 1-\varepsilon \right) ^{2}}\left\| \left( T\left( \mathbf{1}\right) \right) ^{2}f^{-1}\right\| \\
& \leq 4\varepsilon \frac{1+\varepsilon }{\left( 1-\varepsilon \right) ^{2}}\left\| f^{-1}\right\| ,\end{aligned}$$ so by (\[6.10\]) we have $$\left| T\left( \mathbf{1}\right) \left( x\right) -f\left( x\right) \cdot
a_{0}(\varphi \left( x\right)) \right| \leq 4\varepsilon \frac{1+\varepsilon }{\left( 1-10\varepsilon \right) \left( 1-\varepsilon \right)
^{2}}\left\| f^{-1}\right\| \left\| f\right\| \text{, for }x\in \Omega _{0}.
\label{KJ5}$$ From (\[6.1\]) and (\[KJ5\]) we obtain $$\left| T\left( a\right) \left( x\right) -f\left( x\right) \cdot \left(
aa_{0}\right)(\varphi \left( x\right)) \right| \leq \varepsilon
^{\prime }\left\| a\right\| \text{, for }a\in A,\quad x\in \Omega _{0},
\label{KJ6}$$ where $\varepsilon ^{\prime }=4\varepsilon \frac{1+\varepsilon }{\left(
1-10\varepsilon \right) \left( 1-\varepsilon \right) ^{2}}\left\|
f^{-1}\right\| \left\| f\right\| +4\varepsilon \frac{1+\varepsilon }{1-\varepsilon }$.
Let $b_{0}\in A$ be such that $Tb_{0}=f$. If $\varepsilon $ is sufficiently small, then (\[KJ6\]) gives $$\begin{aligned}
\left\| \mathbf{1}-\left( a_{0}b_{0}\right) \circ \varphi \right\| _{\Omega
_{0}}& \leq \left\| f^{-1}\right\| \left\| f-f\cdot (\left( a_{0}b_{0}\right)
\circ \varphi) \right\| _{\Omega _{0}} \\
& =\left\| f^{-1}\right\| \left\| Tb_{0}-f\cdot (\left( a_{0}b_{0}\right)
\circ \varphi) \right\| _{\Omega _{0}} \\
& <\frac{1}{2}\; \; .
\end{aligned}$$
Hence $Re \left( a_{0}b_{0}\right) \left( x\right) \geq
\frac {1}{2}$ for any $x\in Ch(A)$, and consequently for any $x$ in the maximal ideal space of $A $. It follows that $a_{0}$ and $b_{0}$ are invertible in $A$. By (\[KJ5\]) and (\[KJ6\]) the function $\frac{T\left( \mathbf{1}\right) }{f}\cdot\frac{T\left( a_{0}^{-2}\right) }{f}$ is approximately equal, on $\Omega_{0} $, to $(a_{0}\circ\varphi)\cdot (a_{0}^{-1}\circ\varphi)=1$. Since $\bar{\Omega_0}$ contains $\partial A$, the function $\frac{T\left( \mathbf{1}\right) }{f}\cdot\frac{T\left(
a_{0}^{-2}\right) }{f}-\mathbf{1}$ is approximately equal to zero on the maximal ideal space of $A$. Thus, if $\varepsilon$ is sufficiently small, $\frac{T\left( \mathbf{1}\right) }{f}\cdot\frac{T\left( a_{0}^{-2}\right) }{f}
$ is an invertible element of $A$. Consequently $\frac{T\left( \mathbf{1}\right) }{f}$ is invertible.
We can now define an $A$-module isomorphism $S:A\rightarrow Af$ by $$Sa=T\left( \mathbf{1}\right) a\text{, for }a\in A\text{.}$$ Fix $a\in A$ and let $\widetilde{a}\in A$ be such that $T\widetilde
{a}=Sa=T\left( \mathbf{1}\right) a$. By Theorem \[into\], $T\widetilde{a}$ is close, on $\Omega_{0}$, to $T\left( \mathbf{1}\right) \widetilde{a}\circ\varphi$. It follows that $a\approx\widetilde{a}\circ\varphi$ on $\Omega_{0}$. Consequently by (\[6.11\]) and the fact that $T$ is an ‘almost isometry’, we get $\left\| a\right\| \approx\left\| \widetilde{a}\circ\varphi\right\| =\left\|
\widetilde{a}\right\| \approx\left\| T\widetilde{a}\right\| =\left\|
Sa\right\|$. Thus $S$ is also an ‘almost isometry’.
Now assume that $\Omega$ is equal to the Shilov boundary. By (\[6.10\]) we have that $ \left| T\left( \mathbf{1}\right) \right| \approx1$ on $\overline{\Omega }_{0}=\partial A.$ As we proved before, $\frac{T\left( \mathbf{1}\right) }{f}$ is an invertible element of $A$. It follows that $f\in\overline {Q}^{+}.$
We have the following complement to the previous corollary (cf. Corollary \[new2\]):
\[new\] Suppose that $A$ is a uniform algebra on $\Omega$, and that $f \in C(\Omega)^+$. Then $Af \cong A$ almost $A$-isometrically if and only if $Af \cong (Af)_{|(\partial A)}$ isometrically via the restriction map, and $f_{|(\partial A)} \in \bar{Q}^+(\partial A)$.
This proceeds almost identically to the proof of \[new2\]. As in the proof of Corollary \[NEED\], it follows that for all $\epsilon > 0$, there exists a $h_\epsilon \in A^{-1}$ such that $$(1-\epsilon)\Vert a \Vert \leq
\Vert a h_\epsilon f \Vert \leq (1+\epsilon) \Vert a \Vert \; \; ,$$ and that the latter statement implies that $|(h_\epsilon f)_{|(\partial A)}| \approx 1$ on $\partial A$, and $|hf| \leq 1+ \epsilon$ on $\Omega$. Thus $f_{|(\partial A)} \in \bar{Q}^+(\partial A)$. Moreover for $w \in \Omega$, we have $$|(af)(w)| \leq (1+ \epsilon) |(a h_{\epsilon}^{-1})(w)|
\leq (1+ \epsilon) \Vert a h_{\epsilon}^{-1} \Vert_{\partial A}
\leq (1 + \epsilon)^2 \Vert (af)_{|(\partial A)} \Vert \;
.$$ Since $\epsilon > 0$ is arbitrary, $\Vert af \Vert =
\Vert af \Vert_{(\partial A)}$.
\[notA\] For any strictly positive $f\in C(\Omega )$, the following are equivalent.
- $f\in \bar{Q}^{+}$,
- $A\cong Af\cong Af^{-1}$ almost $A$-isometrically,
- $A\cong Af\cong Af^{-1}$ linearly almost isometrically.
We will only prove $(ii)\Rightarrow \left( i\right)$. That $(iii)$ implies $(ii)$ follows from what we just did. The easy implications $(i) \Rightarrow (ii) \Rightarrow (iii)$ are left to the reader.
If $T_{1}:A\rightarrow Af$ and $T_{2}:A\rightarrow Af^{-1}$ are module isomorphisms, let $t_1 = T_{1}(\mathbf{1})$ and $t_2 =
T_{2}(\mathbf{1})$. Then $T_{1}(a)=t_{1} a$, $T_{2}(a)=t_{2} a$, and $t_{1} f^{-1}$, $t_{2} f$ are invertible elements of $A$. If $\left\| T_{1}\right\| \leq 1+\varepsilon ,\left\| T_{2}\right\| \leq
1+\varepsilon ,$ then $$\left| t_{1}\left( x\right) \right| \leq
1+\varepsilon \quad \text{and\quad }\left| t_{2}
\left( x\right) \right| \leq 1+\varepsilon ,\quad \text{ for }x\in \Omega
\label{KJ7}$$ and consequently $$\left| t_{1}\left( x\right) t_{2}\left( x\right)
\right| \leq \left( 1+\varepsilon \right) ^{2},\quad
\text{ for }x\in \Omega .$$ Suppose that $\left\| T_{1}^{-1}\right\| \leq 1,\left\| T_{2}^{-1}\right\| \leq 1$. If $\left| t_{1}\left( x_{0}\right) \right| =1-r<1$ for some $x_{0}\in Ch(A)$, then by Lemma \[ga\] there is $a\in A$ such that $a\left( x_{0}\right) =1=\left\| a\right\| $ and $\left|
a\left( x\right) \right| <\frac{1}{2+\varepsilon }$ for $x\in \left\{ x:\left|
t_{1}\left( x\right) \right| >1-\frac{r}{2}\right\} $. We get $\left\| T_{1}\left( a\right) \right\|
=\left\| t_{1} a\right\| \leq \max \left\{ \frac{1}{2+\varepsilon
}\left\| T_{1}\right\| ,1-\frac{r}{2}\right\} <1$. This contradicts the assumption that $\left\| T_{1}^{-1}\right\| \leq 1$.
Hence $$\left| t_{1}\left( x\right) \right| \geq 1\quad
\text{and\quad }\left| t_{2}\left( x\right) \right|
\geq 1,\quad \text{ for }x\in Ch(A),$$
so $$\left| t_{1} \left( x\right) t_{2}\left( x\right)
\right| \geq 1,\quad \text{ for }x\in Ch(A).$$ However $t_{1} t_{2}$ is invertible, so it attains minimum on $Ch(A)$. Hence $$1\leq \left| t_{1} \left( x\right) t_{2}
\left( x\right) \right| \leq \left( 1+\varepsilon
\right)^{2},\quad \text{ for }x\in \Omega .$$ By (\[KJ7\]) it follows that $$\frac{1}{1+\varepsilon }\leq \left| t_{2} \left(
x\right) \right| \leq 1+\varepsilon ,\quad \text{ for }x\in \Omega .$$ Thus $\left\| f-\left| t_{2} \right| f \right\| \leq
\varepsilon \left\| f\right\| $; and since $\varepsilon >0$ is arbitrary we get $f\in \bar{Q}^{+}$.
\[prev\] If $X$ is a Banach $A$-module over a function algebra $A$, and if $X$ is almost $A$-isometric to a function $A$-module, then $X$ is a function $A$-module.
We use the injective tensor norm characterization of function modules [@BLM]. Suppose that $T_\epsilon$ is the $\epsilon$-isomorphism. Then for $a_1, \cdots , a_n \in A$, and $x_1, \cdots , x_n \in X$, we have: $$\Vert \sum_i a_i T_\epsilon(x_i) \Vert \leq
\Vert \sum_i a_i \otimes T_\epsilon(x_i) \Vert_\lambda
\leq \Vert T_\epsilon \Vert \Vert \sum_i a_i \otimes x_i\Vert_\lambda \; \; ,$$ where $\lambda$ is the injective tensor norm. Thus: $$\Vert \sum_i a_i x_i \Vert = \lim_{\epsilon \rightarrow 0}
\Vert T_\epsilon (\sum_i a_i x_i) \Vert
\leq \Vert \sum_i a_i \otimes x_i\Vert_\lambda \; \; .$$ By [@BLM], $X$ is a function $A$-module.
\[Motat21\] Let $X$ be a Banach $A$-module. The following are equivalent:
- $X$ is an algebraically singly generated, faithful function $A$-module, and $X \cong A$ almost linearly isometrically,
- $X \cong A$ almost $A$-isometrically,
- There exists $f \in \bar{Q}^+$ such that $X \cong Af$ $A$-isometrically.
Assume (i). As we said early in the introduction, the first two conditions in (i), together with a corollary in §3 of [@Bsh], shows that $X \cong Af$ $A$-isometrically, where $f \in C(\Omega)^+$, for some $\Omega$ on which $A$ acts as a function algebra. Now (ii) follows from Corollary \[got\]. By Corollary \[new\], $Af \cong (Af)_{|(\partial A)}$ and $f_{|(\partial A)} \in
\bar{Q}^+(\partial A)$, showing (iii).
Given (ii), it follows by Lemma \[prev\] that $X$ is a function $A$-module, and now (i) is clear. Clearly (iii) implies (i).
A similar result and proof holds with almost isometries replaced by isometries.
We noticed that the results in this section have been stated for a single function $f$, rather then for a pair of functions $f_{1},f_{2}$, like results in the previous section concerning isometries. There are analogous results describing almost isometries between modules $Af_{1}$ and $Af_{2}$, however they are much more technical and involve not a single automorphism $\varphi :\Omega
\rightarrow \Omega $ but a sequence of homeomorphisms between the Choquet boundaries of $Af_{1}$ and $Af_{2}$. The following theorem, which may be of an independent interest, can be proven using methods similar to that of the proof of Theorem 6.1 of [@J2]. It may then be used to extend the last few results to almost isometries between modules $Af_{1}$ and $Af_{2}.$
Assume that $\Omega$ is the Shilov boundary of a uniform algebra $A$, and that $f_{1},f_{2}$ are strictly positive continuous functions on $\Omega$. Suppose that $T:Af_{1}\rightarrow Af_{2}$ is a surjective linear isomorphism such that $\left\| T\right\| \leq1+\varepsilon$ and $\left\| T^{-1}\right\|
\leq1+\varepsilon$ where $\varepsilon<\varepsilon_{0}$ (an absolute constant). Then there is a dense subset $S$ of $\Omega$ containing the Choquet boundary of $A,$ a continuous bijection $\varphi:S\rightarrow Ch(A),$ and a continuous unimodular function $\chi$ such that $$\left\| T\left( af_{1}\right) -\chi\cdot\left( af_{1}\right) \circ
\varphi\right\| \leq\varepsilon^{\prime}\left\| af_{1}\right\| \text{,\quad\
for }a\in A\text{,}$$ where $\varepsilon^{\prime}\rightarrow0$ as $\varepsilon\rightarrow0$.
If the functions $f_1,f_2 \in {\mbox{${\mathcal M}$}}_A$, then a simple version of an ‘almost isometry’ result may be found in Theorem \[gen\].
No doubt there are also versions of all these results for a pair of function algebras $A$ and $B$, and function modules $Af_1$ and $Bf_2$, but that will take us a little further afield from our main concerns.
Part B.
Some observations concerning approximations in modulus
======================================================
In this section we make various observations concerning the sets $P,Q,\mathcal{G},\mathcal{F,} G_{A}(\Omega), H_{A}(\Omega )$ and ${\mbox{${\mathcal M}$}}_A(\Omega)$ defined in the Introduction, Notation and Definitions section. These sets will play crucial role in the next section when we study singly generated bimodules.
We already observed that $C(\Omega )^{+}
= \bar{Q}^{+}$ if and only if $A$ is logmodular on $\Omega $. The question of when $C(\Omega )_{+}=\bar{P}$ (resp. $C(\Omega )_{+}=\mathcal{G}$) has been studied by Mlak, Glicksberg, Douglas and Paulsen, and others. An algebra with this property is called ‘approximating in modulus’ (resp. ‘convexly approximating in modulus’). Just as in the usual proof that $A$ may only be logmodular on $\partial A$, one can show using Urysohn’s lemma, Lemma \[ga\], and the fact that a function in $A$ has maximum modulus achieved on $\partial A$, that either of these ‘approximating in modulus’ properties forces $\Omega = \partial A$.
For example, Dirichlet algebras, such as the disk algebra $A(\mathbb{D})$ considered as functions on the circle, are logmodular, and approximating in modulus. Glicksberg gave the following sufficient condition: If the inner functions (that is, functions in $A$ which have constant modulus 1 on $\Omega $) separate points of $\Omega $, then $A$ is approximating in modulus.
By the Stone-Weierstrass theorem it is easy to see that for any function algebra $A$ on $\Omega $ the set $\mathcal{F}-\mathcal{F}$ is dense in $C_{{\mbox{$\mathbb{R}$}}}(\Omega)$. However this does not imply that $\mathcal{F}$ is dense in $C(\Omega )_{+}$, or equivalently, that $A$ is convexly approximating in modulus. Indeed we have the following:
Let $A$ be a function algebra. If $Ch(A)\neq\partial A$ then $A$ is not convexly approximating in modulus on $\partial A$. That is, $\mathcal{G}\neq
C\left( \partial A\right) _{+}$.
Let $x_{0}\in \partial A,$ let $V\subset \partial A$ be an open neighborhood of $x_{0},$ and let $f\in C\left( \partial A\right) _{+}$ be such that $$f\left( x_{0}\right) =1=\left\| f\right\| \text{, and }f\left( x\right) =0\text{ for }x\in \partial A-V\text{.}$$ Assume that $\mathcal{G}=C\left( \partial A\right) _{+}$, and let $g_{1},...,g_{n}\in A$ be such that $$\left| f-\sum_{j=1}^{n}\left| g_{j}\right| ^{2}\right| \leq \frac{1}{4}.$$ Multiplying $g_{1},...,g_{n}$ by appropriate numbers of absolute value $1$ we may assume that $g_{j}\left( x_{0}\right) \in \mathbb{R}_{+}$. Then $$\frac{3}{4}\leq \left( \sum_{j=1}^{n}\left| g_{j}\right| ^{2}\right) \left(
x_{0}\right) =\sum_{j=1}^{n}g_{j}\left( x_{0}\right) ^{2}\leq \frac{5}{4}.$$ Put $$g=\frac{\sum_{j=1}^{n}g_{j}^{2}}{\sum_{j=1}^{n}g_{j}\left( x_{0}\right) ^{2}}\in A.$$ We have $$g\left( x_{0}\right) =1\text{, }\left\| g\right\| \leq \frac{5}{3}\text{,
and }\left| g\left( x\right) \right| \leq \frac{1}{3}\text{ for }x\in
\partial A-V\text{.}$$ By the Bishop ”$\frac{1}{4}-\frac{3}{4}$” criterion ([@Gam] Th. 11.1 page 52 and remark on p. 59) $x_{0}$ is a p-point of $A$. That is $x_{0}\in Ch(A)$.
\[Gmax\] Suppose that $f^2 \in {\mbox{${\mathcal G}$}}(\Omega)$. Then $f$ achieves its norm on $\partial A$. Indeed $Af \cong
(Af)_{|(\partial A)}$ $A$-isometrically.
Suppose that $K_n$ are $A$-tuples with $\Vert K_n(\cdot)
\Vert_2$ converging uniformly to $f$. Given $\epsilon > 0$, we have $\Vert K_n(w) \Vert_2 \leq f(w) + \epsilon$ for all $w \in \Omega$ and sufficiently large $n$. For such $w , n$, and for any complex Euclidean unit vector $z$ we have $$|z.(a(w)K_n(w))| \leq \Vert z.(a(\cdot)K_n(\cdot)) \Vert_{\partial A}
\leq \Vert af \Vert_{\partial A} + \epsilon \Vert a \Vert \; \; .$$ Thus $\Vert a(w)K_n(w) \Vert_2 \leq
\Vert af \Vert_{\partial A} + \epsilon \Vert a \Vert$. Letting $n \rightarrow \infty$ gives $\Vert af \Vert_{\Omega} \leq
\Vert af \Vert_{\partial A} + \epsilon \Vert a \Vert$. Since $\epsilon > 0$ was arbitrary, we get the result.
Using this we can sharpen an earlier result:
\[well\] Suppose that $f \in C(\Omega)^+$, that $f^{-2} \in {\mbox{${\mathcal G}$}}(\Omega)$, and that $Af \cong A$ $A$-isometrically (resp. almost $A$-isometrically). Then $f \in Q(\Omega)$ (resp. $f \in \bar{Q}^+(\Omega)$).
By the Lemma and Corollary \[new2\] (resp. \[new\]), $f_{|(\partial A)}
\in Q(\partial A)$ (resp. in $\bar{Q}^+(\partial A)$), and $Af^{-1} \cong
(Af^{-1})_{|(\partial A)} \cong A$. Now use corollary \[NEED\] (resp. \[notA\]).
Note that for the disk algebra $A(\mathbb{D})$ considered as functions on the closed disk, or more generally for any function algebra containing no nontrivial inner functions it is clear that $P\cap P^{-1}=Q$. Thus for certain good examples, we will have to look among algebras with many inner functions. It is also easy to see that $P(M_A) \cap P^{-1}(M_A)
= Q(M_A)$.
It will be significant to us that $Q,\bar{Q}^{+}$ and ${\mbox{${\mathcal M}$}}$ are (abelian) groups. We will write $Q'$ for the quotient group $\bar{Q}^{+}/Q$. Of course for some function algebras, it can happen that $Q=\bar{Q}^{+}$, for example for $H^{\infty }(\mathbb{D})$, where $\bar{Q}^{+} = L^{\infty }\left( \mathbb{T}\right)^{+}$ ([@Su], Th. 5.26). On the other hand, for $A(\mathbb{D})$, the disk algebra, we have that $Q$ is a proper subset of $\bar{Q}^{+}=C(\mathbb{T})^{+}$.
A good way of producing interesting functions in $\bar{Q}^{+}$, which works in any nonselfadjoint function algebra goes as follows: By a result of Hoffman and Wermer [@sto], $Re A$ is not uniformly closed. If $g\in (Re \; A)^{\bar{}}\setminus Re A$, set $f=e^{g}$. If $a_{n}\in
A,\; Re a_{n}\rightarrow g$ uniformly, then $e^{Re \; a_{n}}=|e^{a_{n}}|\rightarrow f$. Hence $f\in \bar{Q}^{+}$. If the set of invertible elements in $A$ is connected, or equivalently ([@Gam] p. 91), if the first Cech cohomology group $H^{1}(M_{A},\mathbb{Z})=0$ of the maximal ideal space $M_{A}$ is zero, then $h\in Q$ if and only if $h=e^{Re \; a}$ for some $a\in A$; thus we can definitely assert that $f\notin
Q $ in this case. Thus we have proved:
\[con\] Suppose that $A$ is a nonselfadjoint function algebra, such that $H^{1}(M_{A},\mathbb{Z})=0$, then $Q^{\prime }=\bar{Q}^{+}/Q\neq 0$. Hence there exist nontrivial function $A$-modules which are almost $A$-isometric to $A$.
We shall see later that this corollary also gives the existence of nontrivial rank 1 strong Morita equivalence bimodules for any such $A$.
We now turn to the class ${\mbox{${\mathcal M}$}}_A(\Omega)$ defined in the introduction. We have $$G_{A}(\Omega )\subset H_{A}(\Omega )\subset \log \mathcal{M}_{A}(\Omega ).$$ In Part C we shall see that all three of the above classes have the five properties of ‘harmonic classes’ with respect to $A$ described in the introduction. We shall not use the following, but state it for interests sake. Its proof follows from the definition of ${\mbox{${\mathcal M}$}}$ in the introduction, and is left to the reader.
If $f \in {\mbox{${\mathcal M}$}}_A(\Omega)$ and if $t > 0$ , then $tf \in {\mbox{${\mathcal M}$}}$. Moreover, ${\mbox{${\mathcal M}$}}$ is closed in the relative topology from $C(\Omega)^+$. Thus $\log {\mbox{${\mathcal M}$}}_A(\Omega)$ is uniformly closed.
Note that ${\mbox{${\mathcal M}$}}$ is not complete in the norm topology (since if $f \in {\mbox{${\mathcal M}$}}$ and $t > 0$ , then $tf \in {\mbox{${\mathcal M}$}}$, but $\lim_{t \rightarrow 0} tf = 0
\notin {\mbox{${\mathcal M}$}}$).
The authors do not know if $H_{A}(\Omega )=\log \mathcal{M}_{A}(\Omega )$ in general. In any case, either answer to the question seems very interesting. If $H(\Omega )=\log \mathcal{M}_{A}(\Omega )$ then we obtain from what we do later, amongst other things, a neat description of all topologically singly generated Morita equivalence $A-A$-bimodules; but if $H(\Omega )\neq \log \mathcal{M}_{A}(\Omega )$ in general, then $\log \mathcal{M}_{A}(\Omega )$ seems to be a genuinely new and interesting class of harmonic functions with respect to $A$. A uniform algebra is called ‘Dirichlet’ if $G_{A}(\Omega )=$ $C_{{\mbox{$\mathbb{R}$}}}(\Omega)$, is called ‘logmodular’ if $H_{A}(\Omega )= C_{{\mbox{$\mathbb{R}$}}}(\Omega)$; therefore we will call an algebra *logMorita* if $\log \mathcal{M}_{A}(\Omega )=
C_{{\mbox{$\mathbb{R}$}}}(\Omega)$. Every logmodular algebra is logMorita. The next result shows that logMorita algebras share many properties with logmodular algebras.
\[Omes\] If $A$ is a logMorita function algebra on $\Omega$, then
- $\Omega $ is the Shilov boundary of $A$,
- Every $\phi \in M_A$ has a unique representing measure,
- If $\Pi $ is a Gleason part for $A$, then either $\Pi $ is a singleton, or is an analytic disk, in the sense that there is a bijective continuous map $\Phi :\mathbb{D}\rightarrow \Pi $ , such that if $f\in A$ then $\hat{f}\circ \Phi $ is holomorphic on $\mathbb{D}$.
\(i) is proved as for Dirichlet and logmodular algebras, using Urysohns lemma and the fact that for an $A$-tuple $H \in A^{(n)}$, the function $\Vert H(\cdot )\Vert_{2}$ achieves its maximum modulus on the Shilov boundary. (The latter fact may be seen by considering $z.H(w)$ for $z \in {\mbox{$\mathbb{C}$}}^n_1$). Similarly, (ii) follows the classical line of proof: Suppose that $\mu ,\nu $ are representing measures for $\phi \in M_{A}$, and that $H,K$ are $A$-tuples with $1=H(w)\cdot K(w)$ for all $w\in \Omega $. By Fubini and Cauchy-Schwarz, we have $$1=\hat{H}(\phi )\cdot \hat{K}(\phi )=\int_{\Omega \times \Omega }H(w)\cdot
K(z)\;d(\mu \times \nu )\leq (\int \Vert H(w)\Vert _{2}d\mu )(\int \Vert
K(w)\Vert _{2}d\nu )\;.$$ The remainder of the proof follows 17.1 in [@sto]. Finally, (iii) follows from (ii) by 17.1 of [@sto].
Approximations in modulus and equivalence bimodules
===================================================
In [@BMP] the notion of *strong Morita equivalence* is defined for a pair of operator algebras $A$ and $B$. Its theory and consequences have been worked out there and in other papers of ours (see [@BMN; @BOMD] for example). A related notion, *strong subequivalence*, was recently defined in [@BOMD]. It was shown to have many of the properties of strong Morita equivalence. One of our objectives here is to show that it is not the same as strong Morita equivalence. It will not be necessary for us to state the general definitions of these notions here, we will simply say that they involve a pair of bimodules $X$ and $Y$, called *equivalence bimodules*. In fact we shall restrict our attention here to the special case where the operator algebras are function algebras, and in this case the two definitions can be simplified. Indeed, for a bimodule of the form $Af$, where $f$ is a strictly positive function on $\Omega$, and considering the canonical pairing $Af \times Af^{-1}
\rightarrow A$, it is easy to translate the definitions from [@BMP; @BOMD], using Lemma 2.8 of [@BMP], into the following precise form:
\[prec\] Suppose that $A$ is a function algebra on a compact space $\Omega$, and $f$ is a strictly positive continuous function on $\Omega$.
- We say that $Af$ is a *strong Morita equivalence bimodule (with inverse bimodule $Af^{-1}$)*, if whenever $\epsilon >0$ is given, then we can write $1=\sum_{i=1}^{n}x_{i}y_{i}$ as functions on $\Omega $, with $x_{i}\in
Af,y_{i}\in Af^{-1}$, and $\sum_{i=1}^{n}|x_{i}(\omega )|^{2}\leq 1+\epsilon
$, and $\sum_{i=1}^{n}|y_{i}(\omega )|^{2}\leq 1+\epsilon $, for all $\omega
\in \Omega $. We will say $Af$ is rank 1, if $n=1$ in the above.
- We say that $Af$ is a *strong subequivalence bimodule* if whenever $\epsilon >0$ is given, there are $x_{i}\in Af,y_{i}\in Af^{-1}$, such that $1-\epsilon \leq \sum_{i=1}^{n}|x_{i}(\omega )|^{2}\leq 1+\epsilon,
$ and $1-\epsilon \leq \sum_{i=1}^{n}|y_{i}(\omega )|^{2}\leq 1+\epsilon $, for all $\omega \in \Omega $.
- We say that $Af$ is a *unitary subequivalence bimodule* if (ii) holds, but with $\epsilon =0$ and $n=1$.
- We shall say that $Af$ is a *Shilov subequivalence bimodule* if it is a strong subequivalence bimodule and $\Omega$ is the Shilov boundary of $A$.
All these definitions are in [@BMP; @BOMD] except (iii). Also, strictly speaking, in (i) we should say that $(A,A,Af,Af^{-1},\cdot,\cdot)$ *is a strong Morita context* (see [@BMP] Definition 3.1). Here ‘$\cdot$’ refers to multiplication of scalar functions on $\Omega$. However, to avoid this somewhat cumbersome notation, we will use the looser convention of (i). Also, concerning (iv), we used the word ‘minimal’ instead of ‘Shilov’ in [@BOMD] Definition 5.7.
It is clear in the definitions above, that $(iii)\Rightarrow (ii)$ and that $(i)\Rightarrow (ii)$.
\[ma\] Suppose that $f\in C(\Omega )^{+}$. Then:
- $Af$ is a unitary subequivalence bimodule if and only if $f\in
P\cap P^{-1}$. That is, iff $f\in P$ and $f^{-1}\in P$.
- $Af$ is a strong subequivalence bimodule if and only if $f^{2}\in \mathcal{G}\cap \mathcal{G}^{-1}$
- $f\in \bar{Q}^{+}$ if and only if $Af$ is a rank 1 strong Morita equivalence bimodule (with inverse bimodule $Af^{-1}$).
- $f \in {\mbox{${\mathcal M}$}}(\Omega)$ if and only if $Af$ is a strong Morita equivalence bimodule (with inverse bimodule $Af^{-1}$).
<!-- -->
- Clearly $Af$ is a unitary subequivalence bimodule if and only if $Af$ and $Af^{-1}$ both contain inner functions (functions of constant modulus 1 on $\Omega $), which is clearly equivalent to $f\in P\cap
P^{-1}$.
- Suppose that $\epsilon >0$ is given, and that $x_{i},y_{i}$ are as in (ii). Then $\Vert 1-\sum_{i}|x_{i}|^{2}\Vert _{\infty
}\rightarrow 0$ as one allows $\epsilon \rightarrow 0$. Hence $f^{-2}\in
\mathcal{G}$. Similarly $f^{2}\in \mathcal{G}$. For the converse, notice that the argument given is reversible.
- If $|g_{m}|\rightarrow f$ uniformly as $m\rightarrow
\infty $, where $g_{m},g_{m}^{-1}\in A$, then we may write $1=(g_{m}^{-1}f)(g_{m}f^{-1})$, and clearly $\Vert g_{m}^{-1}f\Vert _{\infty
}\rightarrow 1$ and $\Vert g_{m}f^{-1}\Vert _{\infty }\rightarrow 1$. This obviously implies the condition in (i) of Definition \[prec\] with $n=1$. Also, this argument is reversible.
- This is obvious.
[**Remark.**]{} To those readers who are familiar with notions in [@BOMD], we remark that the proof of (b) shows the following. Consider $Z = C(\Omega)$ as a module over itself. It is clearly generated as a $C(\Omega)$-module by the $A$-submodule $Af$. The proof of (b) shows, in the language of [@BOMD], that $Z$ is the [*C$^*-$dilation*]{} of $Af$ if and only if $f^{-2} \in {\mbox{${\mathcal G}$}}$.
\[oop\] Suppose that $A$ is a function algebra which is antisymmetric on $\Omega$, or for which the number of pieces in the antisymmetric decomposition[^2] of $A$ is finite. Suppose that $f$ is a function such that $Af$ satisfies Definition \[prec\] (i) with $\epsilon = 0$. Then $\log f \in
H(\Omega)$, and indeed $f \in Q$.
First suppose that $A$ is antisymmetric on $\Omega$. By hypothesis, we have $f = \Vert K(\cdot) \Vert_2$ and $f^{-1} = \Vert H(\cdot) \Vert_2$, for $A$-tuples $H,K$ with $1 = H.K = (Hf).(Kf^{-1})$. By the converse to Cauchy-Schwarz, $Hf = K^*f^{-1}$, where ‘$*$’ is the complex conjugate. Thus $h_ik_i = |k_i|^2 f^{-2} \geq 0$. By antisymmetry, $h_i k_i = c_i$ a nonnegative constant. Therefore, if $c_1 \neq 0$ then $h_1,k_1 \in A^{-1}$. We have $c_1 = h_1 k_1 = |k_1|^2 f^{-2}$, so that $f = \frac{1}{\sqrt{c_1}} |k_1| \in Q$.
If $\Omega$ is a disjoint union of a finite number of antisymmetric pieces $\Omega_i$, then each $\Omega_i$ is open and compact, and by the first part we have $f_{|\Omega_i} = |b_i|$, for an invertible $b_i \in A_{|\Omega_i}$. Put $b(x) = b_i(x)$ if $x \in
\Omega_i$, then $b \in C(\Omega)$. It follows from [@sto] Theorem 12.1 say, that $b \in A$ and $b^{-1} \in A$. Thus $f = |b| \in Q$.
Comparing \[ma\] (c) and \[wasla\] shows that for any $f\in \bar{Q}^{+}\setminus Q$, provided say by \[con\], we have that $Af$ is a nontrivial strong Morita equivalence bimodule.
From the next theorem we will be able to give examples of strong subequivalence bimodules which are not strong Morita equivalence bimodules.
Let $A$ be a uniform algebra on $\Omega$. Suppose that $w_{1},w_{2}$ are distinct points of $\Omega$, and that $G$ is an invertible element of $A$ such that $G\left(
w_{1}\right) =1=-G\left( w_{2}\right) $. Put $A_{0}=\left\{ a\in A:a\left(
w_{1}\right) =a\left( w_{2}\right) \right\}$, and set $f=\left| G\right| $. Then $f \in {\mbox{${\mathcal M}$}}_{A_0}$ if and only if $w_{1},w_{2}$ are in different Gleason parts of $A$, and if and only if $f \in \bar{Q}^+_{A_0}$.
Assume that $f \in {\mbox{${\mathcal M}$}}_{A_0}$. We will show that $w_{1},w_{2}$ are in different Gleason parts of $A$. In the definition in the introduction of ${\mbox{${\mathcal M}$}}_{A_0}$, take $\varepsilon=\frac{1}{m}$ for a natural number $m$, and choose the corresponding $H_{m}=[h_{1}^{m}\cdots h_{n_{m}}^{m}],K_{m}=[k_{1}^{m}\cdots
k_{n_{m}}^{m}]\in A^{n_{m}}$ with $\left\langle H|K^{\ast}\right\rangle
=\sum_{i=1}^{n_{m}}h_{i}^{m}k_{i}^{m}=1$. By that definition we may assume that $\Vert K_{m}(w)\Vert_{2}\leq c_{m}f(w)$ , and $\Vert H_{m}(w)\Vert
_{2}f(w)\leq c_{m}$, for all $w\in\Omega$. Here $c_{m}$ is a sequence of real numbers decreasing to 1. Multiplying the functions $h_{i}^{m},k_{i}^{m} $ by constants with absolute value one we may also assume that $$h_{i}^{m}\left( w_{2}\right) \geq 0 \; \;
\text{, for all }i=1,...,n_{m}. \label{pos}$$ Write $\Theta_{m}(w)=\frac{K_{m}(w)}{G\left( w\right) }$, and $\Pi
_{m}(w)=H_{m}(w)G(w)$. So $\Vert\Theta_{m}(w)\Vert_{2}\leq c_{m}$ and $\Vert\Pi_{m}(w)\Vert_{2}\leq c_{m}$. By the Cauchy-Schwarz Inequality we get $$1=|\sum_{i=1}^{n_{m}}h_{i}^{m}k_{i}^{m}|\leq\Vert\Theta_{m}(w)\Vert_{2}\Vert\Pi_{m}(w)\Vert_{2}\leq c_{m}\Vert\Theta_{m}(w)\Vert_{2}$$ Hence $$\frac{1}{c_{m}}\leq\Vert\Theta_{m}(w)\Vert_{2}\leq c_{m}\text{, for all }w\in\Omega.$$ Obviously, the same formula holds with $\Theta_{m}$ replaced by $\Pi_{m}$. Expanding out the following square as an inner product and using (\[pos\]) we get $$\Vert\Theta_{m}\left( w_{2}\right) -\Pi_{m}\left( w_{2}\right) \Vert
_{2}^{2}=\Vert\Theta_{m}\left( w_{2}\right) -\Pi_{m}\left( w_{2}\right)
^{\ast}\Vert_{2}^{2}\leq2c_{m}^{2}-2\rightarrow0.$$ Hence, by the Pythagorean Identity $$\Vert\Theta_{m}(w_{2})+\Pi_{m}(w_{2})\Vert_{2}\rightarrow2\quad\text{ as }m\rightarrow\infty.$$ Put $W_{n}=\frac{1}{2}\left( \Theta_{m}+\Pi_{m}\right) $. Since $h_{i}^{m},k_{i}^{m}\in A_{0}$ and $G\left( w_{1}\right) =1=-G\left(
w_{2}\right) $ we get $$\Vert W_{m}(w_{2})- W_{m}(w_{1})\Vert
_{2}=\Vert2\left( W_{m}\right) (w_{2})\Vert_{2}\rightarrow2\quad\text{ as }m\rightarrow\infty.$$
On the other hand, for any Euclidean vector $z$ of norm 1 we have $$|(z.W_m)(w_{2}) - (z.W_m)(w_1)| \leq \left\| w_{2}-w_{1}\right\|
\Vert z.W_m \Vert \leq \left\| w_{2}-w_{1}\right\|
\sup \{ \Vert W_m(w) \Vert_2 : w \in \Omega \}\; ,$$ where $\left\| w_{2}-w_{1}\right\|$ is the norm of $w_{2}-w_{1}$ considered as a functional on $A$. From the above it follows that $|(z.W_m)(w_{2}) - (z.W_m)(w_1)|
\leq \left\| w_{2}-w_{1}\right\| c_m$. Thus $$\left\| W_{m}(w_{2})- W_{m}(w_{1})\right\|
_{2}\leq\left\| w_{2}-w_{1}\right\| c_m \; .$$ Since $c_{m}\rightarrow1$, we get $\left\|
w_{2}-w_{1}\right\| =2$. Thus $w_{1},w_{2}$ lie in different Gleason parts of $A $.
To prove the other implication assume now that $w_{1},w_{2}$ are in different Gleason parts of $A$. Choose a sequence of functions $r_{n}$ analytic in the disc $\bar{{\mbox{$\mathbb{D}$}}}=\left\{ w\in \mathbb{C}:\left| w\right|
\leq 1\right\} $ with $r_{n}(1)=1,r_{n}(-1)=-1,$ and $1-1/n<|r_{n}(w)|<1+1/n$ for all $w\in \mathbb{D}$. Such functions can be found by taking a conformal equivalence of $\mathbb{D}$ with a ‘smile shaped region’ inside the annulus $1-1/2n<|w|<1+1/2n$, with the two tips of the smile at $-1$ and $1$. Since $w_{1},w_{2}$ are in different Gleason parts of $A$, for any $n\in \mathbb{N}$ large enough, there is an $a_{n}$ in $A$ such that $\left\| a_{n}\right\| \leq 1-\frac{1}{2n}$, $\left| (r_{n}\circ a_{n})\left( w_{1}\right) -\left( 1-\frac{1}{2n}\right)
\right| <\frac{1}{2n}$, $\left| (r_{n}\circ a_{n})\left( w_{2}\right) +\left(
1-\frac{1}{2n}\right) \right| <\frac{1}{2n}$. Also the norm, on $\left\{
a\in A:a\left( w_{1}\right) = 0 \right\} ,$ of the “evaluation at $w_{2}$” functional is equal to one. Thus there is a $\widetilde{a}\in A$ such that $\left\|
\widetilde{a}\right\| \leq \frac{1}{n},$ $\widetilde{a}\left( w_{1}\right)
=0$ and $\widetilde{a}\left( w_{2}\right) =
(r_{n}\circ a_{n})\left( w_{1}\right) +
(r_{n}\circ a_{n})\left(w_{2}\right)$. Put $b_{n}=r_{n}\circ a_{n}-\widetilde{a}\in A$. Notice that $b_{n}\left( w_{1}\right) G\left(
w_{1}\right) =b_{n}\left( w_{2}\right) G\left( w_{2}\right) $ so $b_{n}G\in
A_{0}$. Moreover the spectrum of $b_{n}$ is contained inside the annulus $1-2/n<|w|<1+2/n$. Hence $b_{n}$ is an invertible element of $A$, and consequently $b_{n}G$ is an invertible elements of $A_{0}$. Thus $f=\lim
\left| b_{n}G\right| \in \bar{Q}_{A_{0}}$. By Proposition \[ma\] this proves that $A_{0}f$ is a strong Morita equivalence bimodule.
[**Example.**]{} Let $A$ be a function algebra. First take $\Omega = M_A$ and suppose that $A_0, G, f$ and $w_1,w_2$ are as in the last theorem. Now, for any $\Omega$ on which $A$ sits as a function algebra (such as $\partial A$), suppose further that there exists a function $h \in A$, such that $|h| = 1$ on $\Omega$ and $h(w_1) = - h(w_2)$. We are not assuming $w_1,w_2 \in \Omega$ here, they are points in $M_A$. Then $f = |G| = |hG| \in P(\Omega)$, and $f^{-1} = |G^{-1}| = |\frac{h}{G}| \in P(\Omega)$. Thus the submodule $A_0f$ of $C(\Omega)$ is a strong subequivalence $A_0-A_0$-bimodule, indeed a unitary subequivalence bimodule. However if $w_1, w_2$ are in the same Gleason part of $M_A$, then $A_0f$ is not a strong Morita equivalence $A_0-A_0$-bimodule.
For a very concrete example, let $A({\mbox{$\mathbb{D}$}})$ for the disk algebra, and set $\Omega = {\mbox{$\mathbb{T}$}}$, the unit circle. Choose two points $\alpha, \beta \in {\mbox{$\mathbb{D}$}}$. Then there exists an inner function $h$ in $A({\mbox{$\mathbb{D}$}})$ such that $h(\alpha) = -h(\beta)$. For example, if $\alpha = -\beta = \frac{1}{2}$, let $h(z) = z$. Choose $G \in A({\mbox{$\mathbb{D}$}})^{-1}$ such that $G(\alpha) = - G(\beta)$. Then the submodule $A_0|G|$ of $C({\mbox{$\mathbb{T}$}})$ is a strong subequivalence bimodule, which is not a strong Morita equivalence $A_0-A_0$-bimodule.
Question: Suppose $Af$ is a subequivalence bimodule where $\Omega$ is the maximal ideal space of $A$. Then is $Af$ a strong Morita equivalence bimodule? In other words, is ${\mbox{${\mathcal G}$}}(M_A) \cap {\mbox{${\mathcal G}$}}(M_A)^{-1} = {\mbox{${\mathcal M}$}}^2$? Recall that $P(M_A) \cap
P(M_A)^{-1} = Q(M_A)$. By the way, it follows easily from this latter fact, and the example above, that $P \cap P^{-1}$ is not a harmonic class.
In the spirit of the calculation in the proof above, we end this section by giving some alternative descriptions of $\mathcal{M}_{A}(\Omega )$.
For a uniform algebra $A$ on a compact set $\Omega $, the class $\mathcal{M}_{A}(\Omega )$ coincides with the set of those functions $f$ $\in C(\Omega )^{+}$ for which there exist a sequence of positive integers $\{n_{m}\}$ and a sequence of $A$-tuples $H_{m},K_{m}\in A^{n_{m}}$, such that the following three limits
- $\Vert K_{m}(w)\Vert _{2}\rightarrow f(w)$ ,
- $\Vert H_{m}(w)\Vert _{2}\rightarrow f(w)^{-1}$ , and
- $\Vert K_{m}(w)-H_{m}(w)^{\ast }f(w)^{2}\Vert _{2}\rightarrow 0$
are valid uniformly over $w\in \Omega $. Here the ‘$\ast $’ represents the complex conjugate of the vector.
Suppose that $f \in \mathcal{M}_{A}(\Omega )$. As in the proof of the previous theorem, take $\epsilon =\frac{1}{m}$ for a natural number $m$, and choose $K_{m}, H_{m}, c_m$ as in that proof. Write $\Theta _{m}(w)=\frac{K_{m}(w)}{f(w)}$, and $\Pi _{m}(w)=H_{m}(w)f(w)$. So $\Vert \Theta _{m}(w)\Vert _{2}\leq c_{m}$ and $\Vert \Pi _{m}(w)\Vert _{2}\leq c_{m}$. Again by Cauchy-Schwarz we get $$\frac{1}{c_{m}}\leq \Vert \Theta _{m}(w)\Vert _{2}\leq c_{m}$$ for all $w\in \Omega $; and the same formula holds for $\Pi _{m}$. By expanding out the following square as an inner product, we see again that $$\Vert \Theta _{m}(w)-\Pi _{m}(w)^{\ast }\Vert ^{2}\leq
2c_{m}^{2}-2\rightarrow 0$$ uniformly as $m\rightarrow \infty $.
This gives the one direction of the proposition. However, the argument is reversible until at the end we obtain that $H_{m}(w)\cdot
K_{m}(w)\rightarrow 1$ uniformly in $w\in \Omega $. Thus $b_{m}=H_{m}(w)\cdot K_{m}(w)$ is invertible in $A$, and $(b_{m}^{-1}H_{m}(w))\cdot K_{m}(w)=1$.
Notice that (i) and (ii) in the proposition say exactly that $Af$ is a strong subequivalence bimodule, or that $f^2 \in {\mbox{${\mathcal G}$}}\cap {\mbox{${\mathcal G}$}}^{-1}$. Secondly, notice in fact that (iii) together with either of (i) or (ii), implies the other condition. However it is convenient to state it as such.
The last proposition may be loosely phrased as saying that a function $f \in C(\Omega)^+$ is in ${\mbox{${\mathcal M}$}}$ if and only if there exists a sequence of $A$-tuples $K_m$, such that (i) holds uniformly, and $\frac{K_m(\cdot)^*}{\Vert K_m(\cdot) \Vert^2_2}$) is ‘eventually, uniformly, an $A$-tuple’.
In view of (i) and (ii) we may replace (iii) by the more appealing looking condition:
- $\Vert K'_m(w) - H'_m(w)^*
\Vert_2 \rightarrow 0$
uniformly on $\Omega$, where $v' = \frac{v}{\Vert v \Vert}$ is the unit vector in the same direction as $v$, for any vector $v$ in complex Euclidean space.
The Picard group.
=================
Some material in this section, and indeed in the rest of the paper, requires some technical knowledge of several papers. The reader who desires further background is directed to [@Bnat] for a leisurely introduction to our work, and to [@BMP; @BMN; @Bhmo] for more specific details. The general reader is advised to simply read the main results below.
We begin by discussing strong Morita equivalence of function algebras. As we said earlier, we shall not give the general definition, but simply say that it involves a pair of bimodules $X$ and $Y$, called *equivalence bimodules*. A well know property of (strong) Morita equivalence is that if $A$ and $B$ are unital, commutative and (strongly) Morita equivalent then $A$ is (isometrically) isomorphic to $B$. So we shall assume that $A=B$. However this is not the end of the story, for the question remains as to which $A-A$-bimodules implement such a ‘self-equivalence’ of $A$. The collection of such bimodules, with two such bimodules identified if they are (completely) isometrically $A-A$-isomorphic, is a group, with multiplication being the Haagerup tensor product $\otimes _{hA}$ (see [@BMP]). We call this the *strong Picard group*, and write $Pic_{s}(A)$. For background on the Picard group, see any algebra text covering Morita equivalence (for example [@F] Chapter 12), or [@BGR; @DG] for a discussion of the C$^{\ast }-$algebraic version. For example we will show below that for $A=A(\mathbb{D})$, the disk algebra, this Picard group is the direct product of the Mobius group and the abelian group $C(\mathbb{T})/Re\;A$.
In fact for the most part we will only consider a certain subgroup of $Pic_{s}(A)$, namely the singly generated bimodules. We will show that for a function algebra $A$ on a compact (Hausdorff) space $\Omega $, every such bimodule is essentially of the form $Af$, where $f$ is a strictly positive continuous function on $\Omega $. Indeed $f$ may be chosen in ${\mbox{${\mathcal M}$}}(\Omega)$, and by \[gen\] it follows that $f$ is unique up to the coset $Q$.
Indeed, even in the non-singly generated case, if the second Cech cohomology group of $\Omega $ vanishes, we shall see that every strong Morita equivalence $A-A$-bimodule is a finitely generated submodule of $C(\Omega )$. Of course, the non-singly generated case is probably much more interesting, but will no doubt require a much deeper analysis. Our intention here is mainly to point out very clearly the features of the singly generated case.
As we said earlier, the multiplication on the Picard group of $A$, is given by the module Haagerup tensor product $\otimes_{hA}$. See [@BMP] for details about this tensor product. However in the case that the modules are singly generated (as just discussed), then this tensor product becomes rather trivial:
\[prod\] Suppose that $f \in {\mbox{${\mathcal M}$}}(\Omega)$, and $g \in C(\Omega)^+$. Then $Af \otimes_{hA} Ag
\cong A(f g)$ (completely) $A-A$-isometrically.
We omit the proof of this, which is a special case of the more general Lemma \[sg\], which we prove later.
A standard type of strong Morita equivalence bimodule for $A$ comes from taking an isometric automorphism $\theta : A \rightarrow A$, and defining the module $A_\theta = A$, with the usual left module action, and with right module action $b \cdot a = b \theta(a)$. It is easy to check this is a strong Morita equivalence bimodule for $A$. The collection of (equivalence classes of) this type of equivalence bimodule is a subgroup of $Pic_s(A)$, which is isomorphic to the group $Aut(A)$ of isometric automorphisms of $A$. This, in the case that $A$ is a function algebra, corresponds to a group of homeomorphisms of the maximal ideal space of $A$, which restrict to homeomorphisms of the Shilov boundary of $A$.
More generally, if $X$ is a strong Morita equivalence $A-A$-bimodule, and if $\theta \in Aut(A)$, then $X_\theta$ is also a a strong Morita equivalence $A-A$-bimodule. Here $X_\theta$ is $X$ but with right action changed to $x \cdot a = x \theta(a)$. One way to see this is to note that $X_\theta \cong X \otimes_{hA} A_\theta$.
An $A-A$-bimodule $X$ will be called ‘symmetric’ if $a x = x a$ for all $a \in A , x \in X$.
\[semidi\] For a function algebra $A$, $Pic_s(A)$ is a semidirect product of $Aut(A)$ and the subgroup of $Pic_s(A)$ consisting of symmetric equivalence bimodules. Thus, every $V \in Pic_s(A)$ equals $X_\theta$, for a symmetric $X \in Pic_s(A)$, and for some $\theta \in Aut(A)$.
This follows essentially as in pure algebra ([@F] Chapter 12.18). Suppose that $X$ is any strong Morita equivalence $A-A$-bimodule. Then by the basic Morita theory, any right $A$-module map $T : X \rightarrow X$ is simply left multiplication by a fixed element of $A$. Indeed, via this identification, we have $A \cong CB_A(X)$ isometrically and as algebras ([@BMP] 4.1 and 4.2). For fixed $a \in A$, the operator $x \mapsto xa$ on $X$, is a right $A$-module map, with completely bounded norm $= \Vert a \Vert$. Therefore by the above fact, there is a unique $a' \in A$ such that $a'x = x a$ for all $x \in X$. The map $a \mapsto a'$ is then seen to be an isometric unital automorphism $\theta$ of $A$. In this way we have defined a surjective group homomorphism $Pic_s(A) \rightarrow Aut(A)$. This homomorphism has a 1-sided inverse $Aut(A) \rightarrow Pic_s(A)$, namely $\theta \rightarrow A_\theta$. In this way, we see the ‘semidirect product’ statement. Note that $X =
(X_{\theta^{-1}})_\theta$, and $X_{\theta^{-1}}$ is symmetric.
More generally, if $X$ is a strong subequivalence $A-A$-bimodule in the sense of [@BOMD], then a similar argument works. Since we will not use this result here we will not give full details, but the idea goes as follows: Suppose that $X$ corresponds (using the notation of [@BOMD] §5) to a subcontext $(A,A,X,Y)$ of a C$^*$-Morita context $({\mbox{${\mathcal C}$}},{\mbox{${\mathcal D}$}},W,Z)$, where ${\mbox{${\mathcal C}$}}= C(\Omega)$. It follows that ${\mbox{${\mathcal D}$}}\cong {\mbox{${\mathcal C}$}}$ isometrically. Now use the argument of the proof above, combined with Theorem 5.6 in [@BOMD].
Thus we may assume henceforth that $X$ is symmetric. Then $X$ dilates to a strong Morita equivalence ${\mbox{${\mathcal C}$}}-{\mbox{${\mathcal C}$}}$-bimodule $W = {\mbox{${\mathcal C}$}}\otimes_{hA} X$. From [@Bhmo] Theorem 6.8, we know that $W$ contains $X$ completely isometrically. It is helpful to consider the inclusion $$\left[ \begin{array}{ccl}
A & X \\
Y & A
\end{array} \right] \subset
\left[ \begin{array}{ccl}
{\mbox{${\mathcal C}$}}& W \\
Z & {\mbox{${\mathcal C}$}}\end{array} \right]$$ of linking algebras. Note that since $W \cong
{\mbox{${\mathcal C}$}}\otimes_A X$ we have $wa = aw$ for all $w \in W, a \in A$. Similarly for $Z \cong Y \otimes_A {\mbox{${\mathcal C}$}}$ we have $za = az$. Since $Z = W^*$, we have $w a^* = (a w^*)^* = (w^* a)^* = a^* w$. Therefore $x w = w x$ for all $w \in W, x
\in {\mbox{${\mathcal C}$}}$. Thus $W$ is a symmetric element of $Pic(C(\Omega))$, and consequently (see the appendix of [@Rae], or [@Ri2; @BGR; @DG]) $W$ may be characterized as the space of sections of a complex line bundle over $\Omega$.
Suppose that $A$ is a function algebra on $\Omega$, and that $X$ is a symmetric subequivalence $A-A-$bimodule (or strong Morita equivalence bimodule). If every complex line bundle over $\Omega$ is trivial, or equivalently, if the Cech cohomology group $H^2(\Omega,{\mbox{$\mathbb{Z}$}}) = 0$, then $W \cong C(\Omega)$ completely $A-A$-isometrically. Thus $X$ is $A-A$-isometric to a finitely generated $A-A$-submodule of $C(\Omega)$.
The hypothesis $H^2(\Omega,{\mbox{$\mathbb{Z}$}}) = 0$ applies, for example, if $\Omega = \bar{{\mbox{$\mathbb{D}$}}}$ or ${\mbox{$\mathbb{T}$}}$, or more generally, any multiply connected region in the plane.
\[nice\] Suppose that $A$ is a function algebra on $\Omega$. Every singly generated strong Morita equivalence (resp. subequivalence) $A-A$-bimodule is completely $A-A$-isometrically isomorphic to $(Af)_\theta$, for some $\theta \in Aut(A),
f \in C(\Omega)^+$. The converse of this statement is also true, if $A$ is logmodular or if it is a logMorita algebra (resp. convexly approximating in modulus).
We use the notation and facts established in the second paragraph below the proof of Proposition \[semidi\]. If $X$ is singly generated over $A$, then so is $W$ over ${\mbox{${\mathcal C}$}}$, and again it follows that $W \cong {\mbox{${\mathcal C}$}}$ completely $A-A$-isometrically. Here is one way to see this (which adapts to the noncommutative case): if $x$ is the generator, then $x^*x$ is a strictly positive element in ${\mbox{${\mathcal C}$}}$ (see for example [@BMP] theorem 7.13). Thus $x^*x$ is invertible, so that $f = |x|$ is invertible in ${\mbox{${\mathcal C}$}}$. Thus $W = {\mbox{${\mathcal C}$}}x \cong {\mbox{${\mathcal C}$}}f = {\mbox{${\mathcal C}$}}$, and $X = Ax \cong Af$.
The ‘converse’ assertion follows from the earlier correspondences between the classes $\bar{Q}^+, {\mbox{${\mathcal M}$}}$ and ${\mbox{${\mathcal G}$}}$, and strong Morita equivalence and subequivalence bimodules.
It can be shown that for $X$ a singly generated strong Morita equivalence $A-A$-bimodule, the Morita ‘inverse bimodule’ $\tilde{X} \cong CB_A(X,A)$ may be identified with $\{ c \in {\mbox{${\mathcal C}$}}: c f \in A \} = A f^{-1}$. That is, the strong Morita context (in the sense of [@BMP]) associated with $X$ may be identified with $(A,A,Af,Af^{-1})$. The latter is a subcontext (in the sense of [@BOMD]) of $({\mbox{${\mathcal C}$}},{\mbox{${\mathcal C}$}},{\mbox{${\mathcal C}$}},{\mbox{${\mathcal C}$}})$. Also, $f \in {\mbox{${\mathcal M}$}}_A$. These facts may be probably be shown directly from what we have done in the proof, and [@Bhmo] Theorem 5.10 say. However another proof of these facts is given in Corollary \[Motat\] in the next section.
We turn now to ‘rank one’ modules. We illustrate this concept first at the level of pure algebra. Suppose that $A$ is a commutative unital algebra, and that $X$ is a (purely algebraic) Morita equivalence $A-A$-bimodule, with ‘inverse bimodule’ $Y$ (see [@F]). As above, we say that $X$ is symmetric if $ax = xa$ for all $x \in X, a \in A$. We shall say that $X$ is [*algebraically rank 1*]{}, if $1 = (x',y') = [y'',x'']$ for some $x', x'' \in X, y',y'' \in Y$; here $(\cdot)$ and $[\cdot]$ are the Morita pairings [@F]. It is easy to show that this implies that $(x,y) = [y,x]$ for all $x \in X, y \in Y$. Simple algebra shows that a symmetric bimodule $X$ is an algebraically rank 1 Morita equivalence $A-A$-bimodule if and only if $X \cong A$ as $A$-modules.
Now suppose that $X$ is a symmetric strong Morita equivalence $A-A$-bimodule. We say that $X$ is [*rank one*]{}, if $X$ is algebraically rank 1, and for any $\epsilon > 0$, the $x', y'$ above may be chosen with norms within $\epsilon$ of 1. It follows, that if we define $T(x) = (x,y')$ on $X$, and $S(a) = ax'$ on $A$, then $S = T^{-1}$. Also the norms $\Vert T \Vert_{cb} = \Vert T \Vert$, and $\Vert S \Vert_{cb}
= \Vert S \Vert$, are close to 1. Thus it follows that $X \cong A$ almost completely $A$-isometrically, and we see that $X$ is a $MIN$ space (that is, its operator space structure is that of a subspace of a commutative C$^*-$algebra).
We can therefore add to our earlier Corollary \[Motat21\], the following:
\[Motat2\] Let $X$ be a Banach $A$-module. The following are equivalent:
- $X \cong A$ almost $A$-isometrically,
- $X$, with the symmetric bimodule action, is a rank one strong Morita equivalence $A-A$-bimodule,
- There exists $f \in \bar{Q}^+$ such that $X \cong Af$ $A$-isometrically.
We just saw that $(ii) \Rightarrow (i)$. As we saw in \[Motat21\], condition (i) is equivalent to (iii). Proposition \[ma\] (c) shows that (iii) implies (ii).
Thus the new definition of rank one is equivalent to our earlier definition (\[prec\] (i)) of $Af$ being a rank one strong Morita equivalence bimodule.
\[gen\]
- If $f_1, f_2 \in {\mbox{${\mathcal M}$}}$ then $Af_1 \cong Af_2$ $A$-isometrically if and only if $f_1 = h f_2$ for some $h \in Q$.
- The map $[f] \mapsto Af$ gives an injective group homomorphism ${\mbox{${\mathcal M}$}}' = {\mbox{${\mathcal M}$}}/Q \rightarrow Pic_s(A)$. The range of this homomorphism consists of all the topologically singly generated symmetric elements of $Pic_s(A)$. Thus $Pic_s(A)$ contains as a subgroup, the direct product of ${\mbox{${\mathcal M}$}}'$ and $Aut(A)$.
- If one restricts the group homomorphism in (ii) to $Q' = \bar{Q}^+/Q$, then its range consists of the rank 1 strong Morita equivalence $A-A$-bimodules.
- If $f_1, f_2 \in {\mbox{${\mathcal M}$}}$ then $Af_1 \cong Af_2$ almost $A$-isometrically if and only if $f_1 = h f_2$, for some $h \in \bar{Q}^+$.
By the ‘harmonicity’ associated with the ${\mbox{${\mathcal M}$}}$-class, we can assume that $f_1, f_2 \in C(\partial A)^+$. Then (i) follows immediately from Corollary \[wasla\]. However here is another argument, which adapts immediately to give (iv) too. Suppose that $Af_1 \cong Af_2$ $A$-isometrically, where $f_1, f_2 \in {\mbox{${\mathcal M}$}}$. Then by \[prod\] we have $A \cong Af_1^{-1} \otimes_{hA} Af_1 \cong
Af_1^{-1} \otimes_{hA} Af_2 \cong A(f_2f_1^{-1})$. This implies that $f_2f_1^{-1} \in Q$ by the last theorem. Thus $[f_1] = [f_2]$. This also gives (ii), in view of \[prod\]. A similar argument to (i) also proves (iv).
For many common function algebras, every strong Morita equivalence $A-A$-bimodule $X$ is singly generated. We shall illustrate this for $A = A({\mbox{$\mathbb{D}$}})$. Such an $X$ is algebraically finitely generated and projective ([@F] 12.7). Then by Theorem 3.6 in [@Bhmo], $X$ is completely boundedly $A$-isomorphic to a closed $A$-complemented submodule of $A^{(n)}$ . The associated projection $P : R_n(A) \rightarrow R_n(A)$ may be thought of as an analytic projection valued function $f_P : {\mbox{$\mathbb{D}$}}\rightarrow
M_n$, and since the disk is contractible, $f_P$ is homotopic to the constant function $f_P(0)$. Thus (see [@Bla] 4.3.3) $P$ is similar to a constant projection in $M_n$. We originally heard this last argument from P. Muhly. Hence $X \cong
A^{(m)}$ algebraically, and if $m > 1$ then $W =
{\mbox{${\mathcal C}$}}\otimes_A X \cong {\mbox{${\mathcal C}$}}^{(m)}$ algebraically, which is impossible. So $X$ is singly generated. Putting this together with Theorem \[gen\] (ii), and the fact that $A({\mbox{$\mathbb{D}$}})$ is a Dirichlet algebra (so that $\bar{Q}^+ = {\mbox{${\mathcal M}$}}= C({\mbox{$\mathbb{T}$}})^+$), we have:
\[name\] The Picard group of $A = A({\mbox{$\mathbb{D}$}})$, is the direct product of the Mobius group and the abelian group $C_{{\mathbb R}}({\mbox{$\mathbb{T}$}})/ Re \; A $.
[**Remark.**]{} We have proved elsewhere that the strong Picard group of $A$ is isomorphic to the group of category equivalences of $_AOMOD$ with itself, where $_AOMOD$ is the category of left operator modules over $A$.
Part C.
Rigged modules over function algebras.
======================================
A (left) $A$-Hilbertian module is a left operator module $X$ over $A$, such that there exists a net of natural numbers $n_\alpha$ , and completely contractive $A$-module maps $\phi_\alpha
: X \rightarrow R_{n_\alpha}(A)$ and $\psi_\alpha :
R_{n_\alpha}(A) \rightarrow X$, such that $\psi_\alpha \circ \phi_\alpha
(x) \rightarrow x$, for all $x \in X$. Here $R_n(A)$ is $A^{(n)}$, viewed as an operator space by considering it as the first row of $M_n(A)$. The name ‘Hilbertian’ is due to V. Paulsen. An $A$-rigged module is an $A$-Hilbertian module with the additional property that for all $\beta$ we have $\phi_\beta \psi_m \circ \phi_m
\rightarrow \phi_\beta$ in cb-norm. If in addition, $X$ is singly generated, it follows that we can take the net in the definition of $A$-Hilbertian, to be a sequence $n_m, \phi_m, \psi_m$, $m \in {\mbox{$\mathbb{N}$}}$. We will write $e_m = \psi_m \circ \phi_m$, a completely contractive module map $X \rightarrow X$. We will also say that $X$ is [*rank 1 Hilbertian*]{} if we can take $n_m = 1$ for all $m \in {\mbox{$\mathbb{N}$}}$.
Our main purpose in this section is to show that a singly generated operator module is $A$-Hilbertian if and only if it is $A$-rigged, and to attempt to thoroughly understand such modules. It is clear from the definitions, that every strong Morita equivalence $A-A$-bimodule is a left $A$-rigged module, and every left $A$-rigged module is an $A$-Hilbertian module. For C$^*-$algebras, we proved in [@Bna; @BMP] that the converse is true.
If $E$ is a closed subset of $\Omega$, we will write $J_E$ for the ideal $\{ f \in A : f(x) = 0 \; \text{for all} \;
x \in E \}$ of $A$. The following result is important for us:
[@Sm; @HWW; @ER] \[Smith\] Let $A$ be a function algebra on compact $\Omega$. If $J$ is an ideal in a function algebra $A$, then the following are equivalent:
- $J$ has a contractive approximate identity
- $J$ has a bounded approximate identity
- $J$ is an M-ideal of $A$,
- $J = J_E$, for a p-set $E$ for $A$ in $\Omega$.
We refer the reader to [@Gam] for details on p-sets and peak sets. We allow $\emptyset$ as a p-set. A p-set is an intersection of peak sets. We will not define the term ‘M-ideal’ here, and it will not play a role.
In (iv), it is clear that the $E$ is unique. Also $F = E \cap \partial A$ is a p-set for $A$ on the Shilov boundary, and $J_F = J_E$.
It is easy then to see from the definition of a rigged module above that we have:
\[co\] If $E$ is a p-set for a function algebra $A$, then the ideal $J_E$, is an $A$-rigged module.
Let $e_\alpha$ be a c.a.i. for $J_E$, and define $\psi_\alpha = \phi_\alpha$ to be multiplication by $e_\alpha$. These satisfy the requirements for a rigged module.
The proof of the following corollary requires some technical knowledge of rigged and C$^*-$modules.
\[rig\] Suppose that $A \subset
C(\Omega)$ is a function algebra on a compact space $\Omega$, and that $X$ is a singly generated left $A$-Hilbertian module. Then there is a nonnegative continuous function $f$ on $\Omega$, such that $X \cong (Af)^{\bar{}}$ $A$-isometrically.
As observed in [@Bna] §7, $W = C(\Omega) \otimes_{hA} X$ is a (singly generated) left C$^*-$module over $C(\Omega)$. Also $X$ may be regarded as an $A$-submodule of $W$ (in the obvious way). Let $f$ be the single generator of $X$ and $W$. Suppose that $I$ is the ideal in ${\mbox{${\mathcal C}$}}= C(\Omega)$ generated by the range of the inner product. Now $W$ is a full C$^*-$module over $I$, and so $IW$ is dense in $W$, since C$^*-$modules are automatically nondegenerate. Since ${\mbox{${\mathcal C}$}}f$ is dense in $W$, $I {\mbox{${\mathcal C}$}}= I$, and $IW$ is dense in $W$, we see that $If$ is dense in $W$. Hence $W$ is singly generated by $f$ over $I$. The obvious map $F: I \rightarrow If \subset W$ is adjointable and $F, F^*$ have dense range, so by the basic theory of C$^*-$modules (see e.g. [@La] Prop. 3.8), $W \cong I$, $I$-isometrically. Hence $W \cong I$ , ${\mbox{${\mathcal C}$}}$-isometrically. Since $W$ is singly generated by $f$, $|f|$ is a strictly positive element in $I$, by the argument before Theorem \[nice\] above. Of course $|f|$ is not strictly positive on $\Omega$, unless $I = A$. Clearly $X$ is $A$-isometric to the closure of the submodule $A|f|$ of $I$.
It will be useful to have the following
\[mmag\] Suppose that $f \in C(\Omega)_+$, and that $K \in A^{(n)}$. Then $\Vert a(w) K(w) \Vert_2
\leq \Vert af \Vert_{\Omega}$ for all $a \in A$ and $w \in \Omega$, if and only if $\Vert K(w) \Vert_2 \leq f(w)$ for all $w \in \partial A$.
$(\Leftarrow)$: For $w \in \partial A$, we have $$\Vert a(w)K(w) \Vert_2 \leq |a(w)f(w)| \leq \Vert af
\Vert_{\Omega} \; \;.$$ Since $aK$ achieves its maximum modulus on $\partial A$, we have proved this direction.
$(\Rightarrow)$: If $n=1$, then this is well known, following by the usual Choquet boundary point argument (such as we’ve seen, for example, in \[NEED\]). For general $n$, fix $z \in {\mbox{$\mathbb{C}$}}^n_1$. Then we have $|a(w)(z.K(w))| \leq \Vert af \Vert_{\Omega}$, for all $w \in \Omega$. Thus by the $n=1$ case, $|z.K(w)| \leq f(w)$ for all $w \in \partial A$, which gives what we need.
\[Hisr\] For a singly generated operator module $X$ over a function algebra $A$, $X$ is $A$-Hilbertian if and only if $X$ is $A$-rigged.
Suppose that $X$ is $A$-Hilbertian. We may suppose, by \[rig\], that $X = (Af)^{\bar{}}$ in $C(\Omega)$. We will take $\Omega$ to be the Shilov boundary of $A$. We use the notation of the beginning of the section, but assume, as we may, that the norms of $\phi_m ,
\psi_m$ are strictly less than 1. The map $\psi_n$ may be written as $[a_1, \cdots , a_k] \mapsto \sum_i a_i x_i$ for some $x_i \in X$, and without loss of generality, we can assume that $x_i = h_i f$, with $h_i \in A$. Thus $\psi_n$ may be associated with an $A$-tuple $H_m$, and without loss of generality, $f(w) \Vert H_m(w) \Vert_2 \leq 1$ for all $w \in \Omega$. See [@Bhmo] Prop. 2.5(ii).
Also, $\phi_m$ is completely determined by its action on $f$, so we can associate $\phi_m$ with a unique $A$-tuple $K_m$. Now it is easily seen that without loss of generality, $e_m$ may be regarded as (multiplying by) the element $H_m(w) . K_m(w)$ of $A$. We have $f e_m \rightarrow f$ uniformly.
Next we note that by Lemma \[mmag\], the identity $\Vert \phi_n(af) \Vert \leq
\Vert af \Vert_{\Omega}$ for all $a \in A$, implies that $\Vert K_m(w) \Vert_2
\leq f(w)$ for all $w \in \Omega$. Thus it follows that for $a \in A$, we have $$\Vert e_m a \Vert_\Omega \leq C_m \Vert af \Vert_\Omega$$ for a constant $C_m$ that does not depend on $a$.
Set $\phi'_m = \phi_m e_m, \psi'_m = \psi_m$. These give new factorization nets, but now we have that $$\Vert \phi'_m - \phi'_m
\psi'_n \phi'_n \Vert_{cb} =
\Vert \phi_m e_m (1 - e_n^2) \Vert_{cb}
\leq \Vert e_m(1 - e_n^2) \Vert_\Omega
\leq C_m \Vert f(1 - e_n^2) \Vert_\Omega
\rightarrow 0$$ as $n \rightarrow \infty$. This says that $X$ is a rigged module. The converse is trivial.
\[arr\] If $E$ is a closed subset of $\Omega$, we define ${\mbox{${\mathcal R}$}}_E(\Omega)$ to be the set of functions $f \in C(\Omega)$ which vanish exactly on the set $E$, and for which there exists two sequences $H_n$ and $K_n$ of $A$-tuples, such that
- $ \Vert K_n(w) \Vert_2 \leq f(w)
\leq \Vert H_n(w) \Vert_2^{-1}$ for all $w \in \Omega \; $ (interpreting $\frac{1}{0} = \infty)$,
- $e_m(w) = H_m(w) . K_m(w) \rightarrow 1$ uniformly on compact subsets $C$ of $\Omega \setminus E$ .\
We will write ${\mbox{${\mathcal R}$}}(\Omega)$ for the combined collection of all the ${\mbox{${\mathcal R}$}}_E(\Omega)$ classes.
If we write $g = \log f$, then $f \in {\mbox{${\mathcal R}$}}_E(\Omega)$ if and only if $g$ is finite precisely on $E$, and there exist $H_n, K_n$ as above, satisfying (ii) and $$(i') \; \; \log \Vert K_n(w) \Vert_2 \leq g(w) \leq
- \log \Vert H_n(w) \Vert_2 \; \; .$$ It follows from these, that $\log \Vert K_n(w) \Vert_2$ and $- \log \Vert H_n(w) \Vert_2$ converge uniformly to $g$, on compact subsets of $\Omega \setminus E$ (see Lemma \[sil\] below). Thus we see that $g$ is an upper and lower envelope of ‘sub- and superharmonic’ functions. See Proposition \[sil2\] for more on this. This definition is therefore somewhat reminiscent of the Perron process of solving the Dirichlet problem.
\[req\] ${\mbox{${\mathcal R}$}}_{\emptyset}(\Omega) = {\mbox{${\mathcal M}$}}(\Omega)$.
We leave this as an exercise for the reader.
\[H\] Let $A$ be a function algebra on $\Omega$. Then
- If ${\mbox{${\mathcal R}$}}_E(\Omega) \neq \emptyset$ then $E$ is a peak set for $A$ in $\Omega$.
- If $f \in {\mbox{${\mathcal R}$}}_E(\Omega)$, $(Af)^{\bar{}}$ is a rigged module.
- If $\Omega = \partial A$ and $f \in C(\Omega)_+$, then $(Af)^{\bar{}}$ is a rigged module if and only if $f \in {\mbox{${\mathcal R}$}}_E(\Omega)$ for some peak set $E$.
If one studies the proof of Corollary \[Hisr\], one sees that the ideas there yield that if $f \in {\mbox{${\mathcal R}$}}_E(\Omega)$ then $(Af)^{\bar{}}$ is a rigged module. If $E = f^{-1}(0)$, then $K_n$ vanishes on $E$. By (ii) of Definition \[arr\], the functions $e_m(w) = H_m(w) . K_m(w)$ form a c.a.i. for $C_0(\Omega \setminus E)$, and also for $J_E$. We deduce from Theorem \[Smith\] that $E$ is a p-set. Since $f$ is strictly positive on $\Omega \setminus E$, it follows that $E$ is a $G_\delta$, from which we deduce, using [@Gam] II.12.1, that $E
$ is a peak set. This gives (i) and (ii).
If $(Af)^{\bar{}}$ is a rigged module then the ideas of the proof of Corollary \[Hisr\] show that (i) of Definition \[arr\] holds for $w \in \partial A$. Set $E = f^{-1}(0)$, then $K_m$ vanishes on the set $F = E \cap \partial A$. We saw in that Corollary that $f e_m \rightarrow f$ uniformly on $\partial A$. By the Stone-Weierstrass theorem $C_0(\partial A \setminus F) f$ is dense in $C_0(\partial A \setminus F)$. Thus the functions $e_m(w) = H_m(w) . K_m(w)$, which are in $J_F$, form a c.a.i. for $C_0(\partial A \setminus F)$. Thus by Urysohn’s lemma, the restriction of $f$ to $\partial A$ is in ${\mbox{${\mathcal R}$}}_F(\Omega)$.
\[ops\] The singly generated rigged modules over a function algebra $A$, are exactly (up to $A$-isometric isomorphism), the modules of form $(Af)^{\bar{}}$ for $f \in {\mbox{${\mathcal R}$}}_E(\Omega)$.
If a subset $E \subset \Omega$ is a peak set for $A$ in $\Omega$, then clearly $F = E \cap \partial A$ is a peak set for $A$ in $\partial A$ . Conversely any peak set $F$ for $A$ in $\partial A$ may be written $F = E \cap \partial A$ for a unique peak set $E$ for $A$ in $\Omega$. We will therefore sometimes be sloppy, and write ${\mbox{${\mathcal R}$}}_F(\Omega)$ or ${\mbox{${\mathcal R}$}}_D(\Omega)$ for ${\mbox{${\mathcal R}$}}_E(\Omega)$ , where $D$ is the unique peak set for $A$ in $M_A$ with $D \cap \Omega = E$.
\[sil\] If $f, H_n , K_n$ is as in the definition of ${\mbox{${\mathcal R}$}}_E(\Omega)$ , then $\Vert K_n(w) \Vert_2 \rightarrow
f(w)$ uniformly on $\Omega$, and $\Vert H_n(w) \Vert_2
\rightarrow f(w)^{-1}$ uniformly on compact subsets of $\Omega \setminus E$. Also, $\Vert a f \Vert_{\partial A}
= \Vert a f \Vert_\Omega$ for all $a \in A$.
If $C$ is a compact subset of $\Omega \setminus E$, and $\epsilon > 0$ is given, then $$(1 - \epsilon) f(w)
\leq f(w) |H_m(w) . K_m(w)| \leq \Vert K_m(w) \Vert_2 \leq f(w)$$ uniformly for $w \in C$ and $m$ sufficiently large. From this one easily sees that $\Vert K_n(w) \Vert_2 \rightarrow
f(w)$ uniformly on $\Omega$, since $f$ is bounded away from zero on $C$. The second statement is similar. Finally, for $w \in \Omega$, we see that $$|a(w)f(w)| \leq \sup_m \Vert a(w)K_m(w) \Vert_2
\leq \sup_m \sup \{ \Vert a(x) K_m(x) \Vert_2 : x \in
\partial A \}
\leq \Vert a f \Vert_{\partial A} \; \; .$$
If $H$ is an $A$-tuple, then it will be useful to think of $\log \Vert H(w) \Vert_2$ as a [*subharmonic*]{} function. The logarithm of a function $f \in {\mbox{${\mathcal R}$}}_E$ on the other hand should be thought of as being [*harmonic*]{}, as we mentioned briefly before. The following few results begin to justify these assertions:
\[sil2\] Let $A$ be a uniform algebra on compact $\Omega$, and suppose that $f \in {\mbox{${\mathcal R}$}}_E(\Omega)$. Then:
- $f_{|_{\partial A}} \in {\mbox{${\mathcal R}$}}_{E \cap \partial A}(\partial A)$.
- If $g = \log f$, then $g$ achieves its maximum and minimum on $\partial A$.
- If there is a domain $R \subset {\mbox{$\mathbb{C}$}}^n$, and an inclusion $R \subset \Omega$, such that all functions in $A$ are analytic functions on $R$, then $g = \log f$ is harmonic (in the usual sense on $R$) whenever it is finite (that is, on $R \setminus E$).
It is possible that in (iii) above, $R \setminus E = R$ automatically, if $R$ is a (connected) domain as in (iii). This is the case if $A$ is the disk algebra (see comments after Example \[ex4\]). We have not checked this in general though.
\(i) is obvious. (ii): Let $K_n, H_n$ be $A$-tuples as in Definition \[arr\]. For any unit vector $z$ in the complex Euclidean space of the appropriate dimension, and any $w \in \Omega$, we have $$|K_n(w).z| \leq \Vert K_n(\cdot).z \Vert_{\partial A}
\leq \Vert f \Vert_{\partial A} \; \; .$$ Thus $\log \Vert K_n(w) \Vert_2 \leq \log \Vert f \Vert_{\partial A}$. Letting $n \rightarrow \infty$, gives $g(w) \leq \sup_{\partial A} g$. To get the other inequality, we may assume that $g$ is bounded below. Thus $E = \emptyset$. As above, we obtain $\Vert H_n(w) \Vert_2 \leq
\sup_{\partial A} |f|^{-1}$. Hence $- \log \Vert H_n(w) \Vert_2
\geq \inf_{\partial A} g$. Now let $n \rightarrow \infty$.
(iii): A function is harmonic on a domain in ${\mbox{$\mathbb{C}$}}^n$ if and only if it satisfies the Mean Value Principle. Let $H_n, K_n, z$ be as in (ii), and fix $w_0 \in R \setminus E$. Since $K_n(w).z$ is analytic for $w \in R$, we have that $\log |K_n(w).z|$ is subharmonic on $R$. Thus, for any ball $B$ center $w_0$ in $R$, we have $$\log |K_n(w_0).z| \leq \frac{1}{m(B)} \int_B \log
|K_n(w).z| \leq \frac{1}{m(B)} \int_B \log
\Vert K_n(w) \Vert_2 \leq \frac{1}{m(B)} \int_B \log f(w) \; \; .$$ Since this is true for all such $z$ we have $\log \Vert K_n(w_0) \Vert_2 \leq \frac{1}{m(B)} \int_B \log f(w)$. Taking the limit as $n \rightarrow \infty$ gives $\log f(w_0) \leq \frac{1}{m(B)} \int_B \log f(w)$. A similar argument using $H_n$ gives the other direction of the Mean Value Property.
We state the following obvious fact since it will be referred to several times:
\[sil3\] Let $A$ be a uniform algebra on compact $\Omega$, and suppose that $H$ and $K$ are $A$-tuples, such that $\Vert H(w) \Vert_2 \Vert K(w) \Vert_2 \leq 1$ for all $w \in \Omega$, or all $w \in \partial A$. Then the same inequality holds for all $w \in M_A$.
Let $z,y$ be vectors in complex Euclidean space. Then we have $|(z \cdot H(w))(y \cdot K(w))| \leq
1$ for all $w \in \Omega$, and consequently for all $w \in M_A$.
Suppose that $E$ is a peak set for $A$ , and that $f \in {\mbox{${\mathcal R}$}}_{E}(\Omega)$. Let $D$ be the unique peak set in $M_A$ with $D \cap \Omega = E$. Then there is a unique function $\tilde{f}
\in {\mbox{${\mathcal R}$}}_{D}(M_A)$ such that $\tilde{f}$ restricted to $\Omega$ equals $f$. Also, $\Vert af \Vert_\Omega = \Vert a \tilde{f} \Vert_{M_A}$ for all $a \in A$, so that $(Af)^{\bar{}}$ in $C(\Omega)$ is $A$-isometric to $(A\tilde{f} )^{\bar{}}$ in $C(M_A)$.
As we saw above, condition (ii) of definition \[arr\] is equivalent to saying that the functions $e_m(w) = H_m(w) . K_m(w)$, which are in $J_E$, form a c.a.i. for $C_0(\Omega \setminus E)$. However, $J_E = J_D $, so that $e_m$ is a c.a.i. for $J_D$. Hence by [@Gam] II.12.5 and II.12.7, taking the function $p$ there to be a continuous strictly positive function which is $1$ on $D$ and $< \epsilon$ on a compact subset $C$ of $M_A \setminus D$, we see that $ \; e_m(w) \rightarrow 1$ uniformly on $C$.
Now by lemma \[sil3\], we have $\vert e_m(w) \vert \leq \Vert H_m(w) \Vert_2 \Vert K_m(w) \Vert_2
\leq 1 $ for all $w \in \Omega$. This then implies that $\lim_m \Vert H_m(w) \Vert_2 \Vert K_m(w) \Vert_2
= 1$ uniformly on any compact subset $C$ of $M_A \setminus D$. Since $\Vert H_m(w) \Vert_2 $ and $\Vert K_m(w) \Vert$ are uniformly bounded above on $C$, we deduce that they are also uniformly bounded away from zero on $C$. By Lemma \[sil3\], we have $$\Vert H_m(w) \Vert_2 \Vert K_n(w) \Vert_2 \leq 1 \; \;
\text{for all } w \in M_A \; . \label{star}$$ Thus we have $\Vert H_m(w) \Vert_2
\Vert H_n(w) \Vert^{-1}_2 \leq 1+\epsilon$ uniformly on $C$, for $m,n$ large enough. Thus the sequence $\Vert H_n(w)\Vert_2$, and by symmetry the sequence $\Vert K_n(w)
\Vert_2$, are uniformly Cauchy on $C$. Let $\tilde{f}$ be the uniform limit of $\Vert K_n(w) \Vert_2$ on $C$. Varying over all compact $C$ gives a well defined continuous $\tilde{f}$ on $M_A \setminus D$. Clearly $\tilde{f}$ extends $f$.
We next show that $\tilde{f}
\in C_0(M_A \setminus D)$. Let $\epsilon > 0$ be given, and let $C = \{ w \in \partial A : f(w) \geq \epsilon \}$. For $\gamma > 0$ to be determined, choose (by [@Gam] II.12 again) $a \in A$ with $a \equiv 1$ on $D$, $\Vert a \Vert
\leq 1 + \gamma$, and $|a| < \gamma$ on $C$. For $x \in \partial A, m \in {\mbox{$\mathbb{N}$}}$ and any Euclidean vector $z$ of norm 1, we have $|(z \cdot K_m(x))a(x)| \leq 2 \epsilon$ , if $\gamma$ is smaller than a certain constant which depends only on $\Vert f \Vert_{\partial A}$ and $\epsilon$. Hence $|(z \cdot K_m(w))a(w)| \leq 2 \epsilon$, for all $w \in M_A$. Hence $\Vert K_m(w) \Vert_2 |a(w)| \leq 2 \epsilon$. Letting $m \rightarrow \infty$ we see that $\tilde{f}(w) |a(w)| \leq 2 \epsilon$. In particular, for $w \in U = |a|^{-1}((1-\epsilon,\infty))$, we have $\tilde{f}(w) \leq \frac{2 \epsilon}{1-\epsilon}$. Thus indeed $\tilde{f} \in C_0(M_A \setminus D)$.
Define $\tilde{f}$ to be zero on $D$. Clearly $\Vert K_m(w) \Vert_2 \rightarrow \tilde{f}(w)$ for all $w \in M_A$. For $w \notin D$ we obtain from (\[star\]), that $\Vert H_m(w) \Vert \tilde{f}(w) \leq 1$, and also $\Vert K_m(w) \Vert_2 \leq \tilde{f}(w)$. Thus $\tilde{f} \in {\mbox{${\mathcal R}$}}_E({\mbox{${\mathcal M}$}}_A)$.
Finally, for the uniqueness, we suppose that $f_1$ with $H^1_m, K^1_m$, and $f_2$ with $H^2_m, K^2_m$, both fulfill the definition of ${\mbox{${\mathcal R}$}}_E(M_A)$. If $f_1(x) = f_2(x)$ for all $x \in \partial A$, then $\Vert H^1_m(w) \Vert_2 \Vert K^2_m(w) \Vert_2 \leq 1$ on $\partial A$, and hence, by Lemma \[sil3\], on $M_A $. Hence $f_2(w) \leq f_1(w)$ for any $w \notin D$. This obviously implies what we want, by symmetry.
In view of the previous result we may simply write ${\mbox{${\mathcal R}$}}_E$ , for ${\mbox{${\mathcal R}$}}_E(\Omega)$, if we wish. As remarked earlier, we may switch $E$ for the corresponding peak set in $M_A$ or $\partial A$.
\[ex4\]
The nontrivial p-sets $E$ for the disk algebra $A({\mbox{$\mathbb{D}$}})$ coincide with the peak sets, and they are exactly the closed subsets of ${\mbox{$\mathbb{T}$}}$ of Lebesgue measure $0$ (see [@HWW] for example). By the well known version of Beurlings theorem for the disk algebra, $J_E$ is a singly generated $A({\mbox{$\mathbb{D}$}})$-module. Hence, for $E \subset {\mbox{$\mathbb{T}$}}$ with $|E| = 0$, we have by Corollary \[co\] that $J_E$ is an example of a singly generated rigged module over $A({\mbox{$\mathbb{D}$}})$. Indeed every closed ideal $I$ of $A({\mbox{$\mathbb{D}$}})$ is isometrically isomorphic to some $J_E$, and consequently is an $A$-rigged module. This is because, by Beurlings theorem, $I = J_E g$ for a fixed inner function $g$ and peak set $E$.
Next we find all singly generated rigged modules over $A({\mbox{$\mathbb{D}$}})$. They all turn out to be ‘rank one Hermitian’.
\[ad\] Suppose that $f \in C({\mbox{$\mathbb{T}$}})_+$. The following are equivalent:
- $\log f$ is integrable on ${\mbox{$\mathbb{T}$}}$,
- $f = |\phi|$ for a function $\phi \in H^\infty \setminus
\{0 \}$,
- $(A({\mbox{$\mathbb{D}$}}) f)^{\bar{}}$ is a rigged module over $A({\mbox{$\mathbb{D}$}})$,
- $f \in {\mbox{${\mathcal R}$}}({\mbox{$\mathbb{T}$}})$.
Moreover, every singly generated rigged module over $A({\mbox{$\mathbb{D}$}})
$ is $A({\mbox{$\mathbb{D}$}})$-isometric to one of the form in (iii).
The equivalence of (i) and (ii) is classical ([@Ho2] p. 53).
Suppose that $X$ is a singly generated rigged module over $
A({\mbox{$\mathbb{D}$}})$. By Corollary \[ops\] and the fact mentioned in the previous paragraph, we have $X \cong (A({\mbox{$\mathbb{D}$}})f)^{\bar{}}$ $A$-isometrically, where $E$ is a subset of ${\mbox{$\mathbb{T}$}}$ of Lebesgue measure $0$, and $f \in {\mbox{${\mathcal R}$}}_E({\mbox{$\mathbb{T}$}})$. With the earlier notation, we have $\Vert K_m(w) \Vert_2 \leq f(w)$, and $\Vert K_m(w) \Vert_2 \rightarrow f(w)$, for all $w \in {\mbox{$\mathbb{T}$}}$. This implies that there is a nonzero function $K \in A({\mbox{$\mathbb{D}$}})$, such that $|K| \leq f$ on ${\mbox{$\mathbb{T}$}}$. Then $\log |K| \leq \log f$, a.e. on ${\mbox{$\mathbb{T}$}}$. Since $\log |K|$ is integrable on ${\mbox{$\mathbb{T}$}}$, so is $\log f$ .
Conversely, let $\phi \in H^\infty$, with $f = |\phi|$ continuous on $\bar{{\mbox{$\mathbb{D}$}}}$. Let $E$ be the subset of ${\mbox{$\mathbb{T}$}}$ on which $\phi$ vanishes, which is a a closed subset of measure $0$. Let $w = \log |\phi|$, then $w$ is integrable. We choose a function $k_1$ on ${\mbox{$\mathbb{T}$}}$ such that $w - \epsilon \leq k_1 \leq w$, and such that $k_1$ is continuously differentiable wherever it is finite. Indeed, one may assume that $k_1$ lies in a thin strip about $w - \frac{\epsilon}{2}$. We define $$k(z) = \exp \left( \frac{1}{2\pi} \int_{-\pi}^{\pi}
\frac{e^{i \theta} + z}{e^{i \theta} - z} \; k_1(\theta) \;
d \theta
\right) \; \; .$$ Then $k \in A({\mbox{$\mathbb{D}$}})$, and $\vert k \vert \leq \vert
\phi \vert$ on ${\mbox{$\mathbb{T}$}}$.
Next, choose an open subset $U$ of ${\mbox{$\mathbb{T}$}}$ containing $E$, such that $\int_U |k_1| < \epsilon$, and such that $k_1 < -1$ on $U$. We may suppose that $U$ is a finite collection of disjoint open intervals, of total combined length $< \epsilon$. Choose $k_2 = k_1 + \epsilon$ outside $U$ . On $U$ we define $k_2$ so that $k_2$ lies between $0$ and $w$ on $U$, and so that $k_2$ is finite and continuously differentiable on all of ${\mbox{$\mathbb{T}$}}$. It is not hard to see that this is possible. Then we have $\int_U (k_2 - k_1) \leq \int_U |k_1| \leq \epsilon$. We define $h \in A({\mbox{$\mathbb{D}$}})$ by the formula defining $k$ above, but with $k_1$ replaced by $-k_2$. Then $h$ is nonvanishing on ${\mbox{$\mathbb{T}$}}$, and $\vert k \vert \leq \vert \phi \vert
\leq |h|^{-1}$ on ${\mbox{$\mathbb{T}$}}$. Setting $r = k_2 - k_1$, we may write, for fixed $z \in {\mbox{$\mathbb{D}$}}$: $$\int_{{\mbox{$\mathbb{T}$}}} \frac{e^{i \theta} + z}{e^{i \theta} - z} \; r(
e^{i \theta})
\; = \epsilon \int_{{\mbox{$\mathbb{T}$}}}
\frac{e^{i \theta} + z}{e^{i \theta} - z} \; - \;
\epsilon \int_U \frac{e^{i \theta} + z}{e^{i \theta} - z}
\; +
\; \int_U \frac{e^{i \theta} + z}{e^{i \theta} - z} \; r(e^{i \theta})
\; \; .$$ The first of the three terms on the right equals $2 \pi \epsilon$. Supposing that $d(z,U) \geq \delta$, we have $$|\frac{e^{i \theta} + z}{e^{i \theta} - z}| \leq \frac{2
}{\delta} \; \; ,$$ for $e^{i \theta} \in U$, whence $$| \frac{1}{2\pi} \int_{-\pi}^{\pi}
\frac{e^{i \theta} + z}{e^{i \theta} - z} \; (k_2-k_1)(e^{i
\theta})
\; d \theta | \; \leq \; \epsilon (1 + \frac{4}{\delta}) \;
.$$
We now check that (ii) of Definition \[arr\] holds. By an easy compactness argument, we may assume that the compact subset $C \subset \Omega \setminus E$ there, is a finite closed interval. Pick $\epsilon$ so small in relation to $d(C,E)$, that for any $z$ close enough to $C$, we have $d(z,s) \geq \sqrt{\epsilon}$ for any $s \in
{\mbox{$\mathbb{T}$}}$ with $d(s,E) < \epsilon$. For the $k_1, k_2$ associated with this $\epsilon$, we have, for $z$ close to $C$, that $$\vert \frac{1}{2\pi} \int_{-\pi}^{\pi}
\frac{e^{i \theta} + z}{e^{i \theta} - z} (k_2 - k_1)(e^{i \theta})
\; d \theta \vert \leq 5 \sqrt{\epsilon} \; \; .$$ Thus $$\left| \exp \left( \frac{1}{2\pi} \int_{-\pi}^{\pi}
\frac{e^{i \theta} + z}{e^{i \theta} - z} (k_1 - k_2)(e^{i \theta})
\; d \theta \right)
\; \; - \; \; 1 \right| \leq 6 \sqrt{\epsilon} \; \; ,$$ for all $z$ close enough to $C$ . Thus $ |k(e^{i \theta})h(e^{i \theta}) - 1| \leq 6
\sqrt{\epsilon} $, on $C$. From this it is clear that the conditions of Definition \[arr\] are met, so that $A f$ is a rigged module over $A({\mbox{$\mathbb{D}$}})$.
One may always choose the $\phi$ in the last theorem to be an outer function, and then $|\phi|$ will also be the ‘unique harmonic extension’ of $f$ to $M_A = \bar{{\mbox{$\mathbb{D}$}}}$.
In this case of $A = A({\mbox{$\mathbb{D}$}})$, we see that any $f \in {\mbox{${\mathcal R}$}}$ is nonvanishing inside ${\mbox{$\mathbb{D}$}}$, and has a harmonic logarithm on all of ${\mbox{$\mathbb{D}$}}$.
\[sg\] The set ${\mbox{${\mathcal R}$}}(\Omega)$ is a unital semigroup. Indeed, if $f_1 \in {\mbox{${\mathcal R}$}}(\Omega)$ and $f_2 \in C(\Omega)_+
$ then $(Af_1)^{\bar{}} \otimes_{hA} (Af_2)^{\bar{}}
\cong (Af_1 f_2)^{\bar{}}$ (completely) $A$-isometrically. If $f_1 \in {\mbox{${\mathcal R}$}}_{E_1}$, and if $f_2 \in {\mbox{${\mathcal R}$}}_{E_2}$, then $f_1 f_2 \in {\mbox{${\mathcal R}$}}_{E_1 \cup E_2}$.
Clearly the multiplication map $\Phi :
(Af_1)^{\bar{}} \otimes_{hA} (Af_2)^{\bar{}}
\rightarrow (Af_1 f_2)^{\bar{}}$ is completely contractive, and has dense range. Conversely, choose $H_m, K_m$ as in Definition \[arr\], and let $e_m = H_m \cdot K_m$ as before. For $a \in A$, define $\theta_m(af_1 f_2)
= e_m f_1 \otimes f_2 a = H_m f_1 \odot K_m f_2 a $. The $\odot$ notation here, is commonly used with reference to the Haagerup tensor product. Namely, for two finite tuples $x =(x_i), y = (y_i)$ , the expression $x \odot y$ means $\sum_i x_i \otimes y_i$. If $g = \chi_E f_1^{-1}$, then $\theta_m(af_1f_2) = H_m f_1 \odot K_m g (af_1f_2)$. From the definition of the Haagerup tensor product [@BMP] we see from this that $\theta_m$ is well defined and completely contractive. It is easy to see that $\theta_m (\Phi(u)) \rightarrow u$ , for all $u \in Af_1 \otimes_A Af_2$. Hence $\Phi$ is a complete isometry.
We leave it to the reader to check that if $f_1, f_2 \in {\mbox{${\mathcal R}$}}(\Omega)$, then $f_1f_2 \in {\mbox{${\mathcal R}$}}(\Omega)$. Thus ${\mbox{${\mathcal R}$}}(\Omega)$ is a unital semigroup. The last assertion is also an easy exercise.
Thus $\log {\mbox{${\mathcal R}$}}_E$ and $\log
{\mbox{${\mathcal R}$}}(\Omega)$ are ‘harmonic classes’ in the sense of the introduction.
Applications to Morita bimodules.
=================================
\[whatname\] If $X$ is an algebraically singly generated faithful function module (or equivalently, of the form $Ag$ for some $g \in
C(\Omega)^+$), the following are equivalent:
- $X \cong Af$ $A$-isometrically, for some $f \in {\mbox{${\mathcal M}$}}$;
- For any $\epsilon > 0$, there exists $n \in {\mbox{$\mathbb{N}$}}$, and $A$-module maps $\varphi : X \rightarrow A^{(n)}$ and $\psi : A^{(n)} \rightarrow X$, with $\psi \circ \varphi =
Id_X$, and $\Vert \varphi \Vert_{cb} \leq 1 + \epsilon$ and $\Vert \psi \Vert_{cb} = \Vert \psi \Vert \leq 1 + \epsilon$;
- $X$ is $A$-rigged.
In (ii), the operator space structure on $A^{(n)}$ is $R_n(A)$, as in the definition of $A$-Hilbertian.
$(i) \Rightarrow (ii)$: This follows from the definition of ${\mbox{${\mathcal M}$}}$; given $\epsilon > 0$ there exist $A$-tuples $H, K$ with $H.K = 1$, such that the norms of $Hf$ and $Kf^{-1}$ are close to 1. These may be associated with maps $\varphi$ and $\psi$ as in the proof of Corollary \[Hisr\].
$(ii) \Rightarrow (iii)$: Follows from the definition of $A$-Hilbertian, and the fact that such $X$ is $A$-rigged if and only if it is $A$-Hilbertian.
$(iii) \Rightarrow (ii)$: The maps $\phi_m, \psi_m$ in the definition of $A$-Hilbertian, may be associated as in the proof of Corollary \[Hisr\], with certain $A$-tuples $H_m, K_m$. We have $(H_m.K_m) f = \psi_m(\phi_m(f)) \rightarrow
f$. Hence $(H_m.K_m) \rightarrow 1$, so that by the common Neumann series trick, $H_m.K_m$ is an invertible element $b^{-1}$ of $A$, and $|b| \approx 1$. Replace $K_m$ by $K_m b$. Correspondingly we get an adjusted $\phi'_m, \psi'_m$, which now satisfies (ii).
$(ii) \implies (i)$ we state as the next result.
The following result is highly analagous to Theorem \[new\]:
\[new3\] Suppose that $A$ is a uniform algebra on compact $\Omega$, and that $f \in C(\Omega)^+$. Then $X = Af$ satisfies condition (ii) (or equivalently (iii)) of the above theorem, if and only if $Af \cong (Af)_{|\partial A}$ isometrically via the restriction map, and $f_{|\partial A}
\in {\mbox{${\mathcal M}$}}(\partial A)$.
The $(\Leftarrow)$ direction is ($(i) \Rightarrow (ii)$) of the previous theorem.
Supposing (ii) of the Theorem, we proceed as in the proof of Corollary \[Hisr\], to associate with $\varphi$ and $\psi$, $A$-tuples $H$ and $K$. We have that $1=H.K$, and as in that proof we get $f(\cdot) \Vert H(\cdot) \Vert_2 \leq 1+\epsilon$ on $\Omega$, and $\Vert K(\cdot) \Vert_2 \leq
(1+\epsilon) f(\cdot)$ on $\partial A$. Applying Cauchy-Schwarz to $1 = |(Hf).(Kf^{-1})|$, and using these inequalities, shows that $(1+\epsilon) \Vert K(\cdot) \Vert_2 \geq f$ on $\Omega$, and $\Vert H(\cdot) \Vert_2 \approx f^{-1}$ on $\partial A$. Thus $f_{|\partial A}
\in {\mbox{${\mathcal M}$}}(\partial A)$. That $Af \cong (Af)_{|\partial A}$ follows as in the proof of \[Gmax\].
\[wenmo\] Suppose that $X$ is a [*singly generated*]{} left $A$-rigged module. Then the following are equivalent:
- The peak set $E$ associated with $X$ is the empty set,
- $X$, with the obvious (symmetric) right module action, is a strong Morita equivalence $A-A$-bimodule.
- $X$ is algebraically singly generated.
If $M_A$ is connected, then the above are also equivalent to:
- $X$ is [*algebraically*]{} finitely generated and projective as a left $A$-module.
$(i) \Rightarrow (ii)$: Follows since ${\mbox{${\mathcal M}$}}= {\mbox{${\mathcal R}$}}_{\emptyset}$.
$(ii) \Rightarrow (i)$ : Assuming (ii), then by Theorem \[nice\], $X \cong Af$ $A$-isometrically, where $f \in C(\partial A)^+$. Since every strong Morita equivalence bimodule is a rigged module, Corollary \[H\] (iii) now gives (i).
$(ii) \Rightarrow (iv)$: This is true for any algebraic Morita equivalence $A-A$-bimodule [@F].
$(iv) \Rightarrow (i)$: It follows by [@Bhmo] Theorem 3.6 (6), and \[just\] below, that ${\mbox{$\mathbb{K}$}}(X) = J_E$ is unital. Thus $\chi_E$ is continuous, so that $E$ is closed and open. If $\Omega$ is connected, it follows that $E$ is the empty set.
$(iii) \Rightarrow (ii)$: Every singly generated $A$-rigged module is of the form $(Af)^{\bar{}}$, which is a faithful function $A$-module. Now appeal to ($(iii) \Rightarrow (i)$) of Theorem \[whatname\].
The proof of $(ii) \Rightarrow (i) \Rightarrow (ii)$ above, together with the basic theory of strong Morita equivalence (see [@BMP] section 4), gives a strengthened form of Theorem \[nice\]:
\[Motat\] Let $X$ be a singly generated symmetric strong Morita equivalence $A-A$-bimodule. Then there exists an $f \in {\mbox{${\mathcal M}$}}$ such that $X \cong Af$ $A$-isometrically, $\tilde{X} \cong B_A(X,A) \cong Af^{-1}$ $A$-isometrically, and the strong Morita context associated with $X$ may be identified with $(A,A,Af,Af^{-1})$.
Finally, we show that every singly generated rigged module over a function algebra, is a strong Morita equivalence bimodule (over a possibly different algebra):
\[just\]
- If $X$ is a singly generated operator module over a function algebra $A$ on a compact space $\Omega$, then $X$ is an $A$-rigged module if and only if there exists a peak set $E$ in $\Omega$ for $A$, and a function $f \in {\mbox{${\mathcal R}$}}_E(\Omega)$, such that $X \cong (J_E f)^{\bar{}}$ $A$-isometrically.
- If the equivalent conditions in (i) hold, then $X$ , with the obvious (symmetric) right module action, is a strong Morita equivalence $J_E-J_E$-bimodule.
- Conversely, if $E$ is a p-set, then any strong Morita equivalence $J_E-J_E$-bimodule, or more generally any $J_E$-rigged module, is an $A$-rigged module.
The proof requires some technical knowledge of rigged modules [@Bhmo].
Here is one way to see (iii). If $X$ is a left $J_E$-rigged module, then from Corollary \[co\] and §6 of [@Bhmo], $J_E \otimes_{hJ_E} X$ is a left $A$-rigged module. But $J_E \otimes_{hJ_E} X \cong X$, $A$-isometrically.
To get (i) and (ii), suppose that $X$ is a singly generated $A$-rigged module. Then by the previous results, there is a peak set $E$ and a function $f \in {\mbox{${\mathcal R}$}}_E(\Omega)$ such that $X \cong (Af)^{\bar{}}$ $A$-isometrically. Since $f$ is a strictly positive element of the ideal $I$ in Lemma \[rig\], we see that $I = \{ p \in C(\Omega) :
p(x) = 0 \; \text{for all} \; x \in E \}$. Suppose that $Y = \tilde{X}$ is the dual rigged module of $X$ (see [@Bhmo; @BMP] for details), where the reader may also find the definition of ${\mbox{$\mathbb{K}$}}(X)$, which we shall need shortly. Since $W$ may be taken to be $I$, it follows that the linking C$^*-$algebra for $W$ is $M_2(I)$. Thus we can make the following deductions from Theorem 5.10 in [@Bhmo]. Firstly, ${\mbox{$\mathbb{K}$}}(X)$ may be identified with a closed subalgebra $J$ of $I$, and $J$ has a contractive approximate identity which is a contractive approximate identity for $I$. Also, $Y$ may be regarded as a subspace of $I$, and the canonical pairings $X \times Y \rightarrow A $, and $Y \times X \rightarrow
{\mbox{$\mathbb{K}$}}(X) = J$ may be regarded as the commutative multiplication in $I$ . Thus it follows that $J$ is the closure of the span of the range of the canonical pairing $X \times Y \rightarrow
A$ . Thus $J$ is also a subset, indeed a closed ideal, of $A$. By the commutativity of the multiplication in $I$, $X$ is a strong Morita equivalence $J-J$-bimodule. We also see that $JX$ is dense in $X$.
Since $J$ has a contractive approximate identity, we may appeal to Theorem \[Smith\] to see that $J = J_{E'}$ for a p-set $E'$. Since $J$ contains a c.a.i. for $I$, it is clear that $E' \subset E$. On the other hand, if $w \in E \setminus E'$, then there is a peak set $F$ containing $E'$, with $w \notin F$. Choose $a \in A$ with $a \equiv 0$ on $E'$ and $a(w) \neq 0$. Then $a \in J$, so $a \in I$. This is impossible, so that $E = E'$.
Note that $Af$ is dense in $X$, $JX$ is dense in $X$ and $JA = J$. Hence $Jf$ is dense in $X$. Thus $X = (Jf)^{\bar{}}$.
We thank C. Le Merdy, R. R. Smith, and V. Paulsen for helpful conversations.
[99]{}
W. B. Arveson, [*Subalgebras of* ]{}$C^{*}-$[*algebras,*]{} I Acta Math. [**123** ]{}(1969), 141-22; II [**128**]{} (1972), 271-308.
E. Behrends and M. Cambern, * An isomorphic Banach-Stone theorem,* Studia Math. **90** (1988), 15–26.
B. Blackadar, [*K-theory for Operator Algebras,*]{} 2nd Ed., Math. Sci. Research Inst. Publications, Cambridge University Press (1998).
D. P. Blecher, [*A generalization of Hilbert modules,*]{} J. Funct. Analysis, [**136**]{} (1996), 365-421.
D. P. Blecher, [*A new approach to Hilbert C$^*-$modules,*]{} Math Ann. [**307**]{} (1997), 253-290.
D. P. Blecher, [*Some general theory of operator algebras and their modules,*]{} in [*Operator algebras and applications*]{}, A. Katavalos (editor), NATO ASIC, Vol. 495, Kluwer , Dordrecht, 1997.
D. P. Blecher, [*Modules over operator algebras, and the maximal C$^*-$dilation,*]{} Preprint (1998) To appear J. F. A..
D. P. Blecher, [ *The Shilov boundary of an operator space - and applications to the characterization theorems and Hilbert $C^*$-modules*]{}, Preprint (1999).
D. P. Blecher and C. Le Merdy, [*On function and operator modules,*]{} Preprint (1999).
D. P. Blecher, P. S. Muhly and Q. Na, [*Morita equivalence of operator algebras and their C$^*-$envelopes,*]{} to appear Bulletin of the London Math. Soc.
D. P. Blecher, P. S. Muhly and V. I. Paulsen, [*Categories of operator modules - Morita equivalence and projective modules*]{}, (1998 Revision), To appear Memoirs of the A.M.S.
L. Brown, P. Green, and M. Rieffel, [*Stable isomorphism and strong Morita equivalence of* ]{}$C^{\ast }-$[*algebras*]{}, Pacific J. Math. [**71**]{} (1977), 349-363.
R. G. Douglas and V. I. Paulsen, [*Hilbert modules over function algebras,*]{} Pitman (1989).
M. J. Dupre and R. M. Gillette, [*Banach bundles, Banach modules and automorphisms of C$^*-$algebras*]{}, Pitman (1983).
E. G. Effros and Z.-J. Ruan, On non-self adjoint operator algebras, Proc. Amer. Math. Soc., 110(1990), 915-922.
C. Faith, [*Algebra I: Rings, Modules, and Categories,*]{} Springer-Verlag Berlin Heidelberg New York (1981).
T. W. Gamelin, [*Uniform algebras,*]{} Prentice Hall (1969).
A. M. Gleason, [*Function algebras,*]{} Seminar on Analytic functions, vol. II, Inst. for Adv. Study, Princeton, N.J. 1957.
P. Harmand, D. Werner and W. Werner, [*M-ideals in Banach spaces and Banach algebras,*]{} Springer-Verlag, Berlin - New York (1993).
K. Jarosz, *Perturbations of uniform algebras*, Bull. London Math. Soc. **15** (1983), 133-138.
K. Jarosz, *Perturbations of Banach Algebras*, Springer-Verlag Lecture Notes in Math. 1120 (1985).
K. Jarosz, *Small isomorphisms of $C(X,E)$ spaces*, Pacific J. Math. **138** (1989), 295–315.
K. Hoffman, [*Analytic functions and logmodular Banach algebras,*]{} Acta Math. [**108**]{} (1962), 271-317.
K. Hoffman, [*Banach spaces of analytic functions*]{}, Dover (1988).
E. C. Lance, [*Hilbert C$^*$-modules - A toolkit for operator algebraists,*]{} London Math. Soc. Lecture Notes, Cambridge University Press (1995).
M. Nagasawa, *Isomorphisms between commutative Banach algebras with application to rings of analytic functions*, Kodai Math. Sem. Rep. Math. **11** (1959), 182-188.
I. Raeburn, [*On the Picard group of continuous trace C$^*$-algebras,*]{} Trans. Amer. Math. Soc. [**263**]{} (1981), 183-205.
M. Rieffel, [*Morita equivalence for operator algebras,* ]{}Proceedings of Symposia in Pure Mathematics [**38**]{} Part 1 (1982), 285-298.
R. Rochberg, *Deformation of uniform algebras on Riemann surfaces,* Pacific J. Math. **121** (1986), 135-181.
R. R. Smith, [*An addendum to M-ideal structure in Banach algebras,*]{} J. Funct. An. [**32**]{} (1979), 269-271.
E. L. Stout, *The theory of uniform algebras,* Bogden and Quigley (1971).
I. Suciu, *Function algebras*, Noordhoff Int. Pub. (1975).
[^1]: A left module $X$ is faithful if $aX = 0$ implies $a = 0$.
[^2]: See [@sto] Theorem 12.1 for example. An algebra with finite antisymmetric decomposition, is a direct sum of finitely many uniform algebras.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Ruohan Gao^1^, Rogerio Feris^2^, Kristen Grauman^3^'
bibliography:
- 'ref\_RG.bib'
title: |
Learning to Separate Object Sounds by\
Watching Unlabeled Video
---
**Acknowledgements:** This research was supported in part by an IBM Faculty Award, IBM Open Collaboration Research Award, and DARPA Lifelong Learning Machines. We thank members of the UT Austin vision group and Wenguang Mao, Yuzhong Wu, Dongguang You, Xingyi Zhou and Xinying Hao for helpful input. We also gratefully acknowledge a GPU donation from Facebook.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
In this paper we consider the multi-dimensional pressureless Euler system and we tackle the problem of existence and uniqueness of sticky particle solutions for general measure-type initial data. Although explicit counterexamples to both existence and uniqueness are known since [@BN], the problem of whether one can still find sticky particle solutions for a large set of data and of how one can select them was up to our knowledge still completely open.
In this paper we prove that for a comeager set of initial data in the weak topology the pressureless Euler system admits a unique sticky particle solution given by a free flow where trajectories are disjoint straight lines.
Indeed, such an existence and uniqueness result holds for a broader class of solutions decreasing their kinetic energy, which we call dissipative solutions, and which turns out to be the compact weak closure of the classical sticky particle solutions. Therefore any scheme for which the energy is l.s.c. and is dissipated will converge, for a comeager set of data, to our solution, i.e. the free flow.
author:
- 'Stefano Bianchini[^1]'
- 'Sara Daneri[^2]'
title: 'On the sticky particle solutions to the multi-dimensional pressureless Euler equations'
---
Introduction
============
We consider the pressureless Euler system in $[0,T]\times{\mathbb{R}}^d$ $$\label{eq:E}
\left\{\begin{aligned}
&\partial_t\rho+\div(\rho v)=0 \\
&\partial_t(\rho v)+\div(\rho v\otimes v)=0,
\end{aligned}\right.$$ where $\rho$ is the distribution of particles and $v$ is their velocity.
Such a model has been proposed by Zeldovich [@Zel] as a simplified model for the early stages of the formation of galaxies, when a dust of particles moving without pressure should start to collide and aggregate into bigger and bigger clusters.
Since then, several authors devoted attention to the search of sticky particle solutions, namely solutions to which satisfy the following adhesion principle: if two particles of fluid do not interact, then they move freely keeping constant velocity, otherwise they join with velocity given by the balance of momentum.
The great majority of the results in the literature (see e.g. [@BGSW; @BG; @CSW1; @ERS; @Gre; @HW; @NS; @NT1]) are concerned with the one-dimensional pressureless dynamics. In this case, exploiting the density of finite particle solutions, one can obtain from quite general initial data a global measure solution of satisfying a suitable entropy condition (see [@ERS] and independently [@Gre]). For a different approach see also [@BJ; @PR]. Improvements of this results regarding uniqueness of solutions were given among others in [@BG; @HW; @NT1]. In [@BG] an equivalent formulation of is given, proving that the cumulative distribution function of the particle density is the entropy solution of a scalar conservation law. In [@NT1] it is shown that the velocity field of such solution satisfies the Oleinik condition. In [@HW] uniqueness of solutions for Radon measure initial data is shown. In [@NS] the authors give an alternative characterization of the evolution of observing that the monotone rearrangement $X_t$ of $\rho_t$ satisfies $X_t=P_{K}(X_0+tv_0)$ and $v_t=\dot X_t$, where $P_K$ is the projection operator on the cone $K$ of monotone maps. This allows the authors to investigate finer properties of solutions, and in particular their connections with gradient flows in Wasserstein spaces. A different more direct proof of the equivalence of the formulation introduced in [@NS] has been given in [@CSW1]. These approaches show that the velocity of particles is uniquely determined and the sticky particle condition is satisfied. See also [@Bo; @Sob] for viscous approximations of and [@NT2] for a study of with an additional viscosity.
In general dimension, much less is known. For initial data given by a finite number of particle pointing each in a given direction, it is easy to show that a global sticky particle solution always exists and is unique. However, in dimension $d\geq2$, one sees immediately already from a finite number of particles that the sticky particle solutions do not depend continuously on the initial data.
In [@BN] it is shown that, in general, both existence and uniqueness might fail: it is indeed possible to build initial data of non-existence or non-uniqueness for the sticky particle solutions, in contrast to what had been erroneously stated in [@Sev]. In particular, in dimension $d\geq2$ one cannot hope for a well-posedness of the Cauchy problem in the set of sticky particle solutions for all measure-type initial data as in the one-dimensional case.
In [@CSW2] measure valued solutions to on a compactification of the state space have been constructed for general initial data as limits of variational in time discretizations. Such solutions dissipate the total kinetic energy, and the approximating maps are constructed in the spirit of the adhesive dynamics, but no sticky particle property for the limit measure valued solutions is given.
Thus the natural question of whether one can still find particle solutions for a large class of data (hence excluding the counterexamples in [@BN]) remained unanswered. In this paper we give a positive answer to this question.
In order to state our main result, define $$\mathcal P_{2,1}({\mathbb{R}}^d\times{\mathbb{R}}^d):=\Bigl\{\nu_0\in\mathcal P({\mathbb{R}}^d\times{\mathbb{R}}^d):\,\int|x|^2{\mathtt{p}_x}_{\#}\nu_0\leq1,\,\int|v|^2{\mathtt{p}_v}_{\#}\nu_0\leq1\Bigr\},$$ where $\mathcal P({\mathbb{R}}^d\times{\mathbb{R}}^d)$ are the probability measures on ${\mathbb{R}}^d\times{\mathbb{R}}^d$, $(x,v)\in{\mathbb{R}}^d\times{\mathbb{R}}^d$ the position-velocity coordinates and $\mathtt {p}_x$ $(\mathtt {p}_v)$ the projection operators on the first (last) $d$ coordinates. Moreover, we consider the problem of existence and uniqueness in a larger class of solutions which we call *dissipative* since in particular their kinetic energy is decreasing but their trajectories might cross without joining at later times. By *free flow* we mean a flow in which trajectories are disjoint straight lines which never intersect.
Our main result is the following:
\[thm:freeflow0\]
There is a set $D_0\subset\mathcal P_{2,1}({\mathbb{R}}^d\times{\mathbb{R}}^d)$ such that, for any $\nu_0\in D_0$ there exists a unique dissipative solution $\eta$ with initial data $\nu_0$ and it is given by a free flow. Such a set is a dense $G_{\delta}$ set (i.e. of second category) in the weak topology on $\mathcal P_{2,1}({\mathbb{R}}^d\times{\mathbb{R}}^d)$.
Since our notion of dissipative solution includes the classical sticky particle solutions, the above theorem implies that, even though the sticky particle solutions are not well-posed for every measure-type initial data, there exists a comeager set of initial data in the weak topology giving rise to a unique sticky particle solution. Moreover, for any of these initial data the sticky particle solution is unique also in the larger class of dissipative solutions (where trajectories are allowed to cross) and is given by a trivial free flow concentrated on trajectories which do not intersect. In particular for such initial data there is only one dissipative solution and its dissipation is equal to zero. Thus, for a comeager set of initial data the problem of finding sticky particle solutions is well-posed, but the dynamics that one sees is trivial.
Both the concepts of dissipative and classical sticky particle solutions are defined at a Lagrangian level as measures on the space of curves with finite energy. The class of dissipative solutions turns out to be the compact weak closure of the set of classical sticky particle solutions (see Theorem \[thm:stickydense\]).In Section \[sec:diss\] we introduce the concept of dissipative solution and we show that this class is compact and includes the classical sticky particle solutions. Then in Section \[sec:approx\] we show that dissipative solutions can be approximated in the weak topology by classical sticky particle solutions. In Section \[S:PDE\_formul\] we give a kinetic and PDE formulation of our notion of dissipative solution. Given the approximation result of Section \[sec:approx\], in Section \[sec:gdelta\] we use the fact that the fact that the dimension is greater than or equal to $2$ to modify the initial data of such finite particle solutions in order to have the trivial free flow as unique solution, while staying close in the weak topology. Such data in particular have the property that every dissipating solution starting from them has zero dissipation. The fact that such initial data are a $G_\delta$ set follows from the compactness of the set of dissipative solutions and the upper semicontinuity of the dissipation.\
The construction of this dense $G_\delta$-set relies on some natural assumptions on the approximation scheme, in particular that the energy is l.s.c. and that if the dissipation of energy is $0$ then the only solution is the free flow, see Remark \[Rem:other\_G\_delta\] for details. Hence one concludes that for a dense $G_\delta$-set of initial data the weak solutions constructed by any reasonable approximation scheme coincide with our dissipative solutions, i.e. the free flow.
Preliminaries
=============
We define the following space of curves. For a fixed $T>0$, $$\Gamma := \Big\{ \gamma \in L^2 \big( (-1,T), {\mathbb{R}}^d \big):\, \gamma \llcorner_{(-1,0)} \ \text{affine} \Big\}.$$
\[Rem:other\_possible\_spaces\] The above choice is in order to avoid assigning the initial speed $W_0(\gamma)$ and considering the space ${L^2}((0,T);{\mathbb{R}}^d)) \times {\mathbb{R}}^d$ with the product topology. The solutions we consider will actually be in $W^{1,2}((-1,T),{\mathbb{R}}^d))$ (see Lemma \[Lem:calM\_right\] below): of course the map $$\Gamma \ni \gamma \mapsto \big( \gamma \llcorner_{(0,T)},\gamma(0) - \gamma(-1) \big) \in W^{1,2}((0,T);{\mathbb{R}}^d) \times {\mathbb{R}}^d$$ is a bijection, with inverse $$W^{1,2}((0,T);{\mathbb{R}}^d) \times {\mathbb{R}}^d \ni (\gamma,W_0) \mapsto \begin{cases}
\lim_{s \searrow 0} \gamma(s) - t W_0 & t \in (-1,0], \\
\gamma(t) & 0 < t < T.
\end{cases}$$ Actually these maps are bounded linear operators [when we consider the weak or strong topology of $W^{1,2}$]{}.
For every $\gamma\in\Gamma$, we define the initial velocity field as
$$\label{eq:v0}
W_0(\gamma) = \gamma(0) - \gamma(-1).$$
This function is continuous in the topology of $\Gamma$.
We denote by $\mathcal P(\Gamma)$ the set of Borel probability measures on $\Gamma$.
On $\mathcal P(\Gamma)$ we consider the topology induced by the narrow convergence, namely $\eta^n\rightharpoonup\eta$ in $\mathcal P(\Gamma)$ if $\int\phi(\gamma)\eta^n(d\gamma)\to\int\phi(\gamma)\eta(d\gamma)$ for any bounded $\phi\in C^0(\Gamma)$.
We define the following closed subset of $\mathcal P(\Gamma)$ $$\mathcal M (\Gamma) = \bigg\{ \eta \in \mathcal P(\Gamma) : \int |\gamma(0)|^2 \eta(d\gamma)\leq 1,\, \int \|\dot \gamma\|_{L^2(-1,T)}^2 \eta(d\gamma) \leq 1 \bigg\}.$$ [Note that $\gamma(0)$ is defined because $\gamma \in W^{1,2}$ $\eta$-a.e.]{}
The set $\mathcal M(\Gamma)$ satisfies the following properties.
\[Lem:calM\_right\] The set $\mathcal M(\Gamma)$ is tight.
For each $n\in\N$, the set $$\Gamma(n):=\Big\{ \gamma : |\gamma(0)|^2\leq n, \|\dot \gamma\|^2_{L^2(-1,T)} \leq n \Big\}$$ is bounded and therefore compact in $(\Gamma,d_{\Gamma})$. [In particular the set of bounded measures supported on $\Gamma(n)$ is compact in the narrow topology]{}.
Now notice that $\forall\, \eta \in \mathcal M(\Gamma)$ by Chebyshev’s inequality $$\max\bigg\{\eta \bigg( \bigg\{ \gamma : |\gamma(0)|^2>n\bigg\}\bigg),\, \eta \bigg( \bigg\{ \gamma :\int_{-1}^T |\dot \gamma(t)|^2 dt > n \bigg\} \bigg)\bigg\} < \frac{1}{n},$$ so that $\mathcal M(\Gamma)$ is tight.
\[Cor:calM\_compact\_metric\] The space $\mathcal M(\Gamma)$ with the topology of $\mathcal P(\Gamma)$ is compact metrizable.
For the restrictions of the measures in $\mathcal M(\Gamma)$ to the compact sets $(\Gamma(n), d_{\Gamma})$, $n\in\N$, metrize the narrow convergence with the usual Lèvy Prokorhov distance $\hat d_n$. Then the distance metrizing the narrow convergence between two measures in $\mathcal M(\Gamma)$ is defined by $$d_{\mathcal M(\Gamma)}(\eta_1,\eta_2)=\sum_n2^{-n}\hat d_n(\eta_1\llcorner_{\Gamma(n)},\eta_2\llcorner_{\Gamma(n)}). \qedhere$$
Being concentrated on $W^{1,2}((-1,T),{\mathbb{R}}^d)$, the measures $\eta \in \mathcal M(\Gamma)$ satisfy the following property.
\[Lem:conti\_weak\_cont\] Let $\phi : W^{1,2}((-1,T),{\mathbb{R}}^d) \to {\mathbb{R}}$ is bounded and continuous w.r.t. the weak topology of $W^{1,2}((-1,T),{\mathbb{R}}^d)$, i.e. the $L^2$-topology on $\gamma$ and the weak topology on $\dot \gamma$. Then if $\eta_n \rightharpoonup \eta$ narrowly, then $$\int \phi(\gamma) \eta_n(d\gamma) \to \int \phi(\gamma) \eta(d\gamma).$$
Since $\Gamma(m)$ is compact in both topologies, it follows that the $L^2$-metric and the metrization of the weak topology of $W^{1,2}((-1,T),{\mathbb{R}}^d)$ are equivalent. Let $\tilde \phi_m(\gamma)$ be an $L^2$-continuous extension of $\phi \llcorner_{\Gamma(m)}$ to $L^2$, with the same bound as $\phi$. Hence $$\int \tilde \phi_m \eta_n \to \int \tilde \phi_m \eta,$$ and by tightness of $\mathcal M(\Gamma)$ we conclude.
Dissipative solutions {#sec:diss}
=====================
In this section we first give a definition of dissipative solutions to the system in the Lagrangian formulation, namely as a subset of $\mathcal M(\Gamma)$. Such solutions dissipate the total kinetic energy, but trajectories are allowed to intersect at a certain time without joining in the future. The so-called sticky particle (or adhesive) solutions constitute a subset of dissipative solutions whose trajectories, whenever intersecting at some time, must coincide for all subsequent times.
Definition of dissipative solution {#Ss:definit_sticky}
----------------------------------
For any $t\in(0,T)$, define the space of curves $$\Gamma_t := {L^{2}} \big( (t,T), {\mathbb{R}}^d \big).$$
On $\Gamma_t$ we put the [$L^2$-metric]{}.
Let $T_t:\Gamma\to \Gamma_t$ be the restriction map
$$T_t(\gamma)=\gamma\llcorner_{(t,T)}.$$
[This map is a contraction.]{} The map $T_t$ induces the following equivalence relation on $\Gamma$ $$\gamma\sim_t\gamma'\quad {\Longleftrightarrow} \quad T_t(\gamma)=T_t(\gamma').$$
and the corresponding disintegration of measures $\eta\in\mathcal M(\Gamma)$ $$\label{eq:disinte_eta_at_time_t}
\eta = \int \omega^t_{\gamma'} {T_t}_{\#}\eta(d\gamma'), \qquad {T_t}_{\#}\eta = (T_t)_\sharp \eta, \ \gamma'\in\Gamma_t,$$
where $\omega^t_{\gamma'}\in\mathcal M(\Gamma)$ satisfies $\omega^t_{\gamma'}(T_t^{-1}(\gamma'))=1$, namely it is strongly consistent according to the notation used in [@Fr].
Observe the following
\[Lem:dot\_gamma\_eq\_proje\] It holds for $\mathcal L^1 \times \eta$-a.e. $(t,\gamma)\in(-1,T)\times\Gamma$ $$\dot \gamma(t) = \int \dot \gamma'(t) \omega^t_{T_t(\gamma)}(d\gamma').$$
Let $t\in(-1,T)$ be such that $\dot\gamma(t)$ exists for $\eta$-a.e. $\gamma$. If $T_t(\gamma')=T_t(\gamma)$ and $\gamma'$ is differentiable in $t$, then $\dot\gamma(t)=\dot\gamma'(t)$. Thus removing a set of $\mathcal L^1\times\eta$-measure $0$, it follows that $\dot \gamma( t)$ is constant on equivalence classes, from which the above formula follows.
Moreover, one has the following
The map $$\eta \mapsto \eta_t = (T_t)_\sharp \eta$$ is continuous from $(\mathcal M(\Gamma), d_{\mathcal M(\Gamma)})$ to $(\mathcal M(\Gamma_t), d_{\mathcal M(\Gamma_t)})$.
The map $T_t$ is a [contraction]{}. Hence for any continuous function $\phi$ on $\Gamma_t$ the map $\phi\circ T_t$ is continuous on $\Gamma$ and the statement of the lemma follows from the definition of push-forward.
Recalling the definition of $W_0$ given in , define for all $t\in[0,T]$
$$\label{eq:sticky_particle_property}
V_t(\gamma)=\int W_0(\gamma') \omega^t_{T_t(\gamma)}(d\gamma').$$
[This function is defined for $\eta$-a.e. $(t,\gamma)$.]{}
Notice that in general it may happen that $V_0(\gamma)\neq W_0(\gamma)$, and therefore the velocity of the curves in $\Gamma$ we consider may have an initial jump at time $t=0$. Since, as we will see in Corollary \[cor:BV\], for the kind of solutions $\eta\in\mathcal M(\Gamma)$ we consider the map $t\mapsto V_t$ defined from a vector field $W_0$ will be right continuous in $L^2_{\eta}$, one could equivalently modify the curves $\gamma$ on $(-1,0)$ (and therefore $W_0$) defining $\dot\gamma_{\llcorner(-1,0]}=\lim _{t\searrow 0}V_t(\gamma)$.
We are now ready to give our definition of dissipative solution.
\[def:sticky\] We say that $\eta\in\mathcal M(\Gamma)$ is a *dissipative solution* of if it holds $$\label{eq:diss}
\dot \gamma(t) = V_t(\gamma) \qquad \mathcal L^1 \times \eta\text{-a.e. on $[0,T]\times \Gamma$.}$$
The above condition can be rewritten as follows: for every continuous bounded function $\phi(t,\gamma)$ on $[0,T]\times\Gamma$ it holds $$\int \bigg[ \int \phi(t,\gamma) \dot \gamma(t) dt \bigg] \eta(d\gamma) = \int \bigg[ \int \phi(t,\gamma) V_t(\gamma) \eta(d\gamma) \bigg] dt.$$
The fact that, for $t>s$, the map $T_t$ induces a coarser partition than $T_s$ (or, in other words, that it induces a descending in time filtration on $\Gamma$), can be expressed at the level of disintegrations of $\eta$ at different times in the following way.
For $s<t$, let $T_{s\to t}:\Gamma_s\to\Gamma_t$ be the restriction map such that $T_{s\to t}\circ T_s=T_t$. Then [by disintegrating again]{}
$${(T_s)}_{\#}\eta(d\gamma')=\int\omega^{s\to t}_{\gamma''}(d\gamma'){(T_t)}_{\#}\eta(d\gamma'')$$
and
$$\begin{aligned}
\eta(d\gamma)&=\int\omega^s_{\gamma'}(d\gamma){(T_s)}_{\#}\eta(d\gamma')\notag\\
&=\int\int\omega^s_{\gamma'}(d\gamma)\omega^{s\to t}_{\gamma''}(d\gamma')(T_{s\to t}\circ T_s)_{\#}\eta(d\gamma'')\notag\\
&=\int\int\omega^s_{\gamma'}(d\gamma)\omega^{s\to t}_{\gamma''}(d\gamma'){(T_t)}_{\#}\eta(d\gamma'')\notag\\
&=\int\omega^t_{\gamma''}(d\gamma){(T_t)}_{\#}\eta(d\gamma'').\end{aligned}$$
Therefore, by the uniqueness of the disintegration, [for $(T_t)_\sharp \eta$-a.e. $\gamma''$]{} $$\label{eq:rhoomega}
\omega^t_{\gamma''}(d\gamma)=\int\omega^s_{\gamma'}(d\gamma)\omega^{s\to t}_{\gamma''}(d\gamma').$$
The terminology used for this kind of solutions is consistent with the following
\[prop:diss\] Let $\Psi:{\mathbb{R}}^d\to{\mathbb{R}}$ be a convex function and let $\eta\in\mathcal M(\Gamma)$ be a dissipative solution. Then, the map $$\label{eq:psidiss}
t\mapsto\int\Psi(V_t(\gamma))\eta(d\gamma)$$ is nonincreasing on $(-1,T]$, where by convention we set $V_t=W_0$ if $t\in (-1,0)$. In particular, taking $\Psi=|\cdot|^2$ one has the dissipation balance $$\label{eq:squarediss}
\int|V_s(\gamma)-V_t(\gamma)|^2\eta(d\gamma)=\int|V_s(\gamma)|^2\eta(d\gamma)-\int|V_t(\gamma)|^2\eta(d\gamma), \quad\forall\,s\leq t, s,t\in(-1,T].$$
Recalling the disintegration formula and applying Jensen’s inequality one has that $$\begin{aligned}
\int\Psi(V_t(\gamma))\eta(d\gamma)&=\int\Psi\biggl(\int W_0(\gamma')\omega^t_{T_t(\gamma)}(d\gamma')\biggr){(T_t)}_{\#}\eta(d\gamma)\notag\\
&=\int\Psi\biggl(\int W_0(\gamma')\omega^s_{\gamma''}(d\gamma')\omega^{s\to t}_{T_t(\gamma)}(d\gamma'')\biggr){(T_t)}_{\#}\eta(d\gamma)\notag\\
&\leq \int\Psi(V_s(\gamma''))\omega^{s\to t}_{T_t(\gamma)}(d\gamma''){(T_t)}_{\#}\eta(d\gamma)\notag\\
&=\int\Psi(V_s(\gamma)){(T_s)}_{\#}\eta(d\gamma)\notag\\
&=\int\Psi(V_s(\gamma))\eta(d\gamma).
\end{aligned}$$
To prove that the dissipation balance holds, we observe that
$$\begin{aligned}
\int V_t(\gamma)\cdot V_s(\gamma)\eta(d\gamma)&=\int\int V_t(\gamma)\cdot V_s(\gamma)\omega^s_{\gamma'}(d\gamma){(T_s)}_{\#}\eta(d\gamma')\notag\\
&=\int V_t(\gamma')\cdot V_s(\gamma'){(T_s)}_{\#}\eta(d\gamma')\notag\\
&=\int\int V_t(\gamma')\cdot V_s(\gamma')\omega^{s\to t}_{\gamma''}(d\gamma'){(T_t)}_{\#}\eta(d\gamma'')\notag\\
&=\int V_t(\gamma'')\cdot V_t(\gamma''){(T_t)}_{\#}\eta(d\gamma'')\notag\\
&=\int|V_t(\gamma)|^2\eta(d\gamma).
\end{aligned}$$
Let us also define the total dissipation of a measure $\eta\in\mathcal M(\Gamma)$ as follows.
We define the total dissipation of a measure $\eta\in\mathcal M(\Gamma)$ as $$\label{eq:dissipation}
D(\eta)=T\|W_0\|_{L^2_\eta}^2-\int\int_0^T|V_t(\gamma)|^2dt\eta(d\gamma).$$
Notice that for any dissipative solution $\eta$, $D(\eta)\geq0$. Moreover, for any initial datum $(e_0,W_0)_{\#}\eta\in\mathcal P_2({\mathbb{R}}^d\times{\mathbb{R}}^d)$, being $e_0:\Gamma\to{\mathbb{R}}^d$ the evaluation map $e_0(\gamma)=\gamma(0)$, there exists always a dissipative solution concentrated on straight lines of constant velocity $W_0(\gamma)$, which has zero dissipation.
We now give a precise definition of the concept of sticky particle solution. For every $t\in[0,T]$, let $e_t:\Gamma\to{\mathbb{R}}^d$ be the evaluation map $e_t(\gamma)=\gamma(t)$.
We say that $\eta\in\mathcal M(\Gamma)$ is *sticky particle solution* of if $\eta$ is concentrated on a subset of $\Gamma$ on which, for all $t\in[0,T]$, the maps $T_t$ and $e_t$ induce the same equivalence relation.
Two particles which collide and stick or intersect without joining.
Compactness of the set of dissipative solution {#Ss:conve_sticky}
----------------------------------------------
The aim of this section is to show that the set of dissipative solutions is closed w.r.t. weak convergence. In order to pass to the limit in the relation we embed the dissipative solutions into a larger compact space of Young measures generated by the disintegrations of the measures $\eta\in\mathcal M(\Gamma)$ w.r.t. the restriction maps $T_t$.
For any $\eta\in\mathcal P(\Gamma)$, define the map $$F(\eta):[0,T]\times\Gamma\to\mathcal P(\Gamma), \quad F(\eta)(t,\gamma)=\omega^t_{T_t(\gamma)},$$ where $\{\omega^t_{\gamma'}\}_{\gamma'\in\Gamma_t}$ is the disintegration of $\eta$ w.r.t. the restriction map $T_t$. [This map is defined $\mathcal L^1 \times \eta$-a.e..]{} Let then $$\label{eq:mueta}
\mu(\eta)=\big(\Id_{[0,T]\times\Gamma}\times F(\eta)\big)_{\#}\big(\mathcal L^1\times \eta\big).$$ By definition, $$\mu(\eta)\in \mathcal P([0,T]\times\Gamma\times\mathcal P(\Gamma)).$$
With the disintegration we can rewrite $\mu(\eta)$ as $$\int \phi(t,\gamma,\omega) \mu(\eta)(dtd\gamma d\omega) = \int \phi(t,\gamma,\omega^t_{T_t(\gamma')}) \eta(d\gamma) dt.$$ for every continuous function $\phi$.
The set $$\biggl\{\mu(\eta)\in\mathcal P([0,T]\times\Gamma\times\mathcal P(\Gamma)):\,\eta\in\mathcal M(\Gamma) \biggr\}$$ is tight.
Define the set $$\mathcal M_n(\Gamma) = \bigg\{ \eta\in\mathcal P(\Gamma):\,\int|\gamma(0)|^2\eta(d\gamma)\leq n,\,\int\|\dot\gamma\|_2^2\eta(d\gamma)\leq n \bigg\}.$$ Then the set $$K(n)=[0,T]\times\Gamma(n)\times\mathcal M_n(\Gamma)$$ is compact in $[0,T]\times\Gamma\times\mathcal P(\Gamma)$ [(just adapt the proofs of Lemma \[Lem:calM\_right\] and Corollary \[Cor:calM\_compact\_metric\] to $\mathcal M_n(\Gamma)$)]{}. Let now $\eta\in\mathcal M(\Gamma)$ and $\mu(\eta)$ defined as in .
On the one hand, $$\mu(\eta)([0,T]\times\Gamma(n)^c\times\mathcal P(\Gamma))=\bigl(\mathcal L^1\times \eta\bigr)([0,T]\times\Gamma(n)^c)\leq\frac{T}{n}.$$
On the other hand, $$\begin{aligned}
\mu(\eta)(&[0,T]\times\Gamma(n)\times\mathcal M_n(\Gamma)^c)\leq\int_0^T \eta \big( \{\gamma\in\Gamma:\,\omega^t_{T_t(\gamma)}\in\mathcal M_n(\Gamma)^c\} \big) dt \notag\\
&=\int_0^T\eta\bigg(\bigg\{\gamma\in\Gamma:\int|\gamma'(0)|^2\omega^t_{T_t(\gamma)}(d\gamma')>n \text{ or } \int\|\dot{\gamma'}\|_2^2 \omega^t_{T_t(\gamma)}(d\gamma')>n\biggr\}\biggr) dt.\end{aligned}$$ Since $\eta\in\mathcal M(\Gamma){= \mathcal M_1(\Gamma)}$, $$1\geq\int|\gamma(0)|^2\eta(d\gamma)=\int\int|\gamma'(0)|^2\omega^t_{T_t(\gamma)}(d\gamma'){(T_t)}_{\#}\eta(dT_t(\gamma)),$$ therefore $$\eta\biggl(\biggl\{\gamma\in\Gamma:\,\int|\gamma'(0)|^2\omega^t_{T_t(\gamma)}(d\gamma')>n\biggr\}\biggr)<\frac1n.$$ Similarly, one proves that $$\eta\biggl(\biggl\{\gamma\in\Gamma:\,\int\|\dot{\gamma'}\|_2^2\omega^t_{T_t(\gamma)}(d\gamma')>n\biggr\}\biggr)\leq\frac{1}{n}$$ and the lemma is proved.
The closure of the set $\biggl\{\mu(\eta)\in\mathcal P([0,T]\times\Gamma\times\mathcal P(\Gamma)):\,\eta\in\mathcal M(\Gamma) \biggr\}$ is [contained in]{} the compact set
$$\begin{split}
\mathcal D=\biggl\{\mu\in\mathcal P([0,T]\times\Gamma\times\mathcal P(\Gamma))&:\,\mathit {p_{[0,T]\times\Gamma}}_{\#}\mu=\mathcal L^1\times\eta,\,\eta\in\mathcal M(\Gamma), \\
& \ \int\int \frac{|\gamma'(0)|^2}{T} \omega(d\gamma')\mu(dtd\gamma d\omega) \leq 1, \int\int \|\dot{\gamma'}\|_2^2\omega(d\gamma')\mu(dtd\gamma d\omega) \leq 1 \biggr\}.
\end{split}$$
Indeed, the fact that each $\mu \in \mathcal D$ is tight follows from the same proof of the previous lemma. The fact that every narrow limit $\mathcal D \ni \mu_n \to \mu$ belongs to $\mathcal D$ is a consequence of the observation that $$\omega \mapsto \int \|\dot{\gamma'}\|_2^2 \omega(d\gamma')$$ is l.s.c. w.r.t. the narrow convergence, being $\gamma' \to \|\dot{\gamma'}\|_2^2$ convex l.s.c..
Our aim is now to prove that if $\eta^n\rightharpoonup\eta$ with $\eta^n$ dissipative solutions, then also $\eta$ is dissipative. We will do it looking at the weak limit of the $\mu(\eta^n)$ in $\mathcal D$. In particular, we will use a dense family of smooth test functions of the form
$$\phi_0(t)\tilde{\phi}(T_{\bar t}(\gamma))\phi(\gamma)\psi(\omega),$$ where $\bar t>t$ for all $t\in\supp\phi_0$.
Any weak limit $\mu$ of a weakly convergent sequence $\{\mu(\eta^n):\eta^n\in\mathcal M(\Gamma),\,n\in\N\}$ satisfies $$\label{eq:test1}
\int\phi_0(t)\tilde{\phi}(T_{\bar t}(\gamma))\phi(\gamma)\psi(\omega)\mu(dtd\gamma d\omega)=\int\phi_0(t)\tilde{\phi}(T_{\bar t}(\gamma))\psi(\omega)\biggl(\int\phi(\gamma')\omega(d\gamma')\biggr)\mu(dtd\gamma d\omega).$$
One has that $$\begin{aligned}
\int\phi_0(t)\tilde{\phi}(T_{\bar t}(\gamma))\phi(\gamma)\psi(\omega)\mu(\eta)(dtd\gamma d\omega)&=\int\int\phi_0(t)\tilde{\phi}(T_{\bar t}(\gamma))\phi(\gamma)\psi(\omega^t_{T_t(\gamma)})dt\eta(d\gamma)\notag\\
&=\int\phi_0(t)\biggl(\int\int\tilde{\phi}(T_{\bar t}(\gamma))\phi(\gamma)\psi(\omega^t_{T_t(\gamma)})\omega^t_{\gamma'}(d\gamma){(T_t)}_{\#}\eta(\d\gamma')\biggr)dt\notag\\
&=\int\phi_0(t)\biggl(\int \tilde \phi(T_{t,\bar t}(\gamma'))\biggl(\int\phi(\gamma)\psi(\omega^t_{\gamma'})\omega^t_{\gamma'}(d\gamma)\biggr){(T_t)}_{\#}\eta(d\gamma')\biggr)dt\notag\\
&=\int\phi_0(t)\int\tilde{\phi}(T_{\bar t}(\gamma))\int\phi(\gamma')\psi(\omega^t_{T_t(\gamma)})\omega^t_{T_t(\gamma)}(d\gamma')\eta(d\gamma)dt\notag\\
&=\int\phi_0(t)\tilde{\phi}(T_{\bar t}(\gamma))\biggl(\int\phi(\gamma')\omega(d\gamma')\biggr)\psi(\omega)\mu(\eta)(dtd\gamma d\omega).
\end{aligned}$$ Since $$\omega \mapsto \int \phi(\gamma') \omega(d\gamma')$$ is continuous for $\phi$ continuous, then the set of $\mu$ with the property above is closed, and thus is contains the closure of $\{\mu(\eta),\eta \in \mathcal M(\Gamma)\}$.
Now, using the fact that $\phi_0$ and $\bar t$ are arbitrary [under the condition $\supp \phi_0 \subset (-\infty,\bar t)$]{}, from it follows that, for $\mathcal L^1$-a.e. $t \in [0,T]$
$$\label{eq:test2}
\int\tilde{\phi}(T_t(\gamma))\phi(\gamma)\biggl(\int\psi(\omega)\mu_{t,\gamma}(d\omega)\biggr)\eta(d\gamma)=\int\tilde{\phi}(T_t(\gamma))\biggl(\int\psi(\omega)\biggl(\int\phi(\gamma')\omega(d\gamma')\biggr)\mu_{t,\gamma}(d\omega)\biggr)\eta(d\gamma),$$
where $$\mu(dtd\gamma d\omega)=\int\mu_{t,\gamma}(d\omega)dt\eta(d\gamma).$$ Indeed by letting $\phi_0$ vary, one obtains for [$\mathcal L^1$-a.e. $t$ and]{} all $\bar t > t$ $$\int \tilde{\phi}(T_{\bar t}(\gamma)) \phi(\gamma) \biggl( \int \psi(\omega) \mu_{t,\gamma}(d\omega) \biggr) \eta(d\gamma) = \int \tilde{\phi}(T_{\bar t}(\gamma)) \biggl( \int \psi(\omega) \biggl( \int \phi(\gamma') \omega(d\gamma') \biggr) \mu_{t,\gamma}(d\omega) \biggr) \eta(d\gamma).$$ If $\tilde \phi : \Gamma_t \to {\mathbb{R}}$ is a continuous function, one can test the above equation with $$\tilde \phi'(T_{\bar t}(\gamma)) = \tilde \phi \big( \gamma(\bar t) \chi_{[t,\bar t)} + T_{\bar t}(\gamma)\chi_{[\bar t,T]} \big).$$ The above function $\tilde \phi' : \Gamma_{\bar t} \to {\mathbb{R}}$ is continuous because of the embedding $W^{1,2}([0,T])$ in $C^0$, and as $\bar t \searrow t$ the function $\tilde{\phi}'\circ T_{\bar t}$ converges pointwise to $\tilde \phi\circ T_t$. By the bounded convergence theorem we conclude that holds.
Replacing $\eta$ in with its disintegration w.r.t. $T_t$ and thanks to the fact that $\tilde{\phi}$ is arbitrary, one obtains that, for ${T_t}_{\#}\eta$-a.e. $\gamma'\in\Gamma_t$
$$\begin{aligned}
\int\phi(\gamma)\biggl(\int\psi(\omega)\mu_{t,\gamma}(d\omega)\biggr)\omega^t_{\gamma'}(d\gamma)=\int\biggl(\int\psi(\omega)\biggl(\int\phi(\gamma')\omega(d\gamma')\biggr)\mu_{t,\gamma}(d\omega)\biggr)\omega^t_{\gamma'}(d\gamma).\end{aligned}$$
Taking now $\psi=1$, one gets $$\int \phi(\gamma)\omega^t_{\gamma'}(d\gamma)=\int\int\biggl(\int\phi(\gamma')\omega(d\gamma')\biggr)\mu_{t,\gamma}(d\omega)\omega^t_{\gamma'}(d\gamma)$$ and by the arbitrariness of $\phi$ $$\label{eq:test3}
\omega^t_{\gamma'}(d\gamma)=\int\omega(d\gamma)\mu_{t,\gamma''}(d\omega)\omega^t_{\gamma'}(d\gamma'').$$
We will use condition to show the following proposition.
Let $\{\eta^n\}\subset\mathcal M(\Gamma)$ be a sequence of dissipative solutions such that $\eta^n\rightharpoonup\eta\in\mathcal M(\Gamma)$ as $n\to\infty$. Then $\eta$ is a dissipative solution.
Let $\phi:[0,T]\times\Gamma\to{\mathbb{R}}$ be a continuous function.
On the one hand, as $n\to\infty$ $$\int\phi(t,\gamma)\dot{\gamma}(t)\eta^n(d\gamma)dt\to\int\phi(t,\gamma)\dot{\gamma}(t)\eta(d\gamma)dt.$$
Indeed, $t \mapsto \int \phi(t,\gamma) \dot \gamma(t)dt$ is continuous in $\Gamma$ [w.r.t. the weak convergence, and then one applies Lemma \[Lem:conti\_weak\_cont\].]{}
On the other hand,
$$\begin{aligned}
\int\phi(t,\gamma)\dot{\gamma}(t)\eta^n(d\gamma)\dt&=\int\int\phi(t,\gamma)\int W_0(\gamma')\omega^{n,t}_{T_t(\gamma)}(d\gamma')\eta^n(d\gamma)dt\notag\\
&=\int\int\phi(t,\gamma)\int W_0(\gamma')\omega(d\gamma')\mu(\eta^n)_{t,\gamma}(d\omega)\eta^n(d\gamma)dt\notag\\
&\to\int\int\phi(t,\gamma)\int W_0(\gamma')\omega(d\gamma')\mu_{t,\gamma}(d\omega)\eta(d\gamma)dt \quad\text{ as $n\to\infty$}.
\end{aligned}$$
Hence, $$\label{eq:test4}
\dot{\gamma}(t)=\int\int W_0(\gamma')\omega(d\gamma')\mu_{t,\gamma}(d\omega)\quad\mathcal L^1\times\eta{\text{-a.e. $(t,\gamma)$}}.$$ Integrating w.r.t. $\omega^t_{\gamma''}$ and then using one gets $$\begin{aligned}
\int \dot{\gamma}(t)\omega^t_{\gamma''}(d\gamma)&=\int\int\int W_0(\gamma')\omega(d\gamma')\mu_{t,\gamma}(d\omega)\omega^t_{\gamma''}(d\gamma)\notag\\
&=\int W_0(\gamma)\omega^t_{\gamma''}(d\gamma).
\end{aligned}$$ Finally, since $\dot{\gamma}(t)$ is constant on $T_t^{-1}(\gamma'')$, we deduce that $$\dot{\gamma}(t)=\int W_0(\gamma')\omega^t_{T_t(\gamma)}(d\gamma') \quad\mathcal L^1\times\eta\text{-a.e.}\ (t,\gamma),$$ namely that $\eta$ is a dissipative solution.
We have therefore proved the following
\[thm:sticky\] The set of dissipative solutions in $\mathcal M(\Gamma)$ is compact.
Notice instead that the set of sticky particle solutions in $\mathcal M{(\Gamma)}$ is not closed. Take for example the free flow of two particles which do not interact up to time $T$ and change their directions so that in the limit their trajectories intersect for some time $t\in(0,T)$. Or, even worse, see the Example 4 in [@BN], where the only solution is the trivial one [(i.e. the free flow)]{} with zero dissipation (not sticky), as limit of sticky particle solutions for the initial data obtained removing the particles in a smaller and smaller neighbourhood of the origin.
Therefore [it is justified]{} the necessity to consider the notion of dissipative solution for general initial data.
Properties of dissipative solutions
-----------------------------------
\[cor:BV\] Let $\eta\in\mathcal M(\Gamma)$ be a dissipative solution. Then, the map $t\mapsto V_t$ belongs to $BV^{1/2}([0,T];L^2_\eta(\Gamma;{\mathbb{R}}^d))$ and it is right continuous.
Let $0=t_0\leq t_1\leq\dots\leq t_N=T$ be any finite partition of $[0,T]$. Then, by , $$\sum_{i=0}^{N-1}\|V_{t_{i+1}}-V_{t_i}\|_{L^2_\eta}^2=\|W_0\|_{L^2_\eta}^2-\|V_T\|_{L^2_\eta}^2\leq\|W_0\|_{L^2_\eta}^2.$$
As for the right continuity, the $BV^{1/2}([0,T];L^2_\eta(\Gamma;{\mathbb{R}}^d))$-property implies that for all $\bar t$ there exists the limit $$\lim_{t \searrow \bar t} V_t = \hat V_{\bar t},$$ and that it coincides with $V_{\bar t}$ for $\mathcal L^1$-a.e. $\bar t \in [0,T]$. Indeed, since all $\sigma$-algebras generated by $T_{t}$, $t \geq \bar t$ are contained in the one generated by $T_{\bar t}$, it follows that $\hat V_{\bar t}$ is measurable w.r.t. the $\sigma$-algebra of generated by $T_{\bar t}$. Now for every $t > \bar t$ and for every continuous function $\phi : \Gamma_{\bar t} \to {\mathbb{R}}$ the function $$\tilde \phi(T_{ t}(\gamma)) = \phi \big( \gamma(t) \chi_{[\bar t,t)} + T_{t}(\gamma)\chi_{[t,T]} \big)$$ is continuous and for all $\bar t \leq s \leq t$ $$\int V_s(\gamma) \tilde \phi(T_{ t}(\gamma)) \eta(d\gamma) = \int V_t(\gamma) \tilde \phi(T_{t}(\gamma)) \eta(d\gamma),$$ because $\tilde \phi$ depends only on $T_t(\gamma)$. In particular it is constant in $[\bar t,t]$. Hence we conclude that by taking ${s} \searrow \bar t$ $$\int \hat V_{\bar t}(\gamma) \tilde \phi(T_{t}(\gamma)) \eta(d\gamma) = \int V_t(\gamma) \tilde \phi(T_{\bar t}(\gamma)) \eta(d\gamma) = \int W_0(\gamma) \tilde \phi(T_{ t}(\gamma)) \eta(d\gamma),$$ and since as $t \searrow \bar t$ the function $\tilde \phi\circ T_{t}$ converges pointwise to $\phi\circ T_{\bar t}$, it follows that $$\int \hat V_{\bar t}(\gamma) \phi(T_{\bar t}(\gamma)) \eta(d\gamma) = \int W_0(\gamma) \phi(T_{\bar t}(\gamma)) \eta(d\gamma),$$ which implies $\hat V_{\bar t} = V_{\bar t}$ being $\phi$ arbitrary.
Approximations of dissipative solutions {#sec:approx}
=======================================
The goal of this section is to prove Theorem \[thm:stickydense\], namely that dissipative solutions can be approximated by finite sticky particle solutions, which are sticky particle solutions concentrated on a finite number of trajectories.
We will obtain this result in four steps. In the first step we will approximate a dissipative solution by a dissipative solution whose velocity field is finitely piecewise constant in time ([called]{} discrete in time dissipative solution). Then we will show that discrete in time dissipative solutions can be approximated by dissipative countable particle solutions, namely dissipative solutions concentrated on a countable number of disjoint trajectories. Then we approximate dissipative countable particle solutions with dissipative finite particle solutions. Finally, we approximate dissipative finite particle solutions with finite sticky particle solutions. In each of the first three steps we will apply a general procedure which is resumed in Lemma \[lemma:discrete\_sticky\]. We start with the following definitions.
A dissipative solution $\eta\in\mathcal M(\Gamma)$ is discrete in time if there exists a finite partition $0=t_0<t_1<\dots<t_N=T$ such that $V_t=V_{t_i}$ for all $t\in[t_i,t_{i+1})$.
\[def:countsticky\] A dissipative countable particle solution is a discrete in time dissipative solution $\eta$ with the property that there exist[s]{} a countable number of trajectories $\{\gamma_n\}_{n\in\N}\subset\Gamma$ such that $\dot\gamma_n(t)=\dot{\gamma}_n(t_i)$ if $t\in[t_i,t_{i+1})$, $\inf_{n\neq m}|\gamma_n(T)-\gamma_m(T)|>0$ and $\eta(\cup_n\{\gamma_n\})=1$.
\[def:finitesticky\] A dissipative finite particle solution is a discrete in time dissipative solution $\eta$ with the property that there exist[s]{} a finite number of trajectories $\{\gamma_n\}_{n=1,\dots,N}\subset\Gamma$ such that $\dot\gamma_n(t)=\dot{\gamma}_n(t_i)$ if $t\in[t_i,t_{i+1})$ and $\eta(\cup_{n=1}^N\{\gamma_n\})=1$.
In the next lemma we exhibit a general procedure which allows to find, given a measure $\eta\in\mathcal M(\Gamma)$ (not necessarily dissipative) and a finite number of descending in time equivalence relations on $\Gamma$, a discrete in time dissipative solution.
Such a procedure will be used several times in the following subsections to be able to approximate general dissipative solutions with dissipative finite particle solutions.
\[lemma:discrete\_sticky\] Let $\eta\in\mathcal M(\Gamma)$, let $0=t_0<t_1<\dots<t_N=T$ be a finite partition of $[0,T]$, let $\{E_{t_i}\}_{i=0,\dots,N}$ be a family of Borel equivalence relations on $\Gamma$ with partition maps $G_{t_i}:\Gamma\to [0,1]$ such that $$\label{eq:gti}
G_{t_i}(\gamma)=G_{t_i}(\gamma')\quad\Rightarrow\quad G_{t_j}(\gamma)=G_{t_j}(\gamma')\quad\forall\,j>i,\,\gamma,\gamma'\in\Gamma,$$ or equivalently $E_{t_i}\subset E_{t_j}$ [as graphs for $t_i\leq t_j$]{}, and let $\{y_{T,\alpha}\}_{\alpha \in E_{T}}\subset{\mathbb{R}}^d$. Let $\eta(d\gamma)=\int\omega^{t_i}_{y}(d\gamma){G_{t_i}}_{\#}\eta(y)$ be the disintegrations w.r.t. the partitions $\{E_{t_i}\}$, $V_{t_i}(\gamma)=\int W_0(\gamma')\omega^{t_i}_{G_{t_i}(\gamma)}(d\gamma')$ and $\bar V_t=V_{t_i}$ if $t\in[t_i,t_{i+1})$, $\bar V_t=\bar V_0$ if $t\in (-1,0)$.
Define $$\tilde F:\Gamma\to\Gamma, \quad \tilde F(\gamma)=y_{T,G_T(\gamma)}-\int_t^T\bar V_s(\gamma)ds.$$
Then, the measure $\tilde{\eta}=\tilde F_{\#}\eta$ is sticky.
\[rem:v0\] Observe that $W_0(\tilde F(\gamma))=\int W_0(\gamma')\omega^0_{G_0(\gamma)}(d\gamma')$ for all $\gamma\in\Gamma$. Indeed, $$W_0(\tilde F(\gamma))=\tilde F(\gamma)(0)-\tilde F(\gamma)(-1)=\int_{-1}^0\bar V_s(\gamma)ds=\bar V_0(\gamma)=\int W_0(\gamma')\omega^0_{G_0(\gamma)}(d\gamma').$$
As a preliminary step, notice that, by construction, the partition induced on $\Gamma$ by $G_{t_i}$ is a refinement of the one induced by $T_{t_i}\circ \tilde F$. Indeed, if for some $\gamma'$ $$\int W_0(\gamma'')\omega^{t_j}_{G_{t_j}(\gamma')}(d\gamma'')=\int W_0(\gamma'')\omega^{t_j}_{G_{t_j}(\gamma)}(d\gamma''),\quad\forall\,j\geq i,$$ then $T_{t_i}\circ \tilde F(\gamma)=T_{t_i}\circ \tilde F(\gamma')$ even if possibly $G_{t_i}(\gamma)\neq G_{t_i}(\gamma')$.
Therefore there exists a Borel map, say $F {: [0,1] \to \Gamma_{t_i}}$, such that $T_{t_i}\circ \tilde F=F\circ G_{t_i}$ and one has the disintegrations $$\begin{aligned}
\eta(d\gamma)&=\int\omega^{t_i}_{y}(d\gamma){G_{t_i}}_{\#}\eta(dy)\notag\\
&=\int\omega^{t_i}_{y}(d\gamma)\nu^{t_i}_{\tilde{\gamma}}(dy){T_{t_i}}_{\#}\tilde F_{\#}\eta(d\tilde{\gamma}).
\end{aligned}$$
In particular, applying $\tilde F_{\#}$ to both sides $$\begin{aligned}
\tilde F_{\#}\eta(d\tilde\gamma)
&=\int\tilde F_{\#}\omega^{t_i}_{y}(d\tilde\gamma)\nu^{t_i}_{\tilde{\gamma}'}(dy){T_{t_i}}_{\#}\tilde F_{\#}\eta(d\tilde{\gamma}').
\end{aligned}$$
On the other hand, let $\tilde{\omega}^{t_i}_{\tilde \gamma}$ be the disintegration of $\tilde{\eta}=\tilde F_{\#}\eta$ w.r.t. the restriction map $T_{t_i}$. Then, $$\tilde F_{\#}\eta(d\tilde\gamma)
=\int\tilde\omega^{t_i}_{\tilde{\gamma}'}(d\tilde\gamma){T_{t_i}}_{\#}\tilde F_{\#}\eta(d\tilde{\gamma}').$$
Hence,
$$\label{eq:rhopush}
\tilde \omega^{t_i}_{T_{t_i}\circ\tilde F(\gamma)}(d\tilde{\gamma}')=\int\tilde F_{\#}\omega^{t_i}_{y}(d\tilde\gamma')\nu^{t_i}_{T_{t_i}\circ \tilde F(\gamma)}(dy).$$
Let us now check the dissipation condition for $\tilde \eta$ first when $t=t_i$ for some $i\in\{0,\dots,N\}$, considering here the right derivative because of Corollary \[cor:BV\]. Recall that the vector field associated to $\tilde\eta$ is given by $$\tilde V_t(\tilde{\gamma})=\int W_0(\tilde{\gamma}')\tilde{\omega}^{t_i}_{T_{t_i}(\tilde{\gamma})}(d\tilde{\gamma}'), \qquad {t \in [t_i,t_{i+1})}.$$
One has that
$$\begin{aligned}
{\dot{(\widetilde{F}(\gamma))}}(t_i)&=\bar V_{t_i}(\gamma)=\int W_0({\gamma}')\omega^{t_i}_{ G_{t_i}(\gamma)}(d\gamma')\notag\\
&=\int W_{0}(\gamma')\omega^{t_i}_{y}(d\gamma')\nu^{t_i}_{T_{t_i}\circ \tilde F(\gamma)}(dy),\end{aligned}$$
where in the last equality we have used the fact $\int W_{0}(\gamma')\omega^{t_i}_{(\cdot)}(d\gamma')$ is constant on $F^{-1}(T_{t_i}\circ\tilde F(\gamma))$.
Finally, by Remark \[rem:v0\] and $$\begin{aligned}
{\dot{(\widetilde{F}({\gamma}))}}(t_i)&=\int W_{0}(\gamma')\omega^{t_i}_{y}(d\gamma')\nu^{t_i}_{T_{t_i}\circ \tilde F(\gamma)}(dy)\notag\\
&=\int W_0(\tilde{\gamma}')\tilde F_{\#}\omega^{t_i}_{y}(d\gamma')\nu^{t_i}_{T_{t_i}\circ \tilde F(\gamma)}(dy)\notag\\
&=\int W_0(\tilde{\gamma}')\tilde \omega^{t_i}_{T_{t_i}\circ\tilde F(\gamma)}(d\tilde{\gamma}')\notag\\
&=\tilde V_{t_i}(\tilde F(\gamma)).\end{aligned}$$
Let us now assume $t\in[t_i,t_{i+1})$. Then, $${\dot{(\widetilde F(\gamma))}}(t)=\bar V_t(\gamma)=\bar V_{t_i}(\gamma)=\tilde V_{t_i}(\tilde F(\gamma)).$$
Let us now prove that for $t\in[t_i,t_{i+1})$ $\tilde V_t(\tilde F(\gamma))=\tilde V_{t_i}(\tilde F(\gamma))$.
This follows from the following observation $$\label{eq:tti}
T_t\circ\tilde F(\gamma)=T_t\circ\tilde F(\gamma')\quad\Rightarrow\quad T_{t_i}\circ \tilde F(\gamma)=T_{t_i}\circ \tilde F(\gamma'),$$ which can be easily checked from the definition of $\tilde F$.
From one has that $\tilde \omega^t_{T_t\circ\tilde F(\gamma)}(d\tilde{\gamma})=\tilde \omega^{t_i}_{T_{t_i}\circ\tilde F(\gamma)}(d\tilde{\gamma})$, hence $\tilde V_t(\tilde F(\gamma))=\tilde V_{t_i}(\tilde F(\gamma))$.
Discrete in time approximation
------------------------------
We now apply the general construction of Lemma \[lemma:discrete\_sticky\] in order to prove that, given a dissipative solution $\eta$, there exists a discrete in time dissipative solution whose vector field is close to the vector field of $\eta$ at any time in the $L^2_{\eta}$ topology.
\[prop:discretesticky\] Let $\eta\in\mathcal M(\Gamma)$ be a sticky particle solution and $\eps>0$. Then there exists $\tilde \eta^\eps=\tilde F^\eps_{\#}\eta\in\mathcal M(\Gamma)$ discrete in time sticky particle solution such that $$\label{eq:etaepsprop}
W_0(\tilde F^\eps(\gamma))=W_0(\gamma),\quad\|\tilde V^\eps_t\circ \tilde F^\eps-V_t\|_{L^2_\eta}^2\leq \eps^2, \quad\int|\tilde F^\eps(\gamma)(t)-\gamma(t)|^2\eta(d\gamma)\leq {\eps^2} T.$$
In particular, as $\eps\to0$, the measures $\eta^\eps$ converge in $\mathcal M(\Gamma)$ to $\eta$.
By Corollary \[cor:BV\], we [k]{}now that the velocity field [$t\mapsto V_t$]{} of $\eta$ belongs to $BV^{1/2}([0,T];L^2_\eta(\Gamma;{\mathbb{R}}^d))$ and it is right continuous. Therefore, given $\eps>0$, there exists a finite partition $0=t_0<t_1<\dots<t_N=T$ such that either $$\begin{aligned}
\label{eq:part1}
\|V_{t_i}\|_{L^2_\eta}^2-\|V_{t_{i+1}}\|_{L^2_\eta}^2<\eps^2
\end{aligned}$$ or $$\begin{aligned}
\label{eq:part2}
\|V_{t_i}\|_{L^2_\eta}^2-\lim _{s\nearrow t_{i+1}}\|V_s\|_{L^2_\eta}^2<\eps^2,\qquad\lim_{s\nearrow t_{i+1}}\|V_s\|_{L^2_\eta}^2-\|V_{t_{i+1}}\|_{L^2_\eta}^2\geq\eps^2/2.
\end{aligned}$$ Moreover, there are at most $2\|W_0\|_{L^2_\eta}^2/\eps^2$ points where the last situation occurs.
We apply now Lemma \[lemma:discrete\_sticky\] to $\eta$, $0=t_0<t_1<\dots<t_N=T$, $G_{t_i}=T_{t_i}$, $y_{T,G_T(\gamma)}=\gamma(T)$ and find $\tilde \eta^\eps=\tilde F^\eps_{\#}\eta$ discrete in time dissipative solution.
In particular, since $G_{t_i}=T_{t_i}$, then the first property in holds.
Let us now check the [remaining]{} properties in . Let $t\in[t_i,t_{i+1})$: by ,
$$\begin{aligned}
\|\tilde V^\eps_t\circ \tilde F^\eps-V_t\|_{L^2_\eta}^2=\|V_{t_i}-V_t\|_{L^2_\eta}^2=\|V_{t_i}\|_{L^2_\eta}^2-\|V_t\|_{L^2_\eta}^2<\eps^2.
\end{aligned}$$
Moreover, using and and Jensen inequality, $$\begin{aligned}
\int|\tilde F^\eps(\gamma)(t)-\gamma(t)|^2\eta(d\gamma)\leq \int(T-t) {\biggl(} \int_t^T|\bar V_s(\gamma)-V_s(\gamma)|^2ds {\biggr)}\eta(d\gamma)\leq\eps^2T. \end{aligned}$$
Dissipative countable particle approximation
--------------------------------------------
Given a discrete in time dissipative solution, we now construct a discrete in time dissipative approximation with the additional property of being concentrated on a countable set of trajectories.
\[prop:countsticky\] Let $\eta$ be a discrete in time dissipative solution and $\delta>0$. Then there exists $\hat{\eta}^\delta=\hat F^\delta_{\#}\eta$ discrete in time dissipative solution such that
$$\label{eq:etadeltaprop}
|V_t(\hat F^\delta(\gamma))-V_t(\gamma)|\leq\delta, \quad|\hat F^\delta(\gamma)(t)-\gamma(t)|\leq \delta (1+T).$$
Moreover, $\hat{\eta}^\delta$ is a dissipative countable particle solution in the sense of Definition \[def:countsticky\].
Let $\{Q_k\}_{k\in\N}$ be a countable partition of ${\mathbb{R}}^d$ into cubes of side length $\delta$, $Q_k=x_k+[-\delta/2,\delta/2)^d$. If $0=t_0<t_1<\dots<t_N=T$ is the partition associated to $\eta$, define $\forall\,i=0,\dots,N$ a partition map $G_{t_i}$ on $\Gamma$ defining the following equivalence relation[:]{} $$G_{t_i}(\gamma)=G_{t_i}(\gamma')\quad\Leftrightarrow\quad \forall\,j\geq i,\,\exists\,k_j: V_{t_j}(\gamma)\in Q_{k_j},\,V_{t_j}(\gamma')\in Q_{k_j}\text{ and }\exists \bar k_T: \gamma(T)\in Q_{\bar k_T}, \gamma'(T)\in Q_{\bar k_T}.$$ In particular, the maps $G_{t_i}$ satisfy condition . Let moreover $y_{T,G_T(\gamma)}=x_{\bar k_T}$ if $\gamma(T)\in Q_{\bar k_T}$, observing that $\inf_{n\neq m}|\gamma_n(T)-\gamma_m(T)|\geq\delta>0$. Denote by $\bar V^\delta_t$ the discrete in time vector field associated with such partitions.
Apply now Lemma \[lemma:discrete\_sticky\] finding a discrete in time dissipative solution $\hat{\eta}^\delta=\hat F^\delta{\#}\eta$.
By construction, $\hat{\eta}^\delta$ is a dissipative countable particle solution and it satisfies .
Dissipative finite particle approximation
-----------------------------------------
Now we want to approximate dissipative countable particle solutions with dissipative finite particle solutions.
\[prop:finitesticky\] Let $\eta\in\mathcal M(\Gamma)$ be a dissipative countable particle solution. Then, for every $\sigma>0$, there exists $\eta^\sigma$ dissipative finite particle solution with the property that $\eta^\sigma\rightharpoonup\eta$ as $\sigma\to0$.
Since $\eta\in\mathcal M(\Gamma)$ is a dissipative solution, $$\int|\gamma(T)|^2\eta(d\gamma)\leq\int|\gamma(0)|^2\eta(d\gamma)+T^2\int|W_0(\gamma)|^2\eta(d\gamma)\leq 1+T^2$$ Hence, for any $\Lambda>0$ $$\eta(\{\gamma:\,|\gamma(T)|>\Lambda\})\leq\frac{1+T^2}{\Lambda^2}.$$
As a first step, take then $\bar{\eta}^{\sigma}=\eta_{\llcorner\{\gamma:\,|\gamma(T)|\leq\Lambda\}}/\eta(\{\gamma:|\gamma(T)|\leq\Lambda\})$ for $\Lambda$ large enough so that $\|\bar{\eta}^\sigma-\eta\|_{\mathcal M(\Gamma)}\leq\sigma/3$.
Then we want to apply Lemma \[lemma:discrete\_sticky\] to $\bar{\eta}^\sigma$ with the partition given by the maps $T_{t_i}$, the finite end points $y_{T, T_T(\gamma)}=\gamma(T)$ and vector field $\bar V^{\Lambda}_t$ defined as usual starting from a modification $W_0^\Lambda$ of $W_0$ defined as follows[:]{} $$W_0^\Lambda(\gamma_n)=W_0(\gamma_n) \text{ if $|W_0(\gamma_n)|\leq\Lambda$ },\quad W_0^\Lambda(\gamma_n)=\frac{W_0(\gamma_n)}{|W_0(\gamma_n)|}\Lambda\text{ if $|W_0(\gamma_n)|>\Lambda$ }.$$
In this way one obtains a countable sticky particle solution $\hat{\eta}^\sigma$ which, if $\Lambda$ is large enough, satisfies $\hat V^\sigma_t\circ\hat F^\sigma(\gamma)=\bar V_t^\Lambda(\gamma)$, and is close to $\bar{\eta}^{\sigma}$ in the weak topology.
Moreover, since now the velocity fields $\hat V^\sigma_{t_i}(\hat F^{\sigma}(\gamma_n))$ are all contained in the ball of radius $\Lambda$ in ${\mathbb{R}}^d$, the same construction performed in Proposition \[prop:countsticky\] with $\delta\leq\sigma/3$ leads to a finite sticky particle solution $\eta^\sigma$ with the desired properties.
Finite sticky particle solutions
--------------------------------
Finally we can prove the density of finite sticky particle solutions in the set of dissipative solutions.
\[thm:stickydense\] Let $\eta\in\mathcal M(\Gamma)$ be a dissipative solution. [T]{}hen there [exist]{} $\eta^\nu$ sticky particle solutions with finitely many trajectories (finite sticky particle solutions) with the property that $\eta^\nu\rightharpoonup\eta$ as $\nu\to 0$.
By Propositions \[prop:discretesticky\], \[prop:countsticky\] and \[prop:finitesticky\] we can assume that $\eta\in\mathcal M(\Gamma)$ is a dissipative finite particle solution. Then we proceed as in the construction of Lemma \[lemma:discrete\_sticky\], where at each step $[t_i,t_{i+1})$ we perturb the speed $V_{t_i}$ into a vector field $\tilde V_{t_i}$ in order to have that the trajectories $\gamma_n(t_{i+1}) - \tilde V_{t_i}(\gamma_n) t$ do not intersect in $[t_i,t_{i+1})$. Being the intersection conditions a set of codimension $d-1$, it is fairly easy to see that we can assume $\tilde V_{t_i}$ arbitrarily close to $V_{t_i}$.
PDE formulations {#S:PDE_formul}
================
In this section we give a kinetic and PDE formulation of our notion of solution.
Define the kinetic measure $\varpi_t \in \mathcal P({\mathbb{R}}^d \times {\mathbb{R}}^d)$ $$\int \phi(x,v) \varpi_t(dxdv) = \int \phi(\gamma(t),V_t(\gamma)) \eta(d\gamma).$$
\[prop:kinetic\_equation\] The measure $\varpi_t$ satisfies the PDE $$\label{eq:kinetic_varpi}
\partial_t \varpi_t + v \cdot \nabla_x \varpi_t + \mathrm{div}_v \pi = 0,$$ where $\pi$ is a distribution such that $$\langle\phi, \pi\rangle \leq \|\nabla_v \phi\|_{C^0}$$ and for every function $\Psi(t,x,v)$ convex in $v$ with $\Psi(t,x,v) \leq C \psi(t,x) (1+|v|^2)$ for some $C > 0$, $\psi \in C^1_c({\mathbb{R}}^d \times {\mathbb{R}}^+)$ it holds $$\label{eq:dissip_varpi}
\langle\Psi, \mathrm{div}(\pi)\rangle \geq 0.$$
The requirement that $\Psi$ has quadratic growth and compact support in $(t,x)$ implies that it can be used as a test function for .
Since the weak formulation is invariant for weak limits, we can write the PDE for the approximate finite sticky particle solutions found in Theorem \[thm:stickydense\], and pass to the limit in the estimates obtained. For finite sticky particle solutions $$\varpi_t = \sum_n c_n \delta_{(x_n(t),v_n(t))},$$ so that $$\begin{split}
\partial_t \varpi_t + v \cdot \nabla_x \varpi_t &= \sum_{i} \sum_n c_n \big[ \delta_{(x_n(t_i),v_n(t_i))} - \delta_{(x_n(t_i),v_n(t_{i-1}))} \big] \\
&= \sum_i \sum_{x_{ij}} \sum_{x_n(t_i) = x_{ij}} c_n \big[ \delta_{(x_{ij},v_n(t_i))} - \delta_{(x_{ij},v_n(t_{i-1}))} \big],
\end{split}$$ where in the last equality we have used the fact that the variation of speed occurs only at the times $t_i$ in finite many points $x_{ij}$. We can write the r.h.s. also in divergence form as $$\partial_t \varpi_t + v \cdot \nabla_x \varpi_t = \sum_i \sum_{x_{ij}} \sum_{x_n(t_i) = x_{ij}} c_n \mathrm{div}_v \bigg( \frac{v_n(t_{i-1}) - v_n(t_i)}{|v_n(t_{i-1}) - v_n(t_i)|} \mathcal H^1 \llcorner_{\{(1-\ell) v_n(t_{i}) + \ell v_n(t_{i-1}),\ell \in [0,1]\}} \bigg),$$ which shows that the equation for $\varpi_t$ is in divergence form as in .
We now show that $\pi$ is a distribution which can be computed on $C^1$-functions (i.e. it is a first order distribution). Recall that for a $C^1$ function $\bar \phi$ $$\begin{split}
\bar\phi(v) - \bar \phi(\bar v) &= \bigg( \int_0^1 \nabla \bar \phi \big( (1-l) v_1 + l v_2 \big) dl \bigg) \cdot (v_2 - v_1),
\end{split}$$ so that testing $$- \pi = \sum_i \sum_{x_{ij}} \sum_{x_n(t_i) = x_{ij}} c_n \bigg( \frac{v_n(t_{i-1}) - v_n(t_i)}{|v_n(t_{i-1}) - v_n(t_i)|} \mathcal H^1 \llcorner_{\{(1-\ell) v_n(t_{i}) + \ell v_n(t_{i-1}),\ell \in [0,1]\}} \bigg)$$ with a $C^1$ function $\phi=\phi(t,x,v)$ and using $$\label{eq:distrte_classic_aver_sped}
v_n(t_i) = \frac{\sum_{x_n(t_i) = x_{ij}} c_n v_n(t_{i-1})}{\sum_{x_n(t_i) = x_{ij}} c_n}.$$ one obtains $$\begin{split}
- \int \phi \pi &= \sum_i \sum_{x_{ij}} \sum_{x_n(t_i) = x_{ij}} c_n \big( v_n(t_{i-1}) - v_n(t_i) \big) \int_0^1 \phi \big( t_i,x_n(t_i),(1 -\ell) v_n(t_i) + \ell v_n(t_{i-1}) \big) d\ell \\
&= \sum_i \sum_{x_{ij}} \sum_{x_n(t_i) = x_{ij}} c_n \big( v_n(t_{i-1}) - v_n(t_i) \big) \phi(t_i,x_n(t_i),v_n(t_i)) \\
& \quad + \sum_i \sum_{x_{ij}} \sum_{x_n(t_i) = x_{ij}} c_n \big( v_n(t_{i-1}) - v_n(t_i) \big) \\
&{ \qquad \qquad \qquad \qquad \bigg[ \bigg( \int_0^1 (1 - \ell) \nabla_v \phi \big( t_i,x_n(t_i),(1 - \ell) v_n(t_i) + \ell v_n(t_{i-1}) \big) d\ell \bigg) \cdot \big( v_n(t_{i-1}) - v_n(t_i) \big) \bigg] }\\
&= \sum_i \sum_{x_{ij}} \sum_{x_n(t_i) = x_{ij}} c_n \big( v_n(t_{i-1}) - v_n(t_i) \big) \\
& \qquad \qquad \qquad \qquad \bigg[ \bigg( \int_0^1 (1 - \ell) \nabla_v \phi \big( t_i,x_n(t_i),(1 - \ell) v_n(t_i) + \ell v_n(t_{i-1}) \big) d\ell \bigg) \cdot \big( v_n(t_{i-1}) - v_n(t_i) \big) \bigg].
\end{split}$$ Using the dissipation of energy proved in Proposition \[prop:diss\] it follows that $${\big|} \langle \phi, \pi\rangle {\big|} \leq \|\nabla_v \phi\|_{C^0} \|W_0\|_{L^2_\eta}^2.$$ This proves the first claim about $\pi$, since $\eta \in \mathcal M(\Gamma)$.
Testing $\div_v\pi$ with a function $\Psi(t,x,v)$ convex w.r.t. $v$ we obtain $$\begin{split}
\langle\Psi,\div_v\pi\rangle=-
\sum_i \sum_{x_{ij}} \sum_{x_n(t_i) = x_{ij}} c_n \big[ \Psi(t_i,x_n(t_i),v_n(t_i)) - \Psi(t_{i-1}, x_n(t_{i-1}), v_n(t_{i-1})) \big] \geq 0
\end{split}$$ by Jensen inequality and . This concludes the proof.
Now that we have a kinetic formulation one can give the following Eulerian formulation. Define $$\varpi_t(dxdv) = \int \varpi_{t,x}(dv) \rho_t(dx), \quad \rho_t = (\mathtt p_x)_\sharp \varpi_t,$$ and $$u_t(x) = \int v \varpi_{t,x}(dv), \quad w_t(x) = \int (v \otimes v) \varpi_{t,x}(dv),$$ we can observe that implies that $$\label{Equa:Euler_equation}
\partial_t \rho_t + \mathrm{div}_x (\rho_t u_t) = 0, \quad \partial_t (\rho_t u_t) + \mathrm{div}_x (\rho_t w_t) = 0{,}$$ which is the Eulerian formulation of the sticky particle system. Using instead $\phi(t,x) |v|^2$ we deduce that $$\phi(t,x)\langle |v|^2, \mathrm{div}_v \pi\rangle = \mu \in \mathcal M^+({\mathbb{R}}^+\times {\mathbb{R}}^d),$$ which embodies the dissipation of energy. However in general we cannot compute the third order moment [$\int |v|^2v \varpi_{t,x}(dv)$]{} as a Lebesgue integral, so the energy balance cannot be written in terms of $\rho_t$ integrable functions.
A $G_\delta$ dense set of initial data {#sec:gdelta}
======================================
In this section we prove that there is a $G_\delta$ dense set of initial data for which any sticky particle solution departing from them is a flow in which particles do not interact. In order to make this statement precise, we need to introduce the following definitions.
We say that a dissipative solution $\eta\in\mathcal M(\Gamma)$ is a free flow if $\eta$ is concentrated on a set of straight lines with empty mutual intersection.
Let $W_0:\Gamma\to{\mathbb{R}}^d$ be the continuous map defined by $W_0(\gamma)=\gamma(0)-\gamma(-1)$. We say that a probability measure $\nu_0\in\mathcal P({\mathbb{R}}^d\times{\mathbb{R}}^d)$ is an initial data of a dissipative solution if there exists $\eta\in\mathcal M(\Gamma)$ dissipative solution s.t. $\nu_0=(e_0,W_0)_{\#}\eta$, i.e. $$\int\phi(x,v)\nu_0(dxdv)=\int\phi(\gamma(0),W_0(\gamma))\eta(d\gamma).$$
Notice that, by compactness of the set of dissipative solutions and by continuity of the map $W_0$, the set of initial data of dissipative solutions is compact as well. Moreover, since finite convex combinations of Dirac deltas on ${\mathbb{R}}^d$ pointing in different directions give always rise to a sticky particle solutions, by density it actually coincides with $$P_{2,1}({\mathbb{R}}^d\times{\mathbb{R}}^d)=\Bigl\{\nu_0\in\mathcal P({\mathbb{R}}^d\times{\mathbb{R}}^d):\,\int|x|^2{\mathtt{p}_x}_{\#}\nu_0\leq1,\,\int|v|^2{\mathtt{p}_v}_{\#}\nu_0\leq1\Bigr\}.$$
The main result of this section is Theorem \[thm:freeflow0\], which we recall below:
\[thm:freeflow\] There is a set $D_0\subset\mathcal P_{2,1}({\mathbb{R}}^d\times{\mathbb{R}}^d)$ such that, for any $\nu_0\in D_0$ there exists a unique dissipative solution $\eta$ with initial data $\nu_0$ and it is given by a free flow. Such a set is a dense $G_{\delta}$ set (i.e. of second category) in the weak topology on $\mathcal P_{2,1}({\mathbb{R}}^d\times{\mathbb{R}}^d)$.
Being the map $e_0\times W_0$ continuous, also $(e_0\times W_0)^{-1}(D_0)$ is a $G_\delta$-set. But it cannot be dense in the set of dissipative solutions: just consider a finite sticky particle solution where trajectories do interact.
In order to prove Theorem \[thm:freeflow\] we need the following preliminary lemmas.
\[lemma:freeflowfinite\] If $$\nu_0(dxdv) = \sum_{n=1}^N c_n \delta_{(x_n,v_n)}(dxdv)$$ and the trajectories $x_n + {\mathbb{R}}^+ v_n$ do not intersect, then there is a unique $\eta\in\mathcal M(\Gamma)$ dissipative solution s.t. $\nu_0=(e_0,W_0)_{\#}\eta$ given by $$\eta(d\gamma) = \sum_{n=1}^N c_n \delta_{\{x_n + t v_n\}}(d\gamma).$$
\[lemma:straight\] Let $\eta\in\mathcal M(\Gamma)$ be such that $D(\eta)=0$. Then $\eta$ is concentrated on straight lines of the form $\gamma(0)+tW_0(\gamma)$.
Define
$$\label{eq:D_0}
D_0 = \Big\{ \nu_0\in\mathcal P_{2,1}({\mathbb{R}}^d\times{\mathbb{R}}^d):\, D(\eta)=0\quad\forall\,\eta\in\mathcal M(\Gamma)\text{ dissipative solutions s.t. }(e_0,W_0)_{\#}\eta=\nu_0 {\Big\}}.$$
\[lemma:nodiss\] If $\nu_0\in D_0$, then $\nu_0$ is concentrated on a graph $(x_0,v_0(x_0))\subset{\mathbb{R}}^d\times{\mathbb{R}}^d$ of a map $v_0$ such that the straight lines $[0,T]\ni t\mapsto x_0+tv_0$ do not intersect and there is a unique $\eta\in\mathcal M(\Gamma)$ dissipative solution with $(e_0,W_0)_{\#}\eta=\nu_0$ given by the free flow concentrated on these straight lines.
Thanks to Lemma \[lemma:nodiss\], it is sufficient to prove that the set $D_0$ defined in is a dense $G_\delta$ set in the weak topology.
Let us first prove that $D_0$ is dense in $\mathcal P_{2,1}({\mathbb{R}}^d\times{\mathbb{R}}^d)$.
Consider then any initial data $\nu_0=(e_0,W_0)_{\#}\eta\in\mathcal P_{2,1}({\mathbb{R}}^d\times{\mathbb{R}}^d)$. Applying first Proposition \[prop:discretesticky\], then Proposition \[prop:countsticky\] and finally Proposition \[prop:finitesticky\] we find a sequence of finite sticky particle solutions $\eta^\sigma$ weakly converging to $\eta$ in $\mathcal M(\Gamma)$. By continuity of $W_0$, the measures $\nu_0^\sigma=(e_0,W_0)_{\#}\eta^\sigma$ converge weakly to $\nu_0$.
Since $\eta^\sigma$ is a [sticky]{} finite particle solution and the dimension $d$ is greater or equal than $2$, it is not difficult to see that it is possible to modify slightly the initial datum $\nu_0^\sigma$ to obtain an initial datum $\bar\nu_0^\sigma=(e_0,W_0)_{\#}\bar\eta^\sigma$ in $D_0$. Indeed, let
$$\nu_0^\sigma = \sum_n \eta^\sigma(\{\gamma_n\}) \delta_{(\gamma_n(0),W_0(\gamma_n))}(dxdv),$$
be the initial datum of a finite sticky particle solution. The set of initial speeds $\tilde W_0$ for which $$\exists n \not= m \text{ s.t. } \dist \Big( \big\{\gamma_n(0) + t W_0(\gamma_n),\, t\in [-1,T]\big\}, \big\{\gamma_m(0) + t W_0(\gamma_m),\, t\in[-1,T]\big\} \Big) = 0$$ is [contained in]{} the set $$\exists n \not= m \text{ s.t. } \gamma_n(0) - \gamma_m(0) \parallel W_0(\gamma_n) - W_0(\gamma_m),$$ which is closed and has codimension $d-1$. Therefore any initial datum $\nu_0$ of a finite sticky particle solution with $W_0$ belonging to the open and dense complement of the above set has a free flow solution with a finite number of trajectories at a strictly positive mutual distance. Now we apply Lemma \[lemma:freeflowfinite\] to obtain that $\bar{\nu}_0^\sigma$ generates a unique dissipative solution given by a free flow, namely $\bar{\nu}_0^\sigma\in D_0$.
Having proved the density of $D_0$, let us prove it is given by the intersection of countably many open sets.
Define, for all $k\in\N$, the sets $$D_{1/k} = {\Big\{} \nu_0\in\mathcal P_{2,1}({\mathbb{R}}^d\times{\mathbb{R}}^d):\, D(\eta)<1/k\quad\forall\,\eta\in\mathcal M(\Gamma)\text{ dissipative solutions s.t. }{(e_0,W_0)}_{\#}\eta=\nu_0 {\Big\}}.$$
It is clear that $$D_0=\bigcap_{k}D_{1/k}.$$
We claim that the sets $D_{1/k}$ are open sets. Indeed, if this were not the case, there would be $\nu_0\in D_{1/k}$ and a sequence of initial data $\nu_0^n\rightharpoonup\nu_0$ each generating a dissipative solution $\eta^n$ such that $D(\eta^n)\geq1/k$. By compactness of the set of dissipative solutions, up to extracting a subsequence and relabelling it the measures $\eta^n$ converge to a dissipative solution $\eta$ with $(e_0,W_0)_{\#}\eta=\nu_0$. By definition, it is not difficult to check that the total dissipation is upper semicontinuous w.r.t. weak converge of measures: indeed the measure $$\int\phi(t,x,v)\varpi_n(dtdxdv)=\int{\biggl[}\int\phi(t,\gamma(t),\dot{\gamma}(t)){\biggr]}\eta^n(d\gamma)$$ converges weakly for all $\phi$ continuous, i.e. $W_0$, and being $|v|^2$ convex $$\int\int_0^T|V_t(\gamma)|^2dt\eta(d\gamma)=\int|v|^2{\varpi}\leq\liminf_n\int|v|^2\varpi_n=\liminf_n\int\int_0^T|V_t(\gamma)|^2dt\eta(d\gamma).$$ Therefore one should have that $D(\eta)\geq1/k$, contrary to the assumption on the initial data $\nu_0$.
Let us now prove the series of preliminary lemmas.
First of all, by Proposition \[prop:diss\], it follows that $\eta$ is concentrated on the set of trajectories $\gamma$ such that $$\|\dot \gamma\|_\infty \leq \max_n |v_n| = \bar V.$$ Hence if $\bar d$ is the minimal distance among the sets $\{x_n + {\mathbb{R}}^+ v_n\}$, a trajectory $\gamma$ starting in $x_n$ needs a time of order $\bar t = \bar d/2\bar V$ before interacting with a trajectory $\gamma'$ starting in $x_m \not= x_n$. In the interval of time $[0,\bar t]$ the partition $\gamma \mapsto \gamma(0)$ is then the least sharp partition allowed, and being the initial one one concludes that $\dot \gamma(t) = \dot \gamma(0)$. The statement is obtained by repeating the argument for every interval of time $[n,n+1] \bar t$.
If $D(\eta)=0$, then the map $(-1,T]\ni t\mapsto\int|V_t(\gamma)|^2\eta(d\gamma)$ is constant, where we have used the usual convention $V_t=W_0$ if $t\in(-1,0)$. Recalling the definition of $V_t$, this implies that $$\int|W_0(\gamma')|^2\omega_{T_t(\gamma)}(d\gamma')= {\biggl|} \int W_0(\gamma')\omega_{T_t(\gamma)}(d\gamma') {\biggr|}^2$$ that in turn by Jensen inequality implies that $W_0$ is constant on $T_t^{-1}(T_t(\gamma))$ for all $\gamma$ and for all $t$. Hence, $$\dot{\gamma}(t)=V_t(\gamma)=W_0(\gamma) \quad\mathcal L^1\times \eta\text{{-}a.e. in $(-1,T]\times\Gamma$},$$ namely $\eta$ is concentrated on straight lines with velocity $W_0(\gamma)$.
Let $\eta\in\mathcal M(\Gamma)$ dissipative solution with $(e_,W_0)_{\#}\eta=\nu_0\in D_0$. By Lemma \[lemma:straight\] we know that $\eta$ is concentrated on a set $\Delta\subset\Gamma$ of straight lines.
From now on, we identify straight lines $\gamma(t)=x_0+v_0t\in\Gamma$ with points $(x_0,v_0)\in{\mathbb{R}}^d\times{\mathbb{R}}^d$ according to our convenience.
Define the set $$H = \Big\{ \big( x_0,v_0,x_0',v_0' \big) \in {\mathbb{R}}^{2d} \times {\mathbb{R}}^{2d} : (x_0,v_0) \not= (x_0',v_0') \ \text{and} \ \exists t \in [0,T] \text{ s.t. } x_0 + v_0 t = x_0' + v_0' t \Big\}.$$
We will use also the notation $$\gamma(t) = x_0 + v_0 t, \quad \gamma'(t) = x_0' + v_0' t,$$ and in some cases consider $H$ as a subset of $\Gamma \times \Gamma$. Let $t_{\gamma,\gamma'}$ be the first crossing time in the definition of $H$.
For every $(x_0,v_0,x_0',v_0') \in H$ consider the map $$\label{Equa:gamma_gamma_prime_map}
\left. \begin{array}{c}
\gamma(t) = x_0 + v_0 t, \\
\gamma'(t) = x_0' + v_0' t,
\end{array} \right\}
\quad \mapsto \quad
\begin{cases}
\tilde \gamma_{\gamma'}(t) = \gamma(t) \chi_{t \leq t_{\gamma,\gamma'}} + \frac{\gamma(t) + \gamma'(t)}{2} \chi_{t > t_{\gamma,\gamma'}}, \\
\tilde \gamma'_{\gamma}(t) = \gamma'(t) \chi_{t \leq t_{\gamma,\gamma'}} + \frac{\gamma(t) + \gamma'(t)}{2} \chi_{t > t_{\gamma,\gamma'}}.
\end{cases}$$ The effect of the map is to replace the curves $\gamma,\gamma'$ at the first crossing point with the line with their average speed.
We will use the duality results of [@kel:duality Proposition 3.3]: if $B$ is Borel (or analytic) $$\label{Equa:kellerer_1}
\sup_{\pi \in \mathrm{adm}(\mu_1,\mu_2)} \pi(B) = \inf \bigg\{ \mu_1(B_1) + \mu_2(B_2):\, B_1 \times X \cup X \times B_2 \supset B \bigg\}.$$ We have used the notation $$\mathrm{adm}(\mu_1,\mu_2) = \Big\{ \pi \in \mathcal M^+(X \times Y):\, (\mathtt p_x)_\sharp \pi \leq \mu_1, (\mathtt p_y)_\sharp \pi \leq \nu_2 \Big\}.$$
Consider an admissible transference plan $\pi$ concentrated in $H\cap \Delta\times\Delta$: the plan is said to be admissible if $$(\mathtt p_{x_0,v_0})_\sharp \pi \leq \nu_0, \quad (\mathtt p_{x_0',v_0'})_\sharp \pi \leq \nu_0.$$ Being $H$ symmetric, we can assume that $\pi$ is also symmetric, i.e. $\pi(A \times B) = \pi(B \times A)$. Define the disintegration $$\begin{aligned}
\pi(d\gamma d\gamma')& = \int \pi_\gamma(d\gamma')\eta_{\llcorner\mathtt p_{x_0,v_0}(H\cap \Delta\times\Delta)}(d\gamma) \notag\\
&=\int \pi_\gamma(d\gamma')\eta(d\gamma), \end{aligned}$$ where we set $\pi_{\gamma}=0$ if $\gamma\notin\mathtt p_{x_0,v_0}(H\cap \Delta\times\Delta)$.
Define the new Lagrangian representation $\tilde \eta$ as follows: $$\label{Equa:tilde_eta_def}
\int \phi(\gamma) \tilde \eta(d\gamma) = \int \phi(\gamma) {(1 - \|\pi_\gamma\|)} \eta(d\gamma) + \int \bigg[ \int \phi(\tilde \gamma'_{\gamma}) \pi_{\gamma}(d\gamma') \bigg] \eta(d\gamma).$$ The meaning of $\tilde \eta$ is that part of the curve $\gamma$ is replaced with the curve $\tilde \gamma'_{\gamma}$ constructed in with weight according to $\pi_{\gamma}(d\gamma')$. $\tilde{\eta}$ is clearly a dissipative solution.
Notice that $(e_0,W_0)_{\#}\tilde{\eta}=\nu_0$.
Indeed, since $$\tilde \gamma'_\gamma(0) = \gamma(0), \quad \tilde \gamma'_\gamma(-1) = \gamma(-1), $$ the claim is proved.
Now let us compute the dissipation for $\tilde{\eta}$. One has that
$$\begin{aligned}
\int|W_0(\gamma)|^2\tilde{\eta}(d\gamma)-\int|\dot{\gamma}(t)|^2\tilde{\eta}(d\gamma)=\int|W_0(\gamma)|^2{\|\pi_{\gamma}\|}\eta(d\gamma)-\int{\biggl[}\int|\dot{\tilde{\gamma}}'_{\gamma}(t)|^2\pi_{\gamma}(\gamma'){\biggr]}\eta(d\gamma).
\end{aligned}$$
Since $$|\dot{\tilde{\gamma}}'_{\gamma}(t)|^2=\frac{|W_0(\gamma)+W_0(\gamma)'|^2}{4}<\frac{|W_0(\gamma)|^2}{2}+\frac{|W_0(\gamma')|^2}{2}\quad \forall\,t\geq t_{\gamma,\gamma'},$$
one has that $D(\tilde{\eta})>0$.
In particular, under the assumptions of the lemma, we conclude that $H$ is negligible for all admissible plans $\pi$. Thus by there are two sets $N_1,N_2 \subset \Delta$ such that $H \subset N_1 \times \Gamma \cup N_2 \times \Gamma$. Removing $N_1 \cup N_2$ from $\Delta$ we obtain that the remaining trajectories are disjoint.
\[Rem:converse\_not\_intersecting\_not\_true\] The fact that there exists a dissipative solution $\eta$ with $D(\eta)=0$ does not imply that $(e_0,W_0)_{\#}\eta\in D_0$. Indeed, the Example 3 of Bressan and Nguyen [@BN] consists in constructing a sequence of particles whose mass is decreasing such that the intersection with of the $i$-th and $i+1$-th occurs only if the intersection of the $i+1$-th with $i+2$-th occurs. In this example $\eta$ is concentrated on a set of trajectories which if prolonged have empty pairwise intersection, but since the perturbation needed in order to make them to intersect becomes negligible also the stricly dissipating solution is a solution.
Thanks to the above Propositions, it follows that for a dense $G_\delta$ set of initial data $\varpi_t$ is a measure concentrated on a graph, i.e. $\varpi_t = (u_t)_\sharp \rho_t$, with $u_t \in \L^2(\rho_t)$, and the distribution $\pi = 0$ up to divergence free distributions. Moreover, it is fairly easy to see that for every curve $\gamma(t) = x_0 + v_0 t$ one has $$\int \phi(t,\gamma(t)) dt \leq \|\phi\|_{C^0} \frac{2 \diam(\supp(\phi))}{1 + |v_0|},$$ so that $$\begin{split}
\int \phi(t,x) |u_t|^3 dxdt &= \int \bigg[ \int \phi(t,\gamma(t)) dt \bigg] \eta(d\gamma) \\
&\leq 2 \|\phi\|_{C_0} \diam(\supp(\phi)) \int \frac{|W_0(\gamma)|^3}{1 + |W_0(\gamma)|} \eta(d\gamma) < \infty.
\end{split}$$ Thus we can compute also the third moment of $\varpi_t$ and it belongs to $L^3_\loc({\mathbb{R}}^+\times{\mathbb{R}}^d) $, resulting into the complete pressureless Euler system (without Young measures) $$\partial_t \rho_t + \mathrm{div}_x (\rho_t u_t) = 0, \quad \partial_t (\rho_t u_t) + \mathrm{div}_x(\rho_t u_t \times u_t) = 0, \quad \partial_t (\rho |u_t|^2) + \mathrm{div}_x (\rho_t |u_t|^2 u_t) = 0.$$
\[Rem:other\_G\_delta\] As a final observation, we note that, for a generic approximation scheme, the only requirements for the existence of a dense $G_\delta$-set as in Theorem \[thm:freeflow\] are that
1. for a dense set of initial data (e.g. finitely may $\delta$’s as in Lemma \[lemma:freeflowfinite\]) the only solution is the free flow;
2. the scheme dissipates a convex entropy (e.g. $D(\eta)$ is our case), l.s.c. w.r.t. to convergence of the initial data;
3. if the dissipation is $0$, the only solution is the free flow.
We observe the intersection of dense $G_\delta$-sets is a dense $G_\delta$ set, so that we scan say that up to a set of first category the solution contructed by this scheme coincide with the dissipative solution of Definition \[def:sticky\].
[10]{}
L. Boudin. A solution with bounded expansion rate to the model of viscous pressureless gases. 32 (1), 172–193 (2002).
Y. Brenier, W. Gangbo, G. Savaré and M. Westdickenberg. Sticky particle dynamics with interactions. 99 (5), 577–617 (2013).
Y. Brenier and E. Grenier. Sticky particles and scalar conservation laws. 35 2317–2328 (1998).
F. Bouchut and F. James Equations de transport unidimensionnelles a coefficients discontinus 320 1097–1102 (1995).
A. Bressan and T. Nguyen. Non-existence and non-uniqueness for multidimensional sticky particle systems. 7 (2), 205–218 (2014).
F. Cavalletti, M. Sedjro and M. Westdickenberg. A simple proof of global existence for the 1D pressureless gas dynamics equations. 47, 66–79 (2015).
F. Cavalletti, M. Sedjro and M. Westdickenberg. A variational time discretization for the compressible Euler equations. 371 (2019).
W. E, Y.G. Rykov and Y.G. Sinai. Generalized variational principles, global weak solutions and behaviour with random initial data for systems of conservation laws arising in adhesion particle dynamics. 177, 349–380 (1996).
D. H. Fremlin. Measure theory (2010).
E. Grenier. Existence globale pour le systeme des gaz sans pression. 321, 171–174 (1995).
F. Huang and Z. Wang. Well-posedness for pressureless flows. 222, 117–146 (2001).
H.G. Kellerer. Duality theorems for marginals problems. 67-4, 399–432 (1984).
L. Natile and G. Savaré. A Wasserstein approach to the one-dimensional sticky particle system. 41, 1340–1365 (2009).
T. Nguyen and A. Tudorascu. Pressureless Euler/Euler-Poisson system via adhesion dynamics and scalar conservation laws. 40 (2), 754–775 (2008).
T. Nguyen and A. Tudorascu. One-dimensional pressureless gas systems with/without viscosity. 40 (9), 1619–1665 (2015).
F. Poupaud and M. Rascle Measure solutions to the linear transport equations with nonsmooth coefficients 22 337–358 (1997).
M. Sever. An existence theorem in the large for zero-pressure gas dynamics. 14 (9), 1077–1092 (2001).
A. Sobolevskii. The small viscosity method for a one-dimensional system of equations of gas dynamic type without pressure. 56 (2) 707–709 (1997).
Ya. B. Zeldovich. Gravitational instability: an approximate theory for large density perturbations. 5, 84–89 (1970).
[^1]: [email protected]
[^2]: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We introduce a class of discrete time stationary trawl processes taking real or integer values and written as sums of past values of independent ‘seed’ processes on shrinking intervals (‘trawl heights’). Related trawl processes in continuous time were studied in Barndorff-Nielsen (2011) and Barndorff-Nielsen et al. (2014), however in our case the i.i.d. seed processes can be very general and need not be infinitely divisible. In the case when the trawl height decays with the lag as $j^{-\alpha}$ for some $1< \alpha < 2 $, the trawl process exhibits long memory and its covariance decays as $j^{1-\alpha}$. We show that under general conditions on generic seed process, the normalized partial sums of such trawl process may tend either to a fractional Brownian motion or to an $\alpha$-stable Lévy process.'
author:
- 'P. Doukhan[^1], A. Jakubowski[^2], S.R.C. Lopes[^3] and D. Surgailis[^4]'
title: 'Discrete-time trawl processes with long memory '
---
[*Keywords:*]{} trawl process, integer and continuous-valued time series, long memory, fractional Brownian motion, Lévy process.\
[*AMS Classification subjects 2010*]{}\
60G22 Fractional processes, including fractional Brownian motion\
60G51 Processes with independent increments; Lévy processes\
60G99 Trawl process\
60K99 Long range memory process
Introduction {#sec1}
============
The present paper introduces a class of stationary random processes of the form $$\label{g1}
X_k\ =\ \sum_{j=0}^\infty \gamma_{k-j}(a_j), \qquad k \in \Z$$ where $\gamma_k = \{\gamma_k (u), u \in \R\}$ are i.i.d. copies of a generic process $\gamma = \{\gamma (u), u \in \R\}$ tending to zero in probability as $u \to 0$, and $a_j \in \R$ for $j \in \N, \, \lim_{j \to \infty} a_j =0$ are deterministic numbers. Clearly, includes the class of causal moving averages $X_k = \sum_{j=0}^\infty a_j \xi_{k-j} $ in i.i.d. r.v.s $\{\xi, \xi_k\}$, which correspond to a trivial process $\gamma = \{\gamma (u) = \xi u, u \in \R\}$. In as follows, we call $X = \{X_k, k \in \Z\} $ the [*trawl process*]{} corresponding to the [*seed process* ]{} $\gamma = \{\gamma (u), u \in \R\}$ and [*trawl* ]{} $ a = \{a_j, j \ge 0\} $. The above terminology is borrowed from Barndorff-Nielsen et al. (2014) [@bar2014] which considered a related class of trawl processes in continuous time represented as stochastic integrals $$\label{g11}
Y_t\ =\ \int_{(-\infty, t] \times \R} \1( x \in (0, d_{t-s})) L(\d x, \d s), \qquad t \in \R$$ where $L(\d x, \d s) $ is a homogeneous Lévy measure on $\R^2 $, with independent values on disjoint sets, and $\{d_t, t\in \R_+ \} $ is a deterministic function satisfying certain conditions. In the case when this function takes constant values $d_t = a_j,$ if $ t \in (j, j+1],$ for $ j = 0,1, \dots $, the discretized process $\{Y_k, k \in \Z\} $ in coincides with $\{ X_k, k \in \Z\}$ in with independent increment (Lévy) seed process $\Big\{\zeta (u) = \int_{(0, u] \times (0,1]} L(\d x, \d s), u \in \R \Big\} $. Clearly, an integer-valued seed process $\gamma = \{\gamma (u), u \in \R\}$ in results in an integer-valued trawl process $\{ X_k, k \in \Z\}$, similarly as in the case of continuous-time trawl processes of studied in [@bar2014]. On the other hand, the discrete-time set-up allows us to consider very general seed processes $\gamma$ which need not be infinitely divisible or have independent increments as in [@bar2014].\
([@bar2011], page 22) note that trawl processes represent a flexible class of stochastic processes which can be used to model serially dependent count data and other stationary time series, where the marginal distribution and the autocorrelation structure can be modeled independently from each other. Particularly, trawl processes can exhibit long memory or long-range dependence, which is usually associated with the divergence of the covariance series: $\sum_{k\in \Z} |\cov (X_0, X_k)| = \infty$, see [@gir2012], and which occurs in models and when the trawl function decays sufficiently slowly with the lag, see [@bar2014] and § \[sec2\] below. ([@bar2014], figure 6) exhibit sample paths and autocorrelation graphs of integer-valued trawl process with long-memory trawl function showing a remarkably slow decay and a disagreement between true and sample autocorrelations based on a very large sample length.\
The main question studied in this paper, which is also one of the basic questions for statistical applications of trawl processes, is the rate of convergence and the limit distribution of the sample mean. We prove that for trawl process with long-memory trawl function $a_j$ decaying as $j^{-\alpha}, 1< \alpha < 2$ this limit distribution is either $\alpha$-stable or Gaussian, moreover, a non-Gaussian stable limit is typical for integer valued seed (and trawl) process, while a Gaussian limit occurs for ‘continuous’ seed processes, e.g. diffusions or stochastic volatility processes. We note that our non-Gaussian result contradicts the conjecture in ([@bar2014], page 708) about a Gaussian partial sums limit for long-memory trawl process in . In particular, for a standard Poisson seed process $\gamma $ and $a_j \sim c_0 j^{-\alpha}, 1 < \alpha < 2 $ we obtain, with $H={(3-\alpha)}/2$, a sequence of processes $$Z_n(t)=\frac 1{n^{H}}\sum_{j=1}^{[nt]}(X_j-\E X_j)$$ whose second order moments converge to those of a fractional Brownian motion, $B_H$ with index $H$: $$\lim_{n\to \infty}\cov (Z_n(s),Z_n(t))= \cov (B_H(s),B_H(t)), \qquad \forall s,t.$$ Moreover, $Z_n(t)\to 0$ in probability (the process is evanescent) but $n^{H-\frac1\alpha}Z_n(t)$ converges to a non-trivial limit which is an $\alpha$-stable Lévy process. (Note $H-\frac1\alpha=\frac{(2-\alpha)(\alpha-1)}{2\alpha}>0$ since $1<\alpha<2$.)\
A similar phenomenon (convergence of the partial sums process to a Lévy stable process) occurs for a number of long-range dependent stationary processes with finite variance, see [@TaL1986], [@TaWS1997], [@Konst1998], or [@miko2002], [@Wi2003], [@ls2003], [@su2004], [@kajt2008], [@pils2014] and the references therein, although in most of the literature this convergence is limited to finite-dimensional distributions. For $M/G/\infty$ queue with heavy-tailed activity periods, the adequate functional convergence was proved in [@ResnickVdB2000]. Since the limiting stable processes in these works have independent increments, the above behavior is sometimes called ‘distributional short-range dependence’ in contrast to ‘distributional long-range dependence’ occurring when the limit of the partial sums process has dependent increments. See [@deh2002], [@lps2005]. See also [@lif2014] for a nice discussion of stable and Gaussian limits under long-range dependence.
Discrete-time trawl process {#sec2}
===========================
Existence of discrete-time trawl process {#sec21}
----------------------------------------
Let $\gamma_k = \{\gamma_k (u), u \in \R\}$ be i.i.d. copies of a generic [*seed process*]{} $\gamma = \{\gamma (u), u \in \R\}$ with finite variance $g(u) = \v (\gamma (u))$ and mean $\mu(u) = \E \gamma (u)$ tending to zero as $u \to 0 $ so that $\gamma(0) = 0$ and $\gamma (u) \to_\P 0 $ as $u \to 0$. A trawl $ a = \{a_j \ge 0, j \in \N\} $ is a deterministic sequence such that $\lim_{j \to \infty} a_j = 0$. We shall assume that $$\label{mug}
|\mu(u)| = {\cal O}(g(u)) \to 0 \quad (u \to 0)$$ and $$\label{ga}
\sum_{j=0}^\infty g(a_j) < \infty.$$ The trawl process $X = \{ X_k, k \in \Z \} $ corresponding to trawl $ a = \{a_j \ge 0, j \in \N\} $ and seed process $\gamma = \{\gamma (u), u \in \R\}$ is defined as $$\label{Xk}
X_k\ =\ \sum_{j=0}^\infty \gamma_{k-j}(a_j), \qquad k \in \Z.$$ Let $$\label{rho1}
\rho(u,v) = \cov (\gamma(u), \gamma(v)), \qquad (u,v \in \R )$$ denote the covariance function of the seed process $\gamma $. The following statement is an easy consequence of the Kolmorogov three series theorem.
Let conditions and be satisfied. Then the series in converges a.s. and in mean square for any $k \in \Z $, and defines a stationary process with mean $\E X_k = \sum_{j=0}^\infty \mu(a_j)$ and covariance function $$\label{rhoX}
{\cov} (X_0,X_k) \ = \ \sum_{j=0}^\infty \rho (a_j, a_{j+k}), \qquad k \in \N.$$
Clearly, if the seed process takes integer values: $\gamma(u) \in \Z, \, u \in \R$, this property also holds for the trawl process: $X_k \in \Z \, \ (\forall \, k \in \Z) $. The following examples show that the class of trawl processes is very large.
\[Random line seed process\] \[ex1\] [Let $\gamma (u) = \xi u, u \in \R $, where $\xi $ is a r.v. with zero mean and variance $\sigma^2<\infty $. Then $\mu(u) = 0,$ $g(u) = \sigma^2 u^2 $ and condition translates to $\sum_{j=0}^\infty a^2_j < \infty $. Then $X$ in is a moving-average: $$\label{MA}
X_k = \sum_{j=0}^\infty a_j \xi_{k-j},$$ where $\{\xi_k, k \in \Z\}$ are i.i.d. copies of $\xi$. ]{}
\[Brownian motion seed process\] \[ex2\] [Let $\gamma (u) = B(u), u \in \R_+ $, where $B$ is a Brownian motion with zero mean and covariance $\E B(u) B(v) = u \wedge v $ and $a_j \ge 0$. Then $X$ in is a stationary Gaussian process with zero mean and covariance $\cov (X_0, X_k) = \sum_{j=0}^\infty a_j \wedge a_{k+j}, k \in \N$. Particularly, if $a_j = a^j, a \in (0,1)$ then $\cov (X_0, X_k) = a^k/(1-a) $ and $X$ in agrees with an AR(1) process written as a moving-average in with Gaussian innovations $\xi_k \sim {\cal N}(0,\sigma^2) $ and $\sigma^2 = 1+ a $. ]{}
\[Poisson and Bernoulli seed processes\] \[ex3\] [Let $\gamma (u) = P(u)$, $u \in \R_+ $, where $P$ is a Poisson process with mean $\mu(u) = u$, covariance $\rho(u,v) = \cov (P(u), P(v)) = u \wedge v $ and $a_j \ge 0$. Then $X$ in is a stationary process with mean $\E X_k = \sum_{j=0}^\infty a_j $ and the same covariance as in Example \[ex2\]. Moreover, $X_k $ takes integer values and has a Poisson marginal distribution with mean $\E X_0$.\
The above example can be generalized by considering a mixed Poisson seed process $\gamma(u)=P(u\zeta)$, where $P$ is as above and $\zeta > 0$ is a random variable with $\E \zeta < \infty $, independent of $P$. Particularly, [@fok] proved that when $\zeta$ is exponentially distributed then $P(u\zeta)$ has negative binomial marginal distribution.\
The Bernoulli seed process is defined by $\gamma(u)={\1}(U \le u) $, where $U\sim {\cal U}[0,1]$ is a uniformly distributed random variable. Note also $$\begin{aligned}
\mu(u)&=u\E\zeta, &\rho(u,v)&= (u \wedge v)\E\zeta + uv \v(\zeta)\ &\mbox{($\gamma $ \ is a mixed Poisson process)}, \\
\mu(u)&=u, &\rho(u,v) &= u\wedge v-uv \ &\mbox{($\gamma $ \ is a Bernoulli process).}\end{aligned}$$ ]{}
Further examples of trawl processes can be found in § \[31\] (Examples \[Ex4\]-\[Ex5\]) and § \[32\] (Example \[Rem1\]). As explained in § \[sec1\], this paper is focused on long memory properties and the behavior of the partial sums process of stationary trawl process $X$ in .
Second order properties of discrete-time trawl process {#sec22}
------------------------------------------------------
The covariance function $\cov (X_0,X_k)$ in depends both on the trawl $a = \{ a_j \} $ and on the covariance function $\rho (u,v) $ of the seed process. In order to characterize long memory property in terms of the trawl $a = \{ a_j \} $ alone, it is convenient to impose a linear growth condition on the variance $g (u) = \v (\gamma(u))$ at the origin $u=0$: $$\label{g0}
g(u) = |u|( 1 + o(1)), \qquad u \to 0.$$ Under , condition is equivalent to the summability of the trawl sequence: $$\label{G12}
\sum_{j=0}^\infty |a_j| < \infty.$$ Moreover, for obtaining more precise decay of the covariance function in we also assume that $$\begin{aligned}
\label{G11}
\rho(u,v)&=&(|u| \wedge |v|) (1 + o(1)), \quad \text{as} \quad u, v \to 0, \ uv >0.\end{aligned}$$ Clearly, the trawl processes in Examples \[ex2\] and \[ex3\] satisfy and provided the seed processes in these examples are suitably extended to negative $u <0$. Denote by $S_n = \sum_{k=1}^n X_k$ the partial sums process of the trawl process in .
\[propLM\] (i) Assume conditions , , and $$\label{g4}
a_j \ = \ c_0 j^{-\alpha} (1+ o(1)), \quad j \to \infty \quad (\exists \, c_0 \ne 0, \, 1 < \alpha < 2).$$ Then $$\label{covX}
\cov (X_0, X_k) \ = \ c_1 k^{1-\alpha}(1+ o(1)), \qquad k \to \infty$$ and $$\label{g5}
\v (S_n) = \sum_{k, l=1}^n \cov (X_k, X_l) \ \sim \ c_2 \, n^{3-\alpha} \gg n, \quad n \to \infty,$$ where $c_1 = c_0/(\alpha -1)$, and $ c_2 = 2c_1/(2-\alpha)(3-\alpha)$.
\(ii) Assume conditions , , $$\label{rhobdd}
|\rho(u,v)| \le C (|u|\wedge |v|) \qquad (u, v \in \R)$$ and $$\label{G4}
\sum_{j=1}^\infty j |a_j| \ < \ \infty.$$ Then $$\label{G5}
\sum_{k=1}^\infty |\cov (X_0, X_k)| \ < \ \infty$$ and $$\label{g6}
\v (S_n) = n\sum_{|k|<n}\Big(1-\Big|\frac kn\Big|\Big) \cov (X_k, X_0) \sim \sigma^2 \, n,$$ where $\sigma^2 =
\sum_{k\in \Z} \cov (X_0, X_k)$.
[The estimation of the parameter of interest needs additional work: it will be considered in further papers. ]{}
[*Proof.*]{} (i) Let $c_0 >0$ in , the case $c_0 < 0$ follows analogously. Then $a_j >0$, and $a_{k+j} >0$ hold for all $k \ge 1 $ and $j > j_0$, where $j_0$ is large enough. Moreover, for any $\epsilon >0$ there exists $j_0< j_\epsilon < \infty $ such that $$\label{ajk}
a_{j+k} < a_j, \quad \text{for all} \quad \forall \ j_\epsilon < j < k/2\epsilon, \quad \forall \, k \ge 2\epsilon j_\epsilon.$$ Indeed, by we have that for any $\epsilon >0$ there exists $j_\epsilon > j_0> 0$ such that $a_j > c_0j^{-\alpha}(1-\epsilon)$, $a_{k+j} < c_0 (j+k)^{-\alpha} (1+ \epsilon)$ and therefore $$\Big( \frac{a_{j+k}}{a_j}\Big)^{\frac1\alpha} < \frac j{j+k}\Big(\frac{1+ \epsilon}{1-\epsilon}\Big)^{\frac1\alpha}, \qquad \forall \,
j > j_\epsilon, \quad \forall \, k\ge 1.$$ Since $((1+ \epsilon)/(1-\epsilon))^{\frac1\alpha} < 1 + 2\epsilon$ if $\epsilon >0$ is small enough, relation follows since $ j/(j+k) \le 1/(1 + 2\epsilon)$ for $1 \le j < k/2\epsilon $. Next, for sufficiently large $k$ ($k > 2\epsilon j_\epsilon$) split $k^{\alpha -1} \cov (X_0, X_k)
= \sum_{j=0}^\infty k^{\alpha -1} \rho(a_j, a_{k+j}) = \sum_{i=1}^3 I_{i,k}$, where $$I_{1,k} = \sum_{0\le j \le j_\epsilon} \dots, \qquad
I_{2,k} = \sum_{j_\epsilon < j < k/2\epsilon} \dots, \qquad
I_{3,k} = \sum_{j \ge k/2\epsilon } \dots.$$ By , and Cauchy-Schwartz inequality, for any fixed $\epsilon >0$ and $1 \leq j \leq j_\epsilon$, $$|\rho(a_j, a_{k+j})| \le g(a_j)^{\frac12} g(a_{k+j})^{\frac12}
\le C |a_{k+j}|^{\frac12} \le C k^{-\frac\alpha2}, \qquad k \to \infty$$ implying $$|I_{1,k}| \le C k^{\alpha-1} k^{-\frac \alpha2} = O(k^{ -(1-\frac \alpha2)}) = o(1), \qquad k \to \infty.$$ Next, by and , $|\rho(a_j, a_{j+k})| \le C |a_j| \wedge |a_{j+k}|
\le C j^{-\alpha}, \, (\forall \, j, k \ge 1) $ and therefore $$I_{3,k} \le C k^{\alpha -1} \sum_{j \ge k/2\epsilon } j^{-\alpha}
\le C \epsilon^{\alpha -1 }$$ can be made arbitrarily small uniformly in $k \ge 1 $ by choosing $\epsilon >0$ small enough. Finally, by and , $$\label{Ik}
I_{2,k} = c_0 k^{\alpha-1} \sum_{j_\epsilon < j < k/2\epsilon } \frac{1 + \delta_{j,k}}{ (k+j)^{\alpha}} ,$$ where $\sup_{j \ge 1} |\delta_{j,k}| = 0$ as $k \to \infty $. Note that for each $\epsilon >0$, as $k \to \infty $ $$\begin{aligned}
\label{Jk}
J_{k}(\epsilon)&:=&k^{\alpha-1} \sum_{j_\epsilon < j < k/ 2\epsilon} (k+j)^{-\alpha} \
= \ \frac{1}{k} \sum_{\frac {j_\epsilon}k < \frac jk < 1/2\epsilon} \frac{1}{ \left(1 + \frac jk\right)^\alpha} {\nonumber}\\
&\to&\int_0^{1/2\epsilon } \frac{\d x}{ (1 + x)^\alpha}\ = \ \frac{1}{\alpha -1} \left(1 - (2\epsilon)^{\alpha -1} \right).\end{aligned}$$ According to and , for any $\delta >0$ and any $\epsilon_0>0$ one can find $0<\epsilon < \epsilon_0 $ and $K_0 >0 $ such that $|I_{2,k} - c_0/(\alpha -1)| < \delta $ holds for all $k > K_0$. This proves while follows from , see e.g. ([@gir2012], proposition 3.3.1).
\(ii) It suffices to prove since follows from and the dominated convergence theorem. According to , , , $$\begin{aligned}
\sum_{k=1}^\infty |\cov (X_0,X_k)|
&\le&C\sum_{k=1}^\infty \sum_{j=0}^\infty |a_j| \wedge |a_{j+k}| \\
&\le&C \sum_{k=1}^\infty \sum_{j=0}^\infty |a_{j+k}| \ \le \ C \sum_{k=1}^\infty k |a_k| \ < \ \infty.\end{aligned}$$ Proposition \[propLM\] is proved. $\Box$
Partial sums limits of trawl processes
======================================
We shall consider two typical cases of the seed process $\gamma $ in :
[**Case 1**]{}:
: $\gamma(u), u \ge 0 $ is centered: $\mu (u) = 0$ and a.s. continuous (e.g., a Brownian motion).
[**Case 2**]{}:
: $\gamma(u), u \ge 0 $ is a pure jump process (a typical example is a Poisson process with $\mu (u) = g(u) = u $).
Particularly, in Examples \[ex2\] and \[ex3\] of $\gamma $ (Brownian motion and Poisson process) and a regularly decaying trawl $a = \{a_j\}$ in with exponent $1< \alpha < 2 $ the conditions of Proposition \[propLM\] (i) are satisfied and the covariance function of the trawl process decays as $ k^{1-\alpha }$, see . The last fact implies that the variance of $S_n = \sum_{k=1}^n X_k $ grows faster than $n$, see .
In the following subsections we detail conditions on the seed process $\{\gamma (u), u \in \R\}$ which guarantee that the partial sums process of the trawl process $\{X_k\}$ with regularly decaying trawl tends to either a Gaussian process (fractional Brownian motion with Hurst parameter $H = (3-\alpha)/2 \in (1/2, 1) $ (Case 1) or to a $\alpha$-stable Lévy process (Case 2). The following decomposition of the partial sums process as a sum of independent random variables is crucial for the proofs of Theorem \[thmgauss\] and Theorem \[thmstable\].
\[Decomposition\] \[decomplem\] We have $$\begin{aligned}
\label{Zn}
S_n&=&\sum_{k=1}^n X_k \ = \ \sum_{s=-\infty}^n Z_{s,n}, \quad \text{where} \quad
Z_{s,n}\ =\ \sum_{k=1\vee s}^n \gamma_s(a_{k-s}).\end{aligned}$$ Then the random variables $(Z_{s,n})_{s \le n}$ are independent.
Write $\to_{f.d.d.}$ for the weak convergence of finite-dimensional distributions and $\to_{{\mathcal D}(J_1)}$ and $\to_{{\mathcal D}(M_1)} $ for the weak convergence of random elements in the Skorohod space $D[0,1]$ endowed with the $J_1$-topology and the $M_1$-topology, respectively. For the definition of these topologies, see Skorohod [@Skorokhod1956] or [@Bill99], [@LouhichiRio2011], [@ResnickVdB2000]. Denote $|\mu|_{2+\delta}(u) = \E |\gamma(u)|^{2+\delta} $ the absolute $(2+\delta)$-moment of the seed process.
Gaussian scenario (Case 1) {#31}
--------------------------
\[thmgauss\]
- Assume $\mu(u) = \E \gamma(u) = 0$, , , and $$\label{G2}
|\mu|_{2+\delta}(u) = {\cal O}(|u|^{\frac{2+\delta}2}), \qquad (u \to 0, \, \exists \, \delta >0).$$ Then $$\label{cltD}
\frac 1{n^H} S_{[nt]} \ \to_{{\mathcal D}(J_1)} \ \sqrt{ c_2} \,B_H(t), \qquad H = \frac{3-\alpha}2$$ where $B_H$ is fractional Brownian motion with variance $\E B^2_H (t) = t^{2H} $ and $c_2$ is defined in .
- Assume $\mu(u) = \E \gamma(u) = 0$, , , and $\sigma^2 =
\sum_{k\in \Z} \cov (X_0, X_k) \ne 0$. Then $$\label{cltD2}
\frac 1{\sqrt{n}} S_{[nt]} \ \to_{f.d.d.} \ \sigma \,B(t),$$ where $B$ is a Brownian motion with variance $\E B^2(t) = t$.\
In addition, if $\sum_{k=1}^\infty \sqrt{|a_k|} < \infty $, then the finite dimensional convergence in can be replaced by $\to_{{\mathcal D}(J_1)}$.
- Assume the same conditions as in (ii) except that is replaced by $$\label{mu0}
|\mu|_{2+\delta}(u) = {\cal O}(u) \quad (u \to 0) \quad \text{and} \quad \sum_{j=0}^\infty |a_j|^{\frac1{2+\delta}} < \infty$$ for some $\delta >0$. Then all statements in part (ii) remain valid.
[*Proof.*]{} (i) Consider the convergence of one-dimensional distributions: $$\label{clt}
\frac 1{\sqrt{n^{3-\alpha}}} S_{n} \ \to_{law} \ {\cal N}(0,c_2).$$ In view of and Lemma \[decomplem\], relation follows by Lindeberg’s theorem provided $$\label{lind}
L_n \ := \ \sum_{s=-\infty}^n \E |Z_{s,n}|^{2+\delta} \ = \ o\big(n^{\frac{(3-\alpha)(2+\delta)}2}\big).$$ By Minkowski’s inequality and assumptions and we obtain $$\begin{aligned}
\label{Zsn}
\E |Z_{s,n}|^{2+\delta}&\le&\left( \sum_{k=1\vee s}^n (\E |\gamma(a_{k-s})|^{2+\delta})^{\frac1{2+\delta}} \right)^{2+\delta} \\
&\le&C\left( \sum_{k=1\vee s}^n |a_{k-s}|^{\frac12} \right)^{2+\delta} \
\le\ C\left( \sum_{k=1\vee s}^n |k-s|_+^{-\frac\alpha2} \right)^{2+\delta} {\nonumber}\end{aligned}$$ and therefore $L_n \le C(L^-_n + L^+_n)$, where $$\begin{aligned}
L^-_n&=&\sum_{s=-\infty}^0 \left( \sum_{k=1}^n |k-s|_+^{-\frac\alpha2} \right)^{2+\delta}
\ = \ \sum_{s=0}^\infty \left( \sum_{k=1}^n (k+s)^{-\frac\alpha2} \right)^{2+\delta}, \\
L^+_n&=&\sum_{s=1}^n \left( \sum_{k=1}^n k^{-\frac\alpha2} \right)^{2+\delta} \ = \ n \left( \sum_{k=1}^n k^{-\frac\alpha2} \right)^{2+\delta}.\end{aligned}$$ Here, $L^+_n = {\cal O}\left( n \big( n^{1-\frac \alpha2})^{2+\delta}\right) = o\big(n^{\frac{(3-\alpha)(2+\delta)}2}\big)$. The same relation for $L^-_n $ follows from $$\begin{aligned}
L^-_n&\le&\int_{0}^\infty \d x \left( \int_0^n (x+y)^{-\frac\alpha2} \d x \right)^{2+\delta}\ = \ c n \left( n^{1- \frac\alpha2}\right)^{2+\delta}, \quad \mbox{ with}\\
c& =& \int_{0}^\infty \d x \left( \int_0^1 (x+y)^{-\frac\alpha2} \d x \right)^{2+\delta} < \infty.\end{aligned}$$ This proves and the one-dimensional convergence in . Finite-dimensional convergence in follows similarly using Cramér-Wold device. Finally, the tightness in ${\mathcal D}(J_1)$ of the partial sums process in follows by Kolmogorov’s criterion and from property (see, e.g. [@gir2012], proposition 4.2.2). This proves part (i).
\(ii) Again, it suffices to prove the convergence of one-dimensional distributions: $$\label{clt2}
n^{-1/2} S_{n} \ \to_{law} \ {\cal N}(0,\sigma^2).$$ By writing $S_n$ as in and using Lindeberg’s theorem relation follows from $$\label{lind2}
L_n \ = \ \sum_{s=-\infty}^n \E |Z_{s,n}|^{2+\delta} \ = \ o\big(n^{\frac{2+\delta}2}\big).$$ Using Minkowski’s inequality and assumptions and similarly as in part (i) we obtain $$\begin{aligned}
\label{Zbdd}
\E |Z_{s,n}|^{2+\delta}&\le&C\Big( \sum_{k=1\vee s}^n |a_{k-s}|^{\frac12} \Big)^{2+\delta} \\
&\le&C\Big( \sum_{k=1\vee s}^n |(k-s)a_{k-s}| \Big)^{\frac{2+\delta}2} \Big( \sum_{k=1\vee s}^n (k-s)^{-1} \Big)^{\frac{2+\delta}2} \nonumber \\
&\le&C\Big( \sum_{k=1\vee s}^n (k-s)^{-1} \Big)^{\frac{2+\delta}2}.\end{aligned}$$ and hence $$\begin{aligned}
\sum_{s=-n}^n \E |Z_{s,n}|^{2+\delta}&\le&C n (\log n)^{\frac{2+\delta}2} \ = \ o \big(n^{\frac{2+\delta}2}\big), \\
\sum_{s=-\infty}^{-n} \E |Z_{s,n}|^{2+\delta}
&\le&C \sum_{s=n}^\infty \Big( \sum_{k=1}^n \frac1{k+s} \Big)^{\frac{2+\delta}2} \ \le \
C \sum_{s=n}^\infty ( n s^{-1} )^{\frac{2+\delta}2} \ \le \ C n \ = \ o \big(n^{\frac{2+\delta}2}\big),\end{aligned}$$ proving and . To show the last statement of (ii) (the tightness in $D[0,1]$), it suffices to prove the bound $$\label{Sp}
\E |S_n|^{2+\delta} \ \le \ C n^{\frac {2+\delta}{2}},$$ see ([@gir2012], proposition 4.4.4). By Rosenthal’s inequality, $$\E |S_n|^{2+\delta} \ \le \ C \Big( \sum_{s=-\infty}^n (\E |Z_{s,n}|^{2+\delta})^{\frac{2}{2+\delta}} \Big)^{\frac{2+\delta}{2}}.$$ Using and $\sum_{k=1}^\infty |a_k|^{\frac12} < \infty $, we get $\max_{|s| \le n} \E |Z_{s,n}|^{2+\delta} < C $ and $$\begin{aligned}
\sum_{s=-\infty}^{-n} (\E |Z_{s,n}|^{2+\delta})^{\frac2{2+\delta}}
&\le&C\sum_{s=n}^\infty \Big(\sum_{k=1}^n |a_{k+s}|^{\frac12} \Big)^2 {\nonumber}\\
&\le&C\sum_{k_1, k_2=1}^n \sum_{s=n}^\infty |a_{k_1+s}|^{\frac12} |a_{k_2+s}|^{\frac 12} \ \le \ Cn. \label{2nd}\end{aligned}$$ This proves and part (ii), too.
\(iii) Similarly as in and using we get $$\E |Z_{s,n}|^{2+\delta} \le C\Big( \sum_{k=1\vee s}^n |a_{k-s}|^{\frac{1}{2+\delta}} \Big)^{2+\delta} \le C\sum_{k=1\vee s}^n |a_{k-s}|^{\frac{1}{2+\delta}} \le C$$ for any $-\infty < s \le n$ and hence $$\begin{aligned}
\label{Zsn0}
\sum_{s=-\infty}^{-n} \E |Z_{s,n}|^{2+\delta}&\le&C\sum_{s=n}^\infty \sum_{k=1}^n |a_{k+s}|^{\frac{1}{2+\delta}} \ \le \ Cn, \\
\sum_{s=-\infty}^{-n} (\E |Z_{s,n}|^{2+\delta})^{\frac{2}{2+\delta}}&\le&C
\sum_{s=n}^\infty \big( \sum_{k=1}^n |a_{k+s}|^{\frac{1}{2+\delta}} \big)^2 \ \le \ Cn\end{aligned}$$ as in . Hence, and follow, proving part (iii) and completing the proof of Theorem \[thmgauss\]. $\Box$
[The crucial condition for Gaussian partial sums limit under long-range dependence assumption in Theorem \[thmgauss\] (i) is . Clearly this condition is satisfied for the Brownian motion $\gamma (u) = B(u) $, in which case $|\mu|_{2+\delta}(u) = \E |B(u)|^{2+\delta} = |u|^{\frac{2+\delta}2} \E |B(1)|^{2+\delta} $. On the other hand, condition is not satisfied for most jump processes. Particularly, if $\gamma (u) = P(u) - u, u \ge 0$ is a centered Poisson process with intensity $\E P(u) = u$, then $$|\mu|_{2+\delta}(u) = u \e^{-u} |1-u|^{2+\delta} + {\cal O}(u^{2+ \delta} + u^2)
\sim u \quad (u \to 0)$$ and fails, but the first condition in is satisfied. In particular, in the case of Poisson seed process, the trawl process satisfies Donsker’s theorem if the trawl decays fast enough so that holds. ]{}
Let us present further examples of seed processes satisfying the conditions in Theorem \[thmgauss\].
\[Ex4\] [Set $\gamma(u) = \e^{ B(u) - u/2} -1, u \ge 0$, where $B$ is a standard Brownian motion as above. We have $\E \gamma(u) = 0$ and (if $u\le v$) $$\begin{aligned}
\rho(u,v)&=&\E\exp\{ B(u) + B(v) - \frac{u+v}2\} -1 \\
&=&\exp\Big\{ (\frac12 \E (B(u) + B(v))^2 - \frac{u+v}2\Big\} - 1 \\
&=&\exp\Big\{ (\frac12 (u + v + 2 u) - \frac{u+v}2\Big\} - 1 \\
&=&\e^u -1 = u\wedge v + O \big( (u\wedge v)^2\big), \quad u\wedge v \to 0.\end{aligned}$$ Therefore conditions , are satisfied. We also have by Taylor’s expansion that $|\mu|_{4}(u) = \E \left| \e^{ B(u) - u/2} -1\right|^{4}
= \e^{6u} - 4 \e^{3u} + 6 \e^u - 3= {\cal O}(u^2), \ u \to 0 $ so that is satisfied with $\delta = 2 $. ]{}
\[Diffusion process\]\[Ex5\] [$$\gamma(u) = \int_0^u b(v) \d B(v)$$ with $B $ a Brownian motion, and $(b(v))_{v \ge 0}$ a random predictable process with $\lim_{v\to 0}\E b^2(v)= C >0 $. Then $g(u) = \int_0^u \E b^2(v) \d v \sim C u \, (u \to 0) $ and $\rho(u,v)= g(u), \ 0 \le u \le v $ so that conditions and are satisfied. Moreover, if $\E |b(v)|^{2+\delta} \le C$ then by the moment inequality for Brownian integrals (see, e.g. [@kwa1992], theorem 9.9.2) $$\begin{aligned}
|\mu|_{2+\delta}(u)&\le&C\E \left(\int_0^u b^2(v) \d v\right)^{\frac{2+\delta}2} \\
&\le&C \left(\int_0^u \E |b(v)|^{2+\delta} \d v \right) \left( \int_0^u 1 \, \d v \right)^{\frac{2+ \delta}2 - 1} \
\le\ Cu^{\frac{2+\delta}2},\end{aligned}$$ hence assumption holds, too. ]{}
Stable scenario (Case 2) {#32}
------------------------
We assume now that seed process $\gamma = \{\gamma(u), u \ge 0 \}$ is a piecewise constant nondecreasing process starting at $\gamma(0)=0$ with unit jumps at points $0 = \tau_0 < \tau_1 < \tau_2 < \cdots $: $$\label{Z0}
\gamma(u)= \sum_{k=0}^\infty k\cdot \1(\tau_k \le u < \tau_{k+1}) $$ and such that the distribution of the first jump-point $\tau_1 >0$ has a bounded probability density $\theta(u)$: $$\label{tau1}
\P(0 < \tau_1 \le u) = \int_0^u \theta(y) \d y \qquad \text{and} \qquad \lim_{u \to 0} \theta(u) = 1.$$ Moreover, we shall assume that there exists $\delta >2(\alpha-1)$ such that $$\begin{aligned}
&&\E \gamma(u)^{2+ \delta}\ <\ \infty, \qquad \qquad \qquad \forall \ u >0, \label{Z1} \\
&&\E \gamma(u)^2 \1 (\tau_2 \le u)\ =\ {\cal O}(u^2), \qquad u \to 0. \label{Z2}\end{aligned}$$
[The second condition in can be replaced by $\lim_{u \to 0} \theta(u) = C >0$ without loss of generality. Conditions - are very general and there are satisfied by many jump processes $\gamma$ as this was sketched in the introduction. As shown below, these conditions also imply the conditions on $\gamma $ in Proposition \[propLM\].\
Remark that $(\tau_1\le u)=(\gamma(u)\ge 1)$, $(\tau_2\le u)=(\gamma(u)\ge2)$ and therefore an alternative way to set condition is $\E\gamma^2(u){\1}(\gamma(u)>1)= {\cal O}(u^2)$, as $ u \to 0$. ]{}
For the seed process $\gamma$ in , conditions - imply the assumptions and of Proposition \[propLM\] (i). In addition, if $$\begin{aligned}
&&\E \gamma(v) \1 (\tau_1 \le u, \tau_2 \le v)\ =\ o(u), \qquad 0 \le u \le v \to 0, \label{Z3}\end{aligned}$$ then is satisfied.
[*Proof.*]{} From we have $$\label{zetabdd}
\1 (\tau_1 \le u) \ \le \ \gamma(u) \ \le \ \1 (\tau_1 \le u) + \gamma(u) \1 (\tau_2 \le u)$$ and hence $$\P (\tau_1 \le u) \ \le \ \mu(u) \ \le \ \P(\1 (\tau_1 \le u) + \E \gamma(u) \1 (\tau_2 \le u)$$ From , $\P(0 < \tau_1 \le u) = u (1 + o(1)) $ and from , $$\E \gamma(u) \1 (\tau_2 \le u) \le \E \gamma^2(u) \1 (\tau_2 \le u) = {\cal O}(u^2).$$ Therefore, $$\label{mu1}
\mu (u)\ = \ u (1 + o(1)) + {\cal O}(u^2) \ = \ u (1 + o(1)) \qquad (u \to 0).$$ Similarly, for the second moment $\mu_2(u) = \E \gamma^2(u)$ from , , we obtain $$\P (\tau_1 \le u) \ \le \ \mu_2(u) \ \le \ \P(\1 (\tau_1 \le u) + 2\E \gamma(u) \1 (\tau_2 \le u) +
\E \gamma^2(u) \1(\tau_2 \le u),$$ implying $\mu_2(u) = u (1 + o(1)) + {\cal O}(u^2) = u (1 + o(1)) \ (u \to 0)$ and $$\label{mu3}
g(u)\ = \ \mu_2(u) - \mu^2(u)
\ = \ u (1 + o(1)) \qquad (u \to 0).$$ Clearly, and imply and . Consider assumption . Since $$\rho (u,v) = \E \gamma(u) \gamma(v) - \mu(u) \mu (v)
= \E \gamma(u) \gamma(v) - u v (1 + o(1)) = \E \gamma(u) \gamma(v) + o(u \wedge v),$$ as $ 0 < u \le v \to 0$, condition follows from $$\label{Ebdd}
\E \gamma (u) \gamma(v) = u(1 + o(1)), \qquad 0 < u \le v \to 0.$$ From for $0< u \le v $ we obtain $$\begin{aligned}
\P (\tau_1 \le u)&\le&\E \gamma (u) \gamma(v) \\
&\le&\P(\tau_1 \le u)
+\E \gamma(u) \1 (\tau_2 \le u) +
\E \gamma(v) \1(\tau_1 \le u, \tau_2 \le v) + \E \gamma(u) \gamma(v) \1 (\tau_2 \le u)\end{aligned}$$ where $\E \gamma(u) \gamma(v) \1 (\tau_2 \le u) \le (\E \gamma^2(u) \1 (\tau_2\le u))^{\frac12}
(\E \gamma^2 (v))^{\frac12} \le C u (\E \gamma^2 (v))^{\frac12} $ and $\E \gamma^2 (v) = \mu_2(v) = {\cal O}(v), $ see , . Hence and from we have that $$\E \gamma(u) \1 (\tau_2 \le u) +
\E \gamma(v) \1(\tau_1 \le u, \tau_2 \le v) + \E \gamma(u) \gamma(v) \1 (\tau_2 \le u) = o(u)$$ implying and , too. $\Box$
\[thmstable\] Assume that $a_j \ge 0$ satisfy the regular decay condition in with exponent $1< \alpha < 2 $ and that the seed process in satisfies conditions -. Then $$\label{cltL}
n^{-\frac1\alpha} (S_{[nt]} - \E S_{[nt]}) \ \to_{f.d.d.} \ L_\alpha(t),$$ where $L_\alpha (t), t \ge 0$ is a homogeneous $\alpha$-stable Lévy process with characteristic function $$\label{chfL}
\E \e^{ \i z L_\alpha (t)} \ = \ \exp \left\{ - t |z|^\alpha \frac{c_0 \Gamma (2-\alpha)}{1-\alpha}
\left( \cos (\pi \frac\alpha2) - \i\cdot {\rm sgn}(z) \sin (\pi \frac\alpha 2) \right) \right\}, \quad z \in \R.$$
[*Proof.*]{} Denote $$\label{Zrv}
Z \ = \ \sum_{j=0}^\infty \gamma (a_j), \quad Z^* \ = \ \sum_{j= 0}^\infty \1(\gamma(a_j) \ge 1) \ = \ \# \{ j\ge 0: a_j \ge \tau_1 \},
\quad Z^{**} \ = \ Z - Z^*.$$ Then $Z \ge Z^* \ge 0$ and the series for $Z$ in converges a.s. in view of and has finite mean: $$\E Z = \sum_{j=0}^\infty \mu (a_j) \le C \sum_{j=0}^\infty a_j < \infty.$$ We shall prove that the tail d.f. of r.v. $Z$ decays regularly with exponent $\alpha \in (1,2)$: $$\label{Ztail}
\P (Z > y) \ = \ c_0 \, y^{-\alpha}(1 + o(1)), \qquad \text{as} \ \ y \to \infty.$$ Relation follows from and $$\label{Z1tail}
\P (Z^* > y) \ = \ c_0 \, y^{-\alpha}(1 + o(1)) \quad \text{and} \quad
\P (Z^{**} > y) \ = \ o(y^{-\alpha}), \quad \text{as} \ \ y \to \infty,$$ Consider the first relation in . Since $\P (Z^* > k-1) \ge \P (Z^* > y) \ge \P (Z^* > k) $ when $k-1 \le y \le k $, it suffices to show for $y = k-1 $, or the probability $\P (Z^* \ge k), k \in \N_+$. As noted in the proof of Proposition \[propLM\], for any $\epsilon >0$ there exists $j_0 > 0$ such that $c_0 (1-\epsilon) j^{-\alpha} < a_j < c_0 (1+ \epsilon) j^{-\alpha}, \, \forall \, j\ge j_0$. Clearly, for any $k \ge 1 $ we have $ \P (Z_- \ge k+j_0) \le \P (Z^* \ge k) \le \P(Z_+ \ge k-j_0)$, where $$\begin{aligned}
&Z_+ = \sum_{j=j_0}^\infty \1 (\tau_1 \le c_0(1+\epsilon)j^{-\alpha}) = \# \{j \ge j_0: \tau_1 \le c_0(1+\epsilon)j^{-\alpha}\}, \\
&Z_- = \sum_{j=j_0}^\infty \1 (\tau_1 \le c_0(1-\epsilon)j^{-\alpha}) = \# \{j \ge j_0: \tau_1 \le c_0(1+\epsilon)j^{-\alpha}\}.\end{aligned}$$ According to , as $k \to \infty$, $$\P (Z_+ \ge k-j_0) \ = \ \P (\tau_1 < c_0 (1+\epsilon) k^{-\alpha} )\ = \ \int_0^{c_0 (1+\epsilon) k^{-\alpha}} \theta (y) \d y
\ \sim \ c_0(1+\epsilon) \, k^{-\alpha}$$ and, similarly, $$\P (Z_- \ge k+j_0) \ = \ \P (\tau_1 < c_0 (1-\epsilon) (k + 2j_0 -1)^{-\alpha} )
\ \sim \ c_0(1-\epsilon) \, k^{-\alpha}.$$ Therefore, $c_0 (1- \epsilon) \le \liminf k^{\alpha} \P (Z^* \ge k) \le \limsup k^{\alpha} \P (Z^* \ge k) \le c_0 (1+ \epsilon)$, where $\epsilon >0$ is arbitrary small, proving the first relation in . To prove the second relation in , note $Z^{**} \le \sum_{j=0}^\infty \gamma(a_j) \1(a_j \ge \tau_2) $ and then by and Minkowski’s inequality we obtain $$\begin{aligned}
\E^{\frac12} (Z^{**})^2
&\le&\sum_{j=0}^\infty \big( \E \gamma^2(a_j) \1(a_j \ge \tau_2)\big)^{\frac12} \
\le\ C \sum_{j=0}^\infty |a_j| \ < \ \infty
\label{Z5}\end{aligned}$$ proving the second relation in and hence as well. In turn, implies that the distribution of r.v. $Z$ belongs to the domain of attraction of asymmetric $\alpha$-stable law, viz., $$\label{cltZ}
n^{-\frac 1\alpha} \sum_{k=1}^{[nt]} (Z_k - \E Z_k) \ \to_{f.d.d.} \ L_\alpha(t),$$ where $Z_k = \sum_{j=0}^\infty \gamma_k(a_{j}), \, k \in \Z$ are i.i.d. copies of r.v. $Z$ in and $L_\alpha $ is the $\alpha$-stable Lévy process in -. See e.g. ([@ibr1971], theorem 2.6.7).\
Relation follows from if we show that the partial sums process in can be approximated by the partial sums process in , in the sense that $$\label{ES}
\E |S_n - \widetilde S_n| \ = \ o(n^{\frac1\alpha}), \qquad \text{where} \quad \widetilde S_n
= \sum_{k=1}^n Z_k.$$ We have $\widetilde S_n - S_n = R'_n - R''_n$, where $$R'_n = \sum_{1 \le s \le n} \sum_{j > n-s} \gamma_s (a_j), \qquad
R''_n = \sum_{s \le 0} \sum_{1\le k \le n} \gamma_s (a_{k-s}),$$ then $ R'_n \ge 0, R''_n \ge 0$.\
Using and we obtain $$\begin{aligned}
\E R'_n&=&\sum_{1 \le s \le n} \sum_{j > n-s} \E \gamma_s (a_j) \ = \ \sum_{1 \le s \le n} \sum_{j > n-s} \mu(a_j) \\
&\le&C \sum_{1 \le s \le n} \sum_{j > n-s} j^{-\alpha} \ = \ {\cal O}(n^{2-\alpha}), \\
\E R''_n &=&\sum_{s \le 0} \sum_{1 \le k \le n} \E \gamma_s (a_{k-s}) \ = \ \sum_{s \le 0} \sum_{1 \le k\le n} \mu (a_{k-s}) \\
&=&\sum_{s \ge 0} \sum_{1\le k \le n} (k+s)^{-\alpha} \ = \ {\cal O}(n^{2-\alpha}),\end{aligned}$$ implying since $ 2- \alpha < 1/\alpha$ for $1 < \alpha < 2 $. Theorem \[thmstable\] is proved. $\Box$
\[Jump processes and the assumptions of Theorem \[thmstable\]\] \[Rem1\]
For such a jump process $(\gamma(u)=k)=(\tau_k\le u<\tau_{k+1})$. Conditions in - on the seed process $\{\gamma(u), u \ge 0\} $ in Theorem \[thmstable\] are rather weak and essentially involve the distribution of the first jump-point $\tau_1$ provided the second jump $\tau_2$ cannot occur very fast after $\tau_1$. Particularly,
- The Bernoulli process is very simple: in this case $\tau_2 = \infty $ thus $\P(\tau_2\le u)= \P(\tau_2\le v)
= 0$,
- The Poisson process in Example \[ex3\]. Indeed, for $\gamma (u) = P(u) $ holds since $$\begin{aligned}
&\E \gamma(u)^2 \1 (\tau_2 \le u)\
\ =\ \E \gamma(u)^2 - \P(\gamma(u) =1)\ = \ u + u^2 - u \e^{-u} \ = \ {\cal O}(u^2).\end{aligned}$$ Verification of for $\gamma (u) = P(u) $ is slightly more involved, as follows. Let $b(u,v) = \E \gamma(v) \1 (\tau_1 \le u, \tau_2 \le v)
= b_1(u,v) + b_2(u,v)$, where $b_1(u,v) = \E \gamma(v) \1 ( \tau_2 \le u) \le \P^{2/3}(\tau_2 \le u) \E^{1/3} \gamma^3(v) =
{\cal O}(u^{4/3}) = o(u) $ since $\P(\tau_2 \le u) = \P(\gamma(u) \ge 2) = {\cal O}(u^2), u \to 0$. Next, since $\tau_2 >u$ implies $\gamma(u) = 1$ so $b_2(u,v) = \E \gamma(v) \1 (\tau_1 \le u, u< \tau_2 \le v) = \P (\tau_1 \le u, u< \tau_2 \le v)
+ \E (\gamma(v) - \gamma(u)) \1 (\gamma(u)=1, \gamma(v) \ge 1), $ where $ \P (\tau_1 \le u, u< \tau_2 \le v) = \P(\gamma(u) =1, \gamma(v) -\gamma(u) \ge 1) = {\cal O}\big(u(v-u)\big) = o(u)$ and, similarly $\E (\gamma(v) - \gamma(u)) \1 (\gamma(u)=1, \gamma(v) \ge 1)
= \P (\gamma(u) = 1) \E \gamma (v-u) = {\cal O}(u(v-u)) = o(u), \, 0< u \le v \to 0$, proving .
- Other examples of jump processes satisfying - include mixed Poisson processes (Example \[ex3\]) and renewal process with independent intervals $\tau_1$ and $\tau_2 - \tau_1 $ and $\P (\tau_2 - \tau_1 \le x) = {\cal O}(x) $ since $$\P(\tau_2 \le u) = \int_0^u \theta(y) \P (\tau_2 - \tau_1\le u-y) \d y
\le C \int_0^u (u-y) \d y = {\cal O}(u^2)$$ as in the Poisson case. The same conditions also holds for mixed Poisson processes driven by some random variable $\zeta>0$ (Example \[ex3\]). (thus again the case of negative binomials fits our result as sketched in [@fok]).
We note that the functional convergence in is open and may not hold in the $J_1$-topology. At the cost of additional structure we can prove the convergence in Skorohod’s $M_1$-topology. For definitions and properties related to association of random variables we refer to [@EPW1967].
\[thmfunct\] Suppose that all assumptions of Theorem \[thmstable\] hold. If the jump random variables $\tau_1, \tau_2, \ldots $ are [*associated*]{} (in particular, if they are sums of independent positive random variables) then the finite-dimensional convergence (\[cltL\]) can be strengthened to $$\label{cltLF}
n^{-\frac1\alpha} (S_{[nt]} - \E S_{[nt]}) \ \to_{{\mathcal D}(M_1)} \ L_\alpha(t),$$
[*Proof.*]{} By ([@LouhichiRio2011], theorem 1) it suffices to verify that $X_1, X_2, X_3, \ldots $ are associated random variables. By ([@EPW1967], property $P_5$) it is enough to check association of $$\sum_{j=0}^N \gamma_{1-j}(a_j), \quad\sum_{j=0}^N \gamma_{2-j}(a_j), \ldots,\quad \sum_{j=0}^N \gamma_{k-j}(a_j),$$ for each $N\in \N$ and $k\in \N$, where $\gamma_j(\cdot)$ are independent copies of (\[Z0\]). This in turn is implied by ([@EPW1967], properties $P_4$ and $P_2$), provided the family $
\gamma(a_1), \gamma(a_2), \gamma(a_3),\ldots,
$ is associated. But $$\gamma(u)= \sum_{k=0}^\infty k\cdot \1(\tau_k \le u < \tau_{k+1}) = \sum_{k=1}^{\infty} \1(\tau_k \le u),$$ and by arguments already presented above it is enough to prove association of random variables $$\label{assoarray}
\begin{array}{cccc}
\1(\tau_1 \le a_0), & \1(\tau_2 \le a_0), & \cdots & \1(\tau_k \le a_0), \\
\1(\tau_1 \le a_1), & \1(\tau_2 \le a_1), & \cdots & \1(\tau_k \le a_1), \\
\vdots & \vdots & \ddots & \vdots \\
\1(\tau_1 \le a_N), &\1(\tau_2 \le a_N), & \cdots & \1(\tau_k \le a_N),
\end{array}$$ for each $N\in \N$ and $k\in \N$. Let us notice that $\1(\tau_j \le a_m), = 1 - \1(\tau_j > a_m)$ and that functions $f_m(x) = \1(x > a_m)$ are nondecreasing. Therefore, if $\tau_1, \tau_2, \ldots $ are associated, then also the family $\{ f_m(\tau_j) \,;\, j,m \in \N\}$ is associated. By ([@EPW1967], property $BP_1$) array (\[assoarray\]) is associated as well. $\Box$
\[plusminus\]
Let us consider two families $\{\gamma_j^+\}$ and $\{\gamma_j^-\}$ of processes of the form (\[Z0\]). Consider stationary processes $X_1^+, X_2^+, \ldots $, and $X_1^-, X_2^-, \ldots $, each built according to the recipe (\[Xk\]), and the corresponding partial sum processes $S_{[nt]}^+$ and $S_{[nt]}^-$.
If $\{\gamma_j^+\}$ and $\{\gamma_j^-\}$ are independent and $$\label{fddsing}
n^{-\frac1\alpha} (S^+_{[nt]} - \E S^+_{[nt]}) \ \to_{f.d.d.} \ L^+_\alpha(t), \quad n^{-\frac1\alpha} (S^-_{[nt]} - \E S^-_{[nt]}) \ \to_{f.d.d.} \ L^-_\alpha(t),$$ then also $$\label{fddsum}
n^{-\frac1\alpha} \Big(\big(S^+_{[nt]} - S^-_{[nt]}\big) - \E \big(S^+_{[nt]} - S^-_{[nt]}\big)\Big) \ \to_{f.d.d.} \ L_\alpha(t),$$ where $L_{\alpha} \sim L^+ - L^-$ with independent $L^+ \sim L^+_\alpha$ and $ L^- \sim L^-_\alpha$.
In particular, if $\gamma^+_j$ and $\gamma^-_j$ are identically distributed, then the resulting trawl process is centered and the limiting Lévy process is [*symmetric*]{}. This is the case if e.g. $\gamma^\pm$ are both homogeneous Poisson processes with identical intensities or Bernoulli processes $\gamma^\pm(u)={\1}(U^\pm\le u)$ for independent uniform rvs, $U^\pm$.
[As the example of an ordinary moving average with summable coefficients shows, (\[fddsing\]) may imply (\[fddsum\]) without the assumption of independence of $S^+_{[nt]}$ and $S^-_{[nt]}$ (see e.g. ([@BJL2016], corollary 2.2)). In the functional limit theorem given below we follow this general approach and obtain the functional convergence in the non-Skorohodian $S$ topology (see [@Jakubowski1997]). We shall denote by $\to_{{\mathcal D}(S)}$the weak convergence in the Skorohod space $D[0,1]$ equipped with the $S$ topology). ]{}
\[corStop\] In the framework of Remark \[plusminus\], suppose that both $S^+_{[nt]}$ and $S^-_{[nt]}$ satisfy all assumptions of Theorem \[thmfunct\], so that $$\label{functem}
n^{-\frac1\alpha} (S^+_{[nt]} - \E S^+_{[nt]}) \ \to_{\mathcal{D}(M_1)} \ L^+_\alpha(t), \quad n^{-\frac1\alpha} (S^-_{[nt]} - \E S^-_{[nt]}) \ \to_{\mathcal{D}(M_1)} \ L^-_\alpha(t),$$ for some $\alpha$-stable Lévy motions $L^+_\alpha$ and $L^-_\alpha$.
If for some càdlàg stochastic process $K$ we have $$\label{fddsumS}
n^{-\frac1\alpha} \Big(\big(S^+_{[nt]} - S^-_{[nt]}\big) - \E \big(S^+_{[nt]} - S^-_{[nt]}\big)\Big) \ \to_{f.d.d.} \ K(t),$$ then $$n^{-\frac1\alpha} \Big(\big(S^+_{[nt]} - S^-_{[nt]}\big) - \E \big(S^+_{[nt]} - S^-_{[nt]}\big)\Big) \ \to_{{\mathcal D}(S)}
\ K(t).$$
[*Proof.*]{} By ([@BJL2016], theorem 3.13) (\[functem\]) implies the uniform $S$-tightness of the corresponding processes. The proof of ([@BJL2016], proposition 3.16) gives the uniform $S$-tightness of the differences. A direct application of ([@BJL2016], proposition 3.3) concludes the proof. $\Box$.
#### Acknowledgements.
This study begun with a question from Wilfredo Palma (Santiago de Chile) to the first author: [*how to define LRD integer valued models?*]{} We wish to thank him for considering this problem.\
This work has been developed within the MME-DII center of excellence (ANR-11-LABEX-0023-01) and was partially supported by CNPq-Brazil.\
We also thank the Universities UFRGS (Porto Alegre) and Nicolaus Copernicus (Toruń) for their support.
[99]{}
Balan, R., Jakubowski, A. and Louhichi, S. (2016) Functional convergence of linear processes with heavy-tailed innovations. J. Theoret. Probab. 29, 491–526.
Barndorff-Nielsen, O. E. (2010) Stationary infinitely divisible processes. REBRAPE Braz. J. Probab. Stat. 25, 294–322.
Barndorff-Nielsen, O. E., Benth, F. E. and Veraart, A. E. D. (2011) Recent advances in ambit stochastics. Preprint available at arXiv:1210.1354.
Barndorff-Nielsen, O.E., Lunde, A., Shepard, N. and Veraart, A.E.D. (2014) Integer-valued trawl processes: a class of stationary infinitely divisible processes. Scand. J. Statist. 41, 693–724.
Billingsley, P.. (1999) [*Convergence of Probability Measures*]{}. 2nd ed., Wiley, New York.
Christou, V. and Fokianos, K. (2014) Quasi-likelihood inference for negative binomial time series. J. Time Series Anal. 35, 55–78.
Davydov, Y. A. (1970) The invariance principle for stationary processes. Theor. Probab. Appl. 15, 487–498.
Dehling, H. and Philipp, W. (2002) Empirical process techniques for dependent data. In: H. Dehling, T. Mikosch and M. S[ø]{}rensen (Eds.), [*Empirical Process Techniques for Dependent Data*]{}, pp. 1–113. Birkh[" a]{}user, Boston.
Doukhan, P. , Oppenheim, G. and Taqqu M. S. (Eds.)(2003) [*Theory and Applications of Long-Range Dependence*]{}. Birkhäuser, Boston.
Esary, J.D., Proschan, F. and Walkup, D.W. (1967) Association of random variables, with applications. Ann. Math. Statist. 38, 1466–1474.
Feller, W. (1966) [*An Introduction to Probability Theory and Its Applications*]{}, vol. 2. Wiley, New York.
Giraitis, L., Koul, H. L. and Surgailis, D. (2012) [*Large Sample Inference for Long Memory Processes.*]{} Imperial College Press, London.
Ibragimov, I.A. and Linnink, Y.V. (1971) [*Independent and Stationary Sequences of Random Variables.*]{} Wolters-Noordhoff, Groningen.
Jakubowski, A. (1997) A non-Skorohod topology on the Skorohod space. Electron. J. Probab. 2, 1–21.
Hall, P., Koul, H.L. and Turlach, B.A. (1997) Note on convergence rates of semiparametric estimators of dependence index. Ann. Statist. 25, 1725–1739.
Kaj, I. and Taqqu, M. S. (2008) Convergence to fractional Brownian motion and to the Telecom process: the integral representation approach. In: Vares, M.E. and Sidoravicius, V. (Eds.) [*An Out of Equilibrium 2.*]{} Progress in Probability, vol. 60, pp. 383–427. Birkh[ä]{}user, Basel.
Konstantopoulos, T. and Lin, S.-J. (1998) Macroscopic models for long-range dependent network traffic. Queueing Systems 28, 215–243.
Kwapień, S. and Woyczyński, W. A. (1992) [*Random Series and Stochastic Integrals: Single and Multiple.*]{} Birkhäuser, Boston.
Leipus, R. and Surgailis, D. (2003) Random coefficient autoregression, regime switching and long memory. Adv. Appl. Probab. 35, 737–754.
Leipus, R., Paulauskas, V. and Surgailis, D. (2005) Renewal regime switching and stable limit laws. J. Econometrics 129, 299-327.
Lifshits, M. (2014) [*Random Processes by Example.*]{} World Scientific, New Jersey.
Louhichi, S. and Rio, E. (2011) Functional convergence to stable Lévy motions for iterated random Lipschitz mappings. Electron. J. Probab. 16, 2452–2480.
Mikosch, T., Resnick, S., Rootzén, H. and Stegeman, A. (2002) Is network traffic approximated by stable Lévy motion or fractional Brownian motion? Ann. Appl. Probab. 12, 23–68.
Pilipauskaitė, V. and Surgailis, D. (2014) Joint temporal and contemporaneous aggregation of random-coefficient AR(1) processes. Stochastic Process. Appl. 124, 1011–1035.
Resnick, S. and Van den Berg, E. (2000) Weak convergence of high-speed traffic models. J. Appl. Prob. 37, 375–397.
Skorohod, A.V. (1956) Limit theorems for stochastic processes. Theory Probab. Appl. 1, 261–290.
Surgailis, D. (2004) Stable limits of sums of bounded functions of long memory moving averages with finite variance. Bernoulli 10, 327–355.
Taqqu, M.S. and Levy, J.B. (1986) Using renewal processes to generate long-range dependence and high variability. In: Eberlein, E. and Taqqu, M.S. (Eds.) [*Dependence in Probability and Statistics*]{}, pp. 51–72. Birkhäuser, Boston.
Taqqu, M.S., Willinger, W. and Sherman, R. (1997) Proof of the fundamental result in self-similar traffic modeling. Computer Commun. Rev. 27, 5–23.
Willinger, W., Paxon, V., Riedi, R.H. and Taqqu, M.S. (2003) Long-range dependence and data network traffic. In: Doukhan, P., Oppenheim, G. and Taqqu, M.S. (Eds.) [*Theory and Applications of Long-Range Dependence*]{}, pp. 373–407. Birkhäuser, Boston.
Wolpert, R. L. and Taqqu. M. S. (2005) Fractional Ornstein-Uhlenbeck Lévy processes and the Telecom process: upstairs and downstairs. Signal Process. 85, 1523–1545.
[^1]: UMR AGM8088, University Cergy-Pontoise.
[^2]: Nicolaus Copernicus University, Torun.
[^3]: Federal University of Rio Grande de Sul, UFRGS.
[^4]: Vilnius University.
| {
"pile_set_name": "ArXiv"
} |
There is a great need for improved numerical techniques that are able to treat two-dimensional electronic systems. Fundamental questions such as the existence of superconductivity in the 2D Hubbard and t-J models have not been resolved with present-day analytical or numerical techniques. All the commonly used numerical techniques suffer from shortcomings: exact diagonalization is limited to small lattice sizes due to the exponential growth of states with the system size.[@Dago] Quantum Monte Carlo calculations are unable to reach large fermion systems at low temperatures due to the “sign problem”.[@Loh] DMRG calculations have been very successful at treating one-dimensional systems,[@White1] but accurate results are difficult to obtain for large two-dimensional systems.
It seems most likely that major improvements towards a reliable 2D technique will be made within the DMRG context. Years of effort have resulted in no progress towards solving the fermion “sign problem”.[@Wies] It also seems unlikely that the size of available computer memory will increase fast enough to facilitate exact diagonalization calculations for large systems. Recent DMRG studies,[@White3; @Jongh] on the other hand, have reached the largest 2D systems reported to date. In this Communication I approach the 2D DMRG calculation from a new angle, which hopefully may encourage further research in this direction.
First the basis of standard 1D DMRG and previous 2D DMRG calculations will be reviewed. The source of the difficulties with previous 2D calculations is discussed. Thereafter several new 2D blocking schemes are proposed and tested, keeping only a small number of states per block. Finally a promising algorithm is discussed in more detail. All calculations are performed on the 2D antiferromagnetic Heisenberg model. The ground state parameters of this fundamental quantum-mechanical model are known to a high accuracy,[@Sand] which makes it an ideal testing ground for new numerical techniques.
The central idea in a DMRG calculation is to iteratively increase the system size, but to avoid an exponentially increasing number of states by keeping only a fixed number of the “most important” states at each iteration. In early numerical renormalization calculations the lowest eigenstates of the system were chosen to be the “important states”, but the results were discouraging.[@White4] The major breakthrough came with White’s insight [@White1] to use the density matrix to determine which states to keep. In the superblock method a number of blocks are combined together to form a superblock. The superblock is divided into two parts, the “system block” and the “environment block”. At each iteration the superblock is diagonalized and the density matrix is formed for the system block. The density matrix is diagonalized and the importance of each eigenstate is given by its eigenvalue. The states with the largest eigenvalues are kept and the rest discarded. In the next step of the iteration this system block, with a reduced number of states, will be used in forming the new superblock. Thus the number of sites increases with each iteration, while the number of states kept remains constant.
Using this basic formula many different DMRG algorithms can be defined. An algorithm is characterized by how the superblock is constructed, and by the manner in which the blocks are enlarged. The most commonly used method was proposed in White’s original work. The superblock consists of four blocks, with the two central blocks consisting of single sites, and the two end blocks being reflections of each other, see Fig. \[fig:1D\]. The system block is taken to be half the superblock, that is, one end block and an adjacent single site, here called a building block. Using the density matrix a fixed number of states are kept for the system block, which in the next iteration will be recombined with a building block to form a new system block. In this manner the system blocks are enlarged by the size of the building block (here one site) at each iteration. A variation of this method is to form the superblock out of three blocks, two end blocks and a single site in the middle. The system block is chosen as above, but the environment block consists of a single end block.
If the above procedure is iterated repeatedly one can reach arbitrarily large systems. This is called the infinite system method. In the thermodynamic limit the energy does approach a fixed point, and as the number of states kept is increased, the fixed point approaches the bulk ground state energy for the model. Usually the infinite system method has been used to measure various quantities in the middle of very large systems. In this manner results with an accuracy of up to 13 digits have been reported.[@White2]
0.5cm
The rate of convergence does depend on how the superblock is constructed. If one tries to add the building block at the outer boundary of the superblock, then the local energy will increase as one increases the system size, indicating that a good basis has not been chosen. This can easily be understood since with open periodic boundary conditions the wave function has to vanish at the boundary. If one adds a site to the boundary it is clear that some artifact will remain in the wave function as the system size is increased. This somewhat trivial example shows that one cannot construct an arbitrary blocking scheme and expect that a DMRG calculation will yield a fast convergence. This will become more evident for 2D systems.
In addition to the infinite system method discussed above, a finite system method, also introduced in White’s pioneering work, is commonly used. This algorithm is used to calculate properties of finite size systems to high accuracy. Initially the infinite system method is used to reach a system of desired length $L$. At each iteration the system block is saved, so that when a system of size $L$ is reached, system blocks of sizes 1 to $L/2-1$ are saved. Once the desired total system size has been reached the superblock size is fixed. Next the saved system blocks are used as environmental blocks while the system block size is increased until it has reached the maximum length $L-3$. Now system blocks of sizes 1 to $L-3$ are saved and these blocks can be used as environment blocks for a consequent sweep through the lattice. In this manner the basis kept in the system blocks can be iteratively improved until convergence is reached.
After this brief review of 1D DMRG calculations previous 2D calculations will be considered next. Most previous 2D calculations involve mapping the 2D lattice onto a 1D system with long-range interactions,[@White3; @Liang; @Xiang; @White5] see Fig. \[fig:2D\]. Thereafter the above 1D finite system algorithm is used. Notice that using this mapping it is not possible to use an infinite system algorithm, since one determines the size of the final lattice when doing the mapping. Also, the blocks break the symmetry of the lattice and it is generally not possible to use a reflection of the system block as the environment block. Therefore one has to use some different trick to form the environment block for the initial sweep. The two simplest options are to either leave the environment block empty, or set all long-range interactions to zero in the initial sweep. This method has, however, been able to treat the largest 2D fermion systems to date, up to sizes 11 by 16.[@White3] An alternative approach is to add a row of sites at each iteration. In this manner strips with a width of up to 6 sites and a length of 30 sites have been studied.[@Jongh]
0.5cm
Why have larger systems not been studied ? Liang and Pang [@Liang] found that for a 2D gas of free electrons, the number of states needed to maintain a certain accuracy grows exponentially with the linear system size. This convergence was also confirmed for an algorithm were a row of sites was added at each step.[@Jongh] Although no proof has been given, this statement is often referred to as most probably valid for any 2D DMRG calculation. This statement was, however, made for small finite size systems and it is not clear that it will apply to possible infinite system methods. In an infinite system DMRG calculation a fixed number of states are kept as the size of the system is increased. According to the above statement accuracy should be lost in the process. For a system with open boundary conditions the local energy decreases as the system size is increased. Furthermore, due to the variational character of the technique,[@Ostl] the DMRG energy is an upper bound on the energy of the system. Therefore accuracy would certainly be lost if the DMRG energy increased as the system size is increased, in agreement with the above statement. But if the energy decreased as the system size is increased, then the bound on the system energy is continuously improved, and in the limit of the fixed point the relative accuracy will approach a constant although only a fixed number of states is kept.
It was therefore the goal of this study to investigate whether there exist 2D blocking algorithms for which the energy decreases monotonically as the system size is increased. The algorithm should retain more of the symmetry of the lattice so that it can be used in the infinite system mode, and the fixed point studied directly. The reason for the above mapping of the 2D system to a 1D system is that it is not trivial to construct such an algorithm. Since the superblock of most symmetric 2D algorithms is bound to consist of more blocks than in the 1D case computer memory limitations will also be more severe.
In order to build up a two-dimensional lattice in a more symmetric fashion it seems likely that one has to use building blocks consisting of several sites. The simplest idea is probably to use a row of sites as building block at each iteration.[@Jongh] This method has, however, two shortcomings; it only grows the lattice in one spatial dimension, and the number of states in the added row increases exponentially with the width of the lattice.
In a first attempt to overcome these problems I divided the square 2D lattice up into three blocks, consisting of the diagonal, a triangular block below the diagonal, and the reflection of this block above the diagonal, see Fig. \[fig:2D.1\]. The lower triangular block is used as the system block, the diagonal as the building block and the reflection of the lower triangular block is used as the environment block. At each iteration the whole diagonal is thus added to the lower triangular block. In this manner one of the problems with adding just a row of sites to the system block is overcome; the lattice grows in both spatial dimensions. Furthermore, the procedure retains a high degree of symmetry and could, in principle, be used as an infinite system method. The problem is, of course, that the number of states needed to describe the exact diagonal block still increases exponentially with the linear system size. The method was, however, implemented.
0.5cm
When adding an exact diagonal to the system the energy per site decreased as the system size was increased, until the computer ran out of memory. Having passed this simple test the next problem to be addressed was the exponential increase of states in the diagonal. The natural way to avoid the exponential growth is to select only the most important states in the diagonal block by diagonalizing the density matrix for the diagonal. This was done, and at each iteration a single site was added in one corner. The local energy did, however, start to increase as the system size was increased. As in the 1D case, the reason seemed to be that a site was added at the boundary of the system. If periodic boundary conditions are used this may be a possible blocking formula, but it does not work with open boundary conditions.
A new method was therefore tried where, in analogy with the 1D method, the diagonal was divided into two blocks with the additional site added in the middle. The local energy did, however, still increase as a function of lattice size. A potential problem seemed to be that when working with square lattices, one is forced to construct the wave function for a lattice with an even number of sites from the wave function for a lattice with an odd number of sites. Lattices with odd and even numbers of sites do, however, have quite different wave functions. This issue can be avoided if one studies lattices tilted by 45 degrees. First an attempt was made to use lattices tilted by 45 degrees containing an even number of sites, see Fig. \[fig:2D.2\]. With this geometry the standard 1D DMRG technique can be used for the diagonal, with the exception that the superblock also contains the triangular blocks. The site energy did still not decrease monotonically as the system size was increased. Therefore tilted lattices with an odd number of sites were investigated, see Fig. \[fig:2D.3\], and it was found that the energy decreased monotonically as the lattice size was increased. This was the only blocking scheme found in this study for which this was the case. The fact that there exists such an algorithm is certainly encouraging, and not self-evident, as pointed out above. Since this was the most promising algorithm found in this study the results will be analyzed in more detail next.
0.5cm
0.5cm
In Fig. \[fig:ene2D\] the site energy is shown as a function of the number of iterations. Density matrices are formed both for the diagonal and the triangular blocks, and the number of states kept in these blocks are denoted $m_d$ and $m_t$ respectively. The ground state energy for the 2D Heisenberg model is -0.669437(5).[@Sand] Keeping four states in the diagonal and four blocks in the triangular block the energy levels out around -0.49. Increasing $m_d$ to 16 dramatically improves the energy to -0.57. It seems desirable to keep a higher number of states in the diagonal than in the triangular block. Next the number of states in the triangular block was increased to eight. Then only 16 states could be kept in the triangular block, and for the number of iterations that could be done the results were slightly better than the results obtained when keeping four states in the triangular block and 16 states in the diagonal block.
0.5cm
In order for a DMRG method to be useful one has to be able to keep enough states per block to reach convergence in the quantity studied. Since a 2D blocking algorithm of the kind described above contains more blocks than the traditional 1D blocking method this may prove difficult. The superblock for the above algorithm will contain $2m_d^2m_t^2$ states, and the density matrix for the triangular block will contain $2m_d^2m_t$ states. The programs used in this investigation are, however, far from optimized. Computer memory limited the present study. By using good quantum numbers, like the $z$-component of the spin for the Heisenberg model, all matrices become block diagonal. In this study complete matrices were stored, and one should be able to significantly increase the number of states kept if only the non-zero matrix blocks are stored.
The important issue is thus to study how the bulk ground state energy is approached as one further increases the number of states in the blocks. The reason for the great success of the 1D DMRG method is a very fast convergence. If the 2D calculation shows exponential or power law convergence one may be able to keep enough states to reach accurate results, but if the convergence is slower this may prove difficult.
It is also possible to make a finite system algorithm based on the above blocking procedure. In such a method one could use the infinite system method for the initial steps, saving both the triangular and the diagonal blocks. Further sweeps could use these saved blocks as environment blocks and improve the basis kept in the system blocks, in a manner analogous to the 1D method.
The algorithm presented in this Communication bears some resemblance to a “four-block method” proposed by Bursill.[@Burs] In both methods the building block, which determines the growth of the system block, does not consist of exact sites, as in the original method, but of sites with a reduced number of states.
To conclude, using a new approach I have explored the first fully two-dimensional infinite system DMRG calculation. The fixed point in two-dimensions could be explicitly studied and it was shown that there exists an algorithm for which the local energy for the Heisenberg model decreases monotonically as the system size is increased. Previous results indicated that it is necessary to keep a number of states that grows exponentially with the linear system size to maintain a certain accuracy. This does not appear to be the case with the infinite system algorithm as the fixed point is approached. The method preserves a high degree of the symmetry of the lattice and could be used as a starting point for a finite system algorithm. Further studies are necessary to verify whether the method presented here, or other similar algorithms, exhibit convergence that is fast enough to calculate properties of large two-dimensional electronic systems.
I thank Steven Girvin, Claudio Gazza and Anders Sandvik for helpful conversations. The research was supported by NSF Grants No. CDA-9601632, DMR-9714055 and DMR-9629987. I acknowledge support from Suomalainen Tiedeakatemia and I am thankful for the hospitality of the University of Virginia, where part of the work was done.
Permanent address: National High Magnetic Field Laboratory, Tallahassee, FL 32310. E. Dagotto, Rev. Mod. Phys. [**66**]{}, 763 (1994). E. Y. Loh Jr., J. E. Gubernatis, R. T. Scalettar, S. R. White, D. J. Scalapino and R. L. Sugar, Phys. Rev. B [**41**]{}, 9301 (1990). S. R. White, Phys. Rev. Lett. [**69**]{}, 2863 (1992); Phys. Rev. B [**48**]{}, 10345 (1993). A recent communication claims to have solved the sign problem in some models, see S. Chandrasekharan and U. Wiese, cond-mat/9902128. S. R. White, Phys. Rev. Lett. [**77**]{}, 3633 (1996). M. S. L. du Croo de Jongh and J. M. J. van Leeuwen, Phys. Rev. B [**57**]{}, 8494 (1998). A. W. Sandvik, Phys. Rev. B [**56**]{}, 11678 (1997). S. R. White and R. M. Noack, Phys. Rev. Lett. [**68**]{}, 3487 (1992). S. R. White and D. A. Huse, Phys. Rev. B [**48**]{}, 3844 (1993). S. Liang and H. Pang, Phys. Rev. B [**49**]{}, 9214 (1994). T. Xiang, Phys. Rev. B [**53**]{}, 10445 (1996). S. R. White and D. J. Scalapino, Phys. Rev. Lett. [**80**]{}, 1272 (1998). S. Östlund and S. Rommer, Phys. Rev. Lett. [**75**]{}, 3537 (1995). R. J. Bursill, cond-mat/9812349.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We reformulate the Gross–Pitaevskii equation with an external parabolic potential as a discrete dynamical system, by using the basis of Hermite functions. We consider small amplitude stationary solutions with a single node, called dark solitons, and examine their existence and linear stability. Furthermore, we prove the persistence of a periodic motion in a neighborhood of such solutions. Our results are corroborated by numerical computations elucidating the existence, linear stability and dynamics of the relevant solutions.'
author:
- |
Dmitry E. Pelinovsky$^{\dagger}$ and Panayotis G. Kevrekidis$^{\dagger \dagger}$\
[$^{\dagger}$ Department of Mathematics, McMaster University, Hamilton, Ontario, Canada, L8S 4K1]{}\
[$^{\dagger \dagger}$ Department of Mathematics and Statistics, University of Massachusetts, Amherst, MA 01003 ]{}
title: |
**Periodic oscillations of dark solitons\
in parabolic potentials**
---
Introduction
============
We address the Gross-Pitaevskii (GP) equation with an external parabolic potential $$\label{GP} i U_T = - \frac{1}{2} U_{XX} + \epsilon^2 X^2 U +
\sigma |U|^2 U,$$ where $U(X,T) : \mathbb{R} \times \mathbb{R}_+ \mapsto \mathbb{C}$ is decaying to zero as $|x| \to \infty$, $\epsilon \in \mathbb{R}$ is the strength of the external potential and $\sigma = 1$ ($\sigma = -1$) is normalized for the defocusing (focusing) cubic nonlinearity. This equation is of particular interest in the context of Bose-Einstein condensates, i.e., dilute alkali vapors at near-zero temperatures, where dynamics of localized dips in the ground state trapped by the magnetically induced, confining potential $V(X) = \epsilon^2 X^2$ is studied in many recent papers, see review in [@Konotop]. A question of particular interest concerns whether the localized density dips oscillate periodically near the center point $X = 0$ of the potential $V(X)$. If the motion of a localized dip is truly periodic, the frequency of periodic oscillations is to be found [@Kon05], while if the periodic oscillations are destroyed due to emission of radiation, the gradual change in the amplitude of oscillations is to be followed for sufficiently small $\epsilon$ [@PFK05]. Numerical simulations show radiation and amplitude changes if the confining parabolic potential is perturbed by a periodic potential while no radiation and time-periodic oscillations in the case of purely parabolic confinement [@proukakis].
If $\sigma = 1$, a localized dip on the ground state of the GP equation (\[GP\]) in the formal limit $\epsilon \to 0$ represents the so-called dark soliton of the defocusing nonlinear Schrödinger (NLS) equation, which is the reason why we use the term ”dark soliton" for a localized solution. Persistence and stability of a dark soliton of the defocusing NLS equation in the presence of an exponentially decaying potential $V(X)$ was studied in our previous paper [@PK07], where methods of Lyapunov–Schmidt reductions, Evans functions and the stability theory in Pontryagin space were employed. These methods can not be applied to the potential $V(X) = \epsilon^2 X^2$ since the potential deforms drastically the spectrum of the linearized problem: the continuous spectral band at $\epsilon = 0$ becomes an infinite sequence of isolated eigenvalues for $\epsilon \neq 0$. Therefore, we do not use here the limit $\epsilon \to 0$. Moreover, we transform the GP equation (\[GP\]) to the $\epsilon$-independent form $$\label{GP-zero-epsilon} i u_t = - \frac{1}{2} u_{xx} + \frac{1}{2}
x^2 u + \sigma |u|^2 u,$$ by the scaling transformation $x = \lambda X$, $t = \lambda^2 T$, and $u(x,t) = \lambda^{-1} U(X,T)$ with $\lambda = 2^{1/4}
\epsilon^{1/2}$.
Substitution $u(x,t) = e^{-\frac{i}{2} t - i \mu t} \phi(x)$ reduces equation (\[GP-zero-epsilon\]) to the second-order non-autonomous ODE $$\label{ODE} - \frac{1}{2} \phi''(x) + \frac{1}{2} x^2 \phi(x) +
\sigma \phi^3(x) = \left( \mu + \frac{1}{2} \right) \phi(x),$$ where $\phi : \mathbb{R} \mapsto \mathbb{R}$. A strong solution of the ODE (\[ODE\]) is [*said*]{} to be a dark soliton if $\phi(x)$ is odd on $x \in \mathbb{R}$, has no zeros on $x \in
\mathbb{R}_+$, and decays to zero sufficiently fast as $|x| \to
\infty$. A classification of all localized solutions of the second-order ODE (\[ODE\]) and their construction with a rigorous shooting method is suggested in recent work [@Alfimov].
Substitution of $u(x,t) = e^{-\frac{i}{2} t - i \mu t} \left[ \phi(x)
+ v(x,t) + i w(x,t) \right]$ reduces equation (\[GP-zero-epsilon\]) to the PDE system $$\begin{aligned}
\label{PDE} \left\{ \begin{array}{rcl} v_t & = & {\cal L}_- w + 2
\sigma \phi(x) v w + \sigma (v^2 + w^2) w, \\
-w_t & = & {\cal L}_+ v + \sigma \phi(x) (3 v^2 + w^2) + \sigma
(v^2 + w^2) v, \end{array} \right.\end{aligned}$$ where $(v,w) : \mathbb{R} \times \mathbb{R}_+ \mapsto
\mathbb{R}^2$ and ${\cal L}_{\pm}$ are self-adjoint Schrödinger operators in $L^2(\mathbb{R})$ $$\label{operators} \left\{ \begin{array}{rcl} {\cal L}_+ & = & -
\frac{1}{2}
\partial_x^2 + \frac{1}{2} x^2 - \frac{1}{2} - \mu + 3 \sigma
\phi^2(x), \\ {\cal L}_- & = & - \frac{1}{2} \partial_x^2 +
\frac{1}{2} x^2 - \frac{1}{2} - \mu + \sigma \phi^2(x).
\end{array} \right.$$ Solutions of the PDE (\[GP-zero-epsilon\]) are considered in space $$\label{function-space-1} {\cal H}_1(\mathbb{R}) = \{ u \in
H^1(\mathbb{R}) : \;\; x u \in L^2(\mathbb{R}) \}$$ equipped with the norm $$\label{Sobolev-space-norm} \| u \|^2_{{\cal H}_1} =
\int_{\mathbb{R}} \left( |u'(x)|^2 + (x^2 + 1) |u(x)|^2 \right)
dx.$$ Similarly, the domain of operators ${\cal L}_{\pm}$ in (\[operators\]) is defined in space $$\label{function-space-2} {\cal H}_2(\mathbb{R}) = \{ u \in
H^2(\mathbb{R}) : \;\; x^2 u \in L^2(\mathbb{R}) \}.$$
The PDE system (\[PDE\]) is a Hamiltonian system with the standard symplectic structure and the Hamiltonian function in the form $$\label{Ham-function} H = \frac{1}{2} \left( v,{\cal L}_+ v \right)
+ \frac{1}{2} \left( w,{\cal L}_-w \right) + \sigma (\phi v, v^2 +
w^2) + \frac{\sigma}{4} (v^2+w^2,v^2+w^2),$$ where $(\cdot,\cdot)$ denotes a standard inner product in $L^2(\mathbb{R})$. The Hamiltonian function $H$ is bounded if $v,w
\in {\cal H}_1(\mathbb{R})$ and is constant in time $t$. Due to the gauge invariance of the PDE (\[GP-zero-epsilon\]), there exists an additional quantity $$\label{gauge-fun} Q = 2 (\phi,v) + (v,v) + (w,w),$$ which is constant in time $t$. Global existence of solutions of the initial-value problem associated with the PDE (\[GP-zero-epsilon\]) in space $u \in {\cal H}_1(\mathbb{R})$ for all $t \in \mathbb{R}_+$ has been proved (see Proposition 2.2 in [@Carles]).
Considering the linear part of the PDE system (\[PDE\]), one can separate the variables in the form $v(x,t) = v(x) e^{\lambda t}$, $w(x,t) = w(x) e^{\lambda t}$ and obtain the linear problem $$\label{spectrum} {\cal L}_+ v = - \lambda w, \quad {\cal L}_- w =
\lambda v$$ for the spectral parameter $\lambda \in \mathbb{C}$ and the eigenvector $(v,w) \in L^2(\mathbb{R},\mathbb{C}^2)$. Fix $\mu \in
\mathbb{R}$ such that a stationary solution of the ODE (\[ODE\]) exists and $\phi \in {\cal H}_1(\mathbb{R})$. Then, the linear problem (\[spectrum\]) admits an exact solution $$\label{exact-solution-spectrum} \lambda = \pm i : \qquad v =
\phi'(x), \quad w = \mp i x \phi(x),$$ and $(v,w) \in L^2(\mathbb{R},\mathbb{C}^2)$. Additionally, for any $\mu \in \mathbb{R}$, for which the stationary solution $\phi(x)$ is smooth with respect to parameter $\mu$, the linear problem (\[spectrum\]) admits another exact solution for zero eigenvalue $\lambda = 0$ of geometric multiplicity one and algebraic multiplicity two: $$\label{exact-solution-zero} {\cal L}_- \phi(x) = 0, \qquad {\cal
L}_+ \partial_{\mu} \phi(x) = - \phi(x).$$
The main part of this work is devoted to the study of periodic solutions of the PDE system (\[PDE\]) for values of $\mu$ near $\mu = 1$. This special value corresponds to the second eigenvalue of the linear Schrödinger operator $$\label{L} {\cal L} = - \frac{1}{2} \partial_x^2 + \frac{1}{2} x^2
- \frac{1}{2}$$ with the eigenfunction $\phi(x) = \varepsilon x e^{-x^2/2}$. Here the parameter $\varepsilon$ is arbitrary in the linear problem and it parameterizes the corresponding family of stationary solutions $(\mu,\phi(x))$ of the nonlinear ODE (\[ODE\]), which bifurcates from the small-amplitude eigenmode [@Alfimov].
A periodic solution of the PDE system (\[PDE\]) bifurcates from the linear eigenmodes $v(x) = \delta e^{i \tau} \phi'(x)$ and $w(x) = \mp i \delta e^{i \tau} x \phi(x)$ corresponding to the eigenvalue pair $\lambda = \pm i$ of the linear problem (\[spectrum\]). Here $\delta$ and $\tau$ are two real-valued parameters, which are arbitrary in the linear problem and parameterize the corresponding family of periodic solutions $(v,w)$ of the PDE system (\[PDE\]). An additional parameter $\alpha$ comes from the projection of the solution $(v,w)$ to the geometric kernel of the linear problem (\[spectrum\]) with the eigenmode $v(x) = 0$ and $w(x) = \alpha \phi(x)$. Using this construction, the main result of our paper is described by the following theorem.
\[theorem-main\] Let $\varepsilon$ and $\delta$ be sufficiently small and let $\alpha,\tau$ be arbitrary. There exists a unique family of solutions of the ODE (\[ODE\]) such that $$\label{property-1} \| \phi - \varepsilon x e^{-x^2/2} \|_{{\cal
H}_1} \leq C_1 \varepsilon^3, \quad \left| \mu - 1 - \frac{3
\sigma \varepsilon^2}{\sqrt{32 \pi}} \right| \leq C_2
\varepsilon^4,$$ for some $\varepsilon$-independent constants $C_1,C_2 > 0$. There exists a family of time-periodic space-localized solutions of the PDE system (\[PDE\]) such that $(v,w) \in {\cal
H}_1(\mathbb{R},\mathbb{R}^2)$ for all $t \in \mathbb{R}$, $$\label{property-2} v\left(x,t + \frac{2\pi}{\Omega}\right) =
v(x,t), \quad w\left(x,t + \frac{2\pi}{\Omega}\right) = w(x,t),
\qquad \forall (x,t) \in \mathbb{R}^2,$$ with the bounds $$\begin{aligned}
\label{property-3} \left\| v(\cdot,t) - \delta \phi'(x)
\cos(\Omega t + \tau) \right\|_{{\cal H}_1}
& \leq & C_3 \varepsilon \delta^2, \\
\label{property-4} \left\| w(\cdot,t) - \delta \left[ x \phi(x)
\sin(\Omega t + \tau) + \alpha \phi(x) \right] \right\|_{{\cal
H}_1} & \leq & C_4 \varepsilon \delta^2,\end{aligned}$$ and $|\Omega - 1 | \leq C_5 \varepsilon^2 \delta^2$ for some ($\varepsilon,\delta$)-independent constants $C_3,C_4,C_5 > 0$.
The periodic solution of Theorem \[theorem-main\] has four free parameters $(\varepsilon,\delta,\tau,\alpha)$ which are associated with projections to the four eigenmodes (\[exact-solution-spectrum\]) and (\[exact-solution-zero\]) of the linear problem (\[spectrum\]). Parameters $\tau$ and $\alpha$ can be set to zero due to two obvious symmetries of the PDE (\[GP-zero-epsilon\]): the gauge transformation $u(x,t)
\mapsto u(x,t) e^{i \alpha}$, $\forall \alpha \in \mathbb{R}$ and the reversibility transformation $u(x,t) \mapsto \bar{u}(x,-t)$, $\forall t \in \mathbb{R}$.
Although the eigenmodes (\[exact-solution-spectrum\]) and (\[exact-solution-zero\]) persist for all $\mu \in \mathbb{R}$, existence of periodic orbits of the GP equation (\[GP-zero-epsilon\]) is only proved near $\mu = 1$. This is due to the fact that the non-resonance conditions $n \neq {\rm Im}
\lambda_m$, $\forall n,m \in \mathbb{N}$ are proved to be satisfied only in this domain, where $\lambda_m$ denote other isolated eigenvalues of the linear problem (\[spectrum\]) with ${\rm Im} \lambda_m > 0$ which are different from $\lambda = i$. Thus, resonances do not occur near the value $\mu = 1$. The construction of the periodic orbit is complicated due to the existence of translational eigenmodes associated with the double zero eigenvalue $\lambda = 0$ of the linear problem (\[spectrum\]).
Our main result is in agreement with Theorem 2.1 in [@Gus], where the Newton’s law of particle dynamics is obtained in a more general context of multi-dimensional confining potentials and general nonlinear functions of the GP equation $i \dot{\psi} = -
\nabla^2 \psi + V(x) \psi - f(\psi)$. The Newton’s law is derived for parameters $(a,p)$ of the solitary wave solution of the unperturbed equation with $V(x) \equiv 0$ and it takes the form $$\label{Newton-law} \dot{a} = 2 p, \qquad \dot{p} = - \nabla V(a).$$ Adopting our notations for the time variable and the potential function of the GP equation (\[GP-zero-epsilon\]), we rewrite the Newton’s law (\[Newton-law\]) in the explicit form $\ddot{a}
+ a = 0$, which recovers the frequency $\Omega = 1$ of the periodic solution of Theorem \[theorem-main\] in the linear approximation $\delta \to 0$.
There are several differences between results of Theorem 2.1 in [@Gus] and our Theorem \[theorem-main\]. First, the Newton’s law (\[Newton-law\]) is valid on finite time intervals and in the limit when the localization length of the stationary solution $\phi(x)$ is much smaller than the confinement length of the potential $V(x)$. This situation corresponds to the original GP equation (\[GP\]) in the limit $\epsilon \to 0$. Second, the exact periodicity is not guaranteed by the periodic solutions of the Newton’s law (\[Newton-law\]) because of the remainder terms. Lastly, the frequency $\Omega = 1$ of the Newton’s law is independent of the nonlinear function $f(\psi)$ and the nonlinear corrections in $(a,p)$. In our case, the result of Theorem \[theorem-main\] is valid for all time intervals, the exact periodicity is guaranteed, and the frequency $\Omega$ changes with parameters $\delta$. On the other hand, our results are valid in the limit $\mu \to 1$, which is far from the limit $\epsilon \to
0$ of the GP equation (\[GP\]).
Note that the oscillations of the dark solitons in the GP equation (\[GP-zero-epsilon\]) with the frequency $\Omega = 1$ were predicted from the Ehrenfest Theorem in much earlier works (see references in [@Kon05] and [@Gus]). However, it was argued that this frequency is not observed in numerical simulations of the original GP equation (\[GP\]) with $\sigma = 1$ for sufficiently small $\epsilon$ [@Kon05; @PFK05; @proukakis]. It was suggested in these works (see review in [@Konotop]) that dark solitons oscillate with a smaller frequency $\Omega =
\frac{1}{\sqrt{2}}$. We will show that both frequencies occur in the spectrum of the linear problem (\[spectrum\]) in the corresponding limit but the non-resonance conditions are not satisfied for either frequency in this limit.
Our strategy for the proof of Theorem \[theorem-main\] is to use a complete set of Hermite functions and to reformulate the evolution problem for the PDE (\[GP-zero-epsilon\]) as an infinite-dimensional discrete dynamical system for coefficients of the decomposition (Section 2). Existence of stationary solutions $\phi(x)$ of the ODE (\[ODE\]) and spectral stability of stationary solutions in the linear problem (\[spectrum\]) are studied in the framework of the discrete dynamical system (Section 3). The proof of existence of periodic solutions of the PDE system (\[PDE\]) relies on construction of periodic orbits in the discrete dynamical system (Section 4). The analytical results are verified with numerical approximations of solutions of the ODE (\[ODE\]), eigenvalues of the linear problem (\[spectrum\]) and solutions of the GP equation (\[GP-zero-epsilon\]) (Section 5). Distribution of eigenvalues of the linear problem (\[spectrum\]) in the limit $\mu \to \infty$ for $\sigma = 1$ is also analyzed with formal asymptotic methods (Appendix A).
Formalism of the discrete dynamical system
==========================================
The set of Hermite functions is defined by the standard expressions [@AS]: $$\label{Hermite-function} \phi_n(x) = \frac{1}{\sqrt{2^n n!
\sqrt{\pi}}} H_n(x) e^{-x^2/2}, \qquad \forall n = 0,1,2,3,...,$$ where $H_n(x)$ denote the Hermite polynomials, e.g. $H_0 = 1$, $H_1 = 2x$, $H_2 = 4 x^2 - 2$, $H_3 = 8 x^3 - 12 x$, etc. Since the Hermite functions are eigenfunctions of the linear Schrödinger equation $$\label{Schrodinger-Hermite} -\frac{1}{2} \phi_n''(x) + \frac{1}{2}
x^2 \phi_n(x) = \left( n + \frac{1}{2} \right) \phi_n(x), \qquad
\forall n = 0,1,2,3,...,$$ the Sturm–Liouville theory implies that the set of Hermite functions $\{ \phi_n(x) \}_{n = 0}^{\infty}$ forms an orthogonal basis in $L^2(\mathbb{R})$. The normalization coefficients in the expressions (\[Hermite-function\]) ensure that the Hermite functions have unit $L^2$-norm, such that $$\label{orthogonality} (\phi_n,\phi_m) = \delta_{n,m}, \qquad
\forall n,m = 0,1,2,3,....$$ We represent a solution $u(x,t)$ of the GP equation (\[GP-zero-epsilon\]) by the series of eigenfunctions $$\label{solution-series} u(x,t) = e^{-\frac{i}{2} t} \sum_{n =
0}^{\infty} a_n(t) \phi_n(x)$$ where the components $(a_0,a_1,a_2,...)$ form a vector ${\bf a}$ on $\mathbb{N}$. When the series representation (\[solution-series\]) is substituted to the GP equation (\[GP-zero-epsilon\]), the PDE problem is converted to the discrete dynamical system $$\label{discrete-system} i \dot{a}_n = n a_n + \sigma
\sum_{(n_1,n_2,n_3)} K_{n,n_1,n_2,n_3} a_{n_1} \bar{a}_{n_2}
a_{n_3}, \qquad \forall n = 0,1,2,3,...,$$ where $K_{n,n_1,n_2,n_3} = (\phi_n,\phi_{n_1} \phi_{n_2}
\phi_{n_3})$. We shall use a convention to avoid specifying the range of non-negative integers $(n_1,n_2,n_3)$ and $n$ in the summation signs of the dynamical system (\[discrete-system\]). Let $l^2_s(\mathbb{N})$ be a weighted discrete $l^2$-space equipped with the standard norm $$\label{discrete-norm} \| {\bf a} \|^2_{l^2_s} = \sum_{n =
0}^{\infty} (1 + n)^{2s} |a_n|^2 < \infty, \quad \forall s \in
\mathbb{R}.$$ Since the set $\{ \phi_n(x) \}_{n = 0}^{\infty}$ forms an orthonormal basis in $L^2(\mathbb{R})$, we note the isometry $\| u
\|^2_{L^2} = \| {\bf a}\|^2_{l^2}$, so that $u \in
L^2(\mathbb{R})$ if and only if ${\bf a} \in l^2(\mathbb{N})$. On the other hand, we need an equivalence between the space ${\cal
H}_1(\mathbb{R})$ for the function $u(x)$ and the space $l^2_s(\mathbb{N})$ for the vector ${\bf a}$. In addition, we need to determine the domain and range of the vector field of the discrete dynamical system (\[discrete-system\]). These results are described in Lemmas \[lemma-embedding\] and \[lemma-phase-space\].
Let $u(x) = \sum\limits_{m = 0}^{\infty} a_n \phi_n(x)$. Then $u
\in {\cal H}_1(\mathbb{R})$ if and only if ${\bf a} \in
l^2_{1/2}(\mathbb{N})$. \[lemma-embedding\]
It follows directly that $$\begin{aligned}
\| u \|^2_{{\cal H}_1} & = & \int_{\mathbb{R}} \left(
|u'(x)|^2 + (x^2 + 1) |u(x)|^2 \right) dx \\
& = & \sum_{n_1 = 0}^{\infty} \sum_{n_2 = 0}^{\infty} a_{n_1}
\bar{a}_{n_2} \int_{\mathbb{R}} \left[ \phi_{n_1}'(x)
\phi_{n_2}'(x) + ( x^2 + 1)\phi_{n_1}(x) \phi_{n_2}(x) \right] dx
\\ & = & 2 \sum_{n_1 = 0}^{\infty} \sum_{n_2 = 0}^{\infty}
a_{n_1} \bar{a}_{n_2} (1 + n_2) (\phi_{n_1},\phi_{n_2}) \\
& = & 2 \sum_{n = 0}^{\infty} (1 + n) |a_{n}|^2 = 2 \| {\bf a}
\|^2_{l^2_{1/2}},\end{aligned}$$ where the orthogonality relations (\[orthogonality\]) have been used.
[By the same method, one can prove that $u \in {\cal
H}_2(\mathbb{R})$ if and only if ${\bf a} \in l^2_1(\mathbb{N})$.]{} \[remark-embedding\]
The vector field of the dynamical system (\[discrete-system\]) maps $l^2_{1/2}(\mathbb{N})$ to $l^2_{-1/2}(\mathbb{N})$. \[lemma-phase-space\]
The vector field of the dynamical system (\[discrete-system\]) is decomposed into the linear ${\bf f}({\bf a})$ and nonlinear $\sigma {\bf g}({\bf a})$ parts, where $$f_n = n a_n, \;\; g_n = \sum\limits_{(n_1,n_2,n_3)}
K_{n,n_1,n_2,n_3} a_{n_1} \bar{a}_{n_2} a_{n_3}, \quad \forall n =
0,1,2,3,...$$ The linear unbounded part satisfies the estimate $$\label{linear-vector-field} \| {\bf f}({\bf a}) \|^2_{l^2_s} =
\sum_{n = 0}^{\infty} ( 1 + n)^{2s} n^2 |a_n|^2 \leq \| {\bf a}
\|^2_{l^2_{s+1}},$$ such that ${\bf f} : l^2_{s+1}(\mathbb{N}) \mapsto
l^2_s(\mathbb{N})$ for all $s \in \mathbb{R}$. If ${\bf a} \in
l^2_{1/2}(\mathbb{N})$, then $s = -\frac{1}{2}$. The nonlinear vector part satisfies the estimate $$\begin{aligned}
\| {\bf g}({\bf a}) \|^2_{l^2_s} & = & \sum_{n = 0}^{\infty} ( 1 +
n)^{2s} \sum_{(n_1,n_2,n_3)} \sum_{(m_1,m_2,m_3)}
K_{n,n_1,n_2,n_3} K_{n,m_1,m_2,m_3} a_{n_1} \bar{a}_{n_2} a_{n_3}
\bar{a}_{m_1} a_{m_2} \bar{a}_{m_3} \\
& = & \sum_{n = 0}^{\infty} ( 1 + n)^{2s} \left| \left( \phi_n u,
|u|^2 \right) \right|^2 \leq \left( \sum_{n = 0}^{\infty} ( 1 +
n)^{2s} \| u \phi_n \|^2_{L^2} \right) \| u \|^4_{L^4} \\ & \leq &
\left( \sum_{n = 0}^{\infty} ( 1 + n)^{2s} \| \phi_n \|^2_{L^4}
\right) \| u \|^6_{L^4},\end{aligned}$$ where $u(x) = \sum\limits_{n = 0}^{\infty} a_n \phi_n(x)$ and all $\phi_n(x)$ are real-valued. By the main theorem of [@Freud], there exists a constant $C > 0$ such that $$\label{estimate-2} \| \phi_n \|^4_{L^4} \leq C
\frac{\log(1+n)}{\sqrt{1 + n}}, \qquad \forall n = 0,1,2,...$$ Therefore, the series $\sum_{n = 0}^{\infty} ( 1 + n)^{2s} \|
\phi_n \|^2_{L^4}$ converges for all $s < -\frac{3}{8}$. The value $s = -\frac{1}{2}$ belongs to this interval. Finally, by the Sobolev embedding and Poincare inequality [@Adams], there are constants $C, \tilde{C} > 0$ such that $$\| u \|^4_{L^4} \leq C \| (u^2)' \|^2_{L^2} \leq 4 C \| u
\|^2_{L^{\infty}} \| u' \|^2_{L^2} \leq \tilde{C} \| u \|^4_{H^1}
\leq \tilde{C} \| u \|^4_{{\cal H}_1}.$$ Since the norm in ${\cal H}_1(\mathbb{R})$ for the function $u(x)$ is equivalent to the norm in $l^2_{1/2}(\mathbb{Z})$ for the vector ${\bf a}$ by Lemma \[lemma-embedding\], the estimate for the nonlinear vector field is completed by $$\label{nonlinear-vector-field} \| {\bf g}({\bf a})
\|^2_{l^2_{-1/2}} \leq C_0 \| u \|^6_{L^4} \leq \tilde{C}_0 \|
{\bf a} \|^6_{l^2_{1/2}},$$ for some $C_0, \tilde{C}_0 > 0$. The interpolation argument for the bounds (\[linear-vector-field\]) and (\[nonlinear-vector-field\]) concludes the proof that the nonlinear vector field ${\bf f}({\bf a}) + \sigma {\bf g}({\bf
a})$ maps $l^2_{1/2}(\mathbb{N})$ to $l^2_{-1/2}(\mathbb{N})$.
\[theorem-equivalence\] The discrete dynamical system (\[discrete-system\]) is globally well-posed in the phase space ${\bf a} \in l^2_{1/2}(\mathbb{N})$.
By Proposition 2.2 in [@Carles], the GP equation (\[GP-zero-epsilon\]) is globally well-posed in the phase space $u \in {\cal H}_1(\mathbb{R})$. By Lemma \[lemma-embedding\], the trajectory $u(t) \in {\cal H}_1(\mathbb{R})$ is equivalent to the trajectory ${\bf a}(t) \in l^2_{1/2}(\mathbb{N})$ on $t \in
\mathbb{R}$. By Lemma \[lemma-phase-space\], the vector field of the discrete dynamical system (\[discrete-system\]) is well-defined on $l^2_{1/2}(\mathbb{N}) \subset l^2(\mathbb{N})$, where it is equivalent to the vector field of the GP equation (\[GP-zero-epsilon\]) by virtue of standard orthogonal projections.
Existence and stability of stationary solutions
===============================================
Stationary solutions of the dynamical system (\[discrete-system\]) take the form ${\bf a}(t) = {\bf A} e^{-i
\mu t}$, where ${\bf A}$ is a time-independent vector and $\mu$ is a parameter of the solution. If ${\bf A} \in
l^2_{1/2}(\mathbb{N})$ and $\phi(x) = \sum\limits_{n=0}^{\infty}
A_n \phi_n(x)$, then $\phi \in {\cal H}_1(\mathbb{R})$ is a stationary solution of the GP equation (\[GP-zero-epsilon\]), that is $\phi(x)$ satisfies the ODE (\[ODE\]). The vector ${\bf
A}$ is found as a root of the infinite-dimensional cubic vector field ${\bf F} : l^2_{1/2}(\mathbb{N}) \times \mathbb{R} \mapsto
l^2_{-1/2}(\mathbb{N})$, where the $n$-th component of ${\bf
F}({\bf A},\mu)$ is given by $$\label{stationary-solution} F_n = (\mu - n) A_n - \sigma
\sum_{(n_1,n_2,n_3)} K_{n;n_1,n_2,n_3} A_{n_1} \bar{A}_{n_2}
A_{n_3} = 0, \qquad \forall n = 0,1,2,...$$ The Jacobian operator $D_{\bf A} {\bf F}({\bf 0},\mu)$ is a diagonal matrix with entries $\mu - n$ and it admits a one-dimensional kernel if $\mu = n_0$ for any non-negative integer $n_0$. The corresponding eigenvector is ${\bf e}_{n_0}$, the unit vector in $l^2(\mathbb{N})$. By the local bifurcation theory [@GS], each eigenvector of $D_{\bf A} {\bf F}({\bf 0},n_0)$ can be uniquely continued in a local neighborhood of the point ${\bf A} = {\bf 0} \in l^2_{1/2}(\mathbb{N})$ and $\mu = n_0 \in
\mathbb{R}$. We are particularly interested in the second eigenvalue $n_0 = 1$, which corresponds to the [*dark soliton*]{} $\phi(x)$ with a single zero (node) at $x = 0$. (Other bifurcations of stationary localized solutions $\phi(x)$ are considered in [@Alfimov].) Details of this bifurcation are given in the following proposition.
\[lemma-stationary\] Consider real-valued roots $({\bf
A},\mu)$ of the vector field ${\bf F}({\bf A},\mu)$ such that ${\bf A} \in l^2_{1/2}(\mathbb{N})$. There exists a unique family of solutions near $\mu = 1$ parameterized by $\varepsilon$ such that $$\label{comparison1} \| {\bf A} - \varepsilon {\bf e}_1
\|_{l^2_{1/2}} \leq C_1 \varepsilon^3, \qquad \left| \mu - 1 -
\frac{3 \sigma \varepsilon^2}{\sqrt{32 \pi}} \right| \leq C_2
\varepsilon^4,$$ for some $\varepsilon$-independent constants $C_1, C_2 > 0$ and sufficiently small $\varepsilon$. Moreover, if $\sigma \neq 0$, the solution $({\bf A},\mu)$ is smooth with respect to $\varepsilon$ for sufficiently small $\varepsilon$ and $\frac{d}{d
\mu} Q({\bf A}) \neq 0$, where $Q({\bf A}) = \| {\bf A}
\|^2_{l^2}$.
Both ${\bf F}({\bf A},\mu)$ and $D_{\bf A} {\bf F}({\bf A},\mu)$ are continuous in a local neighborhood of ${\bf A} = {\bf 0} \in
l^2_{1/2}(\mathbb{N})$ and $\mu = 1 \in \mathbb{R}$. At the point ${\bf A} = {\bf 0}$ and $\mu = 1$, the operator has a one-dimensional kernel with the eigenvector ${\bf e}_1 \in
l^2(\mathbb{N})$. By using the method of Lyapunov–Schmidt reductions [@GS], we set ${\bf A} = \varepsilon \left[ {\bf
e}_1 + \tilde{\bf A} \right]$ and $\mu = 1 + \tilde{\mu}$, where $\tilde{\bf A}$ is an orthogonal complement of ${\bf e}_1$ in $l^2(\mathbb{N})$ such that $\tilde{A}_1 = 0$. The orthogonal projection of equation (\[stationary-solution\]) to ${\bf e}_1$ gives a bifurcation equation for $\tilde{\mu}$ $$\begin{aligned}
\tilde{\mu} = \sigma \varepsilon^2 \left[ K_{1;1,1,1} + 3
\sum_{n_1} K_{1;1,1,n} \tilde{A}_{n_1} + 3 \sum_{(n_1,n_2)}
K_{1;1,n_1,n_2} \tilde{A}_{n_1} \tilde{A}_{n_2}+
\sum_{(n_1,n_2,n_3)} K_{1;n_1,n_2,n_3} \tilde{A}_{n_1}
\tilde{A}_{n_2} \tilde{A}_{n_3} \right],\end{aligned}$$ where the index for $(n_1,n_2,n_3)$ in the summation signs runs on the set $\{ 0,2,3,... \}$. Let $P$ be an orthogonal projection from $l^2(\mathbb{N})$ to the orthogonal complement of ${\bf
e}_1$. Then the inverse of $P D_{\bf A} {\bf F}({\bf 0},1) P$ exists and is a bounded operator from $l^2_{1/2}(\mathbb{N})$ to $l^2_{1/2}(\mathbb{N})$. By the Implicit Function Theorem, there exists a unique smooth solution $\tilde{\bf A}$ in the neighborhood of $\tilde{\bf A} = {\bf 0} \in l^2_s(\mathbb{N})$ such that $\| \tilde{\bf A} \|_{l^2_{1/2}} \leq C_1 \varepsilon^2$ for some $C_1 > 0$. By the Implicit Function Theorem, there exists a unique smooth solution $\tilde{\mu}$ of the bifurcation equation in the neighborhood of $\tilde{\mu} = 0$ such that $|\tilde{\mu} -
\varepsilon^2 \sigma K_{1,1,1,1} | \leq C_2 \varepsilon^4$ for some $C_2 > 0$. The value $K_{1,1,1,1} = \| \phi_1\|^4_{L^4} =
\frac{3}{\sqrt{32 \pi}}$ is computed in Table I. Since $Q({\bf A})
= \| {\bf A} \|^2_{l^2} = \varepsilon^2 + {\rm O}(\varepsilon^4)$ and $\mu - 1 = \frac{3 \sigma \varepsilon^2}{\sqrt{32 \pi}} + {\rm
O}(\varepsilon^4)$, then $\frac{d}{d \mu} Q({\bf A}) \neq 0$ near $\mu = 1$ for $\sigma \neq 0$.
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
$\phantom{t}$ $n = 0$ $n = 1$ $n = 2$ $n = 3$ $n = 4$ $n = 5$
--------------- --------------------------- --------------- ----------------------------- ------------------------------- ------------------------------------- -----------------------------------
$K_{n,n,n,n}$ $\frac{1}{\sqrt{2\pi}}$ $\frac{3}{4 $\frac{41}{64 \sqrt{2\pi}}$ $\frac{147}{256 \sqrt{2\pi}}$ $\frac{8649}{16384 \sqrt{2\pi}}$ $\frac{32307}{65536 \sqrt{2\pi}}$
\sqrt{2\pi}}$
$K_{1,n,n,1}$ $\frac{1}{2 \sqrt{2\pi}}$ $\frac{3}{4 $\frac{7}{16 \sqrt{2\pi}}$ $\frac{11}{32 \sqrt{2\pi}}$ $\frac{75}{256 \sqrt{2\pi}}$ $\frac{133}{512 \sqrt{2\pi}}$
\sqrt{2\pi}}$
$K_{0,1,1,n}$ $\frac{1}{2 \sqrt{2\pi}}$ $0$ $ $0$ $-\frac{3 \sqrt{3}}{32 \sqrt{\pi}}$ $0$
\frac{1}{8 \sqrt{\pi}}$
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
[**Table I:**]{} Numerical values for $K_{n,n,n,n} = \| \phi_n
\|^4_{L^4}$, $K_{1,n,n,1} = (\phi_1^2,\phi_n^2)$, and $K_{0,1,1,n}
= (\phi_0 \phi_n,\phi_1^2)$.
Let $({\bf A},\mu)$ be a real-valued root of the nonlinear vector field (\[stationary-solution\]) such that ${\bf
A} \in l^2_{1/2}(\mathbb{N})$. Spectral stability of the stationary solution is studied with the expansion $$\label{linear_stability} {\bf a}(t) = e^{-i \mu t} \left[ {\bf A}
+ \left( {\bf B} - {\bf C}\right) e^{i \Omega t} + \left(
\bar{\bf B} + \bar{\bf C}\right) e^{- i \bar{\Omega} t} + {\rm
O}(\|{\bf B}\|^2 + \| {\bf C} \|^2) \right],$$ where the spectral parameter $\Omega \in \mathbb{C}$ and the eigenvector $({\bf B},{\bf C}) \in l^2(\mathbb{N},\mathbb{C}^2)$ satisfy the linear problem $$\label{eigenvalue} L_+ {\bf B} = \Omega {\bf C}, \qquad L_- {\bf
C} = \Omega {\bf B},$$ associated with matrix operators $L_{\pm}$. Their $n$-th components are defined in the form $$\begin{aligned}
\label{eigenvalue-1-2} \left\{ \begin{array}{ccc} (L_+ {\bf B})_n
& = & (n - \mu) B_n + 3 \sigma
\sum_{n_1} V_{n,n_1} B_{n_1}, \\
(L_- {\bf C})_n & = & (n - \mu ) C_n + \sigma \sum_{n_1} V_{n,n_1}
C_{n_1}, \end{array} \right. \quad \forall n = 0,1,2,3,...,\end{aligned}$$ where $V_{n,n_1} = \sum\limits_{(n_2,n_3)} K_{n,n_1,n_2,n_3}
A_{n_2} A_{n_3}$. We have used here the symmetry of the coefficients $K_{n,n_1,n_2,n_3}$ with respect to the interchange of $(n_1,n_2,n_3)$.
\[lemma-operators\] Let $({\bf A}, \mu)$ be a real-valued root of the vector field ${\bf F}({\bf A},\mu)$ such that ${\bf A} \in
l^2_{1/2}(\mathbb{N})$. Operators $L_+$ and $L_-$ admit closed self-adjoint extensions in $l^2(\mathbb{N})$ with the domain in $l^2_1(\mathbb{N})$.
The diagonal unbounded part of $L_{\pm}$ maps $l^2_1(\mathbb{N})$ to $l^2(\mathbb{N})$. We need to show that the non-diagonal part of $L_{\pm}$ represents a bounded perturbation from $l^2(\mathbb{N})$ to $l^2(\mathbb{N})$ if ${\bf A} \in
l^2_{1/2}(\mathbb{N})$. This is done by using the same ideas as in the proof of Lemma \[lemma-phase-space\]: $$\begin{aligned}
\sum_{n=0}^{\infty} \left| \sum_{n_1} V_{n,n_1} B_{n_1} \right|^2
& = & \sum_{n=0}^{\infty} \sum_{(n_1,n_2,n_3)}
\sum_{(m_1,m_2,m_3)} K_{n,n_1,n_2,n_3} K_{n,m_1,m_2,m_3} A_{n_2}
A_{n_3} A_{m_2} A_{m_3} B_{n_1} \bar{B}_{m_1} \\
& = & \sum_{n = 0}^{\infty} |(\phi_n, u^2 v)|^2 = \| u^2 v
\|^2_{L^2} \leq \| u \|^4_{L^{\infty}} \| v \|^2_{L^2} \leq C^4 \|
u \|^4_{{\cal H}_1} \sum_{n = 0}^{\infty} |B_n|^2,\end{aligned}$$ where $u(x) = \sum_{n=0}^{\infty} A_n \phi_n(x)$ and $v(x) =
\sum_{n=0}^{\infty} B_n \phi_n(x)$.
[The result of Lemma \[lemma-operators\] is obvious from the equivalence between the space ${\cal H}_2(\mathbb{R})$ for the function $v(x)$ and the space $l^2_1(\mathbb{N})$ for the vector ${\bf B}$, see Remark \[remark-embedding\]. We recall that the differential operators ${\cal L}_{\pm}$ given by (\[operators\]) are defined on the domain ${\cal H}_2(\mathbb{R})$ and the matrix operators $L_{\pm}$ given by (\[eigenvalue-1-2\]) represent the action of differential operators on the basis of Hermite functions in ${\cal H}_2(\mathbb{R})$. ]{}
The linear problem (\[eigenvalue\]) has eigenvalue $\Omega = 0$ of geometric multiplicity one and algebraic multiplicity two due to the exact solution $$\label{null-eigenvalue} L_- {\bf A} = {\bf 0}, \qquad L_+
\partial_{\mu} {\bf A} = {\bf A},$$ where the smoothness of ${\bf A}$ with respect to $\mu$ near $\mu
= 1$ is guaranteed by Proposition \[lemma-stationary\].
When ${\bf A} = {\bf 0}$ and $\mu = 1$, the spectrum of the eigenvalue problem (\[eigenvalue\]) is known in the explicit form. It consists of eigenvalues $\Omega = 0$ and $\Omega = \pm 1$ of geometric and algebraic multiplicities two and simple eigenvalues $\Omega = \pm m$ for all $m = 2,3,...$. The double zero eigenvalue persists for any $\varepsilon$ according to the exact solution (\[null-eigenvalue\]), stemming from the underlying $U(1)$ invariance of the system. Splitting of all other eigenvalues in a local neighborhood of ${\bf A} = {\bf 0}$ and $\mu = 1$ is described by the following proposition.
Let $({\bf A},\mu)$ be defined by Proposition \[lemma-stationary\] for sufficiently small $\varepsilon$. Non-zero eigenvalues of the linear problem (\[eigenvalue\]) form a set $\{ \pm \Omega_m \}_{m = 0}^{\infty}$ of simple real symmetric eigenvalue pairs, such that $$\begin{aligned}
\label{eigenvalue-expansion-1} \left| \Omega_0 - 1 \right| \leq
C_0 \varepsilon^4, \qquad \left| \Omega_1 - 1 + \frac{
\varepsilon^2 \sigma}{8 \sqrt{2 \pi}} \right| \leq C_1
\varepsilon^4\end{aligned}$$ and $$\begin{aligned}
\label{eigenvalue-expansion-3} \left| \Omega_m - m + \varepsilon^2
\sigma \left( K_{1,1,1,1} - 2 K_{m+1,1,1,m+1} \right) \right| \leq
C_m \varepsilon^4, \quad \forall m = 2,3,....\end{aligned}$$ for some $\varepsilon$-independent constants $C_0,C_1,C_m > 0$. \[lemma-eigenvalue\]
Since the essential spectrum of the matrix operators $L_{\pm}$ is empty and the potential terms are bounded perturbations to the unbounded diagonal terms, isolated eigenvalues split according to the regular perturbation theory [@Kato]. The formal power series expansion for a simple eigenvalue $\Omega = m = 2,3,...$ is defined by $$\left\{ \begin{array}{rcl} {\bf B} & = & {\bf e}_{m+1} +
\varepsilon^2 \tilde{\bf B} + {\rm O}(\varepsilon^4), \\ {\bf C} &
= & {\bf e}_{m+1} + \varepsilon^2 \tilde{\bf C} + {\rm
O}(\varepsilon^4), \\ \Omega & = & m + \varepsilon^2
\tilde{\Omega} + {\rm O}(\varepsilon^4).
\end{array} \right.$$ Projections to the component $n = m+1$ lead to a linear system at the leading order ${\rm O}(\varepsilon^2)$ $$\begin{aligned}
\left\{ \begin{array}{rcl} m \left( \tilde{B}_{m+1} -
\tilde{C}_{m+1} \right) & = & \sigma \left[ K_{1,1,1,1}
- 3 K_{m+1,1,1,m+1} \right] + \tilde{\Omega} \\
m \left( \tilde{C}_{m+1} - \tilde{B}_{m+1} \right) & = & \sigma
\left[ K_{1,1,1,1} - K_{m+1,1,1,m+1} \right] +\tilde{\Omega}.
\end{array} \right.\end{aligned}$$ The linear system has a solution if and only if $\tilde{\Omega} =
\sigma \left( 2 K_{m+1,1,1,m+1} - K_{1,1,1,1}\right)$. Persistence of the eigenvalue by the perturbation theory results in the expansion (\[eigenvalue-expansion-3\]). The power series expansion for the double eigenvalue $\Omega = 1$ is defined by $$\begin{aligned}
\left\{ \begin{array}{rcl} {\bf B} & = & \alpha {\bf e}_0 + \beta
{\bf e}_2 + \varepsilon^2 \tilde{\bf B} + {\rm O}(\varepsilon^4),
\\ {\bf C} & = & - \alpha {\bf e}_{0} + \beta {\bf e}_2 +
\varepsilon^2 \tilde{\bf C} + {\rm O}(\varepsilon^4), \\
\Omega & = & 1 + \varepsilon^2 \tilde{\Omega} + {\rm
O}(\varepsilon^4),
\end{array} \right.\end{aligned}$$ where $(\alpha,\beta)$ are arbitrary parameters. Projections to the components $n = 0$ and $n = 2$ leads to a linear system at the leading order ${\rm O}(\varepsilon^2)$ $$\begin{aligned}
\left\{ \begin{array}{rcl} \left( \tilde{B}_{0} + \tilde{C}_{0}
\right) & = & \sigma \left[ 3 K_{0;1,1,0}
\alpha + 3 K_{0,1,1,2} \beta - K_{1,1,1,1} \alpha \right] + \tilde{\Omega} \alpha \\
- \left( \tilde{C}_{0} + \tilde{B}_0 \right) & = & \sigma \left[
K_{0,1,1,0} \alpha - K_{0,1,1,2} \beta - K_{1,1,1,1} \alpha
\right] + \tilde{\Omega} \alpha \\
\left( \tilde{B}_2 - \tilde{C}_2 \right) & = & \sigma \left[
K_{1,1,1,1} \beta - 3
K_{2,1,1,0} \alpha - 3 K_{2,1,1,2} \beta \right] + \tilde{\Omega} \beta \\
\left( \tilde{C}_2 - \tilde{B}_2 \right) & = & \sigma \left[
K_{1,1,1,1} \beta + K_{2,1,1,0} \alpha - K_{2,1,1,2} \beta \right]
+ \tilde{\Omega} \alpha \end{array} \right.\end{aligned}$$ The linear system has a solution if and only if $(\alpha,\beta)$ satisfies a homogeneous system $$\begin{aligned}
\label{hom-system-2} \left\{ \begin{array}{rcl} \sigma \left(
K_{1,1,1,1} \alpha - 2 K_{0,1,1,0} \alpha -
K_{0,1,1,2} \beta \right) & = & \tilde{\Omega} \alpha, \\
\sigma \left( -K_{1,1,1,1} \beta + K_{2,1,1,0} \alpha + 2
K_{2,1,1,2} \beta \right) & = & \tilde{\Omega} \beta.
\end{array} \right.\end{aligned}$$ The homogeneous system for $(\alpha,\beta)$ has a non-zero solution if and only if $\tilde{\Omega}$ satisfies a quadratic equation, roots of which are given by $$\tilde{\Omega} = \sigma \left( K_{2,1,1,2} - K_{0,1,1,0} \pm
\sqrt{(K_{1,1,1,1} - K_{0,1,1,0} - K_{2,1,1,2})^2 - K_{0,1,1,2}^2}
\right).$$ It follows from Table I that $\sqrt{(K_{1,1,1,1} - K_{0,1,1,0} -
K_{2,1,1,2})^2 - K_{0,1,1,2}^2} = \frac{1}{16 \sqrt{2 \pi}}$ and $K_{2,1,1,2} - K_{0,1,1,0} = -\frac{1}{16 \sqrt{2 \pi}}$. Persistence of the eigenvalues by the perturbation theory results in the expansion (\[eigenvalue-expansion-1\]).
\[corollary-signatures\] Let $[{\bf B}_m,{\bf C}_m]^T$ be an eigenvector of the linear problem (\[eigenvalue\]) for the eigenvalue $\Omega_m \in \mathbb{R}_+$ for any $m = 0,1,2,3...$ in Proposition \[lemma-eigenvalue\]. For sufficiently small $\varepsilon$, the eigenvalue $\Omega_0$ has positive signature of $\langle {\bf B}_0, L_+ {\bf B}_0 \rangle$, the eigenvalue $\Omega_1$ has negative signature of $\langle {\bf B}_1, L_+ {\bf
B}_1 \rangle$, while all other eigenvalues $\Omega_m$ with $m =
2,3,...$ have positive signature of $\langle {\bf B}_m, L_+ {\bf
B}_m \rangle$, where $\langle \cdot, \cdot \rangle$ denotes a standard inner product in $l^2(\mathbb{N})$.
In the case $\tilde{\Omega} = 0$, the homogeneous system (\[hom-system-2\]) for $(\alpha,\beta)$ has a one-parameter family of solutions with $\beta = - \sqrt{2} \alpha$, such that $\langle {\bf B}_0, L_+ {\bf B}_0 \rangle = - |\alpha|^2 +
|\beta|^2 + {\rm O}(\varepsilon^2) > 0$ for sufficiently small $\varepsilon$. In the case $\tilde{\Omega} \neq 0$, the homogeneous system (\[hom-system-2\]) for $(\alpha,\beta)$ has a one-parameter family of solutions with $\alpha = - \sqrt{2}
\beta$, such that $\langle {\bf B}_1, L_+ {\bf B}_1 \rangle = -
|\alpha|^2 + |\beta|^2 + {\rm O}(\varepsilon^2) < 0$ for sufficiently small $\varepsilon$. In the case of other eigenvalues, it is obvious from the proof of Proposition \[lemma-eigenvalue\] that $\langle {\bf B}_m, L_+ {\bf B}_m
\rangle = m + {\rm O}(\varepsilon^2)$ for $m = 2,3,...$.
[The double zero eigenvalue is associated with the expansions $$\label{expansion-zero-eigenvalue} {\bf A} = \varepsilon {\bf e}_1
+ {\rm O}(\varepsilon^3), \qquad
\partial_{\mu} {\bf A} = \frac{\sqrt{32 \pi}}{3 \sigma \varepsilon}
{\bf e}_1 + {\rm O}(\varepsilon),$$ where $\varepsilon$ is sufficiently small. As a result, $\langle
{\bf A}, \partial_{\mu} {\bf A} \rangle = \frac{\sqrt{32 \pi}}{3
\sigma} + {\rm O}(\varepsilon^2)$.]{} \[remark-eigenvalues\]
\[lemma-diagonalization\] Let $\varepsilon$ be sufficiently small. Let $\{ [{\bf B}_m, {\bf C}_m]^T \}_{m = 0}^{\infty}$ be a set of real-valued eigenvectors of the linear problem (\[eigenvalue\]) for the set of positive eigenvalues $\{\Omega_m\}_{m=0}^{\infty}$. The set of eigenvectors is symplectically orthogonal such that $$\label{orthogonal-projections} \langle {\bf B}_{m'}, {\bf C}_m
\rangle = 0, \;\; \forall m' \neq m \qquad \langle {\bf B}_{m},
{\bf C}_m \rangle \neq 0, \;\; \forall m = 0,1,2,3...$$ In addition, two eigenvectors $\{ [{\bf 0},{\bf A}]^T,
[\partial_{\mu} {\bf A},{\bf 0}]^T \}$ for the double zero eigenvalue $\Omega = 0$ are symplectically orthogonal to other eigenvectors and $\langle {\bf A},\partial_{\mu} {\bf A} \rangle
\neq 0$. The set of eigenvectors $$\label{complete-set} \{ [{\bf B}_m,{\bf C}_m]^T \}_{m =
0}^{\infty} \oplus \{ [{\bf B}_m,-{\bf C}_m]^T \}_{m = 0}^{\infty}
\oplus \{ [{\bf 0},{\bf A}]^T, [\partial_{\mu} {\bf A},{\bf 0}]^T
\}$$ is a basis in $l^2(\mathbb{N},\mathbb{R}^2)$ which is orthogonal with respect to the symplectic projections (\[orthogonal-projections\]).
All eigenvalues $\{ \Omega_m \}_{m = 0}^{\infty}$ are positive and simple for sufficiently small $\varepsilon$ by Proposition \[lemma-eigenvalue\]. Since $L_{\pm}$ are self-adjoint in $l^2(\mathbb{N})$ and $\Omega_m$ is a real eigenvalue, then the eigenvector $[{\bf B}_m,{\bf C}_m]^T$ of the linear problem (\[eigenvalue\]) can be chosen to be real-valued. The orthogonality relations (\[orthogonal-projections\]) follow by direct computations from the linear problem (\[eigenvalue\]) for distinct eigenvalues $\Omega_{m'} \neq \Omega_m$ for all $m' \neq
m$. Values of $\langle {\bf B}_{m}, {\bf C}_m \rangle$ are proportional to the values of $\langle {\bf B}_{m}, L_+ {\bf B}_m
\rangle$ for $\Omega_m \neq 0$ and they are non-zero for sufficiently small $\varepsilon$ by Corollary \[corollary-signatures\]. The value of $\langle {\bf
A},\partial_{\mu} {\bf A} \rangle = \frac{1}{2} \frac{d}{d \mu}
Q({\bf A})$ is non-zero for sufficiently small $\varepsilon$ by Proposition \[lemma-stationary\]. By the proof of Proposition \[lemma-eigenvalue\] and Remark \[remark-eigenvalues\], the eigenvectors of the set (\[complete-set\]) are represented for sufficiently small $\varepsilon$ by the standard basis $\{ {\bf
e}_m \}_{m = 0}^{\infty} \oplus \{ {\bf e}_m \}_{m = 0}^{\infty}$ perturbed by a bounded perturbation in $l^2(\mathbb{N})$ of the order ${\rm O}(\varepsilon^2)$. Also $$\label{asymptotic-distribution-eigenvalues} \Omega_m = m + {\rm
O}(\varepsilon^2), \quad \langle {\bf B}_m, {\bf C}_m \rangle =
\frac{\langle {\bf B}_{m}, L_+ {\bf B}_m \rangle}{\Omega_m} = 1 +
{\rm O}(m^{-1} \varepsilon^2), \quad \forall m = 2,3,...,$$ for sufficiently small $\varepsilon$, uniformly in $m$. Since no other eigenvalues exist, the set of linearly independent eigenvectors (\[complete-set\]) is complete in $l^2(\mathbb{N},\mathbb{R}^2)$. According to the Banach Theorem for non-self-adjoint operators, the set is a basis if and only if the spectral projections are bounded from below by a non-zero constant in the limit $m \to \infty$, which follows from the uniform asymptotic distribution (\[asymptotic-distribution-eigenvalues\]). Therefore, the set (\[complete-set\]) is a basis in $l^2(\mathbb{N},\mathbb{R}^2)$.
\[lemma-persistence\] Fix $\varepsilon \neq 0$ sufficiently small. Simple positive eigenvalues of the set $\{\Omega_m \}_{m =
0}^{\infty}$ lie in the intervals $$\label{distribution-1} \sigma > 0: \;\; \Omega_0 = 1 \;\;
\mbox{and} \;\; m - C_m^- \varepsilon^2 < \Omega_m < m, \;\;
\forall m \in \mathbb{N}$$ and $$\label{distribution-2} \sigma < 0: \;\; \Omega_0 = 1 \;\;
\mbox{and} \;\; m < \Omega_m < m + C_m^+ \varepsilon^2, \;\;
\forall m \in \mathbb{N}$$ for some $\varepsilon$-independent constants $C_m^{\pm} > 0$.
The eigenvalue $\Omega_0 = 1$ persists for any $\varepsilon \in
\mathbb{R}$ due to equivalence of the linear eigenvalue problems (\[spectrum\]) and (\[eigenvalue\]) for $\phi \in {\cal
H}_1(\mathbb{R})$ and ${\bf A} \in l^2_{1/2}(\mathbb{N})$ and the existence of the exact solution (\[exact-solution-spectrum\]) of the linear eigenvalue problem (\[spectrum\]). The corresponding eigenvector $({\bf B},{\bf C})$ of the linear eigenvalue problem (\[eigenvalue\]) is found from the series representation $$\phi'(x) = \sum\limits_{n=0}^{\infty} B_n \phi_n(x), \qquad -x
\phi(x) = \sum\limits_{n=0}^{\infty} C_n \phi_n(x).$$ The eigenvalue $\Omega_1$ satisfies the bounds (\[distribution-1\])–(\[distribution-2\]) due to the explicit bound (\[eigenvalue-expansion-1\]). We use the bound (\[eigenvalue-expansion-3\]) to prove the bounds (\[distribution-1\])–(\[distribution-2\]) for eigenvalues $\Omega_m$ for all $m = 2,3,...$. The values of $K_{1,1,1,1} - 2
K_{m+1,1,1,m+1}$ are positive for the first values of $m =
3,4,...$ as follows from Table I, e.g. $$K_{1,1,1,1} - 2 K_{3,1,1,3} = \frac{1}{16 \sqrt{2\pi}}, \;\;
K_{1,1,1,1} - 2 K_{4,1,1,4} = \frac{21}{128 \sqrt{2\pi}}, \;\;
K_{1,1,1,1} - 2 K_{5,1,1,5} = \frac{59}{256 \sqrt{2 \pi}}.$$ Note that the positive numerical values are monotonically increasing. According to the main theorem in [@Freud], the sequence $\{ \| \phi_n\|_{L^4}\}_{n \in \mathbb{N}}$ is monotonically decreasing to zero with the bound (\[estimate-2\]). Since $K_{m+1,1,1,m+1} \leq \| \phi_1
\|^2_{L^4} \| \phi_{m+1} \|^2_{L^4}$, then $$K_{1,1,1,1} - 2 K_{m+1,1,1,m+1} \geq \| \phi_1 \|^2_{L^4} \left(
\| \phi_1 \|^2_{L^4} - 2 \| \phi_{m+1} \|^2_{L^4} \right), \qquad
\forall m = 2,3,...$$ Since $\| \phi_{m+1} \|^2_{L^4}$ decays monotonically to zero as $m \to \infty$, there exists $M$ sufficiently large, such that the lower bound above is strictly positive for $m \geq M$.
Existence of periodic solutions
===============================
Let $({\bf A},\mu)$ be a real-valued root of the nonlinear vector field ${\bf F}({\bf A},\mu)$ such that ${\bf A} \in
l^2_{1/2}(\mathbb{N})$. We use a decomposition ${\bf a}(t) = e^{-
i \mu t} \left[ {\bf A} + {\bf B}(t) + i {\bf C}(t) \right]$ with real-valued vectors ${\bf B}$ and ${\bf C}$ to rewrite the discrete dynamical system (\[discrete-system\]) in the form $$\begin{aligned}
\label{nonlinear-system} \dot{\bf B} = L_- {\bf C} + \sigma {\bf
N}_-({\bf B},{\bf C}), \qquad - \dot{\bf C} = L_+ {\bf B} + \sigma
{\bf N}_+({\bf B},{\bf C}),\end{aligned}$$ where the operators $L_{\pm}$ are defined by (\[eigenvalue-1-2\]) and the vector fields ${\bf N}_{\pm}({\bf
B},{\bf C})$ contains quadratic and cubic terms with respect to $({\bf B},{\bf C})$. By Theorem \[theorem-equivalence\], the initial-value problem for system (\[nonlinear-system\]) is globally well-posed and the solution set $({\bf B},{\bf C}) \in
l^1_{1/2}(\mathbb{N},\mathbb{R}^2)$ is equivalent to the solution set $(v,w) \in {\cal H}_1(\mathbb{R},\mathbb{R}^2)$ of the PDE system (\[PDE\]). The discrete dynamical system (\[nonlinear-system\]) inherits the Hamiltonian function (\[Ham-function\]) in the form $$\begin{aligned}
\nonumber H & = & \frac{1}{2} \langle {\bf B},L_+ {\bf B} \rangle
+ \frac{1}{2} \langle {\bf C},L_- {\bf C} \rangle + \sigma
\sum_{(n,n_1,n_2,n_3)} K_{n,n_1,n_2,n_3} A_{n_1} (B_{n_2} B_{n_3}
+ C_{n_2} C_{n_3}) B_n
\\ \label{Ham-function-new} & + & \frac{\sigma}{4} \sum_{(n,n_1,n_2,n_3)}
K_{n,n_1,n_2,n_3} \left( B_{n_1} B_{n_2} B_{n_3} B_n + 2 B_{n_1}
B_{n_2} C_{n_3} C_{n} + C_{n_1} C_{n_2} C_{n_3} C_n \right)\end{aligned}$$ and the conserved quantity (\[gauge-fun\]) in the form $$\label{gauge-fun-new} Q = 2 \langle {\bf A},{\bf B} \rangle +
\langle {\bf B},{\bf B} \rangle + \langle {\bf C},{\bf C}
\rangle.$$ Using Lemma \[lemma-diagonalization\], we represent a solution $({\bf B},{\bf C})$ of the discrete system (\[nonlinear-system\]) by the series of eigenvectors (\[complete-set\]) associated with the linear problem (\[eigenvalue\]): $$\begin{aligned}
\label{diagonalization-representaion} \left\{ \begin{array}{ccc}
{\bf B}(t) & = & \sum_{m = 0}^{\infty} b_m(t) {\bf B}_m + \sum_{m
= 0}^{\infty} \bar{b}_m(t) {\bf B}_m + \beta(t) \partial_{\mu} {\bf A}, \\
{\bf C}(t) & = & i \sum_{m = 0}^{\infty} b_m(t) {\bf C}_m - i
\sum_{m = 0}^{\infty} \bar{b}_m(t) {\bf C}_m + \gamma(t) {\bf A},
\end{array} \right.\end{aligned}$$ where $b_0(t)$, ${\bf b}(t) = (b_1,b_2,...)$ are complex-valued and $\beta(t)$, $\gamma(t)$ are real-valued. The linear part of system (\[nonlinear-system\]) becomes block-diagonal in the representation (\[diagonalization-representaion\]), yielding the evolution equations $$\label{non-system-1} \dot{b}_m - i \Omega_m b_m = \sigma
N_m(b_0,{\bf b},\beta,\gamma), \quad \forall m = 0,1,2,3...$$ and $$\label{non-system-2} \dot{\beta} = \sigma S_0(b_0,{\bf
b},\beta,\gamma), \quad \dot{\gamma} + \beta = \sigma S_1(b_0,{\bf
b},\beta,\gamma),$$ where $$\begin{aligned}
N_m(b_0,{\bf b},\beta,\gamma) = \frac{\langle {\bf C}_m, {\bf
N}_-({\bf B},{\bf C}) \rangle + i \langle {\bf B}_m, {\bf
N}_+({\bf B},{\bf C}) \rangle}{2 \langle {\bf C}_m,{\bf B}_m
\rangle}, \;\; \forall m = 0,1,2,3,...\end{aligned}$$ and $$\begin{aligned}
S_0(b_0,{\bf b},\beta,\gamma) = \frac{\langle {\bf A}, {\bf
N}_-({\bf B},{\bf C}) \rangle}{\langle {\bf A}, \partial_{\mu}
{\bf A}\rangle}, \qquad S_1(b_0,{\bf b},\beta,\gamma) = -
\frac{\langle
\partial_{\mu} {\bf A}, {\bf N}_+({\bf B},{\bf C}) \rangle}{
\langle {\bf A}, \partial_{\mu} {\bf A}\rangle}.\end{aligned}$$ Using conservation of $Q$ given by (\[gauge-fun-new\]) and the decomposition (\[diagonalization-representaion\]), one can integrate the first equation of system (\[non-system-2\]) in the form $$\label{non-system-3} \beta = \frac{Q - \| {\bf B} \|^2_{l^2} - \|
{\bf C} \|^2_{l^2}}{2 \langle {\bf A}, \partial_{\mu} {\bf
A}\rangle},$$ where $Q$ is constant in time $t \in \mathbb{R}$. As a result, the second equation of system (\[non-system-2\]) is rewritten explicitly in the form $$\label{non-system-4} \dot{\gamma} = \frac{\| {\bf B} \|^2_{l^2} +
\| {\bf C} \|^2_{l^2} - 2 \sigma \langle
\partial_{\mu} {\bf A}, {\bf N}_+({\bf B},{\bf C}) \rangle - Q}{2 \langle {\bf A},
\partial_{\mu} {\bf A}\rangle}.$$ We are now ready to apply the method of Lyapunov–Schmidt reductions to the proof of Theorem \[theorem-main\].
[*of Theorem \[theorem-main\]*]{}: The vector space $({\bf B},{\bf C}) \in
l^2_{1/2}(\mathbb{N},\mathbb{R}^2)$ is equivalent to the vector space ${\bf b} \in l^2_{1/2}(\mathbb{N})$ because of the asymptotic distribution (\[asymptotic-distribution-eigenvalues\]). For instance, one obtains that $$\sum_{n = 0}^{\infty} (1+n) |B_n| \sim \langle {\bf B}, L_+ {\bf
B} \rangle = 2 \sum_{m=0}^{\infty} \Omega_m \langle {\bf C}_m,
{\bf B}_m \rangle |b_m|^2 + |\beta|^2 \langle {\bf
A},\partial_{\mu} {\bf A} \rangle \sim \sum_{n \in \mathbb{N}}
(1+n) |b_n|^2.$$ We should work in the space of $T$-periodic functions $b_0(t)$, ${\bf b}(t) \in l^2_{1/2}(\mathbb{N})$, $\beta(t)$ and $\gamma(t)$ on $t \in \mathbb{R}$, where $T$ is close to $2 \pi$. This period corresponds to the eigenvalue $\Omega_0 = 1$ which persists for any $\varepsilon \in \mathbb{R}$. By Lemma \[lemma-persistence\], all other eigenvalues of the linear problem (\[eigenvalue\]) satisfy the non-resonance conditions $n
\neq \Omega_m$, $\forall n,m \in \mathbb{N}$ for any fixed $\varepsilon \neq 0$ sufficiently small. As a result, we define periodic functions ${\bf b}(t)$, $\beta(t)$ and $\gamma(t)$ in terms of the periodic function $b_0(t)$, which solves a reduced evolution problem. Let $\delta$ be sufficiently small. We shall prove that there exist solutions of system (\[non-system-1\]), (\[non-system-3\]) and (\[non-system-4\]) which are $T$-periodic on $t \in \mathbb{R}$ satisfying the apriori bounds $$\label{apriori-bound} |b_0(t)| \leq \varepsilon \delta C_0, \;\;
\| {\bf b}(t) \|_{l^2_{1/2}} \leq \varepsilon \delta^2 C_b, \;\;
|\beta(t)| \leq \varepsilon^2 \delta^2 C_{\beta}, \;\; |\gamma(t)
- \delta \alpha| \leq \varepsilon^2 \delta^2 C_{\gamma}, \;\;
\forall t \in \mathbb{R}, \; \forall \alpha \in \mathbb{R},$$ for some ($\varepsilon,\delta$)-independent constants $C_0,C_b,C_{\beta},C_{\gamma} > 0$. If $b_0(t)$, ${\bf b}(t) \in
l^2_{1/2}(\mathbb{N})$, $\beta(t)$ and $\gamma(t)$ are $T$-periodic functions on $t \in \mathbb{R}$ satisfying the bounds (\[apriori-bound\]), then $({\bf B}(t),{\bf C}(t)) \in
l^2_{1/2}(\mathbb{N},\mathbb{R}^2)$ is a $T$-periodic function on $t \in \mathbb{R}$ satisfying the bound $$\| {\bf B}(t)\|_{l^2_{1/2}} + \| {\bf C}(t) \|_{l^2_{1/2}} \leq C
\varepsilon \delta, \qquad \forall t \in \mathbb{R}, \; \forall
\alpha \in \mathbb{R},$$ for some ($\varepsilon,\delta$)-independent constant $C > 0$. Here we recall the expansion (\[expansion-zero-eigenvalue\]) for ${\bf A}$, $\partial_{\mu} {\bf A}$ and the fact that $({\bf
B}_m,{\bf C}_m)^T$ are close to the unit vectors ${\bf e}_m$ for sufficiently small $\varepsilon$. Since ${\bf N}_{\pm}({\bf
B},{\bf C})$ is cubic with respect $({\bf A},{\bf B},{\bf C})$, contains quadratic terms in $({\bf B},{\bf C})$, and maps $l^2_{1/2}(\mathbb{N},\mathbb{R}^2)$ to $l^2_{-1/2}(\mathbb{N},\mathbb{R}^2)$, we obtain the bound $$\label{bound-N-plus} \| {\bf N}_{\pm}({\bf B}(t),{\bf C}(t))
\|_{l^2_{-1/2}} \leq C_{\pm} \varepsilon^3 \delta^2, \qquad
\forall t \in \mathbb{R}, \; \forall \alpha \in \mathbb{R},$$ for some ($\varepsilon,\delta$)-independent constants $C_{\pm} >
0$. By the Implicit Function Theorem to the right-hand-side of equation (\[non-system-4\]), there exists a unique constant $Q$ in the interval $|Q| \leq C_Q \varepsilon^2 \delta^2$ for some $C_Q > 0$, such that the periodic function in the right-hand-side of equation (\[non-system-4\]) has zero mean on $t \in
\mathbb{R}$. In this case, there exists a periodic solution $\gamma(t) = \delta \alpha + \tilde{\gamma}(t)$ of the differential equation (\[non-system-4\]), where $\tilde{\gamma}(t)$ is a uniquely defined varying part and $\delta
\alpha$ is an arbitrary mean part. The varying part $\tilde{\gamma}(t)$ satisfies the last bound in the list (\[apriori-bound\]). The function $\beta(t)$ is uniquely defined by the explicit representation (\[non-system-3\]) and it hence satisfies the third bound in the list (\[apriori-bound\]).
Consider now system (\[non-system-1\]) for $m \in
\mathbb{N}$. Recall that $\Omega_m - m = {\rm O}(\epsilon^2)$ for $m = 1,2,...$ uniformly in $m \in \mathbb{N}$ for sufficiently small $\varepsilon$. By the Implicit Function Theorem, there exists a unique solution ${\bf b}(t) \in l^2_{1/2}(\mathbb{N})$ defined by the periodic function $b_0(t)$ and parameter $\alpha
\in \mathbb{R}$ for sufficiently small $\delta$ provided that the distance $|\Omega_m - m| \neq 0$ and the frequency $\Omega$ of the periodic function $b_0(t)$ is such that $\Omega \to 1$ as $\delta
\to 0$. By the bound (\[bound-N-plus\]) and the distribution $\Omega_m - m = {\rm O}(\epsilon^2)$ for all $m \in \mathbb{N}$, the function ${\bf b}(t)$ satisfies the second bound in the list (\[apriori-bound\]).
Eliminating the components ${\bf b}$, $\beta$ and $\gamma$ from equation (\[non-system-1\]) for $n = 0$, we obtain a reduced evolution problem for $b_0(t)$ in the form $$\label{normal-form} \dot{b}_0 = i b_0 + R(b_0;\alpha),$$ where $R(b_0;\alpha)$ is a remainder term. Explicit computations of $N_0(b_0,{\bf b},\beta,\gamma)$ show that $$\begin{aligned}
\nonumber R(b_0;\alpha) & = & \varepsilon \left[ i
K_1(\varepsilon) b_0^2 + i K_2(\varepsilon) \bar{b}_0^2 + i
K_3(\varepsilon) |b_0|^2 + i K_4(\varepsilon) \delta^2 \alpha^2 +
K_5(\varepsilon) \delta \alpha \bar{b}_0 \right] \\
\label{remainder-term} & \phantom{t} & \phantom{text} + {\rm
O}\left(|b_0|^3,\varepsilon^2 \delta^2 \alpha^2 |b_0|,\varepsilon
|b_0| \| {\bf b}\| \right),\end{aligned}$$ where $K_{1,2,3,4,5}$ are real-valued constants which are bounded for sufficiently small $\varepsilon$. We are looking for $T$-periodic functions $b_0(t)$ which satisfy the evolution problem (\[normal-form\]), have the leading order $b_0 \sim
\varepsilon \delta e^{i t + i \tau}$, where $\tau \in \mathbb{R}$ is arbitrary, and satisfy the first bound in the list (\[apriori-bound\]). By the normal form analysis of the ODE (\[normal-form\]) (see [@K95]), the quadratic terms in the remainder (\[remainder-term\]) do not change the frequency $\Omega$ of oscillations of the periodic function $b_0(t)$ at the leading order and therefore, $|\Omega - 1| \leq C_{\Omega}
\varepsilon^2 \delta^2$ for some $C_{\Omega} > 0$. Since the Hamiltonian function (\[Ham-function-new\]) of system (\[nonlinear-system\]) is constant in time, it remains constant when the function $b_0(t)$ solves the reduced evolution problem (\[normal-form\]) and the functions ${\bf b}(t)$, $\beta(t)$ and $\gamma(t)$ are constructed above. By the normal form analysis of reversible systems, there exists a two-dimensional invariant manifold of system (\[normal-form\]) filled with periodic solutions of frequencies close to $\Omega = 1$ and parameterized by $(\delta,\tau)$ in addition to parameter $(\varepsilon,\alpha)$.
[Theorem \[theorem-main\] is reminiscent of an infinite-dimensional analogue of the Lyapunov Theorem for persistence of periodic orbits in Hamiltonian systems (see Chapter II, Section 45 on pp. 166–180 of [@Lyap]). However, due to the symmetries, a double zero eigenvalue occurs in the linear problem (\[eigenvalue\]), and the proof of Theorem \[theorem-main\] is complicated by the analysis of the associated two-dimensional subspace. Similar theorems on persistence of $k$-dimensional tori in $n$-dimensional Hamiltonian system with $k-1$ additional conserved quantities were studied in the Nekhoroshev–Kuksin Theorems (see Theorem 2.3 on p. 4 of [@BG] and Theorem 1 on p. xiii of [@Kuksin]).]{}
[The periodic solution of Theorem \[theorem-main\] has the smallest frequency in the focusing case $\sigma = -1$, since $\Omega_1 > 1$ in the bound (\[distribution-2\]) for sufficiently small $\varepsilon$. However, it is not the smallest frequency in the defocusing case $\sigma = 1$ since $\Omega_1 < 1$ in the bound (\[distribution-1\]). Persistence of the periodic solution for the smallest frequency $\Omega_1$ can not be proved by a simple application of the Lyapunov Theorem since the bound (\[distribution-1\]) does not guarantee that the non-resonance conditions $n \Omega_1 \neq \Omega_m$ are satisfied for all $n \in
\mathbb{N}$ and $m = 2,3,...$. By the same reason, persistence of quasi-periodic oscillations on the tori with two and more frequencies $\{ 1,\Omega_1,\Omega_2,... \}$ can not be proved for small $\varepsilon$. ]{}
[Persistence of quasi-periodic oscillations on the tori along the Cantor set of parameter values was proved in Section 2.5 on p. 33 of [@Kuksin] for the Hartree nonlinear functions and a perturbation of the parabolic potential $V(x) = \frac{1}{2} x^2$ by a localized potential $V_0(x)$. Our main result is stronger than this application of the main theorem in [@Kuksin] since the periodic orbit is continuous with respect to parameters of the PDE problem rather than along the Cantor set of parameter values.]{}
Numerical Results
=================
We illustrate results of our manuscript with some numerical approximations. First, we identify the relevant branch of stationary solutions of the ODE (\[ODE\]). To do so, we use a fixed point method (Newton-Raphson iteration) to solve a discretized boundary-value problem. A centered-difference scheme is applied to the second-order derivatives with a typical spacing $\Delta x \in [0.025,0.1]$. We are using a sufficiently large computational domain $x \in [-L,L]$ such that the boundary conditions do not affect the approximations within the considered numerical precision. The solutions $\phi(x)$ are obtained, using continuation, as a function of parameter $\mu$. The continuation of the solution branches is performed from the linear limit $\mu =
1$, both for the cases $\sigma=1$ and $\sigma=-1$. The results are shown in Figure \[dfig1\], illustrating the quantity $Q = \|
\phi \|^2_{L^2}$ as a function of $\mu$. The numerical findings are also compared to the asymptotic result (\[comparison1\]) of Proposition \[lemma-stationary\] indicating the good agreement of the latter prediction with our computational results for a fairly wide parametric window.
Once the corresponding numerical solution is identified (for a given $\sigma$ and $\mu$), the linear eigenvalue problem (\[spectrum\]) is approximated numerically. We use again a discretization of differential operators on a finite grid, such that the spectral problem (\[spectrum\]) becomes a matrix eigenvalue problem that is solved through standard numerical linear algebra routines. The relevant lowest eigenvalues are presented in Figure \[dfig2\] and are also compared with the corresponding asymptotic results (\[eigenvalue-expansion-1\])–(\[eigenvalue-expansion-3\]) of Proposition \[lemma-eigenvalue\]. The dashed lines show asymptotic results (\[eigenvalue-expansion-1-asymptotic\])–(\[eigenvalue-expansion-3-asymptotic\]) of Appendix A derived in the limit $\mu \to \infty$ for $\sigma =
1$. Once again, the good agreement offers us a quantitative handle on the relevant eigenvalues.
![Branches of dark solitons versus $\mu$ both for the case of $\sigma=-1$ (when $\mu<1$) and $\sigma=1$ (when $\mu>1$). The numerically obtained solution is shown by solid line and the asymptotic solution (\[comparison1\]) is shown by dash-dotted line.[]{data-label="dfig1"}](paper_fig1.eps){height="8cm"}
![Smallest purely imaginary eigenvalues of the linear eigenvalue problem (\[spectrum\]) versus $\mu$. The numerically obtained eigenvalues are shown by solid lines, the asymptotic results (\[eigenvalue-expansion-1\])-(\[eigenvalue-expansion-3\]) are shown by dash-dotted lines, and the asymptotic results (\[eigenvalue-expansion-1-asymptotic\])–(\[eigenvalue-expansion-3-asymptotic\]) are shown by dashed lines. []{data-label="dfig2"}](paper_fig2.eps){height="8cm"}
Finally, we have also examined periodic oscillations of dark solitons in the numerical simulations of the GP equation (\[GP-zero-epsilon\]). A typical example is shown in Figure \[dfig3\] for $\sigma = 1$ and $\mu = 1.1$ for the initial condition $u(x,0) = \phi(x) + \delta \phi'(x)$ with $\delta =
10^{-3}$. The top left panel shows the space-time contour plot of $|u(x,t)|^2$, clearly highlighting that this is a small (imperceptible, at the scale of this panel) perturbation of a stable stationary solution $\phi(x)$. The bottom left panel shows the space-time contour plot of $|u(x,t)|^2 - \phi^2(x)$, emphasizing the time-periodic oscillations of the perturbation to the stationary solution. The periodic oscillations are also visible on the top right panel where $|u(x_0,t)|^2$ is plotted versus $t$ for $x_0 = 2$. Finally, the bottom right panel illustrates the Fourier transform of the time series of $|u(x_0,t)|^2$ (normalized to its maximum). It shows a high peak of the frequency spectrum near the value $\Omega=1$, in agreement with the results of the main Theorem \[theorem-main\].
![A typical example of the robust time-periodic solution of the Gross-Pitaevskii equation (\[GP-zero-epsilon\]) for $\sigma
= 1$, $\mu = 1.1$ and $u(x,0) = \phi(x) + \delta \phi'(x)$ with $\delta = 10^{-3}$. The top left panel shows the space-time contour plot of $|u(x,t)|^2$, the bottom left panel shows the space-time contour of $|u(x,t)|^2-\phi^2(x)$. The top right panel shows the time evolution of $|u(x_0,t)|^2$ with $x_0 = 2$, while the bottom right panel shows the Fourier transform of the time series of $|u(x_0,t)|^2$, featuring a peak at $\Omega \approx 1$.[]{data-label="dfig3"}](paper_fig3.eps){height="14cm"}
[**Acknowledgement.**]{} D.P. thanks to W. Craig and V. Konotop for useful discussions related to the project. D.P. is supported by the Humboldt and EPSRC fellowships. P.G.K. is supported by NSF through the grants DMS-0204585, DMS-CAREER, DMS-0505663 and DMS-0619492.
Asymptotic distribution of eigenvalues
======================================
Let us consider the case $\sigma = 1$, when the solution $\phi(x)$ of the ODE (\[ODE\]) bifurcates to the interval $\mu > 1$ (see Proposition \[lemma-stationary\] and Figure \[dfig1\]). We are interested in the distribution of eigenvalues of the linear problem (\[spectrum\]) as $\mu \to \infty$, assuming that the solution $\phi(x)$ persists in this limit. It follows from the scaling transformation below equation (\[GP-zero-epsilon\]) that the limit $\mu \to \infty$ of the normalized equation (\[GP-zero-epsilon\]) corresponds to the limit $\epsilon \to 0$ in the original GP equation (\[GP\]). We shall replace $\mu +
\frac{1}{2} = \tilde{\mu}$ and drop tilde notations for the sake of simplicity. We report here formal results based on asymptotic methods. Rigorous justification of these results is beyond the scope of our work.
Denote the ground state of the ODE (\[ODE\]) by $\phi_0(x)$ such that $\phi_0(x)$ is even and positive on $x \in \mathbb{R}$ and it decays to zero as $|x| \to \infty$ sufficiently fast. Using the substitution $\phi_0(x) = \sqrt{\mu q(\xi)}$ and $\xi =
\frac{x}{\sqrt{2 \mu}}$, we obtain an equation for $q(\xi)$, $$\label{ODE-WKB} q = 1 - \xi^2 + \frac{1}{4 \mu^2 \sqrt{q}}
\frac{d^2}{d \xi^2} \sqrt{q}, \qquad \forall \xi \in \mathbb{R},$$ which is solvable with the nonlinear WKB series [@KonKev]. The main result of the formal WKB theory is that there exists a classical solution $q_{\mu}(\xi)$ of the ODE (\[ODE-WKB\]) for sufficiently large $\mu > 1$ such that $$\label{ground-state-limit}
\lim_{\mu \to \infty} q_{\mu}(x) = \left\{ \begin{array}{cc} 1 - \xi^2, & \forall |\xi| \leq 1 \\
0, & \forall |\xi| > 1 \end{array} \right.$$ The linear problem (\[spectrum\]) associated with the ground state $\phi_0(x)$ for $\sigma = 1$ and $\mu + \frac{1}{2} \to \mu$ can be written in variables $v = V(\xi)$, $w = W(\xi)$ and $\lambda = \mu \Lambda$ for sufficiently large $\mu > 1$. In new variables, it takes the form $$\label{linear-problem-WKB} L_+ V = -\Lambda W, \qquad L_- W =
\Lambda V,$$ where $$L_+ = 3 q(\xi) - 1 + \xi^2 - \frac{1}{4 \mu^2} \frac{d^2}{d
\xi^2}, \quad L_- = q(\xi) - 1 + \xi^2 - \frac{1}{4 \mu^2}
\frac{d^2}{d \xi^2}.$$ Eliminating $V(x)$, we close the linear problem (\[linear-problem-WKB\]) at the fourth-order ODE $$L_+ L_- W = \Gamma W, \qquad \Gamma = - \Lambda^2.$$ By using the WKB theory (\[ground-state-limit\]), we consider the auxiliary eigenvalue problem $$\label{spectrum-aux} \frac{1}{16 \mu^4} W^{({\rm iv})} - \frac{(1
-\xi^2)}{2 \mu^2} W'' = \Gamma W(\xi), \qquad \forall \xi \in
[-1,1],$$ for $W \in L^2([-1,1])$. The entire spectrum of the problem (\[spectrum-aux\]) is defined by a set of polynomial solutions $W = P_m(\xi) = \xi^m + \alpha_{m,m-2} \xi^{m-2} + ... +
\alpha_{m,k} \xi^k$, $\forall m \in \mathbb{N}$, where $k = 1$ if $m$ is odd and $k = 0$ if $m$ is even. The balance of the largest term in the ODE (\[spectrum-aux\]) shows that the eigenvalue $\Gamma = \Gamma_m$ is found explicitly as $\Gamma_m =
\frac{m(m-1)}{2 \mu^2}$, while all coefficients $\{
\alpha_{m,m-2k} \}_{k = 1}^{[m/2]}$ are uniquely defined. Converting the values of $\Gamma$ to the values of $\lambda$, we have found that the linear problem (\[spectrum\]) associated with the ground state $\phi_0(x)$ has a set of simple purely imaginary and symmetric eigenvalue pairs $\{ \pm i \Omega_m \}_{m
\in \mathbb{N}}$, such that $$\begin{aligned}
\label{eigenvalue-expansion-1-asymptotic} \lim\limits_{\mu \to
\infty} \Omega_m = \frac{\sqrt{m(m+1)}}{\sqrt{2}}, \qquad \forall
m \in \mathbb{N},\end{aligned}$$ in addition to the double zero eigenvalue $\lambda = 0$.
Finally, the dark soliton $\phi(x)$ of the ODE (\[ODE\]) is obtained asymptotically from the ground state $\phi_0(x)$ by the factorization $\phi(x) = \phi_0(x) \psi(x)$, where $\psi(x)$ is odd on $x \in \mathbb{R}$, positive on $x \in \mathbb{R}_+$ and may approach to the constant values as $|x| \to \infty$ [@PFK05]. Using this factorization and the formal asymptotic analysis, it was shown in [@PFK05] that the spectrum of the linear problem (\[spectrum\]) associated with the dark soliton $\phi(x)$ admits a pair of simple purely imaginary eigenvalues $\pm i \Omega_0$, such that $$\begin{aligned}
\label{eigenvalue-expansion-3-asymptotic} \lim_{\mu \to \infty}
\Omega_0 = \frac{1}{\sqrt{2}}.\end{aligned}$$ Although the analysis of [@PFK05] was directed to the original GP equation (\[GP\]) in the limit of small $\epsilon$ and the eigenvalue pair was found to be $\tilde{\lambda} \to \pm i
\epsilon$, the scaling transformation to the normalized GP equation (\[GP-zero-epsilon\]) implies that $\lambda =
\frac{\tilde{\lambda}}{2^{1/2} \epsilon} \to \pm
\frac{i}{\sqrt{2}}$.
Numerical computations (see Figure \[dfig2\]) suggests that the entire spectrum of the linear problem (\[spectrum\]) associated with the dark soliton $\phi(x)$ is a superposition between an infinite set of eigenvalues (\[eigenvalue-expansion-1-asymptotic\]) of the linear problem (\[spectrum\]) associated with the ground state $\phi_0(x)$ and the additional pair of eigenvalues (\[eigenvalue-expansion-3-asymptotic\]).
Note that the linear eigenmode corresponding to the smallest eigenvalue $\Omega_0 = \frac{1}{\sqrt{2}}$ may not result in the periodic solution of the nonlinear PDE system (\[PDE\]) because the non-resonance condition $n \neq \sqrt{m(m+1)}$ for all $n,m
\in \mathbb{N}$ is violated in the limit $n,m \to \infty$. Similarly, the linear eigenmode corresponding to the second eigenvalue $\Omega_1 = 1$ may not result in the periodic solution of the PDE system (\[PDE\]) because the non-resonance condition $n \neq \frac{\sqrt{m(m+1)}}{\sqrt{2}}$ for all $n,m = 2,3,...$ is violated at least for $n = 6$ and $m = 8$. In both cases, the Lyapunov Theorem for persistence of periodic orbit in Hamiltonian dynamical systems can not be applied [@Lyap].
[99]{}
M. Abramowitz and I.A. Stegun, [*Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables*]{} (Dover, New York, 1965), chapter 22.
R.A. Adams, [*Sobolev Spaces*]{} (Academic Press Inc., San Diego, 1978)
G.L. Alfimov and D.A. Zezyulin, “Nonlinear modes for the Gross–Pitaevskii equation - demonstrative computation approach”, arXiv: nlin.PS/0703006
D. Bambusi and G. Gaeta, “On persistence of invariant tori and a theorem by Nekhoroshev”, Math. Phys. Electr. Journal [**8**]{}, paper I (2002)
V.A. Brazhnyi and V.V. Konotop, “Evolution of a dark soliton in a parabolic potential: application to Bose–Einstein condensates”, Physical Review A [**68**]{}, 043613 (2003)
R. Carles, “Remarks on nonlinear Schrödinger equations with harmonic potential”, Annales Henri Poincare [**3**]{}, 757–772 (2002)
G. Freud and G. Németh, “On the $L_p$-norms of orthonormal Hermite functions”, Studia Scientiarum Mathematicarum Hungarica [**8**]{}, 399-404 (1973)
M. Golubitsky and D.G. Schaeffer, [*Singularities and Groups in Bifurcation Theory*]{}, vol. 1, (Springer-Verlag, New York, 1985)
B.L.G. Jonsson, J. Fröhlich, S. Gustafson, and I.M. Sigal, “Long time motion of NLS solitary waves in a confining potential”, Annales Henri Poincare [**7**]{}, 621–660 (2006)
T. Kato, [*Perturbation theory for linear operators*]{}, (Springer-Verlag, New York, 1976)
V.V. Konotop, “Dark solitons in Bose–Einstein condensates: theory” in [*“Emergent Nonlinear Phenomena in Bose–Einstein Condensates”*]{}, Eds. P.G. Kevrekidis, D.J. Franzeskakis, and R. Carretero–Gonzalez (Springer–Verlag, New York, 2007)
V.V. Konotop and P.G. Kevrekidis, “Bohr–Sommerfeld quantization condition for the Gross–Pitaevskii equation”, Physical Review Letters [**91**]{}, 230402 (2003)
S.B. Kuksin, [*Nearly Integrable Infinite–Dimensional Hamiltonian Systems*]{} (Springer–Verlag, Berlin, 1993)
Y.A. Kuznetsov, [*Elements of Applied Bifurcation Theory*]{}, 2nd ed., Appl. Math. Sci. [**112**]{} (Springer–Verlag, New York, 1998)
M.A. Lyapunov, [*The General Problem of the Stability of Motion*]{} (Taylor and Francis, London, 1992)
D.E. Pelinovsky, D. Frantzeskakis, and P.G. Kevrekidis, “Oscillations of dark solitons in trapped Bose-Einstein condensates”, Physical Review E [**72**]{}, 016615 (2005)
D.E. Pelinovsky and P.G. Kevrekidis, “Dark solitons in external potentials”, Zeitschrift für Angewandte Mathematik und Physik, to be published (2007)
N.G. Parker, N.P. Proukakis, C.F. Barenghi, and C.S. Adams, “Dynamical instability of a dark soliton in a quasi-one-dimensional Bose-Einstein condensate perturbed by an optical lattice”, Journal of Physics B: Atomic Molecular Optical Physics [**37**]{}, S175–S185 (2004)
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Giovanni Sutanto, Katharina Rombach\*, Yevgen Chebotar\*, Zhe Su\*, Stefan Schaal\*, Gaurav S. Sukhatme and Franziska Meier'
bibliography:
- 'references.bib'
title: |
Supervised Learning and\
Reinforcement Learning of\
Feedback Models for\
Reactive Behaviors:\
Tactile Feedback Testbed
---
This research was supported in part by the National Science Foundation grants IIS-1205249, IIS-1017134, EECS-0926052, the Office of Naval Research, and the Okawa Foundation, all of which were issued to the University of Southern California. Moreover, this research was also supported in part by the Max-Planck-Society through funding provided to Giovanni Sutanto, Katharina Rombach, Yevgen Chebotar, Zhe Su, and Stefan Schaal.
APPENDIX {#appendix .unnumbered}
========
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Within the smooth category, an intertwining is exhibited between the global rigidity of irreducible higher-rank ${\mathbb Z}^d$ Anosov actions on $\mathbb T^n$ and the classification of equilibrium-free flows on ${\mathbb T}^n$ that possess nontrivial generalized symmetries.'
author:
- 'Lennard F. Bakker'
title: 'The Katok-Spatzier Conjecture and Generalized Symmetries'
---
Introduction
============
In [@GO], Spatzier communicated the conjecture that any irreducible higher-rank ${\mathbb Z}^d$ Anosov action is $C^\infty$-conjugate to an algebraic action. Later, in [@KS], Kalinin and Spatzier stated a refinement of this conjecture that contends that any irreducible higher-rank ${\mathbb Z}^d$ Anosov action on any compact manifold has a finite cover $C^\infty$-conjugate to an algebraic action. The asserted global rigidity was motivated in part by earlier results of Katok and Lewis in [@KL1] and [@KL2], and a more recent result by Rodriguez Hertz in [@RH]. In the latter, global rigidity has been shown for any higher-rank ${\mathbb Z}^d$ Anosov action on ${\mathbb T}^n$ whose action on homology has simple eigenvalues and whose course Lyapunov spaces are one or two dimensional, plus additional conditions. A partial confirmation of the refined assertion of global rigidity is provided in [@KS] for higher-rank ${\mathbb Z}^d$ Anosov $C^\infty$ actions each of whose course Lyapunov spaces are one-dimensional, plus additional conditions. If a higher-rank ${\mathbb Z}^d$ Anosov action on ${\mathbb T}^n$ is $C^\infty$-conjugate to an algebraic one and that algebraic action has a common real eigenvector, then that higher-rank ${\mathbb Z}^d$ Anosov action preserves a one-dimensional $C^\infty$ foliation of ${\mathbb T}^n$ determined by that common real eigenvector, i.e., generated by a equilibrium-free $C^\infty$ flow.
This paper intertwines the global rigidity of higher-rank ${\mathbb Z}^d$ Anosov $C^\infty$ actions on ${\mathbb T}^n$ with the classification of equilibrium-free $C^\infty$ flows on ${\mathbb T}^n$ that possess nontrivial generalized symmetries. The intertwining centers on presence of a single one-dimensional $C^\infty$ distribution determined by an equilibrium-free $C^\infty$ flow that is invariant under a ${\mathbb Z}^d$ Anosov $C^\infty$ action, without a priori conditions on all the course Lyapunov spaces. As shown in Section 3, any generalized symmetry of an equilibrium-free flow is nontrivial if it is Anosov (see Corollary \[multiplierAnosov\]). Furthermore, any equilibrium-free flow that possesses a nontrivial generalized symmetry does not have any uniformly hyperbolic compact invariant sets (see Corollary \[nonhyper\]). The intertwining juxtaposes an equilibrium-free $C^\infty$ flow that is not Anosov with a smooth ${\mathbb Z}^d$ action that is Anosov. In the $C^\infty$ topology, this is a counterpoint to the result of Palis and Yoccoz in [@PY] on the triviality of centralizers for an open and dense subset of Anosov diffeomorphisms on ${\mathbb T}^n$, and also to the result of Sad in [@Sa] on the local triviality of centralizers for an open and dense subset of Axiom A vector fields that satisfy the strong transversality condition (as applied to vector fields on ${\mathbb T}^n$). In particular, it is quite rare for an equilibrium-free $C^\infty$ flow on ${\mathbb T}^n$ (or more generally, on a closed Riemannian manifold) to possess a nontrivial generalized symmetry (see Corollary \[rare\]).
The first aspect of the intertwining on ${\mathbb T}^n$ relates the global rigidity of a ${\mathbb Z}^d$ Anosov $C^\infty$ action with an equilibrium-free $C^\infty$ flow that is quasiperiodic. As detailed in Section 2, the generalized symmetry group of a $C^\infty$ flow $\Phi$ is the subgroup $S_\Phi$ of ${\rm Diff}^\infty({\mathbb T}^n)$ each of whose elements $R$ sends (via the pushforward) the generating vector field $X_\Phi$ of $\Phi$ to a uniform scalar multiple $\rho_\Phi(R)$ of itself. The multiplier group $M_\Phi$ of $\Phi$ is the abelian group of these scalars. As shown in Section 4, the elements of $M_\Phi\setminus\{1,-1\}$ when $\Phi$ is quasiperiodic (or more generally, minimal) are algebraic integers of degree between $2$ and $n$ inclusively (see Corollary \[algebraicnature\]).
\[Anosovalgebraic\] On ${\mathbb T}^n$, suppose $\alpha$ is a ${\mathbb Z}^d$ Anosov $C^\infty$ action, and $\Phi$ is equilibrium-free $C^\infty$ flow. If $\alpha({\mathbb Z}^d)\subset S_\Phi$ and $\Phi$ is quasiperiodic $($i.e., $C^\infty$-conjugate to an irrational flow$)$, then $\alpha({\mathbb Z}^d)$ is $C^\infty$-conjugate to an affine action, a finite index subgroup of $\alpha({\mathbb Z}^d)$ is $C^\infty$-conjugate to an algebraic action, and $M_\Phi$ contains a ${\mathbb Z}^d$ subgroup.
Relevant definitions and the proof are given in Section 6. The proof holds not only for $d\geq 2$ but also for $d=1$. It uses a semidirect product characterization of the structure of the generalized symmetry group for an irrational flow (as shown in Section 5). It also uses the existence of a common fixed point for a finite index subgroup of the ${\mathbb Z}^d$ Anosov action, a device used in other global rigidity results (for example, see [@KL2]).
The second aspect of the intertwining on ${\mathbb T}^n$ relates the classification of an equilibrium-free $C^\infty$ flow with a ${\mathbb Z}^d$ Anosov $C^\infty$ action that is topologically irreducible. As detailed in Section 7, an irrational flow $\phi$ on ${\mathbb T}^n$ is of Koch type if a uniform scalar multiple of its frequencies form a ${\mathbb Q}$-basis for a real algebraic number field of degree $n$ (also see [@KO] and [@LD]). For an $R\in S_\Phi$ of an equilibrium-free $C^\infty$ flow $\Phi$, the quantity $\log\vert \rho_\Phi(R)\vert$ is the value of the Lyapunov exponent $\chi_R$ of $R$ in the direction of $X_\Phi$ (see Theorem \[Lyapunov\]).
\[flowKoch\] On ${\mathbb T}^n$, suppose $\alpha$ is a higher-rank ${\mathbb Z}^d$ Anosov $C^\infty$ action, and $\Phi$ is equilibrium-free $C^\infty$ flow. If $\alpha({\mathbb Z}^d)\subset S_\Phi$, and $\alpha$ is topologically irreducible and $C^\infty$-conjugate to an algebraic ${\mathbb Z}^d$ action, and for an Anosov element $R\in{\alpha({\mathbb Z}^d})$, the multiplicity of the value $\log\vert \rho_\Phi(R)\vert$ of $\chi_R$ is one at some point of ${\mathbb T}^n$, then $\Phi$ is projectively $C^\infty$-conjugate to an irrational flow of Koch type.
Relevant definitions and the proof are given in Section 7. The proof uses the Oseledets decomposition for an Anosov diffeomorphism (see [@BP] and [@KH]) to show that the flow is $C^\infty$-conjugate to one generated by a constant vector field. Then by the topological irreducibility and results of Wallace in [@WA], the components of a scalar multiple of the constant vector field are shown to form a ${\mathbb Q}$-basis for a real algebraic number field.
Flows with Nontrivial Generalized Symmetries
============================================
Generalized symmetries extend the classical notions of time-preserving and time-reversing symmetries of flows. To simplify notations for these and for proofs of results, it is assumed throughout the remainder of the paper that all manifolds, flows, vector fields, diffeomorphisms, distributions, etc., are smooth, i.e., of class $C^\infty$. Let $P$ be a closed (i.e., compact without bounday) manifold. Let ${\rm Flow}(P)$ denote the set of flows on $P$. Following [@BC], a [*generalized symmetry*]{} of $\psi\in{\rm Flow}(P)$ is an $R\in{\rm Diff}(P)$ such that there is $\mu\in{\mathbb R}^\times = {\mathbb R}\setminus\{0\}$ (the multiplicative real group) for which $$R\psi(t,p) = \psi(\mu t, R(p)){\rm\ for\ all\ } t\in{\mathbb R}{\rm\ and\ all\ } p\in P.$$ It is easy to show that $R$ being a generalized symmetry of $\psi$ is equivalent to $R$ satisfying $$R_*X_\psi = \mu X_\psi{\rm\ for\ some\ }\mu\in{\mathbb R}^\times.$$ Here $X_\psi(p)=(d/dt)\psi(t,p)\vert_{t=0}$ is the vector field that generates $\psi$, and $R_*X_\psi = {\bf T}R X_\psi R^{-1}$ is the push-forward of $X_\psi$ by $R$ where ${\bf T}R$ is the derivative map. The [*generalized symmetry group*]{} of $\psi$ is the set $S_\psi$ of all the generalized symmetries that $\psi$ possesses. There is a homomorphism $\rho_\psi:S_\psi \to {\mathbb R}^\times$ taking $R\in S_\psi$ to its unique multiplier $\rho_\psi(R) = \mu$. The multiplier group of $\psi$ is $M_\psi = \rho_\psi(S_\psi)$.
The generalized symmetry group and the multiplier group of a flow are invariants for the equivalence relation of projective conjugacy. Two $\psi,\phi\in{\rm Flow}(P)$ are [*projectively conjugate*]{} if there are $h\in{\rm Diff}(P)$ and $\vartheta\in{\mathbb R}^\times$ such that $h_*X_\psi = \vartheta X_\phi$. Projective conjugacy is an equivalence relation on ${\rm Flow}(P)$. Projective conjugacy reduces to smooth conjugacy when $\vartheta = 1$. For $h\in{\rm Diff}(P)$, let $\Delta_h$ be the inner automorphism of ${\rm Diff}(P)$ given by $\Delta_h(R) = h^{-1}Rh$ for $R\in{\rm Diff}(P)$. If $h_*X_\psi = \vartheta X_\phi$, then $\Delta_h(S_\phi) = S_\psi$ (see Theorem 4.1 in [@BC] which states that $S_\psi$ is conjugate to the generalized symmetry group for the flow determined by $\vartheta X_\phi$, which is exactly the same as $S_\phi$.) Furthermore, if $\psi$ and $\phi$ are projectively conjugate, then $M_\psi=M_\phi$ (see Theorem 2.2 in [@BA5]), i.e., the multiplier group is an absolute invariant of projective conjugacy.
Any $R\in S_\psi$ is a trivial generalized symmetry of $\psi$ if $\rho_\psi(R)=1$ (i.e., $R$ is time-preserving), or if $\rho_\psi(R)=-1$ (i.e., $R$ is time-reversing). Any $R\in S_\psi$ with $\vert\rho_\psi(R)\vert\ne 1$ is a [*nontrivial generalized symmetry*]{} of $\psi$. A flow $\psi$ (or equivalently its generating vector field $X_\psi$) is said to possess a nontrivial generalized symmetry when $M_\psi\setminus\{1,-1\}\ne\emptyset$.
\[noperiodic\] Let $\psi$ be a flow on a closed Riemannian manifold $P$. If $\psi$ has a periodic orbit and $M_\psi\setminus\{1,-1\}\ne\emptyset$, then $\psi$ has a nonhyperbolic equilibrium.
Suppose for $p_0\in P$ that ${\mathcal O}_\psi(p_0)=\{\psi_t(p_0):t\in{\mathbb R}\}$ is a periodic orbit whose fundamental period is $T_0>0$. (Here $\psi_t(p) = \psi(t,p)$.) Assuming that $M_\phi\setminus\{1,-1\}\ne\emptyset$ implies there is $Q\in S_\psi$ such that $\rho_\psi(Q)\ne \pm 1$. If $\vert \rho_\psi(Q)\vert >1$, then $\vert\rho_\psi(Q^{-1})\vert<1$, since $\rho_\psi$ is a homomorphism. Hence there is $R\in S_\psi$ such that $\vert\rho_\psi(R)\vert<1$. Then $$R(p_0) = R\psi(T,p_0) = \psi\big(\rho_\psi(R)T_0,R(p_0)\big).$$ For $p_1= R(p_0)$, this implies that ${\mathcal O}_\psi(p_1)$ is a periodic orbit with period $T_1 = \vert \rho_\psi(R)\vert T_0$. Suppose that $T_1$ is not a fundamental period for ${\mathcal O}_\psi(p_1)$, i.e., there is $0<T_1^\prime<T_1$ such that $p_1 = \psi(T_1^\prime,p_1)$. This implies that $$R(p_0) = p_1 = \psi(T_1^\prime,p_1) = \psi\big(T_1^\prime,R(p_0)\big) = R\psi(T_1^\prime/\rho_\psi(R),p_0).$$ Since $R$ is invertible, this gives $p_0 = \psi(T_1^\prime/\rho_\psi(R), p_0)$. With $T_0$ as the fundamental period for ${\mathcal O}_\psi(p_0)$, there is $m\in{\mathbb Z}^+$ such that $\vert T_1^\prime/\rho_\psi(R)\vert = mT_0$. Hence $$T_1^\prime = m T_0 \vert \rho_\psi(R)\vert \geq T_0\vert \rho_\psi(R)\vert = T_1,$$ a contradiction to $T_1^\prime < T_1$. Thus, $T_1$ is the fundamental period for ${\mathcal O}_\psi(p_1)$. Since $\vert \rho_\psi(R)\vert<1$, it follows that ${\mathcal O}_\psi(p_0)$ and ${\mathcal O}_\psi(p_1)$ have different fundamental periods, and hence are distinct periodic orbits. Iteration gives a sequence of distinct periodic orbits ${\mathcal O}_\psi(p_i)$, where $p_i = R^i(p_0)$, whose fundamental periods $T_i = \vert \rho_\psi(R)\vert^i T_0$ decrease to $0$ since $\vert \rho_\psi(R)\vert <1$.
Let $l(p_i)$ be the arc length of ${\mathcal O}_\psi(p_i)$. In terms of the Riemannian norm $\Vert\cdot\Vert$ on ${\bf T}P$, the tangent bundle of $P$, $$l(p_i) = \int_0^{T_i} \Vert X_\psi\big(\psi(t,p_i)\big)\Vert \ dt.$$ Continuity of $X_\psi$ on the compact manifold $P$ implies that $M=\sup_{p\in P} \Vert X_\psi(p)\Vert$ is finite. Thus $$l(p_i)\leq MT_i \to 0.$$ By the compactness of $P$, there is a convergence subsequence $p_{i_m}$ with limit $p_\infty$. If $\Vert X_\psi(p_\infty)\Vert\ne 0$, then the smoothness of $X_\psi$ implies by the Flow Box Theorem that there are no periodic orbits of $\psi$ completely contained in a sufficiently small neighborhood $U$ of $p_\infty$. But since $l(p_{i_m})\to0$ and $p_{i_m}\to p_\infty$, there are periodic orbits of $\psi$ completely contained in $U$, a contradiction. Thus $p_\infty$ is an equilibrium for $\psi$. If $p_\infty$ were hyperbolic, then the Hartman-Grobman Theorem would imply that there are no periodic orbits completely contained in a sufficiently small neighborhood of $p_\infty$, again a contradiction. Therefore, $\phi$ has a nonhyperbolic equilibrium.
The flows on a closed Riemannian manifold which possess nontrivial generalized symmetries are rare or non-generic in the sense of Baire category theory. Let ${\mathcal X}(P)$ denote the set of vector fields on $P$. With $P$ compact, there is a one-to-one correspondence between ${\mathcal X}(P)$ and ${\rm Flow}(P)$. Equipped with the usual $C^\infty$ topology, ${\mathcal X}(P)$ is a Baire space. A residual subset of ${\mathcal X}(P)$ is one that contains a countable intersection of open, dense subsets of ${\mathcal X}(P)$. Because ${\mathcal X}(P)$ is a Baire space, any residual subset is dense.
\[rare\] For a closed Riemannian manifold $P$, the set of vector fields on $P$ which do not possess nontrivial generalized symmetries is a residual subset of ${\mathcal X}(P)$.
Let ${\mathcal V}$ be the set of $X$ in ${\mathcal X}(P)$ such that any equilibrium, if any, of the flow $\psi$ induced by $X$ is hyperbolic, and there is at least one periodic orbit for $\psi$. By Theorem \[noperiodic\], none of the vector fields in ${\mathcal V}$ possess nontrivial generalized symmetries. Let ${\mathcal H}$ be the subset of $X$ in ${\mathcal X}(P)$ such that the flow $\psi$ induced by $X$ has periodic orbits all of which are hyperbolic, and any equilibrium it has is hyperbolic. By definition, ${\mathcal H}\subset {\mathcal V}$, and by the Kupka-Smale Theorem, ${\mathcal H}$ is a residual subset of ${\mathcal X}(P)$.
According to Theorem \[noperiodic\], flows which might possess nontrivial generalized symmetries are those that are equilibrium-free and without periodic orbits. The possession of a nontrivial generalized symmetry for an equilibrium-free flow places dynamical restrictions on the compact invariant sets of the flow. An invariant set $\Lambda$ for an equilibrium-free $\Phi\in{\rm Flow}(P)$ is [*uniformly hyperbolic*]{} if there is a continuous ${\bf T}\Phi$-invariant splitting ${\bf T}_p\Lambda = E^s(p) \oplus{\rm Span}(X_\Phi(p))\oplus E^u(p)$ for $p\in\Lambda$ and constants $C\geq 1$ and $\lambda\in(0,1)$ such that $$\Vert {\bf T}_p \Phi_t(v)\Vert \leq C\lambda^t \Vert v\Vert {\rm\ for\ }v\in E^s(p) {\rm\ and\ }t\geq 0$$ and $$\Vert {\bf T}_p \Phi_{-t}(v)\Vert \leq C\lambda^t \Vert v\Vert {\rm\ for\ }v\in E^u(p) {\rm\ and\ }t\geq 0.$$ Recall that if $\Lambda = P$ is uniformly hyperbolic for $\Phi$, then $\Phi$ is called Anosov.
\[nonhyper\] If $\Phi$ is an equilibrium-free flow on a closed Riemannian manifold $P$ which possesses a nontrivial generalized symmetry, then any compact invariant set for $\Phi$ is not uniformly hyperbolic $($and, in particular, $\Phi$ is not Anosov$)$.
Suppose $\Lambda$ is a uniformly hyperbolic compact invariant set of $\Phi$. By the Anosov Closing Lemma, there exists a periodic orbit for $\Phi$. But with $\Phi$ being equilibrium-free and possessing a nontrivial generalized symmetry, Theorem \[noperiodic\] implies that $\Phi$ does not have periodic orbits. Thus $\Lambda$ can not be uniformly hyperbolic.
Multipliers and Lyapunov Exponents
==================================
The multiplier of a generalized symmetry for an equilibrium-free flow is related to the Lyapunov exponents of that generalized symmetry. For a closed Riemannian manifold $P$, the [*Lyapunov exponent*]{} of $R\in{\rm Diff}(P)$ at $p\in P$ is $$\chi_R (p,v) = \limsup_{m\to\infty} \frac{\log\Vert {\bf T}_p R^m(v)\Vert}{m}, \ v\in {\bf T}_p P\setminus\{0\}.$$ It is independent of the Riemannian norm on $P$ because $P$ is compact. As the following result shows, any trivial generalized symmetry for an equilibrium-free flow has zero Lyapunov exponents everywhere.
\[Lyapunov\] If $\psi$ is an equilibrium-free flow on a closed Riemannian manifold $P$, then for each $R\in S_\psi$, the one-dimensional distribution $E\subset {\bf T}P$ determined by $E(p) = {\rm Span}\big(X_\psi(p)\big)$ is $R$-invariant and $$\chi_R\big(p,X_\psi(p)\big) = \log\vert \rho_\psi(R)\vert {\rm \ for\ all\ }p\in P.$$
Each $R\in S_\psi$ satisfies ${\bf T}_pR\big(X_\psi(R^{-1}(p)\big) = \rho_\psi(R) X_\psi(p)$ for all $p\in P$. Then ${\bf T}_pR(X_\psi(p)) = \rho_\psi(R)X_\psi(R(p))$. The distribution $E\subset {\bf T}P$ determined by $E(p) = {\rm Span}\big(X_\psi(p)\big)$ is $R$-invariant and one-dimensional because $\rho_\psi(R)\ne 0$ and $X_\psi(p)\ne 0$ for all $p\in P$. Furthermore, it follows for all $m\geq 1$ that $${\bf T}_pR^m(X_\psi(p)) = [\rho_\psi(R)]^m X_\psi\big(R^m(p)\big),$$ and so $$\Vert {\bf T}_pR^m(X_\psi(p))\Vert = \vert \rho_\psi(R)\vert^m \Vert X_\psi\big(R^m(p)\big)\Vert.$$ The flow $\psi$ being smooth and equilibrium-free on the compact manifold $P$ implies that $\Vert X_\psi\big(R^m(p)\big)\Vert$ is both bounded away from $0$ and bounded above uniformly in $m$ for each $p\in P$. Therefore $$\chi_R\big(p,X_\psi(p)\big) = \limsup_{m\to\infty} \frac{ m \log\vert \rho_\psi(R)\vert + \log \Vert X_\psi\big(R^m(p)\big)\Vert}{m} = \log\vert \rho_\psi(R)\vert$$ for all $p\in P$.
Global uniform hyperbolicity is a dynamical condition on a generalized symmetry of an equilibrium-free flow that guarantees that it is nontrivial. An $R\in{\rm Diff}(P)$ is [*Anosov*]{} if there is a continuous ${\bf T}R$-invariant splitting $T_p P = E^s(p)\oplus E^u(p)$ for $p\in P$, and constants $c>0$ and $\lambda\in(0,1)$ independent of $p\in P$, such that $$\Vert {\bf T}_p R^m(v)\Vert \leq c \lambda^m \Vert v\Vert {\rm \ for\ }v\in E^s(p){\rm\ and \ }m\geq 0,$$ and $$\Vert {\bf T}_p R^{-m}(v)\Vert \leq c \lambda^m \Vert v\Vert {\rm \ for\ }v\in E^u(p){\rm\ and \ }m\geq 0$$ with the angle between $E^s(p)$ and $E^u(p)$ bounded away from $0$ (see [@BP]).
\[multiplierAnosov\] Let $\psi$ be an equilibrium-free flow on a closed Riemannian manifold $P$. If $R\in S_\psi$ is Anosov, then $\vert\rho_\psi(R)\vert \ne 1$.
Suppose $R\in S_\psi$ is Anosov and $\vert\rho_\psi(R)\vert=1$. If $\rho_\psi(R)=-1$, then replacing $R$ with $R^2$ gives $\rho_\psi(R) =1$ with $R$ Anosov. Let $E^s(p)\oplus E^u(p)$ be the continuous ${\bf T}R$-invariant splitting with its associated contraction estimates. The generating vector field for $\psi$ has a continuous decomposition $X_\psi(p) = v^s(p) + v^u(p)$ for $v^s(p)\in E^s(p)$ and $v^u(p)\in E^u(p)$. From the contraction estimates of ${\bf T}_pR$ on $E^s(p)$ and $E^u(p)$ it follows for all $p\in P$ that $$\Vert {\bf T}_pR^m(v^s(p))\Vert \to 0, \ \ \Vert {\bf T}_p R^{-m}(v^u(p))\Vert\to 0{\rm\ as\ } m\to\infty.$$ With $\rho_\psi(R) = 1$ and $X_\psi(p) = v^s(p) + v^u(p)$, the equation $R_*X_\psi = \rho_\psi(R)X_\psi$ becomes $${\bf T}_p R^m (v^s(p) + v^u(p)) = X_\psi(R^m(p)) = v^s(R^m(p))+ v^u(R^m(p))$$ for all $p\in P$ and $m\in {\mathbb Z}$. The ${\bf T}R$-invariance of $E^u(p)$ implies that $${\bf T}_p R^m(v^u(p)) = v^u(R^m(p)).$$ For a fixed $p\in P$, there is by the compactness of $P$, a subsequence ${\rm R}^{m_i}(p)$ converging to a point, say $p_\infty$, as $m_i\to \infty$. Hence $$\begin{aligned}
X_\psi(p_\infty)
& = \lim_{i\to\infty} X_\psi(R^{m_i}(p)) \\
& = \lim_{i\to\infty} {\bf T}_pR^{m_i}(v^s(p)+v^u(p)) \\
& = \lim_{i\to\infty} \big[{\bf T}_p R^{m_i}(v^s(p)) + v^u(R^{m_i}(p))\big] \\
& = v^u(p_\infty).\end{aligned}$$ It now follows that $$\lim_{m\to\infty}X_\psi(R^{-m}(p_\infty)) = \lim_{m\to\infty}{\bf T}_{p_\infty} R^{-m}(X_\psi(p_\infty)) = \lim_{m\to\infty}{\bf T}_{p_\infty} R^{-m}(v^u(p_\infty))=0.$$ Compactness of $P$ implies that $X_\psi$ has a zero. But $\psi$ is an equilibrium-free flow, and therefore $\vert\rho_\psi(R)\vert\ne 1$.
Any equilibrium-free flow with a generalized symmetry that is Anosov does not have any periodic orbits according to Lemma \[noperiodic\] and Theorem \[multiplierAnosov\]. On the other hand, for an equilibrium-free flow that is without nontrivial generalized symmetries, Theorem \[multiplierAnosov\] implies that none of its generalized symmetries can be Anosov. However, the converse of Theorem \[multiplierAnosov\] is false. As illustrated next, a partially hyperbolic diffeomorphism can be a nontrivial generalized symmetry of an equilibrium-free flow without periodic orbits.
\[nonquasi\][Let $P={\mathbb T}^n={\mathbb R}^n/{\mathbb Z}^n$, the $n$-torus, equipped with global coordinates $(\theta_1,\theta_2,\dots,\theta_n)$. The flow $\psi$ generated by $$X_\psi = \frac{\partial}{\partial\theta_1} + \frac{\partial}{\partial\theta_2} + \cdot\cdot\cdot + \frac{\partial}{\partial\theta_{n-2}} + \frac{\partial}{\partial\theta_{n-1}} + \sqrt 2 \frac{\partial}{\partial\theta_n}$$ is equilibrium-free and without periodic orbits. Let $R\in{\rm Diff}({\mathbb T}^n)$ be induced by the ${\rm GL}(n,{\mathbb Z})$ matrix $$B = \begin{bmatrix} 1 & 0 & \hdots & 0 & 0 & 1 \\ 0 & 1 & \hdots & 0 & 0 & 1 \\ \vdots & \vdots & \ddots & \vdots & \vdots & \vdots \\ 0 & 0 & \hdots & 1 & 0 & 1 \\ 0 & 0 & \hdots & 0 & 1 & 1 \\ 0 & 0 & \hdots & 0 & 2 & 1\end{bmatrix},$$ that is, ${\bf T}R = B$. This $R$ is a nontrivial generalized symmetry of $\psi$ because $R_*X_\psi = (1+\sqrt 2)X_\psi$, i.e., $\rho_\psi(R) = 1+\sqrt 2$. However $B$ has an eigenvalue of $1$ of multiplicity $n-2$, and the eigenspace corresponding to this eigenvalue is an $(n-2)$-dimensional center distribution for $R$. The other two eigenvalues of $B$ are $1\pm\sqrt 2$ and the corresponding eigenspaces are the $1$-dimensional unstable and stable distributions for $R$ respectively. Thus $R$ is partially hyperbolic. ]{}
Restrictions on Multipliers
===========================
Algebraic restrictions may occur on $\rho_\Phi(R)$ for a nontrivial generalized symmetry $R$ of an equilibrium-free flow $\Phi$ on a compact manifold $P$ without boundary. This happens when there are interactions beyond $R_*X_\Phi = \rho_\Phi(R)X_\Phi$ between the dynamics of $R$ and $\Phi$ on submanifolds diffeomorphic to ${\mathbb T}^k$ for some $2\leq k\leq {\rm dim}(P)$. A [*real algebraic integer*]{} is the root of a monic polynomial with integer coefficients, and its degree is the degree of its minimal polynomial. As shown by Wilson in [@WI], there is for any integer $k$ with $2\leq k\leq {\rm dim}(P)-2$, an equilibrium-free flow $\Phi$ on $P$ which has an invariant submanifold $N$ diffeomorphic to ${\mathbb T}^k$ on which $\Phi$ is [*minimal*]{}, i.e., every orbit of $\Phi$ in $N$ is dense in $N$. The well-known prototype of a minimal flow on ${\mathbb T}^k$ is an [*irrational flow*]{}, i.e., a $\psi$ on ${\mathbb T}^k$ for which $X_\psi$ is a constant vector field whose components (or frequencies) are rationally independent (see [@FA]), i.e., linearly independent over ${\mathbb Q}$. According to Basener in [@BA], any minimal flow on ${\mathbb T}^n$ is topologically conjugate to an irrational flow.
\[subset\] If a flow $\phi$ on a compact manifold $N$ without boundary is topologically conjugate to a flow $\psi$ on ${\mathbb T}^k$ $(k={\rm dim}(N))$ and $\psi$ is minimal, then $M_\phi\subset M_\psi$ and each $\mu\in M_\phi\setminus\{1,-1\}$ is an algebraic integer of degree between $2$ and $k$ inclusively.
Suppose there is a homeomorphism $h:N\to{\mathbb T}^k$ such that $h\phi_t = \psi_t h$ for all $t\in{\mathbb R}$. Without loss of generality, it is assumed that $\psi$ is an irrational flow, since any minimal flow on ${\mathbb T}^k$ is topologically conjugate to an irrational flow.
Let $V\in S_\phi$ and set $\mu=\rho_\phi(V)$. In terms of the homeomorphism $Q=hVh^{-1}$ on ${\mathbb T}^k$, the multiplier $\mu$ passes through $h$ to $\psi$: $$Q\psi(t,\theta) = hR\phi(t,h^{-1}(\theta)) = h\phi(\mu t,Rh^{-1}(\theta)) = \psi(\mu t, Q(\theta)), {\rm\ for\ all\ }t\in{\mathbb R}, \theta \in{\mathbb T}^k,$$ i.e., $Q\psi_t = \psi_{\mu t}Q$. This does not say yet that $Q$ is a generalized symmetry of $\psi$ with multiplier $\mu$ because $Q$ is only a homeomorphism at the moment.
Following [@AP], the homeomorphism $Q$ of ${\mathbb T}^k$ lifts uniquely to $\hat Q(x)=\hat L(x) + \hat U(x) +\hat c$ on ${\mathbb R}^k$, i.e., $\pi \hat Q = Q\pi$ where $\pi:{\mathbb R}^k\to{\mathbb T}^k$ is the covering map. Here the linear part of this lift is $\hat L(x) = Bx$ for $B\in {\rm GL}(k,{\mathbb Z})$; the periodic part is $\hat U(x)$, i.e., $\hat U(x+\nu) = \hat U(x)$ for all $x\in{\mathbb R}^k$ and all $\nu\in {\mathbb Z}^k$, is continuous and satisfies $\hat U(0)=0$; and the constant part is $\hat c\in[0,1)^k$. A lift of the irrational flow $\psi$ to ${\mathbb R}^n$ is $\hat \psi(t,x) = x+td$ where $d=X_\psi$ and $x\in{\mathbb R}^k$. For all $t\in{\mathbb R}$, a lift of $Q\psi_t$ to ${\mathbb R}^k$ is $\hat Q\hat \psi_t$ and a lift of $\psi_{\mu t}Q$ is $\hat \psi_{\mu t}\hat Q$. These two lifts differ by a constant $m\in{\mathbb Z}^k$ since $Q\psi_t = \psi_{\mu t}Q$ for all $t\in{\mathbb R}$: $$\hat Q\hat \psi(t,x) = \hat \psi(\mu t,\hat Q(x)) + m {\rm\ for\ all\ }t\in{\mathbb R}, x\in{\mathbb R}^k.$$ Since $$\hat \psi(t,x) = Bx + tBd + \hat U(x+td) + \hat c$$ and $$\hat\psi(\mu t,\hat Q(x)) = Bx + \hat U(x) + \hat c + \mu t d,$$ it follows that $$\hat U(x+td) - \hat U(x) = - t(B-\mu I)d + m {\rm\ for\ all\ }t\in{\mathbb R}, x\in{\mathbb R}^k,$$ where $I$ is the identity matrix. Evaluation of this at $x=0$ gives $$\hat U(td) = -t(B-\mu I)d + m {\rm\ for\ all\ }t\in{\mathbb R}.$$ However, $\hat U$ is bounded since it is continuous and periodic. This boundedness implies that $(B-\mu I)d = 0$, and so $\hat U(td) = m$ for all $t\in{\mathbb R}$. Evaluation of this at $t=0$ shows that $m=0$ because $\hat U(0)=0$. Thus $$0=\hat U(td) = \hat U (\hat \psi_t(0)){\rm \ for\ all\ }t\in{\mathbb R}.$$ Since $\hat U$ is periodic and continuous on ${\mathbb R}^n$, it is a lift of a homeomorphism $U$ on ${\mathbb T}^n$. A lift of $U\psi_t$ is $\hat U\hat\psi_t$, and so $$0=\pi \hat U(\hat \psi_t(0)) = U(\psi_t(0)) {\rm\ for\ all\ }t\in{\mathbb R}.$$ The minimality of $\psi$ implies that $U$ is $0$ on a dense subset of ${\mathbb T}^n$. By continuity, $U = 0$ which implies that $\hat U=0$. Thus $\hat Q$ is $C^\infty$, and so $Q$ is a diffeomorphism, whence $Q\in S_\psi$ with $\rho_\psi(Q) = \mu$. Since $\mu\in M_\phi$, then $M_\phi\subset M_\psi$.
The multiplier $\mu \in M_\psi$ is a real algebraic integer of degree at most $k$ because it satisfies $(B-\mu I)d = 0$ for nonzero $d$, i.e., the characteristic polynomial of $B$ is a monic polynomial of degree $k$ with integer coefficients. The only rational roots this characteristic polynomial can have are $\pm 1$ since $\psi$ is an irrational flow, i.e., $M_\psi\cap{\mathbb Q}=\{1,-1\}$ (see Corollary 4.4 in [@BA2]). However, if $\mu \ne \pm 1$, then the minimal polynomial for $\mu$ has degree between $2$ and $k$ inclusively.
\[algebraicmultiplier\] Suppose for an equilibrium-free flow $\Phi$ on $P$ there is a $\Phi$-invariant compact submanifold $N$ without boundary and $R\in S_\Phi$ with $\vert\rho_\Phi(R)\vert \ne 1$ such that $R(N)\cap N\ne \emptyset$. If ${\rm dim}(N)=2$ and $N$ is orientable, then $\rho_\Phi(R)$ is a real algebraic integer of degree $2$. If ${\rm dim}(N)\geq 3$ with $N$ diffeomorphic to ${\mathbb T}^{{\rm dim}(N)}$ and $\Phi\vert N$ is a minimal flow, then $\rho_\Phi(R)$ is a real algebraic integer of degree between $2$ and ${\rm dim}(N)$ inclusively.
Let $\mu=\rho_\Phi(R)$ with $\vert\mu\vert\ne 1$ and $k={\rm dim}(N)$. By Theorem \[noperiodic\] there are no periodic orbits for the equilibrium-free flow $\Phi\vert N$. If $k=2$ and $N$ is orientable, the Poincaré-Bendixson Theorem implies that $N$ is diffeomorphic to ${\mathbb T}^2$, and that $\Phi\vert N$ is minimal. If $k\geq 3$, it is assumed that $N$ is diffeomorphic to ${\mathbb T}^k$ and that $\Phi\vert N$ is minimal.
The submanifold $R(N)$ is $\Phi$-invariant because $N$ is $\Phi$-invariant and $R\in S_\Phi$, i.e., for $p\in R(N)$ and $q= R^{-1}(p)\in N$, $$\Phi(t,p) = \Phi(t,R(q)) = R\Phi(t/\mu,q) \subset R(N) {\rm \ for\ all\ }t\in{\mathbb R}.$$ By hypothesis, there is $\tilde p\in R(N)\cap N$. By the $\Phi$-invariance of $N$ and $R(N)$ and the minimality of $\Phi\vert N$ it follows that $\overline{{\mathcal O}_\Phi(\tilde p)} = N$ and $\overline{{\mathcal O}_\Phi(\tilde p)} = R(N)$. This gives $R(N) = N$, i.e., that $N$ is $R$-invariant.
The nontrivial generalized symmetry $R$ restricts to a nontrivial generalized symmetry of $\Phi\vert N$ because $N$ is $\Phi$-invariant and $R$-invariant. If $V=R\vert N$ and $\phi$ is the flow on $N$ determined by $X_\phi = X_\Phi\vert N$, then $R_*X_\Phi = \mu X_\Phi$ becomes $${\bf T} V X_\phi (p) = \mu X_\phi (V(p)){\rm\ for\ }p\in N.$$ Since $V\in{\rm Diff}(N)$, then $V\in S_\phi$ with $\rho_\phi(V) = \mu$.
By [@BA], minimality of $\phi=\Phi\vert N$ with $N$ diffeomorphic to ${\mathbb T}^n$ implies that $\phi$ is topologically conjugate to an irrational flow $\psi$. Applying Lemma \[subset\] shows that $\mu$ is an algebraic integer of degree between $2$ and $k$ inclusively.
\[algebraicnature\] Suppose $\Phi$ is a minimal flow on $\mathbb T^n$, and $R\in S_\Phi$. If $\vert \rho_\Phi(R)\vert \ne 1$, then $\rho_\Phi(R)$ is an algebraic integer of degree between $2$ and $n$ inclusively.
Any nontrivial generalized symmetry $R$ of $\Phi$ together with $N=\mathbb T^n$ satisfy the conditions of Theorem \[algebraicmultiplier\].
A Group-Theoretical Characterization of Irrational Flows
========================================================
Minimality of an equilibrium-free flow on ${\mathbb T}^n$ places a semidirect product structure the generalized symmetry group of that flow. A group ${\mathcal S}$ is the [*semidirect product*]{} of two subgroups ${\mathcal N}$ and ${\mathcal H}$ if ${\mathcal N}$ is a normal subgroup of ${\mathcal S}$, if ${\mathcal S}={\mathcal N}{\mathcal H}$, and if ${\mathcal N}\cap {\mathcal H}$ is the identity element of ${\mathcal S}$. Notational this is written $${\mathcal S} = {\mathcal N}\rtimes_\Gamma {\mathcal H},$$ where $\Gamma: {\mathcal H}\to {\rm Aut}({\mathcal N})$ is the conjugating homomorphism of the semidirect product, i.e., $\Gamma({\mathfrak h})({\mathfrak n}) = {\mathfrak h}{\mathfrak n}{\mathfrak h}^{-1}$ for ${\mathfrak h}\in {\mathcal H}$ and ${\mathfrak n}\in {\mathcal N}$. A normal subgroup of $S_\psi$ is $\ker \rho_\psi$ for any flow $\psi$. A normal subgroup of ${\rm Diff}(\mathbb T^n)$ is the abelian group ${\rm Trans}(\mathbb T^n)$ of translations. Each [*translation*]{} on ${\mathbb T}^n$ is of the form ${\mathcal T}_c(\theta) = \theta + c$ for $c\in{\mathbb T}^n$. If $\psi$ is a flow on ${\mathbb T}^n$ with $X_\psi$ a constant, then ${\rm Trans}({\mathbb T}^n)\subset {\rm ker}\rho_\psi$ because $({\mathcal T}_c)_* X_\psi = X_\psi$ for all $c\in{\mathbb T}^n$.
\[kernel\] Let $\psi$ be a flow on ${\mathbb T}^n$ for which $X_\psi$ is a constant. If ${\rm ker}\rho_\psi = {\rm Trans}({\mathbb T}^n)$, then $\psi$ is irrational.
Suppose that $\psi$ is not irrational. Then the components of $X_\psi$ are not rationally independent. Up to a permutation of the coordinates $\theta_1,\dots,\theta_n$ on ${\mathbb T}^n$, i.e., a smooth conjugacy, it can be assumed the first $l$ components of $X_\psi$ are the smallest subset of the components that are linearly dependent over ${\mathbb Q}$. Specifically, writing $X_\psi = [a_1,a_2,\dots,a_n]^T$, there is a least integer $l$ with $1\leq l\leq n$ such that $$k_1a_1+\cdot\cdot\cdot + k_la_l =0$$ with $k_i\in{\mathbb Z}\setminus\{0\}$ for all $i=1,\dots,l$. The existence of an $R\in{\rm ker}\rho_\psi\setminus{\rm Trans}({\mathbb T}^n)$ will be exhibited separately in the cases of $l<n$ and $l=n$.
Case $1\leq l\leq n-1$. The $R\in{\rm Diff}({\mathbb T}^n)$ induced by the ${\rm GL}(n,{\mathbb Z})$ matrix $$B = \begin{bmatrix} 1 & 0 & \dots & 0 & 0 & 0 & \dots & 0 & 0 \\ 0 & 1 & \dots & 0 & 0 & 0 & \dots & 0 & 0 \\ \vdots & \vdots & \ddots & \vdots & \vdots & \vdots & \ddots & \vdots & \vdots \\ 0 & 0 & \dots & 1 & 0 & 0 & \dots & 0 & 0 \\ 0 & 0 & \dots & 0 & 1 & 0 & \dots & 0 & 0 \\ 0 & 0 & \dots & 0 & 0 & 1 & \dots & 0 & 0 \\ \vdots & \vdots & \ddots & \vdots & \vdots & \vdots & \ddots & \vdots & \vdots \\ 0 & 0 & \dots & 0 & 0 & 0 & \dots & 1 & 0 \\ k_1 & k_2 & \dots & k_{l-1} & k_l & 0 & \dots & 0 & 1\end{bmatrix}$$ satisfies $R_*X_\psi = X_\psi$. So $R\in {\rm ker}\rho_\psi$, but $R\not\in{\rm Trans}({\mathbb T}^n)$.
Case $l=n$. Here none of the $k_1,k_2,\dots,k_n$ are zero, and the integer-entry matrix $$B = \begin{bmatrix} 1 & 0 & \dots & 0 & 0 \\ 0 & 1 & \dots & 0 & 0 \\ \vdots & \vdots & \ddots & \vdots & \vdots \\ 0 & 0 & \dots & 1 & 0 \\ k_1 & k_2 & \dots & k_{n-1} & k_n\end{bmatrix}$$ induces a $k_n$-to-$1$ smooth surjection $g$ of ${\mathbb T}^n$ to itself. The flow $\phi$ on ${\mathbb T}^n$ generated by $X_\phi = [a_1,\dots,a_{n-1},0]^T$ is a factor of $\psi$ i.e., $g\psi_t = \phi_t g$ for all $t\in\mathbb R$, because $BX_\psi = X_\phi$.
The components $a_1,\dots,a_{n-1}$ of $X_\phi$ are rationally independent. For, if they were not, there would then be integers $k_1^\prime,\dots,k_{n-1}^\prime$, not all zero, such that $k_1^\prime a_1 + \cdot\cdot\cdot + k_{n-1}^\prime a_{n-1}=0$, which would contradict the minimality of $l=n$. With $e_1,\dots,e_n$ as the standard basis for ${\mathbb R}^n$ and $E={\rm Span}(e_1,\dots,e_{n-1})$ an $(n-1)$-dimensional subspace of ${\mathbb R}^n$, the closure of ${\mathcal O}_\phi(0)$ is $\pi(E)$. Thus $\overline{{\mathcal O}_\phi(0)}$ is an embedded submanifold of ${\mathbb T}^n$ diffeomorphic to ${\mathbb T}^{n-1}$.
The semiconjugacy $g$ between $\psi$ and $\phi$ along with $g(0)=0$ imply that $\overline{{\mathcal O}_\psi(0)}$ is an embedded submanifold of ${\mathbb T}^n$ that is diffeomorphic to ${\mathbb T}^{n-1}$. In particular, the $(n-1)$-dimensional subspace $U$ of ${\mathbb R}^n$ determined by $B(U) = E$ satisfies $\pi(U) = \overline{{\mathcal O}_\psi(0)}$, where $U={\rm Span}(u_1,u_2,\dots,u_{n-1})$ for $B(u_i) = e_i$, $i=1,\dots,n-1$. Furthermore, $$X_\psi = a_1u_1 + \cdot\cdot\cdot + a_{n-1}u_{n-1} \in U$$ because $k_1a_1+\cdot\cdot\cdot + k_na_n = 0$.
For $\epsilon>0$, $V$ a nonempty subset of ${\mathbb R}^n$, and $m\in{\mathbb Z}^n$, define $N_\epsilon(V)$ to be the set of points in ${\mathbb R}^n$ less than a distance of $\epsilon$ from $V$, and define $V+m$ to be the translation of $V$ by $m$. By the definition of $E$, if $E+m\ne E$, then $N_{1/2}(E+m)\cap N_{1/2}(E)=\emptyset$. Since $B(U)=E$ and $B({\mathbb Z}^n)\subset {\mathbb Z}^n$, there exists $\epsilon>0$ such that if $U+m\ne U$, then $N_\epsilon(U+m)\cap N_\epsilon(U)=\emptyset$.
The vector $e_n\not \in U$. Let $x_1,\dots, x_n$ be the coordinates on ${\mathbb R}^n$ that correspond to the basis $u_1,\dots,u_{n-1},e_n$. Let $f:{\mathbb R}\to{\mathbb R}$ be a smooth bump function with $f(0)=1$ and whose support has length smaller than $\epsilon$. In terms of the coordinates $x_1,\dots, x_n$, define a smooth vector field on $N_\epsilon(U)$ by $$Y = f(x_n)\frac{\partial}{\partial x_n}.$$ Extend this vector field to all of ${\mathbb R}^n$ by translation to $N_\epsilon(U+m)$ for those $m\in{\mathbb Z}^m$ for which $U+m\ne U$, and to the remainder of ${\mathbb R}^n$ as $0$. The extended vector field is globally Lipschitz, and so determines a flow $\xi$ on ${\mathbb R}^n$.
Since the vector field generating $\xi$ is invariant under translations by $m\in{\mathbb Z}$, the time-one map $\xi_1$ is also invariant under these translations. Thus $\xi_1$ is a lift of an $R\in{\rm Diff}({\mathbb T}^n)$. In terms of the coordinates $x_1,\dots, x_n$, the derivative of $\xi_1$ at any point $x\in{\mathbb R}^n$ is of the form $${\bf T}_x\xi_1 = \begin{bmatrix} 1 & 0 & \dots & 0 & 0 \\ 0 & 1 & \dots & 0 & 0 \\ \vdots & \vdots & \ddots & \vdots & \vdots \\ 0 & 0 & \dots & 1 & 0 \\ 0 & 0 & \dots & 0 & \ast \end{bmatrix},$$ and so ${\bf T}_x\xi _1(u) = u$ for $u\in U$. Since $X_\psi\in U$, this means that ${\bf T}_x \xi_1(X_\psi) = X_\psi$. Since $X_\psi$ is a constant vector field, then $(\xi_1)_* X_\psi = X_\psi$. Since $\xi_1$ is a lift of $R$, it now follows that $R_*X_\psi = X_\psi$. Therefore $R\in {\rm ker}\rho_\psi$ but $R\not\in{\rm Trans}({\mathbb T}^n)$.
The group ${\rm Aut}({\mathbb T}^n)$ of automorphisms of $\mathbb T^n$ is naturally identified with ${\rm GL}(n,{\mathbb Z})$. For $\mathcal T_c\in {\rm Trans}(\mathbb T^n)$ and $B\in {\rm GL}(n,\mathbb Z)$, the composition of $\mathcal T_c$ with $B$ is written $\mathcal T_c B= B+c$.
\[characterization\] Let $\psi$ be a flow on ${\mathbb T}^n$ with $X_\psi$ a nonzero constant vector. Then $\psi$ is irrational if and only if there exists a subgroup $H$ of ${\rm GL}(n,{\mathbb Z})$ isomorphic to $M_\psi$ such that $S_\psi = {\rm Trans}({\mathbb T}^n) \rtimes_\Gamma H$.
Suppose that $\psi$ is irrational. This implies (by Theorem 5.5 in [@BA2]) that $$S_\psi = {\rm ker}\rho_\psi \rtimes_\Gamma H_\psi,$$ where $H$ is a subgroup of ${\rm GL}(n,{\mathbb Z})$ isomorphic to $M_\psi$. Furthermore, the irrationality of $\psi$ implies (by Corollary 4.7 in [@BA2]) that ${\rm ker}\rho_\psi={\rm Trans}({\mathbb T}^n)$.
Now suppose that $\psi$ is not irrational. By Lemma \[kernel\], there is $R\in {\rm ker}\rho_\psi\setminus{\rm Trans}({\mathbb T}^n)$. If there were a subgroup $H$ of ${\rm GL}(n,{\mathbb Z})$ isomorphic to $M_\psi$ such that $S_\psi = {\rm Trans}(\mathbb T^n)\rtimes_\Gamma H$, then $R=\mathcal T_c B$ for some $c\in{\mathbb T}^n$ and $B\in H$. Hence $1=\rho_\psi(R) = \rho_\psi(\mathcal T_c)\rho_\psi(B)= \rho_\psi(B)$. However, since $H$ is isomorphic to $M_\psi$, there is only one element of $H$ which corresponds to the multiplicative identity $1$ of $\mathbb R^\times$, and that is $I$, the identity matrix. This means that $B=I$, and so $R=\mathcal T_c$, a contradiction.
Global Rigidity for Certain ${\mathbb Z}^d$ Anosov Actions
==========================================================
Global rigidity is about when a ${\mathbb Z}^d$ Anosov action, which as is well-known is topologically conjugate to an algebraic ${\mathbb Z}^d$ action (see [@KK]), is smoothly conjugate to an algebraic ${\mathbb Z}^d$ action. A ${\mathbb Z}^d$ action on ${\mathbb T}^n$ is a monomorphism $\alpha:{\mathbb Z}^d\to{\rm Diff}({\mathbb T}^n)$. It is Anosov if there is $m\in{\mathbb Z}^d\setminus\{0\}$ with $\alpha(m)$ Anosov, is [*algebraic*]{} if $\alpha({\mathbb Z}^d)\subset {\rm GL}(n,{\mathbb Z})$, and more generally is [*affine*]{} if $\alpha({\mathbb Z}^d)\subset{\rm Trans}({\mathbb T}^n)\rtimes_\Gamma{\rm GL}(n,{\mathbb Z})$. Algebraic ${\mathbb Z}^d$-actions are found in algebraic number theory (see [@KKS] and [@Sc]).
0.2cm [*Proof of Theorem \[Anosovalgebraic\].*]{} Although well-known, the argument for the existence of a common fixed point of a finite index subgroup of a ${\mathbb Z}^d$ Anosov action is included for completeness. Let $m_0\in{\mathbb Z}^d$ be such that $\alpha(m_0)$ is Anosov. Then $\alpha(m_0)$ is topologically conjugate to a hyperbolic automorphism of ${\mathbb T}^n$ (see [@Ma]), and so $\alpha(m_0)$ has a finite number of fixed points, $f_1,\dots,f_l$. Let $\alpha(m_1),\dots,\alpha(m_d)$ be a generating set for $\alpha({\mathbb Z}^d)$. Since $\alpha({\mathbb Z}^d)$ is abelian, then for all $i=1,\dots,d$ and $j=1,\dots,l$, $$\alpha(m_0)\alpha(m_i)(f_j) = \alpha(m_i)\alpha(m_0)(f_j) = \alpha(m_i)(f_j).$$ This means that $\alpha(m_i)(f_j)$ is one of the finitely many fixed points of $\alpha(m_0)$. Since each $\alpha(m_i)$ is invertible, there is a positive finite integer $r_i$ such that $\alpha(m_i)^{r_i}(f_1) = f_1$. Thus the finite index subgroup of $\alpha({\mathbb Z}^d)$ generated by $\alpha(m_1)^{r_1}, \dots,\alpha(m_d)^{r_d}$ has $f_1$ as a common fixed point.
By hypothesis, there is a $g\in{\rm Diff}({\mathbb T}^n)$ and an irrational flow $\phi$ on ${\mathbb T}^n$ for which $X_\phi = g_* X_\Phi$. By Theorem \[characterization\], there is a subgroup $H_\phi$ of ${\rm GL}(n,\mathbb Z)$ isomorphic to $M_\phi$ such that $S_\phi={\rm Trans}({\mathbb T}^n)\rtimes_\Gamma H_\phi$. For $h\in{\rm Diff}({\mathbb T}^n)$ given by $h={\mathcal T}_{-g(f_1)}\circ g$ it follows that $h_*X_\Phi = X_\phi$ because $({\mathcal T}_c)_*X_\phi = X_\phi$ for any $c\in{\mathbb T}^n$. Then $\Phi$ and $\phi$ are projectively conjugate, and so, as mentioned in Section 2, $\Delta_h(S_\phi) = S_\Phi$. The inclusion $\alpha({\mathbb Z}^d)\subset S_\Phi$ implies that $$\Delta_{h^{-1}}(\alpha({\mathbb Z}^d))\subset {\rm Trans}({\mathbb T}^n)\rtimes_\Gamma H_\phi.$$ This means that $\alpha$ is $C^\infty$-conjugate to an affine ${\mathbb Z}^d$-action.
For each $\alpha(m)$ in the finite index subgroup generated by $\alpha(m_1)^{r_1},\dots,\alpha(m_d)^{r_d}$ there are $B_m\in H_\phi$ and $c_m\in {\mathbb T}^n$ such that $\Delta_{h^{-1}}(\alpha(m)) = B_m + c_m$. Since $h(f_1) = 0$, then $$c_m=(B_m+c_m)(0) = h\circ\alpha(m)\circ h^{-1}(0) = h \circ \alpha(m)(f_1) = h(f_1) = 0.$$ This means that $\Delta_{h^{-1}}(\alpha(m))\in H_\phi$. Hence the finite index subgroup of $\alpha({\mathbb Z}^d)$ generated by $\alpha(m_1)^{r_1},\dots,\alpha(m_d)^{r_d}$, which is isomorphic to ${\mathbb Z}^d$, is $C^\infty$-conjugate to an algebraic ${\mathbb Z}^d$-action. Since $H_\phi$ contains a ${\mathbb Z}^d$ subgroup, and $M_\phi$ is isomorphic to $H_\phi$, then $M_\phi$ contains a ${\mathbb Z}^d$ subgroup. By absolute invariance of the multiplier group under projective conjugacy, $M_\Phi$ contains a ${\mathbb Z}^d$ subgroup. $\Box$
0.2cm Any quasiperiodic flow on ${\mathbb T}^n$ whose generalized symmetry group contains a ${\mathbb Z}^d$ Anosov action must possess nontrivial generalized symmetries since its multiplier group contains a ${\mathbb Z}^d$ subgroup by Theorem \[Anosovalgebraic\] (cf.Theorem \[multiplierAnosov\]). This puts a necessary condition on the quasiperiodic flows to which Theorem \[Anosovalgebraic\] does apply. The quasiperiodic flows of Koch type mentioned in the next section satisfy this necessary condition. However, there are quasiperiodic flows that do not, as is illustrated next.
\[trivialquasi\][On ${\mathbb T}^n$, let $\psi$ be the flow generated by $$X_\psi = \frac{\partial}{\partial\theta_1} + \pi\frac{\partial}{\partial\theta_2} + \cdot\cdot\cdot + \pi^{n-1} \frac{\partial}{\partial\theta_n}.$$ This flow is quasiperiodic, since if its frequencies $1,\pi,\dots,\pi^{n-1}$ were linearly dependent over ${\mathbb Q}$ then $\pi$ would be algebraic. Quasiperiodicity of $\psi$ implies that each $\mu\in M_\psi$ is a real algebraic integer of degree at most $n$, and moreover, $M_\psi\cap{\mathbb Q}=\{1,-1\}$ (see Corollary 4.4 in [@BA2]). Suppose $\mu\in M_\psi\setminus\{1,-1\}$. Then there is $R\in S_\psi$ such that $R_*X_\psi = \mu X_\psi$. Quasiperiodicity of $\psi$ implies that ${\bf T}R=B\in{\rm GL}(n,{\mathbb Z})$ (see Theorem 4.3 in [@BA2]). Then $R_*X_\psi = \mu X_\psi$ becomes $BX_\psi = \mu X_\psi$, and for $B=(b_{ij})$, it follows that $$\mu = b_{11} + b_{12}\pi + \cdot\cdot\cdot + b_{1n}\pi^{n-1}.$$ If $b_{12}=\cdot\cdot\cdot = b_{1n} = 0$, then $\mu = b_{11}\in{\mathbb Z}$. But $M_\psi\cap{\mathbb Q}=\{1,-1\}$, and so $\mu = \pm 1$. This contradiction means that one of $b_{12},\dots,b_{1n}$ is nonzero. Because each multiplier of $\psi$ is a real algebraic integer of degree at most $n$, there is a monic polynomial $l(z)$ in the polynomial ring ${\mathbb Z}[z]$ such that $l(\mu)=0$. But this implies that $\pi$ is a root of a polynomial in ${\mathbb Z}[z]$, making $\pi$ algebraic. This shows that $M_\psi=\{1,-1\}$, and so $\psi$ does not possess nontrivial generalized symmetries. ]{}
Classification of Certain Equilibrium-Free Flows
================================================
Quasiperiodic flows of Koch type are algebraic in nature and provide foliations which are often preserved by a topologically irreducible ${\mathbb Z}^d$ Anosov action (see [@BA5], [@BA6], and [@KKS] for such examples). A flow on ${\mathbb T}^n$ is [*quasiperiodic of Koch type*]{} if it is projectively conjugate to an irrational flow whose frequencies form a ${\mathbb Q}$-basis for a real algebraic number field ${\mathbb F}$ of degree $n$ over ${\mathbb Q}$. For a quasiperiodic flow $\Phi$ of Koch type, the real algebraic number field ${\mathbb F}$ of degree $n$ associated to it is unique, and its multiplier group is a finite index subgroup of the group of units ${\mathfrak o}_{\mathbb F}^\times$ in the ring of integers of ${\mathbb F}$ (see Theorem 3.3 in [@BA5]). By Dirichlet’s Unit Theorem (see [@SD]), there is $d\geq 1$ such that ${\mathfrak o}_{\mathbb F}^\times$ is isomorphic to ${\mathbb Z}_2\oplus{\mathbb Z}^d$, and so every quasiperiodic flow of Koch type always possesses nontrivial generalized symmetries.
Topological irreducibility of a ${\mathbb Z}^d$ action $\alpha$ is a condition on the topological factors that $\alpha$ has. A ${\mathbb Z}^d$ action $\alpha^\prime$ on ${\mathbb T}^{n^\prime}$ is a [*topological factor*]{} of $\alpha$ if there is a continuous surjection $h:{\mathbb T}^n\to{\mathbb T}^{n^\prime}$ such that $h\circ \alpha = \alpha^\prime\circ h$. A topological factor $\alpha^\prime$ of $\alpha$ is [*finite*]{} if the continuous surjection $h$ is finite-to-one everywhere. A ${\mathbb Z}^d$ action $\alpha$ is [*topologically irreducible*]{} if every topological factor $\alpha^\prime$ of $\alpha$ is finite.
For an algebraic ${\mathbb Z}^d$ action $\alpha$, there is a stronger sense of irreducibility, one that uses the group structure of ${\mathbb T}^n$. An algebraic ${\mathbb Z}^d$ action $\alpha^\prime$ on ${\mathbb T}^{n^\prime}$ is an [*algebraic factor*]{} of $\alpha$ if there is a continuous homomorphism $h:{\mathbb T}^n\to{\mathbb T}^{n^\prime}$ such that $h\circ \alpha = \alpha^\prime\circ h$. An algebraic factor $\alpha^\prime$ of $\alpha$ is [*finite*]{} if the continuous homomorphism $h$ is finite-to-one everywhere. An algebraic ${\mathbb Z}^d$ action $\alpha$ is [*algebraically irreducible*]{} if every algebraic factor $\alpha^\prime$ of $\alpha$ is finite. Algebraic irreducibility of a higher rank algebraic $\mathbb Z^d$ action $\alpha$ is equivalent to there being an $m\in \mathbb Z^d$ such that $\alpha(m)$ has an irreducible characteristic polynomial (see Proposition 3.1 on p.726 in [@KKS]; cf.[@BE]).
0.2cm [*Proof of Theorem \[flowKoch\].*]{} Identify ${\bf T}{\mathbb T}^n$ with ${\mathbb T}^n\times{\mathbb R}^n$, and place on the fiber the standard Euclidean norm $\Vert\cdot\Vert$. By the hypotheses, there is $h\in{\rm Diff}({\mathbb T}^n)$ and a hyperbolic $B\in{\rm GL}(n,{\mathbb Z})$ such that $\Delta_{h^{-1}}(\alpha(m_0)) = B$. Every point of ${\mathbb T}^n$ is Lyapunov regular for $B$. The Oseledets decomposition associated with $\chi_B$ is $${\bf T}_\theta {\mathbb T}^n = \bigoplus_{i=1}^k E^i_B,$$ where $E_B^i$, $i=1,\dots,k$, are the invariant subspaces of $B$ which are independent of $\theta$. Since $\Delta_{h^{-1}}(\alpha(m_0))=B$, every point of ${\mathbb T}^n$ is Lyapunov regular for $\alpha(m_0)$. Set $$E^i_{\alpha(m_0)}(\theta) = {\bf T}_{h(\theta)}h^{-1}\big( E^i_B), \ \ \theta\in{\mathbb T}^n.$$ The Oseledets decomposition associated with $\chi_{\alpha(m_0)}$ is then $${\bf T}_\theta {\mathbb T}^n = \bigoplus_{i=1}^k E^i_{\alpha(m_0)}(\theta).$$
For $\mu = \rho_\Phi(\alpha(m_0))$, the hypothesis that the multiplicity of the value $\log\vert \mu \vert$ of $\chi_{\alpha(m_0)}$ is one at a point $\hat \theta\in{\mathbb T}^n$ implies that there is $1\leq l\leq k$ such that ${\rm dim}\big(E^l_{\alpha(m_0)}(\hat\theta)\big) = 1$ and $$\chi_{\alpha(m_0)}(\hat\theta,v) = \log\vert \mu \vert {\rm\ for\ } v\in E^l_{\alpha(m_0)}(\hat\theta)\setminus\{0\}.$$ By the definition $E^l_{\alpha(m_0)}(\hat\theta) = {\bf T}_{h(\hat\theta)}h^{-1}\big( E^l_B)$, it follows that ${\rm dim}\big( E^l_B\big) = 1$. Furthermore, since $\Delta_{h^{-1}}(\alpha(m_0))=B$ and $\chi_B$ is independent of $\theta$, it follows that $\chi_B(\theta,v) = \log\vert\mu\vert$ for all $v\in E^l_B\setminus\{0\}$ and for all $\theta\in{\mathbb T}^n$. The definition $E^l_{\alpha(m_0)}(\theta) = {\bf T}_{h(\theta)}h^{-1}\big( E^l_B)$ and the independence of $E^l_B$ from $\theta$ imply for all $\theta\in{\mathbb T}^n$ that ${\rm dim} \big(E^l_{\alpha(m_0)}(\theta)\big) = 1$ and $\chi_{\alpha(m_0)}(\theta,v) = \log\vert\mu\vert$ for all $v\in E^l_{\alpha(m_0)}(\theta)\setminus\{0\}$. Hence, the multiplicity of $\log\vert\mu\vert$ for $\chi_{\alpha(m_0)}$ is one for all $\theta\in{\mathbb T}^n$.
By Theorem \[Lyapunov\], the $\alpha(m_0)$-invariant one-dimensional distribution $E$ given by $E(\theta) = {\rm Span}(X_\Phi(\theta))$ satisfies $\chi_{\alpha(m_0)}(\theta,X_\Phi(\theta))=\log\vert\mu\vert$ for all $\theta\in{\mathbb T}^n$. If $E(\theta)\ne E^l_{\alpha(m_0)}(\theta)$ at some $\theta\in{\mathbb T}^n$, then $E(\theta)+E^l_{\alpha(m_0)}(\theta)$ is a two-dimensional subspace of ${\bf T}_\theta {\mathbb T}^n$ for which $\chi_{\alpha(m_0)}(\theta, v) = \log\vert\mu\vert$ for all $v\in \big(E(\theta)+ E^l_{\alpha(m_0)}(\theta)\big)\setminus\{0\}$. This contradicts the multiplicity of $\log\vert\mu\vert$ for $\chi_{\alpha(m_0)}$ being one at every $\theta$. Thus $E^l_{\alpha(m_0)}(\theta) = E(\theta) = {\rm Span}(X_\Phi(\theta))$ for all $\theta\in{\mathbb T}^n$.
The vector field $h_*X_\Phi$ satisfies $h_*X_\Phi(\theta)\in E^l_B$ for all $\theta\in{\mathbb T}^n$ because ${\rm Span}(X_\Phi(\theta)) = {\bf T}_{h(\theta)} h^{-1}(E^l_B)$ for all $\theta\in{\mathbb T}^n$. Let $\psi$ be the flow for which $X_\psi = h_*X_\Phi$. Since $E^l_B$ is a one-dimensional invariant subspace of $B$ and $\chi_B(\theta,v) = \log\vert\mu\vert$ for all $v\in E^l_B\setminus\{0\}$, it follows for all $\theta\in{\mathbb T}^n$ that $$\Vert B^kX_\psi(\theta)\Vert = \vert \mu\vert^k \Vert X_\psi(\theta)\Vert {\rm \ for\ all\ }k\in{\mathbb Z}.$$
Hyperbolicity of $B$ implies that there is $\bar\theta\in{\mathbb T}^n$ such that ${\mathcal O}_B(\bar\theta)=\{ B^k(\bar\theta):k\in{\mathbb Z}\}$ is dense in ${\mathbb T}^n$. Since $\alpha(m_0)_*X_\Phi = \mu X_\Phi$ and $X_\psi = h_*X_\Phi$, the matrix $B$ satisfies $BX_\psi =\mu X_\psi B$. Then $B^k X_\psi = \mu^k X_\psi B^k$, and so $X_\psi B^k = \mu^{-k} B^k X_\psi$. Thus, $$\Vert X_\psi(B^k\bar\theta)\Vert = \vert \mu\vert^{-k} \Vert B^k X_\psi(\bar\theta)\Vert = \vert\mu\vert^{-k}\vert\mu\vert^k \Vert X_\psi(\bar\theta)\Vert = \Vert X_\psi(\bar\theta)\Vert {\rm \ for\ all\ }k\in{\mathbb Z}.$$ Denseness of ${\mathcal O}_B(\bar\theta)$ and continuity of $X_\psi$ imply that $\Vert X_\psi(\theta)\Vert = \Vert X_\psi(\bar\theta)\Vert$ for all $\theta\in{\mathbb T}^n$. The one-dimensionality of $E^l_B$ to which $X_\psi$ belongs implies that $X_\psi$ is a constant vector. Thus $X_\psi$ is an eigenvector of $B$.
The assumed topological irreducibility of $\alpha$ and the inclusion $\Delta_{h^{-1}}(\alpha({\mathbb Z}^d))\subset {\rm GL}(n,{\mathbb Z})$ imply that $\Delta_{h^{-1}}(\alpha(\mathbb Z^d))$ is algebraically irreducible. Thus there is $B^\prime\in \Delta_{h^{-1}}(\alpha({\mathbb Z}^d))$ with an irreducible characteristic polynomial. Since $BB^\prime = B^\prime B$ and $B^\prime$ has an irreducible characteristic polynomial, the eigenvector $X_\psi$ of $B$ is an eigenvector of $B^\prime$ too. Then there is $\vartheta\in{\mathbb R}^\times$ such that the components of $\vartheta^{-1}X_\psi$ form a ${\mathbb Q}$-basis for an algebraic number field of degree $n$ over ${\mathbb Q}$ (see Propositions 1 and 8 in [@WA]). Thus the flow $\phi$ determined by $X_\phi = \vartheta^{-1}X_\psi$ is irrational of Koch type for which $h_*X_\Phi = X_\psi = \vartheta X_\phi$. $\Box$
[AA]{}
R.L. Adler and R. Palais, [*Homeomorphic Conjugacy of Automorphisms on the Torus*]{}, Proc. Amer. Math. Soc., Vol. 16, No. 6 (1965), 1222-1225.
L.F. Bakker, [*Structure of Group Invariants of a Quasiperiodic Flow*]{}, Electron. J. of Differential Equations, Vol. 2004 (2004), No. 39, 1-14.
L.F. Bakker, [*Rigidity of Projective Conjugacy of Quasiperiodic Flows of Koch Type*]{}, Colloquium Mathematicum, Vol. 112, No. 2 (2008), 291-312.
L.F. Bakker, [*Measurably Nonconjugate Higher-Rank Abelian Non-Cartan Actions*]{}, to appear in Proceedings of the 5th International Conference of Dynamic Systems and Applications.
L.F. Bakker and G. Conner, [*A Class of Generalized Symmetries of Smooth Flows*]{}, Communications on Pure and Applied Analysis, Vol. 3, No. 2 (2004), 183-195.
L. Barreira and Y.B. Pesin, [*Lyapunov Exponents and Smooth Ergodic Theory*]{}, University Lecture Series, Vol. 23, American Mathematical Society, 2002.
W. Basener, [*Geometry of minimal flows*]{}, Topology and its Applications, 153 (2006), 3627-3632.
D. Berend, [*Multi-Invariant Sets on Tori*]{}, Trans. Amer. Math. Soc., Vol. 280, No. 2 (1983), 509-532.
B.R. Fayad, [*Analytic mixing reparameterizations of irrational flows*]{}, Ergod. Th. & Dynam. Sys., 22 (2002), 437-468.
A. Gorodnik, [*Open problems in dynamics and related fields*]{}, J. of Mod. Dyn., Vol. 1, No. 1 (2007), 1-35.
B. Kalinin and A. Katok, [*Measure rigidity beyond uniform hyperbolicity: Invariant measures for Cartan actions on tori*]{}, J. of Mod. Dyn., Vol. 1, No. 1 (2007), 123-146.
A. Katok and B. Hasselblatt, “Introduction to the Modern Theory of Dynamical Systems”, Encyclopedia of Mathematics and its Applications, Vol. 54, Cambridge University Press, 1995.
A. Katok, S. Katok, and K. Schmidt, [*Rigidity of measurable structures for ${\mathbb Z}^d$-actions by automorphims of a torus,*]{} Comment. Math. Helv. 77 (2002), 718-745.
A. Katok and J.W. Lewis, [*Local rigidity for certain groups of toral automorphisms*]{}, Israel. J. Math 75 (1991), 203-241.
A. Katok and J.W. Lewis, [*Global rigidity results for lattice actions on tori and new examples of volume-preserving actions*]{}, Israel. J. Math. 93 (1996), 253-280. B. Kalinin and R. Spatzier, [*On the classification of Cartan Actions*]{}, Geom. Funct. Anal. Vol. 17, No. 2 (2007), 468-490.
H. Koch, [*A renormalization group for Hamiltonians, with applications to KAM tori,*]{} Ergod. Th. & Dynam. Sys. 19 (1999), 475-521.
J. Lopes Dias, [*Renormalization of flows on the multidimensional torus close to a KT frequency vector*]{}, Nonlinearity 15 (2002), 647-664.
A. Manning, [*There are no new Anosov diffeomorphisms on tori*]{}, Amer. J. of Math., Vol. 96, No. 3 (1974), 422-429.
J. Palis and J.C. Yoccoz, [*Centralizers of Anosov Diffeomorphisms on tori*]{}, Annales scientifiques de l’É.N.S. $4^{e}$ série, tome 22, $n^o$ 1 (1989), 99-108.
F. Rodriguez Hertz, [*Global Rigidity of Certain Abelian Actions by Toral Automorphisms*]{}, J. Mod. Dyn., Vol. 1, No. 3 (2007), 425-442.
P.R. Sad, [*Centralizers of Vector Fields*]{}, Topology Vol. 18 (1979), 97-114.
K. Schmidt, “Dynamical Systems of Algebraic Origin,” Progress in Mathematics Vol. 128, Birkhäuser,1995.
H.P.F. Swinnerton-Dyer, [*A Brief Guide to Algebraic Number Theory*]{}, Cambridge University Press. 2001 Cambridge, 2001.
D.I. Wallace, [*Conjugacy Classes of Hyperbolic Matrices in ${\rm SL}(n,{\mathbb Z})$ and Ideal Classes in an Order*]{}, Trans. Amer. Math. Soc., Vol. 283, No. 1 (1984), 177-184.
F. W. Wilson, Jr., [*On the minimal sets of non-singular vector fields*]{}, Ann. of Math. 2nd Ser., Vol. 84, No. 3 (1966), 529-536.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Causal inference with observational data can be performed under an assumption of no unobserved confounders (unconfoundedness assumption). There is, however, seldom clear subject-matter or empirical evidence for such an assumption. We therefore develop uncertainty intervals for average causal effects based on outcome regression estimators and doubly robust estimators, which provide inference taking into account both sampling variability and uncertainty due to unobserved confounders. In contrast with sampling variation, uncertainty due unobserved confounding does not decrease with increasing sample size. The intervals introduced are obtained by deriving the bias of the estimators due to unobserved confounders. We are thus also able to contrast the size of the bias due to violation of the unconfoundedness assumption, with bias due to misspecification of the models used to explain potential outcomes. This is illustrated through numerical experiments where bias due to moderate unobserved confounding dominates misspecification bias for typical situations in terms of sample size and modeling assumptions. We also study the empirical coverage of the uncertainty intervals introduced and apply the results to a study of the effect of regular food intake on health. An R-package implementing the inference proposed is available.'
author:
- |
**[Minna Genbäck]{}\
CEDAR, Umeå University, 901 87, Umeå, Sweden.\
*email:* [email protected]\
\
**[and]{}\
\
**[Xavier de Luna]{}\
Department of Statistics, USBE, Umeå University, 901 87, Umeå, Sweden.\
*email:* [email protected]******
bibliography:
- 'referenser.bib'
title: Causal inference taking into account unobserved confounding
---
*Keywords:* Average treatment effect; double robust; ignorability assumption; regular food intake; sensitivity analysis; uncertainty intervals.
Introduction
============
In observational studies, a causal effect of a treatment can be identified given the assumption that all variables confounding the effect on the outcome of interest are observed. This unconfoundedness assumption (also called ignorability of the treatment assignment mechanism) is regarded as the Achilles heel of non-experimental studies ([@Liu:2013]) and is not testable without further information (e.g., [@Luna:2006; @LunaJ:2014]). We therefore develop in this paper uncertainty intervals ([@Vansteelandt:2006]) for average causal effects based on outcome regression estimators and doubly robust estimators, which provide inference taking into account sampling variability and uncertainty due to unobserved confounders. The intervals are obtained by deriving the bias of the estimators due to unobserved confounders as a function of a parameter $\rho$ (called bias parameter in the sequel) quantifying the amount of unobserved confounding. Using the bias expressions, we deduce bounds on the average causal effects (an identification set in contrast to point identification available under unconfoundedness). Combining these bounds with sampling variability yields uncertainty intervals that have the property to cover the parameter of interest with higher probability than an a priori chosen level (say 95%). The bounds obtained are useful when information on the parameter $\rho$ is available since they can then be made tighter, in contrast with worst case scenario bounds (e.g., [@M:03], [@HM:06]).
The approach taken here is directly related to the quickly expanding literature on methods to perform a sensitivity analysis to the unconfoundedness assumption ([@R:10], Chap. 14). And, indeed, the uncertainty intervals proposed may be used to perform such a sensitivity analysis whereby the maximum value of the bias parameter is presented for which the uncertainty interval covers zero (no causal effect). Among existing methods to perform sensitivity analyses, many are based on specifying parametric models on how a potential confounder affects the outcome and treatment assignment given the observed covariates, thereby introducing bias parameters (one for the effect of the confounder on the observed outcome and the other for the effect of the confounder on the treatment). Then, typically, using some distributional assumptions for the hypothetical unobserved confounder, the latter is integrated out in order to obtain the bias of an estimator as function of the bias parameters; see, e.g., [@R:10; @Lin:1998; @Robins:2000; @Rosenbaum; @I:03; @VanderWeele:2011] using a frequentist approach, and [@Greenland:2005]; [@Luna:2014] using a Bayesian framework. The confounder and outcome are often assumed binary, but some approaches allow for a continuous confounder and/or outcome (e.g., [@VanderWeele:2011]). A directly related literature deals with sensitivity analyses to departures from the ignorability assumption of a missing outcome data mechanism (missing at random assumption); see, e.g., [@Copas:01; @CE:05; @SDR:03; @DH:08]. In fact, by using the potential outcome framework ([@Rubin:1974]), the estimation of a causal effect can be cast into a problem of missing outcome (unobserved potential outcomes) and sensitivity analyses for the missing at random assumption can readily be used to study deviations from the unconfoundedness assumption. Alternative approaches to parametrising the relation between a potential unobserved confounder and the outcome and treatment, is to define the bias due to non-ignorability (treatment assignment/missingness mechanism) as the bias parameter, and, e.g., put a prior on this bias within a Bayesian framework ([@DH:08; @Josefsson:2016]). Finally, an approach we find appealing from an interpretation point of view is to consider as the bias parameter, the correlation (induced by unobserved confounders) between the treatment assignment and the potential outcomes given the observed covariates; see [@CopasLi] and [@Genback:2015] within a missing outcome context, and [@Imai:2010] within a parametric mediation analysis context. This approach has the advantage of introducing only one bias parameter for each missingness mechanism.
In this paper, we build upon the latter alternative to perform inference on a causal parameter that takes into account uncertainty due to unobserved confounding and sampling variability. When estimating the average causal effect, two data missingness mechanisms must be considered (one for the outcome under treatment and one for the outcome when no treatment is assigned) implying the need for two ignorability assumptions for point-identification, i.e. two bias parameters. On the other hand, if the interest lies solely in an average treatment effect on the treated (or the non-treated), then only one missingness mechanism has to be dealt with, thereby only one bias parameter. We obtain bounds on the causal effect of interest by deducing the bias of the estimators as function of the bias parameter(s). Thus, we are also able to contribute by contrasting the size of the bias of the outcome regression estimator due to, i) violation of the unconfoundedness assumption, and to ii) the misspecification of the models used to explain outcome. Indeed three types of uncertainty can be distinguished: sampling variation, model misspecification and unobserved confounding. Sampling variation decreases as sample size increases. Model misspecification bias may be tackled with double robust estimation, and in any case this bias can in principle be made arbitrarily small with larger samples, e.g. under sparsity assumptions, by increasing the flexibility of the models used. On the other hand, bias due to unobserved confounding, does not disappear with increasing sample size as long as unobserved confounders are omitted, and is therefore essential to take into account in observational studies.
The paper is organized as follows. First, a framework for deducing bounds based on a parametrised model is introduced in Section 2. In Section 3 we focus on outcome regression and double robust estimators of average causal effects. We deduce their bias under confounding and show that confounding bias and model misspecification bias are separable. From confounding bias expressions we obtain bounds, and their corresponding uncertainty intervals for the parameter of interest. The R-package `ui` implements the methods proposed (available at `http://stat4reg.se/software`). In Section 4 simulated experiments are conducted to study the relative size of the biases due to confounding and model misspecification, as well to investigate empirical coverage of the uncertainty intervals proposed. In Section 5 we perform a sensitivity analysis in a real data example. The paper is concluded in Section 6.
Identification and sampling variation
=====================================
Model for point identification
------------------------------
Let $y^1$ and $y^0$ be two potential outcomes, where $y^1$ is the outcome when treatment $1$ is assigned ($z=1$), and $y^0$ is the outcome when treatment $0$ is assigned ($z=0$). The two potential outcomes are defined for each individual in the study although only one is observed ($y^1$ is observed when $z=1$, and $y^0$ is observed when $z=0$). We further assume that a set of covariates $x$ is observed for all individuals. To allow for a more compact notation in the following sections we let the first element of $x$ represent the intercept. The evaluation of a treatment effect on the outcome may be done by considering average effects. In this paper we focus on both $E(y^1-y^0)=\tau$, the average causal effect, and $E(y^1-y^0\mid z=1)=\tau^1$, the average causal effect on the treated.
Without loss of generality, let us write $$\label{outcome.mod}
y^0=f^0(x)+\varepsilon^0,\ \ \ \ \ y^1=f^1(x)+\varepsilon^1,$$ where $f^j(x)$ $j=0,1$, are functions of $x$ and $E(\varepsilon^0\mid x)=E(\varepsilon^1\mid x)=0$. Let further $$\label{assignment.mod}
z^*=g(x)+\eta,\ \mbox{and}\ \ \ z={\bf I}(z^*>0),$$ where $\bf I(\cdot)$ is an indicator function, $z^*$ is not observed, $g(x)$ is a function of $x$ and $E(\eta\mid x)=0$.
We can now give sufficient conditions for point identification of $\tau^1$ and $\tau$.
[Ass]{} \[Ass1\]
\[Ass1a\] $\varepsilon^0{\mbox{\protect $\: \perp \hspace{-2.3ex}\perp$ }}\eta\mid x$.
\[Ass1b\] $\Pr(z=0\mid x)>0, \forall x \in \mathcal{X}$, where $\mathcal{X}$ is the support of $x.$
[Ass]{} \[Ass2\]
\[Ass2a\] $\varepsilon^1{\mbox{\protect $\: \perp \hspace{-2.3ex}\perp$ }}\eta\mid x$.
\[Ass2b\] $\Pr(z=1\mid x)>0, \forall x \in \mathcal{X}$, where $\mathcal{X}$ is the support of $x.$
Assumption \[Ass1a\] and \[Ass2a\] are often called unconfoundedness or ignorability assumptions. We have that $\tau^1$ and $\tau$ are point identified under Assumption \[Ass1\] and under Assumptions \[Ass1\] and \[Ass2\] respectively ([@RR:83]).
Bias parameters and uncertainty intervals {#Identif.sec}
-----------------------------------------
The ignorability assumptions of the treatment assignment mechanism cannot be tested with the observed data unless extra information is available, e.g, instrumental variables; see [@LunaJ:2014]. Thus, unless Assumptions \[Ass1a\] and \[Ass2a\] are true by design of the study, uncertainty about their validity should be taken into account in the inference. For this purpose, it is useful to parametrise deviations from Assumptions \[Ass1a\] and \[Ass2a\] The following parametrization models realistic deviations and is easy to communicate to potential users.
\[Ass3\] Consider model (\[outcome.mod\]-\[assignment.mod\]) with $g(x)= g(x; \gamma)$, for an unknown parameter $\gamma$, and, for $j=0, 1$, let $\sigma_j^2=\mbox{Var}(y^j\mid x)<\infty$, $\eta \sim \mbox{N}(0, 1)$, $\varepsilon^j= \rho_j \sigma_j \eta + \xi_j$, $E(\xi_j)=0$, ${{\mbox{Var}}}(\xi_j)=\sigma_j\sqrt{1-\rho^2}$ and $\xi_j {\mbox{\protect $\: \perp \hspace{-2.3ex}\perp$ }}\eta$.
Here we have introduced the bias parameters $\rho_0=Corr(\varepsilon^0,\eta)$ and $\rho_1=Corr(\varepsilon^1,\eta)$. This model is such that Assumption \[Ass1a\] holds when $\rho_0=0$ and not otherwise, and Assumption \[Ass2a\] holds when $\rho_1=0$ and not otherwise. Hence, these parameters describe departures from ignorability of the treatment assignment mechanism. We call them bias parameters since they tune the bias that will result from assuming ignorability. The normality of $\eta$ corresponds to a choice of link function for (\[assignment.mod\]) which is convenient mathematically in the sequel, but is not otherwise essential in the model. Note, morevoer, that normality may be relaxed to a more general class of distributions ([@Genback:2015], Sec. 3.2).
If we have unmeasured confounders, one way to interpret $\rho_j$ is to rewrite the error terms $\varepsilon^0$, $\varepsilon^1$ and $\eta$ from equations (\[outcome.mod\]) and (\[assignment.mod\]) as the sum of the error that can and cannot be explained by the unmeasured confounder. For instance, if we believe that unmeasured confounder(s) explain $a\cdot100\, \%$ of the variation in $\eta$ and $b\cdot100\, \%$ of the variation in $\varepsilon^j$, and that the unmeasured confounders affect treatment assignment negatively and $y^j$ positively, then $\rho_j=(-\sqrt{a})(+\sqrt{b})$.
The approach proposed here is to deduce an identification interval for the parameter of interest $\tau$ by using an estimator $\hat\tau$ which is unbiased for $\tau$ under unconfoundedness ($\rho_0=\rho_1=0$). Then, the bias of the estimator is computed as a function of the bias parameter $b(\rho_0,\rho_1;\theta)=E(\hat\tau)-\tau$, where $\theta$ is a nuisance parameter vector (containing $f^0$, $f^1$, $\sigma$ and $\gamma$). Finally, this bias expression together with out-of-data information (if any) on $\rho_0$ and/or $\rho_1$ in the form of an interval $\rho_j \in [\rho_j^L ,\rho_j^U]$ yields an identification interval for $\tau$: $$\label{ident.set}
\{\tau: \tau=E_0(\hat\tau)-b(\rho_0,\rho_1;\theta), \rho_0\in [\rho_0^L,\rho_0^U], \rho_1\in [\rho_1^L,\rho_1^U], \theta = \theta_0 \},$$ where $\theta_0$ is the true value of $\theta$, and $E_0$ is the expectation taken over the observed data law, i.e. corresponding to the true but unknown values for $\rho_0$ and $\rho_1$. Note that $\rho_j \in [-1, 1]$ is the no out-of-data information case. In some applications, however, one may have out-of-data information, for instance that the treatment assignment is not negatively correlated with the outcome, $\rho_j \in [0,1 -\delta]$, for some $\delta>0$. Another instance arises when a rich and relevant set of covariates $x$ is available, in which case one may believe that $\rho_j \in [-\delta,\delta]$ for $\delta\geq 0$ small.
In situations where a consistent and asymptotically normal estimator of $E(\hat\tau)-b(\rho_0,\rho_1;\theta)$ is available, denoted $\hat\tau-\hat b(\rho_0,\rho_1)$, with corresponding standard errors, $s.e.(\hat\tau-\hat b(\rho_0,\rho_1))$, an uncertainty interval containing $\tau$ with probability at least $1-\alpha$ ([@Tanja]) is given by: $$UI(\tau; [\rho_0^L,\rho_0^U], [\rho_1^L,\rho_1^U],\alpha)=\bigcup_{\rho_0 \in [\rho_0^L,\rho_0^U],\rho_1\in [\rho_1^L,\rho_1^U]} CI(\tau; \rho_0,\rho_1,\alpha),
\label{UI.eqn}$$ where $$CI(\tau; \rho_0,\rho_1,\alpha)=\left( \hat\tau-\hat b(\rho_0,\rho_1) - c_{\frac{\alpha}{2}}s.e.(\hat\tau-\hat b(\rho_0,\rho_1)), \hat\tau-\hat b(\rho_0,\rho_1) + c_{\frac{\alpha}{2}}s.e.(\hat\tau-\hat b(\rho_0,\rho_1)) \right)
\label{CI.eqn}$$ and $c_{\frac{\alpha}{2}}$ is the $(1-\alpha/2)100 \%$ percentile of the standard normal distribution.
Outcome regression and doubly robust estimators: bias and inference {#Bias.sec}
===================================================================
We now consider two families of estimators and apply the approach described above to deduce their bias and the resulting uncertainty intervals. We will use the following assumption of correctly specified regression models.
[Ass]{} \[Ass4\]
$f^0(x)=\beta^{0'}x$, \[Ass4a\]
$f^1(x)=\beta^{1'}x$, \[Ass4b\]
where $\beta^0$ and $\beta^1$ are parameter vectors, and the first element of $x$ is 1. Assumption 4 can be made very general by replacing $\beta' x$ by $\tilde \beta' w$ where $w$ includes bases functions of the space spanned by $x$, e.g. cubic splines.
Estimators of average causal effects {#Estimators}
------------------------------------
Let us assume that we have a random sample of size $n$ of which $n_1$ are treated and $n_0$ are controls (not treated), and let $\mathcal{I}_1$ be the indexes for the treated and $\mathcal{I}_0$ be the indexes for the controls. We denote $\hat\beta^j_{OLS}=({{\bf X}}_j'{{\bf X}}_j)^{-1}{{\bf X}}_j' {{\bf y}}_j$, for $j=0,1$, where ${{\bf X}}_j$ is a matrix of size $n_j \times (p+1)$ containing the elements $\{x_i' ; i \in \mathcal{I}_j\} $, $p$ is the number of covariates and ${{\bf y}}_j$ is a vector with elements $\{y_i ; i \in \mathcal{I}_j\}$. We consider the following outcome regression estimators for the average causal effect $\tau$ and average causal effect on the treated $\tau^1$ (e.g., [@Tan:2007]): $$\hat\tau^1_{OR} = \frac{1}{n_1}\sum \limits_{i =1}^n z_i\left(y_i-\hat\beta^{0'}_{OLS}x_i\right),$$ $$\hat\tau_{OR}=\frac{1}{n}\sum_{i=1}^{n}\left( \hat\beta^{1'}_{OLS}x_i-\hat\beta^{0'}_{OLS}x_i\right).$$ The outcome regression estimator is an unbiased estimate of $\tau^1$ (or $\tau$) under Assumption \[Ass1\] (and \[Ass2\]), and Assumption \[Ass4a\] (and b.). The doubly robust estimator consists of the outcome regression estimators above with a correction term for the potential misspecification of $f^j(x)$. For $\tau^1$ and $\tau$, the doubly robust estimators are (e.g., [@SRR:99], [@Lunceford:2004] and [@RF:13]): $$\begin{aligned}
\label{DR1.eq}
\hat{\tau}^1_{DR}&=&\hat\tau^1_{OR}-\frac{1}{n_1}\sum \limits_{i =1}^n(1-z_i)\frac{y_i-\hat\beta^{0'}_{OLS}x_i}{1-\hat p(x_i)} ,\end{aligned}$$ $$\label{DR.eq}
\hat{\tau}_{DR}=\hat \tau_{OR}
+\frac{1}{n}\sum \limits_{i =1}^n z_i \frac{y_i-\hat\beta^{1'}_{OLS}x_i}{\hat{p}(x_i)}
-\frac{1}{n}\sum \limits_{i =1}^n (1-z_i)\frac{y_i-\hat\beta^{0'}_{OLS}x_i}{1-\hat{p}(x_i)},$$ where $\hat{p}(x_i)$ is an estimate of the propensity score $p(x_i)=\Pr(z=1\mid x_i)$. The doubly robust estimators are unbiased under Assumption \[Ass1\] (and \[Ass2\]), and Assumption \[Ass4a\] (and b.) and/or a correctly specified propensity score model, see below.
Bias expressions {#Bias.section}
----------------
For the sake of simplicity we denote the (total) bias of an estimator $\hat \tau$ by bias$_T(\hat \tau)=b(\hat \tau, \rho_0,\rho_1;\theta)$. We investigate two sources of bias, the bias due to model misspecification (Assumption \[Ass4\] not fulfilled), bias$_M$, and the bias due to unobserved confounding (non-ignorability of the treatment assignment mechanism, Assumption \[Ass3\]), bias$_C$, as summarized in Table \[tab:bias.overview\]. All proofs are given in Appendix B.
*Bias of OR estimators under correctly specified models.*\
Under Assumptions \[Ass1b\], \[Ass3\], \[Ass4a\] and Regularity Assumption \[RegAss1\] in Appendix A, $$\begin{aligned}
{\lim \limits_{n\to\infty}}\emph{bias}_T (\hat \tau_{OR}^1)&={\lim \limits_{n\to\infty}}\emph{bias}_C (\hat \tau_{OR}^1) =\\
&=\rho_0\sigma_0 \left(E(\lambda_1(g(x; \gamma))\mid z=1)+ E(({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' \boldsymbol \lambda_0) E(x\mid z=1)\right),
\end{aligned}$$ and under Assumptions \[Ass1b\], \[Ass2b\], \[Ass3\] and \[Ass4\], $$\begin{aligned}
\emph{bias}_T (\hat \tau_{OR})= \emph{bias}_C (\hat \tau_{OR})= \frac{1}{n}{{\mbox{E}}}\left[ {\bf1}_n {{\bf X}}\left(\rho_1\sigma_1({{\bf X}}_1'{{\bf X}}_1)^{-1}{{\bf X}}_1' \boldsymbol\lambda_1 +\rho_0\sigma_0({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' \boldsymbol \lambda_0 \right)\right],
\end{aligned}$$ where ${\bf1}_n$ is a vector with all elements 1 of length $n$, and $\boldsymbol \lambda_j$ is a vector of length $n_j$ containing the elements $\{\lambda_j(g(x_i; \gamma)) ; i \in \mathcal{I}_j\} $, $\lambda_j$ is the inverse Mill’s ratio $\lambda_0(g(x_i; \gamma))=\frac{\phi(g(x_i; \gamma))}{1-\Phi(g(x_i; \gamma))}$ and $\lambda_1(g(x_i; \gamma))=\frac{\phi(g(x_i; \gamma))}{\Phi(g(x_i; \gamma))}$, and $\phi$ and $\Phi$ are the normal pdf and cdf. \[Prop.OR1\]
Let us further investigate model misspecification bias in combination with non-ignorability of treatment.
*Bias of OR estimators under model misspecification.*\
Under Assumptions \[Ass1b\] ,\[Ass3\] and Regularity Assumption \[RegAss1\] in Appendix A, $$\begin{aligned}
{\lim \limits_{n\to\infty}}\emph{bias}_T(\hat \tau^1_{OR}) &={\lim \limits_{n\to\infty}}( \emph{bias}_C(\hat \tau^1_{OR})+ \emph{bias}_M(\hat \tau^1_{OR}))=\\
&= \emph{bias}_C(\hat \tau^1_{OR})+ {{\mbox{E}}}\left[ f^0(x) |z=1\right] - E\left[({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' f^0(x) \right] {{\mbox{E}}}(x|z=1),
\end{aligned}$$ and under Assumptions \[Ass1b\], \[Ass2b\] and \[Ass3\], $$\begin{aligned}
\emph{bias}_T(\hat \tau_{OR})&= \emph{bias}_C(\hat \tau_{OR})+ \emph{bias}_M(\hat \tau_{OR})=\\
&=\emph{bias}_C(\hat \tau_{OR})+{{\mbox{E}}}\left(z f^0(x) \right) - \frac{1}{n}{{\mbox{E}}}\left[ {\bf1}_{n_1} {{\bf X}}_1 ({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' \boldsymbol f^0(x)_{|z=0} \right]\\
&\qquad \qquad \quad \;\;\; - {{\mbox{E}}}\left((1-z) f^1(x)\right) +\frac{1}{n}{{\mbox{E}}}\left[ {\bf1}_{n_0} {{\bf X}}_0 ({{\bf X}}_1'{{\bf X}}_1)^{-1}{{\bf X}}_1' \boldsymbol f^1(x)_{|z=1} \right],
\end{aligned}$$ where $\boldsymbol f^j(x)_{|z=j}$ is a vector of length $n_j$ containing the elements $\{f^j(x_i) ; i \in \mathcal{I}_j\} $, $j=0,1$. \[Prop.OR2\]
*Bias of DR estimators.*\
Under Assumptions \[Ass1b\] and \[Ass3\] and Regularity Assumption \[RegAss1\] and \[RegAss2\] in Appendix A, $$\begin{aligned}
{\lim \limits_{n\to\infty}}\emph{bias}_T (\hat \tau^1_{DR}) ={\lim \limits_{n\to\infty}}\emph{bias}_C (\hat \tau^1_{DR}) = \rho_0\sigma_0 \frac{{{\mbox{E}}}(\lambda_0(g(x; \gamma)))}{\Pr(z=1)}.
\end{aligned}$$ Under Assumptions \[Ass1b\], \[Ass2b\], \[Ass3\] and Regularity Assumption \[RegAss2\] in Appendix A, $$\begin{aligned}
\emph{bias}_T (\hat \tau_{DR})=\emph{bias}_C (\hat \tau_{DR})= \rho_1\sigma_1 {{\mbox{E}}}(\lambda_1(g(x; \gamma)))+ \rho_0\sigma_0 {{\mbox{E}}}(\lambda_0(g(x; \gamma))).
\end{aligned}$$ \[Prop.DR\]
-- ---- ----------------------------------- ----------------------------------------------------
OR ${\mbox{bias}}_T=0$ ${\mbox{bias}}_T={\mbox{bias}}_M$
DR ${\mbox{bias}}_T=0$ ${\mbox{bias}}_T=0$
OR ${\mbox{bias}}_T={\mbox{bias}}_C$ ${\mbox{bias}}_T={\mbox{bias}}_C +{\mbox{bias}}_M$
DR ${\mbox{bias}}_T={\mbox{bias}}_C$ ${\mbox{bias}}_T={\mbox{bias}}_C$
-- ---- ----------------------------------- ----------------------------------------------------
: The total bias of the outcome regression and double robust estimators decomposed into bias due to model misspecification (bias$_M$) and bias due to confounding (bias$_C$), see Proposition \[Prop.OR1\]-\[Prop.DR\] for details.
\[tab:bias.overview\]
Since the doubly robust estimator is the outcome regression estimator with a correction term, there is a link between Proposition \[Prop.OR2\] and \[Prop.DR\]: $$\begin{aligned}
{\lim \limits_{n\to\infty}}\mbox{bias}_C(\hat \tau^1_{DR}) - \mbox{bias}_C(\hat \tau^1_{OR}) = \rho_0 \sigma_0 {{\mbox{E}}}\left(\left.\lambda_0(g(x; \gamma)) - ({{\bf X}}_0' {{\bf X}}_0)^{-1} {{\bf X}}_0' {\boldsymbol \lambda_0 } x \right| z=1 \right),\qquad
\end{aligned}$$ and $$\begin{aligned}
\mbox{bias}_C(\hat \tau_{DR}) - \mbox{bias}_C(\hat \tau_{OR}) &= \rho_0 \sigma_0 {{\mbox{E}}}\left(\left.\lambda_0(g(x; \gamma)) - ({{\bf X}}_0' {{\bf X}}_0)^{-1} {{\bf X}}_0' {\boldsymbol \lambda_0 } x \right| z=1 \right)E(z) \;+\quad \\
& \rho_1 \sigma_1 {{\mbox{E}}}\left(\left.\lambda_1(g(x; \gamma)) - ({{\bf X}}_1' {{\bf X}}_1)^{-1} {{\bf X}}_1' {\boldsymbol \lambda_1 } x \right| z=0 \right)E(1-z).
\end{aligned}$$ Since $\lambda_j$ for $j=0,1$ is often close to linear in $x$ ([@P:00]) the expectation of the difference between $\lambda_j(g(x; \gamma))$ and the linear projection of $\lambda_j(g(x; \gamma))$ is small. Hence the difference between the confounding bias of doubly robust and outcome regression estimators is also typically small.
Uncertainty intervals
---------------------
From Proposition \[Prop.OR1\]-\[Prop.DR\] identification intervals and uncertainty intervals as defined in Section \[Identif.sec\] can be derived for $\tau^1$ and $\tau$. The estimation of all elements of ${\mbox{bias}}_C$ (Proposition \[Prop.OR1\]-\[Prop.DR\]) is straightforward with the exception of $\sigma_j$, $j=0,1$. A consistent estimator of $\sigma_j$ is given by: $$\hat \sigma_j=\sqrt{\frac{\hat \sigma_{j, OLS}^2}{1+ (-1)^j \frac{\rho_j^2}{n-p} (g({{\bf x}}_j; \gamma)^T {\boldsymbol \lambda}_j-{\boldsymbol \lambda}_j^T {{\bf X}}_j ({{\bf X}}_j^T {{\bf X}}_j)^{-1}{{\bf X}}_j^T {\boldsymbol \lambda}_j )}},
\label{sigma}$$ where ${\hat \sigma_{OLS,j}^2}$ is the residual sample variance from the OLS fit of the potential outcome $j$ (i.e. ${{\bf y}}^j$ against ${{\bf X}}_j$) and $g({{\bf x}}_j; \gamma)$ is a vector of length $n_j$ containing the elements $\{g(x_i; \gamma)) ; i \in \mathcal{I}_j\} $; a proof of this result for $j=1$ is found in @Tanja [page 9].
From Proposition \[Prop.OR1\]-\[Prop.DR\], identification intervals, assuming $\rho \in [\rho^L, \rho^U]$, can be derived by replacing $\sigma_j$ with $\hat \sigma_j$ from (\[sigma\]). For instance, from Proposition \[Prop.DR\], an estimated identification interval for $\tau^1$ is given by: $$\left\{ \tau^1 : \tau^1=\hat \tau^1_{DR} - \rho_0 \hat\sigma_{0}\frac{\hat{{\mbox{E}}}(\lambda_0(g(x; \gamma)))}{\widehat{\Pr}(z=1)} , \rho_0 \in [\rho_0^L,\rho_0^U]\right\}.
\label{t1DR.IS}$$
Ignoring the sampling variability from $\widehat{{\mbox{bias}}}_T$, and noting that $\hat\tau_{DR}^1$ is asymptotically normally distributed ([@Tsiatis:06]), the lower and upper bound of the uncertainty interval, $UI(\tau_{DR}^1; [\rho_0^L,\rho_0^U],\alpha),$ are respectively (see (\[UI.eqn\])): $$\min \limits_{\rho_0 \in [\rho_0^L,\rho_0^U]} \left( \hat \tau^1_{DR} - \rho_0 \hat\sigma_{0}\frac{\hat{{\mbox{E}}}(\lambda_0(g(x; \gamma)))}{\widehat{\Pr}(z=1)} -c_{\frac{\alpha}{2}}s.e.(\hat \tau^1_{DR}) \right),
\label{UI.ex.l}$$ and $$\max \limits_{\rho_0 \in [\rho_0^L,\rho_0^U]} \left( \hat \tau^1_{DR} - \rho_0 \hat\sigma_{0} \frac{\hat{{\mbox{E}}}(\lambda_0(g(x; \gamma)))}{\widehat{\Pr}(z=1)} +c_{\frac{\alpha}{2}}s.e.(\hat \tau^1_{DR}) \right),
\label{UI.ex.u}$$ where $c_{\frac{\alpha}{2}}$ is the $(1-\alpha/2)100 \%$ percentile of the standard normal distribution. Estimated uncertainty intervals for $\tau$ and the outcome regression estimators are obtained similarly. Standard errors for the outcome regression and doubly robust estimators are given in Appendix C. Note that ignoring the sampling variability from $\widehat{{\mbox{bias}}}_T$ in (\[UI.ex.l\]) and (\[UI.ex.u\]) should have no serious consequences because it is of lower asymptotic order. This is confirmed by the simulation study in Section \[Sim.sec\].
Simulation study {#Sim.sec}
================
The purpose of the simulation experiments is to illustrate the relative sizes of the biases due to model misspecification and confounding, as well as to study the empirical coverages of the proposed uncertainty intervals. The data is generated using four different designs, linear or non-linear, using one or five covariates. For each design, we use two different treatment assignments with different amount of imbalance in the propensity scores between the treated and the non-treated. One with low imbalance (high overlap), L1$=0.18$ and $0.19$, and one with high imbalance (low overlap), L1$=0.37$ and $0.34$. L1 is the area which is not overlaid in a graph with two density histograms of the propensity scores for the treated and untreated, see @L1 [equation (5)]. This measure varies with different bin-size. We use the default of the function [hist]{} in [R]{} statistical software on all the propensity scores (both from treated and untreated) to select bin-size. In all designs we have about $40\%$ treated, and use a linear model for the treatment assignment mechanism, i.e. $g(x; \gamma)= \gamma' x$ in (\[assignment.mod\]). For all four designs we use a sample size of 250 and 500, with 10 000 replications and compute uncertainty intervals based on Proposition \[Prop.OR1\] and \[Prop.DR\], i.e. using the outcome regression estimators adjusting for ${\mbox{bias}}_C$ but not ${\mbox{bias}}_M$ and using the doubly robust estimators adjusting for ${\mbox{bias}}_T={\mbox{bias}}_C$. Finally, we let $$\begin{pmatrix} \eta\\ \varepsilon^0 \\ \varepsilon^1 \end{pmatrix} \sim MVN\left(\begin{pmatrix}0\\0\\0 \end{pmatrix},\begin{pmatrix}
1 & \rho_0& \rho_1 \\
\rho_0& 1 & \rho_0 \rho_1 \\
\rho_1 & \rho_0 \rho_1 & 1 \\
\end{pmatrix}\right),$$ where $\rho_0=\rho_1 =$ 0, 0.05, 0.1, 0.3, and 0.5.
In Design A and B we use $\gamma' = [-0.27, 0.3]$ and $\gamma' = [-0.3, 0.65] $, to generate low and high L1 respectively, and $x \sim {{\mbox{N}}}(0,1)$. In Design A (linear) we use the outcome equations: $f^0(x)= 0.5 + 0.5x $, and $f^1(x)= 2.5 + 1.5x $. In Design B (non-linear) we use $f^0(x)=h^0(x)$ and $f^1(x)=h^1(x)$, where:
[@llll@]{} {[4]{}[3mm]{} &$0.15-x-0.4x^2$ & $x<-1.5$,&\
&$1.5-x+0.5x^2+x^3 $&$-1.5\leq x< 1$,\
&$1.75-0.25x+0.5x^2 $ &$1\leq x< 2$ ,\
&$2.25+0.5x$&$2<x$,\
\
{[4]{}[3mm]{} &$0.2x-0.1x^2$ & $x<-1$,\
&$0.3x $&$-1\leq x< 1$,\
&$0.4-0.1x^2 $ &$1\leq x< 3$,\
&$-0.2-0.1x$&$3<x$.
These choices were made to make polynomial approximation difficult.
In Design C and D we use $\gamma' = (-0.27, 0.2, -0.15, 0.05, 0.15, - 0.1)$ and $\gamma' = (-0.3,0.5, -0.25,\linebreak[0] 0.15, 0.25, - 0.15)$, to generate low and high L1 respectively. The covariates are simulated such that $x_1 \sim {{\mbox{N}}}(0,1)$, $x_2$ and $x_4$ are Bernoulli distributed with probability $0.5+0.05x_1$ and $0.4+ 0.2x_3 $, $x_3 = 0.015x_1 + u_3,$ where $u_3$ is uniformly distributed in $(-0.5, 0.5)$, and $x_5 =0.04x_1+0.15x_2+0.05x_3+u_5,$ where $u_5 \sim {{\mbox{N}}}(0,1)$. In Design C (linear) we use the outcome equations: $f^0(x)= -0.5 + 0.5x_1 +1.0x_2 + 0.5x_3 -1.0x_4 +1.0x_5 $, and $f^1(x)= 1.5 -1.5x_1 + 4.0 x_2-1.5x_3 + 3.0x_5 $. In Design D (nonlinear) we use, $f^0(x)= h^1(x_1) + 0.1 x_2 -0.3 x_3 -0.6 h^1(x_1) \cdot x_4 -0.1x_5 $, $f^1(x)= h^0(x_1) +h^0(x_1) \cdot x_2 + 0.3 x_2 -0.2 x_3 -0.4x_4 +0.6x_5$.
For all designs we fit a correctly specified propensity score model, $g(x;\gamma)$, and $f^0$ and $f^1$ with linear models in $x$, i.e. Assumption 4 is fulfilled in Design A and C but not in Design B and D. We compare width and coverage of 95% confidence intervals for $\tau$ with $UI(\tau; [0,0.2], [0,0.2],0.05)$ and $UI(\tau; [0,0.4], [0,0.4],0.05)$ (the corresponding confidence intervals and uncertainty intervals are used for $\tau^1$).
Results
-------
Figure \[Bias.fig\] displays magnitude of the empirical bias of the outcome regression estimator (both model misspecification and confouning bias), defined in Section \[Bias.section\], for the two non-linear designs with correctly specified propensity score models. We can see that bias$_M(\hat \tau_{OR})$ is larger when the propensity scores of the treated and untreated are more separeted (L1 low). However, even when L1 is high, bias$_M(\hat \tau_{OR})$ and bias$_C(\hat \tau_{OR})$ are of approximately the same magnitude when $\rho_0=\rho_1 = 0.05$, and for $\rho_0=\rho_1 \geq 0.1$ ${\mbox{bias}}_C(\hat \tau_{OR})$ dominates ${\mbox{bias}}_M(\hat \tau_{OR})$. Note also that in Design D ${\mbox{bias}}_C(\hat \tau_{OR})$ and ${\mbox{bias}}_M(\hat \tau_{OR})$ have different signs implying that the total bias is smaller than the confounding bias. Hence, the outcome regression estimator has a smaller total bias than the doubly robust estimator in such a case, since the confounding bias of the two estimators is almost the same.
\[htbp\] ![The magnitude of ${\mbox{bias}}_T(\hat\tau_{OR})$, ${\mbox{bias}}_C(\hat\tau_{OR})$ and ${\mbox{bias}}_T(\hat\tau_{OR})$ with varying $\rho_0=\rho_1=\rho$ for the two non-linear designs, with low and high imbalance in the propensity scores. []{data-label="Bias.fig"}](bias2x2.pdf "fig:"){width="75.00000%"}
The uncertainty intervals are wider than the confidence interval per definition, which is confirmed in Figure \[t.250.fig\] - \[t.500.fig\] and Appendix D. In particular, the uncertainty intervals derived under the assumption that $\rho_0$ and/or $\rho_1 \in [0,0.4]$ are around twice as wide as the corresponding confidence intervals. The empirical coverage of the 95$\%$ uncertainty intervals are, as expected, generally high if the assumption on $\rho_0$ and/or $\rho_1$ is met ($\rho_0$ and/or $\rho_1$ is covered by the pre-specified interval from which the uncertainty interval is derived); see Figure \[t.250.fig\] - \[t.500.fig\] and Appendix D. However, if the assumption is not met the empirical coverage is less than $95 \%$. When using the outcome regression estimator in Design B, we do not necessarily expect 95$\%$ coverage of the UI:s, even if the assumption on $\rho$ is met, because the outcome regression model is misspecified, and ${\mbox{bias}}_M$ has the same sign as ${\mbox{bias}}_C$. However, the empirical coverage is at least 95$\%$ due to three reasons: first, the uncertainty intervals have higher coverage than 95$\%$ if the models are correctly specified; second, $\sigma_0$ is overestimated due to model misspecification; and third, bias$_M$ has the same size or smaller than bias$_C$. Note finally that the empirical coverage of the 95$\%$ confidence intervals assuming no unobserved confounding is too low even for small $\rho$, see Figure \[t.250.fig\] - \[t.500.fig\] and Appendix D.
\[htbp\]
![Boxplot of the width of two 95% uncertainty intervals (assuming $\rho_j \in [0,0.2]$, green, and $\rho_j \in [0,0.4]$, blue, $j=0,1$) and the 95% confidence interval, red, for the doubly robust (DR) and outcome regression (OR) estimator of $\tau$ under design A-D for $\rho_0=\rho_1=0.05$ and $0.3$, with sample size 250. The empirical coverage of each interval is written below each boxplot and the number of outliers that lie outside the window is written at the top of the window above each boxplot.[]{data-label="t.250.fig"}](t_p1_250.pdf){width="\textwidth"}
![Boxplot of the width of two 95% uncertainty intervals (assuming $\rho_j \in [0,0.2]$, green, and $\rho_j \in [0,0.4]$, blue, $j=0,1$) and the 95% confidence interval, red, for the doubly robust (DR) and outcome regression (OR) estimator of $\tau$ under design A-D for $\rho_0=\rho_1=0.05$ and $0.3$, with sample size 250. The empirical coverage of each interval is written below each boxplot and the number of outliers that lie outside the window is written at the top of the window above each boxplot.[]{data-label="t.250.fig"}](t_p2_250.pdf){width="\textwidth"}
\[htbp\]
![Boxplot of the width of two 95% uncertainty intervals (assuming $\rho_j \in [0,0.2]$, green, and $\rho_j \in [0,0.4]$, blue, $j=0,1$) and the 95% confidence interval, red, for the doubly robust (DR) and outcome regression (OR) estimator of $\tau$ under design A-D for $\rho_0=\rho_1=0.05$ and $0.3$, with sample size 500. The empirical coverage of each interval is written below each boxplot and the number of outliers that lie outside the window is written at the top of the window above each boxplot.[]{data-label="t.500.fig"}](t_p1_500.pdf){width="\textwidth"}
![Boxplot of the width of two 95% uncertainty intervals (assuming $\rho_j \in [0,0.2]$, green, and $\rho_j \in [0,0.4]$, blue, $j=0,1$) and the 95% confidence interval, red, for the doubly robust (DR) and outcome regression (OR) estimator of $\tau$ under design A-D for $\rho_0=\rho_1=0.05$ and $0.3$, with sample size 500. The empirical coverage of each interval is written below each boxplot and the number of outliers that lie outside the window is written at the top of the window above each boxplot.[]{data-label="t.500.fig"}](t_p2_500.pdf){width="\textwidth"}
Effect of regular food intake on health
=======================================
SHARE is a longitudinal survey on health, socio-economic status and social networks of individuals aged 50 years or older from several European countries ([@SHARE]). The sampling in SHARE is on a household level where all residents in the household (almost exclusively one individual or one man and one women) are interviewed. We focus this study on women in the 13 countries that participate in both wave 4 and 5 of SHARE, which were collected in 2011 (baseline) and 2013 (follow-up). The observed sample consists of 12 842 individuals. We are interested in investigating the causal effect of regular food intake on health. We define regular food intake as eating at least 3 full meals a day at baseline. A full meal is defined as eating more than 2 items or dishes when you sit down to eat. For example, eating potatoes, vegetables, and meat; or eating an egg, bread, and fruit are both considered full meals. The health outcome used is change in maximum grip strength (in kg, maximum of 4 measures using a dynamometer) from 2011 to 2013. Grip strength is associated with both health-related quality of life, disability and mortality, see e.g. [@GS:Health] and [@GS:Death].
In order to estimate the causal effect of interest we control for covariates measured at baseline. These covariates include health, cognition, lifestyle, and socioeconomic variables as well as other background characteristics. The health variables include self reported health (excellent; very good; good; fair; or poor), number of problems with mobility (such as walking; lifting small objects; lifting heavy objects; etc., maximum 10), number of chronic diseases (such as diabetes; cancer; asthma; etc., maximum 15), depression (number of symptoms of depression, maximum 12, using the EURO-D scale), body mass index ($\mbox{kg/m}^2$) and limitations in daily life due to health problems (yes; no). We measure cognition with the number of animals the subject was able to state during 1 minute. The lifestyle variables consist of high alcohol use (drinking at least one glass of alcohol for women and two glasses for men at least 5 days a week), smoking (smoker; stopped smoking; non-smoker), physical inactivity (if respondents engaged in moderate to vigorous physical activity at most 1 to 3 times a month) and having a social network (have someone to discuss important things with, talk at least several times a week). The socioeconomic variables include education level (level 0-1; 2; 3; 4; or 5-6, using ISCED-97 scale) and whether or not the subject is living in an apartment or freestanding building. Finally, demographic characteristics consist of age, sex and country of residence (Austria; Germany; Sweden; Netherlands; Spain; Italy; France; Denmark; Switzerland; Belgium; Czech republic; Slovenia; or Estonia).
We estimated the causal effects by controlling for all main effects in the two potential outcome models. We used two different treatment assignment models, one with all main effects and one more flexible. The flexible treatment assignment model was fitted with a LASSO to select terms from all main effects together with interactions and quadratic terms. More specifically we use the R package [glmnet]{} and choose the largest value of the tuning parameter such that the mean cross validated error is within one standard error of the minimum, see [@glmnet] for details. The models including the selected terms are then refitted using maximum likelihood. The balance in the propensity scores is fairly similar between the main terms and LASSO based treatment assignment model, see Figure \[prop.sc.hist\].
![Overlaid histograms showing the amount of imbalance in the propensity scores between treated and untreated for the two different sets of covariates, main effects (left) and LASSO (right). This bin size is default from the function [hist]{} in [R]{} statistical software and this bin size was also used to derive L1 (0.23 for main effects and 0.25 for LASSO).](hist_prop_scores.pdf){width="70.00000%"}
\[prop.sc.hist\]
[clllll]{} && coef & CI&UI, $|\rho|\leq0.02$\
&$\tau_{OR}^1$ & 0.28 & (0.07, 0.48)&(-0.11, 0.66)\
&$\tau_{DR}^1$ & 0.26 & (0.05, 0.46)&(-0.12, 0.64)\
&$\tau_{OR}$ & 0.27 & (0.08, 0.46) &(-0.09, 0.63)\
&$\tau_{DR}$ & 0.26 & (0.06, 0.45)&(-0.11, 0.62)\
&$\tau_{OR}^1$ & 0.28 & (0.07, 0.48)&(-0.10, 0.65)\
&$\tau_{DR}^1$ & 0.24 & (0.03, 0.45)&(-0.15, 0.63)\
&$\tau_{OR}$ & 0.27 & (0.08, 0.46) &(-0.09, 0.63)\
&$\tau_{DR}$ & 0.25 & (0.05, 0.44)&(-0.12, 0.62)\
\[ex.tab\]
In Table \[ex.tab\] we can see that the estimates of $\tau$ and $\tau_1$ assuming ignorability of treatment assignment, are significant (95% CI do not cover zero) for all estimators and estimated to between $0.24$ and $0.28$, which can be compared to $0.85$, the average decrease in maximum grip strength of the total study sample. All estimates obtained are fairly similar, in particular when compared to the extra variation introduced by the uncertainty in unobserved confounding (compare UIs with CIs). Indeed, the uncertainty intervals assuming $\rho \in [-0.02, 0.02]$ contain 0. The bounds $\max|\rho|=0.02$ corresponds to unobserved confounding explaining, e.g., 2 % of the unexplained variation in the outcome models and the treatment assignment models (see interpration of $\rho$ given in Section \[Identif.sec\] above). We have no reason to believe that such unobserved confounding is unreasonable. Thus, here, taking into account uncertainty in unobserved confounding yields inconclusive results, i.e. the data does not give us evidence for a positive effect in contrast with the naive conclusion that would typically be taken by only considering sampling uncertainty through classical confidence intervals.
Note, finally, that this analysis has been performed assuming dropout at follow up to be ignorable. Non-ignorable dropout if suspected could be dealt with similarly by introducing a new bias parameter [@Genback:2015], thereby further increasing even more the uncertainty around the estimates obtained.
Discussion
==========
Causal inference from observational data is often based on the assumption of no unobserved confounding variables. This identifying assumption is typically not empirically testable without further assumptions and/or information such as, e.g., the known existence of instrumental variables [@LunaJ:2014]. This paper proposes an inferential approach for outcome regression and doubly robust estimators that takes into account uncertainty on the possible existence of unobserved confounding. The method proposed is computationally fast and easy to apply (the `R`-package `ui` is available at `http://stat4reg.se/software`).
Outcome regression and double robust estimators make model assumptions which, if mistaken, also imply bias. On the other hand, model misspecification can in principle be empirically investigated. In the simulated settings, even though the model misspecification was quite severe for the outcome regression estimator, bias due to unobserved confounding dominated model misspecification bias when $\rho \geq 0.1$ and even more so when propensity scores were not too close to zero or one. More generally, while model misspecification can under some assumptions be made arbitrarily small asymptotically (by increasing model complexity), bias/uncertainty due to unobserved confounding remains unchanged and therefore more relevant when increasing sample size.
We have focused on misspecification of the outcome models instead of the treatment assignment model. The latter is not only more challenging theoretically, but more importantly, one can argue that model building is less difficult for the treatment assignment model than for outcome models since for the former all data is available and no extrapolation is performed, while outcome models are fitted only on one sub-sample at a time (e.g., the controls) and are used to extrapolate on the other sub-sample (e.g., the treated). Extrapolations are thus done on part of the sample space which is sparsly populated, hence, where the model specification is difficult to check. Yet, it has been shown that, for double robust estimators, mild misspecification of both models (for treatment assignment and outcome) may lead to large bias in specific situations, in which case regression outcome estimation may be preferable [@kang:07], or improved versions of the classic double robust estimator used here; see [@RotnitzkyStijn:15] for a review.
The proposed uncertainty intervals can be used to perform a sensitivity analysis. For example, for all the estimators presented in Table \[ex.tab\], the UIs would approximately be bounded below by zero if constructed using $\rho\in [-0.01,0.01]$. Thus, the 5% significance conclusion is here sensitive to unobserved confounding of magnitude $\max|\rho|\geq 0.01$. However, our experience is that sensitivity analyses are difficult to communicate to the layman for whom statistical hypothesis testing may already be a difficult concept. We therefore advocate here the more intuitive interval estimation approach, i.e. providing an UI for the effect of interest given some a priori assumption on unobserved confounding and a desired coverage level.
Acknowledgements {#acknowledgements .unnumbered}
================
We are grateful to Elena Stanghellini, Arvid Sjölander, Anders Lundquist and Anita Lindmark for helpful comments. This work was supported by the Swedish Research Council for Health, Working Life and Welfare and the Marianne and Markus Wallenberg Foundation.
Appendix A {#appendix-a .unnumbered}
==========
\[Lem1\] Under Assumption \[Ass1b\], \[Ass2b\], \[Ass3\] and \[Ass4\] the bias of the ordinary least squares estimate of $\beta^j$, given $\rho_j$ is: $$\label{distortion.eq}
E(\hat\beta^j_{OLS})=\beta^j+ (-1)^{1+j} \rho_j \sigma_{j} E(({{\bf X}}_j' {{\bf X}}_j)^{-1} {{\bf X}}_j' {\boldsymbol \lambda_j } ) ,$$ for $j=0,1$; where ${\boldsymbol \lambda}_j = [\lambda_j(g(x_1; \gamma)), \cdots , \lambda_j(g(x_{n_j}; \gamma))]'$, $\lambda_0(g(x_i; \gamma))= \frac{\phi(g(x_i; \gamma))}{1-\Phi(g(x_i; \gamma))}$, $\lambda_1(g(x_i; \gamma))=\frac{\phi(g(x_i; \gamma))}{\Phi(g(x_i; \gamma))}$. The proof follow from $E(y^j|x,z=j)=E(y^j|x)+ (-1)^{1+j}\rho_j\sigma_j \lambda_j(g(x; \gamma))$ and can be found for $z=1$ in [@Genback:2015], the proof when $z=0$ is similar.
[RegAss]{} \[RegAss1\]
\[RegAss1a\] There exist a constant $c$ such that $E\left(\frac{1}{( \frac{1}{n}\sum_i z_i)^2}\right)<c<\infty, \forall n$ and $E(z)>0.$
\[RegAss1b\] $\left|E(z_i y_i)\right|<\infty$ and $\left|E(\hat\beta^{0'}_{OLS} z_i x_i)\right|<\infty$.
\[Lem2\] Under Regularity Assumption \[RegAss1a\]: $${\lim \limits_{n\to\infty}}E\left(\frac{1}{n_1}\sum_{i=1}^{n}z_i f(x_i,y_i)\right) = \frac{1}{E(z_i)}E\left(z_i f(x_i,y_i)\right),$$ for any function $f$ of $x_i$ and $y_i$ such that $\left| E\left(z_i f(x_i,y_i)\right)\right|<\infty$.
$${\lim \limits_{n\to\infty}}E\left(\frac{1}{n_1}\sum \limits_{i =1}^nz_i f(x_i,y_i)\right)={\lim \limits_{n\to\infty}}E\left(\frac{1}{\frac{1}{n}\sum_i z_i}\right) E\left(\frac{1}{n}\sum \limits_{i =1}^nz_i f(x_i,y_i)\right)=\frac{1}{E(z_i)}E\left(z_i f(x_i,y_i)\right).$$ The first equality follow from: $${{\mbox{Cov}}}\left(\frac{1}{\sum_j z_j}, \sum z_i f(x_i,y_i)\right)= \begin{cases}\frac{1}{n} E\left(z_i f(x_i,y_i)\right) E\left(\frac{-1}{\frac{1}{n^2}(\sum_i z_i))(\sum_i z_i+1)}\right) \mbox{if }z_i=0\\
0\mbox{ if } z_i=1.\end{cases}$$
if there exist a constant $c$ such such that $E\left(\frac{1}{( \frac{1}{n}\sum_i z_i)^2}\right)<c<\infty$ then: $${\lim \limits_{n\to\infty}}\left| {{\mbox{Cov}}}\left(\frac{1}{\sum_j z_j}, \sum z_i f(x_i,y_i)\right) \right| \leq {\lim \limits_{n\to\infty}}\left| E\left(z_i f(x_i,y_i)\right)\right| E\left(\frac{1}{(\frac{1}{n}\sum_iz_i)^2} \right)\frac{1}{n}=0.$$ The second equality follow from: $\frac{1}{n}\sum_i z_i\xrightarrow{p}E(z)$ (weak law of large numbers) and $\frac{1}{\frac{1}{n}\sum_i z_i}\xrightarrow{p}\frac{1}{E(z)}$ if $E(z)>0$ (continuous mapping theorem). By dominated convergence theorem ${\lim \limits_{n\to\infty}}E\left(\frac{1}{\frac{1}{n}\sum_i z_i}\right)=\frac{1}{E(z)}$ (since $E\left(\frac{1}{| \frac{1}{n}\sum_i z_i|}\right)<\sqrt{c}<\infty$), the theorems used can be found for instance in [@Rosenthal].
To calculate the bias of the doubly robust estimator we need a regularity assumption to be able to exchange $\hat p(x)$ with $p(x)$ using the uniform integrability convergence theorem ([@Rosenthal]).
\[RegAss2\] $\sum \limits_{i =1}^n z_i \frac{f(x_i,y_i)-\hat\beta^{1'}_{OLS}x_i}{\hat{p}(x_i)}$, and $\sum \limits_{i =1}^n (1-z_i)\frac{f(x_i,y_i)-\hat\beta^{0'}_{OLS}x_i}{1-\hat{p}(x_i)}$ are uniformly integrable $\forall x \in \mathcal{X}$, the support of $x.$
Appendix B {#appendix-b .unnumbered}
==========
Proof of Proposition \[Prop.OR1\] {#proof-of-proposition-prop.or1 .unnumbered}
---------------------------------
Under, Assumption \[Ass1b\], \[Ass3\], \[Ass4a\] and Regularity Assumption \[RegAss1\]: $$\begin{aligned}
{\lim \limits_{n\to\infty}}\mbox{bias}_T(\hat \tau_{OR}^1)&={\lim \limits_{n\to\infty}}E(\hat\tau^1_{OR})- \tau^1={\lim \limits_{n\to\infty}}E\left(\frac{1}{n_1}\sum \limits_{i =1}^nz_i y_i-\hat\beta^{0'}_{OLS}\frac{1}{n_1}\sum_{i =1}^{n} z_i x_i \right) - \tau^1\\
&= E(y^1\mid z=1) - E\left(\hat\beta^{0'}_{OLS} z_i x_i \right) \frac{1}{E(z)} - \tau^1\\
&= E(y^0\mid z=1)-E(\hat\beta^{0'}_{OLS})E(x\mid z=1) \\
&=E(y^0\mid z=1)-\beta^{0'} E(x\mid z=1) + \rho_0 \sigma_0 E(({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' \boldsymbol \lambda_0) E(x\mid z=1) \\
&=E(y^0\mid z=1)-(E(y^0\mid z=1)-\rho_0\sigma_0 E(\lambda_1(g(x; \gamma)) \mid z=1)) \\
&\qquad\qquad\qquad\qquad+ \rho_0 \sigma_0 E(({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' \boldsymbol \lambda_0) E(x\mid z=1) \\
&= \rho_0\sigma_0 \left( E(\lambda_1(g(x; \gamma))\mid z=1)+ E(({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' \boldsymbol \lambda_0) E(x\mid z=1)\right). \end{aligned}$$ The equality between line 1 and 2 follow from Lemma 2. The equality between line 3 and 4 follow from Lemma 1 and the equality between line 4 and 5 follow from the fact that $E(y^0\mid z=1)=E(\beta^{0'}x_i +\rho_0\sigma_0\lambda_1(g(x_i; \gamma))\mid z_i=1)=\beta^{0'}E(x_i\mid z_i=1) +\rho_0\sigma_0E(\lambda_1(g(x_i; \gamma))\mid z_i=1)$.
Since Assumption \[Ass4\] is fulfilled (the regression model is correctly specified) ${\mbox{bias}}_M(\hat \tau_{OR}^1)=0$, and therefore ${\mbox{bias}}_T(\hat \tau_{OR}^1)={\mbox{bias}}_C(\hat \tau_{OR}^1)$.
Under Assumption \[Ass1b\], \[Ass2b\], \[Ass3\] and \[Ass4\]: $$\begin{aligned}
\mbox{bias}_T(\hat \tau_{OR})&=E\left[\frac{1}{n}\sum \limits_{i =1}^n\left( \hat\beta^{1'}_{OLS}x_i-\hat\beta^{0'}_{OLS}x_i\right)\right] -\tau=E\left[\frac{1}{n}\left({\bf1}_n {{\bf X}}\hat\beta^{1'}_{OLS}- {\bf1}_n {{\bf X}}\hat\beta^{0'}_{OLS}\right)\right] - \tau\\
&=\frac{1}{n}E\left[ {\bf1}_n {{\bf X}}\left( \beta^1 - \beta^0 \right) + {\bf1}_n {{\bf X}}\left(\rho_1\sigma_1({{\bf X}}_1'{{\bf X}}_1)^{-1}{{\bf X}}_1' \boldsymbol\lambda_1 +\rho_0\sigma_0({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' \boldsymbol \lambda_0 \right)\right]-\tau\\
&= \frac{1}{n}E\left[ {\bf1}_n {{\bf X}}\left(\rho_1\sigma_1({{\bf X}}_1'{{\bf X}}_1)^{-1}{{\bf X}}_1' \boldsymbol\lambda_1 +\rho_0\sigma_0({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' \boldsymbol \lambda_0 \right)\right].\end{aligned}$$ Where the equality between line 1 and 2 follow from Lemma 1. Since Assumption \[Ass4\] is fulfilled (the regression model is correctly specified) ${\mbox{bias}}_M(\hat \tau_{OR})=0$, and therefore ${\mbox{bias}}_T(\hat \tau_{OR})={\mbox{bias}}_C(\hat \tau_{OR})$.
Proof Proposition \[Prop.OR2\] {#proof-proposition-prop.or2 .unnumbered}
------------------------------
Under, Assumption \[Ass1b\], \[Ass3\] and Regularity Assumption \[RegAss1\]: $$\begin{aligned}
{\lim \limits_{n\to\infty}}{\mbox{bias}}_T(\hat \tau^1_{OR}) &=E(y^0|z=1) -E(\hat\beta^{0'}_{OLS}) E(x |z=1)\\
&=E[ E(y^0|x,z=1) |z=1] \\
&\qquad\qquad- E\left[({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' f^0(x) - \rho_0 \sigma_0({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' {\boldsymbol \lambda_0 } \right] E(x|z=1) \\
&=E\left[ f^0(x) + \rho_0\sigma_0 \lambda_1(g(x; \gamma)) |z=1\right] \\
&\qquad\qquad- E\left[({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' f^0(x) - \rho_0 \sigma_0({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' {\boldsymbol \lambda_0 } \right] E(x|z=1) \\
&=E\left[ f^0(x) |z=1\right] - E\left[({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' f^0(x) \right] E(x|z=1) +\mbox{bias}_C(\tau_{OR}^1)\\
&=\mbox{bias}_M(\tau_{OR}^1) +\mbox{bias}_C(\tau_{OR}^1),\end{aligned}$$ where the first equality can be seen in proof of Proposition \[Prop.OR1\].
Under, Assumption \[Ass1b\], \[Ass2b\] and \[Ass3\]: $$\begin{aligned}
{\mbox{bias}}_T(\hat \tau_{OR})&=E\left[\frac{1}{n}\sum \limits_{i =1}^n\left( \hat\beta^{1'}_{OLS}x_i-\hat\beta^{0'}_{OLS}x_i\right)\right] - \tau \\
&=E\left[\frac{1}{n}\left({\bf1}_n {{\bf X}}\hat\beta^{1'}_{OLS}- {\bf1}_n {{\bf X}}\hat\beta^{0'}_{OLS}\right)\right] - \tau\\
&= E\left[\frac{1}{n}E\left( \left.{\bf1}_n {{\bf X}}({{\bf X}}_1'{{\bf X}}_1)^{-1}{{\bf X}}_1' \boldsymbol {{\bf y}}_1 - {\bf1}_n {{\bf X}}({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' \boldsymbol {{\bf y}}_0\right| {{\bf X}}\right)\right] -\tau\\
&=\frac{1}{n}E\left[ {\bf1}_n {{\bf X}}\left(({{\bf X}}_1'{{\bf X}}_1)^{-1}{{\bf X}}_1'E\left( \left. \boldsymbol {{\bf y}}_1 \right| {{\bf X}}\right) -({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0'E\left( \left. \boldsymbol {{\bf y}}_0\right| {{\bf X}}\right) \right)\right]- \tau\\
&=\frac{1}{n}E\left( {\bf1}_n {{\bf X}}\left[({{\bf X}}_1'{{\bf X}}_1)^{-1}{{\bf X}}_1' \left( \boldsymbol f^1(x)_{|z=1} +\rho_1 \sigma_1 {\boldsymbol \lambda_1}\right)\right]\right)\\
&\qquad\qquad\qquad\qquad-\frac{1}{n}E\left( {\bf1}_n {{\bf X}}\left[({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0'\left( \boldsymbol f^0(x)_{|z=0} +\rho_0 \sigma_0 {\boldsymbol \lambda_0}\right)\right]\right)- \tau\\
&=\mbox{bias}_C(\tau_{OR})+\frac{1}{n}E\left[ {\bf1}_n {{\bf X}}\left(({{\bf X}}_1'{{\bf X}}_1)^{-1}{{\bf X}}_1' \boldsymbol f^1(x)_{|z=1} \right) \right]\\
&\qquad \qquad \qquad \qquad -\frac{1}{n}E\left[ {\bf1}_n {{\bf X}}\left(({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' \boldsymbol f^0(x)_{|z=0} \right) \right]- \tau\\
&=\mbox{bias}_C(\tau_{OR})+\frac{1}{n}E\left[ {\bf1}_{n_0} {{\bf X}}_0 ({{\bf X}}_1'{{\bf X}}_1)^{-1}{{\bf X}}_1' \boldsymbol f^1(x)_{|z=1} \right]- E\left((1-z) f^1(x)\right)\\
&\qquad \qquad \qquad \qquad + E\left(z f^0(x) \right) - \frac{1}{n}E\left[ {\bf1}_{n_1} {{\bf X}}_1 ({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' \boldsymbol f^0(x)_{|z=0} \right],\end{aligned}$$ where the last equality follow from: $\frac{1}{n}E\left[ {\bf1}_{n_0} {{\bf X}}_0 ({{\bf X}}_0'{{\bf X}}_0)^{-1}{{\bf X}}_0' \boldsymbol f^0(x)_{|z=0} \right]=E((1-z)f^0(x))$ and $\frac{1}{n}E\left[ {\bf1}_{n_1} {{\bf X}}_1 ({{\bf X}}_1'{{\bf X}}_1)^{-1}{{\bf X}}_1' \boldsymbol f^1(x)_{|z=1} \right]=E(zf^1(x))$.
Proof Proposition \[Prop.DR\] {#proof-proposition-prop.dr .unnumbered}
-----------------------------
Under, Assumption \[Ass1b\], \[Ass2b\], \[Ass3\], and Regularity Assumption \[RegAss1\] and \[RegAss2\]: $$\begin{aligned}
{\lim \limits_{n\to\infty}}\mbox{bias}_C(\hat \tau^1_{DR})&={\lim \limits_{n\to\infty}}E(\hat\tau^1_{DR}) - \tau^1\\
&={\lim \limits_{n\to\infty}}\left[\frac{1}{n_1}\sum \limits_{i =1}^n z_i\left(y_i-\hat\beta^{0'}_{OLS}x_i\right)-\frac{1}{n_1}\sum \limits_{i =1}^n(1-z_i)\frac{y_i-\hat\beta^{0'}_{OLS}x_i}{1-\hat p(x_i)} \right]- \tau^1\\
&={\lim \limits_{n\to\infty}}\left[\frac{1}{n_1}\sum \limits_{i =1}^n z_i\left(y_i-\hat\beta^{0'}_{OLS}x_i\right)-\frac{1}{n_1}\sum \limits_{i =1}^n(1-z_i)\frac{y_i-\hat\beta^{0'}_{OLS}x_i}{1-p(x_i)} \right]- \tau^1\\
&=E(y^0 -\hat\beta^{0'}_{OLS} x \mid z=1)\\
&\qquad \qquad \qquad \qquad - \frac{1}{E(z)}E_x\left[\left.E\left((1-z)(y^0-\hat\beta^{0'}_{OLS} x) \right| x \right)\frac{1}{1-p(x)} \right] \\
&=E(y^0 -\hat\beta^{0'}_{OLS} x \mid z=1) \\
&\qquad \qquad \qquad \qquad - \frac{1}{E(z)}E_x\left[\left.E\left((y^0-\hat\beta^{0'}_{OLS} x) \right| x, z=0 \right)\frac{p(z=0|x)}{1-p(x)} \right] \\
&=E(y^0 -\hat\beta^{0'}_{OLS} x \mid z=1) - \frac{1}{E(z)}E_x\left[\left.E\left(y^0\right| x, z=0 \right) -\hat\beta^{0'}_{OLS} x \right] \\
&=E(y^0 -\hat\beta^{0'}_{OLS} x \mid z=1) - \frac{1}{E(z)}E_x\left[E(y^0 | x) - \rho_0\sigma_0\lambda_0(g(x; \gamma))- \hat\beta^{0'}_{OLS} x \right] \\
&=E(y^0 -\hat\beta^{0'}_{OLS} x \mid z=1) - \frac{E(y^0 ) }{E(z)} +\frac{\rho_0\sigma_0\lambda_0(g(x; \gamma))}{E(z)}+\frac{E\left(\hat\beta^{0'}_{OLS} x\right) }{E(z)} \\
&=E(y^0 \mid z=1) -\frac{E(z\hat\beta^{0'}_{OLS} x)}{E(z)}- \frac{E(z y^0 ) +E((1-z) y^0 ) }{E(z)} \\
&\qquad \qquad \qquad \qquad +\frac{\rho_0\sigma_0\lambda_0(g(x; \gamma))}{E(z)}+\frac{E\left( \hat\beta^{0'}_{OLS}x \right) }{E(z)} \\
&=\frac{\rho_0\sigma_0E(\lambda_0(g(x; \gamma)))}{E(z)},\end{aligned}$$ using Lemma 2 and the fact that $\left.E\left(\hat\beta^{0'}_{OLS} \right| x, z=0 \right)= \hat\beta^{0'}_{OLS}= \left.E\left(\hat\beta^{0'}_{OLS} \right| x\right)
$, $E(z y^0)=E(y^0|z)E(z)$, and $E((1-z) y^0 ) =E((1-z) \hat\beta^{0'}_{OLS}x)$.
Under, Assumption \[Ass1b\], \[Ass2b\], \[Ass3\] and Regularity Assumption \[RegAss2\]: $$\begin{aligned}
{\lim \limits_{n\to\infty}}\mbox{bias}_C(\hat \tau_{DR})&={\lim \limits_{n\to\infty}}\left[E(\hat\tau_{DR}) \right]- \tau= {\lim \limits_{n\to\infty}}\left[ \mu_1-\mu_0 \right]- \tau\\ &=\rho_1\sigma_1\lambda_1(g(x; \gamma))+\rho_0\sigma_0\lambda_0(g(x; \gamma)),\end{aligned}$$ where $$\begin{aligned}
{\lim \limits_{n\to\infty}}\mu_1&={\lim \limits_{n\to\infty}}\left[E(\hat\beta^{1'}_{OLS}x) +E\left(\frac{z(y^1-\hat\beta^{1'}_{OLS}x)}{\hat p(x)}\right)\right] \\
&= E(\hat\beta^{1'}_{OLS}x) +E\left[ E\left(\left.\frac{z(y^1-\hat\beta^{1'}_{OLS}x)}{p(x)}\right| x \right)\right] \\
&=E(\hat\beta^{1'}_{OLS}x) +E\left[\frac{1}{p(x)} E\left(\left.z(y^1-\hat\beta^{1'}_{OLS}x)\right| x \right)\right] \\
&=E(\hat\beta^{1'}_{OLS}x) +E\left[\frac{1}{p(x)} E\left(\left.y^1-\hat\beta^{1'}_{OLS}x\right| x, z=1 \right)\Pr(z=1|x)\right] \\
&=E(\hat\beta^{1'}_{OLS}x) +E\left[E\left(\left.y^1\right| x, z=1 \right) -\hat\beta^{1'}_{OLS}x\right] \\
&=E(\hat\beta^{1'}_{OLS}x) +E\left[E\left(\left.y^1\right| x \right)+\rho_1 \sigma_1 \lambda_1(g(x; \gamma)) \right] -E(\hat\beta^{1'}_{OLS}x)\\
&=E\left(f^1(x) \right)+\rho_1 \sigma_1 E\left(\lambda_1(g(x; \gamma)) \right),\end{aligned}$$ and $$\begin{aligned}
{\lim \limits_{n\to\infty}}\mu_0&={\lim \limits_{n\to\infty}}\left[E(\hat\beta^{0'}_{OLS}x) +E\left(\frac{(1-z)(y^0-\hat\beta^{0'}_{OLS}x)}{1-\hat p(x)}\right)\right] \\
&= E(\hat\beta^{0'}_{OLS}x) +E\left[ E\left(\left.\frac{(1-z)(y^0-\hat\beta^{0'}_{OLS}x)}{1-p(x)}\right| x \right)\right]\\
&=E(\hat\beta^{0'}_{OLS}x) +E\left[\frac{1}{1-p(x)} E\left(\left.(1-z)(y^0-\hat\beta^{0'}_{OLS}x)\right| x \right)\right] \\
&=E(\hat\beta^{0'}_{OLS}x) +E\left[\frac{1}{1-p(x)} E\left(\left.y^0-\hat\beta^{0'}_{OLS}x\right| x, z=0 \right)\Pr(z=0|x)\right] \\
&=E(\hat\beta^{0'}_{OLS}x) +E\left[E\left(\left.y^0\right| x, z=0 \right) -\hat\beta^{0'}_{OLS}x\right] \\
&=E(\hat\beta^{0'}_{OLS}x) +E\left[E\left(\left.y^0\right| x \right)-\rho_0 \sigma_0 \lambda_0(g(x; \gamma)) \right] -E(\hat\beta^{0'}_{OLS}x)\\
&=E\left(f^0(x) \right)-\rho_0 \sigma_0 E\left(\lambda_0(g(x; \gamma)) \right). \end{aligned}$$
Appendix C {#appendix-c .unnumbered}
==========
Variance of the outcome regression estimator {#variance-of-the-outcome-regression-estimator .unnumbered}
--------------------------------------------
The variance of the outcome regression estimator can be estimated by either the large sample variance or by the sandwich estimator since it is an m-estimator, see for instance [@Stefanski:2002] for details. In the simulations performed in this paper under Assumption \[Ass4\], both estimators have worked well. However, the sandwich estimator also performed well when Assumption \[Ass4\] was not fulfilled (as expected), and hence it is the one we recommend and use for the simulations and the real data example. Also, the large sample variance for $\hat\tau_{OR}^1$ given below require the estimation of ${{\bf \beta}}^1$ and $E\left(\mbox{Var}(y_i|x_i, z=1)\right)$, i.e. an additional model only used in the variance estimation, which we do not need for the sandwich estimator.
### Sandwich estimator {#sandwich-estimator .unnumbered}
Under Assumption \[Ass1\] the variance of $\hat \tau_{OR}^1$ can be estimated by: $$\widehat{{{\mbox{Var}}}}(\hat{\tau}^1_{OR}) = (A_n^{-1} B_n (A_n^{-1})' )_{(1,1)}/ n$$ where (1,1) stand for the element on row 1 and column 1 and:
$$A_n=\frac{1}{n}\sum \limits_{i =1}^n\left[
\begin{matrix}
z_i& z_i x_i \\
\boldsymbol 0_{(p+1)\times1}&(1-z)x' x \end{matrix} \right], \quad B_n=\frac{1}{n}\sum \limits_{i =1}^n\Psi (y_i, \boldsymbol{ \hat{\theta}})\Psi (y_i, \boldsymbol{ \hat{\theta}})',$$
and $$\Psi (y_i, \boldsymbol{ \hat{\theta}}) =\left[
\begin{matrix}
z_i(y_i- \hat \beta^0_{OLS}x_i -\tau^1) \\
(1-z_i)(y_i- \hat \beta^0_{OLS}x_i) x_i'
\end{matrix} \right].$$
Under Assumption \[Ass1\] and \[Ass2\] the variance of $\hat \tau_{OR}$ can be estimated by: $$\widehat{{{\mbox{Var}}}}(\hat{\tau}_{OR}) = (A_n^{-1} B_n (A_n^{-1})' )_{(1,1)}/ n$$ where (1,1) stand for the element on row 1 and column 1 and:
$$A_n=\frac{1}{n}\sum \limits_{i =1}^n\left[
\begin{matrix}
1& -x_i &x_i \\
\boldsymbol 0_{(p+1)\times1}&zx' x& \boldsymbol 0_{(p+1)\times(p+1)}\\
\boldsymbol 0_{(p+1)\times1}&\boldsymbol 0_{(p+1)\times(p+1)}& (1-z)x' x
\end{matrix} \right],$$
$$B_n=\frac{1}{n}\sum \limits_{i =1}^n\Psi (y_i, \boldsymbol{ \hat{\theta}})\Psi (y_i, \boldsymbol{ \hat{\theta}})',$$
and $$\Psi (y_i, \boldsymbol{ \hat{\theta}}) =\left[
\begin{matrix}
(\hat \beta^1_{OLS}x_i- \hat \beta^0_{OLS}x_i) - \tau \\
z_i(y_i- \hat \beta^1_{OLS}x_i) x_i' \\
(1-z_i)(y_i- \hat \beta^0_{OLS}x_i) x_i'
\end{matrix} \right].$$
### Large sample variance {#large-sample-variance .unnumbered}
The large sample variance of $\hat \tau_{OR}^1$ is given by: $$\mbox{Var}(\hat \tau_{OR}^1) \simeq n_1^{-1} \left[2 \cdot E\left(\mbox{Var}(y_i|x_i, z=1)\right) +(\beta^1-\beta^0)\mbox{Cov}\left(x_i|z_i=1\right)(\beta^1-\beta^0)' \right],$$ which under Assumption \[Ass1\], \[Ass2\] and \[Ass4\] can be estimated by $$\widehat{{\mbox{Var}}}(\hat \tau_{OR}^1) \simeq n_1^{-1} \left[2 \left(\frac{\sum \limits_{i \in \mathcal{I}_1}e_i^2}{n_1-1} \right) +(\hat \beta^1-\hat \beta^0)\widehat{{{\mbox{Cov}}}}\left(x_i|z_i=1\right)(\hat \beta^1-\hat \beta^0)' \right],$$ where $e_i$ and $\hat \beta^1$ are the residuals and coefficients estimates from the OLS regression of ${{\bf X}}_1$ on ${{\bf y}}_1$, and similarly $\hat \beta^0$ are the coefficients estimates from the OLS regression of ${{\bf X}}_0$ on ${{\bf y}}_0$.
Under Assumption \[Ass1\] and \[Ass4a\]: $$\begin{aligned}
\mbox{Var}(\hat \tau_{OR}^1) &\simeq \mbox{Var}(\hat \tau_{OR}^1|{{\bf z}})\\
&=E\left.\left[\mbox{Var}\left(\frac{1}{n_1}\sum \limits_{i =1}^n\left(z_i( y_i -\hat\beta^{0'}_{OLS}x_i) \right)\right| x, {{\bf z}}\right) \right] + \mbox{Var}\left[E\left(\hat \tau_{OR}^1 | x, {{\bf z}}\right) \right]\\
&=n_1^{-2} \left[ \sum \limits_{i =1}^n z_i \left( E\left(\mbox{Var}(y_i|x_i, z=1)\right) \right) +\mbox{Var}\left((\beta^1-\beta^0)\sum \limits_{i =1}^nz_ix_i\right)\right] \\
&+n_1^{-2} \left[ E\left( \sum \limits_{i =1}^n\sum \limits_{j =1}^n z_i z_j\mbox{Cov}(\hat\beta^{0'}_{OLS}x_i,\hat\beta^{0'}_{OLS}x_j | x, {{\bf z}})\right) \right] \\
&=n_1^{-1} \left[ E\left(\mbox{Var}(y_i|x_i, z=1)\right) +(\beta^1-\beta^0)\mbox{Cov}\left(x_i|z_i=1\right)(\beta^1-\beta^0)' \right]\\
&\qquad \qquad\qquad \qquad \qquad + n_1^{-2} E \left( \boldsymbol{1}_{1\times n_1}{{\bf X}}_1\mbox{Cov}(\hat\beta^{0}_{OLS}| x, {{\bf z}}) {{\bf X}}_1'\boldsymbol{1}_{n_1\times1}\right)\\
&=n_1^{-1} \left[2 \cdot E\left(\mbox{Var}(y_i|x_i, z=1)\right) +(\beta^1-\beta^0)\mbox{Cov}\left(x_i|z_i=1\right)(\beta^1-\beta^0)' \right],\end{aligned}$$ since $$\begin{aligned}
\sum \limits_{i =1}^n\sum \limits_{j =1}^n z_i z_j\mbox{Cov}(\hat\beta^{0'}_{OLS}x_i,\hat\beta^{0'}_{OLS}x_j | x, {{\bf z}}) &=
\sum \limits_{i =1}^n\sum \limits_{j =1}^n z_i z_j x_i'\mbox{Cov}(\hat\beta^{0}_{OLS}| x, {{\bf z}}) x_j\\
&=\boldsymbol{1}_{1\times n_1}{{\bf X}}_1\mbox{Cov}(\hat\beta^{0}_{OLS}| x, {{\bf z}}) {{\bf X}}_1'\boldsymbol{1}_{n_1\times1}\\
&=\boldsymbol{1}_{1\times n_1}{{\bf X}}_1\left[\mbox{Var}(y_i|x_i, z=1) \left({{\bf X}}_1' {{\bf X}}_1\right)^{-1} \right] {{\bf X}}_1'\boldsymbol{1}_{n_1\times1}\\
&=n_1 \mbox{Var}(y_i|x_i, z=1).\end{aligned}$$
The large sample variance of $\hat \tau_{OR}$ is given by: $$\begin{aligned}
\mbox{Var}(\hat \tau_{OR})&\simeq n^{-2} \left( E\left[ \boldsymbol{1}_{1\times n}{{\bf X}}\, \left[ \mbox{Cov}\left( \left. \hat\beta^{1}_{OLS} \right| {{\bf X}},{{\bf z}}\right) + \mbox{Cov}\left( \left. \hat\beta^{0}_{OLS} \right| {{\bf X}},{{\bf z}}\right)\right] {{\bf X}}' \, \boldsymbol{1}_{n\times 1} \right] \right)\\
&\qquad \qquad \qquad \qquad + n^{-1}(\beta^1-\beta^0)\mbox{Cov}(x_i)(\beta^1-\beta^0)',\end{aligned}$$ which under Assumption \[Ass1\], \[Ass2\] and \[Ass4\] can be estimated by: $$\begin{aligned}
\widehat{{{\mbox{Var}}}}(\hat \tau_{OR})& = n^{-2} \left( \boldsymbol{1}_{1\times n}{{\bf X}}\, \left[\left(\frac{\sum \limits_{i \in \mathcal{I}_1}e_i^2}{n_1-1} \right){{\bf X}}_1' {{\bf X}}_1 + \left(\frac{\sum \limits_{i \in \mathcal{I}_0}e_i^2}{n_0-1} \right){{\bf X}}_0' {{\bf X}}_0\right] {{\bf X}}' \, \boldsymbol{1}_{n\times 1} \right)\\
&\qquad \qquad \qquad + n^{-1}(\hat\beta^1-\hat\beta^0)\widehat{{\mbox{Cov}}}(x_i)(\hat\beta^1-\hat\beta^0)'.\end{aligned}$$
Under Assumption \[Ass1\] , \[Ass2\] and \[Ass4\]: $$\begin{aligned}
\mbox{Var}(\hat \tau_{OR})&\simeq \mbox{Var}(\hat \tau_{OR}|{{\bf z}})\\
&=E\left[ \mbox{Var}\left( \left. \frac{1}{n}\sum \limits_{i =1}^n\left( \hat\beta^{1'}_{OLS}x_i-\hat\beta^{0'}_{OLS}x_i\right) \right| {{\bf X}}, {{\bf z}}\right) \right]\\
&\qquad \qquad \qquad \qquad + \mbox{Var}\left[ E\left( \left. \frac{1}{n}\sum \limits_{i =1}^n\left( \hat\beta^{1'}_{OLS}x_i-\hat\beta^{0'}_{OLS}x_i\right) \right| {{\bf X}}, {{\bf z}}\right) \right] \\
&= n^{-2} \left( E\left[ \mbox{Var}\left( \left.\sum \limits_{i =1}^n (\hat\beta^{1'}_{OLS}-\hat\beta^{0'}_{OLS})x_i \right| {{\bf X}}, {{\bf z}}\right) \right] \right) \\
&\qquad \qquad \qquad \qquad +n^{-1}(\beta^1-\beta^0)\mbox{Var}(x_i) (\beta^1-\beta^0)' \\
&= n^{-2} \left( E\left[ \boldsymbol{1}_{1\times n}{{\bf X}}\, \left[ \mbox{Cov}\left( \left. \hat\beta^{1}_{OLS} \right| {{\bf X}},{{\bf z}}\right) + \mbox{Cov}\left( \left. \hat\beta^{0}_{OLS} \right| {{\bf X}},{{\bf z}}\right)\right] {{\bf X}}' \, \boldsymbol{1}_{n\times 1} \right] \right)\\
&\qquad \qquad \qquad \qquad + n^{-1}(\beta^1-\beta^0)\mbox{Cov}(x_i)(\beta^1-\beta^0)'.\end{aligned}$$
Variance of the double robust estimator {#variance-of-the-double-robust-estimator .unnumbered}
---------------------------------------
To caluculate the variance of the doubly robust estimators we use the sandwich estimator since they both are m-estimators, see [@Stefanski:2002] for more details.
Under Assumption \[Ass1\] the variance of $\hat \tau_{DR}^1$ can be estimated by: $$\widehat{{{\mbox{Var}}}}(\hat{\tau}^1_{DR}) = (A_n^{-1} B_n (A_n^{-1})' )_{(1,1)}/ n$$ where (1,1) stand for the element on row 1 and column 1 and:
$$A_n=\frac{1}{n}\sum \limits_{i =1}^n\left[
\begin{matrix}
z_i& \left(z_i - \frac{(1-z_i)}{\Phi(\hat {{\bf \gamma}}x_i)}\right) x_i & (1-z_i)(y_i- \hat \beta^0_{OLS}x_i) \frac{\phi(\hat {{\bf \gamma}}x_i)}{\Phi(\hat {{\bf \gamma}}x_i)^2}x \\
\boldsymbol 0_{(p+1)\times1}&(1-z)x' x &\boldsymbol 0_{(p+1)\times(p+1)}\\
\boldsymbol 0_{(p+1)\times1}&\boldsymbol 0_{(p+1)\times(p+1)}& - \frac{d \Psi_3}{d {{\bf \gamma}}}
\end{matrix} \right],$$
$$B_n=\frac{1}{n}\sum \limits_{i =1}^n\Psi (y_i, \boldsymbol{ \hat{\theta}})\Psi (y_i, \boldsymbol{ \hat{\theta}})',$$
and $$\Psi (y_i, \boldsymbol{ \hat{\theta}}) =\left[
\begin{matrix}
\Psi_1 \\
\Psi_2 \\
\Psi_3
\end{matrix} \right]=\left[
\begin{matrix}
z_i(y_i- \hat \beta^0_{OLS}x_i) - (1-z_i)\frac{y_i- \hat \beta^0_{OLS}x_i}{1-\Phi(\hat {{\bf \gamma}}x_i)} - z_i \tau^1 \\
(1-z_i)(y_i- \hat \beta^0_{OLS}x_i) x_i' \\
(z_i\frac{\phi(\hat {{\bf \gamma}}x_i)}{\Phi(\hat {{\bf \gamma}}x_i)}-(1-z)\frac{\phi(\hat {{\bf \gamma}}x_i)}{1-\Phi(\hat {{\bf \gamma}}x_i)})x_i'
\end{matrix} \right].$$
If Assumption \[Ass4\] also holds some of the terms will converge to zero, i.e. the variance expression can be simplified for large samples. However, in the simulations performed for this paper, the simplified estimator have poor precision compared with the estimator given above. For the estimation of the standard error of $\hat \tau_{DR}$, however, a simplified version of the sandwich estimator performs rather good, hence this is what is proposed below.
Under Assumption \[Ass1\], \[Ass2\], \[Ass4\] and a correctly specified propensity score, the standard error of $\hat \tau_{DR}$ can be estimated by: ([@Lunceford:2004]) $$\widehat{{{\mbox{Var}}}}(\hat{\tau}_{DR}) = n^{-2} \sum \limits_{i =1}^n\hat{I}_i^2$$ $$\hat{I}_i= \hat\beta^{1'}_{OLS}x_i-\hat\beta^{0'}_{OLS}x_i
+z_i\, \frac{y_i-\hat\beta^{1'}_{OLS}x_i}{\hat{p}(x_i)}
-(1-z_i)\frac{y_i-\hat\beta^{0'}_{OLS}x_i}{1-\hat{p}(x_i)} -\hat{\tau}_{DR}.$$
Appendix D {#appendix-d .unnumbered}
==========
\[htbp\]
![Boxplot of the width of two 95% uncertainty intervals (assuming $\rho_j \in [0,0.2]$, green, and $\rho_j \in [0,0.4]$, blue, $j=0,1$) and the 95% confidence interval, red, for the doubly robust (DR) and outcome regression (OR) estimator of $\tau^1$ under design A-D for $\rho_0=\rho_1=0.05$ and $0.3$, with sample size 250. The empirical coverage of each interval is written below each boxplot and the number of outliers that lie outside the window is written at the top of the window above each boxplot.[]{data-label="t1.250.fig"}](t1_p1_250.pdf){width="\textwidth"}
![Boxplot of the width of two 95% uncertainty intervals (assuming $\rho_j \in [0,0.2]$, green, and $\rho_j \in [0,0.4]$, blue, $j=0,1$) and the 95% confidence interval, red, for the doubly robust (DR) and outcome regression (OR) estimator of $\tau^1$ under design A-D for $\rho_0=\rho_1=0.05$ and $0.3$, with sample size 250. The empirical coverage of each interval is written below each boxplot and the number of outliers that lie outside the window is written at the top of the window above each boxplot.[]{data-label="t1.250.fig"}](t1_p2_250.pdf){width="\textwidth"}
\[htbp\]
![Boxplot of the width of two 95% uncertainty intervals (assuming $\rho_j \in [0,0.2]$, green, and $\rho_j \in [0,0.4]$, blue, $j=0,1$) and the 95% confidence interval, red, for the doubly robust (DR) and outcome regression (OR) estimator of $\tau^1$ under design A-D for $\rho_0=\rho_1=0.05$ and $0.3$, with sample size 500. The empirical coverage of each interval is written below each boxplot and the number of outliers that lie outside the window is written at the top of the window above each boxplot.[]{data-label="t1.500.fig"}](t1_p1_500.pdf){width="\textwidth"}
![Boxplot of the width of two 95% uncertainty intervals (assuming $\rho_j \in [0,0.2]$, green, and $\rho_j \in [0,0.4]$, blue, $j=0,1$) and the 95% confidence interval, red, for the doubly robust (DR) and outcome regression (OR) estimator of $\tau^1$ under design A-D for $\rho_0=\rho_1=0.05$ and $0.3$, with sample size 500. The empirical coverage of each interval is written below each boxplot and the number of outliers that lie outside the window is written at the top of the window above each boxplot.[]{data-label="t1.500.fig"}](t1_p2_500.pdf){width="\textwidth"}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'From the dawn of civilization, people have used folktales and stories to share information and knowledge. After the invention of printing in the 15th century, technology provided helpful yet complicated utilities to exchange ideas. In the present computerized world, the art of storytelling is becoming more influential through the unprecedented multimedia capabilities of computers. In this article, we introduce a state-of-the-art presentation software by which academicians can present nonlinear topics efficiently and sharpen their storytelling skills. We show how the proposed software can improve the scientific presentation style. We conducted a survey to measure the attractiveness of proposed utility among other alternatives. Results show that academicians prefer the proposed platform to others.'
author:
- Bihter Avşar
- Danial Esmaeili Aliabadi
- Edris Esmaeili Aliabadi
- Reza Yousefnezhad
bibliography:
- 'Ref.bib'
date: 'Accepted: '
subtitle: a New Storytelling Presentation Software for Academic Purposes
title: 'Academic Presenter:'
---
Introduction {#intro}
============
A well-told story can be memorized and recalled quickly. People tend to learn better when the data is transformed into story, and this is the reason societies pass on their values by means of stories to the next generations. As is expressed by @ref1, a story is worth a thousand pictures since an image can talk about a single moment but a story can express the sequence of events. Therefore, developing storytelling skills has a great value. @ref3 claims that storytelling skills can persuade listeners to feel more involved.
Advantages of storytelling induce researchers to exploit storytelling techniques for presentation and educational purposes. For instance, @ref2 examine the performance of an interactive storytelling system for a public archaeology heritage presentation in Belgium. @ref5 create a novel software architecture that couples 3D representation and storytelling for creating engaging linear narrations that can be shared on the web.
The process of information visualization can help us to provide meaningful information for viewer. However, visualization problems can become challenging due to the complexities such as extensive data volumes [@ref7]. @ref4 address the necessity of solving high complexities with visualization problem to relieve the intrinsic limitations of human cognitive capacity and information processing ability. They suggest applying storytelling in the field of information visualization can lead to better information presentation.
Presenting scientific papers requires different qualifications than presenting general topics [@davis2005scientific]. In this study, we introduce new presentation software by which academicians can augment storytelling skills and present nonlinear topics efficiently. Afterward, our proposed software will be called *Academic Presenter*.
The remainder of the manuscript is organized as follows. Section \[literature\] reviews the history of available software products and introduces the proposed software. Section \[problems\] addresses issues related to a scientific presentation that have not previously been completely solved. Section \[problems\] also suggests our solution for each problem. Section \[casestudy\] studies the attractiveness of the proposed software among common presentation utilities. Finally, Section \[conclusion\] concludes.
Related Work {#literature}
============
In this section, we begin with the history of current presentation utilities and discuss the associated advantages and disadvantages of each style. Following this, we introduce the proposed software solution that creates a new paradigm in modeling visual contents by combining previous methods.
Conventional Presentation Utilities
-----------------------------------
From the early stages, multimedia capabilities of computers exhibited a suitability for demanding tasks such as presentation [@keckler2011gpus]. Initially, *Presenter* was released by *Forethought* [@ref8] and in 1987 it was renamed *PowerPoint*. *Microsoft* embedded *PowerPoint* in the *Office* suite in 1990. *PowerPoint* has been designed to create linear presentations through slides. Because of high accessibility, it gained acceptance in academia [@pippert1999multiple]. @ref11 show the positive effect of creating presentations with *PowerPoint* on students’ grades. However, @susskind2005powerpoint claims that *PowerPoint* won’t affect academic performance but enhance students’ attitudes and self-efficacy about the course.
Gradually, high accessibility and linearity create issues, especially in universities [@tufte2003powerpoint]. Speakers create slides merely to present rather than focusing on their messages. Using slide-based presentation software together with students’ lack of experience deteriorates students’ organization skills. Also, the linearity of slide-based software products forces the presenter to simplify sophisticated subjects to a set of bullet items which is misleading for decision-making [@tufte2006cognitive]. Moreover, a linear presentation is not suitable to illustrate the complexity of an issue; nonetheless, @ref9 tried to find a solution for this issue by using a directed graph structure approach.
Another movement in computer graphics started concurrently with slideware. In November 1996, *Macromedia* released the first version of *Flash*. *Flash* is a canvas-based presentation tool that supports vector-based animation. Canvas is like an infinite and borderless workspace in which building blocks form a presentation. Unlike slide-based technology, canvas-based technology offers enough flexibility to create nonlinear presentations[^1]. Yet, it was difficult to create a presentation with *Flash* since it required programming skill. Nowadays, new companies such as *Prezi* [@Prezi] are trying to simplify canvas-technology for building presentations; however, this simplification may confine flexibility. *Prezi* demonstrates positive results in classrooms [e.g., @brock2013tale; @anderson2013diversifying; @vspernjak2014prezi].
Table \[tab:smackdown\] categorizes available presentation software with respect to employed technologies and price. The first and second columns indicate whether the product is canvas-based or slide-based, respectively. The third and fourth columns determine the availability of the corresponding product as a web application (online) or conventional software (offline). Finally, the last column shows which one is free.
\[tab:smackdown\]
As one can see, only two presentation tools offer both canvas-based and slide-based technologies simultaneously. Additionally, the table imply that canvas-technology is less popular than the others, although among all presentation utilities, *Adobe Flash* and *Prezi* are known as revolutionary products. In this table, we added *Academic Presenter* as well.
Proposed Software Solution
--------------------------
*Academic Presenter* combines the potency of slide-based presentation software products with canvas-based [^2]. Users can switch between two common presentation trends based on the level of details; for introducing general topics, they can employ a nonlinear flow and switch to a conventional linear presentation for exhibiting details. Figure \[Fig-Switch\] depicts a sample in which we used both nonlinear and linear flows. From (a) to (b) and then from (b) to (c), a user can zoom, pan, and rotate by using mouse or touch-screen. However at (d), a linear flow can carry the talk to the next topic where the user may switch to a nonlinear flow again. Thanks to the vector-based canvas of *Academic Presenter*, zooming into a particular region will not affect contents’ quality. By taking the advantage of proposed framework, the users can combine even mind-map diagrams and conventional slides.
{width="100.00000%"}
@chou2015prezi investigate the effectiveness of various digital presentation tools (more specifically *PowerPoint* and *Prezi*) on students’ learning performance. Their results show that *Prezi* is a more efficient instructional medium for knowledge acquisition compared with traditional instruction; however, *PowerPoint* demonstrated instructional effectiveness on only the long-term learning retention of the students compared with traditional instruction. Hence; combining the power of slideware (such as *PowerPoint*) and a canvas-based product (such as *Prezi*) can enhance the effectiveness of current digital presentation tools in universities. Although Table \[tab:smackdown\] indicates that *SlideDog* is also offering both presentation technologies, the user has to create *PowerPoint* and *Prezi* projects separately in the mentioned tools.
Our proposed software also enables users to build an engaging presentation by combining different types of audio visual contents: including image, audio, video, vector-based shape, PDF document, LaTeX code, and handwriting.
Because *Academic Presenter* harnesses the power of a video graphics card without an intermediary, it is faster. Figure \[Fig-Video\] shows the interactions among the video graphics card and application to play a video. The bottom line is that *Academic Presenter* is free software, which makes it an interesting option for students on a tight-budget.
![Playing a video with DirectShow[]{data-label="Fig-Video"}](Fig-Video){width="50.00000%"}
In the following sections, we focus on the application of storytelling techniques on a common scientific presentation.
Application of storytelling techniques on a Scientific Presentation {#problems}
===================================================================
In the first subsection, we propose applying mind map diagrams for presenting a typical literature review, and we explain how *Academic Presenter* can help academicians with this. Next, we demonstrate how using animation and storytelling techniques can assist infographics to be more clear and informative. Finally, the effect of *Academic Presenter*’s whiteboard animation on teaching quality will be discussed.
Literature Review with Mind-map
-------------------------------
In any scientific presentation, researchers have to review and discuss published information. Literature review both summarizes and synthesis important information. Unfortunately, common methods to deal with literature review are as follows:
- Listing the most relevant papers as bullet point items.
- Organizing published information inside tables and comparing them with respect to some criteria.
Indeed, these ways of organizing information are not mind-friendly since listeners have to digest and categorize information simultaneously. However, the presentation time is not enough for both thinking deeply and listening carefully. Vector-based canvas of *Academic Presenter* offers another way of organizing information; using mind-map diagrams. Mind mapping has been defined as “visual, nonlinear representations of ideas and their relationships" [@ref14]. Mind-map is also considered as a powerful diagramming tool that plays a significant role in collaborative or group storytelling [@nakamura2010zuzie]. @liu2011enhanced demonstrate the benefits of mind mapping (concept mapping) on students storytelling skills.
By using mind-map, viewers can categorize subjects and find their relationship with the main topic. For example, Figure \[Fig1\] depicts the literature review of a deregulated electricity market using a mind-map diagram. From the central topic toward each branch, more details are added to the parent nodes; thus, doing this provides classification rule to categorize subjects. Each branch ends with a red node containing studies similar to the attached branch. This categorization method is easier to memorize and recall [@ref15]. Moving from one branch to another, a presenter begins by discussing general topics and finishes with more technical information; therefore, viewers might be less likely to lose concentration as a result of listening to details for a long duration. As mentioned, *Academic Presenter* supports both slide-based and canvas-based technologies; therefore, a presenter can switch to slide-mode to explain linear topics inside each node. Interested readers will be invited to watch “Why *Academic Presenter*? (Part 1 - Literature Review)"[^3] for more details.
{width="100.00000%"}
Animated Infographics
---------------------
Information graphics (or infographics) is an innovative medium to visualize data clearly and in an engaging manner. Infographics are enchanting storytelling tools for transforming data into knowledge, as they capture a reader’s attention by utilizing principles of graphic design. These characteristics assist infographics to be highly popular for transferring data to diverse audiences [@bateman2010useful; @borkin2013makes].
However, packing all data and information in a single image can cause a sense of confusion since people may not see the patterns clearly. @harrison2015infographic examine the impact of color and complexity on impression level of audiences and conclude that participants reacted differently to infographics due to the difference in age, educational background, and gender.
A solution to this problem is using timeline animation instead of a single image. Therefore, viewers are gradually becoming familiar with the presented data. The combination of keyframes and infinite canvas in Academic Presenter help designers to prioritize different sections of infographics and add animation to static infographics. Figure \[Fig:info\] displays a sample in where static infographic is converted to an animated one. In Figure \[Fig:info\], the leftmost image is static but the right panel is showing the development of the story with time. Note that designers can zoom and pan in each keyframe to recommend a viewport to audiences.
{width="100.00000%"}
Effect of Handwriting
---------------------
Although typing by computer is easier than writing by hand, there remain many debates about the constructive effects of writing by hand on learning [@ref13]. @ref12 explains why writing by hand can assist learning. There is anecdotal evidence that dynamic sketches together with narration may be more efficient for delivering information than traditional presentations [e.g., @dean2006beyond; @roam2009back]. Consequently, researchers invent new teaching aids compatible with this storytelling technique. For instance, @lee2013sketchstory propose a new narrative visualization (specifically whiteboard animation) device that uses pen and touch interactions to leverage the narrative storytelling attributes. Results confirm that the audience is more engaged by presentations that done with offered tool than *PowerPoint*. Besides, writing by hand allows more flexibility to the writer, especially in abstract courses such as mathematics. Nowadays, many educational websites are using whiteboard animation technique to teach various topics [e.g., @ASAPScience; @RSAAnimate].
In spite of progress in teaching instruments, many professors still prefer to teach by writing on a board. However, by looking at the entire academic career of a professor, one might infer that s/he often teaches almost the same materials each semester to different groups of students. We suggest employing digitizer to utilize the advantages of writing by hand yet alleviate the repetition issue. Nowadays, digitizers are becoming an indispensable part of any computer. Users can record their hand movements on screen by using digitizers. The information which can be retrieved from digitizers is as follows: 2D-position, pressure level, starting time, finishing time, and color. Each time the user draws a line (stroke) on screen, the digitizer records the position of the digitizer’s tip on screen and pressure level. The pressure sensitivity of all digitizers is not the same, but even low-quality digitizers can sense the pressure accurately enough to emulate the movement. Figure \[Fig3\] shows effect of neglecting pressure on a stroke. Figure \[Fig-Pen\] illustrates the employed data structure. The stroke collection consists of strokes and each stroke corresponds to one curve on the canvas.
![Effect of ignoring pressure on a stroke[]{data-label="Fig3"}](Fig4-2.jpg){width="50.00000%"}
![Data structure of digital ink class[]{data-label="Fig-Pen"}](Fig-Pen.jpg){width="50.00000%"}
*Academic Presenter* utilizes a digitizer in presentation, not only to annotate on screen but also to replay the handwriting wherever is necessary. In toolbox, a handful of different pens and highlighters is available at users’ fingertips. Figure \[Fig\_HW\] displays the handwriting toolbox. Every movement is editable and precise. Also, user can increase animation speed to save presentation time. We redirect an interested reader to watch “Why *Academic Presenter*? (Part 5 - Handwriting)" [^4] for more details.
{width="70.00000%"}
Analyzing the Attractiveness of Academic Presenter {#casestudy}
==================================================
*Academic Presenter* is designed for academic environments since presentation has educational and inspirational nature. Potential users are students of universities, teachers in high school and professors. Because our focus is to solve presentation problems related to academic environments, we tried to find flaws in current tools which affect the presentations the most. To analyze the future position of *Academic Presenter* in academia, we exploit Analytic Hierarchy Process (AHP) method [@saaty1988analytic]; therefore, we define four criteria: *Price*, *Number of users*, *Simplicity*, and *Applicability in academia*. We collect quantitative information such as price and number of users from the websites and documents; however for qualitative criterion such as simplicity, we asked from experts in the field of presentation. We select the most significant competitors. The alternatives are listed as follows:
- **Office 365** (including *PowerPoint*) is the brand name adopted by Microsoft for a collection of software plus services subscriptions that provides web-based productivity software and services to its subscribers.
- **Prezi** is a cloud-based storytelling tool for presenting ideas on a virtual canvas. The product employs a zooming user interface, which allows users to zoom in and out of their visual contents, and enables users to navigate through information within a 2.5D space on the Z-axis.
- **SlideShare** is a web-based slide hosting service. Users can upload *PowerPoint*, PDF, and Keynote files privately or publicly. Slide decks can then be viewed on the site itself, on hand held devices or embedded on other websites. *SlideShare* is considered to be similar to YouTube, but for slide shows.
- **PowToon** is a cloud-based for creating animated presentations and animated explainer videos.
- **emaze** is an online presentation platform built on html5 technology. Users can create, manage and share their presentations through their cloud-based system. It offers 3D animations and video backgrounds.
The retrieved information from competitors are displayed in Table \[tab:info\].
-------------------- ----------- ----------------- ----- ---------------
Presentation Price Number of users Applicability
tools (\$/year) (millions) in academia
Academic Presenter 0 0.022 0.5 0.75
Office 365 79.99 15.2 0.5 0.88
Prezi 159 40 0.8 0.58
SlideShare 228 70 1 0.25
PowToon 228 6 0.5 0.50
emaze 178.92 0.011 1 0.58
-------------------- ----------- ----------------- ----- ---------------
\[tab:info\]
As one can perceive from Table \[tab:applicability\], applicability in academia is calculated based on availability of essential features that may help students and professors during their presentations. Also, there are some features with half the unit value for some alternatives which means mentioned feature is not provided at a satisfactory level.
-------------------------- ----------- -------- ------ ------ ------ ------
Applicability Academic Office
in academia Presenter 365
Supporting Images 1 1 1 0 1 1
Supporting Sounds 1 1 1 0 1 1
Supporting Videos 1 1 1 0 1 1
Formula and Latex 1 1 0 0 0 0
Online Presentation 0.5 1 1 1 1 1
Offline Presentation 1 1 1 0 0 0
Nonlinear Presentation 1 0 1 0 0 0
Linear Presentation 1 1 0 1 1 1
Annotation 1 0.5 0 0 0 0
Supporting Second Screen 0 1 0 0 0 0
Charts 0 1 0 0 0 1
Running on different OSs 0.5 1 1 1 1 1
Weight 0.75 0.88 0.58 0.25 0.50 0.58
-------------------------- ----------- -------- ------ ------ ------ ------
\[tab:applicability\]
Furthermore, we invite users to judge about the importance of each criterion. A group of 50 people have attended in a questionnaire. The composition of the attendees are as follows: graduate students 54%, undergraduate students 20%, and instructors 8%. Figure \[Fig-Chart\] delineates the detailed information of the participants on a pie chart.
![The detailed information of the attendees in the survey[]{data-label="Fig-Chart"}](Fig-chart.jpg){width="50.00000%"}
The resulted judgements are reported in Table \[tab:judgement\]. Based upon pairwise comparisons, applicability is the most influential factor. The inconsistency of judgement matrix is 1% which is in acceptable range.
Criterion [Weight]{}
------------------- ------- ------- ------- ------- ------------
[Price]{} 1 1.223 0.820 0.888 0.241
[Num. of users]{} 0.818 1 0.670 0.670 0.193
[Applicability]{} 1.220 1.492 1 1.084 0.294
[Simplicity]{} 1.126 1.377 0.923 1 0.271
\[tab:judgement\]
AHP estimates *Academic Presenter*’s position among competitors regarding retrieved information and pairwise judgements (see Figure \[Fig-Alt\]).
![Alternatives’ ranking based on AHP method[]{data-label="Fig-Alt"}](Fig-Alt.jpg){width="50.00000%"}
As one can see in Figure \[Fig-AHP\], *Academic Presenter* had better off in term of price. Sensitivity analysis of our result is showing that *Academic Presenter*’s rank is relatively stable on simplicity and applicability. Although *Academic Presenter* is showing a promising rank among other alternatives, yet the difference between *Prezi*, *Office 365* and *Academic Presenter* is negligible.
{width="75.00000%"}
Conclusion
==========
Presenting scientific papers need different requirements than presenting general topics. Most of available software solutions are adjusted to meet business presentations’ demands. However, presenting a nonlinear scientific subject is beyond their capabilities. In this paper, we presented a new presentation software that facilitates delivering nonlinear topics. Our free presentation software enables users to enhance their storytelling skill. Users can switch between two common presentation trends based on the level of details; for introducing general topics, they can employ a nonlinear flow and switch to a conventional linear presentation for exhibiting details. Also, we introduce new components in the presented software solution that may help academicians to teach abstract courses more efficiently. Finally, a survey is conducted by asking eligible attendees to prioritize different aspects of a presentation utility. We exploit Analytic Hierarchy Process method to analyze the expected rank of proposed tool among popular alternatives. The results are indicating that the proposed software is more attractive than current software solutions.
Although the proposed utility is the combination of slide-based and canvas-based products and researchers investigated on each technology separately, assessing the effectiveness of proposed tool on the knowledge acquisition of students is a valuable future work.
Authors of this article would like to express their very great appreciation to Daniel Lee Calvey for his valuable and constructive suggestions. Also, the authors would like to thank all those who cooperated and dedicated their valuable times in the survey.
Compliance with ethical standards {#compliance-with-ethical-standards .unnumbered}
=================================
**Conflict of interest** This study was self-funded and no conflict of interest exists.
[^1]: A nonlinear presentation is a presentation style in which user defines paths for illustrating the relationship among concepts by zooming, panning, and rotating screen animations [@good2002zoomable; @bean2012presentation]
[^2]: We refer the interested readers to watch `https://youtu.be/rMG8-wzCaD8` for more details.
[^3]: `https://youtu.be/LUWr8pqJjzg`
[^4]: `https://youtu.be/U-oNFjBtzfE`
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'H. Gómez,'
- 'D. Gibert,'
- 'C. Goy,'
- 'K. Jourde,'
- 'Y. Karyotakis,'
- 'S. Katsanevas,'
- 'J. Marteau,'
- 'M. Rosas-Carbajal,'
- 'A. Tonazzo.'
title: Forward scattering effects on muon imaging
---
Introduction {#sec:Introduction}
============
The idea to use muons produced in the Earth’s atmosphere by cosmic rays as a scanning method of anthropic or geological structures, the so-called muon imaging, was proposed soon after the discovery of these muons [@Anderson1; @Anderson2; @Auger]. Muon imaging leverages the capability of cosmic muons to pass through hundreds of metres or even kilometres of ordinary matter with an attenuation mainly related to the length and density of this matter encountered by the muons along their trajectory before their detection [@Nagamine]. As this attenuation is principally caused by muons absorption and scattering, muon imaging can be mainly performed using two different techniques. The first one is the so-called transmission and absorption muography [@Lesparre_2010; @Marteau_2012]. This technique relies on the well-known radiography concept (widely used, for example, in medicine with X-rays), based on the muon energy loss and its consequently probability to cross a given amount of material. The second is known as deviation muography, which relies on the measurement of the muon track deviation to determine the object density [@Borozdin; @Procureur]. For the first technique, studying all the directions for which muons go through the studied object, and knowing its external shape, it is possible to obtain a 2D mean density image. Thus, muon imaging provides a non-invasive and remote scanning technique utilisable even for large objects, where the detection set-up may be relatively far away from the – potentially dangerous – target (e.g. domes of active volcanoes or damaged nuclear reactors).
One of the first studies performed based on muon imaging dates from 1955, being the scanning of the rock overburden over a tunnel in Australia [@George]. Later, other applications from mining [@Malmqvist] to archaeology were proposed, both in the 70s. For the latter case, some measurements have been already performed, as the exploration of the Egyptian Chephren [@Alvarez] and Khufu [@SPNature] pyramids. Nowadays, thanks to the improvements on the detector performance, and also to their autonomy and portability, muon imaging reveals itself to be a scanning technique competitive and complementary to others non-invasive methods as seismic and electrical resistivity tomography or gravimetry. This has led to its proposal and utilisation in a wide range of fields.
In addition to the above-mentioned applications (archaeology and mining), two others stand out. The first one is related with geophysics, more precisely with the monitoring of volcanoes. This has an important benefit both from a scientific and social point of view. The continuous monitoring of volcanoes helps to understand their internal dynamics, a key feature in the risk assessment. The other application, more related with particle physics, was motivated by the necessity to characterize the overburden of underground laboratories hosting various experiment detectors. It is worth mentioning other applications related to civil engineering and nuclear safety. For the first one, it will be possible for example to scan structures looking for defects. For the case of nuclear safety, set-ups looking for the transport of radioactive materials and wastes already work in cooperation with homeland security agencies. Moreover, the study of nuclear reactors looking for structural damages have been already used, as in the case of the recent Fukushima nuclear power plant accident [@Kume], and it is being considered as a remote scanning method.
As mentioned, the improvement on the detectors used for muon imaging has been one of the main reasons for the renewal of this technique. Better detectors provide a better angular resolution for the muon direction reconstruction and improve the precision of the density radiography. Nonetheless, the background muon flux rejection remains a key procedure for the structural imaging with muons. An important potential noise source, specially in the measurements based on the transmission and absorption muography, is the forward scattering of low energy muons on the object surface, reaching afterwards the detector. This effect mimics through-going particles since the reconstructed direction of the scattered particle points towards the target. The result is an increase of the total number of detected particles, as if the target’s opacity was lower than its actual value, leading to a systematic underestimation of the density [@nishiyama; @3d_souf]. Being produced by muons, these events can not be rejected by particle identification techniques, representing an irreducible background. For this reason, an evaluation of the magnitude of this effect is mandatory to conveniently correct the reconstructed object density.
In this work, a general evaluation of forward scattering of muons has been performed by Monte Carlo simulations. The aim was to develop a versatile tool to be able to evaluate this process for different object geometries and compositions, due to the increasing number of proposed applications based on muon tomography. The main features and results are presented in section \[sec:Scattering\]. Then the impact of this process on the muon imaging capability has been evaluated defining a signal to background parameter. Two physics cases have been studied in section \[sec:SBratio\]. The first one concerns an archaeological target, the Apollonia tumulus near Thessaloniki in Greece, and the second one *La Soufrière* volcano in Guadeloupe Islands of the Lesser Antilles. Finally, a summary of the different results and the main conclusions extracted from them are compiled in section \[sec:conclusions\].
Evaluation of the forward scattering of muons {#sec:Scattering}
=============================================
As mentioned in the introduction, low-energy cosmic muons can change their original direction after interacting with the target or any other object in the surroundings before their detection. As muon imaging is based on the reconstruction of the detected muons direction, these muons would forge the measurement. As a consequence, the determination of the target’s internal structure and the corresponding reconstructed mean density will be affected.
Muons trajectory deviation is mainly driven by their interaction with matter via multiple Coulomb scattering. The resulting deflection angular distribution, theoretically described by the Molière theory [@Bethe], roughly follows a Gaussian,
$$\frac{dN}{d\alpha} = \frac{1}{\sqrt{2\pi}\alpha_{MS}} e^{-\frac{\alpha^2}{2\alpha^{2}_{MS}}}
\label{eq:MoliereGauss}$$
which is centred in zero (i.e. no deflection happens), having a standard deviation $\alpha_{MS}$:
$$\alpha_{MS}=\frac{13.6 MeV}{\beta cp}Q\sqrt{\frac{x}{X_0}}\left (1 + 0.038 ln(x/X_0) \right )
\label{eq:MoliereAngle}$$
where $\beta$ is the relativistic factor, $p$ the muon momentum in MeV/$c$, $x$ the material thickness and $Q$ the absolute electric charge of the muon. $\alpha_{MS}$ also depends on the radiation length ($X_0$) which is empirically given by
$$X_0 \approx \frac{716.4 g/cm^2}{\rho} \frac{A}{Z(Z+1)log(287/\sqrt{(Z)})}
\label{eq:RadialtionL}$$
with $Z$ and $A$ the atomic and mass numbers respectively and $\rho$ the material density. This reveals the relationship of the multiple Coulomb scattering with the properties of the studied material.
Different works (see for example [@Schneider]) provide analytical solutions to the angular distribution of deflected muons after traversing an object with a determined geometry and composition. Besides, other relevant features, as the higher scattering probability for lower energy muons, are also demonstrated in these studies. However, the increasing number of different applications proposed for muon tomography, implies a large variety of objects dimensions, shapes and compositions, being less evident to obtain an analytical estimation of the forward scattering process suitable for all these cases. In this context, Monte Carlo simulations represent a useful tool for the study of muon scattering process, versatile enough to adapt them to the main features of each particular case. As first step on the development of these simulations, a general evaluation of the muons forward scattering has been carried using the Geant4 simulation tool-kit [@Geant4]. It allows the simulation of the 3D muon transport through the defined geometry taking into account the energy loss and trajectory variations due to multiple Coulomb scattering as well as to ionization, bremsstrahlung, pair production and multiple inelastic scattering. Considering these possible processes, results can be compared with the estimations given by the analytical formulas above-mentioned. A scheme of the simulated set-up is shown in figure \[General\_Diff\_Schema\_Img\]. For this case, generated muons are thrown to a fixed point on a standard rock surface (with a density of 2.5 g/cm$^{3}$). In the case of scattered muons, the direction changes, in zenith and/or azimuth angles, can be evaluated.
![Schema of the defined geometry to perform the general studies of forward scattering of muons.[]{data-label="General_Diff_Schema_Img"}](General_Diffusion_Schema.png){width="50.00000%"}
A first set of simulations were performed in order to evaluate the general features of the muon forward scattering. In the previously described set-up, muons up to 10 GeV, with a zenith incident angle ($\theta^{ini}_{det}$) between 70$^{\circ}$ and 90$^{\circ}$ and an azimuth incident angle $\varphi^{ini}_{det}$ = 0$^{\circ}$ were generated. It is worth mentioning that by the set-up definition $\theta^{ini}_{det}$ = 0$^{\circ}$ implies muons perpendicular to the rock surface, while $\theta^{ini}_{det}$ = 90$^{\circ}$ corresponds to tangential ones. Figure \[General\_Diff\_Img\] summarizes the results of this general simulation, leading to some conclusions about the muon forward scattering studied in these simulations. First, it is observed that this process is negligible if the muon energy is higher than 5 GeV, independently of the incident direction. For the lower energy muons, most of the “efficient” scattering processes (i.e. when the scattered particle exits the medium) occur if $\theta^{ini}_{det}$ is higher than 85$^{\circ}$ and do not exist if $\theta^{ini}_{det}$ is lower than 80$^{\circ}$. That means that only low energy muons with incident directions close to the surface tangent are likely to be scattered on the object surface and to induce a signal in the detector. For these muons the angular deviation can reach up to 25$^{\circ}$ both for the zenith and azimuth angles. By the simulation set-up definition, only the azimuth scattering angle ($\Delta\varphi_{det}$) has been registered for the whole angular range. As presented in figure \[General\_Diff\_Img\], the $\Delta\varphi_{det}$ distribution for all the muon energies considered is in agreement with the Gaussian predicted by Molière theory, as well as the other extracted conclusions agree with the analytical predictions [@PDG].
![Summary plots of the results of the general study of forward scattering of muons (see text for details about the study). Top left: Difference on the zenith angle ($\Delta\theta_{det}$ = $\theta^{fin}_{det}$ - $\theta^{ini}_{det}$) with respect to the initial muon energy ($E_{\mu}^{ini}$). Top right: Correlation between the initial and final zenith angles ($\theta^{ini}_{det}$ vs. $\theta^{fin}_{det}$) for all the muon energies considered. Bottom left: Difference on the azimuth angle ($\Delta\varphi_{det}$ = $\varphi^{fin}_{det}$ - $\varphi^{ini}_{det}$) with respect to the initial muon energy ($E_{\mu}^{ini}$). Bottom right: $\Delta\varphi_{det}$ distribution for all the muon energies considered. A Gaussian distribution as predicted by Molière theory (Equations \[eq:MoliereGauss\] - \[eq:RadialtionL\]) is observed.[]{data-label="General_Diff_Img"}](Diff_Gen_Theta_E.png "fig:"){width="45.00000%"} ![Summary plots of the results of the general study of forward scattering of muons (see text for details about the study). Top left: Difference on the zenith angle ($\Delta\theta_{det}$ = $\theta^{fin}_{det}$ - $\theta^{ini}_{det}$) with respect to the initial muon energy ($E_{\mu}^{ini}$). Top right: Correlation between the initial and final zenith angles ($\theta^{ini}_{det}$ vs. $\theta^{fin}_{det}$) for all the muon energies considered. Bottom left: Difference on the azimuth angle ($\Delta\varphi_{det}$ = $\varphi^{fin}_{det}$ - $\varphi^{ini}_{det}$) with respect to the initial muon energy ($E_{\mu}^{ini}$). Bottom right: $\Delta\varphi_{det}$ distribution for all the muon energies considered. A Gaussian distribution as predicted by Molière theory (Equations \[eq:MoliereGauss\] - \[eq:RadialtionL\]) is observed.[]{data-label="General_Diff_Img"}](Diff_Gen_Theta_Comp.png "fig:"){width="45.00000%"} ![Summary plots of the results of the general study of forward scattering of muons (see text for details about the study). Top left: Difference on the zenith angle ($\Delta\theta_{det}$ = $\theta^{fin}_{det}$ - $\theta^{ini}_{det}$) with respect to the initial muon energy ($E_{\mu}^{ini}$). Top right: Correlation between the initial and final zenith angles ($\theta^{ini}_{det}$ vs. $\theta^{fin}_{det}$) for all the muon energies considered. Bottom left: Difference on the azimuth angle ($\Delta\varphi_{det}$ = $\varphi^{fin}_{det}$ - $\varphi^{ini}_{det}$) with respect to the initial muon energy ($E_{\mu}^{ini}$). Bottom right: $\Delta\varphi_{det}$ distribution for all the muon energies considered. A Gaussian distribution as predicted by Molière theory (Equations \[eq:MoliereGauss\] - \[eq:RadialtionL\]) is observed.[]{data-label="General_Diff_Img"}](Diff_Gen_Phi_E.png "fig:"){width="45.00000%"} ![Summary plots of the results of the general study of forward scattering of muons (see text for details about the study). Top left: Difference on the zenith angle ($\Delta\theta_{det}$ = $\theta^{fin}_{det}$ - $\theta^{ini}_{det}$) with respect to the initial muon energy ($E_{\mu}^{ini}$). Top right: Correlation between the initial and final zenith angles ($\theta^{ini}_{det}$ vs. $\theta^{fin}_{det}$) for all the muon energies considered. Bottom left: Difference on the azimuth angle ($\Delta\varphi_{det}$ = $\varphi^{fin}_{det}$ - $\varphi^{ini}_{det}$) with respect to the initial muon energy ($E_{\mu}^{ini}$). Bottom right: $\Delta\varphi_{det}$ distribution for all the muon energies considered. A Gaussian distribution as predicted by Molière theory (Equations \[eq:MoliereGauss\] - \[eq:RadialtionL\]) is observed.[]{data-label="General_Diff_Img"}](MS_Phi_General_Diffusion.png "fig:"){width="45.00000%"}
Taking into account this general information and having checked the agreement between the general simulations and the analytical predictions, a more detailed simulation, optimizing the initial muon sampling was performed. The objective was to establish a probability density function (PDF) to further estimate the background due to forward scattered muons that could be detected during a muon imaging measurement and should be considered in the image analysis. With this aim 10$^{8}$ muons homogeneously distributed up to 5 GeV and with $\theta^{ini}_{det}$ between 85$^{\circ}$ and 90$^{\circ}$ (all with $\varphi^{ini}_{det}$ = 0$^{\circ}$) were generated and simulated in the described Geant4 framework. The generated PDF provides a probability value $\mathcal{P}$($\theta^{ini}_{det}$, $\theta^{fin}_{det}$, E$^{ini}_{\mu}$) depending on the initial and final zenith angle and the initial muon energy. A summary plot of the generated PDF divided in 0.5 GeV energy windows is shown in figure \[Diff\_PDF\_Ewindows\_Img\].
At this point it is worth mentioning that for the studies presented in this work (summarized in section \[sec:SBratio\]), the considered composition of the studied objects are the standard rock used to generate the PDF, but also a definition of soil with different composition and density than the rock ($\rho$ = 2.2 g/cm$^{3}$). Moreover, there exist several types of rocks and soils with different compositions and densities typically, between 2.0 and 2.5 g/cm$^{3}$. For this reason the influence of these two parameters in the PDF generation has been evaluated: a set of dedicated simulations have been performed changing the composition and the density of the target to compare their results. The obtained PDFs, including the standard soil case, agree to better than 97 %. Thus, the PDF presented in figure \[Diff\_PDF\_Ewindows\_Img\] has been used for all the studies.
![Correlation between the initial and final zenith angles ($\theta^{ini}_{det}$ vs. $\theta^{fin}_{det}$) from the general study of forward scattering of muons (see text for details about the study). The correlation plots are divided in 0.5 GeV windows between 0 and 5 GeV for the initial simulated muon energy ($E_{\mu}^{ini}$). These plots, correspond to 10$^{8}$ simulated muons with incident angles between 85$^{\circ}$ and 90$^{\circ}$ and energies between 0 and 5 GeV (both of them homogeneously distributed). They are used as PDF for further estimations on the forward scattered muon flux.[]{data-label="Diff_PDF_Ewindows_Img"}](PDF_Diff_Ewindows_20files.png){width="98.00000%"}
Signal to background ratio estimations {#sec:SBratio}
======================================
The impact of the forward scattered muons in an imaging measurement for a particular object can be evaluated based on simulations as those presented in section \[sec:Scattering\]. This impact can be expressed as a signal to background ratio (S/B) for a given direction $\theta_z$ - $\varphi_z$. These spherical coordinates correspond to those centred at the detector where $\theta_z$ = 0$^{\circ}$ is the vertical direction and $\varphi_z$ = 0$^{\circ}$ points to the main axis of the studied object. The signal S($\theta_z$, $\varphi_z$) is estimated as not scattered muon flux, so their reconstructed direction corresponds to their initial one. The background B($\theta_z$, $\varphi_z$) represents the scattered muons for which the reconstructed direction points towards the target.
As mentioned, these evaluations allow the study of a particular object, with its corresponding composition. For this it is necessary to know its external shape, to assume the object mean density (since this is the observable that can be extracted from a muon tomography measurement), and to determine the muon detector position with respect to this object. This allows the estimation of the object length traversed by muons for each direction as well as its surfaces positions with respect to the detector and to the Earth’s surface (required for the determination of $\theta_z$ and $\varphi_z$).
For this work two cases have been considered, corresponding to two applications of the muon imaging: the archaeology and the volcanology. For the first one a Macedonian tumulus located near Apollonia (Greece) has been studied [@gomez2016arche]. For the second, *La Soufrière* volcano (Guadeloupe island in the Lesser Antilles), already explored by muon imaging, has been taken as reference [@jourde2013experimental; @jourde_nature_2016].
Archaeology: Apollonia tumulus {#sec:Arche}
------------------------------
As quoted in section \[sec:Introduction\], the exploration of archaeological structures is one of the applications for which muon imaging has been proposed since it is non invasive and does not induce any harmful signals (contrary, for example, to vibrations used in seismic tomography). Already suggested in the 60s [@Alvarez], there exist at present different projects based on muon imaging devoted to the study of the internal structure of archaeological constructions (see for example [@ScanPyramids; @SPNature]). The ARCHé project proposes to scan the Apollonia Macedonian tumulus [@Arche]. These tumuli are man-made burial structures where the tomb, placed on the ground, is covered by soil, creating a mound which can also contain internal corridors. The geometry and dimensions of these tumuli are variable but they can always be approximated to a truncated cone. In the case of Apollonia tumulus its height is 17 m while the radius of the base and the top are 46 m and 16 m respectively. With this geometry the angle of the slope of the lateral surface of the tumulus is 29.5$^{\circ}$.
In the present study a standard soil composition, with a density of 2.2 g/cm$^{3}$, has been assumed. The detector has been placed 4 m beside the tumulus base (50 m from the tumulus base centre), so muons with zenith angles $\theta_z$ > 63.4$^{\circ}$ are those which will provide information about the structure of the tumulus. With all these features the signal and background, S($\theta_{z}$, $\varphi_{z}$) and B($\theta_{z}$, $\varphi_{z}$) respectively, can be estimated for a given direction $\theta_{z}$ - $\varphi_{z}$. As already described, these coordinates are centred at the detector and $\theta_{z}$ = 0$^{\circ}$ correspond to vertical muons, while $\varphi_{z}$ = 0$^{\circ}$ points towards the centre of the tumulus base.
From the knowledge of the external shape, it is possible to determine the length of tumulus traversed by muons for a given direction L($\theta_{z}$, $\varphi_{z}$) and thus, the corresponding opacity as the product of this length by the density ($\varrho$ = L $\times \rho$). The required minimal muon energy ($E_{min}$) to cross the target of opacity $\varrho$ can be calculated as
$$E_{min} = \frac{a}{b} \times \left (e^{b \times \varrho} - 1 \right )
\label{eq:Emin}$$
where $a(E)$ and $b(E)$ represent the energy loss coefficients due to ionization and radiative losses respectively. In this case, coefficients corresponding to standard rock summarized in [@PDG] have been used, obtaining E$_{min}$ values as a function of $\varrho$. As a cross-check, these E$_{min}$ values have been also estimated from the CSDA range values of standard rock [@PDG_muons]. The agreement between both methods is better than 95 %.
Hence, the expected signal S($\theta_{z}$, $\varphi_{z}$) corresponds to the muon flux on the studied direction with energies higher than E$_{min}$:
$$S (\theta_{z}, \varphi_{z}) = \int_{E = E_{min}}^{\infty} \phi_{\mu}(E, \theta_{z}, \varphi_{z}) dE
\label{eq:Signal}$$
To compute the background due to muons forward scattered in the same direction, B($\theta_{z}$, $\varphi_{z}$), two assumptions have been done. First, a point-like detector is considered. This implies that for each scattering point on the tumulus surfaces there is a unique final direction reaching the detector. Second, scattering effects in the azimuth angle are neglected. Since the general muon scattering studies (section \[sec:Scattering\]) show that these effects are symmetric and mostly below 5$^{\circ}$ for the azimuth angle (see figure \[General\_Diff\_Img\]), a low influence on the overall estimation is expected, having fade-out effects among the different azimuth directions.
With these two assumptions, B($\theta_{z}$, $\varphi_{z}$) corresponds to the product between the initial flux of muons which can be scattered by the corresponding probability to be scattered with a final zenith angle $\theta_{z}$. As already shown, only muons up to 5 GeV with an incident zenith angle higher than 85$^{\circ}$ with respect to the surface normal need to be considered for the forward scattering studies. This delimits the energy and zenith angle ranges to estimate the initial muon flux. The scattering PDF, $\mathcal{P}$($\theta^{ini}_{det}$, $\theta^{fin}_{det}$, E$^{ini}_{\mu}$), corresponds to the one presented in figure \[Diff\_PDF\_Ewindows\_Img\]. This PDF was generated using the coordinates $\theta_{det}$ - $\varphi_{det}$, centred in the scattering point and orthogonal to the surface. In order to be able to use this PDF with the $\theta_{z}$ - $\varphi_{z}$ coordinates, it is necessary to define the relationship between $\theta_{det}$ and $\theta_{z}$, which is given by $\theta_{det}$ = $\alpha$ + $\theta_{z}$. $\alpha$ represents the elevation angle of the scattering surface (that means, with respect to the Earth’s surface). $\theta_{det}$, $\theta_{z}$ and $\alpha$ angles are presented in figure \[Coordinates\_Img\]. For the case of $\varphi_{z}$ = 0$^{\circ}$, $\alpha$ corresponds to the slope of the lateral surface. For the cases where $\varphi_{z} \neq$ 0$^{\circ}$, it is estimated from the tangent to the tumulus surface at the scattering point. Thus, the expected background B($\theta_{z}$, $\varphi_{z}$) is calculated as:
$$B (\theta_{z}, \varphi_{z}) = \int_{E = 0}^{5} \int_{\theta = 85 - \alpha}^{90 - \alpha} \mathcal{P}(\theta , \alpha + \theta_{z}, E) \phi_{\mu}(E, \theta, \varphi_{z}) dE d\theta
\label{eq:Background}$$
For the different muon flux calculations required to obtain S($\theta_{z}$, $\varphi_{z}$) and B($\theta_{z}$, $\varphi_{z}$), the parametrization proposed in [@Shukla] has been used, corresponding to:
$$\begin{aligned}
\phi_{\mu}(\theta, E) = I_{0} (n-1) E_{0}^{n-1} (E_{0} + E)^{-n} \left (1 + \frac{E}{\epsilon}\right )^{-1} D({\theta})^{-(n-1)}
\label{eq:Shukla1}
\\
D(\theta) = \sqrt{\frac{R^{2}}{d^{2}} cos^{2}\theta + 2\frac{R}{d} + 1} - \frac{R}{d}cos\theta
\label{eq:Shukla2}\end{aligned}$$
where the experimental parameters, summarized in table \[Shukla\_Coeff\] together with other constants used in the equations, have been obtained from the fit of different experimental measurements. This parametrization provides an analytical formula for the muon flux estimation valid for low energy muons and high incident zenith angles.
![Schema showing the relationship between $\theta_{det}$, $\theta_{z}$ and $\alpha$ angles (see text for angles definition), for the use of the muon scattering PDF, $\mathcal{P}$($\theta^{ini}_{det}$, $\theta^{fin}_{det}$, E$^{ini}_{\mu}$), in the B($\theta_{z}$, $\varphi_{z}$) calculation.[]{data-label="Coordinates_Img"}](SB_Coordinates.png){width="70.00000%"}
Parameter Value
------------ --------------- ---------------------------
$E_{0}$ 4.28$\pm$0.05 GeV
$\epsilon$ 854 GeV
$I_{0}$ 88$\pm$2.4 m$^{-2}$s$^{-1}$sr$^{-1}$
$n$ 3.09$\pm$0.03
$R/d$ 174$\pm$12
: Values of the parameters and constants used for the estimation of the muon flux based on equations \[eq:Shukla1\] and \[eq:Shukla2\].[]{data-label="Shukla_Coeff"}
With all these ingredients the S/B ratio for the Apollonia tumulus has been calculated scanning the $\varphi_{z}$ range in 10$^{\circ}$ steps and the corresponding $\theta_{z}$ values for each case in 1$^{\circ}$ steps. The results from these calculations are summarized in figure \[Apollonia\_SB\_Img\] as a function of $\theta_{z}$ and the opacity $\varrho$, which is a more significant variable than $\varphi_{z}$ since the muon flux is basically independent of the azimuth angle. The main conclusion is that for all the studied directions the S/B ratio is higher than 73.9, which means that at most the 1.3 % of the detected muons have been previously scattered on the object surface. It is observed that the directions with the lowest S/B values are those with high $\theta_{z}$ values. For these directions lower values for the signal are expected since they correspond to the most horizontal ones (where the muon flux is lower) and, due to the tumulus geometry, these are cases for which longer tumulus length is traversed. Actually, for directions with $\theta_{z}$ lower than 85$^{\circ}$, the S/B ratio is always higher than 254.8, reducing the contribution of the scattered muons to the total detected to less than 0.4 %. In this region, the obtained S/B values can be considered homogeneous. Differences between directions are basically associated to the uncertainties in the muon scattering PDF. As mentioned in section \[sec:Scattering\], even if the used PDF was generated with another target material than the assumed tumulus composition, it would have a limited effect on the results. This leads to consider that the forward scattered muons on the object surface do not significantly influence the results of the muon imaging for the case of tumuli and, by extension, of other objects with similar dimensions and composition.
![Distribution of the ratio between the non-scattered and scattered detected muons (defined as S/B in the text) with respect to the reconstructed zenith incident angle ($\theta_{z}$) and the opacity ($\varrho$) for the Apollonia tumulus case.[]{data-label="Apollonia_SB_Img"}](histo_SB_Theo_Apollonia_20files_phi_scan.png){width="90.00000%"}
Volcanology: La Soufrière {#sec:Volcanology}
-------------------------
The use of muon imaging for the scanning of volcanoes is another application of this technique, which implies the study of objects with larger dimensions than for archaeology. With this purpose, some projects have already performed measurements in different locations. One of them is the DIAPHANE collaboration [@diaphane], which surveys *La Soufrière* volcano paying special attention to possible variations of the inner liquid/vapour content that can be related to the hydrothermal system dynamics. For this work, *La Soufrière* volcano has been taken as reference to study the impact of the forward scattered muons on the muon imaging reconstruction in volcanology.
As for the case of tumuli, volcanoes geometry can also be approximated to a truncated cone. Based on the topographic plan of *La Soufrière*, their dimensions correspond to a height of 460 m and a base and top radius of 840 m and 160 m respectively. These dimensions lead to a lateral surface with a slope angle of 34.1$^\circ$. In this case a homogeneous composition of standard rock has been considered, with a density of $\rho$ = 2.5 g/cm$^3$, together with 3 different detector positions corresponding to real measurement points of the DIAPHANE project. They are labelled as h-270, h-170 and h-160 respectively because of the height where they are placed. These positions are summarized in table \[Soufriere\_DetPos\_Tab\] taking as reference the centre of the volcano base. Main differences among the positions rely on the distance between the detector and the volcano, going from 5 to 25 m approximately, and on the height with respect to the volcano base, which has a direct influence on the length of the volcano traversed by muons before their detection and, consequently, on the signal S($\theta_z$,$\varphi_z$) computation.
------- --------- ---------- ------- ------------------------
Distance
x (m) y (m) z (m) Detector - Volcano (m)
h-270 384.86 -242.63 270 14.09
h-170 -348.92 -482.49 170 6.72
h-160 189.21 -598.26 160 23.99
------- --------- ---------- ------- ------------------------
: Summary of the detector positions with respect to the base centre of *La Soufrière* volcano model (see text for model details).[]{data-label="Soufriere_DetPos_Tab"}
Both tumulus and volcano have been approximated to the same geometrical shape with their corresponding dimensions. So for volcanoes, the procedure to determine the S/B ratio for different incident directions is equivalent to the described in section \[sec:Arche\] for the tumulus case. The only difference is that for this case the assumed density is $\rho$ = 2.5 g/cm$^3$ instead of $\rho$ = 2.2 g/cm$^3$ (corresponding to the standard soil), affecting in the opacity estimation. Nevertheless, this density variation is expected to have a reduced impact on the results as it has been estimated in section \[sec:Scattering\].
As for the tumulus case, the S/B ratio have been evaluated scanning the $\varphi_{z}$ range in 10$^{\circ}$ steps and the corresponding $\theta_{z}$ values for each case in 1$^{\circ}$ steps. Results have also been represented with respect to $\theta_{z}$ and the opacity $\varrho$. They have been summarized for the three different detector positions in figure \[Soufriere\_SB\_Img\]. For the three cases the S/B ratio takes values significantly lower than for the tumulus, although the corresponding distributions present similar features. For example, the S/B values for directions with $\theta_{z}$ > 85$^\circ$ are again lower than for the rest of the directions. Moreover, for all detector positions, directions with low opacity (corresponding to the volcano contour) present systematically higher values of S/B than those directions pointing to the bulk of the volcano.
![Distribution of the ratio between the non-scattered and scattered detected muons (defined as S/B in the text) with respect to the reconstructed zenith incident angle ($\theta_{z}$) and the opacity ($\varrho$). The distribution is showed for the 3 detectors installed in *La Soufrière* volcano. Numbers correspond to the bin value and have been placed next to the corresponding bin to ease their reading.[]{data-label="Soufriere_SB_Img"}](histo_SB_Theo_Volcano_BR_20files_phi_scan.png "fig:"){width="90.00000%"} ![Distribution of the ratio between the non-scattered and scattered detected muons (defined as S/B in the text) with respect to the reconstructed zenith incident angle ($\theta_{z}$) and the opacity ($\varrho$). The distribution is showed for the 3 detectors installed in *La Soufrière* volcano. Numbers correspond to the bin value and have been placed next to the corresponding bin to ease their reading.[]{data-label="Soufriere_SB_Img"}](histo_SB_Theo_Volcano_SB_10files_phi_scan.png "fig:"){width="90.00000%"} ![Distribution of the ratio between the non-scattered and scattered detected muons (defined as S/B in the text) with respect to the reconstructed zenith incident angle ($\theta_{z}$) and the opacity ($\varrho$). The distribution is showed for the 3 detectors installed in *La Soufrière* volcano. Numbers correspond to the bin value and have been placed next to the corresponding bin to ease their reading.[]{data-label="Soufriere_SB_Img"}](histo_SB_Theo_Volcano_MT_10files_phi_scan.png "fig:"){width="90.00000%"}
Focused on each detector, for the h-270 case, incident directions with high $\theta_{z}$ have S/B ratio values in general below 1, which implies that it is possible to detect more forward scattered muons than those emerging from the volcano, significantly influencing the object density reconstruction. On the opposite side, for the directions where the opacity is smaller than 50$\times$10$^3$ g/cm$^2$, S/B ratio takes values higher than 5, so no more than the 17 % of the detected muons have been previously scattered. If we consider all the other directions, with $\theta_{z}$ < 85$^\circ$ and $\varrho$ > 50 $\times$10$^3$ g/cm$^2$, the S/B distribution is more homogeneous, having a mean value of 2.7 with a standard deviation of 1.4. That means that in average about 27 % of the detected muons are low energy forward scattered muons. In this case, the scattered muons have a significant impact on the volcano density reconstruction. Assuming the percentage of scattered muons constant for all the scanned directions, and estimating the uncertainty of this percentage from the standard deviation of the S/B mean value, it would imply that the reconstructed density should be corrected by multiplying it by a factor 1.4$^{+0.4}_{-0.1}$.
Results for the detector positions labelled as h-170 and h-160 are similar between them. This suggests that S/B ratio has more dependence with the detector height with respect to the volcano base (170 m for h-170 and 160 m for h-160) than with the distance between the detector and the volcano (6.72 m for h-170 and 23.99 m for h-160). Since both detectors are placed lower than the h-270 case, the mean volcano length traversed by non-scattered muons is longer in this case. This leads to smaller S/B ratios mainly because lower S($\theta_{z}$, $\varphi_{z}$) values. As mentioned, the features of the distribution are equivalent: smaller S/B values for $\theta_{z}$ > 85$^\circ$ and higher for low opacities. If only muon directions with $\theta_{z}$ < 85$^\circ$ and $\varrho$ > 50 $\times$10$^3$ g/cm$^2$ are considered, a mean S/B value of 1.1 is obtained for both cases with a standard deviation of 0.9 and 1.0 for the h-170 and h-160 detector positions respectively. These values reveal a high influence of the low energy forward scattered muons in the overall detection (almost half of the detected muons have been previously scattered). Keeping the assumption of a constant S/B value for all the considered directions, the density correction factors are 1.9$^{+4.1}_{-0.4}$ for the h-170 position and 1.9$^{+9.1}_{-0.4}$ for the h-160 case.
Summarizing, for the case of volcanoes, where the length of material to be traversed by muons is longer than for archaeology, the forward scattering of low energy muons and their further detection has a clear influence on the results of the muon imaging. The three studied scenarios and the defined geometry of the volcano as a truncated cone, reveal that the S/B ratio mainly depends on the length of material traversed by non-scattered muons, those considered for S($\theta_z$,$\varphi_z$) computation. Moreover, for a fixed detector position it can be considered S/B as homogeneous for all the incident directions corresponding to the volcano bulk volume, so a global correction factor for the reconstructed density can be applied. The main source of uncertainty of the S/B ratio estimation comes from those associated to the PDF and consequently, B($\theta_{z}$, $\varphi_{z}$). For this reason, as deduced for the h-270 detection position, a higher S/B mean value translates to a more accurate determination of the correction factor for the reconstructed volcano density.
Summary and discussion {#sec:conclusions}
======================
At present, muon imaging is being used and proposed for an increasing number of different applications. This implies that objects with quite different dimensions can be scanned, from some tens to several hundreds of meters as typical sizes. Furthermore, the composition and density of these objects can also vary from one to other. All the experimental approaches to do muon imaging, generally known as transmission and deviation tomography respectively, rely on the direction reconstruction of the detected muons. For this reason, specially for the transmission-based technique, muons changing their direction because of a scattering on the object surface before their detection, represent an irreducible background noise, biasing the object mean density reconstruction. An estimation of the percentage of these forward scattered muons out of all detected, would allow the estimation of correction factors to reconstruct the proper density.
Muons trajectory deviation is mainly driven by the multiple Coulomb scattering. The resulting angular distribution due to this effect is theoretically described by the Molière theory. Besides, some analytical descriptions of the process have been already performed for particular objects and compositions. Nevertheless, the large variety of objects that are currently proposed to be studied by muon tomography, requires more versatile tools to evaluate the forward scattering muons, easily adaptable to each case. With this aim a set of Monte Carlo simulations have been performed using the Geant4 framework.
These simulations provide a general evaluation of the muon forward scattering probability depending on their energy and their incident angle, being in overall agreement with theoretical estimations. They revealed that muons with energies lower than 5 GeV and incident angles above 85$^\circ$ with respect to the normal direction of the surface, are almost the only muons susceptible to be scattered and then detected. The simulations results have been used as PDF to evaluate the influence of scattered muons in different scenarios. To do that, the signal to background ratio (S/B) has been defined. S($\theta_z$,$\varphi_z$) corresponds to the flux of those muons reconstructed in a direction $\theta_z$-$\varphi_z$ without any previous scattering, while B($\theta_z$,$\varphi_z$) is the muon flux of muons reconstructed with the same direction but after a previous forward scattering on the object surface. The S/B evaluation has been presented for two particular cases, corresponding to two of the applications of muon imaging: archaeology and volcanology.
Taking the muon distribution at Earth’s surface proposed in [@Shukla], for the archaeological applications, the Apollonia tumulus has been considered as reference, placing the detector beside the tumulus base. S/B estimations reveals that the percentage of scattered muons detected is never higher than 1.6 %, being lower than 0.5 % if the incident zenith angle is smaller than 85$^\circ$. This leads to conclude that the influence of scattered muons for these cases can be neglected.
This is not the case for volcanology applications. A model based on *La Soufrière* volcano, already scanned inside the DIAPHANE project, has been used together with three different detector positions, corresponding to real measurement points. It has been observed a significant influence of the forward scattered muons in the measurement, which can represent up to 50 % of the detected muons, and even more for incident zenith angles higher than 85$^\circ$. S/B values can be considered homogeneous for the directions corresponding to the bulk volume of the volcano. Main differences on S/B mainly depend on the height of the detector with respect to the volcano base. Due to the volcano geometry, defined as a truncated cone, this is directly related to the muon path length along the volcano. Other features, such as the distance between the detector and the volcano, seem to have smaller influence. With the estimations and numbers obtained in this work, correction factors for the density reconstruction have been computed, taking values from 1.4 to 1.9 depending on the detector position.
Forward scattered muons represent events that are in principle not taken into account, so their detection has a direct impact on the mean density reconstruction. Nonetheless, the observed homogeneity on the S/B ratio for all the considered directions, both in the tumulus and volcano case, leads to think that these muons would not significantly affect fading the resolution of the resulting image.
All these estimations and conclusions are based on simulations of scattered muons on standard rock, which has been demonstrated to produce equivalent results than the standard soil case. In any case, changing accordingly the material composition and properties, this simulation framework can be used to evaluate the influence of forward scattered muons for further muon imaging measurements of other objects and structures.
Authors would like to acknowledge the financial support from the UnivEarthS Labex program of Sorbonne Paris Cité (ANR-10-LABX-0023 and ANR-11-IDEX-0005-02). Data from *La Soufrière* volcano are part of the ANR DIAPHANE project ANR-14-ce04-0001. Part of the project has been funded by the INSU/IN2P3 TelluS and “DEFI Instrumentation aux limites” programmes.
[99]{}
S.H. Neddermeyer, C.D.Anderson. *Note on the nature of cosmic-ray particles*, *Phys. Rev.* **51** (1937) 884
S.H. Neddermeyer, C.D.Anderson. *Cosmic-ray particles of inter-mediate mass*, *Phys. Rev.* **54** (1938) 88
P. Auger. *Les rayons cosmiques*, *PUF, Paris* (1941) 136
K. Nagamine. *Introductory Muon Science*, *Cambridge University Press, Cambridge* (2003)
N. Lesparre et al. *Geophysical muon imaging: feasibility and limits*, *Geophysical Journal International* **183** (2010) 1348
J. Marteau et al. *Muons tomography applied to geosciences, volcanology* *Nucl. Instrum. Meth. A* **695** (2012) 23
K.N. Borozdin et al. *Radiographic imaging with cosmic-ray muons* *Nature* **422** (203) 277
S. Procureur. *Muon imaging: Principles, technologies and applications* *Nucl. Instrum. Meth. A* <http://dx.doi.org/10.1016/j.nima.2017.08.004>
E.P. George. *Cosmic rays measure overburden of tunnel*, *Commonwealth Engineer* **455** (1955)
L.W. Alvarez. *Search for hidden chambers in the pyramids using cosmic rays*, *Science* **167** (1970) 832
K. Morishima et al. *Discovery of a big void in Khufu’s Pyramid by observation of cosmic-ray muons*, *Nature* <http://dx.doi.org/10.1038/nature24647> (2017)
L. Malmqvist et al. *Theoretical studies of in-situ rock density determination using cosmic-ray muon intensity measurements with application in mining geophysics*, *Geophysics* **44** (1979) 1549
N. Kume et al. *Muon trackers for imaging a nuclear reactor*, *J of Instr.* **11** (2016) P09008
R. Nishiyama et al. *Monte Carlo simulation for background study of geophysical inspection with cosmic-ray muons*, *Geophys. J. Int.* **206** (2016) 1039
M. Rosas-Carbajal et al. *Three-dimensional density structure of La Soufri[è]{}re de Guadeloupe lava dome from simultaneous muon radiographies and gravity data*, *Geophysical Research Letters* **44** (2017) 6743
H.A. Bethe. *Molière’s Theory of Multiple Scattering*, *Physical Review* **89** (1953) 1256
U. Schenider et al. *Coulomb scattering and spatial resolution in proton radiography*, *Medical Physics* **21** (1994) 1657
S. Agostinelli et al. (GEANT4 collaboration), *GEANT4: A Simulation toolkit*, *Nucl. Instrum. Meth. A* **506** (2003) 250, *IEEE Trans. Nucl. Sci.* **53** (2006) 270\
J. Allison et al. *Recent developments in GEANT4*, *Nucl. Instrum. Meth. A* **835** (2016) 186
C. Patrignani et al. (Particle Data Group), *Review on Particle Physics*, *Chin. Phys. C* **40** (2016) 100001
H. Gómez et al. *Studies on muon tomography for archaeological internal structures scanning*, *Journal of Physics: Conference Series*, **718(5)** (2016) 052016
K. Jourde et al. *Experimental detection of upward going cosmic particles and consequences for correction of density radiography of volcanoes*, [*Geophysical Research Letters*]{} **40(24)** (2013) 6334
K. Jourde et al. *Muon dynamic radiography of density changes induced by hydrothermal activity at the La Soufrière of Guadeloupe volcano*, [*Scientific Reports*]{} **6** (2016) 33406
ScanPyramids project, [www.scanpyramids.org](www.scanpyramids.org)
H. Gómez et al. *Muon imaging for archaeological applications: feasibility studies of the Apollonia Macedonian Tumulus*, *(In preparation)*
D.E. Groom et al. *Muon stopping-power and range tables*, *Atomic Data and Nuclear Data Tables* **78** (2001) 183
P. Shukla. *Energy and angular distributions of atmospheric muons at the Earth*, arXiv:1606.06907 \[hep-ph\]
Diaphane project, [www.diaphane-muons.com](www.diaphane-muons.com)
| {
"pile_set_name": "ArXiv"
} |
Introduction
============
One of the authors, S. Y., proposed[@ref1; @ref2] the direct electric energy conversion of the heat from plasma by the Nernst effect in a fusion reactor, where a strong magnetic field is used to confine a high temperature fusion plasma. He called[@ref1; @ref2] the element which induces the electric field in the presence of temperature gradient and magnetic field, as Nernst element. In his paper[@ref1; @ref2], he also estimated the figure of merit of the Nernst element in a semiconductor model. In his result[@ref1; @ref2], the Nernst element has high performance in low temperature region, that is, 300 - 500 K. Before his works, the Nernst element was studied in the 1960’s[@ref3]. In those day, induction of the magnetic field had a lot of loss of energy. This is the reason why the Nernst element cannot be used. Nowadays an improvement on superconducting magnet gives us higher efficiency of the induction of the strong magnetic field. We started a measuring system of transport coefficients in the strong magnetic field to estimate efficiency of the Nernst element on a few years ago[@ref4]. As the first candidate of the Nernst element, we choose InSb, which is expected to have the high figure of merit according to the single-band model[@ref5]. The experiment results show that the Nernst coefficient is smaller than the theoretical values. On the other hand, the conductivity, the Hall coefficient and the thermoelectric power has the values expected by the theory. In this paper, we introduce the experimental results and compare the theoretical calculations.
Experiment
==========
Choice of material
------------------
We discuss the principle of transport phenomena in a magnetic field and a temperature gradient. This behavior is written by two phenomenological equations[@ref6] as follows: $$\begin{aligned}
{\mbox {\boldmath $E$} } &=& \frac{ \mbox{\boldmath $J$} }{ \sigma}
+ \alpha \cdot \mbox{\boldmath $\nabla$ } T
+ R_{\rm H} \mbox{\boldmath $B$}
\times \mbox{\boldmath $J$} , \label{eq.1} \\
{\mbox {\boldmath $q$} } &=& \alpha T \mbox{\boldmath $J$}
- \kappa \mbox{\boldmath $\nabla$ } T
+ N T \mbox{\boldmath $B$}
\times \mbox{\boldmath $J$}
+ L \mbox{\boldmath $B$ }
\times \mbox{\boldmath $\nabla$} T, \label{eq.2}\end{aligned}$$ where [$E$]{} is electrical field, [$J$]{} current density, [$B$]{} magnetic field, $T$ temperature, $\sigma$ electrical conductivity, $\alpha$ thermoelectric power, $R_{\rm H}$ Hall coefficient, $N$ Nernst coefficient, $\kappa$ thermal conductivity and $L$ Righi-Leduc coefficient. The Nernst element uses the last term of eq.(\[eq.1\]).
To simplify the discussion of the efficiency, we replace all transport coefficients by averaged quantities, which do not depend on position within a device. In order to estimate efficiency of the Nernst element, it is useful to define the figure of merit $Z_{\rm N}$ as follows[@ref3]: $$Z_{\rm N} \equiv \frac{\sigma B^2 N^2}{\kappa}. \label{eq.3}$$ Using eq.(1), the optimal efficiency of thermomagnetic generators $\varepsilon_{\rm N}$ is given by[@ref3] $$\varepsilon_{\rm N} = \varepsilon_{\rm C}
\left(
\frac{ 1-\delta^{\ast}_{\rm N}
}{ 1 + \frac{ { T_{\rm low} }}{ { T_{\rm high} }}
\delta^{\ast}_{\rm N} }
\right), \label{eq.4}$$ where ${ T_{\rm high} }({ T_{\rm low} })$ is the temperatures of the heating (cooling) block, $\varepsilon_{\rm C}$ the carnot efficiency, $({ T_{\rm high} }- { T_{\rm low} }) / { T_{\rm high} },$ and $$\delta^{\ast}_{\rm N} =
\sqrt{ 1 - Z_{\rm N} \left( \frac{{ T_{\rm low} }+{ T_{\rm high} }}{2}\right)}. \label{eq.5}$$ The value of $\delta^{\ast}_{\rm N}$ must be a real number. This fact impose the following restriction as[@ref3]: $$Z_{\rm N} \left( \frac{{ T_{\rm low} }+{ T_{\rm high} }}{2} \right) \leq 1. \label{eq.6}$$ We plot the normalized efficiency, $\varepsilon_{\rm N}/\varepsilon_{\rm C}$ in Fig. \[fig.1\] as a function of the figure of merit. This figure shows that $\varepsilon_{\rm N}$ increases monotonously as $Z_{\rm N}$ becomes larger. We, therefore, must choose the high-$Z_{\rm N}$ materials. We consider the transport coefficients to choose them. It is known from the Boltzmann equation that both conductivity and Nernst coefficient are proportional to Hall mobility[@ref7; @ref8]. This fact derives the form[@ref1; @ref4]: $$Z_{\rm N} \propto \mu^3.
\label{eq.7}$$ The equation (\[eq.7\]) is a criterion for searching the Nernst element. Under this criterion, we first propose indium antimonide, InSb as a candidate of the Nernst element. To compare the mobility of InSb with the other materials, we summarize the values of the mobilities in Table \[tab.1\].
Measurement of transport coefficients and results
-------------------------------------------------
### Conductivity and Hall coefficient
The carrier concentration of the InSb crystals investigated is $6.6 \times 10^{20}$ \[ m$^{-3}$\] and its mobility 21 \[m$^{2}$/V/s\] at 77K. This sample exhibited intrinsic behavior near room temperature. Copper wires with 50 $\mu$m-diameter are spark-bonded onto a crystal by using a capacitor discharge. Chromel-Alumel thermocouples, 0.5 mm in diameter, are contact to heating and cooling units with silver epoxy. The temperature of the sample is controlled within 270-370K by the heat bath, the water temperature of which is kept a constant. We induced a strong magnetic field up to 4 Tesla by the superconducting coil to measure magnetoresistance of the sample. Analog signals of the thermocouples and voltage source are amplified and converted to digital data. The personal computer acquires these data and draws figures in real time. We use the van der Pauw method[@ref9] to measure the conductivity and Hall coefficients. A geometry for the van der Pauw method is shown in Fig. \[fig.2\]. Figure Fig. \[fig.3\](a) shows the temperature dependence of the resistivity at B=0 Tesla. The temperature dependence of the Hall coefficient is represented in Fig. \[fig.3\](b).
### Thermoelectric power and Nernst coefficient
The sample of measurement of the thermoelectric power and the Nernst element is the same material as the van der Pauw method. However, the shape of the sample is changed from the square to the bridged shape (Fig. \[fig.4\]).
In order to make temperature gradient in the sample, we used thermofoil heater for a heating copper block side, the water temperature of which is controlled by a low temperature incubator, for a cooling one. Using the heating and cooling units, the thermal difference across the sample was within 10-100K. The thermoelectric voltage, $V_{\rm \alpha}$ and the Nernst one, $V_{\rm N}$ have the following relations between the thermoelectric power and the temperature gradient as $$\begin{aligned}
V_{\rm \alpha} &=& L \alpha | \mbox{\boldmath $\nabla $ } T |
\approx \alpha \Delta T, \label{eq.8} \\
V_{\rm N} &=& w N B | \mbox{\boldmath $\nabla $ } T |
\approx \frac{w N B \Delta T }{L}, \label{eq.9} \end{aligned}$$ where $\Delta T$ is $({ T_{\rm high} }- { T_{\rm low} }),$ $w$ the width of the sample and $L$ the length defined by Fig. \[fig.4\]. Here we define the following physical quantity, $\beta$ to compare the thermoelectric power and the Nernst effect: $$\beta \equiv NB, \label{eq.10}$$ which has the same dimension, \[V/K\] as $\alpha.$ The results of the measurement of $\alpha$ and $\beta$ in Figs. \[fig.5\]–\[fig.6\]. The thermoelectric power doesn’t change very much as the magnetic field is induced. On the other hand, the $\beta$ depends on the magnetic field very much. In Fig. \[fig.6\](a), we plot the results as the crosses and the theoretical values as the filled circles. The theoretical values are explained in the later. The difference between the experimental results and the theoretical ones is the order of 10. For the strong magnetic field, the results are shown in Fig. \[fig.6\](b).
Analysis and physical quantities
================================
Carrier concentration
---------------------
In the weak field limit, the Hall coefficient and the carrier concentration has the form[@ref8] $$R_{\rm H } = \frac{3 \pi}{ 8 |e| } \frac{ p-nb^2 }{ \left( p+nb\right)^2 }
\approx - \frac{3 \pi}{8 |e| n} , \label{eq.11}$$ where we define $b = \mu_{\rm n} / \mu_{\rm p}$ and the hole parts are neglected because $b\approx 100$ for InSb[@ref10; @ref11]. Equation (\[eq.11\]) and Fig. \[fig.3\](b) gives the carrier concentration of the electron in Fig. \[fig.7\]. We can fit the following function of the temperature by the least square method as $$n(T) = 3.3 \times 10^{20} T^{1.5} \exp \left( - \frac{2600}{k T}\right). \label{eq.12}$$
We assume that the sample is in the intrinsic region near room temperature. The carrier concentration of the intrinsic semiconductor is written by[@ref8] $$n_{\rm i} (T) = 2\left( \frac{ \sqrt{m_{\rm n} m_{\rm p} } k T}{2\pi \hbar^2 } \right)^{\frac{3}{2} }
\exp \left( - \frac{ E_{\rm G}}{2kT} \right), \label{eq.13}$$ where $E_{\rm G}$ is the energy gap, $m_{\rm n}$ the effective mass of the electron, and $m_{\rm p}$ the effective mass of the hole. Comparing eq. (\[eq.12\]) with eq. (\[eq.13\]), we obtain $$\begin{aligned}
E_{\rm G} &\approx& 2600 \times \frac{k}{|e|} = 0.22 \mbox{ [eV] }, \label{eq.14} \\
m_{\rm n} m_{\rm p} &\approx& \left\{
\left( \frac{3.3\times 10^{20}}{2} \right)^{2/3}
\frac{2\pi \hbar^2}{k} \right\}^2 \approx 1.7\times 10^{-2} m_0^2. \label{eq.15}\end{aligned}$$
Mobility of electron
--------------------
In the weak field limit, the Boltzmann equation for the single parabolic non-degenerated band model gives the conductivity as follows[@ref8]: $$\sigma = 1/\rho = |e| n \mu_{\rm n} \left( 1+ \frac{p}{nb} \right)
\approx |e| n \mu_{\rm n}, \label{eq.16}$$ where we use $b \ll 1$ for InSb. Equations (\[eq.11\]) and (\[eq.16\]) give the mobility of the electron in Fig. \[fig.8\]. The temperature dependence of the mobility of the electron is obtained as $$\mu_{\rm n} \approx 7.5 \times \left( \frac{T}{300} \right)^{-1.50}
\mbox{[m$^2$/V/s]}. \label{eq.17}$$ The exponent -1.5 denotes that the dominant scattering process is acoustic phonon scattering.
Mobility of hole
----------------
In the strong magnetic field limit, the Nernst coefficient of the intrinsic semiconductor becomes[@ref10] $$N \approx \frac{k}{|e|} \mu_{\rm p} \left( 4 +\frac{E_{\rm G} }{kT} \right). \label{eq.18}$$ Substituting the values of the Nernst coefficient given by the experiment at $B =4$ Tesla in the eq. (\[eq.18\]), we obtain the mobility of the hole in Fig. \[fig.9\]. By the least square method, we also have the temperature dependence of the hole mobility as $$\mu_{\rm p} (B=4\mbox{Tesla}) \approx 0.065\times
\left( \frac{T}{300} \right)^{-1.7} \mbox{[m$^2$/V/s]} . \label{eq.19}$$
Fermi level
-----------
The thermoelectric power gives the Fermi level as follows[@ref8]: $$\begin{aligned}
\alpha &=& \alpha_{\rm n} \frac{\sigma_{\rm n} }{ \sigma_{\rm n} + \sigma_{\rm p} }
+ \alpha_{\rm p} \frac{\sigma_{\rm p} }{\sigma_{\rm n} + \sigma_{\rm p} } \nonumber \\
&=& \alpha_{\rm n} \left( \frac{1}{ 1+\frac{p}{nb} } \right)
+ \alpha_{\rm p} \left( \frac{ \frac{p}{nb} }{ 1+\frac{p}{nb} } \right) \nonumber \\
&\approx& \alpha_{\rm n } = - \frac{k}{ |e| }
\left( 2 - \frac{ \zeta_{\rm n} }{kT} \right), \label{eq.20}\end{aligned}$$ where is the Fermi level from the edge of the conduction band and negative. Equation (\[eq.20\]) and the experimental results of the thermoelectric power give the Fermi level in Fig. \[fig.10\]. In the intrinsic region, the Fermi level becomes[@ref8] $$\zeta_{\rm n} = - \frac{E_{\rm G} }{2} + \frac{3kT}{4|e|} \ln
\left( \frac{m_{\rm p}}{m_{\rm n} } \right) \ \ \ \mbox{[eV]}. \label{eq.21}$$ Analysis in Fig. \[fig.10\] gives the temperature dependence of the Fermi level as follows $$\zeta_{\rm n} = - 0.117 + 3.2 \frac{kT}{|e|} \ \ \ \mbox{[eV]}. \label{eq.22}$$ From eqs. (\[eq.21\]) and (\[eq.22\]), the energy gap and the ratio of the effective masses of the electron and the hole is given as $$E_{\rm G} \approx 0.23 \ \ \ \mbox{ [ eV ] },
\ \ \frac{m_{\rm p} }{m_{\rm n} } \approx 73. \label{eq.24}$$
Discussion and conclusions
==========================
We summarize basic physical quantities obtained by the experiment in Table \[tab.2\], where the reference values are also written. This table shows that the experimental results are almost coincident with the previous works. It is concluded that the Hall coefficient, the conductivity and the thermoelectric power of InSb near room temperature in the weak field are given the Boltzmann equation for the non-degenerate parabolic two-band with the acoustic phonon scattering. However, the Nernst coefficient is very smaller than the theoretical value in the weak field. The behavior of the Nernst coefficient in the strong magnetic field is consistent with the two-band model. We try to explain the difference between the experimental and the theoretical values of the Nernst coefficient in the weak field. Moreover we will measure the thermal conductivity in the magnetic field to estimate the figure of merit.
The authors are grateful to Dr. Tatsumi in Sumitomo Electric Industries and Prof. Kuroda in Nagoya university for providing semiconductors. We appreciate Prof. Iiyoshi in the National Institute for Fusion Science for his helpful comments.
S. Yamaguchi, A. Iiyoshi, O. Motojima, M. Okamoto, S. Sudo, M. Ohnishi, M. Onozuka and C. Uesono, [*“Direct Energy Conversion of Radiation Energy in Fusion Energy, Proc. of 7th Int. Conf. Merging Nucl. Energy Systems", Proc. of 7th Int. Conf. Merging Nucl. Energy Systems*]{} (ICENES)“, (1994) 502. S. Yamaguchi, K. Ikeda, H. Nakamura and K. Kuroda, [*“A Nuclear Fusion Study of Thermoelectric Conversion in Magnetic field", 4th Int. Sympo. on Fusion Nuclear Tech.*]{}, ND-P25, Tokyo, Japan, April (1997). T. C. Harman and J. M. Honig, [*“Thermoelectric and Thermomagnetic Effects and Applications",*]{} McGraw-Hill Book Company, (1967), Chap. 7, p. 311. K. Ikeda, H. Nakamura, S. Yamaguchi and K. Kuroda, ”Measurement of Transport properties of Thermoelectric Materials in the Magnetic Field", J. Adv. Sci., [**8**]{}, (1996) 147, (in Japanese). H. Nakamura, K. Ikeda, S. Yamaguchi and K. Kuroda, “Transport Coefficients of Thermoelectric Semiconductor InSb", J. Adv. Sci., [**8**]{}, (1996) 153, (in Japanese). L. D. Landau, E. M. Lifshitz and L. P. Petaevskii, [*“Electrodynamics of Continuous Media",*]{} 2nd Edition, Pergamon Press, (1984)101. E. H. Putly, [*“The Hall Effect and Related Phenomena",*]{} London, Butterworth &Co. Ltd., (1960), Chap. 3, p. 66. K. Seeger, [*“Semiconductor Physics"*]{}, Springer-Verlag, (1989). L. J. van der Pauw, “A Method of Measuring Specific Resistivity and Hall Effect of Discs of arbitrary Shape", Philips Res. Rep., [**13**]{}, (1958) 1. G. Madelung, [*“Physics of III-V Compounds",*]{} J. Wiley & Sons, (1964) 115. J. D. Wiley, [*“Semiconductors and Semimetals",*]{} Vol. 10, ed. by R. K. Willardson and A.C. Beer, Academic Press, (1975) 169. L. Sosnowski, [*“Thermo-Electric and Thermo-Magnetic Effects“, Proc. Int’l School of Physics ”Enrico Fermi" XXII, Semiconductors*]{}, (Academic Press, New York and London 1963) 436. H. J. Hrostowski, F. J. Morin, T. H. Geballe, and G. H. Wheatley, Phys. Rev., [**100**]{}, (1955) 1672. D. M. Zengin, J. Phys., [**D16**]{}, (1983) 635. D. M. S. Bagguley, M. L. A. Pobinson and R. A. Strandling, Phys. Lett., [**6**]{}, (1963) 143.
| {
"pile_set_name": "ArXiv"
} |
The electric $\overline{\alpha}$ and magnetic $\overline{\beta}$ polarizabilities constitute the first-order responses of the internal structure of a nucleon to externally applied electric and magnetic fields. Compton scattering from the proton has been used extensively to determine the polarizabilities of the proton (see Ref. [@Mac-95] and references contained therein). Such measurements are sensitive to the sum (difference) of the polarizabilities for photons scattered at forward (backward) angles. Ref. [@Mac-95] reports the experimental status of the proton polarizabilities, in the usual units of $10^{-4}$ fm$^3$: $$\begin{aligned}
(\overline{\alpha}-\overline{\beta})_p & = & 10.0\pm1.5\pm0.9,
\label{prot-diff}\\
(\overline{\alpha}+\overline{\beta})_p & = & 15.2\pm2.6\pm0.2,\end{aligned}$$ where the first error is the combined statistical and systematic, and the second is due to the model dependence of the dispersion-relation extraction method.
Disperson sum rules relate the sum of the polarizabilities to the nucleon photoabsorption cross section. The generally accepted results [@Pet-81] were $$\begin{aligned}
(\overline{\alpha}+\overline{\beta})_p & = & 14.2 \pm 0.5,
\label{prot-sum} \\
(\overline{\alpha}+\overline{\beta})_n & = & 15.8 \pm 0.5.\end{aligned}$$ There have been recent reevaluations of these sum rules which yield $13.69 \pm 0.14$ ($14.40 \pm 0.66$) [@Bab-98] and $14.0\pm 0.5$ ($15.2\pm 0.5$) [@levchuk1] for the proton (neutron). The proton polarizabilities, obtained from Eqs. \[prot-diff\] and \[prot-sum\], are $$\overline{\alpha}_p = 12.1\pm 0.8\pm 0.5,
\hspace{0.5cm}
\overline{\beta}_p = 2.1\mp 0.8\mp 0.5.
\label{prot-polar}$$
The status of the neutron polarizabilities is much less satisfactory (see Ref. [@Wiss-98] and references contained therein). Measurements of the electric polarizability of the neutron have been done by low-energy neutron scattering from the Coulomb field of a heavy nucleus. The extracted values fall in the range $\alpha_n$=0–19 [@Sch-91; @Koe-95; @Enik-97].
Elastic Compton scattering from the deuteron has also been used to extract information on the nucleon polarizabilities. Using their theoretical model, Levchuk and L’vov [@levchuk1] have reported values of $$(\overline{\alpha}-\overline{\beta})_n = -2 \pm 3,
\hspace{0.5cm}
(\overline{\alpha}+\overline{\beta})_n = 20 \pm 3, \label{neut-diff}$$ from fitting to the data of Refs. [@Lucas-94; @Horn-99]. The large discrepancy between $(\overline{\alpha}-\overline{\beta})_n$ and $(\overline{\alpha}-\overline{\beta})_p$ (Eq. \[prot-diff\]) may be indicative of shortcomings in the theoretical models used to extract the polarizabilities from the $d(\gamma,\gamma)d$ reaction.
The quasi-free Compton scattering reaction $d(\gamma,\gamma^\prime n)p$, in which the scattered photon is detected in coincidence with the recoil neutron, can be used to minimize the model dependence of the extracted polarizabilities. There has been one measurement reported on this reaction using bremsstrahlung photons with an endpoint of 130 MeV [@Rose-90]. A most-probable value of $\overline{\alpha}_n=10.7$ was obtained with an upper limit of 14.0 but with no constraint on the lower limit. In addition, the model dependence is strong at these lower energies.
Levchuk [*et al.*]{} [@Lev-94] have determined that, within the context of their model, the sensitivity to the neutron amplitude is maximized (and model dependence minimized) for E$_\gamma$=200–300 MeV and backward angles for the scattered photons. For $E_\gamma$=247 MeV and $\theta_{\gamma^\prime}$=$-135^\circ$, the contribution of the spectator nucleon modifies the cross section by only $\sim$4% at the quasi-free peak. Furthermore, the free-neutron cross section can be related to the cross section in the center of the neutron quasi-free peak via a spectator formula [@Lev-94; @Wiss-99] $$\frac{ d\sigma(\gamma n\rightarrow \gamma^\prime n) }
{ d\Omega_{\gamma^\prime} } =
\frac{ (2\pi)^3 }{ u^2(0) }
\frac{ E_\gamma E_{\gamma^\prime} }
{ |{\bf p}_n|m{E_{\gamma^\prime}^{(n)2}} }
\frac{ d^3\sigma(\gamma d\rightarrow \gamma^\prime np) }
{d\Omega_{\gamma^\prime}d\Omega_n dE_n },
\label{spectator}$$ where $u(0)$ is the S-wave amplitude of the deuteron wave function at zero momentum, and $E_{\gamma^\prime}^{(n)}$ is the photon energy in the rest frame of the final neutron. The cross section on the right side of the equation must be corrected for the small contribution of the spectator proton (its pole diagram and final state interactions).
The present measurement was performed at the Saskatchewan Accelerator Laboratory (SAL). The $d(\gamma,\gamma^\prime n)p$ and $d(\gamma,\gamma^\prime p)n$ cross sections were measured simultaneously, in kinematics that emphasized the quasi-free reaction. Data were also obtained from the free proton, using the same apparatus, for calibration and normalization purposes. An electron beam of 292 MeV and [$\sim$60%]{} duty factor produced bremsstrahlung photons, which were tagged in the energy range 236–260 MeV, with a resolution of 0.4 MeV. The average, integrated tagged flux was $\sim{6}\times{10^{6}}$ photons/s with a tagging efficiency of $\sim$45%. The total integrated luminosity for the deuterium measurements was $1.25\times 10^{37}$ photons/cm$^{2}$.
The cryogenic target cell was a vertical cylinder (12 cm in diameter), with Mylar walls, containing liquid hydrogen ($p(\gamma,\gamma)$) or liquid deuterium ($d(\gamma,\gamma^\prime N$) and $d(\gamma,pn)$). The quasi-free scattering of 247 MeV photons to $-135^\circ$ from deuterium results in the recoil nucleon being confined to a forward cone centered around 18$^\circ$. These nucleons have a central kinetic energy of $\sim$77 MeV and were detected in a liquid-scintillator array [@Kork-99] situated at 20$^\circ$, subtending a solid angle of approximately 0.11 sr. The array consisted of 85 Lucite-walled cells, filled with BC-505 liquid scintillator. Between the cells and the target were thin plastic scintillators which acted as veto detectors for neutrons or as $\Delta$E detectors for protons.
The neutron efficiency of the array was measured at the beginning of the experiment via the $d(\gamma,pn)$ reaction. A plastic-scintillator $\Delta$E$\cdot$E telescope was placed at 80$^\circ$ to detect the protons while the neutron array was located at $-77^\circ$. The detection efficiency was measured for neutrons with kinetic energies of 50–100 MeV, covering the range of interest for the quasi-free reaction. With the applied threshold of 11 MeV (6 MeV[*electron equivalent*]{}) the efficiency ranged from 5.2–5.9%.
Photons scattered from the target were detected in the large-volume Boston University NaI (BUNI) gamma-ray spectrometer [@Miller-88]. BUNI is composed of five optically-isolated segments of NaI, the core and four surrounding quadrants. Plastic-scintillator detectors were used for rejection of cosmic rays as well as charged particles from the target. BUNI was set at $-135^\circ$ and subtended a solid angle of 0.013 sr. Zero-degree calibrations of BUNI were done at the beginning and end of the experiment, in order to obtain both the lineshape of BUNI and an energy calibration for the NaI core. The NaI quadrants were calibrated periodically with a radioactive source (Th-C).
Figure \[proton\] depicts the BUNI energy spectrum (E$_{\gamma^\prime}$), corrected for photon absorption effects ($\sim$3%), from the free proton. Random coincidences as well as contributions from empty-target backgrounds (36% for $(\gamma,\gamma)$ and 5% for $(\gamma,\gamma p)$) have been subtracted. The spectra are summed over all tagged photon energies and shifted to the maximum incident energy of 260 MeV, with the appropriate kinematic corrections to account for the different scattered photon energy at each incident photon energy.
=3.38in
Figure \[proton\](a) is the energy spectrum from a free proton, without detection of the recoil proton. The Compton scattered events (the peak near 175 MeV) are clearly distinguishable from the large number of events from neutral pion ($\pi^0$) production (below $\sim$165 MeV). The histogram is the result of a simulation of the experiment. The final yield was determined by integrating the data from 170 to 190 MeV and correcting for the Compton events excluded from this region (18%) and the $\pi^0$ events included in this region (1%). For E$_\gamma = 247$ MeV and $\theta_\gamma=-135^\circ$, the cross section for elastic Compton scattering from the free proton ($p(\gamma,\gamma)p$) was determined to be $$\frac{d\sigma}{d\Omega_\gamma} = 94.6 \pm 9.0 \pm 3.8~{\rm nb/sr},
\label{free}$$ where the first error is statistical and the second is systematic. This is in good agreement with recent measurements [@Wiss-99; @Hallin; @legs]. The dispersion calculations of L’vov [*et al.*]{} [@Lvov-97] predict a somewhat lower value of 78.0 nb/sr using $(\overline{\alpha}-\overline{\beta})_p$=10.
The efficiency for the detection of protons in the array was determined in situ from the fraction of the $p(\gamma,\gamma)p$ events (Fig. \[proton\](a)) that included a detected proton (Fig. \[proton\](b)). This correction ($0.58 \pm 0.02$) was accounted for by a combination of the simulation and cuts in the analysis.
=3.38in
For Compton scattering from a bound nucleon, the Fermi motion causes overlap in E$_{\gamma^\prime}$ between the Compton and $\pi^0$ events. In order to obtain better separation, the missing mass (${\rm M_X}$) of the spectator nucleon was calculated. Figure \[miss-mass\] depicts the difference between the missing mass and the spectator mass ($\Delta {\rm M_{miss}}$) for the two quasi-free reactions. The yields are corrected for random coincidences, empty-target backgrounds (12–14%), and detection efficiencies. The histograms are from simulations of the experiment. The thin lines are the individual contributions from the $\pi^0$ and Compton reactions while the thick line is the sum. The simulation assumes that the deuteron reactions proceed via quasi-free kinematics. The spectator nucleon is distributed according to its Fermi distribution and the angular distribution of the $\pi^0$ or photon is determined by the theoretical calculations of Levchuk [*et al.*]{} The simulation gives a good fit to the data, and is relatively insensitive to the input angular distribution ($\sim$2%). The $\chi^2/N_{d.o.f.}$ is 1.4 (proton) and 1.5 (neutron) for $N_{d.o.f.}$=63 ($-20$ to 110 MeV). To extract the yields for the Compton reactions, the data were integrated from $-20$ to +20 MeV and corrected for the Compton events excluded from this region (10–15%) and the $\pi^0$ events included in this region (7–8%).
In order to compare to theoretical predictions, the measured cross sections were interpolated to the quasi-free kinematic point ($P_{spectator}=0$) through use of the simulation. The average and root-mean-square values for $P_{spectator}$ were 48 and 55 MeV/c, respectively. For E$_\gamma = 247$ MeV and $\theta_{\gamma^\prime}=-135^\circ$, the final differential cross sections, at the quasi-free point, are $$\begin{aligned}
\frac{d^3\sigma}{d\Omega_\gamma d\Omega_N dE_N} & = &
56.5 \pm 2.8 \pm 4.5~({\rm proton}),\label{qf-prot}\\
& = & 33.3 \pm 7.2 \pm 3.0~({\rm neutron}),\label{qf-neut}\end{aligned}$$ where the units are nb/sr$^2$/MeV, the first error is statistical, and the second is systematic. The proton result is consistent with the recent results of Wissmann [*et al.*]{} [@Wiss-99].
Since the back-angle Compton cross section is primarily sensitive to the difference of the polarizabilities, $(\overline{\alpha}-\overline{\beta})_p$ was used as a fit parameter in matching the theoretical calculation Levchuk [*et al.*]{} [@Lev-94] to the bound proton data (Fig. \[ratio\](a)). The solid line is the theoretical calculation, the thin line is the central fitted value, and the horizontal dashed lines indicate the combined-error band (the quadratic sum of statistical and systematic). The model dependence of the calculation is small compared to the experimental errors. The fitted value, $$(\overline{\alpha}-\overline{\beta})_p = {14.7^{+4.6}_{-4.0}},$$ is consistent with the free proton result (Eq. \[prot-diff\]), as well as with the quasi-free results of Wissmann [*et al.*]{} [@Wiss-99]. This indicates that the theoretical calculations are adequately describing the nuclear effects.
The theoretical calculations have been carried out using a ‘traditional’ value for the backward-spin polarizability of $\gamma_\pi$=$-36.8$. Using the value $\gamma_\pi$=$-27.1$, suggested by Tonnison [*et al.*]{} [@tonnison], results in a fitted value of $$(\overline{\alpha}-\overline{\beta})_p = 26.6^{+7.8}_{-5.8},$$ which is over two standard deviations from the free value. Our data show no evidence for modification of $\gamma_\pi$, in agreement with the conclusions of Wissmann [*et al.*]{} [@Wiss-99].
The ratio of the bound proton to the bound neutron quasi-free cross section is 1.70$\pm$0.38$\pm$0.13. Using the ratio of the measured cross sections minimizes the effect of systematic errors on the extraction of $(\overline{\alpha}-\overline{\beta})_n$, and reduces the model dependence inherent in extracting information on the free-neutron from quasi-free results. The ratio for the theoretical calculation was obtained by fixing the proton polarizabilities (Eq. \[prot-diff\]) and treating $(\overline{\alpha}-\overline{\beta})_n$ as a fit parameter (Fig. \[ratio\](b)). The best fit was obtained with $(\overline{\alpha}-\overline{\beta})_n = 12$, with a lower limit of $0$. No upper limit can be obtained. The model uncertainty in extracting $(\overline{\alpha}-\overline{\beta})_n$ is expected to be negligible compared to the experimental error. The central value is markedly different from the value extracted from elastic Compton scattering from the deuteron (Eq. \[neut-diff\]). However, the two are within errors. Invoking the sum rule [@levchuk1], the values for the polarizabilities are $\overline{\alpha}_n = 13.6$, with a lower limit of 7.6, and $\overline{\beta}_n$ = 1.6, with an upper limit of 7.6. The upper (lower) limits on $\overline{\alpha}_n$ ($\overline{\beta}_n$) are unconstrained by the present measurement. Combining our results with those from Rose [*et al.*]{} [@Rose-90], we obtain one-sigma constraints of $\overline{\alpha}_n = 7.6-14.0$ and $\overline{\beta}_n = 1.2-7.6$. These can be compared with recent Chiral Perturbation Theory (ChPT) predictions at $O(Q^4)$ [@bernard] $$\overline{\alpha}_n = 13.4 \pm 1.5, \hspace{1cm}
\overline{\beta}_n = 7.8 \pm 3.6.
\label{neut-polar}$$
=3.38in
From Eqs. \[spectator\] and \[qf-prot\] the effective free-proton cross section at $\theta_{\gamma^\prime}=-135^\circ$ can be determined from that of the bound proton ($d(\gamma,\gamma^\prime p)n$). The result is $$\frac{ d\sigma(\gamma p\rightarrow \gamma^\prime p) }
{ d\Omega_\gamma } = 66.8 \pm 3.4 \pm 5.3~{\rm nb/sr},$$ corresponding to E$_\gamma=244.6$ MeV for the free proton. Adjusting Eq. \[free\] to this lower energy results in a $\sim$5% drop in the measured cross section to 90.1$\pm$8.6$\pm$3.6 nb/sr. The ratio of the measured to effective free values is $1.35\pm 0.18$, indicating they are not in good agreement. Model-dependence in extracting effective free-neutron values from quasi-free should be minimized by using the ratio of quasi-free cross sections, as discussed above. The dispersion calculations of L’vov [*et al.*]{} [@Lvov-97] predict a free value of 74.2 nb/sr for $(\overline{\alpha}-\overline{\beta})_p$=10.
Eqs. \[spectator\] and \[qf-neut\] can also be used to determine the effective free cross section for the neutron at $\theta_{\gamma^\prime}=-135^\circ$: $$\frac{ d\sigma(\gamma n\rightarrow \gamma^\prime n) }
{ d\Omega_\gamma } = 39.5 \pm 8.5 \pm 3.6~{\rm nb/sr},$$ corresponding to E$_\gamma=244.6$ MeV for the free neutron. This is the first experimental determination of the cross section for Compton scattering from the free neutron. An alternate method for extracting the free-neutron cross section from the measured values is to take the ratio of free to quasi-free cross sections for the proton (Eqs. \[free\] and \[qf-prot\]) and multiply by the quasi-free neutron cross section (Eq. \[qf-neut\]), resulting in 55.8$\pm$13.5$\pm$4.9. The ratio of the extracted free values $1.41\pm 0.47$ is consistent with unity, albeit with large errors. The dispersion calculations of L’vov [*et al.*]{} [@Lvov-97] predict a free value of 45.7 nb/sr (for $(\overline{\alpha}-\overline{\beta})_n$=10) in agreement with the experimental results.
In summary, the polarizabilities of the bound proton and neutron have, for the first time, been simultaneously extracted from quasi-free Compton scattering measurements. The values for the neutron (Eq. \[neut-polar\]) are consistent with those known for the free proton (Eq. \[prot-polar\]), as expected from charge-symmetry arguments, and are in accord with recent ChPT predictions.
The authors would like to thank M.I. Levchuk, A.I. L’vov, and V.A. Petrun’kin for supplying the results from their theoretical calculations. N.R.K. would like to thank M.I.L. for many useful discussions.
This work was supported in part by grants from the Natural Science and Engineering Research Council of Canada and the National Science Foundation (USA).
B.E. MacGibbon [*et al.*]{}, , 2097 (1995). V.A. Petrun’kin, Sov. J. Part. Nucl. [**11**]{}, 278 (1981). D. Babusci, G. Giordano, and G. Matone, , 291 (1998). M.I. Levchuk and A.I. L’vov, Nucl. Phys. [**A**]{} (in press). F. Wissmann, M.I. Levchuk, M. Schumacher, Eur. Phys. J. A [**1**]{}, 193 (1998). J. Schmiedmayer, P. Reihs, J.A. Harvey, and N.W. Hill, , 1015 (1991). L. Koester [*et al.*]{}, , 3363 (1995). T.L. Enik, L.V. Mitsyna, V.G. Nikolenko, A.B. Popov, and G.S. Samosvat, Sov. J. Nucl. Phys. [**60**]{}, 567 (1997). M.A. Lucas, Ph.D. Thesis, University of Illinois, 1994. D.L. Hornidge [*et al.*]{}, , 2334 (2000). K.W. Rose [*et al.*]{}, Nucl. Phys. [**A514**]{}, 621 (1990). M.I. Levchuk, A.I. L’vov, and V.A. Petrun’kin, Few-Body Sys. [**16**]{}, 101 (1994). F. Wissmann [*et al.*]{}, Nucl. Phys. [**660**]{}, 232 (1999). E. Korkmaz [*et al.,*]{} Nucl. Instrum. Methods Phys. Res. Sect. A [**431**]{}, 446 (1999). J.P. Miller [*et al.*]{}, Nucl. Instrum. Methods Phys. Res. Sect. A [**270**]{}, 431 (1988). E.L. Hallin [*et al.,*]{}, , 1497 (1993). A.M. Sandorfi, J. Tonnison, S. Hoblit, Few Body Sys., Suppl. [**11**]{}, 144 (1999); S. Hoblit, private communication. A.I. L’vov, V.A. Petrun’kin, M. Schumacher, , 359 (1997). J. Tonnison, A.M. Sandorfi, S. Hoblit, A.M. Nathan, Phys. Rev. Lett. [**80**]{}, 4382 (1998). V. Bernard, N. Kaiser, U.-G. Mei$\beta$ner, Int. J. Mod. Phys. E[**4**]{}, 193 (1995).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Pion Virtual Compton Scattering (VCS) via the reaction $\pi^-{e}\rightarrow\pi^-{e}\gamma$ was observed in the Fermilab E781 SELEX experiment. SELEX used a $600\,\mbox{GeV/c}$ $\pi^-$ beam incident on target atomic electrons, detecting the incident $\pi^-$ and the final state $\pi^-$, electron and $\gamma$. Theoretical predictions based on chiral perturbation theory are incorporated into a Monte Carlo simulation of the experiment and are compared to the data. The number of reconstructed events (9) and their distribution with respect to the kinematic variables (for the kinematic region studied) are in reasonable accord with the predictions. The corresponding $\pi^-$ VCS experimental cross section is $\sigma=38.8\pm{13}\,\mbox{nb}$, in agreement with the theoretical expectation $\sigma=34.7\,\mbox{nb}$.'
address: |
$^1$Ball State University, Muncie, IN 47306, U.S.A.\
$^2$Bogazici University, Bebek 80815 Istanbul, Turkey\
$^3$Carnegie-Mellon University, Pittsburgh, PA 15213, U.S.A.\
$^4$Centro Brasiliero de Pesquisas Físicas, Rio de Janeiro, Brazil\
$^5$Fermilab, Batavia, IL 60510, U.S.A.\
$^6$Institute for High Energy Physics, Protvino, Russia\
$^7$Institute of High Energy Physics, Beijing, P.R. China\
$^8$Institute of Theoretical and Experimental Physics, Moscow, Russia\
$^9$Max-Planck-Institut für Kernphysik, 69117 Heidelberg, Germany\
$^{10}$Moscow State University, Moscow, Russia\
$^{11}$Petersburg Nuclear Physics Institute, St. Petersburg, Russia\
$^{12}$Tel Aviv University, 69978 Ramat Aviv, Israel\
$^{13}$Universidad Autónoma de San Luis Potosí, San Luis Potosí, Mexico\
$^{14}$Universidade Federal da Paraíba, Paraíba, Brazil\
$^{15}$University of Bristol, Bristol BS8 1TL, United Kingdom\
$^{16}$University of Iowa, Iowa City, IA 52242, U.S.A.\
$^{17}$University of Michigan-Flint, Flint, MI 48502, U.S.A.\
$^{18}$University of Rome “La Sapienza” and INFN, Rome, Italy\
$^{19}$University of São Paulo, São Paulo, Brazil\
$^{20}$University of Trieste and INFN, Trieste, Italy\
author:
- |
A. Ocherashvili$^{12}$$^{,\P\P}$, G. Alkhazov$^{11}$, A.G. Atamantchouk$^{11}$, M.Y. Balatz$^{8}$$^{,\ast}$, N.F. Bondar$^{11}$, P.S. Cooper$^{5}$, L.J. Dauwe$^{17}$, G.V. Davidenko$^{8}$, U. Dersch$^{9}$$^{,\dag}$, A.G. Dolgolenko$^{8}$, G.B. Dzyubenko$^{8}$, R. Edelstein$^{3}$, L. Emediato$^{19}$, A.M.F. Endler$^{4}$, J. Engelfried$^{13,5}$, I. Eschrich$^{9}$$^{,\ddag}$, C.O. Escobar$^{19}$$^{,\S}$, A.V. Evdokimov$^{8}$, I.S. Filimonov$^{10}$$^{,\ast}$, F.G. Garcia$^{19,5}$, M. Gaspero$^{18}$, I. Giller$^{12}$, V.L. Golovtsov$^{11}$, P. Gouffon$^{19}$, E. Gülmez$^{2}$, He Kangling$^{7}$, M. Iori$^{18}$, S.Y. Jun$^{3}$, M. Kaya$^{16}$, J. Kilmer$^{5}$, V.T. Kim$^{11}$, L.M. Kochenda$^{11}$, I. Konorov$^{9}$$^{,\P}$, A.P. Kozhevnikov$^{6}$, A.G. Krivshich$^{11}$, H. Krüger$^{9}$$^{,\parallel}$, M.A. Kubantsev$^{8}$, V.P. Kubarovsky$^{6}$, A.I. Kulyavtsev$^{3}$$^{,\ast\ast}$, N.P. Kuropatkin$^{11}$, V.F. Kurshetsov$^{6}$, A. Kushnirenko$^{3}$, S. Kwan$^{5}$, J. Lach$^{5}$, A. Lamberto$^{20}$, L.G. Landsberg$^{6}$, I. Larin$^{8}$, E.M. Leikin$^{10}$, Li Yunshan$^{7}$, M. Luksys$^{14}$, T. Lungov$^{19}$$^{,\dag\dag}$, V.P. Maleev$^{11}$, D. Mao$^{3}$$^{,\ast\ast}$, Mao Chensheng$^{7}$, Mao Zhenlin$^{7}$, P. Mathew$^{3}$$^{,\ddag\ddag}$, M. Mattson$^{3}$, V. Matveev$^{8}$, E. McCliment$^{16}$, M.A. Moinester$^{12}$, V.V. Molchanov$^{6}$, A. Morelos$^{13}$, K.D. Nelson$^{16}$$^{,\S\S}$, A.V. Nemitkin$^{10}$, P.V. Neoustroev$^{11}$, C. Newsom$^{16}$, A.P. Nilov$^{8}$, S.B. Nurushev$^{6}$, A. Ocherashvili$^{12}$, Y. Onel$^{16}$, E. Ozel$^{16}$, S. Ozkorucuklu$^{16}$, A. Penzo$^{20}$, S.I. Petrenko$^{6}$, P. Pogodin$^{16}$, M. Procario$^{3}$$^{,\P\P\P}$, V.A. Prutskoi$^{8}$, E. Ramberg$^{5}$, G.F. Rappazzo$^{20}$, B.V. Razmyslovich$^{11}$, V.I. Rud$^{10}$, J. Russ$^{3}$, P. Schiavon$^{20}$, J. Simon$^{9}$$^{,\ast\ast\ast}$, A.I. Sitnikov$^{8}$, D. Skow$^{5}$, V.J. Smith$^{15}$, M. Srivastava$^{19}$, V. Steiner$^{12}$, V. Stepanov$^{11}$, L. Stutte$^{5}$, M. Svoiski$^{11}$, N.K. Terentyev$^{11,3}$, G.P. Thomas$^{1}$, L.N. Uvarov$^{11}$, A.N. Vasiliev$^{6}$, D.V. Vavilov$^{6}$, V.S. Verebryusov$^{8}$, V.A. Victorov$^{6}$, V.E. Vishnyakov$^{8}$, A.A. Vorobyov$^{11}$, K. Vorwalter$^{9}$$^{,\dag\dag\dag}$, J. You$^{3,5}$, Zhao Wenheng$^{7}$, Zheng Shuchen$^{7}$, R. Zukanovich-Funchal$^{19}$\
0.50cm 0.50cm
title: 'First Measurement of $\pi^-{e}\rightarrow\pi^- e \gamma$ Pion Virtual Compton Scattering'
---
Introduction
============
The electric $(\bar{\alpha})$ and magnetic $(\bar{\beta})$ pion polarizabilities characterize the pion’s deformation in an electromagnetic field, as occurs during $\gamma\pi$ Compton scattering. They depend on the rigidity of the pion’s internal structure as a composite particle, and are therefore important dynamical quantities to test the validity of theoretical models. Based on QCD chiral dynamics, the chiral perturbation theory effective Lagrangian, using data from radiative pion beta decay, predicts the pion electric and magnetic polarizabilities $\bar{\alpha}_{\pi}$ = -$\bar{\beta}_{\pi}$ = 2.7 $\pm$ 0.4, expressed in units of $10^{-43}\,\mbox{cm}^3$ [@Pennington; @Pole; @ho90]. Other theoretical predictions are also available [@Pennington].
The pion polarizabilities are usually investigated via their effect on the shape of the measured $\gamma\pi\rightarrow\gamma\pi$ Real Compton Scattering (RCS) angular distribution, as in Ref. [@kbp]. Since pion targets are unavailable, pion RCS is approximated using different artifices, as shown in Fig. \[fig.polpos\]: the $\pi^- Z\rightarrow\pi^-{Z}\gamma$ Primakoff [@Antipov] and $\gamma p\rightarrow\gamma\pi^+ n$ radiative pion photoproduction reactions [@aibergenov]; or by the crossing symmetry $\gamma\gamma\rightarrow\pi^+ \pi^-$ two-photon reaction [@Pole; @Boyer]. In the Primakoff scattering, a high energy pion scatters from a (virtual, practically real) photon in the Coulomb field of the target nucleus. Values of $\bar{\alpha}$ measured by these experiments are given in Table \[tab:mespol\].
Reaction $\bar{\alpha}\,\,[10^{-43}\,\mbox{cm}^3]$ Reference
--------------------------------------- ------------------------------------------- ------------------
$\pi^-Z\rightarrow \pi^-Z\gamma$ $6.8\pm 1.4\pm 1.2$ [@Antipov]
$\gamma p \rightarrow \gamma \pi^+ n$ $20\pm 12$ [@aibergenov]
$\gamma\gamma \rightarrow \pi^+\pi^-$ $2.2\pm 1.6$ [@Pole; @Boyer]
: Experimental values of $\bar{\alpha}$.[]{data-label="tab:mespol"}
They cover a large range of values and have large uncertainties. New high precision pion polarizability measurements are therefore needed.
Electromagnetic studies with virtual photons have the advantage that the energy and three-momentum of the virtual photon can be varied independently. In the pion $\gamma^*\pi\rightarrow\gamma\pi$ VCS reaction, where the initial state photon is virtual (far from the quasi-real photons of Primakoff scattering) and the final state photon is real, polarizabilities can be measured in the space like region, inaccessible by RCS [@Scherer_I]. We thereby access the so-called electric $\bar{\alpha}(q^2)$ and magnetic $\bar{\beta}(q^2)$ generalized polarizabilities of the pion [@newalpha2], where $\bar{\alpha}(0)$ and $\bar{\beta}(0)$ correspond to the RCS $\bar{\alpha}$ and $\bar{\beta}$ pion Compton polarizabilities. The $q^2$-dependent $\bar{\alpha}(q^2)$ determines the change $\Delta{F}(q^2)$ in the pion charge form factor $F(q^2)$ in the presence of a strong electric field. The Fourier transform of $\bar{\alpha}(q^2)$ provides a picture of the local induced pion charge polarization density [@pionffc]. Similarly, first experiments [@dhos] and calculation [@pas] have been carried out for proton VCS via $e p
\rightarrow e p \gamma$.
We study experimentally the feasibility of extracting the “pion VCS” reaction: $$\label{react}
\pi^- e\rightarrow\pi^- e \gamma,$$ as a step in developing pion VCS as a new experimental tool for pion polarizability measurements. The data were taken with the Fermilab E781 SELEX spectrometer [@prop]. We used a $600\,\mbox{GeV/c}$ $\pi^-$ beam incident on target atomic electrons, detecting the incident $\pi^-$ and the final state $\pi^-$, electron and $\gamma$. Theoretical predictions [@pvcs; @scherer] based on chiral perturbation theory are incorporated into a Monte Carlo simulation of the experiment and are compared to the data. With this theory, we calculated the total cross section (as described later) of the Eq. (\[react\]) process for a limited kinematic region discussed later \[Eq. (\[s1s2r\])\], using the Monte Carlo integration program VEGAS [@vegas]. The result is $\sigma(total)=34.7\pm{0.1}\,\mbox{nb}$. For the kinematic range considered, the integrated cross sections are not sensitive to the polarizabilities. We nonetheless chose this region in order to obtain sufficient statistics for a first study of the reaction.
VCS kinematics and theoretical differential cross section {#sec:vcsk}
=========================================================
We study reaction (\[react\]) in terms of the following five independent invariant variables: $$\begin{aligned}
\label{ivar}
s=(p_i+k)^2, \nonumber \\
s_1=(k'+q')^2, &
s_2=(p_f+q')^2, \\
r^2=(p_i-p_f)^2, &
q^2=(k'-k)^2. \nonumber\end{aligned}$$ Here $p_i$ is the 4-momentum of the incoming pion, $k$ is the 4-momentum of the target electron, and $p_f$, $k'$, $q'$ are the 4-momenta of the outgoing pion, electron, and photon, respectively, as shown in Fig. \[fig.kin\].
The differential cross section of reaction (\[react\]) in the convention of Ref. [@bdrel] reads as: $$\begin{aligned}
\label{bdcross}
d \sigma=\frac{m_e^2}{8E_fE_{k'}E_{q'}}
\frac{1}{\sqrt{(p_i\cdot
k)^2-M_{\pi}^2m_e^2}}\frac{1}{(2\pi)^5}\cdot \nonumber\\
|\overline{{\cal M}}
|^2 \delta^4(p_i+k-p_f-k'-q')d^3p_fd^3k'd^3q'. \end{aligned}$$ The invariant amplitude $\cal{M}$ contains the complete information on the dynamics of the process. The quantity $|\overline{\cal{M}}|^2$ indicates the sum over the final states and the average over the initial spin states. The fourfold differential cross section in terms of independent invariant variables of Eq. (\[ivar\]) involves a kinematical function $\lambda$ [@kinematic] and a Jacobian matrix $\Delta_4$ defining the phase space of the physical areas, and is given by: $$\label{eq:xsection}
\frac{d\sigma}{ds_1ds_2dq^2dr^2}=
\frac{1}{{(2\pi)}^5}
\frac{2m_e^2}{\lambda{(s,m_e^2,m_{\pi}^2)}}
\frac{\pi}{16}
\frac{1}{{(-\Delta_4)}^{1/2}}
{|\bar{\cal M}|}^2.$$ Since the variable $s$, involving the energies of the beam pion and of the target electron, is fixed, differential cross section (\[eq:xsection\]) actually depends on four variables.
In reaction (\[react\]), the final photon can be emitted either by the electron or by the pion, as shown in Fig. \[fig.fdiag\]. The first process is described by the Bethe-Heitler (BH) amplitude (Fig. \[fig.fdiag\]a, b), calculable from quantum electro-dynamics.
The second process is described by the VCS (Fig. \[fig.fdiag\]c) amplitude. Since the source of the final photon emission is indistinguishable, one obtains the following form of the matrix element of reaction (\[react\]) [@pvcs]: $$\begin{aligned}
\label{mbhmvcs}
|\overline{\cal M}|^2=\frac{1}{2}\sum
({\cal M}^{VCS}+{\cal M}^{BH})({\cal M}^{VCS*}+{\cal M}^{BH*})=\nonumber\\
|\overline{{\cal M}^{BH}}|^2+|\overline{{\cal M}^{VCS}}|^2+\nonumber\\
|\overline{{\cal M}^{VCS}{\cal M}^{BH*}+{\cal M}^{VCS*}{\cal M}^{BH}}|,\end{aligned}$$ where ${\cal M}^{BH}$ and ${\cal M}^{VCS}$ are amplitudes of the BH and VCS processes.
The general features of the four-fold differential cross section can be inferred from Eq. (\[eq:xsection\]) and matrix element calculations. The $s_1$-dependence is dominated by the $(s_1-m^2_e)^{-1}$ pole of BH, the cross section varies accordingly, and is only slightly modified by the $s_1$ dependence of the VCS amplitude. The $s_2$ dependence is dominated by the $(s_2-M^2_{\pi})^{-1}$ pole of the VCS amplitude with modification of $s_2$ dependence of the BH amplitude, but in this case the modification is not as small as in the case of $s_1$. The $r^2$ dependence is determined by the $r^{-4}$ pole of the BH amplitude, and the $q^2$ dependence follows the $q^{-4}$ pole behavior of typical electron scattering. The energy of the outgoing pion is expected to be high while the angle is expected to be small according to the $r^{-4}$ behavior of the cross section. The energy of the outgoing photon is mainly expected to be low, as follows from the infrared divergence of the BH amplitude. The angular behavior of the outgoing photon is determined by the $(s_1-m^2_e)^{-1}$ and $(s_2-M^2_{\pi})^{-1}$ poles of the BH and VCS amplitudes. The more interesting photons related to the generalized polarizabilities are expected to have higher energies. The behavior of the outgoing electron is completely determined by the $q^{-4}$ behavior of the cross section.
Experimental apparatus and trigger
==================================
Our data were taken with the Hadron-Electron (HE) scattering trigger [@trigger] of experiment E781/SELEX at Fermilab. SELEX uses a negatively charged beam of $600\,\mbox{GeV/c}$ with full width momentum spread of dp/p=$\pm$ 8%, and an opening solid angle of 0.5 $\mu$sr. The beam consisted of approximately 50% $\pi^-$ and 50% $\Sigma^-$. SELEX used Copper and Carbon targets, totaling 4.2% of an interaction length, with target electron thicknesses of $0.676\,\mbox{barn}^{-1}$ and $0.645\,\mbox{barn}^{-1}$, respectively.
The experiment focused on charm baryon hadroproduction and spectroscopy at large-$x_F$. The spectrometer hosted several projects which exploit physics with a small number of tracks compared to charm. SELEX had good efficiency for detecting all particles in the final state since the produced particles and decay fragments at large-$x_F$ production are focused in a forward cone in the laboratory frame. Other requirements in the charm program for background suppression include good vertex resolution and particle identification over a large momentum range.
Four dipole magnets divide SELEX into independent spectrometers (Beam, M1, M2 and M3) each dedicated to one special momentum region. Each spectrometer had a combination of tracking detectors. The M1, M2 and M3 spectrometers included electromagnetic calorimeters. The $\pi-e$ separation for hadron-electron scattering used the M2 particle identification transition radiation detector and also the electromagnetic calorimeter.
The HE scattering trigger was specialized for separation of hadron-electron elastic scattering events. The trigger used information from a charged particle detectors just downstream of the target and, from a hodoscope just downstream of M2, to determine charged particle multiplicity and charge polarity. For the HE requirement, no electromagnetic calorimeter information was included. Therefore, the data collected via this trigger include hadron-electron elastic and inelastic scattering events.
Monte Carlo Simulation
======================
Monte Carlo simulations were carried out for $\pi^-$ VCS signal (\[react\]) and background event distributions with respect to the four invariants s$_1$, s$_2$, q$^2$, r$^2$. We used the SELEX GEANT package GE781 [@mc2]. The Monte Carlo (MC) study was carried out in four steps:
1. A VCS event generator was written to search for the regions of phase space where the data are sensitive to pion structure. Several event generators were made to simulate a variety of expected backgrounds.
2. The event generators were implemented into the simulation package. We studied the resolution, detection efficiency, geometric and trigger acceptances for the signal and background.
3. The offline analysis procedure was developed and tuned to devise software cuts eliminating background while preserving a VCS signal.
4. Finally we estimated the expected number of $\pi^-$ VCS events.
The VCS event generator is written, based on differential cross section (\[eq:xsection\]), matrix element calculations, and 3-body final state kinematics. The acceptance-rejection method [@ARM] is used for event generation. The VCS cross section increases rapidly when the direction of the outgoing real photon is close to the direction of one of the outgoing charged particles (due to the $(s_1-m^2_e)^{-1}$ pole of BH and the $(s_2-M^2_{\pi})^{-1}$ pole of VCS), or when the energy of the outgoing real photon comes close to zero (due to the infrared divergence of the BH ($(s_1-m^2_e-q^2+r^2)^{-1}$ pole). Therefore, if events are generated in the pole region, then the efficiency of the acceptance-rejection method can be rather low, for the more interesting non-pole regions. In order to generate events at an acceptable rate, the pole regions are eliminated. Invariants are generated in the following regions: $$\begin{aligned}
\label{s1s2r}
1000m^2_e\leq{s_1}\leq{M^2_{\rho}}, \nonumber \\
2M^2_{\pi}\leq{s_2}\leq{M^2_{\rho}}, \nonumber \\
-0.2\;GeV^2<q^2<-0.032\;GeV^2, \nonumber \\
-0.2\;GeV^2<r^2<-2m_e E_{\gamma}(min)+q^2+s_1-m^2_e.\end{aligned}$$ For the photon minimum energy, we choose $E_{\gamma}(min)=5$ GeV to cut the infrared divergence of BH, and to be above the calorimeter noise. Since the VCS calculation does not explicitly include the $\rho$ resonance, we choose upper limits of $M^2_{\rho}$ for $s_1$ and $s_2$. In Fig. \[fig.geninv\], we show the generated distribution of events plotted with respect to the Mandelstam invariants, without correction for any acceptances. The solid and dashed curves are for BH+VCS and BH respectively. The VCS amplitude clearly affects the shape of these distributions for s$_1$ and s$_2$. For q$^2$ and r$^2$ the effect is difficult to see for the statistics shown. Taken together, the experiment therefore is potentially sensitive to the pion VCS amplitude.
Since $\pi{/e}$ separation is not 100% efficient, all interactions which produce two negatively charged tracks and a photon in the final state can generate the pattern of the pion VCS; i.e., can create background to the required measurement. For the background simulation, as well as in case of the VCS simulation, we require 5 GeV minimum energy for the photon.
The dominant background process for pion VCS is $\pi^-{e}$ elastic scattering followed by final state interactions of the outgoing charged particles, such as Bremsstrahlung. The MC simulation shows that 28% of the original $\pi^-{e}$ elastic scattering events generate more than 5 GeV Bremsstrahlung photons somewhere in the SELEX apparatus. We therefore consider the $s_3$ invariant mass of the outgoing $\pi^-{e}$ system: $$\label{s3dec}
s_3=(p_f+k')^2.$$ We expect $s_3=s$ for elastic scattering, and $s_3=s_1-s_2+m_e^2+M^2_{\pi}$ for VCS.
Fig. \[fig.s3talk\] shows generated and reconstructed elastic and VCS events subjected to the same charged particle track reconstruction and trigger requirements as the data. The data shown are only in the kinematic region of Eq. \[s1s2r\]. As seen from Fig. \[fig.s3talk\], the events in the data at low $s_3$ do not arise from $\pi^- e$ elastic or $\pi^-$ VCS. To describe the events with low $s_3$, we simulate backgrounds (with corrections for acceptances) from the interactions: $$\begin{aligned}
\label{rhoe}
\pi^-~e\rightarrow{M}~e,\end{aligned}$$ $$\begin{aligned}
\label{meson}
\pi^-~A(Z)\rightarrow{M}~A(Z),\end{aligned}$$ where $M$ in Eqs. (\[rhoe\]) and (\[meson\]) is an intermediate meson state which can decay via $\pi^-\pi^0$, $\pi^-\eta$, $\pi^-\omega$, etc. Considering the threshold energies of reaction (\[rhoe\]), only $\rho$ meson production is allowed at SELEX energies. Fig. \[fig.s3mesontalk\] shows the simulated $s_3$ distributions for reactions (\[rhoe\]-\[meson\]), with the same acceptance requirements as in Fig. \[fig.s3talk\].
SELEX GE781 allows us to estimate the background rate from $\pi^- e$ elastic scattering. However, it is very difficult to estimate the background rate from all neutral meson ($\pi^0$, etc.) production reactions, such as those shown in Fig. \[fig.s3mesontalk\]. The data of Fig. \[fig.s3talk\](bottom) are qualitatively well described by a combination of elastic (Fig. \[fig.s3talk\] top) and meson production (Fig. \[fig.s3mesontalk\] bottom) channels. The contributions of VCS (Fig. \[fig.s3talk\] middle) and $\rho$ production (Fig. \[fig.s3mesontalk\] top) are relatively much lower. We do not show the quantitative sum of all contributions, because of the difficulty to estimate the absolute yields of all meson production channels. Instead, we seek a set of cuts which remove as “completely” as possible the background from all $\gamma$ sources, arising from neutral meson decays.
Since we measure all outgoing particles, the reaction kinematics are overdetermined. Therefore, in the data analysis, a constrained $\chi^2$ fitting procedure [@brandt] is used. A veto condition is used based on a 2-body final state constrained $\chi^2$ kinematic fit for reduction of the background from $\pi^-{e}$ elastic scattering [@mpi]. A 3-body final state constrained $\chi^2$ kinematic fit is used for extracting the pion VCS signal.
A final state electron can arise from photon conversion. For this background, the electron is not created at the same vertex as the pion. Consequently, the quality of the 3-track vertex reconstruction should be low. Also, the simulation shows that no VCS outgoing particles hit the upstream photon calorimeter. Therefore, in addition to the constrained kinematic fitting cuts, we use restrictions on the vertex quality and on the total energy deposit in the upstream photon calorimeter to suppress backgrounds.
An additional way to reduce the background from $\pi^-{e}$ elastic scattering is to use the $\gamma{e}$ invariant mass $s_1$ and the $\Theta_{s_1}$ angle (angle between $p_i$ and $(q'+k')$; see Fig. \[fig.kin\]). The simulation shows that if the final photon is emitted via electron bremsstrahlung, the value of $s_1$ should be low. On the other hand if the photon and electron are produced via $\pi^0$, $\eta$, or $\omega$ meson, $s_1$ will “remember” the mass of the parent particle. Holding the value of the $s_1$ invariant to be between the squared mass of $\pi^0$ and $\eta$ mesons cuts the background from the electron bremsstrahlung and from reactions (\[meson\]). Fig. \[fig.s1ths1\] shows the distributions of VCS and background simulations. It is seen that the region with higher $\sqrt{s_1}$ and lower $\Theta_{s_1}$ are mostly populated with pion VCS.
Another possibility to reduce the background from interactions (\[rhoe\]-\[meson\]) is to cut on the invariant $M$, defined as: $$\label{mass}
M=\sqrt{(P_{\pi}+P_e-P_{e'})^2}.$$ For elastic events $M=M_{\pi}$, for VCS $M=\sqrt{s_2}$, for $\rho$ production $M=M_{\rho}$ (see Fig. \[fig.mass\]).
The set of the final cuts follows:
- [**1**]{}, fulfillment of the pion VCS pattern with: identified electron; E$_{\gamma}\ge$5 GeV for photons observed at laboratory angles less than 20 mrad in the downstream electromagnetic calorimeters; no additional tracks and vertices; the total energy deposit is less than 1 GeV in the upstream electromagnetic calorimeter, covering detection angles greater than 30 mrad.
- [**2**]{}, Eq. (\[s1s2r\]) ranges for the invariants, with $s_1$(min)=0.0225 GeV$^2$.
- [**3**]{}, $\chi^2_{elastic}>20$, following a constrained fit procedure [@brandt; @mpi].
- [**4**]{}, $\chi^2_{VCS}\le5$, following a constrained fit procedure [@brandt; @pvcs].
- [**5**]{}, $\Theta_{s_1}<$2 mrad, $M\le0.625$ GeV.
To estimate the number of expected VCS events, as well as the yields of other $\pi^-{e}$ elastic or inelastic scattering events, we use: $$\label{piel}
N_{\pi{e}}=N_{\pi}\cdot\sigma\cdot{N_T}\cdot\epsilon_{ex}\cdot\epsilon_r.$$ Here $N_{\pi{e}}$ is the number of events observed for a particular $\pi{e}$ reaction, $N_{\pi}$ is the number of incident beam pions ($\sim 4.4\cdot 10^{10}$ as measured by beam scalers and including particle identification), $\sigma$ is the cross section for the particular reaction, $\epsilon_r$ is the offline reconstruction efficiency (36.6% for elastic, 2.65% for VCS) of the studied process, and $N_T$ is the target electron density. Since not all experimental properties are implemented in GEANT, an additional efficiency factor $\epsilon_{ex}$ is included in Eq. \[piel\]. This efficiency factor is common to $\pi{e}$ elastic and pion VCS reactions. The value $\epsilon_{ex}$ is calculated by comparison of the actual number of observed $\pi e$ elastic scattering events to the expected number of events. The common efficiency arises since these two reactions have practically the same $q^2$ dependence; $q^2$ being the only kinematical parameter relevant for the trigger and first order data reduction procedure. We calculate the experimental efficiency $\epsilon_{ex}$ (13.4%) from $\pi{e}$ elastic scattering, as described in Ref. [@pvcs]. For extraction of the reference $\pi{e}$ elastic scattering events, we employ the cuts of the SELEX $\pi{e}$ elastic scattering analysis [@mpi]. The cuts described above, considering the studied sources of background, improve the signal/noise ratio from less than 1/400 to more than 361/1. For these estimates, we used the following cross sections: $\sigma_{VCS}=34.7\,\mbox{nbarn}$ and $\sigma_{elastic}=4.27\,\mu\mbox{barn}$ for $\pi e$ scattering; $\sigma_{Primakoff}(C\;target)=0.025$ mbarn and $\sigma_{Primakoff}(Cu\;target)=0.83\,\mbox{mbarn}$ for the $Z^2$-dependent Primakoff scattering. Based on the calculated pion VCS cross section, we expect $\sim$8 events in the $\pi^-$ data sample.
The effect of the above enumerated cuts on the VCS signal, and on the backgrounds coming from $\pi{e}$ elastic scattering and Primakoff meson production, are listed in Table \[tab:mc\_cut\_acc\]. The results are based on the estimated relative cross section for these three processes [@pvcs]. The cuts are very effective in removing backgrounds while retaining signal events, such that the final event sample is essentially pure pion VCS.
Cuts VCS Elastic Meson production
------ ------ ---------- ----------------------
1 29.8 32.1 0.24
2 9.97 9.21 0.03
3 8.89 0.03 0.03
4 3.58 0.003 0.004
5 2.56 $<0.001$ $<3. \times 10^{-6}$
: Percentages of the remaining events after using the cuts for the MC simulated $\pi$ VCS and background events.[]{data-label="tab:mc_cut_acc"}
Data Analysis
=============
In the first stage analysis, events containing one identified electron are selected. For these events, particle trajectories are checked if they form a vertex inside the target material. An event is accepted if it contains exactly three tracks, including the beam particle and an electron candidate, and forms one vertex in the target. We consider data only in the kinematic region of Eq. \[s1s2r\]. On the accepted data set, we applied the system of the cuts discussed above. The working statistics with these cuts on the data are given in Table \[tab:fincutsrej\] and in Fig \[fig.effs3data\], where we show the effects of the cuts on the $s_3$ invariant. The effect of the cuts on the data (Table \[tab:fincutsrej\]) is comparable to the effect on the simulated VCS events (Table \[tab:mc\_cut\_acc\]).
cuts % of remain events
------ --------------------
1 26.8
2 13.9
3 9.93
4 0.61
5 0.13
: Percentages of events remaining after using the cuts.[]{data-label="tab:fincutsrej"}
Finally 9 events (with a statistical uncertainty $\pm3$) were accepted as pion VCS. The corresponding $\pi$ VCS experimental cross section based on Eq. (\[piel\]) under the assumption that the background has been completely eliminated is $\sigma=38.8\pm{13}\,\mbox{nb}$, in agreement with the theoretical expectation $\sigma=34.7\,\mbox{nb}$. The error given is only statistical, and does not include possible systematic uncertainties in the efficiency product $\epsilon_{ex}\cdot\epsilon_r$ in Eq. (\[piel\]).
The comparisons of reconstructed (data) and generated (theory) event distributions, normalized to unit area are shown in Fig. \[fig.invpim\] with respect to the four invariants $s_1,\;s_2,\;q^2$ and $r^2$ of Eq. (\[ivar\]), shown with binning that matches the experimental resolutions. The resolution for each variable was found by Monte Carlo simulation, comparing reconstructed and generated events.
To check whether the data and theory (MC) distributions are consistent with each other, we use the K-S test of Kolmogorov and Smirnov [@ks]. The K-S test is based on normalized cumulative distribution functions (CDF). We use the K-S $D$ statistic, as a measure of the overall difference between the two CDFs. It is defined as the maximum value of the absolute difference between two CDFs. The significance level for a value $D$ (as a disproof of the null hypothesis that the distributions are the same) is given by the probability $P(D)$ [@ks]. A high value of $P(D)$ means that the data and theory CDF are consistent with one another.
Following the K-S procedure, we calculate the normalized cumulative distributions of data and theory, corresponding to Figs. \[fig.invpim\]a-d.The K-S $D$ statistic, and the K-S probabilities for consistency of theory/data distributions, are given in Table \[tab:ks\] and Figs. \[fig.discomprob\]- \[fig.s3compks\].
variable $D$ $P(D)$
---------- ------ --------
$s_1$ 0.18 0.90
$s_2$ 0.27 0.83
$s_3$ 0.27 0.92
$q^2$ 0.18 0.99
$r^2$ 0.07 0.99
: Value of K-S $D$ statistic, and probabilities $P$, for comparison of data with theory for $\bar{\alpha}=2.7$. The comparison of data with theory for $\bar{\alpha}=6.8$ gives practically the same result.[]{data-label="tab:ks"}
The experimental and theoretical CDFs for s$_1$ and r$^2$ look similar. For s$_2$ and q$^2$, some regions of q$^2$ (s$_2$) have the experimental CDFs larger (smaller) than the theoretical CDFs. The values of P(D) (see Table \[tab:ks\]) are sufficiently high for all five CDF’s, as expected if the experimental data are consistent with theory. In addition, the prediction of a total of 8 events is in agreement with the observed $9\pm{3}$ data events. This further supports the conclusion from the K-S test that we observe pion VCS events.
From the limited statistics and sensitivity of this first pion VCS experiment, we cannot determine the $\bar\alpha$ polarizability value, nor can we determine which value of $\bar\alpha$ is preferred. In a future experiment, the sensitivity to pion polarizability may be increased [@pvcs] by achieving a data set in which the final $\gamma$ ($\pi$) has higher (lower) energies. However, such data correspond to a lower cross section, and therefore require a high luminosity experiment.
Conclusions
===========
The pion Virtual Compton Scattering via the reaction $\pi{e}\rightarrow\pi{'}{e'}\gamma$ is observed. We developed and implemented a simulation with a VCS event generator. We defined cuts that allow background reduction and VCS signal extraction. The measured number of reconstructed pion VCS events, and their distributions with respect to the Mandelstam invariants, are in reasonable agreement with theoretical expectations. The corresponding $\pi$ VCS experimental cross section is $\sigma=38.8\pm{13}\,\mbox{nb}$, in agreement with the theoretical expectation $\sigma=34.7\,\mbox{nb}$.
Acknowledgments
===============
The authors are indebted to the staffs of Fermi National Accelerator Laboratory, the Max–Planck–Institut für Kernphysik, Carnegie Mellon University, Petersburg Nuclear Physics Institute and Tel Aviv University for invaluable technical support. We thank Drs. C. Unkmeir and S. Scherer for the VCS matrix element calculation.
This project was supported in part by Bundesministerium für Bildung, Wissenschaft, Forschung und Technologie, Consejo Nacional de Ciencia y Tecnología [(CONACyT)]{}, Conselho Nacional de Desenvolvimento Científico e Tecnológico, Fondo de Apoyo a la Investigación (UASLP), Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP), the Israel Science Foundation founded by the Israel Academy of Sciences and Humanities, Istituto Nazionale de Fisica Nucleare (INFN), the International Science Foundation (ISF), the National Science Foundation (Phy \#9602178), NATO (grant CR6.941058-1360/94), the Russian Academy of Science, the Russian Ministry of Science and Technology, the Turkish Scientific and Technological Research Board (TÜBİTAK), the U.S. Department of Energy (DOE grant DE-FG02-91ER40664 and DOE contract number DE-AC02-76CHO3000), and the U.S.-Israel Binational Science Foundation (BSF).
Acknowledgments
===============
The authors are indebted to the staffs of Fermi National Accelerator Laboratory, the Max–Planck–Institut für Kernphysik, Carnegie Mellon University, Petersburg Nuclear Physics Institute and Tel Aviv University for invaluable technical support. We thank Drs. C. Unkmeir and S. Scherer for the VCS matrix element calculation.
This project was supported in part by Bundesministerium für Bildung, Wissenschaft, Forschung und Technologie, Consejo Nacional de Ciencia y Tecnología [(CONACyT)]{}, Conselho Nacional de Desenvolvimento Científico e Tecnológico, Fondo de Apoyo a la Investigación (UASLP), Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP), the Israel Science Foundation founded by the Israel Academy of Sciences and Humanities, Istituto Nazionale de Fisica Nucleare (INFN), the International Science Foundation (ISF), the National Science Foundation (Phy \#9602178), NATO (grant CR6.941058-1360/94), the Russian Academy of Science, the Russian Ministry of Science and Technology, the Turkish Scientific and Technological Research Board (TÜBİTAK), the U.S. Department of Energy (DOE grant DE-FG02-91ER40664 and DOE contract number DE-AC02-76CHO3000), and the U.S.-Israel Binational Science Foundation (BSF).
deceased Present address: Infineon Technologies AG, München, Germany Now at Imperial College, London SW7 2BZ, U.K. Now at Instituto de Física da Universidade Estadual de Campinas, UNICAMP, SP, Brazil Now at Physik-Department, Technische Universität München, 85748 Garching, Germany Present address: The Boston Consulting Group, München, Germany Present address: Fermilab, Batavia, IL Now at Instituto de Física Teórica da Universidade Estadual Paulista, São Paulo, Brazil Present address: SPSS Inc., Chicago, IL Now at University of Alabama at Birmingham, Birmingham, AL 35294 Present address: Medson Ltd., Rehovot 76702, Israel Present address: DOE, Germantown, MD Present address: Siemens Medizintechnik, Erlangen, Germany Present address: Deutsche Bank AG, Eschborn, Germany
0.25cm J. Portales, M. R. Pennington, hep-ph/9407295,\
D. Morgan and M. R. Pennington, Phys. Lett. [**B272**]{} (1991) 134. D. Babusci, S. Bellucci, G. Giordano, G. Matone, A.M. Sandorfi, M.A. Moinester, Phys. Lett. [**B277**]{}, 158 (1992). B. R. Holstein, Comments Nucl. Part. Phys. [**19**]{} (1990) 221. A. Klein, Phys. Rev. [**99**]{} (1955) 988;\
A. M. Baldin, Nucl. Phys. [**18**]{} (1960) 310;\
V. A. Petrun’kin, Sov. JETF. [**40**]{} (1961) 1148. Yu. M. Antipov et al., Phys. Let. [**B121**]{} (1983) 445;\
Yu. M. Antipov et al., Z.Phys.C [**26**]{} (1985) 495. T. A. Aibergenov, P. S. Baranov, O. D. Beznisko, S. N. Cherepniya, L. V. Filkov, A. A. Nafikov, A. I. Osadchii, V. G. Raevsky, L. N. Shtarkov, E.I. Tamm, Czech. J. Phys. [**36**]{}, 948 (1986). J. Boyer et al., Phys. Rev. [**D42**]{} (1990) 1350. S. Scherer, Few Body Syst. Suppl. [**11**]{} (1999) 327; Czech. J. Phys. [**49**]{} (1999) 1307. C. Unkmeir, S. Scherer, A.I. L’vov, D. Drechsel, Phys. Rev. [**D61**]{}, 034002 (2000). A.I. L’vov, S. Scherer, B. Pasquini, C. Unkmeir, D. Drechsel, Phys. Rev. [**C64**]{}, 015203 (2001). S. Kerhoas et al., Nucl. Phys. [**A666-667**]{} (2000) 44,\
J. Roche et al., Phys. Rev. Lett. [**85**]{} (2000) 708. B. Pasquini, S. Scherer, D. Drechsel, Phys. Rev. [**C63**]{}, 025205 (2001) ; B. Pasquini, S. Scherer, D. Drechsel, nucl-th/0105074;\
and references therein. J. Russ et al., A proposal to construct SELEX, Proposal PE781, Fermilab, 1987, http://fn781a.fnal.gov/. C. Unkmeir, A. Ocherashvili, T. Fuchs, M.A. Moinester, S. Scherer, Phys. Rev. [**C65**]{} (2002) 015201.\
A. Ocherashvili, Pion Virtual Compton Scattering, Ph.D. thesis, Sept. 2000, Tel Aviv University, http://muon.tau.ac.il/$\sim$aharon/phd.html. C. Unkmeir, Pion Polarizabilities in the reaction of radiative pion photoproduction on the proton, Ph.D. thesis, Johannes Gutenberg University, Mainz, Sept. 2000;\
T. Fuchs, B. Pasquini, C. Unkmeir, S. Scherer, hep-ph/0010218. G.P. Lepage, J. Comp. Phys. [**27**]{} (1978) 192. J. D. Bjorken, S. D. Drell, Relativistic Quantum Mechanics, McGraw-Hill, New York, 1964. E. Byckling K. Kajantie, [*Particle Kinematics*]{}, State, JohnWiley & Son Ltd, 1973, p. 102-145. G. Davidenko et al., GE781: a Monte Carlo package for fixed target experiments, in: R. Shellard and T. D. Nguyen, eds., [*Proceedings of the International Conference on Computing in High Energy Physics’95*]{} (World Scientific, Singapore, 1996) p.832,\
G. Dirkes, H. Kr$\ddot{u}$ger, N. P. Kuropatkin, J. Simon, data and GE781 comparison for elastic scattering physics, MPI Heidelberg, 1999 (Unpublished),\
V. Steiner, GE781 code development for Hadron Electron topics, SELEX Collaboration Meeting Presentations and Private communications,\
I. Giller, Measurement of the pion charge radius, M.Sc. thesis, Tel Aviv University, 1999,\
I. Eschrich et al., Phys. Lett. [**86**]{} (2001) 5243. C. Caso et al. (PDG), Eur. Phys. J. [**C3**]{} (1998) 1-794. S. Brandt, [*Data Analysis: Statistical and Computational Methods for Scientists and Engineers*]{}, Springer Verlag, 1998. J. Simon, Messung des elektromagnetischen Ladungsradius des sigma- bei $600\,\mbox{GeV/c}$ (in German), Ph.D. thesis, MPI Heidelberg, 2000,\
G. Dirkes, Messung der elektromagnetischen Formfaktoren von Pionen im SELEX Experiment (in German), M.Sc. thesis, MPI Heidelberg, 1999. K. Vorwalter, Determination of the pion charged radius with a silicon microstrip detector system, Ph.D. thesis, MPI f. Kernphysik / Univ. Heidelberg, 1998,\
H. Kr$\ddot{u}$ger, Untersuchung der elastischen Hadron-Elektron-Streuung bei 540 GeV/c zur Messung des elektromagnetischen Ladungsradius des Protons (in German), Ph.D. thesis, MPI Heidelberg, 1999. P. Mathew, Construction and evaluation of a high resolution silicon microstrip tracking detector, and utilization to determine interaction vertices, Ph.D. thesis, Carnegie Mellon University, 1997,\
J. L. Langland, Hyperon and anti-hyperon production in $p-Cu$ interactions. Ph.D. thesis, University of Iowa, 1996, UMI-96-03058,\
J. L. Langland, Hyperon beam flux parameterization for E781 based on E497 data, H-note 693, SELEX Internal Report, 1994. W. H. Press, S. A. Teukolsky, W. T. Vetterling, B. P. Flannery, [*Numerical Recipes in FORTRAN: The Art of Scientific Computing*]{}, Second edition, University of Cambridge, 1992, p. 617-622.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
Philip Baringer, Kyoungchul Kong, Mathew McCaskey, Daniel Noonan\
Department of Physics and Astronomy, University of Kansas, Lawrence, KS 66045 USA\
E-mail:
title: 'Revisiting Combinatorial Ambiguities at Hadron Colliders with $M_{T2}$'
---
Introduction {#sec:intro}
============
The Tevatron Run II at Fermilab and the recent turn-on of the Large Hadron Collider (LHC) at CERN are beginning to explore the physics of the Terascale. There are sound theoretical reasons to believe that new physics beyond the Standard Model (BSM) is going to be revealed in those experiments. Perhaps the most compelling phenomenological evidence for BSM particles and interactions at the TeV scale is provided by the dark matter problem. It is a tantalizing coincidence that a neutral, weakly interacting massive particle (WIMP) in the TeV range can explain all of the observed dark matter in the Universe. A typical WIMP does not interact in the detector and can only manifest itself as missing energy. The WIMP idea therefore greatly motivates the study of missing energy signatures at the Tevatron and the LHC. Recently, new ideas of various kinematic methods for the determination of masses, spins, and couplings have attracted a lot of interest in missing energy signatures at colliders (see [@Barr:2010zj; @Barr:2011xt; @Barr:2011im] for recent reviews). They not only offer methods for the determination of masses and spins but also provide good background rejection and enhance discovery potential.
Unlike lepton colliders, the environment at hadron colliders is much more complex and often make such tasks difficult. One issue more prevalent in hadron colliders is the combinatorial ambiguities in events, especially in events with jets. In general, the event topology at hadron colliders will typically contain a number of jets. Some come from the decays of heavier particles in the signal and others may originate from initial state radiation (ISR). The problem gets worse due to detector limitations such as finite resolution, inability of distinguishing quarks from anti-quarks, etc. All these jets pose a severe combinatorics problem: which one of the many jets in an event is the correct one to assign to a particular decay?
Some of the existing studies in the literature simply take for granted that the correct jet can be somehow identified. Others select the jet by matching to the true quark jet in the event generator output, which of course is unobservable. The severity of the jet combinatorics problem is rather model dependent and in practice how well it can be dealt with depends on the individual case at hand. For example, if the mass spectrum is relatively broad, one might expect a jet from the decay of a heavier particle to be among the hardest in the event. This information can be used to improve the purity of the sample. Fortunately, there exists a method (the mixed event technique) which should, at least in principle, remove the effects from incorrect jet combinations [@Hinchliffe:1996iu]. This method has been successfully applied to measuring SUSY masses at the SPS1a study point [@Ozturk:2007ap] (see [@Dutta:2011gs] for a more recent study). However, this only works statistically and does not discriminate on an event-by-event basis. The existence of invisible particles adds difficulty in resolving combinatorial issues since it prohibits us from reconstructing the whole signal.
Combinatorial issues appear at all different levels of an event (i.e. from ISR, decays of heavier particles, etc.). A general solution to [*all*]{} these problems would be very difficult. We may be able to achieve this statistically by relying on a likelihood analysis or matrix element methods [@Alwall:2010cq]. However, this task requires knowledge of new physics which may not be available at the time of discovery. Therefore, it is desirable to find a solution that is model-independent. In fact, some studies on kinematic variables in the literature already indicate that these variables may be useful for mitigating combinatorial difficulties associated with ISR and the signal jets [@Alwall:2009zu; @Nojiri:2010mk] [^1].
The ISR (rather than always proving to be an obstacle) can actually be helpful in the study of BSM physics. This has already been demonstrated in several recent studies. In Ref. [@Alwall:2008va], ISR was shown to make BSM events more prominent by giving pair-produced new-physics states something to recoil against, thus increasing both $\displaystyle{\not} E_T$ and $H_T$. For the $M_{T2}$ variable (defined in Section $3.1$ below) [@Lester:1999tx; @Barr:2003rg], it is known that the transverse momentum of ISR makes the kink structure more pronounced (see references in [@Barr:2010zj]). Usually ISR overestimates the expected end point, which changes depending on the size of the transverse momentum. To make better measurements one needs to understand systematically the effect of ISR as a function of transverse momentum [@Matchev:2009iw; @Burns:2008va; @Konar:2009wn; @Matchev:2009ad].
In this paper, we concentrate on combinatorial ambiguities among the particles in the signal. It is important to resolve or mitigate this combinatorial background issue. Solutions can be directly used to improve experimental measurements such as the forward-backward asymmetry in the top quark pair production, the $W$-helicity, top mass measurements, etc. In terms of BSM, they obviously become more important since most BSM models predict a WIMP candidate which, when produced, will show up as missing transverse momentum in the detector. An appropriate understanding of combinatorics may even be able to reduce SM backgrounds.
In resolving combinatorial ambiguities in signal events, recently Ref. [@Rajaraman:2010hy] argued that the proposed $p_T$ versus invariant mass method performs better than the well known hemisphere method [@Ball:2007zza]. This was shown for parton level events without detector effects, backgrounds or ISR. The results in the previous study are quoted in terms of the efficiency and purity of the samples. For the example point taken in [@Rajaraman:2010hy] (the 3 body decay of 600 GeV gluino ($\tilde{g}$) into 2 jets and a neutralino ($\tilde{\chi}_1^0$) of 100 GeV), the authors have obtained 91.7% purity and 3.1% efficiency for the diquark system with an invariant mass less than 500 GeV and a transverse momentum greater than 450 GeV.
The purpose of this paper to revisit the proposed method in [@Rajaraman:2010hy] and improve the obtained efficiency and purity by utilizing the $M_{T2}$ variable in addition to invariant mass. Section \[sec:pTMmethod\] begins with a review of the method described in [@Rajaraman:2010hy]. In Section \[sec:gg\] we propose our new method for resolving combinatorial ambiguities in the context of gluino pair production with 4 jets $+{\not \!\! E_T}$ in the final state. Following the spirit of Ref. [@Rajaraman:2010hy] we analyze our method under similar assumptions in order to make a suitable comparison. Then we determine the effectiveness of this method by considering different mass splittings, possible uncertainties in the gluino and neutralino masses, and ISR. Next, in Section \[sec:ttbar\], we apply the same method to the $t\bar{t}$ dilepton system to resolve the two-fold ambiguity in pairing a b-tagged jet ($b$) and a lepton ($\ell$). Section \[sec:conclusions\] is reserved for our conclusions.
The $p_T$ versus $M$ method {#sec:pTMmethod}
===========================
In this section we briefly review the $p_T$ versus $M$ method in Ref. [@Rajaraman:2010hy]. For the discussion on combinatorial ambiguities at hadron colliders, the authors consider gluino pair production with each gluino decaying to the lightest neutralino and two quarks: $\tilde g \tilde g \to q q \bar{q} \bar{q} \tilde{\chi}_1^0 \tilde{\chi}_1^0$. All quarks are treated at parton level and the effects of ISR and parton showering are not considered. Following the example used in [@Rajaraman:2010hy], we consider the off-shell squark case ($m_{\tilde{q}} > m_{\tilde{g}}$) with a gluino mass $m_{\tilde{g}} = 600$ GeV and a neutralino mass $m_{\tilde{\chi}_{1}^{0}} = 100$ GeV (the on-shell case is similar).
Without any ISR, we are left with $4$ jets in the final state. In order to do any proper analysis, it needs to be determined which of the $3$ possible pairings of jets is the correct combination. In Ref. [@Rajaraman:2010hy], the authors explored the possibility of extracting the correct jet combination using the invariant mass and transverse momentum of each diquark system. They noticed two prominent features. First, they noted the kinematic edge in the diquark invariant mass distribution, which is especially visible for combinations with high $p_T$. The excess of diquark combinations with an invariant mass larger than the kinematic edge value ($m_{\tilde{g}}-m_{\tilde{\chi}_{1}^{0}}=500$ GeV) must all be incorrect combinations. Second, those combinations which have invariant mass larger than the kinematic edge (which are all incorrect combinations) tail off quickly towards larger diquark $p_T$. Events below the edge (which are a mixture of correct and incorrect combinations) extend to higher $p_T$. Therefore, for the events with larger $p_T$, the ratio of correct to incorrect combinations increases.
![The invariant mass (a) and $p_{T}$ distributions (b) of the jet combinations in the final state of the gluino pair production. (c) Diquark transverse momentum versus invariant mass for the correct (in black) and incorrect (in yellow) jet combinations. The vertical and horizontal lines represent the cuts used to determine the correct jet combination. These distribution show the motivation for the cuts proposed in [@Rajaraman:2010hy]. []{data-label="fig:gluino_pt_inv"}](FIGURES/gluino_inv_new.eps "fig:"){width="32.00000%"} ![The invariant mass (a) and $p_{T}$ distributions (b) of the jet combinations in the final state of the gluino pair production. (c) Diquark transverse momentum versus invariant mass for the correct (in black) and incorrect (in yellow) jet combinations. The vertical and horizontal lines represent the cuts used to determine the correct jet combination. These distribution show the motivation for the cuts proposed in [@Rajaraman:2010hy]. []{data-label="fig:gluino_pt_inv"}](FIGURES/gluino_pT_new.eps "fig:"){width="32.00000%"} ![The invariant mass (a) and $p_{T}$ distributions (b) of the jet combinations in the final state of the gluino pair production. (c) Diquark transverse momentum versus invariant mass for the correct (in black) and incorrect (in yellow) jet combinations. The vertical and horizontal lines represent the cuts used to determine the correct jet combination. These distribution show the motivation for the cuts proposed in [@Rajaraman:2010hy]. []{data-label="fig:gluino_pt_inv"}](FIGURES/pt_invmass_new.eps "fig:"){width="32.00000%"}
To reproduce the results of Ref. [@Rajaraman:2010hy] we use the MadGraph/MadEvent package [@Alwall:2011uj] to generate a total of $10$K events at the LHC with a center of mass energy of $7$ TeV. Simulating the effects caused by detector resolution in the hadronic calorimeter, the $4$-momenta of the final state jets are smeared via $$\left(\frac{\delta E}{E}\right)^{2} = \frac{a^{2}}{E} + b^{2} \, ,
\label{eq:LHCres}$$ where $a = 0.5~(0.1)~\sqrt{\text{GeV}}$ and $b = 0.03~(0.007)$ for jets (leptons) [@RichterWas:2009wx; @Bayatian:2006zz; @Ball:2007zza].
Fig. \[fig:gluino\_pt\_inv\] shows (a) the $p_{T}$ and (b) invariant mass distributions for the correct (blue, solid), incorrect (red, dashed) and all (black, dot-dashed) combinations of jets. The different normalizations of the three histograms are understood since there are two incorrect pairs and one correct pair in each event. In the invariant mass distribution of the correct jet combination there is a clear cutoff at $m_{\tilde{g}}-m_{\tilde{\chi}_{1}^{0}}=500$ GeV. A small fraction of events beyond this kinematic edge occur due to detector resolution. In the $p_{T}$ distribution at high value ($p_{T} \gtrsim$ 500 GeV) the ratio of correct to incorrect jet combinations becomes $\gtrsim 1$. The $p_{T}$ cut is chosen such that only $5\%$ of the combinations with an invariant mass above the kinematic edge (which are likely to be incorrect combinations) survive. As suggested, a cut on combinations with the invariant mass below the kinematic edge and above the $p_T$ cut can guarantee an event sample dominated by correct combinations. To extract the correct jet combination the following cuts are made:
- [The invariant mass of the diquark pair is less than $500$ GeV.]{}
- [The $p_T$ of the diquark pair is greater than $450$ GeV.]{}
If only one combination of jets passes these cuts then the event is accepted, otherwise the event is discarded. Fig. \[fig:gluino\_pt\_inv\](c) shows the diquark transverse momentum versus invariant mass of the correct (black) and incorrect (yellow) jet combinations. The vertical and horizontal lines represent expected cuts to reduce the number of incorrect combinations. Counting those events that pass two cuts ([*i.e.,*]{} the number of events in the left-upper corner of Fig. \[fig:gluino\_pt\_inv\](c)), we find an event efficiency of $3.1\%$ with an event purity of $91.7\%$. This is consistent with the results in Ref. [@Rajaraman:2010hy]. Furthermore, a higher efficiency can be achieved at the cost of lower purity by imposing a lower $p_T$ cut. It is also shown in [@Rajaraman:2010hy] that this method performs better than the hemisphere method.
Improved method {#sec:gg}
===============
We note that the method discussed in the previous section has a low efficiency while the corresponding purity is relatively high. This is due to the long tail in the $p_T$ distribution of the diquark system. In general, for a large mass splitting between the mother and daughter particles, the $p_T$ of the resulting diquark system is relatively large. Conversely, it is smaller for a narrower spectrum, [*i.e.,*]{} the peak position in the $p_T$ distribution is shifted to a lower value in a narrower mass spectrum. Therefore there is a correlation between the location of the peak in the distribution and the mass difference. However the amount of shift (or the location of the peak) is not well enough predicted quantitatively for this observation to be useful, since the diquark system inherits some kinematic information of the mother particles at the production level. An ISR jet is also problematic since the mother particle system will be boosted in the opposite direction of the ISR resulting in modification of the $p_T$ spectra. Unlike the invariant mass distribution, the $p_T$ distribution does not exhibit any kinematic edge. The distribution of the background (incorrect combinations) has a similar shape and, as argued, it decreases faster than the distribution of the correct combination above a certain value (Fig. \[fig:gluino\_pt\_inv\](b)). On the other hand, the invariant mass distribution shows a clear kinematic edge in the distribution of the correct combination but not in that of the incorrect combination, making the two distributions distinct (Fig. \[fig:gluino\_pt\_inv\](a)).
One can recall a similar situation in the leptonic decay of the $W$ boson at hadron colliders. The mass of the $W$ boson in principle can be measured from the $p_T$ distribution of a decay product (a lepton). It shows a nice kinematic edge at half of the $W$ mass when the $W$ boson is produced at rest. However, in reality the transverse mass is used instead. It shows a kinematic edge at the mass of the $W$ and the value of the end point is stable under the presence of initial state radiation. What variable would do such a job for the example in the previous section while increasing the event efficiency? In this paper, instead of cutting on $p_T$, we propose to use $M_{T2}$ as the second cut in addition to the invariant mass. It is the most natural extension of the transverse mass in the final state when more than one missing particle is present. As advertised, it shows a clear kinematic edge and is stable ([*i.e.,*]{} keeping the end point at the same value) under a boost of the whole system. One should investigate how well it performs in terms of the event efficiency and purity.
We consider the same process ($4$ jets and $2$ neutralinos) as in the previous section without any ISR and discuss a method to improve the event efficiency and purity. Then, we discuss how our results change with different mass spectra and the effect of uncertainty in mass measurements. Finally, we look at the same process with ISR and discuss how our results change if we are correctly able to identify the ISR jet. First, we begin this section with a review of $M_{T2}$.
Brief review of $M_{T2}$ {#sec:review}
------------------------
The traditional $M_{T2}$ variable is defined as follows [@Lester:1999tx; @Barr:2003rg]. Consider a pair production of a mother particle, $P$, and their decays down to a daughter particle of mass $M_d$. For any given event, one can construct the transverse mass, $M_{iT}$, of each parent: $$M_{iT}^2\equiv m_i^2+M_d^2+2(E_{iT} E_{iT}^d-\vec{p}_{iT}\cdot\vec{p}^{\, d}_{iT})\, ,
\label{MTparent}$$ where $$E_{iT} \equiv \sqrt{m_i^2+|\vec p_{iT}|^2}, \quad
E_{iT}^d \equiv \sqrt{M_d^2+|\vec p_{iT}^{\,d}|^2},
\label{E}$$ is the transverse energy of the visible particles, $V_i$, and daughter particle, $d$, in the decay branch of each mother particle, correspondingly. $\vec p_{iT}$ and $m_i$ are the transverse momentum and invariant mass of the visible particles in branch $i$ ($i=1,2$). The individual momenta, $\vec p_{iT}^{\,d}$, of the missing daughter particles, $d$, are unknown, but they are constrained by the measured missing transverse momentum, ${\not \!\! \vec{P}_T}$, in the event: $$\vec p_{1T}^{\,d} + \vec p_{2T}^{\,d} = {\not \!\! \vec{P}_T}\equiv - \vec{P}_T - \vec p_{1T}- \vec p_{2T}.
\label{PTmiss}$$ Here $\vec{P}_T$ is the Upstream Transverse Momentum (UTM), which is the transverse momentum of all other visible particles that are not considered in the decay chains. For the true values of the missing momenta, $\vec p_{iT}^{\,d}$, each transverse mass in Eq. \[MTparent\] is bounded above by the true parent mass, $M_P$.
This fact can be used in a rather ingenious way to define the Cambridge $M_{T2}$ variable [@Lester:1999tx; @Barr:2003rg]. One takes the larger of the two quantities in Eq. \[MTparent\] and minimizes it over all possible partitions of the unknown daughter momenta, $\vec p_{iT}^{\,d}$, subject to the constraint in Eq. \[PTmiss\]: $$M_{T2}
(\tilde M_d, \vec{P}_T, \vec{p}_{i T})
\equiv
\min_{\vec p_{1T}^{\,d} + \vec p_{2T}^{\,d} = {\not \!\! \vec{P}_T}}
\left\{\max\left\{M_{1T},M_{2T}\right\} \right\}\ .
\label{eq:mt2def}$$ Since the masses of the mother and daughter particles are not known in advance, one should treat $M_{T2}$ as a function of the trial mass ($\tilde M_d$) of the daughter particle. For a given $P_T$, the endpoint $M_{T2}^{max}$ of this distribution gives the parent mass, $\tilde M_P$, as a function of the input trial daughter mass, $\tilde M_d$. An important property of $M_{T2}$ is $$\tilde{M}_P ( \tilde{M}_d = M_d) = M_{T2}^{max} ( \tilde{M}_d = M_d) = M_P \, .$$ Then one is able to obtain a one dimensional relation between two masses. In cases with non-zero UTM and composite visible system, a kink structure appears, allowing simultaneous determination of both masses (see [@Barr:2010zj] and references therein).
In the case of $\vec{P}_T = 0$, the analytic expression for $M_{T2}$ can be obtained, while for $\vec{P}_T \neq 0$ case there is no analytic solution to minimization and one has to rely on a numerical code for computation. A clever trick to get around this has been proposed in Ref. [@Konar:2009wn]. The idea is to project further the transverse quantities with respect to the non-zero UTM, in which case the momentum conservation in the direction perpendicular to the UTM is independent of the UTM (by construction) and one is able to use an analytic expression for the computation of projected $M_{T2}$. $M_{T2}$ has been extended to cases with non-identical mothers [@Barr:2009jv] and non-identical daughters [@Konar:2009qr]. In principle, one can use it for an associated production as well as processes with two different daughter particles.
Case without ISR
----------------
As an alternative method, instead of using transverse momentum, one uses $M_{T2}$ to help determine the correct jet combination. Fig. \[fig:gluino\_mt2\_inv\](a) shows the $M_{T2}$ distribution for the correct (blue, solid), incorrect (red, dashed) and all (black, dot-dashed) combinations of jets. Note that the normalization of the distribution of incorrect (all) pairs is twice (three times) higher than that of the correct distribution. The distribution for the correct jet combination has a clear cutoff at the gluino mass ($m_{\tilde{g}}=600$ GeV). We apply the same cut that was used in the $p_T$ versus $M$ method for the invariant mass of the jet combinations but now we also require require $M_{T2} < 600$ GeV. As before, we keep the events where only one combination of jets passes both cuts. Fig. \[fig:gluino\_mt2\_inv\](b) shows the correct (black) and incorrect combinations (yellow) in $M_{T2}$ versus invariant mass. For events where only one combination of jets pass the two cuts, we obtain an event efficiency of $20\%$ with an event purity of $95\%$ as shown in Table \[tab:mt2\_combos\]. Thus we have a concrete example showing that an appropriate choice of kinematic variable cuts enhances both the efficiency and purity of our signal.
![(a) The $M_{T2}$ distribution of the correct and incorrect combinations of jets in the gluino pair production. For the correct jet combination a clear cutoff appears at the gluino mass. (b) The $M_{T2}$ versus invariant mass for the correct (in black) and incorrect (in yellow) jet combinations. The vertical and horizontal lines represent the cuts used to determine the correct jet combination. []{data-label="fig:gluino_mt2_inv"}](FIGURES/gluino_mT2_new.eps "fig:"){width="49.00000%"} ![(a) The $M_{T2}$ distribution of the correct and incorrect combinations of jets in the gluino pair production. For the correct jet combination a clear cutoff appears at the gluino mass. (b) The $M_{T2}$ versus invariant mass for the correct (in black) and incorrect (in yellow) jet combinations. The vertical and horizontal lines represent the cuts used to determine the correct jet combination. []{data-label="fig:gluino_mt2_inv"}](FIGURES/mt2_invmass_new.eps "fig:"){width="49.00000%"}
Unlike the cuts using $p_{T}$, a very small fraction of events with correct combinations are ruled out completely by our cuts ([*i.e.*]{}, many events have $2$ or $3$ combinations that pass the $M_{T2}$ and invariant mass cuts). Table \[tab:mt2\_combos\] shows the number of events passing our cuts broken down by the number of combinations that pass. With these cuts only $2\%$ of the correct combinations are discarded and 98% of correct combinations survive, so it is possible to use more refined cuts to extract the correct combination from the events with $2$ and $3$ combinations which could potentially increase the event efficiency even above $20\%$. In principle, one can apply a $p_{T}$ cut at this point. However, as argued, $p_{T}$ is dependent on the mass spectrum and ISR, and perhaps it should be the last variable to make a cut on. Instead, we further increase the efficiency and purity of this procedure by simply taking a look at the events where only two combinations pass the cuts so far.
Making further cuts allows us to extract the correct combination from the two that are left over. We discuss two such ways of extracting the correct combination using the invariant mass and $M_{T2}$ distributions. In Figs. \[fig:gluino\_pt\_inv\](a) and \[fig:gluino\_mt2\_inv\](a) we can see that for invariant masses and $M_{T2}$ that are below the kinematic cutoff the correct combination tends to be closer to the cutoff than the incorrect combination. We can take advantage of this by introducing a second cutoff. In the case of the $M_{T2}$ distribution we require that one combination have an $M_{T2}$ above this cutoff and the other is below. We then take the combination that is above this cutoff as the correct one. To select the cutoff we maximize the sensitivity $\epsilon D^2$, where the $\epsilon$ is the efficiency, $D= 2P-1$ is the dilution, and $P$ is the purity. This is a commonly used method for optimization in such situations [@YenChu]. Fig. \[fig:2ndcut\](a) shows that the maximum sensitivity occurs at an $M_{T2}$ cutoff at $460$ GeV which would give us an overall combined efficiency of $33\%$ and purity of $84\%$. In the case of the invariant mass distribution, each combination has a pair of invariant masses so the cuts need to be modified slightly. From the two jet combinations we take the pair of invariant masses that have the largest separation between them. Placing a cut on the invariant mass we require that one invariant mass is above and the other is below the cutoff. We then take the combination with the invariant mass pair above the cutoff as the correct combination. Fig. \[fig:2ndcut\](b) shows the scan of this invariant mass cutoff. We see that a cutoff of $230$ GeV gives the largest sensitivity to this set of events. Combined with the results from the first set of cuts we obtain a total event efficiency of $41\%$ and a purity of $79\%$.
![The sensitivity (solid, black), efficiency (blue, dot-dashed) and purity (red, dashed) as functions of the (a) $M_{T2}$ and (b) invariant mass cutoff. The cutoffs are used to pick the correct combination out of the events where only two combinations passed the original cuts. We see that a $M_{T2}$ cut (invariant mass cut) of $460$ GeV ($230$ GeV) produces the largest sensitivity. Combining the results of the first set of cuts we have an overall efficiency of $33\%$ and a purity of $84\%$ ($41\%$ and a purity of $79\%$) with additional $M_{T2}$ cut (invariant mass cut). []{data-label="fig:2ndcut"}](FIGURES/mt2_2ndcut.eps "fig:"){width="48.00000%"} ![The sensitivity (solid, black), efficiency (blue, dot-dashed) and purity (red, dashed) as functions of the (a) $M_{T2}$ and (b) invariant mass cutoff. The cutoffs are used to pick the correct combination out of the events where only two combinations passed the original cuts. We see that a $M_{T2}$ cut (invariant mass cut) of $460$ GeV ($230$ GeV) produces the largest sensitivity. Combining the results of the first set of cuts we have an overall efficiency of $33\%$ and a purity of $84\%$ ($41\%$ and a purity of $79\%$) with additional $M_{T2}$ cut (invariant mass cut). []{data-label="fig:2ndcut"}](FIGURES/inv_2ndcut.eps "fig:"){width="48.00000%"}
Uncertainty in particle masses
------------------------------
Up to this point we have assumed the exact masses of the gluino and the lightest neutralino. What if the masses are not known with absolute certainty? To investigate the effects due to an uncertainty in mass measurements on the event efficiency and purity in finding the correct pairs, we generate events where both the masses of the gluino and neutralino fluctuate by $10\%$. Then using exactly the same cuts as before ($m_{jj} < 500$ GeV and $M_{T2} < $ 600 GeV for all mass spectra) we compute the event efficiency and purity, as shown in Table \[tab:mass\_uncertainty\].
A $10\%$ increase in the gluino mass in general increases the invariant mass and $M_{T2}$ of the jet pairs, effectively reducing the probability that the jet pair will pass the cuts. In other words, some events with all three combinations surviving cuts end up having only one or two combinations filtered by this variation. This results in an increasing event efficiency, but the contamination lowers the event purity. On the other hand a gluino mass that is $10\%$ smaller decreases the invariant mass and $M_{T2}$ of the jet pairs, which increases the probability that a jet combination will pass all the cuts. This will lower the event efficiency but will raise the event purity due to the fact that in events with only one combination passing all the cuts, that combination is more likely to be the correct one.
The uncertainty in the neutralino mass produces the opposite effect on the event efficiency and purity. Because the neutralino is a daughter particle, increasing (decreasing) the neutralino mass will increase (decrease) the probability of a jet combination passing our cuts producing the opposite effect to the event efficiency and purity than a similar change in gluino mass. However, it can be seen that the effect on the event efficiency and purity is much milder than that of the gluino due to the smaller change in neutralino mass. Therefore, overestimating the mass of the mother particle (or underestimating the mass of daughter particle) decreases efficiency and increases purity, while underestimating the mother particle mass increases efficiency and decreases purity.
Dependence on mass spectrum
---------------------------
In principle, this method does not require a large mass splitting between the gluino and neutralino in order to be effective. To quantify this we scan the gluino and neutralino mass over a range of values. Assuming that the gluino and neutralino masses are known, we take the following cuts on the invariant mass and $M_{T2}$: $$\begin{aligned}
m_{jj} &<& m_{\tilde{g}}-m_{\tilde{\chi}^{0}_{1}} \, , \label{eq:cut1} \\
M_{T2} &<& m_{\tilde{g}} \, . \label{eq:cut2}\end{aligned}$$ The results are summarized in Table \[tab:purity\].
In the increasing $m_{\tilde{g}}$ direction, the invariant mass and $M_{T2}$ distributions of the correct and incorrect jet combinations becomes more spread apart. In general this would increase the chance that a certain jet combination will be cut, reducing the event efficiency. However, the events that survive with only one passed combination will more likely be the correct combination, leading to a larger event purity. In the increasing $m_{\tilde{\chi}_{1}^{0}}$ direction, the invariant mass and $M_{T2}$ distributions of the correct and incorrect jet combinations becomes more compressed, resulting in higher efficiency and lower purity.
Effect of ISR
-------------
Since gluino pair production gives a multi-jet final state, it is important to address an issue with isolating ISR jets. The ability to identify ISR is essential for using the discussed procedure: if the jets from ISR are not correctly isolated then it would be impossible to find the correct jet combination. Several different methods have been proposed to identify an ISR jet [@Alwall:2010cq; @Alwall:2009zu; @Nojiri:2010mk; @Konar:2008ei; @Konar:2010ma; @Krohn:2011zp; @Alwall:2007fs; @Plehn:2005cq]. To analyze the gluino pair production with a single ISR jet we concentrate on the procedure introduced in Ref. [@Alwall:2009zu]. In order to identify the ISR jet we use the following procedure.
1. [The jets with the highest momentum are not considered to be the ISR jet and are automatically placed in different gluino decay chains.]{}
2. [For the three other jets, each one is taken out and $M_{T2}$ is calculated from all the possible combinations of the remaining jets. The minimum $M_{T2}$ of the possible jet combinations is identified as $M_{T2}^{(i)}$.]{}
3. [The jet associated with the minimum of the $M_{T2}^{(i)}$’s is taken to be the ISR jet.]{}
Using this procedure alone we find that the ISR jet can be correctly identified $24\%$ of the time. This number can be improved slightly by putting a minimum cutoff for the $M_{T2}$ of min$(M_{T2}) > 500$ GeV. In general, if a correct jet is taken out then the resulting $M_{T2}$ becomes much smaller so this minimum $M_{T2}$ cutoff is designed to prevent the small $M_{T2}$ that arises from taking out a hard jet. With this minimum cut included we find that the correct ISR jet can be identified $36\%$ of the time. Both of these numbers are consistent with the results given in Ref. [@Alwall:2009zu]. If we look at the events that have correctly identified the ISR jet and apply the same cuts as before to isolate the correct combination of jets, we can see how our procedure fares with the addition of ISR. The results are given in Table \[tab:mt2\_combos\_isr\]. We find that for the events where the ISR was identified without the $M_{T2}$ minimum cutoff, the efficiency of finding the correct combination is $19\%$ with a purity of $92\%$. Including the minimum $M_{T2}$ cut the efficiency is found to be $16\%$ also with a purity of $92\%$. Though the efficiency drops slightly there are more events in this group where the correct ISR jet is identified. It is very important to note that these results were obtained from the events where the ISR jet was correctly identified. If we were to include the events where the incorrect ISR jet was identified then our purity will suffer. This exemplifies the importance of correctly identifying the ISR jet in this process.
Application: the $b$-$\ell$ ambiguity in the $t \bar{t}$ dilepton channel {#sec:ttbar}
=========================================================================
In this section, we apply the same idea as in Section \[sec:gg\] to resolve the two fold ambiguity that arises in $t\bar{t}$ production in the dilepton channel at the Tevatron and the LHC. Since there are two leptons associated in this analysis, the corresponding background is already small. Requiring one or two b-tagged jets makes the background negligible. As an illustration, we estimated the cross sections for the signal and SM backgrounds at the 7 TeV LHC. We have used $m_t=173$ GeV. Assuming the cross sections summarized in Table \[table:xsections\], we simulate the signal ($t\bar{t}$) and background events with PYTHIA [@Sjostrand:2006za] and further process them with PGS [@PGS] for detector effects. We require two leptons, at least two jets, and missing transverse energy. The default cuts in each MC event generator are applied along with an extra dilepton invariant mass cut, 40 GeV $<$ $m_{\ell\ell}$ $<$ 200 GeV. Table \[table:nevents\] summarizes the number of signal (S) and background (B) events in each channel for zero, one, and two b-tagged jets. The background cross sections are reduced significantly by requiring at least two leptons ($S0/B0 \sim 3.27$) and b-tagging further reduces the backgrounds ($S1/B1 \sim 24$ for one b-tagged jet and $S2/B2 \sim 75$ for two b-tagged jets). Although this conclusion with a naive estimation of the background cross section may be affected slightly due to systematic uncertainties and higher order corrections, we do not expect significantly different results in this two-lepton and one or two $b$-tagged final state. Therefore we will not address background issues in the rest of this section and will assume the background is negligible.
For the analysis in the rest of this section, we generate 10K $t\bar{t}$ events using the MadGraph/MadEvent package at both the Tevatron and the 7 TeV LHC. We force the $W$-boson to decay leptonically into either an electron or muon. For events simulated at the LHC, detector resolution was simulated as in Eq. \[eq:LHCres\]. For the Tevatron, the detector resolution is simulated following the parametrization in PGS [@PGS], $$\begin{aligned}
\frac{\delta E}{E} &=& \frac{0.8}{ \sqrt{E} } ~~~ ~~~~~~~~~{\rm for~jets}\, , \\
\frac{\delta E}{E} &=& \frac{0.2}{ \sqrt{E}} +0.01 ~~~ {\rm for~leptons}\, .
\label{eq:Tevatronres}\end{aligned}$$ We also assume that the ISR is correctly isolated. This can be justified given that we require two $b$-tagged jets for our study on combinatorial issues. To separate the correct and incorrect combinations, we use the same method using the invariant mass of each $b$-$\ell$ pairing as well as the $M_{T2}$ variable. The distributions for the correct and incorrect combinations for $M_{T2}$ are shown in Fig. \[fig:mt2\_ttbar\] for both Tevatron (in (a)) and LHC (in (b)). The correct combinations have a cutoff at $m_{top}$, while the incorrect combinations have a tail extending beyond this cutoff. Fig. \[fig:invmass\_ttbar\] shows the distributions for $m_{b\ell}$, the invariant mass of the $b$-jet lepton pairings at the Tevatron (in (a)) and the LHC (in (b)). Both show a kinematic edge at $\sqrt{m_{top}^2 - m_W^2} \approx 153$ GeV.
Specific cut values, $M_{T2} <$ 176 GeV and $m_{b\ell} <$ 156 GeV, are found by maximizing the sensitivity, defined as $S = \epsilon {\left( 2 P - 1 \right)}^2 $. Events are selected if only one of the combinations passes these cuts. The efficiency ($\epsilon$) determines the fraction of $t\bar{t}$ events passing this criterion and the purity ($P$) is defined as the fraction of lepton-jet pairs that are correctly determined. One can optimize these cuts differently, depending on whether higher purity (but lower efficiency) or higher efficiency (but lower purity) is desired. Table \[tab:ttbar\_results\_LHC\] (Table \[tab:ttbar\_results\_Tevatron\]) shows the results at the LHC (Tevatron) broken down by the number of combinations that pass the invariant mass and $M_{T2}$ cuts in each event. When the restriction that only one combination passes the cuts is applied, we obtain an efficiency of 51.7% (39.9%) and a purity of 94.9% (91.9%).
The CDF collaboration uses several methods to correctly pair the $b$-jet and the lepton [@YenChu]: [$m_{b\ell}^{max}$]{}, the kinematic method, and the neutrino phi weighting method. Let us explain each method briefly.
[*[$m_{b\ell}^{max}$]{}method*]{}: Since there are two leptons and two $b$-jets in each event, there are two exclusive pairing options so there are in total four $m_{b\ell}^2$ values. One takes the pairing option having the maximum $m_{b\ell}^2$ as the incorrect one. As shown in Fig. \[fig:invmass\_ttbar\], above a certain value, larger $m_{b\ell}^2$ comes often from the incorrect combination. To improve the purity of pairing, the events that have maximum $m_{b\ell}^2$ larger than a certain cut value are selected. A larger cut value improves the purity at the cost of reduced efficiency. In order to find the best cut value (leading to maximum sensitivity, $\epsilon (2P-1)^2$) on [$m_{b\ell}^{max}$]{}, one can scan through various maximum invariant mass cut values and select the events having maximum invariant mass larger than the cut. The efficiency and purity are calculated for each cut value. [*Kinematic method*]{}: the kinematic method is a variation of the method used in the top mass analysis in the dilepton channel. This method is based on resolving the full energy-momentum conservation equation set for all the particles in the final sates. By a process of eliminating variables one is left with a 4th-order polynomial equation. Therefore up to 4 real solutions are possible. The CDF collaboration uses a numerical method to solve equations and look for up to two solutions. If there are two solutions found, they choose the one having smaller $m_{t\bar{t}}$. A variation of this method has been used in studies of spin correlation of the $t\bar{t}$ system [@CDF9824]. The kinematic method has slightly higher efficiency than the [$m_{b\ell}^{max}$]{}method but its purity is significantly lower [@YenChu]. Performance of the neutrino phi weighting method is similar to the kinematic method [@YenChu]. The matrix element method can be also used but we do not find any comments on the pairing purity in the literature. As a comparison with our method, we have considered the [$m_{b\ell}^{max}$]{}method (denoted as ($m_{b\ell}$, none) in Figs. \[fig:CDFsensitivity\] and \[fig:combinedsensitivity\], and Tables \[tab:CDF-Tevatron\] and \[tab:CDF-LHC\]) used by the CDF collaboration. Fig. \[fig:CDFsensitivity\] shows the purity-efficiency relation with the sensitivity at the Tevatron (in (a)) and the LHC (in (b)). This one-dimensional relation is due to a degree of freedom in choosing the invariant mass cut. The original CDF method is denoted as ($m_{b\ell}$, none) (solid, red) while ($M_{T2}$, none) (dashed, black) represent the same scheme with $m_{b\ell}$ replaced by $M_{T2}$. At the Tevatron when a cut of $m_{b\ell} >$ 151 GeV is made, the maximum sensitivity is obtained and this method results in an efficiency of 37.2% and a purity of 96.3% for top quark production. The corresponding cuts for ($m_{b\ell}$, none) are are shown in Table \[tab:CDF-Tevatron\]. These are comparable to results obtained with our proposal. For the LHC, both purity and efficiency are higher (see Fig. \[fig:CDFsensitivity\](b) and Table \[tab:CDF-LHC\]).
While the initial requirement of only one combination passing the cuts for our method results in reasonable values for efficiency, we can still improve this by making use of the events in which both combinations pass the cuts (nearly 50 percent of the events for the LHC). It is possible to implement another reconstruction method on these events. One example would be to combine our method with the [$m_{b\ell}^{max}$]{}method to improve the total number of events in which a good combination can be selected, while not causing too large of an adverse effect on the purity. In this case, however, we do not find a noticeable improvement over the original method.
Instead, we take a different approach to increase the event efficiency and purity. The CDF procedure cuts out events in which the combination with the highest $m_{b\ell}$ or $M_{T2}$ is not over the cut value, thus giving a lower efficiency. We look at those events and apply a second cut on the variable not used in the first case, thus applying the method with $M_{T2}$ for events not passing the $m_{b\ell}$ selection, ($m_{b\ell}$, $M_{T2}$), and vice versa, ($M_{T2}$, $m_{b\ell}$). We find that the $M_{T2}$ and invariant mass together with the CDF scheme lead to much better efficiency and purity than those obtained with one of them only. Fig. \[fig:combinedsensitivity\] shows contours of the efficiency (red, dashed), purity (blue, dotted) and sensitivity (black, solid) in the $M_{T2}$-$m_{b\ell}$ plane at the Tevatron (a) and the LHC (b). The cuts that correspond to the maximum sensitivity are shown in Table \[tab:CDF-LHC\]. Similar results are obtained for both ($m_{b\ell}$, $M_{T2}$) and ($M_{T2}$, $m_{b\ell}$) cases, and this shows the order of the two cuts is not crucial. With both kinematic cuts, we improve the efficiency by 20% while keeping the purity at the same level or slightly higher. While in this section we have discussed the combinatorial ambiguity in the $b$-$\ell$ pairings in the dilepton channel, the same method can be used in principle in the semi-leptonic channel as well. In the latter case, often a $\chi^2$ (or a likelihood function) is defined with mass-shell constraints of the top quark and the $W$ boson on the both hadronic and leptonic sides, assuming the missing transverse momentum is solely due to the neutrino from the $W$ decay. In the procedure, from the on-shell condition of the leptonic $W$, an additional two-fold ambiguity in sign for the $z$-component (along the beam direction) of the neutrino momentum is necessarily introduced. The correct combination and sign are determined by minimizing the $\chi^2$ (or maximizing the likelihood function). The kinematic variables such as $M_{T2}$ and invariant mass should be able to help to increase the probability of finding the correct combination, when they are properly included in the $\chi^2$ or the likelihood functions.
We have assumed both $b$-jets are tagged but the proposed method here should apply in the final state with one $b$-tagged jet as well as in the dilepton channel with no $b$-tagged jets. In these cases, the candidates for the $b$-jet can be determined by different selection criteria. For instance, in the final state with one $b$-tagged jet, the hardest remaining jet could be a good candidate for the other $b$-jet while for the final state with no $b$-tagged jets, the two hardest jets are good choices. However the purity of the $t\bar{t}$ sample in the latter case will decrease significantly.
Summary and conclusions {#sec:conclusions}
=======================
The $M_{T2}$ variable is originally introduced to measure masses of semi-invisibly decaying particles. It is the most natural extension to the transverse mass that was used for the $W$ discovery and mass measurement. For the last few years, there have been many studies to determine masses, spins and couplings. Most of them utilize the kinematics of events with missing energy. The $M_{T2}$ variable is especially useful for relatively short decay chains where a traditional invariant mass method is limited. The subsystem $M_{T2}$ allows for measurement of all masses in the decay chain, assuming combinatorial issues are sorted out. It is also shown that the $M_{T2}$ variable shows an interesting kink structure due to the compositeness of the visible sector, the existence of ISR, or upstream transverse momentum from the heavier particle decays. It has been extended to include different mother/daughter particles masses. Finally it has been proposed as a potential solution for isolating ISR from signals with multiple jets.
In this paper we have investigated the feasibility of the use of $M_{T2}$ to resolve combinatorial issues at hadron colliders. We concentrated on a 4-jets signal where 3 possible partitions exist. We have compared the performance of this method with the $p_T$ versus invariant mass method and showed that efficiency increased up to a factor of 5 for the same purity of the sample. One of the advantages of the kinematic methods is that there is little dependence on the mass spectrum (model dependence), unlike a choice of $p_{T}$. We have found that our results for efficiency and purity remain similar for different choices of mass parameters. We showed that the kinematic variable suggested for mass determination is also useful for resolving combinatorial issues in signals, as well as for isolating ISR jets from signal jets.
A similar idea is applied to the $t\bar{t}$ dilepton system and we found that the obtained efficiency and purity are comparable to those using the current CDF method. A variation of their method with $M_{T2}$ results in improvement of the efficiency by 20% at the Tevatron and 25% at the LHC, while the corresponding purity remains the same. It is desirable and important for experimental collaborations to revisit the method with a full detector simulation.
We thank Yen-Chu Chen for helpful discussion and providing relevant references regarding $b$-$\ell$ ambiguity in the $t\bar{t}$ production. This work is supported in part by US Department of Energy grants DE-FG02-04ER14308 and by National Science Foundation grants PHY-0653250. KK is supported partially by the National Science Foundation under Award No. EPS-0903806 and matching funds from the State of Kansas through the Kansas Technology Enterprise Corporation. KK acknowledges the hospitality of TASI-2011 at the University of Colorado, where some of this work was undertaken.
[999]{}
A. J. Barr, C. G. Lester, “A Review of the Mass Measurement Techniques proposed for the Large Hadron Collider,” J. Phys. G [**G37**]{}, 123001 (2010). \[arXiv:1004.2732 \[hep-ph\]\].
A. J. Barr, T. J. Khoo, P. Konar, K. Kong, C. G. Lester, K. T. Matchev, M. Park, “Guide to transverse projections and mass-constraining variables,” \[arXiv:1105.2977 \[hep-ph\]\].
A. J. Barr, T. J. Khoo, P. Konar, K. Kong, C. G. Lester, K. T. Matchev, M. Park, “A Storm in a ‘T’ Cup,” \[arXiv:1108.5182 \[hep-ph\]\].
I. Hinchliffe, F. E. Paige, M. D. Shapiro, J. Soderqvist and W. Yao, “Precision SUSY measurements at LHC,” Phys. Rev. D [**55**]{}, 5520 (1997). \[arXiv:hep-ph/9610544\]. N. Ozturk \[ATLAS Collaboration\], “SUSY Parameters Determination with ATLAS,” arXiv:0710.4546 \[hep-ph\].
B. Dutta, T. Kamon, N. Kolev, A. Krislock, “Bi-Event Subtraction Technique at Hadron Colliders,” \[arXiv:1104.2508 \[hep-ph\]\].
J. Alwall, A. Freitas, O. Mattelaer, “The Matrix Element Method and QCD Radiation,” Phys. Rev. [**D83**]{}, 074010 (2011). \[arXiv:1010.2263 \[hep-ph\]\].
J. Alwall, K. Hiramatsu, M. M. Nojiri [*et al.*]{}, “Novel reconstruction technique for New Physics processes with initial state radiation,” Phys. Rev. Lett. [**103**]{}, 151802 (2009). \[arXiv:0905.1201 \[hep-ph\]\].
M. M. Nojiri, K. Sakurai, “Controlling ISR in sparticle mass reconstruction,” Phys. Rev. [**D82**]{}, 115026 (2010). \[arXiv:1008.1813 \[hep-ph\]\].
P. Konar, K. Kong and K. T. Matchev, “$\sqrt{\hat{s}}_{min}$ : A Global inclusive variable for determining the mass scale of new physics in events with missing energy at hadron colliders,” JHEP [**0903**]{}, 085 (2009). \[arXiv:0812.1042 \[hep-ph\]\].
P. Konar, K. Kong, K. T. Matchev, M. Park, “RECO level $\sqrt{s}_{min}$ and subsystem $\sqrt{s}_{min}$: Improved global inclusive variables for measuring the new physics mass scale in $E_T$ events at hadron colliders,” JHEP [**1106**]{}, 041 (2011). \[arXiv:1006.0653 \[hep-ph\]\].
D. Krohn, L. Randall, L. -T. Wang, “On the Feasibility and Utility of ISR Tagging,” \[arXiv:1101.0810 \[hep-ph\]\].
J. Alwall, S. Hoeche, F. Krauss [*et al.*]{}, “Comparative study of various algorithms for the merging of parton showers and matrix elements in hadronic collisions,” Eur. Phys. J. [**C53**]{}, 473-500 (2008). \[arXiv:0706.2569 \[hep-ph\]\].
T. Plehn, D. Rainwater, P. Z. Skands, “Squark and gluino production with jets,” Phys. Lett. [**B645**]{}, 217-221 (2007). \[hep-ph/0510144\].
J. Alwall, M. -P. Le, M. Lisanti [*et al.*]{}, “Model-Independent Jets plus Missing Energy Searches,” Phys. Rev. [**D79**]{}, 015005 (2009). \[arXiv:0809.3264 \[hep-ph\]\].
C. G. Lester and D. J. Summers, “Measuring masses of semi-invisibly decaying particles pair produced at hadron colliders,” Phys. Lett. B [**463**]{}, 99 (1999). \[arXiv:hep-ph/9906349\].
A. Barr, C. Lester and P. Stephens, “m(T2): The truth behind the glamour,” J. Phys. G [**29**]{}, 2343 (2003). \[arXiv:hep-ph/0304226\].
M. Burns, K. Kong, K. T. Matchev and M. Park, “Using Subsystem MT2 for Complete Mass Determinations in Decay Chains with Missing Energy at Hadron Colliders,” JHEP [**0903**]{}, 143 (2009). \[arXiv:0810.5576 \[hep-ph\]\].
K. T. Matchev, F. Moortgat, L. Pape, M. Park, “Precise reconstruction of sparticle masses without ambiguities,” JHEP [**0908**]{}, 104 (2009). \[arXiv:0906.2417 \[hep-ph\]\].
P. Konar, K. Kong, K. T. Matchev, M. Park, “Superpartner Mass Measurement Technique using 1D Orthogonal Decompositions of the Cambridge Transverse Mass Variable $M_{T2}$,” Phys. Rev. Lett. [**105**]{}, 051802 (2010). \[arXiv:0910.3679 \[hep-ph\]\].
K. T. Matchev, M. Park, “A general method for determining the masses of semi-invisibly decaying particles at hadron colliders,” Phys. Rev. Lett. [**107**]{}, 061801 (2011). \[arXiv:0910.1584 \[hep-ph\]\].
A. Rajaraman, F. Yu, “A New Method for Resolving Combinatorial Ambiguities at Hadron Colliders,” Phys. Lett. [**B700**]{}, 126-132 (2011). \[arXiv:1009.2751 \[hep-ph\]\].
G. L. Bayatian [*et al.*]{} \[ CMS Collaboration \], “CMS technical design report, volume II: Physics performance,” J. Phys. G [**G34**]{}, 995-1579 (2007).
J. Alwall, M. Herquet, F. Maltoni, O. Mattelaer, T. Stelzer, “MadGraph 5 : Going Beyond,” JHEP [**1106**]{}, 128 (2011). \[arXiv:1106.0522 \[hep-ph\]\].
E. Richter-Was \[ Atlas Collaboration \], “Prospect for the Higgs searches with the ATLAS detector,” Acta Phys. Polon. [**B40**]{}, 1909-1930 (2009). \[arXiv:0903.4198 \[hep-ex\]\].
G. L. Bayatian [*et al.*]{} \[ CMS Collaboration \], “CMS physics: Technical design report,”
A. J. Barr, B. Gripaios, C. G. Lester, “Transverse masses and kinematic constraints: from the boundary to the crease,” JHEP [**0911**]{}, 096 (2009). \[arXiv:0908.3779 \[hep-ph\]\].
P. Konar, K. Kong, K. T. Matchev and M. Park, “Dark Matter Particle Spectroscopy at the LHC: Generalizing MT2 to Asymmetric Event Topologies,” JHEP [**1004**]{}, 086 (2010). \[arXiv:0911.4126 \[hep-ph\]\].
T. Sjostrand, S. Mrenna and P. Skands, “PYTHIA 6.4 Physics and Manual,” JHEP [**0605**]{}, 026 (2006). \[arXiv:hep-ph/0603175\].
J. Conway, http://www.physics.ucdavis.edu/ conway/research/software/pgs/pgs4-general.htm
N. Kidonakis, “Top quark cross sections and differential distributions,” \[arXiv:1105.3481 \[hep-ph\]\].
K. Melnikov and F. Petriello, http://www.hep.wisc.edu/ frankjp/FEWZ.html
J. Campbell and K. Ellis, http://mcfm.fnal.gov/
http://www-cdf.fnal.gov/physics/new/top/confNotes/cdf9824spincorr2.8fb-1.pdf, CDF note 9824
[^1]: See [@Konar:2008ei; @Konar:2010ma; @Krohn:2011zp; @Alwall:2007fs; @Alwall:2010cq; @Plehn:2005cq] for different approaches.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The quasi-stationary method for black hole binary inspiral is an approximation for studying strong field effects while suppressing radiation reaction. In this paper we use a nonlinear scalar field toy model (i) to explain the underlying method of approximating binary motion by periodic orbits with radiation; (ii) to show how the fields in such a model are found by the solution of a boundary value problem; (iii) to demonstrate how a good approximation to the outgoing radiation can be found by finding fields with a balance of ingoing and outgoing radiation (a generalization of standing waves).'
address:
- 'Institut für theoretische Physik, Universität Bern, Sidlerstrasse 5, CH-3012 Bern, Switzerland'
- '§Department of Physics, University of Utah, Salt Lake City, UT 84112'
author:
- 'John T Whelan, William Krivan§ and Richard H Price§'
title: 'Quasi-stationary binary inspiral II: Radiation-balanced boundary conditions'
---
Submitted to: [*Class. Quantum Grav.*]{}
Introduction and overview {#sec:intro}
=========================
The inspiraling of black hole binaries is receiving much recent attention both because it is an exciting potential source of detectable gravitational waves[@fh], and due to its inherent interest as a strong field gravitational interaction. The process of inspiral and merger is being investigated with a number of techniques. Newtonian and post-Newtonian computations[@PN] are appropriate to the early stages of inspiral; numerical relativity[@numrel] and black hole perturbation theory[@CL] are used for the late strong field stage of the inspiral. The present paper deals with aspects of the intermediate phase, when strong field effects are too important for post-Newtonian methods to be useful, but in which the full power of numerical relativity is not required. Detweiler and collaborators[@det] have drawn attention to the possible value of an approximation based on the comparison of the orbital time $\tau_{orb}$ for the binary holes with the time $\tau_{rad}$ on which gravitational radiation acts to change the orbit. [From]{} a dimensional analysis of an equal mass binary of mass $M$ and separation $a$ this ratio is $$\label{tauratio}
\frac{\tau_{orb}}{\tau_{rad}}
\propto\left(
\frac{
GM/c^{2}}
{a}
\right)^{5/2}\ .$$ The factor $GM/(ac^{2})$ on the right is an indicator of “how relativistic” the gravitational interaction is between the two holes. When $a$ is on the order of 30 or so times $GM/c^{2}$, the ratio in (\[tauratio\]) will be small, and the orbits may be quasiperiodic, but nonradiative relativistic effects may still be important.
It is useful to focus on a particular strong field relativistic effect of great importance: the innermost stable circular orbit (ISCO). The particle limit in gravitation theory treats the mass $\mu$ of one of the orbiting objects as much smaller than $M$, the mass of the other. In this approximation, there is a minimum radius for circular orbital motion, at a value $r=6GM/c^{2}$ in the case of a nonrotating hole. The existence of this limiting orbit is a purely relativistic effect; there is no such limit in Newtonian theory. It is, furthermore, unrelated to radiation. In the particle limit radiation reaction is a force of order $(\mu/M)^{2}$ and is ignored; the particle moves on a geodesic. There is no such justification for neglecting radiation reaction for the case of binary motion of equal mass holes. Since all orbits are being degraded by gravitational radiation, there is no meaning to “stability” in principle. But there is an important practical question: Do relativistic forces arise in the late binary motion that drive orbiting particles to plunge inward on a time scale much shorter than the time scale due to radiation? It is very plausible that this sort of “practical” meaning can be given to the question of stability. For radial infall of equal mass holes, we know[@misprob] that the radiated energy is a very small fraction of the mass energy of the system as the holes fall into each other from moderate separations. Gravitational radiation reaction, therefore, cannot be a significant modification of the motion of the holes.
This and other questions can be investigated in the absence of radiation reaction by seeking an approximation to the slowly evolving spacetime which is periodic, or, in the case of circular orbits, stationary. It turns out that this greatly changes the mathematical nature of the solution process. The problem of evolving Cauchy data is converted into a boundary value problem. Past experience with this type of problem indicates that this boundary value problem suffers from none of the instability difficulties of numerical nonlinear evolution. The investigation of such questions has usually been based on an [*ansatz*]{}[@Cooknorad; @WMM] for suppressing radiative degrees of freedom in general relativity (GR), but such approaches by necessity only solve a subset of the full Einstein equations. We propose a very different approach. We do not attempt to suppress radiation fields, rather we suppress radiation reaction by requiring that the energy lost to outgoing radiation be replaced by a corresponding amount of ingoing radiation.
Since the goals of this paper are to introduce the general ideas behind a new approximation scheme, and to demonstrate the numerical implementation of this scheme, we devote our attention here to a toy model rather than to GR, with its added complexities. The toy model is the simplest system containing the relevant features of radiation, nonlinearity, and decaying orbits, namely that of a nonlinear scalar field in 2+1 dimensions. The choice of two rather than three spatial dimensions is made for two reasons: First and foremost, it makes the problem of finding a stationary solution less computationally intensive, since the wave equation has to be solved on a two- rather than three-dimensional grid. Second, a scalar field theory in 2+1 dimensions is equivalent to the same theory in 3+1 dimensions with all the sources and fields required to be translationally invariant in one spatial dimension. This theory with line-like sources is analogous to the problem of orbiting line-like sources in 3+1 GR. (Note that the scalar theory in 2+1 dimensions is [*not*]{} analogous to a problem in 2+1 GR. In 2+1 GR, gravity, i.e., Riemann curvature, vanishes outside gravitating bodies.) The formalism for 3+1 line-like sources in GR has already been developed[@paperI], and the spacetime they generate will be the topic of a subsequent paper.
The analogy, with either point-like or line-like sources, between the scalar field problem and full GR is of course only a qualitative one, and we consider the scalar field results only as a test of the general method and not in any way a realistic model of the problem in GR. By the same token, the study now in progress of orbiting line sources in GR will not serve as a realistic simulation of the astrophysical problem with localized sources, but rather as a toy model that incorporates the added complexity of a gravitational problem while remaining computationally similar. Line-like sources in GR have features not found with localized sources, such as the lack of an ISCO for test particles and the lack of an asymptotically flat region in which gravitational waves have familiar properties[@cyl]. On the other hand, the analogy between the 2+1 and 3+1 problems in scalar field theory is much closer, as illustrated analytically in \[app:threeD\] in the absence of nonlinearities. We therefore consider the difference between looking numerically at 2+1 rather than 3+1 *scalar* field theory to be one primarily of computational complexity. Finally,
Our toy model consists of two particles, each with a charge that couples to the nonlinear scalar field. To understand our method it is useful to consider three different solutions for the particle motion and the scalar field: I. The radiation is outgoing, and as a result of the loss of energy due to the radiation, the orbiting particles spiral inward. II. The radiation is outgoing, but due to constraining forces the particles remain in circular periodic orbits. III. Again, the particles move in circular periodic orbits, but now the scalar radiation is balanced and the waves are some generalization of standing waves; there is as much ingoing radiation energy flux as outgoing. The solution of type I is the (scalar field, 2+1) analog of the problem of binary inspiral in GR. The solution of type II is a reasonable approximation of the type-I problem when $\tau_{orb}/\tau_{rad}\ll1$. This type of solution makes sense for a scalar field model; the constraining forces that maintain the periodic motion can be invoked [*ad hoc*]{}. Such forces need not couple to the scalar field, so they can be specified to have no effect on the problem other than to maintain the periodic circular orbits. In GR, on the other hand, all interactions couple to gravitation, and a solution of type II does not make sense. Solutions with periodic orbits and outgoing radiation should therefore not exist in GR. Solutions of type III, however, are not [*a priori*]{} ruled out by such considerations.
There are two principal goals of this paper. The first is to show that the type-II fields can be found by solving a boundary value problem, and to suggest that such a problem is much more easily solved than a Cauchy evolution problem. Our second goal is to demonstrate that a solution of the type-III problem can also be computed without evolution, and that the computed solution gives a good approximation to the fields of the type-II problem, when the conditions of our approximation are valid. The implication is that the physical problem in GR, the type-I problem, can be approximated from the “easily” solved type-III problem.
The remainder of this paper will be organized as follows. The basic toy model, a nonlinear scalar field will be introduced in Section \[sec:toy\] and solutions, both analytic and numerical, will be presented for periodic fields with outgoing boundary conditions. In Section \[sec:balanced\] we discuss the meaning of radiation balanced fields and we make a particular choice (the “TSGF”) of such fields for our nonlinear model. Issues related to the application of the quasi-stationary approximation to GR are discussed in Section \[sec:disc\]. \[app:gfsoln\] gives some technical details of the general solution to the linear wave equation with equal flux of ingoing and outgoing radiation. \[app:threeD\] presents the solution to a linear scalar field problem in 3+1 dimensions, as an illustration of the relationship of the 3+1 and 2+1 problems. Finally, \[app:force\] demonstrates explicitly (in the linear case) the vanishing of the radiation reaction force for the radiation-balanced solutions.
Nonlinear scalar field with periodic orbits {#sec:toy}
===========================================
Our toy model is based on a nonlinear field $\psi$ in 2+1 dimensions satisfying $$\label{gennonlin}
\nabla^{2}\psi-\partial_{t}^{2}\psi+\lambda{\cal F}(\psi,\rho)
=-\sigma\ .$$ where $\sigma$ is a scalar source for the field. The spacetime for the field is Minkowskian and we use polar spatial coordinates $\rho,\phi$. The term $\lambda{\cal F}(\psi,\rho)$ is a term nonlinear in $\psi$ that may be an explicit function of $\rho$, but not an explicit function of $\phi$ or $t$. For later computational convenience we require that the nonlinear term satisfy the symmetry condition $$\label{piflip}
{\cal F}(\psi,\rho)=-{\cal F}(-\psi,\rho)\ .$$ The constant $\lambda$ is a parameter that governs the strength of the nonlinear term. We now specialize to the source $$\label{lines}
\sigma=a^{-1}Q\delta(\rho-a)
\left[
\delta(\phi-\Omega t-\pi/2)-\delta(\phi-\Omega t-3\pi/2)
\right]$$ representing two point sources (in two spatial dimensions) of scalar charge density $\pm Q$, moving around each other in circular orbits of radius $a$, at angular speed $\Omega$.
If we were to treat (\[gennonlin\]) as an evolution problem, we would have to specify Cauchy data and integrate forward in time to find $\psi(\rho,\phi,t)$. Instead we concern ourselves only with the steady state solution, a solution that would evolve after transients associated with the initial conditions are radiated away. This steady state solution would embody the symmetries of the source. In particular, the source depends on $\phi$ and $t$ only in the combination ${\varphi}\equiv \phi-\Omega t$ and we seek a solution with the same symmetry. That is, we look for a solution $\psi(\rho,{\varphi})$. Since the left hand side of (\[gennonlin\]) does not have an explicit dependence on $\phi$ or $t$, such a solution is allowed.
When (\[gennonlin\]) is restricted to solutions of the form $\psi(\rho,{\varphi})$ it reduces to $$\begin{aligned}
\fl\frac{1}{\rho}\frac{\partial}{\partial \rho}\left(
\rho\frac{\partial\psi}{\partial\rho} \right)+\left[
\frac{1}{\rho^{2}}-\Omega^{2}
\right]\frac{\partial^{2}\psi}{\partial{\varphi}^{2}}+\lambda{\cal
F}(\psi,\rho)=-\sigma(\rho,{\varphi})\nonumber\\
=a^{-1}Q\delta(\rho-a)
\left[
\delta({\varphi}-3\pi/2)-\delta({\varphi}-\pi/2)
\right]
\ .
\label{reducedwveq}\end{aligned}$$
=.5
The coordinate regions for this equation are shown in Figure \[coordgrid\]. Although this figure misrepresents the topology (a disk) of the physical problem, it is the appropriate description of the ($\rho,{\varphi}$) grid on which (\[reducedwveq\]) is to be solved as a finite difference equation. The specific conditions to be used in solving the problem are the following: (i) Due to the symmetry of our source, and to condition (\[piflip\]), the solution is to have the symmetry $\psi(\rho,{\varphi})= -\psi(\rho,{\varphi}+\pi)$. This allows us to restrict the numerical solution to the range $0<{\varphi}<\pi$. (ii) Due to the antisymmetry under ${\varphi}\rightarrow{\varphi}+\pi$, the solution for $\psi$ must vanish at all $\rho=0$ grid points. (iii) Outgoing Sommerfeld boundary conditionsare imposed at some maximum radius ${\rho_{\rm max}}$ by requiring $$\label{outgoing}
\left.\left(\frac{\partial\psi}{\partial\rho}
-\Omega\frac{\partial\psi}{\partial{\varphi}}\right)
\right|_{\rho={\rho_{\rm max}}}=0
\ .$$ If (\[reducedwveq\]) and the above conditions (i-iv) are put on a grid of $N_{r}$ radial lines and $N_{{\varphi}}$ lines, containing $N\equiv N_{r}\times N_{{\varphi}}$ grid points with [*a priori*]{} unknown values of $\psi$, a system of $N$ equations for these unknowns is found. The solution of this system is the finite difference solution of our physical problem.
The finite difference procedure just outlined is relatively straightforward, but it has an unusual feature. Note that the nature of the differential equation in (\[reducedwveq\]) changes at the “light cylinder” $\rho=1/\Omega$, shown as a line with dots and dashes in Figure \[coordgrid\]. For $\rho<1/\Omega$ the equation is formally elliptic, while for $\rho>1/\Omega$ it is hyperbolic. Typically elliptic partial differential equations are solved as boundary value problems, with auxiliary data given on a closed boundary surrounding the region of the solution, but hyperbolic equations are given Cauchy data on an “initial” hypersurface. The common wisdom is that a hyperbolic problem with data specified on a closed surface can have more than a single solution[@mathwalk]. Despite this we treat the entire coordinate region ($0\leq{\varphi}<\pi, 0\leq r \leq {\rho_{\rm max}}$) as a boundary value problem. We have not attempted to give a rigorous proof that the boundary value approach to (\[reducedwveq\]) and (\[outgoing\]) is well posed, but several nonrigorous justifications are worth mentioning: (i) Although nonuniqueness is a possible feature for a hyperbolic equations with boundary values, whether or not a particular problem suffers from this difficulty depends on details of the problem, not only on whether it is hyperbolic. (ii) The physical problem described by the boundary value problem appears to be well posed. (iii) Numerical solutions of the boundary value problem are stable and insensitive to numerical grid size.
Here, as in the next section, it will be useful to separate the complexities of nonlinearity from other issues. To do this we temporarily set $\lambda$ to zero so that (\[reducedwveq\]) becomes a linear equation. The outgoing radiation solution for this problem is easily found, with standard techniques, in the form of a series of Bessel $J_{m}$ and Neumann $N_m$ functions. For $\rho\geq a$, this series is $$\label{Bessum}
\fl\psi_{\rm out}=Q\sum_{m=1,3,5,\ldots{}}(-1)^{(m+1)/2}J_{m}(m\Omega a)
\left[
N_m(m\Omega\rho)\sin{m{\varphi}}-J_{m}(m\Omega\rho)\cos{m{\varphi}}
\right]\ .$$ This solution shows, by example, that there are no hidden difficulties in finding a solution to (\[reducedwveq\]) in the case $\lambda=0$, and hence there is no fundamental problem in using a boundary value approach to solve a problem with outgoing radiation. It also demonstrates explicitly that, at least in the $\lambda=0$ case, the light cylinder $\rho=1/\Omega$ is not a special surface in the problem. Note that (\[Bessum\]) gives the solution for waves that are “outgoing at infinity.” Solving the linear problem for waves that are “outgoing” at a finite radius, i.e., for the boundary conditions in (\[outgoing\]), is almost as simple as for true outgoing waves. For $\rho\geq a$, the solution is $$\label{outwall} \psi=\psi_{\rm
out}-Q\sum_{m=1,3,5,\ldots{}}(-1)^{(m+1)/2}J_{m}(m\Omega a)
J_{m}(m\Omega\rho)\,{\rm Re}\left(\gamma_{m}\rme^{i m{\varphi}}\right)\ ,$$ where $$\label{def:gamout}
\gamma_{m}=\left.-\frac{H^{(1)}_{m}(z)+i \frac{d}{dz}H^{(1)}_{m}(z)}
{J_{m}(z)+i \frac{d}{dz}J_{m}(z)}\right|_{z=m\Omega {\rho_{\rm max}}
}\ .$$ Here $H^{(1)}_{m}$ indicates the Hankel function of type 1. From a well known property of Hankel functions[@abramstegun] the numerator of (\[def:gamout\]) vanishes as $m\Omega {\rho_{\rm
max}}\rightarrow\infty$. This demonstrates that (aside from roundoff and truncation error) a solution on a grid of finite radial extent approaches the true outgoing solution as the radial extent of the grid becomes infinite. We have also checked that there are no difficulties, with truncation or otherwise, in solving (\[reducedwveq\]) with finite difference methods. The solution found in this way was compared with (\[Bessum\]). The agreement was excellent, and the expected second order convergence of the finite difference scheme was confirmed.
To investigate a nonlinear model numerically a specific choice must be made for the nonlinear term ${\cal F}$. To suit our purposes this term must satisfy several criteria in addition to the symmetry condition in (\[piflip\]). One criterion is that this nonlinear term be small enough at the outer boundary ${\rho_{\rm max}}$, so that the outgoing condition (\[outgoing\]) is a good approximation at a large finite radius. The nature of this requirement can be seen if we choose ${\cal F}$ simply to be $\psi$, so that the nonlinear term in (\[reducedwveq\]) becomes $\lambda\psi$, and (\[reducedwveq\]) is the Helmholtz equation. The “outgoing” solution to this linear problem is that given in (\[Bessum\]), except that the arguments of the Bessel and Neumann functions have $\sqrt{m^{2}\Omega^{2}+\lambda}$ in place of $m\Omega$. At large $\rho$ this solution will satisfy the condition (\[outgoing\]), for a particular angular Fourier mode, if the $\Omega$ is replaced by $\sqrt{m^{2}\Omega^{2}+\lambda}\,/m$. The standard outgoing condition is then a good approximation only if $|\lambda|\ll \Omega^{2}$. To get a rough idea of the effect of nonlinearities on boundary conditions we can view the nonlinear term $\lambda{\cal F}$ as $\lambda_{\rm eff}\psi$ with $\lambda_{\rm eff}$, the effective $\lambda$, taken to be $\lambda\langle{\cal
F}\rangle/\left<\psi\right>$. Here “$\langle\rangle$” indicates some sort of average (perhaps an r.m.s average over all ${\varphi}$ and one wavelength). A rough criterion for a nonlinear term that is compatible with the boundary condition (\[outgoing\]) is that $\lambda_{\rm eff}$, at ${\rho_{\rm max}}$, be much smaller than $\Omega^{2}$. For our toy model to have interesting nonlinear effects, however, there is a somewhat contradictory requirement: the nonlinear term must be significant, even dominant, at small radius. A nonlinear term like ${\cal F}=\psi^{3}$ can be strong at small radii and weak at large radii if the field $\psi$ falls off quickly enough. For our line sources, the radiation fields fall off only as $r^{-1/2}$. This slow fall off causes computational difficulties in the application of boundary conditions. For this reason we include a factor $\exp{(-\left(\alpha\rho/a \right)^2)}$ in the nonlinear term. To be certain that the nonlinear source is sufficiently well behaved near the point sources at the points $(a,\pi/2)$ and $(a,3\pi/2)$, we choose a source term that does not diverge at those points. The specific choice used in our numerical investigations has been $$\label{gaussource}
{\cal F}=\exp(-(\alpha\rho/a)^{2})
\frac{\psi^3}{\left(\psi^{2}+1\right)^{3/2}}\ .$$
The outgoing wave pattern for $\psi$ is illustrated in Figure \[psiout\]. The field $\psi$ is shown as a function of $\rho
\cos \phi$ and $\rho \sin \phi$ for two different times: (a) $\Omega
t=0$, and (b) $\Omega t=\pi$. The figure illustrates how the rotation of the “rigid” pattern sends waves outward. The results were obtained from numerical runs with $961 \times 41$ gridpoints in $\rho
\in [0,40]$ and ${\varphi} \in [0,\pi]$. The parameter values are $Q=1$, $\Omega =0.5$, $a = 0.5$, $\lambda =1500$, $\alpha=1$. At $\Omega t=0$, the positively charged particle is at $\phi=\pi/2$ and at $\Omega t=\pi$ it is at $\phi=3 \pi/2$.
=3.3in =3.3in
Figure \[inset\] shows the importance of nonlinear effects for the outgoing fields of Figure \[psiout\]. The field $\psi$ is shown as a function of $\rho$ at ${\varphi}=\pi/2$, for $Q=1$, $\Omega =0.5$, and $a = 0.5$. The solid curve shows the solution of the linear problem for $\lambda$ set to zero; the dashed curve shows the nonlinear fields for $\lambda=1500$, $\alpha=1$. Details of the fields near the source particles are shown in the insert. Though the nonlinear terms are negligible in the wave zone, the amplitude of the waves is more than doubled by the increase in the effective source strength produced by the nonlinearity in the central region.
Radiation-balanced boundary conditions {#sec:balanced}
======================================
The universality of gravitational coupling in GR means that periodic orbits and outgoing radiation are not compatible. In this section we investigate a choice of solution for our toy model that [*does*]{} seem applicable to periodic orbits in GR. We introduce boundary conditions that do not carry net energy away from the inner region of the orbiting objects, boundary conditions containing equal measures of ingoing and outgoing radiation. There is a temptation to relate this idea to that of “standing waves,” and to impose Dirichlet boundary conditions that $\psi$ vanishes at some “wall” located at $\rho={\rho_{\rm wall}}$. As in the previous section, it is useful to separate issues of nonlinearity from other issues, and to look at the $\lambda=0$ case first. For this linear problem the solution, for $\rho>a$, with $\psi$ vanishing at $\rho={\rho_{\rm wall}}$ is $$\label{standing}
\fl\psi=
Q\sum_{m=1,3,5,\ldots{}}(-1)^{(m+1)/2}J_{m}(m\Omega a)
\left[
N_m(m\Omega\rho)+\beta_{m}J_{m}(m\Omega\rho)
\right]\sin{m{\varphi}}\ ,$$ where $$\label{betastand}
\beta_{m}=-\frac{N_m(m\Omega {\rho_{\rm wall}})}
{J_{m}(m\Omega {\rho_{\rm wall}})}
\ .$$ The expressions in (\[standing\]) and (\[betastand\]) do not describe a well behaved solution. In general, the denominator for $\beta_{m}$ will come arbitrarily close to zero, as larger and larger values of $m$ are included in the sum. We have confirmed that the finite difference solution to (\[standing\]) and (\[betastand\]) is not stable for small changes in the location of ${\rho_{\rm
wall}}$, or for a change in the number of grid points. \[Note that increasing the number of angular grid divisions is roughly equivalent to increasing the maximum value of $m$ included in the sum in (\[standing\]).\]
In considering fields $\psi$ that are acceptable mixtures of ingoing and outgoing waves, it is again helpful to look at the linear ($\lambda=0$) problem. In this case, a more-or-less obvious acceptable choice of $\psi$ can be constructed simply by averaging the solutions to (\[reducedwveq\]) with ingoing waves at infinity and with outgoing waves at infinity. The result, for $\rho>a$, is $$\label{halfoutin}
\psi=Q\sum_{m=1,3,5,\ldots{}}(-1)^{(m+1)/2}J_{m}(m\Omega a)
N_m(m\Omega\rho)\sin{m{\varphi}}\ .$$ Although this is not the most general solution of the linear problem with equal amounts of in and outgoing waves, it is the most natural solution, as discussed in \[app:gfsoln\].
The field in (\[halfoutin\]) is a solution to (\[reducedwveq\]) in the $\lambda=0$ case, but not for any simply stated boundary conditions. In particular, it does not correspond to the vanishing of $\psi$ at some specific finite radius. Nor is it the limit in which the “wall” of (\[standing\]), (\[betastand\]) is at $\infty$; no such limit exists. Rather, it is a superposition of two solutions each of which has a simply stated boundary condition (ingoing or outgoing), and superposition is not valid for solutions of nonlinear equations. We therefore reformulate our toy model so that fields can be found that are the nonlinear equivalent of (\[halfoutin\]). In place of a partial differential equation, we introduce an integral equation. The first step in doing this is to rewrite (\[reducedwveq\]) as $$\label{rewrite}
{\cal L}\psi
=\sigma_{\rm eff}(\rho,{\varphi},\psi)\ ,$$ where $$\label{Ldef}
{\cal L}\equiv
-\frac{1}{\rho}\frac{\partial}{\partial\rho}
\rho\frac{\partial}{\partial\rho}
-\left(
\frac{1}{\rho^{2}}-\Omega^{2}
\right)\frac{\partial^{2}}{\partial{\varphi}^{2}}$$ and $$\label{rhoeff}
\sigma_{\rm eff}(\rho,{\varphi},\psi)\equiv\sigma(\rho,{\varphi})
+\lambda{\cal F}(\psi,\rho)
\ .$$ We take the solution of (\[rewrite\]-\[rhoeff\]) to be the time symmetric Green function (TSGF) solution given by $$\label{GF}
\psi_{\rm TSGF}(\rho,{\varphi})
=\int\int G_{\rm TS}(\rho,{\varphi};\rho',{\varphi}')
\sigma_{\rm eff}(\rho',{\varphi}'\psi(\rho',{\varphi}'))
\rho'\rmd\rho'\rmd{\varphi}'
\ ,$$ where the Green function $G_{\rm TS}(\rho,{\varphi};\rho',{\varphi}')$ is the time symmetric inverse to the linear operator ${\cal L}$ corresponding to equal mixtures of ingoing and outgoing waves. In principle, this Green function can be written explicitly as $$\begin{aligned}
\fl G(\rho,{\varphi};\rho',{\varphi}')_{\rm TS}=% \left\{
\cases{
% \begin{array}{ll}
-\frac{1}{2\pi}\ln({\rho}/{\rho'})
-\frac{1}{2}\sum_{m=1}^\infty
J_m(m\Omega \rho')N_m(m\Omega\rho)\cos m({\varphi}-{\varphi}')
% \quad
& $\rho>\rho'$\\
% &\\
-\frac{1}{2}\sum_{m=1}^\infty
N_m(m\Omega \rho')J_m(m\Omega\rho)\cos m({\varphi}-{\varphi}')
& $\rho<\rho'$
}
%\ .
% \end{array}
% \right.
\nonumber\\
\label{GFexplicit} \end{aligned}$$ In practice, this series form of the time symmetric Green function is not directly applicable to our numerical method, and it is more useful to write the same radiation balanced solution with the following symbolic notation. We denote the solution to the nonlinear problem with outgoing boundary conditions as $$\label{outonly}
\psi_{\rm out}={\cal L}^{-1}_{\rm out}\sigma_{\rm eff}\ ,$$ where ${\cal L}^{-1}_{\rm out}$ is the inverse to ${\cal L}$ for outgoing boundary conditions, i.e., the retarded-time Green function. The ingoing solution $\psi_{\rm in}$ is written in a parallel manner using the advanced-time Green function ${\cal L}^{-1}_{\rm in}$. The linear superposition of the ingoing and outgoing (LSIO) solutions is not itself a solution since the problem is nonlinear.
To arrive at a field $\psi_{\rm TSGF}$ that [is]{} a solution, and that corresponds to the solution in (\[GF\]), we superpose the operators ${\cal L}^{-1}_{\rm in}$ and ${\cal L}^{-1}_{\rm out}$ and take our radiation balanced solution to be $$\label{RBsolution}
\psi_{\rm TSGF}=\frac{1}{2}\left( {\cal L}^{-1}_{\rm in}+{\cal
L}^{-1}_{\rm out} \right)\sigma_{\rm eff}\equiv
{\cal L}^{-1}_{\rm TSGF}\sigma_{\rm eff}.$$ This integral equation is to be solved by iteration. From $\psi_{{\rm
TSGF},n}$, the $n^{\rm th}$ iteration for $\psi_{{\rm TSGF}}$, we construct $\sigma_{{\rm eff},n}$, and then find the $(n+1)^{\rm th}$ iteration of the solution from $$\label{iteration}
\psi_{{\rm TSGF},n+1}={\cal L}^{-1}_{\rm TSGF}\sigma_{{\rm eff},n}\ .$$ This method is well suited to implementation as a finite difference solution to a boundary value problem on a grid of $N$ points, like that in Figure \[coordgrid\]. In this implementation, $\psi_{\rm
TSGF}$ and $\sigma_{\rm eff}$ are one dimensional vectors of length $N$, and ${\cal L}$, along with the chosen boundary conditions, forms an $N\times N$ matrix. Since the form of this matrix depends on boundary conditions, the numerical inverse ${\cal L}^{-1}$ is also specific to the boundary conditions. The numerical radiation balanced operator ${\cal L}_{\rm TSGF}^{-1}$ is simply the average of the matrix inverses of ${\cal L}^{-1}$ for the ingoing and outgoing problems. Results of this numerical solution are shown in Figure \[standwvs\]. This figure, when compared with Figure \[psiout\] nicely illustrates the symmetry of the radiation field with respect to reversal of ${\varphi}$. It is not so effective in illustrating the fact that there is no radius at which $\psi$ vanishes at all values of ${\varphi}$. This is due to the strong dominance of the $m=1$ multipole of the field. For larger values of the source velocity $v=a\Omega$ the $m=3,5,\ldots{}$ multipoles make stronger contributions, but for those values of $v$ for which we could get accurate solutions (up to $v$ around 0.8) the $m=1$ multipole continued to dominate the appearance of the fields in a plot like that in Figure \[standwvs\].
We now argue that within the scope of our approximation, the TSGF solution in the wave zone is close to a linear superposition of the ingoing and outgoing (LSIO) solutions for the same orbiting sources. It is important to recall that due to the nonlinearity, the LSIO is [*not*]{} a solution of (\[reducedwveq\]), but it does have the convenient property that if the nonlinearities are weak in the wave zone, the outgoing and ingoing solutions can easily be extracted from the LSIO. We are claiming then that an approximate outgoing solution to the problem, in the wave zone, can be found from the TSGF solution. We emphasize that we are [*not*]{} claiming that nonlinear effects are weak. We will show that this approximation method is successful for problems with very strong nonlinear effects. The reason for this success can be seen most easily if (\[RBsolution\]) is rewritten as $$\label{RBrewrite}
\psi_{\rm TSGF}=\frac{1}{2}{\cal L}_{\rm in}^{-1}
\sigma_{\rm eff}(\psi_{\rm TSGF})
+ \frac{1}{2}{\cal L}^{-1}_{\rm out}\sigma_{\rm eff}(\psi_{\rm TSGF})$$ and compared with the LSIO ([*not*]{} a solution of the nonlinear problem), $$\label{LSIO}
\psi_{\rm LSIO}=\frac{1}{2}{\cal L}_{\rm in}^{-1}
\sigma_{\rm eff}(\psi_{\rm in})
+ \frac{1}{2}{\cal L}^{-1}_{\rm out}\sigma_{\rm eff}(\psi_{\rm out})\ .$$ The TSGF solution and the LSIO superposition differ due to the nonlinear terms contained within the effective source. Those nonlinear terms will be significant only in the small radius inner regions of the physical space, where the fields are strong. [*But in the inner, strong field, regions the solution should not be highly sensitive to whether ingoing or outgoing boundary conditions are imposed.*]{} Indeed, this last statement is a way of viewing the underlying idea in our approach. If the fields near the orbiting objects are not significantly influenced by the distant boundary conditions, then radiation reaction cannot be important. We are now assuming the converse: with parameters for which radiation reaction is not important, the fields near the sources will not be sensitive to the boundary conditions.
If the nonlinear contributions to $\sigma_{\rm eff}$ are nonnegligible only in the strong field region near the orbiting objects, and if the fields there are insensitive to boundary conditions, then we can conclude that $\sigma_{\rm eff}(\psi_{\rm in})$ and $\sigma_{\rm
eff}(\psi_{\rm out})$ differ negligibly. It is then plausible that they are also negligibly different from $\sigma_{\rm eff}(\psi_{\rm
TSGF})$, and hence that the LSIO is approximately the same as the TSGF solution. This conclusion, furthermore, should be valid to the same degree that it is valid to ignore radiation reaction (more specifically to ignore the sensitivity of the fields near the source to the boundary conditions on the waves).
For the nonlinearity given by (\[gaussource\]), with $\lambda=1500$, $\alpha=1$, it turns out that the approximation of TSGF by LSIO is [*too*]{} good for an effective illustration. In this case a plot shows no discernible difference between the TSGF and LSIO, even though the nonlinearity plays a strong role, as shown in Figure \[inset\]. A discernible difference can be seen if the $\alpha$ parameter is reduced. The effect is to spread the region of strong nonlinearity to larger radius where the fields are somewhat sensitive to the boundary conditions. The comparison of the TSGF and the LSIO fields, for $\alpha=0.1$ is shown in Figure \[TSGFvsLSIO\]. The field $\psi$ is shown as a function of $\rho$ both for ${\varphi}=\pi/2$, the angular position of a source particle, and at ${\varphi}=\pi/4$. No difference between the TSGF and the LSIO fields can be seen at small radii. In the radiation zone a small phase shift can be seen between the LSIO and the TSGF and the LSIO waves are seen to have a slightly smaller amplitude.
In our approximation approach, the idea is to find the solution of the outgoing wave amplitude from the TSGF solution, by considering the TSGF solution in the wave zone to be (approximately) an equal mixture of ingoing and outgoing radiation. The specific application of this idea requires fitting the TSGF radiation field to a sum of multipoles of the form $$\psi_{\rm
TSGF}\approx\sum_{m=1,3,5\ldots{}}C_{m}N_{m}(m\Omega\rho)\sin{m{\varphi}},$$ and from the amplitudes $C_{m}$ inferred from the fit, writing the outgoing solution as $$\psi_{\rm out}\approx\sum_{m=1,3,5\ldots{}}C_{m}
\left[N_{m}(m\Omega\rho)\sin{m{\varphi}}
-J_{m}(m\Omega\rho)\cos{m{\varphi}}\right]\ .$$ Using our toy nonlinear model, we can solve for both the TSGF and the outgoing solution, to check the accuracy of this method. For the model with parameters $Q=1$, $a=0.5$, $\Omega=0.5$, $\lambda=150$, $\alpha=0.1$, fitting the TSGF from $\rho=20$ to $\rho=40$, gives a wave amplitude for outgoing radiation that is larger than the true amplitude by 24%. For $\alpha=1$ the discrepancy is only 0.04%.
Conclusion and discussion {#sec:disc}
=========================
For our nonlinear toy scalar field model, we have demonstrated that a radiation balanced field, the time symmetric Green function (TSGF) solution, can be found by solving a numerical problem similar to a boundary value problem. Although the character of the differential operator is elliptical inside the light cylinder and hyperbolic outside, no special treatment of this surface was necessary, and the solution was found with boundary value methods usually associated with elliptical equations, despite the outer boundary being in the “hyperbolic” region. We have also shown that to the extent that radiation reaction forces can be ignored, the TSGF solution is a good approximation to the solution with outgoing boundary conditions.
We must now ask how related methods might be brought to bear on the problem of orbiting objects in GR. An important feature of the nonlinear scalar toy model is that the nonlinearities in the theory do not occur in the wave operator. This allowed us directly to recast the problem as an integral equation by using the time-symmetric Green function for the wave operator. It is not at all clear how the GR equations can be cast in a form with a linear wave operator, or whether it is in principle impossible. But it is not necessary that we follow these same steps in dealing with the GR equations. What is required, most generally, is any method for specifying a time symmetric solution.
For such a method in GR it is expected that some features of the basic physics will be the same as in the scalar model. In particular, for motions $\tau_{orb}/\tau_{rad}\ll 1$ the nonlinear terms in the effective source should be insensitive to boundary conditions. It can then be supposed that the radiation balanced solution is a good approximation to the linear superposition of ingoing and outgoing solutions (LSIO) and that from the LSIO we can infer the outgoing solution. There is, however, an important difference between this claim as it applies to GR, and the (demonstrably true) claim for the scalar field model. In the scalar field model we were comparing LSIO fields for periodic motion with the TSGF solution for periodic motion. In GR, there can be no LSIO solution for periodic motion, since ingoing, or outgoing, energy flux would be incompatible with periodic motion. For GR, the analogy needs to be made directly to the solution with outgoing waves and a slow rate of orbital decay due to radiation reaction. By constructing a spacetime with only outgoing radiation from the TSGF solution, we arrive at an approximation in which the radiation fields are periodic, with a period that is constant, not slowly drifting in time. In the notation introduced in Section \[sec:intro\], the quasi-stationary method in GR requires that the type-III problem be solved and used to produce an approximation to the type-I physics.
An additional complication is that the spacetime of the TSGF fields cannot be asymptotically flat in GR, since there is an infinite amount of energy contained in the radiation fields[@gibbonsstewart]. This apparent pathology can be viewed as irrelevant if we think of our TSGF, or outgoing, solutions as being an approximation only in a finite central region of the space. It is worth noting that this viewpoint is consistent with the nature of the numerical solution that is based on boundary conditions applied at finite radius. The important question that remains is whether the inner region, the region in which the problem is solved, is large enough to include a zone in which waves are weak and in which the ingoing and outgoing waves of the approximate LSIO can be disentangled. To investigate this, we note that the gravitational wave luminosity of the binary is of order $L_{\rm GW} \sim (GM/ac^2)^5 (c^5/G)$, where we are using the same notation as for (\[tauratio\]). The gravitational wave energy $E_{\rm GW}$ contained within a sphere of radius ${\rho_{\rm max}}$ is of order $L_{\rm GW}{\rho_{\rm max}}/c$, and hence the ratio of $E_{\rm GW}$ to the orbital energy $GM^{2}/a$ of the binary, is of order $$\frac{E_{\rm GW}}{GM^2/a}\sim \left( \frac{GM}{ac^{2}}
\right)^3\frac{{\rho_{\rm max}}}{a}\ .$$ The nature of our approximation requires that $GM/(ac^{2})$ be small, so we conclude that the gravitational wave energy contained within ${\rho_{\rm max}}$ will be much less than the orbital energy, even for values of ${\rho_{\rm max}}$ large compared to $a$.
This last conclusion is important if we are to hope to use versions of our method to make inferences about the ISCO. By investigating the dependence of the binary energy on radius, we can study whether there is an instability like that for particles. The “energy” of the orbit must be computed as a surface integral at a large radius. If this integral were significantly influenced by the energy contained in the waves, conclusions about orbital stability would be suspect.
Whether or not the methods of this paper can be applied to the ISCO question, the use of these methods in GR would provide an important addition to the tools needed to understand black hole inspiral. A direct and obvious use of the method would be to provide Cauchy data for numerical relativity.
We would like to thank Steven Detweiler for very helpful discussions. We thank also Patrick Brady, Teviet Creighton, Éanna Flanagan, Scott Hughes, Kip Thorne, and Alan Wiseman for useful discussions at meetings on the intermediate black hole problem at Caltech. We are grateful also to Jiří Bičák and John Friedman for helpful suggestions. This work was partially supported by the National Science Foundation under grant PHY9734871. One of us (JTW) acknowledges support by the Swiss Nationalfonds, and by the Tomalla Foundation, Zürich.
General Radiation-Balanced Solution {#app:gfsoln}
===================================
We present here some of the details of the general solution for the linear scalar field theory with equal flux of ingoing and outgoing radiation. The Green function $G(\rho,{\varphi};\rho',{\varphi}')$ for the linear problem satisfies $$\fl\left[\frac{1}{\rho}\frac{\partial}{\partial\rho}
\rho\frac{\partial}{\partial\rho}
+\left(
\frac{1}{\rho^{2}}-\Omega^{2}
\right)
\frac{\partial^{2}}{\partial{\varphi}^{2}}
\right]G(\rho,{\varphi};\rho',{\varphi}')
=-\rho'^{-1}\delta(\rho-\rho')\delta({\varphi}-{\varphi}').$$ When the ${\varphi}$ dependence is represented in a Fourier series, and the Green function is written as $G(\rho,{\varphi};\rho',{\varphi}')$ =Re$\left[\sum{\cal
G}_m(\rho;\rho',{\varphi}')\rme^{im{\varphi}}\right]$, the equation becomes $$\label{generalRB}
\left[\frac{1}{\rho}\frac{\partial}{\partial\rho}
\rho\frac{\partial}{\partial\rho}+m^2
\left(
\Omega^{2}-
\frac{1}{\rho^{2}}
\right)
\right]
{\cal
G}_m(\rho;\rho',{\varphi}')
=-\frac{1}{2\pi\rho'}
\delta(\rho-\rho')\rme^{-im{\varphi}'}$$ Reality of the Green function $G(\rho,{\varphi};\rho',{\varphi}')$ means that ${\cal G}_{-m}(\rho;\rho',{\varphi}')={\cal
G}_m(\rho;\rho',{\varphi}')^*$, and the $m=0$ mode is irrelevant for the sources we consider, so we only need the general solution to (\[generalRB\]) for $m>0$, which is $$\fl
{\cal G}_m(\rho;\rho',{\varphi}')=\cases{
%\left\{
% \begin{array}{ll}
-\frac{1}{4}
J_m(m\Omega \rho')N_m(m\Omega\rho)\rme^{-im{\varphi}'}
%&\\
%\quad
+\Gamma_m(\rho',{\varphi}')J_m(m\Omega\rho)
& $\rho>\rho'$\\
%
% &\\
%
-\frac{1}{4}
N_m(m\Omega \rho')J_m(m\Omega\rho)\rme^{-im{\varphi}'} % &\\
% \quad
+\Gamma_m(\rho',{\varphi}')J_m(m\Omega\rho)
& $\rho<\rho'$
}
% \end{array}
% \right.$$ For $\rho>\rho'$, ${\cal G}_m$ can be written in terms of Hankel functions as $${\cal G}_m=A_mH^{(1)}_m(m\Omega\rho)+B_mH^{(2)}_m(m\Omega\rho)\ ,$$ where $$\begin{aligned}
A_m&=&\frac{1}{2}\Gamma_m(\rho',{\varphi}')
-\frac{1}{8i}J_m(m\Omega\rho)\rme^{-im{\varphi}'}
\nonumber\\
B_m&=&\frac{1}{2}\Gamma_m(\rho',{\varphi}')
+\frac{1}{8i}J_m(m\Omega\rho)\rme^{-im{\varphi}'}\label{AandB} \ .\end{aligned}$$ The condition for equal flux of ingoing and outgoing radiation is $|A_m|=|B_m|$, for all $m$. This, and the expressions in (\[AandB\]), require that $\Gamma_m(\rho',{\varphi}')$ be a real multiple of $J_m(m\Omega\rho)\rme^{-im{\varphi}'}$. The proportionality constant may be different for each $m$, so the general radiation balanced solution is $$\begin{aligned}
\fl
{\cal G}_m(\rho;\rho',{\varphi}')=
\cases{
% \left\{
% \begin{array}{ll}
-\frac{1}{4}
J_m(m\Omega \rho')N_m(m\Omega\rho)\rme^{-im{\varphi}'}% &\\
% \quad
+g_mJ_m(m\Omega \rho')J_m(m\Omega\rho)\rme^{-im{\varphi}'}
& $\rho>\rho'$\\
%
% &\\
%
-\frac{1}{4}
N_m(m\Omega \rho')J_m(m\Omega\rho)\rme^{-im{\varphi}'}
% &\\
% \quad
+g_mJ_m(m\Omega \rho')J_m(m\Omega\rho)\rme^{-im{\varphi}'}
& $\rho<\rho'$
}
% \end{array}
% \right.
\nonumber\\\end{aligned}$$ Any choice of the set of real constants $g_m$ gives a radiation balanced solution. The choice made for the TSGF solution, $g_m=0$, for all $m$, is convenient for a numerical implementation that does not explicitly use Fourier decomposition. The choice $g_m=0$ also seems to be the most natural one if one views the $\Gamma_m$ terms in (\[generalRB\]) as free waves, disconnected from the source.
Three-plus-One Dimensional Theory {#app:threeD}
=================================
Here we present the solution of a linear toy model for point-like sources. We use standard spherical coordinates ($r,\theta,\phi$) and suppose that two particles of opposite scalar charge $\pm Q$ orbit at frequency $\Omega$ in the equatorial plane ($\theta=\pi/2$), at radius $a$, with angular separation $\Delta\phi=\pi$. In the linear differential equation of this toy model $$\label{genlin}
\fl\nabla^{2}\psi-\partial_{t}^{2}\psi=
-a^{-2}Q\delta(r-a)\delta(\cos\theta)
\left[\delta(\phi-\Omega t-\pi/2)-\delta(\phi-\Omega t-3\pi/2) \right]
\ ,$$ we make the [*ansatz*]{} that the solution is not of the general type $\psi(r,\theta,\phi,t)$, but rather of the type $\psi(r,\theta,\varphi)$ where $\varphi\equiv\phi-\Omega t$. The differential equation then reduces to the point-like linear analog of (\[reducedwveq\]): $$\begin{aligned}
\fl\frac{1}{r^2}\frac{\partial}{\partial r}\left(
r^2\frac{\partial\psi}{\partial r} \right)
+\frac{1}{r^2\sin\theta}\frac{\partial}{\partial\theta}\left(
\sin\theta\frac{\partial\psi}{\partial\theta}
\right)
+\left[
\frac{1}{r^{2}\sin^2\theta}-\Omega^{2}
\right]\frac{\partial^{2}\psi}{\partial{\varphi}^{2}}
\nonumber\\
=a^{-2}Q\delta(\rho-a)\delta(\cos\theta)
\left[\delta({\varphi}-\pi/2)-\delta({\varphi}-3\pi/2) \right]\ .
\label{ptlike}\end{aligned}$$ If $\psi(r,\theta,\varphi)$ is decomposed as a sum of spherical harmonics $Y_{\ell m}(\theta,\varphi)$ it is straightforward to find the solution that is well behaved at $r=0$ and that corresponds to outgoing waves at infinity. This solution, the point-like equivalent of (\[Bessum\]), is $$\label{ptsoln}
\psi(r,\theta,\varphi)
=Q\Omega\sum_{\ell m}\kappa_{\ell m}Y_{\ell m}(\theta,\varphi)
\times\cases{
% \left\{
% \begin{array}{ll}
j_\ell(m\Omega a) h^{(1)}_\ell(m\Omega r) & $r>a$\\
%
% &\\
%
j_\ell(m\Omega r) h^{(1)}_\ell(m\Omega a) &$r<a$
}
% \end{array}
% \right. \ .$$ Here $j_\ell$ and $h^{(1)}_\ell$ indicate spherical Bessel and Hankel functions. The sum in (\[ptsoln\]) is over odd $\ell$ and odd $m$, and the coefficients $\kappa_{\ell m}$ are given by $$\fl\kappa_{\ell m}=
im\left[Y_{\ell m}^{*}(\pi/2,\pi/2)-Y_{\ell m}^{*}(\pi/2,3\pi/2)\right]
=(-1)^{(m-1)/2}2mY_{\ell m}(\pi/2,0)
\ .$$ The point-like equivalent of (\[halfoutin\]), the radiation balanced TSGF solution of the problem, is $$\label{ptsolnmerb}
\psi(r,\theta,\varphi)
=iQ\Omega\sum_{\ell m}\kappa_{\ell m}Y_{\ell m}(\theta,\varphi)
\times\cases{
% \left\{
% \begin{array}{ll}
j_\ell(m\Omega a) n_\ell(m\Omega r) & $r>a$\\
%
% &\\
%
j_\ell(m\Omega r) n_\ell(m\Omega a) &$r<a$
}
% \end{array}
% \right. \ ,$$ where $n_\ell$ is the spherical Neumann function. These solutions serve as explicit illustrations that for point-like sources in 3+1 dimensions the field embodies the same symmetry as the sources. That is, the field rotates “rigidly.” The dependence on time and on azimuthal angle appears only in the combination $\phi-\Omega t$.
Net Force on the Orbiting Particles {#app:force}
===================================
In this appendix, we show that for a general radiation-balanced solution to the problem of orbiting particles, the proscribed circular orbits of the particles are consistent with the scalar field equations of motion without the need for external forces. (This is not true in the case of purely outgoing radiation.) We limit attention to the linear theory, so that we can easily separate out the forces due to each particle on the other, without worrying about any self-force.
The covariant force law for a scalar-charged particle with charge $Q$, mass $m$, and energy-momentum vector $p$ moving in a scalar field $\psi$ gives a force of $$m^{-1} p^\nu \nabla_\nu p^\mu = \frac{\rmd p^\mu}{\rmd\tau}
+ m^{-1} \Gamma^\mu_{\nu\lambda} p^\nu p^\lambda
= -Q \nabla^\mu \psi
\ .$$ If we work in co-rotating coordinates, in which the particle is momentarily at rest, the metric is $$\rmd s^2 = -(1-\Omega^2\rho^2)\rmd t^2
+\rmd\rho^2+\rho^2\rmd\varphi^2
+2\Omega\rho^2 \rmd\varphi\rmd t$$ and the only non-zero component of the energy-momentum vector is $$p^t = (1-\Omega^2a^2)^{-1/2} m =: \gamma m
\ ,$$ so that the only relevant Christoffel symbol is $\Gamma^\rho_{tt}$. (In the corresponding calculation in the 3+1 dimensional theory, only $\Gamma^r_{tt}$ is relevant.)
The equations of motion for $p^\rho$ and $p^\varphi$ in the 2+1 dimensional theory thus become $$\begin{aligned}
\label{radforce}
\frac{\rmd p^\rho}{\rmd\tau}
= - Q \psi_{,\rho} + \gamma^2m\rho\Omega^2\\
\label{angforce}
\frac{\rmd p^\varphi}{\rmd\tau}
= - Q g^{\varphi\varphi} \psi_{,\varphi}
\ ,\end{aligned}$$ where we have used the form of the inverse metric $g_{\mu\nu}$ in co-rotating coordinates and the fact that $(\partial\psi/\partial
t)_\varphi=0$. The second term in (\[radforce\]) represents the fictitious centrifugal force that the particle feels in the co-rotating reference frame. The radial momentum $p^\rho$ can remain zero if the repulsive effect of this term cancels the attraction due to the field, described in the first term. This implies a relationship among the orbital radius $a$, angular frequency $\Omega$, and charge-to-mass ratio $Q/m$. (The corresponding relationship is given for orbiting charged particles in the presence of a half-advanced, half-retarded electromagnetic potential in [@schild].)
The angular force in (\[angforce\]), which represents radiation reaction force, will vanish if (and only if) the scalar field due to one particle has vanishing $\varphi$ derivative at the location of the other particle. For concreteness, we consider the force on the positively-charged particle at $\rho=a$, $\varphi=\pi/2$ due to the negatively-charged particle at $\rho=a$, $\varphi=3\pi/2$. The field due to the latter particle is $$\psi(\rho,\varphi) = -Q G(\rho,\varphi;a,3\pi/2)
\ ,$$ where $G(\rho,\varphi;\rho',\varphi')$ is the Green function used to construct the solution. Using the Fourier expansion of the Green function given in \[app:gfsoln\], we find that $$\fl
\psi_{,\varphi}(a,\pi/2) = -{\rm Re}
\left[
\sum_m im
\left(
\frac{1}{4}J_m(m\Omega a)N_m(m\Omega a)(-1)^m
+\Gamma_m(a,3\pi/2)\rme^{im\pi/2}
\right)
\right]
\ .$$ This vanishes if the quantity in large parentheses on the right-hand side is real; for a general radiation-balanced solution this is the case, since $\Gamma_m(a,3\pi/2)=g_m J_m(m\Omega a) \exp(-im3\pi/2)$ with $g_m$ real. Thus *the radiation-balanced solutions have zero radiation reaction force*.
However, in the case of purely outgoing radiation, $\Gamma_m(\rho',\varphi')=J_m \rme^{-im\varphi'}/4i$, and we can explicitly find the radiation reaction force as $$\psi_{,\varphi}(a,\pi/2)
= -\sum_m (-1)^m\frac{m}{4} [J_m(m\Omega a)]^2 \ne 0
\ .$$
A similar demonstration can be made in the 3+1-dimensional case using the outgoing-radiation and radiation-balanced fields (\[ptsoln\]) and (\[ptsolnmerb\]), using the symmetries of the spherical Bessel functions under changes of sign in their arguments to illustrate that terms in the radiation-reaction force due to terms with opposite signs of $m$ cancel each other out in the radiation-balanced mode expansion but not in the outgoing-radiation one.
References {#references .unnumbered}
==========
[99]{}
Flanagan É É and Hughes S A 1998 4535 (Flanagan É É and Hughes S A 1997 [*Preprint*]{} gr-qc/9701039)
E. Poisson 1998 [*Preprint*]{} gr-qc/9801038
Numerical relativity applied to black hole inspiral has been the subject of a Grand Challenge computational project. For information visit: [**http://www.npac.syr.edu/projects/bh/**]{}.
Pullin J 1998 [*Preprint*]{} gr-qc/9803005
Blackburn J K and Detweiler S 1992 2318 Detweiler S 1994 4929
Anninos O, Hobill D, Seidel E, Smarr L, and Suen W-M 1993 2851
Cook G B, Shapiro S L and Teukolsky S A 1996 5533 Cook G B, Shapiro S L and Teukolsky S A 1995 [*Preprint*]{} gr-qc/9512009 Wilson J R and Mathews G J 1995 4161 Wilson J R, Mathews G J and Marronetti P 1996 1317 Wilson J R, Mathews G J and Marronetti P 1996 [*Preprint*]{} gr-qc/9601017
Whelan J T “Quasistationary binary inspiral III: Gravitational waves on a cylindrically symmetric background” (unpublished)
Whelan J T and Romano J D 1999 084009 Whelan J T and Romano J D 1998 [*Preprint*]{} gr-qc/9812041
Gibbons G W and Stewart J M 1983 [*Classical General Relativity*]{} ed Bonnor W B [*et al*]{} (Cambridge: Cambridge University Press)
Mathews J and Walker R L 1970 [*Mathematical Methods of Physics*]{} sec 8.2 (Reading: W. A. Benjamin)
Olver F W J 1964 [*Handbook of Mathematical Functions*]{} ed Abramowitz M and Stegun I A (Washington: National Bureau of Standards)
Schild A 1963 , 2762
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study WDC sets, which form a substantial generalization of sets with positive reach and still admit the definition of curvature measures. Main results concern WDC sets $A\subset {{\mathbb R}}^2$. We prove that, for such $A$, the distance function $d_A= {{\rm dist}\,}(\cdot,A)$ is a “DC aura” for $A$, which implies that each locally WDC set in ${{\mathbb R}}^2$ is a WDC set. An another consequence is that compact WDC subsets of ${{\mathbb R}}^2$ form a Borel subset of the space of all compact sets.'
author:
- Dušan Pokorný
- Luděk Zajíček
title: Remarks on WDC sets
---
[^1]
Introduction
============
In [@PR] (cf. also [@FPR], [@Fu2] and [@PRZ]), the authors introduced the class of WDC sets which form a substantial generalization of sets with positive reach and still admit the definition of curvature measures. The following question naturally arises (see [@FPR Question 2, p. 829] and [@Fu2 10.4.3]).
Is the distance function $d_A= {{\rm dist}\,}(\cdot,A)$ of each WDC set $A\subset {{\mathbb R}}^d$ a DC aura for $F$ (see Definition \[def:WDC\])?
We answer this question positively in the case $d=2$ (Theorem \[thm:distanceAura\] below); it remains open for $d\geq 3$. The proof is based on a characterization (proved in [@PRZ]) of locally WDC sets in ${{\mathbb R}}^2$ and the main result of [@PZ] which asserts that $$\label{dcgr}
\text{$d_A$ a DC function if $A\subset {{\mathbb R}}^2$ is a graph of a $DC$ function $g: {{\mathbb R}}\to {{\mathbb R}}$.}$$ Recall that a function is called DC, if it is the difference of two convex functions and note that each set $A$ as in is a WDC set. Theorem \[thm:distanceAura\] easily implies that each locally WDC set in ${{\mathbb R}}^2$ is WDC. Further, we use Theorem \[thm:distanceAura\] to prove that compact WDC subsets of ${{\mathbb R}}^2$ form a Borel subset of the space of all compact sets of ${{\mathbb R}}^2$ (Theorem \[complwdc\] (i)). The importance of this result is the fact that it suggests that (at least in ${{\mathbb R}}^2$) a theory of point processes on the space of compact WDC sets (analogous to the concept of point processes on the space of sets with positive reach introduced in [@Zah86]) can be build.
Concerning the compact WDC subsets of ${{\mathbb R}}^d$ for $d>2$, we are able to prove only a weaker fact that they form an analytic set (Theorem \[complwdc\] (ii)) which is not probably sufficient for the above mentioned application.
Preliminaries
=============
Basic definitions {#basic}
-----------------
The symbol $\q$ denotes the set of all rational numbers. In any vector space $V$, we use the symbol $0$ for the zero element. We denote by $B(x,r)$ ($U(x,r)$) the closed (open) ball with centre $x$ and radius $r$. The boundary and the interior of a set $M$ are denoted by $\partial M$ and ${\mathrm{int}}M$, respectively. A mapping is called $K$-Lipschitz if it is Lipschitz with a (not necessarily minimal) constant $K\geq 0$.
The metric space of all real-valued continuous functions on a compact $K$ (equipped with the usual supremum metric $\rho_{\sup}$) will be denoted $C(K)$.
In the Euclidean space ${{\mathbb R}}^d$, the norm is denoted by $|\cdot|$ and the scalar product by $\langle \cdot,\cdot\rangle$. By $S^{d-1}$ we denote the unit sphere in ${{\mathbb R}}^d$.
The distance function from a set $A\subset {{\mathbb R}}^d$ is $d_A\coloneqq {{\rm dist}\,}(\cdot,A)$ and the metric projection of $z\in {{\mathbb R}}^d$ to $A$ is $\Pi_A(z)\coloneqq \{ a\in A:\, {{\rm dist}\,}(z,A)=|z-a|\}$.
DC functions {#dc}
------------
Let $f$ be a real function defined on an open convex set $C \subset {{\mathbb R}}^d$. Then we say that $f$ is a [*DC function*]{}, if it is the difference of two convex functions. Special DC functions are semiconvex and semiconcave functions. Namely, $f$ is a [*semiconvex*]{} (resp. [*semiconcave*]{}) function, if there exist $a>0$ and a convex function $g$ on $C$ such that $$f(x)= g(x)- a \|x\|^2\ \ \ (\text{resp.}\ \ f(x)= a \|x\|^2 - g(x)),\ \ \ x \in C.$$
We will use the following well-known properties of DC functions.
\[vldc\] Let $C$ be an open convex subset of ${{\mathbb R}}^d$. Then the following assertions hold.
1. If $f: C\to {{\mathbb R}}$ and $g: C\to {{\mathbb R}}$ are DC, then (for each $a\in {{\mathbb R}}$, $b\in {{\mathbb R}}$) the functions $|f|$, $af + bg$, $\max(f,g)$ and $\min(f,g)$ are DC.
2. Each locally DC function $f:C \to {{\mathbb R}}$ is DC.
3. Each DC function $f:C \to {{\mathbb R}}$ is Lipschitz on each compact convex set $Z\subset C$.
4. Let $f_i: C \to {{\mathbb R}}$, $i=1,\dots,m$, be DC functions. Let $f: C \to {{\mathbb R}}$ be a continuous function such that $f(x) \in \{f_1(x),\dots,f_m(x)\}$ for each $x \in C$. Then $f$ is DC on $C$.
Property (i) follows easily from definitions, see e.g. [@Tuy p. 84]. Property (ii) was proved in [@H]. Property (iii) easily follows from the local Lipschitzness of convex functions. Assertion (iv) is a special case of [@VeZa Lemma 4.8.] (“Mixing lemma”).
It is well-known (cf. [@PZ]) that if $\emptyset \neq A\subset {{\mathbb R}}^d$ is closed, then $d_A$ need not be DC; however (see, e.g., [@CS Proposition 2.2.2]), $$\label{loksem}
\text{$d_A $ is locally semiconcave (and so locally DC) on ${{\mathbb R}}^d \setminus A$.}$$
Clarke generalized gradient {#clar}
---------------------------
If $U\subset {{\mathbb R}}^d$ is an open set, $f:U\to{{\mathbb R}}$ is locally Lipschitz and $x\in U$, we denote by $\partial_C f(x)$ the [*generalized gradient of $f$ at $x$*]{}, which can be defined as the closed convex hull of all limits $\lim_{i\to\infty}f'(x_i)$ such that $x_i\to x$ and $f'(x_i)$ exists for all $i\in\en$ (see [@C S1.1.2]; $\partial_C f(x)$ is also called *Clarke subdifferential of $f$ at $x$* in the literature). Since we identify $({{\mathbb R}}^d)^*$ with ${{\mathbb R}}^d$ in the standard way, we sometimes consider $\partial_C f(x)$ as a subset of ${{\mathbb R}}^d$. We will repeatedly use the fact that the mapping $x\mapsto\partial_C f(x)$ is upper semicontinuous and, hence (see [@C Theorem 2.1.5]), $$\label{uzcl}
v\in\partial_C f(x)\ \ \text{ whenever}\ \ x_i\to x,\ v_i\in\partial_C f(x_i)\ \ \text{ and}\ \ v_i\to v.$$ We also use that $|u|\leq K$ whenever $u\in\partial_C f(x)$ and $f$ is $K$-Lipschitz on a neighbourhood of $x$. Obviously, $$\label{cls}
\partial_C (\alpha f)(x) = \alpha \partial_C f(x).$$
Recall that $$\label{clder1}
f^0(x,v)\coloneqq \limsup_{y\to x, t\to 0+} \ \frac{f(y+tv)-f(y)}{t}$$ and (see [@C]) $$\label{clder2}
f^0(x,v) = \sup \{ \langle v, \nu\rangle:\ \nu\in \partial_Cf(x)\}.$$ We will need the following simple lemma.
\[clod\] Let $f$ be a Lipschitz function on an open set $G \subset {{\mathbb R}}^d$, $x \in G$ and ${\varepsilon}>0$.
1. If ${{\rm dist}\,}(0, \partial_Cf(x)) \geq 2{\varepsilon}$, then $$\label{derpod}
\exists v\in S^{d-1}, \rho>0 \ \forall y \in U(x,\rho), 0<\alpha< \rho:\ \frac{f(y+\alpha v) -f(y)}{\alpha} \leq - {\varepsilon}.$$
2. If holds, then ${{\rm dist}\,}(0, \partial_Cf(x)) \geq {\varepsilon}$.
(i) Let ${{\rm dist}\,}(0, \partial_Cf(x)) \geq 2{\varepsilon}$. Since $\partial_Cf(x)$ is convex, there exists (see e.g. [@DL Theorem 1.5.]) $v\in S^{d-1}$ such that $${{\rm dist}\,}(0, \partial_Cf(x))= - \sup\{\langle v, \nu\rangle:\ \nu \in \partial_Cf(x)\}.$$ So, by , $f^0(x,v) \leq -2{\varepsilon}$ and thus implies .
(ii) If holds, choose corresponding $v\in S^{d-1}$ and $ \rho>0$. Then $f^0(x,v) \leq -{\varepsilon}$ by . Consequently, by , $- |\nu| \leq \langle v, \nu\rangle \leq -{\varepsilon}$ for each $\nu \in \partial_Cf(x)$ and so ${{\rm dist}\,}(0, \partial_Cf(x)) \geq {\varepsilon}$.
WDC sets {#wdc}
---------
WDC sets (see the definition below) which provide a natural generalization of sets with positive reach were defined in [@PR] using Fu’s notion of an “aura” of a set (see, e.g., [@Fu2] for more information). Note that the notion of a DC aura were defined in [@PR] and [@FPR] by a formally different but equivalent way (cf. [@PRZ Remark 2.12 (v)]).
\[def:WDC\] Let $U\subset{{\mathbb R}}^d$ be open and $f:U\to{{\mathbb R}}$ be locally Lipschitz. A number $c\in {{\mathbb R}}$ is called a [*weakly regular value*]{} of $f$ if whenever $x_i\to x$, $f(x_i)>c=f(x)$ and $u_i\in\partial_C f(x_i)$ for all $i\in\en$ then $\liminf_i|u_i|>0$.
A set $A\subset{{\mathbb R}^d}$ is called [*WDC*]{} if there exists a DC function $f:{{\mathbb R}^d}\to [0,\infty)$ such that $A=f^{-1}\{0\}$ and $0$ is a weakly regular value of $f$. In such a case, we call $f$ a [*DC aura*]{} (for $A$).
A set $A\subset {{\mathbb R}}^d$ is called locally WDC if for any point $a \in A$ there exists a WDC set $A^*\subset {{\mathbb R}}^d$ that agrees with $A$ on an open neighbourhood of $a$.
(Note that a weakly regular value of $f$ need not be in the range of $f$, and so $\emptyset$ is clearly a WDC set by our definition.)
Note that a set $A \subset {{\mathbb R}}^d$ has a positive reach at each point if and only there exists a DC aura for $A$ which is even semiconvex ([@Ba]).
Distance function of a WDC set in ${{\mathbb R}}^2$ is an aura {#distwdc}
==============================================================
First we present (slightly formally rewritten) [@PRZ Definition 7.9].
\[sektory\]
1. A set $S \subset {{\mathbb R}}^2$ will be called a [*basic open DC sector*]{} (of radius $r$) if $S= U(0,r) \cap \{(u,v) \in {{\mathbb R}}^2:\ u\in (-\omega, \omega), v >f(u)\}$, where $0<r<\omega$ and $f$ is a DC function on $(-\omega, \omega)$ such that $f(0)=0$, $R(u)\coloneqq \sqrt {u^2 + f^2(u)}$ is strictly increasing on $[0,\omega)$ and strictly decreasing on $(-\omega,0]$.
By an [*open DC sector*]{} (of radius $r$) we mean an image $\gamma(S)$ of a basic open DC sector $S$ (of radius $r$) under a rotation around the origin $\gamma$.
2. A set of the form $\gamma( \{(u,v) \in {{\mathbb R}}^2:\ u\in [0, \omega), g(u) \leq v \leq f(u)\} ) \cap U(0,r)$, where $\gamma$ is a rotation around the origin, $0<r<\omega$ and $f,g: {{\mathbb R}}\to\er$ are DC functions such that $g \leq f$ on $[0,\omega)$, $f(0)=g(0)=f'_+(0)=g'_+(0)=0$ and the functions $R_f(u)\coloneqq \sqrt {u^2 + f^2(u)}$, $R_g(u)\coloneqq \sqrt {u^2 + g^2(u)}$ are strictly increasing on $[0,\omega)$, will be called a [ *degenerated closed DC sector*]{} (of radius $r$).
We will use the following complete characterization of locally WDC sets in ${{\mathbb R}}^2$ ([@PRZ Theorem 8.14]).
[Theorem PRZ]{}\[T:characterisation\] Let $M$ be a closed subset of $\er^2$. Then $M$ is a locally WDC set if and only if for each $x\in\partial M$ there is $\rho>0$ such that one of the following conditions holds:
1. \[cond:singlePointNew\] $M\cap U(x,\rho)=\{x\}$,
2. \[cond:oneConeNew\] there is a degenerated closed DC sector $C$ of radius $\rho$ such that $$M\cap U(x,\rho)=x+C,$$
3. \[cond:manyConesNew\] there are pairwise disjoint open DC sectors $C_1,\dots,C_k$ of radius $\rho$ such that $$\label{kolac}
U(x,\rho) \setminus M =\bigcup_{i=1}^k \left(x+C_i\right).$$
\[lem:distranceFromGraphWR\] Let $f$ be an $L$-Lipschitz function on $\er$. Denote $d\coloneqq {{\rm dist}\,}(\cdot,{\operatorname{graph}}f)$. Then $|\xi_2|\geq \frac{1}{\sqrt{L^2+1}}$ whenever $\xi=(\xi_1,\xi_2)\in\partial_C d(x)$ and $x\in\er^2\setminus{\operatorname{graph}}f$.
Pick $x\in\er^2\setminus{\operatorname{graph}}f$. Without any loss of generality we can assume that $x=0$. We will assume that $f(0)<0$; the case $f(0)>0$ is quite analogous. Denote $r\coloneqq d(0)$ and $P\coloneqq \Pi_{{\operatorname{graph}}f}(0)$. Set $g(u)\coloneqq-\sqrt{r^2-u^2}$, $u\in[-r,r]$. Clearly $f\leq g$ on $[-r,r]$ and $(u,v)\in P$ if and only if $f(u)=g(u)=v$. We will show that $$\label{odhu}
\text{$|u|\leq \frac{Lr}{\sqrt{1+L^2}}$ whenever $(u,v)\in P$.}$$ To this end, suppose $(u,v)\in P$. If $u>0$, then $$L\geq \frac{f(t)-f(u)}{t-u} \geq \frac{g(t)-g(u)}{t-u}\ \ \text{for each}\ \ 0<t<u,$$ and consequently $L\geq g'_-(u)$. Therefore $u<r$ and $L \geq u (r^2-u^2)^{-1/2}$. Analogously considering $g'_+(u)$, we obtain for $u<0$ that $u>-r$ and $u (r^2-u^2)^{-1/2} \geq -L$. In both cases we have $L \geq |u| (r^2-u^2)^{-1/2}$ and an elementary computation gives .
Using we obtain that if $(u,v)\in P$ then $$\label{eq:estimateg(u)}
v=g(u)\leq -\sqrt{r^2-\left(\frac{Lr}{\sqrt{1+L^2}}\right)^2}
=-\frac{r}{\sqrt{1+L^2}}.$$ By [@Fu Lemma 4.2] and we obtain $$\partial_C d(0)={\operatorname{conv}}\left\{ \frac{1}{r}(-u,-v): (u,v)\in P \right\}\subset \er \times \left[\frac{1}{\sqrt{L^2+1}},\infty\right)$$ and the assertion of the lemma follows.
\[thm:distanceAura\] Let $M\ne \emptyset$ be a locally ${{W\! DC}}$ set in $\er^2$. Then the distance function $d_M$ is a DC aura for $M$. In particular, $M$ is a WDC set.
Denote $d\coloneqq d_M$. For each $x\in\partial M$ choose $\rho= \rho(x)$ by Theorem PRZ. We will prove that
1. \[cond:DC\] $d$ is DC on $U\left( x,\frac{\rho}{3} \right)$,
2. \[cond:weaklyRegular\] there is ${\varepsilon}= {\varepsilon}(x)>0$ such that $|\xi|\geq{\varepsilon}$ whenever $y\in U\left( 0,\frac{\rho}{3} \right)\setminus M$ and $\xi\in\partial_C d(y)$.
Without any loss of generality we can assume that $x=0$.
If case \[cond:singlePointNew\] from Theorem PRZ holds, then $d(y)=|y|$, $y\in U\left( 0,\frac{\rho}{3} \right)$, and so $d$ is convex and therefore DC on $U\left(0,\frac{\rho}{3} \right)$. Similarly, condition \[cond:weaklyRegular\] holds as well, since if $y\in U\left( 0,\frac{\rho}{3} \right)\setminus M$ and $\xi\in\partial_C d(y)$ then $\xi=\frac{y}{|y|}$ and so $|\xi|=1$.
If case \[cond:oneConeNew\] from Theorem PRZ holds, we know that $M\cap U(0,\rho)$ is a degenerated closed DC sector $C$ of radius $\rho$. Let $\gamma$, $f$, $g$ and $\omega$ be as in Definition \[sektory\]. Without any loss of generality we may assume that $\gamma$ is the identity map.
By Lemma \[vldc\] (iii) we can choose $L>0$ such that both $f$ and $g$ are $L$-Lipschitz on $[0,\rho]$ and define $$\tilde f(u)\coloneqq
\begin{cases}
f(u)& \text{if}\quad 0\leq u \leq \rho,\\
f(\rho)& \text{if}\quad \rho<u,\\
2Ly& \text{if}\quad u<0.
\end{cases}
\ \ \ \
\text{and}
\ \ \ \
\tilde g(u)\coloneqq
\begin{cases}
g(u)& \text{if}\quad 0\leq u \leq \rho,\\
g(\rho)& \text{if}\quad \rho<u,\\
-2Lu& \text{if}\quad u<0.
\end{cases}$$
It is easy to see that both $\tilde f$ and $\tilde g$ are $2L$-Lipschitz and they are DC by Lemma \[vldc\] (iv).
Put $$M_0\coloneqq \left\{ (u,v)\in\er^2: u\geq0, \; \tilde g(u)\leq v\leq\tilde f(u)\right\},$$ $$M_1\coloneqq \left\{ (u,v)\in\er^2: u\geq0, \; \tilde f(u)< v\right\}\cup
\left\{ (u,v)\in\er^2: u<0, \; -\frac{u}{2L}< v\right\},$$ $$M_2\coloneqq \left\{ (u,v)\in\er^2: u\geq0, \; \tilde g(u)> v\right\}\cup
\left\{ (u,v)\in\er^2: u<0, \; \frac{u}{2L}> v\right\}$$ and $$M_3\coloneqq
\left\{ (u,v)\in\er^2:\frac uL< v< -\frac uL\right\}.$$ Clearly $\er^2= M_0\cup M_1\cup M_2\cup M_3$ and $M_1$, $M_2$, $M_3$ are open.
Set $\tilde d\coloneqq{{\rm dist}\,}(\cdot, M_0)$ and, for each $y \in {{\mathbb R}}^2$, define $$d_0(y)=0,\ \ d_1(y)\coloneqq {{\rm dist}\,}(y,{\operatorname{graph}}\tilde f),\ \
d_2(y)\coloneqq {{\rm dist}\,}(y,{\operatorname{graph}}\tilde g),
\ \ d_3(y)\coloneqq |y|.$$ Functions $d_1$ and $d_2$ are DC on $\er^2$ by , $d_0$ and $d_3$ are convex and therefore DC on $\er^2$.
Using (for $K= 1/L, -1/L, 1/(2L), -1/(2L)$) the facts that the lines with the slopes $K$ and $-1/K$ are orthogonal and $M_0\subset \{(u,v): u\geq 0,\; -Lu\leq v \leq Lu \}$, easy geometrical observations show that $$\label{mjdt}
\tilde d(y)= d_i(y)\ \ \text{ if}\ \ y \in M_i,\ 0\leq i \leq 3,$$ and so Lemma \[vldc\] (iv) implies that $\tilde d$ is DC.
Now pick an arbitrary $y\in \er^2\setminus M_0=M_1\cup M_2\cup M_3$ and choose $\xi=(\xi_1,\xi_2)\in\partial_C \tilde d(y)$. Using , we obtain that if $y\in M_3$ then $\xi=\frac{y}{|y|}$ and so $|\xi|=1$. Using Lemma \[lem:distranceFromGraphWR\], we obtain that if $y\in M_1\cup M_2$, then $|\xi|\geq |\xi_1|\geq \frac{1}{\sqrt{4L^2+1}}$.
Now, since $d=\tilde d$ on $U\left( 0,\frac{\rho}{3} \right)$ both \[cond:DC\] and \[cond:weaklyRegular\] follow.
It remains to prove \[cond:DC\] and \[cond:weaklyRegular\] if case \[cond:manyConesNew\] from Theorem PRZ holds. Let $C_i$, $i=1,\dots, k$, be the open DC sectors as in \[cond:manyConesNew\]. Denote $A_i\coloneqq \er^2\setminus C_i$ and define $\delta_i\coloneqq {{\rm dist}\,}(\cdot,A_i)$, $i=1,\dots, k$.
Note that, for $y\in U\left( 0,\frac{\rho}{3} \right)$, one has $$d(y)=
\begin{cases}
\delta_i(y)& \text{if}\; y\in C_i,\\
0&\text{if}\; y\in M.
\end{cases}$$ Therefore (by Lemma \[vldc\] (iv)) it is enough to prove that \[cond:DC\] and \[cond:weaklyRegular\] hold with $d$ and $M$ being replaced by $\delta_i$ and $A_i$, respectively ($i=1,\dots, k$). Fix some $i\in \{1,\dots, k\}$. Without any loss of generality we can assume that $C_i$ is an basic open DC sector of radius $\rho$ with corresponding $f_i$ and $\omega_i$. Now define $$\tilde f_i(u)\coloneqq
\begin{cases}
f_i(u)&\text{if}\; u\in[-\rho,\rho],\\
f_i(-\rho)&\text{if}\; u<-\rho,\\
f_i(\rho)&\text{if}\; u>\rho.
\end{cases}$$ Then $\tilde f_i$ is Lipschitz and DC on $\er$. Put $\tilde d_i(y)={{\rm dist}\,}(y,{\operatorname{graph}}\tilde f_i)$. Then $\tilde d_i$ is DC by and $0$ is a weakly regular value of $\tilde d_i$ by Lemma \[lem:distranceFromGraphWR\]. And since $d_i=\tilde d_i$ on $U\left( 0,\frac{\rho}{3} \right)$ we are done.
Since $d$ is locally DC on $\er^2\setminus M$ by and on the interior of $M$ (trivially), (a) implies that $d$ is locally DC and so DC by Lemma \[vldc\] (ii). Further, (b) immediately implies that $0$ is a weakly regular value of $d$ and thus $d=d_M$ is an aura for $M$.
By Theorem \[thm:distanceAura\], in $\er^2$ locally ${{W\! DC}}$ sets and WDC sets coincide. This gives a partial answer to the part of [@PR Problem 10.2] which asks whether the same is true in each ${{\mathbb R}}^d$.
Complexity of the system of WDC sets {#compl}
====================================
In the following, we will work in each moment in an ${{\mathbb R}}^d$ with a fixed $d$, and so for simplicity we will use the notation, in which the dependence on $d$ is usually omitted.
The space of all nonempty compact subsets of ${{\mathbb R}}^d$ equipped with the usual Hausdorff metric $\rho_H$ is denoted by ${\mathcal{K}}$. It is well-known (see, e.g., [@Sri98 Proposition 2.4.15 and Corollary 2.4.16]) that ${\mathcal{K}}$ is a separable complete metric space. For a closed set $M\subset \er^d$, we set ${\mathcal{K}}(M)\coloneqq \{K \in {\mathcal{K}}: K \subset M\}$, which is clearly a closed subspace of ${\mathcal{K}}$. The set of all nonempty compact ${{W\! DC}}$ subsets of $M\subset \er^d$ will be denoted by ${{W\! DC}}(M)$.
In this section, we will prove the following theorem.
\[complwdc\]
1. $WDC({{\mathbb R}}^2)$ is an $F_{\sigma\delta\sigma}$ subset of ${\mathcal{K}}({{\mathbb R}}^2)$.
2. $WDC({{\mathbb R}}^d)$ is an analytic subset of ${\mathcal{K}}({{\mathbb R}}^d)$ for each $d \in \en$.
Before the proof of this theorem, we introduce some spaces, make a number of observations, and prove a technical lemma.
First observe that $WDC({{\mathbb R}}^d)= \bigcup_{n=1}^{\infty} WDC(B(0,n))$ and so, to prove Theorem \[complwdc\], it is sufficient to prove that, for each $r>0$, $$\begin{aligned}\label{staci2}
&\text{for $d=2$ (resp. $d \in \en$), $WDC(B(0,r))$}\\
&\text{is an $F_{\sigma\delta\sigma}$ (resp. analytic) subset of ${\mathcal{K}}(B(0,r))$.}
\end{aligned}$$
Further observe that it is sufficient to prove for $r=1$. Indeed, denoting $H(x)\coloneqq x/r,\ x \in {{\mathbb R}}^d$, it is obvious that $H^*: K \mapsto H(K)$ gives a homeomorphism of ${\mathcal{K}}(B(0,r))$ onto ${\mathcal{K}}(B(0,1))$ and $H^*(WDC(B(0,r))=WDC(B(0,1))$ (clearly $f$ is an aura for $K$ if and only if $f \circ H^{-1}$ is an aura for $H^*(K)$).
To prove for $r=1$, we will consider the space $X$ of all $1$-Lipschitz functions $f:B(0,4)\to [0,4]$ such that $f\geq 1$ on $B(0,4) \setminus U(0,3)$, equipped with the supremum metric $\rho_{\sup}$. Obviously, $X$ is a closed subspace of $C(B(0,4))$ and so it is a separable complete metric space.
The motivation for introducing $X$ is the fact that $$\label{moti}
\text{if $K\in {\mathcal{K}}(B(0,1))$, then $f_K\coloneqq d_K\restriction_{B(0,4)} \in X$.}$$ Since we are interested in $K \in WDC(B(0,1))$, we define also two subspaces of $X$: $$A\coloneqq \{f\in X:\ 0\ \ \text{ is a weakly regular value of }\ f|_{U(0,4)}\},$$ $$D\coloneqq \{f\in X:\ f=g-h\ \ \text{for some convex Lipschitz functions}\ \ g, h\ \ \text{on}\ \ B(0,4)\}.$$ Their complexity is closely related to the complexity of $WDC(B(0,1))$, as the following lemma indicate.
\[lem:auraINXd\] Let $\emptyset \neq K\subset B(0,1)\subset\er^d$ be compact. Then:
1. K is WDC if and only if there is a function $g\in D\cap A$ such that $K=g^{-1}(0)$.
2. If $d=2$, then K is WDC if and only if $f_K\coloneqq d_K\restriction_{B(0,4)} \in D \cap A$.
(i) Suppose first that $K$ is WDC and $f$ is an aura for $K$. Using Lemma \[vldc\](iii), we can chose $\alpha>0$ so small that the function $\alpha f$ is $1$-Lipschitz on $B(0,4)$ and $0\leq \alpha f(x) \leq 4$ for $x \in B(0,4)$. Set $$h(x)\coloneqq \max(|x|-2, \alpha f(x)), \ x \in {{\mathbb R}}^d,\ \ \text{and}\ \ g\coloneqq f\restriction_{B(0,4)}.$$ Then clearly $K=g^{-1}(0)$. Since $h$ is DC on ${{\mathbb R}}^d$ by Lemma \[vldc\](i), we obtain $g \in D$ by Lemma \[vldc\](iii). Finally, $g \in A$, since $g=\alpha f$ on $U(0,2)$.
Conversely, suppose that $K=g^{-1}(0)$ for some $g\in A\cap D$ and set $$f(x)\coloneqq
\begin{cases}
\min(g(x), 1), & \mbox{if } x\in U(0,4),\\
1 & \mbox{otherwise}.
\end{cases}$$ Since $f$ is DC on $U(0,4)$ by Lemma \[vldc\] (i) and $f=1$ on ${{\mathbb R}}^d \setminus B(0,3)$, we see that $f$ is locally DC and so DC by Lemma \[vldc\] (ii). Since $0$ is clearly a weakly regular value of $f$, we obtain that $f$ is an aura for $K$.
(ii) If $K$ is WDC, first note that $f_K \in X$ (see ). Since $d_K$ is an aura for $K$ by Theorem \[thm:distanceAura\], we obtain immediately that $f_K \in A$, and also $f_K \in D$ by Lemma \[vldc\] (iii).
If $f_K \in A \cap D$, then $K$ is WDC by (i).
For the application of Lemma \[lem:auraINXd\](ii) we need the simple fact that $$\label{spojpsi}
\text{$\Psi: K \mapsto f_K$, \ $K \in {\mathcal{K}}(B(0,1))$, \ \text {is a continuous mapping into}\ \ $X$.}$$ Indeed, if $K_1, K_2 \in {\mathcal{K}}(B(0,1))$ with $\rho_H(K_1, K_2) <{\varepsilon}$ and $x \in B(0,4)$, then clearly $ d_{K_1}(x) < d_{K_2}(x) + {\varepsilon}$, $ d_{K_2}(x) < d_{K_1}(x) + {\varepsilon}$, and consequently $\rho_{\sup}(f_{K_1}, f_{K_2}) \leq {\varepsilon}$.
Further observe that $$\label{bod}
D\ \ \text{is an}\ \ F_{\sigma}\ \ \text{subset of}\ \ \ X.$$ To prove it, for each $n \in \en$ set $$C_n\coloneqq \{g\in C(B(0,4)):\ g \ \text{is convex}\ n\text{-Lipschitz and}\ \ |g(x)| \leq 4n+4,\ x \in B(0,4)\}.$$ Now observe that if $f \in D$ then we can choose $n \in \en$ and convex $n$-Lipschitz functions $g$, $h$ such that $f=g-h$, $g(0)=0$ and consequently $\|g\|\leq 4n$, $\|h\|\leq 4n+4$, and so $g,\, h \in C_n$. Consequently, $D = X \cap \bigcup_{n=1}^{\infty} (C_n-C_n)$. Each $C_n$ is clearly closed in $C(B(0,4))$ and so it is compact in $C(B(0,4))$ by the Arzel\` a-Ascoli theorem. Consequently also $C_n-C_n = \sigma(C_n \times C_n)$, where $\sigma$ is the continuous mapping $\sigma: (g,h) \mapsto g-h$, is compact, and follows.
The most technical part of the proof of Theorem \[complwdc\] is to show that $A$ is an $F_{\sigma\delta\sigma}$ subset of $X$. To prove it, we need some lemmas.
\[swr2\] Let $f\in X$. Then $f\in A$ if and only $$\label{wr2}
\exists\, 0<{\varepsilon}\;
\forall x\in f^{-1}(0,{\varepsilon}),\; \nu\in\partial_C f(x)\;
:|\nu|\geq {\varepsilon}.$$
If holds, then we easily obtain $f \in A$ directly from the definition of a weakly regular value.
To prove the opposite implication, suppose that $f \in A$ and does not hold. Then there exist points $x_n \in f^{-1} (0, 1/n),\ n\in \en,$ and $\nu_n\in\partial_C f(x_n)$ such that $|\nu_n| < 1/n$. Choose a subsequence $x_{n_k} \to x \in B(0,4)$. Since $0\leq f(x_{n_k}) < 1/n_k$, we have $f(x_{n_k}) \to f(x)=0$, and consequently $x \in U(0,4)$. Since $\nu_{n_k} \to 0$, we obtain that $0$ is not a weakly regular value of $f|_{U(0,4)}$, which contradicts $f \in A$.
Denote $\qp\coloneqq \q\cap (0,1)$ and for every ${\varepsilon}\in\qp$ and $d\in\en$ pick a finite set $\S^d_{{\varepsilon}}\subset S^{d-1}$ such that for every $v\in S^{d-1}$ there is some $\nu\in\S^d_{{\varepsilon}}$ satisfying $|v-\nu|< {\varepsilon}$.
\[lem:A\_dFormula\] Let $f$ be a function from $X$. Then $f\in A$ if and only if $$\label{eq:A_dFormula}
\begin{aligned}
\exists {\varepsilon}\in\qp\;
\forall p,q\in\qp, 0<p<q<{\varepsilon}\;
\exists \rho\in\qp \;
\forall x\in U(0,4):\;\\
(f(x)\notin (p,q) \vee
\exists\nu\in\S^d_{{\varepsilon}}\,\forall y\in U(x,\rho), 0<\alpha<\rho: f(y+\alpha \nu)-f(y)\leq-{\varepsilon}\alpha).
\end{aligned}$$
First suppose that holds and choose ${\varepsilon}\in \qp$ by . We will show that $$\label{hv}
\forall x\in f^{-1}(0,{\varepsilon}),\; \nu\in\partial_C f(x)\;
:|\nu|\geq {\varepsilon}.$$ To this end, consider an arbitrary $x\in f^{-1}(0,{\varepsilon})$ and choose $p,q\in\qp$ such that $0<p<q<{\varepsilon}$ and $ f(x) \in (p,q)$. Choose $\rho \in \qp$ which exists for ${\varepsilon}, p, q$ by . So there exists $ \nu\in\S^d_{{\varepsilon}}$ such that $$\forall y\in U(x,\rho), 0<\alpha<\rho: f(y+\alpha v)-f(y)\leq-{\varepsilon}\alpha.$$ Therefore Lemma \[clod\] (ii) gives that $|\nu|\geq {\varepsilon}$ for each $\nu\in\partial_C f(x)$. Thus holds and so $f\in A$ by Lemma \[swr2\].
Now suppose $ f \in A$. Using , we can choose ${\varepsilon}\in\qp$ such that $$\label{4hv}
\forall x\in f^{-1}(0,{\varepsilon}),\; \nu\in\partial_C f(x)\;
:|\nu|\geq 4 {\varepsilon}.$$ To prove , consider arbitrary $ p,q\in\qp$, $0<p<q<{\varepsilon}$. Using Lemma \[clod\] (i), we easily obtain that for each $z \in K\coloneqq f^{-1}([p,q])$ there exist $\rho(z)>0$ and $v(z)\in S^{d-1}$ such that $$\label{ct}
\forall y\in U(z,\rho(z)), 0<\alpha<\rho(z): f(y+\alpha v(z))-f(y)\leq-2{\varepsilon}\alpha.$$ Choose $\rho \in \qp$ as a Lebesgue number (see [@E]) of the open covering $\{U(z,\rho(z))\}_{z\in K}$ of the compact $K$. For an arbitrary $x \in U(0,4)$, either $f(x)\notin (p,q)$ or $x \in K$. In the second case, by the definition of Lebesgue number, there exists $z\in K$ such that $U(x,\rho)\subset U(z,\rho(z))$. Then clearly $\rho < \rho(z)$ and so implies $$\label{ct2}
\forall y\in U(x,\rho), 0<\alpha<\rho: f(y+\alpha v(z))-f(y)\leq-2{\varepsilon}\alpha.$$ By the choice of $\S_{\varepsilon}^d$ there is some $\nu\in\S^d_{\varepsilon}$ such that $|v(z)-\nu| < {\varepsilon}$. By , for each $y\in U(x,\rho)$ and $0<\alpha< \rho$, $$f(y+\alpha v(z))-f(y)\leq-2{\varepsilon}\alpha.$$ Consequently, using $1$-Lipschitzness of $f\in X$, we obtain $$\begin{aligned}
f(y+\alpha \nu)-f(y)&\leq f(y+\alpha v(z))-f(y)+|f(y+\alpha \nu)-f(y+\alpha v(z))|\\
&\leq f(y+\alpha v(z))-f(y) + |\nu-v(z)|\alpha
\leq -2{\varepsilon}\alpha +{\varepsilon}\alpha
= - {\varepsilon}\alpha,
\end{aligned}$$ and so holds.
\[cor:AFsigmaDeltaSigma\] The set $A$ is an $F_{\sigma\delta\sigma}$ subset of $X$.
For each quadruple $y \in \er^d$, $\nu \in S^{d-1}$, $\alpha>0$, ${\varepsilon}>0$ we set $$C(y,\nu,\alpha, {\varepsilon}) \coloneqq \{ f \in X:\ f(y+\alpha \nu)-f(y)\leq-{\varepsilon}\alpha\}.$$ (Of course, we have $C(y,\nu,\alpha, {\varepsilon})= \emptyset$ if $y \notin U(0,4)$ or $y+\alpha \nu\notin U(0,4)$.) Further, for each triple $x\in U(0,4)$, $0<p<q$, we set $$D(x,p,q)\coloneqq \{ f \in X:\ f(x) \notin (p,q)\}.$$ It is easy to see that both $C(y,\nu,\alpha, {\varepsilon})$ and $ D(x,p,q)$ are always closed subsets of $X$. It is easy to see that Lemma \[lem:A\_dFormula\] is equivalent to $$\begin{aligned}
A=\bigcup_{ {\varepsilon}\in\qp}\;
\bigcap_{\substack{p,q\in\qp,\\0<p<q<{\varepsilon}} } \;
\bigcup_{ \rho\in\qp} \;
\bigcap_{ x\in U(0,4)}\;
\left(D(x,p,q)\ \cup \
\bigcup_{ \nu\in\S^d_{{\varepsilon}}}\ \bigcap_{\substack{y\in U(x,\rho),\\ 0<\alpha<\rho} } \ C(y,\nu,\alpha, {\varepsilon}) \right).
\end{aligned}$$ Therefore, since $\qp$ is countable and each $\S^d_{{\varepsilon}}$ is finite, we obtain that $A$ is an $F_{\sigma\delta\sigma}$ subset of $X$.
[**The proof of Theorem \[complwdc\].**]{} We know that it is sufficient to prove for $r=1$.
Suppose $d=2$. Then Lemma \[lem:auraINXd\](ii) gives that $WDC(B(0,1))= \psi^{-1}(A \cap D)$, where $\psi: {\mathcal{K}}(B(0,4)) \to X$ is the continuous mapping from . Since $A\cap D$ is an $F_{\sigma\delta\sigma}$ subset of $X$ by Corollary \[cor:AFsigmaDeltaSigma\] and , we obtain for $r=1$ and $d=2$, and thus also assertion (i) of Theorem \[complwdc\].
To prove assertion (ii) of Theorem \[complwdc\], it is sufficient to prove that (in each ${{\mathbb R}}^d$) $WDC(B(0,1))$ is an analytic subset of ${\mathcal{K}}(B(0,1))$. To this end, consider the following subset $S$ of ${\mathcal{K}}(B(0,1)) \times X$: $$S\coloneqq \{(K,f)\in {\mathcal{K}}(B(0,1)) \times X:\ f^{-1}(0)=K ,\ f\in A \cap D\}.$$ By Lemma \[lem:auraINXd\](i), $WDC(B(0,1)) = \pi_1(S)$ (where $\pi_1(K,f)\coloneqq K$) and so it is sufficient to prove that $S$ is Borel. Denoting $$Z\coloneqq \{(K,f)\in {\mathcal{K}}(B(0,1)) \times X:\ K = f^{-1}(0),\ f\in X\},$$ we have $S= Z \cap ({\mathcal{K}}(B(0,1)) \times (A\cap D))$. So, since $A\cap D$ is Borel by Corollary \[cor:AFsigmaDeltaSigma\] and , to prove that $S$ is Borel, it is sufficient to show that $Z$ is Borel in ${\mathcal{K}}(B(0,1)) \times X$. To this end, denote for each $n\in \en$ $$P_n\coloneqq \{ (K,f) \in {\mathcal{K}}(B(0,1)) \times X:\ \exists x \in K: f(x) \geq 1/n\},$$ $$Q_n\coloneqq \{ (K,f) \in {\mathcal{K}}(B(0,1)) \times X:\ \exists x \in B(0,4): \ {{\rm dist}\,}(x,K)\geq 1/n, f(x)=0\}.$$ Since clearly $$Z= ({\mathcal{K}}(B(0,1)) \times X) \setminus (\bigcup_{n=1}^{\infty} P_n \cup \bigcup_{n=1}^{\infty} Q_n),$$ it is sufficient to prove that all $P_n$ and $Q_n$ are closed.
So suppose that $(K_i,f_i) \in {\mathcal{K}}(B(0,1)) \times X, \ i=1,2,\dots,$ $(K,f) \in {\mathcal{K}}(B(0,1)) \times X$, $\rho_H(K_i,K) \to 0$ and $\rho_{\sup}(f_i,f) \to 0$.
First suppose that $n\in \en$ and all $(K_i,f_i) \in P_n$. Choose $x_i \in K_i$ with $f_i(x_i) \geq 1/n$. Choose a convergent subsequence $x_{i_j}\to x \in {{\mathbb R}}^d$. It is easy to see that $x \in K$. Since $|f_{i_j}(x_{i_j}) - f(x_{i_j})| \to 0$ and $f(x_{i_j}) \to f(x)$, we obtain $f_{i_j}(x_{i_j})\to f(x)$, and consequently $f(x)\geq 1/n$. Thus $(K,f) \in P_n$ and therefore $P_n$ is closed.
Second, suppose that $n\in \en$ and all $(K_i,f_i) \in P_n$. Choose $x_i \in B(0,4)$ such that ${{\rm dist}\,}(x_i,K_i)\geq 1/n$ and $f_i(x_i)=0$. Choose a convergent subsequence $x_{i_j}\to x \in B(0,4)$. Since $|f_{i_j}(x_{i_j}) - f(x_{i_j})| \to 0$ and $f(x_{i_j}) \to f(x)$, we obtain $f(x)=0$. Now consider an arbitrary $y \in K$ and choose a sequence $y_j \in K_{i_j}$ with $y_j \to y$. Since $|x_{i_j} - y_j| \geq 1/n$ and $x_{i_j}\to x$, we obtain that $|y-x|\geq 1/n$ and consequently ${{\rm dist}\,}(x,K) \geq 1/n$. Thus $(K,f) \in Q_n$ and therefore $Q_n$ is closed.
[99]{}
V. Bangert: Sets with positive reach, [*Arch. Math.*]{} (Basel) [**38**]{} (1982), 54–57.
P. Cannarsa, C. Sinestrari: Semiconcave Functions, Hamilton-Jacobi Equations, and Optimal Control, Progress in Nonlinear Differential Equations and their Applications 58, Birkhäuser, Boston (2004).
F. Clarke: [*Optimization and Nonsmooth Analysis.*]{} SIAM, Philadelphia, 1990
R.A. DeVore, G.G Lorentz: *Constructive Approximation*, Springer, 1993.
R. Engelking: *General Topology* (rev. and compl. ed.), Heldermann Verlag, Berlin, 1989.
J.G.H. Fu: Tubular neighborhoods in Euclidean spaces, [*Duke Math. J.*]{} [**52**]{} (1985), no. 4, 1025–1046
J.G.H. Fu: Integral geometric regularity, In: Kiderlen, M., Vedel Jensen, E.B. (eds.) Tensor Valuations and Their Applications in Stochastic Geometry and Imaging, 261–299, Lecture Notes in Math. 2177, Springer, 2017.
J.G.H. Fu, D. Pokorný, J. Rataj: Kinematic formulas for sets defined by differences of convex functions, [*Adv. Math.*]{} [**311**]{} (2017), 796–832
P. Hartman: On functions representable as a difference of convex functions. [*Pacific J. Math.*]{} [**9**]{} (1959), 707-713
D. Pokorný, J. Rataj: Normal cycles and curvature measures of sets with d.c. boundary. [*Adv. Math.*]{} [**248**]{}, (2013), 963–985 DOI:10.1016/j.aim.2013.08.022
D. Pokorný, J. Rataj, L. Zaj' iček: On the structure of WDC sets. [*Math. Nachr.*]{} (online) DOI:10.1002/mana.201700253
D. Pokorný, L. Zaj' iček: On sets in ${{\mathbb R}}^d$ with DC distance function, preprint (https://arxiv.org/abs/1904.12223).
J. Rataj, L. Zaj' iček: On the structure of sets with positive reach. [*Math. Nachr.*]{} [**290**]{} (2017), 1806–1829
S.M. Srivastava: [*A course on Borel sets*]{}, Graduate Texts in Mathematics 180, Springer-Verlag, New York, 1998.
H. Tuy: [*Convex Analysis and Global Optimization*]{}, 2nd ed., Springer Optimization and Its Applications 110, Springer, 2016.
L. Veselý, L. Zajíček: Delta-convex mappings between Banach spaces and applications, [*Dissertationes Math.*]{} (Rozprawy Mat.) [289]{} (1989), 52 pp.
M. Zähle: Curvature measures and random sets, II. [*Probab. Theory Relat. Fields*]{} [**71**]{} (1986), 37-–58
[^1]: The research was supported by GAČR 18-11058S
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Cooperative behaviour is widespread in nature, but explaining how cooperation evolves constitutes a major scientific challenge. Computer simulations have shown that social network structure plays a key role for the viability of cooperation. However, not much is known about the importance of correlations between cooperative strategy and network position for the evolution of cooperation. Here, we investigate how cooperation is affected by correlations between cooperative strategy and individual connectedness in different networks, by means of simulation models. We find that placing cooperators on well-connected (high-degree) nodes can - as expected - enhance cooperation, but will only do so under certain conditions. The effect depends nontrivially on the combination of the social network structure (degree distribution and assortativity), the type of social dilemma, and the presence of *Trojan horses*, i.e. defectors placed within clusters of cooperators. Interestingly, the largest enhancement is observed in Poisson networks with increased degree assortativity, where cooperation does better than under scale-free degree distribution. The study provides new knowledge about the conditions under which cooperation may evolve and persist. The results are furthermore important to consider in regard to cooperation experiments, as they imply that stochastic initial correlations between cooperativeness and network position can affect the results of such experiments.\
**Keywords:** Evolution of cooperation, cooperator positions, evolutionary game theory, degree assortativity, social networks, social connectedness
author:
- Josefine Bohr Brask
- Jonatan Bohr Brask
bibliography:
- 'coop\_networks\_initcons.bib'
title: 'Evolution of cooperation in networks: well-connected cooperators can enhance cooperation but are counteracted by Trojan horses'
---
[^1]
Introduction
============
Cooperation, understood as behaviour where individuals help others at a cost to themselves, has long constituted a major conundrum for science. Cooperative behaviour seemingly contradicts the central prediction of Darwinian evolutionary theory that individuals will behave in ways that maximise their own gain. Nevertheless, cooperation is found widely across species, including humans, other mammals, fish, birds, insects, and microscopic organisms [@Carter2013; @Rutte2007; @Milinski1987; @Voelkl2015; @Queller1998; @Turner1999]. Explaining the evolution of cooperation has been called one of the biggest scientific challenges of our time [@Pennisi2005].
During recent decades, it has become clear that social network structure (i.e. who interacts with whom) plays an essential role for understanding the evolution of cooperation. Computer simulations have shown that while unstructured populations are not conducive to cooperation, certain network structures can promote and stabilise it (e.g. [@Nowak1992; @Santos2005; @Santos2006evol; @Ohtsuki2006; @Assenza2008]). Across animal species, social networks found in nature are indeed structured [@Krause2015], and the study of cooperation in networks thus provides a promising avenue towards understanding how cooperation may evolve in nature.
A large body of work has investigated the effect of different structural network parameters on the evolution of cooperation via computer simulation (see [@Szabo2007; @Roca2009] for reviews). Some factors which have been found to have important effects include degree distribution [@Santos2005; @Santos2006], degree heterogeneity [@Santos2006evol; @Santos2012], average degree [@Ohtsuki2006], degree assortativity [@Rong2007], network clustering [@Assenza2008], and modularity [@Voelkl2009]. However, while the role of network structure in the evolution of cooperation thus has been extensively investigated, little is known about the role of the positions of cooperative and defective individuals in the network (in other words, the role of correlations between strategy and node properties such as connectedness; [@Chen2008]). Real-world social networks across species are characterized by considerable heterogeneity in the social positions of individuals, where some are more socially connected than others. If cooperative individuals tend to have specific social network positions, this may affect the spread and persistence of cooperation. In particular, it may intuitively be expected that cooperation should fare better in situations where cooperative individuals have more social connections than defective individuals, as their higher connectedness could potentially help them propagate their strategy. However, our knowledge about the importance of such effects for the evolution of cooperation is currently very limited.
Here, we investigate how correlations between strategy and connectedness affect the evolution of cooperation in different types of networks, by means of computer simulations based on evolutionary game theory. This can elucidate the importance of the network positions of cooperative and defective individuals, and thereby increase our general understanding of the conditions under which cooperation can persist. The investigations are also relevant for situations where social networks are constructed. For example, in experiments with humans playing cooperation games in artificial network structures (e.g. [@Cassar2007; @Grujic2010; @Gracia2012; @Rand2014]), initial stochastic correlations between cooperativeness and network position could potentially have a significant effect and could influence the conclusions of the experiments, in particular because the number of replications in such experiments for practical reasons is typically low.
Evolutionary game theory provides a common framework for the study of cooperation [@Smith1973; @Smith1982]. Here, the interaction between individuals is formalised in terms of games where each player adopts one of a limited number of strategies, such as *cooperative* or *defective* (selfish). The game is played repeatedly and players adapt their strategies to optimise their performance. Alternatively, each iteration can be interpreted as a reproductive generation, so that the adaptation is genetic. With this framework, it becomes possible to study the stability and dynamics of cooperation in populations over time (including evolutionary time).
Most of our current knowledge about the role of social structure for the evolution of cooperation comes from simulation of game-theoretical models in networks [@Szabo2007; @Roca2009]. In such models, each individual is represented by a network node, and the links in the network determine who interacts with whom. A key outcome of the models is the frequency of individuals adopting a cooperative strategy (cooperators) in the population over time. The simulations most often begin with the strategies being randomly assigned to the nodes. In other words, usually cooperators and defectors are initially randomly positioned in the network [@Szabo2007; @Roca2009].
We specifically study the effect of correlations between strategy and node degree (i.e. the number of links connected to the node). We focus on degree because this is a fundamental measure of network position that is easily interpretable (as the number of interaction partners an individual has). To study this, we run simulations where cooperators are initially placed on high-degree nodes (stochastically or deterministically), and compare the results to those of corresponding simulations where the cooperators are placed randomly in the network. We note that the simulations can be interpreted both as: 1) simulating systems where the correlations between strategy and network position is a result of the networks being deliberately constructed, for example as in the above-mentioned human cooperation experiments; and 2) simulating systems where the correlations between strategy and network position is a result of previous evolution. Thus, the positioning of cooperators on well-connected nodes can be viewed either as the system’s actual initial conditions, or as a transient state. In either case, the simulations show how cooperators’ occupation of well-connected nodes affects subsequent strategy evolution.
We study the effects of the cooperator positions in Poisson networks (also called random networks) and scale-free networks, two network types commonly used in models of cooperation and characterised by different degree distributions [@Szabo2007; @Roca2009]. In Poisson networks, most individuals have degrees close to the mean degree, whereas in scale-free networks, most individuals have low degrees and a few individuals have very high degrees. For each type of network, we use standard versions of the networks as well as versions with increased degree assortativity (i.e. where nodes with the same number of links are preferentially connected to each other; also known as degree correlation or degree assortment). Real-world social networks are generally characterised by increased degree assortativity [@Newman2002], and such assortativity is particularly likely to affect the evolution of cooperation when strategy is correlated to degree, because it then affects the extent to which cooperators are connected to each other. The networks with increased degree assortativity are therefore likely to give us results that are more relevant for real social networks, whereas the standard networks provide us with results that are more directly comparable to other studies of cooperation in networks [@Szabo2007; @Roca2009].
We study the evolution of cooperation in these networks for two commonly used formalisations of cooperative interactions, the Prisoner’s Dilemma game and the Snowdrift game. The Prisoner’s Dilemma game formalises a situation where behaving cooperatively in itself is not beneficial to the actor, and cooperation cannot survive in a well-mixed population (without special mechanisms). This game embodies the paradox of the evolution of cooperation. The Snowdrift game models a weaker social dilemma where behaving cooperatively in itself provides a benefit to the actor, and a well-mixed population in equilibrium can contain both cooperators and defectors.
Methods
=======
In order to investigate the effect of correlations between strategy and connectedness on the evolution of cooperation, we built an agent-based simulation model. In the following we summarize the model framework, and we describe the networks, the strategy assignment rules used to place the cooperators and defectors in the networks, and our simulation procedures. The general model framework follows commonly used standards for simulations of cooperation in networks. For further details and mathematical description of the model, we refer the reader to Section V.
Simulation model summary
------------------------
The model simulates the evolution of cooperation in networks, where each node in the network corresponds to an individual and the links between the nodes determine who interacts with who. Each individual is either a cooperator or defector. While the network structure does not change throughout a simulation, the individuals change their strategies over time, and the main outcome of the simulation is the frequency of cooperators in the population after a set number of timesteps. Each timestep consists of an *interaction phase*, where all individuals connected by a direct link interact pairwise, and an *update phase*, where all individuals update their strategy adaptively.
In the interaction phase, each indiviual plays a single round of a cooperation game with each of its network neighbours (i.e. the individuals to whom it has a direct link), and gains game payoffs that depend on its own strategy and that of the other player. Each individual accumulates the payoffs from all its games within that timestep. Individuals follow their current strategy, i.e. cooperators only cooperate and defectors only defect. The payoff values for each combination of strategies are determined by the game, which is either the Prisoner’s Dilemma game or the Snowdrift game (i.e. separate simulations are run for each of the two games). Both games formalise a situation where it is of advantage for the individual to defect (i.e. a defector playing against a cooperator gains the highest payoff), but if both individuals defect they are worse off than if they both cooperate (formal definitions of the games can be found in Section V). To make our results comparable to those of seminal papers in the field (e.g. [@Nowak1992; @Hauert2004; @Santos2005; @Santos2006]), we use one-parameter versions of the two games, where the severity of the social dilemma (i.e. how hard it is for cooperation to evolve, everything else equal) is determined by a single parameter. In the Prisoner’s Dilemma, this parameter is called $b$ and corresponds to the payoff a defector gains from playing against a cooperator (i.e. the *benefit to defectors*); in the Snowdrift game the parameter is called $\rho$ and corresponds to the cost-to-benefit ratio of mutual cooperation (see Section V for details).
The simulation proceeds to the update phase when all network neighbours have interacted. Here, each individual decides whether to change its strategy, based on how well it did in the interaction phase in terms of its accumulated game payoffs (i.e. its *fitness*). To make this decision, the individual follows the *proportional imititation* update rule ([@Santos2005; @Hauert2004], formalization given in Section V) and compares its own fitness to that of a randomly picked network neighbour. If the individual’s fitness is higher than or equal to that of the neighbour, it keeps its current strategy. If it has lower fitness than the neighbour, it copies the strategy of the neighbour with a probability that is proportional to the difference between the neighbour’s fitness and its own.
Networks
--------
We use four types of networks: standard versions of Poisson and scale-free networks, and versions of these networks with the same degree distributions but with increased degree assortativity. The standard Poisson networks are of the Erdös-Renyi (ER) random network type [@Erdos1960] and the standard scale-free networks are of the Barabási-Albert (BA) type [@Barabasi1999]. To generate versions of these networks with increased degree assortativity, we apply the algorithm introduced by Xulvi-Brunet and Sokolov [@Xulvi2004], which preserves the degree distribution of the network (see Appendix A for details). All networks have $N=10^3$ nodes and an average degree of $\bar{d}=10$, and we use only networks where all nodes are contained in a single component, i.e. all individuals are at least indirectly connected to each other.
Strategy assignment
-------------------
In the beginning of each simulation, a subset of nodes are assigned the cooperator strategy and the rest the defector strategy. We create correlations between strategy and connectedness by using different strategy assignment methods. Denoting the total number of nodes by $N$ and the number of cooperators by $N_c$, the fraction of cooperators in the population is $r = N_c/N$. The initial fraction of cooperators is $r_{in}$. We take $r_{in} = 1/2$ (i.e. initially, half of the nodes are cooperators), and we use the following three strategy assignment rules:
- *Uniform*. $N r_{in}$ nodes are picked uniformly at random among all nodes and assigned the cooperator strategy. The remaining nodes are assigned the defector strategy. The probability for any given node of being a cooperator thus equals $r_{in}$. This strategy assignment method gives a baseline for our investigations, where there is no correlation induced between strategy and connectedness.
- *Stochastic by degree*. Nodes to be assigned the cooperator strategy are selected sequentially based on their relative degree. The first cooperator node is drawn among all nodes, with the probability of drawing node $i$ given by $d_i / \sum_j d_j$, where $d_i$ is the degree of node $i$. Each subsequent cooperator node is drawn from the remaining set of nodes according to $d_i / \sum_{j\notin\mathcal{C}} d_j$, where $\mathcal{C}$ is the set of nodes which have already been selected. This is iterated until $N r_{in}$ nodes have been assigned the cooperator strategy. The remaining nodes are assigned the defector strategy [^2]. This strategy assignment method creates correlation between strategy and connectedness, with cooperators placed preferentially on high-degree nodes, but with stochasticity in the placement.
- *Deterministic by degree*. The $N r_{in}$ nodes of highest degree are assigned the cooperator strategy while the rest are assigned the defector strategy. This strategy assignment method results in perfect correlation between strategy and connectedness, where all cooperators have higher degree than the defectors.
Simulation procedures
---------------------
We run simulations for all combinations of the four network types, the two games, and the three types of strategy assignment described above. For each of these combinations we run simulations for varying values of the game parameter $b$ or $\rho$ (i.e. for different severities of the social dilemma). We run 50 replications for each setting, and all simulations have a total of $t_{max}=10^4$ timesteps. The final fraction of cooperators for a given setting is calculated as the average fraction in the last 100 timesteps of the 50 replications.
An example of a simulation run (i.e. a single replication) is shown in [Fig. \[fig.cfracexample\]]{}. In this particular example, the cooperator fraction $r$ drops from the initial value of 0.5 to close to zero at the end of the simulation, i.e. cooperation approaches extinction. The example is for Prisoner’s Dilemma in a standard scale-free network, with *stochastic by degree* strategy assignment. The insets indicate that higher-degree nodes tend to be more likely to be cooperators, as expected under this strategy assignment method.
![An example of a single simulation run: the fraction of cooperator nodes over time in a standard scale-free (BA) network for Prisoner’s Dilemma with $b=2$ and initial strategy positions created by the *stochastic by degree* assignment rule. The red shaded region indicates the last 100 generations used to compute the final fraction of cooperators. *Insets:* Snapshots of the cooperator fraction vs. node degree at timesteps 1, 3000, and 10000, for degree 1-20 (above 20 there are only few nodes per degree). []{data-label="fig.cfracexample"}](cfrac_example.png){width="\columnwidth"}
Results
=======
We find that correlations between strategy and connectedness can have considerable effects on the evolution of cooperation, and the effect depends on the network structure, the social dilemma, and stochasticity in the strategy placement. Placing cooperators preferentially on well-connected (high-degree) nodes generally promotes cooperation in scale-free networks, whereas it has limited effects in the Poisson networks, with an interesting exception: placing cooperators on well-connected nodes can strongly promote cooperation in degree-assorted Poisson networks under the Prisoner’s Dilemma, but this effect is sensitive to *Trojan horses*, i.e. defectors placed within clusters of cooperators.
{width="\textwidth"}
The simulation results are shown in [Fig. \[fig.cfracres\]]{}. For the standard networks ([Fig. \[fig.cfracres\]]{}, top row of plots), the connectedness of cooperators has an effect on the evolution of cooperation only in scale-free networks, for both games. For standard scale-free (BA) networks (Fig. 2a,c), when the strategy placement is changed from uniform to degree-correlated (stochastically or deterministically), the average final cooperator fraction is increased for all $b>1$ for Prisoner’s Dilemma and for all $\rho>0.125$ for the Snowdrift game. For standard Poisson (ER) networks (Fig. 2b,d), changing the type of strategy placement has negligible effect for both games. Also, for each of the standard networks, the final cooperator fractions for the stochastic and deterministic degree-correlated placements are almost indistinguishable. In these networks, the effect of the cooperator positions thus mainly depends on the degree distribution (scale-free vs. Poisson), with similar results for the two games and the two types of degree-correlated strategy placement.
The picture is somewhat different for the networks with increased degree assortativity ([Fig. \[fig.cfracres\]]{}, bottom row of plots), which corresponds to more realistic social networks [@Newman2002]. For degree-assorted scale-free networks, both of the degree-correlated cooperator placements enhance cooperation for Prisoner’s Dilemma, with the deterministic distribution having a larger effect (Fig. 2e). For the Snowdrift game, however, the effects in this network type are small and inconsistent (Fig. 2g). For degree-assorted Poisson networks, degree-correlated cooperator placement does not affect cooperation when there is stochasticity in the placement (Fig. 2f,h). In contrast, the deterministic degree-correlated cooperator placement has a large, positive effect for all $b>1$ for Prisoner’s Dilemma, and also enhances cooperation for large $\rho$ for the Snowdrift game. Hence, for the networks with increased degree assortativity (where individuals of similar connectedness cluster together), the effect of placing cooperators on high-degree nodes depends non-trivially on the combination of degree distribution, the game, and the presence of stochasticity in the placement.
Discussion
==========
Our results support the expectation that well-connected cooperators can enhance the spread and persistence of cooperation, but also show that they only do so under certain conditions.
Enhancement of cooperation when cooperators are positioned on well-connected nodes stems from the fact that individuals of higher degree have higher maximal fitness, as a higher number of social partners means increased opportunity to gain benefits from cooperative interactions (c.f. Eq. in Section V). Since each individual copies strategies only from neighbours with higher fitness, a cooperator can turn a defector neighbour into a cooperator only if it has higher fitness than the defector. Hence, intuitively, a higher maximal fitness for cooperators is likely to facilitate the spread of the cooperative strategy.
However, we see that placing cooperators on high-degree nodes does not always enhance cooperation. To better understand when such placement will have an effect, we may ask how the social neighbourhoods of cooperators and defectors should differ in order for cooperators to attain higher fitness than defectors. The answer is that a cooperator generally needs to have considerably more cooperator neighbours than a defector, to get a higher fitness than the defector (see Appendix B for a derivation). A higher degree for the cooperator will not always lead to this requirement being satisfied, and does therefore not necessarily give the cooperator advantage over the defector. But placing cooperators on high-degree nodes increases the chance that the requirement is satisfied and cooperators can withstand invasion from defector neighbours and spread their own strategy. Furthermore, the cooperators’ higher number of links means that they are more likely to be picked as role models in the strategy-update phase (while this is not true in the limit of perfect assortativity, we checked that it holds true for all the network types used here). Combined with higher fitness this increases the cooperators’ control over their neighbourhoods.
Given the above considerations, the fact that we observe effects of correlation between strategy and connectedness more often in scale-free than in Poisson networks may be attributed primarily to the difference in degree variation between these networks. Scale-free networks are characterized by the presence of nodes with a degree much higher than the average (so-called hubs). This means that the placement of cooperators on high-degree nodes gives them a greater advantage in standard scale-free networks than in standard Poisson networks. Our results show that the cooperators’ benefits from higher degrees is too small to affect the final outcome in standard Poisson networks, whereas in standard scale-free networks the advantage cooperators obtain from being placed on hubs leads to considerable positive effects on the evolution of cooperation.
![Examples of Poisson networks with increased degree assortativity, showing positions of cooperators and defectors created by *stochastic by degree* (left) and *deterministic by degree* (right) strategy assignments. Node size indicates degree, and node color indicates strategy.[]{data-label="fig.networkplots"}](network_plots_light.png){width="\columnwidth"}
The difference in effect of cooperator positions between the two types of degree distribution is less clear when degree assortativity is increased, and the pattern of effects is overall less straightforward for the degree-assorted networks. This may be explained by what we will call the *Trojan-horse effect*. Under increased degree assortativity, nodes of similar degree are preferentially connected to each other, and placing cooperators on higher-degree nodes will therefore tend to create clusters of cooperators. Such clustering is in itself advantageous for cooperators, as they are then exposed to fewer defectors and obtain benefits from each other. However, direct connections between hubs also means that a defector hub can more easily exploit a cooperator hub, and this leads to degree assortativity generally having a net negative effect on cooperation [@Rong2007]. In our simulations, degree assortativity generally decreases cooperation ([Fig. \[fig.cfracres\]]{}) in line with previous findings [@Rong2007], with one notable exception, namely for the Prisoners Dilemma under Poisson degree distribution and deterministic placement of cooperators on high-degree nodes (Fig. 2f). Here, cooperation does considerably better in the degree-assorted network than in the corresponding standard network (Fig. 2b,f). Furthermore, in the degree-assorted network (Fig. 2f), cooperation is strongly enhanced under the deterministic placement compared to both of the other types of placement - the stochastic placement of cooperators on high-degree nodes has no observable effect here. This pattern is likely to stem from the fact that when the placement is deterministic, cooperators and defectors are neighbouring each other only in one of the clusters created by the increased degree assortativity, i.e. there is a clear invasion front, with the remaining cooperator clusters being shielded from defectors. In contrast, when the placement of the cooperators on high-degree nodes is not perfect, the higher-degree clusters of cooperators are likely to contain *Trojan horses* in the form of defectors that have by chance been placed on high-degree nodes. Hence, in this case invasion by defectors can not only happen from the invasion front but also from within the higher-degree cooperator clusters. The difference between the deterministic and stochastic placement of cooperators on high-degree nodes is shown in [Fig. \[fig.networkplots\]]{}. It can be seen that for the deterministic placement, cooperators and defectors only meet in one of the network clusters. In contrast, for the stochastic placement, there is no clear invasion front and even the highest-degree cooperator cluster contains Trojan horses. Increased degree assortativity thus tends to induce differences between the effect of stochastic and deterministic placement of cooperators on high-degree nodes. Our results show that the presence or absence of stochasticity in the strategy placement can have major effects on the evolution of cooperation, in particular in degree-assorted networks because this is where the Trojan-horse effect can make a difference.
Interestingly, the cooperator fractions in the degree-assorted Poisson network under the Prisoner’s dilemma and deterministic placement of cooperators on high-degree nodes are higher than in the corresponding scale-free network (Fig. 2e,f). This is opposite to the pattern found for standard Poisson and scale-free networks (Fig. 2a,b and previous studies, e.g. [@Santos2006]), where cooperation does much better under scale-free degree distribution than under Poisson distribution. This implies that the generally observed positive relationship between degree heterogeneity and cooperation [@Santos2006evol; @Santos2012] is under some conditions overruled.
Our study implies that correlation between strategy and network position is an important factor to consider both for our understanding of the evolution of cooperation in the real world, and when designing and conducting scientific investigations of cooperation.
Empirical experiments of humans playing cooperation games in constructed networks constitutes an important tool in the study of cooperation (e.g. [@Cassar2007; @Grujic2010; @Gracia2012; @Rand2014]). Given that the number of replicates for such experiments are for practical reasons often low, random variation in initial conditions can have significant effects. Our results imply that if cooperative individuals are by chance placed in more well-connected network positions, this may affect the experimental results. If extensive replication is not possible, such effects can therefore be important to consider when conducting network cooperation experiments.
In addition to experimental networks, real-world human networks may also sometimes be deliberately constructed (e.g. some collaboration networks). Our study suggests that cooperation could in such cases potentially be enhanced by placing cooperative individuals in central network positions, and that this will only work under certain conditions.
We have looked at the evolution of cooperation in static networks. This allowed us to investigate how the effect of correlations between strategy and connectedness depends on the network structure. While social systems of humans and many other species are dynamic in the sense that they consist of series of time-limited social interactions, temporal changes in the emerging social network structures (i.e. termination or emergence of social bonds) may be slow. Indeed, long-term stability in the structure of social networks have been found in multiple species [@prehn2019; @borgeaud2017; @kerth2011; @godfrey2013]. When changes in the real social structures are much slower than the individuals’ rate of strategy update, then static social networks approximate the systems well in investigations of the evolution of cooperation. Our results imply that correlations between connectedness and strategy may contribute to the persistence and spread of cooperation in real-world networks with stable social structures. Such effects may also be relevant in unstable networks, depending on the nature of the social link dynamics [@zimmermann2005; @santos2006coop]. Investigations of the role of correlations between strategy and network position for the evolution of cooperation in real and simulated systems with different levels of stability constitute an exiting avenue for future research.
Model details
=============
The model simulates the dynamics of a cooperative strategy in static network structures, with interactions between individuals (nodes) occurring across the network links. Each individual adopts one of multiple strategies defined by a game which models their interaction. The time evolution is discrete, with each timestep consisting of an interaction phase and a strategy update phase.
We focus on two-player, symmetric games with a binary choice of strategies, as is commonly done in models of cooperation. The strategies are labelled ‘cooperate’ and ‘defect’, and the game is determined by the following payoff matrix $$\label{eq.payoffM}
\begin{array}{cc}
\begin{blockarray}{cccl}
& & \text{\scriptsize{cooperate}} & \text{\scriptsize{defect}} \\
\begin{block}{c r[cc]}
\multirow{2}{*}{$M = $\,\,} & \text{\scriptsize{cooperate}}\,\, & R & S \\
& \text{\scriptsize{defect}}\,\, & T & P \\
\end{block}
\end{blockarray}
\end{array}$$ Here, the row and column determine the strategies of the two players respectively, and the entries define the payoffs for the row player. The game is symmetric in the sense that the payoffs for the other player are given by an identical matrix. For each node $i$ in the network, we will denote the strategy adopted by the corresponding individual by $s_i$. The payoff for individual $i$ when playing against individual $j$ is then $M_{s_i s_j}$. Within a timestep, each individual plays a single game round against each neighbour, and we define an individual’s *fitness* in a given timestep to be the sum of the payoffs received in all the games against its neighbours in that timestep. That is, for an individual defined by a node $i$, the fitness is $$\label{eq.fitness}
F_i = \sum_{j \in \mathcal{N}_i} M_{s_i s_j} ,$$ where $\mathcal{N}_i$ is the set of neighbouring nodes of $i$.
Two well-known instances of games of the above form are the Prisoner’s Dilemma game and the Snowdrift game. These are both characterized by having $T$ as the highest payoff, giving a temptation to defect. In Prisoner’s Dilemma, the worst outcome is to be defected upon while cooperating, with the order of the payoffs being $T>R>P>S$, whereas in the Snowdrift game, the worst is to be defected upon while defecting, with the payoff order being $T>R>S>P$. Note that in Prisoner’s Dilemma, the strategy with the highest individual payoff is to defect regardless of the opponent’s strategy. In unstructured populations (corresponding to networks where all nodes are connected directly to each other), evolution therefore selects for defection and cooperation does not survive. In the Snowdrift game, the best payoff depends on the opponent’s strategy, and cooperation and defection can co-exist in unstructured populations.
In our simulations, we use common one-parameter versions of the two games (e.g. [@Nowak1992; @Hauert2004; @Santos2005; @Santos2006]). For Prisoner’s Dilemma, we set $R = 1$ and $P=S=0$, and the game is parametrised by the benefit to defectors $b=T$. For $b=1$ there is no dilemma while larger values represent larger temptation to defect (making it harder for cooperation to evolve). As is often done, we take $1 \leq b \leq 2$. The Snowdrift game is parameterized by the cost-to-benefit ratio of mutual cooperation $0 < \rho \leq 1$, with $T=\frac{1}{2}(\rho^{-1}+1)$, $R=\frac{1}{2}\rho^{-1}$, $S=\frac{1}{2}(\rho^{-1} - 1)$, and $P=0$. In unstructured populations, $1-\rho$ is the equilibrium fraction of cooperators (for replicator dynamics).
Strategy update is synchronous and follows the frequently used *proportional imitation* update rule [@Santos2005; @Hauert2004]. For an individual defined by node $i$, a neighbour $j$ is chosen uniformly at random from the set of neighbours $\mathcal{N}_i$. If the neighbour has higher fitness than $i$, that is $F_j > F_i$, then $i$ adopts its strategy with probability $$\label{propimit}
\frac{F_{j}-F_{i}}{\max\{d_i,d_j\} D} ,$$ where $d_i$ denotes the degree of node $i$, and $D$ is the largest possible difference in payoffs for two players in a single game round (i.e. $D=T-S$ for Prisoner’s Dilemma and $D=T-P$ for Snowdrift). The denominator ensures normalisation of the probability. We note that the above update rule corresponds to replicator dynamics adjusted to structured, finite populations [@Santos2005; @Hauert2004; @Hofbauer1998]. Also note that the update phase can alternatively be interpreted as reproduction, in which case each timestep is a generation.
Funding
=======
Josefine Bohr Brask was funded by a Postdoctoral Internationalisation Fellowship from the Carlsberg Foundation. Jonatan Bohr Brask was supported by the Independent Research Fund Denmark.
Acknowledgements
================
We thank Tim Fawcett, Andrew Higginson, and Sylvia Dimitriadou for helpful comments on an earlier version of the manuscript.
Appendix
========
Generating networks with increased degree assortativity
-------------------------------------------------------
To generate Poisson and scale-free networks with increased degree assortativity, we apply the algorithm introduced in [@Xulvi2004]. The algorithm consists of iterated rewiring rounds. In each rewiring round, two links of the network are chosen uniformly at random and one of two rewiring schemes are carried out: *(i)* with probability $p$ the links are rewired such that one link connects the two nodes of highest degree and one connects the two nodes of lowest degree (if this is not already the case); *(ii)* with probability $1-p$ the links are rewired at random. The degree assortativity of the network can thus be controlled by varying $p$. We use $p=1$ (i.e. maximal degree assortativity given the degree distribution and the condition of all nodes belonging to the same component).
The rewiring procedure must be repeated sufficiently many times that almost all links have been rewired, i.e. such that every link has been selected for rewiring with high probability. Denoting the total number of links in the network by $L$, after $k$ iterations the probability that a given link has *not* yet been selected is $(1-2/L)^k \approx e^{-2k/L}$ for large $L$. The number of links not yet selected is thus approximately $L e^{-2k/L}$. Requiring this number to be of order unity, we see that we need $k \approx L \log(L)/2$ iterations. To make sure we reach maximum assortativity for a given network, we take $k = 10 L\log(L)$.
Nieghbourhoods and fitness
--------------------------
Consider a cooperator node of degree $d$ with $n$ cooperator neighbours and a defector node of degree $d'$ with $n'$ cooperator neighbours. The fitness of two such nodes for games with payoffs such as those used here (formalised in Eq. in Section V) will be respectively $$F = n R + (d-n) S ,$$ and $$F' = n' T + (d'-n') P .$$ The cooperator has higher fitness than the defector when $F>F'$. For the one-parameter Prisoner’s Dilemma game, the payoffs are $S=P=0$, $R=1$, and $T=b$, and thus $F>F'$ if and only if $$n > b n' .$$ That is, a cooperator must have $b$ times as many cooperator neighbours as a defector to gain a higher fitness. For the one-parameter Snowdrift game, we have $T=\frac{1}{2}(\rho^{-1}+1)$, $R=\frac{1}{2}\rho^{-1}$, $S=\frac{1}{2}(\rho^{-1} - 1)$, and $P=0$. In this case $F>F'$ when $$n \rho + d(1-\rho) > n' (1+\rho) .$$ We see that for the Snowdrift game, the cooperator also benefits from a high number of cooperator neighbours, but a sufficiently high degree can compensate for a low number of cooperator neighbours. The inequality is always fulfilled for $d=n'(1+\rho)/(1-\rho)$. However, for cost-to-benefit ratios approaching 1 this diverges, and so in this regime $n$ must again be larger than $n'$. For the most severe instances of both games, i.e. for the parameter settings making it hardest for cooperation to evolve ($b=2$ and $\rho=1$), cooperators need to have more than twice as many cooperator neighbours as defectors to achieve higher fitness.
[^1]: [email protected]
[^2]: Note that this is reminiscent of probability-proportional-to-degree sampling without replacement.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Entanglement is known to serve as an order parameter for true topological order in two-dimensional systems. We show how entanglement of disconnected partitions defines topological invariants for one-dimensional topological superconductors. These order parameters quantitatively capture the entanglement that is possible to distill from the ground state manifold, and are thus quantized to 0 or $\log 2$. Their quantization property is inferred from the underlying lattice gauge theory description of topological superconductors, and is corroborated via exact solutions and numerical simulations. Transitions between topologically trivial and non-trivial phases are accompanied by scaling behavior, a hallmark of genuine order parameters, captured by entanglement critical exponents. These order parameters are experimentally measurable utilizing state-of-the-art techniques.'
author:
- 'P. Fromholz$^*$'
- 'G. Magnifico$^*$'
- 'V. Vitale'
- 'T. Mendes-Santos'
- 'M. Dalmonte'
bibliography:
- 'biblioentang.bib'
title: 'Entanglement topological invariants for one-dimensional topological superconductors'
---
#### Introduction. –
In recent years, entanglement has emerged as a groundbreaking diagnostic to characterize and classify many-body quantum phenomena in- and out-of-equilibrium [@Amico2008; @Calabrese_2009; @Eisert2010; @fradkinbook]. An archetypal example is the possibility of unambiguously detecting topological order in two-dimensional systems via the topological entanglement entropy (TEE) [@Hamma_2005; @2006PhRvL..96k0405L; @Kitaev_2006]. The latter spots the presence of ‘long-range’ entanglement which is not distillable via local operations; consequently, it defines a genuine entanglement order parameter, that distinguishes phases depending on their quasiparticle content [@Wen19]. This insight has been widely employed in the characterization of topologically ordered states in numerical simulations [@Depenbrock:2012aa; @Jiang:2012aa; @Isakov:2011aa], and has stimulated the search for experimentally realistic entanglement probes [@Abanin:2012aa; @Daley_2012; @Elben_2018; @Vermersch_2018; @Pichler:2016aa; @Dalmonte:2017aa; @Cornfeld_2019].
While the definition of the TEE naturally emerges from gauge theories in two-dimensions, the existence of topological invariants based solely on entanglement properties in one-dimensional (1D) topological matter - e.g., in the form of an order parameter - is presently not clear. In 1D, bipartite entanglement of *connected* partitions does not display informative scaling corrections [@Calabrese_2009; @Eisert2010], and even its finer structure - captured by the entanglement spectrum -, while providing sharp *sine qua non* [@Pollmann:2010aa; @Fidkowski:aa; @Turner:2011aa], is not able to distinguish the topological character of wave functions [^1]. At the field theory level, the ultimate reason for this is that, in 1D, connected bipartite entanglement is strongly influenced by ultra-violet contributions due to edges, and is thus not immediately linked to ’universal’ information.
[Cartoon\_Fig1.pdf]{} (0,10) [a)]{}
[phase\_diagram\_finalv3.pdf]{} (0,50) [b)]{}
In this work, we show how entanglement and Rényi entropies of *disconnected partitions* provide a set of entanglement order parameters for one-dimensional topological superconductors (TSCs) [@Kitaev01; @Beenakker:2013aa]. These order parameters satisfy the following properties: *(i)* are quantized to 0 or $\log 2$ depending on the phase being topologically trivial or not-trivial, and are thus able to detect the single entanglement bit - an ebit - that can be distilled from the ground state manifold; *(ii)* display scaling behavior when approaching quantum phase transitions, and thus allow for the definition of entanglement critical exponents that describe the build-up of non-local quantum correlations across such transitions; *(iii)* are experimentally measurable in- and out-of-equilibrium utilizing recently introduced [@Elben_2018; @Vermersch_2018] and demonstrated [@Brydges:2019aa] techniques based on random measurement methods [@Enk:2012aa].
Following Ref. [@Casini:2004ab], we consider the $F$-function between two partitions $A, B$, which compensate for all edges and volume contributions in 1D. These properties are required to avoid non-universal effects: as we discuss below, the significance of our diagnostics relies on an underlying gauge theory description, which naturally calls for quantities that are divergence free in the continuum limit. In order to diagnose the presence of non-local correlations in the system, we choose two partitions with different connectivity, as shown in Fig. \[fig:cuts\] [^2]. The resulting disconnected $n$-entropies $S^D_n$ read: $$S^D_{n}= S_{A, n} + S_{B, n} - S_{A \cup B, n} - S_{A \cap B, n},
\label{eq:sqtopo}$$ where $S_{F, n}$ is the Rényi entropy of order $n$ of the partition $F$. The case $n=1$, that we denote as $S^D$, corresponds to the von Neumann entropy. $S^D>0$ because of strong subaddictivity. This entropy has been considered in Ref. [@Zeng:aa], which pointed out a strong analogy between bosonic symmetry-protected topological phases (SPTPs) and error-correcting codes. Here, our focus is instead on fermionic phases where topology is stems from an underlying fundamental symmetry (parity) which cannot be broken by any Hamiltonian perturbation. As we discuss below, this fact plays a crucial role in defining a gauge-theory picture describing the entanglement content of such states: the non-local correlations introduced by the fermionic algebra via the Jordan-Wigner string are responsible for such analogy.
At a qualitative level, the key player in $S^D_{n}$ is the disconnected partition $B$: all other terms are complementary, and only required to eliminate non-universal boundary and volume terms. In Fig. \[fig:cuts\] b, we show the finite-size behavior of $S^D$ across the phase diagram of interacting Kitaev chains: this plots illustrates graphically how, even at modest system and partition sizes, $S^D$ clearly distinguish topological from trivial phases. In the following, we define as $L_{\alpha}$ the size of a given partition, and as $L_D$ the distance between the two different parts of $B$.
#### Model Hamiltonian. –
We consider the interacting version of Kitaev p-wave superconductor, defined on an open chain of length $L$, and whose Hamiltonian reads: $$\begin{split}
H&=\sum_{j=1}^{L-1}\left[-t (a_j^\dagger a_{j+1}+\textrm{h.c.})+ (\Delta a_j a_{j+1}+\textrm{h.c.}) \right. \\
&+\left.4U\left(n_j-\frac{1}{2} \right)\left(n_{j+1}-\frac{1}{2} \right)\right]-\mu\sum_{j=1}^{L}n_j,
\end{split}
\label{eq:hamKitint}$$ where $a_j^\dagger$ ($a_j$) are the creation (annihilation) operators of the spinless fermion on site $j$, $n_j=a^\dagger_ja_j$, $t$ is the hopping amplitude, $\Delta$ is the superconducting amplitude, $U$ is a Hubbard-like interaction, and $\mu$ is the chemical potential. The phase diagram of the model is known [@Stoudenmire11; @Katsura15] and displays a TSC phase, in addition to topologically trivial phases, including a band insulator, a Mott insulator, and a incommensurate charge-density-wave (ICDW) phase. For any state in the Hilbert space, the bipartite properties of a simply connected partition are equivalent to the ones of the XYZ spin chain, which is obtained from Eq. \[eq:hamKitint\] after applying a Jordan-Wigner transformation. As such, they are uninformative about the topological origin of a given phase.
#### Disconnected entropies. –
Given the above, the goal is to find a combination of entropies able to unambiguously capture the influence of non-locality in the ground state properties, and that it is able to identify the amount of information - in this case, a single ebit - that can be stored in the ground state manifold. $S^D$ introduced above contains the simplest non-trivial disconnected partition, $S_{B}$: all other terms are introduced in order to compensate from possible volume and edge effects.
For conformal phases, $S^D_n$ is immediately given by conformal field theory [@Caraglio:2008aa; @Furukawa:2009aa; @Calabrese_2009], and vanishes in the thermodynamics limit. For gapped phases, one has to distinguish between topologically trivial and non-trivial phases. We analyze here the limiting cases.
*(i) $t=\Delta=U=0, \mu>0$*: in the band insulator case, the density matrix of arbitrary partitions has rank 1 both in fermionic and spin systems. This immediately gives $S^D_n=0$. The same result holds in the Mott insulator phase.
*(ii) $t=\Delta=1\gg|\mu|, U=0$*: this regime is representative of the TSC phase. Its correspondent in the XYZ model is a ferromagnetic phase, which we analyze first as a representative of a symmetry-broken phase. There, the lowest energy states at any finite size are equal weight superpositions of the two ferromagnetic states, $|\Psi_{XYZ}\rangle = (|\uparrow\uparrow\uparrow...\rangle\pm|\downarrow\downarrow\downarrow...\rangle/\sqrt{2}$, separated by a gap $\delta\propto e^{-L}$. For both states, any reduced density matrix of an arbitrary spatial partition is equivalent, and thus $S^D_n=0$.
For the TSC, the situation is different. While $S_A, S_{A\cap B}, S_{A\cup B}$ are the same as in the spin model, $S_{B}$ has a sharply different behavior. In this regime, the ground state is two-fold degenerate (again, up to a gap $\delta\propto e^{-L}$): Each of the two states $|\Psi\rangle_\pm$ can be written as an equal weight superposition of states with a given parity $|\psi\rangle_\pm$, i.e., $|\Psi\rangle_\pm = (1/2^{L-1})\sum_\psi|\psi\rangle_\pm$. By taking the proper fermionic trace into account, the entanglement structure of arbitrary partitions is straightforwardly evaluated [@supmat], and one obtains $S_{B}=2\log 2$. This returns a disconnected entropy $S^D = \log 2$.
While the behavior of these cartoon wave functions sharply distinguishes the TSC phase with respect to all other phases, the question whether this is a property of a phase, and whether the value of $S^D$ remains quantized requires to go beyond these oversimplified picture. Before presenting numerical results in support of these findings, we now illustrate how the quantization of $S^D$ emerges naturally when utilizing a lattice gauge theory (LGT) description of the Kitaev chain.
#### Gauge theory characterization of entanglement properties. –
The starting point is the exact relation [@MCCOY1983278] between Eq. \[eq:hamKitint\] and a $\mathbb{Z}_2$ LGT, that we schematically review. The $\mathbb{Z}_2$ LGT describes the coupling between $\mathbb{Z}_2$ gauge fields residing on bonds (represented here by Pauli matrices, $\sigma^\alpha_{j,j+1}$), and hard-core Higgs fields $\varphi_j$, with $n_j=\varphi_j^\dagger\varphi_j$ The gauge invariant Hilbert space is defined as the set of states where the local parity $P_j = (1-2n_j ) \sigma^z_{j-1,j}\sigma_{j,j+1}^z$ is fixed to 1 (see Fig. \[fig:Z2LGT\]). Under open boundary conditions (OBC), we can fix the value of the first gauge field without loss of generality to $\sigma^z_{0,1}=1$. The value of the last gauge field $\sigma^z_{L,L+1}=P$ equals the total parity of the system due to gauge invariance.
 Schematics of the correspondence between the Kitaev chain and $\mathbb{Z}_2$ lattice gauge theories. *a)*: Hilbert space structure and gauge invariant building blocks. *b)*: three examples of the mapping between states in the fermionic (left), gauge theory (center), and string representation (right). *c)*: string representation of $|\Psi\rangle_+$. ](CartoonZ2.pdf){width=".48\textwidth"}
In the gauge theory language, the ground state wave functions $|\Psi\rangle_\pm$ can be described in terms of either fermionic or gauge fields, since, in 1D, those are mutually fixed by Gauss law. In the latter language, the ground states are equal weight-superpositions of all possible string states of arbitrary length, and compatible with the boundary conditions: a sample of those are depicted in Fig. \[fig:Z2LGT\] for the case $P=1$. This picture describes a 1D gauge theory in a phase with strongly fluctuating gauge fields, and is strongly reminiscent of the loop description of 2D $\mathbb{Z}_2$ LGT [@Lacroix2010; @Zeng19; @Wen19].
Evaluating entanglement entropies in this phase is straightforward by exploiting gauge invariance:
*i)* The entropy of each connected partition is $\log 2$. Let us define $\sigma_L^z, \sigma^z_R$ as the two boundary spins of the partition. Their product is equal to the parity of the partition: the density matrix of the partition is block-diagonal in this conserved quantum number. If the correlation length is much smaller than the partition length, both positive and negative parities are equally probable and all states count with equal weight. The corresponding von Neumann entropy is thus $\log 2$.
*ii)* the entropy of disconnected partitions is $(N_c-1)\log 2$, where $N_c$ is the number of partitions. Let us define as $\sigma_{L,h}^z, \sigma^z_{R,h}$ the gauge fields at the boundaries of the partition $h$. As long as the length of each partition is larger than the correlation length, each patch is a equal weight superposition of all possible states, under the condition that $P_h=\pm1$ for partitions with or without an out-coming flux. Fixed the total parity, there are $2^{N_c-1}$ finite, equal values of the corresponding density matrix, which indeed returns an entropy equal to $(N_c-1)\log 2$.
We emphasize that the gauge theory description enables a simple calculation of the entropies (by replacing fermionic statistics with a $\mathbb{Z}_2$ gauge field), and, at the same time provides a simple, compelling physical picture, that might be extended to more exotic types of order.
#### Numerical results. –
We now turn to a numerical investigation of Eq. \[eq:hamKitint\]. We used free fermion techniques [@Peshel03] to investigate the non-interacting case $U=0$, and density-matrix-renormalization group (DMRG) [@White1992; @Schollwock2005] for $U\neq0$. Since DMRG does not give immediate access to $S_{B}$, we performed separate simulations to obtain this quantity, by modifying the lattice connectivity [@supmat]. We kept up to 1200 states after truncation, and performed at least 30 sweeps. Typical discarded weights at the end of the simulation were of order $10^{-8}$.
The phase diagram in Fig. \[fig:cuts\] shows how, even at very modest partition sizes, $S^D$ is large and finite only the TSC phase; analogous results hold for the $U=0$ plane [@supmat]. In Fig. \[fig:FSS\], we show the finite-size-scaling behavior of $S^D$ for representative points in the TSC ($\mu/t=1.0, 1.5$) and topologically trivial ($\mu/t=4$) phase. The asymptotic values are quantized within numerical accuracy of our fits to $\log 2$ and 0, respectively, in agreement with the theoretical discussion above. The inset shows how, in the TSC phase, quantization is approached exponentially fast in system size.
Both results hold not only for $S^D$, but also for its low-order Rényi counterparts. As a representative of these results, we show in Fig. \[fig:RenyiPD\] the behavior of $S^D_2$ in the $\Delta-\mu$ plane: again, the value of this entropy clearly distinguish topologically trivial and non-trivial phases.
![ \[fig:FSS\] (Color online)Finite size scaling properties of $S^D$ for a chain with $L_A=L_B=(L_{A \cap B}+L_{A \cup B})/2$, and $U=0, \Delta/t=1$. In the topologically trivial phase, $S^D$ quickly vanishes. Oppositely, in the topological phase ($\mu=1.0, 1.5$), $S^D$ increases as a function of system size, and approaches its thermodynamic value exponentially fast when increasing $L_A$, as shown in the inset.](scaling_final_2.pdf){width=".36\textwidth"}
![ \[fig:RenyiPD\] (Color online) Top panel: $S^D_2$ in the non-interacting phase diagram of the Kitaev chain with $U=0$. Here, $L=50, L_A=L_B =12, L_D = 32$.](renyi_phase_diagram.pdf){width=".36\textwidth"}
[scaling\_1\_1.pdf]{} (10,35) [a)]{} (55,35) [b)]{}
[scaling\_2\_2.pdf]{} (10,35) [c)]{} (55,35) [d)]{}
#### Universal behavior and entanglement critical exponents. –
Since $S^D$ captures universal properties of each phase, it is natural to wonder whether such quantities can display universal scaling behavior when crossing a quantum phase transition. Here, we focus on the transition between TSC and band insulator, which belongs to the Ising universality class.
Fixing $L_A=L_B=L/2$, $L_{A\cap B}=3L/4$ in order to avoid effects due to aspect ratios of the different partitions, we fit $S^D$ as one would for an order parameter around the phase transition using a phenomenological finite-size scaling ansatz: $$S^{D}L^{\frac{a}{b}}=\lambda\left(L^{\frac{1}{b}}(\alpha-\alpha_{c})\right),\label{eq:scaling_formula}$$ where $\alpha= \mu$ or $U$ is the parameter chosen, $a$ and $b$ (*a priori* different depending on the chosen parameter) take place of the usual critical exponent $\beta$ and $\nu$, and $\lambda(x)$ is a scaling function. We extract these parameters using curve intersections and collapse shown in Figs. \[fig:scaling\] for two cuts indicated in Fig. \[fig:cuts\] with yellow arrows. The results of the collapse scaling locates correctly the transition point (with errors $10^{-4}$). Most surprisingly, we find that the entanglement critical exponents satisfying $a = b =1$ irrespectively of where the transition line is crossed, a sharp signature of universal behavior. The quality of the collapse scaling in both non-interacting (b) and interacting (d) case is good already for modest system sizes, further corroborating such universal behavior.
#### Topological invariance under coherent dynamics. –
A key signature of topological invariants is the fact that, in the thermodynamic limit, those cannot change under unitary evolution (as long as specific symmetries are not broken explicitly [@Caio:2015aa; @DAlessio:2015aa; @McGinley:aa]). In order to check that $S^{D}$ is a true topological invariant, we performed an extensive investigation based on quantum quenches starting within the topological phase, and quenching with arbitrary values of the Hamiltonian parameters.
A representative sample of our results is presented in Fig. \[fig:quench\]. In panel *a)*, we plot the time evolution of $S^{D}$ for a quench from an initial value of the pairing amplitude $\Delta=0.5$ to a final value $\Delta=1.5$. Different lines correspond to different system sizes. For each size, one can sharply distinguish two regimes. At short times, $S^{D}$ does not change with time, and exhibits a plateau up to a time $t_p$ that depends on $L_A$. After this timescale, quantization is lost, and the dynamics is dictated by non-universal dynamics. In order to understand whether quantization is a robust feature, we perform a finite size scaling analysis in panel *b)*: our results show that $t_p$ (defined as the time where $S^D=0.95$) grows approximately linearly with system size, and that its extrapolated value to the thermodynamic limit diverges. This behavior confirms the topological invariant nature of $S^D$.
[main\_quench\_plot]{} (1,30) [a)]{} (50,30) [b)]{}
#### Experimental measurement and comparison to other diagnostics. –
Two key properties of $S^D_n$ are that they are informative already for modest partition sizes, and that Rényi entropies can be used. The proposals in Ref. [@Elben_2018] discuss how to perform measurements of Rényi-2 entropies in synthetic quantum systems: in particular, within that setting, the complexity of the measurement is not sensitive to the connectivity of the partition itself, but only to its total size. Given that a large $L_D$ allows to distill the correct information from the wave function, measuring $S^D$ is as complex as measuring its largest partition $A$. We note that partitions of sizes up to 10 spins have already been probed in experiments [@Brydges:2019aa].
Finally, we comment on the relation between our topological invariant and other diagnostics. Topological invariants such as the many-body Chern number are unrelated to bipartite entanglement properties, as they do not depend solely on the spectrum of density matrices, but also on their eigenfunctions. For specific symmetries, *ad hoc* topological invariants can be defined [@Zeng19] (and potentially experimentally measured [@Elben:aa]) utilizing the matrix-product-state classification of SPTPs [@Chen:2012aa]; these quantities are sensitive to the response of a state to specific (symmetric) operations, and not immediately connected to entanglement. From the theoretical viewpoint, all these diagnostics represent complementary tools, that give access to qualitatively different features characterizing topological matter - response of wave functions under changing boundary conditions, properties with respect to protecting symmetry, and non-local entanglement content of wave functions, respectively.
#### Conclusions. –
We have shown how entanglement of disconnected partitions uniquely distinguishes topological superconducting phases in one-dimensional systems. This distinction is naturally interpreted within a lattice gauge theory framework, and leads to key footprints both at the ground state level, and in quantum quenches. The entanglement order parameters display universal scaling behavior when crossing phase transitions, characterized by entanglement critical exponents. Our findings show that modest partition sizes - of the order of what has been already experimentally demonstrated - are sufficient to uniquely characterize topological superconductors via entanglement. It would be intriguing to investigate whether other forms of quantum correlations between disconnected partitions, such as discord [@Ollivier:2001aa] or quantum coherences [@Frerot:2016aa], display similar characteristic features, and if entanglement topological invariants can be used to characterize the real time dynamics of interesting topological matter [@McGinley:aa].
We thank S. Barbarino, F. Illuminati, N. Lindner, L. Pastori, F. Surace, X. Turkeshi, and B. Vermersch for useful discussion, and J. Budich for suggesting the investigation of the real-time dynamics. This work is partly supported by the ERC under grant number 758329 (AGEnTh), and has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 817482. G.M. is partially supported through the project “QUANTUM” by the Istituto Nazionale di Fisica Nucleare (INFN) and through the project “ALMAIDEA” by the University of Bologna.
[^1]: One simple example is the equivalence between the entanglement spectra of the ground state of finite Ising and Kitaev chains.
[^2]: Note that, at the field theory level, the entanglement properties of these partitions are known to depend on the full operator content [@Caraglio:2008aa; @Calabrese_2009].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This paper quantifies the effect of speed cameras on road traffic collisions using an approximate Bayesian doubly-robust (DR) causal inference estimation method. Previous empirical work on this topic, which shows a diverse range of estimated effects, is based largely on outcome regression (OR) models using the Empirical Bayes approach or on simple before and after comparisons. Issues of causality and confounding have received little formal attention. A causal DR approach combines propensity score (PS) and OR models to give an average treatment effect (ATE) estimator that is consistent and asymptotically normal under correct specification of either of the two component models. We develop this approach within a novel approximate Bayesian framework to derive posterior predictive distributions for the ATE of speed cameras on road traffic collisions. Our results for England indicate significant reductions in the number of collisions at speed cameras sites (mean ATE = -30%). Our proposed method offers a promising approach for evaluation of transport safety interventions.'
author:
- 'Daniel J. Graham'
- Cian Naik
- 'Emma J. McCoy'
- Haojie Li
bibliography:
- 'SC-ABDR.bib'
title: Quantifying the causal effect of speed cameras on road traffic collisions via an approximate Bayesian doubly robust estimator
---
[*Keywords:* Doubly robust; Bayesian inference; propensity score; average treatment effect; speed cameras; casualties]{}.
Introduction
============
Fixed speed limit enforcement cameras are a common intervention used to encourage drivers to comply with maximum legal speed limits. The cameras are installed at sites on selected links in order to detect speed limit violations, which can subsequently be punished with monetary fines, driver licence disqualification points, or prosecution. Since the introduction of speed cameras (SCs) there has been considerable debate about their effects on road traffic collisions (RTCs). At various times claims have been made that SCs serve to reduce RTCs, that they have no effect, or even that they increase RTCs by encouraging more erratic driving behaviour.
A number of academic studies of the effect of speed cameras on RTCs have been undertaken [for a review see @Li/et/al:2013]. Most studies find that speed cameras have led to a reduction in RTCs, but the range of estimated effects is large (from 0% to -55%). Variation in estimates is to be expected given that study results pertain to diverse empirical contexts, but it is also the case that a number of different methods have been applied which can have a critical influence on results obtained. In particular, since SCs are not randomly assigned, it is essential that any adopted method recognises that the observed relationship between SCs on RTCs may be subject to confounding. Confounding arises when the characteristics that influence treatment assignment (i.e. whether a site is ‘treated’ and ‘untreated’ with an SC) also matter for outcomes (i.e. RTCs). Regression to the mean (RTM), for instance, is a well known manifestation of confounding that arises via ‘selection bias’.
The extent to which confounding has been recognised and addressed in existing studies varies considerably. Some studies have simply ignored it, using simple before-and-after methods with control groups [e.g. @Christie/et/al:2003; @Cunningham/et/al:2008; @DePauw/et/al:2014; @Gains/et/al:2004; @Gains/et/al:2005; @Goldenbeld/vanSchagen:2005; @Jones/et/al:2008; @Maher:2015]. Others have used the empirical Bayes (EB) method as suggested by @Hauer/et/al:2002, largely to adjust for effects of confounding that arise via RTM [e.g. @Chen/et/al:2002; @Elvik:1997; @Hoye:2015; @Mountain/et/al:2004; @Mountain/et/al:2005; @Shin/et/al:2009]. Finally, there are a small number of studies that have used time-series methods, either interrupted time-series analyses with control groups or ARIMA, to test for changes in outcome rates [@Carnis/Blais:2013; @Hess/Polak:2003; @Keall/et/al:2001 e.g.]. Where studies have attempted to address confounding this has been done via the inclusion of covariates in outcome regression (OR) models, typically using Poisson or negative binomial Generalised Linear Models (GLMs).
In a previous paper we adopted a propensity score (PS) matching approach to evaluate the effectiveness of speed cameras [see @Li/et/al:2013]. A key advantage of the PS over OR approach is that it provides an effective way of isolating a valid control group by ensuring that the distribution of pre-treatment covariates matches those of the treated group and that genuine overlap in the support of the covariates exists between the two groups. However, as with the OR approach, valid inference from PS models crucially depends on the unknown PS model being correctly specified.
In this paper we build on our previous work by developing and applying an estimation approach which we believe has much to offer in evaluating the effectiveness of road safety interventions. Our approach uses the principle of doubly-robust (DR) estimation, which provides robustness to model misspecification by combining both OR and PS models to derive an average treatment effect (ATE) estimator which is consistent and asymptotically normal under correct specification of just one of the two component models. The DR approach is attractive for our application because the PS and OR models we can construct make different assumptions about the nature of confounding. For the PS model, we are able to faithfully represent via measured covariates the formal criteria that exist for the assignment of speeds cameras to sites. For the OR model, we can difference our response variable before and after treatment to allow for the existence of site level time-invariant unobserved effects in addition to measured confounders.
To avoid common sources of misspecification error, we estimate our component models using semiparametric Generalized Additive Mixed Models (GAMMs) which make minimal a priori assumptions on the functional form of the relationships under study. We also use a matching algorithm prior to forming the DR model to establish a valid control group. Thus, in our approach, potential biases from confounding are addressed by combining three compatible modelling tools: via matching to achieve comparability between treated and control sites, via a regression model for RTCs, and via a model for the treatment assignment mechanism.
DR estimators have been studied and applied extensively in the frequentist setting [e.g. @Robins:2000; @Robins/et/al:2000c; @Robins/Rotnitzky:2001; @VanDerLaan/Robins:2003; @Lunceford/Davidian:2004; @Bang/Robins:2005; @Kang/Schafer:2007]. A further contribution of the paper is that we develop our binary DR estimator within the Bayesian paradigm. A Bayesian representation of the DR model has proven difficult to formulate in previous work because DR estimators are typically constructed as solutions to estimating equations based on a set of moment restrictions that do not imply fully specified likelihood functions. We choose the Bayesian paradigm for three main reasons. First, DR estimation of the ATE involves prediction and extrapolation over covariate distributions with underlying uncertainty in parameter estimates. Bayesian inference provides a suitable framework for prediction that explicitly addresses such uncertainty in the sense that both the predicted observations, and the relevant parameters for prediction, have the same random status. Second, by deriving a posterior predictive distribution for the ATE, rather than a fixed value, we can make probability statements about the causal quantity of interest allowing us to discuss findings in relation to specific hypotheses or in terms of credible intervals which can offer a more intuitive understanding of the effects of SCs for public policy formulation. Finally, we develop an approximate Bayesian approach that can utilise prior information about the parameters of interest, which could be useful in evaluating safety interventions when historical data or training data from other regions are available.
The paper is structured as follows. Section two outlines broad trends in road traffic casualties for Britain and then sets out a formal causal modelling framework to estimate the effects of SCs on RTCs. Section three describes our approximate Bayesian DR approach and presents some simulations that demonstrate its properties. Section four describes the data available for estimation and outlines our chosen model specifications. Results are then presented in section five and conclusions are drawn in the final section.
A causal inference framework to quantify the effects of speed cameras
=====================================================================
Road traffic casualties in Britain
----------------------------------
For the year ending September 2016 the UK DfT recorded a total of 182,560 causalities on British roads of which 25,160 were classified as killed or seriously injured (KSIs) [@DfT:2017]. Since 2010 the annual numbers of fatalities and KSIs have not changed significantly, following several years in which road safety was improving. The average number of fatal road traffic incidents over the period 2010 to 2016 is approximately 1,800. Since the volume of road traffic has continued to grow over this period, however, the number of fatalities per vehicle mile driven has been falling [@DfT:2016].
The DfT argue that there is good evidence to suggest that while the absolute number of fatalities on British roads now appears to be relatively static, overall absolute casualty numbers are continuing to fall. In short, levels of safety appear to be improving in relative terms and not deteriorating in absolute terms. Given the changes that have occurred in vehicle technology, medical care, and road safety interventions, however, the DfT also note that a comprehensive causal understanding of the factors underpinning casualty trends is currently out of reach. In this paper we attempt to contribute to such an understanding by quantifying the causal impact of one type of safety intervention: speed cameras (SCs).
ATE estimation within the potential outcomes framework
------------------------------------------------------
Our sample comprises $n$, $i=1,...,n$, links on the road network. Some links have a SC other do not. We define $D_i\in\{1,0\}$ as a binary random variable indicating the presence or otherwise of a SC and we refer to this as the treatment variable. We are interested in the effect of the treatment on an outcome $Y_i$, which measures collision frequency. We define $Y_i(1)$ and $Y_i(0)$ as the [*potential outcomes*]{} for unit $i$ under treated and control status respectively. Recognising that SCs are not assigned randomly, we also define $X_i$ as a random vector of pre-treatment covariates that capture characteristics of links that are relevant to whether a SC was assigned or not, and are also relevant for outcomes. Thus, the data we observe for each link takes the form of a random vector, $z_i = (y_i,d_i,x_i)$, where $y_i$ denotes a response, $d_i$ the treatment received, and $x_i$ a vector of pre-treatment covariates.
Ideally, we would assess the effects of SCs on each link by calculating the individual causal effect (ICE): $\tau_i=Y_i(1)-Y_i(0)$, but the observed data reveal only actual outcomes not potential outcomes. Thus we observe the random variable $Y_i=Y_i(1) I_{1}(D_i)+Y_i(0)(1-I_{1}(D_i))$, where $I_{1}(D_i)$ is the indicator function for receiving the treatment, but we do not observe the joint density, $f(Y_i(0),Y_i(1))$, since a SC cannot be both present and absent on a link simultaneously. Instead, our target of inference is the ATE, defined as $$\tau=\Expect[Y_i(1)]-\Expect[Y_i(0)],$$ which measures the difference in expected outcomes under treatment and control status.
A key insight of the potential outcomes approach is that if we focus on estimating the ATE then we do not have to observe all potential outcomes, even under a non-random treatment assignment, as long as three key assumptions hold. First, the potential outcomes for unit $i$ must be conditionally independent of the treatment assignment given a (sufficient) set of observed covariates $X_i$: $Y_i(0),Y_i(1)) {\protect\mathpalette{\protect\independenT}{\perp}}I_{1}(D_i)|X_i$. Second, the support of the conditional distribution of $X_i$ given a particular treatment status must overlap with that of $X_i$ given any other treatment status: $0<\P(I_{1}(D_i)=1|X_i=x)<1, \ \forall \ x$. Third, the relationship between observed and potential outcomes must comply with the Stable Unit Treatment Value Assumption (SUTVA) [e.g. @Rubin:1978; @Rubin:1980; @Rubin:1986; @Rubin:1990], which requires that the observed response under a given treatment allocation is equivalent to the potential response under that treatment allocation: $Y_i=I_{1}(D_i)Y_i(1)+ (1-I_{1}(D_i))Y_i(0)$ for all $i=1,...,n$.
The three assumptions defined above, which are together referred to by @Rosenbaum/Rubin:1983b as [*strong ignorability*]{}, allow for identification of causal effects from observational data because if they hold the ATE can be derived as,
\[iden\] $$\begin{aligned}
\tau =& \Expect_i(Y_i(1)-Y_i(0))= \Expect_X\left[\Expect_i(Y_i(1)|X_i=x)-\Expect_i(Y_i(0)|X_i=x)\right]\\
=&\Expect_X\left[\Expect_i(Y_i(1)|X_i=x, I_{1}(D_i)=1)-\Expect_i(Y_i(0)|X_i=x, I_{1}(D_i)=0)\right]\\
=& \Expect_X\left[\Expect_i(Y_i|X_i=x, I_{1}(D_i)=1)-\Expect_i(Y_i|X_i=x,I_{1}(D_i)=0)\right].\end{aligned}$$
Conditional independence justifies the equality of (\[iden\]a) and (\[iden\]b), the SUTVA allows the substitution of observed for potential outcomes to give (\[iden\]c), and overlap ensures that the population ATE in (\[iden\]c) is estimable since there are units in both the treated and untreated groups.
Thus, if strong ignorability holds, the potential outcomes approach offers a route to obtaining valid causal estimates of the ATE of SCs. To proceed we need to estimate the relevant expectations in (\[iden\]c) above.
Causal estimators
-----------------
Using the notation of @Tsiatis/Davidian:2007, we define joint densities of the observed data of the form $$f_Z(z)=f_{Y|D,X}(y|d,x)f_{D|X}(d|x)f_X(x).$$ Given strong ignorability, estimation of the ATE of SCs can proceed in one of the following ways;
1. Outcome regression (OR) model - leave $f_{D|X}(d|x)$ and $f_{X}(x)$ unspecified and posit a model for $\Expect[Y_i|D_i,X_i])$; the mean of the conditional density of the response given the covariates, using an OR model $\Psi^{-1}\{m(D_i,X_i;\beta)\}$, for known link function $\Psi$, regression function $m()$, and unknown parameter vector $\beta$. If the OR is correctly specified for the mean response then the ATE can be consistently estimated by. $$\hat{\tau}_{OR}=\frac{1}{n}\sum_{i=1}^n\left[\Psi^{-1}\{m(1,X_i; \hat{\beta})\}-\Psi^{-1}\{m(0,X_i; \hat{\beta})\}\right].$$
2. Propensity score (PS) model - leave $f_{Y|D,X}(y|d,x)$ and $f_X(x)$ unspecified but assume a model for $f_{D|X}(d|x)$; the conditional density of treatment assignment given covariates. This is a propensity score (PS) model, denoted $\pi(D_i|X_i; \alpha)$, which can be used to form a number of different nonparametric estimators but of primary interest here is its use in the weighting estimator attributed to @Horovitz/Thompson:1952 $$\hat{\tau}_{IPW}=\frac{1}{n}\sum_{i=1}^n \left[\frac{I_{1}(D_i) \cdot Y_i}{\pi(D_i|X_i; \hat{\alpha})} - \frac{[1-I_{1}(D_i)] \cdot Y_i}{1-\pi(D_i|X_i; \hat{\alpha})}\right],$$ which is consistent under correct specification of the PS by virtue of the fact that $\Expect[Y_i(1)]=\Expect\{\left[Y_i(1) \cdot I_{1}(D_i)\right]/\pi(D_i|X_i; \alpha)\}$ and similar for control treatment status.
3. Doubly-robust (DR) model - leave $f(x)$ unspecified but assume both an OR model and a PS model and combine them to form a DR estimator. This is achieved by weighting or augmenting the OR model with a function of the inverse of the estimated PS to give a DR model. In this paper we estimate the weighted model $$e(D_i,X_i;\xi)=\Psi^{-1}\{m(D_i,X_i; \xi)\}$$ where the unknown parameter vector $\xi$ is obtained by weighting the model with $$\widehat{\kappa}_i(D_i,X_i)=\frac{I_{1}(D_i)}{\widehat{\pi}(D_i| X_i; \widehat{\alpha})}+\frac{1-I_{1}(D_i)}{1-\widehat{\pi}(D_i| X_i; \widehat{\alpha})}.$$ This model will consistently estimate $\Expect[Y_i|D_i,X_i])$ if the model $\Psi^{-1}\{m(I_{1}(D_i),X_i;\beta)\}$ is correct because while weighting may induce inefficiency it will leave the consistency and asymptotic normality of the OR estimates unchanged. If the OR model is incorrectly specified, but the PS is correctly specified, the model is still consistent because weighting gives rise to estimating equations of the form $$\sum_{i=1}^{n}\widehat{\kappa}_i(D_i,X_i)\frac{1}{\phi} \frac{\partial e\left(d_i,x_i;\xi\right)}{\partial \xi^{\sf{T}}} \left[y_i-e\left(d_i,x_i;\xi\right)\right]=0,$$ where $\phi_i \equiv \phi (D_i,X_i)$ is a working conditional variance for $Y_i$ given $(D_i,X_i)$, which effectively correct for the bias in approximating $\Expect[Y_i|D_i,X_i]]$ using $\Psi^{-1}\{m(D_i,X_i;\beta)\}$ .
We use estimates of $\xi$ to form the DR estimator $$\hat{\tau}_{DR}=\frac{1}{n}\sum_{i=1}^n\left[\Psi^{-1}\{m(1,X_i; \hat{\xi})\}-\Psi^{-1}\{m(0,X_i; \hat{\xi})\}\right].$$
Approximate Bayesian doubly-robust estimation
=============================================
So far we have discussed DR estimation within the context of frequentist semiparametric inference. As mentioned in the introduction to the paper there are good reasons why a Bayesian inferential approach is particularly beneficial for estimation of road safety interventions. Bayesian inference has, however, proven difficult to apply for DR estimators because they are based on a set of moment restrictions which do not provide fully specified likelihood functions. Here, we make some improvements to the approach proposed by @Graham/et/al:2016 in the context of continuous treatment. In contrast to that paper we focus on binary treatments using PS weighting rather than augmentation to achieve the DR model and we implement ways of incorporating prior information into the posterior distribution of the ATE. The basic theory underpinning approximate Bayesian inference in this context is covered comprehensively in @Graham/et/al:2016 and so we provide only a brief summary here.
The Bayesian bootstrap was first introduced by @Rubin:1981 and applied in weighted likelihood models by @Newton/Raftery:1994. The basic idea is to create new datasets by repeatedly re-weighting the original data in order to obtain the posterior distribution for some parameter of interest. If we treat our observed data, $z_i$ say, as effectively coming from a multinomial distribution with distinct values $a_k$, $k=(1,...,K)$, and attach a probability to each distinct value $\theta=(\theta_1,...,\theta_k)$, then by placing an improper Dirichlet prior on $\theta$ $$\pi(\theta)\propto \prod^{K}_{k=1}\theta_{k}^{-1}.$$ the posterior density also has a Dirichlet distribution $$p(\theta|v)\propto \prod^{K}_{k=1}\theta_{k}^{n_{k}-1}.$$ with parameter $n_{k}$. This posterior can be stimulated via the weighted likelihood $$\widetilde{L}(\theta)=\prod^{n}_{i=1}f(z_i;\theta)^{w_{i}^{ }},$$ in which the weights $w^{ }= (w_{1}^{ },...,w_{n}^{ } )$ are distributed according to the uniform Dirichlet distribution and simulated as $n$ independent standard exponential (i.e. gamma(1,1)) variates and standardised. The weighted likelihood reduces to $$\widetilde{L}(\theta)= \prod_{i=1}^n \left\{ \prod_{k=1}^K \theta_k^{I_{k}(z_i)} \right\}^{w_i} = \prod_{k=1}^{K}\theta_{k}^{\sum\limits_{i=1}^n w_i I_{k}(z_i)} = \prod^{K}_{k=1}\theta_{k}^{n \gamma_{k}},$$ say, where $n \gamma_{k}$ is the sum of the weights $w_i$ for which $z_i = a_k$. Since the vector $\gamma=(\gamma_1,...,\gamma_K)$ has a Dirichlet distribution with parameters $n_k=(n_1,...,n_K)$, $$p(\gamma)\propto \prod_{k=1}^K\gamma_k^{n_k-1}$$ and since at the point of maximisation of $\widetilde L (\theta)$ is $\widetilde \theta = \gamma$, then the solutions to the maximised weighted likelihood function with repeatedly sampled uniform Dirichlet weights $w^{(l)}$ represent a sample from the posterior of $\theta$ under the improper prior $\prod_k \theta^{-1}_{k}$.
To apply the Bayesian bootstrap to our DR model we estimate $$e(D_i,X_i;\xi)=\Psi^{-1}\{m_A(D_i,X_i; \xi)\}$$ with weights $$w_i^{(l)} \cdot \widehat{\kappa}_i(D_i,X_i).$$ The maximiser of $\widetilde{L}(\xi)$, which we denote $\widetilde{\xi}$, implies a solution to $$\label{BBAPO}
\sum_{i=1}^{n}w^{(l)}_{i}\cdot \widehat{\kappa}_i(D_i,X_i) \cdot \frac{1}{\phi} \frac{\partial e\left(d_i,x_i;\xi\right)}{\partial \xi^{\sf{T}}} \left[y_i-e\left(d_i,x_i);\xi\right)\right]=0,$$ which as noted above has the DR property. We repeatedly draw sets of random weights $\{w^{(l)}_i\}^{n}_{i=1}$ as $n$ standardised independent standard exponential variates and solve (\[BBAPO\]) to build up an empirical posterior density of $\widetilde \xi$, denoted $p_n(\widetilde \xi)$, from which the sampled values $\widetilde \xi^{(l)}$ are consistent with the DR estimating equations.
@Newton/Raftery:1994 apply sampling-importance resampling (SIR) to improve accuracy of the weighted bootstrap approach, but this improvement requires a fully specified likelihood function. Instead, for our restricted moment model, we use the resampling scheme proposed by @Muliere/Secchi:1996 which extends Rubin’s bootstrap in a general Bayesian nonparametric context. Two attractive features of Muliere and Secchi’s approach for causal modelling are that it ensures that predictive distributions are not constrained to be concentrated on observed values and it allows us to take prior opinions into account. The posterior predictive distribution of the ATE, incorporating prior information, is obtained in the following way.
1. Estimate the PS model $\pi(D_i|X_i; \alpha)$, and form $$\widehat{\kappa}_i\left(d_i,x_i;\widehat{\alpha}\right)=\frac{I_1(d_i)}{\widehat{\pi}\left(d_i| x_i; \widehat{\alpha}\right) }+\frac{1-I_1(d_i)}{1-\widehat{\pi}\left(d_i| x_i; \widehat{\alpha}\right) }.$$
2. Draw a single set of random weights $\{w^{(l)}_i\}^{n}_{i=1}$ and form the combined weights $w_i^{(l)} \cdot \widehat{\kappa}_i\left(d_i| x_i; \widehat{\alpha}\right)$ and estimate the weighted model $$\Psi^{-1}\left\{m_A\left(d_i,x_i; \xi^{(l)} \right)\right\}.$$
3. Repeatedly compute (ii) using new weights $\{w^{(l)}_i\}^{n}_{i=1}$ to obtain the empirical posterior distribution $p_n(\widetilde \xi)$.
4. Introduce a prior distribution $p_0$ for $\xi$ and a positive number $k$, the ‘measure of faith’ that we have in this prior. This can range anywhere from 1 to a size comparable to the number of samples of $\xi$.
5. Generate $m$ observations $x_1^*,...,x_m^*$ from $\frac{kp_0+Lp_n}{k+L}$, where $p_n$ is as above. We choose $m=L$ in our case.
6. For $i=1,...,m$ generate $v_i$ from a $\Gamma\bigg(\frac{L+k}{m},1\bigg)$ distribution.
7. Sample new parameters $\widetilde \xi_{MS}$ from $x_1^*,...,x_m^*$ using the weights $v_1,...,v_m$ to form the posterior $p_m(\widetilde \xi)$.
8. Resample $V$ values of the covariate vector uniformly over the observed values and a single vector $\xi^{(m)}$ from $p_m(\widetilde \xi)$.
9. Form a sampled value of the ATE random variable as $$\tau^{(m)}_{BDR}= \frac{1}{V}\sum_{v=1}^{V} \left[\Psi^{-1} \{m_A(1, x_v; \widetilde{\xi}^{(m)})\}-\Psi^{-1} \{m_A(0, x_v; \widetilde{\xi}^{(m)})\}\right] .$$
10. Repeat this procedure $M$ times, $m=(1,...,M)$, to obtain the posterior predictive distribution.
Simulations {#C6S4}
-----------
In this subsection we present some simulation to demonstrate the DR properties of our approximate Bayesian approach. The simulations are based on the following data generating process: a binary treatment $D$ is assigned as a function of covariate $X$, and the outcome of interest $Y$ depends on both treatment $D$ and covariate $X$ $$\begin{aligned}
X \sim \text{Normal}(0,10)\\
D \sim \text{Bernoulli}(\text{expit}(\alpha_0+\alpha_1 X))\\
Y \sim \text{Normal}(\beta_0 + \beta_1 D + \beta_2 X, 5)\end{aligned}$$ where $\alpha_0=2$, $\alpha_1=0.2$, $\beta_0=10$, $\beta_1=5$, $\beta_2=0.2$. The true ATE is given by parameter $\beta_1$, that is $\tau = 5.0$.
The following models are tested:
- $\widehat{\tau}_{BOR1}$ - an approximate Bayesian OR model based on the correctly specified model: $\Expect[Y|D,X]=\beta_0+\beta_{1}D+\beta_{2}X$. The point estimate reported in the simulations is the mean value of the ATE posterior predictive distribution, i.e. $$\widehat{\tau}_{BOR}= \frac{1}{L}\sum_{l=1}^{L}\left[\frac{1}{V}\sum_{v=1}^{V} \left[\Psi^{-1} \left\{m\left(1, x_v; \widetilde{\beta}^{(l)}\right)\right\}-\Psi^{-1} \left\{m\left(0, x_v; \widetilde{\beta}^{(l)}\right)\right\}\right] \right].$$
- $\widehat{\tau}_{BOR2}$ - same as \[1.\] except based based on an incorrectly specified OR model with covariate $X$ excluded.
- $\widehat{\tau}_{BDR1}$ - an approximate Bayesian DR model based on an incorrectly specified OR model ($X$ excluded) but with weights based on the correct PS model $$\widehat{\tau}_{BDR}= \frac{1}{L}\sum_{l=1}^{L}\left[\frac{1}{V}\sum_{v=1}^{V} \left[\Psi^{-1} \left\{m_A\left(1, x_v; \widetilde{\xi}^{(l)}\right)\right\}-\Psi^{-1} \left\{m_A\left(0, x_v; \widetilde{\xi}^{(l)}\right)\right\}\right] \right].$$
- $\widehat{\tau}_{BDR2}$ - an approximate Bayesian DR model based on a correctly specified OR model but with weights based on an incorrect PS model, which is generated randomly from the continuous uniform distribution: $\widehat{\pi}(D|X) \sim \text{Uniform}(0,1)$.
- $\widehat{\tau}_{BDR3}$ - an approximate Bayesian DR model based on an incorrectly specified OR model weighted with weights based on an incorrect PS model.
The simulations are based on 1000 runs on generated datasets of size 1,000. In each case, we place a Normal prior on the treatment coefficient $\beta_1$, with mean equal to the true value (5 in this case). We set the measure of faith $k$ to be relatively low so as not to overly affect the results. Table \[sims\] shows our simulation results. Mean values and variances of the point estimates obtained (i.e. means and variances of the ATE distributions) and the mean squared error (MSE) are reported.
Av. Est. Emp. Var. MSE
------ ---------- ----------- -------
BOR1 5.004 0.036 0.036
BOR2 5.350 0.036 0.157
BDR1 5.008 0.046 0.046
BDR2 5.018 0.862 0.862
BDR3 5.360 0.946 1.074
The mean of the posterior distribution for the ATE from the correctly specified OR model, $\widehat{\tau}_{BOR1}$, provides a good approximation to the true value of $\tau$. The incorrectly specified OR model, BOR2, fails to address confounding and consequently $\widehat{\tau}_{BOR2}$ provides a poor approximation to the true ATE. Weighting the incorrectly specified OR model with weights $\widehat{\kappa}(D,X)$, based on a correctly specified PS model, as in the BDR1 model, provides correction for misspecification bias with an average point estimate very close to the true value, but slightly larger posterior variances relative to the correctly specified OR model. The BDR2 model simulation also produces valid point estimates because weighting by weights based on an incorrectly specified PS model does not does not induce bias, but it does increase variance. Finally, if both the OR and PS models are wrongly specified as in BDR3, the model fails to produce a good point estimate of the mean ATE.
Data and model specifications
=============================
Treatment and outcome variable
------------------------------
We have data on the location of fixed speed cameras for 771 camera sites in the following English administrative districts: Cheshire, Dorset, Greater Manchester, Lancashire, Leicester, Merseyside, Sussex and the West Midlands. These sites form our group of treated units. To select potential control sites we randomly sampled a total of 4787 points on the network within our eight administrative districts. The large ratio of potential control to treated units is adopted to ensure that we have a sufficient number of control units after we apply a matching algorithm.
Our outcome variable is the number of personal injury collisions (PICs) per kilometre as recorded from the location of the speed cameras, or in the case of control groups, from the randomly selected point. The PIC data are taken from records completed by police officers each time that an incident is reported to them. The individual police records are collated and processed by the UK Department for Transport as the ‘STATS 19’ data. The location of each PIC is recorded using the British National Grid coordinate system and can be located on a map using Geographical Information System (GIS) software. Because the established dates of speed cameras vary from 2002 to 2004, the period of analysis is from 1999 to 2007 to ensure the availability of collision data for the years before and after the camera installation for every camera site.
Covariates
----------
To adequately adjust for confounding we require a set of measured covariates that adequately represent the characteristics of units that simultaneously determine treatment assignment and outcome. For the UK there exists a formal set of site selection guidelines for fixed speed cameras [see @Gains/et/al:2004] that are extremely valuable in choosing covariates. The criteria are as follows
1. Site length: between 400-1500 m.
2. Number of fatal and serious collisions (FSCs): at least 4 FSCs per km in last three calendar years.
3. Number of personal injury collisions (PICs): at least 8 PICs per km in last three calendar years.
4. 85th percentile speed at collision hot spots: 85th percentile speed at least 10% above speed limit.
5. Percentage over the speed limit: at least 20% of drivers are exceeding the speed limit.
Criteria one to three are primary guidelines for site section and criteria four and five are of secondary importance. There are sites that do not meet the above the above criteria that will still be selected as enforcement sites, mainly for reasons such as community concern and engineering factors.
Selection of the speed camera sites was primarily based on collision history. collision data can be obtained from the STATS 19 database and located on the map using GIS. However, secondary criteria such as the 85th percentile speed and percentages of vehicles over the speed limit are normally unavailable for all sites on UK roads. If speed distributions differ between the treated and untreated groups, then the failure to include the speed data could bias the estimation, an issue discussed in previous research [e.g. @Mountain/et/al:2005; @Gains/et/al:2004]. For untreated sites with the speed limit of 30 mph and 40 mph, the national average mean speed and percentages of speeding are similar to the data for the camera sites. The focus groups for this study are sites with the speed limit of 30 mph and 40 mph throughout the UK. It is reasonable to assume that there is no significant difference in the speed distribution between the treated and untreated groups and hence exclusion of the speed data will not affect the accuracy of the propensity score model.
It is also possible that drivers may choose alternative routes to avoid speed cameras sites. collision reduction at camera sites may include the effect induced by a reduced traffic flow. The benefits of speed cameras will therefore be overestimated without controlling for the change in traffic flow. The annual average daily flow (AADF) is available for both treated and untreated roads and the effect due to traffic flow is controlled for in this study by including the AADF in the propensity score model.
In addition to the criteria that strongly influence the treatment assignment, factors that affect the outcomes should also be taken into account when the propensity score model is specified. We further include road characteristics such as: road types, speed limit, and the number of minor junctions within site length, which are suggested as important factors when estimating the safety impact of speed cameras [@Gains/et/al:2005; @Christie/et/al:2003].
Component model specifications
------------------------------
The outcome variable of interest is the number of collisions per site. For the OR model the response is specified in differenced form, i.e. the number of collisions in the post-treatment period minus the number of collisions in the pre-treatment period. Differencing allows for the existence of unit level time-invariant effects, which could be random or fixed. The PS model is estimated using a logit Generalized Additive Mixed Model (GAMM) specification. Matching and overlap is achieved using nearest neighbour matching via the [MatchIt]{} package in [R]{}. The weighted OR model is then estimated on the trimmed dataset, which satisfies matching and overlap conditions, using a Gaussian GAMM specification. We use GAMMs to avoid making a-priori assumptions on the functional form of the relationships under study.
As mentioned in the introduction, the DR approach is particularly attractive for our application because of the differences inherent in our PS and OR model specifications. Due to the existence of formal criteria for SC assignment we have a high degree of confidence in the ability of our covariates to eliminate confounding via the PS model. For the OR model, differencing of the response variable before and after treatment allows for the existence of site level time-invariant unobserved effects in addition to measured confounders. Thus, there are subtle differences in the way we model the ATE via the PS or OR approaches. A degree of robustness is offered using a DR approach since we will obtain a consistent estimate of the ATE if just one of the component models is well specified.
Results
=======
The objective of our application is to estimate the [*marginal*]{} effect of SCs on RTCs, having adjusted for baseline confounders. We estimate the following models: an OR model, an IPW model, a DR model comprising an OR model weighted with the inverse PS covariate (DR), and a na[ï]{}ve model which is simply the OR model without covariates. For the na[ï]{}ve model we report results using the matched and full samples. All models are repeatedly estimated using the approximate Bayesian approach outlined above. In addition to the posterior predictive distribution for the ATE we report point estimates at the mean of the posterior. For comparison, we also report Frequentist results.
The results are shown in table 2 below including means and credible intervals of the ATE distributions. Our causal models (OR, IPW and DR) indicate that the presence of speed cameras corresponds with an average change in the number of RTCs of -29% to -31% . Note that the approximate Bayesian and Frequentist point estimates are very similar, which is what we would expect for linear models with uninformative priors. In comparison, the Na[ï]{}ve model which does not adjust for confounding, finds a higher ATE of -35% using the matched sample and -39% using the unmatched sample. Figure \[appfig\] below shows the posterior predictive distribution derived from the DR model.
\[results\]
---------------------------- ---------------- -------- -------------------- --------- --------
posterior mean s.d. 95% cred. int. Est. s.e.
OR -28.981 7.002 (-42.912, -15.971) -28.746 7.933
IPW -31.313 11.107 (-53.083, -9.543) -31.103 10.453
DR -28.912 7.041 (-43.240, -15.267) -28.490 8.003
Na[ï]{}ve (matched sample) -35.039 10.045 (-55.284, -16.029) -36.260 10.617
Na[ï]{}ve (full sample) -39.144 6.504 (-48.681, -16.037) -40.297 4.213
---------------------------- ---------------- -------- -------------------- --------- --------
Thus, it would appear that correcting for potential sources of confounding serves to reduce the magnitude of our ATE estimates, but we still find a substantial reduction in RTCs associated with presence of speed cameras. The difference in estimated ATE between the na[ï]{}ve and causal models makes sense given that the formal criteria used to assign SCs favours sites that have exhibited high rates of collisions in the past. Crucially, our causal models imply that SCs do make a real difference to RTCs over and above the modelled effect of confounding from non random assignment.
Conclusions
===========
In this paper we have the quantified the causal effect of speed cameras on road traffic collisions via an approximate Bayesian doubly robust approach. This is the first time such an approach has been applied to study road safety outcomes. The method we propose could be used more generally for estimation of crash modification factor (CMF) distributions. Simulations demonstrate that the approach is doubly-robust for average treatment effect estimation. Our results indicate that speed cameras do cause a significant reduction in road traffic collisions, by as much as 30% on average for treated sites. This is an important result that could help inform public policy debates on appropriate measures to reduce RTCs. The adoption of evidence based approaches by public authorities, based on clear principles of causal inference, could vastly improve their ability to evaluate different courses of action and better understand the consequences of intervention.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We argue that a class of strongly spin-orbit coupled materials, including some pyrochlore iridates and the inverted band gap semiconductor HgTe, may be described by a minimal model consisting of the Luttinger Hamiltonian supplemented by Coulomb interactions, a problem studied by Abrikosov and collaborators. It contains two-fold degenerate conduction and valence bands touching quadratically at the zone center. Using modern renormalization group methods, we update and extend Abrikosov’s classic work and show that interactions induce a quantum critical non-Fermi liquid phase, stable provided time-reversal and cubic symmetries are maintained. We determine the universal power-law exponents describing various observables in this “Luttinger Abrikosov Beneslavskii” state, which include conductivity, specific heat, non-linear susceptibility and magnetic Gruneisen number. Furthermore, we determine the phase diagram in the presence of cubic and/or time-reversal symmetry breaking perturbations, which includes topological insulator and Weyl semi-metal phases. Many of these phases possess an extraordinarily large anomalous Hall effect, with the Hall conductivity scaling [*sub-linearly*]{} with magnetization, $\sigma_{xy}\sim M^{0.51}$.'
author:
- 'Eun-Gook Moon'
- Cenke Xu
- Yong Baek Kim
- Leon Balents
title: 'Non-Fermi liquid and topological states with strong spin-orbit coupling '
---
= 10000
Divining the nature of critical non-Fermi liquid (NFL) phases of electrons is one of the most outstanding problems in correlated electron systems, such as cuprates, pnictides, and heavy fermion materials [@louis; @hirsch; @chubukov; @gegenwart]. Theory, spurred largely by NFL behavior in the cuprates, has focused on models with large Fermi surfaces, appropriate to these materials. Though there have been promising technical advances[@max], this approach has not provided clear resolution of experimental puzzles. Recent theory and experiment have uncovered a new frontier for correlated phenomena: Mott correlation physics in materials with strong spin-orbit coupling (SOC) [@pesin; @bjkim1]. Of particular interest in this regard are the 5d transition metal oxides, where compelling evidence has been built for Mott phenomena driven cooperatively by SOC and Coulomb interactions [@pesin; @bjkim1; @takagi]. Here we consider the possibility of distinct NFL states in the strong SOC regime.
We are particularly motivated by the pyrochlore iridates, A$_2$Ir$_2$O$_7$, where A is a lanthanide element [@maeno; @matsuhira]. These materials display $T>0$ metal-insulator transitions, with the metal-insulator transition temperature decreasing with increasing A radius. The progression culminates with Pr$_2$Ir$_2$O$_7$, which is a highly unconventional metal down to the lowest temperatures [@nakatsuji; @machida]. It displays logarithmic NFL behavior of the magnetic susceptibility, and a remarkable enormous zero field anomalous Hall effect (AHE), in the absence of any measurable magnetization [@nakatsuji; @machida].
With these studies as motivation, we utilize prior studies of the electronic structure of pyrochlore iridates[@ashvin; @ybkim1; @ybkim2] to show that a minimal description for the electronic states in their paramagnetic phase is a storied Hamiltonian from semiconductor physics: the Luttinger model of inverted band gap semiconductors [@luttinger]. This model has gained recent notoriety for its relevance to HgTe, the starting material for some topological insulators [@TI; @qi; @zhang; @zhang2]. While HgTe is a weakly correlated material where band structure alone provides a sufficient description, in the 5d materials the Luttinger Hamiltonian must be supplanted by interactions.
In this paper, we carry out such an analysis, rediscovering and extending a storied analysis by Abrikosov and Beneslavskii of Coulomb forces on the Luttinger problem [@ab1; @ab2]. Using modern renormalization group (RG) techniques, we confirm Abrikosov’s conclusion that long-range Coulomb interactions convert the quadratic band touching into a quantum critical NFL, prove the stability of the state within an $\epsilon$ expansion, and calculate the full set of anomalous dimensions characterizing the state. Consequently, we call the resulting phase a [*Luttinger-Abrikosov-Beneslavskii*]{} (LAB) state. While the LAB phase is stable in the presence of time-reversal and cubic symmetries, we show that it is a “parent” state for other exotic states that can be reached by breaking either or both of these: metallic and double-Weyl semi-metallic phases with enormous anomalous Hall effects (AHEs), and topological insulators. We discuss the implications for the iridates at the end of the paper.
The LAB phase itself has striking properties. Its NFL nature is revealed directly by algebraic singularities in the electron spectral function (probed in angle-resolved photoemission) and in optical conductivity, as well as indirectly through many power law thermodynamic and response functions.
![ Phase diagram of the perturbed LAB in the space of renormalized strain to Zeeman field ratio, $\Delta \equiv
(\delta/H)_R$, versus cubic Zeeman angle, $\theta$. Here“Weyl” denotes the (double) Weyl semimetal, “Ins.” insulator, and “Metal” a metallic phase which has Weyl points shifted from the Fermi energy in the region below the dashed line. For $H=0$, the insulator is a topological insulator. []{data-label="fig1"}](generalPD3.pdf){width="3.3in"}
We now turn to the exposition of these results. We consider the [ *paramagnetic*]{} band structure based on prior work [@ashvin; @ybkim1; @ybkim2], which argued that the states at the zone center ($\Gamma$ point) near the Fermi energy are comprised of the four-dimensional representation, which can be described by “angular momentum” operators $\vec{J}$ (which are $j=3/2$ matrices) transforming as the T$_2$ representation of the cubic group. In our minimal model, we assume only these states close to $\Gamma$ are important. Then $k\cdot p$ theory and cubic symmetry determines the band structure in its vicinity to be precisely described by the Luttinger Hamiltonian with three effective mass parameters [@luttinger; @murakami], $$\begin{aligned}
\mathcal{H}_0(k)&=&\frac{k^2}{2 \tilde{M}_0} + \frac{ \frac{5}{4}k^2-(\vec{k} \cdot \vec{J})^2}{2m} - \frac{ (k_x^2 J_x^2 + k_y^2 J_y^2 +k_z^2 J_z^2)}{2 M_c} .\nonumber\end{aligned}$$ This describes doubly degenerate bands with energy $$\begin{aligned}
E_{\pm}(k)= \frac{k^2}{2M_0} \pm \sqrt{\left(\frac{k^2}{2m}\right)^2+ \frac{m+2 M_c}{4 m M_c^2} p_c(k) },\end{aligned}$$ where $ p_c(k) = \sum_{i} k_i^4 -\sum_{i \neq j} k_i^2 k_j^2$ and $M_0 = (4 M_c \tilde{M}_0)/(4M_c-5 \tilde{M}_0)$. Henceforth we assume $M_0 > m$, which describes conduction and valence bands touching quadratically at $E=k=0$, where the chemical potential for the undoped material crosses.
The LAB is obtained by adding to this the long-range Coulomb interaction. We implement the latter by a scalar potential $\varphi$, which in the Euclidean path integral formalism gives the action $$\begin{aligned}
&& {S}_{L} = \int d \tau d^d x \left\{ \psi^{\dagger} \Big[\partial_{\tau}
-i e \, \varphi +\hat{\mathcal{H}}_0\Big] \psi + \frac{c_0}{2} (\partial_i
\varphi)^2 \right\}\!, \label{eq:2}\end{aligned}$$ with $\hat{\mathcal{H}}_0 = \mathcal{H}_0(-i\vec\nabla)$ and $c_0=1/4\pi$. Here $\psi$ is a four-component spinor, but subsequently we will artificially add an additional $U(N_f)$ flavor index, which allows a check on our calculations by large $N_f$ methods; the physical case is $N_f=1$. Eq. contains in addition to the three mass parameters, the Coulomb coupling constant $e$. For $e=0$, scale invariance is manifest, with the scaling dimensions $[x^{-1}]=1 \, , \, [\tau^{-1}]=z
\, , \, [\psi] = \frac{d}{2} \, , \, [\frac{1}{m}]=z-2$, $[\varphi]=(d+z-2)/2$. Here we introduce the dynamic critical exponent ($z$), which is naturally $z=2$ with $e=0$, but will become non-trivial with interactions.
Directly in the physical case $d=3$, the dimension of the coupling constant is $[e^2] =1$, so Coulomb interactions are strongly relevant. Therefore we employ the $\varepsilon=4-d$ expansion to control the RG analysis. As familiar from quantum electrodynamics, three one loop Feynmann diagrams contribute to leading order in $\varepsilon$: the fermion self-energy, boson self-energy, and vertex correction. Here we show that the relevance of Coulomb interactions signals, rather than a flow to strong coupling and a symmetry breaking instability, the formation of a new [*stable*]{} interacting fixed point, which describes the critical non-Fermi liquid LAB state (Abrikosov’s analysis tacitly assumes this stability).
The RG is carried out perturbatively in $e$, but non-perturbatively in the mass parameters. Thus a full treatment gives non-trivial and complete beta functions for the two dimensionless mass ratios $m/M_0$, $m/M_c$; these are given in the Supplementary Material. The analysis of the full RG shows, however, that there is a single stable [*isotropic*]{} fixed point corresponding to $m/M_0=m/M_c=0$, so for simplicity we quote in the main text only the results in the vicinity of this point.
In this limit, the leading contribution to the bosonic self energy becomes $$\begin{aligned}
\frac{1}{N_f}\Sigma_{\varphi}(q, 0)
&=& -(2m) e^2 \left[ \int \! \frac{d^dk}{(2\pi)^d}\, \frac{1}{k^4}\right] \times q^2 ,\end{aligned}$$ where we took the $\omega \rightarrow 0$ limit because frequency dependence is subdominant. The divergence should be absorbed by rescaling the bosonic field, $\varphi \rightarrow e^{-\eta_b
d\ell}\varphi$ upon reduction of the hard momentum cutoff $\Lambda
\rightarrow e^{-d\ell}\Lambda$, which defines the RG parameter $\ell$. This gives the bosonic anomalous dimension $\eta_b = 2
N_f u$ [@large_n], where the dimensionless coupling constant is $u= \frac{m\, e^2 }{ 8\pi^2 c_0 \Lambda^{4-d}}$, which has the physical meaning in $d=3$ of the ratio of the real space cutoff to the effective Bohr radius. The frequency dependence of the one loop fermionic self-energy and the vertex correction both vanish, the result of a Ward identity. For $k\neq 0$, the fermion self-energy gives mass corrections, e.g. $\delta (1/m) = 8u/(15m) \times d\ell$ to leading order. Detailed analysis is given in the Supplementary Material. Given these calculations, we choose $z=2- 8u/15$ to keep the mass $m$ fixed, which gives the RG equations, to lowest order in $m/M_c, m/M_0$: $$\begin{aligned}
\label{betaftn} %EG
&& \frac{d }{d \ell} u = \varepsilon \, u - \frac{30N_f+8}{15}
u^2, \\
&& \frac{d}{d \ell}\left(\tfrac{m}{M_c}\right ) = -0.152 u
\left(\tfrac{m}{M_c}\right) , \frac{d}{d \ell}\left(\tfrac{m}{M_0}\right ) = - \tfrac{8}{15}u
\left(\tfrac{m}{M_0}\right) .\nonumber \end{aligned}$$
From the first equation above we find the fixed point coupling and hence dynamical exponent, $$u^* = \frac{15}{30 N_f+8 } \varepsilon
\, ,\, z=2-\frac{4}{15 N_f+4 } \varepsilon ,$$ and since $u^*>0$, the second line in Eq. implies both $m/M_0$ and $m/M_c$ are irrelevant. This establishes the existence and nature of the stable, isotropic fixed point describing the LAB phase. As a check, we have carried out a large $N_f$ expansion, which gives the same bosonic anomalous dimension as in the $\varepsilon$ expansion at the one-loop level, supporting the stability of the LAB phase.
The presence of the stable interacting fixed point can be understood physically as a balance of partial dynamical screening of the Coulomb interactions by electron-hole pairs and mass enhancement of the same quasiparticles by pairs. This situation is in sharp contrast to the case of a vanishing [ *indirect*]{} band gap, for which to leading order in the long-range Coulomb interaction electrons and holes are separately conserved, so there is no screening by virtual electron-hole pairs, and exciton formation destabilizes the putative gapless state[@halperin].
Using the RG, we can evaluate the anomalous dimension of any physical operator. By charge conservation, $[\psi^{\dagger}\psi]=d$. Due to the isotropy of the fixed point, there are only two non-trivial values for the other charge-conserving fermion bilinears. We obtain , $[\psi^{\dagger} \Gamma_a \psi] =d+
\eta_1$ , $[\psi^{\dagger} \Gamma_{ab} \psi]=d+\eta_{12} $, , where $\Gamma_a$ are the (time-reversal invariant) Dirac gamma matrices, $\Gamma_{ab} = -\frac{i}{2}[\Gamma_a,\Gamma_b]$ are time-reversal odd, and $a,b=1,2,\cdots,5$. Using the standard operator insertion technique, we find $\eta_{1} = - \frac{6 }{15 N_f +4}\varepsilon$ and $\eta_{12} =- \frac{3 }{15 N_f +4}\varepsilon$. These operators describe many physical observables, [*e.g.*]{} the “angular momentum" operator $J_z \sim
\psi^{\dagger} (-\Gamma_{34} -\frac{1}{2} \Gamma_{12}) \psi $. The negative anomalous dimension of these operators suggests the a schematic picture of power-law excitons due to electron-hole attraction. For pairing channels, we find positive anomalous dimensions, consistent with this view. The local pairing channel has $\eta_{pairing} = \frac{u^*}{5}= \frac{3}{30 N_f + 8}\epsilon$.
Using these results, we obtain thermodynamic responses such as the specific heat, $c_v \sim T^{d/z}\approx T^{1.7}$ and the spin susceptibility $\chi(T) \sim a+b \,T^{(d-z+2\eta_{12})/z}\approx a + b \,
T^{0.5}$, with some constants $a,b$. Interestingly, the non-linear susceptibility $\chi_3 = \left.\partial^3 M/\partial H^3\right|_{H=0}
\sim T^{-(3z-4\eta_{12}-d)/z} \approx T^{-1.7}$ diverges, as in spin glasses but with completely different physics. Comparing the scaling of current and electric field gives the usual result $[\sigma_{ij}] =d-2$. Consequently, the temperature and frequence dependence of the conductivity is $\sigma(\omega,T) \sim T^{1/z}
\mathcal{F}(\omega/T)$, and a [*clean, undoped*]{} LAB is therefore a power-law insulator. We now turn to the effect of applied strain and Zeeman field upon the LAB. These perturbations break cubic/time-reversal symmetries, and thus destabilize the LAB. Due to the isotropic nature of the LAB fixed point, the response to the Zeeman field alone is to leading order independent of its direction (the cubic mass $1/M_c$ can be “dangerously irrelevant”, however – see below), so we take it to lie along the (001) direction. We consider for simplicity tetragonal strain which preserves $C_4$ rotation about this axis (in the absence of Zeeman field, the direction of strain is again unimportant). This leads to the perturbations $$\begin{aligned}
{\mathcal H}' = -\delta (J_z^2-\frac{5}{4}) -H (\cos(\theta) J_z + \sin(\theta) J_z^3) ,\label{pert}\end{aligned}$$ where $\delta$ parametrizes the strain, $h$ is the Zeeman field, and $\theta$ controls the strength of the cubic Zeeman term allowed by the cubic symmetry [@kane_fu; @ewald]. Using the RG results, the dimensions of these perturbations are $[\delta]=z-\eta_1 \approx 2.1$ and $[H]=z-\eta_{12}\approx 1.9$; [*i.e.*]{} strain is slightly enhanced while Zeeman field is slightly suppressed by interactions. However, both dimensions are positive and close to $2$, so that they are strongly [*relevant*]{}. They flow to strong coupling under the RG, and the fate of the system must be re-analyzed in the limit.
To do so, we assume, and check self-consistently, that interactions have weak effects at strong coupling, and simply solve the quadratic Hamiltonian (with $m/M_0=m/M_c=0$) in the presence of the renormalized $\mathcal{H}'$. The result depends upon the dimensionless quantities $\theta$ and the renormalized coupling ratio $\Delta = (\delta/H)_R \sim
\delta/H^{(z-\eta_1)/(z-\eta_{12})}$. For $H=0$ ($\Delta=\infty$), we have time-reversal invariance, and we recover the known result that strain $\delta>0$ induces a gapped, 3d TI phase, as observed in HgTe [@zhang]. The situation in applied Zeeman field is more interesting. Notice that for $\vec{k} = k\hat{z}$, $J_z$ is a good quantum number, and there is no level repulsion between bands of different $J_z$. This allows (non-degenerate) bands to cross along this axis, which indeed occurs when $|\Delta|$ is not too large. Further analysis in the Supplementary Material shows that these crossings correspond to a pair of [*double Weyl points*]{}, with linear dispersion along the $z$ axis and quadratic dispersion normal to it. These points are strength $\pm 2$ monopoles in momentum space. Away from the $k_z$ axis, electron and hole pockets may accidentally cross the Fermi energy. If this does not occur, one has a pristine double Weyl semimetal, which occurs for the angular range $\theta_1 \le \theta \le \theta_2$, where $\theta_1 =
-\tan^{-1}(\frac{8+4 \sqrt{3}}{7 \sqrt{3} +26})$ and $\theta_2 =
\tan^{-1}(\frac{8-4 \sqrt{3}}{7 \sqrt{3} -26})$ for $\delta=0$, as shown in horizontal axis of Fig. \[fig1\]. When $0<|\Delta|<\infty$, we observe insulating, double Weyl semimetal, and Weyl metal (with coexisting electron-hole pockets) phases, as shown in Figure \[fig1\]. Note that in all these phases, the Coulomb interactions become either unimportant (in the insulator), screened (in the metal), or marginally irrelevant (in the Weyl semimetal), justifying our treatment of the phase diagram to a first approximation.
More subtle effects may make small modifications to this picture. Coulomb interactions can destabilize some of the quantum phase transitions in Figure \[fig1\], leading to intermediate phases. When the magnetic field is applied along a low symmetry axis, the double Weyl points can split into multiple single Weyl points, once the effects of the cubic mass $1/M_c$ is included, which is dangerously irrelevant in this case.
A striking experimental consequence of this phase diagram is the AHE due to the Zeeman field, which could originate either from an external magnetic field or as an exchange field due to local moments in the material. The latter is particularly interesting in light of the experimental results on Pr$_2$Ir$_2$O$_7$, which shows a large AHE in a regime where the magnetization $M$ is immeasurably small [@nakatsuji; @machida]. On symmetry grounds, $\sigma_{xy}
\neq 0$ [*implies*]{} $M_z \neq 0$, but evidently $\sigma_{xy}$ is unusually large relative to $M$. This behavior is in fact characteristic of the LAB: since the Hall conductivity has dimensions of inverse length, we expect $\sigma_{xy} \sim H^{1/(z-\eta_{12})} \approx H^{0.51}$. In the situation relevant for Pr$_2$Ir$_2$O$_7$, the Zeeman field is generated by Kondo exchange with the Pr moments, so $H \sim J_K M$, with $M$ the (dimensionless) Pr magnetization, which implies a highly unconventional sublinear dependence of $\sigma_{xy}$ on $M$ for the pristine LAB. If the Fermi level is displaced from the band touching by an amount $\epsilon_F$, then we expect, treating the above power as a square root, $\sigma_{xy} \sim
\frac{e^2}{h} \sqrt{\frac{m H}{\hbar^2}} \mathcal{S}(H/\epsilon_F)$, where $\mathcal{S}$ is a scaling function (see Supp. Mat.). This gives an order of magnitude quantitative estimate $$\label{eq:8}
\sigma_{xy} \sim 10^3 \Omega^{-1} cm^{-1} \times
\left\{ \begin{array}{cc} \sqrt{\frac{m}{m_e}} \sqrt{\frac{J_K}{\rm Ry}}
\sqrt{M}, & J_K M \gg \epsilon_F \\
\sqrt{\frac{m}{m_e}} \frac{J_K}{\sqrt{\epsilon_F {\rm Ry}}} M,
& J_K M \ll \epsilon_F \end{array}\right. \; ,$$ where Ry=13.6eV is the Rydberg, and $m_e$ is the electron mass. The lower regime gives $\sigma_{xy} \sim 0.1 \Omega^{-1} cm^{-1}$, within an order of magnitude of observations in Pr$_2$Ir$_2$O$_7$,[@machida] for parameters $m \sim 20m_e$ (estimated from the calculations in Ref. ) , $\epsilon_F \sim $10meV, $J_K \sim$100K (estimated from the measured Curie-Weiss temperature), and $M \sim 0.01$.
Another interesting experimental observation in Pr$_2$Ir$_2$O$_7$ is a diverging magnetic Gruneisen number $\Gamma_H = \frac{1}{T}
\left. \frac{\partial T}{\partial H}\right|_S$ in the zero field limit [@si; @gruneisen; @227]. In a purely electronic system with no local moment contribution to the entropy, we can readily obtain the behavior of the LAB in the low temperature limit, $$\begin{aligned}
\Gamma_H(H,T) = - \frac{d- y_0 z}{y_0 (z+\eta_H)} \frac{1}{H} ,\label{eq:3}\end{aligned}$$ which depends upon the exponent $y_0$ defining the temperature dependence of the specific heat, $C \sim T^{y_0}$, of the LAB phase. In the Weyl metal, $y_0=1$, and $\Gamma_H<0$, while for isolated double Weyl points, $y_0=2$, and $\Gamma_H>0$. Thus one may imagine a sign change of the Gruneisen number when field or strain is varied. Note that in Pr$_2$Ir$_2$O$_7$, there is certainly a large local moment contribution to the entropy, so that Eq. is not literally applicable. Nevertheless, the LAB physics may play some role in this quantity.
In conclusion, we have described a novel NFL phase occurring in correlated strong SOC systems, with a natural connection to the pyrochlore iridates. Even with weak correlation effects, some of the phenomena discussed here can be observed with only minor modifications, and it would be interesting to search for them in HgTe. Future theoretical studies should include more comprehensive treatment of breaking of cubic symmetries, and the effects of disorder and doping.
We thank T. Hsieh, L. Fu, and P. Gegenwart for discussions on Luttinger model and Gruneisen number. This work was supported by NSF Grant DMR-1151208 and Hellman Family Foundation (EGM, CX), the David and Lucile Packard Foundation, the Alfred P. Sloan Foundation (CX), NSF–DMR–1206809 (LB), and NSERC, CIFAR (YBK). YBK and LB acknowledge the support and hospitality from the Aspen Center for Physics, funded by NSF grant PHY-1066293, and the KITP, funded by NSF grant PHY–1125915.
[99]{}
L. Taillefer, Annu. Rev. Cond. Mat. [**1**]{}, 51 (2010). P.J. Hirschfeld, M.M. Korshunov, and I.I. Mazin, Rep. Prog. Phys. [**74**]{}, 124508 (2011). A. Chubukov, Annu. Rev. Cond. Mat. Phys. [**3**]{}, 57 (2012). P. Gegenwart [*et al.*]{}, Nat. Phys. [**4**]{}, 186 (2008). See, for example, M. Metlitski and S. Sachdev, Phys. Rev. B [**82**]{}, 075127 (2010); [*ibid*]{} 075128 (2010). D. Pesin and L. Balents, Nat. Phys. [**6**]{}, [376]{} (2010). B. J. Kim [*et al.*]{}, Phys. Rev. Lett., [**101**]{}, 076402 (2008). B. J. Kim [*et al.*]{}, Science [**323**]{}, 1329 (2009). D. Yanagishima and Y. Maeno, J. of Phys. Soc. Japan [**70**]{}, 2880 (2001). K. Matsuhira [*et al.*]{}, J. of Phys. Soc. Japan [**76**]{}, 043706 (2007). S. Nakatsuji [*et al.*]{}, Phys. Rev. Lett. [**96**]{}, 087204 (2006). Y. Machida [*et al.*]{}, Nature [**463**]{}, 210 (2010).
X. Wan [*et al.*]{}, Phys. Rev. B, [**83**]{}, 205101 (2011). W. Witczak-Krempa and Y. B. Kim, Phys. Rev. B [**85**]{}, 045124 (2012). B.-J. Yang and Y. B. Kim, Phys. Rev. B [**82**]{}, 085111 (2010).
J. M. Luttinger, Phys. Rev. [**102**]{}, 1030 (1956).
M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. [**82**]{}, 3045 (2010). X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. [**83**]{}, 1057 (2011). B. A. Bernevig, T. L. Hughes, and S.-C Zhang, Science [**314**]{}, 1757 (2006). M. Konig [*et. al.*]{}, Science [**318**]{}, 766 (2007).
A. A. Abrikosov and S. D. Beneslavskii, Sov. Phys. JETP [**32**]{}, 699 (1971). A. A. Abrikosov, Sov. Phys. JETP [**39**]{}, 709 (1974). S. Murakami, N. Nagaosa, and S.-C.Zhang, Phys. Rev. B [**69**]{}, 235206 (2004). In this paper, we perform the $\varepsilon=4-d$ expansion using the $4 \times 4$ gamma matrices. Alternatively, one could use $8 \times 8$ matrices in $d=4$ with a half of the number of fermion flavors and calculate logarithmic divergences directly. B. I. Halperin and T. M. Rice, Solid State Physics [**21**]{}, 115 (1968).
L. Fu and C. Kane, Phys. Rev. B [**76**]{}, 045302 (2007). B. J. Roman and A. W. Ewald, Phys. Rev. B [**5**]{}, 3914 (1972).
A. A. Burkov and L. Balents, Phys. Rev. Lett. [**107**]{}, 127205 (2011). L. Zhu, [*et al.*]{}, Phys. Rev. Lett. [**91**]{}, 066404 (2003). P. Gegenwart [*et al.*]{}, J. Low Temp. Phys. [**161**]{}, 117 (2010). P. Gegenwart, private communication.
Supplementary material
======================
Representations of the Hamiltonian
----------------------------------
In this section, we provide more information about the different representations used in the main text. The Hamiltonian reads $$\begin{aligned}
\mathcal{H}_0(k)&=&\alpha_1 k^2 + \alpha_2 (\vec{k} \cdot \vec{J})^2 + \alpha_3 (k_x^2 J_x^2 + k_y^2 J_y^2 +k_z^2 J_z^2) \nonumber \\
&=& \frac{k^2}{2 \tilde{M}_0} + \frac{ (\frac{5}{4}k^2-\vec{k} \cdot \vec{J})^2}{2m} - \frac{ (k_x^2 J_x^2 + k_y^2 J_y^2 +k_z^2 J_z^2)}{2 M_c} \nonumber \\
&=&\frac{d_a(k)}{2m} \Gamma^a + \frac{k^2}{2M_0} + \frac{d_4(k) \Gamma^4 + d_5(k) \Gamma^5}{2M_c} .\end{aligned}$$ The first line uses the conventional Luttinger parameters ($\alpha_{1,2,3}$) in the $j=3/2$ matrix representation, and the second line is the form used in the main text. For the purpose of computations, it is convenient to introduce the Clifford gamma matrices ($\Gamma_{a}$) in the third line as in the paper by Murakami [*et al.*]{} [@murakami]. $$\begin{aligned}
&&\epsilon_a(k)= \frac{d_a(k)}{2m} , \nonumber \\
&& d_1(k)= -\sqrt{3}k_y k_z \, , \,\, d_2(k)= -\sqrt{3}k_x k_z \,, \,\, d_3(k)= -\sqrt{3}k_x k_y , \nonumber \\
&& d_4(k) = \frac{-\sqrt{3}}{2}(k_x^2 -k_y^2) \,, \,\, d_5(k) =
\frac{-1}{2}(2 k_z^2 - k_x^2 -k_y^2) .\end{aligned}$$ It is straightforward to relate the masses used in the main text and the Luttinger $\alpha_i$ parameters. This can be done by expressing the spin operators in terms of gamma matrices, using the equalities $$\begin{aligned}
&& J_x = \frac{\sqrt{3}}{2} \Gamma_{15} - \frac{1}{2} (\Gamma_{23} - \Gamma_{14}) \ ,\nonumber \\
&& J_y = -\frac{\sqrt{3}}{2} \Gamma_{25} + \frac{1}{2} (\Gamma_{13} + \Gamma_{24}) \ ,\nonumber \\
&& J_z = - \Gamma_{34} - \frac{1}{2} \Gamma_{12} \ ,\end{aligned}$$ where $\Gamma_{ab} = \frac{1}{2 i} [\Gamma_a, \Gamma_b]$ is used.
Weyl semimetal
--------------
Here we consider how the Weyl semimetal appears, taking for simplicity the case $\delta =0 , \theta_1 < \theta < \theta_2$. It is straightforward to generalize this to the cases with $\delta\neq 0$. For convenience, we set $m=1/2$. For $\vec{k}=(0,0,k_z)$, the Hamiltonian in the presence of the Zeeman field, $h$ (along the same axis), becomes $$\begin{aligned}
\mathcal{H}(k_z \hat{z})&=& k_z^2 (\frac{5}{4}-J_z^2)-H ( \cos(\theta) J_z +\sin(\theta) J_z^3). \nonumber \end{aligned}$$ Clearly, the energy eigenstates can be labeled by both $k_z$ and $J_z=\pm 1/2, \pm 3/2$. One can readily see that level crossings occur between the two bands with $J_z=1/2$ and $J_z=-3/2$. In the vicinity of these crossing points, the other states with $J_z=-1/2, +3/2$ can be discarded, and the Hamiltonian projected onto the two level subspace of low energy states. We introduce Pauli matrices in this subspace, so that $\tau^z =
|\tfrac{1}{2}\rangle\langle\tfrac{1}{2}| -
|-\tfrac{3}{2}\rangle\langle-\tfrac{3}{2}|$, $\tau^+ = (\tau^x + i
\tau^y)/2 = |\tfrac{1}{2}\rangle\langle-\tfrac{3}{2}| =
(\tau^-)^\dagger$, and $\tau^0 =
|\tfrac{1}{2}\rangle\langle\tfrac{1}{2}| +
|-\tfrac{3}{2}\rangle\langle-\tfrac{3}{2}|$, which is the identity matrix in the $2\times 2$ subspace. We define two energy parameters, $\epsilon_{-3/2} = H (\frac{3}{2}\cos(\theta) +
\frac{27}{8} \sin(\theta))$ and $\epsilon_{1/2} =-H
(\frac{1}{2}\cos(\theta) + \frac{1}{8} \sin(\theta)) $. Then the reduced Hamiltonian becomes $$\begin{aligned}
&&\mathcal{H}_{2} = \epsilon_{+} \tau^0+ (\epsilon_{-}+d_5(k)) \tau^z +d_4(k) \tau^x - d_3(k) \tau^y, \nonumber \\
&&\epsilon_{\pm}=(\epsilon_{-3/2}\pm\epsilon_{1/2})/2.\end{aligned}$$ There are level crossing points at $k_x=k_y=0$ and $k_z=\pm
K$, with $K=\sqrt{\epsilon_-}$. We expand around these points, letting $k_x=p_x$, $k_y=p_y$ and $k_z =\pm
K+ p_z$, which gives $\mathcal{H}_2(\pm K\hat{z}+\vec{p}) = \epsilon_+ \tau^0 + %EG
\mathcal{H}_{2}^\pm(\vec{p})$, with, to leading order in $\vec{p}$, $$\begin{aligned}
\mathcal{H}_{2}^\pm = \mp v\, p_z \tau^z +d_4(p) \tau^x + d_3(p) \tau^y \ , \nonumber\end{aligned}$$ with $v=2\sqrt{\epsilon_-}$. The energy spectrum is $$\begin{aligned}
E(p)=\pm \sqrt{v^2 p_z^2 + \frac{3}{4}(p_x^2+p_y^2)^2}. \nonumber\end{aligned}$$ We see that the electrons disperse linearly along the field and quadratically orthogonal to it, near the touching point. This can be understood as a consequence of 4-fold rotational symmetry around the $z$ axis. Since $|J_z\rangle \rightarrow e^{i \pi J_z/2} |J_z\rangle$ under such a rotation, the operators $\tau^\pm$ carry a net angular momentum of $\pm 2$, and therefore must couple to the “d-wave” combinations of $p_x$ and $p_y$, which are precisely given by $d_3(p)$ and $d_4(p)$.
Though the quadratic dispersion normal to the field is due to symmetry, the touching itself has a topological character. To see it, it is convenient to define the reduced Hamiltonian in the form $$\label{eq:4}
\mathcal{H}_{2}^\pm = \vec{b}_\pm(\vec{p}) \cdot \vec\tau,$$ with $$\label{eq:5}
\vec{b}_\pm(\vec{p}) = \left( \begin{array}{c} -\frac{\sqrt{3}}{2}(p_x^2-p_y^2) \\
\sqrt{3} p_x p_y \\ \mp v p_z \end{array}\right) .$$ From this, one can define the $U(1)$ Berry flux, which is analogous to a magnetic field in momentum space, $$\label{eq:6}
\mathcal{B}_\pm^\mu = \frac{1}{8\pi}\epsilon^{\mu\nu\lambda} \hat{b}_\pm
\cdot \partial_\nu \hat{b}_\pm \times \partial_\nu \hat{b}_\pm,$$ where $\hat{b}_\pm = \vec{b}_\pm/|\vec{b}_\pm|$, and the derivatives are with respect to $\vec{p}$. By construction, the magnetic field is divergenceless, $\partial_\mu \mathcal{B}_\pm^\mu=0$, away from points of singularity where $\vec{b}_\pm$ vanishes. However, the band touching point is such a singularity, and it is in fact a source of Berry flux. To see this, one may compute the integral of the flux of $\mathcal{B}^\mu$ through a sphere around one of the touching points. It is straightforward to compute this integral, and by doing so one finds $$\label{eq:7}
\partial_\mu \mathcal{B}^\mu(\vec{p}) = \pm 2 \delta(\vec{p}).$$ Thus each band touching is a source of [*two*]{} quanta of Berry flux, and can be therefore considered a [*double Weyl point*]{}. The net Berry flux of both double Weyl points added together vanishes, which is required as the Brillouin zone is a closed manifold without any boundary through which a net flux may escape.
The Berry flux of each double Weyl point is directly related to the Hall conductivity. Indeed, one can show quite generally that the Hall conductivity is proportional to separation of the points,[@anton] $$\begin{aligned}
\sigma_{xy} = &=&2\times \frac{e^2 }{h} \times \frac{2K}{2\pi} \ , \nonumber \end{aligned}$$ where the first factor of $2$ is due to the doubled nature of the Weyl points. This is valid at zero temperature when the Fermi level is at the energy of the double Weyl points. Neglecting the small corrections to the exponent due to Coulomb interactions, we have $K
\sim \sqrt{m H}$, which explains the estimate in Eq. of the main text.
RG equations
------------
Here we describe the renormalization of the fermionic propagator, which was not presented in the main text. We start with the Hamiltonian $$\begin{aligned}
\mathcal{H}_0(k)&=& \frac{k^2}{2M_0} \Gamma^0+\epsilon_a(k) \Gamma^a + \frac{d_4(k) \Gamma^4 + d_5(k) \Gamma^5}{2M_c} \ , \nonumber\end{aligned}$$ where we defined $\Gamma^0$ as the $4\times 4$ unit matrix to emphasize the matrix structure. The corresponding effective Lagrange density is $$\begin{aligned}
\mathcal{L}_{\rm eff} = \psi^{\dagger} (-i \omega_n \Gamma^0
+\mathcal{H}_0(k) - \Sigma_{f}(k,i \omega_n))\psi \nonumber .\end{aligned}$$ Here we have included the one-loop self energy needed for the $\varepsilon$ expansion. It is given by $$\begin{aligned}
&& \Sigma_{\psi}(k, i \omega_n) = \\
&& -g^2 \int_q T \sum_{\Omega_n} G_f(k+q, i \omega_n+i \Omega_n) G_{\varphi} (q,i \Omega_n) \ ,\nonumber \end{aligned}$$ where $G_f^{-1}(k,i \omega) = -i \omega \Gamma^0 +\mathcal{H}_0(k)$, $G_{\varphi}^{-1}(q,i \Omega) = c_0 q^2$ are used. Note that the bosonic propagator does not have frequency dependence since it represents the instantaneous Coulomb interaction. By integrating out high momentum degrees of freedom, we investigate the renormalization of fermion parameters.
Frequency and momentum dependences of the self energy determine anomalous dimensions of the fermion field and mass terms, respectively. At one-loop level, the fermion self energy does not have frequency dependence due to the instantaneous Coulomb interaction, so no anomalous dimension of the fermion field appears. Moreover, the frequency-independence makes one mass coupling ($1/M_0$) unchanged at one loop level and it is because frequency and this mass term have the same unity matrix structure. Note that both corrections naturally appear in higher-loop contributions.
On the other hand, non-trivial renormalizations of isotropic mass ($m$) and anisotropic mass ($M_c$) appear even at one-loop level. For example, the renormalization of isotropic mass term can be read off by calculating the third component ($\Gamma^3$) of the self energy, $$\begin{aligned}
&&\gamma^3(k) \equiv {\rm Tr}\left(\Gamma^3 \frac{\partial^2}{\partial k_x \partial k_y} \Sigma_{\psi}(k, 0) \right) / {\rm Tr} (\Gamma^3 \Gamma^3) \nonumber \\
&=& -\frac{g^2}{c_0}\frac{\partial}{\partial k_x \partial k_y} \int_q \frac{\epsilon_3(k+q)}{ \sqrt{\left(\frac{(k+q)^2}{2m}\right)^2+ \frac{m+2 M_c}{4 m M_c^2} p_c(k+q) }} \frac{1}{q^2}. \nonumber \end{aligned}$$ It can be understood as the self energy corrrection to the original Hamiltonian, $$\begin{aligned}
\epsilon_3(k) &\rightarrow& \epsilon_3(k)- \gamma^3(0) k_x k_y
% {\rm Tr}\left(\Gamma^3 \frac{\partial^2}{\partial k_x \partial k_y} \Sigma_{\psi}(k, 0) \right) / {\rm Tr} (\Gamma^3 \Gamma^3)
\nonumber \\
&=&\epsilon_3(k)\left(1+ u f_1(\frac{m}{M_c}) \log(\frac{\Lambda}{\mu}) \right).\end{aligned}$$ In the last line, the momentum integration is done for $\mu < q < \Lambda$. The correction to the isotropic mass term, $\delta (1/m) = \frac{u}{m} f_1(\frac{m}{M_c}) \log(\frac{\Lambda}{\mu})$ can be regularized in the standard way. Along the same line, one can determine the renormalization of the anisotropic mass term by calculating the fourth component ($\Gamma^4$) of the self-energy.
These considerations lead to the full RG equations as follows. $$\begin{aligned}
&&\frac{d}{d l} (\frac{1}{m}) =\left(z-2 + u f_1(\frac{m}{M_c})\right) (\frac{1}{m}) \ , \nonumber \\
&& \frac{d}{d l}(\frac{m}{M_0}) = -u f_1(\frac{m}{M_c}) \frac{m}{M_0} \ , \nonumber \\
&& \frac{d}{d l}(\frac{m}{M_c}) = -u F_2(\frac{m}{M_c}) \ , \nonumber \\
&&\frac{d }{d l} (u) = \varepsilon \, u - N_f F_1(\frac{m}{M_c}) u^2 \ . \nonumber \label{beta}\end{aligned}$$ Here the following functions of anisotropy are introduced. $$\begin{aligned}
&&f_1 (x) = \frac{2}{ \pi} \int d \Omega_{k} \frac{\hat{k}_x^2\hat{k}_y^2}{\sqrt{1+x(2+x) p_c(\hat{k})} } \ , \nonumber \\
&&f_2 (x) = \frac{1}{2 \pi} \int d \Omega_k\frac{(\hat{k}_x^2-\hat{k}_y^2)^2 }{\sqrt{1+x(2+x)p_c(\hat{k})} } \ ,\nonumber \\
&&f_3 (x) = \frac{3}{8 \pi} \int d \Omega_k \frac{x^4 k_x^2 (\hat{k}_y^2-\hat{k}_z^2)^2 }{\Big(\sqrt{1+x(2+x)p_c(\hat{k})}\Big)^5} + O(\frac{1}{x}) \ , \nonumber \\
&&F_1 (x) = 2 f_3 (x)+ \frac{1}{N_f} f_1 (x) \ , \nonumber \\
&&F_2 (x)= ( 1+x)\left(f_1 (x) -f_2 (x)\right) \ , \nonumber \end{aligned}$$ where $p_c(k) = \sum_{i=1}^3 k_i^4 -\sum_{i \neq j} k_i^2 k_j^2$ is used. The first two functions, $f_1$ and $f_2$, are from the fermion self energy and the third function, $f_3$, is from the boson self-energy calculation. The asymptotic form of $f_3$ is shown above while its complete form can be obtained by evaluating the bosonic self energy function directly. Numerical evaluations of these functions are illustrated in Figs. \[f12\] and \[f3\]. Useful numerical values are $$\begin{aligned}
&&N_f F_1(0)= \frac{30 N_f +8}{15} \ , \quad F_2(0) = 0 \ , \quad F_2^{'} (0) \sim 0.152 \ , \nonumber \\
&&f_1(0)=f_2(0)=\frac{8}{15} \ , \quad f_3(0) = 1\ , \nonumber \\
&&\lim_{x \rightarrow \infty} f_1(x) =\lim_{x \rightarrow \infty} f_2(x) =0 \ , \nonumber \\
&&\lim_{x \rightarrow \infty} f_3(x) \sim 0.888. \nonumber\end{aligned}$$
![ Numerical evaluations of the renormalization functions, $f_1(x), f_2(x)$. The solid (red) and dotted (blue) lines are for $f_1(x)$ and $f_2(x)$ ($f_1(0)=f_2(0) = 8/15$). One can show that $f_1(x)-f_2(x)$ decays as $1/x$ in the large $x$ limit. []{data-label="f12"}](Px12.pdf){width="3.0in"}
![ Numerical evaluations of the universal function, $f_3(x)$. One can see that $f_3(x)$ becomes saturated as $x$ becomes large. []{data-label="f3"}](Pf3B.pdf){width="3.0in"}
We note that there is one subtle issue in the $\varepsilon$ expansion for regularization of the loop corrections. Since the gamma matrices are defined in specific dimensions, there is some ambiguity for the choice of gamma matrices in dimensional regularization. There have been some suggestions as to how to apply the $\epsilon$ expansion to Gamma matrices. [@ab2]
In this letter, we use one of possible regularization schemes; the angle dependence is first integrated in the original dimension and then dimensional regularization is used for the integrals over the magnitude of momentum. One reason for using this scheme is our need to keep track of the cubic anisotropy, which is specifically defined in three spatial dimensions. Formally, the momentum shell integration gives $
\int_{\Lambda e^{-dl}}^{\Lambda} \frac{d^d q}{(2 \pi)^d} \frac{1}{q^4} \rightarrow \frac{d l}{8 \pi^2}
$. Note that different regularization schemes only change numerical coefficients and do not change important physics.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We have constructed a LDA+DMFT method in the framework of the iterative perturbation theory (IPT) with full LDA Hamiltonian without mapping onto the effective Wannier orbitals. We then apply this LDA+DMFT method to ferromagnetic bcc-Fe and fcc-Ni as a test of transition metal, and to antiferromagnetic NiO as an example of transition metal oxide. In Fe and Ni, the width of occupied 3d bands is narrower than those in LDA and Ni $6~{\mbox{eV}}$ satellite appears. In NiO, the resultant electronic structure is of charge-transfer insulator type and the band gap is $4.3~{\mbox{eV}}$. These results are in good agreement with the experimental XPS. The configuration mixing and dynamical correlation effects play a crucial role in these results.'
author:
- 'Oki Miura$^{1,2}$'
- 'Takeo Fujiwara$^{1,2,3}$'
title: ' Electronic structure and effects of dynamical electron correlation in ferromagnetic bcc-Fe, fcc-Ni and antiferromagnetic NiO '
---
Introduction
============
Much attention has been focused on strongly correlated electron systems, since these systems show anomalous physical properties such as various spin, charge and orbital order, metal-insulator transition and so on. The local density approximation (LDA) based on the density functional theory (DFT) is hardly applicable to these fruitful physical properties in strongly correlated electron systems. LDA+U method could not treat metallic phase near metal-insulator transition since it has been developed in order to discuss magnetic insulators. [@re:LDA+U2] Fluctuation of charge or/and spin densities plays an important role in low-energy excitations near the metal-insulator transition. To discuss realistic materials of strongly correlated electron systems, another method is needed for the “dynamical” electron correlation.
The dynamical mean field theory (DMFT) has been developed and applied to model systems of strongly correlated electrons, which led us to a unified picture of low- and high-energy excitations in anomalous metallic phase near metal-insulator transition. [@re:Rev-DMFT] DMFT is based on mapping of many electron systems in bulk onto single impurity atom embedded in effective medium, namely the single impurity problem. In this mapping procedure, the on-site dynamical correlation is included and the inter-atomic correlation is neglected. The combination of DMFT with LDA, called LDA+DMFT, has been developed in order to discuss realistic systems. [@re:Ans; @re:Lic] LDA or DFT meet their own new stage with the GW approximation, [@re:GW-Arya] which is based on the many body perturbation theory and the random phase approximation (RPA). The combination of DMFT with the GW approximation (GW+DMFT) was proposed to include both on-site and inter-atomic Coulomb interaction. [@re:GW]
To solve the mapped single impurity problem within DMFT, one can use several computational schemes such as the quantum Monte Carlo method (QMC), [@re:Rozen-QMC; @re:Geo-QMC; @re:QMC-difficult; @re:PQMC; @re:CTQMC] the iterative perturbation theory (IPT), [@re:Kaj-IPT; @re:Fuji; @re:miura; @re:MOIPT] the non-crossing approximation (NCA) [@re:NCA] and the exact diagonalization (ED). [@re:Geo-ED]
Hirsch-Fye QMC [@re:Hirsch-Fye-QMC] is widely used as a solver for the mapped single impurity problem within DMFT. However, it is not applicable in the low temperature limit. Moreover, Hirsch-Fye QMC has serious difficulty in application of multi-orbital systems with spin-flip and pair-hopping terms of the exchange interactions, since one cannot apply the Hubbard-Stratonovich transformation to these systems. [@re:QMC-difficult] To go beyond such problems of Hirsch-Fye QMC, several QMC methods have been proposed. Projective QMC [@re:PQMC] has been developed to calculate in the low temperature limit and Continuous Time QMC (CTQMC) [@re:CTQMC] to calculate multi-orbital systems with spin-flip and pair-hopping terms.
In spite of development of those QMC, the exact calculations such as QMC and ED can only be suitable for simple Hamiltonians with relatively small size because of their computational costs. Several LDA+DMFT methods with QMC adopt the projected effective Hamiltonian of Wannier-like functions with a few adopted bands for reducing computational costs. With the use of the effective Hamiltonian of Wannier-like functions in LDA+DMFT, only a few adopted bands are discussed with fixing the hybridization mixing and could not describe a possible change of hybridization due to local Coulomb interaction.
To carry out calculations in realistic materials with multiple-orbitals, approximate calculation schemes of IPT or NCA could be more suitable because of its efficient CPU-time, though IPT is not applicable to cases of large Coulomb interactions and NCA cannot yield the Fermi liquid behavior at low energies and in low temperature limit. [@re:NCA-error]
The IPT method was developed by Kajueter and Kotliar for non-degenerate orbital and the results show good agreements with the ED results. [@re:Kaj-IPT] However, it was reported that the quasiparticle peak in the IPT results tends to be sharper [@re:Rev-DMFT] and that the IPT results in La$_{1-x}$Sr$_x$TiO$_3$ does not reprduce quasiparticle peak at 1000K. [@re:Nekrasov-LaSrTiO3]
We generalized IPT for multi-orbital bands and applied this method to doubly degenerated ${\mbox{e}_{g}}$ bands and triply degenerated ${\mbox{t}_{2g}}$ bands on simple cubic and body-centered cubic lattices. [@re:Fuji; @re:miura] We also verified that this DMFT with IPT scheme is fairly applicable to arbitrary electron occupation cases with different Coulomb interaction $U$. The spectrum shows electron affinity and ionization levels with different electron configurations of different occupation numbers and the system becomes insulating state at an integer filling in sufficiently large $U$. [@re:Fuji; @re:miura] IPT for multi orbitals was also proposed by Laad, Craco and Müller-Hartmann and applied to several realistic materials. [@re:MOIPT]
In this paper, we apply this DMFT with IPT scheme to realistic materials, where we adopt full LDA Hamiltonian without reducing its size and IPT as a solver for the mapped single impurity problem. In the following, we notify this DMFT with IPT as “LDA+DMFT-IPT(1)”. The goal of the present paper is as follows:
\(i) To generalize LDA+DMFT-IPT(1) which is applicable to various realistic strongly correlated materials, both metallic and insulating, multi-atom (compound), spin-polarized and strongly hybridized cases between s, p and d-bands.
\(ii) To apply LDA+DMFT-IPT(1) to ferromagnetic bcc-Fe and fcc-Ni and to antiferromagnetic NiO. For ferromagnetic bcc-Fe and fcc-Ni, we will discuss whether LDA+DMFT-IPT(1) reproduces the accurate width of occupied 3d bands and the observed satellite of spectrum at $6~{\mbox{eV}}$ below the Fermi energy in Ni (“$6~{\mbox{eV}}$ satellite”). Application of LDA+DMFT-IPT(1) to ferromagnetic bcc-Fe and fcc-Ni is a test in transition metals since previous LDA+DMFT [@re:Fe-Ni-UJ-paper; @re:Fe-Ni-KKR+DMFT] has been applied to those metals and reproduced those physical properties well. For antiferromagnetic NiO, we will discuss whether LDA+DMFT-IPT(1) reproduces the accurate band gap and the correct description of the charge-transfer insulator. These physical properties are not well reproduced in the previous LSDA calculation.
\(iii) To verify the applicability of LDA+DMFT-IPT(1) to various realistic strongly correlated materials. Actually, the physical properties mentioned in (ii) are reproduced fairly well by LDA+DMFT-IPT(1). We will discuss the origin of drastic changes of those physical properties and show the validity of LDA+DMFT-IPT(1) by comparing with other theoretical scheme including other LDA+DMFT method [@re:Fe-Ni-UJ-paper; @re:Fe-Ni-KKR+DMFT; @re:Kunes-NiO-DMFT] and experiments. We will discuss the origin of nickel $6~{\mbox{eV}}$ satellite, which has been concluded as hole-hole scattering process, [@re:Ni-satellite] by using the spectrum of Ligand Field Theory (LFT) obtained in the procedure of LDA+DMFT-IPT(1). In bcc-Fe and fcc-Ni case, the narrowing of occupied 3d bands in LDA+DMFT-IPT(1) occurs due to the on-site dynamical electron correlation. In NiO case, the appearance of charge-transfer insulator and accurate band-gap in LDA+DMFT-IPT(1) is due to the change of hybridization between Ni-3d and O-2p bands caused by local Coulomb interaction of nickel 3d bands.
In Sec. \[sec:FORM\], the general formulation of LDA+DMFT-IPT(1) will be given. In Sec. \[sec:result-Fe-Ni\], we test LDA+DMFT-IPT(1) in systems of ferromagnetic bcc-Fe and fcc-Ni. In the energy spectra, one can see nickel $6~{\mbox{eV}}$ satellite caused by hole-hole scattering process and the narrowing of occupied 3d bands caused by on-site dynamical electron correlation. Section \[sec:result-NiO\] will be devoted to calculation of the electronic structures in antiferromagnetic NiO by LDA+DMFT-IPT(1). Local Coulomb interaction of nickel 3d bands enhances the hybridization between Ni-3d and O-2p bands and system becomes the charge-transfer insulator with accurate band gap. In both sections, we will present the energy spectra and ${\bm k}$-resolved spectrum and LFT spectra of single isolated ion. Section \[sec:CONC\] is the summary. We discuss the applicability of LDA+DMFT-IPT(1) to various realistic strongly correlated materials by comparing with other LDA+DMFT method and experiments.
Formulation {#sec:FORM}
===========
Hamiltonian and Coulomb interactions
------------------------------------
We proceed with the Hubbard-type Hamiltonian as follows : $$\begin{aligned}
\! \! \! H &=& H_{{\mbox{\scriptsize{LDA}}}}^{dc} + H_{int}~,
\label{eqn:ham}\\
\! \! \! H_{{\mbox{\scriptsize{LDA}}}}^{dc} \! \!
&=& \! \! \! \! \! \! \! \! \! \! \! \sum_{j(m{\sigma})j'(m')} \! \! \! \! \! \! \!
(h^{jj'({\mbox{\scriptsize{LDA}}})}_{mm'{\sigma}}
-\Delta h^{{\sigma}}_{dc}\delta_{jj'}\delta_{mm'}) c_{jm{\sigma}}^{\dag}c_{j'm'{\sigma}}~,
\label{eqn:HLDAdc}\\
\! \! \! H_{int} &=& \! \! {\frac{1}{2}} \! \! \sum_{j(m{\sigma})}\! \!
U_{m_1m_2:m_3m_4}c_{jm_1{\sigma}_1}^{\dag}c_{jm_2{\sigma}_2}^{\dag}c_{jm_4{\sigma}_4}c_{jm_3{\sigma}_3}~.
\nonumber \\
\label{eqn:int}\end{aligned}$$ The Hamiltonian $H$ is divided into two parts: the unperturbed part ($H^{dc}_{{\mbox{\scriptsize{LDA}}}}$) and the interaction part ($H_{int}$). $h^{jj'({\mbox{\scriptsize{LDA}}})}_{mm'{\sigma}}$ in Eq. (\[eqn:HLDAdc\]) is the full LDA Hamiltonian constructed by the Tight-Binding LMTO (TB-LMTO) method, [@re:LMTO] without projecting onto any kind of effective local orbitals. The use of original full LMTO Hamiltonian in LDA+DMFT-IPT(1) enables us to describe the change of hybridization caused by local Coulomb interaction. $\Delta h^{{\sigma}}_{dc}$ in Eq. (\[eqn:HLDAdc\]) is the double-counting term included in LDA Hamiltonian as averaged Coulomb and exchange interactions, because we need to subtract $\Delta h^{{\sigma}}_{dc}$ from $h^{jj'({\mbox{\scriptsize{LDA}}})}_{mm'{\sigma}}$ in order to construct the unperturbed part of the Hamiltonian $H^{dc}_{{\mbox{\scriptsize{LDA}}}}$. Note that an index $j$ in the summation of Eqs. (\[eqn:HLDAdc\]) and (\[eqn:int\]) runs over atomic sites, $\{m\}$ orbital indices and $\{{\sigma}\}$ spins.
$U_{m_1m_2:m_3m_4}$ in Eq. (\[eqn:int\]) is the on-site electron-electron interaction matrix and presented by using the Slater integrals $F^k$ as $$\begin{aligned}
U_{m_1m_2:m_3m_4} &=& \sum_{0\leq k\leq 2l}a_k(m_1m_2:m_3m_4)F^k ,
\label{eqn:U}\\
a_k(m_1m_2:m_3m_4) &=& {\frac{4\pi}{2k+1}} \sum_{q=-k}^{k} \langle lm_1|Y_{kq}|lm_2\rangle \nonumber\\
&\times & \langle lm_3|Y_{kq}^{*}|lm_4\rangle,
\label{eqn:ak}\end{aligned}$$ where $\{|lm\rangle\}$ are the basis set of complex spherical harmonics. Note that Eqs. (\[eqn:U\]) and (\[eqn:ak\]) are the general expression and invariant under any rotational operation. [@re:LDA+U2] For d-orbitals we only need $F^0, F^2$ and $F^4$. The averaged Coulomb and exchange parameters $U$ and $J$ for 3d-orbitals are defined in terms of the Slater integrals $F^0, F^2$ and $F^4$. Here we regard these equations as a definition of the Slater integrals $F^0, F^2$ and $F^4$ in terms of $U$ and $J$: $$\begin{aligned}
U &=& F^0, \label{eqn:F0}\\
J &=& \frac{1}{14}(F^2+F^4), \label{eqn:F2pF4} \end{aligned}$$ with a ratio of $F^2$ and $F^4$ for 3d orbitals as [@re:F2F4-ratio1] $$\begin{aligned}
F^4/F^2 &\sim& 0.625. \label{eqn:F2F4}\end{aligned}$$ Therefore, once we obtain experimentally or assume the values $U$ and $J$ of 3d transition metal elements, the values of the Slater integrals $F^0$, $F^2$ and $F^4$ can be evaluated by using Eqs. (\[eqn:F0\]), (\[eqn:F2pF4\]) and (\[eqn:F2F4\]) and then each matrix element of the on-site electron-electron interaction $U_{m_1m_2:m_3m_4}$ can be evaluated by using Eq. (\[eqn:U\]) and (\[eqn:ak\]).
As discussed above, we treat the interaction Hamiltonian Eq. (\[eqn:int\]) for full 3d-orbitals on transition metals within the framework of DMFT. Thus, the present Hamiltonian in Eq. (\[eqn:int\]) does work for the cases with partially filled $\mbox{e}_{g}$ and $\mbox{t}_{2g}$-orbitals of ferromagnetic bcc-Fe, fcc-Ni and antiferromagnetic NiO. On the other hand, the case of triply degenerated ${\mbox{t}_{2g}}$-orbitals and doubly degenerated Hubbard model were treated within the framework of DMFT in Refs. and , respectively.
Dynamical mean field theory
---------------------------
The matrices of the lattice Green’s function and the local Green’s function are presented on the basis of the non-orthogonal local base set as $$\begin{aligned}
&& [G({\bm k}, {i{\omega}_n})]^{-1}
\nonumber\\
&&\ \ \ \ =[({i{\omega}_n}+\mu)O({\bm k})-\{H^{dc}_{{\mbox{\scriptsize{LDA}}}}({\bm k})+{\Sigma}({i{\omega}_n})\}],
\label{eqn:Gkw}\\
&& G({i{\omega}_n})= {\frac{1}{V}}\int d{\bm k}~G({\bm k},{i{\omega}_n}) ,
\label{eqn:Gw}\end{aligned}$$ where $H^{dc}_{{\mbox{\scriptsize{LDA}}}}({\bm k})$, $O({\bm k})$ and $\mu$ are the Hamiltonian matrix of Eq. (\[eqn:HLDAdc\]), the overlap matrix, both in the ${\bm k}$-space, and the chemical potential, respectively. Here we neglect the ${\bm k}$-dependence of the self-energy ${\Sigma}$ within the framework of DMFT. The chemical potential $\mu$ is determined to satisfy Luttinger’s theorem. [@re:Mul] All the matrices in Eqs. (\[eqn:Gkw\]) and (\[eqn:Gw\]) have suffices $\{jm\sigma, j^\prime m^\prime \sigma^\prime\}$, where $j$, $m$ and $\sigma$ stand for atomic sites in a unit cell, orbitals and spins, respectively.
We assume that the Green’s function in the effective medium $G^0({i{\omega}_n})$ may be expressed as the self consistent equation of DMFT: $$\begin{aligned}
G^0({i{\omega}_n})^{-1}=G({i{\omega}_n})^{-1}+{\Sigma}({i{\omega}_n})+\tilde{\mu}-\mu,~
\label{eqn:Gw-local}\end{aligned}$$ where $\tilde{\mu}$ is the chemical potential in the effective medium.
Iterative Perturbation Theory {#sec:IPT}
-----------------------------
In this paper, IPT method is adopted as a solver for the mapped single impurity problem, generalized by Fujiwara [*et al.*]{} for multi-orbital bands on arbitrary electron occupation with different Coulomb $U$. [@re:Fuji; @re:miura] In the IPT scheme, the self-energy is determined with the interpolation scheme between the high frequency limit and the strong interaction limit (the atomic limit) by using the second order self-energy.
The second order self-energy is calculated as $$\begin{aligned}
\!\!\!\!\!\!\!\!\Sigma^{(2)}_{m{\sigma}}(\tau) \!\!\!\! &=& \!\!\!\! - \!\!\!\!\!\!\!\!\!\!\sum_{m_1m_2m_3{\sigma}'}\!\!\!\!\!\!\!(U_{m_1m_2:mm_3}-U_{mm_2:m_3m_1}\delta_{{\sigma}{\sigma}'})\nonumber\\
&\times& \!\!\!\! U_{m_1m_3:mm_2}G^0_{m_1{\sigma}}(\tau)G^0_{m_3{\sigma}'}(\tau)G^0_{m_2{\sigma}'}(-\tau)
\label{eqn:2nd-pertu-Sig}\\
\!\!\!\!\!\!\!\!\Sigma^{(2)}(i\omega_{n}) \!\!\!\! &=& \!\!\!\!\! \int_{0}^{\beta} \!\!\!\! d\tau
e^{i\omega_{n}\tau}\Sigma^{(2)}(\tau) \ . \label{eqn:2nd-pertu-Sig-fourier} \end{aligned}$$ By using the second order self-energy $\Sigma^{(2)}({i{\omega}_n})$, the self-energy is then expressed in a matrix form as $$\begin{aligned}
\! \! \Sigma(i\omega_n) \! = \! \Sigma^{{\mbox{\scriptsize{HF}}}} \! \! \!
+\! A\Sigma^{(2)}(i\omega_n)[1\! -\! B(i\omega_n)\Sigma^{(2)}(i\omega_n)]^{-1},
\label{eqn:sig-IPT} \end{aligned}$$ where $\Sigma^{{\mbox{\scriptsize{HF}}}}$ is the Hartree-Fock energy. The matrices $A$ and $B(i\omega_n)$ are determined by requiring the self-energy ${\Sigma}({i{\omega}_n})$ to be exact in the high-frequency limit $({i{\omega}_n}\to \infty)$, and in the atomic limit $(U\to \infty)$ [@re:Fuji] as $$\begin{aligned}
\!\!\!\!\!\! A \!\! &=& \!\! \lim_{i{\omega}_n\to\infty}\{\Sigma(i\omega_n)-\Sigma^{HF}\}\Sigma^{(2)}(i\omega_n)^{-1},
\label{eqn:Sig-IPT-A}\\
\!\!\!\!\!\! B(i{\omega}_n) \!\! &=& \!\! \Sigma^{at(2)}(i\omega_n)^{-1}-\{\Sigma^{at}(i\omega_n)-\Sigma^{HF}\}^{-1}A ,
\label{eqn:Sig-IPT-B} \end{aligned}$$ where $\Sigma^{at}(i\omega_n)$ and $\Sigma^{at(2)}(i\omega_n)$ in Eq. (\[eqn:Sig-IPT-B\]) are the self-energy and the second order self-energy in the atomic limit. These values are calculated exactly within the framework of the ligand field theory (LFT), which is discussed in Section \[sec:LF\].
The self-energy in Eq. (\[eqn:sig-IPT\]) also includes spin-flip and pair-hopping terms of the exchange interactions since the second order self-energy in Eq. (\[eqn:2nd-pertu-Sig\]) includes those terms. In addition, the IPT can be achieved within efficient CPU time since the self-energy in Eq. (\[eqn:sig-IPT\]) is obtained by only matrix calculation and hence the IPT is applicable to realistic materials with larger size by using full-LDA Hamiltonian. Form the discussion in this subsection, present IPT is much appropriate impurity solver and present IPT appropriately treats on-site dynamical correlation effects including both hole-hole and electron-electron scattering as well as electron-hole scattering.
In the following Sections (Secs. \[sec:result-Fe-Ni\] and \[sec:result-NiO\]), we assume the cubic symmetry around transition metal ions and, therefore, the self-energy can be assumed as $$\begin{aligned}
\Sigma_{jm\sigma j'm'\sigma '}(i\omega_n)=
\Sigma_{jm\sigma}(i\omega_n)\delta_{mm'}\delta_{jj'}\delta_{\sigma\sigma '}.
\label{eqn:sig-scalar} \end{aligned}$$ For a cubic symmetry, the local Green’s function becomes diagonal after the ${\bm k}$-integration.
Calculation in the atomic limit based on the ligand field theory {#sec:LF}
----------------------------------------------------------------
The Hamiltonian mapped onto an isolated atom is defined as $$\begin{aligned}
H_{atom} &=& H^{dc}_{atom}+H_{int} \ ,
\label{eqn:HAM-atom}\\
H^{dc}_{atom} &=& \sum_{(m{\sigma})}{\epsilon}_{mm'{\sigma}}c_{m{\sigma}}^{\dag}c_{m'{\sigma}} \ ,
\label{eqn:HAM-atom-dc}\\
{\epsilon}_{mm'{\sigma}}&=&h^{({\mbox{\scriptsize{LDA}}})}_{mm'{\sigma}}-\Delta h^{{\sigma}}_{dc}\delta_{mm'} \ .
\label{eqn:ep-atom}\end{aligned}$$ The first term $H^{dc}_{atom}$ in Eq. (\[eqn:HAM-atom\]) is an unperturbed one-electron part. The Coulomb interaction $H_{int}$ is the same as Eq. (\[eqn:int\]). $h^{({\mbox{\scriptsize{LDA}}})}_{mm'{\sigma}}$ is the on-site element of LDA Hamiltonian and therefore $h^{({\mbox{\scriptsize{LDA}}})}_{mm'{\sigma}}$ includes the information of bulk matrix with electron transfer, crystal field and orbital hybridization. $\Delta h^{{\sigma}}_{dc}$ is the double counting term same as in Eq. (\[eqn:HLDAdc\]). Then one electron energy ${\epsilon}_{mm'{\sigma}}$ in Eq. (\[eqn:ep-atom\]) includes the crystal field splitting. Once we obtain ${\epsilon}_{mm'{\sigma}}$ and the Slater integrals, many-electron eigenstates of Eq. (\[eqn:HAM-atom\]) should be obtained within the framework of the ligand field theory (LFT). [@re:LFT] Full configuration interaction (CI) calculation is carried out by using the basis of all the multiplet of d electrons, the number of which is $2^{10}=1024$. Using Full-CI calculation yields an exact solution for local self-energy of Eq. (\[eqn:sig-IPT\]) in the atomic limit. The Green’s function, the self-energy and the second-order self-energy in the atomic limit are then defined as $$\begin{aligned}
G^{at}_{mm'{\sigma}}(i\omega_{n})
&=& {\frac{1}{Z}}\sum_{\eta \nu }
{\frac{\langle \eta|c_{m{\sigma}}|\nu\rangle \langle \nu|c^{\dag}_{m'{\sigma}}|\eta\rangle}{i{\omega}_n +(E_{\eta}-\mu N_{\eta})-(E_{\nu}-\mu N_{\nu})}}
\nonumber\\
&\:\:& \:\: \:\: \:\: \:\: \times \{e^{-\beta( E_{\eta}-\mu N_{\eta})}+e^{-\beta(E_{\nu}-\mu N_{\nu})}\},
\label{eqn:G-atom}\\
{\Sigma}^{at}_{mm'{\sigma}}(i{\omega})
&=& i{\omega}-({\epsilon}_{mm'{\sigma}}-\mu)-[G^{at}(i{\omega})]^{-1}_{mm'{\sigma}},
\label{eqn:Sig-atom}\\
\Sigma^{at(2)}_{m{\sigma}}(i\omega)
&=& \sum_{m_1m_2m_3{\sigma}'}(U_{mm_2:m_1m_3}-U_{mm_2:m_3m_1}\delta_{{\sigma}{\sigma}'})\nonumber\\
\times && \! \! \! \! \! \! U_{m_1m_3:mm_2}\nonumber\\
\times && \! \! \! \! \! \! {\frac{ f_{m_2{\sigma}'}(1-f_{m_1{\sigma}}-f_{m_3{\sigma}'})+f_{m_1{\sigma}}f_{m_3{\sigma}'}}{i{\omega}- {\epsilon}_{m_1{\sigma}} - {\epsilon}_{m_3{\sigma}'} + {\epsilon}_{m_2{\sigma}'} +{\tilde{\mu}}}} \ ,
\label{eqn:sig2-atom}\end{aligned}$$ where $|\eta\rangle$, $E_{\eta}$, $N_{\eta}$, $f_{m{\sigma}}$ are the many-electron eigenstates in the Hilbert space of $2^{10}$ multiplets, the energy eigenvalue, the total electron number of the eigenstate $|\eta\rangle$ and the Fermi distribution function obtained from full-CI calculation within the framework of LFT.
One can assign, with a help of the atomic spectrum in LFT, the origin of peaks in LDA+DMFT spectra, e.g. the initial and final multiplets of the corresponding transitions. Generally, peak structures in LFT spectra would suggest the existence of strong scattering channels in LDA+DMFT spectra at around these particular energies.
From discussion in Sections \[sec:IPT\] and \[sec:LF\], the IPT is a appropriate approximation method and is worth being extended to the application of realistic materials in following four reasons: (1) Efficient computational cost (2) Applicability of realistic materials with larger size and more complicated hybridization by using full-LDA Hamiltonian \[This is followed by (1)\] (3) Care of spin-flip and pair-hopping terms of the exchange interactions in multi-orbital systems (4) Direct comparison of LFT spectra within the IPT method with the LDA+DMFT spectra and assignment of the origin of peaks in LDA+DMFT spectra
Energy spectrum and ${\bm k}$-resolved spectrum {#sec:Eq-Gw-Gkw}
-----------------------------------------------
The energy spectrum and the ${\bm k}$-resolved spectrum may be more physically understandable if we present the Green’s function in the orthogonal base, subject to the sum rule of the energy integration of the local Green’s function. The lattice Green’s function and the local Green’s function in the orthogonal base are defined as $$\begin{aligned}
\!\!\!\!\!\!\!\!\![{\mathscr{G}}({\bm k}, {\omega})]^{-1}\!\!\!\! &=& \!\!\! O^{-\frac{1}{2}}({\bm k})G^{-1}({\bm k}, {\omega})O^{-\frac{1}{2}}({\bm k})
\label{eqn:Gkw-ortho-1}\\
&=&\!\!\! [({\omega}+\mu)
\nonumber\\
&-&\!\!\! O^{-\frac{1}{2}}({\bm k})
\{H^{dc}_{{\mbox{\scriptsize{LDA}}}}({\bm k})+{\Sigma}({\omega})\}O^{-\frac{1}{2}}({\bm k})],
\label{eqn:Gkw-ortho-2}\\
{\mathscr{G}}({\omega})&=& {\frac{1}{V}}\int d{\bm k}~{\mathscr{G}}({\bm k},{\omega}).~
\label{eqn:Gw-ortho-1}\end{aligned}$$ The energy spectrum will be shown with the imaginary part of ${\mathscr{G}}({\omega})$. ${\bm k}$-resolved spectrum will be also presented with an imaginary part of ${\mathscr{G}}({\bm k},{\omega})$, which may be compared with the angle-resolved photoemission spectra (ARPES).
Computational Details {#sec:Comp-Detals}
---------------------
The $8 \times 8 \times 8$ ${\bm k}$-points are used in the ${\bm k}$-integration within the whole Brillouin zone. The ${\bm k}$-integration is carried out by using a generalized tetrahedron method. [@re:Fuji] The Padé approximation is adopted for analytic continuation of the Green’s function from the Matsubara frequencies ${i{\omega}_n}$ to the real ${\omega}$-axis. We adopt $2^{11}=2048$ Matsubara frequencies.
Here, all the calculation were carried out at $T=1000~{\mbox{K}}$ to reduce the number of adopted Matsubara frequency. To consider temperature dependence of magnetic order transition, one should include entropy calculation as well as total energy calculation. [@re:Biermann-Ce] Here, we do not include entropy calculation within LDA+DMFT scheme. Thus temperature $T=1000{\mbox{K}}\sim 0.1{\mbox{eV}}$ causes the spectrum broadening by at most $0.1~{\mbox{eV}}$ and that do not change the magnetic order.
We first carry out LDA calculation to obtain input LDA Hamiltonian and then carry out DMFT calculation. We adopt the converged DMFT results as output. We do not feed back the density by DMFT to the LDA calculation.
In NiO case, on-site self-energy of nickel 3d bands on each sublattice is only included and inter-atom self-energy between nickel atoms on different sublattices are neglected. Thus, our calculation is “single-site” DMFT, but not cluster-DMFT. [@re:rev-cluster-DMFT]
Ferromagnetic bcc-Fe and fcc-Ni {#sec:result-Fe-Ni}
===============================
LSDA calculation has achieved a great success on overall understanding of the ground state electronic structures of crystalline transition metals, including bcc-Fe and fcc-Ni. However, LDA overestimates the width of occupied 3d bands for both ferromagnetic bcc-Fe and fcc-Ni and could not produce the observed satellite of spectrum at $6~{\mbox{eV}}$ below the Fermi energy in Ni (“$6~{\mbox{eV}}$ satellite”). On the other hand, LDA+DMFT calculation [@re:Fe-Ni-UJ-paper; @re:Fe-Ni-KKR+DMFT] gives the width of occupied 3d bands for ferromagnetic bcc-Fe and fcc-Ni and the Ni $6~{\mbox{eV}}$ satellite, in good agreement with XPS experimental results. [@re:Fe-expt-XPS1; @re:Ni-expt-XPS1]
The aim of the calculation for ferromagnetic bcc-Fe and fcc-Ni in this section is to see whether LDA+DMFT-IPT(1) scheme can reproduce a reasonable width of occupied 3d bands and Ni $6~{\mbox{eV}}$ satellite. In fact, we will show that LDA+DMFT-IPT(1) presents the width of occupied 3d bands for bcc-Fe and fcc-Ni and the $6~{\mbox{eV}}$ satellite in fcc-Ni to be in good agreement with both experimental ARPES [@re:Fe-expt-XPS1; @re:Ni-expt-XPS1] and also with previous LDA+DMFT results. [@re:Fe-Ni-UJ-paper; @re:Fe-Ni-KKR+DMFT]
Hamiltonian and $U$, $J$ values
-------------------------------
Lattice constants, atomic sphere radii in the TB-LMTO method and the averaged values of Coulomb and exchange interactions, $U$ and $J$, for ferromagnetic bcc-Fe and fcc-Ni to construct the Hamiltonian are summarized in Table \[tab:parameters\_and\_results\]. The values of $U$ and $J$ are obtained with the constraint-LDA calculation. [@re:Fe-Ni-UJ-paper] The basis set of the Hamiltonian consists of full valence bands of 4s, 4p and 3d orbitals, totally nine orbitals.
lattice constant $a$ (Å) atomic sphere radius $s_0$ (Å) $U$ (eV) $J$ (eV)
-------- ---------------------------- -------------------------------------- ------------------------- ------------------------------
fcc-Ni 3.5233 1.3768 3.0[@re:Fe-Ni-UJ-paper] 0.9[@re:Fe-Ni-UJ-paper]
bcc-Fe 2.8708 1.4128 2.0[@re:Fe-Ni-UJ-paper] 0.9[@re:Fe-Ni-UJ-paper]
NiO 4.1948 1.2318 (Ni), 1.0699 (O), 0.8882 (ES) 7.0[@re:NiO-fujimori2] 0.9[@re:Anisimov-NiO-LSDA+U]
LSDA GW[@re:yamasaki-GW-TM] LDA+DMFT LDA+DMFT-IPT(1) Expt.
------------------------------- -------- ------------------------ ---------- -------------------------- ------- ------------------------------
$\mu_{\rm spin}(\mu_B)$ bcc-Fe 2.17 2.31 2.28[@re:Fe-Ni-KKR+DMFT] 2.16 2.13[@re:Fe-Ni-expt-mag-mom]
fcc-Ni 0.47 0.55 0.57[@re:Fe-Ni-KKR+DMFT] 0.47 0.57[@re:Fe-Ni-expt-mag-mom]
$W_{\rm d, occ}({\mbox{eV}})$ bcc-Fe 3.7 3.4 3.4 3.3[@re:Fe-expt-XPS1]
fcc-Ni 4.5 3.3 3.5 3.2[@re:Ni-expt-XPS1]
LSDA LSDA+U[@re:Anisimov-NiO-LSDA+U] GW[@re:NiO-GW-Ferdi; @re:NiO-GW-Nohara] QPscGW[@re:NiO-GW-Kotani] LDA+DMFT[@re:Kunes-NiO-DMFT] LDA+DMFT-IPT(1) Expt.
----------------------------- ------ --------------------------------- ----------------------------------------- --------------------------- ------------------------------ ----------------- ----------------------------
$E_{\rm gap}({\mbox{eV}}) $ 0.2 3.7 0.2 4.8 4.3 4.3 4.3[@re:NiO-expt]
$\mu_{\rm spin}(\mu_B)$ 1.00 1.59 1.00 1.72 1.85 1.00 1.64[@re:NiO-expt-mag-mom]
Energy Spectrum
---------------
Figure \[fig:dos-Fe-Ni\] shows the energy spectrum $-\mbox{Im} {\mathscr{G}}({\omega})$ obtained by LDA and LDA+DMFT-IPT(1) with the atomic spectrum by LFT. The occupied 3d-band becomes narrower in LDA+DMFT-IPT(1) than in LDA, narrower by $0.3~{\mbox{eV}}$ and $1.0~{\mbox{eV}}$ for bcc-Fe and fcc-Ni, respectively. Particularly, the spectra near the Fermi energy, in the region $-1.5~{\mbox{eV}}<{\omega}<0.0~{\mbox{eV}}$ for both bcc-Fe and fcc-Ni in LDA+DMFT-IPT(1), becomes narrower than that in the low energy region. These results are attributed to strong renormalization of the quasiparticle caused by the on-site dynamical electron correlation within the framework of DMFT. The precise width of occupied 3d bands is discussed in Sec. \[sec:Gkw-Fe-Ni-result\].
In addition, the satellite appears in LDA+DMFT-IPT(1) spectrum of fcc-Ni at $6~{\mbox{eV}}$ below the Fermi energy, in good agreement with experimental XPS results. [@re:Ni-expt-XPS1; @re:Ni-expt-spinXPS] Moreover, this satellite structure is strongly spin-dependent, much enhanced in the spectrum of the majority spin. These effects mainly come from the multiplet scattering of ${\mbox{d}}^8 {\rightarrow}{\mbox{d}}^7$, which is assigned by spin-dependent peak of LFT spectrum of ${\mbox{d}}^8 {\rightarrow}{\mbox{d}}^7$ at around $-5~{\mbox{eV}}$ in Fig. \[fig:dos-Fe-Ni\]-(2c). The satellite is due to hole-hole scattering process, [@re:Ni-satellite] and can appear only when one treats whole on-site dynamical correlation effects. It should be noted that $6~{\mbox{eV}}$ satellite is not observed in GW approximation (GWA) [@re:yamasaki-GW-TM] since GWA includes the dynamical correlation only within RPA (electron-hole excitations) and thus not the electron-electron and hole-hole scattering processes.
Let us focus on the atomic spectrum obtained by LFT in Fig. \[fig:dos-Fe-Ni\]. The atomic spectra are separately shown in Figs. \[fig:dos-Fe-Ni\]-(1c) (2c) for ionization spectra and Figs. \[fig:dos-Fe-Ni\]-(1d) (2d) for affinity spectra. The initial state is a mixture of ${\mbox{d}}^6$, ${\mbox{d}}^7$ in Fe and ${\mbox{d}}^8$, ${\mbox{d}}^9$ in Ni, since the total occupation numbers are non-integers, $n^{{\scriptsize{\mbox{Fe}}}}_{\rm tot}=6.829$ and $n^{{\scriptsize{\mbox{Ni}}}}_{\rm tot}=8.799$ from the results of bulk systems. The excited state is then a mixture of ${\mbox{d}}^5$, ${\mbox{d}}^6$, ${\mbox{d}}^7$, ${\mbox{d}}^8$ in Fe and ${\mbox{d}}^7$, ${\mbox{d}}^8$, ${\mbox{d}}^9$, ${\mbox{d}}^{10}$ in Ni. Small components appear in ionization spectra above $E_F$ and in affinity spectra below $E_F$. Moreover, characteristic atomic spectra at around $-0.5~{\mbox{eV}}$ in Fe and Ni originate from both ionization and affinity process; ${\mbox{d}}^6 {\rightarrow}{\mbox{d}}^7$ and ${\mbox{d}}^7 {\rightarrow}{\mbox{d}}^6$ in Fe in Figs. \[fig:dos-Fe-Ni\]-(1c)(1d) and ${\mbox{d}}^8 {\rightarrow}{\mbox{d}}^9$ and ${\mbox{d}}^9 {\rightarrow}{\mbox{d}}^8$ in Ni in Figs. \[fig:dos-Fe-Ni\]-(2c)(2d). These effects comes from the fact that higher energy occupied states are not fully occupied.
Characteristic structure in both atomic ionization and affinity spectra originate from various multiplet scattering; ${\mbox{d}}^7 {\rightarrow}{\mbox{d}}^6$ and ${\mbox{d}}^6 {\rightarrow}{\mbox{d}}^5$ at around $-4.0~{\mbox{eV}}$ and $-6.0~{\mbox{eV}}$ and ${\mbox{d}}^7 {\rightarrow}{\mbox{d}}^8$ and ${\mbox{d}}^6 {\rightarrow}{\mbox{d}}^7$ at around $1.5~{\mbox{eV}}$ in Fe of Figs. \[fig:dos-Fe-Ni\]-(1c)(1d) and ${\mbox{d}}^9 {\rightarrow}{\mbox{d}}^8$ and ${\mbox{d}}^8 {\rightarrow}{\mbox{d}}^7$ at around $-5.5~{\mbox{eV}}$ in Ni of Figs. \[fig:dos-Fe-Ni\]-(2c)(2d). This is due to the multiplet scattering caused by the exchange interaction $J$. The primary splitting of multiplets comes from the Coulomb interaction $U$, for example the transition spectrum of ${\mbox{d}}^8 {\rightarrow}{\mbox{d}}^7$ should locate in the lower energy region than that of ${\mbox{d}}^9 {\rightarrow}{\mbox{d}}^8$ of Ni spectra. Then additional multiplet splitting is caused by $J$. In LFT spectra in $-6.0~{\mbox{eV}}< {\omega}< 1.5~{\mbox{eV}}$, the energetic order of the spectral positions does not follow the above first simple rule and this fact implies that the scattering process by $J$ changes the multiplet spectra drastically. We conclude that the structures in LDA+DMFT-IPT(1) spectra is broadened and smoothed due to the multiplet scattering of ${\mbox{d}}^7 {\rightarrow}{\mbox{d}}^6$ and ${\mbox{d}}^6 {\rightarrow}{\mbox{d}}^5$ at $-6.0~{\mbox{eV}}< {\omega}< -2.0~{\mbox{eV}}$ and that of ${\mbox{d}}^7 {\rightarrow}{\mbox{d}}^8$ and ${\mbox{d}}^6 {\rightarrow}{\mbox{d}}^7$ at around $1.5~{\mbox{eV}}$ in bcc-Fe and that of ${\mbox{d}}^9 {\rightarrow}{\mbox{d}}^8$ and ${\mbox{d}}^8 {\rightarrow}{\mbox{d}}^7$ at around $-6.0~{\mbox{eV}}< {\omega}<-2.0~{\mbox{eV}}$ in fcc-Ni.
${\bm k}$-resolved spectrum and magnetic moment {#sec:Gkw-Fe-Ni-result}
-----------------------------------------------
Figure \[fig:Gkw-Fe-Ni\] shows the ${\bm k}$-resolved spectrum $- \mbox{Im}\,{\mathscr{G}}({\bm k},{\omega})$ by LDA+DMFT-IPT(1) with the LDA energy bands. LDA results overestimate the width of occupied 3d valence bands, which is defined to be the energy difference between the Fermi energy and the energy eigenvalue at P-point (Fe) or L-point (Ni), in comparison with experimental results. The width of occupied 3d valence bands and the magnetic moment in LDA+DMFT-IPT(1) are shown in Table \[tab:parameters\_and\_results\] in comparison with those by LDA and the experiments. [@re:Fe-Ni-expt-mag-mom; @re:Fe-expt-XPS1; @re:Ni-expt-XPS1] The valence band width of LDA+DMFT-IPT(1) is in reasonable agreement with experiments.
One can observe flat branches in bcc-Fe at around $-1.5~{\mbox{eV}}$ of majority spin and that at around $-2~{\mbox{eV}}$ of minority spin. These flat bands correspond to the local multiplet excitations of ${\mbox{d}}^7 {\rightarrow}{\mbox{d}}^6$. One can also observe flat branches in fcc-Ni at $-6~{\mbox{eV}}$ of both majority and minority spins. These flat bands at $-6~{\mbox{eV}}$ are due to the local multiplet excitations of ${\mbox{d}}^8 {\rightarrow}{\mbox{d}}^7$, where the intensity of the ${\bm k}$-resolved spectrum of majority spin is larger than that of minority spin. This intensity difference causes the strong spin dependence of $-6~{\mbox{eV}}$ satellite in the energy spectrum in Fig. \[fig:dos-Fe-Ni\]-(2b).
The ${\bm k}$-resolved spectrum in bcc-Fe is more diffusive than that in fcc-Ni at around $-3.0~{\mbox{eV}}\sim 0.0~{\mbox{eV}}$ in spite of smaller $U$ for bcc-Fe than fcc-Ni. This comes from that the scattering process by $J$ compared with $U$ is stronger in bcc-Fe than in fcc-Ni since $J/U$ in bcc-Fe is larger than that in fcc-Ni.
The magnetic moment $\mu_{\rm spin}$ of both bcc-Fe and fcc-Ni is almost the same as LDA result. $\mu_{\rm spin}$ of bcc-Fe in LDA+DMFT-IPT(1) result is also in good agreement with experiment, while that of fcc-Ni is slightly smaller than experimental result. This is an artifact due to insufficient number of ${\bm k}$-points to integrate the lattice Green’s function $G({\bm k}, {\omega})$ to obtain the local Green’s function $G({\omega})$ by using generalized tetrahedron method. The adopted value of the total number of the ${\bm k}$-points in LDA+DMFT-IPT(1) is much smaller than that in TB-LMTO method and the discrepancy of $\mu_{\rm spin}$ for fcc-Ni may be improved by increasing the total number of the ${\bm k}$-points in LDA+DMFT-IPT(1).
Comparison with previous LDA+DMFT results {#sec:Compare_LDA+DMFT_for_FeNi}
-----------------------------------------
Here, we compare the results for ferromagnetic bcc-Fe and fcc-Ni obtained by LDA+DMFT-IPT(1) with other previous LDA+DMFT [@re:Fe-Ni-UJ-paper; @re:Fe-Ni-KKR+DMFT]. Lichtenstein [*et al.*]{} [@re:Fe-Ni-UJ-paper] has used the LDA+DMFT with QMC as an impurity solver. Minár [*et al.*]{} [@re:Fe-Ni-KKR+DMFT] has used the KKR+DMFT with perturbative SPTF (spin-polarized T-matrix+FLEX) as an impurity solver. The energy spectra in LDA+DMFT-IPT(1) show a good agreement with those LDA+DMFT calculations. The existence of Ni 6 eV satellite is also very similar to those LDA+DMFT calculations, presumably much better coincident with the position of the observed spectra. The magnetic moment of bcc-Fe is in good agreement with those LDA+DMFT results. Slightly smaller value of the magnetic moment of fcc-Ni in LDA+DMFT-IPT(1) than previous LDA+DMFT is not due to the use of perturbative IPT approach as an impurity solver but due to insufficient number of ${\bm k}$-points mentioned above. Thus, we can conclude that LDA+DMFT-IPT(1) reproduces reasonable results for ferromagnetic bcc-Fe and fcc-Ni and that LDA+DMFT-IPT(1) is applicable to realistic metallic materials in strongly correlated electron systems as well as other previous LDA+DMFT methods. [@re:Fe-Ni-UJ-paper; @re:Fe-Ni-KKR+DMFT]
Antiferromagnetic NiO {#sec:result-NiO}
=====================
NiO is a type-II antiferromagnetic insulator with Néel temperature $523~{\mbox{K}}$. The experimentally observed band gap is $4.3~{\mbox{eV}}$.[@re:NiO-expt] Resonance photoemission experiments [@re:NiO-kyoumei] show that electronic structure of NiO should be of the charge-transfer type and the low energy satellite mainly consists of nickel 3d bands.
Various theoretical methods have been applied to NiO. The band gap $E_{\rm gap}$ and the magnetic moment $\mu_{\rm spin}$ are summarized in Table \[tab:NiO-past-result\]. LSDA calculation shows that NiO is Mott-Hubbard type insulator with a small band gap of $0.2~{\mbox{eV}}$ for antiferromagnetic phase, [@re:Terakura-NiO-LSDA] in which oxygen 2p bands are located at lower energy region than the occupied nickel 3d bands.
The LSDA+U method was applied and the resultant band gap is almost good agreement with experiment and a system becomes charge-transfer insulator. [@re:Anisimov-NiO-LSDA+U] However, bonding states of nickel ${\mbox{e}_{g}}$ bands are observed at around $8~{\mbox{eV}}$ below the Fermi energy and this causes less components of the occupied main peak and too much components of the occupied satellite peak, compared with experimental XPS spectrum. These problems mainly come from the static potential correction with orbital dependence in LSDA+U and this implies that one should include dynamical correlation effects.
GW approximation was applied and the resultant band gap still remains very small (0.2). [@re:NiO-GW-Ferdi; @re:NiO-GW-Nohara] Quasiparticle self-consistent GW (QPscGW) approximation [@re:NiO-GW-Kotani] was also applied and a system becomes charge-transfer insulator. However, the resultant band gap is overestimated (4.8) and the bonding states of nickel ${\mbox{e}_{g}}$ bands are observed at around $6~{\mbox{eV}}$ below the Fermi energy. These fact implies that the dynamical electron correlation should play a more crucial role in NiO than treated by RPA in GW approximation.
LDA+DMFT was applied to paramagnetic NiO [@re:Kunes-NiO-DMFT; @re:Ren-NiO-DMFT] and the resultant band gap is in good agreement with experimentally observed XPS result. [@re:NiO-expt] and the system becomes charge-transfer insulator. However, the top of valence bands mainly consists of nickel 3d bands, while the cluster-model CI calculation [@re:NiO-fujimori2; @re:NiO-vanElp] and oxygen x-ray absorption of Li$_x$Ni$_{1-x}$O [@re:NiO-Kuiper-expt] show that the top of valence bands is mainly based on oxygen 2p bands. This difference comes from that these LDA+DMFT calculations use projected effective Hamiltonian constructed by Wannier-like functions with fixing the hybridization mixing. Moreover, no calculation has been carried out for antiferromagnetic NiO by LDA+DMFT.
The target of LDA+DMFT calculation is to get electronic structures of antiferromagnetic NiO with (i) the accurate band gap, (ii) the correct description of the charge-transfer type insulator and (iii) the occupied main peak corresponding to oxygen 2p states and the occupied satellite peak corresponding to nickel 3d states.
Hamiltonian, the Coulomb and exchange interactions $U$ and $J$
--------------------------------------------------------------
The structure of NiO is not of the dense packing and we put empty atom spheres in vacant region of the lattice in LMTO formalism. The lattice constants, atomic sphere radii of each atom and the averaged values of Coulomb and exchange interactions, $U$ and $J$, are summarized in Table \[tab:parameters\_and\_results\]. The total number of atom spheres in an antiferromagnetic unit cell is eight; two Ni, two O and four empty atoms (ES). We use the full LDA Hamiltonian of the LMTO formalism and adopted muffin-tin orbitals are 4s, 4p and 3d in Ni and 2s, 2p in O and 1s, 1p in ES and, thus, the total number of basis is 42. $U$ refers to the experimental value of $U=7.0~{\mbox{eV}}$ [@re:NiO-fujimori2] and $J$ to the constraint-LDA calculation of $J=0.9.~{\mbox{eV}}$ [@re:Anisimov-NiO-LSDA+U] The difference of $U$-values in fcc-Ni and NiO is due to the difference of screening mechanism in metals and insulators.
The value of $U$ obtained by constraint LDA is $U=8.0~\mbox{eV}$, [@re:Anisimov-NiO-LSDA+U] which is slightly larger than experimental value. Within the framework of constraint LDA, all the screening channels are switch off. When we evaluate suitable value of $U$ for 3d-orbitals required in the framework of LDA+DMFT, only the screening channels associated with 3d electrons should be switch off and those associated with 4s and 4p electrons should remain. In this sense, a suitable value of $U$ is slightly smaller than that for constraint LDA. Thus, the adopted value of $U=7.0~\mbox{eV}$ in the present paper is reasonable within the framework of LDA+DMFT.
Energy Spectrum
---------------
Figure \[fig:dos-NiO\] shows the energy spectrum $-\mbox{Im} {\mathscr{G}}({\omega})$ of antiferromagnetic NiO by LDA+DMFT-IPT(1) with experimental XPS spectrum. [@re:NiO-expt] Experimental XPS spectrum mainly consists of three parts: a main peak at $4~{\mbox{eV}}$, a main peak at $-3.5~{\mbox{eV}}< {\omega}< -0.5~{\mbox{eV}}$ and a satellite peak at $-10.0~{\mbox{eV}}< {\omega}< -6.0~{\mbox{eV}}$ as in Fig. \[fig:dos-NiO\]-(4). With the cluster model CI calculation, [@re:NiO-fujimori2] these three structures are assigned to ${\mbox{d}}^9$, ${\mbox{d}}^8{\mbox{\underline{L}}}$ and ${\mbox{d}}^7$ final state, respectively, where ${\mbox{\underline{L}}}$ is a ligand hole created in oxygen 2p orbitals.
The energy spectrum by LDA+DMFT-IPT(1) reproduces these three structures fairly well, the positions of two main peaks are in good agreement with experimentally observed XPS result, but the position of satellite peak at $-10.0~{\mbox{eV}}< {\omega}< -6.0~{\mbox{eV}}$ shifts slightly upward in comparison with experimental results. [@re:NiO-expt] Appreciable component of oxygen 2p bands appears just at the top of the valence bands. The main peak of the occupied states at $-1.5$ eV and the satellite at $-6.0~{\rm eV}\sim -7.5~{\rm eV}$ originate from t$_{2g}$ orbitals with minority spin of Ni and e$_g$+t$_{2g}$ orbitals with majority spin of Ni, respectively. The main peak of conduction bands comes from the Ni-e$_g$ orbitals with minority spin. The spectrum of LDA+DMFT-IPT(1) in Fig. \[fig:dos-NiO\] shows that the electronic structure is of the charge-transfer insulator type, while the electronic structure in LDA is of Mott-Hubbard type. In fact, the hybridization mixing between oxygen 2p bands and nickel 3d bands is much enhanced in LDA+DMFT-IPT(1) in comparison with that in LDA, though the Coulomb matrix elements between Ni-3d and O-2p bands and among O-2p bands are not included.
On-site Coulomb interaction between nickel ${\mbox{e}_{g}}$ bands makes the occupied nickel ${\mbox{e}_{g}}$ bands shift to the lower energy side and the unoccupied nickel ${\mbox{e}_{g}}$ bands to the higher energy side. Due to the shift of the occupied nickel ${\mbox{e}_{g}}$ bands to lower energy side, the hybridization between Ni-${\mbox{e}_{g}}$ and Ni-${\mbox{t}_{2g}}$ increases and the occupied Ni-${\mbox{t}_{2g}}$ bands shift to the lower energy side. Since the occupied Ni-${\mbox{e}_{g}}$ and ${\mbox{t}_{2g}}$ bands shift to the lower energy side, the hybridization between Ni-3d and O-2p bands increases and whole bands are broadened in the region $-8.5~{\mbox{eV}}<{\omega}<-0.5~{\mbox{eV}}$. These effects enhance the hybridization mixing and change the character of NiO to be of the charge-transfer insulator type. The energy spectrum and the electronic structure by LDA+DMFT-IPT(1) is in good agreement with XPS experiments.
Atomic spectrum for single nickel ion obtained by LFT is shown in Fig. \[fig:dos-NiO\]-(3). Initial state is ${^3\mbox{A}_{2g}}({\mbox{d}}^8)$ of the electron configuration ${{\mbox{t}^{3}_{2g}}}_{{\uparrow}}{{\mbox{t}^{3}_{2g}}}_{{\downarrow}}{{\mbox{e}^{2}_{g}}}_{{\uparrow}}$. The excited states of atomic ionization spectra a1 and a2 are ${^4\mbox{T}_{1g}}({\mbox{d}}^7)$, b1 and b2 are ${^2\mbox{E}_{g}}({\mbox{d}}^7)$ and c1 and c2 are ${^2\mbox{T}_{1g}}({\mbox{d}}^7)$, respectively. This assignment is perfectly consistent with previous LFT calculation. [@re:NiO-fujimori2]
In the atomic calculation of LFT, a single nickel ion is considered and inter-atomic electron transfer is not allowed. Therefore, present atomic calculation does not include explicitly the ${\mbox{d}}^8{\mbox{\underline{L}}}$ final state. The initial state in the LFT spectra is fixed in ${\mbox{d}}^8$ since the total electron number of Ni ion is eight in NiO. Therefore, the unoccupied and occupied spectra in Fig. \[fig:dos-NiO\]-(3) originate from the transition of ${\mbox{d}}^8 {\rightarrow}{\mbox{d}}^9$ and ${\mbox{d}}^8 {\rightarrow}{\mbox{d}}^7$, respectively. Though the ${\mbox{d}}^8{\mbox{\underline{L}}}$ final state is not included in LFT, of Fig. \[fig:dos-NiO\]-(3), the occupied main peak in the spectrum of LDA+DMFT-IPT(1) is in good agreement with experimental XPS result. In fact, the initial states in LDA+DMFT-IPT(1) is a mixture of ${\mbox{d}}^8$ and ${\mbox{d}}^9{\mbox{\underline{L}}}$, since the electron occupation numbers of nickel 3d bands and oxygen 2p bands in LDA+DMFT-IPT(1) are 8.17 and 5.14, respectively. Therefore, we can assign, in the spectrum of LDA+DMFT-IPT(1), the unoccupied main peak and occupied satellite peak to be the ${\mbox{d}}^9,~{\mbox{d}}^7$ final state, respectively. The occupied main peak is also assigned to be the ${\mbox{d}}^8{\mbox{\underline{L}}}$ final state.
The splitting between ${\mbox{d}}^8{\mbox{\underline{L}}}$ and ${\mbox{d}}^7$ final state configurations is small and, in Fig. \[fig:dos-NiO\]-(4), the occupied satellite peak in LDA+DMFT-IPT(1) appear at higher energy region ($ -6 \sim -7$ eV) than that in experimental XPS result ($-8 \sim -9$ eV). This may be due to the fact that we do not include the ${\mbox{d}}^8{\mbox{\underline{L}}}$ final state in LFT calculation. The cluster model CI-calculation instead of LFT calculation would give more precise position of occupied satellite peak.
Unoccupied ${\mbox{e}_{g}}$ bands of nickel majority spin locates at slightly higher energy region than that of minority spin and oxygen 2p bands in Fig. \[fig:dos-NiO\]-(1). It would not be the case, if we include the inter-atomic component of the self-energy or cluster CI-calculation.
The band gap $E_{\rm gap}$ and the magnetic moment $\mu_{\rm spin}$ are summarized in Table \[tab:NiO-past-result\], The band gap in LDA+DMFT-IPT(1) is in fairly good agreement with experimental result. On the other hand, the calculated magnetic moment is almost the same as the calculated result by LDA and much smaller than the observed one. The magnetic moment of antiferromagnetic NiO comes from the electron occupation of nickel ${\mbox{e}_{g}}$ bands since nickel ${\mbox{t}_{2g}}$ bands are almost fully occupied. The present result of smaller value of the magnetic moment in LDA+DMFT-IPT(1) than the experimental result comes from that unoccupied ${\mbox{e}_{g}}$ bands of nickel majority spin has some intensity. In the present calculation, we neglect the inter-atomic and inter-spin components of the self-energy and the discrepancy may be attributed to this approximation, which should be left as a future study.
${\bm k}$-resolved Spectrum
---------------------------
The ${\bm k}$-resolved spectrum $- \mbox{Im}\,{\mathscr{G}}({\bm k},{\omega})$ by LDA+DMFT-IPT(1) with both the energy bands by LDA (dashed lines) and the angle-resolved photoemission spectrum (dots) is shown in Fig. \[fig:Gkw-NiO\]. LDA band at $0.0~{\mbox{eV}}< {\omega}< 1.5~{\mbox{eV}}$ is shifted to the LDA+DMFT-IPT(1) band at $3.0~{\mbox{eV}}< {\omega}< 4.0~{\mbox{eV}}$. These spectra correspond to unoccupied nickel ${\mbox{e}_{g}}$ bands in Fig. \[fig:dos-NiO\]-(1). These unoccupied nickel 3d bands in LDA+DMFT-IPT(1) may be more localized than that in LDA.
In LDA band structure, the occupied bands mainly consists of two parts: (i) $-2.5~{\mbox{eV}}< {\omega}< 0.0~{\mbox{eV}}$ bands structure mainly based on Ni-3d bands and (ii) $-8.5~{\mbox{eV}}< {\omega}< -3.5~{\mbox{eV}}$ bands structure mainly based on O-2p bands. In LDA+DMFT-IPT(1) results, (i) and (ii) have been shifted to lower and higher energy side, respectively. This shift of these results causes the increase of hybridization between Ni-3d and O-2p bands. Especially, strong hybridization occurs at L and $\Gamma$ points.
We observe a broad and flat diffusive structure at $-5~{\mbox{eV}}$ in LDA+DMFT-IPT(1) spectrum, which corresponds to occupied satellite peak of Ni-${\mbox{t}_{2g}}$ bands and also Ni-${\mbox{t}_{2g}}$ state in LFT in Figs. \[fig:dos-NiO\]-(2)(3). Therefore, we conclude that Ni-${\mbox{t}_{2g}}$ bands may be more localized due to the strong scattering by $U$.
One can see flat branches at $-5.5eV< \omega <-5.0~{\mbox{eV}}$and at $ -6.5~{\mbox{eV}}< \omega < -6.0~{\mbox{eV}}$ in ${\bm k}$-resolved spectrum of LDA+DMFT-IPT(1). This corresponds to occupied satellite peak of Ni-${\mbox{e}_{g}}$ bands in Fig. \[fig:dos-NiO\]-(1). This flat band implies that the occupied satellite peak of Ni-${\mbox{e}_{g}}$ bands is not due to the band structure but to the multiplet scattering, particularly ${\mbox{d}}^8 {\rightarrow}{\mbox{d}}^7$ assigned by the atomic calculation of LFT.
The ${\bm k}$-resolved spectrum at $-5.5~{\mbox{eV}}< {\omega}< -2.5~{\mbox{eV}}$ of LDA+DMFT-IPT(1) is in good agreement with experimental ARPES result [@re:NiO-ARPES] along $\Gamma$ to X point, which is due to the inclusion of dynamical correlation within DMFT, which causes strong hybridization between Ni-3d and O-2p bands.
Finally, we should mention that much enhanced intensity profile can be seen in $- \mbox{Im}\,G({\bm k},{\omega})$ but $- \mbox{Im}\,{\mathscr{G}}({\bm k},{\omega})$ becomes more diffusive due to overlap integrals.
Comparison with other calculations and experiments {#sec:Compare_with_others_NiO}
--------------------------------------------------
Here, we compare the results for antiferromagnetic NiO obtained by LDA+DMFT-IPT(1) with other previous calculations. LDA+DMFT-IPT(1) reproduces $4.3~{\mbox{eV}}$ band gap and the characteristics of the charge-transfer type insulator. Moreover, the results obtained by LDA+DMFT-IPT(1) shows no Ni-${\mbox{e}_{g}}$ bonding states below the Fermi energy, which is observed in LSDA+U [@re:Anisimov-NiO-LSDA+U] and QPscGW results. [@re:NiO-GW-Kotani] LDA+DMFT-IPT(1) also reproduces the top of valence bands with mainly based on oxygen 2p bands, while application of the LDA+DMFT method using Hamiltonian constructed by Wannier-like function to paramagnetic NiO [@re:Kunes-NiO-DMFT; @re:Ren-NiO-DMFT] shows that the top of valence bands is mainly based on nickel ${\mbox{e}_{g}}$ bands. Those effects obtained by LDA+DMFT-IPT(1) are in good agreement with experimental XPS results. Those drastic change in LDA+DMFT-IPT(1) is due to enhancement of hybridization between Ni-3d and O-2p bands caused by on-site Coulomb interaction of nickel 3d bands within DMFT scheme.
The assignment of final states of the main and satellite occupied peaks and the unoccupied main obtained in LDA+DMFT spectrum is consistent with that in the cluster-model CI calculation. [@re:NiO-fujimori2] This comes from the use of LFT spectrum within the IPT method and hence the assignment of the origin of peaks in LDA+DMFT spectra by using LFT spectra is a great advantage of the IPT method, compared with other impurity solvers.
However we still have several problems, slightly higher energy positions of the occupied satellite peak and the unoccupied main peak and smaller value of the magnetic moment. From discussion in the previous two subsections, these are mainly due to neglect of the inter-atomic and inter-spin components of the self-energy within DMFT scheme. Particularly, dynamical correlation of nickel 3d bands between two different sublattices and that of nickel 3d and oxygen 2p bands are not included since DMFT neglects all the inter-atomic dynamical correlation.
To improve those problems in antiferromagnetic NiO, one may adopt cluster DMFT [@re:rev-cluster-DMFT], which is based on mapping of many electron systems in bulk onto single cluster impurity problem with including the intra-cluster dynamical Coulomb interaction and neglecting the inter-cluster Coulomb interaction. In LDA+DMFT-IPT(1), the extension of LDA+DMFT to LDA+cluster DMFT is consistent with the extension of LFT of single nickel isolated ion to the cluster model CI-calculation of NiO$_6$ cluster similar to Ref. and . The use of the cluster model CI-calculation will reproduce better positioning of occupied satellite peak and direct assignment of occupied main peak with ${\mbox{d}}^8{\mbox{\underline{L}}}$ final state.
From above discussion, we can conclude that LDA+DMFT-IPT(1) reproduces reasonable results for antiferromagnetic NiO within the DMFT scheme. and that LDA+DMFT-IPT(1) is applicable to realistic compound cases and insulating cases as well as LDA+DMFT with nonperturbative impurity solver.
Conclusion {#sec:CONC}
==========
In summary, we have proposed LDA+DMFT-IPT(1) where we use the full LDA Hamiltonian and IPT as an impurity solver. We then applied LDA+DMFT-IPT(1) to ferromagnetic bcc-Fe, fcc-Ni and antiferromagnetic NiO.
For bcc-Fe and fcc-Ni case, on-site dynamical correlation effect causes the narrowing of the width of occupied 3d bands. The multiplet scattering effect of ${\mbox{d}}^8 {\rightarrow}{\mbox{d}}^7$ assigned by the LFT spectrum causes the spin dependent $6~{\mbox{eV}}$ satellite for fcc-Ni. The energy spectra and magnetic moment in LDA+DMFT-IPT(1) show a good agreement with previous LDA+DMFT calculations [@re:Fe-Ni-UJ-paper; @re:Fe-Ni-KKR+DMFT].
For antiferromagnetic NiO case, on-site Coulomb interaction of nickel 3d bands enhances the hybridization between Ni-3d and O-2p bands and this causes the band gap of $4.3~{\mbox{eV}}$, the charge-transfer type insulator, and the top of valence bands with mainly oxygen 2p bands. Those effects are all in good agreement with experiments. The drastic change of hybridization between nickel 3d and oxygen 2p bands is due to the use of full-LDA Hamiltonian and on-site dynamical electron correlation of nickel site within DMFT. The successful application of LDA+DMFT-IPT(1) to antiferromagnetic NiO implies that LDA+DMFT-IPT(1) is fairly applicable to larger compound cases with more complicated hybridization between more than two atoms.
In addition, the assignment of characteristic peaks of NiO in LDA+DMFT spectrum is consistent with that in the cluster-model CI calculation. [@re:NiO-fujimori2] Thus, the assignment of the origin of peaks by using LFT spectra is a great advantage of the IPT method in understanding the existence of strong scattering channels in LDA+DMFT spectra.
Several remained problems in antiferromagnetic NiO, positions of the occupied satellite and the unoccupied main peak, would be solved by including the inter-atomic dynamical correlation. Thus, we can conclude that LDA+DMFT-IPT(1) is applicable to various realistic materials, such as both metallic and insulating cases, multi-atom (compound) cases, spin-polarized cases and strongly hybridized cases between s, p and d-bands.
[99]{}
A.I. Lichtenstein, V.I. Anisimov and J. Zaanen, Phys. Rev. B [**52**]{} R5467 (1995).
A. Georges, G. Kotliar, W. Krauth and M.J. Rozenberg, Rev. Mod. Phys. [**68**]{} 13 (1996).
V.I. Anisimov, A.I. Poteryaev, M.A. Korotin, A.O. Anokhin and G. Kotliar, J. Phys: Condens. Matter [**9**]{} 7539 (1997).
A.I. Lichtenstein and M.I. Katsnelson, Phys. Rev. B [**57**]{} 6884 (1998).
L. Hedin, Phys. Rev. [**139**]{} A796 (1965).
S. Biermann, F. Aryasetiawan and A. Georges, Phys. Rev. Lett. [**90**]{} 086402 (2003).
M.J. Rozenberg, X. Y. Zhang and G. Kotliar, Phys. Rev. Lett. [**69**]{} 1236 (1992).
A. Georges and G. Krauth, [Phys. Rev. Lett.]{} [**69**]{} 1240 (1992).
K. Held and D. Vollhardt, [Eur. Phys. J. B]{} [**5**]{} 473 (1998).
M. Feldbacher, K. Held, and F. F. Assaad, [Phys. Rev. Lett.]{} [**93**]{} 136405 (2003).
A. N. Rubtsov, V. V. Savkin and A. I. Lichtenstein, [Phys. Rev. B]{} [**72**]{} 035122 (2005).\
P. Werner, A. Comanac, L. de’ Medici, M. Troyer and A. J. Millis, [Phys. Rev. Lett.]{} [**97**]{} 076405 (2006).\
S. Sakai, R. Arita, K. Held and H. Aoki, [Phys. Rev. B]{} [**74**]{}, 155102 (2006).
H. Kajueter and G. Kotliar, [Phys. Rev. Lett.]{} [**77**]{} 131 (1996).
T. Fujiwara, S. Yamamoto and Y. Ishii, [J. Phys. Soc. Jpn.]{} [**72**]{} 777 (2003).
O. Miura and T. Fujiwara, J. Phys. Soc. Jpn. [**75**]{} 014703 (2006).
M. S. Laad, L. Craco, and E. Müller-Hartmann, [Phys. Rev. Lett.]{} [**91**]{} 156402 (2003).\
M. S. Laad, L. Craco, and E. Müller-Hartmann, [Phys. Rev. B]{} [**73**]{}, 045109 (2006).
Th. Pruschke and N. Grewe, [Z. Phys. B]{} [**74**]{} 439 (1989).
A. Georges and G. Krauth, [Phys. Rev. B]{} [**48**]{} 7167 (1993).
J.E. Hirsch, Phys. Rev. B [**31**]{}, 4403 (1985);\
J. E. Hirsch and R.M. Fye, Phys. Rev. Lett. [**56**]{}, 2521 (1986).
Y. Kuramoto and H. Kojima, [Z. Phys. B]{} [**57**]{} 95 (1984).
I.A. Nekrasov, K. Held, N. Blümer, A.I. Poteryaev, V.I. Anisimov, and D. Vollhardt, [Eur. Phys. J. B]{} [**18**]{}, 55 (2000).
A.I. Lichtenstein, M.I. Katsnelson and G. Kotliar, Phys. Rev. Lett. [**87**]{} 067205 (2001).
J. Minár, L. Chioncel, A. Perlov, H. Ebert, M.I. Katsnelson, and A.I. Lichtenstein, Phys. Rev. B [**72**]{}, 045125 (2005).
J. Kuneš, V.I. Anisimov, A.V. Lukoyanov, and D. Vollhardt, Phys. Rev. B [**75**]{}, 165115 (2007).
A. Liebsch, Phys. Rev. B [**23**]{} 5203 (1981);\
D.R. Penn, Phys. Rev. Lett. [**42**]{}, 921 (1979).
O.K. Andersen and O. Jepsen, Phys Rev. Lett. [**53**]{} 2571 (1984);\
O.K. Andersen, O. Jepsen, and D. Glötzel, in [*Canonical Description of the Band Structures of Metals*]{}, Proceedings of the International School of Physics, “Enrico Fermi”, Course LXXXIX, Varenna, 1985, edited by F. Bassani, F. Fumi, and M.P. Tosi (North-Holland, Amsterdam, 1985), p. 59.
F.M.F. de Groot, J.C. Fuggle, B.T. Thole, and G.A. Sawatzky, [Phys. Rev. B]{} [**42**]{} 5459 (1990).
T. Pruschke and R. Bulla, [Eur. Phys. J. B]{} [**44**]{}, 217 (2005).
E. Müller-Hartmann, [Z. Phys. B-Condens. Matter]{} [**76**]{} 211 (1989).
Y. Tanabe and S. Sugano, J. Phys. Soc. Jpn. [**9**]{} 753,766 (1954). B. Amadon, S. Biermann, A. Georges, and F. Aryasetiawan, Phys. Rev. Lett. [**96**]{}, 066402 (2006).
T. Maier, M. Jarrell, T. Pruschke and M. H. Hettler, Rev. Mod. Phys. [**77**]{}, 1027 (2005).
H. Höchst, A. Goldmann and S. Hüfner, Z. Phys. B 24, 245 (1976) ;\
R.E. Kirby, E. Kisker, F.K. King and E.L. Garwin, Solid State Commun. [**56**]{}, 425 (1985) ;\
H. Höchst, S. Hüfner and A. Goldmann , Phys. Lett. 57A, 265 (1976).
H. Höchst, S. Hüfner and A. Goldmann, Z. Phys. B 26, 133 (1977) ;\
M. Lädeniemi, E. Ojala and M. Okoochi, Phys. Status Solidi B [**108**]{}, K61 (1981).
A. Fujimori and F. Minami, Phys. Rev. B [**30**]{}, 957 (1984).
V.I. Anisimov, J. Zaanen and O.K. Andersen, Phys. Rev. B [**44**]{}, 943 (1991).
T. Kinoshita, T. Ikoma, A. Kakizaki, T. Ishii J. Fujii, H. Fukutani K. Shimada, A. Fujimori, T. Okane and S. Sato, Phys. Rev. B [**47**]{}, 47, 6787(1993).
A. Yamasaki and T. Fujiwara, J. Phys. Soc. Jpn. [**72**]{} 607 (2003).
M. B. Stearns, [*in Magnetic Properties in Metals*]{}, edited by H.P.J. Wijn, Landolt-Börnstein, New Series, Vol. III, Pt. 19a (Springer, Berlin, 1984) ;\
D. Bonnenberg, K. A. Hempel and H. P. J. Wijn, [*ibid*]{}.
G.A. Sawatzky and J.W. Allen, Phys. Rev. B [**53**]{} 2339 (1984).
M.R. Thuler, R.L. Benbow, and Z. Hurych, Phys. Rev. B [**27**]{}, 2082 (1983).
F. Aryasetiawan and O. Gunnarsson, Phys. Rev. Lett. [**74**]{}, 3221 (1995).
Y. Nohara and T. Fujiwara, private communication.
S. V. Faleev, M. van Schilfgaarde and T. Kotani, Phys. Rev. Lett. [**93**]{}, 126406 (2004).
A.K. Cheetham and D. A. O. Hope, Phys. Rev. B [**27**]{}, 6964 (1983).
K. Terakura, T. Oguchi, A.R. Williams and J. Kuebler, Phys. Rev. B [**30**]{}, 4734 (1984).
X. Ren, I. Leonov, G. Keller, M. Kollar, I. Nekrasov, and D. Vollhardt, Phys. Rev. B [**74**]{}, 195114 (2006).
J. van Elp, H. Eskes, P. Kuiper, and G. A. Sawatzky, Phys. Rev. B [**45**]{}, 1612 (1992).
P. Kuiper, G. Kruizinga, J. Ghijsen, G. A. Sawatzky, and H. Verweij, Phys. Rev. Lett. [**62**]{}, 221 (1989).
Z.-X. Shen, R.S. List, D.S. Dessau, B.O. Wells, O. Jepsen, A.J. Arko, R. Barttlet, C.K. Shih, F. Parmigiani, J.C. Huang and P.A.P. Lindberg, Phys. Rev. B [**44**]{}, 3604 (1991).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A well-established observation in nuclear physics is that in neutron-rich spherical nuclei the distribution of neutrons extends farther than the distribution of protons. In this work, we scrutinize the influence of this so called neutron-skin effect on the centrality dependence of high-$p_{\rm T}$ direct-photon and charged-hadron production. We find that due to the estimated spatial dependence of the nuclear parton distribution functions, it will be demanding to unambiguously expose the neutron-skin effect with direct photons. However, when taking a ratio between the cross sections for negatively and positively charged high-$p_{\rm T}$ hadrons, even centrality-dependent nuclear-PDF effects cancel, making this observable a better handle on the neutron skin. Up to 10 % effects can be expected for the most peripheral collisions in the measurable region.'
author:
- Ilkka Helenius
- Hannu Paukkunen
- 'Kari J. Eskola'
date: 'Received: date / Accepted: date'
title: 'Neutron-skin effect in direct-photon and charged-hadron production in Pb+Pb collisions at the LHC'
---
[2]{}(12.35,0.3) LU TP 16-33\
June 2016
Introduction
============
In ultra-relativistic heavy-ion collisions the concept of centrality plays an important role in phenomena such as the jet energy loss [@Abelev:2013kqa; @Chatrchyan:2012nia; @Aad:2014bxa] or the systematics of azimuthal anisotropies [@Aad:2014eoa; @Chatrchyan:2012xq]. Experimentally, the centrality of a collision is usually defined according to the amount of energy seen in a specific part of the detector, typically at large pseudorapidities [@Aad:2014eoa; @Chatrchyan:2011sx; @Abelev:2013qoq]: the more energy observed, the more central the collision. The theoretical centrality categorizations are based on Glauber models [@Miller:2007ri], in which the centrality is related to impact parameter (optical Glauber) or to the number of nucleon–nucleon collisions (Monte-Carlo Glauber). While there is no direct, unambiguous relation between the experimental and theoretical prescriptions, it is yet generally accepted that a correspondence exists in collisions of two heavy nuclei. In nucleon–nucleus collisions, however, the same experimental procedure has led to rather unexpected results [@Chatrchyan:2014hqa; @ATLAS:2014cpa; @Adare:2015gla] and it is now commonly believed that such a centrality classification induces a non-trivial bias on the hard process whose centrality dependence was to be measured [@Martinez-Garcia:2014ada; @Bzdak:2014rca; @Alvioli:2014eda; @Perepelitsa:2014yta; @Armesto:2015kwa].
The Glauber models take the nuclear density distribution as an input and it is typically assumed to be identical for protons and neutrons. However, the measurements at lower energies indicate that the tail of the neutron density distribution extends farther than that of the proton density [@Tarbert:2013jze; @Tsang:2012se; @Zenihiro:2010zz]. While this so-called neutron-skin (NS) effect [@Horowitz:2000xj] should not have a great importance in the centrality classification itself, it leads to a growth of the relative number of neutrons at high impact parameters and thereby influences the observables sensitive to electroweak effects in peripheral (large impact parameter) collisions of two heavy nuclei. The impact of the NS effect to $W^{\pm}$ production in Pb+Pb and p+Pb collisions at the LHC was studied in Ref. [@Paukkunen:2015bwa].
In this work, we extend the study of Ref. [@Paukkunen:2015bwa] to direct-photon and charged-hadron production at high transverse momenta ($p_{\rm T}$) in Pb+Pb collisions at the LHC. The goal is to study whether the NS effect has a measurable impact on these observables and to quantify at which centralities and kinematics (transverse momentum, rapidity) the effect would be most pronounced. Our hope is that, later on, the NS effect could help to calibrate the centrality classification in collisions involving heavy ions.
Centrality-dependent hard-process cross section
===============================================
Centrality classification is done here using the optical Glauber model as in Refs. [@Paukkunen:2015bwa; @Helenius:2012wd]. For the nuclear density distribution we use the two-parameter Fermi (2pF) distribution, $$\rho^{A}(\mathbf{r}) = \rho_0^{A}/(1 + \mathrm{e}^{(|\mathbf{r}| - d_{A})/a_{A}}),$$ where $d_{A}$ describes the radius of the nucleus and $a_{A}$ the thickness of the nuclear surface (skin) in nucleus with a mass number $A$. To account for the NS effect the nuclear density is written as $
\rho^{A}(\mathbf{r}) = \rho^{\mathrm{p},A}(\mathbf{r}) + \rho^{\mathrm{n},A}(\mathbf{r})
$ where now the parameters of the 2pF distribution are different for protons and neutrons. Here we use the parameters from Ref. [@Tarbert:2013jze], $d_\mathrm{n, Pb} = 6.70 \pm 0.03~\mathrm{fm}$ and $a_\mathrm{n, Pb} = 0.55 \pm 0.03~\mathrm{fm}$, for neutrons and $d_\mathrm{p, Pb} = 6.680~\mathrm{fm}$ and $a_\mathrm{p, Pb} = 0.447~\mathrm{fm}$ for protons.[^1]
The hard-process cross section in an $A+B$ collision for a given centrality class $\mathcal{C}_k$ corresponding to an impact parameter interval $b_k\leq b < b_{k+1}$ (where $b=|\mathbf{b}|$) can be calculated from $$\begin{aligned}
\mathrm{d}\sigma_{AB}^{\mathrm{hard}}(\mathcal{C}_k) =& \,2\pi \int_{b_{k}}^{b_{k+1}} \mathrm{d}b\, b \int \mathrm{d}^2\mathbf{s} \sum_{i,j} T_A^i(\mathbf{s_1})\,T_B^j(\mathbf{s_2}) \notag \\ &\mathrm{d}\sigma_{ij}^{\mathrm{hard}}(A,B,\mathbf{s_1}, \mathbf{s_2}),
\label{eq:masterformula}\end{aligned}$$ where the nuclear thickness functions $T_A^i(\mathbf{s})$ are obtained by integrating the density over the longitudinal (i.e. beam) direction, $$T^i_{A}({\bf s}) \equiv \int_{-\infty}^{\infty} \mathrm{d}z \rho^{{\rm i},A}({\bf r}),$$ and $\mathbf{s_{1,2}} = \mathbf{s} \pm \mathbf{b}/2$ are defined according to Fig. 20 of Ref. [@Helenius:2012wd]. The indices $i$ and $j$ run over combinations $(i,j) = (\mathrm{p},\mathrm{p}),\, (\mathrm{p},\mathrm{n}),$\
$\, (\mathrm{n},\mathrm{p})\,\text{and}\,(\mathrm{n},\mathrm{n}) $. The impact-parameter intervals required in Eq. (\[eq:masterformula\]) correspond to the fractions of the total inelastic cross section $\sigma^{\rm inel}_{AB}(\sqrt{s})$, obtained as in Refs. [@Paukkunen:2015bwa; @Helenius:2012wd] by $$\sigma^{\rm inel}_{AB}(\sqrt{s}) = \int_{-\infty}^{\infty} \mathrm{d}^2{\bf b} \left[ 1 - \mathrm{e}^{-T_{AB}({\bf b}) \, \sigma^{\rm inel}(\sqrt{s})}\right],$$ where $$T_{AB}({\bf b}) \equiv \int_{-\infty}^{\infty} \mathrm{d}^2{\bf s}
\sum_{i,j}T_A^{j}(\mathbf{s_1})\, T_B^{i}(\mathbf{s_2}).
%\left[T_A^{\rm p}({\bf s}_1) + T_A^{\rm n}({\bf s}_1) \right] \left[T_B^{\rm p}({\bf s}_2) + T_B^{\rm n}({\bf s}_2) \right],$$ We take $\sigma^{\rm inel}({\sqrt{s}=5~\mathrm{TeV}})=70~\mathrm {mb}$ [@Antchev:2013iaa]. The spatial dependence of the hard-process cross section $\mathrm{d}\sigma_{ij}^{\mathrm{hard}}$ arises here from the spatial dependence of nPDFs, $$\begin{gathered}
\mathrm{d}\sigma_{ij}^{\mathrm{hard}}(A,B,\mathbf{s_1}, \mathbf{s_2}) = \sum_{k,l}f_k^{i/A}(x_1, Q^2, \mathbf{s_1})\\ \otimes f_l^{j/B}(x_2, Q^2, \mathbf{s_2}) \otimes\mathrm{d}\hat{\sigma}^{kl\rightarrow {\rm observable}}, \label{eq:hxsec}\end{gathered}$$ where $\mathrm{d}\hat{\sigma}^{kl\rightarrow {\rm observable}}$ are perturbative coefficient functions and $k$ and $l$ are parton flavour indices. The nPDFs appearing in Eq. (\[eq:hxsec\]) above are defined as $$f_k^{i/A}(x, Q^2, \mathbf{s}) = r_k^{i/A}(x,Q^2,\mathbf{s})f_k^{i}(x, Q^2),$$ where $f_k^{i}(x, Q^2)$ is the free nucleon PDF (here CT10NLO [@Lai:2010vv]) and $r_k^{i/A}(x,Q^2,\mathbf{s})$ the nuclear modification which depends on the transverse position of the nucleon inside the nucleus.[^2] Here we use <span style="font-variant:small-caps;">EPS09s</span> nuclear modifications from Ref. [@Helenius:2012wd] in which $$r_k^{{\rm p}/A}(x,Q^2,\mathbf{s}) = 1 + \sum_{j=1}^4 c^{j}_{k}(x,Q^2) \left[T_A^{\rm p}(\mathbf{s}) + T_A^{\rm n}(\mathbf{s})\right]^j,
\label{eq:eps09s}$$ where the coefficients $c^{j}_{k}(x,Q^2)$ are obtained by analyzing the $A$-dependence of the <span style="font-variant:small-caps;">EPS09</span> [@Eskola:2009uj] nPDFs. The neutron PDFs $f_k^{{\rm n}/A}(x, Q^2, \mathbf{s})$ are obtained from the proton PDFs $f_k^{{\rm p}/A}(x, Q^2, \mathbf{s})$ by the isospin symmetry. By combining all, Eq. (\[eq:masterformula\]) factorizes into purely geometric and purely momentum-dependent parts which can be evaluated separately thereby reducing the dimensions of the required numerical integrations. We use the <span style="font-variant:small-caps;">Incnlo</span> program [@Aurenche:1987fs; @Aversa:1988vb; @Aurenche:1998gv; @Aurenche:1999nz] to calculate the momentum-dependent parts at next-to-leading order in perturbative QCD.
Results
=======
Direct-photon production
------------------------
{width="49.00000%"} {width="49.00000%"}
Direct photons are produced either in the hard process or by the fragmentation of high-$p_{\rm T}$ partons from the hard process. To obtain the latter contribution we convolute the partonic spectra with the BFG (set II) parton-to-photon fragmentation functions (FFs) [@Bourhis:1997yu]. Since the photon coupling is stronger in the case of up-type quarks than with the down-type quarks, the production rate of direct photons is larger in p+p collisions than in n+n collisions. This leads to a lower per-nucleon rate of direct photons in heavy-ion collisions than in p+p collisions due to the presence of neutrons. This is often referred to as the isospin effect and it becomes important at large values of $x$ where the valence quarks dominate. Furthermore, since the relative fraction of neutrons grows towards the edge of nucleus due to the NS effect, an additional suppression of direct photons in peripheral collisions is expected.
{width="49.00000%"} {width="49.00000%"} {width="49.00000%"} {width="49.00000%"}
A canonical way to quantify the nuclear effects is to compute the nuclear modification factor, defined in a given centrality class $\mathcal{C}_k$ of a Pb+Pb collision as $$R_{\mathrm{PbPb}}^{\gamma}(\mathcal{C}_k) = \frac{1}{T_{\mathrm{PbPb}}(\mathcal{C}_k)}\frac{\mathrm{d}\sigma_{\mathrm{PbPb}}^{\gamma}(\mathcal{C}_k)}{\mathrm{d}p_{\mathrm{T}}\mathrm{d}\eta} \bigg/ \frac{\mathrm{d}\sigma_{\mathrm{pp}}^{\gamma}}{\mathrm{d}p_{\mathrm{T}}\mathrm{d}\eta},$$ where the normalization is related to the amount of interacting nuclear matter, $$T_{\mathrm{PbPb}}(\mathcal{C}_k) = 2 \pi \int_{b_k}^{b_{k+1}}\mathrm{d}b\,b\int\mathrm{d}^2\mathbf{s}\, \sum_{i,j}T_{\rm Pb}^{j}(\mathbf{s_1})\, T_{\rm Pb}^{i}(\mathbf{s_2}),$$ where the impact parameters $b_k$ and $b_{k+1}$ define the centrality class $\mathcal{C}_k$ as in Eq. (\[eq:masterformula\]). For the cross-section calculations we have set the renormalization, factorization and fragmentation scales to photon $p_{\rm T}$. The uncertainties related to the scale ambiguities are not considered here in more detail since they largely cancel out in the ratio, especially at large values of $p_{\rm T}$ relevant here [@Helenius:2013bya]. The isolation criterion, often used by experiments to suppress secondary photons from hadronic decays, is not applied here since the effect to $R_{\mathrm{PbPb}}^{\gamma}$ is negligible at the very high values of $p_{\mathrm{T}}$ considered here.[^3]
Figure \[fig:RPbPbgamma\] shows $R_{\mathrm{PbPb}}^{\gamma}$ at mid-rapidity for two centrality classes, 70–80 % and 90–100 % with and without the NS effect, compared also to the minimum-bias (0–100 %, MB) result and to the isospin effect. The uncertainties considered here are the <span style="font-variant:small-caps;">EPS09s</span> uncertainty (light colour band) and the one related to the uncertainty of neutrons 2pF parameters (dark colour band), obtained by evaluating $R_{\mathrm{PbPb}}^{\gamma}$ with the quoted parameter variations, and adding the differences to the central prediction in quadrature.
The different $p_{\rm T}$ regions are sensitive to different nPDF effects. First, comparing the MB result to the result without nPDF effects (only isospin), at $p_{\rm T} < 30~\mathrm{GeV/c}$ some suppression due to shadowing is observed which then turns into an enhancement due to anti-shadowing. At $p_{\rm T} > 300~\mathrm{GeV/c}$ a suppression due to the EMC effect is observed. The spatial dependence of the nPDFs always decreases the nuclear effects towards more peripheral collisions whereas the NS effect generates additional suppression with increasing $p_{\rm T}$. Therefore, at high values of $p_{\rm T}$ where the impact of NS is more pronounced, these two effects pull towards opposite directions thereby “softening” the aggregate centrality dependence.
As the nuclear modifications of the PDFs gradually disappear with increasing peripherality, also the uncertainty becomes smaller for more peripheral events. However, even in the 90–100 % bin the nPDF uncertainty is of the same order as the NS effect which further complicates the separation of different effects. The non-zero nPDF effects even at the most peripheral bin are due to the power series ansatz in <span style="font-variant:small-caps;">EPS09s</span>, see Eq. (\[eq:eps09s\]), which, by construction, gives zero nuclear modifications only when $b\rightarrow \infty$. The uncertainty from the neutron 2pF parametrization turns out to be rather small as the contribution of the n+n channel is inferior e.g. to the contribution of the p+p channel and thus the variations in the neutron density are not that important. Since the fraction of neutrons (and therefore the n+n channel contribution) grows towards more peripheral collisions, also the uncertainty grows accordingly.
The centrality dependence can also be studied using the central-to-peripheral ratio $R_{\rm CP}$ defined as $$R_{\rm CP} = \frac{T_{\mathrm{PbPb}}(P)}{ T_{\mathrm{PbPb}}(C)}\frac{\mathrm{d}\sigma_{\mathrm{PbPb}}^{\gamma}(C)}{\mathrm{d}p_{\mathrm{T}}\mathrm{d}\eta} \bigg/ \frac{\mathrm{d}\sigma_{\mathrm{PbPb}}^{\gamma}(P)}{\mathrm{d}p_{\mathrm{T}}\mathrm{d}\eta}.$$ The advantage is that there is no need for a separate p+p baseline measurement and also that some uncertainties are expected to cancel out. Here, we have used the bin 0–10 % as the central result and compared it to the 70–80 % and 90–100 % bins. The results are shown in Fig. \[fig:RCPgamma\] again with and without the NS effect. Since the peripheral bins are now in the denominator, the NS effect increases the ratio and therefore decreases the centrality dependence at high-$p_{\rm T}$ region ($R_{\rm CP}$ closer to unity). The nPDF originating uncertainties are now larger with the most peripheral bin red (90–100 %) because the uncertainties in the central bin (similar to the MB uncertainty in Fig. \[fig:RPbPbgamma\]) do not cancel here as effectively as with the less peripheral bin (70–80 %). Even though the interpretation of this observable is easier, the NS effect is still of the same order as the nPDF uncertainties.
At forward/backward rapidities (the lower panels in Fig. \[fig:RCPgamma\]) the nPDF uncertainties are smaller. This is because here the dominant contribution comes from $q$+g initial state where the gluon is at shadowing region with only mild uncertainty (at high factorization scale), and the quark is also a well-constrained high-$x$ valence quark. The modifications, however, are quite small and since there is an additional uncertainty due to modelling of the spatial dependence of the nPDFs, it is difficult to unambiguously study the NS effect with this observable. The most accurate centrality-dependent measurement for photons in Pb+Pb comes from the ATLAS collaboration [@Aad:2015lcb]. However, their most peripheral bin 40–80 % is still too central, and also the experimental uncertainties are large, to see any effects of NS.
Charged-hadron production
-------------------------
{width="49.00000%"} {width="49.00000%"} {width="49.00000%"} {width="49.00000%"}
An observable in which the nPDF effects should cancel out very efficiently but yet be sensitive to the NS effect, is the ratio between negatively and positively charged hadrons, $$\frac{h^-}{h^+}(\mathcal{C}_k) = \frac{\mathrm{d}\sigma_{\mathrm{PbPb}}^{h^-}(\mathcal{C}_k)}{\mathrm{d}p_{\mathrm{T}}\mathrm{d}\eta} \bigg/\frac{\mathrm{d}\sigma_{\mathrm{PbPb}}^{h^+}(\mathcal{C}_k)}{\mathrm{d}p_{\mathrm{T}}\mathrm{d}\eta}.
\label{eq:Rhposneg}$$ Since the relative number of neutron-involving (p+n, n+p, n+n) collisions is higher in peripheral than in central collisions, the increased d-quark contribution produces less positively charged hadrons and more negatively charged hadrons during the fragmentation. Here we do not consider any additional final-state effects that may affect the hadron production even though a significant suppression for the production of high-$p_{\mathrm{T}}$ hadrons has been observed [@Abelev:2014laa; @Adam:2015kca; @Khachatryan:2016odn] in all centralities. Indeed, the measurements in Refs. [@Abelev:2014laa; @Adam:2015kca] show that the suppression at high $p_{\rm T}$ ($p_{\mathrm{T}} \gtrsim 10~\mathrm{GeV/c}$) is very similar for all light charged hadrons (pions, kaons, protons) and, consequently, the particle ratios $(\mathrm{K}^++\mathrm{K}^-)/(\pi^++\pi^-)$ and $({\rm p}+\overline{{\rm p}})/(\pi^++\pi^-)$ are the same in p+p and Pb+Pb collisions. This motivates us to conjecture that final-state effects would have only a relatively small influence on the ratio of Eq. (\[eq:Rhposneg\]). Moreover, at the very high-$p_{\mathrm{T}}$ region ($p_{\rm T} \gg 100~\mathrm{GeV/c}$) where the current measurements are still statistically limited [@Khachatryan:2016odn], the suppression effect in peripheral bins may be even negligible.
The cross section for hadron production is calculated by convoluting the partonic spectra with non-perturbative parton-to-hadron FFs. We consider three options, <span style="font-variant:small-caps;">dss</span> [@deFlorian:2007ekg], <span style="font-variant:small-caps;">kretzer</span> [@Kretzer:2000yf] and <span style="font-variant:small-caps;">akk08</span> [@Albino:2008fy]. To better understand the variations seen using different FFs, the $h^-/h^+$ ratios in p+p and n+n collisions at $\sqrt{s} = 5.0~\mathrm{TeV}$ are shown in Fig. \[fig:RnegPosHadr\]. The first observation is that with <span style="font-variant:small-caps;">akk08</span> FFs the ratio in p+p actually turns negative at high-$p_{\mathrm{T}}$, caused by the cross section for $h^-$ becoming negative. This clearly unphysical result implies that the considered kinematic region is out of the validity region of <span style="font-variant:small-caps;">akk08</span>. The results using <span style="font-variant:small-caps;">dss</span> and <span style="font-variant:small-caps;">kretzer</span> are not that different in p+p collisions but for n+n collisions almost a factor of two difference at the very highest values of $p_{\rm T}$ is observed. These differences between the FF analyses generate some further theoretical uncertainty for the considered observable. Turning this around, a measurement of $h^-/h^+$ in p+Pb or Pb+Pb collisions would clearly provide additional constraints for future FF analyses (modulo the possible final-state effects in Pb+Pb).
The $h^-/h^+$ ratios in Pb+Pb collisions at $\sqrt{s_{\mathrm{NN}}} = 5.0~\mathrm{TeV}$ for centrality classes 70–80 % and 90–100 % are shown in the upper right panel of Fig. \[fig:RnegPosHadr\] together with the MB result with <span style="font-variant:small-caps;">dss</span> and <span style="font-variant:small-caps;">kretzer</span> FFs. The nPDF effects, including the centrality dependence and the uncertainties, are found to be negligible as expected. The uncertainty in the 2pF parametrization is negligible for the MB case but increases towards more peripheral collisions. The uncertainties are larger than in the case of direct-photon production as the cross section for $h^-$ now gets a large contribution from the n+n channel and thus carries more sensitivity to the parameter uncertainties in the neutron density. More importantly, the centrality dependence from the NS effect is clearly visible in this observable. However, the different FFs still yield rather different results but normalizing the ratio with the MB result, the FF dependence largely cancels out. This is demonstrated in the lower panels of Fig. \[fig:RnegPosHadr\], where the ratios in 70–80 % and 90–100 % classes are normalized with the 0–100 % result for $\eta = 0$ and $|\eta| = 2$. Some FF dependence persists with $|\eta| = 2$ but it is still smaller or of the same order than the uncertainty in 2pF parametrization. Also the nPDFs yield a few-percent uncertainty for the observable.
To estimate the achievable experimental precision for the $h^-/h^+$ ratios discussed above, we multiply the cross sections from Eq. (\[eq:masterformula\]) by the nominal Pb-Pb nucleon–nucleon luminosity of $\mathcal{L}_{\rm nn} = 1~\mathrm{nb}^{-1}$ (also with $\mathcal{L}_{\rm nn}=10~\mathrm{nb}^{-1}$ which would correspond to the luminosity targeted after the LHC Long Shutdown 2 [@highlumi]) [^4]. From the resulting number of events $N$ we compute the relative statistical uncertainty by $1/\sqrt{N}$. To have better statistics, we consider here the rapidity bin $1 < |\eta| < 3$ (excluding the mid-rapidity to have a larger effect from the neutron skin) and suitably wide $p_{\rm T}$ intervals. The results are shown in Fig. \[fig:stats\] where the total statistical uncertainty follows from combining the statistical uncertainties for $h^+$ and $h^-$ quadratically. As can be seen from Fig. \[fig:stats\], the realistically measurable region is $p_{\rm T} < 200~\mathrm{GeV/c}$ for the 70–80 % bin and $p_{\rm T} < 100~\mathrm{GeV/c}$ for the 90–100 % bin. We would expect that systematic uncertainties are small in these ratios like they are e.g. in the case of W charge asymmetry.
![The expected statistical precision for $1~\mathrm{nb}^{-1}$ (light vertical bars) and for $10~\mathrm{nb}^{-1}$ (darker vertical bars) nucleon–nucleon luminosity in 70–80 % (green) and 90–100 % (blue) centrality classes. The shaded boxes show the 2pF uncertainty as in the lower panels of Fig. \[fig:RnegPosHadr\]. The Kretzer FFs have been used.[]{data-label="fig:stats"}](Fig4.pdf){width="\linewidth"}
Summary and Outlook
===================
We have studied the impact of the NS effect to direct-photon and charged-hadron production in Pb+Pb collisions at the LHC. In the case of photon production the NS effect has a 5–10 % impact on $R_{\rm PbPb}^{\gamma}$ though the uncertainties in the nPDFs and their spatial dependence are of the same order or even larger than the expected effect. With $R_{\rm CP}^{\gamma}$ some of the nPDF uncertainties cancel out making the NS effect more transparent. Also, going to larger rapidities decreases the nPDF uncertainties, but still the smallness of the NS effect and the ambiguities due to the centrality dependence of the nPDFs makes the direct-photon production a challenging observable to study the NS effect.
A more promising observable is the ratio between negatively and positively charged high-$p_{\mathrm {T}}$ hadrons, for which we find up to 10 % effects in the statistically relevant $p_{\mathrm {T}}$ region. In this case, the spatial dependence of the nPDFs cancel out very efficiently and, in general, the NS effect has a more pronounced impact than in the case of direct photons. The downsides here are the sensitivity to the applied fragmentation functions and, towards smaller $p_{\mathrm {T}}$, possible final-state modifications due to the produced strongly interacting medium. The first one can be cured by normalizing the ratio with the minimum bias result, but for a more detailed study of the latter, further modelling would be required. However, as discussed, there are indications that the final-state effects may largely disappear when considering particle ratios like the ones we have done here and, after all, the disparity between the amount of initial-state up and down quarks should strongly correlate with the balance of produced negatively and positively charged hadrons, irrespectively of the exact way the produced hard partons hadronize.
We hope that in near future the NS effect could provide an additional handle to control the centrality classification and help to bridge the theoretical and experimental centrality definitions. As a further prospect, we plan to study the NS effect in the future high-luminosity lepton-ion colliders.
I. H. has been supported by the MCnetITN FP7 Marie Curie Initial Training Network, Contract PITN-GA-2012-315877 and has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (Grant Agreement No 668679). This research was supported by the European Research Council Grant HotLHC ERC-2011-StG-279579 and by Xunta de Galicia (Conselleria de Educacion)–H. P. is part of the Strategic Unit AGRUP2015/11.
[99]{}
B. Abelev [et al.]{} (ALICE Collaboration), JHEP [**1403**]{} (2014) 013 doi:10.1007/JHEP03(2014)013 \[arXiv:1311.0633 \[nucl-ex\]\].
S. Chatrchyan [et al.]{} (CMS Collaboration), Phys. Lett. B [**712**]{} (2012) 176 doi:10.1016/j.physletb.2012.04.058 \[arXiv:1202.5022 \[nucl-ex\]\].
G. Aad [et al.]{} (ATLAS Collaboration), Phys. Rev. Lett. [**114**]{} (2015) no.7, 072302 doi:10.1103/PhysRevLett.114.072302 \[arXiv:1411.2357 \[hep-ex\]\].
G. Aad [et al.]{} (ATLAS Collaboration), Eur. Phys. J. C [**74**]{} (2014) no.8, 2982 doi:10.1140/epjc/s10052-014-2982-4 \[arXiv:1405.3936 \[hep-ex\]\].
S. Chatrchyan [et al.]{} (CMS Collaboration), Phys. Rev. Lett. [**109**]{} (2012) 022301 doi:10.1103/PhysRevLett.109.022301 \[arXiv:1204.1850 \[nucl-ex\]\].
S. Chatrchyan [et al.]{} (CMS Collaboration), Phys. Rev. C [**84**]{} (2011) 024906 doi:10.1103/PhysRevC.84.024906 \[arXiv:1102.1957 \[nucl-ex\]\].
B. Abelev [et al.]{} (ALICE Collaboration), Phys. Rev. C [**88**]{} (2013) no.4, 044909 doi:10.1103/PhysRevC.88.044909 \[arXiv:1301.4361 \[nucl-ex\]\].
M. L. Miller, K. Reygers, S. J. Sanders and P. Steinberg, Ann. Rev. Nucl. Part. Sci. [**57**]{} (2007) 205 doi:10.1146/annurev.nucl.57.090506.123020 \[nucl-ex/0701025\].
S. Chatrchyan [*et al.*]{} (CMS Collaboration), Eur. Phys. J. C [**74**]{} (2014) no.7, 2951 doi:10.1140/epjc/s10052-014-2951-y \[arXiv:1401.4433 \[nucl-ex\]\].
G. Aad [et al.]{} (ATLAS Collaboration), Phys. Lett. B [**748**]{} (2015) 392 doi:10.1016/j.physletb.2015.07.023 \[arXiv:1412.4092 \[hep-ex\]\].
A. Adare [et al.]{} (PHENIX Collaboration), Phys. Rev. Lett. [**116**]{} (2016) no.12, 122301 doi:10.1103/PhysRevLett.116.122301 \[arXiv:1509.04657 \[nucl-ex\]\].
G. Martinez-Garcia, arXiv:1408.3108 \[hep-ph\].
A. Bzdak, V. Skokov and S. Bathe, Phys. Rev. C [**93**]{} (2016) no.4, 044901 doi:10.1103/PhysRevC.93.044901 \[arXiv:1408.3156 \[hep-ph\]\].
M. Alvioli, B. A. Cole, L. Frankfurt, D. V. Perepelitsa and M. Strikman, Phys. Rev. C [**93**]{} (2016) no.1, 011902 doi:10.1103/PhysRevC.93.011902 \[arXiv:1409.7381 \[hep-ph\]\].
D. V. Perepelitsa and P. A. Steinberg, arXiv:1412.0976 \[nucl-ex\].
N. Armesto, D. C. Gülhan and J. G. Milhano, Phys. Lett. B [**747**]{} (2015) 441 doi:10.1016/j.physletb.2015.06.032 \[arXiv:1502.02986 \[hep-ph\]\].
C. M. Tarbert [et al.]{}, Phys. Rev. Lett. [**112**]{} (2014) no.24, 242502 doi:10.1103/PhysRevLett.112.242502 \[arXiv:1311.0168 \[nucl-ex\]\].
M. B. Tsang [et al.]{}, Phys. Rev. C [**86**]{} (2012) 015803 doi:10.1103/PhysRevC.86.015803 \[arXiv:1204.0466 \[nucl-ex\]\].
J. Zenihiro [et al.]{}, Phys. Rev. C [**82**]{} (2010) 044611. doi:10.1103/PhysRevC.82.044611
C. J. Horowitz and J. Piekarewicz, Phys. Rev. Lett. [**86**]{} (2001) 5647 doi:10.1103/PhysRevLett.86.5647 \[astro-ph/0010227\].
H. Paukkunen, Phys. Lett. B [**745**]{} (2015) 73 doi:10.1016/j.physletb.2015.04.037 \[arXiv:1503.02448 \[hep-ph\]\].
I. Helenius, K. J. Eskola, H. Honkanen and C. A. Salgado, JHEP [**1207**]{} (2012) 073 doi:10.1007/JHEP07(2012)073 \[arXiv:1205.5359 \[hep-ph\]\].
G. Antchev [et al.]{} (TOTEM Collaboration), Europhys. Lett. [**101**]{} (2013) 21004. doi:10.1209/0295-5075/101/21004
H. L. Lai, M. Guzzi, J. Huston, Z. Li, P. M. Nadolsky, J. Pumplin and C.-P. Yuan, Phys. Rev. D [**82**]{} (2010) 074024 doi:10.1103/PhysRevD.82.074024 \[arXiv:1007.2241 \[hep-ph\]\].
K. J. Eskola, H. Paukkunen and C. A. Salgado, JHEP [**0904**]{} (2009) 065 doi:10.1088/1126-6708/2009/04/065 \[arXiv:0902.4154 \[hep-ph\]\].
P. Aurenche, R. Baier, M. Fontannaz and D. Schiff, Nucl. Phys. B [**297**]{} (1988) 661. doi:10.1016/0550-3213(88)90553-6
F. Aversa, P. Chiappetta, M. Greco and J. P. Guillet, Nucl. Phys. B [**327**]{} (1989) 105. doi:10.1016/0550-3213(89)90288-5
P. Aurenche, M. Fontannaz, J. P. Guillet, B. A. Kniehl, E. Pilon and M. Werlen, Eur. Phys. J. C [**9**]{} (1999) 107 doi:10.1007/s100529900018 \[hep-ph/9811382\].
P. Aurenche, M. Fontannaz, J. P. Guillet, B. A. Kniehl and M. Werlen, Eur. Phys. J. C [**13**]{} (2000) 347 doi:10.1007/s100520000309 \[hep-ph/9910252\].
L. Bourhis, M. Fontannaz and J. P. Guillet, Eur. Phys. J. C [**2**]{} (1998) 529 doi:10.1007/s100520050158 \[hep-ph/9704447\].
I. Helenius, K. J. Eskola and H. Paukkunen, JHEP [**1305**]{} (2013) 030 doi:10.1007/JHEP05(2013)030 \[arXiv:1302.5580 \[hep-ph\]\].
G. Aad [et al.]{} (ATLAS Collaboration), Phys. Rev. C [**93**]{} (2016) no.3, 034914 doi:10.1103/PhysRevC.93.034914 \[arXiv:1506.08552 \[hep-ex\]\].
B. B. Abelev [et al.]{} (ALICE Collaboration), Phys. Lett. B [**736**]{} (2014) 196 doi:10.1016/j.physletb.2014.07.011 \[arXiv:1401.1250 \[nucl-ex\]\].
J. Adam [et al.]{} (ALICE Collaboration), Phys. Rev. C [**93**]{} (2016) no.3, 034913 doi:10.1103/PhysRevC.93.034913 \[arXiv:1506.07287 \[nucl-ex\]\].
V. Khachatryan [et al.]{} (CMS Collaboration), arXiv:1611.01664 \[nucl-ex\].
D. de Florian, R. Sassot and M. Stratmann, Phys. Rev. D [**76**]{} (2007) 074033 doi:10.1103/PhysRevD.76.074033 \[arXiv:0707.1506 \[hep-ph\]\].
S. Kretzer, Phys. Rev. D [**62**]{} (2000) 054001 doi:10.1103/PhysRevD.62.054001 \[hep-ph/0003177\].
S. Albino, B. A. Kniehl and G. Kramer, Nucl. Phys. B [**803**]{} (2008) 42 doi:10.1016/j.nuclphysb.2008.05.017 \[arXiv:0803.2768 \[hep-ph\]\].
A. Uras, PoS LHCP [**2016**]{} (2016) 177
[^1]: In the analysis of Ref. [@Tarbert:2013jze], the proton density was taken as fixed when fitting the neutron parameters to the data. Therefore, the proton density has no uncertainty here.
[^2]: Currently, there is no coherent way to treat the PDF nuclear modifications within the Monte-Carlo Glauber model. This is actually why, in this work, we stick to the optical version of the Glauber model.
[^3]: The valence quark-gluon channel dominates irrespectively of the isolation.
[^4]: In these rough estimates we do not consider in detail the uncertainties related e.g. to the use of different FFs, next-to-NLO corrections, choices for the fragmentation/factorization/renormalization scales, or suppression of the hadron yields in Pb+Pb relative to p+p baseline.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Light dark matter with mass smaller than about 10 GeV is difficult to probe from direct detection experiments. In order to have the correct thermal relic abundance, the mediator of the interaction between dark matter and the Standard Model (SM) should also be relatively light, $\sim 10^2$ GeV. If such a light mediator couples to charged leptons, it would already be strongly constrained by direct searches at colliders. In this work, we consider the scenario of a leptophobic light $Z''$ vector boson as the mediator, and study the the prospect of searching for it at the 8 TeV Large Hadron Collider (LHC). To improve the reach in the low mass region, we perform a detailed study of the processes that the $Z''$ is produced in association with jet, photon, $W^\pm$ and $Z^0$. We show that in the region where the mass of $Z''$ is between 80 and 400 GeV, the constraint from associated production can be comparable or even stronger than the known monojet and dijet constraints. Searches in these channels can be complementary to the monojet search, in particular if the $Z''$ couplings to quarks ($g_{Z''}$) and dark matter ($g_D$) are different. For $g_D < g_{Z''}$, we show that there is a larger region of parameter space which has correct thermal relic abundance and a light $Z''$, $M_{Z''} \sim 100 $ GeV. This region, which cannot be covered by the mono-jet search, can be covered by the resonance searches described in this paper.'
author:
- |
**Haipeng An$^a$, Ran Huo$^b$ and Lian-Tao Wang$^{b,c}$**\
\
*$^a$Perimeter Institute, Waterloo, Ontario N2L 2Y5, Canada*\
*$^b$Enrico Fermi Institute, University of Chicago, Chicago, IL 60637*\
*$^c$Kavli Institute for Cosmological Physics, University of Chicago, Chicago, IL, 60637*\
title: |
\
Searching for Low Mass Dark Portal at the LHC\
---
Introduction
============
Dark Matter (DM) consists of 24% of the energy density of our universe. However, the nature of it is one of the outstanding mysteries. The most popular class of candidates for DM are the stable Weakly Interacting Massive Particles (WIMPs), and the annihilation of WIMPs in the early universe determined the observed relic abundance of DM. This scenario provides a promising approach to detect the WIMPs directly [@Goodman:1984dc], with many recent results [@Aalseth:2011wp; @Akerib:2010pv; @Ahmed:2010wy; @Aprile:2010um; @Angloher:2011uu; @Bernabei:2008yi; @Aprile:2011hi; @Felizardo:2010mi; @Aprile:2012nq]. The direct detection experiments are more sensitive for WIMPs with masses heavier than tens of GeV. Assuming the WIMP-nucleon interaction is spin independent, the limit can be as strong as $ \sigma_{\rm WIMP-nucleon} < 10^{-43}$ to $10^{-45}$ cm$^2$. For light WIMPs, the energy transfer during a WIMP-nucleus scattering is proportional to the WIMP mass. If the WIMP is sufficiently light, the energy transfer is too small to be detected, resulting in a much weaker limit. For example, the upper limit is weaker than $10^{-39}$ cm$^2$ for $M_{\rm WIMP} \sim 5$ GeV.
Recently, the low mass dark matter scenario has received more attention. For example, there is an intriguing connection between the WIMP number density and baryon number density in this scenario [@Nussinov:1985xr; @Kaplan:2009ag; @Farrar:2005zd; @Kitano:2004sv; @Agashe:2004ci; @An:2009vq; @Shelton:2010ta; @Davoudiasl:2010am; @Bell:2011tn; @Gu:2010ft; @Blennow:2010qp; @Dutta:2010va; @Kang:2011wb; @Cheung:2011if; @MarchRussell:2011fi; @Frandsen:2011kt; @Ibe:2011hq; @Kamada:2012ht; @Feng:2012jn], which is motivated by the fact that the energy densities of dark matter and baryons are of the same order of magnitude. High energy colliders can play a significant role in the search for light WIMPs. The simplest approach is to assume that the interaction between WIMPs and Standard Model (SM) particles can be written as effective contact operators [@Goodman:2010yf; @Bai:2010hh; @Goodman:2010ku; @Fortin:2011hv; @Rajaraman:2011wf; @Shoemaker:2011vi; @Graesser:2011vj; @Friedland:2011za], with the mediators between the WIMP and the SM sector integrated out. The simplest observable signal would be WIMP pair production associated with a jet, which is often referred as the monojet + missing transverse energy (MET) search channel. On the other hand, it is a generic possibility that the mediator can be light. In this case, it cannot be integrated out while considering scattering process at the LHC. Such an effect has been studied in Refs. [@Bai:2010hh; @Goodman:2010ku; @Fortin:2011hv; @Rajaraman:2011wf; @Shoemaker:2011vi; @Graesser:2011vj; @Friedland:2011za; @An:2012va]. A light mediator is particularly motivated by reproducing the observed thermal relic abundance, which requires a WIMP annihilation cross section is of order picobarn. If WIMP annihilation mainly proceeds through s-wave , the cross section can be written as $$\label{sigmav}
\langle\sigma v\rangle \approx \frac{ g^2_{Z'} g_D^2 N_C N_f M_D^2}{2\pi M_{Z'}^4}=3\times10^{-26}\left(\frac{g_{Z'}}{0.27}\right)^2\left(\frac{g_D}{0.27}\right)^2\left(\frac{M_D}{5\mbox{GeV}}\right)^2\left(\frac{100\mbox{GeV}}{M_{Z'}}\right)^4\thinspace\mbox{cm}^3\mbox{s}^{-1},$$ where we have assumed that the interaction between WIMP and the SM quarks is mediated by a vector boson $Z'$. This is the so called dark portal. The relevant Lagrangian can be written as $$\mathcal{L}\ni g_{Z'}\bar{q}\gamma^\mu q Z_\mu' + g_D\bar{\chi}\gamma^\mu\chi Z_\mu',$$ where we have assumed that the dark matter particle, denoted by $\chi$, is a Dirac fermion. $g_{Z'}$ and $g_D$ are the coupling of the mediator to quarks and to WIMP, respectively. $N_f$ is the number of SM *flavors* that are kinematically available. For such a low mass resonance its coupling to leptons must be strongly suppressed, otherwise it would be ruled out by LEP. That’s the reason that in our model the $Z'$ couples only to quarks in SM (leptophobic).
Such a light mediator with $M_{Z'} \sim \mathcal{O}(100)$ GeV is within the reach of the Large Hadron Collider (LHC). The most straightforward way to search for a leptophobic $Z'$ is a direct search for dijet resonance, a “bump hunt”, which is summarized in Ref. [@An:2012va]. Although there is some model dependence, it turns out that resonance searches for the vector mediator with masses between 250 GeV and 4 TeV at the colliders and mapping to DM direct detection constraint can provide stronger constraints than the monojet searches. On the other hand, the constraints become much weaker for low mass mediator. Since the jets from lighter $Z'$ decay would be softer, the signal of a light $Z'$ suffers from very low trigger efficiencies.
In this work, we study a class of alternative channels in which the $Z'$ boson is produced in association with a hard jet, a hard photon, or a massive gauge boson in the SM. We also consider the case of $Z'$ pair production. The rates of the SM background for the associated productions can be significantly lower. Therefore, we can lower the trigger threshold and enhance the signal efficiency. In the following we will demonstrate that this is indeed a promising way of finding or constraining the leptophobic $Z'$ mediator, and eventually the DM particle.
We present a detailed description of the Monte Carlo simulation in Sec. 2. In Sec. 3, we scan the parameter space and show the reach for each channel and their combination. In Sec. 4, we relate the $Z'$ search to the collider search of dark matter. We summarize the results in the Sec. 5.
Simulation and Event Selection
==============================
Signal and Background Simulation
--------------------------------
The signal and background are both generated by `MadGraph/MadEvents 5.1.3.9` [@Alwall:2011uj]. We used the $k_T$ MLM Matrix Element-Patron Shower (ME-PS) matching algorithm. The decay width of $Z'$ is calculated by `BRIDGE 2.24` [@Meade:2007js]. The following decay and hadronization procedure are performed by `pythia-pgs 2.1.8`. The jets are constructed by `fastjet 3.0.0` [@Cacciari:2011ma] with $R=0.6$ anti-$k_T$ jet algorithm, and smearing is performed for jets, leptons, photons with the CMS-like energy resolution [@Ball:2007zza] $$\begin{aligned}
\label{smear}
\frac{\delta E_J}{E_J}&=&\frac{0.84}{\sqrt{E_J/\mbox{GeV}}}\oplus0.07,\\
\frac{\delta E_\ell}{E_\ell}&=&\frac{0.028}{\sqrt{E_\ell/\mbox{GeV}}}\oplus0.003\oplus\frac{0.125}{E_J/\mbox{GeV}},\\
\frac{\delta E_\gamma}{E_\gamma}&=&\frac{0.01}{\sqrt{E_\gamma/\mbox{GeV}}}\oplus0.007.\end{aligned}$$ Throughout this paper we assume a 15 fb$^{-1}$ total integrated luminosity at centre-of-mass energy of 8 TeV.
The next-to-leading order correction K-factor $K=\sigma_{NLO}/\sigma_{LO}$ will not significantly change our results. In particular, it is calculated by `MCFM 6.2` [@Campbell:MCFM] to be roughly one, both for the $Z'\thinspace+W^\pm$ channel and the $Z'\thinspace+Z^0$ channel. For the $Z'\thinspace+$ jet and the $Z'\thinspace+\thinspace\gamma$ channels we do not preform a calculation, but [@Bern:2011ep] suggests the K factor for pure QCD 3 jets should be about 1.3 or so, and in [@Bern:2011pa] the K factor for $\gamma+$ 2 jets is about 1.2 or so. All the K factors are close to one, so we expect counting the higher order corrections will not significantly change our reach.
We approximate $Z'$ as a Breit-Wigner resonance. The decay width of $Z'$ can be written as $$\label{width_ana}
\Gamma\simeq \frac{g_{Z'}^2}{12\pi}M_{Z'}(N_f N_C + r^2),$$ where $N_f=5$ or 6 if $M_{Z'}$ is above the ditop threshold, $N_C=3$ is the number of colors in QCD, and $r\equiv g_D/g_{Z'}$. As we will see in the following sections, the narrow width condition $\Gamma_{Z'}\ll M_{Z'}$ is always satisfied in the parameter region we are interested in. Therefore, the Breit-Wigner approximation is valid in our case. We will not consider the interference effect of the $Z'$ signal with the SM background. Such an effect is only important within the $Z'$ width. So in our narrow width $Z'$ scenario the effect is small, and it will be completely washed out due to the jet energy resolution. We verified that this is indeed the case.
Event Selection
---------------
The most important class of event selection cuts for our signal are the jet energy thresholds and acceptance cuts. We begin with the $Z'$ + jet channel. Due to the large QCD background, we have to adopt relatively high jet energy threshold. With our choice, the leading order SM background cross section corresponds to an event rate of a few Hz or so, with the current peak instantaneous luminosity. Since the energy of the jets from $Z'$ decay is closely correlated with the $Z'$ mass, we use two complementary sets of $p_{TJ}$ selection cuts, so that in combination they give us good signal efficiency for a large range of $M_{Z'}$. For light $Z'$ ($M_{Z'}<350$ GeV) we require $$\label{QCDlight}
p_{TJ1}>350\enspace\mbox{GeV}\qquad\qquad p_{TJ2},p_{TJ3}>70\enspace\mbox{GeV}.$$ In this case, the second and third hardest jets are dominantly coming from $Z'$ decay. We choose to use a relatively low threshold for them to enhance signal efficiency. At the same time, we have to require the hardest jet to be very energetic to suppress the rate of QCD background. For heavier $Z'$ ($M_{Z'} > 350$ GeV), the decay products of $Z'$ give two hard jets that are most likely to be the two leading jets. The two hardest jets, if coming from $Z'$ decay, are typically close in $p_T$. Therefore, equal thresholds for the first two leading jets are desirable. We also require the jet produced in association with $Z'$ to be energetic to suppress the QCD background. In particular, we require $$\label{QCDheavy}
p_{TJ}>160\enspace\mbox{GeV for three leading jets}.$$
In the other channels, $Z' + \gamma$ and $Z' + W^\pm/Z^0$, the signal also contains at least two jets coming from $Z'$ decay. At the same time, the additional hard object in the event, such as a hard photon or hard leptons from $W^\pm/Z^0$ decay, can be used to efficiently trigger on this class of events. Therefore, we can afford to use lower thresholds for jets $$\label{PTJCut}
p_{TJ}>50\enspace\mbox{GeV for two leading jets}.$$
We will focus on central jets with a good energy resolution. For all the channels considered in this paper, we impose an acceptance cut $$\label{EtaJCut}
|\eta_{J}|<2.5.$$ We apply this cut not only to jets but also to $\gamma$s and charged leptons.
We also note that the two jets from $Z'$ decay tend to be central. At the same time, the background QCD jets are more forward with larger rapidity gaps. Therefore, we impose an $\eta$ separation cut $$\begin{aligned}
\label{EtaJSeparation}
|\Delta\eta_{JJ}|<1.7,\end{aligned}$$ for the two jets which are used to reconstruct the $M_{Z'}$.
We have also considered different jet energy thresholds in both light and heavier $Z'$ cases. The results are tabulated in Table \[TrigVar\_1\] and Table \[TrigVar\_2\] in Appendix \[sec:jet\_trig\]. While we use the relatively conservative choices of the jet energy threshold, lower them can certainly increase the reach.
\[InvMass\]
$Z'$ Mass (GeV) 60 80 100 120 150 250 350
----------------- --------- --------- --------- --------- --------- ---------- ---------
$M_{JJ}$ (GeV) 50-70 60-90 80-110 100-130 120-160 210-270 290-370
$Z'$ Mass (GeV) 450 550 650 750 850 950
$M_{JJ}$ (GeV) 370-480 450-590 530-700 610-800 700-900 800-1000
: Invariant mass window for each hypothetical $Z^\prime$ mass. We optimize this with signal Monte Carlo, where the boundaries correspond to the bins with half of the peak bin height.
To search for the $Z'$ resonance, we have to identify two jets as candidates for $Z'$ decay products. In the $Z' + $jet channel with $M_{Z'} < 350$ GeV, we choose the second and third hardest jet. For $M_{Z'} > 350 $ GeV, we choose the two leading jet instead. In $Z' + \gamma$ and $Z' + W^\pm/Z^0$ channels, we also choose the two leading jets. After identifying the two candidate jets, we require their invariant mass to be within the mass window around the target $Z'$ mass. The width of the window has two origins, one is the natural line width or the physical $Z'$ total decay width, and the other is due to the detector finite energy resolution. In our case, the latter one always dominates. We optimized the mass window for a set of $Z'$ masses based on Monte Carlo simulation. The complete list of the invariant mass windows used in our analysis is Table \[InvMass\].
We also impose selection cuts on the other (non-jet) hard objects in the signal. For the $Z^\prime$ + $\gamma$ channel, we require $$p_{T\gamma}>50\enspace\mbox{GeV}.$$ For the $Z^\prime$ + $W^\pm$ channel, we focus on the leptonic channel. We require $$\begin{aligned}
p_{T\ell}>25\enspace\mbox{GeV},\\
p\hspace{-0.4em}\slash_{T}>25\enspace\mbox{GeV}.\end{aligned}$$ For the $Z^\prime$ + $Z^0$ channel, we consider two different $Z^0$ decay modes. The first channel is the charged leptonic $Z^0$ decay mode. In this case, there are two opposite sign electrons or muons which correctly reconstruct a $Z^0$. Specifically, we require $$\begin{aligned}
p_{T\ell}>25\enspace\mbox{GeV for two opposite sign leptons},\\
85\enspace\mbox{GeV}<M_{\ell\ell}<97\enspace\mbox{GeV}.\end{aligned}$$ The second channel is the invisible decay $Z^0>\nu\tilde{\nu}$. We require $$\slashed{p}_{T}>60\enspace\mbox{GeV},$$ and veto any charged leptons. They should be replaced by the practical triggering conditions at the LHC.
In our study, we have not considered $Z' \to b \bar b$ as a possible decay channel, mainly for simplicity. The results are already quite encouraging without $b\bar b$ decay channel, and including it will certainly enhance the discovery reach. Extending the analysis in this paper to this case is straightforward, after properly taking into account the b-tagging. For heavier $Z'$, decaying into $t \bar t$ would also give another signal channel. At the same time, identifying the top requires different strategies, depending on $M_{Z'}$ [@Altheimer:2012mn].
Reach of Different Channels
===========================
In general, the couplings of $Z'$ to the left and right-handed quarks can be different. In the $Z'\thinspace+$ jet and $Z'\thinspace+\thinspace\gamma$ channels, since QCD and QED are vector-like, only the combination $g=(g_{Z'L}^2+g_{Z'R}^2)^{1/2}$ is relevant, where $g_{Z'L}$ and $g_{Z'R}$ are the couplings of $Z'$ to the left and right-handed quarks. Only the left handed coupling is relevant for the $Z'\thinspace+\thinspace W^\pm$ channel. The most complicated channel is $Z'\thinspace+\thinspace Z^0$, in which both the $g_{Z'L}$ and $g_{Z'R}$ are relevant. Further complication comes with two kinds of decay channels under consideration. However, we will see in the following that the reach is mainly due to the non-chiral $Z'\thinspace+$ jet channel. The chiral $Z'\thinspace+\thinspace W^\pm$ contribution is nearly the same as the non-chiral $Z'\thinspace+\thinspace\gamma$ contribution. $Z'\thinspace+\thinspace Z^0$ has a very small contribution. Therefore, we expect the ratio $g_{Z'L}/g_{Z'R}$ will not play a significant role, and we will only show the results of $g_{Z'L}=g_{Z'R}\equiv g_{Z'}$.
QCD 3 jet dijet$+ \gamma$ dijet$+W^\pm$ dijet+$(Z^0 \to \ell^+\ell^-)$ dijet$+(Z^0 \to \nu\tilde{\nu})$
--------------------------- ----------------------------------------------------------------------------- ----------------- --------------- -------------------------------- ----------------------------------
$p_{TJ}$, $\eta$ cuts $\hspace{-0.4em}\begin{array}{l}487~\mbox{pb} \\ 412~\mbox{pb} \end{array}$ 327 pb 160 pb 7.6 pb 20 pb
$\Delta\eta$ cut $\hspace{-0.4em}\begin{array}{l}370~\mbox{pb} \\ 284~\mbox{pb} \end{array}$ 226 pb 109 pb 4.2 pb 14 pb
$M_{JJ}\in[120-160]$ GeV 42 pb 49 pb 13 pb 0.95 pb 2.5 pb
$M_{JJ}\in[290-370]$ GeV 67 pb/11 pb 18 pb 4.0 pb 0.32 pb 1.3 pb
$M_{JJ}\in[800-1000]$ GeV 25 pb 0.76 pb 0.21 pb 0.04 pb 0.11 pb
: \[tab:channel\] Leading order standard Model background rate for each of the channels, after various selection cuts. In the first row the basic $p_T$ thresholds and $\eta<2.5$ cut for *all particles* in an event are implemented. Then in the second row we add $\Delta\eta_{12}$ cuts for two leading jets, as well as the invariant mass window for $e^+e^-$ or $\mu^+\mu^-$ in the dijet+ $(Z^0 \to \ell^+\ell^-)$ channel to reconstruct $Z^0$. The event selection cuts are designed in particular to make the corresponding rates here no more than a few Hz for each channel. Lastly invariant mass windows are applied for three hypothetical $Z'$ masses. In the QCD 3 jet channel, we list the two cross sections, which correspond separately to the 350-70-70 $p_{TJ}$ triggering (the former) and the 160-160-160 $p_{TJ}$ triggering (the latter).
$Z'+$ jet $Z'+ \gamma$ $Z'+W^\pm$ $Z'+(Z^0 \to \ell^+\ell^-)$ $Z'+(Z^0 \to \nu\tilde{\nu})$
------------------ ----------------- -------------- ------------ ----------------------------- -------------------------------
$M_{Z'}=150$ GeV 0.80 pb 0.76 pb 0.45 pb 0.031 pb 0.054 pb
$M_{Z'}=350$ GeV 0.72 pb/0.15 pb 0.30 pb 0.14 pb 0.0096 pb 0.021 pb
$M_{Z'}=950$ GeV 0.20 pb 0.013 pb 0.006 pb 0.0004 pb 0.0011 pb
: \[channel\_sig\] Leading order $Z'$ signal rate for each of the channels. All selection cuts in Table 2 (including the mass window) are applied. Here we are assuming $g_L=g_R=0.35$. Again in the $Z'\thinspace+$jet channel we list the two cross sections, which correspond separately to the 350-70-70 $p_{TJ}$ threshold (the former) and the 160-160-160 $p_{TJ}$ threshold (the latter).
As illustration, a list of cross sections for various background channels and signal channels with $g_{Z'}=0.35$ are shown separately in Table \[tab:channel\] and Table \[channel\_sig\]. The expected $2\sigma$ constraints from different channels are shown in Fig. \[fig:Z4channels\], where the red, green, blue, purple and black curves correspond to the upper limits from the $Z'+{\rm jet}$, $Z'+\gamma$, $Z'+W^{\pm}$, $Z'+Z^0$ channels and the combined constraint of the four channels, respectively. Here we have assumed here $g_D = 0$. In realistic models considered later, the decay branching ratio to jets will induce a suppression to the signal rate in each channel, however, the suppression is small at least for $g_D\sim g_{Z'}$[^1]. A detailed discussion is in Sec. \[sec:darkportal\]. For $M_{Z'}\gtrsim1000$ GeV the background QCD dijet rate is low enough that the more efficient direct dijet resonance search provides a better limit [@Aad:2011fq], so we only consider the cases with $M_{Z'} < 1000$ GeV here.
From Fig. \[fig:Z4channels\], we can see that the constraint from $Z'+$jet channel is stronger than other channels, especially in the large $M_{Z'}$ region. The dominant reason is the QCD coupling is much larger than the other relevant gauge couplings. For the $Z'+\gamma$ channel, both the signal and the background are suppressed by the fine structure constant of electromagnetic interaction, $\alpha_{\rm em}$. Whereas in the case of $Z'$+jet channel, it is replaced by $\alpha_S$, which is much larger. As a result, both the signal and the background increase by roughly the same amount in comparison with the $Z'+\gamma$ channel, and therefore the $S/\sqrt{B}$ for the $Z'$+jet channel is larger than that for the $Z'+\gamma$ channel. This is exactly what happens in the large $M_{Z'}$ in Fig. \[fig:Z4channels\]. In the small $M_{Z'}$ region, as shown in Tables \[tab:channel\] and \[channel\_sig\], the cuts make the cross sections of $Z'+$jet channel and $Z'+\gamma$ channels comparable in both signal and background processes. Therefore, as shown in Fig. \[fig:Z4channels\], the constraints on $g_{Z'}$ is comparable in the region $M_{Z'}\sim150$ GeV.
In the $Z'+$jet channel, in the small $M_{Z'}$ region where $M_{Z'} \lesssim 100$ GeV, to balance the transverse momentum of the hardest jet, the $Z'$ resonance is highly boosted. Therefore, the two jets from the decay of the $Z'$ resonance are close to each other, and cannot be distinguished from a single fat jet. Due to the same reason, the small $M_{23}$ region in the background is also removed, where $M_{23}$ is the invariant mass of the second and third hardest jets. As a result, the events of both the signal and the background are cut off by the requirements in Eq. (\[QCDlight\]) and the statistics becomes poor, and therefore the limit on $g_{Z'}$ from $Z'+$jet channel weakens significantly in the region where $M_{Z'}\lesssim 80$ GeV. A quantitative analysis which shows the scale of this effect can be found in Appendix \[appendixB\]. For example, for $M_Z' \sim 80$ GeV which is discussed in Appendix \[appendixB\], if we use the $R=0.1$ anti-$k_T$ jet algorithm rather than the default $R=0.6$ one in a hadron level analysis, we will get about 20 times more events with a dijet invariant mass in the window of $60-90$ GeV. Jet substructure techniques should be able to help in this regime, and we leave the details in this direction for a future study. In contrast, for the $Z'+\gamma$ and $Z'+W$ channels, there is no such kinematic configuration. Therefore, in the small $M_{Z'}$ region, the constraint is mainly from these two channels.
In the $Z'\thinspace+\thinspace\gamma$ channel, it is possible for jets to fake photons. For example, there can be a hard jet which accidentally becomes a $\pi^0$ and passes the neutral electromagnetic trigger to mimic a photon. The jet fake gamma rate is tiny (less than 0.3%), but given the QCD jets are much more abundant, the fake trigger is still considerable. Taking this into consideration, the background will increase by about 20%.
The $Z'\thinspace+\thinspace W^\pm$ channel has comparable $S/\sqrt{B}$ ratio with the $Z'\thinspace+\thinspace\gamma$ channel over all the $Z'$ mass region, which is just a coincidence of various physical reasons contributing in different directions. For example, the weak coupling is larger than the electromagnetic coupling, the $W^\pm$ is massive so that the production is relatively suppressed. $W^\pm \to \ell^{\pm} $ has further suppression from the leptonic decay branching ratio. At the same time, the $\gamma$s are concentrated in the collinear region and hard to pass the $p_{T\gamma}>50$ GeV selection cut, while the acceptance of a $W^\pm \to \ell^{\pm}$ is higher.
For the $Z'\thinspace+\thinspace Z^0$ channel, we have to combine its two decay channels. The neutrino channel has a larger branching ratio ($\sim$20%) than the chagerd lepton channel ($\sim$6.7%), and the former has a slightly better $S/\sqrt{B}$. However, even the combined signal significance is much smaller than other channels, so this channel is less interesting.
We have also checked the $Z'\thinspace+\thinspace Z'$ pair production channel. The $S/\sqrt{B}$ ratio is always much less than 1 in the region of mass and coupling we focus on. One reason is that the couplings are all relatively small, which leads to small production cross section. At the same time, the signal is in a pure 4-jet final state, which is overwhelmed by the dominant QCD background. For a similar background rate with the 3-jet case we have to use nearly the same jet $p_{TJ}$ thresholds, but the signal cross section is further suppressed by a small factor of $g_{Z'}^2$. Therefore, the reach in this channel would be much weaker, and we will not provide the full analysis here.
![Reach plots for $S/\sqrt{B}$ for the $Z'$ + jet (red), the $Z'\thinspace+\thinspace\gamma$ (green), the $Z'\thinspace+\thinspace W^\pm$ (blue) and the $Z'\thinspace+\thinspace Z^0$ (purple) channels, for 15 fb$^{-1}$ integrated luminosity and $S/\sqrt{B}=2$ which corresponds to 95% confidence level. All the channel is assuming $g_{Z'L}=g_{Z'R}=g_{Z'}$, and in the $Z'\thinspace+\thinspace Z^0$ channel the reach from two charged leptons decay product and from two neutrinos are combined. Eventually their combination for exclusion is shown as the black thick line.[]{data-label="fig:Z4channels"}](Z_4channels.eps){height="3in"}
![Comparison between the combined constraint of different channels shown in Fig. \[fig:Z4channels\] and the constraints from dijet and monojet searches. We have assumed $r \equiv g_{D}/g_Z' =1$. The thick black curves shows the combined constraint as in Fig. \[fig:Z4channels\]. The red curve corresponds to the 95% C.L. ATLAS monojet upper limit. The green and blue curves correspond to 95% C.L. upper limits from CDF and ATLAS dijet searches, respectively. The purple curve corresponds to 90% C.L. upper limits from UA2 dijet search.[]{data-label="fig:compareR"}](compare_r_1.eps){height="3in"}
![Comparison of the $g_{Z'}=g_D$ reach of our $Z'$ association production (solid and dashed black curves for $2\sigma$ expected constraints and $5\sigma$ reach, respectively) with other experiments. The red curve is the constraint from ATLAS monojet search with 10 fb$^{-1}$ integrated luminosity at 8 TeV. The green curve is the bound from dijet resonance search by CDF with 1.13 fb$^{-1}$ integrated luminosity. The blue curve is the constraint from ATLAS dijet resonance search with 1 fb$^{-1}$ data set.[]{data-label="fig:dijetplus"}](Z_compare.eps){height="3in"}
Fig. \[fig:compareR\] shows the comparison between the constraints from dijet search with associated products, monojet search and direct dijet searches. To make connection with dark matter detection, we will now consider $g_D \neq 0$ and introduce $r \equiv g_{D}/g_{Z'} $ to parameterize its size. As an illustration, we assume $r \equiv g_{D}/g_{Z'} = 1$ in Fig. \[fig:compareR\] . The red curve is the 95% C.L. upper limit from ATLAS monojet searches [@Atlas10fb], the green and blue curves are the 95% C.L. upper limit from the CDF [@Aaltonen:2008dn] and ATLAS [@Aad:2011fq] dijet resonance searches, and the purple curve shows the 90%C.L. upper limit from UA2 dijet resonance search [@Alitti:1993pn], respectively. For dijet searches, colliders with smaller centre-of-mass energy give stronger constraints, since when $M_{Z'}$ is much smaller than the centre-of-mass energy of the collider, the constraint suffers from large QCD background due to the peak of the gluon parton distribution function at low $x$. For the comparison between the associated dijet constraint and the monojet constraint, a large amount of the background events can be removed with the help of the invariant mass window cut, so the associated dijet constraint can be stronger than the monojet constraint. On the other hand, as discussed before, in the very light $Z'$ region ($M_Z'\lesssim80$ GeV), the two jets from the decay of $Z'$ are either highly boosted and cannot be distinguished from a single jet, or probably cut by the $p_{TJ}>50$ GeV threshold which is roughly half of the resonance mass, whereas the invisible decay of $Z'$ is only characterized by large missing transverse energy. Therefore, in this region the monojet constraints can be stronger than the associated dijet constraint.
Before the end of this section, we briefly mention some existing results in the $W^\pm\thinspace+$ dijet resonance channel. Mainly motivated by checking the CDF $W^\pm\thinspace+$ dijet anomaly, the ATLAS [@Atlas102fb] and more recently CMS group [@:2012he] have performed searches in the same channel. In [@Atlas102fb] based on 5 fb$^{-1}$ integrated luminosity a leptophobic $Z'$ of 150 GeV and $g_{Z'}\simeq0.2$ [@Buckley:2011vc] is excluded. This is in broad agreement with our results shown in Fig. \[fig:Z4channels\]. Possible strategies to enhance the LHC reach in the $W^\pm\thinspace+$ dijet has also been studied in [@Eichten:2012hs]. However, the kinematics of this model in which the djiet resonance and $W$ are decay products of a heavier new resonance, is very different from the scenario considered in this paper.
$Z'$ as a Portal between the SM and Dark Matter {#sec:darkportal}
===============================================
The $Z'$ can mediate interaction between dark matter and SM particles, forming the so called dark portal. In this case, the constraint on $g_{Z'}$ can be mapped onto the constraints on DM direct detection cross section. The direct detection cross section for a nucleon (proton or neutron) is $$\label{cxdirect}
\sigma_{SI}\simeq \frac{9g_{Z'}^2g_D^2M_N^2M_D^2}{\pi M_{Z'}^4(M_D+M_N)^2}\simeq7.7\times10^{-40}\left(\frac{g_{Z'}}{0.1}\right)^2\left(\frac{g_D}{0.1}\right)^2\left(\frac{100\mbox{GeV}}{M_{Z'}}\right)^4\thinspace\mbox{cm}^2,$$ where $M_N$ is the mass of the nucleons, and $M_D=5$ GeV is assumed.
For $g_{Z'}=g_D$, the constraints on direct detections cross section are shown in Fig. \[fig:dijetplus\]. The major improvement is in the region with $Z'$ lighter than limit from the CDF dijet pole search. The constraint can be as strong as a few $\times10^{-42}$ cm$^2$. Assuming $g_{Z'}=g_D$ and $M_{Z'}>80$ GeV, limits from associated production are also stronger than those from the ATLAS monojet search. The current bound assumes an integrated luminosity of 15 fb$^{-1}$, this constraint will becomes stronger if the assumed integrated luminosity increases.
Relaxing $g_{Z'}=g_D$ leads to interesting scaling behavior. The production rate of $Z'$ is proportional to $g_{Z'}^2$. The decay branching ratio of $Z'$ into dijet final states in the case of $g_{Z'}\neq g_D$ can be written as $g_{Z'}^2N_f N_C/(g_{Z'}^2N_f N_C + g_D^2)$. On the other hand, the decay branching ratio of $Z'$ into DM can be written as $g_D^2/(g_{Z'}^2N_f N_C + g_D^2)$. Therefore, for a general $r \equiv g_{D}/g_Z'$, the cross sections for monojet and dijet processes can be written as $$\begin{aligned}
\label{scaling}
\sigma_{\rm monojet} (r) &=& \sigma_{\rm monojet}^{(0)} \times \frac{N_C N_f +1}{N_C N_f +r^2} r^2 \propto g_{Z'} g_D\times r \frac{N_C N_f + 1}{N_C N_f + r^2}\ ; \nonumber \\
\sigma_{\rm dijet} (r) &=& \sigma_{\rm dijet}^{(0)} \times \frac{N_C N_f + 1}{N_C N_f +r^2} \propto g_{Z'} g_D \times \frac{1}{r} \frac{N_C N_f + 1}{N_C N_f +r^2} \ ,\end{aligned}$$ where $\sigma_{\rm monojet}^{(0)}$ and $\sigma_{\rm dijet}^{(0)}$ are the cross sections for $r=1$. Therefore, as long as the $Z'$ is narrow-widthed and light enough so that it can be produced on shell, one can get the constraints on the coupling for a general value of $r$ by scaling the constraints shown in Fig. \[fig:compareR\] using Eq. (\[scaling\]).
![Comparison of the 5$\sigma$ reach for the discovery of a 5 GeV WIMP by the $Z'$ associated production search (all solid curves) and monojet search (all dash-dotted curves), for different ratios of $r\equiv g_D/g_{Z'}$. The region of parameter space above a certain curve either has been excluded or can be constrained by the corresponding search channel. All curves in red correspond to $g_D/g_{Z'} = 0.3$, all curves in green correspond to $g_D/g_{Z'} =0.7$, and all curves in blue correspond to $g_D/g_{Z'} =1$. In shaded region, the $\chi\bar{\chi}\leftrightarrow Z'$ process generates the correct relic abundance, namely requiring the cold dark matter density of the universe $\Omega_{\rm cdm} h^2 = 0.111\pm0.006$ [@pdg]. []{data-label="fig:withrelic"}](relic_zprime.eps){height="3in"}
For fixed $g_{Z'}g_D$, the constraints of the production cross section from monojet and dijet searches have approximately opposite dependence on $r$. Therefore, the monojet method and our associated dijet method are complementary: the former works for large $r$ whereas the latter works for small $r$. We can also take into account the theoretical consideration of DM thermal production, or the so called “WIMP miracle”. If the $Z'$ is the only portal between the DM and the SM sectors, and the relic abundance of DM is determined by thermal freeze-out, the DM annihilation cross section to SM particles through the $Z'$ will be given in Eq. (\[sigmav\]), where we can see that $\langle\sigma v\rangle$ depends on the couplings only through the product $g_{Z'} g_D$ in the case that $M_D \ll M_{Z'}$. To get the observed relic abundance, $\langle\sigma v\rangle$ is fixed to be around $3\times10^{-26}$ cm$^3$s$^{-1}$; as a result, the product $g_{Z'}g_D$ becomes a function of $M_{Z'}$ for fixed $M_D$, which is shown as the grey band in Fig. \[fig:withrelic\].
In the region that $M_{Z'}$ and $r$ are both small, the monojet constraint becomes weaker, as expected from Eq. (\[scaling\]). It cannot probe the shaded region in Fig. \[fig:withrelic\] where the correct relic abundance is generated. In the region $r<0.7$, from the crossing of the solid and dash dotted green curves in Fig. \[fig:withrelic\], we can see that the correct relic abundance can be generated with parameters that satisfy the monojet constraint and can be reached by the associated dijet search with a significance larger than $5\sigma$ at the same time. In the region of $M_{Z'}<130$ GeV, from the red curves in Fig. \[fig:withrelic\], we can see that if $r$ is smaller than about 0.3, the associated dijet resonance search can reach the full region of parameter space with correct relic abundance, with more than $5\sigma$ significance. Whereas in the region of $M_{Z'}>130$ GeV, the associated dijet bump search cannot reach $5\sigma$ significance due to the constraint from UA2 dijet resonance search as shown in Fig. \[fig:compareR\].
Summary and discussions
=======================
In this paper, we have suggested a new systematic way of finding/constraining the general leptophobic $Z'$ gauge boson. To overcome the large QCD background at the LHC, we considered the channels in which $Z'$s are produced in association with a hard jet, an energetic photon, or a $W^\pm/Z^0$. We performed a detailed study of the potential of LHC search in these channels, and show that the $Z^\prime$ + jet, $Z'\thinspace+\thinspace\gamma$ and the $Z'\thinspace+\thinspace W^\pm$ channels are independently promising, and a combination can significantly improve the reach.
Assuming this $Z'$ also couples to dark matter, forming the so-called dark portal, this bound can be mapped onto a dark matter direct detection bound. We demonstrate that there is improvement in the low $Z'$ mass region ($M_{Z'}<400$ GeV) from the use of associated production. The collider constraint of DM direct detection cross section can be extended to a smaller hypothetical $Z'$ mass. For $g_{Z'}>g_D$, the $Z'$ associated production method holds an advantage over the monojet method. For example, if the $Z'$ mediates the only channel for DM to annihilate into SM particles, with small value of $r$, the monojet constraints become weak and the parameter space for generating the correct relic abundance with small 5 GeV $M_D$ and weak scale $Z'$ is still allowed, as shown in Fig. \[fig:withrelic\]. At the same time, large region of this parameter space can be covered by the associated dijet bump search discussed in this paper. In particular, we can discover such a $Z'$ in this search channel with a significance larger than $5 \sigma$, if $g_D/g_{Z'}$ is smaller than about 0.7.
Acknowledgments {#Acknowledgments}
===============
We would like to acknowledge useful information from Antonio Boveia about the triggering of the ATLAS experiment. R.H. wish to thank Peter Skands and Xiaohui Liu for useful discussions. H.A. is supported by in part by the Government of Canada through NSERC and by the Province of Ontario through MEDT. L.T.W. is supported by the NSF under grant PHY-0756966 and the DOE Early Career Award under grant DE-SC0003930.
Variations on jet energy thresholds {#sec:jet_trig}
===================================
-------------- ----------- ----------- ----------- ----------- -----------
Threshold 350-70-70 300-70-70 400-70-70 350-60-60 350-80-80
Background 370 pb 767 pb 200 pb 410 pb 333 pb
$S/\sqrt{B}$ 15.0 19.4 8.9 15.2 14.8
-------------- ----------- ----------- ----------- ----------- -----------
: \[TrigVar\_1\] Comparison of small variation of the first set of jet energy thresholds for the 3-jet signal. For the leading order event rate we also impose the $|\Delta\eta_{23}|<1.7$ cut, but no invariant mass window choice. In $S/\sqrt{B}$ we are assuming $g_L=g_R=0.35$, $M_{Z'}=150$ GeV and 15 fb$^{-1}$ integrated luminosity.
--------------- ------------- ------------- ------------- ------------- -------------
Trigger level 160-160-160 150-150-150 140-140-140 170-170-170 180-180-180
Background 286 pb 406 pb 579 pb 205 pb 148 pb
$S/\sqrt{B}$ 10.8 11.2 12.6 8.9 8.0
--------------- ------------- ------------- ------------- ------------- -------------
: \[TrigVar\_2\] Comparison of small variation of the second set of jet energy thresholds for the 3-jet signal. For the leading order event rate we also impose the $|\Delta\eta_{12}|<1.7$ cut, but no invariant mass window choice. In $S/\sqrt{B}$ we are assuming $g_L=g_R=0.35$, $M_{Z'}=450$ GeV and 15 fb$^{-1}$ integrated luminosity.
We compare different choices of jet energy thresholds, and the results are tabulated here.
Low $M_{Z'}$ detection thresholds for different channels {#appendixB}
========================================================
We start with considering an event with three jets in the final state, labelled by $J_1$, $J_2$ and $J_3$, with transverse momenta $\vec p_{T1}$, $\vec p_{T2}$ and $\vec p_{T3}$, respectively. Here, we require that $p_{T1}\geq p_{T2} \geq p_{T3}$. Then, the invariant mass of $J_2$ and $J_3$ can be written as $$M_{23}^2 = 2(p_{T2}^2+p_{z2}^2)^{1/2} (p_{T3}^2+p_{z3}^2)^{1/2} - 2 \vec p_{T2} \cdot \vec p_{T3} - 2 p_{2z}p_{3z} \ ,$$ where we neglect the invariant mass of a single jet. $p_{2z}$ and $p_{3z}$ are the longitudinal momenta of $J_2$ and $J_3$, respectively. We can always work in a frame where $p_{z3} = 0$ such that the above expression can be simplified as $$\begin{aligned}
M_{23}^2 &=& 2(p_{T2}^2+p_{z2}^2)^{1/2} p_{T3}- 2 \vec p_{T2} \cdot \vec p_{T3} \nonumber \\
&=& 2 p_{T2} p_{T3} \left[ \left( 1 + \frac{p_{z2}^2}{p_{T2}^2}\right)^{1/2} - 1\right] + 2p_{2T}p_{3T}(1-\cos\phi) \ ,\end{aligned}$$ where $\phi$ is the angle between $\vec p_{2T}$ and $\vec p_{3T}$. It is easy to see that $2 p_{T2}p_{T3} > 2 p_{T3}^{\rm min} (p_{T1}^{\rm min} - p_{T3}^{\rm min})\approx (200 {\rm GeV})^2$, where $p_{T1}^{\rm min}$, $p_{T2}^{\rm min}$ and $p_{T3}^{\rm min}$ are the cuts we impose on $p_{T1}$, $p_{T2}$ and $p_{T3}$. In practice, for the region $M_{Z'} < 350$ GeV, we choose $p_{T1}^{\rm min} = 350$ GeV, $p_{T2}^{\rm min}= p_{T3}^{\rm min}= 70$ GeV. Therefore, in the region $M_{23}^2\ll (200 {\rm GeV})^2$ we have $$\begin{aligned}
M_{23}^2 &\approx& 2p_{T2} p_{T3} \left( \frac{p_{z2}^2}{2p_{T2}^2} + \frac{1}{2} \phi^2\right) \nonumber \\
&\approx& p_{T2} p_{T3} (y^2 + \phi^2) \ ,\end{aligned}$$ where $y$ is the rapidity of $J_3$. In the anti-k$_T$ algorithm, the distance between $J_2$ and $J_3$ is defined as $$d_{23} = \min\left({p^{-2}_{T2}},{p^{-2}_{T3}}\right) \frac{y^2 + \phi^2}{R^2} \approx \frac{M_{23}^2}{p_{T2}^3 p_{T3} R^2} \ .$$ If $d_{23} < 1/p_{T2}^2$, $J_2$ and $J_3$ will be identified as a single jet. Therefore, it requires $d_{23} > 1/p_{T2}^2$ to identify $J_2$ and $J_3$ as two separate jets, which means $$M_{23} > R\sqrt{p_{2T}p_{3T}} \geq R\sqrt{p_{T3}^{\rm min} (p^{\rm min}_{T1} - p^{\rm min}_{T3})} \approx 80 {~\rm GeV} \ ,$$ where $R = 0.6$ has been used. Therefore, the cuts we imposed on the three-jet final state configuration remove both the background and the signal in small $M_{23}$ region as shown in Fig. \[fig:distribution\]. We should note that this argument is based on parton level analysis. Taking the parton shower and realistic detector effects in to account, a small amount of events from both background and signal in the region $M_{23} < R\sqrt{p_{T3}^{\rm min} (p^{\rm min}_{T1} - p^{\rm min}_{T3})}$ may pass the cuts and leak into the signal region. However, the statistics in the small $M_{23}$ region remains very poor, and so does the constraint on the coupling.
![Background distributions of dijet invariant mass in $Z'$+jet, $Z'+\gamma$ and $Z'+W$ channels. For $Z'+$jet channel, it is the invariant mass of the second and third hardest jets shown in the plot. []{data-label="fig:distribution"}](distribution.eps){height="3in"}
For the $Z'+\gamma$ channel, as shown in Fig. \[fig:distribution\], since we don’t require a very large transverse momentum for the photon, the limit of the invariant mass of the two jets can simply be written as $$M_{12} \geq 2 p^{\rm min}_{TJ} R / 2 \approx 30 ~{\rm GeV} \ ,$$ where $p^{\rm min}_{TJ} = 50$ GeV is used. Therefore, the constraint from this channel dominates over the $Z'+$jet channel in the small $M_{Z'}$ region.
For the $Z'+W$ channel, since we only consider the leptonic channel, $W$ boson does not have to be produced with a large transverse boost to pass the selection cuts. Therefore, $Z'$ don’t have to be very boosted, and there will be no limitation on $M_{12}$ in the small $M_{Z'}$ region. For the $Z'+Z^0$ channel, however, the dominant decay channel of $Z^0$ we are considering is to a pair of neutrinos and the $\not\!\!p_T > 60$ is required, which indicates that the $p_T$ of $Z^0$ boson must be larger than 60 GeV. For fixed dijet invariant mass $M_{12}$, in the $Z'+W$ channel, ${\hat s}^{1/2}_{\rm min} = M_{W}+ M_{12}$, where ${\hat s}^{1/2}_{\rm min}$ is the minimal parton level centre-of-mass energy to make this process possible. Whereas in the $Z'+Z^0$ channel, ${\hat s}^{1/2}_{\rm min} = \sqrt{M_Z^2 + {p_{TZ}^{\rm min}}^2} + \sqrt{M_{Z'}^2 + {p_{TZ}^{\rm min}}^2}$. Therefore, for both signal and background, in the small $M_{Z'}$ region, ${\hat s}^{1/2}_{\rm min}$ decreases faster in the $Z'+W$ channel than in the $Z'+Z^0$ channel with the decreasing of $M_{Z'}$. As a result, the parton luminosity increases faster in the $Z'+W$ channel than in the $Z'+Z^0$ channel with the decreasing of $M_{Z'}$. Thus, the upper limit on $g_{Z'}$ from the $Z+Z'$ channel increases more rapidly in the small $M_{Z'}$ region with the decreasing of $M_{Z'}$ as shown in Fig. \[fig:Z4channels\].
[99]{}
M. W. Goodman and E. Witten, “Detectability of Certain Dark Matter Candidates,” Phys. Rev. D [**31**]{}, 3059 (1985). C. E. Aalseth, P. S. Barbeau, J. Colaresi, J. I. Collar, J. Diaz Leon, J. E. Fast, N. Fields and T. W. Hossbach [*et al.*]{}, “Search for an Annual Modulation in a P-type Point Contact Germanium Dark Matter Detector,” Phys. Rev. Lett. [**107**]{}, 141301 (2011) \[arXiv:1106.0650 \[astro-ph.CO\]\]. D. S. Akerib [*et al.*]{} \[CDMS Collaboration\], “A low-threshold analysis of CDMS shallow-site data,” Phys. Rev. D [**82**]{}, 122004 (2010) \[arXiv:1010.4290 \[astro-ph.CO\]\]. Z. Ahmed [*et al.*]{} \[CDMS-II Collaboration\], “Results from a Low-Energy Analysis of the CDMS II Germanium Data,” Phys. Rev. Lett. [**106**]{}, 131302 (2011) \[arXiv:1011.2482 \[astro-ph.CO\]\]. E. Aprile [*et al.*]{} \[XENON100 Collaboration\], “First Dark Matter Results from the XENON100 Experiment,” Phys. Rev. Lett. [**105**]{}, 131302 (2010) \[arXiv:1005.0380 \[astro-ph.CO\]\]. G. Angloher, M. Bauer, I. Bavykina, A. Bento, C. Bucci, C. Ciemniak, G. Deuter and F. von Feilitzsch [*et al.*]{}, “Results from 730 kg days of the CRESST-II Dark Matter Search,” Eur. Phys. J. C [**72**]{}, 1971 (2012) \[arXiv:1109.0702 \[astro-ph.CO\]\]. R. Bernabei [*et al.*]{} \[DAMA Collaboration\], “First results from DAMA/LIBRA and the combined results with DAMA/NaI,” Eur. Phys. J. C [**56**]{}, 333 (2008) \[arXiv:0804.2741 \[astro-ph\]\]. E. Aprile [*et al.*]{} \[XENON100 Collaboration\], “Dark Matter Results from 100 Live Days of XENON100 Data,” Phys. Rev. Lett. [**107**]{}, 131302 (2011) \[arXiv:1104.2549 \[astro-ph.CO\]\]. M. Felizardo, T. Morlat, A. C. Fernandes, T. A. Girard, J. G. Marques, A. R. Ramos, M. Auguste and D. Boyer [*et al.*]{}, “First Results of the Phase II SIMPLE Dark Matter Search,” Phys. Rev. Lett. [**105**]{}, 211301 (2010) \[arXiv:1003.2987 \[astro-ph.CO\]\]. \[XENON100 Collaboration\], “Dark Matter Results from 225 Live Days of XENON100 Data,” arXiv:1207.5988 \[astro-ph.CO\]. S. Nussinov, Phys. Lett. B [**165**]{}, 55 (1985). D. E. Kaplan, M. A. Luty and K. M. Zurek, “Asymmetric Dark Matter,” Phys. Rev. D [**79**]{}, 115016 (2009) \[arXiv:0901.4117 \[hep-ph\]\]. G. R. Farrar and G. Zaharijas, “Dark matter and the baryon asymmetry,” Phys. Rev. Lett. [**96**]{}, 041302 (2006) \[hep-ph/0510079\]. R. Kitano and I. Low, “Dark matter from baryon asymmetry,” Phys. Rev. D [**71**]{}, 023510 (2005) \[hep-ph/0411133\]. K. Agashe and G. Servant, “Warped unification, proton stability and dark matter,” Phys. Rev. Lett. [**93**]{}, 231805 (2004) \[hep-ph/0403143\]. H. An, S. -L. Chen, R. N. Mohapatra and Y. Zhang, “Leptogenesis as a Common Origin for Matter and Dark Matter,” JHEP [**1003**]{}, 124 (2010) \[arXiv:0911.4463 \[hep-ph\]\]. J. Shelton and K. M. Zurek, “Darkogenesis: A baryon asymmetry from the dark matter sector,” Phys. Rev. D [**82**]{}, 123512 (2010) \[arXiv:1008.1997 \[hep-ph\]\]. H. Davoudiasl, D. E. Morrissey, K. Sigurdson and S. Tulin, “Hylogenesis: A Unified Origin for Baryonic Visible Matter and Antibaryonic Dark Matter,” Phys. Rev. Lett. [**105**]{}, 211304 (2010) \[arXiv:1008.2399 \[hep-ph\]\]. N. F. Bell, K. Petraki, I. M. Shoemaker and R. R. Volkas, “Pangenesis in a Baryon-Symmetric Universe: Dark and Visible Matter via the Affleck-Dine Mechanism,” Phys. Rev. D [**84**]{}, 123505 (2011) \[arXiv:1105.3730 \[hep-ph\]\]. P. -H. Gu, M. Lindner, U. Sarkar and X. Zhang, “WIMP Dark Matter and Baryogenesis,” Phys. Rev. D [**83**]{}, 055008 (2011) \[arXiv:1009.2690 \[hep-ph\]\]. M. Blennow, B. Dasgupta, E. Fernandez-Martinez and N. Rius, “Aidnogenesis via Leptogenesis and Dark Sphalerons,” JHEP [**1103**]{}, 014 (2011) \[arXiv:1009.3159 \[hep-ph\]\]. B. Dutta and J. Kumar, “Asymmetric Dark Matter from Hidden Sector Baryogenesis,” Phys. Lett. B [**699**]{}, 364 (2011) \[arXiv:1012.1341 \[hep-ph\]\]. Z. Kang, J. Li, T. Li, T. Liu and J. Yang, “Asymmetric Sneutrino Dark Matter in the NMSSM with Minimal Inverse Seesaw,” arXiv:1102.5644 \[hep-ph\]. C. Cheung and K. M. Zurek, “Affleck-Dine Cogenesis,” Phys. Rev. D [**84**]{}, 035007 (2011) \[arXiv:1105.4612 \[hep-ph\]\]. J. March-Russell and M. McCullough, “Asymmetric Dark Matter via Spontaneous Co-Genesis,” JCAP [**1203**]{}, 019 (2012) \[arXiv:1106.4319 \[hep-ph\]\]. M. T. Frandsen, S. Sarkar and K. Schmidt-Hoberg, “Light asymmetric dark matter from new strong dynamics,” Phys. Rev. D [**84**]{}, 051703 (2011) \[arXiv:1103.4350 \[hep-ph\]\]. M. Ibe, S. Matsumoto and T. T. Yanagida, “The GeV-scale dark matter with B-L asymmetry,” Phys. Lett. B [**708**]{}, 112 (2012) \[arXiv:1110.5452 \[hep-ph\]\]. K. Kamada and M. Yamaguchi, “Asymmetric Dark Matter from Spontaneous Cogenesis in the Supersymmetric Standard Model,” Phys. Rev. D [**85**]{}, 103530 (2012) \[arXiv:1201.2636 \[hep-ph\]\]. W. -Z. Feng, P. Nath and G. Peim, “Cosmic Coincidence and Asymmetric Dark Matter in a Stueckelberg Extension,” Phys. Rev. D [**85**]{}, 115016 (2012) \[arXiv:1204.5752 \[hep-ph\]\]. J. Goodman, M. Ibe, A. Rajaraman, W. Shepherd, T. M. P. Tait and H. -B. Yu, “Constraints on Light Majorana dark Matter from Colliders,” Phys. Lett. B [**695**]{}, 185 (2011) \[arXiv:1005.1286 \[hep-ph\]\]. Y. Bai, P. J. Fox and R. Harnik, “The Tevatron at the Frontier of Dark Matter Direct Detection,” JHEP [**1012**]{}, 048 (2010) \[arXiv:1005.3797 \[hep-ph\]\]. J. Goodman, M. Ibe, A. Rajaraman, W. Shepherd, T. M. P. Tait and H. -B. Yu, “Constraints on Dark Matter from Colliders,” Phys. Rev. D [**82**]{}, 116010 (2010) \[arXiv:1008.1783 \[hep-ph\]\]. J. -F. Fortin and T. M. P. Tait, “Collider Constraints on Dipole-Interacting Dark Matter,” Phys. Rev. D [**85**]{}, 063506 (2012) \[arXiv:1103.3289 \[hep-ph\]\]. A. Rajaraman, W. Shepherd, T. M. P. Tait and A. M. Wijangco, “LHC Bounds on Interactions of Dark Matter,” Phys. Rev. D [**84**]{}, 095013 (2011) \[arXiv:1108.1196 \[hep-ph\]\]. I. M. Shoemaker and L. Vecchi, “Unitarity and Monojet Bounds on Models for DAMA, CoGeNT, and CRESST-II,” arXiv:1112.5457 \[hep-ph\]. M. L. Graesser, I. M. Shoemaker and L. Vecchi, “A Dark Force for Baryons,” arXiv:1107.2666 \[hep-ph\]. A. Friedland, M. L. Graesser, I. M. Shoemaker and L. Vecchi, “Probing Nonstandard Standard Model Backgrounds with LHC Monojets,” Phys. Lett. B [**714**]{}, 267 (2012) \[arXiv:1111.5331 \[hep-ph\]\]. H. An, X. Ji and L. -T. Wang, “Light dark matter and $Z'$ dark force at colliders,” arXiv:1202.2894 \[hep-ph\]. J. Alwall, M. Herquet, F. Maltoni, O. Mattelaer and T. Stelzer, “MadGraph 5 : Going Beyond,” JHEP [**1106**]{}, 128 (2011) \[arXiv:1106.0522 \[hep-ph\]\]. P. Meade and M. Reece, “BRIDGE: Branching ratio inquiry / decay generated events,” hep-ph/0703031. M. Cacciari, G. P. Salam and G. Soyez, “FastJet user manual,” Eur. Phys. J. C [**72**]{}, 1896 (2012) \[arXiv:1111.6097 \[hep-ph\]\]. G. L. Bayatian [*et al.*]{} \[CMS Collaboration\], “CMS technical design report, volume II: Physics performance,” J. Phys. G G [**34**]{}, 995 (2007). J. Campbell, K. Ellis and C. Williams, http://mcfm.fnal.gov/. Z. Bern, G. Diana, L. J. Dixon, F. Febres Cordero, S. Hoeche, D. A. Kosower, H. Ita and D. Maitre [*et al.*]{}, “Four-Jet Production at the Large Hadron Collider at Next-to-Leading Order in QCD,” arXiv:1112.3940 \[hep-ph\]. Z. Bern, G. Diana, L. J. Dixon, F. Febres Cordero, S. Hoche, H. Ita, D. A. Kosower and D. Maitre [*et al.*]{}, “Driving Missing Data at Next-to-Leading Order,” Phys. Rev. D [**84**]{}, 114002 (2011) \[arXiv:1106.1423 \[hep-ph\]\].
A. Altheimer, S. Arora, L. Asquith, G. Brooijmans, J. Butterworth, M. Campanelli, B. Chapleau and A. E. Cholakian [*et al.*]{}, J. Phys. G [**39**]{}, 063001 (2012) \[arXiv:1201.0008 \[hep-ph\]\]. \[ATLAS Collaboration\], ATLAS-CONF-2012-147
T. Aaltonen [*et al.*]{} \[CDF Collaboration\], Phys. Rev. D [**79**]{}, 112002 (2009) \[arXiv:0812.4036 \[hep-ex\]\]. G. Aad [*et al.*]{} \[ATLAS Collaboration\], Phys. Lett. B [**708**]{}, 37 (2012) \[arXiv:1108.6311 \[hep-ex\]\]. J. Alitti [*et al.*]{} \[UA2 Collaboration\], Nucl. Phys. B [**400**]{}, 3 (1993). \[ATLAS Collaboration\], ATLAS-CONF-2011-097 S. Chatrchyan [*et al.*]{} \[CMS Collaboration\], “Study of the dijet mass spectrum in $pp \to W +$ jets events at $\sqrt{s}=7$ TeV,” arXiv:1208.3477 \[hep-ex\]. M. R. Buckley, D. Hooper, J. Kopp and E. Neil, “Light Z’ Bosons at the Tevatron,” Phys. Rev. D [**83**]{}, 115013 (2011) \[arXiv:1103.6035 \[hep-ph\]\]. E. Eichten, K. Lane, A. Martin and E. Pilon, Phys. Rev. D [**86**]{}, 074015 (2012) \[arXiv:1206.0186 \[hep-ph\]\]. ÊJ. Beringer [*et al.*]{} Ê\[Particle Data Group Collaboration\], Ê“Review of Particle Physics (RPP),” ÊPhys. Rev. D [**86**]{}, 010001 (2012). Ê
[^1]: Note that due to the large number of degrees of freedom in the quark sector, and our assumption that there is only one species of dark matter, $g_D = g_{Z'} $ only leads to a small invisible width, BR$_{\rm inv} \sim 1/16$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Unified gas kinetic scheme (UGKS) is an asymptotic preserving scheme for the kinetic equations. It is superior for transition flow simulations, and has been validated in the past years. However, compared to the well known discrete ordinate method (DOM) which is a classical numerical method solving the kinetic equations, the UGKS needs more computational resources. In this study, we propose a simplification of the unified gas kinetic scheme. It allows almost identical numerical cost as the DOM, but predicts numerical results as accurate as the UGKS. Based on the observation that the equilibrium part of the UGKS fluxes can be evaluated analytically, the equilibrium part in the UGKS flux is not necessary to be discretized in velocity space. In the simplified scheme, the numerical flux for the velocity distribution function and the numerical flux for the macroscopic conservative quantities are evaluated separately. The simplification is equivalent to a flux hybridization of the gas kinetic scheme for the Navier-Stokes (NS) equations and conventional discrete ordinate method. Several simplification strategies are tested, through which we can identify the key ingredient of the Navier-Stokes asymptotic preserving property. Numerical tests show that, as long as the collision effect is built into the macroscopic numerical flux, the numerical scheme is Navier-Stokes asymptotic preserving, regardless the accuracy of the microscopic numerical flux for the velocity distribution function.'
author:
- Songze Chen
- Zhaoli Guo
- Kun Xu
title: A simplification of the unified gas kinetic scheme
---
Introduction
============
In recent years, multiscale computation is recognized as a powerful tool for studying the interaction on different scales and/or different hierarchies. It has become an active research field and has been applied in many areas, for instance, rarefied gas dynamics, radioactive, plasma, and phonon transfer.
In the rarefied gas dynamics, the physical scales are characterized by the typical geometric length ($L$) and mean free path ($\lambda$). The ratio of these two characteristic lengths is known as the Knudsen number (Kn$=\lambda/L$). When the Knudsen number is much smaller than 1, it is well known that the Navier-Stokes equations are established and govern the fluid behavior. But when the Knudsen number is not too small, the Navier-Stokes equations do not provide accurate physics, and the kinetic equation should be adopted as the governing equation. The simplest kinetic equation for monatomic gas is the BGK equation [@BGK1954], which takes the following form, $$\begin{aligned}
\frac{\partial f}{\partial t} + \mathbf{u}\cdot \nabla f = \frac{g-f}{\tau}, \label{eq:BGK}\end{aligned}$$ where $f$ represents the particle velocity distribution function depending on the location ($\mathbf{x}$), the time ($t$), and particle velocity ($\mathbf{u}$), $g$ denotes the corresponding equilibrium state shown as follows, $$\begin{aligned}
g &=& \mathcal{G}(W) = \rho\left\{\frac{2RT}{\pi}\right\}^{3/2}\exp(-\frac{1}{2RT}(\mathbf{u-U})^2),\label{eq:equilibrium}\\
W &=& (\rho, \rho \mathbf{U}, \rho E)^T. \label{eq:macroVar}\end{aligned}$$ where $\rho$ is the gas density, $T$ is the gas temperature, $R$ is the gas constant, and $E$ denotes the total energy. Since the collision process is conserved, $g$ and $f$ share identical conservative quantities, that is, $$\begin{aligned}
\langle \psi g \rangle = \langle \psi f \rangle, \ \psi = (1, \mathbf{u}, \frac{1}{2}\mathbf{u}^2)^T.\end{aligned}$$ The symbol $\langle f \rangle$ is defined as, $\langle f \rangle = \int_{-\infty}^{+\infty} f d\mathbf{u}$.
Typically, the flow regimes can be categorized into four regimes: continuum flow ($\mbox{Kn} < 0.001$), slip flow ($0.001 < \mbox{Kn} < 0.1$), transition flow ($0.1 < \mbox{Kn} < 10$), and free molecular flow ($\mbox{Kn} > 10$). The Navier-Stokes equations are only validated in the continuum flow regime, and can be further extended to solve a small portion of slip flow problems by considering slip boundary condition. For the other flow regimes, the kinetic theory, including Boltzmann equation and other kinetic equations, must be adopted to take account of the delicate molecular motion. For example, when a vehicle travels through the atmosphere, the density of ambient gas changes dramatically. In another scenario, the gas is driven by the temperature gradient, goes through different chambers in multistage Knudsen pump. The mean free path enlarges as the density decreases, and the Knudsen number enlarges accordingly. The Navier-Stokes equations fail to predict the flow fields somewhere in these two applications. Thus the kinetic equation is necessary to take over in the domain where NS equations break down. An intuitive idea is the domain decomposition method, in which the flow field is solved on different subdomains by appropriate numerical solvers, the Navier-Stokes solvers or the kinetic solvers. But the major difficulty of this method is the information exchange in the buffer zone or overlap region between two numerical methods on different scales. Moreover, in many multiscale problems, the Knudsen number varies both in space and time. Single domain decomposition is incapable for such problems.
Another promising multiscale approach is the asymptotic preserving scheme that can recover large scale system from small scale simulation uniformly[@Jin2012]. When the Knudsen number goes to zero, the numerical scheme for the kinetic equation should be an analogue of the analytical asymptotic analysis of the kinetic equation. In 1991, Coron and Perthame [@Coron1991] proposed a scheme which is asymptotic preserving in terms of Euler equations. After this study, variants AP schemes for the rarefied gas system are proposed in the last two decades, including implicit scheme for the collision terms [@Pieraccini2007; @Filbet2011], penalization method [@Gabetta1997; @Filbet2010; @Yan2013], exponential relaxation method [@Coron1991; @Dimarco2011], unified gas kinetic schemes [@Xu2010; @ugks1_1; @ugks2; @ugks3], and discrete unified gas kinetic scheme [@guo2013discrete; @guo2014discrete; @wang2015comparative] etc.
From the previous literatures, two key ingredients of the asymptotic preserving scheme can be concluded. The first key ingredient is the special treatment of the collision term (RHS of Eq.(\[eq:BGK\])). The traditional DOM solves the collision term explicitly. It is always restricted by the Knudsen number, and cannot obtain physical solution in near continuum and continuum flow regime unless using infinite computation resources. Actually, the stiffness of the collision term due to the small parameter makes the explicit schemes for the kinetic equation useless in the continuum flow regime. Therefore, the exponential collision solver [@Coron1991; @Dimarco2011; @li2014exponential] and implicit treatment of the collision term [@Gabetta1997; @Filbet2010; @Caflisch1997] are proposed to remove the stiffness of the collision term.
The other ingredient of the AP scheme is that the completed kinetic equation must be employed to solve the numerical flux at cell interface and the body force inside a cell in order to attain the correct Navier-Stokes limit [@chen2015comparative]. Bennoune *et al*.[@Bennoune2008] investigated the influence of the implicit schemes for the collision term, and found that, if operator splitting method is employed to evaluate the collision term, the resulting distribution function will be too close to the equilibrium state, thus the schemes cannot attain the physical viscosity. Chen and Xu[@chen2015comparative] studied the Navier-Stokes asymptotic preserving property and concluded that not only the body force needs both convection and collision terms, the numerical fluxes also need these two terms in order to obtain the correct Navier-Stokes limit.
In the early stage, the operator splitting method is employed to simplify the numerical scheme. The governing equation is modified for different purposes. For solving the interfacial numerical flux, the convection term is reserved, but the source term is discarded. Governing equation becomes, $$\begin{aligned}
\frac{\partial f}{\partial t} + \mathbf{u}\cdot \nabla f = 0. \label{eq:collisionless}\end{aligned}$$ For solving the body force, only the source term is reserved, while convection term is abandoned. $$\begin{aligned}
\frac{\partial f}{\partial t} = \frac{g-f}{\tau}. \label{eq:onlyCollision}\end{aligned}$$ It is found that the use of incomplete governing equation will induce large error when simulate continuum flows [@chen2015comparative; @Bennoune2008].
In 2010, Xu *et al* proposed the unified gas kinetic scheme, which couples the collision and convection terms by a local analytical solution of the complete governing equation (\[eq:BGK\]). When approaching the Navier-Stokes limit, the numerical flux turns to the Chapman-Enskog expansion gradually. Therefore, the collision and free transport are all built into the numerical flux and the numerical body force. Theoretically, UGKS can recover the NS limit and Euler limit. With the same spirit, Guo *et al.* proposed a discrete unified gas kinetic scheme (DUGKS) which replaces the local integral solution by a discrete time integral. Although, the discrete approximation is adopted, the DUGKS still possess the NS AP property.
Theoretically, the unified schemes can recover the continuum regime. However, quadrature which accounts for the numerical integral in discrete velocity space is an obstacle for attaining correct asymptotic limit in the continuum flow regime. As we know, in the free molecular flow regime, the Newton-Cotes quadrature is more suitable compared to the Gauss-Hermite quadrature because the distribution function deviates largely from equilibrium state. But in the continuum flow, the Gauss-Hermite quadrature is always used due to its high accuracy for the integral of exponential function. If different quadratures are employed, massive interpolations will be needed to exchange data on different velocity points. And it will introduce additional numerical errors. As a result, it is inconvenient to change the quadrature method automatically according to the flow condition. Therefore, unsuitable quadrature might induce large error or large computational cost in a unified AP scheme.
In practice, a unified scheme is still burdensome to reproduce the continuum flow limit. On the other hand, the Navier-Stokes equations can be derived from the kinetic equation, and the traditional numerical schemes for the Navier-Stokes equations are highly efficient. Why do we derive asymptotic limit from massive high dimensional distribution function in numerical scheme? If we use more degree of freedom to simulate a lower dimensional problem, then there must be something can be simplified. In this study, we revisit the unified gas kinetic scheme and estimate the contribution of each term in asymptotic limit. For the part which can be calculated by traditional Navier-Stokes solver, we use analytical results instead of the discrete velocity representation and propose several simplification of the UGKS.
The article is organized as follows. In Sec. 2, the unified gas kinetic scheme is introduced briefly; in Sec. 3, we analyze the behavior of the UGKS in different flow regime and propose three different simplification strategies; in Sec. 4, the numerical discretization and the boundary condition are introduced; in Sec. 5, numerical comparisons are provided, from which the key ingredient of the unified scheme and the best simplification strategy are identified for the industrial applications. Finally, we conclude this study in Sec. 6.
**Remark:** We shall emphasize the terminologies used above. As the operator splitting method has been prevailing for many years, the numerical flux is always correlated with the convection term (Eq.(\[eq:collisionless\])), and the body force is correlated with the collision term (Eq.(\[eq:onlyCollision\])). However, during a finite time interval, the interfacial flux is not only influenced by the convection term, but is also influenced by collision term, and so is the body force. In this paper, we do not use the ’convection’ and ’collision’ to illustrate the two procedures in the numerical scheme. Actually, considering a control volume, the quantities changing inside the control volume equals to the interfacial flux through the interface plus the body force exerted on the volume. The interfacial flux and the body force are only geometric concepts in the finite volume schemes. Thus, the terminologies, ’interfacial flux’ and ’body force’, are precise to describe the two procedures in the unified schemes.
Unified gas kinetic scheme
==========================
In this paper, we only consider the finite volume schemes. We will fix the numerical method for the body force, and compare different interfacial fluxes. Before discussing the AP property of the UGKS, we briefly recall the numerical flux of the conventional DOM for the kinetic equation. As mentioned in the introduction, the collisionless kinetic equation (Eq.(\[eq:collisionless\])) is taken as the governing equation to evaluate the numerical flux. The solution at the interface ($\mathbf{x}=0$) is then, $$\begin{aligned}
f(0,t,\mathbf{u}) = f(-t\mathbf{u},0,\mathbf{u}).\end{aligned}$$ Considering first order spatial expansion, we have, $$\begin{aligned}
f_{dom}(0,t,\mathbf{u}) = f(0,0,\mathbf{u}) - t\mathbf{u}\cdot \nabla f.\end{aligned}$$ The numerical flux for the distribution function is then, $$\begin{aligned}
\mathcal{F}_{dom} &=& \int_{0}^{\Delta t} u_k f_{dom}(0,t,\mathbf{u}_k) dt = u_k (\Delta t f_k-\frac{1}{2}\Delta t^2 \mathbf{u}_k\cdot \nabla f_k). \label{eq:domflux}\end{aligned}$$ For simplicity, we ignore the arguments of the distribution function $f$ and assume that $u$ is aligned with the normal direction of the cell interface. The numerical flux of the DOM is very simple, only the numerical fluxes for the distribution function are considered in the DOM. In order to compare with the UGKS, the equivalent numerical fluxes for macroscopic variables are derived by taking the moments of the numerical microscopic flux, $$\begin{aligned}
\mathcal{F}^W_{dom} &=& \langle \mathcal{F}_{dom} \rangle_k = \sum_k u_k \psi_k (\Delta t f_k-\frac{1}{2}\Delta t^2 \mathbf{u}_k\cdot \nabla f_k), \label{eq:domfluxW}\end{aligned}$$ where the superscript $W$ denotes the macroscopic flux. The symbol $\langle f \rangle_k$ denotes taking moments of $f$ in discrete velocity space, namely, the summation $\langle f \rangle_k = \sum_k \omega_k f_k$, where $\omega_k$ is the weight function at velocity point $\mathbf{u}_k$. The mechanism of the above formulations for the macroscopic fluxes is equivalent to kinetic flux vector splitting (KFVS) method for the Euler equations.
The unified gas kinetic scheme is an asymptotic preserving scheme benefiting from the local analytical solution of kinetic equation. Integrating along the characteristic of the BGK equation (Eq.(\[eq:BGK\])), a local analytical solution can be derived. $$\begin{aligned}
f(0,t,\mathbf{u}) &=& e^{-t/\tau}f(-\mathbf{u}t,0,\mathbf{u}) \nonumber\\
&& + \frac{1}{\tau}\int_0^{t}g(-\mathbf{u}(t-t'),t', \mathbf{u})e^{-(t-t')/\tau}dt'. \label{eq:localsolution}\end{aligned}$$ The forepart is the non-equilibrium part. When the system approaches equilibrium, $e^{-t/\tau}$ will become zero asymptotically, i.e., the non-equilibrium contribution will vanish. Meanwhile, the second term on the right hand side, which represents the equilibrium part, will dominate.
Suppose, after the numerical reconstruction, the physical quantities are linearly distributed around the cell, and are expressed as follows, $$\begin{aligned}
f(\mathbf{x},0,\mathbf{u}) &=& f(0,0,\mathbf{u}) + \mathbf{x} \cdot \nabla f, \\
g(\mathbf{x},t,\mathbf{u}) &=& g(0,0,\mathbf{u}) + \mathbf{x} \cdot \nabla g+ g_t t.\end{aligned}$$ Substitute these formulas into the analytical solution, $$\begin{aligned}
f_{ugks}(0,t,\mathbf{u}) &=& e^{-t/\tau}(f(0,0,\mathbf{u}) - t\mathbf{u}\cdot \nabla f) + (1-e^{-t/\tau})g(0,0,u) \nonumber\\
&& +(-\tau+(\tau+t)e^{-t/\tau}) \mathbf{u}\cdot \nabla g + (t-\tau+\tau e^{-t/\tau}) g_t. \nonumber\end{aligned}$$ This is the distribution function at the cell interface. The numerical microscopic flux is, $$\begin{aligned}
\mathcal{F}_{ugks} &=& \int_{0}^{\Delta t} u_k f_{ugks}(0,t,\mathbf{u}_k) dt. \label{eq:ugksflux0}\end{aligned}$$ Then taking moments of above solution, we get the numerical macroscopic flux at cell interface. $$\begin{aligned}
\mathcal{F}^W_{ugks} &=& <\mathcal{F}_{ugks}>_k.\end{aligned}$$
As a standard finite volume method, the quantities inside a cell are updated by considering both the numerical flux and the body force. Because of the conservation constraint on the collision term, the source terms for conservative variables are zero, $$\begin{aligned}
\langle \psi(f-g) \rangle = 0 \quad \mbox{or} \quad \langle \psi(f-g) \rangle_k = 0.\end{aligned}$$ Therefore, the conservative variables can be updated by only taking account of the numerical macroscopic flux, $$\begin{aligned}
W^{n+1} = W^{n} - \nabla\cdot \mathcal{F}^W. \label{eq:macroConserve}\end{aligned}$$ After obtaining $W^{n+1}$, the equilibrium state $g^{n+1}$ is known through the formula (Eq.(\[eq:equilibrium\])). The time discretization of the kinetic equation (Eq.(\[eq:BGK\])) can be written as,
$$\begin{aligned}
\frac{f_{k}^{n+1}-f_{k}^{n}}{\Delta t} + \frac{1}{\Delta t}\nabla\cdot \mathcal{F}_{k} = \frac{g_{k}^{n+1} - f_{k}^{n+1}}{\tau}.\end{aligned}$$
Then solve the distribution function at $n+1$ step, $$\begin{aligned}
f_{k}^{n+1} = \frac{\tau}{\tau+\Delta t}(f_{k}^{n}-\nabla\cdot\mathcal{F}_{k}) + \frac{\Delta t}{\tau+\Delta t}g^{n+1}_{k}. \label{eq:lastTimeStepCE}\end{aligned}$$
As shown above, the convection term $\nabla \mathcal{F}$ is also considered when evaluating the body force. The strong coupling of collision and convection term in the scheme is the main distinguishing feature compared to the operator splitting DOM. The UGKS take the complete equation to evaluate the numerical flux and the body force. This is the reason why the UGKS is an NS AP scheme.
In this study, we use ’DOM’ to denote the numerical scheme which couples the collisionless flux (Eq.(\[eq:domflux\])) and the implicit time discretization (Eq.(\[eq:lastTimeStepCE\],\[eq:domfluxW\])) for the body force. The time discretization (Eq.(\[eq:lastTimeStepCE\])) is adopted as a common ingredient of all the numerical schemes compared in this paper.
The numerical fluxes stem from the equilibrium and non-equilibrium parts
------------------------------------------------------------------------
The numerical fluxes of the unified gas kinetic scheme are composed of the equilibrium and non-equilibrium terms. The competition of all these terms determines the asymptotic behavior of the numerical schemes in different flow regimes. This issue has been discussed by Mieussens [@Mieussens2013] for the UGKS of radiative transfer equation. We will investigate every term in detail and deduce the asymptotic coefficient of each term. The numerical flux (Eq.(\[eq:ugksflux0\])) can be further unfolded as follows, $$\begin{aligned}
\mathcal{F}_{ugks} &=& u_k \{\gamma_0^{ugks}f_k + \gamma_1^{ugks} \mathbf{u}_k\cdot \nabla f_k \nonumber\\
&& +\gamma_2^{ugks}g_k +\gamma_3^{ugks} (\mathbf{u}_k\cdot \nabla g_k + \frac{\partial g_k}{\partial t}) + \gamma_4^{ugks} \frac{\partial g_k}{\partial t}\}, \label{eq:ugksflux}\\
\mathcal{F}^W_{ugks} &=& \gamma_0^{ugks}\langle u \psi f \rangle_k + \gamma_1^{ugks}\langle u \psi \mathbf{u}\cdot \nabla f \rangle_k \nonumber\\
&& +\gamma_2^{ugks}\langle u \psi g \rangle_k + \gamma_3^{ugks}\langle u \psi (\mathbf{u}\cdot \nabla g +g_t) \rangle_k + \gamma_4^{ugks}\langle u \psi g_t \rangle_k.\label{eq:ugksfluxW}\end{aligned}$$ For the sake of simplicity, we define the coefficients in the UGKS flux as follows, $$\begin{aligned}
\gamma^{ugks}_0 &=& \tau(1-e^{-\beta}), \nonumber\\
\gamma^{ugks}_1 &=& - \tau(-\Delta t e^{-\beta}+\tau-\tau e^{-\beta}), \nonumber\\
\gamma^{ugks}_2 &=& \Delta t - \tau(1-e^{-\beta}), \\
\gamma^{ugks}_3 &=& \tau(-\Delta t e^{-\beta} - 2\tau e^{-\beta} +2\tau-\Delta t), \nonumber\\
\gamma^{ugks}_4 &=& (\frac{1}{2}\Delta t^2 + \tau((\Delta t +\tau) e^{-\beta} -\tau)), \nonumber\end{aligned}$$ where $\beta$ is defined as the ratio of the time step $\Delta t$ to the relaxation time $\tau$, namely, $\beta = \Delta t/\tau$. The first two terms on the right hand side of the Eq.(\[eq:ugksflux\]) are the non-equilibrium parts which are deduced from the non-equilibrium initial condition at the beginning of the time step. The last three terms on the right hand side stemming from the collision term represent the Navier-Stokes flux. As shown above, the non-equilibrium part $f_0$ does not vanish directly when $\frac{\Delta t}{\tau}\rightarrow +\infty$. A small term ($O(\tau)$) still influences the numerical fluxes. Xu provided a profound perspective of the asymptotic behavior of the numerical flux [@Xu2001]. He showed that proper initial condition (Chapman-Enskog expansion) of each time step should be assumed to deduce correct numerical flux in the continuum flow regimes. Following this idea, we consider a specific expression of the non-equilibrium part. For the sake of the implicit discretization (Eq.(\[eq:lastTimeStepCE\])) of the collision term, the following assumption seems natural and rational. The initial condition $f$ deviates from the equilibrium by $O(\tau)$, namely, $$\begin{aligned}
f = g+O(\tau). \label{eq:assumptionO}\end{aligned}$$ After some derivations, we can get more precise estimation for the initial condition [@chen2015comparative], that is, $$\begin{aligned}
\begin{array}{ccl}
f(0,0,\mathbf{u}) &=& g(0,0,\mathbf{u}) - \tau(\mathbf{u}\cdot \nabla g + g_t) + O(\tau\Delta t), \\
\end{array} \label{eq:assumptionCE}\end{aligned}$$ where we choose the approximate Chapman-Enskog expansion (Eq.(\[eq:assumptionCE\])) as the initial condition for the UGKS. Then substituting the estimation (Eq.(\[eq:assumptionCE\])) into the Eq.(\[eq:ugksflux\]) and Eq.(\[eq:ugksfluxW\]), the numerical flux becomes, $$\begin{aligned}
\mathcal{F}_{ugks} &=& u_k(\Delta t g_k - \Delta t \tau(\mathbf{u}_k\cdot \nabla g_k + \frac{\partial g_k}{\partial t}) + \frac{\Delta t^2}{2}\frac{\partial g_k}{\partial t}) +O(\tau^2), \nonumber \\
\mathcal{F}^W_{ugks} &=& \langle u\psi(\Delta t g - \Delta t \tau(\mathbf{u}\cdot \nabla g + \frac{\partial g}{\partial t}) + \frac{\Delta t^2}{2}\frac{\partial g}{\partial t}) \rangle_k + O(\tau^2).\label{eq:ugksCE}\end{aligned}$$ The Chapman-Enskog expansion for the Navier-Stokes equation is exactly recovered. Please note that, $\beta$ is not required to approach zero as we derive the Chapman-Enskog expansion. As we know, the numerical scheme must converge as the time step goes to zero. In this sense, the asymptotic behavior when $\tau\rightarrow 0,\ \Delta t\rightarrow 0$, and $\beta$ is finite, is more important to the numerical scheme.
Under the more precise assumption (Eq.(\[eq:assumptionCE\])), the estimation of the numerical fluxes in the DOM is written as, $$\begin{aligned}
\mathcal{F}_{dom} &=& u_k(\Delta t g_k - \Delta t(\tau+\frac{\Delta t}{2})(\mathbf{u}_k\cdot \nabla g_k + \frac{\partial g_k}{\partial t}) + \frac{\Delta t^2}{2}\frac{\partial g_k}{\partial t}) +O(\tau^2), \nonumber\\
\mathcal{F}^W_{dom} &=& \langle u \psi(\Delta t g - \Delta t(\tau+\frac{\Delta t}{2})(\mathbf{u}\cdot \nabla g + \frac{\partial g}{\partial t}) + \frac{\Delta t^2}{2}\frac{\partial g}{\partial t})\rangle_k + O(\tau^2).\label{eq:domCE}\end{aligned}$$ The equivalent viscosity in Eq.(\[eq:domCE\]) is enlarged by the free streaming. We use $\alpha_{dom}$ to denote the enlarging factor, which is $$\alpha_{dom} = (\tau+\Delta t/2)/\tau = 1+\frac{1}{2}\beta.$$ It is close to KFVS-NS for a discontinuous flow. When $\beta$ varies from 0 to $\infty$, $\alpha_{dom}$ diverges . The enlarged viscosity is an analogue to the numerical viscosity in the lattice Boltzmann method [@chen1992recovery] before the remedy of the viscosity.
Simplification of the unified gas kinetic scheme
================================================
In the UGKS fluxes (Eq.(\[eq:ugksflux\],\[eq:ugksfluxW\])), the last three terms which stem from the collision term, are also discretized in velocity space. Therefore, it takes huge computational resources compared to the traditional Navier-Stokes solvers. In fact, more than half portion of computation resource is taken to evaluate the equilibrium part. Actually, the quadratures£¬ $\langle g \rangle_k, \langle \nabla g\rangle_k$, and $\langle g_t \rangle_k$ are only approximation of $\langle g \rangle, \langle \nabla g\rangle$, and $\langle g_t \rangle$. The quadrature of $g$ and its derivatives can be calculated analytically, for instance, $\langle g\rangle = (\rho, \rho U, \rho E)^T$. If the quadratures of the equilibrium state $g$ and its derivatives are handled in traditional way in terms of analytical macroscopic flux [@Xu2001], the unified scheme will be much more efficient. Therefore, we propose the first simplification (S1), that is, using traditional DOM to calculate the flux for distribution function and using the macroscopic gas kinetic scheme [@Xu2001] to evaluate the last three terms in Eq.(\[eq:ugksfluxW\]), $$\begin{aligned}
\mathcal{F}_{s1} &=& u_k (\Delta t f_{k} - \frac{1}{2}\Delta t^2 \mathbf{u}_k\cdot \nabla f_{k}), \nonumber\\
\mathcal{F}_{s1}^W &=& \gamma_0^{s1}\langle u \psi f \rangle_k + \gamma_1^{s1}\langle u \psi \mathbf{u}\cdot \nabla f \rangle_k \nonumber\\
&& +\gamma_2^{s1}\langle u \psi g \rangle + \gamma_3^{s1}\langle u \psi (\mathbf{u}\cdot \nabla g +g_t) \rangle + \gamma_4^{s1}\langle u \psi g_t \rangle. \label{eq:s1flux} \\
\gamma_i^{s1} &=& \gamma_i^{ugks}, \quad i = 0,1,2,3,4. \nonumber\end{aligned}$$ Compared to the numerical macroscopic flux of the UGKS (Eq.(\[eq:ugksfluxW\])), the equilibrium part is solved analytically (note the different symbols $\langle \cdot \rangle$ and $\langle \cdot \rangle_k$), and the numerical microscopic flux (Eq.(\[eq:ugksflux\])) is replaced by the traditional DOM (Eq.(\[eq:domflux\])). With the assumption (Eq.(\[eq:assumptionCE\])), if the difference between $\langle\cdot\rangle$ and $\langle\cdot\rangle_k$ is ignored, the numerical microscopic flux becomes, $$\begin{aligned}
\mathcal{F}_{s1} &=& u_k (\Delta t g_{k} - (\tau+\frac{1}{2}\Delta t)\Delta t(\mathbf{u}_k\cdot \nabla g_k + \frac{\partial g_k}{\partial t}) + \frac{1}{2}\Delta t^2 \frac{\partial g_k}{\partial t})+ O(\tau^2), \nonumber\\
\mathcal{F}^W_{s1} &=& \langle u\psi(\Delta t g - \Delta t \tau(\mathbf{u}\cdot \nabla g + \frac{\partial g}{\partial t}) + \frac{\Delta t^2}{2}\frac{\partial g}{\partial t}) \rangle + O(\tau^2).\end{aligned}$$ If the quadrature $(\langle\cdot\rangle_k)$ is accurate, the numerical macroscopic flux of the S1 scheme is identical to the macroscopic flux of the UGKS. Only the flux for the distribution function is different. We will present some numerical comparisons to demonstrate that the inaccurate microscopic numerical flux has very little influence to the NS AP property of the numerical scheme. This simplification only reduces the computational cost, but the formula and the coding are still complicated. Hence, we propose a second simplified method (S2), which is barely a combination of the DOM and the gas kinetic scheme for the Navier-Stokes equations. The numerical fluxes are given as follows. $$\begin{aligned}
\mathcal{F}_{s2} &=& u_k (\Delta t f_{k} - \frac{1}{2}\Delta t^2 \mathbf{u}_k\cdot \nabla f_{k}), \nonumber \\
\mathcal{F}_{s2}^W &=& e^{-\beta} \langle u \psi (\Delta t f - \frac{1}{2}\Delta t^2 \mathbf{u}\cdot \nabla f) \rangle_k \label{eq:s2flux}\nonumber\\
&& + (1-e^{-\beta}) \langle u\psi(\Delta t g_0 - \Delta t\tau(\mathbf{u}\cdot \nabla g + \frac{\partial g}{\partial t}) + \frac{1}{2}\Delta t^2 \frac{\partial g}{\partial t}) \rangle \label{eq:s2flux}\end{aligned}$$
formula $\beta \rightarrow \infty$ $\beta \rightarrow 0$
------------------- ------------------------------------------------------------------- -------------------------------------------------- -----------------------------------------------------------------
$\gamma_0^{ugks}$ $\tau(1-e^{-\beta})$ $\tau + O(e^{-\beta})$ $\Delta t -\frac{\Delta t^2}{2\tau} + O(\beta^2)$
$\gamma_0^{s2}$ $\Delta t e^{-\beta}$ $O(e^{-\beta})$ $\Delta t -\frac{\Delta t^2}{\tau} + O(\beta^2)$
$\gamma_1^{ugks}$ $- \tau(-\Delta t e^{-\beta}+\tau-\tau e^{-\beta})$ $-\tau^2 + O(e^{-\beta})$ $-\frac{\Delta t^2}{2} + \frac{\Delta t^3}{3\tau} + O(\beta^2)$
$\gamma_1^{s2}$ $-\frac{1}{2}\Delta t^2 e^{-\beta}$ $O(e^{-\beta})$ $-\frac{\Delta t^2}{2} + \frac{\Delta t^3}{2\tau} + O(\beta^2)$
$\gamma_2^{ugks}$ $\Delta t - \tau(1-e^{-\beta})$ $\Delta t - \tau + O(e^{-\beta})$ $\frac{\Delta t^2}{2\tau} + O(\beta^2)$
$\gamma_2^{s2}$ $(1-e^{-\beta})\Delta t$ $\Delta t + O(e^{-\beta})$ $\frac{\Delta t^2}{\tau}+O(\beta^2)$
$\gamma_3^{ugks}$ $\tau(2\tau-\Delta t - (\Delta t + 2\tau) e^{-\beta} )$ $ 2\tau^2 -\tau\Delta t + O(e^{-\beta})$ $-\frac{\Delta t^3}{6\tau} + O(\beta^2)$
$\gamma_3^{s2}$ $-\tau\Delta t(1-e^{-\beta})$ $-\tau\Delta t + O(e^{-\beta})$ $\Delta t^2-\frac{\Delta t^3}{2\tau} + O(\beta^2)$
$\gamma_4^{ugks}$ $\frac{1}{2}\Delta t^2 + \tau((\Delta t +\tau) e^{-\beta} -\tau)$ $\frac{1}{2}\Delta t^2 - \tau^2 + O(e^{-\beta})$ $\frac{\Delta t^3}{3\tau} + O(\beta^2)$
$\gamma_4^{s2}$ $\frac{1}{2}\Delta t^2(1-e^{-\beta})$ $\frac{1}{2}\Delta t^2+O(e^{-\beta})$ $\frac{\Delta t^3}{2\tau} + O(\beta^2)$
: The coefficients of the numerical flux in UGKS and S2 scheme, where $\beta = \Delta t/\tau$.[]{data-label="tab:coe"}
This method is very simple. We can easily combine two existing flux solvers to construct a unified scheme for gas kinetic equation. Assume that the initial condition at the beginning of the time step satisfies the near equilibrium assumption, namely, Eq.(\[eq:assumptionCE\]) is applied. The numerical flux of the second simplified method becomes, $$\begin{aligned}
\mathcal{F}_{s2} &=& u_k (\Delta t g_{k} - (\tau+\frac{\Delta t}{2})\Delta t(\mathbf{u}_k\cdot \nabla g_k + \frac{\partial g_k}{\partial t}) + \frac{1}{2}\Delta t^2 \frac{\partial g_k}{\partial t})+ O(\tau^2), \nonumber\\
\mathcal{F}^W_{s2} &=& \langle u\psi(\Delta t g - \Delta t (\tau-\frac{\Delta t}{2} e^{-\beta})(\mathbf{u}\cdot \nabla g + \frac{\partial g}{\partial t}) + \frac{1}{2}\Delta t^2\frac{\partial g}{\partial t}) \rangle+ O(\tau^2). \nonumber\end{aligned}$$ Obviously, in the continuum flow regime, this simplification is accurate enough to lead to the Navier-Stokes numerical flux. The viscosity of the S2 scheme is enlarged by a factor, $$\begin{aligned}
\alpha_{s2} = (\tau-\Delta t e^{-\beta}/2)/\tau = 1-\frac{1}{2}\beta e^{-\beta}.\end{aligned}$$ It only varies inside the interval $[1-e^{-1}/2,1]$. The minimum is attained when $\beta = 1$. The table \[tab:coe\] compares the coefficients of the UGKS and the second simplified scheme. As shown in the second column, the formulas are apparently different when $\beta$ has a finite value. When $\beta$ goes to infinity and $\tau$ goes to zero, i.e., in the continuum flow regime, the coefficients are identical up to $O(\tau)$. The S2 scheme approaches the equilibrium state a little faster than the UGKS because the coefficients of the non-equilibrium part, $\gamma_0^{s2}$ and $\gamma_1^{s2}$, approach to zero more rapidly. Consider the free molecular flow limit, namely, $\beta$ goes to zero and $\tau$ goes to infinity. The coefficients of the UGKS and the S2 scheme are identical up to $O(\beta)$, except $\gamma_3^{s2}$. It deviates from $\gamma_3^{ugks}$ in free molecular flow regime largely. This means the simple combination cannot recover the free molecular flow regime. We will also find large discrepancy generated from $\gamma_3^{s2}$ in the numerical comparison section.
![The enlarging factor of the viscosity for the simplified scheme (S2, S3).[]{data-label="fig:enlargingFactor"}](figures/enlargingFactor.eps)
Therefore, we propose a third simplified method (S3) which modified the coefficient in front of the Navier-Stokes viscous term. The basic idea is to construct a coefficient $\gamma_3^{s3}$ which can preserve the asymptotic limit of $\gamma^{ugks}_3$. The third simplified method is, $$\begin{aligned}
\mathcal{F}_{s3} &=& u_k (\Delta t f_{k} - \frac{1}{2}\Delta t^2 \mathbf{u}_k\cdot \nabla f_{k}) \nonumber\\
\mathcal{F}_{s3}^W &=& e^{-\beta}\langle u \psi (\Delta t f - \frac{1}{2}\Delta t^2 \mathbf{u}\cdot \nabla f)\rangle_k \nonumber\\
&& + (1-e^{-\beta}) \langle u\psi(\Delta t g_0 - \Delta t r_{\tau} \tau (\mathbf{u}\cdot \nabla g + \frac{\partial g}{\partial t}) + \frac{1}{2}\Delta t^2 \frac{\partial g}{\partial t}) \rangle \label{eq:s3flux}\\
r_{\tau} &=& \frac{1-e^{-\beta}-\frac{\Delta t}{2\tau}(1+e^{-\beta})e^{-\beta}}{1-e^{-\beta}}. \label{eq:rTau}\end{aligned}$$ The coefficient $\gamma_3^{s3}$ becomes, $$\begin{aligned}
\gamma_3^{s3} &=& -\tau\Delta t(1-e^{-\beta}-\frac{\Delta t}{2\tau}(1+e^{-\beta})e^{-\beta}), \nonumber\\
\gamma_3^{s3} &=& -\tau\Delta t + O(e^{-\beta}),\quad \mbox{when}\ \frac{\Delta t}{\tau}\rightarrow \infty \nonumber\\
\gamma_3^{s3} &=& -\frac{\Delta t^3}{\tau}+O(\beta^2),\quad \mbox{when}\ \frac{\Delta t}{\tau}\rightarrow 0 \nonumber\end{aligned}$$ The only difference from the second simplified method is that, the coefficient in front of the viscous term is multiplied by a factor $r_{\tau}$. Therefore, the third method can also be taken as a simple combination between the DOM and the Navier-Stokes solver. As we can see, the coefficient $\gamma_3^{s3}$ has the same limit in free molecular flow regime up to $O(\beta)$. Then considering the continuum flow regime, with the assumption Eq.(\[eq:assumptionCE\]), the third simplified method becomes, $$\begin{aligned}
\mathcal{F}_{s3} &=& u_k (\Delta t g_{k} - (\tau+\frac{1}{2}\Delta t)\Delta t(\mathbf{u}_k\cdot \nabla g_k + \frac{\partial g_k}{\partial t}) + \frac{1}{2}\Delta t^2 \frac{\partial g_k}{\partial t})+ O(\tau^2), \nonumber\\
\mathcal{F}^W_{s3} &=& \langle u\psi(\Delta t g - \Delta t \tau(\mathbf{u}\cdot \nabla g + \frac{\partial g}{\partial t}) + \frac{\Delta t^2}{2}\frac{\partial g}{\partial t}) \rangle+ O(\tau^2) \nonumber\\
&&- \langle u\psi(\Delta t (\frac{1}{2}\Delta t e^{-\beta}+\tau(1-e^{\beta})(r_{\tau}-1))(\mathbf{u}\cdot \nabla g + \frac{\partial g}{\partial t}))\rangle \\
&=& \langle u(\Delta t g - \Delta t (\tau-\frac{\Delta t}{2} e^{-2\beta})(\mathbf{u}\cdot \nabla g + \frac{\partial g}{\partial t}) + \frac{\Delta t^2}{2}\frac{\partial g}{\partial t}) \rangle+ O(\tau^2). \nonumber\end{aligned}$$ For the third simplified method, the equivalent viscosity is enlarged by, $$\begin{aligned}
\alpha_{s3} = (\tau-\Delta t e^{-2\beta}/2)/\tau =1-\frac{1}{2}\beta e^{-2\beta}.\end{aligned}$$ It only varies inside the interval $[1-e^{-1}/4,1]$. The minimum is attained when $\beta = 1/2$. Figure \[fig:enlargingFactor\] shows the enlarging factor $\alpha$ versus $\beta$. The numerical flux of the UGKS is based on the analytical solution. Therefore, its viscosity is unchanged in the second order temporal discretization (Eq.(\[eq:ugksCE\])), i.e., $\alpha_{ugks}=1$. The simplified schemes somehow modify the viscous coefficient. As shown in the figure, the S3 scheme is more accurate than the S2 scheme in terms of the viscosity coefficient.
We analyze the behavior of these simplified numerical schemes. The S1 scheme replaces the quadrature related to the equilibrium state by the analytical solution. Although it has correct asymptotic limits and less computational cost, the scheme is still complicated in terms of coding. The S2 scheme is a simple combination of Navier-Stokes solver and traditional DOM. It cannot reproduce the free molecular flow regime. The S3 scheme has correct asymptotic limits in free molecular flow regime, and also in the continuum flow regime. For the transition flow regime, the coefficients are apparently different from the analytical solution. We will use the numerical experiment to investigate the performance of different simplifications.
Numerical discretization
========================
The previous section introduced the numerical flux expression in terms of time. Several simplified numerical fluxes are constructed based on the unified gas kinetic scheme. In this section the spatial discretization and the boundary condition are provided.
Spatial discretization
----------------------
The value and its spatial derivative of a certain quantity are needed in the expressions of the numerical flux (for example Eq. (\[eq:ugksflux\])). For the velocity distribution function, we adopt the third order WENO to interpolate its value at the cell interface $(i+1/2)$, where $i$ denotes the index along the interpolation direction. The formula is given below, $$\begin{aligned}
f_{l} = \frac{w_{-1} f^{-1} + w_{0}f^{0}}{w_{-1}+w_0}, \quad f_{r} = \frac{w_0 f^0 + w_{1}f^{1}}{w_0+w_{1}}, \nonumber\end{aligned}$$ where the subscript ’$l$’ and ’$r$’ represent left side and right side respectively, and $w$ denotes the weight. Their formulas are written as follows, $$\begin{aligned}
w_{-1} = \frac{1}{4(s_{i-1}^2+\varepsilon)},\quad w_0 = \frac{3}{4(s_i^2+\varepsilon)}, \quad w_{1} = \frac{1}{4(s_{i+1}^2+\varepsilon)}, \nonumber\end{aligned}$$ where $\varepsilon = 1\times 10^{-6}$ is used to prevent zero denominator, and $s_i = f_{i+1}-f_{i}$, $$\begin{aligned}
f^{-1} = \frac{3}{2}f_{i}-\frac{1}{2}f_{i-1},\quad f^0 = \frac{1}{2}f_{i+1}+\frac{1}{2}f_{i}, \quad f^{1} = \frac{3}{2}f_{i+1}-\frac{1}{2}f_{i+2} \nonumber\end{aligned}$$
For high speed flow, the 3rd order WENO is also employed to calculate the macroscopic variables at the cell interface, owing to the discontinuous shock wave in the flow field. For low speed flow, the macroscopic conservative variables are interpolated by the central difference method, that is, $$\begin{aligned}
W_{i+1/2} = \frac{1}{2}(W_{i}+W_{i+1}).\end{aligned}$$
The derivatives of the microscopic and macroscopic variables are evaluated by a second order central difference method.
Boundary condition
------------------
Boundary condition is another crucial ingredient for AP schemes. At first, we recall the diffusion boundary condition for the traditional DOM in free molecular flow regime. The distribution function of the reflecting particles is subjected to the Maxwell distribution. Since no penetration occurs during the collision with the wall, the mass flux of the particle can be written as follows, $$\begin{aligned}
\int_0^{\Delta t}\int_{0}^{+\infty} u f^{in} d\mathbf{u}dt + \rho_{dom}\int_0^{\Delta t}\int_{-\infty}^{0} u g^* d\mathbf{u}dt = 0, \label{eq:nopenetration}\end{aligned}$$ where $f^{in}$ represents the incident molecular distribution function which is interpolated from the interior of the flow field, $\rho_{dom}$ is the density of the reflecting molecular stream. The reflecting molecular distribution function is assumed to be the Maxwell equilibrium on the wall, which reads, $$\begin{aligned}
g^* = \left(\frac{2RT^*}{\pi}\right)^{3/2} e^{-\frac{1}{2RT^*}\mathbf{u}^2}, \label{eq:wallEquilibrium}\end{aligned}$$ where $T^*$ denote the temperature of the boundary. According to Eq.(\[eq:nopenetration\]), the density of the reflecting distribution is determined, that is, $$\begin{aligned}
\rho_{dom} = -\frac{\sum\limits_{u_k>0} \omega_k u_k f_k}{\sum\limits_{u_k\leq 0}\omega_k u_k g_k}.\end{aligned}$$ The velocity distribution function at the wall for the microscopic variables is, $$\begin{aligned}
f_{dom} = \left\{
\begin{array}{c}
f^{in}, \quad u > 0, \\
\rho_{dom} g^*, \quad u \leq 0.
\end{array}\right.\end{aligned}$$ The numerical fluxes are written as follows, $$\begin{aligned}
\left\{\begin{array}{ccl}
\displaystyle\mathcal{F^*}_{dom} = u_k f_{dom,k}, \\
\displaystyle\mathcal{F^*}^W_{dom} = \langle u \psi f_{dom} \rangle_k.
\end{array}\right.\end{aligned}$$ The diffusion boundary condition is valid in free molecular flow regime, but cannot automatically recover the no slip boundary condition in the continuum flow regime. The boundary condition for the simplified method (S2, S3) should be designed carefully to preserve the asymptotic limits. Fortunately, this task is very easy to fulfill, since the simplified scheme is a simple combination of existing schemes. Here we just combine the boundary flux of the diffusion boundary condition and the boundary flux of the gas kinetic scheme for the Navier-Stokes equations to develop a boundary condition for the simplified scheme.
We modify the non-equilibrium bounce back boundary condition [@guo2002extrapolation] to implement the isothermal boundary condition for gas kinetic scheme. We adopt the extrapolation from the interior, then construct the NS distribution at the cell interface as the incident distribution function. $$\begin{aligned}
f^{in}_{gks} = g^{in}-r_{\tau}\tau(\mathbf{u}\cdot \nabla g^{in} + g^{in}_t) + g^{in}_t t, \quad\mbox{for}\ u > 0,\end{aligned}$$ where $r_{\tau}$ is defined in Eq.(\[eq:rTau\]). The reflecting distribution function is constructed as follows, $$\begin{aligned}
f^{out}_{gks}(u) = 2\rho_{gks} g^*(u) - f^{in}_{gks}(-u), \quad\mbox{for}\ u \leq 0.\end{aligned}$$ Then the complete velocity distribution function in the gas kinetic scheme is $$\begin{aligned}
f_{gks} = \left\{
\begin{array}{c}
f_{gks}^{in}, \quad u > 0, \\
f_{gks}^{out}, \quad u \leq 0.
\end{array}\right.\end{aligned}$$ The no penetration condition is also employed to determine the density at the wall boundary. $$\begin{aligned}
\rho_{gks} = \sqrt{\frac{2\pi}{RT^*}}\int_0^{+\infty} u f^{in}_{gks} d\mathbf{u}.\end{aligned}$$
The numerical flux for the conservative variables and for the distribution function are given respectively. $$\begin{aligned}
\left\{\begin{array}{ccl}
\mathcal{F}_k &=& \mathbf{u}_k f_{dom,k}, \\
\mathcal{F}^W &=& e^{-\beta} \sum\limits_k u_k \psi_k f_{dom} + (1-e^{-\beta}) <u \psi f_{gks}>. \label{eq:simplifiedBC}
\end{array}\right.\end{aligned}$$
We have tested another choice of $r_{\tau}$, say, $r_{\tau} = 1$ for the second simplified method (S2). When applying this boundary condition, in the free molecular flow regime, there were large oscillation near the boundary, since the coefficient $\gamma_3^{s2}$ is inconsistent with the analytical solution ($\gamma_3^{ugks}$). Therefore, only the simplified boundary condition (Eq.(\[eq:simplifiedBC\])) is adopted for all the numerical simulations in the next section.
Numerical comparison
====================
In all the following numerical tests, the CFL number is 0.4. And all the numerical setting are exactly identical except the numerical flux for different numerical schemes.
Sod shock tube
--------------
At first, the one dimensional shock tube problem is tested under different Knudsen numbers. $$\begin{aligned}
\mbox{Kn} = \frac{\mu_{ref}\sqrt{RT_{ref}}}{p_{ref}L}. \label{eq:Kn}\end{aligned}$$ The computational domain is $[0,1]$ in $x$ direction. And it is discretized into 200 cells. The initial condition is given as follows, $$\begin{aligned}
\left\{
\begin{array}{llll}
\rho_l = 1.0,& U_l = 0.0,& p_l = 1.0,&\quad \mbox{for}\ x \leq 0.5, \\
\rho_r = 0.125,& U_r = 0.0,& p_r = 0.1,&\quad \mbox{for}\ x > 0.5.
\end{array}\right.\end{aligned}$$ The quantities on the right-half domain are selected to define the Knudsen number. We use 150-point uniform grid in the velocity space $[-6,6]$. The computation stops at $t = 0.15$. Figure \[fig:sod1\] and \[fig:sod2\] show the numerical results for Kn$=0.0001, 0.01, 1, 10$. Five different flux solvers are employed to simulate this problem. As expected, all the methods provide very good results.
In the free molecular flow regime, say, the Knudsen number is 10, we find that, except the S2 scheme, all the numerical schemes predict the same density and momentum profile in figure \[fig:sod1\]. This is because the leading order terms are identical for all these schemes (Tab.(\[tab:coe\]), Eq.(\[eq:domflux\],\[eq:ugksflux\],\[eq:s1flux\])). Inaccurate results from the S2 scheme verify that, a simple combination of the DOM and a Navier-Stokes solver cannot lead to correct asymptotic limit. The quantities plotted in figure \[fig:sod1\] are the macroscopic variables updated by the Eq.(\[eq:macroConserve\]). When the relaxation time $\tau$ goes to infinity, the evolution of the distribution function (Eq.(\[eq:lastTimeStepCE\])) are totally independent to the evolution of the macroscopic variables (Eq.(\[eq:macroConserve\])), since the collision term vanishes. As a result, though the macroscopic variables are incorrect in the S2 results, the distribution function derived in the same simulation is identical to the other methods. We will demonstrate it in next two dimensional simulation. In the transition flow regime, the results derived from different schemes are still indistinguishable. In the continuum flow regime, the S1, S2, S3, DOM and UGKS provide almost identical solution. It testified that, the inaccuracy of the initial distribution function affects little to the numerical performance in the continuum flow regime.
These numerical observation are consist with our analysis in the previous section. The discrepancy is hardly noticed in all the flow regimes. All the numerical methods (except the S2) converge to the Euler solution in the continuum regime, and converge to collisionless solution in the free molecular flow regime.
Lid-driven cavity flow
----------------------
The one dimensional numerical results show that all the numerical schemes converge to the Euler solution at Kn$\rightarrow 0$. However, as mentioned in the reference [@chen2015comparative], the one dimensional numerical experiment cannot distinguish the NS AP scheme from the Euler AP scheme. Thus we simulate a two dimensional lid-driven cavity flow which is characterized by strong viscous effect. The gas flow is confined in a square domain whose extent is $[0,1]\times[0,1]$. Each edge of the computational domain is uniformly discretized by 61 nodes. The top boundary moves from left to right with a constant velocity, 0.2. The gas pressure is 1; and the density is also 1. The Mach number based on the velocity of the top wall is about 0.15. The Knudsen number is defined as Eq.(\[eq:Kn\]).
As mentioned in the last subsection, the conservative variables $W$ and the distribution function $f$ are evolving separately in the free molecular flow regime. Therefore, we use the $W$-based variable to denote the macroscopic variable deduced from the conservative variables $W$, and use $f$-based variable to denote the macroscopic variable deduced from the distribution function $f$. Figure \[fig:cavity1\] shows the $W$-based temperature and the $f$-based temperature derived from the S2 scheme. The flow condition is $\mbox{Kn}=2$ and $\mbox{Re} = 0.1$, and the velocity space $[-5,5]\times[-5,5]$ is discretized into $100\times100$. Since the DOM is accurate at high Knudsen number, we choose the $f$-based temperature derived from the DOM as benchmark solution, and plot it on the background. As shown in figure \[fig:cavity1\], the $W$-based temperature deviates from the DOM solution. Meanwhile, the $f$-based temperature is identical to the DOM solution. This is because that, $\mathcal{F}^{W}_{s2} \neq \langle \mathcal{F}_{s2} \rangle$ when $\beta$ approaches zero, namely, the macroscopic flux is inconsistent with the flux of distribution function. More specifically, this is the immediate consequence of the incorrect asymptotic coefficient $\gamma_3^{s2}$ in the S2 scheme (Tab.(\[tab:coe\])). After remedying the coefficient, the S3 scheme has the same asymptotic limit as the analytical solution. As we can see in figure \[fig:cavity2\], the results obtained from the S3 scheme, both $f$-based and $W$-based temperatures coincide with the results derived from the DOM. The results from all the considered numerical methods collapse to the DOM results in figure \[fig:cavity3\].
But remarkable discrepancies are observed when the Reynolds number increases to 1000. In this case, we only use 8 velocity points in one direction to discretize the velocity space ranging from -5 to 5. And the rectangular quadrature in velocity space is adopted. All these numerical settings are on the purpose of illustrating the influence of the inaccurate quadrature in velocity space. Central difference interpolation is adopted for both microscopic and macroscopic variables. As shown in Fig. \[fig:cavity4\](b,d), the DOM cannot simulate the continuum flow properly, therefore, the DOM’s results are not shown in Fig. \[fig:cavity4\](a,c). The UGKS and S1 schemes obtained much better numerical results which are closer to the reference data [@Ghia1982]. However, due to the inaccuracy of the quadrature in the velocity space, the numerical results are not as good as the numerical results in the previous literatures[@ugks2; @wang2015comparative; @zhu2015performance]. The numerical contour lines oscillate near the boundaries (Fig. \[fig:cavity4\](a,c)). On the other hand, the simplified schemes (S2, S3) perform best in this test case.
The asymptotic limits of the numerical schemes coincide with our analysis in the previous section. For the transition flow regime, the numerical results are shown in figure \[fig:cavity5\]. The Reynolds number is 10, and the Knudsen number is 0.02. We use 61 points in physical space, and use 60 points in velocity space. The velocity contours are almost identical for all the schemes. Only minor differences can be noticed in density contour and temperature contour.
From the above results, we demonstrate that the simplified schemes proposed in the paper possess correct asymptotic limit in free molecular flow regime and the continuum flow regime, and provide enough accurate numerical results in transition flow regime.
The high efficiency of the simplified methods
---------------------------------------------
In the Eq.(\[eq:s1flux\]), three out of five terms are evaluated by analytical formulas. These computational costs are infinitesimal compared to the quadrature in velocity space. We also observe that the S1 reduces about half computation time compared to the UGKS. And the DOM, S1, S2, and S3 schemes have almost identical computational efficiency.
On the other hand, the numerical results derived from the simplified methods are closer to the results from NS solver in the continuum flow regime. It is worth noting that, the coefficient for S2, say, $\gamma_0^{s2}$, deviates from $0$ by a exponential truncation error, while $\gamma_0^{ugks}$ and $\gamma_0^{s1}$ preserve $\tau$ as the leading order term. As illustrated in Eq.(\[eq:ugksfluxW\],\[eq:assumptionCE\],\[eq:ugksCE\]), the physical asymptotic process is not simply attained by vanishing the non-equilibrium terms, $f_0$ and $\mathbf{u}\cdot \nabla f$. The non-equilibrium terms still contribute a little $(O(\tau))$ to the total distribution function, and the remaining terms of non-equilibrium part are canceled by the equilibrium part, then result in the Chapman-Enskog expansion. Such balance is very delicate and sophisticated. It is definitely computationally burdensome or clumsy to simulate this subtle asymptotic process in velocity space. The simplified methods proposed in this study circumvent the delicate balance, instead, use more rapid decaying coefficients in front of the non-equilibrium terms. The quadrature of $f_0$ in velocity space impose almost nothing on the numerical macroscopic flux which means less numerical error in the scheme. As we can see in the numerical comparisons, the S2 and S3 schemes provide more accurate numerical results in the continuum flow regime, since the delicate balance between the non-equilibrium part and the equilibrium part are replace by a prior knowledge and circumvent the numerical simulation of asymptotic process. The quadrature of the distribution function is totally replaced by the analytical expression. Hence the simplified schemes lead to more accurate results, and less discrete points in velocity space.
Conclusion
==========
In this study, we analyzed the asymptotic behavior of the unified gas kinetic scheme, and reduced the unnecessary quadrature in the UGKS numerical flux for the equilibrium part. In the first simplified scheme, the quadrature in velocity space for the equilibrium part is replaced by the analytical results. The numerical comparison shows that this replacement reduces about half computation load and does not effect numerical results. Based on the asymptotic expression of the coefficients in the UGKS flux, several other simplification strategies have been proposed. The numerical comparisons demonstrated that simple combination (S2) of a kinetic flux and the macroscopic flux cannot obtain correct asymptotic limit in the free molecular flow regime. With a rescaled viscosity coefficient, the simplified scheme (S3) possesses correct asymptotic limit both in the free molecular flow regime and in the continuum flow regime. Moreover, it can be constructed by combining two existing flux solvers which handle the kinetic equation and the Navier-Stokes equations respectively. The simplified scheme (S3) is efficient in terms of coding and computing, hence, is a promising approach for engineering application. Its accuracy is also acceptable and controllable. The flux hybrid strategy proposed in this study can be further extended to the other multiscale problems.
Acknowledgements {#acknowledgements .unnumbered}
================
This work was supported by NSF91530319, Hong Kong Research Grant Council (620813, 16211014, 16207715), and HKUST (PROVOST13SC01, IRS15SC29, SBI14SC11).
References {#references .unnumbered}
==========
[26]{}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 [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [ ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [ ()]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We consider the evolution of a flat, isotropic and homogeneous Friedmann-Robertson-Walker Universe, filled with a causal bulk viscous cosmological fluid, that can be characterized by an ultra-relativistic equation of state and bulk viscosity coefficient obtained from recent lattice QCD calculations. The basic equation for the Hubble parameter is derived under the assumption that the total energy in the Universe is conserved. By assuming a power law dependence of bulk viscosity coefficient, temperature and relaxation time on energy density, an approximate solution of the field equations has been obtained, in which we utilized equations of state from recent lattice QCD simulations QCD and heavy-ion collisions to derive an evolution equation. In this treatment for the viscous cosmology, we found no evidence for singularity. For example, both Hubble parameter and scale factor are finite at $t=0$, $t$ is the comoving time. Furthermore, their time evolution essentially differs from the one associated with non-viscous and ideal gas. Also thermodynamic quantities, like temperature, energy density and bulk pressure remain finite as well. In order to prove that the free parameter in our model does influence the final results, qualitatively, we checked out other particular solutions.'
author:
- |
A. Tawfik$^{1}$, M. Wahba$^1$, H. Mansour$^2$ and T. Harko$^3$\
[$^1$Egyptian Center for Theoretical Physics (ECTP), MTI University, Cairo-Egypt]{}\
[$^2$Department of Physics, Faculty of Science, Cairo University, Giza-Egypt]{}\
[$^3$Department of Physics and Center for Theoretical and Computational Physics,]{}\
[University of Hong Kong, Hong Kong]{}\
title: |
\
Viscous Quark-Gluon Plasma in the Early Universe
---
Introduction {#sec:intro}
============
The dissipative effects, including both bulk and shear viscosity, are supposed to play a very important role in the early evolution of the Universe. The first attempts at creating a theory of relativistic fluids were those of Eckart [@Ec40] and Landau and Lifshitz [@LaLi87]. These theories are now known to be pathological in several respects. Regardless of the choice of equation of state, all equilibrium states in these theories are unstable and in addition signals may be propagated through the fluid at velocities exceeding the speed of light. These problems arise due to the first order nature of the theory, that is, it considers only first-order deviations from the equilibrium leading to parabolic differential equations, hence to infinite speeds of propagation for heat flow and viscosity, in contradiction with the principle of causality. Conventional theory is thus applicable only to phenomena which are quasi-stationary, i.e. slowly varying on space and time scales characterized by mean free path and mean collision time.
A relativistic second-order theory was found by Israel [@Is76] and developed by Israel and Stewart [@IsSt76], Hiscock and Lindblom [HiLi89]{} and Hiscock and Salmonson [@HiSa91] into what is called “transient” or “extended” irreversible thermodynamics. In this model deviations from equilibrium (bulk stress, heat flow and shear stress) are treated as independent dynamical variables, leading to a total of 14 dynamical fluid variables to be determined. For general reviews on causal thermodynamics and its role in relativity see [@Ma95].
Causal bulk viscous thermodynamics has been extensively used for describing the dynamics and evolution of the early Universe or in an astrophysical context. But due to the complicated character of the evolution equations, very few exact cosmological solutions of the gravitational field equations are known in the framework of the full causal theory. For a homogeneous Universe filled with a full causal viscous fluid source obeying the relation $\xi \sim \rho ^{1/2}$, with $\rho $ the energy density of the cosmological fluid, exact general solutions of the field equations have been obtained in [@ChJa97; @MaHa99a; @MaHa99b; @MaHa00a; @MaTr97]. It has also been proposed that causal bulk viscous thermodynamics can model on a phenomenological level matter creation in the early Universe [@ChJa97]. Exact causal viscous cosmologies with $\xi \sim \rho ^{s},s\neq 1/2$ have been considered in Ref. [@MaHa99a].
Because of technical reasons, most investigations of dissipative causal cosmologies have assumed Friedmann-Robertson-Walker (FRW) symmetry (i.e. homogeneity and isotropy) or small perturbations around it [@MaTr97]. The Einstein field equations for homogeneous models with dissipative fluids can be decoupled and therefore are reduced to an autonomous system of first order ordinary differential equations, which can be analyzed qualitatively [@CoHo95].
The role of a transient bulk viscosity in a FRW space-time with decaying vacuum has been discussed in [@AbVi97]. Models with causal bulk viscous cosmological fluid have been considered recently [@ArBe00]. They obtained both power-law and inflationary solutions, with the gravitational constant an increasing function of time. The dynamics of a viscous cosmological fluids in the generalized Randall-Sundrum model for an isotropic brane were considered in [@Chen01]. The renormalization group method was applied to the study of homogeneous and flat FRW Universes, filled with a causal bulk viscous cosmological fluid, in [@Be03]. A generalization of the Chaplygin gas model, by assuming the presence of a bulk viscous type dissipative term in the effective thermodynamic pressure of the gas, was investigated recently in [@Pun08].
Recent RHIC results give a strong indication that in the heavy-ion collisions experiments, a hot dense matter can be formed [@reff1]. Such an experimental evidence might agree with the “new state of matter” as predicted in the Lattice QCD simulations [@reff5]. However, the experimentally observed elliptic flow in peripheral heavy-ion collisions seems to indicate that a thermalized collective QCD matter has been produced. In a addition to that, the success of ideal fluid dynamics in explaining several experimental data e.g. transverse momentum spectra of identified particles, elliptic flow [@reff6], together with the string theory motivated that the shear viscosity $\eta$ to the entropy $s$ would have the lower limit $\approx 1/4\pi$ [@reff7] leading to a paradigm that in heavy- ion collisions, that a [*nearly*]{} perfect fluid likely be created and the quarks and gluons likely go through relatively rapid equilibrium characterized with a thermalization time less than $1$ fm/c [@mueller1].
According to recent lattice QCD simulations [@mueller2], the bulk viscosity $
\xi $ is not negligible near the QCD critical temperature $T_c$. It has been shown that the bulk and shear viscosity at high temperature $T$ and weak coupling $\alpha_s
$, $\xi\sim \alpha_s^2 T^3/\ln \alpha_s^{-1}$ and $\eta\sim T^3/(\alpha_s^2 \ln
\alpha_s^{-1})$ [@mueller3]. Such a behavior obviously reflects the fact that near $T_c$ QCD is far from being conformal. But at high $T$, QCD approaches conformal invariance, which can be indicated by low trace anomaly $(\epsilon-3p)/T^4$ [@karsh09], where $\epsilon$ and $p$ are energy and pressure density, respectively. In the quenched lattice QCD, the ratio $\zeta/s$ seems to diverge near $T_c$ [@meyer08].
To avoid the mathematical difficulties accompanied with the Abel second type non-homogeneous and non-linear differential equations [@TawCosmos], one used to model the cosmological fluid as an ideal (non-viscous) fluid. No doubt that the viscous treatment of the cosmological background should have many essential consequences [@taw08]. The thermodynamical ones, for instance, can profoundly modify the dynamics and configurations of the whole cosmological background [@conseq1]. The reason is obvious. The bulk viscosity is to be expressed as a function of the Universe energy density $\rho$ [@conseq2]. Much progress has been achieved in relativistic thermodynamics of dissipative fluids. The pioneering theories of Eckart [@Ec40] and Landau and Lifshitz [@LaLi87] suffer from lake of causality constrains. The currently used theory is the Israel and Stewart theory [@Is76; @IsSt76], in which the causality is conserved and theory itself seems to be stable [@HiLi89; @Ma95].
In this article, we aim to investigate the effects that bulk viscosity has on the Early Universe. We consider a background corresponding to a FRW model filled with ultra-relativistic viscous matter, whose bulk viscosity and equation of state have been deduced from recent heavy-ion collisions experiments and lattice QCD simulations.
The present paper is organized as follows. The basic equations of the model are written down in Section \[field\]. In Section \[approx\] we present an approximate solution of the evolution equation. Section \[part1\] is devoted to one particular solution, in which we assume that $H=const.$ The results and conclusions are given in Sections \[final\] and \[final2\], respectively.
Evolution equations {#field}
===================
We assume that geometry of the early Universe is filled with a bulk viscous cosmological fluid, which can be described by a spatially flat FRW type metric given by $$\label{1}
ds^{2}=dt^{2}-a^{2}\left( t\right) \left[ dr^{2}+r^{2}\left( d\theta
^{2}+\sin ^{2}\theta d\phi ^{2}\right) \right] .$$ The Einstein gravitational field equations are: $$R_{ik}-\frac{1}{2}g_{ik}R=8\pi GT_{ik}. \label{ein}$$ In rest of this article, we take into consideration natural units, i.e., $c=1$, for instance.
The energy-momentum tensor of the bulk viscous cosmological fluid filling the very early Universe is given by $$T_{i}^{k}=\left( \rho +p+\Pi\right) u_{i}u^{k}-\left( p+\Pi\right)
\delta_{i}^{k},\label{1_a}$$ where $i,k$ takes $0,1,2,3$, $\rho$ is the mass density, $p$ the thermodynamic pressure, $\Pi $ the bulk viscous pressure and $u_{i}$ the four velocity satisfying the condition $u_{i}u^{i}=1$. The particle and entropy fluxes are defined according to $N^{i}=nu^{i}$ and $S^{i}=sN^{i}-\left( \tau\Pi^{2}/2\xi T\right)
u^{i}$, where $n$ is the number density, $s$ the specific entropy, $T\geq0$ the temperature, $\xi$ the bulk viscosity coefficient, and $\tau\geq0$ the relaxation coefficient for transient bulk viscous effect (i.e. the relaxation time), respectively.
The evolution of the cosmological fluid is subject to the dynamical laws of particle number conservation $N_{\text{ };i}^{i}=0$ and Gibbs’ equation $Td\rho=d\left( \rho /n\right) +pd\left( 1/n\right) $. In the following we shall also suppose that the energy-momentum tensor of the cosmological fluid is conserved, that is $T_{i;k}^{k}=0$.
The bulk viscous effects can be generally described by means of an effective pressure $\Pi $, formally included in the effective thermodynamic pressure $p_{eff}=p+\Pi $ [@Ma95]. Then in the comoving frame the energy momentum tensor has the components $T_{0}^{0}=\rho ,T_{1}^{1}=T_{2}^{2}=T_{3}^{3}=-p_{eff}$. For the line element given by Eq. (\[1\]), the Einstein field equations read $$\begin{aligned}
\label{2}
\left( \frac{\dot{a}}{a}\right)^{2} &=& \frac{8\pi}{3}G \;\rho, \\
\frac{\ddot{a}}{a} &=& -\frac{4\pi}{3}G \; \left( 3p_{eff}+\rho \right),
\label{3}\end{aligned}$$ where one dot denotes derivative with respect to the time $t$, $G$ is the gravitational constant and $a$ is the scale factor.
Assuming that the total matter content of the Universe is conserved, $T_{i;j}^j=0$, the energy density of the cosmic matter fulfills the conservation law: $$\label{5}
\dot{\rho}+3H\left( p_{eff}+\rho \right) =0,$$ where we introduced the Hubble parameter $H=\dot{a}/a$. In presence of bulk viscous stress $\Pi $, the effective thermodynamic pressure term becomes $p_{eff}=p+\Pi $. Then Eq. (\[5\]) can be written as $$\label{6}
\dot{\rho}+3H\left( p+\rho \right) =-3\Pi H.$$
For the evolution of the bulk viscous pressure we adopt the causal evolution equation [@Ma95], obtained in the simplest way (linear in $\Pi)$ to satisfy the $H$-theorem (i.e., for the entropy production to be non-negative, $S_{;i}^{i}=\Pi^{2}/\xi T\geq0$ [@Is76; @IsSt76]). According to the causal relativistic Israel-Stewart theory, the evolution equation of the bulk viscous pressure reads [@Ma95] $$\label{8}
\tau \dot{\Pi}+\Pi =-3\xi H-\frac{1}{2}\tau \Pi \left( 3H+\frac{\dot{\tau}}{\tau }-\frac{\dot{\xi}}{\xi }-\frac{\dot{T}}{T}\right).$$ In order to have a closed system from equations (\[2\]), (\[6\]) and (\[8\]) we have to add the equations of state for $p$ and $T$.
As shown in Appendix A, the equation of state, the temperature and the bulk viscosity of the quark-gluon plasma (QGP), can be determined approximately at high temperatures [@karsch07] from recent lattice QCD calculations [@Cheng:2007jq], as $$\label{13}
P = \omega \rho,\hspace*{1cm}T = \beta \rho^r,\hspace*{1cm}\xi = \alpha \rho + \frac{9}{\omega_0} T_c^4,$$ with $\omega = (\gamma-1)$, $\gamma \simeq 1.183$, $r\simeq 0.213$, $\beta\simeq 0.718$, $$\alpha = \frac{1}{9\omega_0} \frac{9\gamma^2-24\gamma+16}{\gamma-1},$$ and $\omega_0 \simeq 0.5-1.5$ GeV. In the following we assume that $\alpha \rho >> 9/\omega_0 T_c^4$, and therefore we take $\xi \simeq \alpha \rho$. In order to close the system of the cosmological equations, we have also to give the expression of the relaxation time $\tau $, for which we adopt the expression [@Ma95], $$\label{tau}
\tau=\xi\rho^{-1}\simeq\alpha .$$
Eqs. (\[13\]) are standard in the study of the viscous cosmological models, whereas the equation for $\tau$ is a simple procedure to ensure that the speed of viscous pulses does not exceed the speed of light. Eq. (\[tau\]) implies that the relaxation time in our treatment is constant but strongly depends on EoS. These equations are without sufficient thermodynamical motivation, but in the absence of better alternatives, we shall follow the practice of adopting them in the hope that they will at least provide some indication of the range of bulk viscous effects. The temperature law is the simplest law guaranteeing positive heat capacity.
With the use of Eqs. (\[8\]), (\[13\]) and (\[tau\]), respectively, we obtain the following equation describing the cosmological evolution of the Hubble function $H$ $$\begin{aligned}
\label{init}
\ddot H + \frac{3}{2} [1+(1-r) \gamma] H\dot H + \frac{1}{\alpha}\dot H - (1+r) H^{-1} \dot H^2 + \frac{9}{4}(\gamma -2) H^3 +
\frac{3}{2}\frac{\gamma}{\alpha} H^2 &=& 0.\end{aligned}$$
An approximate solution {#approx}
=======================
We introduce the transformation $u=\dot{H}$, so that Eq. (\[init\]) is transformed into a first order ordinary differential equation, $$\label{init2}
u\frac{du}{dH}-(1+r)H^{-1}u^{2}+\left(\frac{3}{2}[1+(1-r)\gamma ]H+\alpha
^{-1}\right) u+\frac{9}{4}\frac{1}{(\gamma)}H^{3}+\frac{3}{2}\frac{\gamma}{\alpha}
H^{2}=0.$$ We can rewrite Eq. (\[init2\]) in the form $$\label{OmegH1}
\Omega \frac{d\Omega }{dH} = F_1(H)\Omega + F_0(H),$$ where $$\begin{aligned}
\Omega &=& u \; E \;\; = u\; \exp\left(-\int \frac{1+r}{H} dH\right), \nonumber \\
F_1(H) &=& -\left( \frac{3}{2} [1+(1-r)\gamma]H + \frac{1}{\alpha}\right)E, \nonumber \\
F_0(H) &=& -\left(\frac{9}{4}(\gamma-2) H^3 + \frac{3}{2} \frac{\gamma}{\alpha} H^2\right)E^2. \nonumber\end{aligned}$$ By introducing a new independent variable $z=\int F_1(H)\,dH$, we obtain $$\Omega \frac{d\Omega}{dz} - \Omega = g(z),$$ with $g(z)$ is defined parametrically as, $$\label{gofzz1}
g(z) = \frac{F_0}{F_1}.$$ As shown in Appendix B, $g(z)$ can be approximated as a simple function of $z$ $$g(z)\approx {\cal C}\; z,$$ where ${\cal C}$ is a constant. We proceed with this approximation to get solvable differential equations. Keeping the parametric solution of $g(z)$, Eq. (\[fullgofz\]), results in much more complicated differential equations. This would be the subject of a future work.
From the definitions of $\Omega$ and $z$ we have $$\begin{aligned}
\Omega &=& H^{1+r}\dot H, \label{Eq1} \\
z &=& H^{2+r} \left(\frac{-3[1+(1-r)\gamma]H}{2(1-r)} +\frac
{1}{\alpha r}\right), \label{Eq2}\end{aligned}$$ Analogous to the solution of reduced Abel type canonical equation, $$\label{abel1}
y \frac{dy}{dx} - y = a x$$ (see Appendix C) we obtain the relation $\Omega = z/{\cal P}$. Therefore, from Eqs. (\[Eq1\]) and (\[Eq2\]) we obtain the following first order differential equation for Hubble parameter $H$, $$\label{init-polyn}
{\cal P} \dot H = \frac{-3[1+(1-r)\gamma]}{2(1-r)} H^2 +\frac{1}{\alpha r}H$$ with the solution $$H(t) = \frac{B}{\exp(-Bt/{\cal P})-A} \label{eq:mysolut1}$$ where $$\label{paramsab}
A=\frac{-3[1+(1-r)\gamma]}{2(1-r)}, \hspace*{1cm} B=\frac{1}{\alpha r},$$ and ${\cal P}$ is taken as a free parameter. We can assign any real value to ${\cal P}$. For the results presented in this work, we used a negative value. This negative sign is necessarily to overcome the sign from the integral limits. The geometric and thermodynamic quantities of the Universe read $$\begin{aligned}
a(t) &=& a_0\left(\frac{\exp(-B t/{\cal P})}{\exp(-B
t/{\cal P})-A}\right)^{{\cal P}/A} \label{approx-a}, \\
\rho(t) &=& 3\, H^2= 3 \left(\frac{B}{\exp(-Bt/{\cal P})-A}\right)^2 \label{approx-rho},\\
T(t) &=& \beta \rho^{r}=\beta \left(3 \frac{B^{2}}{[\exp(-Bt/{\cal P})-A]^{2}}\right)^r \label{approx-T},\\
\Pi(t) &=& -2\dot{H}-3\gamma H^{2}=-\frac{B^2}{{\cal P}} \left(\frac{2\exp(-Bt/{\cal P})+3\gamma{\cal P}}{[\exp(-Bt/{\cal P})-A]^2}\right) \label{approx-Pi},\\
q(t)&=&\frac{d}{dt}H^{-1}-1=-\frac{1}{{\cal P}}\exp(-Bt/{\cal P})-1. \label{approx-q}\end{aligned}$$ $a_0$ is an arbitrary constant of the integration. The sign of $q$ indicates whether the Universe decelerates (positive) or accelerates (negative). $q$ can also be given as a function of the thermodynamic, gravitational and cosmological quantities $q(t)=[\rho(t)
+3p(t)+3\Pi(t)]/2\rho(t)$ [@kolbBook].
de Sitter Universe {#part1}
===================
Besides the approximation in $g(z)$, previous solution apparently depends on the free parameter ${\cal P}$. In this section, we suggest a particular solution to overcome ${\cal P}$. Eq. (\[init\]) can easily be obtained by assuming that $H$ doesn’t depend one $t$, i.e, de Sitter Universe. With a simple calculation, we get an estimation for $H$ $$\label{partcH}
H=\frac{4}{9}\frac{\alpha ^{-1}\gamma}{2-\gamma }.$$ The geometric and thermodynamic parameters of the Universe are given by $$\begin{aligned}
a(t) &=& a_{0}\exp \left[ \frac{4\alpha ^{-1}\gamma }{9(2-\gamma )}t\right], \label{partca} \\
\rho(t) &=& 3\left[ \frac{4\alpha ^{-1}\gamma }{9(2-\gamma )}\right] ^{2}, \label{partcrho}\\
T(t) &=& 3^{r}\beta \left[ \frac{4\alpha ^{-1}\gamma }{9(2-\gamma )}\right]
^{2r}, \label{partcT}\\
\Pi(t) &=& -3\gamma \left[ \frac{4\alpha
^{-1}\gamma }{9(2-\gamma )}\right] ^{2}, \label{partcPi}\\
q(t) &=& -1. \label{partcq}\end{aligned}$$ Although we have assumed here that the cosmic background is filled with viscous matter, the assumption that $H=const$ results in an exponential scale parameter, Eq. (\[partca\]). This behavior characterizes the de Sitter space, when $\Lambda=k=0$. $\rho$ and $T$ are finite at small $t$ as given in Fig. \[Figg2\].
Particular Solution {#part2}
===================
Another particular solution for Eq. (\[init\]) can be obtained, when assuming that the dependence of $u$ on $H$ can be given by the polynomial in Eq. (\[init-polyn\]) $$\label{init-partc2}
u=b_{1}H^{2}+b_{2}H,$$ where $b_{1}$ and $b_{2}$ are constants. Some simple calculations show that this form is a solution of the initial equation, Eq. (\[init2\]), if $$\begin{aligned}
b_1 &=&-\frac{3}{2}\frac{1+\gamma }{1-r}, \\
b_2 &=& \frac{1}{r\alpha }.\end{aligned}$$ $b_2$ is identical to $B$ in Eq. (\[paramsab\]). $r$ and $\gamma$ have to satisfy the compatibility relation $$r=\frac{2-\gamma}{2+\gamma^2}.$$ Integrating Eq. (\[init-partc2\]) results in $$\begin{aligned}
\label{Eq:Ht}
H(t) &=& \frac{b_2\exp(-b_2 t)}{1-b_1\exp(-b_2 t)},\end{aligned}$$ where minus sign in the exponential function refers to flipping the integral limits. This was not necessary while deriving the expressions given in Section \[approx\]. The free parameter [P]{} compensates it. The geometric and thermodynamic quantities of the Universe read $$\begin{aligned}
a(t)&=&a_0\left(\frac{\exp(b_2t)-b_1}{\exp(b_2t)}\right)^{1/b_1}, \label{partc2a} \\
\rho(t)&=& 3 \left(\frac{b_2\exp(-b_2 t)}{1-b_1\exp(-b_2 t)}\right)^2, \label{partc2rho}\\
T(t)&=& 3^r\;\beta \left(\frac{b_2\exp(-b_2 t)}{1-b_1\exp(-b_2 t)}\right)^{2r}, \label{partc2T}\\
\Pi(t)&=& \frac{b_2^2 \left[2\exp(b_2t)-3\gamma\right]}{\left[\exp(b_2t)-b_1\right]^2}, \label{partc2Pi}\\
q(t) &=& \exp(b_2t)-1. \label{partc2q}\end{aligned}$$ Obviously , we notice that the scale parameter in Eq. (\[partc2a\]) looks like Eq. (\[approx-a\]), which strongly depends on the free parameter ${\cal
P}$. The other geometric and thermodynamic quantities find similarities in Eq. (\[approx-rho\]) - (\[approx-q\]), respectively. Deceleration parameter $q$ seems to be positive everywhere.
Results {#final}
=======
In present work, we have considered the evolution of a full causal bulk viscous flat, isotropic and homogeneous Universe with bulk viscosity parameters and equation of state taken from recent lattice QCD data and heavy-ion collisions. Three classes of solutions of the evolution equation have been obtained.
In Fig. \[Figg1\], $H(t)$ and $a(t)$ are depicted in dependence on the comoving time $t$. We compare $H(t)$, given by Eq. (\[eq:mysolut1\]), and $a(t)$, given by Eq. (\[approx-a\]), with the counterpart parameters obtained in the case when the background matter is assumed to be an ideal and non-viscous fluid, described by the equations of state of the non-interacting ideal gas, $$\begin{aligned}
H(t) &=& \frac{1}{2t} \label{htideal1}, \\
a(t) &=& \sqrt{t}. \label{atideal1}\end{aligned}$$
In the left panel of Fig. \[Figg1\], $H(t)=\dot a/a$ has an exponential decay, whereas in the non-viscous case, $H(t)$ is decreasing according to Eq. (\[htideal1\]). The latter is much slower than the former, reflecting the nature of the exponential and linear dependencies. The other difference between the two cases is obvious at small $t$. We notice a divergence, or singularity, associated with the ideal non-viscous fluid, Eq. (\[htideal1\]). The viscous fluid results in finite $H$ even at vanishing $t$, as can be seen from Eq. (\[eq:mysolut1\]).
The scale factor $a(t)$ also shows differences in both cases. $a(t)$ in a Universe with an ideal and non-viscous background matter depends on $t$ according to Eq. (\[atideal1\]), which simply implies that $a(t)$ is directly proportional to $t$, and $a(t)$ vanishes at $t=0$, which shows the existence of a singularity of $H$. Assuming that the background matter is described by a viscous fluid results in different $a(t)$-behaviors with increasing $t$. At $t=0$, $a(t)$ remains finite. Correspondingly, $H(t)$ remains also finite. In general, the dependence on $t$ is much more complicated than in Eq. (\[atideal1\]). Here we have an $A/{\cal P}$ root of an exponential function. If $\exp(-Bt/{\cal P})>>A$, $a$ remains constant.
Fig. \[Figg2\] illustrates the dependence of the two thermodynamical quantities, $\rho $ and $T$, on the comoving time. The non-viscous Universe shows a singular behavior in $\rho$ at vanishing $t$, as shown in the left panel of Fig. \[Figg2\]. This is not obvious in the case where we have taken into consideration a finite viscosity coefficient, i.e., $\rho$ is finite at $t=0$. In both cases, $\rho$ is decreasing with increasing $t$, reflecting that the Early Universe was likely expanding. Also the life time of the thermal viscous Universe seems to be shorter than for the non-viscous Universe. Almost the same behavior is observed in the right panel of Fig. \[Figg2\]. The temperature $T$ seems to be finite at vanishing $t$ in the viscous Universe. The $T$-singularity is only present, if we assume that the background matter is non-viscous ideal gas.
In left panel of Fig. \[Figg3\], we show the dependence of the bulk viscous pressure $\Pi$ on $t$. $\Pi$ takes negative values at very small $t$. Then it switches to positive values at some values of $t$. After reaching the maximum value, $\Pi$ decays exponentially with increasing $t$. At larger $t$, $\Pi$ entirely vanishes. The deceleration parameter $q$, given by Eq. (\[approx-q\]), is depicted in the right panel of Fig. \[Figg3\], and it is compared with $q$ for a non-viscous fluid, $q=-3$. The approximate solution, given by Eq. (\[approx-q\]), results in negative $q$ at small $t$, referring to expansion era. $q$ from the particular solution, Eq. (\[partcq\]) is negative everywhere.\
For the particular solution, only the scale factor depends on $t$, Eq. (\[partca\]). The results are given in the right panel of Fig. \[Figg1\]. All cosmological and thermodynamical quantities given by Eq. (\[partcH\]) and Eqs. (\[partcrho\])-(\[partcq\]) are constant in time.
Conclusions {#final2}
===========
It is obvious that the bulk viscosity plays an important role in the evolution of the Early Universe. Despite of the simplicity of our model, it shows that a better understanding of the dynamics of our Universe is only accessible, if we use reliable equation of state to characterize the matter filling out the cosmic background.
We conclude that the causal bulk viscous Universe described by the approximate solution starts its evolution from an initial non-singular state with a non-zero initial value of Hubble parameter $H(t)$ and scale factor $a(t)$, where $t$ is the comoving time. In this treatment, $t$ is given in GeV$^{-1}$. Also the thermodynamical quantities, energy density $\rho$ for instance, are finite at vanishing $t$. Even the temperature $T$ itself shows no singularity at $t=0$. The Hubble parameter $H$ decreases monotonically with $T$ similar to $\rho$. The bulk viscous pressure $\Pi$ likely satisfies the condition that $\Pi<0$ at very small $t$ indicating to inflationary era. Then $\Pi $ switches to positive value. It reaches a maximum value and then decays and vanishes, exponentially, at large $t$. The deceleration parameter $q$ shows an expanding behavior in the case of non-viscous ideal gas and first particular solution. For second particular solution, $q$ starts from zero and increases, exponentially. According to this solution, the Universe was decelerating. The approximate solution shows an interesting behavior in $q(t)$, Eq. (\[approx-q\]). At small $t$, the values of $q$ are negative, i.e. the Universe was accelerating (expansion). At larger $t$, a non-inflationary behavior sets on, $q>0$, i.e., the Universe switched to a decelerating evolution.
In this treatment, we assumed that the Universe is flat, $k=0$, and the background geometry is filled out with QCD matter (QGP) with a finite viscosity coefficient. The resulting Universe is obviously characterized by a shortly increasing and afterward constant scale factor and a fast vanishing Hubble parameter. At $t=0$, both $a(t)$ and $H(t)$ remain finite, i.e., there is no singularity. The validity of our treatment depends on the validity of the equations of states, Eq. \[13\], which we have deduced from the lattice QCD simulations at temperatures larger than $T_c\approx 0.19~$GeV. Below $T_c$, as the Universe cooled down, not only the degrees of freedom suddenly increase [@Tawfik03] but also the equations of state turn to be the ones characterizing the hadronic matter. Such a phase transition - from QGP to hadronic matter - would characterize one end of the validity of our treatment. The other limitation is the very high temperatures (energies), at which the strong coupling $\alpha_s$ entirely vanishes.
[00]{} C. Eckart, Phys. Rev. [**58**]{}, 919 (1940). L. D. Landau and E. M. Lifshitz, [*Fluid Mechanics*]{}, Butterworth Heinemann (1987). W. Israel, Ann. Phys. [**100**]{}, 310 (1976). W. Israel and J. M. Stewart, Phys. Lett. [**A58**]{}, 213 (1976). W. A. Hiscock and L. Lindblom, Ann. Phys. [**151**]{}, 466 (1989). W. A. Hiscock and J. Salmonson, Phys. Rev. [**D43**]{}, 3249 (1991). R. Maartens, Class. Quantum Grav. [**12**]{}, 1455 (1995); R. Maartens, [*Causal thermodynamics in relativity*]{}, [astro-ph/9609119]{} (1996). L. P. Chimento and A. S. Jakubi, Class. Quantum Grav. [**14**]{}, 1811 (1997) ; L. P. Chimento and A. S. Jakubi, Int. J. Mod. Phys. [**D7**]{}, 177 (1998); M. K. Mak and T. Harko, Gen. Rel. Grav. [**30**]{}, 1171 (1998); Gen. Rel. Grav. [**31**]{}, 273 (1999); J. Math. Phys. [**39**]{}, 5458 (1998). T. Harko and M. K. Mak, Int. J. Theor. Phys. [**38**]{}, 1561 (1999). M. K. Mak and T. Harko, Aust. J. Phys. [**52**]{}, 659 (1999). M. K. Mak and T. Harko, Int. J. Mod. Phys. [**D9**]{}, 97 (2000); Aust. J. Phys. [**53**]{}, 241 (2000); Int. J. Mod. Phys. [**D9**]{}, 475 (2000). R. Maartens and J. Triginer, Phys. Rev. [**D56**]{}, 4640 (1997). A. A. Coley and R. J. van den Hoogen, Class. Quantum Grav. [**12**]{}, 1977 (1995) ; A. A. Coley and R. J. van den Hoogen, Phys. Rev. [**D54**]{}, 1393 (1996) ; A. Di Prisco, L. Herrera and J. Ibanez, Phys. Rev. [**D63**]{}, 023501 (2001). Abdussatar and R. G. Vishwakarma, Class. Quantum Grav. [**14**]{}, 945 (1997).
A. I. Arbab and A. Beesham, Gen. Rel. Grav. [**32**]{}, 615 (2000).
C. M. Chen, T. Harko, and M. K. Mak, Phys. Rev. [**D64**]{}, 124017 (2001); T. Harko and M. K. Mak, Class. Quant. Grav. [**20**]{}, 407 (2003). J. A. Belinchon, T. Harko, and M. K. Mak, Class. Quant. Grav. [**19**]{}, 3003 (2002). C. S. J. Pun, L. A. Gergely, M. K. Mak, Z. Kovacs, G. M. Szabo, and T. Harko, Phys.Rev. [**D77**]{}, 063528 (2008).
BRAHMS Collaboration, I. Arsene [*et al.*]{}, Nucl. Phys. A [**757**]{}, 1 (2005). PHOBOS Collaboration, B. B. Back [*et al.*]{}, Nucl. Phys. A [**757**]{}, 28 (2005). PHENIX Collaboration, K. Adcox [*et al.*]{}, Nucl. Phys. A [**757**]{} (2005) STAR Collaboration, J. Adams [*et al.*]{}, Nucl. Phys. A [**757**]{} (2005) F. Karsch, E. Laermann, P. Petreczky, S. Stickan and I. Wetzorke, 2001 [*Proccedings of NIC Symposium*]{}, Ed. H. Rollnik and D. Wolf, John von Neumann Institute for Computing, Jülich, NIC Series, [**9**]{}, (2002). P. F. Kolb and U. Heinz, in [*Quark-Gluon Plasma 3*]{}, edited by R. C. Hwa and X.-N. Wang, World Scientific, Singapore, (2004). G. Policastro, D. T. Son and A. O. Starinets, Phys. Rev. Lett. [**87**]{}, 081601 (2001);
JHEP [**0209**]{}, 043 (2002) R. J. Fries, J. Phys. G [**34**]{}, S851 (2007). D. Kharzeev and K. Tuchin, JHEP [**0809**]{}, 093 (2008). F. Karsch, D. Kharzeev, and K. Tuchin, Phys. Lett. B [**663**]{}, 217 (2008). P. Arnold, C. Dogan, and G. D. Moore, Phys. Rev. D [**74**]{}, 085021 (2006). A. Bazavov, [*et al.*]{}, Phys. Rev. [**D80**]{}, 014504 (2009). H. Meyer, Phys. Rev. Lett. [**100**]{}, 162001 (2008). A. Tawfik, H. Mansour and M. Wahba, Invited talk given at the 7th international conference on “Modern Problems of Nuclear Physics”, 22-25 September 2009, Tashkent-Uzbekistan, arXiv:0911.4105 \[gr-qc\];
Talk given at 12th Marcel Grossmann Meeting on “General Relativity”, Paris-France, 12-18 July 2009, arXiv:0912.0115 \[gr-qc\].
A. Tawfik, AIP Conf. Proc. [**1115**]{}, 239 (2009), arXiv:0809.3825 \[hep-ph\]. A. Di Prisco, L. Herrera, J. Ibanez, Phys. Rev. D [**63**]{}, 023501 (2001).
V. Belinskii, E. Nikomarov, I. Khalantikov, Sov. Phys. JETP, [**50**]{}, 213 (1979). E. W. Kolb and M. S. Turner, , Addison-Wesley Publ. Co. (1990).
M. Cheng [*et al.*]{}, arXiv:0710.0354 \[hep-lat\]. A. Tawfik and D. Toulban, Phys. Lett. B [**623**]{}, 48 (2005).
F. Karsch, K. Redlich and A. Tawfik, Eur. Phys. J. [**C29**]{} 549 (2003), e-Print: hep-ph/0303108; Phys. Lett. [**B571**]{} 67 (2003), e-Print: hep-ph/0306208;
K. Redlich, F. Karsch and A. Tawfik, J. Phys. [**G30**]{} S1271 (2004), e-Print: nucl-th/0404009;
A. Tawfik, Phys. Rev. [**D71**]{} 054502, (2005), e-Print: hep-ph/0412336.
Appendix A: Viscosity coefficient $\xi (T)$ from LQCD {#App:C .unnumbered}
=====================================================
Following the discussion presented in [[@Cheng:2007jq]]{}, the bulk viscosity of QGP can be calculated from the lattice QCD by Eq. (13) in that paper. We assume that the decay factors for pions and kaons are vanishing above the critical temperature of the phase transition QGP-hadrons. The quark-antiquark condensates can be neglected at temperatures higher than the critical one [@TawDom]. Therefore, Eq. (22) of Ref. [[@Cheng:2007jq]]{} would be reduced to $$\label{ze}
9\,\omega_0\,\xi = T\, s\, \left(\frac{1}{c_s^2}-3\right)-4(\rho-3p) +16|\epsilon_v|$$ where $\rho$ is the energy density and $c_s^2=dp/d\rho $ is the square of the speed of sound. The parameter $\omega_o$ is a scale depending on the temperature $T$, and defines the validity of the underlying perturbation theory. In this relation, the viscosity is assumed to have a thermal part which can be determined through lattice calculations, and a vacuum contributing part, which can be fixed using quark and gluon condensates. The vacuum part would take the value $$16 |\epsilon_v| (1 + \frac{3}{8} \cdot 1.6) \simeq (560\ {\rm MeV})^4
\simeq (3 \,T_c)^4 \,$$ Our algorithm is the following. Using lattice QCD results on trace anomaly, $(\epsilon-3p)/T^4$, and other thermodynamical quantities, we can determine the bulk viscosity. To make use of the lattice QCD results, it is useful to make a suitable fit to the data at high temperatures. Then we obtain the following equations of state $$\begin{aligned}
\label{EoS}
p &=&\omega \rho, \hspace*{2cm} T =\beta \rho^r, \hspace*{2cm} c_s^2 = \omega \nonumber\end{aligned}$$ where $\omega=0.319$, $\beta=0.718\pm 0.054$ and $r=0.23\pm 0.196$. Using the equations of state, Eq. (\[EoS\]) in Eq. (\[ze\]), we obtain $$\label{zeta}
\xi(\epsilon)=\frac{1}{9\omega_o}\frac{9\gamma^2-24\gamma+16}{\gamma-1}\rho+\frac{9}{\omega_o}T_c^4.$$
Appendix B: Estimations of $g(z)$ {#App:A .unnumbered}
=================================
For analytical purposes, the function $g(z)$, which is defined in $z$ parameter as $g(z)=F_0/F_1$ in Eq. (\[gofzz1\]), can be numerically estimated depending on the parameter $z$ by using the following procedure. First, we plot it parametrically depending on the parameter $H$, Fig. (\[Figg4\]). Then we fit the resulting curve to various functions. Based on least-square fit, best choice would be a mixture of polynomial and exponential functions, $$\label{fullgofz}
g(z)= a + b\, z + c \frac{\exp(d\, z)+e}{\left[\exp(d\, z)+f\right]^2},$$ where the coefficients read $a=-2.078\pm0.117$, $b=0.091\pm0.007$ and $c=17.332\pm1.553$, $d=0.189\pm0.003$, $e=-0.814\pm0.162$ and $f=2.849\pm0.02$. At small values of $z$, it is clear that the dependence is linear, $$\label{lineargofz2}
g(z) = c + {\cal C} z.$$ Obviously, the intersect $c$ is much smaller than the slope ${\cal C}$. The sign of $g(z)$ can be flipped regarding to the sign of its independent variable $z$. Accordingly, we get $$\label{lineargofz}
g(z)\approx {\cal C} z.$$
To prove this dependence, algebraically, we try to estimate $g(z)$ directly from the division of $F_0$ by $F_1$, which can be approximated by including their first terms only, i.e. $$\label{eq.A1}
g(H)\approx \frac{3(\gamma-2)}{2[1+(1-r)\gamma]}\; H^{1-r},$$ Then, we approximate $z(H)$ to the form, $$\label{eq.A2}
z(H)\approx-\frac{3[1+(1-r)\gamma]}{2(1-r)}\; H^{1-r}.$$ Finally, we now able to derive an approximate estimation for $g(z)$. According to Eq. (\[eq.A1\]) and (\[eq.A2\]), we get $$g(z)\approx \frac{(1-r)(\gamma-2)}{[1+(1-r)\gamma]^2}\; z$$ Amazingly, this expression looks the same as the one we obtained from the numerical approximation with $${\cal C} = \frac{(1-r)(\gamma-2)}{\left[1+(1-r)\gamma\right]^2}.$$
Appendix C: Solution of Abel equation $y\dot y -y = ax$ {#App:B .unnumbered}
=======================================================
To solve Eq. (\[abel1\]) we divide the whole equation by $y^3$ and introduce a new variable $v=1/y$. Then Eq. (\[abel1\]) reads $$\frac{dv}{dx}+v^{2}+axv^{3}=0.$$ We then introduce the function $v=w/x$. $$x\frac{dw}{dx}=w-w^{2}-aw^{3}, \label{abel2}$$ Previous differential equation can be solved by separation of variables $$\int \frac{dw}{w-w^{2}-aw^{3}}=\ln C^{-1}x,$$ where $C$ is an arbitrary constant of integration. To calculate the integral, we write the function to be integrated as $$\frac{1}{w-w^{2}-aw^{3}}=\frac{1}{w}-\frac{aw}{aw^{2}+w-1}-\frac{1}{aw^{2}+w-1}.$$ Let us assume that $\Delta =1+4a>0$ (this implies that $a>0$). $$\int \frac{dw}{w-w^{2}-aw^{3}}=-\frac{1}{2\sqrt{\Delta }}\ln \frac{2aw-\sqrt{\Delta }+1}{2aw+\sqrt{\Delta }+1}-\frac{1}{2}\ln \left( aw^{2}+w-1\right)
+\ln w.$$
Therefore the general solution of Eq. (\[abel2\]) can be written as $$x=C\frac{w}{\sqrt{aw^{2}+w-1}}\left( \frac{2aw+\sqrt{\Delta }+1}{2aw-\sqrt{\Delta }+1}\right) ^{1/2\sqrt{\Delta }},$$ leading to $$y=\frac{1}{v}=\frac{x}{w}=C\frac{1}{\sqrt{aw^{2}+w-1}}\left( \frac{2aw+\sqrt{\Delta }+1}{2aw-\sqrt{\Delta }+1}\right) ^{1/2\sqrt{\Delta }}.$$
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We study strange and charmed baryon productions, $\pi + p
\rightarrow K(D) + Y(Yc)$, where $Y(Yc)$ is a strange(charmed) baryon of a ground or an excited state. We propose a new production mechanism where two quarks in the baryon participate in the reaction (two-quark process), which enables excitation of both $\lambda$- and $\rho$-modes of the baryons. To deal with the two-quark process, we consider the ’t Hooft six-quark interaction from the instanton model. We study production rates in relation with the structure of charmed baryons for the forward angle scattering.
author:
- 'Sang-In Shim'
- Atsushi Hosaka
- 'Hyun-Chul Kim'
title: Strange and Charmed Baryon Productions with an Instanton Interaction
---
Introduction
============
Recently, as new hadrons including heavy quarks such as charm and bottom have been observed continuously, experimental and theoretical researches on them have been actively conducted [@Hosaka:2016pey]. It is necessary to fix their properties such as spin-parity, structure, and reaction mechanism of hadrons. In this context, a heavy baryon can be a good testing ground for a research. In a baryon containing one heavy quark, two types of excited states from the heavy quark-diquark description, the so-called $\lambda$- and $\rho$-modes, are distinguished, providing unique features for heavy baryon structure together with heavy quark symmetry [@Yoshida:2015tia].
Based on the above background, an experimental plan for charmed baryon productions, $\pi^-+p \rightarrow D^*+Y_c$, has been made at J-PARC [@e50] and the theoretical studies on this experiment have been carried out [@Kim:2014qha]. However, in the previous theoretical work, only the $\lambda$-modes are discussed among the two kinds of excited states of heavy baryons, a further theoretical study is required to investigate both $\lambda$- and $\rho$-mode excitations.
In the present work we propose a reaction mechanism for heavy baryon productions, $\pi^-+p \rightarrow K^0 + Y$ or $D^- +Y_c$, which can excite both $\lambda$- and $\rho$-modes. We calculate matrix elements for the reaction by using a non-relativistic constituent quark model with a three-quark interaction derived from the QCD instanton model. We also calculate the production rates of various heavy baryons.
Formalism
=========
Two-quark process
-----------------
In Fig. \[fig:1\], a quark line representation for two-quark process is given. In the two-quark process, an antiquark in the pion interacts with two quarks in the proton. The two-quark process excites both $\lambda$- and $\rho$-modes. Here, $\lambda$-modes are excitations of relative motion between a diquark and a heavy quark and $\rho$-modes are excitations of diquark itself in the heavy baryon. One can also consider one-quark process in which one quark in the baryon interacts with an antiquark in the pion and possible excitations are only $\lambda$-modes [@Kim:2014qha].
In Fig. \[fig:1\], various momentum fractions carried by various quarks are shown; the momenta of the initial and the final state baryons consist of the momenta of the three quarks inside of the baryons, $\vec{P}_N=\vec{p}_1+\vec{p}_2+\vec{p}_3$, $\vec{P}_Y=\vec{p}_1+\vec{p'}_2+\vec{p'}_3$, where $\vec{p}_i$ and $\vec{p'}_i=\vec{p}_i+\vec{q}_i$ ($i=1,2,3$) are the quark momenta inside of the baryons and $\vec{q}_i$ is the transferred momentum from the initial pion to i-th quark in the heavy baryon. In the two quark process the momentum transfer $\vec{q}$ is shared by two quarks (2,3) such that $\vec{q} = \vec{P}_Y-\vec{P}_p = \vec{q}_2 + \vec{q}_3$ is the transferred momentum from the pion to the heavy baryon.
Three-quark interaction
-----------------------
For the description of two-quark processes we need a suitable interaction where three quarks participate in. Here we employ a point-like interaction of three quarks, the ’t Hooft interaction, as inspired by the instanton dynamics of QCD. The interaction works well for low-energy hadron properties including up, down and strange quarks. However, a naive extension to the charm sector may not be available, while some discussions have been made [@Chernyshev:1994-1995]. Thus, the present study is most likely to be applied to the strange sector, though some features may be discussed for the charm productions. Therefore, results will be shown both for strangeness and charm productions.
The use of the zero-range three-body interaction has a virtue that actual computations become simple. For a more realistic description, finite-range nature of the interaction may be included in terms of form factors.
The relevant ’tHooft interaction is given by [@tHooft:1976]. $$\begin{aligned}
\mathcal{L}_{tH}
= c \,\mathrm{det}[\bar{q}_i (1+\gamma_5)q_j] + H.c.
= c \left| \begin{array}{ccc}
\bar{u}(1+\gamma_5)u & \bar{u}(1+\gamma_5)d & \bar{u}(1+\gamma_5)s \\
\bar{d}(1+\gamma_5)u & \bar{d}(1+\gamma_5)d & \bar{d}(1+\gamma_5)s \\
\bar{s}(1+\gamma_5)u & \bar{s}(1+\gamma_5)d & \bar{s}(1+\gamma_5)s
\end{array} \right| + H.c.,
\label{eq:tHooft}\end{aligned}$$ where $c$ is a coupling constant, which, however, is not important in the present study, since it is difficult to determine the absolute values of the reaction cross sections. Thus, we focus here only on relative production rates.
Matrix elements and production rates
====================================
In a non-relativistic quark model, baryon wave functions can be written as a product of the plane wave for the center-of-mass motion and bound state wave functions in internal coordinates, Jacobi coordinates $\lambda$ and $\rho$. Then, the transition amplitudes for the reaction $\pi^- p \to M Y$ can be written as $$\begin{aligned}
&{\langle}\, Y \, M | \mathcal{L}_{tH}
\, | \, N \, \pi^- {\rangle}\cr
&\propto\,\delta^{(3)}\big(\vec{P}_Y -\vec{P}_N-\vec{q}\big)
\cr
&\hspace{0.3cm}\times \int d^3 q_2\,d^3 q_3
\,\delta^{(3)}\big(\vec{q} -\vec{q}_2-\vec{q}_3\big)
\int d^3 \rho e^{i \vec{q}_\rho \cdot \vec{\rho}}
\psi^{\rho *}_{l_{\rho}}(\vec{\rho}) \psi^{\rho }_{0} (\vec{\rho})
\int d^3 \lambda e^{i \vec{q}_\lambda \cdot \vec{\lambda}}
\psi^{\lambda' *}_{l_{\lambda}}(\vec{\lambda}) \psi^{\lambda}_{0} (\vec{\lambda})
\nonumber \\
&\hspace{0.2cm}
+ (1 \leftrightarrow 2, \, \vec{\rho} \rightarrow -\vec{\rho}) \end{aligned}$$ Having performed the integration with respect to $q_2$, $q_3$, and $\rho$, $\lambda$, we obtain the matrix elements for heavy baryon productions as $$\begin{aligned}
{\langle}Y(l_\rho, l_\lambda) \hspace{0.1cm} M | \mathcal{L}_{tH} |
p\hspace{0.1cm} \pi^- {\rangle}&=C_{YM} \, I_{l} \,
(2\pi)^3\delta^{(3)}\big(\vec{P}_Y
-\vec{P}_p-\vec{q}\big)
\label{eq:TransAmpGn}\end{aligned}$$ where $I_{l}$ is given as $$\begin{aligned}
I_{0} &\equiv
\left(
\frac{16\pi \alpha^2_{\rho} \alpha_{\lambda'}\alpha_{\lambda}}{
B^2}
\right)^{\frac{3}{2}}
e^{-q^2_{eff}/(4 B^2)},\end{aligned}$$ for $l=0$ $$\begin{aligned}
I_{l_\lambda=1} &\equiv
\frac{i\sqrt{2}\alpha_{\lambda'} |\vec{q}_{eff}|}{
2 B^2}
\left(
\frac{16\pi \alpha^2_{\rho} \alpha_{\lambda'}\alpha_{\lambda}}{
B^2}
\right)^{\frac{3}{2}}
e^{-q^2_{eff}/(4 B^2)}\end{aligned}$$ and $$\begin{aligned}
I_{l_\rho=1} &\equiv
\frac{-i\sqrt{2}\alpha_{\rho} |\vec{q}_{eff}|}{
B^2}
\left(\frac{16\pi \alpha^2_{\rho} \alpha_{\lambda'}\alpha_{\lambda}}{
B^2}\right)^{\frac{3}{2}}
e^{-q^2_{eff}/(4 B^2)}\end{aligned}$$ for $l=1$. The coefficients $C_{YM}$ are the prefactors from spin and isospin calculations. The effective momentum transfer, $\vec{q}_{eff}$ and $B^2$ are defined as $$\begin{aligned}
\vec{q}_{eff}
&\equiv\frac{m_d }{m_d + m_q}\vec{P}_N - \frac{m_d }{m_d + m_Q}\vec{P}_Y,
\hspace{0.2cm}
B^2 \equiv \frac{8 \alpha_\rho^2+\alpha_{\lambda'}^2+\alpha_{\lambda}^2}{2} \end{aligned}$$ where $m_d$, $m_q$, and $m_Q$ are the masses of a diquark, light quarks ($u$, $d$ quark), and the heavy quark, respectively. $\alpha_{\rho}$, $\alpha_{\lambda}$, and $\alpha_{\lambda'}$ are the oscillator parameters for the $\rho-$modes, initial and final state $\lambda-$modes, respectively. Here, except for the delta function, the matrix elements Eq. (\[eq:TransAmpGn\]) depend on $\vec{q}_{eff}$ instead of $\vec{q}$ because of the so-called recoil effect due to the change in the masses of particles before and after the interaction.
In the center of mass frame, the differential cross sections for the heavy baryon productions can be written as $$\begin{aligned}
\mathcal{R}\big(Y(J^{p}, J_z) \big)
=\frac{1}{4|p_{i}|\sqrt{s}} |C_{YM}|^2 |I_l|^2 \frac{|\vec{p}_f|}{4\pi \sqrt{s}}.
\label{eq:ProdRates}\end{aligned}$$
Discussion
==========
The production rates of hyperons and charmed baryons are given in Table \[tab:1\]. In the present work, we calculate the production rates of various heavy baryon productions. Since this work is the first attempt of the two-quark process, we consider a simple case, i.e. the forward angle scattering. Those at finite angles is left for future works. To demonstrate the production rates, we set the momentum of the pion at $k^{Lab}_\pi = 5\, \mathrm{GeV}$ for the hyperons and $k^{Lab}_\pi = 20\, \mathrm{GeV}$ for charmed baryons. These momenta provide sufficient energies to create $s\bar s$ or $c \bar c$ pair. In the two-quark process, the momentum $\vec{q}$ is shared by the heavy quark and a light quark in the heavy baryon, which may excite both $\lambda$- and $\rho$-modes. This contrasts with the one-quark process where only one quark receives the transferred momentum and possible excitations are only $\lambda$-modes.
A more detailed discussion about momentum-transfer and structure dependences of the production rates will be given elsewhere.
[lccccccc]{}\
$l=0$ & $\Lambda\left(\frac{1}{2}^+\right)$ & $\Sigma\left(\frac{1}{2}^+\right)$ & $\Sigma\left(\frac{3}{2}^+\right)$ & & & &\
$\mathcal{R}$($Y_s$) & 1 & 3.2 & 0 & & & &\
$\mathcal{R}$($Y_c$) & 1 & 2.9 & 0 & & & &\
\
$l_\lambda=1$ & $\Lambda\left(\frac{1}{2}^-\right)$ & $\Lambda\left(\frac{3}{2}^-\right)$ & $\Sigma\left(\frac{1}{2}^-\right)$ & $\Sigma\left(\frac{1}{2}^-\right)$ & $\Sigma\left(\frac{3}{2}^-\right)$ & $\Sigma\left(\frac{3}{2}^-\right)$ & $\Sigma\left(\frac{5}{2}^-\right)$\
& $j=1$ & $j=1$ & $j=0$ & $j=1$ & $j=1$ & $j=2$ & $j=2$\
$\mathcal{R}$($Y_s$) & 0.004& 0.010& 0.007& 0.015& 0.007& 0.038 & 0\
$\mathcal{R}$($Y_c$) & 0.10 & 0.20 & 0.12 & 0.23 & 0.12 & 0.58 & 0\
\
$l_\rho=1$ & $\Lambda\left(\frac{1}{2}^-\right)$ & $\Lambda\left(\frac{1}{2}^-\right)$ & $\Lambda\left(\frac{3}{2}^-\right)$ & $\Lambda\left(\frac{3}{2}^-\right)$ & $\Lambda\left(\frac{5}{2}^-\right)$ & $\Sigma \left(\frac{1}{2}^-\right)$ & $\Sigma \left(\frac{3}{2}^-\right)$\
& $j=0$ & $j=1$ & $j=1$ & $j=2$ & $j=2$ & $j=1$ & $j=1$\
$\mathcal{R}$($Y_s$) & 0.017& 0.039 & 0.018& 0.10 & 0 & 0.016& 0.032\
$\mathcal{R}$($Y_c$) & 0.22 & 0.43 & 0.22 & 1.1 & 0 & 0.20 & 0.41\
\[tab:1\]
Acknowledgments {#acknowledgments .unnumbered}
===============
SIS and AH are supported in part by Grant-in-Aid for Scientific Research on Innovative Areas, Clustering as a window on the hierarchical structure of quantum systems. HChK is supported by the Basic Science Research Program through the National Research Foundation (NRF) of Korea funded by the Korean government (MEST): Grant No. 2018R1A5A1025563.
[9]{}
A. Hosaka, T. Iijima, K. Miyabayashi, Y. Sakai and S. Yasui, PTEP [**2016**]{}, no. 6, 062C01 (2016) T. Yoshida, E. Hiyama, A. Hosaka, M. Oka and K. Sadato, Phys. Rev. D [**92**]{}, no. 11, 114029 (2015) J-PARC P50 Proposal, “Charmed Baryon Spectroscopy via the $(\pi^-,D^{*-})$ reaction”, Jan, 2012:\
`*+http://www.j-parc.jp/researcher/Hadron/en/Proposal_e.html#1301+`
S. H. Kim, A. Hosaka, H. C. Kim, H. Noumi and K. Shirotori, PTEP [**2014**]{}, no. 10, 103D01 (2014) S. Chernyshev, M. A. Nowak and I. Zahed, Phys. Rev. D [**53**]{}, 5176 (1996) S. Chernyshev, M. A. Nowak and I. Zahed, Phys. Lett. B [**350**]{}, 238 (1995) G. ’t Hooft, Phys. Rev. Lett. [**37**]{}, 8 (1976); Phys. Rev. D [**14**]{}, 3432 (1976) Erratum: \[Phys. Rev. D [**18**]{}, 2199 (1978)\].
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
[^1]\
H. Niewodniczański Institute of Nuclear Physics, Polish Academy of Sciences,\
ul. Radzikowskiego 152, 31-342 Krakow, Poland\
E-mail:
- |
A. Kusina\
Southern Methodist University, Dallas, TX 75275, USA\
E-mail:
- |
M. Skrzypek\
H. Niewodniczański Institute of Nuclear Physics, Polish Academy of Sciences,\
ul. Radzikowskiego 152, 31-342 Krakow, Poland\
E-mail:
- |
M. Slawinska\
H. Niewodniczański Institute of Nuclear Physics, Polish Academy of Sciences,\
ul. Radzikowskiego 152, 31-342 Krakow, Poland\
E-mail:
title: |
NLO parton shower for LHC physics -\
hard processes and beyond
---
Introduction
============
The Large Hadron Collider (LHC) at CERN provides rich harvest of experimental data. The proper understanding and interpretation of these data, possibly leading to discovery of new phenomena, requires perfect mastering of the “trivial” effects due to the multiple emissions of soft and collinear gluons and quarks. Perturbative Quantum Chromodynamics (pQCD) [@GWP; @Gross:1974cs; @Georgi:1951sr], supplemented with clever modelling of the low energy nonperturbative effects, is an indispensable tool for disentangling the Standard Model physics component in the data. This work presents part of the global effort of improving quality of the pQCD calculations for LHC experiments.
Most of the results presented here are described in refs. [@Jadach:2011cr] and [@Jadach:2012vs]. Although this work elaborates on the improved method of the pQCD calculation combining NLO-corrected hard process and LO parton shower Monte Carlo (MC), it should be regarded as the first step towards NNLO-corrected hard process combined with the NLO parton shower MC [@IFJPAN-IV-2012-7].
Basic LO parton shower MC
=========================
The multigluon distribution of the single initial state ladder, which is a building block of our parton shower MC, is represented by the integrand of the “exclusive/unintegrated PDF”, which in the LO approximation is the following: $$\label{eq:LOMC}
\begin{split}
&
D(t,x)
=\int dx_0\; dZ\;
\delta_{x=x_0 Z}\;
d_0(\hat{t}_0,x_0)\; G(t, \hat{t}_0- \ln x_0 | Z),
\\&
G(t, t_0 | Z)=
e^{-S_F} \sum_{n=0}^\infty
\bigg(
\prod_{i=1}^n
\int d^3{{\cal E}}(\bar{k}_i)\;
\theta_{\xi_i>\xi_{i-1}}
\frac{2C_F\alpha_s}{\pi^2} \bar{P}(z_i)
\bigg)
\\&
\qquad\qquad\quad\times
\theta_{t>\xi_n}
\delta_{Z=\prod_{j=1}^n z_j},
\end{split}$$ where evolution kernel is $\bar{P}(z)=\frac{1}{2}(1+z^2)$, evolution time is $\hat{t}_0=\ln(q_0/\Lambda)$ and the “eikonal” phase space integration element is $
d^3{{\cal E}}(k)=\frac{d^3 k}{2k^0}\;\frac{1}{{{\bf{k}}}^2}
=\pi \frac{d\phi}{2\pi} \frac{d k^+}{k^+} d \xi
$ and $k^\pm = k^0\pm k^3$. We use rapidities $\xi_i =\frac{1}{2}\ln\frac{k^-_i}{k^+_i}\big|_{\rm Rh}$ in the hadron beam rest frame (Rh), and $\eta_i=\frac{1}{2}\ln\frac{k^+_i}{k^-_i}\big|_{\rm RFHP}$ defined in hard process rest frame (RFHP). They are related by $\xi_i=\ln\frac{\sqrt{s}}{m_h}-\eta_i$. Rapidity ordering is now $t=\xi_{\max}>\xi_n>\dots>\xi_i>\xi_{i-1}>\dots>\xi_0=t_0$, where $t_0=\xi_0=\ln(q_0/m_h)-\ln x_0$. The direction of the $z$ axis in the RFHP is pointing out towards the hadron momentum. A lightcone variable of the emitted gluon is defined as $\alpha_i= \frac{2k_i^+}{\sqrt{s}}$ and of the emitter parton (quark) as $x_i=x_0-\sum_{j=0}^{i}\; \alpha_j$ (after $i$ emissions). We also use fractions $z_i=x_i/x_{i-1}$. The Sudakov formfactor $S_F$ comes from the “unitarity” condition[^2] $
\int_0^1 dZ\; G(t, t_0 | Z)=1,
$ which is also instrumental in the Markovian MC implementation used to obtain $D(t,x)$ at any value of $t>t_0$.
The initial distribution $d_0(q_0,x_0)$ related to experiment, to previous steps in the MC ladder, or to PDF in the standard $\overline{MS}$ system is not essential for the following discussion, we only note that the unitarity condition provides baryon number conservation sum rule $\int_0^1 dx\; D(t,x) = \int_0^1 dx_0\; d_0(t_0,x_0)$.
For testing our new method of correcting hard process to the NLO level we use the following simplified MC parton shower, implementing the DY process with two ladders and the hard process: [^3] $$\label{eq:LOMCFBmaster}
\begin{split}
&\sigma_0=
\int d x_{0F} d x_{0B}\;\;
d_0(\hat{t}_0,x_{0F}) d_0(\hat{t}_0,x_{0B})
\sum_{n_1=0}^\infty\;
\sum_{n_2=0}^\infty
\int dx _F\; dx_B\;
\\&~~~~~~~~~\times
e^{-S_{_F}}
\int_{\Xi<\eta_{n_1}}
\bigg(
\prod_{i=1}^{n_1}
d^3{{\cal E}}(\bar{k}_i)
\theta_{\eta_i<\eta_{i-1}}
\frac{2C_F\alpha_s}{\pi^2} \bar{P}(z_{Fi})
\bigg)
\delta_{x_F = x_{0F}\prod_{i=1}^{n_1} z_{Fi}}
\\&~~~~~~~~~\times
e^{-S_{_B}}
\int_{\Xi>\eta_{n_2}}
\bigg(
\prod_{j=1}^{n_2}
d^3{{\cal E}}(\bar{k}_j)
\theta_{\eta_j>\eta_{j-1}}
\frac{2C_F\alpha_s}{\pi^2} \bar{P}(z_{Bj})
\bigg)
\delta_{x_B = x_{0B}\prod_{j=1}^{n_2} z_{Bj}}
\\&~~~~~~~~~\times
d\tau_2(P-\sum_{j=1}^{n_1+n_2} k_j ;q_1,q_2)\;
\frac{d\sigma_B}{d\Omega}(sx_Fx_B,\hat\theta)\;
W^{NLO}_{MC}.
\end{split}$$ In the LO approximation $W^{NLO}_{MC}=1$. Rapidity $\xi$ is translated into $\eta$ – the center of mass system rapidity, in the forward part (F) of the phase space as $\xi_i=\ln\frac{\sqrt{s}}{m_h}-\eta_i$, $\eta_{0F}>\eta_i>\Xi$, and in the backward (B) part as $\xi_i=-\ln\frac{\sqrt{s}}{m_h}+\eta_i$, $\Xi>\eta_i>\eta_{0B}$. The rapidity boundary between the two hemispheres $\Xi=0$ is used, until a more sophisticated version related to rapidity of the produced $W/Z$ is introduced.
Analytical integration of eq. (\[eq:LOMCFBmaster\]) results in the standard factorization formula ($W^{NLO}_{MC}=1$) $$\label{eq:LOfactDY2LO}
\sigma_0 =
\int_0^1 dx_F\;dx_B\;
D_F(t, x_F)\; D_B(t, x_B)\;
\sigma_B(sx_Fx_B).$$ The distributions $D_F(t, x_F)=(d_0\otimes G_F) (t, x_F)$ and $D_B(t, x_B)=(d_0\otimes G_B) (t, x_B)$ are obtained from separate Markovian LO Monte Carlo runs. The above LO formula is exact, and can be tested with an arbitrary numerical precision.
Figure \[fig:etaW\_LO\_7TeV\] represents a “calibration benchmark” for the overall normalization at the LO level. We show there the properly normalized distribution of the variable $\eta_W^*=\frac{1}{2}\ln(x_F/x_B)$, which in the collinear limit approximates the rapidity of $W$ boson. The distribution in the upper plot of Fig. \[fig:etaW\_LO\_7TeV\], representing eq. (\[eq:LOfactDY2LO\]), is obtained using the general purpose MC program FOAM [@foam:2002]. The collinear PDF $D(t,x)$ there has been obtained from a separate high statistics run ($10^{10}$ events) of a Markovian MC (MMC), creating $D(t,x)$ in a form of the 2-dimensional look-up table[^4]. The other distribution in the upper plot of Fig. \[fig:etaW\_LO\_7TeV\] represents eq. (\[eq:LOMCFBmaster\]) in LO approximation. It comes from the full scale MC generation (with four-momenta conservation). The MC run with $10^8$ events was used. The constrained MC (CMC) technique of ref. [@Jadach:2007qa] is used here because of the narrow Breit-Wigner peak due to a heavy boson propagator[^5]. Two CMC modules and FOAM are combined into one MC generating gluon emissions and the $W$ boson production. FOAM is taking care of the generation of the variables $x_F,x_B,x_{F0},x_{B0}$ and the sharp Breit-Wigner peak in $\hat{s}=s x_F x_B$, then two CMC modules are initialized and generate the gluon four-momenta $\bar{k}^\mu_j$. They are mapped into $k^\mu_j$, following the prescription defined in ref. [@Jadach:2011cr], such that the overall energy-momentum conservation is achieved. Figure \[fig:etaW\_LO\_7TeV\] demonstrates a very good numerical agreement between $d\sigma/ d \eta_W^*$ from our full scale LO parton shower MC of eq. (\[eq:LOMCFBmaster\]) and the simple formula of eq. (\[eq:LOfactDY2LO\]), to within 0.5%, as seen from the ratio of the two results in the lower part of the figure.
Introducing NLO corrections to hard process
===========================================
The NLO corrections to hard process are imposed on top of the LO distributions of eq. (\[eq:LOMCFBmaster\]) using a single “monolithic” weight $W^{NLO}_{MC}$ defined exactly as in ref. [@Jadach:2011cr]: $$\label{eq:NLODYMCwt}
\begin{split}
&W^{NLO}_{MC}=
1+\Delta_{S+V}
+\sum_{j\in F}
\frac{{{\tilde{\beta}}}_1(q_1,q_2,\bar{k}_j)} {\bar{P}(z_{Fj})\;d\sigma_B(\hat{s},\hat\theta)/d\Omega}
+\sum_{j\in B}
\frac{{{\tilde{\beta}}}_1(q_1,q_2,\bar{k}_j)} {\bar{P}(z_{Bj})\;d\sigma_B(\hat{s},\hat\theta)/d\Omega},
\end{split}$$ the NLO soft+virtual correction is $
\Delta_{V+S}
=\frac{C_F \alpha_s}{\pi}\; \left( \frac{2}{3}\pi^2 -\frac{5}{4} \right)
$, and the real correction reads: $$\label{eq:DYbeta1FB}
\begin{split}
&{{\tilde{\beta}}}_1(q_1,q_2,k)=
\Big[
\frac{(1-\beta)^2}{2}
\frac{d\sigma_{B}}{d\Omega_q}(\hat{s},\theta_{F})
+\frac{(1-\alpha)^2}{2}
\frac{d\sigma_{B}}{d\Omega_q}(\hat{s},\theta_{B})
\Big]
\\&~~~~~~~~~~~~~~~~~~~~
-\theta_{\alpha>\beta}
\frac{1+(1-\alpha-\beta)^2}{2}
\frac{d\sigma_{B}}{d\Omega_q}(\hat{s},\hat\theta)
-\theta_{\alpha<\beta}
\frac{1+(1-\alpha-\beta)^2}{2}
\frac{d\sigma_{B}}{d\Omega_q}(\hat{s},\hat\theta).
\end{split}$$ The above is the exact ME of the quark-antiquark annihilation into a heavy vector boson with additional single real gluon emission[^6]. The LO component, which is already included in the LO MC, is subtracted here. The variable $\hat{s}=s x_F x_B = (q_1+q_2)^2$ is the effective mass squared of the heavy vector boson. The definition of angle $\hat\theta$ in the LO component is rather arbitrary. We define it in the rest frame of the heavy boson, where $\vec{q}_1+\vec{q}_2=0$, as an angle between the decay lepton momentum $\vec{q}_1$ and the difference of momenta of the incoming quark and antiquark $\hat\theta=\angle(\vec{q}_1,\vec{p}_{0F}-\vec{p}_{0B})$. On the other hand the two angles in the NLO ME are defined quite unambiguously as $\hat\theta_F=\angle(\vec{q}_1,-\vec{p}_{0B})$ and $\hat\theta_B=\angle(\vec{q}_1, \vec{p}_{0F})$. In the above we only need directions of the $\vec{p}_{0F}$ and $\vec{p}_{0B}$ vectors, which are the same as the directions of the hadron beams. The lightcone variables $\alpha_j$ and $\beta_j$ of the emitted gluon are defined in the F and B parts of the phase space as follows[^7]: $$\begin{split}
&\alpha_j=1-z_{Fj},\quad
\beta_j= \alpha_j\; e^{2(\eta_j-\Xi)},\quad ~~
{\rm for}~~~ j\in F,
\\&
\beta_j=1-z_{Bj},\quad
\alpha_j=\beta_j\; e^{-2(\eta_j-\Xi)},\quad
{\rm for}~~~ j\in B.
\end{split}$$
Again, the exact phase space integration of eq. (\[eq:LOMCFBmaster\]) including $W^{NLO}_{MC}$ of eq. (\[eq:NLODYMCwt\]) is feasible, and the resulting compact expression for the total cross section is obtained [@Jadach:2011cr]: $$\label{eq:DYanxch}
\begin{split}
\sigma_1 &=
\int_0^1 dx_F\;dx_B\; dz\;
D_F(t, x_F)\;D_B(t, x_B)\;
\sigma_B(szx_Fx_B)
\big\{
\delta_{z=1}(1+\Delta_{S+V})
+C_{2r}(z)
\big\},
\end{split}$$ where $
C_{2r}(z) =\frac{2C_F \alpha_s}{\pi}\; \left[ -\frac{1}{2}(1-z) \right].
$
Numerical test of NLO correction
--------------------------------
Figure \[fig:etaW\_NLOglu\_7TeV\] represents a principal [*proof of concept*]{} of our new methodology for implementing the NLO corrections to the hard process in the parton shower MC. The plotted NLO correction to the $\eta_W^*$ distribution[^8] comes from the parton shower MC with the NLO-corrected hard process according to eqs. (\[eq:LOMCFBmaster\]) and (\[eq:NLODYMCwt\]). Additionally we also plot there result of a simple collinear formula of eq. (\[eq:DYanxch\]), where two collinear PDFs are convoluted with the analytical coefficient function $C_{2r}(z)$ for the hard process. Both results coincide within the statistical error, see their ratio in the lower part of Fig. \[fig:etaW\_NLOglu\_7TeV\].
Technically, the inclusion of the NLO correction in our parton shower MC is rather straightforward, and is obtained by including $W^{NLO}_{MC}$ weight of eq. (\[eq:NLODYMCwt\]). MC is providing both LO and NLO-corrected results in a single run with weighted events. The NLO weight is strongly peaked near $W^{NLO}_{MC}=1$, positive, and without long-range tails. Its distribution is shown in Fig. \[fig:Canv1w\_7TeV\].
In all numerical results we have set $\Delta_{V+S}=0$, as it is completely unimportant for the presented analysis. The initial distributions $d_0(q_0,x_0)$ are defined in ref. [@Jadach:2012vs].
Simplification of the method and comparison with other methodologies
====================================================================
{width="75mm"} {width="75mm"} \[fig:mcRho1glu\]
Our new method for introducing NLO corrections in the hard process, proposed in ref. [@Jadach:2011cr] and tested in ref. [@Jadach:2012vs], is an alternative to the two well established MC@NLO [@Frixione:2002ik] and POWHEG [@Nason:2004rx; @Frixione:2007vw] methodologies. With MC numerical implementation at hand, let us elaborate on the differences with the above two techniques in particular with the POWHEG technique. We shall also see that it is possible to make our method more efficient in terms of CPU time consumption. This improvement is not so critical in the present case of NLO corrected hard process, but may be quite useful in the case of correcting evolution kernels to the NLO in the ladder parts of the MC [@IFJPAN-IV-2012-7].
The most important differences with the POWHEG and MC@NLO techniques are:
- The summation over all emitted gluons, without deciding which gluon is the one involved in the NLO correction and which ones are merely “LO spectators” in the parton shower.
- The absence of $(1/(1-z))_+$ distributions in the real part of the NLO corrections (virtual+soft correction is kinematically independent).
To explain more clearly how $W^{NLO}_{MC}$ of eq. (\[eq:NLODYMCwt\]) is distributed over the multigluon phase space, we restrict now to single ladder (hemisphere) with a simplified weight: $$\label{eq:NLODYMCwt_sim}
\begin{split}
&W^{NLO}_{MC}= 1+\sum_{j\in F} W^{NLO}_j,\qquad
W^{NLO}_j=
\frac{{{\tilde{\beta}}}_1(q_1,q_2,\bar{k}_j)} {\bar{P}(z_{Fj})\;d\sigma_B(\hat{s},\hat\theta)/d\Omega}.
\end{split}$$ In order to find out the phase space regions specific for NLO corrections we consider inclusive distributions of gluons on the Sudakov logarithmic plane of rapidity $\xi$ and variable $v=\ln(1-z)$. In the left hand side (LHS) of Fig. \[fig:mcRho1glu\] we show gluons inclusive distribution in the LO approximation. The flat plateau there represents IR/collinear singularity[^9] $2C_F\frac{\alpha_S}{\pi} d\xi\frac{dz}{1-z}$ with the drop by factor 1/2 towards $z=0$, due to $\frac{1+z^2}{2}$ factor in the LO kernel. In the right hand side (RHS) of Fig. \[fig:mcRho1glu\] we show contributions from all gluons weighted with the component weight[^10] $-W^{NLO}_j$ of eq. (\[eq:NLODYMCwt\_sim\]). The NLO contribution is concentrated in the area near the hard process rapidity $t=\xi_{\max}$, which has to be true for the genuine NLO contribution [^11]. The completeness of the phase space near this important region ($z=0$, $\xi_{\max}$) is critical for the completeness of the NLO corrections. Both POWHEG and MC@NLO use standard LO MCs which feature an empty “dead zone” in this phase space corner.
Figure \[fig:mcRho1glu\] suggests that the dominant contribution to $\sum_j W^{NLO}_j$ could be from the gluon with the maximum $\ln k_j^T\sim \xi_j+\ln(1-z_j)$, which is closest to the hard process phase space corner. In the MC we may easily relabel generated gluons using new index $K$ such that they are ordered in the variable $\kappa_K=\xi_K+\ln(1-z_K),\; \kappa_{K+1}<\kappa_K$ with $K=1$ being the hardest one.
Figure \[fig:mcCanv1k\] demonstrates a split of the LO inclusive distribution of Fig. \[fig:mcRho1glu\] into the $K=1$ component and the rest $K>1$. The important point is that the $K=1$ component reproduces the original complete distribution over the whole region where the NLO correction is non-negligible! This is exactly the observation on which POWHEG technique is built. According to the POWHEG authors, taking the $K=1$ component is sufficient to reproduce the complete NLO correction (modulo NNLO).
The above statement is checked numerically in Fig. \[fig:mcCanV8k\], where we compare the NLO correction to the $x=\prod_j z_j$ distribution from the complete sum $\sum_j W^{NLO}_j$ and from $ W^{NLO}_{K=1}$. As we see the $K=1$ component saturates the complete sum very well, with the $K=2$ component being negligible in the first approximation.
We can therefore speed up the calculation by means of taking only the $K=1$ contribution. The price will be that the formula of eq. (\[eq:DYanxch\]) will not be exact any more. Our method differs, however, from the POWHEG scheme, where the $K=1$ gluon is generated separately in the first step, and other gluons are generated (by the LO parton shower MC) in the next step. That is easy for LO MC with $k^T$-ordering, while in case of the LO MC with angular-ordering POWHEG requires additional effort of generating the so called vetoed and truncated showers. In our method, there is no need for such vetoed/truncated showers in case of angular ordering.
The reason why POWHEG technique is complicated in case of the angular ordering is illustrated in Fig. \[fig:mcRhoOrdLO\]. We show there the distribution of gluons ordered in rapidity, starting from the gluon with the maximum rapidity, the closest to hard process. The gluon distribution with the highest rapidity $\xi\sim \xi_{\max}$ ($J=1$) has a ridge extending towards the soft region. Notice that, when the IR cut-off $\epsilon\to0$ in $(1-z)<\epsilon$, the width of this ridge also goes to zero. Consequently, the gluon with the highest $\xi$ is unable to reproduce the gluon distribution in the NLO corner, close to hard process. This is why in this case POWHEG requires truncated and vetoed showers, which are not needed in our method.
Summary and outlook
===================
A new method of adding the QCD NLO corrections to the hard process in the initial state Monte Carlo parton shower is tested numerically showing that the basic concept of the new methodology works correctly in the numerical environment of a Monte Carlo parton shower. The differences with the well established methods of MC@NLO and POWHEG are briefly discussed. Also, variants of the new method with better efficiency in terms of CPU time are proposed.
Acknowledgement {#acknowledgement .unnumbered}
===============
This work is partly supported by the Polish National Science Centre grant UMO-2012/04/M/ST2/00240, Foundation for Polish Science grant Homing Plus/2010-2/6, the Research Executive Agency (REA) of the European Union Grant PITN-GA-2010-264564 (LHCPhenoNet), the U.S. Department of Energy under grant DE-FG02-04ER41299 and the Lightner-Sams Foundation.
[99]{}
D. J. Gross and F. Wilczek, [*Phys. Rev. Lett.*]{} [**30**]{} (1973) 1343;\
H. D. Politzer, [*Phys. Rev. Lett.*]{} [**30**]{} (1973) 1346;\
D. J. Gross and F. Wilczek, [*Phys. Rev.*]{} [**D8**]{} (1973) 3633;\
H. D. Politzer, [ *Phys. Rep.*]{} [**14**]{} (1974) 129.
D. J. Gross and F. Wilczek, [*[Asymptotically Free Gauge Theories. 2]{}*]{}, [*Phys. Rev.*]{} [**D9**]{} (1974) 980–993.
H. Georgi and H. D. Politzer, [*[Electroproduction scaling in an asymptotically free theory of strong interactions]{}*]{}, [*Phys. Rev.*]{} [ **D9**]{} (1974) 416–420.
S. Jadach, A. Kusina, W. Placzek, M. Skrzypek, and M. Slawinska, [*[On the inclusion of the QCD NLO corrections in the quark– gluon Monte Carlo shower]{}*]{}, [[1103.5015]{}](http://xxx.lanl.gov/abs/1103.5015).
S. Jadach, M. Jezabek, A. Kusina, W. Placzek, and M. Skrzypek, [*[NLO corrections to hard process in QCD shower – proof of concept]{}*]{}, [[1209.4291]{}](http://xxx.lanl.gov/abs/1209.4291).
S. Jadach, A. Kusina, and Skrzypek, [*[NLO corrections to ladder part of the initial state shower in QCD]{}*]{}, 2012. Report IFJPAN-IV-2012-7, in preparation.
S. Jadach, [*[Foam: A general purpose cellular Monte Carlo event generator]{}*]{}, [*Comput. Phys. Commun.*]{} [**152**]{} (2003) 55–100, \[[[physics/0203033]{}](http://xxx.lanl.gov/abs/physics/0203033)\].
S. Jadach, W. Placzek, M. Skrzypek, and P. Stoklosa, [*[Markovian Monte Carlo program EvolFMC v.2 for solving QCD evolution equations]{}*]{}, [ *Comput. Phys. Commun.*]{} [**181**]{} (2010) 393–412, \[[[0812.3299]{}](http://xxx.lanl.gov/abs/0812.3299)\].
K. Golec-Biernat, S. Jadach, W. Płaczek, and M. Skrzypek, [*[Markovian Monte Carlo solutions of the NLO QCD evolution equations]{}*]{}, [*Acta Phys. Polon.*]{} [**B37**]{} (2006) 1785–1832, \[[[hep-ph/0603031]{}](http://xxx.lanl.gov/abs/hep-ph/0603031)\].
S. Jadach, W. Placzek, M. Skrzypek, P. Stephens, and Z. Was, [*[Constrained MC for QCD evolution with rapidity ordering and minimum kT\*]{}*]{}, [ *Comput.Phys.Commun.*]{} [**180**]{} (2009) 675–698, \[[[hep-ph/0703281]{}](http://xxx.lanl.gov/abs/hep-ph/0703281)\].
T. Sjostrand, [*A model for initial state parton showers*]{}, [*Phys. Lett.*]{} [**B157**]{} (1985) 321.
F. A. Berends and R. Kleiss, [*[Initial State Radiation for e+ e- Annihilation Into Jets]{}*]{}, [*Nucl.Phys.*]{} [**B178**]{} (1981) 141.
S. Alioli, K. Hamilton, and E. Re, [*[Practical improvements and merging of POWHEG simulations for vector boson production]{}*]{}, [*JHEP*]{} [**1109**]{} (2011) 104, \[[[1108.0909]{}](http://xxx.lanl.gov/abs/1108.0909)\].
S. Frixione and B. R. Webber, [*[Matching NLO QCD computations and parton shower simulations]{}*]{}, [*JHEP*]{} [**06**]{} (2002) 029, \[[[hep-ph/0204244]{}](http://xxx.lanl.gov/abs/hep-ph/0204244)\].
P. Nason, [*[A new method for combining NLO QCD with shower Monte Carlo algorithms]{}*]{}, [*JHEP*]{} [**11**]{} (2004) 040, \[[[hep-ph/0409146]{}](http://xxx.lanl.gov/abs/hep-ph/0409146)\].
S. Frixione, P. Nason, and C. Oleari, [*[Matching NLO QCD computations with Parton Shower simulations: the POWHEG method]{}*]{}, [*JHEP*]{} [**0711**]{} (2007) 070, \[[[0709.2092]{}](http://xxx.lanl.gov/abs/0709.2092)\].
[^1]: The partial support of the TH Unit of the CERN PH Division for this author is acknowledged.
[^2]: The usual cutoff $1-z<\epsilon$ regularizing the IR singularity is implicit.
[^3]: Following ref. [@Jadach:2011cr], we adopt $d\tau_2(P;q_1,q_2) =
\delta^{(4)}(P-q_1-q_2)\frac{d^3q_1}{2q_1^0}\frac{d^3q_2}{2q_2^0}$.
[^4]: This MMC run solves the LO DGLAP equation using the MC method, as in refs. [@Jadach:2008nu; @GolecBiernat:2006xw].
[^5]: A backward evolution algorithm of ref. [@Sjostrand:1985xi] could be also used here.
[^6]: We employ here the compact representation of ref. [@Berends:1980jk], which has also been used in POWHEG [@Alioli:2011nr].
[^7]: See ref. [@Jadach:2011cr] for more explanations.
[^8]: Extra minus sign introduced to facilitate visualization.
[^9]: We use constant $\alpha_S$.
[^10]: We again insert a minus sign in order to facilitate visualization.
[^11]: It also vanishes towards the soft limit $z\to 1$.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'N. Okabe'
- 'K. Umetsu'
title: Observational Constraints on the ICM Temperature Enhancement by Cluster Mergers
---
Abstract {#abstract .unnumbered}
========
We present results from a combined weak lensing and X-ray analysis of the merging cluster A1914 at a redshift of $z=0.1712$ based on $R_{\rm}$-band imaging data with Subaru/Suprime-Cam and archival Chandra X-ray data. Based on the weak-lensing and X-ray data we explore the relationships between cluster global properties, namely the gravitational mass, the bolometric X-lay luminosity and temperature, the gas mass and the gas mass fraction, as a function of radius. We found that the gas mass fractions within $r_{2500}$ and $r_{{\rm vir}}$ are consistent with the results of earlier X-ray cluster studies and of cosmic microwave background studies based on the WMAP observations, respectively. However, the observed temperature, $k_B T_{\rm ave}=9.6\pm0.3 {\rm
keV}$, is significantly higher than the virial temperature, $k_B T_{\rm vir}=4.7\pm0.3 {\rm keV}$ derived from the weak lensing distortion measurement. The X-ray bolometric luminosity and temperature ($L_X-T$) relation is consistent with the $L_X-T$ relation derived by previous statistical X-ray studies of galaxy clusters. Such correlations among the global cluster properties are invaluable observational tools for studying the cluster merger physics. Our results demonstrate that the combination of X-ray and weak-lensing observations is a promising, powerful probe of the physical processes associated with cluster mergers as well as of their mass properties.
Weak Lensing and X-ray Analyses {#sec:1}
===============================
### Weak Lensing Analysis {#weak-lensing-analysis .unnumbered}
Weak lensing provides a direct measure of the projected mass distribution in the universe regardless of the physical/dynamical state of matter in the system. Therefore weak lensing enables the direct study of mass in clusters even when the clusters are in the process of (pre/mid/post) merging, where the assumptions of hydrostatic equilibrium or isothermality are no longer valid. We carried out a weak lensing analysis on the merging cluster A1914 with deep $R_{\rm c}$-band data taken with Suprime-Cam on the Subaru telescope, covering the entire cluster region out to the cluster virial radius thanks to the wide field-of-view of $34'\times 27'$. The details of the analysis will be presented in Okabe & Umetsu [@ou07] and Umetsu & Okabe[@uo07]. We define a sample of background galaxies with magnitudes $ 21 {\lower.5ex\hbox{$\; \buildrel < \over \sim \;$}}R_{\rm c} {\lower.5ex\hbox{$\; \buildrel < \over \sim \;$}}26$ and $r_h$ and half-light radii $r_h^* {\lower.5ex\hbox{$\; \buildrel < \over \sim \;$}}r_h {\lower.5ex\hbox{$\; \buildrel < \over \sim \;$}}15 {\rm pixels}$, yielding a mean galaxy number density of $n_g\simeq 48 {\rm arcmin}^{-2}$. We derived a radial profile of the tangential component of reduced gravitational shear, $g_+=\gamma_+/(1-\kappa)$, over the radial range of $3'$ to $17'$, where the irregularity in the mass distribution is less significant than that in the central region. The best-fitting NFW (Navarro, Frenk, & White[@nfw97]) profile to the Subaru distortion data is obtained as follows: virial mass $M_{{\rm vir}}=(7.66\pm0.73)\times 10^{14}
M_{\odot} h_{70}^{-1}$ (or viral radius $r_{\rm vir}=12'.26=2.144 h_{70}^{-1}{\rm Mpc}$); concentration parameter $c=5.07\pm1.75$. The best-fitting mass profile from weak lensing distortion measurements is in good agreement with the luminosity profile of cluster member galaxies, which will be presented in Umetsu & Okabe [@uo07].
![[]{data-label="fig:fgas"}](okabeF1.ps){height="3.5cm"}
![[]{data-label="fig:fgas"}](okabeF2.ps){height="3.5cm"}
### X-ray Analysis {#x-ray-analysis .unnumbered}
We used archival [*Chandra*]{} data to measure physical properties of the ICM in A1914. We performed a spectral fit with a single temperature model within a radius of $5'$. The average temperature and abundance were obtained as $k_B T_{{\rm ave}}(\theta<5')=9.6\pm0.3 {\rm keV}$ $A=0.19\pm0.08$ at a $90\%$ confidence level, respectively. The radial profile fit was performed on the observed X-ray surface brightness distribution ($\theta<6'$) using a single $\beta$ model, where we adopted the same center as for the weak lensing tangential shear measurement. The best fitting parameters were obtained as follows: cluster core radius, $r_c=1.'03\pm0.'06$; slope parameter, $\beta=0.727\pm0.014$; central electron density, $n_{e,0}=(1.46\pm0.26)\times10^{-2} h_{70}^{-2} {\rm cm}^{-3}$.
Global Parameters of Merging Clusters {#sec:1}
=====================================
### Gas Mass Fraction {#gas-mass-fraction .unnumbered}
We show in Figure \[fig:fgas\] the radial profile of the gas mass fraction, $f_{\rm gas}(<r)=M_{\rm gas}(<r)/M_{\rm tot}(<r)$. We used the best-fitting models derived in §1 to calculate $f_{\rm gas}(<r)$. Using the $\beta$ model we extrapolated the gas mass profile outside $\theta=6'$ where the weak lensing mass profile is available. The values within typical radii of $r_{2500}$, $r_{500}$ and $r_{{\rm vir}}$, at which the mean density is $2500$, $500$ and $\delta_{\rm vir}$ times the critical density of the universe, are $f_{2500}=0.108$, $f_{500}=0.133$ and $f_{{\rm vir}}=0.165$, respectively. The central gas mass fraction, $f_{2500}$, is similar to those values for relaxed clusters derived by X-ray data alone, such as $0.091\pm0.002$ (Vikhlinin et al. [@vik06]) and $0.117\pm0.002$ (Allen et al. [@all04]). The values of $f_{500}$ deduced from X-ray studies show a large scatter among different clutters and different observations. We note the values of $f_{500}$ deduced from X-ray studies show a large scatter among different clutters and different observations. We found the virial gas fraction, $f_{\rm vir}$, of A1914 is consistent with the cosmic mean baryon fraction $\Omega_{b}/\Omega_{m}$, constrained by the CMB observations (Spergel et al. [@spe03]). The virial gas fraction, $f_{\rm vir}$, is consistent with the cosmic mean baryon fraction $\Omega_{b}/\Omega_{m}$, constrained by the CMB observations (Spergel et al. [@spe03]). We emphasize that a combined weak-lensing and X-ray study will allow us to determine gas mass fractions even in merging clusters.
### $L_X-T$ relation {#l_x-t-relation .unnumbered}
We derived the X-ray bolometric luminosity $L_{X,{\rm bol}}$ and temperature $T$ with a single-temperature model. The resulting X-ray luminosity within $r_{200}$ is $L_{X,{\rm bol}}=2.5 \times10^{45} h_{70}^{-2}{\rm ergs~s}^{-1}$ with $k_B T_{\rm vir}=4.7^{+0.3}_{-0.3}\left(M/M_{\rm
vir}\right)^{2/3} {\rm keV}$, which is consistent with the observed local $L_X-T$ relation from previous X-ray studies: $L_{\rm X,bol}(<r_{200})=2.3^{+0.3}_{-0.3}\times10^{45}\left(k_B
T/9.6{\rm keV}\right)^{2.88\pm0.15} h_{70}^{-2}$ ${\rm ergs~s}^{-1}$ (Arnaud & Evrard [@arn99]). This indicates that the observed local $L_X-T$ relation holds even in the merging cluster A1914.
### $M-T$ relation {#m-t-relation .unnumbered}
The observed temperature $k_B T=9.6\pm0.3 {\rm keV}$ is significantly higher than the viral temperature, $k_B T_{\rm vir}=4.7^{+0.3}_{-0.3}\left(M/M_{\rm
vir}\right)^{2/3} {\rm keV}$ derived from the weak lensing analysis. If the cluster were virialized before the merging process, then the ICM temperature could be heated up by a factor of two. Based on this scenario, we constrain the heating energy of the ICM induced by the cluster merger as $$\begin{aligned}
\langle \Delta E_{{\rm ICM}}(r<5') \rangle \rangle &=& 4 \pi (k_B T_{{\rm
ave}}- k_B T_{{\rm vir}}) \int \sum_j n_j(r) r^2 dr \sim 2 \times 10^{62} {\rm erg}, \nonumber\end{aligned}$$
with $n_H=0.82n_e$, where we assumed the electron and ion temperatures are the same.
| {
"pile_set_name": "ArXiv"
} |
[T. P. Shestakova]{}[Department of Theoretical and Computational Physics, Southern Federal University,\
Sorge St. 5, Rostov-on-Don 344090, Russia\
E-mail: [[email protected]]{}]{}
Introduction
============
In [@Shest1] we present the outline of Hamiltonian dynamics in extended phase space that can be constructed to be completely equivalent to a Lagrangian formulation for a system with constraints. As a rule, Hamiltonian formulation for gravity is constructed following to the Dirac scheme [@Dirac1; @Dirac2] and is a starting point for most attempts to quantize gravity. However, there are some reasons to doubt that Dirac Hamiltonian formulation for gravitational theory [@Dirac3] can be thought as an equivalent one to the original General Relativity. The reasons are closely connected with the role that given to gauge gravitational degrees of freedom $g_{0\mu}$, $\mu=0, 1, 2, 3$, fixing a reference frame, in these two formulations.
In Einstein (Lagrangian) formulation of General Relativity $g_{0\mu}$ components of metric tensor are treated on an equal basis with the rest of components, $g_{ij}$, $i, j=1, 2, 3$, defining 3-space geometry. The theory is invariant under gauge transformations, an infinitesimal form of which is $$\label{g_transf}
\delta g_{\mu\nu}
=\eta^{\lambda}\partial_{\lambda}g_{\mu\nu}
+g_{\mu\lambda}\partial_{\nu}\eta^{\lambda}
+g_{\nu\lambda}\partial_{\mu}\eta^{\lambda}$$ and which are true for all components of metric tensor ($\eta^{\mu}$ are infinitesimal parameters). In the Dirac approach only $g_{ij}$ components with their conjugate momenta are included into phase space, the transformations for this variables are generated by constraints. The constraints, which do not depend on gauge variables, play a crucial role in quantization procedure, while gauge variables are dropped out of further consideration.
It is well-known that the Wheeler – DeWitt quantum geometrodynamics [@DeWitt], constructed by a direct application of the Dirac quantization scheme to gravity, has faced quite a number of problems (see, for example, [@Isham; @SS1]). At the same time, modern approaches to quantization of gravity, such as Loop Quantum Gravity, keep considering the Wheeler – DeWitt equation as a main constituent of the theory [@Rovelli].
It is worth mentioning that the algebra of constraints in original phase space not including gauge degrees of freedom does depend on parametrization of gravitational variables. As was argued in our previous paper [@Shest1], it creates a serious obstacle to find an algorithm to construct a generator that would give correct transformations for all gravitational variables in the limits of the Dirac approach. Moreover, any transformations of gravitational variables, which touch upon gauge degrees of freedom, are not canonical from the viewpoint of the Dirac formalism. Even a transition from metric tensor components to the Arnowitt – Deser – Misner (ADM) variables [@ADM], quite legitime in the Lagrangian formulation, should be considered as non-canonical [@KK]. In this case the theory turns out to be essentially dependent on a chosen parametrization.
In the framework of the canonical scheme it is not obvious that the original (Lagrangian) formulation of General Relativity and its Hamiltonian dynamics à la Dirac are theories with different groups of transformations. It has become more clear in the Batalin – Fradkin – Vilkovisky (BFV) approach [@BFV1; @BFV2; @BFV3], that aims at reproducing the Dirac results on the path integral level. A Hamiltonian form of the effective action in this approach is determined by algebra of constraints. The algebra of constraints in the case of gravity is open and, as was emphasized in [@BFV1], gauge transformations differ from transformations generated by gravitational constraints. It may lead to a new type of additional Feynman diagrams corresponding to four-ghosts interaction which cannot result from the effective action in the Lagrangian form. The new type of diagram does not play an important role while one is interested in only gauge invariant sector in the $S$-matrix theory, for which the BFV approach was originally proposed. However, we can think of it as a considerable mathematical indication that Lagrangian and Hamiltonian formalism appear to be non-equivalent for the full theory of gravity when one deals with spacetime manifolds of any topology, in particular, without asymptotic states which are implied in the $S$-matrix approach.
The central part in the BFV approach is given to the BRST charge constructed as a series in powers of Grassmannian variables with coefficients given by generalized structure functions of constraints algebra [@Hennaux]. Like the constraints, the BRST charge does not generates correct gauge transformations for all gravitational degrees of freedom including gauge ones. It is not surprising because the form of the BRST charge is determined by constraints algebra and, as was already mentioned, gauge transformations differ from those generated by the constraints.
The purpose of this paper is to present Hamiltonian dynamics in extended phase space which is free from the shortcomings mentioned above and can be thought of as a real alternative for Dirac generalized Hamiltonian dynamics, as well for the BFV formalism. The proposed approach has already been demonstrated for gravitational models with finite degrees of freedom ([@SSV1; @SSV2; @Shest1] and other papers). In this work we apply our approach to generalized spherically symmetric gravitational model which imitates the full gravitational theory much better, so that one can see the way how one can get appropriate results in the case of the full theory. In Section 2 the Lagrangian and Hamiltonian dynamics in extended phase space are derived from the effective action for the model, and the structure of the Hamiltonian function and Hamiltonian equations is analyzed. In Section 3 we shall consider some particular cases that the generalized model embraces, and show that the solutions to the obtained equations comprise solutions to the gauge invariant Einstein equations with corresponding symmetry. We also discuss a possible role of gauge-noninvariant terms. In Section 4 we shall make use of BRST invariance of effective action and construct the BRST charge according to the Noether theorem which generates correct transformations for all gravitational degrees of freedom. The proposed method will be shown to be self-consistent, and the equivalence of the Lagrangian and Hamiltonian formulations can be proved by direct calculations.
The model, its Lagrangian and Hamiltonian dynamics
==================================================
In this paper we shall follow to the ADM parametrization for space-time metric: $$\label{gen.int}
ds^2=\left(-N^2+N_i N^i\right)dt^2+2N_i dt dx^i+g_{ij}dx^i dx^j.$$ Under the condition of spherical symmetry the metric is reduced to $$\begin{aligned}
ds^2&=&\left[-N^2(t,r)+(N^r(t,r))^2V^2(t,r)\right]dt^2+2N^r(t,r)V^2(t,r)dt dr\nonumber\\
\label{mod.int}
&+&V^2(t,r)dr^2+W^2(t,r)\left(d\theta^2+\sin^2\theta d\varphi^2\right).\end{aligned}$$ where $N^r=N^1$ is the only component of the shift vector. In this model we have two gauge variables $N$ and $N^r$ which are fixed by two gauge conditions $$\begin{aligned}
\label{g.c.1}
N&=&f(V,W);\\
\label{g.c.2}
N^r&=&f^r(V,W).\end{aligned}$$ $f(V,W)$, $f^r(V,W)$ are arbitrary functions. In differential form the gauge conditions will introduce missing velocities into the effective Lagrangian, so ensuring [*an actual extension of phase space*]{}: $$\begin{aligned}
\label{d.c.1}
\dot N&=&\frac{\partial f}{\partial V}\dot V+\frac{\partial f}{\partial W}\dot W;\\
\label{d.c.2}
\dot N^r&=&\frac{\partial f^r}{\partial V}\dot V+\frac{\partial f^r}{\partial W}\dot W.\end{aligned}$$
We shall consider the Faddeev – Popov effective action including gauge and ghost sectors as it appears in the path integral approach to gauge field theories, $$\label{full-act1}
S_{(eff)}=S_{(grav)}+S_{(gauge)}+S_{(ghost)}.$$
The gravitational part of the effective action $$\label{grav-act1}
S_{(grav)}=\int d^4x\sqrt{-g}R$$ is invariant under gauge transformations (\[g\_transf\]), but the gravitational Lagrangian involves second derivatives of metric components. To get field equations it is much easier to make use of the Lagrangian which is quadratic in first derivatives of metric components and can be obtained from the original one by omitting total derivatives. However, we shall have to return to the original gravitational Lagrangian when deriving the BRST charge in accordance with the Noether theorem (see Section 4).
The gauge-fixing part of the action is $$\label{gfix-act1}
S_{(gauge)}
=\int dt\int\limits_0^{\infty}dr\left[\lambda_0\left(\dot N-\frac{\partial f}{\partial V}\dot V
-\frac{\partial f}{\partial W}\dot W\right)
+\lambda_r\left(\dot N^r-\frac{\partial f^r}{\partial V}\dot V
-\frac{\partial f^r}{\partial W}\dot W\right)\right].$$
Taking into account gauge transformations for gravitational variables $$\begin{aligned}
\label{delN}
\delta N&=&-\dot N\eta^0-N'\eta^r-N\dot\eta^0+N N^r(\eta^0)';\\
\label{delNr}
\delta N^r&=&-\dot N^r\eta^0-(N^r)'\eta^r-N^r\dot\eta^0-\dot\eta^r
+N^r(\eta^r)'+\frac{N^2}{V^2}(\eta^0)'+(N^r)^2(\eta^0)';\\
\label{delV}
\delta V&=&-\dot V\eta^0-V'\eta^r-V(\eta^r)'-V N^r(\eta^0)';\\
\label{delW}
\delta W&=&-\dot W\eta^0-W'\eta^r,\end{aligned}$$ that follow from (\[g\_transf\]), we get the Faddeev – Popov ghost action in the form $$\begin{aligned}
\label{ghost-act1}
S_{(ghost)}&=&\int dt\int\limits_0^{\infty}dr
\left[\bar\theta_0\frac{d}{dt}\left(-\dot N\theta^0-N'\theta^r-N\dot\theta^0
+N N^r(\theta^0)'\right.\right.\nonumber\\
&-&\left.\frac{\partial f}{\partial V}\left[-\dot V\theta^0-V'\theta^r-V(\theta^r)'-V N^r(\theta^0)'\right]
-\frac{\partial f}{\partial W}\left[-\dot W\theta^0-W'\theta^r\right]\right)\nonumber\\
&+&\bar\theta_r\frac{d}{dt}\left(-\dot N^r\theta^0-(N^r)'\theta^r-N^r\dot\theta^0-\dot\theta^r
+N^r(\theta^r)'+\frac{N^2}{V^2}(\theta^0)'+(N^r)^2(\theta^0)'\right.\nonumber\\
&-&\left.\left.\frac{\partial f^r}{\partial V}\left[-\dot V\theta^0-V'\theta^r-V(\theta^r)'-V N^r(\theta^0)'\right]
-\frac{\partial f^r}{\partial W}\left[-\dot W\theta^0-W'\theta^r\right]\right)\right]\end{aligned}$$ $\bar\theta_0$, $\theta^0$, $\bar\theta_r$, $\theta^r$ are ghost variables. After redefinition $$\label{lambda}
\pi_N=\lambda_0+\dot{\bar\theta_0}\theta^0;\quad
\pi_{N^r}=\lambda_r+\dot{\bar\theta_r}\theta^0$$ we can write the effective Lagrangian in the form without second derivatives: $$\begin{aligned}
\label{full-act2}
S_{(eff)}&=&\int dt\int\limits_0^{\infty}dr\left(\frac{\dot V\dot W W}N
+\frac{V\dot W^2}{2N}
-\frac{N'W'W}V
-\frac{N(W')^2}{2V}
-\frac{NV}2\right.\nonumber\\
&-&\frac{W'\dot WVN^r}N
-\frac{W W'\dot V N^r}N
-\frac{W\dot W V'N^r}N
-\frac{W\dot W V(N^r)'}N\nonumber\\
&+&\frac{W W'V'(N^r)^2}N
+ \frac{W W'V N^r(N^r)'}N
+\frac{(W')^2 V(N^r)^2}{2N}\nonumber\\
&+&\pi_N\left(\dot N-\frac{\partial f}{\partial V}\dot V-\frac{\partial f}{\partial W}\dot W\right)
+\pi_{N^r}\left(\dot N^r-\frac{\partial f^r}{\partial V}\dot V
-\frac{\partial f^r}{\partial W}\dot W\right)\nonumber\\
&+&\dot{\bar\theta_0}\theta^r\left(N'-\frac{\partial f}{\partial V}V'
-\frac{\partial f}{\partial W}W'\right)\nonumber\\
&+&\dot{\bar\theta_0}\left(N\dot\theta^0-NN^r(\theta^0)'-\frac{\partial f}{\partial V}VN^r(\theta^0)'
-\frac{\partial f}{\partial V}V(\theta^r)'\right)\nonumber\\
&+&\dot{\bar\theta_r}\left[N^r\dot\theta^0-\left(\frac{N^2}{V^2}+(N^r)^2\right)(\theta^0)'+\dot\theta^r
-N^r(\theta^r)'+(N^r)'\theta^r\right.\nonumber\\
&-&\left.\left.\frac{\partial f^r}{\partial V}\left(VN^r(\theta^0)'+V(\theta^r)'+V'\theta^r\right)
-\frac{\partial f^r}{\partial W}W'\theta^r\right]\right)\end{aligned}$$
Variation of the effective action with respect to $N$, $N^r$, $V$, $W$ yields the Einstein equations for the model with additional terms resulting from the gauge-fixing and ghost parts of the action: $$\label{g.Ein.eqs}
R_{\mu}^{\nu}-\frac12\delta_{\mu}^{\nu}R=
\kappa\left(T_{\mu(mat)}^{\nu}+T_{\mu(obs)}^{\nu}+T_{\mu(ghost)}^{\nu}\right),$$ where $T_{\mu(mat)}^{\nu}$ stands for the energy-momentum tensor of matter fields which can be included into the model, $T_{\mu(obs)}^{\nu}$ and $T_{\mu(ghost)}^{\nu}$ are obtained by varying the gauge-fixing and ghost action, respectively, so that $T_{\mu(obs)}^{\nu}$ describes the observer in a reference frame. It is clear that $T_{\mu(obs)}^{\nu}$ and $T_{\mu(ghost)}^{\nu}$ are not true tensors since they depend on chosen gauge conditions. Equations (\[g.Ein.eqs\]) can be called [*the gauged Einstein equations*]{}. By adding ghost equations and gauge conditions (\[d.c.1\]), (\[d.c.2\]) to the gauged Einstein equations, we obtain [*the extended set of Lagrangian equations*]{} for our model which is presented in Appendix A. The explicit expressions of $T_{\mu(obs)}^{\nu}$, $T_{\mu(ghost)}^{\nu}$ are given in Appendix C. Below $\kappa=8\pi$ (in special units when $G=c=\hbar=1$).
Now we can find the momenta conjugate to all gravitational and ghost variables: $$\begin{aligned}
\label{PN}
P_N&=&\pi_N;\\
\label{PNr}
P_{N^r}&=&\pi_{N^r};\\
\label{PV}
P_V&=&\frac{W\dot W}N-\frac{W'W N^r}N-\pi_N\frac{\partial f}{\partial V}
-\pi_{N^r}\frac{\partial f^r}{\partial V};\\
\label{PW}
P_W&=&\frac{W\dot V}N+\frac{V\dot W}N-\frac{W'V N^r}N-\frac{W V'N^r}N-\frac{V W(N^r)'}N
-\pi_N\frac{\partial f}{\partial W}-\pi_{N^r}\frac{\partial f^r}{\partial W};\\
\label{Pbt0}
P_{\bar\theta_0}&=&N'\theta^r-\frac{\partial f}{\partial V}V'\theta^r
-\frac{\partial f}{\partial W}W'\theta^r
+N\dot\theta^0-N N^r(\theta^0)'-\frac{\partial f}{\partial V}V N^r(\theta^0)'
-\frac{\partial f}{\partial V}V(\theta^r)';\\
\label{Pt0}
\bar P_{\theta^0}&=&\dot{\bar\theta_0}N+\dot{\bar\theta_r}N^r;\\
P_{\bar\theta_r}&=&N^r\dot\theta^0-\frac{N^2}{V^2}(\theta^0)'-(N^r)^2(\theta^0)'
+\dot\theta^r-N^r(\theta^r)'+(N^r)'\theta^r\nonumber\\
\label{Pbtr}
&-&\frac{\partial f^r}{\partial V}V N^r(\theta^0)'
-\frac{\partial f^r}{\partial V}V(\theta^r)'
-\frac{\partial f^r}{\partial V}V'\theta^r
-\frac{\partial f^r}{\partial W}W'\theta^r;\\
\label{Ptr}
\bar P_{\theta^r}&=&\dot{\bar\theta_r}.\end{aligned}$$
Introducing of the missing velocities by means of the differential form of gauge conditions (\[d.c.1\]), (\[d.c.2\]) enables us to construct a Hamiltonian in extended phase space not applying to the Dirac procedure, by the usual rule $$\label{Ham1}
H=\int\limits_0^{\infty}dr\left(P_N\dot N+P_{N^r}\dot N^r+P_V\dot V+P_W\dot W
+\bar P_{\theta^0}\dot\theta^0+\dot{\bar\theta_0}P_{\bar\theta_0}
+\bar P_{\theta^r}\dot\theta^r+\dot{\bar\theta_r}P_{\bar\theta_r}-L\right);$$ $$\begin{aligned}
H&=&\int\limits_0^{\infty}dr\left[\frac N W P_V P_W
-\frac{N V}{2W^2}P_V^2+P_V V'N^r+P_V V(N^r)'+P_W W'N^r
+\frac{N'W'W}V+\frac{N(W')^2}{2V}+\frac{N V}2\right.\nonumber\\
&+&P_N\frac{\partial f}{\partial V}V'N^r
+P_N\frac{\partial f}{\partial W}W'N^r
+P_N\frac{\partial f}{\partial V}V(N^r)'
+P_{N^r}\frac{\partial f^r}{\partial V}V'N^r
+P_{N^r}\frac{\partial f^r}{\partial W}W'N^r\nonumber\\
&+&P_{N^r}\frac{\partial f^r}{\partial V}V(N^r)'
+\frac N W P_V P_N\frac{\partial f}{\partial W}
+\frac N W P_W P_N\frac{\partial f}{\partial V}
-\frac{N V}{W^2}P_V P_N\frac{\partial f}{\partial V}\nonumber\\
&+&\frac N W P_V P_{N^r}\frac{\partial f^r}{\partial W}
+\frac N W P_W P_{N^r}\frac{\partial f^r}{\partial V}
-\frac{N V}{W^2}P_V P_{N^r}\frac{\partial f^r}{\partial V}
-\frac{N V}{2W^2} P_N^2\left(\frac{\partial f}{\partial V}\right)^2\nonumber\\
&-&\frac{N V}{2W^2}P_{N^r}^2\left(\frac{\partial f^r}{\partial V}\right)^2
+\frac N W P_N^2\frac{\partial f}{\partial V}
\frac{\partial f}{\partial W}
+\frac N W P_{N^r}^2\frac{\partial f^r}{\partial V}
\frac{\partial f^r}{\partial W}\nonumber\\
&+&\frac N W P_N P_{N^r}\frac{\partial f}{\partial V}
\frac{\partial f^r}{\partial W}
+\frac N W P_N P_{N^r}\frac{\partial f}{\partial W}
\frac{\partial f^r}{\partial V}
-\frac{N V}{W^2}P_N P_{N^r}\frac{\partial f}{\partial V}
\frac{\partial f^r}{\partial V}\nonumber\\
&+&\frac1N\bar P_{\theta^0}P_{\bar\theta_0}
+\bar P_{\theta^r}P_{\bar\theta_r}
-\frac{N^r}N\bar P_{\theta^r}P_{\bar\theta_0}
-\frac{N'}N\bar P_{\theta^0}\theta^r
+N^r\bar P_{\theta^0}(\theta^0)'
+N^r\bar P_{\theta^r}(\theta^r)'\nonumber\\
&-&(N^r)'\bar P_{\theta^r}\theta^r
+\frac{N'N^r}N\bar P_{\theta^r}\theta^r
+\frac{N^2}{V^2}\bar P_{\theta^r}(\theta^0)'
+\frac{\partial f^r}{\partial V}V'\bar P_{\theta^r}\theta^r
+\frac{\partial f^r}{\partial W}W'\bar P_{\theta^r}\theta^r\nonumber\\
&+&\frac{\partial f^r}{\partial V}V N^r\bar P_{\theta^r}(\theta^0)'
+\frac{\partial f^r}{\partial V}V\bar P_{\theta^r}(\theta^r)'
+\frac V N\frac{\partial f}{\partial V}\bar P_{\theta^0}(\theta^r)'
+\frac{V N^r}N\frac{\partial f}{\partial V}\bar P_{\theta^0}(\theta^0)'\nonumber\\
&-&\frac{V N^r}N\frac{\partial f}{\partial V}\bar P_{\theta^r}(\theta^r)'
-\frac{V(N^r)^2}N\frac{\partial f}{\partial V}\bar P_{\theta^r}(\theta^0)'
+\frac{V'}N\frac{\partial f}{\partial V}\bar P_{\theta^0}\theta^r\nonumber\\
\label{Ham-EPS}
&-&\left.\frac{V'N^r}N\frac{\partial f}{\partial V}\bar P_{\theta^r}\theta^r
+\frac{W'}N\frac{\partial f}{\partial W}\bar P_{\theta^0}\theta^r
-\frac{W'N^r}N\frac{\partial f}{\partial W}\bar P_{\theta^r}\theta^r\right].\end{aligned}$$
The first line in (\[Ham-EPS\]) is the Hamiltonian for pure gravity that can be presented as a linear combination of Dirac secondary constraints since it is believed that a full derivative with respect to $r$ can be omitted in this expression: $$\label{Ham-D}
H_D=\int\limits_0^{\infty}dr\left[N\left(\frac1W P_V P_W-\frac V{2W^2}P_V^2
-\frac{W W''}V-\frac{(W')^2}{2V}+\frac{V'W W'}{V^2}+\frac V2\right)
+N^r\left(P_W W'-P'_V V\right)\right].$$ However, as it follows from (\[Ham-EPS\]), the Hamiltonian in extended phase space cannot be written down as a linear combination of constraints. Now we can write down the set of Hamiltonian equations in extended phase space presented explicitly in Appendix B.
It is important to emphasized that in this formulation of Hamiltonian dynamics the constraints as well as the gauge conditions have the status of Hamiltonian equations. Indeed, the Hamiltonian equations (\[DN\]), (\[DNr\]) coincide with the gauge conditions (\[d.c.11\]), (\[d.c.22\]), while the equations (\[DPN\]), (\[DPNr\]) reproduce the constraints (\[eLN\]), (\[eLNr\]) in the Lagrangian formalism. The equations (\[DV\]) – (\[DPW\]) for physical gravitational degrees of freedom after some rearrangement can be shown to be equivalent to the dynamical Lagrangian equations (\[eLV\]), (\[eLW\]), and Eqs. (\[Dt0\]) – (\[DbPtr\]) are equivalent to the ghost equations (\[eLbt0\]) – (\[eLtr\]). Thus, in this Section we have got two sets of extended equations for our spherically symmetric model in the Lagrangian and Hamiltonian formalisms, which are proved to be completely equivalent. In the next Section we shall consider some particular solutions to these equations.
Particular solutions to the extended set of equations
=====================================================
Solutions to gauge invariant Einstein equations.
------------------------------------------------
It is obvious enough that if we put all ghost as well as the Lagrange multipliers $\pi_N$, $\pi_{N^r}$ equal to zero in the extended set of equations (\[eLN\]) – (\[eLtr\]), we would return to gauge invariant Einstein equations for the model. The ghost equations (\[eLbt0\]) – (\[eLtr\]) are satisfied identically, so in this case we have just four equations (\[eLN\]) – (\[eLW\]) and two gauge conditions (\[d.c.11\]), (\[d.c.22\]). For further simplification of the equations we choose the condition (\[g.c.2\]) in the form $$\label{g.c.2.1}
N^r=0.$$
Taking into account (\[g.c.2.1\]) we come to the following set of equations: $$\begin{aligned}
\label{eLN1}
0&=&\frac{\dot V W\dot W}{N^2}+\frac{V\dot W^2}{2N^2}
+\frac V2-\frac{(W')^2}{2V}-\frac{W W''}V+\frac{V'W'W}{V^2};\\
\label{eLNr1}
0&=&\frac{W W'\dot V}N-\frac{W V\dot W'}N+\frac{N'WV\dot W }{N^2};\\
\label{eLV1}
0&=&\frac{\dot W^2}{2N}+\frac{W\ddot W}N-\frac{W\dot W\dot N}{N^2}+\frac N2
-\frac{N(W')^2}{2V^2}-\frac{N'W'W}{V^2};\\
\label{eLW1}
0&=&\frac{W\ddot V}N+\frac{\dot V\dot W}N+\frac{V\ddot W}N-\frac{\dot N\dot V W}{N^2}
-\frac{\dot N\dot W V}{N^2}-\frac{W N''}V+\frac{W N'V'}{V^2}
-\frac{W'N'}V+\frac{W'N V'}{V^2}-\frac{W''N}V;\\
\label{g.c.1.1}
0&=&N-f(V,W).\end{aligned}$$
Solutions to Eqs. (\[eLN1\]) – (\[eLW1\]) can be considered as solutions of gauge invariant Einstein equations for the spherically symmetric model, though any solution requires fixing of gauge conditions to be presented in its final form. Eqs. (\[eLN1\]), (\[eLV1\]), (\[eLW1\]) can be obtained from a Lagrangian for the model with the metric (\[mod.int\]) where $N^r(t, r)=0$. Let us note that Eq. (\[eLNr1\]) which is equivalent to the $\left(0\atop 1\right)$ Einstein equation could not be obtained if one just substitutes (\[g.c.2.1\]) to the gravitational part of the action. The second gauge condition (\[g.c.1.1\]) can be imposed on different stages of solving the equations, but in what following we shall start from fixing this condition.
The Hamiltonian set of equations will be also rather simplified: $$\begin{aligned}
\label{DN1}
\dot N&=&\frac N W P_V\frac{\partial f}{\partial W}
+\frac N W P_W\frac{\partial f}{\partial V}
-\frac{N V}{W^2}P_V\frac{\partial f}{\partial V};\\
\label{DPN1}
0&=&-\frac1W P_V P_W+\frac V{2W^2}P_V^2+\frac{W''W}V-\frac{V'W'W}{V^2}
+\frac{(W')^2}{2V}-\frac V2;\\
\label{DNr1}
\dot N^r&=&0;\\
\label{DPNr1}
0&=&(P_V)'V-P_W W';\\
\label{DV1}
\dot V&=&\frac N W P_W-\frac{N V}{W^2}P_V;\\
\label{DPV1}
\dot P_V&=&\frac N{2W^2}P_V^2+\frac{N'W'W}{V^2}
+\frac{N(W')^2}{2V^2}-\frac N2;\\
\label{DW1}
\dot W&=&\frac N W P_V;\\
\label{DPW1}
\dot P_W&=&\frac N{W^2}P_V P_W-\frac{N V}{W^3}P_V^2
+\frac{N''W}V-\frac{N'V'W}{V^2}+\frac{N'W'}V+\frac{N W''}V-\frac{N V'W'}{V^2}.\end{aligned}$$
### The Schwarzschild solution.
The Schwarzschild solution [@Schw] is probably the most famous spherically symmetric solution to the Einstein equations. First of all, we choose the gauge condition (\[g.c.1.1\]) in the form $$\label{g.c.1.2}
N=\frac1V.$$
Following to Schwarzschild [@Schw], we shall seek for a solution that does not depend on time. Then, Eq. (\[eLNr1\]) is identically satisfied, and Eqs. (\[eLN1\]), (\[eLV1\]) under the condition (\[g.c.1.2\]) look like $$\begin{aligned}
\label{eLN2}
0&=&\frac V2-\frac{(W')^2}{2V}-\frac{W W''}V+\frac{V'W'W}{V^2};\\
\label{eLV2}
0&=&\frac V2-\frac{(W')^2}{2V}+\frac{V'W'W}{V^2}.\end{aligned}$$ From (\[eLN2\]), (\[eLV2\]) we get $$\label{Schw.W}
W''=0;\qquad
W(r)=C_1r+C_2.$$ One can put $C_1=1$, $C_2=0$, so that the length of a circle with a center in coordinate origin would be $2\pi r$. Then $W(r)=r$. Eq. (\[eLV2\]) takes the form $$\label{eLV2_1}
0=\frac V2-\frac1{2V}+\frac{V'r}{V^2}.$$ Its solution can be written as $$\label{Schw.sol}
V(r)=\frac1{\sqrt{1-\displaystyle\frac{2GM}r}},$$ where $M$ is associated with a Schwarzschild mass. Eq. (\[eLW1\]) is identically satisfied. It is easy to see that the same solution follows from Eqs. (\[DN1\]) – (\[DPW1\]). In our consideration we have started from fixing the gauge condition which helps to separate out the required solution, meanwhile usually a gauge is imposed on a final stage of the procedure. So we get the Schwarzschild metric: $$\label{Schw.met}
ds^2=-\left(1-\frac{2GM}r\right)dt^2+\frac{dr^2}{1-\displaystyle\frac{2GM}r}
+r^2\left(d\theta^2+\sin^2\theta d\varphi^2\right).$$
### The Lemaitre –- Tolman – Bondi metric.
The Lemaitre –- Tolman – Bondi solution is the solution for dust matter in synchronous and comoving reference frame (see, for example, [@Enqvist] and references therein). Actually one should add energy-momentum tensor of dust into the equations. In this case the only non-zero component of the matter energy-momentum tensor is $T_{0(mat)}^0=\varepsilon$, where $\varepsilon$ is the energy density of dust. After imposing the gauge condition $$\label{g.c.1.3}
N=1$$ Eq. (\[eLN1\]) can be rewritten as $$\label{eLN3}
\frac{2\dot V\dot W}{V W}+\frac{\dot W^2}{W^2}+\frac1{W^2}
-\frac{(W')^2}{V^2W^2}-\frac{2W''}{V^2W}+\frac{2V'W'}{V^3W}
=8\pi\varepsilon.$$ Eq. (\[eLNr1\]) gives $$\label{eLNr3}
0=W'\dot V-V\dot W'.$$ Its solution yields a relation between $V(t,r)$ and $W(t,r)$, $$\label{LTB.sol}
V(t,r)=C_1(r)W'(t,r),$$ where $C_1(r)$ is an arbitrary function. The solution (\[LTB.sol\]) can be obtained from Hamiltonian equations. From (\[DW1\]), (\[DV1\]) one can express $$\label{LTB.mom}
P_V=W\dot W,\qquad
P_W=W\dot V+V\dot W.$$ Substitution (\[LTB.mom\]) into (\[DPNr1\]) gives (\[LTB.sol\]). Finally, the Lemaitre –- Tolman – Bondi metric can be presented as $$\label{LTB.met}
ds^2=-dt^2+\left(C_1(r)W'(t,r)\right)^2 dr^2
+W^2(t,r)\left(d\theta^2+\sin^2\theta d\varphi^2\right).$$
Eq. (\[eLV1\]) after substitution (\[LTB.sol\]) leads to $$\label{eLV3}
\dot W^2=\frac{C_2(r)}W+\frac1{C_1^2(r)}-1,$$ $C_2(r)$ is another arbitrary function. The substitution of (\[LTB.sol\]) into (\[eLN3\]) together with (\[eLV3\]) gives the solution for dust energy density $$\label{dust.sol}
\varepsilon=\frac1{8\pi}\frac{C'_2(r)}{W'W^2}.$$
### The Friedmann closed isotropic model.
We can obtain the Einstein equations for the Friedmann closed isotropic model as a particular case of Eqs. (\[eLN1\]) – (\[g.c.1.1\]). We then should put $$\label{isotr.nod}
V(t,r)=a(t),\qquad
W(t,r)=a(t)\sin r.$$ Eq. (\[eLN1\]) gives the constraint $$\label{eLN4}
\frac{a\dot a^2}{N^2}+a=0.$$ Eq. (\[eLNr1\]) is identically satisfied. Eqs. (\[eLV1\]), (\[eLW1\]) are both reduced to the dynamical equation $$\label{eLa4}
\frac{a\ddot a}N+\frac{\dot a^2}{2N}-\frac{a\dot a\dot N}{N^2}+\frac N2=0.$$ Eqs. (\[eLN4\]), (\[eLa4\]) should be complemented by a gauge condition $N=f(a)$, or $$\label{d.c.1.4}
\dot N=\frac{df}{da}\dot a.$$
The same equations can be obtained from the Hamiltonian set (\[DN1\]) – (\[DPW1\]). From (\[DW1\]), (\[DV1\]) we get $$\label{FRW.mom}
P_V=\frac{W\dot W}N=\frac{a\dot a}N \sin^2r,\qquad
P_W=\frac{W\dot V}N+\frac{V\dot W}N=\frac{2a\dot a}N\sin r.$$ Eq. (\[DN1\]) can be rewritten in the form (\[d.c.1.4\]) keeping in mind that $$\label{ch.var}
\frac{df}{da}=\frac{\partial f}{\partial V}+\frac{\partial f}{\partial W}\sin r.$$ Eq. (\[DPN1\]) leads to the constraint (\[eLN4\]). Eq. (\[DPNr1\]) is identically satisfied. Eqs. (\[DPV1\]), (\[DPW1\]) can be both reduced to the equation (\[eLa4\]).
A possible role of gauge-noninvariant terms.
--------------------------------------------
In this section we shall discuss the role of Lagrange multipliers and ghosts in the extended set of Lagrangian equations, and consider a restricted class of solutions when $\theta^0$, $\bar\theta_0$ are functions of $t$ only while $\theta^r$, $\bar\theta_r$ are functions of $r$. We shall show that even for this restricted class their role can be very different: For example, to obtain the Schwarzschild metric one should choose trivial solution for Lagrange multipliers and ghosts, otherwise the metric components will not have the Galilean form at infinity. On the contrary, in the case of the Lemaitre –- Tolman – Bondi model gauge-noninvariant sector of the model (gauge-fixing and ghost components) can imitate dust matter with zero pressure and energy density given by (\[dust.sol\]). In the end of this section we shall refer to our earlier papers where the role of gauge-noninvariant sector has been analyzed for systems with finite number degrees of freedom.
As it was done in Section 3.1, for simplicity we choose one of the gauge conditions in the form $N^r=0$ (\[g.c.2.1\]). Under this condition Eq. (\[eLbtr\]) reads $$\label{eLbtr_1}
\ddot\theta^r-\frac d{dt}\left(\frac{N^2}{V^2}(\theta^0)'\right)=0.$$ This equation is satisfied if $$\label{tsol}
\theta^0=\theta^0(t),\qquad
\theta^r=\theta^r(r).$$ Indeed, the condition (\[g.c.2.1\]) imposes limitations on possible coordinates transformations, so that the admissible transformations of $t$, $r$ are $\tau=\tau(t)$, $\rho=\rho(r)$. The infinitesimal parameters $\eta^0$, $\eta^r$ in (\[delN\]) – (\[delW\]) must be functions of $t$, $r$, correspondingly: $\eta^0=\eta^0(t)$, $\eta^r=\eta^r(r)$. Since ghost variables are known to compensate residual gauge transformations, it is reasonable to consider ghost solutions of the form (\[tsol\]).
Keeping in mind (\[g.c.2.1\]) and (\[tsol\]), one can rewrite Eq. (\[eLbt0\]) as $$\label{eLbt0_1}
\frac d{dt}\left[N\dot\theta^0-\frac{\partial f}{\partial V}V(\theta^r)'
+\left(N'-\frac{\partial f}{\partial V}V'-\frac{\partial f}{\partial W}W'\right)\theta^r\right]=0.$$ As we remember, Eq. (\[d.c.11\]) is the differential form of Eq. (\[g.c.1\]). As a consequence of the latter we get $$\label{d.c.0}
N'=\frac{\partial f}{\partial V}V'+\frac{\partial f}{\partial W}W',$$ and Eq. (\[eLbt0\_1\]) looks like $$\label{eLbt0_2}
\frac d{dt}\left[N\dot\theta^0-\frac{\partial f}{\partial V}V(\theta^r)'\right]=0.$$
Under the condition (\[g.c.2.1\]) equations for ghosts $\bar\theta_0$, $\bar\theta_r$ (\[eLt0\]), (\[eLtr\]) will be reduced to $$\label{eLt0_1}
\frac d{dt}\left(\dot{\bar\theta_0}N\right)
-\frac d{dr}\left(\dot{\bar\theta_r}\frac{N^2}{V^2}\right)=0$$ and $$\label{eLtr_1}
\frac d{dr}\left(\dot{\bar\theta_0}\frac{\partial f}{\partial V}V\right)-\ddot{\bar\theta_r}=0.$$ Again, let us consider the solutions such that $$\label{btsol}
\bar\theta^0=\bar\theta^0(t),\qquad
\bar\theta^r=\bar\theta^r(r).$$ Thus, the two equations for $\bar\theta_0$, $\bar\theta_r$ are $$\label{eLt0r_2}
\frac d{dt}\left(\dot{\bar\theta_0}N\right)=0,\qquad
\frac d{dr}\left(\dot{\bar\theta_0}\frac{\partial f}{\partial V}V\right)=0.$$
One can notice that $\pi_{N^r}$ enters into Eqs. (\[eLV\]), (\[eLW\]) being multiplied by the derivatives of the function $f^r$ which is equal to zero; $\pi_{N^r}$ also enters into Eq. (\[eLNr\]). The ghost terms are missed out from last equation if (\[tsol\]), (\[btsol\]) are true. One can put $$\label{piNr.sol}
\pi_{N^r}=0.$$ In this case Eq. (\[eLNr\]) will be reduced to the constraint (\[eLNr1\]).
Applying the Bianchi identity to the both sides of (\[g.Ein.eqs\]), we obtain $$\label{BI}
\left(T_{\mu(mat)}^{\nu}+T_{\mu(obs)}^{\nu}+T_{\mu(ghost)}^{\nu}\right)_{;\nu}=0.$$ We can require that the conditions $T_{\mu ;\nu}^{\nu}=0$ would be satisfied separately by the energy-momentum tensor of matter fields and quasi-tensors of the observer and of ghosts. It will give us equations for the Lagrange multiplier $\pi_N$. Taking into account that $N^r=0$, $\pi_{N^r}=0$, the non-zero components of $T_{\mu(obs)}^{\nu}$ are $$\begin{aligned}
\label{obs00_1}
T^0_{0(obs)}&=&\frac1{4\pi}\dot\pi_N\frac1{VW^2};\\
\label{obs11_1}
T^1_{1(obs)}&=&-\frac1{4\pi}\dot\pi_N\frac1{NW^2}\frac{\partial f}{\partial V};\\
\label{obs22_1}
T^2_{2(obs)}=T^3_{3(obs)}&=&-\frac1{8\pi}\dot\pi_N\frac1{N V W}\frac{\partial f}{\partial W}.\end{aligned}$$ Equations $T_{\mu(obs);\nu}^{\nu}=0$ with $\mu=0, 1$ give $$\label{Tobs}
\frac d{dt}\left(\dot\pi_N N\right)=0,\qquad
\frac d{dr}\left(\dot\pi_N\frac{\partial f}{\partial V}V\right)=0.$$
The non-zero components of $T_{\mu(ghost)}^{\nu}$ are $$\begin{aligned}
\label{ghost00_1}
T^0_{0(ghost)}&=&\frac1{4\pi}\frac1{V W^2}\left[\dot{\bar\theta_0}(\theta^r)'
-\dot{\bar\theta_0}\dot{\theta^0}\right];\\
\label{ghost11_1}
T^1_{1(ghost)}&=&\frac1{4\pi}\frac V{N W^2}\frac{\partial^2 f}{\partial V^2}\dot{\bar\theta_0}(\theta^r)';\\
\label{ghost22_1}
T^2_{2(ghost)}=T^3_{3(ghost)}&=&\frac1{8\pi}
\left[\frac1{N W}\frac{\partial^2 f}{\partial V\partial W}\dot{\bar\theta_0}(\theta^r)'
-\frac1{N V W}\frac{\partial f}{\partial W}\dot{\bar\theta_0}(\theta^r)'\right].\end{aligned}$$ As one can check, the equations $T_{\mu(ghost);\nu}^{\nu}=0$ are completely compatible with Lagrangian equations for ghosts.
### The Schwarzschild metric.
In this case the consideration of equations for ghost and the Lagrange multiplier $\pi_N$ does not lead to non-zero solutions for this variables, at least under requirements (\[tsol\]), (\[btsol\]), (\[piNr.sol\]). Indeed, Eqs. (\[eLt0r\_2\]) result in $\dot{\bar\theta_0}=0$, and all components of $T_{\mu(ghost)}^{\nu}$ are zero, thus the ghost sector gives no contribution. It follows from Eqs. (\[Tobs\]) that $$\label{pi.Schw}
\frac{\dot\pi_N}V={\rm const}=\frac{C_3}2.$$ and non-zero components of $T_{\mu(obs)}^{\nu}$ are $$\label{obs00_2}
T^0_{0(obs)}=T^1_{1(obs)}=\frac1{8\pi}\frac{C_3}{W^2}.$$ Then instead of (\[Schw.sol\]) we get $$\label{Schw.sol.1}
V(r)=\frac1{\sqrt{1+C_3-\displaystyle\frac{2GM}r}}.$$ The requirement for metric components to have the Galilean form at infinity means that $C_3=0$ and all components of $T_{\mu(obs)}^{\nu}$ are zero.
### The Lemaitre –- Tolman – Bondi solution.
This solution is rather interesting. From (\[eLbt0\_2\]), (\[eLt0r\_2\]) and (\[Tobs\]) we obtain $$\label{pi-theta}
\dot{\bar\theta_0}={\rm const};\qquad
\dot{\theta_0}={\rm const};\qquad
\dot\pi_N=\varphi(r),$$ and, after the substitution of (\[LTB.sol\]), for non-zero components of $T_{\mu(obs)}^{\nu}$, $T_{\mu(ghost)}^{\nu}$ we have $$\label{obs00_3}
T^0_{0(obs)}=\frac1{4\pi}\frac{\varphi_1(r)}{W'W^2};\qquad
T^0_{0(ghost)}=\frac1{4\pi}\frac{\varphi_2(r)}{W'W^2},$$ $\varphi(r)$, $\varphi_1(r)$, $\varphi_2(r)$ are arbitrary functions. We can see that the both quasi-tensors $T_{\mu(ghost)}^{\nu}$, $T_{\mu(obs)}^{\nu}$ describe a medium with zero pressure and energy density of the form (\[dust.sol\]). The meaningful feature of this consideration is that we do not need to introduce an extra dust matter into the model, while a dust matter is simulated by its ghost and gauge-fixing components.
### Models with finite numbers of degrees of freedom.
In our early papers ([@SSV2] and others) it was demonstrated for models with finite numbers of degrees of freedom that the quasi-tensor $T_{\mu(obs)}^{\nu}$ describes a continual medium with the equation of state essentially depending on the parametrization of the gauge variable and the gauge condition for it. It is easy to see that a change of the gauge variable (the choice of its parametrization), which looks like $N=v(\tilde N, a)$ for the isotropic model, together with an additional condition for a new variable $\tilde N$, say, $\tilde N=1$, fixes the gauge condition for $N$: $f(a)=v(1, a)$. It has been shown in [@Shest2] that the choice of parametrization function $v(\tilde N, a)=\tilde N a^{n-3}$ (which is equivalent to the gauge condition $N=a^{n-3}$) results in the equation of state $p=\left(\displaystyle\frac n3-1\right)\varepsilon$, where $n$ can have various values, for example, $n=0$ corresponds to de Sitter false vacuum, $n=4$ corresponds a radiation dominated universe, etc. It confirms that at least in some cases one can simulate a medium with required properties only by means of introducing a suitable parametrization and gauge conditions. Another component of the integrated system, which includes pure gravity, matter fields and the observer, is ghost fields; analyzing its quasi-tensor $T_{\mu(ghost)}^{\nu}$ for a given gauge one can obtain its equation of state. However, the role of ghost fields has not been studied enough.
The BRST charge
===============
It is known that the Faddeev – Popov effective action possesses a residual global symmetry, the so-called BRST symmetry. In the Lagrangian formalism the BRST transformations for our model are given by (\[delN\]) – (\[delW\]), where infinitesimal parameters $\eta^{\mu}$ should be replaced by $\bar\varepsilon\theta^{\mu}$, $\bar\varepsilon$ is a Grassmannian parameter, $$\begin{aligned}
\label{delN1}
\delta N&=&\bar\varepsilon\left[-\dot N\theta^0-N'\theta^r-N\dot\theta^0+N N^r(\theta^0)'\right];\\
\label{delNr1}
\delta N^r&=&\bar\varepsilon\left[-\dot N^r\theta^0-(N^r)'\theta^r-N^r\dot\theta^0-\dot\theta^r
+N^r(\theta^r)'+\frac{N^2}{V^2}(\theta^0)'+(N^r)^2(\theta^0)'\right];\\
\label{delV1}
\delta V&=&\bar\varepsilon\left[-\dot V\theta^0-V'\theta^r-V(\theta^r)'-V N^r(\theta^0)'\right];\\
\label{delW1}
\delta W&=&\bar\varepsilon\left[-\dot W\theta^0-W'\theta^r\right].\end{aligned}$$ Moreover, $$\begin{aligned}
\label{delt0}
\delta\theta^0&=&\bar\varepsilon\left[\dot\theta^0\theta^0+(\theta^0)'\theta^r\right];\\
\label{deltr}
\delta\theta^r&=&\bar\varepsilon\left[\dot\theta^r\theta^0+(\theta^r)'\theta^r\right];\\
\label{delbt0}
\delta\bar\theta_0&=&-\bar\varepsilon\lambda_0;\\
\label{delbtr}
\delta\bar\theta_r&=&-\bar\varepsilon\lambda_r;\\
\label{dell0}
\delta\lambda_0&=&0;\\
\label{dellr}
\delta\lambda_r&=&0.\end{aligned}$$ The transformations (\[delN1\]) – (\[dellr\]) should be supplemented by coordinated transformations $$\label{coord.tr}
\delta t=\bar\varepsilon\theta^0;\qquad
\delta r=\bar\varepsilon\theta^r.$$ As a consequence of a global symmetry there exists a BRST charge which plays a role of a generator of BRST transformations in extended phase space. As we have already mentioned, in the BFV approach it is constructed as a series in powers of Grassmannian variables with coefficients given by generalized structure functions of constraints algebra: $$\label{gen_BFV}
\Omega=c^{\alpha}U^{(0)}_{\alpha}
+c^{\beta}c^{\gamma}U^{(1)\alpha}_{\gamma\beta}\bar\rho_{\alpha}+\ldots$$ $c^{\alpha}$, $\bar\rho_{\alpha}$ are BFV ghosts, $U^{(n)}$ are $n$th order structure functions, while zero order structure functions $U^{(0)}_{\alpha}$ are Dirac secondary constraints.
Since the BFV prescription of constructing the BRST charge is essentially rely upon constraints algebra, it cannot produce correct transformations for gauge gravitational variables, like a linear combination of constraints cannot produce them in the Dirac approach. In [@Shest1] we have analyzed the algorithm suggested in [@Cast] that aims at modifying the Dirac scheme and constructing a generator producing correct transformations for all variables. However, this algorithm fails to be applied to an arbitrary parametrization of gravitational variables, so it is not general enough and cannot be considered as a required solution to the problem.
At the same time, the existence of global BRST symmetry enables us to propose another method based upon the Noether theorem and the equivalence of Lagrangian dynamics and Hamilton dynamics in extended phase space. In this section we shall apply it to our spherically symmetric model, however nothing prevent one from applying it to any other gravitational model including the full theory of gravity. The fact that gauge degrees of freedom are treated on the equal footing with other variables allows one to make transformations of variables including gauge ones which have been proved to be canonical in extended phase space [@Shest1] and do not affect the algebra of Poisson brackets. So, the proposed method will work for any reasonable parametrization of gravitational variables.
You can find the proof of BRST symmetry of the Faddeev – Popov effective action for Yang – Mills fields in any book on quantum field theory [@Weinberg]. In the case of gravity we deal with space-time symmetry, and we should take into account explicit dependence of the Lagrangian and the measure on space-time coordinates. One can check that the sum of gauge-fixing and ghost parts of the action (\[gfix-act1\]), (\[ghost-act1\]) is not invariant under transformations (\[delN1\]) – (\[coord.tr\]). In some works the BRST invariance is guaranteed by asymptotical boundary conditions for ghosts and Lagrange multipliers [@Hennaux; @Hall]. The legitimacy of asymptotic boundary conditions is questionable in the case of space-time of arbitrary topology. Therefore, we seek for a BRST invariant form of the action without appealing to any additional conditions. One can check that to ensure its BRST invariance we have to add to the action the following term containing only full derivatives and not affecting the set of equations obtained in Section 2: $$\begin{aligned}
S_{(add)}&=&\int dt\int\limits_0^{\infty}dr
\left(\frac d{dt}\left[\bar\theta_0\left(\dot N-\frac{\partial f}{\partial V}\dot V
-\frac{\partial f}{\partial W}\dot W\right)\theta^0\right]
+\frac d{dr}\left[\bar\theta_0\left(\dot N-\frac{\partial f}{\partial V}\dot V
-\frac{\partial f}{\partial W}\dot W\right)\theta^r\right]\right.\nonumber\\
\label{add-act}
&+&\left.\frac d{dt}\left[\bar\theta_r\left(\dot N^r-\frac{\partial f^r}{\partial V}\dot V
-\frac{\partial f^r}{\partial W}\dot W\right)\theta^0\right]
+\frac d{dr}\left[\bar\theta_r\left(\dot N^r-\frac{\partial f^r}{\partial V}\dot V
-\frac{\partial f^r}{\partial W}\dot W\right)\theta^r\right]\right).\end{aligned}$$
As was mentioned in Section 2, the gravitational part of the action in (\[full-act2\]) is not invariant under gauge transformation and, therefore under BRST transformations (\[delN1\]) – (\[delW1\]). Then, we should return to the gravitational action (\[grav-act1\]). Now we deal with the Lagrangian which involves second derivatives of metric components and ghosts. The BRST charge is constructed in accordance with the Noether theorem generalized for theories with high derivatives: $$\label{Noet.BRST}
\Omega=\int d^3x\left[\frac{\partial L}{\partial(\partial_0\phi^a)}\delta\phi^a
+\frac{\partial L}{\partial(\partial _0\partial_{\mu}\phi^a)}\delta(\partial_{\mu}\phi^a)
-\partial_{\mu}\left(\frac{\partial L}{\partial(\partial_0\partial_{\mu}\phi^a)}\right)\delta\phi^a
+\partial _0\left(L x^0\right)\right].$$ $\phi^a$ stands for all variables $N$, $N^r$, $V$, $W$ and ghosts. After some tedious calculations we come to the following expression for the BRST charge in the spherically symmetric model: $$\begin{aligned}
\Omega&=&\int\!dr\left[-{\cal H}\theta^0-P_V V'\theta^r
-P_N\frac{\partial f}{\partial V}V'\theta^r
-P_{N^r}\frac{\partial f^r}{\partial V}V'\theta^r\right.\nonumber\\
&-&P_W W'\theta^r-P_N\frac{\partial f}{\partial W}W'\theta^r
-P_{N^r}\frac{\partial f^r}{\partial W}W'\theta^r\nonumber\\
&-&P_V V N^r(\theta^0)'-P_N\frac{\partial f}{\partial V} V N^r(\theta^0)'
-P_{N^r}\frac{\partial f^r}{\partial V} V N^r(\theta^0)'\nonumber\\
&-&P_V V(\theta^r)'-P_N\frac{\partial f}{\partial V} V(\theta^r)'
-P_{N^r}\frac{\partial f^r}{\partial V} V(\theta^r)'\nonumber\\
\label{BRST1}
&-&\left.\bar P_{\theta^0}(\theta^0)'\theta^r
-\bar P_{\theta^r}(\theta^r)'\theta^r
-P_N P_{\bar\theta_0}-P_{N^r}P_{\bar\theta_r}
-\frac{N W W'(\theta^0)'}V\right],\end{aligned}$$ $\cal H$ is a Hamiltonian density in (\[Ham-EPS\]). It can be directly checked that the charge (\[BRST1\]) generates transformations (\[delN1\]) – (\[delbtr\]). Let us emphasized that the Hamiltonian equations in extended phase space, in particular, constraints and gauge conditions which have the status of Hamiltonian equations, is used to get correct results, for instance, $$\begin{aligned}
\delta N&=&\left\{N,\;\bar\varepsilon\Omega\right\}
=\bar\varepsilon\frac{\delta\Omega}{\delta P_N}\nonumber\\
&=&\bar\varepsilon\left[-\frac{\partial{\cal H}}{\partial P_N}\theta^0
-\frac{\partial f}{\partial V}V'\theta^r
-\frac{\partial f}{\partial W}W'\theta^r
-\frac{\partial f}{\partial V} V N^r(\theta^0)'
-\frac{\partial f}{\partial V} V(\theta^r)'-P_{\bar\theta_0}\right]\nonumber\\
\label{delN2}
&=&\bar\varepsilon\left[-\dot N\theta^0-N'\theta^r-N\dot\theta^0+N N^r(\theta^0)'\right].\end{aligned}$$ Here we used one of the Hamiltonian equations $\dot N=\displaystyle\frac{\delta H}{\delta P_N}$ (\[DN\]), and the expression for $P_{\bar\theta_0}$ (\[Pbt0\]). To check (\[dell0\]), (\[dellr\]) one should firstly find $\delta P_N$, $\delta P_{N^r}$. $$\begin{aligned}
\delta P_N&=&\left\{P_N,\;\bar\varepsilon\Omega\right\}
=-\bar\varepsilon\frac{\delta\Omega}{\delta N}
=\bar\varepsilon\left[-\frac{\partial\Omega}{\partial N}
+\left(\frac{\partial\Omega}{\partial N'}\right)'\right]\nonumber\\
&=&\bar\varepsilon\left[\frac{\partial{\cal H}}{\partial N}\theta^0
-\left(\frac{\partial{\cal H}}{\partial N'}\theta^0\right)'
+\frac{W W'(\theta^0)'}V\right]\nonumber\\
&=&\bar\varepsilon\left[\left(\frac{\partial{\cal H}}{\partial N}
-\left(\frac{\partial{\cal H}}{\partial N'}\right)'\right)\theta^0
-\frac{\partial{\cal H}}{\partial N'}(\theta^0)'
+\frac{W W'(\theta^0)'}V\right]\nonumber\\
&=&\bar\varepsilon\left[-\dot P_N\theta^0-\left(\frac{W W'}V
-\frac1N\bar P_{\theta^0}\theta^r
+\frac{N^r}N\bar P_{\theta^r}\theta^r\right)(\theta^0)'
+\frac{W W'(\theta^0)'}V\right]\nonumber\\
\label{delPN}
&=&\bar\varepsilon\left[-\dot P_N\theta^0
+\frac1N\left(N\dot{\bar\theta_0}+N^r\dot{\bar\theta_r}\right)\theta^r(\theta^0)'
-\frac{N^r}N\dot{\bar\theta_r}\theta^r(\theta^0)'\right]
=\bar\varepsilon\left[-\dot P_N\theta^0+\dot{\bar\theta_0}\theta^r(\theta^0)'\right];\\
\label{delPNr}
\delta P_{N^r}&=&\bar\varepsilon\left[-\dot P_{N^r}\theta^0+\dot{\bar\theta_r}\theta^r(\theta^0)'\right].\end{aligned}$$ Here we also used the Hamiltonian equation $\dot P_N=-\displaystyle\frac{\delta H}{\delta N}
=-\displaystyle\frac{\partial{\cal H}}{\partial N}+\left(\displaystyle\frac{\partial{\cal H}}{\partial N'}\right)'$ (\[DPN\]). Keeping in mind the relations between $P_N$, $P_{N^r}$ and $\lambda_0$, $\lambda_r$ (\[lambda\]), one can be convinced that Eqs. (\[dell0\]), (\[dellr\]) are correct.
One can find transformations for $\delta P_V$: $$\begin{aligned}
\delta P_V&=&\bar\varepsilon\left[-\dot P_V\theta^0-P'_V\theta^r
-P'_N\frac{\partial f}{\partial V}\theta^r
-P'_{N^r}\frac{\partial f^r}{\partial V}\theta^r
+P_N\frac{\partial^2 f}{\partial V^2}V(\theta^r)'
+P_{N^r}\frac{\partial^2 f^r}{\partial V^2}V(\theta^r)'\right.\nonumber\\
&+&P_N\frac{\partial^2 f}{\partial V^2}V N^r(\theta^0)'
+P_{N^r}\frac{\partial^2 f^r}{\partial V^2}V N^r(\theta^0)'
+\frac{\partial f^r}{\partial V}\bar P_{\theta^r}(\theta^0)'\theta^r\nonumber\\
\label{delPV}
&+&\left.\frac1N\frac{\partial f}{\partial V}\bar P_{\theta^0}(\theta^0)'\theta^r
-\frac{N^r}N\frac{\partial f}{\partial V}\bar P_{\theta^r}(\theta^0)'\theta^r
-\frac{N W W'}{V^2}(\theta^0)'\right].\end{aligned}$$ The transformation of (\[PV\]) gives the same result. Similarly one can obtain transformations in extended phase space for $P_W$ and ghosts momenta. They are in correspondence with (\[PW\]) – (\[Ptr\]).
Concluding remarks
==================
In the present paper we have constructed a self-consistent Hamiltonian dynamics for the generalized spherically symmetric model in extended phase space. Our starting point was the Faddeev – Popov effective action with gauge-fixing and ghost terms. Thanks to introducing the missing velocities into the Lagrangian by gauge conditions of special form we do not need to invent some prescription how to construct a Hamiltonian function. Hamiltonian equations are proved to be equivalent to the Lagrangian set of equations. The group of transformations in extended phase space includes the group of gauge transformations for all gravitational degrees of freedom. We also have a clear algorithm how to construct a generator of transformations in extended phase space in accordance with the Noether theorem. The necessary condition for the algorithm to work is to find a BRST invariant form of the action. For the present model we have found the additional terms (\[add-act\]) that guarantees the required BRST invariance. The form of these terms gives us a hint what a BRST invariant form of the action would be in the full gravitational theory. Let us emphasize once again that we do not impose any additional conditions to ensure BRST invariance.
In our opinion, the proposed approach to construct Hamiltonian dynamics for gravity (and, in general, to any constrained system) is of interest by itself, as an alternative to the Dirac approach. On the other hand, it can be considered as a preliminary step to subsequent quantization of the model, and it will be a goal of our further research.
The extended set of Lagrangian equations
========================================
Variation of the effective action (\[full-act2\]) with respect to $N$, $N^r$ gives the constraints in the Lagrangian formalism which are equivalent to $\left(0\atop 0\right)$ and $\left(0\atop 1\right)$ Einstein equations: $$\frac{\partial L}{\partial N}=\partial_0\frac{\partial L}{\partial\dot N}+\partial_r\frac{\partial L}{\partial N'};$$ $$\begin{aligned}
0&=&\frac{\dot V W\dot W}{N^2}+\frac{V\dot W^2}{2N^2}
+\frac V2-\frac{(W')^2}{2V}-\frac{W W''}V+\frac{V'W'W}{V^2}\nonumber\\
&-&\frac{W'\dot W V N^r}{N^2}- \frac{W\dot W V(N^r)'}{N^2}
-\frac{WW'\dot V N^r}{N^2}-\frac{W\dot W V'N^r}{N^2}\nonumber\\
&+&\frac{(W')^2V(N^r)^2}{2N^2}+\frac{W W'V'(N^r)^2}{N^2}+\frac{W W'V N^r(N^r)'}{N^2}\nonumber\\
\label{eLN}
&+&\dot{\pi}_N-\dot{\bar\theta_0}\dot{\theta}^0
+(\dot{\bar\theta_0})'\theta^r+\dot{\bar\theta_0}(\theta^r)'
+\dot{\bar\theta_0}N^r(\theta^0)'+\dot{\bar\theta_r}\frac{2N}{V^2}(\theta^0)';\end{aligned}$$ $$\frac{\partial L}{\partial N^r}
=\partial_0\frac{\partial L}{\partial\dot{N^r}}+\partial_r\frac{\partial L}{\partial(N^r)'};$$ $$\begin{aligned}
0&=&\frac{W W'N'V N^r}{N^2}-\frac{W W'\dot V}N-\frac{W W''V N^r}N+\frac{W V\dot W'}N
-\frac{N'WV\dot W }{N^2}+\frac{W W'V'N^r}N\nonumber\\
\label{eLNr}
&-&\dot{\pi}_{N^r}+\dot{\bar\theta_r}\dot{\theta^0}-(\dot{\bar\theta_r})'\theta^r
-2\dot{\bar\theta_r}(\theta^r)'-2\dot{\bar\theta_r}N^r(\theta^0)'
-\dot{\bar\theta_0}\frac{\partial f}{\partial V}V(\theta^0)'
-\dot{\bar\theta_r}\frac{\partial f^r}{\partial V}V(\theta^0)'
-\dot{\bar\theta_0}N(\theta^0)'.\end{aligned}$$ Variations with respect to $V$, $W$ leads to the equations which are equivalent to dynamical $\left(1\atop 1\right)$ and $\left(2\atop 2\right)$ Einstein equations: $$\frac{\partial L}{\partial V}
=\partial_0\frac{\partial L}{\partial\dot V}+\partial_r\frac{\partial L}{\partial V'};$$ $$\begin{aligned}
0&=&\frac{\dot W^2}{2N}+\frac{W\ddot W}N-\frac{W\dot W\dot N}{N^2}+\frac N2
-\frac{N(W')^2}{2V^2}-\frac{N'W'W}{V^2}\nonumber\\
&-&\frac{N'W W'(N^r)^2}{N^2}+\frac{W W''(N^r)^2}N-\frac{W'\dot W N^r}N
+\frac{WW'N^r\dot N}{N^2}+\frac{N'N^r W\dot W}{N^2}\nonumber\\
&-&\frac{2W\dot W'N^r}N-\frac{WW'\dot N^r}N+\frac{(W')^2(N^r)^2}{2N}+\frac{WW'N^r(N^r)'}N\nonumber\\
&-&\dot\pi_N\frac{\partial f}{\partial V}-\dot\pi_{N^r}\frac{\partial f^r}{\partial V}
-(\dot{\bar\theta_0})'\frac{\partial f}{\partial V}\theta^r
+\dot{\bar\theta_0}\frac{\partial^2 f}{\partial V^2}V N^r(\theta^0)'\nonumber\\
&+&\dot{\bar\theta_0}\frac{\partial f}{\partial V}N^r(\theta^0)'
+\dot{\bar\theta_0}\frac{\partial^2 f}{\partial V^2}V(\theta^r)'
-\dot{\bar\theta_r}\frac{2N^2}{V^3}(\theta^0)'\nonumber\\
\label{eLV}
&+&\dot{\bar\theta_r}\frac{\partial^2 f^r}{\partial V^2}V N^r(\theta^0)'
+\dot{\bar\theta_r}\frac{\partial f^r}{\partial V}N^r(\theta^0)'
+\dot{\bar\theta_r}\frac{\partial^2 f^r}{\partial V^2}V(\theta^r)'
-(\dot{\bar\theta_r})'\frac{\partial f^r}{\partial V}\theta^r;\end{aligned}$$ $$\frac{\partial L}{\partial W}
=\partial_0\frac{\partial L}{\partial\dot W}+\partial_r\frac{\partial L}{\partial W'};$$ $$\begin{aligned}
0&=&\frac{W\ddot V}N+\frac{\dot V\dot W}N+\frac{V\ddot W}N-\frac{\dot N\dot V W}{N^2}
-\frac{\dot N\dot W V}{N^2}-\frac{W N''}V+\frac{W N'V'}{V^2}
-\frac{W'N'}V+\frac{W'N V'}{V^2}-\frac{W''N}V\nonumber\\
&-&\frac{2V\dot W'N^r}N-\frac{W'\dot V N^r}N-\frac{W'V\dot N^r}N+\frac{W'V N^r\dot N}{N^2}
+\frac{\dot W V N^r N'}{N^2}-\frac{\dot W V'N^r}N-\frac{2W\dot V'N^r}N\nonumber\\
&-&\frac{WV'\dot N^r}N+\frac{WV'N^r\dot N}{N^2}-\frac{2W\dot V(N^r)'}N+\frac{W\dot V N^r N'}{N^2}
-\frac{\dot W V(N^r)'}N-\frac{W V(\dot N^r)'}N+\frac{WV\dot N(N^r)'}{N^2}\nonumber\\
&+&\frac{W V''(N^r)^2}N+\frac{3W V'N^r(N^r)'}N-\frac{W V'(N^r)^2 N'}{N^2}
+\frac{W V((N^r)')^2}N+\frac{W V N^r(N^r)''}N\nonumber\\
&-&\frac{W V N'N^r(N^r)'}{N^2}+\frac{W''V(N^r)^2}N
+\frac{W'V'(N^r)^2}N+\frac{2W'V N^r(N^r)'}N-\frac{W'V(N^r)^2 N'}{N^2}\nonumber\\
&-&\dot\pi_N\frac{\partial f}{\partial W}-\dot\pi_{N^r}\frac{\partial f^r}{\partial W}
-(\dot{\bar\theta_0})'\frac{\partial f}{\partial W}\theta^r
-\dot{\bar\theta_0}\frac{\partial f}{\partial W}(\theta^r)'
+\dot{\bar\theta_0}\frac{\partial^2 f}{\partial V\partial W}V N^r(\theta^0)'
+\dot{\bar\theta_0}\frac{\partial^2 f}{\partial V\partial W}V(\theta^r)'\nonumber\\
\label{eLW}
&+&\dot{\bar\theta_r}\frac{\partial^2 f^r}{\partial V\partial W}V N^r(\theta^0)'
+\dot{\bar\theta_r}\frac{\partial^2 f^r}{\partial V\partial W}V(\theta^r)'
-(\dot{\bar\theta_r})'\frac{\partial f^r}{\partial W}\theta^r
-\dot{\bar\theta_r}\frac{\partial f^r}{\partial W}(\theta^r)'.\end{aligned}$$
We also have four equations for two pairs of ghosts: $$\frac{\partial L}{\partial\bar\theta_0}
=\partial_0\frac{\partial L}{\partial\dot{\bar\theta_0}}
+\partial_r\frac{\partial L}{\partial(\bar\theta_0)'};$$ $$\begin{aligned}
0&=&N'\dot\theta^r-\frac{\partial f}{\partial V}V'\dot\theta^r
-\frac{\partial f}{\partial W}W'\dot\theta^r+\dot N'\theta^r
-\frac{\partial^2 f}{\partial V^2}\dot V V'\theta^r\nonumber\\
&-&\frac{\partial^2 f}{\partial V\partial W}V'\dot W\theta^r
-\frac{\partial f}{\partial V}\dot V'\theta^r
-\frac{\partial^2 f}{\partial V\partial W}\dot V W'\theta^r
-\frac{\partial^2 f}{\partial W^2}\dot W W'\theta^r
-\frac{\partial f}{\partial W}\dot W'\theta^r\nonumber\\
&+&\dot N\dot\theta^0+N\ddot\theta^0-\dot N N^r(\theta^0)'-N\dot N^r(\theta^0)'-NN^r(\dot\theta^0)'
-\frac{\partial^2 f}{\partial V^2}\dot V V N^r(\theta^0)'\nonumber\\
&-&\frac{\partial^2 f}{\partial V\partial W}\dot W V N^r(\theta^0)'
-\frac{\partial f}{\partial V}\dot V N^r(\theta^0)'
-\frac{\partial f}{\partial V}V\dot N^r(\theta^0)'
-\frac{\partial f}{\partial V}V N^r(\dot\theta^0)'\nonumber\\
\label{eLbt0}
&-&\frac{\partial^2 f}{\partial V^2}\dot V V(\theta^r)'
-\frac{\partial^2 f}{\partial V\partial W}\dot W V(\theta^r)'
-\frac{\partial f}{\partial V}\dot V(\theta^r)'
-\frac{\partial f}{\partial V}V(\dot\theta^r)';\end{aligned}$$ $$\frac{\partial L}{\partial\theta^0}
=\partial_0\frac{\partial L}{\partial\dot\theta^0}+\partial_r\frac{\partial L}{\partial(\theta^0)'};$$ $$\begin{aligned}
0&=&\ddot{\bar\theta_0}N+\dot{\bar\theta_0}\dot N
-(\dot{\bar\theta_0})'N N^r
-\dot{\bar\theta_0}N'N^r-\dot{\bar\theta_0}N(N^r)'\nonumber\\
&-&(\dot{\bar\theta_0})'\frac{\partial f}{\partial V}V N^r
-\dot{\bar\theta_0}\frac{\partial^2 f}{\partial V^2}V'V N^r
-\dot{\bar\theta_0}\frac{\partial^2 f}{\partial V\partial W}W'V N^r
-\dot{\bar\theta_0}\frac{\partial f}{\partial V}V'N^r
-\dot{\bar\theta_0}\frac{\partial f}{\partial V}V(N^r)'\nonumber\\
&+&\ddot{\bar\theta_r}N^r+\dot{\bar\theta_r}\dot N^r
-(\dot{\bar\theta_r})'\left(\frac{N^2}{V^2}+(N^r)^2\right)
-2\dot{\bar\theta_r}\;\frac{N N'}{V^2}
+2\dot{\bar\theta_r}\;\frac{N^2V'}{V^3}
-2\dot{\bar\theta_r}N^r(N^r)'\nonumber\\
\label{eLt0}
&-&(\dot{\bar\theta_r})'\frac{\partial f^r}{\partial V}V N^r
-\dot{\bar\theta_r}\frac{\partial^2 f^r}{\partial V^2}V'V N^r
-\dot{\bar\theta_r}\frac{\partial^2 f^r}{\partial V\partial W}W'V N^r
-\dot{\bar\theta_r}\frac{\partial f^r}{\partial V}V'N^r
-\dot{\bar\theta_r}\frac{\partial f^r}{\partial V}V(N^r)';\end{aligned}$$ $$\frac{\partial L}{\partial\bar\theta_r}
=\partial_0\frac{\partial L}{\partial\dot{\bar\theta_r}}
+\partial_r\frac{\partial L}{\partial(\bar\theta_r)'};$$ $$\begin{aligned}
0&=&\dot N^r\dot\theta^0+N^r\ddot\theta^0-2\frac{N\dot N}{V^2}(\theta^0)'
+2\frac{N^2\dot V}{V^3}(\theta^0)'-\frac{N^2}{V^2}(\dot\theta^0)'
-2N^r\dot N^r(\theta^0)'\nonumber\\
&-&(N^r)^2(\dot\theta^0)'+\ddot\theta^r-\dot N^r(\theta^r)'-N^r(\dot\theta^r)'
+(\dot N^r)'\theta^r+(N^r)'\dot\theta^r\nonumber\\
&-&\frac{\partial^2 f^r}{\partial V^2}\dot V V N^r(\theta^0)'
-\frac{\partial^2 f^r}{\partial V^2}\dot V V(\theta^r)'
-\frac{\partial^2 f^r}{\partial V^2}\dot V V'\theta^r\nonumber\\
&-&\frac{\partial^2 f^r}{\partial V\partial W}\dot W V N^r(\theta^0)'
-\frac{\partial^2 f^r}{\partial V\partial W}\dot W V(\theta^r)'
-\frac{\partial^2 f^r}{\partial V\partial W}\dot W V'\theta^r\nonumber\\
&-&\frac{\partial f^r}{\partial V}\dot V N^r(\theta^0)'
-\frac{\partial f^r}{\partial V}V\dot N^r(\theta^0)'
-\frac{\partial f^r}{\partial V}V N^r(\dot\theta^0)'
-\frac{\partial f^r}{\partial V}\dot V(\theta^r)'
-\frac{\partial f^r}{\partial V}V(\dot\theta^r)'\nonumber\\
\label{eLbtr}
&-&\frac{\partial f^r}{\partial V}\dot V'\theta^r
-\frac{\partial f^r}{\partial V}V'\dot\theta^r
-\frac{\partial^2 f^r}{\partial V\partial W}\dot V W'\theta^r
-\frac{\partial^2 f^r}{\partial W^2}\dot W W'\theta^r
-\frac{\partial f^r}{\partial W}\dot W'\theta^r
-\frac{\partial f^r}{\partial W}W'\dot\theta^r;\end{aligned}$$ $$\frac{\partial L}{\partial\theta^r}
=\partial_0\frac{\partial L}{\partial\dot\theta^r}+\partial_r\frac{\partial L}{\partial(\theta^r)'};$$ $$\begin{aligned}
0&=&\dot{\bar\theta_0}N'-\dot{\bar\theta_0}\frac{\partial f}{\partial W}W'
+(\dot{\bar\theta_0})'\frac{\partial f}{\partial V}V
+\dot{\bar\theta_0}\frac{\partial^2 f}{\partial V^2}V'V
+\dot{\bar\theta_0}\frac{\partial^2 f}{\partial V\partial W}W'V\nonumber\\
&-&\ddot{\bar\theta_r}+(\dot{\bar\theta_r})'N^r+2\dot{\bar\theta_r}(N^r)'
+(\dot{\bar\theta_r})'\frac{\partial f^r}{\partial V}V\nonumber\\
\label{eLtr}
&+&\dot{\bar\theta_r}\frac{\partial^2 f^r}{\partial V^2}V'V
+\dot{\bar\theta_r}\frac{\partial^2 f^r}{\partial V\partial W}W'V
-\dot{\bar\theta_r}\frac{\partial f^r}{\partial W}W'.\end{aligned}$$ Variation with respect to $\pi_N$, $\pi_{N^r}$ yields the gauge conditions (\[d.c.1\]), (\[d.c.2\]): $$\begin{aligned}
\label{d.c.11}
\dot N&=&\frac{\partial f}{\partial V}\dot V+\frac{\partial f}{\partial W}\dot W;\\
\label{d.c.22}
\dot N^r&=&\frac{\partial f^r}{\partial V}\dot V+\frac{\partial f^r}{\partial W}\dot W.\end{aligned}$$ The equations (\[eLN\]) – (\[d.c.22\]) form the extended set of Lagrangian equations for the generalized spherically symmetric gravitational model.
The set of Hamiltonian equations in extended phase space
========================================================
In this Appendix we present the full set of Hamiltonian equations in extended phase space. $$\begin{aligned}
\dot N&=&\frac{\partial f}{\partial V}V'N^r
+\frac{\partial f}{\partial W}W'N^r
+\frac{\partial f}{\partial V}V(N^r)'
+\frac N W P_V\frac{\partial f}{\partial W}
+\frac N W P_W\frac{\partial f}{\partial V}
-\frac{N V}{W^2}P_V\frac{\partial f}{\partial V}\nonumber\\
\label{DN}
&-&\frac{N V}{W^2}P_N\left(\frac{\partial f}{\partial V}\right)^2
+\frac{2N}W P_N\frac{\partial f}{\partial V}\frac{\partial f}{\partial W}
+\frac N W P_{N^r}\frac{\partial f}{\partial V}\frac{\partial f^r}{\partial W}
+\frac N W P_{N^r}\frac{\partial f}{\partial W}\frac{\partial f^r}{\partial V}
-\frac{N V}{W^2}P_{N^r}\frac{\partial f}{\partial V}\frac{\partial f^r}{\partial V};\\
\dot P_N&=&-\frac1W P_V P_W+\frac V{2W^2}P_V^2+\frac{W''W}V-\frac{V'W'W}{V^2}
+\frac{(W')^2}{2V}-\frac V2\nonumber\\
&-&\frac1W P_V P_N\frac{\partial f}{\partial W}
-\frac1W P_W P_N\frac{\partial f}{\partial V}
+\frac V{W^2}P_V P_N\frac{\partial f}{\partial V}
-\frac1W P_V P_{N^r}\frac{\partial f^r}{\partial W}
-\frac1W P_W P_{N^r}\frac{\partial f^r}{\partial V}\nonumber\\
&+&\frac V{W^2}P_V P_{N^r}\frac{\partial f^r}{\partial V}
+\frac V{2W^2} P_N^2\left(\frac{\partial f}{\partial V}\right)^2
+\frac V{2W^2}P_{N^r}^2\left(\frac{\partial f^r}{\partial V}\right)^2
-\frac1W P_N^2\frac{\partial f}{\partial V}\frac{\partial f}{\partial W}
-\frac1W P_{N^r}^2\frac{\partial f^r}{\partial V}\frac{\partial f^r}{\partial W}\nonumber\\
&-&\frac1W P_N P_{N^r}\frac{\partial f}{\partial V}\frac{\partial f^r}{\partial W}
-\frac1W P_N P_{N^r}\frac{\partial f}{\partial W}\frac{\partial f^r}{\partial V}
+\frac V{W^2}P_N P_{N^r}\frac{\partial f}{\partial V}\frac{\partial f^r}{\partial V}\nonumber\\
&+&\frac1{N^2}\bar P_{\theta^0}P_{\bar\theta_0}
-\frac{N^r}{N^2}\bar P_{\theta^r}P_{\bar\theta_0}
-\frac1N(\bar P_{\theta^0})'\theta^r-\frac1N\bar P_{\theta^0}(\theta^r)'\nonumber\\
&+&\frac{(N^r)'}N\bar P_{\theta^r}\theta^r
+\frac{N^r}N(\bar P_{\theta^r})'\theta^r
+\frac{N^r}N\bar P_{\theta^r}(\theta^r)'
-\frac{2N}{V^2}\bar P_{\theta^r}(\theta^0)'\nonumber\\
&+&\frac V{N^2}\frac{\partial f}{\partial V}\bar P_{\theta^0}(\theta^r)'
+\frac{V N^r}{N^2}\frac{\partial f}{\partial V}\bar P_{\theta^0}(\theta^0)'
-\frac{V N^r}{N^2}\frac{\partial f}{\partial V}\bar P_{\theta^r}(\theta^r)'
-\frac{V(N^r)^2}{N^2}\frac{\partial f}{\partial V}\bar P_{\theta^r}(\theta^0)'\nonumber\\
\label{DPN}
&+&\frac{V'}{N^2}\frac{\partial f}{\partial V}\bar P_{\theta^0}\theta^r
-\frac{V'N^r}{N^2}\frac{\partial f}{\partial V}\bar P_{\theta^r}\theta^r
+\frac{W'}{N^2}\frac{\partial f}{\partial W}\bar P_{\theta^0}\theta^r
-\frac{W'N^r}{N^2}\frac{\partial f}{\partial W}\bar P_{\theta^r}\theta^r;\\
\dot N^r&=&\frac{\partial f^r}{\partial V}V'N^r
+\frac{\partial f^r}{\partial W}W'N^r
+\frac{\partial f^r}{\partial V}V(N^r)'
+\frac N W P_V\frac{\partial f^r}{\partial W}
+\frac N W P_W\frac{\partial f^r}{\partial V}
-\frac{N V}{W^2}P_V\frac{\partial f^r}{\partial V}\nonumber\\
&-&\frac{N V}{W^2}P_{N^r}\left(\frac{\partial f^r}{\partial V}\right)^2
+\frac{2N}W P_{N^r}\frac{\partial f^r}{\partial V}\frac{\partial f^r}{\partial W}
+\frac N W P_N\frac{\partial f}{\partial V}\frac{\partial f^r}{\partial W}\nonumber\\
\label{DNr}
&+&\frac N W P_N\frac{\partial f}{\partial W}\frac{\partial f^r}{\partial V}
-\frac{N V}{W^2}P_N\frac{\partial f}{\partial V}\frac{\partial f^r}{\partial V}';\\
\dot P_{N^r}&=&(P_V)'V-P_W W'
-P_N\frac{\partial f}{\partial W}W'
+(P_N)'\frac{\partial f}{\partial V}V
+P_N\frac{\partial^2 f}{\partial V^2}V V'
+P_N\frac{\partial^2 f}{\partial V\partial W}V W'\nonumber\\
&-&P_{N^r}\frac{\partial f^r}{\partial W}W'
+(P_{N^r})'\frac{\partial f^r}{\partial V}V
+P_{N^r}\frac{\partial^2 f^r}{\partial V^2}V V'
+P_{N^r}\frac{\partial^2 f^r}{\partial V\partial W}V W'\nonumber\\
&+&\frac1N\bar P_{\theta^r}P_{\bar\theta_0}
-\bar P_{\theta^0}(\theta^0)'
-2\bar P_{\theta^r}(\theta^r)'
-(\bar P_{\theta^r})'\theta^r
-\frac{N'}N\bar P_{\theta^r}\theta^r\nonumber\\
&-&\frac{\partial f^r}{\partial V}V\bar P_{\theta^r}(\theta^0)'
-\frac V N\frac{\partial f}{\partial V}\bar P_{\theta^0}(\theta^0)'
+\frac V N\frac{\partial f}{\partial V}\bar P_{\theta^r}(\theta^r)'\nonumber\\
\label{DPNr}
&+&\frac{2V N^r}N\frac{\partial f}{\partial V}\bar P_{\theta^r}(\theta^0)'
+\frac{V'}N\frac{\partial f}{\partial V}\bar P_{\theta^r}\theta^r
+\frac{W'}N\frac{\partial f}{\partial W}\bar P_{\theta^r}\theta^r.\end{aligned}$$ It can be checked by direct calculations that Eqs. (\[DPN\]), (\[DPNr\]) coincide with the constraints equations in the Lagrangian formalism (\[eLN\]), (\[eLNr\]), while Eqs. (\[DN\]), (\[DNr\]) correspond to the gauge conditions (\[d.c.11\]), (\[d.c.22\]). The other Hamiltonian equations are: $$\begin{aligned}
\dot V&=&\frac N W P_W-\frac{N V}{W^2}P_V+V'N^r+V(N^r)'\nonumber\\
\label{DV}
&+&\frac N W P_N\frac{\partial f}{\partial W}
-\frac{N V}{W^2}P_N\frac{\partial f}{\partial V}
+\frac N W P_{N^r}\frac{\partial f^r}{\partial W}
-\frac{N V}{W^2}P_{N^r}\frac{\partial f^r}{\partial V};\\
\dot P_V&=&\frac N{2W^2}P_V^2+(P_V)'N^r+\frac{N'W'W}{V^2}
+\frac{N(W')^2}{2V^2}-\frac N2\nonumber\\
&+&(P_N)'\frac{\partial f}{\partial V}N^r
-P_N\frac{\partial^2 f}{\partial V^2}V(N^r)'
+(P_{N^r})'\frac{\partial f^r}{\partial V}N^r
-P_{N^r}\frac{\partial^2 f^r}{\partial V^2}V(N^r)'\nonumber\\
&-&\frac N W P_V P_N\frac{\partial^2 f}{\partial V\partial W}
-\frac N W P_W P_N\frac{\partial^2 f}{\partial V^2}
+\frac N{W^2}P_V P_N\frac{\partial f}{\partial V}
+\frac{N V}{W^2}P_V P_N\frac{\partial^2 f}{\partial V^2}\nonumber\\
&-&\frac N W P_V P_{N^r}\frac{\partial^2 f^r}{\partial V\partial W}
-\frac N W P_W P_{N^r}\frac{\partial^2 f^r}{\partial V^2}
+\frac N{W^2}P_V P_{N^r}\frac{\partial f^r}{\partial V}
+\frac{N V}{W^2}P_V P_{N^r}\frac{\partial^2 f^r}{\partial V^2}\nonumber\\
&+&\frac N{2W^2} P_N^2\left(\frac{\partial f}{\partial V}\right)^2
+\frac{N V}{W^2} P_N^2\frac{\partial f}{\partial V}\frac{\partial^2 f}{\partial V^2}
+\frac N{2W^2}P_{N^r}^2\left(\frac{\partial f^r}{\partial V}\right)^2
+\frac{N V}{W^2}P_{N^r}^2\frac{\partial f^r}{\partial V}
\frac{\partial^2 f^r}{\partial V^2}\nonumber\\
&-&\frac N W P_N^2\frac{\partial^2 f}{\partial V^2}
\frac{\partial f}{\partial W}
-\frac N W P_N^2\frac{\partial f}{\partial V}
\frac{\partial^2 f}{\partial V\partial W}
-\frac N W P_{N^r}^2\frac{\partial^2 f^r}{\partial V^2}
\frac{\partial f^r}{\partial W}
-\frac N W P_{N^r}^2\frac{\partial f^r}{\partial V}
\frac{\partial^2 f^r}{\partial V\partial W}\nonumber\\
&-&\frac N W P_N P_{N^r}\frac{\partial^2 f}{\partial V^2}
\frac{\partial f^r}{\partial W}
-\frac N W P_N P_{N^r}\frac{\partial f}{\partial V}
\frac{\partial^2 f^r}{\partial V\partial W}
-\frac N W P_N P_{N^r}\frac{\partial^2 f}{\partial V\partial W}
\frac{\partial f^r}{\partial V}
-\frac N W P_N P_{N^r}\frac{\partial f}{\partial W}
\frac{\partial^2 f^r}{\partial V^2}\nonumber\\
&+&\frac N{W^2}P_N P_{N^r}\frac{\partial f}{\partial V}
\frac{\partial f^r}{\partial V}
+\frac{N V}{W^2}P_N P_{N^r}\frac{\partial^2 f}{\partial V^2}
\frac{\partial f^r}{\partial V}
+\frac{N V}{W^2}P_N P_{N^r}\frac{\partial f}{\partial V}
\frac{\partial^2 f^r}{\partial V^2}\nonumber\\
&+&\frac{2N^2}{V^3}\bar P_{\theta^r}(\theta^0)'
+\frac{\partial f^r}{\partial V}(\bar P_{\theta^r})'\theta^r
-\frac{\partial^2 f^r}{\partial V^2}V N^r\bar P_{\theta^r}(\theta^0)'
-\frac{\partial f^r}{\partial V}N^r\bar P_{\theta^r}(\theta^0)'
-\frac{\partial^2 f^r}{\partial V^2}V\bar P_{\theta^r}(\theta^r)'\nonumber\\
&-&\frac V N\frac{\partial^2 f}{\partial V^2}\bar P_{\theta^0}(\theta^r)'
-\frac{N^r}N\frac{\partial f}{\partial V}\bar P_{\theta^0}(\theta^0)'
-\frac{V N^r}N\frac{\partial^2 f}{\partial V^2}\bar P_{\theta^0}(\theta^0)'
+\frac{V N^r}N\frac{\partial^2 f}{\partial V^2}\bar P_{\theta^r}(\theta^r)'\nonumber\\
&+&\frac{(N^r)^2}N\frac{\partial f}{\partial V}\bar P_{\theta^r}(\theta^0)'
+\frac{V(N^r)^2}N\frac{\partial^2 f}{\partial V^2}\bar P_{\theta^r}(\theta^0)'
-\frac{N'}{N^2}\frac{\partial f}{\partial V}\bar P_{\theta^0}\theta^r
+\frac1N\frac{\partial f}{\partial V}(\bar P_{\theta^0})'\theta^r\nonumber\\
\label{DPV}
&-&\frac{(N^r)'}N\frac{\partial f}{\partial V}\bar P_{\theta^r}\theta^r
+\frac{N'N^r}{N^2}\frac{\partial f}{\partial V}\bar P_{\theta^r}\theta^r
-\frac{N^r}N\frac{\partial f}{\partial V}(\bar P_{\theta^r})'\theta^r;\\
\label{DW}
\dot W&=&\frac N W P_V+W'N^r
+\frac N W P_N\frac{\partial f}{\partial V}
+\frac N W P_{N^r}\frac{\partial f^r}{\partial V};\\
\dot P_W&=&\frac N{W^2}P_V P_W-\frac{N V}{W^3}P_V^2+(P_W)'N^r+P_W(N^r)'
+\frac{N''W}V-\frac{N'V'W}{V^2}+\frac{N'W'}V+\frac{N W''}V-\frac{N V'W'}{V^2}\nonumber\\
&+&(P_N)'\frac{\partial f}{\partial W}N^r
+P_N\frac{\partial f}{\partial W}(N^r)'
-P_N\frac{\partial^2 f}{\partial V\partial W}V(N^r)'
+(P_{N^r})'\frac{\partial f^r}{\partial W}N^r
+P_{N^r}\frac{\partial f^r}{\partial W}(N^r)'\nonumber\\
&-&P_{N^r}\frac{\partial^2 f^r}{\partial V\partial W}V(N^r)'
+\frac N{W^2}P_V P_N\frac{\partial f}{\partial W}
-\frac N W P_V P_N\frac{\partial^2 f}{\partial W^2}
+\frac N{W^2}P_W P_N\frac{\partial f}{\partial V}
-\frac N W P_W P_N\frac{\partial^2 f}{\partial V\partial W}\nonumber\\
&-&\frac{2N V}{W^3}P_V P_N\frac{\partial f}{\partial V}
+\frac{N V}{W^2}P_V P_N\frac{\partial^2 f}{\partial V\partial W}
+\frac N{W^2}P_V P_{N^r}\frac{\partial f^r}{\partial W}
-\frac N W P_V P_{N^r}\frac{\partial^2 f^r}{\partial W^2}\nonumber\\
&+&\frac N{W^2}P_W P_{N^r}\frac{\partial f^r}{\partial V}
-\frac N W P_W P_{N^r}\frac{\partial^2 f^r}{\partial V\partial W}
-\frac{2N V}{W^3}P_V P_{N^r}\frac{\partial f^r}{\partial V}
+\frac{N V}{W^2}P_V P_{N^r}\frac{\partial^2 f^r}{\partial V\partial W}\nonumber\\
&-&\frac{N V}{W^3}P_N^2\left(\frac{\partial f}{\partial V}\right)^2
+\frac{N V}{W^2}P_N^2\frac{\partial f}{\partial V}
\frac{\partial^2 f}{\partial V\partial W}
-\frac{N V}{W^3}P_{N^r}^2\left(\frac{\partial f^r}{\partial V}\right)^2
+\frac{N V}{W^2}P_{N^r}^2\frac{\partial f^r}{\partial V}
\frac{\partial^2 f^r}{\partial V\partial W}\nonumber\\
&+&\frac N{W^2}P_N^2\frac{\partial f}{\partial V}
\frac{\partial f}{\partial W}
-\frac N W P_N^2\frac{\partial^2 f}{\partial V\partial W}
\frac{\partial f}{\partial W}
-\frac N W P_N^2\frac{\partial f}{\partial V}
\frac{\partial^2 f}{\partial W^2}\nonumber\\
&+&\frac N{W^2}P_{N^r}^2\frac{\partial f^r}{\partial V}
\frac{\partial f^r}{\partial W}
-\frac N W P_{N^r}^2\frac{\partial^2 f^r}{\partial V\partial W}
\frac{\partial f^r}{\partial W}
-\frac N W P_{N^r}^2\frac{\partial f^r}{\partial V}
\frac{\partial^2 f^r}{\partial W^2}\nonumber\\
&+&\frac N{W^2}P_N P_{N^r}\frac{\partial f}{\partial V}
\frac{\partial f^r}{\partial W}
-\frac N W P_N P_{N^r}\frac{\partial^2 f}{\partial V\partial W}
\frac{\partial f^r}{\partial W}
-\frac N W P_N P_{N^r}\frac{\partial f}{\partial V}
\frac{\partial^2 f^r}{\partial W^2}\nonumber\\
&+&\frac N{W^2}P_N P_{N^r}\frac{\partial f}{\partial W}
\frac{\partial f^r}{\partial V}
-\frac N W P_N P_{N^r}\frac{\partial^2 f}{\partial W^2}
\frac{\partial f^r}{\partial V}
-\frac N W P_N P_{N^r}\frac{\partial f}{\partial W}
\frac{\partial^2 f^r}{\partial V\partial W}\nonumber\\
&-&\frac{2N V}{W^3}P_N P_{N^r}\frac{\partial f}{\partial V}
\frac{\partial f^r}{\partial V}
+\frac{N V}{W^2}P_N P_{N^r}\frac{\partial^2 f}{\partial V\partial W}
\frac{\partial f^r}{\partial V}
+\frac{N V}{W^2}P_N P_{N^r}\frac{\partial f}{\partial V}
\frac{\partial^2 f^r}{\partial V\partial W}\nonumber\\
&+&\frac{\partial f^r}{\partial W}(\bar P_{\theta^r})'\theta^r
+\frac{\partial f^r}{\partial W}\bar P_{\theta^r}(\theta^r)'
-\frac{\partial^2 f^r}{\partial V\partial W}V N^r\bar P_{\theta^r}(\theta^0)'
-\frac{\partial^2 f^r}{\partial V\partial W}
V\bar P_{\theta^r}(\theta^r)'\nonumber\\
&-&\frac V N\frac{\partial^2 f}{\partial V\partial W}\bar P_{\theta^0}(\theta^r)'
-\frac{V N^r}N\frac{\partial^2 f}{\partial V\partial W}\bar P_{\theta^0}(\theta^0)'
+\frac{V N^r}N\frac{\partial^2 f}{\partial V\partial W}
\bar P_{\theta^r}(\theta^r)'\nonumber\\
&+&\frac{V(N^r)^2}N\frac{\partial^2 f}{\partial V\partial W}\bar P_{\theta^r}(\theta^0)'
-\frac{N'}{N^2}\frac{\partial f}{\partial W}\bar P_{\theta^0}\theta^r
+\frac1N\frac{\partial f}{\partial W}(\bar P_{\theta^0})'\theta^r
+\frac1N\frac{\partial f}{\partial W}\bar P_{\theta^0}(\theta^r)'\nonumber\\
\label{DPW}
&-&\frac{(N^r)'}N\frac{\partial f}{\partial W}\bar P_{\theta^r}\theta^r
+\frac{N^r N'}{N^2}\frac{\partial f}{\partial W}\bar P_{\theta^r}\theta^r
-\frac{N^r}N\frac{\partial f}{\partial W}(\bar P_{\theta^r})'\theta^r
-\frac{N^r}N\frac{\partial f}{\partial W}\bar P_{\theta^r}(\theta^r)'.\end{aligned}$$ Eqs. (\[DV\]) – (\[DPW\]) are equivalent to the dynamical Lagrangian equations (\[eLV\]), (\[eLW\]). $$\begin{aligned}
\label{Dt0}
\dot\theta^0&=&\frac1N P_{\bar\theta_0}-\frac{N'}N\theta^r+N^r(\theta^0)'
+\frac V N\frac{\partial f}{\partial V}(\theta^r)'
+\frac{V N^r}N\frac{\partial f}{\partial V}(\theta^0)'
+\frac{V'}N\frac{\partial f}{\partial V}\theta^r
+\frac{W'}N\frac{\partial f}{\partial W}\theta^r;\\
\label{DPbt0}
\dot P_{\bar\theta_0}&=&0;\\
\label{Dbt0}
\dot{\bar\theta_0}&=&\frac1N\bar P_{\theta^0}-\frac{N^r}N\bar P_{\theta^r};\\
\dot{\bar P}_{\theta^0}&=&(N^r)'\bar P_{\theta^0}+N^r(\bar P_{\theta^0})'
+\frac{2N N'}{V^2}\bar P_{\theta^r}-\frac{2N^2V'}{V^3}\bar P_{\theta^r}
+\frac{N^2}{V^2}(\bar P_{\theta^r})'\nonumber\\
&+&\frac{\partial^2 f^r}{\partial V^2}V V'N^r\bar P_{\theta^r}
+\frac{\partial^2 f^r}{\partial V\partial W}V W'N^r\bar P_{\theta^r}
+\frac{\partial f^r}{\partial V}V'N^r\bar P_{\theta^r}
+\frac{\partial f^r}{\partial V}V(N^r)'\bar P_{\theta^r}
+\frac{\partial f^r}{\partial V}V N^r(\bar P_{\theta^r})'\nonumber\\
&+&\frac{V'N^r}N\frac{\partial f}{\partial V}\bar P_{\theta^0}
+\frac{V(N^r)'}N\frac{\partial f}{\partial V}\bar P_{\theta^0}
-\frac{V N'N^r}{N^2}\frac{\partial f}{\partial V}\bar P_{\theta^0}
+\frac{V V'N^r}N\frac{\partial^2 f}{\partial V^2}\bar P_{\theta^0}
+\frac{V W'N^r}N\frac{\partial^2 f}{\partial V\partial W}\bar P_{\theta^0}\nonumber\\
&+&\frac{V N^r}N\frac{\partial f}{\partial V}(\bar P_{\theta^0})'
-\frac{V'(N^r)^2}N\frac{\partial f}{\partial V}\bar P_{\theta^r}
-\frac{2V N^r(N^r)'}N\frac{\partial f}{\partial V}\bar P_{\theta^r}
+\frac{V N'(N^r)^2}{N^2}\frac{\partial f}{\partial V}\bar P_{\theta^r}\nonumber\\
\label{DbPt0}
&-&\frac{V V'(N^r)^2}N\frac{\partial^2 f}{\partial V^2}\bar P_{\theta^r}
-\frac{V W'(N^r)^2}N\frac{\partial^2 f}{\partial V\partial W}\bar P_{\theta^r}
-\frac{V(N^r)^2}N\frac{\partial f}{\partial V}(\bar P_{\theta^r})';\\
\dot\theta^r&=&P_{\bar\theta_r}-\frac{N^r}NP_{\bar\theta_0}+N^r(\theta^r)'
-(N^r)'\theta^r+\frac{N'N^r}N\theta^r+\frac{N^2}{V^2}(\theta^0)'\nonumber\\
&+&\frac{\partial f^r}{\partial V}V'\theta^r
+\frac{\partial f^r}{\partial W}W'\theta^r
+\frac{\partial f^r}{\partial V}V N^r(\theta^0)'
+\frac{\partial f^r}{\partial V}V(\theta^r)'\nonumber\\
\label{Dtr}
&-&\frac{V N^r}N\frac{\partial f}{\partial V}(\theta^r)'
-\frac{V(N^r)^2}N\frac{\partial f}{\partial V}(\theta^0)'
-\frac{V'N^r}N\frac{\partial f}{\partial V}\theta^r
-\frac{W'N^r}N\frac{\partial f}{\partial W}\theta^r;\\
\label{DPbtr}
\dot P_{\bar\theta_r}&=&0;\\
\label{Dbtr}
\dot{\bar\theta_r}&=&\bar P_{\theta^r};\\
\dot{\bar P}_{\theta^r}&=&\frac{N'}N\bar P_{\theta^0}
+2(N^r)'\bar P_{\theta^r}+N^r(\bar P_{\theta^r})'
-\frac{N'N^r}N\bar P_{\theta^r}
-\frac{\partial f^r}{\partial W}W'\bar P_{\theta^r}\nonumber\\
&+&\frac{\partial^2 f^r}{\partial V^2}V V'\bar P_{\theta^r}
+\frac{\partial^2 f^r}{\partial V\partial W}V W'\bar P_{\theta^r}
+\frac{\partial f^r}{\partial V}V(\bar P_{\theta^r})'
-\frac{N'V}{N^2}\frac{\partial f}{\partial V}\bar P_{\theta^0}\nonumber\\
&+&\frac{V V'}N\frac{\partial^2 f}{\partial V^2}\bar P_{\theta^0}
+\frac{V W'}N\frac{\partial^2 f}{\partial V\partial W}\bar P_{\theta^0}
+\frac V N\frac{\partial f}{\partial V}(\bar P_{\theta^0})'
-\frac{V(N^r)'}N\frac{\partial f}{\partial V}\bar P_{\theta^r}\nonumber\\
&+&\frac{V N'N^r}{N^2}\frac{\partial f}{\partial V}\bar P_{\theta^r}
-\frac{V V'N^r}N\frac{\partial^2 f}{\partial V^2}\bar P_{\theta^r}
-\frac{V W'N^r}N\frac{\partial^2 f}{\partial V\partial W}\bar P_{\theta^r}
-\frac{V N^r}N\frac{\partial f}{\partial V}(\bar P_{\theta^r})'\nonumber\\
\label{DbPtr}
&-&\frac{W'}N\frac{\partial f}{\partial W}\bar P_{\theta^0}
+\frac{W'N^r}N\frac{\partial f}{\partial W}\bar P_{\theta^r}.\end{aligned}$$ Eqs. (\[Dt0\]) – (\[DbPtr\]) are equivalent to the equations for ghosts (\[eLbt0\]) – (\[eLtr\]).
The explicit expression for addition terms in the gauged Einstein equations
===========================================================================
In this Appendix the additional terms in the Einstein equation (\[g.Ein.eqs\]), $T_{\mu(obs)}^{\nu}$ and $T_{\mu(ghost)}^{\nu}$, which can be obtained by varying the gauge-fixing and ghost action, respectively, are given in the explicit form. $$\begin{aligned}
\label{obs00}
T^0_{0(obs)}&=&\frac1{4\pi}\left(\dot\pi_N\frac1{VW^2}
+\dot\pi_{N^r}\frac{N^r}{NVW^2}\right);\\
\label{obs01}
T^0_{1(obs)}&=&\frac1{4\pi}\dot\pi_{N^r}\frac1{NVW^2};\\
\label{obs11}
T^1_{1(obs)}&=&-\frac1{4\pi}\left[\dot\pi_N\frac1{NW^2}\frac{\partial f}{\partial V}
+\dot\pi_{N^r}\left(\frac{N^r}{NVW^2}
+\frac1{NW^2}\frac{\partial f^r}{\partial V}\right)\right];\\
\label{obs22}
T^2_{2(obs)}=T^3_{3(obs)}&=&-\frac1{8\pi}\left(\dot\pi_N\frac1{N V W}\frac{\partial f}{\partial W}
+\dot\pi_{N^r}\frac1{N V W}\frac{\partial f^r}{\partial W}\right).\end{aligned}$$ $$\begin{aligned}
\label{ghost00}
T^0_{0(ghost)}&=&\frac1{4\pi}\left[-\frac1{V W^2}\dot{\bar\theta_0}\dot{\theta^0}
+\frac1{V W^2}\left(\dot{\bar\theta_0}\theta^r\right)'
+\frac{2N}{V^3 W^2}\dot{\bar\theta_r}(\theta^0)'
+\frac{2N^r}{V W^2}\dot{\bar\theta_0}(\theta^0)'\right.\nonumber\\
&+&\frac{N^r}{N V W^2}\left(\dot{\bar\theta_r}\right)'\theta^r
+\frac{2N^r}{N V W^2}\dot{\bar\theta_r}(\theta^r)'
-\frac{N^r}{N V W^2}\dot{\bar\theta_r}\dot\theta^0
+\frac{2(N^r)^2}{N V W^2}\dot{\bar\theta_r}(\theta^0)'\nonumber\\
&+&\left.\frac{N^r}{N W^2}\frac{\partial f}{\partial V}\dot{\bar\theta_0}(\theta^0)'
+\frac{N^r}{N W^2}\frac{\partial f^r}{\partial V}\dot{\bar\theta_r}(\theta^0)'\right];\\
\label{ghost01}
T^0_{1(ghost)}&=&\frac1{4\pi}\left[\frac1{V W^2}\dot{\bar\theta_0}(\theta^0)'
+\frac1{N V W^2}\left(\dot{\bar\theta_r}\right)'\theta^r
+\frac2{N V W^2}\dot{\bar\theta_r}(\theta^r)'
-\frac1{N V W^2}\dot{\bar\theta_r}\dot\theta^0\right.\nonumber\\
&+&\left.\frac{2N^r}{N V W^2}\dot{\bar\theta_r}(\theta^0)'
+\frac1{N W^2}\frac{\partial f}{\partial V}\dot{\bar\theta_0}(\theta^0)'
+\frac1{N W^2}\frac{\partial f^r}{\partial V}\dot{\bar\theta_r}(\theta^0)'\right];\\
\label{ghost11}
T^1_{1(ghost)}&=&\frac1{4\pi}\left[-\frac{2N}{V^3W^2}\dot{\bar\theta_r}(\theta^0)'
-\frac{N^r}{N V W^2}\left(\dot{\bar\theta_r}\right)'\theta^r
-\frac{N^r}{V W^2}\dot{\bar\theta_0}(\theta^0)'
-\frac{2N^r}{N V W^2}\dot{\bar\theta_r}(\theta^r)'\right.\nonumber\\
&+&\frac{N^r}{N V W^2}\dot{\bar\theta_r}\dot\theta^0
-\frac{2(N^r)^2}{N V W^2}\dot{\bar\theta_r}(\theta^0)'
-\frac1{N W^2}\frac{\partial f}{\partial V}\left(\dot{\bar\theta_0}\right)'\theta^r\nonumber\\
&-&\frac1{N W^2}\frac{\partial f^r}{\partial V}\left(\dot{\bar\theta_r}\right)'\theta^r
+\frac V{N W^2}\frac{\partial^2 f}{\partial V^2}\dot{\bar\theta_0}(\theta^r)'
+\frac V{N W^2}\frac{\partial^2 f^r}{\partial V^2}\dot{\bar\theta_r}(\theta^r)'\nonumber\\
&+&\left.\frac{V N^r}{N W^2}\frac{\partial^2 f}{\partial V^2}\dot{\bar\theta_0}(\theta^0)'
+\frac{V N^r}{N W^2}\frac{\partial^2 f^r}{\partial V^2}\dot{\bar\theta_r}(\theta^0)'\right];\\
\label{ghost22}
T^2_{2(ghost)}=T^3_{3(ghost)}&=&-\frac1{8\pi}
\left[\frac1{N V W}\frac{\partial f}{\partial W}\left(\dot{\bar\theta_0}\right)'\theta^r
+\frac1{N V W}\frac{\partial f^r}{\partial W}\left(\dot{\bar\theta_r}\right)'\theta^r
+\frac1{N V W}\frac{\partial f}{\partial W}\dot{\bar\theta_0}\left(\theta^r\right)'\right.\nonumber\\
&+&\frac1{N V W}\frac{\partial f^r}{\partial W}\dot{\bar\theta_r}\left(\theta^r\right)'
-\frac{N^r}{N W}\frac{\partial^2 f}{\partial V\partial W}\dot{\bar\theta_0}(\theta^0)'
-\frac{N^r}{N W}\frac{\partial^2 f^r}{\partial V\partial W}\dot{\bar\theta_r}(\theta^0)'\nonumber\\
&-&\left.\frac1{N W}\frac{\partial^2 f}{\partial V\partial W}\dot{\bar\theta_0}(\theta^r)'
-\frac1{N W}\frac{\partial^2 f^r}{\partial V\partial W}\dot{\bar\theta_r}(\theta^r)'\right].\end{aligned}$$ The rest components are equal to zero. When comparing (\[obs00\]) – (\[obs22\]) and (\[ghost00\]) – (\[ghost22\]) with the terms in (\[eLN\]) – (\[eLW\]) that include $\pi_N$, $\pi_{N^r}$ and ghost variables, one should keep in mind that Eq. (\[eLN\]) is a linear combination of $\left(0\atop 0\right)$ and $\left(0\atop 1\right)$ Einstein equations, namely, $$\frac{N^2V W^2}2\left(g^{00}G^0_0+g^{01}G^0_1\right),$$ where $G_{\mu}^{\nu}=R_{\mu}^{\nu}-\displaystyle\frac12\delta_{\mu}^{\nu}R$ is the left-hand side of the Einstein equations (\[g.Ein.eqs\]), Eq. (\[eLNr\]) is the $\left(0\atop 1\right)$ Einstein equation multiplied by $\displaystyle\frac{N V W^2}2$, Eq. (\[eLV\]) is a linear combination of $\left(0\atop 1\right)$ and $\left(1\atop 1\right)$ Einstein equations, $$-\frac{N W^2}{2V^2}\left(g_{10}G^0_1+g_{11}G^1_1\right),$$ and Eq. (\[eLW\]) is the $\left(2\atop 2\right) \left({\rm or} \left(3\atop 3\right)\right)$ Einstein equation multiplied by $(-N V W)$.
[99]{} T. P. Shestakova, [*Class. Quantum Grav.*]{} [**28**]{} (2011), 055009. P. A. M. Dirac, [*Can. J. Math.*]{} [**2**]{} (1950), P. 129–148. P. A. M. Dirac, [*Proc. Roy. Soc.*]{} [**A246**]{} (1958), P. 326–332. P. A. M. Dirac, [*Proc. Roy. Soc.*]{} [**A246**]{} (1958), P. 333–343. B. S. DeWitt, [*Phys. Rev.*]{} [**160**]{} (1967), P. 1113–1148. C. Isham, [*Canonical quantum gravity and the problem of time*]{}, lectures presented at NATO Advanced Study Institute, Salamanca, June 1992, Preprint gr-qc/9210011. T. P. Shestakova and C. Simeone, [*Gravitation & Cosmology*]{} [**10**]{} (2004), P. 161–176. C. Rovelli, [*Class. Quantum Grav.*]{} [**28**]{} (2011), 153002. R. Arnowitt, S. Deser and C. W. Misner, in: [*Gravitation, an Introduction to Current Research*]{}, ed. by L. Witten, John Wiley & Sons, New York (1963) P. 227–284. N. Kiriushcheva and S. V. Kuzmin, [*Central Eur. J. Phys.*]{} [**9**]{} (2011), P. 576–615 E. S. Fradkin and G. A. Vilkovisky, [*Phys. Lett.*]{} [**B55**]{} (1975), P. 224–226. I. A. Batalin and G. A. Vilkovisky, [*Phys. Lett.*]{} [**B69**]{} (1977), P. 309–312. E. S. Fradkin and T. E. Fradkina, [*Phys. Lett.*]{} [**B72**]{} (1978), P. 343–348. M. Hennaux, [*Phys. Rep.*]{} [**126**]{} (1985) P 1–66. V. A. Savchenko, T. P. Shestakova and G. M. Vereshkov, [*Gravitation & Cosmology*]{} [**7**]{} (2001), P. 18–28. V. A. Savchenko, T. P. Shestakova and G. M. Vereshkov, [*Gravitation & Cosmology*]{} [**7**]{} (2001), P. 102–116. K. Schwarzschild, [*Sitzungsber. d. Berl. Akad.*]{} (1916), S. 189–196. K. Enqvist, [*Gen. Rel. Grav.*]{} [**40**]{} (2008), P. 451–466. T. P. Shestakova, [*Gravitation & Cosmology*]{} [**12**]{} (2006), P. 223–226. L. Castellani, [*Ann. Phys.*]{} [**143**]{} (1982), P. 357–371. S. Weinberg, [*Quantum theory of fields*]{}, Cambridge University Press, Cambridge, UK (1996). J. J. Halliwell, [*Phys. Rev.*]{} [**D38**]{} (1988), P. 2468–2481.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The $D^0$ decay into $K^0_s$ and a scalar resonance, $f_0(500)$, $f_0(980)$, $a_0(980)$, is studied obtaining the scalar resonances from final state interaction of a pair of mesons produced in a first step in the $D^0$ decay into $K^0_s$ and the pair of pseudoscalar mesons. This weak decay is very appropriate for this kind of study because it allows to produce the three resonances in the same decay in a process that is Cabibbo allowed, hence the rates obtained are large compared to those of $\bar{B}^0$ decays into $J/\psi$ and a scalar meson that have at least one Cabibbo suppressed vertex. Concretely the $a_0(980)$ production is Cabibbo allowed here, while it cannot be seen in the $\bar{B}^0_s$ decay into $J/\psi a_0(980)$ and is doubly Cabibbo suppressed in the $\bar{B}^0$ decay into $J/\psi a_0(980)$ and has not been identified there. The fact that the three resonances can be seen in the same reaction, because there is no isospin conservation in the weak decays, offers a unique opportunity to test the ideas of the chiral unitary approach where these resonances are produced from the interaction of pairs of pseudoscalar mesons.'
author:
- 'Ju-Jun Xie'
- 'L.R. Dai'
- 'E. Oset'
title: 'The low lying scalar resonances in the $D^0$ decays into $K^0_s$ and $f_0(500)$, $f_0(980)$, $a_0(980)$'
---
Introduction
============
The rates for $D^0$ decay into $K^0_s$ and a scalar resonance, $f_0(980)$, $a_0(980)$ are measured by the CLEO collaboration in Ref. [@Muramatsu:2002jp] and Ref. [@Rubin:2004cq] respectively and the rates are relatively large. The $f_0(980)$ is seen through its decay into $\pi^+ \pi^-$ and the $a_0(980)$ through the $\pi^0 \eta$ channel. Related references on the issue can be seen in the PDG [@pdg]. Theoretical work on these decays is scarce and is mostly devoted to issues related to CP violation or $D^0-D^{*0}$ mixing. In Ref. [@kaminski] a thorough study is done of the $D^0 \to K^0_s \pi^+ \pi^-$ reaction and the amplitude is parametrized in terms of form factors, resonance parameters and different couplings, amounting to a set of 33 free parameters, which are fitted to the Belle [@Poluektov:2010wz] and BaBar [@delAmoSanchez:2010xz] data. The purpose is to have a good amplitude that can be used to determine the $D^0-D^{*0}$ mixing parameters and the Cabibbo-Kobayashi-Maskawa (CKM) angle $\gamma$.
The aim of the present work is different, we only evaluate the part of the $D^0 \to K^0_s \pi^+ \pi^-$ amplitude corresponding to a $K^0_s$ and two pions propagating in $s$-wave, which will show the $f_0(500)$ and $f_0(980)$ resonances. In addition we study the $D^0
\to K^0_s \pi^0 \eta$ amplitude, where the $a_0(980)$ resonance shows up, and relate it to the former one. However, we show that, by using basic symmetries and the chiral unitary approach to deal with the meson meson interaction in coupled channels, one is able to determine the shapes of the different amplitudes and the relative weight to each other with no free parameters. Hence genuine predictions for the shapes of these amplitudes and the relative weights of $f_0(500)$, $f_0(980)$ and $a_0(980)$ can be made and compared with experiment.
The chiral unitary approach for meson meson interaction makes use of the Bethe Salpeter (BS) equation in coupled channels. One takes all possible meson meson channels that couple within $SU(3)$ to certain given quantum numbers and the BS equation guaranties exact unitary. The kernel (potential) for the BS equation is taken from the chiral Lagrangians [@Gasser:1983yg; @Bernard:1995dp] and there is freedom for only some regularization scale in the meson meson loops, which is fitted to the meson meson scattering data. A good agreement with experimental data is obtained up to 1.2 GeV [@npa; @ramonet; @kaiser; @markushin; @juanito; @rios]. One of the consequences of this approach is that the resonances $f_0(500)$, $f_0(980)$, $a_0(980)$ and $\kappa(800)$ are automatically generated from these potentials and the use of the BS equations. In this way these resonances qualify as dynamically generated states, some kind of composite, or molecular, meson meson states, in the same way as the deuteron qualifies as a bound state of a proton and a neutron and not a more exotic object [@Weinberg:1965zz]. The approach not only provides the meson meson amplitudes but has been tested successfully in virtually any reaction where any of the former resonances is produced. The latest test was the study of the $B^0$ and $B^0_s$ decays into $J/\psi f_0(500)$ and $J/\psi f_0(980)$ which was done in Ref. [@weihong] (a list of different reactions where the former resonances are produced can also be found there), where a natural explanation was given of the observed facts that the $\bar{B}^0_s$ decays into $J/\psi f_0(980)$, while no signal is seen for $J/\psi f_0(500)$, and the $\bar{B}^0$ decays into $J/\psi
f_0(500)$ and only a small fraction is seen for the $J/\psi
f_0(980)$.
The $D^0$ decay into $K^0_s$ and a scalar resonance, $f_0(500)$, $f_0(980)$, $a_0(980)$ is a privileged case to test the nature of these resonances. Indeed, as we shall see, the three processes are Cabibbo allowed and the rates of production are big compared to those of the $\bar{B}^0$ decays into $J/\psi$ and one of these resonances, where necessarily one of the vertices, the $V_{cb}$, is Cabibbo suppressed [@Aaij:2011fx; @Aaij:2013zpt; @Stone:2013eaa]. On the other hand, the $a_0(980)$ has not been reported in $\bar{B}^0$, $\bar{B}^0_s$ decays. As one can see in Ref. [@Stone:2013eaa; @weihong], in the decay of $\bar{B}^0_s$ into $J/\psi$ one gets an extra $s \bar s$ pair that has $I=0$ and does not allow the $a_0(980)$ production upon hadronization. On the other hand in the $B^0$ decay into $J/\psi$ one gets an extra $d
\bar d$ pair that could lead to the $a_0(980)$ upon hadronization, but the process is doubly Cabibbo suppressed. It is found there that a signal is seen for the $f_0(500)$ production and only a small fraction is reported for $f_0(980)$ production [@Aaij:2013zpt]. One should expect also a minor rate for $a_0(980)$ production in this case and, in fact, this mode of decay is not reported. In the present case the $a_0(980)$ production is allowed and the rates are large [@Rubin:2004cq]. The fact that we have now weak interactions that allow for isospin violation permit that both the $f_0(980)$ and $a_0(980)$ resonances are produced in the same reaction. This is a novelty with respect to strong interactions that are isospin conserving. The present weak decay presents then a new challenge since one can determine the relative weight of production of each one of these resonances in the same reaction, a new situation with respect to what one has in strong interaction reactions.
Formalism
=========
The process for $D^0 \to K^0_s R$ proceeds at the elementary quark level as depicted in Fig. \[Fig:feyn\] (A). The process is Cabibbo allowed, the $s\bar{d}$ pair produces the $\bar{K}^0$, which will convert to the observed $K^0_s$ through time evolution with the weak interaction. The remaining $u\bar{u}$ pair gets hadronized adding an extra $\bar{q}q$ with the quantum mumbers of the vacuum, $\bar{u}u +
\bar{d}d + \bar{s}s$. This topology is the same as for the $\bar{B}_s \to J/\psi s\bar{s}$ (substituting the $s\bar{d}$ by $c\bar{c}$) [@Stone:2013eaa], that upon hadronization of the $s\bar{s}$ pair leads to the production of the $f_0(980)$ [@weihong], which couples mostly to the hadronized $K
\bar{K}$ components.
![(A): Dominant diagrams for $D^0 \to \bar{K}^0 u \bar{u})$ and (B): hadronization of the $u \bar{u}$ to give two mesons.[]{data-label="Fig:feyn"}](feynd.eps)
The hadronization is implemented in an easy way following the work of Ref. [@alberzou]. One starts with the $q\bar{q}$ matrix $M$
$$\label{eq:1}
M=\left(
\begin{array}{ccc}
u\bar u & u \bar d & u\bar s \\
d\bar u & d\bar d & d\bar s \\
s\bar u & s\bar d & s\bar s \\
\end{array}
\right)$$
which has the property $$\label{eq:2}
M\cdot M=M \times (\bar{u}u + \bar{d}d + \bar{s}s).$$
Hence the $u\bar{u}$ component of Fig. \[Fig:feyn\] (B) can be written as, $$u\bar{u}(\bar{u}u + \bar{d}d + \bar{s}s) = (M \cdot M)_{11}.$$
Next, we rewrite the $q\bar{q}$ matrix $M$ in terms of meson components, and we have $M$ corresponding to the matrix $\phi$ [@bramon; @palomar; @gamermann]
$$\label{eq:phimatrix}
\phi = \left(
\begin{array}{ccc}
\frac{1}{\sqrt{2}}\pi^0 + \frac{1}{\sqrt{3}}\eta + \frac{1}{\sqrt{6}}\eta' & \pi^+ & K^+ \\
\pi^- & -\frac{1}{\sqrt{2}}\pi^0 + \frac{1}{\sqrt{3}}\eta + \frac{1}{\sqrt{6}}\eta' & K^0 \\
K^- & \bar{K}^0 & -\frac{1}{\sqrt{3}}\eta + \sqrt{\frac{2}{3}}\eta' \\
\end{array}
\right)$$
This matrix corresponds to the ordinary one used in chiral perturbation theory [@Gasser:1983yg] with the addition of $\frac{1}{\sqrt{3}}diag(\eta_1,\eta_1,\eta_1)$ where $\eta_1$ is a singlet of $SU(3)$, taking into account the standard mixing between $\eta$ and $\eta'$. The term $\frac{1}{\sqrt{3}}diag(\eta_1,\eta_1,\eta_1)$ is omitted in the chiral Lagrangians because the $[\phi,\partial_{\mu}\phi]$ structure of the Lagrangians renders this term inoperative. In Ref. [@weihong] the ordinary $\phi$ matrix of chiral perturbation theory was also used. Here we consider the full $\phi$ matrix of Eq. (\[eq:phimatrix\]) since we are concerned with physical $\eta$ plus $\pi^0$ production.
Hence upon hadronization of the $u\bar{u}$ component we shall have $$\begin{aligned}
&& u\bar{u}(\bar{u}u + \bar{d}d + \bar{s}s) \equiv (\phi \cdot
\phi)_{11} = \frac{1}{2}\pi^0 \pi^0 \nonumber \\
&& + \frac{1}{3} \eta \eta + \frac{2}{\sqrt{6}} \pi^0 \eta +
\pi^+\pi^- + K^+ K^- , \label{eq:phiphi11}\end{aligned}$$ where we have omitted the $\eta'$ term because of its large mass. This means that upon hadronization of the $u\bar{u}$ component we have $D^0 \to \bar{K}^0 PP$, where $PP$ are the different meson meson components of Eq. (\[eq:phiphi11\]). This is only the first step, because now these mesons will interact among themselves delivering the desired meson pair component at the end: $\pi^+
\pi^-$ for the case of the $f_0(500)$ and $f_0(980)$, and $\pi^0
\eta$ for the case of the $a_0(980)$.
The multiple scattering of the mesons is readily taken into account as shown diagrammatically in Fig. \[Fig:mesonmesonFSI\].
Analytically we shall have $$\begin{aligned}
&& t(D^0 \to \bar{K}^0 \pi^+ \pi^-) = V_P (1 + G_{\pi^+
\pi^-}t_{\pi^+\pi^- \to \pi^+\pi^-} \nonumber \\
&& + \frac{1}{2} \frac{1}{2} G_{\pi^0\pi^0} t_{\pi^0 \pi^0 \to \pi^+
\pi^-} + \frac{1}{3} \frac{1}{2}G_{\eta \eta} t_{\eta \eta \to \pi^+
\pi^-} \nonumber \\
&& + G_{K^+K^-} t_{K^+ K^- \to \pi^+ \pi^-}), \label{eq:fzero}\end{aligned}$$ and $$\begin{aligned}
&& \!\!\!\!\!\!\!\!\! t(D^0 \to \bar{K}^0 \pi^0 \eta) = V_P
(\sqrt{\frac{2}{3}} + \nonumber \\ && \!\!\!\!\!\!\!\!\!
\sqrt{\frac{2}{3}} G_{\pi^0 \eta} t_{\pi^0 \eta \to \pi^0 \eta} +
G_{K^+K^-} t_{K^+ K^- \to \pi^0 \eta}), \label{eq:azero}\end{aligned}$$ where $V_P$ is a production vertex, containing the dynamics which is common to all the terms. $G$ is the loop function of two mesons [@npa] and $t_{ij}$ are the transition scattering matrices between pairs of pseudoscalars [@npa]. The $f_0(500)$, $f_0(980)$, and $a_0(980)$ are produced in $s$-wave where $\pi^0
\pi^0$, $\pi^+ \pi^-$ have isospin $I=0$, hence these terms do not contribute to $\pi^0 \eta$ production ($I=1$) in Eq. (\[eq:azero\]). Note that in Eq. (\[eq:fzero\]) we introduce the factor $\frac{1}{2}$ extra for the identity of the particles for $\pi^0 \pi^0$ and $\eta \eta$.
The $t$ matrix is obtained as $$\label{eq:BSeq}
t = [1-VG]^{-1} V,$$ where $V_{ij}$ are the transition potentials evaluated in Refs. [@npa; @danydan]. Explicit expressions for $I=0$ are given in Ref. [@weihong]. We have the $I=1$ case new here and we present the matrix elements below $$\begin{aligned}
&& \!\!\! \! \!\! \! \! V_{K^+ K^- \to \pi^0 \eta} \! = \frac{-
\sqrt{3}}{12f^2} \! ( 3s \!\! - \! \! \frac{8}{3}m^2_K \!\! - \!\! \frac{1}{3}m^2_{\pi} \!\! - \!\! m^2_{\eta} ), \\
&& \!\!\! \! \!\! \! \! V_{K^0\bar{K}^0 \to \pi^0 \eta} = - V_{K^+ K^- \to \pi^0 \eta} ,\\
&& \!\!\! \! \!\! \! \! V_{\pi^0 \eta \to \pi^0 \eta} = -\frac{1}{3f^2}m^2_{\pi}, \\
&& \!\!\! \! \!\! \! \! V_{K^+K^- \to K^+K^-} = -\frac{1}{2f^2} s , \\
&& \!\!\! \! \!\! \! \! V_{K^+K^- \to K^0 \bar{K}^0} = -\frac{1}{4f^2} s , \\
&& \!\!\! \! \!\! \! \! V_{K^0 \bar{K}^0 \to K^0 \bar{K}^0} =
-\frac{1}{2f^2} s ,\end{aligned}$$ with $f$ the pion decay constant, $f = 93$ MeV, and $s$ is invariant mass square of the meson-meson system.
The loop function $G$ [@npa] is regularized by means of a cut off. When the $\eta \eta$ channel is explicitly taken into account the cut off needed is smaller than in Ref. [@npa] and we follow [@weihong] where it was taken equal to $q_{\rm max} =
600$ MeV.
Finally, the mass distribution for the decay is given by [^1] $$\label{eq:dGamma}
\frac{d \Gamma}{d M_{\rm inv}}=\frac{1}{(2\pi)^3}\frac{ p_{\bar K^0}
\tilde{p}_{\pi} }{4M_{D^0}^2} \left| t_{ D^0 \to \bar K^0 \pi^+
\pi^-} \right|^2,$$ where $p_{\bar K^0}$ is the $\bar{K}^0$ momentum in the global CM frame ($D^0$ at rest) and $\tilde{p}_{\pi}$ is the pion momentum in the $\pi^+ \pi^-$ rest frame, $$\begin{aligned}
\label{eq:pJpsi}
p_{\bar{K}^0} &=& \frac{\lambda^{1/2}(M_{D^0}^2, M_{\bar K^0}^2,
M_{\rm inv}^2)}{2M_{D^0}}, \\
\tilde{p}_{\pi} &=& \frac{\lambda^{1/2}(M_{\rm inv}^2, m_{\pi}^2,
m_{\pi}^2)}{2M_{\rm inv}} ,\end{aligned}$$ and similarly for the case of the $\pi^0 \eta$ production.
Before closing this section we should mention that in a three hadron final state one must look for the interaction of three particles, for which one must in principle deal with Faddeev equations [@Faddeev:1960su]. Most of the applications of Faddeev equations are done for three baryon systems but calculations for three mesons are becoming available [@albertokan]. However, for the purpose of the present work it is instructive to follow the idea in Ref. [@dosreis] for the analogous $D^+ \to K^- \pi^+ \pi^+$ reaction. In this work two body unitarity is imposed on the two body systems and diagrams related to three body unitarity are evaluated perturbatively. They are found relevant close to threshold but fade away rapidly of higher energies. What we have done is in this line and we have unitarized the $\pi^+\pi^-$, $\pi^0 \eta$ (and coupled channels pairs) but the $\bar{K}^0$ has been left as a spectator. In principle we should also look at the interaction of $\bar{K}^0
\pi^-$ which can lead to the $\kappa$ resonance [@ramonet], yet the topology of Fig. \[Fig:mesonmesonFSI\] (a) does not favor $s$-wave interaction of $\bar{K}^0 \pi^-$. And furthermore, the $\kappa$ can also come from a different topology of the diagrams than those considered in Fig. \[Fig:mesonmesonFSI\] (a) for instance producing a $\pi^+$ meson from the $c$ quark via direct conversions of $W$ into $\pi^+$ (see section IV, Fig. \[wexchange\] (A)). This is why the $\kappa$ is better seen in the $D^+ \to K^- \pi^+ \pi^+$ reaction, as discussed in Ref. [@dosreis]. We do not consider the $\pi K$ interaction leading to the $\kappa$, with the argument that the $\kappa$, being a very broad resonance in the $\pi K$ invariant mass, only contributes a smooth background below the $\pi^+ \pi^-$, or $\pi^0
\eta$ invariant mass distribution when one looks for the $f_0(980)$ or $a_0(980)$ signals and is taken into account in experimental analysis of these two latter resonances. In this sense, the diagram of Fig. \[Fig:feyn\] chosen and the interaction that we have considered is also what corresponds to the $K^0_s [\pi^+ \pi^-]_s$, $M_2$ amplitude of Ref. [@kaminski], the one that considers the $s$-wave interaction of the pions and the $f_0(500)$ and $f_0(980)$ resonances, or the $a_0(980)$ when we consider in addition the $K^0_s [\pi^0 \eta]_s$ amplitude.
Results
=======
In Fig. \[Fig:dgamr600\], we show the results of our calculation. We have taken the cut off $q_{\rm qmax} = 600$ MeV as in Ref. [@weihong]. We superpose the two mass distributions $d\Gamma /dM_{\rm inv}$ for $\pi^+ \pi^-$ (solid line) and $\pi^0
\eta$ (dashed line). The scale is arbitrary, since it corresponds to taking $V_p = 1000$ in Eqs. (\[eq:fzero\]) and (\[eq:azero\]), but it is the same for the two distributions, which allows us to compare $f_0(980)$ with $a_0(980)$ production. As we discussed before, it is a benefit of the weak interactions that we can see simultaneously both the $I=0$ $f_0(980)$ and $I=1$ $a_0(980)$ productions in the same $D^0 \to \bar{K}^0 R$ decay.
![The $\pi^+ \pi^-$ (solid line) and $\pi^0 \eta$ (dashed line) invariant mass distributions for the $D^0 \to \bar{K}^0 \pi^+
\pi^-$ decay and $D^0 \to \bar{K}^0 \pi^0 \eta$ decay, respectively. A smooth background is plotted below the $a_0(980)$ and $f_0(980)$ peaks.[]{data-label="Fig:dgamr600"}](dgamr600.eps)
When it comes to compare with the experiment we can see that the $f_0(980)$ signal is quite narrow and it is easy to extract its contribution to the branching ratios by assuming a smooth background (shown in Fig. \[Fig:dgamr600\] by the dotted line) below the $f_0(980)$ peak as a continuation of the $f_0(500)$ broad structure at lower energies. For the case of the $\pi^0 \eta$ distribution we get a clear peak that we associate to the $a_0(980)$ resonance, remarkably similar in shape to the one found in the experiment [@Rubin:2004cq]. Yet it is obvious that not all the strength seen in Fig. \[Fig:dgamr600\] can be attributed to the $a_0(980)$ resonance. One should recall that the chiral unitary approach provides amplitudes, in this case the $\pi^0 \eta$ amplitude, but the amplitudes provide poles that one associates to resonances but also background contributions, and this is the case of the $\pi^0 \eta$ distribution. In order to get a “$a_0(980)$” contribution we subtract a smooth background that we depict by a open dotted line in the figure. By doing that we have a remaining “resonant” shape with an apparent width of $80$ MeV, which is in the middle of the $50 - 100$ MeV of the PDG [@pdg]. Integrating the area below these structures we obtain $$\begin{aligned}
R &=& \frac{\Gamma(D^0 \to \bar{K}^0 a_0(980), a_0(980) \to \pi^0
\eta)}{\Gamma(D^0 \to \bar{K}^0 f_0(980), f_0(980) \to \pi^+\pi^-)}
\nonumber \\
& = & 6.7 \pm 1.3 , \label{ratioth}\end{aligned}$$ where we have added a $20\%$ theoretical error due to uncertainties in the extraction of the background.
Experimentally we find from the PDG and the Refs. [@Muramatsu:2002jp; @Rubin:2004cq], $$\begin{aligned}
&& \Gamma(D^0 \to \bar{K}^0 a_0(980), a_0(980) \to \pi^0 \eta)
\nonumber \\
&&= (6.5 \pm 2.0) \times 10^{-3}, \\
&& \Gamma(D^0 \to \bar{K}^0 f_0(980), f_0(980) \to
\pi^+\pi^-)\nonumber \\
&&= (1.22^{+0.40}_{-0.24}) \times 10^{-3} .\end{aligned}$$
The ratio that one obtains from there is $$\begin{aligned}
R = 5.33^{+2.4}_{-1.9} . \label{ratioex}\end{aligned}$$
The agreement found between Eq. (\[ratioth\]) and Eq. (\[ratioex\]) is good, within errors. This is, hence, a prediction that we can do parameter free.
As we mentioned, the explicit consideration of the $\eta \eta$ channel in the meson meson interaction, required to use a cut off $q_{\rm max} = 600$ MeV [@weihong] to agree with experimental amplitudes, smaller than in Ref. [@npa] where this channel was omitted. We use the same cut off here. Yet, we want to show explicitly that the ratio obtained does not get spoiled even if a wide range of cut offs are used. In Fig. \[Fig:dgamr\], we show the results for five different, higher values of $q_{\rm max}$. The magnitude of the $a_0(980)$ production grows a bit with $q_{\rm
max}$, with the prescription taken above, but the strength of the $f_0(980)$ production also grows as a consequence of an increase in the width. One can also see that the peak of the $f_0(980)$ moves to lower energies, what puts constraints on $q_{\rm max}$, but we see that, even within this broad range of values of $q_{\rm max}$, the ratio of Eq. (\[ratioth\]) remains within the errors of this equation and is a solid prediction.
![(Color online) The $\pi^+ \pi^-$ (black curves) and $\pi^0
\eta$ (blue curves) invariant mass distributions with different cut off $q_{\rm max}$ for the $D^0 \to \bar{K}^0 \pi^+ \pi^-$ decay and $D^0 \to \bar{K}^0 \pi^0 \eta$ decay, respectively.[]{data-label="Fig:dgamr"}](dgamr.eps)
It should not go unnoticed that we also predict a sizeable fraction of the decay width into $D^0 \to \bar{K}^0 f_0(500)$, with a strength several times bigger than for the $f_0(980)$. The $\pi^+
\pi^-$ distributions is qualitatively similar to that obtained in Ref. [@weihong] for the $\bar{B}^0 \to J/\psi \pi^+\pi^-$ decay, although the strength of the $f_0(500)$ with respect to the $f_0(980)$ is relatively bigger in this latter decay than in the present case (almost $50\%$ bigger). The $\bar{B}^0 \to J/\psi
f_0(500),~f_0(500) \to \pi^+ \pi^-$ decay mode, together with the $f_0(980)$ one have been identified in Ref. [@Aaij:2014siy] through a partial wave analysis, and the rates obtained are comparable with the findings of Ref. [@weihong]. Such a partial wave analysis is not available from the work of Ref. [@Muramatsu:2002jp], where the analysis was done assuming a resonant state and a stable meson, including many contributions, but not the $K^0_s f_0(500)$. Yet, a discussion is done at the end of the paper [@Muramatsu:2002jp] in which the background seen is attributed to the $f_0(500)$. With this assumption they get a mass and width of the $f_0(500)$ compatible with other experiments. Further analyses in the line of [@Aaij:2014siy] would be most welcome to separate this important contributions to the $D^0 \to
K^0_s \pi^+ \pi^-$ decay.
Further considerations
======================
Our results are based on the dominance of the quark diagrams of Fig. \[Fig:feyn\]. In the weak decay of mesons the diagrams are classified in six different topologies [@Chau:1982da; @Chau:1987tk]: external emission, internal emission, $W$-exchange, $W$-annihilation, horizontal $W$-loop and vertical $W$-loop. As shown in Ref. [@Cheng:2010vk], only the internal emission graph (Fig. \[Fig:feyn\] of the present work) and $W$-exchange [^2] contribute to the $D^0 \to \bar{K}^0 f_0(980)$ and $D^0 \to \bar{K}^0a_0(980)$ decays. In Ref. [@kaminski] the $D^0 \to \bar{K}^0 \pi^+\pi^-$ decay is studied. Hence, only the $D^0 \to K^0_s f_0(980)$ decay can be addressed, which is accounted for by proper form factors and taken into account by means of the $M_2$ ($K^0_s[\pi^+\pi^-]_s$) amplitude, which contains the tree level internal emission, and $W$-exchange (also called annihilation mechanism). In order to establish connection with the work of Ref. [@kaminski], let us draw the external emission and $W$-exchange diagrams pertinent to the $D^0 \to \bar{K}^0\pi^+\pi^-$ decay, as shown in Fig. \[wexchange\].
![External emission diagram \[(A)\] and the $W$-exchange diagram \[(B)\] for $D^0 \to \bar{K}^0 \pi^+ \pi^-$ decay[]{data-label="wexchange"}](wexchange.eps)
It is also instructive to recall the basic non-leptonic Hamiltonian at the quark level responsible for this transition [@Buchalla:1995vs; @ElBennich:2009da; @Leitner:2010ai] $$\begin{aligned}
H_{\rm W} &=& \frac{G_F}{\sqrt{2}} V_{cs} V_{ud} \bar{c}\gamma_{\mu}
(1 - \gamma_5)s \bar{d}\gamma^{\mu} (1 - \gamma_5)u \nonumber
\\ && + h.c. \label{eq:hweak}\end{aligned}$$ This Hamiltonian transforms as an isospin $I=1$ operator. Consequently the decay amplitude of $D^0 \to K\pi\pi$ is $$\begin{aligned}
T(D^0 \! \to \! K\pi\pi) \! = <K^0_s M_1 M_2|H_{\rm W}|D^0>,\end{aligned}$$ where the two meson system $M_1M_2$ ($\pi^+\pi^-$ here) can have $I=0,~1,~2$. This is the case in the diagram of Fig. \[wexchange\] (A) where the $c \bar{u}$, $\pi^+$ intermediate state can have $I=1/2,~3/2$, which allows the $\pi^+\pi^-$ system to have $I=0,~1,~2$ in the final $\bar{K}^0 \pi^+\pi^-$ state. However, the diagram of Fig. \[wexchange\] (A) will not contribute to our resonance production which requires the $\pi^+\pi^-$ $S$ wave loop, as seen in Fig. \[Fig:mesonmesonFSI\], due to the vector structure of Eq. (\[eq:hweak\]) in the $csW^+$ vertex of Fig. \[wexchange\] (A). This is also the case in the phenomenological analysis of Ref. [@Cheng:2010vk]. Then, in the remaining mechanisms of Fig. \[Fig:mesonmesonFSI\] and Fig. \[wexchange\] (B) the $\pi^+\pi^-$ can only be in $I=0$ or $1$.
In our study we have isolated the $S$ wave of the pions in order to get the $f_0(500)$, $f_0(980)$ resonances, and the $a_0(980)$ in the case of $\pi^0 \eta$. Certainly, the operator of Eq. (\[eq:hweak\]) allows other angular momenta, and indeed experimentally $\rho$ meson and other mesons can be obtained, but the experimental analysis of Refs. [@Muramatsu:2002jp; @Rubin:2004cq] with partial wave analysis separate the contributions of $f_0(980)$ and $a_0(980)$ production, which allows us to compare directly with these data without the need to look into other channels. Also, although in principle the amplitudes depend on two independent Mandelstam variables as seen in Ref. [@kaminski], the fact that we do not consider the $\bar{K}^0 \pi^-$ interaction (leading to the $\kappa$), which would just provide a background in the $\pi^+\pi^-$ mass distribution for the reasons discussed at the end of section II, makes our amplitude dependent upon the invariant mass of $\pi^+\pi^-$ or $\pi^0 \eta$.
Concerning the $W$-exchange diagrams, which we have ignored in our approach, we would like to argue in favor of its relative smallness with two arguments: firstly, in Fig. \[Fig:feyn\] (A) we can see that the $\bar{u}$ quark of the $D^0$ is a spectator. We thus have a one body operator at the $D^0$ quark level. However, in the $W$-exchange one involves the two quarks of $D^0$ and the amplitude squared involves the probability to find two quarks, smaller than that of finding one quark. This situation is typical in nuclear reactions, where the $W$-exchange would have its equivalent in the exchange currents [@Gil:1997bm]. The second argument is that in the $W$-exchange diagram of Fig. \[wexchange\] (B) there is a double hadronization compared to the single hadronization of Fig. \[Fig:feyn\] (B). The hadronization reverts into a decreased rate for two meson production compared to the single meson of the original $q\bar{q}$, which we can estimate in about one order of magnitude from the experimental rate [@pdg; @LHCb:2012ae] (see Ref. [@weihong] for details), $$\begin{aligned}
\frac{\Gamma (\bar{B}^0_s \to J/\psi f_0(980); f_0(980) \to
\pi^+\pi^-)}{\Gamma (\bar{B}^0_s \to J/\psi \phi)} = 0.14 . \nonumber \\\end{aligned}$$
In the literature there is much discussions about the relevance of the $W$-exchange mechanism. In Ref. [@Cheng:2010vk] an empirical analysis is done based on giving a weight to the different topological mechanisms, and the $W$-exchange mechanism (evaluated under the assumption that the $f_0$ and $a_0$ resonances are $q\bar{q}$ or tetraquark states) appears of the same order of the internal conversion, with opposite sign, that makes the $C-E$ combination in $a_0$ production bigger than the $C+E$ combination in $f_0$ production. [^3] However, in the same paper, a factorization approach is followed (see section V of Ref. [@Cheng:2010vk]) in which the $W$-exchange contribution is claimed to be suppressed and is neglected in that approach. The present work neglects the $W$-exchange mechanism and produces a large $a_0(980)$ production relative to $f_0(980)$ due to the mechanism of final state interaction. We should note that in both cases, the intermediate production of $K\bar{K}$ states, and further rescattering to give $\pi^+\pi^-$ or $\pi^0 \eta$ in the final states, is a novelty of our approach compared to other approaches and an essential ingredient in the results due to the strong coupling of the $f_0(980)$ and $a_0(980)$ resonances to $K\bar{K}$.
The dominance of the internal emissions in this kind of processes is also supported in other works [@Stone:2013eaa; @Aaij:2014siy; @Li:2011pg; @Aaltonen:2011nk; @Abazov:2011hv]. In Ref. [@kaminski] a detailed discussion is made of results in different works. The $W$-exchange mechanism in Ref. [@kaminski] depends on two unknown form factors which are fitted to the data and a phase which is unknown. From a fit to the data, a minimal strength of about $20\%$ is obtained for the $W$-exchange mechanism, suggesting that the contribution could be bigger. It is clear that this issue is still open but the relative smallness of the $W$-exchange mechanism has many arguments in favor, and our study, producing a big ratio of $a_0(980)$ versus $f_0(980)$ production due to final state interaction in coupled channels, neglecting the $W$-exchange mechanism, provides extra support for its smallness. Note that this $a_0/f_0$ large ratio was the main reason of the relatively large weight of the $W$-exchange mechanism in the fit of Ref. [@Cheng:2010vk]. Studies along the lines of Ref. [@kaminski] for $D^0 \to \bar{K}^0 \pi^0 \eta$ would help bring extra light into this issue.
Summary and conclusions
=======================
We have studied the decay of the $D^0$ decay into $K^0_s$ and a scalar resonance, $f_0(500)$, $f_0(980)$, $a_0(980)$. For this purpose we have identified the weak mechanism that allows the formation of a $\bar K^0$, that will act as a spectator, and a pair of mesons, $K\bar{K}$, $\pi \pi$, $\pi^0 \eta$, $\eta \eta$, etc., that upon interaction will give rise to the $f_0(500)$, $f_0(980)$, $a_0(980)$ resonances. The first step is the production of a $\bar{K}^0_s$ and a pair of $q \bar q$, which upon hadronization leads to these pairs of mesons. The hadronization is done in an easy way, by looking at the flavor content in meson meson of the hadronized $q \bar q$ pair. This is sufficient in the present case where we only aim at determining the shape of the invariant mass distributions and the relative weight of the different production modes, but not absolute rates. Once the weight of the different $\bar K^0$-meson-meson components has been determined we then allow these meson-meson components to interact, using for it the chiral unitary approach, and they give rise to the $f_0(500)$, $f_0(980)$, $a_0(980)$ resonances. They are seen in the $\pi^+ \pi^-$ invariant mass distributions \[$f_0(500)$, $f_0(980)$\] and the $\pi^0 \eta$ distribution \[$a_0(980)$\], and we not only get the poles of these resonances but also realistic mass distributions that can be compared with experiment. We found the shape of the $\pi^0 \eta$ distribution rather similar to the one found in the experiment, and we obtained a ratio of the branching ratios for $a_0(980)$ and $f_0(980)$ production in good agreement with experiment, all of it accomplished without any free parameter, meaning that the parameters of the theory have been determined before hand in the study of the meson meson interaction.
We emphasized the fact that it is the nature of the weak interactions, that allows for isospin violations, what made possible the production of the $a_0(980)$ and $f_0(980)$ resonances in the same decay. This is a most welcome feature that has allowed to test simultaneously the production of the two resonances in the same reaction offering new test for the chiral unitary approach than allowed in strong interaction reactions, providing yet one more example of support for the dynamically generated nature of the low lying scalar mesons.
Acknowledgments {#acknowledgments .unnumbered}
===============
One of us, E. O., wishes to acknowledge support from the Chinese Academy of Science (CAS) in the Program of Visiting Professorship for Senior International Scientists. This work is partly supported by the Spanish Ministerio de Economia y Competitividad and European FEDER funds under the contract number FIS2011-28853-C02-01 and FIS2011-28853-C02-02, and the Generalitat Valenciana in the program Prometeo, 2009/090. We acknowledge the support of the European Community-Research Infrastructure Integrating Activity Study of Strongly Interacting Matter (acronym HadronPhysics3, Grant Agreement n. 283286) under the Seventh Framework Programme of EU. This work is also partly supported by the National Natural Science Foundation of China under Grant Nos. 11105126, 11375080, and 10975068, and the Natural Science Foundation of Liaoning Scientific Committee under Grant No 2013020091. The Project Sponsored by the Scientific Research Foundation for the Returned Overseas Chinese Scholars, State Education Ministry.
[999]{}
H. Muramatsu [*et al.*]{} \[CLEO Collaboration\], Phys. Rev. Lett. [**89**]{}, 251802 (2002); \[Erratum-ibid. [**90**]{}, 059901 (2003)\]. P. Rubin [*et al.*]{} \[CLEO Collaboration\], Phys. Rev. Lett. [**93**]{}, 111801 (2004). J. Beringer [*et al.*]{} \[Particle Data Group Collaboration\], Phys. Rev. D [**86**]{}, 010001 (2012).
J.-P. Dedonder, R. Kaminski, L. Lesniak and B. Loiseau, Phys. Rev. D [**89**]{}, 094018 (2014).
A. Poluektov [*et al.*]{} \[Belle Collaboration\], Phys. Rev. D [**81**]{}, 112002 (2010). P. del Amo Sanchez [*et al.*]{} \[BaBar Collaboration\], Phys. Rev. Lett. [**105**]{} (2010) 081803. J. Gasser and H. Leutwyler, Annals Phys. [**158**]{}, 142 (1984). V. Bernard, N. Kaiser and U. -G. Meissner, Int. J. Mod. Phys. E [**4**]{}, 193 (1995).
J. A. Oller and E. Oset, Nucl. Phys. A [**620**]{}, 438 (1997) \[Erratum-ibid. A [**652**]{}, 407 (1999)\].
J. A. Oller, E. Oset and J. R. Pelaez, Phys. Rev. D [**59**]{}, 074001 (1999) \[Erratum-ibid. D [**60**]{}, 099906 (1999)\] \[Erratum-ibid. D [**75**]{}, 099903 (2007)\]. N. Kaiser, Eur. Phys. J. A [**3**]{}, 307 (1998). M. P. Locher, V. E. Markushin and H. Q. Zheng, Eur. Phys. J. C [**4**]{}, 317 (1998). J. Nieves and E. Ruiz Arriola, Nucl. Phys. A [**679**]{}, 57 (2000).
J. R. Pelaez and G. Rios, Phys. Rev. Lett. [**97**]{}, 242002 (2006). S. Weinberg, Phys. Rev. [**137**]{}, B672 (1965).
W. H. Liang and E. Oset, Phys. Lett. B [**737**]{}, 70 (2014).
R. Aaij [*et al.*]{} \[LHCb Collaboration\], Phys. Lett. B [**698**]{}, 115 (2011). R. Aaij [*et al.*]{} \[LHCb Collaboration\], Phys. Rev. D [**87**]{}, no. 5, 052001 (2013). S. Stone and L. Zhang, Phys. Rev. Lett. [**111**]{}, no. 6, 062001 (2013).
A. Martinez Torres, L. S. Geng, L. R. Dai, B. X. Sun, E. Oset and B. S. Zou, Phys. Lett. B [**680**]{}, 310 (2009). A. Bramon, A. Grau and G. Pancheri, Phys. Lett. B [**283**]{}, 416 (1992). L. Roca, J. E. Palomar and E. Oset, Phys. Rev. D [**70**]{}, 094006 (2004). D. Gamermann, E. Oset and B. S. Zou, Eur. Phys. J. A [**41**]{}, 85 (2009). D. Gamermann, E. Oset, D. Strottman and M. J. Vicente Vacas, Phys. Rev. D [**76**]{}, 074016 (2007).
L. D. Faddeev, Sov. Phys. JETP [**12**]{}, 1014 (1961) \[Zh. Eksp. Teor. Fiz. [**39**]{}, 1459 (1960)\]. A. Martinez Torres, K. P. Khemchandani, L. S. Geng, M. Napsuciale and E. Oset, Phys. Rev. D [**78**]{}, 074031 (2008). P. C. Magalhaes, M. R. Robilotta, K. S. F. F. Guimaraes, T. Frederico, W. de Paula, I. Bediaga, A. C. d. Reis and C. M. Maekawa [*et al.*]{}, Phys. Rev. D [**84**]{}, 094001 (2011). R. Aaij [*et al.*]{} \[LHCb Collaboration\], Phys. Rev. D [**90**]{}, 012003 (2014). L. L. Chau, Phys. Rept. [**95**]{}, 1 (1983). L. L. Chau and H. Y. Cheng, Phys. Rev. D [**36**]{}, 137 (1987). H. Y. Cheng and C. W. Chiang, Phys. Rev. D [**81**]{}, 074031 (2010). G. Buchalla, A. J. Buras and M. E. Lautenbacher, Rev. Mod. Phys. [**68**]{}, 1125 (1996). B. El-Bennich, A. Furman, R. Kaminski, L. Lesniak, B. Loiseau and B. Moussallam, Phys. Rev. D [**79**]{}, 094005 (2009); \[Erratum-ibid. D [**83**]{}, 039903 (2011)\]. O. Leitner, J.-P. Dedonder, B. Loiseau and R. Kaminski, Phys. Rev. D [**81**]{}, 094033 (2010); \[Erratum-ibid. D [**82**]{}, 119906 (2010)\]. A. Gil, J. Nieves and E. Oset, Nucl. Phys. A [**627**]{}, 543 (1997). R. Aaij [*et al.*]{} \[LHCb Collaboration\], Phys. Rev. D [**86**]{}, 052006 (2012). J. Li [*et al.*]{} \[Belle Collaboration\], Phys. Rev. Lett. [**106**]{}, 121802 (2011). T. Aaltonen [*et al.*]{} \[CDF Collaboration\], Phys. Rev. D [**84**]{}, 052012 (2011). V. M. Abazov [*et al.*]{} \[D0 Collaboration\], Phys. Rev. D [**85**]{}, 011103 (2012).
[^1]: The decay amplitude $t_{\bar D^0 \to \bar K^0 \pi^+
\pi^-}$ depends on the invariant mass, $M_{\rm inv} = \sqrt{s}$, of the meson-meson system.
[^2]: The $W$-exchange and $W$-annihilation are often referred together as weak annihilation diagrams.
[^3]: The $C$ and $E$ are the contributions of the internal conversion and $W$-exchange, and $C-E$ and $C+E$ the combinations found in Ref. [@Cheng:2010vk] for $a_0$ and $f_0$ production, respectively.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This paper proposes a design for a hybrid, city-wide urban navigation system for moving agents demanding dedicated assistance. The hybrid system combines GPS and vehicle-to-vehicle communication from an ad-hoc network of parked cars, and RFID from fixed infrastructure –such as smart traffic lights– to enable a safely navigable city. Applications for such a system include high-speed drone navigation and directing visually impaired pedestrians.'
author:
- 'Oisín Moran\*, Robert Gilmore\*, Rodrigo Ordóñez-Hurtado and Robert Shorten [^1][^2] [^3]'
bibliography:
- 'References.bib'
title: Hybrid Urban Navigation for Smart Cities
---
Introduction
============
Driven by increasing urban populations, ageing citizens, and an ever increasing need to make better use of existing resources, the smart city agenda has become one of the most defining research paradigms of recent times. At the heart of this research agenda is the need for services that require precise localisation and rapid navigation. Examples of such services include drones to assist emergency response units, and providing assistance to citizens with special needs. Thus, it is very likely that locating, controlling and feedback will be key enablers for a plethora of future city services.
Despite the obvious need for such services, their realisation is not without significant challenges. Many services that we wish to develop require precise and rapid navigation (drones to support emergency response) but should function even in the partial absence of power or sensor coverage, and at the same time be non-invasive and respect the privacy concerns of modern society.
Current navigation systems rely heavily on Global Navigation Satellite Systems (GNSSs) such as GPS and GLONASS, despite the fact that in urban environments GNSSs’ signals are in general inaccurate. A consequence of such inaccuracies is increased power consumption (drifting around desired paths or waypoints will result in longer times to successfully complete a journey). As a result, drone navigation in urban scenarios tends to be vision-based, which is seen by many as a major privacy concern. Thus the objective of this paper is to propose a viable alternative to vision based systems that addresses some of these drawbacks.
Combining multiple types of sensing could provide the accuracy needed for dedicated autonomous navigation in cities, this paper focuses on creating such a system. As it would be virtually impossible (or at least impractical) to instrument an entire city with battery-powered devices using traditional infrastructure, we propose merging conventional and non-conventional approaches to create a system that is a hybrid of high and low power sensing devices, and a hybrid of high and low precision positioning technologies. We argue that this type of system is the key to enabling large-scale robust, accurate, and rapid localisation of moving agents within a city. For this, we propose to use an Ultra Wide Band (UWB) based system for the dense coverage of high power sensing, and propose to use a passive (cheap and zero power on the agent side) Radio Frequency Identification (RFID) based system for the sparser coverage of low power [*“last-mile”*]{} sensing.
The proposed hybrid system of wireless devices could be created by installing elements in the current city environment by either taking advantage of current infrastructure or by adding new sensors to the fleet of vehicles involved in ad-hoc networks. The latter approach, following [@Ordonez-Hurtado2017] also enables the possibility of alternative monetization platforms for connected vehicles, with a car owner potentially receiving micro-payments to participate in the network. Providing such a network of collaborative technologies offers a broader coverage of an urban environment. Such a dense and expansive network provides the possibility of creating accurate, real-time sensory systems using multiple technologies to improve localisation, which results in the creation of new, safe and robust navigation opportunities. Using passive RFID tags, for example, has the benefit of ensuring an agent is not restricted by the battery life of a sensor. This in effect produces a “Zero Power" consuming sensor as the reflected radio waves in well-designed applications are enough to give adequate localisation in the event of a loss of power, representing a great advantage over traditional applications. In addition to this, this paper also investigates the viability of using a new style of static but flexible infrastructure, namely ad-hoc networks of parked cars, and how the potential services they provide may be able to support a multi-technology sensory system for localisation and navigation.
In summary, we aim to make possible rapid urban navigation for specialised moving agents, such as drones for emergency response and priority delivery or fully autonomous vehicles, by the adoption of a hybrid tracking/localisation system which enables wider coverage than conventional systems and a contingency method in cases of power shortages on the agent side. The efficacy of the proposed system is demonstrated by means of two important use-cases: (i) rapid drone navigation in an urban environment; and (ii) the development of a smart traffic light system to assist citizens with special needs.
Enabling Technologies
=====================
This paper investigates a number of technologies that could collect the data required to localise and track moving agents for control purposes. The infrastructure in which such technologies could be implemented, and the motivation for such a system, derive in part from the idea of a fluid infrastructure – a system that can provide complete coverage from permanently fixed and temporarily stationary network nodes (road side units and parked cars respectively). This system would provide the opportunity to instrument a city and localise an agent at virtually every point of an agent’s traversal through an urban environment.
The road side units could be adorned with RFID technology to allow “Zero Power” sensing in critical locations, while UWB technology could be implemented as part of a new fluid infrastructure utilising parked cars. The system and technologies are described in depth in the following section.
Parked Cars as Infrastructure
-----------------------------
Cars are parked for up to 95% of their lifespan [@sommer2014] and practically serve no purpose in such a state other than the economic drain of occupying valuable space. However, as an underutilised network of dense sensors, these otherwise idle cars lend themselves to new, creative and useful applications. For example they have been proposed for a wide variety of dedicated tasks from finding missing keys to detecting gas leaks [@Zhuk2016]. Additionally, with an increasing number of sensors, greater connectivity, and a move toward the car becoming a service platform, these potential use cases will only increase.
Traditional infrastructure is generally expensive to set up and also to repair and maintain. Using cars as infrastructure effectively eliminates these costs, as the fleet is gradually refreshed. This system also has the extra benefit of upgrading itself over time in contrast to the stagnation or degradation associated with traditional infrastructure. The large number of cars in cities and their spatiotemporal stability (staying in the same place for a long time) will also prove to be of great benefit. The implementation in question here is the integration of UWB antenna nodes within vehicles to act as data collectors and processing units for accurate position estimation purposes.
Ultra Wide Band Technology
--------------------------
With the recent explosion in the use of unmanned aerial vehicles (UAV), we have witnessed the legal landscape change (often not rapidly enough) in response to the real and perceived threats of drones. With vision-based navigation approaches proving so popular, the potential for the erosion of privacy becomes a growing and valid concern, with the additional downside that the performance of vision-based approaches rapidly deteriorates under poor weather conditions. For applications where the end goal is not necessarily itself a visual task (such as delivering a package), and where privacy issues may be of great concern, it would be helpful to have a precise and rapid approach to navigation that is not dependent on the availability of a camera or other vision sensors. Even in cases requiring visual feedback, such as a fire station wishing to ascertain the existence or severity of a fire, having a system capable of navigating to the intended location without the need for a trained operator would save precious time for the firefighters who would now only need to be present for the final stage.
Currently, the closest option available to a precise and rapid non-vision-based approach would be the use of GPS which is not without its own technical drawbacks. A number of factors such as satellite position, signal attenuation, and clocking errors cause typical GPS applications to experience positioning errors of up to 30m on average, while the advanced version DGPS still yields an average error of 5m to 10m in a variety of environments [@kuo2013]. These positioning errors, specially those from conventional GPS, cannot be neglected in comparison with the average dimension of a drone (tens of centimeters). As with any traditional GNSS system, the accuracy of GPS can also be greatly affected by atmospheric/environmental conditions, shadowing by buildings, and multipath propagation which is especially problematic in dense urban environments [@Ordonez-Hurtado2014; @Ordonez-Hurtado2017]. While there exist different demonstrations of GPS-only (i.e. non-cooperative) approaches for drone navigation, such as [@santana2015], these are all invariably done in favourable conditions such as in open fields with no tall buildings. As any inaccuracies in position estimation convert directly to speed reductions for waypoint tracking tasks, any increase in accuracy is highly desirable for time-critical applications. For emergency applications such as the defibrillator drone designed by Alec Momont [@dronesforgood], these time savings could make all the difference for a victim of an out-of-hospital cardiac arrest where every minute that passes without treatment (CPR, defibrillatory shock, or definitive care) decreases his or her survival rate by around 5.5% [@larsen1993].
Our proposed system will thus attempt to solve both challenging issues: navigation speed and privacy. It will do so by using stationary cars as anchor (reference) nodes in a cooperative positioning approach. The cars will average out measurements of their GPS location, thereby reducing the error in their estimation of their location, and will only be added to the network after they have been stationary for a sufficient period of time (decided by a suitable performance index). Involved cars and the drone are required to be equipped with UWB-based systems to accurately estimate the distances between them and cooperatively localize the drone. In this paper we use UWB as a means to emulate current V2V communication systems based on Dedicated Short Range Communications (DSRC), but a UWB module could be easily added to the growing sensory array found in cars. Thus, as there is no imaging equipment involved in this approach, then no privacy concerns arise.
However, UWB-based systems depend on the availability of local power, so there is a gap in the system when power shortages occur in the host agent (i.e. UWB-based system is turned off due to the risk of a low battery level). In addition, as the density of cars in a city varies over space and time, there also exists a possibility that the network of parked cars may not always be sufficient to support such a localisation system. From a business district which may become car-sparse at night, to a pedestrian crossing where there may be no parking spaces nearby, holes may arise in the otherwise connected system. A system without a fail-safe mechanism to allow, for example, a visually impaired person to cross a pedestrian crossing under any circumstance would be neither robust nor safe. We thus propose that a viable method to fill these gaps and thereby make the system more robust, must rely on “Zero Power” technology such as that found in RFID systems.
RFID Technology
---------------
RFID systems are now an ever present part of modern life (generally hidden from sight), with applications that range from collecting tolls and paying bills to gaining access to buildings [@RFID_History]. In general, the traditional way of using RFID system is purely for identification purposes; however, by creating solutions that make use of different parameters, we find they are capable of much more. One such parameter that we will investigate in particular is the Received Signal Strength Indicator (RSSI) of a signal.
For the case of urban navigation, we want to be able to distinguish the RSSI values from different tags to estimate the distances to them, and then to work towards the goal of implementing control systems based on that information. Despite the fact that the RSSI readings are generally noisy, some tactics can be employed to improve/augment the quality of the collected signals, and thus ensure that the information found is trustworthy and distinguishable enough to create a robustly controlled feedback system. Some relevant aspects when using RFID systems are the technical issues when working under Non Line of Sight (NLOS) scenarios, this is, when there are obstructions between the tag and the receiver, either a varying or stationary obstruction. This paper does not investigate how to alleviate the concern of moving obstacles such as other road users (e.g., cars or people); however, there is already other research in this area (which can be further incorporated into our proposed approach). This paper focuses on applications in which RFID technology will not be affected by NLOS environments. In general, the environment will have a great impact on the measured RSSI values, however, with a suitable initial setup this issue can be alleviated.
Deploying RFID antennas capable of producing RF signals with sufficient power to produce identifiable backscattered signals at critical locations (e.g., traffic lights, uncontrolled intersections) is where this technology most clearly shows its benefits. At these points, both RFID and UWB could be used collaboratively to provide more accurate estimations of an agent’s location. However, significantly at these critical locations, RFID could be used alone to determine the location and orientation of an agent in the event of a contingency related to power availability on the agent’s side. Passive RFID tags allow for “Zero Power" sensing on the agent’s side, which is of huge potential benefit as it alleviates the dangers of loss of power in areas where safety is of utmost concern (pedestrian direction along a pedestrian crossing) or at the end point of a journey (drone navigation).
Proposed Implementation
=======================
Using both of the above technologies, namely UWB and RFID, integrated with parked cars as infrastructure, brings the potential to virtually cover a city with enough units/sensors to make it fully navigable using a cooperative approach, and with a contingency strategy for critical scenarios. A critical location is, in general, a location at which agents are most at risk, and where the risk of losing signal or battery power would have the greatest effect. A pedestrian crossing is a good example of such a location. Crossing, already a dangerous activity, has been made even more dangerous with the advent of “quiet cars" as more electric and hybrid cars take to the roads. Thus, reading the user’s location using both UWB and RFID technologies in conjunction could direct a user safely across a road.
![Potential infrastructure for urban navigation: pedestrians (moving agents), UWB-based systems (blue circles marked with A), RFID antennas (blue circles marked with B), and critical location (pedestrian crossing).[]{data-label="fig:InfrastructureDiagram"}](Fig1_InfrastructureDiagram.jpg)
Figure \[fig:InfrastructureDiagram\] illustrates a use-case scenario in which a visually impaired road user (a blind pedestrian) is directed using the UWB-based system for the largest section of their journey and then using both RFID and UWB, or RFID alone in critical zones. In the diagram, the cars are stationary (i.e. parked) and house UWB transceivers while the traffic light system would house the RFID antenna, thus adding functionality to existing infrastructure. In the case of battery power loss, a visually impaired road user will be safe when on the pavement and can continue their journey with confidence as they can re-orient themselves with reference to a known point – the edge of the pavement. However, at the critical and most dangerous point (the road crossing), no such reference is available. This is where the RFID system could provide feedback to the road user without need for any battery power on the agent’s side. Figure \[fig:DroneInfra\] visualises a system that could be of interest to the implementation of package delivery using drones. Many companies are investigating the feasibility of drone delivery systems and the approaches discussed in this paper may be able to accurately control and position drones for this use. Again, the UWB positioning would be used for large-scale movement tracking, while RFID and UWB in conjunction could be used for more accurate positioning and orientation for docking, to ensure safe and timely delivery of products. Again, parked cars can be used as UWB anchors and RFID antennas could be installed like a traditional letterbox. RFID is useful in this use-case application for its potential even to control the orientation of the drone. Thus, in the case of a battery level at potentially critical level, then the UWB-based system could be turned off and the “Zero Power" RFID-based sensing system would be hugely beneficial to complete an agent’s journey successfully.
![Potential infrastructure for drone delivery system: drone (moving agent), UWB-based systems (blue circles marked with A), and RFID antennas (blue circles marked with B).[]{data-label="fig:DroneInfra"}](Fig2_DroneInfra.png)
Experimental Results
====================
Cooperative V2V navigation.
---------------------------
Consistent indoor waypoint tracking has been successfully achieved using UWB-based Pozyx[^4] positioning system.The drone was able to navigate to within a 25cm radius of each waypoint in a list, as demonstrated in Figure \[fig:drone\_flight\] and Movie A in Section \[Demonstrations\].
![Position of drone during indoor waypoint navigation task as estimated by the Pozyx system. Position of drone is shown in blue and waypoints of radius 25cm are shown in grey.[]{data-label="fig:drone_flight"}](Fig3_Drone_flight.pdf){width="\linewidth"}
Concerning tests in an urban-like scenario, we first established a GPS-only system as a benchmark for comparison reasons with respect to our cooperative proposed approach. The used location (periphery of UCD’s Science Centre) was chosen to be representative of an urban environment (see Figure \[fig:GPSreadings\]) and as such was very close to tall buildings. As the GPS readings at this location (solid line in Figure \[fig:GPSreadings\])) contain positioning errors of high magnitude, the GPS-only system was unworkable in such an scenario, which provides a stronger justification to the use of our approach. So far, in the steps toward a full proof of concept, we used the whole set of collected GPS data (4 hours of latitude, longitude, and HDOP readings, collected every 10s) to obtain an improved position estimation by means of a weighted average described in [@Ordonez-Hurtado2017]. Thus, every sample $i$ is weighted by the factor $\lambda_i$ given by $$\lambda_i = \frac{HDOP_i^{-2}}{\sum_{j=1}^N HDOP_j^{-2}}$$ to finally obtain the average value represented with a blue circle in Figure \[fig:GPSreadings\], which is within a proximity of 2m from the real location (yellow star). With these results, our next step will be the validation of the waypoint tracking in the outdoor scenario using the hybrid (GPS+UWB) positioning system, where the relative positioning can be obtained with high accuracy using the UWB anchors, and then the absolute position is estimated by merging the information from the weighted average of GPS positions. Such a validation process is currently being prepared.
![GPS data (solid line) recorded every 10s over 4 hours, versus weighted average (blue circle) and true location (yellow star). Black areas correspond to tall buildings (UCD’s Science Centre, Belfield, Dublin 4, Ireland).[]{data-label="fig:GPSreadings"}](Fig4_GPSreadings.png){width="\linewidth"}
RFID localisation using RSSI measurements
-----------------------------------------
Characterising the relationship between RSSI readings and distance/orientation of moving agents carrying RFID tags is of utmost importance; this is to provide evidence that the backscattered RF signals are reliable and accurate enough to be used in navigational control systems. As mentioned previously, there are some limitations in the use of RFID technology, one of which being that the environment in which it is operated greatly impacts the readings obtained from tags. We have investigated this experimentally with a Parrot AR.Drone 2.0 covered by few tags, and the resulting curves displayed together in Figure \[Curve3\]. The differences in the returned RSSI values can be taken into account before making a decision in any control system, or the system could be constructed in a known location in which the environment is suitable. Something to note is that applications in a smart city context will be more likely to involve outdoors environment where backscattering and multipath effects of RF waves will not be of much concern. The locations of such systems can be chosen to minimize dynamic changes in the environment, such as elevated locations including the top of traffic lights and lamp posts.
Figure \[Curve3\] shows the standard average over 20s of RSSI readings per location, resulting in over 100 readings per location. This test was undertaken in multiple environments (enclosed/open) and operation modes (flying/held) to gain insight on how the location of the controllable agent would affect the returned RSSI values. For the enclosed laboratory, it is seen that for the first 2m, the curve can be characterised as a straight line, whereas at further distances it would be more difficult to determine where the tag is located. There is a one-to-many relationship caused by multipath effects in such an environment. This would not be the case in an outdoor environment as proven by the straight line relationship seen in the more open and unobstructed laboratory. Furthermore, it is useful to note that the RSSI values are of larger magnitude in the open lab, which is likely due to the reduction in multipath reflected return signals that would reduce the average value over the experimental test period.
### Distance Estimation
The results displayed in Figure \[Curve2\] are for distance estimation of a drone carrying multiple tags with respect to two RFID antennas. The antennas were situated at either end of a straight line, with the tags on the drone orientated perpendicularly with respect to the plane of the surface of both antennas. It should be noted here that despite the fact that maximum/minimum readings have a large variation the inter-quartile ranges are relatively small and do not overlap. In this experiment the RFID tags were interrogated for 10s per data point which equated to no less than 80 returned RSSI values.
Now that we can characterise the relationship between RSSI and distance, a Java script was developed to record the RFID-based position estimations. The information shown in Table \[ErrorTable\] reveals that the system is accurate to within 13cm, which is relevant enough for the dimensions of the test drone (51.7cm x 51.7cm).
[|>c |c|c|c|c|c|]{} True Dist (m) & 0.5 & 1 & 1.5 & 2 & 2.5\
Avg Recorded Dist (m) & 0.6023 & 1.0449 & 1.3011 & 1.4738 & 2.1338\
Error (m) & -0.1023 & -0.0449 & -0.0511 & 0.0262 & -0.1338\
S.Deviation (m) & 0.0489 & 0.0503 & 0.0422 & 0.0212 & 0.0622\
\[ErrorTable\]
Figure \[Path1\] displays the information recorded during a straight line flight test, in which a simple control system is implemented. A destination point was set and the drone placed by the user at a point away from the destination. In this case, the starting point was 0.8m from the reference antenna, and the destination was 1.25m from the reference antenna. The RFID system collected the RSSI values and converted them to a distance; this estimation using RFID is shown in Figure \[Path1\]. For comparison, the approximate ground truth was measured by the Pozyx system. A link to a video of this in operation can be found in Section \[Demonstrations\].
### Angle Estimation
For many applications, the orientation of an agent is useful or even necessary. Investigations into how the comparison of multiple tags positioned (at known locations) on one agent has been undertaken here. Emphasis is placed on analysing the differences in returning RF signals, that when processed could track orientation in real-time. The results prove that not only distance can be accurately measured using RFID, but a relationship can be formed that converts the differences in RSSI to corresponding angles. An example of this data collected from 2 different tags on one side of the drone is shown in Figure \[anglegraph2\], in which an roughly linear relationship is observed. Furthermore, Table II shows that the error between the true angle and the recorded angle using RFID technology is on average less than 1.1 degrees over a time period of 5s.
When implemented in a control system, the drone could be rotated to any orientation within the range: $\pm$ 45 degrees. However, it was noted that the accuracy of the system seems to improve the closer the drone’s orientation was to zero degrees (i.e. perpendicular to the antenna). Figure \[anglepath\] displays the orientation as recorded by the RFID reader and by the drone’s IMU. A link to a video of this in operation can be found in Section \[Demonstrations\].
[| >c |c|c|c|c|c|]{} True Angle ($^{\circ}$) & 0 & -30 & 30 & -15 & 15\
Avg Angle Found ($^{\circ}$) & 0.6570 & -30.2382 & 30.7499 & -16.0704 & 15.6625\
Error ($^{\circ}$) & -0.6570 & 0.2382 & -0.7499 & 1.0704 & -0.6625\
S. Deviation ($^{\circ}$) & 1.1135 & 1.2762 & 1.0035 & 0.8030 & 0.6567\
\[angletable\]
### Control Systems and Applications
With the information collected and proof that there is enough accuracy to recognise both distance and angle, it is clear that passive RFID tags can be used as zero power sensors in highly critical areas, alleviating fears of loss of power. From this, ideas of personal and airborne navigation through adding tags to moving agents are both feasible and accurate. A system has already been developed using the results found in this paper, in which the tags alone can distinguish at which angle and position an agent is located, and offer navigation assistance to direct that agent. As discussed above, we imagine that this could be used to aid road crossing with auditory or vibratory feedback given through a user’s smartphone or the traffic light system itself. We further imagine that the sequence of traffic lights could be altered to ensure safety of the pedestrian.
Using the difference in the RSSI values, it has been found that an agent can be directed to a centre line between 2 antennas, and from there the agent can be accurately maneuvered along such an straight line, as illustrated in Figure \[TagSetup\]. Furthermore, using RSSI to estimate the distance of an agent from an antenna accurately in a straight line means that an agent can be controlled and moved/directed to a set location. This has been achieved in airborne tests using a Parrot AR Drone 2.0 and AutoFlight [^5] drone control software.
![Example of RFID setup for feedback control of a moving agent.[]{data-label="TagSetup"}](Fig10_Tagsetup.png)
Demonstrations {#Demonstrations}
==============
We have prepared a set of movies to demonstrate the key salient points of our results.
[**Movie A (Indoor Waypoint Navigation):**]{} Video demonstrating indoor waypoint tracking to a tolerance of 25cm. Position is estimated by using 4 Pozyx anchors (UWB) to localize one tag located on the drone.
Link: `https://youtu.be/pnRgJtCT_Uw`
[**Movie B (Location Control of Drone using RFID):**]{} Video demonstrating control of drone’s location in one dimension to a tolerance of 10cm, using RSSI readings from two tags, from two antennas.
Link: `https://youtu.be/cm7VHIjtH_U`
[**Movie C (Orientation Control of Drone using RFID):**]{} Video demonstrating control of drone’s orientation to a tolerance of 3 degrees, using RSSI readings from two tags, from one antenna.
Link: `https://youtu.be/bQjGTuhpIBQ`
Conclusions
===========
This paper explores the potential to use a number of technologies in a new and complementary way, to redefine how we look at possible infrastructures and implementations to provide city-wide navigation and control. There is huge scope for extension to this work, in refinement and in developing practical working systems that could be put into operation in minimal time. Using both UWB and RFID technologies, allows for a robust and reliable system that reduces the risk in the case of low battery or power shortages situations, maintaining safety at all times. Overall the work completed here has great potential for improving the daily lives of commuters and residents through the implementation of smart services within the city of the future.
[^1]: This work was partially supported by Science Foundation Ireland grant 11/PI/1177.
[^2]: \*Joint first authors.
[^3]: O. Moran, R. Gilmore, R. Ordóñez-Hurtado and R. Shorten are with University College Dublin, School of Electrical, Electronic and Communications Engineering, Belfield, Dublin 4, Ireland. Emails: [email protected], [email protected], [email protected], [email protected].
[^4]: [https://www.pozyx.io/]{}
[^5]: [https://electronics.kitchen/autoflight]{}
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'N. Lodieu [^1], F. Allard , C. Rodrigo , Y. Pavlenko , A. Burgasser , Y. Lyubchik , B. Kaminsky , D. Homeier'
bibliography:
- '../../AA/mnemonic.bib'
- '../../AA/biblio\_old.bib'
date: 'Received ; accepted '
title: 'Metallicity, temperature, and gravity scales of M subdwarfs'
---
[The aim of the project is to define a metallicity/gravity/temperature scale vs spectral types for metal-poor M dwarfs.]{} [We obtained intermediate-resolution ultraviolet (R$\sim$3300), optical (R$\sim$5400), and near-infrared (R$\sim$3900) spectra of 43 M subdwarfs (sdM), extreme subdwarfs (esdM), and ultra-subdwarfs (usdM) with the X-shooter spectrograph on the European Southern Observatory Very Large Telescope. We compared our atlas of spectra to the latest BT-Settl synthetic spectral energy distribution over a wide range of metallicities, gravities, and effective temperatures to infer the physical properties for the whole M dwarf sequence (M0–M9.5) at sub-solar metallicities and constrain the latest state-of-the-art atmospheric models. ]{} [The BT-Settl models reproduce well the observed spectra across the 450–2500 nm wavelength range except for a few regions. We find that the best fits are obtained for gravities of $\log$(g)=5.0–5.5 for the three metal classes. We infer metallicities of \[Fe/H\]=$-$0.5, $-$1.5, and $-$2.0$\pm$0.5 dex and effective temperatures of 3700–2600K, 3800–2900K, and 3700–2900K for subdwarfs, extreme subdwarfs, and ultra-subdwarfs, respectively. Metal-poor M dwarfs tend to be warmer by about 200$\pm$100K and exhibit higher gravity than their solar-metallicity counterparts. We derive abundances of several elements (Fe, Na, K, Ca, Ti) for our sample but cannot describe their atmospheres with a single metallicity parameter. Our metallicity scale expands the current scales available for midly metal-poor planet-host low-mass stars. Our compendium of moderate-resolution spectra covering the 0.45–2.5 micron range represents an important legacy value for large-scale surveys and space missions to come. ]{}
Introduction {#SpClass_sdM:intro}
============
Mass, gravity, temperature, and metallicity constitute key parameters to understand the formation and evolution of any type of stars. M dwarfs represent the largest community among members of the solar neighbourhood [@henry06; @bochanski10; @kirkpatrick12] and have also become attractive targets to search for Earth-like planets [e.g. @nutzman08; @quirrenbach12a; @sozzetti14]. While the metal content of solar-type stars can be measured with great accuracies [e.g. @adibekyan12a], the metallicity of M dwarfs is more difficult to ascertain due to the significant amount of absorption bands [@kirkpatrick99]. Several groups conducted complementary surveys to assess the metallicity of M dwarfs based on photometry and colours [@bonfils05a; @johnson09a; @schlaufman10a; @neves12; @hejazi15; @dittmann16] and high-resolution spectra or spectral synthesis of multiple systems composed of a solar-type primary and a M dwarf secondary at optical and infrared wavelengths [@woolf05; @woolf06; @bean06a; @woolf09; @rojas_ayala10; @rojas_ayala12; @muirhead12; @terrien12a; @onehag12; @neves13; @neves14; @hejazi15; @newton15a; @lindgren16a]. These surveys usually focus on slightly metal-poor M dwarfs, with metallicities above $-$1.0 dex. An extension to lower metallicities is needed to complement the precise distances that the Gaia satellite [@deBruijne12] will provide for a great number of spectroscopically-confirmed thick disk and halo M dwarfs [@savcheva14] but also understand the role of metallicity on the molecules and dust grains present in the atmosphere of low mass stars.
Subdwarfs are population II dwarfs which appear bluer than solar-metallicity stars due to the dearth of metals in their atmospheres [@baraffe97]. They exhibit thick disk or halo kinematics, including high proper motions or large heliocentric velocities [@gizis97a]. They belong to the first generations of stars and are important tracers of the chemical enrichment history of the Galaxy. The original classification for M subdwarfs (sdM) and extreme subdwarfs (esdM) developed by @gizis97a has been revised and extended by @lepine07c. A new class of subdwarfs, the ultra-subdwarfs (usdM), has been added to the sdM and esdM. The new scheme is based on a parameter, $\zeta_{\rm TiO/CaH}$, which quantifies the weakening of the strength of the TiO band (in the optical) as a function of metallicity. Subdwarfs are easily distinguished from dwarfs with solar abundances because they exhibit stronger metal-hydride absorption bands (FeH, CrH) and metal lines (Ca, Fe) as well as blue infrared colours caused by collision-induced H$_{2}$–H$_{2}$ absorption [@gizis97a; @lepine07c; @lodieu17a]. An independent classification scheme based on metallicity, gravity, and temperature has been proposed for M dwarfs by @jao08 and an extension to the L dwarf regime proposed by @kirkpatrick14 and @zhang17a.
In this paper, we present moderate-resolution 0.45–2.5 $\mu$m spectroscopy of 15 sdM, 16 esdM, and 12 usdM to infer their metallicities and T$_{\rm eff}$ by direct comparison with the latest BT-Settl synthetic models [@allard12]. In Section \[MplusTeff\_sdM:sample\] we present our sample of metal-poor M dwarfs drawn from the literature. In Section \[MplusTeff\_sdM:spec\_obs\] we describe our spectroscopic observations. In Section \[MplusTeff\_sdM:Models\] we introduce the BT-Settl models used to infer the physical parameters of metal-poor M dwarfs. In Section \[MplusTeff\_sdM:Comparison\] we infer the spectral type vs.metallicity/gravity/T$_{\rm eff}$ relation of the three M dwarf metal classes and compare it to independent but complementary studies as well as solar-type M dwarfs. In Section \[MplusTeff\_sdM:discussion\] we discuss our results and the peculiarities of some of the spectra.
Sample selection {#MplusTeff_sdM:sample}
================
To select our sample of subdwarfs, we used the SDSS spectroscopic database which contains a wealth of high-quality spectra covering the 5000–9200Å range at a spectral resolution of $\sim$2000. Hundreds of objects have been classified as sdM, esdM, or usdM following the scheme developed by @lepine07c. This classification is publicly available through the SDSS archive and we have taken advantage of it to retrieve a large sample of low-metallicity M dwarfs.
We selected a sub-sample of low-metallicity stars which represent a sequence going from M0 to M9.5 from optical spectra. The objects were specifically chosen to be the (or among the) brightest of their subclass and be observable either in August or January from the Southern hemisphere to achieve the best signal-to-noise ratio possible over a wide wavelength range. We completed our sample with LHS377 [sdM7; @gizis97a], observed with the same telescope/instrument but at higher spectral resolution [@rajpurohit16a]. Our final sample contains 15 sdM0–sdM9.5, 16 esdM0.0–esdM8.5, and 12 usdM0.0–usdM8.5) with almost one object per spectral sub-type. We are missing the sdM7.5, sdM8, and sdM9 in the subdwarf sequence, the esdM2.5 and esdM8 in the extreme subdwarf sequence, and more usdM (usdM1.5, usdM2, usdM3.0, usdM6.5, usdM7, and usdM8). We note that these targets were included in our original sample but not observed during the ESO service runs. Table \[tab\_MplusTeff\_sdM:log\_obs\_Xshooter\] lists their coordinates, optical SDSS$i$ magnitudes, optical spectral types on the @lepine07c scheme, dates of observations, mean airmass, exposure times in all three arms, and numbers of AB cycles for all 43 subdwarfs.
VLT/X-shooter spectroscopy {#MplusTeff_sdM:spec_obs}
==========================
We carried out spectroscopy from the UV- to the $K$ band with the X-Shooter spectrograph [@dOdorico06; @vernet11] mounted on the Cassegrain focus of the Very Large Telescope (VLT) Unit 2. Observations were conducted in service mode by the European Southern Observatory (ESO) staff over the course of three semesters, between August 2012 and June 2014 (Table \[tab\_MplusTeff\_sdM:log\_obs\_Xshooter\]). The conditions at the time of the observations met the clear sky request, with an airmass less than 1.6, grey conditions, and a seeing better than 1.2 arcsec. We set the individual on-source integration times according to the magnitudes of the targets and used the multiple AB patterns to correct for the sky contribution (mainly) in the near-infrared. All observations were done with the slit oriented at parallactic angle. We list all the M subdwarfs observed with X-shooter in Table \[tab\_MplusTeff\_sdM:log\_obs\_Xshooter\] along with a summary of the logs of the observations.
X-Shooter is a multi wavelength cross–dispersed echelle spectrograph made of three arms covering simultaneously the ultraviolet (UVB; 0.3–0.55 $\mu$m), visible (VIS; 0.55–1.0 $\mu$m), and near–infrared (NIR; 1.0–2.48 $\mu$m) wavelength ranges thanks to the presence of two dichroics splitting the light. The spectrograph is equipped with three detectors: a 4096$\times$2048 E2V CCD44-82, a 4096$\times$2048 MIT/LL CCID20, and a 2096$\times$2096 Hawaii 2RG for the UVB, VIS, and NIR arms, respectively. We set the read-out mode to 400k and low gain without binning. We used the 1.6 arcsec slit in the UVB, 1.5 arcsec in the VIS, and 1.2 arcsec NIR, yielding resolving powers of 3300 (9.9 pixels per full-width-half-maximum), 5400 (9.7 pixels per full-width-half-maximum), and 3900 (5.8 pixels per full-width-half-maximum) in the UVB, VIS, and NIR arms, respectively.
We ran the latest version of the X-shooter pipeline (2.8.0)[^2] on the raw data downloaded from the ESO archive with their associated raw calibration files from our three programmes: 089.C-0140(A), 091.C-0264(A), and 093.C-0610(A). All spectra have their instrumental signature removed, including bias and flat-field. The spectra are wavelength-calibrated, sky-subtracted and finally flux-calibrated. The output products include a 2D spectrum associated with a 1D spectrum. However, the optimal extraction of the 1D spectrum not being yet implemented in the version 2.8.0 of the pipeline, we extracted the UVB, VIS, and NIR spectra with the task under IRAF [@tody86; @tody93]. We note that most of the targets have little flux in the UVB arm, resulting in low signal-to-noise ratio below 550nm and very little flux (or no flux) below 400nm. We corrected the VIS and NIR spectra for telluric bands/lines with the package distributed by ESO [@kausch15; @smette15][^3] mainly because the telluric standards were not necessarily taken at the same airmass as our targets. The regions corrected are: 625–632, 686–696, 716–732, 758–770, 812–834, 893–920, 928–980nm in the VIS arm and 1105–1220, 1253–1280, 1310–1510, 1730–1995, 2000–2035, 2045–2085, 2200–2470nm in the NIR arm.
The final spectral energy distributions (SEDs) of sdM, esdM, and usdM are displayed in figures in Appendix \[MplusTeff\_sdM:appendix\_model\_fits\]. For display purposes, we shifted the observed spectra to the BT-Settl models because of the large velocities of our subdwarfs taking into account that the wavelength scale of models is in vacuum and the observed spectra in the air system (see Section \[MplusTeff\_sdM:Comparison\]). Our sample increases by a factor of five the sample of subdwarfs presented in @rajpurohit16a and extend that recent work to the full sequence of the three metallicity classes of M subdwarfs. We will make publicly available all spectra through the late-type subdwarf archive at http://svo2.cab.inta-csic.es/vocats/ltsa/ [@lodieu17a].
[@c @c c c c c c c@]{} R.A (J2000) & dec (J2000) & $i'$ & SpT & Date & Airm & ExpT & Cycles hh:mm:ss.ss & dd:mm:ss.s & mag & & & & sec/sec/sec & 23:46:59.87 & $-$00:59:43.9 & 15.166 & sdM0.0 & 13Aug2012 & 1.30 & 90/66/66 & 2AB 21:22:02.76 & $+$00:44:56.8 & 14.996 & sdM0.5 & 21May2012 & 1.13 & 90/66/66 & 2AB 00:48:00.05 & $+$00:28:49.4 & 15.260 & sdM1.0 & 13Aug2012 & 1.22 & 90/66/66 & 2AB 00:51:25.68 & $-$00:37:16.9 & 16.666 & sdM1.5 & 18Aug2012 & 1.10 & 216/216/240 & 2AB 23:57:32.54 & $-$01:10:36.3 & 15.692 & sdM2.0 & 17Aug2012 & 1.32 & 90/66/66 & 2AB 01:17:46.53 & $-$00:01:05.3 & 16.196 & sdM2.5 & 13Aug2012 & 1.36 & 90/66/66 & 2AB 03:24:49.81 & $-$00:15:05.0 & 14.942 & sdM3.0 & 15Aug2012 & 1.14 & 90/66/66 & 2AB 02:12:08.59 & $+$00:37:01.5 & 14.918 & sdM3.5 & 19Aug2012 & 1.11 & 90/66/66 & 2AB 03:46:01.58 & $+$00:55:11.6 & 15.872 & sdM4.0 & 19Aug2012 & 1.14 & 90/66/66 & 2AB 22:57:48.05 & $+$14:29:39.9 & 18.336 & sdM4.5 & 07Jul2013 & 1.39 & 276/292/300 & 5AB 00:48:36.45 & $+$00:09:31.7 & 17.175 & sdM5.0 & 10Jan2012 & 1.32 & 90/66/66 & AB 23:53:36.94 & $+$00:47:34.1 & 18.025 & sdM5.5 & 14Jul2013 & 1.26 & 276/292/300 & 5AB 00:25:52.59 & $+$01:09:24.9 & 17.797 & sdM6.0 & 19Aug2012 & 1.18 & 266/266/290 & 5AB 16:10:28.99 & $-$00:40:53.0 & 15.903 & sdM6.5 & 10Jun2012 & 1.11 & 90/66/66 & 2AB 02:05:33.75 & $+$12:38:24.0 & 18.107 & sdM8.0 & 11Aug2012 & 1.30 & 266/266/290 & 5AB 10:13:07.35 & $-$13:56:20.4 & 16.010 & sdM9.5 & 09May2013 & 1.34 & 126/142/150 & 1AB 01:37:55.30 & $-$09:39:41.9 & 17.488 & esdM0.0 & 08Sep2013 & 1.45 & 276/292/300 & 5AB 23:25:41.30 & $+$00:04:19.6 & 16.203 & esdM0.5 & 18Aug2012 & 1.13 & 90/66/66 & 2AB 00:38:02.86 & $+$00:50:14.2 & 17.520 & esdM1.0 & 14Jul2013 & 1.20 & 276/292/300 & 5AB 23:40:39.45 & $-$00:51:18.4 & 16.263 & esdM1.5 & 12Aug2012 & 1.49 & 90/66/66 & 2AB 12:55:29.19 & $-$03:30:55.8 & 17.536 & esdM2.0 & 11Jul2013 & 1.18 & 276/292/300 & 5AB 12:53:53.35 & $+$12:12:48.7 & 16.531 & esdM3.0 & 11Jul2013 & 1.30 & 126/142/150 & 3AB 14:52:04.63 & $+$10:18:26.3 & 17.892 & esdM3.5 & 30Jun2014 & 1.22 & 276/292/300 & 5AB 00:40:18.18 & $-$10:41:55.9 & 17.405 & esdM4.0 & 18Aug2012 & 1.04 & 276/292/300 & 5AB 01:09:54.11 & $-$10:12:12.6 & 14.766 & esdM4.5 & 04Aug2013 & 1.05 & 126/142/150 & 1AB 13:31:51.15 & $-$00:09:19.0 & 18.180 & esdM5.0 & 15Jun2013 & 1.16 & 276/292/300 & 5AB 09:03:07.95 & $+$08:42:43.2 & 15.988 & esdM5.5 & 18Apr2014 & 1.20 & 126/142/150 & 1AB 15:36:47.08 & $+$02:55:01.5 & 19.071 & esdM6.0 & 12Aug2012 & 1.14 & 276/276/300 & 6AB 01:33:46.24 & $+$13:28:22.4 & 17.830 & esdM6.5 & 04Aug2013 & 1.37 & 276/292/300 & 5AB 02:35:57.61 & $+$01:08:00.5 & 19.190 & esdM7.0 & 09Sep2013 & 1.11 & 456/472/480 & 4AB 05:58:58.91 & $-$29:03:26.7 & 16.320 & esdM7.5 & 27Aug2012 & 1.34 & 216/216/240 & 2AB 04:52:09.94 & $-$22:45:08.4 & 17.220 & esdM8.5 & 17Aug2012 & 1.13 & 216/216/240 & 2AB 03:27:28.10 & $-$00:50:01.4 & 17.462 & usdM0.0 & 16Spe2012 & 1.09 & 266/266/290 & 5AB 15:34:04.63 & $+$09:36:22.5 & 17.984 & usdM0.5 & 09Aug2013 & 1.22 & 266/266/290 & 5AB 15:12:18.36 & $+$09:30:40.7 & 17.890 & usdM1.0 & 23Jul2013 & 1.23 & 276/292/300 & 5AB 20:59:20.57 & $+$00:00:33.4 & 17.853 & usdM2.5 & 11Aug2012 & 1.12 & 266/266/290 & 5AB 10:41:07.20 & $+$06:33:04.7 & 18.062 & usdM3.0 & 31May2013 & 1.27 & 246/262/270 & 5AB 15:35:40.74 & $+$08:21:43.3 & 18.008 & usdM4.0 & 10Aug2013 & 1.21 & 276/292/300 & 5AB 14:17:48.62 & $+$07:11:05.4 & 17.658 & usdM4.5 & 12Jul2013 & 1.18 & 276/292/300 & 5AB 12:04:26.91 & $+$13:29:23.3 & 16.549 & usdM5.0 & 12Jun2012 & 1.33 & 216/216/240 & 2AB 16:27:54.22 & $+$00:37:14.0 & 18.384 & usdM5.5 & 17Aug2012 & 1.15 & 266/266/290 & 5AB 16:41:23.73 & $+$24:49:42.4 & 17.233 & usdM6.0 & 15Jul2013 & 1.58 & 126/142/150 & 3AB 08:22:33.69 & $+$17:00:19.9 & 17.300 & usdM7.5 & 03Apr2013 & 1.39 & 126/142/150 & 3AB 12:27:05.06 & $-$04:47:20.7 & 16.630 & usdM8.5 & 11Jul2013 & 1.35 & 126/142/150 & 3AB \[tab\_MplusTeff\_sdM:log\_obs\_Xshooter\]
BT-Settl synthetic spectra {#MplusTeff_sdM:Models}
==========================
To infer the range of physical parameters (gravities i.e. $\log$(g), metallicities \[M/H\], and T$_{\rm eff}$) for our sequence of subdwarfs, we employed the BT-Settl models [@allard03a; @allard07a; @allard10a; @allard11; @allard12] available for retrieval at France Allard’s webpage[^4]. The BT-Settl models account for TiO [@plez98; @plez08] and H$_{2}$O ([@barber06a] among other opacities using the @caffau11a abundance values and mixing information for the CO$^{5}$BOLD code [@steiner07; @freytag10a]. These models are valid for a wide range of T$_{\rm eff}$ (400–8000K), gravities ($\log$(g)=2.5–6.0 dex), and metallicities (from $-$5.0 to solar). The synthetic SEDs span the wavelength range from 10Å up to 1000$\mu$m. The stellar metallicity is defined by the total iron content of a star because iron is the easiest species to measure spectroscopically. The abundance ratio \[Fe/H\] is defined as the logarithm of the ratio of a star’s iron abundance compared to that of the Sun, where $-$1.0 dex means that a star has one-tenth of the solar metallicity. The determination of the metallicity is only valid if the abundance of all elements follow the abundance of iron. However, this statement might not be totally true in the case of halo dwarfs that suffered different nucleosynthesis events.
We downloaded the spectra from the CIFIST2011 grid[^5] and considered the ranges that encompass the spectral types of the three metallicity classes: T$_{\rm eff}$=4000–2500K, gravity ($\log$g=4.5–5.5 dex), and metallicity \[M/H\] (from $-$2.5 dex to solar). These models assumes solar abundance values (Z=0.0153, Z/X=0.0209) from @caffau11a with alpha enhancement taken into account as follows: \[alpha/H\]=$+$0.2 relative to solar for \[M/H\]=$-$0.5 dex and \[alpha/H\]=$+$0.4 for lower metallicities. For direct comparison, we limited the wavelength range to 450–2500 nm and smoothed the synthetic SEDs with the Interactive Data Language (IDL) function[^6] to the spectral resolution of our X-shooter spectra.
Throughout the paper, we use the term metallicity \[Fe/H\] which is given by the BT-Settl models. We do not infer abundances or metallicities of single element but the global values given by the synthetic spectra. The former can differ from the latter as shown for a metal-poor low-mass binary [@pavlenko15]. The study of abundances of individual elements in subdwarfs is beyond the scope of this paper but will be investigated in a future publication.
Comparison: observations vs. models {#MplusTeff_sdM:Comparison}
===================================
We fitted the observed spectra with the BT-Settl SEDs with a chi-square ($\chi^2$) minimisation procedure described in Sect. \[MplusTeff\_sdM:Comparison\_chi2\]. We derived the temperature (Sect. \[MplusTeff\_sdM:Comparison\_Teff\]) and metallicity (Sect. \[MplusTeff\_sdM:Comparison\_MH\]) scales from the best model fits.
Chi-square fitting {#MplusTeff_sdM:Comparison_chi2}
------------------
We performed a $\chi^2$ fit to compare the observed spectra with those in the BT-Settl theoretical library. We considered the following ranges in temperature, gravity, and metallicity: 4000–2500K, $\log$g=4.5–6.0 dex, and \[M/H\] between $-$3.0 and 0.0 dex, as expected for old low-mass M-type dwarfs. The steps are 100K and 0.5 dex in temperatures and gravity $+$ metallicity, respectively. We ignored regions of the observed spectra strongly affected by telluric bands, in particular the 530–570 nm, 928–1010 nm, 1110–1150 nm, 1340–1460 nm, 1790–1970 nm wavelength ranges.
We shifted our observed spectra to the wavelength of the models. We calculated the heliocentric radial velocities for our sample of metal-poor M dwarfs using a set of about 15 lines (potassium, sodium, iron, and calcium) in the 610–840 nm wavelength range, the exact number depending on the quality of the spectrum and strength of the line (Table \[tab\_MplusTeff\_sdM:table\_param\]). We measured consistent shifts between these five strong lines for all sources, with dispersions of the order of a few km/s. We assume that the true error bars are set by the resolution of our X-shooter spectra (3900–6700), corresponding to radial velocity uncertainties of approximately 15–25 km/s.
Then, we minimized the $\chi^2$ value for each observed and theoretical spectrum, as:
$$\label{eq:chi2}
\chi^2 = \frac{1}{N} \sum_i\left\{ \frac{ ({\rm F}_{\rm obs,i} - A \ {\rm F}_{\rm mod,i})^2}{(\Delta{\rm F}_{\rm obs,i})^2}\right\}$$
The scale factor $A$ is calculated to minimize $\chi^2$ for each case as:
$$\label{eq:chi2b}
A = \frac{ \sum_i({\rm F}_{\rm obs,i} {\rm F}_{\rm mod,i}/ \Delta{\rm F}_{\rm obs,i}^2)}{\sum_i({\rm F}_{\rm mod,i}^2/\Delta{\rm F}_{\rm obs,i}^2)}$$
In both expressions the sum is performed over the full wavelength range. ${\rm F}_{\rm obs,i}$ and $\Delta{\rm F}_{\rm obs,i}$ are the observed values of the flux and its associated errors, respectively. ${\rm F}_{\rm mod,i}$ are the corresponding values of the theoretical spectrum.
To avoid the fact that points with small observational errors have an excessive weight in the fitting process, we calculated the average of the error values in each spectrum as:
$$\label{eq:chi2c}
|\Delta{\rm F}_{\rm obs}| = \frac{1}{N} \sum_i\Delta{\rm F}_{\rm obs,i}$$
Then, we fixed the error to be half the average for those points with smaller errors during the fitting process.
Overall, we find that the BT-Settl reproduce well the SEDs of metal-poor M dwarfs and in particular the main molecular bands. However, we noticed that some of the main atomic lines are not well reproduced: ((Figs. \[fig\_MplusTeff\_sdM:full\_XSH\_model\_lines\_sdM\], \[fig\_MplusTeff\_esdM:full\_XSH\_model\_lines\_esdM\], \[fig\_MplusTeff\_usdM:full\_XSH\_model\_lines\_usdM\]). The lines of the sodium doublet at $\sim$820 nm predicted by the BT-Settl models appear too broad for spectral types later than approx M5. The calcium lines seem too narrow for ultra-subdwarfs but look correctly reproduced for subdwarfs and extreme subdwarfs. This mis-match between observations and models yields over or under estimates of the abundances of these elements. The disappearance of these elements in other molecules (e.g. CaOH) could also explain the observed discrepancy. We note that the potassium doublet at around 760/790nm is well reproduced by the models, except for the latest spectral types ($\geq$M7) and the coolest sources (Figs. \[fig\_MplusTeff\_sdM:full\_XSH\_model\_lines\_sdM\], \[fig\_MplusTeff\_esdM:full\_XSH\_model\_lines\_esdM\], \[fig\_MplusTeff\_usdM:full\_XSH\_model\_lines\_usdM\]).
As a consequence, we opted for four fitting procedures, described below, to gauge the uncertainties on the physical parameters derived from the synthetic spectra. We describe them below:
1. “FF” corresponds to the fit of the full SED of each subtype and metal class from 450 to 2500 nm (Figs. \[fig\_MplusTeff\_sdM:full\_XSH\_model\_FF\_sdM\]–\[fig\_MplusTeff\_sdM:full\_XSH\_model\_FF\_usdM\]).
2. “LL” corresponds the fitting procedure of a few lines in the optical spectra (K and Na). The model spectra shown correspond to the physical parameters derived from the line fits (Figs. \[fig\_MplusTeff\_sdM:full\_XSH\_model\_LL\_sdM\]–\[fig\_MplusTeff\_sdM:full\_XSH\_model\_LL\_usdM\]).
3. “FL”: in this case we fix the temperature derived from the full fit and adjust gravity and metallicity to converge towards the best fit of the aforementioned lines (Figs. \[fig\_MplusTeff\_sdM:full\_XSH\_model\_FL\_sdM\]–\[fig\_MplusTeff\_sdM:full\_XSH\_model\_FL\_usdM\]).
4. “LF”: we fix the gravity and metallicity from the fits of the line and adjust the effective temperature fitting the full SED of each subtype and subclass. (Figs. \[fig\_MplusTeff\_sdM:full\_XSH\_model\_LF\_sdM\]–\[fig\_MplusTeff\_sdM:full\_XSH\_model\_LF\_usdM\]).
In general, we find that the “FF” fitting procedure reproduces best all observed spectra. Therefore, we conclude that the physical parameters derived from this option are the most probable.
We list the model-dependent physical parameters in Table \[tab\_MplusTeff\_sdM:table\_param\] and display the best fits provided by the synthetic SEDs (red lines) to observed X-shooter spectra (black lines) for each metal class in Appendix \[MplusTeff\_sdM:appendix\_model\_fits\].
We have also looked at the physical parameters derived from the model fit to the optical region of the X-shooter spectra (600–1000nm) from which the spectral classification of subdwarfs is based [@gizis97a; @lepine07c]. We find that on average the optical spectra give equal or cooler effective temperatures, lower gravites, and/or lower metallicities (Table \[tab\_MplusTeff\_sdM:table\_best\_fits\_VV\] in Appendix \[MplusTeff\_sdM:appendix\_model\_fits\]).
[@c @c @c c c c c c c c c c c c c c c@]{} & & & & SpT & RV & $\log$g & T$_{\rm eff}$ & \[Fe/H\] & Mass & RV & $\log$g & T$_{\rm eff}$ & \[Fe/H\] & Mass & RV & $\log$g & T$_{\rm eff}$ & \[Fe/H\] & Mass & km// & dex & K & dex & M$_{\odot}$ & km/s & dex & K & dex & M$_{\odot}$ & km/s & dex & K & dex & M$_{\odot}$ M0.0 & $+$122.6 & 4.5 & 3600 & $-$1.5 & 0.133 & $-$9.2 & 5.0 & 3800 & $-$1.0 & 0.275 & $+$122.6 & 4.5 & 3500 & $-$2.0 & 0.106 M0.5 & $-$45.1 & 5.0 & 3700 & $-$1.0 & 0.216 & $-$97.3 & 5.0 & 3700 & $-$1.5 & 0.153 & $-$117.1 & 5.5 & 3700 & $-$2.5 & 0.125 M1.0 & $-$33.7 & 5.0 & 3700 & $-$0.5 & 0.352 & $-$259.7 & 5.5 & 3800 & $-$0.5 & 0.421 & $-$15.9 & 5.5 & 3700 & $-$1.0 & 0.216 M1.5 & $-$78.7 & 4.5 & 3600 & $-$0.0 & 0.600 & $-$46.7 & 5.0 & 3600 & $-$1.0 & 0.178 & — & — & — & — & — M2.0 & $-$133.8 & 5.5 & 3600 & $-$0.5 & 0.272 & $+$40.6 & 5.5 & 3600 & $-$1.0 & 0.178 & — & — & — & — & — M2.5 & $-$39.7 & 5.5 & 3600 & $-$0.5 & 0.272 & — & — & — & — & — & $-$138.8 & 5.5 & 3600 & $-$1.5 & 0.133 M3.0 & $-$52.6 & 5.0 & 3500 & $-$0.0 & 0.450 & $+$35.3 & 5.5 & 3500 & $-$1.5 & 0.120 & $+$178.0 & 6.0 & 3800 & $-$0.5 & 0.421 M3.5 & $-$21.7 & 5.0 & 3400 & $-$0.5 & 0.177 & $-$52.5 & 5.5 & 3500 & $-$1.5 & 0.120 & — & — & — & — & — M4.0 & $-$67.2 & 5.0 & 3400 & $-$1.0 & 0.129 & $-$81.3 & 5.5 & 3400 & $-$1.5 & 0.110 & $-$19.8 & 4.5 & 3200 & $-$1.5 & 0.098 M4.5 & $-$470.7 & 5.5 & 3300 & $-$1.0 & 0.117 & $-$60.7 & 5.5 & 3400 & $-$1.0 & 0.129 & $-$102.5 & 5.5 & 3400 & $-$2.0 & 0.100 M5.0 & $-$31.6 & 5.0 & 3200 & $-$0.0 & 0.150 & $+$21.7 & 4.5 & 3200 & $-$1.5 & 0.098 & $+$180.0 & 5.5 & 3500 & $-$2.5 & 0.106 M5.5 & $-$12.5 & 5.5 & 3200 & $-$1.0 & 0.107 & $+$318.1 & 5.5 & 3300 & $-$2.0 & 0.098 & $-$25.1 & 4.5 & 3300 & $-$2.0 & 0.098 M6.0 & $-$87.8 & 5.5 & 3200 & $-$1.0 & 0.107 & $-$38.5 & 5.5 & 3300 & $-$2.0 & 0.098 & $+$104.8 & 5.5 & 3300 & $-$2.0 & 0.098 M6.5 & $-$57.2 & 5.5 & 2900 & $-$0.0 & 0.063 & $-$3.7 & 5.5 & 3300 & $-$2.0 & 0.098 & — & — & — & — & — M7.0 & — & 5.0 & 3000 & $-$1.0$^{a}$ & 0.096 & $-$167.1 & 5.5 & 3200 & $-$0.0 & 0.096 & — & — & — & — & — M7.5 & — & — & — & — & — & $+$191.1 & 5.0 & 3000 & $-$2.0 & 0.090 & $+$122.1 & 5.5 & 3100 & $-$2.5 & 0.092 M8.0 & $-$208.9 & 5.5 & 2900 & $-$2.0 & 0.089 & — & — & — & — & — & — & — & — & — & — M8.5 & — & — & — & — & — & $+$72.7 & 5.5 & 3000 & $-$2.0 & 0.090 & $+$85.0 & 5.5 & 3100 & $-$2.5 & 0.092 M9.5 & $+$64.7 & 5.0 & 2800 & $-$2.0 & 0.088 & — & — & — & — & — & — & — & — & — & — \[tab\_MplusTeff\_sdM:table\_param\]
Temperature scale {#MplusTeff_sdM:Comparison_Teff}
-----------------
We derived comparable T$_{\rm eff}$ intervals for all three metallicity classes (Table \[tab\_MplusTeff\_sdM:table\_param\]) fitting the full SEDs (Tables \[tab\_MplusTeff\_sdM:table\_best\_fits\_FF\] and \[tab\_MplusTeff\_sdM:table\_best\_fits\_FL\]). The T$_{\rm eff}$ range from $\sim$3800K for the earlier M subdwarfs down to $\sim$2600K for the latest spectral types. We can hardly distinguish the three classes in the T$_{\rm eff}$ vs. spectral type diagram shown in Fig. \[fig\_MplusTeff\_sdM:plot\_SpT\_Teff\] within the error bars of 100K set by the steps available in the models. We overplotted the temperature scale of field M dwarfs (solid black line in Fig. \[fig\_MplusTeff\_sdM:plot\_SpT\_Teff\])from the latest relation of @rajpurohit13, whose trend is comparable to earlier studies within error bars [@bessell91; @leggett96; @leggett00a; @testi09a]. Dwarfs with spectral types earlier than M2 are indistinguishable in the temperature parameter space. Overall, the temperatures of metal-poor M dwarfs are similar to those of solar-type M dwarfs with an offset of 200$\pm$100K towards warmer temperatures. They follow a linear trend with some spectral types being off by 100K or 200K, which may be due to the error on the spectral classification or binarity which is mainly based on a spectral index measuring the strength of the CaH and TiO bands in the optical [@lepine07c].
We note that the temperature scale of the line fitting option tends to infer lower effective temperatures by at least 100K with a similar interval for temperatures hotter than 3400K, producing ranges of 3700–2700K. The agreement is better at lower temperatures and within the uncertainty of 100K (Tables \[tab\_MplusTeff\_sdM:table\_best\_fits\_LL\] and \[tab\_MplusTeff\_sdM:table\_best\_fits\_LF\]). We also find that the fit to the optical region only yields typically lower effective temperature by 100K to 200K (Table \[tab\_MplusTeff\_sdM:table\_best\_fits\_VV\] in Appendix \[MplusTeff\_sdM:appendix\_model\_fits\]).
![T$_{\rm eff}$ as a function of spectral type for solar-metallicity M dwarfs (black diamonds with line), sdM (red dots), esdM (green triangles) and usdM (blue squares). Error bars on spectral types and model parameters are marked for sdM only for clarity purposes. Error bars on spectral type and temperatures are $\pm$0.25 and $\pm$100K, respectively. The point for the sdM7 subdwarf corresponds to LHS377 from @rajpurohit16a. []{data-label="fig_MplusTeff_sdM:plot_SpT_Teff"}](plot_subdwarfs_SpT_Teff.png){width="\linewidth"}
Gravity scale {#MplusTeff_sdM:Comparison_grav}
-------------
Subdwarfs are old low-mass stars that belong to the thick diskor halo of our Galaxy. On average, they are much older than their solar-metallicity counterparts. @monteiro06a identified several white dwarf-subdwarf systems in the thick disk and derived ages of 6–9 Gyr for two of them with M subdwarf companions. From the comparison with FGK stellar templates selected as benchmarks for the Gaia mission [@jofre14a], @scholz15b inferred a possible age of 12 Gyr for an ancient metal-poor ($-$2.0$\pm$0.2 dex) F-type star member of the Galactic halo due to its large tangential velocity. On average, we expect our sample to be older than 5 Gyr.
From the fit of the BT-Settl models to the full VIS$+$NIR spectral energy distribution of our M subdwarfs, we infer gravities of 4.5–5.5 dex for all metal classes, except for one object with $\log$(g)=6.0 dex (usdM3.0, see discussion section). However, we observe a possible trend of increasing mean gravity with lower metallicity, going from 5.0 dex for sdM to 5.5 dex for esdM and usdM (Fig. \[fig\_MplusTeff\_sdM:plot\_SpT\_logg\]). We note that the range of gravities of the CIFIST models is limited to 6.0 dex, higher gravities are desirable to corroborate this statement. We also note that the fit to the optical region only yields typically lower gravities by 0.5 to 1.0 dex (Table \[tab\_MplusTeff\_sdM:table\_best\_fits\_VV\] in Appendix \[MplusTeff\_sdM:appendix\_model\_fits\]).
If we average the gravities for the metallicities derived from the model fit independently from the metal class (i.e. sdM, esdM, usdM), we observe that targets with metallicities of $-$0.5 dex have on average lower gravities ($\sim$4.9 dex) than more metal-poor objects where mean gravities lie between 5.2 and 5.3 dex. The difference in gravity between field M dwarfs and M subdwarfs is below our error bar of 0.5 dex, while the difference between metallicities is five times smaller than our error bars.
![Gravity as a function of spectral type for sdM (red dots), esdM (green triangles) and usdM (blue squares). Error bars on spectral type and gravity are $\pm$0.5 and $\pm$0.5 dex, respectively. The point for the sdM7 subdwarf corresponds to LHS377 from @rajpurohit16a. []{data-label="fig_MplusTeff_sdM:plot_SpT_logg"}](plot_subdwarfs_SpT_logg.png){width="\linewidth"}
![Metallicity as a function of spectral type for sdM (red dots), esdM (green triangles) and usdM (blue squares). Error bars on spectral type and metallicity are $\pm$0.5 and $\pm$0.5 dex, respectively. []{data-label="fig_MplusTeff_sdM:plot_SpT_FeH"}](plot_subdwarfs_SpT_FeH.png){width="\linewidth"}
Metallicity scale {#MplusTeff_sdM:Comparison_MH}
-----------------
Our work is extending the current metallicity scale of M dwarfs to metallicities below $-$0.5 dex with an accuracy on the calibration of the order of 0.25–0.5 dex. The classification is available in the optical for subdwarfs [@gizis97a; @lepine07c; @jao08; @savcheva14; @kesseli17a] but this is the first time this is attempted for a large sample of M0–M9 dwarfs [see also @rajpurohit16a]. The determination of the metallicity scale of subdwarfs is key for several reasons. First, the range of metallicities for M subdwarfs is poorly constrained up to now for these objects that represent the first generation of stars and are key sources to study the chemical evolution of our Galaxy. On the other hand, M dwarfs are becoming popular to search for low-mass planets by the radial velocity and transit techniques, leading independent groups to look at their metallicity. Nonetheless, all the studies referenced in the introduction focussed on M dwarfs with metallicity slightly below solar with calibrations accurate to $<$0.15 dex either photometrically [@bonfils05a; @johnson09a; @schlaufman10a; @neves12; @hejazi15; @dittmann16] or spectroscopically [@woolf05; @woolf06; @bean06a; @woolf09; @rojas_ayala10; @rojas_ayala12; @muirhead12; @terrien12a; @onehag12; @neves13; @neves14; @hejazi15; @newton15a; @lindgren16a].
The number of M dwarf hosts below 0.5 M$_{\odot}$ and metallicities lower than $-$0.5 dex is extremely small ($\leq$0.2%) compared to the 3396 confirmed planets[^7]. Only three metal-poor M dwarfs have one or several planets: Kapteyn [sdM1.0; Fe/H=$-$0.86 dex; @anglada_escude14; @robertson15a], GJ667 [K3V$+$K5V; Fe/H=$-$0.55 dex; @feroz14; @anglada_escude13a; @delfosse13], and Kepler1124 [Fe/H=$-$0.59 dex; @morton16a]. Another four Kepler M stars with masses in the range 0.5–0.6 M$_{\odot}$ and metallicities between $-$0.5 and $-$1.0 harbour planets.
We present VLT/X-shooter spectra at a resolution of 5400 and 3900 in the VIS and NIR arms, respectively. We estimate metallicity that rely on models and synthetic spectra whose steps are 0.5 dex, which we set as our error bars. Overall, we find that the overall SEDs of subdwarfs, extreme subdwarfs, and ultra-subdwarfs are best reproduced by metallicities between solar and $-$1.0 dex, $-$1.0 and $-$2.0 dex, and $-$1.0 and $-$2.5 dex, respectively (Fig. \[fig\_MplusTeff\_sdM:plot\_SpT\_FeH\]). We note that a few sdM have solar metallicities, including the sdM6.5 (1610$-$0040), a known astrometric binary [@koren16] with a mix of spectral features typical of both L subdwarfs and solar-type M dwarfs [@lepine03b; @reiners06a; @cushing06]. We also emphasise that the two coolest subdwarfs (sdM8 and sdM9.5) in our sample have on average lower metallicities than subdwarfs earlier than sdM7, suggesting saturation of the metallicity index [@lepine07c] used in the classification of subdwarf [@zhang17a]. We conclude that there is a trend towards lower mean metallicity from subdwarfs ($-$0.5$\pm$0.5 dex) to extreme subdwarfs ($-$1.5$\pm$0.5 dex), and ultra-subdwarfs ($-$2.0$\pm$0.5 dex), consistent with the original definition of these three metal classes. However, we observe variations from one object to the other, preventing assigning a given M/H for each metal class. We note that the metallicity scale of the line fitting option differs significantly from the fit of the SEDs, producing lower metallicities by at least 1.0 dex. (Tables \[tab\_MplusTeff\_sdM:table\_best\_fits\_LL\] and \[tab\_MplusTeff\_sdM:table\_best\_fits\_LF\]). We find the same trend if we consider only the optical range (600–1000 nm) for our fitting procedure.
We compared our global results to the recent analysis of @rajpurohit16a also based on VLT/X-shooter spectra but for a much smaller sample of subdwarfs: six sdM (five sdM0.5–M3 and one sdM7), two esdM (esdM2 and esdM4), and a usdM4.5. In their Table 2, they list two possible fits, which usually differ by either 100K or 0.5 dex in gravity or metallicity. Our physical parameters agree with their results within 1–2$\sigma$ of the steps of the synthetic SEDs (Table \[tab\_MplusTeff\_sdM:table\_param\]). The difference can be explained by the uncertainties on the spectral types, typically 0.5 sub-type. We conclude that both T$_{\rm eff}$ vs. spectral type relations are consistent within current error bars on observational and theoretical sides. Moreover, as stated in @rajpurohit16a, these results are fully consistent with the physical parameters derived from higher resolution optical spectroscopy [@rajpurohit14].
We can also compare our work to the physical parameters of the metal-poor binary G224-58AB composed of a esdK5 and a esdM5.5. @pavlenko15 derived the temperature and gravity for the secondary (T$_{\rm eff}$=3200$\pm$100 K, $\log$g=5.0$\pm$0.5), and inferred the metallicity from the primary (\[Fe/H\]=$-$1.92 dex). Their parameters are in good agreement with our values for the esdM5.5 (SDSSJ09030795$+$0842432) in our sample: T$_{\rm eff}$=3300$\pm$100K, $\log$g=5.5$\pm$0.5 dex, and \[Fe/H\]=$-$1.5$\pm$0.5 dex (Table \[tab\_MplusTeff\_sdM:table\_param\]). Their abundance analysis suggests that many elements (e.g. calcium; \[Ca/H\]=$-$1.39$\pm$0.03) are over-abundant compared to iron. We do also find that, although the BT-Settl SEDs reproduce well the molecular bands of the observed spectra, they fail to fit the atomic lines of many elements like Fe, K, Na, and Ca. In other words, the atomic lines from the BT-Settl models are too broad and over-estimate the abundances of single elements (see Sect. \[MplusTeff\_sdM:Comparison\_Abund\]). The disappearance of these elements in other molecules (e.g. CaOH) could also explain the observed discrepancy.
Masses {#MplusTeff_sdM:Comparison_Mass}
------
We inferred model-dependent masses for our sample of subdwarfs from their temperatures looking at the predictions of the NextGen models [@baraffe97], assuming ages of 10 Gyr for these old objects. Although this approach is not correct due to the different inputs physics in the NextGen and BT-Settl grids [computation available on France Allard’s webpage; @baraffe15], it provides estimates on the masses of these low-metallicity M dwarfs. We derived masses below 0.35 M$_{\odot}$ for the most massive M subdwarfs down to $\sim$0.087 M$_{\odot}$ for the coolest objects in our sample, i.e. close to the stellar/substellar boundary with uncertainties of the order of 10% (Table \[tab\_MplusTeff\_sdM:table\_param\]). We note that the gravities quoted by the @baraffe97 models are in the range $\log$(g)=5.0–5.5 dex, consistent with the aforementioned gravities derived from the synthetic SEDs within the error bars of 0.5 dex.
Wavelength range Elements \# lines
--------------------------------------------- ---------------- ------------
5750–5995Å Na, Ti, Ca, Fe 2, 1, 1, 1
7640–7730Å K 2
8080–8350Å Na, Fe 3, 1
8300–8850Å Ca, Ti, Fe 3, 10, 7
\[MplusTeff\_sdM:elements\_for\_abundance\]
: Spectral regions, elements, and their numbers of lines considered to derive abundances of our sample of subdwarfs
[@l @c @c @c @c @c @c@]{} SpT & Model & Na & K & Ca & Ti & Fe sdM0.5 & 3700-4.5-1.0 & 0.3$_{-0.2}^{+0.3}$ & 0.7$_{-0.3}^{+0.4}$ & 1.2$_{-0.0}^{+0.0}$ & 0.9$_{-0.1}^{+0.1}$ & 0.9$_{-0.0}^{+0.1}$ sdM1.0 & 3700-5.0-0.5 & 0.2$_{-0.2}^{+0.2}$ & 0.3$_{-0.3}^{+0.3}$ & 0.5$_{-0.2}^{+0.1}$ & 0.4$_{-0.1}^{+0.1}$ & 0.4$_{-0.1}^{+0.1}$ sdM1.5$^{*}$ & 3400-4.5-1.5 & 0.4$_{-0.1}^{+0.1}$ & 0.6$_{-0.2}^{+0.2}$ & 1.0$_{-0.0}^{+0.0}$ & 0.5$_{-0.1}^{+0.1}$ & 0.5$_{-0.1}^{+0.1}$ sdM2.0$^{*}$ & 3500-4.5-1.0 & 0.7$_{-0.2}^{+0.2}$ & 1.0$_{-0.3}^{+0.3}$ & 0.9$_{-0.0}^{+0.0}$ & 0.5$_{-0.1}^{+0.1}$ & 0.5$_{-0.1}^{+0.1}$ sdM2.5$^{*}$ & 3500-4.5-1.0 & 0.6$_{-0.2}^{+0.2}$ & 0.9$_{-0.3}^{+0.3}$ & 0.7$_{-0.0}^{+0.0}$ & 0.6$_{-0.1}^{+0.1}$ & 0.5$_{-0.1}^{+0.1}$ sdM3.0 & 3500-5.0-0.0 & $-$0.2$_{-0.1}^{+0.1}$ & 0.6$_{-0.2}^{+0.2}$ & 0.2$_{-0.0}^{+0.0}$ & 0.0$_{-0.1}^{+0.1}$ & 0.1$_{-0.0}^{+0.0}$ sdM3.5 & 3400-5.0-0.5 & 0.2$_{-0.2}^{+0.2}$ & 0.3$_{-0.3}^{+0.2}$ & 0.3$_{-0.0}^{+0.1}$ & 0.3$_{-0.2}^{+0.2}$ & 0.1$_{-0.1}^{+0.1}$ sdM4.0 & 3400-5.5-0.5 & $-$0.1$_{-0.2}^{+0.3}$ & 0.2$_{-0.3}^{+0.3}$ & 0.8$_{-0.0}^{+0.0}$ & 0.4$_{-0.1}^{+0.2}$ & 0.2$_{-0.1}^{+0.1}$ sdM4.5 & 3100-4.5-1.5 & 0.5$_{-0.2}^{+0.2}$ & 0.6$_{-0.3}^{+0.3}$ & 0.6$_{-0.0}^{+0.0}$ & 0.9$_{-0.3}^{+0.3}$ & 0.4$_{-0.1}^{+0.1}$ sdM5.0$^{*}$ & 3200-5.5-0.0 & 0.5$_{-0.2}^{+0.2}$ & 0.3$_{-0.3}^{+0.4}$ & 0.0$_{-0.0}^{+0.0}$ & 0.3$_{-0.2}^{+0.3}$ & 0.1$_{-0.1}^{+0.2}$ sdM5.5 & 3200-5.5-1.0 & 0.0$_{-0.2}^{+0.2}$ & 0.1$_{-0.3}^{+0.3}$ & 0.4$_{-0.0}^{+0.2}$ & 0.6$_{-0.2}^{+0.2}$ & 0.1$_{-0.1}^{+0.1}$ sdM6.0 & 3000-4.5-1.5 & 0.6$_{-0.2}^{+0.2}$ & 0.3$_{-0.3}^{+0.4}$ & 1.4$_{-0.0}^{+0.0}$ & 0.7$_{-0.2}^{+0.3}$ & 0.4$_{-0.1}^{+0.2}$ sdM6.5$^{*}$ & 2900-5.5-0.0 & 0.2$_{-0.4}^{+0.5}$ & 0.4$_{-0.5}^{+0.6}$ & 0.0$_{-0.0}^{+0.0}$ & 0.0$_{-0.2}^{+0.3}$ & 0.0$_{-0.1}^{+0.2}$ sdM8.0 & 2800-5.0-2.5 & 0.4$_{-0.2}^{+0.2}$ & 0.3$_{-0.2}^{+0.2}$ & 0.7$_{-0.0}^{+0.0}$ & 0.6$_{-0.2}^{+0.2}$ & 0.0$_{-0.1}^{+0.1}$ sdM9.5 & 2600-4.5-2.5 & 0.6$_{-0.2}^{+0.2}$ & 0.4$_{-0.2}^{+0.2}$ & 0.5$_{-0.0}^{+0.0}$ & 0.8$_{-0.2}^{+0.2}$ & 0.5$_{-0.1}^{+0.1}$ esdM0.0 & 3600-4.5-1.5 & 0.2$_{-0.2}^{+0.2}$ & 0.4$_{-0.2}^{+0.3}$ & 0.6$_{-0.0}^{+0.0}$ & 0.6$_{-0.1}^{+0.2}$ & 0.3$_{-0.0}^{+0.1}$ esdM0.5 & 3600-4.5-1.5 & 0.2$_{-0.2}^{+0.2}$ & 0.6$_{-0.2}^{+0.3}$ & 0.5$_{-0.0}^{+0.0}$ & 0.6$_{-0.1}^{+0.2}$ & 0.3$_{-0.0}^{+0.1}$ esdM1.0 & 3600-4.5-1.5 & 0.3$_{-0.2}^{+0.2}$ & 0.6$_{-0.2}^{+0.3}$ & 0.6$_{-0.0}^{+0.0}$ & 0.7$_{-0.1}^{+0.2}$ & 0.1$_{-0.0}^{+0.1}$ esdM1.5 & 3400-4.5-1.5 & 0.2$_{-0.2}^{+0.2}$ & 0.4$_{-0.2}^{+0.3}$ & 0.7$_{-0.0}^{+0.0}$ & 0.7$_{-0.1}^{+0.2}$ & 0.4$_{-0.0}^{+0.1}$ esdM2.0 & 3400-4.5-1.5 & 0.3$_{-0.2}^{+0.2}$ & 0.5$_{-0.2}^{+0.3}$ & 0.8$_{-0.0}^{+0.0}$ & 0.9$_{-0.1}^{+0.2}$ & 0.5$_{-0.0}^{+0.1}$ esdM3.0 & 3400-5.0-1.5 & 0.2$_{-0.2}^{+0.2}$ & 0.5$_{-0.2}^{+0.2}$ & 0.7$_{-0.0}^{+0.0}$ & 0.9$_{-0.2}^{+0.2}$ & 0.4$_{-0.1}^{+0.1}$ esdM3.5 & 3400-5.0-1.5 & 0.0$_{-0.2}^{+0.2}$ & 0.4$_{-0.2}^{+0.2}$ & 0.4$_{-0.0}^{+0.0}$ & 0.7$_{-0.2}^{+0.2}$ & 0.2$_{-0.1}^{+0.1}$ esdM4.0 & 3200-4.5-2.0 & 0.5$_{-0.2}^{+0.2}$ & 0.7$_{-0.2}^{+0.2}$ & 0.4$_{-0.0}^{+0.0}$ & 0.9$_{-0.2}^{+0.2}$ & 0.4$_{-0.1}^{+0.1}$ esdM4.5 & 3200-4.5-1.5 & 0.6$_{-0.2}^{+0.2}$ & 0.4$_{-0.2}^{+0.2}$ & 0.9$_{-0.0}^{+0.0}$ & 0.8$_{-0.2}^{+0.2}$ & 0.4$_{-0.1}^{+0.1}$ esdM5.0 & 3200-4.5-1.5 & 0.4$_{-0.2}^{+0.2}$ & 0.7$_{-0.2}^{+0.3}$ & 0.7$_{-0.0}^{+0.0}$ & 0.9$_{-0.2}^{+0.2}$ & 0.5$_{-0.1}^{+0.1}$ esdM5.5 & 3100-4.5-2.0 & 0.6$_{-0.2}^{+0.2}$ & 0.8$_{-0.2}^{+0.2}$ & 0.8$_{-0.0}^{+0.0}$ & 1.2$_{-0.3}^{+0.3}$ & 0.7$_{-0.1}^{+0.2}$ esdM6.0 & 3200-5.0-2.0 & 0.1$_{-0.2}^{+0.2}$ & 0.6$_{-0.2}^{+0.2}$ & 0.4$_{-0.0}^{+0.0}$ & 0.8$_{-0.2}^{+0.2}$ & 0.4$_{-0.1}^{+0.1}$ esdM6.5 & 3100-4.5-2.0 & 0.3$_{-0.3}^{+0.3}$ & 0.7$_{-0.2}^{+0.2}$ & 0.7$_{-0.0}^{+0.0}$ & 1.0$_{-0.3}^{+0.3}$ & 0.4$_{-0.1}^{+0.1}$ esdM7.5 & 3000-5.0-2.0 & 0.2$_{-0.2}^{+0.2}$ & 0.6$_{-0.2}^{+0.2}$ & 0.7$_{-0.0}^{+0.0}$ & 0.8$_{-0.2}^{+0.2}$ & 0.3$_{-0.1}^{+0.1}$ esdM8.5 & 2800-4.5-2.5 & 0.4$_{-0.2}^{+0.2}$ & 0.6$_{-0.2}^{+0.2}$ & 0.6$_{-0.0}^{+0.0}$ & 0.8$_{-0.2}^{+0.3}$ & 0.1$_{-0.1}^{+0.1}$ usdM0.0 & 3500-4.5-2.5 & 0.5$_{-0.3}^{+0.4}$ & 1.2$_{-0.4}^{+0.5}$ & 1.2$_{-0.0}^{+0.0}$ & 0.6$_{-0.2}^{+0.3}$ & 0.2$_{-0.2}^{+0.2}$ usdM0.5$^{*}$ & 3600-5.0-2.5 & $-$0.4$_{-0.2}^{+0.3}$ & 0.6$_{-0.4}^{+0.4}$ & 0.4$_{-0.0}^{+0.0}$ & 0.5$_{-0.2}^{+0.2}$ & 0.1$_{-0.1}^{+0.1}$ usdM1.0 & 3600-5.0-1.5 & $-$0.2$_{-0.2}^{+0.3}$ & 0.6$_{-0.4}^{+0.4}$ & 0.5$_{-0.0}^{+0.0}$ & 0.3$_{-0.2}^{+0.2}$ & 0.2$_{-0.1}^{+0.1}$ usdM2.5 & 3400-4.5-2.0 & 0.1$_{-0.2}^{+0.2}$ & 0.8$_{-0.4}^{+0.4}$ & 1.0$_{-0.0}^{+0.0}$ & 0.7$_{-0.2}^{+0.2}$ & 0.8$_{-0.1}^{+0.1}$ usdM3.0 & 3400-4.5-2.0 & $-$0.2$_{-0.2}^{+0.2}$ & 1.2$_{-0.4}^{+0.4}$ & 0.9$_{-0.0}^{+0.0}$ & 0.5$_{-0.2}^{+0.4}$ & 0.3$_{-0.1}^{+0.1}$ usdM4.0$^{*}$ & 3500-5.5-2.0 & 0.1$_{-0.2}^{+0.3}$ & 0.8$_{-0.3}^{+0.4}$ & 0.9$_{-0.0}^{+0.0}$ & 0.8$_{-0.2}^{+0.2}$ & 0.5$_{-0.1}^{+0.1}$ usdM4.5 & 3300-5.0-2.0 & $-$0.1$_{-0.2}^{+0.3}$ & 0.6$_{-0.3}^{+0.3}$ & 0.5$_{-0.0}^{+0.0}$ & 0.5$_{-0.2}^{+0.2}$ & 0.3$_{-0.1}^{+0.1}$ usdM5.0 & 3400-5.5-2.5 & $-$0.4$_{-0.2}^{+0.2}$ & 0.6$_{-0.3}^{+0.3}$ & 0.6$_{-0.0}^{+0.0}$ & 0.5$_{-0.2}^{+0.2}$ & 0.3$_{-0.1}^{+0.1}$ usdM6.0 & 3200-5.0-2.0 & 0.1$_{-0.1}^{+0.2}$ & 1.0$_{-0.3}^{+0.3}$ & 0.4$_{-0.0}^{+0.0}$ & 0.6$_{-0.1}^{+0.1}$ & 0.3$_{-0.1}^{+0.1}$ usdM7.5 & 3000-5.0-2.5 & $-$0.1$_{-0.2}^{+0.2}$ & 1.0$_{-0.3}^{+0.3}$ & 0.4$_{-0.1}^{+0.1}$ & 0.6$_{-0.1}^{+0.1}$ & 0.2$_{-0.1}^{+0.2}$ usdM8.5 & 3000-5.5-2.5 & $-$0.2$_{-0.2}^{+0.2}$ & 0.8$_{-0.3}^{+0.3}$ & 0.3$_{-0.1}^{+0.1}$ & 0.4$_{-0.2}^{+0.2}$ & 0.0$_{-0.1}^{+0.2}$ \[MplusTeff\_sdM:results\_abundance\]
{width="0.48\linewidth"} {width="0.48\linewidth"} {width="0.48\linewidth"} {width="0.48\linewidth"} {width="0.48\linewidth"}
Abundances {#MplusTeff_sdM:Comparison_Abund}
----------
We decided to work with the temperatures, gravities, and metallicities derived from the best fits to the optical regions only (Table \[tab\_MplusTeff\_sdM:table\_best\_fits\_VV\] in Appendix \[MplusTeff\_sdM:appendix\_model\_fits\]) to derive the abundances of several elements (Table \[MplusTeff\_sdM:elements\_for\_abundance\]). The relatively low resolution and reduced signal-to-noise of the observed spectra limit the accuracy of our results. Moreover, we work here with metal-deficient stars, in which spectra are not critically affected by blending effects. We carried out the minimisation procedure using the PHOENIX model atmospheres to compute synthetical spectra with the WITA program [@pavlenko97a]. In these computations, we accounted for the main molecular opacity sources, as described in @pavlenko15 and @pavlenko14. We employed the atomic line information from the the VALD2 and VALD3 databases [@kupka99; @ryabchikova11]. We selected a few spectral ranges with well defined atomic lines and fitted them in framework of the LTE approach to determine abundances of several elements listed in Table \[MplusTeff\_sdM:elements\_for\_abundance\].
To determine the abundances, we compared the residuals of the fluxes and reduced to the local continuum or pseudo-continuum of the observed spectra. We determined these continua/pseudo-continua for all spectral ranges given in Table \[MplusTeff\_sdM:elements\_for\_abundance\]. We carried out the computations for 20 points abundance grid with a step 0.1 dex around the mean abundance for each metal and spectral subclass. The abundances correspond to over-abundances if they are positive, fixing the iron abundances to the metallicity derived from the model fit of the optical region, assuming a solar abundance of iron of $-$4.55 dex [@asplund09].
When several lines are present in a given spectral range, we averaged the values of each abundance. We give the uncertainties on the abundances rounded to the nearest integer for a range of $\pm$100K in effective temperature, which represents the error on our model fits (and the step of the models). We note that the sodium, potassium, and neutral calcium lines are the most sensitive to small changes in effective temperature due to the low ionisation potentials. We estimate the formal accuracy of the abundance determination to be $\pm$0.1 dex due to several uncertainties in the procedure, such as low signal-to-noise ratio and the determination of the levels of the continuum and pseudo-continuum. We list the final abundances of five elements (Na, K, Ca, Ti, and Fe) with their uncertainties in Table \[MplusTeff\_sdM:results\_abundance\]. Some of the stars, marked with an asterisk, have low signal-noise spectra so the abundances should be interpreted with caution.
We should note that any significant changes in the abundance of alkali metals affect the fluxes of the continuum, hence, the strength of the lines. Alkali metals (Na, K, Ca, Mg) are known donors of free electrons in atmospheres of late-type stars. In our case, we re-computed the continuum fluxes for all the abundances, so, in a first approach, we accounted for the effect of opacity changes. However, these changes of abundances may affect the temperature structure of the model atmosphere, changes not accounted for because we employed model atmospheres with a fixed temperature structure.
Our abundance analysis was performed for the model parameters obtained from the fits to observed SEDs (Section \[MplusTeff\_sdM:Models\]). Furthermore, resolution of the fitted spectra was not high enough to get high quality results. On the second hand, we work here with metal-deficient stars, in which spectra are not critically affected by blending effects. We determined a few spectral ranges with well defined atomic lines and fitted them in framework of the LTE approach. Despite all factors constraining the accuracy of our analysis we get a few confident enough results:
- we see a shift of Ti abundance distribution toward larger metallicities with respect to Fe for the lowest metallicities (top left panel of Fig. \[fig\_MplusTeff\_sdM:hists\_abundances\]).
- we observe a Ca over-abundance in the atmospheres of all subdwarfs (top right panel of Fig. \[fig\_MplusTeff\_sdM:hists\_abundances\]). This result agrees with the known enhancement of $\alpha$ elements in the atmospheres of halo dwarfs [e.g. @francois86a; @nissen94a; @gratton94a; @fuhrmann95a].
- we find that K tends to be over-abundant at higher metallicities, opposite effect to Na (middle panels of Fig. \[fig\_MplusTeff\_sdM:hists\_abundances\]). This difference increases with lower metallicity and is most notable for ultra-subdwarfs. Perhaps, these elements have different chemical histories. The difference is well seen in the comparison plot of the distribution of both species where the three metal classes are fairly well separated (bottom panel of Fig. \[fig\_MplusTeff\_sdM:hists\_abundances\].
- we cannot describe the abundances in the atmospheres of our targets with one parameter metallicity (Table \[MplusTeff\_sdM:results\_abundance\] and Fig. \[fig\_MplusTeff\_sdM:hists\_abundances\]).
Discussion {#MplusTeff_sdM:discussion}
==========
We observe that on average the optical range of the subdwarf spectra yields cooler temperatures, lower gravities, and metallicities. Hence, the availability of the near-infrared slope is quite important to determine the physical parameters of the three metal classes and break down the temperature/metallicity degeneracy present in the analysis of the sole visible range. The temperature sequence is well determined for all metal class, with the earliest spectral types being the warmest cases. The metallicity is on average lower for the ultra-subdwarfs, with all three metal class being more metal poor than solar as expected. Those results are, however, global with the physical parameters sensitive to steps of the models in temperature, gravity, and metallicity. Smaller changes in those parameters are most likely limited by the quality (signal-to-noise ratio) and resolution of the X-shooter spectra.
We checked the few cases where our best fits suggest solar metallicity. There are four sdM in this case (including the sdM6.5 specifically treated below) and only one esdM (esdM7.0) also discussed below. For the sdM1.5 we find that the best “FF” fits indicate solar metallicity while the other procedures suggest $-$0.5 dex. The inspection of the width of the lines (K, Na, Ca) is inconclusive to decide on one or the other metallicity (Fig. \[fig\_MplusTeff\_sdM:full\_XSH\_model\_lines\_sdM\]). For the sdM3.0, all four fits are identical and suggest solar metallicity while the third best fit of the “FF” procedure yields $-$0.5 dex. The case of sdM5.0 indicates solar metallicity too but all other fits suggest $-$0.5 dex, confirmed by the width of the sodium and potassium lines, which advocate metallicity lower than solar with a temperature 3200K. In the case of these three subdwarfs, we conclude that they might be peculiar somehow and require further investigation with better quality data.
We also reviewed the four extreme subdwarfs (esdM1.0, esdM1.5, esdM2.0, and esdM4.5) with differences of 200K or more in temperature and $\geq$ 1.0 dex in metallicity comparing the four fitting procedure described in Section \[MplusTeff\_sdM:Comparison\_chi2\]. We note that the three best fits of the “FF” procedure yield results within the step of the models. Therefore, we inspected the sodium, potassium, and calcium lines where we noticed that the best model fits occurred for the lowest metallicity values in all cases (Fig. \[fig\_MplusTeff\_esdM:full\_XSH\_model\_lines\_esdM\]).
Finally, we observe four notable outliers discussed below after checking the best fits derived from the different procedures. We emphasise that we obtained one single spectrum per target.
- [*[sdM0.0]{}*]{} is not included in our abundance analysis because the sodium and potassium lines are clearly resolved in our X-shooter spectrum. We confirm this target as a spectroscopic binary, which will be discussed in a separate paper. Any fit to spectral lines and abundance analysis on the spectrum taken at this specific epoch is therefore flawed.
- [*[sdM6.5]{}*]{} (1610$-$0040) is a known astrometric binary [@koren16] with a mix of spectral features typical of both L subdwarfs and solar-type M dwarfs [@lepine03b; @reiners06a; @cushing06]. We derive a temperature of 2900K with a gravity of 5.5 dex and solar metallicity with a reasonable “FF” model fit (Fig. \[fig\_MplusTeff\_sdM:full\_XSH\_model\_FF\_sdM\]).
- [*[esdM7.0]{}*]{} show solar metallicity in the case of the full fitting procedure but lower metallicities with the other three cases. We inspected the sodium, potassium, and calcium lines of this object and conclude that the lines are broader than model predictions at solar metallicity, favouring the metal poor solution. (Fig. \[fig\_MplusTeff\_esdM:full\_XSH\_model\_lines\_esdM\]). We note that this object has a low quality spectrum too.
- [*[usdM3.0]{}*]{} exhibits a large difference when inspecting the best three chi$^{2}$ fits using the “FF” procedure, with variations of up to 200K in temperature and 2.0 dex in metallicity. These differences are larger than any other object in our sample. We inspected the width and depth of the main sodium and potassium lines and find that they appear shallower with a potential double peak, which may indicate binarity (Fig. \[fig\_MplusTeff\_usdM:full\_XSH\_model\_lines\_usdM\]). However, we obtained one single spectrum for this source so we cannot exclude other phenomenon like rotation, flare activity or spots that can affect our results. A few additional spectra with a minimum spectral resolution of 10000 are required to confirm this possibility, which would explain the large variations in the determination of the physical parameters.
Conclusions {#MplusTeff_sdM:conclusions}
===========
Our atlas of VLT/X-shooter 0.45–2.5 $\mu$m moderate-resolution spectra represents an important database to classify metal-poor subdwarfs and has a legacy value for future large-scale surveys [WEAVE, 4MOST, LSST; @dalton12; @dalton14; @deJong14a; @ivezic08a] and space missions to come such as the James Webb Space Telescope [@gardner09] and Euclid [@mellier16a]. We derived radial velocities for all subdwarfs from the shift of the strongest optical lines. We inferred physical parameters (metallicities, temperatures, and gravities) for 43 metal-poor M dwarf by comparing their spectral energy distributions over the 450–2500nm range to the latest BT-Settl synthetic spectra. The main results of our analysis are:
- the best gravity range for M subdwarfs is $\log$(g)=5.0–5.5 dex.
- the metallicities inferred from the BT-Settl models for subdwarfs, extreme subdwarfs, and ultra-subdwarfs are $-$0.5, $-$1.5, and $-$2.0 dex with errors of 0.5 dex, respectively.
- the ranges in T$_{\rm eff}$ for subdwarfs, extreme subdwarfs, and ultra-subdwarfs are comparable and lie in the intervals 3700–2600K, 3800–2900K, and 3700–2900K with uncertainties of 100K, respectively.
- the Ca and Ti elements show an over-abundance while Na behaves in an opposite manner when compared to the iron abundance.
Improvements in the determination of the physical parameters and abundances of metal-poor low-mass dwarfs require refined and uniform/complete model atmosphere grids to improve the fits. More advanced procedures are needed to improve the quality of the fits to observed SEDs treating atomic lines in a self-consistent approach, with model atmospheres and spectra are computed for one set of input parameters. Models should take into account the enhancement of C/O and improve the modelling of single lines updating abundances of the various elements present in cool atmospheres. To truly determine the physical parameters of M subdwarfs and test evolutionary models at low metallicity, the discovery of metal-poor low-mass transiting eclipsing binaries is key.
NL was funded by the Ramón y Cajal fellowship number 08-303-01-02, and supported by the grants numbers AYA2010-19136 and AYA2015-69350-C3-2-P from Spanish Ministry of Economy and Competitiveness (MINECO). We thank Victor Béjar, Yakiv Pavlenko, and ZengHua Zhang for constructive comments on this work. The research leading to these results has received funding from the French Programme National de Physique Stellaire and the Programme National de Plan- etologie of CNRS (INSU). FA thanks financial support from the Fundación Jesús Serra for a 2 month stay (Jan–Feb 2015) as a visiting professor at the Instituto de Astrofísica de Canarias (IAC) in Tenerife.
This work is based on observations collected with X-shooter on the VLT at the European Southern Observatory, Chile, under programmes 089.C-0140(A), 091.C-0264(A), 092.D-0600(A), and 093.C-0610(A).
This research has made use of the Simbad and Vizier databases, operated at the Centre de Données Astronomiques de Strasbourg (CDS), and of NASA’s Astrophysics Data System Bibliographic Services (ADS).
Funding for the Sloan Digital Sky Survey IV has been provided by the Alfred P. Sloan Foundation, the U.S. Department of Energy Office of Science, and the Participating Institutions. SDSS-IV acknowledges support and resources from the Center for High-Performance Computing at the University of Utah. The SDSS web site is www.sdss.org. SDSS-IV is managed by the Astrophysical Research Consortium for the Participating Institutions of the SDSS Collaboration including the Brazilian Participation Group, the Carnegie Institution for Science, Carnegie Mellon University, the Chilean Participation Group, the French Participation Group, Harvard-Smithsonian Center for Astrophysics, Instituto de Astrofísica de Canarias, The Johns Hopkins University, Kavli Institute for the Physics and Mathematics of the Universe (IPMU) / University of Tokyo, Lawrence Berkeley National Laboratory, Leibniz Institut für Astrophysik Potsdam (AIP), Max-Planck-Institut für Astronomie (MPIA Heidelberg), Max-Planck-Institut für Astrophysik (MPA Garching), Max-Planck-Institut für Extraterrestrische Physik (MPE), National Astronomical Observatory of China, New Mexico State University, New York University, University of Notre Dame, Observatário Nacional / MCTI, The Ohio State University, Pennsylvania State University, Shanghai Astronomical Observatory, United Kingdom Participation Group, Universidad Nacional Autónoma de México, University of Arizona, University of Colorado Boulder, University of Oxford, University of Portsmouth, University of Utah, University of Virginia, University of Washington, University of Wisconsin, Vanderbilt University, and Yale University.
Model fit to VLT/X-shooter spectra {#MplusTeff_sdM:appendix_model_fits}
==================================
{width="\linewidth"}
{width="\linewidth"}
{width="\linewidth"}
{width="\linewidth"}
{width="\linewidth"}
{width="\linewidth"}
{width="\linewidth"}
{width="\linewidth"}
{width="\linewidth"}
{width="\linewidth"}
{width="\linewidth"}
{width="\linewidth"}
{width="0.90\linewidth"}
{width="0.90\linewidth"}
{width="\linewidth"}
Tables with physical parameters of subdwarfs {#MplusTeff_sdM:appendix_tables}
============================================
[c c c c c c c c]{} SpT & Range & T$_{\rm eff}$ & $\log$g & M/H & chi$^{2}$ & Factor & Model\
sdM0.0 & FF & 3600 & 4.5 & -2.0 & 1.336 & 2.263e-22 & lte036.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat sdM0.5 & FF & 3700 & 5.0 & -1.0 & 72.07 & 1.782e-21 & lte037.0-5.0-1.0a+0.4.BT-Settl.spec.7.dat sdM1.0 & FF & 3700 & 5.0 & -0.5 & 67.22 & 1.609e-21 & lte037.0-5.0-0.5a+0.2.BT-Settl.spec.7.dat sdM1.5 & FF & 3600 & 4.5 & -0.0 & 66.4 & 3.168e-22 & lte036.0-4.5-0.0a+0.0.BT-Settl.spec.7.dat sdM2.0 & FF & 3600 & 5.5 & -0.5 & 46.01 & 9.281e-22 & lte036.0-5.5-0.5a+0.2.BT-Settl.spec.7.dat sdM2.5 & FF & 3600 & 5.5 & -0.5 & 35.09 & 7.689e-22 & lte036.0-5.5-0.5a+0.2.BT-Settl.spec.7.dat sdM3.0 & FF & 3500 & 5.0 & -0.0 & 88.51 & 2.208e-21 & lte035.0-5.0-0.0a+0.0.BT-Settl.spec.7.dat sdM3.5 & FF & 3400 & 5.0 & -0.5 & 161.6 & 3.916e-21 & lte034.0-5.0-0.5a+0.2.BT-Settl.spec.7.dat sdM4.0 & FF & 3400 & 5.0 & -1.0 & 58.22 & 1.54e-21 & lte034.0-5.0-1.0a+0.4.BT-Settl.spec.7.dat sdM4.5 & FF & 3300 & 5.5 & -1.0 & 59.07 & 2.072e-22 & lte033.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat sdM5.0 & FF & 3200 & 5.0 & -0.0 & 62.52 & 5.826e-22 & lte032.0-5.0-0.0a+0.0.BT-Settl.spec.7.dat sdM5.5 & FF & 3200 & 5.5 & -1.0 & 60.94 & 2.764e-22 & lte032.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat sdM6.0 & FF & 3200 & 5.5 & -1.0 & 52.62 & 3.327e-22 & lte032.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat sdM6.5 & FF & 2900 & 5.5 & -0.0 & 828.4 & 4.589e-21 & lte029.0-5.5-0.0a+0.0.BT-Settl.spec.7.dat sdM8.0 & FF & 2900 & 5.5 & -2.0 & 49.54 & 3.094e-22 & lte029.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat sdM9.5 & FF & 2800 & 5.0 & -2.0 & 26.84 & 1.071e-21 & lte028.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat esdM0.0 & FF & 3800 & 5.0 & -1.0 & 46.36 & 1.789e-22 & lte038.0-5.0-1.0a+0.4.BT-Settl.spec.7.dat esdM0.5 & FF & 3700 & 5.0 & -1.5 & 28.12 & 4.847e-22 & lte037.0-5.0-1.5a+0.4.BT-Settl.spec.7.dat esdM1.0 & FF & 3800 & 5.5 & -0.5 & 50.66 & 1.683e-22 & lte038.0-5.5-0.5a+0.2.BT-Settl.spec.7.dat esdM1.5 & FF & 3600 & 5.0 & -1.0 & 24.61 & 6.996e-22 & lte036.0-5.0-1.0a+0.4.BT-Settl.spec.7.dat esdM2.0 & FF & 3600 & 5.5 & -1.0 & 54.63 & 2.229e-22 & lte036.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat esdM3.0 & FF & 3500 & 5.5 & -1.5 & 82.5 & 6.294e-22 & lte035.0-5.5-1.5a+0.4.BT-Settl.spec.7.dat esdM3.5 & FF & 3500 & 5.5 & -1.5 & 38 & 1.883e-22 & lte035.0-5.5-1.5a+0.4.BT-Settl.spec.7.dat esdM4.0 & FF & 3400 & 5.5 & -1.5 & 53.24 & 2.04e-22 & lte034.0-5.5-1.5a+0.4.BT-Settl.spec.7.dat esdM4.5 & FF & 3400 & 5.5 & -1.0 & 260.6 & 3.594e-21 & lte034.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat esdM5.0 & FF & 3200 & 4.5 & -1.5 & 39.72 & 1.952e-22 & lte032.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat esdM5.5 & FF & 3300 & 5.5 & -2.0 & 87.47 & 1.533e-21 & lte033.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat esdM6.0 & FF & 3300 & 5.5 & -2.0 & 27.35 & 8.065e-23 & lte033.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat esdM6.5 & FF & 3300 & 5.5 & -2.0 & 84.99 & 2.61e-22 & lte033.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat esdM7.0 & FF & 3200 & 5.5 & -0.0 & 8.431 & 8.349e-23 & lte032.0-5.5-0.0a+0.0.BT-Settl.spec.7.dat esdM7.5 & FF & 3000 & 5.0 & -2.0 & 75.73 & 8.239e-22 & lte030.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat esdM8.5 & FF & 3000 & 5.5 & -2.0 & 46.61 & 3.916e-22 & lte030.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM0.0 & FF & 3500 & 4.5 & -2.0 & 81.52 & 2.637e-22 & lte035.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat usdM0.5 & FF & 3700 & 5.5 & -2.5 & 35.49 & 8.444e-23 & lte037.0-5.5-2.5a+0.4.BT-Settl.spec.7.dat usdM1.0 & FF & 3700 & 5.5 & -1.0 & 34.9 & 1.245e-22 & lte037.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat usdM2.5 & FF & 3600 & 5.5 & -1.5 & 40.94 & 1.724e-22 & lte036.0-5.5-1.5a+0.4.BT-Settl.spec.7.dat usdM3.0 & FF & 3800 & 6.0 & -0.5 & 40.92 & 9.605e-23 & lte038.0-6.0-0.5a+0.2.BT-Settl.spec.7.dat usdM4.0 & FF & 3200 & 4.5 & -1.5 & 8.887 & 2.545e-22 & lte032.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat usdM4.5 & FF & 3400 & 5.5 & -2.0 & 79.52 & 2.428e-22 & lte034.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM5.0 & FF & 3500 & 5.5 & -2.5 & 128.5 & 5.667e-22 & lte035.0-5.5-2.5a+0.4.BT-Settl.spec.7.dat usdM5.5 & FF & 3300 & 4.5 & -2.0 & 12.52 & 5.669e-23 & lte033.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat usdM6.0 & FF & 3300 & 5.5 & -2.0 & 40.62 & 4.176e-22 & lte033.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM7.5 & FF & 3100 & 5.5 & -2.5 & 25.7 & 3.749e-22 & lte031.0-5.5-2.5a+0.4.BT-Settl.spec.7.dat usdM8.5 & FF & 3100 & 5.5 & -2.5 & 49.69 & 4.503e-22 & lte031.0-5.5-2.5a+0.4.BT-Settl.spec.7.dat \[tab\_MplusTeff\_sdM:table\_best\_fits\_FF\]
[c c c c c c c c]{} SpT & Range & T$_{\rm eff}$ & $\log$g & M/H & chi$^{2}$ & Factor & Model\
sdM0.0 & LL & 3600 & 5.5 & -2.5 & 1.274 & 2.182e-22 & lte036.0-5.5-2.5a+0.4.BT-Settl.spec.7.dat sdM0.5 & LL & 3500 & 4.5 & -1.5 & 65.2 & 2.303e-21 & lte035.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat sdM1.0 & LL & 3600 & 5.0 & -0.5 & 44.52 & 1.843e-21 & lte036.0-5.0-0.5a+0.2.BT-Settl.spec.7.dat sdM1.5 & LL & 3600 & 5.0 & -0.5 & 53.5 & 2.944e-22 & lte036.0-5.0-0.5a+0.2.BT-Settl.spec.7.dat sdM2.0 & LL & 3300 & 4.5 & -1.5 & 54.06 & 1.418e-21 & lte033.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat sdM2.5 & LL & 3400 & 5.0 & -1.0 & 33.27 & 1.013e-21 & lte034.0-5.0-1.0a+0.4.BT-Settl.spec.7.dat sdM3.0 & LL & 3500 & 5.0 & -0.0 & 86.06 & 2.228e-21 & lte035.0-5.0-0.0a+0.0.BT-Settl.spec.7.dat sdM3.5 & LL & 3300 & 5.0 & -1.0 & 100 & 4.559e-21 & lte033.0-5.0-1.0a+0.4.BT-Settl.spec.7.dat sdM4.0 & LL & 3300 & 5.0 & -1.0 & 43 & 1.854e-21 & lte033.0-5.0-1.0a+0.4.BT-Settl.spec.7.dat sdM4.5 & LL & 3200 & 5.0 & -1.5 & 64.01 & 2.467e-22 & lte032.0-5.0-1.5a+0.4.BT-Settl.spec.7.dat sdM5.0 & LL & 3200 & 5.5 & -0.5 & 64.17 & 5.346e-22 & lte032.0-5.5-0.5a+0.2.BT-Settl.spec.7.dat sdM5.5 & LL & 3100 & 5.0 & -1.5 & 62.38 & 3.231e-22 & lte031.0-5.0-1.5a+0.4.BT-Settl.spec.7.dat sdM6.0 & LL & 3000 & 4.5 & -1.5 & 68.92 & 4.55e-22 & lte030.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat sdM6.5 & LL & 2600 & 4.5 & -2.5 & 360.4 & 4.65e-21 & lte026.0-4.5-2.5a+0.4.BT-Settl.spec.7.dat sdM8.0 & LL & 2800 & 5.0 & -2.5 & 68.93 & 3.399e-22 & lte028.0-5.0-2.5a+0.4.BT-Settl.spec.7.dat sdM9.5 & LL & 2700 & 5.0 & -2.5 & 33.25 & 1.083e-21 & lte027.0-5.0-2.5a+0.4.BT-Settl.spec.7.dat esdM0.0 & LL & 3800 & 5.0 & -1.0 & 52.51 & 1.807e-22 & lte038.0-5.0-1.0a+0.4.BT-Settl.spec.7.dat esdM0.5 & LL & 3600 & 5.0 & -2.0 & 36.42 & 5.423e-22 & lte036.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat esdM1.0 & LL & 3500 & 4.5 & -2.0 & 43.43 & 2.508e-22 & lte035.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM1.5 & LL & 3400 & 4.5 & -2.0 & 32.73 & 9.175e-22 & lte034.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM2.0 & LL & 3300 & 4.5 & -2.0 & 41.6 & 3.457e-22 & lte033.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM3.0 & LL & 3400 & 5.0 & -1.5 & 54.27 & 7.476e-22 & lte034.0-5.0-1.5a+0.4.BT-Settl.spec.7.dat esdM3.5 & LL & 3400 & 5.0 & -2.0 & 30.57 & 2.211e-22 & lte034.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat esdM4.0 & LL & 3300 & 5.0 & -1.5 & 60.87 & 2.371e-22 & lte033.0-5.0-1.5a+0.4.BT-Settl.spec.7.dat esdM4.5 & LL & 3100 & 4.5 & -2.0 & 188.1 & 5.634e-21 & lte031.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM5.0 & LL & 3100 & 4.5 & -2.0 & 36.16 & 2.286e-22 & lte031.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM5.5 & LL & 3200 & 5.0 & -2.0 & 65.67 & 1.831e-21 & lte032.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat esdM6.0 & LL & 3100 & 4.5 & -2.0 & 29.34 & 1.116e-22 & lte031.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM6.5 & LL & 3200 & 5.0 & -2.0 & 62.31 & 3.104e-22 & lte032.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat esdM7.0 & LL & 3000 & 4.5 & -2.0 & 2.347 & 9.268e-23 & lte030.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM7.5 & LL & 3000 & 5.0 & -2.0 & 93.43 & 8.48e-22 & lte030.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat esdM8.5 & LL & 2900 & 5.0 & -2.5 & 54.67 & 4.499e-22 & lte029.0-5.0-2.5a+0.4.BT-Settl.spec.7.dat usdM0.0 & LL & 3600 & 5.5 & -2.5 & 52.74 & 2.186e-22 & lte036.0-5.5-2.5a+0.4.BT-Settl.spec.7.dat usdM0.5 & LL & 3700 & 5.5 & -2.0 & 52.29 & 8.588e-23 & lte037.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM1.0 & LL & 3600 & 5.0 & -2.0 & 38.42 & 1.449e-22 & lte036.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat usdM2.5 & LL & 3500 & 5.0 & -2.0 & 35.32 & 1.999e-22 & lte035.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat usdM3.0 & LL & 3600 & 5.5 & -2.0 & 38.99 & 1.27e-22 & lte036.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM4.0 & LL & 3400 & 5.0 & -2.0 & 0.7288 & 1.663e-22 & lte034.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat usdM4.5 & LL & 3300 & 5.0 & -2.0 & 81.28 & 2.949e-22 & lte033.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat usdM5.0 & LL & 3500 & 5.5 & -2.0 & 91.84 & 5.981e-22 & lte035.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM5.5 & LL & 3500 & 5.5 & -2.0 & 11.58 & 3.992e-23 & lte035.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM6.0 & LL & 3200 & 5.0 & -2.0 & 44.5 & 5.027e-22 & lte032.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat usdM7.5 & LL & 3000 & 5.0 & -2.0 & 39.86 & 4.725e-22 & lte030.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat usdM8.5 & LL & 3000 & 5.5 & -2.5 & 60.58 & 5.247e-22 & lte030.0-5.5-2.5a+0.4.BT-Settl.spec.7.dat \[tab\_MplusTeff\_sdM:table\_best\_fits\_LL\]
[c c c c c c c c]{} SpT & Range & T$_{\rm eff}$ & $\log$g & M/H & chi$^{2}$ & Factor & Model\
sdM0.0 & FL & 3600 & 5.5 & -2.5 & 1.274 & 2.182e-22 & lte036.0-5.5-2.5a+0.4.BT-Settl.spec.7.dat sdM0.5 & FL & 3700 & 5.0 & -0.5 & 67.14 & 1.791e-21 & lte037.0-5.0-0.5a+0.2.BT-Settl.spec.7.dat sdM1.0 & FL & 3700 & 5.0 & -0.5 & 45.51 & 1.627e-21 & lte037.0-5.0-0.5a+0.2.BT-Settl.spec.7.dat sdM1.5 & FL & 3600 & 5.0 & -0.5 & 53.5 & 2.944e-22 & lte036.0-5.0-0.5a+0.2.BT-Settl.spec.7.dat sdM2.0 & FL & 3600 & 5.0 & -0.0 & 60.82 & 9.902e-22 & lte036.0-5.0-0.0a+0.0.BT-Settl.spec.7.dat sdM2.5 & FL & 3600 & 5.0 & -0.0 & 35.7 & 8.194e-22 & lte036.0-5.0-0.0a+0.0.BT-Settl.spec.7.dat sdM3.0 & FL & 3500 & 5.0 & -0.0 & 86.06 & 2.228e-21 & lte035.0-5.0-0.0a+0.0.BT-Settl.spec.7.dat sdM3.5 & FL & 3400 & 5.0 & -0.0 & 102.6 & 4.24e-21 & lte034.0-5.0-0.0a+0.0.BT-Settl.spec.7.dat sdM4.0 & FL & 3400 & 5.0 & -0.5 & 49.18 & 1.64e-21 & lte034.0-5.0-0.5a+0.2.BT-Settl.spec.7.dat sdM4.5 & FL & 3300 & 5.5 & -1.0 & 65.18 & 2.143e-22 & lte033.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat sdM5.0 & FL & 3200 & 5.5 & -0.5 & 64.17 & 5.346e-22 & lte032.0-5.5-0.5a+0.2.BT-Settl.spec.7.dat sdM5.5 & FL & 3200 & 5.5 & -1.0 & 64.82 & 2.837e-22 & lte032.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat sdM6.0 & FL & 3200 & 5.5 & -1.0 & 71.23 & 3.416e-22 & lte032.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat sdM6.5 & FL & 2900 & 5.5 & -1.0 & 457.5 & 3.619e-21 & lte029.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat sdM8.0 & FL & 2900 & 5.5 & -2.0 & 70.63 & 3.246e-22 & lte029.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat sdM9.5 & FL & 2800 & 5.5 & -2.0 & 36.7 & 1.048e-21 & lte028.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat esdM0.0 & FL & 3800 & 5.0 & -1.0 & 52.51 & 1.807e-22 & lte038.0-5.0-1.0a+0.4.BT-Settl.spec.7.dat esdM0.5 & FL & 3700 & 5.5 & -2.0 & 36.68 & 4.752e-22 & lte037.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat esdM1.0 & FL & 3800 & 5.5 & -1.0 & 51.58 & 1.689e-22 & lte038.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat esdM1.5 & FL & 3600 & 5.5 & -1.0 & 35.37 & 6.833e-22 & lte036.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat esdM2.0 & FL & 3600 & 5.5 & -1.0 & 48.16 & 2.251e-22 & lte036.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat esdM3.0 & FL & 3500 & 5.5 & -1.5 & 61.37 & 6.495e-22 & lte035.0-5.5-1.5a+0.4.BT-Settl.spec.7.dat esdM3.5 & FL & 3500 & 5.5 & -2.0 & 34.19 & 1.914e-22 & lte035.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat esdM4.0 & FL & 3400 & 5.5 & -1.0 & 67.85 & 2.034e-22 & lte034.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat esdM4.5 & FL & 3400 & 5.5 & -1.0 & 211.3 & 3.624e-21 & lte034.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat esdM5.0 & FL & 3200 & 5.0 & -1.5 & 40.53 & 1.967e-22 & lte032.0-5.0-1.5a+0.4.BT-Settl.spec.7.dat esdM5.5 & FL & 3300 & 5.5 & -1.5 & 74.79 & 1.598e-21 & lte033.0-5.5-1.5a+0.4.BT-Settl.spec.7.dat esdM6.0 & FL & 3300 & 5.5 & -2.5 & 31.14 & 8.05e-23 & lte033.0-5.5-2.5a+0.4.BT-Settl.spec.7.dat esdM6.5 & FL & 3300 & 5.5 & -1.5 & 72.68 & 2.71e-22 & lte033.0-5.5-1.5a+0.4.BT-Settl.spec.7.dat esdM7.0 & FL & 3200 & 5.5 & -2.5 & 2.51 & 6.41e-23 & lte032.0-5.5-2.5a+0.4.BT-Settl.spec.7.dat esdM7.5 & FL & 3000 & 5.0 & -2.0 & 93.43 & 8.48e-22 & lte030.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat esdM8.5 & FL & 3000 & 5.5 & -2.0 & 54.82 & 4.189e-22 & lte030.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM0.0 & FL & 3500 & 4.5 & -2.0 & 54.08 & 2.592e-22 & lte035.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat usdM0.5 & FL & 3700 & 5.5 & -2.0 & 52.29 & 8.588e-23 & lte037.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM1.0 & FL & 3700 & 5.5 & -2.0 & 38.67 & 1.27e-22 & lte037.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM2.5 & FL & 3600 & 5.5 & -2.0 & 35.69 & 1.745e-22 & lte036.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM3.0 & FL & 3800 & 5.5 & -0.5 & 46.05 & 1.005e-22 & lte038.0-5.5-0.5a+0.2.BT-Settl.spec.7.dat usdM4.0 & FL & 3200 & 4.5 & -2.0 & 0.772 & 2.22e-22 & lte032.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat usdM4.5 & FL & 3400 & 5.5 & -2.0 & 85.36 & 2.539e-22 & lte034.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM5.0 & FL & 3500 & 5.5 & -2.0 & 91.84 & 5.981e-22 & lte035.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM5.5 & FL & 3300 & 5.0 & -2.0 & 11.84 & 5.247e-23 & lte033.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat usdM6.0 & FL & 3300 & 5.5 & -2.0 & 47.7 & 4.314e-22 & lte033.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM7.5 & FL & 3100 & 5.5 & -2.0 & 40.37 & 4.046e-22 & lte031.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM8.5 & FL & 3100 & 5.5 & -2.0 & 72.9 & 4.879e-22 & lte031.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat \[tab\_MplusTeff\_sdM:table\_best\_fits\_FL\]
[c c c c c c c c]{} SpT & Range & T$_{\rm eff}$ & $\log$g & M/H & chi$^{2}$ & Factor & Model\
sdM0.0 & LF & 3700 & 5.5 & -2.5 & 3.692 & 2.021e-22 & lte037.0-5.5-2.5a+0.4.BT-Settl.spec.7.dat sdM0.5 & LF & 3600 & 4.5 & -1.5 & 87.73 & 1.999e-21 & lte036.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat sdM1.0 & LF & 3700 & 5.0 & -0.5 & 67.22 & 1.609e-21 & lte037.0-5.0-0.5a+0.2.BT-Settl.spec.7.dat sdM1.5 & LF & 3500 & 5.0 & -0.5 & 93.49 & 3.502e-22 & lte035.0-5.0-0.5a+0.2.BT-Settl.spec.7.dat sdM2.0 & LF & 3500 & 4.5 & -1.5 & 60.84 & 1.051e-21 & lte035.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat sdM2.5 & LF & 3600 & 5.0 & -1.0 & 37.11 & 7.791e-22 & lte036.0-5.0-1.0a+0.4.BT-Settl.spec.7.dat sdM3.0 & LF & 3500 & 5.0 & -0.0 & 88.51 & 2.208e-21 & lte035.0-5.0-0.0a+0.0.BT-Settl.spec.7.dat sdM3.5 & LF & 3400 & 5.0 & -1.0 & 169.9 & 3.924e-21 & lte034.0-5.0-1.0a+0.4.BT-Settl.spec.7.dat sdM4.0 & LF & 3400 & 5.0 & -1.0 & 58.22 & 1.54e-21 & lte034.0-5.0-1.0a+0.4.BT-Settl.spec.7.dat sdM4.5 & LF & 3300 & 5.0 & -1.5 & 66.55 & 2.115e-22 & lte033.0-5.0-1.5a+0.4.BT-Settl.spec.7.dat sdM5.0 & LF & 3100 & 5.5 & -0.5 & 98.69 & 6.395e-22 & lte031.0-5.5-0.5a+0.2.BT-Settl.spec.7.dat sdM5.5 & LF & 3100 & 5.0 & -1.5 & 66.45 & 3.139e-22 & lte031.0-5.0-1.5a+0.4.BT-Settl.spec.7.dat sdM6.0 & LF & 3100 & 4.5 & -1.5 & 66.98 & 3.779e-22 & lte031.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat sdM6.5 & LF & 2800 & 4.5 & -2.5 & 8854 & 5.129e-21 & lte028.0-4.5-2.5a+0.4.BT-Settl.spec.7.dat sdM8.0 & LF & 3000 & 5.0 & -2.5 & 83.22 & 2.962e-22 & lte030.0-5.0-2.5a+0.4.BT-Settl.spec.7.dat sdM9.5 & LF & 3000 & 5.0 & -2.5 & 63.45 & 8.95e-22 & lte030.0-5.0-2.5a+0.4.BT-Settl.spec.7.dat esdM0.0 & LF & 3800 & 5.0 & -1.0 & 46.36 & 1.789e-22 & lte038.0-5.0-1.0a+0.4.BT-Settl.spec.7.dat esdM0.5 & LF & 3700 & 5.0 & -2.0 & 29.55 & 4.919e-22 & lte037.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat esdM1.0 & LF & 3600 & 4.5 & -2.0 & 72.12 & 2.124e-22 & lte036.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM1.5 & LF & 3500 & 4.5 & -2.0 & 29.95 & 7.937e-22 & lte035.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM2.0 & LF & 3500 & 4.5 & -2.0 & 91.63 & 2.571e-22 & lte035.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM3.0 & LF & 3500 & 5.0 & -1.5 & 97.21 & 6.376e-22 & lte035.0-5.0-1.5a+0.4.BT-Settl.spec.7.dat esdM3.5 & LF & 3400 & 5.0 & -2.0 & 44.81 & 2.134e-22 & lte034.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat esdM4.0 & LF & 3400 & 5.0 & -1.5 & 55 & 2.057e-22 & lte034.0-5.0-1.5a+0.4.BT-Settl.spec.7.dat esdM4.5 & LF & 3200 & 4.5 & -2.0 & 439 & 4.69e-21 & lte032.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM5.0 & LF & 3200 & 4.5 & -2.0 & 43.03 & 1.99e-22 & lte032.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM5.5 & LF & 3200 & 5.0 & -2.0 & 92.35 & 1.738e-21 & lte032.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat esdM6.0 & LF & 3200 & 4.5 & -2.0 & 30.38 & 9.303e-23 & lte032.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM6.5 & LF & 3200 & 5.0 & -2.0 & 88.16 & 2.966e-22 & lte032.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat esdM7.0 & LF & 3000 & 4.5 & -2.0 & 10.14 & 1.032e-22 & lte030.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM7.5 & LF & 3000 & 5.0 & -2.0 & 75.73 & 8.239e-22 & lte030.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat esdM8.5 & LF & 3000 & 5.0 & -2.5 & 50.75 & 4.173e-22 & lte030.0-5.0-2.5a+0.4.BT-Settl.spec.7.dat usdM0.0 & LF & 3600 & 5.5 & -2.5 & 232.8 & 2.315e-22 & lte036.0-5.5-2.5a+0.4.BT-Settl.spec.7.dat usdM0.5 & LF & 3800 & 5.5 & -2.0 & 36.52 & 7.582e-23 & lte038.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM1.0 & LF & 3600 & 5.0 & -2.0 & 39.32 & 1.403e-22 & lte036.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat usdM2.5 & LF & 3500 & 5.0 & -2.0 & 45.49 & 1.938e-22 & lte035.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat usdM3.0 & LF & 3600 & 5.5 & -2.0 & 43.27 & 1.199e-22 & lte036.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM4.0 & LF & 3300 & 5.0 & -2.0 & 10.1 & 2.222e-22 & lte033.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat usdM4.5 & LF & 3400 & 5.0 & -2.0 & 103 & 2.461e-22 & lte034.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat usdM5.0 & LF & 3500 & 5.5 & -2.0 & 206.1 & 5.591e-22 & lte035.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM5.5 & LF & 3400 & 5.5 & -2.0 & 13.36 & 4.883e-23 & lte034.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM6.0 & LF & 3300 & 5.0 & -2.0 & 49.56 & 4.229e-22 & lte033.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat usdM7.5 & LF & 3100 & 5.0 & -2.0 & 62.7 & 3.727e-22 & lte031.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat usdM8.5 & LF & 3100 & 5.5 & -2.5 & 49.69 & 4.503e-22 & lte031.0-5.5-2.5a+0.4.BT-Settl.spec.7.dat \[tab\_MplusTeff\_sdM:table\_best\_fits\_LF\]
[c c c c c c c c]{} SpT & Range & T$_{\rm eff}$ & $\log$g & M/H & chi$^{2}$ & Factor & Model\
sdM0.0 & VV & 3500 & 4.5 & -2.5 & 1.506 & 2.581e-22 & lte035.0-4.5-2.5a+0.4.BT-Settl.spec.7.dat sdM0.5 & VV & 3700 & 4.5 & -1.0 & 72.9 & 1.816e-21 & lte037.0-4.5-1.0a+0.4.BT-Settl.spec.7.dat sdM1.0 & VV & 3700 & 5.0 & -0.5 & 66.12 & 1.636e-21 & lte037.0-5.0-0.5a+0.2.BT-Settl.spec.7.dat sdM1.5 & VV & 3400 & 4.5 & -1.5 & 48.5 & 3.784e-22 & lte034.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat sdM2.0 & VV & 3500 & 4.5 & -1.0 & 48.26 & 1.082e-21 & lte035.0-4.5-1.0a+0.4.BT-Settl.spec.7.dat sdM2.5 & VV & 3500 & 4.5 & -1.0 & 35.1 & 9.046e-22 & lte035.0-4.5-1.0a+0.4.BT-Settl.spec.7.dat sdM3.0 & VV & 3500 & 5.0 & -0.0 & 89.26 & 2.224e-21 & lte035.0-5.0-0.0a+0.0.BT-Settl.spec.7.dat sdM3.5 & VV & 3400 & 5.0 & -0.5 & 153.6 & 4.033e-21 & lte034.0-5.0-0.5a+0.2.BT-Settl.spec.7.dat sdM4.0 & VV & 3400 & 5.5 & -0.5 & 47.24 & 1.627e-21 & lte034.0-5.5-0.5a+0.2.BT-Settl.spec.7.dat sdM4.5 & VV & 3100 & 4.5 & -1.5 & 64.16 & 2.845e-22 & lte031.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat sdM5.0 & VV & 3200 & 5.5 & -0.0 & 57.52 & 5.62e-22 & lte032.0-5.5-0.0a+0.0.BT-Settl.spec.7.dat sdM5.5 & VV & 3200 & 5.5 & -1.0 & 69.85 & 2.79e-22 & lte032.0-5.5-1.0a+0.4.BT-Settl.spec.7.dat sdM6.0 & VV & 3000 & 4.5 & -1.5 & 60.23 & 4.532e-22 & lte030.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat sdM6.5 & VV & 2900 & 5.5 & -0.0 & 367.4 & 4.09e-21 & lte029.0-5.5-0.0a+0.0.BT-Settl.spec.7.dat sdM8.0 & VV & 2800 & 5.0 & -2.5 & 55.29 & 3.348e-22 & lte028.0-5.0-2.5a+0.4.BT-Settl.spec.7.dat sdM9.5 & VV & 2600 & 4.5 & -2.5 & 28.39 & 1.454e-21 & lte026.0-4.5-2.5a+0.4.BT-Settl.spec.7.dat esdM0.0 & VV & 3600 & 4.5 & -1.5 & 51.33 & 2.339e-22 & lte036.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat esdM0.5 & VV & 3600 & 4.5 & -1.5 & 33.95 & 5.464e-22 & lte036.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat esdM1.0 & VV & 3600 & 4.5 & -1.5 & 54.39 & 2.181e-22 & lte036.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat esdM1.5 & VV & 3400 & 4.5 & -1.5 & 31.73 & 9.081e-22 & lte034.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat esdM2.0 & VV & 3400 & 4.5 & -1.5 & 65.67 & 2.973e-22 & lte034.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat esdM3.0 & VV & 3400 & 5.0 & -1.5 & 81.18 & 7.466e-22 & lte034.0-5.0-1.5a+0.4.BT-Settl.spec.7.dat esdM3.5 & VV & 3400 & 5.0 & -1.5 & 43.41 & 2.206e-22 & lte034.0-5.0-1.5a+0.4.BT-Settl.spec.7.dat esdM4.0 & VV & 3200 & 4.5 & -2.0 & 61.63 & 2.746e-22 & lte032.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM4.5 & VV & 3200 & 4.5 & -1.5 & 326.9 & 4.852e-21 & lte032.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat esdM5.0 & VV & 3200 & 4.5 & -1.5 & 45.1 & 1.963e-22 & lte032.0-4.5-1.5a+0.4.BT-Settl.spec.7.dat esdM5.5 & VV & 3100 & 4.5 & -2.0 & 88.56 & 2.14e-21 & lte031.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM6.0 & VV & 3200 & 5.0 & -2.0 & 29.11 & 9.46e-23 & lte032.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat esdM6.5 & VV & 3100 & 4.5 & -2.0 & 86.41 & 3.646e-22 & lte031.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM7.0 & VV & 3000 & 4.5 & -2.0 & 5.132 & 9.173e-23 & lte030.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat esdM7.5 & VV & 3000 & 5.0 & -2.0 & 89.47 & 8.325e-22 & lte030.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat esdM8.5 & VV & 2800 & 4.5 & -2.5 & 52.36 & 5.534e-22 & lte028.0-4.5-2.5a+0.4.BT-Settl.spec.7.dat usdM0.0 & VV & 3500 & 4.5 & -2.5 & 51.01 & 2.583e-22 & lte035.0-4.5-2.5a+0.4.BT-Settl.spec.7.dat usdM0.5 & VV & 3600 & 5.0 & -2.5 & 42.57 & 9.734e-23 & lte036.0-5.0-2.5a+0.4.BT-Settl.spec.7.dat usdM1.0 & VV & 3600 & 5.0 & -1.5 & 40.24 & 1.445e-22 & lte036.0-5.0-1.5a+0.4.BT-Settl.spec.7.dat usdM2.5 & VV & 3400 & 4.5 & -2.0 & 50.77 & 2.315e-22 & lte034.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat usdM3.0 & VV & 3400 & 4.5 & -2.0 & 40.27 & 1.684e-22 & lte034.0-4.5-2.0a+0.4.BT-Settl.spec.7.dat usdM4.0 & VV & 3500 & 5.5 & -2.0 & 2.578 & 1.425e-22 & lte035.0-5.5-2.0a+0.4.BT-Settl.spec.7.dat usdM4.5 & VV & 3300 & 5.0 & -2.0 & 83.97 & 2.923e-22 & lte033.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat usdM5.0 & VV & 3400 & 5.5 & -2.5 & 131.2 & 6.592e-22 & lte034.0-5.5-2.5a+0.4.BT-Settl.spec.7.dat usdM5.5 & VV & 3800 & 6.0 & -0.5 & 16.71 & 2.727e-23 & lte038.0-6.0-0.5a+0.2.BT-Settl.spec.7.dat usdM6.0 & VV & 3200 & 5.0 & -2.0 & 44.45 & 4.976e-22 & lte032.0-5.0-2.0a+0.4.BT-Settl.spec.7.dat usdM7.5 & VV & 3000 & 5.0 & -2.5 & 28.43 & 4.579e-22 & lte030.0-5.0-2.5a+0.4.BT-Settl.spec.7.dat usdM8.5 & VV & 3000 & 5.5 & -2.5 & 51.61 & 5.169e-22 & lte030.0-5.5-2.5a+0.4.BT-Settl.spec.7.dat \[tab\_MplusTeff\_sdM:table\_best\_fits\_VV\]
[^1]: Based on observations collected at the European Southern Observatory, Chile, under programmes 089.C-0140(A), 091.C-0264(A), 092.D-0600(A), and 093.C-0610(A)
[^2]: http://www.eso.org/sci/software/pipelines/
[^3]: http://www.eso.org/sci/software/pipelines/skytools/molecfit
[^4]: http://perso.ens-lyon.fr/france.allard/
[^5]: phoenix.ens-lyon.fr/Grids/BT-Settl/CIFIST2011/SPECTRA/
[^6]: http://astro.uni-tuebingen.de/software/idl/aitlib/misc/gaussfold.html
[^7]: last update on 10 October 2016 at exoplanetarchive.ipac.caltech.edu
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In the framework of the $k_T$-factorization approach we have calculated in the fragmentation model the $p_T$-spectra of $B_c$ mesons at the energies of the Tevatron and the LHC Colliders and at the large $p_T$ domain. We compare the obtained results with the existing experimental data and with the predictions obtained in the collinear parton model.'
---
[**Production of $B_c$ mesons via fragmentation in the $k_T$-factorization approach**]{}
[*V.A. Saleev*]{}[^1]\
[Samara State University, 443011 Samara, Russia]{}\
and\
[*D.V. Vasin*]{}[^2]\
[II. Institut f" ur Theoretische Physik, Universit" at Hamburg, 22761 Hamburg, Germany[^3]]{}\
Introduction
============
Recently, $B_c$ mesons were observed experimentally [@CDF1; @CDF2]. These particles have a special place in the doubly heavy meson family, because they consist of quarks of different flavors. The $B_c$ meson spectroscopy has the same features as the spectroscopies of charmonia and bottomonia, however the decay and production properties of $B_c$ mesons are unique. There is no annihilation channel for $B_c$ mesons into hadrons via the gluons and, consequently, the lifetime of the ground state of the $\bar b c$ system is large, given by $\tau \approx 0.46$ ps [@CDF2]. The main channel of $B_c$ meson decay occurs through the weak decays of the $b$-quark or the $c$-quark, which are respectively about 25% and 65% of the total decay width.
The $B_c$ meson hadroproduction cross section is approximately 100 times smaller than the ($b\bar b$)-pair production cross section, because the two pairs of the heavy quarks are produced in a parton subprocess. The mass spectrum and the decay properties of the $B_c$ mesons are studied in detail in the framework of potential quark models [@QPM], in the QCD sum rules methods [@SRQCD] and in the OPE approach [@OPE]. The predictions at leading order in $\alpha_s$ for the $B_c$ meson production rates in the $\gamma \gamma$, $\gamma p$ and $pp$ interactions were obtained in the collinear parton model [@Likhoded; @Ruckl; @Chang; @Baranov; @MS].
The $B_c$ meson production cross section became large enough to study experimental ($\sim 1$ nb) only in the region of very high energy or in the region of very small $x\sim M/\sqrt{s}\sim
10^{-2}$, where $x$ is the argument of the gluon distribution function in a proton (the quark contribution to the hadronic production of $B_c$ mesons is very small). In this region the $k_T$-factorization approach [@Amp; @KT] is more adequate for describing the perturbative evolution of the gluon distribution function, which satisfies the BFKL [@BFKL] or CCFM [@CCFM] evolution equations, in comparison with the collinear parton model, which is based on the DGLAP equation [@DGLAP].
In this Latter we have calculated $B_c$ meson $p_T$-spectra at the energy range of the Tevatron and the LHC Colliders in the fragmentation model and in the framework of the $k_T$-factorization approach. In the fragmentation model we have taken into consideration only the main contribution originating from fragmentation of a b-quark into a $B_c$ meson [@Cheung].
The $k_T$-factorization approach
================================
In the $k_T$-factorization approach [@Amp; @KT], which generalizes the collinear parton model to the region of small $x$, the hadronic and partonic cross sections are related as follows: $$\begin{aligned}
&&\sigma^{\mathrm{KT}}(p \bar p\to b \bar b X,
s)=\int{\frac{dx_1}{x_1}}\int{d|\vec
k_{1T}|^2}\int{\frac{d\varphi_1}{2\pi}} \Phi(x_1,|\vec
k_{1T}|^2,\mu^2)\times\nonumber\\&&\times\int{\frac{dx_2}{x_2}}
\int{d |\vec k_{2T}|^2}\int{\frac{d\varphi_2}{2\pi}}
\Phi(x_2,|\vec k_{2T}|^2,\mu^2)\hat\sigma (g^\star g^\star \to b
\bar b, \hat s)
\mbox{,}\end{aligned}$$ where $\Phi(x,|\vec k_T|^2,\mu^2)$ is the unintegrated gluon distribution function in a proton (unintegrated refers to the transverse momentum), $|{\vec k}_{i,T}|^2$ is the virtuality of the initial reggeized gluon, $k_{i}=x_{i} p_{i}+k_{i,T}$ is the 4-momentum of the initial gluon, $k_{i,T}=(0,\vec k_{i,T},0)$ is the transverse momentum of the initial gluon, $\varphi_i$ is the angle between the gluon transverse momentum and the fixed axis $OX$ in the plane $XOY$, $\hat s=x_1 x_2 s-|\vec k_{1T}|^2-|\vec k_{2T}|^2$. In the numerical calculations we have used the following parameterizations for the unintegrated gluon distribution function in a proton: JB by Blumlein [@JB], JS by Jung and Salam [@JS] and KMR by Kimber, Martin and Ryskin [@KMR].
To calculate the amplitude of partonic processes in the $k_T$-factorization approach the polarization vector for the reggeized gluon in the initial state is taken to be [@Amp; @KT] $$\begin{aligned}
\varepsilon^\mu(k_T)=\frac{k_T^\mu}{|\vec k_T|}\mbox{.}\end{aligned}$$ The squared amplitude for one of the processes in $B_c$ meson production in the $k_T$-factorization approach, namely $$\begin{aligned}
g^\star+g^\star \to b + \bar b\end{aligned}$$ was obtained earlier in [@Amp]. Here we use our original results from Ref. [@SVpub] $$\begin{aligned}
\overline{|M(g^\star+g^\star\to Q +\bar Q)|^2}=8 \pi^2 \alpha_s^2 (M_{11}+M_{22}+M_{33}+2 M_{12}+2
M_{13}+2 M_{23})\mbox{,}\end{aligned}$$ where $$\begin{aligned}
M_{11}&=&\frac{1}{3 {\tilde t}^2}
\Bigl(
\kappa_1 (4 \beta_1 - \kappa_1)(\kappa_1^2 + \kappa_2^2 - 4 \beta_2 \kappa_2)
+ 4 \lambda \beta_2 (\kappa_1 - 2 \beta_1)(4 \kappa_2^2 + {\tilde t} )
+ 4 \kappa_1^3 (\lambda \kappa_2 - \beta_1)
\nonumber\\&+&
4 \beta_2^2
(\kappa_1^2 + 4 \beta_1 \kappa_1 - 4 \lambda \kappa_1 \kappa_2)
+ 4 \lambda \kappa_2 (\lambda \kappa_2 - 2 \beta_1)(\kappa_1^2 -4 \beta_2^2)
- 4 \beta_1^2 (2 \beta_2 - \kappa_2)^2
+ \kappa_1^4
\nonumber\\&+&
4 \lambda \kappa_2 (2 \lambda \beta_2 + \beta_1)(2 \kappa_2^2 + {\tilde t})
+ 8 \lambda \beta_1 \kappa_2^3
+ 4 \lambda \kappa_2 ({\hat s} + {\tilde u})(\kappa_1 + \lambda \kappa_2)
+ {\tilde t} {\tilde u} \Bigr)\mbox{,}\end{aligned}$$ $$\begin{aligned}
M_{22}&=&\frac{1}{3 {\tilde u}^2}
\Bigl(\kappa_2^4 - 4 \beta_2 \kappa_2^3 - 4 \beta_2^2 (\kappa_1^2+\kappa_2^2)
+ 16 \beta_1 \beta_2^2 (\kappa_1 - \beta_1)
+ 4 \lambda \bigl(4 \beta_1 \kappa_1^2 (\kappa_2 - 2 \beta_2 )
+{\tilde t} {\tilde u}
\nonumber\\&+&
\kappa_1 (\kappa_2^3 - 2 \beta_2 \kappa_2^2
+ (\kappa_2 - \beta_2)({\hat s} + {\tilde t} - 4 \beta_1^2)
- \beta_2 {\tilde u})
-\beta_1 (2 \beta_2 - \kappa_2) {\tilde u}\bigr)
- 16 \lambda^2 \beta_1^2 \kappa_1^2
\nonumber\\&+&
4 \lambda^2 \kappa_1 (4 \beta_1 \kappa_1^2 - \kappa_1 (\kappa_1^2 + {\tilde u})
+2 \beta_1 {\tilde u})
+\kappa_2 (4 \beta_2 - \kappa_2)((2 \beta_1^2 - \kappa_1)^2 + \kappa_2^2)
+ 4 \kappa_2^2 \beta_2^2
\Bigr)\mbox{,}\end{aligned}$$ $$\begin{aligned}
M_{33}&=&\frac{3}{4 {\hat s}^2}
\Bigl((\kappa_1^2 +\kappa_2^2) {\hat s}
-4 (\beta_2 \kappa_1 - \beta_1 \kappa_2)^2
+ 4 \lambda \beta_1 \kappa_2(\kappa_1^2 - \kappa_2^2 - {\hat t} +
{\hat u})
\nonumber\\&-&
\lambda^2 \bigl(\kappa_1^4 + \kappa_2^4 - 8 \beta_1 \kappa_1^3
- 8 \beta_2 \kappa_2^3 - {\hat s}^2 + {\hat t}^2 +
8 \beta_1 \kappa_1 (\kappa_2^2
- 4 \beta_2 \kappa_2 + {\hat t} - {\hat u})
- 2 {\hat t} {\hat u}
+ {\hat u}^2
\nonumber\\&-&
2 \kappa_1^2 (\kappa_2^2 - 4 \beta_2 \kappa_2 - 8 \beta_1^2 + {\hat t} - {\hat u})
+ 2 \kappa_2^2 (8 \beta_2^2 + {\hat t} - {\hat u})
- 8 \beta_2 \kappa_2 ({\hat t} - {\hat u})\bigr)
\nonumber\\&-&
2 \lambda \bigl(8 \beta_2^2 \kappa_1 \kappa_2 + 2 \beta_2
(\kappa_1^3 - 4 \beta_1 \kappa_1^2 - 4 \beta_1 \kappa_2^2
- \kappa_1 (\kappa_2^2 + {\hat t} - {\hat u})) +
\kappa_1 \kappa_2 (8 \beta_1^2 - {\hat s})\bigr)\Bigr)\mbox{,}\end{aligned}$$ $$\begin{aligned}
M_{12}&=&-\frac{1}{48 {\tilde t} {\tilde u}} \biggl(
4 \beta_1 \kappa_1^3
+ \kappa_2^2 ( {\hat s}- 8 \beta_1^2 - {\tilde t} - {\tilde u}
)
+ 32 \beta_1 \beta_2 (\beta_1 \kappa_2 - \beta_1 \beta_2)
+ {\hat s}^2
- {\tilde t}^2
- {\tilde u}^2
\nonumber\\&-&
\kappa_1^2 (\kappa_2^2 -\kappa_1^2
- 8 \beta_2 \kappa_2
+ 8 \beta_2^2 - 2 {\hat s})
+ 4 \beta_1 \kappa_1 (2 \kappa_2^2
- 8 \beta_2 \kappa_2
+ 8 \beta_2^2 - \kappa_1^2)
+
2 \lambda^2 \Bigl(\kappa_1^4
\nonumber\\&+& \kappa_2^4
- {\hat s}^2 +
{\tilde t}^2 + {\tilde u}^2 + 4 \beta_2 \kappa_2 {\tilde u}
+ 2 \kappa_2^2 {\tilde t} + 2 \kappa_1^2 ( 4 \beta_2 \kappa_2
+ {\tilde u})
+ 4 \beta_1 \kappa_1 (2 \kappa_2^2 - 4 \beta_2 \kappa_2
\nonumber\\&+&
{\tilde t})\Bigr) -
4 \lambda \Bigl(4 \beta_2^2 \kappa_2 (\kappa_1 - 2 \beta_1 ) -
\kappa_2 (-4 \beta_1^2 \kappa_1
+ \beta_1 (3 \kappa_1^2
+ \kappa_2^2 - {\hat s}) + \kappa_1 {\hat s})
- \beta_2 \bigl(\kappa_1^3
\nonumber\\&-&
4 \beta_1 \kappa_1^2
+ \kappa_1 (3 \kappa_2^2
+ 8 \beta_1^2 - {\hat s}) +
2 \beta_1 ({\hat s} -3 \kappa_2^2 - \kappa_1^2)\bigr)\Bigr)\biggr)\mbox{,}\end{aligned}$$ $$\begin{aligned}
M_{13}&=&-\frac{3}{16 {\hat s} {\tilde t}}
\biggl(2 (\lambda^2-1) \kappa_1^4 + 2 \kappa_1^3
(2 \beta_1 - \lambda \kappa_2
- 3 \beta_1 \lambda^2
+ 3 \lambda \beta_2 ) +
2 \beta_1 \lambda (2 \kappa_2^3
+ 4 \beta_2 \kappa_2^2
\nonumber\\&+&
4 \beta_2 ({\hat t} - {\hat u})
- \kappa_2 (16 \beta_2^2 - 2 {\tilde u})) -
\kappa_1^2 (2 (1 + \lambda^2) \kappa_2^2
- 2 ((5 \lambda^2 - 2) \beta_2
+ 6 \beta_1 \lambda) \kappa_2
\nonumber\\&+&
24 \lambda \beta_1 \beta_2 - 8 \beta_2^2 + 2 {\hat s} -
\lambda^2 {\hat s}
+ \lambda^2 {\tilde t} + 2 {\tilde u} - 3 \lambda^2 {\tilde u})
+ 2 \kappa_2 \beta_2 (
8 \beta_1^2
+ {\tilde t} + {\tilde u})
- 4 \beta_1^2 \kappa_2
\nonumber\\&-&
\kappa_2 {\tilde u} +
\lambda^2 (4 \kappa_2^4 - 22 \beta_2 \kappa_2^3 - {\hat s}^2 + ({\hat t} - {\hat u})^2 -
\kappa_2^2 ({\hat s} - 32 \beta_2^2
- 5 {\tilde t} + 3 {\tilde u})
+ 2 \beta_2 \kappa_2 ({\hat s}
\nonumber\\&-&
7 {\tilde t} + 5 {\tilde u})) +
2 \kappa_1 (\beta_1 \lambda^2 (5 \kappa_2^2
- 16 \beta_2 \kappa_2 - 3 {\hat s} + {\tilde t} - 3 {\tilde u})
+ \beta_1 (4 \kappa_2^2 - 8 \beta_2^2
+ 2 {\hat s}
+ {\tilde t}
\nonumber\\&+&
{\tilde u}) +
\lambda (3 \kappa_2^3 - 11 \beta_2 \kappa_2^2 +
\kappa_2 ( 16 \beta_2^2 - 8 \beta_1^2
+ 2 {\tilde t}
- {\tilde u}) +
\beta_2 (16 \beta_1^2 + {\hat s} - 3 {\hat t} + 3 {\hat u})))\biggr)\mbox{,}\end{aligned}$$ $$\begin{aligned}
M_{23}&=& - \frac{3}{16 {\hat s} {\tilde u}}
\biggl( \lambda^2 (4 \kappa_1^4
+ 2 \kappa_2^4 - 22 \beta_1 \kappa_1^3
- {\hat s}^2 +
2 \beta_1 \kappa_1 (5 \kappa_2^2
- 16 \beta_2 \kappa_2 + {\hat s} + 5 {\tilde t} - 7 {\tilde u})
\nonumber\\&-&
\kappa_1^2 (2 \kappa_2^2 - 16 \beta_2 \kappa_2
- 32 \beta_1^2 + {\hat s} + 3 {\tilde t} - 5 {\tilde u}) +
\kappa_2^2 ({\hat s} + 3 {\tilde t} - {\tilde u}) + ({\hat t} - {\hat u})^2
+ 8 \beta_2 \kappa_2 {\tilde u} )
\nonumber\\&-&
2 (\kappa_1^2 (\kappa_2^2 - 3 \beta_2 \kappa_2
+ 4 \beta_2^2 + {\tilde t}) +
\kappa_2 (\kappa_2^3 - \beta_2 \kappa_2^2 + \kappa_2 ({\hat s}
- 4 \beta_1^2 + {\tilde t})
+ \beta_2 (8 \beta_1^2 - {\hat s}))
\nonumber\\&-&
\beta_1 \kappa_1 (8 \beta_2^2 -2 \kappa_2^2
+ {\tilde t} + {\tilde u})) +
2 \lambda (\kappa_1^3 (\kappa_2 + 2 \beta_2 )
+ \beta_1 \kappa_1^2 (4 \beta_2 - 11 \kappa_2)
+ \beta_1 (4 \beta_2 ({\hat u} - {\hat t})
\nonumber\\&+&
3 \kappa_2^3 - 12 \beta_2 \kappa_2^2
+ \kappa_2 (16 \beta_2^2 + {\hat s} + 3 {\hat t} - 3 {\hat u}))
- \kappa_1 (3 \kappa_2^3 - 8 \beta_2 \kappa_2^2 -
\kappa_2 (16 \beta_1^2
- 8 \beta_2^2
- 2 {\hat s}
\nonumber\\&-&
3 {\tilde t} + {\tilde u}) +
\beta_2 (16 \beta_1^2 - {\tilde t})))\biggr)\mbox{,}\end{aligned}$$ where $$\begin{aligned}
{\tilde t} &=& {\hat t} - m_Q^2 \mbox{,}\\
{\tilde u} &=& {\hat u} - m_Q^2 \mbox{,}\\
\beta_1 &=& |{\vec p}_{T}| \cos{\varphi_1}\mbox{,}\\
\beta_2 &=& |{\vec p}_{T}| \cos{\varphi_2}\mbox{,}\\
\kappa_1 &=& |{\vec k}_{1T}|\mbox{,}\\
\kappa_2 &=& |{\vec k}_{2T}|\mbox{,}\\
\lambda &=& \cos{(\varphi_1-\varphi_2)}\end{aligned}$$ and ${\vec p}_T$ — transverse momentum of anti-quark. Since these are more suitable for numerical calculations. In those calculations, which were in the framework of the collinear parton model, we used the GRV [@GRV] parameterization for the gluon distribution function.
The fragmentation model
=======================
The analysis of the $B_c$ meson gluon-gluon production in the framework of the collinear parton model shows [@Likhoded; @Ruckl; @Chang] the dominant role of the fusion mechanism up to $p_T\thickapprox 30$ GeV. It takes into account the total gauge invariant set of the diagrams which describes the parton process $$\begin{aligned}
g+g \to B_c + b + \bar c\mbox{.} \label{eq:ggBcbc}\end{aligned}$$ Thus, the fragmentation approximation, incorporating factorization of the heavy quark production process and the soft process of a meson created in the final state, applies only at $p_T > 30$ GeV. Of course, at the Tevatron Collider this region of $p_T$ is far from being available for experimental study. On the other hand, at the LHC Collider the $p_T\le 50$ GeV region will be under consideration.
In the fragmentation model, the $B_c$ meson production cross section can be presented as follows: $$\begin{aligned}
d \sigma(pp\to B_c X)=\sum_i \int{dz} D_{i \to B_c}(z)\cdot d \hat
\sigma(pp\to i)\mbox{,}\end{aligned}$$ where the sum is taken over all the relevant partons $i=c$, $b$ and $g$ (but the main contribution comes from $b$-quark fragmentation).
The fragmentation functions for the $b$-quark splitting into a vector $B_c^*$ and pseudoscalar $B_c$ meson at the starting scale of the perturbative QCD evolution were obtained in Ref. [@Braaten; @Likhoded-z0]: $$\begin{aligned}
D_{\bar b\to B_c}(z,\mu_0)&=&\frac{8 \alpha_s^2(\mu_0)
|\Psi_{B_c}(0)|^2}{81 m_c^3} \frac{r (1 - z)^2 z}{(1 - (1 - r)
z)^6} \Bigl(6 - 18 (1 - 2 r) z
+ (21 - 74 r + 68 r^2) z^2 \nonumber\\
&-&2 (1 - r) (6 - 19 r + 18 r^2) z^3
+3 (1 - r)^2 (1 - 2 r + 2 r^2) z^4\Bigr)\mbox{,}\\
D_{\bar b\to B_c^\star}(z,\mu_0)&=&\frac{8 \alpha_s^2(\mu_0)
|\Psi_{B_c}(0)|^2}{27 m_c^3} \frac{r (1 - z)^2 z}{(1 - (1 - r)
z)^6} \Bigl(2 - 2 (3 - 2 r) z + 3 (3 - 2 r + 4 r^2) z^2
\nonumber\\&-& 2 (1 - r) (4 - r + 2 r^2) z^3 + (1 - r)^2 (3 - 2
r + 2 r^2) z^4\Bigr)\mbox{,}\end{aligned}$$ where $r=\frac{m_c}{m_c+m_b}$. $|\Psi_{B_c}(0)|^2$ is the squared $B_c$ meson wave function at the origin, which can be calculated using the nonrelativistic potential quark model [@Psi0]. In the numerical calculations, we used the result $\Psi_{B_c}(0)=\sqrt{\frac{m_{B_c}}{12}}f_{B_c}$, where $f_{B_c}=490-560$ MeV is the $B_c$ meson leptonic decay constant.
The QCD evolution of the fragmentation functions $D_{\bar b\to
B_c}$ and $D_{\bar b\to B_c^\star}$ are described by the DGLAP [@DGLAP] evolution equation $$\begin{aligned}
\mu^2\frac{\partial D}{\partial
\mu^2}(z,\mu^2)=\frac{\alpha_s(\mu^2)}{2\pi}\int_z^1
\frac{dx}{x}P_{q\to q}(\frac{x}{z}) D(x,\mu^2)\mbox{,}\end{aligned}$$ where $P_{q\to q}(z)$ is the standard quark-quark splitting function.
The results
===========
The results of our calculation for the $p_T$-spectra of the $B_c^\star$ meson are shown in Fig. \[fig:Tev\_FR\] for the energy of the Tevatron Collider $\sqrt{s}=1.8$ TeV, and in the Fig. \[fig:LHC\_FR\] for the energy of the LHC Collider $\sqrt{s}=14$ TeV. The spectra that we obtain are compared with those obtained earlier in the collinear parton model via the fragmentation mechanism. The curves in Fig. \[fig:Tev\_FR\] correspond to a choice of the parameters from Ref. [@Cheung]: $m_c=1.5$ GeV, $m_b=4.9$ GeV, $\alpha_s = \alpha_s(p_T^2+ m_{B_c}^2)$ and $f_{B_c}=490$ MeV. The curves in Fig. \[fig:LHC\_FR\] were obtained using the set of parameters from Ref. [@Likhoded]: $m_c=1.5$ GeV, $m_b=5.1$ MeV, $\alpha_s
\simeq 0.23$ and $f_{B_c}=560$ MeV.
Fig. \[fig:Tev\_FR\] shows that our result obtained in the collinear parton model agrees well with the result obtained earlier in Ref. [@Cheung]. The curves obtained in the $k_T$-factorization approach lie approximately 2—5 times higher than the collinear parton model prediction in the region $p_T> 10$ GeV. The maximum value of the $B_c$ meson production cross section is obtained using the JS [@JS] parameterization of the unintegrated gluon distribution function, and the minimum value is obtained using the JB [@JB] parameterization.
Note that the slopes of the $p_T$-spectra obtained in the collinear parton model and in the $k_T$-factorization approach are equal, and the situation is the same as in the case of $D^\star$ meson production in $ep$ interactions [@SV; @ZBJ].
At present, there is no experimental information on the $p_T$-spectra of $B_c$ mesons. It is known that the integrated production cross section in the region $p_T > 6$ GeV, $|y| < 1$ and $\sqrt{s}=1.8$ TeV is given by $\sigma_{B_c} \simeq 10 \pm 6$ nb [@CDF1; @CDF2]. As shown in Ref. [@Likhoded], the uncertainties in the calculations are about $\sim 50$%. This value depends on the choice of the masses, the $\alpha_s$ constant and the leptonic decay constant $f_{B_c}$. Furthermore, our calculations performed in the $k_T$-factorization approach show that the $B_c$ meson production cross section has a strong dependence on the choice of the unintegrated gluon distribution function — a variation by a factor of 2 was observed.
The obtained values of the integrated $B_c$ and $B_c^\star$ meson production cross section at the energies of the Tevatron and the LHC Colliders are presented in Table \[tab:1\]. One can see that, with the fragmentation model, the results obtained in the collinear parton model are smaller to those obtained in the $k_T$-factorization approach. The theoretical predictions are shown together with their uncertainties, which are significantly larger in the $k_T$-factorization approach. The absolute value of the cross section obtained in the collinear parton model ($\sigma_{B_c}=1.7\pm 0.8$ nb) is much smaller than the CDF experimental data ($\sigma_{B_c}=10\pm 6$ nb) [@CDF1; @CDF2]. On the other hand, the prediction of the $k_T$-factorization approach agrees well with the data ($\sigma_{B_c}=7.4\pm 5.4$ nb).
Taking into account the relative roles of the fusion and fragmentation mechanisms in $B_c$ meson hadroproduction, one can suppose that the cross section calculated in the fusion model and in the $k_T$-factorization approach will be more large than that obtained in the fragmentation model and in the $k_T$-factorization approach. The calculation of the $B_c$ meson hadroproduction rates via the partonic subprocess (\[eq:ggBcbc\]) with the reggeized initial gluons will be presented in our forthcoming paper.
Acknowledgments
===============
We thank B. Kniehl and A. Likhoded for useful discussions about the results obtained. The work is supported by the Russian Federal Agency of Education under Grant A04-2.9-52. The part of the D.V. work was done in the framework of the Grant “Mikhail Lomonosov”, which is supported by DAAD and by Russian Ministry of Education. D.V. thanks the International Center of Fundamental Physics in Moscow and Dynastiya Foundation for the financial support received while this work was done.
[8.]{}
CDF Collaboration, F. Abe, [at al.]{}, Phys. Rev. Lett. [81]{} (1998) 2432;
CDF Collaboration, F. Abe, [at al.]{}, Phys. Rev. D [58]{} (1998) 112004.
K. Anikeev, [at al.]{}, FERMILAB-Pub-01/197 (2001) 480.
S.S. Gershtein, V.V. Kiselev, A.K. Likhoded, S.R. Slabospitsky, A.V. Tkabladze, Yad. Fiz. [48]{} (1988) 515;
Y.-Q. Chen, Y.-P. Kuang, Phys. Rev. D [46]{} (1992) 1165;
E. Eichten, F. Feinberg, Phys. Rev. D [23]{} (1981) 2724.
V.V. Kiselev, A.K. Likhoded, A.O. Onishenko, Nucl. Phys. B [569]{} (2000) 473;
V.V. Kiselev, A.K. Likhoded, A.E. Kovalsky, Nucl. Phys. B [585]{} (2000) 353.
I. Bigi, Phys. Lett. B [371]{} (1996) 103;
M. Beneke, G. Buchalla, Phys. Rev. D [53]{} (1996) 4991.
A.V. Berezhnoy, A.K. Likhoded, M.V. Shevlyagin, Yad. Fiz. [58]{} (1995) 730;
A.V. Berezhnoy, A.K. Likhoded, O.O. Yushenko, Yad. Fiz. [59]{} (1996) 742;
A.V. Berezhnoy, V.V. Kiselev, A.K. Likhoded, Z. Phys. A [356]{} (1996) 79.
K. Kolodziej, R. Rückl, Nucl. Instrum. Methods A [408]{} (1998) 33;
K. Kolodziej, A. Leike, R. Rückl, Phys. Lett. B [355]{} (1995) 337.
C.-H. Chang, [at al.]{}, Phys. Lett. B [364]{} (1995) 78;
C.-H. Chang, Y.-Q. Chen, R.J. Oakes, Phys. Rev. D [54]{} (1996) 4344.
S.P. Baranov, Phys. Rev. D [56]{} (1997) 3046.
A.P. Martynenko, V.A. Saleev, Phys. Rev. D [54]{} (1996) 1891;
A.P. Martynenko, V.A. Saleev, Yad. Fiz. [59]{} (1996) 747.
J.C. Collins, R.K. Ellis, Nucl. Phys. [360]{} (1991) 3.
L.V. Gribov, E.M. Levin, M.G. Ryskin, Phys. Rep. [100]{} (1983) 1;
S. Catani, M. Ciafoloni, F. Hautmann, Nucl. Phys. B [366]{} (1991) 135;
V.S. Fadin, L.N. Lipatov, Nucl. Phys. B [477]{} (1996) 767.
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Sov. Phys. JETP [44]{} (1976) 443;
I.I. Balitsky, L.N. Lipatov, Sov. J. Nucl. Phys. [28]{} (1978) 822.
M. Ciafaloni, Nucl. Phys. B [296]{} (1988) 49;
S. Catani, F. Fiorani, G. Marchesini, Phys. Lett. B [234]{} (1990) 339;
G. Marchesini, Nucl. Phys. B [445]{} (1995) 49.
V.N. Gribov, L.N. Lipatov, Sov. J. Nucl. Phys. [15]{} (1972) 438;
Y.L. Dokshitzer, Sov. Phys. JETP [46]{} (1977) 641;
G. Altarelli, G. Parisi, Nucl. Phys. B [126]{} (1977) 298.
K. Cheung, T.C. Yuan, Phys. Rev. D [53]{} (1996) 1232;
K. Cheung, Phys. Rev. Lett. [71]{} (1993) 3413.
J. Blumlein, DESY 95-121 (1995).
H. Jung, G. Salam, Eur. Phys. J. C [19]{} (2001) 351.
M.A. Kimber, A.D. Martin, M.G. Ryskin, Phys. Rev. D [63]{} (2001) 114027.
V.A. Saleev, D.V. Vasin, Yad. Fiz. (2005) to be published.
M. Gluck, E. Reya, A. Vogt, Z. Phys. C [67]{} (1995) 433.
E. Braaten, K. Cheung, T.C. Yuan, Phys. Rev. D [48]{} (1993) 5049.
V.V Kiselev, A.K. Likhoded, M.V. Shevlyagin, Z. Phys. C [63]{} (1994) 77.
E.J. Eichten, C. Quigg, Phys. Rev. D [52]{} (1995) 1726.
V.A. Saleev, D.V. Vasin, Phys. Lett. B [548]{} (2002) 161; erratum ibid., hep-ph /0209220, v. 2.
S.P. Baranov, H. Jung, L. Jonsson, S. Padhi, N. P. Zotov, Eur. Phys. J. C [24]{} (2002) 425;
S.P. Baranov, H. Jung, N.P. Zotov, Nucl. Phys. A (Proc. Suppl.) [99]{} (2001) 192.
------------- --------- -------------- --------------- --------------------- --------------
$\sqrt{s},$ $|y|$ $p_{T,min}$, Parton $k_T$-factorization Experimental
TeV GeV model approach data
$1.8$ $< 1$ $6$ $1.7 \pm 0.8$ $7.4 $10 \pm 6$
\pm 5.4$
$14$ $< 2.5$ $10$ $28 \pm 14$ $122 \pm 90$ —
\[1mm\]
------------- --------- -------------- --------------- --------------------- --------------
: The summed $B_c$ and $B_c^\star$ meson production cross sections in the different models at the energies of the Tevatron and LHC Colliders. The data from the CDF Collaboration[@CDF1; @CDF2] are shown. The cross sections are in nb.
\[tab:1\]
[^1]: Email: [email protected]
[^2]: Email: [email protected]; [email protected]
[^3]: On leave from Samara State University, 443011 Samara, Russia
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The conventional formal tool to detect effects of the financial persistence is in terms of the Hurst exponent. A typical corresponding result is that its value comes out close to $0.5$, as characteristic for geometric Brownian motion, with at most small departures from this value in either direction depending on the market and on the time scales involved. We study the high frequency price changes on the American and on the German stock markets. For both corresponding indices, the Dow Jones and the DAX respectively, the Hurst exponent analysis results in values close to $0.5$. However, by decomposing the market dynamics into pairs of steps such that an elementary move up (down) is followed by another move up (down) and explicitly counting the resulting conditional probabilities we find values typically close to $60\%$. This effect of persistence is particularly visible on the short time scales ranging from 1 up to 3 minutes, decreasing gradually to $50\%$ and even significantly below this value on the larger time scales. We also detect some asymmetry in persistence related to the moves up and down, respectively. This indicates a subtle nature of the financial persistence whose characteristics escape detection within the conventional Hurst exponent formalism.'
address: |
$^1$Institute of Physics, University of Rzeszów, PL–35-310 Rzeszów, Poland\
$^2$Institute of Nuclear Physics, Polish Academy of Sciences,\
PL–31-342 Kraków, Poland
author:
- 'R.Rak$^1$, S. Drożdż$^{1,2}$, J. Kwapień$^2$, P. Oświȩcimka$^2$'
title: Detecting subtle effects of persistence in the stock market dynamics
---
=by -1
Introduction
============
The financial dynamics results in fluctuations whose nature is, as pointed out by Bachelier [@Bach] already in 1900, of the Brownian character. By now we know that it is much more complex and fascinating than just the ordinary Brownian motion. Already the distribution of stock market returns is far from being Gaussian and at the short time scales large fluctuations develop heavy power law asymptotics with an exponent $\alpha=3$ [@Gop], well outside the Levy stable regime [@Dro1], however. The autocorrelation function of returns drops down very quickly and after a few minutes it reaches the noise level. At the same time however the volatility autocorrelation function decays very slowly with time [@Dro1], largely according to the power law, and remains positive for many months. On a more advanced level of global quantification, the financial dynamics appears to be describable in terms of multifractality both in the transaction-to-transaction price increments and in the inter-trade waiting times [@Osw]. This indicates a hierarchically convoluted self-similar organization of the market dynamics. One related issue is an effect of persistence. Its commonly adopted measure - the Hurst exponent - is the mode of each multifractal spectrum. The Hurst exponent, however, is a global measure while the effects of persistence may in principle depend on the market phase. Below we address this issue using the high-frequency records (years 1998-99) of the two from among the world leading stock market indices, the Dow Jones Industrial Average (DJIA) for the United States and the Deutsche Aktienindex (DAX) for Germany.
Conventional methods
====================
There exist two commonly accepted and best-known methods to evaluate the long-range dependences in the statistical series. The older one is the so-called rescaled range or R/S analysis [@Man]. This method originates from previous hydrological analysis of Hurst [@Hur] and allows to calculate the self-similarity parameter $H$. A drawback of this method however is that it may spuriously detect some apparent long-range correlations that result from non-stationarity. A method that avoids such artifacts is the Detrended Fluctuation Analysis (DFA) [@Peng]. In this method one divides a time series $g(t_i)$ of length $N$ $(i=1,...N)$ into $M$ disjoint segments $\nu$ of length $n$ and calculates the signal profile $$Y_{\nu} (i)=\sum_{k=1}^i (g(k)-\langle g \rangle),~~~~i=1,...,N
\label{sp}$$ where $\langle ... \rangle$ denotes the mean. For each segment $\nu$ the local trend is then estimated by least-squares fitting the straight line (in general a polynomial) ${\tilde Y}_{\nu}(i)$ and the corresponding variance $$F^2(\nu, n) = {1 \over n} \sum_{j=1}^n \{ Y[(\nu - 1)n + j] -
{\tilde Y}(j)\}. \label{var}$$ Finally, one calculates the mean value of the root mean square fluctuations over all the segments $\nu$: $${\bar F}(n) = {1\over M} \sum_{\nu = 1}^M F (\nu, n) \label{msf}$$ The power-law scaling of the form $${\bar F}(n) \sim n^H \label{pl}$$ indicates self-similarity and is considered to provide a measure of persistence. If the process is white noise then $H=0.5$. If the process is persistent then $H > 0.5$; if it is anti-persistent then $H < 0.5$.
The above procedure applied to the returns $$g(t) = \ln P(t + \Delta t) - \ln P(t), \label{ret}$$ where $P(t)$ represents the price time series, results in numbers as listed in the last column of Table 1 for $\Delta t$ ranging from 1 min up to 30 min. In addition to the DAX and the DJIA this Table includes also what for brevity we here call Nasdaq30 and what for the purpose of this work is constructed as a simple sum of the prices of 30 high-capitalization companies belonging to the Nasdaq Composite basket. As one can see from the Table 1, typically the so-calculated Hurst exponents $H$ point to a trace of anti-persistence but in fact they do not deviate much from 0.5, especially that an error involved in estimation equals about $0.4\%$ for $\Delta t = 1$ min and increases up to $1.5\%$ for $\Delta t = 30$ min due to an effective shortening of the series.
Still this result does not eliminate a possibility that there exist some more local effect of persistence that simply average out when estimated from the longer time intervals. In fact, some proposals to calculate the local counterparts of $H$, based on variants of DFA, are already present in the literature [@Gre; @Car] and point to such effects indeed. The accuracy of the related methods is however not yet well established. Furthermore, observations and experience prompt a possibility that the financial persistence may happen to occur asymmetrically, i.e., a move up may be followed by another move up more often than a move down by another move down, or vice versa. Such effects may carry a very valuable information about the dynamics but remain indistinguishable within the conventional methods and unexplored so far. In order therefore to explore a possibility and character of such effects we return to the very definition of persistence.
Measuring persistence by conditional probabilities
==================================================
Given a time series of price returns $g(t_i)$, where $t_i$ denotes the consecutive equidistant moments of time, to each $i$ we assign +1 if $g(t_i)$ is positive (price goes up), -1 if it is negative (price goes down) and 0 if it happens to be 0 (price remains unchanged). We then explicitly count the number $N_{\alpha,\beta}$ of all the neighboring pairs $\{ g(t_i), g(t_{i+1})\}$ of the type $\alpha, \beta = \{-1, 0, +1 \}$ for fixed values of $\alpha$ and $\beta$ and do so for all the nine combinations of different $\alpha$ and $\beta$. Finally, we calculate $$p_{\alpha, \beta} = N_{\alpha, \beta} / \sum_{\beta'=-1, 0, +1}
N_{\alpha, \beta'},\label{cp}$$ which corresponds to a conditional probability that a return of the type $\beta$ is preceded by a return of the type $\alpha$. This procedure can of course be performed on any time scale $\Delta t = t_{i+1} - t_i$.
Six combinations of $p_{\alpha, \beta}$ corresponding to $\alpha =
\pm 1$ and to all the three possible values of $\beta$ are listed in Table 1 for several values of $\Delta t$ starting from 1 up to 30 min.\
[|c|c|c|c|c|c|c|c|]{} **Index/Scale & $\bf P_{11}$ & $\bf
P_{1-1}$ & $\bf P_{10}$ & $\bf P_{-1-1}$ & $\bf P_{-11}$ & $\bf P_{-10}$ & **[Hurst exp.]{}\
**DAX/1min & 0.567 & 0.427 & 0.005 & 0.562 & 0.431 & 0.005 & 0.493\
**DAX/2min & 0.568 & 0.428 & 0.002 & 0.561 & 0.435 & 0.003 & 0.496\
**DAX/3min & 0.554& 0.444 & 0.0002 & 0.548 & 0.449 & 0.0014 & 0.497\
**DAX/4min & 0.539 & 0.459 & 0.001 & 0.529 & 0.469 & 0.001 & 0.498\
**DAX/5min & 0.528 & 0.470 & 0.0006 & 0.514 & 0.484 & 0.0008 & 0.5\
**DAX/10min & 0.493 & 0.507 & 0.0002 & 0.476 & 0.522 & 0.0006 & 0.498\
**DAX/15min & 0.483 & 0.515 & 0.0004 & 0.460 & 0.538 & 0.0009 & 0.498\
**DAX/30min & 0.483 & 0.516 & 0.0002 & 0.459 & 0.540 & 0.0002 & 0.495\
**DJIA/1min& 0.558& 0.399& 0.042& 0.558& 0.398& 0.043& 0.502\
**DJIA/2min& 0.555& 0.416& 0.027& 0.561& 0.413& 0.026& 0.499\
**DJIA/3min& 0.526& 0.452& 0.021& 0.531& 0.449& 0.019& 0.498\
**DJIA/4min& 0.504& 0.479& 0.016& 0.504& 0.478& 0.016& 0.498\
**DJIA/5min& 0.498& 0.487& 0.013& 0.497& 0.488& 0.014& 0.495\
**DJIA/10min& 0.497& 0.491& 0.011& 0.498& 0.493& 0.008& 0.491\
**DJIA/15min& 0.502& 0.491& 0.006& 0.487& 0.504& 0.007& 0.491\
**DJIA/30min& 0.506& 0.487& 0.0066& 0.472& 0.521& 0.0068& 0.491\
**NQ30/1min& 0.539& 0.454& 0.006& 0.54& 0.455& 0.005& 0.5\
**NQ30/2min& 0.547& 0.449& 0.003& 0.546& 0.45& 0.003& 0.499\
**NQ30/3min& 0.539& 0.458& 0.003& 0.529& 0.468& 0.003& 0.501\
**NQ30/4min& 0.532& 0.464& 0.002& 0.518& 0.478& 0.002& 0.499\
**NQ30/5min& 0.53& 0.467& 0.002& 0.515& 0.481& 0.002& 0.501\
**NQ30/10min&0.538& 0.46& 0.001& 0.511& 0.487& 0.0006& 0.502\
**NQ30/15min&0.526& 0.472& 0.001& 0.497& 0.5& 0.001& 0.5\
**************************************************
Quite interestingly - and somewhat unexpectedly in view of the corresponding values of the Hurst exponents (last column in Table 1) that are very close to 0.5 like for the white noise - both the DAX and the DJIA show significant effects of persistence on the small time scales. A move up (down) is followed by another move up (down) significantly more often than by a move in opposite direction. For $\Delta t$ larger than 5-10 min we observe a crossover: the fluctuations become anti-persistent and, what is particularly interesting, this effect is visibly asymmetric towards moves down as a systematically observed relation $p_{-1,-1}$ and $p_{+1,+1}$ indicates. For the basket of the Nasdaq stocks this crossover also takes place, though on the somewhat larger time scales. Quite interestingly - and somewhat unexpectedly in view of the corresponding values of the Hurst exponents (last column in Table 1) that are very close to 0.5 like for the white noise - both the DAX and the DJIA show significant effects of persistence on the small time scales. A move up (down) is followed by another move up (down) significantly more often than by a move in opposite direction. For $\Delta t$ larger than 5-10 min we observe a crossover: the fluctuations become anti-persistent and, what is particularly interesting, this effect is visibly asymmetric towards moves down as a systematically observed relation $p_{-1,-1}$ and $p_{+1,+1}$ indicates. For the basket of the Nasdaq stocks this crossover also takes place, though on the somewhat larger time scales.
The time period (01.12.1997 - 31.12.1999) of the stock market variability studied here displays a richness of phases. In the first half of this period it for instance includes a spectacular draw up (DAX more than $50\%$) followed by an even faster draw down to the original level. Previous study [@Dro2] based on the correlation matrix formalism provides a serious indication that the dynamics of long-term stock market increases is more competitive and less collective (as far as correlations among the individual stocks forming an index is concerned) than during long-term decreases.
It is thus interesting to inspect if and how our indicators of persistence correlate with the different phases of the market dynamics. Two principal such coefficients (rectangles), $p_{+1,+1}$ and $p_{-1,-1}$, calculated for $\Delta t =1$ min over one trading day time intervals for all the consecutive days covering our 01.12.1997 - 31.12.1999 time period, versus the corresponding DAX changes are shown in Fig. 1. The correlation is visible indeed.
12.5cm 12.5cm
The long-term increases systematically lead to a decrease of persistence both in $p_{+1,+1}$ and in $p_{-1,-1}$. A sharp increase of the DAX in the end of the period here analyzed pulls these two coefficients even below 0.5. On the contrary, the decreases are seen to be lifting our persistency coefficients up to as high as $\sim 0.7$. Even more, changes of the trend in $p_{+1,+1}$ and $p_{-1,-1}$ are somewhat shifted in phase relative to each other which is another manifestation of asymmetry in persistence. That all such effects may reflect a general logic of the stock market dynamics can be seen also from Fig. 2 which displays the same quantities for the DJIA in the same period of time and qualitatively analogous correlations can be postulated.
12.5cm 12.5cm
More focus on this last issue is given in Figs. 3 and 4 and the corresponding Tables 2 and 3. Two sizable periods of the global market increases and decreases, respectively, both for the DAX and for the DJIA, are here extracted and the corresponding conditional probability coefficients $p_{\alpha, \beta}$ calculated for those periods separately. Again one sees that the related fluctuations are persistent and that these persistency effects are even stronger during decreases. Furthermore, even during the same market phase (either global increase or decrease) the asymmetry in persistence between the moves up and down may occur.
12.5cm
------------------------------------------------------------------- -- -- -- -- -- -- --
**Data& $\bf P_{11}$& $\bf P_{1-1}$& $\bf P_{10}$& $\bf
P_{-1-1}$& $\bf P_{-11}$& $\bf P_{-10}$& **[Hurst exp.]{}\
**DAX(increase) &0.562 &0.433 &0.004 &0.57 &0.424 &0.006 &0.491\
**DAX(decrease) &0.615 &0.382 &0.003 &0.585 &0.412 &0.002 &0.504\
********
------------------------------------------------------------------- -- -- -- -- -- -- --
12.5cm
-------------------------------------------------------------------------------------------------------------------- -- -- -- -- -- -- --
**Data& $\bf P_{11}$& $\bf P_{1-1}$& $\bf P_{10}$& $\bf P_{-1-1}$& $\bf P_{-11}$& $\bf P_{-10}$& **[Hurst exp.]{}\
**DJIA(increase) &0.534 &0.415 &0.051 &0.523 &0.424 &0.052 &0.495\
**DJIA(decrease) &0.609 &0.363 &0.028 &0.629 &0.342 &0.029 &0.505\
********
-------------------------------------------------------------------------------------------------------------------- -- -- -- -- -- -- --
Conclusion
==========
The above observations are intriguing and of course demand a much more systematic study as they carry a potential to shed more light on mechanism of the stock market dynamics. The present study however already indicates direction concerning this specific issue. There definitely exist higher order correlations in the financial dynamics that escape detection within the conventional methods. In this connection the wavelet based formalism, due to its ability to focus on local effects, seems to offer a promising frame to develop consistent related methodology, such that a link to multifractality perhaps can also be traced. The wavelet based formalism can also be generalised to account for the asymmetry in persistence - an effect identified above. Finally and ultimately, one needs to develop a realistic theoretical model of the financial dynamics such that also the above effects can be incorporated. A variant of the generalised Weierstrass random walk as developed by Kutner [@Kut] may appear an appropriate solution especially that the Weierstrass-type functions may incorporate log-periodicaly a kind correlations that underlay the financial dynamics [@Dro3; @Dro4].
[2209]{}
L. Bachelier, Ann. Sci. de l’Ecole Norm. Sup. [**III-17**]{}, 21 (1900)
P. Gopikrishnan, V. Plerou, L.A. Nunes Amaral, M. Mayer, H.E. Stanley, Phys. Rev. [**E60**]{}, 5305 (1999)
S. Drożdż, J. Kwapień, F. Grümmer, F. Ruf, J. Speth, Acta Phys. Pol. B [**34**]{}, 4293 (2003)
P. Oświȩcimka, J. Kwapień, S. Drożdż, Physica A [**347**]{}, 626 (2005)
B.B. Mandelbrot, J.R. Wallis, Water Resources Res. [**5**]{}, 967 (1969)
H.E. Hurst, Trans. Am. Soc. Civil Engineers [**116**]{}, 770 (1951)
C.-K. Peng, S.V. Buldyrev, A.L. Goldberger, S. Havlin, M. Simons, H.E. Stanley, Phys. Rev. E [**47**]{}, 3730 (1993)
D. Grech, Z. Mazur, Physica A [**336**]{}, 133 (2004)
A. Carbone, G. Castelli, H.E. Stanley, Physica A [**344**]{}, 267 (2004)
S. Drożdż, F. Grümmer, A.Z. Górski, F. Ruf, J. Speth, Physica A [**287**]{}, 440 (2000)
R. Kutner, Physica A [**264**]{}, 84 (1999); Physica A [**314**]{}, 786 (2002)
S. Drożdż, F. Ruf, J. Speth, M. Wójcik, Eur. Phys. J. B10, 589 (1999).
M. Bartolozzi, S. Drożdż, D.B. Leinweber, J. Speth, A.W. Thomas,\
cond-mat/0501513, International Journal of Modern Physics C in print.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'It is well known that the matrix exponential of a non-normal matrix can exhibit transient growth even when all eigenvalues of the matrix have negative real part, and similarly for the powers of the matrix when all eigenvalues have magnitude less than 1. Established conditions for the existence of these transient effects depend on properties of the entire matrix, such as the Kreiss constant, and can be laborious to use in practice. In this work we develop a relationship between the invariant subspaces of the matrix and the existence of transient effects in the matrix exponential or matrix powers. Analytical results are obtained for two-dimensional invariant subspaces and Jordan subspaces, with the former causing transient effects when the angle between the subspace’s constituent eigenvectors is sufficiently small. In addition to providing a finer-grained understanding of transient effects in terms of specific invariant subspaces, this analysis also enables geometric interpretations for the transient effects.'
address: 'Department of Applied Mathematics and Statistics & Institute for Advanced Computational Science, Stony Brook University, Stony Brook, New York 11794, United States'
author:
- 'Matthew G. Reuter'
bibliography:
- 'refs.bib'
title: 'Associating the Invariant Subspaces of a Non-Normal Matrix with Transient Effects in its Matrix Exponential or Matrix Powers'
---
Matrix exponential ,matrix powers ,transient behavior ,non-normal matrices ,numerical range\
AMS Codes: 15A16 ,15A60
Introduction {#sec:intro}
============
The matrix exponential $e^{\mathbf{M}t}$, for $\mathbf{M}\in\mathbb{C}^{N\times N}$ and $t\in\mathbb{R}$ (often $t\ge0$), appears in many contexts [@moler-3-2003; @bk:trefethen-2005], notably when solving a system of first-order, linear ordinary differential equations. The matrix powers $\mathbf{M}^n$ ($n\ge0$) also have wide applicability [@bk:trefethen-2005]; for instance in Markov processes or in condensed matter physics [@reuter-053001-2017]. When $\mathbf{M}$ is normal (*i.e.*, $\mathbf{M}$ has a complete set of orthogonal eigenvectors), its eigenvalues completely determine the behavior of $e^{\mathbf{M}t}$ and of $\mathbf{M}^n$ [@bk:trefethen-2005]. That is, if all eigenvalues of $\mathbf{M}$ have negative real part, then $\|e^{\mathbf{M}t}\|$ decays monotonically with $t$; likewise, $\| \mathbf{M}^n \|$ decays with $n$ when all eigenvalues of $\mathbf{M}$ have magnitude less than 1.
The situation is more complicated when $\mathbf{M}$ is not normal [@bk:trefethen-2005]. The eigenvalues of $\mathbf{M}$ correctly predict asymptotic behaviors as $t\to\infty$ or $n\to\infty$, but can qualitatively fail for small and intermediate values of $t$ or $n$. For example, there can be transient growth in $e^{\mathbf{M}t}$ such that $\|e^{\mathbf{M}t}\|>1$ for some $0 < t \ll \infty$ even though all eigenvalues of $\mathbf{M}$ have negative real part. Similar effects can be observed in $\|\mathbf{M}^n\|$. Numerous applications of these transient effects — *i.e.*, $\|e^{\mathbf{M}t}\|>1$ or $\| \mathbf{M}^n \|>1$ — are discussed in [@bk:trefethen-2005]. Note that this definition of transience, although taken from [@bk:trefethen-2005], is different from “transience” in some applications, where it may mean that the system has not yet reached asymptotic or steady-state conditions. Throughout this work, transience will mean that there is transient growth in the matrix exponential or matrix powers when none would be predicted from the matrix’s eigenvalues.
Pseudospectral analyses [@bk:trefethen-2005] help predict the existence of transient growth, but have two limitations. First, they rely on quantities that can be laborious to obtain in practice, such as the condition number of the eigenvector matrix of $\mathbf{M}$ or the Kreiss constant of $\mathbf{M}$ (see chapter 14 of [@bk:trefethen-2005]). Second, they do not usually provide sharp conditions for when $\mathbf{M}$ will or will not display transient effects. Furthermore, better predictions of transient effects typically require the more computationally intensive properties of $\mathbf{M}$, *e.g.*, the Kreiss constant.
The goal of this work is to determine more practical conditions for the existence of transient effects in the matrix exponential or matrix powers of $\mathbf{M}$. Our approach differs significantly from previous studies. Rather than relating “holistic” properties of $\mathbf{M}$ (such as the Kreiss constant) to transient effects, we examine the $k$-dimensional ($k\ge 2$) invariant subspaces of $\mathbf{M}$ [@bk:gohberg-2006] and show that transient effects can be associated with invariant subspaces that satisfy certain conditions. This ultimately provides a more fine-grained explanation of transient effects and potentially introduces a new line of inquiry for characterizing non-normal matrices in terms of their invariant subspaces (see Thm. \[thm:subspaces\]).
The main results are stated in Thms. \[thm:transient-exp\] and \[thm:transient-power\] for $e^{\mathbf{M}t}$ and $\mathbf{M}^n$, respectively, and are proven in section \[sec:higherD\]. Most of these conditions come from analytical results for the 2-dimensional invariant subspaces of $\mathbf{M}$. The layout of this paper is as follows. First, section \[sec:prelims\] introduces our notation and reviews key concepts from matrix analysis. Section \[sec:2by2\] then derives conditions for transient effects when $\mathbf{M}\in\mathbb{C}^{2\times2}$. These results are generalized to $\mathbf{M}\in\mathbb{C}^{N\times N}$ in section \[sec:higherD\], where they apply to the 2-dimensional invariant subspaces of $\mathbf{M}$. Results for higher-dimensional Jordan subspaces are also presented. Additionally, there are several geometric interpretations for these results, which are discussed in section \[sec:geom\]. We finally offer concluding remarks and ideas for future studies in section \[sec:conclusions\].
\[thm:transient-exp\] Let $\mathbf{M}$ be a $N\times N$ complex matrix ($N\ge 2$) with all of its eigenvalues in the open left half of $\mathbb{C}$; that is, $\mathrm{Re}(\lambda)<0$ if $\lambda$ is an eigenvalue of $\mathbf{M}$. Let $S$ be an invariant subspace of $\mathbf{M}$. Then $e^{\mathbf{M}t}$ will have transient effects when at least one of the following conditions is satisfied.
1. $S$ is a $M$-dimensional ($M\ge2$) Jordan subspace associated with defective eigenvalue $\lambda$ and $$\mathrm{Re}(\lambda) > -\cos\left( \frac{\pi}{M+1} \right).
\label{eq:defective-lambda-exp}$$
2. $S$ is a 2-dimensional subspace spanned by eigenvectors associated with eigenvalues $\lambda_1$ and $\lambda_2$, $\lambda_1\neq\lambda_2$, $\theta\in(0,\pi/2]$ is the angle between the eigenvectors \[in the Hermitian sense of Eq. \], and $$\theta < \arctan\left( \frac{\left| \lambda_1 - \lambda_2 \right|}{2 \sqrt{\mathrm{Re}(\lambda_1) \mathrm{Re}(\lambda_2)}} \right).
\label{eq:theta-exp}$$
\[thm:transient-power\] Let $\mathbf{M}$ be a nonsingular $N\times N$ complex matrix ($N\ge 2$) with all of its eigenvalues in the open unit circle of $\mathbb{C}$; that is, $0<|\lambda|<1$ if $\lambda$ is an eigenvalue of $\mathbf{M}$. Let $S$ be an invariant subspace of $\mathbf{M}$. Then $\mathbf{M}^n$ will have transient effects when at least one of the following conditions is satisfied.
1. $S$ is a $M$-dimensional ($M\ge2$) Jordan subspace.
2. $S$ is a 2-dimensional subspace spanned by eigenvectors associated with eigenvalues $\lambda_1$ and $\lambda_2$, $\lambda_1\neq\lambda_2$, $\theta\in(0,\pi/2]$ is the angle between the eigenvectors \[in the Hermitian sense of Eq. \], and $$\theta < \arctan\left( \frac{\left| \ln (\lambda_1) - \ln(\lambda_2) \right|}{2 \sqrt{ \ln |\lambda_1| \ln |\lambda_2| } } \right).
\label{eq:theta-power}$$ The branch of $\ln(z)$ can always be chosen to make the imaginary part of the numerator of Eq. be in $[-\pi,\pi)$.
Preliminaries and Notation {#sec:prelims}
==========================
Several standard results from matrix analysis will be needed for the discussion and proofs of Thms. \[thm:transient-exp\] and \[thm:transient-power\]. We overview them in this section, which also introduces our notation. Throughout this section, we assume $\mathbf{M}\in\mathbb{C}^{N\times N}$. The angle between two vectors will be calculated in the Hermitian sense [@galantai-589-2006] with $$\cos(\theta) = \frac{\left|v_1 \cdot{} v_2\right|}{\left\| v_1 \right\| \left\| v_2 \right\|}
\label{eq:vector-angle}$$ such that $\theta\in[0,\pi/2]$.
Schur Decomposition, Unitary Invariance, and Matrix Norm {#sec:prelims:schur}
--------------------------------------------------------
Every $\mathbf{M}$ is unitarily similar to an upper triangular matrix; that is, $\mathbf{M} = \mathbf{QTQ}^{-1}$, where $\mathbf{Q}$ is unitary and $\mathbf{T}$ is upper triangular. $\mathbf{T}$ is called a Schur form of $\mathbf{M}$ [@bk:golub-2013]. It is trivial to show that $\mathbf{M}^n=\mathbf{QT}^n\mathbf{Q}^{-1}$ and, by use of the matrix exponential’s Taylor series, $e^{\mathbf{M}t}=\mathbf{Q}e^{\mathbf{T}t}\mathbf{Q}^{-1}$ [@bk:higham-2008]. Then, assuming the 2-norm or the Frobenius norm for the matrix norm, we get that $\|e^{\mathbf{M}t}\| = \|e^{\mathbf{T}t}\|$ and $\|\mathbf{M}^n\|=\|\mathbf{T}^n\|$ because these matrix norms are invariant to unitary transformations [@bk:golub-2013]. It is therefore sufficient to consider a Schur form of $\mathbf{M}$ in our analyses.
Invariant Subspaces and Restrictions {#sec:prelims:subspaces}
------------------------------------
A subspace $S\subseteq\mathbb{C}^N$ is an invariant subspace of $\mathbf{M}$ if $\mathbf{M}x\in S$ for every $x\in S$ [@bk:gohberg-2006]. The invariant subspaces of $\mathbf{M}$ are closely tied to the Jordan decomposition of $\mathbf{M}$; every invariant subspace of $\mathbf{M}$ is spanned by a set of eigenvectors and generalized eigenvectors of $\mathbf{M}$. In this way, each invariant subspace of $\mathbf{M}$ can be associated with eigenvalues of $\mathbf{M}$. An invariant subspace of $\mathbf{M}$ spanned by eigenvectors and generalized eigenvectors that all belong to the same Jordan chain is called a Jordan subspace of $\mathbf{M}$. A 2- or higher-dimensional Jordan subspace of $\mathbf{M}$ is thus associated with a sole defective eigenvalue of $\mathbf{M}$.
We will also need the restriction of $\mathbf{M}$ onto an invariant subspace $S$, denoted $\left.\mathbf{M}\right|_S:S\to S$ and defined by $$\left.\mathbf{M}\right|_S x = \mathbf{M}x$$ for $x\in S$. Thus, $\left.\mathbf{M}\right|_S$ can be regarded as a $\mathrm{dim}(S)\times\mathrm{dim}(S)$ matrix. If $f$ is a well-behaved function, such as the exponential, then the Jordan form of $\mathbf{M}$ and its relation to $S$ shows that [@bk:higham-2008] $$\left.f(\mathbf{M})\right|_S = f\left( \left. \mathbf{M} \right|_S \right).$$
Numerical Range and Numerical Abscissa {#sec:prelimes:nr}
--------------------------------------
The numerical range of $\mathbf{M}$ is the compact, convex subset of $\mathbb{C}$ given by [@bk:horn-1991] $$W(\mathbf{M}) = \left\{ x \cdot{} \mathbf{M} x : x\in\mathbb{C}^N, \| x \| = 1 \right\}.
\label{eq:numerical-range}$$ The numerical abscissa, $$\omega(\mathbf{M}) = \max \left\{\mathrm{Re}(z) : z\in W(\mathbf{M}) \right\},
\label{eq:numerical-abscissa}$$ comes from the boundary of $W(\mathbf{M})$ and relates to transient effects in $e^{\mathbf{M}t}$ [@bk:trefethen-2005]. An example of the numerical range and numerical abscissa is displayed in Figure \[fig:nr-ellipse\].
Transient Effects in the Matrix Exponential and Matrix Powers {#sec:prelims:transients}
-------------------------------------------------------------
Transient effects in the matrix exponential $e^{\mathbf{M}t}$ occur when all eigenvalues of $\mathbf{M}$ have negative real part and $\|e^{\mathbf{M}t}\| > 1$ for some $t>0$. Clearly, $\| e^{\mathbf{M}t} \|=1$ when $t=0$ such that a sufficient condition for transient behavior is $$\left. \frac{\mathrm{d}}{\mathrm{d}t} \left\| e^{\mathbf{M}t} \right\| \right|_{t=0^+} > 0.$$ From Eq. (14.2) in [@bk:trefethen-2005], $$\left. \frac{\mathrm{d}}{\mathrm{d}t} \left\| e^{\mathbf{M}t} \right\| \right|_{t=0^+} = \omega(\mathbf{M}),$$ Thus, a sufficient condition for the existence of transient effects in $e^{\mathbf{M}t}$ is $$\omega(\mathbf{M}) > 0.
\label{eq:exist-transient-exp}$$ Transient effects in the matrix exponential are consequently related to the boundary of $W(\mathbf{M})$.
Other bounds for predicting the existence of transient effects in $e^{\mathbf{M}t}$ can be found in chapter 14 of [@bk:trefethen-2005], and involve (for example) the pseudospectra of $\mathbf{M}$ or the Kreiss constant of $\mathbf{M}$. These quantities are “holistic”, meaning they examine $\mathbf{M}$ in its entirety. Our analyses in sections \[sec:2by2\] and \[sec:higherD\] will look at the relationship between specific invariant subspaces of $\mathbf{M}$ and transient effects, thereby providing a more fine-grained analysis of transient effects.
Transient effects in $\mathbf{M}^n$ are similar. In this case the eigenvalues of $\mathbf{M}$ are less than one in magnitude and $\| \mathbf{M}^n \|>1$ for some $n>0$. As before, a discussion of transient effects in the matrix powers can be found in chapter 14 of [@bk:trefethen-2005], which includes bounds that depend on holistic properties of $\mathbf{M}$.
Analytical Results on $2\times 2$ Matrices {#sec:2by2}
==========================================
Many of the conditions in Thms. \[thm:transient-exp\] and \[thm:transient-power\] come from analytical results on transient effects in $2\times2$ matrices. This section develops these results. First in section \[sec:2by2:nr\] we discuss the numerical range of $2\times 2$ matrices [@uhlig-541-1985; @bk:horn-1991]. Section \[sec:2by2:parameterize\] then parameterizes $\mathbb{C}^{2\times 2}$ in a way that facilitates our analyses. The key parameter is the angle between the matrix’s eigenvectors (assuming the matrix is not defective). Finally, sections \[sec:2by2:exp\] and \[sec:2by2:power\] derive sufficient conditions for the matrix exponential and matrix powers of a $2\times 2$ matrix, respectively, to display transient behavior.
The Numerical Range {#sec:2by2:nr}
-------------------
The numerical range of a $2\times 2$ matrix $\mathbf{M}$ is a (possibly-degenerate) ellipse with boundary given by Thm. \[thm:nr-ellipse\] [@uhlig-541-1985; @bk:horn-1991]. The center of the ellipse is $\mathrm{tr}(\mathbf{M})/2$, the eigenvalues of $\mathbf{M}$ are the foci of the ellipse, and the major and minor axes are stated in Lemma \[lemma:axes-ellipse\]. Many of the ensuing results will come from analyzing the geometry of this ellipse, for which an example is displayed in Figure \[fig:nr-ellipse\].
In the complex plane $\mathbb{C}\simeq \mathbb{R}^2$, the boundary curve of the numerical range of $\mathbf{M}\in\mathbb{C}^{2\times2}$ is $$\left[ x - \mathrm{Re}\left( \frac{\mathrm{tr}(\mathbf{M})}{2} \right), y - \mathrm{Im}\left( \frac{\mathrm{tr}(\mathbf{M})}{2} \right) \right] \mathbf{S} \left[ \begin{array}{c} x - \mathrm{Re}\left( \frac{\mathrm{tr}(\mathbf{M})}{2} \right) \\ y - \mathrm{Im}\left( \frac{\mathrm{tr}(\mathbf{M})}{2} \right) \end{array} \right] = \frac{\det(\mathbf{S})}{4},
\label{eq:nr-ellipse}$$ where $\left[ \begin{array}{c} x \\ y \end{array} \right]\in\mathbb{R}^2$, $$\mathbf{S} = \left[ \begin{array}{cc}
\| \mathbf{M}_0 \|_\mathrm{F}^2 + 2 \, \mathrm{Re}(\det(\mathbf{M}_0)) & 2 \, \mathrm{Im}(\det(\mathbf{M}_0)) \\
2 \, \mathrm{Im}(\det(\mathbf{M}_0)) & \| \mathbf{M}_0 \|_\mathrm{F}^2 - 2 \, \mathrm{Re}(\det(\mathbf{M}_0))
\end{array} \right],$$ $\mathbf{M}_0 = \mathbf{M}-(\mathrm{tr}(\mathbf{M})/2)\mathbf{I}$, and $\|\cdot\|_\mathrm{F}$ is the Frobenius norm. \[thm:nr-ellipse\]
The major and minor axes of the elliptical numerical range of $\mathbf{M}\in\mathbb{C}^{2\times 2}$ are
$$\label{eq:nr-major-axis}
2\sqrt{\left\| \mathbf{M}_0 \right\|_\mathrm{F}^2 + 2 \left| \mathrm{det}(\mathbf{M}_0) \right|}$$
and $$\label{eq:nr-minor-axis}
2\sqrt{\left\| \mathbf{M}_0 \right\|_\mathrm{F}^2 - 2 \left| \mathrm{det}(\mathbf{M}_0) \right|},$$
respectively, where $\mathbf{M}_0 = \mathbf{M}-(\mathrm{tr}(\mathbf{M})/2)\mathbf{I}$, and $\|\cdot\|_\mathrm{F}$ is the Frobenius norm. \[lemma:axes-ellipse\]
Parameterizing $2\times 2$ Non-Normal Matrices {#sec:2by2:parameterize}
----------------------------------------------
Before developing the results for $2\times 2$ matrices, we first need to parameterize $\mathbb{C}^{2\times 2}$ using quantities that will facilitate our analysis. Appealing to the Jordan form, there are three cases.
First is when the matrix is diagonalizable and has a degenerate eigenvalue. In this case the matrix is a scalar multiple of the identity matrix; it is normal and not of interest to this discussion. We will not consider this case further.
Second is when the matrix is diagonalizable but with distinct eigenvalues, $\lambda_1$ and $\lambda_2$. We will denote this matrix by $\mathbf{A}$. From section \[sec:prelims:schur\] we can assume without loss of generality that $\mathbf{A}$ is upper triangular: $$\mathbf{A} = \left[ \begin{array}{cc}
\lambda_1 & a \\
0 & \lambda_2
\end{array} \right],$$ where $a,\lambda_1,\lambda_2\in\mathbb{C}$ and $\lambda_1\neq\lambda_2$. It is straightforward to see that $\left[ \begin{array}{cc} 1 & 0 \end{array} \right]^T$ is a right eigenvector associated with $\lambda_1$ and that $\left[ \begin{array}{cc} a & \lambda_2-\lambda_1 \end{array} \right]^T$ is a right eigenvector associated with $\lambda_2$. Then, the angle between the eigenvectors \[Eq. \] is $$\cos(\theta) = \frac{|a|}{\sqrt{|a|^2+|\lambda_2-\lambda_1|^2}},
\label{eq:matrix-a:a-theta}$$ which yields $$|a| = |\lambda_2-\lambda_1|\cot(\theta).$$ Then, for $0\le\varphi<2\pi$, we get $$\mathbf{A} = \left[ \begin{array}{cc}
\lambda_1 & e^{i\varphi} \left|\lambda_2-\lambda_1\right| \cot(\theta) \\
0 & \lambda_2
\end{array} \right].
\label{eq:matrix-a}$$ The angle between the eigenvectors can thus be used to parameterize the non-normality of $\mathbf{A}$. This idea of examining the angle between eigenvectors comes from the condition number of an eigenvalue [@bk:heath-2002], which essentially considers the angle between the eigenvalue’s left and right eigenvectors.
Third is when the matrix is defective, with sole eigenvalue $\lambda$. This matrix will be called $\mathbf{D}$. The Schur form is the $2\times2$ Jordan block because the generalized eigenvector can always be chosen to be orthogonal to the eigenvector. Thus, $$\mathbf{D} = \left[ \begin{array}{cc}
\lambda & 1 \\
0 & \lambda
\end{array} \right].
\label{eq:matrix-d}$$
Some useful properties of $\mathbf{A}$ and $\mathbf{D}$ are stated in the following Lemmas.
The major and minor axes of $W(\mathbf{A})$ are
$$2\sqrt{\left( | \lambda_1 | + | \lambda_2 | \right)^2 + \left| \lambda_1 - \lambda_2 \right|^2 \cot^2(\theta)}
\label{eq:major-axis-a}$$
and $$2\sqrt{\left( | \lambda_1 | - | \lambda_2 | \right)^2 + \left| \lambda_1 - \lambda_2 \right|^2 \cot^2(\theta)},
\label{eq:minor-axis-a}$$
respectively. \[lemma:axes-a\]
The proof is trivial following from Eq. and Lemma \[lemma:axes-ellipse\].
The numerical abscissa of $\mathbf{A}$ is $$\omega(\mathbf{A}) = \frac{\sqrt{\left| \lambda_1 - \lambda_2 \right|^2 \left( 1+2\cot^2(\theta) \right) + \mathrm{Re}\left[ (\lambda_1-\lambda_2)^2 \right]}}{2\sqrt{2}} + \frac{\mathrm{Re}(\lambda_1 + \lambda_2)}{2}.
\label{eq:num-abscissa-a}$$ \[lemma:num-abscissa-a\]
$\mathbf{S}$ in Thm. \[thm:nr-ellipse\] is $$\mathbf{S} = \left[ \begin{array}{cc}
\left| \lambda_1 - \lambda_2 \right|^2 \left( \frac{1}{2} + \cot^2(\theta) \right) - \frac{1}{2} \mathrm{Re}\left[ \left(\lambda_1-\lambda_2\right)^2 \right] & - \frac{1}{2} \mathrm{Im}\left[ \left(\lambda_1-\lambda_2\right)^2 \right] \\
- \frac{1}{2} \mathrm{Im}\left[ \left(\lambda_1-\lambda_2\right)^2 \right] & \left| \lambda_1 - \lambda_2 \right|^2 \left( \frac{1}{2} + \cot^2(\theta) \right) + \frac{1}{2} \mathrm{Re}\left[ \left(\lambda_1-\lambda_2\right)^2 \right]
\end{array} \right].$$ Then, recast Eq. as a constrained optimization problem: Maximize $x$ subject to Eq. . This can be solved using standard techniques, *e.g.*, Lagrange multipliers, which produces the result.
\[lemma:num-abscissa-d\] The numerical abscissa of $\mathbf{D}$ is $\omega(\mathbf{D}) = \mathrm{Re}(\lambda) + 1/2$.
The proof is very similar to that of Lemma \[lemma:num-abscissa-a\]. The matrix $\mathbf{S}$ from Thm. \[thm:nr-ellipse\] is the identity matrix such that the boundary of $W(\mathbf{D})$ is $$\left(x-\mathrm{Re}(\lambda) \right)^2 + \left( y - \mathrm{Im}(\lambda) \right)^2 = \frac{1}{4}.$$ The right-most point of the boundary curve is trivially $(\mathrm{Re}(\lambda)+1/2,\mathrm{Im}(\lambda))$; thus, $\omega(\mathbf{D}) = \mathrm{Re}(\lambda)+1/2$. Note that a circular numerical range is expected for nilpotent operators [@karaev-2321-2004] and $W(\mathbf{D})$ having a radius of $1/2$ is consistent with [@haagerup-371-1992; @wu-351-1998]; see Thm. \[thm:nr-jordan\] below.
Matrix Exponential ($2\times 2$ Matrices) {#sec:2by2:exp}
-----------------------------------------
As discussed in section \[sec:prelims:transients\], $e^{\mathbf{M}t}$ has transient behavior when $\omega(\mathbf{M}) > 0$. We can then examine the non-defective matrix $\mathbf{A}$ in Eq. and the defective matrix $\mathbf{D}$ in Eq. using this condition.
\[lemma:2by2:theta-exp\] Let $\mathbf{A}$ be given as in Eq. with $\lambda_1\neq\lambda_2$ and $\mathrm{Re}(\lambda_1),\mathrm{Re}(\lambda_2)<0$. Then, $e^{\mathbf{A}t}$ will have transient effects when $$\theta < \arctan\left( \frac{\left| \lambda_1 - \lambda_2 \right|}{2 \sqrt{\mathrm{Re}(\lambda_1) \mathrm{Re}(\lambda_2)}} \right).$$
Use Lemma \[lemma:num-abscissa-a\] to solve $\omega(\mathbf{A}) > 0$ for $\theta$.
\[cor:2by2:a-theta\] Let $\mathbf{M}\in\mathbb{C}^{2\times2}$ have the form (Schur form) $$\mathbf{M} = \left[ \begin{array}{cc}
\lambda_1 & a \\
0 & \lambda_2 \\
\end{array} \right],$$ with $\lambda_1\neq\lambda_2$ and $\mathrm{Re}(\lambda_1),\mathrm{Re}(\lambda_2)<0$. Then, $e^{\mathbf{M}t}$ will have transient effects when $|
a|>2\sqrt{\mathrm{Re}(\lambda_1)\mathrm{Re}(\lambda_2)}$.
The earlier discussion \[Eq. \] relates $a$ to $\theta$, the angle between the eigenvectors of $\mathbf{M}$. Using trigonometry, $$\theta = \arctan\left( \frac{|\lambda_2-\lambda_1|}{|a|} \right),$$ which provides the result when combined with Lemma \[lemma:2by2:theta-exp\].
\[lemma:2by2:defective-exp\] Let $\mathbf{D}$ be given as in Eq. with $\mathrm{Re}(\lambda)<0$. Then, $e^{\mathbf{D}t}$ will have transient effects when $\mathrm{Re}(\lambda)>-1/2$.
Use Lemma \[lemma:num-abscissa-d\] to solve $\omega(\mathbf{D}) > 0$ for $\mathrm{Re}(\lambda)$.
Matrix Powers ($2\times 2$ Matrices) {#sec:2by2:power}
------------------------------------
Let us now turn our attention to matrix powers $\mathbf{M}^n$. Because the existence of $\mathbf{M}^n$ is complicated for singular $\mathbf{M}$, we will restrict our attention to nonsingular $\mathbf{M}$. The results here follow straightforwardly from Lemmas \[lemma:2by2:theta-exp\] and \[lemma:2by2:defective-exp\] for the matrix exponential.
\[lemma:2by2:theta-power\] Let $\mathbf{A}$ be nonsingular and given as in Eq. with $\lambda_1\neq\lambda_2$ and $0<|\lambda_1|,|\lambda_2|<1$. Then, $\mathbf{A}^n$ will have transient effects when $$\theta < \arctan\left( \frac{\left| \ln (\lambda_1) - \ln(\lambda_2) \right|}{2 \sqrt{ \ln |\lambda_1| \ln |\lambda_2| } } \right).$$
Because $\mathbf{A}$ is not singular, we can write $\mathbf{A}^n=e^{n\ln(\mathbf{A})}$ by choosing the logarithmic branch cut to avoid both eigenvalues [@bk:higham-2008]. The result then trivially follows from Lemma \[lemma:2by2:theta-exp\] by using $\ln(\lambda_1)$ and $\ln(\lambda_2)$ as eigenvalues. Note that the logarithmic branch cut can always be chosen to make the imaginary part of $\ln(\lambda_1)-\ln(\lambda_2)$ be in $[-\pi,\pi)$.
\[lemma:2by2:defective-power\] Let $\mathbf{D}$ be nonsingular and given as in Eq. with $0<|\lambda|<1$. Then, $\mathbf{D}^n$ will have transient effects.
The proof is similar to that of Lemmas \[lemma:num-abscissa-d\] and \[lemma:2by2:theta-power\]. Because $\lambda\neq0$, $\mathbf{D}^n=e^{n\ln(\mathbf{D})}$. It can be shown that [@bk:higham-2008] $$\ln(\mathbf{D}) = \left[ \begin{array}{cc} \ln(\lambda) & \lambda^{-1} \\ 0 & \ln(\lambda) \end{array} \right],$$ which leads to $\mathbf{S} = \mathbf{I}/|\lambda|^2$ in Thm. \[thm:nr-ellipse\]. Noting that $\mathrm{Re}(\ln(\lambda)) = \ln|\lambda|$, the boundary of $W(\ln(\mathbf{D}))$ is $\left(x-\ln|\lambda|\right)^2 + \left( y - \mathrm{arg}(\lambda) \right)^2 = 1/(4|\lambda|^2)$. Thus, $$\omega(\ln(\mathbf{D}))=\ln|\lambda| + \frac{1}{2|\lambda|}.$$ One can verify that $\omega(\ln(\mathbf{D}))>0$ for $0<|\lambda|<1$, implying that there will always be transient effects.
Lemmas \[lemma:2by2:theta-exp\], \[lemma:2by2:defective-exp\], \[lemma:2by2:theta-power\], and \[lemma:2by2:defective-power\] collectively summarize sufficient conditions for $\mathbf{M}\in\mathbb{C}^{2\times 2}$ to have transient effects in its matrix exponential or matrix powers ($\mathbf{M}$ nonsingular assumed for the matrix powers). Note that Lemmas \[lemma:2by2:theta-power\] and \[lemma:2by2:defective-power\] consider $n$ to be continuous such that transient effects may not be noticeable when $n$ is restricted to positive integers.
Although not proven, numerical experiments suggest that the conditions in Lemmas \[lemma:2by2:theta-exp\], \[lemma:2by2:defective-exp\], and \[lemma:2by2:theta-power\] are also necessary for transient effects in $e^{\mathbf{M}t}$ or $\mathbf{M}^n$ when $\mathbf{M}\in\mathbb{C}^{2\times 2}$. Future work should be performed to confirm or refute this conjecture.
Higher-Dimensional Matrices (Proofs of Thms. \[thm:transient-exp\] and \[thm:transient-power\]) {#sec:higherD}
===============================================================================================
The conditions for transient effects in $2\times 2$ matrices that we developed in the previous section readily generalize to $N\times N$ matrices by considering the 2-dimensional invariant subspaces of $\mathbf{M}\in\mathbb{C}^{N\times N}$. Theorem \[thm:subspaces\] states the relationship.
Let $\mathbf{M}\in\mathbb{C}^{N \times N}$ and let $S\subseteq\mathbb{C}^N$ be an invariant subspace of $\mathbf{M}$. If $e^{\left.\mathbf{M}\right|_S t}$ admits transient behavior, then so will $e^{\mathbf{M}t}$. A similar statement holds for $\left( \left. \mathbf{M} \right|_S \right)^n$ and $\mathbf{M}^n$. \[thm:subspaces\]
Transient effects in $e^{\mathbf{M}t}$ require $\|e^{\mathbf{M}t}\|>1$ for some $t>0$. Because $\left.\mathbf{M}\right|_S$ admits transient behavior, we know that $1<\| e^{\left.\mathbf{M}\right|_St} \|$ for some $t>0$. Then, using the definition of the matrix norm, $$1 < \left\| e^{\left. \mathbf{M} \right|_S t} \right\| = \max_{\|x\|=1} \left\| e^{\left. \mathbf{M} \right|_S t}x \right\| = \max_{x\in S, \|x\|=1} \left\| e^{\mathbf{M}t}x \right\| \le \max_{\|x\|=1} \left\| e^{\mathbf{M}t}x \right\| = \left\| e^{\mathbf{M}t} \right\|.$$ Thus, if $\left.\mathbf{M}\right|_S$ admits transient behavior in the matrix exponential, so must $\mathbf{M}$. Similar logic is used to show that if $\left( \left. \mathbf{M} \right|_S \right)^n$ has transient effects, then $\mathbf{M}^n$ will as well.
Using Thm. \[thm:subspaces\], Lemma \[lemma:2by2:theta-exp\] proves the second condition of Thm. \[thm:transient-exp\] and Lemmas \[lemma:2by2:defective-power\] and \[lemma:2by2:theta-power\] prove the first and second conditions of Thm. \[thm:transient-power\], respectively. Lemma \[lemma:2by2:defective-exp\] shows the first condition of Thm. \[thm:transient-exp\] for 2-dimensional Jordan subspaces. [@haagerup-371-1992; @wu-351-1998] allow us to consider Jordan subspaces of arbitrary dimension because the numerical range is analytically known for these cases.
Let $\mathbf{J}$ be a $M\times M$ Jordan block ($M\ge 2$) with eigenvalue $\lambda$. Then $W(\mathbf{J})$ is all points $z\in\mathbb{C}$ satisfying $$\left| z - \lambda \right| \le \cos\left( \frac{\pi}{M+1} \right).$$ Consequently, $$\omega(\mathbf{J})= \mathrm{Re}(\lambda) + \cos\left( \frac{\pi}{M+1} \right).$$ \[thm:nr-jordan\]
The first condition in Thm. \[thm:transient-exp\] trivially follows from Eq. and Thm. \[thm:nr-jordan\]. This completes the proofs of Thms. \[thm:transient-exp\] and \[thm:transient-power\].
Geometric Interpretations {#sec:geom}
=========================
The results in section \[sec:2by2\] largely came from analyzing a matrix $\mathbf{M}\in\mathbb{C}^{2\times 2}$ in terms of the angle between its eigenvectors (assuming $\mathbf{M}$ is not defective). This alone leads to new geometric interpretations for transient effects in the matrix exponential and matrix powers of $\mathbf{M}$, but two other geometric observations can be made as well.
First regards the numerical detection of a defective eigenvalue. Due to numerical instabilities in computing the eigenvalues of a non-normal matrix [@bk:trefethen-2005] — especially a defective eigenvalue — it can be very difficult to distinguish (*e.g.*) two eigenvalues that are close to each other from a defective eigenvalue. A recent discussion of this issue appears in [@zeng-798-2016]. Let $S$ be the 2-dimensional invariant subspace of $\mathbf{M}$ corresponding to these two eigenvalues. An examination of $W\left(\left. \mathbf{M}\right|_S \right)$ provides insight into the situation. Without loss of generality [@uhlig-541-1985], assume that $\mathrm{tr}\left( \left. \mathbf{M} \right|_S \right)=0$ such that $\lambda$ and $-\lambda$ are the eigenvalues of $\left.\mathbf{M}\right|_S$. From Lemma \[lemma:axes-a\], the major and minor axes of $W\left( \left. \mathbf{M} \right|_S \right)$ are $4|\lambda|\csc(\theta)$ and $4|\lambda|\cot(\theta)$, respectively. As $\theta\to0^+$, meaning $\left.\mathbf{M}\right|_S$ approaches a defective matrix, $W\left( \left. \mathbf{M} \right|_S \right)$ will tend toward a circle. Examining the shape of the restricted numerical range for the invariant subspace may help determine if it is a Jordan subspace or not.
Second is a relation to antieigenvalue analyses [@bk:gustafson-2012]. When $\mathbf{M}$ is a self-adjoint, positive-definite matrix (and thus normal), its antieigenvector is the vector that is rotated the most by $\mathbf{M}$. The antieigenvalue is the corresponding angle of rotation, which will be in $[0,\pi/2)$. If $\mathbf{M}$ is $2\times 2$, self-adjoint, and negative-definite with eigenvalues $\lambda_1$ and $\lambda_2$, its antieigenvalue is [@bk:gustafson-2012] $$\arcsin\left( \frac{|\lambda_1 - \lambda_2|}{|\lambda_1 + \lambda_2|} \right),$$ where the angle of rotation is measured in the Hermitian sense of Eq. . Compare this against the critical angle where $\omega(\mathbf{A})=0$ in Lemma \[lemma:2by2:theta-exp\], which is $$\arctan\left( \frac{\left| \lambda_1 - \lambda_2 \right|}{2\sqrt{\lambda_1\lambda_2}} \right)$$ when $\lambda_1,\lambda_2\in\mathbb{R}$ and $\lambda_1,\lambda_2<0$. Some basic trigonometry reveals that these two angles are the same; the critical angle for transient effects in $e^{\mathbf{M}t}$ is exactly the antieigenvalue of a normal matrix with the same real eigenvalues. That is, transient effects appear in $e^{\mathbf{M}t}$ when the normal matrix with the same eigenvalues can rotate vectors more than the angle between the eigenvectors of $\mathbf{M}$. Future work may consider generalizing this connection to matrices with complex eigenvalues, for which an antieigenvalue analysis is not well developed.
Concluding Remarks {#sec:conclusions}
==================
In this work we have established a relationship (Thm. \[thm:subspaces\]) between the invariant subspaces of a non-normal matrix $\mathbf{M}\in\mathbb{C}^{N\times N}$ and transient behavior in $e^{\mathbf{M}t}$ ($t>0$) and $\mathbf{M}^n$ ($n>0$). This then allowed us to derive sufficient conditions for transient effects in the exponential and powers of $\mathbf{M}$, as stated in Thms. \[thm:transient-exp\] and \[thm:transient-power\], respectively. Most of the conditions came from analytic results for 2-dimensional invariant subspaces of $\mathbf{M}$, where the angle between the two associated eigenvectors is a key parameter. We should note that the conditions in Thms. \[thm:transient-exp\] and \[thm:transient-power\] are sufficient but not necessary for transient effects; it is possible that transient effects can only be associated with higher-dimensional invariant subspaces.
Consequently, one immediate next step is to analyze higher-dimensional invariant subspaces for conditions on transient effects. Although not as straightforward as the case for $2\times 2$ matrices, the numerical range of $3\times 3$ matrices is discussed in [@keeler-115-1997]. Their results can be used with Eq. to develop appropriate conditions. Along these lines, the Jordan principal angles [@galantai-589-2006] may be useful when generalizing the angle between eigenvectors to angles between invariant subspaces of arbitrary dimensions.
In a broader sense such investigations are examining structure within $W(\mathbf{M})$ [@uhlig-055019-2008; @dunkl-2042-2011]. For instance, what is the lowest-dimensional invariant subspace $S$ of $\mathbf{M}$ that satisfies $\omega\left(\left.\mathbf{M}\right|_S\right)>0$, thereby inducing transient behavior? Ideas on the restricted numerical range [@das-35-1987; @gawron-102204-2010] may help answer this question. Finally, we end with a practical comment. All of our results make use of the Jordan decomposition of a non-normal $\mathbf{M}$, which can be difficult to compute accurately [@bk:trefethen-2005]. Although the results are theoretically useful, additional work needs to be performed to establish their numerical utility.
Acknowledgments {#acknowledgments .unnumbered}
===============
I thank Jay Bardhan, Thorsten Hansen, Christopher DeGrendele, and Jonathan Kazakov for helpful conversations.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'K. Schultheiss'
- 'R. Verba'
- 'F. Wehrmann'
- 'K. Wagner'
- 'L. Körber'
- 'T. Hula'
- 'T. Hache'
- 'A. Kakay'
- 'A.A. Awad'
- 'V. Tiberkevich'
- 'A.N. Slavin'
- 'J. Fassbender'
- 'H. Schultheiss'
title: Excitation of whispering gallery magnons in a magnetic vortex
---
Whispering gallery magnons can only live in systems with rotational symmetry. This not only applies to the geometry of the magnetic element but also to the magnetisation texture therein. For that reason, we study a Ni$_{81}$Fe$_{19}$ disc that inherently exhibits a magnetic vortex structure[@Shinjo; @Novosad2002; @Guslienko2004; @Novosad2005; @Guslienko2008]. The arrows in Fig. \[fig1\]a depict the generic features of such a vortex state: the magnetic moments curl in plane along circular lines around the vortex core, a nanoscopic region in the center of the disc where the magnetisation tilts out of plane. According to this rotational symmetry, the magnon eigenmodes in a vortex are characterised by mode numbers ($n,m$), with $n = 0, 1, 2, ...$ counting the number of nodes across the disc radius and $m = 0, \pm1, \pm2, ...$ counting the number of nodes in azimuthal direction over half the disc [@Kalinikos1986; @Buess2004].
Other than commonly known waves, like sound, water or electromagnetic waves, magnons exhibit a strongly anisotropic dispersion relation in in-plane magnetised thin films[@Kalinikos1986]. In a vortex, this results in increasing (decreasing) mode energies for increasing $n$ ($m$) as shown by the analytic calculations in Fig. \[fig1\]b. The four exemplary intensity profiles for the eigenmodes $(0,0)$, $(0,10)$, $(0,20)$, and $(0,30)$, that are shown in Fig. \[fig1\]c, reveal the character of whispering gallery magnons: the larger $m$, the more the magnon intensity is pushed toward the perimeter of the disc which can be understood intuitively by the reduction of exchange energy: Leaving an extended area around the vortex core with zero amplitude avoids a strong tilt of neighbouring spins close to the vortex core and, therefore, reduces the total energy.
Even though magnon spectra in magnetic vortices have been intensively studied in the past [@Buess2004; @Zivieri2005; @Guslienko2008; @Awad2010], magnons with large azimuthal wave vectors have not yet been measured experimentally and were only observed in micromagnetic simulations[@Taurel2016]. The challenge to generate such magnons and, thereby, to reach out to whispering gallery magnons is finding an efficient excitation mechanism in a micron-sized vortex. Here, we tackle this problem via nonlinear 3-magnon scattering. In this process, one magnon splits in two new magnons under conservation of energy and momentum. The rotational symmetry of the vortex texture implies specific selection rules for the scattering process which we will describe in context with the experimental data.
In order to selectively drive magnetisation dynamics, we apply microwave currents to an $\Omega$-shaped antenna that encloses the vortex. Inside the $\Omega$ loop, a spatially uniform magnetic field is generated that is oriented perpendicularly to the disc as shown in Fig. \[fig1\]d. The rotational symmetry of this magnetic field prohibits direct coupling to magnons with $m\neq0$. However, because of the small diameter of the antenna, strong magnetic fields can be generated so that these magnons can be indirectly driven in the nonlinear regime via multi-magnon scattering processes.
We track these nonlinear processes by measuring magnon spectra as a function of the applied microwave frequency using Brillouin light scattering (BLS) microscopy[@Sebastian2015]. We would like to emphasise that even though the system is driven with one microwave frequency at a time, the BLS technique allows us to detect the dynamic magnetic response in a broad frequency range. In Fig. \[fig2\]a-c we plot the measured BLS spectra between 2 and 11GHz ($y$-axis) for each excitation frequency ($x$-axis) at microwave powers of 1, 10, and 200mW. The magnon intensity is encoded using the same logarithmic scale shown as an inset in Fig. \[fig2\]a.
At the lowest microwave power of 1mW (Fig. \[fig2\]a), magnons are driven in the linear regime, which is corroborated by the fact that magnons are only observed at the BLS frequency that matches the applied microwave frequency $f_\mathrm{BLS}=f_\mathrm{0}$. Hence, the measured intensities strictly follow the diagonal, dashed line. Four distinct resonances emerge at 5.55, 7.40, 8.75, and 9.65GHz, which we identify as the well known first four radial modes[@Buess2004; @Vogt2011] by spatially-resolved BLS microscopy (insets in Fig. \[fig2\]a).
For a power of 10mW (Fig. \[fig2\]b) the excitation field is already strong enough to drive magnons in the nonlinear regime. Hence, we observe strong off-diagonal signals that appear at BLS frequencies symmetrically spaced around half the excitation frequency $f_\mathrm{0}/2$ (straight line with slope 0.5). These satellite peaks are the result of 3-magnon splitting processes. In order to conserve energy the initial magnon with frequency $f_{0}$ splits in two magnons with frequencies $f_{1}=f_{0}/2-\delta f$ and $f_{2}=f_{0}/2+\delta f$. Moreover, the rotational symmetry of the vortex requires conservation of the momentum component in azimuthal direction. For an initial magnon with $m=0$ this implies that the split modes have azimuthal mode numbers with the same modulus but opposite sign: $m_{1}=-m_{2}$. Our analytic calculations further show that the split modes may not share the same radial index, [*i.e.*]{}, $n_{1} \neq n_{2}$ (see methods). All three selection rules drastically restrict the possible scattering channels within the discrete eigenmode spectrum of the vortex (see Fig. \[fig1\]b). At the maximum applied microwave power of 200mW, the number of off-diagonal signals increases further (Fig. \[fig2\]c). Especially, for excitation frequencies between $6$ and $7$GHz, we do not just measure two satellite peaks with frequencies $f_{1}$ and $f_{2}$ but a total number of ten additional modes. Their presence is attributed to avalanche processes of higher-order multi-magnon scattering. Their frequencies are given by combinations of the three initial magnons, e.g., $2 f_{1}$, $2 f_{2}$, $f_{0}+f_{1}$. Furthermore, the significant line broadening (additional noise) of the directly excited mode and the split modes in the frequency range between $5.3$ and $5.9$GHz can be attributed to 4-magnon scattering[@Schultheiss2012]. However, this article solely focusses on the study of the initial 3-magnon scattering process which clearly dominate in intensity.
To better illustrate the power dependence of the observed modes, we plot the BLS intensity integrated over different frequency windows as a function of the excitation frequency in Fig. \[fig2\]d. The black data resembles the BLS intensity of the direct excitation. With increasing power the initially sharp resonances become broader and show the characteristic nonlinear foldover to higher frequencies [@Suhl; @Janantha]. The red and blue data in Fig. \[fig2\]d show the intensities of the split modes below (red data) and above $f_{0}/2$ (blue data), which overall broaden in range and shift to higher frequencies with increasing power. To further elucidate the threshold character of the 3-magnon splitting, we plot a more detailed power dependence of the magnon intensities in Fig. \[fig2\]e for $f_\mathrm{0}=6.1$GHz. While the mode at 6.1GHz can be observed over a large power range it is evident that the split modes $f_{1}$ and $f_{2}$ appear only above a certain threshold power. Furthermore, we observe a pronounced frequency shift of these two split modes with increasing microwave power. For a quantitative comparison, we integrate the BLS intensity in narrow frequency windows around the directly and indirectly excited modes, respectively (Fig. \[fig2\]f). The double-logarithmic scale reveals the linear growth of the direct excitation at $6.1$GHz starting at 0.1mW. However, the intensities of the satellite peaks around $f_\mathrm{1}=2.65$GHz and $f_\mathrm{2}=3.48$GHz abruptly increase above 10mW which demonstrates the threshold character of the splitting process.
In order to reveal the spatial structure of the eigenmodes that are generated via 3-magnon scattering, we simultaneously mapped the profiles of the directly excited mode and the split modes (Fig. \[fig3\]b,d-f). Additionally, we compare the experimental results for the mode with highest intensity at 6.1GHz with micromagnetic simulations (Fig. \[fig3\]c). The first thing to realise is that all of the split modes show a clear azimuthal character and confirm the analytical calculations and the selection rules imposed by the rotational symmetry: pure radial modes with ($n,0$) split in modes with $m_{1}=-m_{2}$ and $n_{1}\neq n_{2}$. As far as possible, we label the modes according to their radial and azimuthal mode numbers ($n,m$). We resolve azimuthal mode numbers up to 14, to our knowledge the first time to observe vortex modes with such high $m$. For higher $n$, an unambiguous identification of the modes was not possible due to limited spatial resolution. However, the radial mode number can still be retrieved by comparing the measured frequencies to the analytic calculations in Fig. \[fig1\]b. We counted the azimuthal mode numbers and plotted the measured frequencies as black dots in the calculated spectrum. From this comparison we then determined the radial mode numbers (red labels in Fig. \[fig3\]). The micromagnetic simulation for excitation at 6.1GHz reveals the splitting into magnons with the same mode numbers as in the experiment, however, with slightly different frequencies of the split modes. Reasons for this frequency shift may be attributed to variations in the strength and symmetry of the exciting magnetic field or the material parameters.
Note that we only measure stationary mode profiles which implies that, essentially, all split modes are a superposition of modes counter propagating in azimuthal direction. Therefore, we conclude that the two splitting processes $(n_{0}, 0) \rightarrow (n_{1}, m) + (n_{2}, -m) $ and $(n_{0}, 0) \rightarrow (n_{1}, -m) + (n_{2}, m)$ occur with equal probability.
It is remarkable, that the higher $m$ for a given $n$, the stronger the mode is localised at the outer circumference of the disc, resembling intensity distributions of optical whispering gallery modes[@Yang]. The most beautiful example in our dataset is the intensity distributions of the split mode $(0,12)$ at the excitation frequency 6.1GHz shown in Fig. \[fig3\]b, which exhibits a distinct hole in its center.
In summary, we shed light on the nonlinear conversion of magnons in a confined system with rotational symmetry by analysing their spectral and spatial characteristics. We showed how this mechanism can be utilised to generate magnons with unprecedented high azimuthal wave vectors and localisation at the discs perimeter, which resembles the character of whispering gallery modes. The underlying 3-magnon scattering processes show a high degree of tunability regarding the frequency and spatial distribution of the split modes. We believe that this advanced control of the generation of whispering gallery magnons is a missing key towards the realisation of an efficient hybridisation of magnons and other quantum particles as found in circular optical cavities and mechanical quantum resonators.
Methods {#methods .unnumbered}
=======
[**Sample Preparation.**]{} Using electron-beam lithography and conventional lift-off techniques, we patterned magnetic discs with diameters of 5.1$\mu$m from a Ti(2)/Ni$_{81}$Fe$_{19}$(50)/Ti(5) film, that was deposited via electron-beam evaporation on a SiO$_{2}$ substrate. All thicknesses are given in nanometer. In a second step, we designed $\Omega$-shaped antennas with an inner and outer diameter of 9 and 11$\mu$m, respectively, which separately enclose each individual disc. The antennas are patterned from a Ti(2)/Au(200) layer, also employing e-beam lithography, e-beam evaporation and lift-off techniques, and can be connected to a microwave source via picoprobes.
[**BLS microscopy.**]{} All measurements were performed at room temperature. The spin-wave intensity is locally recorded by means of BLS microscopy. This method is based on the inelastic scattering of light and spin waves. Light from a continuous wave, single-frequency 532-nm solid-state laser is focused on the sample surface using a high numerical aperture microscope lens giving a spatial resolution of 250nm. The laser power on the sample surface is typically about 1mW. The frequency shift of the inelastically scattered light is analysed using a six-pass Fabry-Perot interferometer TFP-2 (JRS Scientific Instruments). To record two-dimensional maps of the spin-wave intensity distribution, the sample is moved via a high-precision translation stage (10-nm step size, Newport). The sample position is continuously monitored with a CCD camera using the same microscope lens. A home-built active stabilisation algorithm based on picture recognition allows for controlling the sample position with respect to the laser focus with a precision better than 20nm.
**Theory.** Here, we briefly describe the derivation of the selection rules for 3-magnon splitting in a vortex-state disc; a detailed and general consideration of this problem will be published elsewhere. We use the standard Hamiltonian formalism for nonlinear magnon interaction (see detailed description, e.g. in Refs. ). Neglecting effects of the vortex core (since it results in effects about $R_\mathrm{disc}/R_\mathrm{core} \sim 250$ times smaller than the out-of-core area), we introduce the canonical variable $a(\vec\rho, t)$ as $-(iM_\rho + M_z) \approx \sqrt{2} M_s a \left(1 - aa^*/4 \right)$, $M_\phi = M_s(1 - aa^*)$, where the polar coordinate system $\rho, \phi, z$ is used [@Galkin_2006], $\vec\rho$ is the in-plane radius vector and $M_s$ is the saturation magnetisation. The canonical variable is expanded into a series of linear magnon modes $a(\vec\rho, t) = \sum_{n, m} a_{nm}(t) g_{nm}(\rho) e^{im\phi}$, where $n = 0,1,2,...$ and $m = 0, \pm1, \pm2, ...$ are the radial and azimuthal mode numbers, $a_{nm}$ is the mode amplitude and the functions $g_{nm}(\rho)$ describe the modes profile in radial direction.
The Hamiltonian $\H = \gamma W/(M_sV)$ ($W$ is the total magnetic energy and $V$ is the disc volume) in the studied case consists of exchange and dipolar contributions, $$\label{e:H}
\begin{split}
\H_\mathrm{ex} & = - \frac{\gamma \mu_0 \lambda_{ex}^2}{2M_s \pi R^2} \int \sum\limits_{\alpha = x,y,z} (\nabla M_\alpha)^2 d\vec\rho \,, \\
\H_\mathrm{dip} & = \frac{\gamma \mu_0}{2M_s \pi R^2} \int d\vec\rho d\vec\rho' \vec M(\vec\rho) \cdot \mat G(\vec\rho,\vec\rho') \cdot \vec M(\vec\rho') \,,
\end{split}$$ where $\lambda_{ex}$ is the exchange length, $R$ is the disc radius and $\mat G$ is the magnetostatic Green’s function. Substituting definition of the canonical variable and its eigenmode expansion into Eq. and making straightforward algebra the 3-wave part of the transformed Hamiltonian is derived in the form $\H^{(3)} = \sum_{0,1,2} (V_{12,0} a_1a_2a_0^* + \text{c.c.})\Delta(m_1+m_2-m_0)$, where we use short notations $a_1 = a_{n_1 m_1}$. The term $V_{12,0}a_1a_2a_0^*$ describes the splitting of mode $a_0$ into $a_1$ and $a_2$. The delta function represents the selection rule for the azimuthal mode numbers $m_1+m_2 = m_0$, which is a consequence of the momentum conservation law. Calculations show, that in our case the exchange interaction does not contribute to the 3-magnon coefficient $V_{12,0}$, since $V_{12,0}^{(ex)} \sim m_0 = 0$. Using the expression for Green’s function in polar coordinates [@Guslienko_2000], one finds the dipolar contribution in the case $m_0=0$, $m_1=-m_2=m$ equal to $$\label{e:Vmm0}
\begin{gathered}
V_{12,0} = \frac{im\omega_M}{2\sqrt{2}R^2} \iiint \left(J_{m+1}(k\rho) - J_{m-1}(k\rho)\right) J_m(k\rho')\times \\
f(kL) g_0(\rho') \left[g_1(\rho) g_2(\rho') - g_2(\rho) g_1(\rho') \right] \rho d\rho d\rho'dk \,,
\end{gathered}$$ where $f(x) = 1-(1-e^{-|x|})/|x|$, $L$ is the disc thickness and $J_m$ is the Bessel function. The last term in brackets in Eq. (\[e:Vmm0\]) is nonzero only if $g_{n_1,m}(\rho) \neq g_{n_2,-m}(\rho)$. Radial profiles of modes having the same radial number $n$ and opposite azimuthal numbers are the same $g_{n,m}(\rho) = g_{n,-m}(\rho)$, (except for the case $m = \pm 1$, when the hybridisation with the gyrotropic mode appears [@Guslienko2008], but this difference is small being proportional to $R_\mathrm{core}/R$). Thus, 3-magnon interaction efficiency is nonzero only if $n_1 \neq n_2$. Consequently, the selection rules for splitting of $(n_0,0)$ mode into $(n_1, m_1)$ and $(n_2, m_2)$ are: $m_1=-m_2$ (consequence of momentum conservation law) and $n_1 \neq n_2$ (result of the symmetry of the dipolar interaction). The spectra of spin-wave excitations and the corresponding profiles (Fig. \[fig1\]b,c) were calculated as numerical solution of linearised Landau-Lifshitz equation using the projection method [@Buess2005].
[**Micromagnetics.**]{} Micromagnetic simulations are carried out using a custom version of MuMax3 [@MuMax] which uses a finite difference approach to solve the Landau-Lifshitz-Gilbert equation of motion on a rectangular grid. As a midway between accuracy and computational performance, the discs of 5.1$\mu$m diameter and 50nm thickness are modelled with a cell size of 10nm x 10nm x 5nm using the material parameters listed below. To determine the spatial mode amplitudes of the satellites, a harmonic out-of-plane magnetic field is applied at $f_0$ for a total duration of 100ns. The time-dependent magnetisation is sampled at a rate of 10ps and subsequently Fourier-transformed in the time domain at each cell. From the resulting frequency spectrum, the peak positions of the satellites $f_1$ and $f_2$ are determined and the respective mode amplitudes are calculated via backwards Fourier transform at these frequencies.
Material parameters used for analytical calculations as well as for micromagnetic simulations: thickness 50nm, diameter 5.1$\mu$m, saturation magnetisation $M_{s}=810$kA/m, $\gamma= 1.86\times 10^{11}$rad/(s T), exchange constant $A = 1.3\times 10^{-11}$J/m.
References {#references .unnumbered}
==========
[99]{}
Debye, P. Der Lichtdruck auf Kugeln von beliebigem Material. [*Ann. Phys.*]{} [**30**]{}, 57Ð136 (1909).
Cai, M., Painter, O., Vahala, K. J., & Sercel, P. C. Fiber-coupled microsphere laser. [*Opt. Lett.*]{} [**25**]{}, 1430-1432 (2000).
Oraevsky, A. N. Whispering-gallery waves. [*Quant. Electron.*]{} [**32**]{}, 377-400 (2002).
Vahala, K. J. Optical microcavities. [*Nature*]{} [**424**]{}, 839-846 (2003).
Mie, G. Beiträge zur Optik trüber Medien, speziell kolloidaler Metallösungen. [*Ann. Phys.*]{} [**25**]{}, 377-445 (1908).
Tabuchi, Y. et al. Hybridizing Ferromagnetic Magnons and Microwave Photons in the Quantum Limit. [*Phys. Rev. Lett.*]{} [**113**]{}, 083603 (2014).
Osada, A. Cavity Optomagnonics with Spin-Orbit Coupled Photons. [*Phys. Rev. Lett.*]{} [**116**]{}, 223601 (2016).
Haigh, J. A., Nunnenkamp, A., Ramsay, A. J., & Ferguson, A. J., [*Phys. Rev. Lett.*]{} [**117**]{}, 133602 (2016).
Kusminskiy, S. V., Tang, H. X., & Marquardt, F., Coupled spin-light dynamics in cavity optomagnonics. [*Phys. Rev. Appl.*]{} [**94**]{}, 033821 (2016).
Zhang X., Y. et al. Magnon dark modes and gradient memory. [*Nat. Comm.*]{} [**6**]{}, 8914 (2015).
Suhl, H. The theory of ferromagnetic resonance at high signal powers. [*J. Phys. Chem. Solids.*]{} [**1**]{}, 209 (1957).
Ordonez-Romero, C. [*et al.*]{} Three-magnon splitting and confluence processes for spin-wave excitations in yttrium iron garnet films: Wave vector selective Brillouin light scattering measurements and analysis. [*Phys. Rev. B*]{} [**79**]{}, 144428 (2009).
Schultheiss, H. [*et al.*]{} Direct Current Control of Three Magnon Scattering Processes in Spin-Valve Nanocontacts. [*Phys. Rev. Lett.*]{} [**103**]{}, 157202 (2009).
Camley, R.E. Three-magnon processes in magnetic nanoelements: Quantization and localized mode effects. [*Phys. Rev. B*]{} [**89**]{}, 214402 (2014).
Shinjo, T., Okuno, T., Hassdorf, R., Shigeto, K., & Ono, T., Magnetic vortex core observation in circular dots of permalloy. [*Science*]{} [**289**]{}, 930 (2000).
Novosad, V. [*et al.*]{} Spin excitations of magnetic vortices in ferromagnetic nanodots. [*Phys. Rev. B*]{} [**66**]{}, 052407 (2002).
Guslienko, K.Y. & Novosad, V. Vortex state stability in soft magnetic cylindrical nanodots. [*J. Appl. Phys.*]{} [**96**]{}, 4451 (2004).
Novosad, V., [*et al.*]{} Magnetic vortex resonance in patterned ferromagnetic dots. [*Phys. Rev. B*]{} [**72**]{}, 024455 (2005).
Guslienko, K. Y., Slavin, A. N., Tiberkevich, V., & Kim, S.-K. Dynamic origin of azimuthal modes splitting in vortex-state magnetic dots. [*Phys. Rev. Lett.*]{} [**101**]{}, 247203 (2008).
Kalinikos, B. A. & Slavin, A. N. The dipole-exchange spin wave spectrum for anisotropic ferromagnetic films with mixed exchange boundary conditions. [*J. Phys. C*]{} [**19**]{}, 7013 (1986).
Buess, M. [*et al.*]{} Fourier transform imaging of spin vortex eigenmodes. [*Phys. Rev. Lett.*]{} [**93**]{}, 077207 (2004).
Zivieri, R. & Nizzoli, F. Theory of spin modes in vortex-state ferromagnetic cylindrical dots. [*Phys. Rev. B*]{} [**71**]{} (2005).
Awad, A. [*et al.*]{} Spin excitation frequencies in magnetostatically coupled arrays of vortex state circular Permalloy dots. [*Appl. Phys. Lett.*]{} [**97**]{}, 132501 (2010).
Taurel, B. [*et al*]{}. Complete mapping of the spin-wave spectrum in a vortex-state nanodisk. [*Phys. Rev. B*]{} [**93**]{}, 184427 (2016).
Sebastian, T., Schultheiss, K., Obry, B., Hillebrands, B. & Schultheiss, H. Micro-focused Brillouin light scattering: imaging spin waves at the nanoscale. [*Front. Phys*]{}. [**3**]{}, 35 (2015).
Vogt, K. [*et al.*]{} Optical detection of vortex spin-wave eigenmodes in microstructured ferromagnetic disks. [*Phys. Rev. B*]{} [**84**]{}, 174401 (2011).
Schultheiss, H., Vogt, K. Hillebrands, B. Direct observation of nonlinear four-magnon scattering in spin-wave microconduits. [*Phys. Rev. B*]{} [**86**]{}, 054414 (2012).
Suhl, H. Foldover effects caused by spin wave interactions in ferromagnetic resonance. [*J. Appl. Phys.*]{}, [**31**]{}, 935 (1960).
Janantha, P. A. P., Kalinikos, B., and Wu, M. Foldover of nonlinear eigenmodes in magnetic thin film based feedback rings. [*Phys. Rev. B*]{} [**95**]{}, 064422 (2017).
Yang, J. J., Huang, M., Yu, J., & Lan, Y. Z. Surface whispering-gallery mode. [*Europhys. Lett.*]{} [**96**]{}, 57003 (2011).
V. S. L’vov, *Wave Turbulence under Parametric Excitation* (Springer-Verlag, New York, 1994).
P. Krivosik and C. E. Patton, Hamiltonian formulation of nonlinear spin-wave dynamics: Theory and applications. [*Phys. Rev. B*]{} [**82**]{}, 184428 (2010).
K. Livesey, Nonlinear Behavior in Metallic Thin Films and Nanostructures. in *Handbook of Surface Science, vol. 5*, ed. by R. E. Camley, Z. Celinski, R. L. Stamps, (North-Holland, 2015).
A. Yu. Galkin, B. A. Ivanov, and C. E. Zaspel, Collective modes for an array of magnetic dots in the vortex state. [*Phys. Rev. B*]{} [**74**]{}, 14441 (2006).
K. Y. Guslienko and A. N. Slavin, Spin-waves in cylindrical magnetic dot arrays with in-plane magnetization. [*J. Appl. Phys.*]{} [**87**]{}, 6337 (2000).
Buess, M. [*et al.*]{} Excitations with negative dispersion in a spin vortex. [*Phys. Rev. B*]{} [**71**]{}, 104415 (2005).
Vansteenkiste, A. [*et al.*]{} The design and verification of MuMax3. [*AIP Adv.*]{} [**4**]{}, 107133 (2014).
acknowledgments {#acknowledgments .unnumbered}
===============
The authors acknowledge fruitful discussions with S.V. Kusminskiy. Financial support by the Deutsche Forschungsgemeinschaft is gratefully acknowledged within program SCHU2922/1-1. K.S. acknowledges funding within the Helmholtz Postdoc Programme. Samples were fabricated at the Nanofabrication Facilities (NanoFaRo) at the Institute of Ion Beam Physics and Materials Research at HZDR. We thank B. Scheumann for film deposition and L. Bischoff for the thickness measurement.
Author contributions {#author-contributions .unnumbered}
====================
K.S. and H.S. conceived and designed the experiments. F.W. and K.W. fabricated the samples. R.V., V.T., and A.N.S. did analytical calculations. L.K. and A.K. performed micromagnetic simulations. A.A.A. performed the COMSOL simulations. K.S., F.W., K.W., T.H., L.K., and T.H. performed BLS experiments and analysed the data. K.S. and H.S. wrote the manuscript. All authors discussed the results and commented on the manuscript.
Additional information {#additional-information .unnumbered}
======================
The authors declare no competing financial interests. Reprints and permission information is available online at http://www.nature.com/reprints. Correspondence and requests for materials should be addressed to K.S.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present a new, fully analytical point scattering model which can be applied to arbitrary anisotropic magneto-electric dipole scatterers, including split ring resonators (SRRs), chiral and anisotropic plasmonic scatterers. We have taken proper account of reciprocity and radiation damping for electric and magnetic scatterers with any general polarizability tensor. Specifically, we show how reciprocity and energy balance puts constraints on the electrodynamic responses arbitrary scatterers can have to light. Our theory sheds new light on the magnitude of cross sections for scattering and extinction, and for instance on the emergence of structural chirality in the optical response of geometrically non-chiral scatterers like SRRs. We apply the model to SRRs and discuss how to extract individual components of the polarizability matrix and extinction cross sections. Finally, we show that our model describes well the extinction of stereo-dimers of split rings, while providing new insights in the underlying coupling mechanisms.'
author:
- Ivana Sersic
- Christelle Tuambilangana
- Tobias Kampfrath
- 'A. Femius Koenderink'
title: 'Magneto-electric point scattering theory for metamaterial scatterers'
---
Introduction
============
Research in the field of metamaterials is driven by the possibility to control the properties of light on the nanoscale by using coupled resonant nanoscatterers to create optical materials with very unusual effective medium parameters. Engineering arbitrary values for the effective permittivity $\epsilon$ and permeability $\mu$ would allow new forms of light control based on achieving negative index materials [@Veselago68; @Pendry00; @Pendry01], or transformation optics media [@Pendry06] that arbitrarily reroute light through space. In order to reach such control over $\epsilon$ and $\mu$, many metamaterial building nanoblocks have previously been identified as having an electric and magnetic response to incident light, including split ring resonators (SRRs) [@Smith00; @Enkrich05; @Rockstuhl06; @Klein06; @Sersic09; @Lahiri10], rod-pairs [@Shalaev05], cut-wire pairs [@Dolling05], fishnet structures [@Dolling06; @Dolling07; @Valentine08] and coaxial waveguides [@Waele10]. In many instances, the nanoscatterers are not only interesting as building blocks in subwavelength lattices of designed $\epsilon$ and $\mu$. The building blocks are in fact very strong scatterers with large cross sections [@Husnik08; @Rockstuhl06b; @Corrigan08; @Pors10], comparable to the large cross sections of plasmonic structures. Therefore, metamaterial building blocks are excellently suited to construct magnetic antennas, array waveguides and gratings in which electric and magnetic dipoles couple and form cooperative excitations, in analogy to the functionality imparted by plasmon hybridization [@NordlanderScience]. Experiments outside the domain of effective media have appeared only recently. These experiments include experiments by Husnik *et al*. [@Husnik08], and Banzer *et al*. [@Banzer10] that quantify the extinction cross section of single split rings under differently polarized illumination, as well as a suite of experiments on coupled systems. These experiments include extinction measurements on split ring dimers [@Feth10] that point at resonance hybridization, as well as reports of magnetization waves [@decker09], structural and geometrical chirality in arrays, as evident in e.g. massive circular dichroism [@Gansel09; @Plum09; @Plum09b; @Zhang09; @Wang09; @plum07; @decker07; @decker10], and chiral effects in split ring stereo-dimers studied by Liu *et al*. [@Giessen09].
In order to understand the light-metamatter interaction in systems of strongly coupled magneto-electric scatterers, it is important to understand how individual metamaterial building blocks are excited and how they scatter. So far, explanations of the observed phenomena have mainly rested on two pillars. On the one hand, data are compared to brute force finite-difference time-domain (FDTD) simulations of Maxwell’s equations, usually showing good correspondence [@Smith00; @Enkrich05; @Rockstuhl06b; @Klein06; @Husnik08; @Gansel09; @Plum09]. While the FDTD method is a rigorous method, such numerical experiments do not in themselves provide insight into how split rings scatter or hybridize in coupled systems. There is general consensus that to lowest order, metamaterial interactions in lattices of scatterers like SRRs must be described by magneto-electric point-dipole interactions. Hence, simple LC circuit models with dipolar coupling terms are the second main interpretative tool to predict, e.g., frequency shifts due to electric and magnetic dipole-dipole interactions in lattices and oligomers [@Sersic09; @Giessen09; @Feth10; @guo07]. To rationalize this LC circuit intuition, several authors have analyzed current distributions obtained by FDTD simulations in order to retrieve the microscopic parameters (*i.e.*, the polarizability) underlying such a dipolar interaction model, and in order to estimate multipolar corrections [@Giessen10; @Rockstuhl06b; @Rockstuhl07; @Zhou07; @Corrigan08; @Pors10; @RockstuhlMulti].
While there is general consensus that to lowest order, metamaterial interactions must essentially be magneto-electric point dipole interactions, we note that the dipolar circuit models in use so far have some significant shortcomings. Strictly speaking, the electric circuit theories lack the velocity of light $c$ as a parameter. Hence, they contain no retardation or interference, they violate the optical theorem, do not predict quantitative cross sections, and fail to predict the effects of super- and subradiant damping on resonance linewidths. A fair comparison of intuitive point-dipole ideas with actual data is therefore impossible, unless a fully electrodynamic theory for magneto-electric point dipoles is derived. Such a theory would generalize the electric point scattering theory that is well known as very effective means to describe random media, extraordinary transmission and plasmon particle arrays [@Lagendijk96; @Pedro98; @Abajo07]. In this paper we derive exactly such a theory for general magneto-electric scatterers. We show how reciprocity and energy conservation restrict the full magneto-electric response via Onsager constraints [@Landau; @Lindell_biisotropicbook], and a new magneto-electric optical theorem for the full polarizability tensor. This tensor not only includes an electric (magnetic) response to electric (magnetic) driving, but also off-diagonal coupling in which a magnetic (electric) response results from electric (magnetic) driving. While our theory sheds no light on the microscopic origin of the polarizability [@MerlinPNAS], we show how electrodynamic polarizability tensors can be directly constructed from LC circuit models. Furthermore we predict how extinction measurements and measurements of radiation patterns (i.e., differential scattering cross section) can be used to quantify the polarizability tensor.
The paper is structured in the following way: Firstly, in Section \[section:generaltheory\] we derive the general theory, taking into full account reciprocity, the optical theorem and radiation damping. In Section \[section:singleSRR\] we apply this theory to set up the polarizability of the archetypical metamaterial building block, a single SRR. In Section \[section:singleSRRexp\] we show which set of experiments can be used to retrieve the tensor polarizability $\boldsymbol{\alpha}$. We find that magneto-electric coupling directly implies circular dichroism in the extinction of single split rings, evidencing the utility of our theory to describe structural chirality [@Gansel09; @Plum09; @Plum09b; @Zhang09; @Wang09; @plum07; @decker07; @decker10]. Thirdly, we show in Section \[section:coupledSRR\] that the theory can be simply applied to obtain quantitative scattering spectra of coupled systems. By way of example we examine the case of two coupled resonators in the stereodimer configuration reported by Liu et al. [@Giessen09].
Magneto-electric point scatterer\[section:generaltheory\]
=========================================================
Polarizability
--------------
A paradigm in scattering theory is the point dipole scatterer [@Lagendijk96; @Pedro98; @Weber04; @Koenderink06; @Abajo07] to model scattering by very small, but strongly scattering particles. Generally, incident fields $\bm{E}$ and $\bm{H}$ induce a (complex) current distribution in an arbitrary scatterer. It is the express point of this paper to assess what the scattering properties are of subwavelength scatterers with strong electric and magnetic dipole moments, as this represents the physics expected of metamaterial building blocks [@Garcia-Garcia05; @Sersic09; @Giessen09; @Feth10; @guo07]. Therefore we retain only electric and magnetic dipole terms, neglecting higher order multipoles. In such a theory, each scatterer is approximated as an electric dipole with an electric dipole moment $\bm{p}=\alpha_{EE}\bm{E}$ that is proportional to the driving electric field $\bm{E}$. The proportionality constant is the polarizability $\alpha_{EE}$. In this paper, we derive a generalized point scattering theory for metamaterials that includes a magnetic dipole moment $\bm{m}$ on an equal footing with the electric dipole moment $\bm{p}$. In the most general case, the electric dipole moment $\bm{p}$ and magnetic dipole moments $\bm{m}$ are induced by both the external electric and magnetic fields $\bm{E}$ and $\bm{H}$ according to $$\left(
\begin{array} {c}
\bm{p}\\
\bm{m}
\end{array}\right)
=\boldsymbol{\alpha} \left(
\begin{array} {c}
\bm{E}_{\mathrm{in}}\\
\bm{H}_{\mathrm{in}}
\end{array}\right).
\label{Eq:polarizdef}$$ Throughout this paper we suppress harmonic time dependence $e^{-i\omega t}$. We use a rationalized unit system that significantly simplifies all equations and is fully explained in Appendix \[appendix\]. In Eq. (\[Eq:polarizdef\]), $\boldsymbol{\alpha}$ is a 6$\times$6 polarizability tensor, which consists of four 3$\times$3 blocks, each of which describes part of the dipole response to the electric or magnetic component of the incident light $$\boldsymbol{\alpha}=
\left(
\begin{array}{c c}
\boldsymbol{\alpha}_{EE} & \boldsymbol{\alpha}_{EH}\\
\boldsymbol{\alpha}_{HE} &
\boldsymbol{\alpha}_{HH}
\end{array} \right).
\label{Eq:splitpolariz}$$ Here, $\boldsymbol{\alpha}_{EE}$ quantifies the electric dipole induced by an applied electric field. The tensorial nature of $\boldsymbol{\alpha}_{EE}$ is well appreciated in scattering theory for anisotropic particles, such as plasmonic ellipsoids [@BohrenHuffman]. By analogy with the electric response to electric driving quantified by $
\boldsymbol{\alpha}_{EE}$, the tensor $
\boldsymbol{\alpha}_{HH}$ quantifies the magnetic dipole induced by a driving magnetic field. Finally, the off-diagonal blocks represent magneto-electric coupling. The lower diagonal $ \boldsymbol{\alpha}_{HE}$ quantifies the magnetic dipole induced by an incident electric field, and $
\boldsymbol{\alpha}_{EH}$ quantifies the electric dipole induced by an incident magnetic field. Such magneto-electric coupling is well known to occur in the constitutive tensors of metamaterials [@Rockstuhl06]. Indeed, the first metamaterials consisted of split ring resonators, in which there is a magnetic response without any driving magnetic field in normal incidence experiments [@APLSoukoulis]. However, the relative strength of magneto-electric coupling in the polarizability, i.e., $
\boldsymbol{\alpha}_{EH}$, and $\boldsymbol{\alpha}_{HE}$ have not been experimentally quantified.
Electrodynamic Onsager relation
-------------------------------
There are several constraints on $\boldsymbol{\alpha}$. In addition to any symmetry of the scatterer itself that may impose zeros in the polarizability tensor, these constraints are due to reciprocity and to energy conservation. We start by examining the constraints imposed by reciprocity. It is well known from the field of bi-anisotropic materials [@Lindell_biisotropicbook] that reciprocity imposes so-called Onsager constraints on the most general constitutive tensors relating ($\bm{D},\bm{B}$) to ($\bm{E},\bm{H}$). Already García-García *et al.*[@Garcia-Garcia05] proposed that such Onsager constraints carry over directly to electrostatic polarizabilities. Here we rigorously derive Onsager relations for electrodynamic magneto-electric point scatterers. By, definition, the electric and magnetic fields due to the induced $\bm{p}$ and $\bm{m}$ are equal to $$\left(
\begin{array}{c}
\bm{E}_{\mathrm{out}}\\
\bm{H}_{\mathrm{out}}
\end{array}\right)
=\boldsymbol{G}^0(\bm{r},\bm{r}')\left(
\begin{array}{c}
\bm{p}\\
\bm{m}
\end{array}\right),
\label{Eq:dipoleG}$$ with a dyadic Green tensor $\boldsymbol{G}^0$ that describes the field at position $\bm{r}=(x,y,z)$ due to a dipole at $\bm{r'}=(x',y',z')$. The 6$\times$6 Green dyadic of free space can be divided in four 3$\times$3 blocks $$\boldsymbol{G}^0(\bm{r},\bm{r}') =\left(
\begin{array} {c c}
\boldsymbol{G}^0_{EE}(\bm{r},\bm{r}') & \boldsymbol{G}^0_{EH}(\bm{r},\bm{r}')\\
\boldsymbol{G}^0_{HE}(\bm{r},\bm{r}') & \boldsymbol{G}^0_{HH}(\bm{r},\bm{r}')
\end{array} \right)
\label{Eq:freeG}$$ The 3$\times$3 diagonals correspond to the familiar known electric field Green dyadic [@Pedro98; @Abajo07] and magnetic field Green dyadic of free space, which in our unit system (see Appendix) both equal $$\boldsymbol{G}^0_{EE}(\bm{r},\bm{r}')=\boldsymbol{G}^0_{HH}(\bm{r},\bm{r}')=
(\mathbb{I}k^2+\nabla\nabla)
\frac{e^{ik|\bm{r}-\bm{r'}|}}{|{\bm{r}-\bm{r'}}|}.
\label{Eq:freeGEE}$$ The off diagonal blocks correspond to the mixed dyadics that specify the electric field at $\bm{r}$ due to a magnetic dipole at $\bm{r'}$, respectively the magnetic field at $\bm{r}$ due to an electric dipole at $\bm{r'}$. Explicitly: $$\begin{aligned}
\boldsymbol{G}^0_{EH}(\bm{r},\bm{r}') & =&
-\boldsymbol{G}^0_{HE}(\bm{r},\bm{r}') \nonumber \\ & =& ik \left(
\begin{array} {c c c}
0 & \partial_z & -\partial_y\\
-\partial_z & 0 & \partial_x\\
\partial_y & -\partial_x & 0
\end{array} \right)\frac{e^{ik|\bm{r}-\bm{r'}|}}{|{\bm{r}-\bm{r'}}|}.
\nonumber\\ \label{Eq:freeGEH}\end{aligned}$$ In this work we focus solely on scatterers made from reciprocal constituents, as is commonly true for the metallic scatterers that constitute metamaterials. Since the materials that compose our scatterers (typically gold and silver) satisfy reciprocity microscopically, the polarizability tensor must also lead to a scattering theory that satisfies reciprocity. To derive reciprocity constraints on $\boldsymbol{\alpha}$, it is sufficient to examine the Green function in the presence of just one point scatterer at the origin. This Green function that quantifies the field at $\bm{r}_2$ due to a source at $\bm{r}_1$ in presence of a single scatterer at $\bm{r}_s$ can be written as [@Lagendijk96; @Pedro98; @tfootnote] $$\boldsymbol{G}(\bm{r}_1,\bm{r}_2)=\boldsymbol{G}^0(\bm{r}_1,\bm{r}_2)+\boldsymbol{G}^0(\bm{r}_2,\bm{r}_s)\boldsymbol{\alpha}\boldsymbol{G}^0(\bm{r}_s,\bm{r}_1),
\label{Eq:reciprocity}$$ Reciprocity requires for any Green function $\boldsymbol{G}$ (similarly split in four blocks) that $$\begin{aligned}
\left(
\begin{array} {c c}
\boldsymbol{G}_{EE}(\bm{r}_2,\bm{r}_1) & \boldsymbol{G}_{EH}(\bm{r}_2,\bm{r}_1)\\
\boldsymbol{G}_{HE}(\bm{r}_2,\bm{r}_1) & \boldsymbol{G}_{HH}(\bm{r}_2,\bm{r}_1)
\end{array}\right) = \qquad\qquad \nonumber\\
\qquad\qquad \left(
\begin{array} {c c}
\boldsymbol{G}_{EE}(\bm{r}_1,\bm{r}_2) & -\boldsymbol{G}_{EH}(\bm{r}_1,\bm{r}_2)\\
-\boldsymbol{G}_{HE}(\bm{r}_1,\bm{r}_2) & \boldsymbol{G}_{HH}(\bm{r}_1,\bm{r}_2)
\end{array}\right)^T \nonumber \\
\label{Eq:reciprocTRANSPOSE}\end{aligned}$$ which is equivalent to noting that swapping source and detector leaves the detected field unchanged, up to a change in sign. Specifically, Lorentz reciprocity requires a transpose for the diagonal 3$\times$3 blocks, meaning that swapping a detector and source of like character leaves the detected field unchanged. An extra minus occurs for the off-diagonal terms, *i.e.*, when swapping a magnetic (electric) detector with an electric (magnetic) source. It is easy to verify that Eq. (\[Eq:reciprocTRANSPOSE\]) is indeed satisfied by the free space Green function $\boldsymbol{G}^0$.
Using this fact, we evaluate Eq. (\[Eq:reciprocTRANSPOSE\]) for the Green function in Eq. (\[Eq:reciprocity\]) to find if reciprocity constrains $\boldsymbol{\alpha}$. Since reciprocity is clearly satisfied for the first term in Eq. (\[Eq:reciprocity\]), we now focus on the second term $$\boldsymbol{G}^0(\bm{r}_2,\bm{r}_s)\boldsymbol{\alpha}\boldsymbol{G}^0(\bm{r}_s,\bm{r}_1)=\boldsymbol{G}^0(\bm{r}_1,\bm{r}_s)
\boldsymbol{\alpha}\boldsymbol{G}^0(\bm{r}_s,\bm{r}_2).
$$ Expanding the matrix products in Eq. (\[Eq:reciprocTRANSPOSE\]) while making use of the reciprocity of the free Green function results in the Onsager relations for the dynamic polarizability: $$\begin{aligned}
\boldsymbol{\alpha}_{EE}=\boldsymbol{\alpha}_{EE}^T,
\quad
\boldsymbol{\alpha}_{HH}=\boldsymbol{\alpha}_{HH}^T,
\quad
\mbox{and} \quad
\boldsymbol{\alpha}_{EH}=-\boldsymbol{\alpha}_{HE}^T
\qquad
\nonumber \\
\label{Eq:onsager}\end{aligned}$$ These relations are identical in form to the Onsager relations for constitutive tensors [@Lindell_biisotropicbook], but are now derived on very different grounds for the polarizability of electrodynamically consistent point scatterers. This gratifying result shows that the general point dipoles proposed in this work can be used as microscopic building blocks for an exact scattering theory that describes the formation of bi-anisotropic media from dense lattices of scatterers in the effective medium limit. Indeed, since the point scattering building blocks fulfill the Onsager constraints, they are natural building blocks to derive effective media constitutive tensors by homogenization that also satisfy the Onsager relations.
Optical theorem
---------------
It is well known in point scattering theory for electric dipoles that polarizability tensors are not solely limited by reciprocity and spatial symmetry, but also fundamentally by energy conservation. Indeed, energy conservation imposes an ’optical theorem’ that constrains the polarizability of an electric dipole scatterer to ensure that (in absence of material absorption) extinction equals scattering [@Pedro98]. We proceed to examine these constraints imposed on $\boldsymbol{\alpha}$. Let us first recapitulate the well known case of a scalar electric scatterer [@Lagendijk96; @Pedro98]. An electric scatterer will absorb and scatter part of the incoming light, that together make up the extinction of a dipole. Extinction for an electric scatterer corresponds to the work done by the incident field $\bm{E}$ in order to drive the dipole $\bm{p}$. The work per optical cycle needed to drive $\bm{p}$ equals $W= \ll\mathrm{Re}{\bm{E}}\cdot \mathrm{Re}{\frac{d\bm{p}}{dt}}\gg$, where $\ll\gg$ indicates cycle averaging. Evaluating the work per cycle, and dividing it by the incident intensity $I_{\mathrm{in}}=|\bm{E}|^2/2Z$ ($Z$ the impedance of the host medium) leads to $\sigma_{\mathrm{ext}}=W/I_\mathrm{in}=4\pi k
\mathrm{Im}\alpha_{EE}$. Scattering corresponds to far field radiation radiated by the dipole $\bm{p}$. According to Larmor, the cycle-averaged scattered power is [@jacksonbook] $P=\frac{4\pi k^4}{3 Z}|\bm{p}|^2.$ Hence one obtains the well known cross sections $$\sigma_{\mathrm{ext}}=4\pi k \mathrm{Im}\alpha_{EE}
\quad\mbox{and}\quad
\sigma_{\mathrm{scatt}}=\frac{8\pi}{3}k^4|\alpha_{EE}|^2.\label{Eq:scalarcrossect}$$ Equating extinction to scattering for nonabsorbing particles to impose energy conservation, gives rise to the optical theorem for the polarizability $$\mathrm{Im}\alpha=\frac{2}{3}k^3|\alpha_{EE}|^2
\label{Eq:scalarOpticalTheorem}$$ This equation for instance shows the well-known fact that a real (electrostatic) $\alpha_0$, such as Rayleigh’s polarizability $\alpha=3V(\epsilon -1)/(\epsilon +2)$ for a small sphere of dielectric constant $\epsilon$, never satisfies the optical theorem [@vdHulst]. An electrostatic $\alpha_0$ can be made to satisfy the optical theorem by adding radiation damping [@Pedro98; @Abajo07] to obtain the dynamic polarizability $$\frac{1}{\alpha}=\frac{1}{\alpha_{0}}-i\frac{2}{3}k^3.
\label{Eq:scalarRadiationDamping}$$ It is easy to verify that the albedo of a scatterer with polarizability given by Eq. (\[Eq:scalarRadiationDamping\]) is $$a=\frac{\sigma_\mathrm{scat}}{\sigma_\mathrm{ext}}=\frac{1}{1+\frac{2}{3}k^3\mathrm{Im}\alpha_0},$$ confirming that radiation damping indeed transforms any lossless electrostatic polarizability ($\mathrm{Im} \alpha_0=0$) into a scatterer that satisfies the optical theorem. Also material loss included in $\alpha_0$ via $\epsilon$ evidently leads to a lossy scatterer $a<1$, as expected. Many alternative derivations of Eq. (\[Eq:scalarRadiationDamping\]) have appeared, for instance by making a size parameter expansion of dipolar Mie coefficients [@wokaun].
Inspired by the case of a simple electric dipole, we now generalize the optical theorem and the concept of radiation damping to the full 6x6 tensorial polarizability of arbitrary magneto-electric scatterers. In this case, the work done per unit cycle by the incident field $\bm{E}_\mathrm{in}$ and $\bm{H}_\mathrm{in}$ to drive $\bm{p}$ and $\bm{m}$ is equal to $${W}=\quad \ll\mathrm{Re}\bm{E}_{in}\cdot\mathrm{Re}\frac{d\bm{p}}{dt}+\mathrm{Re}\bm{H}_{in}\cdot\mathrm{Re}\frac{d\bm{m}}{dt}\gg
$$ which evaluates to $${W}= \frac{2\pi}{Z} k \mbox{Im} \left[ \left(
\begin{array} {c c}
\bm{E}_\mathrm{in} & \bm{H}_\mathrm{in}
\end{array}\right)^{*}
\boldsymbol{\alpha} \left(
\begin{array} {c}
\bm{E}_\mathrm{in}\\
\bm{H}_\mathrm{in}
\end{array} \right)\right],
$$ where $(\cdot)^{*}$ indicates complex conjugate. The power per solid angle radiated by the induced dipoles in a direction specified by a unit vector $\bm{\hat{r}}$ is easily found by calculating the far-field Poynting vector from Eq. (\[Eq:dipoleG\]). The result is composed of three terms: $$\frac{dP}{d\Omega}=\frac{dP_p}{d\Omega}+\frac{dP_m}{d\Omega}+\frac{k^4}{2Z}\mathrm{Re}(\bm{p}\times
\bm{m})\cdot {\hat{\bm{r}}},\ \label{Eq:radiationpattern}$$ The first term in Eq. (\[Eq:radiationpattern\]) represents the scattered radiation of just the electric dipole $\bm{p}$, which integrates to a total scattered power given by Larmor’s formula. The second term in Eq. (\[Eq:radiationpattern\]) represents the radiation pattern of just the magnetic dipole $\bm{m}$, again given by Larmor’s formula. Note that both terms simply represent the well known $\sin^2\theta$ donut shaped radiation pattern for $\bm{p}$ and $\bm{m}$. The third term, however, can completely change the radiation pattern, as it contains the interference between the fields of $\bm{p}$ and $\bm{m}$. Hence the relative phase between the induced $\bm{p}$ and $\bm{m}$ is important for the differential scattering cross section. To obtain the total scattered power, one should integrate Eq. (\[Eq:radiationpattern\]) over all solid angle. The interference term integrates to 0, as is easily seen from the fact it is an odd function of $\hat{\bm{r}}$. Therefore, Larmor’s formula immediately generalizes, and the scattering cross section equals: $$P=\frac{4\pi}{3Z}k^4\left\|
\begin{array} {c} \bm{p}\\
\bm{m}\end{array}\right\|^2.\label{Eq:tensorLarmor}$$ Equating extinction to scattering results in a condition that must be satisfied for any incident field $(\bm{E}_\mathrm{in},\bm{H}_\mathrm{in})$ $$\begin{aligned}
\mathrm{Im}\left[\left(
\begin{array} {c c}
\bm{E}_\mathrm{in} & \bm{H}_\mathrm{in}
\end{array} \right)^{*}\boldsymbol{\alpha}\left(
\begin{array} {c}
\bm{E}_\mathrm{in}\\
\bm{H}_\mathrm{in}
\end{array} \right)\right]
= \qquad\qquad \qquad\qquad& \nonumber \\
\qquad\qquad\qquad\qquad
\frac{2}{3}k^3\left[\left(
\begin{array} {c c}
\bm{E}_\mathrm{in} & \bm{H}_\mathrm{in}
\end{array}\right)^{*}\boldsymbol{\alpha}^{*T}\boldsymbol{\alpha}\left(
\begin{array}
{c} \bm{E}_\mathrm{in}\\
\bm{H}_\mathrm{in}
\end{array} \right)\right], &
\nonumber \\ \label{Eq:trickyOpticalTheorem}\end{aligned}$$ Due to the tensorial character of $\boldsymbol{\alpha}$ it is not immediately evident how to extract a useful optical theorem that constrains just the polarizability tensor $\boldsymbol{\alpha}$ without reference to any incident field $(\bm{E}_\mathrm{in},\bm{H}_\mathrm{in})$. In order to eliminate $(\bm{E}_\mathrm{in},\bm{H}_\mathrm{in})$ we make the assumption (verified below for split rings) that $\boldsymbol{\alpha}$ can be diagonalized. We call the eigenvectors $\bm{v}_i$, and denote the eigenvalues, which we will refer to as ‘eigenpolarizabilities’, with $A_i$. Expanding the incident field at the position of the origin in the orthogonal eigenvectors $$\left(
\begin{array} {c}
\bm{E}_\mathrm{in}\\
\bm{H}_\mathrm{in}
\end{array} \right)=\sum_i c_i \bm{v}_i,$$ and with $
\boldsymbol{\alpha} \bm{v}_i=A_i
\bm{v}_i$ and $\langle \bm{v}_i|\bm{v}_j\rangle=\delta_{ij}$, Eq. (\[Eq:trickyOpticalTheorem\]) reduces to $$\frac{2}{3}k^3{\sum_{i=1}^6}|c_i|^2|A_i|^2\geq{\sum_{i=1}^6}|c_i|^2\mathrm{Im}A_i,\label{Eq:almostOpticalTheorem}$$ with strict equality for lossless scatterers. Since this equation must be satisfied for any choice of incident wave (i.e., any combination of $c_i$), we find a generalized optical theorem for 6$\times$6 polarizability tensors that can be expressed in terms of the eigenpolarizabilities as $$\frac{2}{3}k^3|A_i|^2\geq \mathrm{Im} A_i \qquad \forall i=1\ldots
6, \label{Eq:TensorpticalTheorem}$$ again with strict equality for lossless scatterers. Eq. (\[Eq:TensorpticalTheorem\]) implies that the polarizability tensor represents an energy conserving scatterer, if and only if each of its 6 eigenpolarizabilities are chosen to satisfy the simple scalar optical theorem (Eq. (\[Eq:scalarOpticalTheorem\])) derived for electric scatterers. This general optical theorem highlights the importance of two new quantities: the eigenpolarizabilities, and the corresponding eigenvectors of the point scatterer polarizability.
In Eq. (\[Eq:scalarRadiationDamping\]) we reviewed the well-known addition of radiation damping required to make electrostatic polarizabilities satisfy the optical theorem. Since metamaterial scatterers are frequently treated via electrostatic circuit models, it would be extremely fruitful to generalize this method to general 6$\times$6 electrostatic polarizability tensors. It is now evident, that we can simply apply the scalar recipe to each eigenpolarizability separately. An alternative notation for this method is: $$\boldsymbol{\alpha}^{-1}=\boldsymbol{\alpha}_0^{-1}-\frac{2}{3}k^3i\mathbb{I}
\label{Eq:TensorRadiationDamping}$$ We note that this expression, which is identical to Eq. (\[Eq:scalarRadiationDamping\]) upon replacement of $1/(\cdot)$ by matrix inversion, provides a unique relation to translate a magneto-/electrostatic polarizability tensor $\boldsymbol{\alpha}_0$ derived from RLC circuit theory, to the corresponding electrodynamic polarizability that satisfies the optical theorem. We can hence consistently assess how intuitive ideas based on a microscopic RLC circuit model for electrostatic dipoles lead to quantitative predictions for extinction, scattering, as well as resonance hybridization, diffraction and super/sub radiant damping in coupled systems, such as periodic systems, or arbitrary finite clusters.
Polarizability of split ring resonators\[section:singleSRR\]
============================================================
{width="\textwidth"}
Symmetry
--------
As an example of our general theory we consider the specific example of split ring resonators. The electrostatic polarizability of split ring resonators was discussed for instance by Garcìa-Garcìa et al [@Garcia-Garcia05]. We consider the LC resonance of an infinitely thin split ring in the $xy$ plane, with split oriented along the $x$ axis, as shown in Fig. \[Fig1:radpats\](a). Incident electric field polarized along the $x$ direction gives rise to an electric dipole $\bm{p}=(\alpha_{EE}^{xx}E_x,0,0)$ oriented along the split of the SRR. As in an LC circuit, the charge separation generated over the capacitive split relaxes as a circulating current, hence giving rise to a magnetic dipole $\bm{m}=(0,0,\alpha_{HE}^{zx}E_x)$ in the $z$ direction, in response to a driving E-field along $x$ [@Burresi09]. The same is valid vice versa: an applied magnetic field along $z$ induces a magnetic dipole moment $\bm{m}=(0,0,\alpha_{HH}^{zz} H_z)$ along the $z$ direction. The associated current accumulates at the gap, giving rise to an electric dipole moment $\bm{p}=(\alpha_{HE}^{xz}H_z,0,0)$ driven by $H_z$. If we assume that the LC resonance really only involves $\bm{p}_x$ and $\bm{m}_z$, we find that the polarizability tensor is filled with zeros, except for the four contributions described above. Hence $$\boldsymbol{\alpha}_{SRR} = \left(
\begin{array} {c c c c c}
\alpha_{EE}^{xx} & 0 & ... & 0 & \alpha_{EH}^{xz}\\
0 & & & & 0\\
\vdots & & \ddots & & \vdots\\
0 & & & & 0\\
\alpha_{HE}^{zx} & 0 & ... & 0 & \alpha_{HH}^{zz}
\end{array} \right). \label{Eq:SRRsymmetry}$$ The symmetry constraints that set which elements of $\boldsymbol{\alpha}_{SRR}$ are zero, are valid both for the electrodynamic and electrostatic polarizability of split rings.
Quasi-electrostatic RLC model
-----------------------------
We will now construct the electrodynamic polarizability by starting from an electrostatic polarizability derived from a single resonant RLC equation of motion. Therefore we take a common resonant frequency dependence out of the tensor elements, writing $$\boldsymbol{\alpha}_{SRR}^{\mathrm{static}}=\alpha(\omega)
\left(
\begin{array}{c c c c c}
\eta_{E} & 0 & ... & 0 & i\eta_{C}\\
0 & & & & 0\\
\vdots & & \ddots & & \vdots\\
0 & & & & 0\\
-i\eta_{C} & 0 & ... & 0 &\eta_{H}
\end{array}
\right),\label{Eq:SRRpolarizsimple}$$ where $\eta_E$, $\eta_C$ and $\eta_H$ are constant and $\alpha(\omega)$ is a Lorentzian prefactor $$\alpha(\omega)=\frac{\omega_0^2
V}{\omega_0^2-\omega^2-i\omega\gamma}. \label{Eq:prefactorSRR}$$ Here, $\omega_0$ is the SRR resonance frequency $\omega_0\approx
\frac{1}{\sqrt{LC}}$, $\gamma$ is the damping rate due to the Ohmic loss of gold and $V$ is the physical particle volume. As in the plasmonic case, this approximation is coined ‘quasi-static’, as it does contain frequency $\omega$, but does not contain the velocity of light $c$. The polarizability obtained from the quasi-static polarizability once the radiation damping term is added (section \[magnetoelectricsection\] Eq. (\[Eq:TensorRadiationDamping\])) is called ‘dynamic polarizability’. In this formulation, all the frequency dependence, and the units of $\boldsymbol{\alpha}_{SRR}$ are contained in $\alpha(\omega)$. The parameters $\eta_E$, $\eta_H$ and $\eta_C$ are dimensionless. For a lossless split ring $\eta_E$, $\eta_H$ and $\eta_C$ are all real. We assume that all losses are introduced via $\gamma$. To determine the sign of $\eta_E$, $\eta_H$ and $\eta_C$, we expect that for very slow driving the charge (current) on the capacitor directly follows the driving $E$ ($H$)-field, implying $\eta_E>0$ and $\eta_H>0$. The sign of $\eta_C$ follows similar reasoning, After charge build-up, charge associated with a $\bm{p}_x=\mathrm{Re}(\alpha(\omega) \eta_E e^{-i \omega t} E_x)$ relaxes as counter-clockwise current, giving rise to a negative $\bm{m}_z=\mathrm{Re}(\alpha(\omega) i \eta_C e^{-i \omega t} E_x)$, implying that $\mathrm{sign}\,\eta_C=\mathrm{sign}\,\eta_E$.
Limit on magneto-electric coupling\[magnetoelectricsection\]
------------------------------------------------------------
Having constructed an electrostatic polarizability in accordance with RLC circuit models proposed in earlier reports, we apply radiation damping according to Eq. (\[Eq:TensorRadiationDamping\]) to obtain a scatterer that has a correct energy balance: $$\boldsymbol{\alpha}_{{SRR}}^{-1}=(\boldsymbol{\alpha}_{SRR}^{\mathrm{static}})^{-1}-\frac{2}{3}k^3\mathbb{I}.\label{Eq:SRRradiationdamp}$$ So far we have not explicitly discussed absorption loss, except through the inclusion of the material damping constant $\gamma$ in the quasi-static polarizability. Starting from a quasi-static polarizability with quasi-static eigenpolarizabilities $A^{\mathrm{static}}_i$, the albedo for each eigenillumination $\bm{v}_i$ can be expressed as $$a_i=\frac{1}{1+\frac{2}{3}k^3\mathrm{Im}A^{\mathrm{static}}_i}.\label{Eq:albedo}$$ It follows that for any lossy scatterer the imaginary part of each eigenvalue $A^\mathrm{static}_i$ of the electrostatic polarizability tensor must be positive to ensure $0\leq a\leq1$. In the case of a tensorial $\boldsymbol{\alpha}$ with loss included as in Eq. (\[Eq:SRRpolarizsimple\]), (\[Eq:prefactorSRR\]), one needs to explicitly verify that each eigenvalue has positive imaginary part. The eigenvalues of Eq. (\[Eq:SRRpolarizsimple\]) are$A^\mathrm{static}_\pm=\alpha(\omega)\lambda_\pm$ with $\lambda_\pm=\frac{\eta_E+\eta_H\pm\sqrt{(\eta_E-\eta_H)^2+4\eta_C^2}}{2}$. Since $\mathrm{Im}(\alpha(\omega))\geq0$ and $\lambda_{\pm}$ are real, we find that both eigenvalues have positive imaginary part only if both $\lambda_+$ and $\lambda_-$ are positive. Thus, loss sets an additional constraint on the polarizability tensor, and limits the magneto-electric coupling to $$|\eta_{C}|\leq\sqrt{\eta_E\eta_H}. \label{Eq:crosscouplelimit}$$ This result implies a very important limitation on magneto-electric scatterers: it states that a magneto-electric cross coupling ($\eta_C$) can only be generated if there is a sufficiently strong directly electric, and directly magnetic response. We note that this constraint is very similar to the constraint on the magneto-electric cross coupling in constitutive tensors derived for homogeneous bi-anisotropic media in Ref. that recently attracted attention in the framework of proposals for repulsive Casimir forces [@silveirinha; @soukouliscasimir]. While our derivation was specific for split rings, we note that similar constraints hold for all magneto-electric scatterers. In the presence of material loss, the magneto-electric coupling terms are limited by the fact that all electrostatic eigenpolarizabilities must have positive imaginary part.
Predicted scattering properties of single split rings\[section:singleSRRexp\]
=============================================================================
In the remainder of the paper we discuss some insights that the proposed magneto-electric point scattering theory provides in how split rings scatter. In this section we will consider the eigenmodes and the radiation patterns of a single SRR for $\boldsymbol{\alpha}$ given by Eq. (\[Eq:SRRradiationdamp\]). Next, we predict which set of experiments will provide full information on the elements of the polarizability tensor. We will show how the extinction cross sections can be translated back to retrieve SRR polarizabilities and magneto-electric cross polarizabilities of a single SRR. Although the results we present are general, we use a specific set of parameters for all the figures presented in this paper. These parameters are chosen to fit to the properties of split rings that are resonant at $\lambda=1.5~\mu$m ($\omega_0/2\pi= 200$ THz and that consist of 200 by 200 nm gold split rings with a thickness of 30 nm and a gap width of 90 nm. Thus we take $V=200\times 200\times
30$ nm$^3$. We set the damping rate to be that of gold $\gamma
=1.25\cdot 10^{14}$ s$^{-1}$ as fitted to optical constants tabulated in in Ref. . We use $\eta_E=0.7$, $\eta_H=0.3$ and $\eta_C=0.4$. These parameters were chosen because (A) they reproduce quantitatively the extinction cross section under normal incidence along the $z$-axis measured by Husnik *et al.* [@Husnik08], and (B) they fit well to our transmission data on arrays of different densities of split rings taken at normal incidence [@Sersic09] and as a function of incidence angle [@Sersictopublish]. The chosen values correspond to on-resonance polarizabilities $\alpha_{EE}= 4.6 V$, $\alpha_{HH}=2.1V$ and $\alpha_{EH}=2.5 V$, all well in excess of the physical SRR volume $V$ as is typical for strong scatterers. Finally, we note that the calculated albedo fits well to the albedo $a=0.5$ to $0.75$ calculated by FDTD by Husnik et al. [@Husnik08].
Radiation patterns and eigenvectors of the polarizability tensor
----------------------------------------------------------------
In Fig. \[Fig1:radpats\], we consider the eigenstates of the split ring polarizability tensor presented in Eq. (\[Eq:SRRradiationdamp\]). We first assume that the cross coupling terms are absent, i.e., $\eta_C=0$, in which case the polarizability tensor is diagonal, with eigenpolarizabilities $\alpha(\omega) \eta_E$ and $\alpha(\omega)\eta_H$. The corresponding orthogonal eigenmodes are $(p_x,m_z)=(1,0)$ and $(p_x,m_z)=(0,1)$. Figures \[Fig1:radpats\] (b) and (c) show radiation patterns of the two eigenmodes. Figure \[Fig1:radpats\](b) shows the radiation pattern of the purely electric eigenmode $(p_x,m_z)=(1,0)$ and Fig. \[Fig1:radpats\](c) shows the radiation pattern of the purely magnetic eigenmode $(p_x,m_z)=(0,1)$. Note that both $\bm{p}_x$ and $\bm{m}_z$ radiate as simple dipoles with a $\sin^2\theta$ far field radiation pattern [@jacksonbook]. The two eigenmodes can be selectively excited by impinging with a plane wave incident along the $z$-axis with $x$-polarized $E$-field (electric eigenmode), or with a plane wave incident along the $x$-axis with $y$-polarization ($z$-polarized $H$-field, magnetic eigenmode). The extinction cross section of a single split ring at these two incidence conditions is set by $\sigma_{\mathrm{ext}}=4 \pi k \mathrm{Im}(\alpha_{EE})$ and $\sigma_{\mathrm{ext}}=4 \pi k \mathrm{Im}(\alpha_{HH})$.
Next, we consider extinction and eigenmodes for arbitrary values of the cross coupling. It is easy to see that the extinction cross section at the two special illumination conditions (incident along $z$, $x$-polarized, respectively, incident along $x$, with $y$-polarization) remain equal to $\sigma_{ext}=4 \pi k
\mathrm{Im}(\alpha_{EE})$ and $\sigma_{ext}=4 \pi k
\mathrm{Im}(\alpha_{HH})$. However, for nonzero $\eta_C$, these incidence conditions and polarizabilities do not correspond anymore to the eigenvalues and eigenvectors of the polarizability tensor, which now have mixed magneto-electric character. In the extreme case of strongest magneto-electric coupling ($\eta_C=\sqrt{\eta_E\eta_H}$), the eigenvectors reduce to $(p_x,m_z)=(1,i\sqrt{\eta_E/\eta_H})$ and $(p_x,m_z)=(1,-i\sqrt{\eta_H/\eta_E})$. The associated far-field radiation patterns of these eigenvectors correspond to coherent superpositions of the radiation pattern of an $x$-oriented electric dipole, and a $z$-oriented magnetic dipole, with a quarter wave phase difference. Figures \[Fig1:radpats\](d,e) show the on-resonance radiation pattern, assuming $\eta_E=0.7$, $\eta_H=0.3$ and $\eta_C=$0.4`.` Note that these parameters are close to the limit of strongest possible magneto-electric coupling. Figures \[Fig1:radpats\](d,e) reveal that the radiation pattern of each eigenmode is non-dipolar. Rather than a $\sin^2\theta$ donut-shaped pattern, an elongated radiation pattern occurs, with maximum extent in the $y$-direction. The polarization in the far field is linear for directions along the cartesian axis, but is generally elliptical.
Extinction cross sections to measure polarizability
---------------------------------------------------
{width="\figwidth"}
Figure \[Fig2:extinction\] shows the extinction cross section predicted by our point scattering model of a single split ring for different incidence conditions. In Fig. \[Fig2:extinction\](a), the incident wave vector is swept from the $z$-direction to the $y$-direction, while maintaining $x-$ polarized light. For this set of incidence conditions the resulting extinction cross sections only depend on $\alpha_{EE}$ and $\alpha_{HH}$, and are entirely independent of the off-diagonal coupling strength $\alpha_{EH}$. The cross section increases from $\sigma_{\mathrm{ext}}= 4 \pi k\mathrm{Im}\alpha_{EE}$ as the split ring is only driven by the incident $E_x$ field when light is incident along $z$, to $\sigma_{\mathrm{ext}}= 4 \pi
k(\mathrm{Im}\alpha_{EE}+ \mathrm{Im}\alpha_{HH})$, as the split ring is driven by the incident $E_x$ field plus the incident $H_z$ field. When the wavevector is rotated to the the $x$-axis, the extinction cross section diminishes to $4\pi k \mathrm{Im}\alpha_{HH}$, as the split ring is only driven by $H_z$. The chosen values $\eta_E=0.7,
\eta_H=0.3$ and $\eta_C=0.4$ that we also used for Fig. \[Fig1:radpats\](d,e) yield extinction cross sections $\sigma_{\mathrm{ext}}= 4 \pi k\mathrm{Im}\alpha_{EE}=0.29~\mu$m$^2$ and $\sigma_{\mathrm{ext}}= 4 \pi
k\mathrm{Im}\alpha_{HH}=0.13~\mu$m$^2$. The predicted $\sigma_{\mathrm{ext}}= 4 \pi k\mathrm{Im}\alpha_{EE}=0.29~\mu$m$^2$ is consistent with the measurement ($\sigma_{\mathrm{ext}}=0.3~\mu$m$^2$) reported by Husnik et al. [@Husnik08]. It is important to note that measurements along cartesian incidence directions and with linear cartesian polarizations yield only the diagonal elements of the polarizability tensor. Indeed, the proposed measurements form a redundant set of measurements of $\alpha_{EE}$, $\alpha_{HH}$, and $(\alpha_{EE}+\alpha_{HH})$, but do not provide any insight into the magneto-electric cross coupling in the electrodynamic polarizability tensor.[@diagonalnote]
In order to measure the eigenpolarizabilities, it is necessary to selectively address the eigenvectors of the polarizability tensor. As noted above, the eigenvectors in the case of strong magneto-electric coupling $\eta_C\approx\sqrt{\eta_E\eta_H}$ tend to $(p_x,m_z)= (1,i\sqrt{\eta_E/\eta_H})$ and $(1,-i\sqrt{\eta_H/\eta_E})$. These eigenvectors require simultaneous driving by $E_x$ and $H_z$, with a quarter wave phase difference. We note that such fields can be generated by circularly polarized light with incident wave vector constrained to the $xz$-plane. Indeed, at maximally strong magneto-electric coupling and $\eta_E=\eta_H$, circularly polarized light incident at $45^\circ$ from the $z$-axis would selectively excite exactly one eigenmode. Therefore, we expect angle-resolved extinction measurements for oppositely handed circularly polarized beams to reveal the eigenpolarizabilities. Figure \[Fig2:extinction\](b) plots the extinction cross section for right handed circular polarization, as a function of angle of incidence in the $z$-plane, for illumination tuned to the LC resonance frequency. Naturally, at normal incidence the extinction is exactly half the extinction obtained for linear polarization, as a consequence of the fact that $\textit{E}_y$ does not interact with the split ring at all. Strikingly, the extinction cross section is predicted to behave asymmetrically as a function of incidence angle. The extinction increases when going to positive angle and decreases when going to negative angle. Changing handedness is equivalent to swapping positive and negative angles. A detailed analysis shows that the maximum in extinction corresponds to the largest eigenvalue of the polarizability tensor ($\sigma_{ext} = 2 \pi k \mathrm{Im}
\alpha_+$), while the minimum in extinction corresponds to the smallest eigenvalue ($\sigma_{ext} = 2 \pi k \mathrm{Im}
\alpha_-$). Therefore, circularly polarized measurements reveal the eigenvalues of the polarizability tensor. Combining such circularly polarized extinction measurements with the measurements under cartesian incidence in Fig. \[Fig2:extinction\](a), therefore allows to extract all components of the polarizability tensor. In addition to the contrast in extinction, the angle at which the maximum circular dichroism occurs is a second, independent measure for the magneto-electric coupling strength. The measurements in Fig. \[Fig2:extinction\](a) and (b) together hence provide full, even redundant, information on $\eta_E$, $\eta_H$ and $\eta_C$.
Structural chirality
--------------------
The results plotted in Fig. \[Fig2:extinction\](b) show that magneto-electric coupling in the 6$\times$6 polarizability tensor directly implies structural chirality. It is exhilarating that this interesting phenomenon first reported by [@Plum09; @plum07] for the transmission of arrays of scatterers is naturally present in the theory. However, while previous analysis of structural chirality focused on transmission through periodic arrays, we predict that circular dichroism already appears in the extinction cross section of a single split ring, with a strength set by how close the magneto-electric coupling strength is to its limit $\sqrt{\eta_E,\eta_H}$. The circular dichroism in extinction occurs independently of whether there is material loss, as opposed to, e.g., asymmetric transmission phenomena through arrays, that are claimed to require dissipation [@plum07]. For maximally magneto-electrically coupled systems, the smallest eigenvalue is identically zero, implying that such a scatterer is transparent for one circular polarization, and achieves its strongest scattering for the opposite handedness. We expect that our 6$\times$6 polarizability tensor can be successfully used to describe all structurally chiral scatterers reported today, as well as clusters and periodic arrays thereof.
{width="\textwidth"}
A coupled system: Split ring dimers\[section:coupledSRR\]
=========================================================
So far, this manuscript has focused purely on the scattering properties of single magneto-electric point scatterers. In the remainder of the paper we illustrate that our method can be easily used to analyze multiple scattering by magneto-electric scattering clusters. In order to calculate the response of a system of coupled magneto-electric dipoles, we generalize the general self-consistent equation that describes scattering of clusters of electric dipoles $\bm{p}$ as reviewed in [@Abajo07]. Assuming a system of $N$ magneto-electric point scatterers situated at positions $\bm{r}_1\ldots\bm{r}_{N}$, the response upon illumination by an incident field $(\bm{E}_{\mathrm{in}}(\mathbf{r}),\bm{H}_{\mathrm{in}}(\mathbf{r}))$ is determined by a set of $N$ self consistent equations for the induced dipole moments in each scatterer. The dipole moment induced in scatterer $n$ with polarizability tensor $\boldsymbol{\alpha}_n$ is $$\left(
\begin{array} {c}
\bm{p}_n \\
\bm{m}_n
\end{array} \right)
=\boldsymbol{\alpha}_n\left[\left(
\begin{array} {c}
\bm{E}_{\mathrm{in}}(\mathbf{r}_n)\\
\bm{H}_{\mathrm{in}}(\mathbf{r}_n)
\end{array} \right)
+\sum_{\substack{q=1\ldots N \\ q\neq n}}
\boldsymbol{G }^0(\bm{r}_n,\bm{r}_q) \left(
\begin{array} {c}
\bm{p}_q\\
\bm{m}_q
\end{array}
\right)\right]
\label{Eq:selfconsistent}$$ Using this equation we can attempt to reinterpret recent measurements that evidence significant coupling in split rings in 2D arrays, as well as in oligomers [@Sersic09; @Giessen09; @Feth10; @guo07]. Here we focus on the extinction of a dimer of split rings in socalled ‘stereodimer’configuration, first studied by Liu et al. [@Giessen09]. Figure \[Fig3:stereosystem\] shows such a ‘stereodimer’, consisting of two SRRs in vacuum ($V=200\times 200 \times 30$ nm$^3$, resonant at a wavelength around 1500 nm), both parallel to the $xy$ plane, vertically stacked with a small height difference of 150 nm. The upper SRR is rotated by a twist angle $\psi$ around the $z$-axis. On the basis of the report by Liu et al. [@Giessen09], we expect two resonance peaks with an angle dependent splitting, which can be explained in an LC model as the summed effect of electric dipole-dipole coupling and magnetic dipole-dipole coupling.
We calculate the extinction versus twist angle and wavelength of an incident beam incident from the $+z$ direction, with $x$-polarization. This beam directly excites $\bm{p}_x$ in both rings, which also drive each other. We first analyze the experiment assuming that there is no magneto-electric coupling term (setting $\eta_C=$0, although we keep $\eta_E=$0.7 and $\eta_H=0.3$). As Fig. \[Fig3:stereosystem\](b) shows, the extinction shows a single strong resonance that is blueshifted relative to the single SRR resonance at 200 THz. As a function of twist angle, this broad resonance redshifts to 200 THz at a twist of 90$^\circ$, and shifts back to 220 THz at a twist of 180$^\circ$. There is no sign of a second resonance, which might be hidden below the strong resonance. To bring out the second resonance more clearly, we reduce the loss in Fig. \[Fig3:stereosystem\](b), to a 10 times lower value $\gamma=1.25\cdot 10^{13}~$s$^{-1}$) for gold in Fig. (c) and (d). For this almost absorption-free system, Fig. \[Fig3:stereosystem\](c) indeed shows two resonances in extinction. The blue shifted resonance is now observed to cross with a narrow red shifted resonance. The crossing is symmetric around $90^\circ$ and is consistent with the hybridization of an electric dipole fixed along $x$, with a second one above it twisted by an amount $\psi$. The two branches have a very different width and strength, consistent with the fact that a symmetric configuration of dipoles couples more strongly to external fields (blue shifted resonance), than an antisymmetric ‘dark’ configuration (red shifted resonance).
To verify whether the two resonances observed in Fig. \[Fig3:stereosystem\](a) are all resonances in the system, we change the angle of incidence to 45$^\circ$ in the $xz$ plane, so that the exciting field has an $H_z$ component to drive the split rings, in addition to an $E_x$ component. Figure \[Fig3:stereosystem\](d) shows that in this case four resonances occur in extinction. In addition to the two curved bands excited by $E_x$, there are also two non-dispersive bands with a twist independent splitting. Obviously, these bands are due to the coupling of two magnetic dipoles in symmetric (broad and intense band) and antisymmetric head-to-tail configuration. The existence of four instead of two modes is a new insight compared to LC circuit models [@Giessen09; @Giessen10], but is logical in view of the fact that split rings have both a magnetic and an electric response, which are decoupled under the assumption $\eta_C=0$.
Next we analyze the extinction in presence of magneto-electric coupling, setting $\eta_C= 0.4$. Again, we first examine the extinction in presence of realistic loss ($\gamma=1.25\cdot
10^{14}~$s$^{-1}$) for gold in Fig. \[Fig3:stereosystem\](e). As also predicted by FDTD simulations by Liu et al. [@Giessen09], there appear to be two bands. The blue-shifted band is again very broad, but now has a frequency shift away from the single SRR resonance that is significantly larger for twist angle $180^\circ$ than for $0^\circ$. These effects were explained by Liu et al. as due to an additive (subtractive) correction to the dominant electric hybridization at twist angle $180^\circ$ ($0^\circ$) that occurs due to magnetic dipole coupling. A surprise is that the diagram is not symmetric anymore around $90^\circ$ twist as in the case of zero magnetic coupling. Instead, the extinction appears to show an anticrossing at twist angle $60^\circ$ These features were also predicted by FDTD simulations by Liu et al. [@Giessen09] However, the presence of an anticrossing at twist angle $\psi=60^\circ$ could not be interpreted Liu et al [@Giessen09] within an LC electrostatic circuit model, except by invoking higher order multipolar corrections. Here we see that a purely dipolar model may also explain all features of the experiment provided that magneto-electric coupling is accounted for. While we do not claim that multipolar effects are not present in actual experiments, it is an important insight that split ring polarizabilities with magneto-electric coupling terms may provide much richer physics then expected from electrostatic circuit theory. A main advantage of point dipole theory is that the underlying mode structure does not need to be recouped from FDTD simulations, but is easily resolved by repeating a calculation of extinction cross sections with low loss (as done in Fig. \[Fig3:stereosystem\]), or by analyzing the poles of the coupling matrix in Eq. (\[Eq:selfconsistent\]) that relates $(\bm{p},\bm{m})$ to $(\bm{E}_{\mathrm{in}},\bm{H}_\mathrm{in})$. The computational effort for $N$ split rings is equivalent to diagonalizing or inverting a $6N\times 6N$ matrix.
To more clearly bring out all the resonances we artificially reduce the damping $\gamma=1.25\cdot 10^{13}~$s$^{-1}$ to ten times less than the damping of gold, and plot the response of the system under normal incidence (f) and $45^\circ$ incidence (g) in Fig. \[Fig3:stereosystem\] (f,g). The anticrossing at twist angle $\psi=60^\circ$ appears to be due to the coupling of four modes, as opposed to the intuition from LC circuit theory that only two resonances anticross. The existence of four, rather than two modes in a split ring dimer appears surprising and is a second indication of the rich physics of magneto-electric scatterers. Intuition from LC circuits is that although the subspace of driving fields is two dimensional ($E_x$ and $H_z$), nonetheless only one mode per split ring exists. The usual reasoning in LC models is that the relation between electric and magnetic dipole moment is completely fixed and independent of driving, since the loop current and accumulated charge are directly related. Such a constraint is not general: in electrodynamic multipole expansions, magnetic polarizabilties are determined independently from the electric ones. The intuition from LC theory that there is only one mode per scatterer is only retrieved in our model right at the limit of strongest magneto-electric coupling $\eta_C=\sqrt{\eta_E\eta_H}$, since in that case one polarizability is identically zero. We note that the values $\eta_E=0.7,
\eta_H=0.3, \eta_C=0.4$ used in this work (that we fitted to our angle-resolved transmission experiments on 200x200 nm Au split rings on glass) are close to the limit of strong magneto-electric coupling. Whether a general argument exists why physical scatterers are or are not exactly at the limit of strongest magneto-electric coupling $\eta_C=\sqrt{\eta_E\eta_H}$ is a question outside the scope of this paper.
Conclusion
==========
In conclusion, we have developed a new multiple scattering theory by means of which we can calculate scattering and extinction for any magneto-electric scatterer with known polarizability tensor, as well as for arbitrary finite clusters. As opposed to LC circuit models, our new model obeys energy conservation, contains all interference effects, and allows quantitative prediction of absolute cross sections, spectral linewidths and lineshapes. While outside the scope of this paper, the theory is readily extended to deal with arbitrary periodic lattices by generalizing Ewald lattice sums [@Abajo07] to deal with both $\bm{E}$ and $\bm{H}$. Since the electrodynamic polarizability tensor can be directly constructed from electrostatic circuit theory, we expect that our model is readily applicable to many current experiments using chiral and nonchiral metamaterial building blocks for which electrostatic models have been proposed.
Our model does not give any insight into whether the response of a given structure is truly dipolar or not. Also, our model does not provide any insight or quantitative predictions based on microscopic considerations for the magnitude of the polarizability. For such microscopic considerations, based on, e.g., current density distributions derived from full wave simulations, we refer to [@Rockstuhl06b; @Rockstuhl07; @Zhou07; @Corrigan08; @Pors10; @RockstuhlMulti]. Rather, our model allows one to verify if specific data or microscopic calculations are consistent at all with point dipole interactions, allowing to verify or falsify common intuitive explanations in literature that have sofar always been based on electrostatic considerations. Also, our model allows one to assess if a single polarizability tensor indeed can describe a range of different experiments with, e.g., split ring clusters, as should be expected from a consistent model. Finally, our model is the simplest electrodynamical model to consistently describe how metamaterials and photonic crystals are formed from magneto-electric scatterers. A first step is to confirm the parameters used in this work for $\eta_E, \eta_H$ and $\eta_C$ by targeted experiments. While the value for $\eta_E$ used in this work is consistent with the extinction cross section measured by Husnik et al. [@Husnik08], we propose that the new insight that magneto-electric coupling is far stronger than the magnetic polarizability be confirmed by off-normal circularly polarized extinction measurements as proposed in section \[section:singleSRRexp\].
The most important property of our theory is that a polarizability tensor validated for a single scatterer can readily be used to predict all quantitative scattering properties of composite lattices and antennas. We hence expect that new insights can be obtained in effective medium constants of metamaterial arrays. Our analytical model not only facilitates design, but will also for the first time allow to determine rigorously whether, even in the ideal case (no loss, no multipole corrections), metamaterial building blocks can give rise to a desired $\epsilon$ and $\mu$, despite the large importance of electrodynamic corrections [@RockstuhlWAT; @Rockstuhl06; @Sersic09]. In addition to generating new insights for metamaterials, our theory also opens new design routes for gratings and antennas with unprecedented polarization properties. As an example, in this paper we analyzed the four mode anticrossing due to magneto-electric coupling in stereo-dimers. This analysis is easily extended to magneto-electric Yagi-Uda antennas, diffractive gratings of chiral building blocks, and magneto-inductive waveguides that may provide new ways to control the propagation and emission of light [@Koenderink06; @Koenderink09; @Genovguide].
We thank Ad Lagendijk for stimulating and inspirative insights, as well as Dries van Oosten and Lutz Langguth for discussions. This work is part of the research program of the “Stichting voor Fundamenteel Onderzoek der Materie (FOM),” which is financially supported by the “Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO).”
Unit system\[appendix\]
=======================
Throughout this paper we used units that significantly simplify notation throughout, as they maximimize the interchangeability of electric and magnetic fields. Conversion to SI units is summarized in Table \[table\].
[|l||c|l|]{}\
Quantity & Symbol & Relation to SI\
Electric field & $\bm{E}$ & $\bm{E}_{\mathrm{SI}}$\
Magnetic field & $\bm{H}$ & $ Z \bm{H}_{\mathrm{SI}}$\
& &\
Electric dipole moment & $\bm{p}$ & $ \bm{p}_{\mathrm{SI}}/(4\pi\epsilon)$\
Magnetic dipole moment & $\bm{m}$ & $ \bm{m}_{\mathrm{SI}} (Z/(4\pi))$\
& &\
Electric-electric polarizability & $\boldsymbol{\alpha}_{EE}$ & $ \boldsymbol{\alpha}^{\mathrm{SI}}_{EE}/(4\pi\epsilon)$\
Magnetic-magnetic polarizability & $\boldsymbol{\alpha}_{HH}$ & $\boldsymbol{\alpha}^{\mathrm{SI}}_{HH}/(4\pi)$\
Electric-magnetic polarizability & $\boldsymbol{\alpha}_{EH}$ & $\boldsymbol{\alpha}^{\mathrm{SI}}_{EH}(c/(4\pi))$\
Magnetic-electric polarizability & $\boldsymbol{\alpha}_{HE}$ & $\boldsymbol{\alpha}^{\mathrm{SI}}_{HE}(Z/(4\pi))$\
& &\
Electric-electric Green tensor & $\boldsymbol{G}_{EE}$ & $ 4\pi\epsilon \boldsymbol{G}^{\mathrm{SI}}_{EE}$\
Magnetic-magnetic Green tensor & $\boldsymbol{G}_{HH}$ & $ 4\pi \boldsymbol{G}^{\mathrm{SI}}_{EE}$\
Electric-electric Green tensor & $\boldsymbol{G}_{EH}$ & $ 4\pi/Z\, \boldsymbol{G}^{\mathrm{SI}}_{EE} $\
Magnetic-magnetic Green tensor & $\boldsymbol{G}_{HE}$ & $ 4\pi/c\, \boldsymbol{G}^{\mathrm{SI}}_{EE}$\
\
\
For the conversion in Table \[table\], we use $\epsilon$ for the host dielectric constant, $c$ for the velocity of light, and $Z$ for the impedance of the background medium. In this unit system, a plane wave has $|\bm{E}|/|\bm{H}|=1$, and intensity $I=|\bm{E}|^2/(2Z)$, since the Poynting vector is $\bm{S}=1/(2Z)
\mathrm{Re}(\bm{E}^{*}\times\bm{H})$. In these units, the cycle-averaged work done by an electric field $\bm{E}$ to drive an oscillating $\bm{p}$ equals $W=2\pi k/Z \mathrm{Im}(\bm{E}\cdot
\bm{p})$. The magnetic counterpart is $W=2\pi k/Z
\mathrm{Im}(\bm{H}\cdot \bm{m})$
V. G. Veselago, Sov.Phys. USPEKHI **10**, 509-514 (1968).
J. B. Pendry, Phys. Rev. Lett. **85**, 3966 (2000).
J. B. Pendry, Physics World **14**, 47 (2001); C. M. Soukoulis, S. Linden, and M. Wegener, Science **315**, 47 (2007); V. M. Shalaev, Nature Photonics **1**, 41 (2007).
U. Leonhardt, Science **312**,1777 (2006); J. B. Pendry, D. Schurig, and D. R. Smith, *ibid*, 1780 (2006).
D. R. Smith, W. J. Padilla, D. C. Vier, S. C. Nemat-Nasser, and S. Schultz, Phys. Rev. Lett. **84**, 4184 (2000); W. J. Padilla, A. J. Taylor, C. Highstrete, M. Lee, and R. D. Averitt, *ibid.* **96**, 107401 (2006); S. Linden, C. Enkrich, M. Wegener, J. Zhou, T. Koschny, and C. M. Soukoulis, Science **306**, 1351 (2004).
C. Enkrich, M. Wegener, S. Linden, S. Burger, L. Zschiedrich, F. Schmidt, J. F. Zhou, T. Koschny, and C. M. Soukoulis, Phys. Rev. Lett. **95**, 203901 (2005).
C. Rockstuhl, T. Zentgraf, H. Guo, N. Liu, C. Etrich, I. Loa, K. Syassen, J. Kuhl, F. Lederer, and H. Giessen, Appl. Phys. B **84**, 219 (2006).
M. W. Klein, C. Enkrich, M. Wegener, C. M. Soukoulis, and S. Linden, Opt. Lett. **31**, 1259 (2006).
I. Sersic, M. Frimmer, E. Verhagen and A. F. Koenderink, Phys. Rev. Lett. **103**, 213902 (2009).
B. Lahiri, S. G. McMeekin, A. Z. Khokhar, R. M. De La Rue, and N. P. Johnson, Opt. Expr. **18**, 3210 (2010).
V. M. Shalaev, W. Cai, U. K. Chettiar, H.-K. Yuan, A. K. Sarychev, V. P. Drachev, and A. V. Kildishev, Opt. Lett. **30**, 3356 (2005).
G. Dolling, C. Enkrich, M. Wegener, J. F. Zhou, C. M. Soukoulis, and S. Linden, Opt. Lett. **30**, 23 (2005).
G. Dolling, C. Enkrich, M. Wegener, C. M. Soukoulis, and S. Linden, Opt. Lett. **31**, 12 (2006).
G. Dolling, C. Enkrich, M. Wegener, C. M. Soukoulis, and S. Linden, Science **312**, 892 (2007)
J. Valentine, S. Zhang, T. Zentgraf, E. Ulin-Avila, D. A. Genov, G. Bartal, and X. Zhang, Nature **455**, 376 (2008).
S. P. Burgos, R. de Waele, A. Polman, and H. A. Atwater, Nature Mat. **9**, 407 (2010).
M. Husnik, M. W. Klein, N. Feth, M. König, J. Niegemann, K. Busch, S. Linden and M. Wegener, Nature Photonics **2**, 614 (2008).
C. Rockstuhl, F. Lederer, C. Etrich, T. Zentgraf, J. Kuhl, and H. Giessen, Opt. Expr. **14**, 8827 (2006).
T. D. Corrigan, P. W. Kolb, A. B. Sushkov, H. D. Drew, D. C. Schmadel, and R. J. Phaneuf, Opt. Expr. **16**, 19850 (2008).
A. Pors, M. Willatzen, O. Albrektsen, and S. I. Bozhevolnyi, J. Opt. Soc. Am. B **27**, 1680 (2010).
E. Prodan, C. Radloff, N. J. Halas and P. Nordlander, Science **302**, 419 (2003).
P. Banzer, U. Peschel, S. Quabis, and G. Leuchs, Opt. Expr. **18**, 10905 (2010). N. Feth, M. König, M. Husnik, K. Stannigel, J. Niegemann, K. Busch, M. Wegener, and S. Linden, Opt. Express **18**, 6545 (2010).
M. Decker, S. Burger, S. Linden, and M. Wegener, Phys. Rev. B **80**, 193102 (2009).
J. K. Gansel, M. Thiel, M. S. Rill, M. Decker, K. Bade, V. Saile, G. von Freymann, S. Linden, and M. Wegener, Science **325**, 1513 (2009). E. Plum, J. Zhou, J. Dong, V. A. Fedotov, T. Koschny, C. M. Soukoulis, and N. I. Zheludev, Phys. Rev. B **79**, 035407 (2009). E. Plum, X.-X. Liu, V. A. Fedotov, Y. Chen, D. P. Tsai, and N. I. Zheludev, Phys. Rev. Lett. **102**, 113902 (2009).
S. Zhang, Y.-S. Park, J. Li, X. Lu, W. Zhang, and X. Zhang, Phys. Rev. Lett. **102**, 023901 (2009).
B. Wang, J. Zhou, T. Koschny, M. Kafesaki, and C. M. Soukoulis, J. Opt. A: Pure Appl. Opt **11**, 114003 (2009).
E. Plum, V. A. Fedotov and N. I. Zheludev, J. Opt. A: Pure Appl. Opt. **11**,074009:1-7 (2009).
M. Decker, M. W. Klein, M. Wegener, S. Linden, Opt. Lett. **32**, 856 (2007).
M. Decker, R. Zhao, C. M. Soukoulis, S. Linden, M. Wegener, Opt. Lett. **35**, 1593 (2010). N. Liu, H. Liu, S. Zhu and H. Giessen, Nature Photonics **3**, 157 (2009).
H. Guo, N. Liu, L. Fu, T. P. Meyrath, T. Zentgraf, H. Schweizer, and H. Giessen, Opt. Expr. **15**, 12095 (2007). J. Petschulat, J. Yang, C. Menzel, C. Rockstuhl, A. Chipouline, P. Lalanne, A. Tüennermann, F. Lederer, and T. Pertsch, Opt. Express **18**, 14454 (2010). H. Liu, J. X. Cao, S. N. Zhu, N. Liu, R. Ameling and H. Giessen, Phys. Rev. B **81**, 241403(R) (2010).
C. Rockstuhl, T. Zentgraf, E. Pshenay-Severin, J. Petschulat, A. Chipouline, J. Kuhl, T. Pertsch, H. Giessen, and F. Lederer, Opt. Expr. **15**, 8871 (2007). J. Zhou, Th. Koschny, and C. M Soukoulis, Opt. Expr. **15**, 17881 (2007).
A. Lagendijk and B. A. van Tiggelen, Phys. Rep. **270**, 143 (1996). P. de Vries, D. V. van Coevorden and A. Lagendijk, Rev. Mod. Phys. **70**, 2 (1998). F. J. García de Abajo, Rev. Mod. Phys. **79**, 1267 (2007).
L. D. Landau and E. M. Lifshitz, *Electrodynamics of Continuous Media*, Pergamon, Oxford (1960).
I. V. Lindell, A. H. Sihvola, S. A. Tretyakov, and A. J. Viitanen, *Electromagnetic Waves in Chiral and Bi-Isotropic Media*, Artech House, Norwood MA (1994).
R. Merlin, Proc. Nat. Acad. Sci. **106**, 1693 (2009).
W. H. Weber and G. W. Ford, Phys. Rev. B. **70**, 125429 (2004).
A. F. Koenderink and A. Polman, Phys. Rev. B **74**, 033402 (2006). (2007).
J. Garcìa-Garcìa, F. Martìn, J. D. Baena, R. Marqès and L. Jelinek, J. Appl. Phys. **98**, 033103 (2005).
C. F. Bohren and D. R. Huffman, *Absorption and Scattering of Light by Small Particles*, John Wiley & Sons, New York (1983).
N. Katsarakis, T. Koschny, M. Kafesaki, E. N. Economou, and C. M. Soukoulis, Appl. Phys. Lett. **84**, 2943 (2004).
Note that the above equation should strictly be written with $\boldsymbol{\alpha}$ replaced by the $t$-matrix, which is directly proportional to the dynamic polarizability $\boldsymbol{\alpha}$ for point scatterers [@Lagendijk96].
J. D. Jackson, *Classical Electrodynamics (3rd ed.*, John Wiley & Sons, New York (1999).
H. C. van de Hulst, [*Light Scattering by Small Particles*]{} (Dover, New York, 1981).
M. Meier and A. Wokaun, Opt. Lett. **8**, 581 (1983); K. T. Carron, W. Fluhr, A. Wokaun and H. W. Lehmann, J. Opt. Soc Am. B **3**, 420 (1986); K. L. Kelly, E. Coronado, L. L. Zhao and G. C. Schatz, J. Phys. Chem. B **107**, 668 (2003); A. Wokaun, J. P. Gordon and P. F. Liao, Phys. Rev. Lett. **48**, 1574 (1982).
M. Burresi, D. van Oosten, T. Kampfrath, H. Schoenmaker, R. Heideman, A. Leinse and L. Kuipers, Science **326**, 550 (2009); M. Burresi, T. Kampfrath, D. van Oosten, J. C. Prangsma, B. S. Song, S. Noda and L. Kuipers, Phys. Rev. Lett. **105**, 123901 (2010).
M. G. Silveirinha,Phys. Rev. B **82**, 085101 (2010)
R. Zhao, J. Zhou, Th. Koschny, E. N. Economou, and C. M. Soukoulis, Phys. Rev. Lett. **103**, 103602 (2009); M. G. Silveirinha and S. I. Maslovski, Phys. Rev. Lett. **105**, 189301 (2010); R. Zhao, J. Zhou, Th. Koschny, E. N. Economou, and C. M. Soukoulis, Phys. Rev. Lett. **105**, 189302 (2010).
P. B. Johnson and R. W. Christy, Phys. Rev. B **6**, 4370 (1972).
I. Sersic, A. Opheij and A. F. Koenderink, *in preparation*.
Note that the diagonal elements of the electrodynamic polarizability do contain contributions due to off-diagonal elements in the electrostatic tensor. The radiation damping correction in Eq. (\[Eq:SRRradiationdamp\]) mixes $\eta_C$ onto the diagonal.
C. Menzel, T. Paul, C. Rockstuhl, T. Pertsch, S. Tretyakov, and F. Lederer, Phys. Rev. B **81**, 035320 (2010).
A. F. Koenderink, Nano Lett. **9**, 4228 (2009).
H. Liu, D. A. Genov, D. M. Wu, Y. M. Liu, J. M. Steele, C. Sun, S. N. Zhu, and X. Zhang, Phys. Rev. Lett. **97**, 243902 (2006).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The influences of intense coherent laser fields on the transport properties of a single layer graphene are investigated by solving the time-dependent Dirac equation numerically. Under an intense laser field, the valence band and conduction band states mix via the optical Stark effect. The chiral symmetry of Dirac electrons is broken and the perfect chiral tunneling is strongly suppressed. These properties might be useful in the fabrication of an optically controlled field-effect transistor.'
author:
- 'Jiang-Tao Liu'
- 'Fu-Hai Su'
- Hai Wang
- 'Xin-Hua Deng'
title: The influence of the optical Stark effect on chiral tunneling in graphene
---
Graphene has attracted much attention due to its remarkable electronic properties [@11KS; @12KS; @13AH]. The low-energy quasiparticles, which have linear dispersion and nontrivial topological structure in their wave function, can be described by using a Dirac-like equation. This unique band structure of graphene leads to many important potential applications in nanoelectronics [@16VV; @17MI; @17ZZ; @17VH; @17PM; @17EP].
One of the peculiar transport phenomena in graphene is the chiral tunneling [@17MI; @16VV; @A1WR]. In single layer graphene a perfect transmission through a potential barrier in the normal direction is expected. This unique tunneling effect can be explained by the chirality of the Dirac electrons within each valley, which prevents backscattering in general. This kind of reflectionless transmission is independent of the strength of the potential, which limits the development of graphene-based field-effect transistors (FET) [@17MI]. The perfect transmission can be suppressed effectively when the chiral symmetry of the Dirac electrons is broken. For instance, in a magnetic field, a quantized transmission can be observed in graphene *p-n* Junctions [@20DA]. Recently, Elias et. al. proposed that the hydrogenation could convert the semimetal graphene into an insulator material [@A3DC].
The intense optical field can also break the chiral symmetry of Dirac electrons in graphene, e.g., Fistul and Efetov have shown that when the n-p Junctions in graphene is irradiated by an electromagnetic field in the resonant condition, the quasiparticle transmission is suppressed [@21MV]. The optical field control on carrier transport offers several advantages. Optical fields can control not only the charge carriers but also the spin carriers, especially which can be performed over femtosecond time scale. Another fundamental method of optical control is the optical Stark effect (OSE) [@1AM; @2SS; @3CE; @4MC; @A2DF]. The OSE in traditional semiconductors is due to a dynamical coupling of excitonic states by an intense laser field. The OSE have shown many useful applications in optoelectronics and spintronics [@5MC; @6CE; @7SS; @9WY; @10JT].
In graphene, the valence band and conduction band states can also mix strongly via OSE. Thus the chirality of Dirac electrons will be completely changed, or even disappear. Unlike the resonant case [@21MV], in OSE the coherent excitons are virtual excitons, which exist only when the optical field is present. Thus the light-induced shift lasts only for the duration of the pump pulse, which allows for optical gates that might only exist for femtoseconds. Furthermore, since there is no real absorption in the nonresonant case, the absorption of photons is quite small and low power consumption is expected.
![(color online). (a) Schematic of the spectrum of Dirac electrons in single-layer graphene. The optical field is propagated perpendicular to the layer surface and and is linearly polarized along the $Y$ direction. (b) Schematic of the scattering of Dirac electrons by a square potential. $B_{a}$, $B_{in}$, and $B_{out}$ denote the absorbing boundary, incident boundary, and output boundary, respectively. (c) Schematic of the one-dimensional Yee lattice in graphene.[]{data-label="fig1"}](fig1.eps){width="0.98\columnwidth"}
In this Letter, we study the tunneling rate of Dirac electrons in graphene through a barrier with an intense electromagnetic field. We consider a rectangular potential barrier with height $V_{0}$, width $D$ in the $X$ direction, and infinite length in the $Y$ direction \[see Fig. 1 (a) and Fig. 1 (b)\]. The Fermi level (dashed lines) lies in the valence band in the barrier region and in the conduction band outside the barrier. The gray filled areas indicate the occupied states. The optical field is propagated perpendicular to the layer surface and is linearly polarized along the $Y$ direction with a detuning $\Delta_{0}=2
E_{b}-\hbar \omega$. We choose $\Delta_{0}>0$ to ensure that there is no interband absorption inside the barrier. Meanwhile, $\hbar
\omega \ll 2 E_{k}$ is used to guarantee that the influence of the optical field outside the barrier can be neglected.
Since the Coulomb interaction between electrons and holes in OSE is negligible when the detuning is large [@4MC; @5MC], we did not take into account the electron-hole Coulomb interaction or many body effect in our calculation. Thus, neglecting the scattering between different valleys, the scattering process of Dirac electrons in $K$ point is described by the time-dependent Dirac equation $$i\hbar \frac{\partial }{\partial t}\mathbf{\Psi }\left(
\mathbf{r},t\right) =\left[ \mathbf{H}_{0}+V_{0}\left(\mathbf{r}
\right) \mathbf{ I}+\mathbf{H}_{int}\right] \mathbf{\Psi
}\left(\mathbf{r},t\right),\label{eq1}$$ where $\mathbf{\Psi }\left( \mathbf{r}, t\right)=[C_{A}(\mathbf{r},
t),C_{B}(\mathbf{r},t)] $ is the wave function, $\mathbf{H}_{0}=-i\hbar v_{F}\mathbf{\sigma}\bullet \nabla$ is the unperturbed Dirac Hamiltonian, $\mathbf{\sigma}=(\sigma_{x},\sigma_{y})$ are the Pauli matrices, $v_{F}\approx 10^{6}m/s$ is the Fermi velocity, $V_{0}(\mathbf{r})$ is the height of the potential barrier, $\mathbf{I}$ is the unit matrix, and $\mathbf{H}_{int}$ is the interaction Hamiltonian. $\mathbf{H}_{int}$ can write as [@28EJ] $$\mathbf{H}_{int}=-\hbar ev_{F}\left[ A (x,t)\sigma _{x}+A (y,t)\sigma
_{y}\right] =\hbar \left(
\begin{array}{cc}
0 & V_{12}(t) \\
V_{21}(t) & 0%
\end{array}%
\right),$$ where $e$ is the electron charge and $[A (x,t),A (y,t)]=[A_{x}e^{i\omega t}, A_{y}e^{i\omega t}]$ are the vector potentials of the electromagnetic field. When the Dirac electrons is incident on the barrier perdenicularly, we can rewrite Eq. (\[eq1\]) as a set of partial differential equations
$$\begin{aligned}
%\begin{align}
i\partial C_{A}(x,t)/\partial t=&-iv_{F}\partial C_{B}(x,t)/\partial
x +V_{0}C_{A}(x,t) \nonumber\\
&+V_{12}(t)C_{B}(x,t), \label{eqO1}\\
%\end{align}
%\begin{align}
i\partial C_{B}(x,t)/\partial t=&-iv_{F}\partial C_{A}(x,t)/\partial
x+V_{0}C_{B}(x,t)\nonumber\\
&+V_{21}(t)C_{A}(x,t). \label{eqO2}
%\end{align}\end{aligned}$$
Since the tunneling time is sub-picosecond and the potential $V_{12}(t)$ and $V_{21}(t)$ vary as fast as the frequency of incident light beams, this scattering process is strongly time-dependent. In order to study such a strongly time-dependent scattering process, we employ the finite-difference time-domain (FDTD) method to solve Eq. (\[eqO1\]) and Eq. (\[eqO2\]) numerically in the time-domain [@26KS]. In the traditional FDTD method, the Maxwell’s equations are discretized by using central-difference approximations of the space and time partial derivatives. As a time-domain technique, the FDTD method can demonstrate the propagation of electromagnetic fields through a model in real time. Similar to the discretization of Maxwell’s equations in FDTD, we denote a grid point of the space and time as $(i,k)=(i\Delta x,k \Delta t)$ \[see Fig. 1(c)\], and for the any function of space and time $F(i\Delta x,k \Delta t)=F^{k}(i)$. the first order in time or space partial differential can be expressed as $$\begin{aligned}
&\frac{\partial F\left( x,t\right) }{\partial x}|_{x=i\Delta x}\approx
\frac{F^{k}\left( i+1/2\right) -F^{k}\left( i-1/2\right) }{\Delta x},\\
&\frac{\partial F\left( x,t\right) }{\partial t}|_{t=k\Delta t}\approx
\frac{F^{k+1/2}\left( i\right) -F^{k-1/2}\left( i\right) }{\Delta t}.\end{aligned}$$ Thus the Eq. (\[eqO1\]) and Eq. (\[eqO2\]) can be replaced by a finite set of finite differential equations
$$\begin{aligned}
C_{A}^{k+1/2}(i)&\left[\frac{1}{\Delta t}-\frac{V_{0}(i)}{2i}\right] =\left[\frac{%
1}{\Delta t}+\frac{V_{0}(i)}{2i}\right]C_{A}^{k-1/2}(i) \nonumber \\%
&-\left[ \frac{ v_{F}}{\Delta x}-\frac{V_{12}^{k}(i+1/2)}{2i}\right]
C_{B}^{k}(i+1/2)\nonumber\\%
&+\left[ \frac{v_{F}}{\Delta x}+\frac{V_{12}^{k}(i-1/2)}{2i}\right]
C_{B}^{k}(i-1/2),\label{fdtda}\end{aligned}$$
$$\begin{aligned}
C_{B}^{k+1}&(i+1/2)\left[ \frac{1}{\Delta t}-\frac{V_{0}(i+1/2)}{2i}\right] =%
\left[ \frac{1}{\Delta t}+\frac{V_{0}(i+1/2)}{2i}\right]\times\nonumber \\& C_{B}^{k}(i+1/2)%
-\left[ \frac{v_{F}}{\Delta x}-\frac{V_{21}^{k+1/2}(i+1)}{2i}\right]
C_{A}^{k+1/2}(i+1)\nonumber \\
&+\left[ \frac{v_{F}}{\Delta x}+\frac{V_{21}^{k+1/2}(i)}{2i}%
\right] C_{A}^{k+1/2}(i),\label{fdtdb}\end{aligned}$$
For computational stability, the space increment $\Delta x$ and the time increment $\Delta t$ need to satisfy the relation $\Delta x>v_{F}\Delta t$ [@26KS]. Furthermore, the space increment $\Delta x$ must far smaller than the wavelength of electrons $\Delta x<\lambda_{e}/8$, and the time increment $\Delta t$ must be far smaller than the period of the electromagnetic field $T_{l}$.
At the boundary $B_{a}$, one-dimensional Mur absorbing boundary conditions are used [@29GM]. At the input boundary $B_{in}$, a Gaussian electronic wave packet is injected $$C_{A}=C_{B}=\frac{1}{\sqrt{2}}\exp \left[ -\frac{4\pi (t-t_{0})^{2}}{\tau
^{2}}\right],$$ where $t_{0}$ and $\tau$ denote the peak position and the pulse width, respectively.
![(color online). (a) numerical simulations of a wave packet tunneling through a barrier without pump beams. (b)-(d) Time sequence of a wave packet tunneling through a barrier with pump intensity $I_{\omega}=3$ MW/cm$^{2}$, $\Delta_{0}=5meV$, and $D=300$ nm. The light grey shows the barrier area.[]{data-label="fig2"}](fig2.eps){width="0.9\columnwidth"}
Thus, by solving Eq. (\[fdtda\]) and Eq. (\[fdtdb\]) directly in the time domain we can demonstrate the propagation of a wave packet through a barrier in real time. Numerical simulations are shown in Fig. 2. The following parameters are used in our calculation: the peak position $t_{0}=1.5$ ps, the pulse width $\tau=1.0$ ps, the space increment $\Delta x=0.1$ nm, the time increment $\Delta
t=5\times10^{-5}$ ps, and the height of the potential barrier $V_{0}=400$ meV. When there is no pump beams, a perfect chiral tunneling can be found \[see Fig. 2(a)\]. This result is consistent with that of Geim et. al. [@17MI]. But when the sample is irradiated by an intense nonresonant laser beam, a reflected wave packet appears \[see Fig. 2(d)\]. The perfect transmission is suppressed. By analyzing the the transmitted wave packet and the reflected wave packet, we can obtain the tunneling rate.
To explain the suppression of chiral tunneling, We first investigate the OSE in the barrier within a rotating-wave approximation [@2SS; @9WY; @10JT]. Figure 2(a) shows the renormalized band as a function of momentum $k$ with intensity $I_{\omega}=30$ MW/cm$^{2}$. In the case of nonresonant excitation, $\hbar \omega <2E_{b}$ and the dressed states are blue shifted. With increasing detuning, the light-induced shift decreases, and the dressed states asymptotically approach the unperturbed states. The intense electromagnetic field can also induce a strong band mixing. Near the absorption edge, a maximum fermion distribution function $n_{k}\approx 0.44$ can be observed \[see Fig. 1(b)\].
![(color online). (a) Sketch of the renormalized band energies (solid lines) and the unperturbed band energies (dashed lines) as a function of momentum $k$. (b) Sketch of the fermion distribution function $n_{k}$ as a function of momentum $k$. (c) The reflectance (red circles) and the transmittance (black squares) of the barrier as a function of the detuning for $I_{\omega}=30$ MW/cm$^{2}$ and $D=300$ nm. (d) The transmittance as a function of pump intensity for $\Delta_{0}=5$ meV with different barrier width.[]{data-label="fig3"}](fig3.eps){width="0.98\columnwidth"}
Under intense light beams, the dressed states are strongly mixed with valence states and conduction states. Therefore, the chiral symmetry of Dirac electrons in graphene can be broken. For instance, at very small detuning, the wave functions of these dressed states can be approximately written as the superposition of unperturbed conduction and valence wave function, $\Psi=(\Psi_{+}+\Psi_{-})/\sqrt{2}=(1,0)$. These dressed states are not the eigenstates of the helicity operator. The chiral symmetry is broken and perfect chiral tunneling is strongly suppressed. Numerical results are shown in Fig. 2(c) with pump intensity $I_{\omega}=30$ MW/cm$^{2}$ and $D=300$ nm. From Fig. 2(c) we can find that the transmission is strongly suppressed, even with lager detuning (e.g., $\Delta_{0}=10$ meV, the transmittance is about 0.025). When detuning increases, the light-induced mixing becomes weak \[see Fig. 2(b)\], the reflectance decreases, and the transmittance increases. Fig. 2(d) shows the transmittance as a function of pump intensity with different barrier widths. The strong laser field can enhance band mixing and reduce the transmittance. From Fig. 2(d) we also see that the wide barrier can prolong the interaction time between electrons and photons, reduce the tunneling rate, and lower the threshold of the pump laser power.
In conclusion, we have calculated the influence of the OSE on the chiral tunneling in graphene by using the FDTD method. We find that perfect tunneling can be strongly suppressed by the optically induced band mixing, even at large detuning. These properties might be useful in device applications, such as the fabrication of an optically controlled field-effect transistor that has ultrafast switching times and low power consumption.
This work was supported by the NSFC Grant Nos. 10904059, 10904097, and 11004199, the NSF from Jiangxi Province 2009GQW0017, the Open Research Fund of State Key Laboratory of Millimeter Waves No. K200901.
[99]{}
K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva, A. A. Firsov, Science **306**, 666 (2004).
K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. Katsnelson, I. V. Grigorieva, S. V. Dubonos, A. A. Firsov, Nature **438**, 197 (2005).
A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. **81**, 109 (2009).
M. I. Katsnelson, K. S. Novoselov, and A. K. Geim, Nature Physics **2**, 620 (2006).
V. V. Cheianov, V. Fal’ko, and B. L. Altshuler, Science **315**, 1252 (2007).
Z. Z. Zhang, K. Chang, and K. S. Chan, Appl. Phys. Lett. **93**, 062106 (2008).
V. H. Nguyen, A. Bournel, V. L. Nguyen, and P. Dollfus, Appl. Phys. Lett. **95**, 232115 (2009).
P. Michetti, M. Cheli, and G. Iannacconea, Appl. Phys. Lett. **96**, 133508 (2010).
E. Prada, P. San-Jose, and H. Schomerus Phys. Rev. B **80**, 245414 (2009).
W.-R. Hannes and M. Titov, Europhysics Letters **89**, 47007 (2010).
D. A. Abanin and L. S. Levitov, Science **317**, 641 (2007).
D. C. Elias, R. R. Nair, T. M. G. Mohiuddin, S. V. Morozov, P. Blake, M. P. Halsall, A. C. Ferrari, D. W. Boukhvalov, M. I. Katsnelson, A. K. Geim, K. S. Novoselov, Science **323**, 610 (2009).
M. V. Fistul and K. B. Efetov, Phys. Rev. Lett. **98**, 256803 (2007).
A. Mysyrowicz, D. Hulin, A. Antonetti, A. Migus, W. T. Masselink, and H. Morkoc, Phys. Rev. Lett. **56**, 2748 (1986).
S. Schmitt-Rink, D. S. Chemla, and H. Haug, Phys. Rev. B **37**, 941 (1988).
C. Ell, J. F. Müller, K. El Sayed, and H. Haug, Phys. Rev. Lett. **62**, 304 (1989).
M. Combescot and R. Combescot, Phys. Rev. B **40**, 3788 (1989).
D. Fröhlich, B. Uebbing, T. Willms, and R. Zimmermann, Europhys. Lett. **23** 489 (1993)
M. Combescot, Physics Reports **221**, 168 (1992).
C. E. Pryor and M. E. Flatté, Appl. Phys. Lett. **88**, 233108 (2006).
S. Sanchez, C. De Matos, and M. Pugnet, Appl. Phys. Lett. **89**, 263510 (2006).
W. Yao, A. H. MacDonald, and Q. Niu, Phys. Rev. Lett. **99**, 047401 (2007).
J. T. Liu, F. H. Su, and H. Wang, Phys. Rev. B **80**, 113302 (2009).
E. J. Mele, P. Král, and D. Tománek, Phys. Rev. B **61**, 7669 (2000).
K. S. Yee, IEEE Transactions on Antennas and Propagation, **14** 302 (1966).
G. Mur, IEEE Transactions on Electromagnetic Compatibility **EMC-23**, 377 (1981).
| {
"pile_set_name": "ArXiv"
} |
\#1 \#1[| \#1 |]{}
- .375 in
TRI-PP-93-86\
Nov. 1993
-.375 in
Robert Garisto
-.375 in
*TRIUMF, 4004 Wesbrook Mall, Vancouver, B.C., V6T 2A3, Canada*
It is well known that supersymmetry (SUSY) gives neutron and electron electric dipole moments ($d_n$ and $d_e$) which are too large by about $10^{3}$. If we assume a SUSY model cannot contain fine-tunings or large mass scales, then one must require that the SUSY breaking mechanism give real soft breaking parameters, in which case the minimal SUSY model has no $CP$ violation other than from the CKM matrix (besides possible strong $CP$ violating effects). We show that in non-minimal SUSY models, a moderate amount of $CP$ violation can be induced through one loop corrections to the scalar potential, giving an effective phase of order $10^{-3}$, and thus implying $d_n$ and $d_e$ can be near their current experimental bounds $naturally$. This moderate amount of SUSY $CP$ violation could also prove important for models of electroweak baryogenesis. We illustrate our results with a specific model.
Introduction
============
Predictions for $CP$ violating effects in supersymmetric (SUSY) theories have often been discussed with a certain ambiguity. On the one hand, it is well known that when the complex quantities in the theory are allowed to have phases of order unity, the predicted neutron and electron electric dipole moments ($d_n$ and $d_e$) are typically too large by perhaps $10^{3}$ [@hist; @dn; @recent; @dn; @Dugan; @etal; @Kizukuri; @arnowitt]. In order to avoid this, the relevant quantities are often chosen to be real, in which case the theory predicts no non-Standard Model $CP$ violation (CPV) and negligible $d_n$ and $d_e$ [@foot; @strong]. On the other hand, it has often been assumed that the observation of $d_n$ around the current limit of $10^{-25}$[e$\,$cm]{} [@dnexpt] could easily be accommodated by a SUSY theory with the phases somehow reduced by just the right amount. These ideas are clearly in conflict: one cannot have a theory which avoids fine-tunings by setting the SUSY parameters real, and at the same time expect $d_n$ near its current upper bound. The purpose of this paper is to describe a mechanism by which a moderate amount of SUSY $CP$ violation can naturally appear in a theory in which the soft SUSY breaking terms have been taken real.
The superpotential of the Minimal Supersymmetric Standard Model (MSSM) contains the Yukawa sector of the theory, $W_Y$, and a Higgs mixing term,
$$W_{MSSM} = W_Y + \mu {\rm H}_u {\rm H}_d,
\label{MSSM superpot}$$
where ${\rm H}_u$ and ${\rm H}_d$ are Higgs doublet superfields. If the soft breaking terms come from the superpotential, as in (\[L soft\]), then one can use ${\rm H}_u$ and ${\rm H}_d$ to rotate away the phase of $\mu$.
In order to avoid an additional hierarchy problem brought on by $\mu/M_{GUT} \ll 1$ [@Kim; @Nilles], the MSSM is often extended by adding a singlet superfield ${\rm N}$, whose scalar component’s vacuum expectation value (VEV) generates the Higgs mixing term (see [@Ellis; @etal] and references therein). We refer to this model as the N+MSSM. One can use an $R$ symmetry to forbid $B$ and $L$ violating terms in $W_Y$, and to allow only cubic terms involving N, so that the superpotential can be written as
$$W_{N+MSSM} = W_Y + h {\rm N} {\rm H}_u{\rm H}_d + a {\rm N}^3 .
\label{NMSSM superpot}$$
Note that we can use the Higgs and singlet ${\rm N}$ superfields to rotate away the phases of $h$ and $a$. This again assumes that the soft SUSY breaking Lagrangian can be written in the low energy supergravity (SUGRA) parametrization [@Raby]:
$$-{\cal L}_{soft} = \abs{m_i}^2 \abs{\varphi_i}^2 +
\Bigl( {1 \over2} \sum_\lambda\tilde m_\lambda \lambda \lambda +
\bar A \left[ W^{(3)} \right]_\varphi +
\bar B \left[ W^{(2)} \right]_\varphi + h.c.\Bigr),
\label{L soft}$$
where $\varphi_i$ are the scalar superpartners, $\lambda$ are the gauginos, and $[\ ]_\varphi$ means take the scalar part. Here $W^{(2)}$ and $W^{(3)}$ are the quadratic and cubic pieces of the superpotential, so that in the MSSM, $W^{(3)} = W_Y$, and $W^{(2)}=\mu{\rm H}_u{\rm H}_d$; and in the N+MSSM, $W^{(3)}=W$. We have defined the soft breaking parameters $\bar A\equiv A m_0^*$ and $\bar B\equiv B m_0^*$ to include a mass scale $m_0$. The parameters $A$, $B$, their mass scale $m_0$, and the gaugino masses $\tilde m_\lambda$, can all be complex. These parameters contribute to $d_n$ at the order of $10^{-22} \tilde\varphi / \tilde M^2$[e$\,$cm]{}, where $\tilde \varphi$ is a combination of the phases of the parameters, and $\tilde M^2$ is a combination of superpartner masses, normalized to the weak scale. The only known ways to make such a large $d_n$ compatible with the experimental upper bound are to fine-tune the phase $\tilde \varphi$ to order $10^{-3}$; have superpartner masses of order a few TeV; or somehow require all the phases to naturally be zero [@SUSYdn]. Both the first and second approach eliminate much of the attractiveness of SUSY [@Kane; @etal]. For example, having large superpartner masses virtually eliminates the possibility of radiative breaking of $SU_2\times U_1$, which was one of the major successes of SUSY. Losing this is especially undesirable now that the top mass is large enough to make it work. So we will henceforth assume that $A$, $B$, $m_0$ and $\tilde m_\lambda$ are all real [@foot; @physical]. We do $not$ have an explanation for how these conditions will be satisfied, but merely state that any complete SUSY model which has superpartners of order the weak scale must either satisfy these criteria, or provide an explanation for how their phases could naturally be of order $10^{-3}$ [@foot; @high; @E; @spont].
Imposing these ‘no fine-tuning criteria’ means that the only source of CPV in either the MSSM or the N+MSSM is the CKM phase [@foot; @strong]. CKM contributions to $d_n$ and $d_e$ from renormalization group running [@Dugan; @etal] and from finite effects [@arnowitt; @SUSYdn] are below $10^{-30}$ [e$\,$cm]{}, and are thus unobservably small. So if $d_n$ or $d_e$ were detected in the near future, could a SUSY theory with superpartners of order the weak scale explain them without resorting to fine-tunings?
There have also been some interesting models of baryogenesis at the electroweak scale [@Turok; @Zadrozny; @Dine; @BAU; @Cohen; @Nelson], which require CPV beyond the CKM phase [@foot; @Shaposhnikov]. Could one construct a model with sufficient CPV for electroweak baryogenesis, while satisfying the upper bounds on $d_n$ and $d_e$, without fine-tunings?
With these questions in mind, we describe a mechanism by which a moderate amount of CPV can naturally arise in a non-minimal SUSY theory through loop corrections to the Higgs potential. The idea is that a phase which is unobservable at tree level can introduce an observable effective phase through loop effects. This effective phase will always be smaller than a tree level observable phase because of the usual factors of suppression associated with loops. Such a phase can make moderate contributions to $d_n$ and $d_e$, and may be useful in explaining the observed baryon asymmetry.
In Section II, we present an illustrative model which provides an existence proof for this moderate CPV mechanism. In that section, we show that at tree level in the scalar potential, SUSY CPV is essentially unobservable. In Section III, we show how this CPV can appear in the observable sector in the one loop effective potential. The magnitude of this CPV will be suppressed by loop coefficients, so that $d_n$ and $d_e$ can naturally be near their current experimental bounds.
The Model
=========
Let us construct a model which has complex couplings only to terms which contain singlet scalar fields which have zero VEVs. We also need these particles to have no tree level couplings to quarks or leptons. This means that the Higgs scalar potential will be $CP$ conserving, as will all tree level vertices outside the neutral Higgs sector. At this order, there is no one loop contribution to $d_n$ or $d_e$. After one loop corrections to the scalar potential ($V$), a small phase can be induced into these vertices, which generates moderate $d_n$ and $d_e$. To do this in a model which is technically natural, one needs to add at least two such singlets $(N',\ N'')$ to the N+MSSM. In order that they have zero VEVs, we impose a discrete symmetry on their superfields: $({\rm N}',{\rm N}'') \rightarrow -({\rm N}',{\rm N}'')$. Then the most general cubic superpotential respecting this additional symmetry is:
$$W = W_Y +
h {\rm N} {\rm H}_u{\rm H}_d + a {\rm N}^3 + c' {\rm N}{{\rm N}'}^2 + c''
{\rm N}{{\rm N}''}^2
+ b {\rm N}{\rm N}'{\rm N}'' .
\label{model superpot}$$
One sees that the fields ${\rm N}'$ and ${\rm N}''$ have no direct couplings to quarks, leptons, or gauge particles. We will see below that they can each have zero VEVs, and thus their couplings will not affect the tree level minimum of the scalar potential. They do not affect $CP$ violating observables studied to date (at tree level in $V$), so we term this sector [*invisible*]{}. This is merely nomenclature. It should be possible to detect these particles, and perhaps even to see $CP$ violating effects directly in processes in which they are produced, but they are certainly invisible when considering one loop processes involving only external quarks, leptons, and gauge bosons.
Notice that we do not have enough freedom to rotate away all the phases in (\[model superpot\]), and that after making the visible sector $CP$ conserving, the reparametrization invariant $b^2 {c'}^* {c''}^*$ can be complex [@foot; @alt; @choice]. This is the phase which will produce CPV in the one loop scalar potential. Our first task is to be sure that this phase does not produce any CPV in the visible sector at tree level in $V$, else $d_n$ will again be too large.
All supersymmetric contributions to $d_n$ come from the mass matrices of squarks and gauginos—if the mass matrices can all be made real, the SUSY contribution to $d_n$ disappears. If they are complex, the gaugino-squark-quark couplings become complex and contribute to $d_n$ through loop diagrams [@hist; @dn]. Let us write the down squark mass matrix in a partially diagonalized basis: $$\pmatrix{ {\mu_{dL}}^2 {\bf 1} + \hat M_D^2 &
(\bar A^* - hn e^{i \theta_1}\, \tan\beta) \hat M_D\cr
(\bar A^* - hn e^{i \theta_1}\, \tan\beta)^* \hat M_D&
{\mu_{dR}}^2 {\bf 1} + \hat M_D^2\cr } , \label{MDsqk}$$
where $\hat M_D$ is the diagonal, real, $N_F \times N_F$ quark mass matrix (where $N_F$ is the number of families), and ${\mu_{q\, L,R}}^2 \sim \abs{m_{3/2}}^2$. Here $n=\abs{\vev{N}}$ (so that $hn$ takes the place of $\mu$ of the MSSM), and $\tan\beta$ is the ratio of Higgs VEVs. The angle $\theta_1$ is one of the relative phases between the three VEVs, and is defined in (\[theta defs\]). If (as we have assumed) the soft breaking parameters are real, and the minimum of the scalar potential $V$ is $CP$ conserving, then this matrix is real.
Next we can write the chargino mass matrix, $M_{\chi^+}$, $$\pmatrix { \tilde m_{W} & g_2v_2 \cr
g_2v_1 & h n e^{i \theta_1} \cr} ,
\label{Mchargino}$$ in the basis of [@HaK] (with the argument of the Higgs VEVs rotated into $\theta_1$). Here $\tilde m_{W}$ is the $SU_2$ soft breaking gaugino mass, and $g_2$ is the $SU_2$ coupling constant. Again, if the minimum of $V$ is $CP$ conserving, then this matrix is real.
Since we have added three neutral fields to the MSSM (or two to the N+MSSM), the neutralino mass matrix, $M_{\chi^0}$, is $7\times7$. We extend the basis of [@HaK] with $\psi_N,\ \psi_{N'},$ and $\psi_{N''}$: $$\pmatrix{
\tilde m_{B} & 0 & -g_1v_1/\sqrt2 & g_1v_1/\sqrt2
& 0 & 0 & 0 \cr
0 & \tilde m_{W} & g_2v_1/\sqrt2 &-g_2v_2/\sqrt2
& 0 & 0 & 0 \cr
-g_1v_1/\sqrt2 & g_2v_1/\sqrt2 & 0 &-h ne^{i\theta_1}
& -h v_1 & 0 & 0 \cr
g_1v_2/\sqrt2 & -g_2v_2/\sqrt2 & -h ne^{i\theta_1}& 0
& -h v_2 & 0 & 0 \cr
0 & 0 & -h v_1 & -h v_2
& 3ane^{i(\theta_3- 2\theta_1)} & X & Y \cr
0&0&0&0& X & c'n & bn \cr
0&0&0&0& Y & bn & c''n \cr
},
\label{Mneutralino}$$
where $g_1$ is the $U_1$ coupling constant, and $\tilde m_{B}$ is the $U_1$ gaugino mass. The angle $\theta_3$ is also defined in (\[theta defs\]). The cross terms $X$ and $Y$, which mix $\psi_{N'}$ and $\psi_{N''}$ with the visible sector, are proportional to the VEVs of $N'$ and $N''$, so that if these VEVs are zero, $\psi_{N'}$ and $\psi_{N''}$ decouple from the visible sector. The resulting $5\times5$ visible sector matrix is real, if $\sin\theta_1=\sin\theta_3=0$. In that case, all of the mass matrices are real, and there is no new SUSY contribution to $d_n$.
To see if this is the case, we must consider the scalar potential $V$. We define two Higgs doublets of the same hypercharge, and their VEVs, $$\vev{\phi_1} \equiv \vev{{H_d}^c} \equiv
{\left\lgroup \matrix{0 \cr v_1 \cr }\right\rgroup} ,\
\vev{\phi_2} \equiv \vev{H_u} \equiv
{\left\lgroup \matrix{0 \cr v_2 e^{i \xi}\cr} \right\rgroup},
\label{defofphis}$$
and the VEVs of the singlet fields
$$\vev{N} \equiv n e^{i \varphi},\
\vev{N'} \equiv n' e^{i \varphi'},\
\vev{N''} \equiv n'' e^{i \varphi''}.$$
It turns out in this model that if $n'=n''=0$, there are only three combinations of these VEV phases,
$$\theta_1 \equiv \xi + \varphi,\ \,
\theta_2 \equiv \theta_1 - \theta_3 =\xi - 2\varphi,\ \,
\theta_3 \equiv 3 \varphi,
\label{theta defs}$$
which appear in the tree level scalar potential. Elsewhere, a more general linear combination of $\theta_1$ and $\theta_3$ (with integer coefficients) can appear.
Let us write the scalar potential for our model,
$$\begin{aligned}
V &&\!\!\!\!\!= (h^2 \abs{N}^2 + m_1^2) \abs{\phi_1}^2 +
(h^2 \abs{N}^2 + m_2^2) \abs{\phi_2}^2
\nonumber\\
&&\!\!\!\!\!- \left[ \left( \bar A h N + 3 a h {N^*}^2 + h {c'}^* {N'}^{*2}
+ h {c''}^* {N''}^{*2} \right) (\phi_1^\dagger \phi_2) + H.c. \right]
\nonumber\\
&&\!\!\!\!\!+\lambda_1 (\phi_1^\dagger \phi_1)^2
+ \lambda_2 (\phi_2^\dagger \phi_2)^2 +
\lambda_3 (\phi_1^\dagger \phi_1)(\phi_2^\dagger \phi_2) +
\lambda_4 (\phi_1^\dagger \phi_2)(\phi_2^\dagger \phi_1)
\nonumber\\
&&\!\!\!\!\!+ m_0^2 \abs{N}^2 + m_0^2 \abs{N'}^2 + m_0^2 \abs{N''}^2
+ 9a^2 \abs{N}^4 + \abs{c'}^2 \abs{N'}^4 + \abs{c''}^2 \abs{N''}^4 +
\label{scalar potential}\\
&&\!\!\!\!\!+ (4 \abs{c'}^2 + \abs{b}^2) \abs{NN'}^2
+ (4 \abs{c''}^2 + \abs{b}^2) \abs{NN''}^2
\nonumber\\
&&\!\!\!\!\!+ \Bigl[ \bar A a N^3 + \bar A c' N {N'}^2 + \bar A c'' N {N''}^2
+ \bar A b NN'N''
\nonumber\\
&&\!\!\!\!\!+ 2(b^* c' + b {c''}^*) \abs{N}^2 N' {N''}^*
+ 3 ac' {N^*}^2 {N'}^2 + 3 ac'' {N^*}^2 {N''}^2
+ c' {c''}^* {N'}^2 {N''}^{*2} + H.c. \Bigr]
\nonumber\end{aligned}$$
where [@GaH]
$$\lambda_1=\lambda_2 = (g_2^2 + g_1^2)/8,\ \lambda_3 = (g_2^2 - g_1^2)/4,\
\lambda_4= h^2 - g_2^2/2,$$
which is just the scalar potential for the N+MSSM [@Ellis; @etal] plus terms which involve $N'$ and $N''$. The minimum of $V$ can be written as
$$\begin{aligned}
\vev{V} &= \vev{V_{N+MSSM}} + \, K_{20} {n'}^2 + K_{11} n'n'' + K_{02} {n''}^2
&\nonumber\\
&+K_{40} {n'}^4 + K_{22} {n'}^2{n''}^2 + K_{04} {n''}^4,
&\label{vev V}\end{aligned}$$
where $K_{ij}$ depend upon all the other parameters. One can show that for any choice of the parameters in $V_{N+MSSM}$, there exists a set of $\{b,c',c''\}$ such that $n'=n''=0$ is a true minimum. We will assume that this condition is satisfied, so that $\vev{V} = \vev{V_{N+MSSM}}$.
Finally we must be sure there is no problem with spontaneous CPV. As we said, the potential depends only upon the three angles $\theta_{{\rm 1-3}}$ (only two of which are independent). We can write
$$\vev{V} = \alpha_0 - \alpha_1 \cos\theta_1
- \alpha_2 \cos\theta_2
- \alpha_3 \cos\theta_3,$$
where the $\alpha_i$ are functions of the magnitudes of the three VEVs. Differentiating with respect to $\theta_1$ and $\theta_3$, we see that one solution to $\vev{V'}=0$ is $\sin\theta_i=0$. Rom$\tilde{\rm a}$o [@Romao] showed that for this potential ($i.e.$ $\vev{V_{N+MSSM}}$), this is the only stable minimum—that the spontaneous $CP$ violating solution is actually a saddle point. Babu and Barr [@Babu; @Barr] make the interesting claim that this can be made into a minimum by large radiative corrections to the Higgs mass matrix, but these require very heavy squark masses (in which case hard CPV need not be suppressed by fine-tuning [@Kizukuri]), small charged Higgs mass, and $\tan\beta \sim {\cal O}(1)$. These conditions make satisfying the CLEO bound on $b \rightarrow s \gamma$ nearly impossible [@my; @BSG]. It is also unlikely that a model satisfying these conditions could be consistent with such things as Grand Unification and solutions to the Dark Matter problem [@Kane; @etal]. Anyway, we can certainly choose parameters such that the minimum of $V$ is $CP$ conserving, and such that $n'=n''=0$, so that all the SUSY mass matrices (\[MDsqk\])-(\[Mneutralino\]) are real at tree level.
A Loop Induced Observable Phase
===============================
It would seem that since the tree level potential is $CP$ conserving, one could not have a one loop potential which is $CP$ violating. The important point to remember is that even though the [*visible*]{} sector has no CPV, there are still $CP$ violating couplings to $N'$ and $N''$. Consider, for example, Figure 1, which gives a purely finite contribution to a new term in $V$, $\delta \lambda_5 (\phi_1^\dagger \phi_2)^2 + H.c$ (this term is not present in the tree level potential, so it must be finite). The vertices are proportional to ${c'}^*$ and ${c''}^*$, and the mixing between $N'$ and $N''$ contains pieces proportional to $b$. Thus $\delta \lambda_5 \sim b^2 {c'}^* {c''}^*$, which has a reparametrization invariant phase. For $b$, $c'$, $c''$ of order $1/2$, $\delta \lambda_5$ can be of order $10^{-3}$.
Actually, the operator (to which Figure 1 contributes) is more accurately written as $k \vev{N}^2 (\phi_1^\dagger \phi_2)^2$, which gives a contribution to $\vev{\delta V}$ of $2 \abs{k} n^2 v_1^2 v_2^2 \cos(2 \theta_1 + {\rm Arg} k)$, where ${\rm Arg} k$ is just $\theta_{CP} \equiv {\rm Arg}(b^2 {c'}^* {c''}^*)$. We can write the general correction to $\vev{V}$ as
$$\vev{\delta V} = \sum_{x,y,z}^{integers}
\kappa_{x,y,z} \cos(x \theta_1 + y \theta_3 + z \theta_{CP}),
\label{delta V}$$
where the $\kappa_{x,y,z}$ are real coefficients, with the subscripts $x,\ y,\ z$ taking on all integral values, though the $\kappa_{x,y,z}$ become negligible for large integers. Since our model is renormalizable [@renorm] and $\vev{V}$ is $CP$ conserving at tree level, all one loop terms in (\[delta V\]) with $z \neq 0$ must be finite. Note that Figure 1 gives a finite contribution to (\[delta V\]) with $(x,y,z) = (2,0,1)$.
The perturbation in (\[delta V\]) means that $\sin\theta_i = 0$ is no longer a solution to $\vev{V(\theta_1,\theta_3)'}=0$. Since the $\kappa_{x,y,z}$ are small, the solution will lie close to this, so we can define $\theta_i = {\theta_i}_0 + \varepsilon_i$, where $\sin{\theta_i}_0=0$. The minimization condition can then be written in terms of the hessian, and the perturbation:
$$\pmatrix
{ \partial^2 V/\partial \theta_1^2 &
\partial^2 V/\partial \theta_1 \partial\theta_3 \cr
\partial^2 V/\partial \theta_1 \partial\theta_3 &
\partial^2 V/\partial\theta_3^2 \cr}
\pmatrix { \varepsilon_1 \cr \varepsilon_3 \cr}
\simeq
\pmatrix {
\sum_{x,y,z}^{integers} x
\kappa_{x,y,z} \sin(x {\theta_1}_0 + y {\theta_3}_0 + z \theta_{CP}) \cr
\sum_{x,y,z}^{integers} y
\kappa_{x,y,z} \sin(x {\theta_1}_0 + y {\theta_3}_0 + z \theta_{CP}) \cr
}
\label{epsilon eqs}$$
and solved for $\varepsilon_i$. Note that $\sin(x {\theta_1}_0 + y {\theta_3}_0 + z \theta_{CP}) =$ $\pm \sin(z \theta_{CP})$. To find the effective $CP$ violating coefficient, recall that $d_n$ gets a contribution from the imaginary part of left–right squark mixing, which goes as $\sin\theta_1 \simeq \varepsilon_1$. There will be finite contributions to $\varepsilon_1$ from several terms in $\delta V$, but they will be of the same order or smaller than that of $\delta \lambda_5$ from Figure 1. From (\[epsilon eqs\]), one finds that Figure 1 gives $\varepsilon_1 \sim \abs{\delta\lambda_5} {v^2 \over \bar A \mu}
\sin{2\beta} \sin\theta_{CP}$. The squark mixing, gluino mediated contribution to $d_n$ [@SUSYdn] due to this $\varepsilon_1$ can be written as
$$d_n \simeq 10^{-22} {\rm e\, cm}\ {\left( {100 \mbox{GeV} \over \tilde M}
\right)}^2
\abs{\delta \lambda_5}
{v^2 \over A m_0^2}
\sin\theta_{CP} ,
\label{dn expr}$$
where we have defined a SUSY mass scale
$$\tilde M^2 \equiv {\tilde m_d^4 \over m_0 \, \tilde m_g} \ .
\label{overallMsusy}$$
There is a similar neutralino mediated contribution to $d_e$ which is suppressed by $m_e/m_d$, and $\alpha_w/\alpha_s$, but enhanced by the fact that sleptons tend to be lighter than squarks. If we take $A=1$, colored superpartners $\sim 300$GeV, sleptons $\sim 150$GeV, and all other superpartners $\sim 100$GeV, one can have
$$\begin{aligned}
&&d_n \sim 10^{-26} \, \sin\theta_{CP} \, {\rm e\, cm} ,
\label{dn est}\\
&&d_e \sim 10^{-27} \, \sin\theta_{CP} \,
{\rm e\, cm} .
\label{de est}\end{aligned}$$
These estimates depend upon the parameters and the mass scales in the theory, but the point is that the contributions entering at one loop are naturally much smaller than those from SUSY phases which contribute through tree level vertices.
Concluding Remarks
==================
We have considered supersymmetric models which avoid excessively large contributions to $d_n$ and $d_e$ by requiring the ‘no fine-tuning criteria’ to be satisfied, $i.e.$ that $A$, $B$, $m_0$ and $\tilde m_\lambda$ must be real [@SUSYdn]. We showed that it is possible for moderate $CP$ violating effects to be induced at one loop in models which have singlets with zero VEVs. We used an illustrative model with superpotential (\[model superpot\]) and found that a stable minimum exists at $\vev{N'}=\vev{N''}=0$. This means that using the tree level scalar potential, no $CP$ violating effects would be detectable in conventional $CP$ violating observables because all of the SUSY mass matrices are real. We demonstrated that this model introduces small $CP$ violating phases into the one loop effective potential, so that one is left with a moderate contribution to $d_n$ and $d_e$. One could easily have $d_n$ and $d_e$ near their current experimental bounds in such a model, without the need for fine-tuning or large superpartner mass scales.
Note that SUSY contributions to $d_n$ from three gluon operators [@three; @gluon] do not affect our conclusions. Assuming that the no fine tuning criteria are satisfied, such operators will also give a negligible contribution to $d_n$. After one loop corrections to $V$ in our model, there will be small contributions to $d_n$ from these operators, but they will probably be smaller in magnitude than the quark EDM contribution [@arnowitt]. Thus (\[dn est\]) and (\[de est\]) are reasonable estimates of the natural size of SUSY CPV possible in a model such as ours.
We have discussed the issue of [*spontaneous*]{} CPV in Section II in the context of our model and concluded that we can easily choose the minimum of $V$ to be $CP$ conserving. It is worth noting that Maekawa [@Maekawa] considered generating spontaneous CPV at one loop in the MSSM, though Pomarol [@Pomarol; @one] showed that such a model requires a $CP$ odd Higgs which is too light. Pomarol also made the interesting point that a N+MSSM model (which does not rule out ${\rm H}_u{\rm H}_d$, ${\rm N}$, or ${\rm N}^2$ terms by a symmetry) with a strictly $CP$ conserving Lagrangian might violate $CP$ spontaneously at tree level with a phase of order $10^{-2}$, and might be able to explain the $\varepsilon$ parameter as well as give $d_n$ near the current experimental bound [@Pomarol; @two]. The trouble is that the fine-tuning needed by such a model of spontaneous CPV is actually much worse than that for hard CPV because the condition which must be satisfied is of the form
$$\cos\theta = \abs{{X \over Y}} \simeq 1 - {1\over 2} \theta^2,$$
where $\theta$ (or $\pi - \theta$) is the relevant spontaneous CPV phase, and $X$ and $Y$ are some combination of parameters and VEVs. We need $\theta$ to be small to satisfy the bound on $d_n$, which we can achieve only if $\delta \equiv (Y-X)/Y$ is of order $\theta^2$. For example, if we need $\theta\sim 10^{-2}$, then $\delta$ must be fine-tuned to be of order $10^{-4}$, which is completely unacceptable.
As we alluded to in the Introduction, having a moderate amount of CPV is necessary in models which generate the baryon asymmetry at the electroweak scale [@foot; @Shaposhnikov]. A recent interesting model of electroweak baryogenesis used CPV from the Higgs scalar mixing coefficient $\mu_{12}^2$ [@Turok; @Zadrozny], which can be defined as the coefficient of the $\vev{\phi_1^\dagger \phi_2}$ term in $\vev{V}$. It was pointed out [@Dine; @BAU] that $\mu_{12}^2$ can be rotated out of the Higgs potential, but the resulting phase which appears in the gaugino mass matrices was then used by [@Cohen; @Nelson]. They found that with the small phase allowed by the limit imposed by $d_n$, there is probably sufficient CPV for the observed baryon asymmetry. Our results change these conclusions in two ways. At tree level, there is [*no*]{} phase in the gaugino mass matrices (after imposing the no fine-tuning criteria), and no way for ${\rm Arg}\mu_{12}^2$ to cause CPV. Then, using the one loop effective potential in a model such as ours, there can be an effective phase $\varepsilon_1$ ($\equiv \theta_1 - {\theta_1}_0$) introduced into the gaugino mass matrix of order $\varepsilon_1 \sim 10^{-3} \theta_{CP}$. Although this is a large suppression, $\theta_{CP}$ can be of order unity and so $\varepsilon_1$ should generate the same level of CPV as the phase used in [@Cohen; @Nelson], which was bounded by $d_n$ anyway [@foot; @CaN].
From the standpoint of explaining the baryon asymmetry at the electroweak scale, or $d_n$ and $d_e$ near their current experimental bounds, a loop induced observable phase provides an attractive alternative to the fine-tuning needed in the MSSM. If $d_n$ or $d_e$ were observed in the near future, and if superpartners were determined to be of order the weak scale, SUSY model builders would have to appeal to a mechanism such as ours, which naturally explains small effective SUSY phases.
-.375 in
ACKNOWLEDGEMENTS
I sincerely appreciate helpful comments from Gordy Kane, who was also involved in the early stages of this work. I also appreciate helpful conversations with H. Haber and J. Soares. This work was supported in part by a grant from the National Sciences and Engineering Research Council of Canada.
[99]{} J. Polchinski & M. Wise, Phys. Lett. [**B125**]{}, 393 (1983); F. del Aguila, M. Gavela, J. Grifols & A. Mendez, Phys. Lett. [**B126**]{}, 71 (1983);W. Buchmuller & D. Wyler, Phys. Lett. [**B121**]{}, 321 (1983). W. Fischler, S. Paban, S. Thomas, Phys. Lett. [**B289**]{}, 373 (1992); P. Nath, Phys. Rev. Lett. [**66**]{}, 2565 (1991); L. Hall & L. Randall, Nucl. Phys. [**B352**]{}, 289 (1991); R. Mohapatra in [*CP Violation*]{}, edited by C. Jarlskog, World Scientific, Singapore (1989); I. Bigi & F. Gabbiani, Notre Dame preprint UND-HEP90-BIG06; T. Kurimoto, Prog. Theor. Phys., [**73**]{}, 209 (1985); M. Vysotsky, Sov. Phys. Usp, [**28**]{}, 667 (1985); J. G' erard, W. Grimus, A. Raychaudhuri, & G. Zoupanos, Phys. Lett. [**B140**]{}, 349 (1984). M. Dugan, B. Grinstein & L. Hall, Nucl. Phys. [**B255**]{}, 413 (1985). Y. Kizukuri & N. Oshimo, Phys. Rev. [**D46**]{}, 3025 (1992).
R. Arnowitt, M. Duff, & K. Stelle, Phys. Rev. [**D43**]{}, 3085 (1991); R. Arnowitt, J. Lopez, & D. V. Nanopoulos, Phys. Rev. [**D42**]{}, 2423 (1990). We will ignore possible strong CPV effects. While $d_n$ depends on the amount of strong CPV, a non-negligible $d_e$ could not be explained by CKM and strong CPV effects alone.
K. Smith $et\ al$, Phys. Lett. [**B234**]{}, 191 (1990).
J. Kim & H. Nilles, Phys. Lett. [**B138**]{}, 150 (1984).
J. Ellis, J. Gunion, H. Haber, L. Roszkowski, F. Zwirner, Phys. Rev. [**D39**]{}, 844 (1989).
S. Raby, Santa Cruz ’86 TASI proceedings (1986).
For a complete statement of the criteria, see R. Garisto, TRI-PP-93-24 (1993) hep-ph/9306318.
G. Kane, C. Kolda, L. Roszkowski & J. Wells, University of Michigan preprint UM-TH-93-24 (1993).
There are only two new physical phases in these parameters, but if the physical phases are zero, we can always write the theory such that $A$, $B$, $m_0$ and $\tilde m_\lambda$ are real [@SUSYdn].
It may be possible to introduce small phases at the weak scale by a spontaneous symmetry breaking at high energy. See M. Dine, R. Leigh, & A. Kagan, Phys. Rev. [**D48**]{}, 4269 (1993); and K. Babu & S. Barr, Bartol Research Inst. preprint BA-93-48 (1993) hep-ph/9309249.
N. Turok & J. Zadrozny, Nucl. Phys. [**B369**]{}, 729 (1992). M. Dine, P. Huet & R. Singleton Jr., Nucl. Phys. [**B375**]{}, 625 (1992). A. Cohen & A. Nelson, Phys. Lett. [**B297**]{}, 111 (1992); see also A. Cohen, D. Kaplan, A. Nelson, Phys. Lett. [**B294**]{}, 57 (1992).
There is a recent claim by G. Farrar and M. Shaposhnikov, Phys. Rev. Lett. [**70**]{}, 2833 (1993), that the CKM phase is sufficient for electroweak baryogenesis, but the general consensus is that CPV from the CKM phase is insufficient to explain the observed asymmetry.
For simplicity we choose the basis with $h$ and $a$ real. If one chooses to keep the phase in $a$ instead, one can show that observable CPV still occurs only at one loop in $V$, and the results are the same as in our convention.
H. Haber & G. Kane, Phys. Repts. [**117**]{}, 75 (1985). J. Gunion & H. Haber, Nucl. Phys. [**B272**]{}, 1 (1986).
J. Rom$\tilde{\rm a}$o, Phys. Lett. [**B173**]{}, 309 (1986). K. Babu & S. Barr, Bartol Research Inst. preprint BA-93-42 (1993) hep-ph/9308217.
R. Garisto & J. N. Ng, Phys. Lett. [**B315**]{}, 372 (1993).
Our model has a complete, $N=1$ supersymmetric Lagrangian, and is thus renormalizable (see for example M. Sohnius, Phys. Reports [**128**]{}, 2 (1985)).
S. Weinberg, Phys. Rev. Lett. [**63**]{}, 2333 (1989); J. Dai, H. Dykstra, R. Leigh, S. Paban & D. Dicus, Phys. Lett. [**B237**]{}, 216 (1990).
N. Maekawa, Phys. Lett. [**B282**]{}, 387 (1992). A. Pomarol, Phys. Lett. [**B287**]{}, 331 (1992). A. Pomarol, Phys. Rev. [**D47**]{}, 273 (1992).
The phase of $\mu$ used by [@Cohen; @Nelson] does not contribute to CPV unless $B m_0^*$ is complex and fine-tuned to satisfy the bound on $d_n$.
-.375 in
FIGURE CAPTION
Fig 1: One loop contribution to the operator $\delta \lambda_5 (\phi_1^\dagger \phi_2)^2$. The ‘X’ indicates $N'$–$N''$ mixing, which is required if $\delta \lambda_5$ is to contain a reparametrization invariant phase.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Spectral analysis of electron spin resonance (ESR) is a powerful technique for various investigations including characterization of spin systems, measurements of spin concentration, and probing spin dynamics. The nitrogen-vacancy (NV) center in diamond is a promising magnetic sensor enabling improvement of ESR sensitivity to the level of a single spin. Therefore, understanding the nature of NV-detected ESR (NV-ESR) spectrum is critical for applications to nanoscale ESR. Within this work we investigate the linewidth of NV-ESR from single substitutional nitrogen centers (called P1 centers). NV-ESR is detected by a double electron-electron resonance (DEER) technique. By studying the dependence of the DEER excitation bandwidth on NV-ESR linewidth, we find that the spectral resolution is improved significantly and eventually limited by inhomogeneous broadening of the detected P1 ESR. Moreover, we show that the NV-ESR linewidth can be as narrow as 0.3 MHz.'
author:
- Benjamin Fortman
- Susumu Takahashi
title: Understanding the Linewidth of the ESR Spectrum Detected by a Single NV Center in Diamond
---
Introduction
============
Electron spin resonance (ESR) spectroscopy is a powerful technique to investigate properties of magnetic systems and their local environments. [@abragram1970EPRtransitionmetals; @schweiger2001principles; @poole1983electron; @Weil2007Electron] In particular, ESR spectral analysis, in which the position, intensity, and lineshape of the ESR spectrum is carefully analyzed to extract spin parameters including g-values, hyperfine and spin-spin couplings, zero-field splittings and rotational correlation times of systems, is widely and routinely used for characterizations and investigations in science and engineering fields. Examples include identification of paramagnetic defect contents in semiconductors, [@Feher59; @Watkins98; @Loubser1978] investigations of structures and conformational dynamics of biological molecules, [@Hubbell2000; @freed11; @britt15] and characterizations of photochemical reactions. [@puluektov17; @waseilewski03; @britt00]
The nitrogen-vacancy (NV) center is of significant interest in quantum sensing due to its unique properties. [@Gruber1997; @Degen2008; @Balasubramanian2008; @Maze2008] The NV center is an $S = 1$ spin system consisting of a single vacancy site located adjacent to a nitrogen atom in the diamond lattice. The electronic structure of the NV center enables optical readout and initialization of the spin system through optically detected magnetic resonance (ODMR). [@Gruber1997] In addition, the long-lived quantum coherence of the NV centers’ spin states, [@balasubramanian08; @Takahashi2008] provides the NV center with high sensitivity to external magnetic fields, with fields as small as 100 fT being reported. [@Wolf2015] Its atomic size combined with high sensitivity to magnetic fields gives the NV center a sensing radius in the nanometer range, allowing ESR sensitivity to improve to the level of a single electron spin. [@Grinolds2013] Using a single NV center, nanoscale ESR detection of several types of spins in solid state and biological systems has been demonstrated. [@Sushkov2014; @Grinolds2013; @Abeywardana2016; @Shi2013; @Shi2015; @Mamin2012; @DeLange2012; @Rosenfeld2018; @meriles12] For applications of NV-detected ESR (denoted NV-ESR in this manuscript) spectral analysis, it is critical to understand the nature of NV-ESR lineshape and to establish a procedure to obtain a high resolution spectrum representing intrinsic properties of the sample.
Here we investigate the nature of an NV-ESR spectrum of single substitutional nitrogen defects in diamond (called P1 centers). The NV-ESR spectrum is obtained using a double electron-electron resonance (DEER) pulse sequence, which utilizes pulses at two distinct microwave (MW) frequencies to coherently control the NV center and target spins. By studying the spectral linewidth as a function of the DEER pulse length, we identify a significant contribution of the DEER excitation bandwidth to the observed NV-ESR linewidth at short pulse lengths. At long pulse lengths, we observe that the ESR linewidth is limited by inhomogeneous broadening of the detected P1 ESR frequency ($T_2^*$-limit), representing intrinsic spin dynamics of P1 spins. Moreover, by employing a long DEER pulse, we observed that the ESR linewidth is as narrow as 0.3 MHz and, with the improvement of the spectral resolution, we clearly resolve a small splitting (2 MHz) in P1 ESR that originates from the anisotropic hyperfine coupling and four different orientations of the P1 spins.
Materials and Methods
=====================
Diamond sample
--------------
A single crystal (2.0 $\times$ 2.0 $\times$ 0.3 mm$^3$) of (111)-cut high pressure high temperature type-Ib diamond (purchased from Sumitomo electric industries) was used in this study.
115 GHz ESR spectroscopy
------------------------
The 230 GHz/115 GHz ESR system employs a high-power solid-state source consists of a 9-11 GHz synthesizer, pin switch, microwave amplifiers, and frequency multipliers. The output power of the source system is 100 mW at 230 GHz and 700 mW at 115 GHz. The 230/115 GHz excitation is propagated using a quasioptical bridge and a corrugated waveguide and couples to a sample located at the center of a 12.1 T cryogenic-free superconducting magnet. ESR signals are isolated from the excitation using induction mode operation. [@Smith98] For ESR detection, we employ a superheterodyne detection system in which 115 GHz is down-converted into the intermediate frequency (IF) of 3 GHz then down-converted again to in-phase and quadrature components of dc signals. Details of the system have been described elsewhere. [@Cho2014; @Cho2015] In the present experiment, the magnetic field modulation strength was adjusted to maximize the intensity of ESR signals without distorting the lineshape (typical modulation amplitude of 0.01 mT with modulation frequency of 20 kHz).
ODMR spectroscopy
-----------------
The ODMR system is based on a homebuilt confocal microscope system. A 100-mW 532-nm laser (Crystalaser) is passed through an acousto-optic modulator (Isomet 1250C) before being directed through a low-pass filter (Omega) and into a single mode fiber (Thorlabs). The output of the fiber is directed through a dichroic mirror and up through a microscope objective (Zeiss 100X) to the sample stage. Fluorescence (FL) is detected by an avalanche photodiode (Excelitas) through a high-pass filter (Omega) and another single mode fiber. The autocorrelation measurement is performed with a Hanbury Brown-Twiss interferometer. [@Brown1956] For ODMR, microwave (MW) excitation is directed from the sources (Stanford Research Systems SG386 and Rohde-Schwarz SML03) through a power combiner, and a high gain amplifier to the sample stage. A 20 $\mu$m gold wire is placed on the surface of the diamond for MW excitation and coherent control of the NV centers.
Results and Discussion
======================
![CW ESR spectrum of P1 centers taken at 115 GHz at room temperature. The inset graph shows the spectrum of the $m_I=-1$ ESR signal. The ESR linewidth is $0.15 \pm 0.02$ mT. A field modulation of 0.01 mT at 20 kHz and a field sweep rate of $0.01$ mT/s were used.[]{data-label="fig:HF_ESR"}](Fig1.eps)
We first perform 115 GHz continuous wave ESR (CW ESR) spectroscopy to identify impurity contents within the diamond sample. Figure \[fig:HF\_ESR\] shows CW ESR data of the diamond sample with application of an external magnetic field along the \[111\] direction. We observe five pronounced ESR signals from P1 centers ($S = 1/2$, $I = 1$, ${A_{x,y}} = 82$ MHz and ${A_z} = 114$ MHz). [@Loubser1978] The intensity of 115 GHz wave excitation was reduced to avoid the saturation of the ESR signal and the intensity of magnetic field modulation was carefully adjusted to maximize the signal-to-noise ratio of the ESR signal without distortion. The P1 spectrum consists of five ESR signals due to the four possible orientations of P1 and the anisotropic hyperfine coupling. Namely, the signals at 4.104, 4.108 and 4.113 T correspond to the ESR of P1 centers oriented along the \[111\] direction while the signals at 4.105, 4.108 and 4.112 T are from the other three orientations, \[$\Bar{1}11$\], \[$1\Bar{1}1$\], and \[$11\Bar{1}$\]. As shown in the inset of Fig. \[fig:HF\_ESR\], the linewidth of the observed ESR is $0.15 \pm 0.02$ mT for the signal at 4.113 T.
![ODMR experiment to identify a NV center. (a) Diagram of the experimental setup. (b) Spatial FL image with NV1 indicated in the solid red circle. (c) Autocorrelation measurement of NV1. The dotted red line drawn at 0.5 indicates the threshold for single quantum emitters. (d) CW ODMR signals from the lower ($\ket{0} \leftrightarrow \ket{-1}$) and upper ($\ket{0} \leftrightarrow \ket{+1}$) transitions. The signal is normalized by the FL intensity without MW excitation.[]{data-label="fig:Flimage"}](Fig2.eps)
We next measure P1 ESR using a single NV center in diamond. For NV-ESR measurements, we employ a homebuilt ODMR system as shown in Fig. \[fig:Flimage\](a). Figure \[fig:Flimage\](b) shows a FL image of the diamond sample as well as an isolated FL peak for the present experiment (denoted as NV1). An autocorrelation measurement, as shown in Fig. \[fig:Flimage\](c), shows a dip in the signal at $t = 0$ that proves the FL emission is from a single quantum emitter. CW ODMR measurements, where the FL intensity is monitored while sweeping the microwave frequency, are then performed on the isolated FL spot, as shown in Fig. \[fig:Flimage\](d). The observed ODMR signals correspond to the $m_S = -1$ and $m_S = 0$, and the $m_S = +1$ and $m_S = 0$ transitions of the NV center ($S = 1$, $g = 2.0028$ and $D = 2.87$ GHz). Therefore, we determine this spot to be a single NV center. From the observed ODMR frequencies we determined the applied magnetic field to be 33.4 mT with a polar angle of $6.1 \pm 0.1$ degrees from the \[111\] axis.
![Pulsed ODMR data collected from NV1 at 33.4 mT. (a) Measurement of Rabi oscillations. The Rabi oscillations show a $\pi$ pulse time of 40 ns. The pulse sequence is shown in the inset. The data is normalized to reflect the probability of the NV center being in the $m_S=0$ state (P$\ket{m_S=0}$).[@Abeywardana2016] For all pulsed ODMR presented, a 5 $\mu$s laser pulse is used to initialize the spin state while a 300 ns laser pulse is used for readout. Microwave pulses (shown as blue rectangles) are applied to drive the $\ket{0} \leftrightarrow \ket{-1}$ transition. Each pulse sequence is repeated 10$^4$-10$^6$ times for an unweighted averaging of each data point. (b) Spin echo measurement. The spin echo data shows a spin decoherence time ($T_2$) of 40 $\mu$s for NV1. Data is shown in agreement with $I(t) = exp[-(t/T_2)^3]$.[@DeLange2010; @Wang2013; @Abeywardana2016][]{data-label="fig:DataSE"}](Fig3.eps)
![NV-ESR of NV1 with $\tau = 5.7$ $\mu$s. (a) NV-ESR spectrum obtained for NV1 with a 56-ns MW2 $\pi$-pulse. SE intensity at the same $\tau$ is shown as a reference. The DEER pulse length was chosen to maximize signal for the axial P1 orientation. ESR frequencies calculated from the P1 spin Hamiltonian are shown in the stick spectrum. (b) Rabi oscillations of P1 centers measured by NV-ESR. The NV-ESR signal is plotted against $t_p$ (solid black). The distance to the end of the sequence is indicated by $T$ and was 5.7 (2.0) $\mu$s for the upper (lower) data. SE data (Solid blue) is shown as a reference. The simulation using Eq. \[eq:deerIntViktor\] is shown in red. The pulse sequence used for the NV-ESR Rabi experiment is shown to the right.[]{data-label="fig:DEER"}](Fig4.eps)
Next, we perform the NV-ESR experiment. We first conduct Rabi oscillation and spin echo (SE) measurements to determine pulse lengths and the spin coherence time ($T_2$) for NV-ESR. As shown in the inset of Fig. \[fig:DataSE\](a), the Rabi measurement is performed by first initializing the spin state into $\ket{0}$ with a long laser pulse before applying a variable length MW pulse. The final spin state is then read out using a short laser pulse to induce FL from the NV center (see the inset of Fig. \[fig:DataSE\](a)). $T_2$ is measured using a Hahn spin echo sequence (see the inset of Fig. \[fig:DataSE\](b)). We determined the $T_2$ of NV1 to be 40 $\mu$s. After the Rabi and SE experiments, we perform NV-ESR using a DEER technique, as shown in Fig. \[fig:DEER\](a). NV-ESR is performed by measuring the change in a coherent state of the NV center as a function of the frequency of the DEER pulse. The coherent state change is induced by a shift of the magnetic dipole field of target spins due to the population inversion of target spins induced by the DEER pulse. For this measurement, a $\tau$ of 5.7 $\mu$s was chosen to reduce decoherence of the NV center. As shown in Fig. \[fig:DEER\](a), the resulting spectrum exhibits five peaks, in agreement with P1 ESR. In the measurement, the MW intensity is adjusted to ensure the DEER pulse length performs a $\pi$ rotation of the P1 center spins. Figure \[fig:DEER\](b) shows Rabi oscillations of P1 centers measured by NV-ESR with different $T$ values (see the sequence in Fig. \[fig:DEER\](b)). To explain the results, we consider the following NV-ESR model which describes the spin dynamics of an ensemble of two-level systems. [@Stepanov2016] Using this model, the intensity of NV-ESR is given by, $$I_{NV-ESR} = \exp \bigg[\frac{-2\pi \mu_0 \mu_B^2 g_{NV} g_B T}{9\sqrt{3}\hbar} n \: \bigg\langle \sin^2 \frac{\theta}{2} \bigg \rangle_L \bigg]
\label{eq:deerIntViktor}$$ where $\mu_0$ is the vacuum permeability, $\mu_B$ is the Bohr magneton, $g_{NV}$ is the $g$-value of the NV center, $g_B$ is the g value of target spins, $T$ is the time for phase to accumulate after application of the DEER pulse, $\hbar$ is the reduced Planck constant, and $n$ is the concentration of target spins. The $\langle \sin^2 \frac{\theta}{2} \rangle_L$ term represents the effective population inversion of the DEER pulse given as, [@Salikhov1981] $$\begin{aligned}
\bigg\langle \sin^2 \frac{\theta}{2} \bigg \rangle_L = \nonumber
\int_{-\infty}^{+\infty}
\frac{\Omega^2}{(\xi -\omega)^2 +\Omega^2} \\
\times \sin^2\bigg(\sqrt{(\xi -\omega)^2 + \Omega^2 }\frac{t_p}{2}\bigg)
L(\xi; \Delta \omega)d\xi \label{eq:SinSqZeta}\end{aligned}$$ where $\Omega$ is the Rabi frequency of the target spins, $\omega$ is the frequency of MW2, $t_p$ is the applied pulse length, and $L(\xi; \Delta \omega)$ is an intrinsic ESR line of P1 spins where $\Delta \omega$ represents the linewidth. Therefore, Eq. \[eq:SinSqZeta\] includes the effects of the MW excitation and the ESR line on the NV-ESR signal. The P1 Rabi data were simulated with Eq. \[eq:deerIntViktor\] by fixing $T$ while allowing $n$ to vary. As shown in Fig. \[fig:DEER\](b), the simulations were found to be in good agreement with the experiments. We found that NV-ESR intensity depends on the value of $T$. As shown in Fig. \[fig:DEER\](b), NV-ESR with $T=2.0$ $\mu$s exhibits a high intensity contrast between $\pi$ and $\pi/2$ pulses.
![Dependence of DEER pulse length on NV-ESR linewidth. (a) NV-ESR spectra taken using various $\pi$-pulse lengths; $\pi$-pulse lengths are indicated in the legend. (b) The NV-ESR linewidth as a function of the pulse length. The red solid line is the result of a nonlinear least squares regression using Eq. \[eq:deerIntViktor\] and $\Delta \omega = 1.6$ (90 $\%$ confidence bounds of $(1.4, 1.8)$) MHz. Fitting was done with weights $1/\sigma^2$. The blue and green dashed lines show partial contributions. The blue dashed line is the MW excitation bandwidth, while the green dashed line shows $\Delta \omega$. The inset graph shows the spectrum taken using a $\pi$-pulse of 1 $\mu$s. The spectrum was normalized for the probability of the NV $\ket{0}$ state. The simulated spectrum based on a linewidth of $1.6$ MHz is shown in red. (c) Ramsey measurement using NV-ESR to measure $T_2^*$. Pulses were applied 2 $\mu$s before the end of the sequence ($T=2$ $\mu$s). The pulse sequence used for NV-ESR Ramsey is shown to the right. The spacing between the pulses ($t$) was varied in the NV-ESR Ramsey measurement (Sig.). A reference experiment was performed concurrently with the position of a single $\pi$-pulse being varied by $t$. NV center (P1 center) pulse times were 40 (64) and 24 (32) ns, for the $\pi$ and $\pi/2$ pulses respectively. The fit is shown in red.[]{data-label="Fig:DeerSpec"}](Fig5.eps)
We seek to determine the origin of the observed linewidth in Fig. \[fig:DEER\](a) by extracting the contribution from the MW excitation bandwidth. The contribution is studied by analyzing the NV-ESR linewidth as a function of the DEER pulse length ($t_p$). In the experiment, the MW power is adjusted to maintain a $\pi$-pulse for all pulse lengths. As shown in Fig. \[Fig:DeerSpec\](a), the spectrum narrows and the shape of the spectrum changes as the pulse length of the DEER pulse increases. In order to characterize the linewidth, we fit each spectrum to a sum of two Lorentzians with resonance positions for all orientations of P1 centers ([*i.e.*]{}, $m_I=0$ transitions for the \[111\] orientation and other three orientations at $943$ and $945$ MHz, respectively). The extracted full widths at half maximum (FWHM) are summarized in Fig. \[Fig:DeerSpec\](b) where the linewidths strongly depend on $t_p$ below a $t_p$ of 0.4 $\mu$s. To explain the dependence of the pulse length on the linewidth, we analyze contributions to the linewidth by fitting FWHM calculated using Eq. \[eq:deerIntViktor\] with the experimental FWHM where $\Delta \omega$ (Lorentzian linewidth) is a fit parameter. As shown in Fig. \[Fig:DeerSpec\](b), we found excellent agreement with the observed linewidths with $\Delta \omega = 1.6$ MHz with 90$\%$ confidence bounds of $(1.4,1.8)$ MHz. In the figure, we also show partial contributions of the MW excitation and $\Delta \omega$ where the contribution of the MW excitation bandwidth is obtained by numerical calculation of FWHM using $L(\xi;\Delta \omega) = \delta(\xi)$ in Eq. \[eq:deerIntViktor\]. This analysis verified that the MW excitation bandwidth is a major contribution of the NV-ESR linewidth when t$_p$ is shorter than $\sim$0.4 $\mu$s. As shown in the inset of Fig. \[Fig:DeerSpec\](b), the observed NV-ESR spectrum is well-explained by the simulation using Eq. \[eq:deerIntViktor\] with $\Delta \omega = 1.6$ MHz. Moreover, the obtained high spectral resolution NV-ESR spectrum allows clear identification of the $m_I = 0$ P1 ESR signals for the \[111\] orientation and the non-\[111\] orientations which are separated by only 2 MHz. This small splitting was not resolved in a previous experiment performed at a similar magnetic field. [@Abeywardana2016] The splitting is due to the contribution of the anisotropic hyperfine interaction comparable to the Zeeman energy at the low magnetic field which sets the ESR frequency from the non-\[111\] orientations to $945$ MHz at 33.4 mT while the \[111\] orientation remains at $943$ MHz. We next confirm the nature of the intrinsic linewidth $\Delta \omega = 1.6$ MHz by comparing with the spin dephasing time ($T_2^*$). $T_2^*$ relaxation time originates from an inhomogeneous distribution of ESR frequencies and represents the linewidth in many conventional ESR experiments. Given the $\pi$ and $\pi/2$ pulse lengths with $T = 2$ $\mu$s as shown in Fig. \[fig:DEER\](b), we perform a DEER Ramsey experiment to measure $T_2^*$. To confirm the observed signal, a concurrent experiment varying the position of the $\pi$-pulse (Ref.) was performed with the sequence shown in Fig. \[Fig:DeerSpec\](c). We observed exponential behavior from the Ramsey measurement, as shown in Fig. \[Fig:DeerSpec\](c). The observed signal was then analyzed by fitting the data with Eq. \[eq:deerIntViktor\] where $\langle \sin^2 \frac{\theta}{2} \rangle_L = - \exp(t/T_2^*)$. We observed a $T_2^*$ = 118 $\pm$ 34 ns from the analysis for NV1. The value of $T_2^*$ corresponds to a FWHM of $2.7 \pm 1.0$ MHz, a value in reasonable agreement with the $\Delta \omega$ extracted from frequency measurements. Furthermore, from the analysis of the NV-ESR intensity at 943 MHz, the detected magnetic dipole field ($B_{Dip}$) is $\approx$420 nT. [@Abeywardana2016] This strength of the magnetic field corresponds to an axially aligned single spin at a distance of $\sim$16 nm.
Moreover, we investigate NV-ESR spectroscopy with other single NV centers (NV2-5). As summarized in Fig. \[fig:SINV\], NV2-5 also exhibit a strong pulse length dependence similar to that observed for NV1. For NV2, $\Delta \omega$ of 0.9 (0.9, 1.0) MHz was observed in agreement with the measured $T_2^*$ of $240 \pm 125$ ns, as shown in Fig. \[fig:SINV\](a). The linewidth observed for NV3 ($\Delta \omega = 1.0$ MHz) was similar to that of NV2 (Fig. \[fig:SINV\](b)). The linewidth observed for NV4 and NV5 ($\Delta \omega = 0.3$ MHz) was similar in magnitude, but significantly smaller than NV1-3 (Fig. \[fig:SINV\](c) and (d)). For NV4, we measure the $m_I=0$ ESR transition and resolve a very narrow linewidth that allowed for clear resolution of the two peaks originated from the \[111\] and other orientations of P1 spins. In NV5, the NV-ESR linewidth with a pulse length of 2 $\mu$s was only 0.3 (0.2, 0.4) MHz. Overall, the results from NV1-5 provide clear examples of nanoscale ESR investigation of the inhomogeneity in ESR signals. This is shown by variation in ESR linewidths as measured by different NVs located within the same diamond crystal. Moreover, the observed linewidths of NV-ESR are much smaller than that in HF ESR (see Fig. \[fig:HF\_ESR\]). This is most likely because of the significant difference in the sample size between the two experiments. Conventional ESR obtains signal from all spins within the millimeter scale sample while the sample volume in NV-ESR is confined within several to a-few-tens of nanometers from the NV center. This significantly smaller size of the sample volume limits the number of detected P1 spins. In the present case, P1 ESR in the nanometer-scaled sample volume has significantly smaller inhomogeneity compared with conventional ESR. As shown by the previous conventional ESR investigations, there are two major contributions to the P1 ESR linewidth; hyperfine couplings to $^{13}$C nuclear spin baths and magnetic dipole couplings to P1 spin baths. [@Stepanov2016; @vanWyk97] When the P1 concentration is low, the ESR linewidth as narrow as $\sim$0.3 MHz is broadened by the hyperfine couplings to the $^{13}$C nuclear spin baths. On the other hand, when the P1 concentration is high, the linewidth is broader due to the coupling of the P1 spin baths and depends on the P1 concentration. Therefore the present result strongly suggests that the variation of observed P1 linewidths is due to inhomogeneity of densities and spatial configurations of P1 spin baths within the detected nanoscale volume. Furthermore, for NV 4 and 5, the contribution from the P1 spin baths is negligible on the linewidths ($\sim$0.3 MHz) while the hyperfine coupling to $^{13}$C spin baths is the major contribution.
Conclusion
==========
Within this article we investigated the nature of the NV-ESR linewidth by studying P1 ESR. We found that the spectral resolution depends strongly on the length of the DEER pulse. This was particularly evident when pulse lengths are shorter than 0.4 $\mu$s. Upon using long pulse lengths, the minimum resolved linewidth was found to be limited by inhomogeneous broadening of P1 ESR ($T_2^*$-limit). This linewidth was found to vary between NV centers, indicating spatial inhomogeneity of local magnetic fields surrounding each NV center. Since NV-ESR is useful for investigation of many spin systems with single spin sensitivity, the ability to perform high-resolution NV-ESR is critical. The present work provides important context into the improvement of NV-ESR spectral resolution. In particular, we demonstrated resolution of a small ESR splitting (2 MHz) by improving the spectral resolution and identified dominant coupling between P1 and surrounding electron and nuclear spins. Furthermore, the present technique will be applicable for various NV-ESR investigations including identification of multiple types of spins and study of spin dynamics.
Acknowledgements
================
This work was supported by the National Science Foundation (DMR-1508661 and CHE-1611134), the USC Anton B. Burg Foundation and the Searle scholars program (ST).
@ifundefined
[39]{}
Abragam, A.; Bleaney, B. *Electron paramagnetic resonance of transition ions*; Oxford University Press: Oxford, 1970 Schweiger, A.; Jeschke, G. *Principles of pulse electron paramagnetic resonance*; Oxford University Press: New York, 2001 Poole, C. *Electron spin resonance*; John Wiley and Sons, Inc.: New York, 1983 Weil, J. A.; Bolton, J. R. *Electron paramagnetic resonance*; John Wiley and Sons, Inc.: New Jersey, 2007 Feher, G. Electron spin resonance experiments on donors in silicon. I. Electronic structure of donors by the electron nuclear double resonance technique. *Phys. Rev.* **1959**, *114*, 1219–1244 Watkins, G. D. EPR and ENDOR studies of defects in semiconductors. *Semicond. Semimetals* **1998**, *51A*, 1–43 Loubser, J. H. N.; [van Wyk]{}, J. A. Electron spin resonance in the study of diamond. *Rep. Prog. Phys.* **1978**, *41*, 1201–1248 Hubbell, W. L.; Cafiso, D. S.; Altenbach, C. Identifying conformational changes with site-directed spin labeling. *Nat. Struct. Biol.* **2000**, *7*, 735–739 Misra, S. K.; Freed, J. H. Molecular Motions. In *Multifrequency electron paramagnetic resonance*; Misra, S. K., Ed.; Wiley-VCH: New York, 2011; Chapter 11, pp 497–542 Tao, L.; Stich, T. A.; Butterfield, C. N.; Romano, C. A.; Spiro, T. G.; Tebo, B. M.; Casey, W. H.; Britt, R. D. Mn(II) binding and subsequent oxidation by the multicopper oxidase mnxG investigated by electron paramagnetic resonance spectroscopy. *J. Am. Chem. Soc.* **2015**, *137*, 10563–10575 Niklas, J.; Poluektov, O. G. Charge transfer processes in OPV materials as revealed by EPR spectroscopy. *Adv. Energy Mater.* **2017**, *7*, 1602226 Shaakov, S.; Galili, T.; Stavitski, E.; Levanon, H.; Lukas, A.; Wasielewski, M. R. Using spin dynamics of covalently linked radical ion pairs to probe the impact of structural and energetic changes on charge recombination. *J. Am. Chem. Soc.* **2003**, *125*, 6563–6572 Britt, R. D.; Peloquin, J. M.; Campbell, K. A. Pulsed and parallel-polarization EPR characterization of the photosystem II oxygen-evolving complex. *Annu. Rev. Biophys. Biomol. Struct.* **2000**, *29*, 463–495 Gruber, A.; Dr[ä]{}benstedt, A.; Tietz, C.; Fleury, L.; Wrachtrup, J.; [von Borczyskowski]{}, C. Scanning confocal optical microscopy and magnetic resonance on single defect centers. *Science* **1997**, *276*, 2012–2014 Degen, C. L. Scanning magnetic field microscope with a diamond single-spin sensor. *Appl. Phys. Lett.* **2008**, *92*, 243111 Balasubramanian, G.; Chan, I. Y.; Kolesov, R.; Al-Hmoud, M.; Tisler, J.; Shin, C.; Kim, C.; Wojcik, A.; Hemmer, P. R.; Krueger, A. [et al.]{} Nanoscale imaging magnetometry with diamond spins under ambient conditions. *Nature* **2008**, *455*, 648–651 Maze, J. R.; Stanwix, P. L.; Hodges, J. S.; Hong, S.; Taylor, J. M.; Cappellaro, P.; Jiang, L.; Dutt, M. V. G.; Togan, E.; Zibrov, A. S. [et al.]{} Nanoscale magnetic sensing with an individual electronic spin in diamond. *Nature* **2008**, *455*, 644–647 Balasubramanian, G.; Neumann, P.; Twitchen, D.; Markham, M.; Kolesov, R.; Mizuochi, N.; Isoya, J.; Achard, J.; Beck, J.; Tissler, J. [et al.]{} Ultralong spin coherence time in isotopically engineered diamond. *Nat. Mater.* **2008**, *8*, 383–387 Takahashi, S.; Hanson, R.; van Tol, J.; Sherwin, M. S.; Awschalom, D. D. Quenching spin decoherence in diamond through spin bath polarization. *Phys. Rev. Lett.* **2008**, *101*, 047601 Wolf, T.; Neumann, P.; Nakamura, K.; Sumiya, H.; Ohshima, T.; Isoya, J.; Wrachtrup, J. Subpicotesla diamond magnetometry. *Phys. Rev. X* **2015**, *5*, 041001 Grinolds, M. S.; Hong, S.; Maletinsky, P.; Luan, L.; Lukin, M. D.; Walsworth, R. L.; Yacoby, A. Nanoscale magnetic imaging of a single electron spin under ambient conditions. *Nat. Phys.* **2013**, *9*, 215–219 Sushkov, A. O.; Lovchinsky, I.; Chisholm, N.; Walsworth, R. L.; Park, H.; Lukin, M. D. Magnetic resonance detection of individual proton spins using quantum reporters. *Phys. Rev. Lett.* **2014**, *113*, 197601 Abeywardana, C.; Stepanov, V.; Cho, F. H.; Takahashi, S. Electron spin resonance spectroscopy of small ensemble paramagnetic spins using a single nitrogen-vacancy center in diamond. *J. Appl. Phys.* **2016**, *120*, 123907 Shi, F.; Zhang, Q.; Naydenov, B.; Jelezko, F.; Du, J.; Reinhard, F.; Wrachtrup, J. Quantum logic readout and cooling of a single dark electron spin. *Phys. Rev. B* **2013**, *87*, 195414 Shi, F.; Zhang, Q.; Wang, P.; Sun, H.; Wang, J.; Rong, X.; Chen, M.; Ju, C.; Reinhard, F.; Chen, H. [et al.]{} Single-protein spin resonance spectroscopy under ambient conditions. *Science* **2015**, *347*, 1135–1138 Mamin, H. J.; Sherwood, M. H.; Rugar, D. Detecting external electron spins using nitrogen-vacancy centers. *Phys. Rev. B* **2012**, *86*, 195422, G.; [van der Sar]{}, T.; Blok, M.; Wang, Z.-H.; Dobrovitski, V.; Hanson, R. Controlling the quantum dynamics of a mesoscopic spin bath in diamond. *Sci. Rep.* **2012**, *2*, 1–5 Rosenfeld, E. L.; Pham, L. M.; Lukin, M. D.; Walsworth, R. L. Sensing coherent dynamics of electronic spin clusters in solids. *Phys. Rev. Lett.* **2018**, *120*, 243604 Laraoui, A.; Hodges, J. S.; Meriles, C. A. Nitrogen-vacancy-assisted magnetometry of paramagnetic centers in an individual diamond nanocrystal. *Nano Lett.* **2012**, *12*, 3477–3482 Smith, G. M.; Lesurf, J. C. G.; Mitchell, R. H.; Riedi, P. C. Quasi-optical cw mm-wave electron spin resonance spectrometer. *Rev. Sci. Instrum.* **1998**, *69*, 3924–3937 Cho, F. H.; Stepanov, V.; Takahashi, S. A high-frequency electron paramagnetic resonance spectrometer for multi-dimensional, multi-frequency, and multi-phase pulsed measurements. *Rev. Sci. Instrum.* **2014**, *85*, 075110 Cho, F. H.; Stepanov, V.; Abeywardana, C.; Takahashi, S. 230/115 GHz electron paramagnetic resonance/double electron-electron resonance spectroscopy. *Methods Enzymol.* **2015**, *563*, 95–118 Brown, R. H.; Twiss, R. Q. Correlation between photons in two coherent beams of light. *Nature* **1956**, *177*, 27–29 de Lange, G.; Wang, Z. H.; Ristè, D.; Dobrovitski, V. V.; Hanson, R. Universal dynamical decoupling of a single solid-state spin from a spin bath. *Science* **2010**, *330*, 60–63 Wang, Z.-H.; Takahashi, S. Spin decoherence and electron spin bath noise of a nitrogen-vacancy center in diamond. *Phys. Rev. B* **2013**, *87*, 115122 Stepanov, V.; Takahashi, S. Determination of nitrogen spin concentration in diamond using double electron-electron resonance. *Phys. Rev. B* **2016**, *94*, 024421 Salikhov, K. M.; Dzuba, S. A.; Raitsimring, A. M. The theory of electron spin-echo signal decay resulting from dipole-dipole interactions between paramagnetic centers in solids. *J. Magn. Reson.* **1981**, *42*, 255–276 van Wyk, J. A.; Reynhardt, E. C.; High, G. L.; Kiflawi, I. The dependences of ESR line widths and spin-spin relaxation times of single nitrogen defects on the concentration of nitrogen defects in diamond. *J. Phys. D: Appl. Phys.* **1997**, *30*, 1790–1793
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'It is well-known that if $R$ is a domain with finite character, each locally principal nonzero ideal of $R$ is invertible. We address the problem of understanding when the converse is true and survey some recent results.'
address: 'Dipartimento di Matematica, Università degli Studi Roma Tre, Largo S. L. Murialdo, 1, 00146 Roma, Italy '
author:
- Stefania Gabelli
title: Locally principal ideals and finite character
---
{#section .unnumbered}
Throughout all the paper, $R$ will be an integral domain and $K$ its field of fractions. To avoid trivialities, we will assume that $R\neq K$.
An $R$-submodule $M$ of $K$ is called *fractional* if $(R:M):=\{x\in K;\, xM{\subseteq}R\} \neq (0)$, equivalently, if there exists a nonzero $d\in R$ such that $dM{\subseteq}R$. If there is such a $d$, then $dM:=J$ is an ideal of $R$ and $M=d^{-1}J$; thus $M$ is also called a *fractional ideal*.
If $I$ is a fractional ideal of $R$, we call $I$ simply an *ideal* and if $I{\subseteq}R$ we say that $I$ is an *integral ideal*.
An *overring* of $R$ is a domain $D$ such that $R{\subseteq}D{\subseteq}K$. It is easy to see that $D$ is a fractional overring if and only if $D$ is the endomorphism ring of a nonzero integral ideal $I$ of $R$, that is, $D=E(I):=(I:I):=\{x\in K;\, xI{\subseteq}I\}$. In fact, if $I{\subseteq}R$ is a nonzero ideal, $E(I):=(I:I)$ is an overring and $(0) \neq I{\subseteq}(R:E(I))$. Conversely, if $D$ is a fractional overring and $I:=dD{\subseteq}R$, $d\neq 0$, we have that $I$ is an ideal and $D=E(I):=(I:I)$.
An ideal $I$ of $R$ is called *locally principal* if $I_M:=IR_M$ is principal, for each maximal ideal $M$ of $R$. Note that, if $I$ is a locally principal nonzero ideal, its endomorphism ring $E(I):=(I:I)$ is equal to $R$. In fact $(I_M:I_M)=R_M$, for each $M\in \operatorname{Max}(R)$. Hence, $$E(I)=(I:I)=\bigcap_{M\in \operatorname{Max}(R)} (I:I)_M=\bigcap_{M\in \operatorname{Max}(R)} (I_M:I_M)= \bigcap_{M\in \operatorname{Max}(R)} R_M =R.$$
By checking it locally, it is also easy to see that a locally principal nonzero ideal is *cancellative*, that is, whenever $IB = IC$ for ideals $B$ and $C$ of $R$, then $B = C$ [@gilmer Section 6]. As proved by D. D. Anderson and M. Roitman, the converse is also true.
[@AR Theorem] A nonzero ideal $I$ of $R$ is cancellative if and only if it is locally principal.
The set of nonzero ideals of $R$ is a multiplicative semigroup, with unity $R$. A nonzero ideal $I$ is *invertible* if $IJ=R$, for some ideal $J$. If $I$ is invertible, its inverse is $(R:I):=\{x\in K; \,xI{\subseteq}R\}$ [@gilmer Section 7]. Clearly an invertible ideal is cancellative, hence locally principal.
For a converse, we need that $I$ be finitely generated. In fact, if $IJ=R$ and $1=a_1b_1+ \dots + a_nb_n$, $a_i\in I, b_i\in J$, the ideal $I$ is generated by $a_1, \dots, a_n$. For a discussion on the minimal number of generators of an invertible ideal one can see [@OR].
[@gilmer Corollary 7.5] \[inv\] A nonzero ideal $I$ of $R$ is invertible if and only if it is finitely generated and locally principal.
Thus, a nonzero finitely generated ideal is invertible if and only if is locally principal, if and only if is cancellative. In particular, locally principal nonzero ideals of Noetherian domains are invertible.
However, a locally principal ideal need not be finitely generated. In fact there are examples of domains such that $R_M$ is a one-dimensional discrete valuation domain, for each $M\in \operatorname{Max}(R)$ (these domains are called *almost Dedekind domains*), yet some of the maximal ideals are not finitely generated [@gilmer Section 36].
Thus one can ask:
\[Q1\] When is a locally principal nonzero ideal invertible (equivalently finitely generated)?
A first answer is the following.
[@AZ Theorem 3] \[prop1\] Let $I$ be a locally principal nonzero ideal of $R$. Then $I$ is invertible if and only if $(R:I)_M=(R_M:I_M)$, for each $M\in \operatorname{Max}(R)$.
If $I$ is invertible, it is finitely generated. Hence by flatness $(R:I)_M=(R_M:I_M)$. Conversely, if $I$ is locally principal and $(R:I)_M=(R_M:I_M)$, we have: $$I(R:I)=\bigcap_{M\in \operatorname{Max}(R)}I(R:I)R_M= \bigcap_{M\in \operatorname{Max}(R)}I_M(R_M:I_M)=\bigcap_{M\in \operatorname{Max}(R)}R_M=R.$$
Following D.D. Anderson and M. Zafrullah, for short we call $R$ an *LPI-domain* (or say that $R$ has the *LPI-property*) if each locally principal nonzero ideal of $R$ is invertible [@AZ].
If $(R:I)_M=(R_M:I_M)$, for each nonzero ideal $I$ and each maximal ideal $M$ of $R$, then $R$ is an LPI-domain.
If $\{R_{\alpha}\}$ is a family of overrings of $R$ such that $R= \bigcap_{\alpha}R_{\alpha}$, we say that the intersection $\bigcap_{\alpha}R_{\alpha}$ has *finite character* if each nonzero element of $R$ is not invertible in at most finitely many $R_{\alpha}$. If the intersection $R=\bigcap_{M\in \operatorname{Max}(R)}R_M$ has finite character, we say that $R$ has finite character.
An *$h$-local* domain is a domain with finite character such that each nonzero prime ideal is contained in a unique maximal ideal. If $R$ is $h$-local, we have $(X:Y)_M=(X_M:Y_M)$, for each couple of $R$-submodules $X$, $Y$ of $K$ such that $X{\subseteq}Y$ and each $M\in \operatorname{Max}(R)$ [@BS2 Lemma 2.3]. Thus by Proposition \[prop1\] we get:
[@O3 Lemma 3.9] An $h$-local domain is an LPI-domain.
Another sufficient condition for the equality $(R:I)_M=(R_M:I_M)$ is that $I$ be $v$-finite.
Recall that, if $I$ is a nonzero fractional ideal of a domain $R$, the *divisorial closure* of $I$ is the star operation defined by $I_v:=(R:(R:I))$. The ideal $I$ is called *divisorial* if $I=I_v$ and a divisorial ideal is called *$v$-finite* if $I=J_v$ for some finitely generated ideal $J$. Invertible ideals are divisorial. An introduction to star operations and divisorial closure is [@gilmer Section 44].
Now, observing that $(R:J_v)=(R:J)$, if $I=J_v$, with $J$ finitely generated, we have $$(R:I)_M {\subseteq}(R_M:I_M) {\subseteq}(R_M:J_M)=(R:J)_M=(R:I)_M.$$ Hence $(R:I)_M = (R_M:I_M)$.
[@A Theorem 2.1] \[And1\] Let $I$ be a nonzero locally principal ideal of $R$. Then $I$ is invertible if and only if $I$ is divisorial $v$-finite.
Any invertible ideal is divisorial an finitely generated, hence $v$-finite. Conversely, if $I=J_v$, with $J$ finitely generated, as above we have $(R:I)_M = (R_M:I_M)$. Thus, if $I$ is also locally principal, $I$ is invertible by Proposition \[prop1\].
A class of domains where each divisorial ideal is $v$-finite is the class of Mori domains. $R$ is called a *Mori domain* if it satisfies the ascending chain condition on divisorial integral ideals. Clearly Noetherian domains and Krull domains are Mori. For the main properties of Mori domains the reader is referred to [@Bar].
From Proposition \[And1\], we immediately see that:
A Mori domain is an LPI-domain.
Since a local domain is trivially an LPI-domain, each domain is intersection of LPI-domains. Thus, another way of approaching Problem \[Q1\] is to investigate when an intersection of LPI-domains with the same quotient field is an LPI-domain. It turns out that this is true if the intersection has finite character.
[@AZ Theorem 4] Let $\{R_{\alpha}\}$ be a family of overrings of $R$ such that $R=\bigcap R_{\alpha}$. If each $R_{\alpha}$ is an LPI-domain and the intersection has finite character, then $R$ is an LPI-domain.
Let $I$ be a locally principal nonzero ideal and let $R_i:=R_{{\alpha}_i}$, $i=1, \dots, n$, be the overrings of the family containing $I$. Since $R_i$ is an LPI-domain, $IR_i$ is finitely generated, for each $i=1, \dots, n$. Thus, by the finite character, we can find a finitely generated ideal $J{\subseteq}I$ such that $JR_{\alpha}=IR_{\alpha}$, for each ${\alpha}$. By the properties of star operations, this implies that $I=J_v$. Thus $I$ is invertible by Proposition \[And1\].
Now, let us introduce some more notation. The $t$-*operation* is the star operation of finite type associated to the divisorial closure and it is therefore defined by setting $$I_t:=\bigcup\{J_v; \mbox{ $J$ finitely generated and $J{\subseteq}I$}\},$$ for each nonzero ideal $I$ of $R$. If $I=I_t$, $I$ is called a *$t$-ideal*. Note that $I_t{\subseteq}I_v$; thus each divisorial ideal is a $t$-ideal and $I_t=I_v$ if $I$ is finitely generated.
By Zorn’s Lemma each $t$-ideal is contained in a $t$-maximal ideal, which is prime, and we have $$R=\bigcap \{R_{M};\; M \mbox{ $t$-maximal ideal}\}.$$ We say that $R$ has *$t$-finite character* if this intersection has finite character.
\[FC\] [@AZ Corollary 1], [@PT Proposition 1.3] A domain with finite character or $t$-finite character is an LPI-domain.
Since Mori domains have $t$-finite character [@Bar Theorem 3.3], from Corollary \[FC\] we recover that, as seen above, $h$-local domains and Mori domains have the LPI-property.
Observe that a locally principal nonzero ideal that is contained just in finitely many maximal ideals need not be invertible. In fact, as already observed, a maximal ideal of an almost Dedekind domain is always locally principal but need not be finitely generated. A sufficient condition for a locally principal nonzero ideal $I$ to be finitely generated is that $I$ contains an element belonging to only finitely many maximal ideals [@gilmer Lemma 37.3]. In addition, if $I$ contains such an element, $I$ can be generated by two elements [@GH Corollary 1]. This is a direct way of proving that the finite character implies the LPI-property.
However, the LPI-property does not imply neither the finite character nor the $t$-finite character. In fact any Noetherian domain is an LPI-domain, but need not have finite character. On the other hand, there are examples of local domains (hence, LPI-domains) with infinitely many $t$-maximal ideals with nonzero intersection (hence, without $t$-finite character)[@FPT Example 1.13]. Thus we ask:
\[Qt\] When the LPI-property implies the $t$-finite character?
\[Q2\] When the LPI-property implies the finite character?
To study Problem \[Qt\], it is natural to generalize the LPI-property by means of the $t$-operation.
Recall that a nonzero ideal $I$ of $R$ is *$t$-invertible* if there is an ideal $J$ such that $(IJ)_t=R$. A digression on $t$-invertibility can be found in [@Z].
To our purpose, we recall that $t$-invertible ideals have a characterization similar to the one given in Proposition \[inv\] for invertible ideals. That is, a $t$-ideal $I$ is $t$-invertible if and only if is $v$-finite (i.e., $I=J_v$ with $J$ finitely generated) and is $t$-locally principal (i.e., $IR_M$ is principal, for each $t$-maximal ideal $M$) [@Z Theorems 1.1 and 2.2].
We say that $R$ has the *$t$LPI-property* if each $t$-ideal which is $t$-locally principal is $t$-invertible. The $t$LPI-property implies the LPI-property [@FPT Proposition 2.2].
A *Prüfer domain* is a domain whose localizations at prime ideals are valuation domains, equivalently, a domain with the property that finitely generated nonzero ideals are invertible [@gilmer Chapter IV]. The $t$-analog of Prüfer domains are the so called *Prüfer $v$-multiplication domains*, for short P$v$MDs, defined by the property that each localization at a prime $t$-ideal is a valuation domain, equivalently, by the property that each finitely generated nonzero ideal is $t$-invertible [@MZ].
Since almost Dedekind domains are Prüfer, not all Prüfer domains are LPI-domains. S. Bazzoni conjectured that Prüfer domains with the LPI-property were of finite character [@B1 p. 630]. This conjecture was then proved by W. Holland, J. Martinez, W. McGovern and M. Tesemma in [@HM]. (A simplified proof is in [@McG]). F. Halter-Koch gave independently another proof, in the more general context of ideal systems. In particular, he proved the analog of Bazzoni’s conjecture for P$v$MDs, that is: a P$v$MD with the $t$LPI-property has $t$-finite character [@HK Theorem 6.11]. With different methods, the same result was obtained by M. Zafrullah [@Za Proposition 5].
Other contributions to Problem \[Qt\] were given by C.A. Finocchiaro, G. Picozza and F. Tartarone. In [@FPT], they extended several results of Halter-Koch and Zafrullah in the context of star operations of finite type and proved a suitable generalization of Bazzoni’s conjecture for $v$-coherent domains. Recall that $R$ is called *$v$-coherent* if, whenever $I$ is finitely generated, $(R:I)$ is $v$-finite. The class of $v$-coherent domains properly include coherent domains, Mori domains and P$v$MDs [@N Example 1], [@GaH].
Since we do not know any example of $t$LPI-domain without $t$-finite character, we ask:
[@Za p. 657]Ê\[Qt2\] Assume that $R$ is a $t$LPI-domain. Is it true that $R$ has $t$-finite character?
Besides that for P$v$MDs, we know that the answer is positive when each prime $t$-ideal of $R$ is contained in a unique $t$-maximal ideal [@GP1 Proposition 5.7].
Now we go back to the LPI-property. The following theorem gives useful characterizations of domains with finite character and of LPI-domains.
\[teoFPT\]
1. [@FPT Proposition 1.6], [@ZD Corollary 4] $R$ has finite character if and only if any family of pairwise comaximal finitely generated ideals with nonzero intersection is finite.
2. Ê[@Za Theorem 1] If $R$ is an LPI-domain, any family of pairwise comaximal invertible ideals with nonzero intersection is finite.
By means of this theorem, to answer Problem \[Q2\] one can try the following strategy. Assume, by way of contradiction, that $R$ does not have finite character and consider an infinite family of maximal ideals, or an infinite family of pairwise comaximal finitely generated ideals, with nonzero intersection. When, from one of these families, it is possible to recover an infinite family of pairwise comaximal invertible ideals with nonzero intersection, then $R$ cannot be an LPI-domain.
Since finitely generated nonzero ideals of Prüfer domains are invertible, in this way immediately we get:
\[Bazzoni’s Conjecture\] A Prüfer LPI-domain has finite character.
To go further, we enlarge the class of Prüfer domains to finitely stable domains.
Recall that an ideal $I$ of $R$ is called *stable* if $I$ is invertible in its endomorphism ring $E(I):=(I:I)$. Invertible ideals are stable. The domain $R$ is said to be (*finitely*) *stable* if each nonzero (finitely generated) ideal $I$ is stable. Stability was thoroughly investigated by B. Olberding [@O3; @O1; @O2]. In particular, he proved that a domain $R$ is stable if and only if it has finite character and $R_M$ is stable for each $M\in \operatorname{Max}(R)$ [@O2 Theorem 3.3].
Of course stability and finite stability coincide for Noetherian domains; but in general these two notions are distinct. In fact, an integrally closed finitely stable domain is precisely a Prüfer domain. (To see this, just recall that $R$ is integrally closed if and only if $R=(I:I)$ for each finitely generated ideal $I$.) On the other hand, a valuation domain $V$ is stable if and only if $PV_P$ is principal, for each nonzero prime ideal $P$ [@O3 Proposition 4.1].
This shows that a finitely stable domain need not have neither the LPI-property nor finite character. Also observe that, for finitely stable domains, finite character and $t$-finite character coincide. In fact, when $R$ is finitely stable, each maximal ideal is a $t$-ideal; this follows from [@tlink Lemma 2.1 and Theorem 2.4] since a finitely stable domain has Prüfer integral closure [@Rush Proposition 2.1].
Thus one is lead to investigate whether, as in the Prüfer case, a finitely stable LPI-domain has finite character.
[@B4 Question 4.6] \[LPI\] Assume that $R$ is a finitely stable LPI-domain. Is it true that $R$ has finite character?
By using Theorem \[teoFPT\], we are able to answer this question in positive under some additional hypotheses (see also [@G]). Observe that when $R$ is Mori, the answer is trivially affirmative. In fact Mori domains are LPI-domains (Proposition 7) and Mori finitely stable domains have always finite character, because Mori domains have $t$-finite character [@Bar Theorem 3.3] and, as said before, the maximal ideals of a finitely stable domain are $t$-ideals.
\[LPI-FC\] Let $R$ be a finitely stable LPI-domain verifying one of the following conditions:
- Each nonzero finitely generated ideal $I$ is principal in $E(I)$.
- Each maximal ideal of $R$ is stable.
- [@O3 Lemma 3.9] Each nonzero prime ideal of $R$ is contained in a unique maximal ideal (e.g., $R$ is one-dimensional).
- Each fractional overring of $R$ is an LPI-domain.
Then $R$ has finite character.
\(a) Let $\{I_{\alpha}\}$ be a family of pairwise comaximal finitely generated ideals of $R$ with nonzero intersection. By hypothesis, for each fixed ${\alpha}$, $I_{\alpha}=x_{\alpha}E(I_{\alpha})$, $x_{\alpha}\in I_{\alpha}$, and so $I_{\alpha}^2=x_{\alpha}I_{\alpha}$. Hence $I_{\alpha}^2{\subseteq}x_{\alpha}R{\subseteq}I_{\alpha}$. It follows that the principal ideals $x_{\alpha}R$ are pairwise comaximal. If $y\in R$ is a nonzero element contained in each $I_{\alpha}$, then $y^2\in I_{\alpha}^2{\subseteq}x_{\alpha}R$, for each ${\alpha}$. By the LPI-property of $R$, the set of indexes ${\alpha}$ must be finite (Theorem \[teoFPT\](2)). Hence $R$ has finite character (Theorem \[teoFPT\](1)).
\(b) Let $M$ be a maximal ideal of $R$. Since $M$ is stable, $MR_M=x(MR_M:MR_M)$ for some $x\in M$ [@O2 Lemma 3.1]. Then $M^2R_M{\subseteq}xMR_M{\subseteq}xR_M$. It follows that the ideal $I:=xR_M\cap R$, containing $M^2$, is not contained in any maximal ideal of $R$ different from $M$. Since $IR_M=xR_M$, $I$ is locally principal and so it is invertible by the LPI-property.
Now, suppose that $y$ is a nonzero element of $R$ contained in infinitely many maximal ideals $M_{\alpha}$. For each ${\alpha}$, consider the ideal $I_{\alpha}$ constructed in the preceding paragraph. Then $y^2\in M_{\alpha}^2{\subseteq}I_{\alpha}$, for each ${\alpha}$. Since $\{I_{\alpha}\}$ is an infinite family of pairwise comaximal invertible ideals, this contradicts Theorem \[teoFPT\](2).
\(c) Let $x$ be a nonzero element of $R$ contained in infinitely many maximal ideals $M_{\alpha}$ and consider the ideals $I_{\alpha}:=xR_{M_{\alpha}}\cap R$. By hypothesis, for a fixed ${\alpha}$ and maximal ideal $N$, if $N\neq M_{\alpha}$, there is no nonzero prime ideal contained in the intersection $N\cap M_{\alpha}$. Thus $M_{\alpha}$ is the unique maximal ideal containing $I_{\alpha}$. Since $I_{\alpha}R_{M_{\alpha}}=xR_{M_{\alpha}}$, $I_{\alpha}$ is locally principal and so it is invertible by the LPI-property. It follows that $\{I_{\alpha}\}$ is an infinite family of pairwise comaximal invertible ideals containing $x$, in contradiction with Theorem \[teoFPT\](2).
\(d) Assume that $R$ does not have finite character. Then, by Theorem \[teoFPT\](1), there exists a nonzero element $x\in R$ which is contained in infinitely many pairwise comaximal finitely generated ideals $I_{\alpha}$. Set $E_{\alpha}:=(I_{\alpha}:I_{\alpha})$ and consider the $R$-module $E:=\sum_{\alpha}E_{\alpha}$. Since $E$ is contained in the integral closure of $R$ and $R$ is finitely stable, $E$ is an overring of $R$ [@Rush Proposition 2.1]. In addition, since $x\in I_{\alpha}$ for each ${\alpha}$, $xE{\subseteq}\sum_{\alpha}I_{\alpha}{\subseteq}R$ and so $E$ is a fractional overring of $R$.
We claim that $\{I_{\alpha}E\}$ is an infinite family of pairwise comaximal invertible ideals of $E$ containing $x$. Since, by hypothesis, $E$ has the LPI-property, this contradicts Theorem \[teoFPT\](2).
First of all, since $R$ is finitely stable, each ideal $I_{\alpha}$ is stable, so that $I_{\alpha}E$ is invertible in $E$. Then, since $E$ is integral over $R$, each maximal ideal of $R$ is contained in a maximal ideal of $E$. It follows that $I_{\alpha}E\neq E$ for each ${\alpha}$. In addition, for ${\alpha}\neq {\beta}$ the ideals $I_{\alpha}E$ and $I_{\beta}E$ are comaximal, since the contraction of a prime ideal of $E$ is a prime ideal of $R$.
A particular case of Theorem \[LPI-FC\](b) is when each maximal ideal of $R$ is invertible. In this case, the conclusion follows directly from Theorem \[teoFPT\](2). We remark that a domain whose prime ideals are all stable is not necessarily stable, even if it is finitely stable [@O1 Section 3].
By using the techniques developed in [@GP1], it is possible to show that Theorem \[LPI-FC\] holds in the more general context of star operations spectral and of finite type (see for example [@GP1 Proposition 5.5] and the proof of [@GP1 Theorem 5.2]).
S. Bazzoni noted that the LPI-property can be strengthened by the condition that each nonzero ideal $I$ of $R$ that is locally stable (i.e., $IR_M$ is stable, for each maximal ideal $M$) is stable. She called this condition the *local stability property*. B. Olberding proved that a domain with finite character has the local stability property [@O3 Lemma 4.3] and S. Bazzoni proved the converse [@B4 Theorem 4.5] for finitely stable domains. We can recover her result from Theorem \[LPI-FC\] in the following way.
[@B4 Lemma 3.2] \[LSP\] A domain with the local stability property is an LPI-domain.
Assume that $R$ has the local stability property. If $I$ is a nonzero locally principal ideal of $R$, then $I$ is locally stable. Hence $I$ is stable. In particular $I$ is finitely generated in $E(I)$ and so $E(I)_M=(I:I)_M=(I_M:I_M)=R_M$, for each maximal ideal $M$. It follows that $E(I)=R$. Hence $I$ is invertible.
\[LSover\] Let $R$ be a finitely stable domain. If $R$ has the local stability property, each fractional overring of $R$ has the local stability property.
Let $E$ be a fractional overring of $R$. Given a maximal ideal $M$ of $R$, $R_M$ is finitely stable and so $E_M$ is semilocal (i.e., it has finitely many maximal ideals); in fact any overring of a finitely stable domain is finitely stable and any overring of a local finitely stable domain is semilocal [@O2 Lemma 2.4 and Corollary 2.5]. It follows that if $I$ is a locally stable ideal of $E$, then $IE_M$ is stable for each $M\in \operatorname{Max}(R)$. Hence, by the local stability property of $R$, the ideal $I=IE$ is stable.
\[corFC\] Let $R$ be a finitely stable domain. The following conditions are equivalent:
- $R$ has finite character;
- $R$ has the local stability property;
- Each fractional overring of $R$ has the local stability property;
- Each fractional overring of $R$ is an LPI-domain.
\(i) ${\Rightarrow}$ (ii) by [@O3 Lemma 4.3]. (ii) ${\Rightarrow}$ (iii) by Proposition \[LSover\]. (iii) ${\Rightarrow}$ (iv) by Lemma \[LSP\]. (iv) ${\Rightarrow}$ (i) by Theorem \[LPI-FC\](d).
[@B4 Theorem 4.5] A finitely stable domain with the local stability property has finite character.
By Corollary \[corFC\], we see that Problem \[LPI\] can be reformulated in the following way:
\[QuLSI\]ÊIs it true that every fractional overring of a finitely stable LPI-domain is an LPI-domain?
Note that the local stability property and the LPI-property coincide for Prüfer domains [@B4 Proposition 3.3] and that the answer to Problem \[QuLSI\] (equivalently, to Problem \[LPI\]) is positive if and only if these two properties are more generally equivalent for finitely stable domains [@B4 Question 6].
[9]{}
D.D. Anderson, *Globalization of some local properties in Krull domains*, Proc. A.M.S., [**85**]{} (1982), 141–145.
D.D. Anderson and M. Roitman, *A characterization of cancellation ideals*, Proc. A.M.S [**125**]{} (1997), 2853–2854.
D.D. Anderson and M. Zafrullah, *Integral domains in which nonzero locally principal ideals are invertible*, Comm. Algebra [**39**]{} (2011), 933–941.
V. Barucci, *Mori domains*, Non-Noetherian commutative ring theory, 57–73, Math. Appl., 520, Kluwer Acad. Publ., Dordrecht, 2000.
S. Bazzoni, *Class semigroups of Prüfer domains*, J. Algebra [**184**]{} (1996), 613–631.
S. Bazzoni, *Finite character of finitely stable domains*, J. Pure Appl. Algebra [**215**]{} (2011), 1127–1132.
S. Bazzoni and L. Salce, *Warfield domains*, J. Algebra [**185**]{} (1996), 836–868.
T. Dumitrescu and M. Zafrullah, *Characterizing domains of finite $\ast$-character*, J. Pure Appl. Algebra **214** (2010), 2087–2091
D.E. Dobbs, E.G. Houston, T.G. Lucas, M. Roitman and M. Zafrullah, *On $t$-linked overrings*, Comm. Algebra **20** (1992), 1463–1488.
C.A. Finocchiaro, G. Picozza and F. Tartarone *Star-Invertibility and $t$-finite character in Integral Domains*, J. Algebra Appl. **10** (2011), 755–769.
S. Gabelli, *Finite character, local stability property and local invertibility property*, Actes des rencontres du CIRM (Troisième rencontre internationale autour des polynomes à valeurs entières et problèmes d’algèbre commutative, Marsiglia, 2010), **2** (2010), 119–122.
S. Gabelli and E. Houston, *Coherent-like conditions in pullbacks*, Michigan Math. J., [**44**]{} (1997), 99–123.
S. Gabelli and G. Picozza *Stability and regularity with respect to star operations*, Comm. Algebra [**40**]{} (2012), 3558–3582.
R. Gilmer, *Multiplicative Ideal Theory*, Dekker, New York, 1972.
R. Gilmer and W. Heinzer, *On the number of generators of an invertible ideals*, J. Algebra [**14**]{} (1970), 139–151.
F. Halter-Koch, *Clifford semigroups of ideals in monoids and domains*, Forum Math. [**21**]{} (2009), 1001–1020.
W.C. Holland, J. Martinez, W.Wm. McGovern and M. Tesemma, *Bazzoni’s Conjecture*, J. Algebra [**320**]{} (2008), 1764–1768.
W.Wm. McGovern, *Prüfer domains with Clifford Class semigroup*, J. Commut. Algebra **3** (2011), 551–559.
ÊJ. Mott and M. Zafrullah, *On Prüfer $v$-multiplication domains*, Manuscripta Math. [**35**]{} (1981), 1–26.
D. Nour el Abidine, *Sur un théorème de Nagata*, Comm. Algebra [**20**]{} (1992), 2127–2138.
B. Olberding, *Globalizing local properties of Prüfer domains*, J. Algebra **205** (1998), 480–504.
B. Olberding, *On the classification of stable domains*, J. Algebra **243** (2001), 177–197.
B. Olberding, *On the structure of stable domains*, Comm. Algebra **30** (2002), 877–895.
B. Olberding and M. Roitman, *The minimal number of generators of an invertible ideal*, Multiplicative ideal theory in commutative algebra, 349–367, Springer, New York, 2006.
G. Picozza and F. Tartarone, *Flat ideals and stability in integral domains*, J. Algebra **324** (2010),1790–1802.
D. E. Rush, *Two-generated ideals and representations of abelian groups over valuation rings*, J. Algebra [**177**]{} (1995), 77–101.
M. Zafrullah, [*Putting $t$-invertibility to use*]{}, Non-Noetherian Commutative Ring Theory, 429–457, Math. Appl., 520, Kluwer Acad. Publ., Dordrecht, 2000.
M. Zafrullah, *$t$-Invertibility and Bazzoni-like statements*, J. Pure Appl. Algebra [**214**]{} (2010), 654–657.
| {
"pile_set_name": "ArXiv"
} |
---
bibliography:
- 'main.bib'
title: 'Long Term Temporal Context for Per-Camera Object Detection'
---
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We give a brief survey of recent results on word maps on simple groups and polynomial maps on simple associative and Lie algebras. Our focus is on parallelism between these theories, allowing one to state many new open problems and giving new ways for solving older ones.'
address:
- 'Kanel-Belov: Department of Mathematics, Bar-Ilan University, 5290002 Ramat Gan, ISRAEL'
- 'Kunyavskii: Department of Mathematics, Bar-Ilan University, 5290002 Ramat Gan, ISRAEL'
- 'Plotkin: Department of Mathematics, Bar-Ilan University, 5290002 Ramat Gan, ISRAEL'
author:
- 'Alexey Kanel-Belov, Boris Kunyavskiĭ, Eugene Plotkin'
title: |
Word equations in simple groups and\
polynomial equations in simple algebras
---
\[remarks\][Remark]{}
[[F]{}]{} [[Q]{}]{}
Å[A]{}
Keywords: word map; simple group; engel words; dominant map; group width; polynomial map.
Introduction {#intro}
============
In this paper we discuss word maps $$\label{wordmap}
w\colon G^d\to G,$$ induced on any group $G$ by a group word $w=w(x_1,\dots,x_d)$ in $x_1,x_1^{-1},\dots ,x_d,x_d^{-1}$ (= an element of the free group $\cF_d$). We also consider polynomial maps $$\label{polmap}
P\colon A^d\to A,$$ induced on any associative (or Lie) algebra over a field $k$ (or a ring $R$) by an associative (or Lie) polynomial $P=P(X_1,\dots,X_d)$ in $d$ variables (= an element of the free associative (or Lie) algebra).
Both and are evaluation maps: one substitutes $d$-tuples of elements of the group $G$ (algebra $A$) instead of the variables and computes the value by performing all group (algebra) operations. We are interested in surjectivity of maps (\[wordmap\]) and (\[polmap\]), or, more generally, in description of their images. In lowbrow terms, we are interested in solvability of equations of the form
$$\label{word}
w(x_1,\dots ,x_d)=g,$$
or $$\label{pol}
P(X_1,\dots ,X_d)=M,$$
for every right-hand side, or for some “typical” right-hand side, or whether every element of the group (algebra) admits a representation as a product (sum) of finite number values of the word (polynomial) map, etc.
This setting is a particular case of the following one. Let $\Theta$ be a variety of algebras, $H$ be an algebra in $\Theta$, $W(X)$ be a free finitely generated algebra in $\Theta$ with generators $x_1,
\dots, x_d$. Fix $w\in W(X)$ and consider the word map $w\colon H^d
\to H.$ Varying $\Theta$ and $H\in\Theta$, we arrive at the problem of solvability of equations in different varieties over different algebras. In this note we restrict ourselves to the varieties of groups, associative algebras, and Lie algebras.
Whereas in the group case the theory has been intensely developing and led to several spectacular results, including answering some old-standing questions, on the ring-theoretic side much less is known, though some new approaches to not less old questions have been recently found. Our main goal in this survey, which does not pretend to be comprehensive, is to emphasize parallels between the two theories. We hope that this may bring cross-fertilization effects in near future. With an eye towards such a unification, we put here more questions than answers.
The interested reader is referred to the monograph [@Se] and surveys [@Sh1], [@Ni], [@BGK] in what concerns the group case. Some references on the algebra case can be found in [@KBMR], [@BGKP]. We leave aside extremely interesting questions on the number of solutions of equations , (or, in other words, on the structure of the fibres of maps , ). See [@BGK] for an overview of some results in this direction.
Value sets
==========
Given a word map , it is important to distinguish between the following objects:
- the value set in strict sense: $w(G)=\{g\in G: \exists
(g_1,\dots ,g_d) \quad w(g_1,\dots ,g_d)=g\}$;
- the symmetrized value set $w(G)^{\pm}$ consisting of the elements of $w(G)$ and their inverses;
- the verbal subgroup $\left<w(G)\right>$ of $G$ generated by $w(G)$.
Respectively, given a polynomial map of $k$-algebras, we distinguish between the value set $P(A)=\{a\in A: \exists
(a_1,\dots ,a_d) \quad P(a_1,\dots ,a_d)=a\}$ and the vector space $\left<P(A)\right>$ spanned by $P(A)$ over $k$.
It is usually much easier to describe $\left<w(G)\right>$ and $\left<P(A)\right>$ than the actual value sets.
Surjectivity
------------
We start with the group case and ask whether word map is surjective. More precisely,
- given a class of groups $\mathcal G$, we want to describe words $w$ for which map is surjective;
- given a class of words $\mathcal W$, we want to describe groups $G$ for which map is surjective.
In each of these set-ups arising problems range from very easy to extremely difficult, depending on the choice of the class. Here are some examples. Let us start with the first approach and take $\mathcal G$ to be the variety of all groups. Then the needed description is given by [@Se Lemma 3.1.1], where such words are called universal: they are all of the form $w=x^{e_1}\dots
x_d^{e_d}w'$ where $w'$ is a product of commutators and gcd$(e_1,\dots,e_d)=1$. Here is a parallel question for associative algebras:
What are polynomials $P$ such that the map $P\colon A^d\to A$ is surjective for all associative algebras $A$?
When we restrict the class $\mathcal G$, we arrive at more interesting and difficult questions. Answers heavily depend on this choice. For instance, a theorem of Rhemtulla (see [@Se Theorem 3.1.2]) says that if $\mathcal G$ consists of the free groups (and maybe also of some free products adjoined), then the behaviour of [*any*]{} non-universal word is very far from surjectivity: in the terminology explained below, any such word is of infinite width for any group of $\mathcal G$. That is why in this survey we prefer to stay away from equations in free (and close to free) groups; see, e.g., [@CRK], particularly the introduction, for a comprehensive bibliographical survey of vast literature on this fascinating theory.
We mainly focus on another extreme case of [*simple*]{} groups which can also be viewed as a building block for some more general theory. Here is our first question.
\[sur-adj\] Let $\mathcal G$ be the class of simple groups $G$ of the form $G=\mathbb G(k)$ where $k=\bar k$ is an algebraically closed field and $\mathbb G$ is a semisimple adjoint linear algebraic group. Is it true that for all non-power words $w\ne 1$ word map is surjective?
- If we drop the assumption $k=\bar k$, or allow $G$ to be not of adjoint type (i.e., consider quasisimple groups), there are easy counter-examples to surjectivity [@My], [@Bo]. On the other hand, if $G=SU(n)$ and $w(x,y)$ is any word not belonging to the second derived subgroup of $\cF_2$, then the induced word map is surjective for infinitely many $n$ [@ET]. However, if $w$ does belong to the second derived subgroup of $\cF_2$, it may be far from surjective (its image can be arbitrary small in the real topology of $G$) [@Th]).
- If $\mathcal G$ is some infinite family of finite simple groups, then any power word $w=x^n$ gives rise to the word map which is not surjective for infinitely many groups (those whose order is not prime to $n$). A conjecture of Shalev, asserting that such phenomenon may arise only for power maps, turned out to be over-optimistic, see [@JLO] for a counter-example. For simple algebraic groups from Question \[sur-adj\], a power word may not be surjective either. See [@Stb], [@Cha1]–[@Cha4] for details.
- Going over to analogues of Question \[sur-adj\] for finite-dimensional simple associative and Lie algebras over an algebraically closed field, one immediately gets obvious obstructions: there are nontrivial (associative and Lie) polynomials that vanish identically; in addition, an associative polynomial map of matrix algebras may take only central or trace-zero values; even apart from these obvious obstructions, there are counter-examples to surjectivity, see [@BGKP] and [@KBMR]; if the ground field is not algebraically closed, the situation is even more complicated (for example, it is an interesting question which of the surjectivity results of [@KBMR], obtained over quadratically closed fields, survive over $\mathbb R$). Nevertheless, some of techniques developed for associative polynomials may turn out to be useful for attacking Question \[sur-adj\], see Remarks \[rem-am\] and \[rem-uni\] below.
The opposite direction for studying surjectivity is more tricky, even if the class of words $\mathcal W$ consists of only one word. The first non-universal word to be considered is commutator.
### Commutator: Ore and beyond
Let $w=xyx^{-1}y^{-1}\in \cF_2$. The obvious necessary condition for the surjectivity of map is $\left<w(G)\right>=G$, i.e., $G$ must be perfect. It is very far from being sufficient, even within the class of finite groups, see, e.g., [@Is]. Making the further assumption that $G$ is simple, one can say something much more positive, namely that map is surjective in each of the following cases:
- $G$ is finite (Ore’s problem, whose long history has been finished in [@LOST1]);
- $G=S_{\infty}$, the infinite symmetric group [@Or], [@DR] (more generally, [*any*]{} non-power word is surjective on $S_{\infty}$ [@Ly]);
- $G=\mathbb G(k)$, where $\mathbb G$ is a semisimple adjoint linear algebraic group over an algebraically closed field $k$ [@Ree] (and for many other similar cases, such as semisimple Lie groups, etc., see [@VW]);
- $G$ is the automorphism group of some “nice” topological or combinatorial object (e.g., the Cantor set), as listed in [@DR]; see also [@BM].
In connection with (iv), it is worth quoting I. Rivin (see <http://mathoverflow.> [net/questions/77398/how-did-ores-conjecture-become-a-conjecture](net/questions/77398/how-did-ores-conjecture-become-a-conjecture)): “It is a conjecture I attribute to myself, but probably goes back to the ancients, that in every reasonable simple group every element is a commutator”. Of course, the whole point here is in the word “reasonable”: as any meaningful principle, Rivin’s principle is subject to breaking, see Section \[monster\] for counter-examples. In a positive direction, we suggest the following question:
For each of the infinite simple groups listed in [@DR], is it true that any word map is surjective (or, at least, has a “large” image)? (In other words, is it true that in each such group one can solve word equations with “generic” right-hand side?)
We also note that the Ore property is not just a standing alone phenomenon in group theory. It has many interesting applications in algebraic topology (see [@DR] and references therein) and birational geometry [@Ku].
\[beyondOre\] There are several ways to go beyond Ore.
- First, representing $1\ne g\in G$ in the form $g=[x,y]$, one may require to make the choice of $x$ as “uniform” as possible. A typical result of such kind is a theorem of Gow [@Gow]: every finite simple group $G$ of Lie type contains a regular semisimple conjugacy class $C$ such that each semisimple $g\in G$ can be represented in the form $g=[x,y]$ with $x\in C$. A similar result in the case where $G$ is a split semisimple Chevalley group over an infinite field follows from the prescribed Gauss decomposition [@EG]. The case where $G$ is not split seems open.
- One can require that $x,y$ satisfy some additional properties. For example, in [@DR] Ore’s theorem on the infinite symmetric group has been strengthened by choosing $x$ and $y$ so that they generate a transitive subgroup of $S_{\infty}$. It would be interesting to get similar results in the other cases listed above. For instance, one can ask how big can the subgroup $\left<x,y\right>\subset G$ be. Note that we cannot guarantee $\left<x,y\right>=G$, there are counter-examples among alternating groups (we thank A. Shalev for this remark).
- Another well-known generalization is Thompson’s conjecture asserting the existence of a conjugacy class $C$ in every simple group $G$ such that $C^2=G$. This problem has a positive solution for the split semisimple algebraic groups [@EG]. However, it is still open for finite simple groups, in spite of the major breakthrough made in [@EG] (see also [@YCW] and some more recent results (see, e.g., [@BGK] for relevant references)). The case of the infinite symmetric group has been settled in [@Gr]. It is a natural question whether there is a “gap” between Ore’s and Thompson’s conjectures, i.e., whether there is a simple group where Ore’s conjecture holds and Thompson’s does not. N. Gordeev pointed out that such a group exists: take $G=A_{\infty}$, the finitary alternating group. Then evidently each element of $G$ is a commutator but in every fixed conjugacy class $C$ of $G$ any element moves a fixed number $N$ of points, hence any product of two elements of $C$ moves at most $2N$ points. Therefore $C^2\ne G$.
- In the case where $L$ is a “classical” Lie algebra (i.e., the Lie algebra of a Chevalley group $G$ over a sufficiently large field), most of the statements discussed above admit easy analogues when the bracket stands for the Lie operation. Say, the analogue of Ore’s conjecture has been established in [@Br], and an analogue of Gow’s theorem can easily be obtained looking at the linear map $\varphi_x\colon L\to L$, $y\mapsto [x,y]$: its kernel is the centralizer of $x$, and one can conclude that in the representation $g=[x,y]$ the element $x$ can be chosen from a fixed $G$-orbit (with respect to the adjoint action of $G$ on $L$). Further, for the Lie algebras of such kind the analogue of Thompson’s conjecture also holds true [@Gorde]. However, all such generalizations seem unknown for non-split simple classical Lie algebras, simple Lie algebras of Cartan type in positive characteristic, and for infinite-dimensional Lie algebras. The first case to be explored is that of Kac–Moody algebras (as well as Kac–Moody groups, where the prescribed Gauss decomposition is known [@MP]).
- Let now $A$ be an associative algebra, and the polynomial under consideration be the additive commutator $P(X,Y)=XY-YX$. Let $[A,A]$ denote the subalgebra additively generated by the values of $P$. If $A\ne [A,A]$, there is no hope for the surjectivity of the map $P\colon A^2\to A$. This happens whenever $A$ is finitely generated over $\mathbb Q$ or, more generally, if $A$ is any (not necessarily finitely generated) PI-algebra [@Be], [@Mes]. So reasonably interesting questions only arise for the kernel of the reduced trace map (trace zero matrices, for brevity). If $A=M_n(D)$, where $D$ is a division algebra over a field and $n\ge 2$, then every trace zero matrix is an additive commutator [@AR]. If, however, $n=1$, the question is open (except for the special case of a central division algebra of prime degree over a local field [@Ros]), and it is hard to believe in an easy affirmative answer (that would imply the affirmative solution of an old open problem on the cyclicity of any central simple algebra of prime degree $p$ over a field of characteristic $p$, see [@Ros] for details). From discussions with L. Rowen we got an impression that one should rather expect a negative answer, and a counter-example could arise in the algebra of generic matrices (see below). This shows that the questions posed in (iv) for non-split simple Lie algebras are probably answered in the negative. Note that a too straightforward attempt to extend the theorem of Amitsur–Rowen to $M_n(R)$, where $R$ is an arbitrary ring, fails in general [@RR2], [@Ros], [@Mes]. As to infinite-dimensional simple algebras, the questions seem to be totally unexplored, to the best of our knowledge.
### Engel words
Apart from the commutator, there are some other interesting words for which the surjectivity is known in several cases; see, e.g., [@BGK] for some relevant references. Here we only want to mention the most natural generalization of the commutator, the family of Engel words $e_n(x,y)=[[[x,y],y]\dots y]$. Their surjectivity has been only established for groups $G=\PSL(2,q)$ (under some assumptions on $n$ and $q$) [@BGG], $G=\SU(n)$ [@ET], $G=\PSL_2(\CC)$ [@Kl], [@BGG] and for simple classical Lie algebras [@BGKP]. An answer to the following question would be a natural continuation of [@Ree].
\[engel\] Let $G$ be a connected semisimple adjoint linear algebraic group over an algebraically closed field. Is it true that every Engel word induces a surjective map $G^2\to G$?
Dominance: the Deligne–Sullivan trick and Amitsur’s theorem
-----------------------------------------------------------
We start with a theorem of Borel [@Bo] providing a sketch of a somewhat new approach (the proof given in [@La] is essentially the same as in the original paper). Our proof is based on using the generic division algebra (see [@Fo2] for details on this important object, including the history of its creation). Let $F$ be a field (for simplicity, assumed of characteristic zero), let $n$ be a positive integer, let $x^k_{ij}$ $(1\le i,j\le n$, $k\in \mathbb
N$) be independent commuting indeterminates. The $F$-subalgebra of $M_n(F[x^k_{ij}])$ generated by the matrices $X_k=(x^k_{ij})$ is called a ring of generic matrices. Denote it by $R=F\{X\}$. It is a domain [@Am2], and its ring of fractions $Q(R)$ is a central division algebra of dimension $n^2$ over its centre $Z(Q(R))$ [@Am1]. The centre $Z(R)$ of $R$ consists of the central polynomials (and is hence nontrivial for every $n$ [@Fo1],[@Ra]). The centre $C(Q(R))$ is the field of quotients of $C(R)$. It is a long-standing open problem whether the field $C(Q(R))$ is a purely transcendental extension of $F$, see the surveys [@Fo2], [@LB], [@ABGV].
[@Bo] \[Borel\] Let $w\in \cF^d$ ($d\ge 2)$ be a nontrivial word, and let $G$ be a connected semisimple algebraic group over a field $F$. Then the induced morphism $w\colon G^d\to G$ of underlying algebraic varieties is dominant (i.e., its image is Zariski dense).
We present a sketch of proof in the case where $F$ is of characteristic zero. Several parts of the proof are exactly the same as in the original one. First, as dominance is compatible with any extension of the ground field, we may and will assume $F$ to be algebraically closed of infinite transcendence degree. Next, we may assume that $G$ is simple of type $A_n$ (the reduction to the simple case and the passage from $\SL_n$ to the other types are as in [@Bo]). Further, it is enough to prove the dominance for the case where $w$ is a product of commutators. Indeed, suppose that the theorem is proven for such words, and let $w\colon G^d\to G$ be an arbitrary word map. Consider the map $\tilde w\colon G^{2d}\to G$ defined as follows: $\tilde w(x_1,y_1,\dots ,x_d,y_d) :=
w([x_1,y_1],\dots ,[x_d,y_d])$. Then the image of $\tilde w$ is dense, hence so is the image of $w$. Thus henceforth we assume $w\in
[\cF_d,\cF_d]$. Furthermore, it is enough to prove the dominance for the map $$\label{gl}
w\colon (\GL_n)^d\to \SL_n.$$ Indeed, the image of this map coincides with $w(\SL_n)$ because every $g\in\GL_n$ can be replaced with $g/\det(g)$.
Let us now argue by induction on the rank. The case of rank 1 is treated exactly as in [@Bo]. The key point is the induction step. Assume that map is dominant for the rank $n-1$ and prove that it is dominant for the rank $n$. As in the original proof, it is enough to prove the existence of a matrix $C$ in the image of $w$ such that none of its eigenvalues equals 1.
To this end, it suffices to prove that the image contains a [*generic*]{} matrix with this property. Indeed, as soon as this is established, by specialization arguments (which are legitimate because of the assumption on the transcendence degree of $F$) we obtain a non-empty Zariski open set of needed matrices.
To prove the existence of a generic matrix with the required property, assume the contrary. Let $C$ denote a generic matrix from the image of $w$. Since it has an eigenvalue 1, the characteristic polynomial $p_C(t)$ is divisible by $t-1$, so $p_C(t)=(t-1)p_1(t)$. The Cayley–Hamilton theorem then gives $(C-I_n)p_1(C)=0$, which is in a contradiction with Amitsur’s theorem stating that the ring of generic matrices is a domain.
\[rem-am\] In the proof presented above, the core induction argument is based on Amitsur’s theorem, instead of the Deligne–Sullivan argument used in the original proof (which is based on going over to an anisotropic form of $G$ and dates back to the unitary trick of Weyl). We believe that this interrelation between matrix groups and algebras is very important, and its potential is not exhausted. In particular, such an approach may be useful for proving the surjectivity of the word map for semisimple groups over algebraically closed fields and, more generally, for getting a more precise description of the image of the word map.
Let $w=\prod_j a_{i(j)}^{\pm 1}.$ Consider the values of $w$ on invertible matrices, which will be denoted by the letters $a_i$, $i=1,\dots, k$. The matrix $a_i$ satisfies the Cayley–Hamilton equation $$\label{eqHmCl}
a_i^n-\xi_1(a_i)a_i^{n-1}+\dots+(-1)^l\xi_l(a_i)a_i^{n-l}+\dots+(-1)^n\xi_n(a_i)=0$$ where $\xi_l$ is the $l^{th}$ characteristic coefficient of the matrix $a_i$. In particular, $\xi_1=\operatorname{tr}$, $\xi_n=\det$. In the zero characteristic case, the $\xi_l$ are expressed via traces of powers. Equation (\[eqHmCl\]) can be rewritten in the form $$(\sum_{l=1}^{n-1}(-1)^{n-l}a_i^{n-l}\xi_{l-1}(a_i))\det(a_i)^{-1}=a_i^{-1}.$$ Set $\psi(a_i)=\sum_{l=1}^{n-1}(-1)^{n-l}a_i^{n-l}\xi_{l-1}(a_i)$ and rewrite the product $w=\prod_j a_{i(j)}^{\pm 1}$, replacing $a_{i(j)}^{-1}$ with $\psi(a_{i(j)})$ and leaving the factors $a_{i(j)}$ unchanged. We will obtain a polynomial (in the signature enlarged by the characteristic coefficients, in other words, a polynomial with forms), which is a product of polynomials in one variable.
There are some optimistic considerations regarding the study of images of such polynomials. Let us look at the construction of homogeneous polynomials whose image does not coincide with the whole space [@KBMR]. Let $P$ be an arbitrary matrix polynomial, and let $c\in k$. Consider the product $$F=(\lambda_1(P)-c\lambda_2(P))(\lambda_2(P)-c\lambda_1(P))P=
((1+c^2+2c)\det(P)-c\operatorname{tr}(P)^2)P.$$ It vanishes if the ratio of the eigenvalues $\lambda_i(P)$ equals $c^{\pm 1}$ or they are both zero. Therefore the image of $F$ does not contain nonzero matrices with such a ratio of eigenvalues, and this construction is essential for the nature of the problem. If we are given a nontrivial product of polynomials in one variable in which several variables enter, then a requirement on the resulting relation between eigenvalues cannot be determined by one factor (if its determinant is not equal to zero). Of course, $\psi$ is obtained by canceling the determinant, but nevertheless there is some ground for optimism regarding making Borel’s theorem more precise.
\[rem-uni\] Let us give an example of a more complicated argument showing the existence of a unipotent element in the image of a multilinear polynomial $P$ in the space of matrices of second order (assuming it has a noncentral value with nonzero trace).
So let $P(X_1,\dots,X_n)$ be a multilinear polynomial, and assume the contrary, i.e., $P$ has no unipotent values. Fix all variables except for $X_1$. Then $P_{X_2,\dots,X_n}(X_1)=P(X_1,\dots,X_n)$ is a linear in $X_1$ function, meeting the discriminant surface at points of total multiplicity 2. Since we assume the absence of unipotent values, each such point is either scalar or nilpotent. If one point, $X_1^0$, is a nonzero scalar, and the other, $X_1^1$, is nilpotent, then at the point $X_1^0+X_1^1$ we obtain a matrix proportional to a unipotent one with some coefficient $\lambda\ne
0$, and then at the point $(X_1^0+X_1^1)/\lambda$ we shall obtain a unipotent value.
Further, if at both points we have a scalar value, then the original value of $P$ is scalar, and if both values are nilpotent, then the original value has zero trace. Therefore, if the value of $P(X_1,\dots,X_n)$ is not scalar and has nonzero trace, then while moving any coordinate we observe coincidence of the intersection points with the discriminant surface, i.e., tangency to the discriminant surface.
Then one can extract square root of the discriminant $(\lambda_1-\lambda_2)^2$. In other words, $q=\lambda_1(P(X_1,\dots,X_n))- \lambda_2(P(X_1,\dots,X_n))$ is a polynomial in entries of $X_1,\dots,X_n$. The group $\SL_2$ acts by simultaneous conjugations on the $X_i$ and is connected. Then, under such an action, $q$ is mapped to $\pm q$, and connectedness guarantees the invariance of $q$, by the first fundamental theorem (established in [@Pr] in characteristic zero and in [@Do] in positive characteristic).
Since $q$ is a polynomial in characteristic values of the products of the $X_i$, so are $\lambda_1(P), \lambda_2(P)$ because $\lambda_1(P)+\lambda_2(P)=\operatorname{tr}(P)$.
By the Cayley–Hamilton theorem, $(P-\lambda_1(P))(P-\lambda_2(P))=0$ which also contradicts Amitsur’s theorem on zero divisors. Hence $P$ does have unipotent values.
We hope that such kind of reasoning can be helpful in getting a more precise description of the image of the word map in the set-up of Borel’s theorem.
Note that an idea of using generic matrices has been recently used, in a slightly different context of “universal localization”, in [@HSVZ1] and [@Step], for getting subtle information on commutators in Chevalley groups over rings.
There are results of Borel’s flavour, stating that for some infinite groups $G$ any “generic” element $g\in G$ falls into the value set of any non-power word $w$ [@Ma], [@DT].
In the case of Lie algebras, an analogue of Borel’s theorem for Lie polynomials $P$ which are not identically zero, was established in [@BGKP] for semisimple Chevalley algebras (modulo several exceptions over fields of small characteristic), under the additional assumption that $P$ is not an identity of $\mathfrak{sl}(2)$. It is not clear whether this assumption can be removed. The answer heavily depends on whether one can extend a construction of so-called $3$-central polynomials (see, e.g., [@Row Theorem 3.2.21]) to the Lie case. (Recall that a polynomial $P$ is called $n$-central is $P^n$ is central but $P$ is not.) If such polynomials exist, they provide an example of a map which is not dominant. Probably, multilinear Lie polynomials with such a property do not exist, and in this case one can drop the assumption mentioned above.
The case of associative algebras is much more tricky. First, one has to take into account the obvious obstructions to dominance, and assume that the polynomial $P$ is non-central and contains at least one value with nonzero trace. Even under these assumptions, there are counter-examples to dominance for maps on $M_2(K)$; to avoid them, one has to make additional assumptions on $P$ (say, to assume that it is semi-homogeneous). This assumption is not enough for $3\times 3$-matrices. One can show that if $P$ is a nonscalar polynomial such that not all its values on $A=M_3(K)$ are 3-scalar or traceless, then its value set $P(A)$ is dense in $A$ (moreover, if $N$ is the set of non-diagonalizable matrices, then $P(A)\cap N$ is dense in $N$). If $P(A)$ lies in $\mathfrak{sl}_3(K)$ and contains a matrix which is not 3-scalar, then $P(A)$ is dense in $\mathfrak{sl}_3(K)$.
If $P$ is assumed multilinear, the main open question is the validity of the Kaplansky–L${}^{\prime}$vov conjecture which states that the image of $P$ is either 0, or $K$, or $\mathfrak{sl}_n(K)$, or $M_n(K)$. As a first step, we would suggest to look at a downgraded version of this conjecture where we weaken it by allowing the image to be a dense subset of either $\mathfrak{sl}_n(K)$, or $M_n(K)$. As above, one of the key problems is the existence of $n$-central multilinear polynomials. This problem is related to subtle properties of division algebras and is not discussed in this survey. See [@KBMR] for some details.
To sum up, we present a list of questions which seem to be a good starting point.
Let $P\colon M_2(K)^d\to M_2(K)$.
- If $P$ is multilinear, what are its possible images for $K=\mathbb R$? In particular, does the Kaplansky–L${}^{\prime}$vov conjecture hold in this case?
- Let $K$ be quadratically closed and $P$ be an arbitrary (not necessarily homogeneous) polynomial. Is it true that its value set is either the set $\{F([x,y])\}$ of values of a traceless polynomial, or the whole $M_2(K)$ (up to Zariski closure)?
For arbitrary $n\ge 3$ we ask weaker questions.
Let $P\colon M_n(K)^d\to M_n(K)$ be a multilinear polynomial.
- Can it be $n$-central?
- Suppose that $P$ is not $n$-central. Is it true that its image is then dense provided it contains a matrix with nonzero trace? Can it contain non-diagonalizable matrices?
Similar questions can be asked for traceless polynomials.
Being even more modest, one can start with multilinear Lie polynomials, asking similar questions. Even the case of algebras of small rank is open. Let us mention an analogue of the Kaplansky–L${}^{\prime}$vov conjecture.
Let $L$ be a Chevalley Lie algebra, and let $P$ be a multilinear Lie polynomial. Is it true that the image of $P$ on $L$ is either $0$, or $L$?
Even a further downgrading might be of interest, in view of eventual generalizations of Makar-Limanov’s Freiheitsatz [@ML] to the case of positive characteristic: given a polynomial map $P\colon
M_n(K)^d\to M_n(K)$ such that $n\gg\deg P$, can one bound from above the codimension of its image by a “reasonable” function in $n$?
Words with small image
----------------------
If $G$ is a finite simple group, the verbal subgroup $\left<w
(G)\right>$ equals either 1 or $G$. However, for any $G$ one can expect the existence of $w$ such that the actual image $w(G)$ is fairly small (though different from 1). Such examples were recently constructed in [@KN] and [@Le].
(N. Nikolov) [^1] Let $G$ be a finite simple group, and let $X\subset G$ be a union of conjugacy classes. Does there exist $w$ such that $w(G)=X$?
Note that a similar question for polynomials on matrix algebras over finite rings was answered in the affirmative in [@Chu].
Width {#sec:Engel}
=====
General words and polynomials
-----------------------------
In the situation where map is not surjective (or this is not known), but $\left< w(G)\right>=G$, one can ask how many elements of $w(G)$ (or of $w(G)^{\pm}$) are needed to represent every element of $G$. (In the case where $\left< w(G)\right>\ne G$, one represents every element of $\left< w(G)\right>$.) The smallest such number is called the $w$-width of $G$. We denote it by $\wid_w(G)$. If $G$ does not have finite $w$-width, it is said to be of infinite $w$-width. The reader can find comprehensive discussions of this notion in [@Se], [@Ni] (see also [@BGK] for a survey of some more recent developments in the case where $G$ is a finite simple group). We focus here, as above, on parallels with the case of associative algebras.
First note that as soon as we establish, for some topological group $G$, any kind of Borel’s dominance theorem, in the sense that the image of $w$ contains a dense open subset, we immediately conclude by a standard argument that $\wid_w(G)\le 2$. Thus this is the case if $G$ is (the group of points of) a connected semisimple algebraic group over an algebraically closed field. For arbitrary algebraic groups over fields, one can still prove their finite width with respect to any nontrivial word and any element of the verbal subgroup [@Mer]. If, however, we go over to algebraic groups over arbitrary rings, the situation changes dramatically, see the next section.
In the class of finite simple groups, dominance arguments do not make much sense. However, the observation made in [@La] that word maps still have, in a sense, a large image when viewed within an infinite family of finite simple groups, gave rise to a series of wonderful results on uniform word width, christened by Shalev “Waring type properties”. Making the long story short, we just mention the papers [@LS], where the existence of such a uniform bound was established, and [@LST1], [@LST2], where, as a culmination of efforts of many people, the uniform bounds $\wid_w(G)\le 2$ (resp. $\wid_w(G)\le 3$) have been established for all sufficiently large finite simple (resp. quasisimple) groups and all words $w\ne 1$. See [@BGK] for more references and details. Note that quite recently Shalev with his collaborators extended many results of this type to some simple algebraic groups over $p$-adic integers, see [@Sh2] for a list of relevant questions. In the case of polynomial maps of algebras, we propose a notion parallel to word width in the group case.
Let $P(X_1,\dots ,X_d)$ be an associative (resp. Lie) polynomial, let $A$ (resp. $L$) be an associative (resp. Lie) algebra over a ring $R$, let $P\colon A^d\to A$ (resp. $P\colon L^d \to L$) denote the induced map, and let $V_P$ be the $R$-module spanned by the image of $P$. If there exists a positive integer $m$ such that every element $v\in V_P$ can be represented as a sum of $m$ values of $P$, we call the least $m$ with such property the $P$-width of the algebra and denote it $\wid_P(A)$ (resp. $\wid_P(L)$). Otherwise, we say that the algebra is of infinite $P$-width.
We are not aware of any general results concerning this notion, beyond those that follow from the surjectivity or dominance (see, however, next sections for some particular cases). The following question seems conceptually important.
\[alg-width\] Let $k$ be an infinite field. Let $\mathcal A$ denote the class of finite-dimensional central simple $k$-algebras. Let $\mathcal P$ denote the class of associative polynomials such that none of them is central for some $A\in\mathcal A$. Is it true that all $A\in\mathcal A$ are of finite $P$-width for all $P\in \mathcal P$? If yes, is it true that $\sup_{P\in\mathcal P, A\in\mathcal
A}\wid_P(A)<\infty$?
This question makes sense in more restrictive set-ups, when either some class of polynomials or some class of algebras, or both, are fixed (and may be narrow enough, even consisting of one element, see some examples below). Similar questions may be posed for Lie polynomials on finite-dimensional simple Lie algebras. Finally, in both cases (associative and Lie) one can also consider infinite-dimensional simple algebras, finitely or infinitely generated. All this seems completely unexplored. Below we consider a couple of more concrete settings.
Commutator width
----------------
The case of the width of various groups $G$ with respect to $w=xyx^{-1}y^{-1}$ got much attention in the literature. Without pretending to giving a comprehensive survey (various aspects are reflected in [@Se], [@Ni], [@HSVZ2], [@BGK]), we only present some references.
- As mentioned above, the finite simple groups have commutator width 1. As to finite quasi-simple groups, it is at most 2, and all the cases when it actually equals 2 are listed [@LOST2].
- As mentioned above, if $G$ is a Chevalley group over an infinite field, its commutator width is at most 2, in view of Borel’s theorem. However, this breaks down for Chevalley groups over rings, which tend to have very few commutators (the type of behaviour called “anti-Ore” in [@HSVZ1]). In view of examples of groups of infinite commutator width [@DV], we can also call it “anti-Waring”. The situation improves if we go over to infinite matrices: most Chevalley groups of “infinite rank” return to Waring (if not to Ore) behaviour: it is known that their commutator width is at most 2 [@HS], [@DV]; similar results were recently obtained for other groups of infinite matrices, see [@GH] and references therein. This gives rise to a natural question:
Is it true that the groups of commutator width at most 2 listed in [@DV] are of finite word width with respect to any nontrivial word $w$? Is the image of $w$ “large” (i.e., do we have any analogue of Borel’s theorem)?
Here are some parallel results and questions for the additive commutator $$P(X,Y)=XY-YX$$ in associative algebras.
- The commutator width of a matrix algebra $A=M_n(R)$ over any ring is at most two ([@AR] for the case where $R$ is a division algebra, [@Ros] for an arbitrary commutative ring $R$, [@Mes] in general).
- Moreover, in the representation $M=[X,Y]+[Z,T]$ one can fix $X$ and $Z$ which are good for every $M\in A$ (in [@AR] only $Z$ was fixed, and this was strengthened in [@RR1], [@Ros] and [@Mes] using genericity arguments).
This gives rise to the following notion, slightly resembling the notion of one-and-a-half-generation of simple groups [@Stei], [@GK].
\[fracwid-alg\] Let $A$ be an associative algebra of commutator width $m$. If there exist $X_1,\dots ,X_s\in A$ such that every $M\in A$ can be represented in a form $$M=[X_1,Y_1]+\dots + [X_s,Y_s]+[Z_{s+1},Y_{s+1}]+\dots +[Z_m,Y_m],$$ and $s$ is maximal with this property, we say that $A$ is of fractional commutator width $m-\frac{s}{2m}$.
Thus the matrix algebras are of fractional commutator width one and a half (the result of [@AR] gave only one and three-quarters).
One can define fractional commutator width for groups in a similar fashion and ask the following question.
Which of the groups $G$ having commutator width 2 are of fractional commutator width one and a half? one and three-quarters?
One can start with Chevalley groups of infinite rank [@DV] and 14 finite quasisimple groups of width 2 from the list of [@LOST2].
One can define fractional width for more general words and polynomials. We leave this to the interested reader.
Power width
-----------
Again, we only briefly quote some results on power width in groups. For finite simple groups, the story which started with establishing the existence of uniform finite width in [@MZ], [@SW] led to almost conclusive results in [@GM] where for many powers it was proved that this width equals 2. The uniform width for general finite groups was established in [@NS]. For infinite groups (e.g., for Chevalley groups over rings), the problem is almost unexplored. As in the commutator case, some pathologies have been discovered (see the next section).
For matrix algebras $M_n(R)$, we only mention the pioneering paper [@Va] where the problem of representing a matrix over a commutative ring as a sum of $d^{th}$ powers was considered, and some general results of finite power width flavour were obtained (see [@KG] for the history of the problem and more references), and [@Pu2] for some results for matrices over noncommutative rings, in particular, for central simple algebras.
Monsters {#monster}
--------
It was unknown for a long time whether there exists a simple group of commutator width greater than 1. The first counter-example [@BG] appeared in the context of symplectic geometry and gave a simple (infinitely generated) group of infinite commutator width. Later on, using various contexts, there were constructed simple groups of infinite commutator width which are finitely generated ([@Mu1], using small cancellation theory) or finitely presented ([@CF], among quotients of Kac–Moody groups); among groups appearing in [@Mu1], there are those of arbitrary large finite commutator width. In [@Mu2], there were constructed simple groups of infinite square width (and hence infinite commutator width, because of the equality $[x,y]=x^2(x^{-1}y)^2(y^{-1})^2$, showing that finite commutator width implies finite square width). We refer the interested reader to [@GG] for a walk along the zoo of monsters with such anti-Ore and anti-Waring behaviour (and some conceptual explanations of such phenomena), which grew up in differential-geometric environment, following the spirit of [@BG]; they are fed up using some advanced techniques, including quasi-morphisms and quantum cohomology.
It is an interesting question whether such monsters exist among Lie algebras. The case of finite-dimensional algebras of Cartan type over fields of positive characteristic was already mentioned above. As to infinite-dimensional algebras in characteristic zero, discussions with E. Zelmanov give a hint that there are infinitely generated simple Lie algebras of infinite width and finitely generated ones of arbitrary large finite width. Elaborating such examples could give a clue to understanding the situation with associative algebras.
As a final remark, we should mention that apart from Lie algebras, little is known on the problems discussed in this paper in the case of non-associative algebras; see, however, [@Gordo] and [@Pu1] for some particular results of this flavour. In our opinion, it would be interesting to study the image of more general polynomial maps for some classical examples, such as Cayley octonions, simple quadratic Jordan algebras, simple exceptional Jordan algebras $HC_3$, and the like.
[*Acknowledgements*]{}. This research was supported by the Israel Science Foundation, grant 1207/12. Kunyavskiĭ and Plotkin were supported in part by the Minerva Foundation through the Emmy Noether Research Institute of Mathematics. Some questions raised in this paper were formulated after discussions with N. L. Gordeev, L. H. Rowen, A. Shalev, N. A. Vavilov and E. I. Zelmanov to whom we are heartily grateful.
[LOST2]{}
S. A. Amitsur, [*On rings with identities*]{}, J. London Math. Soc. 20 (1955), 464–470.
S. A. Amitsur, [*The T-ideals of the free ring*]{}, J. London Math. Soc. 20 (1955), 470–475.
S. A. Amitsur, L. H. Rowen, [*Elements of reduced trace 0*]{}, Israel J. Math. 87 (1994), 161–179.
A. Auel, E. Brussel, S. Garibaldi, U. Vishne, [*Open problems on central simple algebras*]{}, Transform. Groups 16 (2011), 219–264.
T. Bandman, S. Garion, F. Grunewald, [*On the surjectivity of Engel words on $\PSL(2,q)$*]{}, Groups Geom. Dyn. 6 (2012), 409–439.
T. Bandman, S. Garion, B. Kunyavskiĭ, [*Equations in simple matrix groups: algebra, geometry, arithmetic, dynamics*]{}, Central Europ. J. Math., to appear.
T. Bandman, N. Gordeev, B. Kunyavskiĭ, E. Plotkin, [*Equations in simple Lie algebras*]{}, J. Algebra 355 (2012), 67–79.
J. Barge, E. Ghys, [*Cocycles d’Euler et de Maslov*]{}, Math. Ann. 294 (1992), 235–265.
A. Basmajan, B. Maskit, [*Space form isometries as commutators and products of involutions*]{}, Trans. Amer. Math. Soc. 364 (2012), 5015–5033.
A. Ya. Belov, [*No associative PI-algebra coincides with its commutant*]{}, Sibirsk. Mat. Zh. 44 (2003), 1239–1254; English transl. in Siberian Math. J. 44 (2003), 969–980.
A. Borel, [*On free subgroups of semisimple groups*]{}, Enseign. Math. 29 (1983), 151–164; reproduced in [Œ]{}uvres - Collected Papers, vol. IV, Springer-Verlag, Berlin–Heidelberg, 2001, pp. 41–54.
G. Brown, [*On commutators in a simple Lie algebra*]{}, Proc. Amer. Math. Soc. 14 (1963), 763–767.
P.-E. Caprace, K. Fujiwara, [*Rank-one isometries of buildings and quasi-morphisms of Kac–Moody groups*]{}, Geom. Funct. Anal. 19 (2010), 1296–1319.
M. Casals-Ruiz, I. Kazachkov, [*On systems of equations over free partially commutative groups*]{}, Mem. Amer. Math. Soc. 212 (2011), no. 999.
P. Chatterjee, [*On the surjectivity of the power maps of algebraic groups in characteristic zero*]{}, Math. Res. Lett. 9 (2002), 741–756.
P. Chatterjee, [*On the surjectivity of the power maps of semisimple algebraic groups*]{}, Math. Res. Lett. 10 (2003), 625–633.
P. Chatterjee, [*On the power maps, orders and exponentiality of $p$-adic algebraic groups*]{}, J. reine angew. Math. 629 (2009), 201–220.
P. Chatterjee, [*Surjectivity of power maps of real algebraic groups*]{}, Adv. Math. 226 (2011), 4639–4666.
C.-L. Chuang, [*On ranges of polynomials in finite matrix rings*]{}, Proc. Amer. Math. Soc. 110 (1990), 293–302.
P. Deligne, D. Sullivan, [*Division algebras and the Hausdorff–Banach–Tarski paradox*]{}, Enseign. Math. 29 (1983), 145–150.
R. K. Dennis, L. N. Vaserstein, [*Commutators in linear groups*]{}, K-Theory 2 (1989), 761–767.
S. Donkin, [*Invariants of several matrices*]{}, Invent. Math. 110 (1992), 389–401.
M. Droste, I. Rivin, [*On extension of coverings*]{}, Bull. Lond. Math. Soc. 42 (2010), 1044–1054.
M. Droste, J. K. Truss, [*On representing words in the automorphism group of the random graph*]{}, J. Group Theory 9 (2006), 815–836.
A. Elkasapy, A. Thom, [*About Gotô’s method showing surjectivity of word maps*]{}, <arXiv:1207.5596>.
E. W. Ellers, N. Gordeev, [*On the conjectures of J. Thompson and O. Ore*]{}, Trans. Amer. Math. Soc. 350 (1998), 3657–3671.
E. Formanek, [*Central polynomials for matrix rings*]{}, J. Algebra 23 (1972), 129–132.
E. Formanek, [*The ring of generic matrices*]{}, J. Algebra 258 (2002), 310–320.
J.-M. Gambaudo, E. Ghys, [*Commutators and diffeomorphisms of surfaces*]{}, Ergodic Theory Dynam. Systems 24 (2004), 1591–1617.
N. Gordeev, [*Sums of orbits of algebraic groups*]{}, I, J. Algebra 295 (2006), 62–80.
S. R. Gordon, [*Associators in simple algebras*]{}, Pacific J. Math. 51 (1974), 131–141.
R. Gow, [*Commutators in finite simple groups of Lie type*]{}, Bull. London Math. Soc. 32 (2000), 311–315.
A. B. Gray, Jr., [*Infinite symmetric groups and monomial groups*]{}, Ph.D. Thesis, New Mexico State Univ., 1960.
C. K. Gupta, W. Ho[ł]{}ubowski, [*Commutator subgroup of Vershik–Kerov group*]{}, Linear Algebra Appl. 436 (2012), 4279–4284.
R. M. Guralnick, W. M. Kantor, [*The probability of generating a simple group*]{}, J. Algebra 234 (2000), 743–792.
R. Guralnick, G. Malle, [*Products of conjugacy classes and fixed point spaces*]{}, J. Amer. Math. Soc. 25 (2012), 77–121.
P. de la Harpe, G. Skandalis, [*Sur la simplicité essentielle du groupe des inversibles et du groupe unitaire dans une $C^*$-algèbre simple*]{}, J. Funct. Anal. 62 (1985), 354–378.
R. Hazrat, A. Stepanov, N. Vavilov, Z. Zhang, [*The yoga of commutators*]{}, Zap. Nauch. Sem. POMI 387 (2011), 53–82; English transl. in J. Math. Sci. (New York) 179 (2011), 662–678.
R. Hazrat, A. Stepanov, N. Vavilov, Z. Zhang, [*Commutator width in Chevalley groups*]{}, <arXiv:1206.2128>.
I. M. Isaacs, [*Commutators and the commutator subgroup*]{}, Amer. Math. Monthly 84 (1977), 720–722.
S. Jambor, M. W. Liebeck, E. A. O${}^{\prime}$Brien, [*Some word maps that are non-surjective on infinitely many finite simple groups*]{}, <arXiv:1205.1952>.
A. Kanel-Belov, S. Malev, L. Rowen, [*The images of non-commutative polynomials evaluated on $2\times 2$ matrices*]{}, Proc. Amer. Math. Soc. 140 (2012), 465–478.
M. Kassabov, N. Nikolov, [*Words with few values in finite simple groups*]{}, Quart. J. Math., to appear.
S. A. Katre, A. S. Garge, [*Matrices over commutative rings as sums of $k$-th powers*]{}, Proc. Amer. Math. Soc. 141 (2013), 103–113.
E. Klimenko, Engel maps for $\PSL_2(\CC)$, Preprint.
B. Kunyavskii, [*The Bogomolov multiplier of finite simple groups*]{}, in: “Cohomological and Geometric Approaches to Rationality Problems” (F. Bogomolov, Yu. Tschinkel, Eds.), Progr. Math., vol. 282, Birkhäuser, Boston, MA, 2010, pp. 209–217.
M. Larsen, [*Word maps have large image*]{}, Israel J. Math. 139 (2004), 149–156.
M. Larsen, A. Shalev, P. H. Tiep, [*Waring problem for finite simple groups*]{}, Ann. Math. 174 (2011), 1885–1950.
M. Larsen, A. Shalev, P. H. Tiep, [*Waring problem for finite quasisimple groups*]{}, Intern. Math. Res. Notices, to appear.
L. Le Bruyn, [*Centers of generic division algebras, the rationality problem 1965–1990*]{}, Israel J. Math. 76 (1991), 97–111.
M. Levy, [*Word maps with small image in simple groups*]{}, <arXiv:1206.1206>.
M. W. Liebeck, E. A. O${}^{\prime}$Brien, A. Shalev, P. H. Tiep, [*The Ore conjecture*]{}, J. Europ. Math. Soc. 12 (2010), 939–1008.
M. W. Liebeck, E. A. O${}^{\prime}$Brien, A. Shalev, P. H. Tiep, [*Commutators in finite quasisimple groups*]{}, Bull. London Math. Soc. 43 (2011), 1079–1092.
M. W. Liebeck, A. Shalev, [*Diameters of finite simple groups: sharp bounds and applications*]{}, Ann. Math. 154 (2001), 383–406.
R. C. Lyndon, [*Words and infinite permutations*]{}, in: “Mots. Langue, Raisonnement, Calcul”, Hermès, Paris, 1990, pp. 143–152.
L. Makar-Limanov, [*Algebraically closed skew fields*]{}, J. Algebra 93 (1985), 117–135.
J. A. Maroli, [*Representation of tree permutations by words*]{}, Proc. Amer. Math. Soc. 110 (1990), 859–869.
C. Martinez, E. Zelmanov, [*Products of powers in finite simple groups*]{}, Israel J. Math. 96 (1996), 469–479.
Ju. I. Merzljakov, [*Algebraic linear groups as full groups of automorphisms and the closure of their verbal subgroups*]{}, Algebra i Logika Sem. 6 (1967), no. 1, 83–94. (Russian.)
Z. Mesyan, [*Commutator rings*]{}, Bull. Austral. Math. Soc. 74 (2006), 279–288.
J. Morita, E. Plotkin, [*Gauss decompositions of Kac–Moody groups*]{}, Comm. Algebra 27 (1999), 465–475.
A. Muranov, [*Finitely generated infinite simple groups of infinite commutator width*]{}, Internat. J. Algebra Computation 17 (2007), 607–659.
A. Muranov, [*Finitely generated infinite simple groups of infinite square width and vanishing stable commutator length*]{}, J. Topol. Anal. 2 (2010), 341–384.
J. Mycielski, [*Can one solve equations in groups?*]{}, Amer. Math. Monthly 84 (1977), 723–726.
N. Nikolov, [*Algebraic properties of profinite groups*]{}, <arXiv:1108.5130>.
N. Nikolov, D. Segal, [*Powers in finite groups*]{}, Groups Geom. Dyn. 5 (2011), 501–507.
O. Ore, [*Some remarks on commutators*]{}, Proc. Amer. Math. Soc. 2 (1951), 307–314.
C. Procesi, [*The invariant theory of $n\times n$ matrices*]{}, Adv. Math. 19 (1976), 306–381.
S. Pumplün, [*Sums of squares in octonion algebras*]{}, Proc. Amer. Math. Soc. 133 (2005), 3143–3152.
S. Pumplün, [*Sums of $d$-th powers in non-commutative rings*]{}, Beiträge Algebra Geom. 48 (2007), 291–301.
Yu. Razmyslov, [*On a problem of Kaplansky*]{}, Izv. Akad. Nauk SSSR. Ser. Mat. 73 (1973), 483–501; English transl. in: Math USSR Izv. 7 (1973), 479–496.
R. Ree, [*Commutators in semi-simple algebraic groups*]{}, Proc. Amer. Math. Soc. 15 (1964), 457–460.
M. Rosset, [*Elements of trace zero and commutators*]{}, Ph.D. Thesis, Bar-Ilan Univ., 1997.
M. Rosset, S. Rosset, [*Elements of trace zero in central simple algebras*]{}, in: “Rings, Extensions, and Cohomology (Evanston, IL, 1993)” (A. R. Magid, Ed.), Lecture Notes Pure Appl. Math., vol. 159, Marcel Dekker, New York, 1994, pp. 205–211.
M. Rosset, S. Rosset, [*Elements of trace zero that are not commutators*]{}, Comm. Algebra 28 (2000), 3059–3072.
L. H. Rowen, [*Polynomial identities in ring theory*]{}, Academic Press, New York, 1980.
J. Saxl, J. S. Wilson, [*A note on powers in simple groups*]{}, Math. Proc. Cambridge Phil. Soc. 122 (1997), 91–94.
D. Segal, [*Words: notes on verbal width in groups*]{}, London Math. Soc. Lecture Notes Ser., vol. 361, Cambridge Univ. Press, Cambridge, 2009.
A. Shalev, [*Commutators, words, conjugacy classes and character methods*]{}, Turkish J. Math. 31 (2007), 131–148.
A. Shalev, [*Applications of some zeta functions in group theory*]{}, in: “Zeta Functions in Algebra and Geometry” (A. Campillo [*et al.*]{}, Eds.), Contemp. Math., vol. 566, Amer. Math. Soc., Providence, RI, 2012, pp. 331–344.
A. Stein, [*$1\frac{1}{2}$-generation of finite simple groups*]{}, Beiträge Algebra Geom. 39 (1998), 349–358.
R. Steinberg, [*On power maps in algebraic groups*]{}, Math. Res. Lett. 10 (2003), 621–624.
A. Stepanov, [*Universal localization in algebraic groups*]{}, preprint, 2010, available at <http://alexei.stepanov.spb.ru/papers/formal.pdf>.
A. Thom, [*Convergent sequences in discrete groups*]{}, Canad. Math. Bull. 56 (2013), 424–433.
L. N. Vaserstein, [*On the sum of powers of matrices*]{}, Linear Multilin. Algebra 21 (1987), 261–270.
L. N. Vaserstein, E. Wheland, [*Commutators and companion matrices over rings of stable rank $1$*]{}, Linear Algebra Appl. 142 (1990), 263–277.
S.-K. Ye, S. Chen, C.-S. Wang, [*Gauss decomposition with prescribed semisimple part in quadratic groups*]{}, Comm. Algebra 37 (2009), 3054–3063.
[^1]: This question was answered in the affirmative by A. Lubotzky when the paper was in print. See: A. Lubotzky, [*Images of word maps in finite simple groups*]{}, arXiv:1211.6575.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Super-Earths, objects slightly larger than Earth and slightly smaller than Uranus, have found a special place in exoplanetary science. As a new class of planetary bodies, these objects have challenged models of planet formation at both ends of the spectrum and have triggered a great deal of research on the composition and interior dynamics of rocky planets in connection to their masses and radii. Being relatively easier to detect than an Earth-sized planet at 1 AU around a G star, super-Earths have become the focus of worldwide observational campaigns to search for habitable planets. With a range of masses that allows these objects to retain moderate atmospheres and perhaps even plate tectonics, super-Earths may be habitable if they maintain long-term orbits in the habitable zones of their host stars. Given that in the past two years a few such potentially habitable super-Earths have in fact been discovered, it is necessary to develop a deep understanding of the formation and dynamical evolution of these objects. This article reviews the current state of research on the formation of super-Earths and discusses different models of their formation and dynamical evolution.'
author:
- Nader Haghighipour
title: 'The Formation and Dynamics of Super-Earth Planets'
---
epsf.tex epsf.def psfig.sty
planetary system: formation, planetary system: dynamics
INTRODUCTION {#sec:introduction}
============
The discovery of planets around other stars has undoubtedly revolutionized our understanding of the formation and dynamical evolution of planetary systems. The diverse and surprising characteristics of these objects, both in orbital configuration and physical properties, have confronted astronomers with many new challenges and have reinvigorated the fields of planet formation and dynamics.
One surprising characteristic of the currently known extrasolar planets is the range of their masses. Unlike in the Solar System, where planets belong to two distinct categories of terrestrial (with masses equal to that of Earth or slightly smaller) and giant \[$\sim 14$ Earth masses $({M_\oplus})$ and larger\], many extrasolar planets have masses in an intermediate range, from slightly larger than Earth to 10 ${M_\oplus}$. Dubbed super-Earths, these objects present a new class of planetary bodies with physical and dynamical properties that for the past few years have been the focus of research among many planetary scientists.
The first super-Earth around a main sequence star was discovered by Rivera et al. (2005) using the radial velocity technique. \[Note that in 1992, Wolszczan & Frail (1992) discovered at least two terrestrial-class planets around the pulsar PSR 1257+12.\] Thanks to ground-based observational projects such as the HARPS Search for Southern Extrasolar Planets[^1], the California Planet Survey (CPS)[^2], the Lick-Carnegie Exoplanet Survey (LCE), M2K (Clubb et al. 2009), and the MEarth Project (Nutzman & Charbonneau 2008; Irwin et al. 2009a,b)[^3], and the ongoing success of the CoRoT[^4] and Kepler[^5] space telescopes, to date, the number of these objects has exceeded 90. Tables 1 and 2 show the masses and orbital elements of the currently known super-Earths. As shown, the vast majority of these objects have orbital periods smaller than 50 days. A survey of the parent stars of these bodies indicates that more than half of these stars are hosts to multiple planets. This implies that super-Earths may be more likely to form in short-period orbits and in systems with multiple bodies —- two characteristics that play important roles in developing models of their formation and dynamical evolution.
Among the currently known super-Earths, a few have gained special attention. CoRoT-7 b, the seventh planet discovered by the CoRoT space telescope (Léger et al. 2009; Queloz et al. 2009; Hatzes et al. 2010, 2011), and GJ 1214 b, the first super-Earth discovered by transit photometry around an M star (Charbonneau et al. 2009), are the first super-Earths for which the values of mass and radius have been measured \[CoRoT-7 b: 2.3–8 ${M_\oplus}$, 1.65 Earth radii $({R_\oplus})$; GJ 1214 b: 5.69 ${M_\oplus}$, 2.7 ${R_\oplus}$\]. This major achievement has enabled theoreticians to develop models for the evolution of super-Earths’ interiors (e.g., Valencia et al. 2006, 2007a,b,c, 2009, 2010; O’Neill & Lenardic 2007; Sotin & Schubert 2009; Tackley & van Heck 2009) and their possible atmospheric properties (e.g., Miller-Ricci et al. 2009; Seager & Deming 2009; Bean et al. 2010; Miller-Ricci & Fortney 2010; Rogers & Seager 2010a,b; Bean et al. 2011; Désert et al. 2011; Heng & Vogt 2011; Berta et al. 2012; Menou 2012; Fraine et al. 2013). The three super-Earth-class bodies GL 581 d (Mayor et al. 2009, Forveille et al. 2011), GL 581 g (Vogt et al. 2010, 2012), and GJ 667C c (Anglada-Escudé et al. 2011) have also made headlines. These planets are the first terrestrial-class objects that have been discovered in their respective habitable zones.
For the past few years, the formation and characteristics of super-Earths have been the subject of extensive research.This is primarily because being slightly larger than a typical terrestrial planet, these objects have the capability of developing moderate atmospheres and may have dynamic interiors with plate tectonics —- two conditions that would render a super-Earth potentially habitable if its orbit were in the habitable zone of its host star (see Haghighipour 2011 for a complete review). Also, unlike Earth-sized planets, super-Earths are relatively easy to detect. Current observations of super-Earths have indicated that these objects seem to be more common around cool and low-mass stars (see, e.g., Dressing & Charbonneau 2013, Swift et al. 2013), where the habitable zone is in closer orbit. Two prime examples of such systems are GL 581, an M3V star with one or two potentially habitable super-Earths (Mayor et al. 2009; Vogt et al. 2010, 2012; Forveille et al. 2011), and the M1.5 star GJ 667C, with a 4.5 ${M_\oplus}$ planet in its habitable zone (Anglada-Escudé et al. 2011).
Given the success of observational techniques in detecting potentially habitable super-Earths, and that during the past two years the number of these objects increased twofold, it would be natural to expect that many more habitable super-Earths will be detected in the near future. It is, therefore, imperative to develop a thorough understanding of the formation and dynamical evolution of these bodies, particularly in connection with their habitability. This article presents a review of the current state of research on this topic.
Since there are no super-Earths in the Solar System, it is important to know whether the formation of these objects requires developing new models of planet formation or whether one can use the models of the formation of planets in the Solar System to explain the formation of super- Earths. In the latter case, these models will require major revisions. For instance, one characteristic of super-Earths that presents a challenge to the theories of planet formation is their close-in orbits. While some models suggest that super-Earths were formed at large distances and migrated to their present locations, other models present the possibility of their in-place formation. Fortunately, the physical characteristics of super-Earths, namely their densities,when considered within the context of different planet formation scenarios, present a potential pathway for differentiating between these models. In that respect, the study of super-Earths plays an important role in identifying the most viable planet formation mechanism. The rest of this article presents a review of the current state of research on this topic.
I begin in Section 2 by briefly reviewing the models of planet formation in the Solar System. In Section 3, I discuss in detail the application of these models to the formation of super-Earths, and I conclude in Section 4.
MODELS OF PLANET FORMATION {#sec:FormationModels}
==========================
Explaining the formation of planets is one of the most outstanding problems in planetary astronomy. Despite centuries of efforts to explain the formation of the planets of the Solar System, this problem is still unresolved, and planet formation is still an open question. The discovery of extrasolar planets has added even more to these complexities. As explained in Section 1, many of these objects have physical and orbital properties that are unlike those of the planets in the Solar System and are not well explained by the current models of Solar System formation and dynamics.
Although the diversity of extrasolar planets has been a continuous challenge to the models of planet formation, a common practice in explaining the formation of these objects has been to modify, revise, and/or complement the models of planet formation in the Solar System in such a way that they would be applicable to other planetary bodies. This suggests that to understand the formation of extrasolar planets (such as super-Earths), it is necessary to develop a deep understanding of the models of giant and terrestrial planet formation in the Solar System. This section is devoted to this task. I begin by explaining the growth of dust particles to larger bodies, then discuss different phases of planet growth until a full giant or terrestrial planet is formed.
It is widely accepted that planet formation begins in a circumstellar disk of gas and dust known as a nebula by the growth of dust particles to larger objects. This process, highly dependent on the mass and dynamical properties of the nebula, proceeds in four stages:
- coagulation of dust particles through gentle hitting and sticking, which results in the formation of centimeter- and decimeter-sized objects;
- growth of centimeter- and decimeter-sized bodies to kilometer-sized planetesimals;
- collision and accretion of planetesimals to planetary embryos (moon- toMars-sized objects) in the inner part of the Solar System and to the cores of giant planets in the outer parts; and
- the accretion of gas and formation of giant planets followed by the collisional growth of planetary embryos to terrestrial-class bodies.
The first stage of this process is well understood. Dust grains at this stage undergo different types of random and systematic motions (Weidenschilling 1977) and frequently collide with one another. Particles smaller than 100 $\mu$m are mainly subject to Brownian motion and collide with relative velocities smaller than 1 mm s−1. Larger objects, although slightly faster, are still strongly coupled to the gas, and their dynamics is governed by the gravitational attraction of the central star, nongravitational forces such as radiation pressure, and their interaction with the nebula through gas drag. Gas molecules, however, are subject to pressure gradient (which is necessary for maintaining the gas at hydrostatic equilibrium), and as a result, their velocities are slightly smaller than Keplerian. The slight velocity differences between dust particles and gas molecules cause dust grains to drift inward and approach one another with small relative velocities (Safronov 1969; Weidenschilling 1980; Nakagawa et al. 1981, 1986; Supulver & Lin 2000; Dullemond & Dominik 2005). Turbulence also causes dust grains to collide and is more effective among same-sized particles. As the collisions of dust particles are gentle, van der Waals forces act between their surfaces and stick the dust particles to one another. As shown by laboratory experiments and computational simulations, such gentle collisions result in the fractal growth of dust grains to larger aggregates (Figure 1) (Smoluchowski 1916; Dominik & Tielens 1997; Blum et al. 1998; Wurm & Blum 1998; Blum & Wurm 2000; Krause & Blum 2004; Blum 2006, 2010; Wada et al. 2007).
While the process of the growth of micrometer-sized dust grains to millimeter- and centimeter-sized objects is well understood, the growth of the latter bodies to larger sizes (i.e., kilometer size) is still a big mystery. Simulations have shown that as dust particles grow, their coupling to the gas weakens (i.e., their velocities relative to the gas molecules increase), and they show more of their independent dynamics (Weidenschilling 1977). At this stage, differential vertical settling (Safronov 1969), radial drift (Whipple 1972), and turbulence (Völk et al. 1980, Mizuno et al. 1988, Ormel & Cuzzi 2007) play important roles in driving particles’ relative velocities. The latter causes objects to approach each other rapidly and increases their impact velocities. Results of laboratory experiments and computational simulations have shown that as objects grow to centimeters in size, their sticking efficiency drops dramatically (Blum & Münch 1993), and their relative velocities become so large that their collisions may result in bouncing (bouncing barrier) and/or erosion and fragmentation (fragmentation barrier) (Blum & Wurm 2008, Güttler et al. 2009, Zsom et al. 2010, Beitz et al. 2011).
The above-mentioned bouncing and fragmentation barriers are not the only obstacles in the formation of planetesimals. The sub-Keplerian rotational velocities of gas molecules result in the transfer of angular momentum from solid bodies to the gas and the subsequent drift of these objects toward the central star. The rate of this radial drift is approximately proportional to the size of an object, implying that as an object grows, it approaches the central star in a shorter time. Numerical simulations have indicated that meter-sized bodies have the fastest radial drifts. Combined with turbulence and differential settling, this radial drift increases the relative velocities of solid objects and causes many of them to collide with one another at large speeds. Given that large objects are more prone to collisional destruction (the sticking properties of solid bodies weaken as they grow), it is expected that many of these impacts result in the breakage of the colliding bodies. This process, known as the meter-size barrier, implies that even if the centimeter-size bouncing barrier is overcome, the impact velocities of solid objects become so large that their collisions result in their breaking into small fragments, which subsequently halts their growth to larger sizes. These fragments, even if reaccumulated, will go through the same above-mentioned process and ultimately drift into the central star, leaving the nebula devoid of the solid material necessary for the formation of planetesimals.
Interestingly, despite all these difficulties, planets do exist and so do many kilometer-sized bodies, such as the asteroids and Kuiper belt objects. This implies that during the early stages of planet formation, Nature succeeded in finding a way to overcome the centimeter-sized and meter-sized barriers. It may be that kilometer-sized planetesimals did not form as a result of the mere collisional growth of dust grains; other mechanisms may have also contributed.
A planet-forming nebula is a dynamic environment whose properties and structure vary with time. These variations, in particular in a gaseous disk, may manifest themselves as different structures in the nebula. For instance, regions may appear where the pressure of the gas is locally enhanced. The appearance of such structures will immediately affect the motions of particles in their surroundings. As opposed to a nebula with a monotonic radial pressure profile where gas drag and pressure gradient cause inward migration of solids, in the vicinity of pressure-enhanced regions, the velocity differences between solid objects and gas molecules cause solid particles to undergo inward and outward migrations and to accumulate around the locations of pressure maxima (Haghighipour & Boss 2003a,b; Haghighipour 2005).
In a gaseous disk, the turbulent eddies created by magnetorotational instability are examples of such high-pressure regions. As Johansen et al. (2006, 2007, 2008) have shown, the formation of these turbulent eddies causes small centimeter- and decimeter-sized objects to accumulate in their vicinities and increases the local density of solid material. As the accumulation of solid objects continues, their local spatial density increases until their region becomes gravitationally unstable and the accumulated bodies fragment into several 100–1,000-km-sized planetesimals. This mechanism, known as streaming instability, has been presented as a scenario for planetesimal formation. \[See Chiang & Youdin (2010) for a review and Cuzzi et al. (2008) and Weidenschilling (2010) for alternative viewpoints.\]
It is important to note that as shown by Shariff & Cuzzi (2011), the local enhancement of solid to gas surface density necessary for the onset of instability is achievable only when the turbulence is extremely weak. These authors indicate that when the effect of turbulent mass diffusivity is taken into account, streaming instability becomes inefficient, and the growth rate of planetesimals reduces significantly.
Other mechanisms of the formation of planetesimals include trapping dust particles in vortices (Barge & Sommeria 1995, Klahr & Henning 1997, Lyra et al. 2009a), trapping particles in pressure enhanced regions created by the evaporation front of water in the protoplanetary disk (Kretke & Lin 2007; Brauer et al. 2008a,b; Lyra et al. 2009b), turbulent concentration of solids (Chambers 2010), turbulent clustering of protoplanetary bodies (Pan et al. 2011), concentration of solid objects at the snowline (the region beyond which water is in the permanent state of ice) as a result of the sublimation of drifting ice aggregates (Aumatell & Wurm 2011), trapping of solid objects in dead zones (Gressel et al. 2012) and at the boundary between steady super/sub-Keplerian flow created by inhomogeneous growth of magnetorotational instabilities (Kato et al. 2012), rapid coagulation of porous dust aggregates outside the snowline (Okuzumi et al. 2012), and planetesimal formation in self-gravitating disks (Gibbons et al. 2012, Shi & Chiang 2013).
The four stages of planet formation outlined above share one interesting feature: The underlying physics of each stage is almost distinct from that of the other phases. This makes it possible to study each phase separately. Once the dust grains have grown and kilometer-sized planetesimals are formed, although the circumstellar disk still contains gas and dust, its dynamics is now mainly driven by the interaction of planetesimals with one another. These interactions are primarily gravitational, although gas drag also plays a role. At this stage, because the planetesimals are the main components populating the disk, collisions among these objects are frequent, which results in low eccentricities and low inclinations for these bodies. Because the relative velocity between two bodies is an increasing function of their orbital eccentricities, lowering the eccentricity of planetesimals due to their mutual collisions and dynamical friction, combined with their almost coplanar orbits, reduces their relative velocities. The latter facilitates the merging of these objects and enhances the rate of their accretion to larger bodies.
As a planetesimal grows, the influence zone of its gravitational field expands and as a result, it attracts more material from its surroundings. In other words, more material will be available for the planetesimal to accrete, and the rate of its growth increases. Known as runaway growth, this process results in the growth of kilometer-sized planetesimals to larger bodies in a short time (Safronov 1969; Greenberg et al. 1978; Wetherill & Stewart 1989, 1993; Ida & Makino 1993; Kokubo & Ida 1996, 2000; Weidenschilling et al. 1997).
Runaway growth is a local process. Since the collision of two objects is more likely to result in their coalescence when their relative velocity is small, the effectiveness of this process in producing larger bodies, and the type and size of the resulting objects, varies at different distances from the central star. At large distances (e.g., $> 5$ AU from the Sun), where the rotational velocities are small, planetesimals approach each other with small relative velocities, and their impacts are likely to result in accretion. Also, because the temperature in the circumstellar disk is low at such distances, the bulk material of such planetesimals is primarily ice, which increases the efficiency of their sticking at the time of their collision. As a result, planetesimals at large distances grow to objects of a few Earth masses in a short time. As this process occurs while the nebular gas is still present, a growing object gradually attracts gas from its surroundings, forming a large body with a thick gaseous envelope and a mass equal to a few hundred Earth masses. At this state, a gas-giant planet is formed. This scenario, known as the core-accretion model, has been proposed as a mechanism for the formation of gas-giant planets in the Solar System (Pollack et al. 1996, Hubickyj et al. 2005, Lissauer et al. 2009, Movshovitz et al. 2010).
As the giant planets form at large orbits, the runaway accretion takes a slightly different path in the inner parts of the disk. Similar to the formation of the cores of gas-giant planets, the collisions of planetesimals at this stage may result in their growth to larger bodies. However, because the orbital motions of planetesimals are faster, they may approach each other with larger relative velocities. Also, many of these objects may lose their surface ices and other volatiles at closer distances, and as a result, when they collide with one another, the efficiency of their accretion will not be as high as for those at larger orbits. Simulations of the collision and growth of planetesimals in the inner part of the Solar System have shown that instead of forming objects as big as the cores of giant planets, accretion of these bodies results in the formation of several hundred moon- to Mars-sized objects known as planetary embryos. Computational simulations (Bromley & Kenyon 2006) and analytical analysis (Goldreich et al. 2004) have shown that when the masses of these embryos reach lunar mass, the dynamical friction of the swarm of planetesimals can no longer dampen their orbits, and their runaway growth ends. At this stage, the gravitational perturbation of the resulting planetary embryos, combined with the perturbation of giant planets, strongly affects the dynamics of smaller planetesimals and causes many of them to collide at high velocities and shatter one another, and/or their orbits become highly eccentric, and they subsequently scatter to large distances where they may leave the gravitational field of the system. This growth and clearing process continues until terrestrial planets are formed and the smaller remaining bodies (asteroids) are in stable orbits (Figure 2) (Wetherill 1990a,b, 1994, 1996; Kokubo & Ida 1995, 1998, 2007; Chambers & Wetherill 1998, 2001; Agnor et al. 1999; Morbidelli et al. 2000, 2012; Chambers 2001; Chambers & Cassen 2002; Levison & Agnor 2003; Raymond et al. 2004, 2005a,b, 2006b, 2007, 2009; Kokubo et al. 2006; O’Brien et al. 2006; Hansen 2009; Schlichting et al. 2012; Torres et al. 2013; Haghighipour et al. submitted; Izidoro et al. submitted). Since the accretion and reaccretion of bodies in smaller orbits are not as efficient as in the outer regions, unlike the growth of gas-giant planets, the formation of terrestrial bodies will take several hundred million years. Figure 2 shows the time evolution of a sample simulation of terrestrial planet formation (Haghighipour et al. submitted; Izidoro et al. submitted). The planet formation models as explained above, although capable of explaining many features of the Solar System, face several complicated challenges. The core-accretion model, for instance, requires the nebular gas to be available for $\sim 10$ Ma while the core of Jupiter grows and accretes gas from its surroundings (Pollack et al. 1996). However, the observational estimates of the lifetimes of disks around young stars suggest a lifetime of 0.1–10 Ma, with 3 Ma being the age at which half the stars show evidence of disks (Strom et al. 1993, Haisch et al. 2001, Chen & Kamp 2004, Maercker et al. 2006). These simulations also suggest a solid core for Jupiter with a mass of $\sim 10 {M_\oplus}$. Computational modeling of the interiors of Jupiter and Saturn, however, has indicated different possible values for the cores of these objects, ranging from 0 to as large as $14 {M_\oplus}$ (Guillot 2005, Militzer et al. 2008). It is unclear what the actual masses of the cores of our gas-giant planets are, and if smaller than $10 {M_\oplus}$, how they accumulated their thick envelopes in a short time. I refer the reader to a review by Guillot (2005) for more details.
To overcome these difficulties, the core-accretion model has undergone several improvements. Hubickyj et al. (2005) and Lissauer et al. (2009) have shown that increasing the surface density of the nebula to higher than that suggested by Pollack et al. (1996) significantly reduces the time of the giant planet formation. An improved treatment of grain physics as given by Podolak (2003), Movshovitz & Podolak (2008), and Movshovitz et al. (2010) has also indicated that the value of the grain opacity in the envelope of the growing Jupiter in the original core-accretion model (Pollack et al. 1996) is too high, and a lower value has to be adopted. This lower opacity has led to a revised version of the core-accretion model in which the time of giant planet formation is considerably smaller (Hubickyj et al. 2005, Movshovitz et al. 2010). Most recently, Bromley & Kenyon (2011) have developed a new hybrid N-body-coagulation code that has enabled the authors to form Saturn- and Jupiter-sized planets in $\sim 1$ Ma.
An alternative model for the formation of gas-giant planets addresses this issue by proposing rapid formation of giant planets in a gravitationally unstable nebula (Boss 2000a,b, 2003; Mayer et al. 2002, 2004, 2007; Durisen et al. 2007; Boley 2009; Boley et al. 2010; Cai et al. 2010). Known as the disk-instability scenario, this model suggests that local gravitational instabilities in the solar nebula may result in the fragmentation of the disk to massive clumps that subsequently contract and form gas-giant planets in a short time. Boss’s (2000a,b) and Mayer et al.’s (2002, 2003, 2004) results show that an unstable disk can break up into giant gaseous protoplanets in as short a time as $\sim 1,000$ years. Although this mechanism presents a fast track to the formation of a gas-giant planet, it suffers from the lack of an efficient cooling process necessary to take energy away from a planet-forming clump in a sufficiently short time before it disperses.
FORMATION OF SUPER-EARTHS {#Super-EarthFormation}
=========================
The extent to which current planet formation scenarios can be used to explain the formation of super-Earths varies with the mass and orbital architecture of these objects. Since the dynamics and characteristics of planet-forming nebulae are different for stars with different spectral types, the parent stars of super-Earths also play an important role. The range of masses for the currently known super-Earths, when considered within the context of giant and terrestrial planet formation scenarios, points to two general pathways for the formation of these objects. The low-mass super-Earths could have formed in place following a similar process as the formation of terrestrial planets in the Solar System (see, e.g., Chiang & Laughlin 2012). The larger super-Earths, with masses close to their upper limit, may be the result of an unsuccessful and incomplete giant planet formation (see, e.g., Rogers et al. 2011). In this scenario, the super-Earths’ larger than terrestrial masses, combined with the fact that many of these objects are in short-period orbits, point to a formation scenario in which super-Earths are formed at large distances (where more material is available for their growth) and either migrate to their current locations as they interact with the protoplanetary disk (Kennedy & Kenyon 2008b) or are scattered to their current orbits as a result of interactions with other cores and/or planets (Terquem & Papaloizou 2007). In other words, the formation of these objects may have occurred while their orbital elements were evolving (Terquem & Papaloizou 2007; Kennedy & Kenyon 2008a,b). This mechanism naturally favors the core-accretion model of gas-giant planet formation, although attempts have also been made to explain the formation of super-Earths via the disk-instability scenario (see Section 3.3). As mentioned above, super-Earths owe their popularity to their masses and sizes, which under favorable conditions may render them habitable. While planet formation models allow for the formation of super-Earths around all types of stars (either as a failed core of a giant planet or as a slightly larger terrestrial-class object), because of the current sensitivity of detection techniques, a great deal of interest exists in super-Earths in the habitable zones of cool and low-mass stars (e.g., M dwarfs). For this reason, I devote the rest of this article to presenting a review of the models of super-Earth formation around M stars.
Formation of Super-Earths Around Low-Mass Stars
-----------------------------------------------
The discovery of planets of different sizes, from Jovian-type \[e.g., GJ 876 b, c, and e (Rivera et al. 2010); HIP 57050 b (Haghighipour et al. 2010); GL 581 b (Bonfils et al. 2005); KOI-254 b (Johnson et al. 2012); Kepler-32 d (Swift et al. 2013)\] to small super-Earths \[e.g., GL 581 c, d, e, and g (Udry et al. 2007, Mayor et al. 2009, Vogt et al. 2010); GJ 667C c (Anglada-Escudé et al. 2011); Kepler-32 b and c (Swift et al. 2013)\] around M dwarfs indicates that both giant and terrestrial planet formation can proceed efficiently around low-mass stars. This implies that the circumstellar disks around these stars can accommodate the formation of super-Earths both as a failed core of a giant planet through the gas-giant planet formation process, and also as small terrestrial-class objects through direct collisional growth of protoplanetary bodies and planetary embryos. These mechanisms have to also account for the short periods of super-Earths, whether through planet migration, planet-planet scattering, or a combination of both. I begin this section by considering the core-accretion model as the mechanism for the formation of super-Earths. As mentioned above, the discovery of super-Earths can be taken as strong evidence in support of this model. However, as is explained at the end of the next section, this mechanism alone cannot explain the formation and orbital architecture of all the currently known super- Earths. Other effects such as the evolution of the central star and planet migration have to be taken into consideration as well. I discuss these effects in the next section and conclude this article by reviewing the formation of super-Earths through the disk-instability model.
The Core-Accretion Model
------------------------
As mentioned in Section 2, the efficiency of the core-accretion model and the rate of the growth of the cores of giant planets increase with the disk surface density. Around low-mass stars, where the surface density of the disk is smaller than around the Sun, the solid material (i.e., the planetesimals) is more spatially scattered, and as a result, the collisions among planetesimals and planetary embryos are less frequent. This smaller rate of collision prolongs the growth of planetesimals to larger sizes, and causes the time of the core growth around low-mass stars to be several times longer than the time of the formation of Jupiter around the Sun. As shown by Laughlin et al. (2004), in disks around stars with masses smaller than 0.5 solar masses $({M_\odot})$, the core-accretion mechanism can produce planets ranging from terrestrial-class to Neptune sizes. However, the time for the formation of these objects is much longer than the time for the formation of Jupiter in the Solar System through the core-accretion model. During this time, around M stars, for instance, the gaseous component of the circumstellar disk disperses, leaving the slowly growing core with much less gas to accrete.
The short lifetime of the gas in circumstellar disks around M stars can be attributed to two important factors:
- the high internal radiation of young M stars (at this stage, these stars are almost as bright as Sun-like stars), and
- external perturbations from other close-by stars.
The latter is primarily due to the fact that most stars are formed in clusters (Lada & Lada 2003), and as such, their circumstellar disks are strongly affected by the gravitational perturbations and the radiations of other stars (Adams et al. 2004). For M stars, this causes the circumstellar disk to receive a high amount of radiation from both the central star and external sources. This high amount of radiation combined with the low masses of M stars, which points to their small gravitational fields, increases the effectiveness of the photoevaporation of the gaseous component of the circumstellar disk by up to two orders of magnitude. As a result, the majority of the gas leaves the disk at the early stages of giant planet formation, leaving a still-forming core with not much gas to accrete.
### Effect of stellar evolution
Although the growth of giant planets’ cores through collision and accretion of planetesimals is similar in disks around solar-type and low-mass stars, the fact that around smaller stars this process takes longer introduces a fundamental difference in the formation of giant planets in these two environments. As opposed to young Sun-like stars whose luminosities stay almost constant during the formation of giant and terrestrial planets (e.g., 10-100 Ma), the luminosity of a premain sequence, low-mass star (e.g., $0.5 {M_\odot}$) fades by a factor of 10 to 100 during this process (Hayashi 1981). This causes the internal temperature of the circumstellar disk to decrease, which subsequently causes the disk’s snowline to move toward the central star and to close distances. The forward migration of the snowline results in an increase in the population of icy materials (kilometer-sized and larger planetesimals) in the outer regions of the disk, which in turn increases the efficiency of the collisional growth of these objects to protoplanetary bodies (as mentioned in Section 2, sticking is more efficient among icy bodies). As shown by Kennedy et al. (2006), around a 0.25-$M_\odot$ star, the moving snowline causes rapid formation of planetary embryos within a few million years (also see Kennedy et al. 2007). Subsequent collisions and interactions among these objects result in the formation of super-Earths in approximately 50-500 Ma.
### Effect of planet migration
As mentioned above, one of the major developments in the field of planetary dynamics that was a direct consequence of the detection of extrasolar planets is the concept of planet migration. Although previously post-formation migration had been proposed as a mechanism to explain the orbital architecture of small bodies in the Solar System (e.g., moons of giant planets and Kuiper belt objects), the migration of planets during their formation had not been incorporated into the models of planetary formation. In other words, the planet formation scenarios mentioned above were developed assuming that planets form in place. The discovery of extrasolar planets, almost from the beginning, challenged this assumption. The detection of the first hot Jupiter in a 4-day orbit around the star 51 Pegasi (Mayor & Queloz 1995) revealed that planet migration is an inseparable part of the evolution of a planetary system and prompted astronomers to revisit this concept and to incorporate it into their models of planet formation. Today, planet migration is well developed and widely accepted as part of a comprehensive planet formation scenario.
Planetary and satellite migration has long been recognized as a major contributor to the formation and orbital architecture of planets, their moons, and other minor bodies in the Solar System. As shown by Greenberg et al. (1972) and Greenberg (1973), mean-motion resonances (i.e., commensurable orbital periods[^6]) among the natural satellites of giant planets (e.g., Titan and Hyperion, satellites of Saturn) may have been the result of the radial migration of these objects due to their tidal interactions with their parent planets (Goldreich 1965). The dynamical architecture of Galilean satellites, with their three-body, Laplace resonance, has also been attributed to the migration of these objects. It is accepted that these satellites migrated inward during their formation as a result of interacting with the circumplanetary disk of satellitesimals around Jupiter (Canup & Ward 2002), and subsequently by tidal forces after their formation (Peale & Lee 2002). The lack of irregular satellites between Callisto, the outermost Galilean satellite, and Themisto, the innermost irregular satellite of Jupiter, also can be explained by a dynamical clearing process that occurred during the formation and migration of Galilean satellites (Haghighipour & Jewitt 2008). Among the planets of our Solar System, the post-formation, planetesimal-driven migration of giant planets has been proposed as a mechanism to explain the current state of the asteroid belt (Tsiganis et al. 2005; Minton & Malhotra 2009, 2011; see also Gomes 1997), late heavy bombardment (Gomes et al. 2005), the origin of Jupiter Trojan asteroids (Morbidelli et al. 2005), the effects of secular resonances on terrestrial planet formation (Agnor & Lin 2012), and the small mass and size of Mars (Walsh et al. 2011). I refer the reader to Morbidelli et al. (2012) for a review on these topics.
The idea of the migration of planetary bodies was first proposed by Fernandez & Ip (1984). These authors suggested that after the dispersal of the nebular gas, fully formed giant planets may drift from their original orbits due to the exchange of angular momentum with the disk of planetesimals. As a result of this post-formation migration, small bodies either are scattered out of the Solar System or may reach other regions where they may reside in long-term stable orbits. As shown by Malhotra (1993, 1995), this mechanism can explain the peculiar orbit of Pluto (highly eccentric, inclined, and long-term chaotic), and as shown by Malhotra (1996) and Hahn & Malhotra (2005), it can also explain the dynamical structure of Kuiper belt objects.
The past two decades have witnessed major developments in the theories of planet migration. Simulations of the formation of planetary bodies and their interactions with circumstellar disks have shown that planet migration does not have to occur necessarily after the planets are fully formed. In fact, planets can migrate while they are forming as a result of exchanging angular momentum with their surrounding environment. This naturally suggests that the physical and dynamical characteristics of a planet and its circumstellar disk will play an important role in this process. For instance, the planet may undergo type I migration, in which case it does not accrete nebular material as it migrates (Figure 3a). Conversely, the planet may be large and accrete nebular material, in which case it may create a gap in the disk as it migrates (Figure 3b). This type of migration is known as type II migration. Planet migration may occur in other forms as well.[^7]
The contribution of planet migration to the formation of close-in super-Earths may appear in different forms. The most common scenario involves the inward migration of a fully formed giant planet in a disk of planetesimals and planetary embryos. The giant planet in this scenario affects the dynamics of protoplanetary bodies interior to its orbit by either increasing their orbital eccentricities and scattering them to larger distances or causing them to migrate to closer orbits. The migrating protoplanets may be shepherded by the giant planet into small close-in regions, where they are captured in mean-motion resonances. As Zhou et al. (2005), Fogg & Nelson (2005, 2006, 2007a,b, 2009), and Raymond et al. (2008) have shown, around Sun-like stars, the shepherded protoplanets may also collide and grow to terrestrial-class and super-Earth objects (see, e.g., Figure 6b). Studies of the back-scattered objects in the simulations of disks around massive stars have shown that these bodies may also collide and grow to planetary sizes (Mandell & Sigurdsson 2003, Raymond et al. 2006a, Mandell et al. 2007).
While around Sun-like stars, despite the out-scattering of protoplanetary bodies during the migration of a giant planet, the formation of super-Earths through the collision and growth of planetesimals and planetary embryos proceeds efficiently, around low-mass stars this scenario is not always the case. Simulations of the dynamics of protoplanetary bodies at distances smaller than 0.2 AU around a 0.3 ${M_\odot}$ star have shown that during the inward migration of one or several giant planets (the latter involves migrating planets in mean-motion resonances), the majority of the protoplanets leave the system and do not contribute to the formation of close-in Earth-sized bodies and/or super-Earths (Figure 4) (Haghighipour & Rastegar 2011). These results suggest that the currently known small planets around M stars might have formed at larger distances and were either scattered to their current close-in orbits (e.g., GJ 876 d; see Figure 5) or migrated into their orbits while captured in a mean-motion resonance with a migrating planet.
The above-mentioned scenario for the formation of close-in super-Earths is based on the fact that giant planets are formed long before the protoplanetary bodies grow to larger sizes. The underlying assumption in this scenario is that the giant planet does not migrate during its formation, and the migration of the planetary embryos (the moon- to Mars-sized objects) is also ignored. However, not only do the cores of still-forming giant planets migrate (Alibert et al. 2004), so too do the planetary embryos. While migrating, the embryos may undergo orbital crossing and collisional merging, which may result in their growth to a few super-Earths, especially in mean motion resonances. Simulating the interactions of 25 protoplanetary objects with masses ranging from 0.1 to 1 $M_\oplus$, Terquem & Papaloizou (2007) have shown that a few close-in super-Earths may form in this way with masses up to $12 {M_\oplus}$. The results of these simulations suggest that in systems in which merging of migrating cores results in the formation of super-Earths and Neptune-like planets, such planets will always be accompanied by giant bodies and most likely will be in mean-motion resonances. Similar results have also been reported by Haghighipour & Rastegar (2011).
Interestingly, several planetary systems have been discovered in which central stars host only small Neptune-sized objects and super-Earths (e.g., HD 69830, GL 581). The planets in these systems do not have a Jupiter-like companion that could have migrated to facilitate their formation. Such systems seem to imply that a different mechanism may be responsible for the formation of their super-Earth bodies. Kennedy & Kenyon (2008a) and Kenyon & Bromley (2009) have suggested that the migration of protoplanetary embryos may be the key in facilitating the close-in accretion of these objects. These authors considered a circumstellar disk with a density enhancement at the region of its snowline and simulated the dynamics and growth of its planetary embryos. They showed that while interacting with one another (colliding and accreting), many of these objects may migrate toward the central star. Around a solar-type star, the time of such migrations for an Earth-sized planet at 1 AU is $\sim {10^5} - {10^6}$ years - much smaller than the time for the chaotic growth of a typical moon- or Mars-sized embryo ($10^8$ years) (Goldreich et al. 2004). This implies that most of the migration occurs prior to the onset of the final growth. Depending on their relative velocities, the interactions among the migrating embryos may result in their growth, scattering, and/or shepherding, as in the case of a migrating giant planet. Simulations by Kennedy & Kenyon (2008b) and Kenyon & Bromley (2009) have shown that super-Earth objects with masses up to 8 ${M_\oplus}$ may form in this way around stars ranging from 0.25 to 2 $M_\odot$ (Figure 6).
The Disk-Instability Model
--------------------------
-5pt The formation of super-Earths through the mechanisms explained above, particularly when those mechanisms are used to explain the formation of these objects at the higher end of their mass range, naturally favors the core-accretion model of giant planet formation. However, the fact that Jovian-type planets have been discovered around low-mass stars (e.g., GJ 876, with three planets ranging from 1 Uranus mass to 2.2 Jupiter masses in $\sim$120-, 60-, and 30-day orbits; HIP57050, with a Saturn-mass planet in a $\sim$40-day orbit) suggests that the disk-instability model may also be able to form close-in super-Earths, especially those that are considered as failed cores of giant planets. As explained above, given the low masses of the circumstellar disks around M stars, the existence of giant planets around these stars suggests that they might have formed at large distances and migrated to their current orbits. This is because in a planet-forming nebula, more nebular material is available at outer regions that can then facilitate the formation of a giant planet through the core-accretion model. The availability of more mass at outer distances in a disk may also trigger the formation of giant planets around M stars through the disk-instability scenario. Recall that in this scenario, clumps, formed in an unstable gaseous disk, collapse and form gas-giant planets (e.g., Boss 2000b, Mayer et al. 2002). After the giant planets are formed, a secondary process is needed to remove their gaseous envelopes. As Boss (2006) has shown, such collapsing clumps can form around a 0.5 $M_\odot$ star at a distance of $\sim$8 AU (Figure 7). This author suggests that, as most stars are formed in clusters and in high-mass, star-forming regions, intense far/extreme UV radiations from nearby O stars may rapidly (within 1 Ma) photoevaporate the gaseous envelopes around giant planets, leaving them with large super-Earth cores. Similar mechanisms have been suggested for the formation of Uranus and Neptune in the Solar System (Boss et al. 2002). A subsequent migration, similar to that suggested by Michael et al. (2011), may then move these cores to close-in orbits.
CONCLUDING REMARKS
==================
-5pt As evident from this review, it is generally accepted that super-Earths are formed through a combination of a core accumulation process and planetary migration. Modeling the formation of these objects requires the simulation of the collisional growth of planetary embryos and their subsequent interactions with the protoplanetary disk. A realistic model requires global treatment of the disk and inclusion of large numbers of planetesimals and planetary embryos. In practice, such simulations are computationally expensive. To avoid such complications, most of the current models of super-Earth formation include only small numbers of objects (e.g., cores, progenitors, protoplanets, planetesimals). As shown by McNeil & Nelson (2010), in systems with large numbers of bodies (e.g., several thousand planetesimals and larger objects), the combination of traditional core accretion and type I planet migration may not produce objects larger than 3-4 $M_\oplus$ in close-in (e.g., $\leq 0.5$ AU) orbits. Although the systems studied carry some simplifying assumption, McNeil & Nelson’s results point to an interesting conclusion: While the combination of core accretion and planet migration seems to be a viable mechanism for the formation of close-in super-Earths, the formation of these objects is still an open question, and a comprehensive theory for their formation requires more sophisticated computational modeling, with possibly entirely new physics, as yet to be discovered.
ACKNOWLEDGMENT
==============
I am grateful to Jörgen Blum, Alan Boss, Andre Izidoro, Scott Kenyon, Frédéric Massét, and Ji-Lin Zhou for kindly providing figures.
REFERENCES
==========
Adams FC, Hollenbach D, Laughlin G, GortiU. 2004. Photoevaporation of circumstellar disks due to external far-ultraviolet radiation in stellar aggregates. Astrophys. J. 611:360-79
0.5pt Agnor CB, Canup RM, Levison HF. 1999. On the character and consequences of large impacts in the late stage of terrestrial planet formation. Icarus 142:219-37
0.5pt Agnor CB, Lin DNC. 2012. On the migration of Jupiter and Saturn: constraints from linear models of secular resonant coupling with the terrestrial planets. Astrophys. J. 745:143
0.5pt Alibert Y, Mordasini C, Benz W. 2004. Migration and giant planet formation. Astron. Astrophys. 417:L25-28
0.5pt Anglada-Escudé G, Arriagad P, Vogt SS, Rivera E, Butler RP. 2011. A planetary system around the nearby M dwarf GJ 667C with at least one super-Earth in its habitable zone. Astrophys. J. 751:L16
0.5pt Armitage PJ. 2010. The early evolution of planetary systems. In Astrophysics of Planet Formation, pp. 218-62. Cambridge, UK: Cambridge Univ. Press
0.5pt Aumatell G, Wurm G. 2011. Breaking the ice: planetesimal formation at the snowline. MNRAS 418:L1-5
0.5pt Barge P, Sommeria J. 1995. Did planet formation begin inside persistent gaseous vortices? Astron. Astrophys. 295:L1-4
0.5pt Baruteau C, Mass´et F. 2013. Recent developments in planet migration theory. In Tides in Astronomy and Astrophysics (Lecture Notes in Physics), ed. J Souchay, S Mathis, T Tokieda, p. 201. Berlin: Springer-Verlag
0.5pt Bean JL, Désert J-M, Kabath P, Stalder B, Seager S, et al. 2011. The optical and near-infrared transmission spectrum of the super-Earth GJ 1214b: further evidence for a metal-rich atmosphere. Astrophys. J. 743:92
0.5pt Bean JL, Miller-Ricci Kempton E, Homeier D. 2010. A ground-based transmission spectrum of the super-Earth exoplanet GJ 1214b. Nature 468:669-72
0.5pt Beitz E, Güttler C, Blum J, Meisner T, Teiser J,Wurm G. 2011. Low-velocity collisions of centimeter-sized dust aggregates. Astrophys. J. 736:34
0.5pt Berta ZK, Charbonneau D, Désert J-M, Miller-Ricci Kempton E, McCullough PR, et al. 2012. The flat transmission spectrum of the super-Earth GJ1214b from Wide Field Camera 3 on the Hubble Space Telescope. Astrophys. J. 747:35
0.5pt Blum J. 2006. Dust agglomeration. Adv. Phys. 55:881-947
0.5pt Blum J. 2010. Dust growth in protoplanetary disks—a comprehensive experimental/theoretical approach. Res. Astron. Astrophys. 10:1199-214
0.5pt Blum J, Münch M. 1993. Experimental investigations on aggregate-aggregate collisions in the early solar nebula. Icarus 106:151-67
0.5pt Blum J, Wurm G. 2000. Experiments on sticking, restructuring, and fragmentation of preplanetary dust aggregates. Icarus 143:138-46
0.5pt Blum J, Wurm G. 2008. The growth mechanisms of macroscopic bodies in protoplanetary disks. Annu. Rev. Astron. Astrophys. 46:21-56
0.5pt Blum J, Wurm G, Poppe T, Heim L-O. 1998. Aspects of laboratory dust aggregation with relevance to the formation of planetesimals. Earth Moon Planets 80:285-309
0.5pt Boley AC. 2009. The two modes of gas-giant planet formation. Astrophys. J. 695:L53-57
0.5pt Boley AC, Hayfield T,Mayer L, Durisen RH. 2010. Clumps in the outer disk by disk instability: why they are initially gas giants and the legacy of disruption. Icarus 207:509-16
0.5pt Bonfils X, Forveille T, Delfosse X, Udry S, Mayor M. 2005. The HARPS search for southern extra-solar planets. VI. A Neptune-mass planet around the nearby M dwarf Gl 581. Astron. Astrophys. 443:L15-18
0.5pt Boss AP. 2000a. Formation of extrasolar giant planets: core accretion or disk instability? Earth Moon Planets 81:19-26
0.5pt Boss AP. 2000b. Possible rapid gas-giant planet formation in the solar nebula and other protoplanetary disks. Astrophys. J. 536:L101-4
0.5pt Boss AP. 2003. Rapid formation of outer giant planets by disk instability. Astrophys. J. 599:577-81
0.5pt Boss AP. 2006. Rapid formation of super-Earths around M dwarf stars. Astrophys. J. 644:L79-82
0.5pt Boss AP,Wetherill GW, Haghighipour N. 2002. Rapid formation of ice giant planets. Icarus 156:291-95
0.5pt Brauer F, Dullemond CP, Henning T. 2008a. Coagulation, fragmentation and radial motion of solid particles in protoplanetary disks. Astron. Astrophys. 480:859-77
0.5pt Brauer F, Henning T, Dullemond CP. 2008b. Planetesimal formation near the snow line in MRI-driven turbulent protoplanetary disks. Astron. Astrophys. 487:L1-4
0.5pt Bromley BC, Kenyon SJ. 2006. Terrestrial planet formation. I. The transition from oligarchic growth to chaotic growth. Astron. J. 131:1837-50
0.5pt Bromley BC, Kenyon SJ. 2011. A new hybrid N-body-coagulation code for the formation of gas giant planets. Astrophys. J. 731:101
0.5pt Cai K, Pickett MK, Durisen RH, Milne AM. 2010. Giant planet formation by disk instability: a comparison simulation with an improved radiative scheme. Astrophys. J. 716:L176-80
0.5pt Canup RM, Ward WR. 2002. Formation of the Galilean satellites: conditions of accretion. Astron. J. 124:3404-23
0.5pt Chambers JE. 2001. Making more terrestrial planets. Icarus 152:205-24
0.5pt Chambers JE. 2009. Planetary migration: What does it mean for planet formation? Annu. Rev. Earth Planet. Sci. 37:321-44
0.5pt Chambers JE. 2010. Planetesimal formation by turbulent concentration. Icarus 208:505-17
0.5pt Chambers JE, Cassen P. 2002. The effect of surface density profile and giant planet eccentricities on planetary accretion in the inner solar system. Meteorit. Planet. Sci. 37:1523-40
0.5pt Chambers JE, Wetherill GW. 1998. Making the terrestrial planets: N-body integrations of planetary embryos in three dimensions. Icarus 136:304-27
0.5pt Chambers JE, Wetherill GW. 2001. Planets in the asteroid belt. Meteorit. Planet. Sci. 36:381–99 Charbonneau D, Berta ZK, Irwin J, Burke CJ, Nutzman P, et al. 2009. A super-Earth transiting a nearby low-mass star. Nature 462:891-94
0.5pt Chen CH, Kamp I. 2004. Are giant planets forming around HR 4796A? Astrophys. J. 602:985-92
0.5pt Chiang E, Laughlin G. 2013. The minimum-mass extrasolar nebula: in-situ formation of close-in super-Earths. MNRAS 413:3444-55
0.5pt Chiang E, Youdin AN. 2010. Forming planetesimals in solar and extrasolar nebulae. Annu. Rev. Earth Planet. Sci. 38:493-522
0.5pt Clubb K, Fischer D, Howard A, Marcy G, Henry G. 2009. M2K: a search for planets orbiting early M and late K dwarf stars. Bull. Am. Astron. Soc. 41:192
0.5pt Cuzzi JN, Hogan RC, Shariff K. 2008. Towards planetesimals: dense chondrule clumps in the protoplanetary nebula. Astrophys. J. 687:1432-47
0.5pt Danby JMA. 1992. Fundamentals of Celestial Mechanics. Richmond, VA: Willmann-Bell. 2nd ed.
0.5pt Désert J-M, Bean J, Miller-Ricci Kempton E, Berta ZK, Charbonneau D, et al. 2011. Observational evidence for a metal-rich atmosphere on the super-Earth GJ1214b. Astrophys. J. 731:L40
0.5pt Dominik C, Tielens AGGM. 1997. The physics of dust coagulation and the structure of dust aggregates in space. Astrophys. J. 480:647-73
0.5pt Dressing CD, Charbonneau D. 2013. The occurrence rate of small planets around small stars. Astrophys. J. 767:95
0.5pt Dullemond CP, Dominik C. 2005. Dust coagulation in protoplanetary disks: a rapid depletion of small grains. Astron. Astrophys. 434:971-86
0.5pt Durisen RH, Boss AP, Mayer L, Nelson AF, Quinn T, Rice WKM. 2007. Gravitational instabilities in gaseous protoplanetary disks and implications for giant planet formation. In Protostars and PlanetsV, ed. B Reipurth, D Jewitt, K Keil, pp. 607-22. Tucson: Univ. Ariz. Press
0.5pt Fernandez JA, Ip W-H. 1984. Some dynamical aspects of the accretion of Uranus and Neptune-the exchange of orbital angular momentum with planetesimals. Icarus 58:109-20
0.5pt Fogg MJ, Nelson RP. 2005. Oligarchic and giant impact growth of terrestrial planets in the presence of gas-giant planet migration. Astron. Astrophys. 441:791-806
0.5pt Fogg MJ, Nelson RP. 2006. On the possibility of terrestrial planet formation in hot-Jupiter systems. Int. J. Astrobiol. 5:199-209
0.5pt Fogg MJ, Nelson RP. 2007a. On the formation of terrestrial planets in hot-Jupiter systems. Astron. Astrophys. 461:1195-208
0.5pt Fogg MJ, NelsonRP. 2007b. The effect of type I migration on the formation of terrestrial planets in hot-Jupiter systems. Astron. Astrophys. 472:1003-15
0.5pt Fogg MJ, Nelson RP. 2009. Terrestrial planet formation in low-eccentricity warm-Jupiter systems. Astron. Astrophys. 498:575-89
0.5pt Forveille T, Bonfils X, Delfosse X, Alonso R, Udry S, et al. 2011. The HARPS search for southern extrasolar planets. XXXII. Only 4 planets in GL 581 system. arXiv:1109.2505. http://arxiv.org/abs/1109.2505
0.5pt Fraine JD, Deming D, Gillon M, Jehin E, Demory B-O, et al. 2013. Spitzer transits of the super-Earth GJ 1214 b and implications for its atmosphere. Astrophys. J. 765:127
0.5pt Gibbons PG, Rice WKM, Mamatsashvili GR. 2012. Planetesimal formation in self-gravitating discs. MNRAS 426:1444-54
0.5pt Goldreich P. 1965. An explanation of the frequent occurrence of commensurable mean motions in the Solar System. MNRAS 130:159-81
0.5pt Goldreich P, Lithwick Y, Sari R. 2004. Final stages of planet formation. Astrophys. J. 614:497-507
0.5pt Gomes R, Levison HF, Tsiganis K, Morbidelli A. 2005. Origin of the cataclysmic Late Heavy Bombardment period of the terrestrial planets. Nature 435:466-69
0.5pt Gomes RS. 1997. Dynamical effects of planetary migration on the primordial asteroid belt. Astron. J. 114:396-401
0.5pt Greenberg RJ. 1973. Evolution of satellite resonances by tidal dissipation. Astron. J. 78:338-46
0.5pt Greenberg RJ, Counselman CC, Shapiro II. 1972. Orbit-orbit resonance capture in the Solar System. Science 178:747-49
0.5pt Greenberg R, Hartmann WK, Chapman CR, Wacker JF. 1978. Planetesimals to planets-numerical simulations of collisional evolution. Icarus 35:1-26
0.5pt Gressel O, Nelson RP, Turner NJ. 2012. Dead zones as safe havens for planetesimals: influence of disc mass and external magnetic field. MNRAS 422:1140-59
0.5pt Guillot T. 2005. The interiors of giant planets: models and outstanding questions. Annu. Rev. Earth Planet. Sci. 33:493-530
0.5pt Güttler C, Blum J, Zsom A, Ormel CW, Dullemond CP. 2009. The first phase of protoplanetary dust growth: the bouncing barrier. Geochim. Cosmochim. Acta 73:A482
0.5pt Haghighipour N. 2005. Growth and sedimentation of dust particles in the vicinity of local pressure enhancements in a solar nebula. MNRAS 362:1015-24
0.5pt Haghighipour N. 2011. Super-Earths: a new class of planetary bodies. Contemp. Phys. 52:403-38
0.5pt Haghighipour N, Boss AP. 2003a. On pressure gradients and rapid migration of solids in a nonuniform solar nebula. Astrophys. J. 583:996-1003
0.5pt Haghighipour N, Boss AP. 2003b. On gas drag–induced rapid migration of solids in a nonuniform solar nebula. Astrophys. J. 598:1301-11
0.5pt Haghighipour N, Jewitt D. 2008. A region void of irregular satellites around Jupiter. Astron. J. 136:909-18
0.5pt Haghighipour N, Rastegar S. 2011. Implications of the TTV-detection of close-in terrestrial planets around M stars for their origin and dynamical evolution. Proc. Haute-Provence Obs. Colloq., St. Michel l’Observatoire, France, Aug. 23–27, 2010, ed. F Bouchy, RF Diaz, C Moutou, 11:04004. Les Ulis, France: EPJ Web of Conferences. http://dx.doi.org/10.1051/epjconf/20101104004
0.5pt Haghighipour N, Vogt SS, Butler RP, Rivera EJ, Laughlin G, et al. 2010. The Lick-Carnegie Exoplanet Survey: a Saturn-mass planet in the habitable zone of the nearby M4V star HIP 57050. Astrophys. J. 715:271-76
0.5pt Hahn JM, Malhotra R. 2005. Neptune’s migration into a stirred-up Kuiper belt: a detailed comparison of simulations to observations. Astron. J. 130:2392-414
0.5pt Haisch KE Jr, Lada EA, Lada CJ. 2001. Disk frequencies and lifetimes in young clusters. Astrophys. J. Lett. 553:L153-56
0.5pt Hansen BMS. 2009. Formation of the terrestrial planets from a narrow annulus. Astrophys. J. 703:1131-40
0.5pt Hatzes AP, Dvorak R, Wuchterl G, Guterman P, Hartmann M, et al. 2010. An investigation into the radial velocity variations of CoRoT-7. Astron. Astrophys. 520:93-108
0.5pt Hatzes AP, Fridlund M, Nachmani G, Mazeh T, Valencia D, et al. 2011. The mass of CoRoT-7 b. Astrophys. J. 743:75
0.5pt Hayashi C. 1981. Structure of the solar nebula, growth and decay of magnetic fields and effects of magnetic and turbulent viscosities on the nebula. Prog. Theor. Phys. 70:S35-53
0.5pt Heng K, Vogt SS. 2011. Gliese 581g as a scaled-up version of Earth: atmospheric circulation simulations. MNRAS 415:2145-57
0.5pt Hubickyj O, Bodenheimer P, Lissauer JJ. 2005. Accretion of the gaseous envelope of Jupiter around a 5-10 Earth-mass core. Icarus 179:415-31
0.5pt Ida S, Makino J. 1993. Scattering of planetesimals by a protoplanet: slowing down of runaway growth. Icarus 106:210-27
0.5pt Irwin J, Charbonneau D, Nutzman P, Falco E. 2009a. The MEarth Project: searching for transiting habitable super-Earth planets around nearby M-dwarfs. Proc. 15th Cambridge Workshop Cool Stars Stellar Syst. Sun, St. Andrews, Scotl., July 21-25, 2008. AIP Conf. Proc. 1094:445-48. Melville, NY: AIP
0.5pt Irwin J, Charbonneau D, Nutzman P, Falco E. 2009b. The MEarth Project: searching for transiting habitable super-Earths around nearby M-dwarfs. Proc. Int. Astron. Union Symp., Boston, May 19-23, 2008. 253:37-43. Cambridge, UK: Cambridge Univ. Press
0.5pt Johansen A, Brauer F, Dullemond C, Klahr H, Henning T. 2008. A coagulation-fragmentation model for the turbulent growth and destruction of preplanetesimals. Astron. Astrophys. 486:597-611
0.5pt Johansen A, Klahr H,Henning T. 2006. Gravoturbulent formation of planetesimals. Astrophys. J. 636:1121-34
0.5pt Johansen A, Oishi JS, Mac Low M-M, Klahr H, Henning T, Youdin A. 2007. Rapid planetesimal formation in turbulent circumstellar disks. Nature 448:1022-25
0.5pt Johnson JA, Gazak JZ, Apps K, Muirhead PS, Crepp JR, et al. 2012. Characterizing the cool KOIs. II. The M dwarf KOI-254 and its hot Jupiter. Astron. J. 143:111
0.5pt Kato MT, Fujimoto M, Ida S. 2012. Planetesimal formation at the boundary between steady super/sub-Keplerian flow created by inhomogeneous growth of magnetorotational instability. Astrophys. J. 747:11-20
0.5pt Kennedy GM, Kenyon SJ. 2008a. Planet formation around stars of various masses: the snow line and the frequency of giant planets. Astrophys. J. 673:502-12
0.5pt Kennedy GM, Kenyon SJ. 2008b. Planet formation around stars of various masses: hot super-Earths. Astrophys. J. 682:1264-76
0.5pt Kennedy GM, Kenyon SJ, Bromley BC. 2006. Planet formation around low-mass stars: the moving snow line and super-Earths. Astrophys. J. 650:L139-42
0.5pt Kennedy GM, Kenyon SJ, Bromley BC. 2007. Planet formation around M-dwarfs: the moving snow line and super-Earths. Astrophys. Space Sci. 311:9-13
0.5pt Kenyon SJ, Bromley BC. 2009. Rapid formation of icy super-Earths and the cores of gas-giant planets. Astrophys. J. 690:L140-43
0.5pt Klahr HH, Henning T. 1997. Particle-trapping eddies in protoplanetary accretion disks. Icarus 128:213-29
0.5pt Kokubo E, Ida S. 1995. Orbital evolution of protoplanets embedded in a swarm of planetesimals. Icarus 114:247-57
0.5pt Kokubo E, Ida S. 1996. On runaway growth of planetesimals. Icarus 123:180-91
0.5pt Kokubo E, Ida S. 1998. Orbital oligarchic growth of protoplanets. Icarus 131:171–78
0.5pt Kokubo E, Ida S. 2000. Formation of protoplanets from planetesimals in the solar nebula. Icarus 143:15-27
0.5pt Kokubo E, Ida S. 2007. Formation of terrestrial planets from protoplanets. II. Statistics of planetary spin. Astrophys. J. 671:2082-90
0.5pt Kokubo E, Kominami J, Ida S. 2006. Formation of terrestrial planets from protoplanets. I. Statistics of basic dynamical properties. Astrophys. J. 642:1131-39
0.5pt Krause M, Blum J. 2004. Growth and form of planetary seedlings: results from a sounding rocket microgravity aggregation experiment. Phys. Rev. Lett. 93:021103
0.5pt Kretke KA, Lin DNC. 2007. Grain retention and formation of planetesimals near the snow line inMRI-driven turbulent protoplanetary disks. Astrophys. J. 664:L55-58
0.5pt Lada CJ, Lada EA. 2003. Embedded clusters in molecular clouds. Annu. Rev. Astron. Astrophys. 41:57-115
0.5pt Laughlin G, Bodenheimer P, Adams FC. 2004. The core-accretion model predicts few Jovian-mass planets orbiting red dwarfs. Astrophys. J. 612:L73-76
0.5pt Léger A, Rouan D, Schneider J, Barge P, Fridlund M, et al. 2009. Transiting exoplanets from the CoRoT space mission. VIII. CoRoT-7b: the first super-Earth with measured radius. Astron. Astrophys. 506:287-302
0.5pt Levison HF, Agnor C. 2003. The role of giant planets in terrestrial planet formation. Astron. J. 125:2692-713
0.5pt Lissauer JJ, Hubickyj O, D’Angelo G, Bodenheimer P. 2009. Models of Jupiter’s growth incorporating thermal and hydrodynamic constraints. Icarus 199:338-50
0.5pt Lyra W, Johansen A, Klahr H, Piskunov N. 2009a. Standing on the shoulders of giants. Trojan earths and vortex trapping in low mass self-gravitating protoplanetary disks of gas and solids. Astron. Astrophys. 493:1125-39
0.5pt Lyra W, Johansen A, Zsom A, Klahr H, Piskunov N. 2009b. Planet formation bursts at the borders of the dead zone in 2D numerical simulations of circumstellar disks. Astron. Astrophys. 497:869-88
0.5pt Maercker M, Burton MG, Right WCM. 2006. L-band (3.5 μm) IR-excess in massive star formation. II. RCW 57/NGC 3576. Astron. Astrophys. 450:253-63
0.5pt Malhotra R. 1993. The origin of Pluto’s peculiar orbit. Nature 365:819-21
0.5pt Malhotra R. 1995. The origin of Pluto’s orbit: implications for the Solar System beyond Neptune. Astron. J. 110:420-29
0.5pt Malhotra R. 1996. The phase space structure near Neptune resonances in the Kuiper belt. Astron. J. 111:504-16
0.5pt Mandell AM, Raymond SN, Sigurdsson S. 2007. Formation of Earth-like planets during and after giant planet migration. Astrophys. J. 660:823-44
0.5pt Mandell AM, Sigurdsson S. 2003. Survival of terrestrial planets in the presence of giant planet migration. Astrophys. J. 599:L111-14
0.5pt Massét F, Snellgrove M. 2001. Reversing type II migration: resonance trapping of a lighter giant protoplanet. MNRAS 320:L55-59
0.5pt Mayer L, Lufkin G, Quinn T, Wadsley J. 2007. Fragmentation of gravitationally unstable gaseous protoplanetary disks with radiative transfer. Astrophys. J. 661:L77-80
0.5pt Mayer L, Quinn T, Wadsley J, Stadel J. 2002. Formation of giant planets by fragmentation of protoplanetary disks. Science 298:1756-59
0.5pt Mayer L, Quinn T, Wadsley J, Stadel J. 2003. Simulations of unstable gaseous disks and the origin of giant planets. In ASP Conference Series. Vol. 294: Scientific Frontiers in Research on Extrasolar Planets, ed. D Deming, S Seager, pp. 281–86. San Francisco: Astron. Soc. Pac.
0.5pt Mayer L, Quinn T, Wadsley J, Stadel J. 2004. The evolution of gravitationally unstable protoplanetary disks: fragmentation and possible giant planet formation. Astrophys. J. 609:1045-64
0.5pt MayorM, Bonfils X, Forveille T, Delfosse X, Udry S, et al. 2009. The HARPS search for southern extrasolar planets. XVIII. An Earth-mass planet in the GJ 581 planetary system. Astron. Astrophys. 507:487-94
0.5pt Mayor M, Queloz D. 1995. A Jupiter-mass companion to a solar-type star. Nature 378:355-59
0.5pt McNeil DS, Nelson RP. 2010. On the formation of hot Neptunes and super-Earths. MNRAS 401:1691-708
0.5pt Menou K. 2012. Atmospheric circulation and composition of GJ 1214 b. Astrophys. J. 744:L16
0.5pt Michael S, Durisen RH, Boley AC. 2011. Migration of gas giant planets in gravitationally unstable disks. Astrophys. J. 737:L42
0.5pt Militzer B, Hubbard WB, Vorberger J, Tamblyn I, Bonev SA. 2008. A massive core in Jupiter predicted from first-principles simulations. Astrophys. J. 688:L45-48
0.5pt Miller-Ricci E, Fortney JJ. 2010. The nature of the atmosphere of the transiting super-Earth GJ 1214 b. Astrophys. J. Lett. 716:L74-79
0.5pt Miller-Ricci E, Seager S, Sasselov D. 2009. The atmospheric signatures of super-Earths: how to distinguish between hydrogen-rich and hydrogen-poor atmospheres. Astrophys. J. 690:1056-67
0.5pt Minton DA, Malhotra R. 2009. A record of planet migration in the main asteroid belt. Nature 457:1109-11
0.5pt Minton DA, Malhotra R. 2011. Secular resonance sweeping of the main asteroid belt during planet migration. Astrophys. J. 732:53
0.5pt Mizuno H, Markiewicz WJ, Völk HJ. 1988. Grain growth in turbulent protoplanetary accretion disks. Astron. Astrophys. 195:183-92
0.5pt Morbidelli A, Chambers J, Lunine JL, Petit JM, Robert F, et al. 2000. Source regions and timescales for the delivery of water to Earth. Meteorit. Planet. Sci. 35:1309-20
0.5pt Morbidelli A, Levison HF, Tsiganis K, Gomes R. 2005. Chaotic capture of Jupiter’s Trojan asteroids in the early Solar System. Nature 435:462-65
0.5pt Morbidelli A, Lunine JI, O’Brien DP, Raymond SN, Walsh KJ. 2012. Building terrestrial planets. Annu. Rev. Earth Planet. Sci. 40:251-75
0.5pt Movshovitz N, Bodenheimer P, Podolak M, Lissauer JJ. 2010. Formation of Jupiter using opacities based on detailed grain physics. Icarus 209:616-24
0.5pt Movshovitz N, Podolak M. 2008. The opacity of grains in protoplanetary atmospheres. Icarus 194:368-78
0.5pt Murray CD, Dermott SF. 1999. Solar System Dynamics. Cambridge, UK: Cambridge Univ. Press
0.5pt Nakagawa Y, Nakazawa K, Hayashi C. 1981. Growth and sedimentation of dust grains in the primordial solar nebula. Icarus 45:517-28
0.5pt Nakagawa Y, Sekiya M, Hayashi C. 1986. Settling and growth of dust particles in a laminar phase of a low-mass solar nebula. Icarus 67:375-90
0.5pt Nelson RP, Papaloizou JCB, Massét F, Kley W. 2001. The migration and growth of protoplanets in protostellar discs. MNRAS 318:18-36
0.5pt Nutzman P, Charbonneau D. 2008. Design considerations for a ground-based transit search for habitable planets orbiting M dwarfs. Publ. Astron. Soc. Pac. 120:317-27
0.5pt O’Brien DP, Morbidelli A, Levison HF. 2006. Terrestrial planet formation with strong dynamical friction. Icarus 184:39-58
0.5pt Okuzumi S, Tanaka H, Kobayashi H, Wada K. 2012. Rapid coagulation of porous dust aggregates outside the snow line: a pathway to successful icy planetesimal formation. Astrophys. J. 752:106-23
0.5pt O’Neill C, Lenardic A. 2007. Geological consequences of super-sized Earths. Geophys. Res. Lett. 34:L19204
0.5pt Ormel CW, Cuzzi JN. 2007. Closed-form expressions for particle relative velocities induced by turbulence. Astrophys. Astron. 466:413-20
0.5pt Pan L, Padoan P, Scalo J, Kritsuk AG, Norman ML. 2011. Turbulent clustering of protoplanetary dust and planetesimal formation. Astrophys. J. 740:6
0.5pt Papaloizou JCB, Terquem C. 2006. Planet formation and migration. Rep. Prog. Phys. 69:119-80
0.5pt Peale SJ, Lee MH. 2002. A primordial origin of the Laplace relation among the Galilean satellites. Science 298:593-97
0.5pt Podolak M. 2003. The contribution of small grains to the opacity of protoplanetary atmospheres. Icarus 165:428-37
0.5pt Pollack JB, Hubickyj O, Bodenheimer P, Lissauer JJ, Podolak M, Greenzweig Y. 1996. Formation of the giant planets by concurrent accretion of solids and gas. Icarus 124:62-85
0.5pt Queloz D, Bouchy F, Moutou C, Hatzes A, Hébrard G, et al. 2009. The CoRoT-7 planetary system: two orbiting super-Earths. Astron. Astrophys. 506:303-19
0.5pt Raymond SN, Barnes R, Mandell AM. 2008. Observable consequences of planet formation models in systems with close-in terrestrial planets. MNRAS 384:663-74
0.5pt Raymond SN, Mandell AM, Sigurdsson S. 2006a. Exotic earths: forming habitable worlds with giant planet migration. Science 313:1413-16
0.5pt Raymond SN, O’Brien DP, Morbidelli A, Kaib NA. 2009. Building the terrestrial planets: constrained accretion in the inner Solar System. Icarus 203:644-62
0.5pt Raymond SN, Quinn T, Lunine JI. 2004. Making other earths: dynamical simulations of terrestrial planet formation and water delivery. Icarus 168:1-17
0.5pt Raymond SN, Quinn T, Lunine JI. 2005a. Terrestrial planet formation in disks with varying surface density profiles. Astrophys. J. 632:670-76
0.5pt Raymond SN, Quinn T, Lunine JI. 2005b. The formation and habitability of terrestrial planets in the presence of close-in giant planets. Icarus 177:256-63
0.5pt Raymond SN, Quinn T, Lunine JI. 2006b. High-resolution simulations of the final assembly of Earth-like planets 1: terrestrial accretion and dynamics. Icarus 183:265-82
0.5pt Raymond SN, Quinn T, Lunine JI. 2007. High resolution simulations of the final assembly of Earth-like planets 2: water delivery and planetary habitability. Astrobiol. J. 7:66-84
0.5pt Rivera EJ, Laughlin G, Butler RP, Vogt SS, Haghighipour N, Meschiari S. 2010. TheLick-CarnegieExoplanet Survey: a Uranus-mass fourth planet for GJ 876 in an extrasolar Laplace configuration. Astrophys. J. 719:890-99
0.5pt Rivera EJ, Lissauer JJ, Butler RP, Marcy GW, Vogt SS, et al. 2005. A $\sim 7.5 {M_\oplus}$ planet orbiting the nearby star, GJ 876. Astrophys. J. 634:625-40
0.5pt Rogers LA, Bodenheimer P, Lissauer J, Seager S. 2011. Formation and structure of low-density exo-Neptunes. Astrophys. J. 738:59
0.5pt Rogers LA, Seager S. 2010a. A framework for quantifying the degeneracies of exoplanet interior compositions. Astrophys. J. 712:974-91
0.5pt Rogers LA, Seager S. 2010b. Three possible origins for the gas layer on GJ 1214b. Astrophys. J. 716:1208-16
0.5pt Roy AE. 1982. Orbital Motion. Bristol, UK: Adam Hilger
0.5pt Safronov VS. 1969. Evolution of Protoplanetary Cloud and Formation of the Earth and Planets. Moscow: Nauka
0.5pt Schlichting E, Warren PH, Yin Q-Z. 2012. The last stages of terrestrial planet formation: dynamical friction and the late veneer. Astrophys. J. 752:8-15
0.5pt Seager S, Deming D. 2009. On the method to infer an atmosphere on a tidally locked super Earth exoplanet and upper limits to GJ 876d. Astrophys. J. 703:1884-89
0.5pt Shariff K, Cuzzi JN. 2011. Gravitational instability of solids assisted by gas drag: slowing by turbulent mass diffusivity. Astrophys. J. 738:73
0.5pt Shi J-M, Chiang E. 2013. From dust to planetesimals: criteria for gravitational instability of small particles in gas. Astrophys. J. 764:20
0.5pt Smoluchowski MV. 1916. Drei Vortrage uber diffusion, brownsche bewegung und koagulation von Kolloidteilchen. Physik. Zeit. 17:557-85
0.5pt Sotin C, Schubert G. 2009. Mantle convection and plate tectonics on Earth-like exoplanets. Presented at Am. Geophys. Union Fall Meet., Dec. 14-18, San Francisco (Abstr. P42B-02)
0.5pt Strom SE, Edwards S, Skrutskie MF. 1993. Evolutionary time scales for circumstellar disks associated with intermediate- and solar-type stars. In Protostars and Planets III, ed. EH Levy, JI Lunine, pp. 837-66. Tucson: Univ. Ariz. Press
0.5pt Supulver KD, Lin DNC. 2000. Growth and sedimentation of dust grains in the primordial solar nebula. Icarus 146:525-40
0.5pt Swift J, Johnson JA, Morton TD, Crepp JR, Montet BT, et al. 2013. Characterizing the cool KOIs. IV.Kepler-32 as a prototype for the formation of compact planetary systems throughout the galaxy. Astrophys. J. 764:105
0.5pt Tackley PJ, van Heck H. 2009. Mantle convection, stagnant lids and plate tectonics on super-Earths. Geochim. Cosmochim. Acta 73:A1303
0.5pt Terquem C, Papaloizou JCB. 2007. Migration and the formation of systems of hot super-Earths and Neptunes. Astrophys. J. 654:1110-20
0.5pt Torres K, Winter OC, Izidoro A, Haghighipour N. 2013. A compound model for the origin of Earth’s water. Astrophys. J. 767:54
0.5pt Tsiganis K, Gomes R, Morbidelli A, Levison HF. 2005. Origin of the orbital architecture of the giant planets of the Solar System. Nature 435:459-61
0.5pt Udry S, Bonfils X, Delfosse X, Forveille T, Mayor M, et al. 2007. The HARPS search for southern extra-solar planets. XI. Super-Earths (5 and 8 $M_\oplus$) in a 3-planet system. Astron. Astrophys. 469:L43-47
0.5pt Valencia D, Ikoma M, Guillot T, Nettelmann N. 2010. Composition and fate of short-period super-Earths. The case of CoRoT-7b. Astron. Astrophys. 516:A20
0.5pt Valencia D, O’Conell RJ. 2009. Convection scaling and subduction on Earth and super-Earths. Earth Planet. Sci. Lett. 286:492-502
0.5pt Valencia D, O’Connell RJ, Sasselov DD. 2006. Internal structure of massive terrestrial planets. Icarus 181:545-54
0.5pt Valencia D, O’Connell RJ, Sasselov DD. 2007a. Inevitability of plate tectonics on super-Earths. Astrophys. J. 670:L45-48
0.5pt Valencia D, Sasselov DD, O’Connell RJ. 2007b. Detailed models of super-Earths: How well can we infer bulk properties? Astrophys. J. 665:1413-20
0.5pt Valencia D, Sasselov DD, O’Connell RJ. 2007c.Radius and structure models of the first terrestrial super-Earth planets. Astrophys. J. 656:545-51
0.5pt Vogt SS, Butler RO, Haghighipour N. 2012. GJ 581 update: additional evidence for a super-Earth in the habitable zone. Astron. Nachr. 333:561-75
0.5pt Vogt SS, Butler RP, Rivera EJ, Haghighipour N, Henry GW, Williamson MH. 2010. The Lick-Carnegie Exoplanet Survey: A $3.1 {M_\oplus}$ planet in the habitable zone of the nearby M3V star Gliese 581. Astrophys. J. 723:954-65
0.5pt Völk HJ, Jones FC, Morfill GE, Roeser S. 1980. Collisions between grains in a turbulent gas. Astron. Astrophys. 85:316-25
0.5pt Wada K, Tanaka H, Suyama T, Kimura H, Yamamoto T. 2007. Numerical simulation of dust aggregate collisions. I. Compression and disruption of two-dimensional aggregates. Astrophys. J. 661:320-33
0.5pt Walsh KJ, Morbidelli A, Raymond SN, O’Brien DP, Mandell AM. 2011. A low mass for Mars from Jupiter’s early gas-driven migration. Nature 475:206-9
0.5pt Weidenschilling SJ. 1977. Aerodynamics of solid bodies in the solar nebula. MNRAS 180:57-70
0.5pt Weidenschilling SJ. 1980. Growth and sedimentation of dust grains in the primordial solar nebula. Icarus 44:172-89
0.5pt Weidenschilling SJ. 2010. Particles in the nebular midplane: collective effects and relative velocities. Meteorit. Planet. Sci. 45:276-88
0.5pt Weidenschilling SJ, Spaute D, Davis DR, Marzari F, Ohtsuki K. 1997. Accretional evolution of a planetesimal swarm. Icarus 128:429-55
0.5pt Wetherill GW. 1990a. Comparison of analytical and physical modeling of planetesimal accumulation. Icarus 88:336-54
0.5pt Wetherill GW. 1990b. Formation of earth. Annu. Rev. Earth Planet. Sci. 18:205-56
0.5pt Wetherill GW. 1994. Provenance of the terrestrial planets. Geochim. Cosmochim. Acta 58:4513-20
0.5pt Wetherill GW. 1996. The formation and habitability of extra-solar planets. Icarus 119:219-38
0.5pt Wetherill GW, Stewart GR. 1989. Accumulation of a swarm of small planetesimals. Icarus 77:330-57
0.5pt Wetherill GW, Stewart GR. 1993. Formation of planetary embryos-effects of fragmentation, low relative velocity, and independent variation of eccentricity and inclination. Icarus 106:190-209
0.5pt Whipple FL. 1972. On certain aerodynamic processes for asteroid and comets. Proc. Nobel Symp., 21st, Saltsjöbaden, Sweden, Sept. 6-10, 1971, ed. A Elvius, pp. 211-32. New York: Wiley
0.5pt Wolszczan A, Frail DA. 1992. A planetary system around the millisecond pulsar PSR 1257+12. Nature 355:145-47
0.5pt Wurm G, Blum J. 1998. Experiments on preplanetary dust aggregation. Icarus 132:125-36
0.5pt Zhou J-L, Aarseth SJ, Lin DNC, Nagasawa M. 2005. Origin and ubiquity of short-period Earth-like planets: evidence for the sequential accretion theory of planet formation. Astrophys. J. 631:L85-88
0.5pt Zsom A, Ormel CW, Güttler C, Blum J, Dullemond CP. 2010. The first phase of protoplanetary dust growth: the bouncing barrier. Astron. Astrophys. 513:A57
[@lllllll]{} Planet & $M (M_\oplus$) & $P$ (day) & $a$ (AU)& $\>e$ & Stellar Type & ${M_*} (M_\odot)$\
KOI-55 c &$\>\>$ 0.6678 &$\>\>$ 0.34289 &$\>\>$ 0.0076 & - &$\>\>$ 0.496 &$\>\>$ sdB\
Kepler-42 d &$\>\>$ 0.954 &$\>\>$ 1.856169 &$\>\>$ 0.0154 & - &$\>\>$ 0.13 &$\>\>$ -\
Kepler-42 c &$\>\>$ 1.908 &$\>\>$ 0.45328509 &$\>\>$ 0.006 & - &$\>\>$ 0.13 &$\>\>$ -\
Gl 581 e &$\>\>$ 1.9398 &$\>\>$ 3.14945 &$\>\>$ 0.028 & 0.32 &$\>\>$ 0.31 &$\>\>$ M2.5V\
Kepler-11 f &$\>\>$ 2.301366 &$\>\>$ 46.68876 &$\>\>$ 0.25 & 0 &$\>\>$ 0.95 &$\>\>$ G\
HD 20794 c &$\>\>$ 2.4168 &$\>\>$ 40.114 &$\>\>$ 0.2036 & 0 &$\>\>$ 0.7 &$\>\>$ G8V\
HD 20794 b &$\>\>$ 2.703 &$\>\>$ 18.315 &$\>\>$ 0.1207 & 0 &$\>\>$ 0.7 &$\>\>$ G8V\
HD 215152 b &$\>\>$ 2.7666 &$\>\>$ 7.2825 &$\>\>$ 0.0652 & 0.34 &$\>\>$ - &$\>\>$ K0\
Kepler-42 b &$\>\>$ 2.862 &$\>\>$ 1.2137672 &$\>\>$ 0.0116 & - &$\>\>$ 0.13 &$\>\>$ -\
HD 215152 c &$\>\>$ 3.0846 &$\>\>$ 10.866 &$\>\>$ 0.0852 & 0.38 &$\>\>$ - &$\>\>$ K0\
Kepler-20 e &$\>\>$ 3.0846 &$\>\>$ 6.098493 &$\>\>$ 0.0507 & &$\>\>$ 0.912 &$\>\>$ G8\
MOA-2007-BLG &$\>\>$ 3.18 &$\>\>$ - &$\>\>$ 0.66 & - &$\>\>$ 0.06 &$\>\>$ M\
$\qquad\quad$-192-L b\
Kepler-32 b &$\>\>$ 3.4 &$\>\>$ 5.90 &$\>\>$ 0.0519 & - &$\>\>$ 0.54&$\>\>$ M1V\
HD 85512 b &$\>\>$ 3.498 &$\>\>$ 58.43 &$\>\>$ 0.26 & 0.11 &$\>\>$ 0.69 &$\>\>$ K5V\
HD 39194 b &$\>\>$ 3.7206 &$\>\>$ 5.6363 &$\>\>$ 0.0519 & 0.2 &$\>\>$ - &$\>\>$ K0V\
Kepler-32 c &$\>\>$ 3.8 &$\>\>$ 8.75 &$\>\>$ 0.067 & - &$\>\>$ 0.54 &$\>\>$ M1V\
PSR 1257 +12 d &$\>\>$ 3.816 &$\>\>$ 98.2114 &$\>\>$ 0.46 & 0.025&$\>\>$ - &$\>\>$ -\
PSR 1257 +12 c &$\>\>$ 4.134 &$\>\>$ 66.5419 &$\>\>$ 0.36 & 0.018&$\>\>$ - &$\>\>$ -\
HD 156668 b &$\>\>$ 4.1658 &$\>\>$ 4.646 &$\>\>$ 0.05 & 0 &$\>\>$ 0.772 &$\>\>$ K3V\
HD 40307 b &$\>\>$ 4.1976 &$\>\>$ 4.3115 &$\>\>$ 0.047 & 0 &$\>\>$ 0.77 &$\>\>$ K2.5V\
GJ 667C c &$\>\>$ 4.2612 &$\>\>$ 28.13 &$\>\>$ 0.1251 & 0.34 &$\>\>$ 0.33 &$\>\>$ M1.5V\
Kepler-11 b &$\>\>$ 4.30254 &$\>\>$ 10.30375 &$\>\>$ 0.091 & 0 &$\>\>$ 0.95 &$\>\>$ G\
KOI-55 b &$\>\>$ 4.452 &$\>\>$ 0.2401 &$\>\>$ 0.006 & - &$\>\>$ 0.496 &$\>\>$ sdB\
Kepler-10 b &$\>\>$ 4.5474 &$\>\>$ 0.837495 &$\>\>$ 0.01684 & 0 &$\>\>$ 0.895 &$\>\>$ G\
HD 20794 d &$\>\>$ 4.77 &$\>\>$ 90.309 &$\>\>$ 0.3499 & 0 &$\>\>$ 0.7 &$\>\>$ G8V\
CoRoT-7 b &$\>\>$ 4.8018 &$\>\>$ 0.853585 &$\>\>$ 0.0172 & 0 &$\>\>$ 0.93 &$\>\>$ K0V\
61 Vir b &$\>\>$ 5.088 &$\>\>$ 4.215 &$\>\>$ 0.050201 & 0.12 &$\>\>$ 0.95 &$\>\>$ G5V\
HD 39194 d &$\>\>$ 5.1516 &$\>\>$ 33.941 &$\>\>$ 0.172 & 0.2 &$\>\>$ - &$\>\>$ K0V\
HD 136352 b &$\>\>$ 5.2788 &$\>\>$ 11.577 &$\>\>$ 0.0933 & 0.18 &$\>\>$ - &$\>\>$ G4V\
Gl 581 c &$\>\>$ 5.406 &$\>\>$ 12.9182 &$\>\>$ 0.073 & 0.07 &$\>\>$ 0.31 &$\>\>$ M2.5V\
[@lllllll]{} Planet & $M (M_\oplus$) & $P$ (day) & $a$ (AU)& $\>e$ & Stellar Type & ${M_*} (M_\odot)$\
OGLE-2005-390L b &$\>\>$ 5.406 &$\>\>$ 3500 &$\>\>$ 2.1 & - &$\>\>$ 0.22 &$\>\>$ M\
GJ 667C b &$\>\>$ 5.46324 &$\>\>$ 7.199 &$\>\>$ 0.0504 & 0.09 &$\>\>$ 0.33 &$\>\>$ M1.5V\
GJ 433 b &$\>\>$ 5.7876 &$\>\>$ 7.3709 &$\>\>$ 0.058 & 0.08 &$\>\>$ 0.48 &$\>\>$ M1.5\
HD 1461 c &$\>\>$ 5.9148 &$\>\>$ 13.505 &$\>\>$ 0.1117 & 0 &$\>\>$ 1.08 &$\>\>$ G0V\
HD 39194 c &$\>\>$ 5.9466 &$\>\>$ 14.025 &$\>\>$ 0.0954 & 0.11 &$\>\>$ - &$\>\>$ K0V\
Gl 581 d &$\>\>$ 6.042 &$\>\>$ 66.64 &$\>\>$ 0.22 & 0.25 &$\>\>$ 0.31 &$\>\>$ M2.5V\
Kepler-11 d &$\>\>$ 6.10242 &$\>\>$ 22.68719&$\>\>$ 0.159 & 0 &$\>\>$ 0.95 &$\>\>$ G\
HD 154088 b &$\>\>$ 6.1374 &$\>\>$ 18.596 &$\>\>$ 0.1316 & 0.38 &$\>\>$ - &$\>\>$ K0IV\
GJ 1214 b &$\>\>$ 6.36 &$\>\>$ 1.58040482 &$\>\>$ 0.014 & 0.27 &$\>\>$ 0.153 &$\>\>$ M\
HD 215497 b &$\>\>$ 6.36 &$\>\>$ 3.93404 &$\>\>$ 0.047 & 0.16 &$\>\>$ 0.87 &$\>\>$ K3V\
HD 97658 b &$\>\>$ 6.36 &$\>\>$ 9.4957 &$\>\>$ 0.0797 & 0.13 &$\>\>$ 0.85 &$\>\>$ K1V\
Gl 876 d &$\>\>$ 6.678 &$\>\>$ 1.93778 &$\>\>$ 0.0208 & 0.21 &$\>\>$ 0.334 &$\>\>$ M4 V\
HD 40307 c &$\>\>$ 6.8688 &$\>\>$ 9.62 &$\>\>$ 0.081 & 0 &$\>\>$ 0.77 &$\>\>$ K2.5V\
Kepler-18 b &$\>\>$ 6.9006 &$\>\>$ 3.504725 &$\>\>$ 0.0447 & - &$\>\>$ 0.972 &$\>\>$ -\
GJ 3634 b &$\>\>$ 6.996 &$\>\>$ 2.64561 &$\>\>$ 0.0287 & 0.08 &$\>\>$ 0.45 &$\>\>$ M2.5\
Kepler-9 d &$\>\>$ 6.996 &$\>\>$ 1.592851 &$\>\>$ 0.0273 & - &$\>\>$ 1 &$\>\>$ -\
HD 181433 b &$\>\>$ 7.5684 &$\>\>$ 9.3743 &$\>\>$ 0.08 & 0.39 &$\>\>$ 0.78 &$\>\>$ K3IV\
HD 1461 b &$\>\>$ 7.6002 &$\>\>$ 5.7727 &$\>\>$ 0.063 & 0.14 &$\>\>$ 1.08 &$\>\>$ G0V\
HD 93385 b &$\>\>$ 8.3634 &$\>\>$ 13.186 &$\>\>$ 0.1116 & 0.15 &$\>\>$ - &$\>\>$ G2V\
CoRoT-7 c &$\>\>$ 8.3952 &$\>\>$ 3.698 &$\>\>$ 0.046 & 0 &$\>\>$ 0.93 &$\>\>$ K0V\
Kepler-11 e &$\>\>$ 8.40474 &$\>\>$ 31.9959&$\>\>$ 0.194 & 0 &$\>\>$ 0.95 &$\>\>$ G\
GJ 176 b &$\>\>$ 8.427 &$\>\>$ 8.7836 &$\>\>$ 0.066 & 0 &$\>\>$ 0.49 &$\>\>$ M2.5V\
55 Cnc e &$\>\>$ 8.586 &$\>\>$ 0.7365449 &$\>\>$ 0.0156 & 0.06 &$\>\>$ 0.905 &$\>\>$ K0IV-V\
Kepler-20 b &$\>\>$ 8.586 &$\>\>$ 3.6961219 &$\>\>$ 0.0453 & 0.32 &$\>\>$ 0.912 &$\>\>$ G8\
HD 96700 b &$\>\>$ 9.0312 &$\>\>$ 8.1256 &$\>\>$ 0.0774 & 0.1 &$\>\>$ - &$\>\>$ G0V\
HD 40307 d &$\>\>$ 9.1584 &$\>\>$ 20.46 &$\>\>$ 0.134 & 0 &$\>\>$ 0.77 &$\>\>$ K2.5V\
HD 7924 b &$\>\>$ 9.222 &$\>\>$ 5.3978 &$\>\>$ 0.057 & 0.17 &$\>\>$ 0.832 &$\>\>$ KOV\
HD 134606 b &$\>\>$ 9.2856 &$\>\>$ 12.083 &$\>\>$ 0.102 & 0.15 &$\>\>$ - &$\>\>$ G6IV\
HD 136352 d &$\>\>$ 9.54 &$\>\>$ 106.72 &$\>\>$ 0.411 & 0.43 &$\>\>$ - &$\>\>$ G4V\
HD 189567 b &$\>\>$ 10.0488 &$\>\>$ - &$\>\>$ 14.275 & 0.11 &$\>\>$ 0.23 &$\>\>$ G2V\
HD 93385 c &$\>\>$ 10.1124 &$\>\>$ - &$\>\>$ 46.025 & 0.21 &$\>\>$ 0.24 &$\>\>$ G2V\
[^1]: http://www.eso.org/sci/facilities/lasilla/instruments/harps/
[^2]: http://www.exoplanets.org/cps.html
[^3]: http://www.cfa.harvard.edu/MEarth/Welcome.html
[^4]: http://smsc.cnes.fr/COROT/index.htm
[^5]: http://kepler.nasa.gov/
[^6]: Orbital commensurability is necessary for two planets to be in a mean-motion resonance; however, it is not sufficient. Other constraints have to exist between the angular elements of their orbits as well. For more details, the reader is referred to Roy (1982), Danby (1992), and Murray & Dermott (1999).
[^7]: I do not discuss these mechanisms here, as they may not be entirely relevant to the formation and dynamical evolution of super-Earths. Instead, I refer the reader to numerous articles that have been published on these subjects. Unfortunately, the richness of the literature does not allow me to cite all these articles here, but among them, one can refer to Nelson et al. (2001), Massét & Snellgrove (2001), Papaloizou & Terquem (2006), Chambers (2009), Armitage (2010), and a recent review by Baruteau & Massét (2013).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present an extension of the time-dependent Density Matrix Renormalization Group (t-DMRG), also known as Time Evolving Block Decimation algorithm (TEBD), allowing for the computation of dynamically important excited states of one-dimensional many-body systems. We show its practical use for analyzing the dynamical properties and excitations of the Bose-Hubbard model describing ultracold atoms loaded in an optical lattice from a Bose-Einstein condensate. This allows for a deeper understanding of nonadiabaticity in experimental realizations of insulating phases.'
author:
- Mateusz Łcki
- Dominique Delande
- Jakub Zakrzewski
title: 'Numerical Computation of Dynamically Important Excited States of Many-Body Systems'
---
Introduction
============
Properties of many-body quantum systems are difficult to compute, because the eigenstates are usually unknown and the large size of the Hilbert space makes even approximate descriptions difficult to manipulate. Some numerical methods - such as variational methods or mean-field theories - are intrinsically approximate yielding simple expressions describing most of the interesting physical properties, especially in the thermodynamic limit. The so-called “quasi-exact” methods do not have such restrictions, and are supposed to converge to the exact quantum result when some parameter tends to infinity. Those include exact diagonalization methods and quantum Monte-Carlo methods [@cep86; @bauer11] (the limiting parameter being the statistical sampling). For one-dimensional (1D) systems, particularly effective is the Density Matrix Renormalization Group (DMRG) approach [@white92; @hallberg06]. The states, in modern implementations of the method, are represented by the so called Matrix Product States (MPS) (for a recent smooth introduction, see [@scholl]). In numerous quantum systems, the ground state is very well approximated by a MPS, allowing its efficient description with a relatively small number of parameters, which, moreover, does not increase exponentially with the system size. To some extent, DMRG shares the advantages of both variational and quasi-exact methods.
While the original application of these methods considered stationary properties, recent years brought a big progress in addressing the dynamics. Arguably, it was the formulation of the Time Evolving Block Decimation (TEBD) algorithm [@vidal03], enabling study of real time dynamics of 1D systems of a reasonable size, which triggered a rapid progress in the field. Soon, a time-dependent DMRG (t-DMRG) method was proposed [@daley; @white04] with a fast realization of a close equivalence between TEBD and t-DMRG. Various applications of the method addressed dynamics of quantum phase transitions [@clark04; @cuciap; @zakrzewski09], quantum quenches and thermalizations [@anna; @bernier11; @biroli11; @bernier12], periodic driving [@kollath06; @huo11], to give just a few examples.
As soon as the evolution of the system ceases to be adiabatic, the excited states become important in the dynamics. The global properties of whole groups of excited states may be obtained from evaluation of e.g. the spectral function. Following the pioneering work of Hallberg [@hallberg95], a lot of effort has been put in developing efficient schemes for spectral functions evaluations (see e.g. [@white08; @pereira09; @barthel09; @dargel11; @holzner11] and references therein).
In the present paper, our aims are somehow complementary. We show how one may obtain selected, dynamically important excited eigenstates of strongly correlated systems in a relatively simple way. Instead of addressing spin chain systems, as typical for tests of various DMRG improvements, we shall immediately turn our attention to cold atoms settings and consider a Bose-Hubbard model as realized by cold atoms in 1D optical lattice potentials. In [@zakrzewski09], we have shown on such a specific example how to extract the energies of the few lowest excited states of the system, with, however, no access to their properties. In the present paper, we explain how to use the TEBD (or t-DMRG) method in order to extract some of the excited states with quite good accuracy. Those states are expressed also in terms of MPS making the calculations of expectations values and correlation functions easy. While the area law theorem [@area] assures that the ground state may be efficiently represented in MPS representation this is by no means guaranteed for a generic excited state. Our method will work only for those states for which such an efficient MPS representation exists. We shall show, however, that this restriction leaves plenty of interesting ground to explore as, most often, many excited states up to a reasonably large excitation energy, are also efficiently expressed as MPSs. Interestingly, the method selects states that are important in the systems’ dynamics. It seems more suited for low lying excited states (although we give other examples too). Still, in many-body systems, the density of states increases usually very fast with energy, and even relatively small systems (such as e.g. encountered in ultracold atomic gases) have plenty of excited states lying not far above the ground state energy.
While we use as an example the Bose-Hubbard (BH) model describing an ultra-cold atomic bosonic gas loaded in a one-dimensional optical lattice, the method proposed is quite general and can be used for any system where TEBD (t-DMRG) works for sufficiently long times. The latter restriction is necessary since the method is based on Fourier Transform techniques with resolution being dependent on the integration time. We shall show later that the method has an unexpected self purifying property allowing sometimes to use much longer integration times than anticipated in standard applications of the algorithm [@vidal03; @daley]. Possible future improvements of the method are discussed in the concluding section.
In the following sections, we first describe the method in some details. Then, we apply it for extracting selected excited states populated either by the periodic driving or by the non adiabatic dynamics in the disordered Bose-Hubbard system. The illustrative examples of excited states show the power and the limitations of our approach.
Extraction of eigenstates by Fourier Transform
==============================================
The basic idea is to start from an initial state which is a “wavepacket”, that is a superposition of various eigenstates, described by a MPS, to propagate it in real time with a time-independent Hamiltonian, finally to combine the MPS states at various times to reconstruct the excited states. The evolution of a state by a time-independent Hamiltonian with eigenbasis [$|e_i\rangle$]{} is given by $|\psi(t)\rangle = \sum_i \exp(-i E_i t/\hbar) c_i |e_i\rangle.$ where $c_i=\langle e_i|\psi(0)\rangle,$ The Fourier transform (FT) of the autocorrelation function $C(t)=\langle \psi(0)|\psi(t)\rangle = \sum_i |c_i|^2 \exp(-i E_i t/\hbar)$ over a time interval $T$ is the autocorrelation spectrum: $$\tilde{C}_T(E) = \frac{1}{T}\!\int_{-T/2}^{T/2}{\!\!\!\!{\mathrm e}^{\frac{i E t}{\hbar}} C(t) dt} = \sum_i{|c_i|^2\
\mathrm{sinc}\frac{(E-E_i)T}{2\hbar}}
\label{eq:ce}$$ where $\mathrm{sinc}\,x=\sin x/x.$ In the limit of long time $T$, it yields narrow peaks at the $E_i$’s with weights $|c_i|^2,$ that allows to determine the energy spectrum as shown in [@zakrzewski09].
Here we show that it is actually possible to extract from the dynamics also the excited eigenstates with large overlap - those that contribute most to the dynamical wavepacket. We perform a FT [*directly*]{} on the many-body wavefunction $|\psi(t)\rangle:$ $$|\phi_T(E)\rangle = \frac{1}{T}\! \int_{-T/2}^{T/2}{\!\!\!\!{\mathrm e}^{\frac{i E t}{\hbar}}|\psi(t)\rangle dt}
= \sum_i c_i\ \mathrm{sinc}\frac{(E-E_i)T}{2\hbar} |e_i\rangle
\label{eq:phie}$$ For long $T, |\phi_T(E_i)\rangle \to c_i|e_i\rangle,$ providing us with the targeted eigenstate. The method selects excited states *relevant for the dynamics*, although myriads of other states may be present in the same spectral region. This paves the way to study the properties of individual excited many-body states.
To evaluate $|\phi_T(E_i)\rangle,$ the integral is approximated by a discrete series. To sum the series, a procedure to perform the sum of two MPSs as a MPS is required, which we now briefly sketch. Any state $|\psi\rangle$ of a 1D system of M sites may be expressed in the MPS representation [@vidal03; @verstraete]: $$|\psi\rangle\! = \!\!\!\!\!\!\sum\limits_{\genfrac{}{}{0pt}{}{\alpha_1,\ldots,\alpha_M}{i_1,\ldots,i_M}}\!\!\!\!\!
\Gamma_{1\alpha_1}^{[1]
,i_1}\lambda^{[1]}_{\alpha_1}\Gamma^{[2],i_2}_{\alpha_1\alpha_2}\ldots\Gamma^{[M],i_M}_{\alpha_{n-1}1} |i_1,\dots,i_M\rangle
=:\mathcal{M}(\Gamma,\lambda)
\label{eqn:MPS}$$ where $\Gamma^{[l],i_l}$ are matrices and $\lambda^{[l]}$ vectors and $|i_j\rangle$ span a local Hilbert space on site $j$. When the $\Gamma$’s and the $\lambda$’s satisfy simple orthogonality relations detailed in [@shividal], their physical interpretation is simple. For example, the spectrum of the reduced density matrix associated with bipartite splitting $L:R = \{1,\ldots,l\}:\{l+1,\ldots,M\}$, $\text{Tr}_R |\psi \rangle \langle \psi|$ is nothing but $\left(\lambda^{[l]}_{\alpha_l}\right)^2$ for $\alpha_l=1,2\ldots.$ The corresponding entanglement entropy given by $$S_l = - \sum_{\alpha}{ (\lambda^{[l]}_{\alpha})^2 \ln(\lambda^{[l]}_{\alpha})^2 },
\label{eq:ent_ent}$$ is an important tool to characterize the strongly correlated systems, e.g. topological phases [@kitaev06]. The collection of $\lambda$ values forms the entanglement spectrum.
To describe exactly a generic state in terms of a MPS, a large number (exponentially increasing with $M$) of $\alpha_l$ values is needed. However, quite often, physical states created in many experiments are only slightly entangled so that $\lambda^{[l]}_{\alpha_l=1,2\ldots}$ are rapidly decaying numbers, which allows for introduction of a rather small cutoff $\chi$ in all sums above, resulting in tractable numerical computation [@vidal03]. Our approach is limited to such states only.
Following [@mcculloch] one has that: $\mathcal{M}(\Gamma,\lambda)+\mathcal{M}(\Gamma',\lambda')=\mathcal{M}(\Gamma\oplus\Gamma',\lambda\oplus\lambda').$ The result does not satisfy orthogonality relations that are vital for achieving efficient MPS algorithms. The algorithm to restore them [@shividal] ensures that the sum is stored in a memory-efficient way. Applying the above two-step algorithm to perform elementary addition suffices to obtain $|\psi_T(E_i)\rangle.$ The technical details are presented in the Appendix.
Our goal is to obtain an eigenstate as pure as possible, not merely to detect its presence. This requires long time integration which may affect the accuracy of the TEBD algorithm [@vidal03; @daley]. The integration time may be significantly shortened using few simple ideas which we now describe.
For simplicity let us restrict to a subspace spanned by two eigenstates $|\psi\rangle= \alpha |e_0\rangle + \beta |e_1\rangle$ as a special case. Performing the Fourier Transform as in eq. (\[eq:phie\]) gives: $$|\psi_T(E_0)\rangle = \alpha |e_0\rangle + \beta(T) |e_1\rangle,
\label{eq:psie2}$$ with $\beta(T)=\beta\ \mathrm{sinc}\frac{T\Delta E}{2 \hbar}, \Delta E=E_1 -E_0.$ Clearly $\beta(+\infty)=0,$ therefore large enough $T$ guarantees convergence of $|\psi_T(E_0)\rangle $ to $|e_0\rangle$. Another possibility is to choose $T$ such that $\beta(T)=0,$ i.e. a multiple of $\frac{4\pi \hbar}{\Delta}.$ This ensures that $|\psi_T(E_0)\rangle=|e_0\rangle$ despite relatively short evolution time. When other eigenstates contribute significantly to the wavepacket, there is no possibility to choose such a time $T$ that contributions from other eigenstates vanish simultaneously. However, if for some state $|\psi \rangle,$ and its eigenstate $|e_{i_0}\rangle$ there exists an eigenstate $|e_{i_1}\rangle$ with nonzero overlap on $|\psi\rangle,$ such that $|E_{i_0}- E_{i_1}|
\ll |E_{i_0}-E_j|$ if $j\neq i_0,i_1,$ then choosing $T$ as a multiple of $\frac{4\pi \hbar}{\Delta E}$ is an optimal choice.
If $\beta\gg \alpha$ in eq. (\[eq:psie2\]), a realization of $\beta(T) \ll 1$ is more difficult. One solution is to restart the Fourier Transform evaluation from a partially converged result. Let us first define a map $f_{E,T}(\psi)=\psi_T(E).$ One may verify that for a two-level system, the $n$ fold composition of $f_{E,T}$ satisfies:
$$f^n_{E,T}(\psi) = f(f^{n-1}_{E,T}(\psi))= \alpha |e_0\rangle + \beta \left(\mathrm{sinc}\frac{\Delta E T}{2\hbar}\right)^n |e_1\rangle$$ This requires performing evolution for time $nT$ and results in a power-$n$ decay of $\langle e_1| \psi \rangle/\langle e_0 | \psi\rangle.$ This observation remains true in the full Hilbert space with almost no modification.
Another way to obtain the power-$n$ decay of other eigenstates is to use a proper [*window function*]{} [@harris]. Instead of performing FT as in (\[eq:phie\]), one calculates $$|\phi_T^w(E)\rangle = \frac{1}{T}\! \int_{-T/2}^{T/2}{\!\!\!\!{\mathrm e}^{\frac{i E t}{\hbar}}|\psi(t)\rangle w_T(t) dt}
\label{eq:phiw}$$ with $w_T(t)=0.5\left(1-\cos\left(\frac{2 \pi t}{T}\right)\right), $ called a Hahn window, being one possibility. This window gives a $1/T^2$ convergence (instead of $1/T$ without a window) of $\beta(T).$
Further possible improvements will be apparent when discussing specific examples below.
Bose-Hubbard model
==================
We now illustrate the method using the Bose-Hubbard (BH) model [@fisher89; @jaksch98] as realized in a gas of ultracold atoms in an optical lattice. We are aware of limitations of the tight-binding Bose Hubbard description of the system for deep optical lattices [@dutta11; @sengst12]. The BH model provides, however, a realistic and commonly used simplification of the problem. A pioneer experiment [@greiner02] observed a quantum phase transition (QPT) from a superfluid (SF) phase to a Mott insulator (MI), and stimulated research on interacting many-body systems with external perturbations, for a review see [@lewen2007]. A similar QPT was observed in a 1D realization of the BH model [@stoeferle04]. A $^{87}$Rb condensate was loaded in a deep two-dimensional ”transverse” optical lattice potential realizing an array of one-dimensional atomic tubes. A shallower optical potential along the tubes coming from a pair of counter-propagating beams $V/E_R=s \sin^2(2\pi x/\lambda)$ (in units of the recoil energy $E_R=h^2/2m_{Rb}\lambda^2$) was superimposed on the system. After preparing the system at different potential depth, the lattice was modulated periodically at different frequencies measuring the absorbed energy (for details see [@stoeferle04]). In another realization of this essentially 1D experiment, the effects created by disorder were analyzed [@fallani07]. We shall consider both these situations in the following.
Periodic lattice oscillations
-----------------------------
Let us consider first the excitations created by a periodic modulation of the lattice. The problem has been addressed in [@kollath06] where the energy gain due to such a modulation was studied using a 1D Bose-Hubbard model in a parabolic trap and TEBD for the time evolution, obtaining results which were in a qualitative agreement with the experimental spectra [@stoeferle04]. Recently, a novel extended study appeared [@huo11] which analyzes in detail different possible excitations in the same model using both t-DMRG and a perturbative approach. This later calculation considers smaller and simpler system so, for pedagogical reasons, we start our analysis for cases discussed in [@huo11].
The 1D Bose-Hubbard Hamiltonian reads: $$\!H\! =\! -J \sum_{j}^{M-1} b_j^\dagger b_{j+1} + h.c. +\! \frac{U}{2} \sum_{j}^M n_j
\left(n_j - 1 \right) +\! \sum_{j}^M \epsilon_j n_j,$$ where $M$ is the number of sites, $b_j$ ($b_j^\dagger$) is the destruction (creation) operator of an atom at site $j,$ $n_j=b_j^\dagger b_j$ the number operator. The tunneling amplitude $J$ and the interaction energy $U$ depend on the lattice depth $s$ (see below), $\epsilon_j=c(j-r_0)^2$ is the energy offset of a given site, due to the harmonic trap with $r_0$ the position of its minimum. When the lattice depth is modulated, $s(t)=s_0[1+\delta \cos(\omega t)],$ the energy absorption is enhanced at frequencies leading to resonant excitation of some excited states.
The time-dependence of $s(t)$ maps onto the corresponding time dependence of the tunneling amplitude $J(t)$ and interaction strength $U(t)$ using, as in [@kollath06; @huo11], the approximate formulae due to Zwerger [@zwerger03]: J(t)=s(t)\^(-2) and U(t) = 4 s\^[1/4]{}(t)s\^[1/2]{}\_ where recoil energy units are used and $s_{\perp}$ is the depth of “transverse” potential creating the 1D tubes. Following [@huo11], we denote by $U_0$ the value of $U$ in the absence of modulation ($\delta=0$) and by $J_0$ the corresponding tunneling amplitude. Also, we take $s_{\perp}=30$, $a_s=5.45$nm, $\lambda=825$nm and the curvature of the harmonic trap $c=0.0123$ as in [@huo11].
Consider first a system of $N=36$ particles (corresponding to Fig. 2 of [@huo11]) in a deep lattice with $s_0=15$. The corresponding characteristic interaction energy is $U_0\approx 0.714,$ while the tunneling amplitude is $J_0/U_0\approx 0.01$.
![(color online) Absorption spectrum obtained for a 8% modulation (i.e. $\delta=0.08$), after integration time $t=100$ (smooth red line) and $t=300$ (filled circles connected by the black line). The data for $t=100$ are multiplied by three. The higher resolution obtained for longer times allows partially to separate various components due to individual excited states. Circles correspond to numerical data, black line connects them to guide the eye. The inset shows the average occupation numbers of the initial ground state of the system. []{data-label="fig:ts3c"}](abs_ts3cg.eps){width="0.9\columnwidth"}
For any state in MPS representation - be it the ground state, an excited state or a wavepacket - is it easy to compute expectation values of simple operators, such as the average occupation number $\langle n_l \rangle$ at site $l,$ as well as its standard deviation $\Delta n_l=\sqrt{\langle n_l^2\rangle-\langle
n_l\rangle^2}.$ The ground state of the system is obtained by the TEBD algorithm with imaginary time evolution. The average occupation number of sites, displayed in the inset of Fig. \[fig:ts3c\], shows two Mott plateaus with single and double occupancy. By simulating a periodic driving of the lattice using a real time TEBD evolution, we find the absorption spectrum. We take the modulation to be 8%, that is 8 times bigger than in [@huo11]; this creates bigger excitation in the system, making extraction of the excited states easier. For the same purpose, we shift slightly the lattice and the bottom of the harmonic trap taking $r_0=18.62345$. This breaks the reflection symmetry of the system with respect to the center of the trap, making the analysis and visualization of the excitations easier. Still, the spectrum for shorter evolution time, $t=100$ (for $\hbar=1,$ the unit of time is the inverse of the recoil energy) resembles the perturbative theory curve in Fig. 2 of [@huo11]. Longer integration time $t=300$ gives a better frequency resolution and several sharp peaks emerge. The absorption for time $t=100$ multiplied by a factor of 3 seems to reproduce quite nicely the global amplitude of the broad structure around $\omega=U_0,$ showing that, for such times, absorption in this frequency region is linear. This is not true for isolated peaks at low frequency as well as for the peak around $\omega=1.75U_0,$ suggesting that these peaks correspond to isolated single levels (with the corresponding Rabi-type evolution). The integration is performed with time step $\Delta t=0.02$ using a standard second order Trotter approach assuming maximal occupation at the sites $i_{max}=6$ and $\chi=50$ (those values are highly sufficient for such a deep lattice and mean occupation of sites less than four).
The absorption spectrum now serves as a guide for extracting the excited states. Suppose we are interested in excited states contributing to the first small peak around $\omega/U_0=0.24$. We take the wavepacket obtained at $t=300$ with the modulation at frequency $\omega/U_0=0.24$ as a starting wavepacket for the analysis. We first evolve this wavepacket for $T=10000$ with time step $\Delta t=0.05$ using a [*time independent*]{} Hamiltonian (i.e. $\delta=0$) and find the autocorrelation function $\tilde C_T(E)$. The latter shows a doublet around energy $E/U_0\approx 0.24,$ corresponding to two local excitations, one on the left side and one on the right side of the trap center. From this autocorrelation function, we obtain accurate eigenenergies. In order to extract the eigenstate with the chosen eigenenergy $E,$ we perform a Fourier transform on MPS states during a time sufficiently long for all states outside the peak at $E/U_0\approx 0.24$ to be significantly damped (in this specific case, $T_1=3000$ is convenient). This provides us with a wavepacket $|\psi_1\rangle$ with a large overlap with the desired eigenstate (typically 0.7-0.9, while the overlap of the eigenstate with the initial wavepacket was in the range 0.0001-0.01). The last step is to “purify” the wavepacket $|\psi_1\rangle$ recursively: the wavepacket $|\psi_i\rangle$ is propagated for a relatively long time ($T_i=6000$ in our specific case). The Fourier transform of the time series at energy $E$ again which yields a new wavepacket $|\psi_{i+1}\rangle,$ where the weight of all other eigenstates has been decreased (a Hahn window is used to speed up this decrease). This leads to an exponentially fast convergence of the wavepacket $|\psi_i\rangle$ towards the desired eigenstate. We observed that numerical inaccuracies may cap the exponential convergence of this procedure after 15-30 steps. In all propagating steps, we have again used time step $\Delta t=0.05$. In general, exact computational setup is adjusted taking into account the initial overlap, details of the spectrum and numerical stability.
Altogether, one iteration takes about 4 hours, using 8 threads on 4 core Intel i7 processors (the program is multi-threaded). We have performed 6 iterations to check the accuracy; a single additional iteration suffices for a 1% accuracy on the occupation numbers.
![(color online) Average occupation numbers per site for some selected excited states of the system analyzed in figure \[fig:ts3c\]. The black line with circles is the ground state, while red lines with squares show selected excited states. One of the two states corresponding to the absorption peak at $\omega/U_0=0.24$ is shown in panel (a) (the other state is symmetric w.r.t. the trap center). Panel (b) gives the density for one of the states associated to the very small peak around $\omega/U_0=0.37$. Panels (c) and (d) present two states associated to the “high” frequency peak at $\omega/U_0=1.75$. []{data-label="fig:exc1"}](excts3c1.eps){width="0.9\columnwidth"}
As mentioned above, two closely spaced in energy states correspond to the absorption peak at $\omega/U_0=0.24$, they differ from the ground state by the hopping of one particle between two Mott domains with $n=1$ and $n=2$ occupations occurring either on the left or on the right side of the trap center. Had the trap be placed symmetrically with respect to the optical lattice, the two elementary excitations would have been degenerate and the evolution from a symmetric ground state under a symmetric Hamiltonian would yield the symmetric combination of two excitations. A system with broken mirror symmetry allows for separation of these excitations, the left one is shown in Fig. \[fig:exc1\]a. Fig. \[fig:exc1\]b shows a similar excitation where a particle hops from an occupied to an empty side (this time shown for a “right” excitation). The corresponding energy is larger since such a situation occurs further from the center, where the harmonic trap is steeper. Two other excitations shown in Fig. \[fig:exc1\]c and Fig. \[fig:exc1\]d are the “left” and “right” excitations associated to the high frequency peak at $\omega/U_0=1.75$ and correspond to a particle hopping from single to doubly occupied Mott zone creating a site with triple occupancy.
![(color online) Average occupation numbers for another set of excited states of the system analyzed in Fig. \[fig:ts3c\]. The black line with circles represent the ground state while red lines with squares show selected excited states. []{data-label="fig:exc2"}](excts3c2.eps){width="0.9\columnwidth"}
All these excitations were correctly identified in [@huo11], based on energy considerations (possible for simple situations occurring in the deep insulating regime for large lattice depths). Our excited states provide a direct verification. Similarly, we have checked that the broad structure around $\omega/U_0=1$ (in fact consisting of several distinct peaks as visualized by $t=300$ absorption spectra in Fig. \[fig:ts3c\]) corresponds to single particle-hole excitations in the central Mott domain as identified by [@huo11].
Fig. \[fig:exc2\] presents another set of excited states. Fig. \[fig:exc2\]c resembles a typical particle hole excitation in the central Mott domain, but not between neighboring sites. Thus, the corresponding excitation energy lies not in the broad structure but on its wings. All other plots in this figure present excited states not analyzed in [@huo11]. They correspond to absorption peaks occurring on the left shoulder (a-c) and on the right (d-f) shoulder of the central structure. They show either particle-hole excitations occurring in the lower $n=1$ Mott domain or more complicated situations.
It is apparent that our method provides reliably excited states in the localized regime. All these excited states are well represented by an MPS with relatively small $\chi$. To discuss a more challenging situation, we move to another example, the absorption spectra studied in [@kollath06] where, due to a shallower optical lattice, some excitations may occur in the superfluid regions. In particular, two structures in the absorption spectra around $\omega/U_0\approx 1$ as well as $\omega/U_0\approx 2-2.3$ were observed, in agreement with the experiment [@stoeferle04]. The former resonance corresponds to the gap in the MI phase (i.e. to the particle-hole excitation in the Mott phase). Its appearance was interpreted as an evidence for the presence of the Mott phase. The latter has no such easy explanation; it was argued heuristically [@kollath06] that it corresponds to a particle-hole excitation in the superfluid component.
![(color online) Absorption spectrum for a system of 65 particles on 40 sites, in an harmonic trap with curvature $c=0.006,$ centered around $r_0=20.62345$. The mean lattice depth is $s_0=12$ (corresponding to $U_0=0.6752$ recoil energy). It is modulated sinusoidally with 1.67% relative amplitude. The smooth black line with open circles correspond to a modulation duration $t=50/E_R$; for longer modulation time $t=200/E_R,$ sharper peaks become resolved (red thicker line). The absorbed energy is divided by 4 in the later case, to compensate for a longer perturbation. The thick (green online) dashed line shows the absorption curve for a much stronger 20% modulation for $t=50/E_R$, rescaled by the intensity factor $12^2$.[]{data-label="fig:akol"}](Koll_absorb.eps){width="0.9\columnwidth"}
Our method allows to test the correctness of this claim. We take parameters of Ref. [@kollath06] with a slight modification. As in the previous example, we break the symmetry of the problem by shifting the lattice minima with respect to the harmonic trap. Such a situation seems generic from experimental point of view and allows to separate individual excitations. The absorption spectrum, similar to that of Fig. 4 of Ref. [@kollath06], is shown in Fig. \[fig:akol\]. The initial state - ground state at $s_0=12$ - is excited by a $\delta=0.2$ (20%) modulation for a duration $t=50/E_R$ - the result is shown as a thick dashed line. We observe a broad peak around modulation frequency $\omega/U_0=1$ as well as a much smaller broad peak around $\omega/U_0=2$ [^1]. Fig. \[fig:akol\] shows the absorption spectra for a more gentle driving, i.e. $1.67\%$ modulation. For an easier comparison, the strong modulation curve is rescaled by $12^2$ factor (square of the relative amplitude of modulation). The comparison of the two curves show that the structure around $\omega/U_0=2$ behaves perturbatively, while the broad resonance around $\omega/U_0=1$ shows signature of saturation.
The wavepacket obtained for a given frequency, at the end of the modulation, is a starting wavepacket for our procedure. The strong 20% modulation, as used in [@kollath06], leads to a ”too excited” wavepacket for the TEBD algorithm to work effectively over the long time required for a high-resolution FT (see, however, a discussion below). Of course, the excited states do not depend on the strength of the modulation (while the shape of the absorption curve in the nonperturbative regime does). For that reason, we use the wavepacket obtained after a “gentle” modulation with a smaller (1.67%) amplitude.
Fig. \[fig:akol\] shows also the absorption curve for a longer modulation interval $t=200/E_R$ which allows for a partial resolution of broad peaks. In particular, the structure around $\omega/U_0=2$ shows narrower partially resolved peaks around $\omega/U_0=1.84,\ 2.06,\ 2.25$.
Let us concentrate first on the wavepacket modulated with frequency $\omega/U_0=2.3$. We perform a FT of the autocorrelation function over a long time and observe several partially overlapping peaks. As shown in the inset of fig. \[fig:koll\], we identify a doublet of states, relatively well separated from the others, significantly excited from the ground state by the modulation. Fig. \[fig:koll\] shows the average occupation numbers $\langle n_l\rangle$ at site $l$ for the two excited states, extracted by our method, together with the ground state. Observe that, indeed, these states correspond to a particle-hole excitation in the SF regime (with average occupation numbers between 2 and 3) confirming the suggestion of [@kollath06]. The two states are quasi-symmetric images; they have slightly different energies, because of the choice $r_0=20.62345$ which breaks parity w.r.t. the trap center.
![(color online) Black circles: average occupation number for the ground state of 65 particles in a lattice with depth $s=12$ (40 sites) and an harmonic trap with curvature $c=0.006,$ centered around $r_0=20.62345$ (to separate left and right excited states w.r.t. to the trap center). The filled squares (red online) connected by a dashed line and the crosses (green online) connected by a dotted line show occupation numbers for two extracted excited states showing a particle-hole excitation in the SF region. The excess energies (energies measured with respect to the ground state energy) of the states are $\Delta E/U_0=2.2502$ (resp. $\Delta E/U_0=2.2434$) for the “red squares” (resp. “green crosses) in excellent agreement with the positions of two closely spaced peaks in the FT of the autocorrelation function shown in the inset. The lower panel shows the standard deviation of the occupation number on each site, for the ground state (open circles) and the excited states, the latter shows reduced fluctuations. []{data-label="fig:koll"}](koll4.eps){width="0.9\columnwidth"}
We have found that, in order to obtain a fully converged excited state (e.g. the excitation appearing on the right hand side of the center of the trap represented as crosses in Fig. \[fig:koll\]), another technique is useful. Knowing from the preliminary runs where the local (in space) excitation should be created, we prepare a new wavepacket by a local modulation, i.e. local chemical potential is modulated for a few nearby sites (around site 25 for this particular case). Explicitly $\mu_i(t)= \mu_i(1+\delta \exp[-(i-i_0)/w^2] \sin (\omega t))$ where $w$ is the spatial width of the modulation. Experimentally, such a modulation could be in principle created by an additional focused laser. Here it is used in the numerical simulation to speed up the convergence as well as to isolate excitations quasi-degenerate in energy but spatially well separated.
The data in Fig. \[fig:koll\] show some interesting differences when compared to the Mott regime previously discussed. While, in Figs. \[fig:exc1\]c and \[fig:exc1\]d, the excited states display a transfer of particles towards the trap center - explaining why they have an excess energy (compared to the ground state) smaller than $2U_0$ - it is the opposite for states in Fig. \[fig:koll\] where particles are transferred further from the trap center, and the associated excess energy is larger than $2U_0.$ The MPS representation of the excited states makes it easy to calculate several observables, not only the average occupation number of sites. The lower plot in Fig.\[fig:koll\] shows the standard deviation of the occupation of sites $\Delta n_l$, both for the ground state and the excited one. A small (resp. large) standard deviation is a signature of a Mott insulator (resp. superfluid region). The excited state has lower $\Delta n_l$ in the region of excitation, proving that it is more an “excited Mott”. This is confirmed by the occupation numbers themselves which are close to integer values for the excited state, with larger fractional part for the ground state.
To explore other excited states contributing to the structure around $\omega/U_0=2,$ we prepare the wavepacket with modulation frequency $\omega/U_0=2$. The first attempt to calculate the autocorrelation function by propagating such an excited wavepacket with $\chi=70$ (and a local Hilbert space on each site containing occupation numbers up to imax=7) failed, as e.g. the energy in the system was not conserved. We chose, however, to simply run the algorithm for a much longer time and observed an unexpected behavior depicted in Fig. \[fig:conver\].
![(color online) Long time evolution of a wavepacket prepared by a modulation at frequency $\omega/U_0=2$ and then evolved with the TEBD algorithm. The excess energy over the ground state is plotted in top frame (a). It is not conserved but, after the initial drop, it stabilizes. Similar behavior occurs for the entanglement entropy (b), with additional oscillations because the wavepacket is a sum of relatively few localized states. The FT of the autocorrelation function (c) does not display any clear peaks if performed in the \[0,5000\] time interval (black thin line), but definitely shows several well defined excited states over the time interval \[15000,20000\] (thick line, red online).[]{data-label="fig:conver"}](purig.eps){width="0.9\columnwidth"}
Fig. \[fig:conver\]a presents the energy of the wavepacket versus time (in recoil units, multiplication by $U_0=0.6752$ give time in units of $1/U_0$). After the initial decrease, we observe [*stabilization*]{} of the energy at some value above the ground state. Fig. \[fig:conver\]b presents the corresponding time dependence of the entropy of entanglement in the system defined as $S=\mathrm{max}_l\ S_l$ with $S_l$ given by Eq. (\[eq:ent\_ent\]). It also decays and stabilizes. This behavior can be understood by performing the FT of the wavepacket autocorrelation function (shown in Fig. \[fig:conver\]c). While the FT over the time interval \[0,5000\] displays broad erratic structures, the FT over the time interval \[15000,20000\] gives sharp peaks at some frequencies. The heuristic understanding of this behavior is the following. The initial excited wavepacket is a linear combination of eigenstates of the system. Some of them are well represented as an MPS, some other ones are represented poorly. During the time evolution the “well represented” part evolves accurately while the remaining part produces signatures of TEBD algorithm breakdown. During the course of evolution, a [*self-purification*]{} of the wavepacket occurs with the non-converged fraction seemingly disappearing.
The information on the non-converged fraction is actually lost in the TEBD evolution, when some $\lambda_{\alpha}^{[l]}$ values are discarded (those with $\alpha>\chi$.) No, or little, information is lost on the well represented excited states. Thus, unconverged TEBD time evolution filters out information which anyway cannot be obtained, purifying the wavepacket.
![(color online) Average occupation numbers of exemplary excited states showing a significant superfluid-type excitation, when compared to the ground state (black open circles). Red squares connected by the dashed line correspond to the excited state with excess energy $\Delta E/U_0=2.2418$ (lower energy peak of the doublet in the inset of Fig. \[fig:koll\]). Green crosses connected by dotted lines correspond to a state (excess energy $\Delta E/U_0=1.9504$) with significant excitation around the trap center. The lower panel shows the standard deviation of the occupation of sites for the ground state (open circles) and two excited states. The latter show regions of increased superfluid-like fluctuations. []{data-label="fig:kollsf"}](kollsf.eps){width="0.9\columnwidth"}
A further confirmation comes from the fact that the peaks of the FT provide the energies of excited states which can now be extracted. Fig.\[fig:kollsf\] show examples of excited states involving excitations in the superfluid regime. Contrary to the previous examples, these excited states show increased particle number fluctuations, see lower panel of Fig.\[fig:kollsf\].
![(color online) Average occupation numbers of exemplary excited states showing localized particle-hole excitations, compared to the ground state (black open circles). Red squares connected by dashed lines correspond to the excited states. (a) Excitation similar to those discussed for the smaller system, i.e. particle-hole excitation on the transition region between two Mott plateaus (excess energy $\Delta E/U_0=1.8417$). (b) and (c) show two particles being transfered between sites, the excess energies are $\Delta E/U_0=1.9988$ and $\Delta E/U_0=2.0539$. (d) gives an example of higher lying excited state $\Delta E/U_0=7.5280$ with a partially melted, right $\langle n_l \rangle=2$ Mott plateau. []{data-label="fig:kollmo"}](kollmo.eps){width="0.9\columnwidth"}
![(color online) Standard deviations of the occupations number, for the states presented in Fig. \[fig:kollmo\]. Local excitations lead to an increase of particle number fluctuations. []{data-label="fig:kollmosd"}](kollmosd.eps){width="0.9\columnwidth"}
States presented in Fig. \[fig:kollmo\]a–Fig. \[fig:kollmo\]c are other examples, well localized on the edge between the $\langle n_l \rangle=1$ and $\langle n_l \rangle=2$ zones. The corresponding standard deviations shown in Fig. \[fig:kollmosd\] show that these excitations, nevertheless, lead to an increase of local particle number fluctuations. Fig. \[fig:kollmo\]d shows a state with much larger excitation energy, almost $8U_0$. It shows that our method may produce states with quite high and complicated excitations. This state displays several particle transfers as well as almost a transfer of a part of the right hand side Mott $\langle n_l \rangle=2$ plateau into a superfuid region.
In effect a vast majority of states identified by the FT of the autocorrelation function can be extracted. Naturally, most of these states correspond to local excitations and not too high entanglement entropy.
Florence experiment
-------------------
A second important example is the 1D experiment of the Florence group [@fallani07] where a second, weaker lattice potential (incommensurate with the primary lattice) is added to simulate a disorder. Its strength is denoted by $s_2$. Theoretically, the addition of a disordered potential reveals a new insulating, compressible and gapless phase, called Bose glass (BG) [@fisher89]. Experimental results confirm its existence [@fallani07; @white09], but are not fully conclusive: the system is prepared from a low temperature BEC by ramping up an optical lattice, starting from a SF initial state. Preparing a MI or a BG requires, generally non-adiabatic, passage through a quantum phase transition [@dziarmaga02; @polkovnikov05], thereby exciting the system.
The Florence experiment is described in [@fallani07]. We simulated the experiment as detailed in [@zakrzewski09]: we integrate numerically the evolution of the system when $s$ increases using the TEBD algorithm [@vidal03], starting from the SF ground state at $s=4.$ At the same time, a secondary shallow lattice creating the disorder is turned on. Its effect is to modify the energy offset of the harmonic wells at different sites yielding now: $$\epsilon_j= c(j-r_0)^2 + s_2 E_{R2} \sin^2 \left(\frac{\pi j\lambda}{\lambda_2} + \phi
\right).
\label{trap}$$ The curvature of the trap, $c$ depends on the harmonic trap and on the additional trapping potential created by the transverse profile of the laser beams [@fallani07].
As $s$ increases, we found [@zakrzewski09] that the overlap of the dynamically evolved wave packet on the ground state for that $s,$ rapidly decreases in the region of the SF-MI transition (around $s=8-9$ for the experimental parameters) down to less than 10% for the final $s=14,$ a signature of broken adiabaticity. The situation is even worse (with the final overlap of the order of $10^{-6}$) for the disordered system. Disorder enriches the energy level structure around the ground state, killing adiabaticity both for shallow [@edwards08] and deep [@zakrzewski09] lattices.
To identify the induced excitations, we compute the various excited states building the dynamically created wavepacket, by evolving it further at the [*constant*]{} final $s$ value and calculating the FT in Eqs. (\[eq:ce\],\[eq:phie\]), as explained in the previous Section. In particular we characterize the states obtained using similar observables (easily obtained in the MPS representation [@mcculloch]): the average occupation number $\langle n_l\rangle$ at site $l$ and its standard deviation $\Delta n_l.$ As a by-product, we can also read off the average number of particles in the left part of the system $N_l=\sum_{i=1\ldots l} n_i$ and its standard deviation $\Delta N_l.$ This quantity is relatively easy to measure in a experiment. In our system it qualitatively resembles the entanglement entropy, as shown below [@klich].
![Autocorrelation spectra, Eq. (\[eq:ce\]), obtained dynamically for $s$=14 after switching on the optical lattice, for disorder strengths $s_2$= (a) 0, (b) 0.4375, (c) 2.1875. All parameters are taken to approximate the experimental situation [@fallani07]. The energy levels of the system appear as peaks (origin at the zero-point energy), with height $|c_i|^2$ from eigenbasis expansion. Peak labels are for further reference.[]{data-label="fig:autocorr_spectra"}](FT.eps){width="0.9\columnwidth"}
![(color online) Properties of states of the trapped BEC in a deep optical lattice ($s$=14), without disorder ($s_2$=0), $r_0$=0.12345 in Eq. (\[trap\]) to break the parity symmetry w.r.t. the trap center. The black line refers to the ground state, the red line to the dynamically prepared wavepacket, the brown (with crosses) and blue lines to the two excited states with the largest populations denoted as 1 and 2 in Fig. \[fig:autocorr\_spectra\]a. (a): Average occupation number $\langle n_l \rangle$ on each site. (b): Difference $D_l=\langle n_l \rangle - \langle n_l \rangle_{\mathrm{Gnd\ state}}$ (c): $\Delta n_l=\sqrt{\langle
n_l^2\rangle-\langle n_l\rangle^2}.$ (d): Entanglement entropy, Eq. (\[eq:ent\_ent\]); almost zero in the Mott plateaus for eigenstates - implying their approximate separability; large for the wavepacket. (e): Variance of the number of atoms to the left of site $l$. It is akin to the entanglement entropy, showing a marked difference between the eigenstates and the wavepacket. \[fig:full\_s0\]](summaryd00shi.eps){width="1.0\columnwidth"}
Fig. \[fig:autocorr\_spectra\] shows the autocorrelation spectrum, eq. (\[eq:ce\]), of the dynamically created wavepackets, at increasing disorder strengths. In the absence of disorder ($s_2=0$), about ten states are significantly excited. In Fig. \[fig:full\_s0\], we show various relevant quantities for the ground state, few excited states and the wavepacket. The average occupation number $\langle n_l\rangle$ on each site $l$ has the well known “wedding cake” structure, with large MI regions with integer $\langle n_l\rangle$ separated by narrow SF regions. Because the energy excess brought by non adiabatic preparation is small, all significantly populated excited states have similar shapes. Clearly, all excitations take place in or around the SF regions: these are transfers of one atom from the edge of a Mott plateau to the edge of another Mott plateau or to the neighboring SF region (“melting” of the Mott plateau).
The standard excitations in an homogeneous system (without trapping potential) such as a particle-hole excitation in a Mott plateau are absent in the stationary states because they are dynamically unstable. They are also absent in the wavepacket because they are energetically too costly: a particle-hole excitation costs at least one interaction energy $U\approx0.6E_R$, much larger than the excess energy of the wavepacket $0.1E_R$. Observe, therefore, that we are in a regime significantly different from the one discussed in the previous Section where excited states populated by modulation of the lattice depth have been considered. Here we are much “closer” in energy to the ground state.
The description used in [@gerbier05] where the ground state is contaminated by local particle-hole excitations is not compatible with our findings. A description in terms of melting of the MI [@gerbier07] seems more relevant. The local occupation fluctuation $\Delta n_l$ confirms the existence of large MI regions with low $\Delta n_l$ separated by SF peaks with larger $\Delta n_l.$ While the ground state and excited states look very similar, the wavepacket is different, with higher $\Delta n_l$ in the SF peaks. An even greater difference is revealed by the entanglement entropy. For the eigenstates, it is essentially zero in the MI regions and displays sharp peaks in the SF regions. In stark contrast, the $S_l$ of the wavepacket is also non-zero in the Mott plateaus. Similarly, $\Delta N_l$ is much larger in the Mott plateaus for the wavepacket than for the eigenstates. This suggests the existence of long-range entanglement between SF interfaces linking different MI regions caused by long range particle dislocations, absent in the eigenstates.
![(color online) Same as Fig. \[fig:full\_s0\] but for small disorder ($s_2$=0.4375, $r_0$=0 and $\phi$=0.5432 in Eq.(\[trap\])). Brown (blue) lines correspond to excitations A and B indicated in, Fig. \[fig:autocorr\_spectra\](b). Mott insulating regions are clearly visible for all eigenstates, more vague for the wavepacket. \[fig:full\_s05\]](summaryd05shi.eps){width="1.0\columnwidth"}
Disorder strongly modifies the properties of the system. Possible phases are analyzed in details in [@Roscilde08]. We consider first a small disorder $s_2$=0.4375. The breakdown of adiabaticity is stronger in presence of disorder, with twice larger excess energy - still below the particle-hole excitation energy - and more states significantly excited. The properties of various states are shown for small $s_2$ in Fig. \[fig:full\_s05\]. Several MI regions - identified by plateaus in $\langle n_l\rangle$ - still exist, separated by intermediate parts which are either SF or a BG. Remarkably, the ground state and the excited states do have very similar structures, with several visible Mott plateaus. In contrast, these plateaus are less visible (only the central one seems to survive, with a reduced size) for the wavepacket. This can be interpreted as a partial melting of the BG and MI phases, producing a thermal insulator [@gerbier07].
1truecm ![(color online) Properties of states of bosons trapped in a deep optical lattice ($s$=14) in the presence of strong disorder $s_2$=2.1875. Here $r_0$=0.1075 and $\phi$=0.0304 in Eq.(\[trap\])). Brown (blue) lines correspond to excitations X and Y in Fig. \[fig:autocorr\_spectra\](c). The entanglement entropy is much larger for the dynamically created wavepacket than for stationary states, and $\Delta n_l$ has many more peaks, indicating a significant melting of the Bose glass. \[fig:full\_s25\]](summaryd25shi.eps "fig:"){width="1.0\columnwidth"}
For strong disorder, $s_2=2.1875$, no MI state exists and the ground state of the system forms a BG. The exemplary excitations are shown in Fig. \[fig:full\_s25\]. The occupations of various sites strongly fluctuate, low lying excitations seem quite similar to the ground state. Excitations are [*local*]{} in character modifying the occupation number and its variance in selected (not always close) sites only. The entanglement entropy is small, except in small SF pockets. Again, the wavepacket has large entanglement entropy and $\Delta n_l,$ with numerous peaks indicating melted regions.
Conclusions
===========
In conclusion, we have built an extension of the TEBD algorithm capable of extracting, from a non stationary state (created dynamically), the eigenstates dynamically populated, and the nature of the “defects” created. In the specific case of the BH model realized in a ultracold atomic gas, a quasi-adiabatic quench populates excited states that are essentially similar to the ground state. In contrast, the dynamical wavepacket has a markedly different character, with a significant entanglement across the whole sample. By modulation spectroscopy, higher excited states may also be selectively reached and analyzed.
It is clear that our method will not provide an access to all excited states of a given many body system. By construction, only those excited states which are well represented by a MPS within a given assumed size of the internal Hilbert space $\chi$ may be obtained. On the other hand, the TEBD algorithm seems to create a “self-purification” mechanism which enables to reach integration times much longer than originally expected [@vidal03; @daley]. Naturally the result of this integration is not a good approximation of the true dynamics but contains information about the well represented excited states. Thus our method seems to have larger range of applicability than expected on the basis of standard limitations of the TEBD algorithm.
In that context, wavepackets with low excess energy w.r.t. the ground state created via quasi-adiabatic quench simulating the Florence experiment with disorder [@fallani07] behave exceptionally well. Higher excitations reached in modulation spectroscopy were more computer intensive and revealed that the method cannot be used as a kind of “black-box” approach but has to be tuned to individual cases. While this is certainly some drawback, the method potential capabilities (as exemplified with the highly excited state shown in fig. \[fig:kollmo\]d) seem quite encouraging.
Acknowledgements
================
We are grateful to M. Fleischhauer and his colleagues for critical remarks on an early version of this paper. Support within Polish Government scientific funds for 2009-2012 as a research project is acknowledged. M[Ł]{} acknowledges support from Jagiellonian University International Ph.D Studies in Physics of Complex Systems (Agreement No. MPD/2009/6) provided by Foundation for Polish Science and cofinanced by the European Regional Development Fund. Computer simulations were performed at supercomputer Deszno, purchased thanks to the financial support of the European Regional Development Fund in the framework of the Polish Innovation Economy Operational Program, contract no. POIG.02.01.00-12-023/08 (JZ), ACK Cyfronet AGH as a part of the POIG PL-Grid project (MŁ) and at ICM UW under Grant No. G29-10 (JZ and MŁ).
Appendix: implementation of MPS-addition
========================================
The addition of two MPS, an elementary step in the FT, produces $\lambda$ vectors with increased size. Thus it seems that the $\chi$ value required for representing successive sums should increase fast, ruining the efficiency of the MPS representation. This is not the case: the $\lambda^{[l]}_{\alpha_l=1,2\ldots}$ for the partial sums decrease relatively fast so that the smallest components can be safely discarded [@verstraete], preventing the cutoff to blow up, a key point for the success of the method. Altogether, several thousands states can be summed, at the price of roughly doubling the $\chi$ value to preserve the accuracy of the MPS representation: when the series is dynamically created, as in eq. (\[eq:phiw\]), the result is close to an eigenstate, i.e. simpler than the summands which are wavepackets, and the vectors added together are closely related.
To be able to compute the sum of several thousands of MPS-vectors two issues have to be resolved: restoring the canonical form of a MPS vector and performing truncation in $\chi.$ The former has been described in [@shividal]. The truncation from $\chi=\chi_2$ to $\chi=\chi_1$ can be done in two nonequivalent ways: ”blunt” and ”canonical”. The canonical truncation is done in parallel with restoring the canonical form. The algorithm described in [@shividal] processes sites sequentially restoring orthogonality relations at each of them one at a time. The ”left” and ”right” Schmidt vectors for site $k$ are:
$$|\psi^{1\ldots k}_{\alpha_k}\rangle = \!\!\!\!\!\!\sum\limits_{\genfrac{}{}{0pt}{}{\alpha_1,\ldots,\alpha_k}{i_1,\ldots,i_k}}\!\!\!\!\!
\Gamma_{1\alpha_1}^{[1]
,i_1}\lambda^{[1]}_{\alpha_1}\Gamma^{[2],i_2}_{\alpha_1\alpha_2}\ldots\Gamma^{[k],i_k}_{\alpha_{k-1}\alpha_k} |i_1,\dots,i_k\rangle
\label{eqn:MPS2}$$
$$|\psi^{k+1\ldots M}_{\alpha_k}\rangle = \!\!\!\!\!\!\sum\limits_{\genfrac{}{}{0pt}{}{\alpha_{k+1}, \ldots,\alpha_M}{i_{k+1},\ldots,i_M}}\!\!\!\!\!
\Gamma_{\alpha_k\alpha_{k+1}}^{[k+1]
,i_{k+1}}\lambda^{[k+1]}_{\alpha_{k+1}}\ldots\Gamma^{[M],i_M}_{\alpha_{M-1}\alpha_1} |i_{k+1},\dots,i_M\rangle
\label{eqn:MPS3}$$
$$|\psi\rangle = \sum\limits_{\alpha_k}\lambda_{\alpha_k}^{[k]} |\psi^{1\ldots k}_{\alpha_k}\rangle\otimes |\psi_{\alpha_k}^{k+1,\ldots,M}\rangle$$
For a MPS vector in its canonical form, these vectors are orthonormal eigenvectors of the reduced density matrices $\rho^{[1\ldots,k]},$ $\rho^{[k+1\ldots,M]}.$ The algorithm in [@shividal] restores orthonormality of left and right Schmidt vectors at site $k$ assuming it has already been applied for sites $1,\ldots,k-1.$ After the orthogonality has been restored at site $k,$ and the $\lambda^{[k]}$ sorted by decreasing magnitude, we zero all $\lambda^{[k]}_{\alpha_k}$ for $\chi_1 <\alpha_k \leq \chi_2.$ The truncation error is estimated by $\sum_{\alpha_k=\chi_1+1}^{\chi_2}
(\lambda^{[k]}_{\alpha_k})^2$ [@verstraete]. After all sites have been processed, at each site left Schmidt vectors are orthonormal eigenvectors of the appropriate reduced density matrix. Orthonormality of right Schmidt vectors at site $k$ is lost when truncation of $\lambda^{[k+1]}$ takes place. If the algorithm were applied again (without truncation), both left and right Schmidt vectors would be canonical for any site. We do not perform this second part since we do not use orthonormality of right Schmidt vectors for the addition procedure.
The blunt way to truncate the MPS state is to first use the previous algorithm with no parallel truncation at all, only then perform truncation of the vectors $\lambda^{[k]}$ at all sites. This method is mathematically flawed as, starting from the second site truncation, the canonical form is lost. However, when the $\chi$ value used for FT evaluation is larger than $\chi$ for the temporal evolution, both methods agree well.
In that respect, wavepackets with low excess energy w.r.t. the ground state created via quasi-adiabatic quench simulating the Florence experiment with disorder [@fallani07] behaved exceptionally well. The converged results could be obtained with $\chi=80$ as tested and compared with $\chi=120$ or $\chi=150$.
In contrast, extraction of selected eigenstates from wavepackets created by harmonic modulation is more demanding. Too strong modulation creates wavepackets containing apparently strongly entangled contributions. In such a case, a long time propagation of such a wavepacket is not possible even for large $\chi$ values. Lowering the amplitude of the modulation and adjusting its frequency is a key ingredient for successful extraction of a given excited state. The results presented in this paper have been obtained for $\chi=120$ again testing the convergence by comparison with $\chi=150$.
The convergence may be also tested by monitoring the growth of the smallest $\lambda^{[l]}_{\chi}$ values.
[99]{}
D. Ceperley and B. Alder, Science [**231**]{},555 (1986); J. Anderson, [*Quantum Monte Carlo*]{}, Oxford University Press, Oxford 1997. B. Bauer et al. J. Stat. Mech. P05001 (2011).
S.R. White, Phys. Rev. Lett. [**69**]{}, 2863 (1992); Phys. Rev. B[**48**]{}, 10345 (1993). K. A. Hallberg, Adv. Phys. **55**, 477 (2006).
U. Schollwöck, Ann. Phys. **326**, 96 (2011). G. Vidal, Phys. Rev. Lett. **91**, 147902 (2003);[*ibid.*]{} **93**, 040502 (2004). A. J. Daley, C. Kollath, U. Schollwöck, G. Vidal, J. Stat. Mech. [**2004**]{}, P04005, (2004). S.R. White and A.E. Feiguin, Phys. Rev. Lett. **93**, 076401 (2004). S. R. Clark and D. Jaksch, Phys. Rev. A **70**, 043612 (2004). F. M. Cucchietti, B. Damski, J. Dziarmaga, and W. H. Zurek, Phys. Rev. A **75**, 023603 (2007).
J. Zakrzewski and D. Delande, Phys. Rev. A **80**, 013602 (2009). O. Romero-Isart, K. Eckert, C. Rodó, and A. Sanpera, J. Phys. A: Math. Theor [**40**]{}, 8019 (2007). J.-S. Bernier, G. Roux, and C, Kollath, Phys. Rev. Lett. **106**, 200601 (2011). G. Biroli, C. Kollath, and A. M. Läuchli, Phys. Rev. Lett. **105**, 250401 (2011). J.-S. Bernier, D. Poletti, P. Barmettler, G. Roux, and C. Kollath, Phys. Rev. A **85**, 033641 (2012). C. Kollath et al.,Phys. Rev. Lett. **97**, 050402 (2006).
Jia-Wei Huo, Fu-Chun Zhang, Weiqiang Chen, M. Troyer, and U. Schollwoeck, Phys. Rev. A **84**, 043608 (2011). K. Hallberg, Phys. Rev. B [**5**2]{}, 9827 (1995). S.R. White and I. Affleck, Phys. Rev. B [**7**7]{}, 134427 (2008). R.G. Pereira, S.R. White, and I. Affleck, Phys. Rev. B [**7**9]{}, 165113 (2009). T. Barthel, U. Schollwoeck, and S. R. White, Phys. Rev. B [**7**9]{}, 245101 (2009). P. E. Dargel, A. Honecker, R. Peters, R. M. Noack, and T. Pruschke, Phys. Rev. B [**8**3]{}, 161104 (2011). A. Holzner, A. Weichselbaum, I. P. McCulloch, U. Schollwöck, and J. von Delft, Phys. Rev. B [**8**3]{}, 195115 (2011). J. Eisert, M. Cramer, and M.B. Plenio, Rev. Mod. Phys. [**82**]{}, 277 (2010). F. Verstraete, J. I. Cirac , Phys. Rev. B [**73**]{}, 094423 (2006). Y.Y. Shi, L.M. Duan and G. Vidal, Phys. Rev. A [**74**]{}, 022320 (2006). A. Kitaev and J. Preskill, Phys. Rev. Lett. **96**, 110404 (2006). I. P. McCulloch, J. Stat. Mech. P10014 (2007). F. J. Harris, Proc. IEEE [**66**]{}, 51 (1978). D. Jaksch et al., Phys. Rev. Lett. **81**, 3108 (1998). M.P.A. Fisher et al. Phys. Rev. B **40**, 546 (1989). O. Dutta, A. Eckardt, P. Hauke, B. Malomed, and M. Lewenstein, New J. Phys. [**13**]{}, 023019 (2011). D.-S. Lühmann, O. Jürgensen, and K. Sengstock, New J. Phys. [**1**4]{}, 033021 (2012). M. Greiner et al., Nature **415**, 39 (2002). M. Lewenstein et al., Adv. Phys. **56**, 243 (2007). T. Stöeferle et al., Phys. Rev. Lett. **92**, 130403 (2004). L. Fallani et al., Phys. Rev. Lett. **98**, 130404 (2007). W. Zwerger, J. Opt. B Quantum Semiclass. Opt. [**5**]{}, S9 (2003).
M. White et al., Phys. Rev. Lett. **102**, 055301 (2009). J. Dziarmaga et al., Phys. Rev. Lett. [**88**]{}, 167001 (2002). A. Polkovnikov, Phys. Rev. B[**72**]{} 161201 (2005). E.E. Edwards et al., Phys. Rev. Lett. [**101**]{}, 260402 (2008). I. Klich, G. Rafael, A. Silva, Phys. Rev. A[**74**]{}, 032306 (2011).
F. Gerbier et al., Phys. Rev. Lett. [**95**]{}, 050404 (2005); Phys. Rev. A[**72**]{}, 053606 (2005). F. Gerbier, Phys. Rev. Lett. [**99**]{}, 120405 (2007). T. Roscilde, Phys. Rev. A[**77**]{}, 063605 (2008); G. Roux et al., Phys. Rev. A[**78**]{}, 023628 (2008); X. Deng et al., Phys. Rev. A[**78**]{}, 013625 (2008).
[^1]: In [@kollath06], the relative amplitude of these two peaks differs from our result. We have tried to reproduce their curve also for precisely the same situation, i.e. a with $r_0=20.5,$ the symmetric case, but still the discrepancy persisted.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The interaction of quantum emitters with one-dimensional photon-like reservoirs induces strong and long-range dissipative couplings that give rise to the emergence of so-called Decoherence Free Subspaces (DFS) which are decoupled from dissipation. When introducing weak perturbations on the emitters, e.g., driving, the strong collective dissipation enforces an effective coherent evolution within the DFS. In this work, we show explicitly how by introducing single-site resolved drivings, we can use the effective dynamics within the DFS to design a universal set of one and two-qubit gates within the DFS of two-level atom-like systems. Using Liouvillian perturbation theory we calculate the scaling with the relevant figures of merit of the systems, such as the Purcell Factor and imperfect control of the drivings. Finally, we compare our results with previous proposals using atomic $\Lambda$ systems in leaky cavities.'
address:
- '$^1$ Max-Planck-Institute of Quantum Optics, Hans-Kopfermann-Strasse 1, 85748 Garching, Germany'
- '$^2$ Norman Bridge Laboratory of Physics 12-33, California Institute of Technology, Pasadena, CA 91125, USA'
- '$^3$ Institute for Quantum Information and Matter, California Institute of Technology, Pasadena, CA 91125, USA'
author:
- 'V. Paulisch$^1$, H. J. Kimble$^{2,3}$ and A. González-Tudela$^1$'
bibliography:
- 'Sci.bib'
- 'books.bib'
title: Universal Quantum Computation in Waveguide QED using Decoherence Free Subspaces
---
Introduction
============
Recent theoretical and experimental work has shown that an attractive configuration to engineer strong collective dissipation is given by one-dimensional (1d) photonic-like systems such as photonic crystal waveguides [@joannopoulos_book95a; @laucht12a; @goban13a; @yu14a; @tiecke14a; @sollner14a; @arcari14a; @goban15a; @young15a], optical fibers [@vetsch10a; @goban12a; @mitsch14a; @petersen14a; @beguin14a], metal [@chang07a; @dzsotjan10a; @gonzaleztudela11a; @bermudez15a] and graphene plasmonic [@koppens11a; @huidobro12a; @christensen11a; @martinmoreno15a] waveguides or superconducting circuits [@mlynek14a]. Their interaction with quantum emitters, usually referred to as waveguide QED, offers interesting characteristics: *i)* the density of modes of the waveguide is inversely proportional to the group velocity $1/v_g(\omega_{{{\rm a}}})$, and therefore is strongly enhanced when the atomic frequency is in a region of slow light, e.g., in photonic crystal waveguides close to a band edge. This enhancement implies achieving regions of a large decay rate into the waveguide, $\Gamma_\oned$, compared to other decay channels, denoted by $\Gamma^*$, characterized through the *Purcell Factor*, $P_\oned=\Gamma_\oned/\Gamma^*$; *ii)* the 1d guided modes retain a small modal area $\lesssim \lambda_{{\rm a}}^2$, for propagation lengths $L_{\mathrm{prop}} \gg \lambda_{{\rm a}}$ (the wavelength of the 1d mode associated to the atomic frequency considered); *iii)* the interaction is *strongly* long-range, favoring individual adressing, and it can even be homogeneous if the positions of the atom-like systems are chosen properly [@lehmberg70a; @lehmberg70b], in contrast to 2d or 3d system. This collective dissipation leads to the emergence of subradiant states that form the so-called Decoherence-Free Subspace (DFS) [@zanardi97a; @lidar98a].
Previous works have already considered how to use the DFS of two atoms trapped in leaky cavities to design one and two-qubit gates using three-level $\Lambda$-type schemes [@beige00a; @beige00b; @tregenna02a], where two atomic hyperfine levels are used to encode the qubit. In the light of the variety of systems available nowadays that allows to engineer robust one-dimensional DFS [@joannopoulos_book95a; @laucht12a; @goban13a; @yu14a; @tiecke14a; @sollner14a; @arcari14a; @goban15a; @young15a; @vetsch10a; @goban12a; @mitsch14a; @petersen14a; @beguin14a; @chang07a; @dzsotjan10a; @gonzaleztudela11a; @bermudez15a; @koppens11a; @huidobro12a; @christensen11a; @martinmoreno15a; @mlynek14a], which may couple to different types of quantum emitters, e.g., atoms, quantum dots, NV centers or superconducting qubits, it is interesting to revisit the problem and fill some of the gaps that have not been considered so far, namely,: *i)* how to encode decoherence-free qubits using only two-level systems (TLS) (as $\Lambda$-schemes might not be available for all platforms); *ii)* extend the proposal to systems with more than two atoms; *iii)* analyze the scaling of the fidelities with the relevant figures of merit of the system, e.g., $P_{\oned}$, that in previous works [@beige00a; @beige00b; @tregenna02a] was mainly done through numerical analysis; *iv)* discuss the consequences of imperfect addressing on the fidelity of the gates.
In this work, we show an implementation of universal quantum gates by using $N$ TLS strongly coupled to 1d photon-like reservoirs. First, we show that by *pairing* the TLS, we can define decoherence free-qubits in the singlet (i.e., antisymmetric) states of each pair. The combination of these singlets form the so-called *computational subspace* where we define our operations. Then, we explicitly show how to build single qubit (e.g., phase gates and Pauli-$X$ gates) and two-qubit (e.g., controlled $Z$-gates) operations within the computational subspace without coupling to the other states in the DFS. By using Liouvillian perturbation theory, we obtain analytical expressions for the scaling of the fidelities of the operation ($1-F\propto 1/\sqrt{P_\oned}$) and estimate the error when increasing the number of atoms. Finally, we revisit the problem of the implementation with $\Lambda$ systems in leaky cavities [@beige00a; @beige00b; @tregenna02a] and show how both lead to the similar scaling.
The paper is divided as follows: in section \[sec1\], we introduce the set-up where we implement our proposal and establish the general formalism that we use to characterize the operations. In section \[sec3\], we describe the logical qubits and computational subspace and show how to build a set of universal quantum gates in the ideal case, that is, without considering decay into other non-guided modes or deviations from Quantum Zeno dynamics [@zanardi97a; @lidar98a; @facchi02a]. Then, in sections \[sec4\] and \[sec5\], we analyze possible error sources, both analytically and numerically, including spontaneous emission and imperfect addressing for the different gates of our proposal. Finally, in section \[sec6\], we compare the scaling with the proposal of three-level atoms in leaky cavities already explored in the literature [@beige00a; @beige00b; @tregenna02a].
General set-up and formalism {#sec1}
============================
Set-up: waveguide QED
---------------------
The general set-up that we consider is depicted in figure \[fig1\]a; namely $N$ TLSs, $\{ |g\rangle_n$, $| e \rangle_n \}_{n = 1 \dots N}$, placed at positions $z_n$ and coupled to a 1d field with bosonic annihilation operators $a_q$. Due to the variety of implementations available nowadays, we will keep the discussion as general as possible without making further assumptions on the nature of the TLS and/or 1d waveguides.
![(a) TLS (in blue) trapped along a one-dimensional waveguide, addressed by single-site resolved control fields. In green, we depict the pairing that we will use to engineer the computational qubits.(b) Level structure of a TLS with a coherent driving with amplitude (detuning) $\Omega_n$ ($\Delta_n$) with an additional level to which transitions can be driven off-resonantly to engineer the $\Delta_n \sigma_{ee}^n$ term of the hamiltonian. (c) A three-level system in which the excited state is driven far off-resonantly can be made approximately equivalent to a TLS with modified parameters as shown in the legend.[]{data-label="fig1"}](figure1v3.pdf){width="98.00000%"}
The composite system is described by the Hamiltonian $H = H_0 + H_\II$, where $H_0$ is the free term given by $H_0 = H_\mathrm{qb} + H_\mathrm{field}$, (using $\hbar = 1$) $$H_\mathrm{qb} = \omega_{{\rm a}}\sum_{n=1}^N \sigma^{n}_{ee}, \
H_\mathrm{field} = \sum_q \omega_q a^\dagger_q a_q,
\label{H0}$$ where $\omega_{{\rm a}}$ is the TLS energy, $\sigma_{ij}^n={|i\rangle}_n{\langle j|}_n$ are atomic operators, and $\omega_q$ is the energy dispersion relation of the waveguide modes. We consider a dipolar coupling of the form $$H_\II = \sum_{n} \left( \sigma_{ge}^n E(z_n) + \mathrm{H.c.} \right),
\label{Hint}$$ with $E(z) = \sum_q g_q (a_q e^{i q z}+{{a_q^{\dagger}}} e^{-i q z})$, and $g_q$ the single photon coupling constant. When the system-reservoir coupling is weak (Born approximation) and the evolution of the 1d-reservoir is much faster that the one of the system (Markov approximation), the evolution of $\rho$, the reduced density matrix for the atom-like system, can be described by a Markovian master equation of the form $d \rho /d t = {\cal L}\left[\rho\right]$ [@lehmberg70a; @lehmberg70b; @chang12a; @gardiner_book00a], with the superoperator $$\label{mequation1}
{\cal L}\left[\rho\right] =
\sum_{n,m} \Gamma_{n,m} \left( \sigma_{ge}^n \rho \sigma_{eg}^m - \rho \sigma_{eg}^m \sigma_{ge}^n \right)
+ \mathrm{H.c.} \,,$$ where $$\Gamma_{n,m} = \frac{\Gamma_{\oned}}{2} e^{i q(\omega_{{\rm a}}) |z_n - z_m|}\,,$$ where $\Gamma_\oned$, the decay into waveguide modes, that we will assume to be larger than the rate of spontaneous emission into all other modes, $\Gamma^* \ll \Gamma_\oned$ as this is the regime we are interested in. Moreover, as the propagation lengths of the waveguide modes for many implementations are long ($L_\mathrm{p} \gg \lambda_\mathrm{a}$), the atoms can be separated several wavelengths apart and can therefore be individually addressed as depicted in figure \[fig1\](a). In particular, we assume to control the TLS state through the Hamiltonian (in the interaction picture with respect to $H_\mathrm{qb}$) $$H_\mathrm{las}= \sum_{n} \frac{1}{2} \left( \Omega_n \sigma_{ge}^n + \mathrm{H.c.} \right) + \Delta_n \sigma_{ee}^n,
\label{eq:Hlas}$$ where $\Omega_n$ is the amplitude of the coherent driving (that we consider to be resonant, i.e., $\omega_L=\omega_{{\rm a}}$) which controls the number of excitations of the system, and $\Delta_n$ is a phase shift interaction term. The latter can be obtained, e.g., in atomic systems, by adding an off-resonant driving to another excited state ${|e'\rangle}$, as depicted in figure \[fig1\](b), which results in an Stark shift $\Delta_n=|\Omega'|^2/(\omega_{{\rm a}}-\omega_L')$. In general, the way of implementing $\Omega_n$ and $\Delta_n$ will depend on the particular system.
For completeness, it is worth mentioning that $\Lambda$ systems can also be mapped to effective TLS by using an off-resonant Raman transition as depicted in figure \[fig1\](c). By adiabatically eliminating the excited state ${|e'\rangle}$, one can formally project the dynamics to the two metastable states, $\{{|g\rangle},{|e\rangle}\}$, and find a similar light-matter hamiltonian as the one of equation \[Hint\], with the advantage that the effective TLS defined by $\{{|g\rangle},{|e\rangle}\}$ will be long-lived as they are encoded in metastable states. For example, by switching both $\Omega_g$ and $\Omega_e$ at the same time with detuning $\delta(\gg |\Omega_g|,|\Omega_e|)$ as depicted in figure \[fig1\](c), we can implement a coherent driving term with effective $\Omega = \frac{\Omega_g \Omega_e^*}{4 \delta}$. By switching $\delta$ in this case big enough one can neglect spontaneous emission processes as they will be proportional to $\Gamma^* \left(\frac{|\Omega_e|^2+|\Omega_g|^2}{4\delta^2}\right)$. Moreover, if we switch only $\Omega_e$ and adiabatically eliminate the photonic modes we also obtain an irreversible transition from ${|e\rangle}\rightarrow{|g\rangle}$, but with a renormalization of the decay rates $\Gamma_\oned \rightarrow \Gamma_\oned | \frac{\Omega_e}{2 \delta} |^2$ and $\Gamma^* \rightarrow \Gamma^* | \frac{\Omega_e}{2\delta}|^2$. Hence, the Purcell factor $P_\oned = \Gamma_\oned / \Gamma^*$ is unchanged. In that situation our analysis is an alternative implementation to the one developed in Refs. [@beige00a; @beige00b; @tregenna02a].
Decoherence-Free Subspaces
--------------------------
In the case of equidistant spacing at positions commensurate with the wavelength of the guided mode, i.e. $z_n =n 2\pi/q(\omega_{{\rm a}})$, the effective interaction induced by the waveguide modes yields a pure Dicke model [@dicke54a] decay described by $${\cal L}_\DD \left[ \rho \right]
= \frac{\Gamma_\oned}{2} \left(S_{ge} \rho S_{eg} - S_{eg} S_{ge} \rho \right) + \mathrm{H.c.},
\label{Dicke}$$ where we have introduced the collective spin operator $S_{ge} = \sum_{n=1}^N \sigma^n_{ge}$. The states satisfying $S_{ge} {|\Psi\rangle}=0$ are *decoherence-free* with respect to the collective dissipation $\mathcal{L}_\DD$. These states can be easily described in the collective spin basis $\{| J, m_J, \alpha_J \rangle \}$, that is the eigenstates of the collective operators $S^2 =\sum_{i={x,y.z}}S_{i}^2$ and $S_z$ with $$\begin{aligned}
S^2{|J,m_J,\alpha_J\rangle}&=J(J+1){|J,m_J,\alpha_J\rangle}\,,\\
S_z{|J,m_J,\alpha_J\rangle}&=m_J {|J,m_J,\alpha_J\rangle}\,,\end{aligned}$$ where $J=N/2,N/2-1,\dots$, $m_J=-J, -J+1,\dots,J$. The index $\alpha_J$ is introduced because the states in the collective spin basis are degenerate, with degeneracy given by: $\alpha_J = 1,\dots, { {N \choose J} }-{ {N \choose J-1} }$. It is easy to observe in this basis that the states ${|J,-J,\alpha_J\rangle}$ are decoherence free, and therefore span the *decoherence-free subspace* (DFS).
The DFS has a dimension of ${ {N \choose N/2} }$ (assuming even atom number $N$), and is composed of all the possible states which are antisymmetric with the permutation of two atoms. Thus, an alternative way of characterizing the DFS is to consider all possible (tensor products) of singlet states $${|A_{m,n}\rangle} = \left({|e\rangle}_m \otimes {|g\rangle}_n-{|g\rangle}_m \otimes {|e\rangle}_n\right)/\sqrt{2}\,,$$ where $m,n$ denote the atomic positions of the pair of atoms that form the singlet. This characterization makes it more difficult to describe an orthonormal basis of the DFS. However, we show in the next section that it is convenient to define our computational subspace.
Quantum Zeno dynamics using Liouvillian perturbation theory {#sec2}
-----------------------------------------------------------
We are interested in the regime where the collective dissipation induced by $\mathcal{L}_\mathrm{D}$, with characteristic timescale $\Gamma_{\oned}^{-1}$, dominates over any possible perturbation of the system, $\mathcal{L}_\mathrm{pert}$, with characteristic timescale $\tau\gg\Gamma_{\mathrm{1d}}^{-1}$. Under these assumptions, any state outside of the DFS will only be virtually populated due to the strong dissipation and therefore the dynamics will be restricted to the slow subspace, i.e., the DFS. Mathematically, we formalize this intuitive picture by defining a projection superoperator $\mathbb{P}$ (with $\mathbb{P}^2 = \mathbb{P}$) satisfying: $\mathbb{P}{\cal L}_\DD={\cal L}_\DD \mathbb{P}=0$ that projects out the fast dynamics yielding only the effective evolution in the slow subspace. It is then possible to integrate out the fast dynamics (see Appendix A) arriving to an effective master equation given by $$\frac{\partial \PP\rho}{\partial t}=\mathcal{L}_\mathrm{eff} [\PP\rho]= \Big(\PP \mathcal{L}_\mathrm{pert} \PP +\PP \mathcal{L}_\mathrm{pert} \QQ \frac{1}{{-\cal L}_\DD}\QQ \mathcal{L}_\mathrm{pert} \PP +\mathrm{O}\left(\tau^{-3}/\Gamma_\oned^2 \right)\Big) \rho\,.\label{eq:effemastereq}$$ This result to first order (left hand term in the brackets) accounts for the ideal *Quantum Zeno dynamics* [@zanardi97a; @lidar98a; @facchi02a]. The second order in perturbation theory then yields correction terms mainly coming from slightly populating the (super)radiant states.
In our case, there will be two types of perturbations, namely,
- The Hamiltonian $\mathcal{L}_\mathrm{pert}[\cdot]= - \rmi \left[H_{\mathrm{las}}, \cdot \right]$ to control the atomic state. This results to first order in an effective Hamiltonian $H_{\mathrm{eff}}=\mathcal{P} H_{\mathrm{las}} \mathcal{P}$ that couples only atomic states within the DFS. Here, we introduced the projection onto the DFS for pure states $\mathcal{P} = \sum_i {|\mathrm{d}_i\rangle}{\langle \mathrm{d}_i|}$, where the states ${|\mathrm{d}_i\rangle}$ form an orthonormal basis of the DFS. We use this effective laser coupling to control the atomic state of the ensemble. Besides, there is a second order correction resulting from $H_{\mathrm{las}}$ that will be relevant for the analysis of the error probability of our proposal as we show in section \[sec4\].
- The contribution of the emission of photons to other radiative modes different from the guided mode of the waveguide that we embed into a single decay rate, $\Gamma^*$ and describe through the Liouvillian $$\mathcal{L}_\mathrm{pert}[\rho]=\mathcal{L}_{*}[\rho]=\sum_n \frac{\Gamma^*}{2}(\sigma_{ge}^n\rho\sigma_{eg}^n-\rho \sigma_{ee}^n+\mathrm{H.c.}).
\label{eq:SpEm}$$ This contribution is relevant for the error analysis of the gates in section \[sec4\].
Universal Single- and Two-Qubit Gates {#sec3}
=====================================
In this Section, we show how to engineer a set of universal gates, i.e., defined by any arbitrary single-qubit rotation and a controlled gate [@nielsen_book00a], using the effective evolution $H_{\mathrm{eff}}$ within the DFS that appear in our waveguide QED setup. Firstly, due to the large degeneracy of the DFS, we need to define a set of logical qubits that will expand our computational subspace. Then, we show how to choose $\{\Omega_n,\Delta_n\}$ such that they define a set of universal one and two-qubit gates, namely, the phase and Pauli-$X$ (and $Y$) gate and the controlled-$(-Z)$ gate. A summary of the parameters for these gates can be found in table \[tab:GateSum\]. For completeness, we also give the parameters for other gates such as the Hadamard or SWAP gates. The former can be easily constructed because all single qubit rotations can be performed and the latter is constructed through the same idea as the controlled-$(-Z)$ gate. Due to the degeneracy of the DFS, the challenge lies in defining operations within the computational subspace, without populating the rest of the states within the DFS. In section \[sec4\], we revisit the problem and consider the effect of spontaneous emission and second order corrections to the Zeno dynamics that ultimately limit the fidelity of the operations.
[l|cccccc|c|c]{} Gate & $\Omega_{12}^-$ & $\Omega_{12}^+$ & $\Delta_{12}^+$ & $\Delta_{12}^-$ & $\Delta_{34}^+$ & $\Delta_{34}^-$ & $T$ & $\Delta_{n\geq5}$\
$ T_1$ & 0 & 0 & -$\Delta_\mathrm{T}$ & 0 & 0 & 0 & $\frac{\pi}{4 \Delta_\mathrm{T}}$& 0\
$ Z_1$ & 0 & 0 & $\Delta_\mathrm{Z}$ & 0 & 0 & 0 & $\frac{\pi}{\Delta_Z}$ & 0\
$ X_1$ & $\Omega_\mathrm{X} \in \mathbb{R}$ & 0 & 0 & 0 & $\Delta_\mathrm{D}$ & 0 & $\frac{\pi}{\sqrt{2} \Omega_\mathrm{X}}$ & $\Delta_\mathrm{D}$\
$ Y_1$ & $\Omega_\mathrm{Y} \in \rmi \mathbb{R}$ & 0 & 0 & 0 & $\Delta_\mathrm{D}$ & 0 & $\frac{\pi}{\sqrt{2} |\Omega_\mathrm{Y}|}$ & $\Delta_\mathrm{D}$\
$ \propto H_1$ & -$\Omega_\mathrm{H} \in \mathbb{R}$ & 0 & $\sqrt{2} \Omega_\mathrm{H}$ & 0 & $\Delta_\mathrm{D}$ & 0 & $\frac{\pi}{2 \Omega_\mathrm{H}}$ & $\Delta_\mathrm{D}$\
SWAP$_{12}$ & 0 & 0 & 0 & $\Delta_\mathrm{S}$ & 0 & $\Delta_\mathrm{S}$ & $\frac{\pi}{\Delta_\mathrm{S}}$ & $\Delta_\mathrm{D}$\
C($-Z$)$_{12}$ & 0 & 0 & 0 & $\Delta_\mathrm{C}$ & 0 & 0 & $\frac{\pi 2 \sqrt{2}}{\Delta_\mathrm{C}}$ & $\Delta_\mathrm{D}$\
Definition of Logical Qubits and Computational Subspace
-------------------------------------------------------
Before finding the appropriate gates, we first need to define the logical qubits within the DFS. In principle, assuming an even number of atoms $N$, the dimension of the DFS is ${ {N \choose N/2} }$ and therefore, it is possible to encode $\log_2 { {N \choose N/2} }$ logical qubits. However, it is more useful to restrict the computational subspace to a smaller set of states in order to achieve universal quantum computation. As the DFS is spanned by all (tensor) products of singlet states over two atoms, it is natural to define the logical qubits as $$\begin{aligned}
{|0\rangle}_{j}^{\mathrm{L}} \equiv & {|G_{j,j+1}\rangle} = {|g\rangle}_j \otimes {|g\rangle}_{j+1} \label{eq:logicalQB1} \\
{|1\rangle}_{j}^{\mathrm{L}} \equiv & {|A_{j,j+1}\rangle} = \left({|e\rangle}_j \otimes {|g\rangle}_{j+1}-{|g\rangle}_j \otimes {|e\rangle}_{j+1}\right)/\sqrt{2}.
\label{eq:logicalQB2}\end{aligned}$$ It is instructive to consider particular examples to see how the DFS and computational space look, i.e., for the case of $N=2$ and $N=4$ atoms.
#### **Two atoms:**
In this case it is easy to plot the complete Hilbert space (including states outside DFS) as it consists only of 4 states as depicted in figure \[fig2\](a). The separation into DFS states and non-DFS states is easily done in the familiar singlet-triplet basis. The DFS consists of two states: the one with two atoms in the ground state and the singlet state, i.e., the antisymmetric combination of one single excited state. Thus, we can encode one logical decoherence free qubit. The other two states are superradiant, i.e., they decay with an enhanced decay rate of $2 \Gamma_\oned$.
#### **Four atoms:**
The complete Hilbert space consists of $2^4=16$ states, and the dimension of the DFS, shown in figure \[fig2\](b), is ${ {4 \choose 2} } = 6$. As aforementioned, we want to use as computational subspace the tensor product of the antisymmetric pairs described in equation \[eq:logicalQB1\] and \[eq:logicalQB2\] , which consist only of $2^{2}=4$ states. This is why in this situation we have to distinguish within the DFS between the computational space and the additional states that must be either decoupled or used as auxiliary states. In particular, for $N=4$ the auxiliary states are ${|A_{1,2}G_{3,4}+A_{3,4}G_{1,2}\rangle}/\sqrt{3}$ and ${|A_{1,3}A_{2,4}+A_{1,4}A_{2,3}\rangle}/\sqrt{2}$.
#### **For $N>4$ atoms:**
In general (for $N>2$) the dimension of the DFS, ${ {N \choose N/2} }$ (for even $N$), is larger than the one of the computational subspace, $2^{N/2}$. Thus, one can split the projection onto the DFS, $\mathcal{P}$, into two orthogonal projections, i.e. one into the computational subspace $\mathcal{P}_\mathrm{CS}$ and its orthogonal counterpart $\mathcal{Q}_\mathrm{CS}$: $$\begin{aligned}
\mathcal{P} =& \mathcal{P}_\mathrm{CS} + \mathcal{Q}_\mathrm{CS},\\
\mathcal{P}_\mathrm{CS} =& \sum_{j\ \mathrm{odd}} {|0\rangle}^\mathrm{L}_j {\langle 0|} + {|1\rangle}^\mathrm{L}_j {\langle 1|}\,,\end{aligned}$$ such that the effective Hamiltonian can be written as follows $$H_\mathrm{eff} =
\mathcal{P}_\mathrm{CS} H_{\mathrm{las}} \mathcal{P}_\mathrm{CS}
+ \left( \mathcal{P}_\mathrm{CS} H_{\mathrm{las}} \mathcal{Q}_\mathrm{CS} + \mathrm{H.c.} \right)
+ \mathcal{Q}_\mathrm{CS} H_{\mathrm{las}} \mathcal{Q}_\mathrm{CS},$$ which separates the transitions within the computational \[auxiliary\] subspace $\mathcal{P}_\mathrm{CS} H \mathcal{P}_\mathrm{CS}$ \[$\mathcal{Q}_\mathrm{CS} H \mathcal{Q}_\mathrm{CS}$\] and the coupling between these two subspaces: $\mathcal{P}_\mathrm{CS} H\mathcal{Q}_\mathrm{CS}$. This separation will be useful to argue that we can make operations within the DFS even in the situations with $N>4$ as we will show afterwards. For a general situation, it is easy to show that by projecting $H_{\mathrm{las}}$ into the computational subspace we obtain an effective evolution inside the computational subspace given by: $$\mathcal{P}_\mathrm{CS} H_{\mathrm{las}}\mathcal{P}_\mathrm{CS}
= \sum_{j \ \mathrm{odd}}
\left(\frac{\Omega_{j,j+1}^{-}}{\sqrt{2}} {|1\rangle}^\mathrm{L}_j{\langle 0|} + \mathrm{h.c.} \right)
+ \Delta_{j,j+1}^+ {|1\rangle}^\mathrm{L}_j {\langle 1|},$$ where we used the following notation $x_{i,j}^\pm = \frac{1}{2} \left(x_i \pm x_{j}\right)$ to abbreviate the combination of parameters.
![(a) Hilbert space of two TLS in the basis of non-DFS and DFS states with energies in the interaction picture with respect to $H_\mathrm{qb}$. The blue arrows denote the transitions necessary for the Pauli-X gate. The triplet states ${|E\rangle} = {|ee\rangle}$ and ${|S\rangle} = {|eg+ge\rangle}/\sqrt{2}$ are not inside the DFS. (b) DFS of 4 TLS consists of 6 states that split into the $2^2$-dimensional computational subspace and two states in the auxiliary subspace. The blue \[green\] arrows denote the transitions necessary for the Pauli-X \[and Controlled-(-Z)\] gate.[]{data-label="fig2"}](figure2v1.pdf){width="98.00000%"}
Single-Qubit Gates
------------------
The goal is to find the $\{\Omega_n,\Delta_n\}$ such that they define both the phase and Pauli-$X$ (and $Y$) gates over the computational subspace.
#### **Two atoms:**
This is the simplest situation because the size of the computational space is the same as the one of the DFS. In this case (see also [@beige00b]), a phase shift gate on the logical qubit ($\alpha {|0\rangle}^\mathrm{L} + \beta {|1\rangle}^\mathrm{L}\rightarrow \alpha {|0\rangle}^\mathrm{L} + \beta \mathrm{e}^{-\rmi \phi} {|1\rangle}^\mathrm{L}$) is obtained by applying $\Omega_{12}^-= 0$, $\Delta_{12}^+ \neq 0$ for a time $T = \frac{\phi}{\Delta_{12}^+}$. Pauli-$X$ rotations (plus a phase) are obtained for $0 \neq \Omega_{12}^- \in \mathbb{R}$, $\Delta_{12}^+ = 0$ and time $T = \frac{\pi}{\sqrt{2} \Omega_{12}^-}$. Note that to avoid errors in both cases, one should also set $\Omega_{12}^+ = \Delta_{12}^- = 0$ as will be discussed in section \[sec4\]. The Pauli $Y$ can be obtained as the $X$ just by using $i\Omega_{12}^- \in\ \mathbb{R}$, so that we will not discuss it further.
#### **Four atoms:**
In this case, the way to do phase gates and rotations is the same as for the two atom case. However, in the case of the rotations, states within the computational subspace couple to auxiliary states for more than two atoms. In particular, the state ${|10\rangle}^\mathrm{L}$ is coupled to the auxiliary state ${|A_{1,3}A_{2,4}+A_{1,4}A_{2,3}\rangle}/\sqrt{2}$ for $\Omega_{12}^- \neq 0$ as shown in figure \[fig2\]. However, this transition can be made far off-resonance by setting $|\Delta_{34}^+| \gg |\Omega_{12}^-|$. This results in an additional error rate $\frac{|\Omega_{12}^-|^2}{2 \Delta_{34}^+}$ that will be considered when calculating the fidelity of the operation.[^1]
#### **For $N>4$ atoms:**
Again in the case of rotations, transitions to states outside the computational subspace in the ideal case ($\Omega_{12}^+ = \Delta_{12}^- = 0$) are possible when $\Omega_{12}^- \neq 0$, that is when $$\mathcal{Q}_\mathrm{CS} H \mathcal{P}_\mathrm{CS}
=\mathcal{Q}_\mathrm{CS} H_\mathrm{eff} \mathcal{P}_\mathrm{CS} \neq 0\,,$$ where we use that $\mathcal{P}_\mathrm{CS} \mathcal{P} = \mathcal{P}_\mathrm{CS}$. However, the transitions to these states can be made far off-resonant by setting $$\Omega_n = 0, \ \mathrm{and} \ |\Omega_{12}^-| \ll \Delta_n
= \Delta_\mathrm{D} \ll \Gamma_\oned, \ n \geq 3,$$ because the auxiliary states inside the DFS that the computational subspace couples to extend over more than two atoms [^2] and can therefore be detuned as $$\mathcal{Q}_\mathrm{CS} H \mathcal{Q}_\mathrm{CS} \sim \Delta_D \mathcal{Q}_\mathrm{CS}\,,
\label{eq:DeltaD}$$ while keeping the desired transition driven by $\Omega_{12}^-$ as resonant. One has to make sure that the Stark-shift introduced by this off-resonant transition is small and possibly correct the detuning that it will induce by choosing appropriately the applied laser frequency $\omega_L$.
Controlled Pauli-Z
------------------
For universal quantum computation, a controlled two-qubit gate is required. In this case, the minimal system to encode the operation is the $N=4$ atom case, where two decoherence-free logical qubits can be obtained.
#### **Four atoms:**
In order to build the controlled-$Z$ gate, we use one of the auxiliary states, ${|A_{1,2}G_{3,4}+A_{3,4}G_{1,2}\rangle}/\sqrt{2}$. Now, it is possible to drive only the transition between this state and ${|10\rangle}^\mathrm{L}$ without affecting the other states within the DFS by the choice $\Omega_n = 0$, $\Delta_3=\Delta_4 = 0$ and $\Delta_1 = -\Delta_2 \neq 0$. A $\pi$-pulse on the state ${|10\rangle}^\mathrm{L}$ leads to a relative phase of $-1$ on this state, i.e. $${|10\rangle}^\mathrm{L}_{1,3} \longrightarrow -{|10\rangle}^\mathrm{L}_{1,3},$$ for $\frac{1}{\sqrt{2}} \Delta_{12}^- T = 2 \pi$ without affecting the other states of the computational subspace. Hence, we have defined a a controlled controlled-$(-Z)$ gate which is equivalent up to single qubit unitaries to a CNOT-gate [@nielsen_book00a].
#### **For $N>4$ atoms:**
One can restrict the dynamics to the subspace of four atoms in a similar way as for the single-qubit rotations. With the choice of $$\Omega_n = 0, \ \mathrm{and} \ |\Delta_{12}^-| \ll \Delta_n =\Delta_\mathrm{D} \ll \Gamma_\oned , \ n\geq 5,$$ transitions to states over more than four atoms are far off-resonant. As before, this adds an error rate proportional to $\frac{|\Delta_{12}^- |^2}{\Delta_\mathrm{D}}$ with a proportionality factor depending on the coupling strength after the projection onto the DFS.
Error analysis: spontaneous emission and imperfect addressing. {#sec4}
==============================================================
So far we have considered only the interaction within the ideal Quantum Zeno Dynamics, where the only possible sources of error were due to the larger dimension of the DFS with respect to the computational space. In this section, we take into account other sources of errors that will be present in most of the implementations, namely, i) errors coming from spontaneous emission to other modes, with rate $\Gamma^*$, included through ${\cal L}_{*}\left[\rho\right] $ as in equation \[eq:SpEm\]; ii) errors from deviations from the Zeno Hamiltonian, attributed to photons emitted to the waveguide from the small population present in the states outside the DFS; iii) errors that may arise from an imperfect control of the laser parameters $\{\Omega_n,\Delta_n\}$. In what follows, we assume to work in a regime with $P_{\oned}\gg 1$, such that the following parameter hierarchy can be satisfied: $\Gamma^*\ll ||H_\mathrm{eff}|| \ll \Gamma_\oned$.
This section discusses, for each gate separately, the numerical results and their analytical approximation of the fidelity between the theoretical final (goal) state, ${|\psi_f\rangle}$, and the real atomic state, $\rho$, obtained after the gate operation, i.e., $F = \langle \psi_\mathrm{f} | \rho | \psi_\mathrm{f} \rangle^{1/2}$. The numerical results are obtained by solving the master equation in second order perturbation theory (see equation \[eq:effemastereq\]). We have checked numerically that this is a good approximation in the parameter ranges considered throughout this manuscript. To obtain the analytical approximations, we used the effective non-hermitian Hamiltonian that can be obtained from the second order master equation (see appendix A).
Phase Shift Gate
----------------
![Infidelities of single and two qubit gates for $N=4$ atoms. (a)-(d) Infidelity of a $\pi/2$ phase shift on the first logical qubit of four atoms on the state ${|10+00\rangle}^\mathrm{L}/\sqrt{2}$ for (a) $\Delta_{12}^- = 0$ and $\Delta_{34}^+ = 0$, (b) $\Delta_{12}^- = 0.1 \Delta_{12}^+$ and $\Delta_{34}^+ = 0$, (c) $\Delta_{12}^- = \Delta_{12}^+$ and $\Delta_{34}^+ = 0$, (d) $\Delta_{12}^- = \Delta_{12}^+$ and $\Delta_{34}^+ = 20 \Delta_{12}^+$. (e) Infidelity of a single qubit Pauli-X gate on the first qubit of four atoms on the state ${|00\rangle}^\mathrm{L}$ for $\Delta_{34}^+ = 10 \Omega_{12}^-$. (f) Infidelity of the controlled-$(-Z)$ gate for four atoms (2 qubits) acting on the state ${|10+11\rangle}^\mathrm{L}/\sqrt{2}$. The black lines in (d)-(f) represent the scaling of the coupling strength $\Delta_{12}^-$, $\Omega_{12}^-$ and $\Delta_{12}^-$, respectively, for the minimal infidelity with $P_\oned^{-1/2}$.[]{data-label="fig3"}](figure3v1.pdf){width="98.00000%"}
For the phase shift gate we must set $\Omega_n = 0$ for all $n$ and $\Delta_n = \Delta_\mathrm{D} \gg \Delta_{12}^-$ for all $n \geq 3$ to avoid errors from transitions to auxiliary states. By choosing $\Delta_1 = \Delta_2$, i.e., $\Delta_{12}^-=0$, no errors (from second order perturbation theory) occur because the computational states do not couple to the radiant ones. However, it is instructive to consider the errors that appear for situations where $\Delta_{12}^-\neq 0$, e.g, because of imperfect addressing, as this yields a useful understanding about how to deal with situations where the second order correction cannot be avoided.
#### **Two atoms:**
For the simplest situation the additional errors due to imperfect addressing, i.e., $\Delta_{12}^-\neq 0$, enter at a rate proportional to $|\Delta_{12}^-|^2/\Gamma_\oned$ through the same error channel as the spontaneous emission into all other modes with rate $\Gamma^*$, that is, via the quantum jump operator ${|0\rangle}^L {\langle 1|}$. Then, the infidelity, i.e., $1-F$, for a $\pi/2$ phase shift of the first logical qubit on the normalized state $\alpha {|0\rangle}^\mathrm{L} + \beta {|1\rangle}^\mathrm{L}$ can be approximated by $$1-F
\approx \frac{|\beta|^2}{4} \frac{\pi}{\Delta_{12}^+}
\left( \Gamma^* + 4 \frac{| \Delta_{12}^-|^2}{\Gamma_{1\mathrm{D}}} \right).
\label{eq:FidAppPh}$$ One observes, that in the ideal case, $\Delta_{12}^- = 0$, the infidelity can be arbitrarily close to $0$ for large $\Delta_{12}^+$. If $\Delta_{12}^-$ is not negligible, the transition strength $\Delta_{12}^+$ cannot be chosen arbitrarily large to decrease the infidelity. For example, in the worst case scenario where $\Delta_{12}^- = \Delta_{12}^+$ this results in an optimal infidelity scaling $\frac{|\beta|^2 \pi }{2} P_\oned^{-1/2}$ for $\Delta_{12}^+ = \Delta_{12}^-=\frac{1}{2} \sqrt{\Gamma^* \Gamma_\oned}$.
#### **Four atoms:**
A similar behaviour can be obtained by choosing $\Delta_{12}^- = 0$ such that the infidelity is arbitrarily close to $0$ (see figure \[fig3\]a). Slight deviations from this ideal value do not change this behaviour drastically (see figure \[fig3\]b). However, when $\Delta_{12}^-$ is not negligible, it leads to two types of errors that decrease the fidelity (see figure \[fig3\]c): i) virtual population of non-DFS states, which leads to an error rate proportional to $|\Delta_{12}^-|^2/\Gamma_\oned$ as for two atoms; and ii) transitions to auxiliary states, in particular ${|A_{1,2}G_{3,4}+A_{3,4}G_{1,2}\rangle}/\sqrt{2}$. The latter can be made far off-resonance by applying a detuning on the second qubit such that $|\Delta_{12}^-| \ll |\Delta_{34}^+| \ll \Gamma_\oned$. With a large off-resonance ratio $r_\Delta = |\Delta_{34}^+ / \Delta_{12}^-| \gg 1$, one still achieves a small infidelity (see figure \[fig3\]d).
As shown in figures \[fig3\]a-d, the detuning of the third and fourth atom is important when $\Delta_{12}^-$ cannot be neglected. As expected, the infidelity decreases by increasing the off-resonance ratio $r_\Delta$ (see figure \[figSM1\]a). For large enough $r_{\Delta}$, the infidelity can be analytically approximated by$$1-F
\stackrel{r_\Delta \rightarrow \infty}{\longrightarrow} \frac{\pi}{8}
\left( \frac{\Gamma^*}{|\Delta_{12}^+|} + 2 \frac{|\Delta_{12}^+|}{\Gamma_\oned} \right).$$ This leads to a minimal infidelity $\propto P_\oned^{-1/2}$ (see figure \[figSM1\]b) for $\Delta_{12}^+ = \Delta_{12}^- = \sqrt{\Gamma^* \Gamma_\oned/2}$.
![Dependence of infidelity on off-resonance ratio $r_\Delta=|\Delta_{34}^+ / \Delta_{12}^-|$ for $\pi/2$ phase shift on first qubit with $\Delta_{12}^- = \Delta_{12}^+$. (a) Infidelity varied over the coupling strength $\Delta_{12}^+$ for $P_\oned=10^4$. The results correspond to $r_\Delta=5$ (black), $r_\Delta=10$ (red) and $r_\Delta=15$ (blue). (b) Minimal infidelity depending on the Purcell Factor $P_\oned$. The numerical results (points) corresponding to the values as in (a) fit well with the approximation (line) of $(1-F)_\mathrm{min} \propto P_\oned^{-1/2}+ C(r_\Delta)$, where the $C(r_\Delta)$ is a number which does not depend on $P_\oned$.[]{data-label="figSM1"}](figure4v2.pdf){width="98.00000%"}
Pauli-X Gate
------------
For rotations around the $x$-axis, we set $\Delta_{1} = \Delta_2= 0$, and $\Delta_n = \Delta_\mathrm{D} \gg \Omega_{12}^-$ for all $n\geq 3$ to avoid transitions to auxiliary states. In contrast to the phase shift gate, even in the ideal case, $\Omega_{12}^+ = 0$, errors will occur because $\Omega_{12}^-$ couples to state outside the DFS, as shown schematically in figure \[fig2\](a) for the two atom case. Moreover, we also include a short discussion on deviations due to imperfect control on $\Delta_{1(2)} \neq 0$ and $\Omega_{12}^+$.
#### **Two atoms:**
Using $\Omega_{12}^+ = 0$, the error rate from deviations from the Zeno Hamiltonian enters in the same way as from the spontaneous emission into all other modes, that is via the quantum jump operator ${|0\rangle}_\mathrm{L}{\langle 1|}$. The corresponding decay rate is $( |\Delta_{12}^-|^2 + |\Omega_{12}^-|^2 / 2) /\Gamma_\oned$. The error from $\Omega_{12}^+ \neq 0$ enters differently, but can still be included in the estimation of the infidelity. Neglecting the errors from $\Delta_{12}^+ \neq 0$, the infidelity for a Pauli-$X$ gate ($\frac{1}{\sqrt{2}} |\Omega_{12}^-| T = \pi/2$) on state ${|1\rangle}_\mathrm{L}$ can be approximated by $$1-F
\approx \frac{1}{2} \frac{\pi}{\sqrt{2} |\Omega_{12}^-|} \left( \Gamma^* + \frac{| \Omega_{12}^- |^2}{2 \Gamma_{1\mathrm{D}}} + \frac{| \Delta_{12}^- |^2}{\Gamma_{1\mathrm{D}}} + \frac{| \Omega_{12}^+ |^2}{2 \Gamma_{1\mathrm{D}}}\right) \equiv \varepsilon_0 \,.$$ In the ideal case of perfect control of addressing parameters, i.e., $\Delta_{12}^- = \Omega_{12}^+ = 0$, the minimal value of the infidelity, proportional to $P_\oned^{-1/2}$, is obtained at $|\Omega_{12}^-| = \sqrt{2 \Gamma^* \Gamma_{1\mathrm{D}}}$ as shown in red circles of figure \[fig4\]. Note, that even for $\Omega_{12}^- = \Omega_{12}^+$ and $\Delta_{12}^- = 0$, the infidelity is still proportional to $P_\oned^{-1/2}$.
#### **Four atoms:**
In this case apart from the transitions out of the DFS, $\Omega_{12}^-$ also couples states inside the DFS, but out of the computational space (see figure \[fig2\]) such that we need to detune these processes to achieve the rotations. As already explained in the previous section, this can be done by setting $|\Omega_{12}^-| \ll |\Delta_{34}^+| \ll \Gamma_\oned$. As expected, the infidelity decreases when increasing the off-resonance ratio $r_\Omega= |\Delta_{34}^+ / \Omega_{12}^-|$ (see figure \[figSM2\]). For large enough ratios $r_\Omega (\gtrsim 4)$, the infidelity can be approximated by $$1-F \approx \varepsilon_0 + \frac{\alpha}{r_\Omega^2},$$ where the constant $\alpha = \mathcal{O}(1)$ can be obtained through a numerical fit. The infidelity of a $\pi/2$-pulse on the state ${|00\rangle}^\mathrm{L}$ is plotted in figure \[fig3\]e, whereas the minimal infidelity is shown to scale with $P_\oned^{-1/2}$ in figure \[fig4\].
![(a) Numerical calculation of infidelity of Pauli-X gate on the first qubit on the state ${|00\rangle}^\mathrm{L}$ for $N=4$ atoms with $\Delta_{34}^+ = 10 \Omega_{12}^-$ and $P_\oned = 10^4$ for different values of the off-resonance ratio, that is $r_\Omega=2$ (black), $r_\Omega=5$ (red) and $r_\Omega=10$ (blue). (b) Scaling of the minimal infidelity for the same values as in (a) for different Purcell Factors, that is $P_\oned = 10^3$ (black), $P_\oned = 10^4$ (red) and $P_\oned = 10^5$ (blue).[]{data-label="figSM2"}](figure5v2.pdf){width="98.00000%"}
Controlled Pauli-Z
------------------
For the controlled-$(-Z)$ gate, we set $\Omega_{n} = 0$ for all $n$, $\Delta_{12}^+ =0 $ and $\Delta_{n} = \Delta_\mathrm{D} \gg |\Delta_{12}^-|$ for all $n\geq 5$. As $\Delta_{12}^-$ couples ${|10\rangle}^\mathrm{L}$ and ${|11\rangle}^\mathrm{L}$ also to states outside the DFS, the fidelity shows a similar behaviour as the Pauli-X gate (see figure \[fig3\](f) for example with $N=4$), i.e., there is an optimal $\Delta_{12}^-$ that sets the maximum fidelity.
The infidelity can be approximated similarly to equation \[eq:FidAppPh\], i.e., after a controlled Pauli-$Z$ gate ($|\Delta_{12}^-| T/\sqrt{2} = \pi$) acting on the state $\left( {|10+11\rangle}^\mathrm{L} \right)/\sqrt{2}$ can be approximated by $$1-F \approx \frac{3 \pi}{2 \sqrt{2} |\Delta_{12}^-|}
\left(\Gamma^* + \frac{3}{4} \frac{| \Delta_{12}^- |}{\Gamma_{1\mathrm{D}}} \right),$$ which attains its minimal value, $3 \pi/\sqrt{2 P_\oned}$, for $|\Delta_{12}^-| = \sqrt{4 \Gamma^* \Gamma_\oned/3}$. As for the single qubit gates, the infidelity scales with $P_\oned^{-1/2}$, shown in blue circles figure \[fig4\].
Summary of analysis
-------------------
Summing up, from the explicit analysis with two and four TLS, we have shown both numerically and analytically that both the single-qubit rotations and the control (-Z) gate show a scaling of the infidelity as $P_\oned^{-1/2}$ (see figure \[fig4\]). Only for small values of the Purcell Factor $P_\oned$ does the minimal infidelity deviate slightly from the theoretical analysis because the hierarchy $\Gamma^* \ll \Omega_{12}^-, \Delta_{12}^- \ll \Gamma_\oned$ is no longer well satisfied.
Moreover, in the $N=4$ case, we also showed how to deal with the errors that come from the larger size of the DFS with respect to the computational one. For single qubit rotations in a system of four emitters, the choice $ |\Omega_{12}^-|, |\Delta_{12}^+| \ll |\Delta_{34}^+| \ll \Gamma_\oned$ ensures that the dynamics can be restricted to two atoms. In the extreme case where $|\Delta_{34}^+| \gg \Gamma_\oned$ the perturbation analysis is no longer valid. However, in this case the levels are so strongly shifted, that they are decoupled from the collective dissipation, so that the system can be described as a system of only two emitters. The same is true, if the emitters can be completely decoupled from the waveguide by other means available in a particular implementation. For more atoms the same arguments hold as the second order correction introduced from deviations from Zeno dynamics satisfies $$\|\PP \mathcal{L}_{\mathrm{pert}} \QQ \frac{1}{\mathcal{L}_\DD}\QQ \mathcal{L}_{\mathrm{pert}} \PP \rho \|
\ll \frac{\| \mathcal{L}_{\mathrm{pert}} \|^2}{\Gamma_\oned},$$ where $\mathcal{L}_{\mathrm{pert}}$ is the perturbation to the purely collective decay and $\| \cdot \|$ denotes the maximum norm. This is independent of the atom number $N$, because $\| \mathcal{L}_{\mathrm{pert}} \|$ does not increase with the number of atoms for one and two-qubit gates. So there is an upper limit on the second order correction, which leads to the $P_\oned^{-1/2}$-scaling. Finally, the error rate stemming from spontaneous emission of the logical states ${|1\rangle}_L={|A\rangle}$ is proportional to the number of excited states in the system. Therefore, the gate fidelity does depend on the full state, and can be upper bounded by considering the worst-case state, that is the state with ${|1\rangle}_L$ in all other computational qubits, which indeed will depend on the atom number.
![Scaling behaviour of the minimal infidelity for the Pauli-X (red squares) and controlled-$(-Z)$ gates (blue circles) with the same parameters as in figure \[fig3\]e-f. The scaling fits well with the scaling $P_\oned^{-1/2}$ (black line) for large enough values of the Purcell Factor $P_\oned$. For the phase shift gate the infidelity is arbitrarily close to $0$ in the ideal case.[]{data-label="fig4"}](figure6v2.pdf){width="49.00000%"}
Further error analysis: finite propagation length of 1d modes. {#sec5}
==============================================================
For completeness, it is interesting to consider another source of error that may be very relevant for some implementations with short propagation lengths, e.g., plasmonic waveguides [@chang07a; @dzsotjan10a; @gonzaleztudela11a; @bermudez15a]. The finite propagation length enters into the decay matrix [@gonzaleztudela11a] as $$\Gamma_{n,m} = \frac{\Gamma_\oned}{2} \ee^{\ii q(\omega_\mathrm{a}) |z_n - z_m|}
\ee^{- |z_n - z_m|/L_\mathrm{prop}}
= \frac{\Gamma_\oned}{2} \ee^{- x |n-m|},$$ if the atoms are equidistantly placed a multiple of a wavelength apart, $d$ and where we introduced $x = d/L_\mathrm{prop}$ as the perturbation parameter. For simplicity, we restrict our discussion to the case with $N=2$, where analytical expressions can be obtained. In that situation, the finite propagation length only leads to the replacements $$\begin{aligned}
\Gamma^* \rightarrow& \Gamma^* + \Gamma_\oned (1-\ee^{-x}) \approx \Gamma^* + \Gamma_\oned x, \\
2 \Gamma_\oned \rightarrow& \Gamma_\oned \left(2-(1-\ee^{-x})\right) \approx \Gamma_\oned \left(2-x \right)\,,\end{aligned}$$ when $x\ll 1$. Therefore, the scaling of the infidelity is then given by $$1-F
\propto \sqrt{\frac{\Gamma^* + \Gamma_\oned x}{\Gamma_\oned \left(2-x \right)}}
\approx P_\oned^{-1/2} + \frac{1}{2} P_\oned^{1/2} x,$$ which scales as $1/\sqrt{P_{\oned}}$ as long as $x P_\oned \ll 1$,that is, that the distance between neighbouring emitters satisfies $d=|z_n-z_{n+1}|\ll L_\mathrm{prop} P_\oned^{-1}$. For more atoms, it is difficult to obtain the analytical scaling as the superradiant state is not an eigenstate of the modified decay matrix and thus the DFS states change as well. However, because the finite propagation length enters as $\rme^{-|z_m -z_n|/L_\mathrm{prop}} \approx 1 - |z_m -z_n|/L_\mathrm{prop}$ it can be treated as a perturbation to the Liouvillian of equation \[Dicke\] that will be kept small as long as $Nd\ll L_{\mathrm{prop}}$ [@gonzaleztudela15a].
Depending on the particular implementation other errors have to be considered, e.g., for atoms trapped close to a dielectric waveguide the separation condition $|z_n-z_m| = n 2\pi /q(\omega_{{\rm a}})$ might not be satisfied exactly or because its position is changing over time due to atomic motion. However, its main effect can be approximated as an effective increase of $\Gamma^*$ that is small with current state of the art parameters for photonic crystal waveguides, as discussed in reference [@gonzaleztudela15a].
Comparison to Three-Level Atoms {#sec6}
===============================
The use of the DFS of atomic $\Lambda$-systems in cavity QED setups has already been considered in detail in the literature [@beige00a; @beige00b; @tregenna02a]. In that case, a three-level system with a $\Lambda$-type level structure is used to define a logical qubit in the two metastable states ${|0\rangle}$ and ${|1\rangle}$. The excited state ${|e\rangle}$ decays to one of the metastable states, say ${|1\rangle}$. When two atoms are inside the cavity an additional decoherence-free state emerges, i.e., $\left({|1e\rangle}-{|e1\rangle}\right)/\sqrt{2}$, that can be used to define a CNOT gate in the so-called bad-cavity limit, where the atom-cavity coupling ($g$) is smaller than the cavity losses $\kappa$, but the decay into the cavity ($g^2/\kappa$) is still bigger than into the rest of the decay channels ($\Gamma^*$). The ratio between the good/bad processes is the so-called cooperativity $C=\frac{g^2}{\kappa \Gamma^*}$, which therefore plays a similar role as $P_{\mathrm{1d}}$ in our proposal. The errors in the CNOT gate come both from $\Gamma^*$, and from deviations from the Zeno Hamiltonian, giving rise to an optimal infidelity proportional to $1/\sqrt{C}$, which is similar to the one that we found using only TLSs.
We note that using TLS the computational qubits have a finite lifetime compared to the implementations using atomic metastable states. However, i) there are situations in which one would like to use gates to build a given atomic state within the DFS in order to map it immediately into a photonic state in the waveguide [@gonzaleztudela15a] such that long lifetimes are not required; ii) some of the implementations have extremely long-lived qubits, e.g. superconducting systems [@mlynek14a]. iii) Moreover, if $\Lambda$ schemes are available, as in atoms, we can also implement our single and two-qubit gates with metastable states with the equivalence shown in figure \[fig1\]c. In that case, our proposal just constitutes a complementary way of doing universal quantum computation within DFS.
Conclusion & Outlook.
=====================
Summing up, we have shown how to implement a universal set of quantum gates using the decoherence-free subspaces appearing within TLS interacting with one-dimensional photon-like reservoirs. We have given an explicit construction of single and two-qubit gates for logical qubits defined in the DFS and analyzed possible sources of errors such as spontaneous emission to other modes, coupling to states outside of the DFS, imperfect addressing and finite propagation lengths. Through both analytical and numerical analysis, we have shown the fidelities of the gates scale generally with $(1-F)_\mathrm{min}\propto P_\oned^{-1/2}$, analogous to the one using $\Lambda$ schemes [@beige00a; @beige00b; @tregenna02a]. Thus, this work widens up the zoology of quantum emitters that can be used to implement quantum gates within waveguide QED setups. An interesting outlook for the application of these gates is to use them for generating entangled states of many emitters within the DFS, which afterwards can be mapped into waveguide multiphoton states in a very efficient way [@gonzaleztudela15a].
We gratefully acknowledge discussions with I. Cirac. The work of AGT and VP was funded by the European Union integrated project *Simulators and Interfaces with Quantum Systems* (SIQS). AGT also acknowledges support from Alexander Von Humboldt Foundation and Intra-European Marie-Curie Fellowship NanoQuIS (625955). HJK acknowledges funding by the Institute of Quantum Information and Matter, a National Science Fundation (NSF) Physics Frontier Center with support of the Moore Foundation, by the Air Force Office of Scientific Research, Quantum Memories in Photon-Atomic-Solid State Systems (QuMPASS) Multidisciplinary University Research Initiative (MURI), by the Department of Defense National Security Science and Engineering Faculty Fellows (DoD NSSEFF) program, by NSF PHY1205729 and support as a Max Planck Institute for Quantum Optics Distinguished Scholar.
Appendix A: Second order Liouvillian perturbation theory.\[sec:apA\] {#appendix-a-second-order-liouvillian-perturbation-theory.secapa .unnumbered}
====================================================================
When the system evolves under a very strong collective decay, the driving term $H_\mathrm{las}$ and the decay into other bath modes may be treated as a perturbation to the collective dissipation given by $\mathcal{L}_\mathrm{D}$ [@reiter12a; @kessler12b]. In order to describe these perturbations as generally as possible, we denote them by a Liouvillian $\mathcal{L}_\mathrm{pert}$, and assume that it has a relevant timescale $\tau$. If the timescale satisfies, $\tau\gg 1/\Gamma_{\oned}$, the dynamics of the atomic system can be formally projected into the DFS of the Liovillian ${\cal L}_\DD$, by using a projector operator $\mathbb{P}$ satisfying: $\mathbb{P}{\cal L}_\DD={\cal L}_\DD \mathbb{P}=0$. This projector can be found via the right (left) eigenoperators $\rho_{ij}$ ($\chi_{ij}$) corresponding to the eigenvalue 0 of $\mathcal{L}_\mathrm{D}$, which are combined to $$\mathbb{P}\rho = \sum_{i,j} \rho_{ij} \langle \chi_{ij}, \rho \rangle,$$ where $\langle A,B \rangle = \Tr \left(A^\dagger B\right)$ is the inner product on the space of density matrices. The orthogonal eigenoperators are indexed such that $ \langle \chi_{ij}, \rho_{kl} \rangle = \delta_{i,k} \delta_{j,l} = \langle \rho_{ij}, \rho_{kl} \rangle$. We also define the orthogonal part of $\mathbb{P}$, $\mathbb{Q}=1-\mathbb{P}$. The left eigenoperators can be derived from the right ones by $$\chi_{ij} = \rho_{ij} + \alpha_{ij}^{(1)} S_{eg} \rho_{ij} S_{ge} + \alpha_{ij}^{(2)} S_{eg} S_{eg} \rho_{ij} S_{ge} S_{ge} + \dots,$$ where the coefficients $\alpha_{ij}^{(n)}$ are determined by $\langle \chi_{ij}, \mathcal{L}_\DD \rho \rangle = 0$. With this choice, the projector is independent of the choice of $\rho_{ij}$ and hence, one can choose $\rho_{ij} = {|\mathrm{d}_i\rangle} {\langle \mathrm{d}_j|}$, where ${|\mathrm{d}_i\rangle}$ are orthonormal states from the DFS. Using these projectors, one can formally integrate out the fast dynamics outside the DFS, described by $\QQ \rho$: $$\begin{aligned}
\frac{\dd}{\dd t} \QQ \rho
&= \mathbb{Q}\ \left( \mathcal{L}_\DD + \mathcal{L}_\mathrm{pert} \right) \mathbb{Q} \rho + \mathbb{Q} \mathcal{L}_\mathrm{pert} \mathbb{P} \rho \, , \\
\mathbb{Q} \rho(t)
&= \int_0^t \rmd \tau\ \mathrm{exp}\left[ \mathbb{Q} (\mathcal{L}_\DD+ \mathcal{L}_\mathrm{pert}) \mathbb{Q} (t-\tau) \right]\mathbb{Q} \mathcal{L}_\mathrm{pert} \mathbb{P} \rho(\tau) \nonumber \\
& \approx \mathbb{Q} (-\mathcal{L}_\DD^{-1} ) \mathbb{Q} \mathcal{L}_\mathrm{pert} \mathbb{P} \rho + \mathcal{O}(\tau^{-2} / \Gamma_\oned^2),
\label{eq:Qrho}\end{aligned}$$ where the last approximation is obtained by i) applying a Markov approximation $\rho(\tau) \approx \rho(t)$ in the integral; ii) neglecting terms of higher order in $\tau^{-1}/\Gamma_\oned$; and iii) extending the integral to infinity.
Plugging this into the equation for the DFS-part of the state, that is $$\label{eqA:meq}
\frac{\dd}{\dd t} \PP \rho
= \PP \mathcal{L}_\mathrm{pert} \PP \rho + \PP \mathcal{L}_\mathrm{pert} \QQ \rho
= \mathcal{L}_\mathrm{eff} \PP \rho ,$$ yields an effective Liouvillian of the atomic system within the DFS given (up to second order in $\tau^{-1}/\Gamma_\oned$) by
$$\mathcal{L}_\mathrm{eff}
= \PP \mathcal{L}_\mathrm{pert} \PP
+\PP \mathcal{L}_\mathrm{pert} \QQ \frac{1}{{-\cal L}_\DD}\QQ \mathcal{L}_\mathrm{pert} \PP
+\mathrm{O}\left(\tau^{-3}/\Gamma_\oned^2 \right)\,.$$
The first order of this Liouvillian, i.e., $\PP \mathcal{L}_\mathrm{pert} \PP$, is the effective evolution induced within the DFS induced by the strong collective dissipation. This is commonly referred to as the ideal Quantum Zeno dynamics [@zanardi97a; @lidar98a; @facchi02a] as it can be understood as the effective dynamics enforced by the continuous monitoring of the atomic system due to the interaction of the waveguide modes. The second order term stems from slight population of (super)radiant modes that generates some corrections on the ideal Quantum Zeno dynamics.
It is instructive to write the effective master equation derived in Eq. \[eqA:meq\] in a form that separates the non-hermitian evolution dynamics and the contribution coming from quantum jump processes. For our particular situation, considering the perturbation of $\mathcal{L}_*$ and $H_\mathrm{las}$ as defined in the main text, it can be shown after some algebra that $$\label{eqA:effmas}
\dot{\rho}
= - \ii \left[H_\mathrm{eff},\rho \right]
+ \mathbb{P} \mathcal{L}_* \rho
+ \mathbb{P} \left( o_1 \rho o_2^\dagger + o_2 \rho o_1^\dagger \right)
- o_2^\dagger o_1 \rho - \rho o_2^\dagger o_1,$$ where $H_\mathrm{eff}=\mathcal{P} H_\mathrm{las} \mathcal{P}$, where we used the projection $\mathcal{P}$ \[$\mathcal{Q}$\] for pure states inside \[outside\] of the DFS as defined in the main text. Furthermore, $o_1 = \mathcal{Q} \left(\frac{\Gamma_\oned}{2} S^+ S^-\right)^{-1} \mathcal{Q} H_\mathrm{las} \mathcal{P}$, $o_2 =\mathcal{Q} H_\mathrm{las} \mathcal{P}$ can be obtained by noting that the second order term reduces to simple matrix multiplication in the corresponding subspace because $\mathbb{P} \mathcal{P} A \mathcal{Q} = 0 = \mathbb{P} \mathcal{Q} A \mathcal{P}$ for all operators $A$.[^3] Although this does not look like a Liouvillian in Lindblad form, it is trace-preserving, as $$\Tr \mathbb{P} A
= \sum_{i,j} \Tr \rho_{ij} \langle \chi_{ij}, A \rangle
= \sum_i \langle \chi_{ii}, A \rangle
= \Tr A,$$ because $\sum_{i} \chi_{ii} = \mathbf{1}$. From Eq. \[eqA:effmas\], it is straightforward to define a non-Hermitian Hamiltonian from the above master equation, that is $$H_\mathrm{nh}
= \mathcal{P} \left( H_\mathrm{eff}
- \ii \frac{\Gamma^*}{2} \sum_n \sigma^n_{eg} \sigma_{ge}^n - \ii o_2^\dagger o_1 \right)\mathcal{P},$$ which describes the no-jump evolution, and that we use to get the analytical estimations of the infidelity.
References {#references .unnumbered}
==========
[^1]: In fact, this argument can be reversed to excite the auxiliary state from the computational state ${|10\rangle}^\mathrm{L}$ with the choice $\Delta_{34}^+ = 0$ and $\Delta_{12}^+ \gg \Omega_{12}^-$.
[^2]: The auxiliary states necessarily extend over more than two atoms, because it is orthogonal to the logical qubits and therefore contains excited (superradiant) triplet states in the “pairing” of the atoms. An antisymmetric combination of such states can be in the DFS, but not in the computational subspace, and necessarily extends over multiple atom “pairs”.
[^3]: We want to note that if it was possible to detect the photons emitted to the waveguide and post-select, the term $\mathbb{P} \left( o_1 \rho o_2^\dagger + o_2 \rho o_1^\dagger \right)$ would vanish and the system would be described by pure states except for the spontaneous emission into non-guided modes.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We prove asymptotically isometric, coarsely geodesic metrics on a toral relatively hyperbolic group are coarsely equal. The theorem applies to all lattices in $SO(n,1)$. This partly verifies a conjecture by Margulis. In the case of hyperbolic groups/spaces, our result generalizes a theorem by Furman and a theorem by Krat.
We discuss an application to the isospectral problem for the length spectrum of Riemannian manifolds. The positive answer to this problem has been known for several cases. All of them have hyperbolic fundamental groups. We do not solve the isospectral problem in the original sense, but prove the universal covers are $(1,C)$-quasi-isometric if the fundamental group is a toral relatively hyperbolic group.
author:
- 'Koji Fujiwara[^1]'
title: Asymptotically isometric metrics on relatively hyperbolic groups and marked length spectrum
---
\[section\] \[thm\][Lemma]{} \[thm\][Corollary]{} \[thm\][Proposition]{} \[thm\][Question]{}
\[thm\][Remark]{} \[thm\][Important Remark]{} \[thm\][Definition]{} \[thm\][Example]{} \[thm\][Fact]{} \[thm\][Convention]{}
§[[S]{}]{} Ł[[L]{}]{}
Introduction
============
Asymptotically isometric metrics
--------------------------------
Suppose a group $G$ acts on a space $X$ with two metrics $d_1$ and $d_2$ that are $G$-invariant. We say $(X,d_1)$ and $(X,d_2)$ are
1. [*coarsely equal*]{} if there exists $C$ such that for all $x,y \in X$, $$|d_1(x,y) -d_2(x,y) | \le C.$$
2. [*asymptotically isometric*]{} (or jsut [*asymptotic*]{}) if $d_1(x,y) \to \infty$ if and only if $d_2(x,y) \to \infty$, and $$\frac{d_1(x,y)}{d_2(x,y)} \to 1 \,\,\, {\rm as} \,\,\, d_2(x,y) \to \infty.$$
3. [*weakly asymptotically isometric*]{} (or [*weakly asymptotic*]{}) if $(X,d_1)$ and $(X,d_2)$ are quasi-isometric, $g \in G$ is hyperbolic for $d_1$ if and only if it is hyperbolic for $d_2$, and that for every $g \in G$ that is hyperbolic, we have $$\lim _{n \to \infty} \frac{d_1(x, g^n(x))}{d_2(x,g^n(x))}=1.$$ This property does not depend on the choice of $x$.
4. $d_1$ and $d_2$ have the [*same marked length spectrum*]{} with respect to the $G$-actions if $|g|_1=|g|_2$ for all $g \in G$, where $|g|_i$ is the [*translation length*]{} of $g$ for $d_i$ defined by $$|g|_i= \lim_{n\to \infty} \frac{d_i(x,g^n(x))}{n}.$$ $g$ is [*hyperbolic*]{} on $(X,d_i)$ if and only if $|g|_i>0$.
Clearly, (1) $\Rightarrow$ (2) $\Rightarrow$ (3) $\Leftrightarrow$ (4). We discuss the other implications in this paper. Some remarks are in order. (1) is same as the identity map is a $(1,C)$-quasi-isometry. This is stronger than $d_1$ and $d_2$ are $(1,C)$-quasi-isometric. Even if $d_1$ and $d_2$ are isometric, (1) may not hold. (2) implies that the identity map is a quasi-isometry. In view of that, if we concern the implication from (3) to (1) or (2), then we should look at $G$-equivariant maps from $(X,d_1)$ to $(X,d_2)$, not the identity map.
We write $A \sim _C B$ if $|A-B|\le C$. A metric space $(X,d)$ is [*$C$-coarsely geodesic*]{} if for any $x,y \in X$, there is a path $\gamma$ from $x$ to $y$ such that for all $t, s$, we have $|t-s| \sim _C |\gamma(t)-\gamma(s)|$. $\gamma$ does not have to be continuous. If $C=0$, $X$ is geodesic. We may suppress the constant $C$ and just say [*coarsely geodesic*]{}.
We write $d_1 \sim _C d_2$ if for all $x,y$ we have $d_1(x,y) \sim _C d_2(x,y)$. We may simply write $d_1 \sim d_2$. This is nothing but they are coarsely equal.
Recall that a map $f : X \to Y$ between two metric spaces $(X, d_X)$ and $(Y, d_Y )$ is called a $(L,C)$-[*quasi-isometry*]{} if $Ld_X(a, b) - C \le d_Y (f(a), f(b)) \le d_X(a, b)/L + C$, for all $a, b \in X$, and every $y \in Y$ is at distance at most $C$ from some element of $f(X)$. If such $f$ exists for some $L \ge 1, C\ge 0$, then $X$ and $Y$ are quasi-isometric.
Burago [@burago] proved for $G={\Bbb Z}^n$ and $X=\R^n$, (2) $\Rightarrow$ (1) for $G$-invariant Riemannian metrics. His argument applies to a pair of coarsely geodesic metrics on ${\Bbb Z}^n$ (Corollary \[burago2\]).
Krat [@krat] proved an analogous result when $X$ is $\delta$-hyperbolic, in particular, which implies (2) $\Rightarrow$ (1) for two left invariant metrics on a hyperbolic group $G$ that are quasi-isometric to a word metric. Furman [@furman] proved (3) $\Rightarrow$ (1) in the same setting. His argument is different from hers. We will modify her argument and prove (3) $\Rightarrow$ (1) for toral relatively hyperbolic groups, which are more general than hyperbolic groups.
Abels and Margulis [@abels] proved (2) $\Rightarrow$ (1) for “word metrics" on reductive Lie groups. It is asked by Margulis in [@margulis] if (2) $\Rightarrow$ (1) holds in general on a finitely generated group $G$. Breuillard [@breu] answered this question in the negative. He found a counter example, which are two word-metrics on $H_3(\Bbb Z) \times \Bbb Z$, where $H_3(\Bbb Z)$ is the three dimensional discrete Heisenberg group. In fact, the two metrics are not coarsely equal on some cyclic (undistorted) subgroup in his example. Also, those two metrics are not even $(1,C)$-quasi-isometric to each other for any $C$, [@breu2]. It was known that (2) $\Rightarrow$ (1) on $H_3(\Bbb Z)$, [@krat].
Main results
------------
An isometric action on a metric space $(X,d)$ by a group $G$ is [*cobounded*]{} if there exists a bounded set $B$ in $X$ such that $G.B=X$, and [*proper*]{} if for any $x \in X$ and $R>0$ there exist at most finitely many $g \in G$ with $d(x, g.x) \le R$.
The following is the main result.
Assume $(G, \mathcal H)$ is a relatively hyperbolic group such that for each $H_i \in \mathcal H$, $H_i$ contains $\Z^{n_i}$ as a finite index subgroup. Assume $G$ acts on $X$ properly and co-boundedly by isometries for geodesic metrics $d_1, d_2$ (or more generally, $d_2$ is a coarsely geodesic metric). If they are weakly asymptotically isometric, then they are coarsely equal.
The following are examples of toral relatively hyperbolic groups (see [@hruska] and Theorem 1.2.1 therein):
- all lattices in $SO(n,1)$ (uniform ones are hyperbolic).
- CAT(0) groups with isolated flats ([@hruska]). In particular, the fundamental group of a closed, irreducible $3$-manifold such that each piece of its JSJ-decomposition is atoroidal (namely, hyperbolic).
- Limit groups in the sense of Sela.
It seems it is an open question if the conclusion of the theorem holds for non-uniform lattices in the Lie group $SU(n,1)$. See the discussion in Section \[complex\].
A merit to show (3) $\Rightarrow$ (1) is it has an application to the marked length spectrum problem since (3) $\Leftrightarrow$ (4).
Let $(M_1,d_1), (M_2,d_2)$ be closed Riemannian manifolds with the isomorphic fundamental group $G$ that is toral relatively hyperbolic. Assume they have the same marked length spectrum. Then there is a $G$-equivariant $(1,C)$- quasi-isometry map $f:\tilde M_1 \to \tilde M_2$.
It is easy to see that $M_1$ and $M_2$ have the same marked length spectrum if the conclusion in the corollary holds. Therefore we rephrase the marked length spectrum problem as follows. If there is a $G$-equivariant $(1,C)$- quasi-isometry map from $\tilde M_1$ to $\tilde M_2$, then is $M_1$ isometric to $M_2$ ? Notice that if $C=0$, then $M_1$ and $M_2$ are isometric.
The iso-spectral problems for the marked length spectrum has been solved for several families of Riemannian manifolds, but in all of those cases, the fundamental group is hyperbolic (see Section \[marked\]). The novel part of our result is that we put the marked length spectrum problem into context for a broader class of groups.
We close the introduction with a discussion on a question by Gromov. In [@gromov.asym 2C$_2$(c)] he asks if the Hausdorff distance of two manifolds $X_1$ and $X_2$ is finite if they are acted by $G$ properly and co-compactly and that “$AL \, {\rm Dist}(X_1,X_2)=0$". Here, AL is for asymptotically Lipschitz and $AL \, {\rm Dist}(X_1,X_2)=0$ if for any $a>0$ there is a $(1+a, C_a)$-quasi-isometry $f_a$ between $X_1$ and $X_2$ for some $C_a$. Our property (2) implies $AL \, {\rm Dist}(X_1,X_2)=0$ since we can take the identity map as $f_a$ for any $a>0$ with sufficiently large $C_a$. (1) implies that the Hausdorff distance between $X_1$ and $X_2$ is at most $C$ via the identity map. Our result affirmatively answers a version of the question by Gromov with both the assumption and the conclusion $G$-equivariant when $G$ is toral relatively hyperbolic.
The case of hyperbolic spaces
=============================
We first prove the following. Krat [@krat] proved the result under the assumption that $d_1, d_2$ are asymptotic, but our assumption is weaker.
\[krat\] Let $d_1, d_2$ be coarsely geodesic metrics on $X$ on which $G$ acts by isometries, co-boundedly with respect to both $d_1$ and $d_2$. Suppose $(X,d_1)$ is $\delta$-hyperbolic. Assume that $d_1$ and $d_2$ are weakly asymptotically isometric. Then, $d_1 \sim d_2$.
The outline of our argument is same as the one by Krat. Define $\D(x,y)=d_1(x,y)-d_2(x,y)$. To argue by contradiction, assume $\D$ is not bounded. We will prove that then $\liminf_{n} \D(x,g^n(x))/d_2(x,g^n(x)) \not= 0$ as $n \to \infty$ for some $g \in G$. This contradicts to the assumption.
Here are two elementary lemmas. The first one is straightforward from the triangle inequality for $d_1$ and $d_2$.
\[triangle\] Let $x,y,z$ be points in $X$. Then,
$|\D(x,y) - \D(x,z) | \le d_1(y,z) +d_2(y,z)$.
\[basic.additive\] Let $\gamma_1$ be a $d_1$-geodesic and $\gamma_2$ a $d_2$-geodesic from $x$ to $y$. Let $z \in \gamma_1$ be a point such that there exists $z' \in \gamma_2$ with $d_2(z,z')\le C$. Then, $\D(x,z)+\D(z,y) \sim _{2C} \D(x,y)$. Moreover, the conclusion holds if $\gamma_1$ and $\gamma_2$ are $L$-coarse geodesics (with a constant that is larger than $2C$ depending on $L$).
We say that $\D$ is [*almost additive at $z$*]{}. Suppose $\gamma_1, \gamma_2$ are geodesics. $\D(x,z)=d_1(x,z)-d_2(x,z) \sim_{C} d_1(x,z) -d_2(x,z')$, and similarly $\D(z,y) \sim_{C} d_1(z,y)-d_2(z',y)$. Now, $\D(x,z)+\D(z,y) \sim_{2C}
d_1(x,z) -d_2(x,z') +(d_1(z,y)-d_2(z',y))=d_1(x,y)-d_2(x,y)=\D(x,y).$
A similar argument applies when $\gamma_i$ are coarse geodesics and we omit details.
We go back to the proof of the theorem. In the following we assume that $d_1$ and $d_2$ are geodesic metrics. We can easily modify each argument when they are only coarsely geodesic with extra constants, but we leave it to the readers.
Since $d_1$ and $d_2$ are asymptotic, therefore quasi-isometric to each other, any $d_1$-geodesic is a $d_2$-quasi-geodesic with a controlled quasi-isometric constants, vice-versa. There exists a constant $C$ (by the Morse lemma, [@bridson Theorem 1.7]) such that a $d_1$-geodesic and a $d_2$-geodesic with the same endpoints are in the $C$-neighborhood (for both $d_1$ and $d_2$) of each other. Therefore by Lemma \[basic.additive\], $\Delta$ is almost additive, for a uniform constant, on any $d_1$-geodesic at any point.
By assumption, a metric ball of radius, say, $D$ covers $X$ by the $G$-action. Fix a base point $x$ and we write $g(x)$ as $g$. Write $\Delta(1,g)$ as $\D(g)$. Notice $\D(g)=\D(g^{-1})$. By assumption there is $g$ with $\D(g) >> \delta, C, D$. Take a $d_1$-geodesic $\gamma$ from $1$ to $g$, then there is $h\in G$ with $\D(h)$ is approximately $\D(g)/2$. This is possible since $\D$ is almost continuous on a geodesic. Set $k=h^{-1}g$. Then $\D(k)$ is approximately $\D(g)/2$ by Lemma \[basic.additive\] and Lemma \[triangle\]. Let $[p,q]$ denote a $d_1$-geodesic from $p$ to $q$. For $g \in G$, let $g^*.x$ denote the piecewise geodesic $\cup_{n \ge 0}[g^n.x, g^{n+1}.x]$ starting at $x$.\
[**Claim**]{}. At least one of the following piecewise geodesic is a $(2,10\delta+D)$-quasi-geodesic on $X$: $k^*.x$ or $h^*.x$ or $(hk)^*.x$.
This is a standard fact. See Lemma 8.1.A in [@gromov]. Since the points $1,h,g=hk$ is almost on a geodesic, $d_1(x,g.x) \sim d_1(x,h.x) + d_1(h.x,g.x)$ and both $d_1(x,h.x)$ and $d_1(h.x, g.x)$ are large, since $\Delta(g)$ is large. This is enough to apply Lemma 8.1.
This claim is the new ingredient than the argument by Krat, who stated that a piecewise geodesic of a similar property, not necessarily periodic, exists (Lemma 1.12). The periodicity is crucial under our assumption.
Let $f^*.x$ be one of the paths we obtain from the claim. On the path, $\D(x,f^n.x)$ grows roughly linearly on $n$ since on each geodesic piece it increases at least by $\D(h)-2C$ or $\D(g)-2C$, depending on $f^*.x$, by Lemma \[basic.additive\]. We used that $\D(h) \sim \D(k) >> \delta, C, D$. It follows $\liminf_n \D(x,f^n(x))/d_1(x,f^n(x)) >0$.
Toral relatively hyperbolic groups
==================================
In this section we will generalize Theorem \[krat\] to relatively hyperbolic groups.
Asymptotically tree graded spaces and relatively hyperbolic groups
------------------------------------------------------------------
We review a few key notions and results from [@DS]. Let $X$ be a complete geodesic metric space and let $\mathcal P$ be a collection of closed geodesic subsets (called [*pieces*]{}) with the following properties ([@DS]):\
(T1) every two different pieces have at most one common point.\
(T2) Every simple geodesic triangle (a geodesic triangle that is a simple loop) in $X$ is contained in one piece.
Let $X$ be a metric space and $\mathcal A$ a collection of subsets in $X$. $X$ is [*asymptotically tree-graded with respect to $\mathcal A$*]{} if every asymptotic cone, $Con^{\omega}(X)$, of $X$ is tree-graded with respect to a certain collection of subsets $\mathcal A_{\omega}$, which are defined from the collection $\mathcal A$ (see [@DS Definition 3.19] for the precise definition). Here, $\omega$ is an ultra filter. See the definition of the asymptotic cone in [@DS Definition 3.8]. In this paper, we do not use the definitions of (asymptotically) tree-graded spaces, but only quote geometric properties of those spaces from [@DS] and [@sisto]. We will state them later.
There is more than one way to define relatively hyperbolic groups. The following definition is one of the main theorems in [@DS]. We say a finitely generated group $G$ is [*relatively hyperbolic*]{} with respect to a collection of subgroups $\mathcal H=\{H_1, \cdots, H_m\}$ if $G$ is asymptotically tree-graded with respect to the subgroups $\mathcal H$, namely, the Cayley graph of $G$, $\Gamma(G,S)$, with respect to some (and any) finite set, $S$, of generators is asymptotically tree-graded with respect to the collection of left cosets $\{gH_i | g \in G, i=1, \cdots, m\}$. The subgroups $H_i$ are called [*peripheral*]{} subgroups. If they are all finitely generated virtually abelian groups, $G$ is called a [*toral*]{} relatively hyperbolic group. We do not assume that a toral relatively hyperbolic group is torsion-free.
Farb [@farb] defined that $G$ is [*weakly relatively hyperbolic*]{} with respect to $\mathcal H$ if the Cayley graph $\Gamma(G,S \cup \mathcal H)$ is hyperbolic, where $S \cup \mathcal H$ means the union of the elements in $S$ and the elements in $H_i$. Drutu-Sapir [@DS Theorem 8.6] proved the relative hyperbolicity in the above sense implies the weak relative hyperbolicity.
Toral relatively hyperbolic groups
----------------------------------
We generalize Theorem \[krat\] to toral relatively hyperbolic groups.
\[main\] Assume $(G, \mathcal H)$ is a relatively hyperbolic group such that for each $H_i \in \mathcal H$, $H_i$ contains $\Z^{n_i}$ as a finite index subgroup. Assume $G$ acts on $X$ properly and co-boundedly by isometries for geodesic metrics $d_1, d_2$ (or more generally, $d_2$ is a coarsely geodesic metric). If they are weakly asymptotically isometric, then they are coarsely equal.
Since $d_2$ may be coarsely geodesic, the theorem for example applies to two word metrics on $G$ (take the Cayley graph for one of the two metrics as $X$). We assume $d_1$ is geodesic since we apply [@DS] and [@sisto] to $d_1$. They only discuss geodesic metrics. It looks likely that the results we use from those two papers hold for coarsely geodesic metrics.
Notice that $\Gamma(G,S)$ and $X$ are quasi-isometric with respect to both $d_1, d_2$ ([@bridson Proposition 8.19]. The statement therein is for co-compact group actions on a geodesic space, but the argument uses only co-boundedness, and also it applies to a coarse geodesic space).
In the proof of the theorem we will first apply the theorem by Burago [@burago] to $(H,d_1|H), (H, d_2|H)$, and want to conclude $d_1 \sim d_2$ on $H$. A little issue is that $d_i|H$ is not a geodesic metric. So we first modify his result then argue that the modification is enough for us.
\[burago\] Let $d$ be a coarsely geodesic metric on $\Z^n$ that is invariant by the left action of $\Z^n$. Then
1. for each $g \in \Z^n$, $\lim_{n \to \infty} \frac{d(1,g^n)}{n}$ exists. We write the limit by $|g|_d$.
2. There is a constant $C$ such that for all $g$, $|d(1,g) - |g|_d| \le C$.
3. $|d(1,g^n) - n |g|_d| \le C$ for all $n>0$ and $g$.
We first prove that the limit exists. The argument is a modification of the proof of [@burago Theorem 1]. We explain the change we need. We only need to modify Lemma 4 in [@burago].
Embed $\Z^n < \R^n$ as a subgroup. Fix $x \in \Z^n$. The claim of Lemma 4 is for $h \in \Z^n$, $2d(x, h.x) \sim d(x, h^2(x))$, where a bound does not depend on $h$. To prove it, join $x, h^2.x$ by a coarse geodesic $\gamma:[0,L] \to \Z^n$. Approximate it by a continuous path $\gamma':[0,L] \to \R^n$ such that at each time, the points on the two paths are boundedly apart such that for each point $g=\gamma(t)$ on $\gamma$, there exists $t'$ such that $\gamma(t')=\gamma'(t')$ (also $|t-t'|$ is bounded). In other words, $\gamma'$ and $\gamma$ visit the same points in $\Z^n$ at the same time (for each point).
Apply Lemma 2 in [@burago] to $\gamma'$ and divide it into at most $n$ segments at points in $\Z^n$ and rearrange, then get a path $\gamma'':[0,L]\to \R^n$ from $x$ to $h^2.x$ such that the distance from $h.x$ to $\gamma''$ is bounded. In the original case it exactly passes $h.x$ since we can divide the path anywhere. In our setting, we approximate the original dividing points by nearby points in $\Z^n$. Since $\Z^n$ is commutative and we cut only at most $n$ times, we get a uniform bound. From $\gamma''$, by approximating it by nearby points in $\Z^n$, we get a coarse geodesic $\gamma''':
[0,L] \to \Z^n$ from $x$ to $h^2(x)$. It follows that $2d(x, h.x)=d(x, h.x) + d(h.x, h^2.x) \sim h(x, h^2(x)).$ Lemma 3 in [@burago] is proved, therefore (1) and (2) are proved as in [@burago]. For (3), notice that by the definition of $|g|_d$, we have $|g^n|_d = |n||g|_d$, therefore, $|d(1,g^n) - n |g|_d| \le C$ for all $n>0$ and $g$.
\[burago2\] Let $d_1, d_2$ be a coarsely geodesic metric on $\Z^n$ that is invariant by the left action of $\Z^n$. Assume that if $d_2(1,g^n)$ or $d_1(1,g^n) \to \infty$ as $n \to \infty$, then $\lim_{n \to \infty} \frac{d_1(1,g^n)}{d_2(1,g^n)}=1$. Then $d_1 \sim d_2$.
Moreover, the result is true if we replace $\Z^n$ by a group which contains $\Z^n$ as a finite index subgroup.
In other words, if $d_1, d_2$ are weakly asymptotically isometric, then they are coarsely equal. By Theorem \[burago\], there is $C$ such that for all $n>0, g$, we have $|d_1(1,g^n) - n |g|_{d_1}| \le C$ and $|d_2(1,g^n) - n |g|_{d_2}| \le C$. Now by our assumption, $|g|_{d_1}=|g|_{d_2}$. Again, by the theorem, for all $g$, $|d_1(1,g) - d_2(1,g)|\le 2C$. Since both $d_1, d_2$ are $\Z^2$ invariant, we have $d_1 \sim d_2$.
For the moreover part, we already know $d_1 \sim d_2$ on $\Z^n$. But any element in $H$ is at bounded distance, say $D$, from a subgroup isomorphic to $\Z^n$ (for both $d_1, d_2$), therefore $d_1\sim_{C+2D} d_2$ on $H$.
Now, here is a lemma that will assure that $d|_{H_i}$ is coarsely geodesic.
\[coarse\] Assume $H$ acts on a coarsely geodesic space $X$ with a point $x\in X$. Define a metric on $H$ by $d(a,b)=d_X(a.x, b.x)$. Assume that there is $C$ such that for any $a,b \in H$, there is a $X$-geodesic between $a.x, b.x$ which is in the $C$-neighborhood of $H.x$. Then $d$ is a coarse geodesic metric on $H$.
This is straightforward from the definition. First assume that $X$ is geodesic. Let $\gamma(t), 0 \le t \le L$ be a geodesic in $X$ from $a.x$ to $b.x$. For each $t$, choose $h_t \in H$ with $d(\gamma(t),h_t.x) \le C$. Define a path $\alpha$ in $H$ from $a$ to $b$ by $\alpha(t)=h_t, 0 \le t \le L$. Now for any $0 \le t \le s \le L$, we have $d(\alpha(t), \alpha(s))=d(h_t,h_s) =d_X(h_t.x, h_s.x)
\sim_{2C}d_X(\gamma(t), \gamma(s)) =|t-s|$. Therefore $\alpha$ is a $2C$-coarsely geodesic. If $X$ is coarsely geodesic, then start with a coarse geodesic $\gamma$ and argue.
We quote several results on asymptotically tree graded spaces and relatively hyperbolic groups. First, a peripheral subgroup is almost convex, therefore Lemma \[coarse\] applies to $d|_H$.
\[almost.convex\][@DS Lemma 4.3] Let $(G,\mathcal H)$ be relatively hyperbolic. Then each $H \in \mathcal H$ is almost convex in $G$, in the sense that any $(K,L)$-quasi-geodesic joining two points of $H$ in $\Gamma(G,S)$ is in the $C$-neighborhood of $H$, where $C$ depends on $K, L$ but not on the quasi-geodesic.
We remark that in [@DS] the lemma is stated only for geodesics, which is sufficient to apply Lemma \[coarse\], but the lemma holds for quasi-geodesics. The reason is that two distinct $aH$ and $bH'$ stay close only in a bounded set (Lemma \[disjoint\]), so that the claim follows using the next lemma.
Two quasi-geodesics in $X$ with common end points stay close to each other in the following sense. We say that the two quasi-geodesics [*fellow travels*]{}. This is a version of Morse lemma for asymptotically tree graded spaces.
\[bcp\][@DS Theorem 1.12] Let $x,y \in X$, and $\alpha$ a $(K,L)$-quasi-geodesic and $\gamma$ a geodesic both between $x,y$ in $(X,d_1)$. Then there exists a constant $C(K,L)$ such that they $C$-fellow travel up to $aH.x$’s, ($H \in \mathcal H$). More precisely, $\alpha$ is in the $C$-neighborhood of $\gamma$ except for the union of some (long) sub-quasi-geodesics of $\alpha$ each of which is contained in the $C$-neighborhood of some $aH.x$ such that $aH.x$ is distance at most $C$ from $\gamma$. In this case, the end points of each of the sub-quasi-geodesics are in the $C$-neighborhood of $\gamma$.
In [@DS] the above lemma is stated for $\Gamma(G,S)$, but $\Gamma(G,S)$ and $(X,d_1)$, as well as $(X,d_2)$, are quasi-isometric therefore the results hold for $(X,d_1), (X,d_2)$ as well. In the above lemma, we may assume the long subpaths of $\alpha$ are disjoint by the following lemma.
\[disjoint\][@DS Lemma 4.7] Fix $x \in X$. For each $D$, the diameter of the intersection of the $D$-neighborhood of $aH_i.x$ and the $D$-neighborhood of $bH_j.x$ in $X$ is uniformly bounded unless $aH_i=bH_j$. The bound depends only on $D$.
Following [@DS], the (almost) [*projection*]{} to $aH$, $\pi_{aH}$, in $\Gamma(G,S)$ is defined as follows for $H \in \mathcal H$: for $g \in G$, $\pi_{aH}(g)$ is the subset of points in $aH$ whose distance from $g$ is less than $d(g,aH)+1$.
The following result also holds for the projection to $aH.x$ in $X$ as well (the proof is same).
\[projection\][@sisto Lemma 1.13 (1) and Theorem 2.14] Let $(G,\mathcal H)$ be relatively hyperbolic. Let $H \in \mathcal H$ and $\pi$ be the projection to $aH$ in the Cayley graph with $a \in G$. Then
1. Any $(K,L)$-quasi geodesic from a point $g$ in $\Gamma(G,S)$ to a point in $aH$ passes the $C$-neighborhood of $\pi(g)$. The constant $C$ depends on $K, L$, but not on $a, H$ and $g$.
2. The diameter of $\pi(g)$ is uniformly bounded. The bound does not depend on $a, H$ and $g$.
We may write $\Gamma(G, S \cup \mathcal H)$ as $G'$. We denote the distance on $\Gamma(G, S \cup \mathcal H)$ by $d_{G'}$. $\Gamma(G, S \cup \mathcal H)$ is hyperbolic (see [@DS Section 8]) and $G$ acts on it. Each edge in $\Gamma(G, S \cup \mathcal H)$ that is not in $\Gamma(G,S)$ joins two points in $aH$ for some $H \in \mathcal H$. Given a geodesic $\gamma$ in $\Gamma(G, S \cup \mathcal H)$, a [*lift*]{} is a path in $\Gamma(G,S)$ obtained by replacing each edge of $\gamma$ that is not in $\Gamma(G,S)$ by a geodesic in $\Gamma(G,S)$ connecting the two end points of the edge.
\[lift.geodesic\][@sisto Prop 1.14] If $\gamma$ is a geodesic in $\Gamma(G, S \cup \mathcal H)$ then its lift is a quasi-geodesic in $\Gamma(G,S)$ with uniform quasi-geodesic constants.
The above lemma does not hold for quasi-geodesics $\gamma$ in general.
\[projection.geodesic\][@DS Prop 8.25] Let $\gamma$ be a quasi-geodesic in $\Gamma(G,S)$ between $x,y$ and $\gamma'$ a quasi-geodesic in $\Gamma(G, S \cup \mathcal H)$ between $x,y$. Then they are in a bounded Hausdorff-distance in $\Gamma(G, S \cup \mathcal H)$. The bound depends only on the quasi-geodesic constants.
We start the proof of Theorem \[main\]. Fix $x \in X$. Define a left invariant (pseudo-)metric $d_1(g,h)=d_1(g.x, h.x)$ on $G$, and also $d_2$ in the same way. Both $(G,d_1), (G,d_2)$ are quasi-isometric to $\Gamma(G,S)$. As before set $\D=d_1-d_2$ on $X$ and $G$. Note that $\D(g,h)=\D(g.x,h.x)$.
We summarize what we know on each peripheral subgroup $H$ by now. On $H$, $d_1$ and $d_2$ are coarsely geodesic metrics by Lemma \[coarse\] and Lemma \[almost.convex\]. By assumption they are weakly asymptotic on $H$. Therefore $d_1 \sim d_2$ on $H$ by Corollary \[burago2\]. The conclusion holds for $aH$ as well.
\[constant\] There exists $L$ such that on each $aH.x$ and $z \in X$, $\D(z,y)$ varies at most $L$ for $y \in aH.x$. $L$ does not depend on $a, H$ and $z$.
Moreover, the statement holds when $y$ is in the $K$-neighborhood of $aH.x$ ($L$ depends on $K$).
Also, the statement holds on $G$, namely, on each $aH$ and $g \in G$, $\D(g,h)$ varies at most $L$ for $h$ that is in the $K$-neighborhood of $aH$.
As usual we assume $d_2$ is geodesic. We omit details for the coarse geodesic case. Fix a point $A \in \pi_{aH.x}(x)$, where $\pi_{aH.x}$ is defined in $(X,d_1)$. We claim that $\D(x,y) \sim \D(x,A)$ for all $y \in aH.x$ such that the constant for $\sim$ does not depend on $y$ or $aH$. Let $\gamma_1, \gamma_2$ be geodesics from $x$ to $y$ for $d_1, d_2$. Notice that $\gamma_2$ is a $d_1$-quasi-geodesic with controlled constants. By Lemma \[projection\], they pass $C$-neighborhood (in both $d_1$ and $d_2$, which are quasi-isometric to each other) of $A$. $C$ does not depend on $a, H, y$. Now take $q_1, q_2$ on each geodesic with $d_i(q_1,A) \le C$ and $d_i(q_2,A) \le C$ with $i=1,2$. Then, by Lemma \[basic.additive\] and Lemma \[triangle\], we have $\D(x,y) \sim_{4C} \D(x,q_1) + \D(q_1, y)
\sim_{8C} \D(x,A) +\D(A,y)$.
But we already know $d_1\sim d_2$ on $aH$, therefore, we have $\D(A,y) \sim 0$. It follows $\D(x,y) \sim \D(x,A)$. The argument is complete when $y \in aH.x$. The moreover part now follows from Lemma \[triangle\]. The argument for the group $G$ is same.
Here is a consequence.
\[estimate\] There exists $P$ such that for any $x,y,z \in G$, we have $$|\D(x,y)-\D(x,z)| \le P d_{G'}(y,z).$$
Let $N=d_{G'}(y,z)$ and $\gamma$ a geodesic from $y$ to $z$ in $\Gamma(G, S \cup \mathcal H)$ with vertices $y=y_0, y_1, \cdots, y_N=z$. $y_n$ and $y_{n+1}$ are joined by an edge. There exists $K$ such that $|\D(x,y_n)-\D(x,y_{n+1})| \le K$ if the edge is in $\Gamma(G,S)$. On the other hand, if the edge is not in $\Gamma(G,S)$, then $y_n, y_{n+1}$ are in some $aH$, therefore $|\D(x,y_n)-\D(x,y_{n+1})| \le L$ by Lemma \[constant\]. Now it follows that $|\D(x,y)-\D(x,z)| \le (L+K)N$. Set $P=L+K$.
\[subadditive\] Let $\gamma$ be a $d_2$-quasi-geodesic in $X$. Then, $\D$ is almost additive on $\gamma$. Namely, let $x,z,y$ be points on $\gamma$ in this order, then $\D(x,z)+\D(z,y) \sim_B \D(x,y)$, where $B$ depends on the quasi-geodesic constants of $\gamma$.
The statement holds for quasi-geodesics on $\Gamma(G,S)$ as well.
By assumption $\gamma$ is a $d_2$-quasi-geodesic with controlled constants. Then it is a $d_1$-quasi-geodesic with controlled constants as well. Let $\gamma'$ be a $d_1$-geodesic in $X$ from $x$ to $y$. By Lemma \[bcp\], there exists $C$ that depends on the quasi-geodesic constants such that $\gamma$ and $\gamma'$ stay close to each other except for subsegments in $\gamma$ each of which stays in the $C$-neighborhood of one $aH.x$ for a long time, but the end points of those segments are $C$-close to $\gamma'$. If $z$ is outside of those segments, then $z$ is close to $\gamma'$, which implies the almost additivity at $z$ by Lemma \[basic.additive\].
Now assume $z$ is contained in one of the subsegments, say, $[z_1,z_2]$. Each $z_i$ is $C$-close to $\gamma'$, therefore by Lemma \[basic.additive\], we have $\D(x,y) \sim \D(x,z_1) + \D(z_1,y)$. On the other hand, since $z,z_1,z_2$ is in the $C$-neighborhood of $aH.x$, by Lemma \[constant\] and Lemma \[triangle\], we have $ \D(x,z_1) + \D(z_1,y) \sim \D(x,z) + \D(z,y)$ (Lemma \[constant\] applies to $\D(*,y)$). Combining them, $\D$ is almost additive at $z$. The argument is complete. Since $\Gamma(G,S)$ and $X$ are quasi-isometric, we also have the almost additivity in $\Gamma(G,S)$.
We go back to the proof of the theorem. We want to show that $\D(x,y)$ is bounded on $X$, which is equivalent to that $\D(g,h)$ is bounded on $G$. To argue by contradiction, assume not. We will find $f \in G$ such that $\D(x, f^n.x)=\D(1,f^n)$ grows roughly linearly on $n$, which will be a contradiction since $d_1$ and $d_2$ are weakly asymptotic.
Take $g$ such that $\D(1, g)$ is very large. Let $\gamma$ be a geodesic from $1$ to $g$ in $\Gamma(G,S)$, and let $h \in G$ be such that $h$ is on $\gamma$ and $\D(1,h)$ is approximately $\D(1,g)/2$. This is possible since $\D(1,y)$ is almost continuous when we vary $y$ on $\gamma$.
Let $\alpha$ be a geodesic from $1$ to $g$ in $\Gamma(G, S \cup \mathcal H)$. By Lemma \[projection.geodesic\], the Hausdorff distance between $\gamma$ and $\alpha$ in $\Gamma(G, S \cup \mathcal H)$ is bounded. In particular, $h$ is at bounded distance from $\alpha$ in $\Gamma(G, S \cup \mathcal H)$. Notice that $d_{G'}(1,h), d_{G'}(h, g), d_{G'}(1, g)$ are all large since $\D(1,h), \D(h,g), \D(1,g)$ are all large (use Lemma \[estimate\]). Since $\Gamma(G, S \cup \mathcal H)$ is hyperbolic, as before, one of the paths $h^*, k^*, (hk)^*$ (we use the same notation as in the proof of Theorem \[krat\] with $x=1$) is a quasi-geodesic in $\Gamma(G, S \cup \mathcal H)$ with uniform quasi-geodesic constants. Denote it by $\gamma=f^*$.
First, we assume that $\gamma$ is a geodesic, and argue. Take a lift, $\beta$, of $\gamma$ in $\Gamma(G,S)$. Then $\beta$ is a quasi-geodesic in $\Gamma(G,S)$ with controlled constants by Lemma \[lift.geodesic\]. Since $\D(1,f)$ is very large (approximately $\D(1, g)$ or $\D(1, g)/2$), by Lemma \[subadditive\], $\D(1, f^n)$ grows roughly linearly on $n$ (apply the lemma at each point $f^n$).
In general, $\gamma$ is only a quasi-geodesic in $\Gamma(G, S \cup \mathcal H)$ with the quasi-geodesic constants controlled. In this case, for each $N>0$, take a geodesic $\gamma'$ from $1$ to $f^N$ in $\Gamma(G, S \cup \mathcal H)$, so that $1,f, \cdots, f^{N}$ are in a bounded neighborhood of $\gamma'$ in $\Gamma(G, S \cup \mathcal H)$. This is because $\Gamma(G, S \cup \mathcal H)$ is hyperbolic. For each $f^n$, let $y_n \in \gamma'$ be a closest point on $\gamma'$ for $d_{G'}$. Take $y_0=1, y_N=f^N$. By Lemma \[estimate\], for each $n$, $\D(y_n, y_{n+1}) \sim \D(f^n, f^{n+1})
= \D(1, f)$. Take a lift of $\gamma'$, denoted by $\beta$, which is a quasi-geodesic in $\Gamma(G,S)$ with controlled constants. The points $y_n$ are on $\beta$, and by Lemma \[subadditive\], $\D(y_0, y_n)$ grows roughly linearly on $n$ (roughly the slope is $\D(1, f)$, which is much larger than any constants for $\sim$ in the above argument). Again by Lemma \[estimate\], $\Delta(1,f^n)$ grows roughly linearly for $0 \le n \le N$, with the slope roughly $\D(1, f)$. Since $N$ was arbitrary, $\D(1,f^n)$ grows roughly linearly on $0 \le n$. This finish the argument.
More general case {#complex}
-----------------
In the proof of Theorem \[main\], what we need from the peripheral subgroups $H$ is the property that any two weakly asymptotic, coarsely geodesic metrics on $H$ are coarsely equal (see the discussion in the beginning of the proof). We verified this property in Corollary \[burago2\] for virtually abelian groups. We restate Theorem \[main\] in this more general form. The proof is identical and we omit it.
\[main2\] Let $(G, \mathcal H)$ be a relatively hyperbolic group. Assume that each $H_i \in \mathcal H$ satisfies the property such that any two weakly asymptotic, coarsely geodesic metrics on $H_i$ are coarsely equal. Suppose $G$ acts on $X$ properly and co-boundedly by isometries for geodesic metrics $d_1, d_2$ (or more generally, $d_2$ is a coarsely geodesic metric). If $d_1$ and $d_2$ are weakly asymptotically isometric on $X$, then they are coarsely equal.
One potential application would be to non-uniform lattices in the Lie group $SU(n,1)$. It is known that such lattice is relatively hyperbolic with peripheral subgroups virtually nilpotent. As we mentioned in the introduction, the assumption in the theorem does not hold in general for all nilpotent groups.
Marked length spectrum {#marked}
======================
In this section we discuss an application to the marked length spectrum problem. We state a variant of Theorem \[main\].
\[rel.hyp\] Let $(X_1, d_1), (X_2,d_2)$ be geodesic metric spaces on which $G$ acts by isometries, co-boundedly and properly with respect to both $d_1$ and $d_2$. Suppose the action on $X_1$ is free. We allow $X_1$ to be coarsely geodesic. Assume $|g|_1=|g|_2$ for any hyperbolic element $g \in G$. If $G$ is toral relatively hyperbolic group, then there exists a $G$-equivariant, $(1,C)$-quasi-isometry map $f:X_1\to X_2$.
We follow the argument for Theorem \[main\]. We fix points $x_1 \in X_1$ and $x_2 \in X_2$, and define $\D(g,h)=d_1(g.x_1, h.x_1)-d_2(g.x_2, h.x_2)$ for $g,h \in G$. We claim that $\D$ is bounded. We will prove this later, but once this is known, then there exists a desired map $f$. In other words, $\D(p,q)=d_1(p,q)-d_2(f(p),f(q))$ is bounded. Indeed, set $f(g.x_1)=g.x_2$ for each $g \in G$ (use the $G$-action on $X_1$ is free). Then, $f$ is $G$-equivariant and $\D(p,q)$ is bounded for $p,q \in G.x_1$. Moreover, we can extend $f$ to $X_1$, $G$-equivariantly, such that $\D$ is bounded. Indeed, choose a point $p$ on each $G$-orbit in $X_1$. There is $g \in G$ such that $d_1(p,g.x_1)$ is bounded. Define $f(p)=g.x_2$, and extend $f$, $G$-equivariantly, to the $G$-orbit of $p$. It is clear that $\D$ is bounded on $X_1$.
We are left to show $\D$ is bounded on $G.x_1$. Again, we repeat the argument for Theorem \[main\]. When needed, we replace $X$ with $X_2$ and map any objects in $X_1$ to $X_2$ by $f$ and argue on $X_2$. Notice that the map $f$ defined above is a quasi-isometry from $X_1$ to $X_2$. For example, in the proof of Lemma \[constant\], $\gamma_1$ is a geodesic from $x_1$ to $y_1=ah.x_1 \in aH.x_1$, and $\gamma_2$ is a geodesic from $x_2$ to $y_2 =ah.x_2\in aH.x_2$. Then $f(\gamma_1)$ is a quasi-geodesic from $x_2$ to $y_2$. Apply the argument to $\gamma_2$ and $f(\gamma_1)$. Lemma \[subadditive\] is similar. We omit details.
We apply the result to the marked length spectrum problem for manifolds.
\[manifold\] Let $(M_1,d_1), (M_2,d_2)$ be closed Riemannian manifolds with the isomorphic fundamental group $G$ that is toral relatively hyperbolic. Assume they have the same marked length spectrum. Then there is a $G$-equivariant $(1,C)$- quasi-isometry $f:\tilde M_1 \to \tilde M_2$.
Moreover, if there is a homeomorphism $H: M_1 \to M_2$ that induces the isomorphism on $G$ and lifts to a $G$-equivariant homeomorphism $h:\tilde M_1 \to \tilde M_2$, then $h$ is a $(1,C')$- quasi-isometry.
Let $X_i$ be the universal cover of $M_i$, respectively. Each of them have the lift of $d_i$, which we also write by $d_i$. Each action by $G$ on $X_i$ is free. By assumption, we can apply Theorem \[rel.hyp\] and we get a desired map $f$.
For the moreover part, it suffices to show that there is a constant $L$ such that for any point $p \in \tilde M_1$, $d_2(f(p),h(p)) \le L$. To see this, fix $x_1 \in X_1$ and set $x_2=h(x_1)$. Let $\pi:X_1 \to M_1$ be the covering map. Given $p \in X_1$, join $\pi(x_1)$ to $\pi(p)$ by a shortest geodesic $\gamma$ in $M_1$. Lift it to a geodesic $\tilde \gamma$ from $p$ to $g.x_1$ with $g \in G$. Then $h(\tilde \gamma)$ is a path from $h(p)$ to $h(g.x_1)=g.(h(x_1))=g.x_2$. By the argument for Theorem \[rel.hyp\], $f(p)=g.x_2$ (to be precise, we can extend $f$ in this way). That means that $h(p)$ and $f(p)$ are joined by $h(\tilde \gamma)$, but the length of this path is bounded since the length of $\tilde \gamma$ is bounded and $h$ is continuous and $G$-equivariant.
For example, if $M_1$ has non-positive curvature, dimension is not 3 nor 4, and $M_2$ is aspherical (for example $M_2$ has non-positive curvature), then each isomorphism of $G$ is induced by a homeomorphism $H$ by Farrell-Jones (see [@farrell]), therefore its lift $h$ is a $(1,C)$-quasi-isometry map.
The case where $G$ is word-hyperbolic in Corollary \[manifold\] is proved by Furman [@furman Theorem 2]. His argument is different from ours, and uses Patterson-Sullivan measures for hyperbolic groups with respect to word metrics constructed by Coornaert [@co], and does not seem to apply to prove Theorem \[rel.hyp\].
If $C=0$ for some $f$ in Theorem \[manifold\], $M_1$ and $M_2$ are isometric. This is the conclusion that [*the marked length spectrum problem/conjecture*]{} concerns (see the conjecture in [@katok 3.1] for negative curvature case). Several cases are known to have the positive answer, for example for surfaces of negative curvature [@otal]. Interestingly, Bonahon [@bonahon] gave examples of geodesic metrics $d_1, d_2$ on a hyperbolic surface such that $d_1$ is Riemannian with constant negative curvature and that they have the same marked length spectrum, but are not isometric to each other. Theorem \[rel.hyp\] applies to his example. For higher dimension there is a result by Hamenstädt ($M_1$ is a rank-1 locally symmetric space and $M_2$ is negatively curved) using a theorem of Besson-Courtois-Gallot [@bcg] on the volume entropy. In all of those cases, the fundamental group is hyperbolic. Our result put the marked length spectrum problem (for manifolds) into context for a broader class of groups. Although this is not for manifolds, another case where the isospectral length problem is solved is $\R$-trees under the assumption that the action is minimal and semi-simple, [@culler].
[krat]{}
H. Abels and G. Margulis. Coarsely geodesic metrics on reductive groups. In Modern dynamical systems and applications, 163–-183. Cambridge Univ. Press, Cambridge, 2004.
G. Besson, G. Courtois, S. Gallot, Entropies et rigidités des espaces localement symétriques de courbure strictement négative. Geom. Funct. Anal. 5 (1995), no. 5, 731-–799.
Francis Bonahon, Surfaces with the same marked length spectrum. Topology Appl. 50 (1993), no. 1, 55–-62.
E. Breuillard. Geometry of groups of polynomial growth and shape of large balls. arXiv:0704.0095, 2007. Preprint.
E. Breuillard, E. Le Donne. On the rate of convergence to the asymptotic cone for nilpotent groups and subFinsler geometry. Proc Natl Acad Sci U S A. 2012 Dec 17.
Martin R. Bridson, André Haefliger, Metric spaces of non-positive curvature. Grundlehren der Mathematischen Wissenschaften 319. Springer-Verlag, Berlin, 1999.
D. Yu. Burago, Periodic metrics. Representation theory and dynamical systems, 205–-210, Adv. Soviet Math., 9, Amer. Math. Soc., Providence, RI, 1992.
K. Burns, A. Katok, Manifolds with nonpositive curvature. Ergodic Theory Dynam. Systems 5 (1985), no. 2, 307–-317.
Michel Coornaert, Mesures de Patterson-Sullivan sur le bord d’un espace hyperbolique au sens de Gromov. Pacific J. Math. 159 (1993), no. 2, 241–-270.
Marc Culler, John W. Morgan, Group actions on $\R$-trees. Proc. London Math. Soc. (3) 55 (1987), no. 3, 571–-604.
Cornelia Drutu, Mark Sapir, Tree-graded spaces and asymptotic cones of groups. With an appendix by Denis Osin and Sapir. Topology 44 (2005), no. 5, 959–-1058.
B. Farb. Relatively hyperbolic groups. Geom. Funct. Analysis 8 (1998), 810–840.
F. T. Farrell, L. E. Jones, P. Ontaneda, Negative curvature and exotic topology. Surveys in differential geometry. Vol. XI, 329–-347, Surv. Differ. Geom., 11, Int. Press, Somerville, MA, 2007.
Alex Furman, Coarse-geometric perspective on negatively curved manifolds and groups. Rigidity in dynamics and geometry (Cambridge, 2000), 149–-166, Springer, Berlin, 2002.
M. Gromov, Hyperbolic groups. Essays in group theory, 75–-263, Math. Sci. Res. Inst. Publ., 8, Springer, New York, 1987.
M. Gromov, Asymptotic invariants of infinite groups. Geometric group theory, Vol. 2 (Sussex, 1991), 1–295, London Math. Soc. Lecture Note Ser., 182, Cambridge Univ. Press, Cambridge, 1993.
G. Christopher Hruska, Bruce Kleiner, Hadamard spaces with isolated flats. With an appendix by the authors and Mohamad Hindawi. Geom. Topol. 9 (2005), 1501-–1538.
S. A. Krat, On pairs of metrics invariant under a cocompact action of a group. Electron. Res. Announc. Amer. Math. Soc. 7 (2001), 79–-86.
Gregory Margulis, Metrics on Reductive Groups. Mathematisches Forschungsinstitut Oberwolfach Report No. 33/2006, Geometric Group Theory, Hyperbolic Dynamics and Symplectic Geometry, 2004 – 2006.
G. A. Noskov, Coarsely geodesic metrics on reductive groups (after H. Abels and G. A. Margulis). Mathematical structures and modeling. No. 15, 5–-17, Omsk. Gos. Univ., Omsk, 2005.
Jean-Pierre Otal, Le spectre marqué des longueurs des surfaces à courbure négative. Ann. of Math. (2) 131 (1990), no. 1, 151–162.
Alessandro Sisto. Projections and relative hyperbolicity. Enseign. Math. 59 (2013), 165-181.
[^1]: The author is supported in part by Grant-in-Aid for Scientific Research (No. 23244005)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'There can be performance and vulnerability concerns with block ciphers, thus stream ciphers can used as an alternative. Although many symmetric key stream ciphers are fairly resistant to side-channel attacks, cryptographic artefacts may exist in memory. This paper identifies a significant vulnerability within OpenSSH and OpenSSL and which involves the discovery of cryptographic artefacts used within the ChaCha20 cipher. This can allow for the cracking of tunneled data using a single targeted memory extraction. With this, law enforcement agencies and/or malicious agents could use the vulnerability to take copies of the encryption keys used for each tunnelled connection. The user of a virtual machine would not be alerted to the capturing of the encryption key, as the method runs from an extraction of the running memory. Methods of mitigation include making cryptographic artefacts difficult to discover and limiting memory access.'
author:
-
bibliography:
- 'main.bib'
title: Deriving ChaCha20 Key Streams From Targeted Memory Analysis
---
network traffic; decryption; memory analysis; Virtual machine introspection; Secure Shell; Transport Layer Security; stream ciphers; ChaCha20
Introduction
============
There is an increasing challenge between the rights of citizens to privacy and the rights of society to protect itself from adversaries [@bauman2014after][@Iphofen]. The breaking of encryption tunnels is thus one of the major debating points, and where law enforcement agencies often aim to gather tools and methods which break these tunnels, or where we fix vulnerabilities in tools in order to avoid these tunnels from being broken. In most cases we now perform a key negotiation phase – typically with ECDH (Elliptic Curve Diffie-Hellman) – and then use a symmetric key method to encrypt the traffic within the tunnel.
The cracking of the key exchange process and of the symmetric key used in the tunnel are, in most cases, too costly to crack. Unfortunately, the key exchange process can leave behind trails of evidence in memory which can provide significant clues to the symmetric key being used. While this has been demonstrated for block ciphers, such as for the Advanced Encryption Standard (AES) [@halderman2009lest][@maartmann2009persistence], this paper outlines how well-used applications such as OpenSSH and OpenSSL allow for every generated key in the ChaCha20 stream cipher to be revealed within a fairly fast discovery time. As virtualized environments enable access to virtual machine resources from more privileged levels such as hypervisors or hypervisor consoles, applications operating at that level can extract live virtual machine memory. Extraction is most effective when a virtual machine is paused but it is not necessary. So, virtualized environments present an opportunity to find keys without impacting the target and for target applications to be unaware of extraction.
The rest of the paper is structured as follows. Section II discusses related research including side-channel studies and background on stream ciphers and ChaCha20 cipher implementations is presented in Section III. Section IV provides relevant details of the framework and its implementation is given in Section V. The results are presented and discussed in Section VI and conclusions drawn in Section VII.
Related Work
============
This paper focuses on the decrypting network traffic encrypted with ChaCha20-Poly1305 cipher. Prior studies have investigated potential vulnerabilities in cipher design and in cipher implementation. Researchers have found no vulnerabilities in ChaCha20 design. For example, differential attacks using techniques such as identifying significant key bits only succeeded with reduced cipher rounds and significant volumes of plaintext-ciphertext pairs [@aumasson2008new][@maitra2016chosen]. Combined linear and differential analysis improves performance, but is similarly restricted [@choudhuri2016differential].
ChaCha20 implementations may be vulnerable to side-channel attacks. While the cipher design may prevent timing attacks [@bernstein2008salsa20], correlating power electromagnetic radiation when specific cryptographic activities are performed may leak key stream information [@jungk2017don][@adomnicai2017bricklayer]. Engendering instruction skips, for example by using a laser or electromagnetic pulse, could potentially produce the key stream but timing the activity would be challenging [@kddi]. Furthermore, these approaches may be impractical in real-world scenarios.
Cryptographic artefacts have been found in device memory. For instance, RSA keys may be discovered in virtual machine images [@Rocha2011] [@Saxon2015]. Studies have also discovered DES and AES cipher keys in cold-boot attacks [@halderman2009lest], Skipjack and Twofish key blocks in virtual memory [@maartmann2009persistence], and AES session keys in virtual memory [@taubmann2016tlskex]. Although these approaches use entropy measures to determine possible keys, they do not decrypt ciphertext encrypted with ciphers such as AES in Counter mode and ChaCha20 which require nonces/initialization vectors. This study builds on the TLSkex [@taubmann2016tlskex] and MemDecrypt studies [@mclarenssh2019] which used privileged monitors to extract identified virtual machine process memory to identify TLS 1.2 AES keys, and SSH AES keys and initialization vectors, respectively. Instead, this study uses a different algorithm to find ChaCha20 cipher keys and nonces in device memory enabling SSH and TLS sessions to be decrypted in a non-invasive manner. The approach may enable decryption of Adiantum encrypts [@crowley2018adiantum], the Google disk-encryption algorithm based on XChaCha20, an extension to ChaCha20 and Salsa20 with a longer nonce [@bernstein2011extending].
Stream ciphers
==============
Secure protocols use encryption to provide confidentiality for secure communications between parties. While asymmetric encryption ciphers are used in secure protocol set-up stages, for performance symmetric ciphers encrypt the confidential information. Symmetric ciphers are commonly classified as being stream ciphers, where plaintext is encrypted bit-by-bit or byte-by-byte or block ciphers, where blocks of a specific size are encrypted. This paper focuses on stream ciphers.
Stream ciphers generate a random key stream from an evolving state [@biryukov2004block]. The key stream is then typically XOR-ed with the plaintext to generate ciphertext [@biryukov2004block]. Software implementations for stream cipher have been found to be faster than block ciphers although possibly more difficult to implement [@klein2013stream]. Stream ciphers have typically been used in embedded technologies such as Internet of Things (IoT) devices and smartphones [@manifavas2016survey]. Stream ciphers, as well as block ciphers, have been supported by secure protocols. However, with vulnerabilities leading to the planned deprecation of the RC4 stream cipher for protocols such as Secure Shell (SSH) [@Camara] and Transport Layer Security (TLS) [@popova], alternative stream ciphers have been under consideration. In particular, the ChaCha20 stream cipher with the Poly1305 authenticator [@nir] has been adopted in secure protocol implementations such as OpenSSH [@OpenSSH], and OpenSSL [@OpenSSL], as well for Google Chrome on smartphones [@Ianix]. The discovery of the key stream or inputs to key stream generation may be an unacceptable vulnerability for stream ciphers including ChaCha20-Poly1305.
Stream cipher cryptographic artefacts are memory-resident at points in time. Furthermore, these artefacts, as well as the key stream are on a program stack, in the heap or in shared memory. If timely acquisition of target device memory is obtained, the stream or artefacts may be discovered. With the growth of forensic technologies that enable memory access to targets such as desktops, servers, smartphones, and IoT devices, an opportunity exists to discover these artefacts. Although researchers have investigated ChaCha20 side-channel attacks with intermediate state leakage [@jungk2017don] and without [@adomnicai2017bricklayer], no studies analyze memory so this paper presents a new approach to decrypting secure communications encrypted with the ChaCha20 cipher. A single memory extract suffices to find the cryptographic artefacts for decryption. Furthermore, the approach is faster and less invasive than methods that use side-channel techniques.
OpenSSH and PuTTY implement [email protected] while OpenSSL [@Ianix] versions above 1.1 implement variations where the asymmetric cipher used in key exchange vary, e.g. ECDHE-RSA-CHACHA20-POLY1305 [@langley2016chacha20]. Both implementations adhere to the RFC *ChaCha20 and Poly1305 for IETF Protocols* [@nir] which is based on Bernstein’s ChaCha20 cipher proposal [@bernstein2008chacha], a variation of the earlier Salsa20 cipher [@bernstein2008salsa20].
ChaCha20 ciphering process
--------------------------
The ChaCha20 cipher generates key streams of 64-bytes. Inputs to key stream generation are independent of plaintext or ciphertext similar to other eSTREAM proposals [@Hutchison] but unlike stream ciphers such as Helix [@bernstein2008salsa20]. This enables parallel ciphertext generation with consequent performance improvement. 20 rounds of mathematical calculations using XOR, addition and rotation using as inputs four 4-byte constants, a random 32-byte key, a 4-byte counter, and a 12-byte nonce (Bernstein originally specified the nonce and counter lengths to be eight). However, this is not a material difference in the investigations). The 4-byte constants are 0x61707865, 0x3320646e, 0x79622d32, and 0x6b206574 or in ASCII ‘apxe’, ‘3 dn’, ‘yb-2’, and ‘k et’. In ChaCha20 these strings are concatenated. The counter, which typically starts at 0 or 1 increments for each 64-byte plaintext block [@nir].
In the [email protected] implementation, memory is allocated to hold a structure comprising key stream input fields, the 64-byte key stream itself, and an index pointer. The packet lengths are encrypted separately from the remainder of the payload so four structures are required: two for encrypting outgoing lengths and payloads, and the same again for incoming encrypted data. The memory contents of the concatenated constant string are *expand 32-byte k*. The nonce is a sequence number and counters for the encrypted packet lengths and payloads are zero and unity, respectively.
The OpenSSL implementation of ChaCha20-Poly1305 differs in a number of respects as required by IETF RFC ChaCha20-Poly1305 Cipher Suites for Transport Layer Security (TLS) [@langley2016chacha20]. As encrypting packet length is not required in TLS two memory structures are used for encrypting outgoing messages and decrypting incoming messages. Memory used for ChaCha20 inputs is temporary as the encryption structure is assembled from other sources. Also, the nonce is an XOR of the sequence number and the vector generated during the initial handshake when the cipher keys are obtained.
ChaCha20 takes a 256-bit key and a 32-bit nonce (and which includes a counter). This creates a key stream which is then XOR-ed with the plaintext stream as illustrated in Figure \[fig:ChaCha\]. In software-only implementations, it is often more than three times faster than AES [@langley2016chacha20], and is well suited to lower-powered devices and in real-time communications. ChaCha20 operates on 32-bit words at a time with a key of 256 bits (K=($k_0$, $k_1$, $k_2$, $k_3$, $k_4$, $k_5$, $k_6$, $k_7$). This outputs blocks of 512 bits for the key stream (Z), and which is XOR-ed with the plaintext stream. The state of the encryption is stored within 16x32-bit word values and arranged as a 4x4 matrix: $$\begin{bmatrix}
x_0 & x_1 & x_2 & x_3\\
x_4 & x_5 & x_6 & x_7\\
x_8 & x_9 & x_{10} & x_{11}\\
x_{12} & x_{13} & x_{14} & x_{15}\\
\end{bmatrix}$$
The initial state contains 16x32-bit values with constant values (0x61707865, 0x3320646e, 0x79622d32, 0x6b206574) the key ($k_0$, $k_1$, $k_2$, $k_3$, $k_4$, $k_5$, $k_6$, $k_7$), the counter ($c_0$) and the nonce ($n_0$,$n_1$,$n_2$,$n_3$): $$\begin{bmatrix}
\scriptscriptstyle 0x61707865 & \scriptscriptstyle 0x3320646e & \scriptscriptstyle 0x79622d32b & \scriptscriptstyle 0x6b206574\\
k_0 & k_1 & k_2 & k_3\\
k_4 & k_5 & k_{6} & k_{7}\\
c_{0} & n_{0} & n_{1} & n_{2}\\
\end{bmatrix}$$
The counter thus has 32-bits (1 x 32 bits), and the nonce has 96-bits (3 x 32 bits). ChaCha20 then defines a quarter round as shown in Algorithm \[alg:ChaChaQuarterRound\].
$a = a + b$ $d = d \oplus a$ $d = (d)<<16$ $c = c + d$ $b = b \oplus c$ $b = (b)<<12$ $a = a + b$ $d = d \oplus a$ $d = (d)<<8$ $c = c + d$ $b = b \oplus c$ $b = (b)<<7$
There are then 20 rounds (10 for column rounds and 10 for diagonal rounds) as shown by Algorithm \[alg:ChaChaKeyStream\].
$y \longleftarrow X$ $ Z \longleftarrow X + y\; $
Decryption Framework
====================
The decryption framework is comprised of data capture, analysis, and decrypt components. Each component is modifiable or replaceable so that different devices, target operating systems, ciphers, and protocols can be addressed. Details of the component design for ChaCha20 are presented in the following paragraphs.
**Data Collection**. Target device network packets and volatile memory are extracted. Complete SSH and TLS sessions originating from the target device are captured for later examination. The indicator for extraction of target volatile is the client transmission of a protocol message indicating that the initialization phase is complete. These messages are *New Keys* and *Client Finished* for SSH and TLS Version 1.2 respectively. Cryptographic artefacts have then been generated and are likely to be memory-resident and memory can be extracted for any outgoing message in the network session.
**Memory Analysis**. Candidate cryptographic artefacts are discovered in memory extracts. Initially, the component searches for the constant string *expand 32-byte k* to discover candidate ChaCha20 data structures. Although unlikely for the string to be present in non-base structures, a second step assesses whether the 32-byte block after the constant string in a candidate base structure’s might be a key. Encryption keys must be unpredictable, i.e. random. Key randomness can be evaluated using the Shannon entropy definition [@shannon1948mathematical]:\
$$H=-\sum_{i=1}^{n} p(i) \log_2 p(i)$$
where $p_i$ is the normalized frequency of the $i$th byte in the message i.e. $p(i)=f(i)/n$. So, where the entropy of the 32-byte block exceeds a threshold, a candidate base structure is identified and the key, nonce and counter are retained.
**Decrypt Analysis**. Cryptographic artefacts output by the memory analysis component are input parameters in decrypt analysis. In each instance, the candidate key and nonce group is decrypted and verified according to the protocol used. For SSH, the groups are used to decrypt incoming and outgoing packet lengths and payloads. The packet length groups are identified if the decrypted packet length meets Equation (2) for short packets, typically in the authentication and channel set-up phases. For larger packets, the decrypted packet length supports SSH packet reassembly. Decrypts with the remaining cryptographic groups are analyzed to establish compliance with SSH protocol specifications. For TLS, decrypts are analyzed to establish compliance with TLS. protocol specifications.
$$\begin{split}
packet\ data\ length\ = \\
& decrypted\ packet\ length\ + \\
& size(packet\ length field)\ + \\
& size(MAC field)
\end{split}$$
Implementation
==============
The framework is implemented in a virtualized environment as illustrated in Figure \[fig:Framework\] . Implementations on other technologies which facilitate packet capture and target memory access should be possible. The Xen hypervisor [@Xen] offers benefits over alternatives including the presence of LibVMI [@LibVMI] and PyVMI [@Payne] libraries providing access to the volatile memory of live virtual machines. Because of its small trusted computing base, Xen is managed by a privileged virtual machine, which runs or initiates the framework components. The privileged virtual machine also provides the unprivileged guests with network and disk access.
The data collection component inspects virtual machine network traffic. Each packet is redirected to a local queue using an iptables rule and NetFilterQueue 0.8.1 [@Kerkhoff], and protocol fields to determine whether SSH or TLS sessions have been initiated. Each SSH and TLS packet is written to file. Memory extraction uses PyVMI, LibVMI, and Volatility libraries [@volatilityfoundation].
Memory analysis obtains candidate ChaCha20 cryptographic artefacts. Each memory extract file is searched for the constant string. Although false positives are unlikely, additionally the entropy of the following 32-byte block is compared with a threshold, experimentally found to be 4.5. If the threshold is exceeded, a base structure has been identified and the key, nonce, and counter fields are retained for decryption as groups.
The decrypt analysis component iterates through the key and nonce groupings. For SSH, the first four encrypted bytes in packets are decrypted using the Chacha20poly1305 package with counter value 0 [@Chacha20poly1305]. For a valid group the first four bytes represent the packet length and Equation (2) holds. For ChaCha20, the decrypts with other groups are validated. If the decrypted padding length obeys Equation (3) as specified in SSH Transport Layer Protocol [@ylonen2005secure] the following blocks are decrypted to evaluate compliance with the SSH authentication and channel set-up specifications [@ylonen2005auth][@ylonen2005connectoin]. For TLS, the entire data block is decrypted with cryptographic artefact groups. Although TLS supports other higher-level protocols such as SMTP, HTTP-over-TLS is perhaps the most commonly use and so compliance of the decrypt with the HTTP 1.1 specification [@fielding1999hypertext] is assessed. Valid decrypts are retained for user inspection.\
$$4 \leq padding\ length \leq 255$$
The physical environment for experimentation is a Core 2 Duo Dell personal computer with 40GB of disk storage and 3GB of RAM. This hosts the hypervisor, the privileged virtual machine, an unprivileged Windows virtual machine, and an unprivileged Ubuntu virtual machine. The hypervisor is Xen Project 4.4.1 and the hypervisor console is Debian release 3.16.0-4-amd64 Version 1. Tests run on untrusted Windows client and Linux server virtual machines. The Windows clients runs the Windows 10 (10.0.16299) operating system with 2GB of memory and 40 GB of disk. A Linux virtual machine runs an Ubuntu 14.04 build (“Trusty”) with 512MB of allocated memory and 4GB of disk storage. The PuTTY suite [@PuTTY] is used for SSH client application testing with SSH server functionality provided by openssh-server. OpenSSL 1.1.0h provides TLS client and server functionality.
Evaluation
==========
For SSH evaluation the PuTTY ‘pscp’ program is executed from the Windows command line using requests of the form:
*pscp -P nnnn ’filename’ @ipaddress:/home/name*
where *nnnn* is the target port, *‘filename’* is the file being transmitted, *name* is a user account on the target Ubuntu server, *ipaddress* is the target server IP address and */home/name* is the Ubuntu server target folder for the transmitted file. An Ubuntu service is started from the bash command line to listen to client SSH messages with requests of the form:
*/usr/sbin/sshd -f /root/sshd\_config -d -p nnnn*
where *nnnn* is the port number the service listens on and *sshd\_config* contains configuration details. Here, *sshd\_config* contains the string ‘ciphers [email protected]’. Client and server tests execute OpenSSL from the command line. The OpenSSL server emulates a web server with:
*openssl s\_server -accept 443 -cert crt.pem -key key.pem -WWW*
where *crt.pem* and *key.pem* are certificate and key files. The client connects to the OpenSSL server with:
*openssl s\_client \[-cipher CIPHER\] -connect a.b.c.d:443*
where *CIPHER* identifies the encryption cipher, key exchange and authentication algorithms (here, ECDHE-RSA-CHACHA20-POLY1305) and *a.b.c.d* the OpenSSL server IP address. Client input simulates browser requests, e.g. ‘GET / HTTP/1.1’, ‘Host: a.b.c.d.’, Accept-Encoding: gzip, deflate’, Accept: \*/\*.
ChaCha20 base structures were found in 100% of instances. For the SSH [email protected] protocol four base stuctures were discovered in Windows and Linux application memory. For OpenSSL, one base structure was discovered for ECDHE-RSA-CHACHA20-POLY1305 as shown in the highlighted section of Figures \[fig:SSHbase\] and \[fig:TLSbase\] respectively. These discoveries leads to key stream generation and the rapid decryption of complete SSH sessions including server user credentials, file names and uploaded file contents between 150 bytes and 1 MB as well as the decryption of outgoing client TLS traffic. Differences between SSH and TLS relates to their implementations. PuTTY/OpenSSH structures are heap-resident so memory extracts for successful decryption may not be linked to client SSH message transmission. By contrast, OpenSSL structures are stack-resident and may therefore be overwritten. This limitation prevents Linux server successful searches for the constant string although high-entropy regions identify candidate encryption keys. Consequently, full SSH sessions are decryptable whereas only outgoing TLS sessions are currently decryptable when applying the algorithms. A SSH decrypt is illustrated in Figure \[fig:SSHdecrypt\] and example output from the analysis stages is presented in Figure \[fig:MemDecrypt\].
Memory and decrypt analysis components decrypt rapidly. The memory analysis durations are less than 0.5 (SSH) and 0.1 (TLS) seconds as shown in Table \[tab:my\_label\]. As single extract analysis requires a maximum 0.018 seconds, parallel processing memory extract analysis offers significant performance opportunities. Although memory analysis is independent of file size, decrypt analysis durations are proportionate to the volume of encrypted traffic.
SSH TLS
-------------------- ------- -------
Maximum 0.407 0.027
Minimum 0.007 0.011
Mean 0.144 0.021
Standard Deviation 0.153 0.006
: Memory Analysis Durations
\[tab:my\_label\]
Countermeasures
===============
Fortunately, countermeasures to discovering the ChaCha20 basic structures exist, and hiding the constant string makes discovery more challenging. Possible measures are copying the constant string segments to registers and assembling the structure in the encryption routine, encrypting the constant string, or randomly segmenting the constant string. Perhaps, the most effective approach is assembling the base structure on the stack, as for OpenSSL, and clearing stack contents immediately after the encryption process. The cryptographic artefacts can still be discovered by searching for high-entropy measures but the process is comparatively slower with durations exceeding 4 minutes.
Conclusions
===========
Implementations of ChaCha20-Poly1305 encryption using commonly used applications and libraries for SSH and TLS communications are vulnerable to decrypt analysis on a single memory extract. As memory analysis identifies cryptographic artefacts with 100% success, the artefacts could be retained with network sessions for later decryption. This may benefit entities, such as cloud vendors, to assist state agencies in decrypting criminals’ communications, without conflicting with local privacy laws. To achieve this aim, future work should focus on performance improvements such as multi-threading and pipe-lining, as well as investigating other protocols, encryption ciphers and modes, and technologies that uses encrypted communications channels.
| {
"pile_set_name": "ArXiv"
} |
---
address:
- 'Department of Mathematics, Keio University, 3-14-1 Hiyoshi, Kouhoku-ku, Yokohama, JAPAN'
- 'NWF-I Mathematik, Universität Regensburg, 93040 Regensburg, Germany'
author:
- Kenichi Bannai
- Guido Kings
title: '$p$-adic Beilinson conjecture for ordinary Hecke motives associated to imaginary quadratic fields'
---
[^1]
Introduction
============
The purpose of this article is to give an overview of the series of papers [@BK1] [@BK2] concerning the $p$-adic Beilinson conjecture of motives associated to Hecke characters of an imaginary quadratic field $K$, for a prime $p$ which splits in $K$. The $p$-adic $L$-function for such $p$ interpolating *critical values* of $L$-functions of Hecke characters associated to imaginary quadratic fields was first constructed by Vishik and Manin [@VM], and a different construction using $p$-adic Eisenstein series was given by Katz [@Ka1]. The $p$-adic Beilinson conjecture, as formulated by Perrin-Riou in [@PR], gives a precise conjecture concerning the *non-critical values* of $p$-adic $L$-functions associated to general motives. The purpose of our research is to investigate the interpolation property at non-critical points of the $p$-adic $L$-function constructed by Vishik-Manin and Katz.
For simplicity, we assume in this article that the imaginary quadratic field $K$ has class number *one* and that the Hecke character $\psi$ we consider corresponds to an elliptic curve with complex multiplication defined over $\bbQ$. Let $a$ be an integer $>0$. The main theorem of this article (Theorem \[thm: main\]) is a proof of the $p$-adic Beilinson conjecture for $\psi^a$ (see Conjecture \[conj: PR\]), when the prime $p \geq 5$ is an *ordinary* prime. The authors would like to thank the organizers Takashi Ichikawa, Masanari Kida and Takao Yamazaki for the opportunity to present our research at the RIMS “Algebraic Number Theory and Related Topics 2009" conference.
The $p$-adic Beilinson conjecture {#section: p Beilinson}
=================================
Assume that $K$ is an imaginary quadratic field of *class number one*. Let $E$ be an elliptic curve defined over $\bbQ$. We assume in addition that $E$ has complex multiplication by the ring of integers $\cO_K$ of $K$. We let $\psi := \psi_{E/K}$ be the Grossencharacter of $K$ associated to $E_K := E \otimes_\bbQ K$ by the theory of complex multiplication, and we denote by $\frf$ the conductor of $\psi$.
We let $M(\psi)$ be the motive over $K$ with coefficients in $K$ associated to the Grossencharacter $\psi$. Then we have $M(\psi) = H^1(E_K)$, where $H^1(E_K)$ is the motive associated to $E_K$. The Hasse-Weil $L$-function of $M(\psi)$ is a function with values in $K \otimes_\bbQ \bbC$ given by $$L(M(\psi), s) = ( L(\psi_\tau,s))_{\tau : K \hookrightarrow \bbC},$$ where $\tau: K \hookrightarrow \bbC$ are the embeddings of the coefficient $K$ of $M(\psi)$ into $\bbC$ and $L(\psi_\tau,s)$ is the Hecke $L$-function $$L(\psi_\tau,s) = \prod_{(\frq,\frf)=1} \left(1 - \frac{\psi_\tau(\frq)}{N\frq^s} \right)^{-1}$$ associated to the character $
\psi_\tau: \bbA^\times_K \xrightarrow\psi K \overset{\tau}\hookrightarrow \bbC.
$ Here, the product is over the prime ideals $\frq$ of $K$ which are prime to $\frf$. For integers $a>0$ and $n$, we let $M^a = M(\psi^{a}) := M(\psi)^{\otimes_K a}$, which is a motive over $K$ with coefficients in $K$. Then the Hasse-Weil $L$-function $L(M^a, s)$ is given by the Hecke $L$-function $$L(M^a, s) = (L(\psi_\tau^a, s))_{\tau: K \hookrightarrow \bbC}$$ with values in $K \otimes_\bbQ\bbC$. We let $M^a_B$ be the Betti realization of $M^a$, which is a $K$-vector space of dimension *one*. We fix a $K$-basis $\omega^a_B$ of $M^a_B$. The de Rham realization $M^a_\dR(n)$ of $M^a(n)$ is the rank one $K \otimes_\bbQ K$-module $$M^a_\dR(n) = K \omega^{n-a,n} \bigoplus K \omega^{n,n-a},$$ with Hodge filtration given by $$F^m M^a_\dR(n) =
\begin{cases}
M^a_\dR(n) & m \leq -n \\
K\omega^{n,n-a} & -n < m \leq a-n \\
0 & \text{otherwise}.
\end{cases}$$
In what follows, we consider the case when $n>a$, which implies in particular that our motive is *non-critical*. We have in this case $F^0 M^a_\dR(n) = 0$. The tangent space of our motive is given by $$t^a_{n} := M^a_\dR(n)/ F^0 M^a_\dR(n) \cong M^a_\dR(n),$$ which is again a $K \otimes_\bbQ K$-module of rank one. Note that $\omega^a_{\tg,n} : = \omega^{n-a,n} + \omega^{n,n-a}$ gives a basis of $t^a_n$ as a $K\otimes_\bbQ K$-module.
We denote by $V^a_\infty(n)$ the $\bbR$-Hodge realization of $M^a(n)$. The Beilinson-Deligne cohomology $
H^1_\sD(K \otimes_\bbQ \bbR, V^a_\infty(n) )
$ is given as the cokernel of the natural inclusion $$%\thealpha_{M^a} :
M^a_B(n) \otimes_\bbQ\bbR \rightarrow t^a_n \otimes_\bbQ \bbR.$$ The Beilinson regulator map gives a homomorphism $$\label{eq: Beilinson regulator}
r_\infty : H^1_\mot(K, M^a(n)) \rightarrow H^1_\sD( K \otimes_\bbQ \bbR, V^a_\infty(n)),$$ from the motivic cohomology $H^1_\mot(K, M^a(n))$ of $K$ with coefficients in $M^a(n)$ to $H^1_\sD( K \otimes_\bbQ \bbR, V^a_\infty(n))$. Then $r_\infty \otimes_\bbQ \bbR$ is known to be surjective and is conjectured to be an isomorphism. We let $c^a_n$ be an element of $H^1_\mot(K, M^a(n))$ such that $r_\infty(c^a_n)$ generates $H^1_\sD( K_\frp, V^a_\infty(n))$ as a $K_\infty := K \otimes_\bbQ \bbR$-module. We define the complex period $\Omega_\infty(n)$ of $M^a(n)$ to be the determinant of the exact sequence $$\label{eq: complex determinant}
0 \rightarrow M^a_B(n) \otimes_\bbQ \bbR \rightarrow t^a_n \otimes_\bbQ \bbR
\rightarrow H^1_\sD(K \otimes_\bbQ \bbR, V^a_\infty(n) ) \rightarrow 0$$ for the basis $r_\infty(c^a_n)$, $\omega_{\tg,n}^a$, and $\omega^a_B$. The complex period is an element in $K_\infty$ and is independent of the choice of the basis up to multiplication by an element in $K^\times$. The value $L(M^a,n)$ is in $K \otimes_\bbQ \bbR$, and the weak Beilinson conjecture for $M^a(n)$ as proved by Deninger [@Den] gives the following (see Theorem \[thm: Deninger\] and Corollary \[cor: Deninger\] for the precise statement.)
\[thm: Deninger one\] The value $$\frac{L(M^a, n)}{\Omega_\infty(n)}$$ is an element in $K^\times$.
For any prime $p$, the étale realization $V^a_p(n)$ of our motive is a $K \otimes_\bbQ \bbQ_p$-vector space with continuous action of $\Gal(\ol K/K)$. We fix a prime $p \geq 5$ relatively prime to $\frf$ such that $E$ has good *ordinary* reduction at $p$. In this case, the ideal generated by $p$ splits as $(p) = \frp\frp^*$ in $K$. We fix a prime ideal $\frp$ of $K$ above $p$. Then the Bloch-Kato exponential map gives an isomorphism $$\label{eq: exponential map}
\exp_p : t^a_n \otimes_K K_\frp \xrightarrow\cong H^1_f( K_\frp, V^a_p(n)),$$ and the inverse of this isomorphism is denoted by $\log_p$. The $p$-adic étale regulator map gives a homomorphism $$\label{eq: p-adic regulator}
r_p : H^1_\mot(K, M^a(n)) \rightarrow H^1_f( K_\frp, V^a_p(n)),$$ and the map $r_p \otimes \bbQ_p$ is conjectured to be an isomorphism. Assuming that the $p$-adic regulator map $r_p$ is *injective*, we define the $p$-adic period $\Omega_p(n)$ of $M^a(n)$ to be the determinant of the map $\log_p$ for the basis $r_p(c^a_n)$ and $\omega_{\tg,n}^a$. In other words, $\Omega_p(n)$ is an element in $K_p := K \otimes_\bbQ K_\frp \cong K_\frp \bigoplus K_{\frp^*}$ satisfying $$\label{eq: p-adic period}
\log_p \circ\,\, r_{p}(c^a_n) = \Omega_p(n) \,\omega_{\tg,n}^a.$$ The $p$-adic period $\Omega_p(n)$ is independent of the choice of basis up to multiplication by an element in $K^\times$.
We need to assume the injectivity of the $p$-adic regulator $r_p$ to insure that the $p$-adic period $\Omega_p(n)$ is non-zero. Kato has proved in [@Kato] 15.15 the weak Leopoldt conjecture for any Hecke character of $K$. Hence by a result of Jannsen ([@Jan], Lemma 8), we may then conclude that $$H^2(\cO_K[1/p\frf_a],V^a_p(n))=0$$ for almost all $n$. This implies that $r_p$ is injective for such $n$.
The $p$-adic Beilinson conjecture as formulated by Perrin-Riou (see [@Col] Conjecture 2.7) specialized to our setting is given as follows.
\[conj: PR\] Let $a$ be an integer $>0$. Then there exists a $p$-adic pseudo-measure $\dmu^a$ on $\bbZ_p$ with values in $K_p$ such that the value $$L_p(\psi^a \otimes \chi^n_\cyc) := \int_{\bbZ^\times_p} w^n \dmu^a(w)$$ in $K_p$ for any integer $n > a$ satisfies $$\frac{L_p(\psi^a \otimes \chi^n_\cyc)}{\Omega_p(n)}
= \left(1 - \frac{\psi(\frp)^a}{p^n} \right) \left(1 - \frac{\ol{\psi}(\frp^*)^a}{p^{a+1-n}} \right) \frac{\Gamma(n) L(\psi^a, n)}{\Omega_\infty(n)},$$ where $\frp$ is a fixed prime in $K$ above $p$.
If $\frf_a\neq(1)$ for the conductor $\frf_a$ of $\psi^a$, then $\dmu^a$ should in fact be a $p$-adic measure. Note that the dependence of the pseudo-measure on the choices of the basis $\omega^a_{\tg,n}$ and $c^a_n$ cancel, where as the pseudo-measure depends on the choice of the basis $\omega^a_B$. The main goal of our research is to prove that the $p$-adic measure constructed by Vishik-Manin and Katz gives the pseudo-measure of the above conjecture when the prime $p$ is split in $K$.
The main theorem of this article (Theorem \[thm: main\]) is the proof of the above conjecture for integers $n$ such that the corresponding $p$-adic regulator map $r_p$ is injective.
Construction of the Eisenstein class {#section: Eisenstein class}
====================================
The main difficulty in the proof of the Beilinson and $p$-adic Beilinson conjectures is to construct the element $c^a_n \in H^1_\mot(K, M^a(n))$ for $M^a: = M(\psi^a)$ and to calculate the images $r_\infty(c^a_n)$ and $r_p(c^a_n)$ with respect to the Beilinson-Deligne and $p$-adic regulator maps. We will use the Eisenstein symbol as constructed by Beilinson.
We fix an integer $N \geq 3$, and let $M(N)$ be the modular curve defined over $\bbZ[1/N]$ parameterizing for any scheme $S$ over $\bbZ[1/N]$ the pair $(E, \levelstructure)$, where $E$ is an elliptic curve over $S$ and $$\levelstructure: (\bbZ/N\bbZ)^2 \xrightarrow\cong E[N]$$ is a full level $N$-structure on $E$, where $E[N]$ is the group of $N$-torsion points of $E$. We let $\pr : \wt E \rightarrow M$ be the universal elliptic curve over $M$ with universal level $N$-structure $\wt\levelstructure : (\bbZ/N\bbZ)^2\cong \wt E[N]$, and consider the motivic sheaf $\bbQ(1)$ on $\wt E$. We let $$\label{eq: H}
\sH := R^1 \pr_* \bbQ(1),$$ and we denote by $\Sym^k \sH$ the $k$-th symmetric product of $\sH$. Let $\varphi = \sum_{\rho \in \wt E[N] \setminus \{0\}} a_\rho [\rho]$ be a $\bbQ$-linear sum of non-zero elements in $\wt E[N]$. For any integer $k>0$, the Eisenstein class $\Eis_\mot^{k+2}(\varphi)$ is an element $$\Eis_\mot^{k+2}(\varphi) \in H^1_\mot(M, \Sym^k \sH(1)).$$ Although the formalism of mixed motivic sheaves or motivic cohomology with coefficients have not yet been fully developed, one can give meaning to the above sheaves and cohomology (see [@BL], [@BK1] for details).
Then the class $c^a_n$ may be constructed from the Eisesntein class as follows. Let $K$ be an imaginary quadratic field of class number one, and let $E$ be an elliptic curve defined over $\bbQ$ with complex multiplication by the ring of integers $\cO_K$ of $K$. We denote again by $\psi$ the Hecke character of $K$ corresponding to $E_K$ with conductor $\frf$. We take $N \geq 3$ such that $N$ is divisible by $\frf$. For the extension $F: = K(E[N])$ of $K$ generated by the coordinates of the points in $E[N]$, we let $G_{F/K} := \Gal(F/K)$ the Galois group of $F$ over $K$. We fix a level $N$-structure $\levelstructure: (\bbZ/N\bbZ)^2\cong E[N]$ of $E$ over $F$, and we denote by $\levelstructure^\sigma$ the composition of $\levelstructure$ with the action of $\sigma \in G_{F/K}$. Then for any $\sigma \in G_{F/K}$, we denote by $\iota^{\sigma*}$ the pull-back with respect to the $F$-valued point $\iota^\sigma: \Spec\, F \rightarrow M$ of $M$ corresponding to $(E,\levelstructure^\sigma)$. Then the image of the sum $
\iota^* :=\sum_{\sigma \in G_{F/K}} \iota^{\sigma*}
$ is invariant by the action of the Galois group, hence gives a pull-back morphism $$H^1_\mot(M, \Sym^k \sH(1)) \xrightarrow{\iota^{*}} H^1_\mot(K, \Sym^k \iota^*\!\sH(1)).$$ Note that on $\Spec\, K$, the motivic sheaf $\iota^*\!\sH$ is given by the motive $H^1(E)(1)$, which by definition corresponds to the motive $M(\psi)(1)$. The structure of $K$-coefficients on $\iota^*\!\sH$ gives the following decomposition.
\[lem: decomposition\] For integers $j$ satisfying $0 \leq j \leq k/2$, we have the decomposition of motives $$\Sym^k \iota^* \!\sH =
\bigoplus_{0 \leq j \leq \frac{k}{2}} M(\psi^{k-2j})\left(k-j\right),$$ where we take the convention that for $k=2j$, we let $M(\psi^0)(k/2)$ be the Tate motive $\bbQ(k/2)$ with coefficients in $\bbQ$.
Let $a>0$ be an integer and we let $\frf_a$ be the conductor of $\psi^a$. We let $F_a := K(E[\frf_a])$ be the extension of $K$ generated by the coordinates of the points in $E[\frf_a]$, and we let $w_{F/F_a}$ be the order of the Galois group $\Gal(F/F_a)$. The Eisenstein classes $\Eis^{k+2}_\mot(\rho)$ are defined for points $\rho \in \wt E[N]\setminus \{0\}$ but is *not* defined for $\rho =0$. Hence in defining $c^a_n$, we differentiate between the case when $\frf_a \neq (1)$ and $\frf_a =(1)$.
\[def: varphi\] We define $\varphi_a$ as follows.
1. If $\frf_a \neq (1)$, then we fix a primitive $\frf_a$-torsion point $\rho_a$ of $E$ and let $$\varphi_a := \frac{1}{w_a w_{F/F_a}} [\rho_a],$$ where we denote again by $\rho_a$ the $N$-torsion point of $\wt E$ corresponding to $\rho_a$ through $\nu$ and $\wt\nu$, and $w_a$ is the number of units in $\cO_K$ which are congruent to *one* modulo $\frf_a$.
2. If $\frf_a=(1)$, then we let $$\varphi_a := \frac{1}{w_a w_{F/F_a}} \sum_{\rho\in\wt E[N]\setminus\{0\}} [\rho].$$
We define the class $c^a_n$ as follows.
\[def: can\] For any integer $a, n$ such that $n > a >0$, we let $k = 2n - a -2$. Then the motive $M^a(n) := M(\psi^a)(n)$ is a direct summand of $\Sym^k \iota^*\!\sH(1)$. We define the motivic class $c^a_n$ to be the image of $\Eis^{k+2}_\mot(\varphi_a)$ with respect to the projection $$H^1_\mot(K, \Sym^k \iota^*\!\sH(1)) \rightarrow H^1_\mot(K, M^a(n)),$$ where $\varphi_a$ is as in Definition \[def: varphi\].
Let $p$ be a rational prime which does not divide $\frf$, and we take $N \geq 3$ to be an integer divisible by $\frf$ and prime to $p$. In order to prove the $p$-adic Beilinson conjecture, it is necessary to calculate the images of $c^a_n$ with respect to the Beilinson-Deligne and $p$-adic regulator maps. The image $r_\infty(c^a_n)$ by the Beilinson-Deligne regulator map was calculated by Deninger [@Den]. We will calculate the image $r_p(c^a_n)$ by the $p$-adic regulator map using rigid syntomic cohomology.
Denote by $M^a_\cris(n)$ the crystalline realization of $M^a(n)$, which is a filtered module with a $\sigma$-linear action of Frobenius, and let $H^1_\syn(K_\frp, M^a_\cris(n))$ be the syntomic cohomology of $K_\frp$ with coefficients in $M^a_\cris(n)$. Then noting that $t^a_n \otimes_\bbQ \bbQ_p= M^a_\cris(n)$ in this case, there exists a canonical isomorphism $$\label{eq: equation one}
t^a_n \otimes_\bbQ \bbQ_p\xrightarrow\cong H^1_\syn(K_\frp, M^a_\cris(n)).$$ If we let $V^a_p(n)$ be the $p$-adic étale realization of $M^a(n)$, then we have a canonical isomorphism $$\label{eq: p-adic comparison}
H^1_\syn(K_\frp, M^a_\cris(n)) \xrightarrow\cong H^1_f(K_\frp,V^a_p(n)),$$ which combined with gives the exponential map . The syntomic regulator map $$r_\syn : H^1_\mot(K, M^a(n)) \rightarrow H^1_\syn(K_\frp, M^a_\cris(n))$$ defined by Besser ([@Bes1] §7) is compatible with the $p$-adic regulator $r_p$ through the isomorphism ([@Bes1] Proposition 9.9). Therefore, in order to calculate $\log_p \circ \, r_p(c^a_n)$, it is sufficient to calculate the image of $r_\syn(c^a_n)$ with respect to . We will calculate this image using the explicit determination of the syntomic Eisenstein class given in [@BK1].
Eisenstein class and $p$-adic Eisenstein series
===============================================
In this section, we review the explicit description of the syntomic Eisenstein class in terms of $p$-adic Eisenstein series given in [@BK1]. Let $M := M(N)$ be the modular curve over $\bbZ[1/N]$ given in the previous section. We will first describe a certain real analytic Eisenstein series $E^\infty_{k+2,l,\varphi}$.
Let $\Gamma \subset \bbC$ be a lattice, and we denote by $A$ the area of the fundamental domain of $\Gamma$ divided by $\thepi := 3.14159\cdots$. For any integer $a$ and complex number $s$ satisfying $\operatorname{Re}(s) > a/2+1$, the Eisenstein-Kronecker-Lerch series $K^*_a(z,w,s ; \Gamma)$ to be the series $$K^*_a(z,w,s ; \Gamma) :={ \sum_{\gamma \in \Gamma}}^* \frac{(\ol z+\ol\gamma)^a}{|z+\gamma|^{2s}}
\pair{\gamma,w}$$ where $\sum^*$ denotes the sum over $\gamma \in \Gamma$ satisfying $\gamma \not=-z$ and $\pair{z, w} := \exp((\ol w z - w \ol z)/A)$. By [@We] VIII §12 (see [@BKT] Proposition 2.4 for the case $a < 0$), this series for $s$ continues meromorphically to a function on the whole $s$-plane, holomorphic except for a simple pole at $s=1$ when $a=0$ and $w \in \Gamma$. This function satisfies the functional equation $$\label{eq: functional equation}
\Gamma(s) K^*_a(z,w,s; \Gamma) = A^{a+1-2s} \Gamma(a+1-s) K^*_a(w,z,a+1-s) \pair{w,z}.$$
We fix a level $N$-structure $\levelstructure: (\bbZ/N\bbZ)^2 \cong \frac{1}{N}\Gamma/\Gamma$, and let $\rho \in \frac{1}{N}\Gamma/\Gamma$. For integers $k$ and $l$, we define the real analytic Eisenstein series $E^\infty_{k+2,l,\rho}$ to be the modular form on $M_\bbC := M(N) \otimes_\bbQ \bbC$ whose value at the test object $(\bbC/\Gamma, dz, \levelstructure)$ is given by $$\label{eq: Eisenstein series}
E^\infty_{k+2,l,\rho}(\bbC/\Gamma, dz, \levelstructure):=A^{-l}\left. \Gamma(s) K^*_{k+l+2}(0, \rho, s; \Gamma)\right|_{s=k+2}.$$ We let $
E^\infty_{k+2,l,\varphi} := \sum_\rho a_\rho E^\infty_{k+2,l,\rho}
$ for the $\bbQ$-linear sum $\varphi = \sum_\rho a_\rho [\rho]$.
When $l=0$, then $E^\infty_{k+2,0,\varphi}$ is a holomorphic Eisenstein series of weight $k+2$ on $M_\bbC$. From the $q$-expansion, we see in this case that this Eisenstein series is defined over $\bbQ$, and hence defines a section $E_{k+2,0,\varphi}$ in $\Gamma(M_\bbQ, \omega^{\otimes k} \otimes \Omega^1_{M_\bbQ})$ for $\omega := \pr_* \Omega^1_{\wt E/M}$. Denote by $\sH_\dR$ the de Rham realization of $\sH$, which is the coherent $\cO_{M_\bbQ}$-module $R^1 \pr_* \Omega^\bullet_{\wt E}$ with Gauss-Manin connection $$\nabla: \sH_\dR \rightarrow \sH_\dR \otimes \Omega^1_{M_\bbQ},$$ and let $\Sym^k \sH_\dR$ be the $k$-th symmetric product of $\sH_\dR$ with the induced connection. From the natural inclusion $\omega^{\otimes k} \hookrightarrow \Sym^k \sH_\dR$, we see that $E_{k+2,0,\varphi}$ defines a section in $\Gamma(M_\bbQ, \Sym^k \sH_\dR \otimes \Omega^1_{M_\bbQ})$.
Let $p$ be a prime number not dividing $N$. We denote by $\sH_\rig$ the filtered overconvergent $F$-isocrystal associated to $\sH$ on $M_{\bbZ_p}$, which is given by $\sH_\dR$ with an additional structure of Hodge filtration and Frobenius. Let $H^1_\syn(M_{\bbZ_p}, \Sym^k \sH_\rig(1))$ be the rigid syntomic cohomology of $M_{\bbZ_p}$ with coefficients in $\Sym^k \sH_\rig(1)$. The rigid syntomic regulator is a map $$r_\syn: H^1_\mot(M, \Sym^k \sH(1)) \rightarrow H^1_\syn(M_{\bbZ_p}, \Sym^k \sH_\rig(1)),$$ and we define the syntomic Eisenstein class $\Eis^{k+2}_\syn(\varphi)$ to be the image by the syntomic regulator of the motivic Eisenstein class. We let $M^\ord_{\bbZ_p}$ be the ordinary locus in $M_{\bbZ_p}$, and $M^\ord_{\bbQ_p}: = M^\ord_{\bbZ_p} \otimes_{\bbZ_p} \bbQ_p$. By [@BK1] Proposition A.16, a class in $
H^1_\syn(M^\ord_{\bbZ_p}, \Sym^k \sH_\rig(1))
$ is given by a pair $(\cohomologyclass, \xi)$ of sections $$\label{eq: classes}
\begin{split}
\cohomologyclass &\in \Gamma(M^\ord_{\bbQ_p}, j^\dagger \Sym^k \sH_\rig(1)) \\
\xi & \in \Gamma(M^\ord_{\bbQ_p}, \Sym^k \sH_\dR \otimes_{\bbQ_p} \Omega^1_{M^\ord})
\end{split}$$ satisfying $\nabla(\cohomologyclass) = (1 - \phi^*) \xi$. The $\cohomologyclass$ for the class $(\cohomologyclass, \xi)$ corresponding to the restriction to the ordinary locus of the syntomic Eisenstein class $\Eis^{k+2}_\syn(\varphi)$ is given as follows.
We let $p \geq 5$ be a prime not dividing $N$, and we let $\cM$ be the $p$-adic modular curve defined over $\bbZ_p$ parameterizing the triples $(E_B, \eta, \levelstructure)$ consisting of an elliptic curve $E_B$ over a $p$-adic ring $B$, an isomorphism $$\label{eq: moduli}
\eta: \wh\bbG_m \cong \wh E_B$$ of formal groups over $B$, and a level $N$-structure $\levelstructure$. The ring of $p$-adic modular forms $V_p(\bbQ_p, \Gamma(N))$ is defined as the global section $$V_p(\bbQ_p, \Gamma(N)) := \Gamma(\cM, \cO_\cM) \otimes_{\bbZ_p} \bbQ_p.$$ The $q$-expansion gives an injection $$V_p(\bbQ_p, \Gamma(N)) \hookrightarrow \bbQ_p(\zeta_N)[[q]].$$ There exists a Frobenius action $\phi^*$ on $V_p(\bbQ_p, \Gamma(N))$ given on the $q$-expansion as $\phi^* = \Frob \otimes \sigma$, where $\Frob(q) = q^p$ and $\sigma$ is the the absolute Frobenius acting on $\bbQ_p(\zeta_N)$. The Eisenstein series $E_{k+2,0, \varphi}$ naturally defines an element in $V_p(\bbQ_p, \Gamma(N))$, and using the fact that the differential $\partial_{\log q} := q \frac{d}{dq}$ preserves the space of $p$-adic modular forms, we let for any integer $l \geq 0$ $$E_{k+l+2,l,\varphi} := \partial_{\log q}^l E_{k+2,0,\varphi}.% \in V_p(\bbQ_p, \Gamma(N)).$$ We let $
E^{(p)}_{k+2,0,\varphi} := (1 - \phi^*) E_{k+2,0,\varphi}
$ and $
E^{(p)}_{k+l+2,l,\varphi} := \partial_{\log q}^l E^{(p)}_{k+2,0,\varphi}% \in V_p(\bbQ_p, \Gamma(N)).
$ for any integer $l \geq 0$. Then the calculation of the $q$-expansion shows that we have $$\label{eq: Frobenius removal}
E^{(p)}_{k+l+2,l,\varphi} = (1 - p^l \phi^*) E_{k+l+2,l,\varphi}.$$ Following the method of Katz [@Ka1], we may construct a $p$-adic measure on $\bbZ_p \times \bbZ_p^\times$ with values in $V_p(\bbQ_p, \Gamma(N))$ satisfying the following interpolation property.
\[theorem: Katz measure\] There exists a $p$-adic measure $\dmu_\varphi$ on $\bbZ_p \times \bbZ_p^\times$ with values in $V_p(\bbQ_p, \Gamma(N))$ such that $$\int_{\bbZ_p \times \bbZ_p^\times} x^{k+1} y^l \dmu_\varphi(x,y) = E^{(p)}_{k+2,l,\varphi}$$ for integers $k > 0$, $l \geq 0$.
Using this measure, we define $E^{(p)}_{k+2,l,\varphi}$ for $l<0$ as follows.
\[def: p Eisenstein\] Let $k$ be an integer $ \geq -1$. We let $$E^{(p)}_{k+2,l,\varphi} := \int_{\bbZ_p \times \bbZ_p^\times} x^{k+1} y^l \dmu_\varphi(x,y)
\,\,\in V_p(\bbQ_p, \Gamma(N)),$$ where $l$ is any integer in $\bbZ$.
The $p$-adic Eisenstein series satisfies the differential equation $$\partial_{\log q} E^{(p)}_{k+2,l,\varphi} = E^{(p)}_{k+3,l+1,\varphi},$$ and the weight of $E^{(p)}_{k+2,l,\varphi}$ is $k+l+2$. The syntomic Eisenstein class may be described using these $p$-adic Eisenstein series. The moduli problem for $\cM$ implies that there exists a universal trivialization $$\eta: \wh\bbG_m \cong \wh{\wt E}$$ of the universal elliptic curve on $\cM$, which gives rise to a canonical section $\wt \omega$ of $\ul \omega := \pr_* \Omega^1_{\wt E/\cM}$corresponding to the invariant differential $d \log (1+T)$ on $\wh \bbG_m$. Since $\cM$ is affine, there exists sections $x$ and $y$ of $\wt E$ such that the elliptic curve $\wt E_{\bbQ_p} := \wt E \otimes \bbQ_p$ is given by the Weierstrass equation $$\wt E_{\bbQ_p}: y^2 = 4x^3-g_2x-g_3, \quad g_2, g_3 \in V_p(\bbQ_p, \Gamma(N))$$ satisfying $\wt \omega = dx/y$. Then the pull back of the $F$-isocrystal $\sH_\rig$ to $\cM_{\bbQ_p}$ is given as $$\sH_\rig = \cO_{\cM_{\bbQ_p}}\wt\omega^\vee \oplus \cO_{\cM_{\bbQ_p}}\wt u^\vee,$$ with connection $\nabla(\wt u^\vee) = \wt \omega^\vee \otimes d \log q$, $\nabla(\wt \omega^\vee) = 0$, Frobenius $\phi^*(\wt\omega^\vee) = p^{-1} \wt\omega^\vee$, $\phi^*(\wt u^\vee) = \wt u^\vee$ and Hodge filtration $\Fil^{-1} \sH_\rig = \sH_\rig$, $\Fil^0 \sH_\rig = \cO_{\cM_{\bbQ_p}}\wt u^\vee$, $\Fil^1 \sH_\rig = 0$ (See [@BK1] §4.3). If let $\wt\omega^{m,n} := \wt\omega^{\vee m} \wt u^{\vee n}$, then the filtered $F$-isocrystal $\Sym^k \sH_\rig(1)$ on $\cM_{\bbQ_p}$ is given by the coherent module $$\Sym^k \sH_\rig(1) = \bigoplus_{j=0}^k \cO_{\cM_{\bbQ_p}}\wt\omega^{k-j, j}(1)$$ with connection $\nabla(\wt\omega^{k-j, j}(1)) = j\,\wt\omega^{k-j +1, j-1}(1) \otimes d\log q$, Frobenius $$\phi^*(\wt\omega^{,k-j, j}(1)) = p^{j-k-1} \wt\omega^{k-j, j}(1),$$ and Hodge filtration $$\begin{aligned}
\Fil^m (\Sym^k \sH_\rig(1)) = \bigoplus_{j=m+k+1}^k \cO_{\cM_{\bbQ_p}}\wt\omega^{k-j, j}(1).\end{aligned}$$ If we let $\wt\cohomologyclass^{k+2}_\Eis$ be the section $$\wt\cohomologyclass^{k+2}_\Eis(\varphi) := \sum_{j=0}^k \frac{(-1)^{k-j}}{j!} E^{(p)}_{j+1,j-k-1,\varphi}\, \wt\omega^{k-j, j}(1),$$ then we have $$\nabla(\wt\cohomologyclass^{k+2}_\Eis(\varphi)) =\frac{(1 - \phi^*) E_{k+2,0,\varphi}}{k!}\, \wt\omega^{0,k}(1) \otimes d\log q.$$ The main result of [@BK1] is the following.
\[thm: BK\] For any integer $k>0$, the syntomic Eisenstein class $$\Eis^{k+2}_\syn(\varphi) \in H^1_\syn(M_{\bbZ_p}, \Sym^k \sH_\rig(1))$$ restricted to the ordinary locus $H^1_\syn(M^\ord_{\bbZ_p}, \Sym^k \sH_\rig(1))$ is represented by the pair $(\cohomologyclass, \xi)$ as in , where $\xi = E_{k+2,0,\varphi} \wt\omega^{0,k}(1)/k! \otimes d \log q$ and $\cohomologyclass$ is a section which maps to $\wt\cohomologyclass^{k+2}_{\Eis}(\varphi)$ in $\Gamma(\cM_{\bbQ_p}, \Sym^k \sH_\rig(1))$.
The main ingredient in the proof of the above theorem is the characterization of $\xi$ by the residue, which by [@BL] 2.2.3 (see also [@HK1] C.1.1) and the compatibility of the Beilinson-Deligne regulator map with the residue morphism shows that $\xi$ represents the de Rham Eisenstein class in de Rham cohomology. See [@BK1] Proposition 3.6 and Proposition 4.1 for details concerning this point.
Special values of Hecke $L$-functions {#section: Hecke}
=====================================
In this section, we give in Propositions \[pro: L-value\] and \[pro: L-value-2\] the precise relation between the special values of the Hecke $L$-function $L(\psi^a,s)$ and Eisenstein-Kronecker-Lerch series. Assume that $K$ is an imaginary quadratic field of class number one, and let $E$ be an elliptic curve over $\bbQ$ with good ordinary reduction at a prime $p$ with complex multiplication by the ring of integers $\cO_K$ of $K$. We let $\psi$ be the Grossencharacter of $K$ associated to $E_K := E \otimes_\bbQ K$, and we denote by $\frf$ the conductor of $\psi$. We fix an invariant differential $\omega$ of $E$ defined over $K$. We fix once and for all a complex embedding $\tau: K \hookrightarrow \bbC$ of the base field $K$ into $\bbC$, and we let $\Gamma$ be the period lattice of $E: = E \otimes_{K, \tau} \bbC$ with respect to $\omega$. Then we have a complex uniformization $$\label{eq: complex uniformization}
\bbC/\Gamma \xrightarrow\cong E(\bbC)$$ such that the pull-back of the invariant differential $\omega$ coincides with $dz$. Note that since $E$ has complex multiplication, we have $\Gamma = \Omega \cO_K$ for some complex period $\Omega \in \bbC^\times$.
By abuse of notation, we will denote by $\psi$ and $\ol\psi$ the complex Hecke characters $\psi_\tau$ and $\ol{\psi_\tau}$ associated to $\psi$, where $\tau$ is the fixed embedding given above. Let $-d_K$ denote the discriminant of $K$, so that $K = \bbQ(\sqrt{-d_K})$. The Hecke character $\psi$ is of the form $\psi(( u)) = \varepsilon( u) u$ for any $ u \in \cO_K$ prime to $\frf$, where $\varepsilon: (\cO_K/\frf)^\times \rightarrow K^\times$ is a primitive character on $(\cO_K/\frf)^\times$.
Let $\chi: (\cO_K/\frf_\chi)^\times \rightarrow K^\times$ be a primitive character of conductor $\frf_\chi$, and let $f_\chi$ be a generator of $\frf_\chi$. Then for any $u$ in $\cO_K$, we define the *Gauss sum* $G(\chi, u)$ by $$G(\chi, u) : = \sum_{ v \in \cO_K/\frf_\chi} \ol\chi( v)
\exp\left( 2 \thepi i \Tr_{K/\bbQ}\left( u v/f_\chi\sqrt{-d_K}\right)\right)$$ (see [@L] Chapter 22 §1), where we extend $\chi$ to a function on $\cO_K/\frf_\chi$ by taking $\chi(\thealpha) :=0$ for any $\thealpha \in \cO_K$ not prime to $\frf_\chi$. We let $G(\chi) := G(\chi,1)$. Then the standard fact concerning Gauss sums are as follows (see for example [@L] Chapter 22 §1.)
\[lem: Gauss sum\] Let the notations be as above.
1. We have $|G(\chi)|^2= N(\frf_\chi)$.
2. For any $u \in \cO_K$, we have $
G(\chi, u) = \chi(u) G(\chi).
$
As in §\[section: Eisenstein class\], we let $a>0$ be an integer and $\frf_a$ be the conductor of $\psi^a$. Then the finite part $\varepsilon^a$ of $\psi^a$ is a primitive character $\varepsilon^a: (\cO_K/\frf_a)^\times \rightarrow \bbC^\times$ of conductor $\frf_a$. We fix a generator $f_a$ of $\frf_a$ and we denote by by $G(\varepsilon^a,u)$ the corresponding Gauss sum for any $u$ in $\cO_K$.
We let the notations be as in §\[section: Eisenstein class\]. In particular, we let $w_a$ be the number of units in $\cO_K$ which are congruent to *one* mod $\frf_a$, and we let $w_{F/F_a}$ be the order of the Galois group $\Gal(F/F_a)$. We again let $N \geq 3$ be a rational integer divisible by $\frf$ and prime to $p$, and $F := K(E[N])$. We fix an isomorphism $\levelstructure: (\bbZ/N\bbZ)^2\cong\frac{1}{N}\Gamma/\Gamma$.
We first consider the case when $\frf_a \neq (1)$. We let $\rho_a: = \Omega/f_a$ be a primitive $\frf_a$-torsion point, which corresponds through the uniformization to a point $\rho_a \neq 0 \in E(\ol K)$. We then have the following.
\[pro: L-value\] Suppose $\frf_a \neq (1)$. Then we have $$\frac{(-1)^a}{w_a w_{F/F_a}} \sum_{\sigma \in \Gal(F/K)} E^\infty_{n,a-n, \rho_a^\sigma}(\bbC/\Gamma, dz, \nu) =
\frac{G(\varepsilon^a)\ol\Omega^a}{A^{a-n}|\Omega|^{2n}} \Gamma(s) L(\psi^a,s)|_{s=n}.$$
Let $w_0$ be the number of units in $\cO_K$. By definition, we have $$L(\psi^a, s) = \frac{1}{w_0} \sum_{\thealpha \in \cO_K} \frac{\psi^a(\thealpha)}{N(\thealpha)^s}
%= \frac{1}{w_0} \sum_{\thealpha \in \cO_K/\frf_a} \varepsilon^a(\thealpha)
%\sum_{\gamma\in\frf_a} \frac{(\thealpha+\gamma)^a}{N(\thealpha+\gamma)^s}.
= \frac{1}{w_0} \sum_{\thealpha \in \cO_K} \frac{\varepsilon^a(\thealpha)\thealpha^a}{N(\thealpha)^s}$$ Then Lemma \[lem: Gauss sum\] (2) gives the equality $\varepsilon^a(\thealpha) = G(\varepsilon^a, \thealpha)/G(\varepsilon^a)$. If we expand the definition of the Gauss sum, we see that $$L(\psi^a, s) = \frac{1}{w_0} \sum_{\substack{\thealpha \in \cO_K\\ \thebeta\in\cO_K/\frf_a}}
\frac{ \ol\varepsilon^a(\thebeta)\thealpha^a}{N(\thealpha)^s}
\exp\left( 2\pi i \Tr_{K/\bbQ}\left( \frac{\thealpha \thebeta}{f_a \sqrt{-d_K}} \right) \right).$$ Noting that $\cO_K$ is preserved by complex conjugation, we see that the above is equal to $$\frac{(-1)^a}{w_0} \sum_{\thebeta \in \cO_K/\frf_a } \sum_{\thealpha\in \cO_K }
\frac{ \ol\varepsilon^a(\thebeta)\ol \thealpha^a}{|\thealpha|^{2s}} \exp\left( \frac{2\pi}{\sqrt{d_K}}
\left(\frac{\thealpha\ol{\thebeta}}{\ol f_a}- \frac{\ol\thealpha \thebeta}{f_a} \right) \right).$$ For any $\sigma \in \Gal(F/K)$, we have $\rho_a^\sigma = \rho_a^{\sigma'}$, where $\sigma'$ is the class of $\sigma$ in $\Gal(F_a/K)$. If $\sigma'_\thebeta := (\thebeta, F_a/F)$ is the element in $\Gal(F_a/K)$ corresponding to $\thebeta \in (\cO_K/\frf_a)^\times$ through the *inverse* of the Artin map, then by the theory of complex multiplication, we have $
\rho_a^{\sigma'_{\thebeta}} = \psi(\thebeta) \rho_a.
$ Hence $$\begin{gathered}
\sum_{\sigma \in \Gal(F/K)} K_a^*(0, \rho_a^\sigma, s; \Gamma)
= w_{F/F_a} \sum_{\sigma' \in \Gal(F_a/K)} K_a^*(0, \rho_a^{\sigma'}, s; \Gamma)\\
= w_{F/F_a} \frac{w_a}{w_0} \sum_{\thebeta \in (\cO_K/\frf_a)^\times} \sum_{\gamma \in \Gamma} \frac{\ol\gamma^a}{|\gamma|^{2s}}
%\exp\left(\frac{\gamma\ol{\psi(\thebeta)\rho_a} - \ol\gamma \psi(\thebeta) \rho_a}{A}\right)\\
\pair{\gamma, \psi(\thebeta)\rho_a}\\
= w_{F/F_a} \frac{w_a}{w_0}
\sum_{\thebeta \in (\cO_K/\frf_a)^\times} \sum_{\gamma \in \Gamma} \frac{\ol\varepsilon^a(\thebeta)\ol\gamma^a}{|\gamma|^{2s}}
%\exp\left(\frac{\gamma\ol{\thebeta\rho_a} - \ol\gamma \thebeta \rho_a}{A}\right).
\pair{\gamma, \thebeta \rho_a}.
\end{gathered}$$ Our assertion follows from the fact that $\Gamma = \Omega \cO_K$, $A = |\Omega|^2\sqrt{d_K}/2 \thepi$ and the definition of the Eisenstein-Kronecker-Lerch series.
The right hand side of Proposition \[pro: L-value\] may be used to express the Hecke $L$-function on the other side of the functional equation as follows.
\[lem: pre functional\] We have $$\begin{gathered}
\label{eq: pre functional}
\frac{1}{w_a w_{F/F_a}}\sum_{\sigma \in \Gal(F/K)} E^\infty_{n,a-n, \rho_a^\sigma}(\bbC/\Gamma, dz, \nu)\\=
\frac{A^{1-n} N(\frf_a)^{a+1-n}\ol\Omega^a}{\ol f_a^a |\Omega|^{2(a+1-n)}} \left.\Gamma(s) L(\ol\psi^a, s) \right|_{s=a+1-n}.
\end{gathered}$$
We have by definition $$\begin{gathered}
\sum_{\thebeta\in\cO_K/\frf_a} K^*_{a}(\psi(\thebeta)\rho_a, 0 , a+1-s; \Gamma) = \sum_{\thebeta\in\cO_K/\frf_a}
{\sum_{\gamma \in \Gamma}}^* \frac{(\ol{\psi(\thebeta) \rho_a} + \ol\gamma)^a}{|\psi(\thebeta)\rho_a+\gamma|^{2(a+1-s)}}\\
= \frac{ N(\frf_a)^{a+1-s} \ol\Omega^a}{\ol f_a^a |\Omega|^{2(a+1-s)}} L(\ol\psi^a, a+1-s)
\end{gathered}$$ for $\operatorname{Re}(s) < a/2$, hence for any $s \in \mathbb{C}$ by analytic continuation. Our assertion follows from the functional equation $$\Gamma(s) K^*_{a}(0, \psi(\thebeta) \rho_a, s; \Gamma) =
A^{a+1-2s}\Gamma(a+1-s) K^*_{a}(\psi(\thebeta)\rho_a, 0 , a+1-s; \Gamma)$$ and the definition of the Eisenstein-Kronecker-Lerch series.
The case when $\frf_a=(1)$ is given as follows.
\[pro: L-value-2\] Suppose $\frf_a = (1)$. Then we have $$\begin{gathered}
\frac{1}{w_a} \sum_{\rho \in E[N] \setminus \{0\}}
E^\infty_{n,a-n,\rho}(\bbC/\Gamma, dz, \nu)
\\= \left( \frac{N^{a+2}}{N^{2n}} -1 \right) \frac{\ol\Omega^a}{A^{a-n} |\Omega|^{2n}}\Gamma(s) L(\psi^a,s)|_{s=n}.
\end{gathered}$$
By definition, we have $$%\begin{multline*}
\sum_{\rho \in \frac{1}{N} \Gamma/\Gamma} K^*_a(0, \rho, s; \Gamma) =
\sum_{\gamma\in \Gamma}\sum_{\rho \in \frac{1}{N} \Gamma/\Gamma}
\frac{\ol\gamma^a}{|\gamma|^{2s}} \pair{\gamma,\rho} \\
= \frac{N^{a+2}}{N^{2s}} \sum_{\gamma\in \Gamma} \frac{\gamma^a}{|\gamma|^{2s}},$$ where the last equality follows from the equality $$\sum_{\rho\in\frac{1}{N}\Gamma/\Gamma} \pair{\gamma,\rho} =
\begin{cases}
N^2 & \gamma \in N\Gamma\\
0 & \text{otherwise}
\end{cases}$$ and the fact that complex conjugation acts bijectively on $\Gamma$. Our assertion follows from the definition of the Eisenstein-Kronecker-Lerch series.
Similarly to Lemma \[lem: pre functional\], we have the following.
We have $$\begin{gathered}
\label{eq: pre functional-2}
\frac{1}{w_a} \sum_{\rho \in E[N] \setminus \{0\}} E^\infty_{n,a-n,\rho}(\bbC/\Gamma, dz, \nu))\\=
\left( \frac{N^{a+2}}{N^{2n}} -1\right)
\frac{A^{1-n} \ol\Omega^a}{ |\Omega|^{2(a+1-n)}} \left.\Gamma(s) L(\ol\psi^a, s) \right|_{s=a+1-n}.
\end{gathered}$$
The Main Result
===============
In this section, we give an outline of the proof of our main theorem. We will mainly deal with the case when $\frf_a \neq (1)$, as the case for $\frf_a = (1)$ is essentialy the same except for the factor $(N^{a+2}/N^{2n}-1)$. We first calculate the $p$-adic and complex periods $\Omega_p(n)$ and $\Omega(n)$. From the definition of $c^a_n$ and from the compatibility of the syntomic regulator with respect to pull-back morphisms, the restriction of the syntomic Eisenstein class through the decomposition of Lemma \[lem: decomposition\] gives the image by the syntomic regulator of the element $c^a_n$ in $H^1_\mot(K, M^a(n))$.
Let the notations be as in the previous section. We denote by $\omega^*$ the class in $H^1_\dR(E/\bbC)$ corresponding to $d \ol z/A$, which is in fact a class in $H^1_\dR(E/K)$. Let $k = 2n - a -2$. Then $\omega^{k-j+1, j+1} := \omega^{\vee k-j} \omega^{*\vee j}(1)$ for $0 \leq j \leq k$ form a basis of $\Sym^k \iota^*\sH_\dR(1)$. The relation between the basis $\wt\omega^{m,n}$ and $\omega^{m,n}$ is given by $\wt\omega^{m,n} =\Omega_\frp^{n-m}\omega^{m,n}$. In what follows, let $
\varphi_a
$ be as in Definition \[def: varphi\]. By Theorem \[thm: BK\], the pull-back of the syntomic Eisenstein class $\Eis^{k+2}_{\syn}(\varphi_a)$ to $H^1_\syn(K_\frp, \Sym^k\iota^*\sH(1))$ is expressed by the element $$\iota^* \widetilde\cohomologyclass^{k+2}_{\Eis}(\varphi_a) = \sum_{j=0}^k \frac{(-1)^{k-j}}{j!} \Omega_\frp^{2j-k}
E^{(p)}_{j+1, j-k-1,\varphi_a}(E, \omega, \levelstructure)\omega^{k-j+1, j+1}.$$ Hence the element $r_\syn(c^a_n)$ in $H^1_\syn(K_\frp, M^a_\cris(n))$ corresponding by definition to the direct factor $j = n-1$ and $j=n-a-1$ is represented by $$\begin{gathered}
\frac{ (-1)^{n-a+1}}{\Gamma(n)} \Omega_\frp^{a} E^{(p)}_{n, a-n,\varphi_a}(E, \omega, \levelstructure) \omega^{n-a,n} \\+
\frac{ (-1)^{n-1}}{\Gamma(n-a)} \Omega_\frp^{-a} E^{(p)}_{n-a, -n,\varphi_a}(E, \omega, \levelstructure) \omega^{n,n-a}.\end{gathered}$$ By definition of the exponential map, the element in $t^a_n \otimes \bbQ_p$ corresponding to $r_\syn(c^a_n)$ through the isomorphism is $$\begin{gathered}
\frac{ (-1)^{n-a+1}}{\Gamma(n)}\Omega_\frp^{a} E_{n, a-n,\varphi_a}(E, \omega, \levelstructure) \omega^{n-a,n} \\
+ \frac{ (-1)^{n-1}}{\Gamma(n-a)}\Omega_\frp^{-a} E_{n-a, -n,\varphi_a}(E, \omega, \levelstructure) \omega^{n,n-a},\end{gathered}$$ where $E_{n, a-n,\varphi_a}(E, \omega, \levelstructure)$ is the element in $\widehat K^{\ur}_\frp$ satisfying $$\label{eq: no p factor}
\left( 1 - p^{a-n} \sigma^* \right)E_{n, a-n,\varphi_a}(E, \omega, \levelstructure) = E^{(p)}_{n, a-n,\varphi_a}(E, \omega, \levelstructure).$$ From the definition of $c^a_n$ and the discussion at the end of §\[section: Eisenstein class\], this shows that we have $$\begin{gathered}
\log_p \circ \, r_p(c^a_n) = \frac{ (-1)^{n-a+1}}{\Gamma(n)} \Omega_\frp^{a} E_{n, a-n,\varphi_a}(E, \omega, \levelstructure) \omega^{n-a,n} \\
+ \frac{ (-1)^{n-1}}{\Gamma(n-a)}\ \Omega_\frp^{-a} E_{n-a, -n,\varphi_a}(E, \omega, \levelstructure) \omega^{n,n-a}.\end{gathered}$$ This gives the following.
\[theorem: p period\] Let $n$ be an integer $>a$ and assume that the $p$-adic regulator $r_p$ is injective. Then the $p$-adic period $\Omega_p(n) \in K_\frp \bigoplus K_{\frp^*}$ of the motive $M^a(n)$ is given by $$\begin{gathered}
\Omega_p(n) = \frac{ (-1)^{n-a+1}}{\Gamma(n)} \Omega_\frp^{a} E_{n, a-n,\varphi_a}(E, \omega, \levelstructure) \\
\bigoplus\frac{ (-1)^{n-1}}{\Gamma(n-a)} \Omega_\frp^{-a} E_{n-a, -n,\varphi_a}(E, \omega, \levelstructure).
\end{gathered}$$
The theorem follows from the definition given in of the $p$-adic period, noting that we have an isomorphism $$(K \otimes_\bbQ \bbQ_p)\, \omega^a_{\tg,n} \cong \bbQ_p \, \omega^{n,n-a} \bigoplus \bbQ_p\, \omega^{n-a,n}$$ induced from the canonical splitting $K \otimes \bbQ_p = K_\frp \bigoplus K_{\frp^*} \cong \bbQ_p \bigoplus \bbQ_p$.
The calculation of the complex period, originally due to Deninger [@Den] may be done in a similar fashion. If we let $\Gamma = \Omega\cO_K$ as in the previous section, then the Betti homology of $E$ is given by $
H_1^B(E(\mathbb{C}), \bbZ) = \Gamma.
$ We let $\gamma_1 := \Omega \in \Gamma$, which is a generator of $\Gamma$ as a $\cO_K$-module. If we fix a $\thebeta \in \cO_K$ such that $\cO_K := \bbZ \oplus \bbZ\thebeta$ and if we let $\gamma_2: = \thebeta(\gamma_1)$ where $\thebeta$ acts through the complex multiplication of $E$, then we have $\Gamma := \bbZ \gamma_1 \oplus \bbZ \gamma_2$ as a $\bbZ$-module. The period relation gives the equality $$\begin{pmatrix}
\gamma_1 \\ \gamma_2
\end{pmatrix}
=
\begin{pmatrix}
\Omega & \ol\Omega/A\\
\tau\Omega & \ol\tau\ol\Omega/A
\end{pmatrix}
\begin{pmatrix}
\omega^\vee \\ \omega^{*\vee}
\end{pmatrix}.$$ The $K$-basis $\gamma_1$ induces a $K$-basis of $M^a_B(n) \subset \Sym^k H^1_B(E(\bbC), \bbQ(1))$, which we denote by $\omega_B$. Then the inclusion $$M^a_B(n) \otimes_\bbQ \bbR \hookrightarrow t^a_n \otimes_\bbQ \bbR$$ maps $\omega_B$ to $$%\begin{multline*}
\Omega^{n-a} \frac{\ol\Omega^n}{A^n} \omega^{n-a,n} + \Omega^n \frac{\ol\Omega^{n-a}}{A^{n-a}} \omega^{n,n-a}. % \\
%= \frac{1}{\Omega^{a}} \left( \frac{2\thepi}{\sqrt{d_K}}\right)^n \omega^{n-a,n} + \Omega^{a}
%\left( \frac{2\thepi}{\sqrt{d_K}}\right)^{n-a}\omega^{n,n-a}.$$Furthermore, one may prove the following.
\[thm: Deninger\] The image $ r_\infty(c^a_n) $ in $$H^1_\sD(K \otimes_\bbQ \bbR, V^a_\infty(n)) \cong (t^a_n \otimes_\bbQ \bbR) / (M^a_B(n) \otimes_\bbQ \bbR)$$ of $c^a_n$ by the Beilinson regulator is represented by the element $$\begin{gathered}
\frac{(-1)^{n-1}}{\Gamma(n)} A^{n-a} E^\infty_{n,a-n,\varphi_a}(\bbC/\Gamma, dz, \levelstructure)\, \omega^{n-a,n}\\+
\frac{(-1)^{n-a+1}}{\Gamma(n-a)} A^n E^\infty_{n-a,-n,\varphi_a}(\bbC/\Gamma, dz, \levelstructure)
\, \omega^{n,n-a}%\pmod{M^a_N(B) \otimes_\bbQ \bbR}
\end{gathered}$$ in $t^a_n \otimes_\bbQ \bbC$.
The Eisenstein class in this paper defined using the elliptic polylogarithm is related to the Eisenstein class defined by Beilinson and Deninger. The theorem is then a special case of the weak Beilinson conjecture for Hecke character associated to imaginary quadratic fields proved by Deninger [@Den] (see also [@DW] for the case of an elliptic curve defined over $\bbQ$ with complex multiplication.) The theorem may also be proved by explicitly calculating the Hodge realization of the elliptic polylogarithm [@BL] (see also [@BKT] Theorem A 29.)
By taking the determinant of the complex with respect to the basis $r_\infty(c^a_n)$, $\omega_B$, $\omega^{n-a,n}$ and $\omega^{n,n-a}$, the above calculation and the definition of the complex period give the following.
\[prop: period\] The complex period $\Omega_\infty(n)$ of $M^a(n)$ in $K \otimes_\bbQ \bbR$ is given by $$%\begin{multline*}
\Omega_\infty(n) = (-1)^{a+n-1} G(\varepsilon^a) L(\psi^a,n) \, \bigoplus \,(-1)^{n+1} G(\ol\varepsilon^a) L(\ol\psi^a,n )$$ in $K \otimes_\bbQ \bbC \cong \bbC \bigoplus \bbC$ if $\frf_a \neq (1)$. A similar formula holds for the case when $\frf_a = (1)$, but with a factor $(N^{a+2}/N^{2n}-1)$ multiplied to the $L$-value.
The assertion follows from Theorem \[thm: Deninger\] by explicit calcuation, using the definition of $\varphi_a$ (Definition \[def: varphi\]), the calculation of the complex period above, and the relation between Eisenstein-Kronecker-Lerch series and special values of $L$-functions (Proposition \[pro: L-value\] if $\frf_a\neq(1)$, or Proposition \[pro: L-value-2\] if $\frf_a = (1)$).
This gives the following corollary, which we stated in Theorem \[thm: Deninger one\].
\[cor: Deninger\] If $\frf_a \neq (1) $, then we have $$\frac{L(\psi^a,n)}{\Omega_\infty(n)} =\frac{ (-1)^{a+n-1}}{G(\varepsilon^a)} \,\in\, K^\times \subset \,K \otimes_\bbQ \bbC.$$ A similar formula holds for the case when $\frf_a = (1)$, but with multiplication by $(N^{a+2}/N^{2n}-1)^{-1}$ on the right hand side.
The equalitiy follow from the calculation of the complex period in Proposition \[prop: period\]. Since $\varepsilon^a$ is a primitive Hecke character with values in $K$, we see that this value is in $K$.
We next construct the $p$-adic measure $\dmu^a$ which appears in the formulation of Conjecture \[conj: PR\]. Since $E$ has good ordinary reduction at $p$, the prime $p$ splits as $p =\frp \frp^*$ in $K$. In what follows, we fix once and for all complex and $p$-adic embeddings of our coefficient $K$ as follows. We let $\tau: K \hookrightarrow \bbC$ as in §\[section: Hecke\] and an embedding $K \hookrightarrow \bbC_p$ mapping $\frp$ to a prime in $\bbC_p$. With this convention, we may regard the complex and $p$-adic periods as elements respectively in $\bbC$ and $\bbC_p$, by taking the first components in Proposition \[prop: period\] and Theorem \[theorem: p period\].
Let $\wh E$ be the formal group of $E$ over $\cO_K$, and let $\wh K_\frp^\ur$ be the $p$-adic completion of the maximal unramified extension $K^\ur_\frp$ of $K_\frp$, which we regard as a subfield of $\bbC_p$ through our fixed embedding. Since $p$ is an ordinary prime, there exists an isomorphism of formal groups $\eta$ over $\cO_{\wh K_\frp^\ur}$ $$\eta: \wh E \xrightarrow\cong \wh\bbG_m$$ given by a power series $\eta(t) = \exp(\lambda(t)/\Omega_\frp)-1$, where $\Omega_\frp$ is a $p$-adic period of $E$ which is an element in $\cO_{\wh K_\frp^\ur}$ satisfying $$\label{eq: Frobenius action}
\Omega_\frp^{\sigma-1} = \psi(\frp)p^{-1}.$$ The above isomorphism gives the equality $$\label{eq: normalization}
\eta^*\left(d \log(1+T)\right) =\omega/\Omega_\frp.$$
Again let $N \geq 3$ be an integer as in §\[section: Eisenstein class\] divisible by $\frf$ and prime to $p$. By [@Ka1] 5.10.1, the value of the $p$-adic Eisenstein series $E_{k+2,l,\varphi_a}$ at the test object $(E, \omega, \levelstructure)$ is defined by $$\label{eq: value of specialization}
E_{k+2,l,\varphi_a}(E, \omega, \levelstructure) := \Omega_\frp^{k+l+2} E_{k+2,l,\varphi_a}(E, \eta, \levelstructure).$$ In addition, the comparison theorem [@Ka1] 8.0.9 states that this value for integers $k>0, l\geq 0$ is an element in $F:= K(\frf)$ satisfying the equality $$E_{k+2,l,\varphi_a}(E, \omega, \levelstructure) = E^\infty_{k+2,l,\varphi_a}(E, \omega, \levelstructure).$$ Then the calculation above and Theorem \[theorem: Katz measure\] gives the following.
We let $\varphi_a$ be as in Definition \[def: varphi\], and we denote again by $\dmu_{\varphi_a}$ the $p$-adic measure on $\bbZ_p \times \bbZ_p^\times$ obtained as the value of $\dmu_{\varphi_a}$ of Theorem \[theorem: Katz measure\] at $(E, \omega, \levelstructure)$. If $\frf_a\neq (1)$, then we have $$%\begin{multline*}
\frac{(-1)^a}{\Omega_\frp^a} \int_{\bbZ_p \times \bbZ_p^\times} x^{n-1} y^{a-n} \dmu_{\varphi_a}(x,y) = %E^{(p)}_{n,a-n,\varphi_a}
G(\varepsilon^a)\left(1 - \frac{\psi(\frp)^a}{p^n} \right)
\frac{\ol\Omega^a \Gamma(n)L(\psi^a, n)}{ A^{a-n} |\Omega|^{2n} }$$ for integers $a \geq n > 0$. A similar formula holds for the case when $\frf_a = (1)$, but with multiplication by $(N^{a+2}/N^{2n}-1)$ on the right hand side.
The relation between the action of the Frobenius on $V_p(\bbQ_p, \Gamma(N))$ and its specialization is given by $$\phi^*(E_{k+2,l,\varphi} )(E, \eta, \levelstructure) = E_{k+2,l,\varphi} (E,\eta, \levelstructure^{\sigma_\frp}),$$ since $(E,\omega)$ is define over $K$ and hence $E^{\sigma_\frp} = E$ and $\omega^{\sigma_\frp} = \omega$. Then from the definition of the specialization of $p$-adic modular forms and the action of the Frobenius on the $p$-adic period , we have $$\begin{gathered}
p^l \phi^*(E_{k+2,l,\varphi} )(E, \omega, \levelstructure)
= p^l \left(\Omega_\frp^{\sigma_\frp}\right)^{k+l+2} E_{k+2,l,\varphi} (E, \eta, \levelstructure^{\sigma_\frp})\\
= \psi(\frp)^{k+l+2} p^{-k-2} \Omega_\frp^{k+l+2} E_{k+2,l,\varphi^{\sigma_\frp}} (E, \eta, \levelstructure).
%= \psi(\frp)^{k+l+2} p^{-k-2} \Omega_\frp^{a} E_{k+2,l,\varphi^{\sigma_\frp}} (E, \omega, \levelstructure).
\end{gathered}$$ Applying the above calculation to the case $a = k+l+2$ and $n= k+2$, our assertion now follows from Theorem \[theorem: Katz measure\], noting the definition of the $p$-adic Eisenstein series , the definition of $\varphi_a$, and the fact that the sum over all $\sigma \in G_{F/K}$ of $\varphi_a^\sigma$ is invariant by the action of $\sigma_\frp$.
\[pro: measure\] Let $\bbZ_p^\times \times \bbZ_p^\times \rightarrow \bbZ_p^\times$ be the surjection defined by $\rho: (x,y) \mapsto x/y$. We define the measure $\dmud^a$ on $\bbZ^\times_p$ by $$\int_{\bbZ_p^\times} f(w) \dmud^a(w) =
\frac{(-1)^a}{G(\varepsilon^a)} \int_{\bbZ^\times_p \times \bbZ_p^\times} x^{-1} y^a \rho^*(f)(x, y) \dmu_{\varphi_a}(x,y).$$ If $\frf_a \neq (1)$, then this measure satisfies the interpolation property $$\frac{1}{\Omega^a_\frp} \int_{\bbZ_p^\times} w^n \dmud^a(w) = %\frac{L_p(\psi^a \otimes \chi_\cyc^n) }{\Omega_\frp^a}=
\left(1 - \frac{\psi(\frp)^a}{p^n} \right)
\left(1 - \frac{\ol\psi(\frp^{*})^a}{p^{a+1-n}} \right)
\frac{\ol\Omega^a \Gamma(n) L(\psi^a, n)}{A^{a-n} |\Omega|^{2n} }.$$ for any integers $a$ and $n$ such that $a \geq n > 0$. A similar formula holds for the case when $\frf_a = (1)$, but with multiplication by $(N^{a+2}/N^{2n}-1)$ on the right hand side.
The equality is obtained from the definition of $\dmud^a$ and in calculating the restriction of the measure $\dmu_{\varphi_a}$ to $\bbZ^\times_p \times \bbZ^\times_p$. The calculation directly follows from Katz [@Ka1] 8.7.6, using the functional equation (see Remark \[rem: functional equation\] below.) One may also do the calculation using an alternative construction of Katz $p$-adic measure ([@BKo] Proposition 3.5 and Theorem 3.7), again after using the functional equation.
\[rem: functional equation\] Combining Proposition \[pro: L-value\] and (or if $\frf_a=(1)$, then Proposition \[pro: L-value-2\] and ), we obtain the functional equation $$%\begin{multline*}
\frac{\Gamma(n) L(\psi^a,n)}{A^{a-n} |\Omega|^{2n} }
= \frac{ N(\frf_a)^{a+1-n} \Gamma(a+1-n) L(\ol\psi^a, a+1-n) }{(-1)^aG(\varepsilon^a)\ol f_a^aA^{n-1}|\Omega|^{2(a+1-n)}}.$$ We regard $f_a$ and $\ol f_a$ in $\cO_K$ as elements in $\bbZ_p^\times$ through the canonical isomorphism $\cO_{K_\frp} \cong \bbZ_p$. Denote by $\dwtmu_{\varphi_a}$ the measure on $\bbZ_p^\times \times \bbZ_p^\times$ obtained as the pull-back of $f_a^{-1} \dmu_{\varphi_a}$ through the isomorphism $$\bbZ_p^\times \times \bbZ_p^\times \xrightarrow\cong \bbZ_p^\times \times \bbZ_p^\times$$ given by $(x,y) \mapsto (\ol f_a x,f_a^{-1} y)$. Since $A = |\Omega|^2 \sqrt{d_K}/2 \thepi$, if we let $k_1 = a+1-n$ and $k_2 = 1-n$, then the interpolation property of $\dwtmu_{\varphi_a}$ at $(E, \omega, \levelstructure)$ becomes $$\begin{gathered}
\frac{1}{\Omega_\frp^{k_1-k_2}} \int_{\bbZ_p^\times \times \bbZ_p^\times} x^{-k_2} y^{k_1-1}
\dwtmu_{\varphi_a}(x,y) \\= %E^{(p)}_{n,a-n,\varphi_a}
\left(1 - \frac{\psi(\frp)^{k_1-k_2}}{p^{1-k_2}} \right)
\left(1 - \frac{\ol{\psi}(\frp^{*})^{k_1-k_2}}{p^{k_1}} \right) \left( \frac{\sqrt{d_K}}{2\thepi}\right)^{k_2}
\frac{\Gamma(k_1) L(\ol\psi^{{k_1-k_2}}, k_1)}{\Omega^{k_1-k_2}}
\end{gathered}$$ for $k_1 > -k_2 \geq 0$. This coincides with the interpolation property of the two-variable $p$-adic measure constructed by Katz and Yager (see [@Ya] §1.)
If $\frf_a \neq (1)$, then the measure $\dmud^a$ defined in Proposition \[pro: measure\] satisfies the condition of Conjecture \[conj: PR\]. If $\frf_a = (1)$, then we need to cancel the factor $(N^{a+2}/N^{2n}-1)$ which appears in the interpolation formula.
\[def: measure\] We define the pseudo-measure $\dmu_a$ on $\bbZ_p^\times$ as follows.
1. If $\frf_a \neq (1)$, then we let $\dmu^a := \dmud^a$.
2. If $\frf_a = (1)$, then we let $\dmu^a_N$ be the measure on $\bbZ_p^\times$ defined by $$\int_{\bbZ_p^\times} x^n \dmu^a_N = \left(\frac{N^{a+2}}{N^{2n}}-1 \right)$$ for any integer $n$. We define $\dmu^a$ to be the pseudo-measure on $\bbZ_p^\times$ obtained as the quotient of $\dmud^a$ by $\dmu^a_N$ (see for example [@Col] §1.2 for the definition of a pseudo-measure).
When $\frf_a \neq (1)$, then $\dmu_a$ is by definition a $p$-adic measure on $\bbZ_p^\times$.
We now have the following.
\[thm: main\] Let $a>0$ be an integer and let $\dmu^a$ be the pseudo-measure on $\bbZ_p^\times$ defined in Definition \[def: measure\]. If we let $$%\begin{equation}\label{eq: p-adic L}
L_p(\psi^a \otimes \chi_\cyc^n) := \int_{\bbZ_p^\times} w^n \dmu^a(w),$$ then we have $$\frac{L_p(\psi^a \otimes \chi_\cyc^n)}{\Omega_p(n)}=
\left(1 - \frac{\psi(\frp)^a}{p^n} \right) \left(1 - \frac{\ol\psi(\frp^{*})^a}{p^{a+1-n}} \right) \frac{\Gamma(n) L(\psi^a,n)}{\Omega_\infty(n)}$$ for any integer $n > a$ such that the $p$-adic regulator map $r_p$ is injective.
We first consider the case when $\frf_a\neq(1)$. By definition of the $p$-adic Eisenstein series in Definition \[def: p Eisenstein\], the moments of the measure $\dmu_{\varphi_a}$ constructed in Theorem \[theorem: Katz measure\] is given by $p$-adic Eisenstein series. As in , let $E_{n,a-n,\varphi_a}(E, \omega, \levelstructure)$ be the element in $\wh K_\frp^\ur$ satisfying $$\left( 1 - p^{a-n} \sigma^* \right)E_{n, a-n,\varphi_a}(E, \omega, \levelstructure) = E^{(p)}_{n, a-n,\varphi_a}(E, \omega, \levelstructure).$$ Then the compatibility between the Frobenius on the modular curve and a point, as well as the restriction of the measure on $\bbZ_p \times \bbZ_p^\times$ to $\bbZ_p^\times\times\bbZ_p^\times$ shows that we have the relation $$\begin{gathered}
%\label{eq: formula two}
\frac{1}{\Omega^a_\frp} \int_{\bbZ_p^\times} w^n \dmu^a(w)
\\ = \frac{1}{G(\varepsilon^a)}
\left(1 - \frac{\psi(\frp)^a}{p^n} \right) \left(1 - \frac{\ol\psi(\frp^{*})^a}{p^{a+1-n}} \right) E_{n,a-n,\varphi_a}(E, \omega, \levelstructure).
\end{gathered}$$ The calculation of the $p$-adic period in Theorem \[theorem: p period\] shows that $$\Omega_p(n) =\frac{ (-1)^{a-n+1}}{\Gamma(n)} \Omega^a_\frp E_{n,a-n,\varphi_a}(E, \omega, \levelstructure)$$ for our fixed embedding $K \hookrightarrow \bbC_p$. This proves in particular that $$\frac{(-1)^{a-n+1}}{\Omega_p(n)} \int_{\bbZ_p^\times} w^n \dmu^a(w) =
\left(1 - \frac{\psi(\frp)^a}{p^n} \right) \left(1 - \frac{\ol\psi(\frp^{*})^a}{p^{a+1-n}} \right) \frac{\Gamma(n)}{G(\varepsilon^a)}.$$ Our assertion now follows from Corollary \[cor: Deninger\]. The case for $\frf_a=(1)$ follows in a similar fashion, noting the interpolation property of $\dmud^a$ and $\dmu^a_N$.
[999]{} , $p$-adic elliptic polylogarithm, $p$-adic Eisenstein series and Katz measure, arXiv:0707.3747v1 \[math.NT\], to appear in the American Math. Journal. , in preparation. , Algebraic theta functions and $p$-adic interpolation of Eisenstein-Kronecker numbers, Duke Math. J. **153** no. 2 (2010), 229-295. , On the de Rham and $p$-adic realizations of the elliptic polylogarithm for CM elliptic curves, Annales scientifiques de l’ENS **43**, fascicule 2 (2010), 185-234. , Syntomic regulators and $p$-adic integration I: Rigid syntomic regulators, Israel Journal of Math. **120** (2000), 291-334. , The Elliptic Polylogarithm, *Motives*, (Seattle, WA, 1991), 123-192. , Fonctions $L$ $p$-adiques (Exposé 851), Astérisque **266** (2000), 21-58. C. Deninger, Higher regulators and Hecke $L$-series of imaginary quadratic fields I. Invent. math. **96**, 1–69 (1989). , Extensions of motives associated to symmetric powers of elliptic curves and to Hecke characters of imaginary quadratic fields, In: F. Catanese (ed.) *Proc. Arithmetic Geometry, Cortona 1994*, Symp. Math. **37**, 99-137, Cambridge Univ. Press (1997). C. Deninger and K. Wingberg, On the Beilinson conjectures for elliptic curves with complex multiplication, In: Rapoport, M., Schappacher, N., Schneider, P. (ed.) *Beilinson’s conjectures on special values of $L$-functions*, Perspect. Math. vol. **4**, 249–272, Academic Press, (1988). , Degeneration of $l$-adic Eisenstein classes and of the elliptic polylog, Invent. Math. **135** (1999), no. 3, 545–594. , On the $l$-adic cohomology of varieties over number fields and its Galois cohomology, In: *Galois groups over $\bbQ$*, Springer (1989), 315–360. , $p$-adic Hodge theory and values of zeta functions of modular forms, In: *Cohomologies $p$-adiques et applications arithmétiques (III)*, Astérisque **295** (2004), 117–290. , $p$-adic interpolation of real analytic Eisenstein series, Ann. of Math. (2) **104** (1976), no. 3, 459–571. , *Elliptic Functions*, Second Edition, GTM **112**, Springer 1987. , *Fonctions $L$ $p$-adiques des représentations $p$-adiques*, Astérisque **229**, (1995). , [$p$-adic Hecke series for quadratic imaginary fields]{}, Math. Sbornik, V, **95** (137), No. 3 (11), 1974. , [*Elliptic Functions according to Eisenstein and Kronecker*]{}, Springer-Verlag, 1976. , [On two variable $p$-adic $L$-functions]{}, Ann. of Math. **115** (1982), 411-449.
[^1]: This work was supported by KAKENHI 21674001.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The dynamics of matter-wave solitons in Bose-Einstein condensates (BEC) is considerably affected by the presence of a surrounding thermal cloud and by condensate depletion during its evolution. We analyze these aspects of BEC soliton dynamics, using time-dependent Hartree-Fock-Bogoliubov (TDHFB) theory. The condensate is initially prepared within a harmonic trap at finite temperature, and solitonic behavior is studied by subsequently propagating the TDHFB equations without confinement. Numerical results demonstrate the collapse of the BEC via collisional emission of atom pairs into the thermal cloud, resulting in splitting of the initial density into two solitonic structures with opposite momentum. Each one of these solitary matter waves is a mixture of condensed and noncondensed particles, constituting an analog of optical random-phase solitons.'
author:
- 'H. Buljan,$^{1,2}$ M. Segev,$^{1}$ and A. Vardi$^{3}$'
title: 'Incoherent matter-wave solitons'
---
The physics of quantum-degenerate, interacting bose gases closely resembles the behavior of light in nonlinear media. The dynamics of Bose-Einstein condensate (BEC) at zero-temperature is well-described by the Gross-Pitaevskii (GP) mean-field theory, given by the nonlinear Schrödinger equation (NLSE) for the condensate order parameter. The same equation describes the evolution of coherent light in nonlinear Kerr media. This analogy has opened the way for the field of nonlinear atom optics [@Lens; @Rolston] with striking demonstrations of familiar nonlinear optics phenomena such as four wave mixing [@Deng], superradiant Rayleigh scattering [@Inouye1], and matter-wave amplification [@Kozuma; @Inouye2], carried out with matter-waves. One such phenomenon is the formation of matter-wave solitons [@Ruprecht; @PerezG; @Burger; @Denschlag; @Busch; @Smerzi; @Khajkovic; @Strecker; @Salasnich; @Eiermann]. Experimentally, dark solitons [@Burger; @Denschlag] and bright gap solitons [@Eiermann] were observed in BECs with repulsive interactions, whereas bright solitons [@Khajkovic; @Strecker] were demonstrated in systems with attractive interactions. These experimental results are augmented by extensive theoretical work including predictions on bright [@Ruprecht; @PerezG] and dark [@Busch] matter-wave solitons, lattice solitons [@Smerzi], and soliton trains [@Salasnich]. The vast majority of previous theoretical efforts on matter-wave solitons have utilized the zero-temperature GP mean-field theory. However, in a realistic system, elementary excitations arising from thermal and/or quantum fluctuations are always present [@Griffin], and the BEC dynamics may be considerably affected by the motion of excited atoms around it (thermal cloud), and by its dynamical depletion [@Castin], giving rise to new nonlinear matter-wave phenomena.
Here we analyze these aspects of BEC soliton dynamics by using the time-dependent Hartree-Fock-Bogoliubov (TDHFB) theory [@Proukakis1; @Holland; @Rey]. Soliton dynamics is analyzed by first calculating the finite-temperature ground state of the attractively interacting gas within an harmonic trap [@Griffin]. The harmonic confinement is then suddenly turned off, and the partially condensed Bose gas starts to dynamically evolve. Within the TDHFB model, we find a characteristic pattern of evolution of the system whereby pairs of atoms are collisionally excited from the BEC into the thermal cloud causing the initial density to eventually split into two solitonic structures with opposite momentum. Both solitons constitute a mixture of condensed and noncondensed particles. We emphasize that the observed composite waves are a truly novel type of matter-wave solitons, where localization is attained not only in spatial density, but also in spatial correlations. This type of solitons are reminiscent of composite incoherent optical solitons [@Mitchell; @Equ; @Mitchell1], thus highlighting an analogy between incoherent light behavior in nonlinear media and BECs at finite-temperatures.
Starting with a near-unity condensate fraction at very low temperatures, the GP dynamics reproduces, under proper conditions, the well-known zero-temperature BEC solitons, thus demonstrating the condensate’s mechanical stability. However, the evolution of the same initial nearly pure BEC with TDHFB clearly illustrates BEC depletion through pairing, causing these coherent solitons to disintegrate in a characteristic fashion into incoherent solitary matter waves. In all cases, when both the trap and the interparticle interactions are turned off simultaneously, we observe fast matter-wave dispersion.
We consider a system of $N$ interacting bosons placed in a quasi one-dimensional (Q1D) cigar-shaped harmonic potential $V_{ext}(x,y,z)=(\omega_x x^2+\omega_{\perp} y^2+\omega_{\perp} z^2)/2$, where $\omega_{\perp}\gg\omega_x$ denote the transverse and the longitudinal frequencies of the trap, respectively. The interparticle interaction is approximated by the Q1D contact potential $V(x_1-x_2)=g_{1D}\delta(x_1-x_2)$, where $g_{1D}=-2\hbar^2/ma_{1D}$, $a_{1D}\approx -a_{\perp}^2/a_{3D}$ is the effective 1D scattering length [@Dunjko; @Kheruntsyan1D; @Moritz], $m$ is the particle mass, $a_{\perp}=\sqrt{\hbar/m\omega_{\perp}}$ is the size of the lowest transverse mode, while $a_{3D}$ is the 3D scattering length. At finite temperatures, the equilibrium state of the system can be described by the HFB theory [@Griffin]:
$$H_{sp} \Phi^{s}+
g_{1D}[n_c^{s}(x)+2\tilde n^{s}(x)]\Phi^{s}
+g_{1D} \tilde m^{s}(x)\Phi^{s*}
=\mu \Phi^{s}(x),
\label{statcon}$$
$$\left [
\begin{array}{c c}
\mathcal{L}^{s}(x) &
\mathcal{M}^{s}(x) \\
-\mathcal{M}^{s^*}(x) &
-\mathcal{L}^{s}(x)
\end{array}
\right ]
\left [
\begin{array}{c}
u_j^{s}(x) \\
v_j^{s}(x)
\end{array}
\right ]
=E_j
\left [
\begin{array}{c}
u_j^{s}(x) \\
v_j^{s}(x)
\end{array}
\right ].
\label{statuv}$$
Here, $H_{sp}=-\frac{\hbar^2}{2m}\frac{\partial^2}{\partial x^2}
+\frac{1}{2}m\omega_x^2 x^2$ and $\mu$ is the chemical potential. The superscript $s$ denotes the static HFB calculation, e.g. the static order parameter is denoted by $\Phi^{s}(x)$ and $n_c^{s}(x)=|\Phi^{s}(x)|^2$ denotes the static HFB condensate density [@quasicondensate]. The normal density is $\tilde n^{s}(x)=\sum_j {|u_j^{s}(x)|^2 N_j
+ |v_j^{s}(x)|^2 (N_j+1)}$ , whereas $\tilde m^{s}(x)=-\sum_j
u_j^{s}(x)v_j^{s*}(x) (2N_j+1)$ is the anomalous density [@Griffin]. The population of excited states at temperature $T$ follows the Bose distribution $N_j=(e^{E_j/kT}-1)^{-1}$. In Eq. (\[statuv\]) we denote $\mathcal{L}^{s}(x)=H_{sp}+2g_{1D}[n_c^{s}(x)+\tilde n^{s}(x)]-\mu$ and $\mathcal{M}^{s}(x)=-g_{1D}[\Phi^{s2}(x)+\tilde m^{s}(x)]$. Two variants of the HFB formalism are the Hartree-Fock (HF) approach, altogether neglecting anomalous terms, and the HFB-Popov approximation wherein the [*noncondensate*]{} anomalous density is explicitly dropped, i.e. $\mathcal{M}^{s}(x)\approx -g_{1D}\Phi^{s2}(x)$ [@Griffin]. Unlike the static HFB, the HF and HFB-Popov approximation do not have an unphysical gap in their excitation spectra [@Griffin].
We use the solution of Eqs. (\[statcon\]) and (\[statuv\]) as the initial conditions to study dynamics without confinement within the TDHFB approximation in the modal form [@Rey]
$$i\hbar \frac{\partial \Phi(x,t)}{\partial t}=
H_{sp} \Phi+g_{1D}[n_c(x,t)+2\tilde n(x,t)]\Phi
+g_{1D} \tilde m(x,t)\Phi^*,
\label{con}$$
$$i\hbar\frac{\partial}{\partial t}
\left [
\begin{array}{c}
u_j(x,t) \\
v_j(x,t)
\end{array}
\right ]
=
\left [
\begin{array}{c c}
\mathcal{L}(x,t) &
\mathcal{M}(x,t) \\
-\mathcal{M}^{*}(x,t) &
-\mathcal{L}(x,t)
\end{array}
\right ]
\left [
\begin{array}{c}
u_j \\
v_j
\end{array}
\right ],
\label{dynuv}$$
where $\mathcal{L}(x,t)=H_{sp}+2g_{1D}[n_c(x,t)+\tilde n(x,t)]$, $\mathcal{M}(x,t)=-g_{1D}[\Phi^{2}(x,t)+\tilde m(x,t)]$, $\tilde n(x,t)=\sum_j {|u_j|^2 N_j + |v_j|^2 (N_j+1)}$ is the normal density, and $\tilde m(x,t)=-\sum_j
u_jv_j^{*}(2N_j+1)$ is the anomalous density; at $t=0$, all dynamical quantities are identical to their static counterparts, e.g., $\Phi(x,0)=\Phi^{s}(x)$ etc. The TDHFB model is usually given in the form of coupled equations for the condensate order parameter and the single particle density matrix (e.g. see Ref. [@Holland]). A tedious but straightforward calculation [@Derivation] shows its full equivalence to the modal form (\[dynuv\]). As expected, when the time-dependence of the BEC and qusiparticle functions is $\Phi(x,t)=\Phi^{s}(x)\exp(-i\mu t/\hbar)$, $u_j(x,t)=u_j^{(s)}(x)\exp(-i(E_j+\mu)t/\hbar)$, and $v_j(x,t)=v_j^{(s)}(x)\exp(-i(E_j-\mu)t/\hbar)$, the equations of motion (\[con\]) and (\[dynuv\]) reduce to the time-independent HFB equations (\[statcon\]) and (\[statuv\]).
In what follows we present numerical results based on the described formalism, demonstrating the effect of thermal particles and condensate depletion on matter-wave soliton dynamics. The parameters of the calculation are chosen to resemble the experimental parameters of Ref. [@Khajkovic]. We consider $N=2.2\ 10^4$ $^{7}$Li atoms in a harmonic trap with $\omega_{\perp}=4907$ Hz ($a_{\perp}=\sqrt{\hbar/m\omega_{\perp}}\approx 1.35\ \mu$m). The 3D scattering length $a_{3D}=-3.1\ 10^{-11}$ m corresponds to a nonlinear parameter of $N|a_{3D}|\approx 0.68 \ \mu$m, and is tunable by the Feshbach resonance technique [@Khajkovic]. We present two calculations for different $T$ and $\omega_x$ to show the influence of excitations on the soliton dynamics.
![(color online) (a) Total density (solid line), and condensate density (circles) at $t=0$; total (dot-dashed line) and condensate density (dashed line) at $t\omega_\perp=180$. (b) The dipole-shaped first excited state, $u_1^s(x)$ (solid line) and $v_1^s(x)$ (dashed line). (c) Evolution of the density with interactions present, and (d) without interactions. (e) The complex degree of coherence $\mu(x,x',t)$ of a partially-coherent matter-wave at $t=0$ (dot-dashed line), and $t\omega_\perp=180$ \[$Re\ \mu(x,x',t)$ solid line, and $Im\ \mu(x,x',t)$ dashed line\]; $x'$ is placed at the position of the left peak. (f) The condensate fraction (solid line) and the total population within the first 60 excited states (dashed line) during evolution. []{data-label="fig1"}](imws_hfb_fig1s.ps){width="50.00000%"}
First we consider the gas at higher temperature, where the condensate fraction is approximately $25$%. The trapping frequency is $\omega_{\perp}=141$ Hz ($a_{x}=\sqrt{\hbar/m\omega_{x}}\approx 8.0\ \mu$m), while $k_BT/\hbar \omega_{\perp}=60$. Fig. \[fig1\](a) illustrates the total density $n_c^{s}(x)+n_t^{s}(x)$ of the stationary HFB-Popov calculation. The total density profile is double humped, which is a clear signature of the significant population of the first excited state (approximately $\approx 10$% of atoms), which has a dipole-like shape shown in Fig. \[fig1\](b). We have numerically checked the stability [@Kagan1] of the solution with respect to small perturbations; the stability is underpinned by the use of parameters resembling experiment [@Khajkovic]. When the trap is turned off, the system is suddenly taken out of equilibrium, and consequently starts to evolve; we simulate the dynamics with the full TDHFB model. In the spirit of Ref. [@Khajkovic], we compare the $x$-unconfined dynamics of the system in the presence of interparticle interactions \[Fig. \[fig1\](c)\] to its time evolution when both the confinement in $x$ and the interactions are turned off \[Fig. \[fig1\](d)\]. In the absence of interactions, we clearly observe fast matter-wave dispersion. In contrast, when interactions are present, the two humps begin to separate, because the trapping potential which provided a balance to the kinetic energy term is no longer present \[Fig. \[fig1\](c)\]. During evolution, the condensate is slowly depleted \[Fig. \[fig1\](f)\]. Consequently, at one point during the evolution the two humps split, each forming a solitonic structure with opposite momentum. Each of these solitonic structures contains both Bose-condensed atoms, and a significant portion of the non-condensed particles, i.e., they are partially coherent matter-wave solitons. The uniqueness of such random-phase structures is ellucidated by their complex degree of coherence, $\mu(x,x',t)=\rho(x,x',t)/\sqrt{\rho(x,x,t)\rho(x',x',t)}$, where $\rho(x,x',t)=\Phi(x',t)^{*}\Phi(x,t)+
\langle\hat \Psi^{\dagger}(x',t)\hat \Psi(x,t)\rangle$, plotted at different times in Fig. \[fig1\](e). We observe that spatial correlation is localized in space at $t=0$. Correlations change as the two humps split. While retaining spatial localization, the phases at the two separated peaks are well-correlated, and out of phase, resembling the behavior of out-of-phase adjacent solitons from Ref. [@Strecker]. We emphasize that for zero-temperature GPE solitons, the pair correlation function factorizes as $\rho(x,x')=\Phi^*(x)\Phi(x')$, which yields $\mu(x,x')=1$, corresponding to coherent matter-waves. Our incoherent matter-wave solitons are thus rather special in that they correspond to localization of [*entropy*]{} and spatial correlation, as well as to localization of density.
While the temperature $k_BT$ in the previous example is higher than the transverse level spacing $\hbar \omega_{\perp}$ whereas a ’true’ 1D geometry calls for $k_BT<\hbar \omega_{\perp}$ [@Moritz], the use of a Q1D formalism is still justified because the first $\omega_x/\omega_{\perp}\sim 35$ states are essentially 1D (they are in the lowest state of the transverse Hamiltonian). Furthermore, only condensed atoms, and atoms from lower excited states determine the outcome of the motion. Therefore, a proper inclusion of the transverse dimension in the calculation would lead to some rescaling of the parameters, but would not influence the dynamics observed in our quasi-1D calculation. Moreover, the simulations as well as the experiment of [@Khajkovic], are all in the weak interaction regime $N|a_{1D}|/a_x\sim 10^8\gg 1$ [@Dunjko], thus justifying the use of a mean-field approach.
Next, we consider a gas prepared at near-zero temperature, where the condensate fraction is $99$%. The trapping frequency is $\omega_{\perp}=439$ Hz ($a_{x}=\sqrt{\hbar/m\omega_{x}}\approx 4.51\ \mu$m), while $k_BT/\hbar \omega_{\perp}=5$. As previously, the trap is turned off and the dynamics is calculated using both the GP equation (Fig. \[fig2\]a) and the TDHFB formalism (Fig. \[fig2\]b). The mechanical stability of the evolving BEC is demonstrated by propagating the time-dependent GP equation, showing small oscillations of the condensate width, rather than a mechanical collapse to a point \[Fig. \[fig2\](a)\]. This motion corresponds to a coherent matter wave soliton. However, as clearly evident from propagation of the TDHFB equations \[Fig. \[fig2\](b)\], allowing BEC depletion \[Fig. \[fig2\](c)\], the BEC collapses via a [*pairing*]{} instability [@Imry; @Jeon] whereby pairs of atoms are collisionally pulled out of the BEC into the thermal cloud, thus gaining a mean-field energy which goes into their relative motion. Such pairing collapse with little or no mechanical shrinking is indeed observed in 3D collapse experiments [@Donley; @Milstein]. Our results show that in 1D it results in two separate solitonic structures of opposite momentum, containing both a condensed part and a significant thermal population. We note that similar structures were observed in stochastic simulations of molecular BEC dissociation in 1D geometry [@Kheruntsyan], indicating that incoherent matter-wave solitons can also be produced in this system. Figure \[fig2\](d) shows that the correlations of each solitonic structure are localized, and the two structures are partially correlated.
Before closing, we note that within the time-dependent Hartree-Fock (TDHF) approximation [@Proukakis1], where condensate depletion is not accounted for during the dynamics, we find solutions with double-humped density profile that begin to split as in TDHFB model. However, as the condensate is not depleted within TDHF, the two peaks may be pulled back and merge to almost recover the initial density profile. Such motion is also characteristic of composite incoherent solitons in optics.
In conclusion, we have used the time-dependent Hartree-Fock-Bogoliubov theory to analyze the influence of the thermal cloud and condensate depletion onto the dynamics of BEC solitons. We find that condensate depletion induced by pairing, and the presence of a thermal cloud cause the particle density to split into two solitonic structures, each being a mixture of condensed and non-condensed particles. The predicted incoherent matter-wave structures represent novel correlation solitons which resemble localized second-sound entropy waves. Such random-phase matter-wave solitons correspond to incoherent solitons in nonlinear optics [@Mitchell; @Equ; @Mitchell1], which points at the analogy between partially condensed Bose gases and nonlinear partially coherent optical waves.
[99]{}
G. Lens, P. Meystre, and E.M. Wright, Phys. Rev. Lett. [**71**]{}, 3271 (1993).
S. L. Rolston and W. D. Phillips, Nature [**416**]{}, 219 (2002).
L. Deng [*et al.*]{}, Nature [**398**]{}, 218 (1999).
S. Inouye [*et al.*]{}, Science [**285**]{}, 571 (1999).
M. Kozuma [*et al.*]{}, Science [**286**]{}, 2309 (1999).
S. Inouye [*et al.*]{}, Nature [**402**]{}, 641 (1999).
P.A. Ruprecht [*et al.*]{}, Phys. Rev. A [**51**]{}, 4704 (1995).
V.M. Perez-Garcia, H. Michinel, and H. Herrero, Phys. Rev. A [**57**]{}, 3837 (1998).
S. Burger [*et al.*]{}, Phys. Rev. Lett. [**83**]{}, 5198 (1999).
J. Denschlag, [*et al.*]{}, Science [**287**]{}, 97 (2000).
Th. Busch, and J.R. Anglin, Phys. Rev. Lett. [**84**]{}, 2298 (2000).
A. Trombettoni and A. Smerzi, Phys. Rev. Lett. [**86**]{}, 2353 (2001).
L. Khaykovich [*et al.*]{}, Science [**296**]{}, 1290 (2002).
K.E. Strecker, [*et al.*]{}, Nature [**417**]{}, 150 (2002).
L. Salasnich, A. Parola, and L. Reatto, Phys. Rev. Lett. [**91**]{}, 080405 (2003).
B. Eiermann [*et al.*]{}, Phys. Rev. Lett. [**92**]{}, 230401 (2004).
A. Griffin, Phys. Rev. B [**53**]{}, 9341 (1996).
Y. Castin, and R. Dum, Phys. Rev. Lett. [**79**]{} 3553 (1997).
N.P. Proukakis and K. Burnett, J. Res. Natl. Inst. Stand. Technol. [**101**]{} 457 (1996); N.P. Proukakis, K. Burnett, and H.T.C. Stoof, Phys. Rev. A [**57**]{}, 1230 (1998).
M. Holland, J. Park, and R. Walser, Phys. Rev. Lett. [**86**]{}, 1915 (2001).
A.M. Rey [*et al.*]{}, Phys. Rev. A [**69**]{} 033610 (2004); Appendix A of that paper contains the modal form of TDHFB in the discrete Bose-Hubbard model.
M. Mitchell [*et al.*]{}, Phys. Rev. Lett. [**77**]{}, 490 (1996); M. Mitchell and M. Segev, Nature (London) [**387**]{}, 880 (1997).
D.N. Christodoulides [*et al.*]{}, Phys. Rev. E [**63**]{}, 035601 (2001).
M. Mitchell [*et al.*]{}, Phys. Rev. Lett. [**79**]{}, 4990 (1997).
V. Dunjko, V. Lorent, and M. Olshanii, Phys. Rev. Lett. [**86**]{}, 5413 (2001).
K.V. Kheruntsyan [*et al.*]{}, Phys. Rev. Lett. [**91**]{}, 040403 (2003).
H. Moritz [*et al.*]{}, Phys. Rev. Lett. [**91**]{}, 250402 (2003).
While strictly speaking the 1D critical temperature is $T_c=0$, there is significant increase in the population of the lowest energy state below $k_BT_c=\hbar\omega_xN/\ln(2N)$, see W. Ketterle, and N.J. van Drutten, Phys. Rev. [**54**]{}, 656 (1996).
H. Buljan [*et al*]{}, unpublished.
Yu. Kagan, A.E. Muryshev, and G. Schlyapnikov, Phys. Rev. Lett. [**81**]{}, 933 (1998).
W.A.B. Evans and Y. Imry, Nuovo Cimento B [**63**]{} 155 (1969).
G.S. Jeon, L. Yin,1 S.W. Rhee, and D.J. Thouless, Phys. Rev. A [**66**]{} 011603 (2002).
E. A. Donley [*et al*]{}, Nature (London) [**412**]{}, 295 (2001).
J. N. Milstein, C. Menotti, and M. J. Holland, New J. Phys. [**5**]{}, 52 (2003).
K. V. Kheruntsyan and P. D. Drummond, Phys. Rev. A [**66**]{}, 031602 (2002).
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'A. Ecuvillon, G. Israelian, N. C. Santos, M. Mayor, V. Villar'
- 'G. Bihain'
date: 'Received 22 April 2004 / Accepted 17 June 2004'
title: 'C, S, Zn and Cu abundances in planet-harbouring stars[^1]'
---
Introduction
============
Since the first success in the search for exoplanets with the discovery of the planet orbiting 51 Peg (Mayor & Queloz [@May95]), studies on the formation and evolution of planetary systems have increased. Several spectroscopic analyses of stars with planets have recently been carried out. Most of them used iron as the reference element (Gonzalez [@Gonz97]; Murray & Chaboyer [@Mur02]; Laws et al. [@Law03]; Santos et al. [@San01], [@San03b], [@San04a]; for a review see Santos et al. [@San03a]), while only a few studies determined abundance trends for other metals (Gonzalez & Laws [@Gonz00]; Gonzalez et al. [@Gonz01]; Santos et al. [@San00]; Bodaghee et al. [@Bod03]; Takeda et al.[@Tak01]; Sadakane et al. [@Sad02]; for a review see Israelian [@Isr03b]).
One of the most remarkable results is that planet-harbouring stars are on average more metal-rich than solar-type disc stars (Gonzalez [@Gonz97]; Gonzalez et al. [@Gonz01]; Laws et al. [@Law03]; Santos et al. [@San01], [@San03b], [@San04a]). Two main explanations have been suggested for linking this metallicity excess with the presence of planets. The first of these, the “self-enrichement” hypothesis, attributes the origin of the observed overabundance of metals to the accretion of large amounts of metal-rich H- and He-depleted rocky planetesimal materials on to the star (Gonzalez [@Gonz97]). The opposite view, the “primordial” hypothesis, considers the metallicity enhancement to be caused by the high metal content of the protoplanetary cloud from which the planetary system formed (Santos et al. [@San00], [@San01]). Discriminating between these two possibilities will help in understanding how planetary systems form.
Light elements may give fundamental information about the mixing, diffusion and angular momentum history of exoplanets hosts, as well as stellar activity caused by interaction with exoplanets (Santos et al. [@San02], [@San04b]; Israelian et al. [@Isr04]). Studies of Be, Li and the isotopic ratio / could give evidences to distinguish between different planet formation theories (Sandquist et al. [@Sand02]).
Murray & Chaboyer ([@Mur02]) concluded that an accretion of 5.0 $M_\oplus$ of iron while on the main sequence would explain the overabundance of metals found in planet host stars. On the other hand, Pinsonneault et al. ([@Pin01]) rejected the “self-enrichement” scenario since the metallicity enhancement does not show the expected $T_\mathrm{eff}$ dependence; recent results by Vauclair ([@Vau04]) may have refuted, however, this argument. Furthermore, several studies have shown that planetary frequency is rising as a function of \[Fe/H\] (Santos et al. [@San01], [@San04a]; Reid [@Reid02]), result which seems to support the primordial metallicity enhancement scenario. However, evidences of pollution have been found for a few cases (Israelian et al. [@Isr01], [@Isr03a]; Laws & Gonzalez [@Law01]).
{height="6cm"} {height="6cm"} {height="6cm"} {height="6cm"}
{height="6cm"} {height="6cm"} {height="6cm"} {height="6cm"}
{height="6cm"} {height="6cm"} {height="6cm"} {height="6cm"}
Abundance trends of different elements may provide important clues to the question. Volatile elements (with low condensation temperature $T_{\rm C}$) are expected to be deficient in accreted materials because of the high temperatures near the star. Therefore, the “self-enrichement” scenario should lead to a relative overabundance of refractories, such as the $\alpha$-elements Si, Mg, Ca, Ti and the iron-group elements compared to volatiles, such as CNO, S and Zn. Smith et al. ([@Smi01]) found a small subset of planet host stars which bore this accretion signature since these stars exhibited an accentuated trend of increasing \[$X$/H\] with increasing $T_{\rm C}$ for a given element $X$. Nevertheless, subsequent studies have obtained a similar behaviour in volatiles as in refractory elements (Takeda et al. [@Tak01]; Sadakane et al. [@Sad02]).
Other evidence in favour of a “primordial” high metallicity has been provided by several studies, which conclude that the abundances of volatiles in planet host stars do not reveal significant differences from those in field dwarfs of the same metallicity (Santos et al. [@San00]; Gonzalez et al. [@Gonz01]; Takeda et al. [@Tak01]; Sadakane et al. [@Sad02]). However, all these authors used inhomogeneous comparison samples of field dwarfs from the literature, which could be a source of systematic errors. A sample of stars with no known planets has been prepared by Santos et al. ([@San01]) and analysed in the same way as the planet host set in several studies of the stellar and kinematic parameters of stars with planets (Santos et al. [@San03b]) and about their metal-rich nature (Santos et al. [@San01], [@San04a]). Other uniform and unbiased comparisons have been carried out for abundances of elements other than iron – some $\alpha$- and Fe-group elements (Bodaghee et al. [@Bod03]) and the volatile nitrogen (Ecuvillon et al. [@Ecu04]) – and conclude that the abundance trends observed in planet host stars are almost identical to those in the comparison sample.
Sulphur is a volatile $\alpha$ element which behaves as true primary; it is well established that it is produced mainly by Type II supernovae (SNe II) of massive stars (e.g. Timmes, Woosley & Weaver [@Tim95]; Carretta, Gratton & Sneden [@Car00]). For Cu and Zn the situation is rather confused; massive (M$>$8M$_\odot$), low- (M$<$4M$_\odot$) and intermediate-mass stars (4M$_\odot$$<$M$<$8M$_\odot$) are likely contributors, although their relative weights are still uncertain (e.g. Luck & Bond [@Luc85]; Sneden et al. [@Sne91]; Mishenina et al. [@Mis02]; Bihain et al. [@Bih04]). A number of different production sites have been suggested for carbon: supernovae, novae, Wolf-Rayet stars, intermediate- and low-mass stars. Recently, Kobulnicky & Skillman ([@Kob98]) concluded that low- and intermediate- mass stars are the significant contributors of carbon. On the other hand, Prantzos et al. ([@Pra94]) and Gustafsson et al. ([@Gus99]) found massive stars to be significant sites for the production of carbon. At the present the situation is not clear.
This work presents a systematic, uniform and detailed study of the volatile elements C, S and Zn and of the refractory Cu in two large samples, a set of planet-harbouring stars and a volume-limited comparison sample of stars with no known planetary mass companions. We studied C, S, Zn and Cu abundances in several optical lines, by measuring [EW]{} in the case of carbon and by the synthesis technique for the other elements. We compare our results with those of other recent studies (Santos et al. [@San00]; Gonzalez et al. [@Gonz01]; Takeda et al. [@Tak01]; Sadakane et al. [@Sad02]).
[lcccr]{} Star & $V$ & Obser. run & S/N & Date\
& 5.0 & UVES & 850 & July 2001\
& 4.1 & UES & 500 & Feb. 2001\
& 6.8 & UES & 800 & Oct. 2001\
& 5.4 & UVES & 950 & July 2001\
& 5.1 & UVES & 1100 & Aug. 2001\
& 3.7 & UES & 500 & Oct. 2001\
& 6.3 & UVES & 1200 & Mar. 2001\
& 6.4 & UVES & 1000 & Mar. 2001\
& 6.5 & UVES & 950 & Mar. 2001\
& 5.7 & UES & 680 & Feb. 2001\
& 7.0 & UVES & 950 & Mar. 2001\
& 7.3 & UVES & 1350 & Mar. 2001\
& 4.5 & UVES & 1100 & Mar. 2001\
& 7.6 & UVES & 900 & Mar. 2001\
& 6.9 & UES & 800 & Oct. 2001\
& 7.7 & UVES & 1100 & June 2001\
& 6.2 & UES & 500 & Oct. 2001\
\[tab1\]
[lccc]{} Species & $\lambda$ (Å) & $\chi_l$ (eV) & $\log{gf}$\
& 5380.340 & 7.68 & $-1.71$\
& 5052.160 & 7.68 & $-1.42$\
& 6743.440 & 7.87 & $-1.27$\
& 6743.531 & 7.87 & $-0.92$\
& 6743.640 & 7.87 & $-0.93$\
& 6757.007 & 7.87 & $-0.81$\
& 6757.177 & 7.87 & $-0.33$\
& 4810.537 & 4.08 & $-0.13$\
& 4722.160 & 4.03 & $-0.37$\
& 5218.204 & 3.82 & $-1.33$\
& 5218.206 & 3.82 & $-0.85$\
& 5218.208 & 3.82 & $-0.98$\
& 5218.210 & 3.82 & $-0.26$\
& 5218.214 & 3.82 & $-0.48$\
& 5218.216 & 3.82 & $-0.48$\
& 5218.219 & 3.82 & $ 0.14$\
& 5782.032 & 1.64 & $-3.53$\
& 5782.042 & 1.64 & $-3.84$\
& 5782.054 & 1.64 & $-3.14$\
& 5782.064 & 1.64 & $-3.19$\
& 5782.073 & 1.64 & $-3.49$\
& 5782.084 & 1.64 & $-2.79$\
& 5782.086 & 1.64 & $-3.14$\
& 5782.098 & 1.64 & $-3.14$\
& 5782.113 & 1.64 & $-2.79$\
& 5782.124 & 1.64 & $-2.79$\
& 5782.153 & 1.64 & $-2.69$\
& 5782.173 & 1.64 & $-2.34$\
\[tab2\]
{height="6cm"} {height="6cm"}
{height="6cm"} {height="6cm"}
{height="6cm"} {height="6cm"}
{height="6cm"} {height="6cm"}
Observations
============
Most of the analysed spectra were collected during several observational campaigns with the CORALIE spectrograph on the 1.2 m Euler Swiss telescope, the FEROS spectrograph on the 2.2 m ESO/MPI telescope (both at La Silla, Chile), the UVES spectrograph on the VLT/UT2 Kueyen telescope (Paranal Observatory, ESO, Chile), the SARG spectrograph on the 3.5 m TNG and the UES spectrograph on the 4.2 m WHT (both at the Roque de los Muchachos Observatory, La Palma, Spain). Part of these data were used in previous articles to derive precise stellar parameters (Santos et al. [@San01], [@San03b], [@San04a]) and abundances of C, O, S, Si, Ca and Ti (Santos et al. [@San00]), as well as to study trends of nine refractory elements (Bodaghee et al. [@Bod03]) and of the volatile nitrogen (Ecuvillon et al. [@Ecu04]). We refer the reader to these papers for a description of the data.
New spectra with high S/N ratios from the UVES and UES spectrographs, on the VLT/UT2 Kueyen telescope (Paranal Observatory, ESO, Chile) and on the 4.2 m WHT (Roque de los Muchachos Observatory, La Palma, Spain), respectively, were used. These are listed in Table \[tab1\]. The S/N ratios for these spectra are in most cases higher than 800 at a resolution of 55000 and 110000 for the UES and the UVES spectrograph, respectively.
Analysis
========
The abundance analysis was carried out in standard local thermodinamic equilibrium (LTE) using a revised version of the spectral synthesis code MOOG (Sneden [@Sne73]) and a grid of Kurucz ([@Kur93]) ATLAS9 atmospheres. All the atmospheric parameters, $T_\mathrm{eff}$, $\log {g}$, $\xi_t$ and \[Fe/H\], and the corresponding uncertainties, were taken from Santos et al. ([@San04a]). The adopted solar abundances for carbon, sulphur, zinc, copper and iron were respectively $\log{\epsilon}\,({\rm C})_{\odot}$ = 8.56dex, $\log{\epsilon}\,({\rm S})_{\odot}$ = 7.21dex , $\log{\epsilon}\,({\rm Zn})_{\odot}$ = 4.60dex, $\log{\epsilon}\,({\rm Cu})_{\odot}$ = 4.21dex (Anders & Grevesse [@And89]) and $\log{\epsilon}\,({\rm Fe})_{\odot}$ = 7.47dex (as used in Santos et al. [@San04a]).
Carbon abundance analysis was carried out by [EW]{} measurement of two lines 5380.3 Å and at 5052.2 Å. Wavelength and excitation energies of the lower level were taken from VALD (Kupka et al. [@Kup99]) and solar gf values were computed using [EW]{}s (21.0 mÅ and 34.2 mÅ, respectively) obtained from the Kurucz Solar Atlas (Kurucz et al. [@Kur84]), and a solar model with $T_\mathrm{eff}$ = 5777K, $\log{g}$ = 4.44 and $\xi_t$ = 1.0kms$^{-1}$. All the adopted parameters for each spectral line are presented in Table \[tab2\]. [EW]{}s were determined by Gaussian fitting using the SPLOT task of IRAF and abundances were computed with the ABFIND driver of MOOG.
Sulphur, zinc and copper abundances were derived by fitting synthetic spectra to the data. For sulphur, we analysed two features: the first at 6743.5 Å consisting of three lines, at 6743.44 Å, 6743.53 Å and at 6743.64 Å, and the second at 6757.1 Å consisting of two lines, at 6757.01 Å and at 6757.18 Å. Fits were carried out in the spectral regions 6742.5–6744.5 Å and 6756.0–6758.0 Å. The line lists for each spectral region were taken from VALD (Kupka et al. [@Kup99]) and $\log{gf}$ values were slightly modified in order to achieve a good fit to the Kurucz Solar Atlas (Kurucz et al. [@Kur84]). The adopted atomic data are listed in Table \[tab2\]. For the instrumental broadening we used a Gaussian function with [FWHM]{} of 0.07 Å and 0.11 Å, depending on the instrumental resolution, and a rotational broadening function with $v\sin{i}$ values from the CORALIE database. For targets without a CORALIE $v\sin{i}$ determination, we chose the $v\sin{i}$ values which fitted the best the spectral lines in the synthesized regions. These values were between 1 and 5kms$^{-1}$, since all our targets are slow rotators. Two examples of the fitting of the two features are shown in Figure \[fig1\]. The red side of the SI feature at 6757 Å could not be perfectly reproduced because of an unknown feature between 6757.25 and 6757.45 Å (there are no spectral lines in the VALD database at these wavelengths). Our differential analysis relative to the Sun cancels out this source of uncertainty since all our targets are affected in the same way as the Sun. Moreover, we have verified that an $EW$ analysis yields the same final abundances, and therefore we consider this effect to be negligible.
Zinc abundances were derived from the two lines at 4810.5 Å and at 4722.2 Å. We computed 5Å wide synthetic spectra. The atomic lists for elements other than zinc were provided by VALD (Kupka et al. [@Kup99]), while atomic data for the lines were taken from Gurtovenko & Kostyk ([@Gurt89]). Copper abundances were derived from the lines at 5218.2 Å and at 5782.1 Å, by the synthesis of the spectral regions 5213–5220 Å and 5780–5785Å. The atomic lists for elements other than copper were provided by VALD (Kupka et al. [@Kup99]), while atomic data (hyperfine structure components) for the lines were taken from Steffen ([@Ste85]). A differential analysis relative to the Sun was carried out for the two elements. Some wavelengths were slightly modified in order to fit the solar spectrum better. Table \[tab2\] presents the adopted values for both elements. Figures \[fig2\] and \[fig3\] show two examples of fits for the and lines, respectively.
Uncertainties in the atmospheric parameters are of the order of 50K in $T_\mathrm{eff}$, 0.12dex in $\log {g}$, 0.08kms$^{-1}$ in $\xi_t$ and 0.05dex in \[Fe/H\] (see Santos et al. [@San04a]). We estimated the effect of changes in atmospheric parameters on the resulting abundances in the following way. For each atmospheric parameter, we selected a set of three stars having similar values of all the parameters, excepting the considered one, which must vary within the sample. We then tested abundance sensitivity to changes in the parameter ($\pm$100K for $T_\mathrm{eff}$, $\pm$0.3dex for $\log {g}$ and \[Fe/H\], $\pm$0.5kms$^{-1}$ for $\xi_t$). The results are shown in Table \[tab3\]. Uncertainties in the abundances of all the elements were determined adding in quadrature the dispersion of each abundance from the mean value, 0.05 dex of the continuum determination and the errors due to the abundance sensitivities to changes in the atmospheric parameters.
Dependences on $T_\mathrm{eff}$ and on $\log {g}$ of \[C/Fe\], \[S/Fe\], \[Zn/Fe\] and \[Cu/Fe\] are represented in Fig. \[fig4\], Fig. \[fig5\], Fig. \[fig6\] and Fig. \[fig7\], respectively. We note that no characteristic trends appear in any of the cases. This means that our results are almost free from systematic errors. Only in the case of Zn does \[Zn/Fe\] decrease for $T_\mathrm{eff}$ greater than 6000K. This could be due to an NLTE effect, usually not very strong in metal-rich stars for complex atoms such as iron (see Edvardsson et al. [@Edv93]; Thévenin & Idiart [@The99]). We verified that using only stars with $T_\mathrm{eff}$ lower than 6000K would not change the resulting trends.
[ccccc]{} & Star & & &\
& ($T_\mathrm{eff}$; $\log {g}$ ; \[Fe/H\] ; $\xi_t$) & (5073; 4.43; -0.13; 1.05) & (5601; 4.41; -0.08; 0.99) & (6222; 4.27; -0.04; 1.79)\
C: & $\Delta T_\mathrm{eff}=\pm100$K & $\mp0.09$ & $\mp0.07$ & $\mp0.05$\
S: & $\Delta T_\mathrm{eff}=\pm100$K & $\mp0.10$ & $\mp0.10$ & $\mp0.05$\
Zn: & $\Delta T_\mathrm{eff}=\pm100$K & $\pm0.03$ & $\pm0.04$ & $\pm0.05$\
Cu: & $\Delta T_\mathrm{eff}=\pm100$K & $\pm0.03$ & $\pm0.05$ & $\pm0.05$\
& Star & & &\
& ($T_\mathrm{eff}$; $\log {g}$ ; \[Fe/H\]; $\xi_t$) & (5641; 4.05; 0.14; 1.13) & (5801; 4.22; 0.15; 1.34) & (5656; 4.45; 0.22; 1.01)\
C: & $\Delta \log{g}=\pm0.3$dex & $\mp0.10$ & $\mp0.10$ & $\mp0.10$\
S: & $\Delta \log{g}=\pm0.3$dex & $\mp0.10$ & $\mp0.10$ & $\mp0.10$\
Zn: & $\Delta \log{g}=\pm0.3$dex & $\pm0.07$ & $\pm0.07$ & $\pm0.06$\
Cu: & $\Delta \log{g}=\pm0.3$dex & $\pm0.02$ & $\pm0.03$ & $\pm0.03$\
& Star & & &\
& ($T_\mathrm{eff}$; $\log {g}$ ; \[Fe/H\]; $\xi_t$) & (5835; 4.60; -0.52; 1.53) & (5954; 4.44; 0.06; 1.30) & (5636; 4.23; 0.40; 1.12)\
C: & $\Delta$\[Fe/H\]$ =\pm0.3$dex & $\mp0.03$ & $\mp0.02$ & $\mp0.02$\
S: & $\Delta$\[Fe/H\] $=\pm0.3$dex & $\mp0.00^1$ & $\mp0.00^1$ & $\mp0.00^1$\
Zn: & $\Delta$\[Fe/H\] $=\pm0.3$dex & $\mp0.20$ & $\mp0.20$ & $\mp0.20$\
Cu: & $\Delta$\[Fe/H\] $=\pm0.3$dex & $\mp0.30$ & $\mp0.20$ & $\mp0.20$\
& Star & & &\
& ($T_\mathrm{eff}$; $\log {g}$ ; \[Fe/H\]; $\xi_t$) & (5644; 4.37; 0.33; 0.89) & (5518; 4.42; 0.26; 1.22) & (6190; 4.19; 0.24; 1.54)\
C: & $\Delta \xi_t=\pm0.5$kms$^{-1}$ & $\mp0.01$ & $\mp0.01$ & $\mp0.01$\
S: & $\Delta \xi_t=\pm0.5$kms$^{-1}$ & $\mp0.00^1$ & $\mp0.00^1$ & $\mp0.00^1$\
Zn: & $\Delta \xi_t=\pm0.5$kms$^{-1}$ & $\mp0.10$ & $\mp0.10$ & $\mp0.10$\
Cu: & $\Delta \xi_t=\pm0.5$kms$^{-1}$ & $\mp0.10$ & $\mp0.10$ & $\mp0.10$\
\[tab3\]
Results {#Res}
=======
Several studies have been recently published about abundances of elements other than iron in planet host stars (e.g. Santos et al. [@San00], [@San02]; Gonzalez et al. [@Gonz01]; Takeda et al. [@Tak01]; Sadakane et al. [@Sad02]; Bodaghee et al. [@Bod03]; Israelian et al. [@Isr04]; Ecuvillon et al. [@Ecu04]). Studies concerning such volatile elements as CNO, S and Zn were carried out, but most of them did not count on a large set of planet-harbouring targets and a homogeneous comparison sample of stars without known planetary companions.
Santos et al. ([@San00]) compared the abundances of C, O and S for eight planet host stars with results for field dwarfs from the literature without finding significant differences. They pointed out the need to use a homogeneous comparison sample of stars without planetary mass companions. Also Gonzalez et al. ([@Gonz01]) found no deviation of the C and O abundances of twenty planet hosts from the trends traced by field stars taken from the literature. Takeda et al. ([@Tak01]) and Sadakane et al. ([@Sad02]) presented abundances of twelve and nineteen elements, respectively, among which the volatiles C, N, O, S and Zn and the refractory Cu, for fourteen and twelve planet-harbouring stars, respectively. Their results revealed no anomalies between planet host and comparison samples. Ecuvillon et al. ([@Ecu04]) presented a homogeneous analysis of N abundances in 91 solar-type stars, 66 with and 25 without known planets, and confirmed that the two samples show the same behaviour.
In this paper we analysed two optical lines of in 91 targets with planets and in 31 comparison sample stars. All atmospheric parameters, [EW]{} values with uncertainties and abundance results for both samples are listed in Tables \[tab4\], \[tab5\] and \[tab6\]. For sulphur, zinc and copper, we synthesized two optical lines in 84, 68 and 73 planet host stars, and in 31, 41 and 41 comparison sample stars, respectively. Atmospheric parameters, abundance results and the corresponding uncertainties, are listed in Tables \[tab7\] to \[tab15\].
{height="6cm"} {height="6cm"}
{height="6cm"} {height="6cm"}
{height="6cm"} {height="6cm"}
Figure \[fig8\] presents \[C/H\] and \[C/Fe\] as functions of \[Fe/H\]. Both samples, stars with and without planets, behave quite similarly. Since targets with planets are on average more metal rich than comparison sample stars, their abundance distributions correspond to the extensions of the comparison sample trends at high \[Fe/H\]. Planet host stars do not present anomalies in carbon abundances with respect to comparison sample dwarfs.
Similar results are obtained for sulphur and zinc. Figures \[fig9\] and \[fig10\] show the \[$X$/H\] and \[$X$/Fe\] vs. \[Fe/H\] plots for the two elements. The abundance trends of sulphur and zinc in planet host stars are similar to those of the comparison sample. In fact, no discontinuity is seen in the overlap region of both samples.
\[$X$/Fe\] vs. \[Fe/H\] plots for the three volatiles show decreasing trends with increasing \[Fe/H\]. Although this effect is more evident for \[C/Fe\] and \[S/Fe\] than for \[Zn/Fe\], they all show a negative slope . This means that, unlike nitrogen (see Ecuvillon et al. [@Ecu04]), none of these elements keeps pace with iron; an excess of carbon, sulphur and zinc exists in the more metal-poor tail with respect to stars with high \[Fe/H\]. Since no differences appear to be related to the presence of planets, the observed slopes may be a by-product of Galactic chemical evolution. However, no detailed models are available to explain trends such as these.
\[Cu/H\] and \[Cu/Fe\] vs. \[Fe/H\] trends are provided in Figure \[fig11\]. No discontinuities appear between the two samples. In the \[Cu/Fe\] vs. \[Fe/H\] plot, the planet host star set seems to produce a steeper fit than that corresponding to the comparison sample. Although this may be seen as a signature of the presence of planets, it seems more likely that the \[Cu/Fe\] slope increases at \[Fe/H\] $>$ 0.1, and that its relation to the presence of planets is solely due to the high metallicity of the parent stars. The \[Cu/Fe\] trend corresponding to both samples is on average slightly overabundant with respect to \[Fe/H\].
The \[$X$/H\] distributions for the four elements are presented in Figure \[fig12\]. Volatile (C, S and Zn) distributions differ in their shapes. The planet host distributions are strongly asymmetrical, with peaks at high \[$X$/H\]. The comparison samples show different behaviours: the same asymmetric distribution for sulphur and a mirror shape for carbon. In the case of zinc, the comparison sample distribution looks more symmetrical. The \[Cu/H\] distributions for the two sets, stars with and without planets, present similar shapes, both quite symmetric. The cumulative distributions show that differences between planet host and comparison sample stars are statistically significant.
The average values of \[$X$/H\] for the samples with and without planets for each element, as well as the rms dispersions and the differences between the mean \[$X$/H\] values, are listed in Table \[TabAve\]. The differences vary from 0.16 to 0.31dex. These discrepancies are not very significant because of the high dispersion around the mean value.
We note that the differences between average values of \[$X$/H\] for the two samples are smaller for carbon and sulphur than for zinc and copper. For nitrogen, another volatile element, a difference of the order of 0.3dex was found between the average \[N/H\] values for planet host and comparison sample stars (Ecuvillon et al. [@Ecu04]). Even though these discrepancies are not statistically significant, we cannot completely exclude that this characteristic, if confirmed, may be related with the different condensation temperatures of the elements.
{height="6cm"} {height="6cm"}
{height="6cm"} {height="6cm"} {height="6cm"} {height="6cm"} {height="6cm"} {height="6cm"} {height="6cm"} {height="6cm"}
{height="6cm"} {height="6cm"} {height="6cm"} {height="6cm"}
Comparison with the literature
==============================
Previous studies concerning abundances of refractory and volatiles elements (see Sect. \[Res\]) have analysed planet host stars common to our study. In order to carry out a homogeneous comparison, we gathered all these abundance results and scaled them to our atmospheric parameters using the published sensitivities for each atomic line.
Santos et al. ([@San00]), Gonzalez et al. ([@Gonz01]), Takeda et al. ([@Tak01]) and Sadakane et al. ([@Sad02]) analysed one of the two lines we used to compute carbon abundances, the line at 5380 Å. We could therefore directly compare the measured [EW]{}s. Table \[tab16\] shows all the atmospheric parameters and the [EW]{} values for the 46 compared targets. All the measurements are in good agreement, especially for values of Gonzalez et al. ([@Gonz01]) from Keck spectra, with differences tipically lower than 2 mÅ. In some cases, our data differ from the values taken from Takeda et al. ([@Tak01]), but the other sources agree better with our results (see and ). Only for is our measurement larger than the two values in the literature.
For sulphur, comparisons for 31 targets were possible. All the atmospheric parameters and S abundances are listed in Table \[tab17\]. \[S/H\] values from Santos et al. ([@San00]) and Sadakane et al. ([@Sad02]) are in good agreement with our results in most cases, with discrepancies lower than 0.2dex. However, significant disagreements, of the order of 0.4dex, appear when comparing with S abundances from Takeda et al. ([@Tak01]). Using different lines and atomic parameters could be the cause of this discrepancy. Sadakane et al. ([@Sad02]) synthesized the same lines as we did. Our results agree very well. Since Gonzalez et al. ([@Gonz01]) did not include a sensitivity study in their work, we could not scale their results to our atmospheric parameters, thus preventing a homogeneous comparison.
Zinc abundances were compared with literature values in nineteen targets (see Table \[tab18\]). We obtained similar results in almost all cases, with differences lower than 0.2dex. Only , and show clear discrepancies with results from Takeda et al. ([@Tak01]). As for S, the cause of these differences could be that different methods, with different sets of lines, were used.
Sadakane et al. ([@Sad02]) obtained copper abundances in eight planet host stars common to our work. The comparison is presented in Table \[tab19\]. The results are in very good agreement, with differences lower than 0.13dex, except for and . In these two cases, the results may diverge because of the different methods used. All the comparisons are represented in Fig. \[fig13\].
Discussion and conclusions
==========================
This article presents the abundances of one refractory (Cu) and three volatile (C, S and Zn) elements in a large set of planet-harbouring stars and in an unbiased volume-limited comparison sample of solar-type dwarfs with no known planetary mass companions. An independent and uniform study of the two samples was performed for each element using atmospheric parameters derived from a detailed spectroscopic analysis by Santos et al. ([@San04a]). We have carried out a careful comparison of our results with those already published by several authors and obtained a good agreement on the whole. The result is a significant study for the completeness and homogeneity of the four elements analysed.
Comparing trends and searching for differences between the two samples, stars with and without known planets, can provide clues toward clarifying the formation and evolution of planetary systems. In particular, the behaviour of volatile elements can be very informative for discriminating between the “self-enrichment” and the “primordial” hypotheses, and the relative importance of the differential accretion (Gonzalez [@Gonz97]; Santos et al. [@San00], [@San01]; Smith et al. [@Smi01]). If the accretion of large amounts of metal-rich rocky material were the principal reason for the observed enhancement in iron abundance in planet host stars, then volatiles would not show as much overabundance as refractories do in these targets.
Our results show that abundances of volatile elements have the same behaviour in stars with and without planets. The abundance trends for planet host and comparison sample stars are nearly indistinguishable. The planet host distributions are the natural extensions of the comparison sample trends toward high metallicities. This may imply that the accretion of rocky materials is not the principal cause of the metal-rich nature of stars with planets. Although the possibility of pollution is not excluded (see Laws & Gonzalez [@Law01]; Israelian et al. [@Isr01], [@Isr03a]), the hypothesis of a primordial metal-rich cloud out of which planetary systems would have formed seems likelier than the “self-enrichment” scenario, as an explanation for the observed metallicity enhancement.
Most of the evidence suggests a “primordial” origin for the metallicity excess (Pinsonneault et al. [@Pin01]; Santos et al. [@San01], [@San03b], [@San04a]). Previous studies of volatiles have already led to the same conclusion. Santos et al. ([@San00]) found no significant differences in \[C/Fe\] and \[O/Fe\] trends between field stars from literature and eight planet host stars. Takeda et al. ([@Tak01]) and Sadakane et al. ([@Sad02]) arrived at the same conclusion by comparing refractory and volatile elements and searching for a $T_{\rm C}$ dependence. Ecuvillon et al. ([@Ecu04]) found that nitrogen abundances show the same behaviour in a large set of stars with and without known planets.
Furthermore, we have obtained negative slopes in the \[$X$/Fe\] trends for the three volatiles, while the \[N/Fe\] trend was previously found flat (Ecuvillon et al. [@Ecu04]). Contrary to results obtained in previous studies (see Takeda et al. [@Tak01]; Sadakane et al. [@Sad02]), C, S and Zn abundances do not scale with that of iron. Since no differences are seen between the two samples, this behaviour must be evidence of the chemical evolution of the Galactic disc, with no link to the presence of planets. In fact, several studies of abundances in the Galactic disc have revealed just such a clear linear behaviour for carbon (Friel & Boesgaard [@Fri92]; Andersson & Edvardsson [@And94]; Gustafsson et al. [@Gus99]; Shi et al. [@Shi02]) and for sulphur (Takada-Hidai et al. [@Tak02]; Chen et al. [@Che02]) in the metallicity range $-0.8<$ \[Fe/H\] $<$ 0.3. The \[Zn/Fe\] vs. \[Fe/H\] plot shows a decreasing trend for $-0.6<$ \[Fe/H\] $<$ 0 and a slight rise at metallicities above solar. A similar result has been recently obtained by Bensby et al. ([@Ben03]) for 66 disc stars.
Concerning the refractory Cu, \[Cu/Fe\] vs. \[Fe/H\] plot reveals a behaviour similar to Zn, with a slight decrease at $-$0.6 $<$ \[Fe/H\] $<$ 0 and a rise above the solar metallicity, steeper than for Zn. \[Cu/Fe\] is on average overabundant with respect to the solar value. The larger value of the slope resulting from fitting only stars with planets might be due to the observed \[Cu/Fe\] rise at \[Fe/H\] $>$ 0, probably related to Galactic chemical evolutionary effects rather than to the presence of planets. In fact, planet host stars do not show different behaviour with respect to the comparison sample. Since all the available studies of Cu Galactic trends do not include the solar metallicity range, we cannot compare our results with those of the literature for this element.
The \[$X$/Fe\] trends must be compared with detailed models of Galactic chemical evolution to distinguish possible effects due to the presence of planets. In this framework, our work provides new additional data of C, S, Zn and Cu abundances to check Galactic chemical evolution models in the high \[Fe/H\] region. Because of the lack of detailed abundance studies reaching solar metallicities, our results can be very informative to improve our present understanding of stellar nucleosynthesis and chemical evolution.
In the future, it will be very important to manage uniform studies of other volatile and refractory elements, with a wide range of condensation temperatures $T_{\rm C}$. Possible systematic trends of \[$X$/H\] in terms of $T_{\rm C}$ will give us conclusive information about the relative importance of differential accretion on the metallicity excess as a whole.
IRAF is distributed by the National Optical Astronomy Observatories, operated by the Association of Universities for Research in Astronomy, Inc., under contract with the National Science Foundation, USA.
Anders, E., & Grevesse, N. 1989, Geochim et. Cosmochim. Acta, 53, 197
Andersson, H., & Edvardsson, D. 1994, A&A, 290, 590
Bensby, T.,Feltzing, S., & Lundström, I. 2003, A&A, 410, 527
Bihain, G., Israelian, G., Rebolo, R., Bonifacio, P., & Molaro, P. 2004, A&A, submitted
Bodaghee, A., Santos, N. C., Israelian, G. & Mayor, M. 2003, A&A, 404, 715
Carretta, E., Gratton, R. G., & Sneden, C. 2000, A&A, 356, 238
Chen, Y. Q., Nissen, P. E., Zhao, G., & Apslund, M. 2002, A&A, 390, 225
Ecuvillon, A., Israelian, G., Santos, N.C., Mayor, M., R. J. García López, Randich, S., 2004, A&A, 418, 703
Edvardsson, B., Andersen, J., Gustafsson, B., Lambert, D. L., Nissen, P.E., & Tomkin, J. 1993, A&A, 275, 101
Friel, E. D., & Boesgaard, A. M. 1992, ApJ, 387, 170
Gonzalez, G. 1997, MNRAS, 285, 403
Gonzalez, G., & Laws, C. 2000, AJ, 119, 390
Gonzalez, G., Laws, C., Tyagi, S., & Reddy, B. E. 2001, AJ, 121, 432
Gurtovenko, E. A., & Kostyk, R. I. 1989, Fraunhoffer spectrum and system of solar oscillator strengths, KiIND, 200
Gustafsson, B., Karlsson, T., Olsson, E., Edvardsson, B., & Ryde, N. 1999, A&A, 342, 426
Israelian, G., Santos, N. C., Mayor, M., & Rebolo, R. 2001, Nature, 411, 163
Israelian, G., Santos, N. C., Mayor, M., & Rebolo, R. 2003a, A&A, 405, 753
Israelian, G. 2003b, in IAU S219: Stars as Suns: Activity, Evolution, and Planets, ed. A. K. Dupree (San Francisco: ASP), in press
Israelian, G., Santos, N. C., Mayor, M., & Rebolo, R. 2004, A&A, 414, 601
Kobulnicky, H. A., Skillman, E. D. 1998, ApJ, 497, 601
Kupka, F. G., Piskunov, N. E., Ryabchikova, T. A., Stempels, H. C., Weiss, W. W., 1999, A&AS, 138, 119
Kurucz, R. L. 1993, CD-ROMs, ATLAS9 Stellar Atmospheres Programs and 2 $\mathrm{km}\,\mathrm{s}^{-1}$ Grid (Cambridge: Smithsonian Astrophys. Obs.)
Kurucz, R. L., Furenlid, I., Brault, J., Testerman, L. 1984, Solar Flux Atlas from 296 to 1300 nm, NOAO Atlas No. 1
Laws, C. & Gonzalez, G. 2001, ApJ, 553, 405
Laws, C., Gonzalez, G., Walker, K. M., Tyagi, S., Dodsworth, J., Snider, K. & Suntzeff, N. 2003, AJ, 125, 2664
Luck, R. E., & Bond, H. E. 1985, ApJ, 292, 559
Mayor, M. & Queloz, D. 1995, Nature, 378, 355
Mishenina, T. V., Kovtyukh, V. V., Soubiran, C., Travaglio, C., Busso, M., 2002, A&A, 396, 189
Murray, N., & Chaboyer, B. 2002, ApJ, 566, 442
Pinsonneault, M. H., DePoy, D. L., & Coffee, M., 2001, ApJ, 556, L59
Prantzos, N., Vangioni-Flam E., Chauveau, S. 1994, A&A, 285, 132
Reid, I. N. 2002, PASP, 114, 306
Sadakane, K., Ohkubo, Y., Takeda, Y., Sato, B., Kambe, E., & Aoki, W. 2002, PASJ, 54, 911
Sandquist, E. L., Dokter, J. J., Lin, D. N. C., & Mardling, R. 2002, ApJ, 572, 1012
Santos, N. C., Israelian, G., & Mayor, M. 2000, A&A 363, 228
Santos, N. C., Israelian, G., & Mayor, M. 2001, A&A 373, 1019
Santos, N. C., García López, R. J., Israelian, G., Mayor, M., Rebolo, R., García-Gil, A., Pérez de Taoro, M. R., & Randich, S. 2002, A&A, 386, 1028
Santos, N. C., Mayor, M., Udry, S., et al., 2003a, in IAU S219: Stars as Suns: Activity, Evolution, and Planets, ed. A. K. Dupree (San Francisco: ASP), in press
Santos, N. C., Israelian, G., Mayor, M., Rebolo, R., & Udry, S. 2003b, A&A,398,363
Santos, N. C., Israelian, G., & Mayor, M. 2004a, A&A, 415, 1153
Santos, N.C., Israelian, G., García López, R. J., Mayor, M., Rebolo, R., Randich, S., Ecuvillon, A., & Cerdeña, C. 2004b, A&A, submitted
Shi, J. R., Zhao, G., & Chen, Y. Q. 2002, A&A, 381, 982
Smith, V. V., Cunha, K., & Lazzaro, D., 2001, AJ, 121,3207
Sneden, C. 1973, Ph.D. thesis, University of Texas
Sneden, C., Gratton, R. G., Crocker, D. A. 1991, A&A, 246, 354
Steffen, M. 1985, A&AS, 59, 403
Takada-Hidai, M., Takeda, Y., Sato, S., Honda, S., Sadakane, K., Kawanomoto, S., Sargent, W. L. W., Lu, L., & Barlow, T. 2002, ApJ, 573, 614
Takeda, Y., Sato, B., Kambe, E., Aoki, W., Honda, S., Kawanomoto, S., & Masuda, S., et al. 2001, PASJ, 53, 1211
Thévenin, F. & Idiart, T. P. 1999, ApJ, 521, 753
Timmes, F. X., Woosley, S. E., & Weaver, T. A. 1995, ApJS, 98, 617
Vauclair, S. 2003, ApJ, to be published in ApJ, April 20
[lccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & EW$_{5380}$ & EW$_{5052}$ & \[C/Fe\]\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & (mÅ) & (mÅ) &\
& $6302\pm56$ & $4.34\pm0.13$ & $ 1.86\pm0.17$ & $ 0.14\pm0.07$ & 52.1 & 69.1 & $ 0.10\pm0.08$\
& $5536\pm50$ & $4.56\pm0.12$ & $ 1.33\pm0.06$ & $ 0.12\pm0.06$ & 16.7 & 27.6 & $-0.06\pm0.05$\
& $5976\pm51$ & $4.45\pm0.10$ & $ 1.26\pm0.07$ & $ 0.32\pm0.06$ & 33.3 & 52.8 & $-0.13\pm0.05$\
& $5173\pm35$ & $4.37\pm0.12$ & $ 0.74\pm0.05$ & $ 0.12\pm0.04$ & 12.9 & 20.1 & $-0.01\pm0.07$\
& $5636\pm40$ & $4.23\pm0.14$ & $ 1.12\pm0.05$ & $ 0.40\pm0.05$ & 41.0 & 55.9 & $ 0.05\pm0.07$\
& $5626\pm32$ & $4.49\pm0.10$ & $ 0.95\pm0.06$ & $-0.24\pm0.04$ & 20.5 & 22.2 & $ 0.23\pm0.19$\
& $5835\pm59$ & $4.60\pm0.12$ & $ 1.53\pm0.27$ & $-0.52\pm0.08$ & 19.0 & 21.0 & $ 0.40\pm0.19$\
& $6151\pm57$ & $4.51\pm0.10$ & $ 1.45\pm0.15$ & $ 0.06\pm0.07$ & 33.4 & 45.4 & $-0.01\pm0.09$\
& $6212\pm64$ & $4.26\pm0.13$ & $ 1.69\pm0.16$ & $ 0.13\pm0.08$ & 47.8 & 64.6 & $ 0.07\pm0.07$\
& $6130\pm31$ & $4.48\pm0.08$ & $ 1.40\pm0.08$ & $-0.03\pm0.04$ & 27.2 & 41.2 & $-0.01\pm0.04$\
& $5641\pm28$ & $4.05\pm0.05$ & $ 1.13\pm0.03$ & $ 0.14\pm0.04$ & 27.4 & 43.2 & $-0.01\pm0.03$\
& $5702\pm36$ & $4.33\pm0.08$ & $ 1.05\pm0.04$ & $ 0.36\pm0.05$ & 32.3 & 52.2 & $-0.05\pm0.05$\
& $5163\pm00$ & $4.52\pm0.00$ & $ 0.72\pm0.00$ & $-0.24\pm0.00$ & 12.9 & 12.7 & $ 0.31\pm0.24$\
& $5801\pm30$ & $4.22\pm0.12$ & $ 1.34\pm0.04$ & $ 0.15\pm0.04$ & 28.2 & 41.2 & $-0.08\pm0.06$\
& $6252\pm53$ & $4.61\pm0.16$ & $ 1.18\pm0.10$ & $ 0.26\pm0.06$ & 47.7 & 61.4 & $ 0.02\pm0.10$\
& $6190\pm00$ & $4.19\pm0.00$ & $ 1.54\pm0.00$ & $ 0.24\pm0.00$ & 61.6 & 74.9 & $ 0.15\pm0.09$\
& $6138\pm79$ & $4.53\pm0.22$ & $ 1.22\pm0.16$ & $ 0.17\pm0.10$ & 23.2 & 39.6 & $-0.28\pm0.09$\
& $5073\pm42$ & $4.43\pm0.08$ & $ 1.05\pm0.06$ & $-0.13\pm0.05$ & 9.9 & 14.5 & $ 0.19\pm0.10$\
& $5959\pm46$ & $4.35\pm0.12$ & $ 1.20\pm0.10$ & $-0.11\pm0.06$ & 23.1 & 36.3 & $ 0.03\pm0.06$\
& $6108\pm36$ & $4.25\pm0.10$ & $ 1.30\pm0.05$ & $ 0.31\pm0.05$ & 44.5 & 65.7 & $-0.05\pm0.04$\
& $4825\pm107$ & $3.55\pm0.32$ & $1.18\pm0.12$ & $ 0.39\pm0.13$ & 16.8 & 26.0 & $-0.05\pm0.13$\
& $5656\pm44$ & $4.45\pm0.08$ & $ 1.01\pm0.06$ & $ 0.22\pm0.05$ & 27.8 & 42.4 & $ 0.03\pm0.05$\
& $5587\pm00$ & $4.29\pm0.00$ & $ 1.08\pm0.00$ & $ 0.38\pm0.00$ & 37.6 & 54.6 & $ 0.08\pm0.02$\
& $6046\pm49$ & $4.71\pm0.09$ & $ 1.79\pm0.19$ & $-0.08\pm0.06$ & 24.4 & 32.7 & $ 0.28\pm0.11$\
& $5546\pm30$ & $4.50\pm0.03$ & $ 0.80\pm0.07$ & $-0.38\pm0.04$ & 13.0 & 19.9 & $ 0.29\pm0.06$\
& $5674\pm40$ & $3.94\pm0.12$ & $ 1.38\pm0.05$ & $ 0.40\pm0.06$ & 45.0 & 53.3 & $-0.06\pm0.14$\
& $5991\pm27$ & $4.42\pm0.10$ & $ 1.24\pm0.04$ & $ 0.10\pm0.04$ & 30.7 & 43.8 & $-0.03\pm0.06$\
& $6145\pm42$ & $4.31\pm0.15$ & $ 1.29\pm0.09$ & $ 0.21\pm0.05$ & 34.7 & 55.8 & $-0.11\pm0.07$\
& $5268\pm55$ & $4.41\pm0.16$ & $ 0.97\pm0.06$ & $ 0.20\pm0.06$ & 20.8 & 27.0 & $ 0.08\pm0.14$\
& $5644\pm54$ & $4.37\pm0.07$ & $ 0.89\pm0.07$ & $ 0.33\pm0.06$ & 25.9 & 34.9 & $-0.19\pm0.09$\
& $6026\pm30$ & $4.41\pm0.13$ & $ 1.11\pm0.06$ & $ 0.01\pm0.04$ & 28.3 & 40.7 & $-0.00\pm0.07$\
& $6105\pm00$ & $4.28\pm0.00$ & $ 1.36\pm0.00$ & $ 0.23\pm0.00$ & 43.6 & 66.9 & $ 0.04\pm0.03$\
& $5666\pm31$ & $4.53\pm0.09$ & $ 1.06\pm0.05$ & $-0.12\pm0.04$ & 12.3 & 21.4 & $-0.07\pm0.04$\
& $5988\pm52$ & $4.45\pm0.15$ & $ 1.25\pm0.08$ & $ 0.36\pm0.06$ & 35.7 & - & $-0.14\pm0.06$\
& $5995\pm45$ & $4.30\pm0.07$ & $ 1.42\pm0.09$ & $ 0.03\pm0.06$ & 23.4 & 36.7 & $-0.14\pm0.04$\
& $5518\pm49$ & $4.42\pm0.12$ & $ 1.22\pm0.06$ & $ 0.26\pm0.06$ & 24.8 & 30.6 & $-0.08\pm0.14$\
& $5699\pm49$ & $4.27\pm0.12$ & $ 1.26\pm0.06$ & $ 0.27\pm0.06$ & 28.7 & 43.2 & $-0.10\pm0.06$\
& $6112\pm39$ & $4.34\pm0.10$ & $ 1.38\pm0.07$ & $ 0.16\pm0.05$ & 39.1 & 54.9 & $-0.00\pm0.06$\
& $6143\pm53$ & $4.42\pm0.13$ & $ 1.53\pm0.09$ & $ 0.28\pm0.07$ & 37.6 & 56.6 & $-0.13\pm0.06$\
& $5279\pm62$ & $4.37\pm0.18$ & $ 0.98\pm0.07$ & $ 0.33\pm0.07$ & 21.6 & 29.6 & $-0.02\pm0.12$\
& $5737\pm34$ & $4.25\pm0.14$ & $ 1.18\pm0.04$ & $ 0.41\pm0.05$ & 32.3 & 46.1 & $-0.20\pm0.07$\
& $5574\pm72$ & $4.46\pm0.20$ & $ 1.14\pm0.09$ & $ 0.32\pm0.09$ & 25.2 & 42.4 & $-0.04\pm0.09$\
& $6015\pm00$ & $4.46\pm0.00$ & $ 1.13\pm0.00$ & $ 0.30\pm0.00$ & 37.6 & 66.8 & $ 0.02\pm0.11$\
& $5454\pm61$ & $4.33\pm0.17$ & $ 1.08\pm0.08$ & $ 0.35\pm0.08$ & 27.7 & 41.0 & $-0.00\pm0.08$\
& $6234\pm45$ & $3.98\pm0.05$ & $ 1.62\pm0.08$ & $ 0.22\pm0.05$ & 51.4 & 75.1 & $ 0.03\pm0.04$\
& $5821\pm41$ & $4.45\pm0.06$ & $ 1.16\pm0.05$ & $ 0.32\pm0.05$ & 37.0 & 47.2 & $-0.05\pm0.11$\
& $5954\pm25$ & $4.44\pm0.10$ & $ 1.30\pm0.04$ & $ 0.06\pm0.03$ & 25.9 & 42.5 & $-0.03\pm0.04$\
& $5899\pm35$ & $4.34\pm0.07$ & $ 1.08\pm0.06$ & $-0.01\pm0.05$ & 23.1 & 37.0 & $-0.04\pm0.03$\
& $6248\pm42$ & $4.49\pm0.16$ & $ 1.35\pm0.08$ & $ 0.20\pm0.05$ & 37.5 & 56.5 & $-0.07\pm0.06$\
& $5596\pm42$ & $4.37\pm0.12$ & $ 0.89\pm0.05$ & $ 0.23\pm0.05$ & 27.5 & 36.3 & $-0.02\pm0.10$\
& $5494\pm26$ & $4.50\pm0.10$ & $ 0.84\pm0.05$ & $-0.36\pm0.04$ & 10.2 & 16.1 & $ 0.17\pm0.06$\
& $5886\pm36$ & $4.28\pm0.13$ & $ 1.25\pm0.09$ & $-0.25\pm0.05$ & 16.2 & 32.8 & $ 0.05\pm0.10$\
& $5884\pm34$ & $4.22\pm0.02$ & $ 1.31\pm0.17$ & $-0.70\pm0.04$ & 12.2 & 21.2 & $ 0.33\pm0.03$\
& $5098\pm36$ & $4.45\pm0.11$ & $ 0.74\pm0.05$ & $ 0.09\pm0.04$ & 17.2 & 22.6 & $ 0.24\pm0.13$\
& $5560\pm34$ & $4.07\pm0.05$ & $ 1.18\pm0.05$ & $-0.06\pm0.05$ & 18.0 & 29.4 & $-0.01\pm0.03$\
& $6339\pm73$ & $4.19\pm0.10$ & $ 1.70\pm0.16$ & $ 0.23\pm0.07$ & 59.9 & 85.1 & $ 0.19\pm0.08$\
\[tab4\]
[lccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & EW$_{5380}$ & EW$_{5052}$ & \[C/Fe\]\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & (mÅ) & (mÅ)&\
& $6075\pm40$ & $4.64\pm0.12$ & $ 1.31\pm0.07$ & $ 0.16\pm0.05$ & 29.7 & 41.8 & $-0.10\pm0.08$\
& $5392\pm36$ & $4.48\pm0.06$ & $ 0.85\pm0.05$ & $ 0.03\pm0.04$ & 14.5 & 19.5 & $-0.03\pm0.11$\
& $5776\pm29$ & $4.36\pm0.07$ & $ 1.09\pm0.04$ & $ 0.30\pm0.04$ & 36.3 & 55.2 & $ 0.03\pm0.03$\
& $6222\pm39$ & $4.27\pm0.15$ & $ 1.79\pm0.12$ & $-0.04\pm0.05$ & 36.9 & 53.4 & $ 0.07\pm0.06$\
& $4775\pm113$ & $3.09\pm0.40$ & $ 1.78\pm0.11$ & $0.13\pm0.14$ & 17.3 & 25.5 & $ 0.08\pm0.16$\
& $5909\pm39$ & $4.51\pm0.08$ & $ 1.13\pm0.06$ & $ 0.10\pm0.05$ & 28.2 & 49.4 & $ 0.07\pm0.06$\
& $6045\pm44$ & $4.53\pm0.08$ & $ 1.12\pm0.07$ & $ 0.21\pm0.05$ & 24.5 & 41.0 & $-0.23\pm0.05$\
& $5853\pm25$ & $4.41\pm0.15$ & $ 1.35\pm0.07$ & $-0.21\pm0.04$ & 21.1 & 33.5 & $ 0.16\pm0.05$\
& $5311\pm87$ & $4.42\pm0.18$ & $ 0.92\pm0.10$ & $ 0.43\pm0.08$ & 24.5 & 34.4 & $-0.01\pm0.11$\
& $5883\pm25$ & $4.51\pm0.05$ & $ 1.18\pm0.04$ & $ 0.06\pm0.04$ & 22.0 & 32.5 & $-0.11\pm0.06$\
& $5961\pm27$ & $4.50\pm0.10$ & $ 1.11\pm0.06$ & $-0.01\pm0.04$ & 19.1 & 30.3 & $-0.13\pm0.04$\
& $5798\pm33$ & $4.31\pm0.08$ & $ 1.19\pm0.04$ & $ 0.32\pm0.04$ & 35.2 & 54.2 & $-0.04\pm0.04$\
& $5617\pm35$ & $4.22\pm0.05$ & $ 1.21\pm0.05$ & $ 0.06\pm0.05$ & 26.7 & 42.3 & $ 0.13\pm0.03$\
& $5601\pm33$ & $4.41\pm0.12$ & $ 0.99\pm0.05$ & $-0.08\pm0.05$ & 22.5 & 29.3 & $ 0.17\pm0.12$\
& $6299\pm41$ & $4.10\pm0.02$ & $ 1.42\pm0.09$ & $ 0.21\pm0.05$ & 48.2 & 70.5 & $-0.02\pm0.04$\
& $4804\pm77$ & $3.57\pm0.17$ & $ 1.14\pm0.09$ & $ 0.33\pm0.09$ & 23.9 & 25.0 & $ 0.15\pm0.24$\
& $5600\pm42$ & $4.44\pm0.08$ & $ 0.95\pm0.05$ & $ 0.27\pm0.05$ & 23.0 & 39.0 & $-0.08\pm0.05$\
& $6260\pm43$ & $4.43\pm0.05$ & $ 1.41\pm0.09$ & $ 0.22\pm0.05$ & 40.5 & 70.6 & $ 0.00\pm0.11$\
& $5772\pm25$ & $4.40\pm0.07$ & $ 1.07\pm0.04$ & $ 0.08\pm0.04$ & 23.8 & 37.0 & $-0.03\pm0.03$\
& $5845\pm22$ & $4.42\pm0.07$ & $ 1.10\pm0.03$ & $ 0.13\pm0.03$ & 26.2 & 39.1 & $-0.07\pm0.04$\
& $5325\pm00$ & $3.90\pm0.00$ & $ 1.11\pm0.00$ & $-0.26\pm0.00$ & 12.6 & 18.4 & $ 0.07\pm0.07$\
& $5584\pm36$ & $4.37\pm0.06$ & $ 1.07\pm0.05$ & $ 0.24\pm0.05$ & 30.5 & 41.0 & $ 0.05\pm0.09$\
& $5859\pm31$ & $4.32\pm0.07$ & $ 1.27\pm0.05$ & $ 0.09\pm0.04$ & 29.0 & 39.8 & $-0.03\pm0.07$\
& $5918\pm44$ & $4.35\pm0.13$ & $ 1.39\pm0.06$ & $ 0.22\pm0.05$ & 38.5 & 66.3 & $ 0.11\pm0.10$\
& $5752\pm53$ & $4.50\pm0.09$ & $ 1.01\pm0.06$ & $ 0.35\pm0.06$ & 27.1 & 37.6 & $-0.21\pm0.08$\
& $5532\pm00$ & $4.29\pm0.00$ & $ 1.03\pm0.00$ & $ 0.19\pm0.00$ & 27.7 & 38.2 & $ 0.05\pm0.07$\
& $5984\pm33$ & $4.25\pm0.10$ & $ 1.25\pm0.05$ & $ 0.17\pm0.05$ & 36.1 & 55.5 & $-0.00\pm0.04$\
& $5938\pm42$ & $4.12\pm0.05$ & $ 1.28\pm0.06$ & $ 0.24\pm0.05$ & 46.5 & 63.0 & $ 0.07\pm0.05$\
& $5887\pm32$ & $4.30\pm0.07$ & $ 1.31\pm0.04$ & $ 0.25\pm0.04$ & 37.0 & 55.9 & $ 0.01\pm0.03$\
& $5423\pm41$ & $4.40\pm0.13$ & $ 1.01\pm0.05$ & $ 0.26\pm0.04$ & 20.0 & 28.2 & $-0.11\pm0.09$\
& $5804\pm36$ & $4.42\pm0.07$ & $ 1.20\pm0.05$ & $ 0.20\pm0.05$ & 34.3 & 48.2 & $0.05\pm0.06$\
& $5645\pm00$ & $4.31\pm0.00$ & $ 1.06\pm0.00$ & $ 0.37\pm0.00$ & 29.3 & 44.0 & $-0.13\pm0.02$\
& $5732\pm31$ & $4.40\pm0.05$ & $ 0.99\pm0.04$ & $0.17\pm0.04$ & 19.8 & 29.2 & $-0.23\pm0.06$\
& $4916\pm70$ & $3.36\pm0.21$ & $ 1.27\pm0.06$ & $ 0.16\pm0.08$ & 12.6 & 20.7 & $-0.19\pm0.09$\
& $5843\pm38$ & $4.45\pm0.07$ & $ 1.03\pm0.06$ & $ 0.05\pm0.05$ & 22.2 & 38.8 & $-0.03\pm0.05$\
\[tab5\]
[lccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & EW$_{5380}$ & EW$_{5052}$ & \[C/Fe\]\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & (mÅ) & (mÅ) &\
& $ 5956\pm44$ & $4.39\pm0.13$ & $1.07\pm0.09$ & $-0.14\pm0.05$ & 18.8 & 30.7 & $-0.04\pm0.05$\
& $ 5878\pm53$ & $4.74\pm0.15$ & $1.13\pm0.10$ & $-0.03\pm0.06$ & 18.9 & 35.8 & $ 0.07\pm0.08$\
& $ 6140\pm41$ & $4.39\pm0.16$ & $1.50\pm0.08$ & $ 0.18\pm0.05$ & 39.9 & 62.2 & $ 0.03\pm0.07$\
& $ 5344\pm29$ & $4.57\pm0.09$ & $0.91\pm0.06$ & $-0.52\pm0.04$ & 7.0 & 12.6 & $ 0.33\pm0.04$\
& $ 5368\pm24$ & $4.55\pm0.05$ & $0.88\pm0.05$ & $-0.47\pm0.03$ & 7.5 & 10.2 & $ 0.22\pm0.12$\
& $ 5180\pm56$ & $4.44\pm0.13$ & $1.33\pm0.08$ & $ 0.06\pm0.07$ & 9.2 & 15.6 & $-0.10\pm0.07$\
& $ 6275\pm57$ & $4.40\pm0.37$ & $2.41\pm0.41$ & $-0.19\pm0.07$ & 29.1 & 45.0 & $ 0.08\pm0.13$\
& $ 5733\pm31$ & $4.55\pm0.10$ & $1.09\pm0.06$ & $-0.21\pm0.04$ & 14.0 & 26.3 & $ 0.10\pm0.05$\
& $ 5444\pm31$ & $4.47\pm0.07$ & $0.98\pm0.06$ & $-0.38\pm0.04$ & 10.1 & 18.0 & $ 0.25\pm0.03$\
& $ 5843\pm26$ & $4.47\pm0.10$ & $1.17\pm0.06$ & $-0.23\pm0.04$ & 16.2 & 30.4 & $ 0.11\pm0.06$\
& $ 5074\pm60$ & $3.77\pm0.16$ & $1.08\pm0.06$ & $ 0.13\pm0.08$ & 17.0 & 30.2 & $ 0.07\pm0.08$\
& $ 5176\pm45$ & $4.41\pm0.17$ & $1.03\pm0.06$ & $ 0.06\pm0.05$ & 8.5 & 12.7 & $-0.19\pm0.10$\
& $ 5126\pm34$ & $4.51\pm0.08$ & $0.60\pm0.07$ & $-0.31\pm0.04$ & 11.5 & 14.4 & $ 0.42\pm0.15$\
& $ 5868\pm30$ & $4.55\pm0.10$ & $1.24\pm0.05$ & $ 0.02\pm0.04$ & 18.4 & 31.6 & $-0.10\pm0.04$\
& $ 5479\pm37$ & $4.61\pm0.07$ & $1.12\pm0.05$ & $ 0.00\pm0.05$ & 11.6 & 25.0 & $-0.01\pm0.09$\
& $ 5752\pm32$ & $4.53\pm0.07$ & $1.26\pm0.07$ & $-0.23\pm0.05$ & 12.1 & 24.4 & $ 0.02\pm0.07$\
& $ 5633\pm35$ & $4.48\pm0.07$ & $1.24\pm0.05$ & $-0.01\pm0.04$ & 13.8 & 24.0 & $-0.12\pm0.03$\
& $ 5594\pm36$ & $4.41\pm0.09$ & $1.05\pm0.04$ & $ 0.10\pm0.05$ & 21.1 & 32.2 & $-0.01\pm0.06$\
& $ 5825\pm20$ & $4.37\pm0.10$ & $1.20\pm0.04$ & $-0.19\pm0.03$ & 19.3 & 33.1 & $ 0.12\pm0.04$\
& $ 5260\pm31$ & $4.35\pm0.11$ & $0.74\pm0.05$ & $-0.26\pm0.04$ & 11.0 & 14.8 & $ 0.18\pm0.13$\
& $ 5979\pm31$ & $4.59\pm0.12$ & $1.36\pm0.10$ & $-0.29\pm0.04$ & 21.0 & 35.6 & $ 0.25\pm0.05$\
& $ 5410\pm26$ & $4.38\pm0.07$ & $0.89\pm0.03$ & $-0.03\pm0.04$ & 14.2 & 24.2 & $ 0.05\pm0.03$\
& $ 5242\pm28$ & $4.50\pm0.09$ & $0.69\pm0.05$ & $-0.37\pm0.04$ & 8.5 & 17.4 & $ 0.35\pm0.05$\
& $ 5000\pm55$ & $4.55\pm0.13$ & $1.07\pm0.08$ & $-0.03\pm0.06$ & 13.7 & 20.9 & $ 0.40\pm0.09$\
& $ 5803\pm29$ & $4.50\pm0.08$ & $1.02\pm0.04$ & $ 0.14\pm0.04$ & 23.3 & 38.5 & $-0.07\pm0.03$\
& $ 6167\pm37$ & $4.35\pm0.10$ & $1.42\pm0.09$ & $-0.03\pm0.05$ & 33.7 & 50.9 & $ 0.07\pm0.04$\
& $ 5765\pm24$ & $4.52\pm0.05$ & $1.22\pm0.05$ & $-0.23\pm0.04$ & 17.5 & 25.7 & $ 0.13\pm0.06$\
& $ 5069\pm49$ & $4.38\pm0.19$ & $0.79\pm0.07$ & $-0.01\pm0.05$ & 16.7 & 18.6 & $ 0.28\pm0.21$\
& $ 5435\pm39$ & $4.48\pm0.08$ & $0.91\pm0.07$ & $-0.29\pm0.05$ & 11.5 & 15.8 & $ 0.15\pm0.12$\
& $ 5910\pm24$ & $4.42\pm0.05$ & $1.14\pm0.04$ & $ 0.00\pm0.04$ & 19.9 & 34.1 & $-0.10\pm0.03$\
& $ 5890\pm30$ & $4.51\pm0.07$ & $1.12\pm0.07$ & $-0.17\pm0.04$ & 18.4 & 30.4 & $ 0.07\pm0.03$\
\[tab6\]
[lcccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & \[S/Fe\]$_1$ & \[S/Fe\]$_2$ & \[S/Fe\]\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & & &\
& $6302\pm56$ & $4.34\pm0.13$ & $1.86\pm0.17$ & $ 0.14\pm0.07$ & $-$ & $-0.39$ & $ -0.39 \pm 0.09 $\
& $5536\pm50$ & $4.56\pm0.12$ & $1.33\pm0.06$ & $ 0.12\pm0.06$ & $-$ & $-0.22$ & $ -0.22 \pm 0.08 $\
& $5976\pm51$ & $4.45\pm0.10$ & $1.26\pm0.07$ & $ 0.32\pm0.06$ & $-$ & $-0.17$ & $ -0.17 \pm 0.08 $\
& $5173\pm35$ & $4.37\pm0.12$ & $0.74\pm0.05$ & $ 0.12\pm0.04$ & $ 0.13$ & $ 0.08$ & $ 0.11 \pm 0.08 $\
& $5636\pm40$ & $4.23\pm0.14$ & $1.12\pm0.05$ & $ 0.40\pm0.05$ & $-0.20$ & $-0.20$ & $ -0.20 \pm 0.08 $\
& $5626\pm32$ & $4.49\pm0.10$ & $0.95\pm0.06$ & $-0.24\pm0.04$ & $-0.16$ & $-0.11$ & $ -0.14 \pm 0.07 $\
& $5835\pm59$ & $4.60\pm0.12$ & $1.53\pm0.27$ & $-0.52\pm0.08$ & $-$ & $-0.18$ & $ -0.18 \pm 0.09 $\
& $6151\pm57$ & $4.51\pm0.10$ & $1.45\pm0.15$ & $ 0.06\pm0.07$ & $-0.11$ & $-0.11$ & $ -0.11 \pm 0.08 $\
& $6212\pm64$ & $4.26\pm0.13$ & $1.69\pm0.16$ & $ 0.13\pm0.08$ & $-0.33$ & $-$ & $ -0.33 \pm 0.09 $\
& $6143\pm31$ & $4.48\pm0.08$ & $1.40\pm0.08$ & $-0.03\pm0.04$ & $-0.12$ & $-0.12$ & $ -0.12 \pm 0.06 $\
& $5641\pm28$ & $4.05\pm0.05$ & $1.13\pm0.03$ & $ 0.14\pm0.04$ & $-0.19$ & $-$ & $ -0.19 \pm 0.06 $\
& $5702\pm36$ & $4.33\pm0.08$ & $1.05\pm0.04$ & $ 0.36\pm0.05$ & $-0.26$ & $-0.26$ & $ -0.26 \pm 0.07 $\
& $5163\pm00$ & $4.52\pm0.00$ & $0.72\pm0.00$ & $-0.24\pm0.00$ & $ 0.04$ & $-0.06$ & $ -0.01 \pm 0.07 $\
& $5801\pm30$ & $4.22\pm0.12$ & $1.34\pm0.04$ & $ 0.15\pm0.04$ & $-0.10$ & $-0.15$ & $ -0.12 \pm 0.08 $\
& $6252\pm53$ & $4.61\pm0.16$ & $1.18\pm0.10$ & $ 0.26\pm0.06$ & $-0.26$ & $-0.26$ & $ -0.26 \pm 0.09 $\
& $6190\pm00$ & $4.19\pm0.00$ & $1.54\pm0.00$ & $ 0.24\pm0.00$ & $-0.29$ & $-0.29$ & $ -0.29 \pm 0.05 $\
& $5073\pm42$ & $4.43\pm0.08$ & $1.05\pm0.06$ & $-0.13\pm0.05$ & $-0.02$ & $ 0.03$ & $ 0.00 \pm 0.08 $\
& $5959\pm46$ & $4.35\pm0.12$ & $1.20\pm0.10$ & $-0.11\pm0.06$ & $-0.24$ & $-0.24$ & $ -0.24 \pm 0.08 $\
& $6108\pm36$ & $4.25\pm0.10$ & $1.30\pm0.05$ & $ 0.31\pm0.05$ & $-0.26$ & $-0.21$ & $ -0.23 \pm 0.08 $\
& $5656\pm44$ & $4.45\pm0.08$ & $1.01\pm0.06$ & $ 0.22\pm0.05$ & $-0.12$ & $-0.07$ & $ -0.09 \pm 0.08 $\
& $5587\pm00$ & $4.29\pm0.00$ & $1.08\pm0.00$ & $ 0.38\pm0.00$ & $-$ & $-0.23$ & $ -0.23 \pm 0.05 $\
& $6046\pm49$ & $4.71\pm0.09$ & $1.79\pm0.19$ & $-0.08\pm0.06$ & $-0.12$ & $-0.12$ & $ -0.12 \pm 0.08 $\
& $5546\pm30$ & $4.50\pm0.03$ & $0.80\pm0.07$ & $-0.38\pm0.04$ & $ 0.13$ & $ 0.18$ & $ 0.16 \pm 0.07 $\
& $5674\pm40$ & $3.94\pm0.12$ & $1.38\pm0.05$ & $ 0.40\pm0.06$ & $-0.35$ & $-0.30$ & $ -0.33 \pm 0.08 $\
& $5991\pm27$ & $4.42\pm0.10$ & $1.24\pm0.04$ & $ 0.10\pm0.04$ & $-0.15$ & $-0.20$ & $ -0.18 \pm 0.07 $\
& $6145\pm42$ & $4.31\pm0.15$ & $1.29\pm0.09$ & $ 0.21\pm0.05$ & $-0.21$ & $-0.21$ & $ -0.21 \pm 0.08 $\
& $5268\pm55$ & $4.41\pm0.16$ & $0.97\pm0.06$ & $ 0.20\pm0.06$ & $ 0.05$ & $ 0.00$ & $ 0.03 \pm 0.10 $\
& $4554\pm85$ & $2.48\pm0.23$ & $1.82\pm0.08$ & $-0.54\pm0.12$ & $ 0.29$ & $ 0.24$ & $ 0.26 \pm 0.13 $\
& $5644\pm54$ & $4.37\pm0.07$ & $0.89\pm0.07$ & $ 0.33\pm0.06$ & $-0.23$ & $-0.18$ & $ -0.20 \pm 0.08 $\
& $6026\pm30$ & $4.41\pm0.13$ & $1.11\pm0.06$ & $ 0.01\pm0.04$ & $-0.31$ & $-0.36$ & $ -0.34 \pm 0.08 $\
& $6105\pm00$ & $4.28\pm0.00$ & $1.36\pm0.00$ & $ 0.23\pm0.00$ & $-0.23$ & $-0.18$ & $ -0.21 \pm 0.06 $\
& $5666\pm31$ & $4.53\pm0.09$ & $1.06\pm0.05$ & $-0.12\pm0.04$ & $-0.03$ & $-0.03$ & $ -0.03 \pm 0.07 $\
& $5988\pm52$ & $4.45\pm0.15$ & $1.25\pm0.08$ & $ 0.36\pm0.06$ & $-0.21$ & $-0.16$ & $ -0.19 \pm 0.09 $\
& $5995\pm45$ & $4.30\pm0.07$ & $1.42\pm0.09$ & $ 0.03\pm0.06$ & $-0.28$ & $-0.23$ & $ -0.26 \pm 0.08 $\
& $5518\pm49$ & $4.42\pm0.12$ & $1.22\pm0.06$ & $ 0.26\pm0.06$ & $-0.21$ & $-0.26$ & $ -0.24 \pm 0.09 $\
& $5699\pm49$ & $4.27\pm0.12$ & $1.26\pm0.06$ & $ 0.27\pm0.06$ & $-0.27$ & $-0.17$ & $ -0.22 \pm 0.09 $\
& $6112\pm39$ & $4.34\pm0.10$ & $1.38\pm0.07$ & $ 0.16\pm0.05$ & $-0.31$ & $-0.26$ & $ -0.29 \pm 0.08 $\
& $6143\pm53$ & $4.42\pm0.13$ & $1.53\pm0.09$ & $ 0.28\pm0.07$ & $-0.33$ & $-0.28$ & $ -0.31 \pm 0.09 $\
& $5279\pm62$ & $4.37\pm0.18$ & $0.98\pm0.07$ & $ 0.33\pm0.07$ & $-0.23$ & $-0.18$ & $ -0.21 \pm 0.10 $\
& $5737\pm34$ & $4.25\pm0.14$ & $1.18\pm0.04$ & $ 0.41\pm0.05$ & $-0.31$ & $-0.31$ & $ -0.31 \pm 0.08 $\
& $5574\pm72$ & $4.46\pm0.20$ & $1.14\pm0.09$ & $ 0.32\pm0.09$ & $-0.22$ & $-0.22$ & $ -0.22 \pm 0.11 $\
& $6015\pm00$ & $4.46\pm0.00$ & $1.13\pm0.00$ & $ 0.30\pm0.00$ & $-0.20$ & $-0.20$ & $ -0.20 \pm 0.05 $\
& $5454\pm61$ & $4.33\pm0.17$ & $1.08\pm0.08$ & $ 0.35\pm0.08$ & $-$ & $-0.15$ & $ -0.15 \pm 0.10 $\
& $5821\pm41$ & $4.45\pm0.06$ & $1.16\pm0.05$ & $ 0.32\pm0.05$ & $-$ & $-0.22$ & $ -0.22 \pm 0.07 $\
& $5954\pm25$ & $4.44\pm0.10$ & $1.30\pm0.04$ & $ 0.06\pm0.03$ & $-0.16$ & $-0.16$ & $ -0.16 \pm 0.06 $\
& $5899\pm35$ & $4.34\pm0.07$ & $1.08\pm0.06$ & $-0.01\pm0.05$ & $-$ & $-0.09$ & $ -0.09 \pm 0.07 $\
& $6248\pm42$ & $4.49\pm0.16$ & $1.35\pm0.08$ & $ 0.20\pm0.05$ & $-0.25$ & $-0.25$ & $ -0.25 \pm 0.08 $\
& $5494\pm26$ & $4.50\pm0.10$ & $0.84\pm0.05$ & $-0.36\pm0.04$ & $ 0.06$ & $ 0.11$ & $ 0.08 \pm 0.07 $\
& $5884\pm34$ & $4.22\pm0.22$ & $1.31\pm0.17$ & $-0.70\pm0.04$ & $ 0.15$ & $ 0.15$ & $ 0.15 \pm 0.09 $\
& $5098\pm36$ & $4.45\pm0.11$ & $0.74\pm0.05$ & $ 0.09\pm0.04$ & $-$ & $ 0.01$ & $ 0.01 \pm 0.07 $\
& $5560\pm34$ & $4.07\pm0.05$ & $1.18\pm0.05$ & $-0.06\pm0.05$ & $ 0.11$ & $-0.04$ & $ 0.04 \pm 0.10 $\
& $6339\pm73$ & $4.19\pm0.10$ & $1.70\pm0.16$ & $ 0.23\pm0.07$ & $-0.18$ & $-0.18$ & $ -0.18 \pm 0.09 $\
& $6075\pm40$ & $4.64\pm0.12$ & $1.31\pm0.07$ & $ 0.16\pm0.05$ & $-0.16$ & $-0.16$ & $ -0.16 \pm 0.08 $\
& $5392\pm36$ & $4.48\pm0.06$ & $0.85\pm0.05$ & $ 0.03\pm0.04$ & $-0.05$ & $-0.08$ & $ -0.06 \pm 0.07 $\
& $5776\pm29$ & $4.36\pm0.07$ & $1.09\pm0.04$ & $ 0.30\pm0.04$ & $-0.20$ & $-0.15$ & $ -0.18 \pm 0.07 $\
& $6222\pm39$ & $4.27\pm0.15$ & $1.79\pm0.12$ & $-0.04\pm0.05$ & $-0.21$ & $-0.21$ & $ -0.21 \pm 0.08 $\
\[tab7\]
[lccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & \[S/Fe\]$_1$ & \[S/Fe\]$_2$ & \[S/Fe\]\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & & &\
& $5909\pm39$ & $4.51\pm0.08$ & $1.13\pm0.06$ & $ 0.10\pm0.05$ & $-0.15$ & $-0.20$ & $ -0.18 \pm 0.07 $\
& $6045\pm44$ & $4.53\pm0.08$ & $1.12\pm0.07$ & $ 0.21\pm0.05$ & $-0.06$ & $-0.11$ & $ -0.09 \pm 0.08 $\
& $5853\pm25$ & $4.41\pm0.15$ & $1.35\pm0.07$ & $-0.21\pm0.04$ & $-0.04$ & $-0.04$ & $ -0.04 \pm 0.07 $\
& $5311\pm87$ & $4.42\pm0.18$ & $0.92\pm0.10$ & $ 0.43\pm0.08$ & $-0.13$ & $-0.13$ & $ -0.13 \pm 0.12 $\
& $5883\pm25$ & $4.51\pm0.05$ & $1.18\pm0.04$ & $ 0.06\pm0.04$ & $-$ & $-0.21$ & $ -0.21 \pm 0.06 $\
& $5798\pm33$ & $4.31\pm0.08$ & $1.19\pm0.04$ & $ 0.32\pm0.04$ & $-0.17$ & $-0.17$ & $ -0.17 \pm 0.07 $\
& $5617\pm35$ & $4.22\pm0.05$ & $1.21\pm0.05$ & $ 0.06\pm0.05$ & $-0.01$ & $-0.06$ & $ -0.04 \pm 0.07 $\
& $5601\pm33$ & $4.41\pm0.12$ & $0.99\pm0.05$ & $-0.08\pm0.05$ & $ 0.13$ & $-$ & $ 0.13 \pm 0.07 $\
& $6299\pm41$ & $4.10\pm0.02$ & $1.42\pm0.09$ & $ 0.21\pm0.05$ & $-0.31$ & $-0.31$ & $ -0.31 \pm 0.06 $\
& $5600\pm42$ & $4.44\pm0.08$ & $0.95\pm0.05$ & $ 0.27\pm0.05$ & $-0.12$ & $-0.07$ & $ -0.10 \pm 0.08 $\
& $6260\pm43$ & $4.43\pm0.05$ & $1.41\pm0.09$ & $ 0.22\pm0.05$ & $-0.12$ & $-0.17$ & $ -0.15 \pm 0.07 $\
& $5772\pm25$ & $4.40\pm0.07$ & $1.07\pm0.04$ & $ 0.08\pm0.04$ & $-0.13$ & $-0.13$ & $ -0.13 \pm 0.06 $\
& $5845\pm22$ & $4.42\pm0.07$ & $1.10\pm0.03$ & $ 0.13\pm0.03$ & $-0.28$ & $-0.18$ & $ -0.23 \pm 0.08 $\
& $5325\pm00$ & $3.90\pm0.00$ & $1.11\pm0.00$ & $-0.26\pm0.00$ & $ 0.06$ & $-0.04$ & $ 0.01 \pm 0.07 $\
& $5584\pm36$ & $4.37\pm0.06$ & $1.07\pm0.05$ & $ 0.24\pm0.05$ & $-0.14$ & $-0.14$ & $ -0.14 \pm 0.06 $\
& $4947\pm58$ & $4.51\pm0.20$ & $0.86\pm0.09$ & $-0.02\pm0.06$ & $-0.08$ & $-0.08$ & $ -0.08 \pm 0.10 $\
& $5859\pm31$ & $4.32\pm0.07$ & $1.27\pm0.05$ & $ 0.09\pm0.04$ & $-0.24$ & $-0.24$ & $ -0.24 \pm 0.06 $\
& $5752\pm53$ & $4.50\pm0.09$ & $1.01\pm0.06$ & $ 0.35\pm0.06$ & $-0.20$ & $-0.25$ & $ -0.23 \pm 0.08 $\
& $6117\pm26$ & $4.48\pm0.08$ & $1.40\pm0.06$ & $ 0.02\pm0.03$ & $-0.22$ & $-0.22$ & $ -0.22 \pm 0.06 $\
& $5532\pm00$ & $4.29\pm0.00$ & $1.03\pm0.00$ & $ 0.19\pm0.00$ & $-0.04$ & $-0.09$ & $ -0.07 \pm 0.06 $\
& $5984\pm33$ & $4.25\pm0.10$ & $1.25\pm0.05$ & $ 0.17\pm0.05$ & $-0.27$ & $-$ & $ -0.27 \pm 0.07 $\
& $5938\pm42$ & $4.12\pm0.05$ & $1.28\pm0.06$ & $ 0.24\pm0.05$ & $-$ & $-0.14$ & $ -0.14 \pm 0.07 $\
& $5887\pm32$ & $4.30\pm0.07$ & $1.31\pm0.04$ & $ 0.25\pm0.04$ & $-$ & $-0.15$ & $ -0.15 \pm 0.06 $\
& $5423\pm41$ & $4.40\pm0.13$ & $1.01\pm0.05$ & $ 0.26\pm0.04$ & $-0.01$ & $-0.01$ & $ -0.01 \pm 0.08 $\
& $5804\pm36$ & $4.42\pm0.07$ & $1.20\pm0.05$ & $ 0.20\pm0.05$ & $-0.15$ & $-0.10$ & $ -0.13 \pm 0.07 $\
& $5645\pm00$ & $4.31\pm0.00$ & $1.06\pm0.00$ & $ 0.37\pm0.00$ & $-0.27$ & $-0.22$ & $ -0.25 \pm 0.06 $\
& $4916\pm70$ & $3.36\pm0.21$ & $1.27\pm0.06$ & $ 0.16\pm0.08$ & $-$ & $-0.11$ & $ -0.11 \pm 0.11 $\
& $5843\pm38$ & $4.45\pm0.07$ & $1.03\pm0.06$ & $ 0.05\pm0.05$ & $-0.25$ & $-$ & $ -0.25 \pm 0.07 $\
\[tab8\]
[lccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & \[S/Fe\]$_1$ & \[S/Fe\]$_2$ & \[S/Fe\]\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & & &\
& $5956\pm44$ & $4.39\pm0.13$ & $1.07\pm0.09$ & $-0.14\pm0.05$ & $-0.16$ & $-0.11$ & $ -0.28 \pm 0.08 $\
& $5878\pm53$ & $4.74\pm0.15$ & $1.13\pm0.10$ & $-0.03\pm0.06$ & $-$ & $-0.02$ & $ -0.02 \pm 0.09 $\
& $4911\pm54$ & $4.49\pm0.18$ & $0.71\pm0.11$ & $-0.17\pm0.06$ & $-0.23$ & $-0.23$ & $ -0.23 \pm 0.09 $\
& $6140\pm41$ & $4.39\pm0.16$ & $1.50\pm0.08$ & $ 0.18\pm0.05$ & $-0.18$ & $-0.18$ & $ -0.18 \pm 0.08 $\
& $4970\pm40$ & $4.49\pm0.10$ & $0.76\pm0.07$ & $-0.26\pm0.04$ & $-$ & $ 0.06$ & $ 0.06 \pm 0.07 $\
& $5344\pm29$ & $4.57\pm0.09$ & $0.91\pm0.06$ & $-0.52\pm0.04$ & $ 0.02$ & $ 0.12$ & $ 0.07 \pm 0.08 $\
& $5368\pm24$ & $4.55\pm0.05$ & $0.88\pm0.05$ & $-0.47\pm0.03$ & $ 0.07$ & $ 0.07$ & $ 0.07 \pm 0.06 $\
& $5180\pm56$ & $4.44\pm0.13$ & $1.33\pm0.08$ & $ 0.06\pm0.07$ & $-0.16$ & $-0.11$ & $ -0.14 \pm 0.09 $\
& $6275\pm57$ & $4.40\pm0.37$ & $2.41\pm0.41$ & $-0.19\pm0.06$ & $ 0.04$ & $-0.01$ & $ 0.02 \pm 0.15 $\
& $5733\pm31$ & $4.55\pm0.10$ & $1.09\pm0.06$ & $-0.21\pm0.04$ & $-$ & $-0.09$ & $ -0.09 \pm 0.07 $\
& $5444\pm31$ & $4.47\pm0.07$ & $0.98\pm0.06$ & $-0.38\pm0.04$ & $-$ & $ 0.03$ & $ 0.03 \pm 0.06 $\
& $5843\pm26$ & $4.47\pm0.10$ & $1.17\pm0.06$ & $-0.23\pm0.04$ & $-0.07$ & $-0.07$ & $ -0.07 \pm 0.07 $\
& $5074\pm60$ & $3.77\pm0.16$ & $1.08\pm0.06$ & $ 0.13\pm0.08$ & $-$ & $-0.18$ & $ -0.18 \pm 0.09 $\
& $5176\pm45$ & $4.41\pm0.17$ & $1.03\pm0.06$ & $ 0.06\pm0.05$ & $-$ & $-0.16$ & $ -0.16 \pm 0.09 $\
& $5868\pm30$ & $4.55\pm0.10$ & $1.24\pm0.05$ & $ 0.02\pm0.04$ & $-0.12$ & $-0.12$ & $ -0.12 \pm 0.07 $\
& $5479\pm37$ & $4.61\pm0.07$ & $1.12\pm0.05$ & $ 0.00\pm0.05$ & $-$ & $-0.05$ & $ -0.05 \pm 0.07 $\
& $5752\pm32$ & $4.53\pm0.07$ & $1.26\pm0.07$ & $-0.23\pm0.05$ & $-0.02$ & $-0.02$ & $ -0.02 \pm 0.06 $\
& $5633\pm35$ & $4.48\pm0.07$ & $1.24\pm0.05$ & $-0.01\pm0.04$ & $-0.19$ & $-0.14$ & $ -0.17 \pm 0.07 $\
& $5594\pm36$ & $4.41\pm0.09$ & $1.05\pm0.04$ & $ 0.10\pm0.10$ & $-0.05$ & $-0.05$ & $ -0.05 \pm 0.07 $\
& $5825\pm20$ & $4.37\pm0.10$ & $1.20\pm0.04$ & $-0.19\pm0.03$ & $-0.01$ & $-0.01$ & $ -0.01 \pm 0.06 $\
& $5260\pm31$ & $4.35\pm0.11$ & $0.74\pm0.05$ & $-0.26\pm0.04$ & $ 0.16$ & $ 0.16$ & $ 0.16 \pm 0.07 $\
& $5979\pm31$ & $4.59\pm0.12$ & $1.36\pm0.10$ & $-0.29\pm0.04$ & $-$ & $ 0.04$ & $ 0.04 \pm 0.07 $\
& $5410\pm26$ & $4.38\pm0.07$ & $0.89\pm0.03$ & $-0.03\pm0.04$ & $-0.12$ & $-0.12$ & $ -0.12 \pm 0.06 $\
& $5000\pm55$ & $4.55\pm0.13$ & $1.07\pm0.08$ & $-0.03\pm0.06$ & $-$ & $-0.07$ & $ -0.07 \pm 0.09 $\
& $5803\pm29$ & $4.50\pm0.08$ & $1.02\pm0.04$ & $ 0.14\pm0.04$ & $-0.04$ & $-0.04$ & $ -0.04 \pm 0.06 $\
& $6167\pm37$ & $4.35\pm0.10$ & $1.42\pm0.09$ & $-0.03\pm0.05$ & $-0.07$ & $-0.07$ & $ -0.07 \pm 0.07 $\
& $5765\pm24$ & $4.52\pm0.05$ & $1.22\pm0.05$ & $-0.23\pm0.04$ & $-$ & $-0.07$ & $ -0.07 \pm 0.06 $\
& $5069\pm49$ & $4.38\pm0.19$ & $0.79\pm0.07$ & $-0.01\pm0.05$ & $-$ & $-0.29$ & $ -0.29 \pm 0.09 $\
& $5435\pm39$ & $4.48\pm0.08$ & $0.91\pm0.07$ & $-0.29\pm0.05$ & $-$ & $-0.01$ & $ -0.01 \pm 0.07 $\
& $5910\pm24$ & $4.42\pm0.05$ & $1.14\pm0.04$ & $ 0.00\pm0.04$ & $-0.05$ & $-0.05$ & $ -0.05 \pm 0.06 $\
& $5890\pm30$ & $4.51\pm0.07$ & $1.12\pm0.07$ & $-0.17\pm0.04$ & $-0.03$ & $-0.03$ & $ -0.03 \pm 0.06 $\
\[tab9\]
[lcccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & \[Zn/Fe\]$_1$ & \[Zn/Fe\]$_2$ & \[Zn/Fe\]\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & & & &\
&$6302\pm56 $&$ 4.34\pm0.13 $&$1.86\pm0.17 $&$ 0.14\pm0.07 $&$-0.25 $&$ -0.16 $&$ -0.21 \pm 0.10 $\
&$5976\pm51 $&$ 4.45\pm0.10 $&$1.26\pm0.07 $&$ 0.32\pm0.06 $&$-0.05 $&$ -0.03 $&$ -0.04 \pm 0.07 $\
&$5636\pm40 $&$ 4.23\pm0.14 $&$1.12\pm0.05 $&$ 0.40\pm0.05 $&$0.00 $&$ 0.02 $&$ 0.01 \pm 0.07 $\
&$5626\pm32 $&$ 4.49\pm0.10 $&$0.95\pm0.06 $&$ -0.24\pm0.04 $&$0.12 $&$ 0.22 $&$ 0.17 \pm 0.08 $\
&$5835\pm59 $&$ 4.60\pm0.12 $&$1.53\pm0.27 $&$ -0.52\pm0.08 $&$0.05 $&$ 0.07 $&$ 0.06 \pm 0.10 $\
&$6151\pm57 $&$ 4.51\pm0.10 $&$1.45\pm0.15 $&$ 0.06\pm0.07 $&$-0.17 $&$ -0.08 $&$ -0.13 \pm 0.10 $\
&$6212\pm64 $&$ 4.26\pm0.13 $&$1.69\pm0.16 $&$ 0.13\pm0.08 $&$-0.15 $&$ -0.08 $&$ -0.12 \pm 0.10 $\
&$6143\pm31 $&$ 4.48\pm0.08 $&$1.40\pm0.08 $&$ -0.03\pm0.04 $&$-0.10 $&$ -0.16 $&$ -0.13 \pm 0.07 $\
&$5641\pm28 $&$ 4.05\pm0.05 $&$1.13\pm0.03 $&$ 0.14\pm0.04 $&$0.08 $&$ 0.12 $&$ 0.10 \pm 0.06 $\
&$5702\pm36 $&$ 4.33\pm0.08 $&$1.05\pm0.04 $&$ 0.36\pm0.05 $&$0.08 $&$ 0.17 $&$ 0.13 \pm 0.08 $\
&$5163\pm00 $&$ 4.52\pm0.00 $&$0.72\pm0.00 $&$ -0.24\pm0.00 $&$0.20 $&$ 0.21 $&$ 0.21 \pm 0.06 $\
&$5801\pm30 $&$ 4.22\pm0.12 $&$1.34\pm0.04 $&$ 0.15\pm0.04 $&$-0.05 $&$ -0.08 $&$ -0.07 \pm 0.07 $\
&$6252\pm53 $&$ 4.61\pm0.16 $&$1.18\pm0.10 $&$ 0.26\pm0.06 $&$-0.04 $&$ -0.08 $&$ -0.06 \pm 0.08 $\
&$6190\pm00 $&$ 4.19\pm0.00 $&$1.54\pm0.00 $&$ 0.24\pm0.00 $&$-0.15 $&$ -0.08 $&$ -0.12 \pm 0.09 $\
&$6138\pm79 $&$ 4.53\pm0.22 $&$1.22\pm0.16 $&$ 0.17\pm0.10 $&$-0.10 $&$ -0.08 $&$ -0.09 \pm 0.11 $\
&$5073\pm42 $&$ 4.43\pm0.08 $&$1.05\pm0.06 $&$ -0.13\pm0.05 $&$-0.15 $&$ -0.08 $&$ -0.12 \pm 0.08 $\
&$5959\pm46 $&$ 4.35\pm0.12 $&$1.20\pm0.10 $&$ -0.11\pm0.06 $&$-0.10 $&$ -0.05 $&$ -0.08 \pm 0.08 $\
&$6108\pm36 $&$ 4.25\pm0.10 $&$1.30\pm0.05 $&$ 0.31\pm0.05 $&$-0.05 $&$ -0.08 $&$ -0.07 \pm 0.07 $\
&$4825\pm107$&$ 3.55\pm0.32 $&$1.18\pm0.12 $&$ 0.39\pm0.13 $&$-0.18 $&$ -0.08 $&$ -0.13 \pm 0.15 $\
&$5656\pm44 $&$ 4.45\pm0.08 $&$1.01\pm0.06 $&$ 0.22\pm0.05 $&$0.10 $&$ 0.07 $&$ 0.09 \pm 0.07 $\
&$5587\pm00 $&$ 4.29\pm0.00 $&$1.08\pm0.00 $&$ 0.38\pm0.00 $&$0.05 $&$ 0.07 $&$ 0.06 \pm 0.07 $\
&$6046\pm49 $&$ 4.71\pm0.09 $&$1.79\pm0.19 $&$ -0.08\pm0.06 $&$-0.32 $&$ -0.25 $&$ -0.29 \pm 0.09 $\
&$5546\pm30 $&$ 4.50\pm0.03 $&$0.80\pm0.07 $&$ -0.38\pm0.04 $&$0.15 $&$ 0.32 $&$ 0.24 \pm 0.11 $\
&$5674\pm40 $&$ 3.94\pm0.12 $&$1.38\pm0.05 $&$ 0.40\pm0.06 $&$-0.17 $&$ -0.10 $&$ -0.14 \pm 0.08 $\
&$5991\pm27 $&$ 4.42\pm0.10 $&$1.24\pm0.04 $&$ 0.10\pm0.04 $&$ 0.00 $&$ -0.03 $&$ -0.02 \pm 0.07 $\
&$5115\pm68 $&$ 4.32\pm0.19 $&$0.93\pm0.09 $&$ 0.04\pm0.07 $&$-0.15 $&$ 0.00 $&$ -0.08 \pm 0.12 $\
&$5268\pm55 $&$ 4.41\pm0.16 $&$0.97\pm0.06 $&$ 0.20\pm0.06 $&$0.10 $&$ 0.17 $&$ 0.14 \pm 0.09 $\
&$5644\pm54 $&$ 4.37\pm0.07 $&$0.89\pm0.07 $&$ 0.33\pm0.06 $&$-0.05 $&$ -0.01 $&$ -0.03 \pm 0.07 $\
&$6026\pm30 $&$ 4.41\pm0.13 $&$1.11\pm0.06 $&$ 0.01\pm0.04 $&$0.10 $&$ 0.12 $&$ 0.11 \pm 0.07 $\
&$6105\pm00 $&$ 4.28\pm0.00 $&$1.36\pm0.00 $&$ 0.23\pm0.00 $&$-0.07 $&$ -0.10 $&$ -0.09 \pm 0.07 $\
&$5608\pm34 $&$ 4.35\pm0.08 $&$1.24\pm0.04 $&$ 0.26\pm0.04 $&$-0.10 $&$ -0.04 $&$ -0.07 \pm 0.07 $\
&$6112\pm39 $&$ 4.34\pm0.10 $&$1.38\pm0.07 $&$ 0.16\pm0.05 $&$-0.05 $&$ -0.03 $&$ -0.04 \pm 0.07 $\
&$5279\pm62 $&$ 4.37\pm0.18 $&$0.98\pm0.07 $&$ 0.33\pm0.07 $&$0.15 $&$ 0.17 $&$ 0.16 \pm 0.09 $\
&$5574\pm72 $&$ 4.46\pm0.20 $&$1.14\pm0.09 $&$ 0.32\pm0.09 $&$0.05 $&$ 0.02 $&$ 0.04 \pm 0.10 $\
&$5821\pm41 $&$ 4.45\pm0.06 $&$1.16\pm0.05 $&$ 0.32\pm0.05 $&$0.10 $&$ 0.12 $&$ 0.11 \pm 0.07 $\
&$5954\pm25 $&$ 4.44\pm0.10 $&$1.30\pm0.04 $&$ 0.06\pm0.03 $&$-0.03 $&$ 0.02 $&$ 0.00 \pm 0.07 $\
&$5899\pm35 $&$ 4.34\pm0.07 $&$1.08\pm0.06 $&$ -0.01\pm0.05 $&$-0.10 $&$ -0.16 $&$ -0.13 \pm 0.07 $\
&$6248\pm42 $&$ 4.49\pm0.16 $&$1.35\pm0.08 $&$ 0.20\pm0.05 $&$-0.15 $&$ -0.13 $&$ -0.14 \pm 0.08 $\
&$4804\pm61 $&$ 4.36\pm0.28 $&$0.57\pm0.12 $&$ -0.08\pm0.06 $&$0.30 $&$ 0.12 $&$ 0.21 \pm 0.13 $\
&$5098\pm36 $&$ 4.45\pm0.11 $&$0.74\pm0.05 $&$ 0.09\pm0.04 $&$0.15 $&$ 0.20 $&$ 0.18 \pm 0.07 $\
&$6075\pm40 $&$ 4.64\pm0.12 $&$1.31\pm0.07 $&$ 0.16\pm0.05 $&$0.04 $&$ 0.06 $&$ 0.05 \pm 0.07 $\
&$5392\pm36 $&$ 4.48\pm0.06 $&$0.85\pm0.05 $&$ 0.03\pm0.04 $&$-0.05 $&$ 0.17 $&$ 0.06 \pm 0.13 $\
&$5776\pm29 $&$ 4.36\pm0.07 $&$1.09\pm0.04 $&$ 0.30\pm0.04 $&$0.15 $&$ 0.22 $&$ 0.19 \pm 0.07 $\
&$6222\pm39 $&$ 4.27\pm0.15 $&$1.79\pm0.12 $&$ -0.04\pm0.05 $&$-0.25 $&$ -0.15 $&$ -0.20 \pm 0.09 $\
&$4775\pm110$&$ 3.09\pm0.40 $&$1.78\pm0.11 $&$ 0.13\pm0.14 $&$-0.36 $&$ -0.26 $&$ -0.31 \pm 0.16 $\
&$5909\pm39 $&$ 4.51\pm0.08 $&$1.13\pm0.06 $&$ 0.10\pm0.05 $&$-0.05 $&$ 0.02 $&$ -0.02 \pm 0.08 $\
&$5853\pm25 $&$ 4.41\pm0.15 $&$1.35\pm0.07 $&$ -0.21\pm0.04 $&$0.07 $&$ 0.12 $&$ 0.10 \pm 0.08 $\
&$5311\pm87 $&$ 4.42\pm0.18 $&$0.92\pm0.10 $&$ 0.43\pm0.08 $&$0.10 $&$ 0.22 $&$ 0.16 \pm 0.11 $\
&$5883\pm25 $&$ 4.51\pm0.05 $&$1.18\pm0.04 $&$ 0.06\pm0.04 $&$-0.10 $&$ -0.08 $&$ -0.09 \pm 0.06 $\
&$5798\pm33 $&$ 4.31\pm0.08 $&$1.19\pm0.04 $&$ 0.32\pm0.04 $&$0.00 $&$ 0.00 $&$ 0.00 \pm 0.06 $\
&$5617\pm35 $&$ 4.22\pm0.05 $&$1.21\pm0.05 $&$ 0.06\pm0.05 $&$0.15 $&$ 0.17 $&$ 0.16 \pm 0.07 $\
&$5601\pm33 $&$ 4.41\pm0.12 $&$0.99\pm0.05 $&$ -0.08\pm0.05 $&$0.10 $&$ 0.17 $&$ 0.14 \pm 0.08 $\
&$4804\pm77 $&$ 3.57\pm0.17 $&$1.14\pm0.09 $&$ 0.33\pm0.09 $&$-0.08 $&$ 0.10 $&$ 0.01 \pm 0.13 $\
&$5588\pm115$&$ 4.46\pm0.20 $&$0.82\pm0.14 $&$ 0.24\pm0.10 $&$0.01 $&$ 0.03 $&$ 0.02 \pm 0.11 $\
&$6260\pm43 $&$ 4.43\pm0.05 $&$1.41\pm0.09 $&$ 0.22\pm0.05 $&$-0.39 $&$ -0.38 $&$ -0.39 \pm 0.07 $\
&$5772\pm25 $&$ 4.40\pm0.07 $&$1.07\pm0.04 $&$ 0.08\pm0.04 $&$0.10 $&$ 0.12 $&$ 0.11 \pm 0.06 $\
\[tab10\]
[lccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & \[Zn/Fe\]$_1$ & \[Zn/Fe\]$_2$ & \[Zn/Fe\]\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & & &\
&$5845\pm22 $&$ 4.42\pm0.07 $&$1.10\pm0.03 $&$ 0.13\pm0.03 $&$0.10 $&$ 0.05 $&$ 0.08 \pm 0.06 $\
&$5584\pm36 $&$ 4.37\pm0.06 $&$1.07\pm0.05 $&$ 0.24\pm0.05 $&$0.05 $&$ 0.07 $&$ 0.06 \pm 0.07 $\
&$4947\pm58 $&$ 4.51\pm0.20 $&$0.86\pm0.09 $&$ -0.02\pm0.06 $&$0.06 $&$ 0.07 $&$ 0.07 \pm 0.09 $\
&$5859\pm31 $&$ 4.32\pm0.07 $&$1.27\pm0.05 $&$ 0.09\pm0.04 $&$-0.05 $&$ -0.08 $&$ -0.07 \pm 0.06 $\
&$5918\pm44 $&$ 4.35\pm0.13 $&$1.39\pm0.06 $&$ 0.22\pm0.05 $&$0.00 $&$ 0.02 $&$ 0.01 \pm 0.07 $\
&$5532\pm00 $&$ 4.29\pm0.00 $&$1.03\pm0.00 $&$ 0.19\pm0.00 $&$0.05 $&$ 0.12 $&$ 0.09 \pm 0.07 $\
&$5984\pm33 $&$ 4.25\pm0.10 $&$1.25\pm0.05 $&$ 0.17\pm0.05 $&$-0.05 $&$ -0.03 $&$ -0.04 \pm 0.07 $\
&$5938\pm42 $&$ 4.12\pm0.05 $&$1.28\pm0.06 $&$ 0.24\pm0.05 $&$-0.05 $&$ -0.03 $&$ -0.04 \pm 0.07 $\
&$5887\pm32 $&$ 4.30\pm0.07 $&$1.31\pm0.04 $&$ 0.25\pm0.04 $&$0.00 $&$ 0.02 $&$ 0.01 \pm 0.06 $\
&$5804\pm36 $&$ 4.42\pm0.07 $&$1.20\pm0.05 $&$ 0.20\pm0.05 $&$-0.02 $&$ -0.01 $&$ -0.02 \pm 0.07 $\
&$5645\pm00 $&$ 4.31\pm0.00 $&$1.06\pm0.00 $&$ 0.37\pm0.00 $&$0.00 $&$ 0.07 $&$ 0.04 \pm 0.07 $\
&$5843\pm38 $&$ 4.45\pm0.07 $&$1.03\pm0.06 $&$ 0.05\pm0.05 $&$-0.05 $&$ -0.03 $&$ -0.04 \pm 0.07 $\
\[tab11\]
[lccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & \[Zn/Fe\]$_1$ & \[Zn/Fe\]$_2$ & \[Zn/Fe\]\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & & &\
&$5956\pm44 $&$ 4.39\pm0.13 $&$1.07\pm0.09 $&$ -0.14\pm0.05 $&$0.09 $&$ 0.06 $&$ 0.08 \pm 0.08 $\
&$5878\pm53 $&$ 4.74\pm0.15 $&$1.13\pm0.10 $&$ -0.03\pm0.06 $&$-0.05 $&$ -0.08 $&$ -0.07 \pm 0.08 $\
&$4911\pm54 $&$ 4.49\pm0.18 $&$0.71\pm0.11 $&$ -0.17\pm0.06 $&$-0.05 $&$ -0.03 $&$ -0.04 \pm 0.08 $\
&$6140\pm41 $&$ 4.39\pm0.16 $&$1.50\pm0.08 $&$ 0.18\pm0.05 $&$-0.05 $&$ -0.13 $&$ -0.09 \pm 0.08 $\
&$4970\pm40 $&$ 4.49\pm0.10 $&$0.76\pm0.07 $&$ -0.26\pm0.04 $&$0.00 $&$ 0.05 $&$ 0.03 \pm 0.07 $\
&$5344\pm29 $&$ 4.57\pm0.09 $&$0.91\pm0.06 $&$ -0.52\pm0.04 $&$0.15 $&$ 0.19 $&$ 0.17 \pm 0.07 $\
&$5368\pm24 $&$ 4.55\pm0.05 $&$0.88\pm0.05 $&$ -0.47\pm0.03 $&$0.03 $&$ 0.09 $&$ 0.06 \pm 0.06 $\
&$5180\pm56 $&$ 4.44\pm0.13 $&$1.33\pm0.08 $&$ 0.06\pm0.07 $&$-0.10 $&$ -0.11 $&$ -0.11 \pm 0.08 $\
&$6275\pm57 $&$ 4.40\pm0.37 $&$2.41\pm0.41 $&$ -0.19\pm0.06 $&$-0.35 $&$ -0.28 $&$ -0.32 \pm 0.14 $\
&$5733\pm31 $&$ 4.55\pm0.10 $&$1.09\pm0.06 $&$ -0.21\pm0.04 $&$0.00 $&$ 0.02 $&$ 0.01 \pm 0.06 $\
&$5444\pm31 $&$ 4.47\pm0.07 $&$0.98\pm0.06 $&$ -0.38\pm0.04 $&$0.20 $&$ 0.20 $&$ 0.20 \pm 0.06 $\
&$5843\pm26 $&$ 4.47\pm0.10 $&$1.17\pm0.06 $&$ -0.23\pm0.04 $&$-0.05 $&$ -0.05 $&$ -0.05 \pm 0.06 $\
&$5074\pm60 $&$ 3.77\pm0.16 $&$1.08\pm0.06 $&$ 0.13\pm0.08 $&$-0.05 $&$ 0.07 $&$ 0.01 \pm 0.11 $\
&$4975\pm55 $&$ 4.48\pm0.16 $&$0.77\pm0.09 $&$ -0.11\pm0.06 $&$-0.12 $&$ 0.03 $&$ -0.05 \pm 0.11 $\
&$5176\pm45 $&$ 4.41\pm0.17 $&$1.03\pm0.06 $&$ 0.06\pm0.05 $&$-0.08 $&$ -0.03 $&$ -0.06 \pm 0.08 $\
&$5126\pm34 $&$ 4.51\pm0.08 $&$0.60\pm0.07 $&$ -0.31\pm0.04 $&$0.30 $&$ 0.22 $&$ 0.26 \pm 0.07 $\
&$5868\pm30 $&$ 4.55\pm0.10 $&$1.24\pm0.05 $&$ 0.02\pm0.04 $&$-0.13 $&$ -0.08 $&$ -0.11 \pm 0.07 $\
&$5479\pm37 $&$ 4.61\pm0.07 $&$1.12\pm0.05 $&$ -0.00\pm0.05 $&$-0.08 $&$ -0.15 $&$ -0.12 \pm 0.08 $\
&$5752\pm32 $&$ 4.53\pm0.07 $&$1.26\pm0.07 $&$ -0.23\pm0.05 $&$-0.05 $&$ -0.08 $&$ -0.07 \pm 0.07 $\
&$4805\pm52 $&$ 4.37\pm0.37 $&$0.49\pm0.12 $&$ -0.30\pm0.05 $&$0.10 $&$ 0.12 $&$ 0.11 \pm 0.11 $\
&$5633\pm35 $&$ 4.48\pm0.07 $&$1.24\pm0.05 $&$ -0.01\pm0.04 $&$-0.10 $&$ -0.08 $&$ -0.09 \pm 0.06 $\
&$5594\pm36 $&$ 4.41\pm0.09 $&$1.05\pm0.04 $&$ 0.10\pm0.05 $&$0.05 $&$ 0.07 $&$ 0.06 \pm 0.07 $\
&$4658\pm56 $&$ 4.32\pm0.24 $&$0.64\pm0.15 $&$ -0.04\pm0.07 $&$0.00 $&$ 0.07 $&$ 0.04 \pm 0.10 $\
&$5825\pm20 $&$ 4.37\pm0.10 $&$1.20\pm0.04 $&$ -0.19\pm0.03 $&$0.00 $&$ 0.02 $&$ 0.01 \pm 0.06 $\
&$5260\pm31 $&$ 4.35\pm0.11 $&$0.74\pm0.05 $&$ -0.26\pm0.04 $&$0.15 $&$ 0.22 $&$ 0.19 \pm 0.08 $\
&$5979\pm31 $&$ 4.59\pm0.12 $&$1.36\pm0.10 $&$ -0.29\pm0.04 $&$0.05 $&$ 0.17 $&$ 0.11 \pm 0.09 $\
&$5410\pm26 $&$ 4.38\pm0.07 $&$0.89\pm0.03 $&$ -0.03\pm0.04 $&$0.03 $&$ 0.07 $&$ 0.05 \pm 0.06 $\
&$5242\pm28 $&$ 4.50\pm0.09 $&$0.69\pm0.05 $&$ -0.37\pm0.04 $&$0.10 $&$ 0.09 $&$ 0.10 \pm 0.06 $\
&$5000\pm55 $&$ 4.55\pm0.13 $&$1.07\pm0.08 $&$ -0.03\pm0.06 $&$-0.10 $&$ -0.08 $&$ -0.09 \pm 0.08 $\
&$5803\pm29 $&$ 4.50\pm0.08 $&$1.02\pm0.04 $&$ 0.14\pm0.04 $&$0.07 $&$ 0.07 $&$ 0.07 \pm 0.06 $\
&$6167\pm37 $&$ 4.35\pm0.10 $&$1.42\pm0.09 $&$ -0.03\pm0.05 $&$-0.10 $&$ -0.08 $&$ -0.09 \pm 0.07 $\
&$5765\pm24 $&$ 4.52\pm0.05 $&$1.22\pm0.05 $&$ -0.23\pm0.04 $&$0.00 $&$ 0.02 $&$ 0.01 \pm 0.06 $\
&$5005\pm45 $&$ 4.38\pm0.25 $&$0.67\pm0.09 $&$ -0.55\pm0.06 $&$0.25 $&$ 0.17 $&$ 0.21 \pm 0.10 $\
&$5069\pm49 $&$ 4.38\pm0.19 $&$0.79\pm0.07 $&$ -0.01\pm0.05 $&$0.10 $&$ 0.17 $&$ 0.14 \pm 0.09 $\
&$5435\pm39 $&$ 4.48\pm0.08 $&$0.91\pm0.07 $&$ -0.29\pm0.05 $&$0.10 $&$ 0.07 $&$ 0.09 \pm 0.07 $\
&$5910\pm24 $&$ 4.42\pm0.05 $&$1.14\pm0.04 $&$ 0.00\pm0.04 $&$-0.08 $&$ -0.13 $&$ -0.11 \pm 0.07 $\
&$4629\pm77 $&$ 4.36\pm0.19 $&$0.42\pm0.25 $&$ -0.06\pm0.08 $&$0.10 $&$ 0.09 $&$ 0.10 \pm 0.11 $\
&$5890\pm30 $&$ 4.51\pm0.07 $&$1.12\pm0.07 $&$ -0.17\pm0.04 $&$-0.05 $&$ -0.08 $&$ -0.07 \pm 0.07 $\
&$4555\pm87 $&$ 4.53\pm0.26 $&$0.66\pm0.28 $&$ -0.01\pm0.09 $&$0.04 $&$ 0.03 $&$ 0.04 \pm 0.12 $\
&$4747\pm58 $&$ 4.48\pm0.22 $&$0.40\pm0.20 $&$ -0.31\pm0.06 $&$0.26 $&$ 0.18 $&$ 0.22 \pm 0.10 $\
&$5260\pm41 $&$ 4.45\pm0.11 $&$0.92\pm0.06 $&$ -0.16\pm0.05 $&$0.04 $&$ 0.03 $&$ 0.04 \pm 0.07 $\
\[tab12\]
[lccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & \[Cu/Fe\]$_1$ & \[Cu/Fe\]$_2$ & \[Cu/Fe\]\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & & &\
&$6302\pm56 $&$ 4.34\pm0.13 $&$1.86\pm0.17 $&$ 0.14\pm0.07 $&$-0.06 $&$ 0.04 $&$-0.01 \pm0.10$\
&$5976\pm51 $&$ 4.45\pm0.10 $&$1.26\pm0.07 $&$ 0.32\pm0.06 $&$ 0.08 $&$ 0.12 $&$ 0.10 \pm0.08$\
&$5636\pm40 $&$ 4.23\pm0.14 $&$1.12\pm0.05 $&$ 0.40\pm0.05 $&$ 0.20 $&$ 0.10 $&$ 0.15 \pm0.08$\
&$5626\pm32 $&$ 4.49\pm0.10 $&$0.95\pm0.06 $&$ -0.24\pm0.04 $&$ 0.06 $&$ 0.07 $&$ 0.07 \pm0.06$\
&$5835\pm59 $&$ 4.60\pm0.12 $&$1.53\pm0.27 $&$ -0.52\pm0.08 $&$ 0.00 $&$ 0.08 $&$ 0.04 \pm0.11$\
&$6151\pm57 $&$ 4.51\pm0.10 $&$1.45\pm0.15 $&$ 0.06\pm0.07 $&$ 0.00 $&$ 0.04 $&$ 0.02 \pm0.09$\
&$6212\pm64 $&$ 4.26\pm0.13 $&$1.69\pm0.16 $&$ 0.13\pm0.08 $&$-0.05 $&$ 0.02 $&$-0.02 \pm0.10$\
&$6143\pm31 $&$ 4.48\pm0.08 $&$1.40\pm0.08 $&$ -0.03\pm0.04 $&$-0.13 $&$ -0.03 $&$-0.08 \pm0.08$\
&$5641\pm28 $&$ 4.05\pm0.05 $&$1.13\pm0.03 $&$ 0.14\pm0.04 $&$ 0.05 $&$ 0.00 $&$ 0.03 \pm0.07$\
&$5702\pm36 $&$ 4.33\pm0.08 $&$1.05\pm0.04 $&$ 0.36\pm0.05 $&$0.20 $&$ 0.12 $&$ 0.16 \pm0.08$\
&$5163\pm00 $&$ 4.52\pm0.00 $&$0.72\pm0.00 $&$ -0.24\pm0.00 $&$0.40 $&$ 0.06 $&$ 0.23 \pm0.18$\
&$5801\pm30 $&$ 4.22\pm0.12 $&$1.34\pm0.04 $&$ 0.15\pm0.04 $&$-0.05 $&$ 0.03 $&$-0.01 \pm0.07$\
&$6252\pm53 $&$ 4.61\pm0.16 $&$1.18\pm0.10 $&$ 0.26\pm0.06 $&$-0.05 $&$ 0.02 $&$-0.02 \pm0.09$\
&$6190\pm00 $&$ 4.19\pm0.00 $&$1.54\pm0.00 $&$ 0.24\pm0.00 $&$0.05 $&$ 0.12 $&$ 0.09 \pm0.06$\
&$6138\pm79 $&$ 4.53\pm0.22 $&$1.22\pm0.16 $&$ 0.17\pm0.10 $&$-0.05 $&$ -0.08$&$-0.07 \pm0.11$\
&$5073\pm42 $&$ 4.43\pm0.08 $&$1.05\pm0.06 $&$ -0.13\pm0.05 $&$0.00 $&$ 0.02 $&$ 0.01 \pm0.07$\
&$5959\pm46 $&$ 4.35\pm0.12 $&$1.20\pm0.10 $&$ -0.11\pm0.06 $&$-0.05 $&$ -0.02$&$-0.04 \pm0.08$\
&$6108\pm36 $&$ 4.25\pm0.10 $&$1.30\pm0.05 $&$ 0.31\pm0.05 $&$0.15 $&$ 0.18 $&$ 0.17 \pm0.07$\
&$4825\pm107$&$ 3.55\pm0.32 $&$1.18\pm0.12 $&$ 0.39\pm0.13 $&$0.20 $&$ 0.24 $&$ 0.22 \pm0.13$\
&$5656\pm44 $&$ 4.45\pm0.08 $&$1.01\pm0.06 $&$ 0.22\pm0.05 $&$0.15 $&$ 0.13 $&$ 0.14 \pm0.07$\
&$5587\pm00 $&$ 4.29\pm0.00 $&$1.08\pm0.00 $&$ 0.38\pm0.00 $&$0.15 $&$ 0.07 $&$ 0.11 \pm0.06$\
&$6046\pm49 $&$ 4.71\pm0.09 $&$1.79\pm0.19 $&$ -0.08\pm0.06 $&$-0.03 $&$ 0.03 $&$ 0.00 \pm0.09$\
&$5546\pm30 $&$ 4.50\pm0.03 $&$0.80\pm0.07 $&$ -0.38\pm0.04 $&$0.08 $&$ 0.09 $&$ 0.09 \pm0.06$\
&$5674\pm40 $&$ 3.94\pm0.12 $&$1.38\pm0.05 $&$ 0.40\pm0.06 $&$0.15 $&$ 0.10 $&$ 0.13 \pm0.08$\
&$5991\pm27 $&$ 4.42\pm0.10 $&$1.24\pm0.04 $&$ 0.10\pm0.04 $&$ 0.05 $&$ 0.09 $&$ 0.07 \pm0.06$\
&$5115\pm68 $&$ 4.32\pm0.19 $&$0.93\pm0.09 $&$ 0.04\pm0.07 $&$0.00 $&$ -0.05 $&$-0.03 \pm0.09$\
&$5268\pm55 $&$ 4.41\pm0.16 $&$0.97\pm0.06 $&$ 0.20\pm0.06 $&$0.14 $&$ 0.14 $&$ 0.14 \pm0.08$\
&$5644\pm54 $&$ 4.37\pm0.07 $&$0.89\pm0.07 $&$ 0.33\pm0.06 $&$-0.13 $&$ -0.11$&$-0.12 \pm0.07$\
&$6026\pm30 $&$ 4.41\pm0.13 $&$1.11\pm0.06 $&$ 0.01\pm0.04 $&$ 0.00 $&$ 0.00 $&$ 0.00 \pm0.06$\
&$6105\pm00 $&$ 4.28\pm0.00 $&$1.36\pm0.00 $&$ 0.23\pm0.00 $&$0.10 $&$ 0.08 $&$ 0.09 \pm0.05$\
&$5608\pm34 $&$ 4.35\pm0.08 $&$1.24\pm0.04 $&$ 0.26\pm0.04 $&$-0.01 $&$ 0.01 $&$ 0.00 \pm0.06$\
&$6112\pm39 $&$ 4.34\pm0.10 $&$1.38\pm0.07 $&$ 0.16\pm0.05 $&$0.05 $&$ 0.08 $&$ 0.07 \pm0.07$\
&$5279\pm62 $&$ 4.37\pm0.18 $&$0.98\pm0.07 $&$ 0.33\pm0.07 $&$0.15 $&$ 0.10 $&$ 0.13 \pm0.09$\
&$5574\pm72 $&$ 4.46\pm0.20 $&$1.14\pm0.09 $&$ 0.32\pm0.09 $&$0.20 $&$ 0.22 $&$ 0.21 \pm0.10$\
&$6234\pm45 $&$ 3.98\pm0.05 $&$1.62\pm0.08 $&$ 0.22\pm0.05 $&$- $&$ 0.00 $&$ 0.00 \pm0.07$\
&$5821\pm41 $&$ 4.45\pm0.06 $&$1.16\pm0.05 $&$ 0.32\pm0.05 $&$0.20 $&$ 0.17 $&$ 0.19 \pm0.07$\
&$5954\pm25 $&$ 4.44\pm0.10 $&$1.30\pm0.04 $&$ 0.06\pm0.03 $&$0.05 $&$ 0.12 $&$ 0.09 \pm0.07$\
&$5899\pm35 $&$ 4.34\pm0.07 $&$1.08\pm0.06 $&$ -0.01\pm0.05 $&$-0.05 $&$ 0.01 $&$-0.02 \pm0.07$\
&$6248\pm42 $&$ 4.49\pm0.16 $&$1.35\pm0.08 $&$ 0.20\pm0.05 $&$-0.05 $&$ 0.00 $&$-0.03 \pm0.08$\
&$5596\pm42 $&$ 4.37\pm0.12 $&$0.89\pm0.05 $&$ 0.23\pm0.05 $&$0.10 $&$ 0.09 $&$ 0.10 \pm0.07$\
&$4804\pm61 $&$ 4.36\pm0.28 $&$0.57\pm0.12 $&$ -0.08\pm0.06 $&$0.13 $&$ 0.10 $&$ 0.12 \pm0.09$\
&$5886\pm36 $&$ 4.28\pm0.13 $&$1.25\pm0.09 $&$ -0.25\pm0.05 $&$0.00 $&$ 0.05 $&$ 0.03 \pm0.08$\
&$5098\pm36 $&$ 4.45\pm0.11 $&$0.74\pm0.05 $&$ 0.09\pm0.04 $&$0.15 $&$ 0.09 $&$ 0.12 \pm0.07$\
&$6075\pm40 $&$ 4.64\pm0.12 $&$1.31\pm0.07 $&$ 0.16\pm0.05 $&$-0.01 $&$ 0.03 $&$ 0.01 \pm0.07$\
&$4835\pm72 $&$ 4.44\pm0.21 $&$0.89\pm0.11 $&$ 0.03\pm0.07 $&$-0.03 $&$ -0.05 $&$-0.04 \pm0.09$\
&$5392\pm36 $&$ 4.48\pm0.06 $&$0.85\pm0.05 $&$ 0.03\pm0.04 $&$0.05 $&$ 0.00 $&$ 0.03 \pm0.07$\
&$5776\pm29 $&$ 4.36\pm0.07 $&$1.09\pm0.04 $&$ 0.30\pm0.04 $&$0.20 $&$ 0.17 $&$ 0.19 \pm0.06$\
&$6222\pm39 $&$ 4.27\pm0.15 $&$1.79\pm0.12 $&$ -0.04\pm0.05 $&$ 0.00 $&$ 0.08 $&$ 0.04 \pm0.08$\
&$4775\pm110$&$ 3.09\pm0.40 $&$1.78\pm0.11 $&$ 0.13\pm0.14 $&$0.07 $&$ 0.09 $&$ 0.08 \pm0.14$\
&$5909\pm39 $&$ 4.51\pm0.08 $&$1.13\pm0.06 $&$ 0.10\pm0.05 $&$0.05 $&$ 0.09 $&$ 0.07 \pm0.07$\
&$5853\pm25 $&$ 4.41\pm0.15 $&$1.35\pm0.07 $&$ -0.21\pm0.04 $&$0.05 $&$ 0.08 $&$ 0.07 \pm0.07$\
&$5311\pm87 $&$ 4.42\pm0.18 $&$0.92\pm0.10 $&$ 0.43\pm0.08 $&$0.20 $&$ 0.17 $&$ 0.19 \pm0.10$\
&$5883\pm25 $&$ 4.51\pm0.05 $&$1.18\pm0.04 $&$ 0.06\pm0.04 $&$-0.12 $&$ -0.07$&$-0.10 \pm0.07$\
&$5961\pm27 $&$ 4.50\pm0.10 $&$1.11\pm0.06 $&$ -0.01\pm0.04 $&$-0.15 $&$ -0.08$&$-0.12 \pm0.07$\
&$5798\pm33 $&$ 4.31\pm0.08 $&$1.19\pm0.04 $&$ 0.32\pm0.04 $&$0.12 $&$ 0.10 $&$ 0.11 \pm0.06$\
&$5617\pm35 $&$ 4.22\pm0.05 $&$1.21\pm0.05 $&$ 0.06\pm0.05 $&$0.15 $&$ 0.13 $&$ 0.14 \pm0.07$\
\[tab13\]
[lccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & \[Cu/Fe\]$_1$ & \[Cu/Fe\]$_2$ & \[Cu/Fe\]\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & & &\
&$5601\pm33 $&$ 4.41\pm0.12 $&$0.99\pm0.05 $&$ -0.08\pm0.05 $&$0.10 $&$ 0.10 $&$ 0.10 \pm0.07$\
&$4804\pm77 $&$ 3.57\pm0.17 $&$1.14\pm0.09 $&$ 0.33\pm0.09 $&$ 0.17 $&$ 0.24 $&$ 0.21 \pm0.10 $\
&$5588\pm115$&$ 4.46\pm0.20 $&$0.82\pm0.14 $&$ 0.24\pm0.10 $&$ -0.14 $&$ -0.17 $&$ -0.16 \pm0.11 $\
&$6260\pm43 $&$ 4.43\pm0.05 $&$1.41\pm0.09 $&$ 0.22\pm0.05 $&$ -0.05 $&$ -0.03 $&$ -0.04 \pm0.07 $\
&$5772\pm25 $&$ 4.40\pm0.07 $&$1.07\pm0.04 $&$ 0.08\pm0.04 $&$ 0.12 $&$ 0.12 $&$ 0.12 \pm0.06 $\
&$5845\pm22 $&$ 4.42\pm0.07 $&$1.10\pm0.03 $&$ 0.13\pm0.03 $&$ 0.05 $&$ 0.09 $&$ 0.07 \pm0.06 $\
&$5584\pm36 $&$ 4.37\pm0.06 $&$1.07\pm0.05 $&$ 0.24\pm0.05 $&$ 0.15 $&$ 0.09 $&$ 0.12 \pm0.07 $\
&$4947\pm58 $&$ 4.51\pm0.20 $&$0.86\pm0.09 $&$ -0.02\pm0.06 $&$ 0.00 $&$ 0.02 $&$ 0.01 \pm0.08 $\
&$5859\pm31 $&$ 4.32\pm0.07 $&$1.27\pm0.05 $&$ 0.09\pm0.04 $&$ 0.00 $&$ 0.05 $&$ 0.03 \pm0.07 $\
&$5918\pm44 $&$ 4.35\pm0.13 $&$1.39\pm0.06 $&$ 0.22\pm0.05 $&$ 0.22 $&$ 0.17 $&$ 0.20 \pm0.08 $\
&$5532\pm00 $&$ 4.29\pm0.00 $&$1.03\pm0.00 $&$ 0.19\pm0.00 $&$ 0.15 $&$ 0.12 $&$ 0.14 \pm0.06 $\
&$5984\pm33 $&$ 4.25\pm0.10 $&$1.25\pm0.05 $&$ 0.17\pm0.05 $&$ 0.08 $&$ 0.07 $&$ 0.08 \pm0.07 $\
&$5938\pm42 $&$ 4.12\pm0.05 $&$1.28\pm0.06 $&$ 0.24\pm0.05 $&$-0.02 $&$ 0.05 $&$ 0.02 \pm0.08 $\
&$5887\pm32 $&$ 4.30\pm0.07 $&$1.31\pm0.04 $&$ 0.25\pm0.04 $&$ 0.10 $&$ 0.12 $&$ 0.11 \pm0.06 $\
&$5804\pm36 $&$ 4.42\pm0.07 $&$1.20\pm0.05 $&$ 0.20\pm0.05 $&$ 0.08 $&$ 0.07 $&$ 0.08 \pm0.07 $\
&$5645\pm00 $&$ 4.31\pm0.00 $&$1.06\pm0.00 $&$ 0.37\pm0.00 $&$ 0.15 $&$ 0.07 $&$ 0.11 \pm0.06 $\
&$5843\pm38 $&$ 4.45\pm0.07 $&$1.03\pm0.06 $&$ 0.05\pm0.05 $&$ 0.09 $&$ 0.09 $&$ 0.09 \pm0.07 $\
\[tab14\]
[lccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & \[Cu/Fe\]$_1$ & \[Cu/Fe\]$_2$ & \[Cu/Fe\]\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & & &\
&$5956\pm44 $&$ 4.39\pm0.13 $&$1.07\pm0.09 $&$ -0.14\pm0.05 $&$-0.01$&$0.03 $&$ 0.01 \pm0.07 $\
&$5878\pm53 $&$ 4.74\pm0.15 $&$1.13\pm0.10 $&$ -0.03\pm0.06 $&$ 0.00$&$0.01 $&$ 0.01 \pm0.08 $\
&$4911\pm54 $&$ 4.49\pm0.18 $&$0.71\pm0.11 $&$ -0.17\pm0.06 $&$0.05 $&$0.00 $&$ 0.03 \pm0.08 $\
&$6140\pm41 $&$ 4.39\pm0.16 $&$1.50\pm0.08 $&$ 0.18\pm0.05 $&$0.05 $&$0.09 $&$ 0.07 \pm0.07 $\
&$4970\pm40 $&$ 4.49\pm0.10 $&$0.76\pm0.07 $&$ -0.26\pm0.04 $&$0.01 $&$-0.03 $&$ -0.01 \pm0.07 $\
&$5344\pm29 $&$ 4.57\pm0.09 $&$0.91\pm0.06 $&$ -0.52\pm0.04 $&$0.02 $&$-0.03 $&$ -0.01 \pm0.07 $\
&$5368\pm24 $&$ 4.55\pm0.05 $&$0.88\pm0.05 $&$ -0.47\pm0.03 $&$0.05 $&$0.00 $&$ 0.03 \pm0.06 $\
&$5180\pm56 $&$ 4.44\pm0.13 $&$1.33\pm0.08 $&$ 0.06\pm0.07 $&$0.00 $&$-0.10 $&$ -0.05 \pm0.09 $\
&$6275\pm57 $&$ 4.40\pm0.37 $&$2.41\pm0.41 $&$ -0.19\pm0.06 $&$-0.05$&$0.09 $&$ 0.02 \pm0.14 $\
&$5733\pm31 $&$ 4.55\pm0.10 $&$1.09\pm0.06 $&$ -0.21\pm0.04 $&$0.05 $&$0.02 $&$ 0.04 \pm0.07 $\
&$5444\pm31 $&$ 4.47\pm0.07 $&$0.98\pm0.06 $&$ -0.38\pm0.04 $&$0.10 $&$0.09 $&$ 0.10 \pm0.06 $\
&$5843\pm26 $&$ 4.47\pm0.10 $&$1.17\pm0.06 $&$ -0.23\pm0.04 $&$0.00 $&$0.02 $&$ 0.01 \pm0.06 $\
&$5074\pm60 $&$ 3.77\pm0.16 $&$1.08\pm0.06 $&$ 0.13\pm0.08 $&$0.18 $&$0.12 $&$ 0.15 \pm0.09 $\
&$4975\pm55 $&$ 4.48\pm0.16 $&$0.77\pm0.09 $&$ -0.11\pm0.06 $&$0.16 $&$0.08 $&$ 0.12 \pm0.09 $\
&$5176\pm45 $&$ 4.41\pm0.17 $&$1.03\pm0.06 $&$ 0.06\pm0.05 $&$0.00 $&$-0.10 $&$ -0.05 \pm0.09 $\
&$5126\pm34 $&$ 4.51\pm0.08 $&$0.60\pm0.07 $&$ -0.31\pm0.04 $&$0.12 $&$0.07 $&$ 0.10 \pm0.07 $\
&$5868\pm30 $&$ 4.55\pm0.10 $&$1.24\pm0.05 $&$ 0.02\pm0.04 $&$-0.05$&$0.03 $&$ -0.01 \pm0.07 $\
&$5479\pm37 $&$ 4.61\pm0.07 $&$1.12\pm0.05 $&$ 0.00\pm0.05 $&$-0.12$&$-0.08 $&$ -0.10 \pm0.07 $\
&$5752\pm32 $&$ 4.53\pm0.07 $&$1.26\pm0.07 $&$ -0.23\pm0.05 $&$-0.03$&$0.00 $&$ -0.01 \pm0.07 $\
&$4805\pm52 $&$ 4.37\pm0.37 $&$0.49\pm0.12 $&$ -0.30\pm0.05 $&$0.00 $&$-0.05 $&$ -0.02 \pm0.09 $\
&$5633\pm35 $&$ 4.48\pm0.07 $&$1.24\pm0.05 $&$ -0.01\pm0.04 $&$-0.12$&$-0.14 $&$ -0.13 \pm0.06 $\
&$5594\pm36 $&$ 4.41\pm0.09 $&$1.05\pm0.04 $&$ 0.10\pm0.05 $&$0.15 $&$0.10 $&$ 0.13 \pm0.07 $\
&$4658\pm56 $&$ 4.32\pm0.24 $&$0.64\pm0.15 $&$ -0.04\pm0.07 $&$0.00 $&$-0.03 $&$ -0.01 \pm0.09 $\
&$5825\pm20 $&$ 4.37\pm0.10 $&$1.20\pm0.04 $&$ -0.19\pm0.03 $&$0.01 $&$0.04 $&$ 0.02 \pm0.06 $\
&$5260\pm31 $&$ 4.35\pm0.11 $&$0.74\pm0.05 $&$ -0.26\pm0.04 $&$0.10 $&$0.04 $&$ 0.07 \pm0.07 $\
&$5979\pm31 $&$ 4.59\pm0.12 $&$1.36\pm0.10 $&$ -0.29\pm0.04 $&$0.08 $&$0.10 $&$ 0.09 \pm0.07 $\
&$5410\pm26 $&$ 4.38\pm0.07 $&$0.89\pm0.03 $&$ -0.03\pm0.04 $&$0.05 $&$-0.03 $&$ 0.01 \pm0.07 $\
&$5242\pm28 $&$ 4.50\pm0.09 $&$0.69\pm0.05 $&$ -0.37\pm0.04 $&$0.07 $&$ 0.04 $&$ 0.06 \pm0.06 $\
&$5000\pm55 $&$ 4.55\pm0.13 $&$1.07\pm0.08 $&$ -0.03\pm0.06 $&$-0.05$&$-0.11 $&$ -0.08 \pm0.08 $\
&$5803\pm29 $&$ 4.50\pm0.08 $&$1.02\pm0.04 $&$ 0.14\pm0.04 $&$0.03 $&$-0.01 $&$ 0.01 \pm0.06 $\
&$6167\pm37 $&$ 4.35\pm0.10 $&$1.42\pm0.09 $&$ -0.03\pm0.05 $&$-0.05$&$0.01 $&$ -0.02 \pm0.07 $\
&$5765\pm24 $&$ 4.52\pm0.05 $&$1.22\pm0.05 $&$ -0.23\pm0.04 $&$0.05 $&$0.04 $&$ 0.05 \pm0.06 $\
&$5005\pm45 $&$ 4.38\pm0.25 $&$0.67\pm0.09 $&$ -0.55\pm0.06 $&$0.15 $&$0.12 $&$ 0.14 \pm0.08 $\
&$5069\pm49 $&$ 4.38\pm0.19 $&$0.79\pm0.07 $&$ -0.01\pm0.05 $&$0.18 $&$0.07 $&$ 0.13 \pm0.09 $\
&$5435\pm39 $&$ 4.48\pm0.08 $&$0.91\pm0.07 $&$ -0.29\pm0.05 $&$0.02 $&$0.04 $&$ 0.03 \pm0.07 $\
&$5910\pm24 $&$ 4.42\pm0.05 $&$1.14\pm0.04 $&$ 0.00\pm0.04 $&$-0.07$&$-0.05 $&$ -0.06 \pm0.06 $\
&$4629\pm77 $&$ 4.36\pm0.19 $&$0.42\pm0.25 $&$ -0.06\pm0.08 $&$0.02 $&$-0.13 $&$ -0.06 \pm0.13 $\
&$5890\pm30 $&$ 4.51\pm0.07 $&$1.12\pm0.07 $&$ -0.17\pm0.04 $&$-0.02$&$0.04 $&$ 0.01 \pm0.07 $\
&$4555\pm87 $&$ 4.53\pm0.26 $&$0.66\pm0.28 $&$ -0.01\pm0.09 $&$-0.01$&$-0.07 $&$ -0.04 \pm0.12 $\
&$4747\pm58 $&$ 4.48\pm0.22 $&$0.40\pm0.20 $&$ -0.31\pm0.06 $&$0.21 $&$0.11 $&$ 0.16 \pm0.10 $\
&$5260\pm41 $&$ 4.45\pm0.11 $&$0.92\pm0.06 $&$ -0.16\pm0.05 $&$0.01 $&$0.03 $&$ 0.02 \pm0.07 $\
\[tab15\]
[lccc]{} Species & $<$\[X/H\]$>\pm$ rms & $<$\[X/H\]$>\pm$ rms & Difference\
& (comp. sample) & (planet hosts) &\
C & $-0.03\pm0.14$ & $0.14\pm0.17$ & 0.17\
S & $-0.18\pm0.14$ & $-0.02\pm0.17$ & 0.16\
Zn & $-0.12\pm0.16$ & $0.13\pm0.22$ & 0.25\
Cu & $-0.12\pm0.17$ & $0.19\pm0.23$ & 0.31\
\[TabAve\]
[lccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & EW & EW$_{liter}$ & Ref.\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & & &\
& $5536\pm50$ & $4.56\pm0.12$ & $1.33\pm0.06$ & $ 0.12\pm0.06$ & $16.7$ & $17.0$ & 1\
& $6212\pm64$ & $4.26\pm0.13$ & $1.69\pm0.16$ & $ 0.13\pm0.08$ & $45.2$ & $41.5$ & 3\
& $5641\pm28$ & $4.05\pm0.05$ & $1.13\pm0.03$ & $ 0.14\pm0.04$ & $27.4$ & $24.9$ & 2\
& $5702\pm36$ & $4.33\pm0.08$ & $1.05\pm0.04$ & $ 0.36\pm0.05$ & $32.3$ & $31.9$ & 2K\
& $5163\pm00$ & $4.52\pm0.00$ & $0.72\pm0.00$ & $-0.24\pm0.00$ & $12.9$ & $ 7.8$ & 1\
& $5073\pm42$ & $4.43\pm0.08$ & $1.05\pm0.06$ & $-0.13\pm0.05$ & $ 9.5$ & $ 9.8$ & 3\
& $5546\pm30$ & $4.50\pm0.03$ & $0.80\pm0.07$ & $-0.38\pm0.04$ & $13.0$ & $11.9$ & 2K\
& $5674\pm40$ & $3.94\pm0.12$ & $1.38\pm0.05$ & $ 0.40\pm0.06$ & $45.0$ & $35.2$ & 2K\
& $5674\pm40$ & $3.94\pm0.12$ & $1.38\pm0.05$ & $ 0.40\pm0.06$ & $45.0$ & $35.5$ & 3\
& $5268\pm55$ & $4.41\pm0.16$ & $0.97\pm0.06$ & $ 0.20\pm0.06$ & $20.8$ & $19.7$ & 2K\
& $6105\pm00$ & $4.28\pm0.00$ & $1.36\pm0.00$ & $ 0.23\pm0.00$ & $43.6$ & $39.4$ & 1\
& $6105\pm00$ & $4.28\pm0.00$ & $1.36\pm0.00$ & $ 0.23\pm0.06$ & $43.6$ & $41.6$ & 2K\
& $6105\pm00$ & $4.28\pm0.00$ & $1.36\pm0.00$ & $ 0.23\pm0.06$ & $43.6$ & $37.1$ & 3\
& $6143\pm53$ & $4.42\pm0.13$ & $1.53\pm0.09$ & $ 0.28\pm0.07$ & $37.6$ & $39.0$ & 1\
& $5279\pm62$ & $4.37\pm0.18$ & $0.98\pm0.07$ & $ 0.33\pm0.07$ & $21.6$ & $19.5$ & 3\
& $6015\pm00$ & $4.46\pm0.00$ & $1.13\pm0.00$ & $ 0.30\pm0.00$ & $37.6$ & $35.4$ & 4\
& $6015\pm00$ & $4.46\pm0.00$ & $1.13\pm0.00$ & $ 0.30\pm0.00$ & $37.6$ & $36.3$ & 1\
& $5454\pm61$ & $4.33\pm0.17$ & $1.08\pm0.08$ & $ 0.35\pm0.08$ & $27.7$ & $26.1$ & 1\
& $6234\pm45$ & $3.98\pm0.05$ & $1.62\pm0.08$ & $ 0.22\pm0.05$ & $50.2$ & $49.3$ & 2\
& $6234\pm45$ & $3.98\pm0.05$ & $1.62\pm0.08$ & $ 0.22\pm0.05$ & $50.2$ & $43.0$ & 3\
& $5821\pm41$ & $4.45\pm0.06$ & $1.16\pm0.05$ & $ 0.32\pm0.05$ & $37.0$ & $31.1$ & 2K\
& $5821\pm41$ & $4.45\pm0.06$ & $1.16\pm0.05$ & $ 0.32\pm0.05$ & $37.0$ & $28.8$ & 4\
& $5954\pm25$ & $4.44\pm0.10$ & $1.30\pm0.04$ & $ 0.06\pm0.03$ & $25.9$ & $25.8$ & 3\
& $5899\pm35$ & $4.34\pm0.07$ & $1.08\pm0.06$ & $-0.01\pm0.05$ & $23.1$ & $21.8$ & 4\
& $5884\pm34$ & $4.22\pm0.22$ & $1.31\pm0.17$ & $-0.70\pm0.04$ & $12.2$ & $10.0$ & 4\
& $5560\pm34$ & $4.07\pm0.05$ & $1.18\pm0.05$ & $-0.06\pm0.05$ & $18.0$ & $14.9$ & 3\
& $6339\pm73$ & $4.19\pm0.10$ & $1.70\pm0.16$ & $ 0.23\pm0.07$ & $59.9$ & $56.8$ & 3\
& $5392\pm36$ & $4.48\pm0.06$ & $0.85\pm0.05$ & $ 0.03\pm0.04$ & $14.5$ & $14.6$ & 2\
& $5392\pm36$ & $4.48\pm0.06$ & $0.85\pm0.05$ & $ 0.03\pm0.04$ & $14.5$ & $13.7$ & 4\
& $5776\pm29$ & $4.36\pm0.07$ & $1.09\pm0.04$ & $ 0.30\pm0.04$ & $36.3$ & $34.8$ & 2\
& $5776\pm29$ & $4.36\pm0.07$ & $1.09\pm0.04$ & $ 0.30\pm0.04$ & $36.3$ & $34.1$ & 4\
& $5909\pm39$ & $4.51\pm0.08$ & $1.13\pm0.06$ & $ 0.10\pm0.05$ & $28.2$ & $26.5$ & 4\
& $5853\pm25$ & $4.41\pm0.15$ & $1.35\pm0.07$ & $-0.21\pm0.04$ & $21.1$ & $15.9$ & 3\
& $5311\pm87$ & $4.42\pm0.18$ & $0.92\pm0.10$ & $ 0.43\pm0.08$ & $24.5$ & $23.9$ & 3\
& $5617\pm35$ & $4.22\pm0.05$ & $1.21\pm0.05$ & $ 0.06\pm0.05$ & $26.7$ & $27.3$ & 2\
& $5601\pm33$ & $4.41\pm0.12$ & $0.99\pm0.05$ & $-0.08\pm0.05$ & $22.5$ & $18.0$ & 4\
& $6299\pm41$ & $4.10\pm0.02$ & $1.42\pm0.09$ & $ 0.21\pm0.05$ & $48.2$ & $45.1$ & 1\
& $5772\pm25$ & $4.40\pm0.07$ & $1.07\pm0.04$ & $ 0.08\pm0.04$ & $23.8$ & $23.9$ & 3\
& $5845\pm22$ & $4.42\pm0.07$ & $1.10\pm0.03$ & $ 0.13\pm0.03$ & $26.2$ & $24.9$ & 4\
& $5325\pm00$ & $3.90\pm0.00$ & $1.11\pm0.00$ & $-0.26\pm0.00$ & $12.6$ & $11.9$ & 4\
& $5752\pm53$ & $4.50\pm0.09$ & $1.01\pm0.06$ & $ 0.35\pm0.06$ & $27.1$ & $22.5$ & 1\
& $5804\pm36$ & $4.42\pm0.07$ & $1.20\pm0.05$ & $ 0.20\pm0.05$ & $34.3$ & $30.2$ & 2\
& $5804\pm36$ & $4.42\pm0.07$ & $1.20\pm0.05$ & $ 0.20\pm0.05$ & $34.3$ & $29.5$ & 3\
& $5645\pm00$ & $4.31\pm0.00$ & $1.06\pm0.00$ & $ 0.37\pm0.00$ & $29.1$ & $27.6$ & 2\
& $5645\pm00$ & $4.31\pm0.00$ & $1.06\pm0.00$ & $ 0.37\pm0.00$ & $29.1$ & $29.4$ & 3\
& $5843\pm38$ & $4.45\pm0.07$ & $1.03\pm0.06$ & $ 0.05\pm0.05$ & $22.2$ & $20.1$ & 2\
\
\
\
\
\[tab16\]
[lccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & \[S/Fe\] & \[S/Fe\]$_{liter}$ & Ref.\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & & &\
& $5536\pm50$ & $4.56\pm0.12$ & $1.33\pm0.06$ & $ 0.12\pm0.06$ & $-0.22\pm0.08$ & $ 0.00$ & 1\
& $6212\pm64$ & $4.26\pm0.13$ & $1.69\pm0.16$ & $ 0.13\pm0.08$ & $-0.33\pm0.09$ & $ 0.06$ & 2\
& $5073\pm42$ & $4.43\pm0.08$ & $1.05\pm0.06$ & $-0.13\pm0.05$ & $ 0.00\pm0.08$ & $ 0.32$ & 2\
& $5674\pm40$ & $3.94\pm0.12$ & $1.38\pm0.05$ & $ 0.40\pm0.06$ & $-0.32\pm0.08$ & $ 0.08$ & 2\
& $6105\pm00$ & $4.28\pm0.00$ & $1.36\pm0.00$ & $ 0.23\pm0.00$ & $-0.20\pm0.06$ & $ 0.06$ & 2\
& $6105\pm00$ & $4.28\pm0.00$ & $1.36\pm0.00$ & $ 0.23\pm0.06$ & $-0.20\pm0.06$ & $-0.12$ & 1\
& $6143\pm53$ & $4.42\pm0.13$ & $1.53\pm0.09$ & $ 0.28\pm0.07$ & $-0.31\pm0.09$ & $-0.21$ & 1\
& $5279\pm62$ & $4.37\pm0.18$ & $0.98\pm0.07$ & $ 0.33\pm0.07$ & $-0.20\pm0.10$ & $ 0.13$ & 2\
& $6015\pm00$ & $4.46\pm0.00$ & $1.13\pm0.00$ & $ 0.30\pm0.00$ & $-0.20\pm0.05$ & $-0.13$ & 1\
& $6015\pm00$ & $4.46\pm0.00$ & $1.13\pm0.00$ & $ 0.30\pm0.00$ & $-0.20\pm0.05$ & $-0.04$ & 3\
& $5454\pm61$ & $4.33\pm0.17$ & $1.08\pm0.08$ & $ 0.35\pm0.08$ & $-0.15\pm0.10$ & $ 0.07$ & 1\
& $5821\pm41$ & $4.45\pm0.06$ & $1.16\pm0.05$ & $ 0.32\pm0.05$ & $-0.22\pm0.07$ & $-0.10$ & 3\
& $5954\pm25$ & $4.44\pm0.10$ & $1.30\pm0.04$ & $ 0.06\pm0.03$ & $-0.26\pm0.06$ & $-0.08$ & 2\
& $5899\pm35$ & $4.34\pm0.07$ & $1.08\pm0.06$ & $-0.01\pm0.05$ & $-0.09\pm0.07$ & $-0.05$ & 3\
& $5884\pm34$ & $4.22\pm0.22$ & $1.31\pm0.17$ & $-0.70\pm0.04$ & $ 0.15\pm0.09$ & $ 0.12$ & 3\
& $5560\pm34$ & $4.07\pm0.05$ & $1.18\pm0.05$ & $-0.06\pm0.05$ & $ 0.03\pm0.10$ & $ 0.10$ & 2\
& $6339\pm73$ & $4.19\pm0.10$ & $1.70\pm0.16$ & $ 0.23\pm0.07$ & $-0.18\pm0.09$ & $ 0.26$ & 2\
& $5392\pm36$ & $4.48\pm0.06$ & $0.85\pm0.05$ & $ 0.03\pm0.04$ & $-0.06\pm0.07$ & $ 0.01$ & 3\
& $5776\pm29$ & $4.36\pm0.07$ & $1.09\pm0.04$ & $ 0.30\pm0.04$ & $-0.17\pm0.07$ & $-0.01$ & 3\
& $5909\pm39$ & $4.51\pm0.08$ & $1.13\pm0.06$ & $ 0.10\pm0.05$ & $-0.18\pm0.07$ & $ 0.06$ & 3\
& $5853\pm25$ & $4.41\pm0.15$ & $1.35\pm0.07$ & $-0.21\pm0.04$ & $-0.04\pm0.07$ & $ 0.15$ & 2\
& $5311\pm87$ & $4.42\pm0.18$ & $0.92\pm0.10$ & $ 0.43\pm0.08$ & $-0.13\pm0.12$ & $ 0.32$ & 2\
& $5601\pm33$ & $4.41\pm0.12$ & $0.99\pm0.05$ & $-0.08\pm0.05$ & $ 0.13\pm0.07$ & $ 0.08$ & 3\
& $6299\pm41$ & $4.10\pm0.02$ & $1.42\pm0.09$ & $ 0.21\pm0.05$ & $-0.31\pm0.06$ & $-0.17$ & 1\
& $5772\pm25$ & $4.40\pm0.07$ & $1.07\pm0.04$ & $ 0.08\pm0.04$ & $-0.13\pm0.06$ & $-0.09$ & 2\
& $5845\pm22$ & $4.42\pm0.07$ & $1.10\pm0.03$ & $ 0.13\pm0.03$ & $-0.23\pm0.08$ & $-0.09$ & 3\
& $5325\pm00$ & $3.90\pm0.00$ & $1.11\pm0.00$ & $-0.26\pm0.00$ & $ 0.01\pm0.07$ & $-0.06$ & 3\
& $5752\pm53$ & $4.50\pm0.09$ & $1.01\pm0.06$ & $ 0.35\pm0.06$ & $-0.22\pm0.08$ & $-0.26$ & 1\
& $6117\pm26$ & $4.48\pm0.08$ & $1.40\pm0.06$ & $ 0.02\pm0.03$ & $-0.22\pm0.06$ & $ 0.00$ & 3\
& $5804\pm36$ & $4.42\pm0.07$ & $1.20\pm0.05$ & $ 0.20\pm0.05$ & $-0.12\pm0.07$ & $-0.01$ & 2\
& $5645\pm00$ & $4.31\pm0.00$ & $1.06\pm0.00$ & $ 0.37\pm0.00$ & $-0.24\pm0.06$ & $ 0.07$ & 2\
\
\
\
\[tab17\]
[lccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & \[Zn/Fe\] & \[Zn/Fe\]$_{liter}$ & Ref.\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & & &\
& $6212\pm64$ & $4.26\pm0.13$ & $1.69\pm0.16$ & $ 0.13\pm0.08$ & $-0.12\pm0.10$ & $ 0.05$ & 1\
& $5073\pm42$ & $4.43\pm0.08$ & $1.05\pm0.06$ & $-0.13\pm0.05$ & $-0.12\pm0.08$ & $-0.16$ & 1\
& $5674\pm40$ & $3.94\pm0.12$ & $1.38\pm0.05$ & $ 0.40\pm0.06$ & $-0.14\pm0.08$ & $ 0.38$ & 1\
& $6105\pm00$ & $4.28\pm0.00$ & $1.36\pm0.00$ & $ 0.23\pm0.00$ & $-0.09\pm0.07$ & $ 0.14$ & 1\
& $5279\pm62$ & $4.37\pm0.18$ & $0.98\pm0.07$ & $ 0.33\pm0.07$ & $ 0.16\pm0.09$ & $ 0.48$ & 1\
& $5821\pm41$ & $4.45\pm0.06$ & $1.16\pm0.05$ & $ 0.32\pm0.05$ & $ 0.11\pm0.07$ & $ 0.29$ & 2\
& $5954\pm25$ & $4.44\pm0.10$ & $1.30\pm0.04$ & $ 0.06\pm0.03$ & $ 0.00\pm0.07$ & $ 0.11$ & 1\
& $5899\pm35$ & $4.34\pm0.07$ & $1.08\pm0.06$ & $-0.01\pm0.05$ & $-0.13\pm0.07$ & $-0.12$ & 2\
& $5392\pm36$ & $4.48\pm0.06$ & $0.85\pm0.05$ & $ 0.03\pm0.04$ & $ 0.06\pm0.13$ & $ 0.01$ & 2\
& $5776\pm29$ & $4.36\pm0.07$ & $1.09\pm0.04$ & $ 0.30\pm0.04$ & $ 0.19\pm0.07$ & $ 0.31$ & 2\
& $5909\pm39$ & $4.51\pm0.08$ & $1.13\pm0.06$ & $ 0.10\pm0.05$ & $-0.02\pm0.08$ & $ 0.05$ & 2\
& $5853\pm25$ & $4.41\pm0.15$ & $1.35\pm0.07$ & $-0.21\pm0.04$ & $ 0.10\pm0.08$ & $-0.10$ & 1\
& $5311\pm87$ & $4.42\pm0.18$ & $0.92\pm0.10$ & $ 0.43\pm0.08$ & $ 0.16\pm0.11$ & $ 0.54$ & 1\
& $5601\pm33$ & $4.41\pm0.12$ & $0.99\pm0.05$ & $-0.08\pm0.05$ & $ 0.14\pm0.08$ & $ 0.02$ & 2\
& $5772\pm25$ & $4.40\pm0.07$ & $1.07\pm0.04$ & $ 0.08\pm0.04$ & $ 0.11\pm0.06$ & $-0.03$ & 1\
& $5845\pm22$ & $4.42\pm0.07$ & $1.10\pm0.03$ & $ 0.13\pm0.03$ & $ 0.08\pm0.06$ & $ 0.08$ & 2\
& $4947\pm58$ & $4.51\pm0.20$ & $0.86\pm0.09$ & $-0.02\pm0.06$ & $ 0.07\pm0.09$ & $ 0.12$ & 2\
& $5804\pm36$ & $4.42\pm0.07$ & $1.20\pm0.05$ & $ 0.20\pm0.05$ & $-0.02\pm0.07$ & $ 0.18$ & 1\
& $5645\pm00$ & $4.31\pm0.00$ & $1.06\pm0.00$ & $ 0.37\pm0.00$ & $ 0.04\pm0.07$ & $ 0.27$ & 1\
\
\
\[tab18\]
[lccccccr]{} Star & $T_\mathrm{eff}$ & $\log {g}$ & $\xi_t$ & \[Fe/H\] & \[Cu/Fe\] & \[Cu/Fe\]$_{liter}$ & Ref.\
& (K) & (cms$^{-2}$) & (kms$^{-1}$) & & & &\
& $5821\pm41$ & $4.45\pm0.06$ & $1.16\pm0.05$ & $ 0.32\pm0.05$ & $ 0.19\pm0.07$ & $ 0.52$ & 1\
& $5899\pm35$ & $4.34\pm0.07$ & $1.08\pm0.06$ & $-0.01\pm0.05$ & $-0.02\pm0.07$ & $-0.04$ & 1\
& $5392\pm36$ & $4.48\pm0.06$ & $0.85\pm0.05$ & $ 0.03\pm0.04$ & $ 0.03\pm0.07$ & $ 0.16$ & 1\
& $5776\pm29$ & $4.36\pm0.07$ & $1.09\pm0.04$ & $ 0.30\pm0.04$ & $ 0.19\pm0.06$ & $ 0.52$ & 1\
& $5909\pm39$ & $4.51\pm0.08$ & $1.13\pm0.06$ & $ 0.10\pm0.05$ & $ 0.07\pm0.07$ & $ 0.13$ & 1\
& $5601\pm33$ & $4.41\pm0.12$ & $0.99\pm0.05$ & $-0.08\pm0.05$ & $ 0.10\pm0.07$ & $ 0.03$ & 1\
& $5845\pm22$ & $4.42\pm0.07$ & $1.10\pm0.03$ & $ 0.13\pm0.03$ & $ 0.07\pm0.06$ & $ 0.20$ & 1\
& $4947\pm58$ & $4.51\pm0.20$ & $0.86\pm0.09$ & $-0.02\pm0.06$ & $ 0.01\pm0.08$ & $ 0.04$ & 1\
\
\[tab19\]
[^1]: Based on observations collected at the La Silla Observatory, ESO (Chile), with the [CORALIE]{} spectrograph at the 1.2-m Euler Swiss telescope and with the [FEROS]{} spectrograph at the 1.52-m and 2.2-m ESO telescopes, at the Paranal Observatory, ESO (Chile), using the UVES spectrograph at the [VLT/UT2]{} Kueyen telescope, and with the UES and SARG spectrographs at the 4-m William Hershel Telescope (WHT) and at the 3.5-m TNG telescope, respectively, both at La Palma (Canary Islands).
| {
"pile_set_name": "ArXiv"
} |
---
bibliography:
- 'persistence\_fp.bib'
---
[**Contractibility of a persistence map preimage**]{}
0.5cm [Jacek Cyranka$^{*,\dag}$ and Konstantin Mischaikow$^{*}$ ]{} 0.5cm [$^*$ Department of Mathematics, Rutgers, The State University of New Jersey,\
110 Frelinghusen Rd, Piscataway, NJ 08854-8019, USA]{}\
0.5cm [$^\dag$ Department of Computer Science and Engineering, University California, San Diego,\
9500 Gilman Drive, La Jolla, CA 92093-0404, USA]{} 0.5cm [[email protected], [email protected]]{} 0.5cm
[**Abstract.**]{} This work is motivated by the following question in data-driven study of dynamical systems: given a dynamical system that is observed via time series of persistence diagrams that encode topological features of solutions snapshots, what conclusions can be drawn about solutions of the original dynamical system? In this paper we provide a definition of a persistence diagram for a point in ${\mathbb{R}}^N$ modeled on piecewise monotone functions. We then provide conditions under which time series of persistence diagrams can be used to guarantee the existence of a fixed point of the flow on ${\mathbb{R}}^N$ that generates the time series. To obtain this result requires an understanding of the preimage of the persistence map. The main theorem of this paper gives conditions under which these preimages are contractible simplicial complexes.
#### Keywords:
[ Topological data analysis, persistent homology, dynamical systems, fixed point theorem.]{}
Introduction {#secintro}
============
*Topological data analysis* (TDA), especially in the form of persistent homology, is rapidly developing into a widely used tool for the analysis of high dimensional data associated with nonlinear structures. That topological tools can play a role in this subject should not be unexpected given the central role of nonlinear functional analysis in the study of geometry, analysis, and differential equations. What is perhaps surprising is that, to the best of our knowledge, there has been no systematic attempt to make use of nonlinear functional analysis as a tool to process information obtained via persistent homology.
Persistent homology is often used as a means of data reduction. A typical example takes the form of a complicated scalar function defined over a fixed domain, where the geometry of the sub-(super)-level sets is encoded via homology. Of particular interest to us are settings in which the scalar function arises as a solution to a partial differential equation, we are interested in tracking the evolution of the function, but experimental data only provides information on the level of digital images of the process. Furthermore, for many problems capturing the dynamics requires a long time series of rather large digital images. Thus, rather than storing the full images, one can hope to work with a time series of persistence diagrams.
The simplest mathematical analogy is that of attempting to describe the evolution of a partial differential equation via observations of the evolution of finite dimensional vectors obtained through a Galerkin projection. Obviously, since this is a central problem in numerical analysis this problem has been addressed in a variety of ways, including functional analysis. However, we are interested in going a step further. One can imagine that a piecewise linear function is associated with each finite dimensional vector, and to continue the analogy, the sublevel set persistence diagram for this function is computed. We would like to draw conclusions about the dynamics of the original partial differential equation from the time series of the persistence diagrams. This is an extremely ambitious goal and far beyond our capabilities at the moment. A much simpler question is the following: if there is an attracting region in the space of persistence diagrams, under what conditions can we conclude that there is a fixed point for the partial differential equation.
This paper represents a first step towards answering the simpler question. Theorem \[thmmaindyn\] shows that given a flow with a global compact attractor ${{\mathcal{A}}}\subset{\mathbb{R}}^N$ and a neighborhood of a particular form in the space of persistence diagrams, if under the dynamics the neighborhood is mapped into itself, then there exists a fixed point for the flow. Returning to the partial differential equation analogy, one could consider a system defined over an interval where the dynamics on the global attractor is obtained via a finite difference approximation.
The challenge, even in the simplified finite dimensional setting, is that to obtain results one must understand the topology of pre-images of sets under persistent homology, a topic for which there are only limited results. That the structure of pre-images is complicated follows directly from the fact that persistent homology can provide tremendous data reduction, but in a highly nonlinear fashion. We emphasize that simple examples show that the preimage set of a straight line path in a persistence diagram is nonconvex, see Figure \[figpre\].
With this in mind the primary goal of this paper is to show that for a reasonable class of bounded piecewise monotone continuous functions on a bounded interval the pre-image of a persistence diagram is composed of contractible, simplicial sets. The importance of this result is that it opens the possibility of applying standard algebraic topological tools, e.g. Lefschetz fixed point theorem, Conley index, to dynamics that is observed through the lens of persistent homology.
To state our goal precisely requires the introduction of notation. Throughout this paper ${{\mathcal{S}_N}}$ denotes a simplicial complex composed out of $N$ vertices and $N-1$ edges. More specifically, ${{\mathcal{S}_N}}= {{\mathcal{S}_N^{(0)}}}\cup {{\mathcal{S}_N^{(1)}}}$ where ${{\mathcal{S}_N^{(0)}}}$ denotes the $0$-skeleton of ${{\mathcal{S}_N}}$, i.e. the set of vertices $${{\mathcal{S}_N^{(0)}}}=\left\{ <i>\ |\ i=1,\dots,N \right\},$$ and ${{\mathcal{S}_N^{(1)}}}$ denotes the set of edges $${{\mathcal{S}_N^{(1)}}} = \left\{ <i,i+1>\ |\ i=1,\dots,N-1\right\}.$$ Observe that ${{\mathcal{S}_N}}$ can be viewed as a simplicial decomposition of a bounded interval $I\subset{\mathbb{R}}$. We are interested in sublevel set filtrations associated with piecewise monotone continuous functions on $I$, and thus we are interested in the following filtration on ${{\mathcal{S}_N}}$.
\[def:Sfiltration\] Let $z\in {\mathbb{R}}^N$. For $\theta\in {\mathbb{R}}$, define $${{\mathcal{S}_N}}(z,\theta) := {\left\{ {\sigma\in{{\mathcal{S}_N}}: f(z,\sigma)\leq \theta}\right\}}$$ where $f\colon {\mathbb{R}}^N\times{{\mathcal{S}_N}}\to {\mathbb{R}}$, is given by $$f(z,\sigma) := \begin{cases}
z_j &\text{if $\sigma=<j>$,}\\
\max{\left\{ {z_j, z_{j+1}}\right\}}&\text{if $\sigma=<j,j+1>$.}
\end{cases}$$
\[defn:sublevelset\] Given $z=(z_1,\ldots,z_N)\in {\mathbb{R}}^N$, we can reorder the coordinates of $z$ such that $$z_{j_1} \leq z_{j_2} \leq \cdots \leq z_{j_N}.$$ The *sublevel-set filtration of ${{\mathcal{S}_N}}$ at $z$* is given by $${{\mathcal{S}_N}}(z,z_{j_1}) \subseteq {{\mathcal{S}_N}}(z,z_{j_2}) \subseteq \cdots \subseteq {{\mathcal{S}_N}}(z,z_{j_N}).$$ and denoted by ${{\mathcal{S}_N^{{{\mathsf F}}}}}(z)$.
Since ${{\mathcal{S}_N^{{{\mathsf F}}}}}(z)$ is a finite filtration of simplicial complexes we can use the classical results from persistent homology [@ZC; @comptop] to compute the persistent homology of ${{\mathcal{S}_N^{{{\mathsf F}}}}}(z)$. Since for fixed $N$, ${{\mathcal{S}_N^{{{\mathsf F}}}}}(z)$ is completely determined by $z$, we treat this as a map $${\mathsf{Dgm}}\colon {\mathbb{R}}^N\to {\mathsf{Per}}$$ where ${\mathsf{Per}}$ denotes the space of all persistence diagrams. Furthermore, there are a variety of topologies that can be put on ${\mathsf{Per}}$ such that ${\mathsf{Dgm}}$ becomes a continuous map [@chazal; @stability].
If we replace $\max{\left\{ {z_j,z_{j+1}}\right\}}$ by $\min{\left\{ {z_j,z_{j+1}}\right\}}$ in the definition of $f$ and set\
${{\mathcal{S}_N}}(z,\theta) := {\left\{ {\sigma\in{{\mathcal{S}_N}}: f(z,\sigma)\geq \theta}\right\}}$, then the appropriate modification of Definition \[defn:sublevelset\] leads to a superlevel-set filtration of ${{\mathcal{S}_N}}$. The results of this paper are equally applicable in this setting, but the proofs must be modified appropriately.
Since ${\mathsf{Per}}$ is not a linear space, ${\mathsf{Dgm}}$ is far from a linear function. However, some restrictions of its image can be obtained from the values of the coordinates of $z$. For this reason it is useful to think of $z$ as a real valued function defined over its coordinate indices, i.e. $z\colon {\left\{ {1,\ldots,N}\right\}}\to {\mathbb{R}}$ where $z(i):=z_i$, $i=1,\ldots,N$.
\[defextr\] Let $z\in{\mathbb{R}}^N$. A *local minimum* of $z$ is a sequence of indices $\left\{i,{i+1},\dots,{i+n}\right\}$ such that $z(i) = z(i+1) = \dots = z(i+n)$, where $0 \leq n < N-1$, and it holds that
- if $i-1\geq 1$, then $z(i-1) > m$,
- if $i+n+1\leq N$, then $z(i+n+1) > m$.
A *local maximum* of $z$ is a sequence of indices $\left\{i,{i+1},\dots,{i+n}\right\}$ such that $z(i) = z(i+1) = \dots = z(i+n)$, where $0 \leq n < N-1$, and it holds that
- if $i-1\geq 1$, then $z(i-1) < m$,
- if $i+n+1\leq N$, then $z(i+n+1) < m$.
A *local extremum* of $z$ is either a local minimum of $z$ or a local maximum of $z$. A *boundary extremum* is a *local extremum* where $i=1$ or $i+n=N$. A local extremum that is not a boundary extremum is an *interior extremum*.
We say that $z$ is *monotone increasing* (*monotone decreasing*) over the set of indices ${\left\{ {k,k+1,\ldots, k+j}\right\}}$ if $z(k)\leq z(k+1) \leq \cdots \leq z(k+j)$ ($z(k)\geq z(k+1) \geq \cdots \geq z(k+j))$.
We make the following assumption for the remainder of this paper
A
: The values of all local extrema of $z$ are distinct.
The important consequence of assumption [**A**]{} is the following proposition that is proven in Section \[sec:background\].
\[prop:singlePoints\] If $z\in{\mathbb{R}}^n$ satisfies assumption [**A**]{}, then each persistence point in ${\mathsf{Dgm}}(z)$ is unique.
As indicated above, given $A\subset {\mathsf{Per}}$ we are interested in the topology of ${\mathsf{Dgm}}^{-1}(A)$. For the moment we focus on the pre-image of a single persistence diagram for which it is useful to catalog the order of the local extrema.
The *critical value set* is given by $$\label{def:E+}
{{\mathcal{E}}}_+ := {\left\{ { h= (h_1,\ldots,h_n)\in {\mathbb{R}}^n : \text{$n$ is odd, and $h_{2k\pm 1} < h_{2k}$ for all $k$} }\right\}}.$$ Set ${{\mathcal{E}}}^n_+:= {\left\{ {h\in \mathcal{E}_+ : h\in{\mathbb{R}}^n}\right\}}$. We will also use the simpler notation ${{\mathcal{E}}}^n = {{\mathcal{E}}}^n_+$.
To obtain results concerning superlevel set persistence one requires the use of $$\label{def:E-}
{{\mathcal{E}}}_- := {\left\{ { h= (h_1,\ldots,h_n)\in {\mathbb{R}}^n : \text{$n$ is odd, and $h_{2k\pm 1} > h_{2k}$ for all $k$} }\right\}}$$
\[defcv\] We define a map $\operatorname{cv}\colon {\mathbb{R}}^N \to {{\mathcal{E}}}_+^n$ as follows. Assume $z\in {\mathbb{R}}^N$ has $n$ local extrema that are either interior extrema or boundary minima. Observe that this implies that $n$ is odd. Let ${i_j}$ be an index associated with the $j$-th extremum ordered such that $i_j < i_{j+1}$. Observe that $$\operatorname{cv}(z) := (z_{i_1},\ldots, z_{i_n})\in {{\mathcal{E}}}^n_+\subset{{\mathcal{E}}}_+.$$
\[ex:cv(z)\] Let $z=(1.5,-0.9,-0.9,1.1,2.1,1.4)\in {\mathbb{R}}^6$. The interior extrema are ${\left\{ {2,3}\right\}}$ and ${\left\{ {5}\right\}}$. The boundary extrema are ${\left\{ {1}\right\}}$ and ${\left\{ {6}\right\}}$. Thus $\operatorname{cv}(z)\in {{\mathcal{E}}}^3_+$ and $(i_1,i_2,i_3) = (2,5,6)$ or $(i_1,i_2,i_3) = (3,5,6)$. Finally, $h=\operatorname{cv}(z) = (-0.9,2.1,1.4)\in{{\mathcal{E}}}^3_+$.
As is discussed in Section \[sec:background\], if $\operatorname{cv}(z) = \operatorname{cv}(z')$, then ${\mathsf{Dgm}}(z) = {\mathsf{Dgm}}(z')$. Given $h\in{{\mathcal{E}}}_+$, set $$\label{eqCh}
C(h) := {\left\{ {z\in{\mathbb{R}}^N : \operatorname{cv}(z) = h}\right\}}.$$ Then, ${\mathsf{Dgm}}$ is constant on $C(h)$.
Using results of [@curry] in Section \[sec:background\] we prove the following result.
Let ${\mathsf{diag}}\in{\mathsf{Per}}$. Then the preimage of ${\mathsf{diag}}$ is composed of a finite number of mutually disjoint components $${\mathsf{Dgm}}^{-1}({\mathsf{diag}}) = \bigcup_{i=1}^{I({\mathsf{diag}})} C(h^i)$$ where $h^i\in {{\mathcal{E}}}_+$ and all the $h^i$, $i=1,\ldots, I({\mathsf{diag}})$ are related by permutations of the coordinates.
The main result of this paper is as follows.
\[thm:main\] Assume $z\in{\mathbb{R}}^N$ satisfies assumption [**A**]{}. Let $${\mathsf{Dgm}}^{-1}({\mathsf{Dgm}}(z)) = \bigcup_{i=1}^{I({\mathsf{Dgm}}(z))} C(\operatorname{cv}(z)^i).$$ Then for each $i$, $C(\operatorname{cv}(z)^i)$ is an union of a finite *contractible* set of *polytopes*.
Moreover, let ${\mathcal{B}}= \left\{z\in{\mathbb{R}}^N\colon z_1\in[-\bar{a},\bar{a}],\ z_N\in[-\bar{b},\bar{b}]\right\}$ for some $\bar{a},\bar{b} > 0$, and $z\in{\mathcal{B}}$ satisfies assumption [**A**]{}, then $C(\operatorname{cv}(z)^i)\subset{\mathcal{B}}$ is a *compact simplicial set*.
As is described below the proof of Theorem \[thm:main\] occupies the majority of this paper (Sections \[sec:background\] through \[seccontr\]). In Section \[preimageregions\] we prove Theorem \[thm:mainNdelta\]. This is an extension of Theorem \[thm:main\] to regions in ${\mathsf{Per}}$ of a particular form and is obtained by providing an explicit homotopy contracting the preimage of the region to the preimage of a single diagram that lies in the region. Theorem \[thmmaindyn\], described above, follows almost directly from Theorem \[thm:mainNdelta\], and is presented in Section \[secapplication\].
The proof of Theorem \[thm:main\] is extremely technical, involving an induction proof based on the number of persistence points or equivalently the number of local extrema pairs and requires substantial bookkeeping of subsets of ${\mathbb{R}}^N$.
We begin in Section \[sec:background\] with the necessary background information on sub-level set filtrations and persistence diagrams. We also recall the Nerve Theorem, which is the main tool that we use for our eventual proof of contractbility of the preimage set. The new content is a basic form of bookkeeping provided by convex polytopes covering the preimage set. By doing so we provide an easy proof that the preimage set $C(\operatorname{cv}(z)^i)$ is a finite union of polytopes (Proposition \[prop:Cpolytope\]).
Unfortunately, we are unable to prove contractibility using the representation presented in Section \[sec:background\]. Thus, in Section \[secindexing\] we introduce more complicated class of coverings of $C(\operatorname{cv}(z)^i)$. Fortunately, this class of coverings has the structure of a meet semi-lattice that we use extensively as a new form of bookkeeping.
Section \[secsubsets\] describes the preimage set in terms of star-shaped sets, generated by semi-lattice morphisms on the set of multi-indices. However, analyzing the topology of the nerve complex defined using those star-shaped sets directly turns out to be hard, and its intersection structure is unclear. Instead, we introduce a covering using more coarse sets.
Section \[secfirststep\] provides an argument for the initial step of the induction argument. To provide the reader with some intuition we illustrate the results with an example for a particular case.
Section \[secmotiv\] provides a motivating example for the contractibility inductive argument. We carry out formally the inductive argument for contractibility of the coarse covering in Section \[seccontr\].
Acknowledgements
----------------
The work of JC and KM was partially supported by grants NSF-DMS-1125174, 1248071, 1521771 and a DARPA contracts HR0011-16-2-0033.
In addition KM was partially supported by DARPA contract FA8750-17-C-0054 and NIH grant R01 GM126555-01.
Preliminaries {#sec:background}
=============
#### Notation
Throughout this paper $n$ and $N$ denote natural numbers where $n\leq N$.
Sublevel-set filtrations and persistence diagrams
-------------------------------------------------
Before we introduce finite dimensional dynamical systems we study in this paper, let us first introduce the crucial tool that we use to describe graphs determined by a finite number of points. We compute *persistent homology* of *sublevel-set filtrations* of one dimensional simplicial complexes. The vertices of the simplicial complex have associated values, determined by a graph. We define the *persistence map* as the a map assigning to each sublevel-set filtration its *persistence diagram* or (equivalently) *barcode*. Existing literature about persistent homology is very rich, see e.g. [@chazal; @comptop; @ZC] and references provided there. Here we briefly recall persistent homology theory to the extent it is required by our theory. We also prove some preliminary results about the preimage (fiber) of the persistence map.
Let $z\in{\mathbb{R}}^N$ be fixed. We define the persistence map as the map associating the filtered complex ${{\mathcal{S}_N^{{{\mathsf F}}}}}(z)$ its persistence diagram, or sometimes referred to as the *barcode* [@ghrist; @ZC].
Observe that ${{\mathcal{S}_N^{{{\mathsf F}}}}}(z)$ is an one dimensional filtered *cell complex*. The persistent homology of cell complexes is computed using by now standard *persistent homology* algorithms provided in [@comptop].
By *persistence diagram* we mean a set of points of the extended plane $$\mathbb{E} = {\mathbb{R}}\times {\mathbb{R}}\cup\{\infty\}.$$ We denote the space of persistence diagrams by $${\mathsf{Per}}\subset \mathbb{E},$$ For any diagram ${\mathsf{diag}}\in{\mathsf{Per}}$, and for each point $(b,d)\in{\mathsf{diag}}$, it holds that $b\leq d$. Moreover, in the one-dimensional setting there is necessarily one point having the coordinates $(b',\infty)$.
Here we do not study in detail the structure and decomposition of the persistence map. We define the *persistence map*, as the map which takes a function $f\colon{\mathbb{R}}^N\times{{\mathcal{S}_N}}\to{\mathbb{R}}$ and a vector $z\in{\mathbb{R}}^N$ to the persistence diagram encoding $0$-th order persistence homology of ${{\mathcal{S}_N^{{{\mathsf F}}}}}(z)$. Also, note that in our setting we deal only with the persistence map of degree $0$, associated with connected components.
It is known that the persistence map is $1$-Lipschitz, after equipping the category of persistent vector spaces with the *interleaving distance* [@chazal].
There exists a vast literature on algorithms computing the persistence diagram for general simplicial inputs, see e.g. [@comptop; @ZC] and references cited there. But, in the literature regarding computational homology the space of admissible functions is chosen in a way admitting a notion of *discrete Morse functions*, which in turn provides the standard notion of critical points. As a consequence, the cells where components of the sub-level set filtration are born/die can be indicated uniquely by algorithms computing the persistent homology. However, the algorithm can be easily altered to make it work in the setting of plateaus.
In our setting motivated by dynamical systems we do not make any assumption that would provide the standard notion of critical points. A function $f$ can include ’*plateaus*’, i.e. pieces of constant value. As a consequence the local extrema, where the components of the sublevel-set filtration are born or die are not necessary unique.
Let $${\mathsf{diag}}\in{\mathsf{Per}}$$ be a persistence diagram composed out of a finite number of points $\{p_1,p_2,\dots,p_n\}$.
Where the first point $p_1$ corresponds to the component of infinite persistence, we have $$\label{finpts}
{\mathsf{diag}}= \left\{p_1 = (p_1,\infty),\ p_2 = (p_2^b,p_2^d),\ \dots,\ p_n = (p_n^b,p_n^d)\right\},$$ where it holds that $n < \infty$, and $$p_i^b\neq p_j^b\text{ for }i\neq j\text{, and }p_j^d > p^d_{j+1}.$$ Due to the Elder Rule [@curry; @comptop] property of the persistence map it must hold that $$\label{p1less}
p_1 < p_j^b,\text{ for all }j>1.$$
The diagram ${\mathsf{diag}}$ has its associated *barcode*, which we denote by $B({\mathsf{diag}})$, and which is composed out of the following intervals $$\label{barcode}
I_1 = [p_1,\infty),\quad I_2 = [p_2^b,p_2^d),\ \quad\dots,\quad I_n = [p_n^b,p_n^d).$$ Observe that from the following inclusion on barcodes must hold $$I_j \subset I_1.$$
We say that a function $f(z,\cdot)$ realizes the barcode $B({\mathsf{diag}})$ if it holds that ${\mathsf{Dgm}}(f,z) = {\mathsf{diag}}$. It was shown in [@curry] that there is at least one piecewise linear function realizing the barcode, and this argument can be transferred directly to our setting.
To distinguish functions realizing the barcode $B({\mathsf{diag}})$, we group them into equivalence classes. We remark that to characterize equivalence classes we can use the notion of *chiral merge tree* as defined in [@curry].
It was shown in \[Corollary 5.5, [@curry]\] that the number of chiral merge trees realizing a barcode $B$ is equal to $2^{N-1}\prod_{j=2}^N{\mu_B(I_j)}$, where $\mu_B(I_j)$ denotes the number of intervals in $B$ containing $I_j$.
Now we recall the known result saying that ${\mathsf{Per}}^{-1}({\mathsf{diag}})$ is composed out of a finite number of components. For the moment we use an abstract notion $C(\operatorname{cv}(z)^i)$, which we will make precise later. Each component is represented by a *chiral merge tree*, which was proved in [@curry] in the case of *Morse* functions. The functions we consider in our setting are not necessarily Morse, below we provide a suitable lemma fitted to our setting.
\[lempreim\] Let ${\mathsf{diag}}\in{\mathsf{Per}}$. Then the preimage ${\mathsf{Dgm}}^{-1}({\mathsf{diag}})\subset {\mathbb{R}}^N$ is composed out of a finite number of components $${\mathsf{Dgm}}^{-1}({\mathsf{diag}}) = \bigcup_{i=1}^{I({\mathsf{Dgm}}(z))}{C(\operatorname{cv}(z)^i)},$$ where ${\mathsf{diag}}= {\mathsf{Dgm}}(z)$. $C(\operatorname{cv}(z)^i)\subset {\mathbb{R}}^N$ is such that for $i\neq j$ it holds that
- $C(\operatorname{cv}(z)^i)$, $C(\operatorname{cv}(z)^j)$ are not path connected, i.e. that for any $z\in C(\operatorname{cv}(z)^i)$ and $z'\in C(\operatorname{cv}(z)^j)$, there is no continuous path connecting $z$ with $z'$ in ${\mathsf{Dgm}}^{-1}({\mathsf{diag}})$.
- each $C(\operatorname{cv}(z)^i)$ is characterized by the equivalence class of the associated chiral merge tree.
The proof follows naturally from the results in [@curry]. If two vectors in ${\mathsf{Dgm}}^{-1}({\mathsf{diag}})$ are such that the *chiral merge trees* associated with filtration functions $f(z,\cdot)$, $f(z',\cdot)$ are different. It follows that $z, z'$ have the same number of extrema, whose heights are determined by the persistence diagram ${\mathsf{diag}}$, but appearing in different permutations. Hance any continuous path modifying $z$ into $z'$ must change the heights of the local minima, and hence, the persistence diagram along this path must not be constant. Therefore, there is no continuous path in ${\mathsf{Dgm}}^{-1}({\mathsf{diag}})$, which connects $z$ with $z'$.
The case of super-level set filtrations is completely analogous, and we leave redoing the super-level case as an exercise for the reader.
Topology of $C(h)$ {#sec:C(h)}
------------------
Let us fix example $h\in {{\mathcal{E}}}^n_+$. In this subsection we show that $C(h)$ defined in is the union of a finite set of polytopes.
The set of *strictly ordered multi-indices* is given by $$L_m^N := \left\{(y_1,\dots,y_m)\in\{1,2,\dots,N\}^m\colon y_j < y_{j+1}\text{ for }1 \leq j < m\right\}.$$
\[defmonotone\] Let $h\in {{\mathcal{E}}}^n_+$ and $y\in L_n^N$. A point $z\in{\mathbb{R}}^N$ is *$(y,h)$-monotone* if the following conditions are satisfied.
1. $z(y_k)=h_k$ for $k=1,\ldots,n$.
2. $z$ is monotone increasing on ${\left\{ {1,\ldots,y_1}\right\}}$ and monotone decreasing on ${\left\{ {y_n,\ldots,N}\right\}}$.
3. If $m$ is odd, then $z$ is monotone increasing on ${\left\{ {y_m,y_{m+1}}\right\}}$.
4. If $m$ is even, $z$ is monotone decreasing on ${\left\{ {y_m,y_{m+1}}\right\}}$.
Observe that if $z$ is $(y,h)$-monotone, then $\operatorname{cv}(z) = h$.
If $z$ is as in Example \[ex:cv(z)\], then $z$ is $(y,\operatorname{cv}(z))$-monotone for $y=(2,5,6)$ or $y = (3,5,6)$.
Given $h\in {{\mathcal{E}}}^n_+$ define $S_h = S_{h,N,n}\colon L_n^N \to 2^{{\mathbb{R}}^{N}}$ by $$\label{slprop}
S_h(y) = S_{h,N,n} (y) := \left\{ z\in{\mathbb{R}}^{N}\colon \text{$z$ is $(y,h)$-monotone} \right\}.$$ Let $$\bar{a},\bar{b}>0$$ denote global bounds imposed on the first and the last components respectively and denote the subset of ${\mathbb{R}}^N$ having boundary elements bounded by $$\label{compactB}
{\mathcal{B}}= {\mathcal{B}}(\bar{a},\bar{b}) = \left\{z\in{\mathbb{R}}^N\colon z_1\in[-\bar{a},\bar{a}],\ z_N\in[-\bar{b},\bar{b}]\right\}.$$ Let $h\in {{\mathcal{E}}}^n_+$, define the compact version of $S_h = S_{h,N,n}\colon L_n^N \to 2^{{\mathcal{B}}}$ by $$\label{slpropcompact}
S_h(y) = S_{h,N,n} (y) := \left\{ z\in{\mathcal{B}}\colon \text{$z$ is $(y,h)$-monotone} \right\}.$$ Because most of the arguments are precisely the same regardless if we consider a compact or noncompact version of $S_h$ (the application in Section \[secapplication\] is the sole exception), we do not provide distinct symbols for these two objects.
Let $n=3$, $N=7$, $h = (0.5,3.5,2)$, and $y=(2,4,6)$. The possible coordinate values of $z\in S_{h,N,n}(y)\subset {\mathbb{R}}^7$ are shown in Figure \[fig:Sh\]. Observe that $S_{h,N,n}(y)$ is an unbounded convex set.
Let us relate a preimage component $C(h)$ with the definition of $(y,h)$-monotonicity. First, observe that
\[eqCh2\] $$\begin{aligned}
C(h) &= \left\{ z\in{\mathbb{R}}^{N}\colon \text{$z$ is $(y,h)$-monotone for any $y\in L_n^N$} \right\},\text{ and}\\
C(h) &= \bigcup_{y\in L_n^N}{S_h(y)}.\label{Chsum}\end{aligned}$$ We define analogously the compact version of $C(h)\colon L_n^N \to 2^{{\mathcal{B}}}$, using compact $S_h$’s.
\[prop:Spolytope\] Given $h\in {{\mathcal{E}}}^n_+$, and $y\in L_n^N$. The set $S_h(y) = S_{h,N,n}(y)$ defined by is a *closed convex polytope*. $S_h(y)\subset{\mathcal{B}}$ defined as is additionally *compact*.
From the definition $S_h(y) = S_{h,N,n}(y)$ is the intersection of a finite number of half-spaces $z_1\geq z_2$, $z_2 \geq z_3,\dots,z_{y_1-1}\geq z_{y_1}=h_1$ etc.. Therefore $S_{h,N,n}(y)$ is a closed convex polytope. In case $z\in{\mathcal{B}}$ it is additionally compact, as the boundary points $z_1,z_N$ have restricted values.
\[prop:Cpolytope\] Given $h\in {{\mathcal{E}}}^n_+$. The set $C(h)$ is a finite union of polytopes. $C(h)\subset {\mathcal{B}}$ is a compact simplicial set.
Follows directly from Prop \[prop:Spolytope\] showing that for each $y$ $S_h(y)$ is a closed convex polytope (and in case $S_h(y)\subset{\mathcal{B}}$, $S_h(y)$ is compact), and the identity .
\[remcompact\] $S_h(y)$ defined by is also a *unbounded convex set*, where only the boundary points take the unbounded values, see Figure \[fig:Sh\]. However, $S_h(y)\subset{\mathcal{B}}$ defined as $S_h(y) := \left\{ z\in{\mathcal{B}}\colon \text{$z$ is $(y,h)$-monotone} \right\}$ is obviously a compact convex set. Analogously, in the unbounded case $C(h)\subset{\mathbb{R}}^N$ is a finite union of polytopes, and $C(h)\subset{\mathcal{B}}$ is additionally compact, hence a simplicial set.
All elements of the proof of Theorem \[thm:main\] are the same regardless if we consider the polytope sets $S_h(y)$ being unbounded (over ${\mathbb{R}}^N$) or compact (over ${\mathcal{B}}$).
(0,0) – (0,5); (0,0) – (8,0);
(-0.1,0.5) – (0,0.5); (-0.1,1) – (0,1); (-0.1,1.5) – (0,1.5); (-0.1,2) – (0,2); (-0.1,2.5) – (0,2.5); (-0.1,3) – (0,3); (-0.1,3.5) – (0,3.5);
(1,-0.1) – (1,0.1); (2,-0.1) – (2,0.1); (3,-0.1) – (3,0.1); (4,-0.1) – (4,0.1); (5,-0.1) – (5,0.1); (6,-0.1) – (6,0.1); (7,-0.1) – (7,0.1);
(2,0.5) circle (2pt); (4,3.5) circle (2pt); (6,2) circle (2pt);
(1,0.5) – (1,5.5); (3,0.5) – (3,3.5); (5,2) – (5,3.5); (7,2) – (7,5.5);
(1,-0.3) node [$z_1$]{}; (2,-0.3) node [$z_2$]{}; (3,-0.3) node [$z_3$]{}; (4,-0.3) node [$z_4$]{}; (5,-0.3) node [$z_5$]{}; (6,-0.3) node [$z_6$]{}; (7,-0.3) node [$z_7$]{};
(-0.3,1) node [$1$]{}; (-0.3,2) node [$2$]{}; (-0.3,3) node [$3$]{};
Nerve Theorem {#secnerve}
-------------
The fundamental tool in our study of the topology of the persistence map preimage is the Nerve Theorem, below we recall it as Theorem \[thmnerve\]. It is elementary to observe that any preimage set of interest ($C(h)$) is a finite union of polytopes, see Proposition \[prop:Cpolytope\]. However, the polytopes constructing $C(h)$ can have arbitrary topology. By finding an appropriate family of sets that covering $C(h)$ (denoted in the sequel by $\Lambda_h$’s), we show that the nerve of that covering is the full simplex, and hence, $C(h)$ is homotopy equivalent to a point (contractible). See also Example \[exn5\]. Let $\|\cdot\|$ denote the support of a simplicial complex, let $[n]=\{1,2,\dots,n\}$. Let ${{\mathcal{A}}}= \{A_1, A_2,\dots, A_n\}$ be a family of sets. The *nerve* records the “intersection pattern” of ${{\mathcal{A}}}$. It is the simplicial complex denoted by ${{\mathcal{N}}}({{\mathcal{A}}})$, with vertex set $[n]$ and with simplices given by $${{\mathcal{N}}}({{\mathcal{A}}}) = \left\{ {{\mathcal{J}}}\subset[n]\colon\,\bigcap_{i\in{{\mathcal{J}}}}{A_i}\neq\emptyset\right\}.$$
\[thmnerve\] Let $K_1, K_2,\dots, K_n$ be subcomplexes of a finite simplicial complex $K$ that together cover $K$ (each simplex of $K$ is in at least one $K_i$), and let $A_i:=\|K_i\|$. Suppose that the intersection $\cap_{i\in {{\mathcal{J}}}}{A_i}$ is empty or contractible for each nonempty ${{\mathcal{J}}}\subset [n]$. Then $$\|{{\mathcal{N}}}(\{A_1,A_2,\dots,A_n\})\| \simeq \|K\|;$$ i.e., the nerve is homotopy equivalent to $K$.
Meet semi-lattices {#secmeet}
------------------
*A *meet-semilattice* $(L,\wedge)$ is a set $L$ with a binary operation $\wedge\colon L\times L\to L$ satisfying*
1. (associativity) $w\wedge (x \wedge y) = (w\wedge x)\wedge y$,
2. (symmetry) $w\wedge x = x\wedge w$, and
3. (idempotency) $w\wedge w = w$.
A classical example of a meet-semilattice comes from set theory. Given a set $X$, let $2^X$ denote its power set. Then, $(2^X,\cap)$ is a meet-semilattice.
Indexing sets {#secindexing}
=============
As indicated in the Introduction the proof of Theorem \[thm:main\] makes use of a complicated class of coverings. The remainder of this section introduces the notation used to index these coverings.
\[def:calL\]
*For $n\leq N$, we denote the set of [*multi-indices*]{} by $\mathbb{L}_{2n}^N\subset \{1,\dots,N\}^{2n}$ and define it to consist of the vectors $w=( w_1,\dots,w_{2n} )$ that satisfy the following four conditions for all $k=1,\ldots, n-1$.*
1. $w_{2k-1}\leq w_{2k}$
2. $w_{2k-1}< w_{2k+1}$
3. $w_{2k} < w_{2k+2}$
4. $w_{2k+1} \leq w_{2k+2}$.
The order relation in $\mathbb{L}_{2n}^N$ is summarized in Figure \[fig:Lorder\]. Condition (ii) implies that the assumption that $n\leq N$ is necessary. Observe that if $n=N$, then $\mathbb{L}_{2n}^N = \{(1,1,2,2,\ldots,N,N)\}$ consists of a single element.
\(1) at (1,0) [$w_1$]{}; (2) at (0,1) [$w_2$]{}; (3) at (2,1) [$w_3$]{}; (4) at (1,2) [$w_4$]{}; (5) at (3,2) [$w_5$]{}; (6) at (2,3) [$w_6$]{}; (7) at (4,3) [$w_7$]{}; (8) at (3,4) [$w_8$]{};
\(1) edge\[-, dashed, shorten <= 2pt, shorten >= 2pt\] (2) (1) edge\[-,shorten <= 2pt, shorten >= 2pt\] (3) (2) edge\[-,shorten <= 2pt, shorten >= 2pt\] (4) (3) edge\[-,dashed,shorten <= 2pt, shorten >= 2pt\] (4) (3) edge\[-,shorten <= 2pt, shorten >= 2pt\] (5) (4) edge\[-,shorten <= 2pt, shorten >= 2pt\] (6) (5) edge\[-,dashed,shorten <= 2pt, shorten >= 2pt\] (6) (5) edge\[-,shorten <= 2pt, shorten >= 2pt\] (7) (6) edge\[-,shorten <= 2pt, shorten >= 2pt\] (8) (7) edge\[-,dashed,shorten <= 2pt, shorten >= 2pt\] (8) ;
\[ex:bbL34\] The simplest nontrivial example of a set of multi-indices is $$\begin{aligned}
\mathbb{L}_{4}^3 & = {\left\{ {(w_1,w_2,w_3,w_4) : w_1\leq w_2, w_3\leq w_4, w_1 < w_3, w_2 < w_4}\right\}} \\
& = {\left\{ {(1122),(1123),(1133),(1223),(1233),(2233)}\right\}}\end{aligned}$$
Define $\wedge \colon \mathbb{L}_{2n}^N\times \mathbb{L}_{2n}^N \to \mathbb{L}_{2n}^N$ by $w\wedge w' = ((w\wedge w')_1, \ldots ,(w\wedge w')_{2n})$ where $$(w\wedge w') _ k = \begin{cases}
\min\{ w_k,w'_k\} & \text{if $k$ odd,} \\
\max\{ w_k,w'_k\} & \text{if $k$ even.}
\end{cases}$$
\[thm:Lsemilattice\] $(\mathbb{L}_{2n}^N ,\wedge)$ is a meet-semilattice.
We begin by showing that if $w,w'\in\mathbb{L}_{2n}^N$, then $w'' = w\wedge w'\in \mathbb{L}_{2n}^N$. We need to show that $w'' = (w''_1,\ldots,w''_{2n})$ satisfies conditions (i)-(iv) of Definition \[def:calL\]. Since $w_{2k-1} \leq w_{2k}$ and $w'_{2k-1} \leq w'_{2k}$ $$w''_{2k-1}:= \min\{w_{2k-1},w'_{2k-1}\} \leq \max\{w_{2k},w'_{2k}\} = w''_{2k}$$ and thus (i) is satisfied. To see that (ii) is satisfied we note that $w_{2k-1} < w_{2k+1}$ and $w'_{2k-1} < w'_{2k+1}$ implies that $$w''_{2k-1}:= \min\{w_{2k-1},w'_{2k-1}\} < \min\{w_{2k+1},w'_{2k+1}\} = w''_{2k+1}.$$ Similar arguments apply to (iii) and (iv).
Checking that $\wedge$ is associative, symmetric, and idempotent is straightforward.
We denote the set of *ordered multi-indices* by $$\label{ordmultind}
\mathcal{L}_{2n}^N := \left\{ ( x_1,\dots,x_{2n} ) \in \{ 1,\dots,N\}^{2n}\colon x_{2k-1}\leq x_{2k}\text{ for $1\leq k \leq n$ and }x_{2k}<x_{2k+1}\text{ for $1\leq k \leq n-1$} \right\}$$ Observe that $\mathcal{L}_{2n}^N\subset \mathbb{L}_{2n}^N$. However, $\mathcal{L}_{2n}^N$ is not a sub meet-semilattice of $\mathbb{L}_{2n}^N$. To see this assume that $x,x'\in \mathcal{L}_{2n}^N$ and that $x'_{2k} < x'_{2k+1} < x_{2k} < x_{2k+1}$. Let $x'' = x'\wedge x$. Then $x''_{2k} = x_{2k} \not< x'_{2k+1} = x''_{2k+1}$, which implies that $x''\not\in\mathcal{L}_{2n}^N$.
\[ex:calL34\] The simplest nontrivial example of a set of ordered multi-indices is $$\begin{aligned}
\mathcal{L}_{4}^3 & = {\left\{ {(x_1,x_2,x_3,x_4) : x_1\leq x_2 < x_3\leq x_4}\right\}} \\
& = {\left\{ {(1122),(1123),(1133),(1233),(2233)}\right\}}\end{aligned}$$
Recall the strictly ordered multi-indices $L_{2n}^N$ defined in Section \[sec:C(h)\] and observe that $$L_{2n}^N\subset\mathcal{L}_{2n}^N\subset \mathbb{L}_{2n}^N.$$ Again, $L_{2n}^N$ is not a sub meet-semilattice of $\mathbb{L}_{2n}^N$.
We now define functions that takes arbitrary multi-indices and produces sets of pairs of strictly ordered multi-indices, i.e., maps $\varphi_i\colon \mathbb{L}^{N-1}_{2(n-1)}\to 2^{L^N_{n}\times L^N_{n}}$, $i=1,\ldots, n-2$. Set $$\label{varphik1}
\varphi_1(w) =
\begin{cases}
\left\{ (y,y)\in L^N_n\times L^N_n : y_1 \leq w_{1},\ w_{2} < y_{2} \leq w_{3}\right\} & \text{if $w_{2} < w_{3}$,} \\
\Big\{ (y,y')\in L^N_n\times L^N_n : y_1 \leq w_{1},\ y_{2} = w_{3},\ y'_{2} = w_{2}+1,\ \text{$y_j=y'_j$ for $j\neq 2$}\Big\} & \text{if $w_{3}\leq w_{2}$.}
\end{cases}$$ For $k=2,\dots,n-3$, define $\varphi_k= \varphi_{k,N,n}\colon \mathbb{L}_{2(n-1)}^{N-1} \to 2^{L_{n}^N}\times 2^{L_{n}^N}$ by $$\label{varphi}
\varphi_k(w) =
\begin{cases}
\left\{ (y,y)\in L^N_n\times L^N_n : w_{2k} < y_{k+1} \leq w_{2k+1}\right\} & \text{if $w_{2k}<w_{2k+1}$,} \\
\Big\{ (y,y')\in L^N_n\times L^N_n : y_{k+1} = w_{2k+1},\ y'_{k+1} = w_{2k}+1,\ \text{$y_j=y'_j$ for $j\neq k+1$}\ \Big\} & \text{if $w_{2k+1}\leq w_{2k}$.}
\end{cases}$$ And $$\label{varphiknm2}
\varphi_{n-2}(w) =
\begin{cases}
\left\{ (y,y)\in L^N_n\times L^N_n : w_{2(n-2)} < y_{n-1} \leq w_{2(n-2)+1},\ w_{2(n-1)} < y_{n} \right\} & \text{if $w_{2(n-2)}<w_{2(n-2)+1}$,} \\
\Big\{ (y,y')\in L^N_n\times L^N_n : y_{n-1} = w_{2(n-2)+1},\ y'_{n-1} = w_{2(n-2)}+1,\ w_{2(n-1)} < y_{n}\\
\qquad\qquad\qquad\qquad\qquad\qquad\quad\quad\quad\quad\quad\quad\text{$y_j=y'_j$ for $j\neq n-1$}\ \Big\} & \text{if $w_{2(n-2)+1}\leq w_{2(n-2)}$.}
\end{cases}$$
[ *If $n=3$ and $N=7$, then $\varphi_1\colon \mathbb{L}_{4}^{6} \to 2^{L_{3}^7}\times 2^{L_{3}^7}$. Furthermore, if $w=(2,4,3,6)$, then $$\varphi_1(w) = \big\{\big((1,3,7),(1,5,7) \big),\big( (2,3,7),(2,5,7) \big) \big\}.$$* ]{}
[ *If $n=4$ and $N=6$, then $\varphi_k\colon \mathbb{L}_{6}^{5} \to 2^{L_{4}^6}\times 2^{L_{4}^6}$, for $k =1,2$. Furthermore, if $w=(1,2,2,3,3,4)$, then $$\begin{aligned}
\varphi_1(w) & = \big\{\big((1,2,4,5),(1,3,4,5) \big),\big( (1,2,4,6),(1,3,4,6) \big),\big( (1,2,5,6),(1,3,5,6) \big) \big\} \\
\varphi_2(w) & = \big\{\big((1,2,3,5),(1,2,4,5) \big),\big( (1,2,3,6),(1,2,4,6) \big) \big\}\end{aligned}$$* ]{}
The action of all the $\varphi_k$ restricted to ordered multi-indices $\mathcal{L}_{2(n-1)}^{N-1}$ is relatively easy to describe. Define $\rho = \rho_{N,n} \colon \mathcal{L}_{2(n-1)}^{N-1} \to 2^{L_{n}^N}$ by $$\label{rhoo}
\rho(x) = \left\{ y \in L_{n}^N : y_1 \leq x_1,\ x_{2(k-1)} < y_k \leq x_{2(k-1)+1},\ x_{2(n-1)} < y_n,\ k=2,\ldots, n-1 \right\}$$
Let $n=4$, $N=6$. Then $\rho\colon {{\mathcal{L}}}_6^5 \to L_4^6$,
$$\rho(112244) = \left\{y\in L_4^6\colon y_1\leq 1,\ 1<y_2\leq 2,\ 2<y_3\leq 4,\ 4<y_4\right\} = \left\{ 1235, 1236, 1245, 1246 \right\}.$$
\[lem:rho\] Let $x\in \mathcal{L}_{2(n-1)}^{N-1}$. If $(y,y')\in \varphi_k(x)$, then $y'=y$ and $$\{(y,y) : y\in \rho(x)\} = \bigcap_{k=1}^{n-2} \varphi_k(x).$$
If $x\in \mathcal{L}_{2(n-1)}^{N-1}$, then $x_{2k}<x_{2k+1}$ and hence $$\varphi_k(x) = \{(y,y) : y_k\leq x_{2k-1},\ x_{2k}<y_{k+1} \leq x_{2k+1},\ x_{2(k+1)}< y_{k+2}\}.$$ Therefore, $$\bigcap_{k=1}^{n-2} \varphi_k(x) = \{(y,y) : y_1\leq x_{1},\ x_{2(k-1)} < y_k \leq x_{2(k-1)+1},\ x_{2(n-1)} < y_n,\ k=2,\ldots, n-1\}$$ which by definition of $\rho(x)$ is equal to the set $\{(y,y): y\in\rho(x)\}$.
For $n < N$, set $$\mathcal{P}_{2n}^{N}:={\left\{ { (x_1,\dots,x_{2n} ) \in \{1,\dots,N\}^{2n} : x_{2k-1}=x_{2k} < x_{2k+1}=x_{2k+2},\ k= 1,\ldots, n-1}\right\}} \subset \mathcal{L}_{2n}^{N}.$$ Observe that $\mathcal{P}_{2n}^{N}\subset \mathcal{L}_{2n}^{N}\subset \mathbb{L}_{2n}^{N}$, but it is not a sub-lattice of $\mathbb{L}_{2n}^N$.
\[ex:calP34\] The simplest nontrivial example is $$\begin{aligned}
\mathcal{P}_{4}^3 & = {\left\{ {(x_1,x_2,x_3,x_4) : x_1 = x_2 < x_3 = x_4}\right\}} \\
& = {\left\{ {(1122),(1133),(2233)}\right\}}\end{aligned}$$
We make use of ordered multi-indices (pairs) on the restricted domain, i.e. $$\mathcal{L}_2^{n-3,N-3} := \left\{ ( x_1,x_2 ) \in \{ n-3,\ldots,N-3\}^{2} : x_1\leq x_2 \right\}.$$ in conjunction with a map $\theta\colon \mathcal{L}_2^{n-3,N-3} \to 2^{\mathcal{P}^{N-1}_{2(n-1)}}$ given by $$\label{thetadef}
\theta((p_1,p_2)) := \left\{ (y_1,\dots, y_{2(n-1)}) : \ y_{2(n-3)}\leq p_1\leq p_2<y_{2(n-3)+1}\right\}.$$
Let $N=6$ and $n=5$. Then, $$\mathcal{L}_2^{n-3,N-3} = \mathcal{L}_2^{2,3} = \left\{ ( x_1,x_2 ) \in \{ 2,3\}^{2} : x_1\leq x_2\right\} =\left\{ ( 22 ), (23), (33) \right\}$$ Then $$\theta\colon \mathcal{L}_2^{2,3} \to 2^{\mathcal{P}^{5}_{8}},$$ and hence we have $$\begin{aligned}
\theta(22) &= \left\{ 11223344, 11223355, 11224455\right\},\\
\theta(23) &= \left\{ 11224455 \right\},\\
\theta(33) &= \left\{ 11224455, 11334455, 22334455\right\}.\end{aligned}$$
\[propthetap\] It holds that $$\label{eqpx}
\bigcup_{\substack{p\in\mathcal{L}_2^{n-3,N-3}\\p=(p_1,p_1)}}{ \theta(p_1,p_2) } = \mathcal{P}_{2(n-1)}^{N-1}.$$
Let $x\in\mathcal{P}_{2(n-1)}^{N-1}$. Let $p\in\mathcal{L}_2^{n-3,N-3}$ be such that $$p_1 = p_2 = x_{2(n-3)},$$ then from the definition it holds that $x\in\theta(p_1,p_2)$. Hence, as $x$ is arbitrary, we showed that it holds that $\bigcup_{\substack{p\in\mathcal{L}_2^{n-3,N-3}\\p = (p_1,p_2) }}{ \theta(p_1,p_1) } = \mathcal{P}_{2(n-1)}^{N-1}$.
\[propKwedgeM\] For any $p,q\in \mathcal{L}_2^{n-3,N-3}$ it holds that $$\theta((p_1,p_2))\cap\theta((q_1,q_2)) = \theta( (p_1, p_2) \wedge (q_1, q_2) ).$$
We need to show that $$\theta\left( (p_1,p_2)\wedge (q_1,q_2) \right) = \theta\left(\min\{p_1,q_1\},\max\{p_2,q_2\}\right) =
\theta\left((p_1,p_2)\right)\cap\theta\left((q_1,q_2)\right).$$ First, observe that from $(p_1, p_2),\,(q_1, q_2)\in \mathcal{L}_2^{n-3,N-3}$ it follows that $$(p_1, p_2) \wedge (q_1, q_2) = (\min\{p_1,q_1\},\max\{p_2,q_2\})\in \mathcal{L}_2^{n-3,N-3},$$ as if $p_1\leq p_2$ and $q_1\leq q_2$, then $\min\{p_1,q_1\}\leq \max\{p_2,q_2\}$.
From the definition of map $\theta$ , we have $$\theta\left((p_1,p_2)\right)\cap\theta\left((q_1,q_2)\right)=
\left\{ (y_1,\dots, y_{2(n-1)}) : \ y_{2(n-3)}\leq \min\{p_1,q_1\}\leq \max\{p_2,q_2\}<y_{2(n-3)+1}\right\}.$$ Therefore, we obtain the claim.
\[lemtheta\] Let $p=(p_1,p_2), q=(q_1,q_2) \in \mathcal{L}_2^{n-3,N-3}$. For any $w\in\theta((p_1,p_2))$, and $z\in\theta((q_1,q_2))$. It holds that $$w\wedge z < y,$$ for some $$y \in \theta((p_1, p_2)) \cap \theta((q_1, q_2)).$$
We have that $$w\wedge z = \left(\min\{ w_1 , z_1 \}, \max\{ w_2, z_2 \}, \min\{ w_3, z_3 \}, \max\{ w_4, z_4 \},\dots, \min\{ w_{2(n-1)-1}, z_{2(n-1)-1} \}, \max\{ w_{2(n-1)}, z_{2(n-1)} \}\right).$$ Let us take $$y = \left(\min\{ w_1 , z_1 \}, \min\{ w_2, z_2 \}, \max\{ w_3, z_3 \}, \max\{ w_4, z_4 \},\dots, \max\{ w_{2(n-1)-1}, z_{2(n-1)-1} \}, \max\{ w_{2(n-1)}, z_{2(n-1)} \}\right).$$
As $$\min\{w_{2(n-3)-1},z_{2(n-3)-1}\},\,\min\{ w_{2(n-3)}, z_{2(n-3)} \} \leq \min\{p_1,q_1\},$$ and $$\max\{p_2,q_2\} < \max\{ w_{2(n-3)+1}, z_{2(n-3)+1} \},\,\max\{ w_{2(n-2)}, z_{2(n-2)} \}.$$ It holds that $$y \in \theta((p_1,p_2)) \cap \theta((q_1,q_2)) = \theta((p_1,p_2)\wedge(q_1,q_2)).$$ And from the definition $$(w\wedge z)\wedge y = w\wedge z.$$ Hence $$w\wedge z < y.$$
Let $p=(p_1,p_2), q=(q_1,q_2) \in \mathcal{L}_2^{n-3,N-3}$. For any $w\in\theta((p_1,p_2))$, and $z\in\theta((q_1, q_2))$ it holds that $$A_h(w\wedge z) \subset A_h(y),$$ for some $$y \in \theta((p_1, p_2)) \cap \theta((q_1, q_2)).$$
Follows from Corollary \[corineq\] and Lemma \[lemtheta\].
Semilattice representation of Euclidean space {#secsubsets}
=============================================
In Section \[sec:C(h)\] it is shown that $C(h)$ is the union of a finite set of polytopes of the form $S_{h,N,n}(y)$. However, to prove Theorem \[thm:main\] we also need to show that $C(h)$ is contractible. We do this by describing $C(h)$ in terms of star-shaped sets, generated by semi-lattice morphisms $A_h$ defined on set of multi-indices $\mathbb{L}$. The focus of this section is on obtaining and characterizing this description.
\[defn:Ah\] Given $h\in {{\mathcal{E}}}^n_+$ define $A_h = A_{h,N,n}\colon \mathbb{L}_{2(n-1)}^{N-1} \to 2^{{\mathbb{R}}^{N}}$ by $$\label{aprop}
A_h(w) = A_{h,N,n}(w) = \bigcap_{k=1}^{n-2} \left( \bigcup_{(y,y')\in \varphi_k(w)} S_h(y)\cap S_h(y') \right)$$
Exactly the same arguments presented in this section are also true in the compact case, i.e. when the presented construction is based on the compact polytopes $S_h\subset{\mathcal{B}}$ defined using . This will ultimately allow for proving Theorem \[thm:main\] in the compact case.
We begin by analyzing how $A_h$ acts on $\mathcal{L}_{2(n-1)}^{N-1}$.
\[lemSintersect\] If $x\in \mathcal{L}_{2(n-1)}^{N-1}$, then $$\bigcap_{k=1}^{n-2}\bigcup_{(y,y)\in\varphi_k(x)}{S_h(y)} = \bigcup_{(y,y)\in\bigcap_{k=1}^{n-2}\varphi_k(x)}{S_h(y)}$$
Since $$\bigcup_{(y,y)\in\bigcap_{k=1}^{n-2}\varphi_k(x)}{S_h(y)}\subset\bigcap_{k=1}^{n-2}\bigcup_{(y,y)\in\varphi_k(x)}{S_h(y)}$$ is trivial, we focus on the opposite inclusion.
Assume $z\in \bigcap_{k=1}^{n-2}\bigcup_{(y,y)\in\varphi_k(x)}{S_h(y)}$. This implies that for each $k=1,\ldots, n-2$ there exists $y^k$ such that $(y^k,y^k)\in \varphi_k(x)$ and $$z\in \bigcap_{k=1}^{n-2} S_h(y^k).$$ Therefore $z$ is $(y^k,h)$ monotone for $k=1,\ldots, n-2$. Define $y\in L^N_n$ by $$y_1 := y_1^1,\quad y_j := y_j^{j-1},\quad y_n := y_n^{n-2}, \quad \text{for $j=2,\ldots,n-1$.}$$ Observe that $y\in \rho(x)$, and therefore, by Lemma \[lem:rho\], $(y,y)\in \bigcap_{k=1}^{n-2}\varphi_k(x)$. From the fact that $z$ is $y^j$ monotone for all $j$, $z$ is also $(y,h)$ monotone, and $$z\in \bigcup_{(y,y)\in\bigcap_{k=1}^{n-2}\varphi_k(x)}{S_h(y)}.$$
\[prop:restrdom\] If $x\in \mathcal{L}_{2(n-1)}^{N-1}$, then $$\label{acall}
A_h(x) = \bigcup_{y\in \rho(x)} S_h (y).$$
By Lemma \[lem:rho\](i) the definition of $A_h(x)$ reduces to $$\begin{aligned}
A_h(x) & = \bigcap_{k=1}^{n-2} \left( \bigcup_{(y,y)\in \varphi_k(x)} S_h(y) \right) \\
& = \bigcup_{(y,y)\in \bigcap_{k=1}^{n-2}\varphi_k(x)} S_h(y) \\
& = \bigcup_{y\in \rho(x)} S_h (y),\end{aligned}$$ where the second equality follows from Lemma \[lemSintersect\] and the final inequality from Lemma \[lem:rho\](ii).
From $\rho(112244) = \left\{ 1235, 1236, 1245, 1246 \right\}$, it follows $$A_h(112244) = S_h(1235)\cup S_h(1236)\cup S_h(1245)\cup S_h(1246).$$
The following proposition relates the sets defined by $A_h$ to $C(h)$.
\[prop:C(h)Ah\] Let $h\in \mathcal{E}_+^n$. Then, $$C(h) = \bigcup_{x\in\mathcal{L}_{2(n-1)}^{N-1}}{A_h(x)}.$$
By Proposition \[prop:restrdom\], if $x\in\mathcal{L}_{2(n-1)}^{N-1}$, then $A_h(x) = \bigcup_{y\in \rho(x)} S_h (y)$. Observe that for any $y\in L_n^N$ we can find some $x\in{{\mathcal{L}}}_{2(n-1)}^{N-1}$, such that $y\in\rho(x)$ (for instance, $x$ satisfies $x_1=y_1$, and $x_{2(k-1)}=y_{k}-1$, and $x_{2(k-1)+1}=y_k$ for $1\leq k < n$). Hence, it holds that $$\bigcup_{x\in\mathcal{L}_{2(n-1)}^{N-1}}{\bigcup_{y\in \rho(x)} S_h (y)}=\bigcup_{y\in L^N_n}{S_h(y)} = C(h),$$ where the latter equality follows from .
\[lemwflat\] Let $w\in \mathbb{L}_{2(n-1)}^{N-1}/\mathcal{L}_{2(n-1)}^{N-1}$ and assume that for some $k$, $w_{2k+1}\leq w_{2k}$. If $(y,y')\in \varphi_k(w)$ and $z\in S_h(y)\cap S_h(y')$, then $$z_i = h_{k+1},\quad \text{for $i = w_{2k+1},\dots, w_{2k}+1$.}$$
By definition of $\varphi_k$ $$y_{k+1} = w_{2k+1}\quad\text{and}\quad y'_{k+1} = w_{2k}+1.$$ By definition of $S_h$, $z$ is $(y,h)$ and $(y',h)$ monotone. Therefore, $z_{y_{k+1}} = h_{k+1} = z_{y'_{k+1}}$. Again invoking monotonicity, $z_j = h_{k+1}$ for all $j=w_{2k+1},\ldots,w_{2k}+1$.
We now turn our attention to the semilattice structure of $A_h$.
\[thmlatmorph\] For all $h\in {{\mathcal{E}}}^n_+$, $A_h\colon \mathbb{L}_{2(n-1)}^{N-1} \to 2^{{\mathbb{R}}^{N}}$ is a semi-lattice morphism.
Before presenting the proof we give an example and two corollaries.
\[ex:ALatticeMorph\] Let $w,w',w''\in\mathbb{L}_8^5$ where $w = 11223344$, $w' = 22334455$, and $w'' = 11223345$. Then the semi-lattice morphism for $A_h$ states that $$A_h(11223344)\cap A_h(22334455) = A_h(11223344\wedge 22334455) = A_h(12233445)$$ and $$A_h(11223344)\cap A_h(11223345) = A_h(11223344\wedge 11223345) = A_h(11223345).$$
The latter equality of Example \[ex:ALatticeMorph\] motivates the following two corollaries.
\[corineq\] Let $w,w'\in\mathbb{L}_{2n}^{N-1}$. If $w\wedge w' = w$, then $$A_h(w) \subseteq A_h(w').$$
From Theorem \[thmlatmorph\] $$A_h(w) = A_h(w\wedge w') = A_h(w)\cap A_h(w') \subseteq A_h(w').$$
\[cor:C(h)Ah2\] Let $h\in{{\mathcal{E}}}_+^n$ where $n\leq N$. Then $$C(h) = \bigcup_{x\in {{\mathcal{P}}}^{N-1}_{2(n-1)}} A_h(x).$$
By Proposition \[prop:C(h)Ah\], $C(h) = \bigcup_{x\in {{\mathcal{L}}}^{N-1}_{2(n-1)}} A_h(x)$. Since ${{\mathcal{P}}}^{N-1}_{2(n-1)}\subset {{\mathcal{L}}}^{N-1}_{2(n-1)}$, $$\bigcup_{x\in {{\mathcal{P}}}^{N-1}_{2(n-1)}} A_h(x) \subset \bigcup_{x\in {{\mathcal{L}}}^{N-1}_{2(n-1)}} A_h(x).$$
To prove the opposite inequality, consider $x\in {{\mathcal{L}}}^{N-1}_{2(n-1)}$. Define $x'\in {{\mathcal{P}}}^{N-1}_{2(n-1)}$, by $x'_{2k-1} = x'_{2k} = x_{2k}$ for $k=1,\ldots, n-1$. By definition $x_{2k-1}\leq x'_{2k-1} = x_{2k}$. Let $x'' := x'\wedge x$ and observe that $x''_{2k}=x'_{2k}$ and $x''_{2k-1}=x_{2k-1}$. Thus $x'\wedge x =x$. Therefore, $A_h(x)\subset A_h(x')$. Since this is true for every $x\in {{\mathcal{L}}}^{N-1}_{2(n-1)}$ $$\bigcup_{x\in {{\mathcal{L}}}^{N-1}_{2(n-1)}} A_h(x) \subset \bigcup_{x\in {{\mathcal{P}}}^{N-1}_{2(n-1)}} A_h(x).$$
The remainder of this section focuses on the proof of Theorem \[thmlatmorph\]. The following result characterizes the kernel of $A_h$.
\[thmemptyint\] Let $w\in \mathbb{L}_{2(n-1)}^{N-1}$, $n>3$. Then, $$A_h(w) = \emptyset\quad\text{if and only if}\quad w_{2k} + 1 \geq w_{2k+3}$$ for some $k=1,2,\dots, n-3$.
In case $n=3$, it holds that $$A_h(w)\neq\emptyset\text{ for all }w\in \mathbb{L}_4^{N-1}.$$
Let $n>3$. Let us pick $k\in\{1,2,\dots,n-3\}$. First, we show (by contradiction) that if $w_{2k}+1 \geq w_{2k+3}$, then $A_h(w) = \emptyset$.
Let us take $z\in S_h(y)\cap S_h(y')\cap S_h(v)\cap S_h(v')$ for some $(y,y')\in\varphi_k(w)$, and $(v,v')\in\varphi_{k+1}(w)$. It holds that
$$\begin{gathered}
\left\{\begin{array}{ll}
z_i = h_{k+1}&\text{ for some } w_{2k} < i \leq w_{2k+1} \text{(if $w_{2k}<w_{2k+1}$)},\\
z_i = h_{k+1}&\text{ for all }i = w_{2k+1},\dots,w_{2k}+1\ \text{(due to Lemma~\ref{lemwflat})}\end{array}\right.,\\
\text{and}\\
\left\{\begin{array}{ll}
z_j = h_{k+2}&\text{ for some } w_{2k+2} < j \leq w_{2k+3} \text{(if $w_{2k+2}<w_{2k+3}$)},\\
z_j = h_{k+2}&\text{ for all }j = w_{2k+3},\dots,w_{2k+2}+1\ \text{(due to Lemma~\ref{lemwflat}}))
\end{array}\right.\end{gathered}$$
The equations above show that we can find indices $(k,l)$, $k<l$, such that $z_k = h_{k+2}$, and $z_l = h_{k+1}$, which contradicts that $z$ is $(y,h)$, $(y',h)$, $(v,h)$, $(v',h)$-monotone. As $(y,y')\in\varphi_k(w)$, and $(v,v')\in\varphi_{k+1}(w)$ were arbitrary, hence $$\bigcup_{(y,y')\in\varphi_k(w)}{S_h(y)\cap S_h(y')}\cap\bigcup_{(v,v')\in\varphi_{k+1}(w)}{S_h(v)\cap S_h(v')} = \emptyset,$$ Therefore, in this case $$A_h(w) = \emptyset.$$
On the other hand we show that if $w_{2k}+1< w_{2k+3}$ for all $k\in\{1,2,\dots,n-3\}$, then we can construct a vector $z\in A_h(w)$. Let us pick multi-indices $y,y', v, v'$ depending on the cases considered below, and fix selected components of $z\in{\mathbb{R}}^N$ accordingly.
If $w_{2k} < w_{2k+1}$ take $(y,y') \in \varphi_k(w)$, from Lemma \[lem:rho\] $(y,y') = (y,y)$, and let $y$ be such that $$y_k = w_{2k}+1,\text{ fix } z_{w_{2k}+1} = h_{k+1},$$ otherwise, take any $(y,y')\in\varphi_k(w)$, and fix $$z_i = h_{k+1},\text{ for }i = w_{2k+1},\dots, w_{2k}+1.$$
If $w_{2k+2} < w_{2k+3}$ take $(v,v')\in\varphi_{k+1}(w)$, from Lemma \[lem:rho\] $(v,v') = (v,v)$, and let $v$ be such that $$v_k = w_{2k+3},\text{ fix }z_{w_{2k+3}}=h_{k+2},$$ otherwise, take any $(v,v')\in\varphi_{k+1}(w)$, and fix $$z_j = h_{k+2},\text{ for }j = w_{2k+3},\dots,w_{2k+2}+1.$$ Observe that in any of the cases considered above, due to the assumption $w_{2k}+1 < w_{2k+3}$, if $(k,l)$ are multi-indices such that $$z_k = h_{k+1},\text{ and }z_l = h_{k+2},\text{ then }k < l.$$ We may pick any values satisfying the monotonicity property for all of the remaining components of $z$.
Hence we showed that there exists $z$ which is $(y,h)$, $(y',h)$, $(v,h)$, $(v',h)$-monotone. Therefore, $$z\in S_h(y)\cap S_h(y')\cap S_h(v)\cap S_h(v').$$ Hence $$\bigcup_{(y,y')\in\varphi_k(w)}{S_h(y)\cap S_h(y')}\cap\bigcup_{(v,v')\in\varphi_{k+1}(w)}{S_h(v)\cap S_h(v')} \neq \emptyset,\text{ if }w_{2k} + 1 < w_{2k+3}.$$ Repeating this argument for all $k=1,2,\dots,n-3$, and assuming $w_{2k}+1 < w_{2k+3}$ for all $k$, we fix different components of $z$. Hence, if $w_{2k} + 1 \geq w_{2k+3}$ for all $k=1,2,\dots, n-3$ we find a $$z\in A_h(w).$$ Hence $$A_h(w) \neq \emptyset.$$
In case $n=3$, applying the argument above for $k=1,2$, we construct analogously a vector $z\in A_h(w)$.
[ *If $n=4$ and $N=6$, then $\varphi_k\colon \mathbb{L}_{6}^{5} \to 2^{L_{4}^6}\times 2^{L_{4}^6}$, for $k =1,2$. Furthermore, if $w=(1,2,2,3,3,4)$, then $$A_h(w) = \left[\bigcup_{(v,v')\in\varphi_1(w)}{S_h(v)\cap S_h(v')}\right]\cap\left[\bigcup_{(y,y')\in\varphi_2(w)}{S_h(y)\cap S_h(y')}\right] = \emptyset,$$ because $w_2 + 1 = 3 = w_5 = 3$.* ]{}
The following result provides the characteristics of nonempty sets in $A_h$ image.
\[threecontr\] Let $h\in{{\mathcal{E}}}^n_+$, $w\in \mathbb{L}_{2(n-1)}^{N-1}$. If $A_h(w)$ is nonempty, then $A_h(w)$ is star-shaped and hence contractible.
Let $h\in{{\mathcal{E}}}^n_+$, $w\in \mathbb{L}_{2(n-1)}^{N-1}$. Also assume that in case $n>3$ it holds that $w_{2k} + 1 < w_{2k+3}$ for all $k=1,2,\dots, n-3$, which from Theorem \[thmemptyint\] is equivalent to $A_h(w)$ being nonempty. We want to show that $A_h(w)$ is star-shaped. We proceed in two steps. First, we show that $\bigcup_{(y,y')\in \varphi_k(w)} S_h(y)\cap S_h(y')$ is star-shaped for each $k$. Observe that $S_h(y)\cap S_h(y')$ is a convex set for all $(y,y')$, as it is an intersection of two convex sets. A vector $\bar{z}\in {\mathbb{R}}^N$, such that $$\bar{z}\in S_h(y)\text{ for some }y\in L_n^N\text{ and }\bar{z}_{j} = h_{k+1}\text{ for }j=\left\{\begin{matrix}w_{2k}+1,\dots,w_{2k+1},&\text{ if }w_{2k}<w_{2k+1}\\
w_{2k+1},\dots,w_{2k}+1,&\text{ if }w_{2k}\geq w_{2k+1}\end{matrix}\right.$$ is $(y,h)$ and $(y',h)$-monotone for all $(y,y')\in\varphi_k(w)$, and hence $\bar{z}\in S_h(y)\cap S_h(y')$ for all $(y,y')\in \varphi_k(w)$, and ${\bigcup_{(y,y')\in \varphi_k(w)} S_h(y)\cap S_h(y')}$ is an union of convex sets having a common vector, and hence is star-shaped.
Second, we show that the whole family of star-shaped sets $\left\{\bigcup_{(y,y')\in \varphi_k(w)} S_h(y)\cap S_h(y')\right\}_{k=1}^{n-2}$ has a common vector. Let $\bar{z}\in{\mathbb{R}}^N$ be such that $$\bar{z}_{j} = h_{k+1}\text{ for }j=\left\{\begin{matrix}w_{2k}+1,\dots,w_{2k+1},&\text{ if }w_{2k}<w_{2k+1}\\
w_{2k+1},\dots,w_{2k}+1,&\text{ if }w_{2k}\geq w_{2k+1}\end{matrix}\right.\text{, and for all }k = 1,\dots, n-2,$$ for all the remaining coordinates $\bar{z}$ takes values, such that it is monotone, observe that conditions $w\in \mathbb{L}_{2(n-1)}^{N-1}$, and if $n>3$ also $w_{2k} + 1 < w_{2k+3}$ for $k=1,2,\dots,n-3$ guarantee existence of such vector $\bar{z}$.
Therefore, $A_h(w)$ is an intersection of star-shaped sets having a common vector, hence $A_h(w)$ is star-shaped itself.
\[lemsuminc\] Let $w,x \in \mathbb{L}_{2(n-1)}^{N-1}$. Let $k\in\{1,\dots,n-2\}$. Assume $$w_{2k-1} \leq x_{2k-1},\text{ and }w_{2k} \geq x_{2k},\text{ and }w_{2k+1} \leq x_{2k+1},\text{ and }w_{2k+2} \geq x_{2k+2}.$$ Then it holds that $$\bigcup_{(y,y')\in\varphi_k(w)}{S_h(y)\cap S_h(y')} \subset \bigcup_{(v,v')\in\varphi_k(x)}{S_h(v) \cap S_h(v')}.$$
Let $(y,y')\in\varphi_k(w)$. In case $w_{2k} < w_{2k+1}$ it holds that $(y,y')=(y,y)$ from Lemma \[lem:rho\], and $$y_k \leq w_{2k-1},\quad w_{2k}< y_{k+1} \leq w_{2k+1},\quad w_{2k+2} < y_{k+2}.$$ Due to the assumption it also holds $$y_k \leq x_{2k-1},\quad x_{2k}< y_{k+1} \leq x_{2k+1},\quad x_{2k+2} < y_{k+2}.$$ Hence $(y,y')\in\varphi_k(x)$.
In case $w_{2k} \geq w_{2k+1}$ it holds that $$y_k \leq w_{2k-1},\quad y_{k+1} = w_{2k+1},\quad y'_{k+1} = w_{2k}+1,\quad w_{2k+2} < y_{k+2},$$ Let us take $z\in S_h(y)\cap S_h(y')$. From Lemma \[lemwflat\], $z$ satisfies $$z_i = h_{k+1}\text{ for }i = w_{2k+1},\dots, w_{2k}+1.$$ And, in particular from the assumption it follows that $$z_i = h_{k+1}\text{ for }
\left\{ \begin{array}{ll}
i = x_{2k+1}, \dots, x_{2k}+1,&\text{ if }x_{2k}\geq x_{2k+1},\\
i \in \{ x_{2k}+1, \dots, x_{2k+1} \}&\text{ otherwise}.
\end{array}\right.$$ Also $z_{y_k} = h_k$, $z_{y_{k+2}} = h_{k+2}$, and due to the assumption $y_k \leq x_{2k-1}$, and $y_{k+2} > x_{2k+2}$. Hence $$z \in{S_h(v) \cap S_h(v')}\text{ for some }(v,v')\in\varphi_k(x).$$ Therefore, $$z \in\bigcup_{(v,v')\in\varphi_k(x)}{S_h(v) \cap S_h(v')}.$$
We need to show that $$\begin{aligned}
A_h(w\wedge x) & = A_h(w)\cap A_h(x) \\
\bigcap_{k=1}^{n-2} \left( \bigcup_{(y,y')\in \varphi_k(w\wedge x)} S_h(y)\cap S_h(y') \right) & =
\bigcap_{k=1}^{n-2} \left( \bigcup_{(v, v')\in \varphi_k(w)} S_h(v)\cap S_h(v') \right) \cap
\bigcap_{k=1}^{n-2} \left( \bigcup_{(u,u')\in \varphi_k(x)} S_h(u)\cap S_h(u') \right)\end{aligned}$$ Observe that it is sufficient to prove that for each $k=1,\ldots, n-2$ $$\label{eqmorphism}
\bigcup_{(y,y')\in \varphi_k(w\wedge x)} S_h(y)\cap S_h(y') =
\left( \bigcup_{(v,v')\in \varphi_k(w)} S_h(v)\cap S_h(v') \right) \cap
\left( \bigcup_{(u,u')\in \varphi_k(x)} S_h(u)\cap S_h(u') \right)$$
For fixed $k$ we break the proof into three cases: $w_{2k} < w_{2k+1}$ and $x_{2k} < x_{2k+1}$, $w_{2k} < w_{2k+1}$ and $x_{2k} \geq x_{2k+1}$, and $w_{2k} \geq w_{2k+1}$ and $x_{2k} \geq x_{2k+1}$. In each case, to show the left from the right inclusion in , we pick $z\in\text{r.h.s. of \eqref{eqmorphism}}$ and we construct multi-indices $(y,y')\in\varphi_k(w\wedge x)$, such that $z\in S_h(y)\cap S_h(y')$.
*Case 1: $w_{2k} < w_{2k+1}$ and $x_{2k} < x_{2k+1}$.* By the definition of $\varphi_k$ and Lemma \[lem:rho\], reduces to $$\label{eq:morphism1}
\bigcup_{(y,y')\in \varphi_k(w\wedge x)} S_h(y) \cap S_h(y') =
\left( \bigcup_{(v,v)\in \varphi_k(w)} S_h(v) \right) \cap
\left( \bigcup_{(u,u)\in \varphi_k(x)} S_h(u) \right).$$ Observe that the inclusion $$\bigcup_{(y,y')\in \varphi_k(w\wedge x)} S_h(y) \cap S_h(y') \subset
\left( \bigcup_{(v,v)\in \varphi_k(w)} S_h(v) \right) \cap
\left( \bigcup_{(u,u)\in \varphi_k(x)} S_h(u) \right)$$ follows from Lemma \[lemsuminc\]. To show the other inclusion, we take $(v,v)\in\varphi_k(w)$, and $(u,u)\in\varphi_k(x)$, then $$\label{uw}
v_k\leq w_{2k-1}\quad\text{and}\quad w_{2k}<v_{k+1}\leq w_{2k+1}\quad\text{and}\quad w_{2k+2}<v_{k+2}$$ and $$\label{ux}
u_k \leq x_{2k-1} \quad\text{and}\quad x_{2k}<u_{k+1}\leq x_{2k+1} \quad\text{and}\quad x_{2k+2}<u_{k+2}$$ respectively.
Observe that from , and it follows that
\[minmax\] $$\begin{aligned}
\min(v_k,u_k) &\leq \min(w_{2k-1},x_{2k-1}) = (w\wedge x)_{2k-1},\label{ineq1}\\
\max(v_{k+1},u_{k+1}) & > \max(w_{2k},x_{2k}) = (w\wedge x)_{2k},\label{ineq2}\\
\min(v_{k+1},u_{k+1}) & \leq \min(w_{2k+1},x_{2k+1}) = (w\wedge x)_{2k+1},\label{ineq3}\\
\max(v_{k+2},u_{k+2}) & > \max(w_{2k+2},x_{2k+2}) = (w\wedge x)_{2k+2}.\label{ineq4}\end{aligned}$$
Let $z \in S_h(v)\cap S_h(u)$. It follows from the monotonicity property (Def. \[defmonotone\]) that $$\begin{aligned}
z_{\min(v_k,u_k)}&=h_k,\\
z_i &= h_{k+1}\text{ for }i=\min(v_{k+1},u_{k+1}),\dots,\max(v_{k+1},u_{k+1}),\\
z_{\max(v_{k+2},u_{k+2})}&=h_{k+2}.\end{aligned}$$ Because $w\wedge x\in \mathbb{L}_{2(n-1)}^{N-1}$ we consider two subcases, either $(w\wedge x)_{2k} \geq (w\wedge x)_{2k+1}$ or $(w\wedge x)_{2k} < (w\wedge x)_{2k+1}$.
*Case A* $(w\wedge x)_{2k} \geq (w\wedge x)_{2k+1}$. Let $\bar{u},\bar{v}\in L_n^N$ be such that $$\begin{aligned}
\bar{u}_{k} = \bar{v}_k &= {\min(v_k,u_k)},\\
\bar{u}_{k+1} &= (x\wedge w)_{2k+1},\\
\bar{v}_{k+1} &= (x\wedge w)_{2k}+1,\\
\bar{u}_{k+2} = \bar{v}_{k+2} &= {\max(v_{k+2},u_{k+2})}.\end{aligned}$$
It holds that $(\bar{u},\bar{v})\in\varphi_k(w\wedge x)$.
Due to the inequalities , it holds that $$z_i = h_{k+1}\text{ also for }i = (w\wedge x)_{2k+1},\dots,(w\wedge x)_{2k}+1.$$
Hence, $z$ is $(\bar{u},h)$, $(\bar{v},h)$-monotone and therefore $$z\in \bigcup_{(\bar{u},\bar{v})\in \varphi_k(w\wedge x)} S_h(\bar{u}) \cap S_h(\bar{v}).$$
*Case B* $(w\wedge x)_{2k} < (w\wedge x)_{2k+1}$. Let $\bar{u}\in L_n^N$ be any multi-index satisfying $$\begin{aligned}
\bar{u}_{k}&= {\min(v_k,u_k)},\\
\bar{u}_{k+1} &\in \left\{ (w\wedge x)_{2k}+1,\dots,(w\wedge x)_{2k+1}\right\},\\
\bar{u}_{k+2} &= {\max(v_{k+2},u_{k+2})}.\end{aligned}$$
It holds that $(\bar{u},\bar{u})\in\varphi_k(w\wedge x)$.
Due to the inequalities , , it also holds that
$z_i = h_{k+1}$ for $i\in \left\{\max\left[\min(v_{k+1},u_{k+1}),(w\wedge x)_{2k}+1\right],\dots,\min\left[\max(v_{k+1},u_{k+1}),(w\wedge x)_{2k+1}\right]\right\}$. Hence, $z$ is $(\bar{u},h)$-monotone, and therefore $$z\in \bigcup_{(\bar{u},\bar{u})\in \varphi_k(w\wedge x)} S_h(\bar{u}).$$
*Case 2: $w_{2k} < w_{2k+1}$ and $x_{2k} \geq x_{2k+1}$ (and the symmetric case).* By the definition of $\varphi_k$ reduces to $$\bigcup_{(y,y')\in \varphi_k(w\wedge x)} S_h(y)\cap S_h(y') =
\left( \bigcup_{(v,v)\in \varphi_k(w)} S_h(v) \right) \cap
\left( \bigcup_{(u,u')\in \varphi_k(x)} S_h(u)\cap S_h(u') \right)$$
The inclusion $$\bigcup_{(y,y')\in \varphi_k(w\wedge x)} S_h(y)\cap S_h(y') \subset
\left( \bigcup_{(v,v)\in \varphi_k(w)} S_h(v) \right) \cap \left(\bigcup_{(u,u')\in \varphi_k(x)} S_h(u)\cap S_h(u') \right),$$ follows from Lemma \[lemsuminc\].
We now proceed to showing the inclusion $$\label{inc2}
\bigcup_{(y,y')\in \varphi_k(w\wedge x)} S_h(y)\cap S_h(y') \supset
\left( \bigcup_{(v,v)\in \varphi_k(w)} S_h(v) \right) \cap
\left( \bigcup_{(u,u')\in \varphi_k(x)} S_h(u)\cap S_h(u') \right).$$ Let us pick $$z\in S_h(v) \cap S_h(u)\cap S_h(u')$$ for some $(v,v)\in \varphi_k(w)$, and $(u,u')\in \varphi_k(x)$. Because $z$ is $(v,h)$, $(u,h)$, $(u',h)$-monotone. From the assumption $x_{2k} \geq x_{2k+1}$, it holds that $u_{k+1}=x_{2k+1}$, and $u'_{k+1}=x_{2k}+1$. And from the monotonicity property (Def. \[defmonotone\]), it holds that
\[mono1\] $$\begin{aligned}
z_{\min(v_k,u_k, u'_k)} &= h_k,\\
z_{i} &= h_{k+1}\text{ for }i = \min(v_{k+1},x_{2k+1}), \dots, \max(v_{k+1},x_{2k}+1),\\
z_{\max(v_{k+2},u_{k+2}, u'_{k+2})} &= h_{k+2}.\end{aligned}$$
Let $\bar{u},\bar{v}\in L_n^N$ be such that $$\begin{aligned}
\bar{u}_{k} = \bar{v}_k &= {\min(v_k,u_k, u'_k)},\\
\bar{u}_{k+1} &= (x\wedge w)_{2k+1},\\
\bar{v}_{k+1} &= (x\wedge w)_{2k}+1,\\
\bar{u}_{k+2} = \bar{v}_{k+2} &= {\max(v_{k+2},u_{k+2}, u'_{k+2})}.\end{aligned}$$ It holds that $$(\bar{u},\bar{v})\in\varphi_k(w\wedge x).$$
From the inequality $w_{2k} < v_{k+1} \leq w_{2k+1}$ it holds that $$\begin{aligned}
\min(v_{k+1},x_{2k+1}) \leq (x\wedge w)_{2k+1},\\
\max(v_{k+1},x_{2k}+1) \geq (x\wedge w)_{2k}+1.\end{aligned}$$ Therefore from it holds that $$z_i = h_{k+1}\text{ for }i = (x\wedge w)_{2k+1},\dots,(x\wedge w)_{2k}+1,$$ Hence $z$ is $(\bar{u},h)$, $(\bar{v},h)$-monotone. And, therefore $$z \in \bigcup_{(\bar{u},\bar{v})\in\varphi_k(x\wedge w)}{S_h(\bar{u})\cap S_h(\bar{v})},$$ i.e. we showed the inclusion .
*Case 3: $w_{2k} \geq w_{2k+1}$ and $x_{2k} \geq x_{2k+1}$.* By the definition of $\varphi_k$ and reduces to $$\bigcup_{(y,y')\in \varphi_k(w\wedge x)} S_h(y)\cap S_h(y') =
\left( \bigcup_{(v,v')\in \varphi_k(w)} S_h(v) \cap S_h(v') \right) \cap
\left( \bigcup_{(u,u')\in \varphi_k(x)} S_h(u) \cap S_h(u') \right)$$ Observe that the inclusion $$\bigcup_{(y,y')\in \varphi_k(w\wedge x)} S_h(y)\cap S_h(y') \subset
\left( \bigcup_{(v,v')\in \varphi_k(w)} S_h(v) \cap S_h(v') \right) \cap
\left( \bigcup_{(u,u')\in \varphi_k(x)} S_h(u) \cap S_h(u') \right)$$ follows from Lemma \[lemsuminc\].
Now, we show the inclusion $$\label{inc3}
\bigcup_{(y,y')\in \varphi_k(w\wedge x)} S_h(y)\cap S_h(y') \supset
\left( \bigcup_{(v,v')\in \varphi_k(w)} S_h(v) \cap S_h(v') \right) \cap
\left( \bigcup_{(u,u')\in \varphi_k(x)} S_h(u) \cap S_h(u') \right)$$
Let us pick $$z\in S_h(v) \cap S_h(v')\cap S_h(u)\cap S_h(u'),$$ for some $(v, v')\in \varphi_k(w)$, and $(u, u')\in \varphi_k(x)$. Because $z$ is $(v,h)$, $(v',h)$, $(u,h)$, $(u',h)$-monotone. From the assumptions $x_{2k} \geq x_{2k+1}$, $w_{2k} \geq w_{2k+1}$ and the monotonicity property (Def. \[defmonotone\]) it holds that $$\begin{aligned}
z_{\min(v_k,v'_k, u_k, u'_k)} &= h_k,\\
z_{i} &= h_{k+1}\text{ for }i = \min(x_{2k+1}, w_{2k+1}), \dots, \max(x_{2k}+1,w_{2k}+1),\\
z_{\max(v_{k+2},v'_{k+2}, u_{k+2}, u'_{k+2})} &= h_{k+2}.\end{aligned}$$ Let $\bar{u},\bar{v}\in L_n^N$ be such that $$\begin{aligned}
\bar{u}_{k} = \bar{v}_k &= z_{{\min(v_k,v'_k, u_k, u'_k)}},\\
\bar{u}_{k+1} &= (x\wedge w)_{2k+1},\\
\bar{v}_{k+1} &= (x\wedge w)_{2k}+1,\\
\bar{u}_{k+2} = \bar{v}_{k+2} &= z_{{\max(v_{k+2},v'_{k+2}, u_{k+2}, u'_{k+2})}}.\end{aligned}$$
It holds that $$(\bar{u},\bar{v})\in\varphi_k(w\wedge x).$$ And from the monotonicity property $$z_i = h_{k+1},\text{ for }i = (x\wedge w)_{2k+1},\dots,(x\wedge w)_{2k}+1,$$ Hence $z$ is $(\bar{u},h)$, $(\bar{v},h)$-monotone, and it holds that $$z \in \bigcup_{(\bar{u},\bar{v})\in\varphi_k(x\wedge w)}{S_h(\bar{u})\cap S_h(\bar{v})}.$$ And we showed the inclusion .
Contractibility for $n=3$ {#secfirststep}
=========================
As indicated in the introduction, the aim of this section is to establish the base case for the induction argument that proves Theorem \[thm:main\]. We state this base case in the form of the following theorem.
\[thm:baseCase\] If $z\in{\mathbb{R}}^N$ satisfies assumption [**A**]{} and $h=\operatorname{cv}(z)\in{{\mathcal{E}}}^3_+$, then $C(h)$ is contractible.
If $z\in{\mathcal{B}}$ satisfies assumption [**A**]{}, then $C(h)\subset{\mathcal{B}}$ is contractible and compact.
Because the proof is rather formal and provides little geometric insight, we first consider the simplest nontrivial example.
Let $N=4$ and $n=3$. Thus, we consider $h\in {{\mathcal{E}}}_+^3$. By Corollary \[cor:C(h)Ah2\] $$C(h) = \bigcup_{x\in\mathcal{P}_4^3}{A_h(x)},$$ where (see Example \[ex:calP34\]) $$\mathcal{P}_4^3 = \left\{ (1122), (1133), (2233) \right\}.$$
The first step is to check that the conditions of the Nerve Theorem (Thm. \[thmnerve\]) are satisfied for the following nerve simplicial complex with vertex set ${{\mathcal{P}}}_4^3$, and ${{\mathcal{A}}}= \{A_h(x)\}_{x\in{{\mathcal{P}}}_4^3}$, i.e. $${{\mathcal{N}}}({{\mathcal{A}}}) = \left\{ {{\mathcal{J}}}\subseteq{{\mathcal{P}}}_4^3\colon\,\bigcap_{x\in{{\mathcal{J}}}}{A_h(x)}\neq\emptyset \right\}.$$
By Theorem \[threecontr\] $A_h(x)$ is contractible for all $x\in \mathcal{P}_4^3$. To determine contractibility of $$\bigcap_{x\in{{\mathcal{J}}}\subset\mathcal{P}_4^3 }A_h(x)$$ we make use of the fact that $A_h$ is a semilattice morphism (Theorem \[thmlatmorph\]) and compute $$A_h\left(\bigwedge_{x\in{{\mathcal{J}}}\subset\mathcal{P}_4^3}x \right).$$ The following table indicates the action of $\wedge$ on ${{\mathcal{P}}}_4^3$.
$\wedge$ 1122 1133 2233
---------- ------ ------ ------
1122 1122 1123 1223
1133 1133 1233
2233 2233
Observe that all products are elements of $\mathbb{L}_4^3$ (see Example \[ex:bbL34\]), and therefore by Theorem \[threecontr\] their images under $A_h$ are contractible. Figure \[figcomplex1\] indicates the simplicial complex associated with the Nerve of the images of $A_h$.
![Simplicial complex nerve representation for the presented example with $N=4$ and $n=3$.[]{data-label="figcomplex1"}](simplex3.pdf){width="30.00000%"}
Using the so far established results we are prepared to show Theorem \[thm:baseCase\]. Observe that in case $n=3$ from Theorem \[thmemptyint\] it follows that $A_h(w)\neq\emptyset$ for any $w\in\mathbb{L}^{N-1}_4$. And then, from Theorem \[threecontr\] $A_h(w)$ is star-shaped for all $w\in\mathbb{L}^{N-1}_4$. We may summarize those facts as
\[corJ\] For any nonempty $\mathcal{J} \subset \mathcal{L}_4^{N-1}$ it holds that $$\bigcap_{x\in\mathcal{J}}{A_h(x)}$$ is contractible.
Let us define the following nerve simplicial complex with vertex set ${{\mathcal{L}}}_4^{N-1}$ for the family of sets ${{\mathcal{A}}}= \left\{ A_h(x)\right\}_{x\in{{\mathcal{L}}}_4^{N-1}}$ $${{\mathcal{N}}}({{\mathcal{A}}}) = \left\{{{\mathcal{J}}}\subseteq {{\mathcal{L}}}_4^{N-1}\colon\,\bigcap_{x\in{{\mathcal{J}}}}{A_h(x)}\neq\emptyset \right\}.$$ By Corollary \[corJ\] ${{\mathcal{N}}}({{\mathcal{A}}})$ is the full $N-1$-simplex, and consequently the set $$\bigcup_{x\in{{\mathcal{L}}}_4^{N-1}}{A_h(x)}$$ from the Nerve Theorem (Thm. \[thmnerve\]) has the homotopy type of a point, i.e. is contractible. By Proposition \[prop:C(h)Ah\] $$C(h) = \bigcup_{x\in\mathcal{L}_4^{N-1}}{A_h(x)},$$ and we are done. As noted earlier all results from Section \[secsubsets\] are also true in the compact case (Remark \[remcompact\]).
Motivating the induction step {#secmotiv}
=============================
As indicated in the introduction our proof of the induction step is fairly cumbersome. To motivate our approach and to give an outline of the argument we consider the simplest case beyond that of the previous section.
\[exn5\] Let $N=6$ and $n=5$. Then, $h\in {{\mathcal{E}}}_+^5$. By Corollary \[cor:C(h)Ah2\] $$C(h) = \bigcup_{x\in\mathcal{P}_8^5}{A_h(x)},$$ where $$\mathcal{P}_8^5 = \left\{ (11223344), (11223355), (11224455), (11334455), (22334455) \right\}.$$ The first step is to check that the conditions of the Nerve Theorem are satisfied for the following nerve simplicial complex with vertex set ${{\mathcal{P}}}_8^5$, and ${{\mathcal{A}}}= \{A_h(x)\}_{x\in{{\mathcal{P}}}_8^5}$, i.e. $${{\mathcal{N}}}({{\mathcal{A}}}) = \left\{ {{\mathcal{J}}}\subseteq{{\mathcal{P}}}_8^5\colon\,\bigcap_{x\in{{\mathcal{J}}}}{A_h(x)}\neq\emptyset \right\}.$$ By Theorem \[threecontr\] $A_h(x)$ is contractible for all $x\in \mathcal{P}_8^5$. To determine contractibility of $$\bigcap_{x\in{{\mathcal{J}}}\subset\mathcal{P}_8^5 }A_h(x)$$ we make use of the fact that $A_h$ is a semilattice morphism (Theorem \[thmlatmorph\]) and compute $$A_h\left(\bigwedge_{x\in{{\mathcal{J}}}\subset\mathcal{P}_8^5}x \right).$$ The following table indicates the action of $\wedge$ on ${{\mathcal{P}}}_8^5$.
$\wedge$ 11223344 11223355 11224455 11334455 22334455
---------- ---------- ---------- ---------- ------------ ------------
11223344 11223345 11223445 *11233445* *12233445*
11223355 11223455 11233455 *12233455*
11224455 11234455 12234455
11334455 12334455
Observe that all products are elements of $\mathbb{L}_8^5$, by Theorem \[thmlatmorph\] $A_h$’s indexed with any multiindex written using regular font is contractible, whereas by Theorem \[thmemptyint\] $A_h$’s indexed with any multiindex written using italics is empty, due to condition $w_{2k}+1\geq w_{2k+3}$ being satisfied for some $k$ (we have $A_h(11233445) = A_h(12233445) = A_h(12233455) = \emptyset$).
Thus, we only need to consider the following triples of $\wedge$ $$\begin{aligned}
11223344 \wedge 11223355 \wedge 11224455 = 11223345 \wedge 11224455 & = 11223445 \\
11223355 \wedge 11224455 \wedge 11334455 = 11223455 \wedge 11334455 & = 11233455 \\
11224455 \wedge 11334455 \wedge 22334455 = 11234455 \wedge 22334455 & = 12334455.\end{aligned}$$ Again, all products are elements of $\mathbb{L}_8^5$, and therefore, by Theorem \[thmlatmorph\] are contractible. For all higher order products $A_h$ is trivial. Therefore, we can apply Theorem \[thmnerve\] and use the nerve complex ${{\mathcal{N}}}({{\mathcal{A}}})$ to compute the homotopy or homology type of $C(h)$. ${{\mathcal{N}}}({{\mathcal{A}}})$ is presented on Figure \[figcomplex3\] and is clearly contractible.
![Simplicial ${{\mathcal{N}}}({{\mathcal{A}}})$ complex nerve representation in Example \[exn5\]. The presented vertices are all multiindices from ${{\mathcal{P}}}_8^5$. The edges and faces represent contractible intersections of corresponding $A_h$’s.[]{data-label="figcomplex3"}](simplex5.pdf){width="50.00000%"}
Unfortunately, the argument used in this example does appear to suggest a principle from which we can determine contractibility in general. With this in mind, we develop a coarser covering, such that the complex used to represent $C(h)$ is a full simplex.
\[defAt\] Let map $\theta$ be given by . Define the following ${\Lambda}_h\colon \mathcal{L}_2^{n-3,N-3} \to 2^{\mathbb{R}^{N}}$ by $${\Lambda}_h((p_1, p_2)) := \bigcup_{x\in \theta((p_1, p_2))} A_h(x).$$
\[propChLambda\] Let $h\in{{\mathcal{E}}}^n_+$, then $$C(h) = \bigcup_{p \in \mathcal{L}_2^{n-3,N-3}}{\Lambda_h((p_1,p_2))}.$$
By Definition $${\Lambda}_h((p_1, p_2)) = \bigcup_{x\in \theta((p_1, p_2))} A_h(x).$$
Hence from Proposition \[propthetap\] $$\bigcup_{p \in \mathcal{L}_2^{n-3,N-3}}{\Lambda_h((p_1,p_2))} = \bigcup_{x\in\mathcal{P}_{2(n-1)}^{N-1}}{A_h\left(x\right)} = C(h),$$ where the last equality follows from Proposition \[prop:C(h)Ah\].
\[thmintersect\] Let ${\Lambda}_h\colon\mathcal{L}_2^{n-3,N-3} \to 2^{\mathbb{R}^{N}}$ be the map defined in Def. \[defAt\].
Then ${\Lambda}_h\colon\mathcal{L}_2^{n-3,N-3} \to 2^{\mathbb{R}^{N}}$ is a semi-lattice morphism.
Let $p=(p_1,p_2), q=(q_1,q_2) \in \mathcal{L}_2^{n-3,N-3}$. We will show that it holds that $${\Lambda}_h((p_1, p_2))\,\cap\,{\Lambda}_h((q_1, q_2)) = \bigcup_{y\in\theta((p_1, p_2))\,\cap\,\theta((q_1, q_2))}{A_h(y)} = \bigcup_{y\in\theta((p_1, p_2)\wedge (q_1, q_2))}{A_h(y)} = \Lambda_h((p_1, p_2)\wedge (q_1, q_2)).$$
It follows from Corollary \[corineq\] and Lemma \[lemtheta\] that $${\Lambda}_h((p_1, p_2))\cap{\Lambda}_h((q_1, q_2)) = \bigcup_{ \substack{w\in\theta((p_1, p_2)) \\ z\in\theta((q_1, q_2))} }{A_h(w\wedge z)} \subset \bigcup_{y\in\theta((p_1, p_2))\,\cap\,\theta((q_1, q_2))}{A_h(y)}.$$ The other inclusion $$\bigcup_{y\in\theta((p_1, p_2))\,\cap\,\theta((q_1, q_2))}{A_h(y)} \subset {\Lambda}_h((p_1, p_2))\cap{\Lambda}_h((q_1, q_2)).$$ holds trivially.
From Proposition \[propKwedgeM\] it follows that $$\bigcup_{y\in\theta((p_1, p_2))\,\cap\,\theta((q_1, q_2))}{A_h(y)} = \bigcup_{y\in\theta((p_1, p_2)\wedge (q_1, q_2))}{A_h(y)} = \Lambda_h((p_1, p_2)\wedge (q_1, q_2)).$$
Let us recall that in the sequel we developed coarser and coarser covering of $C(h)$ in order to show that $C(h)$ is contractible. In we defined convex sets $S_h(y)$, whose union covers $C(h)$. However, it is not clear at all how to show directly contractibility of the nerve complex of the covering $$\bigcup_{y\in\mathcal{L}_n^N}{S_h(y)}.$$
Instead, as demonstrated in Section \[secfirststep\], we use a covering based on $A_h$ semi-lattice morphism, defined in Proposition \[prop:restrdom\].
We emphasize it is still not clear to show that the covering of $C(h)$ provided by $\{ A_h(x) \}_{x\in{{\mathcal{P}}}_{2(n-1)}^{N-1}}$ is contractible by directly investigating the nerve complex ${{\mathcal{N}}}\left(\{ A_h(x) \}\right)$. For the particular case presented as Example \[exn5\] the obtained complex ${{\mathcal{N}}}\left(\{ A_h(x) \}_{x\in{{\mathcal{P}}}_8^5}\right)$ is contractible, which is clearly readable from its picture on Figure \[figcomplex1\]. However, its not clear how to construct a scheme of proving contractibility in general. One natural way of deriving a general scheme would be to invoke the technique of mathematical induction. In fact, we develop an inductive scheme for proving the general case using a more coarse covering based on family $\left\{\Lambda_h((p_1,p_2))\right\}_{p\in\mathcal{L}_2^{{n-3},{N-3}}}$ defined in Definition \[defAt\]. Recall that the union of $A_h(x)$’s and the union of $\Lambda_h((p_1,p_2))$’s both cover $C(h)$,i.e. $$C(h) = \bigcup_{x\in{{\mathcal{P}}}_{2(n-1)}^{N-1}}{A_h(x)} = \bigcup_{p \in \mathcal{L}_2^{{n-3},{N-3}}}{\Lambda_h((p_1,p_2))}.$$ Still, advantage of using $\Lambda_h((p_1,p_2))$’s is that we can show that nerve complex ${{\mathcal{N}}}\left( \left\{\Lambda_h((p_1,p_2))\right\}_{p\in\mathcal{L}_2^{{n-3},{N-3}}}\right)$ is contractible by invoking an inductive scheme.
We now return to Example \[exn5\] and show that it is contractible using the book keeping that $\Lambda_h$ provides. By Proposition \[propChLambda\] $$C(h) = \bigcup_{p \in \mathcal{L}_2^{2,3}}{\Lambda_h((p_1,p_2))},$$ where $${{\mathcal{L}}}_2^{2,3} = \left\{ (2,2), (2,3), (3,3)\right\}.$$
The first step is to check that the conditions of the Nerve Theorem are satisfied for the following nerve simplicial complex with vertex set ${{\mathcal{L}}}_2^{2,3}$, and ${{\mathcal{A}}}= \{\Lambda_h((p_1,p_2))\}_{p\in{{\mathcal{L}}}_2^{2,3}}$, i.e. $${{\mathcal{N}}}({{\mathcal{A}}}) = \left\{ {{\mathcal{J}}}\subseteq{{\mathcal{L}}}_2^{2,3}\colon\,\bigcap_{p\in{{\mathcal{J}}}}{\Lambda_h((p_1,p_2))}\neq\emptyset \right\}.$$
To determine contractibility of $$\bigcap_{p\in{{\mathcal{J}}}\subset\mathcal{L}_2^{2,3} }\Lambda_h((p_1,p_2))$$ we make use of Theorem \[thmintersect\] to compute $$\Lambda_h\left(\bigwedge_{p\in{{\mathcal{J}}}\subset\mathcal{L}_2^{2,3}}p \right).$$ Clearly $$22\wedge 23 = 22\wedge 33 = 22\wedge 23\wedge 33 = 23,$$ and $$\theta(22) = \left\{11223344, 1223355, 11224455 \right\},\ \theta(23) = \left\{ 11224455 \right\},\ \theta(33) = \left\{ 11224455, 11334455, 22334455 \right\}$$ In Section \[thmind1\] we establish an inductive argument relating the contractibility of $\theta(22)$, $\theta(23)$, $\theta(33)$ with the contractibility of smaller complexes for $n=3$, i.e. $\bigcup_{x\in\mathcal{P}_{4}^{p_1}}A_{h',p_1,3}\left({x}\right)$, for $p_1=2,3$, that already were established to be contractible in Section \[secfirststep\]. It follows that all of the intersections $\bigcap_{p\in{{\mathcal{J}}}\subset\mathcal{L}_2^{2,3} }\Lambda_h((p_1,p_2))$ are contractible, and the nerve complex of the covering $C(h) = \bigcup_{p \in \mathcal{L}_2^{2,3}}{\Lambda_h((p_1,p_2))}$ is in fact a very basic one-dimensional simplex, presented on Figure \[figcomplex2\].
![Simplicial complex nerve representation in Example \[exn5\]. The presented vertices are all multiindices from $\mathcal{L}_2^{2,3}$. The edge is the intersection of the corresponding $\Lambda_h$’s.[]{data-label="figcomplex2"}](complexp.pdf){width="25.00000%"}
Contractibility of the preimage set $C(h)$ {#seccontr}
==========================================
The goal of this section is to prove the following theorem providing an inductive argument for contractibility. Let us start with an auxiliary definition
We define $$\begin{gathered}
\Pi_{M}\colon{\mathbb{R}}^{N} \to {\mathbb{R}}^{M},\ \Pi_{M}(x) = (x_1,\dots,x_{M}),\\
I-\Pi_{M}\colon{\mathbb{R}}^{N} \to {\mathbb{R}}^{N - M },\ (I-\Pi_{M})(x) = (x_{M+1},\dots, x_{N}).
\end{gathered}$$ With a slight abuse of notation we will use the same $\Pi$ notation defined above to denote projections of sets of vectors, $\Pi_M\colon 2^{{\mathbb{R}}^{N}}\to 2^{{\mathbb{R}}^{M}}$, and $I-\Pi_{M}\colon2^{{\mathbb{R}}^{N}} \to 2^{{\mathbb{R}}^{N - M }}$.
As suggested in Section \[secmotiv\] to show contractibility of $C(h)$ we first need to show that any of the sets $\Lambda((p_1,p_2))$ forming the coarse covering (Def. \[defAt\]) is contractible for $p\in{{\mathcal{L}}}_2^{n-3,N-3}$. The following theorem is the main result of this section, which relates contractibility of $\Lambda((p_1,p_2))$ to contractibility of a lower dimensional preimage set $C(h')$.
\[thmind1\] Let $\theta\colon \mathcal{L}_{2}^{{n-3},{N-3}} \to 2^{\mathcal{P}_{2(n-1)}^{N-1}}$ be the map defined by . Let $h\in\mathcal{E}^n_+,\ h' = \Pi_{n-2}{h},\ p=(p_1,p_2)\in \mathcal{L}_2^{{n-3},{N-3}},\ n \geq 5$.
If $$\bigcup_{x\in\mathcal{P}_{2(n-3)}^{p_1}}A_{h',p_1,\bold{n-2}}\left({x}\right)$$ is contractible, then $$\Lambda_h((p_1,p_2)) = \bigcup_{x\in\theta((p_1,p_2))}{A_{h,N,\bold{n}}\left(x\right)}$$ is contractible as well.
Theorem \[thmind1\] provides an inductive argument. The induction variable is $n$ (it is highlighted in the statement above using the bold font). $n$ determines the number of persistence points on the considered diagram and the number of local extrema, observe that the number of extrema changes by two for the induction step, which is in turn equivalent to a new persistence point appearing on the sublevel-set filtration diagram.
We have the following immediate corollary from Theorem \[thmind1\] providing a more explicit induction argument.
\[cormain\] Let $h'= \Pi_{n-2}h$. If $$C(h') = \bigcup_{x\in{{\mathcal{P}}}_{2(n-3)}^{p_1}}{A_{h',p_1,n-2}(x)}$$ is contractible for all $n-3\leq p_1\leq N-3$, then $$C(h) = \bigcup_{x\in{{\mathcal{P}}}_{2(n-1)}^{N}}{A_{h,N,n}(x)}$$ is contractible as well.
Assuming contractibility of $\bigcup_{x\in\mathcal{P}_{2(n-3)}^{p_1}}A_{h',p_1,n-2}\left({x}\right)$ for all $n-3\leq p_1\leq N-3$, from Theorem \[thmind1\] we have that $\Lambda_h((p_1,p_2))$ is contractible for all $p=(p_1,p_2)\in \mathcal{L}_2^{{n-3},{N-3}}$. Hence, the nerve complex $${{\mathcal{N}}}\left( \left\{\Lambda_h((p_1,p_2))\right\}_{p\in\mathcal{L}_2^{{n-3},{N-3}}}\right)$$ is the full $N-1$ dimensional simplex (intersection of any nonempty family is contractible), and therefore $C(h)$ is homotopy equivalent to a point.
Now, having already established our theoretical framework, a proof of the main result of this paper is straightforward.
Let $z\in{\mathbb{R}}^N$ be a vector satisfying assumption **A** (or analogously $z\in{\mathcal{B}}$ satisfying assumption **A**, and using $S_h(y)\subset{\mathcal{B}}$ defined as ) . Let $z$ has $n$ (even number) of local extrema, and $h\in{{\mathcal{E}}}^n_+$ be the vector of heights of the local extrema of $z$ fixing their ordering, i.e. $h = \operatorname{cv}(z)^i$. Let $h' = \Pi_3{h}$ be the projection onto the first three coordinates of $h$. From Theorem \[thm:baseCase\] it follows that $C(h')$ is contractible regardless the vector space dimension. Applying the induction from Theorem \[thmind1\] we obtain that $C(h)$ is contractible. Hence, in the case of the compact polytopes $S_h\subset{\mathcal{B}}$, we obviously have that $C(h)\subset{\mathcal{B}}$ is a contractible compact simplicial set.
Exactly the same arguments presented in this section are also true when using the compact building polytopes $S_h$, i.e. $S_h(y)\subset{\mathcal{B}}$, . This will ultimately allows for proving Theorem \[thm:main\] in the compact case.
To prove Theorem \[thmind1\] we need to first show some auxiliary results. First, in Theorem \[thmintersect\] we show that $\Lambda_h$ is a semi-lattice morphism, exactly like $A_h$. Then, we establish the final contractibility argument using several intermediate steps, the two auxiliary Lemma \[lemcontr\],\[lemcontr2\] proved below are necessary for proving the final contractibility argument from Theorem \[thmind1\].
To show the contractibility of $A_h$, we establish series of results proving that the intersection of any family of $\Lambda_h((k_1,k_2))'s$ is contractible.
Let us recall the definition $$\begin{aligned}
\theta&\colon \mathcal{L}_2^{{n-3},{N-3}} \to 2^{\mathcal{P}^{N-1}_{2(n-1)}},\\
\theta(p_1,p_2) &= \left\{ (y_1,\dots, y_{2(n-1)})\colon\ y_{2(n-3)}\leq p_1\leq p_2<y_{2(n-3)+1}\right\}.\end{aligned}$$ To prove the next proposition we define a slightly modified semi-lattice morphism $A_h$
\[defAtilde\] $$\begin{aligned}
\tilde{A}_{h,N,n-2} &\colon \mathbb{L}_{2(n-1)}^{N-1} \to 2^{\mathbb{R}^{N}},\\
\tilde{A}_{h,N,n-2}\left(w\right) &= A_{h,N,n-2}\left(\Pi_{2(n-3)}w\right)\cap\left[ \bigcup_{(y,y')\in\widetilde{\varphi}_{n-3}(w)}{S_{h,N,n-2}(y)\cap S_{h,N,n-2}(y')}\right]=\\
&=\left[\bigcap_{k=1}^{n-4}\bigcup_{(v,v')\in \varphi_k(w)}{S_{h,N,n-2}(v)\cap S_{h,N,n-2}(v')}\right] \cap\left[ \bigcup_{(y,y')\in\widetilde{\varphi}_{n-3}(w)}{S_{h,N,n-2}(y)\cap S_{h,N,n-2}(y')}\right]\end{aligned}$$
where $\widetilde{\varphi}_{n-3}\colon \mathbb{L}_{2(n-1)}^{N-1}\to L^N_{n-2}$ is a slightly modified indexing function $\varphi$ for the purpose of the next proposition, and is defined as follows $${\scriptsize
\label{varphin3}
\widetilde{\varphi}_{n-3}(w) =
\begin{cases}
\left\{ (y,y)\in L^N_{n-2}\times L^N_{n-2} : w_{2(n-3)} < y_{n-2} \leq w_{2(n-3)+1}\right\} & \text{if $w_{2(n-3)}<w_{2(n-3)+1}$,} \\
\Big\{ (y,y')\in L^N_{n-2}\times L^N_{n-2} : y_{n-2} = w_{2(n-3)+1},\ y'_{n-2} = w_{2(n-3)}+1,\ \text{$y_j=y'_j$ for $j\neq n-2$}\ \Big\} & \text{if $w_{2(n-3)+1}\leq w_{2(n-3)}$.}
\end{cases}}$$
\[proppieq\] Let $n\geq 5$, $h\in\mathcal{E}^n$. Let $ A_h\colon \mathbb{L}_{2(n-1)}^{N-1} \to 2^{\mathbb{R}^{N}}$ be the map defined by . Let $\tilde{A}_{h} \colon \mathbb{L}_{2(n-1)}^{N-1} \to 2^{\mathbb{R}^{N}}$ be defined by . Let $w\in \mathbb{L}^{N-1}_{2(n-1)}$.
It holds that $$\Pi_M A_h(w) = \Pi_M\tilde{A}_{h',N,n-2}(w),$$ where $h'=\Pi_{n-2}h$, and $M = \max(w_{2(n-3)}, w_{2(n-3)+1})$.
First, we show that $$\Pi_M\left[\bigcap_{k=1}^{n-3}{\bigcup_{(y,y')\in\varphi_k(w)}{S_h(y)\cap S_h(y')}}\right] = \Pi_M\tilde{A}_{h',N,n-2}(w),$$
From the inequalities provided in Def. \[def:calL\] it follows that $M < \min(w_{2(n-2)}, w_{2(n-2)+1})$, and hence $\varphi_{n-2}(w)$ affects only the monotonicity of the $(I-\Pi_M)$-projection of the vectors. Hence
$\bigcup_{(y,y')\in\varphi_{n-2}(w)}{\Pi_M\left[S_{h}(y)\cap S_{h}(y')\right]}$ is the set of all $(h',v)$-monotone vectors, where $v$ is any stricly ordered multi-index, $v\in{L}_M^{n-2}$, and $h' = \Pi_{n-2}h$ $$\bigcup_{(y,y')\in\varphi_{n-2}(w)}{\Pi_M\left[S_{h}(y)\cap S_{h}(y')\right]} = \bigcup_{v\in{L}_M^{n-2}}{S_{h',M,n-2}(v)},$$ and therefore we obtain our claim from the identity $$\begin{gathered}
\bigcap_{k=1}^{n-2}{\bigcup_{(y,y')\in\varphi_k(w)}{\Pi_M\left[S_{h}(y)\cap S_{h}(y')\right]}} =
\left\{\bigcap_{k=1}^{n-3}{\bigcup_{(y,y')\in\varphi_k(w)}{\Pi_M\left[S_{h}(y)\cap S_{h}(y')\right]}}\right\}\cap \left\{\bigcup_{(v,v')\in\varphi_{n-2}(w)}{\Pi_M\left[S_{h}(v)\cap S_{h}(v')\right]}\right\}\\
=\left\{\bigcap_{k=1}^{n-3}{\bigcup_{(y,y')\in\varphi_k(w)}{\Pi_M\left[S_h(y)\cap S_h(y')\right]}}\right\}\cap\left\{\bigcup_{v\in{L}_M^{n-2}}{S_{h',M,n-2}(v)}\right\}=
\bigcap_{k=1}^{n-3}{\bigcup_{(y,y')\in\varphi_k(w)}{\Pi_M\left[S_h(y)\cap S_h(y')\right]}}.\end{gathered}$$ The last equality holds due to the fact that $\bigcap_{k=1}^{n-3}{\bigcup_{(y,y')\in\varphi_k(w)}{\Pi_M\left[S_h(y)\cap S_h(y')\right]}}\subset \bigcup_{v\in{L}_M^{n-2}}{S_{h',M,n-2}(v)}$.
\[lemcontr\] Let $\theta\colon \mathcal{L}_2^{{n-3},{N-3}} \to 2^{\mathcal{P}_{2(n-1)}^{N-1}}$ be the map defined by . Let $n \geq 5$. Let $h\in\mathcal{E}^n$. Let $p = (p_1,p_2) \in \mathcal{L}_2^{{n-3},{N-3}}$. Let $ A_h\colon \mathbb{L}_{2(n-1)}^{N-1} \to 2^{\mathbb{R}^{N}}$ be the map defined by .
For any $$\label{Jsubsettheta}
\mathcal{J}\subset\theta((p_1, p_2)) \subset \mathcal{P}_{2(n-1)}^{N-1},\ \mathcal{J} \neq \emptyset$$ it holds that the contractibility of $$A_{h,N,n}\left(\wedge_{x\in\mathcal{J}}{x}\right) =
\left[\bigcap_{k=1}^{n-2} \left( \bigcup_{(y,y')\in \varphi_k(\wedge_{x\in\mathcal{J}}{x})} S_h(y)\cap S_h(y') \right)\right]$$ is equivalent to the contractibility of $$\tilde{A}_{h',N,n-2}\left(\wedge_{x\in\mathcal{J}}{x}\right) = A_{h',N,n-2}\left(\Pi_{2(n-3)}(\wedge_{x\in\mathcal{J}}{x})\right)\cap\left[ \bigcup_{(y,y')\in\widetilde{\varphi}_{n-3}(\wedge_{x\in\mathcal{J}}{x})}{S_{h',N,n-2}(y)\cap S_{h',N,n-2}(y')}\right],$$ where $h' = \Pi_{n-2}{h}$.
Let us denote $${x^\wedge}= \wedge_{x\in\mathcal{J}}{x}.$$
And $$M = \max({x^\wedge}_{2(n-3)}, {x^\wedge}_{2(n-3)+1}).$$
First, let us consider the case $$A_h({x^\wedge}) = \emptyset,\text{ or }A_{h',N,n-2}\left(\Pi_{2(n-3)}{x^\wedge}\right)=\emptyset,$$ then $$A_h\left({x^\wedge}\right) = \tilde{A}_{h',N,n-2}\left({{x^\wedge}}\right)= \emptyset.$$ Therefore, the claim holds in this case.
Second, we assume $$\bigcap_{k=1}^{n-3} \left( \bigcup_{(y,y')\in \varphi_k({x^\wedge})} S_h(y)\cap S_h(y') \right) \neq \emptyset.$$
Which from Theorem \[thmemptyint\] means that $$\label{eqindcond}
{x^\wedge}_{2k} + 1 < {x^\wedge}_{2k+3}\text{ for all $k=1,2,\dots, n-4$}.$$
Observe that due to the condition it holds that $$\label{yineq}
{x^\wedge}_{2(n-3)} \leq p_1,\text{ and } p_2 < {x^\wedge}_{2(n-3)+1},$$
It holds that $$\label{emptyset}
\left(\bigcup_{(y,y)\in\varphi_{n-2}({x^\wedge})}{S_h(y)}\right) \cap \left(\bigcup_{(v,v')\in\varphi_k({x^\wedge})}{S_h(v)\cap S_h(v')}\right) \neq \emptyset\text{ for any }k\in\{1,\dots,n-3\},$$ as we assumed , and it holds that $${x^\wedge}_{2(n-3)}+1 < {x^\wedge}_{2(n-3)+3},$$ which holds from ${x^\wedge}_{2(n-3)} < p_1$, and $p_2 < {x^\wedge}_{2(n-3)+3}$. Hence, holds from Theorem \[thmemptyint\].
First, assume that $\tilde{A}_{h',N,n-2}( {x^\wedge})$ is contractible. It means that there exists a vector $\bar{z}\in \tilde{A}_{h',N,n-2}({x^\wedge})$, and a continuous map $H\colon[0,1]\times \tilde{A}_{h',N,n-2}({x^\wedge}) \to \tilde{A}_{h',N,n-2}({x^\wedge})$ a deformation retraction of $\tilde{A}_{h',N,n-2}({x^\wedge})$ to the single point $\bar{z}$, such that for any $z\in \tilde{A}_{h',N,n-2}({x^\wedge})$ it holds that $$H([0,1],z)\in \tilde{A}_{h',N,n-2}({x^\wedge}),\ H(0,z) = z,\text{ and }H(1,z) = \overline{z}.$$
From Prop. \[proppieq\] we have the following identity $$\label{ipiid}
\Pi_{M}\tilde{A}_{h',N,n-2}({x^\wedge})=\Pi_{M}A_h({x^\wedge}),$$
We define the homotopy $G\colon[0,1]\times A_h({x^\wedge})\to A_h({x^\wedge})$ coordinate-wise. Let $z'\in A_h({x^\wedge})$. In general, it does not hold that $z'\in \tilde{A}_{h',N,n-2}({x^\wedge})$, but $\Pi_{M}z' \in \Pi_{M}\tilde{A}_{h',N,n-2}({x^\wedge})$ does hold from . $$\begin{aligned}
G_k(t,z') &= (\Pi_{M}H)_{k}(t, \Pi_{M}z'),\text{ for }k=1,\dots,M,\nonumber\\
G_k(t,z') &= t\cdot\bar{z}'_k + (1-t)z'_k,\text{ for }k=M+1,\dots,N,\nonumber
\end{aligned}$$
where $$\bar{z}' = (\underbrace{h_{n-1},\dots,h_{n-1}}_{{x^\wedge}_{2(n-1)} - M }, \underbrace{h_{n},\dots,h_{n}}_{N-{x^\wedge}_{2(n-1)}}).$$
It holds that $$G(1,z') = (\Pi_{M}{\bar{z}}, \bar{z}').$$ First, we show that $(\Pi_{M}\bar{z},\bar{z}') \in A_h\left({x^\wedge}\right)$. Observe that $$\bar{z}'\in \bigcup_{(y,y')\in\varphi_{n-2}({x^\wedge})}{(I-\Pi_{M})S_h(y)} = (I-\Pi_{M})\left(\bigcup_{(y,y')\in\varphi_{n-2}({x^\wedge})}{S_h(y)}\right) =(I-\Pi_{M})A_h\left({x^\wedge}\right),$$ because clearly $\bar{z}' \in (I-\Pi_{M})S_h(y)$ for all $(y,y')\in\varphi_{n-2}({x^\wedge})$. The last equality holds due to the $(I-\Pi)_{M}$-projection of the vectors in $A_h\left({x^\wedge}\right)$ is determined by $\varphi_{n-2}({x^\wedge})$ exclusively, as for any $(y,y')\in\varphi_{n-3}({x^\wedge})$ it holds that $y,y'\leq M$.
It also holds from that $$\Pi_{M}\bar{z} \in \Pi_{M}\tilde{A}_{h',N,n-2}( {x^\wedge}) = \Pi_{M}A_h({x^\wedge}) .$$
Now we show that $$G([0,1], z')\in A_h({x^\wedge}).$$ From the assumption it follows that $\Pi_MH([0,1],z')\in \Pi_{M}\tilde{A}_{h',N,n-2}({x^\wedge})$, hence
$\Pi_{M}G([0,1], z')\in \Pi_{M}\tilde{A}_{h',N,n-2}({x^\wedge}) = \Pi_{M}A_h({x^\wedge})$.
It reminds to show that ${(I-\Pi_{M})G([0,1],z')\in (I-\Pi_{M})A_h({x^\wedge})}$.
Observe that $$(I-\Pi_{M})G([0,1],z') = t\cdot\bar{z}' + (1-t)(I-\Pi_{M})z' \in \left(\bigcup_{(y,y')\in\varphi_{n-2}({x^\wedge})}{(I-\Pi_{M})S_h(y)}\right),$$ as $\bar{z}', (I-\Pi_{M})z' \in (I-\Pi_{M})S_h(y)$ for all $(y,y')\in\varphi_{n-2}({x^\wedge})$, and $\bigcup_{(y,y')\in\varphi_{n-2}({x^\wedge})}{(I-\Pi_{M})S_h(y)}$ is a union of convex sets, hence $$\begin{gathered}
t\cdot\bar{z}' + (1-t)(I-\Pi_{M})z' \in \left(\bigcup_{(y,y')\in\varphi_{n-2}({x^\wedge})}{(I-\Pi_{M})S_h(y)}\right) = \\=(I-\Pi_{M})\left(\bigcup_{(y,y')\in\varphi_{n-2}({x^\wedge})}{S_h(y)}\right) = (I-\Pi_{M})A_h\left({x^\wedge}\right).
\end{gathered}$$
Finally, the map $G$ is continuous, as it is a product of two continuous functions ($\Pi_{M}G=\Pi_{M}H$ is continuous by the assumption, $(I-\Pi_{M})G$ is a linear homotopy). Hence $G$ is a deformation retraction of $A_h\left({x^\wedge}\right)$ to the single point $(\Pi_{M}{\bar{z}}, \bar{z}')$.
Vice-versa, assuming that $A_h({x^\wedge})$ is contractible, i.e. there exists a continuous map $$G\colon[0,1]\times A_h({x^\wedge})\to A_h({x^\wedge}),$$ such that $$G(0,z) = z,\text{ and }G(1,z) = \bar{z},\text{ for all }z \in A_h({x^\wedge})$$ we define the continuous map $H\colon[0,1]\times \tilde{A}_{h',N,n-2}({x^\wedge}) \to \tilde{A}_{h',N,n-2}({x^\wedge})$ coordinate-wise $$\begin{aligned}
H_k(t,z') &= (\Pi_{M}G)_{k}(t, \Pi_{M}z'),\text{ for }k=1,\dots,M,\nonumber\\
H_k(t,z') &= t\cdot\bar{z}'_k + (1-t)z'_k,\text{ for }k = M+1,\dots,N,\nonumber
\end{aligned}$$ where $$\bar{z}' = \underbrace{(h_{n-1},\dots, h_{n-1})}_{N - M}.$$
Performing the same analysis as in the opposite case, we show that for any $z'\in \tilde{A}_{h',N,n-2}({x^\wedge})$ $$H(1, z) = (\Pi_{M}{\bar{z}}, \bar{z}') \in \tilde{A}_{h',N,n-2}({x^\wedge}),$$ and $$H([0,1],z) \in \tilde{A}_{h',N,n-2}({x^\wedge})$$ (observe that $(1-t)(I-\Pi_{M})z' +t\bar{z}'\in (I-\Pi_{M})\tilde{A}_{h',N,n-2}({x^\wedge})$ as $(1-t)(I-\Pi_{M})z' +t\bar{z}'$ is a non-decreasing vector for all $t\in[0,1]$). Hence $H$ is a deformation retraction of $\tilde{A}_{h',N,n-2}({x^\wedge})$ to a single point.
\[corind\] Let $\theta\colon \mathcal{L}_{2}^{{n-3},{N-3}} \to 2^{\mathcal{P}_{2(n-1)}^{N-1}}$ be the map defined by . Let $h\in \mathcal{E}^n_+$. Let $n \geq 5$.
If $$\bigcup_{x\in\theta((p_1,p_2))}{\tilde{A}_{h',N,n-2}\left({x}\right)}$$ is contractible, then $$\bigcup_{x\in\theta((p_1,p_2))}{A_{h,N,n}\left(x\right)}$$ is contractible as well.
Follows from Lemma \[lemcontr\].
From Lemma \[lemcontr\] for any $\mathcal{J}\subset\theta((p_1, p_2)) \subset \mathcal{P}_{2(n-1)}^{N-1}$, $\mathcal{J} \neq \emptyset$ the contractibility of $$\tilde{A}_{h',N,n-2}\left(\wedge_{x\in\mathcal{J}}{x}\right).$$ is equivalent to the contractibility of $$A_h\left(\wedge_{x\in\mathcal{J}}{x}\right)$$ Hence, as $\mathcal{J}$ is arbitrary the claim follows from the Nerve Theorem.
\[lemcontr2\] Let $\theta\colon \mathcal{L}_2^{{n-3},{N-3}} \to 2^{\mathcal{P}_{2(n-1)}^{N-1}}$ be the map defined by . Let $n \geq 5$. Let $h\in \mathcal{E}^n_+$. Let $p=(p_1,p_2) \in \mathcal{L}_2^{{n-3},{N-3}}$. Let $ A_{h',N,n-2}\colon \mathcal{L}_{2(n-3)}^{N-1} \to 2^{\mathbb{R}^{N}}$ be the map defined by .
For any $$\label{Jsubsettheta2}
\mathcal{J}\subset\theta((p_1, p_2)) \subset \mathcal{P}_{2(n-1)}^{N-1},\ \mathcal{J} \neq \emptyset$$ it holds that the contractibility of $$\tilde{A}_{h',N,n-2}\left(\wedge_{x\in\mathcal{J}}{x}\right) =\\
A_{h',N,n-2}\left(\Pi_{2(n-3)}(\wedge_{x\in\mathcal{J}}{x})\right)\cap\left[ \bigcup_{(y,y')\in\widetilde{\varphi}_{n-3}(\wedge_{x\in\mathcal{J}}{x})}{S_{h',N,n-2}(y)\cap S_{h',N,n-2}(y')}\right]$$ is equivalent to the contractibility of $$A_{h',M,n-2}\left(\Pi_{2(n-3)}\left(\wedge_{x\in\mathcal{J}}{x}\right)\right) =
\left[\bigcap_{k=1}^{n-4} \left( \bigcup_{\substack{(y,y')\in \varphi_k{(\Pi_{2(n-3)}\left(\wedge_{x\in\mathcal{J}}{x}\right))}}} S_{h',M,n-2}(y)\cap S_{h',M,n-2}(y') \right)\right],$$ where $M=\max\left(\left(\wedge_{x\in\mathcal{J}}{x}\right)_{2(n-3)}, \left(\wedge_{x\in\mathcal{J}}{x}\right)_{2(n-3)+1}\right)$, and $h' = \Pi_{n-2}{h}$.
Observe that Lemma \[lemcontr\] and Lemma \[lemcontr2\] together provide the following contractibility equivalence.
The contractibility of $$A_{h,N,n}\left(\wedge_{x\in\mathcal{J}}{x}\right) =
\left[\bigcap_{k=1}^{n-2} \left( \bigcup_{(y,y')\in \varphi_k(\wedge_{x\in\mathcal{J}}{x})} S_h(y)\cap S_h(y') \right)\right].$$ is equivalent to the contractibility of $$A_{h',M,n-2}\left(\Pi_{2(n-3)}\left(\wedge_{x\in\mathcal{J}}{x}\right)\right) =
\left[\bigcap_{k=1}^{n-4} \left( \bigcup_{\substack{(y,y')\in \varphi_k{(\Pi_{2(n-3)}\left(\wedge_{x\in\mathcal{J}}{x}\right))}}} S_{h',M,n-2}(y)\cap S_{h',M,n-2}(y') \right)\right],$$ where $h' = \Pi_{n-2}{h}$.
Let us denote $${x^\wedge}= \wedge_{x\in\mathcal{J}}{x}.$$
It holds that $$\label{PiMN}
A_{h',M,n-2}(\Pi_{2(n-3)}{x^\wedge}) = \Pi_M\tilde{A}_{h',N,n-2}({x^\wedge}),$$ as any $\Pi_Mz\colon z\in \tilde{A}_{h',N,n-2}({x^\wedge})$ is necessarily $(h',\Pi_{2(n-1)-3}{x^\wedge})$-monotone.
The contractibility of $A_{h',M,n-2}\left(\Pi_{2(n-3)}{x^\wedge}\right)$ follows trivially from the contractibility of
$\tilde{A}_{h',N,n-2}\left({x^\wedge}\right)$ (it is a projected set, see ). To show the opposite implication assume that $A_{h',M,n-2}\left(\Pi_{2(n-3)}{x^\wedge}\right)$ is contractible. It means that there exists a vector $\bar{z}\in A_{h',M,n-2}\left(\Pi_{2(n-3)}{x^\wedge}\right)$, and a continuous map $H\colon[0,1]\times A_{h',M,n-2}\left(\Pi_{2(n-3)}{x^\wedge}\right) \to A_{h',M,n-2}\left(\Pi_{2(n-3)}{x^\wedge}\right)$ a deformation retraction of $A_{h',M,n-2}\left(\Pi_{2(n-3)}{x^\wedge}\right)$ to the single point $\bar{z}$, such that for any $z\in A_{h',M,n-2}\left(\Pi_{2(n-3)}{x^\wedge}\right)$ it holds that $$H([0,1],z)\in A_{h',M,n-2}\left(\Pi_{2(n-3)}{x^\wedge}\right),\ H(0,z) = z,\text{ and }H(1,z) = \bar{z}.$$
Let $z'\in \tilde{A}_{h',N,n-2}\left({x^\wedge}\right)$. We define a homotopy $G\colon[0,1]\times \tilde{A}_{h',N,n-2}\left({x^\wedge}\right) \to \tilde{A}_{h',N,n-2}\left({x^\wedge}\right)$ as follows $$\begin{aligned}
G_k(t,z') &= H_k(t,\Pi_Mz')\text{ for }k = 1,\dots,M,\nonumber\\
G_k(t,z') &= (1-t)(I-\Pi_M)z'_k + t\bar{z}'_k\text{ for }k = M+1,\dots,N,\label{Gk}\end{aligned}$$ where $\bar{z}=(\underbrace{h_{n-1},\dots,h_{n-1}}_{N-M})$. It holds that $G(1,z') = (\bar{z},\bar{z}')$. First, it holds that $(\bar{z},\bar{z}')\in \tilde{A}_{h',N,n-2}\left({x^\wedge}\right)$, as $\bar{z}\in \Pi_M\tilde{A}_{h',N,n-2}\left({x^\wedge}\right)$, and also $(I-\Pi_M)(\bar{z},\bar{z}')$ satisfies the monotonicity property ($\bar{z}'$ is nondecreasing), hence $(\bar{z},\bar{z}')$ is $(h',\Pi_{2(n-3)+1}{x^\wedge})$-monotone.
Second, we show that $G([0,1],z')\in \tilde{A}_{h',N,n-2}\left({x^\wedge}\right)$. It holds that $\Pi_MG([0,1],z')\in A_{h',M,n-2}\left(\Pi_{2(n-3)}{x^\wedge}\right) = \Pi_M\tilde{A}_{h',N,n-2}\left({x^\wedge}\right)$. It reminds to show that $(I-\Pi_M)G([0,1],z')\in (I-\Pi_M)\tilde{A}_{h',N,n-2}\left({x^\wedge}\right)$. Observe that $(I-\Pi_M)\tilde{A}_{h',N,n-2}\left({x^\wedge}\right)$ is a convex set (it is essentially set of nondecreasing vectors), hence due to the straight line homotopy within the $(I-\Pi_M)$-projection, it holds that $(I-\Pi_M)G([0,1],z')\in (I-\Pi_M)\tilde{A}_{h',N,n-2}\left({x^\wedge}\right)$.
Finally, the map $G$ is continuous, as it is a product of two continuous functions ($H$ is continuous by the assumption, $(I-\Pi_{M})G$ is a linear homotopy). Hence $G$ is a deformation retraction of $\tilde{A}_{h',N,n-2}\left({x^\wedge}\right)$ to the single point $(\bar{z}, \bar{z}')$.
\[thmind\] Let $\theta\colon \mathcal{L}_{2}^{{n-3},{N-3}} \to 2^{\mathcal{P}_{2(n-1)}^{N-1}}$ be the map defined by . Let $h\in \mathcal{E}^n_+,\ h' = \Pi_{n-2}{h},\ p=(p_1,p_2)\in \mathcal{L}_2^{{n-3},{N-3}},\ n \geq 5$.
If $$\bigcup_{x\in\mathcal{P}_{2(n-3)}^{p_1}}A_{h',p_1,n-2}\left({x}\right)$$ is contractible, then $$\Lambda_h((p_1,p_2)) = \bigcup_{x\in\theta((p_1,p_2))}{A_{h,N,n}\left(x\right)}$$ is contractible as well.
Let $p\in\mathcal{L}_2^{{n-3},{N-3}}$. Let $$\mathcal{J}\subset\theta((p_1, p_2)),\quad \mathcal{J}\neq\emptyset.$$ From Theorem \[thmintersect\] it follows that $$\label{Awedge}
\bigcap_{x\in\mathcal{J}}{A_h(x)} = A_h( \wedge_{x\in\mathcal{J}}{x} ).$$ From Lemma \[lemcontr\] we have that the contractibility of is equivalent to the contractibility of $$\tilde{A}_{h',N,n-2}\left(\wedge_{x\in\mathcal{J}}{x}\right),$$ As $\mathcal{J}\neq\emptyset$ is arbitrary, as a result we obtain that if for all ${{\mathcal{J}}}$ $\tilde{A}_{h',N,n-2}\left(\wedge_{x\in\mathcal{J}}{x}\right)$ is contractible, and hence from Nerve Theorem $$\bigcup_{x\in\theta((p_1,p_2))}{\tilde{A}_{h',N,n-2}(x)},$$ is contractible, then for all ${{\mathcal{J}}}\neq\emptyset$ $A_h( \wedge_{x\in\mathcal{J}}{x})$ is contractible as well, and hence from Nerve Theorem $$\Lambda_h((p_1,p_2)) = \bigcup_{x\in\theta((p_1,p_2))}{A_h(x)}$$ is contractible as well.
Observe that $x_{2(n-3)+1}\leq p_1$. It holds that $$\tilde{A}_{h',N,n-2}(x)\subset \tilde{A}_{h',N,n-2}\left(\Pi_{2(n-3)}x,p_1,(I-\Pi_{2(n-3)+1})x\right)\text{ for all }x\in\theta((p_1,p_2)),$$ which follows from the inclusion $$\bigcup_{(y,y')\in\widetilde{\varphi}_{n-3}({x})}{S_{h',N,n-2}(y)\cap S_{h',N,n-2}(y')}\subset \bigcup_{(y,y')\in\widetilde{\varphi}_{n-3}({\left(\Pi_{2(n-3)}x,p_1,(I-\Pi_{2(n-3)+1})x\right)})}{S_{h',N,n-2}(y)\cap S_{h',N,n-2}(y')},$$ and the obvious equality $$\Pi_{2(n-3)}({x}) = \Pi_{2(n-3)}\left(\Pi_{2(n-3)}{x},p_1,(I-\Pi_{2(n-3)+1}){x}\right),$$ holding for all $x\in\theta((p_1,p_2))$ – compare .
Hence, some sets merge up in the union, and we get $$\bigcup_{x\in\theta((p_1,p_2))}{\tilde{A}_{h',N,n-2}(x)} = \bigcup_{x\in\theta((p_1,p_2))}{\tilde{A}_{h',N,n-2}\left(\Pi_{2(n-3)}x,p_1,(I-\Pi_{2(n-3)+1})x\right)},$$
Moreover, from Lemma \[lemcontr2\] we have that for any $\mathcal{J}\subset\theta((p_1,p_2))$ the contractibility of
$\tilde{A}_{h',N,n-2}\left((\Pi_{2(n-3)}x,p_1,(I-\Pi_{2(n-3)+1})x)\right)$ is equivalent to the contractibility of $A_{h',p_1,n-2}\left(\Pi_{2(n-3)}x\right)$ (it holds that $M=p_1$). As for any $x\in\theta((p_1,p_2))$, we have that $\Pi_{2(n-3)}x\in \mathcal{P}_{2(n-3)}^{p_1}$, finally, we obtain that if $$\bigcup_{y\in\mathcal{P}_{2(n-3)}^{p_1}}A_{h',p_1,n-2}\left({y}\right).$$ is contractible, then $\Lambda_h((p_1,p_2))$ is contractible as claimed.
Contractibility of the preimage of regions in the persistence plane {#preimageregions}
===================================================================
\[defrregion\] Let $${\mathfrak{R}}\subset{\mathsf{Per}}$$ be a set of points on the persistence plane, such that each ${\mathsf{diag}}\in{\mathfrak{R}}$ is composed out of a finite number ($M>2$) of points ${\mathsf{diag}}=\{p_1,p_2,\dots,p_M\}$, $p_1 = (p_1,\infty),\ p_2 = (p_2^b, p_2^d),\ \dots,\ p_M = (p_M^b, p_M^d)$.
Assume that ${\mathfrak{R}}$ has the following structure $${\mathfrak{R}}= \bigcup_{i=1}^M{R_i}\text{, where $R_i\subset \mathbb{E}$, and}$$ $R_1 = [b_1^-, b_1^+]\times\{\infty\}$, $R_i = [b_i^-, b_i^+]\times[d_i^-, d_i^+]$, where $b_i^-<b_i^+$, and $d_i^-<d_i^+$ (when $b_i^-=b_i^+, d_i^-=d_i^+$ then it is a point diagram case considered in the previous sections). We assume that the regions have empty intersections, i.e. it holds that $$\label{bcond}
b_i^+< b_{i+1}^-\text{ for all }i=1,\dots,M-1,$$ and $$\label{dcond}
d_i^+ < d_{i+1}^-\text{ for all }i=2,\dots,M-1\quad(d_1=\infty).$$
\[defdregion\] Let $\delta>0$ be a small parameter. Let $${\mathfrak{D}}_\delta \subset {\mathsf{Per}}$$ be a set of persistence diagrams, such that each ${\mathsf{diag}}\in{\mathfrak{D}}_\delta$ is composed out of a finite $(M_\delta>0)$ number of points $\{r_1,r_2,\dots,r_{M_\delta}\}$ satisfying $0\leq r_i^d - r_i^b \leq \delta$, and $$\label{rcond}
[r_j^b,r_j^d]\nsubseteq [r_i^b,r_i^d]\text{ for all $j\neq i$.}$$
Observe that it may hold that $r_i^d - r_i^b=0$ for a nonempty set of indices, i.e. some of the points in ${\mathsf{diag}}$ can be located precisely on the diagonal.
We use the following notation for the union of persistence plane regions defined in Defs. \[defrregion\],\[defdregion\] $$\label{eqnregion}
{\mathfrak{N}}_\delta = {\mathfrak{R}}\cup {\mathfrak{D}}_\delta.$$ For a given diagram ${\mathsf{diag}}\in{\mathfrak{N}}_\delta$ we denote its projection onto its part contained in ${\mathfrak{R}}$ and ${\mathfrak{D}}_\delta$, by $\Pi_{{\mathfrak{R}}}$ and $\Pi_{{\mathfrak{D}}_\delta}$ respectively.
The motivation behind the conditions , and is that the functions that get mapped to ${\mathfrak{N}}_\delta$ by ${\mathsf{Dgm}}$ have clearly separated extrema. The definition of $\operatorname{cv}$ (Definition \[defcv\]) has a natural extension to the multi-valued map $\operatorname{cv}\colon 2^{{\mathbb{R}}^N}\to 2^{{{\mathcal{E}}}_+}$ for class of sets ${\mathfrak{N}}_\delta$ as for the extrema heights values $z_{i_j}^-, z_{i_j}^+$ it follows from , , that it hold $z_{i_{j-1}}^+ < z_{i_j}^-$, which basically imply that $\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}_\delta))$ is well defined and $\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}_\delta))\subset {{\mathcal{E}}}_+$. All the other sets $\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}_\delta))^j$ for $i\neq j$ are related by permuting the vector of extrema heights (such that it is still in ${{\mathcal{E}}}_+$), the number of such permutations is denoted by $I({\mathsf{Dgm}}^{-1}({\mathfrak{N}}_\delta))$. $z_{i_j}^-$ is the minimal value of the interval of the extremum height values, and $z_{i_j}^+$ is the maximal value of the interval of the extremum height values.
Using this new definition of $\operatorname{cv}$, Lemma \[lempreim\] can be also obtained for the preimage of ${\mathfrak{N}}_\delta$. Namely, we have the following corollary
Let $\delta>0$ be a small parameter. Let ${\mathfrak{N}}= {\mathfrak{N}}_\delta\subset {\mathsf{Per}}$, be a region on persistence plane defined in . Then the preimage ${\mathsf{Dgm}}^{-1}({\mathfrak{N}})\subset {\mathbb{R}}^N$ is composed out of a finite number of components, i.e. $${\mathsf{Dgm}}^{-1}({\mathfrak{N}}) = \bigcup_{i=1}^{I({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))}{C(\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))^i)},$$ $C(\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))^i)\subset {\mathbb{R}}^N$ is such that for $i\neq j$ it holds that
- $C(\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))^i)$, $C(\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))^j)$ are not path connected, i.e. that for any $z\in C(\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))^i)$ and $z'\in C(\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))^j)$, there is no continuous path connecting $z$ with $z'$ in ${\mathsf{Dgm}}^{-1}({\mathfrak{N}})$.
- each $C(\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))^i)$ is characterized by the equivalence class of the associated chiral merge tree.
Now let us focus on analyzing a given component $C(\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))^i)$.
We also naturally extend the definition of map $C$ (Definition \[eqCh\]) for sets ${\mathcal{H}}\subset{{\mathcal{E}}}^n_+$ $$C({\mathcal{H}}) = \left\{ z\in{\mathbb{R}}^{N}\colon \text{$z$ is $(y,h)$-monotone for any $y\in L_n^N$, $h\in{\mathcal{H}}$} \right\},\text{ and}\ C({\mathcal{H}}) = \bigcup_{\substack{y\in L_n^N\\h\in{\mathcal{H}}}}{S_h(y)}.$$ The analogous result to Proposition \[prop:Spolytope\] holds, i.e.
Let $y\in L_n^N$, $\delta>0$ a small parameter. Given ${\mathfrak{N}}={\mathfrak{N}}_\delta$ a set satisfying , and ${\mathcal{H}}=\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))^i \subset {{\mathcal{E}}}_+$ one of the sets defined by map $\operatorname{cv}$ applied to ${\mathfrak{N}}$. The set $\bigcup_{h\in{\mathcal{H}}}{S_{h}(y)}$ is a closed convex polytope.
If $S_h(y)\subset{\mathcal{B}}$ is defined as , and ${\mathcal{H}}\subset {{\mathcal{E}}}_+^n$, then $\bigcup_{h\in{\mathcal{H}}}{S_{h}(y)}\subset{\mathcal{B}}$ is additionally compact.
From the definition of $S_{h}(y)$, $\bigcup_{h\in{\mathcal{H}}}{S_{h}(y)}$ is the intersection of a finite number of half-spaces $z_1\geq z_2,\ z_2\geq z_3,\ \dots\ , z_{y_1-1} \geq g_1,\ \hat{h}_1^+\geq g_1,\ g_1\geq \hat{h}_1^-$ etc. In case $z\in{\mathcal{B}}$ it is additionally compact, as all of $S_h(y)$’s forming the union are compact.
From $$C({\mathcal{H}}) = \bigcup_{\substack{y\in L_n^N\\h\in{\mathcal{H}}}}{S_h(y)}$$ It follows immediately that component $C({\mathcal{H}})$ is a *finite union of polytopes*.
Now, to establish the contractibility of $C({\mathcal{H}})$ we show that it can be continuously contracted to $C\left((z_{i_1}^-, z_{i_2}^+,\dots,z_{i_n}^-)\right)$ (observe that $(z_{i_1}^-, z_{i_2}^+,\dots,z_{i_n}^-)\in{{\mathcal{E}}}_+$ ), which establishes the contractibility.
\[thm:mainNdelta\] Let $y\in L_n^N$. Given ${\mathfrak{N}}= {\mathfrak{N}}_\delta$ a set satisfying . For all $1\leq i\leq I({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))$, preimage component $C(\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))^i)$ is contractible.
Let us fix $1\leq i\leq I({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))$. We define a continuous homotopy $H\colon[0,1]\times{\mathbb{R}}^N\to{\mathbb{R}}^N$, contracting $C(\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))^i)$ to $C\left((z_{i_1}^-, z_{i_2}^+,\dots,z_{i_n}^-)\right)$, whose contractibility is proved using the theory developed in the prequel.
Let us pick any ${\mathsf{diag}}\in{\mathfrak{N}}$, and a vector $z$, such that $\operatorname{cv}(z)\in \operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))^i$, such that ${\mathsf{Dgm}}(z) = {\mathsf{diag}}$. Precisely, we will define $H$, such that $$H(0,z) = z,\text{ and }H(1,z) = w\text{, for some }w\in C((z_{i_1}^-, z_{i_2}^+,\dots,z_{i_n}^-)),$$ homotopy $H$ is constructed as a composition of two ’partial’ homotopies $$\label{defH}
H(\lambda, z) = \left\{\begin{matrix}H_d(2\lambda, z)&\text{if }\lambda\in[0,\frac12],\\H_l(2(\lambda-\frac12),H_d(1, z))&\text{ if }\lambda\in[\frac12,1].\end{matrix}\right.$$ The role of homotopy $H_d$ is to ’flatten out’ the part of $z$ corresponding to the diagram in $\mathcal{D}_\delta$, contained in the neighborhood of the diagonal. Whereas, $H_l$ transforms the local extrema having heights determined by a diagram in $\mathcal{R}$, and their neighborhoods to plateaus having heights determined by $(z_{i_1}^-, z_{i_2}^+,\dots,z_{i_n}^-)$.
Let us first define homotopy $H_d$. Let us pick any ${\mathsf{diag}}\in{\mathfrak{N}}$, and a vector $z\in C(\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))^i)$, such that ${\mathsf{Dgm}}(z) = {\mathsf{diag}}$. $H_d$ works by removing all components of the sublevel set filtration of $(\mathcal{S}, z)$, corresponding to the persistence points found in the defined neighbourhood of the diagonal ${\mathfrak{D}}_\delta$. Consider $r_1=(r_1^b, r_1^d)$ the first point on $\Pi_{{\mathfrak{D}}_\delta}{\mathsf{diag}}$. Let $\{z_{s(r_1)},\dots, z_{e(r_1)}\}$ be the sequence of $z$ components such that $r_1^b\leq z_j\leq r_1^d$ for all $s(r_1)\leq j\leq e(r_1)$, and the corresponding local minimum and local maximum is within this range of indices. The existence of such a unique sequence of indices follows from the second assumption on ${\mathfrak{D}}_\delta$, i.e. for all $j\neq i$ it holds that $[r_j^b,r_j^d]\nsubseteq [r_i^b,r_i^d]$. Also we define $H_d$ locally for this range of indices $$H_d(\lambda, z)_j = (1 - \lambda)z_j + \lambda r_1^b.$$ We define the homotopy $H_d$ analogously within all subsequences of indices determined by all the other intervals $[r_b^2, r_d^2], [r_b^3, r_d^3], \dots$. For all indices $k$ that are not considered above we use the constant homotopy $H_d([0,1], z)_k = z_k$. As $H_d$ creates plateaus in place of all local extrema defining the (noisy) points on ${\mathfrak{D}}_\delta$, hence ${\mathsf{Dgm}}(H_d(1,z))$ has all its points within ${\mathfrak{R}}$, and all the remaining local extrema of $H_d(1,z)$ are defined by the points in ${\mathfrak{R}}$.
Let us now define homotopy $H_l$. Let us pick any ${\mathsf{diag}}\in{\mathfrak{N}}$, such that $\Pi_{{\mathfrak{D}}_\delta}$ consists of only diagonal points (i.e. the persistence noise is flatten out), and a vector $z\in C(\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))^i)$, such that ${\mathsf{Dgm}}(z) = {\mathsf{diag}}$. From the assumption the first birth value of $\Pi_{{\mathfrak{R}}}{{\mathsf{diag}}}$ is in interval $R_1 = [b_1^-, b_1^+)$, hence, the first component appearing in the sublevel set filtration of complex $(\mathcal{S},z)$ is born in the range $[b_1^-, b_1^+)$. We pick $\{ s(b_1),\dots, e(b_1)\}$, the set of indices, such that $$z_{j} \leq b_1^+,\text{ for all } s(b_1)\leq j\leq e(b_1).$$ For such indices $j$, we define $$H_l(\lambda,z)_j = (1-\lambda)z_j + \lambda b_1^-,$$ i.e. the homotopy $H_l$ creates a plateau of height $b_1^-$ in $(\mathcal{S},H_d(1,z))$.
Proceeding further, the birth and death values of the second component of the sublevel set filtration of $(\mathcal{S},z)$ are within $R_2 = [b_2^-,b_2^+)\times[d_2^-,d_2^+)$. Observe that due to assumptions it holds that $[b_2^-,b_2^+)$ is disjoint from $[d_2^-,d_2^+)$. We pick a sequence of $z$ components $\{ z_{s(b_2)},\dots, z_{e(b_2)}\}$, such that $b_2^-\leq z_j \leq b_2^+$ for all $s(b_2)\leq j\leq e(b_2)$, and the local minimum is contained within this sequence (there may be several disjoint sequences within the specified range, but only one of them contains the local extremum). For such indices, we define the homotopy $$H_l(\lambda,z)_j = (1-\lambda)z_j + \lambda b_2^-.$$ Now, we pick a sequence of $z$ components $\{ z_{s(d_2)},\dots, z_{e(d_2)}\}$, such that $d_2^-\leq z_j \leq d_2^+$ for all $s(d_2)\leq j \leq e(d_2)$, and the indices of the local maximum are contained within this sequence (there may be several disjoint sequences within the specified range, but only one of them contains the local extremum). For such indices, we define the homotopy $$H_l(\lambda,z)_j = (1-\lambda)z_j + \lambda d_2^+.$$ We define the homotopy $H_l$ analogously for each of the subsequences of indices containing indices of one of the other local extrema.
For indices $k$ that were not considered above we define $H_l$ as a constant homotopy, i.e. $$H_l([0,1],z)_k = z_k.$$
As $H_l$ creates plateaus having heights $(z_{i_1}^-, z_{i_2}^+,\dots,z_{i_n}^-)$ in place of all local extrema, it holds that $H(1,z)\in C((z_{i_1}^-, z_{i_2}^+,\dots,z_{i_n}^-))$ for $H$ defined in .
Finally, to show that the constructed homotopy $H$ is continuous with respect to the input. Let $\varepsilon>0$ be small. Let $\widetilde{z}_1, \widetilde{z}_2\in C(\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}))^i)$, such that $|\widetilde{x}_1 - \widetilde{x}_2| < \varepsilon$. Observe that the homotopy $H$ is defined such that applied to $\widetilde{z}_1$ and $\widetilde{z}_2$, some of the $H(\lambda, \widetilde{z}_1)$ and $H(\lambda, \widetilde{z}_2)$ components approach the same value, or stay the same. Hence, $H$ cannot increase the distance between $H(\lambda, \widetilde{z}_1)$ and $H(\lambda, \widetilde{z}_2)$ for any $\lambda\in[0,1]$, therefore, it holds that $|H(\lambda, \widetilde{z}_1) - H(\lambda, \widetilde{z}_2)|\leq \varepsilon$ for all $\lambda\in[0,1]$.
On the existence of fixed points for flows {#secapplication}
==========================================
In this section we present an application of our preimage contractibility result presented in Section \[seccontr\] to prove the existence of a fixed point of a finite dimensional dynamical system that is being observed in the space of persistence diagrams.
Let $N\geq 3$. Consider a system of differential equations $\dot{z} = F(z)$ defined on ${\mathbb{R}}^{N}$ that generates a flow on a *compact global attractor* ${{\mathcal{A}}}\subset{\mathbb{R}}^N$ $$\Phi\colon {\mathbb{R}}\times {{\mathcal{A}}}\to {{\mathcal{A}}}.$$ Observe that at each time point $\Phi(t,z) = (\Phi_1(t,z),\ldots,\Phi_{N-1}(t,z),\Phi_{N}(t,z)) \in {{\mathcal{A}}}$, and hence we can consider ${\mathsf{Dgm}}(\Phi(t,z))\in {\mathsf{Per}}$. Thus $\Phi$ induces dynamics on ${\mathsf{Per}}$ that we define by $$\label{eqobsdyn}
\begin{aligned}
\Phi_{{\mathsf{Per}}}\colon{\mathbb{R}}\times{{\mathcal{A}}}&\to {\mathsf{Per}}\\
(t,z) &\mapsto \Phi_{{\mathsf{Per}}}(t,z) := {\mathsf{Dgm}}\left(\Phi(t,z)\right),
\end{aligned}$$ but hasten to point out that $\Phi_{{\mathsf{Per}}}$ need not be a semiflow, let alone a flow.
Let us define the global bounds $\bar{a}, \bar{b}$ as $$\bar{a} = \sup_{z\in{{\mathcal{A}}}}{|z_1|},\qquad \bar{b} = \sup_{z\in{{\mathcal{A}}}}{|z_N|}.$$
Our main result says that if a region ${\mathfrak{N}}_\delta \subset {\mathsf{Per}}$ is positively invariant under $\Phi_{{\mathsf{Per}}}$, then the underlying flow $\Phi$ has an equilibrium point $\bar{z}$ such that ${\mathsf{Dgm}}(\bar{z})\in {\mathfrak{N}}_\delta$.
\[thmmaindyn\] Let $\Phi\colon {\mathbb{R}}\times {{\mathcal{A}}}\to {{\mathcal{A}}}$ be a flow on a compact global attractor ${{\mathcal{A}}}\subset{\mathbb{R}}^N$, and let $\Phi_{\mathsf{Per}}$ be the induced dynamics on ${\mathsf{Per}}$. Let ${\mathcal{B}}= \left\{z\in{\mathbb{R}}^N\colon z_1\in[-\bar{a},\bar{a}],\ z_N\in[-\bar{b},\bar{b}]\right\}$. Let ${\mathfrak{N}}_\delta\subset {\mathsf{Per}}$ .
Assume that for any $z\in{\mathcal{B}}$ satisfying ${\mathsf{Dgm}}( z )\in{\mathfrak{N}}_\delta$ it holds that $$\Phi_{{\mathsf{Per}}}\left([0,\infty), z\right) \subset{\mathfrak{N}}_\delta.$$
Then, there exists $\overline{z} \in{\mathcal{B}}$, such that $${\mathsf{Dgm}}(\overline{z}) \in {\mathfrak{N}}_\delta,$$ and $\overline{z}$ is a fixed point of $\Phi$.
Let us denote $$C_j := C(\operatorname{cv}({\mathsf{Dgm}}^{-1}({\mathfrak{N}}_\delta))^j).$$ From our assumption that $\Phi_{{\mathsf{Per}}}\left([0,\infty), z\right)\subset{\mathfrak{N}}_\delta$ it follows that for any vector $z$, such that $$z \in C_j$$ for some $1\leq j\leq I({\mathsf{Dgm}}^{-1}({\mathfrak{N}}_\delta))$, it holds that $$\Phi_{{\mathsf{Per}}}\left([0,\infty), z\right)\subset C_j,\text{ for all }t\in[0,\infty).$$
Without loss of generality let us fix $j$. Hence, the component $C_j$ gets mapped into itself by the semi-flow $\Phi$, i.e. $$\Phi([0,\infty),C_j) \subset C_j.$$
By Theorem \[thm:mainNdelta\] $C_j\subset{\mathcal{B}}$ is a compact simplicial set, see Remark \[remcompact\].
Now, for any $t>0$ the time-$t$ map $ \Phi_t(z) := \Phi(t,z)$ $$\Phi_t(C_j) \subset C_j,$$ and because $C_j$ is a compact contractible polytope only $0$-th homology of self map $\Phi_t$ is nontrivial, and from the Lefshetz fixed point theorem [@lefshetz1; @lefshetz2] $\Phi_t$ for any $t>0$ admits a fixed point in $C_j$.
Finally passing to the limit $t\to 0$ a fixed point of the semi-flow $\Phi$ is obtained.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Theory of scattering of a quantum-mechanical particle on a cosmic string is developed. S-matrix and scattering amplitude are determined as functions of the flux and the tension of the string. We reveal that, in the case of the nonvanishing tension, the high-frequency limit of the differential scattering cross section does not coincide with the differential cross section for scattering of a classical pointlike particle on a string.'
---
=0.3cm
0.9cm
**On effects of non-Euclidean geometry in quantum theory**
0.3cm
Yu.A. Sitenko $ ^{a\dagger}$, N.D. Vlasii $^{ab\star}$
0.1cm
$^{a}$ *Bogolyubov Institute for Theoretical Physics, National Academy of Sciences of Ukraine,*
*14 b Metrologichna str., Kyiv 03143 , Ukraine*
$^{b}$ *Department of Physics, Taras Shevchenko National University of Kyiv,*
*6 Academician Glushkov ave., Kyiv 03680, Ukraine*
0.2cm
0.1cm
**Keywords**: cosmic string, Bohm-Aharonov effect
0.1cm
(based on talks given at: the International Workshop “Frontiers of Particle Astrophysics”, June 21-24, 2004, Kyiv, Ukraine; the George Gamow Memorial International Conference “Astrophysics and Cosmology after Gamow - Theory and Observations”, August 8-14, 2004, Odessa, Ukraine; the IV International Conference “Non-Euclidean Geometry in Modern Physics and Mathematics”, September 7-11, 2004, Nizhny Novgorod, Russia)
$
\begin{array}{ll}
^{\dagger}\mbox{{\it e-mail address:}} &
\mbox{[email protected]} \\
^{\star}\mbox{{\it e-mail address:}} & \mbox{[email protected]}
\end{array}
$
Introduction {#s1}
============
Usually, the effects of non-Euclidean geometry are identified with the effects which are due to the curvature of space. It can be immediately shown that this is not the case and there are spaces which are flat but non-Euclidean.
A simplest example is given by a twodimensional space (surface) which is obtained from a plane by cutting a segment of a certain angular size and then sewing together the edges. The resulting surface is the conical surface which is flat but has a singular point corresponding to the apex of the cone. To be more precise, the intrinsic (Gauss) curvature of the conical surface is proportional to the twodimensional delta-function placed at the apex; the coefficient of proportionality is the deficit angle. Usual cones correspond to positive values of the deficit angle, i.e. to the situation when a segment is deleted from the plane. But one can imagine a situation when a segment is added to the plane; then the deficit angle is negative, and the resulting flat surface can be denoted as a saddle-like cone. The deleted segment is bounded by the value of $2\pi$, whereas the added segment is unbounded. Thus, deficit angles for possible conical surfaces range from $-\infty$ to $2\pi$.
It is evident that an apex of the conical surface with the positive deficit angle can play a role of the convex lens, whereas an apex of the conical surface with the negative deficit angle can play a role of the concave lens. Really, two parallel trajectories coming from infinity towards the apex from different sides of it, after bypassing it, converge (and intersect) in the case of the positive deficit angle, and diverge in the case of the negative deficit angle. This demonstrates the non-Euclidean nature of conical surfaces. It is interesting that this item provides a basis for understanding such physical objects as cosmic strings. In the present paper we shall discuss peculiarities of quantum theory and its quasiclassical limit, which are due to non-Euclidean geometry of locally flat space-times.
Space-time in the presence of a cosmic string {#concl}
=============================================
Cosmic strings are topological defects which are formed as a result of phase transitions with spontaneous breakdown of symmetries at early stages of evolution of the universe, see, e.g., reviews in Refs.[@Hi; @Vi]. In general, a cosmic string is characterized by two quantities: flux $$\Phi=\int\limits_{\rm core}d^2x\sqrt{g}B^3,$$ and tension $$\mu=\frac{1}{16\pi G}\int\limits_{\rm core}d^2x\sqrt{g}R;$$ here the integration is over the transverse section of the core of the string, $B^3$ is the field strength which is directed along the string axis, $R$ is the scalar curvature, $G$ is the gravitational constant, and units $\hbar=c=1$ are used. The space-time metric outside the string core is $$ds^2=dt^2-(1-4G\mu)^{-1}d{\stackrel{\sim}{r}}^2-
(1-4G\mu)\stackrel{\sim}{r}^2d\varphi^2-dz^2=
dt^2-dr^2-r^2d\stackrel{\sim}{\varphi}^2-dz^2,$$ where $$\stackrel{\sim}{r}=r\sqrt{1-4G\mu}, \quad 0\leq\varphi<2\pi,\quad
0\leq\stackrel{\sim}{\varphi}<2\pi (1-4G\mu).$$ A surface which is transverse to the axis of the string is isometric to the surface of a cone with a deficit angle equal to $8\pi G\mu$. Such space-times were known a long time ago (M. Fierz, unpublished, see footnote in Ref.[@We]) and were studied in detail by Marder [@Ma]. In the present context, as cosmological objects and under the name of cosmic strings, they were introduced in seminal works of Kibble [@Ki] and Vilenkin [@Vil]. A cosmic string resulting from a phase transition at the scale of the grand unification of all interactions is characterized by the values of tension $$\mu\sim(10^{-7}\div10^{-6})G^{-1}.$$ The nonvanishing of the string tension leads to various cosmological consequences and, among them, to a very distinctive gravitational lensing effect. A possible observation of such an effect has been reported recently [@Sa], and this has revived an interest towards cosmic strings.
The flux parameter (1) is nonvanishing for the so-called gauge cosmic strings, i.e. strings corresponding to spontaneous breakdown of local symmetries. If tension vanishes ($\mu=0$), then a gauge cosmic string becomes a magnetic string, i.e. a tube of the magnetic flux lines in Euclidean space. If the tube is impenetrable for quantum-mechanical charged particles, then scattering of the latter on the magnetic string depends on flux $\Phi$ periodically with period $2\pi e^{-1}$ ($e$ is the coupling constant – charge of the particle). This is known as the Bohm-Aharonov effect [@Aha], which has no analogue in classical physics, since the classical motion of charged particles cannot be affected by the magnetic flux from the impenetrable for the particles region. The natural question is, how the nonvanishing string tension ($\mu\neq0$) influences scattering of quantum-mechanical particles on the string. Thus, the subject of cosmic strings, in addition to tantalizing phenomenological applications, acquires a certain conceptual importance.
Quantum scattering on a cosmic string {#concl}
=====================================
Due to nonvanishing flux $\Phi$ and tension $\mu$, the quantum scattering of a test particle by a cosmic string is a highly nontrivial problem. It is impossible to choose a plane wave as the incident wave, because of the long-range nature of the interaction inherent in this problem. A general approach to quantum scattering in the case of long-range interactions was elaborated by Hormander [@Ho]. This approach covers the cases of scattering on a Coulomb center and on a magnetic string ($\mu=0$), but is not applicable to the case of scattering on a cosmic string ($\mu\neq
0$). Therefore the last case needs a special consideration and a thorough substantiation.
When the effects of the core structure of a cosmic string are neglected and the transverse size of the core is negligible, the field strength and the scalar curvature are presented by twodimensional delta-functions. Scattering of a quantum-mechanical particle on an idealized (without structure) cosmic string was considered in Refs.[@Hoo; @De; @So; @Si2]. A general theory of quantum-mechanical scattering on a cosmic string, permitting to take into account the effects of the core structure, was elaborated in Ref.[@Si5]. According to this theory, the $S$-matrix in the momentum representation is $$\begin{aligned}
S(k,\varphi;\,\,k',\varphi')=\frac{1}{2}\frac{\delta(k-k')}{\sqrt{kk'}}\left\{\Delta
(\varphi-\varphi'+\frac{4G\mu\pi}{1-4G\mu})\exp\left[-\frac{ie\Phi}{2(1-4G\mu)}
\right]+\right. \nonumber \\
+\left.\Delta\left(\varphi-\varphi'-\frac{4G\mu\pi}{1-4G\mu}\right)
\exp\left[\frac{ie\Phi}{2(1-4G\mu)}\right]\right\}+\delta(k-k')
\sqrt{\frac{i}{2\pi k}}f(k,\,\,\varphi-\varphi'),\end{aligned}$$ where the initial $({\bf k})$ and final $({\bf k}')$ twodimensional momenta of the particle are written in polar variables, $f(k,\,\,\varphi-\varphi')$ is the scattering amplitude, and $\Delta(\varphi)=
\frac{1}{2\pi}\sum\limits_{n=-\infty}^{\infty}e^{in\varphi}$ is the angular part of the twodimensional delta-function. Note that in the case of short-range interaction one has $2\Delta(\varphi-\varphi')$ instead of the figure brackets in Eq.(6). Thus, one can see that, due to the long-range nature of interaction, even the conventional relation between $S$-matrix and scattering amplitude is changed, involving now a distorted unity matrix (first term in Eq.(6)) instead of the usual one, $\delta(k-k')\Delta(\varphi-\varphi')(kk')^{-1/2}$.
In view of the comparison with the Bohm-Aharonov effect [@Aha], we shall be interested in the situation when the string core is impenetrable for the particle. The scattering amplitude in this case takes form: $$f(k,\,\,\varphi)=f_0(k,\,\,\varphi)-\sqrt{\frac{2}{\pi
ik}}\sum\limits_{n=-\infty}^{\infty}\exp[in\varphi-i(\alpha_n-|n|)\pi]
\frac{J_{\alpha_n}(kr_c)}{H_{\alpha_n}^{(1)}(kr_c)},$$ where $r_c$ is the radius of the string core, $J_\nu(u)$ and $H_\nu^{(1)}(u)$ are the Bessel and the first-kind Hankel functions of order $\nu$, $$\alpha_n=\left|n-\frac{e\Phi}{2\pi}\right|(1-4G\mu)^{-1},$$ and $$\begin{aligned}
f_0(k,\varphi)&=&\frac{1}{\sqrt{2\pi i k}} \left\{
\frac{\exp\left[i[[\frac{e\Phi}{2\pi}]](\varphi+
\frac{4G\mu\pi}{1-4G_\mu})-\frac{ie\Phi}{2(1-4G\mu)}\right]}{1-\exp
\left[-i\left(\varphi+\frac{4G\mu\pi}{1-4G\mu}\right)\right]}-\right.
\nonumber \\
&-&\left.\frac{\exp\left[i[[\frac{e\Phi}{2\pi}]]\left(\varphi-
\frac{4G\mu\pi}{1-4G\mu}\right)+\frac{ie\Phi}{2(1-4G\mu)}\right]}{1
-\exp\left[-i\left(\varphi-\frac{4G\mu\pi}{1-4G\mu}\right)\right]}
\right\}\end{aligned}$$ is the amplitude of scattering on an idealized (without structure) cosmic string, $[[u]]$ is the integer part of $u$. Sum over $n$ in Eq.(7) describes the core structure effects. In the low-frequency limit $(k\rightarrow 0)$ these effects die out, and the differential cross section (i. e. the square of the absolute value of the amplitude) takes form $$\begin{aligned}
\frac{d\sigma}{d\varphi}&=&\frac{1}{4\pi k} \left\{
\frac{1}{2\sin^2\left[\frac{1}{2}\left(\varphi+\frac{4G\mu\pi}{1-
4G\mu}\right)\right]}+\frac{1}{2\sin^2\left[\frac{1}{2}\left(\varphi-
\frac{4G\mu\pi}{1-4G\mu}\right)\right]}-\right. \nonumber \\
&-&\left.
\frac{\cos\left[\frac{e\Phi}{1-4G\mu}-\left(2[[\frac{e\Phi}{2\pi}]]+1\right)
\frac{4G\mu\pi}{1-4G\mu}\right]}{\sin\left[\frac{1}{2}\left(\varphi+
\frac{4G\mu\pi}{1-4G\mu}\right)\right]\sin\left[\frac{1}{2}\left(\varphi-
\frac{4G\mu\pi}{1-4G\mu}\right)\right]} \right\}.\end{aligned}$$
Differential cross section in the limit of high frequency of\
scattered particle {#concl}
=============================================================
In the high-frequency limit $(k\rightarrow \infty)$ the first term in Eq.(7) dies out, and the differential cross section takes form $$\begin{aligned}
\frac{d\sigma}{d\varphi}&=&\frac{1}{2}r_c(1-4G\mu)^2\left|\sum\limits_{l}\sqrt{\cos[\frac{1}{2}
(1-4G\mu)(\varphi-\pi+2l\pi)]}\times \right.\nonumber \\
&\times& \left.\exp\{ie\Phi
l-2ikr_c\cos[\frac{1}{2}(1-4G\mu)(\varphi-\pi+2l\pi)]\}\right|^2,\end{aligned}$$ where the summation is over integer $l$ satisfying condition $$-\frac{\varphi}{2\pi}-\frac{2G\mu}{1-4G\mu}<l<-\frac{\varphi}{2\pi}+1+\frac{2G\mu}{1-4G\mu}.$$ Note that results (10) and (11) are periodic in the value of flux $\Phi$ with period equal to $2\pi e^{-1}$. This feature is common with the scattering on a purely magnetic string ($\mu=0$). The difference is that the Bohm-Aharonov differential cross section in the low frequency limit ($k\rightarrow 0$) diverges in the forward direction, $\varphi=0$, while Eq.(10) diverges in two symmetric directions, $\varphi=\pm4G\mu(1-4G\mu)^{-1}$. The difference becomes much more crucial in the high-frequency limit $(k\rightarrow \infty)$. In the $\mu=0$ case one gets $$\frac{d\sigma}{d\varphi}=\frac{1}{2}r_c\sin\frac{\varphi}{2},$$ which is the cross section for scattering of a classical pointlike particle by an impenetrable cylindrical shell of radius $r_c$; evidently, the dependence on fractional part of $e\Phi(2\pi)^{-1}$ disappears in this limit. In the $\mu\neq 0$ case the dependence survives, see Eq.(11). In particular, if $0<\mu<(8G)^{-1}$, which is most interesting from the phenomenological point of view, then the cross section at $k\rightarrow\infty$ takes the following form in the region of the cosmic string shadow, $-\frac{4G\mu\pi}{1-4G\mu}<\varphi<\frac{4G\mu\pi}{1-4G\mu}$: $$\begin{aligned}
\frac{d\sigma}{d\varphi}&=&r_c(1-4G\mu)^2\left(\cos[\frac{1}{2}(1-4G\mu)\varphi]\sin(2G\mu\pi)
+\right.
\nonumber \\
&+&\left.\sqrt{\sin^2(2G\mu\pi)-\sin^2[\frac{1}{2}(1-4G\mu)\varphi]}\right.
\left.\cos\left\{e\Phi+4kr_c\sin[\frac{1}{2}(1-4G\mu)\varphi]\cos(2G\mu\pi)\right\}\right).\end{aligned}$$ Integrating Eq.(14) over the region of the shadow and the appropriate expression (which is independent of $\Phi$) over the region out of the shadow, we obtain the total cross section in the $k\rightarrow\infty$ limit: $$\sigma_{\rm tot}=2r_c(1-4G\mu).$$ The high-frequency limit is usually identified with the quasiclassical limit. Although this identification is valid for the total cross section, it is found to be invalid for the differential cross section, see Eqs.(11) and (14) revealing the periodic dependence on the flux, which is a purely quantum effect.
These results are generalized to the case of scattering of a particle with spin.
0.2cm
Acknowledgements
=================
This work was supported by the State Foundation for Basic Research of Ukraine (project 2.7/00152).
[99]{} A.Vilenkin, E.P.S.Shellard, [*Cosmic Strings and Other Topological Defects*]{}, Cambridge Univ. Press, Cambridge (1994). M.B.Hindmarsh, T.W.B.Kibble, Rep. Progr. Phys. [**58**]{}, 477 (1995). J.Weber, J.A.Wheeler, Rev. Mod. Phys. [**29**]{}, 509 (1957). L.Marder, Proc. Roy. Soc. London [**A252**]{}, 45 (1959). T.W.B.Kibble, J. Phys. [**A9**]{}, 1387 (1976); Phys. Rep. [**67**]{}, 183 (1980). A.Vilenkin, Phys. Rev. [**D23**]{}, 852 (1981); [**D24**]{}, 2082 (1981). M.Sazhin et al., Mon. Not. Roy. Astron. Soc. [**343**]{}, 353 (2003); astro-ph/0406516. Y.Aharonov, D.Bohm, Phys. Rev. [**115**]{}, 485 (1959). L.Hormander, [*Analysis of Linear Partial Differential Operators IV*]{}, Springer–Verlag, Berlin (1985). G.’t Hooft, Commun. Math. Phys. [**117**]{}, 685 (1988). S.Deser, R.Jackiw, Commun. Math. Phys. [**118**]{}, 495 (1988). P. de Sousa Gerbert, R.Jackiw, Commun. Math. Phys. [**124**]{}, 229 (1989). Yu.A.Sitenko, Nucl. Phys. [**B372**]{}, 622 (1992). Yu.A.Sitenko, A.V.Mishchenko, JETP [**81**]{}, 831 (1995).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We provide a rigorous solution to the problem of constructing a structural evolution for a network of coupled identical dynamical units that switches between specified topologies without constraints on their structure. The evolution of the structure is determined indirectly, from a carefully built transformation of the eigenvector matrices of the coupling Laplacians, which are guaranteed to change smoothly in time. In turn, this allows to extend the Master Stability Function formalism, which can be used to assess the stability of a synchronized state. This approach is independent from the particular topologies that the network visits, and is not restructed to commuting structures. Also, it does not depend on the time scale of the evolution, which can be faster than, comparable to, or even secular with respect to the the dynamics of the units.'
author:
- 'Charo I.'
- Miguel
- Regino
- Stefano
title: Synchronization in dynamical networks with unconstrained structure switching
---
Networked structures, in which sets of distributed dynamical systems interact over a wiring of links with non-trivial topology, are a key tool to investigate the emergence of collective organization in many systems of interest [@Alb02; @Boc06; @Boc14]. The analysis of synchronized states is particularly relevant, as they play a crucial role in many natural systems, such as brain networks [@Var01], or ecological communities [@Ber99]. In the past decade, the emergence of synchronized states has been extensively reported and studied [@Boc02], with a notable emphasis on the effect of complex static topologies on synchronization properties [@Pec98; @Lag00; @Bar02; @Nis03; @Bel04; @Cha05; @Hwa05; @Mot05; @Yao09; @Li13; @Mas13]. Nonetheless, static network models do not adequately describe the processes that arise because of mutations in some biological systems, such as infectious bacterial population, which are known to have adaptive mutation rates that can become very highly elevated [@Den06; @Csa07]. These require, instead, the use of time-dependent topologies whose evolution occurs over time scales that are commensurate with those of the node dynamics [@And88; @Han04]. A powerful tool to assess the stability of the synchronous solution in networks of $N$ identical nodes with diffusive coupling, is the so-called Master Stability Function (MSF) [@Pec98]. If the evolution of such networks is along structures whose Laplacians commute at all times, synchronization can be significantly enhanced. In fact, studies have shown that it can be achieved even when the connection structure at each time would not allow a synchronized state, in the static case [@Boc06_2; @Amr06]. The far more realistic situation of networks whose coupling matrices do not necessarily commute has also generated significant interest. This has led to the development of several results about synchronizability in systems with specific properties, including the study of the synchronous state in the case of moving neighbourhood networks [@Por06], the rigorous derivation of sufficient conditions for synchronization in fast switching networks [@Sti06], and the analysis of the system dynamics in the so-called blinking limit [@Has13; @Has13_2]. In this Article, we propose a framework that allows to assess the stability of a synchronous solution by means of an MSF when the evolution of the topology is fully general and unconstrained, and without assuming any hypothesis on the time scales of the topological evolution.
To this purpose, let us consider a network of $N$ identical systems, evolving according to $$\label{eq1}
\dot{\mathbf{x}}_i = \mathbf{f}\left(\mathbf{x}_i\right) - \sigma \sum_{j=1}^N L_{ij}\left(t\right) \mathbf{h}\left(\mathbf{x}_j\right)\:.$$ Here dot denotes time derivative, $\mathbf{x}_i$ is an $m$-dimensional row vector describing the state of the $i^\mathrm{th}$ node, $\sigma$ is the interaction strength, and $\mathbf f$ and $\mathbf h$ are two vectorial functions describing the local dynamics of the nodes and the output from a node to another, respectively. Also, $L\left(t\right)=S\left(t\right)-W\left(t\right)$ is the Laplacian of the network describing the time evolution of the connections. In this expression, $W$ is the weighted adjacency matrix, and $S$ is the diagonal strength matrix of the network: $S_{ij}=\delta_{i,j}\sum_{k=1}^N W_{ik}$. To fix the ideas, assume that the network has an initial structure, with Laplacian $L_0$, that is constant from time $t=0$ to $t=t_0$. Also assume that, at time $t_1>t_0$, the network is found in a configuration with Laplacian $L_1≠L_0$. Note that in the following we consider Laplacians with Hermitian structure. Also, as $L\left(t\right)$ is a Laplacian matrix, the sum of each row vanishes, its diagonal elements are strictly positive and its off-diagonal elements are non-positive. Now, let $\mathbf{x_s}$ be the state vector indicating the synchronized solution, and define the $mN$-dimensional column vector $\delta\mathbf X=\left(\delta\mathbf x_1,\dotsc,\delta\mathbf x_N\right)^\mathrm{T}$, representing the global deviation from the synchronized state. From Eq. \[eq1\], to linear order in $\delta\mathbf X$, one has $$\label{eq:var}
\delta\dot{\mathbf X} = \left(\mathds 1\otimes\mathrm J
\mathbf f\left(\mathbf{x_s}\right) - \sigma L\left(t\right)\otimes\mathrm J \mathbf h
\left(\mathbf{x_s}\right)\right)\delta\mathbf X\:,$$ where $\mathds 1$ is the $N$-dimensional identity matrix, $\otimes$ denotes the direct product, and $\mathrm J$ is the Jacobian operator. The vector $\delta\mathbf X$ can be written at each time as a sum of direct products of the eigenvectors $\mathbf v_i$ of $L\left(t\right)$ and a time-dependent set of $N$ $m$-dimensional row-vectors $\boldsymbol \eta_i\left(t\right)$: $$\delta\mathbf X=\sum_{i=1}^N\mathbf v_i\left(t\right)\otimes\boldsymbol \eta_i\left(t\right)\:.$$ Multiplying $\mathbf v_j^{\mathrm T}$ from the left to both sides of Eq. \[eq:var\], one gets $$\label{variation}
{\frac{\mathrm d}{\mathrm {dt}}}\boldsymbol \eta_j=\mathbf K_j\boldsymbol \eta_j\ \ \boxed{-\sum_{i=1}^N\mathbf v_j^{\mathrm T}
\left(t\right)\cdot{\frac{\mathrm d}{\mathrm {dt}}}\mathbf v_i\left(t\right)\boldsymbol\eta_i}\:,$$ where, for the sake of brevity, we defined $\mathbf K_j\equiv\left(\mathrm J \mathbf f\left(\mathbf{x_s}\right) - \nu_j\mathrm J \mathbf h(\mathbf{x_s})\right)$ and $\nu_j\equiv\sigma\lambda_j\left(t\right)$, in which $\lambda_j\left(t\right)$ is the $j^\mathrm{th}$ eigenvalue of $L\left(t\right)$. To assess the stability of this dynamical system, one can compute the Master Stability Function, which represents the dependence on $\nu$ of the largest Lyapunov exponent $\Lambda_{\max}$ associated to the equations \[variation\]. Then, the stability criterion for a given $\nu$ is that the time averages of $\Lambda_{\max}$ in the direction of all eigenvectors are negative. This allows to study systems with highly non-trivial behaviour. As an example, one can consider an evolving system where each “frozen” connection topology has at least one direction in which $\Lambda_{\max}$ is positive, but the synchronization manifold is transversely stable [@Boc06_2]. Similarly, one can detect instabilities introduced by the evolving nature of the Laplacian in systems where the synchronization manifold in each frozen configuration is attracting [@Jos08]. This is particularly useful, as it is well-known that non-linear perturbations in a system can destroy the stability of the synchronized state [@Bar02_2].
Note that using the MSF is not the only possible method to assess the stability of a synchronized state. For instance, one could construct the Lyapunov function for the synchronization manifold. This would guarantee the stability since it is a necessary and sufficient condition, while in the general case the MSF is only a necessary one. However, while it is certainly possible to build the Lyapunov function in some specific cases [@Wu07], a general construction method is not known. Also, it should be noted that in the absence of fixed points or other attracting sets far from the synchronization manifold, the MSF provides a sufficient stability condition as well, thereby becoming a widely used approach [@Sor11].
To use the MSF method, one must first note that the boxed term in Eqs. \[variation\] explicitly depends on the time variation of the eigenvectors of $L$. If the Laplacians $L_0$ and $L_1$ commute, one can choose to study the problem in the common basis of eigenvectors. In this reference frame the eigenvectors do not change. Thus, the boxed term vanishes and Eqs. \[variation\] reduce to a set of variational equations. However, if we allow the network to switch to a different structure without imposing the extra requirement of commutativity, the eigenvector variation must be taken into account. Note that this forbids instantaneous jumps between the two structures, as a sudden change in the eigenvectors would cause their derivatives to be not defined. Therefore, the goal is constructing a smooth evolution process from $t=t_0$ to $t=t_1$ to allow the system to evolve between between the two topologies while keeping the eigenvector elements differentiable. To achieve this, we first consider the matrices $A$ and $B$, consisting of the eigenvectors of $L_0$ and $L_1$, respectively, and describe how to transform one into the other via a proper rotation around a fixed axis. Then, we use this framework to find a transformation between $L_0$ and $L_1$. The rotation evolving $A$ into $B$ takes the form of a one-parameter transformation group $G_s$ such that $G_0A=A$ and $G_1A=B$. Note that, as we will use this mapping to build the Laplacian, we must also impose the extra requirement that the vector $\mathbf a$, corresponding to the null eigenvalue, is kept constant by $G_s$ for all $0\leqslant s\leqslant 1$.
In general, the transformation $O$ from $A$ to $B$ is a rotation, which can be found by solving the linear system of $N^2$ equations in $N^2$ variables $OA=B$. It is convenient to work in the basis defined by the matrix $A$. In this basis, $A\equiv\mathds{1}$. Without loss of generality, assume that the conserved vector $\mathbf a$ is the first vector of $A$. Then, the transformation matrix $O$ has the form $$\label{tmat} O=
\begin{pmatrix}
1 & 0 & 0 & \dotsm & 0\\
0 & O_{22} & O_{23} & \dotsm & O_{2N}\\
0 & O_{32} & O_{33} & \dotsm & O_{3N}\\
\vdots & \vdots & \vdots & \ddots & \vdots\\
0 & O_{N2} & O_{N3} & \dotsm & O_{NN}
\end{pmatrix}\:.$$
As $O\in O\left(N\right)$, it is a proper rotation if its determinant is 1, or a composition of rotations and reflections if its determinant is $-1$. The determinant of $O$ equals the determinant of the minor $O'$ obtained by removing the first row and the first column from $O$. Thus, we only need to find a solution to the problem in $N-1$ dimensions. We will henceforth use primes when referring to objects in $N-1$ dimensions.
From the considerations above, it is $G'_0=\mathds{1}'$ $G'_1=O'$, and, of course, $O'\in O\left(N-1\right)$. Thus, the problem is equivalent to determining the possibility of finding a path between the identity and $O'$. If $\left|O'\right|=1$, then $O'\in SO\left(N-1\right)$. But $SO\left(N-1\right)$ is the connected identity component of the orthogonal group, and additionally, since it is a manifold, it is path-connected [@War71]. Thus, for every orthogonal $\left(N-1\right)\times\left(N-1\right)$ matrix in it, there is a continuum of orthogonal matrices of the same dimension connecting it to the identity. Each point along this path corresponds to an orthogonal matrix that can be embedded in $SO\left(N\right)$ by adding a $1$ in the top left corner. Since every such embedded matrix keeps the synchronization manifold vector invariant, a parametrization of the path provides a direct solution to the original problem.
If, instead, $\left|O'\right|=-1$, then $O'\in O\left(N-1\right)\setminus SO\left(N-1\right)$. While $O\left(N-1\right)\setminus SO\left(N-1\right)$ is also a connected topological space, the identity does not belong to it [@War71]. Thus, no path connects the identity to $O'$. However, in our case, the labeling of the vectors is irrelevant. In other words, provided that the vector $\mathbf a$ is kept constant, one can arbitrarily swap two vectors in the basis given by the matrix $A$, obtaining a new matrix $C$. Of course, this imposes a swap of the corresponding columns in the transformation matrix $O$ as well. But swapping two rows in a matrix changes the sign of its determinant. This means that the new matrix $O'\in SO\left(N-1\right)$, and a path connecting it to the identity can be found. The only consequence of the swap is a change of the order of the eigenvalues. As we are considering unlabelled networks, the problem can always be solved. To build an explicit solution, we factor the transformation $O'$ into rotations and reflections in mutually orthogonal subspaces.
Before describing the actual procedure, we recall a useful, twofold result. First, any orthogonal operator $X$ in a normed space over $\mathbb{R}$ induces a 1- or 2-dimensional invariant subspace. To find one such subspace, first define the unitary operator $U$ acting on $\mathbf a+\mathrm i\mathbf b$ as $U\left(\mathbf a+\mathrm i\mathbf b\right)=X\mathbf a+\mathrm iX\mathbf b$, where $\mathbf a$ and $\mathbf b$ belong to $\mathbb{R}^N$. Then, find a non-vanishing eigenvector $\mathbf x=\mathbf {x_R}+\mathrm i\mathbf{x_I}$ of $U$. The span of $\mathbf{x_R}$ and $\mathbf{x_I}$ defines the invariant subspace: applying $X$ to any linear combination of $\mathbf{x_R}$ and $\mathbf{x_I}$ produces a vector that is still a linear combination of $\mathbf{x_R}$ and $\mathbf{x_I}$. Also, if the corresponding eigenvalue is complex, $\mathbf{x_R}$ and $\mathbf{x_I}$ are orthogonal.
Using this, we can describe the following algorithmic procedure to build a transformation $O$ of $A$ into $B$:
1. Express the problem in the basis $A$, in which $O$ has the form of Eq. \[tmat\].
2. Consider the operator $O'$ obtained from $O$ by removing the first row and the first column, and let $d$ be its dimension.
3. Build the operator $U$ that acts on $\mathbf a+\mathrm i\mathbf b$ as $U\left(\mathbf a+\mathrm i\mathbf b\right)=O'\mathbf a+\mathrm iO'\mathbf b$, where $\mathbf a$ and $\mathbf b$ belong to $\mathbb{R}^d$.
4. Find an eigenvector $\mathbf x=\mathbf{x_R}+\mathrm i\mathbf{x_I}$ of $U$, with eigenvalue $\lambda$.
5. Normalize $\mathbf{x_R}$ and $\mathbf{x_I}$.
6. If $\lambda\in\mathbb{R}$ then
1. Pick the non-vanishing component between $\mathbf{x_R}$ and $\mathbf{x_I}$. If both are non-zero, choose one randomly. Without loss of generality, assume this is $\mathbf{x_R}$.
2. Create $d-1$ other orthonormal vectors, all orthogonal to $\mathbf{x_R}$, and arrange all these vectors so that $\mathbf{x_R}$ is the last of them. This set of vectors is an orthonormal basis $C$ of $\mathbb{R}^d$.
3. Change the basis of the $d$-dimensional sub-problem to $C$. In this basis, all the elements in the last row and in the last column of $O'$ will be 0, except the last one, which will be $\pm 1$.
4. If $d>1$, consider a new operator $O'$ obtained from the old $O'$ by removing the last row and the last column. Let $d$ be its dimension, and restart from step 3. Otherwise stop.
If instead $\lambda\notin\mathbb{R}$, then
1. Create $d-2$ other orthonormal vectors, all orthogonal to $\mathbf{x_R}$ and $\mathbf{x_I}$, and arrange all these vectors so that $\mathbf{x_R}$ and $\mathbf{x_I}$ are the first two of them. This set of vectors is an orthonormal basis $C$ of $\mathbb{R}^d$.
2. Change the basis of the $d$-dimensional sub-problem to $C$. In this basis, all the elements in the first two rows and in the first two columns of $O'$ will be 0, except the first two.
3. If $d>2$, consider a new operator $O'$ obtained from the old $O'$ by removing the first two rows and the first two columns. Let $d$ be its dimension, and restart from step 3. Otherwise stop.
At each iteration of the steps above, 1 or 2 dimensions are eliminated from the problem. All the subsequent changes of base leave the already determined elements of $O$ unchanged, because they act on orthogonal subspaces to those already eliminated. The procedure reconstructs $O$ piece by piece with a block-diagonal form, in which the blocks correspond to the action of the orthogonal operator on the invariant subspaces. If the subspace is 1-dimensional, then the block is a single $\pm1$ element. If instead the subspace is 2-dimensional, then its block is either a rotation or a reflection, i.e., it is either $
\begin{pmatrix}
\cos\alpha & -\sin\alpha\\
\sin\alpha & \cos\alpha\\
\end{pmatrix}$ or $
\begin{pmatrix}
\pm 1 & 0\\
0 & \mp 1\\
\end{pmatrix}$. Thus, the 1-dimensional invariant subspaces induced by the operator correspond either to leaving one direction untouched, or to reflecting the system about that direction. Conversely, the 2-dimensional invariant subspaces correspond to rotations in mutually orthogonal planes.
Once this form of $O$ is found, permute the basis vectors from the second onwards so they correspond, in order, first to all the actual rotation blocks, then to the $-1$ elements, and finally to the $+1$ elements. The new form of the transformation matrix, $O_N$, is simply $O_N=TOT^{-1}$, where $T$ is the required change-of-basis matrix.
Note that the determinant of $O_N$ could still be $-1$. However, as seen before, in this case, one can relabel two vectors of the original basis, inducing a swap of the corresponding columns of $O_N$. To perform this, first note that, if $\left|O_N\right|=-1$, then the number of $-1$ elements in $O_N$ must be odd. Then, there are three possible cases.
If $O'_N$ has at least one $+1$ element, swap the basis vectors corresponding to the first $-1$ and the first $+1$ elements in $O'$. Then, the first block after the “$\sin$–$\cos$” blocks is $\begin{pmatrix} 1&0\\0&-1\end{pmatrix}$. Swapping the labels of the the two corresponding vectors, makes the block in $O'$ become $
\begin{pmatrix}
0 & 1\\
-1 & 0
\end{pmatrix}
=\begin{pmatrix}
\cos\left(-\frac{\pi}{2}\right) & -\sin\left(-\frac{\pi}{2}\right)\\
\sin\left(-\frac{\pi}{2}\right) & \cos\left(-\frac{\pi}{2}\right)
\end{pmatrix}$.
If instead $O'_N$ has no $+1$ elements and only one $-1$ element, the basis vectors to swap are those corresponding to the first two vectors in the last $3\times 3$ block of $O'$, that becomes $$M = \begin{pmatrix}
-\sin\vartheta_k & \cos\vartheta_k & 0\\
\cos\vartheta_k & \sin\vartheta_k & 0\\
0 & 0 & -1
\end{pmatrix}.$$ Now, perform one more basis change, leaving all the basis vectors unchanged, but mapping the last three into the eigenvectors of $M$, which are $$V = \begin{psmallmatrix}
0 & \left(-\sec\vartheta_k-\tan\vartheta_k\right)\sin\left(\frac{\pi}{4}-\frac{\vartheta_k}{2}\right) & \left(\sec\vartheta_k-\tan\vartheta_k\right)\sin\left(\frac{\vartheta_k}{2}+\frac{\pi}{4}\right)\\
0 & 1 & 1\\
1 & 0 & 0
\end{psmallmatrix}\:.$$ The new form of $M$ becomes $$M' = V^\mathrm{T}MV = \begin{pmatrix}
-1 & 0 & 0\\
0 & -1 & 0\\
0 & 0 & 1
\end{pmatrix}.$$
Finally, if $O'_N$ has no $+1$ elements and at least three $-1$ elements, swap the basis vectors corresponding to the first two after the “$\sin$–$\cos$” blocks. Their $3\times 3$ block is now $$M = \begin{pmatrix}
0 & -1 & 0\\
-1 & 0 & 0\\
0 & 0 & -1
\end{pmatrix}.$$ Next, do a change of basis as described in the previous case. The eigenvector matrix of $M$ is $$V = \begin{pmatrix}
0 & \frac{1}{\sqrt{2}} & -\frac{1}{\sqrt{2}}\\
0 & \frac{1}{\sqrt{2}} & \frac{1}{\sqrt{2}}\\
1 & 0 & 0
\end{pmatrix}\:,$$ so, after the basis change, the new form of $M$ is once more $$M' = V^\mathrm{T}MV = \begin{pmatrix}
-1 & 0 & 0\\
0 & -1 & 0\\
0 & 0 & 1
\end{pmatrix}.$$
Regardless of the original value of $\left|O_N\right|$, take now every two subsequent $-1$ elements, if there are any, and change their diagonal block into $
\begin{pmatrix}
\cos\pi & -\sin\pi\\
\sin\pi & \cos\pi\\
\end{pmatrix}$. This yields the final general form for the transformation matrix, which can be turned into the required transformation group via the introduction of a parameter $s\in\left[0,1\right]$:
$$G_s=
\begin{psmallmatrix}
1 & 0 & 0 & \dotsm & 0 & 0 & 0 & 0 & 0 & 0 & 0 & \dotsm & 0 & 0\\
0 & \cos\left(\vartheta_1s\right) & -\sin\left(\vartheta_1s\right) & \dotsm & 0 & 0 & 0 & 0 & 0 & 0 & 0 & \dotsm & 0 & 0\\
0 & \sin\left(\vartheta_1s\right) & \cos\left(\vartheta_1s\right) & \dotsm & 0 & 0 & 0 & 0 & 0 & 0 & 0 & \dotsm & 0 & 0\\
\scriptscriptstyle \vdots & \scriptscriptstyle \vdots & \scriptscriptstyle \vdots & \scriptscriptstyle \ddots & 0 & 0 & 0 & 0 & 0 & 0 & 0 & \dotsm & 0 & 0\\
0 & 0 & 0 & 0 & \cos\left(\vartheta_ks\right) & -\sin\left(\vartheta_ks\right) & 0 & 0 & 0 & 0 & 0 & \dotsm & 0 & 0\\
0 & 0 & 0 & 0 & \sin\left(\vartheta_ks\right) & \cos\left(\vartheta_ks\right) & 0 & 0 & 0 & 0 & 0 & \dotsm & 0 & 0\\
0 & 0 & 0 & 0 & 0 & 0 & \cos\left(-\frac{\pi}{2}s\right) & -\sin\left(-\frac{\pi}{2}s\right) & 0 & 0 & 0 & \dotsm & 0 & 0\\
0 & 0 & 0 & 0 & 0 & 0 & \sin\left(-\frac{\pi}{2}s\right) & \cos\left(-\frac{\pi}{2}s\right) & 0 & 0 & 0 & \dotsm & 0 & 0\\
0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & \cos\left(\pi s\right) & -\sin\left(\pi s\right) & 0 & \dotsm & 0 & 0\\
0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & \sin\left(\pi s\right) & \cos\left(\pi s\right) & 0 & \dotsm & 0 & 0\\
\scriptscriptstyle \vdots & \scriptscriptstyle \vdots & \scriptscriptstyle \vdots & \scriptscriptstyle \vdots & \scriptscriptstyle \vdots & \scriptscriptstyle \vdots & \scriptscriptstyle \vdots & \scriptscriptstyle \vdots & \scriptscriptstyle \vdots & \scriptscriptstyle \vdots & \scriptscriptstyle \vdots & \scriptscriptstyle \ddots & \scriptscriptstyle \vdots & \scriptscriptstyle \vdots\\
0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & \dotsm & 1 & 0\\
0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & \dotsm & 0 & 1\\
\end{psmallmatrix}\:.$$
Notice that when $s=0$, $G_s=G_0=\mathds{1}$, and when $s=1$, $G_s=G_1=O$. Also, for every value of $s$, the determinant of $G_s$ is always 1, and the first vector is kept constant, which means that $G_s$ describes a proper rotation around the axis defined by the eigenvector corresponding to the synchronization manifold. Moreover, the first vector has always been left untouched by all the possible basis transformations. Thus, as $s$ is varied continuously between 0 and 1, the application of $G_s$ sends $A$ into $B$ continuously, as needed.
Finally, to describe how to obtain a transformation between the two Laplacians $L_0$ and $L_1$, let $R$ be the change-of-basis matrix resulting from the composition of all basis changes done in step 6 of the method, the permutation of the vectors to obtain $O$, and the possible final adjustment in case of negative determinant. Also, to simplify the formalism, in the following we let $B_0\equiv A$ and $B_1\equiv B$. Since $B_0$ and $B_1$ are matrices of eigenvectors, it is $B_0^{\mathrm T}L_0B_0 = D_0$ and $B_1^{\mathrm T}L_1B_1 = D_1$, where $D_0$ and $D_1$ are the diagonal matrices of the eigenvalues of $L_0$ and $L_1$. Also, as $G_s$ is a group of orthogonal transformations, it is $ G_sRB_0=RB_s\quad\forall0\leqslant s\leqslant 1$, where $B_s$ is a basis of $\mathbb{R}^N$. Multiplying this by $R^{\mathrm T}$ on the left yields $$\label{BTdef}
R^{\mathrm T}G_sRB_0 = R^{\mathrm T}RB_s = B_s\:,$$ where we used $R^{\mathrm T}=R^{-1}$.
Now, note that all the basis changes performed are between orthonormal bases. Thus $R$ is an isometry, as is any $G_s$, since they are all proper rigid rotations, and any $B_s$ defines an orthonormal basis of $\mathbb{R}^N$. Then, the Laplacian for the parameter $s$ is given by the matrix $L_s$ that solves the equation $$\label{BTDT}
B_s^{\mathrm T}L_sB_s=D_s\:,$$ where $D_s$ is a diagonal matrix whose elements are the eigenvalues of $L_s$, and $B_s$ consists of the eigenvectors of $L_s$. However, the equation above has two unknowns, namely $L_s$ and $D_s$.
This provides a certain freedom in describing the evolution of the eigenvalues of the Laplacians. For instance, one can choose the simplest evolution, which is given by a set of linear transformations. Then, for all $1\leqslant i\leqslant N$, $$\label{valueevol}
\lambda_i^{(s)} = \left(1-s\right)\lambda_i^{(0)}+s\lambda_i^{(1)}\:,$$ where $\lambda_i^{(0)}$ and $\lambda_i^{(1)}$ are the $i^\mathrm{th}$ eigenvalues of $L_0$ and $L_1$, respectively. Note that this allows for the possibility of degeneracy of some eigenvalues for some particular value of the parameter $s^\ast$. However, the transformation we described leaves all the eigenvectors distinct and separate throughout the evolution. Thus, the Laplacian can always be diagonalized for any value of $s$. Multiplying Eq. \[BTDT\] on the right by $B_s^{\mathrm T}$ yields $B_s^{\mathrm T}L_sB_sB_s^{\mathrm T}=D_sB_s^{\mathrm T}$, hence $B_s^{\mathrm T}L_s=D_sB_s^{\mathrm T}$. Multiplying this on the left by $B_s$, it is $B_sB_s^{\mathrm T}L_s=B_sD_sB_s^{\mathrm T}$, hence $L_s=B_sD_sB_s^{\mathrm T}$. Substituting Eq. \[BTdef\] into this last equation gives $$\label{LvsD}
\begin{split}
L_s = R^\mathrm{T}G_sRB_0D_s\left(R^\mathrm{T}G_sRB_0\right)^\mathrm{T}\\=R^\mathrm{T}G_sRB_0D_sB_0^\mathrm{T}R^\mathrm{T}G_s^\mathrm{T}R\:.
\end{split}$$
In the equation above, $B_0$ is known, $R$ and $G_s$ have been explicitly built, and $D_s$ is completely determined by Eq. \[valueevol\]. Thus, Eq. \[LvsD\] defines the Laplacian for any given value of the parameter $s$. The evolution of the eigenvalues is imposed by Eq. \[valueevol\], and the evolution of the eigenvectors is given by Eq. \[BTdef\].
It is important to stress that the solution given by the linear evolution of the eigenvalues is not unique. There could be, in principle, many allowed transformations of $L_0$ into $L_1$, each characterized by a specific eigenvalue evolution. However, the difference between solutions is only in the $\mathbf K_j$ term in Eq. \[variation\], since the boxed term does not depend on the eigenvalues. Thus, the specific choice of eigenvalue evolution could change the phenomenology of the system studied, but would not modify how the switching between general structures affects the stability. This is akin to the results presented in Ref. [@Boc06_2], which can be considered a special case of the present treatment that occurs when all the Laplacians commute.
With this result, one can finally describe the system evolution through unconstrained topologies. From $t=0$ to $t=t_0$, the boxed term in Eq. \[variation\] vanishes. To compute its value during the switch, first note that the $i^\mathrm{th}$ eigenvector at time $t$ is the $i^\mathrm{th}$ column of $B_s$, with $s\equiv\frac{t-t_0}{t_1-t_0}$. But then, using Eq. \[BTdef\], the $k^\mathrm{th}$ element of the $i^\mathrm{th}$ eigenvector is $$\begin{split}
\left(\mathbf{v}_i\right)_k = \left(B_s\right)_{ki} = \left(R^\mathrm{T}G_sRB_0\right)_{ki} = \\ = \sum_{r=1}^N\sum_{q=1}^N\sum_{x=1}^N R_{rk}\left(G_s\right)_{rq}R_{qx}\left(B_0\right)_{xi}\:.
\end{split}$$
Notice that in the equation above the only term that depends on time is $\left(G_s\right)_{rq}$, since $R$ is just a change-of-basis matrix, and $B_0$ is the matrix of eigenvectors of $L_0$ at time $t=0$. Therefore, it is $${\frac{\mathrm d}{\mathrm {dt}}}\left(\mathbf{v}_i\right)_k = \sum_{r=1}^N\sum_{q=1}^N\sum_{x=1}^N R_{rk}R_{qx}\left(B_0\right)_{xi}\frac{1}{t_1-t_0}{\frac{\mathrm d}{\mathrm {ds}}}\left(G_s\right)_{rq}\:.$$
{width="45.00000%"} {width="45.00000%"}
This allows a fully explicit expression for the boxed term in Eqs. \[variation\] that accounts for the time variation of the eigenvectors:
$$\label{NCbox1}
-\sum_{i=1}^N\mathbf{v}_j^\mathrm{T}\left(t\right)\cdot{\frac{\mathrm d}{\mathrm {dt}}}\mathbf{v}_i\left(t\right)\boldsymbol \eta_i =
-\frac{1}{t_1-t_0}\sum_{i=1}^N\sum_{k=1}^N\left[\sum_{r=1}^N\sum_{q=1}^N\sum_{x=1}^N R_{rk}\left(G_s\right)_{rq}R_{qx}\left(B_0\right)_{xj}\right]\left[\sum_{r=1}^N\sum_{q=1}^N\sum_{x=1}^N R_{rk}R_{qx}\left(B_0\right)_{xi}{\frac{\mathrm d}{\mathrm {ds}}}\left(G_s\right)_{rq}\right]\boldsymbol\eta_i\:.$$
However, for all practical purposes, one does not need to use the expression above directly. In fact, considering that most elements of $G_s$ are 0, it is quite simple to compute and store $B_s$ in a symbolic form. Similarly, most of the ${\frac{\mathrm d}{\mathrm {ds}}}\left(G_s\right)_{rq}$ terms are 0. In fact, they vanish if $r=1$, if $q=1$, if $\left|r-q\right|>1$, if $r>2b+1$, and if $q>2b+1$, where $b$ is the number of “$\sin$–$\cos$” blocks in $G_s$. Also, for all other cases ${\frac{\mathrm d}{\mathrm {ds}}}\left(G_s\right)_{rq}$ is proportional to a sine or a cosine. Then, define $\dot{G_s}$ to be the matrix whose $\left(rq\right)$ element is $\frac{1}{t_1-t_0}{\frac{\mathrm d}{\mathrm {ds}}}\left(G_s\right)_{rq}$; also, define $\dot{B_s}\equiv R^{\mathrm T}\dot{G_s}RB_0$. Again, $\dot{B_s}$ can be easily computed and stored in a symbolic form. Then, Eq. \[NCbox1\] becomes $$\label{NCred}
-\sum_{i=1}^N\mathbf{v}_j^\mathrm{T}\left(t\right)\cdot{\frac{\mathrm d}{\mathrm {dt}}}\mathbf{v}_i\left(t\right)\boldsymbol \eta_i = -\sum_{i=1}^N\sum_{k=1}^N\left(B_s\right)_{kj}\left(\dot{B_s}\right)_{ki}\boldsymbol \eta_i\:.$$ Once more, the equation above can be computed symbolically, and evaluated at any particular $t$, when needed. Note that, despite the seemingly complex expressions, Eqs. \[NCbox1\] and \[NCred\] are straightforward to deal with. This is due to the fact that $G_s$ and $\dot{G_s}$ are always represented as tridiagonal matrices, and, as mentioned above, many of the non-trivial elements of $\dot{G_s}$ vanish as well. Thus, numerical applications of this approach can benefit not only from a restricted amount of needed memory, but also from sparse matrix methods that result in a small computational complexity.
The treatment we built is valid for every positive, finite switching time $t^\ast\equiv t_1-t_0$ between configurations. As explained above, this time cannot vanish, lest the derivatives in Eq. \[variation\] be not defined. Nonetheless, one can wonder about the behaviour of a system when the switching time becomes very small, although non-zero. To this purpose, first note that this time only appears as a multiplicative factor in Eq. \[NCbox1\]. Thus, a very small $t^\ast$ would have the effect of making the boxed term in Eq. \[variation\] much larger than the purely variational term. In this regime, the effects on the the stability of the synchronous solution are due mostly, if not exclusively, to the switching process. In other words, if the expression in Eq. \[NCbox1\] yields positive results, the synchronized state is made more stable in the corresponding direction, and vice versa for negative results, regardless of the contribution coming from the variational term. Note that this is in agreement with the finding that blinking networks can greatly facilitate synchronization [@Bel04; @Sti06]. Similarly, one can consider the opposite limit, namely that of a secular switching for which $t^\ast$ becomes very large, while still remaining finite. In this case, for large enough switching time, the boxed term in Eq. \[variation\] becomes negligible compared to the rest, and the stability is determined entirely by the variational term. This case is very similar to that of an evolution along commutative structures [@Boc06_2]. In fact, in this regime of quasi-static evolution, the structure at any given time $t$ is, to first order, equal to the structure at time $t+\mathrm{dt}$. Thus, it is $\left[L_t,L_{t+\mathrm{dt}}\right]=\boldsymbol\varepsilon$. Therefore, one can treat this case as the commutative one with the addition of a small perturbation. Note that this perturbation does not change the stability of the synchronized state: for a positive variational term, instability is maintained, and for a negative one, the synchrony remains stable. The only uncertainty happens for the critical condition corresponding to a vanishing variational term, for which the perturbation can have either effect on stability.
 Estimating the largest Lyapunov exponent. The average of the logarithm of the norm of $\boldsymbol\eta$ (Eq. \[etavec\]) for the example system converges to approximately $-0.3$ when $\sigma=1$ (black solid line), and to approximately $0.022$ when $\sigma=0.1$ (red dashed line), indicating that the synchronized state is stable in the first case, and unstable in the second. Note the logarithmic time scale.](maxlya.eps){width="45.00000%"}
To illustrate the use of our method, we consider the example of a weighted network of $N=10$ chaotic Rössler oscillators, switching back and forth between two topologies (Fig. \[topotopotopo\]). Letting the state vector $\mathbf x\equiv\left(x,y,z\right)$, each of the oscillators obeys the local dynamics $$\label{numf}
\mathbf f\left(\mathbf x\right) = \left(-y -z, x + 0.165y, 0.2 + z\left(x-10\right)\right)\:,$$ with the output function $$\label{numh}
\mathbf h\left(\mathbf x\right) = \left(0,y,0\right)\:.$$ The switching times and the time periods for which the network remains in each of the two configurations (permanence times) are all set to $0.1$. We perform two simulations, one with interaction strength $\sigma=1$ and one with $\sigma=0.1$, to illustrate two different cases and the sensitivity of our method. To estimate the largest Lyapunov exponent associated to the system of Eq. \[variation\], we compute the time-average of the logarithm of the norm of the vector $$\label{etavec}
\boldsymbol\eta\equiv\left(\boldsymbol\eta_2,\boldsymbol\eta_3,\dotsc,\boldsymbol\eta_N\right)$$ at each integration step. The value to which $<\log\left|\boldsymbol\eta\right|>$ converges is $\Lambda_{\max}$. The results, in Fig. \[LyaEta\], show that for the $\sigma=1$ case the convergence value is approximately $-0.3$, indicating that the synchronized state is stable. Conversely, when $\sigma=0.1$, the estimated Lyapunov exponent is just positive, with a value of approximately $0.022$, corresponding to an unstable synchronized state. To verify this numerical result, we simulated the actual network evolution for the two cases according to Eq. \[eq1\], and with $\mathbf f$ and $\mathbf h$ given by Eqs. \[numf\] and \[numh\] above. Figure \[GlobErr\] shows the time evolution of the global synchronization error $$\label{chierr}
\chi = \frac{1}{3\left(N-1\right)}\sum_{i=2}^N\left(\left|x_i-x_1\right|+\left|y_i-y_1\right|+\left|z_i-z_1\right|\right)\:.$$ For the $\sigma=1$ case, after a certain transient, the synchronization error decays to 0. When the interaction strength is lowered to $\sigma=0.1$, instead, $\chi$ eventually starts growing and oscillates wildly, always taking non-zero values. These results indicate that the system is indeed able to synchronize in the first case, while it never does in the second, in agreement with the numerical calculations of $\Lambda_{\max}$. Thus, the simulations not only confirm the validity of our treatment, but provide an example for which the stability of the synchronized state can be changed by the tuning the parameters controlling the topological evolution.
In summary, we demonstrated how to explicitly solve the problem of constructing an appropriate time evolution of a system of networked dynamical units switching between different topologies. Our method builds the evolution from a mapping of the eigenvectors of the graph Laplacians of the individual structures, and it ensures that the elements of the eigenvectors are differentiable at each intermediate time. This enables the use of the Master Stability Function for network topologies that evolve in time in a fully general and unconstrained way.
 Stability of the synchronized state. The time evolution of the global synchronization error $\chi$ (Eq. \[chierr\]) shows that the system is able to reach synchronization when $\sigma=1$ (solid black line), while it never synchronizes when $\sigma=0.1$ (dashed red line), confirming the numerical results. Note the logarithmic time scale and the break in the vertical axis.](syncherr.eps){width="45.00000%"}
While the connection pathway is not unique, different solutions only affect the variational part of the linearized system. It has to be remarked that our treatment is valid *regardless* of the time scales involved. There is no restriction on the permanence times of the network in each configuration, and the only constraint on the switching times is that they do not vanish. In addition, our method introduces a numerical advantage, in that one only needs to integrate a set of linear equations coupled with a single non-linear one, rather than having to deal with a system entirely composed of non-linear differential equations. Also, this approach does not rely on particular assumptions concerning the structures visited by the systems, and contains the regimes of blinking networks and commutative evolution as its limiting cases. Thus, our results have a natural application in the study of synchronization events in systems for which the temporal scales of the topology evolution are comparable with (or even secular with respect to) those characterizing the evolution of the dynamics in each networked unit. This is a common occurrence in many real-world systems, such as neural networks, where synchronization can become possible due to mutations [@Noe03], or financial market, where global properties are affected by adaptive social dynamics.
The authors would like to thank Kyle Morris for helpful discussions. CIDG acknowledges support by EINS, Network of Excellence in Internet Science, via the European Commission’s FP7 under Communications Networks, Content and Technologies, grant No. 288021.
[99]{} R. Albert and A.-L. Barabási, Rev. Mod. Phys. **74**, 47 (2002). S. Boccaletti *et al.*, Phys. Rep. **424**, 175 (2006). S. Boccaletti *et al.*, Phys. Rep. **544**, 1 (2014). F. Varela, J.-P. Lachaux, E. Rodriguez and J. Martinerie, Nat. Rev. Neurosci. **2**, 229 (2001). E. L. Berlow, Nature **398**, 330 (1999). S. Boccaletti et al. Phys. Rep. **366**, 1 (2002). L. M. Pecora and T. L. Carroll, Phys. Rev. Lett. **80**, 2109 (1998). L. F. Lago-Fernández, R. Huerta, F. Corbacho and J. A. Sigüenza, Phys. Rev. Lett. **84**, 2758 (2000). M. Barahona and L. M. Pecora, Phys. Rev. Lett. **89**, 054101 (2002). T. Nishikawa, A. E. Motter, Y.-C. Lai and F. C. Hoppensteadt, Phys. Rev. Lett. **91**, 014101 (2003). I. V. Belykh, V. N. Belykh and M. Hasler, Physica D **195**, 188 (2004). M. Chavez, D.-U. Hwang, A. Amann, H. G. E. Hentschel and S. Boccaletti, Phys. Rev. Lett. **94**, 218701 (2005). D.-U. Hwang, M. Chavez, A. Amann and S. Boccaletti, Phys. Rev. Lett. **94**, 138701 (2005). A. Motter, C. Zhou and J. Kurths, EPL **69**, 335 (2005). J. Yao, H. O. Wang, Z.-H. Guan and W. Xu, Automatica **45**, 1721 (2009). F. Li and X. Lu, Neural Netw. **44**, 72 (2013). N. Masuda, K. Klemm and V. M. Eguíluz, Phys. Rev. Lett. **111**, 188701 (2013). R. Denamur and I. Matic, Mol. Microbiol. **60**, 820 (2006). P. Csaba, M. D. Maciá, A. Oliver, I. Schachar and A. Buckling, Nature **450**, 1079 (2007). P. W. Anderson, K. Arrow and D. Pines, *The economy as an evolving complex system* (Addison-Wesley, Redwood City, California, 1988). J.-D. J. Han et al. Nature **430**, 88 (2004). S. Boccaletti et al. Phys. Rev. E **74**, 016102 (2006). R. E. Amritkar and C.-K. Hu, Chaos **16**, 015117 (2006). M. Porfiri, D. J. Stilwell, E. M. Bollt and J. D. Skufca, Physica D **224**, 102 (2006). D. Stilwell, E. Bollt and D. Roberson, SIAM J. Appl. Dyn. Syst. **5**, 140 (2006). M. Hasler, V. N. Belykh and I. V. Belykh, SIAM J. Appl. Dyn. Syst. **12**, 1007 (2013). M. Hasler, V. N. Belykh and I. V. Belykh, SIAM J. Appl. Dyn. Syst. **12**, 1031 (2013). K. Josić and R. Rosenbaum, SIAM Rev. **50**, 570 (2008). L. Barreira and Ya. Pesin, *Lyapunov exponents and smooth ergodic theory* (American Mathematical Society, Providence, RI, 2002). C. W. Wu, *Synchronization in complex networks of nonlinear dynamical systems* (World Scientific, Singapore, 2007). F. Sorrentino and M. Porfiri, EPL **93**, 50002 (2011). F. W. Warner, *Foundations of differentiable manifolds and Lie groups* (Scott, Foresman and Company, Glenview, IL, 1971). J. L. Noebels, Annu. Rev. Neurosci. **26**, 599 (2003).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In deciding on a student’s grade in a class, an instructor generally needs to combine many individual grading judgments into one overall judgment. Two relatively common numerical scales used to specify individual grades are the 4-point scale (where each whole number 0-4 corresponds to a letter grade) and the percent scale (where letter grades A through D are uniformly distributed in the top 40% of the scale). This paper uses grading data from a single series of courses offered over a period of 10 years to show that the grade distributions emerging from these two grade scales differed in many ways from each other. Evidence suggests that the differences are due more to the grade scale than to either the students or the instructors. One major difference is that the fraction of students given grades less than C$-$ was over 5 times larger when instructors used the percent scale. The fact that **each** instructor who used both grade scales gave more than 4 times as many of these low grades under percent scale grading suggests that the effect is due to the grade scale rather than the instructor. When the percent scale was first introduced in these courses in 2006, one of the authors of this paper had confidently predicted that any changes in course grading would be negligible. They were not negligible, even for this instructor.'
author:
- 'David J. Webb'
- 'Cassandra A. Paul'
- 'Mary K. Chessey'
title: 'Relative impacts of different grade-scales on student success in introductory physics'
---
\[sec:Introduction\]Introduction
================================
In higher education, the role of grades is paramount. Students enroll in courses, and the grade that they earn in each course indicates the degree to which the student was successful in that course. Good grades indicate that students are successful and they are encouraged to continue in that course and on to more advanced courses in the same topical area. Poor grades on early assessments in a course may cause students to drop a course. Enough poor grades can cause a student to fail a course, which can in turn affect the student’s time-to-degree, their retention in a major, or even in their retention in college itself. Furthermore, college grade point averages (GPAs) are important as graduate schools and professional schools have minimum requirements for applicants, and some employers request GPA and/or transcript information so course grades influence student career opportunities and pathways even after graduation. Because the stakes are high, it’s important that educators take care to construct the meaning behind their grades and also to understand any implications of chosen grading techniques or philosophies.
Instructors administering grades on many individual assignments and exams must aggregate these data in assigning a single course grade that best describes each student’s understanding and/or skill. While instructors can and do use different factors and criteria when determining a grade [@Henderson2004Grading; @Brookhart1993], two things are important to achieve in this regard, 1) that the student’s grade is a meaningful representation of their achievements in the class, and 2) that the grading process is consistent such that students taking the same course can expect similar grades for similar successes and failures. Choosing how to award grades usually involves selecting a grade-scale, or numerical method, for evaluating students work. Perhaps the most commonly used grade scale is the percent-based grading scale (hereafter referred to as “percent scale”). A student being graded using the percent scale earns a 90 for example if 90% of the work being graded is correct.
Recently, the percent scale has come under criticism [@Guskey2013; @Reeves2004; @Connor2011; @Wormeli2006] and alternative methods of grading have been introduced. Some of these alternative grade scales and methods have been shown to be quite successful [@Brookhart2016], even when directly compared [@Carey2012] to the percent scale. However, despite the wide range of criticisms [@Guskey2013; @Reeves2004; @Connor2011; @Wormeli2006], there has yet to be significant study of the impact, on students, of the problematic issues associated with the percent scale. In this paper, we address this gap in the literature by examining some of the effects of the percent scale compared to an alternative grade scale. Specifically, we address common criticisms of the percent scale, and determine the validity of these criticisms in one particular context as we explore the similarities and differences in student’s course grades resulting from these two grade scales.
\[sec:level1\]Percent Scale
===========================
The percent scale (shown in Table \[tab1\]) is perhaps the most prevalent way of awarding numerical grades to students both in higher education and in the K-12 context in the United States. On the percent scale, 100 is the highest grade students can earn, zero is the lowest, the numbers represent the percent correct, and a value around 65% is the boundary between passing and failing (see Table \[tab1\]). In general, the basic concept is that the number of points awarded indicates the percent of the assignment that is ‘correct.’ So a grade of 85 would mean that 85% of the assignment is evaluated as correct. There is also a somewhat standard way to interpret this numerical grade as a letter grade. For example, above 90 is generally a grade of ‘A’, between 80 and 90 is usually considered a ‘B’ etc. Any grade below 60 or 65 on the percent scale is typically considered failing, and a zero is often awarded for missing work. There are slight variations of this scale depending on the instructor and/or institutional rules, but any scale where a grade is based on percent and students need to earn more than 50% of the points to earn a passing grade can be considered percent scale grading. Despite it’s popularity, there are some notable criticisms of this scale.
------------ --------------- ----------- ----------- -----------
**Letter**
**Grade** Percent Scale 4.0 Scale CLASP10 CLASP4
A+ 97-100 4.0 9.67-10 4.17-4.5
A 93-96 4.0 9.33-9.67 3.83-4.17
A$-$ 90-92 3.7 9.0-9.33 3.5-3.83
B+ 87-89 3.3 8.67-9.0 3.17-3.5
B 83-86 3.0 8.33-8.67 2.83-3.17
B$-$ 80-82 2.7 8.0-8.33 2.5-2.83
C+ 77-79 2.3 7.67-8.0 2.17-2.5
C 73-76 2.0 7.33-7.67 1.83-2.17
C$-$ 70-72 1.7 7.0-7.33 1.5-1.83
D+ 67-69 1.3 6.67-7.0 1.17-1.5
D 65-66 1.0 6.33-6.67 0.83-1.17
D$-$ 6.0-6.33 0.5-0.83
E/F 0-65 0.0 0-6.0 0-0.5
------------ --------------- ----------- ----------- -----------
: Comparison between grade and score different grading schemes. The Letter Grade to Percent Scale and 4.0 Scale conversions are copied from the College Board website. CLASP10 is the specific version of the percent scale whose results are discussed in the paper and CLASP4 is the specific 4-point scale used.[]{data-label="tab1"}
\[sec:level2\]Criticisms of the Percent Scale
=============================================
One criticism of the percent scale is the portion of the scale devoted to failing grades. In a 2013 article, Guskey [@Guskey2013] points out that a larger portion of the scale is devoted to failure (60%) than success (40%). This means that failure (grades equivalent to ‘F’) can be measured in 60 different degrees, while each other letter grade (A, B, C and D) is limited to only ten degrees. Another way of stating this is that the grade space devoted to ‘F’ is 6 times larger than that of any other letter grade. That the majority of the scale is devoted to F is a potentially a philosophical problem; in fact Guskey [@Guskey2013] asks, “What message does that communicate to students?” But the problems are mathematical as well because, as discussed by Connor and Wormeli [@Connor2011] any of the F-grades below 50 tend to skew an averaging procedure [@Reeves2004; @Wormeli2006].
The amount of the scale devoted to ‘F’ grades is particularly important [@Guskey2013; @Connor2011; @Reeves2004; @Wormeli2006] when considering awarding the lowest grade, a grade of 0%. Zero grades are often given to students who skip an assignment. While there are conflicting viewpoints on whether missed assignments should be included in aggregate grades at all (see Chap’s 14-17 of Ref. [@Wormeli2006]), if we take the viewpoint that there are some instances where a earning a 0% on the percent scale is warranted, critics point out that it is a very difficult thing for a student to balance out the effect of even a single zero. Table \[tab2\] shows us that it takes a total of 19 perfect scores to fully erase (i.e. eventually receive an ‘A’) the impact of a single zero, or a minimum of 6 perfect scores to earn a B. In previous work [@Paul2018], we have shown that zeros are more likely to be awarded to first generation students and to students identifying as members of racial/ethnic groups that are underrepresented in physics than to other students. Furthermore, we provided evidence that these zeros are not clearly indicative of lack of understanding and we noted that use of the percent scale makes this situation more of a problem for these students than another grade scale (CLASP4 in Table \[tab1\]). Research [@Wormeli2006; @Selby1992; @Grant2003] has shown that while good grades are motivators of good work, poor grades may not always motivate students to work harder.
**Desired Grade** **N%** **N4**
------------------- -------- --------
D 2 $<$ 1
C 3 1
B 6 2
A 19 8
: After receiving score of zero, how many perfect scores (either 100% or 4.5) are needed to achieve the desired average grade? N% is number needed when using a percent scale and N4 is the number needed when using the CLASP4 scale from Table \[tab1\]).[]{data-label="tab2"}
Another criticism of the percent scale is that because there are so many levels (100) it is prone to inaccuracy and inconsistency. Studies of teachers suggest that different teachers grades assigned to the same student work tend to be distributed over a range of width of order 10 when using the percent scale. This suggests that using fewer than 100 specific grades may lead to less teacher-to-teacher grade variability.
Finally there are a number of criticisms of the scale that are not mathematical in nature, for example the fact that low failing grades (grades below 50% for example) negatively motivate students, or reduce their self-efficacy [@Cavallo2004; @Hazari2010]. These are also important arguments to consider, but our quantitative data does not address these issues and so in this paper we are primarily concerning ourselves with the mathematical implications of grade scale.
\[sec:level1\]Alternative Grade Scales and Practices
====================================================
There are a few different common alternatives to using the percent scale. For example, in order to mitigate the percent scale issue of devastating zeros, the concept of “minimum grading” was conceived [@Connor2011]. Minimum grading is the practice of raising all very low grades to some ‘minimum’ grade (usually 50%) so that students are able to recover from missing work, or just really poor assignment performances. Critics of minimum grading suggest that it may result in students passing even if they have not learned the material. They argue that minimum grading promotes student entitlement (they get something for nothing) and leads to social promotion. Some instructors feel strongly that it’s unfair to give students 50% if they have completed less than 50% of the work. There are also concerns that minimum grading contributes to well-documented grade inflation [@Rojstaczer2012] (a phenomenon that may include a higher average grade, more students being given ‘A’s, or both). Research on minimum grading [@Carey2012] in one school district shows that neither of these things happened. Carey and Carifio [@Carey2012] present an analysis of seven years of grading data collected from a school that implemented minimum grading. They used standardized test results to show that students who earned at least one minimum grade actually outperformed students who did not receive any minimum grades on standardized testing. This was true even though students receiving minimum grades on average had lower classroom grades. This indicates that minimum grading does not cause grade inflation and also that minimum grading still does not entirely make up for the fact that a percent grading scheme potentially under-measures the performance of the struggling students for whom the practice of minimum grading was designed to help. Unfortunately, even though this seems to be a reasonable way to address some concerns of the percent scale, many instructors don’t like it [@Reeves2004], because they do not agree with the concept of giving students scores that they don’t believe they have earned.
Another way to address the percent scale issues is to use the concept of Standards-Based grading (discussed in Brookhart, et al review [@Brookhart2016] and references therein). This method asks students to demonstrate proficiency in certain areas, with the instructor providing ordinal grades such as well below proficiency, approaching proficiency, proficient, and excellent, that describe the path to proficiency [@Anderson2018]. This approach has been used in college physics classes previously [@Beatty2013], but, because the method requires students to be able to have multiple opportunities to attempt the same proficiency, this is difficult to accomplish with the large class sizes that are typical of introductory science courses.
A final alternative to the percent scale is the college 4.0 scale which is typically used to calculate GPA. Each integer in the scale (4, 3, 2, 1, 0) corresponds to a letter grade (A, B, C, D, F). While many college level instructors may use the percent scale and then convert to a letter grade (which has a numeric value tied to the 4.0 scale), others may use a 4.0 scale from the beginning of the course and then report the numbers in the form of letter grades. These numbers can be averaged into a single grade. Because the 4.0 scale allocates the same ‘space’ to each letter grade, the scale avoids most of the issues associated with the percent scale. For example, in Table \[tab2\] we see that after a student earns a zero, it only takes 2 perfect scores for them to earn a grade of “B”, and a total of 8 for them to erase the zero completed. This is much more doable than the 6 and 19 perfect scores required by the percent scale to achieve the same goal. This means that a grade of ‘zero’ is less disastrous for students. This scale is mathematically similar to the practice of minimum grading, and instructors using standards-based grading also sometimes use the integers on a 4.0 scale to mean different levels of proficiency [@Anderson2018; @Beatty2013], so aspects of this scale exist in both of the other alternatives we describe. Because the 4.0 scale mitigates many of the criticisms of the percent-scale, and does not have the any new drawbacks like minimum grading and standards-based grading do, many suggest it as an alternative to percent scale grading.
\[sec:level1\]Research Questions
================================
The active learning introductory physics course CLASP at UC Davis, has been in existence since 1995 [@Potter2014Sixteen]. Originally, all courses used the same grade scale described as “CLASP4” in Table \[tab1\]. This grade scale was based on the standard college 4.0 scale. After many years, instructors began to move away from the CLASP4 grade scale, and began to utilize a percent scale instead. Because the course materials over the years were extremely similar, and some instructors used both types of grade scale in different sections of the same course, the circumstances provide for an ideal opportunity to compare the usage of examples of the two scales. In this paper we consider CLASP10 to be an example usage of the percent scale, and CLASP4 to be an example of a 4-point scale. We use this data to compare use of the percent scale, CLASP10 to the 4-point scale, CLASP4, to examine some common critiques of the percent scale and to further explore similarities and differences between these two grade scales.
**Specifically, regarding the controversies over percent scale grading, we ask:**
1\) Does the 4-point scale lead to course grades that are “inflated” compared to percent scale grading?
2\) How does the distribution of course grades differ between the percent scale and 4-point grading?
3\) How does the distribution of exam-item grades differ between the percent scale and 4-point grading?
4\) How variable are the course grades for each grade scale by instructor and by class?
5\) How does averaging grades under these different numerical scales compare with aggregating grades using the median instead of the average?
When considering these questions it’s important to emphasize that our aim here is to uncover how different grade scales can impact student outcomes. We are not evaluating the philosophy behind either scale, nor making claims about what constitutes an “A” or an “F.” Finally we will make no claims about the connection between assigned grade and student understanding.
In this article we examine 10 years of student grades in a large enrollment introductory physics college course. Instructors in this course sometimes used a percent scale (CLASP10), and sometimes used an alternative scale (CLASP4) based on the standard collegiate 4.0 scale. What we will show is that the fraction of students failing the course is much larger when instructors use the percent scale and that instructors assigned more F grades to their student’s work when using the percent scale. We also find that the increase in students failing is associated mostly with the grade scale used when aggregating grades by averaging and not with the increase in individual assigned F’s. These conclusions seem to be determined by the grade scale rather than the instructor. Furthermore, we find that that using the percent scale leads to more class-to-class variability in grade distribution. Finally, we illustrate how using the arithmetic mean to determine course grades is similar to using the median under the 4-point grade scale but less similar under the percent scale.
\[sec:Methods\]Methods
======================
\[sec:Setting\]Setting and Context
----------------------------------
When calculating a GPA, most colleges use a 4.0 scale which equates each integer (4, 3, 2, 1, 0) to a corresponding letter grade (A, B, C, D, F). The designers of the Collaborative Learning Through Active Sense-making in Physics (CLASP) [@Potter2014Sixteen] curriculum wanted a grading system that was both transparent and non-competitive, and so they directly linked every single graded item (be it an exam question, or the exam itself) to a slightly modified version of this 4-point scale so that students could understand how their performance on a given question related to the expectations of the course instructors. The resulting CLASP4 grade scale, shown in Table \[tab1\], is therefore a version of the 4.0 scale. CLASP4 is a continuous grade scale from 0 to 4.5 with each letter grade region, except for the F region, centered on the appropriate integer. Generally, a grade of 4.5 (the highest A+) was earned when the description/calculation correctly and completely applied an appropriate physical model to the situation described in the exam. Descriptions/calculations that were not correct or were not complete were assigned lower grades with each grade depending on the instructor’s judgment of the quality of the answer. An answer judged to be excellent but not perfect was given a score in the A$-$ to A$+$ range between 3.5 and 4.5, an answer in the B$-$ to B$+$ range was assigned between 2.5 and 3.5, in the C$-$ to C$+$ range between 1.5 to 2.5, etc. Under this grade scale a zero was almost universally reserved for students who did not answer the question at all, the top of the F range of grades was 0.5, and grades between 0 and 0.5 were used for students who gave an answer but whose answer showed almost no familiarity with or understanding of the subject. Multi-part exam problems often had a grade for each part (sub-exam level grades) and these were averaged, with the weight per part determined by the instructor, to determine the exam grade. These exam grades were then averaged, with weight per exam determined by the instructor, to give the course grade on the same 4-point scale. From 1995 until 2005 essentially all CLASP instructors used this same basic grading method for quizzes, exams, and the class grades[^1].
In 2006 several instructors began experimenting with a 10-point grade scale, CLASP10, that was just a re-scaled version of the standard percent scale. With the CLASP10 grade scale, answers in the A$-$ to A$+$ range were given grades 9 to 10 (instead of the 3.5 to 4.5 of the 4-point scale), B$-$ to B$+$ range were given grades 8 to 9, etc. for C and D grade ranges. Again the zero of this grade scale was reserved for students who did not answer the question but now the highest F grade given is 6.0. The result is that, relative to CLASP4 scale, CLASP10 had a much larger grading measure available for F’s (0-6) even though the other grades have the same measure on each scale. For these reasons we will generally refer to CLASP10 as a “percent scale”. The grades on a single exam problem were then, as under CLASP4, averaged to give the exam grade and the exams were similarly averaged to give a course score which then determined the course letter grade.
\[sec:Data Set\]Data Set
------------------------
Data were collected from the first two quarters of the three quarter CLASP series from course archives spanning the ten years 2003-2012. (The “CLASP A” content primarily covers energy and thermodynamics, while the “CLASP B” content focuses on mechanics.) During those years the structure and content of these classes was relatively constant. Over 75% of a student’s time in one of these classes was spent working in discussion/laboratory sections (referred to as DLs in the CLASP curriculum) on activities that only changed slowly over those years. The platform for entering grades in this large enrollment class was centralized using a separate database file for every separate course offering. These course databases include exam scores for each student along with the individual grades that were given as well as the calculations that led to these exam scores. In addition, these databases sometimes included the calculations that led to the actual course grades. Over these ten years there were 133 of these classes, and we have found databases for 96 of them that are identifiably graded as described above using either CLASP4 grading or percent scale grading. This identification was determined by examining the maximum grades given to individual student answers. If the maximum was always 10 then we considered the class to have had percent grading and if the maximum grades were always 4.5 then we considered the class to have had CLASP4 scale. The resulting database contains 773,667 sub-exam level grades given to 15,757 individual students on each part of each exam. Fifty-seven of the included classes are CLASP4 grade scale (including 478,617 grades on individual answers) and the remaining 39 are percent grade scale (including 295,050 grades on individual answers). We also have access, from UC Davis administration, to the recorded course grades from all of these classes as recorded by the Registrar. Course grades were all recorded on the standard 4-point grade scale with letter grades A through D, each of which may include a $+$ or $-$, and F. In each database that included the calculation of course (letter) grades these letter grades were determined using the cutoffs approximately as shown in Table \[tab1\]. Although we will group together the data from the first two courses (CLASP A and CLASP B) in this series of courses because the differences between the grade scales show up in both courses, we will specifically note each situation where the data from one course differ substantially from that of the other.
Each of our research questions requires a different set of comparisons to make. So rather than providing a list of justifications for the comparability of each set in this section, we instead share this information when it can be considered alongside the research question and resulting comparison.
\[sec:Results\]Analysis & Results
=================================
Throughout our analysis we compare the percent scale, CLASP10, to the 4-point scale, CLASP4, used by the instructors of CLASP. In addition to course grades we will also be reporting on individual scores that instructors gave to student answers on exams. These scores will always be referenced by the grade range in which they are contained. For instance, a 3.8 given to a student answer under CLASP4 grading will be considered to be in the A region just the same as a 9.3 given to a student answer under CLASP10 grading. We treat borderline grades (i.e. a 3.5 under CLASP4 grading or the equivalent 9.0 under CLASP10 grading) as being half in each of the bordering letter grade ranges. Most counts, averages, standard errors, etc. were calculated in Excel and most are double-checked with STATA software. We used STATA for the standard statistical tests, calculations, and regressions. The error estimates we give will be standard error of the mean or propagated from standard errors unless otherwise noted.
\[sec:PercentScaleFailsMore\]Percent scale fails more students
--------------------------------------------------------------
When one separates the classes taught according to grade scale, either 4-point or percent scale, a trend in student fail-rates is undeniably present. Figure \[Fig1\] shows the fraction of course grades given that are less than C$-$ as a function of the year. We choose a cutoff of C$-$ to measure because UC Davis allows students receiving less than a C$-$ to repeat a course whereas those with C$-$ or higher cannot. A similar way of noting that C$-$ is an important cutoff is that courses graded Pass/Not-Pass give the Pass grade only to students who would have received a grade of C$-$ or higher. Therefore, the figure shows the fraction of students who have a grade considered low enough to warrant repeating the course. From the averages shown on the figure we see that instructors using the percent grading scale gave $5.3 \pm 0.4$ times as many grades lower than C$-$ than instructors using a 4-point scale. This fraction was $4.5\pm 0.5$ in the first course in the series (CLASP A) and $6.9\pm0.9$ in the second course (CLASP B) so the two courses both showed this grade scale effect. A chi-square test for the entire data set shows this difference between grade scales is significant, $\chi^2(2, N=22,865) = 513.4, P < 0.001$. Of course, it’s conceivable that this difference is primarily an issue of student academic performance rather than of grade scale used. To explore this possibility we will use i) the variable $EnterGPA$, a student’s GPA upon entering the course, as a predictor of that student’s academic performance and ii) the variable $GrScale$ (the grade scale) as a categorical variable in logistic regression models predicting the odds of a student receiving a grade less than C$-$. The logistic regression model including both of these variables is $$\begin{gathered}
ln[odds(CourseGrade < C-)]\\=b_0 + b_1 EnterGPA + b_2 GrScale \end{gathered}$$ where $e^{b}$ is the appropriate odds ratio (note that the odds of receiving a grade less than C$-$ is equal to the probability of receiving a grade less than C$-$ divided by the probability of receiving a grade of C$-$ or higher). First, leaving out $GrScale$ and including **only** $EnterGPA$ as a predictor of the odds of receiving $<$ C$-$, we find an odds ratio for $EnterGPA$ of $=0.095 \pm 0.009$ $(z=-24.9, N = 20,950, P<0.001, Pseudo R^2=0.12)$. This means that an increase in entering GPA by 1 (e.g. from 2.5 to 3.5) lowered the odds of receiving a grade $<$ C$-$ by over 90%. Now we include the categorical grade scale variable $GrScale$ along with $EnterGPA$. The odds ratio for $EnterGPA$ changes to $0.079 \pm 0.008$ and we find that a student graded under a percent scale had $6.5 \pm 0.6$ times higher odds of receiving less than C$-$ than the same student under 4-point grading ($z=-19.9, N = 20,950, P<0.001$ for the variable $GrScale$ in this model) and $Pseudo R^2=0.195$ for this model. We conclude that student academic performance does not explain the large fraction of students with these low grades under percent grading. Finally, we should point out that the student withdrawal/drop rates are essentially independent of grade scale ($0.76\%\pm0.08\%$ under 4-point grading and $0.74\%\pm0.09\%$ under percent grading) in this set of classes. What isn’t shown in these data is why more students were failed when the percent scale was used. In the remainder of this paper we analyze factors that contribute to this phenomenon.
\[sec:GradeInflation\]Is Grade Inflation Happening?
---------------------------------------------------
Because instructors select either the percent scale or the 4-point scale, it’s important to consider the instructors also in this analysis. One possibility is that instructors using the 4-point scale give generally higher grades to their students’ answers than those given by instructors using a percent scale, a situation commonly referred to as “Grade Inflation”. If 4-point scale instructors were simply inflating grades, we would expect to find fewer low grades but we would also expect to find more high grades under 4-point grading as well as a higher average grade.
Figure \[Fig2\] shows the complete course grade distribution for both grade scales. One possible sign of grade inflation in 4-point classes, more high grades, is easily seen to be missing. Instead, we find that students in percent scale classes are the ones receiving more A’s (about 20% more than their peers in 4-point classes).
To look for a shift in the average grade, we computed the average course grade given by each grade scale using the UC Davis method for calculating GPA (A=4.0, A$-$=3.7, B+=3.3, B=3.0, etc.) except that we use A+=4.3 rather than the UC Davis A+=4.0. This amounts to choosing, for each course grade, a value roughly in the middle of the relevant CLASP4 range shown in Table \[tab1\]. We find the average grade given to a student graded under the percent scale was 2.846 (SD = 0.89) and under the 4-point scale was 2.915 (SD = 0.67) for a grade shift of 0.069 (0.01). These average grades (both between a B and a B$-$) are shown in Figure \[Fig2\]. The effect size of this grade shift is about 0.088 so it is a small effect. The difference of 0.07 GPA units is less than half of the class-to-class variation for either grade scale (standard deviation, over the individual classes, of average class grade is 0.19 for 4-point classes and 0.34 for percent scale classes). A t-test of the two distributions shows that this small difference in the two average grades is statistically significant ($t=6.6, df=22863, P<0.001$) and including student GPA as a covariate does not change this conclusion. At this point we should note that the two courses that we have grouped together here showed different results for this comparison when considered individually. The first-quarter course (CLASP A) had essentially the same average grades for the two grade scales ($2.858\pm0.008$ for 4-point grading and $2.874\pm0.012$ for percent grading). The second quarter course (CLASP B) had a lower average under percent grading ($2.935\pm0.009$ for 4-point grading and $2.757\pm0.014$ for percent grading). So there is conflicting evidence for any simply defined “grade inflation.” We find that students graded using the percent scale are more likely than students graded using the 4-point scale to earn “A” grades, but that they are also more likely to fail the class, and these two differences combine in a way that the average grade in courses graded using the 4-point scale is statistically significantly higher but that the difference is small. When we examine the distribution of grades for both courses in figure \[Fig2\], we see evidence that suggests fewer students fail under the 4-point scale not because the distribution of course grades under that grade scale is shifted uniformly toward higher grades but that the course grade distribution under 4-point grading was narrower than it was for percent scale grading.
\[sec:UseofGradeSpace\]Instructor Use of Grade Space
----------------------------------------------------
Since very low grades given under percent scale grading have a much larger effect on course grade than the lowest grades given under 4-point scale grading (Table \[tab2\]), we might expect that this is the main difference between the two grade scales and accounts for the difference in the student fail-rates. However, there is another clear difference in our data that can also lead to more failing grades. We find differences in how instructors allocate grades on the individual answers given by their students on exams that differ between the two scales.
First, we note that students who leave a problem blank (or mostly blank) on an exam receive a zero, a grade that involves no instructor judgment of understanding or skill. Figure \[Fig3\] shows the fractions of scores given on individual exam items for all 96 courses. We see that there is very little difference between the number of zeros earned by students under each scale. However, the fraction of F-grades that are not zero is distinctly dependent on the grade scale. Figure \[Fig3\] shows that instructors using a percent scale are considerably more likely to judge individual student solutions on exam items as non-zero F’s than those instructors using the 4-point scale. This amounts to shifting about 14% of the entire exam item grade weight from higher grades under 4-point grading down to non-zero F’s under percent grading. Notice that none of this shifted grade weight comes from the A’s, which were actually more common under percent grading and likely little of it from B’s because the total fraction of (A’s + B’s) is about the same for the two scales, 56.4% for percent grading and 56.9% for 4-point grading. In order to analyze how these grade shifts affected individual students and whether they might be due to student academic performance, we compute the fraction of non-zero F’s for each student. Averaging this student-level number over all students gives $0.0364\pm 0.0004$ under 4-point grading and $0.1636\pm 0.0014$ under percent grading. Cohen’s d for this difference is 1.3 so this is, not surprisingly, a large effect. The student-level distribution of non-zero F’s is, unfortunately, both non-normal and heteroskedastic but we can still use student GPA, $EnterGPA$, in a linear regression to see if it affects the extra fraction of non-zero F’s seen for percent grading. The regression model we use to model a student’s fraction of non-zero F’s is $$\begin{gathered}
FractionOfNonZeroFs\\=b_0 + b_1 EnterGPA + b_2 GrScale\end{gathered}$$ First, leaving out $GrScale$ and using **only** $EnterGPA$ to predict the exam item fraction of F’s with $N = 20,837$ gives, coefficient for $EnterGPA=-0.056, t=-35, P<0.001, R^2=0.056$. Including $GrScale$ with $EnterGPA$ in our model we find that the percent scales had $12.6\%\pm0.1\%$ of extra grade weight in non-zero F’s compared to 4-point scales, $t=99, P<0.001$ for the variable $GrScale$, and $N=20,837$ and $R^2=0.36$ for this model. Controlling for student GPA did not change the fraction of shifted grade weight very much so we conclude that the extra non-zero F’s under percent scale grading are not due to differences in the students in these courses. Finally, we note that the two courses that we have grouped together here had different amounts of grade weight shifted into non-zero F’s. In the first course of the series (CLASP A) $8.5\%\pm0.2\%$ of grade weight was shifted into nonzero F’s under percent grading and in the second course of the series (CLASP B) that number was $18.6\%\pm0.3\%$.
\[sec:Selection\]Individual Instructors’ Results
------------------------------------------------
One possible explanation for the difference between the failing rates of the two grade scales is that different instructors choose the scale that serves their interest better. Therefore, it is useful to address any selection effects the choice of grade scale may have. Of the 60 instructors involved in these courses over the 10 years in our data set, seven instructors used both the 4-point scale and the percent scale at various times. This gives us seven comparisons between the two grade scales where an instructor[^2] is held constant. Table \[tab3\] shows that these seven instructors gave between 4 and 12 times more course grades less than C$-$ under percent grading than under 4-point grading with an overall average of $5.4\pm 0.5$ times as many grades less than C$-$, a number very similar to the overall result described previously. In addition, each of these instructors had between 4% and 20% extra exam-item grade weight in the non-zero F region under percent grading than that particular instructor had under 4-point grading. These numbers are modified only very slightly if we attempt to account for class-to-class differences in the student academic performance by using their incoming GPA as a covariate. Specifically, the largest shift, caused by controlling for student GPA, in Extra F Weight is from 0.12 to 0.11 for Instructor 4 and, overall, we still find, 4% and 20% more sub-exam-level grade weight in the non-zero F region under percent grading. These data are evidence against either instructor or students as causes of the effects shown in figures \[Fig1\], \[Fig2\], and \[Fig3\].
**Instructor** **N4** **N%** **Fail Ratio** **Extra F Weight**
---------------- -------- -------- ---------------- -------------------- --
1 2824 1700 7.5 (2.0) 0.099 (0.005)
2 1472 952 7.7 (1.7) 0.158 (0.004)
3 1920 267 4.1 (1.1) 0.043 (0.004)
4 591 509 4.2 (1.2) 0.122 (0.006)
5 3558 754 4.5 (0.9) 0.163 (0.004)
6 662 297 6.7 (5.4) 0.076 (0.004)
7 1273 258 12 (4) 0.196 (0.008)
: Seven instructors used both grade scales at various times. N’s are the numbers of students taught under the particular grading regime. The Fail Ratio is the fraction of that instructor’s students with course grades less than C$-$ under percent grading divided by that fraction under 4-point grading. Extra F Weight is the extra fractional weight of nonzero F’s that the instructor gave to individual student exam answers under percent grading above what they gave under 4-point. Standard errors are in parentheses. []{data-label="tab3"}
\[sec:Variability\]Variability of the Percent Scale
---------------------------------------------------
Figure \[Fig1\] shows that the fail-rates for the instructors using the percent scale are more variable by year than the 4-point scale. We can quantify this on a course-to-course basis by examining how the fail rate (i.e. fraction of students with grade $<$ C$-$) varies over the courses offered. Figure \[Fig5\] has a kernel density plot of course failure rate for each grade scale. One finds not only that the failure rate is larger for the percent scale courses but that it is also much more variable over courses. It is possible that the students themselves were more variable in percent scale courses and we can check this using the student’s incoming GPA’s to find the distribution over classes, of the class-average GPA, for each grade scale. For 4-point classes we find a class-GPA averaged over classes of 2.97 with a standard deviation over classes of 0.08 and for the percent scale classes we find a class-GPA average of 2.99 with a standard deviation of 0.08. A t-test of these class-average GPA’s gives $t=1.1, df=93, P=0.26$ so the students under the two grade scales are, in terms of average GPA, indistinguishable from each other. We can also examine whether the fraction of students failing in a class varies with the class-average GPA. We find the linear regression coefficient is not significant for either grade scale (P = 0.74 for 4-point scales and P = 0.16 for percent scales). So the variability in failing fraction is not obviously associated with an underlying variability in the students’ academic performance. Finally, since both of the distributions in Figure \[Fig5\] include classes at the lower bound of zero students failing, we might worry that being pushed against this lower bound has artificially narrowed the distribution for the 4-point scale, a floor-effect. We check this by using a higher grade cutoff. A cutoff of C still had classes from each scale with zero students below cutoff, so we use a cutoff of C+ and find that percent scale classes had an average fraction of 0.22 students with lower than C+ with a standard deviation over classes of 0.12 and 4-point graded classes had an average fraction of 0.12 student with lower than C+ and a standard deviation of 0.07. The percent scale classes are more variable under this measure also.
If the variability in fail-rates is not associated with the students then maybe it is an inter-instructor effect where each instructor tends to fail about the same number of students in each percent course but that different instructors fail very different numbers of students. We can turn, again, to instructors who used both 4-point and percent grading at various times during these 10 years. Figure \[Fig7\] shows the fail rates for two of these instructors. Both of these instructors have considerable variation of the fail-rates in their classes. The fail-rates in their 4-point courses have much smaller variation than the fail-rates in their percent courses. Unfortunately, only four instructors taught at least two courses under each grade scale but each of them had a larger spread in fail rates under percent grading than under 4-point grading (the smallest increase in the spread of fail rates was a factor of 1.7 higher for percent scales and the highest was a factor of 5.5 higher). So, the variability is not obviously an instructor selection effect.
\[sec:Weight vs. F\]Separation of grade weight effects from extra F’s
---------------------------------------------------------------------
From Table \[tab2\] we see that the percent scale requires 3 perfect grades of a student wanting to raise the lowest F-grade (i.e. a zero) to a straight C. Similarly, we note that percent scale F-grades of 25 and 50 require 2 and 1 perfect grades respectively to average to a C. The highest percent scale F grades average in much the same way that F’s do with a 4-point scale. Since the lower F-grades require more perfect grades to cancel them out, we can think of them as F’s that carry more effective “weight” than any F from a 4-point scale. These “weighty” F’s lead to the “skewing” of average grades discussed by Connor and Wormeli [@Connor2011].
As noted earlier, some of the class database files that we have access to not only have all grades recorded but also include all of the calculations that led to the course grades. Courses offered in 2008 and later years do not have all of these calculations but eight of the thirteen percent graded classes offered 2006-2007 do have complete sets of grades and grade calculations. These eight classes gave 1839 course grades to 1272 students (567 students are in two of these particular classes). The eight classes give us a way to separate the effects of the heavy weighing of low F’s from the effects of just giving more F’s on the individual exam answers. Without changing the number of F’s (or any other grade) that were given on exams in these eight classes we map the CLASP10 percent scale grades onto the CLASP4 scale as follows: i) for all grades larger than 6.0 we subtract 5.5 from the grade and ii) in order to be very conservative in our treatment of the F’s we set all original CLASP10 grades less than or equal to 6.0 to a 0 in our 4-point re-scaling (that is, **all** original F’s are set to 0). We then do all the original weighted averages of individual grades into exam grades and exam grades into course grades where point cutoffs determining the letter grades are gotten by subtracting 5.5 from all of the original percent scale grade cutoffs. The original percent graded courses had a fail rate of 8.3% $\pm$ 0.6% and the 4-point re-scaled courses had a fail rate of 1.3% $\pm$ 0.3%. The 4-point rescaled fail rate of 1.3% is consistent with the rest of the 4-point graded classes (see Figures \[Fig1\] and \[Fig5\]) and the increase of a factor of 6.5 $\pm$ 1.4 in the fail rate under percent grading is consistent with both the overall ratio of 5.3 and the individual instructor ratios given in Table \[tab3\]. These overall consistencies from a simple rescaling onto a 4-point scale suggests that **the heavy effective weight that the low-F grades carry is the main factor increasing the failure rates** and that the extra F’s assigned by instructors using the percent scale for individual exam answers aren’t the main difference between the two course grade distributions.
\[sec:Median\]Aggregating Grades - Mean vs Median
-------------------------------------------------
An alternative to using the mean to aggregate student grades is to use the median [@Brookhart2004]. This is perhaps most straightforward with letter grades because they are explicitly ordinal but could also be chosen for numerical grades that are referenced to letter grades when they are given by an instructor whose grading philosophy is that the grades they give are ordinal in nature. For the CLASP courses the numerical grades are chosen from a continuous number line and were always averaged but these individual grades are ultimately connected to letter grades so one could argue that taking a median might be a reasonable choice here in constructing course grades. In fact, the argument that it is best to consider even numerical grades to be ordinal and so use the median to aggregate grades has already been made by others [@Wright1994]. For our grade data an interesting similarity between the two grade scales is that the overall median of all of the grades given to individual student answers under 4-point grading was 3.0 (exactly middle B) and the overall median under CLASP10 was 8.5 (also exactly a middle B). It may help better understand the differences between the two grade scales to use the grade data we have discussed in this paper to compare, at the student level, the two methods, median and average, of determining course grades.
The actual course grades given during the 10 years of these data are always made up of complicated weightings of the individual grading judgments. This complicated weighting can even be student dependent because instructors would almost always drop each student’s lowest quiz score, give the final exam more weight if the student performed better on that exam than on exams during the term, and adjust some student’s grades slightly depending on their performance in their discussion/lab section. This makes it impossible to devise a unique process of using the median function to determine a final grade that can be directly compared to the actual final grades. Nevertheless, it may be useful in understanding the similarities and differences between the two grade scales to just construct a straight unweighted average of each student’s individual grades and then compare those averages with a similarly unweighted median of each student’s grades. We find that 49 of the courses graded under 4-scale grading (11,708 students) and 24 courses under 10-scale grading (5,862 students) have a complete set of grades and so can be used in this way.
Figure \[Fig8\] shows each student’s unweighted average grade as a function of their unweighted median grade for both grade scales. The relationship between an average grade and a median grade, for a set of grades that have both an upper bound and a lower bound, is evident in the figure. Students with many grades at the upper bound can have a median grade equal to the upper bound but any other grades they have will lower their average grade and exactly what other grades they have will determine how much lower their average grade is, so average grades will tend to be lower than median near the upper bound. Conversely, average grades will tend to be higher than median when the median falls near the lower grade boundary. The result of those two general features is that average grades and median grades must be similar for some region in the middle of the grade space. For the grades given to the students in these courses, the average approximates the median in the B- to B region of grades under 4-point grading and at the F to D crossover region of grades under percent grading. This means that under percent grading the region of grades that includes 92% of student grades (median grade $\geq$ D) will tend to give lower grades when the average is used than when the median is used.
Another effect that is obvious from Figure \[Fig8\] is that, for any specific value of the median grade, percent grading gives a much larger spread in average grades. The standard deviation in the distribution of average grades is roughly twice as large for percent as for 4-point grading for any particular letter grade, determined by the median, from D$-$ to A$+$. So, in the courses we are considering here, percent scale grading had both a larger systematic shift, of the average from the median, than 4-point grading and a larger random spread around that systematic shift.
\[sec:Discussion\]Discussion
============================
Considering only the mathematical characteristics of each scale, it is perhaps not surprising that more students fail under the percent scale. After all, if lower grades are given more weight, it’s likely that more students will fail. That said, there are several nuances in the overall grade distributions of each scale uncovered by our analysis that are worth considering by instructors who are considering using either scale.
The above results confirm many of the critiques of the percent scale discussed in Section \[sec:level2\]. A complete understanding of the grade scale entails understanding these critiques so that the percent scale can be used consistently and effectively. In fact, many instructors are aware of some of these critiques, and adjust percent scale grades accordingly (for example, they might consider a lower grade than 65% passing or they might add some number of points to all students grades to increase the class average.) The intention of this discussion is to discuss the nuances of the use of the percent scale in one particular context in order to bring these characteristics attention so that they may be discussed both by the research community and instructors considering their grading philosophy.
\[sec:partial credit\] Considering Partial Credit
-------------------------------------------------
If a student does not complete a problem correctly but shows some small part of understanding, instructors will often award an accordingly small amount of “partial credit.” Figure \[Fig3\] shows that a very large portion of grades earned on exam items graded using the percent scale is devoted to “Non-Zero F’s.” In fact, more students earn “Non-Zero F’s” on the percent scale than any other individual grade besides “A”. Mathematically, there is a big difference between averaging in a 40% and a 10%, but does the instructor see a correspondingly meaningful difference between these two different grades for failing? Perhaps, but is this the difference between these two failing grades equivalent to distinguishing the differences between a grade of D (65%) and an A (95%) which are also 30 points apart? If an instructor thinks that the numerical grades they assign are actually interval in nature, rather than just ordinal [@Stevens1946; @Brookhart2004], and they are averaging the individual grades to determine some aggregate grade, then the same consideration as to whether to award a D or an A should go into determining whether a student earns an F (10%) or an F (40%). Specifically, it seems important that instructors avoid the mind frame of awarding percents lower than 50% and thinking that this is giving the student partial credit if they consider a grade of roughly 60% to be the border between passing and failing. For example, while 30% is obviously better than a zero it is still 30 points below failing, which is the same mathematical difference between an A and an F. The data in our paper show the collective effects of the very low F-grades of the percent scale. And the fact that each instructor who used both grade scales gave more F’s suggests that the grade scale itself might affect a teacher’s grading judgments in cases when a student’s answer does not show much understanding.
\[sec:Considering zeros\] The Meaning of a Zero
-----------------------------------------------
The effect of awarding a zero is greater in the percent scale than in the 4-point scale, as shown by Table \[tab2\] and therefore contribute to the grade weighting issues discussed in section \[sec:Weight vs. F\]. When a student leaves an exam problem blank for any reason, the instructor often awards them a zero for this. The instructor’s justification for this is certainly logical in the sense that the student has provided zero evidence of understanding, and therefore has earned 0% of the possible points. The instructor might also be using the zero for a motivational purpose in the sense that students need to complete the work in order to earn points. However, many studies [@Grant2003; @Shim2005; @Corpus2009] suggest that this simple view of motivation is unwarranted. In addition, as seen in Figure \[Fig3\], our data set shows that even though a zero carries much more weight on the percent scale than on the 4-point scale, the overall fraction of blank problems remains essentially constant. This is tentative evidence in support of the fact that the number of blank responses is not affected by choice of grade scale in this course and possible additional evidence against the motivational use of a zero. Furthermore, leaving a physics problem blank is a behavioral trait that may be more common [@Paul2018] for women, students identifying as underrepresented minorities, and first generation college students. Therefore the practice of awarding zeros for missing work may well contribute to achievement gaps.
As pointed out by many authors [@Guskey2002; @Wormeli2006], a zero that is earned because a student did not complete an assignment is not a measurement, it’s actually missing data. In previous work [@Paul2018] we showed that the number of problems left blank by a student is poorly correlated with other metrics of understanding, so that leaving a blank is by no means predictive of that student’s overall understanding of physics. Averaging in zeros for missing data would be a terrible practice in one’s research and so we might consider alternatives to this practice should this concern us in our teaching.
\[sec:Valuing the Instructor’s Evaluation\]Valuing the Instructor’s Evaluation
------------------------------------------------------------------------------
In figure \[Fig3\] we showed that the instructors using the percent scale gave more non-zero “F” grades than instructors using the 4-point scale and, using Table \[tab3\], argued that these extra F’s were due to the grade scale rather than the instructor. However later, in Section \[sec:Weight vs. F\], we argued that the reason that more students failed the course when graded using the percent scale as compared to the 4-point scale, was due to the mathematics of averaging percent scale grades, and NOT because those instructors actually gave more “F” grades on individual exam items when using the percent scale. In some ways this second point seems to temper the first point but the two effects should really be considered together. For example, regarding the second point, an instructor could argue that giving a very low F grade and using it in an averaging process that gives it a large effective weight is entirely appropriate because that is what their student “earned.” However the first point, that the percent scale seems to have guided seven out of seven instructors into giving more of these F grades, might give this instructor pause regarding their own ability to provide impartial absolute grading judgments of these poor student answers. The set of low F grades which average with higher effective weight are exactly the grades that may be caused by the grade scale itself and so be particularly hard to be confident about.
We further characterized the student-level skewing effects of these low F grades in Section \[sec:Median\]. In Figure \[Fig8\] we used unweighted averages to show how the percent scale might allow for students to earn average grades that are consistently lower than their median grade and sometimes much lower. Of course the actual instructor weightings (like dropping the lowest grade) would likely reduce the scale of some of these effects but will also reduce similar but smaller effects of the 4-point scale and are not likely to change our conclusions that the low F grades available to instructors using the percent scale i) can have very large effects and ii) may only have been given to a student because of the grade scale used.
\[sec:Considering Variation at the course level\]Considering Variation at the course level
------------------------------------------------------------------------------------------
Beyond the student-level effects discussed above, there also appear to be instructor-level/class-level effects associated with the skewing effect of the low F grades in a percent scale. We have already seen that the effective weight of these low F grades leads to the increase in the fail rate so the class-to-class fail rate variability that we saw in Figure \[Fig7\] must be caused by an underlying class-to-class variability in the number of these low F’s and/or various instructor’s responses to the low average grades they see in front of them as they determine course grades for their students. If there are a set of possible grades that carry extra weight when averaging then the actual grades given to the students in that class may well be very sensitive to exactly how those powerful grades are used during the term with this sensitivity giving rise to the class-to-class variability.
\[sec:Conclusions\]Conclusions
==============================
In summarizing our results we again emphasize that we are not casting judgment on use of the percent scale in general or any grading practices in particular, but instead argue that it is essential for instructors to consider the biases of the percent scale when planning their course for the semester, so that they can ensure that their teaching philosophies match their grading philosophies.
The primary purpose of these analyses is to fully understand the impact of the percent scale as compared to another somewhat commonly used scale, the 4-point scale. In fact, even knowing these results, one of the authors has chosen to continue using a version of the percent scale in their small graduate courses. In this use case, work that does not meet expectations earns no lower than 60%, but zeros for missing assignments are used to ensure it’s not possible to pass the course without completing the assignments. This example is shared NOT as an example of an exemplar use of the scale (we do not have such data to support such a claim), but rather to emphasize that the intention of this paper is not to discredit the percent scale, but rather to expose some characteristics of the percent scale that may have previously avoided consideration due to its widespread use.
With respect to our research questions, we draw the following conclusions.
RQ1) The 4-point scale does not appear to inflate grades in the traditional use of the word because the average grades of the two scales are close and students are actually over 30% **less likely** to earn A grades in courses using the 4-point scale.
RQ2) Although the average course grades under the two scales are close, the width of the distribution is much larger under percent grading. This led to many more students receiving failing course grades when the percent scale is used as compared to the 4-point scale. We found that the odds of failing is over 5 times higher under percent scale grading, $P<0.001$. The overall fail rate varies by instructor and by individual class for the percent scale with an average fail rate of about 8%. Nevertheless, **each** instructor who used both scales at various times failed, overall, at least four times as many students under the percent scale.
RQ3) Instructors tend to give out more “F” grades on individual exam items when using the percent scale than when using the 4-point scale ( 13% to 14% of the entire grade weight was shifted down into the F-region under percent grading, effect size = 1.3). We have not seen this sort of effect reported in the literature. However, the number of extra F’s is not found to be the main contributor to the higher course fail rate under percent scales. Rather, the extra “effective weight” of these low F’s in the averaging process is the main contributor. The grade scale is more important in determining the fail rate than the instructor.
RQ4) The percent scale has a much more variable fail rate when compared to the 4-point scale. The class-to-class variation of the fail rate under percent grading is over seven times higher than under 4-point scale grading even though the variation in the students was negligible.
RQ5) The two grade-aggregating methods, mean and median, gave results more consistent with each other under 4-point scales than under percent scales. Under percent scale grading 88% of students had unweighted average grades lower than their unweighted median grade and 12% had higher averages. Conversely, for 4-point scales 56% of students had lower unweighted averages than unweighted means and 44% had higher averages. In addition to the systematically lower averages under percent grading, the distribution of average grades for any particular median grade was about twice as large under percent grading as under 4-point grading.
These results are derived from a data set that comes entirely from two (sequential) courses offered over a period of ten years by one department at a single institution. This course has a fairly low fail rate regardless of which scale is used. Findings would likely vary across institutions. While we have made every effort to account for student and instructor selection effects, this is not a randomized controlled study. It is possible that there are unseen factors contributing to higher fail-rates in the courses graded using the percent scale. Future work will examine similar data sets at other institutions offering CLASP and other Introductory Physics for Life Sciences courses. Furthermore, a big question we have not addressed here is what happens to these students after completing this course. Are students who would have failed under the percent scale, but passed under the 4-point scale successful in future courses? We do not address this question in this paper, as our intent is not to prove one grading method is superior to another, but rather to uncover characteristics of the percent and 4-point scales that are important for instructors to consider when deciding on a grading practice for their courses. We do plan to investigate this question in forthcoming work.
Each college or university level instructor has their own opinion about the quality of a student’s work but this judgment should represent an unbiased opinion of that work. Toward that end, it is useful for instructors to know the origins of possible biases so that they can account for these in assigning grades. Our results indicate that instructor use of the 4-point scale led to many more students passing their introductory physics course as compared with classes using the percent scale. This result was achieved without grade inflation. Our findings align with previous critiques of the percent scale, and indicate that instructors should consider the specific issues we highlight in this paper when using the percent scale.
\[sec:Acknowledgements\]Acknowledgements
========================================
None of these results would have been possible without the organized databases that Wendell Potter set up in 1997 and continued through his retirement in 2006 so, in memoriam, we owe him a debt of gratitude. We would also like to thank the education research groups at UC Davis and San Jose State for useful comments on the research and the manuscript. We also would like to thank Jayson Nissen for providing feedback on an earlier draft of this paper.
[23]{}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 [****, ()](https://doi.org/10.1119/1.1634963) @noop [****, ()]{} [****, ()](https://doi.org/10.1136/bmj.314.7085.970) @noop [****, ()]{} @noop [****, ()]{} @noop [**]{}, ed. (, , ) [****, ()](https://doi.org/10.3102/0034654316672069) [****, ()](https://doi.org/10.3102/0013189X12453309) in @noop [**]{}, (, ) @noop [****, ()]{} @noop [****, ()]{} [****, ()](https://doi.org/10.1111/j.1949-8594.2004.tb18000.x) [****, ()](https://doi.org/10.1002/tea.20363) [****, ()](https://adams.idm.oclc.org/login?url=http://search.ebscohost.com/login.aspx?direct=true{&}db=eue{&}AN=78174629{&}site=ehost-live{&}scope=site) @noop [****, ()]{}, @noop [****, ()]{} [****, ()](https://doi.org/10.1119/1.4857435) [****, ()](https://doi.org/10.1002/tl.165), @noop [****, ()]{} [****, ()](https://doi.org/10.1126/science.103.2684.677) [****, ()](https://doi.org/10.3200/JEXE.73.4.333-349) @noop [****, ()]{} [****, ()](https://doi.org/10.1177/003172170208301013)
[^1]: Homework was generally not graded and there are no homework grades in these databases.
[^2]: Each non-summer course has two instructors, both of whom are responsible for the grades given.
| {
"pile_set_name": "ArXiv"
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.