text
stringlengths 4
2.78M
| meta
dict |
|---|---|
---
author:
- |
Francisco Correa\
Centro de Estudios Científicos (CECs), Valdivia, Chile.\
E-mail:
- |
Mokhtar Hassaine\
Instituto de Matemática y Física, Universidad de Talca,\
Casilla 747, Talca, Chile.\
E-mail:
title: Thermodynamics of Lovelock black holes with a nonminimal scalar field
---
Introduction
============
During the last decades, the interests on higher-dimensional physics have grown up particulary concerning the higher-dimensional General Relativity. In this context, apart from the standard Einstein-Hilbert action, there also exists an interesting gravity theory in dimension $d\geq 5$ involving higher powers of the curvatures such that the field equations for the metric are at most of second-order. This theory known as the Lovelock gravity has been first implemented in five dimensions by Lanczos [@LAN] and then generalized in higher dimension $d\geq 5$ by Lovelock [@LOV]. The resulting action is a $d-$form constructed out of the vielbein, the spin connection and their exterior derivative without using the Hodge dual. The invariance under local Lorentz transformations of the Lovelock Lagrangian can be extended to a local anti-de Sitter (AdS) or Poincaré symmetry through a particular choice of the coefficients appearing in the Lovelock expansion. In both cases, the resulting Lagrangian is a Chern-Simons form whose supersymmetric extensions are also known; see [@Zanelli:2005sa] for a good review on Chern-Simons (super)gravity. The Lovelock gravity or its Chern-Simons particular case have been shown to possess (topological) AdS black hole solutions with interesting thermodynamical properties [@Banados:1993ur; @Cai:1998vy; @Crisostomo:2000bb; @Aros:2000ij] generalizing those obtained in the Einstein-Gauss-Bonnet case [@Boulware:1985wk; @Cai:2001dz].
In order to source the Lovelock gravity with the purpose of obtaining black hole solutions, nonminimally coupled scalar field can be an excellent candidate. Indeed, the nonminimal coupling may be useful to escape the standard no-hair theorems, and black hole solutions are known in this case in standard Einstein gravity (with and without cosmological constant) in four dimensions with a conformal coupling [@Bekenstein:1974sf; @Bocharova:1970; @Martinez:2002ru; @Martinez:2005di]. Recently, there has been a renewal interest concerning this kind of source particularly in order to gain a better understanding of some unconventional superconductors [@Horowitz:2010gk; @Hartnoll:2008kx; @Herzog:2009xv]. Indeed, it is believed that black hole solutions with scalar hair at low temperature that disappears at high temperature will be of particular importance for the unconventional superconductors since they will reproduce the correct behavior of the superconductor phase diagram.
In the present work, we consider a series of particular Lovelock gravity actions indexed by an integer $k$ in dimension $d$, and sourced by a self-interacting nonminimal scalar field. To be more precise, the arbitrary coefficients appearing in the Lovelock series are fixed by requiring that the resulting theory has a unique anti-de Sitter vacuum with a fixed cosmological constant This yields to $k=1,2\cdots, [\frac{d-1}{2}]$ inequivalent gravity theories where $k=1$ corresponds to the standard Einstein-Hilbert action. Note that such model with a mass term potential has been considered in [@Gaete:2013ixa; @Gaete:2013oda], and black hole solutions with planar horizon have been obtained for two particular values of the nonminimal coupling parameter. Here, we generalize these solutions and obtain black hole solutions for any value of the nonminimal coupling parameter. This task is achieved thanks to the introduction of a much more general self-interacting term that depends explicitly on the integer index $k$. More specifically, we show that for each inequivalent Lovelock gravity theory indexed by $k$, there exists a judicious choice for the self-interacting potential that permits to derive black hole solutions for arbitrary value of the nonminimal coupling parameter. The solutions are shown to be uniparametric and reduce to those derived in [@Gaete:2013ixa; @Gaete:2013oda] for the special values of the nonminimal coupling parameter. The thermodynamics study of the solutions realized through the Hamiltonian analysis reveals that these configurations have a zero mass as well as a vanishing entropy. More precisely, we show that the mass contribution coming from the gravity side exactly cancels the mass contribution inherent to the matter source. The vanishing of the entropy can be also corroborated by the fact that the entropy of the solutions is proportional to the lapse function evaluated at the horizon. Since the constant appearing in the black hole solution does not contribute to any conserved charge, we interpret it as a sort of hair which turns out to be inversely proportional to the temperature. Hence, at high temperature, this hair will disappear and this kind of behavior is excepted in the unconventional superconductors in order to correctly reproduce the phase diagram.
The plan of the paper is organized as follows. In the next section, we present the model, the field equations and the solutions. Sec. III is devoted to the thermodynamical analysis while our conclusions and comments are given in the last section.
Lovelock black hole solution with arbitrary nonminimal coupling parameter
=========================================================================
We consider a generalization of the Einstein-Hilbert gravity action in arbitrary dimension $d$ yielding at most to second-order field equations for the metric and known as the Lovelock Lagrangian. This latter is a $d-$form constructed with the vielbein $e^a$, the spin connection $\omega^{ab}$, and their exterior derivatives without using the Hodge dual. The Lovelock action is a polynomial of degree $[d/2]$ (where $[x]$ denotes the integer part of $x$) in the curvature two-form, $R^{ab} = d\,\omega^{ab} + \omega^{a}_{\;c}
\wedge \omega^{cb}$ as
$$\begin{aligned}
&&\int \sum_{p=0}^{[d/2]}\alpha_p~ L^{(p)},\\
&& L^{(p)}=\epsilon_{a_1\cdots a_d} R^{a_1a_2}\cdots
R^{a_{2p-1}a_{2p}}e^{a_{2p+1}}\cdots e^{a_d},\end{aligned}$$
where the $\alpha_p$ are arbitrary dimensionful coupling constants and where wedge products between forms are understood. Here $L^{(0)}$ and $L^{(1)}$ are proportional respectively to the cosmological term and the Einstein-Hilbert Lagrangian. As shown in Ref. [@Crisostomo:2000bb], requiring the Lovelock action to have a unique AdS vacuum with a unique cosmological constant, fixes the $\alpha_p$ yielding to a series of actions indexed by an integer $k$, and given by $$\begin{aligned}
I_k=-\frac{1}{2k(d-3)!}\int \sum_{p=0}^{k}\frac{C^k_p}{(d-2p)}~
L^{(p)},\qquad\qquad 1\leq k\leq \Big[\frac{d-1}{2}\Big], \label{Ik}\end{aligned}$$ where $C^k_p$ corresponds to the combinatorial factor. The global factor in front of the integral is chosen such that the gravity action (\[Ik\]) can be re-written in the standard fashion as $$\begin{aligned}
I_k=\frac{1}2\int d^{d}x\,\sqrt{-g} \Big[&&R+\frac{(d-1)(d-2)}{k}+\frac{(k-1)}{2(d-3)(d-4)}{\cal L}_{GB}\nonumber\\
&&+
\frac{(k-1)(k-2)}{3!(d-3)(d-4)(d-5)(d-6)}{\cal L}_{(3)}+\cdots
\Big], \label{Ik2}\end{aligned}$$ where ${\cal L}_{GB}=R^{2}-4\,R_{\mu \nu}R^{\mu
\nu}+R_{\alpha\beta\mu\nu}R^{\alpha\beta\mu\nu}$ stands for the Gauss-Bonnet Lagrangian, and ${\cal L}_{(3)}$ is given by $$\begin{aligned}
{\cal L}_{(3)}&=&R^3 -12RR_{\mu \nu } R^{\mu \nu } + 16\,R_{\mu
\nu }R^{\mu }_{\phantom{\mu} \rho }R^{\nu \rho } + 24 R_{\mu \nu
}R_{\rho \sigma }R^{\mu \rho \nu \sigma }+ 3RR_{\mu \nu \rho \sigma
} R^{\mu \nu \rho \sigma }
\nonumber \\
&&-24R_{\mu \nu }R^{\mu} _{\phantom{\mu} \rho \sigma \kappa } R^{\nu
\rho \sigma \kappa }+ 4 R_{\mu \nu \rho \sigma }R^{\mu \nu \eta
\zeta } R^{\rho \sigma }_{\phantom{\rho \sigma} \eta \zeta }-8R_{\mu
\rho \nu \sigma } R^{\mu \phantom{\eta} \nu \phantom{\zeta}
}_{\phantom{\mu} \eta \phantom{\nu} \zeta } R^{\rho \eta \sigma
\zeta }.\end{aligned}$$ Note that in odd dimension $d=2n-1$ and for $k=n-1$, the corresponding action $I_{n-1}$ is a Chern-Simons action, that is a $(2n-1)-$form whose exterior derivative can be written as the contraction of an invariant tensor with the wedge product of $n$ curvatures two-forms. In even dimension $d=2n$, the maximal value of $k$ is $n-1$, and in this case the resulting gravity action has a Born-Infeld like structure since it can be written as the Pfaffian of the $2-$form $\bar{R}^{ab}=R^{ab}+e^ae^b$. The gravity theories $I_k$ have been shown to possess black hole solutions with interesting features, in particular concerning their thermodynamics properties, see [@Crisostomo:2000bb] and [@Aros:2000ij].
For each $k\geq 2$, we source the gravity actions $I_k$ with a self-interacting and nonminimally coupled scalar field, that is $$\begin{aligned}
S_k=&I_k-\int d^{d}x\,\sqrt{-g} \Big[
\frac{1}{2}\partial_{\mu}\Phi\partial^{\mu}\Phi+\frac{\xi}2 R\Phi^2
+U_k(\Phi)\Big],\label{Sk}\end{aligned}$$ where $\xi$ denotes the nonminimal coupling parameter and $U_k$ is a potential term which depends explicitly on the index $k$, and whose form will be given below. The field equations read
\[eqsmotionk\] $$\begin{aligned}
&&{\cal G}^{(k)}_{\mu\nu}=T_{\mu\nu},\\
&&\Box \Phi = \xi R \Phi+\frac{d U_k}{d\Phi},\end{aligned}$$
where ${\cal G}^{(k)}_{\mu\nu}$ is the gravity tensor associated to the variation of the action $I_k$ (\[Ik\]), $$\begin{aligned}
{\cal G}^{(k)}_{\mu\nu}=&&G_{\mu\nu}-\frac{(d-1)(d-2)}{2k}g_{\mu\nu}+\frac{(k-1)}{2(d-3)(d-4)}K_{\mu\nu}\\
+&&\frac{(k-1)(k-2)}{3!(d-3)(d-4)(d-5)(d-6)}S_{\mu\nu}+\cdots\end{aligned}$$ where $K_{\mu\nu}$ is the Gauss-Bonnet tensor $$K_{\mu\nu}=
2\big(RR_{\mu\nu}-2R_{\mu\rho}R^{\rho}_{\phantom{\rho}\nu}-2R^{\rho\sigma}R_{\mu\rho\nu\sigma}
+R_{\mu}^{\phantom{\mu}\rho\sigma\gamma}R_{\nu\rho\sigma\gamma}\big)
-\frac{1}{2}\, g_{\mu\nu}\mathcal{L}_{GB}$$ and $S_{\mu\nu}$ arises from the variation of ${\cal L}_{(3)}$, $$\begin{aligned}
S_{\mu\nu}&=&\,3\Big(R^2 R_{\mu \nu } - 4 R R_{\rho \mu } R^{\rho}
_{\phantom{\rho} \nu } - 4R^{\rho \sigma }R_{\rho \sigma }R_{\mu \nu
} +8 R^{\rho \sigma }R_{\rho \mu }R_{\sigma \nu } - 4 R R^{\rho
\sigma }R_{\rho \mu \sigma \nu }
\nonumber \\
&& +8 R^{\rho \kappa }R^\sigma _{\phantom{\sigma }\kappa }R_{\rho
\mu \sigma \nu } -16 R^{\rho \sigma }R^\kappa _{\phantom{\kappa }
(\mu }R_{ |\kappa \sigma \rho | \nu ) } + 2 R R^{\rho \sigma \kappa
}_{\phantom{\rho \sigma \kappa }\mu }R_{\rho \sigma \kappa \nu }
+R_{\mu \nu }R^{\rho \sigma \kappa \eta } R_{\rho \sigma \kappa \eta
}
\nonumber \\
&&- 8 R^\rho _{\phantom{\rho }(\mu }R^{\sigma \kappa \eta
}_{\phantom{\sigma \kappa \eta } |\rho | }R_{|\sigma \kappa \eta |
\nu ) } - 4 R^{\rho \sigma }R^{\kappa \eta }_{\phantom{\kappa \eta }
\rho \mu }R_{\kappa \eta \sigma \nu }+8R_{\rho \sigma }R^{\rho
\kappa \sigma \eta }R_{\kappa \mu \eta \nu } - 8 R_{\rho \sigma
}R^{\rho \kappa \eta }_{\phantom{\rho \kappa \eta }\mu }R^\sigma
_{\phantom{\sigma } \kappa \eta \nu }
\nonumber \\
&&+4 R^{\rho \sigma \kappa \eta }R_{\rho \sigma \zeta \mu
}R_{\kappa \eta \phantom{\zeta } \nu }^{\phantom{\kappa \eta }\zeta
} - 8 R^{\rho \kappa \sigma \eta }R^\zeta _{\phantom{\zeta }\rho
\sigma \mu }R_{\zeta \kappa \eta \nu }- 4R^{\rho \sigma \kappa
}_{\phantom{\rho \sigma \kappa } \eta } R_{\rho \sigma \kappa \zeta
}R^{\eta \phantom{\mu }\zeta }_{\phantom{\eta } \mu \phantom{\zeta
} \nu }\Big)- \frac12 g_{\mu \nu }{\cal L}_{(3)}.\end{aligned}$$ In the matter part of the equations, $T_{\mu\nu}$ stands for the energy-momentum tensor of the scalar field whose expression is given by $$\begin{aligned}
\label{tmunusf} T_{\mu \nu}=&&\partial_{\mu}\Phi\partial_{\nu}
\Phi-g_{\mu
\nu}\Big(\frac{1}{2}\,\partial_{\sigma}\Phi\partial^{\sigma}\Phi+U_k(\Phi)\Big)+\xi\left(g_{\mu
\nu}\Box-\nabla_{\mu}\nabla_{\nu}+G_{\mu\nu}\right)\Phi^{2}.\end{aligned}$$ For each inequivalent theory $k\geq 2$ and for $\xi\not=\frac{1}{4}$, the potential is given by the sum of the six following terms $$\begin{aligned}
\label{potLov}
&&U_k(\Phi)=\frac{1}{(4\xi-1)^2}\left[\alpha_1\Phi^2+\alpha_2\,b\Phi^{\frac{1}{2\xi}}+\alpha_3\,b^2\Phi^{\frac{1-2\xi}{\xi}}+\alpha_4\Phi^{\frac{2k}{k-1}}
+\alpha_5\,b\Phi^{\frac{4\xi+k-1}{2\xi(k-1)}}+\alpha_6\,b^2\Phi^{\frac{2\xi(2-k)+k-1}{\xi(k-1)}}\right]\nonumber\\\end{aligned}$$ and depends of a parameter $b$ and the constants $\alpha_i$ which read $$\begin{aligned}
&&\alpha_1=8\xi d(d-1)(\xi-\xi_d)(\xi-\xi_{d+1}),\qquad \alpha_2=-16\xi^2 (d-1)(\xi-\xi_d),\qquad \alpha_3=2\xi^2\\
&&\alpha_4=-\frac{8\xi^{\frac{k}{k-1}}(1+(k-1)d)((k-1)d+2-k)}{(k-1)k}\left(\xi-\hat{\xi}_{k,d}\right)
\left(\xi-\hat{\xi}_{k,d+1}\right)\\
&&\alpha_5=\frac{16\xi\left(k(d-1)-(d-2)\right)-4d(k-1)+8(k-1)}{(k-1)}\xi^{\frac{2k-1}{k-1}},\qquad\alpha_6=-\frac{8\xi+2(k-1)}{k-1}\xi^{\frac{2k-1}{k-1}}\end{aligned}$$ Here $\xi_{d}$ denotes the conformal coupling in $d$ dimensions and we have defined $\hat{\xi}_{k,d}$ by $$\begin{aligned}
\label{xi}
\xi_d=\frac{d-2}{4(d-1)},\qquad \hat{\xi}_{k,d}=\frac{(d-2)(k-1)}{4\left[(d-1)k-(d-2)\right]}.\end{aligned}$$ Various comments can be made concerning the particular form of the potential (\[potLov\]). We anticipate that this potential naturally emerges looking for solutions of the field equations with an Ansatz of the form $$\begin{aligned}
ds^2=-F(r)dt^2+\frac{dr^2}{F(r)}+r^2d\vec{x}_{d-2}^2,\qquad \Phi=\Phi(r).
\label{ansatz}\end{aligned}$$ We also note that the self-interacting term depends explicitly on a coupling constant $b$ as well as on the index parameter of the Lovelock theories $k$, and this dependence on $k$ only concerns the last three terms of the potential. The first three terms of $U_k$ exactly reproduce the potential that appears in many different contexts involving nonminimally coupled scalar field as for examples for the stealth configuration [@AyonBeato:2004ig] on the BTZ black hole [@Banados:1992wn], or when looking for AdS wave backgrounds [@AyonBeato:2005qq], or for black hole solutions when adding axionic fields coupled with the scalar field [@Caldarelli:2013gqa]. For $b=0$ and $\xi=\hat{\xi}_{k,d}$ or $\xi=\hat{\xi}_{k,d+1}$, the potential reduces to the mass term considered in [@Gaete:2013ixa; @Gaete:2013oda] for which black hole solutions have been found. It is also clear from its expression that $U_k$ is not well defined for $\xi=\frac{1}{4}$, and its derivation for this particular coupling involves logarithmic pieces.
Let us now present some solutions of the field equations (\[eqsmotionk\]). For $k\geq 2$ and $\xi\not=\frac1{4}$, an uniparametric black hole solution with planar horizon is given by $$\begin{aligned}
\label{solk}
&&ds^{2}=- r^2\left(1-\left[\xi(ar+b)^{\frac{4\xi}{4\xi-1}}\right]^{\frac{1}{k-1}}\right)dt^{2}+\frac{dr^2}{ r^2\left(1-\left[\xi(ar+b)^{\frac{4\xi}{4\xi-1}}\right]^{\frac{1}{k-1}}\right)}+{r^{2}}d\vec{x}_{d-2}^2,\nonumber\\
&&\Phi(r)=(ar+b)^{\frac{2\xi}{4\xi-1}},\end{aligned}$$ where $a$ is an integration constant. We first stress that this solution is valid for any arbitrary value of the nonminimal coupling parameter $\xi\not=0$, and for $\xi\in ]0,\frac{1}{4}[$, the solution is asymptotically AdS while for $\xi>\frac{1}{4}$ the asymptotic behavior of the metric is faster that the usual AdS one. It is interesting to note that the singularity $r_s$ is localized at $r_s=-\frac{b}{a}$ but it can always be hidden by the horizon $r_h=-\frac{b}{a}+\frac{\xi^{\frac{1-4\xi}{4\xi}}}{a}$ provided that the constant $a>0$. For $b=0$, and for $\xi=\hat{\xi}_{k,d}$ or for $\xi=\hat{\xi}_{k,d+1}$, the solutions (\[solk\]) reduce to those found in the case of a mass term potential [@Gaete:2013ixa; @Gaete:2013oda]. We would like also to point out an important remark concerning the lapse function appearing in the metric solution that will be of importance in the thermodynamics study of these solutions. The metric function solution can be expressed in term of the scalar field as $$r^2\left(1-\left[\xi(ar+b)^{\frac{4\xi}{4\xi-1}}\right]^{\frac{1}{k-1}}\right)=r^2\left(1-(\xi\Phi^2)^{\frac{1}{k-1}}\right)$$ and, hence the localization of the horizon $r_h$ is such that $$\begin{aligned}
\left(1-\xi\Phi^2\right)|_{r_h}=0.
\label{truc}\end{aligned}$$ However, as we will see below, the entropy of the solutions is always proportional to this quantity (\[truc\]), and consequently the solutions (\[solk\]) will have a zero entropy. We also remark that the existence of these solutions is strongly inherent to the presence of the higher curvature terms present in the Lovelock Lagrangian. Indeed, it is clear from the different expressions obtained here that the standard GR limit $k=1$ is singular, and hence these solutions are only effective for the higher order Lovelock terms $k\geq 2$.
To conclude this section, we present the solution for the coupling $\xi=\frac{1}{4}$. The self-interacting potential is given by $$\begin{aligned}
&&U_{[k,\xi=\frac{1}{4}]}(\Phi)=\frac{1}{8k(k-1)}\Big\{4\left[\ln\left(\frac{\Phi}{b}\right)\right]^2
\left(-4^{\frac{1}{1-k}}k^2\Phi^{\frac{2k}{k-1}}+\Phi^2k(k-1)\right)\\
&&-4k(k-1)(d-1)\ln\left(\frac{\Phi}{b}\right)\left(4^{\frac{1}{1-k}}\Phi^{\frac{2k}{k-1}}-
\Phi^2\right)+(d-2)(d-1)(k-1)\left(-4^{\frac{1}{1-k}}(k-1)\Phi^{\frac{2k}{k-1}}+\Phi^2 k\right)\Big\}.\nonumber\end{aligned}$$ and the equations of motion (\[eqsmotionk\]) admit a solution where the scalar field and the metric are given as follows $$\begin{aligned}
\Phi(r)=b \,e^{ar}, \quad F(r)=r^2\left(1-\left[\frac{1}{4}\, b^2\, e^{2ar} \right]^{\frac{1}{k-1}}\right) \, .
\nonumber\end{aligned}$$
Thermodynamics
==============
The partition function for a thermodynamical ensemble is identified with the Euclidean path integral in the saddle point approximation around the Euclidean continuation of the classical solution [@Gibbons:1976ue]. The Euclidean and Lorentzian action are related by $I_{E}=-iI$ where the periodic Euclidean time is $\tau
=it$. The Euclidean continuation of the class of metrics considered here is given by $$ds^2=N(r)^2F(r)d\tau^2+\frac{dr^2}{F(r)}+r^2\left(dx_1^2+dx_2^2+\cdots+dx_{d-2}^2\right).$$ In order to avoid conical singularity at the horizon in the Euclidean metric, the Euclidean time is made periodic with period $\beta$ and the Hawking temperature $T$ is given by $T=\beta^{-1}$. Here we are interested only in the static solution with a radial scalar field, and hence it is enough to consider a *reduced* action principle. This latter is given by $$\begin{aligned}
\label{redaction}
I_E=&&-\beta\Sigma_{d-2}\int_{r_h}^{\infty}\frac{N}{2k}(d-2)\frac{d}{dr}\left[r^{d-1}\left(1-\frac{F}{r^2}\right)^k\right]\nonumber\\&&+
\beta\Sigma_{d-2}\int_{r_h}^{\infty} N r^{d-2}\Big\{\left(\frac{1-4\xi}{2}\right)F(\Phi^{\prime})^2-
\xi\Phi\Phi^{\prime}\left(F^{\prime}+\frac{2(d-2)}{r}F\right)-2\xi\Phi\Phi^{\prime\prime}F\nonumber\\&&-\frac{(d-2)\xi}{2r}F^{\prime}\Phi^2
+\Phi^2\left(-\frac{\xi}{2r^2}(d-2)(d-3)F\right)+U(\Phi)\Big\}+B\end{aligned}$$ where the radial coordinate $r$ belongs to the range $[r_h,\infty[$ where $r_h$ is the location of the horizon. Here, $\beta$ stands for the inverse of the temperature and $\Sigma_{d-2}$ corresponds to the compactified volume of the planar $(d-2)-$dimensional base manifold. In the reduced action (\[redaction\]), $B$ is a boundary term that is fixed by requiring that the Euclidean action has an extremum, that is $\delta I_{E}=0$ which in turn implies that $$\begin{aligned}
{\delta B}=\beta\Sigma_{d-2}\left({\delta B}_{\mbox{\tiny{gravity}}}+{\delta B}_{\mbox{\tiny{matter}}}\right)\end{aligned}$$ where the first contribution ${\delta B}_{\mbox{\tiny{gravity}}}$ arises from the variation of the gravity part while the second one ${\delta B}_{\mbox{\tiny{matter}}}$ comes from the matter source, and are given by $$\begin{aligned}
{\delta B}_{\mbox{\tiny{gravity}}}=&&\int_{r_h}^{\infty}\frac{N}{2k}(d-2)\frac{d}{dr}\left[r^{d-1}\delta\left(1-\frac{F}{r^2}\right)^k\right],\nonumber\\
\\
{\delta B}_{\mbox{\tiny{matter}}}=&&\int_{r_h}^{\infty}2\xi\Phi Fr^{d-2}\delta\Phi^{\prime}+\int_{r_h}^{\infty}\delta F\left[\frac{\xi(d-2)}{2}r^{d-3}\Phi^2+\xi\Phi\Phi^{\prime}r^{d-2}\right]\nonumber\\
&&+\int_{r_h}^{\infty}\Big\{\delta\Phi\left[F\Phi^{\prime}r^{d-2}(2\xi-1)-\xi F^{\prime}\Phi r^{d-2}\right]\Big\}.\nonumber\end{aligned}$$ In the grand canonical ensemble, the Euclidean action is related with the mass ${\cal M}$ and the entropy ${\cal S}$ by $$\begin{aligned}
I_E=\beta {\cal M}-{\cal S}.
\label{gce}\end{aligned}$$ For the solution (\[solk\]), we have the following asymptotic variation behaviors $$\begin{aligned}
&&\delta F|_{\infty}=-\alpha \xi^{\frac{1}{k-1}}a^{\alpha-1}(\delta a)\left[r^{\alpha+2}+\frac{b(\alpha-1)}{a}r^{\alpha+1}
+\frac{b^2(\alpha-1)(\alpha-2)}{a^2}r^{\alpha}+\cdots\right],\nonumber\\
&&\delta\Phi|_{\infty}=a^{\frac{(k-1)\alpha}{2}}(\delta a)\left[r^{\frac{(k-1)\alpha}{2}}+\frac{(k-1)\alpha b}{2a}r^{\frac{(k-1)\alpha-2}{2}}+\frac{(k-1)\alpha\left((k-1)\alpha-2\right)b^2}{4a^2}r^{\frac{(k-1)\alpha-4}{2}}
+\cdots\right],\\
&&\delta\Phi^{\prime}|_{\infty}=\frac{d}{dr}\left(\delta\Phi|_{\infty}\right),\nonumber
\label{inf}\end{aligned}$$ where for simplicity we have defined $\alpha=\frac{4\xi}{(k-1)(4\xi-1)}$. At the horizon $r_h$ the variations read $$\begin{aligned}
\delta F|_{r_h}=-F^{\prime}|_{r_h}\delta r_h,\qquad
\delta\Phi|_{r_h}=\delta\Phi(r_h)-\Phi^{\prime}|_{r_h}\delta r_h
\label{hor}\end{aligned}$$ It is intriguing that for the solution (\[solk\]), the variation of the gravity part at the infinity exactly cancels at each order the variation of the matter at infinity, that is $${\delta B}_{\mbox{\tiny{gravity}}}(\infty)=-{\delta B}_{\mbox{\tiny{matter}}}(\infty),$$ and hence we have $B(\infty)=0$. At the horizon, a simple computation yields $$\delta B(r_h)=2\pi\Sigma_{d-2}\delta\Big[\left(1-\xi\,\Phi^2(r_h)\right)r_h^{d-2}\Big]$$ but as stressed before (\[truc\]) the quantity between brackets vanishes. Hence, the boundary term vanishes identically, $B=0$, and the identification of the mass and the entropy through (\[gce\]) yields $${\cal M}=0,\qquad {\cal S}=0.$$ It is then clear that the solutions obtained here have the particularity that their entropy is proportional to the lapse function evaluated at the horizon (\[truc\]), and this quantity, by definition of the horizon, precisely vanishes. Consequently, the integration constant $a$ appearing in the solution can be naively interpreted as a sort of hair since it has not conserved charged associated to it. The temperature of the solution is given by $$\begin{aligned}
T=-\frac{\xi^{\frac{8\xi-1}{4\xi}}}{\pi a (k-1)(4\xi-1)}\left(\xi^{\frac{1-4\xi}{4\xi}}-b\right)^2,\end{aligned}$$ and turns out to be positive provided that the nonminimal coupling parameter $\xi<\frac{1}{4}$ which is precisely the range where the solutions behaves asymptotically AdS. Interestingly enough, the temperature being inversely proportional to the hair $a$, this implies that a high temperature the hair will disappear as it is excepted in order to reproduce the phase diagram of the unconventional superconductors.
Conclusions
===========
Here, we have considered some particular Lovelock gravity theories indexed by an integer $k$ whose coefficients are fixed by requiring the existence of a unique AdS vacuum with a matter source given by a self-interacting nonminimally coupled scalar field. We have shown that for each inequivalent Lovelock gravity theories there exists an appropriate choice for the self-interacting potential that permits to obtain black hole solutions for any arbitrary values of the nonminimal coupling parameter. The form of the potential involves six different terms where the first three terms exactly correspond to the potential that usually arises in different situations involving nonminimally coupled scalar fields. It will be interesting to explore in which context the $k-$depending part of the potential $U_k$ (which concerns the last three terms in (\[potLov\])) may emerge. The thermodynamics analysis of the solutions shows that the mass and the entropy of the solutions both vanish, and hence the integration constant $a$ of the solutions can be viewed as a sort of hair. Note that in the context of pure Lovelock gravity in even dimension, black hole solutions with zero mass and zero entropy have been found in [@Anabalon:2011bw]. In our case, the vanishing of the thermodynamical quantities can be viewed as a consequence of the fact that the entropy of the solutions are proportional to the lapse function evaluated at the horizon (\[truc\]). We have shown that the temperature $T$ goes like $T\propto a^{-1}$ and hence at high temperature the hair should disappear as it is excepted in order to reproduce the phase diagram of the unconventional superconductors. It will be more than interesting to explore more intensively the possible applications of the solutions derived here in the context of holographic superconductors. We end with the fact that for a coupling constant $b=0$ and $\xi=\hat{\xi}_{k,d+1}$ as defined in (\[xi\]), the potential reduces to a mass term and the solution becomes a stealth configuration [@Gaete:2013ixa; @Gaete:2013oda], that is a particular solution of the field equations (\[eqsmotionk\]) where both sides of the equations (gravity and matter) vanish identically $${\cal G}^{(k)}=0=T_{\mu\nu}.$$ In this particular case, the solution involves an additional integration constant which is due to the fact that the equation $T_{\mu\nu}=0$ becomes invariant under the rescaling of the scalar field $\Phi\to A\Phi$, where $A$ is a constant. In order to be self-contained, we present the stealth solution as found in [@Gaete:2013ixa; @Gaete:2013oda] for $b=0$ and $\xi=\hat{\xi}_{k,d+1}$ $$\begin{aligned}
\label{ss}
&&ds^{2}=-\left(r^2-(ar)^{\frac{1-d}{k}}\right)+\frac{dr^2}{\left(r^2-(ar)^{\frac{1-d}{k}}\right)}+{r^{2}}d\vec{x}_{d-2}^2,\nonumber\\
&&\Phi=Ar^{\frac{(k-1)(1-d)}{2k}}.\nonumber\end{aligned}$$ Because of this additional constant $A$, the quantity $(1-\xi\Phi^2)$ evaluated at the horizon may not necessarily be zero, and consequently, the entropy may not be zero. We would like to further explore this issue concerning the thermodynamics properties of the stealth solutions arising in the context of nonminimally coupled scalar field.
We thank Moises Bravo, Julio Oliva, Alfredo Perez and Ricardo Troncoso for useful discussions. FC is partially supported by the Fondecyt grant 11121651 and by the Conicyt grant 79112034. MH is partially supported by grant 1130423 from FONDECYT, by grant ACT 56 from CONICYT and from CONICYT, Departamento de Relaciones Internacionales “Programa Regional MATHAMSUD 13 MATH-05”. The Centro de Estudios Científicos (CECs) is funded by the Chilean Government through the Centers of Excellence Base Financing Program of Conicyt.
[99]{}
C. Lanczos, “A remarkable property of the Riemann.Christoffel tensor in four dimensions,” Ann. Math. [**39**]{}, 842 (1938).
D. Lovelock, “The Einstein tensor and its generalizations,” J. Math. Phys [**12**]{}, 498 (1971).
J. Zanelli, “Lecture notes on Chern-Simons (super-)gravities. Second edition (February 2008),” hep-th/0502193.
M. Banados, C. Teitelboim and J. Zanelli, “Dimensionally continued black holes,” Phys. Rev. D [**49**]{}, 975 (1994) \[gr-qc/9307033\]. R. -G. Cai and K. -S. Soh, “Topological black holes in the dimensionally continued gravity,” Phys. Rev. D [**59**]{}, 044013 (1999) \[gr-qc/9808067\]. J. Crisostomo, R. Troncoso and J. Zanelli, “Black hole scan,” Phys. Rev. D [**62**]{}, 084013 (2000) \[hep-th/0003271\]. R. Aros, R. Troncoso and J. Zanelli, “Black holes with topologically nontrivial AdS asymptotics,” Phys. Rev. D [**63**]{}, 084015 (2001) \[hep-th/0011097\].
D. G. Boulware and S. Deser, “String Generated Gravity Models,” Phys. Rev. Lett. [**55**]{}, 2656 (1985).
R. -G. Cai, "Gauss-Bonnet black holes in AdS spaces,” Phys. Rev. D [**65**]{}, 084014 (2002) \[hep-th/0109133\].
J. D. Bekenstein, “Exact Solutions Of Einstein Conformal Scalar Equations,” Annals Phys. [**82**]{}, 535 (1974). N. M. Bocharova, K. A. Bronnikov and V. N. Melnikov, “An exact solution of the system of Einstein equations and mass-free scalar field,” Vestn. Mosk. Univ. Fiz. Astron. [**6**]{}, 706 (1970) \[Moscow Univ. Phys. Bull. [**25**]{}, 80 (1970)\].
C. Martinez, R. Troncoso and J. Zanelli, “De Sitter black hole with a conformally coupled scalar field in four dimensions,” Phys. Rev. D [**67**]{}, 024008 (2003) \[arXiv:hep-th/0205319\]. C. Martinez, J. P. Staforelli and R. Troncoso, “Charged topological black hole with a conformally coupled scalar field,” Phys. Rev. D [**74**]{}, 044028 (2006) \[arXiv:hep-th/0512022\].
G. T. Horowitz, “Introduction to Holographic Superconductors,” arXiv:1002.1722 \[hep-th\]. S. A. Hartnoll, C. P. Herzog and G. T. Horowitz, “Holographic Superconductors,” JHEP [**0812**]{}, 015 (2008)
C. P. Herzog, “Lectures on Holographic Superfluidity and Superconductivity,” J. Phys. A [**42**]{}, 343001 (2009) \[arXiv:0904.1975 \[hep-th\]\].
M. Bravo-Gaete and M. Hassaine, “Topological black holes for Einstein-Gauss-Bonnet gravity with a nonminimal scalar field,” Phys. Rev. D [**88**]{}, 104011 (2013) \[arXiv:1308.3076 \[hep-th\]\].
M. Bravo-Gaete and M. Hassaine, “Planar AdS black holes in Lovelock gravity with a nonminimal scalar field,” JHEP [**1311**]{}, 177 (2013) \[arXiv:1309.3338 \[hep-th\]\].
E. Ayon-Beato, C. Martinez and J. Zanelli, “Stealth scalar field overflying a (2+1) black hole,” Gen. Rel. Grav. [**38**]{}, 145 (2006) \[hep-th/0403228\].
M. Banados, C. Teitelboim and J. Zanelli, “The Black hole in three-dimensional space-time,” Phys. Rev. Lett. [**69**]{}, 1849 (1992) \[hep-th/9204099\].
E. Ayon-Beato and M. Hassaine, “Exploring AdS waves via nonminimal coupling,” Phys. Rev. D [**73**]{}, 104001 (2006) \[hep-th/0512074\].
M. M. Caldarelli, C. Charmousis and M. Hassaine, “AdS black holes with arbitrary scalar coupling,” JHEP [**1310**]{}, 015 (2013) \[arXiv:1307.5063 \[hep-th\]\].
G. W. Gibbons and S. W. Hawking, “Action Integrals and Partition Functions in Quantum Gravity,” Phys. Rev. D [**15**]{}, 2752 (1977).
A. Anabalon, F. Canfora, A. Giacomini and J. Oliva, “Black holes with gravitational hair in higher dimensions,” Phys. Rev. D [**84**]{}, 084015 (2011) \[arXiv:1108.1476 \[hep-th\]\].
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We unravel the nonequilibrium correlated quantum quench dynamics of an impurity traveling through a harmonically confined Bose-Einstein condensate in one-dimension. For weak repulsive interspecies interactions the impurity oscillates within the bosonic gas. At strong repulsions and depending on its prequench position the impurity moves towards an edge of the bosonic medium and subsequently equilibrates. This equilibration being present independently of the initial velocity, the position and the mass of the impurity is inherently related to the generation of entanglement in the many-body system. Focusing on attractive interactions the impurity performs a damped oscillatory motion within the bosonic bath, a behavior that becomes more evident for stronger attractions. To elucidate our understanding of the dynamics an effective potential picture is constructed. The effective mass of the emergent quasiparticle is measured and found to be generically larger than the bare one, especially for strong attractions. In all cases, a transfer of energy from the impurity to the bosonic medium takes place. Finally, by averaging over a sample of simulated in-situ single-shot images we expose how the single-particle density distributions and the two-body interspecies correlations can be probed.'
author:
- 'S. I. Mistakidis'
- 'F. Grusdt'
- 'G.M. Koutentakis'
- 'P. Schmelcher'
title: |
Dissipative correlated dynamics of a moving impurity\
immersed in a Bose-Einstein Condensate
---
Introduction
============
Ultracold atoms offer an excellent platform to study highly imbalanced multicomponent systems, such as impurities immersed in a Bose-Einstein Condensate (BEC) or in a Fermi sea [@Scazza; @Kohstall; @Schirotzek], due to their exquisite degree of controlability. Indeed the interaction between the impurities and host atoms is tunable via Feshbach resonances [@Chin; @Kohler] and the emergent many-body states can be characterized e.g. with the aid of radiofrequency and Ramsey spectroscopy [@Koschorreck; @Kohstall; @Cetina; @Cetina_interferometry] and in-situ measurements with single-site resolution [@Fukuhara; @Sherson]. Mobile impurities interacting with a surrounding quantum many-body environment form quasiparticles, such as polarons [@Massignan; @Schmidt_rev], originally introduced by Landau [@Landau], when the medium consists of neutral atoms that are not exposed to any external field. The dressing of the impurity atoms from the collective excitations of their host leads to alterations of their properties including their effective mass [@Grusdt_1D; @Ardila_MC], induced interactions [@induced_int_artem; @Mistakidis_Fermi_pol], formation of bound states e.g. bipolarons [@Bipolaron; @Massignan; @Schmidt_rev; @Kevin], as well as dramatic changes during their nonequilibrium dynamics [@dynamics_Artem; @Mistakidis_orth_cat; @Mistakidis_two_imp_ferm; @Mistakidis_eff_mass; @Grusdt_RG; @Shchadilova; @Kamar]. Another important feature is that the impurity subsystem constitutes a few-body setting evincing the inescapable necessity of taking correlation effects into account.
The controllable immersion of single or multiple impurities into a many-body environment have recently led to the experimental observation of Bose [@Jorgensen; @Hu; @Catani1; @Fukuhara; @Yan_bose_polarons] and Fermi [@Scazza; @Koschorreck; @Kohstall] polarons. These experiments triggered an intense theoretical activity in order to describe the polaron characteristics by operating within different frameworks [@Grusdt_approaches; @Rath_approaches]. These include, but are not restricted to, the mean-field approximation [@Astrakharchik; @Cucchietti; @Kalas; @Bruderer1], the Fröhlich model [@Bruderer; @Privitera; @Casteels1; @Casteels2; @Kain; @Tempere_path_int], variational methods [@Mistakidis_eff_mass; @Mistakidis_orth_cat; @Mistakidis_Fermi_pol; @Jorgensen; @Ardila_MC], effective Hamiltonian approaches [@Effect_hamilt; @Effect_hamilt1; @induced_int_artem; @dynamics_Artem] and renormalization group techniques [@Grusdt_RG; @Grusdt_strong_coupl; @Grusdt_approaches]. While the majority of these investigations have been mainly focused on the equilibrium properties of the emergent quasiparticles, the dynamics of impurities is far less explored. In this context notable examples include the observation of self-trapping phenomena [@Cucchietti_self_trap; @Schecter_self_trap], orthogonality catastrophe events [@Mistakidis_orth_cat], generation of dark-bright solitons [@Grusdt_1D; @Mistakidis_two_imp_ferm], transport properties of impurities in optical lattices [@Cai_transp; @Johnson_transp] as well as collisional aspects [@few_body_col; @Lausch_col; @Lausch_col1] of an impurity injected into a gas of Tonks-Girardeau bosons [@Burovski_col; @Lychkovskiy_col; @Lychkovskiy_col1; @Lychkovskiy_col2; @Meinert; @Flutter; @Flutter1; @Gamayun_col].
In this latter context, Bloch-oscillations of impurities in the absence of a lattice [@Meinert], long-lived oscillations [@Flutter; @Flutter1], as well as relaxation of moving impurities [@Burovski_col; @Lychkovskiy_col; @Lychkovskiy_col1; @Lychkovskiy_col2] have been observed in one-dimension. The majority of these investigations have been focusing on a Tonk-Girardeau gas of host atoms in homogeneous space. Yet the collisional dynamics of an impurity particle penetrating a weakly or intermediate repulsively interacting quantum bosonic gas being harmonically trapped can be much more involved. Indeed, in this case the dynamics of the impurity might exhibit a completely different behavior compared to the aforementioned settings for the following reasons. First the finite interactions between the host atoms will give rise to fundamentally different scattering properties between the impurity and the bosonic medium. In this sense it would be particularly interesting to examine the dynamical response of the impurity for different interspecies repulsive and attractive interactions and study how the response regimes depend on the velocity (subsonic, sonic and supersonic) of the impurity. Note also that the initial velocity of the impurity is expected to give rise to a much more involved dynamics compared to the zero velocity case since it will trigger multiple scattering events between the impurity and the BEC. Concordantly, one can e.g. infer whether long-lived oscillations occur [@Flutter; @Flutter1; @Meinert] for subsonic or sonic impurities. Moreover, the presence of the external harmonic trap confines the bosonic bath to a finite spatial region and it would be important to inspect under what conditions the impurity can escape the BEC. Another intriguing prospect is to examine if the impurity forms a strongly correlated (entangled) state with the bosonic bath generating a quasiparticle and unveil the correlation effects of its dynamics [@Mistakidis_orth_cat; @Flutter]. Certainly the properties of the generated quasiparticle such as its effective mass are of significant importance.
To address these inquiries in the present work we investigate the interspecies interaction quench dynamics of a moving impurity initially modeled by a coherent state which penetrates a repulsively interacting and harmonically trapped bosonic gas. To simulate the correlated quantum dynamics of the mixture we invoke the Multi-Layer Multi-Configuration Time-Dependent Hartree Method for Atomic Mixtures (ML-MCTDHX) [@MLX; @MLB1], which is a non-perturbative variational method capturing all interparticle correlations. We find that the dynamics of the impurity exhibits different response regimes depending on the value of the postquench interspecies interaction strength [@Mistakidis_orth_cat; @Mistakidis_two_imp_ferm; @Grusdt_1D]. In particular, for weak postquench interspecies repulsive interactions the subsonic impurity undergoes a dipole motion with a larger frequency for increasing coupling. This is in sharp contrast to the behavior of an initially stationary impurity which, following an interspecies interaction quench, performs a breathing motion inside the BEC as analyzed in Ref. [@Mistakidis_orth_cat]. Strikingly enough, at strong interspecies interactions which exceed the bosonic intraspecies ones the impurity moves towards the edge of the BEC background and thereafter equilibrates around its Thomas-Fermi radius. This behavior of the moving impurity observed for strong repulsions takes place independently of its characteristics e.g. initial velocity, prequench position, trapping frequency and mass and occurs due to the presence of correlations. Importantly, the density of the impurity approaches selectively the smaller distant edge of the Thomas-Fermi radius with respect to its prequench position. This result is altered for a zero velocity impurity whose density at such strong repulsions breaks into two fragments which exhibit a dissipative oscillatory motion at the edges of the Thomas-Fermi profile of the bosonic gas, e.g. see Ref. [@Mistakidis_orth_cat]. In all cases, the bosonic bath shows weak distortions from its initial Thomas-Fermi profile and shallow density dips built upon the bosonic density thus imprinting the motion of the impurity. Indeed, the response of the bosonic medium for a zero velocity impurity undergoes weak amplitude collective breathing oscillations as it has been demonstrated in Ref. [@Mistakidis_orth_cat].
Referring to quenches towards attractive interspecies interactions we showcase that the impurity performs a damped oscillatory motion within the bosonic bath, a behavior that becomes more evident for stronger attractions. As a result the BEC develops a density peak at the location of the impurity. An effective potential picture is also developed for each case in order to provide an intuitive understanding of the resulting dynamics of the impurity [@Mistakidis_orth_cat; @Grusdt_1D]. Note that this effective potential is greatly affected by the motion of the impurity, causing significant deformations in the Thomas-Fermi profile of the bosonic medium, a result that is in contrast to the initially stationary impurity case. Inspecting the individual energy contributions of each species we reveal that the impurity dissipates energy into the bosonic medium [@Mistakidis_orth_cat; @Nielsen; @Lampo], a phenomenon that is more enhanced for stronger interactions of either sign. Employing the Von-Neumann entropy we unveil the presence of interspecies correlations in the course of the evolution. It is worth mentioning that energy exchange processes and the development of correlations between the impurity and the BEC are generic phenomena appearing in impurity physics and their emergence is almost independent of the considered quench scenario. Moreover, we estimate the effective mass of the emergent quasiparticle showcasing that for attractive interactions it is larger than the bare one tending to the latter when approaching the non-interacting limit and becoming slightly larger to its bare value for repulsive interactions. Finally, we provide experimental links of our findings by simulating single-shot absorption measurements. We demonstrate that by averaging a sample of in-situ images the quench-induced correlated dynamics can be adequately reproduced on the single-particle density level [@mistakidis_phase_sep; @Erdmann_phase_sep; @Katsimiga_diss_flow]. Also by utilizing the simulated images on the co-moving frame of the impurity we showcase its imprint on the bosonic gas and discuss how the resulting imaging process probes the two-body interspecies correlations [@Koepsell].
This work is organized as follows. Section \[sec:theory\] presents our setup, the employed many-body treatment and different observables that are used for the characterization of the dynamics. Subsequently, the resulting interspecies interaction quench dynamics towards repulsive \[Sec. \[sec:quench\_repulsive\]\] and attractive \[Sec. \[sec:quench\_attractive\]\] interactions is discussed. The effective mass of the emergent quasiparticle is analyzed in Section \[sec:effective\_mass\], while in Section \[sec:single\_shots\] we present the simulation of in-situ single-shot images. We summarize and discuss future perspectives in Section \[sec:conclusions\]. In Appendix \[sec:single\_shot\_algorithm\] we elaborate on the numerical implementation of the single-shot process and in Appendix \[sec:convergence\_numerics\] we demonstrate the convergence of our many-body calculations.
Theoretical Framework {#sec:theory}
=====================
Setup and Quench Protocol {#sec:hamiltonian}
-------------------------
We consider a highly particle imbalanced bosonic mixture consisting of a single impurity atom $N_I=1$ and $N_B=100$ bosons constituting the majority species (bath). The many-body Hamiltonian of the system consisting of mass balanced species, i.e. $m_A=m_B\equiv m$, which are trapped in the same external one-dimensional harmonic oscillator potential of frequency $\omega_A=\omega_B=\omega$ reads $$\begin{split}
\label{eq:hamiltonian}
H = \sum_{\sigma = B,I} \sum_{i = 1}^{N_{\sigma}}\left[-\frac{\hbar^2}{2m}\left(\frac{\mathrm{d}}{\mathrm{d}x_i^{\sigma}}\right)^2 +
\frac{1}{2} m \omega (x_i^{\sigma})^2\right] \\+ g_{BB} \sum_{i<j} \delta(x_i^B - x_j^B) + g_{BI} \sum_{j = 1}^{N_{B}} \delta(x_j^B-x_1^I).
\end{split}$$ Within the $s$-wave scattering limit which is the dominant interaction process in the ultracold regime both the intra- and interspecies interactions are modeled by a contact potential with effective coupling constants $g_{BB}$ and $g_{BI}$. More specifically, the effective one-dimensional coupling strength [@Olshanii] is given by ${g_{\sigma \sigma'}} =\frac{{2{\hbar ^2}{a^s_{\sigma \sigma'}}}}{{\mu a_ \bot ^2}}{\left( {1 - {\left|{\zeta (1/2)} \right|{a^s_{\sigma \sigma'}}}
/{{\sqrt 2 {a_ \bot }}}} \right)^{ -1}}$, where $\sigma,\sigma'=B, I$ and $\mu=\frac{m}{2}$ refers to the corresponding reduced mass. Here, ${a_\bot } = \sqrt{\hbar /{\mu{\omega _ \bot }}}$ is the transversal length scale characterized by a transversal confinement frequency ${{\omega _ \bot }}$. Also, ${a^s_{\sigma \sigma'}}$ is the three-dimensional $s$-wave scattering length within ($\sigma=\sigma'$) or between ($\sigma \neq \sigma'$) the two species. Experimentally $g_{\sigma\sigma'}$ can be adjusted through ${a^s_{\sigma \sigma'}}$ via Feshbach resonances [@Kohler; @Chin] as well as by manipulating ${{\omega _ \bot }}$ by means of confinement-induced resonances [@Olshanii]. Throughout this work we use a trapping frequency $\omega=0.1\approx2\pi\times 20~ {\rm Hz}$, unless it is stated otherwise. To restrict the dynamics to one dimension one can e.g. assume the experimentally relevant transversal confinement $\omega_{\perp}\approx 2\pi \times 200~ {\rm Hz}$ which is typical for one-dimensional experiments [@Bersano; @Jochim_resolved]. Additionally, the intraspecies interactions are kept fixed to $g_{BB}=1.0$ while the interspecies one, $g_{BI}$, varies upon considering a quench taking values in the repulsive or the attractive regime. Due to the above-mentioned assumptions our system can be well approximated by a binary BEC of $^{87}$Rb atoms prepared in the hyperfine states $\Ket{F=1, m_F=-1}$ and $\Ket{F=2, m_F=1}$ [@Egorov].
For convenience, in the following, the many-body Hamiltonian of Eq. (\[eq:hamiltonian\]) is rescaled in units of $\hbar \omega_{\perp}$. Then the corresponding length, time, and interaction strengths are provided in terms of $\sqrt{\frac{\hbar}{m \omega_{\perp}}}$, $\omega_{\perp}^{-1}$ and $\sqrt{\frac{\hbar^3 \omega_{\perp}}{m}}$ respectively. Also, the spatial extension of our system is limited by employing hard-wall boundary conditions at $x_\pm=\pm80$. The location of the latter does not affect the dynamics since we do not observe any significant density population beyond $x_\pm=\pm40$.
![Sketch of the dynamical quench protocol. An impurity modeled initially by a coherent state travels through a harmonically trapped BEC medium with an initial velocity $u_0$. At $t=0$ an interspecies interaction quench is performed from $g_{BI}=0$ to a finite positive or negative value of $g_{BI}$.[]{data-label="fig:1"}](Fig1-eps-converted-to.pdf){width="45.00000%"}
In order to examine the correlation effects arising due to the injection of the impurity into the bosonic gas we follow the protocol outlined below. The bosonic medium is initially prepared into its ground state for $g_{BB}=1$. The impurity resides in a coherent state characterized by an initial velocity $u_0$ and it is instantaneously localized around $x_0$. In particular, its wavefunction assumes the form $$\Psi^I(x^I;t=0)=\bigg(\frac{m\omega}{\pi \hbar}\bigg)^{1/4} e^{-\frac{m\omega}{2\hbar}(x^I-x_0)^2+ik_0(x^I-x_0)}, \label{coherent_state}$$ where $k_0$ is its initial wavenumber and $u_0=k_0/m$ its initial velocity. To trigger the collisional dynamics between the impurity and the bosonic medium we suddenly switch on at $t=0$ the interspecies interaction strength to repulsive \[Section \[sec:quench\_repulsive\]\] or attractive \[Section \[sec:quench\_attractive\]\] interactions and monitor the time-evolution of the system, see Fig. \[fig:1\] for a sketch. Experimentally our protocol can be realized as follows. A magnetic potential gradient that shifts the minimum of the external trap of the impurity with respect to the bath by $\delta x=u_0/\omega$ is first employed [@Du_gradient]. Then, in order to import an initial momentum onto the impurity this gradient is swiched off letting the impurity to evolve and a Feshbach resonance is utilized to perform the interaction quench when the impurity is at position $x_0$.
Many-Body Treatment {#ML_ansatz}
-------------------
To obtain the ground state and most importantly to explore the quench dynamics of the bosonic mixture we numerically solve the underlying many-body Schr[ö]{}dinger equation by employing the ML-MCTDHX [@MLX; @MLB1] method. It is an ab-initio method which rests on expanding the systems’ many-body wavefunction in terms of a time-dependent and variationally optimized basis enabling us to take into account both the inter- and the intraspecies correlations of the binary system.
To include interspecies correlations into our many-body wavefunction ansatz we introduce $k=1,2,\dots,D$ different species functions for each component, namely $\Psi^{B}_k (\vec x^{B};t)$ and $\Psi^{I}_k (x^{I};t)$ respectively. Here, $\vec x^{B}=\left( x^{B}_1, \dots, x^{B}_{N_{B}} \right)$ and $x^{I}$ denote the spatial coordinates of each species with $N_B$ and $N_I\equiv1$ being the number of bath and impurity atoms, respectively. Then the many-body wavefunction, $\Psi_{MB}$, is expressed in the form of a truncated Schmidt decomposition [@Horodecki] of rank $D$ as follows $$\Psi_{MB}(\vec x^B, x^I;t) = \sum_{k=1}^D \sqrt{ \lambda_k(t) }~ \Psi^B_k (\vec x^B;t) \Psi^I_k ( x^I;t).
\label{Eq:WF}$$ The Schmidt coefficients $\lambda_k(t)$ provide a measure of the entanglement between the two species (see also below) and will be referred to, in the following, as the natural species populations of the $k$-th species function. Indeed, the system is termed entangled [@Roncaglia] or interspecies correlated if at least two different $\lambda_k(t)$ possess a nonzero value since in this case $\Psi_{MB}$ is not a direct product of two states.
Moreover, in order to account for the intraspecies correlations of the system we further express each of the above-mentioned species functions $\Psi^{B}_k (\vec x^{B};t)$ with respect to permanents consisting of $d_{B}$ distinct time-dependent single-particle functions (SPFs) $\varphi_1^B,\dots,\varphi_{d_{B}}^B$. As a result, $\Psi^{B}_k (\vec x^{B};t)$ of the bosonic gas reads $$\begin{split}
&\Psi_k^{B}(\vec x^{B};t) = \sum_{\substack{l_1,\dots,l_{d_{B}} \\ \sum l_i=N_B}} c_{k,(l_1,
\dots,l_{d_{B}})}^B(t) \times \\ & \sum_{i=1}^{N_{B}!} \mathcal{P}_i
\left[ \prod_{j=1}^{l_1} \varphi_1^B(x_j;t) \cdots \prod_{j=1}^{l_{d_{B}}} \varphi_{d_{B}}^B(x_{K(d_{B})+j};t) \right],
\label{Eq:SPF_bath}
\end{split}$$ where $\mathcal{P}$ is the permutation operator which exchanges the particle positions $x_{\nu}^{B}$, $\nu=1,\dots,N_{B}$ within the SPFs. Also, $K(r)\equiv \sum_{\nu=1}^{r-1}l_{\nu}$ with $l_{\nu}$ being the occupation number of the $\nu$th SPF and $r\in\{1,2,\dots,d_{B}\}$, while $c_{k,(l_1,\dots,l_{d_{B}})}^B(t)$ are the time-dependent expansion coefficients of a certain permanent. Correspondingly, the species functions $\Psi^{I}_k (x^{I};t)$ of the impurity are expressed as follows $$\begin{split}
&\Psi_k^{I}( x^{I};t) = \sum_{p=1}^{d_I} c_{k,(l_1=0,\dots,l_p=1,\dots,l_{d_{B}}=0)}^I(t) \varphi_{p}^I(x^I;t),
\label{Eq:SPF_impurity}
\end{split}$$ with $c_{k,(l_1=0,\dots,l_p=1,\dots,l_{d_{B}}=0)}^I(t)$ being the respective time-dependent expansion coefficients on the SPFs $\varphi_1^I,\dots,\varphi_{d_{I}}^I$ of the species $I$.
To solve the underlying Schr[ö]{}dinger equation we need to determine the corresponding ML-MCTDHX equations of motion [@MLX; @Kohler_fabian]. The latter can be accomplished by following e.g. the Dirac-Frenkel variational principle [@Frenkel; @Dirac] for the many-body ansatz given by Eqs. (\[Eq:WF\]), (\[Eq:SPF\_bath\]) and (\[Eq:SPF\_impurity\]). This way we arrive at $D^2$ linear differential equations of motion for the coefficients $\lambda_k(t)$ which are coupled to a set of $D[$ ${N_B+d_B-1}\choose{d_B-1}$+$d_I$\] non-linear integro-differential equations for the species functions and $d_B+d_I$ integro-differential equations for the SPFs. Finally, let us remark that within ML-MCTDHX it is also possible to operate at different orders of approximation. For instance, we can retrieve the corresponding mean-field wavefunction [@Pethick_book] of the bosonic mixture in the limit of $D=d_B=d_I=1$. Namely $$\begin{split}
&\Psi_{MF}(\vec x^{B}, x^{I};t) =
\frac{1}{\sqrt{N_B!}}\prod_{j=1}^{N_B}\varphi_1^B(x_j^{B};t)\varphi_1^I(x^{I};t).
\label{Eq:MF}
\end{split}$$ Recall that in this approximation both intra- and interspecies correlations are neglected since the system is described by one single-particle function for each of the species [@Pethick_book; @mistakidis_phase_sep].
Obervables of interest {#observables}
----------------------
To monitor the dynamics of each species after the quench we employ as a spatially resolved measure the $\sigma$-species one-body reduced density matrix $$\rho_\sigma^{(1)}(x,x^\prime;t)=\langle\Psi_{MB}(t)|\hat{\Psi}^{\sigma\dagger}(x)\hat{\Psi}^\sigma(x^\prime)|\Psi_{MB}(t)\rangle.$$ In this expression, $\hat{\Psi}^{\sigma}(x)$ \[$\hat{\Psi}^{\sigma \dagger}(x)$\] denotes the bosonic field operator that annihilates \[creates\] a $\sigma$-species boson at position $x$, satisfying also the standard bosonic commutation relations [@Pethick_book]. To be more specific, in the following, for simplicity we will resort to the corresponding $\sigma$-species one-body density defined as $\rho_\sigma^{(1)}(x;t)\equiv\rho_\sigma^{(1)}(x,x^\prime=x;t)$. Note also that the eigenfunctions and eigenvalues of $\rho_\sigma^{(1)}(x,x^\prime;t)$ are the so-called natural orbitals $\phi^{\sigma}_i(x;t)$ and natural populations $n^{\sigma}_i(t)$ [@mistakidis_phase_sep; @MLX]. The natural orbitals are related with the SPFs \[see Eqs. (\[Eq:SPF\_bath\]), (\[Eq:SPF\_impurity\])\] via a time-dependent unitary transformation that diagonalizes the matrix $\rho_{\sigma;ij}^{(1)}=\int dx dx' \varphi_i^*(x,t)\varphi_j(x',t)\rho_{\sigma}^{(1)}(x,x';t)$, more details can be found in [@MLX; @MLB1]. Moreover, each bosonic subsystem is said to be intraspecies correlated if more than one natural population possesses a macroscopic occupation, otherwise the corresponding subsystem is fully coherent. Indeed it can be easily shown that when $n_1^{\sigma}(t)=1$, $n_{i>1}^{\sigma}(t)=0$ holds, then $\Psi_{MB}(\vec x^B, x^I;t) \to \Psi_{MF}(\vec x^B, x^I;t)$, see also Eq. (\[Eq:WF\]) and Eq. (\[Eq:MF\]). As a result the degree of the $\sigma$ subsystem intraspecies correlations, and therefore the deviation of the many-body state from the mean-field one, can be theoretically quantified via $\lambda_{\sigma}(t) =1 -n_1^{\sigma}(t)$.
To unveil the degree of interspecies correlations or entanglement during the nonequilibrium dynamics of the bosonic mixture we measure the so-called von-Neumann entropy [@Erdmann_phase_sep; @Catani]. Recall that \[see also the discussion in Section \[ML\_ansatz\] and Eq. \[Eq:WF\])\] the presence of interspecies correlations or entanglement can be inferred by the values of the higher than the first Schmidt coefficients, i.e. $\lambda_k(t)$ with $k>1$. The Schmidt coefficients $\lambda_k(t)$ are the eigenvalues of the species reduced density matrix e.g. $\rho^{N_{B}} (\vec{x}'^{B};t)=\int d x^{I} \Psi^*_{MB}(\vec{x}^{B},
x^{I};t) \Psi_{MB}(\vec{x}'^{B}, x^{I};t)$, with $\vec{x}^{B}=(x^{B}_1, \cdots,
x^{B}_{N_{B-1}})$ \[see also Eq. (\[Eq:WF\])\]. They are chosen to be ordered in a monotonically decreasing manner i.e. $\lambda_k>\lambda_{k+1}$, $k=1,2,\dots,D-1$. As a consequence, the system is species entangled or interspecies correlated when more than a single eigenvalue of $\rho^{N_{B}}$ are macroscopically populated, otherwise it is non-entangled. The von-Neumann entropy [@Erdmann_phase_sep; @Catani; @Mistakidis_two_imp_ferm] reads $$\begin{aligned}
S_{VN}(t)=-\sum\limits_{k=1}^D \lambda_{k}(t)\ln[\lambda_k(t)].\label{eq:entropy} \end{aligned}$$ It can be easily deduced that in the mean-field limit $S_{VN}(t)=0$ since $\lambda_1(t)=1$, while for a many-body state where more than one Schmidt coefficients $\lambda_k$ are populated it holds that $S_{VN}(t)\neq0$.
To track the position of the impurity in the course of the evolution we rely on its spatially averaged mean position. This enables us to assess the trajectory of the impurity given by $$\braket{X_I(t)}=\braket{\Psi_{MB}(t)|\hat{x}^I|\Psi_{MB}(t)},\label{mean_position}$$ where the one-body operator $\hat{x}^I=\int_{\mathcal{D}} dx x^I \hat{\Psi}^{\sigma \dagger}(x) \hat{\Psi}^{\sigma}(x)$ with $\mathcal{D}$ being the spatial extension of the impurity. Experimentally, $\braket{X_I(t)}$ can be measured via spin-resolved single-shot absorption images [@Jochim_resolved]. In particular, each image provides an estimate of the impurity position and $\braket{X_I(t)}$ can be obtained by averaging over a sample of such images.
{width="90.00000%"}
Concluding, we remark that our predictions can be directly tested in state-of-the-art experimental settings [@Yan_bose_polarons; @Jorgensen; @Hohmann_Rb_Cs; @Spethmann_Rb_Cs; @Fukuhara; @Meinert; @Catani; @Mayer_doping]. Indeed, the initial state of the impurity is prepared by utilizing a magnetic gradient, while the employed quench protocol can be realized with the aid of a Feshbach resonance. Moreover, the main quantities used to monitor the dynamics such as the single-particle density and the trajectory of the impurity can be experimentally tracked via in-situ single-shot absorption measurements as we discuss in Section \[sec:single\_shots\].
Quench Dynamics Towards Repulsive Interactions {#sec:quench_repulsive}
==============================================
In the following we investigate the collisional dynamics of a moving single impurity, $N_I=1$, inside a harmonically trapped many-body bosonic bath of $N_B=100$ atoms following an interspecies interaction quench to repulsive interactions. The many-body bath is initialized in its ground state with $g_{BB}=1$ exhibiting a Thomas-Fermi profile of radius $R_{TF}\approx25$. On the other hand, the impurity is non-interacting with the bath, $g_{BI}=0$, and resides in a coherent state \[see Eq. (\[coherent\_state\])\]. Its initial velocity is $u_0=-u_c/2$, with $u_c$ being the speed of sound of the BEC background, and therefore it is subsonic. The dynamics is triggered by performing an interspecies interaction quench to positive $g_{BI}$ values at $t=0$ where the impurity is located at position $x_0=0$.
Single-Particle Density Evolution and Effective Potential {#sec:density_repulsive}
---------------------------------------------------------
Let us first inspect the dynamical response of the system upon considering an interspecies quench from $g_{BI}=0$ towards a finite positively valued $g_{BI}$. To achieve a spatially resolved description of the dynamics we resort to the time-evolution of the $\sigma$-species single-particle density $\rho^{(1)}_{\sigma}(x;t)$, see Fig. \[fig:2\]. For a weak interspecies interaction quench, such that $g_{BI}<g_{BB}$, the impurity \[see Fig. \[fig:2\] (b)\] performs almost perfect dipole oscillations of frequency $\omega_R\approx 0.07$ inside the bosonic medium. The deviation from perfect dipole oscillations, caused by the finite value of $g_{BI}$, is manifested in the shape of $\rho^{(1)}_{I}(x;t)$ since it becomes more wide when located close to the edges of $\rho^{(1)}_{B}(x;t)$ than the trap center, see Figs. \[fig:3\] (a)-(c). The bosonic bath remains unperturbed to a large extent \[Fig. \[fig:2\] (a)\] throughout the time-evolution, exhibiting small distortions at the core of its Thomas-Fermi cloud due to its interaction with the impurity. These distortions are directly evident in the instantaneous density profiles of $\rho^{(1)}_{B}(x;t)$ shown in Figs. \[fig:3\] (a)-(c). It is also important to note here that the total external potential of the impurity can be well approximated by the time-averaged effective potential created by the external harmonic oscillator and the density of the bosonic bath. Such an effective potential picture affects the dynamics of the impurity in an essential manner only in the presence of an external trapping since in the homogeneous case the density of the bath is constant in space. More specifically, this effective potential [@Mistakidis_orth_cat; @Hannes] reads $$\bar{V}_I^{eff}(x)=\frac{1}{2}m\omega^2x^2+\frac{g_{BI}}{T}\int_0^T dt \rho_B^{(1)}(x;t), \label{effective_potential_repulsive}$$ where $T$ denotes the considered evolution time. Here, we have used $T=150$. Let us also mention that this averaging process aims to eliminate the observed distortions on the instantanteous bosonic density $\rho^{(1)}_B(x;t)$, which is achieved in our case for $T>100$. These distortions are, of course, caused by the impurity motion and are mainly imprinted as sound waves, see e.g. Figs. \[fig:2\] (c) and (e). This $\bar{V}_I^{eff}(x)$ at $g_{BI}=0.5$ corresponds to a modified harmonic oscillator potential and it is depicted in Fig. \[fig:4\] (a) together with its first few single-particle eigenstates. In this case the impurity undergoes a dipole motion within $\bar{V}_I^{eff}(x)$ and predominantly resides in its energetically lowest-lying state, $n=1$.
Performing a quench to stronger interspecies interaction strengths, e.g. $g_{BI}=1$, the collisional dynamics between the impurity and the bosonic bath \[Figs. \[fig:2\] (c), (d)\] is drastically altered compared to the above-mentioned weakly interacting case \[Figs. \[fig:2\] (a)-(b)\]. In particular, the prominent interspecies interactions greatly affect the motion of the impurity after the quench, see Fig. \[fig:2\] (d). Inspecting $\rho_{I}^{(1)}(x;t)$ we observe that it undergoes an irregular oscillatory motion within the BEC. More specifically, it initially travels to the left edge of the bosonic bath where at $t\approx33$ it is reflected back towards the right edge possessing also a larger width compared to its initial one \[see also Figs. \[fig:3\] (d), (e)\]. This change of the width of $\rho_{I}^{(1)}(x;t)$ is a direct effect of the interaction between the bosonic medium and the impurity and it becomes more pronounced when the impurity reaches the right edge of the bosonic bath and shows a multihump structure, see Fig. \[fig:2\] (d) around $t\approx105$ and also Fig. \[fig:3\] (f). This multihump structure of $\rho_{I}^{(1)}(x;t)$ suggests that the impurity populates a superposition of higher-lying excited states of the corresponding effective potential given by Eq. (\[effective\_potential\_repulsive\]) as we shall discuss in more detail below. The bosonic medium becomes also perturbed due to its interaction with the impurity. As a result slight deviations from the initial Thomas-Fermi profile occur \[Figs. \[fig:3\] (d)-(f)\] while $\rho_{B}^{(1)}(x;t)$ develops shallow density dips at the instantaneous location of the density hump of $\rho_{I}^{(1)}(x;t)$, thus imprinting the motion of the impurity.
![Snapshots of the $\sigma=B,I$ species one-body density at distinct time-instants (see legends) for a varying interspecies interaction strength $g_{BI}$ (see legends). The remaining system parameters are the same as in Fig. \[fig:2\].[]{data-label="fig:3"}](Fig3-eps-converted-to.pdf){width="45.00000%"}
{width="90.00000%"}
Entering to stronger postquench interspecies interactions, e.g. $g_{BI}=2$, which satisfy $g_{BI}>g_{BB}$ reveals a completely different dynamical response of both species, see Figs. \[fig:2\] (e), (f). Remarkably enough the impurity travels towards the left edge of the bosonic bath where it remains locked while exhibiting an oscillatory behavior of negligible amplitude for $t>110$. The fact that the impurity escapes from the bosonic gas and undergoes damped oscillations around its Thomas-Fermi radius is reminiscent of an orthogonality catastrophe phenomenon [@Mistakidis_orth_cat]. Indeed, it has been showcased that upon considering an interspecies interaction quench of a zero velocity impurity atom immersed in a bosonic gas to strong repulsions, the structure factor [@Cetina_interferometry] of the quasiparticle becomes zero and simultaneously the density of the impurity resides at the edges of the BEC [@Mistakidis_orth_cat]. However, an important difference between a zero velocity and a moving impurity is that in the former case after the quench $\rho_{I}^{(1)}(x;t)$ breaks into two fragments, a behavior that does not occur herein. Note also that due to its interaction with the BEC background the impurity emits some small portions of density when it is located well inside $\rho_{B}^{(1)}(x;t)$, see for instance the dashed white rectangles in Fig. \[fig:2\] (f) and the small amplitude density hump of $\rho_{I}^{(1)}(x;t)$ in Fig. \[fig:3\] (g). When the impurity reaches the left edge of the Thomas-Fermi cloud its $\rho_{I}^{(1)}(x;t)$ develops a multihump structure which indicates that it resides in a superposition of several energetically higher-lying excited states of the effective external potential. In this case of strong $g_{BI}$ the effective potential introduced in Eq. (\[effective\_potential\_repulsive\]) is an asymmetric double-well potential which exhibits an energy offset between the left and right wells. We remark here that the effective potential approximation provides for these strong interactions only a very approximate but rather intuitive picture of the impurity dynamics. However, $\rho_{I}^{(1)}(x;t)$ shown in Fig. \[fig:2\] (f) and Figs. \[fig:3\] (g)-(i) has been obtained within the full many-body approach described in Section \[ML\_ansatz\]. The resulting $\bar{V}_{I}^{eff}(x)$ and its first few single-particle eigenstates are illustrated in Fig. \[fig:4\] (b). In terms of this effective picture for $t>100$, the impurity is trapped in the left well of $\bar{V}_{I}^{eff}(x)$ where it predominantly occupies a superposition of the $n=1$, $n=3$ and $n=5$ eigenstates. Furthermore, the motion of the impurity leaves its fingerprints also in the BEC background which as a result becomes perturbed. Indeed, the Thomas-Fermi cloud is disturbed as it can be seen from the corresponding instantaneous density profiles presented in Figs. \[fig:3\] (g)-(i) and in particular when $\rho_{I}^{(1)}(x;t)$ is well inside $\rho_{B}^{(1)}(x;t)$ \[e.g. see Fig. \[fig:3\] (g)\] the latter develops density dips at the instantaneous location of $\rho_{I}^{(1)}(x;t)$. Note also that the small distortions appearing in the spatial region of the barrier of the effective double-well potential depicted in Fig. \[fig:4\] (b) are caused by the existence of beyond mean-field corrections at the core of $\rho_{B}^{(1)}(x;t)$.
Mean Position of the Impurity {#sec:position_repulsive}
-----------------------------
To examine the dependence of the dynamical response of the impurity on the distinct system parameters we next monitor its motion by calculating its mean position $\braket{X_I(t)}$ \[see also Eq. (\[mean\_position\])\] during the dynamics. We first investigate the motion of a subsonic impurity with initial velocity $u_0=-u_c/2$ and the quench is performed when it is located at $x_0=0$. Figure \[fig:5\] (a) shows $\braket{X_I(t)}$ for different postquench interspecies interaction strengths. In line with our discussion in Section \[sec:density\_repulsive\], we observe that for $g_{BI}<g_{BB}$ the impurity oscillates within the BEC background but with an increasing period for a larger $g_{BI}$, e.g. compare $\braket{X_I(t)}$ between $g_{BI}=0.1$ and $g_{BI}=0.5$. However, for quench amplitudes characterized by $g_{BI}>g_{BB}$ the impurity moves towards the left edge of the bosonic bath and subsequently equilibrates \[Fig. \[fig:2\] (f)\], e.g. see $\braket{X_I(t)}$ at $g_{BI}=1.5$ for $t>50$.
{width="80.00000%"}
A natural question that arises is whether this latter behavior of the impurity, namely equilibration at the edge of the BEC background, is an effect of the inclusion of the correlations into the dynamics. To address this question we next present $\braket{X_I(t)}$ at $g_{BI}=2$ in Fig. \[fig:5\] (b) within the fully many-body approach \[Eq. (\[Eq:WF\])\] and the mean-field approximation \[Eq. (\[Eq:MF\])\]. Evidently, $\braket{X_I(t)}$ within the mean-field approximation exhibits an oscillatory behavior for long evolution times meaning that the impurity remains independently of $g_{BI}$ well inside the BEC background. This sharp contrast of the behavior of $\braket{X_I(t)}$ between the many-body and the mean-field treatments occurring at large $g_{BI}$ evinces that the observed equilibration of the impurity at the edge of the bosonic bath is a direct effect of the presence of correlations. Moreover, this behavior of $\braket{X_I(t)}$ taking place at strong $g_{BI}$ occurs even for a decreasing trapping frequency e.g. $\omega=0.05$ (at larger evolution times) as shown in Fig. \[fig:5\] (c). The fact that the phenomenon occurs for larger evolution times can be attributed to the fact that for a decreasing $\omega$, and thus tending to the untrapped case, the corresponding Thomas-Fermi radius of the BEC becomes larger and therefore the impurity needs to travel a longer distance until it reaches the edge of the BEC cloud. This behavior is a direct manifestation of the effect of the strength of the external trapping on the equilibration time of the moving impurity at strong interspecies repulsions. Let us also remark in passing that at short evolution times ($0<t<40$ in Fig. \[fig:5\] (c)) the trajectory, and thus also the corresponding velocity, of the impurity is independent of the harmonic oscillator frequency. This is an expected result since for these short evolution times the impurity lies well inside the bosonic gas and thus experiences an almost homogeneous environment.
Interestingly, the equilibration of the impurity occurring at strong repulsions persists also for a heavier impurity atom as depicted in Fig. \[fig:5\] (d). Here we consider a $^{87}$Rb bosonic gas and a single $^{133}$Cs impurity at the hyperfine states $\Ket{F=1, m_F=0}$ and $\Ket{F=3, m_F=2}$ respectively both trapped in an external harmonic oscillator of the same frequency [@Hohmann_Rb_Cs; @Spethmann_Rb_Cs]. Also, the initial momentum of the subsonic impurity is kept fixed in both cases. As it can be seen, $\braket{X_I(t)}$ reaches the edge of the BEC at almost the same time scales in both mixtures but the $^{133}$Cs atom remains inside the Thomas-Fermi radius to a larger extent than the $^{87}$Rb one. This is an expected result since the velocity $u_0$ of the $^{133}$Cs impurity is smaller than the corresponding $^{87}$Rb one because $m_{Cs}>m_{Rb}$.
Focusing on the strongly interacting regime, e.g. $g_{BI}=2$, we next inspect $\braket{X_I(t)}$ by considering the interaction quench at different locations $x_0$ of the impuritys’ motion with respect to the trap center, see Fig. \[fig:5\] (e). As it can be deduced, $\braket{X_I(t)}$ exhibits a saturated behavior independently of $x_0$. Notice also here that for $x_0>0$ the impurity is repelled by the bosonic cloud to the opposite direction of its motion and finally reaches the right edge of the BEC background, e.g. see $\braket{X_I(t)}$ for $x_0=10$. This is, of course, a manifestation of the exerted force by the BEC on the impurity. Accordingly, we can infer that the density of the impurity approaches selectively the smaller distant edge of the bosonic bath in terms of its prequench position. Note that this result is in contrast to the behavior of a zero velocity impurity whose density at such strong repulsions breaks into two fragments which exhibit a dissipative oscillatory motion around the edges of the bosonic gas [@Mistakidis_orth_cat].
Next, we examine the dependence of the motion of the impurity in the strongly interacting regime, $g_{BI}=2$, on its initial velocity $u_0$ when the quench is performed at position $x_0=0$. Figure \[fig:5\] (f) illustrates $\braket{X_I(t)}$ for initial velocities $u_0\approx-(1/5)u_c$ (subsonic), $-(1/2)u_c$ (subsonic) and $-u_c$ (sonic) with $u_c\approx1.74$ being the speed of sound of the BEC background. We deduce that for an increasing initial velocity, such that $u_0\to -u_c$, the impurity reaches faster the left edge of the bosonic bath where it subsequently equilibrates. Note that this behavior of $\braket{X_I(t)}$ for $u_0\approx -u_c$ is in contrast to the long-lived oscillations reported in homogeneous settings [@Flutter; @Flutter1] but for supersonic ($u_0\gg u_c$) impurities. However for $u_0\ll u_c$, e.g. $u_0=(1/5)u_c$, the impurity performs oscillations through the BEC of a much slower decaying amplitude when compared to the previous cases. This behavior is caused due to its small velocity which generates a lesser amount of excitations to the BEC as compared to large $u_0$.
Degree of Entanglement {#sec:entanglement_repulsive}
----------------------
To quantify the correlated nature of the collisional dynamics between the impurity and the BEC we next measure the degree of entanglement or interspecies correlations by employing the Von-Neumann entropy $S_{VN}(t)$ \[Eq. (\[eq:entropy\])\]. Recall that $S_{VN}(t)\neq0$ signifies the presence of interspecies entanglement, otherwise the system is non-entangled [@mistakidis_phase_sep].
The dynamics of $S_{VN}(t)$ is shown in Fig. \[fig:6\] (a) following an interspecies interaction quench for different values of $g_{BI}$. As it can be seen $S_{VN}(t=0)=0$ since for the initial state of the system $g_{BI}=0$. However, after the quench $S_{VN}(t)$ acquires finite values thus indicating the presence of interspecies correlations. For weak postquench interactions, e.g. $g_{BI}=0.3$, there is only a small amount of interspecies correlations since $S_{VN}(t)$ is suppressed taking very small values. Recall that in this case the impurity performs dipole oscillations within the BEC, see also Fig. \[fig:2\] (b). On the contrary, for stronger postquench interactions such as $g_{BI}=1.5$ $S_{VN}(t)$ increases rapidly at the initial stages of the dynamics where the impurity resides within the BEC while for later times at which the impurity equilibrates at the edge of the bosonic gas $S_{VN}(t)$ tends to saturate to a certain finite value. This behavior of $S_{VN}(t)$ indicates that the underlying many-body state \[Eq. (\[Eq:WF\])\] is strongly entangled. It is worth mentioning that for strong repulsions where the impurity essentially escapes from the BEC, e.g. at $g_{BI}=2$ for $t>80$ \[Fig. \[fig:2\] (f)\], suggesting a break down of the quasiparticle picture $S_{VN}(t)$ acquires an almost constant value \[Fig. \[fig:6\] (a)\]. Also stronger postquench interspecies interactions, $g_{BI}$, result in larger values of $S_{VN}(t)$.
Interspecies Energy Transfer {#sec:energy_repulsive}
----------------------------
To further understand the nonequilibrium dynamics of the impurity immersed in the BEC background for a different postquench $g_{BI}$, below we analyze the distinct energy contributions of the bosonic mixture [@Mistakidis_orth_cat; @Mistakidis_two_imp_ferm; @Nielsen]. The normalized energy of the BEC corresponds to $E_B(t)=\braket{\Psi(t)|\hat{T}_B+\hat{V}_B(x)+\hat{H}_{BB}|\Psi(t)}-\braket{\Psi(0)|\hat{T}_B+\hat{V}_B(x)+\hat{H}_{BB}|\Psi(0)}$, and for the impurity is $E_I(t)=\braket{\Psi(t)|\hat{T}_I+\hat{V}_I(x)|\Psi(t)}$. Moreover, the interspecies interaction energy is $E_{BI}(t)=\braket{\Psi(t)|\hat{H}_{BI}|\Psi(t)}$. In this notation, $\hat{T}_{\sigma}=-\int dx \hat{\Psi}^{\sigma \dagger}(x)\frac{\hbar^2}{2 m} (\frac{d}{dx^{\sigma}})^2 \hat{\Psi}^{\sigma}(x)$ and $\hat{V}_{\sigma}(x)=\int dx \hat{\Psi}^{\sigma \dagger}(x)\frac{1}{2} m \omega^2 x_{\sigma}^2 \hat{\Psi}^{\sigma}(x)$ denote the kinetic and the potential energy operators of the $\sigma=B,I$ species respectively. Also, $\hat{H}_{BB}=g_{BB} \int dx~\hat{\Psi}^{B \dagger}(x) \hat{\Psi}^{B \dagger}(x) \hat{\Psi}^{B} (x)\hat{\Psi}^{B}
(x)$ and $\hat{H}_{BI}=g_{BI}\int dx~\hat{\Psi}^{B \dagger}(x) \hat{\Psi}^{I \dagger}(x) \hat{\Psi}^{I}(x)\hat{\Psi}^{B}(x)$ refer to the operators of the intra- and interspecies interactions with $\hat{\Psi}^{\sigma} (x)$ being the $\sigma$-species field operator.
{width="100.00000%"}
The dynamics of each of the above-described energy contributions is presented in Figs. \[fig:6\] (b), (c) upon considering a quench towards weak and strong interspecies repulsive interactions respectively. Focusing on weak postquench interactions \[Fig. \[fig:6\] (b)\], e.g. $g_{BI}=0.5$, the energy of the impurity $E_I(t)$ and the interspecies interaction energy $E_{BI}(t)$ exhibit an oscillatory behavior. The energy of the bath, $E_{B}(t)$, slightly increases and $E_{B}(t)<E_{I}(t)<E_{BI}(t)$ holds in the course of the evolution. In particular, $E_I(t)$ is minimized at the time-intervals where the impurity is close to the trap center and it is maximized when the impurity travels towards the edges of the BEC. Accordingly, $E_{BI}(t)$ oscillates out-of-phase with $E_I(t)$ since the interspecies interaction is stronger when the impurity is close to the trap center (where $E_I$ is small) and vice versa. Moreover, the fact that $E_{B}(t)$ increases to a minor extent during the dynamics suggests that the impurity devolves a small amount of energy to the BEC. This process is captured by the very weakly decaying amplitude of oscillations of $E_I(t)$.
Referring to strong interspecies interactions, e.g. $g_{BI}=2$ shown in Fig. \[fig:6\] (c) the dynamical behavior of all energy contributions is drastically altered when compared to their weakly interacting counterparts \[compare Figs. \[fig:6\] (b) and (c)\]. At $0<t<40$, $E_{BI}(t)$ reduces while $E_{B}(t)$ and $E_{I}$ increase. Indeed, within this time interval the impurity density moves to the left edge of the BEC \[Fig. \[fig:2\] (b)\] with a large kinetic energy and as a result dissipates energy to the latter. For $40<t<100$, $E_{I}(t)$ and $E_{BI}(t)$ oscillate out-of-phase with respect to one another and in particular $E_{I}(t)$ overall increases while performing small amplitude oscillations. Note that in this time interval $\rho^{(1)}_{I}(x;t)$ oscillates around the left boundary of $\rho^{(1)}_{B}(x;t)$ and still weakly interacts with the BEC. Simultaneously, $E_{B}(t)$ increases when the impurity resides to a large extent in the bosonic bath and decreases when $\rho^{(1)}_{I}(x;t)$ is located at the left edge. As a result the impurity transfers a part of its energy to the bosonic gas. Similar energy exchange processes between the impurity and the host atoms have already been observed e.g. in Refs. [@Mistakidis_orth_cat; @Nielsen]. Deeper in the evolution, $t>100$, all energy components acquire an almost constant value with $E_{B}(t)< E_{BI}(t)<E_{I}(t)$. Recall that for $t>100$ $\rho^{(1)}_{I}(x;t)$ resides at the left edge of $\rho^{(1)}_{B}(x;t)$ and therefore the interaction between the two species is drastically reduced.
Quench Dynamics Towards Attractive Interactions {#sec:quench_attractive}
===============================================
Next, we explore the out-of-equilibrium dynamics of a subsonically moving impurity ($N_I=1$) immersed within a harmonically trapped BEC ($N_B=100$) upon considering an interspecies interaction quench towards attractive interactions. As in the previous Section \[sec:quench\_repulsive\], the BEC is initially prepared into its ground state with $g_{BB}=1$ having a Thomas-Fermi profile of radius $R_{TF}\approx25$. The subsonic impurity is initially modeled as a coherent state \[Eq. (\[coherent\_state\])\] with a velocity $u_0=-u_c/2$ and the impurity-BEC interaction is zero ($g_{BI}=0$) at $t=0$. To induce the dynamics we perform at $t=0$ a quench to negative $g_{BI}$ interaction strengths when the impurity is at $x_0=0$.
Density Evolution and Effective Picture {#sec:density_attractive}
---------------------------------------
To unveil the dynamical response of the system after an interspecies interaction quench to attractive coupling strengths we resort to the time-evolution of the $\sigma$-species single-particle density $\rho_{\sigma}^{(1)}(x;t)$. The emergent evolution of $\rho_{\sigma}^{(1)}(x;t)$ is illustrated in Fig. \[fig:7\] for distinct postquench interspecies interactions ranging from weak to strong negative values. Focusing on weak postquench negative interactions, $g_{BI}=-0.2$ \[Figs. \[fig:7\] (a), (b)\], we observe that due to the small $g_{BI}$ $\rho_{I}^{(1)}(x;t)$ oscillates within $\rho_{B}^{(1)}(x;t)$ with an almost fixed amplitude, see also Figs. \[fig:8\] (a)-(c), and frequency $\omega_{osc}\approx 0.11$. As a result of the motion of the impurity and the weak $g_{BI}$ the Thomas-Fermi cloud is slightly distorted and in particular faint density humps built upon $\rho_{B}^{(1)}(x;t)$ at the location of $\rho_{I}^{(1)}(x;t)$ \[hardly visible in Fig. \[fig:7\] (a)\].
{width="90.00000%"}
![Density profiles of the $\sigma=B,I$ species at various time-instants (see legends) of the evolution for distinct interspecies interaction strengths $g_{BI}$ (see legends). The remaining system parameters are the same as in Fig. \[fig:8\].[]{data-label="fig:8"}](Fig8-eps-converted-to.pdf){width="45.00000%"}
For stronger negative interspecies interactions, e.g. $g_{BI}=-1$, $\rho_{I}^{(1)}(x;t)$ undergoes an oscillatory motion of decaying amplitude within the BEC background and a larger frequency $\omega_{osc}\approx0.14$ compared to the $g_{BI}=-0.2$, see Fig. \[fig:7\] (d). Due to the finite $g_{BI}$ the motion of $\rho_I^{(1)}(x;t)$ in turn results in the development of a density hump on $\rho_B^{(1)}(x;t)$ at the instantaneous position of the impurity, see Fig. \[fig:7\] (c) and Figs. \[fig:8\] (d)-(f). This decaying amplitude oscillatory behavior of the impurity persists and becomes more evident for stronger attractive $g_{BI}$, compare Figs. \[fig:7\] (d) and (f). Note also the additional modulations of the density peak of the impurity caused by its collisions with the excitations of the bosonic gas. The above-mentioned behavior of $\rho_{I}^{(1)}(x;t)$ can be directly captured by inspecting the dynamics of the mean position of the impurity $\braket{X_I(t)}$ for varying $g_{BI}$ shown in Fig. \[fig:9\]. Indeed, we can deduce that $\braket{X_I(t)}$ oscillates with a decaying amplitude in time which is more pronounced deeper in the attractive regime of interactions, compare $\braket{X_I(t)}$ for $g_{BI}=-0.5$ and $g_{BI}=-2$. This attenuation of the oscillation amplitude of $\braket{X_I(t)}$ is a direct effect of the presence of interspecies interactions and the underlying energy transfer process from the impurity to the bath, see also the discussion below and Refs. [@Mistakidis_eff_mass; @Nielsen]. Also, $\rho_{I}^{(1)}(x;t)$ having a sech-like shape tends to be more localized for larger negative values of $g_{BI}$ \[see Figs. \[fig:8\] (g)-(i)\], a result that holds equally for the corresponding density hump building upon $\rho_{B}^{(1)}(x;t)$ \[Fig. \[fig:7\] (e)\]. The latter density hump being directly discernible in $\rho_{B}^{(1)}(x;t)$ is essentially an imprint of the impurity motion inside the BEC. Another interesting observation here is that at large ${\left| g_{BI} \right|}$ the system is strongly correlated and the BEC background is highly excited, as can be inferred from the emission of a large amount of sound waves, see for instance the dashed black circle in Fig. \[fig:7\] (e) and the discussion below. Such a sound wave emission has been extensively reported during the motion of a Gaussian barrier inside a BEC within \[e.g. see Refs. [@sound_wave; @sound_wave1]\] and beyond [@Katsimiga_diss_flow] the mean-field approximation. In the present investigation the impurity plays, of course, the role of the Gaussian barrier. According to these studies the motion of the impurity locally perturbs the initial zero phase of the BEC leading to the formation of small amplitude phase disturbances that lead to sound waves. A similar mechanism takes place also herein where we can identify the existence of sound waves by measuring their velocity at the center of the trap being larger than $0.95 u_c$.
![Position of the impurity in the course of the dynamics for different attractive interspecies interaction strengths (see legend). The remaining system parameters are kept fixed and are the same as in Fig. \[fig:8\].[]{data-label="fig:9"}](Fig9-eps-converted-to.pdf){width="45.00000%"}
{width="90.00000%"}
The above-described dynamical response of the impurity and the BEC taking place at these negative interspecies interactions can be qualitatively understood by invoking an effective potential picture [@Mistakidis_orth_cat; @Hannes]. Indeed, the effective potential acting on the BEC consists of the external harmonic oscillator and the instantaneous density of the impurity namely $$V_B^{eff}(x,t)= V(x)-{\left| g_{BI} \right|} \rho_I^{(1)}(x;t). \label{effective_pot_bath}$$ A schematic illustration of $V_B^{eff}(x,t)$ at $g_{BI}=-2$ is shown in Fig. \[fig:10\] (a) at two distinct time-instants of the evolution. We deduce that $V_B^{eff}(x,t)$ corresponds to an harmonic oscillator like potential possessing also a dip, at the momentary position of the impurity, which is characterized by negative energies. This latter attractive part of $ V_B^{eff}(x,t)$ is responsible for the observed density hump appearing in the dynamics of $\rho_B^{(1)}(x;t)$. Accordingly, the effective potential of the impurity is created by the external harmonic oscillator $V(x)$ and the single-particle density of the BEC. We remark that since $\rho_B^{(1)}(x;t)$ is greatly affected by the impurity motion, a time-average effective potential cannot adequately capture the dynamics of the impurity. In particular, the effective potential of the impurity reads $$V_I^{eff}(x,t)=V(x)-{\left| g_{BI} \right|} \rho_B^{(1)}(x;t). \label{effective_pot_impurity}$$ Figure \[fig:10\] (b) presents $V_I^{eff}(x,t)$ calculated at $g_{BI}=-2$ for two different times in the course of the dynamics. As shown, $V_I^{eff}(x,t)$ is a deformed attractive harmonic oscillator potential having an additional dip around $x\approx0$ due to the presence of the density hump building upon $\rho_B^{(1)}(x;t)$. This attractiveness of $V_I^{eff}(x,t)$ causes the localized sech-like shape of $\rho_I^{(1)}(x;t)$ located around the aforementioned additional potential dip. Most importantly, the observed distinct features of the impurity occurring for stronger attractive interactions can be explained via the behavior of the constructed $V_I^{eff}(x,t)$. Indeed, for increasing ${\left| g_{BI} \right|}$ the effective frequency of $V_I^{eff}(x,t)$ is larger and $V_I^{eff}(x,t)$ becomes more attractive. The former property of $V_I^{eff}(x,t)$ accounts for the decreasing oscillation period of $\braket{X_I(t)}$ for larger ${\left| g_{BI} \right|}$. Additionally, the increasing attractiveness of $V_I^{eff}(x,t)$ is responsible for the reduced width of $\rho_I^{(1)}(x;t)$ for a larger ${\left| g_{BI} \right|}$ and thus its increasing localization tendency, e.g. compare Figs. \[fig:7\] (d) and (f).
To showcase the interconnection between $V^{eff}_I(x;t)$ and $V^{eff}_B(x;t)$ we approximatively decompose the one-body density of the BEC at time $t_0$ according to $$\rho^{(1)}_B(x;t_0) \approx (1-A) \rho^{(1)}_B(x;0) + A N_B \rho^{(1)}_I(x;t_0).
\label{eff_expansion}$$ The first term in Eq. (\[eff\_expansion\]) corresponds to the unperturbed BEC in the absence of the impurity. The second term provides a correction to $\rho^{(1)}_B(x;t_0)$ stemming from the interspecies interaction according to $V_B^{eff}(x,t)$ \[see Eq. (\[effective\_pot\_bath\]) and also Fig. \[fig:10\] (c)\]. Also, $A$ is a real valued parameter bounded in the interval $[0,1]$. In the sense of Eq. (\[eff\_expansion\]) positive values of $A$ encode the back-action of $\rho^{(1)}_I(x;t_0)$ on the density of the BEC. The latter, in turn, forms the effective potential $V^{eff}_I(x;t)$ as dictated by Eq. (\[effective\_pot\_impurity\]) which accordingly determines $\rho^{(1)}_I(x;t_0)$. Indeed, Fig. \[fig:7\] (c)-(d) and \[fig:8\] (g)-(h) indicate that this correction proportional to $A$ is sizable especially in the case of strong attractive interactions, e.g. $g_{BI}=-2$. This is in sharp contrast to an impurity repulsively interacting with a BEC where no sizable corrections of this nature are found, see also Section \[sec:density\_repulsive\] and Ref. [@Mistakidis_orth_cat].
Entanglement Dynamics {#sec:entanglement_attractive}
---------------------
To reveal the correlated character and in particular the degree of entanglement of the quench-induced dynamics we calculate the corresponding Von-Neumann entropy $S_{VN}(t)$ \[see Eq. (\[eq:entropy\])\]. The time-evolution of $S_{VN}(t)$ is demonstrated in Fig. \[fig:11\] (a) for different postquench interaction strengths $g_{BI}$. As in Section \[sec:entanglement\_repulsive\], we again observe that $S_{VN}(t=0)=0$ holds for all cases due to the fact that initially $g_{BI}=0$. However for $t>0$ $S_{VN}(t)\neq 0$ testifying that the many-body state \[Eq. (\[Eq:WF\])\] is entangled. At the initial stages of the dynamics, e.g. $0<t<5$ for $g_{BI}=-1.2$, $S_{VN}(t)$ exhibits its larger growth rate and subsequently shows an overall decreasing behavior tending to approach a constant value for large evolution times $t>120$. The fact that $S_{VN}(t)$ exhibits the aforementioned decreasing trend for $t>8$ can be explained via inspecting the dynamics of the underlying Schmidt coefficients $\lambda_k(t)$ of the many-body wavefunction \[Eq. (\[Eq:WF\])\] (not shown here for brevity). At the early stages of the dynamics $\lambda_1(t)$ drops from unity very quickly, while $\lambda_2(t)$, $\lambda_3(t)$ (with $\lambda_2(t)>\lambda_3(t)$) acquire finite values which become maximal at the time-instant where the impurity emits a small portion of its density, see the dashed rectangle in Fig. \[fig:7\] (f) at $t\approx4$. Thereafter, the central density hump of the impurity is predominantly described by ${\left| \Psi_1^I(x;t) \right|}^2$ while the emitted density portion by a superposition of ${\left| \Psi_2^I(x;t) \right|}^2$ and ${\left| \Psi_3^I(x;t) \right|}^2$. For later evolution times the emitted density portion re-collides with the central hump \[see the dashed rectangle in Fig. \[fig:7\] (f) around $t\approx 20$\]. Simultaneously $\lambda_1(t)$ tends to larger values, while the populations of $\lambda_2(t)$ and $\lambda_3(t)$ decrease. Then, ${\left| \Psi_1^I(x;t) \right|}^2$ provides the dominant contribution to $\ket{\Psi_{MB}(t)}$. The above-described decrease of the higher-lying Schmidt coefficients leads to the decreasing tendency of $S_{VN}(t)$. This decreasing behavior of $S_{VN}$ is more pronounced for larger values of $g_{BI}$ and essentially indicates the attenuation of the oscillation amplitude of the impurity dictated in $\rho_I^{(1)}(x;t)$ and $\braket{X_I(t)}$, see for instance Figs. \[fig:7\] (f) and Fig. \[fig:9\]. It is worth mentioning here that the attenuation of $\braket{X_I(t)}$ is also related to an energy transfer from the impurity to the BEC (see Section \[sec:energy\_attractive\]). Therefore, the decreasing tendency of $S_{VN}(t)$ is reminiscent of a cooling process for the impurity atom. Moreover, by inspecting Fig. \[fig:11\] (a) it becomes evident that entering deeper to the attractive regime of interactions leads to a larger magnitude of entropy, e.g. compare $S_{VN}(t)$ for $g_{BI}=-0.5$ and $g_{BI}=-2$.
{width="100.00000%"}
Interspecies Energy Exchange {#sec:energy_attractive}
----------------------------
To further comprehend the dissipative motion of the impurity through the BEC for attractive interspecies interactions we also investigate the dynamics of the individual energy contributions of the bosonic mixture. The resulting energy parts following a quench to weak attractions, e.g. $g_{BI}=-0.2$, are presented in Fig. \[fig:11\] (b). We observe that the interspecies interaction energy, $E_{BI}(t)$, and the energy of the impurity, $E_I(t)$, oscillate out-of-phase with one another in time with a weak amplitude taking negative and positive values respectively. Also when $E_{BI}(t)$ is maximized the corresponding $E_I(t)$ minimizes since then the impurity resides in regions of lower BEC density. The reverse process occurs when the impurity is close to the trap center, i.e. $E_{BI}(t)$ minimizes and accordingly $E_{I}(t)$ is maximized. On the other hand, the energy of the BEC $E_{B}(t)$ shows a minor increase at the very early stages of the dynamics and subsequently it remains constant. This increasing tendency of $E_{B}(t)$ indicates that the impurity conveys a minor amount of energy to the BEC.
Turning to strong attractive interspecies interactions, e.g. $g_{BI}=-2$ demonstrated in Fig. \[fig:11\] (c) the energy contributions exhibit a completely different behavior. At the very early stages of the dynamics, i.e. $0<t<5$, the energy of the bath $E_{B}(t)$ and the impurity $E_{I}(t)$ increase whilst the interaction energy $E_{BI}(t)$ reduces. The increasing behavior of $E_{I}(t)$ indicates that the impurity gains kinetic energy due to the quench transferring also a part of its energy to the bosonic gas [@Mistakidis_orth_cat; @Nielsen] which creates sound waves, see also Fig. \[fig:7\] (f). For later evolution times $E_{I}(t)$ remains almost constant since the impurity is strongly localized while $E_{B}(t)$ and $E_{BI}(t)$ fluctuate due to the existence of sound waves in the BEC background [@Katsimiga_diss_flow].
Effective Mass {#sec:effective_mass}
==============
Having analyzed the nonequilibrium dynamics of the subsonic impurity which penetrates the BEC we next measure its effective mass $m^{eff}$. We remark that the effective mass of quasiparticles has been measured experimentally based on the collective excitations of the impurities, e.g. their breathing motion [@Catani; @Scazza]. Recall here that for weak repulsive interactions, $0<g_{BI}<0.95$, the impurity moves back and forth with respect to the trap center and remains within the bosonic bath. Entering the strong repulsive regime, $g_{BI}>1$, it probes the left edge of the BEC where it equilibrates for longer evolution times. However for attractive interspecies coupling strengths it performs a damped oscillatory motion within the bosonic medium. In all cases, since the impurity interacts with the BEC it is dressed by the excitations of the latter forming a quasiparticle. To model the motion of the impurity within the BEC we assume that it follows the following effective damped equation of motion $$\ddot{x}+\frac{\gamma^{eff}}{m^{eff}}\dot{x}=-\frac{\omega^{eff}}{m^{eff}}x.\label{effective_equation}$$ Here, $\omega^{eff}$ refers to the effective trapping of the formed quasiparticle due to the combined effect of its interaction with the bath and the presence of the external harmonic confinement. Furthermore, $m^{eff}$ denotes the effective mass of the impurity and $\gamma^{eff}$ is the effective damping parameter of the impurity due to its motion inside the BEC. We also remark that the above effective description inherently involves the assumption that the impurity is effectively trapped by the bosonic bath. Therefore it is valid only for the interaction interval $-2.5<g_{BI}<0.95$ where the impurity does not escape from the Thomas-Fermi radius of the BEC.
To determine the effective mass of the formed quasiparticle as well as its effective trapping frequency and damping parameter within $-2.5<g_{BI}<0.95$, we perform the following analysis. We first calculate the mean position, $\braket{X_I(t)}$, and momentum, $\braket{P_I(t)}$, of the impurity for a fixed interspecies interaction quench solely relying on our numerical calculations described in Sections \[sec:quench\_repulsive\] and \[sec:quench\_attractive\]. Then, by solving Eq. (\[effective\_equation\]) it can be easily shown that the mean position of the impurity reads $$\begin{split}
\braket{X_I(t)}=e^{-\frac{\gamma^{eff}}{2m^{eff}}t}&\Big[x_0 \cos(\omega_{0}t)\\&-\frac{p_0+\frac{\gamma^{eff}}{2}x_0}{m^{eff}\omega_0}\sin(\omega_0t)\Big], \label{position_theory}
\end{split}$$ with $\omega_0=\sqrt{(\omega^{eff})^2-(\frac{\gamma^{eff}}{2})^2}$. Moreover, since we consider that initially the impurity-BEC interaction is zero, i.e. $g_{BI}=0$, we obtain $p_0\equiv\braket{\Psi(0)|\hat{p}|\Psi(0)}= \hbar m u_0$ and $x_0\equiv\braket{\Psi(0)|\hat{x}|\Psi(0)}=0$. Also, the mean momentum of the impurity obeys the following equation $$\begin{split}
&\braket{P_I(t)}=e^{-\frac{\gamma^{eff}}{2m^{eff}}t}\bigg\{ p_0 \cos(\omega_{0}t)\\&+\Big[m^{eff}\omega_0x_0+\frac{\gamma^{eff}(p_0+\frac{\gamma^{eff}}{2}x_0)}{2m^{eff}\omega_0}\Big]\sin(\omega_0t) \bigg\}. \label{momentum_theory}
\end{split}$$ Evidently, in the above equations the unknown parameters that need to be determined are $\omega^{eff}$, $m^{eff}$ and $\gamma^{eff}$. In order to estimate these parameters we perform a fitting of the analytical form of both $\braket{X_I(t)}$ and $\braket{P_I(t)}$ provided by Eqs. (\[position\_theory\]) and (\[momentum\_theory\]) to the corresponding numerically obtained results of $\braket{X_I(t)}$ and $\braket{P_I(t)}$. The values of the parameters $\omega^{eff}$, $m^{eff}$ and $\gamma^{eff}$ obtained via this fitting procedure are shown in Fig. \[fig:12\] for a varying $g_{BI}$ such that $-2.5<g_{BI}<0.95$ where the quasiparticle picture is well defined.
Focusing on the attractive regime of interactions we observe that the effective mass of the emergent quasiparticle is larger than its bare mass and tends to the latter, i.e. $m^{eff}\to m$, as the non-interacting limit is approached. Additionally, the effective trapping frequency of the quasiparticle is larger than the actual frequency of the external harmonic oscillator and overall $\omega^{eff}$ exhibits a decreasing tendency as $g_{BI}\to0$. This result is in line with the previously discussed effective potential picture $V_I^{eff}(x)$, see Section \[sec:density\_attractive\] and Fig. \[fig:10\] (b). Moreover, the effective damping parameter $\gamma^{eff}$ acquires a finite value signaling the dissipative motion of the impurity inside the BEC and it tends to vanish for $g_{BI}\to 0$. Turning to repulsive interactions the quasiparticle effective mass is very close to the bare value, e.g. $m^{eff}\approx 1.001 m$ at $g_{BI}=0.1$, while for increasing repulsion it becomes slightly larger, namely $m^{eff}\approx 1.043 m$ at $g_{BI}=0.5$. Note here that this behavior of $m^{eff}$ in the repulsive regime of interactions is in contrast to the one discussed in Ref. [@Mistakidis_eff_mass] where $m^{eff}$ has been found to become smaller than the bare mass of the impurity. In this latter case the effective potential used to describe the quasiparticle formation did not include a damping parameter, an assumption which has been proved sufficient for the zero velocity impurity. However in the present case the damping term is important for the description of the observed dynamics and it is responsible for the aforementioned discrepancy. Also, the effective trapping frequency is smaller than the one of the external harmonic oscillator and shows a decreasing tendency for larger repulsions. This behavior is in accordance with the effective potential picture introduced in Section \[sec:density\_repulsive\], see also Fig. \[fig:4\] and Eq. (\[effective\_potential\_repulsive\]). Furthermore, $\gamma^{eff}$ takes small values and increases slightly as $g_{BI}$ becomes stronger.
To expose the role of non-perturbative effects in the resulting effective mass of the quasiparticle we further calculate $m^{eff}$ relying on the well-known perturbative expansion of the Fröhlich model [@Grusdt_approaches]. Note that this model operates in the absence of an external confinement, i.e. $\omega=0$. Indeed, it can be shown that the leading order correction of the effective mass [@Grusdt_approaches; @Devreese_notes; @Grusdt_1D] with respect to $g_{BI}$ is given by $$m^{eff}=m_I+4g_{BI}^2A+ \mathcal{O}(g_{BI}^4).\label{frohlich}$$ In this expression, the constant $A=\int_0^{\infty}dk \frac{k^2(V_k/g_{BI})^2}{(\omega_k+k^2/2m_{I})^3}$ with the scattering amplitude defined by $V_k=\sqrt{n_0}(2\pi)^{-1/2}g_{BI}(\frac{(\xi k)^2}{2+(\xi k)^2})^{1/4}$. Also the healing length and the speed of sound of the BEC are $\xi=(2m_Bg_{BB}n_0)^{-1/2}$ and $u_c=\sqrt{\frac{g_{BB}n_0}{m_B}}$ respectively with $n_0$ being the density of the homogeneous bosonic gas. To adequately compare the prediction given by Eq. (\[frohlich\]) with our results which include an external trap we choose $n_0=\rho^{(1)}_B(x=0;t=0)$. Furthermore, the dispersion relation of the elementary excitations of the bosonic gas corresponds to $\omega_k=u_ck \sqrt{1+\frac{1}{2}(k \xi)^2}$. Figure \[fig:12\] (a) shows $m^{eff}$ within the Fröhlich model for varying $g_{BI}$. Strikingly, the predictions of the Fröhlich model and the full many-body approach are in very good agreement with one another both at weak attractive and repulsive interspecies interactions. Therefore we can deduce that for such weak interspecies interaction strengths the external trapping does not play any crucial role for the effective mass of the impurity. This result is not surprising since at these weak interactions the impurity resides well inside the bosonic gas and thus approximately experiences a homogeneous background. Noticeable deviations are observed for strong attractive interactions $g_{BI}>-1.25$, e.g. being of the order of $3\%$ and $9\%$ at $g_{BI}=-2.0$ and $g_{BI}=-2.5$ respectively. Also, small differences on $m^{eff}$ estimated between the two aforementioned approaches occur on the repulsive regime of interactions, and especially for $g_{BI}\geq 0.8$ become larger than $3\%$. These deviations can be partly attributed to the effect of the trap since for an increasing repulsion the density of the impurity probes the edges of the cloud of the BEC. Similarly the effect of the harmonic trap cannot be neglected for large attractive interactions. Indeed the effective potential in this case amplifies any small discrepancies of the BEC density from the homogeneous case that occur around the trap center.
As a final remark we note that the effective mass depends weakly on the initial, i.e. before the quench, velocity $u_0$ of the impurity. For instance, referring to a fixed postquench interspecies interaction strength e.g. $g_{BI}=-1$ the effective mass takes values $m^{eff}\approx1.19$ for $u_0\approx-u_c$, $m^{eff}\approx1.15$ when $u_0=-u_c/2$ and $m^{eff}\approx1.13$ if $u_0=-u_c/5$.
{width="100.00000%"}
Single-Shot Simulations {#sec:single_shots}
=======================
To provide further possible experimental links of our results we simulate in-situ single-shot absorption measurements [@Sakmann_single_shot; @mistakidis_phase_sep] aiming at demonstrating how in-situ imaging can be used to adequately monitor the quench-induced dynamical dressing of the impurity. These measurements probe the spatial configuration of the atoms and therefore the many-body probability distribution which is indeed available within ML-MCTDHX. The corresponding experimental images are obtained via a convolution of the spatial particle configuration with a point spread function that essentially dictates the experimental spatial resolution. Below, we present such simulations by employing a point spread function of Gaussian shape and width $w_{PSF}=1 \ll l\approx 3.16$, with $l=\sqrt{1/\omega}$ being the harmonic oscillator length. Note also that $w_{PSF} > \xi \approx 0.4$, where $\xi=\frac{\hbar}{\sqrt{2}mu_c}$ denotes the healing length of the BEC. For a more elaborated discussion on the details of the numerical implementation of this process in one-dimensional binary systems we refer the reader to Appendix \[sec:single\_shot\_algorithm\] and also to Refs. [@mistakidis_phase_sep; @Erdmann_phase_sep].
{width="100.00000%"}
Having at hand the many-body wavefunction of our system within ML-MCTDHX we reproduce in-situ single-shot images for the BEC medium $B$, $\mathcal{A}^B(\tilde{x};t)$, and the impurity $I$, $\mathcal{A}^I(\tilde{x'}|\mathcal{A}^B(\tilde{x});t_{im})$, at each time-instant of the evolution. Here, $t_{im}$ denotes the time-instant of the imaging. In particular, we consecutively image first the BEC and then the impurity species. Note that the reverse imaging process does not affect the image obtained after averaging over several single-shots, see also Appendix \[sec:single\_shot\_algorithm\]. Further details of the corresponding simulation process of this experimental technique are discussed in Appendix \[sec:single\_shot\_algorithm\]. In the following we analyze the nonequilibrium dynamics of the bosonic mixture for quenches towards strongly repulsive, $g_{BI}=1.5$, and attractive interactions, i.e. $g_{BI}=-2$. We remark that a similar analysis has been followed also for other values of $g_{BI}$ (not shown here for brevity reasons). Before describing the outcome of the images it is noteworthy to mention that a direct correspondence between the single-particle density and only one single-shot image is not possible due to the small particle number of the considered setup, $N_B=100$ and $N_I=1$. Such a resemblance is feasible only when considering large particle numbers, e.g. of the of order $10^5$ particles [@Katsimiga_diss_flow]. Another reason that excludes the possibility of explicitly observing the one-body density within a single-shot image is the presence of multiple orbitals in the system \[Eqs. (\[Eq:SPF\_bath\]) and (\[Eq:SPF\_impurity\])\]. More specifically, the many-body state is expressed as a superposition over multiple orbitals \[see Eqs. (\[Eq:SPF\_bath\]) and (\[Eq:SPF\_impurity\])\] and thus imaging an atom alters the many-body state of the other atoms and as a consequence their one-body density. A more elaborated discussion on this topic is provided in Refs. [@mistakidis_phase_sep; @Katsimiga_bent; @Katsimiga_diss_flow]. Most importantly, it can be demonstrated that the average image e.g. of the BEC ($B$ species) i.e. $\mathcal{\bar{A}}^{B}(\tilde{x})$, over a sample of $N_{shots}$ single-shot images, $\mathcal{A}^{B}(\tilde{x})$, is related to the $B$ species one-body density, namely $\rho_{B}^{(1)} (x_{B}')$, as follows $$\mathcal{\bar{A}}^{B}(\tilde{x})=\frac{N_{B}}{\sqrt{2\pi}w_{PSF}}\int dx_{B}' e^{-\frac{(\tilde{x}-x_{B}')^2}{2w^2_{PSF}}} \rho_{B}^{(1)} (x_{B}').
\label{average_images}$$ In this expression, $\tilde{x}$ are the spatial coordinates within the image and $x_{B}'$ refers to the spatial coordinate of the $B$ species. The width of the employed point spread function is $w_{PSF}$ and the species particle number is $N_{B}$. A similar relation holds for the other species but using the corresponding images.
According to our above discussion in order to retrieve the one-body density of each subsystem we rely on an average of several single-shot images for each species. In particular, we measure $\bar{\mathcal{A}}^B(\tilde{x};t)=1/N_{shots}\sum_{k=1}^{N_{shots}}
\mathcal{A}_k^B(\tilde{x};t)$ for the BEC and $\bar{\mathcal{A}}^I(\tilde{x}^{'}|\mathcal{A}^B(\tilde{x});t)=1/N_{shots}\sum_{k=1}^{N_{shots}}
\mathcal{A}_k^I(\tilde{x}^{'}|\mathcal{A}^B(\tilde{x});t)$ for the impurity atom respectively. Figures \[fig:13\] (a)-(f) show $\bar{\mathcal{A}}^B(\tilde{x};t)$ and $\bar{\mathcal{A}}^I(\tilde{x}^{'}|\mathcal{A}^B(\tilde{x});t)$ for different number of samplings, i.e. $N_{shots}$, upon considering a quench from $g_{BI}=0$ to $g_{BI}=1.5$. Comparing this averaging process for an increasing number of $N_{shots}$ and the actual single-particle density calculated via ML-MCTDHX \[see Figs. \[fig:2\] (e), (f)\] unveils that they become almost the same. More specifically, it can easily be deduced that for $N_{shots}>100$ the $\bar{\mathcal{A}}^A(\tilde{x};t)$ and the $\bar{\mathcal{A}}^B(\tilde{x}^{'}|\mathcal{A}^A(\tilde{x});t)$ tend gradually to $\rho^{(1)}_{B}(x;t)$ and $\rho^{(1)}_{I}(x;t)$ respectively. The same overall phenomenology occurs for the case of a quench towards the attractive interaction regime, here $g_{BI}=-2$, as illustrated in Figs. \[fig:13\] (g)-(k). Indeed, the dissipative motion of the impurity and its imprint on the BEC background are fairly captured even for $N_{shots}=10$, e.g. compare Figs. \[fig:13\] (g)-(h) with Figs. \[fig:7\] (e)-(f).
Utilizing the aforementioned single-shots we can further probe the spatial configuration of the bosonic gas in the co-moving frame of the impurity. This procedure sheds light on the imprint of the impurity motion onto the correlations emanating within the bosonic mixture. Such a protocol has been succesfully experimentally implemented to probe the internal structure of magnetic polarons [@Koepsell]. Within this protocol we shift each of the previously obtained single-shots, $\mathcal{A}_k^{B}(\tilde{x},t)$ by the amount $X^I_k=\int d\tilde{x}' ~ \tilde{x}' \mathcal{A}_k^I(\tilde{x}^{'}|\mathcal{A}^B(\tilde{x});t)$ being the measured position of the impurity at the $k$-th single-shot, i.e. $\mathcal{A}_k^{BI}(x_r,t)=\mathcal{A}_k^{B}(\tilde{x}-X^I_k,t)$. It can be shown that the corresponding average image $\mathcal{\bar{A}}^{BI}(x_r,t)=\sum_{i=1}^{N_{shots}} \mathcal{A}_k^{BI}(x_r,t)$ over $N_{shots}$ is related to the two-body interspecies correlation function as follows $$\begin{split}
\mathcal{\bar{A}}^{BI}(x_r,t)=\frac{N_B}{\sqrt{2\pi}w_{PSF}}&\int dx'_{B} d x'_I e^{-\frac{[x_r-(x_{B}'-x'_I)]^2}{2w^2_{PSF}}} \\& \times \rho_{BI}^{(2)} (x'_{B}, x'_I;t).
\end{split}$$ Here, $\rho_{BI}^{(2)}(x_1,x_2;t)=\langle\Psi_{MB}(t)|\hat\Psi^{\dagger B}(x_1)\hat\Psi^{\dagger I}(x_2)\hat\Psi^{B}(x_1)$ $\hat\Psi^{ I}(x_2)|\Psi_{MB}(t) \rangle$ are the diagonal elements of the two-body interspecies reduced density matrix [@mistakidis_phase_sep; @Sakmann_den_matr]. The latter provides the probability of measuring a $B$- and a $I$-species particle simultaneously at positions $x_1$ and $x_2$ respectively. $\bar{A}^{BI}(x_r,t)$ obtained from $N_{shots}=800$ is presented in Figs. \[fig:13\] (m) and (n) exemplarily for a quench to positive $g_{BI}=0.5$ and negative $g_{BI}=-2$ interactions respectively. Recall that the effective quasiparticle description holds only when the impurity is effectively trapped into the bosonic bath and therefore in our case is valid for $-2.5<g_{BI}<0.95$. For repulsive postquench interactions we observe that $\bar{A}^{BI}(x_r,t)$ shows an overall oscillatory behavior which is a consequence of the mere fact that the impurity undergoes for these interactions an oscillatory motion inside the bosonic gas, see also Fig. \[fig:2\] (b). Focusing on $\bar{A}^{BI}(x_r,t)$ in the vicinity of the impurity, i.e. $x_r \approx 0$, we deduce that the latter repels the particles of the bosonic gas leading to the development of shallow dips in $\bar{A}^{BI}(x_r,t)$, e.g. see the dashed rectangles in Fig. \[fig:13\] (m). Turning to strong attractive interactions, see Fig. \[fig:13\] (n), we discern the formation of a pronounced peak in the vicinity of $x_r=0$ caused by the presence of the impurity \[see also Fig. \[fig:7\] (f)\]. This result is in accordance to the effective potential picture \[Eq. (\[effective\_pot\_bath\]) and Fig. \[fig:10\]\] Also the height of this central peak fluctuates which is hardly discernible in Fig. \[fig:13\] (n). Additionally, an overall oscillatory behavior of $\bar{A}^{BI}(x_r,t)$ occurs since we operate in the co-moving frame of the impurity. Concluding based on $\bar{A}^{BI}(x_r,t)$ we deduce that the interspecies two-body correlations between the impurity and the BEC are much more prevalent in the case of attractive interactions.
Summary and Conclusions {#sec:conclusions}
=======================
We have studied the interspecies interaction quench quantum dynamics of a subsonically moving impurity that penetrates a harmonically trapped Bose-Einstein condensate. Monitoring the time-evolution of the impurity on the single-particle level we identify a variety of response regimes arising for different interaction strengths.
For weak postquench interspecies repulsive interactions the subsonic impurity performs a dipole motion inside the bosonic bath. The latter remains essentially unperturbed exhibiting some small distortions from its initial Thomas-Fermi profile. Increasing the interspecies coupling, the oscillation period of the impurity becomes larger and shallow density dips built upon the bosonic density thus imprinting the impuritys’ motion. However, at strong quench amplitudes such that the interspecies interaction exceeds the bosonic intraspecies one the dynamical behavior of the impurity is significantly altered. More specifically, the impurity travels in the direction of its initial velocity towards the corresponding edge of the BEC background and thereafter fluctuates around the Thomas-Fermi radius. This latter behavior of the impurity at strong repulsive interactions occurs independently of its initial velocity, its prequench position, the trapping frequency and the mass ratio of the atomic species. Most importantly it takes place due to the involvement of correlations since e.g. within the mean-field approximation the impurity undergoes an oscillatory motion inside the bosonic bath throughout the dynamics. Employing the Von-Neumann entropy reveals the development of strong interparticle correlations in the course of the evolution, a result that becomes more pronounced for larger repulsions. Inspecting the individual energy contributions of each species we unveil that the impurity dissipates energy into the bosonic medium, a phenomenon that is more enhanced for increasing interspecies interactions. To interpret the dynamics of the impurity we construct an effective potential which corresponds to a modified harmonic oscillator for weak interactions turning to a double-well when entering the strongly repulsive regime.
Entering attractive interspecies interactions we showcase that the impurity undergoes a damped oscillatory motion inside the bosonic bath. This behavior becomes more pronounced for an increasing attraction where the impurity exhibits a localization tendency and the BEC develops a density peak at the location of the impurity. It is shown that the above response of each species can be intuitively understood in terms of an effective potential picture for the bath and the impurity independently. Moreover, by invoking the energy contributions of each species we find that the impurity transfers a part of its energy to the bosonic medium which in turn generates sound waves being also evident in its single-particle density. Also, an inspection of the Von-Neumann entropy shows the presence of interspecies correlations especially for stronger attractive interactions.
We have estimated the effective mass of the emergent quasiparticle by modeling its damped motion through the medium with an effective dissipative equation of motion. Performing a fitting of our numerical results and the analytical prediction of this dissipative equation we are able to estimate the effective mass, trapping frequency and damping parameter of the impurity. It is found that in the attractive regime of interactions the effective mass and trapping frequency are larger than the bare ones and tend to the latter when approaching the non-interacting limit. Also, the effective damping parameter acquires a finite value and tends to vanish for zero interspecies couplings. For repulsive interactions the quasiparticles’ effective mass is slightly larger than its bare value while the damping parameter acquires small values. The corresponding effective trapping frequency is smaller than the one of the external harmonic oscillator showing a decreasing tendency for larger repulsions. Finally, we have provided possible experimental evidences of the impurity dynamics by simulating in-situ single-shot measurements. In particular, we showcase how an increasing sampling of such images can be used to adequately retrieve the observed dynamics.
There is a variety of possible extensions of the present work in future endeavors. An imperative prospect is to unravel the resultant interspecies interaction quench dynamics upon considering two or more interacting bosonic impurities immersed in a bosonic bath. This study will shed light into the presence of the most probably emergent induced interactions between the impurities and would enable us to systematically explore their role in the time-evolution. Additionally, the inclusion of temperature effects in such an investigation would be very interesting [@Tajima; @Liu]. Another intriguing direction would be to simulate the corresponding radiofrequency spectrum [@Mistakidis_Fermi_pol] or the structure factor of the current setup [@Mistakidis_orth_cat; @Shchadilova] by employing spinor impurities in order to identify the possibly emerging polaronic states and subsequently measure e.g. their lifetime and residue. Certainly the generalization of the present results to higher-dimensional settings is highly desirable.
Technical Details of the Single-Shot Algorithm {#sec:single_shot_algorithm}
==============================================
To perform the simulation of the single-shot procedure we employ a sampling of the many-body probability distribution [@Sakmann_single_shot; @Katsimiga_diss_flow; @mistakidis_phase_sep]. The latter is available in terms of the ML-MCTDHX approach for each time-instant of the evolution. It is important to note at this point that the numerical implementation of this experimental procedure has already been reported and applied to a variety of setups including neutral and spinor atoms [@Katsimiga_bent; @Katsimiga_diss_flow; @Koutentakis_prob] as well as binary mixtures [@mistakidis_phase_sep; @Erdmann_phase_sep]. In this sense, below we briefly outline the corresponding numerical procedure but for more details and extensive discussions we refer the reader to Refs. [@Sakmann_single_shot; @mistakidis_phase_sep; @Katsimiga_bent; @Katsimiga_diss_flow].
As it has already been argued in previous works [@Sakmann_single_shot; @mistakidis_phase_sep; @Katsimiga_bent; @Katsimiga_diss_flow; @Koutentakis_prob], the single-shot procedure for binary mixtures is crucially affected by the systems’ intra- and interspecies correlations. Indeed, for a many-body state the presence of entanglement \[see Eq. (\[Eq:WF\])\] among the distinct species is important regarding the image ordering. This dependence can be understood by resorting to the underlying Schmidt decomposition \[see Eq. \[Eq:WF\]\] since it directly affects the Schmidt coefficients $\lambda_k$. Below, we briefly sketch the numerical process when imaging first the BEC $B$ and subsequently the impurity $I$ species. In this way, we obtain the corresponding absorption images $\mathcal{A}^B(\tilde{x})$ and $\mathcal{A}^I(\tilde{x}'|\mathcal{A}^B(\tilde{x}))$. To avoid any confusion, let us remark that in order to image first the $I$ and then the $B$ species we can follow the same procedure, retrieving the images $\mathcal{A}^I(\tilde{x})$ and $\mathcal{A}^B(\tilde{x}'|\mathcal{A}^I(\tilde{x}))$. It is also worth mentioning that the image ordering plays a role when one is interested in the individual single-shot images. However, in our case that we discuss the average of a sample of single-shots (see the discussion in Section \[sec:single\_shots\]) the image ordering is irrelevant since all the effects stemming from entanglement are averaged out.
To perform the imaging of the $B$ and then of the $I$ species we first annihilate one-by-one all $B$-species bosons. Referring to a specific time-instant of the imaging, e.g. $t_{im}$, a random position is drawn obeying $\rho_{N_B}^{(1)}(x_1')>z_1$ where $z_1$ is a random number taking values in the interval \[$0$, $ \max\lbrace{\rho^{(1)}_{N_B}(x;t_{im})\rbrace}$\]. Then by utilizing the projection operator $\frac{1}{\mathcal{N}}(\hat{\Psi}^B(x_1')\otimes \hat{\mathbb{I}}_I)$ we project the ($N_B+N_I$)-body wavefunction onto the ($N_B-1+N_I$)-body one. The bosonic field operator annihilating a $B$ species boson at position $x_1'$ is $\hat{\Psi}^B(x_1')$ and $\mathcal{N}$ denotes the normalization constant. Evidently, this process affects the Schmidt coefficients, $\lambda_k$, and consequently the densities $\rho^{(1)}_{N_B-1}(t_{im})$ and $\rho^{(1)}_{N_I}(t_{im})$ are altered. As a result, the Schmidt decomposition of the many-body wavefunction following this first measurement reads $$\begin{split}
&\ket{\tilde{\Psi}_{MB}^{N_B-1,N_I}(t_{im})}=\\ &\sum_i \sqrt{\tilde{\lambda}_{i,N_B-1}(t_{im})}\ket{\tilde{\Psi}_{i,N_B-1}^B(t_{im})}\ket{\Psi_i^I(t_{im})}.
\label{Eq:wfn_first_measurement}
\end{split}$$ In this expression, $\ket{\tilde{\Psi}_{i,N_B-1}^B}=\frac{1}{N_i}\hat{\Psi}^B(x_1')\ket{\Psi_i^B}$ is the $N_B-1$ species wavefunction with $N_i=\sqrt{\bra{\Psi_i^B}\hat{\Psi}^{B\dagger}(x_1')\hat{\Psi}^B(x_1')\ket{\Psi_i^B}}$. Moreover, the Schmidt coefficients of the ($N_B-1+N_I$)-body wavefunction are given by $\tilde{\lambda}_{i,N_B-1}=\lambda_i N_i/\sum_i \lambda_i N_i^2$.
The imaging process of the $B$-species is finalized after repeating the above-mentioned steps $N_B-1$ times realizing the following distribution of positions ($x'_1$, $x'_2$,...,$x'_{N_B-1}$). This distribution is then convoluted with a point spread function resulting in the single-shot image of the $B$-species $\mathcal{A}^B(\tilde{x})=\frac{1}{\sqrt{2\pi}w_{PSF}}\sum_{i=1}^{N_B}e^{-\frac{(\tilde{x}-x'_i)^2}{2w_{PSF}^2}}$, where $\tilde{x}$ are the spatial coordinates within the image and $w_{PSF}$ is the width of the point spread function. After annihilating all $N_B$ atoms, the many-body wavefunction acquires the form $$\begin{split}
&\ket{\tilde{\Psi}_{MB}^{0,N_B}(t_{im})}=\\ &\ket{0^B} \otimes\sum_i \frac{\sqrt{\tilde{\lambda}_{i,1}(t_{im})}
\braket{x'_{N_B}|\Phi_{i,1}^B}}{\sum_j{\sqrt{\tilde{\lambda}_{j,1}(t_{im})|\braket{x'_{N_B}|\Phi_{j,1}^B}|^2}}}\ket{\Psi_i^I(t_{im})}.
\label{Eq:A3}
\end{split}$$ Here, the single-particle orbital of the $j$-th mode is $\braket{x'_{N_B}|\Phi_{j,1}^B}\equiv\braket{0^B|\hat{\Psi}^B(x'_{N_B})|\Phi_{j,1}^B}$ and the second term in the cross product of the right-hand side ($\ket{\Psi_{MB}^{N_I}(t_{im})}$) denotes the impurity species wavefunction. The latter is a non-entangled single-particle wavefunction ($N_I=1$) and as a consequence the corresponding single-shot procedure of the $I$ species reduces to that of a single-species [@Sakmann_single_shot; @Katsimiga_bent; @Katsimiga_diss_flow]. Indeed, for an imaging time $t=t_{im}$, we measure $\rho^{(1)}_{N_I}(x;t_{im})$ from $\ket{\Psi^{N_I}_{MB}}\equiv \ket{\Psi(t_{im})}$ and draw a random position $x''_1$ satisfying $\rho^{(1)}_{N_I}(x''_1;t_{im})>z_2$. Here, $z_2$ is a random number bounded in the interval \[$0$, $\rho^{(1)}_{N_I}(x;t_{im})$\]. As a result, the $I$-species particle is annihilated at position $x''_1$ and this position is subsequently convoluted with a point spread function resulting to the single-shot image $\mathcal{A}^I(\tilde{x'}|\mathcal{A}^B(\tilde{x}))$.
Convergence of the Many-Body Simulations {#sec:convergence_numerics}
========================================
To simulate the correlated nonequilibrium quantum dynamics of the considered binary mixture we resort to the Multi-Layer Multi-Configurational Time-Dependent Hartree Method for Atomic Mixtures (ML-MCTDHX) [@MLX; @MLB1], see also Section \[ML\_ansatz\]. It is a variational approach for solving the time-dependent many-body Schr[ö]{}dinger equation of atomic mixtures consisting either of bosonic [@mistakidis_phase_sep; @Mistakidis_eff_mass; @Mistakidis_orth_cat] or fermionic [@Erdmann_phase_sep; @Koutentakis_prob] components that might additionally include spin degrees of freedom [@Mistakidis_orth_cat; @Mistakidis_Fermi_pol]. More specifically, this method relies on the expansion of the many-body wavefunction in terms of a time-dependent and variationally optimized basis. Such a treatment enables us to include all the important inter- and intraspecies correlations into our many-body ansatz utilizing a computationally feasible basis size. In this way, it allows us to span the relevant subspace of the Hilbert space at each time-instant in an efficient manner. The latter is in contrast to methods relying on a time-independent basis where the number of basis states can be significantly larger rendering the simulation of intermediate size systems impossible.
The underlying Hilbert space truncation is inferred from the used orbital configuration space, namely $C=(D;d_B;d_I)$. In this notation, the number of species and single-particle functions of each species are denoted by $D=D_B=D_I$ and $d_B$, $d_I$ respectively, see also Eqs. (\[Eq:WF\]), (\[Eq:SPF\_bath\]) and (\[Eq:SPF\_impurity\]). We remark here that since we use a single impurity then by definition $D=d_I$ holds. Additionally, for our numerical calculations a primitive basis based on a sine discrete variable representation consisting of 1000 grid points is employed. This sine discrete variable representation intrinsically introduces hard-wall boundary conditions which in our case are imposed at $x_\pm=\pm80$. Of course, the location of these boundaries do not affect our results since we do not observe appreciable densities to occur beyond $x_{\pm}=\pm40$.
{width="90.00000%"}
To conclude upon the convergence of our many-body simulations we ensure that all observables of interest become almost insensitive, to a certain degree, when varying the used orbital configuration space i.e. $C=(D;d_B;d_I)$. In our case, a convergent behavior of all the many-body calculations discussed in the main text has been achieved by exploiting the orbital configuration space $C=(8;3;8)$. To testify the convergence of our results for a different number of species and single-particle functions e.g. we examine the mean position of the impurity during the interspecies interaction quench dynamics. In particular, we investigate its absolute deviation between the $C=(8;3;8)$ and other orbitals configurations $C'=(D;d_B;d_I)$ $$\Delta \braket{X_I(t)}_{C,C'} =\frac{{\left| \braket{X_I(t)}_C -\braket{X_I(t)}_{C'} \right|}}{\braket{X_I(t)}_C}. \label{dev_mean_pos}$$ The time-evolution of $\Delta \braket{X_I(t)}_{C,C'}$ is presented in Fig. \[fig:14\] following an interspecies interaction quench from $g_{BI}=0$ to $g_{BI}=2$ \[Fig. \[fig:14\] (a)\] and $g_{BI}=-2$ \[Fig. \[fig:14\] (b)\]. Evidently, a systematic convergence of $\Delta \braket{X_I(t)}_{C,C'}$ in both the repulsive and the attractive regime of interactions can be inferred. Focusing on repulsive interactions, we observe that $\Delta \braket{X_I(t)}_{C,C'}$ between the $C=(8;3;8)$ and $C'=(9;3;9)$ orbital configurations lies below $4.3\%$ for all evolution times. Moreover, $\Delta \braket{X_I(t)}_{C,C'}$ calculated for $C=(8;3;8)$ and $C=(7;3;7)$ is at most $5\%$ during the dynamics. Similar observations can be made by inspecting $\Delta \braket{X_I(t)}_{C,C'}$ in the case of a quench towards attractive interactions, see Fig. \[fig:14\] (b). Indeed, the mean position when $C=(8;3;8)$ and $C'=(9;3;9)$ becomes at most of the order of $4.1\%$ while e.g. for $C=(8;3;8)$ and $C'=(6;4;6)$ it acquires a maximum value of $7.2\%$.
Furthermore we showcase the convergence of the Von-Neumann entropy in the course of the time-evolution. More precisely, we inspect the relative difference of $S_{VN}(t)$ calculated within the $C=(8;3;8)$ and different orbital configurations $C'=(D;d_B;d_I)$ namely $$\Delta S_{VN}(t)_{C,C'} =\frac{{\left| S_{VN}(t)_C -S_{VN}(t)_{C'} \right|}}{S_{VN}(t)_C}. \label{dev_Von_Neum}$$ The dynamics of the relative deviation $\Delta S_{VN}(t)_{C,C'}$ is illustrated in Fig. \[fig:14\] after an interaction quench from $g_{BI}=0$ to $g_{BI}=2$ \[Fig. \[fig:14\] (c)\] and $g_{BI}=-2$ \[Fig. \[fig:14\] (d)\] for different of orbital configurations $C'$ and fixed $C=(8;3;8)$. As it can be seen, convergence is achieved also for $\Delta S_{VN}(t)_{C,C'}$ at both repulsive and attractive postquench interspecies interaction strengths. For repulsive interspecies couplings, e.g. $g_{BI}=2$ presented in Fig. \[fig:14\] (c), the deviation $\Delta S_{VN}(t)_{C,C'}$ with $C=(8;3;8)$ and $C'=(9;3;9)$ \[$C=(7;3;7)$\] is smaller than $4\%$ \[$2.7\%$\] throughout the evolution. Turning to attractive postquench interactions such as $g_{BI}=-2$ \[Fig. \[fig:14\] (d)\], we deduce that $\Delta S_{VN}(t)_{C,C'}$ among the orbital configurations $C=(8;3;8)$ and either $C'=(9;3;9)$ or $C'=(6;4;6)$ takes a maximum value of the order of $6\%$ or $7.2\%$ respectively during the dynamics. We should also mention that a similar investigation has been performed for all other interspecies interaction quench amplitudes discussed in the main text and found to be adequately converged (not shown here for brevity).
Acknowledgements {#acknowledgements .unnumbered}
================
S.I.M. and P.S. gratefully acknowledge financial support by the Deutsche Forschungsgemeinschaft (DFG) in the framework of the SFB 925 “Light induced dynamics and control of correlated quantum systems”. F.G. acknowledges support from the Technical University of Munich - Institute for Advanced Study, funded by the German Excellence Initiative and the European Union FP7 under grant agreement 291763, from the DFG grant No. KN 1254/1-1, and DFG TRR80 (Project F8). S. I. M. gratefully acknowledges financial support by the Lenz-Ising Prize of the University of Hamburg.
[60]{}
C. Kohstall, M. Zaccanti, M. Jag, A. Trenkwalder, P. Massignan, G. M. Bruun, F. Schreck, and R. Grimm, Nature **485**, 615 (2012).
F. Scazza, G. Valtolina, P. Massignan, A. Recati, A. Amico, A. Burchianti, C. Fort, M. Inguscio, M. Zaccanti, and G. Roati, Phys. Rev. Lett. **118**, 083602 (2017).
A. Schirotzek, C. H. Wu, A. Sommer, and M. W. Zwierlein, Phys. Rev. Lett. **102**, 230402 (2009).
C. Chin, R. Grimm, P. Julienne, and E. Tiesinga, Rev. Mod. Phys. **82**, 1225 (2010).
T. K[ö]{}hler, K. G[ó]{}ral, and P. S. Julienne, Rev. Mod. Phys. **78**, 1311 (2006).
M. Cetina, M. Jag, R.S. Lous, J.T. Walraven, R. Grimm, R.S. Christensen, and G.M. Bruun, Phys. Rev. Lett. **115**, 135302 (2015).
M. Koschorreck, D. Pertot, E. Vogt, B. Fr[ö]{}hlich, M. Feld, and M. K[ö]{}hl, Nature **485**, 619 (2012).
M. Cetina, M. Jag, R. S. Lous, I. Fritsche, J. T. Walraven, R. Grimm, J. Levinsen, M. M. Parish, R. Schmidt, M. Knap, and E. Demler, Science **354**, 96 (2016).
T. Fukuhara, A. Kantian, M. Endres, M. Cheneau, P. Schauss, S. Hild, D. Bellem, U. Schollw[ö]{}ck, T. Giamarchi, C. Gross, I. Bloch, and S. Kuhr, Nat. Phys. [**9**]{}, 235 (2013). J. F. Sherson, C. Weitenberg, M. Endres, M. Cheneau, I. Bloch, and S. Kuhr, Nature **467**, 68 (2010).
P. Massignan, M. Zaccanti, and G. M. Bruun, Rep. Progr. Phys. **77**, 034401 (2014).
R. Schmidt, M. Knap, D. A. Ivanov, J. S. You, M. Cetina, and E. Demler, Rep. Progr. Phys. **81**, 024401 (2018).
L. D. Landau, Sov. Phys. JETP, **3**, 20 (1957).
L. P. Ardila, and Giorgini, Phys. Rev. A **92**, 033612 (2015).
F. Grusdt, G. E. Astrakharchik, and E. Demler, New J. Phys. **19**, 103035 (2017).
A. S. Dehkharghani, A. G. Volosniev, and N. T. Zinner, Phys. Rev. Lett. **121**, 080405 (2018).
S.I. Mistakidis, G.C. Katsimiga, G.M. Koutentakis, and P. Schmelcher, New J. Phys. **21**, 043032 (2019).
A. Camacho-Guardian, L. P. Ardila, T. Pohl, and G. M. Bruun, Phys. Rev. Lett. **121**, 013401 (2018).
K. Keiler, S. Krönke, and P. Schmelcher, New J. Phys. **20**, 033030 (2018).
F. Grusdt, K. Seetharam, Y. Shchadilova, and E. Demler, Phys. Rev. A **97**, 033612 (2018).
A. G. Volosniev, H. W. Hammer, and N. T. Zinner, Phys. Rev. A **92**, 023623 (2015).
Y. E. Shchadilova, R. Schmidt, F. Grusdt, and E. Demler, Phys. Rev. Lett. **117**, 113002 (2016).
N. A. Kamar, A. Kantian, and T. Giamarchi, Phys. Rev. A **100**, 023614 (2019).
S. I. Mistakidis, G. C. Katsimiga, G. M. Koutentakis, T. Busch, and P. Schmelcher, Phys. Rev. Lett. **122**, 183001 (2019).
S.I. Mistakidis, A.G. Volosniev, N.T. Zinner, and P. Schmelcher, Phys. Rev. A **100**, 013619 (2019).
S. I. Mistakidis, L. Hilbig, and P. Schmelcher, Phys. Rev. A **100**, 023620 (2019).
N. B. Jørgensen, L. Wacker, K. T. Skalmstang, M. M. Parish, J. Levinsen, R. S. Christensen, G. M. Bruun, and J. J. Arlt, Phys. Rev. Lett. [**117**]{}, 055302 (2016). M.-G. Hu, M. J. Van de Graaff, D. Kedar, J. P. Corson, E. A. Cornell, and D. S. Jin, Phys. Rev. Lett. [**117**]{}, 055301 (2016).
J. Catani, G. Barontini, G. Lamporesi, F. Rabatti, G. Thalhammer, F. Minardi, S. Stringari, and M. Inguscio, Phys. Rev. Lett. **103**, 140401 (2009).
Z. Z. Yan, Y. Ni, C. Robens, and M. W. Zwierlein, arXiv:**1904.02685** (2019).
F. Grusdt, and E. Demler, [*New theoretical approaches to Bose polarons. Quantum Matter at Ultralow Temperatures*]{}, 325-411 (2015).
S. P. Rath, and R. Schmidt, Phys. Rev. A **88**, 053632 (2013).
G. E. Astrakharchik, and L. P. Pitaevskii, Phys. Rev. A [**70**]{}, 013608 (2004).
F. M. Cucchietti, and E. Timmermans, Phys. Rev. Lett. [**96**]{}, 210401 (2006).
R. M. Kalas, and D. Blumme, Phys. Rev. A [**73**]{}, 043608 (2006).
M. Bruderer, A. Klein, S. R. Clark, and D. Jaksch, Europhys. Lett. [**82**]{}, 30004 (2008).
M. Bruderer, A. Klein, S. R. Clark, and D. Jaksch, Phys. Rev. A [**76**]{}, 011605(R) (2007). A. Privitera, and W. Hofstetter, Phys. Rev. A [**82**]{}, 063614 (2010).
W. Casteels, J. Tempere, and J. T. Devreese, Phys. Rev. A [**86**]{}, 043614 (2012). W. Casteels, J. Tempere, and J. T. Devreese, Phys. Rev. A [**88**]{}, 013613 (2013).
B. Kain, and H. Y. Ling, Phys. Rev. A [**94**]{}, 013621 (2016). J. Tempere, W. Casteels, M. K. Oberthaler, S. Knoop, E. Timmermans, and J. T. Devreese, Phys. Rev. B **80**, 184504 (2009).
A. G. Volosniev, and H. W. Hammer, Phys. Rev. A **96**, 031601 (2017).
A. S. Dehkharghani, A. G. Volosniev, and N. T. Zinner, Phys. Rev. A **92**, 031601 (2015).
F. Grusdt, R. Schmidt, Y. E. Shchadilova, and E. Demler, Phys. Rev. A **96**, 013607 (2017).
F. M. Cucchietti, and E. Timmermans, Phys. Rev. Lett. **96**, 210401 (2006).
M. Schecter, D. M. Gangardt, and A. Kamenev, New J. Phys. **18**, 065002 (2016).
Z. Cai, L. Wang, X. C. Xie, and Y. Wang, Phys. Rev. A **81**, 043602 (2010).
T. H. Johnson, S. R. Clark, M. Bruderer, and D. Jaksch, Phys. Rev. A **84**, 023617 (2011).
S. Krönke, J. Knörzer, and P. Schmelcher, New J. Phys. **17**, 053001 (2015).
T. Lausch, A. Widera, and M. Fleischhauer, Phys. Rev. A **97**, 033620 (2018).
T. Lausch, A. Widera, and M. Fleischhauer, Phys. Rev. A **97**, 023621 (2018).
E. Burovski, V. Cheianov, O. Gamayun, and O. Lychkovskiy, Phys. Rev. A **89**, 041601 (2014).
O. Lychkovskiy, Phys. Rev. A **89**, 033619 (2014).
O. Lychkovskiy, Phys. Rev. A **91**, 040101 (2015).
O. Lychkovskiy, O. Gamayun, and V. Cheianov, Phys. Rev. B **98**, 024307 (2018).
O. Gamayun, O. Lychkovskiy, E. Burovski, M. Malcomson, V. V. Cheianov, and M. B. Zvonarev, Phys. Rev. Lett. **120**, 220605 (2018).
F. Meinert, M. Knap, E. Kirilov, K. Jag-Lauber, M. B. Zvonarev, E. Demler, and H. C. Nägerl, Science **356**, 945 (2017).
C. J. Mathy, M. B. Zvonarev, and E. Demler, Nature Phys. **8**, 881 (2012).
M. Knap, C. J. Mathy, M. Ganahl, M. B. Zvonarev, and E. Demler, Phys. Rev. Lett. **112**, 015302 (2014).
L. Cao, V. Bolsinger, S. I. Mistakidis, G. M. Koutentakis, S. Kr[ö]{}nke, J. M. Schurer, and P. Schmelcher, J. Chem. Phys. **147**, 044106 (2017).
L. Cao, S. Kr[ö]{}nke, O. Vendrell, and P. Schmelcher, J. Chem. Phys. **139**, 134103 (2013).
A. Lampo, S. H. Lim, M. A. Garcia-March, and M. Lewenstein, Quantum **1**, 30 (2017).
K.K. Nielsen, L.A.P. Ardila, G.M. Bruun, and T. Pohl, New J. Phys. **21**, 043014 (2019).
S. I. Mistakidis, G. C. Katsimiga, P. G. Kevrekidis, and P. Schmelcher, New J. Phys. **20**, 043052 (2018).
J. Erdmann, S. I. Mistakidis, and P. Schmelcher, Phys. Rev. A **99**, 013605 (2019).
G. C. Katsimiga, S. I. Mistakidis, G. M. Koutentakis, P. G. Kevrekidis, and P. Schmelcher, Phys. Rev. A **98**, 013632 (2018).
J. Koepsell, J. Vijayan, P. Sompet, F. Grusdt, T. A. Hilker, E. Demler, G. Salomon, I. Bloch, and C. Gross, Nature **572**, 358 (2019).
M. Olshanii, Phys. Rev. Lett. **81**, 938 (1998).
T. M. Bersano, V. Gokhroo, M. A. Khamehchi, J. D’Ambroise, D. J. Frantzeskakis, P. Engels, and P. G. Kevrekidis, Phys. Rev. Lett. **120**, 063202 (2018).
A. Bergschneider, V. M. Klinkhamer, J. H. Becher, R. Klemt, G. Z[ü]{}rn, P. M. Preiss, and S. Jochim, Phys. Rev. A [**97**]{}, 063613 (2018).
M. Egorov, B. Opanchuk, P. Drummond, B. V. Hall, P. Hannaford, and A. I. Sidorov, Phys. Rev. A **87**, 053614 (2013).
X. Du, L. Luo, B. Clancy, and J. E. Thomas, Phys. Rev. Lett. **101**, 150401 (2008).
R. Horodecki, P. Horodecki, M. Horodecki, and K. Horodecki, Rev. Mod. Phys. **81**, 865 (2009).
M. Roncaglia, A. Montorsi, and M. Genovese, Phys. Rev. A **90**, 062303 (2014).
F. Köhler, K. Keiler, S. I. Mistakidis, H. D. Meyer, and P. Schmelcher, J. Chem. Phys. **151**, 054108 (2019).
J. Frenkel, [*in Wave Mechanics*]{} 1st ed. (Clarendon Press, Oxford, 1934), pp. 423-428.
P. A. Dirac, Proc. Camb. Phil. Soc. **26**, 376, Cambridge University Press (1930).
C.J. Pethick, and H. Smith, Bose-Einstein condensation in dilute gases. Cambridge University Press 2002.
J. Catani, G. Lamporesi, D. Naik, M. Gring, M. Inguscio, F. Minardi, A. Kantian, and T. Giamarchi, Phys. Rev. A [**85**]{}, 023623 (2012).
M. Hohmann, F. Kindermann, B. Gänger, T. Lausch, D. Mayer, F. Schmidt, and A. Widera, EPJ Quantum Technology , 23 (2015).
N. Spethmann, F. Kindermann, S. John, C. Weber, D. Meschede, and A. Widera, Phys. Rev. Lett. **109**, 235301 (2012).
D. Mayer, F. Schmidt, D. Adam, S. Haupt, J. Koch, T. Lausch, J. Nettersheim1, Q. Bouton, and A. Widera, J. Phys. B: At. Mol. and Opt. Phys. **52**, 015301 (2018).
H. Kiehn, S. I. Mistakidis, G. C. Katsimiga, and P. Schmelcher, Phys. Rev. A **100**, 023613 (2019).
R. Carretero-González, P. G. Kevrekidis, D. J. Frantzeskakis, B. A. Malomed, S. Nandi, and A. R. Bishop, Mathematics and Computers in Simulation, **74**, 361 (2007).
T. Winiecki, J. F. McCann, and C. S. Adams, Phys. Rev. Lett. **82**, 5186 (1999).
J. T. Devreese, Fr" ohlich Polarons. Lecture course including detailed theoretical derivations. arXiv:**1611.06122** (2016).
K. Sakmann, and M. Kasevich, Nature Phys. **12**, 451 (2016).
G. C. Katsimiga, S. I. Mistakidis, G. M. Koutentakis, P. G. Kevrekidis, and P. Schmelcher, New J. Phys. **19**, 123012 (2017).
K. Sakmann, A. I. Streltsov, O. E. Alon, and L. S. Cederbaum, Phys. Rev. A **78**, 023615 (2008).
G. M. Koutentakis, S. I. Mistakidis, and P. Schmelcher, New J. Phys. **21**, 053005 (2019).
H. Tajima, and S. Uchino, New J. Phys. **20**, 073048 (2018).
W. E. Liu, J. Levinsen, and M. M. Parish, Phys. Rev. Lett. **122**, 205301 (2019).
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- |
Yuval Grossman and Matthias Neubert[^1]\
Stanford Linear Accelerator Center, Stanford University\
Stanford, California 94309, U.S.A.\
E-mail: ,
- |
Alexander L. Kagan\
Department of Physics, University of Cincinnati\
Cincinnati, Ohio 45221, U.S.A.\
E-mail:
title: Trojan Penguins and Isospin Violation in Hadronic B Decays
---
Introduction
============
The study of rare decay processes is an important tool in testing the fundamental interactions among elementary particles, exploring the origin of CP violation, and searching for New Physics beyond the Standard Model (SM). Such processes have been explored in great detail, both theoretically and in experimental searches, in the weak interactions of kaons and $B$ mesons, as well as in $K$–$\bar K$, $D$–$\bar D$, and $B$–$\bar B$ mixing. Two prominent examples are the evidence for mixing-induced CP violating in the decay $B\to J/\psi\,K_S$ reported by the CDF Collaboration [@CDF], and the observation of direct CP violation in the decays $K\to\pi\pi$ [@NA31; @E731], which has recently been confirmed by the KTeV and NA48 Collaborations [@KTeV; @NA48].
Flavor-changing neutral current (FCNC) processes, which are forbidden at tree level in the SM, are especially sensitive to any new source of flavor-violating interactions. Already, the absence of experimental signals for new FCNC couplings puts stringent bounds on the parameters of many extensions of the SM such as Supersymmetry (SUSY) [@Yuval]. So far, FCNC processes have been explored mainly in particle–antiparticle mixing and in “semihadronic” weak decays, which permit a clean theoretical description. In the kaon system, examples of the latter type are the decays $K\to\pi\,l^+l^-$ and $K\to\pi\,\nu\bar\nu$. In the $B$ system, the decays that have received the most attention are $B\to X_s\,\gamma$, $B\to X_s\,l^+l^-$ and $B\to X_s\,\nu\bar\nu$, where $X_s$ can be any final state, exclusive or inclusive, containing a strange quark [@BaBar].
In the present paper, we explore in detail how New Physics could affect purely hadronic decays such as $B\to\pi K$, which are sensitive to isospin- or, more generally, SU(3) flavor-violating interactions. In the SM, the main contributions to the decay amplitudes for these processes come from the penguin-induced FCNC transition $\bar b\to\bar s q\bar q$, which by far exceeds a small, Cabibbo-suppressed $\bar b\to\bar u u\bar s$ contribution from $W$-boson exchange. Because of a fortunate interplay of isospin, Fierz and flavor symmetries, the theoretical description of the charged decays $B^\pm\to\pi K$ in the SM is clean despite the fact that these are exclusive nonleptonic decays [@NR; @me]. Isospin violation arises through the small charged-current contribution and through the electroweak penguin operators in the low-energy effective weak Hamiltonian [@Heff]. In the SM, these operators are induced by penguin and box diagrams involving the exchange of weakly interacting $W$ and $Z$ bosons, or of a photon. Here we point out that in a large class of New Physics models such effects can be mediated by “trojan” electroweak penguins, which are neither (pure) penguins nor of electroweak origin. Nevertheless, at low energies their effects are parameterized by an extension of the usual basis of electroweak penguin operators. We will explore examples where trojan penguins are induced by tree-level couplings, e.g., models with an extra $Z'$ boson, or pure strong-interaction processes, e.g., gluino box diagrams in SUSY models.
In the present work we calculate the Wilson coefficients of the hadronic electroweak penguin operators in the effective weak Hamiltonian in an extended operator basis and for a large class of New Physics models. We then explore the phenomenological consequences of these new, isospin-violating contributions for weak-interaction observables. This serves two purposes: first, it allows us to derive bounds on New Physics parameters, which in some cases improve upon existing bounds derived from other processes; secondly, it shows which observables may be interesting to look at as far as searches for New Physics are concerned. We shall address both issues in detail for the particular case of $B^\pm\to\pi K$ decays, which in the context of the SM are of prime importance in determining the weak phase $\gamma=\mbox{arg}(V_{ub}^*)$ of the Cabibbo–Kobayashi–Maskawa (CKM) matrix [@NR; @me; @FM; @Robert; @mat; @Frank]. We stress, however, that trojan penguins could be important in a much wider class of processes. In particular, they may be responsible for a large contribution to the quantity $\epsilon'/\epsilon$ measuring direct CP violation in $K\to\pi\pi$ decays [@epspr; @eps1]. The importance of $B^\pm\to\pi K$ decays in the search for New Physics has been emphasized in [@me; @Mati], and some specific scenarios containing new isospin-violating contributions have been explored in [@CDK; @anom].
In Section \[sec:ops\], we discuss the effective weak Hamiltonian relevant to hadronic $B$ decays. In the presence of generic New Physics contributions, the basis of penguin operators has to be extended from the standard one in several aspects. We discuss the structure of the new operators and their scaling properties under a renormalization-group transformation from a high scale down to low energies. The theory of the rare hadronic decays $B^\pm\to\pi K$ is discussed in Section \[sec:BpiK\], where we indicate how various observables in these decays can be used to test for physics beyond the SM. We derive model-independent bounds on a ratio $R_*$ of CP-averaged $B^\pm\to\pi K$ branching ratios in the presence of New Physics, and show that the value $\gamma_{\pi K}$ of the weak phase extracted in $B\to\pi K$ decays is extremely sensitive to isospin-violating New Physics contributions. Because the theoretical analysis has only small hadronic uncertainties, potential New Physics effects can be detected even if they are 50% smaller than the isospin-violating contributions present in the SM. In Sections \[sec:tree\] and \[sec:loop\], we calculate the Wilson coefficients of the $\bar b\to\bar s q\bar q$ penguin operators in a large class of extensions of the SM, including models with tree-level FCNC couplings of the $Z$ boson, extended gauge models, multi-Higgs models, and SUSY models with and without R-parity conservation. In each case, we explore which region of parameter space can be probed by measuring certain $B^\pm\to\pi K$ observables, and how big a departure from the SM predictions one can expect under realistic circumstances. Section \[sec:concl\] contains a summary of our results and the conclusions.
Effective Hamiltonian for hadronic FCNC processes {#sec:ops}
=================================================
The effective weak Hamiltonians relevant to rare hadronic $B$ decays based on the quark transitions $\bar b\to\bar s q\bar q$ or $\bar b\to\bar d q\bar q$ have been discussed extensively in the literature. Here we consider only the first case, adopting the notations of [@Heff]. A similar discussion (with obvious replacements of indices) would apply to the other case. In the SM, the result can be written in the compact form $$\label{Heff}
{\cal H}_{\rm eff} = \frac{G_F}{\sqrt2}\,\sum_{q=u,c}
\lambda_q\,\bigg[ \sum_{i=1,2} C_i(\mu)\,Q_i^q(\mu)
+ \sum_{i=3\dots 10} C_i(\mu)\,Q_i(\mu)
+ C_{8g}(\mu)\,Q_{8g}(\mu) \bigg] \,,$$ where $\lambda_q=V_{qb}^* V_{qs}$ are combinations of CKM matrix elements obeying the unitarity relation $\lambda_u+\lambda_c+\lambda_t=0$, and $Q_i$ are local operators containing quark and gluon fields. Specifically, $$\label{curcur}
Q_1^q = (\bar b_\alpha q_\beta)_{V-A}\,
(\bar q_\beta s_\alpha)_{V-A} \,, \qquad
Q_2^q = (\bar b_\alpha q_\alpha)_{V-A}\,
(\bar q_\beta s_\beta)_{V-A} \,,$$ summed over color indices $\alpha$ and $\beta$, are the usual current–current operators induced by $W$-boson exchange, $$\begin{aligned}
\label{QCDp}
Q_3 &=& (\bar b_\alpha s_\alpha)_{V-A}\,\sum_q\,
(\bar q_\beta q_\beta)_{V-A} \,, \nonumber\\
Q_4 &=& (\bar b_\alpha s_\beta)_{V-A}\,\sum_q\,
(\bar q_\beta q_\alpha)_{V-A} \,, \nonumber\\
Q_5 &=& (\bar b_\alpha s_\alpha)_{V-A}\,\sum_q\,
(\bar q_\beta q_\beta)_{V+A} \,, \nonumber\\
Q_6 &=& (\bar b_\alpha s_\beta)_{V-A}\,\sum_q\,
(\bar q_\beta q_\alpha)_{V+A} \,,\end{aligned}$$ summed over the light flavors $q=u,d,s,c,b$, are referred to as QCD penguin operators, and $$\begin{aligned}
\label{EWp}
Q_7 &=& \frac32\,(\bar b_\alpha s_\alpha)_{V-A}\,\sum_q\,e_q\,
(\bar q_\beta q_\beta)_{V+A} \,, \nonumber\\
Q_8 &=& \frac32\,(\bar b_\alpha s_\beta)_{V-A}\,\sum_q\,e_q\,
(\bar q_\beta q_\alpha)_{V+A} \,, \nonumber\\
Q_9 &=& \frac32\,(\bar b_\alpha s_\alpha)_{V-A}\,\sum_q\,e_q\,
(\bar q_\beta q_\beta)_{V-A} \,, \nonumber\\
Q_{10} &=& \frac32\,(\bar b_\alpha s_\beta)_{V-A}\,\sum_q\,e_q\,
(\bar q_\beta q_\alpha)_{V-A} \,,\end{aligned}$$ with $e_q$ denoting the electric charges of the quarks, are called electroweak penguin operators. The notation $(\bar q_1 q_2)_{V\pm A}$ implies $\bar q_1\gamma^\mu
(1\pm\gamma_5)q_2$. The terminology of QCD and electroweak penguins is slightly misleading insofar as the Wilson coefficients of the QCD penguin operators also receive small contributions from electroweak penguin and box diagrams. However, in the SM there are no strong-interaction contributions to the coefficients of the electroweak penguin operators. The operator $Q_{8g}
=\frac{g_s m_b}{8\pi^2}\,\bar b\,\sigma^{\mu\nu}(1-\gamma_5)
G_{\mu\nu}s$ in (\[Heff\]) is the chromomagnetic dipole operator. The analogous electromagnetic dipole operator and semileptonic operators containing products of a quark current with a lepton current can be safely discarded from the effective Hamiltonian for hadronic $B$ decays.
In general, physics beyond the SM can induce a much larger set of penguin operators, and it is therefore unavoidable for our purposes to generalize the standard nomenclature reviewed above. However, for all the models we explore below it is sufficient to consider products of vector and/or axial vector currents only. We define a basis of such operators by $$\begin{aligned}
\label{basis}
O_1^q &=& (\bar b_\alpha s_\alpha)_{V-A}\,
(\bar q_\beta q_\beta)_{V+A} \,, \qquad
O_2^q = (\bar b_\alpha s_\beta)_{V-A}\,
(\bar q_\beta q_\alpha)_{V+A} \,, \nonumber\\
O_3^q &=& (\bar b_\alpha s_\alpha)_{V-A}\,
(\bar q_\beta q_\beta)_{V-A} \,, \qquad
O_4^q = (\bar b_\alpha s_\beta)_{V-A}\,
(\bar q_\beta q_\alpha)_{V-A} \,, \nonumber\\
O_5^q &=& (\bar b_\alpha q_\alpha)_{V-A}\,
(\bar q_\beta s_\beta)_{V+A} \,, \qquad
O_6^q = (\bar b_\alpha q_\beta)_{V-A}\,
(\bar q_\beta s_\alpha)_{V+A} \,,\end{aligned}$$ and denote their Wilson coefficients by $c_i^q$. We implicitly assume a regularization scheme that preserves Fierz identities. In a general model, we also need operators of opposite chirality compared to the ones shown above, i.e., with $V-A\leftrightarrow
V+A$ everywhere. We denote these operators by $\widetilde O_i^q$ and their coefficients by $\widetilde c_i^q$. Thus, the most general penguin Hamiltonian considered in this paper takes the form $$\label{Hpeng}
{\cal H}_{\rm peng} = \frac{G_F}{\sqrt 2}\,\sum_i
\sum_q \left[ c_i^q(\mu)\,O_i^q(\mu)
+ \widetilde c_i^q(\mu)\,\widetilde O_i^q(\mu) \right] .$$ To this, one has to add the current–current operators in (\[curcur\]) and the chromo-magnetic dipole operator. It is implicitly understood that for the cases where $q=s$ or $b$ the operators $O_5^q$ and $O_6^q$ as well as $\widetilde O_5^q$ and $\widetilde O_6^q$ are omitted from the list of operators, because they are Fierz-equivalent to the remaining ones.
The QCD and electroweak penguin operators present in the effective weak Hamiltonian of the SM in(\[Heff\]) are linear combinations of the four operators $O_{1\dots 4}^q$. However, in a general model there may be additional penguin operators built out of $O_5^q$, $O_6^q$, and of the opposite-chirality operators $\widetilde O_i^q$. Also, there may be operators which cannot be represented as linear combinations of QCD and electroweak penguin operators as shown in (\[QCDp\]) and (\[EWp\]), because in a general model the flavor symmetry among up- or down-type quarks can be violated. An example would be operators with flavor content $\bar sb(\bar d d-\bar s s)$, which are absent in the SM. Because such operators are of no relevance to our discussion they will not be explored any further here. For completeness, we show how the Wilson coefficients $C_{3\dots 10}$ of the SM penguin operators in (\[Heff\]) can be expressed in terms of the coefficients $c_i^q$. Defining the linear combinations $$\label{cdefs}
c_i^{\rm QCD} \equiv \frac{c_i^u+2c_i^d}{3} \,, \qquad
c_i^{\rm EW} \equiv c_i^u-c_i^d \,,$$ we have $$\begin{aligned}
-\lambda_t\,C_{3,4} &=& c_{3,4}^{\rm QCD} \,, \qquad
-\lambda_t\,C_{5,6} = c_{1,2}^{\rm QCD} \,,
\nonumber\\
-\frac32\,\lambda_t\,C_i &=& c_{i-6}^{\rm EW} \,;
\quad i=7,\dots,10 \,.\end{aligned}$$
The isospin-violating effects induced by the coefficients $c_i^{\rm EW}$ and $\widetilde c_i^{\rm EW}$ are the main focus of this paper. In the SM, the matching conditions for the corresponding electroweak penguin coefficients at the weak scale, and to leading order in perturbation theory, are $C_8(m_W)=C_{10}(m_W)=0$, $C_7(m_W)\approx 0$, and [@Bur93] $$\label{C9match}
C_9(m_W) \approx -\frac{\alpha}{12\pi}\,
\frac{x_t}{\sin^2\!\theta_W} \left( 1 + \frac{3\ln x_t}{x_t-1}
\right) ,$$ where $x_t=(m_t/m_W)^2$. For simplicity, we show only the large electroweak contribution to $C_9(m_W)$ and omit a common, renormalization-scheme dependent electromagnetic contribution to $C_7(m_W)$ and $C_9(m_W)$, which is negligible compared with the contribution in (\[C9match\]). In this approximation, the SM matching conditions for the electroweak penguin coefficients in our basis read $$\begin{aligned}
\label{SMinit}
c_3^{\rm EW,SM}(m_W) &=& \frac{\alpha}{8\pi}\,
\frac{\lambda_t\,x_t}{\sin^2\!\theta_W} \left( 1
+ \frac{3\ln x_t}{x_t-1} \right)
\approx -5.5\times 10^{-4} \,, \nonumber\\
c_{i\ne 3}^{\rm EW,SM}(m_W)
&=& \widetilde c_i^{\rm EW,SM}(m_W) = 0 \,.\end{aligned}$$ For the numerical estimate we have used $m_t=\overline{m}_t(m_t)
=170$GeV, $\alpha=1/129$, and $\lambda_t=V_{tb}^* V_{ts}=-0.04$.
In Sections \[sec:tree\] and \[sec:loop\], we will calculate the Wilson coefficients $c_i^q$ at a high scale, which for simplicity will be identified with the electroweak scale. In phenomenological applications, however, one usually prefers working with coefficients renormalized at a low scale of order $m_b$. The two sets of coefficients are connected by a renormalization-group transformation [@Heff]. Here we discuss the QCD evolution of the electroweak penguin operators at the leading logarithmic order, neglecting next-to-leading corrections as well as QED corrections to the anomalous dimensions of the operators. Under QCD evolution, flavor-nonsinglet combinations of penguin operators mix into flavor-singlet combinations through diagrams in which two light quarks annihilate into a gluon, which then fragments into a pair of light quarks. However, there is no mixing of flavor-singlet operators into flavor-nonsinglet ones. If we restrict ourselves to flavor-nonsinglet combinations of the operators $O_i^q$, each pair in the three lines in (\[basis\]) obeys a separate matrix evolution equation. It follows that the coefficients $c_i^{\rm EW}$ mix pairwise under renormalization. For each pair of coefficients associated with $(V\mp A)\otimes(V\pm A)$ operators we obtain $$\begin{aligned}
\label{RGE-R}
c_1^{\rm EW}(\mu) &=& \kappa^{-3/23}\,c_1^{\rm EW}(m_W) \,,
\nonumber\\
c_2^{\rm EW}(\mu) &=& \frac{\kappa^{24/23}-\kappa^{-3/23}}{3}\,
c_1^{\rm EW}(m_W) + \kappa^{24/23}\,c_2^{\rm EW}(m_W) \,,\end{aligned}$$ and similarly for $(c_5^{\rm EW},c_6^{\rm EW})$, where $\kappa=\alpha_s(\mu)/\alpha_s(m_W)$. For a pair of coefficients associated with $(V\mp A)\otimes(V\mp A)$ operators we obtain instead $$\begin{aligned}
\label{RGE-L}
c_3^{\rm EW}(\mu) &=& \phantom{-}
\frac{\kappa^{12/23}+\kappa^{-6/23}}{2}\,c_3^{\rm EW}(m_W)
- \frac{\kappa^{12/23}-\kappa^{-6/23}}{2}\,
c_4^{\rm EW}(m_W) \,, \nonumber\\
c_4^{\rm EW}(\mu) &=& -\frac{\kappa^{12/23}-\kappa^{-6/23}}{2}\,
c_3^{\rm EW}(m_W) + \frac{\kappa^{12/23}+\kappa^{-6/23}}{2}\,
c_4^{\rm EW}(m_W) \,.\end{aligned}$$ The coefficients $\widetilde c_i^q$ scale in the same way as the $c_i^q$. Since our main focus is on electroweak penguins and their generalizations beyond the SM, we will not discuss the more complicated evolution equations for the coefficients $c_i^q$ themselves, which can however readily be deduced from the literature [@Heff].
Searching for New Physics with $B^\pm\to\pi K$ decays {#sec:BpiK}
=======================================================
In close correspondence with the different types of operators in the effective weak Hamiltonian, one distinguishes three classes of flavor topologies relevant to $B\to\pi K$ decays, referred to as trees, QCD penguins and electroweak penguins. In the SM, the weak couplings associated with these topologies are known. From the measured branching ratios for the various $B\to\pi K$ decay modes it follows that the QCD penguins dominate the decay amplitudes [@Digh], whereas trees and electroweak penguins are subleading and of a similar strength [@oldDesh]. The theoretical description of the two charged modes $B^\pm\to\pi^\pm K^0$ and $B^\pm\to\pi^0 K^\pm$ exploits the fact that the amplitudes for these processes differ in a pure isospin amplitude $A_{3/2}$, defined as the matrix element of the isovector part of the effective Hamiltonian between a $B$ meson and the $\pi K$ isospin eigenstate with $I=\frac 32$. In the SM the parameters of this amplitude are determined, up to an overall strong-interaction phase $\phi$, in the limit of SU(3) flavor symmetry [@NR]. SU(3)-breaking corrections can be calculated in the factorization approximation [@Stech], so that theoretical uncertainties enter only at the level of nonfactorizable SU(3)-breaking corrections to a subleading decay amplitude. Moreover, it has recently been shown that even these nonfactorizable corrections can be calculated in a model-independent way up to terms that are power suppressed in $\Lambda/m_b$ and vanish in the heavy-quark limit [@fact].
General parametrization of the decay amplitudes
-----------------------------------------------
In the presence of New Physics, the analysis of $B^\pm\to\pi K$ decays becomes more complicated. A convenient and completely general parametrization of the two decay amplitudes is $$\begin{aligned}
\label{ampls}
{\cal A}(B^+\to\pi^+ K^0) &=& P\,(1-i\rho\,e^{i\phi_\rho})
\,,\nonumber\\
-\sqrt2\,{\cal A}(B^+\to\pi^0 K^+) &=& P \Big[
1-i\rho\,e^{i\phi_\rho} - \varepsilon_{3/2}\,e^{i\phi}
(e^{i\gamma} - a\,e^{i\phi_a} - ib\,e^{i\phi_b}) \Big] \,,\end{aligned}$$ where $P$ is the dominant penguin amplitude defined as the sum of all CP-conserving terms in the $B^\pm\to\pi^\pm K^0$ decay amplitudes, $\varepsilon_{3/2}$, $\rho$, $a$, $b$ are real hadronic parameters, and $\phi$, $\phi_\rho$, $\phi_a$, $\phi_b$ are strong-interaction phases. The weak phase $\gamma=\mbox{arg}(V_{ub}^*)$ and the terms $i\rho$ and $ib$ change sign under a CP transformation, whereas all other parameters stay invariant. The terms proportional to $\varepsilon_{3/2}$ in (\[ampls\]) parameterize the isospin amplitude $A_{3/2}$. The contribution proportional to $e^{i\gamma}$ comes from the matrix elements of the current–current operators $Q_1^u$ and $Q_2^u$ in the effective Hamiltonian, which mediate the tree process $\bar b\to\bar u u\bar s$. The quantities $a$ and $b$ parameterize the effects of electroweak penguins. It is crucial that only isospin-violating terms can contribute to the amplitude $A_{3/2}$. All isospin-conserving contributions reside in $P$ and $\rho$.
Let us discuss the various terms entering the decay amplitudes in detail. The parameter $\varepsilon_{3/2}$ characterizes the relative strength of tree and QCD penguin contributions. Information about it can be derived by using SU(3) flavor symmetry to relate the tree contribution to the isospin amplitude $A_{3/2}$ to the corresponding contribution in the decay $B^+\to\pi^+\pi^0$. Since the final state $\pi^+\pi^0$ has isospin $I=2$ (because of Bose symmetry), the amplitude for the latter process does not receive any contribution from QCD penguins. Moreover, in the SM electroweak penguins in $\bar b\to\bar d q\bar q$ transitions are negligible, and thus only the tree topology contributes to the $B^+\to\pi^+\pi^0$ decay amplitude. In our analysis we make the plausible assumption that potential New Physics contributions to this amplitude can be neglected.[^2] Even if new electroweak penguin effects would be of comparable strength in $\bar b\to\bar s q\bar q$ and $\bar b\to\bar d q\bar q$ transitions, the latter would have to compete with a Cabibbo-enhanced tree amplitude in order to be significant in $B^+\to\pi^+\pi^0$ decays. We then find that [@NR; @me] $$\label{eps}
\bar\varepsilon_{3/2}
\equiv \frac{\varepsilon_{3/2}}{\sqrt{1+\rho^2}}
= \sqrt2\,R_{\rm SU(3)} \left|\frac{V_{us}}{V_{ud}}\right|
\left[
\frac{\mbox{B}(B^+\to\pi^+\pi^0)+\mbox{B}(B^-\to\pi^-\pi^0)}
{\mbox{B}(B^+\to\pi^+ K^0)+\mbox{B}(B^-\to\pi^-\bar K^0)}
\right]^{1/2} .$$ SU(3)-breaking corrections are described by the factor $R_{\rm SU(3)}=1.22\pm 0.05$, which can be calculated in a model-independent way using the QCD factorization theorem of [@fact]. The quoted error is an estimate of the theoretical uncertainty due to uncontrollable corrections of $O(\frac{1}{N_c}\frac{m_s}{m_b})$. Using preliminary data reported by the CLEO Collaboration [@CLEO] to evaluate the ratio of branching ratios in (\[eps\]), we obtain $$\label{epsval}
\bar\varepsilon_{3/2} = 0.21\pm 0.06_{\rm exp}
\pm 0.01_{\rm th} \,.$$ With a better measurement of the branching ratios the uncertainty in $\bar\varepsilon_{3/2}$ will be reduced significantly.
The parameter $\rho$ in (\[ampls\]) parameterizes the sum of all CP-violating contributions to the $B^+\to\pi^+ K^0$ decay amplitude. In the presence of New Physics, those contributions could arise from QCD as well as electroweak penguin operators. Note that the CP-conserving part of such terms is absorbed, by definition, into the quantity $P$. We will not attempt a theoretical calculation of this quantity (which is difficult even in the SM) and only consider observables that are independent of $P$. In the SM, $\rho\simeq\varepsilon_a\sin\gamma$ [@me] describes a small contribution induced by final-state rescattering from tree or annihilation diagrams [@Blok97; @BFM98; @Ge97; @Ne97; @Fa97; @At97]. In the heavy-quark limit, the parameter $\varepsilon_a$ can be calculated and is found to be of order $-2\%$ [@fact].
Finally, in the SM the parameter $b$ vanishes, while $$\label{delval}
a\,e^{i\phi_a} = \delta_{\rm EW}
= (0.64\pm 0.09)\times\frac{0.085}{|V_{ub}/V_{cb}|}$$ is calculable in terms of fundamental parameters [@NR; @me; @Fl96]. Up to some small SU(3)-breaking corrections, $\delta_{\rm EW}$ is given by the Wilson coefficient $c_3^{\rm EW}(m_W)$ in (\[SMinit\]) divided by $-|\lambda_u|$. There are no additional hadronic uncertainties in this estimate in the SM. In particular, the strong-interaction phase $\phi_a$ is bounded to be less than a few degrees and can be neglected for all practical purposes [@me]. In a general model, the parameters $a$ and $b$ depend on the values of the penguin coefficients $c_i^{\rm EW}$ and $\widetilde c_i^{\rm EW}$ as well as on the hadronic matrix elements of the corresponding operators evaluated between a $B$ meson and the $\pi K$ isospin state with $I=\frac32$. Since our intention here is to look for New Physics effects rather than doing precision calculations, it will be sufficient for our purposes to evaluate these matrix elements in a given New Physics scenario using the naive factorization approximation and neglecting small SU(3)-breaking effects.[^3] Then the strong-interaction phases $\phi_a$ and $\phi_b$ vanish, and we obtain $$\begin{aligned}
\label{abres}
|\lambda_u|\,(a+ib) &\approx&
- \left( \bar c_3^{\rm EW} + \bar c_4^{\rm EW} \right)
+ \frac23\,\kappa^{3/23}
\left( \bar c_1^{\rm EW} + \bar c_5^{\rm EW} \right)
\nonumber\\
&&\mbox{}- \frac{3\chi-1}{4}\,\kappa^{30/23}
\bigg( \bar c_2^{\rm EW} + \frac13\,\bar c_1^{\rm EW}
+ \bar c_6^{\rm EW} + \frac13\,\bar c_5^{\rm EW} \bigg) \,,\end{aligned}$$ where $\bar c_i^{\rm EW}\equiv c_i^{\rm EW}(m_W)
-\widetilde c_i^{\rm EW}(m_W)$, and $$\chi = \frac{2 m_K^2}{(m_s+m_d)\,m_b}
= \frac{2 m_\pi^2}{(m_u+m_d)\,m_b} \,.$$ Parity invariance implies that the relevant hadronic matrix elements of the operators $\widetilde O_i$ have the opposite sign compared with those of the operators $O_i$. The fact that $\bar c_3^{\rm EW}$ and $\bar c_4^{\rm EW}$ enter with the coefficient $-1$ in (\[abres\]) is a model-independent result free of hadronic uncertainties, irrespective of whether these coefficients receive New Physics contributions or not. It follows because the isovector components of the penguin operators $O_3^q$ and $O_4^q$ can be related to the usual current–current operators by a Fierz transformation [@Ne97; @Fl96]. For the numerical analysis we choose the renormalization scale $\mu=m_b$ and take $\kappa=\alpha_s(m_b)/\alpha_s(m_W)=1.83$ and $\chi=1.18$, yielding $$\label{abformula}
|\lambda_u|\,(a+ib) \approx
- (\bar c_3^{\rm EW} + \bar c_4^{\rm EW})
+ 0.26 (\bar c_1^{\rm EW} + \bar c_5^{\rm EW})
- 1.40 (\bar c_2^{\rm EW} + \bar c_6^{\rm EW}) \,.$$
Next we consider the New Physics contributions to the parameter $\rho$. In the factorization approximation, we find $$\label{rhobound}
\frac{\rho}{\sqrt{1+\rho^2}}
\approx \frac{3\bar\varepsilon_{3/2}}{4|\lambda_u|}\,
\mbox{Im}\Big[
(\bar c_5^d - \bar c_4^d - \chi\,\bar c_2^d)
+ \frac13\,(\bar c_6^d - \bar c_3^d - \chi\,\bar c_1^d)
- a_{8g}\,C_{8g} \Big] \,,$$ where $\bar c_i^d\equiv c_i^d(m_b)-\widetilde c_i^d(m_b)$. Because the QCD evolution of the Wilson coefficients $c_i^d$ and $\widetilde c_i^d$ is complicated, we prefer to present the result in terms of coefficients renormalized at the scale $m_b$. Note that by pulling out a factor of $\bar\varepsilon_{3/2}$ on the right-hand side we avoid the difficulty of calculating the overall penguin amplitude $P$ in (\[ampls\]). The contribution of the chromomagnetic dipole operator is formally of next-to-leading order in $\alpha_s$ and thus could be dropped; however, we keep it because in some New Physics models the coefficient $C_{8g}$ can be enhanced with respect to its SM value by an order of magnitude [@glue1; @George; @Alex; @CGGi]. Using the QCD factorization approach of [@fact], we find that $$a_{8g} = \frac{2\alpha_s}{3\pi}\,(1+\chi)(-\lambda_t)
\approx 4\times 10^{-3} \,.$$ Note that the magnitude of the left-hand side in (\[rhobound\]) is bounded by unity. This implies a nontrivial upper bound on the possible CP-violating New Physics contributions to the penguin operators, given by $$\label{rhonumeric}
\left|\,\mbox{Im}\Big[
(\bar c_5^d - \bar c_4^d - \chi\,\bar c_2^d)
+ \frac13\,(\bar c_6^d - \bar c_3^d - \chi\,\bar c_1^d)
- a_{8g}\,C_{8g} \Big] \right| <
\frac{4|\lambda_u|}{3\bar\varepsilon_{3/2}} \,.$$ Using $|V_{ub}/V_{cb}|=0.085\pm 0.015$, corresponding to $|\lambda_u|=(7.5\pm 1.3)\times 10^{-4}$, and taking for $\bar\varepsilon_{3/2}$ the value in (\[epsval\]), we find that the right-hand side of this bound is less than $7.4\times 10^{-3}$ at 90% confidence level. For comparison, we note that in the SM the magnitude of the combination of penguin coefficients entering above is about $3\times 10^{-3}$. However, due to the smallness of the weak phase of $\lambda_t$ the imaginary part of this combination is much smaller.
Model-independent bounds on $R_*$ in the presence of New Physics
----------------------------------------------------------------
The most important observable in the exploration of New Physics effects in $B\to\pi K$ decays is the ratio of the CP-averaged branching ratios for the decays $B^\pm\to\pi^\pm K^0$ and $B^\pm\to\pi^0 K^\pm$, given by $$\label{Rstexp}
R_* =
\frac{\mbox{B}(B^+\to\pi^+ K^0)+\mbox{B}(B^-\to\pi^-\bar K^0)}
{2[\mbox{B}(B^+\to\pi^0 K^+)+\mbox{B}(B^-\to\pi^0 K^-)]}
= 0.75\pm 0.28 \,,$$ where the quoted experimental value is derived from data reported by the CLEO Collaboration [@CLEO]. It will often be convenient to consider a related quantity defined as $$\label{XRval}
X_R = \frac{\sqrt{R_*^{-1}}-1}{\bar\varepsilon_{3/2}}
= 0.72\pm 0.98_{\rm exp}\pm 0.03_{\rm th} \,.$$ Because of the theoretical factor $R_{\rm SU(3)}$ entering the definition of $\bar\varepsilon_{3/2}$ in (\[eps\]) this is, strictly speaking, not an observable. However, the irreducible theoretical uncertainty in $X_R$ is so much smaller than the present experimental error that it is justified to treat this quantity as an observable. The advantage of presenting our results in terms of $X_R$ rather than $R_*$ is that the leading dependence on $\bar\varepsilon_{3/2}$ cancels out (see below). Also, some experimental errors cancel in the ratio in (\[XRval\]) [@Frank].
When writing theoretical expressions for the quantities $R_*$ and $X_R$ we eliminate the two parameters $\varepsilon_{3/2}$ and $\rho$ in favor of the measurable parameter $\bar\varepsilon_{3/2}$ and a “weak phase” $\varphi\in[-90^\circ,90^\circ]$ defined by $$\sin\varphi = \frac{\rho}{\sqrt{1+\rho^2}} \,, \qquad
\cos\varphi = \frac{1}{\sqrt{1+\rho^2}} \,.$$ The most direct way of probing this phase is via the direct CP asymmetry in the decays $B^\pm\to\pi^\pm K^0$, which is given by $$A_{\rm CP}(\pi^+ K^0) =
\frac{\mbox{B}(B^+\to\pi^+ K^0)-\mbox{B}(B^-\to\pi^-\bar K^0)}
{\mbox{B}(B^+\to\pi^+ K^0)+\mbox{B}(B^-\to\pi^-\bar K^0)}
= \sin2\varphi\sin\phi_\rho \,.$$ In the SM, $\sin2\varphi\approx 2\varepsilon_a\sin\gamma$ is of order a few percent, and with realistic values for $\phi_\rho$ of order $10^\circ$–$20^\circ$ [@fact] one expects a very small CP asymmetry. However, in New Physics scenarios with new CP-violating couplings $\sin 2\varphi$ may be significantly larger than in the SM. An experimental finding that $A_{\rm CP}(\pi^+ K^0)=O(10\%)$ would constitute strong evidence for the existence of such an effect.
The exact theoretical expression for $R_*$ is $$\begin{aligned}
\label{Rstar}
R_*^{-1} &=& 1 + 2\bar\varepsilon_{3/2}\cos\varphi
\Big[ a\cos(\phi+\phi_a) - \cos\gamma\cos\phi \Big] \nonumber\\
&&\mbox{}- 2\bar\varepsilon_{3/2}\sin\varphi
\Big[ b\cos(\phi+\phi_b-\phi_\rho)
- \sin\gamma\cos(\phi-\phi_\rho) \Big] \nonumber\\
&&\mbox{}+ \bar\varepsilon_{3/2}^2 \left( 1
- 2a\cos\gamma\cos\phi_a - 2b\sin\gamma\cos\phi_b
+ a^2 + b^2 \right) .\end{aligned}$$ In the SM $b=0$, $a=\delta_{\rm EW}$, $\phi_a=0$, and $\varphi\approx 0$ to very good approximation. Therefore $$\begin{aligned}
\label{Rstbound}
R_*^{-1} &=& 1 + 2\bar\varepsilon_{3/2}\,
(\delta_{\rm EW}-\cos\gamma)\cos\phi
+ \bar\varepsilon_{3/2}^2
(1 - 2\delta_{\rm EW}\cos\gamma + \delta_{\rm EW}^2)
\nonumber\\
&\le& \left( 1 + \bar\varepsilon_{3/2}\,
|\delta_{\rm EW}-\cos\gamma| \right)^2
+ \bar\varepsilon_{3/2}^2\sin^2\!\gamma \,.\end{aligned}$$ In the second step we have used the fact that $|\cos\phi|\le 1$ to obtain an upper bound on $R_*^{-1}$. Similarly, a lower bound can be derived, which is obtained by changing the sign of $\bar\varepsilon_{3/2}$ in the above inequality. These bounds imply nontrivial constraints on $\cos\gamma$ provided that $R_*$ differs from 1 by a significant amount. In Figure \[fig:SMbound\], we show the resulting lower and upper bounds on the quantity $X_R$ versus $\gamma$, obtained by scanning the input parameters in the intervals $0.15<\bar\varepsilon_{3/2}
<0.27$ and $0.49<\delta_{\rm EW}<0.79$. The latter value is obtained by using $|V_{ub}/V_{cb}|=0.085\pm 0.015$ in (\[delval\]). The dependence on the variation of $\bar\varepsilon_{3/2}$ is so small that it would almost be invisible on the scale of the plot. Note that the extremal values $R_*$ can take in the SM are such that $|X_R|\le(1+\delta_{\rm EW})$ irrespective of the value of $\gamma$. A value exceeding this limit would be a clear signal for New Physics [@me; @Mati]. In view of the present large error on $X_R$, this is still a realistic possibility. Because the upper and lower bounds on $X_R$ are, to
a very good approximation, symmetric around $X_R=0$, we will from now on only show the upper bounds, since $X_R>0$ is the region favored by experiment. Also, since the bounds change little under variation of $\delta_{\rm EW}$ and $\bar\varepsilon_{3/2}$, we will work with the central values $\delta_{\rm EW}=0.64$ and $\bar\varepsilon_{3/2}=0.21$. By the time the experimental data will be sufficiently precise to perform the New Physics searches proposed in this work, the errors on these parameters are likely to be reduced by a significant amount.
Let us first discuss the case where New Physics induces arbitrary CP-violating contributions to the $B\to\pi K$ decay amplitudes, while preserving isospin symmetry. Then the only change with respect to the SM would be that the weak phase $\varphi$ may no longer be negligible. We obtain $$\begin{aligned}
\label{Rstphi}
R_*^{-1} &=& 1 + 2\bar\varepsilon_{3/2} \Big[
\cos\varphi\,(\delta_{\rm EW}-\cos\gamma)\cos\phi
+ \sin\varphi\sin\gamma\cos(\phi-\phi_\rho) \Big] \nonumber\\
&&\mbox{}+ \bar\varepsilon_{3/2}^2
(1 - 2\delta_{\rm EW}\cos\gamma + \delta_{\rm EW}^2)
\nonumber\\
&\le& 1 + 2\bar\varepsilon_{3/2} \Big[
\cos\varphi\,|\delta_{\rm EW}-\cos\gamma|
+ |\sin\varphi\sin\gamma| \Big] \nonumber\\
&&\mbox{}+ \bar\varepsilon_{3/2}^2
(1 - 2\delta_{\rm EW}\cos\gamma + \delta_{\rm EW}^2)
\nonumber\\
&\le& \left( 1 + \bar\varepsilon_{3/2}
\sqrt{1 - 2\delta_{\rm EW}\cos\gamma + \delta_{\rm EW}^2}
\right)^2 \,.\end{aligned}$$ In deriving the upper bounds we have varied the strong-interaction phases $\phi$ and $\phi_\rho$ independently. Analogous lower bounds are obtained as previously by changing the sign of $\bar\varepsilon_{3/2}$. The last inequality in (\[Rstphi\]) is remarkable in that it holds for arbitrary isospin-conserving New Physics effects no matter how large they are. The corresponding bound on $|X_R|$ is $$\label{phiarb}
|X_R| \le
\sqrt{1 - 2\delta_{\rm EW}\cos\gamma + \delta_{\rm EW}^2}
\le 1+\delta_{\rm EW} \,.$$ Note that the extremal value is the same as in the SM, i.e., isospin-conserving New Physics effects cannot lead to a value of $|X_R|$ exceeding $1+\delta_{\rm EW}$. In the left-hand plot in Figure \[fig:case12\] we show the upper bound on $X_R$ versus $\gamma$ in the SM, and for New Physics scenarios with different values of $\varphi$. The three choices of $\varphi$ shown correspond to $|\rho|\approx 0.27$, 0.58 and 1. The gray curve shows the upper bound obtained by varying $\varphi$. We observe that isospin-conserving New Physics can enhance the value of $X_R$ relative to the SM, but only by a moderate amount.
Next we consider New Physics effects that violate isospin symmetry, but we first restrict ourselves to the important subclass of models which do not contain significant new CP-violating phases. Then $\varphi$ and $b$ still vanish, and we obtain $$\begin{aligned}
\label{Rsta}
R_*^{-1} &=& 1 + 2\bar\varepsilon_{3/2} \Big[
a\cos(\phi+\phi_a) - \cos\gamma\cos\phi \Big]
+ \bar\varepsilon_{3/2}^2
(1 - 2a\cos\phi_a\cos\gamma + a^2) \nonumber\\
&\le& \left[ 1 + \bar\varepsilon_{3/2} \left(
|a|+|\cos\gamma| \right) \right]^2
+ \bar\varepsilon_{3/2}^2\sin^2\!\gamma \,.\end{aligned}$$ Again, a lower bound can be obtained by changing the sign of $\bar\varepsilon_{3/2}$. In contrast to the previous case, now the maximal value of $|X_R|$ is given by $1+|a|$ and thus can exceed the SM bound provided that $|a|>\delta_{\rm EW}$. This is illustrated in the right-hand plot in Figure \[fig:case12\], where we show the resulting upper bound on $X_R$ versus $\gamma$ for different values of $|a|$. In contrast with the case of isospin-conserving New Physics, even a moderate enhancement of the coefficient $a$ corresponding to a 10%–20% change in the decay amplitudes can lead to a significant increase of the upper bound on $X_R$.
If both isospin-violating and isospin-conserving New Physics effects are present and involve new CP-violating phases, the analysis becomes more complicated. Still, it is possible to derive from (\[Rstar\]) a series of bounds on $R_*^{-1}$. We find $$\begin{aligned}
R_*^{-1} &\le& 1 + 2\bar\varepsilon_{3/2} \Big[
\cos\varphi\,(|a|+|\cos\gamma|)
+ |\sin\varphi|\,(|b|+|\sin\gamma|) \Big] \nonumber\\
&&\mbox{}+ \bar\varepsilon_{3/2}^2 \Big[
(|a|+|\cos\gamma|)^2 + (|b|+|\sin\gamma|)^2 \Big] \nonumber\\
&\le& \left[ 1 + \bar\varepsilon_{3/2}
\sqrt{(|a|+|\cos\gamma|)^2 + (|b|+|\sin\gamma|)^2} \right]^2
\nonumber\\
&\le& \left[ 1 + \bar\varepsilon_{3/2} \left( 1
+ \sqrt{a^2 + b^2} \right) \right]^2 \,,\end{aligned}$$ where in the second and third steps we have eliminated $\varphi$ and $\gamma$, respectively. As before a series of lower bounds is obtained by changing the sign of $\bar\varepsilon_{3/2}$. The corresponding bounds on $X_R$ are $$\label{abbound}
|X_R| \le \sqrt{(|a|+|\cos\gamma|)^2 + (|b|+|\sin\gamma|)^2}
\le 1 + \sqrt{a^2 + b^2}
\le \frac{2}{\bar\varepsilon_{3/2}} + X_R \,,$$ where the last inequality is relevant only in cases where $\sqrt{a^2 + b^2}\gg\delta_{\rm EW}$. With the current values for $\bar\varepsilon_{3/2}$ and $X_R$, the right-hand side is less than 15 at 90% confidence level. The important point to note is that in the most general case, where $b$ and $\rho$ are nonzero, the maximal value $X_R$ can take is no longer restricted to occur at the endpoints $\gamma=0^\circ$ or $180^\circ$, which are disfavored by the global analysis of the unitarity triangle [@Yuvi]. Rather, the maximal value $X_R^{\rm max}
=1+\sqrt{a^2 + b^2}$ now occurs at $$|\tan\gamma| = |\rho| = \bigg| \frac{b}{a} \bigg| \,.$$
The situation is illustrated in Figure \[fig:case3\], where we show the upper bound on $X_R$ for arbitrary New Physics contributions satisfying $a^2+b^2=1$, but for different values of $a$ and $b$ chosen such that the maximum occurs at the endpoints $|\gamma|=0^\circ$ or $180^\circ$ ($b=0$), at the intermediate points $|\gamma|=45^\circ$ or $135^\circ$ ($a=b$), and in the center at $|\gamma|=90^\circ$ ($a=0$). The corresponding values of the New Physics parameter $\rho$ required to reach the maximum are $|\rho|=0$, 1, and $\infty$ (i.e., $|\rho|\gg 1$), respectively.
The present experimental value of $X_R$ in (\[XRval\]) has too large an error to determine whether there is any deviation from the SM prediction. If $X_R$ turns out to be larger than 1 (i.e., only about one third of a standard deviation above its current central value), then an interpretation of this result in the SM would require a large value $|\gamma|>96^\circ$ (see Figure \[fig:SMbound\]), which may be difficult to accommodate. This may be taken as evidence for New Physics. If $X_R>1.3$, one could go a step further and conclude that this New Physics must necessarily violate isospin.
New Physics effects on the determination of $\gamma$
----------------------------------------------------
A value of the observable $R_*$ which violates the SM bound (\[Rstbound\]) would be an exciting hint for new isospin-violating penguin contributions from New Physics. With the current central value of $R_*$ derived from CLEO data this is still a realistic possibility. However, even if a more precise measurement will give a value that is consistent with the SM bound, $B^\pm\to\pi K$ decays still provide an excellent testing ground for physics beyond the SM. In the SM, the weak phase $\gamma$, along with the strong-interaction phase $\phi$, can be determined up to discrete ambiguities by combining measurements of $R_*$ and an asymmetry $\widetilde A$, which is defined as a linear combination of the direct CP asymmetries in the two $B^\pm\to\pi K$ decay channels [@NR; @me]. The discrete ambiguities can be resolved using information on the strong-interaction phase $\phi$ from theoretical approaches such as the QCD factorization theorem derived in [@fact]. The theoretical uncertainty on the value of $\gamma$ is typically of order $10^\circ$. Although New Physics may not be exotic enough to lead to a violation of the general bounds derived in the previous section, it may still cause a significant shift in the extracted value of $\gamma$. This may lead to inconsistencies when the value $\gamma_{\pi K}$ extracted in $B^\pm\to\pi K$ decays is compared with determinations of $\gamma$ using other information.
A global fit of the unitarity triangle combining information from semileptonic $B$ decays, $B$–$\bar B$ mixing, CP violation in the kaon system, and mixing-induced CP violation in $B\to J/\psi\,K_S$ decays will provide information on $\gamma$, which in a few years will determine its value within a rather narrow range [@BaBar; @Yuvi]. Such an indirect determination could be complemented by direct measurements of $\gamma$ using, e.g., $B\to D K^{(*)}$ decays [@Soffer], or using the triangle relation $\gamma=180^\circ-\alpha-\beta$ combined with a measurement of $\alpha$ in $B\to\pi\pi$ or $B\to\pi\rho$ decays [@BaBar]. In our discussion below we will assume that a discrepancy between the “true” $\gamma=\mbox{arg}(V_{ub}^*)$ and the value $\gamma_{\pi K}$ extracted in $B^\pm\to\pi K$ decays of more than $25^\circ$ will be observable after a few years of operation at the $B$ factories. This will set the benchmark for sensitivity to New Physics effects.
In order to illustrate how big an effect New Physics could have on the value of $\gamma$ extracted from $B^\pm\to\pi K$ decays, we assume for simplicity that the strong-interaction phase $\phi$ is small. We expect that this is indeed a good assumption, since the QCD factorization theorem of [@fact] predicts that $\phi=O(\alpha_s,\Lambda/m_b)$. In this case, a measurement of the asymmetry $\widetilde A$ will provide information about $\phi$ (and serve as a test of our assumption), whereas $\gamma$ is determined by $R_*$ alone. In the context of the SM, the solution obtained with $\cos\phi\approx 1$ is $$\label{gamextr}
\cos\gamma_{\pi K} \approx \delta_{\rm EW}
- \frac{R_*^{-1}-1
-\bar\varepsilon_{3/2}^2(1-\delta_{\rm EW}^2)}
{2\bar\varepsilon_{3/2}
(1+\bar\varepsilon_{3/2}\delta_{\rm EW})}
\approx \delta_{\rm EW}
- \frac{R_*^{-1}-1}{2\bar\varepsilon_{3/2}} \,.$$ Given the present uncertainties in $\bar\varepsilon_{3/2}$ and $\delta_{\rm EW}$, this result is a good approximation to the exact solution as long as $|\phi|<25^\circ$. Let us now investigate how New Physics may affect the results of this extraction, neglecting, for the purpose of illustration, all strong-interaction phases. As in the previous section, we focus first on the situation where the New Physics conserves isospin symmetry. Inserting for $R_*^{-1}$ in (\[gamextr\]) the expression given in (\[Rstphi\]), we obtain $$\cos\gamma_{\pi K} = \cos(\gamma+\varphi)
+ \delta_{\rm EW}\,(1-\cos\varphi)
+ O(\bar\varepsilon_{3/2}) \,.$$ For small values of $\varphi$, the result is simply given by $\gamma_{\pi K}\approx\gamma+\varphi$. Therefore, to have a significant shift requires having $|\varphi|>25^\circ$, which corresponds to rather large values $|\rho|>0.5$ and hence an $O(1)$ change in the decay amplitudes. The situation is different if the New Physics contributions violate isospin symmetry. Consider, for simplicity, the case where there are no new CP-violating phases. From (\[Rsta\]) it then follows that $$\cos\gamma_{\pi K} = \cos\gamma - a_{\rm NP}
+ O(\bar\varepsilon_{3/2}) \,,$$ where $$a_{\rm NP}\equiv a - \delta_{\rm EW} \,.$$ Now even a moderate New Physics contribution to the electroweak penguin coefficients can lead to a large shift in $\gamma$. In the most general case, where all New Physics contributions are present, we obtain $$\cos\gamma_{\pi K} = \cos(\gamma+\varphi) + \delta_{\rm EW}
- a\cos\varphi + b\sin\varphi + O(\bar\varepsilon_{3/2}) \,,$$ which again allows for large shifts.
These observations are illustrated in Figure \[fig:contours\], where we show contours of constant $X_R$ versus $\gamma$ for different values of one of the parameters $\varphi$, $a$ and $b$, keeping the other two fixed to their SM values. Without loss of generality we assume that $\gamma>0$. These plots show that even moderate New Physics contributions to the parameter $a$ can induce large shifts in $\gamma$. On the other hand, small values of $\varphi$ lead to much smaller effects. Likewise, the effects induced by the parameter $b$ are much smaller. We stress that the present central value of $X_R\approx 0.7$ is such that negative values of $\varphi$, as well as values of $a$ less than the SM result $a\approx 0.64$, are disfavored since they would require values of $\gamma$ exceeding $100^\circ$, in conflict with the global analysis of the unitarity triangle [@BaBar; @Yuvi].
In Figure \[fig:shifts\], we show the difference $\gamma_{\pi K}-\gamma$ as a function of $X_R$ for the two cases where either $\varphi$ or $a$ are varied with respect to their SM values. The dashed curves in the plots refer to negative values of $\varphi$ or $a_{\rm NP}$, which are disfavored by the present value of $X_R$. This implies that isospin-conserving New Physics can only lead to moderate shifts in $\gamma$, which reach the $30^\circ$ level for large values $\rho=O(1)$. Isospin-violating New Physics effects, on the other hand, can induce very large shifts of $\gamma$ even if they are of moderate size. As an example, consider the case where a future, precise measurement would yield $X_R=1$. An interpretation of this result in the SM would imply a relatively large value $\gamma\approx 106^\circ$, which can be determined with a theoretical uncertainty of about $10^\circ$ or better [@NR; @me]. Imagine that all other information about the unitarity triangle favors a value of $\gamma\approx 75^\circ$, again with a small error. To accommodate this difference, one could either invoke a large, isospin-conserving but CP-violating New Physics contribution such that $\varphi\approx 45^\circ$ (corresponding to $\rho\approx 1$), or an isospin-violating electroweak penguin contribution such that $a$ is twice as large as in the SM. The first solution would imply a New Physics contribution to the decay amplitudes of order 100%, whereas the second would imply only a small contribution of less than 15%.
“Wrong kaon” decays
-------------------
So far, when considering $B\to\pi K$ decays we have implicitly assumed an underlying quark transition of the form $\bar b\to\bar s q\bar q$, in which case the decays with a neutral kaon in the final state are $B^+\to\pi^+ K^0$ and $B^-\to\pi^-\bar K^0$. Indeed, in the SM this is an excellent approximation, because the quark transition $\bar b\to\bar d s\bar d$ leading to the “wrong kaon” decays $B^+\to\pi^+\bar K^0$ and $B^-\to\pi^- K^0$ is highly suppressed. However, this may no longer be the case in the presence of New Physics.[^4]
In practice, only the $B^\pm\to\pi^\pm K_{S,L}$ decay rates can be measured. In particular, the CLEO result for $R_*$ quoted in (\[Rstexp\]) really refers to the ratio $$\label{Xwrong}
R_*^{\rm exp} =
\frac{\mbox{B}(B^+\to\pi^+ K_S)+\mbox{B}(B^-\to\pi^- K_S)}
{\mbox{B}(B^+\to\pi^0 K^+)+\mbox{B}(B^-\to\pi^0 K^-)}
\equiv R_*\,(1+|X_{\rm wrong}|^2) \,,$$ which differs from $R_*$ by the “wrong kaon” contribution $$|X_{\rm wrong}|^2 =
\frac{\mbox{B}(B^+\to\pi^+\bar K^0)+\mbox{B}(B^-\to\pi^- K^0)}
{\mbox{B}(B^+\to\pi^+ K^0)+\mbox{B}(B^-\to\pi^-\bar K^0)}
\,.$$ Note that the presence of this contribution could only enhance the observed value $R_*^{\rm exp}$ with respect to $R_*$. This observation, combined with the fact that $R_*^{\rm exp}$ is not much larger than the expected value for $R_*$ in the SM, allows us to put bounds on the “wrong kaon” contribution in specific New Physics models.
In analogy with (\[basis\]), there are three operators entering the effective Hamiltonian for $\bar b\to\bar d s\bar d$ decays, which we define as $$\begin{aligned}
O_1^{dd} &=& (\bar b_\alpha d_\alpha)_{V-A}\,
(\bar s_\beta d_\beta)_{V+A} \,, \nonumber\\
O_2^{dd} &=& (\bar b_\alpha d_\beta)_{V-A}\,
(\bar s_\beta d_\alpha)_{V+A} \,, \nonumber\\
O_3^{dd} &=& (\bar b_\alpha d_\alpha)_{V-A}\,
(\bar s_\beta d_\beta)_{V-A} \,,\end{aligned}$$ and similarly there may be operators $\widetilde O_i^{dd}$ of opposite chirality. We denote the corresponding Wilson coefficients by $c_i^{dd}$ and $\widetilde c_i^{dd}$, respectively. The renormalization-group evolution of these coefficients can be read off from (\[RGE-R\]) and (\[RGE-L\]) by replacing $c_1^{\rm EW}\to c_1^{dd}$, $c_2^{\rm EW}\to c_2^{dd}$, and $c_{3,4}^{\rm EW}\to c_3^{dd}$. Using factorization, and normalizing the result to $\bar\varepsilon_{3/2}$, we find $$X_{\rm wrong} = \frac{\bar\varepsilon_{3/2}}{|\lambda_u|}
\left[ \bar c_3^{dd} - \frac23\,\kappa^{3/23}\,\bar c_1^{dd}
+ \frac{3\chi-1}{4}\,\kappa^{30/23} \bigg( \bar c_2^{dd}
+ \frac13\,\bar c_1^{dd} \bigg) \right] \,,$$ where $\bar c_i^{dd}\equiv c_i^{dd}(m_W)-\widetilde c_i^{dd}(m_W)$. Inserting this result into (\[Xwrong\]), and performing the evolution to the scale $\mu=m_b$, we obtain $$\label{Xwrongbound}
|\bar c_3^{dd} - 0.26 \bar c_1^{dd} + 1.40\bar c_2^{dd}|
\approx \frac{|\lambda_u|}{\bar\varepsilon_{3/2}}
\sqrt{\frac{R_*^{\rm exp}}{R_*} -1} \,.$$ Using the current values of $\lambda_u$, $\bar\varepsilon_{3/2}$ and $R_*^{\rm exp}$, and taking $R_*>0.45$ corresponding to the smallest possible value in the SM, we find that the right-hand side is less than $5.2\times 10^{-3}$ at 90% confidence level.
Trojan penguins from tree-level processes {#sec:tree}
=========================================
In this and the following sections, we calculate the matching conditions for the penguin coefficients $c_i^q$ and $\widetilde c_i^q$ in a variety of extensions of the SM with new flavor-violating couplings. We focus first on models where such couplings appear at tree level. Loop-mediated New Physics contributions will be discussed in Section \[sec:loop\].
Flavor-changing $Z$-boson exchange
----------------------------------
A generic feature of extensions of the SM with extra nonsequential quarks is the presence of tree-level flavor-changing couplings of the $Z$ boson, such as a $bsZ$ vertex. A detailed discussion of such models can be found, e.g., in [@bsnn]. The relevant terms in the Lagrangian read $$\label{mixcurrent}
{\cal L}_{\rm FCNC} = -\frac{g}{4\cos\theta_W}\,\sum_{i\neq j}
\,\bar d^i\,[\kappa_L^{ij}\gamma^\mu(1-\gamma_5) +
\kappa_R^{ij}\gamma^\mu(1+\gamma_5)]\,d^j Z_\mu \,,$$ where $i,j$ are generation indices. The quantities $\kappa_L^{bs}$ and $\kappa_R^{bs}$ are the two new complex parameters relevant to $b\to s$ transitions. Since the flavor-violating interactions are small (see below), the flavor-diagonal couplings of the $Z$-boson are to leading order the same as in the SM. It follows that at low energies the tree-level $Z$ exchange for the decay $\bar b\to\bar s q\bar q$ leads to the effective Hamiltonian $$\label{ZEWP}
{\cal H}_{\rm eff} = \frac{G_F}{\sqrt2}\,
[\kappa_L^{bs}(\bar b s)_{V-A} + \kappa_R^{bs}(\bar b s)_{V+A}]
\,\sum_q\,[ C_L^q(\bar q q)_{V-A} + C_R^q(\bar q q)_{V+A} ] \,,$$ where $$\begin{aligned}
C_L^u &=& \phantom{-}\frac12 - \frac23 \sin^2\!\theta_W \,,
\qquad
C_R^u = -\frac23 \sin^2\!\theta_W \,, \nonumber\\
C_L^d &=& -\frac12 + \frac13 \sin^2\!\theta_W \,, \qquad
C_R^d = \phantom{-}\frac13 \sin^2\!\theta_W \,.\end{aligned}$$ It is straightforward to match this result with the generic form of the penguin terms in the effective Hamiltonian, and to deduce the corresponding values of the Wilson coefficients. The nonvanishing coefficients at the matching scale are $c_1^q=\kappa_L^{bs}\,C_R^q$, $c_3^q=\kappa_L^{bs}\,C_L^q$, $\widetilde c_1^q=\kappa_R^{bs}\,C_L^q$, and $\widetilde c_3^q=\kappa_R^{bs}\,C_R^q$. In the notation of (\[cdefs\]), this implies $$c_3^{\rm QCD} = - \frac{\kappa_L^{bs}}{6} \,, \qquad
\widetilde c_1^{\rm QCD} = - \frac{\kappa_R^{bs}}{6} \,,
\qquad
c_1^{\rm QCD} = \widetilde c_3^{\rm QCD} = 0 \,,$$ as well as $$\begin{aligned}
c_1^{\rm EW} &=& -\kappa_L^{bs}\sin^2\!\theta_W \,, \qquad
c_3^{\rm EW} = \phantom{-}\kappa_L^{bs}\cos^2\!\theta_W \,,
\nonumber\\
\widetilde c_1^{\rm EW} &=& \phantom{-}\kappa_R^{bs}
\cos^2\!\theta_W \,, \qquad
\widetilde c_3^{\rm EW} = -\kappa_R^{bs}\sin^2\!\theta_W \,.\end{aligned}$$ These electroweak penguin coefficients would be of the same order as the SM result for $c_3^{\rm EW}$ in (\[SMinit\]) if $|\kappa_{L,R}^{bs}|\approx 5\times 10^{-4}$ which, as we will see below, is consistent with experimental bounds. Inserting these results into (\[abres\]) and using $|\lambda_u|\approx
7.5\times 10^{-4}$ yields $$\label{cute}
a_{\rm NP}+ib \approx -1.1\times 10^3\,
(\kappa_L^{bs} + 0.51\kappa_R^{bs}) \,.$$ We stress that the simple model considered here is a prototype of New Physics models in which the electroweak penguin coefficients $c_i^{\rm EW}$ and $\widetilde c_i^{\rm EW}$ are not suppressed relative to the QCD penguin coefficients $c_i^{\rm QCD}$ and $\widetilde c_i^{\rm QCD}$. This property is in contrast with the SM, where electroweak penguins are suppressed by small gauge couplings.
At present, the strongest constraints on $\kappa_L^{bs}$ and $\kappa_R^{bs}$ follow from the experimental bound on the $B\to X_s\,e^+ e^-$ decay rate. Since this bound lies far above the SM prediction for this process, we can neglect the SM contribution and write the effective Hamiltonian for this process as $${\cal H}_{\rm eff} = \frac{G_F}{\sqrt2}\,
[\kappa_L^{bs}(\bar bs)_{V-A} + \kappa_R^{bs}(\bar bs)_{V+A}]
\left[ C_L^e\,(\bar e e)_{V-A}
+ C_R^e\,(\bar e e)_{V+A} \right] \,,$$ where $$C_L^e = -\frac12 + \sin^2\!\theta_W \,, \qquad
C_R^e = \sin^2\!\theta_W \,.$$ It is convenient to normalize the result for the $B\to X_s\,e^+ e^-$ decay rate to the semileptonic rate. Then many common factors cancel, and we obtain $$\frac{\Gamma(B\to X_s\,e^+ e^-)}{\Gamma(B\to X_c\,e^-\bar\nu_e)}
= \frac{|\kappa_L^{bs}|^2+|\kappa_R^{bs}|^2}
{f(m_c/m_b)\,|V_{cb}|^2}
\left[ (C_L^e)^2 + (C_R^e)^2 \right] \approx
157 \left( |\kappa_L^{bs}|^2 + |\kappa_R^{bs}|^2 \right) \,,$$ where we have used $|V_{cb}|\approx 0.04$, and $f(m_c/m_b)\approx
0.5$ for the phase-space factor in the semileptonic decay. Using the upper bound $\mbox{B}(B\to X_s\,e^+ e^-)<5.7\times 10^{-5}$ together with $\mbox{B}(B\to X_c\,e^-\bar\nu_e)=0.105$ [@PDG] yields $$\sqrt{|\kappa_L^{bs}|^2 + |\kappa_R^{bs}|^2}
< 1.9\times 10^{-3} \,.$$ Combining this result with (\[cute\]), we obtain the upper bound $$\sqrt{a_{\rm NP}^2 + b^2} < 2.0 \,,$$ which may be compared with the SM value $a\approx 0.64$. It follows that tree-level $Z$ exchange with new flavor-violating couplings can yield isospin-violating electroweak penguin effects that are up to a factor 3 larger than in the SM.
For completeness, we also mention the resulting bound on the New Physics parameter $\rho$. Neglecting renormalization-group effects, we obtain from (\[rhobound\]) the estimate $$\frac{\rho}{\sqrt{1+\rho^2}} \approx 29\,\Big[
0.8\,\mbox{Im}\,\kappa_L^{bs} + \mbox{Im}\,\kappa_R^{bs}
\Big] \quad
\Rightarrow \quad |\rho| < 0.05 \,, \quad
|\varphi| < 3^\circ \,,$$ indicating that in this model there is no room for large values of $\rho$.
Extended gauge models with a $Z'$ boson
---------------------------------------
A new neutral boson $Z'$ with tree-level flavor-changing couplings to quarks is a generic property of many models with an extended gauge group. The analysis of electroweak penguins in such models is very similar to the flavor-changing tree-level $Z$ exchange discussed above. For simplicity, we assume no significant mixing between the $Z$ and $Z'$ bosons. Then the effective Hamiltonian is a simple generalization of (\[ZEWP\]), i.e. $$\begin{aligned}
\label{Zp-EWP}
{\cal H}_{\rm eff} &=& \frac{g_{U(1)'}^2}{m_{Z'}^2}
[ \kappa_L^{\prime bs}(\bar b s)_{V-A}
+ \kappa_R^{\prime bs}(\bar b s)_{V+A} ]\,
\sum_q\, [ C_L^q(\bar q q)_{V-A} + C_R^q(\bar q q)_{V+A} ] \,,\end{aligned}$$ where $C_{L,R}^q$ now denote the charges of the quarks under the new U$(1)'$ group. Introducing the ratio $$\xi = \frac{g_{{\rm U}(1)'}^2}{g^2}\,\frac{m_W^2}{m_{Z'}^2} \,,$$ and taking into account that, since we neglect $Z$–$Z'$ mixing, the $Z'$ charges are the same for all left-handed fields (in particular, $C_L^u=C_L^d$), we find $$\begin{aligned}
c_1^{\rm QCD} &=& \xi\kappa_L^{\prime bs}\,
\frac{C_R^u+2C_R^d}{3} \,, \qquad
c_3^{\rm QCD} = \xi\kappa_L^{\prime bs} C_L^q \,,
\nonumber\\
\widetilde c_1^{\rm QCD} &=& \xi\kappa_R^{\prime bs} C_L^q \,,
\qquad
\widetilde c_3^{\rm QCD} = \xi\kappa_R^{\prime bs}\,
\frac{C_R^u+2C_R^d}{3} \,,\end{aligned}$$ and $$c_1^{\rm EW} = \xi\kappa_L^{\prime bs} (C_R^u-C_R^d) \,, \qquad
\widetilde c_3^{\rm EW} = \xi\kappa_R^{\prime bs}
(C_R^u-C_R^d) \,, \qquad
c_3^{\rm EW} = \widetilde c_1^{\rm EW} = 0 \,.$$ Inserting these results into (\[abres\]) gives $$a_{\rm NP}+ib \approx \frac{\xi(C_R^u-C_R^d)}{|\lambda_u|}\,
(\kappa_R^{\prime bs}+0.26\kappa_L^{\prime bs}) \,.$$
The allowed range for the relevant parameters in $Z'$ extensions of the SM is largely model dependent. For example, the bounds derived from the upper limit on the $B\to X_s\,l^+ l^-$ branching ratio depend on the lepton charges under the new U$(1)'$ gauge group. In the so-called “leptophobic” $Z'$ models these charges are arranged so as to vanish or be very small [@leptophobic]. Therefore, in general the contributions of the $Z'$ couplings to the electroweak penguin operators can be arbitrarily large. In fact, the best model-independent bound on these couplings follows from the second inequality in (\[abbound\]), which implies $$|\xi (C_R^u-C_R^d)|\,
|\kappa_R^{\prime bs}+0.26\kappa_L^{\prime bs}|
< 0.01 \,,$$ where we have neglected the small SM contribution. Furthermore, assuming $C_{L,R}^q=O(1)$ and no cancellations, the bound (\[rhonumeric\]) gives $|\xi\,\mbox{Im}\,
\kappa_{L,R}^{\prime bs}|<O(10^{-2})$. Turning these observations around, we conclude that in extended gauge models with flavor-changing $Z'$ couplings such that $\xi\kappa_{L,R}^{\prime bs}=O(10^{-2})$ there can be very large New Physics effects in $B\to\pi K$ decays.
SUSY models with R-parity violation
-----------------------------------
In SUSY models with broken R-parity extra trilinear terms are allowed in the superpotential, some of which can give rise to a large enhancement of the electroweak penguin coefficients. Denoting by $L_L^i$, $Q_L^i$, $u_R^i$ and $d_R^i$ the chiral superfields containing, respectively, the left-handed lepton and quark doublets, and the right-handed up- and down-type quark singlets of the $i$-th generation, these terms read $$\label{Wrpb}
W = \lambda'_{ijk}\,L_L^i\,Q_L^j\,\bar d_R^k
+ \lambda''_{ijk}\,\bar u_R^i\,\bar d_R^j\,\bar d_R^k \,.$$ At low energies, slepton and squark exchange can generate local penguin operators. The most general case has been treated in [@CDK]. For simplicity, we neglect left-right sfermion mixing, which is a small effect and does not generate new operators. We then find for the coefficients of the various penguin operators $$\widetilde c_2^u = \sum_{i=1}^3
\frac{\lambda_{i12}^{\prime *} \lambda'_{i13}}
{4\sqrt{2} G_F m_{\tilde e_{iL}}^2} \,, \qquad
- \widetilde c_3^u = \widetilde c_4^u
= \frac{\lambda_{113}^{\prime\prime *} \lambda''_{112}}
{2\sqrt{2} G_F m_{\tilde d_{1R}}^2} \,,$$ and $$\begin{aligned}
c_2^d &=& \sum_{i=1}^3
\frac{\lambda_{i31}^{\prime *} \lambda'_{i21}}
{4\sqrt{2} G_F m_{\tilde\nu_i}^2} \,, \qquad
c_6^d = \sum_{i=1}^3
\frac{\lambda_{i32}^{\prime *} \lambda'_{i11}}
{4\sqrt{2} G_F m_{\tilde\nu_i}^2} \,, \nonumber \\
\widetilde c_2^d &=& \sum_{i=1}^3
\frac{\lambda_{i12}^{\prime *} \lambda'_{i13}}
{4\sqrt{2} G_F m_{\tilde\nu_i}^2} \,, \qquad
\widetilde c_6^d = \sum_{i=1}^3
\frac{\lambda_{i11}^{\prime *} \lambda'_{i23}}
{4\sqrt{2} G_F m_{\tilde\nu_i}^2} \,, \nonumber\\
- \widetilde c_3^d &=& \widetilde c_4^d = \sum_{i=1}^3
\frac{\lambda_{i13}^{\prime\prime *} \lambda''_{i12}}
{2\sqrt{2} G_F m_{\tilde u_{iR}}^2} \,.\end{aligned}$$ Previous authors have investigated bounds on some of these R-parity violating couplings in the context of nonleptonic $B$ decays [@CRS; @BD]. However, in these studies model-dependent predictions for the overall penguin amplitude $P$ in (\[ampls\]) are employed. The only significant bound which has an impact on our analysis comes from a combination of constraints derived from limits on double nucleon decay into two kaons and neutron–antineutron oscillations, yielding $|\lambda_{113}^{\prime\prime *}
\lambda''_{112}|<10^{-9}$ [@CRS]. Therefore, is it safe to neglect $\widetilde c_3^u$ and $\widetilde c_4^u$.
Not all of the above coefficients contribute to the isospin-violating terms parametrized by $a$ and $b$. Up to a small SU(2)$_L$ breaking in the slepton and sneutrino masses we find $\widetilde c_2^u=\widetilde c_2^d$, and thus $\widetilde
c_2^{\rm EW}\approx 0$. Moreover, in (\[abres\]) only the sum $\bar c_3^{\rm EW}+\bar c_4^{\rm EW}$ contributes, which vanishes since $\widetilde c_3^q=-\widetilde c_4^q$. It follows that $$a_{\rm NP}+ib \approx 2.83\times 10^3\,
\sum_{i=1}^3\,\frac{(100\,\mbox{GeV})^2}{m_{\tilde\nu_i}^2}\,
\Big( {\lambda_{i31}^{\prime *} \lambda'_{i21}}
+ \lambda_{i32}^{\prime *} \lambda'_{i11}
- \lambda_{i11}^{\prime *} \lambda'_{i23} \Big) \,.$$ The result for the parameter $\rho$ is more complicated. Setting for simplicity all sfermion masses equal, we find from (\[rhobound\]) $$\begin{aligned}
\frac{\rho}{\sqrt{1+\rho^2}} &\approx& 106\,
\frac{(100\,\mbox{GeV})^2}{m_{\tilde f}^2} \nonumber\\
&\times& \sum_{i=1}^3\,\mbox{Im} \Big[
4 \lambda_{i13}^{\prime\prime *} \lambda''_{i12}
+ 3\chi(\lambda_{i12}^{\prime *} \lambda'_{i13}
- \lambda_{i31}^{\prime *} \lambda'_{i21})
+ \lambda_{i32}^{\prime *} \lambda'_{i11}
- \lambda_{i11}^{\prime *} \lambda'_{i23}
\Big] \,. \qquad\end{aligned}$$ Using the results derived in Section \[sec:BpiK\], we can obtain bounds on several of the R-parity violating couplings. Assuming that only one combination of couplings is dominant, neglecting the SM contribution, and using a common sfermion reference mass of $100\,$GeV, we find from (\[abbound\]) that at 90% confidence level $$\begin{aligned}
\label{bounds1}
\Big| \sum_{i=1}^3\,\lambda_{i31}^{\prime *} \lambda'_{i21}
\Big| &<& 4.9\times 10^{-3} \,, \nonumber\\
\Big| \sum_{i=1}^3\,\lambda_{i32}^{\prime *} \lambda'_{i11}|
\Big| &<& 4.9\times 10^{-3} \,, \nonumber\\
\Big| \sum_{i=1}^3\,\lambda_{i11}^{\prime *} \lambda'_{i23}|
\Big| &<& 4.9\times 10^{-3} \,.\end{aligned}$$ In addition, from (\[rhonumeric\]) we obtain $$\begin{aligned}
\label{bounds2}
\Big| \sum_{i=1}^3\,\mbox{Im}\,(\lambda_{i13}^{\prime\prime *}
\lambda''_{i12}) \Big| &<& 3.7\times 10^{-3} \,, \nonumber \\
\Big| \sum_{i=1}^3\,\mbox{Im}\,(\lambda_{i31}^{\prime *}
\lambda'_{i21}) \Big| &<& 4.1\times 10^{-3} \,, \nonumber \\
\Big| \sum_{i=1}^3\,\mbox{Im}\,(\lambda_{i12}^{\prime *}
\lambda'_{i13}) \Big| &<& 4.1\times 10^{-3} \,, \end{aligned}$$ where we do not present bounds on the imaginary parts of couplings which are weaker than the corresponding bounds on the absolute values in (\[bounds1\]).
It is interesting that SUSY models with R-parity violation provide an example of scenarios in which $\bar b\to\bar d s\bar d$ transitions may not be suppressed relative to $\bar b\to\bar s d\bar d$ transitions. As pointed out in Section \[sec:BpiK\], this can lead to potentially large “wrong kaon” decays of the type $B^+\to\pi^+\bar K^0$ and $B^-\to\pi^- K^0$. In the model considered here, the only nonvanishing coefficients are $$c_2^{dd} = \sum_{i=1}^3
\frac{\lambda_{i31}^{\prime *} \lambda'_{i12}}
{4\sqrt{2} G_F m_{\tilde\nu_i}^2} \,, \qquad
\widetilde c_2^{dd} = \sum_{i=1}^3
\frac{\lambda_{i21}^{\prime *} \lambda'_{i13}}
{4\sqrt{2} G_F m_{\tilde\nu_i}^2} \,.$$ The result (\[Xwrongbound\]) can be used to obtain the bounds $$\label{bounds3}
|\lambda_{i31}^{\prime *} \lambda'_{i12}|
< 3.4\times 10^{-3} \,, \qquad
|\lambda_{i21}^{\prime *} \lambda'_{i13}|
< 3.4\times 10^{-3} \,,$$ again at 90% confidence level.
Our bounds in (\[bounds1\]) and (\[bounds2\]) are stronger than the ones discussed in the literature [@CRS; @BD] and refer to a larger number of $R$-parity violating couplings. Most importantly, however, they are affected by much smaller hadronic uncertainties. The bounds in (\[bounds3\]), on the other hand, are weaker than constraints derived from $B$–$\bar B$ and $K$–$\bar K$ mixing [@Bhat].
Trojan penguins from loop processes {#sec:loop}
===================================
Having considered in the previous section some specific models with tree-level FCNC couplings, we now explore extensions of the SM in which new contributions to the electroweak and QCD penguin operators arise at one-loop order. In particular, we study in detail the structure of electroweak penguins in SUSY models, where isospin-violating $\bar b\to\bar s q\bar q$ transitions can arise due to strong-interaction gluino box diagrams. This provides another realization of models in which electroweak penguins are not suppressed relative to QCD penguins. For completeness, we also discuss two-Higgs–doublet models and models with anomalous gauge-boson couplings. They are simple since there are no new CP-violating phases, so only the parameter $a$ can receive a New Physics contribution. However, there is no parametrical enhancement of the electroweak penguins relative to the SM, and thus the New Physics contributions tend to be small.
SUSY models
-----------
In SUSY extensions of the SM with conserved R-parity, the potentially most important contributions to the Wilson coefficients of the penguin operators in the effective Hamiltonian (\[Hpeng\]) arise from strong-interaction penguin and box diagrams with gluino–squark loops. They can contribute to FCNC processes because the gluinos have flavor-changing couplings to the quark and squark mass eigenstates. Provided there is a significant mass splitting between the right-handed up and down squarks, gluino box diagrams are also the most important source of isospin violation in SUSY models [@epspr]. The corresponding contributions to the electroweak penguin coefficients are then much more important than other SUSY contributions from photon or $Z$ penguins usually discussed in the literature. In fact, in such a scenario SUSY contributions to the coefficients of the electroweak and QCD penguin operators are of the same order and scale like $\alpha_s^2/m_{\rm SUSY}^2$, where $m_{\rm SUSY}$ is a generic mass of the superparticles. Whereas the QCD penguin contributions are typically smaller than in the SM, the electroweak penguin contributions can be important, since their scaling relative to the SM coefficients is controlled by the ratio $(\alpha_s/\alpha)(m_W^2/m_{\rm SUSY}^2)\sim 1$. In our analysis we will consider only these potentially large gluino box and penguin contributions and neglect a multitude of other SUSY diagrams, which are parametrically suppressed by small electroweak gauge couplings. The latter include photon or $Z$ penguins with gluino–squark, chargino–squark, neutralino–squark, or charged-Higgs–quark loops, and various box diagrams containing at least one chargino or neutralino. We have calculated all of these diagrams and, for generic regions in SUSY parameter space, have found their contributions to be largely suppressed relative to the pure gluino diagrams.
Large SUSY contributions to the penguin operators via gluino loops require near maximal mixing between the strange and bottom squarks, so that the squark mass-insertion approximation is not valid. We therefore present our results using the general vertex-mixing method, summing over diagrams with different squark mass-eigenstates in the loops [@berto91; @cho]. We denote by $\Gamma^{D_L}$ the rotation matrices relating the left-handed down-squark interaction states in the quark mass-eigenbasis, $\tilde q_L^I$ ($q=d,s,b$), to the squark mass eigenstates, $\tilde d_i$ ($i=1,\dots,6$), such that $\tilde q_L^I=(\Gamma^{D_L}_{iq})^*\,\tilde{d}_i$, with obvious generalizations for the up- and right-handed squarks. In addition, $x_{\tilde q_i\tilde g}\equiv m_{\tilde q_i}^2/m_{\tilde g}^2$, where $m_{\tilde q_i}$ is the mass of the $i$-th down ($q=d$) or up ($q=u$) squark mass eigenstate. In the operator basis of (\[basis\]), we obtain for the gluino box contributions to the Wilson coefficients at the SUSY matching scale $$\begin{aligned}
\label{gluinobox}
c_{1,{\rm box}}^u
&=& \frac{\alpha_s^2}{2\sqrt2 G_F m_{\tilde g}^2}
\left(\Gamma^{D_L}_{ib}\right)^* \Gamma^{D_L}_{is}
\left(\Gamma^{U_R}_{ju}\right)^* \Gamma^{U_R}_{ju}
\left[ \frac{1}{18}\,
F(x_{\tilde d_i\tilde g},x_{\tilde u_j\tilde g})
- \frac{5}{18}\,
G(x_{\tilde d_i\tilde g},x_{\tilde u_j\tilde g}) \right] ,
\nonumber\\
c_{1,{\rm box}}^d
&=& \frac{\alpha_s^2}{2\sqrt2 G_F m_{\tilde g}^2}
\left(\Gamma^{D_L}_{ib}\right)^* \Gamma^{D_L}_{is}
\left(\Gamma^{D_R}_{jd}\right)^* \Gamma^{D_R}_{jd}
\left[ \frac{1}{18}\,
F(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g})
- \frac{5}{18}\,
G(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g}) \right] ,
\nonumber\\
c_{2,{\rm box}}^u
&=& \frac{\alpha_s^2}{2\sqrt2 G_F m_{\tilde g}^2}
\left(\Gamma^{D_L}_{ib}\right)^* \Gamma^{D_L}_{is}
\left(\Gamma^{U_R}_{ju}\right)^* \Gamma^{U_R}_{ju}
\left[ \frac76\,
F(x_{\tilde d_i\tilde g},x_{\tilde u_j\tilde g})
+ \frac16\,
G(x_{\tilde d_i\tilde g},x_{\tilde u_j\tilde g}) \right] ,
\nonumber\\
c_{2,{\rm box}}^d
&=& \frac{\alpha_s^2}{2\sqrt2 G_F m_{\tilde g}^2}
\left(\Gamma^{D_L}_{ib}\right)^* \Gamma^{D_L}_{is}
\left(\Gamma^{D_R}_{jd}\right)^* \Gamma^{D_R}_{jd}
\left[ \frac76\,
F(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g})
+ \frac16\,
G(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g}) \right] ,
\nonumber\\
c_{3,{\rm box}}^u
&=& \frac{\alpha_s^2}{2\sqrt2 G_F m_{\tilde g}^2}
\left(\Gamma^{D_L}_{ib}\right)^* \Gamma^{D_L}_{is}
\left(\Gamma^{U_L}_{ju}\right)^* \Gamma^{U_L}_{ju}
\left[ -\frac59\,
F(x_{\tilde d_i\tilde g},x_{\tilde u_j\tilde g})
+ \frac{1}{36}\,
G(x_{\tilde d_i\tilde g},x_{\tilde u_j\tilde g}) \right] ,
\nonumber\\
c_{3,{\rm box}}^d
&=& \frac{\alpha_s^2}{2\sqrt2 G_F m_{\tilde g}^2}
\Bigg\{ \left(\Gamma^{D_L}_{ib}\right)^* \Gamma^{D_L}_{is}
\left(\Gamma^{D_L}_{jd}\right)^* \Gamma^{D_L}_{jd}
\left[ -\frac59\,
F(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g})
+ \frac{1}{36}\,
G(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g}) \right]
\nonumber\\
&&\hspace{1.8cm}\mbox{}+
\left(\Gamma^{D_L}_{ib}\right)^* \Gamma^{D_L}_{js}
\left(\Gamma^{D_L}_{jd}\right)^* \Gamma^{D_L}_{id}
\left[ \frac13\,
F(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g})
- \frac{7}{12}\,
G(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g}) \right]
\Bigg\} , \nonumber\\
c_{4,{\rm box}}^u
&=& \frac{\alpha_s^2}{2\sqrt2 G_F m_{\tilde g}^2}
\left(\Gamma^{D_L}_{ib}\right)^* \Gamma^{D_L}_{is}
\left(\Gamma^{U_L}_{ju}\right)^* \Gamma^{U_L}_{ju}
\left[ \frac13\,
F(x_{\tilde d_i\tilde g},x_{\tilde u_j\tilde g})
- \frac{7}{12}\,
G(x_{\tilde d_i\tilde g},x_{\tilde u_j\tilde g}) \right] ,
\nonumber\\
c_{4,{\rm box}}^d
&=& \frac{\alpha_s^2}{2\sqrt2 G_F m_{\tilde g}^2}
\Bigg\{ \left(\Gamma^{D_L}_{ib}\right)^* \Gamma^{D_L}_{is}
\left(\Gamma^{D_L}_{jd}\right)^* \Gamma^{D_L}_{jd}
\left[ \frac13\,
F(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g})
- \frac{7}{12}\,
G(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g}) \right]
\nonumber\\
&&\hspace{1.8cm}\mbox{}+
\left(\Gamma^{D_L}_{ib}\right)^* \Gamma^{D_L}_{js}
\left(\Gamma^{D_L}_{jd}\right)^* \Gamma^{D_L}_{id}
\left[ -\frac59\,
F(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g})
+ \frac{1}{36}\,
G(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g}) \right]
\Bigg\} , \nonumber\\
c_{5,{\rm box}}^d
&=& \frac{\alpha_s^2}{2\sqrt2 G_F m_{\tilde g}^2}
\left(\Gamma^{D_L}_{ib}\right)^* \Gamma^{D_R}_{js}
\left(\Gamma^{D_R}_{jd}\right)^* \Gamma^{D_L}_{id}
\left[ \frac{1}{18}\,
F(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g})
- \frac{5}{18}\,
G(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g}) \right] ,
\nonumber\\
c_{6,{\rm box}}^d
&=& \frac{\alpha_s^2}{2\sqrt2 G_F m_{\tilde g}^2}
\left(\Gamma^{D_L}_{ib}\right)^* \Gamma^{D_R}_{js}
\left(\Gamma^{D_R}_{jd}\right)^* \Gamma^{D_L}_{id}
\left[ \frac76\,
F(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g})
+ \frac16\,
G(x_{\tilde d_i\tilde g},x_{\tilde d_j\tilde g}) \right] , \end{aligned}$$ where repeated indices are summed over, and $c_{5,{\rm box}}^u
=c_{6,{\rm box}}^u=0$. The functions $F(x,y)$ and $G(x,y)$ are given by [@berto91; @cho] $$\begin{aligned}
F(x,y) &=& -\frac{x\ln x}{(x-y)(x-1)^2}
- \frac{y\ln y}{(y-x)(y-1)^2} - \frac{1}{(x-1)(y-1)} \,,
\nonumber\\
G(x,y) &=& \frac{x^2\ln x}{(x-y)(x-1)^2}
+ \frac{y^2\ln y}{(y-x)(y-1)^2} + \frac{1}{(x-1)(y-1)} \,. \end{aligned}$$ The corresponding expressions for the coefficients $\widetilde c_i^q$ of the opposite-chirality operators are obtained via the exchange $L\leftrightarrow R$ in the expressions for $c_i^q$. In practice, $c_{5,{\rm box}}^d$, $c_{6,{\rm box}}^d$ and the second terms in $c_{3,{\rm box}}^d$, $c_{4,{\rm box}}^{d}$ (as well as the corresponding terms in the coefficients of the opposite-chirality operators) can be neglected due to $B_d$–$\bar B_d$ and $K$–$\bar K$ mixing constraints on the off-diagonal (1-3) and (1-2) entries of the $\Gamma^{D_L}$ and $\Gamma^{D_R}$ matrices. Gluino box diagrams which in the mass-insertion approximation would contain left–right squark-mass insertions can also be neglected and have not been included in (\[gluinobox\]). Specifically, these graphs would contain the mass insertions $\delta m^2_{\tilde s_L\tilde b_R}$, $\delta m^2_{\tilde s_R\tilde b_L}$, $\delta m^2_{\tilde d_L\tilde b_R}$ or $\delta m^2_{\tilde d_R\tilde b_L}$, whose magnitudes are tightly constrained by the experimental value of the $B\to X_{s,d}\,\gamma$ branching ratio [@GGMS96]. Further suppression of such graphs can be expected on theoretical grounds since the remaining left–right squark-mass insertions they would contain are suppressed by light quark masses in general supergravity theories [@Louis; @lostAlex] and in SUSY theories of flavor [@Dine; @nirseiberg; @susyflavor].
In addition to the box contributions, the QCD penguin coefficients also receive contributions from gluon penguin diagrams containing gluino–squark loops. These are given at the SUSY matching scale by [@barbi; @abel] $$\begin{aligned}
\label{gluinoQCD}
c_{1,{\rm peng}}^q &=& c_{3,{\rm peng}}^q
= -\frac{c_{2,{\rm peng}}^q}{3}
= -\frac{c_{4,{\rm peng}}^q}{3} \nonumber\\
&=& \frac{\alpha_s^2}{2\sqrt2 G_F m_{\tilde g}^2}
\left( \Gamma^{D_L}_{ib} \right)^* \Gamma^{D_L}_{is}
\left[ \frac12\, A(x_{\tilde d_i\tilde g})
+ \frac29\,B(x_{\tilde d_i\tilde g}) \right] ,\end{aligned}$$ where $$\begin{aligned}
A(x) &=& \frac{1}{2(1-x)} + \frac{(1+2x)\ln x}{6(1-x)^2} \,,
\nonumber \\
B(x) &=& - \frac{11-7x+2x^2}{18(1-x)^3}
- \frac{\ln x}{3(1-x)^4} \,.\end{aligned}$$ The opposite-chirality contributions are again obtained via the substitution $L\rightarrow R$.
FCNC constraints on the off-diagonal entries of the squark mass matrix allow for a simple parametrization of the gluino box and penguin contributions to $\bar b\to\bar s u\bar u$ and $\bar b\to\bar s d\bar d$ transitions, up to small corrections which have a negligible impact on the resulting Wilson coefficients. Let us first consider the down-squark sector. Constraints from $B_d$–$\bar B_d$ and $K$–$\bar K$ mixing imply that, to good approximation, the down squark is decoupled from the strange and bottom squarks. We also neglect the left–right down-squark submatrix, since even in the most general case of supergravity theories with arbitrary Kähler potential its entries are much smaller than the typical squark-mass squared [@Louis; @lostAlex],[^5] and in SUSY theories of flavor its entries are even more suppressed [@susyflavor]. The above simplifications essentially give three “left-handed” and three “right-handed” down-squark mass eigenstates, obtained by diagonalizing the left–left and right–right squark submatrices. Specifically, in the physical down-quark basis $(d_L,s_L,b_L)$ the left–left submatrix takes the form $$M_{d,LL}^2 \simeq \left( \begin{array}{ccc}
m_{11}^2 & 0 & 0 \\
0 & m_{22}^2 & m^2_{23} \\
0 & (m_{23}^2)^* & m_{33}^2
\end{array} \right) \,.$$ Let us denote the left-handed mass eigenstates by $\tilde q_L$ ($q=d,s,b$) and their masses by $m_{\tilde q_L}^2$, making the identification $m_{\tilde d_L}^2=m_{11}^2$. Then the left-handed squark mass-eigenstates take the form $$\tilde d_L\equiv \tilde d_1 = \left( \begin{array}{c}
1 \\ 0 \\ 0 \end{array} \right) \,, \quad
\tilde s_L\equiv \tilde d_2 = \left( \begin{array}{c}
0 \\ \cos\theta_L \\ -\sin\theta_L\,e^{-i\delta_L}
\end{array} \right) \,, \quad
\tilde b_L\equiv \tilde d_3 = \left( \begin{array}{c}
0 \\ \sin\theta_L\,e^{i\delta_L} \\ \cos\theta_L
\end{array} \right) \,,$$ where $\delta_L$ is a new CP-violating phase. We take $|\theta_L|
\le 45^\circ$, so that the squark mass-eigenstate $\tilde s_L$ is more closely aligned with the $s$ quark, and $\tilde b_L$ with the $b$ quark. In the case of the box graphs we also need to consider the up-squark sector. $D$–$\bar D$ mixing bounds [@nirseiberg] imply that, to good approximation, the up squark is decoupled from the charm squark in the sense that including the phenomenologically allowed mixing between the two will lead to negligible modifications of the Wilson coefficients. Without loss of generality, we can also ignore mixing between the up and top squarks, which is a good approximation in SUSY theories of flavor.[^6] Finally, as before we can neglect the mixing between the left- and right-handed up squarks.
Taken together, the above approximations imply $$\begin{aligned}
\label{simpleap}
\left(\Gamma^{D_L}_{ib}\right)^* \Gamma^{D_L}_{is}
&\simeq& \frac12\sin2\theta_L\,e^{i\delta_L}\,
(\delta_{\tilde d_i\tilde b_L}
- \delta_{\tilde d_i\tilde s_L}) \,, \nonumber\\
\left(\Gamma^{D_L}_{ib}\right)^* \Gamma^{D_L}_{id}
&\simeq& 0 \,, \nonumber\\
\left(\Gamma^{D_L}_{jd}\right)^* \Gamma^{D_L}_{jd}
&\simeq& \delta_{\tilde d_j\tilde d_L} \,, \qquad
\left(\Gamma^{U_L}_{ju}\right)^* \Gamma^{U_L}_{ju}
\simeq \delta_{\tilde u_j\tilde u_L} \,.\end{aligned}$$ The diagonalization in the right-handed sector proceeds in a similar way, leading to mass eigenstates parameterized by a mixing angle $\theta_R$ and a weak phase $\delta_R$.
It is now straightforward to reexpress the gluino box and penguin contributions in (\[gluinobox\]) and (\[gluinoQCD\]) in terms of our parametrization. The combined results for the coefficients $c_i^q$ read ($q=u,d$) $$\begin{aligned}
\label{gluinoboxparam}
c_1^q &=& \frac{\alpha_s^2\sin2\theta_L\,e^{i\delta_L}}
{4\sqrt2 G_F m_{\tilde g}^2} \left[
\frac{1}{18}\,F(x_{\tilde b_L\tilde g},x_{\tilde q_R\tilde g})
- \frac{5}{18}\,G(x_{\tilde b_L\tilde g},x_{\tilde q_R\tilde g})
+ \frac12\,A(x_{\tilde b_L\tilde g})
+ \frac{2}{9}\,B(x_{\tilde b_L\tilde g}) \right] \nonumber\\
&&\mbox{}- (x_{\tilde b_L\tilde g}\to x_{\tilde s_L\tilde g})
\,, \nonumber\\
c_2^q &=& \frac{\alpha_s^2\sin2\theta_L\,e^{i\delta_L}}
{4\sqrt2 G_F m_{\tilde g}^2} \left[
\frac76\,F(x_{\tilde b_L\tilde g},x_{\tilde q_R\tilde g})
+ \frac16\,G(x_{\tilde b_L\tilde g},x_{\tilde q_R\tilde g})
- \frac32\,A(x_{\tilde b_L\tilde g})
- \frac{2}{3}\,B(x_{\tilde b_L\tilde g}) \right] \nonumber\\
&&\mbox{}- (x_{\tilde b_L\tilde g}\to x_{\tilde s_L\tilde g})
\,, \nonumber\\
c_3^q &=& \frac{\alpha_s^2\sin2\theta_L\,e^{i\delta_L}}
{4\sqrt2 G_F m_{\tilde g}^2} \left[
- \frac59\,F(x_{\tilde b_L\tilde g},x_{\tilde q_L\tilde g})
+ \frac{1}{36}\,G(x_{\tilde b_L\tilde g},x_{\tilde q_L\tilde g})
+ \frac12\,A(x_{\tilde b_L\tilde g})
+ \frac{2}{9}\,B(x_{\tilde b_L\tilde g}) \right] \nonumber\\
&&\mbox{}- (x_{\tilde b_L\tilde g}\to x_{\tilde s_L\tilde g})
\,, \nonumber\\
c_4^q &=& \frac{\alpha_s^2\sin2\theta_L\,e^{i\delta_L}}
{4\sqrt2 G_F m_{\tilde g}^2} \left[
\frac13\,F(x_{\tilde b_L\tilde g},x_{\tilde q_L\tilde g})
- \frac{7}{12}\,G(x_{\tilde b_L\tilde g},x_{\tilde q_L\tilde g})
- \frac32\,A(x_{\tilde b_L\tilde g})
- \frac{2}{3}\,B(x_{\tilde b_L\tilde g}) \right] \nonumber\\
&&\mbox{}- (x_{\tilde b_L\tilde g}\to x_{\tilde s_L\tilde g})
\,.\end{aligned}$$ The coefficients $\widetilde c_i^q$ are obtained by substituting $L\leftrightarrow R$ above. The coefficients $c_{5,6}^q$ and $\widetilde c_{5,6}^q$ vanish in the approximation (\[simpleap\]).
Let us identify those regions of SUSY parameter space which can give large contributions to the Wilson coefficients $c_i^q$ and $\widetilde c_i^q$. It turns out that a small gluino mass is favored for all of the coefficients. Large contributions also require $m_{\tilde s_{L,R}}^2\gg m_{\tilde b_{L,R}}^2$ and small $m_{\tilde b_{L,R}}^2$, or vice versa. Both options, $m_{\tilde s_{L,R}}^2\gg m_{\tilde b_{L,R}}^2$ or $m_{\tilde b_{L,R}}^2\gg m_{\tilde s_{L,R}}^2$, are equivalent as far as the magnitudes of the new contributions to the Wilson coefficients are concerned. Perhaps the first option is more attractive given that constraints from $K$–$\bar K$ mixing are more stringent than those from $B$–$\bar B$ mixing. In addition, in models where SUSY is broken at high energies, e.g., at the grand-unified theory (GUT) or Planck scales, renormalization tends to make the third-generation squarks the lightest at the weak scale. Large contributions also require near maximal mixing between the strange and bottom squarks, i.e., $|\sin2\theta_L|$ or $|\sin2\theta_R|$ not far below 1. For the left-handed squarks this condition poses an obstacle for model-building due to the requirement that the CKM mixing-angle hierarchy must be reproduced. However, a large mixing between the right-handed strange and bottom squarks poses no such problem. In fact, several SUSY models of flavor utilizing horizontal symmetries exist in which this situation is realized [@nirseiberg; @carone]. It therefore appears unlikely that SUSY contributions to the SM operators could be as large as those to the opposite-chirality operators. Nevertheless, to be fully general we will present numerical results for large left-handed or right-handed mixing. Finally, as mentioned at the outset, significant contributions to the isospin-violating operators require a large mass splitting between the up and down squarks of the first generation. This is possible only in the right-handed sector, since SU(2)$_L$ invariance implies that $m_{\tilde d_L}^2=m_{\tilde u_L}^2$ up to tiny SU(2)$_L$-breaking corrections. Therefore, only $c_1^{\rm EW}$, $c_2^{\rm EW}$, $\widetilde c_3^{\rm EW}$ and $\widetilde c_4^{\rm EW}$ can acquire significant gluino box contributions. The magnitudes of these contributions are symmetric under interchange of $m_{\tilde u_R}^2$ and $m_{\tilde d_R}^2$. One can consider $m_{\tilde d_R}^2\gg m_{\tilde u_R}^2$ and small $m_{\tilde u_R}^2$, or vice versa.
We are now ready to present our numerical results. We begin with the SUSY contributions to the penguin coefficients at the SUSY matching scale, which for simplicity we take to be $m_W$, thus ignoring the slow running of $\alpha_s$ and superpartner masses above the weak scale. We find that the QCD coefficients obey the approximate scaling relation $c_2^{\rm QCD}\sim
c_4^{\rm QCD}\sim -3 c_1^{\rm QCD}\sim -3 c_3^{\rm QCD}$ (provided we use the same masses for left- and right-handed squarks), which according to (\[gluinoQCD\]) is exact for the contributions of the penguin diagrams but only approximate for the box diagrams. The coefficients $\widetilde c_i^{\rm QCD}$ are the same as the $c_i^{\rm QCD}$ if all labels $L\leftrightarrow R$ are interchanged. In the left-hand plot in Figure \[fig:c2\] we show the largest coefficient, $c_2^{\rm QCD}$, for a common mass $m_{\tilde b_L}=m_{\tilde g}=250$GeV as a function of the mass splitting between the left-handed strange and bottom squarks and of the common mass $m_{\tilde u_R}=m_{\tilde d_R}$. Note that only the box contributions in (\[gluinobox\]) depend on the latter two masses, but not the penguin contributions in (\[gluinoQCD\]). We find that these two contributions interfere destructively. For small up- and down-squark masses the box contributions are dominant, whereas for large masses the boxes decouple and the penguin contributions dominate. For intermediate masses there is a region with large destructive interference, where $c_2^{\rm QCD}$ vanishes or takes small values. Note that typical values of the coefficients $c_i^{\rm QCD}$ are of order few times $10^{-4}$ (provided the gluino is as light as 250GeV and there is sufficient mass splitting between the strange and bottom squarks), which is an order of magnitude less than the typical size of QCD penguin coefficients in the SM [@Heff]. If the gluino mass is increased and all mass ratios remain the same, then the SUSY contributions to the Wilson coefficients decrease, scaling like $(250\,\mbox{GeV}/m_{\tilde g})^2$.
We next turn to the coefficients of the electroweak penguin operators, which only receive contributions from the gluino box diagrams. As mentioned above, because of SU(2)$_L$ symmetry only $c_1^{\rm EW}$, $c_2^{\rm EW}$, $\widetilde c_3^{\rm EW}$ and $\widetilde c_4^{\rm EW}$ are important, and we find that for equal strange–bottom mixing and mass splitting in the left-handed and right-handed squark sectors they roughly scale according to $c_2^{\rm EW}\sim \widetilde c_4^{\rm EW}\sim
-3 c_1^{\rm EW}\sim -3\widetilde c_3^{\rm EW}$. In the right-hand plot in Figure \[fig:c2\] we show the largest coefficient, $c_2^{\rm EW}$, as a function of the mass splittings between the left-handed strange and bottom squarks and between the right-handed up and down squarks. Provided both splittings are significant, the typical values of the electroweak penguin coefficients are of order few times $10^{-4}$, which is comparable with the value of the coefficient $c_3^{\rm EW}$ in the SM, given in (\[SMinit\]). Therefore, in certain regions of SUSY parameter space there can be important isospin-violating contributions to the parameters $a$ and $b$ entering the $B^\pm\to\pi K$ decay amplitudes. An important point to notice is that SUSY contributions to the electroweak penguin coefficients are typically of same order as, or can be larger than, the contributions to the QCD penguin coefficients, if there is sufficient mass splitting between the right-handed up and down squarks.
The SUSY contributions to the parameters $a$ and $b$ can be decomposed as $$\begin{aligned}
\label{ALAR}
&&(a_{\rm NP}+ib)_{\rm SUSY} \\
&&= A_L\sin2\theta_L\,e^{i\delta_L}
+ A_R\sin2\theta_R\,e^{i\delta_R} \,, \nonumber\end{aligned}$$ where $A_L$ receives contributions from the electroweak penguin coefficients $c_1^{\rm EW}$ and $c_2^{\rm EW}$, and $A_R$ receives contributions from $\widetilde c_3^{\rm EW}$ and $\widetilde c_4^{\rm EW}$. As previously mentioned, to obtain large values of these parameters requires a significant mass splitting between the right-handed up and down squarks, as well as a substantial splitting between the left- or right-handed strange and bottom squarks. Exchanging strange and bottom or up and down squarks leaves the results invariant up to a sign. In Figure \[fig:abSUSY\] we show the values of $A_L$ and $A_R$ versus $m_{\tilde d_R}-m_{\tilde u_R}$ for two choices of the strange–bottom splitting, such that $m_{\tilde s_{L,R}}/m_{\tilde b_{L,R}}=2$ or 4. We note that for large mass splittings the magnitude of $A_L$ can be up to twice the SM parameter $\delta_{\rm EW}\approx 0.64$, whereas for more moderate splittings $|A_L|\sim\delta_{\rm EW}$ can be obtained. The magnitude of the parameter $A_R$ is typically smaller by a factor of 3. According to Figure \[fig:shifts\], SUSY contributions of this size could lead to shifts of up to $\pm 50^\circ$ in the extracted value of $\gamma_{\pi K}$. Even in the more realistic case where only right-handed strange–bottom squark mixing is large, shifts of up to $\pm 25^\circ$ are possible.
Let us now briefly discuss SUSY contributions to the parameter $\rho$ describing CP-violating but isospin-conserving New Physics effects in $B^\pm\to\pi K$ decays. From (\[rhobound\]) it follows that QCD penguin coefficients of order few times $10^{-4}$ can only lead to rather small values $|\rho|\lsim 0.1$. In analogy with (\[ALAR\]), we define $$\begin{aligned}
&&(\sin\varphi)_{\rm SUSY} \\
&&= S_L\sin2\theta_L\sin\delta_L
- S_R\sin2\theta_R\sin\delta_R \,, \nonumber\end{aligned}$$ where $S_L$ ($S_R$) depends on the mass splitting between the left-handed (right-handed) strange and bottom squarks, and both quantities depend on the masses of the left- and right-handed down squarks. In Figure \[fig:rhoSUSY\] we show the values of these quantities versus the common down-squark mass for two choices of the strange–bottom splitting. We see that, indeed, typical values of $S_L$ and $S_R$ are of order 0.1 or less. The corresponding contributions to $\rho$ are of the same order for maximal weak phases and large strange–bottom mixing. According to Figure \[fig:shifts\], SUSY contributions of this size can only lead to insignificant shifts in the extracted value of $\gamma_{\pi K}$.
Thus far we have only considered contributions to $\rho$ due to the four-quark penguin operators. The largest possible contributions in fact arise if the coefficient $C_{8g}$ of the chromomagnetic dipole operator, or the coefficient $\widetilde C_{8g}$ of the corresponding dipole operator with opposite chirality, have very large magnitudes compared with their values in the SM, implying enhanced $\bar b\to\bar sg$ transitions. This scenario has been discussed in the context of the low semileptonic branching ratio and charm yield in $B$ decays [@George; @Alex; @kaon] and the large $B\to X_s\,\eta'$ branching ratio [@houtseng; @SoniA; @Alexagain]. In SUSY models it is most easily realized via gluino–squark loops containing left–right strange–bottom squark mass insertions [@glue1; @Alex; @CGGi; @berto91]. Constraints on these graphs from $B\to X_s\,\gamma$ decays allow for $\mbox{B}(B\to X_{sg})\lsim 10\%$, which corresponds to $(|C_{8g}|^2+|\widetilde C_{8g}|^2)^{1/2}\approx 1$ (at the scale $m_b$), together with possibly large CP-violating phases in these coefficients [@neub]. From (\[rhobound\]) it follows that large values $\rho=O(1)$ can be obtained in such a scenario. According to Figure \[fig:shifts\], in this extreme case large shifts in the value of $\gamma_{\pi K}$ caused by isospin-conserving New Physics are not excluded.
At present, there are no significant phenomenological constraints on the angles $\theta_L$ and $\theta_R$ parameterizing the mixing between the strange and bottom squarks. In particular, the measured $B\to X_s\,\gamma$ branching ratio does not impose a useful constrain on these parameters. However, an important constraint would emerge if in the future $B_s$–$\bar B_s$ mixing were found to be consistent with, or not much larger than, its predicted value in the SM. For simplicity, we assume that only one of the two mixing angles is large. In the case of left-handed squark mixing, for instance, the relevant gluino box contribution to the $B_s$–$\bar B_s$ mass difference $\Delta m_s$, normalized to the SM contribution, is given by [@berto91] $$\begin{aligned}
\left| \frac{\Delta m_s^{LL}}{\Delta m_s^{\rm SM}} \right|
&=& \frac{\sin^2\!2\theta_L}{|\lambda_t|^2}\,
\frac{\alpha_s^2}{\alpha_W^2}\,\frac{m_W^2}{m_{\tilde g}^2}\,
\frac{1}{C(x_t)}
\nonumber\\
&\times& \Bigg| \frac{11}{18} \Big[
G(x_{\tilde b_L\tilde g},x_{\tilde b_L\tilde g})
+ G(x_{\tilde s_L\tilde g},x_{\tilde s_L\tilde g})
- 2 G(x_{\tilde b_L\tilde g},x_{\tilde s_L\tilde g}) \Big]
\nonumber\\
&&\mbox{}- \frac{2}{9} \Big[
F(x_{\tilde b_L\tilde g},x_{\tilde b_L\tilde g})
+ F(x_{\tilde s_L\tilde g},x_{\tilde s_L\tilde g})
- 2 F(x_{\tilde b_L\tilde g},x_{\tilde s_L\tilde g}) \Big]
\Bigg| \,,\end{aligned}$$ where $x_t=m_t^2/m_W^2$, and $$C(x) = \frac{x^4-12x^3+15x^2-2x+6x^3\ln x}{4(x-1)^3} \,.$$ For right-handed squark mixing the ratio $|\Delta m_s^{RR}/
\Delta m_s^{\rm SM}|$ would be obtained from the above result via the substitution $L\to R$.
In Figure \[fig:Bsmix\] the ratio of the SUSY contribution to $\Delta m_s$ to the SM result is shown as a function of the mass splitting between strange and bottom squarks. The same plot with obvious substitutions $L\to R$ applies to the case of mixing between the right-handed squarks. We observe that $\Delta m_s$ would greatly exceed the SM value in regions of parameter space associated with large SUSY contributions to the penguin coefficients. To gauge the potential impact of a $\Delta m_s$ measurement near the predicted SM value we impose as an example the hypothetical constraint that $|\Delta m_s^{LL}/
\Delta m_s^{\rm SM}|\le 2$. According to Figure \[fig:Bsmix\], it then follows that, e.g., $|\sin2\theta_L|<0.52$ for $m_{\tilde s_L}=2 m_{\tilde b_L}=500$GeV, and $|\sin2\theta_L|<0.28$ for $m_{\tilde s_L}=4 m_{\tilde b_L}
=1000$GeV. Hence, if such a constraint would have to be imposed in the future, the allowed magnitude of the SUSY contributions to the penguin coefficients would be reduced by a significant amount. (Note, however, that the coefficients of the chromomagnetic dipole operators are very weakly constrained by $B_s$–$\bar B_s$ mixing.)
Finally, we comment on implications of naturalness for the large right-handed up–down squark mass splitting necessary to obtain sizable SUSY contributions to the electroweak penguin coefficients. Following [@dimo] we note that in models in which SUSY is broken at high energies, e.g., supergravity, there is a naturalness constraint on the squark and slepton mass spectrum coming from the hypercharge $D$-term. In these models the $Z$-boson mass is given by $$\label{MZsusy}
\frac{m_Z^2}{2}
= \frac{m_1^2-m_2^2\tan^2\!\beta}{\tan^2\!\beta -1} \,,$$ where $m_1^2$, $m_2^2$ and $\tan\beta$ are the usual parameters of the Higgs potential in the minimal SUSY extension of the SM [@Hunter]. The renormalization-group equations give $$\begin{aligned}
\label{Dterms}
m_2^2 &=& -\frac{1-Z_1}{22}\,\mbox{Tr}\,\Big(
m_{\tilde Q_L}^2 + m_{\tilde d_R}^2
- 2 m_{\tilde u_R}^2 - m_{\tilde L_L}^2
+ m_{\tilde e_R}^2 \Big) + \hat m_2^2 \,, \nonumber\\
Z_1 &=& \left( 1 + \frac{33}{20\pi}\,\alpha_{\rm GUT}
\ln\frac{M_{\rm GUT}^2}{m_{\rm SUSY}^2} \right)^{-1}
\approx 0.4 \,, \end{aligned}$$ where $m_A^2$ ($A=\tilde Q_L, \tilde u_R, \tilde d_R, \tilde L_L,
\tilde e_R$) are the values of the squark and slepton masses at the GUT scale, $\hat m_2^2$ contains the dependence on the other soft SUSY breaking masses, and the trace is taken over generation space. It is clear from (\[Dterms\]) that a large mass splitting between the first generation right-handed up and down squarks poses a potential naturalness problem. However, the hypercharge $D$-term can vanish in GUT theories in which hypercharge is embedded in the GUT group. For example, in SU(5) one has the relations $$m_{\tilde Q_L}^2 = m_{\tilde u_R}^2 = m^2_{\tilde e_R}
\equiv m_{10}^2 \,, \qquad
m_{\tilde d_R}^2 = m_{\tilde L_L}^2 = m_{\bar 5}^2 \,,$$ so that it is possible to have large up–down squark-mass splitting without encountering difficulties with naturalness.
To summarize, we have seen that SUSY contributions to the Wilson coefficients of the penguin operators in the effective Hamiltonian for $\bar b\to\bar s q\bar q$ transitions, and in particular of the isospin-violating electroweak penguin operators, can be substantial if the gluino and certain squarks have masses near the weak scale, and other squarks have masses near a TeV. Large left-handed or right-handed strange–bottom squark mixing is also required. The latter option is naturally realized in certain SUSY theories of flavor.
Two-Higgs–doublet models
------------------------
In extensions of the SM containing charged Higgs bosons [@Hunter] there are new photon and Z penguin diagrams contributing to the Wilson coefficients of the penguin operators. Here we consider a general class of two-Higgs–doublet models (2HDMs) discussed in [@Wolf], which contains the conventional type-1 and type-2 2HDMs as special cases. We find that, if terms of order $m_b/m_t$ are neglected, the new penguin contributions only involve the $H t_R b_L$ coupling. Following [@Wolf] we write the corresponding term in the Lagrangian as $${\cal L}_{H t_R b_L} = - \xi_t\,m_t\,
\bar b_{Lj} V_{ji}^\dagger\,t_{Ri}\,H^- + \mbox{h.c.},$$ where $V_{ij}$ is the CKM matrix. In principle, the parameter $\xi_t$ may contain a CP-violating phase. However, the penguin contributions only depend on $|\xi_t|^2$, and thus any weak phase would cancel out. This conclusion holds true even in a wider class of multi-Higgs models [@YuH]. It is therefore sufficient to focus on the conventional type-1 or type-2 two-Higgs–doublet models, for which $|\xi_t|^2=\cot^2\!\beta$.
It follows from the above discussion that in 2HDMs there are no New Physics contributions to the CP-violating parameters $\rho$ and $b$ entering the parametrization of the $B^\pm\to\pi K$ decay amplitudes in (\[ampls\]). Therefore, it is sufficient for our purposes to focus on the electroweak penguin coefficients, which induce a new contribution to the parameter $a$. Including both the photon and Z penguin diagrams, one obtains at the weak scale [@1Higgs; @MUHiggs] $$\begin{aligned}
c_1^{\rm EW} &=& \frac{\alpha\lambda_t}{8\pi} \cot^2\!\beta
\left[ x_t\,f(x_{tH}) + \frac19\,g(x_{tH}) \right]
\,, \nonumber\\
c_3^{\rm EW} &=& \frac{\alpha\lambda_t}{8\pi} \cot^2\!\beta
\left[ - x_t\cot^2\!\theta_W\,f(x_{tH})
+ \frac19\,g(x_{tH}) \right] \,,\end{aligned}$$ where $x_{tH}=(m_t/m_{H^+})^2$, and $$\begin{aligned}
f(x) &=& \frac{x}{1-x} + \frac{x\ln x}{(1-x)^2} \,,
\nonumber\\
g(x) &=& \frac{38x-79x^2+47x^3}{6(1-x)^3}
+ \frac{4x-6x^2+3x^4}{(1-x)^4} \ln x \,.\end{aligned}$$
In this model there is a simple result for the New Physics contribution to the parameter $a$, normalized to the SM contribution. We find $$\frac{a_{\rm NP}}{\delta_{\rm EW}}
= - \cot^2\!\beta\,
\frac{f(x_{tH})}{1+\frac{3\ln x_t}{x_t-1}}
\left\{ 1 - 0.74\sin^2\!\theta_W \left[ 1
+ \frac{g(x_{tH})}{9x_t\,f(x_{tH})} \right] \right\} \,,$$ where the first term in parenthesis is free of hadronic uncertainties, and the second term is numerically small.
In Figure \[fig:Higgs\], we show the ratio $a_{\rm NP}/\delta_{\rm EW}$ in units of $\cot^2\!\beta$ as a function of the Higgs mass. Even for $\cot\beta=1$ a significant contribution to the parameter $a$ requires a small Higgs mass. This possibility is not excluded by direct searches, however in the context of specific models the constraint from the $B\to X_s\,\gamma$ branching ratio often favors a larger mass and $\cot^2\!\beta<1$ [@JoAnne; @BarPh]. Therefore, it appears unlikely that a large new contribution to $a$ can be obtained in 2HDMs.
Models with anomalous gauge-boson couplings
-------------------------------------------
The SU(2)$_L\times \mbox{U(1)}_Y$ gauge symmetry of the SM fully determines the form of the dimension-4 operators that describe the vector-boson self-couplings. New Physics may induce anomalous couplings of the electroweak gauge bosons, which at low energies give rise to higher-dimensional operators, whose effects are suppressed by inverse powers of the New Physics scale $\Lambda$. The effects of anomalous gauge-boson couplings on rare $B$ decays have been investigated in [@bsnn; @Gusta], and their impact on the determination of $\gamma$ from $B^\pm\to\pi K$ decays has been discussed in [@anom]. A general parametrization of the anomalous gauge-boson couplings can be found in [@HPZK]. In low-energy processes, the four new parameters that enter are $\Delta\kappa^\gamma$, $\Delta g_1^Z$, $\lambda^\gamma$ and $g_5^Z$. The first two represent corrections to couplings already present in the SM, whereas the latter two refer to new vertices. Note that these four parameters are real and thus can only contribute to the quantity $a$, but not to $\rho$ and $b$. As in the previous section, we therefore focus only on the coefficients of the electroweak penguin operators. They are [@He-eps] $$\begin{aligned}
c_1^{\rm EW} &=& \frac{\alpha\lambda_t}{8\pi}\,x_t \Big[
\cos^2\!\theta_W\,f_A(x_t) + h_A(x_t) \Big] \,, \nonumber\\
c_3^{\rm EW} &=& \frac{\alpha\lambda_t}{8\pi}\,x_t \left[
- \cot^2\!\theta_W \cos^2\!\theta_W\,f_A(x_t)
+ h_A(x_t) \right] \,,\end{aligned}$$ with $$\begin{aligned}
f_A(x) &=& -3\Delta g_1^Z \ln\frac{\Lambda^2}{m_W^2}
+ \frac{6 g_5^Z}{1-x} \left( 1 + \frac{x\ln x}{1-x}
\right) \,, \nonumber\\
h_A(x) &=& \frac{\Delta\kappa^\gamma}{2}
\ln\frac{\Lambda^2}{m_W^2}
+ \lambda^\gamma \left[ \frac{1-3x}{(1-x)^2}
- \frac{2x^2\ln x}{(1-x)^3} \right] \,, \end{aligned}$$ where the New Physics scale $\Lambda$ acts as an ultraviolet cutoff.
As in the case of the 2HDMs, there is a very simple result for the New Physics contribution to the parameter $a$, normalized to the SM contribution. We find $$\frac{a_{\rm NP}}{\delta_{\rm EW}}
= - \cos^2\!\theta_W\,
\frac{f_A(x_t)}{1+\frac{3\ln x_t}{x_t-1}}
\left\{ 1 - 0.74\sin^2\!\theta_W \left[ 1
+ \frac{h_A(x_t)}{\cos^2\!\theta_W\,f_A(x_t)} \right]
\right\} \,,$$ where as before the first term in parenthesis is free of hadronic uncertainties. For instance, taking $\Lambda=1$TeV gives $$\label{numbs}
\frac{a_{\rm NP}}{\delta_{\rm EW}}
\approx 4.20\Delta g_1^Z - 0.45 g_5^Z
+ 0.19\Delta\kappa^\gamma + 0.03\lambda^\gamma \,.$$ From naive dimensional analysis, one expects that $\Delta g_1^Z,\Delta\kappa^\gamma\sim(g_W v/\Lambda)^2\sim
10^{-2}$ (with $v\approx 246$GeV the Higgs vacuum expectation value), whereas $g_5^Z$ and $\lambda^\gamma$ are expected to be further suppressed [@Gusta; @DaVa]. Potentially the most important contribution in (\[numbs\]) is due to $\Delta g_1^Z$, which is bounded by experiment to lie in the range $-0.113<\Delta g_1^Z<0.126$ [@LEP]. Even when this bound is saturated, $a_{\rm NP}/\delta_{\rm EW}$ cannot exceed 0.5 in magnitude. However, from naive dimensional analysis $a_{\rm NP}$ is naturally an order of magnitude smaller.
Conclusions {#sec:concl}
===========
We have explored how New Physics could affect purely hadronic FCNC transitions of the type $\bar b\to\bar s q\bar q$ focusing, in particular, on isospin violation. Unlike in the Standard Model, where isospin-violating effects in these processes are strongly suppressed by electroweak gauge couplings or small CKM matrix elements, in many New Physics scenarios these effects are not parametrically suppressed relative to isospin-conserving FCNC processes. In the language of effective weak Hamiltonians, this implies that the Wilson coefficients of QCD and electroweak penguin operators are of a similar magnitude. For a large class of New Physics models, we find that the coefficients of the electroweak penguin operators are, in fact, due to “trojan” penguins, which are neither related to penguin diagrams nor of electroweak origin.
We have calculated the Wilson coefficients of the penguin operators in the effective weak Hamiltonian in several New Physics models, extending the usual operator basis where appropriate. We have also included penguin operators mediating the decay $\bar b\to\bar d s\bar d$, which is highly suppressed in the Standard Model. Specifically, we have considered: (a) models with tree-level FCNC couplings of the $Z$ boson, extended gauge models with an extra $Z'$ boson, SUSY models with broken R-parity; (b) SUSY models with R-parity conservation; (c) two-Higgs–doublet models, and models with anomalous gauge-boson couplings. In case (a), the resulting electroweak penguin coefficients can be much larger than in the Standard Model because they are due to tree-level processes. In case (b), these contributions can compete with the Standard Model coefficients because they arise from strong-interaction box diagrams, which scale relative to the Standard Model like $(\alpha_s/\alpha)(m_W^2/m_{\rm SUSY}^2)$. In models (c), on the other hand, isospin-violating New Physics effects are not parametrically enhanced and are generally smaller than in the Standard Model.
We have focused on the rare hadronic decays $B^\pm\to\pi K$, which are particularly sensitive to isospin-violating effects. These decays are especially useful for probing New Physics contributions, since in the Standard Model the theoretical description of such effects is very clean. We have found that the ratio $R_*$ of the CP-averaged branching ratios defined in (\[Rstexp\]) and the value of the weak phase $\gamma_{\pi K}$ extracted from $B^\pm\to\pi K$ decays are the most useful observables for probing isospin-violating New Physics contributions. Using a fully general parametrization of the decay amplitudes, we have derived model-independent bounds on $R_*$ in the presence of New Physics, and have investigated by how much $\gamma_{\pi K}$ could differ from the true value of the CKM angle $\gamma$. We have seen that, depending on the measured value of $R_*$, it may be possible to unambiguously distinguish between isospin-violating and isospin-conserving New Physics contributions. Irrespective of the value of $R_*$, we find that large shifts in $\gamma$ can be caused by even moderate isospin-violating contributions to the decay amplitudes of order 10%. In contrast, significant shifts due to isospin-conserving New Physics effects would require a new $O(1)$ CP-violating contribution to the amplitudes.
For each New Physics model we have explored which regions of parameter space can be probed by the $B^\pm\to\pi K$ observables, and how big a departure from the Standard Model predictions one can expect under realistic circumstances. In Table \[tab:1\], we summarize our estimates of the maximal isospin-violating and isospin-conserving contributions to the decay amplitudes, as parameterized by $|a_{\rm NP}+ib|$ and $|\rho|$, respectively. For comparison, we recall that in the Standard Model $a\approx 0.64$ and $b\approx\rho\approx
0$. We also list the corresponding maximal values of the difference $|\gamma_{\pi K}-\gamma|$. As noted above, in models with tree-level FCNC couplings New Physics effects can be dramatic, whereas in SUSY models with R-parity conservation isospin-violating loop effects can be competitive with the Standard Model. In the case of SUSY models with R-parity violation, we have derived interesting bounds on combinations of the trilinear couplings $\lambda_{ijk}'$ and $\lambda_{ijk}''$, which are given in (\[bounds1\]) and (\[bounds2\]).
It is worth pointing out that isospin- or, more generally, SU(3) flavor-violating New Physics effects in hadronic weak decays could also be important in other processes. For instance, they have been shown to yield a potentially large contribution to the quantity $\epsilon'/\epsilon$ in $K\to\pi\pi$ decays [@epspr]. Moreover, there are other $B$ and $B_s$ decay channels that could be sensitive to flavor-violating New Physics contributions. We look forward to returning to this subject in an extra dimension.
Y.G. and M.N. are supported by the Department of Energy under contract DE–AC03–76SF00515, and A.K. under Grant No. DE-FG02-84ER40153.
[99]{}
A documentation of this measurement can be found at\
[http://www-cdf.fnal.gov/physics/new/bottom/cdf4855]{}.
G.D. Barr et al. (NA31 Collaboration), .
L.K. Gibbons et al. (E731 Collaboration), .
A. Alavi-Harati et al. (KTeV Collaboration), .
A documentation of this measurement can be found at [http://www.cern.ch/NA48]{}.
For a review, see: Y. Grossman, Y. Nir and R. Rattazzi, in: [*Heavy Flavours*]{} (Second Edition), A.J. Buras and M. Lindner eds. (World Scientific, Singapore, 1998) pp. 755 \[\].
For a review, see: [*The BaBar Physics Book*]{}, P.F. Harison and H.R. Quinn eds., SLAC Report No. SLAC-R-504 (1998),\
[ http://www.slac.stanford.edu/pubs/slacreports/slac-r-504]{}.
M. Neubert and J.R. Rosner, \[\];\
\[\].
M. Neubert, \[hep-ph/9812396\].
For a review, see: G. Buchalla, A.J. Buras and M.E. Lautenbacher, \[\].
R. Fleischer and T. Mannel, \[\].
R. Fleischer, \[\];\
A.F. Buras and R. Fleischer, [*A general analysis of $\gamma$ determinations from $B\to\pi K$ decays*]{}, Preprint CERN-TH-98-319, .
M. Neubert, [*Exploring the weak phase $\gamma$ in $B^\pm\to\pi K$ decays*]{}, Preprint SLAC-PUB-8122, , to be published in the Proceedings of the 17th International Workshop on Weak Interactions and Neutrinos (WIN 99), Cape Town, South Africa, 24–30 January 1999.
Y. Gao and F. Würthwein (representing the CLEO Collaboration), [*Charmless hadronic $B$ decays at CLEO*]{}, Preprint CALT-68-2220, , to be published in the Proceedings of the American Physical Society Meeting of the Division of Particles and Fields (DPF 99), Los Angeles, CA, 5–9 January 1999.
A.L. Kagan and M. Neubert, [*Large $\Delta I=3/2$ contribution to $\epsilon'/\epsilon$ in supersymmetry*]{}, Preprint SLAC-PUB-8231, , to appear in [*Phys. Rev. Lett.*]{}
A.J. Buras and L. Silvestrini, \[\].
R. Fleischer and J. Matias, [*Searching for New Physics in nonleptonic $B$ decays*]{}, Preprint CERN-TH-99-164, .
D. Choudhury, B. Dutta and A. Kundu, \[\].
X.-G. He, C.-L. Hsueh and J.-Q. Shi, [*Constraints on the phase $\gamma$ and New Physics from $B\to\pi K$ decays*]{}, Preprint .
A.J. Buras, M. Jamin and M.E. Lautenbacher, \[\].
A.S. Dighe, M. Gronau and J.L. Rosner, \[\].
N.G. Deshpande and X.-G. He, \[\]; erratum: [*ibid.*]{} [**74**]{} (1995) 4099.
M. Neubert and B. Stech, in: [*Heavy Flavours*]{} (Second Edition), A.J. Buras and M. Lindner eds. (World Scientific, Singapore, 1998) pp. 294 \[\].
M. Beneke, G. Buchalla, M. Neubert and C.T. Sachrajda, \[\], and work in preparation.
R. Poling, Rapporteur’s talk presented at the 19th International Symposium on Lepton and Photon Interactions at High Energies, Stanford, California, 9–14 August 1999;\
Y. Kwon et al. (CLEO Collaboration), Conference contribution CLEO CONF 99-14.
B. Blok, M. Gronau and J.L. Rosner, \[\];\
M. Gronau and J.L. Rosner, \[\].
A.J. Buras, R. Fleischer and T. Mannel, \[\].
J.M. Gérard and J. Weyers, \[\];\
D. Delepine, J.M. Gérard, J. Pestieau and J. Weyers, \[\].
M. Neubert, \[\].
A.F. Falk, A.L. Kagan, Y. Nir and A.A. Petrov, \[\].
D. Atwood and A. Soni, \[\].
R. Fleischer, \[\].
S. Bertolini, F. Borzumati and A. Masiero, .
B. Grzadkowski and W.-S. Hou, .
A.L. Kagan, \[\].
M. Ciuchini, E. Gabrielli and G.F. Giudice, \[\]; erratum: [*ibid.*]{} [**393**]{} (1997) 489.
See, e.g.: Y. Grossman, Y. Nir, S. Plaszczynski and M.-H. Schune, \[\].
For a feasibility study, see: A. Soffer, \[\].
K. Huitu, C.-D. Lu, P. Singer and D.-X. Zhang, \[\]; \[\].
Y. Grossman, Z. Ligeti and E. Nardi, \[\]; erratum: [*ibid.*]{} [**480**]{} (1996) 753.
C. Caso et al. (Particle Data Group), .
See, e.g.: T.G. Rizzo, \[\], and references therein.
C.E. Carlson, P. Roy and M. Sher, \[\].
G. Bhattacharyya and A. Datta, \[\].
G. Bhattacharyya and A. Raychaudhuri, \[\].
S. Bertolini, F. Borzumati, A. Masiero and G. Ridolfi, .
P. Cho, M. Misiak and D. Wyler, \[\].
F. Gabbiani, E. Gabrielli, A. Masiero and L. Silvestrini, \[\].
J. Louis and Y. Nir, \[\].
A.L. Kagan, in: Proceedings of the International Workshop on [*Recent Advances in the Superworld*]{}, Woodlands, Texas, April 1993, edited by J.L. Lopez and D.V. Nanopoulos (World Scientific, Singapore, 1994) pp. 215.
M. Dine, A.L. Kagan and R. Leigh, \[\].
Y. Nir and N. Seiberg, \[\].
For a recent compilation of references, see: L. Randall and S. Su, \[\].
R. Barbieri and A. Strumia, \[\].
S.A. Abel, W.N. Cottingham and I.B. Whittingham, \[\].
C.D. Carone, L.J. Hall and T. Moroi, \[\].
A.L. Kagan and J. Rathsman, [*Hints for enhanced $b\to sg$ from charm and kaon counting*]{}, Preprint .
W.-S. Hou and B. Tseng, \[\].
D. Atwood and A. Soni, \[\].
A.L. Kagan and A.A. Petrov, [*$\eta'$ production in $B$ decays: Standard Model versus New Physics*]{}, Preprint ;\
A.L. Kagan, [*The phenomenology of enhanced $b\to sg$*]{}, Preprint UCTP-107-98, , to be published in the Proceedings of the 7th International Symposium on Heavy Flavor Physics, Santa Barbara, CA, 7–11 July 1997.
A.L. Kagan and M. Neubert, \[\]; \[\].
S. Dimopoulos and G.F. Giudice, \[\].
J.F. Gunion, H.E. Haber, G.L. Kane and S. Dawson, [*The Higgs Hunter’s Guide*]{}, (Addison-Wesley, Reading, MA, 1990), and references therein.
L. Wolfenstein and Y.L. Wu, \[\].
Y. Grossman, \[\].
W.-S. Hou and R.S. Willey, ; .
G. Buchalla et al., .
J.L. Hewett, \[\].
V. Barger, M.S. Berger and R.J.N. Phillips, \[\].
G. Burdman, \[\].
K. Hagiwara, R.D. Peccei, D. Zeppenfeld and K. Hikasa, .
X.-G. He and B.H.J. McKellar, \[\];\
X.-G. He, \[\].
S. Dawson and G. Valencia, \[\]; \[\]; \[\].
S. Dhamotharan et al. (ALEPH Collaboration), [*Measurement of triple gauge-boson couplings at 183–189GeV*]{}, Conference contribution ALEPH 99-019, CONF 99-014.
[^1]: On leave from: Newman Laboratory of Nuclear Studies, Cornell University, Ithaca, NY 14853
[^2]: If this were not the case, the New Physics impact on $\bar\varepsilon_{3/2}$ would provide us with another handle on non-standard isospin-violating effects.
[^3]: According to [@fact], naive factorization gives the leading term in the heavy-quark limit.
[^4]: For the related decay $\bar b\to\bar s d\bar s$, this possibility has been explored in [@bssd].
[^5]: The largest entries in the absence of flavor symmetries are suppressed by a factor $m_b/m_{\rm SUSY}$.
[^6]: Although up–top squark mixing can be large in supergravity theories with arbitrary Kähler potential, it would not modify our conclusions qualitatively. Furthermore, a model which would admit large up–top squark mixing but satisfy the $D$–$\bar D$ constraint on up–charm squark mixing would have to be very contrived.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Substantial effort has been devoted in determining the ideal proxy for quantifying the morphology of the hot intracluster medium in clusters of galaxies. These proxies, based on X-ray emission, typically require expensive, high-quality X-ray observations making them difficult to apply to large surveys of groups and clusters. Here, we compare optical relaxation proxies with X-ray asymmetries and centroid shifts for a sample of SDSS clusters with high-quality, archival X-ray data from *Chandra* and *XMM-Newton*. The three optical relaxation measures considered are: the shape of the member-galaxy projected velocity distribution – measured by the Anderson-Darling (AD) statistic, the stellar mass gap between the most-massive and second-most-massive cluster galaxy, and the offset between the most-massive galaxy (MMG) position and the luminosity-weighted cluster centre. The AD statistic and stellar mass gap correlate significantly with X-ray relaxation proxies, with the AD statistic being the stronger correlator. Conversely, we find no evidence for a correlation between X-ray asymmetry or centroid shift and the MMG offset. High-mass clusters ($M_\mathrm{halo} > 10^{14.5}{\,\mathrm{M_{\sun}}}$) in this sample have X-ray asymmetries, centroid shifts, and Anderson-Darling statistics which are systematically larger than for low-mass systems. Finally, considering the dichotomy of Gaussian and non-Gaussian clusters (measured by the AD test), we show that the probability of being a non-Gaussian cluster correlates significantly with X-ray asymmetry but only shows a marginal correlation with centroid shift. These results confirm the shape of the radial velocity distribution as a useful proxy for cluster relaxation, which can then be applied to large redshift surveys lacking extensive X-ray coverage.'
author:
- |
Ian D. Roberts$^{1}$[^1], Laura C. Parker$^{1}$, Julie Hlavacek-Larrondo$^{2}$\
$^{1}$Department of Physics and Astronomy, McMaster University, Hamilton ON L8S 4M1, Canada\
$^{2}$Département de Physique, Université de Montréal, Montréal QC H3C 3J7, Canada\
bibliography:
- 'CompleteManuscriptFile\_v3.bib'
date: 'Accepted XXX. Received YYY; in original form ZZZ'
title: 'Connecting optical and X-ray tracers of galaxy cluster relaxation'
---
\[firstpage\]
galaxies: clusters: general – galaxies: evolution – galaxies: groups: – galaxies: statistics
Introduction {#sec:introduction}
============
The majority of galaxies in the local Universe do not evolve in isolation but instead inhabit dense environments such as groups and clusters [e.g. @geller1983; @eke2005]. In addition to internal processes (e.g. AGN feedback, @dubois2013 [@gurkan2015; @mullaney2015; @bongiorno2016]; bar-driven evolution, @knapen1995 [@kormendy2004; @sheth2005]; morphological quenching, @martig2009; virial gas heating, @birnboim2003 [@cattaneo2006; @gabor2015]; etc.), interactions with local environments play a significant role in shaping the observed properties of galaxies. For example, mechanisms acting in dense environments such as ram-pressure stripping [e.g. @gunn1972] and starvation [e.g. @larson1980; @peng2015] can remove the cold and hot gas components from galaxies, respectively. Galaxy interactions, such as mergers and impulsive high-speed encounters, can drive gas to the central regions and induce star-burst events which may exhaust a galaxies gas reserves [e.g. @mihos1994a; @mihos1994b; @ellison2008; @davies2015]. These interactions can also influence galaxy morphology through the growth of a strong bulge component, and the end products of major mergers tend to be bulge dominated galaxies with classical de Vaucouleurs profiles [e.g. @barnes1989]. Finally, tidal interactions can also influence gas content through direct stripping or by transporting gas outwards allowing it to be more easily stripped by other mechanisms [e.g. @mayer2006; @chung2007]. It’s generally accepted that these mechanisms can act on galaxies in dense environments, though the relative balance between different mechanisms in different environments remains an outstanding question.
Understanding the influence of environment is contingent on being able to identify and quantify galaxy environments. Common environmental measures include the projected number density of galaxies out to the $Nth$ nearest neighbour, the halo mass of a host group or cluster, or the projected separation from the centre of a group or cluster. Star formation and morphology of galaxies correlate well with these environment proxies, with galaxies in high densities regions (or alternatively, high halo mass or small group/cluster-centric radius) being preferentially red, passive, and early type [@dressler1980; @goto2003; @poggianti2008; @kimm2009; @li2009; @wetzel2012; @wilman2012; @fasano2015; @haines2015]. An alternative way to parametrize the environment of a host group or cluster, is to classify the degree to which a system is dynamically relaxed. A relaxed, dynamically old group or cluster should be characterized by a central galaxy which is the brightest (most massive) member by a significant margin [e.g @khosroshahi2007; @dariush2010; @smith2010] and is located near the minimum of the potential well (e.g. @george2012 [@zitrin2012], however also see @skibba2011), satellite galaxies which are distributed in velocity space according to a Gaussian profile [e.g. @yahil1977; @bird1993; @hou2009; @martinez2012], and diffuse X-ray emission which is symmetric about the group/cluster centre [e.g. @rasia2013; @weissmann2013; @parekh2015]. The dynamical state of clusters is related to the age of the halo and the time since infall for member galaxies, which simulations have shown is an important quantity in determining the degree to which galaxy properties are affected by environment [e.g. @wetzel2013; @oman2016; @joshi2017]. Unrelaxed groups and clusters are systems which formed more recently or which have recently experienced a significant merger event, and in either case it would be expected that the time-since-infall onto the current halo for member galaxies will be relatively short. Therefore galaxies in unrelaxed groups may have properties which have been less influenced by environment compared to galaxies in more relaxed systems.
Recent studies have attempted to determine the degree to which galaxy properties depend on the “relaxedness” of a given group or cluster. It has been shown that galaxies in relaxed groups tend to be redder than counterparts in unrelaxed systems, using relaxation definitions based on the presence of a well-defined central galaxy [e.g. @carollo2013] as well as the shape of the satellite velocity distribution [e.g. @ribeiro2010; @ribeiro2013a]. Previously, we have shown that low-mass galaxies in the inner regions of Gaussian (G) groups have reduced star-forming fractions relative to non-Gaussian (NG) groups [@roberts2017]. We have also shown that star-forming and disc fractions for low mass galaxies are enhanced in X-ray underluminous (XRW) groups, and show that galaxies XRW groups have velocity distributions consistent with being unrelaxed systems (at least relative to X-ray strong groups, @roberts2016).
Building from our recent work, here we aim to further investigate the connection between X-ray and optical measures of group relaxedness. The shape of the diffuse X-ray component of a group or cluster is among the most direct probes of the degree to which a group/cluster is relaxed or recently disturbed. The downside, however, is that measuring this morphology requires deep, high-quality X-ray observations which are not available for large surveys containing thousands of groups and clusters. To address this challenge, we use a sample of galaxy clusters with existing X-ray observations to investigate the relationship between the X-ray relaxation and three previously used optical probes of relaxation: the shape of the satellite velocity distribution, the stellar mass gap between the most-massive and second-most-massive group galaxy, and the offset between the position of the most-massive galaxy and the luminosity-weighted centre of the group. We determine the effectiveness of these optical relaxation measures (which are applicable to large redshift surveys) by comparing them to measured X-ray morphology, a more direct probe of relaxation.
The outline of this paper is as follows. In Section 2 we describe the optical group catalogue as well as the archival X-ray data used in this work. In Section 3 we outline the cluster relaxation estimators, both optical and X-ray, that we consider. In Section 4 we present the main results, comparing optical and X-ray cluster relaxation measures. In Section 5 we discuss these results and provide a summary in Section 6.
This paper assumes a flat $\mathrm{\Lambda}$ cold dark matter cosmology with $\Omega_\mathrm{M} = 0.3$, $\Omega_\mathrm{\Lambda} = 0.7$, and $H_0 = 70\,\mathrm{km}\,\mathrm{s^{-1}}\,\mathrm{Mpc^{-1}}$. The $h$-dependence of important calculated properties are: $M_\mathrm{halo} \sim h^{-1}$, $M_\mathrm{\star} \sim h^{-2}$, $R_{500} \sim h^{-1}$.
Data {#sec:data}
====
![Cluster halo mass versus redshift for Yang clusters. Stars correspond to the X-ray matched clusters used in this work, coloured by the number of galaxies identified in each system. Gray contours show the distribution for the parent sample of N > 10 Yang clusters.[]{data-label="fig:z_mh"}](z_mh.pdf){width="0.85\columnwidth"}
Optically identified galaxy clusters
------------------------------------
We use galaxy clusters identified from the seventh release of the Sloan Digital Sky Survey (SDSS DR7; @abazajian2009) by @yang2005 [@yang2007] who construct a group sample using a “halo-based" group finder which aims to improve upon the classic friends-of-friends (FoF) algorithm [e.g. @huchra1982; @press1982]. For a full description of the algorithm see @yang2005 [@yang2007], however in short, the groups are initially populated by connecting galaxies through a standard FoF approach (with very small linking lengths) and group memberships are iteritively updated under the assumption that the distribution of galaxies in phase space follows that of a spherical NFW profile [@navarro1997]. Each iteration yields an updated estimate of the group mass, size, and velocity dispersion and iterations continue until memberships stabilize. Final group halo masses ($M_\mathrm{halo}$) obtained via abundance matching are given in the Yang catalogue (in particular, we use the sample III); we use galaxy stellar masses ($M_\star$) given in the New York University Value-Added Galaxy Catalogue (NYU-VAGC; @blanton2005a) determined using fits to the galaxy spectra and broad-band photometric measurements following the procedure of @blanton2007. We note that the Yang catalogue contains a mixture of what would generally be considered groups ($M_\mathrm{halo} < 10^{14}{\,\mathrm{M_{\sun}}}$) as well as galaxy clusters ($M_\mathrm{halo} \ge 10^{14}{\,\mathrm{M_{\sun}}}$), for the sake of brevity we will refer to all systems as clusters regardless of halo mass as the majority of the systems we consider have $M_\mathrm{halo} \ge 10^{14}{\,\mathrm{M_{\sun}}}$.
Cluster-centric radii are computed for galaxies using the redshift and the angular separation between the galaxy position and the luminosity-weighted centre of the cluster. We normalize all cluster-centric radii by $R_{500}$ (the radius at which the average interior density is 500 times the critical density of the Universe) of each cluster which we compute as $$R_{500} = R_{200m} / 2.7,$$ where, $$R_{200m} = 1.61\,\mathrm{Mpc} \left(\frac{M_\mathrm{halo}}{10^{14}\,{\,\mathrm{M_{\sun}}}}\right)^{1/3} (1 + z_\mathrm{group})^{-1}$$ is the radius at which the average interior density is equal to 200 times the critical mass density of the Universe [@yang2007; @tinker2008], and we have assumed an NFW density profile [@navarro1997] with a concentration given by the concentration-mass relation of @maccio2007 [@wang2014].
Our sample of galaxy clusters is a subset of the Yang catalogue including only clusters with ten or more member galaxies (2559 clusters). The cut-off in membership is chosen in order to be able to classify the shape of the velocity profile for each cluster with relative accuracy [@hou2009]. Fig. \[fig:z\_mh\] shows the $M_\mathrm{halo}$ - redshift distribution for the parent sample (gray contours) with the 58 X-ray matched clusters (see Section \[sec:X-ray\_match\]) overplotted as stars colour-coded by the number of galaxies identified in each cluster. As expected, at fixed redshift the observed cluster richness increases with halo mass and at fixed halo mass the observed cluster richness decreases with redshift. The latter is a selection effect due to increasing incompleteness at higher redshift. To check whether this incompleteness may be biasing our results we repeat our analysis on “low-z” ($z < 0.10$) and “high-z” ($z \ge 0.10$) subsamples (results not shown) and find no difference between the conclusions drawn from either redshift subsample. Therefore moving forward, we consider the entire redshift range.
X-ray matched clusters {#sec:X-ray_match}
----------------------
![Distribution of the offset between the X-ray peak and cluster luminosity-weighted centre (left) and X-ray exposure time (right) for the clusters in our sample, for *Chandra* (gray) and *XMM-Newton* (white).[]{data-label="fig:offset_texp"}](offset_texp.pdf){width="\columnwidth"}
In order to make connections between optical measures of cluster relaxedness and the shape of the cluster extended X-ray profile we searched the *Chandra* and *XMM-Newton* science archives at the positions of the luminosity-weighted centres of each of the 2559 $N > 10$ Yang clusters. Using a search radius of $5\arcmin$ we matched observations of extended X-ray emission to the corresponding optically identified cluster, only including observations with clean exposure times $\ge 10\,\mathrm{ks}$. We also exclude systems where multiple Yang (N > 10) clusters are matched to the same X-ray observation to avoid the potential overlap of X-ray emission from physically distinct systems in projection (this was only the case for $<5$ per cent of matches). This matching results in 58 Yang clusters with X-ray coverage. Fig. \[fig:offset\_texp\]a shows the projected separation between the luminosity-weighted centre and the X-ray centre of each Yang cluster for *Chandra* (gray) and *XMM-Newton* (white), whereas Fig. \[fig:offset\_texp\]b shows the respective filtered exposure times for the observations. X-ray centres are calculated as the position of the brightest pixel in the X-ray image after smoothing using a Gaussian kernel with a bandwidth of $40\,\mathrm{kpc}$ (as in @nurgaliev2013). As shown in Fig. \[fig:offset\_texp\]a, the offset between the optical and X-ray centres is far smaller than the virial radius for all systems.
*Chandra* observations were reprocessed, cleaned, and calibrated using the latest version of <span style="font-variant:small-caps;">ciao</span> (<span style="font-variant:small-caps;">ciao</span> version 4.9, <span style="font-variant:small-caps;">caldb</span> version 4.7.5). Charge transfer inefficiency and time-dependent gain corrections were applied and observations were filtered for background flares using the <span style="font-variant:small-caps;">lc\_clean</span> script with a $3\sigma$ threshold. Exposure corrected images are then created using exposure maps generated at an energy of $1.5\,\mathrm{keV}$, the average peak emission of our sample. Images were created in the $0.5-5\,\mathrm{keV}$ energy band to maximize the ratio between cluster and background flux [@nurgaliev2013]. Point sources are identified using the <span style="font-variant:small-caps;">wavdetect</span> script and are filled with local poisson noise using <span style="font-variant:small-caps;">dmfilth</span>, blank sky background images are generated for each observation using the <span style="font-variant:small-caps;">blanksky</span> and <span style="font-variant:small-caps;">blanksky\_image</span> scripts. All observations are then checked by eye to ensure that no obvious point sources were missed by the algorithm. For systems with multiple observations, combined images and exposure maps were generated with the <span style="font-variant:small-caps;">merge\_obs</span> script and blank sky background images were combined using <span style="font-variant:small-caps;">reproject\_image</span>.
Data reduction for *XMM-Newton* observations was done using the Extended Source Analysis Software (<span style="font-variant:small-caps;">esas</span>) within the *XMM-Newton* Science Analysis System (<span style="font-variant:small-caps;">sas</span>, version 16.0.0). Calibrated event files were generated using the <span style="font-variant:small-caps;">emchain</span> script, and filtered event lists were generated using <span style="font-variant:small-caps;">mos-filter</span>. Exposure corrected images were created in the $0.5-5\,\mathrm{keV}$ band, and point sources were identified with the <span style="font-variant:small-caps;">cheese</span> script and subsequently filled with local poisson noise using the <span style="font-variant:small-caps;">ciao</span> script <span style="font-variant:small-caps;">dmfilth</span>. Again, images are checked by eye to ensure no obvious point sources are missed. For *XMM-Newton* observations we only use the MOS exposures to avoid the complications of the many chip gaps on the PN detector. MOS exposures are combined using the <span style="font-variant:small-caps;">comb</span> script to give merged images, exposure maps, and background images. For systems with multiple observations, images, exposure maps, and background images are merged with the <span style="font-variant:small-caps;">ciao</span> script <span style="font-variant:small-caps;">reproject\_image</span>.
The pixel scale of the resulting images is $0.5\arcsec$ for *Chandra* and $2.5\arcsec$ for *XMM-Newton*. We calculate X-ray asymmetries and centroid shifts at these native resolutions to avoid losing information from the higher resolution *Chandra* images, however we note that binning the *Chandra* images to the *XMM-Newton* resolution does not alter the results. Furthermore, when we compare asymmetries and centroid shifts computed for systems which are observed by both *Chandra* and *XMM-Newton*, we see no bias introduced by the resolution difference.
Cluster relaxation measures {#sec:relax_param}
===========================
Optical
-------
In this study we implement three previously used optical measures to parameterize the relaxation of clusters: the Anderson-Darling statistic, the stellar mass ratio between the second most-massive and the most-massive cluster galaxy ($M_2 / M_1$), and the offset between the position of the MMG and the luminosity-weighted centre of the cluster (MMG offset).
### Anderson-Darling statistic
The Anderson-Darling (AD) test is a statistical normality test which measures the “distance" between the cumulative distribution functions (CDFs) corresponding to the data as well as the ideal case of a normal distribution [@anderson1952]. The distance between the CDFs is parameterized by the AD statistic ($A^2$), in the sense that large values of this statistic correspond to larger deviations from normality. The AD statistic, $A^2$ is given by $$A^2 = -n - \frac{1}{n} \sum_{i=1}^n [2i - 1][\ln \Phi(x_i) + \ln(1 - \Phi(x_{n+1-i}))],$$ where $x_i$ are the length-$n$ ordered data and $\Phi(x_i)$ is the CDF of the hypothetical underlying distribution (Gaussian in this application).
In the context of cluster evolution, it is expected that galaxies in evolved, dynamically old clusters should display projected velocity profiles which are well fit by a normal distribution; conversely more unrelaxed clusters will show larger deviations from normality [@yahil1977; @bird1993; @ribeiro2013a]. The AD test can therefore be applied to the velocity distributions of member galaxies to discriminate between relaxed and unrelaxed clusters [e.g. @hou2009]. In this work we use the AD statistic as a proxy for cluster relaxedness, where increasing values of $A^2$ are indicative of progressively more unrelaxed clusters. It is also common in the literature to use the p-value associated with the AD statistic to define a dichotomy between Gaussian and non-Gaussian clusters [e.g @hou2009; @martinez2012; @roberts2017], which we consider in Section \[sec:G\_NG\].
### Stellar mass gap
The second optical parameter we use to classify the relaxation of galaxy clusters is the stellar mass ratio between the second most-massive and most-massive galaxies in a given cluster. Since the MMG should sit near the centre of the cluster potential, it will progressively grow in stellar mass by dominating gas accretion within the cluster, and more importantly, by cannibalizing galaxies through minor mergers [e.g. @delucia2007; @ruszkowski2009; @lin2013; @mcdonald2016]. This MMG mass growth will therefore drive down $M_2/M_1$ in dynamically old clusters, whereas more unrelaxed systems will have had less time to establish a dominant MMG.
The reliability of $M_2/M_1$ is contingent on correctly identifying both the MMG and $M_2$. A particular concern when using SDSS data is the potential for galaxies missing spectra due to fibre collisions; this has an increasing impact in the dense inner regions of groups and clusters where one would expect to find the MMG and $M_2$. In an attempt to mitigate the effects of fibre collisions we use sample III from the Yang group catalogue which corrects for fibre collisions by assigning fibre collision galaxies the redshift of the galaxy they “collide” with. While this procedure accounts for fibre collisions it also introduces potential impurities to the group catalogue (some fibre collision galaxies will have redshifts which are catastrophically different from the one they are assigned), we delay a more detailed discussion of these effects until Section \[sec:mmg\_discussion\] though we urge the reader to keep these caveats in mind when interpretting results in Section \[sec:xray\_opt\_corr\].
### MMG offset
The final optical relaxation parameter we consider is the projected offset between the MMG and the luminosity-weighted cluster centre, $\delta R_{MMG}$. There is currently no consensus regarding the best observational definition of group centre, with the position of the MMG, the position of the X-ray peak, and the luminosity or mass-weighted centre all being popular choices [e.g. @george2012]. For relaxed clusters it is expected that all of the aforementioned centre definitions will be relatively consistent with one another, however more unrelaxed clusters may show significant offsets between different cluster centre choices. In particular, many unrelaxed clusters host MMGs with large offsets from other cluster centre definitions [e.g. @katayama2003; @sanderson2009; @carollo2013; @khosroshahi2017], therefore the offset between MMG position and luminosity-weighted centre can be a useful measure of cluster relaxation.
As with the stellar mass gap, there are potential complications with regards to interpretting the MMG offset as a relaxation probe. For example, it is based on the assumption that in a relatively relaxed systems the MMG (or brightest galaxy) will be located at rest at the centre of the dark matter potential well – the so-called central galaxy paradigm (CGP). However, some recent studies have called into question whether or not the CGP is valid in all systems [@vandenbosch2005; @coziol2009; @skibba2011; @sehgal2013; @lauer2014; @hoshino2015]. Additionally, even in relaxed systems the MMG may oscillate about the centre of a cored dark-matter potential [e.g. @harvey2017] further complicating the interpretation of the radial offset of the central galaxy. Yet again, we will defer a full discussion of these effects to Section \[sec:mmg\_discussion\].
X-ray
-----
{width="80.00000%"}
To measure the degree of cluster relaxedness from X-ray observations we consider two relaxation proxies: the photon asymmetry ($A_\mathrm{phot}$) and the centroid shift ($w$).
### Photon Asymmetry {#sec:photon_asym}
Photon asymmetry is a novel technique to measure the asymmetry of X-ray profiles which is model-independent and robust across a wide range in X-ray counts and background level [@nurgaliev2013]. In this work we will give a brief discussion of the photon asymmetry computation, however for a complete description, including tests of robustness, we direct the reader to @nurgaliev2013.
The photon asymmetry measures the degree to which the count profile of an X-ray observation is axisymmetric around the X-ray peak, or phrased alternatively, the degree to which the polar angles of X-ray counts are distributed uniformly over the range $0 \le \phi \le 2\pi$. This is accomplished quantitatively using Watson’s U2 test which compares the polar angle CDF for observed counts to a uniform CDF corresponding to an idealized axisymmetric profile [@watson1961]. In a given radial annulus, the distance between the observed count distribution and a uniform distribution is given by $$\hat{d}_{N,C} = \frac{N}{C^2} \left(U_N^2 - \frac{1}{12} \right),$$ where $N$ is the total number of counts within the annulus, $C$ is the number of counts intrinsic to the cluster (ie. above the background) within the annulus, and $U_N^2$ is Watson’s statistic. We follow @nurgaliev2013 and compute Watson’s statistic using the following relation (see @watson1961) $$U_N^2(\phi_0) = \frac{1}{12N} + \sum_{i=0}^{N-1} \left(\frac{2i + 1}{2N} - F(\phi_i) \right)^2 - N \left(\frac{1}{2} - \frac{1}{N} \sum_{i=0}^{N-1} F(\phi_i) \right)^2,$$ where $\phi_i$ are the observed count polar angles, $\phi_0$ is the origin polar angle on the circle, and $F$ is the uniform CDF. To obtain the final value for $U_N^2$ we minimize the statistic over all origin angles on the circle $$U_N^2 = \min_{\mathrm{origin\;on\;circle},\, \phi_0} U_N^2(\phi_0).$$ The final value for the photon asymmetry, $A_\mathrm{phot}$, is given by the cluster count weighted average of $\hat{d}_{N,C}$ in each radial annulus, namely $$A_\mathrm{phot} = 100 \left. \sum_{k=1}^{N_\mathrm{ann}} C_k \hat{d}_{N_k,C_k} \middle/ \sum_{k=1}^{N_\mathrm{ann}} C_k \right..$$
We assume a uniform background which we estimate from blank-sky images for each observation and subsequently compute the number of cluster counts, $C$, by subtracting the expected number of background counts within the annulus from the total number of observed counts. Following @nurgaliev2013 we compute $\hat{d}_{N,C}$ in four radial annuli, which in this work range between $0.05\,\mathrm{R_{500}}$ and $0.5\,\mathrm{R_{500}}$. This choice of four annuli ensures that we will obtain at least hundreds of cluster counts in each annulus for the low-count observations ($\sim$ a few thousand counts). Optimal annuli are selected by requiring an approximately constant number of cluster counts within each annulus. We define the annuli radii as those which minimize the variance in cluster counts across each of the annuli. In Fig. \[fig:opt\_ann\] we show the optimal annuli positions for each of the 58 clusters in the X-ray matched sample. We note that while there is some variation in optimal annuli from cluster to cluster, in general the scatter is relatively small and there is no overlap between the $1\sigma$ scatter of neighbouring annuli. The final annuli edges are taken to be the median values across all of the clusters, which corresponds to $\{0.05, 0.13, 0.23, 0.34, 0.50\}\times \mathrm{R_{500}}$. The inner boundary of $0.05\,\mathrm{R_{500}}$ is set to aviod pixelation artifacts at small radii [@nurgaliev2013], and the outer boundary of $0.5\,\mathrm{R_{500}}$ is chosen to enclose the majority of the emission while still ensuring chip coverage. The large angular sizes of some of the high-mass, low-redshift systems ($R_{500} \sim 15 - 20 \arcmin$) prevents us from computing $A_\mathrm{phot}$ out to a full $\mathrm{R_{500}}$ since they extend beyond the edge of the detector.
Statistical uncertainties on $A_\mathrm{phot}$ are estimated following @nurgaliev2013 by randomly resampling half of the observed counts 500 times and recalculating $A_\mathrm{phot}$ for each iteration. For clusters with both *Chandra* and *XMM-Newton* observations, we compute $A_\mathrm{phot}$ for the *Chandra* and *XMM* data separately and then combine them as a count-weighted average.
### Centroid shift {#sec:centroid_shift}
A commonly used X-ray relaxation proxy is the centroid shift, $w$, which measures the shift of the X-ray surface brightness centroid in different radial apertures. For a system in dynamical equilibrium, the centre of mass of the ICM (ie. the centroid) should be independent of scale, whereas an unrelaxed system with substructure can have a centre of mass which depends on radius [e.g. @mohr1993]. To compute centroid shifts we use the following relation [e.g. @boehringer2010] $$w = \left[\frac{1}{N - 1}\sum_i (\Delta_i - \langle \Delta \rangle)^2 \right]^{1/2} \times \frac{1}{R_\mathrm{max}},$$ where $\Delta_i$ is the offset between the X-ray peak and the centroid position within the $i$th aperture, $N$ is the number of apertures, and $R_\mathrm{max}$ is the radius of the largest aperture. Centroids are determined from the moments of the exposure-corrected X-ray images[^2], and the X-ray peak is considered to be the position of the brightest pixel after smoothing using a Gaussian kernel with a bandwidth of $40\,\mathrm{kpc}$. The smallest aperture that we consider is $R < 0.1\,\mathrm{R_{500}}$ and we progressively increase the aperture radius by $0.05\,\mathrm{R_{500}}$ out to a maximum of $0.5\,\mathrm{R_{500}}$, for a total of 9 apertures. These aperture choices are motivated by previous studies [@boehringer2010; @nurgaliev2013; @rasia2013; @weissmann2013] as well as ensuring chip coverage as was done to measure $A_\mathrm{phot}$ in Section \[sec:photon\_asym\].
As with $A_\mathrm{phot}$, uncertainties for the centroid shift are determined from randomly resampling the X-ray images and recalculating $w$, and for clusters with observations from *Chandra* and *XMM-Newton* $w$ is computed as a count-weighted average.
Results {#sec:results}
=======
Relationship between X-ray and optical relaxation proxies {#sec:xray_opt_corr}
---------------------------------------------------------
To explore the consistency between group relaxation measures in the X-ray and optical, we measure the correlations between photon asymmetry and centroid shifts and the three optical relaxation parameters ($A^2$, $M_2/M_1$, $\delta R_{MMG}$). To quantify the correlations between these parameters, we use two different methods:
1. We fit a simple power-law to each relationship and derive uncertainties on the slope and normalization with bootstrap resampling.
2. We compute Spearman’s rank correlation coefficient, $r_s$, (which is preferred over the Pearson correlation due to its non-parametric nature) for each relationship to quantify the percentile at which the data are consistent with a correlation.
### Photon Asymmetry {#sec:Aphot_corr}
{width="\textwidth"}
In Figs \[fig:asym\_ad\_m12\_offset\]a-c we show the relationship between photon asymmetry and the three optical relaxation proxies. The $A_\mathrm{phot} - A^2$ relationship shows a significant correlation as measured by both the power-law fit and by the Spearman test. The best-fit power law has a positive slope at $3.5\sigma$ and the Spearman test gives a positive correlation at the $>99.9$ per cent level. The $A_\mathrm{phot} - M_2 / M_1$ relationship shows a weaker (but still significant) correlation with a positive slope at $2.9\sigma$ and a Spearman correlation at the $96$ per cent level. In contrast, the $A_\mathrm{phot} - \delta R_{MMG}$ relationship does not display a significant correlation by either measure, with a power law slope consistent with zero and a Spearman p-value of 0.23.
### Centroid shift {#sec:w_corr}
{width="\textwidth"}
In Figs \[fig:centshift\]a-c we now show the relationship between the optical relaxation parameters and the centroid shift as the X-ray relaxation proxy. The results in Fig. \[fig:centshift\] are very similar to those in Fig. \[fig:asym\_ad\_m12\_offset\], with the optical relaxation proxies tracing the centroid shift analagously to photon asymmetry. The $w - A^2$ relationship has a best-fit positive slope at $2.8\sigma$ and the Spearman test gives a positive correlation at the 99.9 per cent level. The $w - M_2/M_1$ again shows a significant correlation as well with a positive slope at $3.6\sigma$ and a positive Spearman correlation at the 97 per cent level. Finally, we find no evidence for a correlation between the centroid shift and the MMG offset, with a power-law slope consistent with zero and a Spearman p-value of 0.62 .
Based on the results from this section we conclude that Anderson-Darling statistic provides the best correlation with X-ray asymmetry among the three optical relaxation measures, as it shows the strongest Spearman correlations with the X-ray relaxation proxies, and the $A_\mathrm{phot} - A^2$ and $w - A^2$ relationships have positive power-law slopes at $\gtrsim 3\sigma$. Modulo scatter, this correlation lends credence to the use of the Anderson-Darling test to quantify cluster relaxation for a large sample, as the shape of the diffuse X-ray profile is an independent (and arguably more direct) probe of the degree to which groups are unrelaxed/disturbed. Therefore, for the remainder of this paper we will focus on the $A_\mathrm{phot} - A^2$ and $w - A^2$ relationships. In the next section, we extend this analysis by investgating the halo mass dependence of the $A_\mathrm{phot} - A^2$ and $w - A^2$ relations.
Halo mass dependence of X-ray-optical relations {#sec:halo_mass}
-----------------------------------------------
In Section \[sec:xray\_opt\_corr\] we presented a significant correlation between the AD statistic for a given cluster and X-ray relaxation parameters (the photon asymmetry and the centroid shift). In this section we further divide the sample into systems with small halo masses and those with large halo masses to investigate if the $A_\mathrm{phot} - A^2$ and $w - A^2$ correlations vary with cluster halo mass. We choose the median halo mass of our cluster sample, $M_{\mathrm{halo},\,\mathrm{med}} = 10^{14.5}{\,\mathrm{M_{\sun}}}$, to make this division.
### Photon Asymmetry {#sec:mh_aphot}
![68 and 95 per cent confidence ellipses for the photon asymmetry vs. AD statistic best-fit power law parameters for low-mass halos ($M_\mathrm{halo} < 10^{14.5}{\,\mathrm{M_{\sun}}}$, dot-dashed), high-mass halos ($M_\mathrm{halo} > 10^{14.5}{\,\mathrm{M_{\sun}}}$, dashed), and the total sample (solid). Spearman correlation coefficients are denoted for each sample.[]{data-label="fig:asym_mh"}](asymCX_ad_mhMed.pdf){width="0.9\columnwidth"}
In Fig. \[fig:asym\_mh\] we show the confidence ellipses (68 and 95 per cent levels) corresponding to the power-law fit results to the $A_\mathrm{phot} - A^2$ relationship for high-mass and low-mass halos (as well as the total sample). The separation between the high- and low-mass ellipses suggests that high- and low-mass clusters follow somewhat different scaling relations between $A_\mathrm{phot}$ and $A^2$. For high-mass halos a significant correlation is still seen, with a best-fit slope of $1.15_{-0.39}^{+0.42}$ and a Spearman correlation significant at the 99.4 per cent level. For low-mass halos the correlation is somewhat weaker with a best-fit power law slope of $0.61_{-0.33}^{+0.44}$ and a Spearman correlation significant at the 94 per cent level. The best-fit slopes for the low- and high-mass halos are equal within uncertainties, however the normalization is larger for high-mass halos at the $> 2\sigma$ level. Additionally, we find that high-mass halos have a slightly larger median $A_\mathrm{phot}$ ($0.08 \pm 0.03$ ) than low-mass halos ($0.04 \pm 0.01$ ).
### Centroid shift {#sec:mh_cs}
![Same as Fig. \[fig:asym\_mh\] however for the centroid shift instead of photon asymmetry.[]{data-label="fig:cs_mh"}](centshiftCX_ad_mhMed.pdf){width="0.9\columnwidth"}
In Fig. \[fig:cs\_mh\] we now show the power-law fit results to the $w - A^2$ relationship for the two halo mass subsamples. We find qualitatively similar results when considering centroid shift instead of photon asymmetry, with the high-mass halos displaying a clear correlation (slope: $0.86_{-0.24}^{+0.27}$ , Spearman p-value: 0.005 ) whereas the correlation for low-mass halos is marginal and not statistically significant (slope: $0.24_{-0.30}^{+0.35}$, Spearman p-value: 0.19 ). We also find that the median centroid shift is larger for high-mass halos ($0.015 \pm 0.002$ ) than low-mass halos ($0.007 \pm 0.001$).
The discrete case: X-ray asymmetry of Gaussian and non-Gaussian groups {#sec:G_NG}
----------------------------------------------------------------------
Thus far we have treated the shape of the velocity distribution in a continuous fashion with the AD statistic, though it is commonplace in the literature to define a dichotomy between “Gaussian” and “non-Gaussian” clusters [@hou2009; @ribeiro2013a; @roberts2017; @decarvalho2017]. We use the AD test and choose a critical p-value of 0.10 to define G and NG groups – where G groups have $p_{AD} \ge 0.10$ and NG groups have $p_{AD} < 0.10$ (though our results are not sensitive to the precise p-value chosen over a reasonable range).
To quantify the relationship between photon asymmetry and whether a cluster is classified as G or NG we employ the method of logistic regression [e.g. @cox1958]. Logistic regression is a classification tool used to estimate the probability of a binary response as a function of one (or many) independent variables, which may be numeric or categorical. For this application, a galaxy cluster is classified as either G or NG (the boolean, dependent variable) and we are interested in the probability of a galaxy cluster being NG as a function of photon asymmetry or centroid shift (the numeric, independent variable). The estimated probability is then
$$\label{eq:log_curve}
\hat{p} = \frac{e^{\beta_1 x + \beta_0}}{1 + e^{\beta_1 x + \beta_0}},$$
where $\beta_0$ and $\beta_1$ are parameters of the fit, and for this work we have $\hat{p} = \hat{p}(\mathrm{NG})$ and $x = \log A_\mathrm{phot}\;\mathrm{or}\; \log w$.
### Photon asymmetry {#sec:G_NG_aphot}
![Estimated probability of a cluster being non-Gaussian as a function of photon asymmetry. Photon asymmetry data points are shown for Gaussian (0, green) and non-Gaussian (1, purple) clusters and the black line shows the best-fit logistic curve. Shaded vertical lines show the median asymmetry and $1\sigma$ standard error for Gaussian and non-Gaussian groups.[]{data-label="fig:log_reg"}](log_reg.pdf){width="0.9\columnwidth"}
In Fig. \[fig:log\_reg\] we show the photon asymmetry for G (0, green) and NG (1, purple) clusters along with the best-fitting logistic curve (black line, equation \[eq:log\_curve\]) describing the probability of being classified as NG as a function of $A_\mathrm{phot}$. It is clear from Fig. \[fig:log\_reg\] that the probability of a cluster being NG increases with photon asymmetry, we obtain a best-fit coefficient of $\beta_1 = 2.1 \pm 0.8$ indicating a significant correlation at $2.6\sigma$. According to our logistic model, the $A_\mathrm{phot}$ value where the probability of being a NG cluster reaches 50 per cent is $A_\mathrm{phot} = 0.14$ and the asymmetry where the probability reaches 75 per cent is $A_\mathrm{phot} = 0.46$. Additionally in Fig. \[fig:log\_reg\] we show the median photon asymmetry and the $1\sigma$ standard error for G and NG clusters, NG clusters have a larger median asymmetry of $0.13 \pm 0.03$ compared to $0.05 \pm 0.02$ for G clusters.
### Centroid shift {#sec:G_NG_cs}
![Same as Fig. \[fig:log\_reg\] however for the centroid shift instead of photon asymmetry.[]{data-label="fig:log_reg_w"}](log_reg_w.pdf){width="0.9\columnwidth"}
In Fig. \[fig:log\_reg\_w\] we show an analagous logistic regression to Fig. \[fig:log\_reg\], with the centroid shift as the numeric variable. Examining Fig. \[fig:log\_reg\_w\] it is clear that the distinction between G and NG clusters is not as strong as it was with photon asymmetry. From the fit we obtain a best-fit coefficient of $\beta_1 = 1.2 \pm 0.8$, indicating only a marginal correlation at $1.5\sigma$. From this fit the value of $w$ where $\hat{p}(\mathrm{NG})$ reaches 50 per cent is $w = 0.04$. We also show the median value for $w$ for G (green) and NG (purple) clusters, and find that while the median centroid shift is slightly larger for NG clusters, this difference is not significant (G: $0.009 \pm 0.001$, NG: $0.012 \pm 0.003$).
Discussion {#sec:discussion}
==========
The Anderson-Darling test as a relaxation proxy
-----------------------------------------------
The primary result from this paper is the strong correlation detected between X-ray relaxation measures and both the AD statistic for a given cluster (Fig. \[fig:asym\_ad\_m12\_offset\]a and Fig. \[fig:centshift\]a), as well as the probability of a system being NG when considering the dichotomy of G and NG clusters (at least for $A_\mathrm{phot}$, Fig. \[fig:log\_reg\]). We argue that this is an important confirmation of the usefulness of the AD test to quantitatively identify unrelaxed/disturbed systems. This, however, is only true in the statistical sense as there is still significant scatter around the $A_\mathrm{phot} - A^2$ and $w - A^2$ relations. The AD test may or may not accurately classify the dynamical state of an individual cluster, but applied to a large statistical sample it is a useful tool to identify systems which are on average relaxed or unrelaxed. It is also worth considering whether the group finder preferentially selects G or NG clusters. The @yang2005 group finder constructs clusters assuming that the galaxy phase-space distribution follows a spherical NFW profile, which could bias the group finder in favour of G clusters (ie. assuming a spherical, symmetric distribution). The analysis presented here does not account for any such bias, but since the clusters are all selected with the same algorithm the correlations found are robust for this sample.
The AD test has become a relatively common tool used to identify unrelaxed systems from large redshift surveys [e.g. @hou2009; @ribeiro2010; @martinez2012; @hou2013; @ribeiro2013a; @roberts2017], though its efficacy has only been tested in detail using Monte Carlo simulations sampling from idealized parent distributions (both Gaussian and non-Gaussian, @hou2009 [@ribeiro2013a]). These tests have provided useful insight into the strengths and limitations of the AD test, but it is also important to test this technique in a more physical setting. The comparision to diffuse X-ray morphology in this work provides one such test in an astronomical context. In an upcoming paper we perform a detailed analysis on the AD test applied to groups and clusters in large, cosmological, N-body simulations. This will allow us to explore things such as the false-positive rate, as well as potential differences in satellite time-since-infall or halo age for G and NG systems in a cosmological context.
The results of Section \[sec:G\_NG\] can also be used to constrain the dividing line between relaxed and unrelaxed clusters. Based on the logistic regression model, the value of $A_\mathrm{phot}$ above which the probability of being a NG cluster exceeds 50 per cent is $A_\mathrm{phot} = 0.14$ and the value above which the probability exceeds 75 per cent is $A_\mathrm{phot} = 0.46$. Correspondingly, the median $A_\mathrm{phot}$ for NG clusters of $0.13 \pm 0.03$, suggesting that $A_\mathrm{phot} \gtrsim 0.10 - 0.50$ may be a useful dividing line between relaxed and unrelaxed clusters, depending on the desired level of purity. In @mcdonald2017 a threshold of $A_\mathrm{phot} < 0.10$ is chosen to identify relaxed clusters, while this threshold was chosen arbitrarily we’ve shown here that this is a reasonable choice based on cluster velocity distribution measurements. The threshold used to identify unrelaxed clusters in @mcdonald2017 is $A_\mathrm{phot} > 0.50$, which also shows excellent agreement with the dividing lines that we derive from velocity measurements. The choice of $A_\mathrm{phot} > 0.50$ is motivated by simulations of cluster major mergers from @nurgaliev2017 who suggest that $A_\mathrm{phot} \gtrsim 0.2 - 0.6$ is a useful threshold to identify disturbed clusters, again corresponding very closely to the range we determine in this work. This shows that using the AD test (in this case with a p-value of 0.10) to identify relaxed and unrelaxed clusters corresponds very closely to previous results using X-ray techniques.
When using the centroid shift, $w$, instead of photon asymmetry, the logistic regression model does not seperate G and NG clusters as distinctly, however we can still use the model to constrain a dividing line. In particular, the regression model suggests that the probability of being a NG cluster reaches 50 per cent at $w = 0.040$. This is larger by a factor of a few than the boundary between regular and disturbed objects in previous X-ray analyses, which ranges between $w \simeq 0.01$ and $w \simeq 0.02$ [@ohara2006; @cassano2010; @weissmann2013]. Given that the logistic regression only detects a marginal correlation between $\hat{p}(\mathrm{NG})$ and $w$, the dividing line that we derive here is likely not be well constrained.
We can also contrast the two different X-ray relaxation proxies by highlighting any differences in the photon asymmetry and centroid shift relationships with the AD statistic. When considering the continuous case in Section \[sec:xray\_opt\_corr\] we see very similar behaviour in the $A_\mathrm{phot} - A^2$ and $w - A^2$ relationships, consistent with the fact that photon asymmetry and centroid shift have been shown to correlate strongly [@nurgaliev2013]. However the discrete case in Section \[sec:G\_NG\] shows that G and NG clusters are more clearly segregated in terms of photon asymmetry than centroid shift, perhaps suggesting that photon asymmetry is a slightly stronger identifier of dynamically unrelaxed clusters, though a larger sample is required to robustly determine this.
Finally, in Section \[sec:halo\_mass\] we explored the halo mass dependence of the $A_\mathrm{phot} - A^2$ and $w - A^2$ relationships by separating the sample into subsamples of low-mass ($M_\mathrm{halo} < 10^{14.5}{\,\mathrm{M_{\sun}}}$) and high-mass clusters ($M_\mathrm{halo} \ge 10^{14.5}{\,\mathrm{M_{\sun}}}$). We find small differences between the low-mass and high-mass relations, namely, both the $A_\mathrm{phot} - A^2$ and $w - A^2$ relationships for high-mass clusters have larger normalizations, whereas the slopes are consistent between the high- and low-mass samples. In addition, the median values for $A_\mathrm{phot}$ and $w$ are slightly larger for high-mass clusters compared to low-mass clusters. @nurgaliev2013 show that the $A_\mathrm{phot}$ and $w$ statistics are robust against varying numbers of X-ray counts above $\sim 2000$ counts ($A_\mathrm{phot}$ is robust even below 2000 counts). All of the systems in this work have $N_\mathrm{counts} > 2000$, therefore it is unlikely that these differences in asymmetry and centroid shift are being driven by the relatively high-count observations of massive systems. This result hints that low-mass and high-mass clusters may follow slightly different scaling relations when it comes to $A_\mathrm{phot}$ or $w$ versus $A^2$, though a larger sample is necessary to build up the statistics required to conclude this with high confidence. In principle, this difference could be explained through simple hierarchical growth where low-mass halos are on average more virialized than higher-mass clusters at the present day. High-mass clusters will be more recently formed through mergers and accretion which can in turn increase $A_\mathrm{phot}$ and $w$ [e.g. @cassano2010; @nurgaliev2017]. From the optical perspective, we also find that high-mass clusters have velocity distributions which are less Gaussian than low-mass systems, in agreement with previous studies [@roberts2017; @decarvalho2017]. Although it is important to note that it is easier to statistically identify departures from normality for high-mass systems with many members.
Interpreting MMG-based relaxation parameters {#sec:mmg_discussion}
--------------------------------------------
The second optical relaxation proxy that shows a significant correlation with X-ray relaxation proxies is the stellar mass gap between the two most-massive cluster galaxies (see Section \[sec:Aphot\_corr\] and Section \[sec:w\_corr\]). The correlations between X-ray relaxation proxies and $M_2 / M_1$ are found to be weaker than for $A^2$ (especially as measured by the Spearman correlation), potentially suggesting that $M_2 / M_1$ is a poorer (though still useful) tracer of diffuse X-ray morphology. This may be expected given that satellite galaxies (i.e. the velocity distribution) and the diffuse hot gas profile should both trace the larger-scale cluster potential relatively directly, whereas central galaxy growth is governed more by dynamical interactions and gas accretion at the cluster centre. It is also possible that the $A_\mathrm{phot} - M_2 / M_1$ and $w - M_2 / M_1$ trends are being affected by selection effects related to the difficulty indentifying the true MMG (and second most massive galaxy) in these clusters. A particular concern regarding the SDSS is the impact of fibre collisions in the dense inner regions of clusters, as it has been estimated that up to 30 per cent of clusters may be missing a spectra for the true BCG [@vonderlinden2007]. In an attempt to mitigate the effect of fibre collisions we use the systems from sample III in the Yang group catalogue which attaches redshifts to galaxies that lack spectra due to fibre collisions by assigning these galaxies the redshift of the galaxy it “collided” with. While this procedure allows the group finder to include galaxies which otherwise would be missed due to fibre collisions, the trade-off is uncertainty regarding whether the added galaxies are true group members. $\sim 60$ per cent of fibre collision galaxies have redshifts within $500\,\mathrm{km}\,\mathrm{s^{-1}}$ of the estimated value [@zehavi2002], though this still leaves a significant number of fibre collision galaxies which may have true redshifts that differ substantially from the assigned value. To ensure that our results are not being affected by the inclusion of these fibre collision galaxies we re-test the $A_\mathrm{phot} - M_2 / M_1$, $w - M_2 / M_1$ and $A_\mathrm{phot} - \delta R_{MMG}$, $w - \delta R_{MMG}$ relationships for correlations, now removing any systems where the MMG (and in the case of $M_2 / M_1$, the second-most-massive galaxy as well) is a fibre collision galaxy. 26 per cent of the clusters in the sample have an MMG which is a fibre collision galaxy and 39 per cent of the sample have either the MMG or the second-most-massive galaxy as a fibre collision galaxy. Re-testing these relationships for correlations leaves the Spearman correlation coefficient virtually unchanged from Sections \[sec:Aphot\_corr\] and \[sec:w\_corr\], suggesting that fibre collision galaxies are not biasing the results.
In Section \[sec:Aphot\_corr\] we find no evidence for a correlation between $A_\mathrm{phot}$ or $w$ and $\delta R_{MMG}$, which suggests that the MMG offset is not a reliable tracer of cluster relaxation. One caveat which is important to consider is the assumptions made to justify the use of $M_2 / M_1$ and $\delta R_{MMG}$ as relaxation proxies, in particular that for relaxed systems the MMG (or brightest galaxy) resides at rest at the centre of the dark matter potential – the so-called central galaxy paradigm (CGP). For example, if the MMG is instead a satellite galaxy then the use of $M_2 / M_1$ as a relaxation measure may not be valid as it is predicated on the MMG being the central and growing through accretion and mergers at the centre of the potential well. Similarly, if the MMG is a satellite then its offset from the luminosity-weighted centre would not be expected to trace cluster relaxation. Many recent studies have called into question the ubiquity of the CGP by highlighting the fact that a substantial fraction of brightest cluster galaxies (BCGs) are significantly offset from the cluster centroid, both in terms of projected distance and velocity [@vandenbosch2005; @coziol2009; @skibba2011; @sehgal2013; @lauer2014; @hoshino2015]. In particular, @skibba2011 find that the fraction of halos where the brightest galaxy is in fact a satellite ($f_\mathrm{BNC}$) ranges from $\sim 25$ per cent in low-mass halos ($10^{12}\,h^{-1} \le M \lesssim 2 \times 10^{13}\,h^{-1}{\,\mathrm{M_{\sun}}}$) to $\sim 40$ per cent in high-mass halos ($M \gtrsim 5 \times 10^{13}\,h^{-1}{\,\mathrm{M_{\sun}}}$). Furthermore, @hoshino2015 find $f_\mathrm{BNC} \sim 20 - 30$ per cent for galaxies in redMaPPer clusters, and in terms of velocity @coziol2009 show that the median peculiar velocity for BCGs in a sample of Abell clusters is $\sim$ one third of the cluster velocity dispersion. It is plausible that systems where the CGP is not valid are diluting stronger trends between $A_\mathrm{phot}$ or $w$ and $M_2/M_1$, or perhaps masking trends between $A_\mathrm{phot}$ or $w$ and $\delta R_{MMG}$. Unfortunately, identifying systems where the CGP is violated is difficult on a case-by-case basis, limited by observing in projection, and is generally done in the statistical sense for large samples (ie. thousands) of groups and clusters [e.g. @vandenbosch2005; @skibba2011]. Therefore we continue to argue that the AD test (or some other measure of the velocity distribution shape, see e.g. @ribeiro2013a) is a better optical relaxation proxy as it is not complicated by CGP assumptions.
Summary {#sec:summary}
=======
In this paper we present a comparison between diffuse X-ray morphology and cluster relaxation proxies based on optical measures. Using the @yang2007 SDSS group catalogue we match optically identified clusters with $N \ge 10$ members to X-ray observations from both the *Chandra* and *XMM-Newton* X-ray observatories. With a sample of 58 X-ray matched clusters we compare X-ray asymmetry and centroid shift to three different optical relaxation probes: the Anderson-Darling statistic, the stellar mass gap, and the MMG offset. The main conclusions of this work are as follows:
1. We detect a significant positive correlation between X-ray relaxation proxies (photon asymmetry, centroid shift) and Anderson-Darling statistic at $\sim 3-4\sigma$ as measured by both a power-law fit and by the Spearman correlation test, and a weaker correlation ($\sim 2-3 \sigma$) between X-ray relaxation proxies and stellar mass gap (between two most-massive cluster galaxies).
2. We do not detect a significant correlation between X-ray asymmetry or centroid shift and the MMG offset.
3. We find that the $A_\mathrm{phot} - A^2$ and $w - A^2$ relationships vary somewhat for low-mass ($M_\mathrm{halo} < 10^{14.5}{\,\mathrm{M_{\sun}}}$) and high-mass ($M_\mathrm{halo} \ge 10^{14.5}{\,\mathrm{M_{\sun}}}$) clusters. Specifically, high-mass clusters have a best-fit relationship with a larger normalization, and the median asymmetry and centroid shift is larger in high-mass systems. However, a definitive measure of the halo mass dependence awaits a larger sample.
4. When considering a dichotomy between Gaussian ($p_{AD} \ge 0.10$) and non-Gaussian ($p_{AD} < 0.10$) clusters we find that the probability of being a non-Gaussian system (as measured by a logistic regression) correlates clearly with X-ray asymmetry. Additionally, the median asymmetry of non-Gaussian clusters is larger than that of Gaussian clusters. When using the centroid shift as the X-ray relaxation proxy the correlation is marginal.
Though the scatter in the above mentioned relations limit the reliability of this approach on a case-by-case basis, these results confirm the effectiveness of the shape of the projected velocity distribution as a proxy for cluster relaxation, when applied to a large sample.
Acknowledgments {#sec:acknowledgments .unnumbered}
===============
We thank the Natural Sciences and Engineering Research Council of Canada for funding. This work made use of many open-source software packages, such as: <span style="font-variant:small-caps;">AstroPy</span> [@astropy2013], <span style="font-variant:small-caps;">Matplotlib</span> [@hunter2007], <span style="font-variant:small-caps;">NumPy</span> [@vanderwalt2011], <span style="font-variant:small-caps;">Pandas</span> [@mckinney2010], <span style="font-variant:small-caps;">Photutils</span> [@bradley2016], <span style="font-variant:small-caps;">SciPy</span> [@jones2001], <span style="font-variant:small-caps;">Statsmodels</span> [@seabold2010], and <span style="font-variant:small-caps;">Topcat</span> [@taylor2005].
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 Observatories 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.
X-ray images
============
{width="\textwidth"}
{width="\textwidth"}
\[lastpage\]
[^1]: E-mail: [email protected]
[^2]: http://photutils.readthedocs.io/en/stable/photutils/centroids.html
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We present an update on the MILC Collaboration’s light hadron spectrum calculation with two flavors of dynamical, staggered quarks. Results are presented for gauge couplings 5.30, 5.415, 5.50 and 5.60, with a range of quark masses for each value of the coupling. We present extrapolations of $m_N/m_\rho$ to the continuum limit for fixed values of $m_\pi/m_\rho$ including the physical one.'
address:
- '[-0.10in[0.07in Department of Physics, Washington University, St. Louis, MO 63130, USA]{}]{}'
- 'Department of Physics, Brookhaven National Lab, Upton, NY 11973, USA'
- 'Physics Department, University of Colorado, Boulder, CO 80309, USA'
- 'Physics Department, University of Utah, Salt Lake City, UT 84112, USA'
- 'Department of Physics, Indiana University, Bloomington, IN 47405, USA'
- 'SCRI, Florida State University, Tallahassee, FL 32306-4130, USA'
- 'Universität Bielefeld, Fakultät für Physik, Postfach 100131, D-33501 Bielefeld, Germany'
- 'Department of Physics, University of California, Santa Barbara, CA 93106, USA'
- 'Department of Physics, University of Arizona, Tucson, AZ 85721, USA'
author:
- 'Claude Bernard, Tom Blum, Thomas A. DeGrand, Carleton DeTar, Steven Gottlieb, Urs M. Heller, Jim Hetrick, $\,\null^{\rm a}$ [^1] Craig McNeile, $\,\null^{\rm c}$ Kari Rummukainen, Robert Sugar [^2] and Doug Toussaint'
title: Update on the hadron spectrum with two flavors of staggered quarks
---
epsf.sty
\#1 \#2 \#3 \#4 \#5
\#1 \#2 \#3 \#4 \#5
\#1 \#2 \#3 \#4 \#5 \#6
Over the past few years the MILC Collaboration has been engaged in a series of spectrum calculations with dynamical quarks and in the quenched approximation, using both standard and improved actions[@EARLIER]. One objective of this work is to obtain sufficient control over systematic errors so that one can make reliable extrapolations to the limits of zero lattice spacing and physical quark masses. Another objective is to make detailed comparisons between the spectra with quenched and dynamical quarks, and with standard and improved actions. Clearly, to reach these objectives requires high statistics calculations for a sufficiently wide range of lattice spacings and quark masses to enable us to make extrapolations to the continuum and chiral limits. It also requires that these calculations be carried out on large enough lattices to avoid finite size effects.
In this note we provide an update on our results for two flavors of dynamical, staggered quarks using the standard gauge and quark actions. Results for the quenched approximation and with improved actions are presented elsewhere in these proceedings[@STEVE]. We have carried out simulations at four values of the gauge coupling, $6/g^2=5.30$, 5.415, 5.50 and 5.60 using at least four values of the quark mass at each coupling. Lattices were generated with the refreshed hybrid molecular dynamics algorithm. Once the lattices were equilibrated the light hadron spectrum was measured every fifth molecular dynamics time unit. Lattices were saved after every tenth time unit for use in other projects. With the exeption of a few runs at large quark mass, measurements were made on at least four hundred lattices at each values of the gauge couping and quark mass.
As one would expect, the hadron masses evaluated in lattice units are strongly dependent on the gauge coupling and bare lattice quark mass, $am_q$. Here $a$ is the lattice spacing. The one exception is the mass of the Goldstone pion which is weakly dependent on the gauge coupling. In Fig. 1, we plot the $\rho$ mass as a function of the bare quark mass for the two weakest values of the gauge coupling that we have studied. For comparison we also include in this figure results from quenched calculations with staggered quarks at gauge couplings 5.70, 5.85 and 6.15. Notice that for fixed values of the gauge coupling, $am_\rho$ decreases more rapidly for decreasing $am_q$ in full QCD than in the quenched approximation. This is due to the dependence of the lattice spacing on the quark mass in full QCD.
3.0in -0.15in [FIG1]{} [fig1.ps]{} [The $\rho$ mass as a function of the quark mass for full and quenched QCD.]{}
3.0in -0.15in [FIG2]{} [fig2.ps]{} [$\delta_\pi$ as a function of $am_q$ for full QCD.]{}
An important question regarding any simulation with staggered quarks is the extent to which flavor symmetry is restored as the lattice spacing and quark mass are decreased. One measure of flavor symmetry violation is the quantity $\delta_\pi =(m_{\pi_2}^2 - m_{\pi}^2) / m_{\rho}^2$. In Fig. 2, we plot $\delta_\pi$ as a function of $am_q$ for each of the couplings we have studied. The trends are as expected. The value of $\delta_\pi$ for $6/g^2=5.60$ is close to, but slightly below that found in the quenched approximation for gauge coupling 5.85.
$$\vbox{
\null\vskip-0.2in\hskip-0.30in \epsfxsize=3.0in \epsfbox[0 0 4096 4096]{fig3.ps}
\vskip -0.3in
\vskip 0.0truein plus0.2truein}$$ 0.1in [Figure 3. The Edinburgh plot for full and quenched QCD.]{} 0.2in
In Fig. 3, we show the Edinburgh plot for our two weakest gauge couplings using the Goldstone pion in the ratio $m_\pi/m_\rho$. Once again we include quenched results for gauge couplings 5.70, 5.85 and 6.15 in this graph. Here too, the full QCD results at 5.60 are quite close to the quenched ones at 5.85.
Because we have carried out calculations with a number of quark masses at each value of the coupling we have studied, we can perform fits to the hadron masses as a function of $am_q$ in order to extrapolate to the chiral limit and to interpolate for comparisons with quenched and improved action calculations. We are analyzing a variety of fitting functions. Here we present results from fits of the nucleon and rho masses to the form $$m=m_0+\alpha m_q+\beta m_q^{3/2}+\gamma m_q^2.$$ In Fig. 4 we show $m_N/m_\rho$ for the four values of the gauge coupling we have studied, in each case interpolating to the value of $am_q$ for which $m_\pi/m_\rho=0.5$. The $x$-axis in this figure is $(am_\rho)^2$ interpolated to the same value of the quark mass. This quantity gives a measure of the square of the lattice spacing.
3.0in -0.15in [FIG4]{} [fig4.ps]{} [$m_N/m_\rho$ as a function of the square of the lattice spacing for $m_\pi/m_\rho=0.5$]{}
Fig. 5 is a repeat of Fig. 4, this time with extrapolations to the value of the quark mass for which $m_\pi/m_\rho$ takes on its experimental value, 0.1753. The errors for the individual points in Figs. 4 and 5 were determined by a jackknife analysis. The solid lines in these figures are fits to the four interpolated or extrapolated points of the form $\alpha + \beta (am_\rho)^2$, where $\alpha$ and $\beta$ are fit parameters. The confidence levels of these fits are shown in the figures. The fits allow an extrapolation of $m_N/m_\rho$ to the continuum limit for fixed values of $m_\pi/m_\rho$, and they provide support of the expectation that for staggered quarks the leading corrections are of order $a^2$. These results are quite encouraging; however, a better understanding of the chiral extrapolation, as well as additional calculations at weaker coupling and smaller quark masses are needed in order to obtain definitive results.
This work was supported by the Department of Energy and the National Science Foundation. Computations were carried out at ORNL, PSC, NCSA, IU, SDSC, CTC and MHPCC.
3.0in -0.15in [FIG5]{} [fig5.ps]{} [$m_N/m_\rho$ as a function of the square of the lattice spacing with $m_\pi/m_\rho$ fixed at its experimental value.]{}
[9]{} The MILC Collaboration, C. Bernard [*et al.*]{}, Nucl. Phys. B (Proc. Suppl.) [**47**]{} (1996) 345; Nucl. Phys. B (Proc. Suppl.) [**53**]{} (1997) 212; Phys. Rev. [**D55**]{} (1997) 1133. The MILC Collaboration, C. Bernard [*et al.*]{}, these proceedings.
[^1]: Current address, Department of Physics, University of the Pacific, Stockton, CA 95211, USA
[^2]: presented by Robert Sugar
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- 'Moein Khajehnejad$^{1}$$^{*}$'
- 'Ahmad Asgharian Rezaei$^{2}$$^{*}$'
- 'Mahmoudreza Babaei$^{3}$'
- |
\
Jessica Hoffmann$^{4}$
- |
Mahdi Jalili$^{2}$Adrian Weller$^{5,6}$\
$^{1}$Monash University\
$^{2}$RMIT University\
$^{3}$Max Planck Institute for Software Systems\
$^{4}$The University of Texas at Austin\
$^{5}$University of Cambridge\
$^{6}$Alan Turing Institute\
[email protected], [email protected], [email protected], [email protected], [email protected], [email protected]
title: Adversarial Graph Embeddings for Fair Influence Maximization over Social Networks
---
[^1]
Acknowledgements {#acknowledgements .unnumbered}
================
AW acknowledges support from the David MacKay Newton Research Fellowship at Darwin College, The Alan Turing Institute under EPSRC grant EP/N510129/1 & TU/B/000074, and the Leverhulme Trust via the CFI.
[ ]{}
[^1]: $^{*}$Authors contributed equally to this work.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'For the superconducting linear collider TESLA a multi purpose detector has been designed. This detector is optimised for the important physics processes expected at a next generation linear collider up to around 1 TeV and is designed for the specific environment of a superconducting collider.'
author:
- Klaus Mönig
date: October 15th 2001
title: The TESLA Detector
---
Snowmass E3029
Talk presented at the 2001 Snowmass Workshop on the Future of Particle Physics July 1-20, 2001
Introduction {#sec:intro}
============
Recently the Technical Design Report for the superconducting linear collider TESLA has been completed [@tdr_tesla]. As a proof that the proposed physics program [@tdr_phys] can be done with a detector of known technology and reasonable price the conceptual design of a multipurpose detector has been included [@tdr_det].
The physics goals of TESLA require from the detector
- very good momentum resolution ($\delta 1/p \sim 4\cdot 10^{-5}/\GeV$) e.g. to measure the Z recoil mass in the process ${\mathrm{e}^+\mathrm{e}^-}\rightarrow {\rm ZH},
{\rm Z}\rightarrow \ell^+ \ell^-$,
- high resolution of the hadronic jet energy ($\Delta E/E \approx 30\%/\sqrt{E}$) to reconstruct multi-jet events with intermediate resonances such as ZHH or ${{\mathrm t\bar{\mathrm t}}}$H,
- superb b-tagging to identify multi-b final states like ZHH or ${{\mathrm t\bar{\mathrm t}}}$H or to separate ${\rm H} \rightarrow {{\mathrm b\bar{\mathrm b}}}, \, {\rm H} \rightarrow {{\rm c\bar{\rm c}}}$ and ${\rm H} \rightarrow gg$,
- optimal hermeticity to reduce backgrounds in missing energy channels, especially to veto two-photon induced events.
At TESLA a bunch train consists of around 2800 bunches with a bunch spacing of more than 300 ns. The relatively long time between bunches makes bunch identification easy and no special fast detectors are needed for this purpose. The only really relevant background at TESLA are ${\mathrm{e}^+\mathrm{e}^-}$ pairs created in the collision. These pairs are concentrated at low angle or low transverse momentum. The low angle component requires a mask in the forward direction, which can, however, to a large part be used for calorimetry. The low $p_t$ component at large angles can be kept at low radius by a strong magnetic field. It limits the radius of the most inner detector layer to around 1.5cm and causes a significant background in that layer of the vertex detector.
Figure \[fig:fuldet\] a) shows the principle layout of the detector. The tracking system and the calorimeters are situated inside a 4T superconducting coil. Sections \[sec:track\] and \[sec:calo\] describe the two main subsystems, tracking and calorimetry. In section \[sec:perf\] the expected performance of the detector is summarised. Needed and planned R&D to build the detector presented here is described in other talks of this session.
![a) Layout of the TESLA detector b) The tracking system of the TESLA detector[]{data-label="fig:fuldet"}](e3_moenig_029_fig1a.eps "fig:"){width="0.45\linewidth"} ![a) Layout of the TESLA detector b) The tracking system of the TESLA detector[]{data-label="fig:fuldet"}](e3_moenig_029_fig1b.eps "fig:"){width="0.45\linewidth"}
The Tracking system {#sec:track}
===================
The layout of the tracking system is shown in figure \[fig:fuldet\] b). It consists of a a precise microvertex detector, a large TPC, silicon tracking in between these two main detectors and a set of forward chambers behind the TPC endplate.
The TPC is in principle similar to existing ones, such as ALEPH, DELPHI or STAR. Due to the higher photon background higher granularity and 100–200 pad rows are needed. For this reason an R&D program has been started to read out the signals with new technologies like GEMs or Micromegas. Not to compromise the momentum resolution and the calorimetric energy reconstruction also the field cage and the endplate will be designed to be as thin as possible.
For the microvertex detector (VTX) several technologies are under study, CCDs, active pixel sensors and CMOS sensors. With the CCDs a point resolution of $3.5 {\mu\mathrm{m}}$ and a layer thickness of $0.12 \%$ of a radiation length can be reached. The main problem of CCD detectors for TESLA is the relatively long readout time. Reading out the columns in parallel with a frequency of 50MHz results in one complete readout every 100 bunch crossings giving a relatively high but acceptable background in the innermost layer. For the other technologies an R&D program is in progress to reach a similar performance as presented for the CCDs.
The intermediate tracking detector consists of two cylinders of silicon strip detectors in the barrel region down to $25^\circ$ (SIT) and three silicon pixel and four silicon strip layers on either side in the forward region (FTD) of the detector. The SIT and FTD use only technologies that have already been used successfully in LEP so that important R&D is not needed beyond finding ways to thin the detectors. It improves significantly the momentum resolution in the full acceptance region and enables a precise angle measurement in the forward region before a lot of material is crossed, which is needed to measure the acolinearity of Bhabha events in the analysis of the luminosity spectra.
To get a reasonably precise momentum measurement below a polar angle of around $12^\circ$, where the projected tracklength in the TPC starts to limit the accuracy, at least one precise space point with a resolution around $50 {\mu\mathrm{m}}$ at maximal distance from the interaction point is needed. This can for example be done with a forward chamber (FCH) consisting of 12 straw tube planes with $100 {\mu\mathrm{m}}$ resolution. The high redundancy helps solving ambiguities in the pattern recognition. The chamber is also extended over the full TPC endplate to help mapping distortions in the TPC and possibly to serve as a preshower device.
(18,12)(0,0) (12,10)[(0,-1)[1.3]{}]{} (11.3,10.1)[LAT]{} (15,11)[(0,-1)[1.5]{}]{} (13.5,11.1)[tungsten]{} (12,2)[(3,2)[4]{}]{} (10.5,1.4)[graphite]{} (14.5,1)[(3,4)[3]{}]{} (13.5,0.4)[LCAL]{}
Calorimetry {#sec:calo}
===========
To measure jet energies the so called “energy flow” concept is proposed. In this scheme the energy is measured as the sum of the charged particle energies, which contribute about 60% to a typical hadronic jet, the neutral electromagnetic energy measured in the electromagnetic calorimeter, around 30% of a jet, and the energy of neutral hadrons measured in the hadronic calorimeter, representing only the remaining 10% of the energy. In principle this method allows a very good jet energy resolution, since the tracking resolution is below a few percent, while the resolution of a typical hadron calorimeter is around $100\%/\sqrt{E}$. However, in this concept one needs a very good spatial resolution of the calorimeters to separate showers from charged and neutral particles.
The setup of the proposed calorimeters can been seen in figure \[fig:fuldet\] . In the barrel and endcap the electromagnetic (ECAL) and hadronic (HCAL) calorimeters, which are inside the coil are followed by the instrumented iron return yoke which is used as a tail catcher and for muon identification. In the very forward region the mask is instrumented for electromagnetic calorimetry.
For the ECAL two options are under study, a SiW calorimeter and a lead scintillator sandwich with shashlik readout. The SiW option offers several advantages. Due to the high density of tungsten one can pack 25 radiation length into 20cm thickness. The small Moliere radius of tungsten ($\sim 1 \, {\mathrm{cm}}$) together with a $ 1 \, {\mathrm{cm}}^2$ Si-pad size offers a superb lateral resolution and with around 40 layers also a very good longitudinal resolution is possible. However the detector is fairly expensive ($\sim 130$[[Meuro]{}]{}). For the shashlik option a readout granularity of $3 \times 3 \, {\mathrm{cm}}^2$ has been studied. Longitudinally a granularity of two is possible using scintillators of different decay times. The shashlik calorimeter is about a factor six cheaper than for SiW, but with worse performance which still has to be quantified.
For the hadronic calorimeter again two options have been studied. One option is a scintillating tile calorimeter with $5 \times 5\, {\mathrm{cm}}^2$ tiles in the front part and $25 \times 25 \, {\mathrm{cm}}^2$ tiles in the back part. The granularity is limited by the possibility to get the fibres out of the detector. As an alternative a so called “digital calorimeter” has been proposed. If the granularity of the calorimeter is high enough the energy resolution using only the number of hit cells is better than by using the total amplitude. Following this observation as a second possibility a “digital calorimeter” has been proposed, consisting of $1 \times 1 \, {\mathrm{cm}}^2$ Geiger counters which are read out binary. The very high granularity is well matched to the SiW-ECAL allowing an almost perfect separation of showers from charged and neutral particles.
The detailed setup of the mask is shown in figure \[fig:mask\]. The two parts labelled “LAT” and “LCAL” are equipped as calorimeters. The LAT is reasonably clean so that it can be used at least to veto two-photon events. The LCAL receives much background from ${\mathrm{e}^+\mathrm{e}^-}$-pairs from beamstrahlung. Probably it is only usable for a fast luminosity measurement for machine tuning.
Detector performance {#sec:perf}
====================
Figure \[fig:mres\] shows the momentum resolution for $250 \GeV$ muons as a function of the polar angle and for particles at $\theta = 90^\circ$ as a function of the momentum. Over the full tracking region a unique charge identification is possible up to highest momenta and in the central region $\Delta \frac{1}{p} \approx 4 \cdot 10^{-5} /\GeV$ is reached. With this resolution the Z- and Z-recoil mass resolution in ${\mathrm{e}^+\mathrm{e}^-}\rightarrow {\rm ZH},\, {\rm Z} \rightarrow {\mu^+\mu^-}$ is dominated by the Z-width and the beam energy spread.
![Momentum resolution of the tracking system, a) for 250GeV muons as a function and b) as a function of the momentum for $\theta = 90^\circ$.[]{data-label="fig:mres"}](e3_moenig_029_fig3a.eps "fig:"){width="0.4\linewidth"} ![Momentum resolution of the tracking system, a) for 250GeV muons as a function and b) as a function of the momentum for $\theta = 90^\circ$.[]{data-label="fig:mres"}](e3_moenig_029_fig3b.eps "fig:"){width="0.4\linewidth"}
With the CCD vertex detector the impact parameter resolution is given by $\sigma = 2.9 \oplus 3.9/(p \sin^{3/2} \theta) {\mu\mathrm{m}}$. For b- and c-tagging the SLD package ZVTOP [@ref:ZVTOP] has been used. As detailed in [@ref:stefania] this allows in Z decays at rest to tag b-quarks with 70% efficiency at 95% purity and to tag c-quark with 40% efficiency at 85% purity.
An energy flow resolution of $\Delta E/\sqrt{E} = 30\%$ had been reached with the SiW-ECAL and the digital HCAL [@ref:eflow]. As can be seen from figure \[fig:eflow\] this resolution is needed for example to measure the Higgs self coupling from ${\mathrm{e}^+\mathrm{e}^-}\rightarrow {\rm ZHH}$. A comparable analysis for the shashlik ECAL does not exist yet.
![Mass-distance variable for ${\mathrm{e}^+\mathrm{e}^-}\rightarrow {\rm ZHH}$ and background assuming a): $\Delta E/E = 60\%(1+|\cos\theta_{\rm{jet}}|)/\sqrt{E}$ or b): $\Delta E/E = 30\%/\sqrt{E}$.[]{data-label="fig:eflow"}](e3_moenig_029_fig4.eps){width="0.7\linewidth"}
conclusions {#sec:conc}
===========
For the TESLA collider a detector has been designed with which the proposed physics from $\sqrt{s}=91\GeV$ to $\sqrt{s}=800\GeV$ can be well measured. The degradation of the physics signals due to the finite resolution is minimal. The necessary R&D for the proposed detector is explained in other contributions to this session. The total cost of the detector has been estimated to be between 160[[Meuro]{}]{} and 280[[Meuro]{}]{} depending on the calorimeter choice.
I would like to thank the members of the TESLA detector group for their work on the detector concept, the pleasant atmosphere in the group and for their help preparing the talk and the manuscript.
[1]{}
F. Richard (ed.), J. R. Schneider (ed.), D. Trines (ed.), and A. Wagner (ed.). . DESY-01-011A.
J. A. Aguilar-Saavedra et al. . DESY-01-011C.
G. Alexander et al. . DESY-01-011D.
D. Jackson. , A388:247, 1997.
S. Xella Hansen, these proceedings.
J.C Brient, H. Videau, these proceedings.
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- 'Alberto Pérez-Cervera$^{1,2}$, Tere M-Seara$^{1}$ and Gemma Huguet$^{1}$\'
bibliography:
- 'bibIsos.bib'
title: 'Global phase-amplitude description of oscillatory dynamics via the parameterization method'
---
**Corresponding author:** Alberto Pérez-Cervera, `[email protected]`\
**Keywords:** Phase-Amplitude variables, non-linear oscillators, Phase and Amplitude Response Functions, isochrons, isostables, parameterization method.\
**MSC2000 codes:** 37D10, 92B25, 65P99, 37N25\
Abstract {#abstract .unnumbered}
========
In this paper we use the parameterization method to provide a complete description of the dynamics of an $n$-dimensional oscillator beyond the classical phase reduction. The parameterization method allows, via efficient algorithms, to obtain a parameterization of the attracting invariant manifold of the limit cycle in terms of the phase-amplitude variables. The method has several advantages. It provides analytically a Fourier-Taylor expansion of the parameterization up to any order, as well as a simplification of the dynamics that allows for a numerical globalization of the manifolds. Thus, one can obtain the local and global isochrons and isostables, including the slow attracting manifold, up to high accuracy, which offer a geometrical portrait of the oscillatory dynamics. Furthermore, it provides straightforwardly the *infinitesimal Phase and Amplitude Response Functions*, that is, the extended infinitesimal Phase and Amplitude Response Curves, which monitor the phase and amplitude shifts beyond the asymptotic state. Thus, the methodology presented yields an accurate description of the phase dynamics for perturbations not restricted to the limit cycle but to its attracting invariant manifold. Finally, we explore some strategies to reduce the dimension of the dynamics, including the reduction of the dynamics to the slow stable submanifold. We illustrate our methods by applying them to different three dimensional single neuron and neural population models in neuroscience.
Introduction
============
Oscillations are ubiquitous in a wide range of physical, chemical and biological processes [@kuramoto2003chemical; @winfree2001geometry; @strogatzbook; @PIK01]. Of particular interest for this paper, are the neural oscillators ranging from single neuron [@izhikevich2007] to neural populations [@buzsaki2006rhythms]. From the mathematical perspective, oscillations correspond to attracting limit cycles in the phase space whose dynamics can be described by a single variable: the phase. The description of the dynamics by means of the phase variable assuming that trajectories remain close to the limit cycle, is known as the phase reduction. This reduction has been extensively used to study weakly coupled oscillators because it simplifies the dynamics of a high-dimensional oscillator to a simple one-dimensional dynamical system [@hoppensteadt2012; @ErmentroutTerman2010; @ErmentroutKopell91; @winfree1967biological].
The phase variable can be extended out of the limit cycle under generic conditions via the concept of asymptotic phase and isochrons [@winfree1974patterns]. When the limit cycle is hyperbolic, it is known that each point on its basin of attraction approaches asymptotically the orbit of a point on the limit cycle [@hirsch1970stable; @guckenheimer1975]. Therefore, we can assign to any point an asymptotic phase corresponding to the phase of the point on the limit cycle whose orbit approaches to. The isochrons are thus defined as the set of points having the same asymptotic phase [@winfree2001geometry]. Brief external perturbations displace trajectories from one isochron to another, thus causing a phase shift. A useful tool characterizing the response of a given oscillator to brief perturbations is the phase response curve (PRC) [@ErmentroutTerman2010; @schultheiss2011phase; @brown2004phase]. The PRC quantifies the phase change of an oscillator due to perturbations acting at different phases of the cycle. If the strength of the perturbation is weak, the PRC can be approximated by its first order term, the infinitesimal PRC (iPRC). Nevertheless, the validity of the phase reduction and the iPRC is limited by different factors such as the time between stimuli, the contraction rate to the limit cycle, or the amplitude of the perturbation.
Due to the widespread oscillatory dynamics, there are many cases in which the limitations of the phase reduction become relevant. Examples can be found in the field of circadian rhythms [@WilsonE19], in control theory [@monga2019optimal] or in the study of the role of brain rhythms in cognitive functions [@buzsaki2006rhythms; @schultheiss2011phase; @canavier2015phase; @tiesinga2010; @PerezSH20]. These examples, amongst many others, motivate the growing mathematical research trying to overcome the limitations inherent to the phase reduction. Amongst the different approaches taken we find the extension of the phase variable to the basin of attraction of the limit cycle through the computation of the global isochrons [@osinga2010continuation; @Detrixheetal16], the computation of PRCs beyond the linear approximation given by the iPRC [@suvak2010quadratic; @takeshita2010higher; @PerezCervera2019; @OprisanPC04], or generalizations of the phase reduction approach for perturbations having specific frequency features [@park2016weakly; @pyragas2015phase].
An alternative strategy, which has gained a lot of interest in the recent years, consists in introducing extra variables which describe the dynamics along the directions transversal to the limit cycle. Several groups have proposed different ways to define the transverse coordinate system [@wedgwood2013phase; @shirasaka2017phase; @bonnin17; @mauroy2018global; @moehliswilsonpre2016; @wilson2018greater; @wilson2020]. Of particular interest for this paper are the works [@guillamon2009computational; @castejon2013phase; @perezrole], where the transverse coordinates, called amplitude variables, parameterize the isochrons. The addition of these variables allows to describe the effects of the perturbations acting on trajectories away from the limit cycle. These effects, the phase changes, are recorded in the so-called Phase Response Functions (PRFs), which extend the PRCs to the basin of attraction of the limit cycle [@guillamon2009computational; @castejon2013phase]. Complementing the isochrons, one can define the level sets of the amplitude variables, called isostables [@mauroy2013isostables] or A-curves in [@castejon2013phase]. They correspond to points on the basin of attraction which share the same asymptotic convergence in a well-defined sense. Analogously to the PRFs one can define the Amplitude Response Functions (ARFs) (or Isostable Response Functions in [@mauroy2018global]) quantifying the change in the amplitude variable due to a perturbation [@castejon2013phase].
In this context, the parameterization method introduced in [@cabre2003parameterization; @cabre2003parameterization2; @cabre2005parameterization] becomes a powerful tool to study the dynamics around a hyperbolic attracting limit cycle using the phase-amplitude variables (see [@haro2016] for a review of the method). This method provides a systematic way to compute the (un)stable manifolds around invariant objects in terms of some variables (the coordinates of the parameterization) with a prescribed dynamics that is chosen as simple as possible. The method provides analytically a Fourier-Taylor expansion of the parameterization, as well as a simplification of the dynamics, that allows for a numerical globalization of the manifolds. We want to emphasize that in the case of an attracting limit cycle its stable manifold corresponds to its basin of attraction. Therefore, the parameterization method can be seen as a change of coordinates to the phase-amplitude variables. It has been successfully applied in [@guillamon2009computational; @huguet2013computation; @castejon2013phase] to compute local and global isochrons, isostables (A-curves), iPRF and iARFs for 2-dimensional oscillators. Furthermore, it has also been applied to compute invariant curves and PRCs in different problems of neuroscience [@PerezCervera2019; @PerezCerveraHS17; @castejon2020phase].
In this paper, we extend the applications of the parameterization method in [@guillamon2009computational; @castejon2013phase; @huguet2013computation] to provide efficient algorithms to compute the full set of phase-amplitude coordinates, as well as local and global isochrons, isostables, iPRF and iARFs for $d$-dimensional oscillators, $d \geq 2$. Although all the methods and algorithms presented herein are completely analogous in higher dimension, for clarity of exposition we present them in dimension 3. The efficiency of the algorithms relies mainly on the use of Floquet theory to solve periodic linear high-dimensional systems [@castelli2015parameterization] and the use of automatic differentiation techniques [@griewank2008evaluating; @haro2016] to compute the compositions of power series with elementary functions, thus, avoiding the computation of high-order derivatives of the vector field. We apply these algorithms and compute an approximate parameterization in Fourier-Taylor series up to any degree, which provides analytically the local isochrons, isostables, iPRF and iARFs in a neighbourhood of the limit cycle. We stress that, in addition to these objects, the parameterization naturally provides the slow submanifold, which corresponds to the manifold associated to the slowest Floquet multiplier [@cabre2003parameterization]. Furthermore, we use a numerical strategy based on backwards integration to globalize the slow manifold and then use it as a skeleton to globalize the rest of the objects in the full basin of attraction. We illustrate the techniques by applying them to a selection of different 3-dimensional models for single neuron and neural populations in neuroscience. We stress that the objects and functions computed are obtained in the full basin of attraction of the limit cycle with a numerical accuracy that we previously established. Finally, we consider a periodically perturbed 3-dimensional single neuron model, and we use the computed iPRF and iARFs to explore different approximations of the stroboscopic map to describe the dynamics. Namely, we consider the full 3D phase-amplitude map, the 2D map that considers the reduction to the slow submanifold, thus considering the phase and the slow amplitude variable, and the classical 1D phase map. We emphasize that the parameterization $K$ and the isochrons provide a geometrical interpretation of the effects of the perturbation and the validity of the different dimensional reductions.
The structure of the paper is the following: in Section \[sec:section2\] we provide the theoretical background to tackle the rest of the paper. In Section \[sec:section3\] we introduce the theoretical and computational methodology to obtain the parameterization of the invariant manifolds of the limit cycle and we use it to globalize the isochrons, isostables, iPRF and iARFs. In Section \[sec:section4\] we illustrate the methodology by computing these objects for different models in neuroscience. In Section \[sec:section5\], we apply the tools developed in the previous Sections to perform a study of the dynamics of a perturbed 3-dimensional single neuron model, while we explore the scope of validity of different dimensional reductions. We end the paper in Section \[sec:section6\], in which we present the conclusions of our work and its relation with other results in the field.
Background and statement of the problem {#sec:section2}
=======================================
Phase variable and Isochrons {#sec:phaseSec}
----------------------------
Consider an autonomous system of ODEs $$\begin{aligned}
\label{eq:mathDef_1}
\dot{x} = X(x), \quad x \in \mathbb{R}^{d}, \quad d \geq 2 ,\end{aligned}$$ whose flow is denoted by $\phi_t(x)$. Assume that $X$ is an analytic vector field and that system has a $T$-periodic hyperbolic attracting limit cycle $\Gamma$, parameterized by the phase variable $\theta = t/T$ as $$\begin{aligned}\label{eq:mathDef_2}
\gamma:\mathbb{T}:= \mathbb{R}/\mathbb{Z} &\to \mathbb{R}^{d}\\
\theta &\mapsto \gamma(\theta),
\end{aligned}$$ so that it has period 1, that is, $\gamma(\theta) = \gamma(\theta +1)$ and $x(t) = \gamma(\frac{t}{T})$ satisfies . Thus, the dynamics of on $\Gamma$ can be reduced to a single equation $$\dot{\theta} = \frac{1}{T}, \quad \quad \theta \in \mathbb{T}.$$
\[rm:varRemakr\] We recall that a periodic orbit is hyperbolic attracting if all its characteristic exponents have negative real part except the trivial one which is 0 (or equivalently, the Floquet multipliers are inside the unit circle except the trivial one which is 1). The characteristic exponents of $\Gamma$ can be obtained by solving the variational equations of system along the solution $\gamma(t/T)$. More precisely, the variational equations are given by the linearisation of the vector field $X$ around $\gamma$, that is, $$\label{eq:mjVarEqs}
\dot{\Phi} = DX(\gamma(t/T)) \Phi, \quad \quad \quad \text{with} \quad \Phi(0) = Id.$$ The solution of the system above $\Phi(t)$ evaluated at $t = T$, i. e. $\Phi(T)$, is known as the monodromy matrix. The eigenvalues of $\Phi(T)$, namely $\mu_i = e^{\lambda_i T}$, $i = 0, ..., d-1$, are known as the Floquet multipliers of the limit cycle $\Gamma$ and the values $\lambda_i$ as the characteristic exponents. The index $i = 0$ will be assigned from now on to the trivial multiplier $\mu_0 = 1$, so $\lambda_0 = 0$.
As we consider a hyperbolic attracting limit cycle $\Gamma$, the orbit of any point $p$ in the basin of attraction $\Omega$ of the limit cycle $\Gamma$ will approach asymptotically the orbit of a point $q$ in $\Gamma$ [@HirschPS77; @guckenheimer1975], that is, $$\begin{aligned}
\label{eq:synapseCurrent}
\lim_{t \to \infty} |\phi_t(q) - \phi_t(p)| = 0.\end{aligned}$$ We will say that the two points $p$ and $q \in \Omega$ have the same asymptotic phase. We define the isochron $\mathcal{I}_{\theta}$ as the set of points having the same asymptotic phase $\theta$, that is, $$\label{eq:isochronsDef}
\mathcal{I}_{\theta} = \{x \in \Omega \mid \ |\phi_t(x) - \phi_t(\gamma(\theta))| = |\phi_t(x) - \gamma\left(\theta + \frac{t}{T}\right)| \rightarrow 0 \quad \text{as} \quad t \rightarrow \infty \}.$$
Thus, the phase can be extended under generic conditions to a neighbourhood of the limit cycle via the concepts of asymptotic phase and isochrons [@winfree1967biological; @guckenheimer1975]. When $\Gamma$ is a hyperbolic attracting periodic orbit, the isochrons correspond to the leaves of the stable manifold $\mathcal{M}$ of $\Gamma$, which coincides with its basin of attraction $\Omega$. The sets of points where the asymptotic phase is not defined are called *phaseless sets* [@guckenheimer1975].
Phase-Amplitude variables and the parameterization method
---------------------------------------------------------
In this Section we present the so-called parameterization method adapted to our problem, which allows us to provide a description of the dynamics of the oscillator in terms of phase and amplitude variables. The parameterization method, introduced in [@cabre2003parameterization; @cabre2003parameterization2; @cabre2005parameterization], is a general functional analytic scheme to study $n$-dimensional invariant manifolds of a dynamical system. The method consists in finding a parameterization of the invariant manifold by means of solving a functional equation. This equation characterizes the invariance of the manifold and expresses the dynamics on this manifold in the coordinates induced by the parameterization. The coordinates will be chosen in such a way that the dynamics expresses in the simplest way possible (in some cases linear). Mathematically, one looks for an embedding $K: \mathcal{U} \subset \mathbb{R}^n \rightarrow \mathbb{R}^d$ and a vector field $\mathcal{X}: \mathcal{U} \subset \mathbb{R}^n \rightarrow \mathbb{R}^n$, $n \leq d$ and $\mathcal{U}$ open, such that $$\label{eq:invariantgen}
DK \cdot \mathcal{X}=X \circ K,$$ so that $$\mathcal{M}:= Range(K)=\{ K(u) \in \mathbb{R}^d \quad |\quad u \in \mathcal{U} \subset \mathbb{R}^n \}$$ is invariant under the flow of $X$ and the vector field $\dot{u}=\mathcal{X}(u)$ describes the dynamics on the invariant manifold $\mathcal{M}$. Thus, the method provides information not only on the embedding but also on the dynamics on the manifold. Moreover, the method leads to efficient numerical algorithms to compute high order Taylor expansions of $K$ and $\mathcal{X}$. The book [@haro2016] contains a detailed description of the analytical and numerical aspects of the method in several contexts as well as a complete list of references.
In our case, we look for a parameterization $K$ of the $d$-dimensional stable manifold $\mathcal{M}$ of the periodic orbit, which coincides with the basin of attraction $\Omega$ of $\Gamma$. This is equivalent to look for a change of variables $K$ that conjugates the vector field $X$ in $\Omega$ to a vector field $\mathcal{X}$ with a simpler expression of the dynamics.
In our case, we look for a local analytic diffeomorphism $$\label{eq:kThetaSigma}
\begin{aligned}
K : \mathbb{T} \times \mathcal{B} \subset \mathbb{T} \times \mathbb{R}^{d-1} &\rightarrow \mathbb{R}^d \\
(\theta, \sigma) &\rightarrow K(\theta, \sigma),
\end{aligned}$$ such that it satisfies the following invariance equation $$\label{eq:mjInvEq}
\frac{1}{T}\frac{\partial}{\partial\theta}K(\theta, \sigma) + \sum_{i=1}^{d-1} \lambda_i \sigma_i \frac{\partial}{\partial \sigma_i}K(\theta, \sigma) = X(K(\theta, \sigma)),$$ where $T$ is the period of the limit cycle $\Gamma$ and $\lambda_1$, ..., $\lambda_{d-1} \in \mathbb{R}$ its non-trivial characteristic exponents. Equation will be the centrepiece of our approach.
We can think of as a change of coordinates. The new coordinates will be the phase $\theta \in \mathbb{T}$ introduced in and the amplitude coordinates $\sigma_1$, ..., $\sigma_{d-1} \in \mathbb{R}$, corresponding to transverse directions to the limit cycle. Thus, the dynamics of the vector field $X$ in expressed in these new variables $(\theta, \sigma) \in \mathbb{T} \times (\mathcal{B} \subset \mathbb{R}^{d-1})$ is given by $$\label{eq:flipoTete}
\dot{\theta} = \frac{1}{T}, \quad \quad \quad \dot{\sigma} = \Lambda \cdot \sigma, \quad \quad \text{with} \quad \quad \Lambda =
\begin{pmatrix}
\lambda_1 & & \\
& \ddots & \\
& & \lambda_{d-1}
\end{pmatrix}.$$ That is, the variable $\theta$ rotates at a constant speed $1/T$, while the variables $\sigma_i$ contract at a rate $\lambda_i$. Note that the vector field in is the vector field that we named $\mathcal{X}$ in . With this choice of $\mathcal{X}$, Eq. in our general presentation becomes Eq. .
In Section \[sec:section3\] we will show that, provided that the characteristic exponents satisfy certain non-resonance conditions, one can indeed solve the functional equation above and find a map $K(\theta, \sigma)$, at least formally. We foresee here that, for the purposes of this paper, we are going to assume that the characteristic exponents are real and distinct, that is, $\lambda_{1}<\cdots<\lambda_{d-1}<0$. Moreover, we are going to assume that they are non-resonant, that is $$\label{eq:resonance}
\sum_{i=1}^{d-1} m_i \lambda_i \neq \lambda_k, \quad \textrm{for any}\enskip k=1,\ldots,d-1,\enskip \textrm{for all} \enskip m_i \in \mathbb{Z} .$$ The case of complex exponents or degeneracies, which is beyond the scope of this paper, is discussed in [@castelli2015parameterization].
Notice that the characteristic exponents are negative because we assume that the limit cycle is hyperbolic attracting. Indeed, this is the interesting case in the neuroscience context. The case of a hyperbolic repelling limit cycle, though, is completely analogous, just reversing time.
\[rem:nres0\] The non-resonance condition will be necessary to solve the invariance equation as we will see in Section \[sec:section3\] (see Remark \[rem:nonres\]). The existence of a resonance at some order does not prevent the existence of the stable invariant manifold and therefore the isochrons (see for instance, [@guckenheimer2013nonlinear]). However, it prevents the existence of an analytic conjugacy that conjugates the vector field $X$ to a linear one of the form . In that case, the parameterization method still works, but it is necessary to conjugate to a polynomial vector field instead of the linear field as done here. For a thorough discussion of resonances and the parameterization method for fixed points and equilibria, see [@cabre2003parameterization].
Notice that Eq. does not have a unique solution. Indeed, if $K$ is a solution then $\tilde K(\theta,\sigma)=K(\theta + \omega, b \sigma)$ is also a solution, for any $\omega \in \mathbb{T}$ and $b \in \mathbb{R}^{d-1}$. The meaning of $\omega$ is the choice of the origin of time, and $b$ corresponds to the choice of units in $\sigma$. In Section \[sec:section32\] we discuss how to choose these constants to guarantee the numerical stability of the method.
The evolution of the flow $\phi_t$ in the coordinates $(\theta,\sigma)$ given in becomes $$\label{eq:aboveEq}
\phi_t(K(\theta, \sigma)) = K\Big(\theta + \frac{t}{T}, e^{\Lambda t}\sigma\Big).$$
The map $K$ in allows to define a scalar function $\Theta$ that assigns the asymptotic phase to any point $x$ in a neighbourhood $\Omega$ of the limit cycle $\Gamma$. Indeed, $$\begin{aligned}
\Theta:\Omega \subset \mathbb{R}^{d} &\to \mathbb{T},\\
x &\mapsto \Theta(x) = \theta \quad \quad \text{where} \enskip x = K(\theta, \sigma),\enskip \textrm{for some} \enskip \sigma \in \mathbb{R}^{d-1}.
\end{aligned}\label{eq:mathDef_3}$$ Notice that $\Theta(\phi_t(x)) = \Theta(x) + \frac{t}{T}$. The level curves of $\Theta$ correspond to the isochrons $\mathcal{I}_{\theta}$ introduced in , that is, $$\mathcal{I}_{\theta} = \{x \in \Omega \quad | \quad \Theta(x) = \theta \}.$$
Analogously, the map $K$ in also allows us to define the scalar functions $\Sigma_i$, for $i=1,\ldots, d-1$ that assign the amplitude variable $\sigma_i$ to any point $x \in$ $\Omega$: $$\label{eq:sigma3Dcase}
\begin{aligned}
\Sigma_i : \Omega \subset \mathbb{R}^{d} &\to \mathbb{R},\\
x &\mapsto \Sigma_i(x) = \sigma_i, \quad \text{where} \enskip x = K(\theta, \sigma), \enskip \textrm{for some} \enskip \theta \in \mathbb{T}.
\end{aligned}$$ Notice that $\Sigma_i(\phi_t(x)) = \Sigma_i(x)e^{\lambda_i t}$, for $i=1,\ldots,d-1$. The level curves of $\Sigma_i$ are called *isostables* (see [@mauroy2018global; @mauroy2013isostables]) or *A-curves* for the 2-dimensional case (see [@castejon2013phase]), and correspond to the sets of points $$\label{eq:aCurvesDef}
\mathcal{A}_{\sigma_i}^i = \{x \in \Omega \quad | \quad \Sigma_i(x) = \sigma_i \}.$$ We will denote by $\Sigma$ the vector-valued function $\Sigma(x):=(\Sigma_1(x),\ldots,\Sigma_{d-1}(x))$.
Phase and Amplitude Response Functions {#sec:phaseAmpFun}
--------------------------------------
Let us consider that an instantaneous pulse of amplitude $A$ is applied to the trajectory at time $t=t_s$ in the direction $\vec v \in \mathbb{R}^{d}$, that is $$\dot{x} = X(x) + A \vec{v} \delta(t-t_s),$$ where $\delta(t)$ is the Dirac delta function. We can assume without loss of generality that $t_s=0$. Such instantaneous perturbation, when acting over a trajectory on a point $x = K(\theta, \sigma) \in \Omega$, will displace the trajectory from the point $x$ to a new point $x + \Delta x=K(\theta_{new},\sigma_{new}) \in \Omega$ having a new phase $\theta _{new}$ and new amplitude $\sigma_{new}$ given by: $$\label{eq:defparf}
\begin{aligned}
\theta_{new} = \Theta(x + \Delta x) = \Theta(x) + PRF(A, \theta,\sigma), \\
\sigma_{new} = \Sigma(x + \Delta x) = \Sigma(x) + ARF(A, \theta, \sigma),
\end{aligned}$$ where the PRF and ARF are the Phase Response Function the Amplitude Response Function, respectively (see [@castejon2013phase; @guillamon2009computational]). These functions quantify the shift in phase and amplitude due to the effect of the perturbation at a given point $x = K(\theta, \sigma) \in \Omega$, respectively.
For the case of an instantaneous pulse of weak amplitude $|A| \ll 1$, then $\Delta x \ll 1$, and therefore $$\begin{aligned}\label{eq:arcs_2}
\theta_{new} &= \Theta(x + \Delta x) = \Theta(x) + \nabla\Theta(x) \cdot \Delta x + O(|\Delta x|^2),\\
\sigma_{new} &= \Sigma(x + \Delta x) = \Sigma(x) + \nabla\Sigma(x) \cdot \Delta x + O(|\Delta x|^2), \quad \quad \quad \text{for} \quad x \in \Omega.
\end{aligned}$$ Notice that the first order approximation of the PRF and the ARFs is given by $\nabla\Theta(x)$ and $\nabla\Sigma(x)$, which correspond to the infinitesimal PRF (iPRF) and the infinitesimal ARFs (iARFs), respectively.
\[rm:prcsRM\] The PRF and the ARFs are the natural extension of the Phase Response Curve (PRC) and Amplitude Response Curves (ARCs). In particular, the PRC and ARCs correspond to the restriction of the PRF and the ARFs to the limit cycle ($\sigma = 0$), respectively. That is, $$\begin{aligned}
PRC(A, \theta) = PRF(A, \theta, 0),\\
ARC(A, \theta) = ARF(A, \theta, 0).
\end{aligned}$$ Moreover, the infinitesimal PRC (iPRC) and infinitesimal ARCs (iARC) correspond to $$\begin{aligned}
iPRC(\theta) = \nabla\Theta(\gamma(\theta)),\\
iARC(\theta) = \nabla\Sigma(\gamma(\theta)).
\end{aligned}$$
\[rem:sobreprf\] Notice that the PRF (resp. ARFs) are defined in as real-valued functions, while the iPRF $\nabla \Theta$ (resp. iARFs $\nabla \Sigma$) are defined in as vector-valued functions. Indeed, its range is a vector with $d$-components. However, in the neuroscience literature, since the perturbation $\Delta x$ occurs usually in the direction of the voltage (typically corresponding to the first component), sometimes the iPRF (resp. iARFs) refer only to the real-valued function corresponding to $\partial \Theta (x)/\partial V$ (resp. $\partial \Sigma (x)/\partial V$). In this paper, abusing language, we will refer to $\nabla \Theta$ (resp. $\nabla \Sigma$) and its first component indistinctly as iPRF (resp. iARFs).
The iPRF $\nabla\Theta(x)$ and the iARFs $\nabla\Sigma(x)$ can be computed by means of the parameterization $K(\theta, \sigma)$ in . Indeed, taking derivatives on both sides at the expression $K(\Theta(x), \Sigma_1(x), ..., \Sigma_{d-1}(x)) = x$, we have for $x = (x_1, ..., x_d) \in \Omega$: $$\label{eq:prodidentity}
Id_{d \times d}
=
\begin{bmatrix} DK(\theta, \sigma) \end{bmatrix} \begin{bmatrix} \nabla\Theta(x) \\ \nabla\Sigma_1(x) \\
\vdots\\
\nabla\Sigma_{d-1}(x) \end{bmatrix},$$ and therefore $$\label{eq:usefulSpace} \begin{bmatrix} \nabla\Theta(x) \\ \nabla\Sigma_1(x) \\
\vdots\\
\nabla\Sigma_{d-1}(x) \end{bmatrix}
=
{\begin{bmatrix} DK(\theta, \sigma) \end{bmatrix}}^{-1}.$$ Moreover, as it is shown in [@guillamon2009computational; @castejon2013phase], for points $x = K(\theta, \sigma) \in \Omega$, $\nabla \Theta$ and $\nabla \Sigma_i$ satisfy the following adjoint equations $$\begin{aligned}\label{eq:prcs_4b}
\frac{d \nabla\Theta(\phi_t(x))}{dt} &= -DX^T(\phi_t(x))\nabla\Theta(\phi_t(x)),\\
\frac{d\nabla\Sigma_i(\phi_t(x))}{dt} &= \left( \lambda_i - DX^T(\phi_t(x)) \right)\nabla\Sigma_i(\phi_t(x)),
\end{aligned}$$ which is specially relevant for computational purposes. Indeed, as we discuss in Section \[sec:section32\], we use Eq. to compute $\nabla \Theta$ and $\nabla \Sigma$ in a local neighbourhood of the limit cycle $\Omega_{loc} \subseteq \Omega$ using the local approximate expression for $K$. Moreover, this computation provides an initial condition for system , so that one can obtain the values of $\nabla \Theta$ and $\nabla \Sigma$ beyond $\Omega_{loc}$ by means of backwards integration.
The classical adjoint method [@ErmentroutKopell91; @brown2004phase; @ErmentroutTerman2010] considers $x \in \Gamma$ in Eq. top, that is, with $\phi_t(x)=\gamma(t/T)$ and looks for a $T$-periodic solution to obtain the iPRC. However, if we want to extend the adjoint method to a neighbourhood of the limit cycle, we cannot impose periodicity conditions. Thus, the problem lacks conditions to be solved uniquely, unless one knows a way to obtain initial conditions. We use that Eq. provides an expression for $\nabla\Theta$ and $\nabla\Sigma$ for points not restricted to the limit cycle, and therefore an initial condition for the adjoint equations to be solve the problem uniquely. So, the computation of $K(\theta, \sigma)$ allows us to extend the use of the adjoint equations beyond the limit cycle.
Methodology and Numerical Algorithm {#sec:section3}
===================================
In this Section, we describe the numerical methodology to solve the invariance equation for the unknowns $K, T$ and $\Lambda$. In particular, in Section \[sec:section31\], we compute a formal expression for the parameterization $K(\theta, \sigma)$ in Fourier-Taylor series. Next, in Section \[sec:sectionNum\], we discuss the numerical implementation of the computation of $K(\theta, \sigma)$. As the methodology used to numerically obtain $K$ involves the truncation of the power series up to a given order $L$, in Section \[sec:section32\] we discuss the domain of accuracy of the truncated $K$. Section \[sec:local\] is devoted to use the local approximation of $K$ to compute local isochrons, isostables, iPRF and iARFs. Finally, in Section \[sec:isoComp3D\] we introduce a method to globalize the local approximation of the map $K(\theta, \sigma)$ to a larger domain which automatically provides the globalization of the isochrons, isostables, iPRF and iARFs. Although the methodology can be applied to any system in $\mathbb{R}^d$, the details are given for the case $d=3$.
A formal solution for the invariance equation {#sec:section31}
---------------------------------------------
The dynamics of system for $d=3$, in terms of the phase-amplitude variables is given by $$\label{eq:thetaSigma3var}
\dot{\theta} = \frac{1}{T}, \quad \quad \quad \dot{\sigma} = \Lambda \cdot \sigma, \quad \quad \text{with} \quad \quad \Lambda =
\begin{pmatrix}
\lambda_1 & 0 \\
0 & \lambda_2
\end{pmatrix},$$ where $\sigma = (\sigma_1, \sigma_2) \in \mathbb{R}^2$, and $\lambda_1$ and $\lambda_2$ are the characteristic exponents of $\Gamma$. We recall that we assume that $\lambda_1, \lambda_2$ are real and distinct (see Eq. ).
The invariance equation in the case $d=3$ writes as $$\label{eq:mjInvEq2}
\frac{1}{T}\frac{\partial}{\partial\theta}K(\theta, \sigma) + \sum_{i=1}^{2} \lambda_i \sigma_i \frac{\partial}{\partial \sigma_i}K(\theta, \sigma) = X(K(\theta, \sigma)).$$
In order to solve the above invariance equation we assume a formal series solution for Eq. of the form: $$\label{eq:mjFourierTaylor}
K(\theta, \sigma) = \sum_{m=0}^{\infty} \sum_{\alpha=0}^{m} K_{\alpha, m-\alpha}(\theta) \sigma^{\alpha}_1 \sigma^{m-\alpha}_2,$$ where the functions $K_{\alpha,m-\alpha}: \mathbb{T} \rightarrow \mathbb{R}^d$, for $\alpha = 0, ..., m$ and $m \in \mathbb{N}$. Therefore, we substitute $K(\theta, \sigma)$ in in Eq. and expand the vector field $X(K(\theta, \sigma))$ in Taylor series with respect to the variable $\sigma$ about $\sigma=0$. To obtain the expression for $K(\theta, \sigma)$ one just has to collect terms with the same power of $\sigma$ and solve the resulting equations. Next, we explain how to solve Eq. for each degree.
For $m = 0$ the term $K_0(\theta) := K_{00}(\theta)$ satisfies the equation $$\label{eq:eqMzero}
\frac{1}{T}\frac{d}{d\theta}K_0(\theta) = X(K_0(\theta)).$$ Clearly, the solution of is the limit cycle itself, that is $K_0(\theta) = \gamma(\theta)$.
\[rm:remark31\] Notice that if $K_0(\theta)$ is a solution, then $K_0(\theta + \omega)$ is also a solution for any $\omega \in [0,1)$. This means that the phase of a given oscillation can be fixed arbitrarily. As we will see in Section \[sec:section4\], for the examples in this paper, we follow the standard criterion in neuroscience which sets the zero phase at the maximum value of the voltage coordinate.
For $m = 1$, the equations for $K_{10}(\theta)$ and $K_{01}(\theta)$ are $$\label{eq:eqMone}
\begin{aligned}
\frac{1}{T}\frac{d}{d\theta}K_{10}(\theta) + \lambda_1 K_{10}(\theta) &= DX(K_0(\theta))K_{10}(\theta), \\
\frac{1}{T}\frac{d}{d\theta}K_{01}(\theta) + \lambda_2 K_{01}(\theta) &= DX(K_0(\theta))K_{01}(\theta),
\end{aligned}$$ respectively. The solutions for these equations are given by $$\label{eq:eqMzeroSols}
K_{10}(\theta) = \Phi(\theta T)e^{-\lambda_1 \theta T }v_1, \quad \quad \quad K_{01}(\theta) = \Phi(\theta T )e^{-\lambda_2 \theta T}v_2,$$ where $\Phi(t)$ is the solution of the variational equations and $v_i$ is the eigenvector of $\Phi(T)$ (the monodromy matrix) associated to the non-trivial $i$-th Floquet multiplier $\mu_i$, for $i = 1,2$.
\[rm:remark32\] Notice that the solutions $K_{10}(\theta)$ and $K_{01}(\theta)$ of are non unique. If $v_1$ and $v_2$ are eigenvectors of the monodromy matrix so are $b_1 v_1$ and $b_2 v_2$ for any $b_1, b_2 \in \mathbb{R}$, giving rise to new solutions $b_1 K_{10}$ and $b_2 K_{01}$. Even though all the choices of $K_{10}$ and $K_{01}$ are mathematically equivalent, the choice affects the numerical properties of the algorithm. See Section \[sec:section32\] for a more detailed discussion.
Finally, for $m \geq 2$, the terms $K_{\alpha, m-\alpha}(\theta)$, $\alpha = 0, ..., m$, satisfy the so-called *homological equations*: $$\label{eq:homologicalEqs}
\frac{1}{T}\frac{d K_{\alpha, m-\alpha}(\theta)}{d\theta} + (\alpha \lambda_1 + (m-\alpha) \lambda_{2}) K_{\alpha, m-\alpha}(\theta) = DX(K_0(\theta)) K_{\alpha, m-\alpha}(\theta) + B_{\alpha, m-\alpha}(\theta),$$ where $B_{\alpha, m-\alpha}(\theta)$ is the coefficient of the term $\sigma_1^\alpha \sigma_2^{m-\alpha}$ in the Taylor expansion of $$\label{eq:Bexpand}
\emph{X}\left( \sum_{n=0}^{m-1} \sum_{\alpha=0}^{n} K_{\alpha, n-\alpha}(\theta) \sigma^{\alpha}_1 \sigma^{n-\alpha}_2 \right).$$ Notice that $B_{\alpha, m-\alpha}(\theta)$ is an explicit polynomial depending only on the terms of order lower than $m$, that is, the functions $K_{\alpha, n-\alpha}(\theta)$ for $n<m$ and whose coefficients are the derivatives of $X$ evaluated at $K_0$. They can be numerically computed using automatic differentiation techniques [@haro2016] (see Appendix \[sec:autoDifAp\]).
The Eq. can be solved assuming that $K_{\alpha, m-\alpha}(\theta)$ can be written in Fourier series so its coefficients are the unknowns. The resulting system of equations for the Fourier coefficients is linear, but it involves a large dimensional matrix which has a high computational cost. To avoid this numerical drawback, in the next Section, we review the method proposed in [@castelli2015parameterization] to solve the homological equation in an efficient way using the Floquet normal form.
### Reducibility of the homological equations via Floquet normal form {#sec:floqSchm}
In this Section we use the Floquet normal form to solve the homological equations . This allows us to transform the homological equations to a linear system with diagonal constant coefficient matrix in Fourier space, following [@castelli2015parameterization]. A similar idea has been also applied in [@huguet2013computation].
To avoid stodgy notation, from now on we will use ${\bm{\alpha}}:=(\alpha,m-\alpha)$.
First of all recall that, by Floquet theory [@floquet1883equations], the fundamental matrix $\Phi(t)$ of system can be written as $$\label{eq:floqNf}
\Phi(t) = \mathcal{Q}(t)e^{t R},$$ where $\mathcal{Q}(t)$ is a $T$-periodic $3 \times 3$ matrix and $R$ is a real-valued $3 \times 3$ matrix. As $\Phi(t)$ is a solution of , using , we have $$\label{eq:floquetDifInv}
\frac{1}{T} \left(\frac{d Q(\theta)}{d \theta} + Q(\theta) TR \right) = DX(\gamma(\theta)) Q(\theta),$$ where $Q(\theta) := \mathcal{Q}(\theta T)$.
Then, we introduce the function $w: \mathbb{T} \rightarrow \mathbb{R}^3$ and write $K_{{\bm{\alpha}}}(\theta)$ as $$K_{{\bm{\alpha}}}(\theta) = Q(\theta)w(\theta),$$ and substituting it in , we have $$\label{eq:mjIneed}
\frac{1}{T} \left(\frac{dQ(\theta)}{d\theta}w(\theta) + Q(\theta)\frac{dw(\theta)}{d\theta} \right) + \Upsilon Q(\theta)w(\theta) = DX(K_0(\theta))Q(\theta)w(\theta) + B_{{\bm{\alpha}}}(\theta),$$ where we have introduced the constant matrix $\Upsilon := (\alpha \lambda_1 + (m-\alpha) \lambda_{2})\cdot Id_{3 \times 3}$.
Then, using in equation , we obtain $$\frac{1}{T} \Big(-Q(\theta)TRw(\theta) + Q(\theta)\frac{d w(\theta)}{d \theta} \Big) + \Upsilon Q(\theta)w(\theta) = B_{{\bm{\alpha}}}(\theta),$$ and multiplying both sides by $Q^{-1}(\theta)$ we have $$\label{eq:batiburillo}
\frac{1}{T}\frac{d w(\theta)}{d \theta} = (-\Upsilon + R)w(\theta) + Q^{-1}(\theta)B_{{\bm{\alpha}}}(\theta).$$
Finally, we assume that the matrix $R$ in can be diagonalized, that is, there exists a matrix $C$ such that $$\label{eq:tereJ}
J = C^{-1}RC = \begin{pmatrix}
\lambda_0 & 0 & 0 \\
0 & \lambda_1 & 0 \\
0 & 0 & \lambda_2
\end{pmatrix}.$$ We make a final coordinate transformation and define $u(\theta)$ as $w(\theta) = Cu(\theta)$, and multiplying both sides by $C^{-1}$, expression reads as $$\label{eq:finalEqU}
\frac{1}{T} \frac{d u(\theta)}{d \theta} = (-\Upsilon + J)u(\theta) + A_{{\bm{\alpha}}}(\theta),$$ where $$\label{eq:Afunction}
A_{{\bm{\alpha}}}(\theta) = C^{-1}Q^{-1}(\theta)B_{{\bm{\alpha}}}(\theta).$$
Finally, we write $u(\theta)$ and $A_{{\bm{\alpha}}}(\theta)$ in Fourier series, that is, $$\label{eq:mjIneedAgain}
u(\theta) = \sum_{k = -\infty}^{\infty} u_{k}e^{2\pi ik\theta}, \quad \quad A_{{\bm{\alpha}}}(\theta) = \sum_{k = -\infty}^{\infty} A_{k}e^{2\pi ik\theta}, \quad \quad \quad A_{k}, u_{k} \in \mathbb{C}^3,$$ and substitute expressions in Eq. . We obtain a linear system for the Fourier coefficients $u_k=(u_k^{(1)},u_k^{(2)},u_k^{(3)}) \in \mathbb{C}^3$ which is diagonal and can be solved componentwise, thus obtaining the following expression for the Fourier coefficients: $$\label{eq:mjFourierCoefs}
u^{(j)}_{k} = \frac{1}{\frac{2\pi ik}{T} + \alpha \lambda_1 + (m-\alpha) \lambda_{2} - \lambda_{j-1}}A_{k}^{(j)},$$ for $j = 1, 2, 3$. Notice that the superindex $(j)$ refers to each component of the vectors $u_k$ and $A_k$. Finally, the solution $K_{\alpha, m-\alpha}(\theta)$ of Eq. is given by $$\label{eq:Kfunction}
K_{\alpha, m-\alpha}(\theta) = K_{{\bm{\alpha}}}(\theta) = Q(\theta)Cu(\theta).$$
\[rem:nonres\] The Fourier coefficients $u^{(j)}_{k}$ in are formally well defined to all orders provided that, for any $k \in \mathbb{N}$ and $m\geq 2$, $\alpha=0,\ldots,m$, we have $$\frac{2\pi ik}{T} + \alpha \lambda_1 + (m-\alpha) \lambda_{2} - \lambda_{j-1} \neq 0, \, \quad j =1,2,3.$$ Notice that this condition is always satisfied since we assumed that the characteristic exponents $\lambda_i$ are real negative and distinct, together with the the non-resonant condition (recall that $\lambda_0=0$).
\[rem:general\] Notice that the Floquet reduction is computed only once and it is then used to find the solution of the homological equations at any degree. Indeed, from expressions --, it is clear that to obtain the terms $K_{{\bm{\alpha}}}$ for different ${\bm{\alpha}}$’s, only the term $B_{{\bm{\alpha}}}$ needs to be recomputed, while the matrices $C$ and $Q$ are always the same.
Numerical computation of $K$ {#sec:sectionNum}
----------------------------
In this Section we explain how to numerically solve Eqs. , and using the methodology described in Section \[sec:section31\] and, thus obtain a local approximation of $K(\theta, \sigma)$.
- For $m=0$ (see Eq. ), we need to compute the periodic solution $\Gamma$. To do so, we construct a Poincaré section and use a Newton method to find a fixed point of the corresponding Poincaré map. By doing this, we obtain a point $x_0 \in \Gamma$ and the period $T$.
We integrate system with initial condition $x(0) = x_0$ and the variational equations altogether for a time $T$ to obtain $x(\theta T) =: K_0(\theta)$ and $\Phi(\theta T)$ for $\theta \in [0, 1)$. We store them for equidistant values of $\theta$; that is $\theta_i = i/N$ for $i = 0, ..., N-1$, which is equivalent to store the coefficients of the Fourier series up to order $N$. Indeed, we can switch between real and Fourier space by means of a Fast Fourier Transform (FFT) algorithm [@brigham1978fast].
- For $m=1$ (see Eq. ), we consider the monodromy matrix $\Phi(T)$ and obtain its eigenvalues $\mu_i$ (Floquet multipliers), the Floquet exponents $\lambda_i = \frac{1}{T} \ln(\mu_i)$ and their respective eigenvectors $v_i$ for $i = 1,2$. We compute $K_{10}$ and $K_{01}$ according to the formulas , and we store them again for the same equidistant values of $\theta$.
- For $m\geq2$, we use the scheme described in Section \[sec:floqSchm\] to solve Eq. . We solve these equations up to order $m=L$, $L \in \mathbb{N}$.
To obtain the functions $B_{{\bm{\alpha}}}$ in the homological equation we need to compute the Taylor expansion of $X(f(\sigma_1, \sigma_2))$, where $f(\sigma_1, \sigma_2)$ is a particular Taylor polynomial in $\sigma_1$, $\sigma_2$ up to order $m-1$ (see Eq. ). As the vector field $X$ is analytic and consists of a combination of elementary functions, we can use automatic differentiation techniques [@griewank2008evaluating; @haro2016] to compute the coefficients $B_{{\bm{\alpha}}}$ of the Taylor expansion of $X$ up to arbitrary order (see Appendix \[sec:autoDifAp\] for more details).
To apply the scheme described in Section \[sec:floqSchm\], we first compute the matrices $\mathcal{Q}$ (and therefore $Q$) and $R$ in and the matrix $C$ satisfying (see Appendix \[sec:floquetComputation\]). We use the formula to obtain the functions $A_{{\bm{\alpha}}}$ in real space and apply the FFT algorithm to obtain the Fourier coefficients $A_k$ (see ). Finally, we use formulae and together with the inverse FFT to obtain the function $K_{{\bm{\alpha}}}$ in real space. Again, we save this function for the same equidistant values of $\theta$.
To check the accuracy of the each of the solutions $K_{{\bm{\alpha}}}$ obtained, we substitute them in their corresponding equation (Eq. for $m=0$, Eq. for $m=1$ and Eq. for $m\geq2$) for discrete values of $\theta$, that is, $\theta_i = i/N$ for $i = 0, ..., N-1$. For each value $\theta_i$, this substitution provides an error value $E_{{\bm{\alpha}}}(\theta_i)$. Finally, we compute the discrete $\ell_1$ norm of the error to get the accuracy, that is, $$\label{eq:errortol}
\Vert E_{{\bm{\alpha}}} \Vert_{\ell_1} = \frac{1}{N} \sum_{i=0}^{N-1} |E_{{\bm{\alpha}}}(\theta_i)|.$$
Numerical errors and domain of accuracy of the approximate solution {#sec:section32}
-------------------------------------------------------------------
Given a Taylor truncation at order $L \in \mathbb{N}$ and a Fourier truncation at order $N \in \mathbb{N}$, the above procedure provides an approximate solution of Eq. of the form $$\label{eq:truncatedSeries3D}
\bar{K}(\theta, \sigma) = \sum_{m=0}^{L} \sum_{\alpha=0}^{m} \bar{K}_{\alpha, m-\alpha}(\theta) \sigma^{\alpha}_1 \sigma^{m-\alpha}_2, \enskip \text{where} \enskip \bar{K}_{{\bm{\alpha}}}(\theta) = \sum_{k=-N/2}^{N/2} c_{{\bm{\alpha},k}} e^{2 \pi ik \theta}, \enskip c_{{\bm{\alpha},k}} \in \mathbb{C}.$$ Recall that we have defined ${\bm{\alpha}}:=(\alpha,m-\alpha)$.
Next we discuss how to choose $L$ and $N$, as well as how to determine how good is the approximate solution and its domain of accuracy.
### Number of Fourier coefficients $N$
To decide how many Fourier coefficients $N$ we have to compute (or equivalently, how many points on real space we have to store), we use the same criterion as in [@guillamon2009computational]. Thus, we pick a value $N$ such that the norm of the series with the last $10\%$ of Fourier coefficients is smaller than a given tolerance $E_{tail}$ (in the examples considered we set it at $10^{-10}$), that is $$\label{eq:theTail}
|K_{{\bm{\alpha}}}^{tail}| = 2 \sum_{k=0.9N/2}^{N/2} |c_{{\bm{\alpha},k}}| < E_{tail},$$ for all computed ${\bm{\alpha}}$’s. We start with a value $N$ for which this condition is satisfied for $K_0$, $K_{10}$ and $K_{01}$, and we compute the rest of the $K_{{\bm{\alpha}}}$, for $m=2 \ldots L$. At the end of the computation we check this condition for the new $K_{{\bm{\alpha}}}$. Whenever it is not satisfied we recompute $K_{{\bm{\alpha}}}$’s again with $2N$.
### Local approximation and number of Taylor coefficients $L$
The solution $\bar{K}$ of is computed as a power series in $\sigma$, yet we do not expect that it is a good approximate solution for all $\vert \sigma \vert > 0$. Rather we expect that it is good only in a neighbourhood of $\sigma=0$ (the limit cycle). Of course, the domain where the approximation is valid will depend on the error tolerance $E_{tol}$ and the order of the approximation $L$.
Let us define the error function $E$ as $$\label{eq:errorEq}
E(\theta, \sigma) := \frac{1}{T}\frac{\partial}{\partial \theta} \bar{K}(\theta, \sigma) + \sum^2_{i = 1} \lambda_i \sigma_i \frac{\partial}{\partial \sigma_i}\bar{K}(\theta, \sigma) - X(\bar{K}(\theta, \sigma)).$$
Therefore, for a given error tolerance $E_{tol} > 0$, we can define a numerical domain of approximation $\Omega_{loc}(E_{tol})$ for the solution $\bar K$ in the following way $$\label{eq:mjOmegaLoc}
\Omega_{loc}(E_{tol}):= \{x \in \Omega \subset \mathbb{R}^3 \mid x=K(\theta, \sigma) \textrm{ for } (\theta,\sigma) \in \mathbb{T} \times \mathbb{R}^2 \textrm{ and } \Vert E(\theta, \sigma) \Vert < E_{tol} \},$$ where $\parallel \cdot \parallel$ is the euclidean norm in $\mathbb{R}^3$. From now on we will refer to the domain $\Omega_{loc}$ without writing explicitly the dependence on $E_{tol}$.
In order to compute numerically the domain $\Omega_{loc}$, we perform the following strategy. For a fixed $\theta$, we write $\sigma_1=r \cos \varphi$ and $\sigma_2=r \sin \varphi$, for $r \in \mathbb{R}^+$ and $\varphi \in [0,2 \pi)$. For a fixed value $\varphi$, we look for the maximum $r$ such that $$x=K(\theta,r \cos \varphi, r \sin \varphi) \in \Omega_{loc}.$$ Thus, we define the functions $R_{\theta}: [0,2\pi) \rightarrow \mathbb{R}^+$, such that $$\label{eq:smax}
x=K(\theta,\sigma_1,\sigma_2) \in \Omega_{loc} \iff \Vert (\sigma_1,\sigma_2) \Vert < R_{\theta} (\varphi), \, \textrm{ where } \tan \varphi=\frac{\sigma_2}{\sigma_1}.$$ Moreover, we define the set $\mathcal{B}_{loc}(\theta)$ as $$\label{eq:bloc}
\mathcal{B}_{loc}(\theta):=\{ (\sigma_1,\sigma_2) \in \mathbb{R}^2 \quad | \quad \Vert (\sigma_1,\sigma_2) \Vert < R_{\theta} (\varphi), \, \textrm{ where } \tan \varphi= \frac{\sigma_2}{\sigma_1}\}.$$
As we already mentioned in Remark \[rm:remark32\], the choice of $v_1$ and $v_2$ in is non unique. Although, theoretically, we can choose any vectors $b_1 v_1$, $b_2 v_2$, for any $b_1, b_2 \in \mathbb{R}$, their choice affects the numerical stability of the method and, more importantly, the size of the local approximation. Indeed, once $b_{1}$ and $b_{2}$ are fixed, the monomial $K_{\alpha, m-\alpha}$ will be multiplied by a factor $b^\alpha_{1}b^{m-\alpha}_{2}$. Therefore, if one chooses small values for $b_{1}$ and $b_{2}$, the terms $K_{\alpha, m-\alpha}$ will become small very fast as $m$ increases, so increasing the order $L$ does not provide any extra information since we are just adding terms that are smaller than the machine error. By contrast, if one chooses large values for $b_{1}$ and $b_{2}$, the functions $K_{\alpha, m-\alpha}$ will blow up fast as $m$ increases and, eventually, those values will be too large to computationally operate with them. Therefore, to obtain a local approximation that extends to a larger neighbourhood, one has to choose appropriate values for $b_{1}$ and $b_{2}$ so that the coefficients $K_{\alpha, m-\alpha}$ can be kept at order 1 and the round-off errors are greatly reduced. We refer the reader to [@huguet2013computation] for a more detailed discussion on the role of $b_1$ and $b_2$.
We determine the value of $b_1$ and $b_2$ and the order of the expansion $L$ by numerical experimentation. We typically have in mind a certain error tolerance for (around $10^{-6}$) and we stop at a certain order $L$ whenever this error is larger than the tolerance and cannot be made smaller by changing $b_1$ and $b_2$. So, we look for solutions computed to the highest order and having the largest accurate domains.
Local isochrons, isostables, iPRF and iARFs {#sec:local}
-------------------------------------------
In the previous section we have discussed the domain of accuracy $\Omega_{loc}$ for the local approximation $\bar K$. Thus, in this domain, we define the local isochrons ${\mathcal{I}_{\theta}^{loc}}$ and the local isostables ($A$-surfaces) ${\mathcal{A}_{c}^{i,loc}}$, $i=1,2$, as $$\begin{aligned}
\label{eq:localIsos}
{\mathcal{I}_{\theta}^{loc}}&:= \{ x \in \Omega_{loc} \quad \mid \quad \Theta(x) = \theta\}, \quad \theta \in \mathbb{T}\\
\label{eq:localIsostables}
{\mathcal{A}_{c}^{i,loc}}&:= \{ x \in \Omega_{loc} \quad \mid \quad \Sigma_i(x) = c\}, \quad c \in \mathbb{R}.\end{aligned}$$ Thus, ${\mathcal{I}_{\theta}^{loc}}$ can be computed by evaluating the function $$\begin{array}{rccc}
K(\theta,\cdot): & \mathbb{R}^2 & \rightarrow & \mathbb{R}^3 \\
& \sigma & \rightarrow & K(\theta,\sigma),
\end{array}$$ for points $\sigma \in \mathcal{B}_{loc}(\theta)$ defined in . In Section \[sec:isoComp3D\] we will discuss how to choose a grid on $\mathcal{B}_{loc}(\theta)$, where we will evaluate the function above.
Moreover, the local iPRF $\nabla \Theta^{loc}$ and the local iARFs $\nabla \Sigma_i^{loc}\enskip i=1,2$, can be computed straightforwardly using formula , for points $x = K(\theta, \sigma) \in \Omega_{loc}$.
The formula uses implicitly the fact that the functions $\nabla \Theta$ and $\nabla \Sigma_i$, $i=1,2$ are known at any order $L$. In some cases, one might be interested in the explicit expression of these functions in power expansions in $\sigma$. Next, we will show that formula provides also the terms in $\sigma$. To do so, notice that using we have $$\begin{array}{rcl}
DK(\theta, \sigma) & = &
\sum_{m=0}^{\infty} \sum_{\alpha=0}^{m}
\big [ \begin{array}{c|c|c}
K'_{\alpha, m-\alpha}(\theta) &
(\alpha+1) K_{\alpha+1, m-\alpha}(\theta) &
(m- \alpha+1) K_{\alpha, m-\alpha+1}(\theta) \\
\end{array}
\big ]
\sigma^{\alpha}_1 \sigma^{m-\alpha}_2 \\
\\
& = &
\big [ \begin{array}{c|c|c}
K'_{0}(\theta) &
K_{10}(\theta) &
K_{01} \\
\end{array} \big]
+
\big [ \begin{array}{c|c|c}
K'_{10}(\theta) &
2K_{20}(\theta) &
K_{11} (\theta)\\
\end{array} \big ]
\sigma_1
\\
\\
&& +
\big [ \begin{array}{c|c|c}
K'_{01}(\theta) &
K_{11}(\theta) &
2 K_{02} (\theta)\\
\end{array} \big ]
\sigma_2
+ \mathcal{O}_2(\sigma_1, \sigma_2),
\end{array}$$ and recall that that $K_{\alpha,m-\alpha}(\theta)$ are 3-dimensional column vectors. Let us introduce the following Taylor expansion $$\begin{bmatrix} \nabla\Theta(x) \\ \nabla\Sigma_1(x) \\ \nabla\Sigma_{2}(x) \end{bmatrix} =
\sum_{m=0}^{\infty} \sum_{\alpha=0}^{m}
\begin{bmatrix}
\nabla \Theta^{(\alpha, m-\alpha)} (\theta) \\
\nabla \Sigma_1^{(\alpha, m-\alpha)} (\theta) \\
\nabla \Sigma_{2}^{(\alpha, m-\alpha)} (\theta)
\end{bmatrix}
\sigma^{\alpha}_1 \sigma^{m-\alpha}_2,$$ where we recall that $\nabla \Theta^{(\alpha, m-\alpha)}$ and $\Sigma_i^{(\alpha, m-\alpha)}$, $i=1,2$, are 3-dimensional row vectors.
Thus, substituting the above Taylor expansions in for $d=3$ and collecting the $0$-th order terms we have that $$\label{eq:0inv}
\big [
\begin{array}{c|c|c}
K'_0 (\theta) & K_{10} (\theta) & K_{01} (\theta) \\
\end{array}
\big ]
\begin{bmatrix}
\nabla\Theta^{(0)} (\theta) \\
\nabla\Sigma_1^{(0)}(\theta) \\
\nabla\Sigma_{2}^{(0)}(\theta)
\end{bmatrix} =
Id_{3x3}.$$ From the expression above we have $$\nabla\Theta^{(0)} (\theta) = \frac{1}{det(K'_0 (\theta),K_{10} (\theta), K_{01} (\theta))} K_{10}(\theta) \times K_{01} (\theta).$$ That is, the iPRC $\nabla \Theta^{(0)}(\theta)$ is orthogonal to the attracting linear eigenspace of the limit cycle spanned by $K_{10}(\theta)$ and $K_{01} (\theta)$. Moreover, $$\nabla\Sigma_1^{(0)} (\theta) = \frac{1}{det(K'_0 (\theta),K_{10} (\theta), K_{01} (\theta))} K_{01}(\theta) \times K'_{0} (\theta),$$ and $$\nabla\Sigma_2^{(0)} (\theta) = \frac{1}{det(K'_0 (\theta),K_{10} (\theta), K_{01} (\theta))} K'_{0}(\theta) \times K_{10} (\theta).$$ Thus, using expressions for $K_{10}(\theta)$ and $K_{01}(\theta)$ and for $K'_{0}(\theta)$, we have an explicit expression for the iPRC and iARCs (see remark \[rm:prcsRM\]).
Consider now the terms of order 1 in $\sigma_1$, we have $$\big [
\begin{array}{c|c|c}
K'_0 (\theta) & K_{10} (\theta) & K_{01} (\theta) \\
\end{array}
\big ]
\begin{bmatrix}
\nabla\Theta^{(1,0)} (\theta) \\
\nabla\Sigma_1^{(1,0)}(\theta) \\
\nabla\Sigma_{2}^{(1,0)}(\theta)
\end{bmatrix}
+
\big [
\begin{array}{c|c|c}
K'_{10}(\theta) &
2K_{20}(\theta) &
K_{11} (\theta)\\
\end{array}
\big ]
\begin{bmatrix}
\nabla\Theta^{(0)} (\theta) \\
\nabla\Sigma_1^{(0)}(\theta) \\
\nabla\Sigma_{2}^{(0)}(\theta)
\end{bmatrix}
= 0,$$ and, therefore, using we have $$\begin{bmatrix}
\nabla\Theta^{(1,0)} (\theta) \\
\nabla\Sigma_1^{(1,0)}(\theta) \\
\nabla\Sigma_{2}^{(1,0)}(\theta)
\end{bmatrix}
=
-
\begin{bmatrix}
\nabla\Theta^{(0)} (\theta) \\
\nabla\Sigma_1^{(0)}(\theta) \\
\nabla\Sigma_{2}^{(0)}(\theta)
\end{bmatrix}
\big [
\begin{array}{c|c|c}
K'_{10}(\theta) &
2K_{20}(\theta) &
K_{11} (\theta)\\
\end{array}
\big ]
\begin{bmatrix}
\nabla\Theta^{(0)} (\theta) \\
\nabla\Sigma_1^{(0)}(\theta) \\
\nabla\Sigma_{2}^{(0)}(\theta)
\end{bmatrix} .$$ Equivalently, collecting terms of order 1 in $\sigma_2$, we have $$\begin{bmatrix}
\nabla\Theta^{(0,1)} (\theta) \\
\nabla\Sigma_1^{(0,1)}(\theta) \\
\nabla\Sigma_{2}^{(0,1)}(\theta)
\end{bmatrix}
=
-\begin{bmatrix}
\nabla\Theta^{(0)} (\theta) \\
\nabla\Sigma_1^{(0)}(\theta) \\
\nabla\Sigma_{2}^{(0)}(\theta)
\end{bmatrix}
\big [
\begin{array}{c|c|c}
K'_{01}(\theta) &
K_{11}(\theta) &
2 K_{02} (\theta)\\
\end{array}
\big ]
\begin{bmatrix}
\nabla\Theta^{(0)} (\theta) \\
\nabla\Sigma_1^{(0)}(\theta) \\
\nabla\Sigma_{2}^{(0)}(\theta)
\end{bmatrix}
.$$
Moreover, using the following remark we can compute the functions $\nabla \Theta$ and $\nabla \Sigma_i$, $i=1,2$ at any order $L$.
\[rem:allorders\] If $$\left ( \sum_{i=0}^{\infty} A_i(\theta) \sigma^i \right ) \left ( \sum_{i=0}^{\infty} B_i(\theta) \sigma^i \right )= Id$$ where $A_i(\theta)$ and $B_i(\theta)$ are real-valued $d \times d$ matrices, and $\sigma \in \mathbb{R}^d$, we have that, $$\begin{array}{rcl}
A_0 B_0 = Id &\Rightarrow & B_0 = A_0^{-1},\\
A_0 B_1 + A_1 B_0 =0 & \Rightarrow & B_1 = - A_0^{-1} A_1 B_0=- B_0 A_1 B_0, \\
\sum_{j=0}^i A_{i-j} B_j =0 & \Rightarrow & B_i = - A_0^{-1} \left ( \sum_{j=0}^{i-1} A_{i-j} B_j \right)= - B_0 \left ( \sum_{j=0}^{i-1} A_{i-j} B_j \right)\quad \textrm{for } i \geq 2.
\end{array}$$
The functions $\nabla \Theta^{(0)}$ and $\nabla \Sigma^{(0)}$ correspond to the iPRC and iARFs. Notice that, using Remark \[rem:allorders\], our method provides the explicit expansions of $\nabla \Theta$ and $\nabla \Sigma$ in $\sigma$ at any order $L$. Although the method is presented only for the case $d=3$, it extends straightforwardly to any dimension $d$.
Globalization of $K$, isochrons, isostables, iPRF and iARFs {#sec:isoComp3D}
-----------------------------------------------------------
In this Section we explain how to extend to a larger domain $\Omega_c \subset \Omega$, the isochrons $\mathcal{I}_\theta$, the isostables $\mathcal{A}^i_c$, the iPRF $\nabla \Theta$ and the iARFs $\nabla \Sigma$, which are known in a local domain $\Omega_{loc}$ (Section \[sec:local\]). To do so, we will use that $K(\theta,\sigma)$ is invariant by the flow $\phi_t$ of the vector field $X$ (see Eq. ). We refer to this procedure as the globalization process.
Let us start by explaining how to extend the isochron ${\mathcal{I}_{\theta}^{loc}}$. Since the flow of the vector field $X$ takes isochrons to isochrons, we can obtain several points on the isochron of phase $\theta$, ${\mathcal{I}_{\theta}}$, by integrating backwards for a time $\Delta t$ points on the isochron of phase $\theta + \Delta t/T$, i.e, $\mathcal{I}_{\theta+ \Delta t/T}$, which is known at least locally, that is, $$\label{eq:isoback}
\phi_{- \Delta t}(\mathcal{I}_{\theta+ \Delta t/T}) \subset \mathcal{I}_{\theta}.$$ Since we are integrating backwards and the backward dynamics is expanding, this procedure extends the isochron to a larger domain than $\Omega_{loc}$. In particular, we can use $\Delta t= n T, \enskip n \in \mathbb{N}$, that is, we can use points on the the same isochron to globalize it. We will present the method for the latter case, but we stress that it can be adapted to consider also points on other isochrons (see [@guillamon2009computational]).
The globalization of isochrons by means of backwards integration using the property in larger dimensional spaces ($d > 2$) presents a fundamental challenge: it involves different expanding directions, each of them associated to a particular Floquet multiplier of $\Gamma$. Therefore, by taking an homogeneous distribution of points on the local isochron ${\mathcal{I}_{\theta}^{loc}}$, most of the points will escape in the direction of the largest multiplier. To avoid this drawback, we have taken different actions:
- The use of higher order expansions for $\bar K$ (large $L$ in Eq. ), which allows us to start the computations at a relatively large distance from the limit cycle. This fact decreases the integration time to expand the isochron. Thus, our methods represent an advantage in front of other methods that consider only first order expansions.
- To choose the points in $\Omega_{loc}$ in a clever way so that one obtains a homogeneous distribution of points on the isochron $\mathcal{I}_{\theta}$. Next, we provide a detailed description of the scheme we follow, adapted from [@simo1990analytical].
Assume without loss of generality that $|\lambda_1| > |\lambda_2|$. Consider the $2$-dimensional isostable $$\label{eq:slowManifold}
\mathcal{S} := \mathcal{A}^1_0 =\{x \in \Omega \quad | \quad \Sigma_1(x) = 0 \}.$$ Notice that $\mathcal{S}$ corresponds to the *slow* stable manifold. Here, following [@cabre2003parameterization; @cabre2005parameterization], the concept of slow manifold refers to the invariant submanifold of the full stable manifold associated to the slowest direction (corresponding to the smallest in modulus Floquet exponent $\lambda_2$). These manifolds are important because while the faster directions get rapidly suppressed, the slowest directions dominate the asymptotics of convergence. In many applications one can use the slow manifold to study possible reductions of the dynamics. This will be done for a periodically forced system in Section \[sec:section5\].
We want to emphasize that the parameterization method provides straightforwardly the slow manifold. Indeed, $$\mathcal{S}= \{ x \in \Omega \quad | \quad x=K(\theta, 0, \sigma_2), \quad \theta \in \mathbb{T}, \quad \sigma_2 \in \mathbb{R}\}.$$
It will be convenient to consider the foliation of the slow manifold $\mathcal{S}$ given by $\{S^{\theta}\}_{\theta \in \mathbb{T}}$, where: $$\label{eq:sltheta}
\mathcal{S}^\theta = \{x \in \Omega \quad | \quad \Theta(x) = \theta, \quad \Sigma_1(x) = 0 \}.$$ Notice that $\mathcal{S}^{\theta} \subset {\mathcal{I}_{\theta}}$. Thus, our method to compute isochrons in $\Omega_c$ starts by computing points on the leaf of the slow manifold $\mathcal{S}^{\theta}$ in the domain $\Omega_c$, and then use it to obtain the isochron ${\mathcal{I}_{\theta}}$ (see Fig. \[fig:globalMethod3d\]). We present first an algorithm to extend the foliation of the slow manifold $\mathcal{S}$ provided by the parameterization $\bar{K}$ locally in the domain $\Omega^{loc}$ to a domain $\Omega_c$. Thus, for each $\mathcal{S}^{\theta}$, we are going to generate a sequence of points ${x_0,...,x_m}$ on it, such that they are at a distance (Euclidean norm in $\mathbb{R}^3$) smaller than some tolerance $\Delta_{max}$. The strategy is depicted in Fig. \[fig:globalMethod3d\]A.
\[alg:slowMani\] **Computation of the slow manifold leaf $\mathcal{S}^\theta$**. Given an approximate solution $\bar K$ (see ) in a local domain $\Omega_{loc}$, a constant $\Delta_{max}$ that determines the maximum distance between points on the manifold $\mathcal{S}^\theta$ and a constant $\sigma_{max}:=R_{\theta}(\pi/2)$, where $R_{\theta}$ is defined in , perform the following operations:
1. Compute the point $x_0 = K_0(\theta)$. Start the list $\{x_0\}$.
2. Set $k = 0$, $n=0$, $\sigma=0$ and $\Delta \sigma=0.8 \sigma_{max}$. Notice that here $\sigma, \Delta \sigma \in \mathbb{R}$.
3. Compute $x_{int} = K(\theta, 0, \sigma+\Delta \sigma)$ and $x_{k+1} = \phi_{-nT}(x_{int})$.
- If $ \Vert x_{k+1} - x_{k} \Vert < \Delta_{max}$ then add $x_{k+1}$ to the list $\{x_0, \ldots, x_{k}\}$ and set $\sigma \leftarrow \sigma + \Delta \sigma$ and $k \leftarrow k+1$.
- Else, divide $\Delta \sigma $ by 2.
Repeat this step until $\sigma + \Delta \sigma > \sigma_{max}$.
4. Then set $\sigma \leftarrow \sigma_{min}:=\sigma_{max} e^{\lambda_2 T}$, $\Delta \sigma=0.8 (\sigma_{max}-\sigma_{min})$, and $n \leftarrow n + 1$, and repeat from step 3 until $x_{k+1}$ is out of the computational domain $\Omega_c$.
Notice that for $n=0$, $x_{k+1}=\phi_0(x_{int})=x_{int}$ in step 3.
To extend the leaf $\mathcal{S}^\theta$ for negative values of $\sigma$ we use Algorithm \[alg:slowMani\] with $\sigma_{max}=-R_{\theta}(3\pi/2)$.
Next, we present an algorithm that extends the local isochron ${\mathcal{I}_{\theta}^{loc}}$ to $\Omega_c$ from the previously computed manifold $\mathcal{S}^{\theta}$. The strategy is the same as in the previous algorithm and is depicted in Fig. \[fig:globalMethod3d\]B.
\[alg:extendIso\] **Computation of the isochron ${\mathcal{I}_{\theta}}$**. Given $\{x_0,..,x_m\}$ points on $\mathcal{S}^\theta$ and pairs\
$\{(\sigma^ {(0)},N_0),\ldots (\sigma^ {(m)},N_m)\}$, where $(\sigma^{(k)},N_k) \in \mathbb{R} \times (\mathbb{N} \cup \{0\})$ such that $x_k=\phi_{-N_k T}(K(\theta,0,\sigma^{(k)}))$, $k=0,\ldots,m$, provided by Algorithm \[alg:slowMani\].
Consider also given an approximate solution $\bar K$ in a local domain $\Omega_{loc}$, a constant $\Delta_{max}$ that determines the maximum distance between points on the manifold $\mathcal{I}_{\theta}$ and a constant $\sigma_{max}$ (which is computed using the functions $R_{\theta}$ defined in ), perform the following operations:
1. Set $k=1$ and $q=0$.
2. Set $x_k^{(0)}=x_k$. Start the list $\{x^{(0)}_k\}$.
3. Set $\sigma=0$, $\Delta t=N_k T$ and $\Delta \sigma=0.8 \sigma_{max}$. Notice that here $\sigma, \Delta \sigma \in \mathbb{R}$.
4. Compute $x_{int} = K(\theta, \sigma+\Delta \sigma,\sigma^ {(k)})$ and $x_{k}^{(q+1)} = \phi_{- \Delta t}(x_{int})$.
- If $\Vert x^{(q+1)}_k - x^{(q)}_{k} \Vert < \Delta_{max}$ then add $x_{k}^{(q+1)}$ to the list $\{x_k^{(0)}, \ldots, x_{k}^{(q)}\}$ and set $\sigma \leftarrow \sigma + \Delta \sigma$ and $q \leftarrow q+1$.
- Else, divide $\Delta \sigma $ by 2.
Repeat this step until $\sigma + \Delta \sigma > \sigma_{max}$.
5. Then set $\sigma^{(k)} \leftarrow \sigma^{(k)} e^{\lambda_2 T}$, recompute $\sigma_{max}$ for the new $\sigma^{(k)}$ and set $N_k \leftarrow N_k + 1$, and repeat from step 3 (with $\sigma \leftarrow \sigma_{max} e^{\lambda_1 T}$) until $x_{k+1}$ is out of the computational domain $\Omega_c$.
6. Set $k \leftarrow k+1$, $q \leftarrow 0$ and repeat from step 2.
Notice that to completely extend the isochron around $\Sigma_2 = c$ one has to repeat algorithm \[alg:extendIso\] with $\Delta \sigma \leftarrow - \Delta \sigma$.
If the contraction in the direction of $\sigma_1$ is strong enough so that the points $x_k^{(q)}$ escape fast from the isostable $\Sigma_1=0$, we have that they fall outside of the computational domain $\Omega_c$ without reaching the value $\sigma_{max}$. Thus step 5 in the previous algorithm does not need to be applied (See Fig. \[fig:globalMethod3d\]B). We will see that this is the case in all the examples considered in Section \[sec:section4\],
Notice that the way we use to globalize the isochrons is to compute first the isostable $\Sigma_1=0$ and later the isostables $\Sigma_2=c$, for $c \in \mathbb{R}$ (see Fig. \[fig:globalMethod3d\]C).
### Computation of the iPRF and iARFs
The strategy used in Algorithm \[alg:extendIso\] also permits the computation of the iPRF $\nabla\Theta(x)$ and the iARFs $\nabla\Sigma(x)$ using expression and equations . Notice first that each point $x$ on the isochron ${\mathcal{I}_{\theta}}$ has been obtained either by direct evaluation of the parameterization $\bar{K}$, that is $x=\bar{K}(\theta, \sigma)$ for some $(\theta, \sigma) \in \mathbb{T} \times \mathbb{R}^2$ or by integrating backwards for a time $\Delta t$ the flow $\phi_t$ of the vector field $X$ starting at a point $x_{int} \in \Omega_{loc}$, which at its turns is obtained from the parameterization $\bar{K}$, that is, $x_{int} = \bar{K}(\theta, \sigma)$ for some $(\theta, \sigma) \in \mathbb{T} \times \mathbb{R}^2$ (see Algorithm \[alg:extendIso\]). Thus, for each computed point $x \in {\mathcal{I}_{\theta}}$, we can obtain $\nabla \Theta (x)$ and $\nabla \Sigma(x)$ in the following way:
- If $x \in \Omega_{loc}$ then $x=K(\theta^*, \sigma^*)$ for some $(\theta^*, \sigma^*) \in \mathbb{T} \times \mathbb{R}^2$. Using the local approximation $\bar K$ and the values $(\theta^*,\sigma^*)$, $\nabla \Theta (x)$ and $\nabla \Sigma(x)$ are obtained from expression .
- If $x \notin \Omega_{loc}$ then $x=\phi_{-\Delta t} (x_{int})$, where $x_{int}=K(\theta^*, \sigma^*)$ for some $(\theta^*, \sigma^*) \in \mathbb{T} \times \mathbb{R}^2$ and $\Delta t$. Using the local approximation $\bar K$ and the values $(\theta^*,\sigma^*)$, we compute $\nabla \Theta (x_{int})$ and $\nabla \Sigma(x_{int})$ using expression and then $$\nabla \Theta (x) = \Psi^{\theta}_{-\Delta t} (\nabla \Theta (x_{int})) \quad \textrm{ and } \quad \nabla \Sigma (x) = \Psi^{\sigma}_{-\Delta t} (\nabla \Sigma (x_{int})),$$ where $\Psi^{\theta}_t, \Psi^{\sigma}_t$ is the flow of .
Thus, expression provides an accurate approximation of these functions in the domain $\Omega_{loc}$, and Eqs. allow to globalize them in $\Omega_c$.
### Computation of isostables $\mathcal{A}^i_{c}$
Algorithm \[alg:slowMani\] provides a relevant isostable (the slow manifold $\mathcal{S}:= \mathcal{A}^1_0$), while Algorithm \[alg:extendIso\] provides the isostables $\mathcal{A}^2_c$, i.e, $\Sigma_2=c$, for $c\in\mathbb{R}$. To compute any other isostable $\mathcal{A}^i_{c}$ for $i=1,2$, defined in , one can adapt straightforwardly the strategy of the mentioned algorithm. Indeed, the isostable $A^i_{c}$ can be obtained from the locally computed isostable $A^i_{c^*}$ (for $|c|>|c^*|> 0$), just integrating the vector field $X$ backwards for a time $$t = \frac{1}{\lambda_i} \ln \left(\frac{c}{c^*}\right).$$
[![Schematic representation for the globalization procedure of the isochrons. (A) Strategy presented in Algorithm \[alg:slowMani\] to globalize the leaf $\mathcal{S}^\theta$ of the slow manifold $\mathcal{S}$. (B) Strategy presented in Algorithm \[alg:extendIso\] to use the points $x_q$ on $\mathcal{S}^\theta$ to extend the isochron $\mathcal{I}_\theta$ along the curve of constant $\sigma_2$, $\Sigma_2(x)=\Sigma_2(x_q)$ (C) By repeating the strategy in (B) for all the points $x_q$ in the leaf $\mathcal{S}^\theta$ of the slow manifold, one obtains the global isochron. See text for more details.[]{data-label="fig:globalMethod3d"}](./images/globalIsos_v3.pdf "fig:"){width="98mm"}]{}
Globalization methods require good numerical precision of the local approximation $\bar{K}$. For instance, in the globalization of the slow manifold, one starts at a point $x_0 = K(\theta, 0, \sigma_2^{(0)})$, but numerically, it corresponds to a value $K(\theta, \varepsilon_1, \sigma_2^{(0)})$, where $\varepsilon_1$ is a small error. Since $|\lambda_1| > |\lambda_2|$, this error may grow as we integrate backwards in time $$\phi_{\substack{-\Delta t}}(x_0) = K(\theta - \Delta t/T, \varepsilon_1 e^{-\lambda_1 \Delta t}, \sigma^{(0)}_2e^{-\lambda_2 \Delta t}),$$ and thus generate isochrons not accurate enough (it is not possible to obtain a homogeneous coverage of the space, specially for points near the manifold $\Sigma_1(x) = 0$).
Thus, a good numerical approximation provides a smaller $\varepsilon_1$ and a high order expansion for $\bar{K}$ provides local points that are far from the invariant object $\Gamma$, thus reducing the integration time $\Delta t$ and therefore the error. Moreover, in the algorithms presented, the time $\Delta t$ is always a multiple of the period $T$, since we globalize the isochron by integrating backwards points on the same isochron. However, one can adapt the algorithms to globalize the isochrons using points on a different isochron (see Eq. ) and therefore integrating for a shorter time $\Delta t$ (see [@guillamon2009computational]).
Numerical Examples {#sec:section4}
==================
We have carried out the numerical implementation of the methodology presented in Section \[sec:section3\] and applied it to some representative single neuron and neuronal population models showing oscillatory dynamics. In this Section we present the results obtained and discuss the relevant aspects.
Models
------
In this Section, we present the four models that we have used to apply the techniques introduced in Section \[sec:section3\]. There are two models for spiking dynamics of single neurons, including a version of the classical Hodgkin-Huxley model, and two models for mean firing rate dynamics of neuronal populations, including a 3D version of the classical Wilson-Cowan model. The examples are chosen either because they are relevant to illustrate the properties of the numerical methodology or because they are representative of classical models. The parameter values and the functions for each model can be found in Appendix \[sec:appendix\].
- A single thalamic neuron model introduced in [@rubin2004high], that we refer to as $RT$, with sodium, potassium and low-threshold calcium currents: $$\begin{aligned}\label{eq:rtEDOs}
C_m \dot{V} &= - I_L(V) - I_{Na}(V, h) - I_{K}(V, h) - I_{T}(V, r) + I_{app}, \\
\dot{h} &= \frac{h_{\infty}(V) - h}{\tau_h(V)}, \\
\dot{r} &= \frac{r_{\infty}(V) - r}{\tau_r(V)},
\end{aligned}$$ where $V$ describes the membrane potential and $h$ and $r$ are the gating variables.
This model is interesting because it has a slow-fast dynamics (notice that as $\tau_r(V) \gg \tau_h(V) \gg 1$, the variable $r$ is much slower than $V$ and $m$) and it allows us to illustrate how the slow-fast dynamics affects the geometric objects (isochrons and isostables). Moreover, as we will see in Section \[sec:section5\], we can take advantage of these dynamics to explore the amplitude-phase description as an alternative to the phase reduction. This model was also studied in [@wilson2018greater].
- A reduced Hodgkin-Huxley-like system that we refer to as $HH$ [@izhikevich2007], with sodium and potassium currents, and two gating variables: $$\begin{aligned}\label{eq:hhEDOs}
C_m \dot{V} &= - I_L(V) - I_{Na}(V, h) - I_{K}(V, n) + I_{app}, \\
\dot{n} &= \frac{n_{\infty}(V) - n}{\tau_n(V)}, \\
\dot{h} &= \frac{h_{\infty}(V) - h}{\tau_h(V)},
\end{aligned}$$ where $V$ describes the membrane potential and $n$ and $h$ are the gating variables.
This model is a 3D reduction of the classical 4D Hodgkin-Huxley model (just setting the fast variable $m$ to its steady-state value and slightly modifying some parameters). It also has a slow-fast nature but less noticed than in the $RT$ model.
- An extension of the Wilson-Cowan equations [@wilson1972excitatory] including dynamics for the inhibitory synapsis that we refer to as $WC_{Syn}$: $$\begin{aligned}\label{eq:wcEDOs3D}
\tau_e \dot{E} &= -E + \delta_E(c_1 E - c_2 s + P), \\
\tau_i \dot{I} &= -I + \delta_I(c_3 E - c_4 s + Q), \\
\tau_d \dot{s} &= -s + \tau_d I,
\end{aligned}$$ where $E$ and $I$ are the mean firing rates of excitatory and inhibitory populations, respectively, whereas $s$ describes the inhibitory synaptic dynamics.
- A model for the mean field activity of a population of heterogeneous quadratic integrate-and-fire neurons [@devalle2017firing] that we refer to as $QIF$: $$\begin{aligned}\label{eq:qfEDOs}
\tau_m \dot{V} &= V^2 - (\pi \tau_m R)^2 - J\tau_mS + \Theta, \\
\tau_m \dot{R} &= -\frac{\Delta}{\pi \tau_m} + 2RV,\\
\tau_d \dot{S} &= -S + R,
\end{aligned}$$ where $V$ is the mean membrane potential, $R$ is the mean firing rate of the population and $S$ is the synaptic activation.
We have chosen this model because it is a representative example of a new generation of neural field models [@montbrio2015macroscopic; @coombes2019next].
Numerical results
-----------------
In the four examples considered, there exists a hyperbolic attracting limit cycle $\Gamma$. Moreover, there exists an unstable equilibrium point of saddle-focus type having a two-dimensional unstable manifold and a one-dimensional stable manifold. We set the zero phase value at the maximum of the voltage variable $V$ (in the $WC_{Syn}$ model it will be the maximum of the $E$ variable). For each model, the limit cycle $\Gamma$ has two characteristic exponents, $\lambda_1$ and $\lambda_2$, with $|\lambda_1| > |\lambda_2|$. The numerical integration has been performed using an 8th-order Runge-Kutta Fehlberg method (rk78) with a tolerance of $10^{-14}$. In the neighbourhood of $\Gamma$ we have computed a Taylor expansion up to order $L$ of the parameterization $K$ as in and we have computed $N+1$ Fourier coefficients for the periodic functions $K_{\alpha,m-\alpha}$, $\alpha=0,\ldots,m$ and $m=0,\ldots,L$. Notice that, as in all the examples the dimension $d=3$, to obtain expansions up to an order $L$ we use $\sum_{n=1}^{L+1} n=(L+1)(L+2)/2$ monomials. Recall that the number of Fourier coefficients chosen was such that the residuals are smaller than $E_{tail}$ (see Eq. ). The domain of the local approximation $\Omega_{loc}$ is computed with an error smaller than $E_{loc}$ (see Eq. ). Table \[table:modelPrms\] contains the values of the parameters described above for each model.\
Model $T$ $\lambda_1$ $\lambda_2$ $L $ $N $ $E_{tail}$ $E_{loc}$ Equilibrium Point
------------ ------- ------------- ------------- ------ ------ ------------ ----------- ------------------------------
$RT$ 8.395 $-0.368$ $-0.022$ 10 2048 $10^{-10}$ $10^{-8}$ (-39.1, 0.38, 1.3 $10^{-5}$)
$HH$ 7.586 $-1.73$ $-0.2$ 10 2048 $10^{-10}$ $10^{-6}$ (-49.1, 0.564, 0.137)
$WC_{Syn}$ 24.43 $-0.445$ $-0.246$ 10 2048 $10^{-10}$ $10^{-8}$ (0.272, 0.033, 0.198)
$QIF$ 27.58 $-0.408$ $-0.06$ 10 2048 $10^{-10}$ $10^{-8}$ (0.018, -0.267, 0.018)
: Numerical values for the different models considered; $T$ period of the periodic orbit $\Gamma$; $\lambda_1$, $\lambda_2$ characteristic exponents associated to $\Gamma$; $L$ order of the Taylor expansion; $N+1$ number of Fourier coefficients; tolerance $E_{tail}$ defined in ; maximal error $E_{loc}$ that establishes the domain of accuracy of the local approximation and coordinates of the unstable equilibrium point of saddle-focus type.[]{data-label="table:modelPrms"}
For each model, we have computed the approximate parameterization $\bar{K}$ , the slow manifold $\mathcal{S}$ (which is the most relevant isostable) and the isochrons $\mathcal{I}_{\theta}$ . In Figs. \[fig:rtPanel\] to \[fig:qfPanel\], we illustrate some aspects of the elements computed, but we stress that our computations have more data than the ones shown. More precisely, each figure corresponds to a different model and has 5 panels that contain the same information for each model. Panel A shows the three components of the parameterization of the limit cycle $\gamma(\theta)$ defined in . Panel B shows the first component of the periodic functions $K_{i,0}$ and $K_{0,i}$ for $i=1,2,5,10$, corresponding to different coefficients of the Taylor expansion of $\bar K$ up to order 10. As the magnitude of the different functions $K_{i,j}$ is very different and depends on the constants $b_{1}$ and $b_{2}$ (see Section \[sec:section32\]), for illustration purposes we have normalized these functions so that the maximum is one (see Table \[table:ksPrms\] for more details). Panel C shows the domain of accuracy $\Omega_{loc}$ of the local approximation $\bar K$ for some values of $\theta$ or, equivalently, the local isochrons ${\mathcal{I}_{\theta}^{loc}}$. We have computed $N=2048$ isochrons, that is $\mathcal{I}^{loc}_{\theta_i}$ for $\theta_i=i/2048$ for $i=0,\ldots,2047$. We have not chosen equidistant values of $\theta$ to plot the isochrons. Alternatively, the chosen values of $\theta_k$ have been adapted to each model, so that the isochrons are uniformly distributed in the phase space along the limit cycle to allow the reader to distinguish the isochrons in the plots. Panel D shows the globalized slow manifold $\mathcal{S}$ and 64 leaves $\mathcal{S}^{\theta}$ , for equidistant values of $\theta$, more precisely for $\theta_i=i 32/2048$, for $i=0,..63$. Finally, panel E shows the globalization of some of the isochrons ${\mathcal{I}_{\theta}}$ computed in panel C from two different perspectives. The isochron computation has been restricted to a domain $\Omega_c$, based on values that are biophysically plausible. Namely, for the $RT$ model, the gating variables $h$ and $r$ are allowed to vary between 0 and 1. The same criterion is applied to the gating variables $n$ and $h$ in $HH$ model, and $V$ was restricted to $V<60$. For the $WC_{Syn}$ model, the three variables $E, I$ and $s$ are restricted between 0 and 1. Finally, for the $QIF$ model, $R$ and $S$ variables can not be negative, and $-6 < V < 6$.
Moreover, we have also computed for each model the iPRF $\nabla \Theta$ and the iARFs $\nabla \Sigma_i, i=1,2$ (Figs. \[fig:rtPanelPrf\] to \[fig:qfPanelPrf\]). In each figure on the right, we show the first component of these vector-valued functions that, abusing language, we denote by the same letters (see Remark \[rem:sobreprf\]). On the left, we plot the restriction of these functions to the limit cycle, the iPRC and iARCs (see Remark \[rm:prcsRM\]). Since the computation of the functions $\nabla \Theta(x)$, $\nabla \Sigma_1(x)$ and $\nabla \Sigma_2(x)$ is done in parallel with the computation of points on the isochrons (see Section \[sec:isoComp3D\]), we have evaluated these functions for those computed points on the isochrons. For illustration purposes, we plot these functions for two or three isochrons indicated in each caption. Moreover, in the figures we just plot a part of the isochron that permits a proper visualization of the functions $\nabla \Theta$ and $\nabla \Sigma_i$, $i=1,2$, even though our computations are done for the whole globalized isochrons.
Model $b_{1}$ $K^{max}_{1,0}$ $K^{max}_{2,0}$ $K^{max}_{5,0}$ $K^{max}_{10,0}$ $b_{2}$ $K^{max}_{0,1}$ $K^{max}_{0,2}$ $K^{max}_{0,5}$ $K^{max}_{0,10}$
------------ --------- ----------------- ----------------- ------------------- -------------------- --------- ----------------- ----------------- ------------------ ------------------
$RT$ 0.5 1.2 0.05 $7\cdot10^{-6}$ $1.5\cdot10^{-11}$ 0.5 51.2 59.5 $1.2\cdot10^{3}$ $1.2\cdot10^{5}$
$HH$ 2 7.8 1.2 $1.3\cdot10^{-2}$ $1.2\cdot10^{-5}$ 2 23.7 9.1 3.3 3.2
$WC_{Syn}$ 1 3.4 7.8 $3\cdot10^{3}$ $4\cdot10^{8}$ 1 0.54 3.64 $1.2\cdot10^{3}$ $1\cdot10^{8}$
$QIF$ 0.2 4.58 5.9 8.6 18.1 1 2.3 2.4 3.05 4.38
: Numerical values for the constants $b_1$, $b_2$ and $K^{max}_{i,j}=\max_{\theta \in \mathbb{T}} |\bar{K}^X_{i,j}(\theta)|$, where we denote by $\bar{K}^X_{i,j}(\theta)$ the first component of the function $\bar{K}_{i,j}(\theta)$.[]{data-label="table:ksPrms"}
As we observed in Remark \[rm:remark32\], the choice of constants $b_{1}$, $b_{2} \in \mathbb{R}$ affects the magnitude of the monomials $K_{\alpha, m-\alpha}$ and therefore the domain of accuracy of $\bar K$. Indeed, depending on the choice of $b_{1}$, $b_{2}$, successive orders of $K_{\alpha, m-\alpha}$ can either blow up or vanish, since the monomial $K_{\alpha, m-\alpha}$ will be multiplied by a factor $b_{1}^\alpha b_{2}^{m-\alpha}$. In Table \[table:ksPrms\] we provide the values of the constants $b_{1}$ and $b_{2}$ that we have used for each model as well as the maximum values of some representative monomials. The criteria for the choice has been to keep the monomials at order one (as much as possible), thus maximizing the domain of accuracy of the local approximation $\bar K$.
For all models we have computed the Taylor expansion up to degree $L=10$. For the $QIF$ model the monomials can be kept of order 1 up to degree 10, but for the $HH$ and $RT$ models some monomials start to show a decay to zero at degree 5. This fact reflects then in the size of the local domain $\Omega_{loc}$ and, in consequence, the size of the local isochrons ${\mathcal{I}_{\theta}^{loc}}$. The local isochrons are smaller for the $RT$ and $HH$ models (compare panel C in for the $RT$ and $HH$ model with panel C for the $QIF$ model). Moreover, the isochrons for the $HH$ model look elongated (panel C in Fig. \[fig:hhPanel\]). Indeed, in the direction of $\sigma_1$, the coefficients $\bar{K}_{ij}$ decay to 0, while in the direction of $\sigma_2$ they remain of order 1.
Besides the purely numerical considerations regarding the coefficients $\bar{K}_{ij}$, we want to highlight that the linear terms $\bar{K}_{10}$ and $\bar{K}_{01}$ (black curves in panel B of Figs \[fig:rtPanel\]-\[fig:qfPanel\]) provide information about the attraction to the limit cycle. More precisely, when the attraction to the limit cycle is not homogeneous along the cycle, the functions $\bar{K}_{10}$ and $\bar{K}_{01}$ show dramatic changes along a period, as it happens for instance for the $RT$ and $HH$ models (see Figs. \[fig:rtPanel\]B and \[fig:hhPanel\]B). Then, this has also consequences in the size and shape of the local isochrons. Take, for instance, the values of the coefficients $\bar{K}_{ij}$ for the first component $\bar{K}^X_{ij}$ shown in panels B. The phases $\theta$ at which $\bar{K}^X_{ij}$ are close to zero (approximately $0.4-0.8$ in Fig. \[fig:rtPanel\]B and $0.2-0.8$ in Fig. \[fig:hhPanel\]B), the corresponding local isochron ${\mathcal{I}_{\theta}^{loc}}$ does not extend along the first component $V$ and it is almost parallel to planes with constant $V$ (see the isochrons in blue colours Fig. \[fig:rtPanel\]C and in blue/green colours in Fig. \[fig:hhPanel\]C). For the same reason, if we look at the 3 components of the vector-valued functions $\bar{K}_{ij}$, for those values of $\theta$ for which the 3 components are close to zero, the local isochrons will be small. Thus, local isochrons ${\mathcal{I}_{\theta}^{loc}}$ show great size and shape differences for different values of $\theta$, whenever the range of the functions $\bar{K}_{ij}$ is wide. However, whenever $\bar{K}_{ij}$ are more uniform along a period the size of the local isochrons is also more homogeneous for the different phases (see for instance Fig.\[fig:wcPanel\]C and their respective values of $\bar{K}_{ij}$ in Fig.\[fig:wcPanel\]B).
The slow manifold $\mathcal{S}$ in panel D illustrates the different geometries underlying the approximation to the limit cycle $\Gamma$. For instance, when the system is slow in one variable as in the $RT$ model, the slow manifold has a cylinder-like shape in the direction of the slow variable $r$ (see Fig. \[fig:rtPanel\]D and Fig. \[fig:hhPanel\]D for negative values of the voltage $V$). The $QIF$ model has also a slow-fast dynamics (see the characteristic exponents $\lambda_1$ and $\lambda_2$ in Table \[table:modelPrms\]), but since the slow manifold is not in the direction of any coordinate axis, its cylindrical shape should became visible after a linear change of variables (Fig. \[fig:qfPanel\]D). Even if for completeness we also show the slow manifold for the $WC_{syn}$ model in Fig. \[fig:wcPanel\], we stress that this manifold has not a special significance in this case since the characteristic exponents $\lambda_1$ and $\lambda_2$ are of the same size (see Table \[table:modelPrms\]).
Furthermore, the slow-fast dynamics is also reflected in the isochron distribution. In general, the isochrons are not homogeneously distributed in phase space for equidistant values of $\theta$. Indeed, they accumulate in the regions where the dynamics is slow, and they appear separated when the dynamics is fast. Notice that in Figs. \[fig:rtPanel\]C, \[fig:hhPanel\]C and \[fig:qfPanel\]C, one can see that isochrons for phases between 0 and 0.2 get distributed in approximately half of the limit cycle (warm coloured isochrons), while to cover the rest of the limit cycle we need the isochrons of phases between 0.2 and 1 (cool coloured isochrons). Consequently, if we consider equidistant values of $\theta$, we would observe that the isochrons accumulate in a part of the limit cycle. This can be appreciated in Panel D of Figs. \[fig:rtPanel\], \[fig:hhPanel\] and \[fig:qfPanel\], where the leaves $\mathcal{S}_{\theta}$, corresponding to the projection of the isochrons ${\mathcal{I}_{\theta}^{loc}}$ onto the slow manifold, are equidistant in $\theta$. This phenomenon is different in Fig. \[fig:wcPanel\]C, where the isochrons are better distributed for equidistant values of $\theta$ (see also Fig. \[fig:wcPanel\]D). This is a consequence of the non slow-fast nature of the model $WC_{syn}$.
The geometry of the isochrons is linked to the iPRF and iARFs shown in Figs. \[fig:rtPanelPrf\]-\[fig:qfPanelPrf\]. For instance, when the isochrons get very close to each other, $\nabla \Theta$ increases dramatically because small perturbations can cause the trajectories to jump to an isochron with a very different phase. Indeed, in a neighbourhood of the unstable fixed point (a phaseless set for these models) the isochrons pack (see Figs. \[fig:rtPanel\]-\[fig:qfPanel\]E) and the amplitude variable $\sigma$ tends to infinity, thus a perturbation acting on points therein generates large shifts both in phase and in amplitude (results not shown). Alongside, for systems with isochrons that are parallel to the coordinate plane containing the $V$-axis (as it happens in the $HH$ model, see Fig. \[fig:hhPanel\]E), the changes in phase due to perturbations in the direction of $V$ are going to be zero. In Figs.\[fig:rtPanelPrf\]-\[fig:qfPanelPrf\]A (right) we plot $\nabla \Theta$ along the isochron. Clearly, we see that $\nabla \Theta$ is close to 0 in Fig. \[fig:hhPanelPrf\]A (right) which corresponds to Fig. \[fig:hhPanel\]E. As we mentioned, when can also appreciate the increment of $\nabla \Theta$ on the isochrons when they accumulate near the phaseless set but, as explained before, we just show a part of the isochron before $\nabla \Theta$ blows up. Notice though that the effect of perturbations acting on points lying on the same isochron can be very different depending whether the point is close or far from the limit cycle.
In Figs. \[fig:rtPanelPrf\]-\[fig:qfPanelPrf\] the magnitude of $\nabla \Sigma$ needs to be put in context. Indeed, since the amplitude variables $\sigma_1$ and $\sigma_2$ are scaled by the factors $b_1$ and $b_2$ one needs to control how the size of $\sigma$ translates to Euclidean distance to the limit cycle. In Section \[sec:section5\] we discuss this issue in more detail for a concrete example.
Study of Perturbations using the Phase-Amplitude Variables {#sec:section5}
==========================================================
The phase reduction is a powerful tool to reduce high dimensional dynamics to a single equation and understand the dynamics that emerges when nonlinear oscillators are weakly perturbed. It has been extensively used to study weakly coupled oscillators and synchronization properties [@hoppensteadt2012], specially in the neuroscience context [@ErmentroutTerman2010], but it has several limitations. Indeed, the phase reduction assumes that the dynamics tkes place on the limit cycle or very close to it. However, if the attraction to the limit cycle is weak or the time between perturbations is short compared to the strength of the perturbation, this strong assumption is not valid anymore and the phase reduction becomes unreliable.
In this Section we show that the phase-amplitude approach based on the parameterization method can account for an accurate description of the phase variation away from the limit cycle $\Gamma$. The phase-amplitude description does not reduce the dimension, but allows us to identify the transversal directions that are relevant for the dynamics and suggests possible dynamical reductions.
In general, we consider perturbations of our original system $X$ in such that the perturbed system is of the form $$\begin{aligned}
\label{eq:sispert}
\dot{x} = X_A(x, t) = X(x) + A p(t),\end{aligned}$$ where $p(t)$ is a vector-valued function on $\mathbb{R}^d$ depending on time. The dynamics of the perturbed system in phase-amplitude variables $(\theta,\sigma)=(\theta,\sigma_1,\ldots,\sigma_{d-1}) \in \mathbb{T} \times \mathbb{R}^{d-1}$ is given by $$\label{eq:systa}
\begin{array}{rcl}
\dot{\theta} &= &\frac{1}{T} + A \nabla \Theta (K(\theta,\sigma))\cdot p(t), \\
\dot{\sigma}_i &=& \lambda_i \sigma_i + A \nabla \Sigma_i (K(\theta,\sigma))\cdot p(t), \quad i=1,\ldots,d-1,
\end{array}$$ where $\nabla \Theta (K(\theta,\sigma))$ and $\nabla \Sigma_i (K(\theta,\sigma))$, for $i=1,...,d-1$, are the iPRF and iARFs, respectively, defined in .
We want to stress that the distance to the limit cycle is not a limitation for our method. Indeed, $\nabla \Theta$ and $\nabla \Sigma$ are known for any point in $\Omega$. Recall that they are known locally as Taylor expansions around the limit cycle up to high order (order $L$) and can be globalized using equations (see Section \[sec:phaseAmpFun\]). Thus, the system above is exact.
For our study, we will use a perturbation consisting of a train of $n$ pulses of size $\varepsilon$ separated by a time $T_s$. Each train of pulses is repeated periodically with period $T_p$ (see Fig.\[fig:perturbation3D\] left). Mathematically, $$\label{eq:stimulus}
p(t)=\varepsilon \vec{v} \sum_{k=0}^{n} \delta(t-kT_s), \quad t \in [0, T_{total}),$$ where $T_{total}=nT_s+T_p$ is the period and $\vec{v}$ is a $d$-dimensional vector representing the direction of the pulsatile perturbation.
Notice that although this perturbation consists of small kicks (size $\varepsilon$), it manages to displace the trajectories away from the limit cycle $\Gamma$ if the interpulse interval $T_s$ is short and $n$ is large (see Fig.\[fig:perturbation3D\] right). Indeed, when $T_s \rightarrow 0$, the perturbation becomes close to a delta perturbation of amplitude $n\cdot\epsilon$.
Since the perturbation is periodic, the dynamics can be described by the stroboscopic map, given by the flow of at time $T_{total}$ starting at $t=0$, which has the following expression $$\label{eq:gemmaMap3D}
F(x):=\phi_{\substack{T_p}} \circ \underbrace{f \circ \cdots \circ f}_{n} (x),$$ where $f(x):=\phi_{\substack{T_s}}(x+\varepsilon \vec v)$ and $\phi_t$ is the flow of the unperturbed vector field $X$. Notice that, knowing the position of the trajectory at a given kick $n$, the map $f$ provides the position of the trajectory at the time immediately preceding the next kick $n+1$.
Since the perturbation is pulsatile, the map $f$ can be obtained analytically in terms of the phase-amplitude variables and has the following expression: $$\label{eq:fullMap3Dnoaprox}
\begin{aligned}
\bar \theta & = \Theta(K(\theta, \sigma) + \varepsilon \vec{v}) + \frac{T_s}{T}, \quad \quad \text{(mod 1)} \quad k \in \mathbb{N} \\
\bar \sigma_i & = \left ( \Sigma_i(K(\theta, \sigma)+\varepsilon \vec{v}) \right )e^{\lambda_i T_s}, \quad \quad i = 1,\ldots, d-1.
\end{aligned}$$
Using that the pulses are of size $|\varepsilon| \ll 1$, the map above can be approximated using the iPRF and iARFs and has the following approximated expression $\bar f$: $$\label{eq:fullMap3D}
\begin{aligned}
\bar \theta & = \theta + \varepsilon \nabla\Theta(K(\theta, \sigma))\cdot \vec{v} + \frac{T_s}{T}, \quad \quad \text{(mod 1)} \quad k \in \mathbb{N} \\
\bar \sigma_i & = \big(\sigma_i + \varepsilon \nabla\Sigma_i(K(\theta, \sigma))\cdot \vec{v}\big)e^{\lambda_i T_s}, \quad \quad i = 1,\ldots, d-1.
\end{aligned}$$
Moreover, recall that the map $\phi_{T_p}$ is described exactly in phase-amplitude variables by the expression . Thus, we have a semi-analytical expression for the map $F$ in phase-amplitude variables.
Expression for the map $f$ is not exact. Indeed, it uses that if $\varepsilon$ is small then the iPRF $\nabla \Theta$ and the iARFs $\nabla \Sigma_i$ provide a good approximation in first order of the PRF and ARFs, respectively (see Eq. ). For higher order approximations of the PRC see [@suvak2010quadratic; @takeshita2010higher].
As we already mentioned, the system with one phase and $d-1$ amplitude variables is not reduced, as the number of variables is the same as in the original system. However, we will use it to show that our computation in the new coordinates can capture the dynamics with high accuracy.
Dynamical Reductions
--------------------
In some cases, the particularities of the system or the perturbation allow for a reduction of the full phase-amplitude system. The *phase reduction* is the most extended reduction. It assumes that perturbations are weak enough so that the trajectories are not displaced far away from the limit cycle. The study of the dynamics is then reduced to control the phase on the limit cycle. Thus, the transversal directions $\sigma_i$ are assumed to be 0 and the map $\bar f|_{\Gamma}$ is just a stroboscopic map of a circle to itself of the form, $$\label{eq:phaseMap}
\bar \theta = \theta + \varepsilon \nabla\Theta(K(\theta, 0))\cdot \vec{v} + \frac{T_s}{T}, \quad \quad \text{(mod 1)},$$ where $\nabla\Theta(K(\theta, 0))$ is the classical iPRC (see Remark \[rm:prcsRM\]).
Nevertheless, when studying large perturbations that displace the trajectories far from the limit cycle there exist more suitable reductions that can account for the dynamics. Since in high dimensional systems ($d > 2$), there exist more than one characteristic exponents, it is often the case that some of them are very small, as it happens close to a bifurcation. Recall that the amplitude variables $\sigma_i$ decay to zero at a rate that depends on its associated Floquet exponent $\lambda_i$. Therefore, those associated to Floquet exponents with large moduli can be assumed to be zero and its equation removed from , thus reducing the dimension of the system. In the particular case that all variables $\sigma_i$ are assumed to be 0, except the one corresponding to the smallest (in modulus) Floquet exponent that we assume to be $\lambda_{d-1}$ without loss of generality, we say that the dynamics is reduced to the slow manifold $\mathcal{S}$ defined in . In this case, the map $\bar f|_{\mathcal{S}}$ writes as: $$\label{eq:variables_prc21}
\begin{aligned}
\bar \theta & = \theta + \varepsilon \nabla\Theta(K(\theta, 0, \ldots, 0,\sigma_{d-1}))\cdot \vec{v} + \frac{T_s}{T}, \quad \quad \text{(mod 1)} \\
\bar \sigma_{d-1} & = \big(\sigma_{d-1} + \varepsilon \nabla\Sigma_{d-1}(K(\theta, 0, \ldots, 0, \sigma_{d-1}))\cdot \vec{v}\big)e^{\lambda_{d-1} T_s}.
\end{aligned}$$ The dynamics, then, reduces to two equations and, furthermore, the number of monomials in $\bar K$ drastically reduces. We refer to this reduction as the *slow manifold reduction*.
Examples
--------
In this Section we illustrate our methodology and compare the different dynamical reductions discussed in the previous Section. To do so, we choose the $RT$ model in , as in [@wilson2018greater]. Recall that it is a $3$-dimensional model with an attracting limit cycle that has two associated Floquet exponents: $\lambda_2=-0.022$ and $\lambda_1=-0.368$, which is approximately 16 times greater (in modulus) than $\lambda_2$ (see Table \[table:modelPrms\]). Therefore, it is reasonable to explore also the slow manifold reduction for this system.
We apply the perturbation (see Fig. \[fig:perturbation3D\]) with $n=100$ pulses of amplitude $\varepsilon=-0.1$, separated by a time interval $T_s=0.001$. The interval between input trains will be set to $T_p=8.394 \approx T$, where $T$ is the period of the unperturbed limit cycle. The perturbation will be in the voltage direction, that is, $\vec{v}=(1,0,0)$. Notice that the sign of $\varepsilon$ is negative, therefore the perturbation is inhibitory. We apply inhibitory perturbations because the changes in $\sigma_2$ are greater for inhibitory perturbations than excitatory, therefore they displace the trajectory further away from the limit cycle as observed in [@wilson2018greater] (results not shown).
We compute the iterates using four different stroboscopic maps. The first one is the 3-dimensional map $F$ in obtained by direct integration of the vector field in the original coordinates $(V,h,n) \in \mathbb{R}^3$. We refer to this map as the *state variables map*. This map provides the exact description of the dynamics. The second one is an approximation of the map $F$ given in the phase-amplitude coordinates $(\theta,\sigma) \in \mathbb{T} \times \mathbb{R}^2$. We use expression of $\bar f$ in to approximate $f$ and Eq. for $\phi_{T_p}$. We denote it by $\bar{F}$ and refer to it as the *Phase-Amplitude map*. Notice that this map is still 3-dimensional. The other two maps are dimensional reductions of the [Phase-Amplitude map]{}. One is obtained by setting the variable $\sigma_1=0$ (see Eq. ), and it is 2-dimensional. We refer to this one as the *slow manifold reduction* and denote it by $\bar{F}|_{\mathcal{S}}$. The other one is the classical 1-dimensional *phase map* $\bar{F}|_{\Gamma}$ and is obtained by setting all amplitude variables to zero (see Eq. ) and keeping only the phase variable.
Next we discuss the performance of the above mentioned maps. For all of them we compute $N=80$ iterates of the map starting on the point with 0-phase on the limit cycle. Notice that this point has $\sigma_1=\sigma_2=0$ amplitudes, therefore, it is well-defined for all maps and reductions. We observe that, for all maps, the iterates tend to a fixed point, but it is different for each map. Fig. \[fig:panellPertRT\] shows the phase coordinate for 80 iterates of the maps $\bar F$, $\bar F|_{\mathcal{S}}$ and $\bar F|_{\Gamma}$ (right) and the evolution of the voltage variable along the corresponding trajectory with the values of the voltage at the last 10 iterates of the maps $F$, $\bar F$, $\bar F|_{\mathcal{S}}$ and $\bar F|_{\Gamma}$ (left). In Table \[table:puntFix\] we present the coordinates of the fixed point obtained for the different maps. We provide the fixed points in state variables (obtained evaluating the parameterization $K$) and phase-amplitude variables, when applicable. Notice that only the phase-amplitude variables provide an accurate description of the dynamics of the system. Nevertheless, the slow manifold reduction can capture the value of the phase variable.
By looking at the $(\theta,\sigma)$ coordinates of the fixed points it might seem that the assumption that some amplitude variable is zero is very strong. However, we want to recall that the amplitude variables $\sigma_1$ and $\sigma_2$ are scaled by the constants $b_1$ and $b_2$, thus, their absolute values do not provide enough information on the distance to the limit cycle. Indeed, in the computation performed for the $RT$ model, we can see that small variations in $\sigma_1$ translate to small variations in the distance to the limit cycle (see Fig. \[fig:proves\] left). A different behaviour occurs for $\sigma_2$, where small variations translate to large variations in the distance to the limit cycle (see Fig. \[fig:proves\] right). Thus, the assumption that $\sigma_1=0$ in the slow manifold reduction is less strong that the assumption that $\sigma_2=0$ in the phase reduction (see Table \[table:puntFix\] and Fig \[fig:proves\]).
![For the *RT* model we show the Euclidean distance to the limit cycle as a function of the value of the amplitude coordinate $\sigma_1$ (left) and $\sigma_2$ (right). More precisely, for $\theta=0.283$, we show $\Vert K(\theta,\sigma_1,0)-K(\theta,0,0)\Vert$ (left) and $\Vert K(\theta,0,\sigma_2)-K(\theta,0,0)\Vert$ (right) []{data-label="fig:proves"}](./images/panellDists.pdf){width="150mm"}
Results in Table \[table:puntFix\] can be interpreted by means of Fig. \[fig:panellSlowMan\], which shows the limit cycle $\Gamma$, the slow manifold $\mathcal{S}$ and the trajectory for a time $T_{total}$ of the perturbed system starting at the fixed point $P$ of the map $F$. We also show the point $P_{pert}$, which provides the position of the trajectory after the perturbation has been applied, that is $P_{pert}=f^n(P)$. Observe that the trajectory $\phi_{t}(P)$ for $t \in [0,T_{total})$ is not on the limit cycle $\Gamma$, which explains the poor description of the dynamics when using the phase map . Notice that the perturbation displaces the trajectory away from the slow manifold (see position of $P_{pert}$), but the trajectory relaxes back to the slow manifold $\mathcal{S}$ when the perturbation is removed (see position of point $P$). In this example, the phase-amplitude map is the most suitable to describe properly the dynamics. We would like to highlight the good performance of the phase-amplitude map (see Table 3). However, if we consider a larger value of $T_s$ so that the trajectories relax back to the slow manifold but not to the limit cycle after every kick, we expect that the slow manifold map will provide as well an accurate description of the dynamics. We stress that for the phase map to provide an accurate description of the dynamics, this time should be much longer.
Map Symbol $V$ $h$ $r$ $\theta$ $\sigma_1$ $\sigma_2$
------------------------- ----------------------------------------------------------- ----------- --------- ----------- ---------- ------------ ------------
State Variables $P = F(P)$ $-57.16$ $0.135$ $0.00383$ —- —- —-
Phase-Amplitude $P_{\Omega} = \bar{F}(P_{\Omega})$ $-57.13$ $0.135$ $0.00377$ $0.283$ $-2.37$ $4.98$
Slow Manifold Reduction $P_{\mathcal{S}} = \bar{F}|_\mathcal{S}(P_{\mathcal{S}})$ $-61.81$ $0.197$ $0.00314$ $0.269$ $0$ $3.439$
Phase Reduction $P_{\Gamma} = \bar{F}|_\Gamma(P_{\Gamma})$ $-60.458$ $0.175$ $0.0017$ $0.15$ $0$ $0$
: Values of the $V$, $h$ and $r$ coordinates of the fixed point $P$ of the stroboscopic map $F$ in the state variables, the fixed point $P_{\Omega}$ of the full phase-amplitude map $\bar{F}$, the fixed point $P_{\mathcal{S}}$ of the slow manifold reduction $ \bar{F}|_\mathcal{S}$ and the fixed point $P_{\Gamma}$ of the phase reduction map $ \bar{F}|_\Gamma$. We also show the values for the coordinates $\theta$, $\sigma_1$ and $\sigma_2$ of the corresponding fixed point.[]{data-label="table:puntFix"}
Conclusions and discussion {#sec:section6}
==========================
In this paper we extend the applications of the parameterization method to study the phase-amplitude description for $d$-dimensional non-linear oscillators with $d \geq 2$. More precisely, we consider systems having a hyperbolic attracting limit cycle $\Gamma$ and we provide a computational method to obtain a parameterization $K$ of its attracting invariant manifold (which in fact is its whole basin of attraction) in terms of the phase-amplitude variables. The use of this method has several advantages. On one hand, it automatically provides a geometrical portrait of the oscillator through the computation of the isochrons and the isostables, including the slow submanifold. On the other hand, it provides the iPRF and iARFs, which allow us to accurately track the effects of a given perturbation $p(t)$ beyond the phase reduction.
The theoretical and numerical methods presented herein extend previous results for planar systems [@guillamon2009computational; @huguet2013computation] to higher dimensional systems ($d \geq 2$), which are based on the classical parameterization method. In this paper, we use efficient numerical tools to obtain a semi-analytical approximation of the parameterization $K$ using Fourier-Taylor expansions around the limit cycle $\Gamma$. Whereas the lower order terms in the Taylor expansion (degree 0 and 1) are straightforward to obtain using variational equations (see for instance [@brown2004phase]), the higher order terms (degree larger than 1) involve solving a homological equation for each term in the Taylor expansion. There are two main challenges to solve these equations. On one hand, to obtain the homological equations we need to compute the Taylor expansion of the composition of the vector field $X$ with the terms of the Taylor expansion of $K$ of lower order previously computed. Some works [@wilson2020] propose a symbolic manipulator to compute these derivatives, but for high orders and high dimensional systems this strategy is computationally very expensive. Alternatively, we propose to use automatic differentiation techniques as it is typical in the parameterization method [@haro2016]. These techniques avoid the use of symbolic differentiation, while provide the exact recurrent formulas for the terms of the Taylor expansion, which are faster and without numerical errors besides roundoff errors. On the other hand, one needs to find a periodic solution of each homological equation. In [@wilson2020] the authors design a Newton method to find a periodic orbit for each equation. In this paper, we adapt the methodology in [@castelli2015parameterization] (see also [@huguet2013computation]), based on Floquet theory, to reduce the problem to solve a diagonal linear system for the Fourier coefficients, which has an exact explicit solution. Of course, we can easily switch between Fourier and real space using the FFT algorithms. Moreover, this method takes advantage of the form of the equations to provide a general solution for all orders where only the computation of the terms $B_{\alpha,m-\alpha}$ in is required at each step (see Remark \[rem:general\]). Thus, our method allows us to obtain high order analytical approximations of the map $K$ at a low computational cost.
Our approach also proposes a numerical strategy to globalize the local isochrons given by the approximate parameterization $\bar K$. There exist several approaches in the literature for computing global (un)stable manifolds of a vector field, see [@Krauskopfetal05; @simo1990analytical]. In [@guillamon2009computational] the authors adapted the method in [@simo1990analytical] which uses backwards integration to compute $1$-dimensional global isochrons. The procedure proposes an efficient way of choosing the points in the local domain to obtain a homogeneous coverage of the isochron when globalizing it. The higher dimensional case $d > 2$, addressed in this paper, is more difficult because there are several expanding directions and using rough backwards integration produces points that are concentrated along the direction of the largest multiplier. In this paper, we solve this drawback and we provide the details for the 3-dimensional case. Our procedure is based on [@simo1990analytical], which computes first the invariant slow submanifold $\mathcal{S}$ , corresponding to the direction of the smallest (in modulus) multiplier. It is known that the trajectories asymptotically approach the limit cycle along this manifold. Furthermore, we use this manifold to globalize the isochron along the isostables (see Fig. \[fig:globalMethod3d\]C). Finally, we use the parameterization $K$ to compute the iPRF and the iARFs analytically for points in the local domain and we globalize these functions to the whole basin of attraction by using a modified adjoint equation . We illustrate the methodology for different models in neuroscience.
Finally, in Section \[sec:section5\] we use the computed parameterization $K$ to extend the study of phase dynamics for perturbed systems beyond the limit cycle. We show first that the full phase-amplitude description captures the perturbed dynamics accurately. Moreover, we have explored the scope of applicability of two useful reductions, namely the slow manifold reduction and the phase reduction. While the limitations of the phase reduction are already well-known (see for instance [@castejon2013phase; @AshComNic2016; @wilson2018greater]) we have been able to identify the limitations of the slow manifold reduction for a particular example. More precisely, we observe that even for systems with two Floquet exponents with a large difference in size and one of them close to zero, the slow manifold reduction might not be accurate enough. Indeed, if the interpulse interval is too short, the slow manifold reduction might not be sufficient and the method would require to consider the full phase-amplitude system. When this period increases, we expect that the performance of the slow manifold is guaranteed.
We acknowledge that we assumed that the Floquet multipliers of the limit cycle are all different and real. We highlight that we can easily tackle the extension to the case with complex conjugate multipliers, just adapting the techniques in [@castelli2015parameterization]. Of special interest is the case where the Floquet exponents are resonant (see Remark \[rem:nres0\] and Remark \[rem:nonres\]). In this case, it is not possible to conjugate the dynamics to a linear vector field and one needs to consider higher order terms in the dynamics of the amplitude variables $\sigma$ (see [@cabre2003parameterization] for a thorough discussion and [@James16] for an application to the case of the stable manifold of an equilibrium point). A detailed study of this case is an interesting topic for future research.
The study of the oscillatory dynamics beyond the phase reduction has become a topic of growing interest in the last decade (see [@AshComNic2016; @ermentroutetal19]). We briefly discuss how our method based on the parameterization method, compares with other approaches in the literature. In [@wilson2018greater] the authors provide a method to compute the iPRF $\nabla \Theta$ and iARFs $\nabla \Sigma_i$, $i=1,\ldots,d-1$ in a neighbourhood of the limit cycle by means of a linear order expansion of these functions in the amplitude variable, using the generalized adjoint equations. In a recent paper [@wilson2020], the author extends the previous results to obtain a higher order expansion of the iPRF and iARFs in the amplitude variables in a neighbourhood of the limit cycle. Moreover, he also writes the classical homological equations for computing the parameterization $K$ [@cabre2003parameterization]. We want to stress that the strategy to solve these equations is different from ours, as discussed above.
Several groups use the spectral properties of the Koopman operator [@budivsic2012applied; @mezic2005spectral] to provide an alternative approach to the study of the phase-amplitude dynamics of a limit cycle. In particular, the Koopman operator has been used to compute global isochrons and isostables of a given limit cycle by forward integration [@mauroy2012use], as well as PRFs and ARFs. This methodology has been used to compute the amplitude coordinate corresponding to the dominant Floquet multiplier to study perturbed oscillatory dynamics [@mauroy2018global; @shirasaka2017phase]. We emphasize that our method provides all transversal directions in the same computation using the same procedure.
We acknowledge here that we only have applied our methods to neural oscillators, even if they are valid for any nonlinear oscillator. Neural oscillatory activity is widely observed at different levels of organization [@buzsaki2006rhythms], so the phase reduction of neural oscillators has been profusely studied from the single neuron to the network level [@rinzel1998analysis; @brown2004phase; @izhikevich2007; @AshComNic2016]. Moreover, the PRCs have been an important tool in neuroscience both from the theoretical and experimental perspective [@smeal2010phase; @gutkin2005phase; @achuthan2009phase; @canavier2015phase; @schultheiss2011phase]. However, other areas of biology, life sciences or control theory offer more examples of oscillatory activity [@strogatzbook; @monga2019phase] to which this methodology can be applied.
Over the last decade, there has been an increasing need of understanding phase dynamics from biological data [@tass2007phase]. Indeed, the interest in therapies focusing on stimulating cellular tissue at particular phases of certain pathological rhythms is increasing since it has been reported to suppress them [@holt2016phasic; @rosenblum2004delayed; @azodi2015phase]. This, on its turn, has stimulated the study and computation of phase dynamics and PRCs in the stochastic domain [@schwabedal2013phase; @thomas2014asymptotic; @rosenblum2019numerical]. Since our methodology provides accurate descriptions and efficient computational techniques, we consider its possible extensions in the stochastic domain as a very promising and interesting area of research.
In conclusion, in this paper we show how the parameterization method provides a clear and solid theoretical framework which can be used in an efficient computable way to obtain both a geometrical and dynamical accurate description of the oscillatory dynamics. We hope that the application of the different techniques herein can be useful to gain insight and alternative perspectives to oscillatory dynamics.
Acknowledgments
===============
This work has been partially funded by the Spanish MINECO-FEDER Grant PGC2018-098676-B-100 (AEI/FEDER/UE), the Catalan Grant 2017SGR1049 (GH, AP, TS), and the Russian Scientific Foundation Grant 14- 41-00044 (TS). GH acknowledges the RyC project RYC-2014-15866. TS is supported by the Catalan Institution for research and advanced studies via an ICREA academia price 2018. AP acknowledges the FPI Grant from project MINECO-FEDER-UE MTM2012-31714.
Computation of the Floquet normal form {#sec:floquetComputation}
======================================
In this Appendix we provide the method to compute the matrices $\mathcal{Q}$ and $R$ that appear in formula . By Floquet theory [@floquet1883equations], the monodromy matrix satisfies $$M = \Phi(T) = \mathcal{Q}(T)e^{TR} = \mathcal{Q}(0)e^{TR} = \Phi(0)e^{TR} = e^{TR},$$ where we have used $\Phi(0) = Id$ and $\mathcal{Q}(T)$ is a T-periodic matrix.
Then, if there exists a matrix $C$ such that $$\Phi(T) = CDC^{-1} \quad \quad \quad \text{where} \quad D = diag(1, \mu_1, \mu_2),$$ one can find the matrix $R$ in as $$R = \frac{1}{T} C \begin{pmatrix}
0 & & \\
& \ln(\mu_1) & \\
& & \ln(\mu_2)
\end{pmatrix} C^{-1} = CJ C^{-1},$$ where $J$ is given in , with $\lambda_0 = 0$, $\lambda_i = \ln(\mu_i)/T$ $i=1,2$. Therefore, the matrix $\mathcal{Q}$ in is given by $$\mathcal{Q}(t) = \Phi(t) e^{-tR} = \Phi(t) C \begin{pmatrix}
1 & & \\
& e^{\frac{-t}{T}\ln(\mu_1)} & \\
& & e^{\frac{-t}{T}\ln(\mu_2)}
\end{pmatrix} C^{-1} = \Phi(t) C \begin{pmatrix}
1 & & \\
& e^{-t \lambda_1} & \\
& & e^{-t \lambda_2}
\end{pmatrix} C^{-1}.$$ We recall that $\mu_1$ and $\mu_2$ are the Floquet multipliers and $\lambda_1$ and $\lambda_2$ the Floquet exponents.
Automatic Differentiation {#sec:autoDifAp}
=========================
In this Section we aim to illustrate the Automatic Differentiation techniques. We recall that we use this technique in Section \[sec:sectionNum\] to obtain the terms $B_{\alpha, m-\alpha}$ in , corresponding to the terms of degree $m$ of the Taylor expansion of $X(K(\theta, \sigma))$ around $\sigma = 0$ (see Eq. \[eq:Bexpand\]). A possible approach to obtain the coefficients of the Taylor expansion is the to compute the appropriate derivatives. Writing and computing explicitly (using symbolic differentiation) the derivatives up to a given order is a costly task. Nevertheless, if the vector field $X$ is analytic, it can be written as a combination of algebraic operations (sum, product, etc.) and elementary transcendental functions (sin, cos, exp, log, power, etc.), and we can overcome this drawback by using automatic differentiation techniques. Automatic differentiation avoids the need of calculating by hand the $n$-th order derivatives by obtaining the terms in using simple recurrent relations. These relations are obtained by applying systematically the chain rule to each of the operations that compose the function. We stress that this procedure, which is faster than symbolic differentiation, provides the terms without numerical error. This appendix aims to show the basics of this technique for a Taylor expansion in two variables. For more details we refer the reader to [@griewank2008evaluating; @haro2016; @JorbaZ05].
Start by considering a function $f$ expressed in power series in the variables $\sigma_1, \sigma_2 \in \mathbb{R}$, $$\label{eq:equacioPerF}
f(\theta, \sigma) = \sum_{m=0}^{\infty} \sum_{\alpha=0}^{m} f_{\alpha, m-\alpha}(\theta) \sigma^{\alpha}_1 \sigma^{m-\alpha}_2,$$ where $f_{\alpha, m-\alpha}(\theta)$ are periodic coefficients in $\theta$.
Given an analytic function $\varphi$, our goal is to find the Taylor expansion of the composition $\varphi(f)$, that is, $$\label{eq:equacioPerPhi}
\varphi(f)(\theta, \sigma) = \sum_{m=0}^{\infty} \sum_{\alpha=0}^{m} [\varphi(f)]_{\alpha, m-\alpha}(\theta) \sigma^{\alpha}_1 \sigma^{m-\alpha}_2.$$
Even if the coefficients of these Taylor expansions depend periodically on the variable $\theta$, this feature does not play any role in the procedure we are going to expose. For this reason, from now on we do not write explicitly this dependence. Therefore, we will write $f(\sigma)$ instead of $f(\theta,\sigma)$ and $f_{\alpha, m-\alpha}$ instead of $f_{\alpha, m-\alpha}(\theta)$.
We introduce the radial derivative, which is defined as $$Rf(\sigma) = \nabla f(\sigma) \cdot \sigma = \sum_{i=1}^{d=2} \frac{\partial f}{\partial \sigma_i}(\sigma) \sigma_i,$$ and has two useful properties. The first one, is that when applied to the Taylor expansion it satisfies the following relationship $$\label{eq:eulerIdentity}
Rf(\sigma) = \sum_{m=0}^{\infty} m \sum_{\alpha=0}^{m} f_{\alpha, m-\alpha} \sigma^{\alpha}_1 \sigma^{m-\alpha}_2,$$ which is known as Euler’s identity. The second one comes up when applying the chain rule to the composition $\varphi(f(\sigma))$ in , $$\label{eq:chainRule}
R \varphi(f(\sigma)) = \varphi'(f(\sigma)) R \varphi(f(\sigma)).$$
Next, we will show that for the case of $\varphi(x)$ being an elementary function, the combination of and , allows us to compute the $n$-th order coefficient of starting by the already known $\varphi(f_0)$ as initial seed. We illustrate the methodology for the example $\varphi(x) = \exp(x)$. Writing the Taylor expansion of the function $\exp(f)(\sigma)$ as $$\label{eq:equacioPerExp}
\exp(f)(\sigma) = \sum_{m=0}^{\infty} \sum_{\alpha=0}^{m} e_{\alpha, m-\alpha} \sigma^{\alpha}_1 \sigma^{m-\alpha}_2,$$ and since $\varphi'(x)= \varphi(x)$, substituting in and using yields $$\label{eq:bigEquation}
\sum_{m=0}^{\infty} m \sum_{\alpha=0}^{m} e_{\alpha, m-\alpha} \sigma^{\alpha}_1 \sigma^{m-\alpha}_2 = \left(\sum_{n=0}^{\infty} \sum_{\beta=0}^{n} e_{\beta, n-\beta} \sigma^{\beta}_1 \sigma^{n-\beta}_2 \right)\left(\sum_{l=0}^{\infty} l \sum_{\mu=0}^{l} f_{\mu, l-\mu} \sigma^{\mu}_1 \sigma^{l-\mu}_2 \right).$$ Then, collecting the terms of the same order, we end up with the following recurrence equation $$\label{eq:recEquation}
e_{\alpha, m-\alpha} = \frac{1}{m} \sum_{k=0}^{\alpha-1} \sum_{l=0}^{m-\alpha} (m - k - l) e_{k,l} f_{\alpha-k, m-\alpha-l}, \quad \quad \quad \text{for} \quad m \geq 1, \alpha=0,\ldots,m.$$ Notice that the computation of $e_{\alpha, m-\alpha}$ only depends on terms $e_{k,l}$, where $0 \leq k+l \leq m-1$.
Recurrence equations for a large variety of transcendental functions can be found in Table 2.1 in [@haro2016].
Model Parameters {#sec:appendix}
================
RT model {#rt-model .unnumbered}
--------
For the single neuron model we have used the following functions $$\begin{aligned}
I_L = g_L(V - V_L), \quad & \quad
I_{Na} = g_{Na}m^3_{\infty}h(V-V_{Na}),\\
I_K = g_K(.75(1-h))^4(V-V_K), \quad & \quad
I_T = g_Tp^2_{\infty}r(V-V_{T}),\\
h_{\infty}(V) = \frac{1}{1 + \exp((v+41)/4)}, \quad & \quad
r_{\infty}(V) = \frac{1}{1 + \exp((v+84)/4)},\\
m_{\infty}(V) = \frac{1}{1 + \exp(-(v+37)/7)}, \quad & \quad
p_{\infty}(V) = \frac{1}{1 + \exp(-(v+60)/6.2)},\\
\tau_r(V) = 28 + \exp(-(v+25)/10.5), \quad & \quad \tau_h(V) = 1/(a_h(V) + b_h(V)),\\
\quad a_h(V) = 0.128 \exp(-(v+46)/18), \quad & \quad
b_h(V) = \frac{4}{1 + \exp(-(v+23)/5)},
\end{aligned}$$ and the following set of parameters $$\begin{split}
\mathcal{P}_{RT} = \{C_m = 1, g_L = 0.05, V_L = -70, g_{Na} = 3, V_{Na} = 50,\\ g_K = 5, V_K = -90, g_T = 5, V_T = 0\}.
\end{split}$$
HH model {#hh-model .unnumbered}
--------
For the $HH$ model we have used the following functions $$I_L = g_L(V - V_L), \quad \quad I_{Na} = g_{Na}m^3_{\infty}h(V-V_{Na}), \quad \quad
I_K = g_K n^4(V-V_K),$$ $$n_{\infty}(V) = \frac{1}{1 + \exp(-(V+53)/15)}, \quad \quad
h_{\infty}(V) = \frac{1}{1 + \exp((V+62)/7)},$$ $$m_{\infty}(V) = \frac{1}{1 + \exp(-(V+40)/9)},$$ $$\tau_h(V) = 7.4 \exp(-((67+V)/20)^2) + 1.2, \quad \quad
\tau_n(V) = 4.7 \exp(-((79+V)/50)^2) + 1.1,$$ and the following set of parameters $$\mathcal{P}_{HH} = \{ C_m = 1, g_L = 0.1, V_L = -75.6, g_{Na} = 30, V_{Na} = 55, g_K = 9, V_K = -77 \}.$$
WC-Syn Model {#wc-syn-model .unnumbered}
------------
For the extension of the Wilson-Cowan equations we have used the following functions $$\delta_{E}(x) = \frac{1}{1 + \exp(-a_E(v-\theta_E))}, \quad
\delta_{I}(x) = \frac{1}{1 + \exp(-a_I(v-\theta_I))},$$ and the following set of parameters $$\begin{split}
\mathcal{P}_{WC} = \{P = 4.5, \tau_e = 3, a = 8, b = 16, a_E = 3, \theta_E = 4, Q = 0, \\d = 3, \tau_i = 3, c = 7, a_I = 2, \theta_I = 3, \tau_r = 1, \tau_d = 6\}.
\end{split}$$
QIF Model {#qif-model .unnumbered}
---------
For the $QIF$ model we have used the following parameters $$\mathcal{P}_{QF} = \{ \tau_m = 10, \Delta = 0.3, J = 21, \Theta = 4, \tau_d = 5 \}.$$
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The behavior of a uniformly magnetized ferronematic slab is investigated numerically in a situation in which an external magnetic field is applied parallel and antiparallel, respectively, to its initial magnetization direction. The employed numerical method allows one to determine hysteresis curves, from which a critical magnetic field strength (i.e., the one at which the ferronematic sample becomes distorted) as function of the system parameters can be inferred. Two possible mechanisms of switching the magnetization by applying a magnetic field in the antiparallel direction are observed and characterized in terms of the coupling constant between the magnetization and the nematic director as well as in terms of the coupling strength of the nematic liquid crystal and the walls of the slab. Suitably prepared walls allow one to combine both switching mechanisms in one setup, such that one can construct a cell the magnetization of which can be reversibly switched off.'
author:
- Grigorii Zarubin
- Markus Bier
- 'S. Dietrich'
date: 5 September 2018
title: Ferronematics in confinement
---
\[sec:Intro\]Introduction
=========================
Ferronematics, i.e., suspensions of anisotropic ferromagnetic particles dispersed in a nematic liquid crystal (NLC), attract both theoretical [@2006_Zadorozhnii; @2007_Zadorozhnii; @2014_Brand; @2014_Zakhlevnykh; @2015_Zakhlevnykh; @2016_Boychuk; @2016_Zakhlevnykh; @2017_Zakhlevnykh] and experimental [@2011_Podoliak; @2013_Mertelj; @2014_Mertelj; @2015_Hess; @2017_Mertelj; @2018_Maximean] interest due to their ability to exhibit fluidity due to the solvent as well as macroscopic magnetization due to the colloidal inclusions. The anisotropic nature of the solvent implies broken rotational symmetry as compared to a simple isotropic liquid. The interaction of the anisotropic ferromagnetic colloids with the solvent depends on the orientation of the former with respect to the nematic director of the latter. As a result, the individual magnetic moments of the colloids become effectively trapped around the two possible orientations of the nematic order. Therefore, suitably prepared samples can exhibit a macroscopically ferromagnetic phase. The phase behavior of this complex system follows from its free energy density. The authors of Ref. [@2013_Mertelj] proposed a phenomenological expression thereof which is formulated in terms of the magnetization $\mathbf{M}$ and the nematic director $\mathbf{n}$. A similar expression was derived analytically starting from a microscopic description of the system [@2018_Zarubin]: $$\label{eq:FN}
f(\mathbf{M}, \mathbf{n})=
\frac{a}{2}|\mathbf{M}|^2 - \frac{1}{2}\gamma\mu_0(\mathbf{M}\cdot\mathbf{n})^2
- \mathbf{M}\cdot\mathbf{B}$$ where $\mu_0=4\pi\times 10^{-7}$ N/A$^2$ is the permeability of vacuum, $a>0$ is a constant which depends on properties of both the nematic medium and the colloids (for an explicit form see Ref. [@2018_Zarubin]), $\gamma \geq 0$ measures the coupling between the magnetization and the nematic director, and $\mathbf{B}=B\mathbf{e}_x$ is the external magnetic field. Both $a$ and $\gamma$ are functions of the microscopic coupling constant $c:=WR/K$ where $W$ is the anchoring strength measuring the interaction energy of the NLC per surface area of a single colloid, $R$ is the radius of a colloidal particle modeled as a thin disc (i.e., a disc whose thickness is much smaller than its radius such that the interaction of the rim with the NLC medium can be disregarded), and $K$ is the elastic constant of the NLC within the one-elastic-constant approximation. (The value of $K=3.5\times10^{-12}$ N, corresponding to the twist elastic constant of 5CB [@2014_Mertelj], is used throughout the current study if not specified otherwise.)
One of the interesting results of the experiments reported in Ref. [@2013_Mertelj] was the observation of a complex response of the ferronematic slab to a uniform external magnetic field, which depends on the initial state of the sample: If, on one hand, the sample was prepared by quenching the NLC solvent from the isotropic into the nematic phase in the absence of an external magnetic field, the colloids formed *various* magnetic domains within which $\mathbf{M}||\mathbf{n}$. If, on the other hand, the NLC solvent was quenched in the presence of a uniform magnetic field, a *single* domain formed with the entire sample being magnetized in one direction with $\mathbf{M}||\mathbf{n}$. Applying thereafter a uniform external magnetic field opposite to the direction of the magnetization of the single-domain sample yielded a complex, optically observable response of a nonuniform director field.
Here we focus on the case of single-domain samples. So far such samples have been thoroughly investigated theoretically in the situation in which the external magnetic field is applied *perpendicular* to the initial magnetization of the sample [@2006_Zadorozhnii; @2007_Zadorozhnii; @2015_Zakhlevnykh]. Also the dynamics of such a configuration was investigated experimentally [@2017_Potisk; @2018_Potisk; @2018_Sebastian]. A *thresholdless* distortion of the nematic was observed. Moreover, the authors of Refs. [@2006_Zadorozhnii; @2007_Zadorozhnii; @2015_Zakhlevnykh] considered infinitely strong anchoring at the walls and external magnetic fields up to magnitudes strong enough to directly interact with the magnetically anisotropic NLC molecules.
Our aim is to investigate in detail the behavior of a monodomain sample exposed to a uniform external magnetic field which is applied in the direction *antiparallel* to the initial magnetization. Moreover, we consider only magnetic fields of small ($\leq 25$ mT) amplitudes, such that the direct magnetic field influence on the NLC can be neglected. It was observed experimentally [@2013_Mertelj; @2014_Mertelj] that, like for common ferromagnets, ferronematics exhibit hysteresis in the magnetic properties as a function of the external field. Moreover, the critical field, i.e., the magnetic field strength at which the ferronematic becomes distorted (for a more precise definition see Sec. \[subsec:I\] below), is another feature of the ferronematic sample. We obtain the hysteresis curves numerically by using a conjugate-gradient technique in order to minimize an appropriate free energy functional of the ferronematic in slab geometry. From the hysteresis curves one can infer the value of the critical magnetic field as function of the parameters of the model and compare them with the expressions derived in Ref. [@2013_Mertelj]. In our previous study [@2018_Zarubin] we derived the dependence of the coupling parameter $\gamma$ on the microscopic coupling $c$ which in turn depends on the particle size. Having obtained the critical field as a function of $\gamma$ allows us to relate it to the size of the colloids and therefore one can potentially tune the value of the critical field by tuning the mean of the size distribution of the particles used.
It turns out that the switching process of the considered ferronematic slab from one phase to the other takes place according to one of two possible scenarios which we shall discuss. In the first scenario, regions nucleate near the system walls in which the magnetization is flipped, whereas the nematic director is kept in place by the walls. In the second scenario, the nematic director follows the magnetization, i.e., it makes a full rotation by 180$^\circ$, everywhere throughout the sample.
Finally, we propose a novel technique which can be used, e.g., in data storage devices. It is based on magnetic fields which control the magneto-optical properties of ferronematic cells, and thus allows one to switch between magnetized and demagnetized states by applying a uniform magnetic field of suitable orientation.
The paper is organized as follows. In Sec. \[sec:Theory\] we introduce the free energy functional in order to describe the system and the numerical method to minimize it. Section \[subsec:I\] contains the description of the first of the two switching mechanism as well as the results concerning the critical magnetic field and its dependence on the parameters of the model. In Sec. \[subsec:II\] we present the second switching mechanism and provide a map which relates the parameters of the model to the character of the switching. In Sec. \[subsec:TwoWalls\] we report that a combination of the two mechanisms leads to a sample the magnetization of which can be reversibly switched off by using the external magnetic field. The role of the phenomenon of segregation is discussed in Sec. \[sec:Segregation\]. In Sec. \[sec:Discussion\] we conclude by discussing the main results.
\[sec:Theory\]Numerical model
=============================
![Sketch of a ferronematic in a slab of width $D$. The $x$-direction corresponds to the (lateral) easy direction of the liquid crystalline medium. Due to lateral translational invariance, all profiles depend only on the normal coordinate $z$. The nematic director $\mathbf{n}(z)$, the magnetization $\mathbf{M}(z)$, and the external magnetic field $\mathbf{B}$ are parallel to the $x$-$z$-plane and their directions with respect to the positive $x$-direction are described by the angles $\varphi(z)$, $\theta(z)$, and $\psi$, respectively.[]{data-label="fig:Slab"}](./slab.eps)
We consider the experimental setup studied in Ref. [@2013_Mertelj]. It consists of a ferronematic confined by two parallel and planar walls at a distance $D$ with $\mathbf{e}_x$ as the so-called easy axis at both surfaces, which imposes a parallel orientation of the nematic director there (see Fig. \[fig:Slab\]). We assume that the sample was prepared in the presence of a homogeneous external magnetic field $\mathbf{B}=B\mathbf{e}_x$ in the direction parallel to the easy axis $\mathbf{e}_x$ of the walls (i.e., $B>0$), thus producing a single domain of the ferromagnetic phase. In the following the effect of applying an external magnetic field in the direction opposite to the one used during this preparation (i.e., $B<0$) is investigated numerically. The system is described by two spatially varying fields: the director field $\mathbf{n}(\mathbf{r})$ and the magnetization field $\mathbf{M}(\mathbf{r})$. We assume the absolute value of the magnetization is spatially constant, $|\mathbf{M}(\mathbf{r})|=\text{const}=m\rho_\text{iso}$ (where $m$ is the absolute value of the magnetic moment of a single colloid, here taken to be $3\times 10^{-18}$ A m$^2$ [@2013_Mertelj], and $\rho_\text{iso}$ is the number density of the colloids dispersed in the isotropic phase of the liquid crystal during the preparation of the sample (see Refs. [@2013_Mertelj] and [@2018_Zarubin]), i.e., segregation effects are assumed to be small [@footnote1]. (For a discussion of the possible influence of segregation see Sec. \[sec:Segregation\].) Due to translational invariance in the lateral $x$-$y$-plane all physical quantities depend on the normal coordinate $z$ only. We consider that both $\mathbf{n}(z)$ and $\mathbf{M}(z)/(m\rho_\text{iso})$ are parallel to the $x$-$z$-plane [@footnote2], so that they can be described by the angles $\varphi(z)$ and $\theta(z)$, respectively (see Fig. \[fig:Slab\]). The initial configuration is given by the uniform profiles $\varphi(z)=0$ and $\theta(z)=0$, which corresponds to an unstable state when a uniform magnetic field $\mathbf{B}$ is applied in the direction $\psi=\pi$ (see Fig. \[fig:Slab\]). In terms of the profiles $\varphi$ and $\theta$ the free energy functional of the system is given by $$\label{eq:SlabF}
\frac{1}{S}\beta\mathcal{F}[\varphi,\theta] =
\beta F_\text{ferr}[\varphi,\theta] + \beta F_\text{elas}[\varphi] +
\beta F_\text{surf}[\varphi],$$ where $S$ is the surface area of one of the glass plates, $\beta:=1/(k_\text{B}T)$, $$\label{eq:Fferr}
\beta F_\text{ferr}[\varphi,\theta]=
\int_0^D\!\!{\ensuremath{\mathop{}\!\mathrm{d}}}z\ \beta f(\mathbf{M}(z),\mathbf{n}(z))$$ with the free energy density $f$ given by Eq. , which is the contribution due to the *ferr*onematic, $$\label{eq:NLCEl}
\beta F_\text{elas}[\varphi]=
\frac{1}{2}\beta K\int_0^D\!\!{\ensuremath{\mathop{}\!\mathrm{d}}}z\ \left( \frac{{\ensuremath{\mathop{}\!\mathrm{d}}}\varphi(z)}{{\ensuremath{\mathop{}\!\mathrm{d}}}z}\right)^2,$$ is the contribution due to the elastic distortions of the liquid crystal, and $$\label{eq:WallC}
\beta F_\text{surf}[\varphi]=
-\frac{1}{2}\beta W_\text{wall}\left(\cos(\varphi(0))^2+\cos(\varphi(D))^2\right)$$ is the contribution due to the coupling of the liquid crystal to the glass plates.
The equilibrium profiles $\varphi(z)$ and $\theta(z)$ correspond to the minimum of the free energy in Eq. , which has been determined numerically by using the Fletcher-Reeves-Polak-Ribiere general function minimization algorithm [@2002_Press]. The absolute value $|\mathbf{M}|$ of the magnetization is assumed to have a constant value $m\rho_\text{iso}$ and is taken to be independent of the external field $\mathbf{B}$ throughout Sec. \[sec:Results\]. For the discussion of the problem in the case of a spatially varying $|\mathbf{M}|$ see Sec. \[sec:Segregation\].
We have used the following parameter values: $K=3.5\times 10^{-12}$ N, $\rho_\text{iso}=1.5\times 10^{19}$ m$^{-3}$, $m=3\times 10^{-18}$ A m$^2$, and $T=300$ K. These values are consistent with experimental data reported in Refs. [@2013_Mertelj; @2014_Mertelj]. The thickness $D$ of the slab is taken to be 20 $\mu$m throughout Sec. \[sec:Results\]. Thicker slabs are considered in Sec. \[sec:Segregation\] [@footnote7; @1995_Burylov].
\[sec:Results\]Results
======================
\[subsec:I\]Switching mechanism I and the critical field
--------------------------------------------------------
The experiments in Refs. [@2013_Mertelj; @2014_Mertelj] demonstrate that, upon applying a uniform external magnetic field to the setup described in Sec. \[sec:Theory\] (see also Fig. \[fig:Slab\]), there is a nonvanishing critical magnetic field strength $\mathbf{B}_\text{cr}=B_\text{cr}\mathbf{e}_x$ such, that for $B<B_\text{cr}<0$ ($B>B_\text{cr}>0$) in the case of an initial magnetization pointing along the positive (negative) $x$-direction, elastic distortions of the liquid crystal matrix occur. The occurrence of such a critical magnetic field strength $B_\text{cr}$ can be explained qualitatively in terms of a diverging relaxation time of the fluctuations of the nematic director field $\mathbf{n}$ (or $\varphi$) (see Ref. [@2013_Mertelj]). Here we aim at exploring the dependence of $B_\text{cr}$ on the coupling constant $\gamma$ and the wall anchoring strength $W_\text{wall}$ [@footnote5]. Moreover, we are also interested in the intermediate metastable states preceding the switches of the magnetization field $\mathbf{M}$ to the ground state parallel to the external field $\mathbf{B}$.
![Hysteresis of the spatially averaged magnetization $\mathcal{M}$ along the $x$-direction (see Eq. ) of the ferronematic slab as function of the $x$-component $B$ of the external magnetic field $\mathbf{B}$. Negative values of $\mathcal{M}$ or $B$ correspond to orientations in the negative $x$-direction (see Fig. \[fig:Slab\]). For initially saturated samples with $\mathcal{M}=1$ (i.e., magnetized in the positive $x$-direction) there is a nonvanishing critical magnetic field $B_\text{cr}<0$ (indicated in the plot) such that for $B\in[B_\text{cr},0]$ the magnetization $\mathcal{M}$ does not respond to the external field. Upon increasing the field in the negative $x$-direction (i.e., for $B<B_\text{cr}$, left branch of the loop) the system evolves through a series of qualitatively distinct metastable states (red circles) corresponding to the profiles displayed in Fig. \[fig:MetaS\] and eventually reaches saturation along the negative $x$-direction (i.e., $\mathcal{M}=-1$). Gradually lowering the magnitude $|B|$ of the magnetic field does not influence the magnetization of the sample (i.e., for $B<0$, see the part of the loop along $\mathcal{M}=-1$). After $B=0$ is crossed, the situation is identical to the one described above up to a change of sign of $B$ and $\mathcal{M}$ (right branch of the loop). Dotted vertical lines separate regions of qualitatively different metastable states. Note that state (1) corresponds to $B_\text{cr}<B<0$. $W_\text{wall}=3.1\times 10^{-5}$ J/m$^2$, $\gamma=240$, $K=3.5\times 10^{-12}$ N, $\rho_\text{iso}=1.5\times 10^{19}$ m$^{-3}$, $m=3\times 10^{-18}$ A m$^2$, and $T=300$ K[]{data-label="fig:Hyst"}](./hyst.eps)
Figure \[fig:Hyst\] shows the hysteresis curve of the spatially averaged projection of the magnetization $\mathcal{M}$ onto the $x$-axis, $$\label{eq:PrettyM}
\mathcal{M}:=\frac{1}{D}\int_0^D\!\!{\ensuremath{\mathop{}\!\mathrm{d}}}z\,\,\cos\theta (z),$$ as function of the component of the external magnetic field $\mathbf{B}$ along the $x$-axis for the particular choice of the coupling constant $\gamma=240$ (which corresponds to a value of the microscopic coupling constant $c\approx 0.035$ [@2018_Zarubin]) and of the wall anchoring strength $W_\text{wall}=3.1\times 10^{-5}$ J/m$^2$; this choice of the parameters is reasonable in the context of available experimental data (see Refs. [@2013_Mertelj], [@2014_Mertelj], and [@2018_Zarubin]). In order to investigate the switching process of the magnetization in the ferromagnetic phase as function of the external magnetic field, Fig. \[fig:MetaS\] displays the orientation profiles $\varphi(z)$ and $\theta(z)$ for a series of intermediate metastable states corresponding to the hysteresis loop in Fig. \[fig:Hyst\].
For initially saturated samples with $\mathcal{M}=1$, the magnetization does not change significantly in the presence of $x$-components of the magnetic field $B > B_\text{cr}$, whereas for $B < B_\text{cr}$ there is a noticable deviation of the $x$-component of the spatially averaged magnetization $\mathcal{M}$ from the initial saturation value (see Fig. \[fig:Hyst\]). This defines a critical magnetic field strength $B_\text{cr} < 0$. For $B > B_\text{cr}$ both the magnetization and the nematic director field profiles, i.e., $\theta(z)$ and $\varphi(z)$, de facto do not deviate from the saturated ones (see Fig. \[fig:MetaS\](a)). While the magnetization tends to align with the external magnetic field, due to the interaction described by the coupling constant $\gamma$ it is dragging the nematic director field along. At $B=B_\text{cr}$ the metastable state corresponding to the unperturbed nematic director becomes unfavorable compared to the metastable state corresponding to the perturbation induced in the interior of the slab (see Fig. \[fig:MetaS\](b)). We note that the saturated sample with $\mathcal{M}=1$ in a magnetic field in the negative $x$-direction with $B=B_\text{cr}$ corresponds to a saddle point of the free energy so that spontaneous symmetry breaking induced by fluctuations leads to perturbations of the magnetization orientation profile $\theta(z)$ with either $\theta(z) > 0$ or $\theta(z)<0$. In the following we focus only on the first case, while the second, conjugated one, follows from changing signs. It is the perturbed nematic director field which manifests itself as a brightening of the sample when viewed with crossed polarizers as in the experiment reported in Ref. [@2013_Mertelj], and it occurs only due to the coupling of the magnetization field $\mathbf{M}$ to the nematic director field $\mathbf{n}$. The external magnetic field imposes a torque onto the magnetization field which in turn leads to a torque onto the nematic director field. The latter is opposing the torque generated by the walls of the cell and which is transmitted due to the elasticity of the NLC (Eq. ). Upon increasing the external magnetic field the variations inside the slab become more and more pronounced for both the magnetization and the nematic director field. However, in the case of soft anchoring [@footnote6] (distinct from the case of infinitely strong anchoring, see Ref. [@1970_Brochard]) at the surface of the colloid, the angle between the magnetization and the nematic director is nonzero for $B < B_\text{cr}$ (see Figs. \[fig:Slab\] and \[fig:MetaS\]), i.e., $\theta\neq\varphi$.
![Explicit magnetization field (red arrows) and nematic director field (black rods) of the pretransitional metastable state (3) (see Figs. \[fig:Hyst\] and \[fig:MetaS\](c)) of the ferronematic in between two glass walls (hatched regions). While in the interior of the slab (white region) the magnetization field (red arrows) is, to a large extent, aligned with the external magnetic field in the negative $x$-direction, close to the walls (grey regions) its orientation interpolates between the configurations being parallel and antiparallel to the magnetic field. This is due to the coupling to the nematic director field (black rods), which is aligned along the easy direction at the walls. Upon further increasing the strength of the external magnetic field in the negative $x$-direction ($B < B_\text{cr} < 0$), the grey regions widen and eventually produce an almost uniformly magnetized sample (see Fig. \[fig:MetaS\](d)). The pale red regions denote thin layers very close to the walls which switch last.[]{data-label="fig:Nucleation"}](./vectors.eps)
Before reaching the magnetic phase with the sample being magnetized along the field in negative $x$-direction, the system passes through the metastable state (3) in Fig. \[fig:Hyst\], in which the magnetization profile $\theta(z)$ has a peculiar form (see Figs. \[fig:MetaS\](c) and \[fig:Nucleation\]). In this metastable state the magnetization in the interior of the sample is aligned along the magnetic field. Within certain transition regions close to the walls the orientation of the magnetization interpolates between the direction along the magnetic field and the opposite direction. These transition regions occur because the magnetization is coupled to the nematic director field, which is aligned along the easy axis ($\varphi =0$) at the walls. The width of these transition regions grows upon increasing the external magnetic field strength so that eventually the minimum of the free energy given by Eq. corresponds to the magnetization being oriented parallel to the external magnetic field in the entire slab ($\theta=\pi$). Concerning the transition regions following observations can be made: (i) Due to the soft coupling between the colloids and the nematic director field of the NLC the ground state, in which the entire sample is magnetized along the external field, is attained by means of “switching” the magnetization locally, i.e., by inverting the direction of the magnetization (and thus of the orientation of the magnetic colloids) without the simultaneous rotation of the local nematic director field. (ii) Layers of the incipient ferronematic phase are nucleated in the regions close to the walls due to the interplay between elastic and magnetic torques and because the coupling energy is invariant with respect to an inversion of the magnetization.
 and (b) for $\gamma\approx 195$ and $W_\text{wall}=4\times 10^{-5}$ J/m$^2$, respectively (the values of the remaining parameters are the same as in Fig. \[fig:Hyst\]).[]{data-label="fig:Phase"}](./phase.eps)
Naturally the question arises concerning the dependences of the critical magnetic field strength $B_\text{cr}$ on the coupling constant $\gamma$ and on the wall anchoring $W_\text{wall}$. Here we define $B_\text{cr}$ as the magnetic field strength at which the spatially averaged magnetization $\mathcal{M}$ equals $0.97$; note that $\mathcal{M}=1$ in the saturated state. This definition differs from the one used in Refs. [@2013_Mertelj; @2014_Mertelj], where $B_\text{cr}$ is defined as the magnetic field strength at which the relaxation time of thermal fluctuations of the direction of $\mathbf{n}$ diverges. Here we do not consider dynamic processes, instead we propose the above alternative definition of $B_\text{cr}$. Obviously, the choice of $0.97$ for the threshold value contains some degree of arbitrariness. However, as can be inferred from the steep slope of the hysteresis loop close to state (1) in Fig. \[fig:Hyst\], no significant changes are expected to occur by choosing different threshold values not too much less than unity. Figure \[fig:Phase\] shows the dependence of $B_\text{cr}$ on the coupling constant $\gamma$ and the wall anchoring strength $W_\text{wall}$. One can infer from Fig. \[fig:Phase\] that for fixed $W_\text{wall}$ the critical field $B_\text{cr}$ increases upon increasing $\gamma$. Indeed, for a given value of $W_\text{wall}$, the magnetization field is aligned with the nematic director field, the rotation of which is opposed by the torque imposed by the walls. The system sustains the alignment for increasing external magnetic field strengths which in their turn are due to an increasing strength of the coupling $\gamma$ between the magnetization and the nematic director field. One can also infer from Fig. \[fig:Phase\] that the critical field $B_\text{cr}$ depends rather weakly on the wall anchoring strength $W_\text{wall}$: Within the considered range of the anchoring strengths, $W_\text{wall}\in[0.5\times 10^{-5},10\times 10^{-5}]\mathrm{J/m^2}$, for fixed $\gamma$ the critical field strength $B_\text{cr}$ varies by $\approx 0.1\text{ mT}$. For large values of $W_\text{wall}$ the critical field reaches a plateau (see Fig. \[fig:Cuts\] (a)) and it becomes independent of the wall anchoring $W_\text{wall}$.
![Dependence of the critical magnetic field strength $B_\text{cr}$ (a) on the wall anchoring strength $W_\text{wall}$ (along the thick vertical black line in Fig. \[fig:Phase\], i.e., for $\gamma\approx 195$), (b) on the coupling constant $\gamma$ (present approach and Eq. along the thick horizontal black line in Fig. \[fig:Phase\], i.e., for $W_\text{wall}=4\times 10^{-5}$ J/m$^2$), and (c) on the microscopic coupling constant $c$ (see see below Eq. , $W_\text{wall}=4\times 10^{-5}$ J/m$^2$). The values of the remaining parameters are the same as in Fig. \[fig:Hyst\].[]{data-label="fig:Cuts"}](./cuts.eps)
Figure \[fig:Cuts\] illustrates these trends of the critical magnetic field strength $B_\text{cr}$ via cuts in Fig. \[fig:Phase\] for $\gamma\approx 195$ (see Fig. \[fig:Cuts\](a)) and $W_\text{wall}=4\times 10^{-5}$ J/m$^2$ (see Fig. \[fig:Cuts\](b)). Moreover, Fig. \[fig:Cuts\](c) displays the dependence of $B_\text{cr}$ on the microscopic coupling constant $c$ (see below Eq. ). Finally, Fig. \[fig:Cuts\](b) compares the values of the critical magnetic field $B_\text{cr}$ as defined in the present approach (circles) with the corresponding expression given in Ref. [@2013_Mertelj], $$\label{eq:BCr}
B^\text{\cite{2013_Mertelj}}_\text{cr}=
\frac{\pi^2 \gamma \mu_0 K M_\text{s}}{\pi^2 K+\gamma\mu_0 M_\text{s}^2D^2},$$ where $M_\text{s}:=m\rho_\text{iso}$ is the magnetization of the saturated sample, obtained in the limit $W_\text{wall}\to\infty$. (Here, we consider the particular NLC used for the experiments in Refs. [@2013_Mertelj; @2014_Mertelj] and therefore a fixed value of the elastic constant $K$. Although the variation of the expression given here as function of the elastic constant of the NLC is interesting, we leave this issue for future work due to the highly non-trivial occurrences of $K$.) Remarkably, $B_\text{cr} \approx B^\text{\cite{2013_Mertelj}}_\text{cr}$ appears to hold although the two definitions of the critical magnetic field strength differ and although $B^\text{\cite{2013_Mertelj}}_\text{cr}$ in Eq. does not take the dependence on $W_\text{wall}$ into account.
\[subsec:II\]Switching mechanism II
-----------------------------------
In Sec. \[subsec:I\] we revealed a mechanism of switching the sample magnetization in the case of large values of the wall anchoring strength $W_\text{wall}$. Here we show another possible mechanism which corresponds, however, to small values of $W_\text{wall}$. For suitable combinations of $\gamma$ and $W_\text{wall}$ the magnetization field is able to drag the nematic director field along, thereby inducing a large change of the angle $\varphi$ compared with the initial configuration. For small values of $W_\text{wall}$ the anchoring at the wall is so weak, that the nematic director field at the surface of the walls is able to deviate from the direction of the easy axis and to rotate with the magnetic field due to the coupling between the magnetization and the nematic director.
We have performed calculations analogous to those described in Sec. \[subsec:I\] but for small values of the anchoring strength $W_\text{wall}$ at the sample walls. It turns out that for values $W_\text{wall} <0.5\times 10^{-5}$ J/m$^2$ there are corresponding values of the coupling constant $\gamma$ which produce a switching mechanism which is qualitatively different from the one described in the previous section. In this mechanism the early stages of the switching are similar to those described in the previous section (see Fig. \[fig:MetaS\](1) and (2)). However, the subsequent stage, as displayed in Fig. \[fig:MetaSII\](3), is qualitatively different in the sense that the system does not separate in distinct spatial regions with different orientations of the magnetization (compare with Fig. \[fig:MetaS\](3)). Obviously, the change in character of the switching mechanism is directly related to the anchoring at the walls being too weak to prevent the liquid crystal from rotating along with the magnetization field. This weakness is revealed also by nonzero angles $\varphi$ and $\theta$ at the walls.
![Regions of dominance for the two switching mechanisms between two ferromagnetic phases in terms of the coupling constant $\gamma$ and the wall anchoring strength $W_\text{wall}$, which is the same for both walls. (The values of the remaining parameters are the same as in Fig. \[fig:Hyst\].) The region denoted as “I” (full red circles) corresponds to switching mechanism I in which the magnetization leaves the nematic director field behind (see Sec. \[subsec:I\]). The region denoted as “II” (full blue circles) corresponds to switching mechanism II in which the nematic director field is weakly coupled to the sample walls and therefore is able to follow the magnetization.[]{data-label="fig:Switch"}](./switch.eps)
At such low anchoring strengths the torque imposed on the liquid crystal director by the walls cannot compete with the drag imposed by the rotating magnetization field and it is energetically more favorable for the director at the walls to flip its orientation. The regions of dominance for the two switching mechanisms are depicted in Fig. \[fig:Switch\], where a map spanned by the coordinates ($\gamma$, $W_\text{wall}=W_\text{wall}^{(1)}=W_\text{wall}^{(2)}$) marks region “I” (corresponding to switching mechanism I, for which the director field returns back to its initial configuration upon increasing the external magnetic field and reaching the saturation of the magnetization in the direction of the field) and region “II” (corresponding to switching mechanism II, for which the director field follows the magnetization). Within our numerical approach switching mechanism II is observed only for values of the wall anchoring $W_\text{wall}$ which are significantly smaller than the one estimated from the experiment [@2014_Mertelj], i.e., $W^\text{exp}_\text{wall}\approx
(3.40\pm 0.11)\times10^{-5}$ J/m$^2$. Therefore, we expect switching mechanism I to be the one realized experimentally.
The segregation of colloids might play an important role. Segregation amounts to a redistribution of the colloids dispersed in the liquid crystal. This effect is caused by the opportunity to lower the free energy of the magnetic colloids in an external magnetic field by migrating away from regions in which the liquid crystal prevents alignment along the external field. Thereby energy is gained by accomplishing alignment at the expense of the entropic contribution due to denser packing which is proportional to $\rho\log\rho$, where $\rho$ is the local number density of the colloids (see Ref. [@2018_Zarubin]). Although segregation is neglected in Sec. \[sec:Results\], we nonetheless do not expect segregation to influence our results qualitatively (see Sec. \[sec:Segregation\]). On the other hand, quantitative changes are conceivable, i.e., the map in Fig. \[fig:Switch\] might be affected.
\[subsec:TwoWalls\]Confining walls with different anchoring strengths
---------------------------------------------------------------------
In this section we study a combination of switching mechanisms I and II described in Secs. \[subsec:I\] and \[subsec:II\], respectively, by considering a strong anchoring strength at one wall and a weak one at the other, sharing the same easy axis. Thus the system is described by three parameters (assuming $a$, $T$, $K$, $D$, $m$, and $\rho_\text{iso}$ to be fixed): (i) coupling constant $\gamma$, (ii) the anchoring strength at one of the walls, and (iii) the ratio of the anchoring strengths at the two walls. Note that introducing different but still uniform and parallel anchorings at the walls renders the system still effectively one-dimensional along the $z$-direction and hence from a numerical point of view its complexity does not change.
Adding a third parameter (i.e., the ratio of the anchoring strengths at the two walls $w:=W_\text{wall}^{(1)}/W_\text{wall}^{(2)}$) introduces a third dimension to the map considered in Fig. \[fig:Switch\]. The cut of this three-dimensional map along $w=1$ produces the two-dimensional map shown in Fig. \[fig:Switch\]. While the two-dimensional map in Fig. \[fig:Switch\] exhibits only two switching regions (I and II), in the three-dimensional parameter space the situation can be more involved. It is reasonable to expect that if both $W_\text{wall}^{(1)}$ and $W_\text{wall}^{(2)}$ become infinitely strong, the magnetization of the sample switches according to mechanism I. If, on the other hand, $W_\text{wall}^{(1)},W_\text{wall}^{(2)}\to 0$ one can expect that the magnetization of the sample switches according to mechanism II. However, pairs $(W_\text{wall}^{(1)},W_\text{wall}^{(2)})$ can exist such that the magnetization in the vicinity of one wall would switch according to mechanism I and the magnetization in the vicinity of the other wall would switch according to mechanism II. Thus, the three-dimensional parameter space consists of *three* regions: dominance of mechanism I, II, and their combination.
![Orientation profiles $\varphi(z)$ (nematic director, black line) and $\theta(z)$ (magnetization, red line) for $B \leq -12$ mT (with the sample initially magnetized along the positive $x$-direction). There is a gradual variation of the nematic director from one wall to the other. The parameters are chosen as $\gamma=240$, $W^{(1)}_\text{wall}=10^{-5}$ J/m$^2$ (see $z=0$), and $W^{(2)}_\text{wall}=0.1\times10^{-5}$ J/m$^2$ (see $z=D$), $K=9\times 10^{-12}$ N [@footnote3]; the values of the remaining parameters are the same as in Fig. \[fig:Hyst\].[]{data-label="fig:TwoWalls"}](./two_walls.eps)
The line separating the two regions in Fig. \[fig:Switch\] becomes a two-dimensional manifold in the three-dimensional parameter space ($W_\text{wall}^{(1)},w,\gamma$). In the vicinity of the plane $w=1$ this manifold, which is separating the regions “I” and “II”, is considered to be perpendicular to the plane $w=1$ and only the two regions “I” and “II” occur. Therefore, if one would like to find a point ($W_\text{wall}^{(1)},w,\gamma$) that belongs to the region, which corresponds to the combination of the two switching mechanisms, it is necessary to pick the value of $w$ significantly different from 1. To this end, for fixed $\gamma$ it seems to be natural to take as an estimate the anchoring at one of the walls from region “I” in Fig. \[fig:Switch\] and the anchoring at the other wall from region “II” in Fig. \[fig:Switch\]. It has turned out that for $D=20$ $\mu$m (for a discussion concerning larger values of $D$ see Sec. \[sec:Segregation\]) the combination $W_\text{wall}^{(1)}=10^{-5}$ J/m$^2$ and $W_\text{wall}^{(2)}=0.1\times 10^{-5}$ J/m$^2$ (i.e., $w=10$) yields profiles $\varphi(z)$ (nematic director) and $\theta(z)$ (magnetization) which consist of one part due to switching mechanism I and another part due to switching mechanism II. Figure \[fig:TwoWalls\] shows the actual profiles (i.e., for magnetic field strengths $B \leq -12$ mT and for the initial magnetization pointing into the positive $x$-direction) of the magnetization and of the nematic director field for $\gamma=240$, $W^{(1)}_\text{wall}=10^{-5}$ J/m$^2$, and $W^{(2)}_\text{wall}=0.1\times10^{-5}$ J/m$^2$, where superscript $(1)$ denotes the wall at $z=0$ and superscript $(2)$ denotes the wall at $z=D$. Both the orientation field $\varphi(z)$ of the nematic director and the orientational field $\theta(z)$ of the magnetization have a nontrivial form. The nematic director field profile exhibits a smooth rotation by an angle of $\pi$ from one wall to the other. According to Fig. \[fig:TwoWalls\], the wall with the strong anchoring at $z/D=0$ is able to align the nematic director along the easy axis there ($\varphi=0$), while the magnetization field is switching to the negative $x$-direction parallel to the external field (compare Sec. \[subsec:I\]). On the other side, the weak anchoring at $z/D=1$ in Fig. \[fig:TwoWalls\] allows the nematic director there to follow the magnetization ($\theta=\pi$ implies $\varphi=\pi$; compare Sec. \[subsec:II\]). This provides a situation in which $\varphi=0$ at one wall and $\varphi=\pi$ at the other. The elastic contribution in Eq. ensures that no singularities occur in the interior of the slab so that there is a smooth crossover between the two boundary values.
![Vector fields corresponding to Fig. \[fig:TwoWalls\]. The notation is the same as in Fig. \[fig:Nucleation\]. Note the distortion of the nematic director field (black rods) in the interior and the interface between two magnetization (red arrows) domains at $z\approx 0.55D$ due to the combination of the switching mechanisms I and II. Basically throughout the whole sample the magnetization has reached the switched state in negative $x$-direction. $\gamma=240$, $W^{(1)}_\text{wall}=10^{-5}$ J/m$^2$ (see $z=0$), and $W^{(2)}_\text{wall}=0.1\times10^{-5}$ J/m$^2$ (see $z=D$), $K=9\times 10^{-12}$ N [@footnote3] (the values of the remaining parameters are the same as in Fig. \[fig:Hyst\]).[]{data-label="fig:TwoWallsVectors"}](./two_walls_vectors.eps)
In the middle ($z\approx 0.55D$) of the sample the magnetization field exhibits an interface between two halves of the slab (see Fig. \[fig:TwoWallsVectors\]). The orientation of the magnetization within the two halves differs only in how the magnetization approaches the value $\theta=\pi$ in the vicinity of the center of the slab. This behavior of the magnetization field profile is caused by the necessity to be compatible with the nematic director profile in the center region. This means that the rotation of the nematic director in the interior of the slab forces the magnetization direction to reach its value $\theta=\pi$ at $z\approx D/2$ either from $\theta>\pi$ at $z\lesssim D/2$ or from $\theta<\pi$ at $z\gtrsim D/2$ (see the red curve in Fig. \[fig:TwoWalls\]). In the situation of Fig. \[fig:TwoWalls\], upon switching off the external magnetic field, we found that the system relaxes into a state with a uniform nematic director field and two domains with the magnetization pointing into opposite directions (see Fig. \[fig:Storage\](c)) [@footnote3]. The position of the interface between these two domains depends on the position of the interface plane formed while the magnetic field was still on [@footnote4]. Application of the external magnetic field to the two-domain configuration opens up two possibilities: (i) If the external magnetic field is applied in the same direction as the field used to create the two-domain sample, the resulting state is identical to the one in Fig. \[fig:TwoWalls\]. (ii) If, on the other hand, the magnetic field is applied opposite to the direction of the magnetic field used to create the two-domain sample, one of the domains (i.e., the one the magnetization of which is opposite to the external field) switches. This yields a uniform sample both in terms of the nematic director and the magnetization field, thus returning the system to its initial state. These steps are summarized in Fig. \[fig:Storage\]. Note that the states shown in Figs. \[fig:Storage\] (a) and \[fig:Storage\] (d) are identical. Also note that the state depicted in Fig. \[fig:Storage\] (a) exhibits saturated magnetization $\mathcal{M}=1$ (Eq. ) whereas the state depicted on Fig. \[fig:Storage\] (c) exhibits $\mathcal{M}\ll 1$. Since one can restore the initial state (see Fig. \[fig:Storage\] (a)) from the two-domain state (see Fig. \[fig:Storage\](c)) by applying an external magnetic field $B>0$, one is able to cycle through three states (see Figs. \[fig:Storage\] (a), (b), and (c)). Accordingly, this ferronematic cell with two walls of different anchoring strength can be put in either of two states (i.e., magnetized or demagnetized) by using an external magnetic field of suitable direction. This opens up the possibility, e.g., to use an array of such cells for storage of binary information with a “bit” being represented by the state of the cell (magnetized/demagnetized) or as a spatially resolving magnetic field detector with memory function.
![Results of the numerical minimization of the functional in Eq. with segregation effects included (see Eqs. and ). The sample thickness is $D=120\,\mathrm{\mu m}$, the wall anchoring is $W_\text{wall}=3.4\times10^{-5}$ J/m$^2$ for both walls, the coupling constant is $\gamma=240$, the external magnetic field is $B=4$ mT, and $a\approx 3.1\times 10^{-4}$ N/A$^2$. (The values of the remaining parameters are the same as in Fig. \[fig:Hyst\].) Due to the large thickness $D$ of the cell the switching mechanism I is now combined with the switching mechanism II even in the case that both walls provide strong anchoring. The particles are expelled (see the blue solid line) from the regions with strong gradients of the nematic director field $\varphi$ (black solid line). The depletion layers (i.e., the minima of the blue solid line) are separated from the walls due to the soft anchoring between the magnetization and the nematic director field. This differs from the situation described in Ref. [@1970_Brochard] in which the depletion layer is located in close vicinity of the wall.[]{data-label="fig:Segregation"}](./seg.eps)
\[sec:Segregation\]Segregation effects
======================================
It was pointed out by Brochard and de Gennes [@1970_Brochard] that anisotropic magnetic colloids tend to move away from regions of the NLC where distortions of the director field prevent them from minimizing their free energy in the external magnetic field. The segregation parameter defined as (see Ref. [@1970_Brochard]) $$s:=\beta mB$$ is greater than unity already for $B\gtrsim 1.4$ mT and therefore one can expect segregation to occur for external fields stronger than $1.4$ mT. In the following we investigate the impact of segregation on the switching mechanisms I and II.
The present theoretical approach (see Eqs. and ) includes the possibility of segregation to occur through the dependence of the magnetization field $\mathbf{M}$ on the spatial coordinate $z$. In particular, we are interested in spatial inhomogeneities of the *absolute value* $|\mathbf{M}(z)|$ of the magnetization vector. It is convenient to introduce the dimensionless quantity $\tau (z) := |\mathbf{M}(z)|/(m\rho_\text{iso})$. So far all our results have been obtained in the limit $\tau(z) = \text{const} =1$. If $\tau (z) \neq 1$, that part of the free energy density, which depends on it (see Eq. and Fig. \[fig:Slab\]), is given by $$\begin{aligned}
\label{eq:FNWithS}
\beta &f(\theta(z), \varphi(z), \tau(z))=\\
&\beta (m\rho_\text{iso})^2\tau(z)^2 \Big(\frac{a}{2}-\frac{1}{2}\gamma\mu_0 \cos(\theta(z)-\varphi(z))^2\Big)\\
&-\beta m\rho_\text{iso}B\tau(z)\cos(\theta(z)-\psi).
\end{aligned}$$ Concerning the segregation effects the value of $a$ matters. According to Ref. [@2018_Zarubin] the value of $\gamma=240$ implies $c\approx 0.035$ and therefore $a\approx 10\times k_\text{B}T/(m^2\rho_\text{iso})\approx 3.1\times 10^{-4}$ N/A$^2$ (concerning the definition of $a$ in terms of $c$ see Ref. [@2018_Zarubin]). Since the sample always contains a fixed number of magnetic colloids (i.e., neglecting aggregation) the field $\tau(z)$ is subject to the constraint (see Appendix \[sec:Appendix\]) $$\label{eq:TauCons}
\frac{1}{D}\int_0^D\!\!{\ensuremath{\mathop{}\!\mathrm{d}}}z\,\,\tau(z)=1.$$ It is convenient to consider deviations $\delta\tau(z)$ from the homogeneous case, i.e., $$\tau(z)=1+\delta\tau(z),
\label{eq:deltataudef}$$ which allows one to rewrite the free energy density in Eq. as the sum of the free energy density evaluated for $\tau(z) = \text{const} =1$ and the contribution due to segregation: $$\begin{aligned}
\label{eq:Split}
\beta &f(\theta(z), \varphi(z), \tau(z))=\nonumber\\
&\beta f(\theta(z),\varphi(z), \tau(z)=1)+\beta f_\text{seg}(\theta(z), \varphi(z), \tau(z)),\end{aligned}$$ where $f_\text{seg}(\theta(z), \varphi(z), \tau(z))$ is defined as $$\label{eq:FSeg}
\begin{aligned}
\beta &f_\text{seg}(\theta(z), \varphi(z), \tau(z)):=\\
&\beta (m\rho_\text{iso})^2\Big(a-\gamma\mu_0 \cos(\theta(z)-\varphi(z))^2\\
&-\beta m\rho_\text{iso}B\cos(\theta(z)-\psi)\Big)\delta\tau(z)\\
&+\beta (m\rho_\text{iso})^2 \Big(\frac{a}{2}-\frac{1}{2}\gamma\mu_0 \cos(\theta(z)-\varphi(z))^2\Big)\delta\tau(z)^2.
\end{aligned}$$ The constraint in Eq. turns into $$\label{eq:DTauCons}
\int_0^D\!\!{\ensuremath{\mathop{}\!\mathrm{d}}}z\,\,\delta\tau(z)=0.$$ We express $\delta\tau(z)$ in terms of a Fourier series: $$\delta\tau(z)=\frac{a_0}{2}+\sum_{n=1}^\infty \Big[a_n\cos\Big(\frac{2\pi nz}{D}\Big)+b_n\sin\Big(\frac{2\pi nz}{D}\Big)\Big].$$ Equation implies $a_0=0$. The functional in Eq. is minimized with respect to the fields $\theta(z)$ and $\varphi(z)$ and the coefficients $a_n$ and $b_n$, $n\in \{1,2,...,N\}$. The number of coefficients $N$ to be taken into account has to be chosen. It is reasonable to set the minimum wavelength in the Fourier series to be larger than the colloid diameter $d\sim 100$ nm. Therefore, $N$ has to be smaller than $N_\text{max}=[D/d\,]$ where $[x]$ denotes the integer part of $x$. For a slab of thickness $D=20$ $\mu$m one has $N_\text{max}=200$.
It turned out that for slab thicknesses $D<60\,\mathrm{\mu m}$ the equilibrium profile $\tau(z)$ obtained from Eq. takes negative values, i.e., $\delta\tau(z)<-1$, which contradicts its physical meaning $\tau(z)\sim|\mathbf{M}(z)|
\geq0$. This behavior is related to the absence of contributions in $\tau(z)$ beyond quadratic order. However, for slab thicknesses $D\geq 60$ $\mu$m the algorithm does provide the profiles $\theta(z)$ and $\varphi(z)$ together with a physically reasonable segregation profile $\delta\tau(z)$. Figure \[fig:Segregation\] shows the calculated profiles for $D= 120$ $\mu$m, equal walls with strong anchoring $W_\text{wall}=3.4\times10^{-5}$ J/m$^2$, coupling constant $\gamma=240$, and external magnetic field $B=4$ mT. It is evident that switching mechanism I is observed even in the presence of segregation effects. The density of magnetic colloids is largely reduced (with $\delta\tau(z)$ close to -1) in the regions of nonzero gradient of the director profile. An important difference to the case of infinitely strong coupling of the colloids to the liquid crystal (see Ref. [@1970_Brochard]) is the *depletion layer* being shifted away from the walls towards the interior of the sample. The functional form of the profiles $\varphi(z)$ and $\theta(z)$ obtained for asymmetric pairs of walls with strong and weak anchoring (see Fig. \[fig:TwoWalls\] in Sec. \[subsec:TwoWalls\]) is found at both walls for sufficiently thick slabs. We have performed a series of calculations for different slab thicknesses $D$ in order to determine for which thickness the pure switching mechanism I turns into a combination of mechanisms I and II (see Fig. \[fig:Segregation\]); we have found $D=95\pm 5$ $\mu$m. This observation opens the possibility to manipulate the sample magnetization in a manner similar to that described in Sec. \[subsec:TwoWalls\] but without the need to use a second, weakly anchored wall. However, the state of nonzero net magnetization is not necessarily saturated, but it might exhibit $\mathcal{M} < 1$. In order to have nonetheless a state with the magnetization $\mathcal{M}\approx 1$, one would need to adjust the system parameters (e.g., wall anchoring, elastic constant of the NLC etc.) such that each region of switched magnetization in the vicinity of the walls (see Fig. \[fig:Segregation\]) takes up $\approx 25\%$ of the slab thickness.
We have also performed a calculation for the case of two walls with equally weak anchoring. We found that segregation did not have a qualitative impact in that case either. Therefore we expect that the results of Secs. \[subsec:I\] and \[subsec:II\] are not affected qualitatively by segregation.
\[sec:Discussion\]Summary and conclusions
=========================================
In this analysis we have studied theoretically a ferronematic confined between two planar, parallel walls which impose an easy axis on the NLC director field. Inspired by the experimental studies reported in Refs. [@2013_Mertelj; @2014_Mertelj], the system is subjected to an external magnetic field. The ferronematic is an anisotropic polar fluid and thus the system is characterized by the relative directions of the NLC director, the easy axes due to the walls, the magnetization, and the external magnetic field. We have considered the situation in which the ferronematic is initially prepared with a uniform magnetization along the easy axis of the NLC. Subsequently an external magnetic field is applied in the direction opposite to the magnetization. This choice of the geometry reduces the theoretical description to an effectively one-dimensional one. The experiments reported in Refs. [@2013_Mertelj; @2014_Mertelj] showed that for such a setup there exists a critical external magnetic field $B_\text{cr}>0$ such that for magnetic field strengths $|\mathbf{B}|<B_\text{cr}$ the sample remains unperturbed. The authors of Ref. [@2013_Mertelj] also provided the expression in Eq. for the critical magnetic field strength in terms of the coupling $\gamma$ between the magnetization and the nematic director field. This critical magnetic field strength, which increases upon increasing $\gamma$, has been determined as that magnetic field strength for which the relaxation rate of long-wavelength fluctuations of the nematic director field vanish.
Here we study the system by numerical minimization of the corresponding free energy functional in Eq. . The numerical minimization is performed by using the Fletcher-Reeves-Polak-Ribiere general function minimization algorithm. It is obvious, that the global minimum of the free energy functional in Eq. before the external magnetic field has been applied is the initial state of the ferronematic being uniformly magnetized along the easy axis. Once the magnetic field is applied in the direction opposite to the initial magnetization, this state becomes only a local (metastable) minimum. The new global minimum is the ferronematic magnetized in the direction of the field. By means of a conjugate gradient algorithm one is able to search for the local minimum of the free energy and therefore to identify metastable states of the system. This is particularly useful in the present context, because this way one can investigate possible intermediate orientation profiles between the initial, now metastable, state with a uniform magnetization in the positive $x$-direction, i.e., opposite to the magnetic field $\mathbf{B}=B\mathbf{e}_x$ pointing into the negative $x$-direction, i.e., $B<0$, and the final stable state with the magnetization in the direction along the magnetic field, i.e., in negative $x$-direction ($B<0$).
Figure \[fig:Hyst\] shows the dependence of the dimensionless magnetization $\mathcal{M}$ (Eq. ) of the metastable state described above on the strength of the external magnetic field for particular values of the coupling constant $\gamma$ and of the anchoring strength $W_\text{wall}$ at a wall. One observes hysteresis of the magnetization for which a critical magnetic field strength $B_\text{cr}$ can be identified as the one for which significant deviations from the saturation magnetization $\mathcal{M}=1$ occur. One can distinguish several, qualitatively different, intermediate states (red circles in Fig. \[fig:Hyst\]). First, for a magnetic field $\mathbf{B}=B\mathbf{e}_x$, with component $B$ in the direction of the initial magnetization ($B>0$), the sample remains practically unperturbed for $B > B_\text{cr}$ ($B_\text{cr}<0$), i.e., for the magnetic field either along the initial magnetization ($B>0$) or sufficiently weak in the direction opposite to the initial magnetization ($B<0$) (see Fig. \[fig:MetaS\](1)). Upon decreasing the component $B$ of the magnetic field $\mathbf{B}=B\mathbf{e}_x$ further (i.e., making it less positive or more negative), the torque imposed on the NLC by the walls is no longer able to keep the ferronematic in the initial unperturbed state and thus the profiles become perturbed (see Fig. \[fig:MetaS\](2)). These states correspond to the brightening of the sample when viewed via crossed polarizers [@2013_Mertelj]. If one decreases the magnetic field component $B$ even further (i.e., making $B$ even more negative), one encounters the interesting metastable state shown in Fig. \[fig:MetaS\](3). In this state, near each wall a layer is formed within which the nematic director is close to the easy axis and the magnetization has inverted its direction, pointing along the external magnetic field (i.e., in negative $x$-direction). This flipping of the magnetization is energetically favorable for a sufficiently large strength $|\mathbf{B}|=|B|$ of the magnetic field pointing in the direction opposite to the initial magnetization (i.e., pointing into the negative $x$-direction), because the contribution to the free energy (Eq. ) of the coupling between the magnetization and the nematic director is invariant upon inversion of the magnetization, $\mathbf{M} \mapsto -\mathbf{M}$, but the contribution of the coupling between the magnetization and the external magnetic field is not. When the external magnetic field becomes even stronger (i.e., $B$ becomes even more negative and $|B|$ even larger), the regions of flipped magnetization expand into the interior of the system, eventually giving rise to the whole sample (except for thin layers in the very vicinity of the walls) being magnetized along the magnetic field. The qualitatively different scenario, which we refer to as switching mechanism II, occurs if the wall anchoring $W_\text{wall}$ is too weak to prevent the nematic director field from following the rotating magnetization. In accordance to Figs. \[fig:MetaS\](3) and \[fig:MetaS\](4), in this scenario the final stages of the switching are not realized. Figure \[fig:MetaSII\] illustrates how the final stages (3) and (4) of the switching mechanism II are realized, according to which the magnetization *and* the director field rotate in parallel.
Figure \[fig:Switch\] shows whether certain combinations of wall anchoring strengths $W_\text{wall}^{(1)}=W_\text{wall}^{(2)}=W_\text{wall}$ as well as of the coupling constant $\gamma$ lead to switching mechanism I or II. According to the map in Fig. \[fig:Switch\] the switching mechanism I is the dominant one in the experiments described in Refs. [@2013_Mertelj; @2014_Mertelj] ($W_\text{wall}^\text{exp}\approx 3.4\times 10^{-5}$ J/m$^2$).
The dependences of the critical magnetic field strength $B_\text{cr}$ on the coupling constant $\gamma$ and on the (equal) wall anchoring strength $W_\text{wall}$ are presented in Figs. \[fig:Phase\] and \[fig:Cuts\]. On one hand, $B_\text{cr}$ increases as function of $W_\text{wall}$ and, on the other hand, it also increases as function of $\gamma$, which is consistent with the results of Ref. [@2013_Mertelj]. In Fig. \[fig:Cuts\](b) a comparison of the critical magnetic field strength $B_\text{cr}$ as defined here with that introduced in Ref. [@2013_Mertelj] (see Eq. ) shows good agreement, although the two expressions involve different properties of the ferronematic.
Within a recently developed theory of ferronematics [@2018_Zarubin] one can relate the coupling coefficient $\gamma$ to the microscopic coupling $c$ (see below Eq. ) which depends on the size of the colloids in the suspension. Figure \[fig:Cuts\](c) shows the dependence of $B_\text{cr}$ on the microscopic coupling $c$ for a particular value of the wall anchoring strength $W_\text{wall}$. This allows one to vary the critical magnetic field by tuning the mean value of the size distribution of the colloids participating in the ferronematic.
Combining two walls with different anchoring strengths allows one to design a sample such that its switching mechanism is a superposition of type I and type II. The resulting nematic director and magnetization field profiles (see Figs. \[fig:TwoWalls\] and \[fig:TwoWallsVectors\]) are obtained by applying an external magnetic field. In turn, switching off this field divides the sample into two domains with opposite magnetizations (see Fig. \[fig:Storage\] (c)), rendering a sample with zero net magnetization. The initial state (i.e., the magnetized slab) can be restored by applying an external magnetic field of suitable direction to the two-domain sample. This cycle can be repeated arbitrarily, thus facilitating the switching between two states (magnetized/demagnetized slab) by using a uniform magnetic field only. This opens application perspectives such as storage of information and magnetic fields detection.
Similar controllable magnetic slabs can be constructed by using two walls with equally strong anchoring for samples of larger thickness ($D\geq 95$ $\mu$m for $\gamma=240$, $K=3.5\times 10^{-12}$ N, and $W_\text{wall}=3.4\times 10^{-5}$ J/m$^2$). We have found segregation to be *quantitatively* different from the case of walls with infinitely strong anchoring and of infinitely strong coupling of the colloids to the NLC (see Ref. [@1970_Brochard]). However, segregation effects do not affect the switching mechanisms qualitatively.
Conflicts of interest {#conflicts-of-interest .unnumbered}
=====================
There are no conflicts to declare.
\[sec:Appendix\]Constraint of the field $\tau(z)$
=================================================
In this Appendix we derive Eq. .
The spatially varying magnetization field $\mathbf{M}(z)$ was defined in Ref. [@2018_Zarubin] as $$\mathbf{M}(z)=\int\!\!{\ensuremath{\mathop{}\!\mathrm{d}}}^2\omega\,\, m{\boldsymbol{\omega}}\,\rho(z,{\boldsymbol{\omega}}),$$ where $m$ is the magnitude and ${\boldsymbol{\omega}}$ the direction of the magnetic moment of a single colloid and $\rho(z,{\boldsymbol{\omega}})$ is the number density of colloids in a layer around point $z$ and oriented in direction ${\boldsymbol{\omega}}$. We *assume* that in a small layer around a given point $z$ all individual magnetic moments point in one direction, i.e., the direction of $\mathbf{M}(z)$: $$\rho(z,{\boldsymbol{\omega}})=g(z)\delta({\boldsymbol{\omega}}-{\boldsymbol{\omega}}_0(z)),$$ with ${\boldsymbol{\omega}}_0(z):=\mathbf{M}(z)/|\mathbf{M}(z)|$ and $g(z)$ is the number density of colloids at point $z$ regardless of their orientation. From the definition of $\rho_\text{iso}$ it follows $$\begin{aligned}
\rho_\text{iso} =
\frac{1}{D}\int_0^D\!\!{\ensuremath{\mathop{}\!\mathrm{d}}}z\int\!\!{\ensuremath{\mathop{}\!\mathrm{d}}}^2\omega\,\,\rho(z,{\boldsymbol{\omega}}) =
\frac{1}{D}\int_0^D\!\!{\ensuremath{\mathop{}\!\mathrm{d}}}z \,\,g(z).
\label{eq:rhoisointg}\end{aligned}$$ Noting that $$\label{eq:AbsM}
|\mathbf{M}(z)|=mg(z)$$ and defining $\tau (z) := |\mathbf{M}(z)|/(m\rho_\text{iso}) = g(z)/\rho_\text{iso}$, Eq. can be written in the form of Eq. : $$\label{eq:IntG}
1 =
\frac{1}{D}\int_0^D{\ensuremath{\mathop{}\!\mathrm{d}}}z\,\,\frac{g(z)}{\rho_\text{iso}} =
\frac{1}{D}\int_0^D{\ensuremath{\mathop{}\!\mathrm{d}}}z\,\,\tau(z).$$ Using the definition $\delta\tau(z) := \tau(z) - 1$ (see Eq. ) this is equivalent to (see Eq. ) $$\int_0^D\!\!{\ensuremath{\mathop{}\!\mathrm{d}}}z\,\,\delta\tau(z)=0.$$
[4]{}
V. I. Zadorozhnii, A. N. Vasilev, V. Yu. Reshetnyak, K. S. Thomas, and T. J. Sluckin. *Nematic director response in ferronematic cells*. EPL **73**, 408 (2006).
V. I. Zadorozhnii, V. Yu. Reshetnyak, A. V. Kleshchonok, T. J. Sluckin, and K. S. Thomas. *Inverse Frederiks effect and bistability in ferronematic cells*. Mol. Cryst. Liq. Cryst. **475**, 221 (2007).
H. R. Brand and H. Pleiner. *Macroscopic behavior of ferronematic gels and elastomers*. Eur. Phys. J. E **37**, 122 (2014).
A. N. Zakhlevnykh and D. A. Petrov. *Weak coupling effects and re-entrant transitions in ferronematic liquid crystals*. J. Mol. Liq. **198**, 223 (2014).
A. N. Zakhlevnykh and D. A. Petrov. *Orientational bistability in ferronematic liquid crystals with negative diamagnetic anisotropy*. J. Magn. and Magn. Mat. **393**, 517 (2015).
A. N. Boychuk, D. V. Makarov and A. N. Zaklevnykh. *Dynamics of liquid-crystalline magnetic suspension in a rotating magnetic field*. Eur. Phys. J. E **39**, 101 (2016).
A. N. Zakhlevnykh and D. A. Petrov. *Orientational transitions in antiferromagnetic liquid crystals*. Phys. Solid State **58**, 1906 (2016).
A. N. Zakhlevnykh, M. S. Lubnin and D. A. Petrov. *A simple model of liquid-crystalline magnetic suspension of anisometric particles*. J. Magn. and Magn. Particles **431**, 62 (2017).
N. Podoliak, O. Buchnev, O. Buluy, G. D’Alessandro, M. Kaczmarek, Y. Reznikov, and T. J. Sluckin. *Macroscopic optical effects in low concentration ferronematics*. Soft Matter **7**, 4742 (2011).
A. Mertelj, D. Lisjak, M. Drofenik, and M. Copic. *Ferromagnetism in suspensions of magnetic platelets in liquid crystal*. Nature **504**, 237 (2013).
A. Mertelj, N. Osterman, D. Lisjak and M. Copic. *Magneto-optic and converse magnetoelectric effects in a ferromagnetic liquid crystal*. Soft Matter **10**, 9065 (2014).
A. J. Hess, Q. Liu and I. Smalyukh. *Optical patterning of magnetic domains and defects in ferromagnetic liquid crystal colloids*. Appl. Phys. Lett. **107**, 071906 (2015).
A. Mertelj and D. Lisjak. *Ferromagnetic nematic liquid crystals*. Liq. Cryst. Rev. **5**, 1 (2017).
D. Manaila Maximean. *New grafted ferrite particles/liquid crystal composite under magnetic field*. J. Magn. and Magn. Mat. **452**, 343 (2018).
G. Zarubin, M. Bier and S. Dietrich. *Effective Landau theory of ferronematics*. J. Chem. Phys. **149**, 054505 (2018).
T. Potisk, D. Svensek, H. R. Brand, H. Pleiner, D. Lisjak, N. Osterman, and A. Mertelj. *Dynamic magneto-optic coupling in a ferromagnetic nematic liquid crystal*. Phys. Rev. Lett. **119**, 097802 (2017).
T. Potisk, A. Mertelj, N. Sebastian, N. Osterman, D. Lisjak, H. R. Brand, H. Pleiner, and D. Svensek. *Magneto-optic dynamics in a ferromagnetic nematic liquid crystal*. Phys. Rev. E **97**, 012701 (2018).
N. Sebastian, N. Osterman, D. Lisjak, M. Copic, and A. Mertelj. *Director reorientation dynamics of ferromagnetic nematic liquid crystals*. Soft Matter, Advance article, DOI: 10.1039/C8SM01377B (2018).
Segregation effects can be neglected if the segregation parameter $\kappa:=
D/l$ is much larger than unity [@2014_Zakhlevnykh], where $l:=\sqrt{K/(k_\text{B}T\rho_\text{iso})}$ is the segregation length [@1970_Brochard]; for $K$ see Eq. . In the present study one has $\kappa\approx 7$.
The NLC in the experiment analyzed in Ref. [@2013_Mertelj] exhibits twist deformations in the $x$-$y$-plane because the twist elastic constant is the smallest Frank modulus of that system. Here we consider the one-elastic-constant approximation so that the system can undergo another type of deformation (i.e., bending) instead. Except for exchanging “twist” by “bend” deformations the results are equivalent because the underlying mathematical model is identical.
W. H. Press, S. A. Teukolski, W. T. Vetterling, and B. P. Flannery. *Numerical recipes in C++. Second edition*. Cambridge University Press (2002).
The results of our study are valid only if the *collective* response mode [@1970_Brochard] of the ferronematic is realized. The minimum number density of magnetic colloids which facilitates collective behavior is [@1995_Burylov] $$\rho_\text{coll}=\frac{1}{cD^2R}$$ which gives $\approx 10^{18}$ m$^{-3}$ for $c=0.03$ (i.e., roughly the microscopic coupling constant in the experiment [@2013_Mertelj; @2014_Mertelj]), $D=20$ $\mu$m, and $R=100$ nm. Due to $\rho_\text{iso}=1.5\times 10^{19}$ m$^{-3}>\rho_\text{coll}$ and because $D=20$ $\mu$m is the minimum thickness used here, the assumption of occurrence of collective behavior is valid.
S. V. Burylov and Y. L. Raikher. *Macroscopic properties of ferronematics caused by orientational interactions on the particle surfaces. I. Extended continuum model*. Mol. Cryst. Liq. Cryst. **258**, 107 (1995).
The positive constant $a$ in Eq. is unimportant as long as $|\mathbf{M}|=m\rho_\text{iso}=\text{const}$. It becomes relevant otherwise, i.e., if segregation is taken into account (see Sec. \[sec:Segregation\]).
In the case of soft anchoring the interaction of the NLC with the wall surface is described by the free energy potential $$F_\text{s}=W\int_\mathcal{\partial\mathcal{V}}\!\!{\ensuremath{\mathop{}\!\mathrm{d}}}\mathbf{s}\,[\mathbf{n}(\mathbf{s})\times{\boldsymbol{\nu}}(\mathbf{s})]^2,$$ where $W>0$ is the anchoring strength at the wall surface $\partial\mathcal{V}$, and ${\boldsymbol{\nu}}(\mathbf{s})$ is the easy axis direction at the surface point $\mathbf{s}$. In the case of a planar wall surface with constant easy axis the above expression reduces to the form used in Eq. . The soft anchoring at the surface of the colloids is taken into account implicitly within the derivation of the free energy density given in Eq. [@2018_Zarubin].
F. Brochard and P. G. de Gennes. *Theory of magnetic suspensions in liquid crystals*. J. Physique **31**, 691 (1970).
The two-domain state required us to tune the elastic constant $K$, because for certain values of $K$ the minimization algorithm produced throughout the sample a perturbed director profile for $B=0$ as a favorable state. In order to make the corresponding elastic distortions more costly, we increased the value of $K$ to $9\times 10^{-12}$ N compared to $K=3.5\times 10^{-12}$ N used in Secs. \[subsec:I\], \[subsec:II\], and \[sec:Segregation\].
The position of the interface depends on the external field *strength* which should be chosen such that the interface is located approximately in the middle of the slab. We used the values $B=12$ mT together with $\gamma\approx 240$, $W_\text{wall}^{(1)}=1\times 10^{-5}$ J/m$^2$, $W_\text{wall}^{(2)}=0.1\times 10^{-5}$ J/m$^2$, and $K=9\times 10^{-12}$ N.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We investigate which three dimensional near-horizon metrics $g_{NH}$ admit a compatible 1-form $X$ such that $(X, [g_{NH}])$ defines an Einstein-Weyl structure. We find explicit examples and see that some of the solutions give rise to Einstein-Weyl structures of dispersionless KP type and dispersionless Hirota (aka hyperCR) type.'
address: |
Department of Mathematics and Statistics\
Faculty of Science, Masaryk University\
Kotlářská 2, 611 37 Brno\
Czech Republic
author:
- Matthew Randall
title: 'Three dimensional near-horizon metrics that are Einstein-Weyl'
---
Introduction
============
Let $M^3$ be a three dimensional smooth manifold equipped with a Lorentzian metric. A near-horizon metric on $M^3$ is a Lorentzian metric of the form $$\label{nhg}
g_{NH}=2 {{\rm d}}\nu \left( {{\rm d}}r+r h(x) {{\rm d}}x+\frac{r^2}{2}F(x) {{\rm d}}\nu\right)+{{\rm d}}x {{\rm d}}x,$$ where $x$, $\nu$ and $r$ are local coordinates and $h(x)$, $F(x)$ are arbitrary functions of $x$. Near-horizon geometries in higher dimensions are studied in relation to the existence of extremal black holes [@dgs], [@KL3], [@lp], [@lrs], [@lsw].
The near-horizon metric (\[nhg\]) is derived as follows. For a smooth null hypersurface $\Si^2$ in $M^3$, there exists an adapted coordinate system called Gaussian null coordinates valid in any neighbourhood of $\Si^2$. Imposing that the normal vector field $N^a$ be Killing in $M^3$ implies that $\Si^2$ is a Killing horizon. We further require that the Killing horizon $\Si^2$ be degenerate. This means that the flow of $N^a$ is along affinely parameterised null geodesics on $\Si^2$, or equivalently that the surface gravity of $\Si^2$ is zero. The scaling limit of the degenerate Killing horizon is called a near-horizon geometry, and the metric is of the form (\[nhg\]).
In dimensions 4 or more, imposing the vacuum Einstein equations on the near-horizon geometry metric $g_{NH}$ give rise to the near-horizon geometry equations on the spatial section of the degenerate Killing horizon. Solutions to this overdetermined system of equations on a compact cross-section are related to the existence of extremal black holes. For further details, see [@KL0], [@KL3] and [@lp].
Similarly, we can require that a 3 dimensional near-horizon geometry metric satisfies Einstein’s equations. This has been done in [@KL3]. In this case, the near-horizon geometry equations reduce to a pair of first order ODEs that can be integrated explicitly. It is also found that global and periodic solutions for $h(x)$ exist on a 1-dimensional cross-section $\cH$ which has the topology of the circle; this is the 1-dimensional analogue of an extremal black hole horizon.
Instead of asking Einstein’s equations to be satisfied, we can ask for a natural generalisation which is to impose the Einstein-Weyl equations instead. This requires an additional structure of a 1-form, to be explained in the next section. Here we investigate whether there exists a 1-form $X$ compatible with (\[nhg\]) such that $(X,g_{NH})$ satisfies the Einstein-Weyl equations.
In the 37th winter school in geometry and physics, held in Srní, Czech Republic in 2017, plenary lectures on non-expanding horizons including near horizon geometries were presented by Jerzy Lewandowski and lectures on Einstein-Weyl geometry and dispersionless integrable systems were presented by Maciej Dunajski. The latter topics were also covered in the lectures by Evgeny Ferapontov. We hope that attendees of the school would find these topics to be interesting and delve into the rich and steadily expanding field that relates integrable PDEs to general relativity and classical differential geometry.
Three dimensional Einstein-Weyl geometries
==========================================
Einstein-Weyl geometries in 3 dimensions play an important role because of its relationship with the geometry of third order ODEs (see [@Nurowski] and [@Tod]), twistor theory (see [@Hitchin] and [@holodisk]) and integrable systems (see [@grass], [@ck], [@DFK] and [@FerKrug]). Let $(M^3,[g])$ be a smooth conformal manifold equipped with a conformal class of metrics $[g]$ of (pseudo)-Riemannian signature. In 3 dimensions, this is either Riemannian or Lorentzian. Any 2 representative metric $g$, $\tilde g \in [g]$ are related via $$\tilde g= \Om^2 g$$ for some smooth positive function $\Om$. A Weyl structure on $(M^3,[g])$ is a torsion-free connection $D$ that preserves the conformal class of metrics. Equivalently, $$D_ag_{bc}=2 X_a g_{bc}$$ for $g_{ab}$ a representative in the conformal class and $X_a$ a 1-form that is not necessarily closed. The Einstein-Weyl equations are the system of equations obtained by requiring that that the symmetric part of the Ricci tensor of the Weyl connection is pure trace, i.e.$$R^{D}_{(ab)}=s g_{ab}$$ for some function $s$. This system of equations is conformally invariant. Writing $R^D_{(ab)}$ in terms of the Levi-Civita connection ${\nabla}$ for the metric $g_{ab}$ and 1-form $X_a$, we get $${\nabla}_{(a}X_{b)}+X_aX_b+\P_{ab}=\La g_{ab},$$ where round brackets denote symmetrisation over indices, $\P_{ab}$ is the Schouten tensor of $g_{ab}$ and the trace term $$\La=\frac{1}{3}\left({\nabla}_aX^b+X_aX^b+\P\right)$$ is a function (more appropriately, a section of a density valued line bundle and we refer to [@conformal] and [@ET] for more details).
In [@Cartan], it is shown that a 3-dimensional Lorentzian Einstein-Weyl structure corresponds to a 2 parameter family of totally-geodesic null hypersurfaces in $M^3$. A twistor description is developed in [@Hitchin], where it is shown that real-analytic Lorentzian 3-dimensional Einstein-Weyl structures arise precisely from the Kodaira deformation space of rational normal curves with normal bundle $\cO(2)$ in a 2-dimensional complex manifold. The 2-form ${{\rm d}}X$ is called the Faraday 2-form of the Weyl structure. If ${{\rm d}}X=0$, then $X$ is locally exact and $g$ can be conformally rescaled to a metric of constant curvature in 3 dimensions. In particular, it implies that $g$ is locally conformally flat. A computation of the Cotton tensor of the near-horizon metric $g_{NH}$ shows that
The metric of the form (\[nhg\]) is locally conformally flat iff $$\label{cotton}
F'(x)=F(x) h(x).$$
Condition (\[cotton\]) implies that the Cotton tensor of $g_{NH}$ is zero, and conversely so.
Results
=======
We shall assume that $(M^3,[g_{NH}])$ is smooth. We consider an ansatz for $X$ of the form $$X=c h(x){{\rm d}}x+X_2 {{\rm d}}r +X_3 {{\rm d}}v,$$ where $c$ is a constant and the functions $X_2=X_2(x,\nu,r)$, $X_3=X_3(x,\nu,r)$ are to be determined. We also require that $X|_{r=0}=c h(x) {{\rm d}}x$, so that the 1-form $X$ is determined up to a constant multiple $c$ by its value on the spatial section of the degenerate Killing horizon. In the following analysis, it turns out that there is a degenerate case when $c=-\frac{1}{2}$. We have the following:
\[nhg-ew-0\] A 3 dimensional near-horizon geometry metric (\[nhg\]) on $M^3$ of the form $$\begin{aligned}
\label{weierstrass}
g_{NH}=&2 {{\rm d}}\nu \left( {{\rm d}}r+r h(x) {{\rm d}}x+\frac{r^2}{2}e^{{{\rmintop\nolimits}h(x) dx}}\wp\left({\rmintop\nolimits}e^{{\frac{1}{2}{\rmintop\nolimits}h(x) dx}}dx+a;0,b\right){{\rm d}}\nu\right)\nonumber\\
&+{{\rm d}}x {{\rm d}}x\end{aligned}$$ where $\wp(z;g_2,g_3)$ is the Weierstrass elliptic function, $a$, $b$ are constants, and a Weyl connection $X$ of the form $$X=-\frac{1}{2}h(x) {{\rm d}}x-2r e^{{{\rmintop\nolimits}h(x) dx}}\wp\left({\rmintop\nolimits}e^{{\frac{1}{2}{\rmintop\nolimits}h(x) dx}}dx+a;0,b\right) {{\rm d}}\nu$$ defines an Einstein-Weyl structure $([g_{NH}],X)$ on $M^3$. This depends on $1$ free function of one variable.
In particular, taking $h(x)$ to be globally defined and periodic allows $\cH$ to have the topology of a circle. In the case that $h(x)=0$, the metric simplifies to $$g=2 {{\rm d}}\nu \left( {{\rm d}}r+\frac{r^2}{2}\wp(x+a;0,b){{\rm d}}\nu\right)+{{\rm d}}x {{\rm d}}x$$ and the 1-form $X$ is given by $$X=-2r \wp(x+a;0,b) {{\rm d}}\nu.$$ We recognise that the function $$u(x,\nu,r)=-\frac{r^2}{2}\wp(x+a;0,b)$$ satisfies the dispersionless Kadomtsev-Petviashvili (dKP) equation given by $$2(u_{\nu}-u u_r)_r=u_{xx}$$ and the Einstein-Weyl structure corresponds to one of dKP type. For further details about such Einstein-Weyl structures we refer to [@dKP]. For generic values of $c$, which also allows for $c=-\frac{1}{2}$, we have
\[nhgewgen\] A 3 dimensional near-horizon geometry metric (\[nhg\]) on $M^3$ and a Weyl connection $X$ of the form $$X=c h(x) {{\rm d}}x+r ((2c+1) h'+c(2c+1) h^2-2 F(x)) {{\rm d}}v$$ defines an Einstein-Weyl structure if and only if $$F(x)=\frac{ h''+4 c h h'+2 c^2 h^3}{2 h}$$ and $h(x)$ satisfies the $4^{\rm th}$ order ODE $$\begin{aligned}
\label{4thode}
&h^3 (h')^2(c-1)^2-\frac{1}{2}(c-1)^2h^4 h''+\frac{9}{4}(c-1)h^2 h' h''\nonumber\\
&-\frac{3}{4}(c-1) h^3 h'''-\frac{1}{2}(h')^2h''+\frac{1}{2}h h' h'''+h(h'')^2-\frac{1}{4}h^2 h''''=0. \end{aligned}$$
Proof
=====
We start with an ansatz of the form $$X=c h(x){{\rm d}}x+X_2 {{\rm d}}r +X_3 {{\rm d}}\nu,$$ where the functions $X_2=X_2(x,\nu,r)$, $X_3=X_3(x,\nu,r)$ are to be determined, and we require that $X|_{r=0}=c h(x) dx$, so that the 1-form $X$ is determined by its value on the spatial section of the degenerate Killing horizon. We find that substituting this ansatz for $X$ into the Einstein-Weyl equations, the ${{\rm d}}r {{\rm d}}r$ component gives $$\partial_r X_2+X_2^2=0,$$ which has solutions $$X_2=\frac{1}{r+f_1(x,\nu)} \qquad \mbox{or} \qquad X_2=0.$$ Since the first solution does not restrict to zero on $\{r=0\}$, we take $X_2=0$ instead. The ${{\rm d}}x {{\rm d}}x$ component gives now $$-2F-\partial_r X_3+c(2c+1) h^2+(2 c+1)h'=0,$$ from which we obtain $$X_3=(-2F+c(2c+1)h^2+(2c+1)h')r+f_2(x,\nu).$$ Once again requiring that $X_3|_{r=0}=0$ implies $f_2(x,\nu)=0$. With this, only the ${{\rm d}}x {{\rm d}}\nu$ and ${{\rm d}}\nu {{\rm d}}\nu$ components remain to be solved in the Einstein-Weyl equations. The ${{\rm d}}x {{\rm d}}\nu$ component gives $$\frac{r}{2}(2c+1)\left(h''-2 h F+4c h h'+2c^2 h^3\right)=0.$$ This vanishes identically when $c=-\frac{1}{2}$. Otherwise, we have $$F(x)=\frac{h''+4 c h h'+2 c^2 h^3}{2 h}.$$ In the first case where $c=-\frac{1}{2}$, the remaining equation in the ${{\rm d}}\nu {{\rm d}}\nu$ component is $$-3 F h^2+5 h F'+2 F h'+12 F^2-2 F''=0,$$ which has solutions $$F(x)=\wp\left({\rmintop\nolimits}\exp\left({\frac{1}{2}{\rmintop\nolimits}h(x) dx}\right) dx+a,0,b\right)\exp\left({{\rmintop\nolimits}h(x) dx}\right)$$ whatever $h(x)$ is. This proves Theorem \[nhg-ew-0\]. For the other case, substituting $$F(x)=\frac{h''+4 c h h'+2 c^2 h^3}{2 h}$$ gives $$X_3=\frac{c h^3+(1-2 c) h h'-h''}{h}r$$ and the only remaining ${{\rm d}}\nu {{\rm d}}\nu$ component of the Einstein-Weyl equations gives the $4^{\rm th}$ order ODE $$\begin{aligned}
&h^3 (h')^2(c-1)^2-\frac{1}{2}(c-1)^2h^4 h''+\frac{9}{4}(c-1)h^2 h' h''\\
&-\frac{3}{4}(c-1) h^3 h'''-\frac{1}{2}(h')^2h''+\frac{1}{2}h h' h'''+h(h'')^2-\frac{1}{4}h^2 h''''=0. \end{aligned}$$ Note that this formula is still well-defined for $c=-\frac{1}{2}$. This proves Theorem \[nhgewgen\].
Explicit solutions
==================
In this section, we find interesting families of solutions to the 4th order ODE (\[4thode\]). Consider the following second order nonlinear ODE $$\begin{aligned}
\label{2ndh}
h''=\alpha h h'+\beta h^3\end{aligned}$$ with $\al$, $\beta$ constant. We find, upon substituting (\[2ndh\]) and its derivatives into (\[4thode\]), that we obtain $$\begin{aligned}
-\frac{h^3}{4}\big(2(c-1)^2+3\al(c-1)+\al^2-\beta\big)\left(\beta h^4+\al h^2h'-2(h')^2\right)=0.\end{aligned}$$ Thus solutions to (\[2ndh\]) automatically satisfy (\[4thode\]) provided $$\label{abc}
\beta=2(c-1)^2+3\al(c-1)+\al^2.$$ Switching independent and dependent variables $(x, h(x)) \mapsto (h, x(h))$, the non-linear second order ODE (\[2ndh\]) is dual to $$\begin{aligned}
x''=-\beta h^3 (x')^3-\al h(x')^2,\end{aligned}$$ which upon setting $y(h)=h^2 x'(h)$, gives an Abel differential equation of the first kind: $$\begin{aligned}
y'=\frac{1}{h}\left(-\beta y^3-\al y^2+2 y\right).\end{aligned}$$ This has solutions given by $$\begin{aligned}
\label{hy}
h=\frac{\ga\sqrt{y}\exp\left(\frac{\al}{2\sqrt{\al^2+8\beta}}\tanh^{-1}\left(\frac{2\beta y+\al}{\sqrt{\al^2+8\beta}}\right)\right)}{(\beta y^2+\al y-2)^{\frac{1}{4}}},\end{aligned}$$ and consequently $x$ viewed as a function of $y$ is given by $$\begin{aligned}
\label{xy}
x(y)=-\frac{1}{\ga}{\rmintop\nolimits}\frac{\exp\left(\frac{-\al}{2\sqrt{\al^2+8\beta}}\tanh^{-1}\left(\frac{2\beta y+\al}{\sqrt{\al^2+8\beta}}\right)\right)}{\sqrt{y}(\beta y^2+\al y-2)^{\frac{3}{4}}}{{\rm d}}y.\end{aligned}$$
When $\al=0$, we obtain $\beta=2(c-1)^2$ from (\[abc\]). In this case, solutions to the nonlinear ODE (\[2ndh\]) $$h''=\beta h^3=2(c-1)^2 h^3$$ are given by the Jacobi elliptic function $$h(x)=a~{\rm sn}\left( a\left(\frac{\sqrt{-2 \beta}}{2}x +b\right), i\right)=a~{\rm sn}\left( a(i(c-1)x +b), i\right),$$ with $a$, $b$ the constants of integration. Here the elliptic modulus is given by $i=\sqrt{-1}$. Alternatively, we obtain $$\begin{aligned}
h=\frac{\ga \sqrt{y}}{(\beta y^2-2)^{\frac{1}{4}}}\end{aligned}$$ from (\[hy\]) and $$\begin{aligned}
x=-\frac{1}{\ga}{\rmintop\nolimits}\frac{1}{\sqrt{y}(\beta y^2-2)^{\frac{3}{4}}}{{\rm d}}y\end{aligned}$$ from (\[xy\]). Passing to $y=\left(\frac{2}{\beta z}\right)^{\frac{1}{2}}$, this gives the expression in terms of hypergeometric functions $$\begin{aligned}
h=\frac{\ga}{\beta^{\frac{1}{4}}(1-z)^{\frac{1}{4}}} \quad \text{and} \quad
x=\frac{\sqrt{2z}}{2\ga \beta^{\frac{1}{4}}}{}_2F_1\left(\frac{1}{2},\frac{3}{4};\frac{3}{2};z\right).\end{aligned}$$
When $\beta=0$, we have $$(2c-2+\al)(c-1+\al)=0$$ from (\[abc\]) and therefore $\al=1-c$ or $\al=2-2c$. For $\al \neq 0$, which implies $c \neq 1$, solutions to (\[2ndh\]) are given by $$h=\frac{1}{\al}\tan\left(\frac{1}{2}\sqrt{2 \ell \al}(x+b)\right)\sqrt{2\ell \al}.$$ This is periodic but not globally defined. For $\al=0$, or equivalently $c=1$, solutions to (\[2ndh\]) are given by $$h=\ell x+b$$ with $b$, $\ell$ the constants of integration. The degenerate Killing horizon in this case has the topology of the real line and the metric is not conformally flat. If we do not assume either $\al$ or $\beta$ is zero, and consider the case when $c=1$, then equation (\[2ndh\]) is $$\begin{aligned}
\label{habel}
h''=\al h h'+\al^2 h^3.\end{aligned}$$ From (\[hy\]), we obtain $$\begin{aligned}
h=\frac{\ga\sqrt{y}}{(-1)^{\frac{1}{4}}(\al y+2)^{\frac{1}{6}}(1-\al y)^{\frac{1}{3}}}.\end{aligned}$$ We consider real solutions by taking $\ga=(-1)^{\frac{1}{4}}\varepsilon$. This gives $$\begin{aligned}
h^6=\frac{\varepsilon y^3}{(\al y +2) (1-\al y)^2},\end{aligned}$$ which implies that $y$ is a solution of the cubic equation $$\varepsilon y^3=h^6(\al y+2)(1-\al y)^2.$$ In the limit $\varepsilon \to 0$, we have $y=-\frac{2}{\al}$ or $y=\frac{1}{\al}$. In these cases we obtain $$x(h)=-{\rmintop\nolimits}\frac{2}{\al h^2}{{\rm d}}h+b=\frac{2}{\al h}+b \mbox{~and~} x(h)={\rmintop\nolimits}\frac{1}{\al h^2}{{\rm d}}h+b=-\frac{1}{\al h}+b.$$ Hence $$h(x)=\frac{2}{\al(x-b)} \mbox{~and~} h(x)=-\frac{1}{\al(x-b)}$$ satisfy (\[4thode\]) with the parameter $c=1$. For these solutions $h(x)$ is singular on the degenerate Killing horizon. These solutions give conformally flat metrics when $\al$ is chosen so that $h(x)=\frac{-2}{x-b}$ or $\frac{1}{x-b}$. In fact when $c=1$, (\[4thode\]) can be integrated to give $$\begin{aligned}
\label{3rdode1}
-\frac{1}{4}h^2 h'''+h h' h''-\frac{1}{2}(h')^3=0.\end{aligned}$$ Passing to $h(x)={\rm e}^{f(x)}$, we see that (\[3rdode1\]) is satisfied iff $f(x)$ satisfies the non-linear $3^{\rm rd}$ order ODE $$\label{nlode}
f'''-f' f''-(f')^3=0.$$ This ODE is none other than (\[habel\]) with $f'$ replacing $h$ and $\al=1$. This gives the following additional solutions to (\[4thode\]) $$h(x)=(x-b)^2 \mbox{~and~} h(x)=\frac{1}{x-b}.$$ The conformal structure for the first $h(x)$ is not flat and $h(x)$ is globally defined on the line horizon, while the conformal structure for the second is flat, which can be rescaled to a metric of constant scalar curvature.
Interestingly, we can relate the solutions to (\[2ndh\]) with $c=-1$, $\al=2$, $\beta=0$ to dispersionless Hirota type or hyperCR Einstein-Weyl structures. For a review of such structures see [@hyperCR], [@hyperCR2]. Recall that a metric of the form $$\begin{aligned}
g=&({{\rm d}}x+H_r d\nu)^2-4\left({{\rm d}}r-H_x {{\rm d}}v\right) {{\rm d}}\nu\\
=&2{{\rm d}}\nu\left(-2{{\rm d}}r+H_{r}{{\rm d}}x+(2H_x+\frac{H_r^2}{2}){{\rm d}}\nu\right)+{{\rm d}}x {{\rm d}}x \end{aligned}$$ where $H=H(x,\nu,r)$ and a Weyl connection of the form $$X=\frac{1}{2}H_{rr}{{\rm d}}x+\frac{1}{2}(H_{r}H_{rr}+2H_{xr}) {{\rm d}}\nu$$ defines a hyperCR Einstein-Weyl structure iff the hyperCR equation $$\begin{aligned}
\label{hypercr}
H_{x}H_{rr}-H_rH_{xr}-H_{xx}+H_{r\nu}=0\end{aligned}$$ is satisfied. A particular family of solutions to this equation can be found: $$H(x,\nu,r)=j\tanh^3\left(\frac{a^2}{b}r+b \nu+ax+e\right)+k \tanh\left(\frac{a^2}{b}r+b \nu+ax+e\right)+l$$ is a $6$ parameter family of solutions satisfying (\[hypercr\]) depending on constants $a$, $b$, $e$, $j$, $k$, $l$. Aligning the 1-form in the hyperCR case with our ansatz for $X$, we require that $$X|_{r=0}=\frac{1}{2}H_{rr}{{\rm d}}x+\frac{1}{2}(H_{r}H_{rr}+2H_{xr}) {{\rm d}}\nu|_{r=0}=c h(x){{\rm d}}x.$$ Hence $$H=c h(x) r^2+f_2(x,\nu)$$ for some function $h(x)$ and $f_2(x,\nu)$. Additionally, to align the hyperCR metric with the near-horizon metric (\[nhg\]), we require $f_2(x,\nu)=0$. Plugging this solution for $H$ back into the metric and 1-form $X$, we discover that the Einstein-Weyl equations are satisfied iff $$\begin{aligned}
h''=&-2 c h h'.
$$ This has solutions given by $$h(x)=\frac{\sqrt{c \ell}}{c}\tanh\left(\sqrt{c \ell}(x+b)\right)$$ for $c \neq 0$. Redefining coordinates $r \mapsto \tilde r=-2r$ and renaming $r$, we have
\[hypercr1\] A 3 dimensional Lorentzian metric on $M^3$ of the form $$\label{tan}
g=2 {{\rm d}}\nu \left( {{\rm d}}r-c h r {{\rm d}}x+\frac{r^2}{2}(c h'+c^2 h^2){{\rm d}}\nu\right)+{{\rm d}}x {{\rm d}}x$$ where $c \neq 0$ with $$h(x)=\frac{\sqrt{c \ell}}{c}\tanh\left(\sqrt{c \ell}(x+b)\right)$$ satisfying the ODE $$h''=-2 c h h'$$ and a Weyl connection $X$ of the form $$\begin{aligned}
X=&c h {{\rm d}}x- c r (c h^2+h'){{\rm d}}\nu\end{aligned}$$ defines a hyperCR Einstein-Weyl structure on $M^3$. Comparing the metric of the form (\[tan\]) to the near horizon metric (\[nhg\]), we see that setting $c=-1$ gives the near-horizon metric $$g_{NH}=2 {{\rm d}}\nu \left( {{\rm d}}r+ h r {{\rm d}}x+\frac{r^2}{2}(- h'+h^2){{\rm d}}\nu\right)+{{\rm d}}x {{\rm d}}x$$ and Weyl connection $$\begin{aligned}
X=&-h {{\rm d}}x+ r (-h^2+h'){{\rm d}}\nu.\end{aligned}$$
In particular, for $c=-1$, the ODE $h''=2 h h'$ that $h(x)$ satisfies agrees with the solution (\[2ndh\]) to (\[4thode\]) with the parameters $\beta=0$, $c=-1$, $\al=1-c=2$.
Further remarks and outlook
===========================
It is curious that solving for the Einstein-Weyl equations for a near-horizon metric (\[nhg\]) give rise to dKP Einstein-Weyl structures for $c=-\frac{1}{2}$ as in Theorem \[nhg-ew-0\] and hyperCR Einstein-Weyl structures for $c= -1$ as in Theorem \[nhgewgen\] and Proposition \[hypercr1\]. Do other parameters of $c$ give rise to other interesting EW structures? The metric in Gaussian null coordinates on $M^3$ with a smooth Killing horizon (see [@KL3]) is given by $$\label{dk}
g=2 {{\rm d}}\nu \left( {{\rm d}}r+r h(x,r) {{\rm d}}x+\frac{r}{2}F(x,r) {{\rm d}}\nu\right)+\gamma(x,r)^2 {{\rm d}}x {{\rm d}}x.$$ In [@descendants], the authors investigated the case where (\[dk\]) is Einstein with non-zero cosmological constant. Similarly, we can investigate whether the metric of the form (\[dk\]) admits a compatible 1-form $X_a$ such that $([g],X)$ is Einstein-Weyl. Some computations have been made but the calculations are considerably more involved than those presented here.
[XX]{}
B. Doubrov, E. V. Ferapontov, B. Kruglikov and V. Novikov, *On the integrability in Grassmann geometries: integrable systems associated with fourfolds $Gr(3,5)$*, arXiv:1503.02274.
D. M. J. Calderbank and B. Kruglikov, *Integrability via geometry: dispersionless differential equations in three and four dimensions*, arXiv:1612.02753.
E. Cartan, *Sur une classe d’espaces de Weyl*, Ann. Sci. École Norm. Sup. (3) [**60**]{} (1943) 1–16.
M. Dunajski, E.V. Ferapontov and B. Kruglikov, *On the Einstein-Weyl and conformal self-duality equations*, Journal of Mathematical Physics [**56**]{} (8) (2015).
M. Dunajski, J. Gutowski, W. Sabra, *Einstein-Weyl Spaces and Near-Horizon Geometry*, arXiv:1610.08953.
M. Dunajski, W. Kryński, *Einstein-Weyl geometry, dispersionless Hirota equation and Veronese webs*, Math. Proc. Cambridge Philos. Soc. [**157**]{} 1 (2014), 139–150.
M. Dunajski, W. Kryński, *Point invariants of third-order ODEs and hyper-CR Einstein-Weyl structures*, J. Geom. Phys. [**86**]{} (2014), 296–302.
M. Dunajski, L.J. Mason, and K.P. Tod, *Einstein-Weyl geometry, the dKP equation and twistor theory*, J. Geom. Phys. [**37**]{} (2001), 63–93.
M. G. Eastwood, *Notes on conformal differential geometry*, Supp. Rend. Circ. Matem. Palermo [**43**]{} (1996), 57–76.
M. G. Eastwood and K. P. Tod, *Local constraints on Einstein-Weyl geometries*, J. Reine Angew. Math., [**491**]{} (1997), 183–198.
E. V. Ferapontov and B. Kruglikov, *Dispersionless integrable systems in 3D and Einstein-Weyl geometry*, J. Diff. Geom. [**97**]{} (2014), 215–254.
N. J. Hitchin, *Complex manifolds and Einstein’s equations*, Twistor geometry and nonlinear systems (Primorsko, 1980), 73–99, Lecture Notes in Math. [**970**]{}, Springer, Berlin-New York (1982).
H. K. Kunduri and J. Lucietti, *A classification of near-horizon geometries of extremal vacuum black holes*, Journal of Mathematical Physics [**50**]{}, 082502, 2009, 41pp.
H. K. Kunduri and J. Lucietti, *Classification of Near-Horizon Geometries of Extremal Black Holes*, Living Rev. Relativity [**16**]{}, 8 , 2013.
C. LeBrun and L. J. Mason, *The Einstein-Weyl equations, scattering maps, and holomorphic disks*, Math. Res. Lett. [**16**]{}, (2009), 291–301.
J. Lewandowski, T. Pawlowski, *Extremal Isolated Horizons: A Local Uniqueness Theorem*, Class. Quantum Grav. [**20**]{} (2003), 587–-606, arXiv:gr-qc/0208032.
J. Lewandowski, I. Racz, A. Szereszeweski, *Near Horizon Geometries and Black Hole Holograph*, arXiv:1701.01704.
J. Lewandowski, A. Szereszeweski, P. Waluk, *When Isolated Horizons met Near Horizon Geometries*, 2nd LeCosPA Symposium Proceedings, “Everything about Gravity" celebrating the centenary of Einstein’s General Relativity, December 14-18 (2015), Taipei, arXiv:1602.01158.
C. Li and J. Lucietti, *Three-dimensional black holes and descendants*, Physics Letters B [**738**]{}, (2014), 48–54.
P. Nurowski, *Differential equations and conformal structures*, J. Geom. Phys. [**55**]{}, no. 1 (2005), 19–49.
K. P. Tod, *Einstein-Weyl spaces and third-order differential equations*, J. Math. Phys. [**41**]{} (2000), 5572–5581.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Giant $\gamma$-ray flares comprise the most extreme radiation events observed from magnetars. Developing on (sub)millisecond timescales and generating vast amounts of energy within a fraction of a second, the initial phase of these extraordinary bursts present a significant challenge for candidate trigger mechanisms. Here we assess and critically analyse the linear growth of the relativistic tearing instability in a globally twisted magnetosphere as the trigger mechanism for giant $\gamma$-ray flares. Our main constraints are given by the observed emission timescales, the energy output of the giant flare spike, and inferred dipolar magnetic field strengths. We find that the minimum growth time of the linear mode is comparable to the $e$-folding rise time, i.e. $\sim10^{-1}$ ms. With this result we constrain basic geometric parameters of the current sheet. We also discuss the validity of the presumption that the $e$-folding emission timescale may be equated with the growth time of an MHD instability.'
author:
- |
C. Elenbaas$^1$[^1], A. L. Watts$^1$, R. Turolla$^{2,3}$, J. S. Heyl$^4$\
$^{1}$Anton Pannekoek Institute for Astronomy, University of Amsterdam, Science Park 904, 1098XH, Amsterdam, The Netherlands.\
$^{2}$Department of Physics and Astronomy, University of Padova, via Marzolo 8, 35131, Padova, Italy.\
$^{3}$Mullard Space Science Laboratory, University College London, Holbury St. Mary,Surrey, RH5 6NT, UK .\
$^{4}$University of British Columbia, Vancouver, Canada.
title: 'The impulsive phase of magnetar giant flares: assessing linear tearing as the trigger mechanism'
---
\[firstpage\]
stars: magnetars – X-rays: bursts – magnetic reconnection
Introduction {#sec:intro}
============
Magnetars are neutron stars (NSs) whose output power is dominated by the decay of an ultra-strong magnetic field (often exceeding the quantum critical field, $B_{\rm qed}\equiv m_e^2c^3/(e\hbar)\simeq4.41\times10^{13}$ G) [@Thompson95, e.g. @Mereghetti08; @Turolla15]. The transient emission properties of such sources include comparatively minor recurrent soft $\gamma$-ray bursts ($E\lesssim10^{42}$ erg) and sporadic giant $\gamma$-ray flares ($E \sim 10^{44}-10^{46}$ erg)[^2]. At present, three giant flares have been observed from independent sources and their lightcurves exhibit remarkably similar characteristics (see Figure \[fig:lcurve\] in Section \[sec:et\]). Giant flares are typically composed of an explosive initial hard $\gamma$-ray spike ($k_{\rm B}T_{\rm spec}\sim175-250$ keV) that develops within (sub)milliseconds and lasts a mere fraction of a second ($\sim0.15-0.5$ s), and a quasi-exponentially abating x-ray tail ($\sim20-30$ keV) that persists for minutes, with superimposed pulsations [see e.g. @Mazets79; @Fenimore96; @Hurley99; @Feroci01; @Palmer05] .
The emission of the decaying tail is argued to be the result of a continuously evaporating and locally magnetically trapped thermal photon-pair fireball. Beamed emission from this moves in and out the line of sight, due to the rotation of the underlying NS [@Thompson95]. The physical process behind the onset, the trigger mechanism, that would clarify the impulsive phase of these energetic flares, remains however a topic of great debate. Here we will discuss one such mechanism, spontaneous tearing of a globally extended equatorial current sheet, in more detail. Typical emission timescales of the observed giant flares play a critical role in resolving this dispute.
Giant flare trigger mechanisms
------------------------------
In this section we briefly explore the various magnetar giant flare trigger mechanisms that have been proposed. We begin with the setup of the system prior to the explosive event and proceed with the triggers, subdivided in internal and external mechanisms.
### Setup: magnetic field formation and evolution
The origin of the strong magnetic field is a non-trivial affair. @Thompson92 have argued that during the transient phase of extensive neutrino cooling moments after gravitational collapse of the progenitor star, entropy-driven convection and differential rotation inside a rapidly spinning (initial spin period: $\Omega_0^{-1}\sim1$ ms) proto-NS may sustain an efficient $\alpha$-$\Omega$ dynamo which could generate an internal magnetic field up to $\sim10^{17}$ G. Alternatively, the massive progenitor may already accommodate a sizeable magnetic field. An ultra-strong field is consequently formed via straightforward flux conservation of the fossil field during implosion [@Ferrario06].
The dynamical timescale of the newly formed ultra-strong field is only seconds or less, and the crystallisation of the outer layer does not set in for another couple of minutes to hours. This allows the field to evolve readily towards a (meta-)stable magneto-hydrodynamic (MHD) equilibrium configuration, likely consisting of a combination of a poloidal- and toroidal component, before its further evolution is constrained by the presence of a highly conductive solid crust [@Flowers77; @Braithwaite06]. The problem of magnetic-field stability, and the respective strengths of the two field components have been studied by e.g. @Braithwaite09, @Lander12, and @Ciolfi13. No consensus has been reached on these matters yet, and further investigations including the effects of superconductivity [@Lander14; @Henriksson13] and the NS crust [@Gourgouliatos14] are required to advance the issue. Subsequent evolution of the strongly twisted field is then determined by ambipolar diffusion, Ohmic decay, and (non-diffusive) Hall drift which occur throughout the interior of the NS (crust and core) and operate on much longer timescales $\gtrsim10^4$ yr [@Thompson96; @Heyl98]. The conductive crust either severely resists the imparted motion of the frozen-in magnetic flux tubes such that Maxwell stress builds up continuously in the system or allows for a constrained transport of magnetic helicity into the magnetosphere, which in turn may develop into a sheared configuration. A reservoir of energy grows (internally or externally) until a certain critical threshold is reached, suddenly releasing the energy in an explosive manner through e.g. a crustal failure or MHD instability of the magnetic field.
### Internal trigger
Motivated by the duration of the impulsive phase ($\sim0.1-1$ s) @Thompson95 initially proposed an internal trigger mechanism whereby a large-scale interchange instability, i.e. a global MHD rearrangement, would take place in the liquid core of the NS and propagate outward on a dynamical timescale, equal to the internal Alfvén crossing time, $$\label{eq:int. Alfven cross time}
\tau^{\rm int}_A=\frac{R_*}{v_A^{\rm int}}\sim0.1~\rm{s},$$ where $R_*\sim10^6$ cm is the typical radius of a NS and $v_A^{\rm int}\sim10^7\,B^{\rm int}_{15}$ cm s$^{-1}$ is the core Alfvén speed for a density $\sim10^{15}$ g cm$^{-3}$ with the core magnetic field strength given by $B^{\rm int}\equiv B^{\rm int}_{15}\times10^{15}$ G. This results in a sudden global displacement of the magnetic footpoints on the surface of the star injecting an ‘Alfvén pulse’ into the magnetosphere, which subsequently induces a relativistic outflow of plasma. The (sub)millisecond rise of the giant flare lightcurve is, they argue, the signature of a reconnection front in the magnetosphere leading the relativistic outflow, which in turn develops on the external Alfvén crossing time, $$\label{eq:ext. Alfven cross time}
\tau_A^{\rm ext}=\frac{R_*}{v_A^{\rm ext}}\sim3\times10^{-2}~\rm{ms},$$ where $v_A^{\rm ext}\sim c$ is the magnetospheric Alfvén speed. Therefore, even though we initially observe the emission from the reconnection front, the trigger nevertheless is given by the onset of the internal instability[^3].
A second trigger mechanism introduced by @Thompson01 involves the force balance between the rigidity of the elastic NS crust and vast magnetic shear-stress, imparted through the anchored magnetic field lines[^4]. Ultimately, the tension of the strongly twisted magnetic field in the crust will become the dominant force and drive the crustal lattice beyond its critical straining threshold $\theta_{\rm crit}$. As the crust yields, the suppressed magnetic energy is allegedly liberated abruptly through a propagating fracture – analogous to an earthquake – producing seismic modes, which in turn couple to magnetospheric Alfvén modes via the pinned magnetic field lines [@Blaes89].
@Thompson01 note however that the storage capacity of elastic energy in the crust $$\label{eq:crust energy}
E^{\rm max}_{\rm elastic}\sim 1.7\times10^{43}\left(\frac{\theta_{\rm crit}}{10^{-2}}\right)^{2}~{\rm erg},$$ which depends on its critical yield strain, is insufficient to explain the observed output power of a giant flare ($E\gtrsim10^{44}$ erg). Accordingly, they argued that the crust merely functions as a gate that assists in the storage and discharge of the internal magnetic energy, rather than as the main energy reservoir. It is important to remark however that they assumed conservatively $\theta_{\rm crit}\lesssim10^{-2}$, yet this value has since been revised by @Horowitz09 through molecular dynamics simulations to be $\theta_{\rm crit}\sim0.1$ [this value has been independently reproduced by @Hoffman12]. With this we obtain $E^{\rm max}_{\rm elastic}\sim10^{45}$ erg \[see Eq. (\[eq:crust energy\])\], which is comparable to the total energy output of the giant flares. Note however that the value for the critical breaking strain decreases significantly, due to defects induced in the crust after the first time it yields [@Hoffman12]. Nonetheless, @Lander15 argue that even a moderate breaking strain of $\sim0.065$ and a fracture extending to the base of the crust can power the most energetic giant flare to date.
Due to the large hydrostatic pressure in the NS crust $P_{\rm crust}$ in comparison to the shear modulus $\mu$, i.e. $P_{\rm crust}\gg \mu$, it is impossible to create a long-lived void necessary for a brittle fracture to occur [@Jones03], regardless of the magnitude of the imparted Maxwell stress. When the crust yields it does not crack, yet rather undergoes a gradual plastic deformation in response to the imparted Lorentz forces, whereby internal currents and associated magnetic helicity are transported outward into the less conductive magnetosphere [@Thompson02].
@Levin12 argue that the presence of a strong magnetic field reinforces the crust, which might strongly impede the formation of a propagating fracture or global slip, altogether. Only under certain specific conditions, where the magnetic flux surface is oriented almost perfectly perpendicular to the direction of shear (within $10^{-3}$ radian), can enough energy be released through a propagating fracture to explain the observed emission.
An important challenge for trigger mechanisms that manifest internally, either a core MHD instability or crustal failure, is the significant impedance mismatch between the internal and external Alfvén velocities [@Link14]. As a result, magnetic energy that dissipates through shear waves cannot be transmitted to the magnetosphere fast enough to explain the (sub)millisecond rise of the initial transient phase of the giant flare. Instead, shear waves get reflected numerous times prior to leaving the stellar interior, extending the outward transmission time considerably.
### External trigger
The aforementioned issues with internal triggers have led to the notion that prior to a giant flare the magnetic energy might be stored in the magnetosphere, rather than in the interior of the NS. @Thompson02 argue that the tightly wound internal magnetic field induces a strong current that in turn closes through a thin surface layer. This local surface layer will experience a Lorentz force, which causes the crust to rotate plastically. Anchored magnetic field lines are dragged along with the gyrating motion and a twist is gradually imparted to the external magnetic field. The twist supporting currents can be composed by charges stripped from the NS surface or – more likely – by pair creation in the magnetosphere [@Beloborodov07]. Subsequently, the non-potential external field reacts to the new boundary conditions and evolves through a series of quasi-equilibria, continuously twisting the external field either locally [@Huang14a; @Huang14b; @Beloborodov09] or globally [@Thompson02].
A local increase of helicity leads to the formation of a helically twisted flux rope embedded in the magnetar magnetosphere, whereby the impulsive phase of the giant flare is associated with an abrupt loss of equilibrium and subsequent catastrophic destabilisation of the flux rope, analogous to the dynamics of coronal mass ejections (CMEs) [@Masada10; @Yu12; @Yu13; @Huang14a; @Huang14b]. Alternatively, a global accumulation of twist may cause the external field to eventually expand outwards, becoming increasingly radial, and admitting a cusp-shaped or Y-type neutral line topology, characterised by a narrow equatorial current sheet where the magnetic shear is most significant [@Mikic94; @Wolfson95; @Parfrey13]. In this narrow yet extended neutral layer the gradients become significant and the MHD approximation breaks down allowing for the field lines to diffuse through the plasma. The onset of the flare is then given by an explosive reconnection event, which may roughly develop on the external Alfvén crossing time $\tau_A^{\rm ext}\sim10^{-2}$ ms \[Eq. (\[eq:ext. Alfven cross time\])\] [@Thompson95], and the expulsion of a relativistic plasmoid. In this paper we investigate specifically the reconnection process in the latter configuration – illustrated in Figure \[fig: mag\_sphere1\].
Both magnetospheric models provide a mechanism for slow build up of an energy reservoir over tens of years caused by the ambipolar diffusion of the internal magnetic field and its subsequent rapid conversion into bulk kinetic energy, particle acceleration, and radiation ($\lesssim$ milliseconds). Observed spectral hardening (softening) and an increase (decrease) in spin down in the pre (post) giant flare stage of SGR 1900+14 and SGR 1806-22 [@Woods99; @Woods01; @Mereghetti05; @Rea05] are consistent with an increase (decrease) of twist and charge density in the external field [@Thompson02; @Lyutikov06]. Moreover, a considerable reduction in harmonic content of the pulse profile of SGR 1900+14 during and following the giant flare suggests a burst mechanism which reduced the twist of the external field significantly [@Woods01].
Distinct reconnection models have been introduced to describe the initial transient phase of the observed giant flares. @Lyutikov03 [@Lyutikov06] and @Komissarov07 suggest the development of the tearing instability in a relativistic force-free current sheet as the trigger mechanism. They argue that the minimum growth time of the linear tearing mode accords with the (sub)millisecond rise (to peak) of the giant flares. Alternatively, @Gill10 propose a fast reconnection model that relies on collisionless Hall reconnection ($\tau_{\rm rec}^{\rm Hall}\sim0.3$ ms) and ascribes a crucial role to the soft precursors ($\gtrsim10^{41}$ erg, $k_{\rm B}T<50$ keV) that have been observed before the last two giant flares[^5]. These precursors facilitate the conditions for collisionless Hall reconnection by introducing a baryon contaminant in the pair dominated magnetosphere, since the former process relies on the Hall effect which in turn requires a non-mass-symmetric plasma composition to operate.\
\
In this article we focus on magnetospheric giant flare trigger mechanisms. In particular we critically analyse the most discussed candidate reconnection mechanism, i.e. impulsive reconnection through the spontaneous development of the tearing instability in a globally sheared external field.[^6] We revise the tearing mode growth time as applied to magnetar magnetospheres by @Lyutikov03 and expand on the rectified result. Characteristic timescales appearing in the giant flare lightcurves have hereby provided necessary constraints. Furthermore, we provide order of magnitude estimates related to the geometry of the reconnection region and discuss the validity of basic assumptions regarding this trigger mechanism.
In Section \[sec:et\] we review typical timescales of the observed giant flare emission and additional relevant data of the phenomena involved. In Section \[sec:tmt\] we summarise previous works on the dynamics behind the relativistic tearing instability and the general expression for its minimum growth time. Subsequently we show that a revised version of the tearing mode growth time for magnetar magnetospheres can in principle explain the (sub)millisecond rise times of giant flares under certain conditions, pertaining to the geometry of the reconnection region. In Section \[sec:pcotrr\], using straightforward theoretical models that rely on the tearing mode timescale, we constrain the height of the reconnection region and thickness of the current sheet in which the tearing mode develops. In section \[sec:ep\] we discuss the relations between the MHD growth time and the radiative timescale, which is directly connected with the observed lightcurve. Throughout the article we adopt a Gaussian-cgs unit system in our calculations.
Emission Timescales {#sec:et}
===================
Currently, three magnetars have produced a giant $\gamma$-ray flare. In chronological order they are the 1979 March 5 flare from SGR 0526-66 [@Mazets79], the 1998 August 27 flare from SGR 1900+14 [@Hurley99], and the 2004 December 27 flare from SGR 1806-20 [@Palmer05]. The energy in radiation emitted during the decaying tail was roughly equal for the three giant flares ($E_{\rm tail}\sim10^{44}$ erg), indicating that the strength of the confining magnetospheric field, which traps the photon-pair fireball, is roughly similar for the three sources since the energy storage capacity of the field is related to its strength [@Mereghetti08]. The inferred surface dipole magnetic field strengths $B_s$ of the sources are given in Table \[tab:gf data\]. The released photon energy during the initial spike was however considerably larger for the most recent giant flare ($E_{\rm spike}\sim10^{46}$ erg), as compared to the first two ($E_{\rm spike}\sim10^{44}$ erg).
Considering the fact that the duration of the hard spike is roughly three orders of magnitude less than the soft tail, it is rather astonishing that the photon energy output of the hard spike is approximately equal to or even much greater than the energy released during the decay of the soft tail. The conversion of such a vast amount of stored magnetic energy into high energy radiation in a considerably limited window of time, requires an extraordinary trigger mechanism indeed, which accordingly may be constrained by the observed photon flux and associated sub(milli)second rise time.
[llllllll]{} SGR & & **0526-66** & & **1900+14** & & **1806-20** &\
Date & &1979 March 5 & & 1998 August 27 & & 2004 December 27 &\
&&& *Ref.* && *Ref.* && *Ref.*\
$\tau_e$ & \[ms\] & $\lesssim1$ & \[4\] & $<1.6,<4$ & \[14\], \[6\] & $\lesssim0.3,\,<1$ & \[13\], \[7\]\
$\tau_{\rm peak}$ & \[ms\] & $\sim15$, $\sim20$ & \[8\], \[2\] & - & & $\sim1.5$ & \[13\]\
$\tau_{\rm spike}$ &\[s\] & $\sim0.1-0.2$ & \[9\] & $\sim0.35, \sim1.0$ & \[10\], \[6\] & $\sim0.2$, $\sim0.5$ & \[7\], \[13\]\
\
$d$ & \[kpc\] & 53.6 & \[5\]$^{\dagger}$ & 12.5 & \[3\]$^{\dagger}$ & 8.7 & \[1\]$^{\dagger}$\
\
$L_{\rm peak}$\* &\[$10^{44}$ erg s$^{-1}$\] & $\sim4.7,\,\sim18$\*\* & \[8\], \[4\] & $>0.64,\,>5.6,\,>160$ & \[6\], \[10\], \[14\] & $\sim7\times10^2$ & \[7\]\
$E_{\rm spike}$\* &\[$10^{44}$ erg\] & $\sim1.1$ & \[8\] & $>0.10,\,>3.0$ & \[10\], \[14\] & $\sim1.2\times10^2$ & \[7\]\
$k_{\rm B}T_{\rm spec}$ & \[keV\] & $246$ & \[4\] & 240 & \[6\] & 175 & \[7\]\
\
$B_{\rm s}$ &\[$10^{14}$ G\] & 5.6 & \[15\]$^{\dagger}$ & 7.0 & \[11\]$^{\dagger}$ & 20 & \[12\]$^{\dagger}$\
[Reference peak luminosities and spike energies have been adjusted according to respective source distances in \[5\], \[3\], and \[1\]. \*\*Peak luminosity from \[8\] (\[4\]) is an average over 200 (10) ms. *References*: \[1\] @Bibby08; \[2\] @Cline80; \[3\] @Davies09; \[4\] @Fenimore96; \[5\] @Haschke12; \[6\] @Hurley99; \[7\] @Hurley05; \[8\] @Mazets79; \[9\] @Mazets81; \[10\] @Mazets99; \[11\] @Mereghetti06; \[12\] @Nakagawa09; \[13\] @Palmer05; \[14\] @Tanaka07; \[15\] @Tiengo09. $^{\dagger}$References obtained through the ‘McGill Online Magnetar Catalog’ [@Olausen14]: <http://www.physics.mcgill.ca/~pulsar/magnetar/main.html>. Burst references can also be found at the Amsterdam Magnetar Burst Catalogue: <http://staff.fnwi.uva.nl/a.l.watts/magnetar/mb.html>.]{}
\[tab:gf data\]
Timescale definitions {#sec:emission timescales}
---------------------
In studying the initial spectrally hard phase of the giant flare lightcurve, the following characteristic emission timescales may be defined[^7] (see Figure \[fig:lcurve\]). The $e$-folding rise time $\tau_e$ describes the exponential rise of the spike out from the continuum (\[$f_\gamma\propto\exp(t/\tau_e)]$, where $f_\gamma$ represents the photon flux). This emission timescale constrains the explosive capability of the trigger mechanism, i.e. the physical process that generates the observed radiation is necessarily required to advance on this timescale. The peak time $\tau_{\rm peak}\equiv |t_{\rm peak}-t_0|$ denotes the time between the onset of the spike $t_0$ and the moment $t_{\rm peak}$ when the spike photon flux peaks $[f_\gamma^{\rm max}(t_{\rm peak})]$ and the spike time $\tau_{\rm spike}\equiv|t_*-t_0|$ represents the duration of the spike, i.e. the timespan of the spectrally hard phase of the giant flare lightcurve, where $t_*$ indicates the end time of the hard spike. The latter timescale may serve to constrain the energy deposition or radiative evaporation time. This timescale will depend on factors such as the extent of the energy reservoir, the rate of energy conversion and radiation production, and/or the effective trapping of the generated radiation.
Observed characteristic timescales and auxiliary parameters {#ssec:observed timescales}
-----------------------------------------------------------
From the giant flare initial spike data listed in Table \[tab:gf data\], we find that the values for the various timescales are typically, $\tau_e\sim0.1-1$ ms, $\tau_{\rm peak}\sim1-10$ ms, and $\tau_{\rm spike}\sim0.1-1$ s. However, the accuracy and precision of the $e$-folding rise time measurements is restricted by the limited time-resolution of the detectors operational at the time. Moreover, the short timescales may have been significantly affected by saturation of the detector and deadtime of the instrument. Both effects, if present, result in an overestimation of the shortest timescales and in particular the $e$-folding rise times. Therefore strictly one should regard these timescales as upper limits.
The listed spectral temperatures $k_{\rm B}T_{\rm spec}$ in Table \[tab:gf data\] are obtained through fitting optically thin thermal Bremsstrahlung (OTTB) or cooling blackbody models to the spectra of the observed giant flare spikes [@Fenimore96; @Hurley99; @Hurley05]. However, the exact physical mechanism that generates the observed spectra remains unknown.
The initial spikes display strong variability on (sub)millisecond timescales and quasi-periodic oscillations (QPOs) with $\nu\sim10^2$ Hz [@Barat83; @Hurley99; @Feroci01; @Terasawa05; @Schwartz05]. Peak luminosities and spike energies in Table \[tab:gf data\] are found assuming isotropic radiation and computed from the observed fluxes using the respective source distances; no spectral bolometric corrections have been applied. In Table \[tab:gf data\], multiple values are quoted at times for various quantities. These values have been sourced from distinct references. They differ because of significant differences in instrumentation, e.g. energy bandwidth and time resolution, and in data analysis techniques. We quote these values to give an indication of the uncertainties involved.
The relativistic tearing mode {#sec:tmt}
=============================
Here we consider the development of the tearing instability in a relativistic current sheet as depicted in Figure \[fig: mag\_sphere1\]. In the presence of finite magnetic resistivity $\eta$ the current sheet will become unstable to transverse tearing modes ($\mathbf{k}\cdot\mathbf{B}=0$) and decompose into many smaller current filaments or magnetic islands [@Furth63] – see Figure \[fig: mag\_sphere1\]. Simultaneously magnetic energy is converted into heat, bulk kinetic energy, and charged particles are accelerated by the induced reconnection electric field $\mathbf{E}=-E_z\hat{\mathbf{z}}$ [see e.g. @Priest00].
In the following we revisit and further analyse the (relativistic) tearing instability as a candidate trigger mechanism for the onset of magnetar giant flares, the groundwork for which has been laid in detail by @Lyutikov03 and @Komissarov07. Here we briefly review the relevant equations of resistive magnetodynamics and the stability analysis of a current sheet in a magnetically dominated magnetosphere, which ultimately results in a minimum growth time of the linear tearing mode. Next we discuss the application of this characteristic timescale to the initial rise of magnetar giant flares and reassess the conclusions of previous work.
Force-free degenerate electrodynamics
-------------------------------------
### Magnetization parameter
To investigate the properties of the magnetar magnetosphere it proves useful to define the dimensionless magnetization parameter, $$\label{eq}
\sigma_{\rm m}\equiv2\frac{u_{\rm B}}{u_{\rm p}}=\frac{B^2}{4\pi\rho c^2},$$ which describes the ratio of magnetic energy density to total particle energy density, where $u_{\rm B}=B^2/8\pi$ and $u_{\rm p}=\rho c^2$, with $B$ the magnitude of the magnetic field, $\rho$ the particle density, and $c$ the speed of light. The magnetization parameter for magnetar magnetospheres is estimated to be $10^{13}\leq\sigma_{\rm m}\leq10^{16}$ [@Komissarov07]. When $\sigma_{\rm m}\gg1$, the magnetosphere is said to be magnetically dominated (the inertia of the particles is negligible, even though they still act as carriers of charge) and relativistic, since the velocity of an Alfvén wave, $$\label{eq}
v_A= c\left(\frac{\sigma_{\rm m}}{1+\sigma_{\rm m}}\right)^{1/2},$$ approaches the speed of light, i.e. $v_A\to c$. Note accordingly that the Alfvén transit time becomes the light crossing time, $\tau_A\to\tau_c=l/c$, where $l$ denotes the typical length scale of the system.
In describing the dynamics of the magnetar magnetosphere, $\sigma_{\rm m}^{-1}$ may be used as a small expansion parameter to approximate the general equations of relativistic magnetohydrodynamics (RMHD) in the limit of vanishing rest-mass density and pressure of matter (force-free approximation), i.e. force-free degenerate electrodynamics (FFDE) or ‘magnetodynamics’ (MD) [@Uchida97; @Komissarov02; @Komissarov07].
### Ohm’s law in resistive FFDE
In FFDE the energy-momentum equation in covariant form, stripped from its matter component, reduces to $$\label{eq:energy-mom}
\nabla_{\mu}T_{\rm em}^{\mu\nu}=0,$$ where $$\label{eq}
T_{\rm em}^{\mu\nu}=\frac{1}{4\pi}\left[F^{\nu\alpha}F^\mu_{~\,\alpha}-\frac{1}{4}g^{\mu\nu}\left(F_{\alpha_\beta}F^{\alpha_\beta}\right)\right],$$ denotes the electromagnetic stress-energy tensor, composed of the electromagnetic field tensor $F^{\mu\nu}$ and the metric tensor $g^{\mu\nu}$. We do not consider the effects of gravitational curvature and assume a Minkowski metric $g^{\mu\nu}\to\eta^{\mu\nu}$ with signature $s=-2$. Combining the energy-momentum equation \[Eq. (\[eq:energy-mom\])\] with the covariant homogeneous and inhomogeneous Maxwell’s equations, respectively $$\label{eq:homogen Maxwell}
\partial_\mu (\star F)^{\nu\mu}=0,$$ and $$\label{eq:inhomogen Maxwell}
\partial_\mu F^{\mu\nu}=\frac{4\pi}{c} J^\nu,$$ where $(\star F)^{\mu\nu}=(1/2)\epsilon^{\mu\nu\sigma\lambda}F_{\sigma\lambda}$ represents the Hodge dual of $F^{\mu\nu}$ and $J^\mu=(c\rho_{\rm ch},\mathbf{j})^T$ is the four-current[^8], we may write the divergence of the stress-energy tensor as $$\label{divTf}
\partial_\nu T_{\rm em}^{\mu\nu}-\frac{1}{c}F^{\mu\nu}J_{\nu}=0,$$ and subsequently find $$\label{covariantFFEeq}
F^{\mu\nu}J_{\nu}=0.$$ The above expression is the so-called force-free condition and implies specifically that the Lorentz force, $$\label{eq}
f^{\mu}=\frac{1}{c}F^{\mu\nu}J_\nu$$ is required to vanish, i.e. that the force-free electromagnetic field is fundamentally degenerate [@Komissarov02]. It follows immediately that the first electromagnetic invariant is zero, $$\label{eq:1st elec variant}
F_{\mu\nu}(\star F)^{\mu\nu}=0,$$ which is known as the degeneracy condition. This means that the inertia of the plasma particles, but not their electromagnetic interaction, is ignored.
In ideal FFDE we wish to describe the plasma velocity in a physical force-free electromagnetic field. To this end we require the plasma velocity field given by $U^{\mu}=\gamma(c,\mathbf{v})^T$ to satisfy $F_{\mu\nu}U^{\nu} = 0$. Since the four-velocity of the plasma is a time-like vector, we demand that the second electromagnetic invariant is positive,$$\label{eq:2nd elec variant}
F_{\mu\nu}F^{\mu\nu} > 0,$$ which necessitates the existence of time-like zero eigenvectors of $F_{\mu\nu}$. This condition implies that there exists a reference frame wherein observers at rest detect a field that is purely magnetic, i.e. where the electric field vanishes entirely [@Uchida97].
Adopting 3+1-notation we find that Eq. (\[covariantFFEeq\]), Eq. (\[eq:1st elec variant\]), and Eq. (\[eq:2nd elec variant\]) become respectively $$\begin{aligned}
\rho_{\rm ch}\mathbf{E}+\frac{1}{c}\mathbf{j}\times\mathbf{B}&=0,\label{eq: force-free cond. 3+1D}\\
\mathbf{E}\cdot\mathbf{B}&=0\label{eq:1st elec variant in 3+1D},\\
B^2-E^2&>0\label{eq: 2nd elec variant in 3+1D},\end{aligned}$$ where $E$ denotes the magnitude of the electric field and $B$ the magnitude of the magnetic field. Incidentally, $F^{0i}J_i=\mathbf{E}\cdot \mathbf{j}=0$ and the electromagnetic energy is conserved, i.e. $$\label{eq: conservation elemag energy}
\partial_t(\mathbf{E}\cdot\mathbf{B})=0.$$
To obtain Ohm’s law, which describes the relation between the current and the electric field, it is convenient to separate the current vector into components parallel and perpendicular to the magnetic field vector, $$\begin{aligned}
\label{jincomponentseq}
\mathbf{j}=\mathbf{j}_\perp+\mathbf{j}_\|,&&\mathbf{j}_\perp=\frac{(\mathbf{B}\times\mathbf{j})\times\mathbf{B}}{B^2},
&&\mathbf{j}_\|=\frac{(\mathbf{B}\cdot\mathbf{j})\,\mathbf{B}}{B^2}.\end{aligned}$$ With the force-free condition Eq. (\[eq: force-free cond. 3+1D\]) we may express the perpendicular component as $$\label{eq}
\mathbf{j}_\perp=\rho_{\rm ch}\mathbf{v}_\perp$$ where along with the requirement expressed by Eq. (\[eq: 2nd elec variant in 3+1D\]) we have defined the electric drift velocity $$\label{eq: drift velocity}
\mathbf{v}_\perp\equiv c\,\frac{\mathbf{E}\times\mathbf{B}}{B^2},$$ which denotes the plasma velocity component across the magnetic field.
In the singular current sheet however the ideal MHD approximation breaks down and the magnetic resistivity becomes finite, i.e. the second electromagnetic invariant \[Eq. (\[eq: 2nd elec variant in 3+1D\])\] becomes negative. Accordingly, the parallel component of Ohm’s law is altered to include the effect of current dissipation, solely along the magnetic field, due to the presence of a resistive electric field. To this end we introduce the relativistic formulation of Ohm’s law in covariant form [@Gedalin96], $$\label{eq: rel. Ohms law}
F^{\mu\nu}U_{\nu}=\frac{4\pi}{c}\,\Theta^{\mu\nu}(\delta^\alpha_\nu-U_\nu U^\alpha)J_\alpha,$$ where $\Theta^{\mu\nu}$ represents the resistivity tensor. This tensor is highly anisotropic in FFDE, since only the currents flowing along the field may experience resistive dissipation. Accordingly we define the resistivity tensor as such $$\label{eq}
\Theta^{\mu\nu}\equiv\eta\frac{b^\mu b^\nu}{b^2},$$ where $b^\mu=(\star F)^{\mu\nu}U_{\nu}$ represents the magnetic four-vector and the scalar resistivity or magnetic diffusivity, which characterises the dissipation of currents, is given by the phenomenological parameter[^9] $\eta=c^2/(4\pi\sigma)$, with $\sigma$ the macroscopic conductivity of the plasma. Subsequently, by convolving Eq. (\[eq: rel. Ohms law\]) with the magnetic four-vector we obtain [@Lyutikov03] $$\label{eq}
F^{\mu\nu}U_{\nu}(\star F)_{\mu\alpha}U^{\alpha}=\frac{4\pi}{c}\eta(\star F)_{\mu\alpha}U^{\alpha}J^{\mu},$$ which in 3+1 notation becomes $$\label{eq}
\gamma^2(\mathbf{B}\cdot\mathbf{E})(c^2-\mathbf{v}\cdot\mathbf{v}) =\frac{4\pi}{c}\eta\gamma\left[\mathbf{j}\cdot(c\,\mathbf{B}-\mathbf{v}\times\mathbf{E})-J^0(\mathbf{B}\cdot\mathbf{v})\right],$$ where $\gamma=[1-(\mathbf{v}\cdot\mathbf{v}/c^2)]^{-1/2}$ is the Lorentz factor. The above expression reduces to $$\label{eq:3+1notation, 2}
\frac{c^2}{4\pi\eta}(\mathbf{B}\cdot\mathbf{E})=\frac{\gamma}{c}\left[\mathbf{j}\cdot(c\,\mathbf{B}-\mathbf{v}\times\mathbf{E})-J^0(\mathbf{B}\cdot\mathbf{v})\right].$$ Upon splitting vectors into components parallel and perpendicular to the magnetic field we can rewrite Eq. (\[eq:3+1notation, 2\]) as $$\label{eq: parallel EJ}
\frac{c^2}{4\pi\eta}(\mathbf{B}\cdot\mathbf{E})=\gamma \left[(\mathbf{B}\cdot\mathbf{j})\left(1-\frac{E_\perp^2}{B^2}\right)-\rho_{\rm ch}(\mathbf{B}\cdot\mathbf{v})\left(1-\frac{E_\perp^2}{B^2}\right)\right].$$ We remove the second term on the r.h.s. by choosing our coordinate system such that $v_\|\equiv0$. Consequently with Eq. (\[eq: drift velocity\]) we have that, $$\label{eq}
\gamma^{-2}=\left(1-\frac{E_\perp^2}{B^2}\right),$$ such that Eq. (\[eq: parallel EJ\]) becomes $$\label{eq}
(\mathbf{B}\cdot\mathbf{j})=\frac{c}{4\pi}\left[\frac{c\gamma}{\eta}(\mathbf{B}\cdot\mathbf{E})\right].$$ Accordingly we use the above result to rewrite the parallel component of the current vector and ultimately obtain the following expression for the current vector, $$\label{resistiveformofOhmslawRMHDeq}
\mathbf{j}=\frac{c}{4\pi}\left[4\pi \rho_{\rm ch}\frac{\mathbf{v}_\perp}{c}+\frac{c\gamma}{\eta}\frac{(\mathbf{B}\cdot\mathbf{E})\mathbf{B}}{B^2}\right],$$ which describes Ohm’s law in resistive FFDE, whereby the electric current is written solely in terms of the electric field components, parallel and perpendicular to the magnetic field. Note that in the plasma rest frame $\mathbf{v}_\perp=0$, the electromagnetic field is no longer purely magnetic, due to the presence of the resistive electric field.\
\
Magnetodynamics near force-free equilibrium
-------------------------------------------
The divergence of the stress-energy tensor Eq. (\[divTf\]) determines the energy- and momentum conservation equations, given in 3+1 notation as follows $$\begin{aligned}
&\partial_tu_{\rm em}+\boldsymbol{\nabla}\cdot\mathbf{S}+\mathbf{E}\cdot\mathbf{j}=0,\label{eq:3+1 energy}\\
&\partial_t\mathbf{p}_{\rm em}-\boldsymbol{\nabla}\cdot\mathbf{T}^{ij}_{\rm em}+\frac{1}{c}\mathbf{j}\times\mathbf{B}+\rho_{\rm ch} \mathbf{E}=0,\label{eq:3+1 moment}\end{aligned}$$ where respectively $$\label{eq}
u_{\rm em}=\frac{B^2+E^2}{8\pi}~~~~\text{and}~~~~\mathbf{p}_{\rm em}=\frac{\mathbf{S}}{c^2},$$ are the electromagnetic energy- and electromagnetic momentum density. The above expressions are written in terms of the Poynting vector, $$\label{eq}
\mathbf{S}=\frac{c}{4\pi}\mathbf{E}\times\mathbf{B},$$ and the Maxwell stress-tensor $$\label{eq}
\mathbf{T}^{ij}_{\rm em}=\frac{1}{4\pi}\left[E^iE^j+B^iB^j-\frac{1}{2}(E^2+B^2)\delta^{ij}\right],$$ where $\delta^{ij}$ is the Euclidean metric of flat space.
To study the dynamical properties of a system near force-free equilibrium, we introduce the relevant timescales via the relativistic Lundquist number, $$\label{eq:Lundqvist number}
\mathcal{S}_l\equiv\frac{\tau_\eta}{\tau_A}=\frac{l c}{\eta},$$ where $\tau_\eta\equiv l^2/\eta$ is the resistive diffusion timescale, $\tau_A\equiv l/v_A\to l/c$ denotes the hydromagnetic timescale or Alfvén transit time (for $\sigma_{\rm m}\gg1$), and $l$ denotes the corresponding typical length scale of the system.
The evolution of the system can be represented by the timescale $\tau$, for which $\tau_A\ll\tau\ll\tau_\eta$. Accordingly, $|\mathbf{v}_\perp|\ll c$ and with Eq. (\[eq: drift velocity\]) we find naturally $E_\perp\ll B$. Immediately we may approximate, $$\begin{aligned}
&\gamma\to1,\nonumber\\
&u_{\rm em}\simeq u_B=\frac{B^2}{8\pi},\nonumber\\
&\mathbf{T}^{ij}_{\rm em}\simeq\frac{1}{4\pi}\left(B^iB^j-\frac{B^2}{2}\delta^{ij}\right).\nonumber\end{aligned}$$
Scaling Eq. (\[eq:3+1 energy\]) and Eq. (\[eq:3+1 moment\]) in terms of the small expansion parameters ($\tau/\tau_\eta$) and ($\tau_A/\tau$) and assuming incompressibility of the plasma, @Komissarov07 derive the following closed set of equations, $$\begin{aligned}
&\boldsymbol{\nabla}\cdot\mathbf{v_\perp}=0,\label{eq:system MD1}\\
&\boldsymbol{\nabla}\cdot\mathbf{B}=0,\\
&\partial_t \mathbf{B}=\boldsymbol{\nabla}\times(\mathbf{v_\perp\times B})+\eta\nabla^2 \mathbf{B},\label{eq:system MD3}\\
&\rho_{\rm em}[\partial_t(\boldsymbol{\nabla}\times\mathbf{v}_\perp)]=\frac{1}{8\pi}\boldsymbol{\nabla}\times(\mathbf{B}\cdot\boldsymbol{\nabla})\mathbf{B},\label{eq:system MD4}\end{aligned}$$ that together govern the dynamics of a system near force-free equilibrium and incidentally closely resemble the equations of non-relativistic resistive incompressible MHD.
Growth time of the (relativistic) tearing mode
----------------------------------------------
### Linear stability analysis
The growth time of the tearing instability may be obtained by performing linear stability analysis on a current sheet described by the following one-dimensional force-free equilibrium profile that represents a rotational discontinuity [@Low73], $$\label{eq:}
\mathbf{B}_0 = B_0\tanh\left(\frac{x}{\delta}\right) \hat{\mathbf{y}} \pm B_0\,{\rm sech}\left(\frac{x}{\delta}\right) \hat{\mathbf{z}},$$ where the magnetic null line is given by the sheared $B_{0y}$-component that goes to zero at $x=0$, whilst the magnitude of the magnetic field vector $|\mathbf{B}_0(x)|$ remains constant under rotation over $\pi$ radian (see Figure \[fig:equi2\]). The vector rotates predominantly within the domain $-\delta<x<\delta$, such that the typical length scale of the system is given by $l\to\delta$, which denotes the (half-)thickness of the current sheet[^10].
Through linearising the dynamical equations of resistive MD[^11] @Komissarov07 demonstrate how to derive the following expressions for the (maximum) wavelength and (minimum) growth time of the fastest growing linear mode, $$\begin{aligned}
\label{eq: tearing mode timescale (general)}
&\lambda^{\rm max}=2\pi\,\delta\mathcal{S}_\delta^{1/4},
&\tau_{\rm tm}^{\rm min}=\tau_A\mathcal{S}_\delta^{1/2}=\left(\tau_A\tau_\eta\right)^{1/2},\end{aligned}$$ where $\mathcal{S}_\delta=c\delta/\eta$ is the relativistic Lundquist number corresponding to the length scale $\delta$ and the minimum growth time of the tearing mode is ascertained to be the geometric mean of the Alfvén- and resistive diffusion timescale, as in the case of non-relativistic resistive incompressible MHD. Further comprehensive and general derivations of tearing mode characteristics may be found in the literature, e.g. @White86, @Goldston95, @Priest00, @Lyutikov03, and @Goedbloed09.
Tearing mode growth time in magnetar magnetospheres
---------------------------------------------------
Here we aim to establish the minimum growth time of the tearing mode prevailing in magnetar magnetospheres. In a globally twisted magnetic field the radial dependency of the magnetic field strength is approximately given by [@Thompson02] $$\label{eq:mag. r dependence}
B_0(r)\simeq B_{\rm s}\left(\frac{r}{R_*}\right)^{-(2+p)},$$ where $B_{\rm s}$ denotes the inferred surface dipole magnetic field strength and $R_*\sim10^6$ cm is the typical NS radius. Also, $0<p<1$ is the radial index which parameterises the net twist angle $0<\Delta\phi<\pi$, where the limiting value $p=1$ ($p=0$) corresponds to a net twist of $\Delta\phi=0$ ($\Delta\phi=\pi$), representing a pure dipole (split monopole) configuration. $B_0(r)$ will function as the background or upstream magnetic field strength of the reconnection region.
We need a qualitative estimate of the local magnetic resistivity $\eta$. We consider a macroscopic description, whereby the resistivity is homogeneous and given by the presence of Langmuir turbulence [as in @Lyutikov03]. In this case the typical turbulent length scale is given by the electron skin depth, $$\label{eq}
\delta_e=\frac{c}{\omega_{p,e}}$$ where $\omega_{p,e}=(4\pi n_\pm e^2/m_e)^{1/2}$ is the electron plasma frequency, with $n_\pm=n_++n_-$ the total number density of the charge carriers, i.e. the sum of positrons $n_+$ and electrons $n_-$, $e$ is the elementary charge unit, and $m_e$ the electron mass. Accordingly, the resultant resistivity of a turbulent plasma with a typical eddy size and turnover velocity of $\delta_e$ and $c$ respectively, is approximately $$\label{eq: resistivity ito plasma freq}
\eta\sim c\,\delta_e=\frac{c^2}{\omega_{p,e}}=c^2\left(\frac{4\pi n_\pm e^2}{m_e}\right)^{-1/2}.$$ In the aforementioned globally twisted dipole model the magnetospheric currents generate a toroidal field component that approaches the strength of the poloidal field, i.e. $B_t\lesssim B_p\sim B_{0}(r)$. Therefore we may apply Ampère’s law to obtain an estimate for the charge number density as a function of the local magnetic field strength $B_0(r)$ and distance from the NS centre $r$ [@Lyutikov02], $$\label{eq: Ampere to plasma freq.}
\boldsymbol{\nabla}\times\mathbf{B}_0=\frac{4\pi}{c}\mathbf{j}=4\pi e\,[\boldsymbol{\beta}_+n_+-\boldsymbol{\beta}_-n_-],$$ where $\boldsymbol{\beta}_+$ and $\boldsymbol{\beta}_-$ are the dimensionless drift velocities of the positrons and electrons, respectively. If we consider the case where $\boldsymbol{\beta}_+=-\boldsymbol{\beta}_-\sim\mathbf{1}$ and $n_+\simeq n_-$, we may simplify[^12] $$\label{eq:}
n_\pm\sim\frac{B_0(r)}{8\pi e\,r}.$$ Accordingly, we obtain an expression for the plasma frequency, $$\label{eq:}
\omega_{p,e}\sim\left(\frac{\omega_{c,e}\,c}{r}\right)^{1/2},$$ with the electron cyclotron frequency given by $\omega_{c,e}\equiv eB_0(r)/(m_e c)$. The resistivity as a function of the surface dipole magnetic field strength and distance to the centre of the NS becomes $$\label{eq: resistivity}
\eta\simeq c^2\left[\frac{eB_{\rm s}}{m_er}\left(\frac{r}{R_*}\right)^{-(2+p)}\right]^{-1/2}.$$ Now together with Eq. (\[eq: tearing mode timescale (general)\]) and Eq. (\[eq:Lundqvist number\]) we may ultimately obtain the minimum growth time of the tearing mode in magnetar magnetospheres, $$\begin{aligned}
\label{eq: tearing mode timescale (general+resistivity)}
&\tau_{\rm tm}^{\rm min}=\left(\frac{\delta^3}{c\eta}\right)^{1/2}\simeq\left(\frac{eR_*^{(2+p)}}{m_e c^6}\right)^{1/4}\,\delta^{3/2}r^{-(3+p)/4}B_{\rm s}^{1/4}.\end{aligned}$$ In order to compare this result with the observed timescales, we rewrite the above result in terms of typical values for the relevant parameters, $$\label{eq:minimum growth time linear tearing mode}
\tau_{\rm tm}^{\rm min}\simeq10^{-1}\,\delta_4^{3/2}r_7^{-(3+p)/4}B_{\rm s, 15}^{1/4}~{\rm ms},$$ where we define $\delta_4\equiv\delta/(10^4$ cm), $r_7\equiv r/(10^7$ cm), $B_{\rm s, 15}\equiv B_{\rm s}/(10^{15}$ G), and $0< p< 1$ (in practice $p$ will always be close to unity). With these scalings, the minimum growth time of the tearing mode agrees nicely with the observed (sub)millisecond $e$-folding rise times $\tau_e$ of the magnetar giant flares.
Note that this timescale differs significantly from the minimum growth time as calculated by @Lyutikov03, essentially due to an error in that calculation (specifically in the inferred expression for the plasma frequency). In addition we have adopted a rather smaller (by a factor of $10^{-2}$) typical size for the thickness of the current sheet $\delta$ than the value used in @Lyutikov03. We do this since for large gradients to develop, the thickness of the current sheet must be significantly less than the global extent of the reconnection region, which in the case of magnetar giant flares is a few times the NS radius. @Komissarov07 argue for a current sheet thickness of $\sim3\times10^{3}$ cm, however we have not been able to reproduce their inferred tearing mode timescale (particularly Eq. (73) in their paper). Without the above modification to the typical value for $\delta$, however, the inferred tearing mode growth time would be $\sim 100$ ms [@Duncan04]. If this were the case, it would entirely rule out the development of the tearing mode as a candidate mechanism to explain the (sub)millisecond rise times of the magnetar giant flares.
In the subsequent section we will assume that the trigger is given by the development of a tearing instability, and that its minimum growth time corresponds to the timescale on which the observed emission is released from the system, i.e. $\tau^{\rm min}_{\rm tm}=\tau_e$. We explore additional constraints on the geometry of the reconnection region that are required for the linear tearing mode to be a plausible mechanism for the giant flares, and discuss how they relate to the constraints derived in this section.
Physical constraints on the reconnection region {#sec:pcotrr}
===============================================
Here we present two straightforward models, respectively based on energy conservation and mechanical equilibrium of the current sheet, that provide order of magnitude estimates for the thickness of the current sheet $\delta$ and height of the base of the reconnection region in terms of the radial distance from the NS centre $r$.
In order to relate the thickness of the tearing unstable current sheet to the global length of the reconnection region $L_y$, we consider the following elementary instability condition: for an unstable mode to be able to develop in a current sheet, its growth time ($\tau_{\rm tm}^{\rm min}$) is required to be less than the time it would take for the perturbation to exit the system ($L_y/c$) [@Shibata01]. We obtain the requirement $$\label{eq: cond1}
\tau_{\rm tm}^{\rm min}<\frac{L_y}{c}.$$ Together with Eq. (\[eq: tearing mode timescale (general+resistivity)\]) we have the following upper limit to the thickness of the current sheet $$\label{eq: cond2}
\delta^{\rm max}=\mathcal{S}_\delta^{-1/2} L_y,$$ and equivalently $$\label{eq: cond3}
\delta^{\rm max}=[\,c\,\eta\,(\tau^{\rm min}_{\rm tm})^2\,]^{1/3}.$$
Conversion of magnetic energy {#subsec:ec}
-----------------------------
Figure \[fig:rec1\] shows the geometry of the reconnection region in the $xy$-plane, whereby the curved (blue) arrows represent the sheared magnetic field that continues to annihilate within the current sheet, which in turn is denoted by the smaller rectangular box ($2\delta\times L_y$). The larger rectangle ($2L_x\times L_y$) describes the size of the total area that proceeds to reconnect, i.e. the extent of magnetic flux that is advected into the current sheet for the duration of the hard $\gamma$-ray spike $\tau_{\rm spike}$. We hypothesise that the magnetic field lines are fed into the diffusion region at a constant rate (this assumption is discussed further in section \[ssec:tearing mode phases\]). The reconnection rate is generally determined by the aspect ratio of the reconnection region through mass flux conservation [e.g. @Pucci13], i.e $$\label{eq:}
\frac{v_{\rm rec}}{c}\simeq\frac{\delta}{L_y}=\mathcal{S}_\delta^{-1/2},$$ which together with Eq. (\[eq: cond1\]) leads to $$\label{eq:}
v_{\rm rec}\simeq\frac{\delta}{\tau_{\rm rec}}.$$ Accordingly, we find that $$\label{eq: Lx approx}
L_x\sim v_{\rm rec}\tau_{\rm spike}=\delta\left(\frac{\tau_{\rm spike}}{\tau_{\rm rec}}\right).$$ For an equatorial current sheet we have $L_z < 2\pi r$ (one may picture the current sheet as a disk around the NS if $L_z=2\pi r$). The entire volume of magnetic flux that reconnects over the course of the initial hard phase of the giant flare is then given by $V=(2L_x)L_yL_z$. The energy contained in this region, that is subsequently released within $\tau_{\rm spike}$ can be estimated as $$\label{eq}
E_{\rm tot}\simeq \zeta\, u_BV=\zeta\,\frac{B_0^2}{8\pi}\left(2L_xL_yL_z\right)=\frac{\zeta\,B_0^2r L_y \delta}{2} \left(\frac{\tau_{\rm spike}}{\tau_{\rm rec}}\right),$$ where $\zeta$ is the fraction of free magnetic energy that is dissipated and we have used $u_B=B_0^2/(8\pi)$ for the local magnetic energy density in terms of the upstream magnetic field $B_0$. Rewriting this equation we obtain $$\label{eq}
\delta(r)\sim \frac{2\,E_{\rm tot}}{\zeta\,B_0^2rL_y}\left(\frac{\tau_{\rm rec}}{\tau_{\rm spike}}\right).$$ Note incidentally that the above general expression does not rely on any particular reconnection mechanism as yet. If we now consider linear tearing as the principal reconnection mechanism, we may set $\tau_{\rm rec}=\tau_{\rm tm}^{\rm min}$ and, through Eq. (\[eq: cond1\]), $L_y= c\tau_{\rm tm}^{\rm min}$. Using Eq. (\[eq:mag. r dependence\]) and adopting $p=1/2$ we end up with $$\label{eq}
\delta(r)\sim\frac{2\,E_{\rm tot}r^4}{\zeta\,cB_s^2R_*^5\tau_{\rm spike}}.$$ Together with the condition stated in Eq. (\[eq: cond3\]), we find an estimate for the height of the reconnection region $$\label{eq:e conserv r}
r_{\rm rec}\sim 10^7\left[\zeta\,B_{s,15}^{11/6} (\tau_{\rm tm,-4}^{\rm min})^{2/3} \left(\frac{\tau_{\rm spike,0.2}}{E_{\rm tot,45}}\right)\right]^{12/41}~{\rm cm},$$ and the thickness of the current sheet at $r_{\rm rec}$, $$\label{eq:e conserv delta}
\delta(r_{\rm rec})\sim10^4\left[\zeta^7\,B_{s,15}^{6} (\tau_{\rm tm,-4}^{\rm min})^{32} \left(\frac{\tau_{\rm spike,0.2}}{E_{\rm tot,45}}\right)^{7}\right]^{1/41}~{\rm cm}.$$ In the above we have made use of Eq. (\[eq: resistivity\]) and Eq. (\[eq: cond1\]) to eliminate $\eta(r)$ and $L_y$ respectively. The solutions depend mildly on $\zeta$.
Mechanical equilibrium
----------------------
Without mechanical equilibrium across the current sheet boundary, the current sheet would disrupt before reconnection could occur effectively [@Uzdensky11]. This requirement is given by the following pressure balance, $$\label{eq}
P_{\rm cs}+\frac{B_{\rm cs}^2}{8\pi}=P_0+\frac{B_0^2}{8\pi},$$ where $P_{\rm cs}$ and $B_{\rm cs}$ respectively are the leptophotonic pressure \[see Eq. (\[eq:leptophotonic pressure\])\] and magnetic field strength inside the current sheet, and $P_0$ and $B_0$ respectively are the local plasma pressure and magnetic field strength in the upstream region. In the upstream region we have $\sigma_{\rm m}\gg1$, such that the plasma beta, $\beta=P_{\rm plasma}/P_{\rm mag}$, is small, i.e. $P_0\ll B_0^2$. Consequently, the above expression simplifies to $$\label{eq: reduced balance}
P_{\rm cs}+\frac{B_{\rm cs}^2}{8\pi}\simeq \frac{B_0^2}{8\pi}.$$ The leptophotonic pressure in the current sheet may be decomposed as $$\label{eq:leptophotonic pressure}
P_{\rm cs}=P_{\rm rad}+P_{\pm},$$ where $P_{\rm rad}$ signifies the radiation pressure and $P_{\pm}$ denotes the pressure as a result of pair production. In a relativistic current sheet $P_{\pm}$ becomes $\sim (7/4)P_{\rm rad}$ [@Uzdensky11], such that $$\label{eq: Pcs is fPrad}
P_\text{cs}\sim \frac{11}{4} P_{\rm rad},$$ and $$\label{eq}
P_{\rm rad}(T_{\rm cs})=\frac{4\,\sigma_{\rm SB}}{3\,c\,k_{\rm B}^4}(k_BT_{\rm cs})^4,$$ where $\sigma_{\rm SB}\equiv\pi^2k_{\rm B}^4/(60\hbar^3c^2)\simeq5.67\times10^{-5}$ erg cm$^{-2}$ s$^{-1}$ K$^{-4}$ is the Stefan-Boltzmann constant and $T_{\rm cs}$ represents the temperature inside the current sheet.
Eq. (\[eq: reduced balance\]) may then be written as $$\label{eq: reduced balance2}
B_0^2-B_{\rm cs}^2=22\pi P_{\rm rad}.$$ Furthermore using Gauss’s law for magnetism $\boldsymbol{\nabla}\cdot\mathbf{B}=0$, we approximate $$\label{eq}
\frac{B_{\rm cs}}{\delta}+\frac{B_y}{L_y}+\frac{B_g}{L_z}\simeq0,$$ where we respectively parameterise the strengths of the guide field and $y$-component of the field as $B_g=qB_0$ and $B_y=(1-q)B_0$, with $0\leq q \leq1/2$ and $\mathbf{B}_0=\mathbf{B}_g+\mathbf{B}_y$. Subsequently, we may write $$\label{eq:}
B_{\rm cs}\simeq B_0 \left[(1-q)\frac{\delta}{L_y}+q\frac{\delta}{L_z}\right].$$ Together with Eq. (\[eq: cond2\]) and the relation for $L_z$ below Eq. (\[eq: Lx approx\]) this becomes $$\label{eq: Bcs as function of B0}
B_{\rm cs}\simeq B_0 \left[(1-q)\,\mathcal{S}_\delta^{-1/2}+q\frac{\delta}{2\pi r}\right],$$ such that we may eliminate $B_{\rm cs}$ from Eq. (\[eq: reduced balance2\]): $$\label{eq: reduced balance3}
B_0^2\left\{1- \left[(1-q)\,\mathcal{S}_\delta^{-1/2}+q\frac{\delta}{2\pi r}\right]^2\right\}=22\pi P_{\rm rad}.$$ The above equation depends on the values for $B_s$, $T_{\rm cs}$, $\delta$ and $r$. To solve Eq. (\[eq: reduced balance3\]) we need to write $\delta$ in terms of $r$ via Eq. (\[eq: cond3\]) and require an estimate for the temperature *inside* the current sheet $k_{\rm B}T_{\rm cs}$. It is however questionable whether $k_{\rm B}T_{\rm spec}$ – listed in Table \[tab:gf data\] – would represent $k_{\rm B}T_{\rm cs}$, since the former may rather correspond to a Lorentz boosted photospheric temperature of a relativistically expanding fireball. Of necessity, we consider here the following reasonable range of temperatures: $k_{\rm B}T_{\rm cs}\sim250-1000$ keV.
Consequently, together with $B_s=10^{15}$ G and $\tau_{\rm tm}^{\rm min}=10^{-4}$ s, we solve Eq. (\[eq: reduced balance3\]) numerically for $r$ and find, $$\label{eq}
r_{\rm rec}\sim(3\times10^6)-10^{7}~{\rm cm},$$ and furthermore with Eq. (\[eq: cond3\]) we have $$\label{eq}
\delta(r_{\rm rec})\sim(4-8)\times10^3~{\rm cm},$$ where the lower (upper) estimates of the above equations correspond to the upper (lower) value for $k_{\rm B}T_{\rm cs}$. These estimates remain equal down to the fourth decimal for the entire range of $q$ and as one can observe from Eq. (\[eq: Bcs as function of B0\]), $B_{\rm cs}\ll B_{0}$, such that the second term on the l.h.s. of Eq. (\[eq: reduced balance2\]) may be neglected to find the following expression (for $p=1/2$), $$\label{eq:}
r_{\rm rec}\sim10^7B_{s,15}^{2/5}\left(\frac{k_{\rm B}T_{\rm cs}}{250\,\text{keV}}\right)^{-4/5}~{\rm cm}.$$ Note that the results agree roughly with those obtained in the previous section. Additionally, we find that the dimensionless reconnection rate is approximately $M_{\rm rec}\equiv\delta/(v_A\tau_{\rm rec})\simeq\delta/(c\tau_{\rm tm}^{\rm min})\sim10^{-3}$, which is comparable to the reconnection rates found for solar flares [e.g @Narukage06]. Moreover, note that the reconnection region is located high up in the magnetosphere, such that the background magnetic field is sub-critical $B_0(r_{\rm rec})\simeq10^{12}\text{ G}\sim10^{-1}B_{\rm qed}$.
Discussion {#sec:ep}
==========
Geometry of the reconnection region
-----------------------------------
The previous calculations provide estimates for the scale of the reconnection region involving spontaneous tearing of a global current sheet; the sheet length \[from Eq. (\[eq: cond1\])\] is $L_y\gtrsim c\,\tau_{\rm tm}^{\rm min}=c\,\tau_e\sim(3\times10^6)-10^7$ cm, the sheet thickness is $\delta\sim10^4$ cm, and the height of the reconnection region is $r\sim10^7$ cm. Here we briefly discuss various consequences of these results.
We have assumed that the resistivity is given by a homogeneous background of Langmuir turbulence, which fundamentally requires that the drift velocity of the current-carrying particles exceeds the thermal velocity of the background plasma. This needs to be the case throughout the extensive reconnection region ($>2\delta\times L_y$) for impulsive tearing to be able to occur on the requisite short timescales.
With an estimate for the thickness of the reconnection region, we can infer the temperature at the photosphere of the current sheet $k_{\rm B}T_{\rm phot}$ [@Uzdensky11]. At the photosphere the optical depth $$\label{eq:optical depth}
\tau\sim\frac{\delta}{\lambda_{\rm mfp}},$$ will be of order unity, where $\lambda_{\rm mfp}$ is the photon mean free path. Assuming that this temperature is sub-relativistic, such that the pair number density is given by $$\label{eq:npairs}
n_{\pm}=\frac{1}{\sqrt{2\pi^3}}\left(\frac{m_ec}{\hbar}\right)^{3}\left(\frac{k_{\rm B}T_{\rm phot}}{m_e c^2}\right)^{3/2}\exp\left[-\frac{m_e c^2}{k_{\rm B}T_{\rm phot}}\right],$$ and considering that the scattering opacity of O-mode (i.e. ordinary mode) photons in the presence of a strong magnetic field remains close to Thompson scattering opacity, $\sigma_{\rm es}($O$)\sim\sigma_T\equiv(8\pi/3)\,e^4/(m_e c^2)^2\simeq6.65\times10^{-25}$ cm$^2$, we have $$\label{eq:}
\lambda_{\rm mfp}(\text{O})\sim\frac{1}{n_{\pm}\,\sigma_T}.$$ Together with Eq. (\[eq:optical depth\]) we find $$\label{eq:delta to Tphot}
\delta\,\sigma_T\,n_{\pm}(k_{\rm B}T_{\rm phot})\sim1,$$ which can be solved for $\delta\sim10^4$ cm to get $k_{\rm B}T_{\rm phot}\sim27$ keV. Note however that $k_{\rm B}T_{\rm phot}$ depends only weakly on $\delta$.
Due to the release of high-energy radiation following the reconnection process, extensive pair-production has resulted in a high photospheric pair density $n_\pm(k_{\rm B}T_{\rm phot}\sim27 \text{ keV})\sim10^{20}$ cm$^{-3}$. Note that this pair density greatly exceeds the charge density that is available prior to the onset of reconnection \[from Eq. (\[eq: Ampere to plasma freq.\]) we establish $n\gtrsim10^{14}B_{\rm s,\,15}r_7^{-7/2}$ cm$^{-3}$\]. It is argued that the observed spectral temperatures $k_{\rm B}T_{\rm spec}$ (see Table \[tab:gf data\]) correspond to a Lorentz-boosted photospheric temperature of a pair fireball that, in the wake of the onset of the flare, expands outward from the stationary reconnection region relativistically [@Lyutikov06; @Uzdensky11], such that $$\label{eq:}
\Gamma\,k_{\rm B}T_{\rm phot}=k_{\rm B}T_{\rm spec},$$ where $\Gamma$ denotes the bulk Lorentz factor of the ejected fireball.[^13] Using the result from Eq. (\[eq:delta to Tphot\]) and assuming that the dimensions of the fireball roughly correspond to those of the initial current sheet, we obtain $\Gamma\sim10$, which is consistent with previous estimates in literature.
Furthermore, considering the required scale of the initial configuration $L_y$, uniquely determined by $\tau_e$, spontaneous tearing seems an unlikely candidate for the smaller recurrent $\gamma$-ray bursts ($\lesssim10^{41}$ erg, $\tau_e\sim1$ ms [@Gogus01; @Gavriil04]), since their $e$-folding rise times are similar to those of the giant flares, such that $L_y\sim(3-10)R_{*}$. These particular bursts may rather demonstrate for instance driven reconnection through an external driver (e.g. sudden crustal motion at magnetic foot points or ideal instabilities in smaller critically sheared magnetic arcades (Browning et al. 2008)), or comprise explosive seismic events without involving magnetospheric reconnection altogether.
Linear tearing and the observed high energy emission
----------------------------------------------------
In discussing linear tearing as a candidate mechanism for explaining the fast initial rise of magnetar giant flare light curves, it has been implicitly assumed throughout the literature that the growth of the resistive instability directly coincides with the conversion of magnetic energy – via Ohmic heating and particle acceleration – into the observed high energy radiation (i.e. $\tau_{\rm tm}^{\rm min}=\tau_e$) [@Lyutikov03; @Duncan04; @Komissarov07]. This conjecture presumes that (i) linear tearing dictates the rate of radiation release and (ii) that during the linear tearing phase a significant amount of magnetic energy is converted efficiently to produce the observed radiation in the first place. Both assumptions will be examined further; in Section \[ssec:nonthermal emission\] we discuss the former requisite (i), and in Sections \[ssec:tearing mode phases\] and \[ssec:coalescence instability\] we address the latter (ii).
### Nonthermal emission from accelerated particles {#ssec:nonthermal emission}
Concerning point (i) above it should be emphasised by observing that even for comparatively well-studied phenomena like solar flares the generation and release of radiation is not unequivocally linked to the reconnection rate. Note that whilst solar flares are not supposed to be directly analogous, the comparison may still be informative. The rapid onset of a solar flare is given by the sudden increase of hard X-ray (HXR) emission due to collisional thick-target Bremsstrahlung interactions of nonthermal particles at the chromospheric footpoints of coronal loop structures undergoing magnetic reconnection [@Shibata11]. Accordingly, the observed radiation timescales are determined by the acceleration timescales of the nonthermal particles.
Proposed acceleration mechanisms include direct acceleration by reconnection induced or field-aligned electric fields [e.g. @Aschwanden06; @Egedal12], acceleration through shocks [@Aschwanden06], and stochastic acceleration through turbulence excited by reconnection outflows at the loop top or cascading Alfvén waves near the footpoints [e.g. @Petrosian04; @Liu08; @Liu09; @Fletcher08]. None of the above processes guarantee a straightforward connection between the timescales of linear tearing and that of radiation release. Moreover, such acceleration mechanisms generally rely on the later phases of reconnection (e.g. nonlinear tearing; see section \[ssec:coalescence instability\]) or rather its large-scale effects, such as reconnection jets that excite MHD turbulence or the catastrophic rearrangement of the global magnetic field topology. In the latter case, the amount and rate of energy release will be determined more by the dynamic restructuring of the field, than on the dissipation of an extended current sheet [@Hoshino12].
Efficient particle acceleration in magnetar magnetospheres may however require local regions where the magnetic field becomes small enough, since considerable synchrotron losses might otherwise impede any significant acceleration. Acceleration through reconnection induced electric fields localised at magnetic x-points seems fitting in this regard, since not only does $B_y\to0$ but the presence of $\mathbf{E}\times\mathbf{B}$-drift also focusses the trajectory of the charged particles in the acceleration region [@Speiser65]. Particle-in-cell (PIC) simulations of relativistic reconnection in pair plasmas disclose short acceleration timescales [@Zenitani01], such that the timescale on which the radiation is generated is the reconnection rate.
Nonetheless, a major complication is given by the copious pair production that will ensue upon release of high energy radiation in the presence of an ultra-strong magnetic field [e.g. @Harding06], causing the reconnection region to become optically thick. The observed radiation timescales will therefore not necessarily represent the timescales associated with reconnection dynamics [@Uzdensky11; @Hoshino12]. To further constrain magnetar burst trigger mechanisms, via emission timescales, will require a better understanding of radiation transport in the magnetar magnetosphere.
### Phases of tearing: linear and nonlinear {#ssec:tearing mode phases}
Exponential growth of the magnetic island proceeds until their half-width $$\label{eq:}
w(t)\propto\exp\left[\frac{t}{\tau_{\rm tm}}\right],$$ becomes comparable to the size of the resistive sublayer $\epsilon\delta$; here nonlinear effects become important. Analytic calculations disclose a transition from exponential to algebraic growth ($\propto t^\alpha$), once this stage is reached [@Rutherford73]. Numerical simulations confirm this strong change in reconnection rate, even though it is less significant when $k\ll1$ and $\mathcal{S}_\delta\gg1$ [@Steinolfson84]. Moreover, it is found that the nonlinear regime sets in very quickly, after only a few $e$-folding times, such that one would expect to observe a considerable change in reconnection rate just moments after the onset of the instability. With $\tau_{\rm tm}^{\rm min}\sim10^{-4}$ ms, the exponential phase of the light curve would only last a few tenths of milliseconds to a millisecond, followed by a notable decline in count rate due to the transition from linear to nonlinear tearing. A break in the increase of the count rate during the initial rise to peak has been observed for the SGR 1806-20 flare by @Terasawa05, @Schwartz05, and @Tanaka07 after a few $e$-folding times. The latter reference also finds a similar break in the SGR 1900+14 giant flare.
Note that the assumption of a constant reconnection rate for the duration of the hard spike ($\tau_{\rm spike}\sim0.1-1$ s), as applied in Section \[subsec:ec\], is suspect in the light of nonlinearity of the mode; the obtained estimates for $r$ and $\delta$ \[Eq. (\[eq:e conserv r\]) and Eq. (\[eq:e conserv delta\])\] are lower limits in this regard.
### Coalescence and impulsive bursty reconnection {#ssec:coalescence instability}
The least stable long-wavelength tearing modes ($\lambda^{\rm max}\sim L_y$) tend to saturate soon after the onset of the nonlinear phase. For efficient reconnection to occur the presence of a significantly strong external driver (e.g. sudden crustal motions at the footpoints of sheared arcades, the onset of an ideal instability, or the catastrophic ejection of a flux rope) may be required, which forces a current sheet to become unstable to shorter wavelength modes ($\lambda^{\rm max}\ll L_y$), such that a chainlike structure of magnetic islands is formed before the nonlinear phase sets in [@Uzdensky14]. This configuration is consequently unstable to the coalescence instability, whereby the magnetic islands approach each other through mutually attractive Lorentz forces since they essentially comprise parallel flowing current concentrations. Island coalescence is typically subdivided into two phases: (1) the ideal MHD phase, where the current loops approach one another, and (2) the resistive reconnection phase, where due to finite resistivity ($\eta\neq0$) and large field gradients between the approaching current loops, the loops merge to form one current loop with an increased cross section, i.e. larger magnetic island. Stability analysis was performed by @Finn77 on a particular periodic island-chain configuration described by a Fadeev force-free equilibrium [@Fadeev65], $$\begin{aligned}
\label{eq:}
&\psi_0=\ln[\cosh(kx)+\epsilon\cos(ky)],\nonumber\\
&\mathbf{B}_0=B_0 \,\boldsymbol{\hat{z}}\times\boldsymbol{\nabla}\psi_0,\nonumber\\
&\nabla^2\psi_0=4\pi j_{z0}=(1-\epsilon^2)k^2\exp\left[-2\psi_0\right],\nonumber\end{aligned}$$ where $\psi_0$ is the equilibrium magnetic flux function, $B_0$ the local (background) magnetic field, $j_{z0}$ the equilibrium current directed perpendicular to the reconnection plane, and $0<\epsilon<1$ the peakedness parameter of the current concentration in the magnetic islands. Subsequent numerical simulations have shown that, for a large range of $S_\delta$, the coalescence growth rate is much greater than the tearing growth rate [@Pritchett79] and that it depends critically on the value for $\epsilon$ [@Bhattacharjee83]; linear tearing corresponding to $\epsilon = 0$.
The coalescence instability is characterised by two timescales associated with its distinct phases [@Kliem95]. During the ideal phase the current loops approach each other on a hydromagnetic timescale, whereby the length scale is given by the separation distance of paired current loops $\lambda_{\rm C}$, $$\label{eq:}
\tau_{\rm C1}\sim\epsilon^{-1}\frac{\lambda_{\rm C}}{v_A}$$ with $\delta\lesssim\lambda_{\rm C}\lesssim\lambda^{\rm max}$. In general $\tau_{\rm C1}\ll\tau_{\rm tm}^{\rm min}$, yet no magnetic energy is dissipated in the process. During the resistive phase, when the current loops merge, ‘anti-reconnection’ occurs in-between the approaching islands. The reconnection rate is enhanced by the external driving forces of the converging current loops, such that in general $\tau_{\rm C2}<\tau_{\rm tm}^{\rm min}$. Moreover, for strongly peaked current concentrations ($\epsilon\to1$), we have $\tau_{\rm C1}\sim\tau_{\rm C2}\ll\tau_{\rm tm}^{\rm min}$. For $\lambda_{\rm C}^{\rm max}\simeq\lambda^{\rm max}\sim10^6$ cm, the coalescence timescale becomes comparable to the magnetospheric light crossing time, i.e. $\tau_{\rm C}\sim \tau_A^{\rm ext}\,\epsilon^{-1}$ ms.
Coalescence following tearing converts the bulk of the free magnetic energy in the current sheet, such that the island growth phase may act as mere prelude to the explosive energy release of merging current loops [@Leboeuf82]. Its rapid development and ability to convert a significant fraction of magnetic energy argue in favour of coalescence, rather than tearing, as an explanation for the impulsive phase of flares [@Tajima82; @Tajima87; @Sakai87; @Kliem95; @Schumacher97]. The observed giant flare emission may therefore be a proxy of the nonlinear, rather than the linear, tearing phase.
Furthermore, for higher values of $\mathcal{S}_\delta$ and $\sigma_{\rm m}$ a nonlinear process known as ‘impulsive bursty reconnection’ may occur, whereby a cycle of slow tearing, rapid coalescence, current sheet thinning, and further secondary tearing (i.e the plasmoid instability) at an increased rate repeats successively [@Leboeuf82; @Priest85; @Uzdensky10]. Consequently, energy is released during separate coalescence events in a fragmentary and quasi-periodic manner. This process is advanced to explain the periodic temporal fine structure of hard X-ray (HXR) emission and coherent drifting radio bursts associated with discrete (bidirectional) electron beams observed during the impulsive phase of solar flares [@Aschwanden95; @Kliem00; @Karlicky04]. Quasi-periodic oscillations (QPOs) of $\nu\sim10^2$ Hz, that might be associated with separate energy injections, have also been detected during the initial phases of the magnetar giant flares: See Subsection \[ssec:observed timescales\]. These distinct energy surges may be interpreted as quasi-periodic peaks in coalescence rates[^14], resulting from impulsive bursty reconnection. Precise timing observations of hypothesised (drifting) radio burst from magnetars [@Lyutikov02] may greatly help to further probe the reconnection substructure (e.g. separate plasmoids), reconnection rate, and density of the acceleration region. Yet even though various magnetars show radio emission [e.g. @Camilo06], no such bursts of coherent radio emission coincident with the (recurrent) $\gamma$-ray bursts have been observed to date.
The total energy release through a multitude of coalescence events may be estimated accordingly [@Kruger89; @Kliem95], $$\label{eq:}
U_{\rm C}^{\rm tot}\simeq\frac{(N_{\rm C}-1)^3}{N_{\rm C}}\frac{\lambda_{\rm C}^2L_zB_0^2}{24\pi^2}\ln\left(\frac{\lambda_{\rm C}}{\delta}\right),$$ where $N_{\rm C}$ is the number of individual coalescence events. If we estimate the total number of coalescence events during the impulsive phase of a giant flares as follows, $$\label{eq:}
N_{\rm C}\simeq\nu\tau_{\rm spike}\frac{L_y}{\lambda_{\rm C}^{\rm max}}\sim10^2,$$ we find for the total energy release through dynamic current sheet reconnection, $$\label{eq:}
U_{\rm C}^{\rm tot}\sim10^{45}\,N_{\rm C,2}^2(\lambda_{\rm C,6}^{\rm max})^2L_{z,6}B_{s,15}^2r_7^{-5}\ln\left(\frac{\lambda^{\rm max}_{\rm C,6}}{\delta_4}\right)~{\rm erg},$$ where $L_{z,6}=L_z/10^6$ cm is the length of a current loop. This estimate is consistent with the observed energy output of the initial spike – see Table \[tab:gf data\].
Summary
=======
To better understand the extreme nature of the explosive onset of magnetar giant flares, we have discussed impulsive reconnection through the spontaneous development of the linear tearing instability in a globally sheared external field as a candidate trigger mechanism. Upon reexamination of previous works on the (relativistic) linear tearing mode, we found that the minimum growth time in magnetar magnetospheres is $\tau_{\rm tm}^{\rm min}\sim10^{-1}$ ms \[Eq. (\[eq:minimum growth time linear tearing mode\])\]. This estimate is consistent with the typical $e$-folding rise times ($\tau_e\sim0.1-1$ ms) of the giant flare light curves (see Table \[tab:gf data\]). Our result differs significantly from the one found by @Lyutikov03 ($\tau_{\rm tm}^{\rm L03}\sim10$ ms). Even though the rescaling of the current sheet thickness (by a factor of $10^{-2}$) has a larger effect on the final result, the difference is however essentially due to an error in that calculation.
Assuming the validity of the assumption that the exponential rise time of the giant flare is a proxy for the linear growth time of the tearing mode $\tau_{\rm tm}^{\rm min}=\tau_e$, we obtained order of magnitude estimates for the thickness of the current sheet and height of the base of the reconnection region, respectively $\delta\sim10^4$ cm and $r\sim10^7$ cm, through elementary pressure balance and energy conservation considerations. Additionally we found that the global length of the current sheet would have to be $L_y\sim(3-10)\,R_*$, which is reasonable for the giant flares, yet problematic for the smaller recurrent bursts, since such large unstable regions would have to develop on very short timescales $\Delta T\sim100$ s, where $\Delta T$ represents the typical waiting time of recurrent bursts [@Gogus99; @Gogus00].
Finally we discussed the obtained constraints on the reconnection geometry and evaluated the soundness of the aforementioned assumption of equating an MHD growth time with an emission timescale. Regarding the latter, it is not apparent whether linear tearing dictates the rate of radiation release and if during the linear tearing phase magnetic field dissipation occurs efficiently enough to generate the observed emission. Considering the impulsive phase of solar flares there is no unequivocal connection between linear tearing and the observed high energy emission that is ultimately radiated by accelerated nonthermal particles. Moreover, substantial pair production in magnetar magnetospheres may obscure the emission resulting from magnetic field dissipation through reconnection, altogether.
Furthermore, nonlinear effects become significant soon after the onset of linear tearing and in general reduce the reconnection rate considerably. Fast and efficient reconnection during the nonlinear impulsive bursty regime that may follow tearing, requires however the presence of a strong external driver e.g. rapid crustal motion or catastrophic loss of equilibrium of external magnetic field configurations. Accordingly, we propose that future research into magnetospheric trigger mechanisms for magnetar (giant) bursts investigate *driven* reconnection scenarios, where the emission timescales may constrain the development of the external driver, the nonlinear reconnection phase, or the intense reconnection aftereffects.\
\
**Acknowledgments:** C.E. acknowledges support from NOVA (Nederlandse Onderzoeksschool voor Astronomie). A.W. acknowledges support from NWO Vidi Grant 639.042.916. The work of R.T. is partially supported by INAF through a PRIN grant. J.H. is supported by the Natural Sciences and Engineering Research Council of Canada. We would like to thank Sam Lander, Lyndsay Fletcher, Maxim Lyutikov, Serguei Komissarov, and the participants of the ‘Integrated Plasma Modelling of Solar Flares’ Lorentz Center workshop (May 2015) for helpful discussions. We also wish to acknowledge the useful and significant feedback from the anonymous referees.
[54]{} natexlab\#1[\#1]{}url \#1[[\#1]{}]{}urlprefix
M.J., [Benz]{} A.O., [Dennis]{} B.R., et al., 1995, , 455, 347
M.J., 2006, Physics of the Solar Corona (Springer & Praxis Publishing Ltd, Chichester, UK)
C., [Hayles]{} R.I., [Hurley]{} K., et al., 1983, , 126, 400
A.M., [Thompson]{} C., 2007, , 657, 967
A.M., 2009, , 703, 1044
A., [Brunel]{} F., [Tajima]{} T., 1983, Phys. of Fluids, 26, 3332
J.L., [Crowther]{} P.A., [Furness]{} J.P., [Clark]{} J.S., 2008, , 386, L23
O., [Blandford]{} R., [Goldreich]{} P., [Madau]{} P., 1989, , 343, 839
J., [Spruit]{} H.C., 2006, , 450, 1097
J., 2009, , 397, 763
F., [Ransom]{} S.M., [Halpern]{} J.P., et al., 2006, , 442, 892
B., [Werner]{} G.R., [Uzdensky]{} D.A., [Begelman]{} M.C., 2014, , 782, 104
R., [Rezzolla]{} L., 2013, , 435, L43
T.L., [Desai]{} U.D., [Pizzichini]{} G., et al., 1980, , 237, L1
B., [Figer]{} D.F., [Kudritzki]{} R.-P., et al., 2009, , 707, 844
R.C., 2004, In: [Höflich]{} P., [Kumar]{} P., [Wheeler]{} J.C., 2004, (eds.) Cosmic Explosions in Three Dimensions, (Cambridge, UK: Cambridge Univ. Press)
J., [Daughton]{} W., [Le]{} A., 2012, Nature Phys. Lett., 8, 321
V.M., [Kvabtskhava]{} I.F., [Komarov]{} N.N., 1965, Nucl. Fusion, 5, 202
E.E., [Klebesadel]{} R.W., [Laros]{} J.G., 1996, , 460, 964
M., [Hurley]{} K., [Duncan]{} R.C., [Thompson]{} C., 2001, , 549, 1021
L., [Wickramasinghe]{} D., 2006, , 367, 1323
J.M., [Kaw]{} P.K., 1977, 20, 72
L., [Hudson]{} H.S., 2008, , 675, 1645
E., [Ruderman]{} M.A., 1977, , 215, 302
H.P., [Killeen]{} J., [Rosenbluth]{} M.N., 1963, Phys. of Fluids, 6, 459
F.P., [Kaspi]{} V.M., [Woods]{} P.M., 2004, , 607, 959
M., 1996, , 76, 3340
R., [Heyl]{} J., 2010, , 407, 1926
J.P., [Keppens]{} R., [Poedts]{} S., 2009, Advanced Magnetohydrodynamics (Cambridge: Cambridge Univ. Press)
E., [Woods]{} P.M., [Kouveliotou]{} C., et al., 1999, , 526, L93
E., [Woods]{} P.M., [Kouveliotou]{} C., et al., 2000, , 532, L121
E., [Kouveliotou]{} C., [Woods]{} P.M., et al., 2001, , 558, 228
P., [Reisenegger]{} A., 1992, , 395, 250
R.J., [Rutherford]{} P.H., 1995, Introduction to Plasma Physics. CRC Press.
K.N., [Cumming]{} A., 2014, Phys. Rev. Let., 112, 171101
A.K., [Lai]{} D., 2009, Rep. Prog. Phys., 69, 2631
R., [Grebel]{} E.K., [Duffau]{} S., 2012, , 144, 107
K.T., [Wasserman]{} I., 2013, , 431, 2986
J.S., [Kulkarni]{} S.R., 1998, , 506, L61
K., [Heyl]{} J., 2012, , 426, 2404
C.J., [Kadau]{} K., 2009, , 102, 191102
M., [Lyubarsky]{} Y., 2012, Space Sci. Rev., 173, 521
L., [Yu]{} C., 2014, , 784, 168
L., [Yu]{} C., 2014, , 796, 3
K., [Cline]{} T., [Mazets]{} E., et al., 1999, , 397, 41
K., [Boggs]{} S.E., [Smith]{} D.M., et al., 2005, , 434, 1098
P.B., 2003, , 595, 342
D., [Milosavljevic]{} M., [Spitkovsky]{} A., 2013, , 774, 41
M., 2004, , 417, 325
B., 1995, Proc-1995-Benz, 93
B., [Karlicky]{} M., [Benz]{} A.O., 2000, , 360, 715
S.S., 2002, , 336, 759
S.S., [Barkov]{} M., [Lyutikov]{} M., 2007, , 374, 415 (KBL07)
A., [Kliem]{} B., [Hildebrandt]{} J., 1989, ESA-SP, 285, 169
S.K., [Jones]{} D.I., 2012, , 424, 482
S.K., 2014, , 437, 424
S.K., [Andersson]{} N., [Antonopoulou]{} D., et al., 2015, , 449, 2047
J.N., [Tajima]{} T., [Dawson]{} J.M., 1982, Phys. of Fluids, 25, 784
Y., [Lyutikov]{} M., 2012, , 427, 1574
B., 2014, , 441, 2676
W., [Petrosian]{} V., [Dennis]{} B.R., [Jiang]{} Y.W., 2008, , 676, 704
S., [Fletcher]{} L., 2009, , 701, L34
B.C., 1973, , 181, 209
M., 2002, , 580, L65
M., 2003, , 346, 540 (L03)
M., 2006, , 367, 1594
Y., [Nagataki]{} S., [Shibata]{} K., [Terasawa]{} T., 2010, , 62, 1093
E.P., [Golenetskii]{} S.V., [Il’Inskii]{} V.N., et al., 1979, , 282, 587
E.P., [Golenetskii]{} S.V., 1981, Astrophys. & Space Sci., 75, 47
., [Cline]{} ., [Aptekar]{} ., et al., 1999, Astron. Lett., 25, 635
S., [Tiengo]{} A., [Esposito]{} P., et al., 2005, , 628, 938
S., [Esposito]{} P., [Tiengo]{} A., et al., 2006, , 653, 1423
S., 2008, , 15, 225
Z., [Linker]{} J.A., 1994, , 430, 898
Y.E., [Mihara]{} T., [Yoshida]{} A., et al., 2009, Astron. Soc. of Japan, 61, 387
V.M., [Melnikov]{} V.F., 2009, Space Sci. Rev., 149, 119
N., [Shibata]{} K., 2006, , 637, 1122
S.A., [Kaspi]{} V.M., 2014, Suppl. Series, 212, 6
D.M., [Barthelmy]{} S., [Gehrels]{} N., et al., 2005, , 434, 1107
K., [Beloborodov]{} A.M., [Hui]{} L., 2013, , 774, 92
V., [Liu]{} S.M., 2004, , 610, 550
E.R., 1985, IAU Symposia, 107, 233
E., [Forbes]{} T., 2000, Magnetic Reconnection: MHD Theory and Applications (New York: Cambridge Univ. Press)
P.L., [Wu]{} C.C., 1979, Phys. of Fluids, 22, 2140
F., [Velli]{} M., 2013, , 780, L19
N., [Tiengo]{} A., [Mereghetti]{} S., et al., 2005, , 627, L133
P.H., 1973, Phys. of Fluids, 16, 1903
J.I., [Ohsawa]{} Y., 1987, Space Sci. Rev., 46, 113
J., [Kliem]{} B., 1997, Phys. of Plasmas, 4, 3533
S.J., [Zane]{} S., [Wilson]{} R.J., et al., 2005, , 627, L129
K., [Tanuma]{} S., 2001, Earth Planets Space, 53, 473
K., [Magara]{} T., 2011, Living Rev. in Solar Phys., 8, 6
L., [Spitkovsky]{} A., 2014, , 783, L21
T.W., 1965, J. of Geophys. Res., 70, 4219
R.S., [van Hoven]{} G., 1984, Phys. of Fluids, 27, 1207
T., [Brunel]{} F., [Sakai]{} J., 1982, , 258, L45
T., [Sakai]{} J., [Nakajima]{} H., et al., 1987, , 321, 1031
Y.T., [Terasawa]{} T., [Kawai]{} N., et al., 2007, , 665, L55
T., [Tanaka]{} Y.T., [Takei]{} Y., et al., 2005, , 434, 1110
C., [Duncan]{} R.C., 1992, , 392, L9
C., [Duncan]{} R.C., 1995, , 275, 255
C., [Duncan]{} R.C., 1996, , 473, 322
C., [Duncan]{} R.C., 2001, , 561, 980
C., [Lyutikov]{} M., Kulkarni S.R., 2002, , 574, 332
A., [Esposito]{} P., [Mereghetti]{} S., et al., 2009, , 399, L74
R., [Zane]{} S., [Watts]{} A., 2015, Rep. on Progress in Phys., 73, 116901
T., 1997, Phys. Rev. E, 56, 2181
D.A., [Loureiro]{} N.F., [Schekochihin]{} A.A., et al., 2010 , Phys. Rev. Lett., 105, 235002
D.A., 2011, , 160, 45
D.A., [Loureiro]{} N.F., 2014, arXiv: 1411.4295
R.B., 1986, Rev. Mod. Phys., 58, 183
R., 1995, , 443, 810
P.M., [Kouveliotou]{} C., [van Paradijs]{} J., et al., 1999, , 518, L103
P.M., [Kouveliotou]{} C., [Göğüş]{} E., et al., 2001, , 552, 748
C., 2012, , 757, 67
C., 2013, , 771, L46
S., [Hoshino]{} M., 2001, , 562, L63
S., [Hoshino]{} M., 2007, , 670, 702
S., [Hoshino]{} M., 2008, , 677, 530
[^1]: E-mail: [email protected]
[^2]: Energy discharge estimates assume an isotropic release of radiation.
[^3]: See also the discussion in @Link14 on the feasibility of such an internal MHD instability mechanism.
[^4]: This mechanism was discussed earlier by @Thompson95 in explaining the physical process behind the less energetic recurrent soft $\gamma$-ray bursts from magnetars.
[^5]: Any precursor of the 1979 March 5 giant flare would have gone by unnoticed due to the lack of detectors operational at the time with sensitivities below $\sim50$ keV.
[^6]: Recent particle-in-cell simulations that describe relativistic reconnection in pair plasmas demonstrate the growth of the drift-kink (DK) instability perpendicular to the plane of reconnection through tearing [@Zenitani07]. For certain initial equilibrium configurations the DK instability dominates over the tearing instability at first and consequently impedes efficient reconnection, thermalises the particles, and broadens the current sheet. However, it is also shown that efficient reconnection leading to significant particle acceleration will occur at a later stage, when the tearing mode regains dominance [@Sironi14]. Moreover, it is found that the DK instability is quenched in the presence of a finite guide field, such that the dynamics of the sheet is dictated by the development of the tearing instability at all stages [@Zenitani08; @Kagan13; @Cerutti14]. The $B_\phi$ component of the globally twisted magnetic field surrounding the neutron star may act as a guide field in the case of an equatorial current sheet.
[^7]: Here we follow the definitions for the typical timescales as described by @Duncan04, section 1.3.
[^8]: Here $\rho_{\rm ch}$ represents the plasma charge density.
[^9]: We do not derive $\eta$ from microscopic plasma processes, but rather assume a simple macroscopic description.
[^10]: In the following we refer to $\delta$ simply as the thickness of the current sheet, even though in principle it only describes half of the total thickness - see Figure \[fig: mag\_sphere1\].
[^11]: The linearised equations of resistive MD are equal to those of resistive MHD, such that the growth time of the linear tearing mode remains equal for both regimes. This similarity was first made explicit by @Komissarov07.
[^12]: Twisted magnetospheres are believed to be threaded by pairs moving at mildly relativistic speeds and with low multiplicity, as required to explain magnetars quiescent emission at X-ray energies [see e.g. @Turolla15 and references therein].
[^13]: Note that the photosphere of the relativistically expanding fireball differs from the stationary emission region associated with the onset of the flare, such that the bulk Lorentz factor of the latter is zero.
[^14]: Quasi-periodic pulsations (QPPs) are ubiquitously observed in solar flares; among self-oscillatory reconnection, a multitude of alternative mechanisms have been proposed to explain these phenomena [@Nakariakov09].
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We prove the existence of Gysin morphisms for hyperplane sections that may not satisfy the usual hypotheses of the Lefschetz hyperplane theorem. As an application, we show the triviality of the Alexander polynomial of a particular class of non-symmetric line arrangements, thus providing positive evidence for a conjecture of Papadima and Suciu.'
author:
- Federico Venturelli
title: 'Gysin morphisms for non-transversal hyperplane sections with an application to line arrangements'
---
Introduction {#introduction .unnumbered}
============
We work over the field of complex numbers ${\mathbb{C}}$. Let $X$ be a closed subvariety of ${\mathbb{P}}^n$ and $Y:=X\cap H$ be a hyperplane section; in order to compare the cohomology groups of $X$ and $Y$ one needs some form of the Lefschetz hyperplane theorem. The typical situations in which this theorem can be used are:
(a) (Classical version) Both $X$ and $Y$ are smooth.
(b) (Modern formulation) There exists a Whitney stratification ${\mathcal{A}}$ of ${\mathbb{P}}^n$ such that $X=\cup_{i=1}^k A_i$ with $A_i\in{\mathcal{A}}$ and $H$ is transversal to $A_i$ for $i=1,\ldots,k$ (see for example [@D1 Theorem 1.6.5] and references therein).
In both cases above, the comparison is provided by Gysin morphisms $H^{k}(Y)\rightarrow H^{k+2}(X)$ which are isomorphisms for $k>\dim(Y)$ and a surjection for $k=\dim(Y)$.
In this paper we prove a version of the Lefschetz hyperplane theorem that allows one to find such Gysin morphisms even in some cases in which neither (a) nor (b) are satisfied. Key to our result are cubical hyperresolutions of varieties, which can be thought of as a way of resolving a variety ‘at all levels’. Indeed, when computing a resolution one usually stops when the exceptional divisor has simple normal crossings, but one could go on and resolve the singularities of the exceptional divisor and so on. Cubical hyperresolutions are a precise formalisation of this ‘inductive resolution’ procedure.
Cubical hyperresolutions were used in [@GNPP] to prove another version of the Lefschetz hyperplane theorem, which was the main motivation for our work. The precise statement is the following:
\[thm:Guillen\] [@GNPP Corollaire III.3.12] Let $X$ be a quasi-projective complex variety and $Y$ be a hyperplane section of $X$ satisfying the following hypotheses:
(i) There exists an augmented $m$-cubical hyperresolution $X_{{\square}}\rightarrow X$ such that $Y_{{\square}}:=X_{{\square}}\times_{X}Y$ is a cubical hyperresolution of $Y$.
(ii) For any ${\alpha}\in{\square}_m^*$, there exists a closed immersion $Y_{{\alpha}}\hookrightarrow X_{{\alpha}}$ of codimension $1$.
Then there exist Gysin morphisms between de Rham cohomology groups
$$H_{DR}^k(Y)\rightarrow H_{DR}^{k+2}(X)$$
which are isomorphisms for $k>\dim(Y)$ and a surjection for $k=\dim(Y)$.
One can check that conditions (i) and (ii) above are satisfied if either (a) or (b) is satisfied.
The main result of this paper is the following:
\[thm:risultato\] Let $X\subset{\mathbb{P}}^n$ be a quasi-projective variety with singular locus $\Sigma_X$, $H\subset{\mathbb{P}}^n$ be a hyperplane and $Y:=X\cap H$ be the corresponding hyperplane section with singular locus $\Sigma_Y$. Denote by $\tilde{X}$ and $\tilde{Y}$ resolutions of $X$ and $Y$, and by $D_X$ and $D_Y$ the corresponding exceptional divisors.
Assume that:
(i) There exist $m$-cubical hyperresolutions $H(Y)_{{\square}}$ and $H(X)_{{\square}}$ of the resolution squares $S(Y)$ and $S(X)$ associated to $Y$ and $X$ and a closed immersion $H(Y)_{{\square}}\hookrightarrow H(X)_{{\square}}$.
(ii) There exists an $m_1$-cubical hyperresolution ${D_Y}_{{\square}}$ (resp. ${D_X}_{{\square}}$) of $D_Y$ (resp. $D_X$) and an $m_2$-cubical hyperresolution ${\Sigma_Y}_{{\square}}$ (resp. ${\Sigma_X}_{{\square}}$) of $\Sigma_Y$ (resp. $\Sigma_X$) such that:
(I) For all $I\in{\square}^*_{m_1}$ there exists a closed immersion ${D_Y}_I\hookrightarrow {D_X}_I$ which restricts to a codimension one closed immersion on each irreducible component of ${D_Y}_I$.
(II) There exists $c\in\{0,1\}$ such that for all $I\in{\square}^*_{m_2}$ there exists a closed immersion ${\Sigma_Y}_I\hookrightarrow {\Sigma_X}_I$ which restricts to a codimension $c$ closed immersion on each irreducible component of ${\Sigma_Y}_I$.
We have the following:
1. If $c=1$ then there exist Gysin morphisms $H^k(Y)\rightarrow H^{k+2}(X)$ which are isomorphisms for $k>\dim(Y)$ and a surjection for $k=\dim(Y)$.
2. if $c=0$ then the conclusion of point 1. holds for $k>2\dim(\Sigma_X)+1$.
For the terminology regarding cubical hyperresolutions and cubical varieties, as well as for the definition of (algebraic) de Rham cohomology groups, we refer to Section 1.
Observe that if $c=1$ then we obtain the same Gysin morphisms provided by the usual Lefschetz hyperplane theorem; one can check that if (b) holds then we are in this situation.
We were able to find an example in which the hypotheses of Theorem \[thm:risultato\] are satisfied, but (a), (b) and those of Theorem \[thm:Guillen\] are not. Our example relies on the fact that both $X$ and $Y$ have isolated and very simple singularities, which makes it easy to control the cubical hyperresolutions; this suggests that as the dimension and complexity of $\Sigma_X$ and $\Sigma_Y$ increase finding examples in which conditions (i) and (ii) of Theorem \[thm:risultato\] are satisfied becomes much harder.
Using Theorem \[thm:risultato\] we were able to prove the triviality of the Alexander polynomial of a particular class of line arrangements in ${\mathbb{P}}^2$. The study of the Alexander polynomial of line arrangements is a vast subject, lying at the crossroads of topology, combinatorics and geometry; for this reason, in this introduction and in Section 3 we focus only on the aspects we care the most about.
The Alexander polynomial $\Delta_C$ of a reduced plane curve $C=V(f)$ of degree $d$ is the characteristic polynomial of the algebraic monodromy acting on $H^1(F,{\mathbb{C}})$, where $F$ is any fibre of the Milnor fibration $f:{\mathbb{C}}^3\setminus f^{-1}(0)\rightarrow{\mathbb{C}}^*$; it is known that $\Delta_C$ depends on the type and relative position of the singular points of ${\mathbb{C}}$, and that it can be written as
$$\Delta_C(t)=(t-1)^{r-1}\prod_{1<k|d}\Phi_k(t)^{e_k}=(t-1)^{r-1}q(t)$$
where $r$ is the number of irreducible components of $C$ and $\Phi_k$ is the $k$-th cyclotomic polynomial. When $f$ factors into linear forms $C$ is called a line arrangement, and is usually denoted by ${\overline{{\mathcal{A}}}}$. Such curves have been intensively studied by mathematicians interested in the Alexander polynomial, as one hopes to relate the latter to the combinatorial structure of the line arrangement encoded in its intersection semilattice $L({\overline{{\mathcal{A}}}})$; in particular, it is natural to ask the following questions:
1. Does $\Delta_{{\overline{{\mathcal{A}}}}}$ depend only on $L({\overline{{\mathcal{A}}}})$?
2. Are there necessary or sufficient conditions in order to have $q(t)\neq 1$ that depend only on $L({\overline{{\mathcal{A}}}})$?
Regarding question 1., in [@PS Conjecture 1.9] Papadima and Suciu formulated the following conjecture:
The Alexander polynomial of a line arrangement ${\overline{{\mathcal{A}}}}$ has the form
$$\Delta_{{\overline{{\mathcal{A}}}}}(t)=(t-1)^{|{\overline{{\mathcal{A}}}}|-1}\Phi_3(t)^{e_3}[\Phi_2(t)\Phi_4(t)]^{e_4}$$
where $e_3=\beta_3({\overline{{\mathcal{A}}}})$ and $e_2=e_4=\beta_2({\overline{{\mathcal{A}}}})$.
The $\beta_i({\overline{{\mathcal{A}}}})$ are the modular Aomoto-Betti numbers of ${\overline{{\mathcal{A}}}}$, and they depend only on $L({\overline{{\mathcal{A}}}})$ and $i$ (see [@PS Section 3]).
As for question 2. many examples in the literature suggest that in order for $q(t)$ to be different from $1$ it is necessary that ${\overline{{\mathcal{A}}}}$ admits a multinet; the latter is a purely combinatorial notion that may be thought of as a formalisation of the idea of ‘highly symmetric arrangement’ (see \[defn:multinets\]).
We prove the following:
\[thm:AP\] Let ${\overline{{\mathcal{A}}}}$ be a line arrangement s.t. any line passes through at least one of two given points $P_1$ and $P_2$. Then the Alexander polynomial of ${\overline{{\mathcal{A}}}}$ is trivial, i.e. $\Delta_{{\overline{{\mathcal{A}}}}}(t)=(t-1)^{|{\overline{{\mathcal{A}}}}|-1}$.
Since arrangements of this type do not admit multinets, this reinforces the idea that the existence of multinets is necessary in order to have $q(t)\neq 1$; moreover, our result is consistent with the conjecture by Papadima and Suciu, as $\beta_2({\overline{{\mathcal{A}}}})=\beta_3({\overline{{\mathcal{A}}}})=0$.
The key steps of the proof of Theorem \[thm:AP\] are the following:
1. We associate to ${\overline{{\mathcal{A}}}}$ a threefold $X\subset{\mathbb{P}}^4$ and a fibred threefold $\psi:X'\rightarrow {\mathbb{P}}^1$ such that any fibre of $\psi$ is isomorphic to a hyperplane section $Y$ of $X$. Then we choose a generic fibre $Y$ of $\psi$, we explicitly compute the monodromy action $\phi$ on it and we show that $X$ and $Y$ satisfy the hypotheses of Theorem \[thm:risultato\].
2. We prove that the surjective Gysin morphism $H^2(Y)\twoheadrightarrow H^4(X)$ remains surjective if we restrict the domain to the fixed part of $H^2(Y)$ under the action of the algebraic monodromy $T_{\phi}$ (Proposition \[prop:monodromy\]) and that it gives a surjective morphism between the primitive parts (Lemma \[lem:primitive\]).
3. We bound the dimension of $H^2(Y)^{T_{\phi}}$ using the inclusion relation between the Hodge and polar filtration on $H^{{\bullet}}({\mathbb{P}}^3\setminus Y)$ and the explicit description of the graded pieces of the latter in terms of differential forms; we then use a Thom-type result (Lemma \[lem:ThomIso\]) to deduce from this bound our result on $\Delta_{{\overline{{\mathcal{A}}}}}$.
The paper is organised as follows. In Section 1 we give a brief account of the theory of cubical hyperresolutions, following [@GNPP] and [@PeSt Section 5], showing in particular a sketch of their standard construction. Section 2 is devoted to the proof of Theorem \[thm:risultato\]. Section 3 deals with the Alexander polynomial of curves and line arrangements, presenting the definition of multinet as well as some interesting known results, and it is meant to give the reader an idea of how the combinatorics of ${\overline{{\mathcal{A}}}}$ can affect $\Delta_{{\overline{{\mathcal{A}}}}}$ as well as of how Theorem \[thm:AP\] fits in the picture; it closes with some sparse facts that we shall need in Section 4, which are placed here for lack of a better alternative. Lastly, Section 4 constitutes the proof of Theorem \[thm:AP\].
Cubical hyperresolutions and de Rham cohomology
===============================================
1. The $n$*-semisimplicial category* is the category ${\triangle}_n$ with objects the sets $[m]:=\{0,\ldots,m\}$ for $0\leq m\leq n$ and with morphisms the strictly increasing maps $[m]\rightarrow [m']$. The $n$*-cubical category* is the category ${\square}_n$ with objects the subsets of $[n-1]$ and with $\text{Hom}(I,J)$ consisting of a single element if $I\subset J$ and empty otherwise. We denote by ${\square}_n^*$ the full subcategory of ${\square}_n$ whose objects are the non-empty subsets of $[n-1]$
2. If ${\mathcal{C}}$ is any category, an $n$*-cubical ${\mathcal{C}}$-object* is a contravariant functor $K_{{\square}}:{\square}_n\rightarrow{\mathcal{C}}$, and morphisms between such objects are morphisms of the corresponding functors; $K^*_{{\square}}$ denotes the restriction of $K_{{\square}}$ to ${\square}_n^*$. Similarly, we can define $n$*-semisimplicial ${\mathcal{C}}$-objects* $K_{{\bullet}}$ and morphisms thereof. We will use the notations $K_I:=K_{{\square}}(I)$ and $K_m:=K_{{\bullet}}([m])$.
3. If $S$ is any object in ${\mathcal{C}}$, the *constant* $n$*-cubical ${\mathcal{C}}$-object* $S$ is the contravariant functor $S_{{\square}}:{\square}_n\rightarrow{\mathcal{C}}$ such that $S_I=S$ for all $I\in{\square}_n$, with all morphisms $S_I\rightarrow S_J$ given by the identity of $S$. An *augmentation* of an $n$-cubical ${\mathcal{C}}$-object $K_{{\square}}$ to $S$ is a morphism of $n$-cubical ${\mathcal{C}}$-objects $K_{{\square}}\rightarrow S$. If we replace ${\square}_n$ by ${\triangle}_n$, we obtain *constant $n$-semisimplicial ${\mathcal{C}}$-objects* and augmentations thereof.
The next observations will be useful in what follows:
\[rmk:cubobj\]
1. If $X_{{\square}}$ is an $n$-cubical ${\mathcal{C}}$-object we can associate to it the augmented $n$-cubical ${\mathcal{C}}$-object ${\varepsilon}:X_{{\square}}\rightarrow X_{\emptyset}$; sometimes we will call this augmentation the *natural augmentation*.
2. Any $(n+1)$-cubical ${\mathcal{C}}$-object $X_{{\square}}$ can be considered as a morphism $Y_{{\square}}\rightarrow Z_{{\square}}$ of $n$-cubical ${\mathcal{C}}$-objects by setting $Z_I:=X_I$ and $Y_I:=X_{I\cup\{n\}}$ for $I\in{\square}_n$; in particular, a $1$-cubical ${\mathcal{C}}$-object is the datum of two objects $X,Y\in{\mathcal{C}}$ and a morphism $f:X\rightarrow Y$ between them.
3. To any $(n+1)$-cubical ${\mathcal{C}}$-object $X_{{\square}}$ we can associate functorially an $n$-semisimplicial ${\mathcal{C}}$-object $X_{{\bullet}}$ and an augmentation ${\varepsilon}:X_{{\bullet}}\rightarrow X_{\emptyset}$. We set:
$$X_k:=\coprod_{|I|=k+1}X_I\hspace{0.5cm}\text{for }k=0,\ldots,n.$$
Let ${\beta}:[s]\rightarrow [r]$ be a strictly increasing map (in particular $r\geq s$). If $I\in{\square}_{n+1}$ has cardinality $r+1$ we can write it as $I=\{i_0,\ldots,i_r\}$ with $i_0<\dots<i_r$; the set $J:={\beta}(I):=\{i_{{\beta}(0)},\ldots,i_{{\beta}(s)}\}$ is contained in $I$, so we have a morphism $d_{JI}:X_I\rightarrow X_J$. We can now define the morphism
$$d_{{\beta}}:X_r\rightarrow X_s\hspace{0.5cm}\text{s.t. }(d_{{\beta}})_{|X_I}=d_{{\beta}(I)I}.$$
Since for any $I\in{\square}_{n+1}$ we have a morphism $d_{\emptyset I}:X_I\rightarrow X_{\emptyset}$ we obtain the desired augmentation by setting ${\varepsilon}_{|X_I}:=d_{\emptyset I}$.
The category $TopAbSh$ has objects the pairs $(X,{\mathcal{F}})$ where $X$ is a topological space and ${\mathcal{F}}$ is a sheaf of abelian groups on $X$, and as morphisms the pairs $(f,f^{\#}):(X,{\mathcal{F}})\rightarrow (Y,{\mathcal{G}})$ where $f:X\rightarrow Y$ is a continuous function and $f^{\#}:{\mathcal{G}}\rightarrow f_*{\mathcal{F}}$ is a morphism of sheaves of abelian groups on $Y$. A *sheaf of abelian groups* ${\mathcal{F}}^{{\bullet}}$ (resp. ${\mathcal{F}}^{{\square}}$) on an $n$-semisimplicial space (resp. on an $n$-cubical space) is just an $n$-semisimplicial (resp. $n$-cubical) $TopAbSh$-object. In a similar manner, we can define complexes and resolutions of sheaves of abelian groups on $n$-semisimplicial or $n$-cubical spaces.
Given a sheaf of abelian groups ${\mathcal{F}}^{{\bullet}}$ on an $n$-semisimplicial space $X_{{\bullet}}$, it is possible to define the cohomology of $X_{{\bullet}}$ with values in ${\mathcal{F}}^{{\bullet}}$; indeed, using the Godement resolutions of each ${\mathcal{F}}^m$ and differentials coming from the face maps of ${\triangle}_n$, one obtains a double complex $F^{{\bullet},{\bullet}}$ and sets
$$H^k(X_{{\bullet}},{\mathcal{F}}^{{\bullet}}):=H^k(s(F^{{\bullet},{\bullet}}))$$
where $s(F^{{\bullet},{\bullet}})$ is the simple complex associated to $F^{{\bullet},{\bullet}}$.
If $Y$ is a constant $n$-semisimplicial space, any sheaf of abelian groups on $Y$ will be denoted by ${\mathcal{F}}$ and not by ${\mathcal{F}}^{{\bullet}}$; likewise, the cohomology groups of $Y$ with values in ${\mathcal{F}}$ will be denoted by $H^k(Y,{\mathcal{F}})$.
Assume that ${\varepsilon}:X_{{\bullet}}\rightarrow Y$ is an augmented $n$-semisimplicial space and ${\mathcal{F}}^{{\bullet}}$ is a sheaf of abelian groups on $X_{{\bullet}}$. The sheaves ${\varepsilon}_*{\mathcal{C}}_{Gdm}^p({\mathcal{F}}^q)$ form a double complex of sheaves of abelian groups on $Y$, whose associated simple complex gives
$$R{\varepsilon}_*{\mathcal{F}}^{{\bullet}}:=s[{\varepsilon}_*{\mathcal{C}}_{Gdm}^{{\bullet}}({\mathcal{F}}^{{\bullet}})].$$
One can prove that the hypercohomology of the latter complex coincides with the cohomology of $X_{{\bullet}}$ with values in ${\mathcal{F}}^{{\bullet}}$, i.e.
$$\label{eq:isocohom}
\mathbb{H}^k(Y,R{\varepsilon}_*{\mathcal{F}}^{{\bullet}})=H^k(X_{{\bullet}},{\mathcal{F}}^{{\bullet}})\hspace{0.5cm}\text{for any }k.$$
[@De Definition 5.3.2] An augmented $n$-semisimplicial space ${\varepsilon}:X_{{\bullet}}\rightarrow Y$ is *of cohomological descent* if for any sheaf of abelian groups ${\mathcal{F}}$ on $Y$ the natural adjunction morphism
$$\label{eq:CohomDesc}
{\mathcal{F}}\rightarrow R{\varepsilon}_*{\varepsilon}^{-1}{\mathcal{F}}$$
is an isomorphism in $D_+(Sh(Y))$.
1. Observe that if ${\varepsilon}:X_{{\bullet}}\rightarrow Y$ is of cohomological descent then $H^k(Y,{\mathcal{F}})\simeq H^k(X_{{\bullet}},{\varepsilon}^{-1}{\mathcal{F}})$ for any $k$ and for any sheaf of abelian groups ${\mathcal{F}}$ on $Y$.
2. If $X_{{\square}}$ is an $(n+1)$-cubical space and ${\mathcal{F}}^{{\square}}$ is a sheaf of abelian groups on $X$, by point 3. of Remark \[rmk:cubobj\] to this data we can associate an augmented $n$-semisimplicial space ${\varepsilon}:X_{{\bullet}}\rightarrow X_{\emptyset}$ and a sheaf of abelian groups ${\mathcal{F}}^{{\bullet}}$ on it. We set
$$\label{eq:cono}
C^{{\bullet}}(X_{{\square}},{\mathcal{F}}^{{\square}}):=\text{Cone}^{{\bullet}}[{\mathcal{F}}^{\emptyset}\rightarrow R{\varepsilon}_*{\varepsilon}^{-1}{\mathcal{F}}^{\emptyset}].$$
From now on, we will take for ${\mathcal{C}}$ the category whose objects are reduced separated schemes of finite type over ${\mathbb{C}}$, which we will simply call varieties, and whose morphisms are morphisms of schemes; this is not fully consistent with the existing literature, in which the term ‘algebraic variety’ is usually reserved for *integral* separated schemes of finite type over some field.
1. An augmented $n$-semisimplicial variety is of cohomological descent if this is the case for the associated augmented $n$-semisimplicial space.
2. Let $X$ be a variety. An $n$-*semisimplicial resolution* of $X$ is an $n$-semisimplicial variety ${\varepsilon}:X_{{\bullet}}\rightarrow X$ augmented to $X$ such that all maps $X_m\rightarrow X$ are proper, all $X_m$ are smooth and ${\varepsilon}$ is of cohomological descent.
3. An $(n+1)$-cubical variety is of cohomological descent (resp. a *cubical hyperresolution*) if the associated augmented $n$-semisimplicial variety is of cohomological descent (resp. an $n$-semisimplicial resolution). By the definition of cohomological descent and (\[eq:cono\]), $X_{{\square}}$ is of cohomological descent if and only if $C^{{\bullet}}(X_{{\square}},{\mathcal{F}}^{{\square}})$ is acyclic for any sheaf of abelian groups ${\mathcal{F}}^{{\square}}$ on $X_{{\square}}$.
We want to show that any variety $X$ admits an $m$-cubical hyperresolution $X_{{\square}}$ for some $m$.
1. A *proper modification* of a variety $S$ is a proper morphism $f:X\rightarrow S$ such that there exists $U\subset S$ open and dense for which $f$ induces an isomorphism $f^{-1}(U)\rightarrow U$; a *resolution* of $S$ is a proper modification with $X$ smooth.
2. The *discriminant* of a proper morphism of varieties $f:X\rightarrow S$ is the minimal closed subset $D\subset S$ such that $f$ induces an isomorphism $X\setminus f^{-1}(D)\rightarrow S\setminus D$.
The notions of proper modification, resolution and discriminant extend immediately to $n$-cubical varieties and morphisms thereof. By [@GNPP Théorème I.2.6] any $n$-cubical variety $X_{{\square}}$ admits a resolution, which is constructed by separating and resolving the irreducible components of each $X_I$ and then ‘patching together’ the pieces in a way prescribed by the cubical structure of $X_{{\square}}$.
Let $f:X_{{\square}}\rightarrow S_{{\square}}$ be a proper modification (resp. resolution) of an $n$-cubical variety. A *discriminant square* (resp. *resolution square*) for $f$ is a commutative diagram
$$\xymatrixcolsep{1.5cm}\xymatrixrowsep{1.5cm}\xymatrix{
E_{{\square}} \ar[r]^j \ar[d] & X_{{\square}} \ar[d]^f\\
D_{{\square}} \ar[r]^i & S_{{\square}}}$$ where the horizontal maps are closed immersions and $f$ induces an isomorphism between $X_{{\square}}-j(E_{{\square}})$ and $S_{{\square}}-i(D_{{\square}})$ (i.e. $i(D_{{\square}})$ contains the discriminant of $f$).
[@PeSt Lemma 5.20] The $(n+2)$-cubical variety defined by a discriminant square for a proper modification of an $n$-cubical variety is of cohomological descent.
We can now state the main result we shall need on cubical hyperresolutions:
\[thm:iperrescub\] Any variety $X$ admits an $(n+1)$-cubical hyperresolution $X_{{\square}}$.
A full proof can be found in [@GNPP Théorème I.2.15] or in [@PeSt Thereom 5.26]; here we are only interested in sketching how such a cubical hyperresolution can be constructed.
- Take a resolution $\pi:\tilde{X}\rightarrow X$ of $X$ and consider the $2$-cubical variety $X^{(1)}_{{\square}}$ given by the associated resolution square:
$$X^{(1)}_{\emptyset}:=X,\hspace{0.5cm}X^{(1)}_{\{0\}}:=\tilde{X},\hspace{0.5cm}X^{(1)}_{\{1\}}:=D,\hspace{0.5cm}X^{(1)}_{\{0,1\}}:=\pi_1^{-1}(D).$$
$X^{(1)}_{{\square}}$ can be seen as a morphism of $1$-cubical varieties $f^{(1)}:Y_{{\square}}^{(1)}\rightarrow Z_{{\square}}^{(1)}$, with $Z_I$ smooth for $I\neq\emptyset$.
- Consider a resolution $\pi_2:\tilde{Y}_{{\square}}^{(1)}\rightarrow Y_{{\square}}^{(1)}$ and the corresponding resolution square; we obtain the diagram
$$\xymatrixcolsep{1.5cm}\xymatrixrowsep{1.5cm}\xymatrix{
E_{\square}^{(1)} \ar[r] \ar[d] & \tilde{Y}_{\square}^{(1)} \ar[d]^{\pi_2} \ar[rd]^{f^{(1)}\circ\pi_2} &\\
D_{\square}^{(1)} \ar[r] & Y_{\square}^{(1)} \ar[r]^{f^{(1)}} & Z_{\square}^{(1)}.}$$ The outer commutative square of $1$-cubical varieties can be considered as a $3$-cubical variety $X_{\square}^{(2)}$ i.e. as a morphism of $2$-cubical varieties $f^{(2)}:Y_{{\square}}^{(2)}\rightarrow Z_{{\square}}^{(2)}$, with $Z_I$ smooth for $I\neq\emptyset$.
- Repeat the previous step enough times.
Observe that if we take for ${\mathcal{C}}$ the category of $n$-cubical varieties and consider $X_{{\square}}\in{\mathcal{C}}$, we can still apply the construction of Theorem \[thm:iperrescub\] to $X_{{\square}}$: at each step we obtain an $m$-cubical variety whose entries are $n$-cubical varieties. More precisely, Theorem \[thm:iperrescub\] implies the following:
\[thm:iperrescubcub\] Any $n$-cubical variety $X_{{\square}}$ admits a hyperresolution by an $m$-cubical variety $Y_{{\square}}$ whose entries are $n$-cubical varieties.
\[rmk:iperrescubcub\] Assume that $X_{{\square}}=\{X_{\emptyset},X_{\{0\}},X_{\{1\}},X_{\{01\}}\}$ is a $2$-cubical variety and $Y_{{\square}}$ is an $m$-cubical hyperresolution of $X_{{\square}}$, then $Y_{{\square}}$ can be thought of as a $2$-cubical variety $Y'_{{\square}}=\{Y'_{\emptyset},Y'_{\{0\}},Y'_{\{1\}},Y'_{\{01\}}\}$ of $(m-2)$-cubical varieties; by construction, for any $I\in{\square}_2$ we have that $Y'_I$ is an $(m-2)$-cubical hyperresolution of $X_I$.
In [@GNPP] cubical hyperresolutions were used to define a cohomology theory for possibly singular algebraic varieties; namely:
\[defn:DeRhamComplex\] Let $X$ be a separated scheme of finite type over a field ${\mathbb{K}}$ of characteristic zero, and let ${\varepsilon}:X_{{\square}}\rightarrow X$ be an $(n+1)$-cubical hyperresolution of $X$ together with its natural augmentation; the *de Rham complex* and $k$-th algebraic de Rham cohomology group of $X$ are defined as
$$DR^{{\bullet}}_X:=R{\varepsilon}_*\Omega^{{\bullet}}_{X_{{\bullet}}}\hspace{1cm}H_{DR}^k(X):=\mathbb{H}^k(X,DR^{{\bullet}}_X).$$
If $V\subset X$ is a closed subset, the $k$-th algebraic de Rham cohomology group of $X$ with supports in $V$ is defined as
$$H_{DR,V}^k(X):=\mathbb{H}^k(V,R\Gamma_V DR^{{\bullet}}_X).$$
In both cases, ${\varepsilon}:X_{{\bullet}}\rightarrow X$ is the augmented $n$-semisimplicial resolution of $X$ associated to $X_{{\square}}$.
This cohomology theory coincides with the one developed by Hartshorne in [@H1] in case of an embeddable scheme $X$ over ${\mathbb{C}}$, since the cohomology groups are defined as the hypercohomology of isomorphic complexes (see [@GNPP Théorème III.1.3]).
The definitions of hyperresolution and de Rham complex $DR^{\bullet}_X$ given in [@GNPP] are actually different from the ones we presented here; if we denote by ${\mathcal{C}}$ the category of separated schemes of finite type over a field ${\mathbb{K}}$ of characteristic zero, we have:
\[defn:hyperrescubGNPP\] [@GNPP Définition I.3.2] If $X_{{\square}}$ is an $n$-cubical hyperresolution of $X\in{\mathcal{C}}$, with its natural augmentation $X_{{\square}}\rightarrow X$, then $X^*_{{\square}}$ has a natural augmentation ${\varepsilon}:X^*_{{\square}}\rightarrow X$ to $X$ too. The latter is an *$n$-cubical hyperresolution* of $X$.
\[defn:DeRhamComplexBis\] [@GNPP Définition III.1.10, Proposition III.1.12] If $X\in{\mathcal{C}}$ and $X^*_{{\square}}$ is an $(n+1)$-cubical hyperresolution of $X$ with its natural augmentation ${\varepsilon}:X^*_{{\square}}\rightarrow X$, the *de Rham complex* of $X$ is
$$DR^{{\bullet}}_X:=R{\varepsilon}_*\Omega^{{\square}}_{X^*_{{\square}}}.$$
Although different, the two definitions of the de Rham complex are equivalent. Indeed, pick $X\in{\mathcal{C}}$, let $X_{{\square}}$ be an $(n+1)$-cubical hyperresolution of $X$ with its natural augmentation ${\varepsilon}:X_{{\square}}\rightarrow X$, and let ${\varepsilon}:X_{{\bullet}}\rightarrow X$ be the augmented $n$-semisimplicial resolution associated to it. Let $X^*_{{\square}}$ be the augmented $(n+1)$-cubical hyperresolution of $X$ as in Definition \[defn:hyperrescubGNPP\], and denote by ${\varepsilon}:X^*_{{\square}}\rightarrow X$ its augmentation. In order to show that Definitions \[defn:DeRhamComplex\] and \[defn:DeRhamComplexBis\] are equivalent, we need to prove that
$$R{\varepsilon}_*\Omega^{{\bullet}}_{X_{{\bullet}}}\simeq R{\varepsilon}_*\Omega^{{\square}}_{X^*_{{\square}}}$$
But this is a consequence of the construction we presented in point 3. of Remark \[rmk:cubobj\]: indeed, that construction does not involve the entry $X_{\emptyset}$ of an $(n+1)$-cubical ${\mathcal{C}}$-object, hence all the entries of $X^*_{{\square}}$ can be found in $X_{{\bullet}}$ too (‘bundled together’ by the coproducts); moreover, the augmentation from the entries of $X_{{\bullet}}$ to $X$ are combinations of the augmentations from the entries of $X^*_{{\square}}$ to $X$.
Proof of Theorem 2
==================
Let us write resolution squares for $X$ and $Y$:
$$S(X):=
\xymatrixcolsep{1.5cm}\xymatrixrowsep{1.5cm}\xymatrix{
D_X \ar[r] \ar[d] & \tilde{X} \ar[d]\\
\Sigma_X \ar[r] & X}\hspace{1cm}
S(Y):=
\xymatrixcolsep{1.5cm}\xymatrixrowsep{1.5cm}\xymatrix{
D_Y \ar[r] \ar[d] & \tilde{Y} \ar[d]\\
\Sigma_Y \ar[r] & Y.}$$
Consider the $m$-cubical hyperresolutions $H(Y)_{{\square}}$ of $S(Y)$ and $H(X)_{{\square}}$ of $S(X)$ provided by hypothesis (i). $H(Y)_{{\square}}$ can be rewritten as a $2$-cubical variety of $(m-2)$-cubical varieties which, by Remark \[rmk:iperrescubcub\], are actually $(m-2)$-cubical hyperresolutions of the corresponding entries of $S(Y)$:
$$H(Y)_{{\square}}=
\xymatrixcolsep{1.5cm}\xymatrixrowsep{1.5cm}\xymatrix{
{D_Y}_{{\square}} \ar[r]^f \ar[d]^a & \tilde{Y}_{{\square}} \ar[d]^b\\
{\Sigma_Y}_{{\square}} \ar[r]^g & Y_{{\square}}.}$$
Being a hyperresolution, $H(Y)_{{\square}}$ is in particular of cohomological descent: hence, if $\underline{{\mathbb{C}}}_{H(Y)_{{\square}}}$ denotes the constant sheaf on $H(Y)_{{\square}}$ then $C^{{\bullet}}(H(Y)_{{\square}},\underline{{\mathbb{C}}}_{H(Y)_{{\square}}})$ is acyclic, and the same is true of $C^{{\bullet}}(H(Y)_{{\square}},\underline{{\mathbb{C}}}_{H(Y)_{{\square}}})[2]$; by [@PeSt Corollary 5.28] we deduce the existence of an isomorphism
$$\begin{aligned}
C^{{\bullet}}(Y_{{\square}},\underline{{\mathbb{C}}}_{Y_{{\square}}})\xrightarrow{\simeq}\text{Cone}^{{\bullet}}&[Rb_*C^{{\bullet}}(\tilde{Y}_{{\square}},\underline{{\mathbb{C}}}_{\tilde{Y}_{{\square}}})\oplus Rg_*C^{{\bullet}}({\Sigma_Y}_{{\square}},\underline{{\mathbb{C}}}_{{\Sigma_Y}_{{\square}}})\xrightarrow{(C(a^{\#}),C(b^{\#}))}\\
&\xrightarrow{(C(a^{\#}),C(b^{\#}))} R(g\circ a)_*C^{{\bullet}}({D_Y}_{{\square}},\underline{{\mathbb{C}}}_{{D_Y}_{{\square}}})][-1].\end{aligned}$$
If we shift by $-1$ the short exact sequence of the cone over the morphism $(C(a^{\#}),C(b^{\#}))$ we obtain
$$\begin{aligned}
0&\rightarrow R(g\circ a)_*C^{{\bullet}}({D_Y}_{{\square}},\underline{{\mathbb{C}}}_{{D_Y}_{{\square}}})[-1]\rightarrow\text{Cone}^{{\bullet}}[Rb_*C^{{\bullet}}(\tilde{Y}_{{\square}},\underline{{\mathbb{C}}}_{\tilde{Y}_{{\square}}})\oplus Rg_*C^{{\bullet}}({\Sigma_Y}_{{\square}},\underline{{\mathbb{C}}}_{{\Sigma_Y}_{{\square}}})\xrightarrow{(C(a^{\#}),C(b^{\#}))}\\
&\xrightarrow{(C(a^{\#}),C(b^{\#}))} R(g\circ a)_*C^{{\bullet}}({D_Y}_{{\square}},\underline{{\mathbb{C}}}_{{D_Y}_{{\square}}})][-1]\rightarrow Rb_*C^{{\bullet}}(\tilde{Y}_{{\square}},\underline{{\mathbb{C}}}_{\tilde{Y}_{{\square}}})\oplus Rg_*C^{{\bullet}}({\Sigma_Y}_{{\square}},\underline{{\mathbb{C}}}_{{\Sigma_Y}_{{\square}}})\rightarrow 0\end{aligned}$$
so using the isomorphism above we get the short exact sequence of objects in $D_+(Sh(Y))$
$$0\rightarrow R(g\circ a)_*C^{{\bullet}}({D_Y}_{{\square}},\underline{{\mathbb{C}}}_{{D_Y}_{{\square}}})[-1]\rightarrow C^{{\bullet}}(Y_{{\square}},\underline{{\mathbb{C}}}_{Y_{{\square}}})\rightarrow Rb_*C^{{\bullet}}(\tilde{Y}_{{\square}},\underline{{\mathbb{C}}}_{\tilde{Y}_{{\square}}})\oplus Rg_*C^{{\bullet}}({\Sigma_Y}_{{\square}},\underline{{\mathbb{C}}}_{{\Sigma_Y}_{{\square}}})\rightarrow 0.$$
Now, since the $(m-2)$-cubical hyperresolution ${\varepsilon}:Y_{{\square}}\rightarrow Y$ is of cohomological descent $C^{{\bullet}}(Y_{{\square}},\underline{{\mathbb{C}}}_{Y_{{\square}}})$ is acyclic; hence, if we denote by $Y_{{\bullet}}$ the $(m-3)$-semisimplicial space associated to $S_{{\square}}$ we can write the following isomorphism in $D_+(Sh(Y))$:
$$\underline{{\mathbb{C}}}_{Y}\xrightarrow{\simeq} R{\varepsilon}_*\underline{{\mathbb{C}}}_{Y_{{\bullet}}}.$$
Since all elements of the $(m-3)$-semisimplicial variety $Y_{{\bullet}}$ are smooth, in $D_+(Sh(Y_{{\bullet}}))$ we have an isomorphism $\underline{{\mathbb{C}}}_{Y_{{\bullet}}}\xrightarrow{\simeq}\Omega^{{\bullet}}_{Y_{{\bullet}}}$, so we can substitute $R{\varepsilon}_*\underline{{\mathbb{C}}}^{{\bullet}}_{Y_{{\bullet}}}$ with $R{\varepsilon}_*\Omega^{{\bullet}}_{Y_{{\bullet}}}$; the same can of course be done with the other $(m-2)$-cubical hyperresolutions in $H(Y)_{{\square}}$.
In this way we obtain a short exact sequence of objects in $D_+(Sh(Y))$
$$\label{eq:SESforY}
0\rightarrow R(g\circ a)_*DR_{D_Y}^{{\bullet}}[-1]\rightarrow DR_Y^{{\bullet}}\rightarrow Rb_*DR_{\tilde{Y}}^{{\bullet}}\oplus Rg_*DR_{{\Sigma_Y}}^{{\bullet}}\rightarrow 0$$
which yields the long exact sequence of algebraic de Rham cohomology groups
$$\label{eq:LECSY}
\dots\rightarrow H_{DR}^{{\bullet}}(\Sigma_Y)\oplus H_{DR}^{{\bullet}}(\tilde{Y})\rightarrow H_{DR}^{{\bullet}}(D_Y)\rightarrow H_{DR}^{{\bullet}+1}(Y)\rightarrow\cdots.$$
We want to apply a similar argument to $H(X)_{{\square}}$. We rewrite it as
$$H(X)_{{\square}}=
\xymatrixcolsep{1.5cm}\xymatrixrowsep{1.5cm}\xymatrix{
{D_X}_{{\square}} \ar[r]^f \ar[d]^a & \tilde{X}_{{\square}} \ar[d]^a\\
{\Sigma_X}_{{\square}} \ar[r]^g & X_{{\square}}}$$
where each entry is an $(m-2)$-cubical hyperresolution of the corresponding entry of $S(X)$. Hypothesis (i) implies in particular that each $(m-2)$-cubical hyperresolution in $H(Y)_{{\square}}$ can be embedded into the corresponding $(m-2)$-cubical hyperresolution in $H(X)_{{\square}}$ as a closed $m$-cubical subvariety; if we pass to semisimplicial objects, we deduce the existence of natural closed immersions
$$H(Y)_{{\bullet}}\hookrightarrow H(X)_{{\bullet}}\hspace{0.5cm}Y_{{\bullet}}\hookrightarrow X_{{\bullet}}\hspace{0.5cm}{\Sigma_Y}_{{\bullet}}\hookrightarrow {\Sigma_X}_{{\bullet}}\hspace{0.5cm}\tilde{Y}_{{\bullet}}\hookrightarrow \tilde{X}_{{\bullet}}\hspace{0.5cm}{D_Y}_{{\bullet}}\hookrightarrow {D_X}_{{\bullet}}$$
and of the corresponding restriction of sections functors, which we shall denote by $\Gamma$.
Now we apply the same argument as before to the complex of sheaves on $H(X)_{{\square}}$ given by $R\Gamma_{H(Y)_{{\square}}}\underline{{\mathbb{C}}}_{H(X)_{{\square}}}$.
We have the following commutative diagram of functors:
$$\xymatrixcolsep{1.5cm}\xymatrixrowsep{1.5cm}\xymatrix{
Sh({D_X}_{{\bullet}}) \ar[r]^{\Gamma_{{D_Y}_{{\bullet}}}} \ar[d]^{{\varepsilon}_*} & Sh({D_X}_{{\bullet}}) \ar[d]^{{\varepsilon}_*}\\
Sh(D_X) \ar[r]^{\Gamma_{D_Y}} & Sh(D_X).}$$ From this we deduce the equality of the total derived functors $R({\varepsilon}_*\circ\Gamma_{{D_Y}_{{\bullet}}})=R(\Gamma_{D_Y}\circ{\varepsilon}_*)$. But pushforwards preserve injective objects, and the same holds for $\Gamma_{{D_Y}_{{\bullet}}}$ because ${D_Y}_{{\bullet}}$ is closed in ${D_X}_{{\bullet}}$; since injective objects are adapted to any functor, we obtain isomorphisms
$$\label{eq:comm1}
R{\varepsilon}_*\circ R\Gamma_{{D_Y}_{{\bullet}}}\simeq R({\varepsilon}_*\circ\Gamma_{{D_Y}_{{\bullet}}})=R(\Gamma_{D_Y}\circ{\varepsilon}_*)\simeq R\Gamma_{D_Y}\circ R{\varepsilon}_*.$$
This commutativity holds for all the restriction of sections functors previously listed.
The $(m-2)$-cubical hyperresolution ${\varepsilon}:X_{{\square}}\rightarrow X$ is of cohomological descent so in $D_+(Sh(X))$ we have an isomorphism $\underline{{\mathbb{C}}}_{X}\xrightarrow{\simeq} R{\varepsilon}_*\underline{{\mathbb{C}}}_{X_{{\bullet}}}$. Moreover all elements of the $(m-3)$-semisimplicial variety $X_{{\bullet}}$ are smooth, so in $D_+(Sh(X_{{\bullet}}))$ we also have an isomorphism $\underline{{\mathbb{C}}}_{X_{{\bullet}}}\xrightarrow{\simeq}\Omega^{{\bullet}}_{X_{{\bullet}}}$. If we combine these facts we obtain isomorphisms
$$R\Gamma_Y\underline{{\mathbb{C}}}_X\simeq R\Gamma_Y R{\varepsilon}_*\underline{{\mathbb{C}}}_{X_{{\bullet}}}\simeq R\Gamma_Y R{\varepsilon}_*\Omega_{X_{{\bullet}}}^{{\bullet}}=R\Gamma_Y DR^{{\bullet}}_X$$
that have counterparts for all the $(m-2)$-cubical hyperresolution in $H(X)_{\square}$. Thus we obtain the short exact sequence of objects of $D_+(Sh(X))$
$$\label{eq:SESforX}
0\rightarrow R(g\circ a)_*R\Gamma_{D_Y} DR_{D_X}^{{\bullet}}[-1]\rightarrow\Gamma_Y DR_X^{{\bullet}}\rightarrow Rb_*\Gamma_{\tilde{Y}}DR_{\tilde{X}}^{{\bullet}}\oplus Rg_*\Gamma_{\Sigma_Y}DR_{\Sigma_X}^{{\bullet}}\rightarrow 0$$
which yields the long exact sequence of algebraic de Rham cohomology groups with supports
$$\label{eq:LECSX}
\dots\rightarrow H_{DR,\Sigma_Y}^{{\bullet}}(\Sigma_X)\oplus H_{DR,\tilde{Y}}^{{\bullet}}(\tilde{X})\rightarrow H_{DR,D_Y}^{{\bullet}}(D_X)\rightarrow H_{DR,Y}^{{\bullet}+1}(X)\rightarrow\cdots.$$
Consider now the $m_1$-cubical hyperresolutions ${D_Y}_{{\square}}$ of $D_Y$ and ${D_X}_{{\square}}$ of $D_X$ provided by hypothesis (ii); we want to use the following result:
\[lem:tracemaps\] [@H1 Lemma II.3.1] Let ${\mathbb{K}}$ be a field of characteristic zero. Let $f:X\rightarrow Y$ be either a smooth morphism or a closed immersion of smooth schemes of finite type over ${\mathbb{K}}$. Let $Z$ be a closed subscheme of $X$ such that the induced map $f:Z\rightarrow Y$ is a closed immersion. Then the trace map gives an isomorphism of complexes in $D_+(Sh(Y))$
$$\text{Tr}_f:f_*R\Gamma_Z\Omega_X^{{\bullet}}[2n]\rightarrow R\Gamma_Z\Omega_Y^{{\bullet}}$$
where $n=\dim(X)-\dim(Y)$ and $\Gamma_Z$ denotes the restriction of sections functor.
Fix any $I\in{\square}_{m_1}^*$ and consider any irreducible component $V$ of ${D_X}_I$; two things can happen:
(I) To $V$ there corresponds a unique irreducible component $U$ of ${D_Y}_I$ that admits a closed immersion $i:U\hookrightarrow V$ of codimension $1$.
(II) To $V$ there corresponds no irreducible component of ${D_Y}_I$.
In the first case we can use Lemma \[lem:tracemaps\] with $Y=V$ and $Z=X=U$ to deduce the isomorphism
$$i_*\Omega_U\xrightarrow{\simeq} R\Gamma_{U}\Omega_V[2]\hspace{0.5cm}\text{in }D_+(Sh(V)).$$
In the second case we have a closed immersion $\emptyset\hookrightarrow V$ and using Lemma \[lem:tracemaps\] we find an isomorphism between trivial complexes. If we repeat the same reasoning for all irreducible components of ${D_X}_I$ we obtain an isomorphism
$$i_*\Omega_{{D_Y}_I}\xrightarrow{\simeq} R\Gamma_{{D_Y}_I}\Omega_{{D_X}_I}^{{\bullet}}[2]\hspace{0.5cm}\text{in }D_+(Sh({D_X}_I)).$$
where $i$ is the closed immersion ${D_Y}_I\hookrightarrow {D_X}_I$. If we do the same for all $I\in{\square}^*_{m_1}$ and then switch to semisimplicial objects, we obtain the isomorphism
$$\label{eq:isosheaves}
i_*\Omega_{{D_Y}_{{\bullet}}}^{{\bullet}}\xrightarrow{\simeq} R\Gamma_{{D_Y}_{{\bullet}}}\Omega_{{D_X}_{{\bullet}}}^{{\bullet}}[2]\hspace{0.5cm}\text{in }D_+(Sh({D_X}_{{\bullet}})).$$
where $i$ is the closed immersion ${D_Y}_{{\bullet}}\hookrightarrow{D_X}_{{\bullet}}$.
Denote by $j$ the closed immersion $D_Y\hookrightarrow D_X$ and by ${\varepsilon}_Y$ and ${\varepsilon}_X$ the natural augmentations ${D_Y}_{{\bullet}}\rightarrow D_Y$ and ${D_X}_{{\bullet}}\rightarrow D_X$; we have a commutative diagram
$$\xymatrixcolsep{1.5cm}\xymatrixrowsep{1.5cm}\xymatrix{
Sh({D_Y}_{{\bullet}}) \ar[r]^{i_*} \ar[d]^{{{\varepsilon}_Y}_*} & Sh({D_X}_{{\bullet}}) \ar[d]^{{{\varepsilon}_X}_*}\\
Sh(D_Y) \ar[r]^{j_*} & Sh(D_X).}$$ From this we deduce the equality of the total derived functors $R(j_*\circ{{\varepsilon}_Y}_*)=R({{\varepsilon}_X}_*\circ i_*)$. Both $i_*$ and $j_*$ are exact, because they are pushforwards of closed immersions, so they coincide with their derived functors; moreover, all pushforwards preserve injective objects, which are adapted to any functor. We thus obtain an isomorphism
$$\label{eq:comm2}
R{{\varepsilon}_X}_*\circ i_*=R{{\varepsilon}_X}_*\circ Ri_*\simeq R({{\varepsilon}_X}_*\circ i_*)=R(j_*\circ {{\varepsilon}_Y}_*)\simeq Rj_*\circ R{{\varepsilon}_Y}_*=j_*\circ R{{\varepsilon}_Y}_*$$
as functors from $D_+(Sh({D_Y}_{{\bullet}}))$ to $D_+(Sh(D_X))$.
If we apply to the sides of (\[eq:isosheaves\]) the corresponding $R{\varepsilon}_*$ and use (\[eq:comm2\]) on the left-hand side and (\[eq:comm1\]) on the right-hand side, we obtain an isomorphism between de Rham complexes $j_*DR^{{\bullet}}_{D_Y}\xrightarrow\simeq R\Gamma_{D_Y}DR^{{\bullet}}_{D_X}[2]$ in $D_+(Sh(D_X))$ and so an isomorphism between de Rham cohomology groups. Since we can repeat the previous reasoning for the $m_2$-cubical hyperresolutions of $\Sigma_Y$ and $\Sigma_X$ provided by hypothesis (ii), we obtain isomorphisms
$$\begin{aligned}
\label{eq:morfismibelli}
\begin{split}
&H_{DR}^{{\bullet}}(D_Y)\xrightarrow{\simeq} H_{DR,D_Y}^{{\bullet}+2}(D_X)\\
&H_{DR}^{{\bullet}}(\Sigma_Y)\xrightarrow{\simeq} H_{DR,\Sigma_Y}^{{\bullet}+2c}(\Sigma_X).
\end{split}\end{aligned}$$
These isomorphisms will allow us to relate the long exact sequences (\[eq:LECSY\]) and (\[eq:LECSX\]).
#### $\underline{c=1}$
For $k\geq 1$ we have the following diagram
$$\label{eq:diag1}
\begin{tikzpicture}
\matrix (m) [matrix of math nodes,row sep=3em,column sep=4em,minimum width=2em]
{
H_{DR}^{k-1}(\tilde{Y})\oplus H_{DR}^{k-1}(\Sigma_Y) & H_{DR,\tilde{Y}}^{k+1}(\tilde{X})\oplus H_{DR,\Sigma_Y}^{k+1}(\Sigma_X)\\
H_{DR}^{k-1}(D_Y) & H_{DR,DY}^{k+1}(D_X)\\
H_{DR}^k(Y) & H_{DR,Y}^{k+2}(X)\\
H_{DR}^k(\tilde{Y})\oplus H_{DR}^k(\Sigma_Y) & H_{DR,\tilde{Y}}^{k+2}(\tilde{X})\oplus H_{DR,\Sigma_Y}^{k+2}(\Sigma_X)\\
H_{DR}^k(D_Y) & H_{DR,D_Y}^{k+2}(D_X)\\};
\path[-stealth]
(m-1-1) edge node [left] {${\alpha}$} (m-2-1)
(m-1-1) edge node [above] {$\simeq$} (m-1-2)
(m-2-1) edge node [left] {$\beta$} (m-3-1)
(m-2-1) edge node [above] {$\simeq$} (m-2-2)
(m-3-1) edge node [left] {$\delta$} (m-4-1)
(m-4-1) edge node [left] {$\sigma$} (m-5-1)
(m-4-1) edge node [above] {$\simeq$} (m-4-2)
(m-5-1) edge node [above] {$\simeq$} (m-5-2)
(m-1-2) edge node [right] {${\alpha}'$} (m-2-2)
(m-2-2) edge node [right] {$\beta'$} (m-3-2)
(m-3-2) edge node [right] {$\delta'$} (m-4-2)
(m-4-2) edge node [right] {$\sigma'$} (m-5-2);
\end{tikzpicture}$$
The two squares are commutative. Indeed, the trace maps are functorial by construction (see [@H2 Chapter VI, Section 4.2]) so the same holds for the isomorphisms of cohomology groups they yield, which are the horizontal maps of this diagram; as the vertical maps are obtained from the hyperresolutions of $S(Y)$ and $S(X)$ they are functorial too, and this gives the commutativity of the squares. From this we deduce that $Ker(\delta)\simeq Ker(\delta')$ so we find an isomorphism $\theta_k:H_{DR}^k(Y)\xrightarrow{\simeq} H_{DR,Y}^{k+2}(X)$; moreover, we can choose the $\theta_k$ in such a way that *all* the squares of (\[eq:diag1\]) commute.
Now, $X\setminus Y$ is affine so $H_{DR}^j(X\setminus Y)=0$ for $j\geq\dim(X)+1$ by [@GNPP Corollaire III.3.11(i)]; writing down the long exact sequence of algebraic de Rham cohomology groups associated to the pair $(X,X\setminus Y)$, we find that the morphism $H_{DR,Y}^{k+2}(X)\rightarrow H_{DR}^{k+2}(X)$ is surjective for $k+2=\dim(X)+1$ and an isomorphism for $k+2>\dim(X)+1$. If we pre-compose these morphisms with the corresponding $\theta_k$ we obtain morphisms $H_{DR}^k(Y)\rightarrow H_{DR}^{k+2}(X)$ that are surjective for $k=\dim(X)-1=\dim(Y)$ and isomorphisms for $k>\dim(Y)$; using the comparison theorem [@H1 Theorem IV.1.1] we can conclude that these morphisms exist for singular cohomology too.
#### $\underline{c=0}$
In this case in order to have a diagram like (\[eq:diag1\]) we need $k>2\dim(\Sigma_X)+1$, but this is the only difference with the previous case.
Alexander polynomial and line arrangements
==========================================
By the works of Milnor [@Mi] and Lê [@Le] we know in particular that if $f\in{\mathbb{C}}[x_0,\ldots,x_n]$ is a homogeneous polynomial then the map $f:{\mathbb{C}}^{n+1}\setminus f^{-1}(0)\rightarrow{\mathbb{C}}^*$ is a smooth locally trivial fibration; its generic fibre, usually denoted by $F$, is called *Milnor fibre*. To $F$ we can associate the geometric monodromy operator $h:F\rightarrow F$ and the induced algebraic monodromy operators $T_i:H^i(F,{\mathbb{C}})\rightarrow H^i(F,{\mathbb{C}})$.
Let $C=V(f)\subset{\mathbb{P}}^2$ be a reduced curve. The *Alexander polynomial* of $C$ is the characteristic polynomial of $T_1$, and is denoted by $\Delta_C$.
If $f$ has degree $d$ then $h$ is given by $\underline{x}\mapsto\eta_d\cdot\underline{x}$, where $\zeta_d$ is a primitive $d$-th root of unity; hence both $h$ and $T$ have order $d$, so $T$ is diagonalisable with roots of unity of order $d$ as eigenvalues. Moreover, the Milnor fibre of $C$ is a $d$-fold cover of $U:={\mathbb{P}}^2\setminus C$ and the geometric monodromy $h$ is a generator of the group of deck transformations of $F$; this implies that
$$\Delta_C(t)=(t-1)^{r-1}\prod_{1<k|d}\Phi_k(t)=(t-1)^{r-1}q(t).$$
where $r$ is the number of irreducible components of $C$. We call $q(t)$ the non-trivial part of $\Delta_C(t)$, and say that $\Delta_C(t)$ is *non-trivial* if $q(t)\neq 1$.
The most general tool for computing $\Delta_C$ is a formula by Libgober (see [@L1]) that involves type and relative position of the singularities of $C$; one can use it to verify one of the striking features of the Alexander polynomial: that it is rather hard to find curves for which it is non-trivial. This has led researchers to look for classes of curves for which the non-triviality of $\Delta_C$ could be detected by easier means, without the need to directly compute the whole polynomial. Line arrangements, which we will denote by ${\overline{{\mathcal{A}}}}$, are one of these classes. The reason for this choice is two-fold: on the one hand, they are curves with the simplest possible singularities; on the other hand, one may try and take advantage of the combinatorial nature of such objects, encoded in their intersection semilattices $L({\overline{{\mathcal{A}}}})$.
Indeed, over the course of the years many examples and results have shown that the non-triviality of $\Delta_{{\overline{{\mathcal{A}}}}}$ might be detected simply by looking at $L({\overline{{\mathcal{A}}}})$; in order to present them properly, we need to introduce the notion of multinet [@FY; @PS]:
\[defn:multinets\] Let ${\overline{{\mathcal{A}}}}$ be a line arrangement, ${\mathcal{N}}$ denote a partition of ${\overline{{\mathcal{A}}}}$ into $k\geq 3$ subsets ${\overline{{\mathcal{A}}}}_1,\ldots,{\overline{{\mathcal{A}}}}_k$, $m$ be a ‘multiplicity function’ $m:{\overline{{\mathcal{A}}}}\rightarrow{\mathbb{N}}$ and ${\mathcal{X}}$ be a subset of the multiple points of ${\overline{{\mathcal{A}}}}$; consider moreover the following conditions:
(i) There exists $d\in{\mathbb{N}}$ such that $\sum_{l\in{\overline{{\mathcal{A}}}}_i}m(l)=d$ for all $i=1,\ldots,k$.
(ii) For any $l\in{\overline{{\mathcal{A}}}}_i$ and $l'\in{\overline{{\mathcal{A}}}}_j$ with $i\neq j$ we have $l\cap l'\in{\mathcal{X}}$.
(iii) For all $p\in{\mathcal{X}}$ the integer $n_p:=\sum_{l\in{\overline{{\mathcal{A}}}}_i,p\in l}m(l)$ does not depend on $i$.
(iv) For all $i=1,\ldots,k$ and any $l,l'\in{\overline{{\mathcal{A}}}}_i$, there is a sequence $l=l_0,\ldots,l'=l_r$ such that $l_{j-1}\cap l_j\notin{\mathcal{X}}$.
The couple $({\mathcal{N}},{\mathcal{X}})$ is called:
- a *weak* $(k,d)$-*multinet* if it satisfies (i)-(iii).
- a $(k,d)$-*multinet* if it satisfies (i)-(iv).
- a *reduced* $(k,d)$-*multinet* if it satisfies (i)-(iv) and $m(l)=1$ for all $l\in{\overline{{\mathcal{A}}}}$.
- a $(k,d)$-*net* if it satisfies (i)-(iv) and $n_p=1$ for all $p\in{\mathcal{X}}$; if $d=1$, the $(k,1)$-net is called a *trivial* $k$-net.
We call ${\overline{{\mathcal{A}}}}_1,\ldots,{\overline{{\mathcal{A}}}}_k$ the *classes* of ${\mathcal{N}}$, ${\mathcal{X}}$ its *base locus* and $d$ its *weight*. If $({\mathcal{N}},{\mathcal{X}})$ is a weak $(k,d)$-multinet on ${\overline{{\mathcal{A}}}}$ and $p$ is a multiple point of ${\overline{{\mathcal{A}}}}$, we define the *support of $p$ with respect to ${\mathcal{N}}$* as
$$supp_{{\mathcal{N}}}(p):=\{{\alpha}\in\{1,\ldots,k\}|p\in l\text{ for some }l\in{\overline{{\mathcal{A}}}}_{{\alpha}}\}.$$
Observe that the notion of multinet is a mathematically precise formalisation of the notion of symmetry.
(0.75,0.5) – (3.25,5.5); (0.5,1) – (5.5,1); (2.75,5.5) – (5.25,0.5); (3,0.5) – (3,5.5); (0.25,0.5) – (4.5,3.33); (1.5,3.33) – (5.75,0.5); (1,1) circle \[radius=2pt\]; (5,1) circle \[radius=2pt\]; (3,5) circle \[radius=2pt\]; (3,2.33) circle \[radius=2pt\];
A $3$-net on the $A_3$ line arrangement.
We have the following result, which can be obtained as a consequence of [@PS Theorem 8.3] or combining [@FY Theorem 3.11] with [@DP Theorem 3.1(i)]:
\[thm:SuffCond\] If ${\overline{{\mathcal{A}}}}$ admits a reduced multinet then its Alexander polynomial is non-trivial.
This sufficient condition for the non-triviality of the Alexander polynomial of a line arrangement ${\overline{{\mathcal{A}}}}$ is not necessary; however, all line arrangements with non-trivial Alexander polynomial known so far admit at least a $(k,d)$-multinet (and if the arrangement is non-central we have $k=3$ or $k=4$ only). This suggests that multinets somehow control the non-triviality of $\Delta_{{\overline{{\mathcal{A}}}}}$; indeed, for some classes of line arrangements such a dependence has been established ([@PS Theorem 1.6, Theorem 1.2]):
Let ${\overline{{\mathcal{A}}}}$ be a line arrangement with only double and triple points, then $\Delta_{{\overline{{\mathcal{A}}}}}(t)=(t-1)^{|{\mathcal{A}}|-1}\Phi_3(t)^{\beta_3({\overline{{\mathcal{A}}}})}(t)$ with $0\leq \beta_3({\overline{{\mathcal{A}}}})\leq 2$ and $\beta_3({\overline{{\mathcal{A}}}})\neq 0$ if and only if ${\overline{{\mathcal{A}}}}$ admits a $3$-net.
This result, together with the many examples gathered throughout the years, led Papadima and Suciu to formulate the following conjecture:
The Alexander polynomial of a line arrangement ${\overline{{\mathcal{A}}}}$ has the form
$$\Delta_{{\overline{{\mathcal{A}}}}}(t)=(t-1)^{|{\overline{{\mathcal{A}}}}|-1}\Phi_3(t)^{\beta_3({\overline{{\mathcal{A}}}})}[\Phi_2(t)\Phi_4(t)]^{\beta_2({\overline{{\mathcal{A}}}})}$$
The numbers $\beta_i({\overline{{\mathcal{A}}}})$ are the modular Aomoto-Betti numbers of ${\overline{{\mathcal{A}}}}$ (see [@PS Section 3]), and they only depend on $L({\overline{{\mathcal{A}}}})$ and $i$. Recent results [@MPP; @D2; @DSt] show that this conjecture is valid for all complex reflection arrangements.
The only known arrangement with $\beta_2\neq 0$ is the Hesse arrangement: it can be constructed considering the nine inflection points of an elliptic curve and taking all lines that contain exactly three such points. We obtain an arrangement with twelve lines and nine point of order four with $\beta_2=2$.
The rest of this section is devoted to collecting some sparse results we shall need in the following one.
\[lem:ThomIso\] Suppose $f(x_0,\ldots,x_n)$ has an isolated singularity at the origin and $g(y_0,\ldots,y_n)$ has an arbitrary singularity at the origin. Call $F$, $G$ and $F\oplus G$ the Milnor fibres of $f$, $g$ and $f+g$ respectively, and denote by $T^i_f$, $T^i_g$ and $T^i_{f+g}$ the monodromy operators on the $i$-th cohomology groups. There is an isomorphism
$$\tilde{H}^{n+k+1}(F\oplus G,{\mathbb{Q}})\simeq \tilde{H}^n(F,{\mathbb{Q}})\otimes\tilde{H}^k(G,{\mathbb{Q}})\hspace{0.8cm}\text{for }k=0,\ldots,n$$
respecting the monodromy operators: $T^{f+g}_{n+k+1}=T^f_n\otimes T^g_k$.
This is a consequence of [@D1 Lemma 3.3.20, Corollary 3.3.21].
If $f(x_0,\ldots,x_n)=0$ is a homogeneous polynomial defining an isolated hypersurface singularity, the *Steenbrink spectrum* of $f$ is the formal sum of rational numbers
$$sp(f):=\sum_{{\alpha}\in{\mathbb{Q}}}{\alpha}\nu({\alpha})$$
where $\nu({\alpha})$ is the dimension of the $e^{-2\pi i{\alpha}}$-eigenspace of the monodromy operator acting on $Gr_F^{{\lfloor n-{\alpha}\rfloor}}H^n(F)$. If $f$ has degree $d$ and weights $w_i$ then
$$\nu({\alpha})=\text{dim }M(f)_{({\alpha}+1)d-w}$$
where $M(f)$ is the Milnor algebra of $f$ and $w$ is the sum of the $w_i$’s. The spectrum is symmetric around $\frac{n-1}{2}$ and $\nu({\alpha})=0$ for ${\alpha}\notin (-1,n)$.
Proof of Theorem 3
==================
Consider a line arrangement ${\overline{{\mathcal{A}}}}=V(f)$ of degree $n$ in ${\mathbb{P}}^2$ having two multiple points $P_1$ and $P_2$ such that any line of the arrangement passes through $P_1$ or $P_2$. We call $p$ the multiplicity of $P_1$, $q$ the multiplicity of $P_2$ and assume, without loss of generality, that $p\geq q$; by our hypothesis on the arrangement, we have $p+q=n$ and all multiple points of ${\overline{{\mathcal{A}}}}$ different from $P_1$ and $P_2$ have multiplicity two. Up to an isomorphism of ${\mathbb{P}}^2$, we can assume that $P_1=(0:0:1)$ and $P_2=(0:1:0)$.
Assume ${\overline{{\mathcal{A}}}}$ admits a weak $(k,d)$-multinet $({\mathcal{N}},{\mathcal{X}})$ with classes ${\overline{{\mathcal{A}}}}_1,\ldots,{\overline{{\mathcal{A}}}}_k$. By definition of weak multinet and support, if $l\in{\overline{{\mathcal{A}}}}$ then $|supp_{{\mathcal{N}}}(l)|\in\{1,k\}$. Thus, any double point of ${\overline{{\mathcal{A}}}}$ is the intersection of lines belonging to the same class; but since all lines of ${\overline{{\mathcal{A}}}}$ contain at least a double point, this means that all lines belong to the same class, which is a contradiction. Hence ${\overline{{\mathcal{A}}}}$ does not admit weak multinets.
A polynomial $f\in{\mathbb{C}}[x_0,x_1,x_2]$ describing an arrangement of this type can be written as
$$f=\prod_{i=1}^p(x_0-\lambda_i x_1)\prod_{i=1}^q(x_0-\mu_i x_2)$$
with at most one of the ${\lambda}_i,\mu_i$ equal to zero. Call $g=y^n+z^n$ and consider the threefold $X:=V(g-f)\subset{\mathbb{P}}^4$: the natural projection map $\pi:X\rightarrow{\mathbb{P}}^2$ s.t. $(y:z:x_0:x_1:x_2)\mapsto (x_0:x_1:x_2)$ is a rational map with discriminant $X\cap V(x_0,x_1,x_2)$.
Any hyperplane $H:=V({\alpha}x_1-\beta x_0)\subset{\mathbb{P}}^4$ cuts a surface from $X$; if we assume ${\alpha}\neq 0$ and call $s:=\beta/{\alpha}$ then this surface, which we denote by $Y_s$, is a hypersurface of ${\mathbb{P}}^3$ defined by the polynomial
$$f_s:=y^n+z^n-h(s)x_0^p\prod_{i=1}^q(x_0-\mu_i x_2)\hspace{1cm}\text{ where }h(s):=\prod_{i=1}^p(1-{\lambda}_i s).$$
If ${\alpha}=0$ we denote the corresponding surface by $Y_{\infty}$, whose defining polynomial as hypersurface of ${\mathbb{P}}^3$ is
$$f_{\infty}:=y^n+z^n-(-1)^n\left(\prod_{i=1}^p{\lambda}_i\prod_{i=1}^q\mu_i\right)x_1^px_2^q.$$
If we call $B$ the blow-up of ${\mathbb{P}}^2$ at $P_1$ and set $X':=X\times_{{\mathbb{P}}^2}B$ we obtain
$$X'\simeq\{(y:z:x_0:x_1:x_2)\times({\alpha}:\beta)\text{ s.t. }x_0\beta=x_1{\alpha}, y^n+z^n-f(x_0,x_1,x_2)=0\}$$
and we can write the following diagram
\(m) \[matrix of math nodes,row sep=3em,column sep=4em,minimum width=2em\] [ X & \^2\
X’ & B\
& \^1\
]{}; (m-1-1) edge \[dashed\] node \[above\] [$\pi$]{} (m-1-2) (m-2-1) edge node (m-1-1) (m-2-1) edge node (m-2-2) (m-2-1) edge node \[below\] [$\psi$]{} (m-3-2) (m-2-2) edge node \[right\] [$\pi_2$]{} (m-1-2) (m-2-2) edge node \[right\] [$\pi_1$]{} (m-3-2);
where $\pi_i$ is the projection from $B$ onto ${\mathbb{P}}^i$, $\psi$ is given by $(y:z:x_0:x_1:x_2)\times ({\alpha}:\beta)\mapsto ({\alpha}:\beta)$ and the maps from $X'$ are the projections. The map $\psi:X'\rightarrow{\mathbb{P}}^1$ is a fibration in surfaces: we have in fact
$$\begin{aligned}
\psi^{-1}(1:s)\simeq Y_s\hspace{1cm}\psi^{-1}(0:1)\simeq Y_{\infty}.\end{aligned}$$
Both the threefold $X$ and the surfaces $Y_s$, $Y_{\infty}$ are singular, with singular loci given by
$$\begin{aligned}
\Sigma_X&=\{P_p,P_q\}\cup\{(0:0:a:b:c)|(a:b:c)\text{ is a double point of }{\overline{{\mathcal{A}}}}\}.\\
\Sigma_{Y_s}&=\begin{cases}
\begin{array}{cl}
\{P_p\}&\text{ if }h(s)\neq 0.\\
L_s&\text{ if }h(s)=0.
\end{array}
\end{cases}\\
\Sigma_{Y_{\infty}}&=\{P_p,P_q\}.\end{aligned}$$
where $P_p:=(0:0:0:0:1)$, $P_q:=(0:0:0:1:0)$ and $L_s:=\{(0:0:a:as:b)\text{ s.t. }(a:b)\in{\mathbb{P}}^1\}$.
Observe that the singularities of $X$ at the points in $\Sigma_X$ different from $P_p$ and $P_q$ are topologically equivalent to $y^n+z^n-v^2-w^2=0$; the singularity in $P_k$ is topologically equivalent to $y^n+z^n-v^k-w^k=0$ for $k=p,q$. The singularity of $Y_{\infty}$ (and $Y_s$ for $h(s)\neq 0$) in $P_p$ is topologically equivalent to $y^n+z^n-v^p=0$, while the one in $P_q$ is topologically equivalent to $y^n+z^n-v^q=0$.
Assume now $s_1$ and $s_2$ are not roots of $h(s)$, then we can find a diffeomorphism $Y_{s_1}\rightarrow Y_{s_2}$. Pick in fact $(y:z:x_0:s_1x_0:x_2)\in Y_{s_1}$, which satisfies $y^n+z^n-h(s_1)x_0^p\prod_{i=1}^q(x_0-\mu_i x_2)=0$: we can find $(\alpha y:\beta z:x_0:s_2x_0:x_2)\in Y_{s_2}$ for simple values of $\alpha$ and $\beta$. Namely, in order to have $(\alpha y:\beta z:x_0s_2x_0:x_2)\in S_{s_2}$ the equation $\alpha^n y^n+\beta^n z^n-h(s_2)x_0^p\prod_{i=1}^q(x_0-\mu_i x_2)=0$ must be satisfied; as $x_0^p\prod_{i=1}^q(x_0-\mu_i x_2)=\frac{y^n+z^n}{h(s_1)}$, we need to find $\alpha$ and $\beta$ satisfying
$$\alpha^n y^n+\beta^n z^n-h(s_2)\frac{y^n+z^n}{h(s_1)}=0\Longleftrightarrow y^n\left(\alpha^n-\frac{h(s_2)}{h(s_1)}\right)=z^n\left(\beta^n-\frac{h(s_2)}{h(s_1)}\right)$$
and this gives $\alpha^n=\beta^n=\frac{h(s_2)}{h(s_1)}=:\gamma$.
If we call $\Delta:=\{(0:1)\}\cup\{(1:s)|h(s)=0\}$ we obtain a locally trivial fibration $T'-\psi^{-1}(\Delta)\rightarrow{\mathbb{P}}^1-\Delta$, with the $Y_s$ with $h(s)\neq 0$ as generic fibre. We now compute the monodromy of $\psi$ around one of its special fibres, i.e. one of the $Y_s$ with $s\in\Delta$:
The geometric monodromy around a special fibre of $\psi$ is given by
$$\phi:Y_s\rightarrow Y_s\text{ s.t. }(y:z:x_0:sx_0:x_2)\mapsto (\eta_ny:\eta_nz:x_0:sx_0:x_2)$$
where $\eta_n$ is an $n$-th primitive root of unity.
Assume the special fibre we are considering is $Y_{\frac{1}{\lambda_1}}$. Consider a loop $s(t)=\frac{1}{\lambda_1}+re^{2\pi it}$ around $\frac{1}{\lambda_1}$: by the above discussion, the diffeomorphism between $Y_{s(0)}$ and $Y_{s(t)}$ is governed by
$$\gamma_t:=\frac{h(s(t))}{h(s(0))}=e^{2\pi it}\prod_{i=2}^p\frac{\lambda_1-re^{2\pi it}\lambda_1\lambda_i-\lambda_i}{\lambda_1-r\lambda_1\lambda_i-\lambda_i}$$
We can choose branch cuts for the $n$-th root function in such a way that, for $r$ small enough, the loop $s(t)$ remains in a zone of the complex plane in which the $n$-th root is a single-valued function. The only indeterminacy lies then in the term $e^{2\pi it}$; since we look for automorphisms $\phi_t:Y_{s(0)}\rightarrow Y_{s(t)}$ giving the identity for $t=0$, we deduce that the monodromy action $\phi$ on $Y_{s(0)}$ is given by $y\mapsto\eta_n y$, $z\mapsto\eta_n z$.
Fix now an $s\neq\infty$ s.t. $h(s)\neq 0$ and write $Y:=Y_s$; what we want to do is the following:
(1) Verify that the hypotheses of Theorem \[thm:risultato\] are satisfied for $Y\subset X$, so that we can find a surjective morphism $H^2(Y)\twoheadrightarrow H^4(X)$.
(2) Show that this morphism specialises to $H^2(Y)_{\text{prim}}^{T_{\phi}}\twoheadrightarrow H^4(X)_{\text{prim}}$, where $T_{\phi}$ denotes the algebraic monodromy.
(3) Bound the dimension of $H^2(Y)_{\text{prim}}^{T_{\phi}}$, and thus of $H^4(X)_{\text{prim}}$, to deduce that the Alexander polynomial of ${\overline{{\mathcal{A}}}}$ is trivial thanks to Lemma \[lem:ThomIso\].
Step 1
------
Observe first that if $Y=X\cap H_s$ with $H_s=V(x_1-sx_0)$ then $\Sigma_Y=\Sigma_X\cap H_s$. We begin by finding explicit resolutions $\tilde{X}$ and $\tilde{Y}$ of $X$ and $Y$: since we have explicit equations for the singularities, this is just a matter of computations. It is straightforward to check that resolving $P_p$ yields as exceptional divisors:
####
On $\tilde{X}$ a smooth surface $E$ and $p$ disjoint planes $W_1,\ldots,W_p$, each of them intersecting $E$ in a line $L_i$, and on $\tilde{Y}$ a smooth curve $F=E\cap H$ such that $F\cap L_i=\emptyset$ for any $i=1,\ldots,p$.
####
On $\tilde{X}$
- A smooth surface $E$.
- Planes $Z^{(t)}_i$ with $i=1,\ldots,p$ and $t=0,\ldots,r$ such that $$\begin{aligned}
\begin{array}{cl}
Z^{(t_1)}_i\cap Z^{(t_2)}_j&=\begin{cases}
\begin{array}{cl}
\text{a line}&\text{if }i=j\text{ and }t_1=t_2\pm 1.\\
\emptyset&\text{otherwise}.
\end{array}
\end{cases}\\
Z^{(t)}_i\cap E&=\begin{cases}
\begin{array}{cl}
\text{a line }&\text{if }t=0.\\
\emptyset&\text{otherwise}.
\end{array}
\end{cases}
\end{array}
\end{aligned}$$
- Planes $Y^{(t)}_i$ with $i=1,\ldots,n$ and $t=0,\ldots,u$ such that $$\begin{aligned}
\begin{array}{cl}
Y^{(t_1)}_i\cap Y^{(t_2)}_j&=\begin{cases}
\begin{array}{cl}
\text{a line}&\text{if }i=j\text{ and }t_1=t_2\pm 1.\\
\emptyset&\text{otherwise}.
\end{array}
\end{cases}\\
Y^{(t)}_i\cap E&=\begin{cases}
\begin{array}{cl}
\text{a line }&\text{if }t=0.\\
\emptyset&\text{otherwise}.
\end{array}
\end{cases}\\
Y^{(t_1)}_i\cap Z^{(t_2)}_j&=\emptyset
\end{array}
\end{aligned}$$
and on $\tilde{Y}$
- A smooth curve $F=E\cap H$.
- Lines $K^{(t)}_i=Y^{(t)}_i\cap H$ with $i=1,\ldots,n$ and $t=0,\ldots,u$ such that $$\begin{aligned}
\begin{array}{cl}
K^{(t_1)}_i\cap K^{(t_2)}_j&=\begin{cases}
\begin{array}{cl}
\text{a point}&\text{if }i=j\text{ and }t_1=t_2\pm 1.\\
\emptyset&\text{otherwise}.
\end{array}
\end{cases}\\
K^{(t)}_i\cap F&=\begin{cases}
\begin{array}{cl}
\text{a point}&\text{if }t=0.\\
\emptyset&\text{otherwise}.
\end{array}
\end{cases}
\end{array}
\end{aligned}$$
The resolution of the points $P_q$ and $(0:0:a:b:c)$ s.t. $(a:b:c:)$ is a double point of ${\overline{{\mathcal{A}}}}$, which belong to $X$ only, yield the same divisors as $P_p$ but with $p$ replaced by $q$ and $2$ respectively.
We write the resolution squares of $X$ and $Y$ as
$$S(X)=
\xymatrixcolsep{1.5cm}\xymatrixrowsep{1.5cm}\xymatrix{
D_X \ar[r] \ar[d] & \tilde{X} \ar[d]\\
\Sigma_X \ar[r] & X}\hspace{1cm}
S(Y)=
\xymatrixcolsep{1.5cm}\xymatrixrowsep{1.5cm}\xymatrix{
D_Y \ar[r] \ar[d] & \tilde{Y} \ar[d]\\
\Sigma_Y \ar[r] & Y.}$$
We use Theorem \[thm:iperrescubcub\] to obtain an $m$-cubical hyperresolution $H(X)_{{\square}}$ of $S(X)$ for some $m$. Recall the process begins by taking a resolution of $S(X)$ as in [@GNPP Théorème I.2.6] (i.e. via separation and resolution of the irreducible components) and then considering the associated resolution square; this construction is iterated until we obtain an $m$-cubical hyperresolution.
We observe the following:
(a) irreducible components of $D_X$ and $D_Y$ and intersections thereof are smooth, and each irreducible component of $D_Y$ is an hyperplane section of an irreducible component of $D_X$.
(b) $\Sigma_X$ and $\Sigma_Y$ are smooth, with the latter being a hyperplane section of the former. The same goes for $\tilde{X}$ and $\tilde{Y}$.
Given how $H(X)_{{\square}}$ is constructed, these facts imply that considering in each entry of $H(X)_{{\square}}$ the corresponding hyperplane section yields an $m$-cubical hyperresolution $H(Y)_{{\square}}$ of $S(Y)$, so there is a natural closed immersion $H(Y)_{{\square}}\hookrightarrow H(X)_{{\square}}$; hence hypothesis (i) of Theorem \[thm:risultato\] is satisfied.
Now we need to find suitable hyperresolutions of two of the four entries of $S(X)$ and $S(Y)$. For $\Sigma_X$ and $\Sigma_Y$ there is nothing to do, because they are smooth and hence coincide with their hyperresolutions: we thus immediately find a closed immersion $\Sigma_Y\hookrightarrow\Sigma_X$ of codimension zero. For $D_X$ and $D_Y$ we distinguish again two cases:
####
To find a cubical hyperresolution of $D_X$ we use Theorem \[thm:iperrescub\] (again, we separate and resolve its irreducible components $Z_i$); all $Z_i$ are smooth, and their pairwise intersections are either lines $L_j$ or the empty set. We find a resolution square (a $2$-cubical hyperresolution) like this
$${D_X}_{{\square}}:=
\xymatrixcolsep{1.5cm}\xymatrixrowsep{1.5cm}\xymatrix{
K'_X \ar[r] \ar[d] & D'_X \ar[d]\\
K_X \ar[r] & D_X}$$
where
$$\begin{aligned}
D'_X&=\coprod Z_i\\
K_X&=\bigcup L_j\\
K'_X&=(\coprod L^0_j)\coprod(\coprod L^1_j).\end{aligned}$$
If $L_j=Z_{i_0}\cap Z_{i_1}$ then $L^0_i$ denotes the line $L_i$ thought of as belonging to $Z_{i_0}$ and $L^1_i$ denotes the line $L_i$ thought of as belonging to $Z_{i_1}$.
Recall that since $p=q$ we have that $D_Y$ is a smooth curve $F$, so the procedure described in Theorem \[thm:iperrescub\] returns $D_Y$ itself as cubical hyperresolution; we will thus need to construct a different cubical hyperresolution. First we consider the $1$-cubical variety given by the identity of $D_Y$, then we choose a point $Q\in D_Y$ and a point $Q'$ belonging to some $L_i=Z_{i_0}\cap Z_{i_1}$, and consider the 2-cubical variety
$${D_Y}_{{\square}}:=
\xymatrixcolsep{1.5cm}\xymatrixrowsep{1.5cm}\xymatrix{
\{Q'\} \ar[r]^a \ar[d]^{id} & D_Y \ar[d]^{id}\\
\{Q'\} \ar[r]^c & D_Y}$$
where $a$ and $c$ send $Q'$ to $Q$. It is of cohomological descent because it is a discriminant square for $D_Y$; moreover, all of its entries are smooth and all the morphisms are proper. This means that ${D_Y}_{{\square}}$ is a cubical hyperresolution of $D_Y$, and it is readily verified that it satisfies the hypothesis (ii) of Theorem \[thm:risultato\].
####
In this case it is enough to apply Theorem \[thm:iperrescub\] to both $D_X$ and $D_Y$ to obtain the desired cubical hyperresolutions, thanks to point (a) above.
Hence hypothesis (ii) is satisfied too, and by part (b) of Theorem \[thm:risultato\] we obtain a surjective morphism $H^2(Y)\twoheadrightarrow H^4(X)$.
Step 2
------
Denote by $\gamma$ the surjective morphism $H^2(Y)\twoheadrightarrow H^4(X)$ provided by Theorem \[thm:risultato\].
\[prop:monodromy\] We have $\gamma(H^2(Y))=\gamma(H^2(Y)^{T_{\phi}})$, so there is a surjective morphism
$$\label{eq:gys1}
H^2(Y)^{T_{\phi}}\twoheadrightarrow H^4(X).$$
We denote by $\psi'$ the natural morphism $\tilde{X}\rightarrow{\mathbb{P}}^1$ whose generic fibre is $\tilde{Y}$, by $\phi'$ the geometric monodromy on $\tilde{Y}$ and by $T_{\phi'}$ the induced automorphisms on the cohomology groups of $\tilde{Y}$. If we denote by $\tilde{\gamma}$ the usual Gysin morphism $H^2(\tilde{Y})\rightarrow H^4(\tilde{X})$ then by the global invariant cycle theorem (see for example [@V2 Theorem 4.24]) we have
$$\tilde{\gamma}(H^2(\tilde{Y}))=\tilde{\gamma}(H^2(\tilde{Y})^{T_{\phi'}}).$$
From the resolution square of $X$ we obtain the exact sequences of MHS (see [@PeSt Definition-Lemma 5.17])
$$\begin{aligned}
\dots\rightarrow H^3(D_X)\rightarrow H^4(X)\rightarrow H^4(\tilde{X})\rightarrow\cdots;\end{aligned}$$
since the Hodge structure on $H^4(X)$ is pure by [@PeSt Proposition 6.33] and $H^3(D_X)$ has weights up to $3$, we deduce that $H^4(X)\rightarrow H^4(\tilde{X})$ is injective.
Now we observe that the diagram
$$\label{eq:diagfin}
\xymatrixcolsep{1.5cm}\xymatrixrowsep{1.5cm}\xymatrix{
H^2(Y) \ar@{->>}[r]^{\gamma} \ar[d] & H^4(X) \ar@{^{(}->}[d]\\
H^2(\tilde{Y}) \ar@{->>}[r]^{\tilde{\gamma}} & H^4(\tilde{X})}$$
is commutative. This can be read off the following diagram (there is a slight abuse of notation: we have switched to singular cohomology, but we maintain the names we gave to morphisms in the algebraic setting):
$$\xymatrixcolsep{1.5cm}\xymatrixrowsep{1.5cm}\xymatrix{
H^2(Y) \ar[r]^{\theta_2} \ar[d]^{\delta} & H_Y^4(X) \ar@{->>}[r] \ar[d]^{\delta'} & H^4(X) \ar@{^{(}->}[d]\\
H^2(\tilde{Y}) \ar[r]^{\simeq} & H_{\tilde{Y}}^4(\tilde{X}) \ar@{->>}[r] & H^4(\tilde{X}).
}$$ The left square is commutative, because it is simply the equivalent, in singular cohomology, of the third square of diagram (\[eq:diag1\]); the right square is commutative too, because the vertical maps are pullbacks and the horizontal maps come from the long exact sequences of the pairs $(X,X\setminus Y)$ and $(\tilde{X},\tilde{X}\setminus\tilde{Y})$ respectively, which are functorial. Since the compositions of the maps on the top and on the bottom give exactly the Gysin morphisms $\gamma$ and $\gamma'$, we obtain the commutativity of diagram \[eq:diagfin\].
The pullback morphism $H^2(Y)\rightarrow H^2(\tilde{Y})$ maps the subspace $V\subset H^2(Y)$ which is not $T_{\phi}$-invariant to the subspace $\tilde{V}\subset H^2(\tilde{Y})$ which is not $T_{\phi'}$-invariant, and the latter is sent to zero by $\tilde{\gamma}$ by the global invariant cycle theorem; since the diagram (\[eq:diagfin\]) is commutative and $H^4(X)\rightarrow H^4(\tilde{X})$ is injective, we deduce that $\gamma(V)=0$.
\[lem:primitive\] The morphism $H^2(Y)\twoheadrightarrow H^4(X)$ specialises to $H^2(Y)_{\text{prim}}\twoheadrightarrow H^4(Y)_{\text{prim}}$.
Assume $H_0$ is the hyperplane of ${\mathbb{P}}^4$ that cuts $Y$ from $X$, and choose another hyperplane $H$ of ${\mathbb{P}}^4$ such that $H\cap \Sigma_X=\emptyset$; we can find a resolution of singularities $\pi_X:\tilde{X}\rightarrow X$ such that $\tilde{Y}:=\tilde{X}\cap\pi_X^{-1}(H_0)$ is smooth. If we call $\pi_Y:\tilde{Y}\rightarrow Y$ the restriction of $\pi_X$ to $\tilde{Y}$, we can write functorial morphisms
$$\begin{aligned}
\begin{split}
\pi_Y^*:H^2(Y)\rightarrow H^2(\tilde{Y})\hspace{0.5cm} & \text{(pullback)}\\
\pi_X^*:H^4(X)\rightarrow H^4(\tilde{X})\hspace{0.5cm} & \text{(pullback)}\\
\tilde{\gamma}:H^2(\tilde{Y})\twoheadrightarrow H^4(\tilde{X})\hspace{0.5cm} & \text{(Gysin).}
\end{split}\end{aligned}$$
Since $\pi_X^{-1}(H)\simeq H$ and $\pi_Y^{-1}(H_0\cap H)\simeq H_0\cap H$ we deduce that $\tilde{\gamma}([\pi_Y^{-1}(H_0\cap H)])=[\pi_X^{-1}(H)]$; moreover, the functoriality of the pullback maps implies that $\pi_X^*([H])=[\pi_X^{-1}(H)]$ and $\pi_Y^*([H_0\cap H])=[\pi_Y^{-1}(H_0\cap H)]$.
The commutativity of (\[eq:diagfin\]) now implies that $\gamma([H_0\cap H])$ can be written as $[H]+Ker(\pi_X^*)$, but since $\pi_X^*$ is injective it must be $\gamma([H_0\cap H])=[H]$; this proves our claim.
The commutativity of (\[eq:diagfin\]) actually allows us to further refine these results. Since $H^2(\tilde{Y})$ is a pure HS of weight $2$, the kernel of $H^2(Y)\rightarrow H^2(\tilde{Y})$ contains $W_1 H^2(Y)$; this, together with the injectivity of $H^4(X)\rightarrow H^4(\tilde{X})$ and the commutativity of (\[eq:diagfin\]), implies that $W_1H^2(Y)\subset Ker(\gamma)$. The same holds true if we restrict first to primitive cohomology groups and then to the invariant part of $H^2(Y)_{\text{prim}}$ under the action of $T_{\phi}$, which proves that
$$\label{eq:trick}
\gamma(H^2(Y)_{\text{prim}}^{T_{\phi}})=\gamma(W_2 H^2(Y)_{\text{prim}}^{T_{\phi}}).$$
Step 3
------
If we call $U:={\mathbb{P}}^4\setminus X$ then from the long exact sequence of MHS associated to the pair $({\mathbb{P}}^4,X)$ we deduce
$$\dots\rightarrow H^4({\mathbb{P}}^4)\rightarrow H^4(X)\rightarrow H^5_c(U)\rightarrow 0.$$
By using Poincaré duality and the isomorphism of homology and cohomology we obtain the isomorphism $H^5_c(U)\simeq H^3(U)^{\vee}$; since the map $H^4({\mathbb{P}}^4)\rightarrow H^4(X)$ is injective we obtain $H^4(X)_{\text{prim}}\simeq H^3(U)^{\vee}$. This implies in particular that $\dim H^4(X)_{\text{prim}}=\dim H^3(F_{g-f})^{T_{g-f}}$.
Similarly, if we call $U':={\mathbb{P}}^3\setminus Y$ from the long exact sequence of MHS associated to the pair $({\mathbb{P}}^3,Y)$ we deduce $$\dots\rightarrow H^2({\mathbb{P}}^3)\rightarrow H^2(Y)\rightarrow H^3_c(U')\rightarrow 0$$ and we obtain $H^2(Y)_{\text{prim}}\simeq H^3(U')^{\vee}$. Since we will need to study in detail the MHS on $H^2(Y)_{\text{prim}}$ we write the Poincaré duality isomorphism at the level of MHS: we have
$$\label{eq:ISOMHS}
H^2(Y)_{\text{prim}}\simeq H^3(U')^{\vee}(-3).$$
In order to simplify notations we call $V:=H^2(Y)_{\text{prim}}$. The isomorphism above implies the following equality of mixed Hodge numbers:
$$\label{eq:isoMHN}
h^{p,q}(V)=h^{3-p,3-q}(H^3(U')).$$
Since $V$ is a mixed Hodge substructure of $H^2(Y)$ it has weights $\leq 2$, and its Hodge filtration can be written as
$$\begin{array}{c}
0=F^3V\subset F^2V\subset F^1V\subset F^0V=V
\end{array}$$ while for $H^3(U')$ we have
$$\begin{array}{c}
0=F^4H^3(U')\subset F^3H^3(U')\subset F^1H^3(U')\subset F^0H^3(U')=H^3(U').
\end{array}$$ On $H^3(U')$ we also have the polar filtration:
$$0=P^4 H^3(U')\subset\dots\subset P^1 H^3(U')=H^3(U').$$
Since the action of $T_{\phi}$ is compatible with all these filtrations, from (\[eq:isoMHN\]), the inclusion $F^k H^3(U')\subseteq P^k H^3(U')$ given by [@D1 Theorem 6.1.31] and the symmetry of mixed Hodge numbers we deduce
$$\begin{aligned}
\label{eq:ineqs}
\begin{split}
h^{2,0}(V^{T_{\phi}})+h^{1,0}(V^{T_{\phi}})+h^{0,0}(V^{T_{\phi}})&\leq\dim P^3H^3(U')^{T_{\phi}}\\
h^{2,0}(V^{T_{\phi}})+2h^{1,0}(V^{T_{\phi}})+h^{0,0}(V^{T_{\phi}})+h^{1,1}(V^{T_{\phi}})&\leq\dim P^2H^3(U')^{T_{\phi}}.
\end{split}\end{aligned}$$
We call now $R:={\mathbb{C}}[y,z,x_0,x_2]$, $f_Y\in R$ the polynomial defining $Y$ and $J_{f_Y}\subset R$ the associated Jacobian ideal; for $t=1,2,3$ we have maps
$$\label{eq:rec}
(R/J_{f_Y})_{tn-4}\twoheadrightarrow Gr_P^{4-t} H^3(U')=P^{4-t}H^3(U')/P^{5-t}H^3(U')$$
Any class in $P^{k}H^3(U')$ has a representative of the form
$$\omega_h:=\frac{h\Omega}{f_Y^k}\hspace{1.55cm}\text{ with }h\in R_{kn-4}$$
(where $\Omega=ydz\wedge dx_0\wedge dx_2-zdy\wedge dx_0\wedge dx_2+x_0dy\wedge dz\wedge dx_2-x_2 dy\wedge dz\wedge dx_0$), and $T_{\phi}$ acts on it by multiplying $y$ and $z$ by $\eta_n$; this means that if $h(y,z,x_0,x_2)$ is an element of $(R/J_{f_S})_{kn-4}$ such that
$$\label{eq:condizione}
h(y,z,x_0,x_2)yzx_0x_2=h(\eta_n y,\eta_n z,x_0,x_2)\eta_n^2 yzx_0x_2$$
then the cohomology class $[\omega_h]\in H^3(U')$ is fixed by $T_{\phi}$. If we denote by $((R/J_{f_Y})_{tn-4})^{T_{\phi}}$ the elements of $(R/J_{f_Y})_{tn-4}$ satisfying condition (\[eq:condizione\]), from (\[eq:rec\]) we deduce
$$\label{eq:mappette}
(R/J_{f_Y})_{tn-4}^{T_{\phi}}\twoheadrightarrow Gr_P^{4-t} H^3(U')^{T_{\phi}}=P^{4-t}H^3(U')^{T_{\phi}}/P^{5-t}H^3(U')^{T_{\phi}}\hspace{0.5cm}\text{for }t=1,2,3.\\$$
Let us compute the dimensions of the $(R/J_{f_Y})_{tn-4}^{T_{\phi}}$. A monomial $y^az^bx_0^cx_2^d$ satisfies condition (\[eq:condizione\]) if and only if $a+b=kn-2$ for some $k\in{\mathbb{Z}}$ (\*); since $J_{f_Y}$ contains $y^{n-1}$ and $z^{n-1}$, a monomial $y^az^bx_0^cx_2^d\in(R/J_{f_Y})_{tn-4}$ can satisfy (\*) only for $k=1$; this implies in particular that $(R/J_{f_Y})_{n-4}^{T_{\phi}}=0$. From this we deduce that
$$Gr_P^3 H^3(U')^{T_{\phi}}=P^3 H^3(U')^{T_{\phi}}=0$$
which implies $Gr_P^2 H^3(U')^{T_{\phi}}=P^2 H^3(U')^{T_{\phi}}$; by (\[eq:ineqs\]) we obtain
$$\begin{aligned}
\dim V^{T_{\phi}}=h^{1,1}(V)^{T_{\phi}}\leq\dim Gr_P^2 H^3(U')^{T_{\phi}}.\end{aligned}$$
Since there are $n-1$ choices of non-negative $a,b<n-1$ that give $a+b=n-2$, we have $(n-1)^2$ monomials in $(R/J_{f_Y})_{2n-4}$ satisfying condition (\[eq:condizione\]); this gives
$$\label{eq:bound}
\dim V^{T_{\phi}}\leq (n-1)^2.$$
Now we compute the dimension of $H^1(F_{y^n+z^n})$ by studying the Steenbrink spectra of the homogeneous isolated singularities of ${\mathbb{C}}$ given by $y^n=0$ and $z^n=0$.
We use the same notations as in Section 3. For $y^n$ we have $d=n$, $w=1$ and $M(y^n)={\mathbb{C}}\oplus{\mathbb{C}}y\oplus\dots\oplus{\mathbb{C}}y^{n-2}$, so the non-zero parts of $M(y^n)$ have weights $0,\dots,n-2$ and dimension $1$. In order to have $({\alpha}+1)n-1=j$ for $j\in[0,n-2]$ we need ${\alpha}=\frac{j+1-n}{n}$, which implies $sp(y^n)=\sum_{j=0}^{n-2}(\frac{j+1-n}{n})$; this means the monodromy operator on $H^0(F_{y^n},{\mathbb{C}})$ has $n-1$ eigenspaces of dimension $1$ with eigenvalues $\zeta_n^a$ for $a\in[1,n-1]$ (and the same goes for $H(F_{z^n})$).
By Lemma \[lem:ThomIso\] we deduce that $H^1(F_{y^n+z^n})$ has dimension $(n-1)^2$ and it is the direct sum of monodromy eigenspaces with eigenvalues $\eta_n^{a+b}$ for $a,b\in[1,n-1]$. The equality $\eta_n^{a+b}=\eta_n^k$ is satisfied by $n-2$ choices of the couple $(a,b)$ for $k\neq 0$, while for $k=0$ the choices are $n-1$: this means that in $H^1(F_{y^n+z^n})$ the fixed part under the monodromy action has dimension $n-1$, while all the other $n-1$ eigenspaces have dimension $n-2$.
Lemma \[lem:ThomIso\] also allows us to write
$$H^3(F_{g-f})^{T_{g-f}}=\bigoplus_{0\leq{\alpha}<1} H^1(F_g)_{1-{\alpha}}\otimes H^1(F_f)_{{\alpha}}$$
where the subscript ${\alpha}$ indicates the eigenspace relative to $e^{2\pi i{\alpha}}$. If we denote by $\epsilon_i$ the dimension of $H^1(F_f)_{\zeta_n^i}$ then $\epsilon_0=n-1$, so we can write the dimension of the right-hand side as
$$(n-1)^2+\sum_{i=1}^{n-1}(n-2)\epsilon_i=(n-1)^2+(n-2)\sum_{i=1}^{n-1}\epsilon_i$$
From (\[eq:trick\]) and (\[eq:bound\]) we deduce that $H^3(F_{g-f})^{T_{g-f}}$ has dimension at most $(n-1)^2$, so $\epsilon_i=0$ for all $i\neq 0$, which means exactly that the Alexander polynomial of the arrangements we consider is trivial.
Using [@PS Lemma 3.1] one can easily check that ${\beta}_2({\overline{{\mathcal{A}}}})={\beta}_3({\overline{{\mathcal{A}}}})=0$ for ${\overline{{\mathcal{A}}}}$ as in Theorem \[thm:AP\].
[90]{} Dimca A. - Singularities and topology of hypersurfaces. Universitext, Springer-Verlag, New York, 1992. Dimca A. - On the Milnor monodromy of the irreducible complex reflection arrangements. arXiv:1606.04048. Dimca A., Papadima S. - Finite Galois covers, cohomology jump loci, formality properties, and multinets. Ann. Sc. Norm. Super. Pisa Cl. Sci. **10** (2011), no. 2, 253-268. Dimca A., Sticlaru G. - On the Milnor monodromy of the exceptional reflection arrangement of type $G_{31}$. arXiv:1606.06615. Deligne P. - Thèorie de Hodge II, III. Publ. Math. IHES **40** (1971), 5-58 and **44** (1974), 5-77. Falk M., Yuzvinsky S. - Multinets, resonance varieties, and pencils of plane curves. Compositio Math. **143** (2007), no. 4, 1069-1088. Guillén F., Navarro Aznar V., Pascual Gainza P., Puerta F. - Hyperrésolutions cubiques et descente cohomologique. Lecture Notes in Mathematics, vol. 1335, Springer-Verlag, Berlin, 1988, Papers from the Seminar on Hodge-Deligne Theory held in Barcelona, 1982. Hartshorne R. - On the de Rham cohomology of algebraic varieties. Publ. Math. I.H.E.S., **45** (1976), 5-99. Hartshorne R. - Residues and duality. Lecture Notes in Math. Springer, **20**, 1966. Lê Dung Trang - Some remarks on relative monodromy. Real and complex singularities (Oslo 1976), Sijthoff and Noordhoff, Amsterdam (1977), 397-403. Libgober A. - Alexander invariants of plane algebraic curves. Singularities, Part 2 (Arcata, Calif., 1981), 135-143, Proc. Sympos. Pure Math., **40**, Amer. Math. Soc., Providence, RI, 1983. Măcinic A., Papadima S., Popescu R. - Modular equalities for complex reflection arrangements. Doc. Math. **22** (2017), 135-150. Milnor J. - Singular points of complex hypersurfaces. Annals of Math. Studies, vol. 61, Princeton Univ. Press, Princeton, NJ, 1968. Papadima S., Suciu A.I. - The Milnor fibration of a hyperplane arrangement: from modular resonance to algebraic monodromy. Proc. of the London Math. Soc **114**, no. 6 (2017), 961-1004. Peters C. A. M., Steenbrink J. - Mixed Hodge Structures. Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics \[Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics\], vol. 52, Springer-Verlag, Berlin, 2008. Voisin C. - Hodge theory and complex algebraic geometry II. Cambridge University Press, **77** (2003).
Department of Mathematics “Tullio Levi-Civita”, University of Padua, Via Trieste 63, 35121 Padova, Italy.
*E-mail address*: `[email protected]`
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'G347.3–0.5 is one of three shell-type supernova remnants in the Galaxy whose X-ray spectrum is dominated by nonthermal emission. This puts G347.3–0.5 in the small, but growing class of SNRs for which the X-ray emission reveals directly the presence of extremely energetic electrons accelerated by the SNR shock. We have obtained new high-resolution X-ray and radio data on G347.3–0.5 using the [*Chandra X-ray Observatory*]{} and the Australia Telescope Compact Array (ATCA) respectively. The bright northwestern peak of the SNR seen in [*ROSAT*]{} and [*ASCA*]{} images is resolved with [ *Chandra*]{} into bright filaments and fainter diffuse emission. These features show good correspondence with the radio morphological structure, providing strong evidence that the same population of electrons is responsible for the synchrotron emission in both bands in this part of the remnant. Spectral index information from both observations is presented. We found significant difference in photon index value between bright and faint regions of the SNR shell. Spectral properties of these regions support the notion that efficient particle acceleration is occurring in the bright SNR filaments. We report the detection of linear radio polarization towards the SNR, which is most ordered at the northwestern shell where particle acceleration is presumably occurring. Using our new [*Chandra*]{} and ATCA data we model the broad-band emission from G347.3–0.5 with the synchrotron and inverse Compton mechanisms and discuss the conditions under which this is a plausible scenario.'
author:
- |
J. S. Lazendic, P. O. Slane, B. M. Gaensler, S. P. Reynolds,\
P. P. Plucinsky and J. P. Hughes
title: 'A high-resolution study of nonthermal radio and X-ray emission from SNR G347.3–0.5'
---
INTRODUCTION
============
Due to the release of an enormous amount of energy ($\sim 10^{51}$ erg) at their creation, supernova remnants (SNRs) have long been considered as a primary source of Galactic cosmic rays with energies up to the “knee” of the spectrum at $\sim 3 \times 10^{15}$ eV [@shklovsky53; @ginzburg57]. Cosmic rays with energies higher than this are believed to be extragalactic in origin [@axford94]. First-order Fermi shock acceleration, also called diffusive shock acceleration, in which particles gain energy from scattering back and forth across the shock, has been suggested as the most probable acceleration mechanism in SNR shocks [see @reynolds81; @blanford87; @jones91]. However, until recently the observational evidence for the production of high energy particles in SNRs was poor and came mainly from the fact that SNRs emit synchrotron radiation in the radio band. The major observational break-through came only recently with the detection of nonthermal X-ray emission from the shell-type SNR SN 1006 [@koyama95]. The featureless X-ray spectrum found at the rim of SN 1006, in contrast to the thermal spectra found towards the interior of the remnant, was fitted well with a power law of photon index $\Gamma \sim 2.2$ (where the photon flux $F$ obeys $F \propto (h\nu)^{-\Gamma}$). Further evidence came with the detection of TeV $\gamma$-ray emission from SN 1006 [@tanimori98]. There are several mechanisms capable of producing TeV energy photons, including inverse Compton (IC) scattering, nonthermal bremsstrahlung and pion-decay. Broad-band modeling of the emission from SN 1006 indicates that IC scattering is responsible for the TeV $\gamma$-ray emission [e.g., @mastichiadis96; @allen01]. However, cosmic rays are comprised mostly of protons and there is not yet clear evidence of proton acceleration in SNRs, aside from the suggestion that the TeV emission results from neutral pion-decay [@aharonian99; @enomoto02]. Another problem in the quest for a cosmic ray origin in SNRs is that the maximum energy of electrons produced by SNRs seems to fall below the “knee”. For example, studies of about 20 SNRs with mostly thermal X-ray emission imply that the maximum energy to which electrons can be accelerated does not exceed [$10^{14}$]{} eV [@reynolds99; @hendrick01].
Two more shell-type remnants with dominant non-thermal X-ray spectra have been identified: G347.3–0.5 [@koyama97; @slane99] and G266.2–1.2 [@slane01]. G347.3–0.5 (RX J1713.7–3946) was first discovered in the [*ROSAT (Röntgensatellit)*]{} All-Sky Survey by @pfeffermann96, who used a thermal plasma model to infer a very high temperature of $kT \sim
4.8$ keV, and column density of $N_H \sim 4.5 \times 10^{21}$ [cm$^{-2}$]{}. Based on this column density the distance to the SNR was estimated to be $\sim 1.1$ kpc, while the derived plasma temperature implied an SNR age of $\sim 2100$ yrs. However, subsequent [*ASCA (Advanced Satellite for Cosmology and Astrophysics)*]{} observations revealed that the X-ray emission from the remnant is predominantly nonthermal [@koyama97; @slane99]. The remnant is $\sim 1$ in diameter and appears to be of a shell–type morphology with the brightest emission in the western region. The SNR is located on the edge of the molecular cloud complex that encompasses the [H<span style="font-variant:small-caps;">ii</span>]{} region G347.61+0.20 located northwest of the SNR (see Figure \[fig-atca\]). Assuming that the SNR is physically associated with the molecular cloud complex, the distance was estimated to be $6.3 \pm 0.4$ kpc from existing observations of the CO 1–0 line emission towards the complex [@slane99]. [*ASCA*]{} observations did not reveal any line emission from the SNR interior, and this lack of thermal emission sets an upper limit on the mean density around the remnant of $< 0.3$ [cm$^{-3}$]{} [@slane99]. Such a low density suggests that the majority of the remnant is still evolving in the interior of the large circumstellar cavity driven by the wind of its massive progenitor. In the northwestern region of G347.3–0.5, where the SNR may be interacting with denser molecular gas, the upper limit on the ambient density is higher ($< 1$ [cm$^{-3}$]{}), which is broadly consistent with the densities estimated around other SNRs that are associated with molecular clouds. Most recently, @pannuti03 reported detection of a thermal component towards the center of the SNR, which implies gas density of 0.05–0.07[cm$^{-3}$]{} in this part of the SNR.
Two point sources have been identified within the boundaries of the remnant shell: 1WGA J1714.4–3945 is believed to be of stellar origin [@pfeffermann96], while 1WGA J1713.4–3949 is located at the center of the SNR, has no obvious optical counterpart, and is a candidate for an associated neutron star [@slane99]. Our [*Chandra*]{} observations also included the latter point source, results for which are presented elsewhere [@lazendic03].
G347.3–0.5 was also detected at TeV energies with the CANGAROO[^1] telescope [@muraishi00; @enomoto02], and models of broad-band emission point to IC scattering as the origin of the TeV photons [@muraishi00; @ellison01]. More recently, follow up TeV $\gamma$-ray observations have led to a different conclusion, suggesting pion-decay as the source of energetic photons [@enomoto02], but the nature of this emission is still uncertain [see @butt02; @reimer02]. An EGRET[^2] source 3EG J1714–3857 [@hartman99] was also detected near the remnant and linked to the SNR interaction with a molecular cloud [@butt01]. This is potentially supported by the identification of the hard X-ray source AX J1714.1–3912, which appears to coincide with the location of one of the molecular clouds in the complex, and possibly with the EGRET source [@uchiyama02]. However, the energetics required to yield the observed X-ray flux of the source appear problematic relative to the total energy budget of the SNR unless the source distance is a factor of five smaller than suggested by the molecular line velocity of the cloud.
To investigate the fine-scale structure of the SNR and its relationship with particle acceleration, we obtained high-resolution X-ray and radio data on G347.3–0.5 using the [*Chandra X-ray Observatory*]{} and the Australia Telescope Compact Array (ATCA). Our goal was to look for spectral variations in different spatial regions, to improve the estimates on thermal emission, to compare the X-ray morphology with high-resolution radio images, to correlate the radio and X-ray spectral indices and to investigate linear polararization. We present these results here, along with modeling of the broad-band spectrum of G347.3–0.5 to investigate the origin of the accelerated high energy particles in this SNR.
In section $\S$ 2 we describe our X-ray observations, and present images and spectral results for the SNR. Radio observations and images of the SNR are presented in section $\S$3, where we also derive a radio spectral index and measure linear polarization in the SNR. In section $\S$4 we discuss the spectral variations across the nortwestern SNR shell and in section $\S$5 we investigate the SNR morphology and the correlation between X-ray and radio images. In section $\S 6$ we present synchrotron and inverse Compton modeling of the broad-band spectrum from G347.3–0.5 and discuss the origin of the TeV emission in this SNR.
X-RAY DATA
==========
G347.3–0.5 was observed with the Advanced CCD Imaging Spectrometer (ACIS) detector on board the [*Chandra X-ray Observatory*]{} on 2000 July 25. ACIS consists of two CCD arrays — the ACIS-I (Imaging) array has four CCDs arranged in a square, while the ACIS-S (Spectroscopic) array has six linearly adjacent CCDs. Only six CCDs can be used at one time. The angular size of G347.3–0.5 is much larger than the ACIS-I field of view ($16\farcm 9 \times 16\farcm
9$), so we observed two diametrically opposed fields covering regions of the SNR shell, as shown in Figure \[fig-chandra+rosat\]. Field 1 (ObsId 736) was centered on the bright northwestern rim at [${\rm RA}={17}^{\rm h}{11}^{\rm m}{49}\fs {8}$]{}, [${\rm Dec.}={-39}^{\circ}{36}\arcmin{18}\farcs
{0}$]{} and consisted of a 30 ks exposure. Field 2 (ObsId 737) was observed for 40 ks, positioned at the fainter eastern SNR rim centered at [${\rm RA}={17}^{\rm h}{15}^{\rm m}{16}\fs {5}$]{}, [${\rm Dec.}={-39}^{\circ}{57}\arcmin{56}\farcs
{0}$]{}. Two ACIS-S chips were also turned on: for field 1, the ACIS-S1 and ACIS-S3 chips were used to include the central point source and a region in the SNR interior, while for field 2 ACIS-S2 and ACIS-S3 were used to cover the eastern SNR rim. Data were taken in full-frame timed–exposure (TE) mode with the standard integration time of 3.2 s.
Data were reduced using standard threads in the [*Chandra*]{} [Interactive Analysis of Observations (CIAO)]{} software package version 2.2.1. To mitigate the degradation in the spectral response of the ACIS chips caused by radiation damage early in the mission [see @prigozhin00], we also used software [@townsley00; @townsley02] to correct for the effect of increased charge transfer inefficiency (CTI), such as gain variations, event grade distortion and degraded energy resolution as a function of row number in each CCD. These corrections increase the number of detected events and improve the assigned event energies and spectral resolution. We applied CTI corrections to the Level 1 processed event list provided by the pipeline processing done at the [*Chandra*]{} X-ray Center (CXC). The data were then screened for “flaring” pixels and filtered with standard [*ASCA*]{} grades (02346). The effective exposure time after data processing was 29.6 ks and 38.9 ks for field 1 and 2, respectively.
X-ray Image
-----------
We used the [CIAO]{} script [merge\_all]{} to combine the data from the two fields into an exposure-corrected X-ray image in the energy range 1–8 keV. Since the effective area of the detector depends both on the energy of incident photons and their positions on the CCD, we applied spectral weighting to the merged image, which takes into account the fraction of the incident flux falling in the particular part of the band. The exposure-corrected image was blanked where the exposure was less than 15% of its maximum value. The final image is shown in Figure \[fig-chandra+rosat\] overlaid with the [*ROSAT*]{} contours. The image was binned in $4\farcs 6 \times 4\farcs 6$ pixels and smoothed with a Gaussian filter with a FWHM of 2. Since CTI corrections were not available for ACIS-S1 chip on which the central source is located, these data were processed separately and then added to the image. We also made soft (0.5–2.1 keV) and hard band (2.1–8.0 keV) images, but they do not show any significant brightness variation with energy.
The northwestern peak seen in [*ROSAT*]{} and [*ASCA*]{} images is resolved with [*Chandra*]{} observations into two bright arcs and fainter diffuse emission, as shown in Figure \[fig-chandra+rosat\]. As we discuss below, this structure bears significant resemblances to what is seen in radio images. Arc 1 (see Figure \[fig-chandra+rosat\] and Figure \[fig-regions\]) appears to be made of thin filaments delineating the SNR shock front, while Arc 2 has a more irregular shape. Diffuse emission along the eastern shell does not show significantly different structure from that seen with [*ROSAT*]{}, although the limb is more clearly defined and extends beyond contours obtained from the [*ROSAT*]{} image.
X-ray Spectra {#sec-xray}
-------------
X-ray spectra were extracted from discrete regions of the ACIS chips in field 1 and field 2, shown in Figure \[fig-regions\]. For each spectrum the counts were grouped with a minimum of 25 counts per spectral channel. We use redistribution matrix files (RMF) appropriate for CTI-corrected data [@townsley00; @townsley02], and created weighted auxiliary response files (ARF) applying the same spectral binning as used in the corresponding RMFs. The background for the spectral fitting was extracted from the northwest corner in field 1 and from the southeast corner in field 2 (see Figure \[fig-regions\]).
Shock acceleration of electrons to X-ray emitting energies is expected to produce a roughly power law distribution with an exponential cutoff above some maximum energy, resulting in a photon spectrum rolling slowly off through the X-ray band. The [*Chandra*]{} bandpass is not large enough in most cases to distinguish this slight curvature from a strait power law. The X-ray slope bears no particular relation to the slope at lower frequencies, but can serve as an indication of the location of the rolloff. We shall use both power law fits (giving photon index $\Gamma$) and cutoff-model fits ([SRCUT]{}, giving a rolloff frequency) to describe our spectra.
Individual spectra were first fitted separately to search for any change in the photon index across the source and for the presence of thermal component. We found small-scale spectral differences in different regions. We could fit those either with variations in column density, and roughly the same photon index ($\sim 2.1$), or, if the column density were fixed, with variations in the photon index. In the first case, we found that the column density was higher towards the bright regions ($\sim 8\times10^{21}$[cm$^{-2}$]{}) than towards the fainter regions ($\sim 5\times10^{21}$[cm$^{-2}$]{}). Thus, the brighter regions have higher column density, which is opposite from what one expects if the brightness variation is caused by varying interstellar absorption. Furthermore, the fact that the radio and X-ray emission have the similar morphologies in field 1 implies that X-ray brightness distribution is not the result of variations in absorption. Thus, it seems plausible that the column density is constant across field 1 and we adopt the second case, fixing the column density at the value found in the spatially averaged fitting and allowing $\Gamma$ to vary. Photon index values derived with the second approach are plotted in Figure \[fig-fg\] and imply that there is significant spectral variation between the bright and the faint SNR regions. The bright emission has values $2.05 < \Gamma <
2.25$, while the faint emission takes broad range of values, with the steepest photon index of $\sim 3$. Representative spectra with power law model fit are shown in Figure \[fig-spectra\]. The average values of photon index across field 1 of $\sim 2.35$ is consistent with that of $2.41$ for the northwestern region from the previous study with [*ASCA*]{} [@slane99].
We then fitted our data with the [SRCUT]{} model [@reynolds99], which calculates a synchrotron spectrum from a power law distribution of electrons modified by an exponentially cut off in a uniform magnetic field. The model can be used to derive the maximum rolloff frequency $\nu_{rolloff}$ related to the maximum electron energy. Derived $\nu_{rolloff}$ values are plotted in Figure \[fig-fg\]. We frozen the radio spectral index $\alpha$ (where the radio flux $S_{\nu} \propto \nu^{-\alpha}$) parameter for each spectrum to the global SNR value of 0.6 and left normalization (i.e., the 1 GHz radio flux density) as a free parameter. The model gave reduced $\chi^2$ value comparable to that of the power law model and the range of values for rolloff frequency between 3[$\times 10^{16}$]{}–2[$\times 10^{18}$]{} Hz, with the fainter regions having, in generall, lower $\nu_{rolloff}$ values.
The SNR spectra from regions we observed with [*Chandra*]{} show no evidence for any emission lines. Adding equilibrium or non-equilibrium thermal models to the power law fits of the individual or joint fit to the spectra does not improve the fit significantly. However, spectra from some regions in both ACIS-S3 chips show a possible line feature around 0.8 keV. Modeling this feature does not provide strong constraints on the thermal emission properties, so deeper observations are needed to follow up this potential detection of thermal emission in G347.3–0.5.
RADIO DATA
==========
Radio observations of G347.3–0.5 were obtained with the ATCA during January, March and April 1998. The array consists of six 22 m antennas that can be configured to give baselines between 31 m and 6 km [see @frater92]. We used the array in three different configurations (375, 750 A and 1.5 A) to provide optimal $uv$-coverage of the observed region. Due to the large extent of the SNR and the complexity of the field around it, the region was imaged in mosaic mode with 10 pointing centers. Data were taken simultaneously at two frequencies, 1.4 and 2.5 GHz, each using a bandwidth of 128 MHz split in 32 4-MHz channels. For all the observations the primary calibrator PKS B1934–638 was used for bandpass and absolute flux calibration. PKS 1740–517 was used as a secondary calibrator for antenna gains and instrumental polarization calibration.
Radio Images
------------
Radio data were reduced using standard procedures of the [MIRIAD]{} software package [@sault99]. In the editing procedure, 4 channels on each side of the band were discarded and the remaining 26 central channels were averaged down to 13 8–MHz channels by Hanning smoothing. A uniform weighting was applied to visibilities to minimize sidelobes in the images. The longest baselines (i.e., all correlations with the sixth antenna) were excluded to enhance the surface brightness sensitivity. Also, amplitude and phase self-calibration were applied to the 1.4 GHz data to improve solutions for antenna gains using a strong point source in the eastern edge of the observed field (outside the image shown here). The images were made using the multi-frequency synthesis [@sault94] and deconvolved using the mosaiced maximum entropy approach [@sault96]. The resulting images were convolved with a Gaussian restoring beam listed in Table \[tab-atca\], and corrected for primary beam response. The rms sensitivity in the images is also listed in Table \[tab-atca\].
In Figure \[fig-atca\] we show the image at 1.4 GHz overlaid with the [*ROSAT*]{} contours. As noted by @slane99, in the radio band the SNR appears as a faint shell with two bright arcs to the west (marked in Figure \[fig-atca\]). A faint inner ring of emission is also evident in the radio image of the SNR (Figure \[fig-atca\]), with a diameter of $\sim 30\arcmin$. The image at 1.4 GHz has angular resolution comparable to the Molonglo Observatory Synthesis Telescope (MOST) image [@slane99] and was published in preliminary form by @ellison01. While the image at 2.5 GHz has improved angular resolution, it shows no significant morphological difference with respect to the other two images, and thus we do not show it here.
We find no obvious radio source at the location of the [*ASCA*]{} source AX J1714.1–3912, believed to be associated with the EGRET source 3EG J1714–3857 in the vicinity of the SNR [@uchiyama02]. There are, however, a few compact radio sources around that location which are probably thermal [H<span style="font-variant:small-caps;">ii</span>]{} regions.
Radio Spectral Index
--------------------
To determine the spectral index of the bright western SNR rim, we used the “spectral tomography” approach [@katz97]. Since source visibilities at different frequencies are sampled with different $u-v$ spacings, we first had to resample the 1.4 GHz data to match the $uv$-coverage of the 2.5 GHz data [see e.g., @crawford01]. We first removed the primary beam attenuation from the 1.4 GHz image, to which we then applied the primary beam attenuation and mosaic pattern of the 2.5 GHz data. The resulting image was Fourier transformed and resampled using the transfer function of the 2.5 GHz data. We then imaged and deconvolved these modified 1.4 GHz data using the same procedures as for the 2.5 GHz data. The modified 1.4 GHz and original 2.5 GHz image were then both smoothed to a resolution of 60. The 2.5 GHz image was scaled by a trial spectral index, $\alpha_t$, and subtracted from the 1.4 GHz image, $I_{\alpha_t}=I_{1.4}-(1.4/2.5)^{\alpha_t}
I_{2.5}$, where $I_{\alpha_t}$ is the difference image, and $I_{1.4}$ and $I_{2.5}$ are the images at 1.4 and 2.5 GHz, respectively. The spectral index $\alpha$ is then found as the value of $\alpha_t$ at which a particular feature in the image blends into the background, while the range of $\alpha_t$ values for which the residuals of the blended feature are significant gives the uncertainty in the spectral index.
We find a spectral index of $0.50 \pm 0.40$ for the bright northwestern region of the SNR. This large uncertainty in the spectral index determination is caused by calibration errors and dynamic range limitations in bright, complex regions. The spectral index of Arc 1 is $0.45 \pm 0.30$ and of Arc 2 is $0.10 \pm 0.40$. The radio emission from other regions in the SNR is generally too faint for spectral index determination. The spectral index of the large [H<span style="font-variant:small-caps;">ii</span>]{} region north-west from the SNR is found to be $-0.10 \pm 0.05$.
Radio Polarization
------------------
Continuum observations with the ATCA provide simultaneos recording of all four Stokes parameters. We therefore made images of the $Q$, $U$ and $V$ Stokes parameters at 1.4 and 2.5 GHz for each of the 13 data channels to minimize bandwidth depolarization. To produce an image of polarized intensity we used the [MIRIAD]{} task [pmosmem]{} which performs a joint maximum entropy deconvolution of the total and polarized intensities for mosaic observations [@sault99-2]. Cleaned $Q$ and $U$ images were then restored with a Gaussian beam, listed in Table \[tab-atca\], and corrected for primary beam response. A linear polarization image, $P_I=(Q^2+U^2)^{1/2}$, was formed for each pair of the Stokes $Q$ and $U$ images and corrected for non-Gaussian noise statistics [@killeen86]. The mean of the 13 $P_I$ planes was formed and then blanked where total or polarized emission fell below the 5$\sigma$ level.
We found significant linear polarization towards the western shell of G347.3–0.5. In Figure \[fig-pol\] we show the fractional polarization towards the SNR at 1.4 GHz. The mean fractional polarization is $\sim 5$% at 1.4 GHz and $\sim 10\%$ at 2.4 GHz, calculated by dividing the sum of polarized intensity by the sum of total intensity. The polarized intensity is strongest towards Arc 1, up to 12% at 1.4 GHz and 30% at 2.5 GHz, where it clearly follows the arc-like distribution of the total intensity. There is weaker ($< 6$%), diffuse polarized emission from Arc 2. The polarized emission towards the rest of the western SNR region is patchy and diffuse, with no particular correlation with the radio continuum, implying that the magnetic field is not very highly ordered there and that most of the remnant has a high degree of depolarization. There is almost no polarization detected towards the rest of the SNR, with exception of a few small regions, e.g., towards the center region and the eastern SNR shell.
The multi-channel continuum capability of the ATCA can be used to derive the rotation measure (RM) across the observing bandwidth, caused by the Faraday rotation. Using the frequencies within 1.4 and 2.4 GHz bandwidths, we get RM values towards Arc 1 of $> +100$ rad m$^{-2}$. Our observed uncertainty in RM, $\Delta
{\rm RM}\sim 20$ rad m$^{-2}$, implies an uncertainty in the intrinsic polarization position angle $\Delta \phi_0 \sim \Delta {\rm RM} c^2
/\nu^2 \sim 1$ rad, when extrapolated from the RM measured within each band. Furthermore, we are unable to calculate the RM between the 1.4 and 2.4 GHz bands, because these observations have different wavelengths and angular resolution, and are thus subject to very different depolarization and Faraday rotation effects by foreground ionized gas [e.g., @gaensler01]. Thus we are unable to determine the intrinsic orientation of polarization from the available data.
Spectral Variations Across G347.3–0.5
=====================================
Recently, @uchiyama03 presented a study of [*Chandra*]{} observations towards the northwestern rim of G347.3–0.5. In their spectral analysis of individual spectra from field 1 they infer that the photon index is same in diffuse and bright SNR regions. However, for their spectral fitting they used a blank sky observations for the background subtraction and varying column density for diffuse and bright regions. We also tried using blank sky observations for the background subtraction and found no difference in spectral fit results from the fitting using the background region from field 1. We also found that leaving column density as a free parameter for each individual spectrum results in very close photon index values. As explained in an earlier section ($\S$ \[sec-xray\]), the possible variation in column density across the northwestern shell of the SNR seems unlikely; our use of a uniform value for $N_{\rm H}$ accounts for the difference in spectral indices between our work and that of @uchiyama03.
We produced two summed spectra, one for the bright and one for the faint SNR regions in field 1, which we fitted with power law and [SRCUT]{} models. The results from the fit are summarized in Table \[tab-joint-fit\]. The spectrum from bright regions has $\Gamma \sim 2.10$, while the spectrum from the faint regions has $\Gamma \sim 2.53$. When fitting the [SRCUT]{} model we froze again the radio spectral index to the global remnant value of 0.6 and left the normalization as a free parameters. The resuting rolloff frequency for the bright regions was $\nu_{rolloff} \sim 6.3\times 10^{17}$ Hz and for the faint regions was $\nu_{rolloff} \sim 1.2\times 10^{17}$ Hz. The radio flux at 1 GHz estimated by [SRCUT]{} of $S_{\nu} \sim 4.2$ Jy for the bright regions and $S_{\nu} \sim 2.9$ Jy for the faint regions is broadly consistent with that measured from our radio data, which gives us confidence in this approach.
These results, as well as those from fitting the individual spectra, imply that the fainter SNR regions have a steeper photon index and lower rolloff frequency than the bright regions. Thus, our data suggest that acceleration of particles to higher energies is more efficient in the brighter regions, perhaps because of the higher magnetic field in those regions. Another possibility is that electrons in the fainter regions are older, i.e., they could be produced in bright regions and then lose energy diffusing away from their origin.
High-resolution Comparison of X-ray and Radio Data
==================================================
Using our high-resolution radio and X-ray images we are able for the first time to address the correlation between the radio and X-ray morphology in a great detail. The [*Chandra*]{} and ATCA images of field 1 are overlaid in Figure \[fig-chandra+atca\] (left panel); we do not compare closely emission in field 2 because the radio emission is extremely weak there and shows no obvious structure. The [*Chandra*]{} image of the field 1 is shown smoothed to resolution of 2 (right panel), as well as convolved to the resolution of 1.4 GHz ATCA image of 47 (contours in the left panel). If the same population of electrons is responsible for the synchrotron emission in both the radio and X-ray band, one would expect similar morphology in these two bands, as appears to be the case for SN 1006 [@allen01; @dyer01]. The correlation between the radio and X-ray images of G347.3–0.5 is not perfect, as noted from the [*ASCA*]{} data [@slane99]. Most notable is the lack of X-ray emission along the well-defined inner radio ring and the lack of radio emission from the southwestern peak present in the X-ray image — these regions were not imaged by us with [ *Chandra*]{} so we compare ATCA and [*ROSAT*]{} images in Figure \[fig-atca\]. However, the broad morphological agreement between the nonthermal X-ray and radio emission from field 1, and Arc 1 in particular (Figure \[fig-chandra+atca\]), provides strong evidence that efficient particle acceleration is occurring in the northwestern part of the SNR. Arc 2 shows mostly diffuse emission, but a bright filament is present in the radio image which does not have a counterpart in the X-ray image. The nonthermal radio nature of Arc 2 was questioned by @ellison01 due to its proximity to a region of thermal emission from the adjacent [H<span style="font-variant:small-caps;">ii</span>]{} region. Indeed, we find that this region has a large uncertainty in the radio spectral index, which implies confusion by the surrounding material. The X-ray spectrum of Arc 2 is clearly nonthermal; we found no indications of thermal X-ray emission towards this region. Furthermore, the polarized intensity in this region, shown in Figure \[fig-pi\], shows emission from Arc 2 which morphologically resembles X-ray data rather than the radio data. That implies that Arc 2’s radio emission is at least partially nonthermal.
The expected relation of radio and synchrotron X-ray morphology in an evolved, complex object like G347.3–0.5 is not simple. While we expect radio and X-ray emitting electrons to be born in the same shock waves, their subsequent evolution can vary greatly, because of different diffusive properties and the factor of 30,000 greater radiative lifetime for a 1 GHz-radiating electron compared to a 1 keV-radiating one. In fact, an electron radiating its peak synchrotron power in 1 keV photons has a half-life against synchrotron losses of $$t_{1/2} = 5.0 \times 10^4 B_{\mu{\rm G}}^{-3/2}
\left( {h \nu \over {1 \ {\rm keV}}} \right)^{-1/2}
\ {\rm yr}$$ where we have averaged over electron pitch angles. (Losses to inverse-Compton scattering from the cosmic microwave background (CMB) dominate for $B < 3.2 \ \mu$G, the field strength with the energy density of the CMB, so the maximum lifetime is about 8600 years for 1-keV-emitting electrons.) In a simple single shock wave, we might then expect radio and X-ray synchrotron emission to appear at the same location, with X-ray emission disappearing a shorter distance behind the shock. Such a morphology is evident in SN 1006 [@long03] and in parts of RCW 86 [@rho02]. If the magnetic field is quite inhomogeneous, as we propose below, the lifetime of an individual electron will also depend strongly on the mean magnetic field along its particular diffusive trajectory. Furthermore, weak shocks may accelerate electrons to radio-emitting but not X-ray-emitting energies, so we should not be surprised to find radio emission without X-ray counterparts. Finally, X-rays can be absorbed by intervening gas at much lower column densities than would be required for free-free absorption of radio emission. In general, while we would hope to find at least some shock-like (thin linear) structures in X-rays and radio, there may not be an extremely close correspondence overall between the two bands.
Figure \[fig-profile\] shows profiles through the most prominent linear feature seen in both radio and X-ray, a thin filament at the inner edge of Arc 1. It is apparent that the turn-on of radio and X-rays is nearly coincident (we recall that the resolution of the radio image is 47), while the radio emission extends considerably further than the X-rays. The width of the X-ray filament is about 40 (1.2 pc at a distance of 6 kpc). If the leading edge of the structure is in fact a shock seen close to edge on, the X-ray width could result from synchrotron losses in a $3 \ \mu$G field if the relativistic electrons are convected downstream (i.e., in the plane of the sky) at the modest velocity of about 140 kms$^{-1}$. As we argue below, the broadband spectrum of G347.3–0.5 can be explained if the magnetic-field structure consists of small regions with $B_1 \sim 15 \ \mu$G occupying about 1% of the volume, with a much lower field ($B_2 < 1 \mu$G) occupying the rest. The mean field sampled by an electron would be fairly close to $B_2$, in this case, unless electrons were somehow trapped in the high-field regions. We therefore expect lifetimes of 1-keV-emitting electrons to be near the CMB limit of about 9000 yr, though in particular regions they could be much shorter. (An electron living all its life in $B_1$ would last only about 140 yr).
Origin of TeV Emission from G347.3–0.5
======================================
An earlier broad-band emission model for G347.3–0.5 by @ellison01 implied that the IC mechanism was responsible for producing TeV emission from the remnant. The derived maximum electron energy of $E_{max} \sim 14 \times 10^{12}$ eV falls below the “knee” energy, but they showed that in addition to efficient acceleration of electrons there can be efficient acceleration of ions (Fe$^{+26}$ in particular), whose maximum energy could reach the desired energy of $\sim 10^{15}$ eV. However, for this study only a single TeV flux measurement from the 3.8-m CANGAROO telescope was available. Most recently, @enomoto02 published a $\gamma$-ray spectrum for G347.3–0.5 obtained with the new 10-m telescope, and suggested that the IC mechanism cannot produce the observed shape of the TeV spectrum. They modeled the TeV spectrum assuming pion-decay, which implies a higher particle density for the ambient medium than allowed by the X-ray emitting gas inside the SNR [@slane99]. If TeV emission is produced in regions outside the boundary of the SNR shock, where particles are upscattered through the diffusive acceleration process, the density values derived by @enomoto02 from the pion-decay model might be plausible since the molecular cloud is so close to the remnant. In this case the bulk of TeV emission would have to originate in the cloud, somewhere beyond the SNR shell. Their pion-decay spectrum is inconsistent with EGRET observations [@butt01; @reimer02], unless the proton spectrum has a cutoff below some energy, as also noted by @uchiyama03. This would result naturally if the protons must diffuse ahead of the shock to the molecular cloud, and if the diffusion coefficient increases with proton energy. In such a scenario only the highest-energy protons would reach the target, so that the pion-decay model would require an even higher target (molecular cloud) density, and demand even greater production efficiency of cosmic-ray protons at the shock. @uchiyama03 applied a two-zone model for the acceleration and diffusion of particles in an effort to explain the irregular morphology along the SNR limb. They considered the bright SNR filaments as locations where particles are accelerated, and the fainter regions where the electrons diffuse to after being accelerated. For a distance of 6 kpc, they derive a field strength of 50$\mu$G, but require that this field be similar in the filamentary and diffuse emission regions. This then implies that the spectral cutoff in the diffuse emission regions is due to radiative losses while that in the filmants is due to diffusive escape of the particles. The resulting IC emission falls below the observed CANGAROO flux, thus requiering some other mechanism to produce the TeV emission. The IC mechanism was also rejected in the most recent study of [*ASCA*]{} data by @pannuti03, because it required unrealistically low magnetic field filling factor (see next section). We wished to see, however, if our new [*Chandra*]{} data really supported ruling out the IC mechanism for the TeV emission.
The Broad-band Model
--------------------
In Figure \[fig-model\] we show measurements used to model the broad-band emission from G347.3–0.5. While the CANGAROO observations have poor angular resolution [for G347.3–0.5 observations estimated to be 023; @enomoto02], we shall assume that the TeV emission originates from particles in a broad region in the northwest of the SNR [see @muraishi00], which includes the synchrotron-emitting regions. The similar radio and X-ray morphology towards the northwest suggests a common origin for the synchrotron-emitting particles at least in this part of G347.3-0.5. To match the spatial resolution of the CANGAROO data we produced a summed [*Chandra*]{} spectrum for the region covered by ACIS-I CCDs in field 1. The resulting unabsorbed X-ray flux in the 0.5–10.0 keV band is $\sim 1.3$[$\times 10^{-10}$]{}[ ergcm$^{-2}$s$^{-1}$]{}. To measure radio flux densities, we convolved the radio images to 60, which are then corrected for variations in the background level determined from the average of several areas around the region of interest. We obtained an integrated flux density from the region corresponding to field 1 of $6.7 \pm 2.0$ Jy at 1.4 GHz, and $5.6 \pm 1.7$ Jy at 2.4 GHz. The EGRET measurements are taken from @reimer02, which used flux values from @hartman99. The EGRET values correspond to the nearby source 3EG J1714–3857, which is either associated with G347.3–0.5 or else provides an upper limit to any such emission from the SNR. The CANGAROO measurements are taken from @enomoto02.
We model the synchrotron and IC emission from G347.3–0.5 using a power law energy distribution modified by an exponential cutoff [e.g., @gaisser98]: $$\frac{dN}{dE} = A_e~E^{- \sigma} exp \left [-
\left ( \frac{E}{E_{max}} \right ) ^{\alpha_{path}} \right ] ,$$ where $A_e$ is the normalization factor, $\sigma$ is the index of the electron distribution and $E_{max}$ is the maximum energy of accelerated particles. We included additional parameter, $\alpha_{path}$, in the exponential function. This is a phenomenological parameter that allows for a broadening of cutoff, which accounts for variations in magnetic field from place to place, or in $E_{max}$, or both. A similar model was also used by @uchiyama03, or in a different form by @ellison01. We use a steady-state model which assumes that the synchrotron and IC emission comes from a single population of relativistic electrons. Details of the model are given in Appendix A. The model yields estimates of the photon spectra produced by synchrotron and IC emission mechanisms in the $\nu F_{\nu}$ format: $$\nu F_{\nu} (synch) = \frac{V_{E}}{4\pi D^{2}_{pc}}~f_B~
\nu~P_{tot}(\nu, E),
\label{eq-synch}$$ $$\nu F_{\nu} (IC) = \frac{V_{E}}{4\pi D^{2}_{pc}}~
E^2~\frac{dn_{\gamma}(\varepsilon_{\gamma})}{dt},$$ where $D_{pc}$ is the distance to the SNR in pc, $V_{E}$ is the electron emission volume (assuming a thin spherical shell geometry) given as [@ellison01]: $$V_{E} = f_{E}~\frac{4 \pi}{3} (R_{O}^{3} - R_{I}^3 ).$$ $R_{O}$ and $R_{I}$ are the outer and inner shock radius respectively, and $f_{E}$ represents the fraction of the shell volume producing the emission seen from the northwestern SNR limb. The magnetic field filling factor $f_B$ in equation (\[eq-synch\]) corresponds to the fraction of the volume of IC-emitting electrons containing the magnetic field responsible for the synchrotron emission [e.g., @allen01]. $P_{tot}(\nu, E)$ and $\frac{dn_{\gamma}(\varepsilon_{\gamma})}{dt}$ are synchrotron and IC emissivities and are given in Appendix A.
Since the magnetic field strength $B$ and maximum particle energy $E_{max}$ cannot be determined independently from the synchrotron spectrum, we first modeled the CANGAROO data with an IC spectrum (treating the EGRET data as an upper limit for the flux detectable from the SNR) which depends only on $E_{max}$ (see Appendix A). In this way we determined the parameters for the electron number distribution: maximum electron energy $E_{max}= 5$ TeV and particle index $\sigma = 2.0$. We then varied the magnetic field strength, which is fixed by the ratio of peak emission frequencies $R_{\nu}$. An estimate good to about 25% is given by (see Appendix B) $$B=9 \times 10^4 C_{\rm path}^{-1} R_{\nu}^{-1} {\rm G}.$$ The parameter $C_{\rm path}$ depends on the radio spectral index and on the parameter $\alpha_{\rm path}$ describing the breadth of the cutoff in the electron spectrum; by experimentation we found that for $\sigma = 2,$ a value of $\alpha_{\rm path} = 0.5$ described the X-ray data well. For these values, $C_{\rm path} = 16$ which, with an estimate for $R_\nu \sim 10^{8.5}$, gives $B \sim 20 \ \mu$G. We used our full model calculation for a more precise assessment and found that in order to match the shape of the X-ray spectrum we require $B \sim 15\,\mu$G. However, this would overpredict the synchrotron portion of the spectrum by a factor of 100. It is this problem that caused @enomoto02 to rule out a synchrotron/IC model. But we can rescue such models by assuming that the magnetic field fills only a portion of the volume occupied by the IC-emitting electrons, and take a value of $\sim 0.01$ for the magnetic field filling factor $f_B$. This filling factor is given by (see Appendix B): $$f_B=C(\sigma) R_{S} B^{(\sigma+1)/2}$$ where $R_S$ is the ratio between synchrotron and IC flux at a frequency in the power-law part of both spectra, and the values of $C(\sigma)$ are given in Appendix B. The values of the parameters used in the model are summarized in Table \[tab-model\].
We note that using the [SRCUT]{}-type model (i.e., setting $\alpha_{path}=
1$), our model gives parameters for the broad-band spectrum consistent with those of @pannuti03, who derived high magnetic field ($\sim 150~\mu$G) and unrealistically low magnetic field filling factor ($\sim 10^{-3}$), the latter being the reason they rejected IC as a plausible mechanism for TeV emission production in G347.3–0.5. However, we find that the X-ray data seem to prefer slightly slower steepening of the cutoff than the sharp exponential cutoff. Thus our model with $\alpha_{path}= 0.5$ results in different broad-band model parameters which are more reasonable from the physical point of view. Furthermore, our results illustrate the strong dependence of inferred parameters on the detailed model used to describe the synchrotron X-rays. This strong dependence means that particular parameter values obtained from this kind of broad-band fitting may be far less well-determined than their statistical uncertainties within a particular model may seem to indicate. Since our purpose here is mainly to illustrate that a synchrotron/IC broadband model is not ruled out by the data, we need only demonstrate that some reasonable model in this class can reproduce the observations, and we feel we have done so.
Thus, we find that, if the CANGAROO spectrum of G347.3–0.5 is correct, we have to decouple the electron emitting volume from the magnetic field volume to model the synchrotron spectrum with the parameters from the IC spectrum. We note that a magnetic field filling factor ($f_B$) value of $\sim 1$% seems very low, but MHD simulations do predict magnetic field enhancements in small-scale features produced by turbulence generated in SNR shocks [@jun99]. This is supported by the radio polarization maps of G348.3-0.5, which show that the magnetic filed in this SNR is very patchy, which could be, in part, due to the turbulent amplification. Also, this is the factor by which the volume occupied by magnetic field producing the bulk of the synchrotron emission is smaller than the volume occupied by ultrarelativistic electrons producing IC photons. In other words, the $15\ \mu$G magnetic field is confined to $\sim 1$% of the region from which the IC emission is emitted. This requires that the rest of the region has much lower magnetic field, $\sim 0.7\,\mu$G, in order not to exceed the total observed synchrotron emission from the SNR.
Our value for magnetic field is consistent with that calculated by @ellison01, but our value for maximum energy of accelerated particles is factor of 3 lower. This is not surprising since $E_{max}$ is constrained by the TeV spectrum and their model was based on a single TeV measurement, which lies in the middle of the TeV spectrum and would thus push $E_{max}$ to the higher values. Using the rolloff frequency obtained from the [SRCUT]{} fit to field 1 spectra and our value for magnetic field, we get $E_{max}\sim
5.0$ TeV, using relationship $$\nu_{rolloff}\approx 1.61 \times 10^{16} \left ( \frac{B}{10
\mu {\rm G}} \right ) \left ( \frac{E}{10 {\rm TeV}} \right )^2 ~{\rm
Hz,}$$ which is the corrected version of the relationship reported in @reynolds99 which contained a confusion of a factor of 0.29 between maximum and characteristic frequencies . This value of $E_{max}$ is thus consistent with our model. In any case, given the uncertainties inherent in this kind of broad-band modeling of a highly inhomogeneous source, even a factor of 3 difference in inferred parameters should be regarded as rough consistency.
Maximum Energy of Accelerated Electrons
---------------------------------------
Our fitted value of $E_{max}$ of 5 TeV, with a magnetic field strength of $15 \ \mu$G, gives a peak emitting frequency of $\nu_{ms} \sim 2
\times 10^{15}$ Hz. This frequency is low enough to put significant constraints on shock-acceleration models producing the electron spectral cutoff by radiative losses. Simple estimates of the cutoff energy [e.g., @reynolds98] in which the electron scattering mean free path is a constant factor $\eta \ge 1$ times the gyroradius give $$E_{max} \sim 20 \ \eta^{-1/2}
\left( B \over {15 \ \mu{\rm G}} \right)^{-1/2} u_8 \ {\rm TeV}$$ where the shock normal is parallel to the mean upstream magnetic field, and $$E_{max} \sim 20 \ \eta^{+1/2}
\left( B \over {15 \ \mu{\rm G}} \right)^{-1/2} u_8 \ {\rm TeV}$$ where the shock normal is more nearly perpendicular. The only difference is in the $\eta$-dependence, due to more rapid acceleration in perpendicular shocks [@jokipii87]. Also, $u_8$ is the shock velocity in units of $10^8$ cm s$^{-1}$, and $B$ is the postshock field strength. The above estimate is weakly dependent on the assumption that the shock compression ratio $r$ is 4.
These expressions give a cutoff frequency $$\nu_{ms}({\rm perp}) \equiv 1.82 \times 10^{18} E_{max}^2 B
= 3 \times 10^{16} \eta~u_8^2 \ {\rm Hz}$$ for a perpendicular shock, and $$\nu_{ms}({\rm par}) = 3 \times 10^{16} \eta^{-1}~u_8^2 \ {\rm Hz}$$ for a parallel shock. These frequencies are independent of the magnetic field strength, since $E_{max} \propto B^{-1/2}$
If the shock velocity in G347.3–0.5 is about 1000 km s$^{-1}$ [@ellison01], it is impossible to produce the observed low cutoff frequency in perpendicular shocks — the acceleration rate is too fast. If the shocks are parallel, larger $\eta$ will slow the acceleration rate and lower the cutoff frequency. For instance, if $u_{\rm sh} = 900$ km s$^{-1}$, then $\eta \sim 10$ would produce our inferred cutoff frequency. In the quasilinear approximation, electron scattering is due to the amplitude of magnetic fluctuations $\delta B$ with wavelengths comparable to the electron’s gyroradius. In this picture, $\eta = (\delta B/B)^{-2}$. In strong turbulence, $\eta$ approaches 1 (Bohm limit). A value of 10 is often assumed for moderately strong ($\delta B/B
\sim 0.3$) turbulence.
The issue of the nature of the cutoff has important implications for the acceleration of ions. If the electron spectrum is limited by radiative losses, the proton spectrum might extend to much higher energies, while any other cutoff mechanism for electrons should affect protons as well. We deduce that if the shocks accelerating the electrons producing synchrotron X-rays in G347.3–0.5 are largely parallel shocks with fairly high levels of MHD turbulence, then the rate of electron acceleration can balance the loss rate for electron energies around the 5 TeV cutoff we obtain from fitting the TeV photon spectrum, and we cannot constrain any turnover in the accelerated proton spectrum. If the shocks do not have these properties, or if the acceleration is due to some other process altogether, we should expect that the proton spectrum also cuts off around 5 TeV.
Thus, assuming that protons are accelerated in the same manner as electrons, we can use the same particle distribution for electrons and protons (i.e., $\sigma=\sigma_p$ and $E_{max}=E_{max,p}$) to calculate total energy content of the relativistic electrons and protons [see @allen01]. We obtained a total electron energy of 3.7[$\times 10^{46}$]{} erg and a total proton energy of 1.7[$\times 10^{49}$]{} erg, which yields total particle energy that is significantly smaller than the SNR energy budget ($\sim 1-2 \times
10^{51}$erg). The total magnetic energy yields 4.6[$\times 10^{46}$]{} erg for $B=15\mu$G and $f_B=0.01$, remarkably close to equipartition with electrons.
CONCLUSIONS
===========
[*Chandra*]{} and ATCA observations of G347.3–0.5 have been used to investigate the SNR’s morphology and spectral properties with high angular resolution. The main results are summarized below.
The X-ray emission from the remnant is dominated by nonthermal power law emission, as found in previous observations. High resolution X-ray observations with [*Chandra*]{} reveal a complex morphology of the northwestern SNR region composed of bright filaments embedded in faint diffuse emission. Our spectral analysis implies that there are significant variations between the spectral properties of the fainter regions, which have steeper spectra ($\Gamma \sim 2.5$), and that of the bright regions, which have flatter spectra ($\Gamma
\sim 2.1$).
To improve an estimate of the thermal emission in the SNR, we studied in particular the emission in the ACIS-S3 detectors which have the best response to the soft emission. We found a possible trace of thermal emission, but were unable to constrain its properties. No thermal component was detected in regions covered by the ACIS-I detectors.
Due to a complex environment the SNR is located in, we were unable to derive an accurate spectral index for radio emission. Using the ATCA data at 1.4 and 2.5 GHz we derived an approximate spectral index of the northwestern SNR region to be $0.50\pm0.40$.
Significant linear polarization of $\sim 5$–10% is detected with the magnetic field being most ordered towards the northwestern SNR filament, where the linear polarization reaches 12–30%. The low mean value is comparable to that found in the historical shell remnants [see references in @reynolds93], and is considerably smaller than often found in larger, older remnants. It is thus consistent with the otherwise surprising result that a remnant as large as G347.3-0.5 could have such strong shock acceleration.
We used the small scale morphology of [*Chandra*]{} data to identify possible regions of particle acceleration in the SNR. The X-ray morphology of the northwestern SNR region corresponds fairly well with the radio morphology, implying that the same population of electrons is responsible for the emission in both bands. We show that synchrotron/IC models cannot be ruled out for the broad-band spectrum, if we allow the possibility that the magnetic field is spatially inhomogeneous and enhanced in small regions. The data require a magnetic field of $\sim 15 \mu$G occupying $\sim 1\%$ of the IC-emitting volume which is filled with relativistic electrons. The maximum energy of accelerated electrons is found to be $\sim 5$ TeV. The derived total electron and magnetic energies are close to equipartition in this model.
While non-thermal bremsstrahlung is not considered as a potential TeV emission mechanism because it would violate the limits on the thermal part of the bremsstrahlung emission from the SNR [@ellison01; @pannuti03], there are objections for both IC and pion-decay processes. IC requires small magnetic field filling factor of nonthermal electrons ($f_B \sim 1\%$), while pion-decay requires extreme gas densities. While the very small filling factor may seem implausible, there is some observational evidence for such regions in G347.3–0.5. Future $\gamma$-ray telescopes with improved spatial resolution will help localize the TeV emission regions more precisely, enabling more accurate broad-band models to be derived.
We thank G. Allen and D. Ellison for invaluable discussions and D. Galloway for help with programing in IDL. This work was supported in part by NASA contract NAS8–39073 (POS), grant GO0–1123X (JSL) and GO2–3080B (JPH).
The Australia Telescope Compact Array is part of the Australia Telescope funded by the Commonwealth of Australia for operation as a National Facility, managed by CSIRO. This research has made use of the NASA/IPAC Infrared Science Archive, which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.
Synchrotron and Inverse Compton Models
======================================
The energy distribution in our synchrotron and inverse Compton (IC) models is taken to be a power law with an exponential cutoff [e.g @gaisser98]: $$\frac{dN}{dE} = A_e~E^{-\sigma} exp \left [-
\left ( \frac{E}{E_{max}} \right )^{\alpha_{path}} \right ] ,$$ where $A_e$ is the normalization factor, $\sigma$ is the index of the particle distribution $E_{max}$ is the maximum energy of accelerated particles, and $\alpha_{path}$ is an empirical parameter allowing a broadening of the cutoff apparently required by observations [see e.g., @ellison01; @uchiyama03].
The total synchrotron power radiated by a single electron is [e.g., @blum70] $$p(\nu) = \frac{\sqrt{3} e^3 B_{\bot}}{m c^2} \frac{\nu}{\nu_c}
\int^{\infty}_{\nu /\nu_c} K_{5/3}(\xi)d\xi ,$$ where $\nu_c$ is the critical frequency given by $$\nu_c = \frac{3 e B_{\bot}}{4 \pi m^3 c^5} E^2,$$ $B_{\bot}$ is the magnetic field component perpendicular to the line of sight, and $K_{5/3}$ is the modified Bessel function of 5/3 order. The total power radiated by the population of electrons with the modified power law distribution $N(E) \equiv dN/dE$ is then $$P_{tot}(\nu ,E)=\int_{E_{min}}^{E_{max}} p(\nu) N(E) dE.$$
For the IC spectrum we follow the formulae of @baring99. The probability that a photon with initial energy $\varepsilon_s m_e c^2$ will collide with an electron of energy $\gamma_e=(E_{kin} - m_e
c^2)/m_e c^2$ and upscatter to the energy $\varepsilon_{\gamma} m_e
c^2$ is given in the general case with the Klein-Nishina cross section as: $$\sigma_{\rm K-N}(\varepsilon_s, \gamma_e; \varepsilon_{\gamma})=
\frac{2 \pi r_{0}^2}{\varepsilon_s \gamma_{e}^{2}}
\left [ 2q~log_e q + 1 + q - 2q^2 +
\frac{\delta_e^2 q^2 (1 - q)}{2(1 + \delta_e q)} \right ],$$ where $r_0 = e^2/(m_e c^2)$ is the electron radius, $\delta_e = 4
\varepsilon_s \gamma_e$ is a parameter that controls the importance of photon recoil and Klein-Nishina effects [@blum70], and $q$ is given as: $$q = \frac{\varepsilon_{\gamma}}{\delta_e (\gamma_e - \varepsilon_{\gamma})},
~~~0\ge q \ge 1.$$ We take the photon field to be the cosmic microwave background (CMB) radiation, which has the spectral distribution of a blackbody at temperature of $T=2.73$ K: $$n_{\gamma} (\varepsilon_s)=\left ( \frac{\varepsilon_{s}^{2}}{\pi^2 \lambda_{c}^{3}}
\right ) \left ( \frac{1}{e^{\varepsilon_s/\Theta} - 1}
\right )$$ where $\Theta=\frac{k T}{m_e c^2}$, and $\lambda_c=\hbar /(m_e c)$ is the Compton wavelength. The IC emissivity is then $$\frac{dn_{\gamma}(\varepsilon_{\gamma})}{dt}=
c \int N_e (\gamma_e) d\gamma_e \int n_{\gamma}(\varepsilon_s) d\varepsilon_s
\sigma_{\rm K-N}(\varepsilon_s, \gamma_e; \varepsilon_{\gamma}),$$ where $N_e (\gamma_e) \equiv dN/dE$ is the electron energy distribution. We have ignored other sources of seed photons for two reasons. First, in typical SNR environments the energy density of the local IR/optical radiation field is an order of magnitude or less than the CMB [@gaisser98]. While no emission in [*IRAS (Infrared Astronomical Satellite)*]{} images can be clearly associated with G347.3–0.5, an upper limit for the flux at 100$\mu$m can be estimated by taking all emission within the remnant boundaries. This gives about 450 Jy, or a mean intensity of about 0.16 Jy arcmin$^{-2}$, implying in the source an energy density $\nu u_\nu
\sim 0.01$ eV cm$^{-3}$. This upper limit is far smaller than that of the CMB (0.26 eV cm$^{-3}$). Upper limits from the shorter-wavelength [*IRAS*]{} bands are even smaller than this. Second, the Klein-Nishina parameter $\delta_e$ is $\sim 10^{-2}$ for our 5 TeV electrons, meaning that the scattering cross-section is essentially the Thompson cross-section. For $\delta_e > 1,$ the cross section begins to drop substantially due to Klein-Nishina suppression. This corresponds to photon wavelengths shortward of about 50 $\mu$m. So mid-to near-IR and optical seed photons would contribute substantially less to the IC emission, even if their energy densities were comparable to that of the CMB. We are thus justified in neglecting IR and higher photon fields in calculating the inverse-Compton emission from G347.3–0.5.
Magnetic Field and Its Filling Factor
=====================================
For a simple homogeneous source filled with relativistic electrons and magnetic field (with a filling factor $f_B$), we can write down simple relations between the synchrotron emission and inverse Compton emission from CMB seed photons. We consider spectra plotted as $\nu
F_\nu$, so that both components rise with the same slope $1-\alpha$ (where $\alpha$ is radio spectral index) to peaks at frequencies $\nu_{ps}$ and $\nu_{pi}$, respectively. Since Compton upscattering increases scattered photon energies by a factor $\sim 2 (E/m_e
c^2)^2$, a photon at the peak of the 2.73 K CMB spectrum with frequency $5.6 \times 10^{10}$ Hz emerges at a frequency $\nu_i = 1.7
\times 10^{23} E^2$ Hz (with electron energy $E$ in erg). Now electrons with energy $E_m$ radiate the peak of their synchrotron radiation (SR) spectrum at a frequency $\nu_{m} = 1.82 \times 10^{18}
E_m^2 B$ [e.g., @pacholczyk70]. However, the peak of the resulting $\nu F_\nu$ spectrum is not necessarily at $\nu_m$. The delta-function approximation to the single-electron synchrotron emissivity, integrated over our modified exponentially cut off electron spectrum, gives a volume emissivity $j_\nu \propto
\nu^{-\alpha} \exp(-(\nu / \nu_m)^{\alpha_{\rm path}/2})$. The function $\nu j_\nu$ has a maximum at $$\nu_{ps} =
\left( {3 - \sigma} \over \alpha_{\rm path} \right)^{2/\alpha_{\rm path}}
\nu_m \equiv C_{\rm path} \nu_m.$$ For $\alpha_{\rm path} = 1,$ $\nu_p = \nu_m,$ but if $\alpha_{\rm path} < 1,$ the increased broadening of the turnover moves the peak in $\nu j_\nu$ to higher frequencies, more so for flatter electron spectra (lower $\sigma$). (For $\sigma = 2$ and $\alpha_{\rm path} = 0.5,
C_{\rm path} = 16.$) We can then use the observed ratio of peak inverse-Compton frequency to peak synchrotron frequency $R_\nu \equiv \nu_{pi} / \nu_{ps}$ to solve for the magnetic field: $$B = 9 \times 10^4 C_{\rm path}^{-1}
\left( {\nu_{pi} \over \nu_{ps}} \right)^{-1} \ {\rm G}
\equiv 9 \times 10^4 C_{\rm path}^{-1} R_\nu^{-1} \ {\rm G}.$$
We can write down the ratio of synchrotron to inverse Compton emissivity at a given frequency (assuming a power law electron spectrum and neglecting Klein-Nishina effects, as is appropriate below the peak frequencies) from, e.g., @pacholczyk70 [@rybicki79], for the case of a thermal photon distribution at temperature $T$ $${j_\nu(SR) \over j_\nu(IC)} = {c_j K B^{1 + \alpha} \nu^{-\alpha}
\over K \left( m_e c^2 \right)^{\sigma - 1} \left( 2 \pi r_0^2 \right)
\left( hc \right)^{-2} \left( kT \right)^{(\sigma + 5)/2} F(\sigma)
\left( h\nu \right)^{-\alpha}}$$ where $K = A_e c^{\sigma + 1}$ (so $dN/dE = K E^{-\sigma}$ far below the cutoff). Here $r_0$ is the classical electron radius, $h$ is Planck’s constant, and $c_j = c_5(\alpha) (1.25 \times
10^{19})^\alpha$ in the notation of @pacholczyk70 ($c_5 \sim
10^{-23}$ where cgs units are used throughout). Also, $$F(\sigma) = {\sigma^2 + 4\sigma + 11 \over \left( \sigma+3 \right)^2
(\sigma + 5) (\sigma + 1)} 2^{\sigma + 3} \Gamma_f\left( {{\sigma
+ 5} \over 2} \right) \zeta\left( {{\sigma + 5} \over 2} \right)$$ where $\Gamma_f(z)$ is the complete gamma-function and $\zeta(z)$ is the Riemann zeta-function [see, e.g., @abramowitz65]. This can be reduced to $${j_\nu(SR) \over j_\nu(IC)} = 7.92 \times 10^{-8} \ { c_j B^{1 +
\alpha} \left( m_e c^2 \right)^{\sigma - 1} \over \left( kT
\right)^{(\sigma + 5)/2} F(\sigma) h^{(\sigma - 1)/2}}.$$ Inserting $T = 2.73$ K for the cosmic microwave background, we obtain finally $${j_\nu(SR) \over j_\nu(IC)} \equiv \left[ C(\sigma) \right]^{-1} B^{1
+ \alpha},$$ defining $C(\sigma)$ for later convenience we have $C(2.2) = 6.8
\times 10^{-14}$, $C(2.0) = 2.1 \times 10^{-14}$, and $C(1.7) = 3.6
\times 10^{-15}$.
Now the emitting volume of synchrotron radiation may be smaller than that of inverse Compton radiation, if the magnetic field occupies a fraction $f_B \le 1$ of the volume. Then the ratio of SR to IC fluxes is smaller by a factor $f_B$. Including this factor and inverting equation (B5) above, we obtain $$f_B = C(\sigma) {F_\nu(SR) \over F_\nu(IC)} B^{-(1 + \alpha)}.$$ Estimates based on similar considerations (but for $f_B \equiv 1$) were obtained by @aharonian99. These results reproduce the results of detailed model fitting to within factors of 25% for $B$ and 2 for $f_B$.
Abramowitz, M., & Stegun, I. 1965, Handbook of Mathematical Functions (New York: Dover)
Aharonian, F. A. & Atoyan, A. M. 1999, , 351, 330
Allen, G. E., Petre, R., & Gotthelf, E. V. 2001, , 558, 739
Axford, W. I. 1994, , 90, 937
Baring, M. G., Ellison, D. C., Reynolds, S. P., Grenier, I. A., & Goret, P. 1999, , 513, 311
Blandford, R. & Eichler, D. 1987, , 154, 1
Blumenthal, G. R. & Gould, R. J. 1970, Reviews of Modern Physics, 42, 237
Butt, Y. M., Torres, D. F., Combi, J. A., Dame, T., & Romero, G. E. 2001, , 562, L167
Butt, Y. M., Torres, D. F., Romero, G. E., Dame, T. M., & Combi, J. A. 2002, , 418, 499
Crawford, F., Gaensler, B. M., Kaspi, V. M., Manchester, R. N., Camilo, F., Lyne, A. G., & Pivovaroff, M. J. 2001, , 554, 152
Dyer, K. K., Reynolds, S. P., Borkowski, K. J., Allen, G. E., & Petre, R. 2001, , 551, 439
Ellison, D. C., Slane, P., & Gaensler, B. M. 2001, , 563, 191
Enomoto, R. et al. 2002, , 416, 823
Frater, R. H., Brooks, J. W., & Whiteoak, J. B. 1992, Journal of Electrical and Electronics Engineering Australia, 12, 103
Hartman, R. C. et al. 1999,ApJS, 123, 79
Hendrick, S. P. & Reynolds, S. P. 2001, , 559, 903
Gaensler, B. M., Dickey, J. M., McClure-Griffiths, N. M., Green, A. J., Wieringa, M. H., & Haynes, R. F. 2001, , 549, 959
Gaisser, T. K., Protheroe, R. J., & Stanev, T. 1998, , 492, 219
Ginzburg, V. L., 1957, Prog. Elem. Particle Cosm. Ray Phys. 4, 339
Katz-Stone, D. M. & Rudnick, L. 1997, , 488, 146
Killeen, N. E. B., Bicknell, G. V., & Ekers, R. D. 1986, , 302, 306
Koyama, K., Petre, R., Gotthelf, E. V., Hwang, U., Matsura, M., Ozaki, M., & Holt, S. S. 1995, , 378, 255
Koyama, K., Kinugasa, K., Matsuzaki, K., Nishiuchi, M., Sugizaki, M., Torii, K., Yamauchi, S., & Aschenbach, B. 1997, , 49, L7
Jokipii, J.R., 1987, ApJ, 313, 842
Jones, F. C. & Ellison, D. C. 1991, Space Science Reviews, 58, 259
Jun, B. & Jones, T. W. 1999, , 511, 774
Lazendic, J. S., Slane, P. O., Gaensler, B. M., Plucinsky, P. P., Hughes, J. P., Galloway, D. K., & Crawford, F. 2003, , 593, L27
Long, K. S., Reynolds, S. P., Raymond, J. C., Winkler, P. F., Dyer, K. K., & Petre, R. 2003, , 586, 1162
Mastichiadis, A. & de Jager, O. C. 1996, , 311, L5
Muraishi, H. et al. 2000, , 354, L57
Pacholczyk, A.G. 1970, Radio Astrophysics (San Francisco: Freeman)
Pannuti, T. G., Allen, G. E., Houck, J. C., & Sturner, S. J. 2003, , 593, 377
Pfeffermann, E. & Aschenbach, B. 1996, Röntgenstrahlung from the Universe, eds. Zimmermann, H.U., Trümper, J., and Yorke, H., MPE Report 263, 267
Prigozhin, G. Y., Kissel, S. E., Bautz, M. W., Grant, C., LaMarr, B., Foster, R. F., Ricker, G. R., & Garmire, G. P. 2000, Proc. SPIE, 4012, 720
Reimer, O. & Pohl, M. 2002, , 390, L43
Reynolds, S. P. & Chevalier, R. A. 1981, , 245, 912
Reynolds, S. P. & Gilmore, D. M. 1993, , 106, 272
Reynolds, S. P. 1998, , 493, 375
Reynolds, S. P. & Keohane, J. W. 1999, , 525, 368
Rho, J., Dyer, K. K., Borkowski, K. J., & Reynolds, S. P. 2002, , 581, 1116
Rybicki & Lightman 1979, Radiative processes in astrophysics (New York: Wiley-Interscience)
Sault, R. J. & Wieringa, M. H. 1994, , 108, 585
Sault, R. J., Staveley-Smith, L., & Brouw, W. N. 1996, , 120, 375
Sault, R. J., Bock, D. C.-J., & Duncan, A. R. 1999, , 139, 387
Sault, R. J. & Killeen, N., 2001, in Miriad Users Guide (Sydney: Australia Telescope National Facility)
Shklovsky, I. S., 1953, Dokl. akad. Nauk. SSSR, 91, 475
Slane, P., Gaensler, B. M., Dame, T. M., Hughes, J. P., Plucinsky, P. P., & Green, A. 1999, , 525, 35
Slane, P., Hughes, J. P., Edgar, R. J., Plucinsky, P. P., Miyata, E., Tsunemi, H., & Aschenbach, B. 2001, , 548, 814
Tanimori, T. et al. 1998, , 497, L25
Townsley, L. K., Broos, P. S., Garmire, G. P., & Nousek, J. A. 2000, , 534, L139
Townsley, L. K., Broos, P. S., Nousek, J. A. & Garmire, G. P. 2002, Nucl. Instrum. Methods Phys. Res. A, in press
Uchiyama, Y., Takahashi, T., & Aharonian, F. A. 2002, , 54, L73
Uchiyama, Y., Aharonian, F. A. & Takahashi, T. 2003, , 400, 567
[ll]{}
$I_{1.4}$ FWHM beam & $46\farcs 3 \times 36\farcs 2$, P.A.=$-3.8$\
RMS noise in $I_{1.4}$ & 0.7 [mJybeam$^{-1}$]{}\
$Q_{1.4}, U_{1.4}$ FWHM beam & $57\farcs 1 \times 48\farcs 8$, P.A.=$-5.1$\
RMS noise in $Q_{1.4}, U_{1.4}$ & 0.2 [mJybeam$^{-1}$]{}\
$I_{2.5}$ FWHM beam & $24\farcs 7 \times 17\farcs 9$, P.A.=$-$1.6\
RMS noise in $I_{2.5}$ & 0.3 [mJybeam$^{-1}$]{}\
$Q_{2.5}, U_{2.5}$ FWHM beam & $32\farcs 5 \times 27\farcs 5$, P.A.=$-$3.1\
RMS noise in $Q_{2.5}, U_{2.5}$ & 0.3 [mJybeam$^{-1}$]{}\
\[tab-atca\]
[lllll]{}
$N_{\rm H}$ ([$10^{21}$]{}[cm$^{-2}$]{}) & &\
$\Gamma$/$\nu_{rolloff}$ (Hz) & $2.10^{+0.04}_{-0.03}$ & $2.53^{+0.07}_{-0.07}$ & $6.3^{+0.8}_{-0.6} \times 10^{17}$ & $1.2^{+0.3}_{-0.4} \times 10^{17}$\
$F_X$/$S_{\nu}$ ([ ergcm$^{-2}$s$^{-1}$]{}/Jy) & $1.12^{+0.01}_{-0.03}$[$\times 10^{-10}$]{} & $2.65^{+0.07}_{-0.13}$[$\times 10^{-11}$]{} & $4.2^{+0.4}_{-0.2}$ & $2.9^{+0.6}_{-0.4}$\
$\chi^{2}_{red}$/dof & &\
\[tab-joint-fit\]
[ll]{} Derived &\
Maximum electron energy, $E_{max}$ & 5.0$_{-0.2}^{+0.1}$ TeV\
Electron spectral index, $\sigma$ & 2.0$_{-0.04}^{+0.05}$\
Broadening factor, $\alpha_{path}$ & 0.5$_{-0.02}^{+0.03}$\
Electron normalization, $A_e$ & 5.5$_{-0.1}^{+0.1}$[$\times 10^{-32}$]{}\
Magnetic field strength, $B$ & 15$_{-0.03}^{+0.05}$ $\mu$G\
$B$ filling factor, $f_B$ & 0.010$^{-0.003}_{+0.002}$\
Fixed &\
SNR volume, $V_{E}$ & 4.9[$\times 10^{59}$]{} [cm$^{-3}$]{}\
Emission filling factor, $f_{E}$ & 0.16\
Outer SNR radius, $R_{O}$ & 40 pc\
Inner SNR radius, $R_{I}$ & 34 pc\
SNR distance, $D_{pc}$ & 6.3 kpc\
reduced $\chi^2$/ degrees of freedom & 1.55/338\
\[tab-model\]
\
[^1]: Collaboration of Australia and Nippon for a Gamma-ray Observatory in the Outback
[^2]: Energetic Gamma Ray Experiment Telescope
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Fairness is becoming an increasingly important concern when designing markets, allocation procedures, and computer systems. I survey some recent developments in the field of multi-agent fair allocation.'
author:
- |
Haris Aziz\
UNSW Sydney and Data61\
Sydney, Australia\
[email protected]
title: 'Developments in Multi-Agent Fair Allocation'
---
Introduction
============
Many important decisions in our lives are increasingly being made by computers. These decisions include whether we get a certain job or loan. It becomes imperative then that computer systems provide guarantees that the decision has been made transparently and fairly. For many settings such as two-sided matching, fairness considerations based on justified envy or diversity guarantees are enshrined in the letter of the law. Even for smaller day to day decisions, the appetite for fairness guarantees is naturally present among people.[^1] At a bigger scale, the Internet leads to peer to peer interactions where fairness comes up as an important concern (see, e.g. @Moul18 and @KRT01a).
The issue of what is fair and how to achieve fairness has been studied in several disciplines including economics [@Moul03a; @Thom11b], political science [@Bram08a; @BaYo82a] and philosophy [@Resc02a; @Rawl71a].[^2]
I mention some recent developments in fair allocation in multi-agent systems. In particular, I discuss allocation of indivisible items (Section \[sec:indiv\]), allocation of divisible items (Section \[sec:div\]), two-sided matching (Section \[sec:match\]), and fairness in other social choice settings (Section \[sec:sc\]). Finally, I wrap up the discussion in Section \[sec:discuss\].
Allocation of Indivisible Items {#sec:indiv}
===============================
When indivisible items are allocated among agents, guaranteeing the existence of fair allocations is impossible for many desirable notions of fairness. For example, if there are two agents and one indivisible valued good, allocating the good to either of the two agents will make the other agent envious. In other words, an envy-free allocation may not exist. Another challenge is computational. Suppose agents have positive additive utilities over the items. Then the problem of checking whether there exists an envy-free allocation is NP-complete even for 1-0 utilities [@AGMW15a]. Similarly the problems of computing allocations that maximize the egalitarian welfare (utility of the worst off agent) or Nash welfare (geometric mean of the utilities of agents) is NP-hard (see, e.g. @NNRJ14a). Given these challenges, a couple of influential fairness concepts were proposed by @Budi11a. The first one is called *maxmin share (MMS)* fairness. The *maxmin fair share* of an agent is the best she can guarantee for herself if she is allowed to partition the items into bundles for the agents but then receives her least preferred bundle. An allocation satisfies MMS fairness if each agent’s allocation is at least as preferred to her as her maxmin fair share. Although an MMS fair allocation is not guaranteed to exist even for additive utilities, the concept gives rise to approximation versions in which each agent aspires to get a certain percentage of her maxmin fair share [@PrWa14a]. This has led to several results including new algorithms that are faster, simpler, or provide better approximation guarantees, or work for more general families of utilities (see e.g. @ARSW17a, @AMNS15a, @BaKr17 and @XiLu19a).
The second solution concept that has been proposed is a relaxation of envy-freeness called *envy-free up to one item (EF1)*. An allocation satisfies EF1 if it is envy-free or any agent’s envy for another agent can be removed if some item is ignored. Under additive utilities, EF1 can be achieved by a simple algorithm called the round-robin sequential allocation algorithm. Agents take turn in a round-robin manner and pick their most preferred unallocated item. The interest in EF1 was especially piqued when @CKM+16a proved that for positive additive utilities, a rule based on maximizing Nash social welfare finds an allocation that is both EF1 and Pareto optimal. The result has led to very interesting followup work. For example, @BKV18a presented a pseudo-polynomial time algorithm for the same setting to compute an Pareto optimal and EF1 allocation. The returned allocation also provides a 1.45 approximation of the maximum Nash welfare.[^3]
A majority of the results on indivisible items assume that the items are goods yielding positive utilities (see e.g. @BCM15a [@LaRo16]). In recent years, there has been a push to obtain similar existence and algorithmic results for negative utilities (see e.g. @Aziz16a) or more general utility functions [@PlRo18; @OPS19a]. For example, @ACI+18 designed an algorithm that returns in polynomial time an EF1 allocation for any class of utility functions.
There are ongoing efforts to develop models and algorithms that take into account real-life features including distributional constraints, dynamic and online settings, distributed settings, and asymmetric agents. An interesting new setting that has been considered is one in which the agents are partitioned into groups and the items are allocated to groups [@SeSu19a]. Researchers have also been exploring stronger or alternative fairness and efficiency concepts. For example, there is a renewed focus on concepts that concern envy between groups of agents [@AzRe19a; @CFS+19; @BCEZ19]. Another direction concerns scenarios where items are viewed as nodes of a graph and only those allocations are considered in which each agent gets a connected coalition [@BCE+17a].
Allocation of Divisible Items {#sec:div}
=============================
When divisible items are allocated among agents, it is easier to guarantee the existence of fair allocations. It can also be relatively easier to compute a fair allocation. Again, consider the example of two agents and one item. If the item is divisible, we can simply allocate half of the item to each agent, guaranteeing many reasonable fairness criteria including envy-freeness.
In contrast to the case of indivisible items, an envy-free and Pareto optimal allocation always exists for the case of divisible items. For positive additive utilities, such an allocation can be computed by maximizing Nash social welfare in polynomial time. When the utilities are additive but negative, there are still important gaps in our understanding of the computational complexity of envy-free and Pareto optimal allocations (see e.g. @BMSY17).
For divisible items, one important class of utility functions is called *Leontief preferences* in which each agent requires fixed proportions of the items (such as 1 CPU and 5 GB RAM to perform one task). @GZH+11a proposed a solution for the problem of allocating as much of each item as possible while equalizing the agents’ use of the fraction of their own dominant (bottleneck) resource. The method satisfies envy-freeness, proportionality, Pareto optimality, and strategyproofness. These solutions have been extended in important ways such as dealing with indivisibilities and dynamic settings [@PPS12a].
Another model of divisible items is called cake-cutting [@RoWe98a; @Proc15a; @Proc12a]. A cake is a unit interval representing a heterogeneous item. An agent may have different utilities for different subintervals, even if they are of equal size. The challenge is to compute fair allocations using queried information from agents. Cake-cutting has a very long history dating back to the inception of the mathematical theory of fair division. Two of the most well-studied concepts are envy-freeness and proportionality (each agent gets at least $1/n$ of her value of the whole cake). While there are well-known and simple algorithms to compute a proportional allocation, finding envy-free allocations is more challenging. In the last few years, new algorithms have been proposed that compute an envy-free allocation in a number of queries that is bounded in the number of agents. The algorithms work for either positive utilities [@AzMa16c] or negative utilities [@DFHY18a]. There is no known bounded algorithm for the case of mixed utilities. It has also been noted by the authors that if the requirement to allocate the whole cake is relaxed, then there are simpler and faster algorithms to compute an allocation that is both envy-free and proportional. In recent years, several new variants of the cake-cutting problems have been studied (see, e.g @SHA15a).
Recently, researchers have approached problems where indivisible items are allocated among agents, but a minimal number of items are shared or treated as divisible to guarantee existence of fair allocations [@SaSe19b]. The approach can be viewed as a natural extension of the Adjusted Winner rule (see e.g. @BrTo96a) that finds an envy-free and Pareto optimal allocation among two agents in which only at most one item is shared.
Money can also be viewed as a divisible item that can be moved around to achieve fairness. A well-studied discrete allocation problem with money is the fair room-rent division problem (see e.g. @ADG91a [@ASU04a; @GMPZ17a]).
Two-Sided Matching {#sec:match}
==================
In two sided matching markets (see e.g. @Manl13a), the goal is to match entities from two different sets. A typical example of the problem is one where students/doctors are matched to school/hospitals. Students have preferences over schools. Schools have priorities over students. An outcome in these settings is a matching of students to schools.
In two-sided matching, an important criterion of desirable outcomes is a fairness property that requires that no student should want to replace another student in a hospital when the hospital has a higher priority for the former student over the latter student. The concept is also based on the idea of envy. The only difference is that an agent can only envy another agent if the former has higher priority.
The property of fairness along with that of non-wastefulness (no student prefers to take some empty slot in a school) are the central properties that underpin much of the work on two-sided matching. For many general two-sided matching markets, the two properties can be simultaneously satisfied by an allocation returned by the well-known Deferred Acceptance algorithm [@Roth08a] or its generalizations. For several important settings such as the allocation of school seats to students [@AbSo03b], one version of the algorithm additionally satisfies strategyproofness.
An important research agenda extends this kind of result to other two-sided matching problems with distributional constraints. These constraints included upper quotas on regions of hospitals (see e.g. @KaKo15 [@BFIM10; @GIKY+16a]) as well as target lower quotas for diverse representation [@AGSW19a; @EHYY14a; @KHIY15a]. These new settings give rise to fundamental questions of tradeoffs between merit, diversity, and segregation. Yet another direction is taking a time-sharing approach to two-sided matching (see e.g. @AzKl19b [@KeUn15a]).
Fairness in other Social Choice Settings {#sec:sc}
========================================
We have mostly focussed on allocations problems where agents get ownerships of their allocation. We can also consider allocation of public goods such as the funding of public projects. Public goods can be selected via voting on the different options. In classical voting settings in which a single alternative is to be selected, it is very difficult to guarantee reasonable representation of sizable minorities. Consider the case where 51$\%$ of the agents most prefer alternative $a$ and 49$\%$ of the agents prefer alternative $b$. Then alternative $a$ that is supported by a majority is selected even though the remaining agents may find it unpalatable.
There are other voting settings where fairness concerns become more meaningful. One such setting is a simple extension of single-winner voting to selecting multiple winners (see e.g. @ABC+16a [@FSST17a; @Aziz19b]). Another setting is probabilistic voting or portioning in which each alternative is given a certain probability weight or a fraction of a budget (see e.g. @AAC+19a [@ABM19a; @Aziz18a; @Bran17a]). In both settings, meaningful fairness axioms have been proposed and algorithms have been designed to satisfy the axioms. A general principle while formalizing the axioms is that of proportional representation: if a large enough group cohesively prefers some outcome(s), then that group should get suitable representation (see e.g. @AzLe19a). Recently, fairness concerns are also being examined in participatory budgeting problems in which projects have costs and there is a constraint on the maximum budget that can be used [@FGM16a; @ALT18a; @FST+17a].
Another collective decision-making setting where fairness becomes more meaningful is the case where there are several issues with each issue having its own candidates outcomes [@CF017]. Agents have preferences over outcomes over each of the issues. The setting is more general than allocation of indivisible items in following way. Each issue corresponds to the decision of allocation of a particular item. The possible outcomes of an issue correspond to giving the item to one of the agents. It turns out that a solution based on maximizing Nash social welfare satisfies both Pareto optimality and a natural relaxation of proportionality.
Discussion {#sec:discuss}
==========
I discussed some settings where fairness plays an important role. Fairness concerns can come up in many other multi-agent settings as well such as auctions and coalition formation. Much of the work in multi-agent fair allocation is centered around formal axioms to capture fairness.[^4] It will be interesting to use data-centric approaches to understand how humans conceptualize fairness or decide on tradeoffs between societal outcomes.
Another area of computer science where fairness is taking center stage is machine learning. Machine learning algorithms learn from data to make actionable suggestions such as giving a loan. Whether these algorithm are biased is a prominent concern for society (see, e.g. @KLR18a). A natural approach to incorporate fairness in these algorithms follows the Aristotelian “equal treatment of equals” principle. It is an interesting challenge to explore ways in which ideas from multi-agent fair allocation can contribute to fairness in machine learning. In general, the interplay of fairness with other desirable aspects such as accountability, transparency, and trust is broad research question (see, e.g. @ShPa19a).
As computers start to make important decisions in our lives, it is timely to revisit the lessons learnt. New computer applications will give rise to novel and challenging problems so it is critical to continue updating our toolkit of axioms and methods for designing and enabling fair multi-agent systems.
Acknowledgements {#acknowledgements .unnumbered}
================
The survey is based on a presentation given by the author at the Dagstuhl-Seminar on Application-Oriented Computational Social Choice (Seminar Number 19381). Aziz is supported by a UNSW Scientia Fellowship, and Defence Science and Technology (DST) under the project “Auctioning for distributed multi vehicle planning” (DST 9190). The author thanks Alex Lam and the reviewers for useful comments.
Abdulkadiro[ğ]{}lu, A., and S[ö]{}nmez, T. 2003. School choice: A mechanism design approach. 93(3):729–747.
Abdulkadiro[ğ]{}lu, A.; S[ö]{}nmez, T.; and [Ü]{}nver, M. U. 2004. Room assignment-rent division: A market approach. 22(3):515–538.
Airiau, S.; Aziz, H.; Caragiannis, I.; Kruger, J.; Lang, J.; and Peters, D. 2019. Portioning using ordinal preferences: Fairness and efficiency. In [*Proceedings of the 28h International Joint Conference on Artificial Intelligence (IJCAI)*]{}.
Alkan, A.; Demange, G.; and Gale, D. 1991. Fair allocation of indivisible goods and criteria of justice. 59:1023–1039.
Amanatidis, G.; Markakis, E.; Nikzad, A.; and Saberi, A. 2015. Approximation algorithms for computing maximin share allocations. In [*Proceedings of the 35th International Colloquium on Automata, Languages, and Programming (ICALP)*]{}, 39–51.
Atkinson, A. 1970. On the measurement of inequality. 2(3):244–263.
Aziz, H., and Klaus, B. 2019. Random matching under priorities: Stability and no envy concepts. 53(2):213–259.
Aziz, H., and Lee, B. 2019. The expanding approvals rule: Improving proportional representation and monotonicity. .
Aziz, H., and Mackenzie, S. 2016. A discrete and bounded envy-free cake cutting protocol for any number of agents. In [*Proceedings of the 57th Symposium on Foundations of Computer Science (FOCS)*]{}, 416–427.
Aziz, H., and Rey, S. 2019. Almost group envy-free allocation of indivisible goods and chores. abs/1907.09279.
Aziz, H.; Gaspers, S.; Mackenzie, S.; and Walsh, T. 2015. Fair assignment of indivisible objects under ordinal preferences. 227:71–92.
Aziz, H.; Brill, M.; Conitzer, V.; Elkind, E.; Freeman, R.; and Walsh, T. 2017a. Justified representation in approval-based committee voting. 48(2):461–485.
Aziz, H.; Rauchecker, G.; Schryen, G.; and Walsh, T. 2017b. Algorithms for max-min share fair allocation of indivisible chores. In [*Proceedings of the 31st AAAI Conference on Artificial Intelligence (AAAI)*]{}, 335–341.
Aziz, H.; Caragiannis, I.; Igarashi, A.; and Walsh, T. 2019a. Fair allocation of indivisible goods and chores. In [*Proceedings of the 28th International Joint Conference on Artificial Intelligence (IJCAI)*]{}.
Aziz, H.; Gaspers, S.; Sun, Z.; and Walsh, T. 2019b. From matching with diversity constraints to matching with regional quotas. In [*Proceedings of the 18th International Conference on Autonomous Agents and Multi-Agent Systems (AAMAS)*]{}.
Aziz, H.; Bogomolnaia, A.; and Moulin, H. 2019. Fair mixing: the case of dichotomous preferences. In [*Proceedings of the 20th ACM Conference on Economics and Computation (ACM-EC)*]{}, 753–781.
Aziz, H.; Lee, B. E.; and Talmon, N. 2018. Proportionally representative participatory budgeting: Axioms and algorithms. In [*Proceedings of the 17th International Conference on Autonomous Agents and MultiAgent Systems, [AAMAS]{} 2018, Stockholm, Sweden, July 10-15, 2018*]{}, 23–31.
Aziz, H. 2016. Computational social choice: Some current and new directions. In [*Proceedings of the 25th International Joint Conference on Artificial Intelligence (IJCAI)*]{}, 4054–4057.
Aziz, H. 2019a. A probabilistic approach to voting, allocation, matching, and coalition formation. In Laslier, J.-F.; Moulin, H.; Sanver, R.; and Zwicker, W. S., eds., [*The Future of Economic Design*]{}. Springer-Verlag. Forthcoming.
Aziz, H. 2019b. A rule for committee selection with soft diversity constraints. 1–8.
Balinski, M., and Young, H. P. 1982. . Yale University Press. (2nd Edition \[with identical pagination\], Brookings Institution Press, 2001).
Barman, S., and Krishnamurthy, S. K. 2017. Approximation algorithms for maximin fair division. In [*Proceedings of the 18th ACM Conference on Economics and Computation (ACM-EC)*]{}.
Barman, S.; Krishnamurthy, S. K.; and Vaish, R. 2018. Finding fair and efficient allocations. In [*Proceedings of the 2018 [ACM]{} Conference on Economics and Computation, Ithaca, NY, USA, June 18-22, 2018*]{}, 557–574.
Benabbou, N.; Chakraborty, M.; Elkind, E.; and Zick, Y. 2019. Fairness towards groups of agents in the allocation of indivisible items. In [*Proceedings of the 28th International Joint Conference on Artificial Intelligence (IJCAI)*]{}.
Biro, P.; Fleiner, T.; Irving, R. W.; and Manlove, D. F. 2010. The college admissions problem with lower and common quotas. 411(34–36):3136–3153.
Bogomolnaia, A.; Moulin, H.; Sandomirskiy, F.; and Yanovskaya, E. 2017. Competitive division of a mixed manna. 85(6):1847–1871.
Bouveret, S.; Cechl[á]{}rov[á]{}, K.; Elkind, E.; Igarashi, A.; and Peters, D. 2017. Fair division of a graph. In [*Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence, [I \]JCAI]{} 2017, Melbourne, Australia, August 19-25, 2017*]{}, 135–141.
Bouveret, S.; Chevaleyre, Y.; and Maudet, N. 2016. Fair allocation of indivisible goods. In Brandt, F.; Conitzer, V.; Endriss, U.; Lang, J.; and Procaccia, A. D., eds., [*Handbook of Computational Social Choice*]{}. Cambridge University Press. chapter 12.
Brams, S. J., and Togman, J. M. 1996. Camp david: Was the agreement fair? 15(1):99–112.
Brams, S. J. 2008. . Princeton University Press.
Brandt, F. 2017. Rolling the dice: [R]{}ecent results in probabilistic social choice. In Endriss, U., ed., [*Trends in Computational Social Choice*]{}. AI Access. chapter 1, 3–26.
Budish, E. 2011. The combinatorial assignment problem: Approximate competitive equilibrium from equal incomes. 119(6):1061–1103.
Caragiannis, I.; Kurokawa, D.; Moulin, H.; Procaccia, A. D.; Shah, N.; and Wang, J. 2016. . In [*Proceedings of the 17th ACM Conference on Economics and Computation (ACM-EC)*]{}, 305–322.
Cole, R., and Gkatzelis, V. 2015. Approximating the [N]{}ash social welfare with indivisible items. In [*Proceedings of the 47th Annual ACM Symposium on Theory of Computing (STOC)*]{}, 371–380. ACM Press.
Conitzer, V.; Freeman, R.; Shah, N.; and Vaughan, J. W. 2019. Group fairness for the allocation of indivisible goods. In [*Proceedings of the 33rd AAAI Conference on Artificial Intelligence (AAAI)*]{}.
Conitzer, V.; Freeman, R.; and Shah, N. 2017. Fair public decision making. In [*Proceedings of the 2017 [ACM]{} Conference on Economics and Computation, [EC]{} ’17*]{}, 629–646.
Dehghani, S.; Farhadi, A.; Hajiaghayi, M. T.; and Yami, H. 2018. Envy-free chore division for an arbitrary number of agents. In [*Proceedings of the Twenty-Ninth Annual [ACM-SIAM]{} Symposium on Discrete Algorithms, [SODA]{} 2018, New Orleans, LA, USA, January 7-10, 2018*]{}, 2564–2583.
Ehlers, L.; Hafalir, I. E.; Yenmezb, M. B.; and Yildirimc, M. A. 2014. School choice with controlled choice constraints: Hard bounds versus soft bounds. 153:648—683.
Fain, B.; Goel, A.; and Munagala, K. 2016. The core of the participatory budgeting problem. In [*Web and Internet Economics - 12th International Conference, [WINE]{} 2016, Montreal, Canada, December 11-14, 2016, Proceedings*]{}, 384–399.
Faliszewski, P.; Skowron, P.; Slinko, A.; and Talmon, N. 2017. Multiwinner voting: A new challenge for social choice theory. In Endriss, U., ed., [*Trends in Computational Social Choice*]{}. chapter 2.
Fluschnik, T.; Skowron, P.; Triphaus, M.; and Wilker, K. 2017. Fair knapsack. abs/1711.04520.
Gal, Y.; Mash, M.; Procaccia, A. D.; and Zick, Y. 2017. Which is the fairest (rent division) of them all? 64(6):39:1–39:22.
Ghodsi, A.; Zaharia, M.; Hindman, B.; Konwinski, A.; Shenker, S.; and Stoica, I. 2011. Dominant resource fairness: Fair allocation of multiple resource types. In [*Proceedings of the 8th USENIX Symposium on Networked Systems Design and Implementation*]{}.
Goto, M.; Iwasaki, A.; Kawasaki, Y.; Kurata, R.; Yasuda, Y.; and Yokoo, M. 2016. Strategyproof matching with regional minimum and maximum quotas. 235:40–57.
Huang, X., and Lu, P. 2019. An algorithmic framework for approximating maximin share allocation of chores. abs/1907.04505.
Kamada, Y., and Kojima, F. 2015. Efficient matching under distributional constraints: Theory and applications. 105(1):67–99.
Kesten, O., and Unver, U. 2015. A theory of school choice lotteries. 543—595.
Kleinberg, J.; Ludwig, J.; Mullainathan, S.; and Rambachan, A. 2018. Algorithmic fairness. 108:22–27.
Kleinberg, J. M.; Rabani, Y.; and Tardos, [É]{}. 2001. Fairness in routing and load balancing. 63(1):2–20.
Kurata, R.; Hamada, N.; Iwasaki, A.; and Yokoo, M. 2015. Controlled school choice with soft bounds and overlapping types. In [*Proceedings of the 29th AAAI Conference on Artificial Intelligence (AAAI)*]{}, 951–957.
Lang, J., and Rothe, J. 2016. Fair division of indivisible goods. In Rothe, J., ed., [*Economics and Computation*]{}. 493–550.
Manlove, D. F. 2013. . World Scientific Publishing Company.
Moulin, H. 2003. . The MIT Press.
Moulin, H. 2019. Fair division in the age of internet. 11:407–441.
Nguyen, N.-T.; Nguyen, T.-T.; Roos, M.; and Rothe, J. 2014. Computational complexity and approximability of social welfare optimization in multiagent resource allocation. 28(2):256–289.
Oh, H.; Procaccia, A. D.; and Suksompong, W. 2019. Fairly allocating many goods with few queries. In [*Proceedings of the 33rd AAAI Conference on Artificial Intelligence (AAAI)*]{}, 2141–2148.
Parkes, D. C.; Procaccia, A. D.; and Shah, N. 2013. Beyond dominant resource fairness: Extensions, limitations, and indivisibilities. In [*Proceedings of the 13th ACM Conference on Electronic Commerce (ACM-EC)*]{}, 808–825.
Plaut, B., and Roughgarden, T. 2018. Almost envy-freeness with general valuations. In [*Proceedings of the 29th Annual ACM-SIAM Symposium on Discrete Algorithms (SODA)*]{}.
Procaccia, A. D., and Wang, J. 2014. Fair enough: Guaranteeing approximate maximin shares. In [*Proceedings of the 15th ACM Conference on Economics and Computation (ACM-EC)*]{}, 675–692. ACM Press.
Procaccia, A. D. 2013. Cake cutting: Not just child’s play. 56(7):78–87.
Procaccia, A. D. 2016. Cake cutting algorithms. In Brandt, F.; Conitzer, V.; Endriss, U.; Lang, J.; and Procaccia, A. D., eds., [*Handbook of Computational Social Choice*]{}. Cambridge University Press. chapter 13.
Rawls, J. 1971. . Harvard University Press.
Rescher, N. 2002. . Transaction Publishers.
Robertson, J. M., and Webb, W. A. 1998. . A. K. Peters.
Roth, A. E. 2008. Deferred acceptance algorithms: history, theory, practice, and open questions. 36:537—569.
Sandomirskiy, F., and Segal-Halevi, E. 2019. Fair division with minimal sharing. abs/1908.01669.
Segal-Halevi, E., and Suksompong, W. 2019. Democratic fair allocation of indivisible goods. 277:103–167.
Segal[-]{}Halevi, E.; Hassidim, A.; and Aumann, Y. 2015. Envy-free cake-cutting in two dimensions. In [*Proceedings of the 29th AAAI Conference on Artificial Intelligence (AAAI)*]{}, 1021–1028.
Shin, D., and Park, Y. J. 2019. Role of fairness, accountability, and transparency in algorithmic affordance. 98:277–284.
Thomson, W. 2011. Chapter twenty-one - fair allocation rules. In Arrow, K. J.; Sen, A.; and Suzumura, K., eds., [*Handbook of Social Choice and Welfare*]{}, volume 2 of [*Handbook of Social Choice and Welfare*]{}. Elsevier. 393 – 506.
Thomson, W. 2016. Introduction to the theory of fair allocation. In Brandt, F.; Conitzer, V.; Endriss, U.; Lang, J.; and Procaccia, A. D., eds., [*Handbook of Computational Social Choice*]{}. Cambridge University Press. chapter 11.
Varian, H. R. 1974. Equity, envy, and efficiency. 9:63–91.
[^1]: See, for example, the interest in fair allocation portal [www.spliddit.org/](www.spliddit.org/)).
[^2]: Despite the richness of fairness concepts and principles in these literatures, I will only focus on a handful of fairness concepts in this paper. Other fairness ideas include indices to measure inequality (see, e.g. @Atki70a) or market-based approaches in which agents have equal budgets (see, e.g. @Vari74a).
[^3]: There is a stream of papers primarily focussed on approximating Nash welfare. For example, @CoGz15a presented a polynomial-time 2.889-approximation algorithm for maximum Nash welfare.
[^4]: In addition to standard axioms such as envy-freeness, a formal theory of fair allocation covers many different axiom for desirable rules or algorithms (see, e.g. @Thom15a.)
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Superfluid $T_c$ of liquid helium-3 and its pressure dependence are calculated by using a relation obtained from our macro-orbital microscopic theory. The results agree closely with experiments. This underlines the accuracy of our relation and its potential to provide superfluid $T_c$ of electron fluid in widely different superconductors and renders experimental foundation to our conclusion ([*cond-mat/0603784*]{}) related to the basic factors responsible for the formation of ([**q**]{}, -[**q**]{}) bound pairs of fermions and the onset of superfluidity in a fermionic system. Since available experimental studies of superconductors pertaining to changes in lattice parameters around their superconducting $T_c$ seem to support a link between lattice strain and the onset of superconductivity, need for similar studies is emphasized.'
---
**Superfluid $T_c$ of Helium-3 and its Pressure Dependence**
**Yatendra S. Jain**
Department of Physics
North-Eastern Hill University, Shillong - 793 022, India
Key-words : helium-3, superfluidity, transition temperature
PACS : 64.70.Ja, 67.57.-z, 67.57.Bc
email: [email protected]
[**1. Introduction**]{}
Liquid $^3He$ has been a subject of extensive theoretical and experimental studies \[1-5\] for the last six decades for several reasons including its superfluid behavior and it appears that it will continue to fascinate the researchers for many more decades to come. However, as remarked by Georges and Laloux \[6\] several aspects of even normal state of the liquid at low temperatures (LT), [*viz.*]{}: (i) increase in inertial mass as revealed by the experimental values of its LT specific heat, (ii) many fold increase in its magnetic susceptibility indicating as if it is at the blink of ferromagnetic instability, and (iii) nearly temperature independent low compressibility need better understanding. Two models, [*viz.*]{}: (i) “almost ferromagnetic” \[7\] and “almost localized” \[8,9\] have been extensively tried to account for these aspects. Identifying that the two models are seemingly contradictory, Georges and Laloux \[6\] propose Mott-stoner liquid model. However, in view of our recently developed microscopic model of a [*system of interacting fermions*]{} (SIF) used to conclude the basic foundations of superconductivity \[10\] both these pictures coexist. We, therefore, have a detailed program to study different aspects of liquid $^3He$ ([*including those listed above*]{}) in the framework of our model which is based on the macro-orbital representation of a particle in a many body system. We also use this representation to conclude the long awaited microscopic theory a system of interacting bosons such as liquid $^4He$ \[11\].
It is evident that the results and inferences of our model \[10\] can be applied to understand the normal and superfluid behavior of liquid $^3He$. We note that some of our conclusions to some extent agree with those of the well known BCS theory \[12\] of superconductivity. For example we find that superconductivity is a consequence of [*bound pairs*]{} of electrons moving with equal and opposite momenta ([**q**]{}, -[**q**]{}). But in variance with BCS theory, the binding of such pairs is basically found \[10\] to be a consequence of the mechanical strain in the lattice forced by the zero-point force of electrons arising from their zero point energy; the electrical polarization of the lattice emphasized by the BCS model may have its $+ve$ or $-ve$ contribution to this binding. In addition our approach reveals a single theoretical framework for the superconductivity of conventional as well as high $T_c$ systems and finds that superconductivity can, in principle, be observed at temperatures as high as room temperature (RT). It renders mathematically simple formulations and microscopic foundations to the well known phenomenological theories ([*viz.*]{} two fluid theory of Landau \[13\] and $\Psi-$theory of Ginzburg \[14\]). It concludes that superfluid and superconducting transitions are a kind of [*quantum phase transitions*]{} which, however, occur at a non-zero $T$ due to proximity effect of quasi-particle excitations. Guided by all these factors, we use our approach to: (i) estimate the value of superfluid $T_c$ of liquid $^3He$ which has been identified to be a difficult problem \[1\], (ii) study its pressure dependence, and (iii) analyze their consistency with experiments. The details of other important properties of the liquid would be analyzed in our forthcoming paper(s).
Theoretical calculations, predicting possible value(s) of $T_c$ of superfluid $^3He$, based on BCS picture were reported within a year of the publication of the BCS theory. While the first few studies \[15, 16\] indicated that the liquid was unlikely to have a superfluid transition, Emery and Sessler \[17\] concluded that a second order transition may occur at a $T$ between 50 to 100 mK. However, when the transition was really observed between 0.92 to 2.6 mK \[18\] (depending on the pressure on the liquid), calculations by Levin and Valls \[19, 20\] not only obtained a $T_c$ close to these values but also found its pressure dependence closely matching with experiments. Almost similar estimates have been reported recently by Rasul and coworkers \[21, 22\]. Widely different inferences and estimated $T_c$ from different theoretical calculations using common picture (BCS Theory) as their central idea seem to indicate the complexity of estimates and lack of reliability. On the other hand, the merit of our theory \[10\] lies with the fact that it does not have any scope to use different considerations to obtain different $T_c$ which indicates its reliability. In addition the fact that our $T_c$ for superfluid $^3He$ its pressure dependence agrees closely with experiments indicates its accuracy.
[**2 Superfluid $T_c$ and its Pressure Dependence**]{}
Using the universal component ($H_o(N)$, Eqn. 2 of \[10\]) of the net hamiltonian $H(N)$ of a SIF (Eqn. 1 of \[10\]), such as electron fluid in a conductor or liquid $^3He$, we find that its particles below $$T_c \approx \epsilon_g/k_B =
\frac{h^2}{8k_Bmd^2}\frac{\Delta d}{d} \eqno(1)$$
assume a state of bound pairs and the system as a whole has a transition to its superfluid state \[10\]. $\epsilon_g$ and $m$ in Eqn. 1, respectively, represent the binding energy (or energy gap) and mass of a fermion with (i) $d = (V/N)^{1/3)}$ and (ii) $\Delta d =$ an increase in $d$ forced by the zero-point force of a fermion occupying its ground state in a cavity (size $=d$) formed by neighboring fermions. It may be noted that for electrons in a conductor $d$ in Eqn. 1 represents diameter $d_c$ of the channels [*through which conduction electrons move in the lattice*]{} \[10\]. In view of the fact revealed from the experimentally observed specific heat values of liquid $^3He$, a particle in an interacting environment of the liquid at a $T$ closed to $T_c$ starts behaving like a quasi-particle of mass $m^*$, we use $$T_c = \frac{h^2}{8k_Bm^*d^2}\frac{\Delta d}{d} \eqno(2)$$
Fig. 1 : Pressure dependence of expansion in $He-He$ bond $\Delta d$ (in Å). While Curve A is obtained from the molar volume data of Kollar and Vollhardt \[24\], Curve B represents the linear fit (Eqn. 3) after excluding point P=0 at Curve A (see text).
to obtain $T_c$ of liquid $^3He$ at different pressures. In this we define $\Delta d = d(T=0) - d_{min}$ with $d_{min} = d$ at the point of maximum density of the liquid for a chosen pressure. As shown for the simple case of a particle trapped in 1-D box \[23\], we identify \[10\] the zero-point force of a particle occupying its ground state in the cavity of neighboring particles as the microscopic reason for the expansion of the liquid on cooling below certain $T < T_F$ (Fermi temperature). We determine $d$ and $\Delta d$ by using molar volume of the liquid recently reported by Kollar and Vollhardt \[24\]. However, as indicated by Kollar and Vollhardt \[24\] themselves and the plot of $\Delta d$ [*vs.*]{} $T$ in Figure 1, their data for $P=0$ seem to have large systematic errors; note that $\Delta d$ at $P=0$ falls considerably away from any logical trend in which other points can be fitted. Consequently, we discarded this point and obtained a linear fit $$P = 5806.15\Delta d - 21.6865 \eqno (3)$$
for all other points by using a standard computer software. In this context not only the remaining points seem to fall closely on the line but a linear change in $\Delta d$ with increasing $P$ is also expected because $\Delta d$ is a kind of strain in $He-He$ bonds. As such we used Eqn.3 to obtain $\Delta d$ values for our calculations of $T_c$ at different pressures including $P = 0$. To obtain $m^*$ that enters in Eqn. 2, we note that as per our theoretical formulation the quasi-particle excitations which contribute to the specific heat of the fermionic system of non-interacting particles have $4m$ mass. Obviously, when the impact of interactions is included, we have $4m^*$ as the mass of the quasi-particle which, obviously, equals the effective mass ($m^*_{sp}$) that we obtain from specific heat data \[25\]. In other words we use $m^* = m^*_{sp}/4$ in Eqn. 2 to obtain our $T_c$ at which the superfluid phase transition is expected as per our theory \[10\]. The $T_c$ values so obtained are tabulated with experimental values in Table I and both are plotted in Fig. 2 for their comparison. The fact that our $m^*$ changes from 0.7525$m(^3He)$ to 1.4233$m(^3He)$ with pressure increasing from 0 to 28.0 bar indicates that inter-particle interaction dominated locally by zero-point repulsion slowly assumes attractive nature (at $\approx$ 10 bar pressure) with $^3He$ atoms having increased electric dipolemoment with increasing pressure.
Fig. 2 : Pressure dependence of superfluid $T_c$ of liquid $^3He$. Curve A (experimental, [*cf.*]{} Column 7, Table I) and Curve B (Eqn 2).
Table I : Calculated and experimental $T_c$ and related data
---------- ------- ------------ ---------------- ----------- -------------- -------------------
Pressure $d$ $\Delta d$ $m^*_{sp}$$^+$ $m^*$ $T_c(eqn.2)$ $T_c(Exp)$$^{++}$
(in bar) $\AA$ $\AA$ $m(^3He)$ $m(^3He)$ mK mK
0.0 3.94 .00374 3.010 0.7525 2.0564 0.92
5.0 3.78 .00460 3.629 0.9073 2.3667 1.60
10.0 3.69 .00546 4.183 1.0458 2.6208 1.99
15.0 3.63 .00632 4.670 1.1675 2.8598 2.21
20.0 3.58 .00718 5.084 1.2710 3.1057 2.37
25.0 3.54 .00804 5.472 1.3680 3.3399 2.47
28.0 3.52 .00856 5.693 1.4233 3.4782 2.52
---------- ------- ------------ ---------------- ----------- -------------- -------------------
$^+$ obtained from graphical plots of $m^*_{sp}$ values \[25\], $^{++}$Zero pressure value from \[26\] and others from \[25\].
[**3. Discussion**]{}
The BCS model, basically formulated to explain superconductivity of conventional superconductors, has been used to understand superfluidity and related aspects of liquid $^3He$ because the electron fluid in conductors and liquid $^3He$ are closely identical SIF; of course suitable modifications ([*e.g.*]{}, fermions participating in Cooper pairing in the latter case have $p-$state not $s-$state) compatible with the model are adopted. This paper uses the same considerations to apply the basic foundations of superconductivity \[10\] revealed from our non-conventional theoretical framework which emphasizes mechanical strain ([*in the crystalline lattice of a superconductor or in inter-atomic bonds in case of liquid $^3He$ type*]{} SIF) as the main source of ([**q**]{}, -[**q**]{}) bound pair formation. As established in \[23, 27\], such strain is [*a basic consequence of zero-point force arising from the wave particle duality*]{} and it ought to be present whenever a particle occupies its ground state in a box or cavity of its neighboring particles or in a channel through which it is free to move. While electrons in superconductors create strain in the lattice structure of the channels through which they move \[10\], a $He$ atom creates this strain in $He-He$ bonds which it makes with its neighboring atoms \[11\]. The experimental fact that liquid $^3He$ as well as liquid $^4He$ show $-ve$ thermal expansion at $T \approx T_o$ ([*the temperature equivalent of the ground state energy of $He$ atom in a cavity of neighboring atoms*]{}) confirms the presence of strain in $He-He$ bonds. Evidently, our theoretical estimate of superfluid $T_c$ of liquid $^3He$ and its pressure dependence ([*cf.*]{}, Table 1 and Figure 2), which have close agreement with experiments \[25,26\], undoubtedly prove the accuracy of Eqn. 2 and conclusions of \[23 and 27\]. It also demonstrates the potential of our theory \[10\] to predict the superfluid $T_c$ of a SIF which has been a difficult task in the framework of conventional BCS theory \[1\]. In other words Eqn. 2 can be used to estimate the superconducting $T_c$ of widely different superconductors (including [*high $T_c$ superconductors*]{}) if accurate values of $d$, $\Delta d$ and $m^*$ are known. Several experimental studies \[[*e.g.*]{}, 28-33\] indicate that the occurrence of lattice strain or related effects such as negative expansion of lattice, hardening of lattice, anomalous or anisotropic change in lattice parameters, [*etc.*]{} around superconducting $T_c$ are common aspects of superconductors. This naturally supports our inference \[10\] regarding the relation between lattice strain and bound pair formation. However, the effect is not seen to be as clean as in liquid $^3He$ because an electron in a superconductor not only interacts with other electrons but also to the ions or atoms which decide their lattice arrangement through complex inter-particle interactions. Naturally, an accurate prediction of superconducting $T_c$ from Eqn.2 depends on the accuracy of the experimentally measured $\Delta d/d$, $d$ and $m^*$ for a chosen superconductor. In view of Eqn.2, $T_c$ increases with increase in $\Delta d/d$ and decrease in $d$ and $m^*$ and, depending on the values of these parameters, superfluid transition in a SIF can, in principle, occur at any temperature. This is corroborated by the facts that: (i) an atomic nucleus exhibits nucleon superfluidity at a $T$ much higher than even room temperature because nucleon-nucleon $d$ is found to be about $10^{-5}$ times shorter than $^3He-^3He$ distance which implies that the typical $T_c$ should be as high as $10^7$K (the mass of a nucleon and $^3He$ atom having same order of magnitude), and (ii) a typical superconducting $T_c$ falls around 10 K ($\approx 10^3$ times the superfluid $T_c$ of liquid $^3He$) because $m_e$ (mass of electron) is about 6000 times smaller than $m(^3He)$ (inter-electron distance or the channel size being nearly equal to $^3He - ^3He$ distance).
We note that Eqn.2 has been obtained by analyzing the universal component, $H_o(N) = H(N) - V'(N)$, of the net hamiltonian $H(N)$ ([*cf.*]{}, Eqn. 1 of \[10\]) of a SIF with $V'(N)$ representing the sum of all non-central potentials including spin-spin interactions; as such it considers only bare fermion-fermion central forces. Evidently, our estimates of $T_c$ of superfluid $^3He$ and its pressure dependence (Table 1 and Figure 2) exclude the contributions from $V'(N)$ (sum of interactions such as spin-spin interactions, electron-phonon interaction induced by electric polarization of the lattice, [*etc.*]{}) and possibly for this reason our estimates are about two times higher than experimental values. In view of these facts our estimates not only establish that the “mechanical strain” forced by zero-point force is the basic cause of bound pair formation in a SIF but also indicates that $V'(N)$ perturbations could be responsible for the supression of $T_c$ below our estimates. We hope that this would be supported by studies related to the impact of these perturbations on $T_c$ and its pressure dependence. Heiselberg [*et. al.*]{} \[34\] have summarized the important inferences of studies related to the impact of induced interactions such as BCS type attraction on the $T_c$ deduced from bare fermion-fermion interaction. They identify that such interactions in liquid $^3He$ are responsible for the ABM state to be energetically more favorable than BM state, while in neutron matter they suppress the superfluid gap significantly. The effect has been studied in dilute spin 1/2 Fermi gas by Gorkov and Melik-Barkhurdarov \[35\] who find that $T_c$ obtained from bare inter-particle interactions gets suppressed by a factor $(4e)^{1/3} \approx 2.2$. Evidently, all such effects of $V'(N)$ interactions (not included in deriving Eqn. 2) can be trusted for reducing the difference of our values of $T_c$ with experiments (Table 1 and Figure 2). We plan to examine these effects in our future course of studies. Interestingly, similar results of pressure dependent $T_c$ reported in \[19-22\] are also about two times higher than experimental values but it appears that these studies leave no factor(s) which could help in getting better agreement with experiments.
Finally, it may be mentioned that we have limited information about the thermal expansion of superconductors \[36\] around $T_c$ while the importance of its detailed and accurate measurements has been emphasized \[37\] soon after the discovery of high $T_c$ superconductors. The observation of negative lattice expansion, anisotropic thermal expansion, change in hardness, [*etc.*]{} around $T_c$ in a number of superconducting systems \[28-33, and 38-40\] not only re-emphasizes the importance of such studies but also indicates a relation of this effect with the onset of superfluidity in fermionic systems which naturally corroborates its mechanism as concluded by our theory \[10\].
[**4. Conclusion**]{}
The paper uses a relation obtained from our recently reported theoretical model \[10\] to estimate superfluid $T_c$ of $^3He$ and its pressure dependence. The close agreement between our estimates and experimental results indicates the accuracy of our model and the microscopic mechanism of superfluidity in a SIF like liquid $^3He$ and electron fluid in widely different superconductors. As suggested in \[28\], we also believe that accurate measurements of different aspects related to modifications in lattice structure, [*viz.*]{} thermal expansion, changes in lattice parameters, hardening, change in sound velocity, [*etc.*]{} around superconducting $T_c$ of widely different superconductors would be of great help in establishing the role of mechanical strain in the lattice as a basic component of the microscopic mechanism of superfluidity of different SIF and we hope that these would support our theory \[10\]. In this context it may be noted that liquid $^3He$ and liquid $^4He$ which do not have various complexities of electron fluid in conductors exhibit $-ve$ thermal expansion around superfluid $T_c$ as predicted by their respective microscopic theories \[10\] and \[11\] based on our macro-orbital approach. In addition it is significant that our approach has no space for subjective considerations which provide widely different estimates of $T_c$ as one may see with different $T_c$ values (0 to 100 mK) \[15-17, 19-22\] estimated by using only one model ([*the BCS picture*]{}) for superfluid transition in liquid $^3He$.
[??]{}
A. J. Leggett, Rev. Mod. Phys. [**76**]{}, 142 (2004); D.M. Lee, Rev. Mod. Phys. [**69**]{}, 645 (1997). D. Vollhardt and P. Wolfle, [*The Superfluid Phases of Helium -3*]{}, Taylor and Francis, London (1990). G.E. Volovik, [*Exotic Properties of Superfluid $^3He$*]{}, World Scientific, Singapore (1992). G.E. Volovik, [*The Universe in a Helium Droplet*]{}, Clarendon Press, Oxford (2003), A. Georges and L. Laloux, [*Normal Helium 3 : a Mott Stoner liquid*]{}, arXiv:cond-mat/9610076 (1996). K. Levin and O.T. Valls, Phys. Rep. [**98**]{}, 1 (1983). P.W. Anderson and W.F. Brinkman in [*The Helium Liquids*]{}, J.G.M. Armitage and I.E. Farqhar, Eds., Academic, New York (1975). D. Vollhardt, Rev. Mod. Phys. [**56**]{}, 99 (1984). Y.S. Jain, [*Basic Foundations of the Microscopic Theory of Superconductivity*]{}, arXiv:cond-mat/0603784 (2006). Y.S. Jain, [*Macro-orbitals and Microscopic Theory of a System of Interacting Bosons*]{}, arXiv:cond-mat/0606571 (2006). J. Bardeen, L.N. Cooper and Schrieffer, Phys. Rev. [**106**]{}, 162 (1957). L.D. Landau, J. Phys. (USSR) [**5**]{}, 71 (1941); English translation published in [*Helium 4*]{} by Z.M. Galasiewicz, Pergamon Press, Oxford (1971), pp 191-233. V.N. Ginzburg, Rev. Mod. Phys. [**76**]{}, 981 (2004). N.N. Bogoliubov, Doklady Akad. Nauk (U.S.S.R.) [**119**]{}, 1 (1958); Sov. Phys. Doklady [**3**]{}, 292 (1958). L.N. Cooper, R.L. Mills and A.M. Sessler, Phys. Rev. [**114**]{}, 1377 (1959). V.J. Emery and A.M. Sessler, Phys. Rev. [**119**]{}, 43 (1960). D.D. Osheroff, R.C. Richardson and D.M. Lee, Phys. Rev. Lett. [**28**]{}, 885 (1972). K. Levin and O.T. Valls, Phys. Rev. B [**17**]{}, 191 (1978). K. Levin and O.T. Valls, Phys. Rev. B [**20**]{}, 105 (1979). J.B. Rasul, T.C. Li and H. Beck, Phys. Rev. B [**39**]{}, 4191 (1989). J.B. Rasul, Phys. Rev. B [**45**]{}, 4191 (1992). Y.S. Jain, [*Untouched Aspects of the Wave Mechanics of a Particle in 1-D Box*]{}, arXiv:quant-ph/0606009 (2006). M. Kollar and D. Vollhardt, Phys. Rev. B [**61**]{}, 15347 (2000). J.C. Wheatley in [*The Helium Liquids*]{}, J.G.M. Armitage and I.E. Farqhar, Eds., Academic, New York (1975). C. Enss and S. Hunklinger, [*Low Temperature Physics*]{} Springer-Verlag, Berlin 2005. Y.S. Jain, [*Wave Mechanics of Two Hard core Particles in 1-D Box*]{}, arXiv:quant-ph/0603233 (2006); Cent. Euro. J Phys. [**2**]{}, 709 (2004). A. J. Mills and K.M. Rabe, Phys. Rev. B [**38**]{}, 8908 (1988). C. Meingast, O. Kraut, T. Wolf, H. Wuhl, A. Erb and G. Muller-Vogt, Phys. Rev. Lett. [**67**]{}, 1634 (1991). G. J. Burkhart and C. Meingast, Phys. Rev. B [**54**]{}, R6865 (1996). J. Kortus, I.I. Mazin, K.D. Belashchenko, V.P. Antropov and L.L. Boyer, Phys. Rev. Lett. [**86**]{}, 4656 (2001). J.M. An and W.E. Pickett, Phys. Rev. Lett. [**86**]{}, 4366 (2001). T. Yildirim et al, Phys. Rev. Lett. [**87**]{}, 037001 (2001). H. Heiselberg, C.J. Pethick, H. Smith and L. Viverit, Phys. Rev. Lett. [**85**]{}, 2418 (2000). L.P. Gorkov and T.K. Melik-Barkhudarov, Sov. Phys. JETP [**13**]{}, 1018 (1961). T.H.K. Barron, J.G. Collins and G.K. White, Adv. Phys. [**29**]{}, 609 (1980). A. de Visser [*et al*]{} Phys. Rev. B [**41**]{}, 7304 (1990). T. Takeuchi, H. Shishido, S. Ikeda, R. Settai, Y. Haga and Y. Onuki, J. Phys.: Condens Matter [**14**]{}, L261 (2002). A.C. Mclaughlin, F. Sher and J.P. Attfield, Nature [**436**]{}, 829(2005). J.D. Jorgensen, D.G. Hinks, P.G. Radaelli, W.I.F. David, R.M. Ibberson, [*Large Anisotropic Thermal Expansion Anomaly near the Superconducting Transition TEmperature in MgB$_2$*]{}, arXiv:cond-mat/0205486 (2002).
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Timely maintenance is an important means of increasing system dependability and life span. Fault Maintenance trees (FMTs) are an innovative framework incorporating both maintenance strategies and degradation models and serve as a good planning platform for balancing total costs (operational and maintenance) with dependability of a system. In this work, we apply the FMT formalism to a [Smart Building]{} application and propose a framework that efficiently encodes the FMT into Continuous Time Markov Chains. This allows us to obtain system dependability metrics such as system reliability and mean time to failure, as well as costs of maintenance and failures over time, for different maintenance policies. We illustrate the pertinence of our approach by evaluating various dependability metrics and maintenance strategies of a Heating, Ventilation and Air-Conditioning system. [^1]'
author:
- Nathalie Cauchi
- Khaza Anuarul Hoque
- Marielle Stoelinga
- Alessandro Abate
bibliography:
- 'Bib\_rep.bib'
title: Maintenance of Smart Buildings using Fault Trees
---
<ccs2012> <concept> <concept\_id>10010520.10010575.10010579</concept\_id> <concept\_desc>Computer systems organization Maintainability and maintenance</concept\_desc> <concept\_significance>300</concept\_significance> </concept> </ccs2012>
Introduction {#sec:Intro}
============
The Internet-of-things has enabled a new type of buildings, termed *Smart Buildings*, which aim to deliver useful building services that are cost effective, reliable, ubiquitous and ensure occupant comfort and productivity (thermal quality, air comfort). Smart buildings are equipped with many sensors such that a high level of intelligence is achieved: light and heating can be switched on automatically; fire and burglar alarms can be more sophisticated; and cleaning services can be connected to the occupancy rate. Maintenance is a key element to keep smart buildings smart: without proper maintenance (cleaning, replacements, etc.), the benefits of achieving greater efficiency, comfort, increased building lifespan, reliability and sustainability are quickly lost.
In this paper, we consider an important element in smart buildings, namely the heating, ventilation and air-conditioning (HVAC) system, responsible for maintaining thermal comfort and ensuring good air-quality in buildings.
One way of improving the lifespan and reliability of such systems is by employing methods to detect faults and to perform preventive and predictive maintenance actions. Techniques for fault detection and diagnosis for Smart Building applications have been developed in [@Yan2017; @boem2017distributed]. Predictive and preventive maintenance strategies are devised in [@cauchi2017model; @macek2017long; @Babishin201610480]. Moreover, a reliability-centered predictive maintenance policy is proposed in [@zhou2007reliability]. This policy is for a continuously monitored system which is subject to degradation due to imperfect maintenance. However, these techniques neglect reliability measurements and focus only on synthesis of maintenance policies in the presence of degradation and faults. [The current industrial standard for measuring a system’s reliability is the use of Fault trees, where the focus is on finding the root causes of a system failure using a top-down approach and do not incorporate degradation of system components and maintenance action [@ammar2016formal; @ruijters2015fault; @Li20151400]. ]{} [@ruijters2016fault] presents the Fault Maintenance Tree (FMT) as a framework that allows to perform planning strategies for balancing total costs and reliability and availability of the system. FMTs are an extension of FT encompassing both degradation and maintenance models. The degradation models represent the different levels of component degradation and are known as Extended Basic Events (EBE). The maintenance models incorporate the undertaken maintenance policy which includes both inspections and repairs. These are modelled using Repair and Inspection modules in the FMT framework.
[In literature, analysis of FMTs is performed using Statistical Model checking (SMC) [@ruijters2016fault], which generates sample executions of a stochastic system according to the distribution defined by the system and computes statistical guarantees based on the executions [@legay2010statistical].]{} In contrast, Probabilistic Model Checking (PMC) provides formal guarantees with higher accuracy when compared with SMC [@younes2006numerical], at a cost of being more memory intensive and may result in a state space explosion. PMC is an automatic procedure for establishing if a desired property holds in a probabilistic system model which encodes the probability of making a transition between states. This allows for making quantitative statements about the system’s behaviour which are expressed as probabilities or expectations [@kwiatkowskaadvances]. Probabilistic model checking has been successfully applied in a different domains so far including:[ aerospace and avionics [@hoque2017formal], optical communication [@siddique2017formal], systems biology [@dannenberg2013dna] and robotics [@feng2015controller]]{}. In this paper we tackle the FMT analysis using PMC. Our contributions can be summarised as follows:
1. We formalise the FMT using Continuous Time Markov Chain (CTMCs) and the dependability metrics of a Heating, Ventilation and Air-Conditioning (HVAC) system, using the Continuous Stochastic Logic (CSL) formalism, such that they can be computed using the PRISM model checker [@PRISM].
2. To tackle the state space explosion problem, we present an FMT abstraction technique which decomposes a large FMT into an equivalent abstract FMT based on a graph decomposition algorithm. This involves an intermediate step where the large FMT is transformed into an equivalent direct acyclic graph and decomposed into a set of small sub-graphs. Each of these small sub-graphs are converted to an equivalent smaller CTMC and analysed separately to compute the required metric, while maintaining the original FMT hierarchy. Using our framework, we are able to achieve a 67$\%$ reduction in the state space size.
3. Finally, we construct a FMT that identifies failure of an HVAC and illustrate the use of the developed framework to construct and analyse the FMT. We also evaluate relevant performance metrics using the PRISM model checker, compare different maintenance strategies and highlight the importance of performing maintenance actions.
[This article has the following structure: Section \[sec:hvac\] introduces the heating, ventilation and air-conditioning (HVAC) set-up under consideration together with the maintenance question we are addressing. This is followed by Section \[sec:Prelim\], which presents the fault maintenance trees and probabilistic model checking frameworks. Next, we present the developed methodology for modelling FMT using CTMCs [and perform model checking in Section \[subsec:FMTFrame:Modelling\]]{}. The framework is then applied to the HVAC system in Section \[sec:CaseStudy\].]{}
Problem formulation {#sec:hvac}
===================
We consider the heating, ventilation and air-conditioning (HVAC) system setup found within the Department of Computer Science, at the University of Oxford. A graphical description is shown in Figure \[fig:hvac-system\]. It is composed of two circuits - the air flow circuitry and the water circuit. The gas boiler heats up the supply water and transfers the supply water into two sections - the supply air heating coils and the radiators. The rate of water flowing in the heating coil is controlled using a heating coil valve, while the rate of water flow in the radiator is controlled using a separate valve. [The outside air is mixed with the air extracted from the zone via the mixer]{}. This is fed into the heating coil, which warms up the input air to the desired supply air temperature. This air is supplied back, at a rate controlled by the Air Handling unit (AHU) dampers, into the zone via the supply fan. The radiators are directly connected to the water circuitry and transfer the heat from the water into the zone. The return water, from both the heating coils and the radiators, [is then passed through the collector ]{} and is returned back to the boiler.
{width=".75\textwidth"}
The correct maintenance of this system is essential to ensure that the building operates with optimum efficiency while user comfort is maintained. The choice on the type of maintenance depends on several factors, including the different costs of maintenance and failures, and the [practical feasibility]{} of performing maintenance. To this end, we aim to address the following maintenance questions: (1) what is the optimal maintenance strategy that [minimises system failures?]{}; (2) what is the best trade-off between cost of inspections, operation and maintenance, vs the system’s number of expected failures?; (3) how frequently should the different maintenance actions such as performing a cleaning or a replacement be performed?; and (4) what is the effect of employing maintenance over a specific time horizon vs not performing maintenance?
Preliminaries {#sec:Prelim}
=============
Fault trees {#subsec:Prelim:FT}
-----------
Fault trees (FT) are directed acyclic graphs (DAG) describing the combinations of component failures that lead to system failures. It consists of two types of nodes: events and gates.
\[def:event\] An [event]{} is an occurrence within the system, typically the failure of a subsystem down to an individual component. Events can be divided into *basic events* (BEs) and intermediate events. BE occur spontaneously and denote the component/system failures while intermediate events are caused by one of or more other events. The event at the top of the tree, called the *top event* (TE), is the event being analysed, modeling the failure of the (sub)system under consideration [(both type of events are highlighted in Figure \[fig:FT\])]{}.
\[def:gates\] The internal nodes of the graph are called *gates* and describe the different ways that failures can interact to cause other components to fail i.e. how failures in subsystems can combine to cause a system failure. Each gate has one output and one or more inputs. The gates in a FT can be of several types and these include the AND gate, OR gate, k/N-gate [@ruijters2016fault]. The output of a gate triggers an event which shows how failures propagate through the fault tree.
Figure \[fig:FT\] depicts a fault tree were the basic events are shown using circles, top and intermediate events are depicted by a rectangle.
![Example of a FT with five basic events (1-5), two intermediate events ($B_1, B_2$) and top event $A$; failures are propagated by the gates ($G_1 - G_3$). []{data-label="fig:FT"}](FT){width="0.55\columnwidth"}
Fault maintenance trees {#subsec:Prelim:FMTFrame}
-----------------------
Fault maintenance trees (FMT) extend fault trees by including maintenance (all the standard FT gates are also employed by the FMTs). This is achieved by making use of:
1. *Extended Basic Events (EBE)* - The basic events are modified to incorporate degradation models of the component the EBE represents. The degradation models represent different discrete levels of degradations the components can be in and are a function of time. The timing diagram showing the progression of degradation within an EBE is shown in Figure \[fig:EBETiming\]. The presented EBE had $N$ discrete degradation levels, initially the EBE is its new state and it gradually moves from one degradation levels, based on the underlying distribution describing the degradation, to the next until the faulty level [$N$]{} is reached.
![Timing diagram of degradation within an EBE. []{data-label="fig:EBETiming"}](EBETimingDiagram){width="0.35\columnwidth"}
2. *Rate Dependency Events* - A new gate, introduced in [@ruijters2016fault] and labelled as *RDEP*, accelerates the degradation rates of [$n$ dependent child nodes]{} and is depicted in Figure \[fig:RDEPGate\]. When the component connected to the input of the RDEP fails, the degradation rate of the dependent components is accelerated with an acceleration factor $\gamma$. The corresponding timing diagram is shown in Figure \[fig:RDEPTiming\]. When the input signal is enabled ($input = 1$), the child EBE moves to the next degradation levels at a faster rate.
![Degradation level evolution of child EBE showing effect of RDEP on degradation rate. [Note, when the input is equal to $1$ the curve representing the degradation rate to go from one degradation level to the next (e.g. going from degradation level 2 to 3) is steeper vs previous degradation level transitions (e.g. going from 0 to 1 or 1 to 2).]{} []{data-label="fig:RDEPTiming"}](RDEPTimingDiagrams){width="0.35\columnwidth"}
‘
3. *Repair and Inspection modules* - The repair module (RM) performs cleaning or replacements actions. These actions can be either carried out using fixed time schedules or when enabled by the inspection module (IM). The RM module performs periodic maintenance actions (clean or replace), independently of the IM. The IM performs periodic inspections and when components fall below a certain degradation threshold a maintenance action is initiated by the IM and performed by the RM (outside of the RM’s periodic maintenance cycle). The IM and RM modules are depicted in Figure \[fig:IMRM\]. The effect of performing a maintenance cleaning or replacement action on the EBE is illustrated in Figure \[fig:RMIMtiming\]. When a cleaning action is performed the EBE moves back to its previous degradation level, while when a replacement is performed the EBE moves back to the initial level.
![High-level description of the inspection and repair modules. [The repair module performs maintenance actions periodically (clean or replace). The inspection module performs inspections periodically and when the degradation level of an EBE reaches *thresh* level, it triggers the repair module to perform a maintenance action immediately. ]{}[]{data-label="fig:IMRM"}](IMRM){width="0.3\columnwidth"}
![Degradation level progression of EBE for different maintenance actions.[]{data-label="fig:RMIMtiming"}](RMIMTimingDiagrams){width="0.8\columnwidth"}
A visual rendering of an FMT is given in Figure \[fig:FMTsmall\]. It is composed of five EBEs located at the bottom of the tree, one RDEP with one dependent child, three gates, one repair and inspection module and three events which show the different fault stages.
![Example of a fault maintenance tree.[]{data-label="fig:FMTsmall"}](FMT_IMRM){width="0.8\columnwidth"}
Probabilistic model checking {#subsec:Prelim:PMC}
----------------------------
Model checking [@Clarke86automaticverification] is a well-established formal verification technique used to verify the correctness of finite-state systems. Given a formal model of the system to be verified in terms of labelled state transitions and the properties to be verified in terms of temporal logic, the model checking algorithm exhaustively and automatically explores all the possible states in a system to verify if the property is satisfiable or not. *Probabilistic model checking* (PMC) deals with systems that exhibit stochastic behaviour and is based on the construction and analysis of a probabilistic model of the system. We make use of CTMCs[, having both transition and state labels, ]{} to perform stochastic modelling. Properties are expressed in the form of Continuous Stochastic Logic (CSL) [@kwiatkowska2007stochastic], a stochastic variant of the well-known Computational Tree Logic (CTL) [@Clarke86automaticverification] which includes reward formulae. [Note, a system can be modelled using multiple CTMCs which represent different sub-components within the whole. Transition labels are then used to synchronise the individual CTMCs representing different parts of a system and in turn obtain the full CTMC representing the whole system.]{}
\[def:CTMC\] [The tuple $C =(S, s_0, {\text{TL}}, {\text{AP}}, L, {\textbf{R}})$ defines a CTMC which is composed of a set of states $S$, the initial state $s_0$, a finite set of transition labels ${\text{TL}}$, a finite set of atomic propositions ${\text{AP}}$, a labelling function $L: S\rightarrow 2^{AP}$ and the transition rate matrix $ {\textbf{R}}: S \times S \rightarrow {\textbf{R}}_{\geq 0}$. ]{}The rate ${\textbf{R}}(s,s')$ defines the delay before which a transition between states $s$ and $s'$ takes place. If ${\textbf{R}}(s,s') \ne 0$ then the probability that a transition between the states $s$ and $s'$ is defined as $1 - e^{-{{{\textbf{R}}}}(s,s') t} $ where $t$ is time. No transitions will trigger if ${\textbf{R}}(s,s') = 0$.
[The logic of CSL specifies state-based properties for CTMCs, built out of propositional logic (with atoms $a \in {\text{AP}}$) , a steady-state operator (${\mathrm{S}}$) that refers to the stationary probabilities, and a probabilistic operator (${\mathrm{P}}$) for reasoning about transient state probabilities. The state formulas are interpreted over states of a CTMC, whereas the path formulas are interpreted over paths in a CTMC. The syntax of CSL is: $$\begin{aligned}
\Phi &::= \texttt{true}\;|\;a\;|\;\Phi \wedge \Phi\;|\; \neg \Phi \;|\; {\mathrm{S}}_{\sim p}[\Phi] \;|\;{\mathrm{P}}_{\sim p}[\phi]\\
\psi &::= \mathrm{X}\; \phi \;|\;\Phi\; \mathrm{U}^{\ge T} \;\Phi
\end{aligned}$$ where $\sim \in \{<,\le, =,\ge,>\}$, $p \in [0,1]$, $T \in {\mathbb{R}}_{\ge 0}$ is the time horizon, $\mathrm{X}$ is the next operator and $\mathrm{U}$ is the until operator. The semantics of CSL formulas is given in [@kwiatkowska2007stochastic].]{} $S_{\sim p}[\Phi]$ asserts that the steady-state probability for a $\Phi$-state meets the bound $\sim p$ whereas $P_{\sim p}[\Phi \;\mathrm{U}^{\le t} \;\Phi]$ asserts that with probability $\sim p$, by the time $t$ a state satisfying $\Phi$ will be reached such that all preceding states satisfy $\Phi$.[ Additional properties can be specified by adding the notion of rewards. ]{}The extended CSL logic adds reward operators, a subset of which are [@kwiatkowska2007stochastic]: $$\begin{aligned}
{\mathrm{R}}_{\sim r} [{\mathrm{C}}^{\le T}] \;|\; {\mathrm{R}}_{\sim r} [{\mathrm{F}}\;\Phi] \end{aligned}$$ where $r,t \in {\mathbb{R}}_{\le 0}$ and $\Phi$ is a CSL formula. A state $s$ satisfies ${\mathrm{R}}_{\sim r} [{\mathrm{C}}^{\le T}]$ if, from state s, the expected reward cumulated up until $T$ time units have elapsed satisfies $\sim r$ and ${\mathrm{R}}_{\sim r} [{\mathrm{F}}\; \Phi] $ is true if, from state $s$, the expected reward cumulated before a state satisfying $\Phi$ is reached meets the bound $\sim r$. Examples of a CSL property with its natural language translation are: (i) $ {\mathrm{P}}_{\geq 0.95} [{\mathrm{F}}~complete]$ - “The probability of the system eventually completing its execution successfully is at least 0.95". Each state (and/or transition) of the model is assigned a real-valued reward, allowing queries such as: (ii) ${\mathrm{R}}_{=?} [{\mathrm{F}}~success] $ - “What is the expected reward accumulated before the system successfully terminates?" Rewards can be used to specify a wide range of measures of interest, for example, the total operational costs and the total percentage of time [during which]{} the system is available.
Formalizing FMTs using CTMCs {#subsec:FMTFrame:Modelling}
============================
FMT Syntax {#subsubsec:FMTFrame:Modelling:Syntax}
----------
To formalise the syntax of FMTs using CTMCs, we first define the set [$\mathcal{F}$]{}, characterizing each FMT element by [type, inputs and rates]{}. We introduce a new element called DELAY which will be used to model the deterministic time delays required by the extended basic events (EBE), repair module (RM) and inspection module (IM). We restrict the set [$\mathcal{F}$]{}to contain the EBE, RDEP gate, OR gate, DELAY, RM and IM modules since these will be the components used in the case study presented in Section \[sec:CaseStudy\].
\[def:FMTSem\] The set [$\mathcal{F}$]{}$=\{EBE, RDEP, OR, DELAY,RM, IM\}$ of FMT elements consists of the following tuples. Here, $n, N \in {\mathbb{N}}$ are natural numbers, $\mathit{thresh}, in, \mathit{trig} \in \{0,1\}$ take binary values, [${T_{cln}}$, ${T_{rplc}},$ ${T_{rep}},{T_{oh}}$, ${T_{insp}}\in {\mathbb{R}}_{\ge 0}$ are deterministic delays, ${T_{deg}}\in {\mathbb{R}}_{\ge 0}$ is a rate and $\gamma \in {\mathbb{R}}_{\ge 0}$ is a factor]{}.
- $(EBE,{T_{deg}},{T_{cln}},{T_{rplc}},N)$ represent the extended basic events with $N$ discrete degradation levels, each of which degrade with a time delay equal to $T_{deg}$. It also takes as inputs the time taken to restore the EBE to the previous degradation level ${T_{cln}}$ when cleaning is performed and the time taken to restore the EBE to its initial state ${T_{rplc}}$ following a replacement action.
- $(RDEP,n,\gamma,in, {T_{deg}})$ represents the RDEP gate with $n$ dependent children, acceleration factor $\gamma$, the input $in$ which activates the gate and ${T_{deg}}$ the degradation rate of the dependent children.
- $(OR,n)$ represents the OR gate with $n$ inputs. [When either one of the inputs reaches the state labelled with $failed$, the OR gate returns a true signal.]{}
- $(RM,n,{T_{rep}},{T_{oh}},{T_{insp}},{T_{cln}}, {T_{rplc}},{\mathit{thresh}}, \mathit{trig})$ represents the RM module which acts on $n$ EBEs (in our case, this corresponds to all the EBEs in the FMT). The RM can either be triggered periodically to perform a cleaning action, every ${T_{rep}}$ delay, or a replacement action, every ${T_{oh}}$ delay, or by the IM when the delay ${T_{insp}}$ has elapsed and the $\mathit{thresh}$ condition is met. The time to perform a cleaning action is ${T_{cln}}$, while the time taken to perform a replacement is ${T_{rplc}}$. The $\mathit{trig}$ signal ensures that when the component is not in the degraded states, no unnecessary maintenance actions are carried out.
- $(IM,n,{T_{insp}},{T_{cln}},{T_{rplc}},\mathit{thresh})$ represents the IM module which acts on $n$ EBEs (in our case, this corresponds to all the EBEs in the FMT). The IM initiates a repair depending on the current state of the EBE. Inspections are performed in a periodic manner, every ${T_{insp}}$. If during an inspection, the current state of the EBE does not correspond to the *new* or *failed* state (i.e. the degradation level of the inspected EBE is below a certain threshold), the $thresh$ signal is activated and is sent to the RM. Once a cleaning action is performed the IM moves back to the initial state with a delay equal to ${T_{cln}}$ or ${T_{rplc}}$ depending on the maintenance action performed.
- $(DELAY, T, N)$ represents the DELAY module which takes two inputs representing the deterministic delay $T \in \{{T_{deg}},$ ${T_{cln}},{T_{rplc}},{T_{rep}},{T_{oh}},{T_{insp}}\}$ to be approximated using an Erlang distribution with $N$ states. This DELAY module can be extended by inclusion of a reset transition label, which when triggered restarts the approximation of the deterministic delay before it has elapsed. The extended DELAY module is referred to as $(DELAY, T, N)_{ext}$.
The FMT is defined as a special type of directed acyclic graph $G=(V,E)$ where the vertices $V$ represent the gates and the events which represent an occurrence within the system, typically the failure of a subsystem down to an individual component level, and the edges $E$ which represent the connections between vertices. The vertices $V$ are labelled instances of elements in [$\mathcal{F}$]{} i.e. $V$ may contain multiple elements of the same component obtained from the set [$\mathcal{F}$]{}which are identified by their common element label. Events can either represent the EBEs or *intermediate* events which are caused by one or more other events. The event at the top of the FMT is the top event (TE) and corresponds to the event being analysed - modelling the failure of the (sub)system under consideration. The EBE are the leaves of the DAG. For $G$ to be a well-formed FMT, we take the following assumptions (i) vertices are composed of the OR, RDEP gates, (ii) there is only one top event, (iii) RDEP can only be triggered by EBEs and (iv) RM and IM are not part of the DAG tree but are modelled separately This DAG formulation allows us to propose a framework in Subsection \[subsec:FMTFrame:Dec\] such that we can efficiently perform probabilistic model checking.
\[def:FMT\] A fault maintenance tree is a directed acyclic graph $G = (V,E)$ composed of vertices $V$ and edges $E$.
Semantics of FMT elements {#subsubsec:FMTFrame:Modelling:SemanticsElements}
--------------------------
Next, we provide the semantics for each FMT element, which are composed using the syntax of CTMC (cf. Definition \[def:CTMC\]). These elements are then instantiated based on the underlying FMT structure to form the semantics of the whole FMT.[ We obtain the semantics of the whole FMT via synchronisation of transition labels between the different CTMCs representing the individual FMT elements. This is further explained in subsection \[subsubsec:FMTFrame:Modelling:Semantics\]]{}.
#### **DELAY** {#par:FMTFrame:Modelling:Clck}
[We define the semantics for the $(DELAY,T,N)$ element using Figure \[fig:ClkNoRst\] and describe the corresponding CTMC using the set of states given by $D =\{d_0,d_1,\dots, d_{N+1}\}$, the initial state $d_0$, the set of transitions labels ${\text{TL}}=\{\texttt{trigger}, \texttt{move}\}$, the set of atomic propositions ${\text{AP}}=\{T\}$ with $L(d_0) = \dots =L(d_{N}) = \emptyset$, and $L(d_{N+1}) = \{T\}$. The rate matrix ${\textbf{R}}$ becomes clear from Figure \[fig:ClkNoRst\] and $$\begin{aligned}
{\textbf{R}}_{ij} = \begin{cases}
\mu & i= 0 \wedge j =1,\\
\frac{N}{T} & ((i \ge 1 \vee i < N+1) \wedge j = i+1) \vee (i=N+1 \wedge j=1), \\
0 & \text{otherwise},
\end{cases}
\end{aligned}$$ with $i$ representing the current state, $j$ is the next state and $\mu$ is a fixed large value corresponding to introducing a negligible delay, which is used to trigger all the DELAY modules at the same time (cf. Definition \[def:CTMC\]). ]{} In Figure \[fig:ClkRst\] we define the semantics of $(DELAY,T,N)_{ext}$. This results in the CTMC described using the state space $D =\{d_0,d_1,\dots, d_{N+1}\}$, the initial state $d_0$, the set of transition labels ${\text{TL}}=\{\texttt{trigger},$ $ \texttt{move}, \texttt{reset}\}$, the set of atomic propositions ${\text{AP}}=\{T\}$, the labelling function $L(d_0) = L(d_1) = \dots =L(d_{N}) = \emptyset$, and $L(d_{N+1}) = \{T\}$ and the rate matrix ${\textbf{R}}$ where $$\begin{aligned}
{\textbf{R}}_{ij} = \begin{cases}
\mu & i= 0 \wedge j =1,\\
1 & (i \ge 2 \vee i < N+1) \wedge j = 1,\\
\frac{N}{T} & ((i \ge 1 \vee i < N+1) \wedge j = i+1) \vee (i=N+1 \wedge j=1), \\
0 & \text{otherwise},
\end{cases}
\end{aligned}$$ with $i$ representing the current state and $j$ is the next state. In both instances, the deterministic delays is approximated using an Erlang distribution [@hoque2015towards] and all DELAY modules are synchronised to start together using the `trigger` transition label. The extended DELAY module have the transition labels `reset` which restarts the Erlang distribution approximation whenever the guard condition is met at a rate of $1 \times {\textbf{R}}_{sync}$ where ${\textbf{R}}_{sync}$ is the rate coming from the use of synchronisation with other modules causing the reset to occur ( as explained in Subsection \[subsubsec:FMTFrame:Modelling:Semantics\]). This is required when a maintenance action is performed which restores the EBE’s state back to the original state and thus restart the degradation process, before the degradation time has elapsed.
\[remark:Er\] A random variable $Z \in {\mathbb{R}}_{+}$ has an Erlang distribution with $k \in {\mathbb{N}}$ stages and a rate $\lambda \in {\mathbb{R}}_{+}, Z \sim Erlang(k,\lambda)$, if $Z = Y_1 + Y_2 + \dots Y_k$ where each $Y_i$ is exponentially distributed with rate $\lambda$. The cumulative density function of the Erlang distribution is characterised using $$\label{eqn:cdf}
f(t;k,\lambda) = 1- \sum_{n=0}^{k-1}\frac{1}{n!} \exp(-\lambda t)(\lambda t)^n \quad \textrm{for } t,\lambda \ge 0,$$ and for $k=1$, the Erlang distribution simplifies to the exponential distribution. In particular, the sequence $Z_k \sim Erlang(k,\lambda k)$ converges to the deterministic value $\frac{1}{\lambda}$ for large $k$. Thus, we can approximate a deterministic delay $T$ with a random variable $Z_k \sim Erlang(k,\frac{k}{T})$ [@bortolussi2012fluid; @hoque2014probabilistic]. Note, there is a trade-off between the accuracy and the resulting blow-up in size of the CTMC model for larger values of $k$ (a factor of $~k$ increase in the model size) [@hoque2015towards]. In this work, the Erlang distribution will be used to model the fixed degradation rates, the maintenance and inspection signals. This is a similar approach taken in [@ruijters2016fault] where degradation phases are approximated by an (k,$\lambda$)-Erlang distribution.
\[fig:Clk\]
#### **Extended Basic Events (EBE)** {#par:FMTFrame:Modelling:EBE}
The EBE are the leaves of the FMT and incorporate the component’s degradation model. EBE are a function of the total number of degradation steps $N$ considered. Figure \[fig:EBE\] shows the semantics of the $(EBE,T_{deg}, T_{cln},T_{rep},N=3)$. The corresponding CTMC [is described by the tuple $(\{s_0,s_1,s_2,s_3\},$ $ s_0,{\text{TL}}_{EBE},$ ${\text{AP}}_{EBE}, L_{EBE}, {\textbf{R}}_{EBE})$ where $s_0$ is the initial state , $$\begin{aligned}
{\text{TL}}_{EBE} &=\{ \texttt{degrade}_{i\in\{0,\dots,N\}}, \texttt{perform\_clean},\texttt{perform\_replace}\},
\end{aligned}$$ the atomic propositions ${\text{AP}}_{EBE} = \{\mathit{new},\mathit{thresh},$ $ \mathit{failed}\}$, the labelling function $L(s_0) = \{new\},$ $ L(s_1) = L(s_2) = \{thresh\}, L(s_3) = \{failed\}$ ]{}and $$\begin{aligned}
{\textbf{R}}_{EBE}= \left[\begin{matrix}
0&1&0&0\\
1&0&1&0\\
1&1&0&1\\
1&0&1&0\\
\end{matrix}\right].
\end{aligned}$$
The deterministic time delays taken as inputs are modelled using three different DELAY modules:
1. an extended DELAY module approximating ${T_{deg}}$ with the transition label `move` replaced with `degrade`$_N $such that synchronisation between the two CTMCs is performed (explained in Subsection \[subsubsec:FMTFrame:Modelling:Semantics\]). When ${T_{deg}}$ has elapsed the transition labelled with `degrade`$_N$ is triggered and the EBE moves to the next state at a rate [^2] equal to $\frac{N}{{T_{deg}}} \times 1$. The `reset` transition label and the corresponding transitions are replicated in extended DELAY module and replaced with `perform_clean` and `perform_replace`. When the the previous state (if cleaning action is carried out) or to the initial state (if replace action is performed).
2. a DELAY module approximating ${T_{cln}}$ with the transition label `move` replaced with `perform_clean`. When ${T_{cln}}$ has elapsed the transition with transition label `perform_clean` is triggered and the EBE moves to the previous state at a rate equal to $\frac{N}{{T_{cln}}}$.
3. a DELAY module approximating ${T_{rplc}}$ with the transition label `move` replaced with `perform_replace`. When ${T_{rplc}}$ has elapsed the transition label `perform_replace` is triggered and the EBE moves to the initial state at a rate equal to $\frac{N}{{T_{rplc}}}$.
The transition labels `perform_clean` and `perform_replace` cannot be triggered at the same time and it is assumed that ${T_{cln}}\neq {T_{rplc}}$. This is a realistic assumption as only one maintenance action is performed at the same time.
#### **RDEP gate** {#par:FMTFrame:Modelling:RDEP}
The RDEP gate has static semantics and is used in combination with the semantics of its $n$ dependent EBEs. When triggered ($input=1$), the associated EBE reaches the state labelled $\mathit{failed}$, the degradation rate of the $n$ dependent children is accelerated by a factor $\gamma$. We model the $input$ signal using $$\begin{aligned}
\label{eqn:in}
input &= \begin{cases}
1 & L(s) = \mathit{failed},\\
0 & \text{otherwise},\\
\end{cases}
\end{aligned}$$ where $L(s)$ is the label of the current state of the associated EBE (cf. Figure \[fig:RDEPTiming\]). [ Similarly, we map the RDEP gate function using $$\begin{aligned}
\label{eqn:rdep}
RA &= \begin{cases}
\gamma T_{deg_1},\dots, \gamma T_{deg_n}& input= 1,\\
T_{deg_1},\dots, T_{deg_n} & \text{otherwise},\\
\end{cases}
\end{aligned}$$ where $T_{deg_i}, i \in {1, \dots n}$ corresponds to the degradation rate of the $n$ dependent children.]{} [^3]
#### **OR gate** {#par:FMTFrame:Modelling:OR}
The OR gate indicates a failure when either of its input nodes have failed and also does not have semantics itself but is used in combination with the semantics of its $n$ dependent input events (EBEs or intermediate events). We use [$$\begin{aligned}
\label{eqn:OR}
FAIL &= \begin{cases}
0 & E_{1} = 1 \wedge \dots \wedge E_{n} = 1, \\
1 & \text{otherwise},\\
\end{cases}
\end{aligned}$$]{} where $E_{i} = 1, i \in 1 \dots n$ corresponds to when the $n$ events (cf. Def. \[def:event\]), connected to the OR gate, represent a failure in the system. In the case of EBEs, $E_1 =1$ occurs when the EBE reaches the $\mathit{failed}$ state .
#### **Repair module (RM)** {#par:FMTFrame:Modelling:RM}
Figure \[fig:RM\] shows the semantics of $(RM,n,$ ${T_{rep}}, {T_{oh}}, {T_{insp}},$ $ {T_{cln}},$ ${T_{rplc}},$ $\mathit{thresh},\mathit{trig})$. The CTMC is described using [ the state space $\{rm_0,rm_1\}$, the initial state $rm_0$, the transition labels $$\begin{aligned}
{\text{TL}}_{RM} = \{\texttt{inspect}, \texttt{check\_clean}, \texttt{check\_replace},\texttt{trigger\_clean}, \texttt{trigger\_replace} \},
\end{aligned}$$ the atomic propositions ${\text{AP}}=\{maintenance\}$, the labelling function $L(rm_{0})=\{ \emptyset \},$ $ L(rm_1) = \{ \mathit{maintenance} \}$ and with ]{} $$\begin{aligned}
{\textbf{R}}_{IM} = \left[\begin{matrix}
1&1\\
1&0\\
\end{matrix}\right].
\end{aligned}$$ [For brevity in Figure \[fig:RM\], we used the transition labels `check_maintenance` and `trigger_maintenance`. The transition label `check_maintenance` and corresponding transitions are replicated and the transition labels replaced by `check_clean` or `check_replace` to allow for both type of maintenance checks. Similarly, the transition label `trigger_maintenance` and corresponding transitions are duplicated and the transition labels replaced by `trigger_clean` or `trigger_replace` to allow the initiation of both type of maintenance actions to be performed. Due to synchronisation, only one of the transitions may trigger at any time instance (as explained in Subsection \[subsubsec:FMTFrame:Modelling:Semantics\]). The transition labels `trigger_clean` or `trigger_replace` correspond to the transition label `trigger` within the DELAY module approximating the deterministic delays ${T_{cln}}$ and ${T_{rplc}}$ respectively. The deterministic delays which trigger `inspect`, `check_clean` or `check_replace` correspond to when the time delays ${T_{insp}}, {T_{rep}}$ and ${T_{oh}}$ respectively, have elapsed. All these signals are generated using individual DELAY modules with the `move` transition label for each module replaced using `inspect`, `check_clean` or `check_replace` respectively. ]{} The $\mathit{thresh}$ signal is modelled using $$\begin{aligned}
\label{eqn:thresh}
\mathit{thresh} &= \begin{cases}
1 & L(s_{j,1}) = \mathit{thresh} \vee \dots \vee L(s_{j,n}) = \mathit{thresh},\\
0 & \text{otherwise},\\
\end{cases}
\end{aligned}$$ where $L(s_{j,i}), j \in 0 \dots N, i \in 1 \dots n$ correspond to the label of the current state $j$ of each of the $n$ EBE. Similarly, we model the $\mathit{trig}$ signal using $$\begin{aligned}
\label{eqn:trig}
\mathit{trig} &= \begin{cases}
1 & L(s_{j,1}) \neq \mathit{new} \vee \dots \vee L(s_{j,n}) \neq \mathit{new} ,\\
0 & \text{otherwise}.\\
\end{cases}
\end{aligned}$$ Both signals act as guards which when triggered determine which transition to perform (cf. Fig. \[fig:RM\]).
#### **Inspection module (IM)** {#par:FMTFrame:Modelling:IM}
The semantics of the $(IM,n,{T_{insp}},$ $ {T_{cln}},{T_{rplc}},$ $\mathit{thresh})$ is depicted in Figure \[fig:IM\].[ The CTMC is defined using the tuple $(\{im_0,im_1\},im_0,{\text{TL}}_{IM}, {\text{AP}}_{IM}, L_{IM}, {\textbf{R}}_{IM})$. Here, $$\begin{aligned}
{\text{TL}}_{IM} = \{\texttt{inspect}, \texttt{perform\_clean},\texttt{perform\_replace} \},
\end{aligned}$$ $ {\text{AP}}_{IM}= \{\emptyset\},$ with $ L(s_0) =L(s_1) = \emptyset$ and $$\begin{aligned}
{\textbf{R}}_{IM} = \left[\begin{matrix}
1&1\\
1&0\\
\end{matrix}\right].\end{aligned}$$]{}The $\mathit{thresh}$ signal corresponds to same signal used by the RM, given using . In Figure \[fig:IM\], for clarity, we use the transition label `perform_maintenance`. This transition label and corresponding transitions are duplicated and the transition labels are replaced by either `perform_clean` or `perform_replace` to allow for both type of maintenance actions to be performed when one of them is triggered using synchronisation. The same DELAY modules used in the RM and EBE to represent the deterministic delays are used by the IM. The DELAY module used to represent the deterministic delays ${T_{cln}}$ and ${T_{rplc}}$ triggers the transition labels `perform_clean` or `perform_replace`. This represents that the maintenance action has completed.
[ Semantics of composed FMT]{} {#subsubsec:FMTFrame:Modelling:Semantics}
------------------------------
[Next, we show how to obtain the semantics of a FMT from the semantics of its elements using the FMT syntax introduced in Subsection \[subsubsec:FMTFrame:Modelling:Syntax\]. We define the DAG $G$ by defining the vertices $V$ and the corresponding events $E$. The leaves of the DAG are the events corresponding to the EBE. The events $E$ are connected to the vertices $V$, which trigger the corresponding auxiliary function used to represent the semantics of the gates. The $Events$ connected to the RM and IM are initiated by triggering the auxiliary functions $\mathit{thresh}$ and $\mathit{trig}$ given using and respectively. Based on the structure of $G$, we compute the corresponding CTMC by applying parallel composition of the individual CTMCs representing the elements of the FMT. The parallel composition formulae are derived from [@hermanns2011concurrency] and defined as follows,]{}
\[def:IntSynch\] The interleaving synchronous product of $C_1 = (S_1,s_{01},$ $ {\text{TL}}_{1},$ $ {\text{AP}}_1, L_1,{\textbf{R}}_1)$ and $C_2 =(S_2,s_{02}, {\text{TL}}_{2}, {\text{AP}}_2, L_2,{\textbf{R}}_2)$ is $C_1 ||C_2 = (S_1 \times S_2, (s_{01} , s_{02}),{\text{TL}}_1 \cup {\text{TL}}_2, {\text{AP}}_1 \cup {\text{AP}}_2, L_1 \cup L_2,{\textbf{R}})$ where ${\textbf{R}}$ is given by: [$$\begin{aligned}
\frac{s_1 \xrightarrow{\alpha_1, \lambda_1} s_1'}{(s_1,s_2) \xrightarrow{\alpha_1,\lambda_1} (s_1',s_2)},~\text{and}~
\frac{s_2 \xrightarrow{\alpha_2, \lambda_2} s_2'}{(s_1,s_2) \xrightarrow{\alpha_2,\lambda_2} (s_1,s_2')},
\end{aligned}$$ ]{} and $s_1,s_1' \in S_1$, $\alpha_1 \in {\text{TL}}_1$, ${\textbf{R}}_1(s_1,s_1') = \lambda_1$, $s_2,s_2' \in S_2$, $\alpha_2 \in {\text{TL}}_2$, ${\textbf{R}}_2(s_2,s_2') = \lambda_2$.
\[def:FullSynch\] The full synchronous product of $C_1 = (S_1,s_{01}, {\text{TL}}_{1}, {\text{AP}}_1, L_1,{\textbf{R}}_1)$ and $C_2 =(S_2,s_{02}, {\text{TL}}_{2}, {\text{AP}}_2, L_2,{\textbf{R}}_2)$ is $C_1 ||C_2 = (S_1 \times S_2, (s_{01} , s_{02}),{\text{TL}}_1 \cup {\text{TL}}_2, {\text{AP}}_1 \cup {\text{AP}}_2, L_1 \cup L_2,{\textbf{R}})$ where ${\textbf{R}}$ is given by: [$$\begin{aligned}
\frac{s_1 \xrightarrow{\alpha, \lambda_1} s_1' ~ \text{and}~ s_2 \xrightarrow{\alpha, \lambda_2} s_2' }{(s_1,s_2) \xrightarrow{\alpha,\lambda_1\times \lambda_2} (s_1',s_2')},
\end{aligned}$$ ]{} and $s_1,s_1' \in S_1$, $\alpha \in {\text{TL}}_1 \wedge {\text{TL}}_2$, ${\textbf{R}}_1(s_1,s_1') = \lambda_1$, $s_2,s_2' \in S_2$, $\alpha_2 \in {\text{TL}}_2$, ${\textbf{R}}_2(s_2,s_2') = \lambda_2$.
For any pair of states, synchronisation is performed either using interleaving or full synchronisation. For full synchronisation, as in Definition \[def:IntSynch\], the rate of a synchronous transition is defined as the product of the rates for each transition. The intended rate is specified in one transition and the rate of other transition(s) is [specified as one]{}. [For instance, the RM synchronises using full synchronisation with the DELAY modules representing ${T_{insp}}$, ${T_{rep}}$ and ${T_{rplc}}$ and therefore, to perform synchronisation between the RM and the DELAY modules, the rates of all the transitions of RM should have a value of one (cf. Fig. \[fig:RM\]), while the rate of the DELAY modules represent the actual rates (cf. Fig \[fig:ClkNoRst\] and Fig \[fig:ClkRst\]). The same principle holds for the EBEs and the IM. We refer the reader to Table \[tab:synch\] to further understand the synchronisation between the FMT components and the method employed for parallel composition.]{}
Consider a simple example showing the time signals and synchronisations required for modelling an EBE and the RM and IM. The EBE has a degradation rate equal to ${T_{deg}}$ and we limit the functionality of the RM and IM by allowing only the maintenance action to perform cleaning. We also need the corresponding DELAY modules generating the degradation rates, ${T_{deg}}$ and the maintenance rates ${T_{cln}},{T_{insp}},{T_{rep}}$. The resulting CTMC is obtained by performing a parallel composition of the components $C_{all} = C_{EBE}||$ $C_{{T_{deg}}}||C_{RM}||C_{IM}||C_{{T_{cln}}}$ $||C_{{T_{insp}}}||C_{{T_{rep}}}.$ The resulting state space is then $S_{all}= S_{EBE} \times S_{{T_{deg}}} \times S_{RM} \times S_{IM} \times S_{{T_{cln}}} \times S_{{T_{insp}}} \times S_{{T_{rep}}} $. The synchronisation between the different components is shown in Figure \[fig:Synch\] and proceeds as follows:
1. All the DELAY modules (except ${T_{cln}}$) start at the same time using the `trigger` transition label.
2. When the extended DELAY module generating the ${T_{deg}}$ time delay elapses, the corresponding EBE moves to the next state through synchronisation with the transition label `degrade`$_N$.
3. The clock signals ${T_{rep}},{T_{insp}}$ represent periodic maintenance and inspection actions and when the deterministic delay is reached, through synchronisation with the transition label `check_clean` or the $\texttt{inspect}$, the RM or IM modules are triggered (cf. Fig. \[fig:RM\] and \[fig:IM\]). If RM triggers a maintenance action, the DELAY representing ${T_{cln}}$ is triggered using the synchronisation labels `trigger_clean`. Once the deterministic delay ${T_{cln}}$ elapses, the EBE, the extended DELAY module representing ${T_{deg}}$ (where the `reset` transition label within the extended DELAY module is replaced with `perform_clean`) and the IM are reset using the transition label `perform_clean`.
![[Block diagram showing the synchronisation connections between one component and the other, together with the corresponding transition label which triggers synchronisation.]{}[]{data-label="fig:Synch"}](cs20.png){width=".5\textwidth"}
\[remark:ss\] One should note that performing synchronisation results in a large state space, which is a function of the number of states used to approximate the deterministic delays. In order to counteract this effect we propose an abstraction framework in Subsection \[subsec:FMTFrame:Dec\].
Metrics {#subsec:FMTFrame:Met}
-------
We use PRISM to compute the metrics of the model described in Subsection \[subsec:Prelim:FMTFrame\]. The metrics can be expressed using the extended Continuous Stochastic Logic (CSL) as follows:
1. *Reliability* : This can be expressed as the complement of the probability of failure over the time $T$, $1-{\mathrm{P}}_{=?}[{\mathrm{F}}^{\le T} failed ]$.
2. *Availability*: This can be expressed as ${\mathrm{R}}_{=?}[C^{\le T} ]/T$, which corresponds to the cumulative reward of the total time spent in states labelled with *okay* and *thresh* during the time $T$.
3. *Expected cost*: This can be expressed using ${\mathrm{R}}_{=?}[C^{\le T}]$, which corresponds to the cumulative reward of the total costs (operational, maintenance and failure) within the time $T$.
4. *Expected number of failure*: This can be expressed using ${\mathrm{R}}_{=?}[C^{\le T} ]$, which corresponds to the cumulative transition reward that counts the number of times the top event enters the *failed state* within the time $T$.
Decomposition of FMTs {#subsec:FMTFrame:Dec}
---------------------
The use of CTMC and deterministic time delays results in a large state space for modelling the whole FMT (cf. Remark \[remark:ss\]). We therefore propose an approach that decomposes the large FMT into an equivalent abstract CTMC that can be analysed using PRISM. The process involves two transformation steps. First we convert the FMT into an equivalent directed acyclic graph (DAG) and split this graph into a set of smaller sub-graphs. Second, we transform each sub-graph into an equivalent CTMC by making use of the developed FMT components semantics (cf. Subsec. \[subsubsec:FMTFrame:Modelling:SemanticsElements\]), and performing parallel composition of the individual FMT components based on the underlying structure of the sub-graph. The smaller sub-graphs are then sequentially composed to generate the higher level abstract FMT. Figure \[fig:Overall\] depicts a high-level diagram of the decomposition procedure.
{width="\textwidth"}
#### **Conversion of the original FMT to the equivalent graph**
The FMT is a DAG (cf. Subsection \[subsec:FMTFrame:Modelling\]) and in this framework we need to apply a transformation to the DAG in the presence of an RDEP gate, such that we can perform the decomposition. The RDEP causes an acceleration of events on dependent children nodes when the input node fails. In order to capture this feature in a DAG, we need to duplicate the input node such that it is connected directly to the RDEP vertex. This allows us to capture when the failure of the input occurs and the corresponding acceleration of the the children. This is reasonable as the same RM and IM are used irrespective of the underlying FMT structure.
#### **Graph decomposition**
We define modules within the DAG as sub-trees composed of at least two events which have no inputs from the rest of the tree and no outputs to the rest except from its output event [@Li20151400]. We can divide the graph into multiple partitions based on the number of modules making up the DAG. We define the following notations to ease the description of the algorithm:
- $V_o$ indicates whether the node is the top node of the DAG.
- $V_g$ indicates the node where the graph split is performed.
- Modules correspond to sub-graphs in DAG.
We set $V_o$ when we construct the DAG from the FMT and then proceed with executing Algorithm \[alg:split\]. We first identify all the sub-graphs within the whole DAG and label all the top nodes of each sub-graph $i$ as $V_{Ti}$. We loop through each sub-graph and its immediate child (the sub-graph at the immediate lower level) and at the point where the sub-graph and child are connected, the two graphs are split and a new node $V_g$ is introduced. Thus, executing Algorithm \[alg:split\] results in a set of sub-graphs linked together by the labelled nodes $V_g$. For each of the lower-level sub-graphs, we now proceed to compute the mean time to failure (MTTF). This will serve as an input to the higher-level sub-graphs, such that metrics for the abstract equivalent CTMC can be computed.
\[alg:ADP\_Nat1\] Identify sub-graphs using ‘depth-first’ traversal Label all top nodes of each sub-graph $i$ as $V_{T_i}$
#### **PMC of sub-graphs**
We start from the bottom level sub-graphs and perform the conversion to CTMC using the formal models presented in Subsection \[subsubsec:FMTFrame:Modelling:SemanticsElements\]. The formal models have been built into a library of PRISM modules and based on the underlying components and structure making up the sub-graph, the corresponding individual formal models are converted into the sub-graph’s equivalent CTMC by performing parallel composition (cf. Subsec. \[subsubsec:FMTFrame:Modelling:Semantics\]). For each sub-graph, we compute the probability of failure $D_e(T)$ at time $T$ , from which we calculate the MTTF [@ruijters2015fault] using $$\mathit{MTTF} = \frac{\ln(1-D_e(T))}{-T}.$$ The MTTF serves as the input to the higher level sub-graph at time $T$. The new node in the higher-level sub-graph, now degrades with the new time delay ${T_{deg}}= \mathit{MTTF}$, which is fed into the corresponding DELAY component. This process is repeated for all the different sub-graphs until the top level node $V_o$ is reached. [Figure \[fig:conv\] depicts the steps needed to perform PMC for one of the sub-graphs. ]{}
![PMC of sub-graphs.[]{data-label="fig:conv"}](SubgraphsPMC){width="\textwidth"}
#### **PMC of final equivalent abstract CTMC**
On reaching the top level node $V_o$, we compute the metrics for the equivalent abstract CTMC for a specific time horizon $T$. For different horizons, the previous step of computing the MTTF for the underlying lower level sub-graphs needs to be repeated. Using this technique, we can formally verify larger FMTs, while using less memory and computational time due to the significantly smaller state space of the underlying CTMCs.
![[The original FMT and the abstract FMT corresponding to the equivalent abstract CTMC generated by the developed framework. The MTTF for the F’ is computed based on the probability of failure of the heating coil.]{}[]{data-label="fig:IncVer"}](IncVer.pdf){width="60.00000%"}
Next, we proceed with an illustrative example comparing the process of directly modelling the large FMT using CTMCs versus the de-compositional modelling procedure. Figure \[fig:IncVer\] presents the FMT composed of two modules and the corresponding abstracted FMT. The abstract FMT is a pictorial representation of the model represented by the equivalent abstract CTMC obtained using the developed decomposition framework (cf. Fig. \[fig:Overall\]). For both the large FMT and the equivalent abstract FMT a comparison between the total number of states for the resulting CTMC models, the total time to compute the reliability metric and the resulting reliability metric is performed. All computations are run on an 2.3 GHz Intel Core i5 processor with 8 GB of RAM and the resulting statistics are listed in Table \[tab:HVAC\_SizeEg\]. The original FMT has a state space with 193543 states, while the equivalent abstract CTMC has a state space with 63937 states. This corresponds to a $67\%$ reduction in the state space size. The total time to compute the reliability metric is a function of the final time horizon and a maximal $73\%$ reduction in computation time is achieved. Accuracy in the reliability metric of the abstract model is a function of the time horizon and the number of states used to approximate the deterministic delay representing the computed MTTF. The larger the number of states the more accurate the representation of the MTTF, but this comes at a cost on the size of the underlying CTMC model. In our case, $N=4$ is chosen. The accuracy of the reliability metric computed by the abstract FMT results in a maximal reduction of $0.61\%$.
Case study {#sec:CaseStudy}
==========
We apply the FMT framework to a Heating, Ventilation and Air-conditioning (HVAC) system used to regulate a building’s internal environment (cf. Sec. \[sec:hvac\]). Based on this HVAC system we construct the corresponding FMT shown in Figure \[fig:FMT\]. The FMT structure follows the structure of the underlying HVAC system, as can be seen from the colour shading used in Figure \[fig:FMT\]. The leaves of the tree are EBE with discrete degradation rates computed using Table \[tab:HVAC\_FM\], approximated by the Erlang distribution where $N$ is the number of degradation phases ($k=N$ for the Erlang distribution) and MTTF is the expected time to failure with $MTTF = 1/\lambda$ (cf. Remark \[remark:Er\]). We choose an acceleration factor $\gamma = 2$ for the RDEP gate. The system is periodically cleaned every ${T_{rep}}$ months and a major overhaul with a complete replacement of all components is carried out once every ${T_{oh}}$ years. Inspections are performed every ${T_{insp}}$ months and return the components back to the previous state, corresponding to a cleaning action. The total time to perform a cleaning action is 1 day (${T_{cln}}= 1\;day$), while performing a total replacement of components takes 7 days (${T_{rplc}}= 7\;days$). The time timing signals $\{{T_{rep}}, {T_{oh}},{T_{insp}},{T_{cln}},{T_{rplc}}\}$ are all approximated using the Erlang distribution with $N = 3$. All maintenance actions are performed simultaneously on all components.
![[FMT for failure in HVAC system with leaves represented using EBE (associated RM and IM not shown in figure). The EBE are labelled to correspond to the component failure they represent using the fault index presented in Table \[tab:HVAC\_FM\]. The EBE and intermediate events are colour coded such that they correspond to the different HVAC components thus showing how the propagation of faults in the HVAC is reflected within the FMT.]{}[]{data-label="fig:FMT"}](HVAC_CONV_FMT.pdf){width="78.00000%"}
[\*5c]{} **Fault Index** & **Failure Mode** & **N** & **MTTF**\
& & &**(years)**\
1 & Broken AHU Damper & 4 & 20\
2 & Fan motor failure & 3 & 35\
3 & Obstructed supply fan& 4 & 31\
4 & Fan bearing failure &6 &17\
5 & Radiator failure & 4 & 25\
6 & Radiator stuck valve & 2 & 10\
7 & Heater stuck valve & 2&10\
8 & Failure in heat pump &4&20\
Quantitative results {#subsec:CaseStudy:Res}
--------------------
In the following subsections, we employ the developed framework (cf. Subsec. \[subsec:FMTFrame:Dec\]) to the FMT representing the failure of the HVAC system (cf. Fig. \[fig:FMT\]) and perform three different experiments. We first demonstrate the use of the developed framework by converting the FMT for the HVAC set-up into an abstract CTMC. For this abstract CTMC we compute the metrics (cf. Sec. \[subsec:FMTFrame:Met\]) using probabilistic model checking to show the type of analysis that can be performed using the set-up. Next, we perform a comparison between different maintenance strategies applied to the same FMT. This allows the user to deduce the optimal strategy for the set-up. Last, we construct a FMT which does not employ the repair and inspection module and compare it with the original FMT (includes the maintenance modules) to further highlight the advantage of incorporating maintenance.
#### **Applying the framework to HVAC set-up** {#subsec:CaseStudy:Res:1}
We convert the FMT representing the failure of the HVAC system into the equivalent abstract CTMC and perform probabilistic model checking over six time horizons $N_r= \{0,5,10,15,20,$ $25\}$ years with the maintenance policy consisting of periodic cleaning every [${T_{rep}}= 2 $ years]{} and inspections every [ ${T_{insp}}= 1$ year]{}. No replacement actions are considered. For this set-up, all the metrics corresponding to the reliability, availability, total costs (maintenance, inspection and operational costs) and the total expected number of failures of the HVAC systems over the time horizon are computed and are shown in Figure \[fig:Rel\_Avail\]. The total maintenance cost to perform a clean is 100 \[GBP\], while an inspection cost 50 \[GBP\]. The maximal time taken to compute a metric using the abstract FMT is 1.47 minutes. It is deduced that the reliability reduces over time. The availability is seen to be nearly constant, while the expected number of failures increases until it reaches a steady state value. This shows that there is a saturation in the number of maintenance actions which one can perform before the system no longer achieves higher performance in reliability and availability. One can further note that, as expected, the maintenance costs increases linearly with time.
#### **Comparison between different maintenance strategies** {#subsec:CaseStudy:Res:2}
In this second experiment, we compare all the metrics (reliability, availability, total costs and expected number of failures) over the time horizon $N_r= \{0,5,10,15,20,25\}$ years when considering different maintenance strategies, such that we can identify the optimal maintenance strategy that minimises cost and achieves the best trade-off in HVAC performance (i.e. with minimal expected number of failures and high reliability and availability). We consider five different maintenance strategies which are listed in Table \[tab:MaintStrat\].
[\*5c]{} **Strategy index** & [${T_{rep}}$]{} & ${T_{oh}}$ & ${T_{insp}}$\
$M_0$ & 2 years & - & 1 year\
$M_1$ & 5 years & - & 2 years\
$M_2$ & 2 years & 5 years & -\
$M_3$ & 2 years & 10 years & 1 year\
$M_4$ & 2 years & 20 years & 6 months\
We select strategies that have a different combination of repair, inspection and replacement strategies to highlight the effect the different maintenance actions have on the HVAC system’s performance. Figure \[fig:NumFailures\] depicts the resulting metrics for the employed strategies.
; ; ; ; ;
\
We can deduce that the worst performing strategy is when cleaning actions are carried out every 5 years with inspection carried out bi-annually and no replacements (corresponding to strategy $M_1$). Strategies $M_2$ and $M_3$ have comparable high performance but with a significant increase in the total costs due to the replacement action. We witness the highest costs using strategy $M_2$ due to the frequent replacement of the HVAC system. Comparing strategies $M_3$ and $M_4$ we can note that $M_3$ has fewer number of failures over the whole time horizon but this comes with higher total costs due to the replacements. Strategies $M_0$ and $M_4$ have similar performance with $M_0$ having a slightly lower availability and higher expected number of failures but with comparable maintenance costs. From this analysis, we can deduce that the optimal strategy which gives the best trade-off between costs and HVAC system’s performance is strategy $M_0$ (i.e. with annual inspections, bi-annual cleaning and no replacements).
#### **Comparison between performing maintenance and no maintenance**
Lastly, we compare the performance of the HVAC system without performing any maintenance actions vs the HVAC system with annual inspections, bi-annual cleaning and a major overhaul after 10 years. We employ the developed framework to represent the FMT of the HVAC system, first without incorporating the repair and inspection modules and then incorporating the repair and inspection modules with ${T_{insp}}= 1$ year, ${T_{rep}}=2$ years and ${T_{oh}}=10$ years. \[subsec:CaseStudy:Res:3\]
; ;
\
The obtained results, depicted in Figure \[fig:CompRMIM\], highlight the importance of maintenance and how appropriate maintenance strategies are required in order to maintain a reliable and available HVAC. When no maintenance is performed, both the reliability and availability of the HVAC system are gradually reduced, while the expected number of failures increases, as the components are degrading with time. This is in contrast to when maintenance is performed where high performance values of reliability and availability are achieved and the expected number of failures are low, throughout the whole time horizon. One should note, that this comes at a price, where the total costs increase when maintenance is applied. Consequently, this further highlights the need to perform an analysis to deduce the optimal maintenance strategy which gives the best trade-off between costs, reliability, availability and the expected number of failures.
Conclusion and Future Works {#sec:Conc}
===========================
The paper presents a methodology for applying probabilistic model checking to FMTs. We model FMTs using CTMCs which simplify the transformation of FMT into formal models that can be analysed using PRISM. We further present a novel technique for abstracting the equivalent CTMC model. The novel decomposition procedure tackles the issue of state space explosion and results in a significant reduction in both the state space size and the total time required to compute metrics. The framework is applied to an HVAC system and a set of different experiments to demonstrate the use of the developed framework and to highlight (i) the importance of performing maintenance and (ii) the effect of applying different maintenance strategies has been presented. The presented framework can be further enhanced by adding more gates to the PRISM modules library which include the Priority-AND, INHIBIT, k/N gates and to incorporate lumping of states as in [@yevkin2015efficient].
[ The author’s would also like to thank Carlos E. Budde and Enno Ruijters for their useful discussion and suggestions. This work has been funded by the [AMBI project]{} under Grant No.: [324432]{}, by the Alan Turing Institute, UK, post-doctoral research grant from [Fonds de Recherche du Quebec - Nature et Technologies (FRQNT)]{} and Malta’s ENDEAVOUR Scholarships Scheme.]{}
[^1]: Parts of this paper have been published in the 4th ACM International Conference on Systems of Energy-Efficient Build Environments (BuildSys 2017) [@cauchi2017efficient].
[^2]: This is a direct consequence of synchronisation and corresponds to ${\textbf{R}}\times {\textbf{R}}_{EBE}$. Refer to Subsection \[subsubsec:FMTFrame:Modelling:Semantics\]
[^3]: Note, this effectively results in changing the deterministic delay being modelled by the DELAY module to a new value if $input = 1$.
|
{
"pile_set_name": "ArXiv"
}
|
---
address:
- 'Institute of Mathematics, Hunan University, Changsha 410082, China'
- 'School of Mathematical Sciences, Dalian University of Technology, Liaoning 116024, China'
author:
- 'Chao-Ping Dong'
- Guobiao Weng
title: 'Minuscule representations, the graded poset $\Delta(1)$, and Panyushev conjectures'
---
Introduction
============
Let ${\mathfrak{g}}$ be a finite-dimensional simple Lie algebra over ${\mathbb{C}}$. Fix a Cartan subalgebra ${\mathfrak{h}}$ of ${\mathfrak{g}}$. The associated root system is $\Delta=\Delta({\mathfrak{g}}, {\mathfrak{h}})\subseteq{\mathfrak{h}}_{{\mathbb{R}}}^*$. Recall that a decomposition $$\label{grading}
{\mathfrak{g}}=\bigoplus_{i\in {\mathbb{Z}}}{\mathfrak{g}}(i)$$ is a *${\mathbb{Z}}$-grading* of ${\mathfrak{g}}$ if $[{\mathfrak{g}}(i), {\mathfrak{g}}(j)]\subseteq {\mathfrak{g}}(i+j)$ for any $i, j\in{\mathbb{Z}}$. In particular, in such a case, ${\mathfrak{g}}(0)$ is a Lie subalgebra of ${\mathfrak{g}}$. Since each derivation of ${\mathfrak{g}}$ is inner, there exists $h_0\in{\mathfrak{g}}(0)$ such that ${\mathfrak{g}}(i)=\{x\in{\mathfrak{g}}\mid [h_0, x]=i x\}$. The element $h_0$ is said to be *defining* for the grading . Without loss of generality, one may assume that $h_0\in{\mathfrak{h}}$. Then ${\mathfrak{h}}\subseteq{\mathfrak{g}}(0)$. Let $\Delta(i)$ be the set of roots in ${\mathfrak{g}}(i)$. Then we can choose a set of positive roots $\Delta(0)^+$ for $\Delta(0)$ such that $$\Delta^+ :=\Delta(0)^+\sqcup \Delta(1)\sqcup \Delta(2)\sqcup \cdots$$ is a set of positive roots of $\Delta({\mathfrak{g}}, {\mathfrak{h}})$. Let $\Pi$ be the corresponding simple roots, and put $\Pi(i)=\Delta(i)\cap
\Pi$. Note that the grading is fully determined by $\Pi=\bigsqcup_{i\geq 0} \Pi(i)$. If $|\Pi(1)|=1$ and $\Pi(i)$ vanishes for $i\geq 2$, we say the ${\mathbb{Z}}$-grading is *$1$-standard*. We refer the reader to Ch. 3, §3 of [@GOV] for generalities on gradings of Lie algebras, see also the paper of Vinberg [@Vin].
In the above setting, each $\Delta(i)$, $i\geq 1$, inherits a poset structure from the usual one of $\Delta^+$. That is, let $\alpha$ and $\beta$ be two roots of $\Delta(i)$, then $\alpha\leq\beta$ if and only if $\beta-\alpha$ is a nonnegative integer combination of simple roots. Among these posets, as revealed by the recent studies of Panyushev [@P; @P2], it turns out that $\Delta(1)$ is of particular importance and carries rich structure.
In [@P], Panyushev raised several beautiful conjectures concerning the $\mathcal{M}$-polynomial, $\mathcal{N}$-polynomial and the reverse operator in $\Delta(1)$. Before stating his conjectures, let us recall necessary notation. When the ${\mathbb{Z}}$-grading is $1$-standard, $\Delta(1)$ has the form $$[\alpha_i]:=\{\alpha\in\Delta^+ \mid [\alpha: \alpha_i]=1\}.$$ Here $\alpha_i$ is a simple root, and $[\alpha: \alpha_i]$ is the coefficient of $\alpha_i$ in $\alpha$. We refer the reader to a paper of Ringel [@R] for vivid pictures of the Hasse diagrams for the posets $\Delta^+$, from which one can figure out those of $[\alpha_i]$ readily, see Section 4.
Recall that a subset $I$ of a finite poset $(P, \leq)$ is a *lower* (resp., *upper*) *ideal* if $x\leq y$ in $P$ and $y\in I$ (resp. $x\in I$) implies that $x\in I$ (resp. $y\in I$). Let $J(P)$ be the lower ideals of $P$, partially ordered by inclusion. A subset $A$ of $P$ is an *antichain* if its elements are mutually incomparable. Note that the following maps give bijections between lower ideals, upper ideals and antichains of $P$: $$I\mapsto P\setminus I \mapsto \min(P\setminus I).$$ Denote by $\mathcal{M}_{P}(t)$ the generating function of lower ideals of $P$. That is, $\mathcal{M}_{P}(t):=\sum_{I} t^{|I|}$, where $I$ runs over the lower ideals of $P$. Denote by $\mathcal{N}_{P}(t)$ the generating function of antichains of $P$. That is, $\mathcal{N}_{P}(t):=\sum_{A} t^{|A|}$, where $A$ runs over the antichains of $P$.
As on p. 244 of Stanley [@St], a finite poset $P$ is said to be *graded* if every maximal chain in $P$ has the same length. In this case, there is a unique rank function $r$ from $P$ to the positive integers $\mathbb{P}$ such that all the minimal elements have rank $1$, and $r(x)=r(y)+1$ if $x$ covers $y$. The model for our concern is $\Delta(1)$, where the height function $\rm{ht}$ gives the rank.
Now Conjecture 5.1 of [@P] is stated as follows.
**Panyushev’s ${\mathcal{M}}$-polynomial conjecture.** *For any ${\mathbb{Z}}$-grading of ${\mathfrak{g}}$, we have $$\label{KM}
\mathcal{M}_{\Delta(1)}(t)=\prod_{\gamma\in\Delta(1)}
\frac{1-t^{\rm{ht}(\gamma)+1}}{1-t^{\rm{ht}(\gamma)}}.$$*
Note that the RHS of traces back to the celebrated Kostant-Macdonald identity (see [@Ko1] and Corollary 2.5 of [@M]), which says that $$\sum_{w\in W}t^{l(w)}=\prod_{\gamma\in\Delta^+}\frac{1-t^{\rm{ht}(\gamma)+1}}{1-t^{\rm{ht}(\gamma)}}.$$ Here $W$ is the Weyl group associated with $\Delta^+$, and $l(\cdot)$ is the length function. Actually, like many other studies, the current one is inspired by the influential paper of Kostant [@Ko] which links (affine) Weyl groups, abelian ideals of a Borel subalgebra, and representation theory of Lie groups.
Note also that when the grading is *abelian* (i.e., when $\Delta(i)$ vanishes for $i\geq 2$), the poset $[\alpha_i^{\vee}]$ in the dual root system $\Delta^{\vee}$ is *minuscule* in the sense of Proctor [@Pr], see Section 3 for more details. According to Exercise 3.170 of Stanley [@St], we call a finite graded poset $P$ *pleasant* if , with $\rm{ht}$ replaced by the rank function $r$, holds for it. Thus Panyshev’s ${\mathcal{M}}$-polynomial conjecture asserts that each $\Delta(1)$ is pleasant.
As noted in [@P], to handle the above conjecture, it boils down to consider the $1$-standard ${\mathbb{Z}}$-gradings. Note further that when the grading is abelian or *extra-special* (i.e., $\Delta(i)$ vanishes for $i\geq 3$ and $|\Delta(2)|=1$), or when $\alpha_i$ is a branching point in the Dynkin diagram, it has been settled in [@P]. However, essentially we do not need these results. Instead, the first aim of this paper is to remark that Panyushev’s ${\mathcal{M}}$-polynomial conjecture follows from Proctor’s Theorem (see Theorem \[thm-Proctor\]) plus some additional effort. The key observation is that for all but seven exceptions (see Section 4.10) these $[\alpha_i]$ bear the pattern $$\label{pattern}
[\alpha_i]\cong [k]\times P,$$ where $k$ is a positive integer, $[k]$ denotes the totally ordered set $\{1, 2, \cdots, k\}$, and $P$ is a connected minuscule poset classified in Theorem \[thm-Proctor\]. As a consequence, we will confirm the following.
\[thm-M-poly-main\] Panyushev’s ${\mathcal{M}}$-polynomial conjecture is true.
Although the pattern is demonstrated case-by-case, the underlying method is the same: There are at most three (sub) connected components of the Dynkin diagram of ${\mathfrak{g}}$ having $\alpha_i$ as an ending point. One component must be $A_k$, where $k=1$ if $\alpha_i$ itself is an ending point, and this component produces the $[k]$ in ; the other component(s) produces the minuscule poset $P$ in , see Section 4. Moreover, we find that there are seven ending points $\alpha_i$’s of the Dynkin diagrams such that the $[\alpha_i]$’s are not minuscule. The pattern is violated exactly in these cases. We shall check that all the seven $[\alpha_i]$’s are pleasant, while none of them is *Gaussian* in the sense of . This finding gives further support to the Gaussian poset conjecture raised by Proctor in 1984 (see Section 9 of [@Pr]) which claims that *all* the connected Gaussian posets are minuscule.
It is obvious that the number ${\mathcal{M}}_{\Delta(1)}(1)$ counts the lower ideals of $\Delta(1)$. Namely, we have the following.
\[cor-thm-M-poly-main\] For any ${\mathbb{Z}}$-grading of ${\mathfrak{g}}$, $$\label{number-antichains}
{\mathcal{M}}_{\Delta(1)}(1)=|J(\Delta(1))|=\prod_{\gamma\in\Delta(1)}\frac{\rm{ht}(\gamma)+1}{\rm{ht}(\gamma)}.$$
The above corollary has been obtained earlier by Panyushev [@P2] using hyperplane arrangements. However, the method there does not lead to a proof of Theorem \[thm-M-poly-main\].
Let $E$ (resp. $F$) be the *multi-set* of the even (resp. odd) heights of $\Delta(1)$. By Theorem \[thm-M-poly-main\], we have $$\label{M-1}
{\mathcal{M}}_{\Delta(1)}(-1)=
\begin{cases}
\prod_{f\in F}(f+1)/\prod_{e\in E} e &
\mbox{ if } |E|=|F|, \\ 0&
\mbox{ otherwise.}
\end{cases}$$ It is far from being evident that the number ${\mathcal{M}}_{\Delta(1)}(-1)$ counts certain lower ideals of $\Delta(1)$ enjoying nice symmetries. Indeed, suppose that $c: P\to P$ is an order-reversing involution on the finite poset$(P, \leq)$. Follow Stembridge [@St], we call the triple $(P, \leq, c)$ a *complemented* poset. In such a case, for any $I\in J(P)$, put $I^c:=P\setminus \{c(x)\mid x\in
I\}$. Then $I\mapsto I^c$ is an order-reversing involution on $J(P)$. This makes $J(P)$ into a complemented poset as well, for which we denote by $(J(P), \subseteq, c)$, or simply by $(J(P), c)$. We call a lower ideal $I\in J(P)$ *self-complementary* if $I=I^c$. In our situation, let $w_0^i$ be the longest element of the Weyl group of ${\mathfrak{g}}(0)$ coming from the $1$-standard ${\mathbb{Z}}$-grading such that $\Pi(1)=\{\alpha_i\}$. Note that $w_0^i(\Delta(1))=\Delta(1)$, and the $w_0^i$ action on $\Delta(1)=[\alpha_i]$ makes it into a complemented poset, for which we denote by $([\alpha_i], w_0^i)$. We denote the corresponding complemented poset structure on $J([\alpha_i])$ by $J([\alpha_i], w_0^i)$.
In Lemma \[lemma-uniform-transfer\], we shall transfer the order-reversing involution on each minuscule weight lattice coming from the $w_0$ action to the corresponding minuscule poset $\Delta(1)$. Here $w_0$ is the longest element of the Weyl group $W({\mathfrak{g}}, {\mathfrak{h}})$. Then we will build up further links between the pattern and the minuscule representations. This makes Stembridge’s “$t=-1$ phenomenon" (see Theorem 4.1 of [@Stem]) applicable, and allows us to prove Conjecture 5.2 of [@P].
\[thm-M-1-main\] Let ${\mathfrak{g}}=\bigoplus_{j\in {\mathbb{Z}}}{\mathfrak{g}}(j)$ be any $1$-standard ${\mathbb{Z}}$-grading of ${\mathfrak{g}}$ such that $\Pi(1)=\{\alpha_i\}$. Then the number ${\mathcal{M}}_{\Delta(1)}(-1)$ counts the self-complementary lower ideals in $J([\alpha_i], w_0^i)$.
Originally, Conjecture 5.2 of [@P] is stated in terms of upper ideals. In Lemma \[lemma-fixed-pt\], we will show that a lower ideal $I$ of $\Delta(1)$ is self-complementary if and only if the upper ideal $\Delta(1)\setminus I$ is self-complementary. Thus we can interpret the above theorem in terms of upper ideals instead. Using an argument similar to the one after Conjecture 5.1 of [@P], one sees that Theorem \[thm-M-1-main\] also holds for $\Delta(1)$ in a general ${\mathbb{Z}}$-grading.
It is interesting to ask that when does the number ${\mathcal{M}}_{\Delta(1)}(-1)$ vanish? A direct answer using is that this happens if and only if $|E|\neq |F|$. A deeper characterization is found as follows.
\[thm-fixed-point-main\] Let ${\mathfrak{g}}=\bigoplus_{j\in {\mathbb{Z}}}{\mathfrak{g}}(j)$ be any $1$-standard ${\mathbb{Z}}$-grading of ${\mathfrak{g}}$. Then ${\mathcal{M}}_{[\alpha_i]}$ vanishes at $-1$ if and only if the $w_0^i$ action on $[\alpha_i]$ has fixed point(s).
Let us turn to the second theme of this paper. Let $P_i$ be the set of elements in the finite graded poset $P$ with rank $i$. The sets $P_i$ are said to be the *rank levels* of $P$. Suppose that $P=\bigsqcup_{i=1}^{d} P_i$. Then $P$ is called *Sperner* if the largest size of an antichain equals $\max\{|P_i|, 1\leq i\leq
d\}$. Algebraic geometry and Lie theory had been used by Stanley in [@St80] to construct families of posets with the Sperner property, which then lead to a proof of the Erdös-Moser conjecture. See also the papers of Proctor [@Pr82; @Pr822].
Now let us state Conjecture 5.12 of [@P].
**Panyushev’s ${\mathcal{N}}$-polynomial conjecture.** *Let ${\mathfrak{g}}=\bigoplus_{i\in {\mathbb{Z}}}{\mathfrak{g}}(i)$ be any $1$-standard ${\mathbb{Z}}$-grading of ${\mathfrak{g}}$. Then ${\mathcal{N}}_{\Delta(1)}$ is palindromic if and only if $\Delta(1)$ has a unique rank level of maximal size.*
When the grading is abelian or extra-special, the above conjecture has been verified in [@P]. The second theme of this paper is to prove Panyushev’s ${\mathcal{N}}$-polynomial conjecture for the remaining cases. More precisely, we will establish the following.
\[thm-N-poly-main\] Let ${\mathfrak{g}}=\bigoplus_{i\in {\mathbb{Z}}}{\mathfrak{g}}(i)$ be any $1$-standard ${\mathbb{Z}}$-grading of ${\mathfrak{g}}$. The following are equivalent:
- $\mathcal{N}_{\Delta(1)}$ is palindromic;
- $\mathcal{N}_{\Delta(1)}$ is monic;
- $\Delta(1)$ has a unique antichain of maximal size;
- $\Delta(1)$ has a unique rank level of maximal size.
In particular, Panyushev’s ${\mathcal{N}}$-polynomial conjecture is true.
We collect the antichains of $P$ as $\mathrm{An}(P)$. For any $x\in P$, let $I_{\leq x}=\{y\in P\mid y\leq x\}$. Given an antichain $A$ of $P$, let $I(A)=\bigcup_{a\in A} I_{\leq a}$. The *reverse operator* $\mathfrak{X}$ is defined by $\mathfrak{X}(A)=\min
(P\setminus I(A))$. Since antichains of $P$ are in bijection with lower (resp. upper) ideals of $P$, the reverse operator acts on lower (resp. upper) ideals of $P$ as well. Note that the current $\mathfrak{X}$ is inverse to the reverse operator $\mathfrak{X}^{\prime}$ in Definition 1 of [@P], see Lemma \[lemma-inverse-reverse-operator\]. Thus replacing $\mathfrak{X}^{\prime}$ by $\mathfrak{X}$ does not affect our forthcoming discussion on orbits. Panyushev firstly observed special properties of $\mathfrak{X}$ on root posets [@P0]. When $P$ is a root poset, as in Armstrong, Stump and Thomas [@AST], we call $\mathfrak{X}$ the *Panyushev complement* and call a $\mathfrak{X}$-orbit a *Panyushev orbit*. The third theme of this paper is the structure of Panyushev orbits of $\Delta(1)$.
Recall that the ${\mathbb{Z}}$-grading is extra-special if $$\label{extra-special}
{\mathfrak{g}}={\mathfrak{g}}(-2)\oplus {\mathfrak{g}}(-1) \oplus {\mathfrak{g}}(0) \oplus {\mathfrak{g}}(1)
\oplus {\mathfrak{g}}(2) \mbox{ and }\dim{\mathfrak{g}}(2)=1,$$ Up to conjugation, any simple Lie algebra ${\mathfrak{g}}$ has a unique extra-special ${\mathbb{Z}}$-grading. Without loss of generality, we assume that $\Delta(2)=\{\theta\}$, where $\theta$ is the highest root of $\Delta^+$. Namely, we may assume that the grading is defined by the element $\theta^{\vee}$, the dual root of $\theta$.
In such a case, we have $$\label{Delta-one}
\Delta(1)=\{\alpha\in\Delta^+\mid (\alpha, \theta^{\vee})=1\}.$$ Recall that $h:=\mathrm{ht}(\theta)+1$ is the *Coxeter number* of $\Delta$. Let $h^*$ be the *dual Coxeter number* of $\Delta$. That is, $h^*$ is the height of $\theta^{\vee}$ in $\Delta^{\vee}$. As noted on p. 1203 of [@P], we have $|\Delta(1)|=2h^*-4$. We call a lower (resp. upper) ideal $I$ of $\Delta(1)$ *Lagrangian* if $|I|=h^*-2$. Write $\Delta_l$ (resp. $\Pi_l$) for the set of *all* (resp. *simple*) *long* roots. In the simply-laced cases, all roots are assumed to be both long and short. Note that $\theta$ is always long, while $\theta^{\vee}$ is always short.
\[thm-main-reverse-operator-orbit\] Conjecture 5.11 of [@P] is true. That is, in any extra-special ${\mathbb{Z}}$-grading of ${\mathfrak{g}}$, the number of Panyushev orbits equals $|\Pi_l|$, and each orbit is of size $h-1$. Furthermore, if $h$ is even (which only excludes the case $A_{2k}$ where $h=2k+1$), then each Panyushev orbit contains a unique Lagrangian lower ideal.
Originally, Conjecture 5.11 of [@P] was stated in terms of upper ideals and $\mathfrak{X}^{\prime}$. Equivalently, we can phrase it using lower ideals and the Panyushev complement $\mathfrak{X}$. Moreover, by the proof of Theorem \[thm-main-reverse-operator-orbit\], one checks easily that for any $1$-standard extra-special ${\mathbb{Z}}$-grading of ${\mathfrak{g}}$, all the statements of Conjecture 5.3 in [@P] hold.
The *cyclic sieving phenomenon (CSP)* was defined by Reiner, Stanton and White [@RSW] as follows: let $X$ be a finite set, let $X(t)$ be a polynomial in $t$ whose coefficients are nonnegative integers and let $C=\langle c\rangle$ be a cyclic group of order $n$ acting on $X$. The triple $(X, X(t), C)$ exhibits the CSP if $$\label{CSP-1}
X(t)\big|_{t=\zeta^k}=\big|\{x\in X \mid c^k(x)=x\}\big|,$$ where $\zeta$ is a primitive $n$-th root of unity. Let $$\label{CSP-2}
X(t)\equiv a_0 + a_1 t+\cdots +a_{n-1}t^{n-1} \mod (t^n-1).$$ By Proposition 2.1 of [@RSW], an equivalent way to define the CSP is to say that $a_i$ equals the number of $C$-orbits in $X$ whose stabilizer order divides $i$. The following result is a slight extension of the main theorems of Rush and Shi [@RS].
\[thm-main-cyclic-sieving\] Let ${\mathfrak{g}}=\bigoplus_{i\in {\mathbb{Z}}}{\mathfrak{g}}(i)$ be any $1$-standard ${\mathbb{Z}}$-grading of ${\mathfrak{g}}$. Then the triple $(\Delta(1)$, ${\mathcal{M}}_{\Delta(1)}(t)$, $\langle\mathfrak{X}_{\Delta(1)}\rangle)$ exhibits the CSP.
We observe certain *duality* within each Panyushev orbit, and make the following.
\[conj-main-duality\] Let ${\mathfrak{g}}=\bigoplus_{j\in {\mathbb{Z}}}{\mathfrak{g}}(j)$ be any $1$-standard ${\mathbb{Z}}$-grading of ${\mathfrak{g}}$. Then there exists an order-reversing involution $c$ of $\Delta(1)$ such that $I\in \mathcal{O}$ if and only if $I^{c}\in \mathcal{O}$, where $I$ is any lower ideal of $\Delta(1)$ and $\mathcal{O}$ is any Panyushev orbit.
Note that Conjecture 5.3 (iii) of [@P] would follow directly from Conjecture \[conj-main-duality\].
To close the introduction, we remark that it seems rather hard to push the powerful Lie-theoretic techniques in [@P] further to handle a general ${\mathbb{Z}}$-grading which is niether abelian nor extra-special. Instead, our approach lives mainly in combinatorics, and the core is minuscule poset. It has demerits. Indeed, we resort to the classification of simple Lie algebras, and adopt computer verifications via `Mathematica` in the following cases: the seven posets violating the pattern for Theorems \[thm-M-poly-main\], \[thm-M-1-main\], \[thm-N-poly-main\], and \[thm-main-cyclic-sieving\]; the exceptional Lie algebras for Theorem \[thm-main-reverse-operator-orbit\]. (The program files are available from the first named author.) But our approach has the merit of being straightforward and effective. Moreover, it echoes Section 11 of Proctor [@Pr] written in 1984, where the ubiquity of minuscule posets in mathematics was demonstrated.
The paper is organized as follows: We prepare some preliminary results in Section 2, and recall Proctor’s Theorem in Section 3. Then we analyze the structure of the posets $[\alpha_i]$ in Section 4, and show Theorem \[thm-M-poly-main\] in Section 5. We make Stembridge’s theorem applicable, and prove Theorems \[thm-M-1-main\] and \[thm-fixed-point-main\] in Section 6. We deduce some results on ${\mathcal{N}}$-polynomials and verify Theorem \[thm-N-poly-main\] in Section 7. Finally, Theorem \[thm-main-reverse-operator-orbit\] and \[thm-main-cyclic-sieving\] are obtained in Section 8.
**Notation.** Let ${\mathbb{N}}=\{0, 1, 2, \dots\}$, and let $\mathbb{P}=\{1, 2, \dots\}$. For each $k\in\mathbb{P}$, the poset $[k]$ is equipped with the order-reversing involution $c$ such that $c(i)=k+1-i$. We denote $J(J(P))$ and $J(J(J(P)))$ by $J^2(P)$ and $J^3(P)$, respectively.
Preliminary results
===================
Let us collect some preliminary results in this section. Let $(P_i,\leq), i=1, 2$ be two finite posets. One can define a poset structure on $P_1\times P_2$ by setting $(u_1, v_1)\leq (u_2, v_2)$ if and only if $u_1\leq u_2$ in $P_1$ and $v_1\leq v_2$ in $P_2$. We simply denote the resulting poset by $P_1 \times P_2$. The following lemma gives all lower ideals of $P_1\times P_2$.
\[lemma-P1P2\] Let $P_1, P_2$ be two finite posets. Let $S$ be a subset of $P_1\times P_2$. For each $u\in P_1$, put $S_u=\{v \in
P_2|(u,v)\in S\}$. Then $S$ is a lower ideal of $P_1\times
P_2$ if and only if $S_u$ is a lower ideal of $P_2$ for each $u\in P_1$, and that $S_{u_1}\supseteq S_{u_2} $ whenever $u_1\leq u_2$ in $P_1$.
It suffices to prove the sufficiency. Given $(u,v)\in S$, take any $(x,y)\in P_1\times P_2$ such that $(x, y)\leq (u,v)$, then $x\leq
u$ and $y\leq v$. Firstly, we have $y\in S_u$ since $S_u$ is a lower ideal of $P_2$ and $v\in S_u$. Secondly, since $x\leq u$, we have $S_x\supseteq S_u$. Hence $y\in S_x$, i.e., $(x,y)\in S$. This proves that $S$ is a lower ideal of $P_1\times P_2$.
As a direct consequence, we have the following well-known result describing the lower ideals of $[n]\times P$.
\[lemma-ideals-CnP\] Let $P$ be a finite poset. Let $I$ be a subset of $[m]\times
P$. For $1\leq i\leq m$, denote $I_i=\{a\in P\mid (i, a)\in I\}$. Then $I$ is a lower ideal of $[m]\times P$ if and only if each $I_i$ is a lower ideal of $P$, and $I_m\subseteq I_{m-1}\subseteq \cdots \subseteq I_{1}$.
The following lemma describes the antichains in $[m]\times P$.
\[lemma-antichain-CnP\] Let $P$ be a finite poset. Let $A$ be a subset of $[m]\times P$. For $1\leq i\leq m$, denote $A_i=\{a\in P\mid (i, a)\in A\}$. Then $A$ is an antichain of $[m]\times P$ if and only if each $A_i$ is an antichain of $P$, and $A_i\subseteq P\setminus I(A_{i+1})$ for $1\leq i\leq m-1$.
This follows readily from the definition of antichain.
Now let us compare the two reverse operators. Let $(P, \leq)$ be a finite poset. For any $x\in P$, let $I_{\geq x}=\{y\in P\mid x\leq
y\}$. For any antichain $A$ of $P$, put $I_{+}(A)=\bigcup_{a\in A}
I_{\geq a}$. Recall that in Definition 1 of [@P], the reverse operator $\mathfrak{X}^{\prime}$ is given by $\mathfrak{X}^{\prime}(A)=\max (P\setminus I_{+}(A))$.
\[lemma-inverse-reverse-operator\] The operators $\mathfrak{X}$ and $\mathfrak{X}^{\prime}$ are inverse to each other.
Take any antichain $A$ of $P$, note that $$I_{+}(\min(P\setminus
I(A)))=P\setminus I(A)\mbox{ and } I(\max(P\setminus
I_{+}(A)))=P\setminus I_{+}(A).$$ Then the lemma follows.
Suppose that $P=\bigsqcup_{j=1}^{d} P_j$ is the decomposition of a finite graded poset $P$ into rank levels. Let $P_0$ be the empty set $\emptyset$. Put $L_i=\bigsqcup_{j=1}^{i} P_j$ for $1\leq j\leq d$, and let $L_0$ be the empty set. We call those $L_i$ *full rank* lower ideals. Recall that the reverse operator acts on lower ideals as well. For instance, $\mathfrak{X}_P(L_i)=L_{i+1}$, $0\leq i<d$ and $\mathfrak{X}_P(L_d)=L_{0}$.
Let $\mathfrak{X}$ be the reverse operator on $[m]\times P$. In view of Lemma \[lemma-ideals-CnP\], we *identify* a general lower ideal of $[m]\times P$ with $(I_1, \cdots, I_m)$, where each $I_i\in J(P)$ and $I_m\subseteq \cdots \subseteq I_{1}$. We say that the lower ideal $(I_1, \cdots, I_m)$ is *full rank* if each $I_i$ is full rank in $P$. Let $\mathcal{O}(I_1, \cdots, I_m)$ be the $\mathfrak{X}_{[m]\times P}$-orbit of $(I_1, \cdots, I_m)$. We prepare the following.
\[lemma-operator-ideals-CmP\] Keep the notation as above. Then for any $n_0\in {\mathbb{N}}$, $n_i\in\mathbb{P}$ ($1\leq i\leq s$) such that $\sum_{i=0}^{s} n_i =m$, we have $$\label{rank-level}
\mathfrak{X}_{[m]\times P}(L_d^{n_0}, L_{i_1}^{n_1}, \cdots, L_{i_s}^{n_s})=
(L_{i_1+1}^{n_0+1}, L_{i_2+1}^{n_1}, \cdots, L_{i_s+1}^{n_{s-1}}, L_0^{n_s-1}),$$ where $0\leq i_s<\cdots <i_1<d$, $L_d^{n_0}$ denotes $n_0$ copies of $L_d$ and so on.
Note that under the above assumptions, $(L_d^{n_0}, L_{i_1}^{n_1}, \cdots, L_{i_s}^{n_s})$ is a lower ideal of $[m]\times P$ in view of Lemma \[lemma-ideals-CnP\]. Then analyzing the minimal elements of $([m]\times P)\setminus (L_d^{n_0}, L_{i_1}^{n_1}, \cdots, L_{i_s}^{n_s})$ leads one to .
\[lemma-operator-types\] Let $(I_1, \cdots, I_m)$ be an arbitrary lower ideal of $[m]\times P$. Then $(I_1, \cdots, I_m)$ is full rank if and only if each lower ideal in the orbit $\mathcal{O}(I_1, \cdots, I_m)$ is full rank.
Use Lemma \[lemma-operator-ideals-CmP\].
Due to the above lemma, we say the $\mathfrak{X}_{[m]\times P}$-orbit $\mathcal{O}(I_1, \cdots, I_m)$ is of *type I* if $(I_1, \cdots, I_m)$ is full rank, otherwise we say $\mathcal{O}(I_1, \cdots, I_m)$ is of *type II*.
For any $n\geq 2$, let $K_{n-1}=[n-1]\oplus([1]\sqcup [1])\oplus
[n-1]$ (the ordinal sum, see p. 246 of [@St]). We label the elements of $K_{n-1}$ by $1$, $2$, $\cdots$, $n-1$, $n$, $n^{\prime}$, $n+1$, $\cdots$, $2n-2$, $2n-1$. Fig. 1 illustrates the labeling for $K_3$. Note that $L_i$ ($0\leq i\leq 2n-1$) are all the full rank lower ideals. For instance, we have $L_{n}=\{1, 2,
\cdots, n, n^{\prime}\}$. Moreover, we put $I_{n}=\{1, \cdots, n-1,
n\}$ and $I_{n^{\prime}}=\{1, \cdots, n-1, n^{\prime}\}$. The following lemma will be helpful in analyzing the $\mathfrak{X}_{[m]\times K_{n-1}}$-orbits of type II.
\[lemma-operator-ideals-CmK\] Fix $n_0\in {\mathbb{N}}$, $n_i\in\mathbb{P}$ ($1\leq i\leq s$), $m_j\in\mathbb{P}$ ($0\leq j\leq t$) such that $\sum_{i=0}^{s} n_i +
\sum_{j=0}^{t} m_j=m$. Take any $0\leq j_t< \cdots<j_1<n\leq
i_s<\cdots <i_1<2n-1$, we have $$\begin{aligned}
\mathfrak{X}_{[m]\times K_{n-1}}&(L_{2n-1}^{n_0}, L_{i_1}^{n_1}, \cdots, L_{i_s}^{n_s}, I_n^{m_0}, L_{j_1}^{m_1}, \cdots, L_{j_t}^{m_t})=\\
&\begin{cases}
( L_{i_1+1}^{n_0+1}, L_{i_2+1}^{n_1}, \cdots, L_{i_s+1}^{n_{s-1}}, I_{n^{\prime}}^{n_s}, L_{j_1+1}^{m_0},
L_{j_2+1}^{m_1}, \cdots, L_{j_t+1}^{m_{t-1}}, L_0^{m_t-1} ) & \mbox { if } j_1 < n-1;\\
( L_{i_1+1}^{n_0+1}, L_{i_2+1}^{n_1}, \cdots, L_{i_s+1}^{n_{s-1}}, L_{n}^{n_s}, I_n^{m_0},
\, \, \, \, L_{j_2+1}^{m_1}, \cdots, L_{j_t+1}^{m_{t-1}}, L_0^{m_t-1} )& \mbox { if } j_1 = n-1.
\end{cases}\end{aligned}$$
Analyzing the minimal elements of $$([m]\times K_{n-1})\setminus (L_{2n-1}^{n_0}, L_{i_1}^{n_1}, \cdots, L_{i_s}^{n_s}, I_n^{m_0}, L_{j_1}^{m_1}, \cdots, L_{j_t}^{m_t})$$ leads one to the desired expression.
Proctor’s Theorem
=================
In this section, we will recall minuscule representations, minuscule posets, and a theorem of Proctor. We continue to denote by ${\mathfrak{g}}$ a finite-dimensional simple Lie algebra over ${\mathbb{C}}$ with rank $n$. Let $V_{\lambda}$ be a finite-dimensional irreducible ${\mathfrak{g}}$-module with highest weight $\lambda$. Denote by $\Lambda_{\lambda}$ the multi-set of weights in $V_{\lambda}$. One says that $V_{\lambda}$ (and hence also ${\lambda}$) is *minuscule* if the action of $W$ on $\Lambda_{\lambda}$ is transitive. By Exercise VI.1.24 of Bourbaki [@B], a minuscule weight $\lambda$ must be a fundamental weight. However, the converse is not true. We refer the reader to the appendix of [@Stem] for a complete list of minuscule weights.
Now let $V_{\varpi_i}$ be a minuscule representation, where $\varpi_i$ is the fundamental weight corresponding to the $i$-th simple root $\alpha_i\in\Pi$. Namely, for any $1\leq j \leq n$, $$\langle \varpi_i, \alpha_j^{\vee} \rangle=\delta_{ij},$$ where $\alpha_j^{\vee}=2\alpha_j /\|\alpha_j\|^2$. Then by Proposition 4.1 of [@P], one knows that the poset $\Lambda_{\varpi_i}$ is a distributive lattice. Thus by Theorem 3.4.1 of [@St], there is a (unique) poset $P_{\varpi_i}$ such that $\Lambda_{\varpi_i}\cong J(P_{\varpi_i})$. Indeed, we point out that $$\label{minu-real}
P_{\varpi_i}\cong [\alpha_i^{\vee}] \mbox{ in } (\Delta^{\vee})^+,$$ where $\Delta^{\vee}$ is the root system dual to $\Delta$. Moreover, these $P_{\varpi_i}$ are exactly the *minuscule posets* in the sense of [@P].
Let us recall from Exercise 3.172 of [@St] that a finite graded poset $P=\{t_1, \dots, t_p\}$ is *Gaussian* if there exists positive integers $h_1, \dots, h_p>0$ such that for each $m\in
{\mathbb{N}}$, $$\label{Gaussian}
\mathcal{M}_{[m]\times
P}(t)=\prod_{i=1}^{p}\frac{1-t^{m+h_i}}{1-t^{h_i}}.$$
Now let us state Proctor’s theorem, which is a combination of Proposition 4.2 and Theorem 6 of [@Pr].
\[thm-Proctor\] *(**Proctor**)* The connected minuscule posets are classified as below: $[n]\times
[m]$, for all $m, n\in \mathbb{P}$; $K_r:=[r]\oplus([1]\sqcup [1])\oplus
[r]$ (the ordinal sum, see p. 246 of [@St]), for all $r\in
\mathbb{P}$; $H_r:=J([2]\times [r])$, for all $r\in \mathbb{P}$; $J^2([2]\times [3])$ and $J^3([2]\times [3])$. Moreover, each minuscule poset is Gaussian.
\[rmk-thm-Proctor\] By Exercise 3.172 of [@St], $P$ is Guassian if and only if $P\times [m]$ is pleasant for each $m\in \mathbb{P}$.
For reader’s convenience, we present the Hasse diagrams of $J^2([2]\times [3])$ and $J^3([2]\times [3])$ in Fig. 2.
The structure of $\Delta(1)$
============================
This section is devoted to understanding the structure of $\Delta(1)=[\alpha_i]$ in a general $1$-standard ${\mathbb{Z}}$-grading. Ringel’s paper [@R] is very helpful on this aspect. The main point is to demonstrate that most of these $[\alpha_i]$ observe the pattern . Those $[\alpha_i]$ violating the pattern are listed in the final subsection.
Although our discussion below is case-by-case, the underlying method is the same. Indeed, let $X_n$ be the type of ${\mathfrak{g}}$. If $\alpha_i$ is not a branching point, then there are two (sub) connected components in the Dynkin diagram of $X_n$ containing $\alpha_i$ as an ending point: one is $A_{k}$, and the other is $Y_{n-k+1}$. Here $k=1$ if $\alpha_i$ itself is an ending point in $X_n$. Then $A_{k}$ produces $[k]$, while the minuscule poset $P$ is related to $Y_{n-k+1}$. Actually, if in $Y_{n-k+1}$ the fundamental weight corresponding to $\alpha_i$ is minuscule, then $P$ is just $[\alpha_i]$ in $Y_{n-k+1}$. If $\alpha_i$ is a branching point, then there are three (sub) connected components: $A_{2}$, $A_{r+1}$ and $A_{s+1}$, where $r+s=n-2$, and we have $[\alpha_i]\cong [2]\times
[r+1] \times [s+1]$.
$A_n$
-----
We fix $\alpha_i=e_i-e_{i+1}$, $1\leq i\leq n$, then $$[\alpha_i]\cong [i]\times [n+1-i], \quad 1\leq i\leq n.$$
$B_n$
-----
We fix $\alpha_i=e_i-e_{i+1}$, $1\leq i\leq n-1$, and $\alpha_n=e_n$ as the simple roots. Then $$[\alpha_i] \cong [i]\times [2n+1-2i], \quad 1\leq i\leq n.$$
$C_n$
-----
We fix $\alpha_i=e_i-e_{i+1}$, $1\leq i\leq n-1$, and $\alpha_n=2e_n$ as the simple roots. Then $[\alpha_n]\cong H_n$, and $$[\alpha_i] \cong [i]\times [2n-2i], \quad 1\leq i\leq n-1.$$
$D_n$
-----
We fix $\alpha_i=e_i-e_{i+1}$, $1\leq i\leq n-1$, and $\alpha_n=e_{n-1}+e_n$ as the simple roots. Then $[\alpha_{n-1}]\cong [\alpha_n]\cong H_{n-1}$, and $$\label{}
[\alpha_i]\cong [i] \times K_{n-i-1}, \quad 1\leq i\leq n-2.$$
$G_2$
-----
Let $\alpha_1$ be the short simple root, and let $\alpha_2$ be the long simple root. Then $[\alpha_1]\cong [2]$ and $[\alpha_2]\cong
[4]$.
$F_4$
-----
The Dynkin diagram is as follows, where the arrow points from long roots to short roots.
Then $[\alpha_1]\cong K_3$, $[\alpha_2]\cong [2]\times [3]$, $[\alpha_3]\cong [2]\times K_2$.
$E_6$
-----
The Dynkin diagram of $E_6$ is given below. Note that our labeling of the simple roots agrees with p. 687 of [@K], while differs from that of [@P].
Then $[\alpha_1]\cong [\alpha_6]\cong J^2([2]\times [3])$, $[\alpha_3]\cong [\alpha_5]\cong [2]\times H_4$, and $[\alpha_4]\cong [2]\times [3]\times [3]$.
$E_7$
-----
The Dynkin diagram is obtained from that of $E_6$ by adding $\alpha_7$ adjacent to $\alpha_6$. Then $[\alpha_3]\cong [2]\times H_{5}$, $[\alpha_4]\cong [2]\times
[3]\times [4]$, $[\alpha_5]\cong [3]\times H_{4}$, $[\alpha_6]\cong
[2]\times J^2([2]\times [3])$, $[\alpha_7]\cong J^3([2]\times [3])$.
$E_8$
-----
The Dynkin diagram is obtained from that of $E_7$ by adding $\alpha_8$ adjacent to $\alpha_7$. Then $[\alpha_3]\cong [2]\times H_6$, $[\alpha_4]\cong [2]\times
[3]\times [5]$, $[\alpha_5]\cong [4]\times H_4$, $[\alpha_6]\cong
[3]\times J^2([2]\times [3])$ and $[\alpha_7]\cong [2]\times
J^3([2]\times [3])$.
Exceptions to the pattern
--------------------------
There are seven such exceptions: $[\alpha_4]$ in $F_4$ (extra-special); $[\alpha_2]$ in $E_6$ (extra-special); $[\alpha_1]$ in $E_7$ (extra-special), $[\alpha_2]$ in $E_7$; $[\alpha_1]$, $[\alpha_2]$ and $[\alpha_8]$ in $E_8$. We present the Hasse diagrams for two of them in Fig. 3 and Fig. 4. Note that each $\alpha_i$ is an ending point in the Dynkin diagram.
A proof of Panyushev’s ${\mathcal{M}}$-polynomial conjecture
============================================================
In this section, we will prove Theorem \[thm-M-poly-main\]. Note that if holds for $[\alpha_i]$, namely, $[\alpha_i]\cong [k]\times P$ for some minuscule poset $P$, then $P$ is Gaussian by Theorem \[thm-Proctor\]. Thus Remark \[rmk-thm-Proctor\] allows us to conclude that $[k] \times P$ is pleasant, as desired. This finishes the proof of Theorem \[thm-M-poly-main\] for those $[\alpha_i]$ bearing the pattern . Now it remains to check the four non-extra-special posets in Section 4.10.
$E_7$
-----
Using `Mathematica`, one can verify that $${\mathcal{M}}_{[\alpha_2]}(t)=\frac{(1-t^8)(1-t^{10})(1-t^{11})(1-t^{12})(1-t^{14})}{(1-t)(1-t^3) (1-t^4)
(1-t^5)(1-t^7)}.$$ Thus $[\alpha_2]$ is pleasant.
The RHS of for $[\alpha_2]\times [6]$ is not a polynomial since the RHS of for it is not an integer. Thus $[\alpha_2]\times [6]$ is not pleasant, and $[\alpha_2]$ is *not* Gaussian.
$E_8$
-----
Using `Mathematica`, one can verify that $$\begin{aligned}
{\mathcal{M}}_{[\alpha_1]}(t)&=\frac{(1 - t^{14}) (1 - t^{17}) (1 - t^{18}) (1 - t^{20}) (1 - t^{23})}{(1 - t) (1 -t^4) (1 - t^6) (1 - t^7) (1 - t^{10})}; \\
{\mathcal{M}}_{[\alpha_2]}(t)&=\frac{(1 - t^{11}) (1 - t^{12}) (1 - t^{13}) (1 - t^{14}) (1 - t^{15}) (1 -
t^{17})}{(1 - t) (1 - t^3) (1 - t^4) (1 - t^5) (1 - t^6) (1 - t^7)}; \\
{\mathcal{M}}_{[\alpha_8]}(t)&= \frac{(1 - t^{20}) (1 - t^{24}) (1 - t^{29})}{(1 - t) (1 - t^6) (1 - t^{10})}.\end{aligned}$$ Thus every poset is pleasant, and the $E_8$ case is finished.
Similar to the previous remark, one can show that none of $[\alpha_1]$, $[\alpha_2]$, $[\alpha_8]$ is Gaussian. Moreover, none of the three extra-special posets in Section 4.10 is Gaussian.
The number ${\mathcal{M}}_{\Delta(1)}(-1)$
==========================================
This section is devoted to proving Theorems \[thm-M-1-main\] and \[thm-fixed-point-main\]. We continue to let ${\mathfrak{g}}$ be a finite-dimensional simple Lie algebra over ${\mathbb{C}}$, and label the simple roots as in Section 4. Let $w_0$ be the longest element of the Weyl group $W=W({\mathfrak{g}}, {\mathfrak{h}})$ of ${\mathfrak{g}}$. The following result is well-known.
\[lemma-w0\]
- $w_0=-1$ in $A_1$, $B_n$, $C_n$, $D_{2n}$, $E_7$, $E_8$, $F_4$ and $G_2$.
- In $A_{n-1}$, $w_0(\alpha_i)=-\alpha_{n-i}$, $1\leq i\leq n-1$.
- In $D_{2n-1}$, $-w_0$ interchanges $\alpha_{2n-2}$ and $\alpha_{2n-1}$, while preserves other simple roots.
- In $E_{6}$, $-w_0$ interchanges $\alpha_1$ and $\alpha_6$, $\alpha_3$ and $\alpha_5$, while preserves $\alpha_2$ and $\alpha_4$.
Let $\varpi_i$ be a minuscule fundamental weight, and let $P_{\varpi_i}$ be the corresponding minuscule poset. Recall that $w_0\in W$ acts as an order-reversing involution on the weight poset $\Lambda_{\varpi_i}\cong J(P_{\varpi_i})$. As on p. 479 of [@Stem], this involution transfers to an order-reversing involution on the poset $P_{\varpi_i}$. Indeed, for every $x\in P_{\varpi_i}$, the lower order ideals $I_{\leq x}=\{y\in P_{\varpi_i}\mid y\leq x\}$ and $I_{< x}=\{y\in P_{\varpi_i}\mid y< x\}$ have the property that $I_{< x}^{c}- I_{\leq x}^{c}=\{x^{\prime}\}$ for some $x^{\prime}\in P_{\varpi_i}$. Then one can easily check that $x\mapsto x^{\prime}$ is indeed an order-reversing involution on $P_{\varpi_i}$. We denote the corresponding complemented poset by $(P_{\varpi_i}, c)$. That is, $J(P_{\varpi_i},
c)$ and $(\Lambda_{\varpi_i}, w_0)$ are isomorphic as complemented posets. Now let us recall Theorem 4.1 of [@Stem].
\[thm-Stembridge\] *(**Stembridge**)* Let $(P_{\varpi_i}, c)$ be a complemented minuscule poset as above. Then ${\mathcal{M}}_{[m]\times P_{\varpi_i}}(-1)$ is the number of self-complementary lower ideals of $[m]\times P_{\varpi_i}$, or equivalently, the number of multi-chains $I_m\subseteq \cdots
\subseteq I_1$ ($I_j\in J(P_{\varpi_i})$) such that $I_j^c=I_{m+1-j}$ (see Lemma \[lemma-ideals-CnP\]). Here recall that $I_j^c:=P_{\varpi_i}\setminus \{c(x)\mid x\in I_j\}$.
The following lemma gives the order-reversing involution $c$ on $P_{\varpi_i}$ explicitly.
\[lemma-uniform-transfer\] In the setting of , denote by $w_0^i$ the longest element of the Weyl group of ${\mathfrak{g}}(0)$ in the $1$-standard ${\mathbb{Z}}$-grading such that $\Pi(1)=\{\alpha_i\}$. Then we have $$\label{minu-transfer}
(\Lambda_{\varpi_i}, w_0)\cong J([\alpha_i^{\vee}], w_0^i).$$
Since we need to pass to Lie subalgebras of ${\mathfrak{g}}$ frequently, let us explicitly give the types to avoid confusion. For instance, $[\alpha_k](A_{n})$ means the $[\alpha_k]$ in ${\mathfrak{g}}$ of type $A_{n}$.
Firstly, note that $[\alpha_k](A_{n})\cong [k]\times [n+1-k]$. Moreover, the order-reversing involution induced by $w_0^k(A_n)$ on $[k]\times [n+1-k]$ is the one sending $(i, j)$ to $(k+1-i, n+2-k-j)$. By Example 4.2 of [@Stem], we have $$\label{transfer-A}
(\Lambda_{\varpi_k}(A_n), w_0(A_n))\cong J([\alpha_k](A_{n}),
w_0^k(A_{n})).$$
Note that $\Lambda_{\varpi_n}(B_n)\cong J(H_n)$, see Example 3.3 of [@Stem]. On the other hand, $[\alpha_n](C_n)\cong H_n$ has only one order-reversing inclusion $(i, j)\mapsto (n+1-j, n+1-i)$, see Example 4.3 of [@Stem]. Thus we must have $$\label{transfer-B}
(\Lambda_{\varpi_n}(B_{n}), w_0(B_n)) \cong J([\alpha_n](C_n), w_0^n(C_{n})).$$ Similarly, one sees that holds for $(\Lambda_{\varpi_{n-1}}, w_0(D_n))$ and $(\Lambda_{\varpi_{n}}, w_0(D_n))$.
Note that $[\alpha_1](B_n)\cong [2n-1]$ has a unique order-reversing involution. Thus we must have $$\label{transfer-C} (\Lambda_{\varpi_1}(C_{n}),
w_0(C_n)) \cong J([\alpha_1](B_n), w_0^1(B_{n})).$$
Now let us prove that for $n\geq 4$, we have $$\label{transfer-D}
(\Lambda_{\varpi_1}(D_n), w_0(D_n))\cong J([\alpha_1](D_n),
w_0^1(D_{n})).$$ Note first that $[\alpha_1](D_n)\cong K_{n-2}$ and $J(K_{n-2})=K_{n-1}$. Moreover, $K_n$ has exactly two order-reversing involutions, one has two fixed points, while the other has none. Now let us proceed according to two cases.
- $n$ is even. Then the $w_0^1(D_{n})$ action on $[\alpha_1](D_n)$ has two fixed points, while the complemented poset $J([\alpha_1](D_n),
w_0(D_{n-1}))$ has none. On the other hand, the first fundamental weight in $D_{n}$ is $e_1$. Since $w_0(D_{n})=-1$, one sees that $(\Lambda_{\varpi_1}(D_n), w_0(D_n))$ has no fixed point as well. Thus holds. Here in the special case that $n=4$, we interpret $D_3$ as $A_3$.
- $n$ is odd. Then the $w_0^1(D_{n})$ action on $[\alpha_1](D_n)$ has no fixed point, while the complemented poset $J([\alpha_1](D_n),
w_0^1(D_{n}))$ has two. On the other hand, using Lemma \[lemma-w0\](c), one sees that $(\Lambda_{\varpi_1}(D_n),
w_0(D_n))$ also has two fixed points. Thus holds.
To sum up, is always true.
Now let us prove that $$\label{transfer-E6}
(\Lambda_{\varpi_1}(E_6), w_0(E_6)) \cong J([\alpha_1](E_6),
w_0^1(E_6)).$$ Note that on one hand in the graded poset $\Lambda_{\varpi_1}(E_6)$ (see the right one of Fig. 2), the middle level consists of three elements with rank $9$: $$s_1 s_3 s_4 s_5 s_2 s_4 s_3 s_1(\varpi_1), s_3 s_4 s_2 s_6 s_5 s_4 s_3 s_1(\varpi_1), s_5 s_4 s_2 s_6 s_5 s_4 s_3 s_1(\varpi_1).$$ All of them are fixed by $w_0(E_6)$. On the other hand, one can check that there are three lower ideals in $([\alpha_1](E_6), w_0^1(E_6))$ with size $8$, and they are all fixed points in $J([\alpha_1](E_6), w_0^1(E_6))$. Then follows directly.
Finally, we mention that $$\label{transfer-E7}
(\Lambda_{\varpi_1}(E_7), w_0(E_7))\cong J([\alpha_1](E_7),
w_0^1(E_7)).$$ We note that Fig. 1 (right) of [@Pr] gives the structure $\Lambda_{\varpi_1}(E_7)$, based on which one can figure out the structure of $(\Lambda_{\varpi_1}(E_7), w_0(E_7))$. In particular there is a unique cube. On the other hand, recall that $[\alpha_1](E_7)\cong J^3([2]\times [3])$. Then one can determine the structure of $([\alpha_1](E_7), w_0^1(E_7))$. Passing to $J([\alpha_1](E_7), w_0^1(E_7))$, we will also get a unique cube. By matching the patterns around the two cubes, one will obtain . We omit the details.
Note that $([\alpha_1](A_n), w_0^1(A_{n}))$ is just the poset $([n],
\leq)$ equipped with the order-reversing involution $j\mapsto
n+1-j$. For simplicity, sometimes we just denote the complemented minuscule poset $([\alpha_1](A_n), w_0^1(A_{n}))$ by $[n]$ instead. Now we are ready to prove Theorem \[thm-M-1-main\].
*Proof of Theorem \[thm-M-1-main\].* Firstly, let us handle those $[\alpha_i]$ bearing the pattern . Similar to Section 4, our discussion is case-by-case, yet the method is the same.
By Lemma \[lemma-uniform-transfer\] and Theorem \[thm-Stembridge\], if the fundamental weight $\varpi_i$ is minuscule, then Theorem \[thm-M-1-main\] holds for $[\alpha_i^{\vee}]$ in $(\Delta^{\vee})^+$.
Let us investigate $[\alpha_k](D_n)$ for $2\leq
k\leq n-3$ in details. Since now $$w_0^k(D_n)=w_0(A_{k-1}) w_0(D_{n-k})$$ (again $A_3$ is viewed as $D_3$) and the two factors commute, we have that $$([\alpha_k](D_{n}), w_0^k(D_n))\cong [k]\times ([\alpha_1](D_{n-k+1}), w_0^1(D_{n-k+1})).$$ Now by applying to $D_{n-k+1}$ and using Theorem \[thm-Stembridge\], one sees that Theorem \[thm-M-1-main\] holds for $[\alpha_k](D_n)$. For some other cases, we list the substitutes for as follows:
- $[\alpha_k](B_n)$, use $J([\alpha_1](B_n), w_0^1(B_{n}))\cong (\Lambda_{\varpi_1}(A_{2n-1}), w_0(A_{2n-1}))$;
- $[\alpha_k](C_n)$, use $J([\alpha_1](C_n), w_0^1(C_{n}))\cong (\Lambda_{\varpi_1}(A_{2n-2}), w_0(A_{2n-2}))$;
- $[\alpha_i]$ where $\alpha_i$ is a branching point, use ;
- $[\alpha_6](E_7)$, use ;
- $[\alpha_7](E_8)$, use .
Secondly, we have used `Mathematica` to check Theorem \[thm-M-1-main\] for the seven posets in Section 4.10.
Before proving Theorem \[thm-fixed-point-main\], we prepare the following.
\[lemma-fixed-pt\] If the $w_0^i$ action on $[\alpha_i]({\mathfrak{g}})$ has fixed point(s), then $([\alpha_i], w_0^i)$ has no self-complementary lower ideal.
Let $I$ be any lower ideal of $[\alpha_i]$. Note that $I$ is self-complementary if and only if $[\alpha_i]\setminus I$ is self-complementary. Indeed, $$\begin{aligned}
I = [\alpha_i]\setminus w_0^i(I) &\Leftrightarrow I \sqcup
w_0^i(I)=[\alpha_i]\\
&\Leftrightarrow ([\alpha_i]\setminus I) \sqcup
w_0^i([\alpha_i]\setminus I)=[\alpha_i] \\
&\Leftrightarrow [\alpha_i]\setminus I = [\alpha_i]\setminus
w_0^i([\alpha_i]\setminus I).\end{aligned}$$ Now suppose that there exists $\gamma\in[\alpha_i]$ such that $w_0^i\gamma=\gamma$. If there was a self-complementary lower ideal $I$ of $[\alpha_i]$, let us deduce contradiction. Indeed, if $\gamma\in I$, then $\gamma=w_0^i(\gamma)\in w_0^{i}(I)$. This contradicts to the assumption that $I$ is self-complementary. If $\gamma\in [\alpha_i]\setminus I$, then one would also get a contradiction since the upper ideal $[\alpha_i]\setminus I$ is self-complementary as well.
Finally, let us deduce Theorem \[thm-fixed-point-main\].
*Proof of Theorem \[thm-fixed-point-main\].* If the $w_0^i$ action on $[\alpha_i]$ has fixed point(s), then $([\alpha_i], w_0^i)$ has no self-complementary lower ideal by Lemma \[lemma-fixed-pt\]. Thus ${\mathcal{M}}_{[\alpha_i]}(-1)=0$ by Theorem \[thm-M-1-main\].
Conversely, if ${\mathcal{M}}_{[\alpha_i]}(-1)=0$, then we need to exhibit the fixed points of the $w_0^i$ action on $[\alpha_i]$. We note that ${\mathcal{M}}_{[\alpha_i]}(-1)=0$ in the following cases:
- $[\alpha_{i}](A_{2n+1})$ for those odd $i$ between $1$ and $2n+1$;
- $[\alpha_{i}](B_{n})$ for $i$ odd;
- $[\alpha_{n}](C_{n})$; $[\alpha_{n-1}](D_{n})$ and $[\alpha_{n}](D_{n})$;
- $[\alpha_{i}](D_{2n})$ for those odd $i$ between $1$ and $2n-3$;
- $[\alpha_2](E_7)$ and $[\alpha_7](E_7)$.
In the classical types, aided by Lemma \[lemma-w0\], one can identify the fixed points easily. We provide the fixed points for the last two cases. $$\begin{aligned}
[\alpha_2](E_7): [1, 1, 1, 2, 1, 1, 0], [1, 1, 2, 2, 1, 0, 0], [0,
1, 1, 2, 1, 1, 1]; \\
[\alpha_7](E_7): [1, 1, 2, 2, 1, 1, 1], [1, 1, 1,
2, 2, 1, 1], [0, 1, 1, 2, 2, 2, 1],\end{aligned}$$ where the roots are expressed in terms of the simple ones.
A proof of Panyushev’s ${\mathcal{N}}$-polynomial conjecture
============================================================
This section is devoted to proving Theorem \[thm-N-poly-main\]. Note that (a) trivially implies (b) since the constant term of any ${\mathcal{N}}$ polynomial is always $1$, while (c) is just a restatement of (b). Since $\Delta(1)$ is Sperner by Lemma 2.6 of [@P] and each rank level $\Delta(1)_i$ is an antichain, one sees that (c) implies (d). Therefore, it remains to show that (d) implies (a). This will be carried out in the remaining part of this section. Firstly, let us prepare the following.
\[lemma-N-poly-known\] We have that
- ${\mathcal{N}}_{[n]\times [m]}(t)=\sum_{i\geq 0}{n \choose i}{m\choose
i}t^i$. The poset $[n]\times [m]$ has a unique rank level of maximal size if and only if $m=n$, if and only if ${\mathcal{N}}_{[n]\times [m]}(t)$ is palindromic.
- ${\mathcal{N}}_{H_n}(t)=\sum_{i\geq 0}{n+1 \choose 2i}t^i$. The poset $H_n$ has a unique rank level of maximal size if and only if $n$ is odd, if and only if ${\mathcal{N}}_{H_n}(t)$ is palindromic.
- The following posets have at least two rank levels of maximal size: $[2]\times H_4$, $[2]\times
[3]\times [3]$; $[2]\times H_5$, $[2]\times
J^2([2]\times [3])$; $[2]\times H_6$, $[4]\times H_4$, $[2]\times [3]\times [5]$, $[3]\times J^2([2]\times
[3])$, $[2]\times J^3([2]\times [3])$.
- The following posets have a unique rank level of maximal size: $[2]\times
[3]\times [4]$, $[3]\times H_4$. Moreover, their ${\mathcal{N}}$ polynomials are palindromic.
Part (a) follows directly from Lemma \[lemma-antichain-CnP\], see also item 1 on p. 1201 of [@P]. Part (b) is item 2 on p. 1201 of [@P]. With the help of Lemma \[lemma-antichain-CnP\], one easily verifies part (c) and the first statement of part (d). For the second statement of (d), we mention that $$\begin{aligned}
{\mathcal{N}}_{[2]\times
[3]\times [4]}(t) &=1+ 24t + 120 t^2+ 200 t^3 + 120 t^4 + 24 t^5 +t^6,\\
{\mathcal{N}}_{[3]\times H_4}(t) &=1+ 30t + 165 t^2+ 280 t^3 + 165 t^4 + 30
t^5 +t^6.\end{aligned}$$
Now let us investigate the ${\mathcal{N}}$-polynomial of $[m]\times K_n$. Suppose now we have exactly one ball labeled $i$ for $1\leq i \neq
n+1 \leq 2n+1$, and two *distinct* balls labeled $n+1$. We want to put them into $m$ boxes arranged from left to right so that there is at most one ball in each box with the only exception that the two balls labeled $n+1$ can be put in the same box, and that the relative order among the labels $1,2, \dots, 2n+1$ under $\leq$ are preserved when we read them off the balls from left to right. Let us denote by $A_{n, m}(i)$ the number of filling $i$ balls into the boxes so that the above requirements are met. By Lemma \[lemma-antichain-CnP\], one sees easily that $$\label{Anmi}
{\mathcal{N}}_{[m]\times K_n}(t)=\sum_{i=0}^{m+1} A_{n, m}(i) t^i.$$
\[thm-N-poly-CmKn\] The following are equivalent:
- The poset $[m]\times K_n$ has a unique rank level of maximal size;
- $m=1$ or $2n+1$;
- ${\mathcal{N}}_{[m]\times K_n}$ is monic;
- ${\mathcal{N}}_{[m]\times K_n}$ is palindromic.
The equivalence between (a) and (b) is elementary, while that between (b) and (c) follows from . Part (d) trivially implies (c). Now it remains to show that (b) implies (d). When $m$=1, we have $${\mathcal{N}}_{K_n}(t)=1+(2n+2)t+t^2,$$ which is palindromic. Now let us show that ${\mathcal{N}}_{[2n+1]\times K_n}$ is palindromic.
To obtain $A_{n, 2n+1}(1)$, we note there are two possibilities: neither of the two balls labeled $n+1$ is chosen; exactly one of the two balls labeled $n+1$ is chosen. This gives $$A_{n, 2n+1}(1)={2n\choose 1}{2n+1\choose 1}+{2\choose 1}{2n+1\choose 1}.$$ To obtain $A_{n, 2n+1}(2n+1)$, we note there are three possibilities: exactly one of the two balls labeled $n+1$ is chosen; both of the two balls labeled $n+1$ are chosen and they are put in the same box; both of the two balls labeled $n+1$ are chosen and they are put in different boxes. This gives $$A_{n, 2n+1}(2n+1)={2\choose 1}{2n+1\choose 2n+1}+{2n \choose 2n-1}{2n+1\choose 2n}+2{2n \choose 2n-1}{2n+1\choose 2n+1}.$$ One sees that $A_{n, 2n+1}(1)=A_{n, 2n+1}(2n+1)$.
Now let $2\leq i\leq n$. To obtain $A_{n, 2n+1}(i)$, we note there are four possibilities: neither of the two balls labeled $n+1$ is chosen; exactly one of the two balls labeled $n+1$ is chosen; both of the two balls labeled $n+1$ are chosen and they are put in the same box; both of the two balls labeled $n+1$ are chosen and they are put in different boxes. Therefore $A_{n, 2n+1}(i)$ is equal to $${2n\choose i}{2n+1\choose i}+{2\choose 1}{2n \choose i-1}{2n+1\choose i}+{2n \choose i-2}{2n+1\choose i-1}+2{2n \choose i-2}{2n+1\choose i}.$$ Thus $$A_{n, 2n+1}(i)={2n \choose i-2}{2n+1\choose i-1}+{2n\choose i}{2n+1\choose i}+2{2n+1 \choose i-1}{2n+1\choose i}.$$ Substituting $i$ by $2n+2-i$ in the above formula gives $$A_{n, 2n+1}(2n+2-i)={2n \choose i}{2n+1\choose i}+{2n\choose i-2}{2n+1\choose i-1}+2{2n+1 \choose i}{2n+1\choose i-1}.$$ Thus $A_{n, 2n+1}(i)=A_{n, 2n+1}(2n+2-i)$ for $2\leq i\leq n$.
To sum up, we have shown that ${\mathcal{N}}_{[2n+1]\times K_n}$ is palindromic. This finishes the proof.
Now we are ready to prove Panyushev’s ${\mathcal{N}}$-polynomial conjecture.
*Proof of Theorem \[thm-N-poly-main\].* As noted in the introduction, it remains to show that if $\Delta(1)$ has a unique rank level of maximal size, then ${\mathcal{N}}_{\Delta(1)}(t)$ is palindromic. By Lemma \[lemma-N-poly-known\] and Theorem \[thm-N-poly-CmKn\], Theorem \[thm-N-poly-main\] holds for those non-abelian or non-extra-special $[\alpha_i]$’s bearing the pattern . Now it remains to handle the four non-extra-special posets in Section 4.10.
For $E_7$, it remains to consider $[\alpha_2]$. Indeed, it has a unique rank level of maximal size. Moreover, using `Mathematica`, we obtain that $${\mathcal{N}}_{[\alpha_2]}(t) =1+ 35t + 140 t^2+ 140 t^3 + 35 t^4 + t^5,$$ which is palindromic.
For $E_8$, it remains to consider $[\alpha_1]$, $[\alpha_2]$, and $[\alpha_8]$. Indeed, each of them has more than one rank level of maximal size. This finishes the proof.
We mention that in $E_8$, one has $$\begin{aligned}
{\mathcal{N}}_{[\alpha_1]}(t)&=1+ 64t + 364 t^2+ 520 t^3 + 208 t^4 + 16
t^5,\\
{\mathcal{N}}_{[\alpha_2]}(t)&=1+ 56t + 420 t^2+ 952 t^3 + 770 t^4 + 216 t^5
+16 t^6,\\
{\mathcal{N}}_{[\alpha_8]}(t)&=1+ 56t + 133 t^2+ 42 t^3.\end{aligned}$$
Structure of the Panyushev orbits
=================================
This section is devoted to investigating the structure of the Panyushev orbits. To be more precise, we shall establish Theorems \[thm-main-reverse-operator-orbit\] and \[thm-main-cyclic-sieving\].
*Proof of Theorem \[thm-main-reverse-operator-orbit\].* We keep the notation of Section 2. In particular, $L_i$’s are the full rank lower ideals of $P$, and $(I_1, I_2)$, where $I_i\in J(P)$ and $I_2\subseteq I_1$, stands for a general lower ideal of $[2]\times P$. Recall that $\mathfrak{X}$ acts on lower ideals as well.
Note that when ${\mathfrak{g}}$ is $A_n$, the extra-special $\Delta(1)\cong [n-1]\sqcup
[n-1]$; when ${\mathfrak{g}}$ is $C_n$, the extra-special $\Delta(1)\cong
[2n-2]$. One can verify Theorem \[thm-main-reverse-operator-orbit\] for these two cases without much effort. We omit the details.
For ${\mathfrak{g}}=B_n$, the extra-special $\Delta(1)= [2]\times [2n-3]$. Now $|\Pi_{l}|=n-1$, $h-1=2n-1$, and $h^*-2=2n-3$. As in Section 2, let $L_i$ ($0\leq i\leq 2n-3$) be the rank level lower ideals. For simplicity, we denote $\mathfrak{X}_{[2]\times [2n-3]}$ by $\mathfrak{X}$. For any $0\leq i\leq n-2$, let us analyze the type I $\mathfrak{X}$-orbit $\mathcal{O}(L_i, L_i)$ via the aid of Lemma \[lemma-operator-ideals-CmP\]: $$\begin{aligned}
\mathfrak{X}(L_i, L_i)&=(L_{i+1}, L_0),\\
\mathfrak{X}^{2n-4-i}(L_{i+1}, L_0)&=(L_{2n-3}, L_{2n-4-i}),\\
\mathfrak{X}(L_{2n-3}, L_{2n-4-i})&=(L_{2n-3-i}, L_{2n-3-i}),\\
\mathfrak{X}(L_{2n-3-i}, L_{2n-3-i})&=(L_{2n-2-i}, L_{0}),\\
\mathfrak{X}^{i-1}(L_{2n-2-i}, L_{0})&=(L_{2n-3}, L_{i-1}),\\
\mathfrak{X}(L_{2n-3}, L_{i-1})&=(L_{i}, L_{i}).\end{aligned}$$ Thus $\mathcal{O}(L_i, L_i)$ consists of $2n-1$ elements. Moreover, in this orbit, $(L_{2n-2-\frac{i+1}{2}}, L_{\frac{i-1}{2}})$ (resp. $(L_{n+\frac{i}{2}-1}, L_{n-\frac{i}{2}-2})$) is the unique lower ideal with size $2n-3$ when $i$ is odd (resp. even). Since there are $(n-1)(2n-1)$ lower ideals in $[2]\times [2n-3]$ by Corollary \[cor-thm-M-poly-main\], one sees that all the $\mathfrak{X}$-orbits have been exhausted, and Theorem \[thm-main-reverse-operator-orbit\] holds for $B_{n}$.
Let us consider $D_{n+2}$, where the extra-special $\Delta(1)\cong
[2]\times K_{n-1}$. We adopt the notation as in Section 2. For simplicity, we write $\mathfrak{X}_{[2]\times K_{n-1}}$ by $\mathfrak{X}$. We propose the following.
**Claim.** $\mathcal{O}(L_i, L_i)$, $0\leq i\leq n-1$, $\mathcal{O}(I_n, I_n)$, and $\mathcal{O}(I_{n^{\prime}},
I_{n^{\prime}})$ exhaust the orbits of $\mathfrak{X}$ on $[2]\times
K_{n-1}$. Moreover, each orbit has size $2n+1$ and contains a unique lower ideal with size $2n$.
Indeed, firstly, for any $0\leq i\leq n-1$, observe that by Lemma \[lemma-operator-ideals-CmP\], we have $$\begin{aligned}
\mathfrak{X}(L_i, L_i)&=(L_{i+1}, I_0),\\
\mathfrak{X}^{2n-i-2}(L_{i+1}, L_0)&=(L_{2n-1}, L_{2n-i-2}),\\
\mathfrak{X}(L_{2n-1}, L_{2n-i-2})&=(L_{2n-i-1}, L_{2n-i-1}),\\
\mathfrak{X}(L_{2n-i-1}, L_{2n-i-1})&=(L_{2n-i}, L_{0}),\\
\mathfrak{X}^{i-1}(L_{2n-i}, L_{0})&=(L_{2n-1}, L_{i-1}),\\
\mathfrak{X}(L_{2n-1}, L_{i-1})&=(L_{i}, L_{i}).\end{aligned}$$ Thus the type I orbit $\mathcal{O}(L_i, L_i)$ consists of $2n+1$ elements. Moreover, in this orbit, $(L_{2n-i+\frac{i-1}{2}},
L_{\frac{i-1}{2}})$ (resp. $(L_{n+\frac{i}{2}},
L_{n-\frac{i}{2}-1})$) is the unique lower ideal with size $2n$ when $i$ is odd (resp. even).
Secondly, assume that $n$ is even and let us analyze the orbit $\mathcal{O}(I_n, I_n)$. Indeed, by Lemma \[lemma-operator-ideals-CmK\], we have $$\begin{aligned}
\mathfrak{X}(I_n, I_n)&=(I_{n^{\prime}}, L_0),\\
\mathfrak{X}^{n-1}(I_{n^{\prime}}, L_0)&=(I_{n}, L_{n-1}),\\
\mathfrak{X}(I_{n}, L_{n-1})&=(L_{n}, I_{n}),\\
\mathfrak{X}^{n-1}(L_{n}, I_{n})&=(L_{2n-1}, I_{n^{\prime}}),\\
\mathfrak{X}(L_{2n-1}, I_{n^{\prime}})&=(I_{n}, I_{n}).\end{aligned}$$ Thus the type II orbit $\mathcal{O}(I_n, I_n)$ consists of $2n+1$ elements. Moreover, in this orbit, $(I_n, I_n)$ is the unique ideal with size $2n$. The analysis of the orbit $\mathcal{O}(I_{n^{\prime}}, I_{n^{\prime}})$ is entirely similar.
Finally, assume that $n$ is odd and let us analyze the orbit $\mathcal{O}(I_n, I_n)$. Indeed, by Lemma \[lemma-operator-ideals-CmK\], we have $$\begin{aligned}
\mathfrak{X}(I_n, I_n)&=(I_{n^{\prime}}, L_0),\\
\mathfrak{X}^{n-1}(I_{n^{\prime}}, L_0)&=(I_{n^{\prime}}, L_{n-1}),\\
\mathfrak{X}(I_{n^{\prime}}, L_{n-1})&=(L_{n}, I_{n^{\prime}}),\\
\mathfrak{X}^{n-1}(L_{n}, I_{n^{\prime}})&=(L_{2n-1}, I_{n^{\prime}}),\\
\mathfrak{X}(L_{2n-1}, I_{n^{\prime}})&=(I_{n}, I_{n}).\end{aligned}$$ Thus the type II orbit $\mathcal{O}(I_n, I_n)$ consists of $2n+1$ elements. Moreover, in this orbit, $(I_n, I_n)$ is the unique ideal with size $2n$. The analysis of the orbit $\mathcal{O}(I_{n^{\prime}}, I_{n^{\prime}})$ is entirely similar.
To sum up, we have verified the claim since there are $(n+2)(2n+1)$ lower ideals in $[2]\times K_{n-1}$ by Corollary \[cor-thm-M-poly-main\]. Note that $|\Pi_{l}|=n+2$, $h=h^*=2n+2$ for ${\mathfrak{g}}=D_{n+2}$, one sees that Theorem \[thm-main-reverse-operator-orbit\] holds for $D_{n+2}$.
Theorem \[thm-main-reverse-operator-orbit\] has been verified for all exceptional Lie algebras using `Mathematica`. We only present the details for $E_6$, where $\Delta(1)=[\alpha_2]$. Note that $|\Pi_l|=6$, $h-1=11$, $h^*-2=10$. On the other hand, $\mathfrak{X}$ has six orbits on $\Delta(1)$, each has $11$ elements. Moreover, the size of the lower ideals in each orbit is distributed as follows:
- $0, 1, 2, 4, 7, \textbf{10}, 13, 16, 18, 19, 20$;
- $3, 4, 5, 6, 9, \textbf{10}, 11, 14, 15, 16, 17$;
- $3, 4, 5, 6, 9, \textbf{10}, 11, 14, 15, 16, 17$;
- $7, 7, 8, 8, 9, \textbf{10}, 11, 12, 12, 13, 13$;
- $5, 6, 6, 8, 9, \textbf{10}, 11, 12, 14, 14, 15$;
- $7, 7, 8, 8, 9, \textbf{10}, 11, 12, 12, 13, 13$.
One sees that each orbit has a unique Lagrangian lower ideal.
*Proof of Theorem \[thm-main-cyclic-sieving\].* Based on the structure of $\Delta(1)$ for $1$-standard ${\mathbb{Z}}$-gradings of ${\mathfrak{g}}$ in Section 3, the desired CSP for the triple $(\Delta(1)$, ${\mathcal{M}}_{\Delta(1)}(t)$, $\langle\mathfrak{X}_{\Delta(1)}\rangle)$ follows from Theorems 1.1, 1.2 and 10.1 of [@RS] combined with Table 1 at the root of the paper, where $n$ is the order of $\mathfrak{X}_{\Delta(1)}$, $[n]_t:= \frac{1-t^n}{1-t}$ and has been applied.
[|c|c|c|c|]{}\
$\Delta(1)$ & $n$ & orbits (size $\times$ number) & ${\mathcal{M}}_{\Delta(1)}(t) \mod (t^n-1)$\
$[\alpha_4](F_4)$ & 11& $11 \times 2$ & $2\times [11]_{t}$\
$[\alpha_2](E_6)$ & 11& $11 \times 6$ & $6 \times [11]_{t}$\
$[\alpha_1](E_7)$ & 17& $17 \times 7$ & $7 \times [16]_{t}$\
$[\alpha_2](E_7)$ & 14& $14 \times 25+ 2\times 1$ & $1+t^7+ 25 \times [14]_{t}$\
$[\alpha_5](E_7)$ & 10& $10 \times 67+ 2\times 1$ & $1+t^5+ 67 \times [10]_{t}$\
$[\alpha_1](E_8)$ & 23& $23 \times 51$ & $51 \times [23]_{t}$\
$[\alpha_2](E_8)$ & 17& $17 \times 143$ & $143 \times[17]_{t}$\
$[\alpha_5](E_8)$ & 11& $11 \times 252$ & $252 \times [11]_{t}$\
$[\alpha_8](E_8)$ & 29& $29 \times 8$ & $8 \times [29]_{t}$\
<span style="font-variant:small-caps;">Acknowledgements</span>
The research is supported by the National Natural Science Foundation of China (grant no. 11571097) and the Fundamental Research Funds for the Central Universities. We thank Dr. Bai, Dr. Wang, and Prof. Stembridge for helpful discussions.
[99]{}
D. Armstrong, C. Stump, H. Thomas, *A uniform bijection between nonnesting and noncrossing partitions*, Tran. Amer. Math. Soc. **365** (2013), 4121–4151.
N. Bourbaki, *Lie groups and Lie algebras*, Chapters 4–6, Springer (2002). Originally published as *Groupes et algebres de Lie*, Paris: Hermann, 1968.
V. V. Gorbatsevich, A. L. Onishchik, E. B. Vinberg, *Lie groups and Lie algebras III* (Encyclopaedia Math. Sci., vol. 41), Berlin: Springer (1994).
A. Knapp, *Lie Groups, Beyond an Introduction*, Birkhäuser, 2nd Edition, 2002.
B. Kostant, *The principal three-dimensional subgroup and the Betti numbers of a complex simple Lie group*, Amer. J. Math. **81** (1959), 973–1032.
B. Kostant, *The set of Abelian ideals of a Borel subalgebra, Catran decompositions and Discrete Series Representations*, Int. Math. Res. Not. **5** (1998), 225–252.
I. Macdonald, *The Poincaré series of a Coxeter group*, Math. Ann. **199** (1972), 161–174.
D. I. Panyushev, *On orbits of antichains of positive roots*, Eur. J. Comb. **30** (2009), 586–594.
D. I. Panyushev, *Antichains in weight posets associated with gradings of simple Lie algebras*, Math. Z. **281** (2015), 1191–1214.
D. I. Panyushev, *Weight posets associated with gradings of simple Lie algebras, Weyl groups, and arrangements of hyperplanes*, J. Alg. Comb., published online with DOI 10.1007/s10801-016-0671-0.
R. A. Proctor, *Reprentations of $sl(2, {\mathbb{C}})$ on posets and the Sperner property*, SIAM. J. Alg. Disc. Meth. **3** (1982), 275–280.
R. A. Proctor, *Solution of two difficult combinatorial problems with linear algebra*, Amer. Math. Monthly **89** (1982), 721–734.
R. A. Proctor, *Bruhat lattices, plane partition generating functions, and minuscule representations*, Eur. J. Comb. **5** (1984), 331–350.
V. Reiner, D. Stanton, D. White, *The cyclic sieving phenomenon*, J. Comb. Theory A **108** (2004), 17–50.
C. M. Ringel, *The $(n-1)$-antichains in a root poset of width $n$*, preprint, arxiv: 1305.1593v1.
D. Rush, X. Shi, *On orbits of order ideals of minuscule posets*, J. Alg. Comb. **37** (2013), 545–569.
R. P. Stanley, *Weyl groups, the hard Lefschetz theorem, and the Sperner property*, SIAM. J. Alg. Disc. Meth. **1** (1980), 168–184.
R. P. Stanley, *Enumerative combinatorics*, Vol. 1, 2nd Edition, Cambridge University Press, 2012.
J. R. Stembridge, *On minuscule representations, plane partitions and involutions in complex Lie groups*, Duke. Math. J. **73** (1994), 469–490.
E. B. Vinberg, *The Weyl group of a graded Lie algebra*, Math. USSR-Izv. **10** (1976), 463–495.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Normalization techniques such as Batch Normalization have been applied successfully for training deep neural networks. Yet, despite its apparent empirical benefits, the reasons behind the success of Batch Normalization are mostly hypothetical. We here aim to provide a more thorough theoretical understanding from a classical optimization perspective. Our main contribution towards this goal is the identification of various problem instances in the realm of machine learning where Batch Normalization can provably accelerate optimization. We argue that this acceleration is due to the fact that Batch Normalization splits the optimization task into optimizing length and direction of the parameters separately. This allows gradient-based methods to leverage a favourable global structure in the loss landscape that we prove to exist in Learning Halfspace problems and neural network training with Gaussian inputs. We thereby turn Batch Normalization from an effective practical heuristic into a provably converging algorithm for these settings. Furthermore, we substantiate our analysis with empirical evidence that suggests the validity of our theoretical results in a broader context.'
bibliography:
- 'batch-norm.bib'
---
Appendix {#appendix .unnumbered}
========
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'A matrix is called acyclic if replacing the diagonal entries with $0$, and the nonzero diagonal entries with $1$, yields the adjacency matrix of a forest. In this paper we show that null space and the rank of a acyclic matrix with $0$ in the diagonal is obtained from the null space and the rank of the adjacency matrix of the forest by multipliying by non-singular diagonal matrices. We combine these methods with an algorithm for finding a sparsest basis of the null space of a forest to provide an optimal time algorithm for finding a sparsest basis of the null space of acyclic matrices with $0$ in the diagonal.'
author:
- |
Daniel A. Jaume[^1] and Adrián Pastine[^2]\
Universidad Nacional de San Luis\
Departamento de Matemática\
San Luis, Argentina
title: 'On the structure of the fundamental subspaces of (not necessarily symmetrical) tree-patterned matrices'
---
Introduction
============
Throughout this article, all graphs are assumed to be finite, undirected and without loops or multiple edges. The vertices of a graph $G$ are denoted by $V(G)$ and its edges by $E(G)$. We also assume that $\mathbb{F}$ denotes an arbitrary field. Following the notation in [@Mohammadian], we denote by $\mathcal{M}_{\mathbb{F}}(G)$ the set of all matrices $M$ over $\mathbb{F}$ with rows and columns indexed by $V(G)$, so that for every two distinct vertices $u,v\in V(G)$, the $(u,v)-$entry of $M$ is non-zero if and only if $\{u,v\}\in E(G)$. Notice that the diagonal entries are allowed to be non-zero. A matrix $M$ over $\mathbb{F}$ is said to be *acyclic* if $M\in \mathcal{M}_{\mathbb{F}}(F)$ for a forest $F$. If the forest $F$ is a tree, then $M$ is said to be *tree-patterned*.
Given a graph $G$, the adjacency matrix of $G$, denoted by $A(G)$, is a $(0,1)-$matrix in $\mathcal{M}_{\mathbb{F}}(G)$ with zero diagonal.
In [@Mohammadian], the following lemma was proved.
\[lemma symm\] Let $M$ be an acyclic matrix over a field $\mathbb{F}$. Then there exist a finite-dimensional field extension $\mathbb{E}$ of $\mathbb{F}$ and a diagonal matrix $D$ over $\mathbb{E}$ such that $D^{-1}MD$ is symmetric.
Due to a previous version of Lemma \[lemma symm\] (which first appeared in [@Parter; @Youngs]), most of the study of acyclic matrices was done on symmetric matrices. The matrices considered in this article do not need to be symmetric. This is done because the proofs work almost identically, and in this way there is no need to calculate the necessary diagonal matrix over the field extension.
The fundamental spaces of a matrix $M$ are the null space, ${\operatorname{\mathpzc{Null}}(M)}$, and the rank, ${\operatorname{\mathpzc{Rank}}(M)}$. The structure of the fundamental spaces of graph-patterned matrices has been studied in depth for symmetric tree-patterned matrices allowing non-zero entries in the diagonal, see for instance [@Fiedler; @Nylen; @Parter; @Wiener]. Most of these papers deal with the dimension of the null space, but none of them give a basis for it.
Given $M\in\mathcal{M}_{\mathbb{F}}(G)$, the *null support* of $M$, denoted ${\operatorname{\mathpzc{Supp}}(M)}$, is the set of vertices of $G$ that have non-zero entries in at least one vector from ${\operatorname{\mathpzc{Null}}(M)}$. In other words, $v\in {\operatorname{\mathpzc{Supp}}(M)}$ if there is a vector $\overrightarrow{x}\in{\operatorname{\mathpzc{Null}}(M)}$ with $\overrightarrow{x}_v\neq 0$, where $\overrightarrow{x}_v$ is the coordinate of $\overrightarrow{x}$ corresponding to the vertex $v$. In [@JM] the null support of adjacency matrices of forests has been studied in depth. The authors provided a decomposition for any forest into an $S$-forest (the forests that have a unique maximum independent set), and an $N$-forest (the forests that have a unique maximum matching). They showed that all the information of the null space of $A(F)$ can be obtained from the $S$-forests and $N$-forests related to $F$. It was implicitly shown that ${\operatorname{\mathpzc{Null}}(A(F))}$ coincides with the intersection of all the maximum independent sets of $F$. In [@JMPS] an optimal time algorithm for finding a *sparsest* (i.e., has the fewest nonzeros) $\{-1,0,1\}$ basis for the null space of a forest has been found. It is important to notice that the problem of finding a sparsest basis of the null space of a matrix is an important problem for numerical applications, which is known to be NP-complete [@MR857589] and even hard to approximate [@MR3537025].
Let $\mathcal{M}_{\mathbb{F},0}(G)$ be the set of matrices in $\mathcal{M}_{\mathbb{F}}(G)$ with zero in the diagonal. In Section \[Section Null\] we show that given a forest $F$, the null space of any matrix in $\mathcal{M}_{\mathbb{F},0}(F)$ can be obtained by multiplying the null space of $F$ by a suitable non-singular diagonal matrix. This will allow the use of all the tools developed in [@JM; @JMPS] for the study of the null space of said matrices. In particular we use the results of [@JMPS] to give an optimal time algorithm for finding a sparsest basis of the null space of the matrix. In Section \[Section Rank\] we prove that the rank of any matrix in $\mathcal{M}_{\mathbb{F},0}(F)$ can be found by multiplying the rank of $F$ by a suitable non-singular diagonal matrix.
The restriction to having zero in the diagonal may seem strong, but several problems (chemistry, electric conductance, flow in networks, etc) can be modeled with this kind of matrices.
On the null space {#Section Null}
=================
The null space of a graph is the direct sum of the null spaces of its connected components. In a similar fashion, the null space of a matrix $M\in\mathcal{M}_{\mathbb{F}}(G)$ is the direct sum of the null spaces of $M$ over the connected components of $G$. Because of this, we study the null space of matrices over trees and obtain results for the null space of acyclic matrices.
Lemma \[suppindep\], which is fundamental for our results, first appeared in [@Mohammadian] as Theorem $8(i)$.
[@Mohammadian]\[suppindep\] Let $F$ be a forest, and $M\in \mathcal{M}_{\mathbb{F},0}(F)$. If $\{v,w\}\in {\operatorname{\mathpzc{Supp}}(M)}$ then $\{v,w\}\not \in E(F)$.
Let $T$ be a tree, $M\in \mathcal{M}_{\mathbb{F},0}(T)$, and $v$ be a vertex of $T$. For each vertex $w\in T$ let $vPw$ be the unique directed path from $v$ to $w$ in $T$. In this sense $vPw$ and $wPv$ are different, because we care about the direction. We define the *$v$-scalation of $M$* as the non-singular diagonal matrix with $$D^{(M,v)}_{w,w}=\prod_{\substack{u\in{\operatorname{\mathpzc{Supp}}(M)},\\ (u,t)\in vPw}}M_{u,t}^{-1}\prod_{\substack{u\in {\operatorname{\mathpzc{Supp}}(M)},\\ (t,u)\in vPw}}M_{t,u}.$$
The following lemma follows from the definition of $D^{(M,v)}$ and Lemma \[suppindep\]
\[cuentasD\] Let $T$ be a tree, $M\in \mathcal{M}_{\mathbb{F},0}(F)$, $u,v\in V(T)$, $vPu=(v=v_0,v_1,\ldots,u_1,u)$ and $vPu_2=(v=v_0,v_1,\ldots,u_1,u,u_2)$ two directed paths in $T$. The following statements are true.
i) If $u_1,u_2\in {\operatorname{\mathpzc{Supp}}(M)}$, $$D^{(M,v)}_{u_2,u_2}=D^{(M,v)}_{u_1,u_1}*M_{u,u_1}^{-1}*M_{u,u_2},$$
ii) if $u_1\in {\operatorname{\mathpzc{Supp}}(M)}$, $u_2\not\in {\operatorname{\mathpzc{Supp}}(M)}$, $$D^{(M,v)}_{u_2,u_2}=D^{(M,v)}_{u_1,u_1}*M_{u,u_1}^{-1},$$
iii) if $u_1\not\in {\operatorname{\mathpzc{Supp}}(M)}$, $u_2\in {\operatorname{\mathpzc{Supp}}(M)}$, $$D^{(M,v)}_{u_2,u_2}=D^{(M,v)}_{u_1,u_1}*M_{u,u_2},$$
iv) if $u_1,u,u_2\not\in {\operatorname{\mathpzc{Supp}}(M)}$, $$D^{(M,v)}_{u_2,u_2}=D^{(M,v)}_{u_1,u_1},$$
v) if $u\in {\operatorname{\mathpzc{Supp}}(M)}$, $$D^{(M,v)}_{u_2,u_2}=D^{(M,v)}_{u_1,u_1}*M_{u_1,u}*M_{u,u_2}^{-1}.$$
As a direct consequence of Lemma \[cuentasD\], the matrices $D^{(M,v)}$ and $\left(D^{(M,v)}\right)^{-1}$ can be obtained in linear time over the number of vertices, as at most one multiplication must be done at each vertex.
\[main\] Given a tree $T$, a matrix $M\in \mathcal{M}_{\mathbb{F},0}(T)$ and a vertex $v\in{\operatorname{\mathpzc{Supp}}(M)}$, a vector $\overrightarrow{x}$ is in ${\operatorname{\mathpzc{Null}}(M)}$ if and only if $D^{(M,v)}\overrightarrow{x}$ is in ${\operatorname{\mathpzc{Null}}(A(T))}$.
Suppose $\overrightarrow{x}=(x_1,\ldots,x_n)\in {\operatorname{\mathpzc{Null}}(M)}$, and let $v\in{\operatorname{\mathpzc{Supp}}(M)}$.
We want to show $A(T)D^{(M,v)}\overrightarrow{x}=0$. Consider $\left(A(T)D^{(M,v)}\overrightarrow{x}\right)_{u}$. If $\overrightarrow{x}_{w}=0$ for all $w\sim u$, then
$$\begin{aligned}
\left(A(T)D^{(M,v)}\overrightarrow{x}\right)_{u}&=\sum_{w\sim u}D^{(M,v)}_{w,w}\overrightarrow{x}_{w}
=0.\end{aligned}$$
Otherwise, let $(w_1,u)\in vPu$. If $\overrightarrow{x}_{w_1}\neq 0$, then applying Lemma \[cuentasD\] we get: $$\begin{aligned}
(A(T)D\overrightarrow{x})_{u}&=D_{w_1,w_1}\overrightarrow{x}_{w_1}+ \sum_{\substack{w\sim u,\\ w\neq w_1}}D_{w,w}\overrightarrow{x}_{w}
\\
&=D_{w_1,w_1}\overrightarrow{x}_{w_1}+\sum_{\substack{w\sim u,\\ w\neq w_1}}\overrightarrow{x}_{w}D_{w_1,w_1}
\left(M_{u,w_1}^{-1}M_{u,w}\right)\\
&=D_{w_1,w_1}M_{u,w_1}^{-1}\left(M_{u,w_1}\overrightarrow{x}_{w_1}+\sum_{\substack{w\sim u,\\ w\neq w_1}}
\overrightarrow{x}_{w}M_{u,w}\right)\\
&=D_{w_1,w_1}M_{u,w_1}^{-1}(M\overrightarrow{x})_{u}\\
&=0\end{aligned}$$
If $\overrightarrow{x}_{w_1}=0$, and $\overrightarrow{x}_{w}\neq 0$ for some $w\sim u$, Lemma \[cuentasD\] implies: $$\begin{aligned}
(A(T)D\overrightarrow{x})_{u}&= \sum_{\substack{w\sim u,\\ w\neq w_1}}D_{w,w}\overrightarrow{x}_{w}\\
&=\sum_{\substack{w\sim u,\\ w\neq w_1}}\overrightarrow{x}_{w}D_{w_1,w_1}M_{u,w}\\
&=D_{w_1,w_1}(M\overrightarrow{x})_{u}\\
&=0\end{aligned}$$
Therefore $A(T)D^{(M,v)}\overrightarrow{x}=0$.
For the reciprocal similar arguments working with $D^{-1}$ instead of $D$ yield the result.
If we consider a forest instead of a tree, then a diagonal matrix can be obtained by choosing one vertex in each connected component, and working in a similar fashion. Let $F$ be a forest, and $U\subset V(F)$ such that $U$ has at most one vertex in each connected component of $F$. We define the *$U$-scalation of $M$* as the non-singular diagonal matrix with $$D^{(M,U)}_{w,w}=\prod_{v\in U} D^{(M,v)}_{w,w},$$ where $D^{(M,v)}_{w,w}=1$ if $v$ and $w$ belong to different connected components of $F$.
In order to generalize Theorem \[main\] we need to have a set $U$ that has elements in all the necessary components of a forest. Given a forest $F$ with connected components $T_1,...,T_k$, a set $U\subset {\operatorname{\mathpzc{Supp}}(M)}$ is *supp-transversal of $M$* if $U\cap V(T_i)\neq \emptyset$ whenever ${\operatorname{\mathpzc{Supp}}(M)}\cap V(T_i)\neq \emptyset$. We have the following.
Given a forest $F$ and a matrix $M\in \mathcal{M}_{\mathbb{F},0}(F)$, a vector $\overrightarrow{x}$ is in ${\operatorname{\mathpzc{Null}}(M)}$ if and only if $D^{(M,U)}\overrightarrow{x}$ is in ${\operatorname{\mathpzc{Null}}(A(F))}$ for every set $U$ supp-transversal of $M$.
\[DADBnull\] Let $F$ be a forest, and $M,N\in \mathcal{M}_{\mathbb{F},0}(F)$. For every set $U_1$ supp-transversal of $M$ and every set $U_2$ supp-transversal of $N$, $\overrightarrow{x}\in {\operatorname{\mathpzc{Null}}(M)}$ if and only if $D^{(M,U_1)}(D^{(N,U_2)})^{-1}\overrightarrow{x}$ is in the nullspace of $N$.
Given a forest $F$, and a pair of matrices $M,N\in \mathcal{M}_{\mathbb{F},0}(F)$, a vertex $v$ is in ${\operatorname{\mathpzc{Supp}}(M)}$ if and only if $v$ it is in ${\operatorname{\mathpzc{Supp}}(N)}$.
If $v\in{\operatorname{\mathpzc{Supp}}(M)}$, there is some $\overrightarrow{x}\in{\operatorname{\mathpzc{Null}}(M)}$ with $\overrightarrow{x}_v\neq 0$. Thus by Corollary \[DADBnull\], the vector $D^{(M,v_1)}(D^{(N,v_2)})^{-1}\overrightarrow{x}$ is in ${\operatorname{\mathpzc{Null}}(N)}$ for some $v_1\in {\operatorname{\mathpzc{Supp}}(M)}$, $v_2\in {\operatorname{\mathpzc{Supp}}(N)}$. But $\left(D^{(M,v_1)}(D^{(N,v_2)})^{-1}\overrightarrow{x}_v\right)$ is not $0$ because $D^{(M,v_1)}$ and $D^{(N,v_2)}$ are non-singular diagonal matrices. Therefore $v\in {\operatorname{\mathpzc{Supp}}(N)}$.
The next two corollaries are given to illustrate the strength of Corollary \[DADBnull\], and the relation between the structure of a forest $F$ and the null space of the matrices in $\mathcal{M}_{\mathbb{F},0}(F)$. In [@JM], the concept of the $S$-set of a tree $T$ was introduced. It is the subgraph induced by ${\operatorname{\mathpzc{Supp}}(T)}\cup N({\operatorname{\mathpzc{Supp}}(T)})$ (where $N({\operatorname{\mathpzc{Supp}}(T)})$ denotes the neighborhood of ${\operatorname{\mathpzc{Supp}}(T)}$) and is denoted $\mathcal{F}_S(T)$. In other words, $\mathcal{F}_S(T)$ is the subgraph induced by the vertices in the null support and the neighbors of the vertices in the null support. One of their main results is the fact that the null space of a tree $T$ is the same as the null space of $\mathcal{F}_S(T)$, extended with $0$ to match the dimensions. The same holds true for matrices in $\mathcal{M}_{\mathbb{F},0}(T)$. And, by doing direct sum, the same holds true for forests.
To help with the cleanness of the next corollary, we introduce some notation. Given a matrix $M\in\mathcal{M}_{\mathbb{F},0}(F)$, and $G$ an induced subgraph of $F$, we denote by $M[G]$ the matrix obtained by deleting the rows and columns of vertices not in $G$. We do the same for vectors, $\overrightarrow{x}[G]$ denotes the vector obtained from $\overrightarrow{x}$ by deleting the coordinates correspoding to vertices not in $G$.
Let $F$ be a forest, $M\in\mathcal{M}_{\mathbb{F},0}(F)$ and $\overrightarrow{x}\in \mathbb{F}^F$. Then $\overrightarrow{x}\in {\operatorname{\mathpzc{Null}}(M)}$ if and only if:
- $\overrightarrow{x}\left[\mathcal{F}_S(F)\right]\in {\operatorname{\mathpzc{Null}}(M\left[\mathcal{F}_S(F)\right])}$, and
- $\overrightarrow{x}\left[V(F)\setminus \mathcal{F}_S(F)\right]=\overrightarrow{0}$.
A helpful result, implicit in [@JM], is the fact that the ${\operatorname{\mathpzc{Supp}}(T)}$ is the intersection of all the maximum independent sets of $T$. Which yields the following.
Let $F$ be a forest, $M\in\mathcal{M}_{\mathbb{F},0}(F)$, and $v\in V(F)$. Then $v\in {\operatorname{\mathpzc{Supp}}(M)}$ if and only if $v$ is in every maximum independent set of $F$.
The next corollary, originally proved in [@Mohammadian], follows directly from Corollary \[DADBnull\] and the fact that dimension of the rank of a tree is twice its matching number (see [@bevis]).
If $F$ is a forest and $M\in\mathcal{M}_{\mathbb{F},0}(F)$, then $\dim{{\operatorname{\mathpzc{Rank}}(M)}}=2m$, where $m$ is the size of a maximum matching in $F$.
One can now use the relation between the null space of a forest $F$ and the null space of any matrix $M\in \mathcal{M}_{\mathbb{F},0}(F)$ to find a basis for the null space of $M$, which is done in Algorithm \[algorithm\]. Finding the forest $F$ given the matrix $M$ takes linear time, because it can be obtained by replacing the entries by $1$, and the matrix has at most $2(n-1)$ nonzero entries (the edges of the forest). As $D^{(M,U)}$ does not change the support of a vector, a sparsest basis for the null space of $F$ provides a sparsest basis for the null space of $M$ once it is multiplied by $(D^{(M,U)})^{-1}$. In [@JMPS] the support of a forest was found in linear time, and a $\{-1,0,1\}$ and sparsest basis for the null space of a forest was found in optimal time.
Using the support, finding $D^{(M,U)}$ and $(D^{(M,U)})^{-1}$ takes linear time on the number of vertices, as for each vertex only one operation needs to be done. Afterwards, multiplying the elements of the basis found using the algorithm from [@JMPS] by $(D^{(M,U)}_{v,v})^{-1}$ takes one operation per each non-zero entry in the vectors of the basis of the forest. Hence a sparsest basis for the null space of $M$ can be found in optimal time.
1. INPUT: $M$, a tree-patterned matrix with $0$ in the diagonal.
2. Find $F$ such that $M\in \mathcal{M}_{\mathbb{F},0}(F)$.
3. Apply the algorithms from [@JMPS] to find a sparsest basis, $\mathcal{B}_F$ of $A(F)$ and ${\operatorname{\mathpzc{Supp}}(F)}$.
4. Find the connected components of $F$.
5. For each $T_i$ connected component of $F$ with ${\operatorname{\mathpzc{Supp}}(A(T_i))}\neq \emptyset$ chose $v_i\in {\operatorname{\mathpzc{Supp}}(A(T_i))}$.
6. Let $U$ be the set of the chosen $v_i$ and calculate $(D^{(M,U)})^{-1}$.
7. Calculate $\mathcal{B}_M=(D^{(M,U)})^{-1}\mathcal{B}_F$
8. OUTPUT $\mathcal{B}_M$
Algorithm \[algorithm\] is important because it expands on the set of matrices for which a sparsest basis of the null space can be found in optimal.
On the rank {#Section Rank}
===========
In the previous section we proved that given a forest $F$ and $M\in \mathcal{M}_{\mathbb{F},0}(F)$, ${\operatorname{\mathpzc{Null}}(M)}$ is a non-singular diagonal multiplication of ${\operatorname{\mathpzc{Null}}(F)}$. In this section show that ${\operatorname{\mathpzc{Rank}}(M)}$ is a non-singular diagonal multiplication of ${\operatorname{\mathpzc{Rank}}(F)}$. In order to do so, first we find a basis for the rank of $M$.
Let $v\not\in {\operatorname{\mathpzc{Supp}}(M)}$, we define its *supported-neighborhood vector*, $\overrightarrow{s}_v$, as $$\overrightarrow{s}_v(M)=\sum_{w\in {\operatorname{\mathpzc{Supp}}(M)}\cap N(v)}M(v,w)\overrightarrow{e}_w,$$ where $\overrightarrow{e}_w$ denotes the vector with $1$ in coordinate $w$ and $0$ elsewhere.
In [@JMS] it was shown that $$B(F):=\bigcup\limits_{v\not\in{\operatorname{\mathpzc{Supp}}(F)}}\{\overrightarrow{e}_v,\overrightarrow{s}_v(F)\}
\setminus\{\overrightarrow{0}\}$$ is a basis for the rank of $F$. We show the same result for $M\in \mathcal{M}_{\mathbb{F},0}(F)$.
If $F$ is a forest and $M\in \mathcal{M}_{\mathbb{F},0}(F)$, then $$B(M):=\bigcup\limits_{v\not\in{\operatorname{\mathpzc{Supp}}(M)}}\{\overrightarrow{e}_v,\overrightarrow{s}_v(M)\}\setminus
\{\overrightarrow{0}\}$$ is a basis for the rank of $M$.
It is easy to see that all columns of $M$ can be written as linear combinations of $$B(M)=\bigcup\limits_{v\not\in
{\operatorname{\mathpzc{Supp}}(M)}}\{\overrightarrow{e}_v,\overrightarrow{s}_v(M)\}\setminus\{\overrightarrow{0}\}.$$ Hence ${\operatorname{\mathpzc{Rank}}(M)}\subset{\operatorname{\mathpzc{Span}}(B(M))}$.
But $\dim({\operatorname{\mathpzc{Null}}(M)})=\dim({\operatorname{\mathpzc{Null}}(F)})$ by Theorem \[main\]. Thus $$\dim({\operatorname{\mathpzc{Rank}}(M)})=\dim({\operatorname{\mathpzc{Rank}}(F)})=|B(F)|=|B(M)|.$$
Therefore $B(M)=\bigcup\limits_{v\not\in{\operatorname{\mathpzc{Supp}}(M)}}\{\overrightarrow{e}_v,\overrightarrow{s}_v(M)\}\setminus
\{\overrightarrow{0}\}$ is a basis for the rank of $M$.
Again, we work on a tree instead of a forest, because the rank is the direct sum of the ranks of the connected components.
Let $T$ be a tree, $M\in \mathcal{M}_{\mathbb{F},0}(T)$ and $v$ a vertex of $F$. For each vertex $w$ let $\pi(v,w)$ be second vertex in $vPw$, where $v$ is the first vertex of the path. We define the *$v$-normalization of $M$* as the non-singular diagonal matrix with $$C^{(M,v)}_{w,w}=M_{v,\pi(v,w)}.$$
We define the *rank-normalization of $M$*, $R^M$, as the product of $C^{(M,v)}$ over all vertices $v\not\in
{\operatorname{\mathpzc{Supp}}(M)}$. $$R^M=\prod_{v\not\in{\operatorname{\mathpzc{Supp}}(M)}}C^{(M,v)}$$
Let $v\in T$, then we say that $v$ is a *core vertex* of $M$ if $N(v)\cap{\operatorname{\mathpzc{Supp}}(M)}\neq \emptyset$. In other words, core vertices are the neighbors of vertices in the null support of $M$. The *core* of $M$, ${\operatorname{\mathpzc{Core}}(M)}$ is the set of all core vertices of $M$.
If $T$ is a tree and $M\in \mathcal{M}_{\mathbb{F},0}(T)$, then $R^M{\operatorname{\mathpzc{Rank}}(T)}={\operatorname{\mathpzc{Rank}}(M)}$.
Let $v\not\in{\operatorname{\mathpzc{Supp}}(M)}$. We have $$\begin{aligned}
R^M\overrightarrow{s}_v(T)=&\sum_{w\in {\operatorname{\mathpzc{Supp}}(M)}\cap N(v)}R^M\overrightarrow{e}_{w}\\
=&\sum_{w\in {\operatorname{\mathpzc{Supp}}(M)}\cap N(v)}C^{(M,v)}\prod_{\substack{u\not\in{\operatorname{\mathpzc{Supp}}(M)},\\ u\neq v}}C^{(M,u)}\overrightarrow{e}_{w}.\\\end{aligned}$$ But if $w\in{\operatorname{\mathpzc{Supp}}(M)}\cap N(v)$ and $u\not\in{\operatorname{\mathpzc{Supp}}(M)}$ with $u\neq v$, then $\pi(u,w)=\pi(u,v)$. Notice that $C^{(M,u)}e_{w}=M(u,\pi(u,v))\overrightarrow{e}_{w}$. Hence $$\begin{aligned}
R^M\overrightarrow{s}_v(T)=&\sum_{w\in {\operatorname{\mathpzc{Supp}}(M)}\cap N(v)}C^{(M,v)}\prod_{\substack{u\not\in{\operatorname{\mathpzc{Supp}}(M)},\\ u\neq v}}C^{(M,u)}\overrightarrow{e}_{w}\\
=&\sum_{w\in {\operatorname{\mathpzc{Supp}}(M)}\cap N(v)}C^{(M,v)}\prod_{\substack{u\not\in{\operatorname{\mathpzc{Supp}}(M)},\\ u\neq v}}M(u,\pi(u,v))\overrightarrow{e}_{w}\\
=&\prod_{\substack{u\not\in{\operatorname{\mathpzc{Supp}}(M)},\\ u\neq v}}M(u,\pi(u,v))\sum_{w\in {\operatorname{\mathpzc{Supp}}(M)}\cap N(v)}C^{(M,v)}\overrightarrow{e}_{w}.\\\end{aligned}$$ On the other hand, if $w\in N(v)$, $C^{(M,v)}\overrightarrow{e}_{w}=M(v,w)\overrightarrow{e}_{w}$. Therefore $$\begin{aligned}
R^M\overrightarrow{s}_v(T)=&\prod_{\substack{u\not\in{\operatorname{\mathpzc{Supp}}(M)},\\ u\neq v}}M(u,\pi(u,v))\sum_{w\in {\operatorname{\mathpzc{Supp}}(M)}\cap N(v)}C^{(M,v)}\overrightarrow{e}_{w}\\
=&\prod_{\substack{u\not\in{\operatorname{\mathpzc{Supp}}(M)},\\ u\neq v}}M(u,\pi(u,v))\sum_{w\in {\operatorname{\mathpzc{Supp}}(M)}\cap N(v)}M(v,w)\overrightarrow{e}_{w}\\
=&\prod_{\substack{u\not\in{\operatorname{\mathpzc{Supp}}(M)},\\ u\neq v}}M(u,\pi(u,v))\overrightarrow{s}_v(M).\end{aligned}$$ Hence ${\operatorname{\mathpzc{Span}}(R^M\overrightarrow{s}_v(T))}={\operatorname{\mathpzc{Span}}(\overrightarrow{s}_v(M))}$. It is easy to see that ${\operatorname{\mathpzc{Span}}( R^M\overrightarrow{e}_v)}={\operatorname{\mathpzc{Span}}(\overrightarrow{e}_v)}$. Therefore $R^M{\operatorname{\mathpzc{Rank}}(T)}={\operatorname{\mathpzc{Rank}}(M)}$.
If instead we consider a forest, then a diagonal matrix can be obtained by having $C^{(M,v)}_{w,w}=1$ when $v$ and $w$ are in different connected components. Hence, we have the following.
Given a tree $F$ and a matrix $M\in \mathcal{M}_{\mathbb{F},0}(F)$, $R^M{\operatorname{\mathpzc{Rank}}(F)}={\operatorname{\mathpzc{Rank}}(M)}$.
The following result is a direct application, because $R^M$ is nonsingular.
Let $F$ be a forest, and $M,N\in \mathcal{M}_{\mathbb{F},0}(F)$. The vector $\overrightarrow{x}$ is in ${\operatorname{\mathpzc{Rank}}(M)}$ if and only if the vector $R^N(R^M)^{-1}\overrightarrow{x}$ is in ${\operatorname{\mathpzc{Rank}}(N)}$.
Conclusion
==========
There is a strong relation between the rank and the null space of a tree-patterned (acyclic) matrix with diagonal $0$, and its underlying tree (forest). It would be interesting to study what happens when non-zero diagonal entries are allowed, or when a different graph is used. We conjecture that there will still be a strong relation, but it will not be so straightforward. For example, having non-zero diagonal entries only in the vertices in ${\operatorname{\mathpzc{Core}}(F)}$ should have no effect in the null space of the matrix.
References
==========
[10]{} Bevis, Jean H and Domke, Gayla S and Miller, Valerie A. . 1995 , 18: 109–119, 1995.
T. F. Coleman and A. Pothen. The null space problem. [I]{}. [C]{}omplexity. , 7(4):527–537, 1986.
M. Fiedler. . , 25 (100): 607–218, 1975.
L.-A. Gottlieb and T. Neylon. Matrix sparsification and the sparse null space problem. , 76(2):426–444, 2016.
D. A. Jaume and G. Molina. Null decomposition of trees. , 341: 835-850, 2018.
D. A. Jaume, G. Molina, A. Pastine, and M. Safe. A $\{-1,0,1\}$- and sparsest basis for the null space of a forest in optimal time, 2017. Preprint available as arXiv:1710.01639.
D. A. Jaume, G. Molina, and R. Sota. S-trees, 2017. Preprint available as arXiv:1709.03865.
A. Mohammadian. Trees and acyclic matrices over arbitrary fields. , 64 (3): 466–476, 2015.
A. Neumaier. The second largest eigenvalue of a tree. , 46: 9–25, 1982.
P. Nylen. . , 279: 153–161, 1998.
S. Parter. . , 8 (2): 376–388, 1960.
S. Parter, J. Youngs. . , 4: 102–110, 1962.
G. Wiener. . , 61: 15–29, 1984.
[^1]: [email protected]
[^2]: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We give a detailed exposition of the “vectorized" notation for dealing with quantum operations. This notation is used to highlight the relationships between representations of completely-positive dynamics. Vectorization considerably simplifies the analysis of different methods of quantum process tomography, and enables us to derive compact representation of the investigated quantum operations in terms of the resulting data.'
author:
- Alexei Gilchrist
- 'Daniel R. Terno'
- 'Christopher J. Wood'
title: Vectorization of quantum operations and its use
---
Introduction
============
Quantum process tomography [@cn] is one of the standard tools of quantum information science, and efficient methods of processing tomographic data are of great practical and conceptual interest. Tomographic data manipulations involve massive amounts of the “qubit/qudit algebra" on the finite-dimensional Hilbert spaces. Linear operators on them form vector spaces of their own, and the Hilbert-Schmidt inner product AB:=(A\^B) makes such a vector space into a Hilbert space. As a result, all the formulas that describe state evolution and measurements can be re-written in this vector form. “Vectorized" (or “double ket" [@baran]) formalism is fruitfully used in quantum information (e.g., [@karol; @pt98; @dariano00; @tyson; @aiello]), However, it is sometimes done on the ad hoc basis using not very transparent notation.
In this article we consistently apply it to various forms of the dynamics of open systems (Sec. III) and quantum process tomography (Sec. IV). It allows us to give an easy derivation of known results, and provides a number of new relationships between the reconstructed data and the investigated processes. To make the discussing self-contained and recognizing that a good notation is a clue to successful derivations we begin with a pedagogical introduction into vectorization of finite-dimensional matrices. This is a well-established area of matrix analysis [@hj-topics], and part of our presentation is devoted to introducing notation which is consistent with other areas of theoretical physics and with group representation theory [@wkt].
Vectorization of Matrices {#ssec:vecmat}
=========================
Conventions
-----------
We distinguish between upper (contravaraint) and lower (covariant) indices. Components of a vector $\psi\in{\mathcal{H}}$ are labeled as $\psi^k$, the basis of ${\mathcal{H}}$ is denoted by $\{{{\mathrm{e}}}_k\}$, $k=1,\ldots d\equiv \dim{\mathcal{H}}$. We employ the Einstein summation convention where the summation goes over identical upper and lower indices, hence $\psi=\sum_k\psi^k{{\mathrm{e}}}_k\equiv\psi^k{{\mathrm{e}}}_k$.
The inner product allows to identify ${{\mathrm{e}}}^k\equiv {{\mathrm{e}}}_k^{~\dag}$. When there is a chance of confusion of the vectorial labels with the components, the former are put inside the brackets: the $a$-th component of the basis vector $k$ is denoted as (\_k)\^a=\_[(k)]{}\^a. We choose the orthonormal bases, so ${{\mathrm{e}}}_{(k)}^a= \delta^a_k$. The Dirac notation is introduced as $|k\9\equiv {{\mathrm{e}}}_k$ and $\6k|\equiv {{\mathrm{e}}}^k$. Since =\_k\^k|k, =\_k\^[k\*]{}, the convention for the components of general vectors is (\^)\_k=\^[k\*]{}. \[vecdual\] Moreover, when the indices are shown explicitly, the sign of a Hermitian conjugation becomes redundant, since positioning of the index indicates weather the object is a ket (vector) or a bra (its dual), hence $\psi_k=\psi^{k*}$.
Matrices of the size $p\times q$ represent the operators between the vector spaces of the corresponding dimensions, M:\_1\_2, \^a=M\^a\_[ b]{}\^b,a=1, …p\_2, b=1,…,q\_1. If the bases of the two Hilbert spaces are denoted by ${{\mathrm{e}}}_l$ and ${{\mathrm{u}}}_k$, respectively, then M=M\^k\_[ l]{}\_k\^l\_[k,l]{}M\^k\_[ l]{}|kl|\_2\^\*\_1, where ${\mathcal{H}}^*$ is the dual Hilbert space of ${\mathcal{H}}$. This equation suggests a basis for the space of $p\times q$ matrices, E\_[k]{}\^[ l]{}, which consists of matrices with all but one entry being zero and the unity at the $(kl)$-[th]{} entry. The matrix $M$ reads then as M=M\^k\_[ l]{}E\_[k]{}\^[ l]{}, k=1,…p, l=1,…, q. Matrices with two lower or two upper indices can be considered as superpositions of direct product states,, as in =F\^[kl]{}\_k\_l\_1\_2. In writing the matrix element $(ab)$ in any upper or lover index combination, such as $M^a_{~b}$, $M_a^{~b}$ or $(E_k^{~l})^a_{~b}$, the row index $a$ appears first.
With the above conventions the relationships between components of a transposed matrix are given by (M\^T)\_[ab]{}=M\_[ba]{}, (M\^T)\_[a]{}\^[ b]{}=M\^b\_[ a]{}, the complex conjugation turns the lower into the upper indices, and vice versa, (M\^\*)\_a\^[ b]{}=M\^[a\*]{}\_[ b]{}, and the Hermitian conjugation satisfies (M\^)\^a\_[ b]{}=M\^[b\*]{}\_[ a]{}=(M\^T)\^[ b\*]{}\_[a]{}=(M\^\*)\_b\^[ a]{} Kronicker’s delta is real and symmetric, $\delta^a_b=\delta^a_{~b}=\delta_b^{~a}$. Consistency requires to set $\delta^a_b{}^*=\delta^b_a$. We also keep in mind that $(E_a^{~b})^*=E_a^{~b}$ and $(E_a^{~b})^T=(E_a^{~b})^\dag=E_b^{~a}$, so M\^\*=\_[a,b]{}M\^\*\_a\^[ b]{}E\_a\^[ b]{}=M\^[a\*]{}\_[ b]{}E\_a\^[ b]{}, M\^=(M\^a\_[ b]{}E\_a\^[ b]{})\^=\_[a,b]{}M\^[a\*]{}\_[ b]{}E\_b\^[ a]{}=M\^\*\_a\^[ b]{}E\_b\^[ a]{}.
Consider a tensor product of two matrices, $K=M\otimes N$. A pair of row indices $a_1$ and $a_2$ is combined into a single row index of $K$, and a pair of column indices $b_1$ and $b_2$ is combined into the new column index according to the lexicographic ordering rule that is described below. Hence K\^\_[ ]{}=K\^[a\_1a\_2]{}[,]{}\_[b\_1b\_2]{}M\^[a\_1]{}\_[ b\_1]{}N\^[a\_2]{}\_[ b\_2]{}, and similarly for K\^[a\_1 b\_2]{}\_[ a\_2,b\_1]{}=M\^[a\_1]{}\_[ b\_1]{}N\_[a\_2]{}\^[ b\_2]{}, etc.
Definition and simple properties
--------------------------------
Representation of matrices as vectors on a higher dimensional Hilbert space is called *vectorization*. It transforms a $p\times q$ matrix $M$ into a $(pq)\times 1$ column vector denoted by $\mathrm{vec}(M)$ or $\overrightarrow{M}$ [@hj-topics]. We choose to do it by ordering matrix elements lexicographically, i.e. by stacking the rows of $M$ on top of each other to form a vector. Our convention agrees with [@karol], but is opposite to [@hj-topics]. We made this choice as to keep the same concatenation rules for both vectorization and taking tensor products.
For example, a $2\times 2$ matrix $M$ is re-arranged to form the four-dimensional vector ${\overrightarrow{M}}$, =(M\^1\_[ 1]{}, M\^1\_[ 2]{},M\^2\_[ 1]{},M\^1\_[ 2]{})\^T.
To automate matrix manipulations we have to lump a pair of matrix indices $ab$, $a=1,\ldots p$, $b=1,\ldots q$ into a single vector index $\alpha$. With the stacking convention that we adopted, =f(a,b)q(a-1)+b. \[indexall\] The inverse of vectorization restores the matrix form, as $\mathrm{mat}({\overrightarrow{M}})=M$. The matrix indices are restored according to a=/q+1, b=q-q/q, \[concatind\] where $\lfloor x\rfloor$ is the largest integer less than or equal to $x$.
We now list some useful properties of vectorized matrices [@hj-topics]. For convenience we consider square matrices that act on the space ${\mathcal{H}}$, $p=\dim{\mathcal{H}}$. A matrix A=\_[kl]{}A\^k\_[ l]{}A\^k\_[ l]{}\_k\^l\^\*, is transformed to the vector =\_[kl]{}A\^k\_[ l]{}\_k\_l=\_l(A\_l)\_l =(A\_p)\_l\_l\_l, \[itself\] where ${\mathbb{I}}_p$ is the identity matrix on ${\mathcal{H}}$.
Vectorization is obviously linear: for matrices $A_\alpha$ and scalars $a^\alpha$, ( a\^ A\_)= a\^.
Vectorization is intrinsically related to the tensor product. Our stacking convention for a matrix $M^a_{~b}$ is the same as the the index concatenation rule for elements of a tensor product space, such as $M^{ab}\mapsto M^\alpha$. This will be useful in the following. It is also easy to discover how the vectorizations of $M^T$, $M^*$, and $M^\dag$ are related to each other. Following the definition of vectorization and Eq , we see that =\^, =\^. \[vecmat\] We set ${\mbox{$\overrightarrow{M}$}}$ and ${\mbox{$\overrightarrow{M^\dag}$}}$ to carry the contravariant indices, so ${\mbox{$\overrightarrow{M^*}$}}$ and ${\mbox{$\overrightarrow{M^T}$}}$ carry the covariant ones. Hence \^M\^=M\^a\_[ b]{}, \_M\_=[M\^\*]{}\_a\^[ b]{}=M\^[a\*]{}\_[ b]{}=\^, and \^=M\^\^a\_[ b]{}=M\^[b\*]{}\_[ a]{}, \_=M\^T\_a\^[ b]{}=M\^b\_[ a]{}=\^.
The Hilbert-Schmidt inner product is equivalent to the usual inner product of vectors, (A\^B)=A\^\*\_b\^[ a]{}B\^b\_[ a]{}=\_\^=\^\_\^=\^,, where $\beta=f(b,a)$.
Vectorization relates the tensor and the outer product of vectors, =(\^), as follows from Eq. .
To apply the summation convention we use the opposite label positioning for the basis elements $\mathrm{vec}(E_a^{~b})\equiv {\mbox{$\overrightarrow{{{\mathrm{E}}}_\alpha}$}}$. Let $\alpha=f(a,b)$ and $\beta=f(k,l)$. Then the basic definition ()\^=\^\_=\^k\_a\^b\_l, and the conjugation rules for vectors and matrices lead, e.g., to ()\_=(\^)\_=\_=\^, which agrees with ()\_k\^[ l]{}=(E\_a\^[ b]{})\^\*\_k\^[ l]{}=(E\_a\^[ b]{})\_[ l]{}\^[k\*]{}=(\_a\^k\^b\_l)\^\*=\^a\_k\_b\^l, Similarly, ()\_k\^[ l]{}=(E\_a\^[ b]{})\^l\_[ k]{}=\_a\^l\_k\^b, and =M\^\_, \^=M\^\^=M\_\^=.
Consider two matrices $A$ and $B$, of the sizes $p\times q$ and $q\times r$, respectively. Their product $C=AB$ can be written as a vector ${\mbox{$\overrightarrow{AB}$}}$ in two ways. The matrix $C$ results from the left action of $A$ on $B$, so in a vectorized notation C\^=[\_A]{}\^\_[ ]{} B\^, where the matrix ${{\mathsf{M}}}_A$ will be now determined. Writing the indices in full and using matrix multiplication rules leads to C\^a\_[ b]{}=[\_A]{}\^[a d]{}\_[ b,c]{}B\^c\_[ d]{}=A\^a\_[ c]{}B\^c\_[ d]{}\^d\_[ b]{}=A\^a\_[ c]{}\^[ d]{}\_[b]{}B\^c\_[ d]{}, \[transa\] so \^[a d]{}\_[ b,c]{}=A\^a\_[ c]{}\^[ d]{}\_b. As a result, =(A\_r). \[lefta\] Now we see that Eq. for ${\mbox{$\overrightarrow{A}$}}$ is a corollary of the above result with $B\mapsto{\mathbb{I}}$. The matrix $C$ also results from the right action of $B$ on $A$, which is expressed as =(\_pB\^T). \[righta\]
The following lemma [@hj-topics] deals with the triple product of matrices:
Let $A, B, C$ and $X$ be $p\times q$, $r\times s$, $p\times s$ and $q\times r$ matrices, respectively. Then the matrix equation AXB=C for $qs$ unknowns $X^l_{~m}$ is equivalent to the system of $ps$ equations (AB\^T)=. That is to say, =(AB\^T). The proof consists in the repeated application of Eqs. (\[lefta\]) and (\[righta\]), =(A\_s)=(A\_s)(\_qB\^T)=(AB\^T). $\square$
Reshuffling {#ssec:reshuffling}
-----------
The SWAP operation changes the order of subsystems in a tensor product: |\_1|\_2(|\_1|\_2)=|\_2|\_1. For two identical systems this operation swaps their quantum states. Consider two vectors $\psi\in{\mathcal{H}}_1$ and $\phi\in{\mathcal{H}}_2$. In the component form we write them as =\^[a\_1]{}\_[a\_1]{}, a\_1=1,…, p=\_1, and =\^[a\_2]{}\_[a\_2]{}, a\_2=1,…, r=\_2,
A single index $\alpha$ of the tensor product $\Psi^\alpha=(\psi\otimes \phi)^\alpha$ is built from the indices of its subsystems as in Eq. , =r(a\_1-1)+a\_2. Similarly, a single vectorial index for $\Phi=\phi\otimes\psi$ is =q(a\_2-1)m+a\_1. The bases of the two tensor products are certain permutations of each other. It is easy to see that this permutation $\beta=\sigma[r,p](\alpha)$ is given by the table (
[cccccc|ccc|ccc]{} 1 & 2 &…& s &…& r & r+1 & r+2 &…&(p-1)r+1 & …& pr\
1 & 1+p&…&1+(s-1)p &…& 1+(r-1)p & 2 & 2+p &…&p &…& p+(r-1) p
), \[permut\] where the first row contains indices of the components of $\Psi$ that are matched with the corresponding indices of $\Phi$ in the second row. The vertical lines separate the first and the last $r$ components of $\Psi$ from the rest, and $s=1,\ldots r$. Its inverse, $\sigma[r,p]^{-1}$ associates to each element of $\Phi$ an element of $\Psi$ and is obtained by interchanging the rows. As a result, if we label the SWAP operator as $S(r,p)$, then ()\^=S(r,p)\^\_[ ]{}()\^where S(r,p)\^\_[ ]{}=\^[()]{}\_[ ]{}. It follows from this construction that $S$ is an orthogonal matrix, and $S(r,p)^T=S(r,p)^{-1}=S(p,r)$. For example, a qubit SWAP is S(2,2)= (
[cccc]{} 1 & 0 & 0 & 0\
0 & 0 & 1 & 0\
0 & 1 & 0 & 0\
0 & 0 & 0 & 1\
), and the SWAP of $2\times 3$ systems is done with S(2,3)= (
[cccccc]{} 1 & 0 & 0 & 0 & 0 & 0\
0 & 0 & 0 & 1 & 0 & 0\
0 & 1 & 0 & 0 & 0 & 0\
0 & 0 & 0 & 0 & 1 & 0\
0 & 0 & 1 & 0 & 0 & 0\
0 & 0 & 0 & 0 & 0 & 1
). If we restore the original indices we see that SWAP can be treated as a matrix transposition \^[a\_2a\_1]{}=\^[a\_1a\_2]{}=S\^[a\_2a\_1]{}\_[ ,b\_1b\_2]{}\^[b\_1b\_2]{}, and S\^[a\_2a\_1]{}\_[ ,b\_1b\_2]{}=\^[a\_2]{}\_[b\_2]{}\^[a\_1]{}\_[b\_1]{}. For two square matrices $M$ and $N$ that act on the spaces ${\mathcal{H}}_1$ and ${\mathcal{H}}_2$, respectively, the SWAP operation results in NM = S(r,p) (MN) S(r,p)\^T.
As we will see in the next sections, it is important to relate $\mathrm{vec}(M)\otimes\mathrm{vec}(N)$ and $\mathrm{vec}(M\otimes N)$. It is done by the operation of *reshuffling* [@karol]. Let the matrices $M$ and $N$ be of the sizes $p\times q$ and $r\times s$, respectively. A matrix element M\^[a\_1]{}\_[ b\_1]{}N\^[a\_2]{}\_[ b\_2]{} can be interpreted either according to “ first vectorize, then tensor", as M\^[a\_1]{}\_[ b\_1]{}N\^[a\_2]{}\_[ b\_2]{}=:C\^[a\_1 a\_2]{}\_[ b\_1, b\_2]{} C\^ C\^A=((M)(N))\^A, or as M\^[a\_1]{}\_[ b\_1]{}N\^[a\_2]{}\_[ b\_2]{}=:\^[a\_1 a\_2]{}\_[ ,b\_1 b\_2]{} \^\_[ ]{} \^B=(MN)\^B, according to “first tensor, then vectorize" precept.
A repeated application of the index concatenation definition gives A=rsq(a\_1-1)+rs(b\_1-1)+s(a\_2-1)+b\_2, B=rsq(a\_1-1)+qs(a\_2-1)+s(b\_1-1)+b\_2. Hence swapping the indices $a_2$ and $b_1$ brings ${\mbox{$\overrightarrow{M}$}}\otimes{\mbox{$\overrightarrow{N}$}}$ to ${\mbox{$\overrightarrow{M\otimes N}$}}$. This can be formalized by the following
For matrices $M$ and $N$ of the size $p\times q$ and $r\times s$, respectively, the reshuffling matrix ${{\mathsf{R}}}(M,N)$ is defined by (M,N)(p,q,r,s)=\_pS(r,q)\_s,
Our discussion established the
Let the matrices $M$ and $N$ be of the size $p\times q$ and $r\times s$, respectively. Then (MN) = (p,q,r,s)(). \[vecten\]
A Mathematica package by Miszczak, Gawron, and Pucha[ł]{}a [@pack] for the analysis of quantum states and operations implements matrix vectorization and resufling for the use in the relevant functions [^1].
Applications of Vectorization to Open Quantum Systems
=====================================================
The simplest use of the vectorization is in analyzing the convex probability domains of quantum measurements [@pt98]. Consider a positive operator valued measure (POVM) with $N$ outcomes. Vectorization makes it nearly obvious that the domain in the probability space formed by probabilities of all outcomes with all possible states of a $n$ dimensional system has at most $n^2-1$ dimensions.
Completely Positive Maps {#ssec:cpaxioms}
------------------------
Evolution of an open quantum system whose initial state is uncorrelated with the environment is described by a completely positive map. Label the initial state of the system (${\mathcal{H}}_1$) as $\rho$, and the state of its environment (${\mathcal{H}}_2$) by $\omega$. If the joint state is given by $\tau_{12}=\rho\otimes\omega$, and the overall unitary evolution is $U_{12}$, then the final state of the system is given by ()=\_[\_2]{}(U\_[12]{}U\_[12]{}\^), \[bighilb\] where the operation of partial tracing is defined by (\_2)\^[a\_1]{}\_[ b\_1]{}=\^[a\_1 a\_2]{}\_[ ,b\_1a\_2]{}. Such a map ${{\mathsf{T}}}$ is trace-preserving, convex-linear and completely positive [@karol]. This latter property means that ${{\mathsf{T}}}$ is positive (maps positive matrices to positive matrices), and also if we introduce an auxiliary system of arbitrary dimension then the map ${{\mathsf{T}}}\otimes{{\mathsf{I}}}$ on the joint system is positive, where ${{\mathsf{I}}}$ is the identity map on the auxiliary system.
A map satisfying these three axioms is referred to as a completely positive, trace preserving (CPTP) map. It is possible to relax the trace condition to ${\mbox{Tr}}[{{\mathsf{T}}}(\rho)]\le{\mbox{Tr}}[\rho]$, allowing for completely positive trace decreasing maps. We will only be concerned with CPTP maps and we may refer to maps satisfying all three of these axioms simply as completely-positive maps (CP maps).
Any CPTP map has a convenient operator sum representation:
\[thm:kraus\] A map ${{\mathsf{T}}}$ acting on density operators of ${\mathcal{H}}$ is CPTP if and only if there exists a set of bounded operators $\{K_n\}$ acting on ${\mathcal{H}}$ such that $${{\mathsf{T}}}(\rho) = \sum_n K_n \rho K_n^\dagger
\quad \mbox{ where }\quad \sum_n K_n^\dagger K_n = {\mathbb{I}}.
\label{eqn:krausform}$$
The operators $K_n$ are called *Kraus matrices* and they satisfy $\sum_n K_n^\dagger K_n = {\mathbb{I}}$, which is known as the *completeness relation*. A Kraus representation of a given process ${{\mathsf{T}}}$ is not unique. This can be useful as different system-environment interactions may still give rise to the same reduced dynamics on the system.
In general, a linear hermiticity-preserving transformation ${{\mathsf{T}}}$ (completely positive or not) acting on the space of density matrices may be represented by the [*dynamical (Choi) matrix*]{} ${{\mathsf{D}}}({{\mathsf{T}}})$ [@karol; @su61; @choi], $$\rho'{}^a_{~b}={{\mathsf{D}}}^{a~~~d}_{~c,b}\rho^c_{~d},\label{introchoi}$$ and a vectorized version of this relation ’\^=|\^\_[ ]{}\^, \[dynres\] uses the reshuffled matrix $\bar{D}$, |\^[a d]{}\_[ b,c]{}=\^[a d]{}\_[ c,b]{}, $\mathrm{vec}(\bar{{{\mathsf{D}}}})={{\mathsf{R}}}{\mbox{$\overrightarrow{{{\mathsf{D}}}}$}}$.
The dynamical matrix has a number of useful properties [@karol]. The trace preserving condition is equivalent to the constraint on the partial trace of the dynamical matrix, $${{\mathsf{D}}}^{a~~~d}_{~c,a}= \delta^d_c \ ,$$ which implies that the eigenvalues sum up to the system’s dimension, $\sum_a\lambda_a=d_1$. Moreover, if the map is unital, i.e., it maps the maximally mixed state into the maximally mixed state, then $${{\mathsf{D}}}^{a~~~s}_{~s,b}=\delta^a_b \ .$$
The dynamical matrix is Hermitian, ${{\mathsf{D}}}^\dag{}^\alpha_{~\beta}={{\mathsf{D}}}^\alpha_{~\beta}$, and due to a theorem of Choi [@choi] its positivity is equivalent to the complete positivity of ${{\mathsf{T}}}$:
\[thm:process\] A quantum operation ${{\mathsf{T}}}$ on a $d$-dimensional system $S$ is CP if and only if its dynamical matrix ${{\mathsf{D}}}({{\mathsf{T}}})$ is positive-semidefinite $({{\mathsf{D}}}({{\mathsf{T}}})\ge0)$.
The proof is based on the eigendecomposition of the Choi matrix, $${{\mathsf{D}}}^{a~~~d}_{~c,b}={{\mathsf{D}}}^\alpha_{~\beta}=\sum_n\lambda_n {\mbox{$\overrightarrow{M_n}$}}{}^\alpha{\mbox{$\overrightarrow{M_n}$}}{}^\dag_\beta=\sum_n\lambda_n{\mbox{$\overrightarrow{M_n}$}}{}^\alpha{\mbox{$\overrightarrow{M^*_n}$}}_\beta=\sum_n\lambda_n M_n{}^a_{~c} M_n^*{}^{~d}_{b}, \label{eigenD}$$ where $\alpha=f(a,c)$, $\beta=f(b,d)$, and if all the eigenvalues $\lambda_n$ are non–negative it is possible to define Kraus matrices by absorbing the eigenvalues, K\_n=M\_n. We also get a compact expression for the reshuffled matrix, |=\_n K\_nK\^\*\_n. \[bartensor\]
Linear Superoperator {#ssec:superop}
--------------------
Vectorization allows a different perspective on Choi matrix and its reshuffled version. We consider a combination of basis matrices $E_a^{~b}$ and ${{\mathsf{T}}}(E_a^{~b})\equiv E'{}_a^{~b}$, with suitably arranged component indices.
A *linear superoperator* $\Phi_{{\mathsf{T}}}$ acts on vectorized density matrices and is defined by $$\Phi_{{\mathsf{T}}}{\mbox{$\overrightarrow{\rho}$}}\equiv{\mbox{$\overrightarrow{{{\mathsf{T}}}(\rho)}$}}.
\label{eqn:sopevo}$$ On the one hand, since \_,\_==(E\_a\^[ b]{}\^E\_c\^[ d]{})=\^a\_c\_b\^d=\^\_, \[phi1\] we have \^=\^=\_. \[linact\] As a result, $$\Phi_{{\mathsf{T}}}=\sum_\alpha{\mbox{$\overrightarrow{E'_\alpha}$}}\,{\mbox{$\overrightarrow{{{\mathrm{E}}}_\alpha}$}}{}^\dag={\mbox{$\overrightarrow{E'_\alpha}$}}\,{\mbox{$\overrightarrow{{{\mathrm{E}}}^\alpha}$}}.
\label{eqn:sodef}$$ A comparison with Eq. identifies the linear superoperator with the reshuffled dynamical matrix, $\Phi_{{\mathsf{T}}}=\bar{{{\mathsf{D}}}}({{\mathsf{T}}})$. The matrix elements of $\Phi_{{\mathsf{T}}}$ are obtained by \_\^\_[ ]{}=\^\_=\_[a,b]{}(E’\_a\^[ b]{})\^k\_[ l]{}(E\_a\^[ b]{}\^\*)\_m\^[ n]{} =(E’\_a\^[ b]{})\^k\_[ l]{}\_m\^a\_b\^n=(E’\_m\^[ n]{})\^k\_[ l]{}=(E’\_)\^. Indeed, $\rho'_\kappa=\Phi_{{\mathsf{T}}}{}^\kappa_{~\mu}\rho^\mu$, so |\^[k n]{}\_[ l,m]{}=(E’\_m\^[ n]{})\^k\_[ l]{}=\_x (K\_x)\^k\_[ m]{} (K\^\*\_x)\_[l]{}\^[ n]{}. We note that when expressed in terms of the Kraus matrices $K_n$ the reshuffled matrix $\bar{{{\mathsf{D}}}}$ is built according to “first tensor, then vectorize" prescription. However, the opposite approach “first vectorize, then tensor" is taken when ${\mbox{$\overrightarrow{\Phi_{{\mathsf{T}}}}$}}$ is expressed in terms of the basis matrices and their transforms.
Process matrix
--------------
We define the *process matrix* $\Lambda_{{\mathsf{T}}}$ for a map ${{\mathsf{T}}}$ as a matrix on a fictitious space ${\mathcal{H}}_1\otimes{\mathcal{H}}_2$, where the two spaces are copies of ${\mathcal{H}}$. Without going through the vectorization of Eqs. , , the transformation $\rho'={{\mathsf{T}}}(\rho)$ can be written as ’=\_[a,b]{}\^c\_[ d]{}E’\_a\^[ b]{}(E\_a\^[ b]{}\^E\_c\^[ d]{})=\^c\_[ d]{}E’\_a\^[ b]{}(E\_b\^[ a]{} E\_c\^[ d]{}). Thus it is given by $${{\mathsf{T}}}(\rho)={\mbox{Tr}}_2\left[\Lambda_{{\mathsf{T}}}^{T_2}({\mathbb{I}}\otimes\rho)\right], \qquad \Lambda_{{\mathsf{T}}}^{T_2}\equiv E'{}_a^{~b} \otimes E_{b}^{~a}. \label{eqn:processmap}$$ For our purposes it is more convenient to use the matrix arrangement of $\Phi_{{\mathsf{T}}}$, and define a new four-indexed object according to the “first tensor, then vectorize" prescription, \_E’\_a\^[ b]{} E\_[b]{}\^[ a]{}\^=\_[a,b]{} E’\_a\^[ b]{} E\_[a]{}\^[ b]{}\^\*, where we keep a complex conjugation as a reminder for the correct index placement. Then ()=\_2, and \_\^[k n]{}\_[ l,m]{}=(\_[a,b]{} E’\_a\^[ b]{} E\_[a]{}\^[ b]{}\^\*)\^[k n]{}\_[ l,m]{}=(E’\_a\^[ b]{})\^k\_[ m]{}\_l\^a\_b\^n=(E’\_l\^[ n]{})\^k\_[ m]{}. A comparison with Eq shows that ${{\mathsf{D}}}({{\mathsf{T}}})=\Lambda_{{\mathsf{T}}}$.
Summary of relationships
------------------------
The Jamio[ł]{}kowski isomorphism [@karol; @jamio] identifies dynamical matrices of CP maps with certain entangled states. Consider a generalization of the Bell state $|\Phi^+\9=(|++\9+|--\9)/2$ on the space ${\mathcal{H}}_1\otimes{\mathcal{H}}_2$ of the previous section. Its density matrix is given by \_+=\_[a,b]{} E\_[a]{}\^[ b]{}E\_[a]{}\^[ b]{}/d, where $d=\dim{\mathcal{H}}$. It is now easy to write the matrix elements of $\rho'=({{\mathsf{T}}}\otimes{{\mathsf{I}}})\rho$, ’\_+\^[kl]{}\_[ ,mn]{}=(E’\_l\^[ n]{})\^k\_[ m]{}/d=\_\^[k n]{}\_[ l,m]{}/d, which establishes
Any linear map ${{\mathsf{T}}}$ acting on the space of mixed states on the Hilbert space ${\mathcal{H}}$ can be associated, via its dynamical matrix ${{\mathsf{D}}}({{\mathsf{T}}})$, with an operator on the enlarged Hilbert spacee ${\mathcal{H}}\otimes{\mathcal{H}}$, ()()\_+=’\_+, where for a completely positive map ${{\mathsf{T}}}$ the matrix $\tau'_+$ is a valid quantum state.
As a computational aid we present a compact summary of the relationships between different representations of a completely positive map ${{\mathsf{T}}}$.
To$\backslash$From ${{\mathsf{D}}}({{\mathsf{T}}}) $ $\Phi_{{\mathsf{T}}}$ $ K_n $
------------------------- -------------------------------------------------------------------------- ------------------------------------------------------------------------------------ ----------------------------------------------------------------------------------
${{\mathsf{T}}}(\rho)=$ ${\mbox{Tr}}_2\left[{{\mathsf{D}}}({\mathbb{I}}_1\otimes\rho^T)\right] $ $\mathrm{mat}(\Phi_{{\mathsf{T}}}{\mbox{$\overrightarrow{\rho}$}})$ $\sum_n K_n\rho K_n^\dagger$
${{\mathsf{D}}}=$ ${{\mathsf{D}}}$ $\mathrm{mat}({{\mathsf{R}}}^{-1}{\mbox{$\overrightarrow{\Phi_{{\mathsf{T}}}}$}})$ $\sum_{n} {\mbox{$\overrightarrow{K_n}$}}{\mbox{$\overrightarrow{K_n}$}}{}^\dag$
$\Phi_{{\mathsf{T}}}=$ $\mathrm{mat}({{\mathsf{R}}}{\mbox{$\overrightarrow{{{\mathsf{D}}}}$}})$ $\Phi_{{\mathsf{T}}}$ $\sum_n K_n\otimes K_n^*$
$K_n =$ $\sqrt{\lambda_n}\mathrm{mat}({\mbox{$\overrightarrow{M_n}$}})$ $\longleftarrow$ $K_n$
: Relationships between mathematical representations of CP maps.[]{data-label="tab:cpreps"}
Two possible viewpoints on both ${{\mathsf{D}}}$ and $\Phi_{{\mathsf{T}}}$ as obtained either according to “first tensor, then vectorize" or “first vectorize, then tensor" approaches (with a counterpart being obtained by the other procedure) are consistent, because the reshuffling matrix involved in the process satisfies ${{\mathsf{R}}}={{\mathsf{R}}}^{-1}$. The latter result follows from Eq. , with all the dimensions being equal to $\dim{\mathcal{H}}$.
Process Tomography using the Superoperator representation
=========================================================
Dual bases
----------
Quantum process tomography is used to characterize the properties of quantum information-processing devices. The standard process tomography [@cn] uses a tomographically complete set of input states that spans the entire set of states. The input states are sent through the investigated device and its action is then reconstructed by analyzing the outputs. A set of dual states $D=\{ D_\nu \}$ for a tomographically complete set $\rho_{{{\mathrm{in}}}}=\{ \rho_\mu\}$, $\mu=1,\ldots, n=d^2$, where $d=\dim{\mathcal{H}}$, plays an important role in the mathematics of reconstruction. It is defined by the orthogonality relation ${\mbox{Tr}}(D_\nu^\dagger \rho_\mu ) =\delta_{\nu\mu}$, $\forall \mu,\nu$.
If we rewrite this in a vectorized notation we obtain [,]{}= \_ hence =\^\_= Introducing the operator P=\_\^, we see that $P{\mbox{$\overrightarrow{D_\nu}$}}={\mbox{$\overrightarrow{\rho_\nu}$}}$, hence =P\^[-1]{}. It is possible to write this relationship between a tomographic set and its dual in a simpler and computationally more efficient way. We write vectors of the set $\rho_{{\mathrm{in}}}$ as the columns of a matrix =(,…,). \[eqn:matvec\] Since its columns are linearly independent, this $d^2\times d^2$ matrix is invertible. Its Hermitian conjugate can be written as $$\begin{aligned}
[\rho_{{{\mathrm{in}}}}]^\dagger &=&\left(
\begin{array}{c}
{\mbox{$\overrightarrow{\rho_1}$}}{}^\dag \\
\vdots \\
{\mbox{$\overrightarrow{\rho_n}$}}{}^\dag
\end{array}
\right).\end{aligned}$$ With this notation we get P=\[\_\]\[\_\]\^. As a result, =\[\_\]\^[-1]{}\^\^[-1]{}. If we introduce $[D]=({\mbox{$\overrightarrow{D_1}$}},\ldots,{\mbox{$\overrightarrow{D_n}$}})$, then an even simpler expression is obtained: =\[\_\]\^[-1]{}\^\^[-1]{}\[\_\]=\[\_\]\^[-1]{}\^.
Now we apply this to the standard process tomography.
Process tomography with $\Phi_{{\mathsf{T}}}$
---------------------------------------------
Let $\rho_{{{\mathrm{in}}}}=\{\rho_\mu\}$ be a tomographically complete set of input states with dual basis $D=\{D_\mu\}$. Then the linear superoperator $\Phi_{{\mathsf{T}}}$ for a CP map ${{\mathsf{T}}}: \rho \mapsto {{\mathsf{T}}}(\rho) $ is given by \_= \_\^ \[phinew\]
Consider an arbitrary state $\rho= p^\mu\rho_\mu$. We have to show $\Phi_{{\mathsf{T}}}{\mbox{$\overrightarrow{\rho}$}}=\mathrm{vec}{\,{{\mathsf{T}}}(\rho)}$. Applying the above expression to ${\mbox{$\overrightarrow{\rho}$}}$ we see that \_()=p\^[D\_,\_]{}=p\^\_\^=p\^=().
We simplify this expression by using the matrix of vectorized input states, Eq. . This leads us to the following
For the set of output states $\rho_{\mathrm{out}}=\{{{\mathsf{T}}}(\rho_\mu)\}$, where $\rho_{{{\mathrm{in}}}}=\{\rho_\mu\}$ is a tomographically complete set of input states, the linear superoperator $\Phi_{{\mathsf{T}}}$ is given by \_= \[\_\]\[ \_\]\^[-1]{}
We use Eq. . \^[-1]{}=\[\_\]\[D\]\^==\_.
We rewrite this result in terms of the probabilities of various experimental outcomes.
Let $\{M_\mu\}$ be a tomographically complete measurement set ($M_\mu\geq 0$, $\sum M_\mu={\mathbb{I}}$) with a dual basis $\{E^\nu\}$, ${\mbox{Tr}}(E_\nu^\dag M_\mu)=\delta^\nu_\mu$. Then \_= \^, \[spt\] where $[E]=({\mbox{$\overrightarrow{E_1}$}},\ldots, {\mbox{$\overrightarrow{E_n}$}})$ is a matrix of the vectorized dual elements, and $[m]=(m_{\mu\nu})$ is a matrix of probabilities, $m_{\mu\nu}={\mbox{Tr}}(M_\mu^\dag\rho'_\nu)$.
Since any $\rho$ is reconstructed according to $\rho=E^\mu m_\mu$, where $m_\mu={\mbox{Tr}}(M_\mu\rho)$, = \[E\]\[m\], and $[D]^\dag=[\rho_{{\mathrm{in}}}]^{-1}.$
Ancilla-assisted process tomography with linear superoperator
-------------------------------------------------------------
Presentation of Jamio[ł]{}kowski isomorphism and the manipulation of data in the ancilla-assisted quantum tomography (AAPT) [@dariano00; @alte] also benifit from the vectorized notation.
We introduce an auxiliary system (ancilla) ${\mathcal{H}}_2$ to our principal system ${\mathcal{H}}_1$, so that the state space of the joint system is given by ${\mathcal{H}}_1\otimes{\mathcal{H}}_2$. AAPT aims to reconstruct the CP map ${{\mathsf{T}}}$ on the states of ${\mathcal{H}}$ the action of ${{\mathsf{T}}}\otimes{{\mathsf{I}}}$ on a single state $\tau_{12}$ of this combined system. Evolution on ${\mathcal{H}}_1$ results in an operation ${{\mathsf{T}}}$ described by a superoperator $\Phi_{{\mathsf{T}}}$, and the ancilla does not evolve. Any initial state $\tau_{12}$ of the joint system can be represented as \_[12]{}=\_w\^\_\_, \_w\^=1 which is in general entangled (it is separable if and only if all $w^\mu\geq 0$).
The joint system dynamical matrix has the form \_=(\_|)\^[-1]{}, where ${{\mathsf{R}}}$ is the reshuffling matrix, which in this case again satisfies ${{\mathsf{R}}}={{\mathsf{R}}}^{-1}$, and the linear superoperator of the identity map is $\bar{{{\mathsf{I}}}}\equiv\Phi_{{\mathsf{I}}}$.
Linearity of the evolution is expressed as ’=()=\_w\^(\_)\_. Hence ’=\_w\^=\_w\^(’) =(\_|)\_w\^. Matrix elements of $\bar{{{\mathsf{I}}}}$ satisfy $\bar{{{\mathsf{I}}}}^{k~~~n}_{~l,m}=\delta^k_m\delta^n_l={\mathbb{I}}^{\alpha}_{~\beta}$. Another reshuffling leads to the desired result, ’=(\_|)\^[-1]{}.
Both vectors \^A=\^(\^[-1]{})\^=(\^[-1]{})\^[k m]{}\_[ l, n]{}= \_w\^\_\^k\_[ l]{}\_\^m\_[ n]{}, and ${\mbox{$\overrightarrow{\tau_{{\mathrm{out}}}}$}}{}^A\equiv({{\mathsf{R}}}^{-1}{\mbox{$\overrightarrow{\tau}$}}')^{\alpha\beta}$ correspond to the right hand side of Eq. . Their relationship is of the forms of Eq. : \^=(\_\^\_[ ]{}\^\_[ ]{})\^. Introducing $\Phi_\tau\equiv\mathrm{mat}({\mbox{$\overrightarrow{\tau_{{\mathrm{in}}}}$}})$ we have =(\_)=(\_\^T). Hence we can recover the dynamical matrix $\Phi_{{\mathsf{T}}}$ from the output state $\tau_{AS}$ when the matrix $\Phi_\tau$ is invertible:
A linear superoperator $\Phi_{{\mathsf{T}}}$ is recovered from the output of AAPT with the initial state $\tau$ according to \^\_[ ]{}=((\_\^[-1]{})\^T)\^\_[ ,]{}\^ \[aapt\]
An important special case is the entanglement-assisted process tomography, where the input state is a maximally entangled =\_+=\_[ij]{}/d. It corresponds to $\Phi_{\tau_+}=\mathrm{mat}({\mbox{$\overrightarrow{\tau_+}$}})={\mathbb{I}}$. Hence we established a useful expression for the dynamical matrix and a dual form of the Jamio[ł]{}kowski isomorphism
In the entanglement assisted process tomography (AAPT with the maximally entangled initial state $\tau_+$) the dynamical matrix the linear superoperator $\Psi_{{\mathsf{T}}}$ is determined by the output state according to =. \[eapt\]
Conclusions and outlook
=======================
Adopting vectorized notation allows transparent and consistent representation of various forms of open system dynamics, isomorphism between states and operations and representation of process tomography. Neat expressions for process tomography , and use already reconstructed output states. Processing of actual state tomographic data is much more involved. In particular, relative frequencies cannot be directly taken as probabilities [@pt98; @alte], assumption of completely positive dynamics should be justified or may not be true [@lidar; @us], matrices may have only generalized inverses [@dariano09]. A transparent and versatile notation is a great asset in dealing with these issues, and we expect that it will simplify some of the existing formulas and bring to light new useful relationship, similarly to presented in this work.
We thank Karol Życzkowski for many useful discussions. A correspondence with Andrea Aiello, Joseph Emerson, Jaroslaw Miszczak, Yutaka Shikano, and Jon Tyson is gratefully acknowledged. The work of DRT was supported in part by the grant from the Australian Academy of science. CJW was supported by the Perimeter Scholars International program.
[99]{}
I. L. Chuang and M. A. Nielsen, J. Mod. Opt. [**44**]{}, 2455 (1997). M. Baranger, Phys. Rev. **111**, 494 (1958). I. Bengtsson and K. Życzkowski, *Geometry of Quantum States* (Cambridge University Press, Cambridge), 2006. A. Peres and D. R. Terno, J. Phys. A **31**, L671 (1998). G. D’Ariano, L. Maccone, and M. Paris, Phys. Lett. A **276**, 25 (2000). J. E. Tyson, J. Phys. A **36**, 10101 (2003) A. Aiello, G. Puentes, D. Voigt, and J. P. Woerdman, Phys. Rev. A **75**, 062118 (2007); A. Aiello, G. Puentes, and J. P. Woerdman, Phys. Rev. A **76**, 032323 (2007). R. A. Horn and C. R. Johnson, *Topics in Matrix Analysis* (Cambridge University Press, 1991). W-K Tung, *Group Theory in Physics*, (World Scientific, Singapore, 1985). J.A. Miszczak, P. Gawron, Z. Pucha[ł]{}a, available at http://zksi.iitis.pl/wiki/projects:mathematica-qi. E. C. G. Sudarshan, P. M. Mathews and J. Rau, Phys. Rev. [**121**]{}, 920 (1961). M.-D. Choi, Linear Algebra Appl. [**10**]{}, 285 (1975). A. Jamio[ł]{}kowski, Rep. Math. Phys. **3**, 275, (1972). J. B. Altepeter *et al.*, Phys. Rev. Lett. **90**, 193601 (2003). A. Shabani and D. Lidar, Phys. Rev. Lett. **102** 100402 (2009). A. Gilchrist *et al.*, in preparation. G.M. D’Ariano, D. F. Magnani, and P. Perinotti, Phys. Lett. A, **373**, 1111 (2009).
[^1]: Note that in [@pack] our vectorization operation is called “reshaping", and the name “vectorization" refers to the column-by-column stacking, as in [@hj-topics].
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In the upcoming LHC Run 2, at $\sqrt{s} \sim13$ TeV, it is suggested to seek unusually charged ($Q= -4/3$ and $+5/3$) quarks with mass $M_Q \sim 3$ TeV which carry lepton number ($L = +2$ and $-2$ respectively) and decay superweakly to a bilepton $Y$ with mass $M_Y\sim2.5$ TeV and a usual quark. These long-lived decays will have displaced decay vertices and produce a striking final state in $pp$ which contains two separated jets together with two pairs of correlated like-sign charged leptons. Such a process was inaccessible energetically in LHC Run 1 with $\sqrt{s} \sim8$ TeV. The simplest theoretical explanation is the 331-model which has new physics necessarily below $4$ TeV and which explains the existence of three families by anomaly cancellation.'
---
April 2015\
.5in
.45in
[**Paul Howard Frampton**]{}[^1]
[*Oxford, UK.*]{}
.4in
Introduction
============
In addition to the theoretical predictions for the LHC of supersymmetry and dark matter, the discovery of either of which would be revolutionary, it is worth being more conservative and to consider instead the ancient art of model-building in gauge theories which extend the standard model and are motivated and testable. In particular, we here suggest that LHC experimentalists seek unusually-charged quarks ($Q=-4/3, +5/3$) which are produced strongly and decay slowly by weaker than weak interactions, are constrained to lie below $4$ TeV, and motivated by an explanation of three families.
The 331 model[@F; @PP] has provoked sufficient interest that there exist a number of studies of its phenomenological ramifications. One aspect which has, however, escaped much attention is the issue of lepton number ($L$) conservation and the role it plays in suppressing the decay rate for the heavy quarks. Although there are reviews of 331 bilepton physics[@CR; @CD], the slow decays of the 331 heavy quarks have not been previously emphasized. The upgraded LHC seems tailor-made for discovery of these heavy quarks and its Run 2 could expose them.
The familiar quarks $(u, d, c, s, t, b)$ have baryon number $B=1/3$ and $L=0$. The familiar leptons $(e-, \mu-, \tau-, \nu_e, \nu_{\mu}, \nu_{\tau})$ have $B=0$ and $L=1$. The exotically-charged quarks of the 331 model carry nonzero $L$ as follows: $D$ and $S$ have $B=1/3$ and $L=2$; $T$ has $B=1/3$ and $L=-2$.
An important final-state particle at the LHC is the penetrating and unstable muon which decays via the weak interaction $$\mu^- \rightarrow e^- + \bar{\nu}_e + \nu_{\mu}
\label{mudecay}$$ with a long lifetime $\tau_{\mu} \sim 2\times10^{-6}$s according [@vRS1; @vRS2] to the tree-level formula $$\tau_{\mu} = \frac{g_2^4 M_{\mu}^5}{12288 \pi M_W^4}
\label{mulifetime}$$ where $g_2$ is the electroweak $SU(2)$ gauge coupling with $g_2^2 = 8M_W^2G_F$. Other than the muon, the only long-lived charged particles in the standard model are the stable electron and proton. But there may be about to appear an entirely new breed of metastable charged elementary particles to enter this small group.
It is well-known and investigated that if there exists a fourth family of quarks, then they can mix only very little with the first three families because the $3\times3$ CKM matrix [@C; @KM] is close to being unitary. Hence, the additional quarks of the fourth family would have interesting long lifetimes as discussed in [@FH; @MRSY]. In the 331-model it is assumed that such sequential quarks do not exist and it is predicted that there are only three families.
Here we are interested in quarks which are long-lived for a different reason, namely $L$ conservation. The consequent superweak interaction is mediated by $Y$ bilepton intermediate vector bosons in the 331-model and a possible particle discovery at LHC is of a sibling to the $W^{\pm}$. For instance $Y^-$ mediates the abnormal muon decay $$\mu^- \rightarrow e^- + {\nu}_e + \bar{\nu}_{\mu}
\label{muotherdecay}$$ which is suppressed relative to the normal decay, Eqs. (\[mudecay\],\[mulifetime\]) by a factor $$f= (M_W/M_Y)^4.
\label{f}$$ and, given $M_W \simeq 80$ GeV, the question is whether the value of $M_Y$ can be arbitrarily large. In the present context the answer is no.
In the 331-model there is an important theoretical upper limit $M_Y \leq 4TeV$ for the symmetry breaking to the standard model, arising from the renormalization group behavior of the electroweak mixing angle and the group embedding. The value $\sin^2 \theta (M_Z) = 0.231$ runs upward with energy scale and reaches $\sin^2 \theta (E) = 0.250$, a singular point of the embedding $SU(2)_L \subset SU(3)_L$, at $E=4$ TeV. This was first analysed in [@F] and has been much more recently confirmed in [@Buras]. This intrinsic 331 upper limit is what underlies the claim that the new physics is at a mass scale especially befitting LHC’s Run 2.
Theoretically then $M_Y { \mathop{}_{\textstyle \sim}^{\textstyle <} }4$ TeV while experimentally $M_Y \geq 1.5$ TeV. We may reasonably take $M_Y=2.5$ TeV ($\simeq 10\sqrt{10} M_W$) as an illustration whereupon the suppression factor $f$ in Eq. (\[f\]) is $f \simeq10^{-6}$. The experimental upper limit for process Eq.(\[muotherdecay\]) is [@PDG] disappointing, the branching ratio being restricted merely to $\leq 1.2\%$. The 331 prediction is that this branching ratio is four orders of magnitude smaller, $\sim 10^{-6}$.
By superweak interaction we therefore mean the weak interaction further suppressed for bilepton mediation by the factor $f$ in Eq.(\[f\]) relative to the $W$ exchange. Superweakness implies that the exotic quarks $(D, S, T)$ are long-lived.
Long-Lived Quarks
=================
In the 331-model which requires exactly three families there are three additional exotic quarks($D, S, T$), one in each family. The gauge group is $SU(3)_C \times SU(3)_L \times U(1)$ and for the first family the quarks are in the triplet and three singlets of $SU(3)_L$ $$\left( \begin{array}{c} u^{\alpha} \\
d^{\alpha} \\
D^{\alpha}
\end{array}
\right)_L ~~~ \bar{D}_{L.\alpha}, ~~~ \bar{d}_{L.\alpha}, ~~~ \bar{u}_{L,\alpha},$$ and similarly for the second family $$\left( \begin{array}{c} c^{\alpha} \\
s^{\alpha} \\
S^{\alpha}
\end{array}
\right)_L ~~~ \bar{S}_{L.\alpha}, ~~~ \bar{s}_{L.\alpha}, ~~~ \bar{c}_{L,\alpha}.$$ The quarks of the third family are assigned differently, in one antitriplet and three singlets $$\left( \begin{array}{c} T^{\alpha} \\
t^{\alpha} \\
b^{\alpha}
\end{array}
\right)_L ~~~ \bar{b}_{L.\alpha}, ~~~ \bar{t}_{L.\alpha}, ~~~ \bar{T}_{L,\alpha}.$$
The established weak gauge bosons $(W^-, W^0, W^-)$ with $W^0 \equiv Zcos\theta + \gamma\sin\theta$ are augmented by five more, a $Z^{'}$ and four bileptons $(Y^+, Y^{++})$ ($L=-2$) and $(Y^-, Y^{--})$ ($L=+2$).
The superweak decays of $D$ are (we exhibit only the muonic decays, the most readily detected) $$D \rightarrow u + Y^{--} \rightarrow u + \mu^- + \mu^-
\label{Ddecay}$$ has a displaced vertex in the silicon detector. There is the alternative equally long-lived decay $$D \rightarrow d + Y^{-} \rightarrow d + \mu^- + \nu_{\mu},
\label{Ddecay2}$$ but the neutrino $\nu_{\mu}$ makes process Eq. (\[Ddecay2\]) far more challenging to detect than Eq.(\[Ddecay\]).
The sequential second family exotic quark $S$ has similar long-lived decays $$\begin{aligned}
S &\rightarrow& c + Y^{--} \rightarrow c + \mu^- + \mu^- \nonumber \\
S &\rightarrow& s + Y^{-} \rightarrow s + \mu^- + \nu_{\mu}
\label{Sdecay}\end{aligned}$$
In the 331 model the third family, on the other hand, the $T$ has $Q=+4/3$ and lepton number $L=-2$ so that its long-lived decays have flipped electric charges $$\begin{aligned}
T &\rightarrow& b + Y^{++} \rightarrow b + \mu^+ + \mu^+ \nonumber \\
T &\rightarrow& t + Y^{+} \rightarrow t + \mu^+ + \bar{\nu}_{\mu}
\label{Tdecay}\end{aligned}$$
The antiquarks ($\bar{D}, \bar{S}, \bar{T}$) have superweak decays into the corresponding charge conjugate final states $$\begin{aligned}
\bar{D} &\rightarrow& \bar{u} + Y^{++} \rightarrow \bar{u} + \mu^+ + \mu^+ \nonumber \\
\bar{D} &\rightarrow& \bar{d} + Y^+ \rightarrow \bar{d} + \mu^+ + \nu_{\mu} \nonumber \\
\bar{S} &\rightarrow& \bar{c} + Y^{++} \rightarrow \bar{c} + \mu^+ + \mu^+ \nonumber \\
\bar{S} &\rightarrow& \bar{s} + Y^+ \rightarrow \bar{s} + \mu^+ + \nu_{\mu} \nonumber \\
\bar{T} &\rightarrow& \bar{b} + Y^{--} \rightarrow \bar{b} + \mu^- + \mu^- \nonumber \\
\bar{T} &\rightarrow& \bar{t} + Y^{-} \rightarrow \bar{t} + \mu^- + \bar{\nu}_{\mu}\end{aligned}$$
We take $M_Y=2.5$ TeV and can assume a normal quark mass hierarchy is $M_T > M_S > M_D$ with $M_D \sim 3$ TeV. The lowest threshold will then be for $\bar{D}D$ pair production by strong interactions so the most interesting event would be $pp \rightarrow \bar{D}D$ + any.
By strong interactions, two gluons can produce the $\bar{D}D$ pair, practicable only at Run 2 of the LHC with $13 $ TeV. The $\bar{D}, D$ quarks being long-lived will travel a macroscopic distance from the production vertex.
The $\bar{D},D$ quarks decay to bileptons and the most striking signature would surely be an event with $$\bar{D} \rightarrow \bar{u} + Y^{++} \rightarrow \bar{u} + \mu^+ + \mu^+
\label{barD}$$ and the counterpart $$D \rightarrow u + Y^{--} \rightarrow u + \mu^- + \mu^-
\label{D}$$
In Eqs.(\[barD\],\[D\]), the light quarks $\bar{u}, u$ will hadronize to high-energy jets which the LHC physicists are well equipped to reconstruct. The two jets will originate from the separate displaced decay vertices for $\bar{D}$ and $D$. The like-sign pairs of muons will centre around an invariant bilepton mass $M_Y \sim 2.5$ TeV, which can be determined.
This bilepton mass can subsequently be confirmed by the other channels $\mu^-\nu_{\mu}$, $e^-e^-$, $e^-\nu_e$, $\tau^-\tau^-$, etc.
Discussion
==========
The two most heralded targets for the LHC, beyond the Higgs boson, were to confirm weak-scale supersymmetry and to produce dark matter. If weak-scale supersymmetry existed, it was expected to appear in the 2009-2013 Run 1 at $\sim 8$ TeV, but did not. The excluded parameters narrow the likelihood of its discovery in Run 2. Once weak-scale supersymmetry is abandoned, the link between the weak scale and dark matter mass is lost. The masses for suggested DM candidates range from axions with mass $\sim 10^{-15}$ GeV to black holes with mass $\sim 10^{62}$ GeV so it would now require remarkably good fortune for it to show up at the LHC. The possibility of extra spatial dimensions large enough to be detected at the LHC is not strongly motivated.
There are not many theoretical models with a strong reason to expect the relevant new physics scale to be specifically in the LHC Run 2 ($13$ TeV) regime, as opposed to the Run 1 ($8$ TeV) one. Among these, the 331 model does naturally contain an multi-TeV scale [@F; @Buras] in its analysis. Its long-lived charged quarks are predicted in the appropriate mass range hence more likely to be produced in Run 2 than Run 1. The signature of such an event is striking and although this would not immediately explain all the parameters of the standard model it will give a second stronger explanation of why there exist three families beyond the ingenious observation in [@KM] that it accommodates the observed CP violation in flavor-changing weak interactions.
If such a discovery is made, what is the next step in the theory? It would suggest even further cousins of the $W$ and $Y$ gauge bosons which could appear in additional $SU(3)$ factors. Nonabelian subsumption of the $U(1)_Y$ gauge group factor is hinted at by avoidance of the Landau pole. At present this remains idle speculation until an electroweak $SU(3)$ has empirical evidence which would, nevertheless, firmly justify the construction of higher energy apparatus to answer further questions.
[100]{} P.H. Frampton, Phys. Rev. Lett. [**69,**]{} 2889 (1992). F. Pisano and V. Pleitez, Phys. Rev. [**D46,**]{} 410 (1992). F. Cuypers and M. Raidal, Nucl. Phys. [**B501,**]{} 3 (1997).\
[arXiv:hep-ph/9704224]{}. F. Cuypers and S. Davidson, Eur. Phys. J. [**C2,**]{} 503 (1998).\
[arXiv:hep-ph/9609487]{}. T. van Ritbergen and R.G. Stuart, Phys. Rev. Lett. [**82,**]{} 488 (1999).\
[arXiv:hep-ph/9808283]{}. T. van Ritbergen and R.G. Stuart, Nucl. Phys. [**B564,**]{} 343 (2000).\
[arXiv:hep-ph/9904240]{}. N. Cabibbo, Phys. Rev. Lett. [**10,**]{} 531 (1963). M. Kobayashi and T. Maskawa, Prog. Theor. Phys. [**49,**]{} 652 (1973). P.H. Frampton and P.Q. Hung, Phys. Rev. [**D58,**]{} 057704 (1998).\
[arXiv:hep-ph/9711218]{}. H. Murayama, V. Rentala, J. Shu and T.T. Yanagida, Phys. Lett.[**B705,**]{} 208 (2011).\
[arXiv:1012.0338\[hep-ph\]]{}. A.J. Buras, F. De Fazio and J. Girrbach, JHEP [**1402,**]{} 112 (2014).\
[arXiv:1311:6729 \[hep-ph\]]{}. K.A. Olive, [*et al.*]{} (Particle Data Group). Chin. Phys. [**C38,**]{} 090001 (2014).
[^1]: e-mail address: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
24.2cm 16.0cm -1.0in -42pt
@rel\#1\#2[@@rel\#1\#2]{} @@rel\#1\#2\#3
\[equation\]
@fontreset@font
to
**Solitons of the Einstein-Yang-Mills Theory[^1]**
Dieter Maison[^2]
*Max-Planck-Institut für Physik\
— Werner Heisenberg Institut —\
Föhringer Ring 6\
80805 Munich (Fed. Rep. Germany)*
Introduction
============
My talk is an overview of results on self-gravitating solitons of the classical Yang-Mills-Higgs (YMH) theory. It is based on analytical and numerical results obtained in collaboration with P. Breitenlohner and P. Forgács [@BFM] . Many other people, who have contributed in establishing our present understanding of this subject will be mentioned in due course.
Let me start by specifying, what precisely I understand under solitons, since this concept is used with various different meanings in the literature. I will adhere to a rather liberal use of this concept, denoting by it any particle-like solution of a non-linear field theory. Particle-like solutions are localized, time-independent solutions of finite total energy (mass) with some stability against perturbations. A typical, maybe the best known example of relativistic solitons are the non-abelian ‘t Hooft-Polyakov monopoles. As a genuine non-linear structure they play an important role in the non-perturbative aspects of the YMH theory. For a long time it was believed, that also Einstein’s theory of General Relativity had such smooth solitonic solutions - called geons. However, as was shown by Lichnerowicz [@Lichner], neither Einstein’s theory in vacuum nor the Einstein-Maxwell theory give rise to regular particle-like solutions. Nevertheless there are alternative candidates - the Schwarzschild resp. Reissner-Nordstr[ø]{}m (RN) black holes. Although they suffer from a physical singularity at their center, this singularity is hidden from the observer behind an event horizon. Black holes have finite mass and behave in many ways like genuine particles. In fact, they may well be considered as “renormalized” point particles, dressed with their gravitational self-field [@Damour].
Regular particle-like solutions were found for models involving gravitating complex scalar fields (“Boson stars”), but they have rather the properties of exotic cosmic objects than those of particles [@Colpi]. Later Lichnerowicz’s No-Go-Theorem could be generalized from the Einstein-Maxwell theory to Kaluza-Klein models and supergravities [@BGM], leading to the belief that no smooth solitons can be found for self-gravitating gauge theories. It is clear that this result cannot apply to self-gravitating versions of flat-space solitons like ’t Hooft-Polyakov monopoles, first studied by van Nieuwenhuizen, Wilkinson and Perry [@Perry]. However for the YM theory without a Higgs field, which has no solitons in flat space, it came as a surprise to many, when Bartnik and McKinnon (BM) [@Bart] discovered a family of regular, localized (finite mass) solutions of the gravitating theory. Although their discovery was based only on “numerical evidence” a rigorous existence proof was found subsequently (with due delay!) [@Yau]. When it was found that they are unstable these “particles” lost some of their glamour. It was realized that they had much in common with the “sphalerons” of the YMH theory [@Wald]. The latter are solutions with a Higgs doublet (for the gauge group $SU(2)$) in contrast to the monopole obtained with a triplet. Varying the strength of the gravitational coupling of the gravitating version of the flat-space sphaleron one obtains a one-parameter family of solutions interpolating between the flat solution and the (first) BM solution [@Greene]. This strongly suggests to interprete the BM solutions as gravitationally bound counter-parts of the Higgs-bound flat sphaleron. Indeed, still other sphalerons can be obtained replacing the gravitational field by a dilaton [@LM].
Besides the gravitating sphalerons I will also discuss the effects of the gravitational self-interaction on the non-abelian monopoles. As to be expected they develop a gravitational instability for sufficiently strong gravitational self-force. Contrary to naive expectation (and to claims in the literature) the static monopoles do however not simply turn into black holes as the strength of the gravitational coupling is increased to its critical value. For values close to the critical strength of the gravitational coupling the space-like hypersurfaces of these solutions develop two distinct regions separated by a long throat. The inner part tends to a kind of ‘cosmological’’ solution representing a closed asymptotically Robinson-Bertotti universe, whereas the outer one becomes the exterior part of the extremal RN black hole.
As already mentioned before there is a second type of soliton in General Relativity - the black hole. Taking the YM resp. YMH model as the matter part one finds a rich spectrum of static black hole solutions. This is to be contrasted with Einstein’s theory in vacuum resp. with the Einstein-Maxwell theory, where according to a theorem of Israel [@Israel] the Schwarzschild resp. RN solution are the only static black holes. In the EYM theory one finds not only the embedding of the abelian RN black hole, but in addition there are genuinely non-abelian (“coloured”) black holes. Their co-existence gives rise to an interesting violation of the “No-Hair-Conjecture”, since they carry the same (magnetic) charge [@Bizon].
Yang-Mills-Higgs in Flat Space {#flat}
==============================
to
Before I come to the effects of gravity I would like to give a short reminder of “particle like” solutions of the YMH system in flat space. For simplicity I restrict myself to the gauge group [*SU(2)*]{} from now on. There are two different cases to be considered, leading to rather different types of solutions. The Higgs field can be either in a triplet or in a doublet representation. In either case the action is $$\label{action}
S=-{1\over4\pi}\int d^4\;x\Bigl[{1\over4g^2}{\rm Tr}F^2+{1\over2}
|D\phi|^2+{\lambda\over8}(|\phi|^2-v^2)\Bigr]\;.$$ It is important to notice that the expression for the action contains two mass scales, the mass $M_W=gv$ of the YM field and the mass $M_H=\sqrt\lambda v$ of the Higgs field. From these we may form the dimensionless ratio $\beta=M_H/M_W$.
The particle like (static, spherically symmetric) solutions in the case of a Higgs triplet are the ’t Hooft-Polyakov magnetic monopoles. They are obtained with the ansatz $$\label{ansatz}
W^a_0=0\quad W^a_i=\epsilon_{iak}{x^k\over r}(W(r)-1)\quad
\phi^a={x^a\over r}H(r)\;.$$ Inserting this ansatz in the action (\[action\]) one gets $$\label{mono}
S=-\int dr\Bigl[{1\over g^2}(W'^2+{(1-W^2)^2\over2r^2})+{r^2\over 2}H'^2+
{\lambda r^2\over8}(H^2-v^2)^2+W^2H^2\Bigr]\;.$$ In order to obtain finite total energy the Higgs field has to tend to its vacuum value $v$ for $r\to\infty$, forcing in turn $W\to 0$ (l.h.s. of Fig. \[figflat\]).
For large values of $M_H$ and hence of $\beta$ the function $H(r)$ rises quickly to its asymptotic value $v$. In the limit $\beta\to\infty$ the Higgs field may be replaced by $v$ for all $r>0$ and its only role is to give a mass to the YM field. The total energy of the solution stays finite in this limit. In fact, it only varies by a factor $\approx 1.8$ as $\beta$ varies from $0$ to $\infty$. There is a second possibility to let $\beta$ go to infinity, holding $M_H$ fixed, but letting $M_W\to 0$ (and hence $W\equiv1$ at the relevant length scale $1/M_H$). This way one obtains the “global monopole” playing the role of a texture in cosmological considerations [@Vil].
Due to the topological character of the magnetic charge, related to the asymptotic vacuum structure of configurations with finite energy, the monopole is a stable solution.
The second possibility is a Higgs field in the doublet representation. The relevant ansatz of the Higgs field is $\Phi^\alpha=H(r)\xi^\alpha$ with some constant spinor $\xi$. Although this ansatz is not itself spherically symmetric it leads to a consistent reduction. The corresponding reduced action is $$\label{sphal}
S=-\int dr\Bigl[{1\over g^2}(W'^2+{(1-W^2)^2\over2r^2})+{r^2\over 2}H'^2+
{\lambda r^2\over8}(H^2-v^2)^2+{1\over4}(W+1)^2H^2\Bigr]\;.$$ The only essential difference of this action to the one for the triplet is the form of the mass term. It destroys the symmetry $W\to -W$ and enforces $W$ to turn to $W=-1$ for $r\to\infty$ in order to have finite total energy (Fig. \[figflat\]). This asymptotic behaviour implies that the solution has no magnetic charge in contrast to the previous case with $W\to 0$.
In contrast to the stable monopole the sphaleron, i.e. the solution minimizing the energy $H=-S$, is unstable. In order to understand this instability it is important to consider the most general spherically symmetric ansatz for the YM field. $$\begin{aligned}
\label{sphericalYM}
W_t^a&=(0,0,A_0)\,,\qquad W_\theta^a&=(W_1,W_2,0)\,\nonumber \\
W_r^a&=(0,0,A_1)\,, \qquad
W_\varphi^a&=(-W_2\sin\theta,W_1\sin\theta,\cos\theta)\,.\end{aligned}$$ The ansatz used above for the monopole and the sphaleron corresponds to a consistent reduction putting $A_0=A_1=W_2=0$ and $W_1=W$. The sphaleron turns out to be stable under variations staying within the minimal reduction, but not if $\delta W_2\neq0$ and $\delta A_1\neq0$. As was discussed by Manton [@Manton] this instability is due to the non-trivial topology of the configuration space of the spherically symmetric YM potential, again related to the asymptotic vacuum structure of configurations with finite energy.
Gravitating Monopoles {#gmon}
=====================
to
A spherically symmetric gravitational field is described by a space-time metric of the form $$\label{Metr}
ds^2=e^{2\nu}dt^2-e^{2\lambda}dR^2
-r^2d\Omega^2\;,$$ where $d\Omega^2=d\theta^2+\sin^2\theta d\varphi^2$ and $\nu,\lambda$ and $r^2$ are functions of the coordinates $t$ and $R$.
The quantity $r^2$ is proportional to the surface area of the invariant 2-spheres and hence has a geometrical significance. Furthermore for static space-times the function $e^\nu$ measures the invariant length of the time-translation Killing vector, thus only the function $e^\lambda$ is gauge dependent, i.e., depends on the choice of the radial coordinate $R$.
A simple gauge choice is $R=r$ (Schwarzschild coordinates), which is however well defined only as long as $dr/dR\neq0$. Another convenient choice is obtained putting $e^\lambda=r$ (isotropic coordinates), leading to an autonomous form of the field equations.
With gravity a new scale comes in through Newton’s constant $G$, which allows us to define the Planck mass $M_{\rm Pl}=1/\sqrt G$. The existence of two different scales, the YM scale given by $M_W$ and the Planck scale $M_{\rm Pl}$ have a very important impact on the structure of the solutions. In particular, studying the limiting case $M_{\rm Pl}>>M_W$ gives important insights for their interpretation, as we will see in the following.
Together with $M_W$ we can form the dimensionless ratio $\alpha=M_W\sqrt G/g=M_W/gM_{\rm Pl}$. As just mentioned, a special role is played by the limiting case $\alpha\to0$, which can however be achieved in two different ways:
1. $G\to0$, $M_W$ fixed, in which the gravitational field decouples (flat space);
2. $v=M_W/g\to0$, $G$ fixed, in which the Higgs field becomes trivial and can be ignored.
The reduced EYMH action can be expressed as $$\label{gaction}
S=-\int dR e^{(\nu+\lambda)}
\Bigl[
{1\over2}\Bigl(1+e^{-2\lambda}((r')^2
+\nu'(r^2)'\Bigr)- e^{-2\lambda}r^2V_1-V_2-V_3
\Bigr]\;,$$ with $$V_1={(W')^2\over r^2}+{1\over2}(H')^2\;,$$
$$V_2={(1-W^2)^2\over2r^2}+
{\beta^2r^2\over8}(H^2-\alpha^2)^2$$
and $$V_3=W^2H^2 \quad {\rm resp.}\quad V_3={1\over4}(W+1)^2H^2$$ for the triplet resp. doublet Higgs. Through a suitable rescaling we have achieved that the action depends only on the dimensionless parameters $\alpha$ and $\beta$.
to
Well-known exact solutions of the coupled Einstein-YM equations, obtained from the action (\[gaction\]), are the Schwarzschild solution with trivial YM and Higgs fields $W\equiv1,H\equiv v$ and the abelian (magnetically charged) Reissner-Nordstr[ø]{}m (RN) solution with $W\equiv0,H\equiv v$. Both describe static black holes with a curvature singularity at the origin. As was already mentioned, there exist no regular solitons of the Einstein-Maxwell theory.
Besides these trivial abelian solutions there is a rich spectrum of non-abelian solutions found by numerical integration of the field equations with suitable boundary conditions [@BFM; @Lee].
To begin with there are the self-gravitating versions of the flat-space non-abelian monopoles, which are recovered in the limit $\alpha\to0$ ($G\to0$) (l.h.s. of Fig. \[figmon\]). These are globally regular solutions with a regular center of symmetry (origin) and finite mass. They exist only up to some maximal value $\alpha_{\rm max}$ (depending on $\beta$) of the mass ratio $\alpha=M_W/gM_{\rm Pl}$. Such a maximal value of $\alpha$ is to be anticipated. We expect the monopole to become gravitationally unstable, when its size $R_{\rm mon}\approx 1/M_W$ becomes comparable to its Schwarzschild radius $R_{SS}=GM_{\rm mon}\approx GM_W/g^2$, i.e. for $GM_W^2/g^2=\alpha^2\approx 1$. As $\alpha$ increases the solutions develop a typical limiting behaviour indicating this instability, which may be characterized as “gravitational confinement” of the monopole. The spatial hyper-surface $t={\rm const.}$ develops an infinite throat separating an interior region with a smooth origin and non-trivial YM field from an exterior extremal RN solution with $W\equiv 0$. This throat is characterized by a finite limiting value $r_l=\alpha$ of the metric function $r$. Geometrically this means that neighbouring radial light-rays become non-divergent. All this is much like the $t={\rm const.}$ surfaces of the extremal RN solution, with the only difference, that the interior part of the throat is not the analytic continuation of the exterior one. In fact the combination of the metric functions $\nu+\lambda$ blows up along the throat coming from the interior, whereas $\nu+\lambda\equiv0$ for the RN solution.
Amazingly, for small values ($\lsim 0.7$) of the parameter $\beta=M_H/M_W$ this kind of singular limiting behaviour does not occur at the maximal value of $\alpha$ but at some critical value $\alpha_c<\alpha_{\rm max}$. Running along the 1-paramter family of solutions starting at $\alpha=0$ and increasing $\alpha$ one runs through a maximum of $\alpha$ before the slightly smaller critical value $\alpha_c$ is reached, i.e.there exist two monopole solutions to the same value of $\alpha$ in the interval $\alpha_c<\alpha<\alpha_{\rm max}$. This double-valuedness can be avoided using a different parameter for this family of solutions. For solutions with a regular origin the YM potential $W$ has the behaviour $W=1-br^2+O(r^4)$ for $r\to 0$. Given $\alpha$ the parameter $b(\alpha)$ is fixed by the requirement of asymptotic flatness. It turns out that the parametrization with $b$ instead of $\alpha$ is one-to-one (compare Fig. \[figbalpha\]).
to
This figure contains also the first two members of a sequence of excited families of monopoles showing a different behaviour of $b(\alpha)$ for $\alpha\to 0$. The corresponding values of $b$ tend to finite values $b_n$ ($n=1,2,...$) as $\alpha$ tends to 0 ($v\to0$) related to the solutions found by Bartnik and McKinnon (BM). The latter are globally regular solutions of the EYM equations without a Higgs field. They are labelled by the number $n$ of zeros of the YM potential $W$. Their mass $M$ is of order one in units of $M_{\rm Pl}/g$ — the only mass scale in this case — tending rapidly to $M=1$ for growing $n$. The parameters $b_n$ converge to $b_\infty\approx 0.7064$. In contrast to the monopole solutions they carry zero magnetic charge, related to their different asymptotic behaviour for $r\to\infty$, where $W\to\pm1$. (Compare Fig. \[figbart\]). The shape of the $n=1$ solution reminds very much of the flat space sphaleron. In fact, it is also unstable and may be understood as a gravitationally bound sphaleron, the gravitational field replacing the Higgs field of the flat sphaleron [@Wald]. That this interpretation makes good sense is underlined by the fact that similar solutions are obtained with a scalar dilaton replacing the gravitational field [@LM].
For finite, but small $\alpha$ the excited monopoles consist essentially of a very small (Planck size) BM-solution sitting inside a large (size $1/M_W$) flat monopole (r.h.s. of Fig. \[figmon\]). Let me recall that for all the monopole solutions $W\to0$ for $r\to\infty$, even though this is not clearly visible for all the curves of the plot.
All these families of excited monopole solutions have a common value of $\alpha_c=\sqrt3/2$, which is also the maximal one in this case. For $\alpha\to\alpha_c$ we observe the same limiting behaviour (infinite throat) as for the fundamental monopole. Thus it seems that the latter describes a rather universal phenomenon indicating gravitational instability of static equilibrium configurations and hence is to be expected to occur also for other gravitating matter systems.
to
The l.h.s. of Fig.\[figmass\] shows the masses of the various monopole solutions as a function of $\alpha$ ($\beta=0$). We find that for $\alpha=\alpha_c$ the mass of the solutions becomes $M=M_c\equiv M_{\rm Pl}/g$, the mass of the extremal RN solution. This is easily understood from the merging of the exterior throat part of the monopoles with the latter solution as $\alpha\to\alpha_c$. As indicated in the r.h.s. of Fig \[figmass\] the mass at $\alpha_{\rm max}$ is slightly bigger than $M_c$. While $\beta$ increases $\alpha_c$ decreases to the limiting value $\alpha_c=\sqrt{2}/2$ for $\beta\to\infty$. Similar to their flat counterparts solutions for $\beta=\infty$ (i.e. $H\equiv v$) may be considered as cosmological textures.
Non-abelian Black Holes {#bh}
=======================
to
Apart from the solutions with a regular origin there are non-abelian, “coloured” black holes, parametrized by their radius $r_h$ (in geometrical units, i.e. the value of $r$ at the event horizon) in addition to $\alpha$ and $\beta$ [@BFM; @Aichel]. For $r_h<< 1/M_W$ these non-abelian black holes may be interpreted as a tiny Schwarzschild black hole sitting inside a monopole. On the other hand, when $r_h$ becomes bigger than $\approx 1/M_W$ this type of solution disappears and only the abelian RN black holes exist. For $r_h\to 0$ the matter fields tend uniformly to those of the globally regular solutions, whereas for the metrical functions this limit is clearly more delicate.
Detailed numerical analysis reveals that non-abelian black holes exist only in a limited domain of the $\alpha$-$r_h$-plane, whose shape undergoes some characteristic changes as $\beta$ varies from 0 to $\infty$. Fig. \[figdomain\] shows these domains. Observe that we use $\alpha r_h$ instead of $r_h$ as the abscissa - equivalent to expressing $r_h$ in units of $1/M_W$ - in order to obtain domains remaining bounded for $\alpha\to0$. In the following I shall discuss in some more detail the structure of these “Phase Diagrams” and the phenomena happening at their boundaries. Let me start with the case $\beta=0$.
It is appropriate to subdivide the relevant sector $\alpha\geq0$, $r_h\geq0$ into the four subregions I-IV (compare Fig. \[figphase\]).
In regions I and II we find coloured black holes. Above the diagonal, i.e. in regions II and III we have the abelian RN solutions, the extremal RN black holes sitting on the diagonal. Below the diagonal the RN solution has a naked singularity and does not represent a black hole. No black holes neither abelian nor non-abelian could be found in region IV. Region I may be subdivided in region ${\rm I_a}$, where only the b.h. version of the fundamental monopole resides and region ${\rm I_b}$, where in addition their radial excitations are found. Thus region ${\rm I_a}$ contains essentially one b.h. solution for given values of $\alpha$ and $r_h$ - apart from a small interval $\alpha_c(r_h)<\alpha<\alpha_{\rm max}(r_h)$, where two solutions exist) - whereas in region ${\rm I_b}$ countably many solutions exist (for given $\alpha$ and $r_h$).
In region II abelian and non-abelian black holes coexist. This establishes an obvious violation of the so-called “No-Hair Conjecture”. According to the latter black holes should (apart from mass and angular-momentum) be uniquely determined through their “gauge charges” - their magnetic charge in the present case. However, abelian and non-abelian black holes carry the same magnetic charge and can also be made degenerate in mass resp. the value of $r_h$. Because in general only one type of the “degenerate” black holes is stable (compare below), a weakened form of the “No-Hair Conjecture”, including the requirement of stability, could be maintained.
As $\beta$ increases from $0$ to $\beta=4$ the structure of the “Phase Diagram” remains essentially the same, the right boundary curve moving in to the left. However, for $\beta>4$ this boundary curve develops a second, concave branch (compare Fig. \[figdomain\]) determined by another mechanism, the formation of a degenerate (inner or outer) horizon.
The boundary curve above the diagonal is essentially characterized by the bifurcation of the non-abelian with the abelian RN solution. For a given value of $\alpha$ this happens at some value $r_{\rm h,c}(\alpha)$. Approaching this value from below the value $W_h$ of $W$ at the horizon tends to zero.
But again there is a slight complication for small values of $\beta$ ($\beta\lsim 1.1$); similar to the existence of $\alpha_{\rm max}>\alpha_c$ there is a $r_{\rm h,max}>r_{\rm h,c}$-phenomenon (compare Fig. \[figrhwh\]).
to
Stability of monopoles and coloured black holes {#stabmon}
===============================================
I shall discuss here only stability against infinitesimal, spherically symmetric perturbations. In view of the time-independence of the solutions this amounts to analyzing the spectrum of perturbations with harmonic time-dependence obeying suitable boundary conditions. Imaginary frequencies correspond to unstable modes of the solution.
As to be expected all the excited regular monopoles turn out to be unstable. The branch of gravitating monopoles connected to the flat space solution is stable up to $\alpha_{\rm max}$, whereas the corresponding upper branch - existing for $\alpha\epsilon[\alpha_c,\alpha_{\rm max}]$ is unstable [@Helia]. This change of stability at the bifurcation point at $\alpha_{\rm max}$ of the massfunction is in agreement with general results on 1-parameter families of solutions and well-known from stellar models [@Wheel].
Analogous results hold for the non-abelian, magnetically charged black holes. It is, however, interesting to observe that the abelian RN black hole is unstable in the framework of the non-abelian theory for $\alpha$ smaller than some value $\alpha(r_h)<\sqrt3/2$ [@Wein; @BFM]. In particular, the extremal RN solution is unstable for $\alpha<\sqrt3/2$ and stable above this value. At the limiting value $\alpha=\sqrt3/2$ the extremal RN solution bifurcates with infinitely many non-abelian solutions and in fact develops infinitely many unstable modes.
Gravitating Sphalerons and Sphaleron Black Holes {#gsphal}
================================================
In complete analogy to gravitating monopoles one may also consider self-gravitating sphalerons, i.e. gravitating versions of the flat-space sphaleron [@Greene]. Although many of the phenomena discussed above for monopoles repeat itself in this case, there are some important differences as far as the domains of existence for sphaleron black holes are concerned. Also the stability properties are clearly different.
to
Again there is a maximal value of the parameter $\alpha$ for which static gravitating sphalerons exist. However, it is a rather different mechanism that is responsible for its existence. For very small values of $\alpha$ the coupling to gravity yields only a small perturbation of the flat-space sphaleron, whose size is $\approx 1/M_W$. As $\alpha$ increases the gravitational attraction leads to shrinking of the sphaleron, until it eventually becomes of size $\approx 1/M_{\rm Pl}$ (l.h.s. of Fig. \[figsph\]).
Following the 1-parameter family of solutions obtained by varying $\alpha$ one finds a similar phenomenon as observed for the gravitating monopoles: $\alpha$ runs through a maximum. In contrast to the situation for the monopole there is however no critical value for $\alpha$ and the family may be continued all the way back to $\alpha=0$. In fact, there is no analogue of the abelian RN solution in this case, since finite mass requires the asymptotic condition $W\to -1$ for $r\to\infty$. The parameter $b(\alpha)$ (compare l.h.s. of Fig. \[figbalf\]) instead increases monotonously and reaches the value $b_1$ of the BM solution from below as $\alpha$ comes back to 0. Thus as $\alpha$ increases the gravitational force becomes stronger and eventually replaces the effect of the Higgs field and we end up with a gravitationally bound sphaleron - the BM solution. The limit $\alpha\to 0$ corresponds to case ii) discussed above. However, Fig. \[figbalf\] shows that there is another branch of solutions starting at $b_1$ with $b(\alpha)\gsim b_1$ for small values of $\alpha$. Looking at the solution for $b\gsim b_1$ we see, that it consists of a tiny BM solution sitting inside an essentially flat-space sphaleron of size $1/M_W$. The solution starts with $W=-1$ at $r=0$, reaches almost $W=1$ within distance $1/M_{\rm Pl}$ and then decreases slowly to $W=-1$ (compare Fig. \[figsph\] r.h.s.). For all these sphaleron solutions $W$ tends to $-1$ for $r\to\infty$ although this is not obvious from the plots, because of the different length scales involved. As we increase $\alpha$ it again runs through a maximum and eventually comes back to 0 at $b=b_2$. This whole story repeats itself, a new branch of solutions starting at each BM solution. The values $b_n$ may be interpreted as points where the $n^{\rm th}$ BM solution bifurcates with the same BM solution having a flat-space sphaleron attached to it at large $r$.
Again there are solutions for $\beta=\infty$ (i.e. $H\equiv v$), which one might call ‘global sphalerons’’ in analogy to the global monopoles of the triplet model.
Besides the globally regular solutions there are also in this case non-abelian black holes. Like their regular counter-parts they carry no charge. As already mentioned above the abelian RN solution is not allowed in this case. For $\alpha\to 0$ these sphaleronic black holes tend to the corrersponding BM-type black holes. Again they exist only in a bounded domain of the $\alpha$-$r_h$-plane ($r_h$ measured in units of $1/M_W$!). The domains for the various radial excitations of the fundamental solution (with $n\geq2$ zeros of $W$) are nested and shrink very quickly with $n$ (compare r.h.s. of Fig. \[figbalf\]).
Stability of sphalerons and sphaleronic black holes {#stabsph}
===================================================
to
Like their flat counter-parts the gravitating sphalerons are unstable with respect to certain variations involving the component $W_2$ of the YM potential [@Bosch].
In the gravitating case we observe however an additional new type of instability involving variations of $W$, i.e. within the minimal ansatz for the YM field. In contrast to the first mentioned instability already present in flat space and which we may call “topological”, the second one may be considered a “gravitational” instability [@Lav]. It sets in at the right turning point of the lowest branch of $b(\alpha)$, i.e. at the maximal value of $\alpha$ of the gravitating version of the flat-space sphaleron. According to a general result of stability theory the whole upper branch between $\alpha_{\rm max}$ and $\alpha=0$ has one unstable mode. At each turning point of the subsequent branches of $b(\alpha)$ another unstable mode appears. This explains the observation that the n$^{\rm th}$ BM solution, obtained when $\alpha\to 0$ on the upper part of the n$^{\rm th}$ branch, has $n$ unstable modes within the minimal ansatz in addition to the (also $n$) topological unstable modes involving $W_2$ [@Volkov].
Analogous results are found for the sphaleronic black holes.
[99]{}
Breitenlohner, P., Forgács, P., and Maison, D.: [[*Nucl. Phys. **B***]{}383]{} (1992) 357\
[[*Nucl. Phys. **B***]{}442]{} (1995) 126
Lichnerowicz, A.: [*Théories Relativiste de la Gravitation et de l’Électromagnétisme*]{},\
Masson, Paris, 1955.
Damour, T.: [*Gravitational Radiation*]{}, N. Deruelle and T. Piran eds., North Holland, Amsterdam, 1983.
M. Colpi, S.L. Shapiro and I. Wasserman: [[*Phys. Rev. Lett. *****]{}57]{} (1986) 2485
Breitenlohner, P., Gibbons, G.W. and Maison, D.: [[*Commun. Math. Phys. *****]{}120]{} (1988) 295
van Nieuwenhuizen, P., Wilkinson, D., and Perry, J.: [[*Phys. Rev. **D***]{}13]{} (1976) 778
Bartnik, R., and McKinnon, J.: [[*Phys. Rev. Lett. *****]{}61]{} (1988) 141
J.A. Smoller, A.G. Wasserman, S.T. Yau, and J.B. McLeod, [[*Commun. Math. Phys. *****]{}143]{} (1991) 115\
Breitenlohner, P., Forgács, P., and Maison, D.: [[*Commun. Math. Phys. *****]{}163]{} (1994) 141
Volkov, M.S. and Gal’tsov, D.V.: [[*Phys. Lett. **B***]{}273]{} (1991) 273\
Sudarsky, D and Wald, R.M.: [[*Phys. Rev. **D***]{}46]{} (1992) 1453
Greene, B.R., Mathur, S.D. and O’Neill, C.M.: [[*Phys. Rev. **D***]{}47]{} (1993) 2242
Lavrelashvili, G. and Maison, D.: [[*Phys. Lett. **B***]{}295]{} (1992) 67
Israel, W.: [[*Phys. Rev. *****]{}164]{} (1967) 1776\
[[*Commun. Math. Phys. *****]{}8]{} (1968) 245
Bizon, P.: [PRL 64]{} (1990) 2844\
Chruściel, P.T.: preprint, gr-qc/9402032
Barriola, M., and Vilenkin, A.: [[*Phys. Rev. Lett. *****]{}63]{} (1989) 341
Manton,N.S.: [PRD 28]{} (1983) 2019
Lee, K.-Y., Nair, V.P., and Weinberg, E J..: [[*Phys. Rev. **D***]{}45]{} (1992) 2751
Aichelburg, P.C., and Bizon, P.: [[*Phys. Rev. **D***]{}48]{} (1993) 607
Hollmann, H.: [[*Phys. Lett. **B***]{}338]{} (1994) 181
Harrison, B.K., Thorne, K.S., Wakano M., and Wheeler, J.A.: [*Gravitation Theory and Gravitational Collapse*]{}.\
Univ. Chicago Press, Chicago, 1965 aeda, K., Tachizawa, T., Torii, T. and Maki, T.: [[*Phys. Rev. Lett. *****]{}72]{} 450 (1994)\
Maeda, K., Tachizawa, T., Torii, T. and Maki, T.: [[*Phys. Rev. Lett. *****]{}72]{} 450 (1994)
Bizon, P. and Wald, R.M.: [[*Phys. Lett. **B***]{}267]{} (1991) 173\
Lee, K., Nair, V.P., and Weinberg, J.: [[*Phys. Rev. Lett. *****]{}68]{} (1992) 1100
Boschung, P., Brodbeck, O., Moser, F., Straumann, N. and Volkov, M.: [[*Phys. Rev. **D***]{}50]{} (1994) 3842
Lavrelashvili, G. and Maison, D.: [[*Phys. Lett. **B***]{}343]{} (1995) 214
Volkov, M.S., Brodbeck, O. Lavrelashvili, G. and Straumann, N.: preprint, hep-th/9502045
[^1]: To appear in the proceedings of the [*Pacific Conference on Gravitation and Cosmology*]{}, Seoul, Feb. 1996
[^2]: e-mail: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The $\mu\nu$SSM solves the $\mu$ problem of the MSSM and explains the origin of neutrino masses by simply using right-handed neutrino superfields. The solution implies the breaking of R-parity. The properties and phenomenology of the model are briefly reviewed.'
author:
- Carlos Muñoz
title: 'Phenomenology of a New Supersymmetric Standard Model: The $\mu\nu$SSM'
---
[ address=[Departamento de Física Teórica UAM and Instituto de Física Teórica UAM/CSIC,\
Universidad Autónoma de Madrid (UAM), Cantoblanco, E-28049 Madrid, Spain]{} ]{}
The LHC will finally start operations at the end of this year. It will be able to answer not only the crucial question about the origin of the mass, but also to clarify whether or not supersymmetry (SUSY) exists. SUSY not only solves several important theoretical problems of the Standard Model, but also has spectacular experimental implications.
However, it is fair to say that SUSY has also its own theoretical problems, and, in particular, a crucial one is the so-called $\mu$-problem of the MSSM. As it is well known, the NMSSM, provides a solution via the introduction of an extra singlet superfield $\hat S$.
On the other hand, neutrino experiments have confirmed during the last years that neutrinos are massive. Thus all models must be modified in order to reproduce this result. The BRpV has been proposed in this context. There, bilinear terms breaking R-parity of the type, $\mu_i \hat L_i \hat H_2$, are added to the MSSM. These induce neutrino masses through the mixing with the neutralinos (actually one mass at tree level and the other two at one loop) withouth including right-handed neutrinos in the model, unlike the MSSM. However, the $\mu$-problem is augmented with the three new bilinear terms.
The “$\mu$ from $\nu$” Supersymmetric Standard Model ($\mu\nu$SSM) was proposed in [@MuNuSSM; @MuNuSSM0] as an alternative to the MSSM, solving the $\mu$-problem and explaining the origin of neutrino masses by simply using right-handed neutrino superfields. The superpotential of the $\mu\nu$SSM contains, in addition to the Yukawas for quarks and charged leptons, Yukawas for neutrinos $\hat H_u\, \hat L \, \hat \nu^c$, terms of the type $\hat \nu^c \hat H_d\hat H_u$ producing an [**effective $\mu$ term**]{} through right-handed sneutrino VEVs of order the electroweak (EW) scale, $\mu\equiv
\lambda_i \langle \tilde \nu^c_i \rangle$, and also terms of the type $\hat \nu^c \hat \nu^c \hat \nu^c$ avoiding the existence of a Goldstone boson and contributing to generate [**effective Majorana masses**]{} ($\sim \kappa \langle \tilde \nu^c_i \rangle$): $$\begin{aligned}
W & = &
\ \epsilon_{ab} \left(
Y_{u_{ij}} \, \hat H_u^b\, \hat Q^a_i \, \hat u_j^c +
Y_{d_{ij}} \, \hat H_d^a\, \hat Q^b_i \, \hat d_j^c +
Y_{e_{ij}} \, \hat H_d^a\, \hat L^b_i \, \hat e_j^c +
Y_{\nu_{ij}} \, \hat H_u^b\, \hat L^a_i \, \hat \nu^c_j
\right)
\nonumber\\
& - &
\epsilon{_{ab}} \lambda_{i} \, \hat \nu^c_i\,\hat H_d^a \hat H_u^b
+
\frac{1}{3}
\kappa{_{ijk}}
\hat \nu^c_i\hat \nu^c_j\hat \nu^c_k \ .
\label{superpotential}\end{aligned}$$ Clearly, the above terms produce the [**explicit breaking of R-parity**]{} (and lepton number) in this model. The size of the breaking can be understood better if we realize that in the limit where neutrino Yukawa couplings $Y_{\nu}$ are vanishing, the $\hat \nu^c$ are ordinary singlet superfields like the $\hat S$ of the NMSSM, without any connection with neutrinos, and this model [**would coincide with the NMSSM**]{} (but with three singlets instead of one) where R-parity is conserved . Once we switch on the $Y_{\nu}$, the $\hat \nu^c$ become right-handed neutrinos, and, as a consequence, R-parity is broken. This breaking has to be small because we have an [**electroweak-scale seesaw**]{}, implying that the values of $Y_{\nu}$ are no larger than $10^{-6}$ (like the electron Yukawa) to reproduce the neutrino masses ($\lsim 10^{-2}$ eV).
Actually, the breaking of R-parity produces the mixing of neutralinos with neutrinos, and as a consequence a generalized $10\times 10$ matrix of the EW seesaw type, $${\mathcal{M}}=\left(\begin{array}{cc}
M & m\\
m^{T} & 0_{3\times3}\end{array}\right),
\label{matrizse}$$ that gives rise at three level to three light eigenvalues corresponding to the neutrino masses [@MuNuSSM; @MuNuSSM2]. Here $M$ is a $7\times 7$ matrix showing the mixing of neutralinos with right-handed neutrinos, and $m$ a $7\times 3$ matrix representing the mixing of neutralinos with left-handed neutrinos, and left- with right-handed neutrinos. The entries of the matrix $M$ are of the order of the EW scale and much larger than the ones of the matrix $m$ which turn out to be of the order of the Dirac masses for the neutrinos ($Y_{\nu} \langle H_u^0 \rangle \lsim 10^{-4}$ GeV). The latter can easily be understood, since the entries of $m$ are proportional to $g\langle \tilde \nu_i \rangle$, $Y_{\nu}\langle \tilde \nu_i^c \rangle$ and $Y_{\nu} \langle H_u^0 \rangle $. On the one hand, $ \langle \tilde \nu_i^c \rangle \sim \langle H_u^0 \rangle$. On the other hand, because of the contribution of $Y_{\nu}\lsim 10^{-6}$ to the minimization conditions for the left-handed neutrinos, the $\langle \tilde \nu_i \rangle$ turn out to be small, and one can check that they are no larger than $Y_{\nu} \langle H_u^0 \rangle$ [@MuNuSSM].
Let us finally remark that having a (dynamical) EW seesaw avoids the introduction of [*ad-hoc*]{} high energy scales in the model, as it occurs in the case of a GUT seesaw.
Notice that the neutrino Yukawas generate, through the $\langle \tilde \nu_j^c \rangle$ three efective bilinear terms, $\mu_i \hat L_i \hat H_2$, with $\mu_i\equiv Y_{\nu_{ij}} \langle \tilde \nu_j^c \rangle \lsim 10^{-4}$ GeV. These [**characterize the BRpV model**]{}, as mentioned above. [**The advantages**]{} of the $\mu\nu$SSM (from our viewpoint) with respect to other popular models proposed in the literature are now more clear. Concerning the $\mu$-problem, one solves it without having to introduce an extra singlet superfield as in the NMSSM. A special form of the Kahler potential as in the Giudice-Masiero mechanism, or specific superpotential couplings to the hidden sector [@condensados; @condensados2], are not necessary either.
Using the eight minimization conditions for the scalar potential (which includes the usual soft, $D$ and $F$ term contributions), one can eliminate e.g. the soft masses $m_{H_d}$, $m_{H_u}$, $m_{\widetilde{L}_{i}}$, and $m_{\widetilde{\nu}_{i}^{c}}$ in favour of the VEVs of the Higgses and neutrinos. We thus consider as independent [**parameters**]{} of the neutral scalar sector [@MuNuSSM2]: $$\lambda, \, \kappa,\, \tan\beta, \, \nu_1, \, \nu_3, \nu^c, \, A_{\lambda}, \, A_{\kappa}, \, A_\nu\ ,
\label{freeparameters2}$$ where $\nu_i\equiv \langle \tilde \nu_i \rangle$, $\nu^c\equiv \langle \tilde \nu^c \rangle $, and we have assumed for simplicity that there is no intergenerational mixing and that in general they have the same values for the three families. In the case of the neutrino parameters, following the discussion in [@MuNuSSM2; @neutrinos], it is sufficient with two generations with different VEVs and couplings in order to reproduce the experimental pattern. Thus we have chosen $Y_{\nu_1} \neq Y_{\nu_2}= Y_{\nu_3}$ and $\nu_1\neq \nu_2=\nu_3$. Actually, we have checked that with $Y_{\nu_2}= Y_{\nu_3} \approx 2 \; Y_{\nu_1} \sim 10^{-6} $ and $\nu_2=\nu_3 \approx 2 \, \nu_1 \sim 10^{-4}$ GeV, the observed neutrino masses and mixing angles are reproduced (thus this result can be obtained even with a diagonal neutrino Yukawa matrix as pointed out in [@Ghosh:2008yh]). As explained in detail in [@neutrinos], this is so easy to get due to the peculiar characteristics of this seesaw, where the relevant scale is the EW one, and R-parity is broken involving not only the right-handed neutrinos in the seesaw but also the MSSM neutralinos. In a sense, this gives an answer to the question why the mixing angles are so different in the quark and lepton sectors. In [@neutrinos], it was also shown that [**spontaneous CP violation**]{} through complex VEVs is possible in the $\mu\nu$SSM at tree level.
The parameter space of the model was analyzed in detail in [@MuNuSSM2], studying the viable regions which avoid false minima and tachyons, as well as fulfill the Landau pole constraint. The structure of the [**mass matrices**]{} and the associated [**particle spectrum**]{} was also computed. The breaking of $R$-parity generates complicated mass matrices and mass eigenstates, as we already saw above for the case of the neutralinos/neutrinos. The charginos mix with the charged leptons giving rise to a 5$\times$5 matrix. Nevertheless, there will always be three light eigenvalues corresponding to the electron, muon and tau. Concerning the scalar mass matrices, the neutral Higgses are mixed with the sneutrinos, and the charged Higgses with the charged sleptons, and we are left with fifteen (eight CP-even and seven CP-odd) neutral states and seven charged states. Notice nevertheless that the three left handed sneutrinos are decoupled from the Higgs-right handed sneutrinos, and also the six charged sleptons from the charged Higgses. The upper bound for the lightest Higgs boson mass turns out to be similar to the one of the NMSSM, about 140 GeV after imposing the Landau pole constraint up to the GUT scale. Obviously, the [**phenomenology**]{} of models where $R$-parity is broken differs substantially from that of models where $R$-parity is conserved. Needless to mention, the lightest supersymmetry particle (LSP) is no longer stable, and therefore not all SUSY chains must yield missing energy events at colliders. In [@Ghosh:2008yh], the decays of the lightest neutralino to two body ($W$-lepton) final states were discussed. The correlations of the decay branching ratios with the neutrino mixing angles were studied as another possible test of the $\mu\nu$SSM at the LHC. The phenomenology of the $\mu\nu$SSM was also analyzed in [@Hirsch0], particularized for one and two generations of right-handed sneutrinos, and taking into account all possible final states when studying the decays of the lightest neutralino. Possible signatures that might allow to distinguish this model from other R-parity breaking models were discussed qualitatively in these two works [@Ghosh:2008yh; @Hirsch0].
Let us finally discuss potential problems of the $\mu\nu$SSM and their possible solutions.
Since R-parity is broken, one could add in the superpotential the usual lepton and baryon number violating terms, $LLe^c + LQd^c$ and $d^cd^cu^c$, producing fast [**proton decay**]{} (the new terms of the $\mu\nu$SSM are obviously harmless with respect to proton decay). Nevertheless, the choice of $R$-parity is [*ad hoc*]{}. There are other discrete symmetries, like e.g. baryon triality which only forbids the baryon violating operators. Obviously, for all these symmetries R-parity is violated. Besides, in superstring constructions the matter superfields can be located in different sectors of the compact space or have different extra $U(1)$ charges, in such a way that some operators violating $R$-parity can be forbidden [@old], but others can be allowed. Let us remark here that even if the terms $LQd^c$ are set to zero at a high-energy scale, one-loop corrections in the $\mu\nu$SSM will generate them. Nevertheless, these contributions are very small, as shown in [@MuNuSSM2].
In the $\mu\nu$SSM the usual bilinear $\mu$ term of the MSSM, as well as Majorana masses for neutrinos are absent from the superpotential (\[superpotential\]), and [**only dimensionless trilinear couplings are present**]{} (the EW scale of the breaking being determined by the soft terms in the scalar potential). For this to happen we can invoke a $Z_3$ symmetry as it is usually done in the NMSSM. Nevetheless, let us recall that this is actually what happens in superstring constructions, where the low-energy limit is determined by the massless superstring modes. Since the massive modes have huge masses, of the order of the string scale, only the trilinear couplings for the massless modes are relevant. Since the superpotential has a $Z_3$ symmetry, one would expect to have also a cosmological [**domain wall**]{} problem like in the NMSSM. Nevertheless, the usual solutions to this problem will also work in this case: non-renormalizable operators can break explicitly the dangerous $Z_3$ symmetry, lifting the degeneracy of the three original vacua, and this can be done without introducing hierarchy problems. In addition, these operators can be chosen small enough as not to alter the low-energy phenomenology.
When lepton number is broken, [**flavour violating processes**]{} are possible. Although there are strong experimental constraints, these are fulfilled in the $\mu\nu$SSM once neutrino data are imposed, similar to the case of BRpV [@mario].
We mentioned above that when R-parity is broken the LSP is not stable. Thus the neutralino or the sneutrino, with very short lifetimes, are no longer candidates for the [**dark matter**]{} (DM) of the Universe. Nevertheless, other SUSY particles such as the gravitino or the axino can still be used since their lifetimes are typically very long. Concerning the [**gravitino**]{}, it has an interaction term in the supergravity Lagrangian with the photon and the photino. Since the photino and the left-handed neutrinos are mixed due to the breaking of R-parity, the gravitino will be able to decay into a photon and a neutrino. The decay is suppressed both by the gravitational interaction and by the small R-parity violating coupling, thus its lifetime can be much longer than the age of the Universe [@Takayama:2000uz]. Therefore, the gravitino can be a DM candidate. Since the gravitino decays producing a monochromatic photon with an energy half of the gravitino mass, the prospects for detecting these gamma rays in satellite experiments were analyzed in the context of bilinear R-parity violation scenarios in the literature. In a recent work [@recentgravitino], gravitino DM and its possible detection in the FERMI satellite were discussed in the context of the $\mu\nu$SSM. Gravitino masses larger than $20$ GeV are disfavored by the diffuse photon background measurements, but a gravitino with a mass range between $0.1 - 20$ GeV gives rise to a signal that might be observed by the FERMI satellite.
[9]{}
D. E. López-Fogliani and C. Muñoz, [*Phys. Rev. Lett.*]{} [**97**]{} (2006) 041801.
C. Muñoz, unpublished notes (1994).
N. Escudero, D. E. Lopez-Fogliani, C. Muñoz and R. R. de Austri, [*JHEP*]{} [**12**]{} (2008) 099. J.E. Kim and H.P. Nilles, [*Phys. Lett.*]{} [**B263**]{} (1991) 79; E.J. Chun, J.E. Kim and H.P. Nilles, [*Nucl. Phys.*]{} [**B370**]{} (1992) 105.
J.A. Casas and C. Muñoz, [*Phys. Lett.*]{} [**B306**]{} (1993) 288.
J. Fidalgo, D. E. López-Fogliani, C. Muñoz and R. R. de Austri, [*JHEP*]{} [**08**]{} (2009) 105.
P. Ghosh and S. Roy, [*JHEP*]{} [**04**]{} (2009) 069. A. Bartl, M. Hirsch, A. Vicente, S. Liebler and W. Porod, [*JHEP*]{} [**05**]{} (2009) 120.
J.A. Casas, E.K. Katehou and C. Muñoz, Oxford preprint, Nov. 1987, Ref: 1/88; [*Nucl. Phys.*]{} [**B317**]{} (1989) 171; J.A. Casas and C. Muñoz, [*Phys. Lett.*]{} [**B212**]{} (1988) 343.
D.F. Carvalho, M.E. Gomez and J.C. Romao, [*Phys. Rev.*]{} [**D65**]{} (2002) 093013.
F. Takayama and M. Yamaguchi, [*Phys. Lett.*]{} [**B485**]{} (2000) 388.
K.Y. Choi, D.E. López-Fogliani, C. Muñoz and R. Ruiz de Austri, arXiv:0906.3681\[hep-ph\].
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'A new non-parametric method based on Gaussian Processes (GP) was proposed recently to measure the Hubble constant $H_0$. The freedom in this approach comes in the chosen covariance function, which determines how smooth the process is and how nearby points are correlated. We perform coverage tests with a thousand mock samples within the $\Lambda$CDM model in order to determine what covariance function provides the least biased results. The function Matérn(5/2) is the best with sligthly higher errors than other covariance functions, although much more stable when compared to standard parametric analyses.'
---
Introduction
============
The difference between the value of the Hubble constant $H_0$ determined by [*Planck*]{} [@planck (Planck Collaboration 2013)] and by local measurements [@riess (Riess et al. 2011)] shows a 2.3$\sigma$ tension. In order to understand what could generate such a discrepancy, many attempts in the literature were done searching for new physics or systematic errors (e.g. [@marra Marra et al. 2013], [@spergel Spergel et al. 2013], [@efs Efstathiou 2014], [@neutrinos Wyman et al. 2014], [@holanda Holanda et al. 2014], [@clarkson Clarkson et al. (2014)]).
Recently, we proposed a new method to determine $H_0$ by applying [*Gaussian Processes*]{} (GP), which is a non-parametric procedure, to reconstruct $H(z)$ data and extrapolating to redshift zero [@bcs2014 (Busti et al. 2014)]. We selected 19 $H(z)$ measurements ([@simon2005 Simon et al. 2005], [@stern2010 Stern et al. 2010], [@moresco2012 Moresco et al. 2012]) based on cosmic chronometers and obtained $H_0= 64.9 \pm 4.2$ km s$^{-1}$ Mpc$^{-1}$, which is compatible with [*Planck*]{} but shows a tension with local measurements.
Here, we use mock samples in order to test our method. Basically, we are interested to see which covariance function adopted in the GP analysis provides the best match with the fiducial cosmological model. As we shall see, the Matérn(5/2) performs better, with the standard squared exponential covariance function underestimating the errors. In the next section GP will be briefly described, showing our results in Sec. 3. We draw our conclusions and discuss future improvements in Sec. 4.
Gaussian Processes (GP)
=======================
A gaussian process allows one to reconstruct a function from data without assuming a parametrisation for it. While a gaussian distribution is a distribution over random variables, a gaussian process is a distribution over functions. We use GaPP (Gaussian Processes in Python)[^1] [@gapp (Seikel et al. 2012))] in order to reconstruct the Hubble parameter as a function of the redshift.
Basically, the reconstruction is given by a mean function with gaussian error bands, where the function values at different points $z$ and $\tilde{z}$ are connected through a covariance function $k(z,\tilde{z})$. This covariance function depends on a set of hyperparameters. For example, the general purpose squared exponential (Sq. Exp.) covariance function is given by $$k(z,\tilde{z}) = \sigma_f^2 \exp\left\{-\frac{(z-\tilde{z})^2}{2l^2}\right\}.$$ In the above equation we have two hyperparameters, the first $\sigma_f$ is related to typical changes in the function value while the second $l$ is related to the distance one needs to move in input space before the function value changes significantly. We follow the steps of [@gapp] and determine the maximum likelihood value for $\sigma_f$ and $l$ in order to obtain the value of the function. In this way, we are able to reconstruct the Hubble parameter as a function of the redshift from $H(z)$ measurements. Many choices of covariance function are possible, and we consider a variety below.
Results
=======
The freedom in the GP approach comes in the covariance function. While in traditional parametric analyses we choose a model to characterise what is our prior belief about the function in which we are interested, with GP we ascribe in the covariance function our priors about the expected function properties (e.g. smoothness, correlation scales etc.).
We consider the Sq. Exp. covariance function and three examples from the Matérn family:
$$k(z,\tilde{z}) = \sigma_f^2 \frac{2^{1-\nu}}{\Gamma(\nu)} \left( \frac{\sqrt{2\nu(z-\tilde{z})^2}}{l} \right)^{\nu} K_{\nu}\left( \frac{\sqrt{2\nu(z-\tilde{z})^2}}{l} \right),$$
where $K_{\nu}$ is a modified Bessel function and we choose $\nu=5/2$, $7/2$ and $9/2$ (see [@sc2013 Seikel & Clarkson 2013 for more discussions]). Writing $\nu = p + 1/2$, each Matérn function is $p$ times differentiable as are functions drawn from it, and the squared exponential is recovered for $\nu \rightarrow \infty$. Increasing $\nu$ increases the width of the covariance function near the peak implying stronger correlations from nearby points for a fixed correlation length $\ell$. For comparison purposes, we also consider two standard parametric models: a flat $\Lambda$CDM model and a flat XCDM model.
The results are shown in Table \[table1\] together with the constraints from the 19 $H(z)$ data. The coverage test of each covariance function and parametric model was performed by creating 1000 mock catalogues of 19 data points with the same redshifts and error-bars of the measured points in a fiducial $\Lambda$CDM model. For each model realisation a value of $H_0$ was derived. The third and fourth columns of Table \[table1\] show the frequency the true value for $H_0$ was recovered inside the $1\sigma$ and $2\sigma$ regions. So, for example, the Matérn$(9/2)$ covariance function captures the true value at 1(2)$\sigma$ about 60%(94%) of the time – alternatively, the 1(2)$\sigma$ region should be interpreted as a 60%(94%) confidence interval. This provides a way to re-normalise the $n\sigma$ intervals for a given covariance function and a prior model assumption, which we show between parentheses for 1$\sigma$(68%) errors in Table \[table1\]. Therefore, this is an attempt to quantify a possible systematic error from the covariance functions *assuming the true model is $\Lambda$CDM*. We also considered some different fiducial models with a time-varying dark energy equation of state, 64 data points in the redshift range $0.1 < z < 1.8$, with coverages showing the same pattern as depicted in Table \[table1\]. It is important to note this is a model-dependent comparison which relies on the knowledge of the true model in advance, which is never the case, and changes with the quality of the data. The coverage can change with a different underlying model as well – but note that the errors are actually much more stable than switching from $\Lambda$CDM to XCDM.
[@ccccc@]{} Method & $H_0$ $\pm$ $1\sigma$ & Coverage & Coverage\
& (km s$^{-1}$ Mpc$^{-1}$) & 1$\sigma$ & 2$\sigma$\
\
Sq. Exp. & 64.9 $\pm$ 4.2(5.9) & 0.527 & 0.905\
Matérn$(9/2)$ & 65.9 $\pm$ 4.5(5.6) & 0.594 & 0.939\
Matérn$(7/2)$ & 66.4 $\pm$ 4.7(5.7) & 0.610 & 0.946\
Matérn$(5/2)$ & 67.4 $\pm$ 5.2(5.5) & 0.665 & 0.959\
$\Lambda$CDM & 68.9 $\pm$ 2.8 & 0.676 & 0.938\
XCDM & 69.0 $\pm$ 6.7 & 0.685 & 0.939\
Conclusions
===========
We have applied GP to reconstruct $H(z)$ data and from it extrapolate to redshift zero to obtain $H_0$ [@bcs2014 (Busti et al. 2014)]. Based on a set of 1000 mock samples, we have tested the method assuming a fiducial flat $\Lambda$CDM model by considering four different covariance functions and applying a coverage test. We have shown Matérn(5/2) represents better the errors, with errors slightly higher than the other covariance functions. A heuristic method to recalibrate the errors for different covariance functions was also provided within the $\Lambda$CDM model.
Possible improvements can be achieved by marginalizing over the hyperparameters and comparing the results using Bayesian model comparison tools, which will allow a direct assessment of performance with no need to rely on a fiducial model.
2014, *MNRAS* (Letters) 441, L11 \[preprint(arXiv:1402.5429)\]
2014, preprint(arXiv:1405.7860)
2014, *MNRAS*, 440, 1138
2014, *MNRAS* (Letters), 443, L74 \[preprint(arXiv:1303.5076)\]
2013, *Phys. Rev. Lett.*, 110, 241305
2012, *J. Cosmol. Astropart. Phys.*, 8, 6
2013, preprint(arXiv:1303.5076)
2011, *ApJ*, 730, 119
2012, *J. Cosmol. Astropart. Phys.*, 6, 36
2013, preprint(arXiv:1311.6678)
2005, *Phys. Rev. D*, 71, 123001
2013, preprint(arXiv:1312.3313)
2010, *J. Cosmol. Astropart. Phys.*, 2, 8
2014, *Phys. Rev. Lett.*, 112, 051302
[^1]: http://www.acgc.uct.ac.za/[.17ex]{}seikel/GAPP/index.html
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We present a generic approach that allows us to develop a fully polynomial-time approximation scheme (FTPAS) for minimizing nonlinear functions over the integer points in a rational polyhedron in fixed dimension. The approach combines the subdivision strategy of Papadimitriou and Yannakakis [@papadimitriou_approximability_2000] with ideas similar to those commonly used to derive real algebraic certificates of positivity for polynomials. Our general approach is widely applicable. We apply it, for instance, to the Motzkin polynomial and to indefinite quadratic forms $\ve x^T Q \ve x$ in a fixed number of variables, where $Q$ has at most one positive, or at most one negative eigenvalue. In dimension three, this leads to an FPTAS for general $Q$.'
author:
- 'Robert Hildebrand[^1]'
- 'Robert Weismantel[^2]'
- 'Kevin Zemmer[^3]'
bibliography:
- 'references.bib'
date: 'October 10, 2015'
title: An FPTAS for Minimizing Indefinite Quadratic Forms over Integers in Polyhedra
---
Introduction
============
Consider the problem $$\label{eq:problem-general}
\min \{ f(\ve x) : \ve x \in P \cap {\mathbb{Z}}^n\}$$ for $f \colon {\mathbb{R}}^n \to {\mathbb{R}}$, $P = \{ \ve x \in {\mathbb{R}}^n \,:\, A \ve x \le \ve b \}$, $A \in {\mathbb{Z}}^{m \times n}$, and $\ve b \in {\mathbb{Z}}^m$. We use the words *size* and *binary encoding length* synonymously. The size of $P$ is the sum of the sizes of $A$ and $\ve b$. We say that Problem can be solved in polynomial time if in time bounded by a polynomial in the size of its input, we can either determine that the problem is infeasible, find a feasible minimizer, or show that the problem is unbounded by exhibiting a feasible point $\ve{\bar x}$ and an integer ray $\ve{\bar r} \in \operatorname{rec}(P) := \{ \ve x\in {\mathbb{R}}^n : A \ve x \leq 0\}$ such that $f(\ve{\bar x} + \lambda \ve{\bar r}) \to -\infty$ as $\lambda \to \infty$.
The main focus of this paper is Problem with $f(\ve x) = \ve x^T Q \ve x$, where $Q \in {\mathbb{Z}}^{n \times n}$ is a symmetric matrix. Note that if $Q$ is not symmetric, then we can replace it by $Q' = \frac{1}{2} Q + \frac{1}{2} Q^T$, which is symmetric and satisfies $\ve x^T Q \ve x = \ve x^T Q' \ve x$. The input size of Problem with $f(\ve x) = \ve x^T Q \ve x$ is the sum of the sizes of $P$ and $Q$. When $Q$ is positive semi-definite, $f(\ve x)$ is convex, whereas it is concave when $Q$ is negative semi-definite. In the former case, Problem with fixed $n$ and bounded $P$ can be solved in polynomial time by [@khackiyan_integer_2000], whereas in the latter case by [@cook-hartmann-kannan-mcdiarmid-1992] (cf. Theorem \[thm:convex-concave\] below).
The computational complexity of mixed-integer polynomial optimization in fixed dimension was surveyed in [@loera_integer_2006; @loera_mixed_integer_2008], and they develop an FPTAS for maximizing non-negative polynomials over integer and mixed-integer points in a polytope, respectively. They also use a weaker notion of approximation algorithm, that has also been used in [@vavasis_approximation_1992], called *fully polynomial-time weak-approximation scheme*. It allows the accuracy to depend on the maximum and minimum values of $f$ on the feasible set. That is, let $f_{\max}$ and $f_{\min}$ be the maximum and minimum values of $f$ on the feasible region. Then a weak $\epsilon$-approximation $\ve x_\epsilon$ of the minimization problem satisfies $$| f(\ve x_\epsilon) - f_{\min} | \leq \epsilon (f_{\max} - f_{\min}).$$
When $n \le 2$ and $f$ is a polynomial of degree $d$, it has been shown in [@delpia_integer_2013] and [@delpia_minimizing_2015] that Problem can be solved in polynomial time for $d \le 3$ and in [@loera_integer_2006] that it is NP-hard for $d = 4$. To the best of our knowledge, the complexity of Problem for fixed $n \ge 3$ and when $f$ is a polynomial of degree $d=2,3$ is still unknown.
We present an FPTAS for Problem with $f(\ve x) = \ve x^T Q \ve x$ and fixed $n$ in two cases: For $Q$ with at most one negative eigenvalue, and for $Q$ with at most one positive eigenvalue. The numbers of positive, negative, and zero eigenvalues of $Q$ from a triple known as the *inertia* of $Q$ and can be computed in polynomial time (see [@bunch_partial_1974] or [@hartung_computational_2007]).
By [@delpia_mixed_2014], for fixed $n$, it is possible to decide in polynomial time if Problem with $f(\ve x) = \ve x^T Q \ve x$ is bounded or not. If it is unbounded, [@delpia_mixed_2014] gives a certificate of unboundedness, whereas if it is bounded, it returns an upper bound $\tau$ on the size of an optimal solution that is polynomial in the input size. Thus we can replace $P$ by $P \cap \{ \ve x \in {\mathbb{R}}^n \,:\, \| \ve x \|_\infty \le 2^{\tau} \}$. Henceforth we assume that $P$ is bounded.
We use a notion of approximation that is common in combinatorial optimization and is akin to the maximization version used in [@loera_integer_2006; @loera_mixed_integer_2008] for maximizing non-negative polynomials over polyhedra, except here, we extend the notion to allow for the objective to take negative values.
Let $\ve x_{{\min}}$ be an optimal solution to Problem and let $\epsilon > 0$. We say that $\ve x_\epsilon$ is an *$\epsilon$-approximate solution* to Problem if $\ve x_\epsilon$ is feasible and one of the following hold:
1. $f(\ve x_{{\min}}) > 0$ and $f(\ve x_\epsilon) \leq (1+\epsilon) f(\ve x_{{\min}})$,
2. $f(\ve x_{{\min}}) < 0$ and $f(\ve x_\epsilon) \leq \frac{1}{1 + \epsilon} f(\ve x_{{\min}})$,
3. $f(\ve x_{{\min}}) = 0$ and $f(\ve x_\epsilon) = f(\ve x_{{\min}})$.
We say an algorithm for Problem is a *fully polynomial-time approximation scheme* if for any $\epsilon > 0$, in polynomial time in $\tfrac{1}{\epsilon}$ and the size of the input, the algorithm correctly determines whether the problem is feasible, and if it is, outputs an $\epsilon$-approximate solution $\ve x_\epsilon$.
In order to develop an FPTAS for classes of nonlinear functions to be minimized over integer points in polyhedra, we propose a framework that combines the techniques of Papadimitriou and Yannakakis [@papadimitriou_approximability_2000] with ideas similar to those commonly used to derive certificates of positivity for polynomials over semialgebraic sets. Generally speaking, in the latter context one is given a finite number of “basic polynomials” $f_1,\ldots,f_m$ which are known to be positive over the integers in a polyhedron $P$. A sufficient condition to prove that another polynomial $f$ is positive over $P \cap {\mathbb{Z}}^n$ is to find a decomposition of $f$ as a sum of products of a *sum of squares* (SOS) polynomial and a basic function $f_i$. A polynomial $p(\ve x)$ is SOS if there exist polynomials $q_1(\ve x), \ldots, q_m(\ve x)$ such that $p(\ve x) = \sum_{i = 1}^m q_i^2(\ve x)$.
We would like to use a similar approach to arrive at an FPTAS. Again we work with classes of “basic functions”. Then, for a given $f$, we try to detect a decomposition of $f$ as a finite sum of products of a so-called “sliceable function” and a basic function $f_i$. Roughly speaking, sliceable functions — thanks to the result of [@papadimitriou_approximability_2000] — can be approximated by subdividing the given polyhedron.
For instance, the set of all convex functions presented by a first order oracle that are nonnegative over $P \cap {\mathbb{Z}}^n$ could serve as a class of basic functions, because we can solve Problem for any member in the class in polynomial time when $n$ is fixed. The nonnegativity assumption implies sign-compatibility, which is a necessary property of the set of basic functions that will be introduced in Section \[sec:sliceable\]. Another example is the set of all concave functions presented by an evaluation oracle that are nonnegative over $P \cap {\mathbb{Z}}^n$. The same property holds true in this case. These two examples demonstrate that we consider not only polynomials $f_i$, but also more general classes of basic functions. In fact, this is key to tackle the quadratic optimization problem. For example, we can decompose the polynomial $x^2 + y^2 - z^2$ as the product of two non-polynomial functions: a basic function $\sqrt{x^2 + y^2} - |z|$ and a sliceable function $\sqrt{x^2 + y^2} + |z|$. Our technique also applies, for instance, to the Motzkin polynomial, but to functions $f$ that are not polynomials as well (see Theorem \[thm:sliceable-product-sum\]).
We postpone the discussion of what we mean by basic and sliceable functions to Section \[sec:sliceable\]. As a consequence of our technique we easily derive the following result that in the optimization community was an open question for quite several years.
\[thm:main\] Let $Q \in {\mathbb{Z}}^{n \times n}$ be a symmetric matrix and let $n$ be fixed. Then there is an FPTAS for Problem with $f(\ve x) = \ve x^T Q \ve x$ in the following cases:
(i) $Q$ has at most one negative eigenvalue; \[itm:main\_min\]
(ii) $Q$ has at most one positive eigenvalue. \[itm:main\_max\]
We present a proof for Theorem \[thm:main\] in Section \[sec:quadratics\]. Note that Theorem \[thm:main\] (\[itm:main\_max\]) is equivalent to maximizing $\ve x^T Q \ve x$ with at most one negative eigenvalue. Together with known results on quasi-convex and quasi-concave optimization, Theorem \[thm:main\] implies that for $n = 3$ there is an FPTAS for general $Q$.
\[cor:dim3\] Let $Q \in {\mathbb{Z}}^{n \times n}$ be a symmetric matrix and $n = 3$. Then there is an FPTAS for Problem with $f(\ve x) = \ve x^T Q \ve x$.
In order to prove this corollary, we need the following known results and notation, which are also used for our other results. Given a convex set $C$, a function $f$ is *quasi-convex* on $C$ if the set $\{ \ve x \in C \,:\, f(\ve x) \le \alpha \}$ is convex for all $\alpha \in {\mathbb{R}}$. Convex functions are quasi-convex on any convex set in their domain. We say that $f$ is *quasi-concave* if $-f$ is quasi-convex. A *semi-algebraic set* in ${\mathbb{R}}^n$ is a subset of the form $\bigcup_{i=1}^s \bigcap_{j=1}^{r_i} \left\{ \ve x \in {\mathbb{R}}^n \,\, | \,\, f_{i,j}(\ve x)\ *_{i,j}\ 0 \right\}$ where $f_{i,j} \colon {\mathbb{R}}^n \to {\mathbb{R}}$ is a polynomial in $n$ variables and $*_{i,j}$ is either $<$ or $=$ for $i = 1, \dots, s$ and $j = 1, \dots, r_i$ (cf. [@bochnak_real_1998]). For a polyhedron $P$, denote by $P_I$ the *integer hull* of $P$, i.e., the convex hull of $P \cap {\mathbb{Z}}^n$.
\[thm:convex-concave\] For $A \in {\mathbb{Z}}^{m \times n}$ and $\ve b \in {\mathbb{Z}}^m$, let $P = \{ \ve x \in {\mathbb{R}}^n \,:\, A \ve x \le \ve b \}$ be a bounded polytope, and let $n$ be fixed.
(i) If $C$ is a convex, semi-algebraic set given by polynomial inequalities of degree at most $d$ and with integral coefficients of size at most $l$, and $f$ is a polynomial which is quasi-convex on $P$, then $\min \{ f(\ve x) : \ve x \in P \cap C \cap {\mathbb{Z}}^n\}$ can be solved in polynomial time in $d$, $l$ and the size of $P$ and the size of the coefficients of all involved polynomials.\[itm:quasi-convex-semi-algebraic\]
(ii) The vertices of $P_I$ can be enumerated in polynomial time in the size of $P$.\[itm:vertices-integer-hull\]
(iii) If a comparison oracle for $f$ is available, i.e., an oracle that given $\ve x, \ve y \in {\mathbb{Z}}^n$ decides whether $f(\ve x) < f(\ve y)$ holds or not, and $f$ is quasi-concave on $P_I$, then Problem can be solved in polynomial time in the size of $P$.\[itm:quasi-concave\]
The inertia of $Q$ can be computed in polynomial time (see [@bunch_partial_1974] or [@hartung_computational_2007]). If $Q$ has no negative eigenvalues, then $f$ is convex because it is the sum of convex functions, so we can apply Theorem \[thm:convex-concave\] (\[itm:quasi-convex-semi-algebraic\]) with $C = P$. If $Q$ has no positive eigenvalues, then $f$ is concave, so we can can apply Theorem \[thm:convex-concave\] (\[itm:quasi-concave\]). If $Q$ has one negative eigenvalue, we can apply Theorem \[thm:main\] (\[itm:main\_min\]), whereas if there are two, we can apply Theorem \[thm:main\] (\[itm:main\_max\]).
The remainder of this paper is outlined as follows: In Section \[sec:sliceable\] we explain basic and sliceable functions and present the main tool of this paper. In Section \[sec:quadratics\] we give some important properties of quadratics and prove Theorem \[thm:main\].
Basic and Sliceable Functions {#sec:sliceable}
=============================
Let $\mathcal P$ be a set of polytopes that is closed under taking subsets, i.e., such that for all $\bar P \subseteq P$, $P \in \mathcal P$ implies $\bar P \in \mathcal P$. A set of *basic functions* $\Omega_{\mathcal P}$ is a class of functions with a common encoding structure for all elements and the following properties:
(i) $\Omega_{\mathcal P}$ is closed under addition and multiplication by $\lambda \ge 0$,
(ii) each $f \in \Omega_{\mathcal P}$ admits a FPTAS for Problem over each polytope $P \in \mathcal P$,
(iii) all functions in $\Omega_{\mathcal P}$ are sign compatible, i.e., $\operatorname{sign}(f(\ve x)) = \operatorname{sign}(g(\ve x))$ for all $f, g \in \Omega_{\mathcal P}$, $\ve x \in P$, $P \in \mathcal P$.
Examples of classes of functions $\Omega_{\mathcal P}$ that satisfy the properties above are the set of sign-compatible convex function over $\mathcal P$ presented by a first order oracle (cf. [@Grotschel1988; @Oertel2014]), sign-compatible concave functions over $\mathcal P$ presented by an evaluation oracle (cf. [@cook-hartmann-kannan-mcdiarmid-1992]), and the set of polynomials that are negative on some polytope over $\mathcal P$ (cf. [@loera_integer_2006; @loera_mixed_integer_2008]).
We next introduce a class of functions that can be combined with basic functions and preserve the property that they admit a FPTAS for Problem . For this, we need the following notation.
For $\ve k \in {\mathbb{Z}}^l$, $\ve c^0 \in {\mathbb{Z}}^l$, $\ve c^1, \ldots, \ve c^l \in {\mathbb{Z}}^n$, and $\epsilon > 0$, define $$\begin{gathered}
B_{\ve k}(\ve c^0, \ve c^1, \ldots, \ve c^l, \epsilon) := \Big\{
\ve x \in {\mathbb{R}}^n
\ :\ \\
\begin{cases}
(1 + \epsilon)^{k_j-1} \le L_j(\ve x) \le (1 + \epsilon)^{k_j}
& \textrm{if } k_j \ge 1 \\
L_j(\ve x) = 0
& \textrm{if } k_j = 0 \\
-(1 + \epsilon)^{-k_j-1} \le L_j(\ve x) \le -(1 + \epsilon)^{-k_j}
& \textrm{if } k_j \le -1
\end{cases} \\
\textrm{for } j = 1, \ldots, l
\Big\}.\end{gathered}$$ where $L_j(\ve x) := \ve c^j \cdot \ve x + c^0_j$. Let $B(\ve c^0, \ve c^1, \ldots, \ve c^l, \epsilon) := \bigcup_{\ve k \in {\mathbb{Z}}^l} B_{\ve k}(\ve c^0, \ve c^1, \ldots, \ve c^l, \epsilon)$. Note that for all $\ve c^0 \in {\mathbb{Z}}^l$, $\ve c^1, \ldots, \ve c^l \in {\mathbb{Z}}^n$, and $\epsilon > 0$ we have that ${\mathbb{Z}}^n \subseteq B(\ve c^0, \ve c^1, \ldots, \ve c^l, \epsilon)$. However, ${\mathbb{R}}^n \not \subseteq B(\ve c^0, \ve c^1, \ldots, \ve c^l, \epsilon)$, because $B_{\ve k}(\ve c^0, \ve c^1, \ldots, \ve c^l, \epsilon)$ is not full-dimensional if any entry of $\ve k$ is equal to zero.
\[def:sliceable\] Let $C \subseteq {\mathbb{R}}^n$. A function $s\colon C \to {\mathbb{R}}_{\ge 0}$ is called *sliceable* over $C$ if there exist $\ve c^0 \in {\mathbb{Z}}^l$, $\ve c^1, \ldots, \ve c^l \in {\mathbb{Z}}^n$ and $\zeta \in {\mathbb{Z}}_{\ge 1}$ such that for all $\epsilon > 0$ and for all $\ve k \in {\mathbb{Z}}^l$, we have that $$\label{eq:sliceable}
s(\ve x) \le (1 + \epsilon) s(\ve y)$$ for all $\ve x, \ve y \in C \cap B_{\ve k}(\ve c^0, \ve c^1, \ldots, \ve c^l, \epsilon / \zeta) \cap {\mathbb{Z}}^n$.
We define the size of a sliceable function $s$ as the sum of the sizes of $\ve c^0, \ve c^1, \ldots, \ve c^l$ and the unary encoding size of $\zeta$.
It follows from Definition \[def:sliceable\] that if a function $s$ is sliceable over $C$ with parameters $\ve c^0, \ve c^1, \ldots, \ve c^l, \zeta$, and $s(\bar{\ve x}) = 0$ for some $\bar {\ve x} \in C \cap B_{\ve k}(\ve c^0, \ve c^1, \ldots, \ve c^l, \epsilon / \zeta) \cap {\mathbb{Z}}^n$, then $s(\ve x) = 0$ for all $\ve x \in C \cap B_{\ve k}(\ve c^0, \ve c^1, \ldots, \ve c^l, \epsilon / \zeta) \cap {\mathbb{Z}}^n$. Regardless of this propagation property, there exist large classes of sliceable functions. We present several examples of non-trivial sliceable functions below. To avoid the propagation of the zeros, we give choices of the parameters $\ve c^0, \ve c^1, \ldots, \ve c^l, \zeta$ such that the zeros of the function lay on intersections of hyperplanes defined by $L_j(\ve x) = 0$ for $j \in J$ for some $J \subseteq 1, \ldots, l$.
The set of sliceable functions is not closed under additive inverses, but is in fact a proper semifield as the following lemma shows.
\[lem:sliceable-properties\] Let $s, r \colon C \to {\mathbb{R}}_{\ge 0}$ be sliceable functions and let $\lambda \in {\mathbb{R}}_{\ge 0}$. Then $\lambda s$, $s + r$ and $s \, r$ are sliceable with size polynomial in the size of $s$ and $r$. Moreover, if $s > 0$, then $\frac{1}{s}$ is sliceable with size polynomial in the size of $s$.
Let $s$ be sliceable with parameters $\ve c^0$, $\ve c^1$, …, $\ve c^{l_s}$, $\zeta_s$ and $r$ with parameters $\ve d^0$, $\ve d^1$, …, $\ve d^{l_r}$, $\zeta_r$. Set $\zeta_{s + r} = \max\{ \zeta_s, \zeta_r\}$ and $\zeta_{sr} = 4 \max\{\zeta_s, \zeta_r\}$. We show that $s+r$ and $sr$ are sliceable with the same parameters $[\ve c^0; \ve d^0], \ve c^1, \ldots, \ve c^{l_s}, \ve d^1, \ldots, \ve d^{l_r}$ and with $\zeta_{s+r}, \zeta_{sr}$, respectively. Then for $\zeta = \zeta_{s + r}, \zeta_{s r}$ and $0 < \epsilon < 1$, $$\begin{gathered}
B_{[\ve k^1; \ve k^2]}([\ve c^0; \ve d^0], \ve c^1, \ldots, \ve c^{l_s}, \ve d^1, \ldots, \ve d^{l_r}, \epsilon/ \zeta)
\\
\subseteq B_{\ve k^1}(\ve c^0, \ve c^1, \ldots, \ve c^{l_s}, \epsilon/\zeta ) \cap B_{\ve k^2}( \ve d^0, \ve d^1, \ldots, \ve d^{l_r}, \epsilon/\zeta).\end{gathered}$$
Let $
\ve x, \ve y \in B_{[\ve k^1; \ve k^2]}\big([\ve c^0; \ve d^0], \ve c^1, \ldots, \ve c^{l_s}, \ve d^1, \ldots, \ve d^{l_r},
\allowbreak
\epsilon/ \zeta\big) \cap {\mathbb{Z}}^n.
$ By the containment, it follows that $$s(\ve x) \leq (1 + \epsilon \tfrac{\zeta_s}{ \zeta}) s(\ve y) \leq (1 + \epsilon) s(\ve y)$$ and $$r(\ve x)\leq (1 + \epsilon \tfrac{\zeta_r}{ \zeta }) r(\ve y) \leq (1 + \epsilon) r(\ve y) .$$
For $\zeta = \zeta_{s + r}$, by combining these, we see that $s(\ve x) + r(\ve x) \leq (1+\epsilon) (s(\ve y) + r(\ve y))$. Hence, $s+r$ is sliceable. Instead, for $\zeta = \zeta_{s r}$, by combining these, we see that $sr$ is also sliceable since $$\begin{aligned}
s(\ve x) r(\ve x)
&\leq (1 + \epsilon \tfrac{\zeta_s}{ \zeta_{sr}}) s(\ve y) (1 + \epsilon \tfrac{\zeta_r}{ \zeta_{sr} }) r(\ve y)
\\
&\leq (1 + \tfrac{\epsilon}{4})^2 s(\ve y) r (\ve y)
\\
&\leq (1+ \epsilon) s(\ve y) r(\ve y).\end{aligned}$$
The function $\lambda s$ is trivially sliceable the same parameters as $s$. Also, if $s > 0$, then it follows $\tfrac{1}{s}$ is sliceable with the same parameters as $s$. Indeed, for $\ve x, \ve y \in B_{\ve k}(\ve c^0, \ve c^1, \ldots, \ve c^{l_s}, \epsilon/\zeta_s ) \cap {\mathbb{Z}}^n$ it holds that $s(\ve y) \leq (1+\epsilon) s(\ve x)$. By dividing, we have $\tfrac{1}{s(\ve x)} \leq (1+ \epsilon) \tfrac{1}{s(\ve y)}$.
Any function that is the sum of even powers of linear forms is sliceable over ${\mathbb{R}}^n$. Rational powers of sliceable functions are also sliceable. For instance, $\sqrt{ s(\ve x)}$ is sliceable for any sliceable function $s$. Moreover, any polynomial that is positive on a polyhedron $P$ is sliceable on $P$. Handelman [@handelman1988] showed that any polynomial $f$ that is positive over a polyhedron has a representation as $f(\ve x) = \sum_{|\alpha| \leq L, \alpha \in {\mathbb{Z}}_+^n} \beta_\alpha \prod_{i=1}^n (\ve a^i \cdot \ve x - b_i)^{\alpha_i}$ for some $L \in {\mathbb{Z}}_+$ and $\beta_\alpha \geq 0$. It follows that $f$ is sliceable over $P$. Unfortunately, the parameter $\zeta$ grows exponentially with $L$. The best bounds on $L$ are given by Powers and Reznick [@Powers2001], which are in terms of the coefficients of $f$ and the smallest value that $f$ attains on $P$. The smallest value is bounded by a polynomial in the input provided the degree and the dimension are fixed [@Jeronimo2010], but this still implies that the best known bound for $L$ is not of polynomial size in the input. Therefore there is no known bound on the size of $\zeta$ based on the input of the polynomial $f$ that would admit an FPTAS for Problem over $P$ with this decomposition.
The polynomial $M(x, y) = x^4 y^2 + x^2 y^4 - 3 x^2 y^2 + 1$ is known as the Motzkin polynomial. While it is not SOS, it is known to be positive because $(x^2 + y^2) \cdot M(x, y)$ has an explicit SOS decomposition (see [@parrilo_polynomial_2012 p. 84]). In particular, for $x,y \neq 0$, we have $$\begin{aligned}
M(x,y)
&= \frac{1}{(x^2 + y^2)} \Big( y^2(1-x)^2(1+x)^2
\\
&+ x^2(1-y)^2(1+y)^2 + x^2 y^2 (x^2 + y^2 - 2)^2 \Big).\end{aligned}$$ Thus, for $x,y \neq 0$, $M(x,y)$ is a composition of sliceable functions and a convex function. That is, the parameters $\ve c^0, \ve c^1, \ldots, \ve c^l$ are defined by $c^j_1 x + c^j_2 y + c^0_j = F_j(x, y)$ for $F_1(x, y), \ldots, F_6(x, y)$ equal to $x,y, 1-x, 1+x, 1-y, 1+y$ respectively, and the convex function is $(x^2 + y^2 - 2)^2$. Sign-compatibility is also given because all involved functions are non-negative. As we will see in Theorem \[thm:sliceable-product-sum\], this implies that Problem with $f(x, y) = M(x,y)$ admits an FPTAS over polytopes.
The following lemma allows us to find approximate solutions by taking the best approximate solution over a decomposition of the feasible region.
\[lem:main-tool\]
Let $C \subseteq {\mathbb{R}}^n$ such that $C \cap {\mathbb{Z}}^n \ne \emptyset$, and for $j=1, \dots, m$ let $s_j \colon C \to {\mathbb{R}}_{\geq 0}$ for $j=1, \dots, m$ and let $g_j \colon C \to {\mathbb{R}}$. Let $0 < \epsilon < 1$ and $\epsilon' = \epsilon / 4$. Let $K \subseteq {\mathbb{Z}}^n$ such that $C \cap {\mathbb{Z}}^n \subseteq \bigcup_{\ve k \in K} B_{\ve k}$, where $B_{\ve k} \subseteq {\mathbb{R}}^n$ are polytopes. Also, let $L_j^{\ve k} \geq 0$ such that $L_j^{\ve k} \leq s_j(\ve x) \leq (1+\epsilon') L_j^{\ve k}$ for all $\ve x \in B_{\ve k} \cap {\mathbb{Z}}^n$ and $j=1, \dots, m$.
Then, for $$\ve x_\epsilon := \arg \min \left\{ \sum_{j=1}^m g_j(\ve{ x}^{\ve k}_{\epsilon'}) s_j(\ve{ x}^{\ve k}_{\epsilon'}) : \ve k \in K \right\}$$ where $\ve{ x}^{\ve k}_{\epsilon'}$ is an $\epsilon'$-approximate solution to $$\min \left\{ \sum_{j=1}^m g_j(\ve x) L_j^{\ve k} : \ve x \in C \cap B_{\ve k} \cap {\mathbb{Z}}^n \right\},$$ we have that $\ve x_\epsilon$ is an $\epsilon$-approximate solution to $$\min \left\{ \sum_{j=1}^m g_j(\ve x) s_j(\ve x) : \ve x \in C \cap {\mathbb{Z}}^n \right\} .$$
Let $$\ve x_{{\min}}:= \arg \min \left\{ \sum_{j=1}^m g_j(\ve x) s_j(\ve x) : \ve x \in C \cap {\mathbb{Z}}^n \right\} ,$$ let $\ve{\bar k} \in K$ such that $\ve x_{{\min}}\in B_\ve{ \bar k}$, and define $f(\ve x) := \sum_{j=1}^m g_j(\ve x) s_j(\ve x)$. Suppose first that $f(\ve{ x}^{\ve{\bar k}}_{\epsilon'}) \ge 0$, which implies that $g_j(\ve{ x}^{\ve{\bar k}}_{\epsilon'}) \ge 0$ for all $j = 1, \ldots, m$, because all functions in $\Omega_{\mathcal P}$ are sign compatible and $s_j \ge 0$. Then $$\begin{aligned}
f(\ve{ x}^{\ve{\bar k}}_{\epsilon'})
&= \sum_{j=1}^m g_j(\ve{ x}^{\ve{\bar k}}_{\epsilon'}) s_j(\ve{ x}^{\ve{\bar k}}_{\epsilon'})
\\
&\le (1 + \epsilon') \sum_{j=1}^m g_j(\ve{ x}^{\ve{\bar k}}_{\epsilon'}) L_j^{\ve k}
\\
&\le (1 + \epsilon')^2 \sum_{j=1}^m g_j(\ve x_{{\min}}) L_j^{\ve k}
\\
&\le (1 + \epsilon')^2 \sum_{j=1}^m g_j(\ve x_{{\min}}) s_j(\ve x_{{\min}})
\\
&= (1 + \epsilon')^2 f(\ve x_{{\min}})
\le (1 + \epsilon) f(\ve x_{{\min}}).\end{aligned}$$ The second inequality follows from approximate minimality of $\ve{x}^{\ve{\bar k}}_{\epsilon'}$, which at the same time implies that $g_j(\ve x_{{\min}}) \ge 0$ for all $j = 1, \ldots, m$, as otherwise $f(\ve{ x}^{\ve{\bar k}}_{\epsilon'}) < 0$. In particular, this implies that $f(\ve x_{{\min}}) \ge 0$.
Suppose now that $f(\ve{ x}^{\ve{\bar k}}_{\epsilon'}) < 0$, which implies that $g_j(\ve{ x}^{\ve{\bar k}}_{\epsilon'}) \le 0$ for all $j = 1, \ldots, m$ because all functions in $\Omega_{\mathcal P}$ are sign compatible and $s_j \ge 0$. Then $$\begin{aligned}
f(\ve{ x}^{\ve{\bar k}}_{\epsilon'})
&= \sum_{j=1}^m g_j(\ve{ x}^{\ve{\bar k}}_{\epsilon'}) s_j(\ve{ x}^{\ve{\bar k}}_{\epsilon'})
\le \sum_{j=1}^m g_j(\ve{ x}^{\ve{\bar k}}_{\epsilon'}) L_j^{\ve k}
\\
&\le \frac{1}{1 + \epsilon'} \sum_{j=1}^m g_j(\ve x_{{\min}}) L_j^{\ve k}
\\
&\le \frac{1}{(1 + \epsilon')^2} \sum_{j=1}^m g_j(\ve x_{{\min}}) s_j(\ve x_{{\min}})
\\
&= \frac{1}{(1 + \epsilon')^2} f(\ve x_{{\min}})
\le \frac{1}{1 + \epsilon} f(\ve x_{{\min}}).\end{aligned}$$ To derive those inequalities, we have again used the assumptions on $L^{\ve k}_j$ and approximate minimality of $\ve{ x}^{\ve{\bar k}}_{\epsilon'}$. In particular, the last inequality holds because $f(\ve x_{{\min}}) \le 0$. The result follows from $f(\ve x_\epsilon) \le f(\ve{ x}^{\ve{\bar k}}_{\epsilon'})$.
Let $\mathcal S(\mathcal P)$ be the set of sliceable functions over all $P \in \mathcal P$ with an evaluation oracle. We define $$\mathcal C(\Omega_{\mathcal P}) := \left\{ \sum_{j=1}^m g_j s_j \,|\, g_j \in \Omega_{\mathcal P}, s_j \in \mathcal S(\mathcal P), m \textrm{ finite} \right\} .$$
We now show that there is an FPTAS for all functions in the class $\mathcal C (\Omega_{\mathcal P})$.
\[thm:sliceable-product-sum\] Let $\mathcal P$, $\Omega_{\mathcal P}$ and $\mathcal C(\Omega_{\mathcal P})$ be as above. Suppose that there is an evaluation oracle for $f = \sum_{j=1}^m g_j s_j \in \mathcal C(\Omega_{\mathcal P})$ where $g_j \in \Omega_{\mathcal P}$, $s_j \in \mathcal S(\mathcal P)$, $j=1, \dots, m$. Then for every $P \in \mathcal P$, there exists an FPTAS for Problem , where there size of the input is the sum of the sizes of $g_j$, $s_j$ for $j = 1, \ldots, m$, and $P$.
Let $f = \sum_{j=1}^m g_j s_j \in \mathcal C(\Omega_{\mathcal P})$, $P \in \mathcal P$, and let $\epsilon > 0$ . Feasibility of $P \cap {\mathbb{Z}}^n$ can be determined in polynomial time by [@lenstra_integer_1983]. Thus we assume henceforth that $P \cap {\mathbb{Z}}^n \ne \emptyset$.
Let the slicing parameters of $s_j$ be $\ve c^0_j, \ve c^1_j, \ldots, \ve c^{l_j}_j, \zeta_j$ for $j=1, \dots,m$ and let $\zeta_{\max} = \max_j \{ \zeta_j\}$. Since $P$ is bounded, there exists $R \in {\mathbb{Z}}_{\ge 1}$ of polynomial size in the size of $P$ and the sizes of $s_j$, $j=1, \dots, m$, such that $| \ve c_j^i \cdot \ve x + c^0_j | \leq R$ for all $\ve x \in P$, $i=1, \ldots, l_j$, and $j=1, \ldots, m$ (see, for example, [@schrijver_theory_1986]). Set $\epsilon' = \tfrac{\epsilon}{4}$ and $N := \left\lceil \frac{\log_2(R)}{\log_2 (1+\epsilon'/ \zeta_{\max})} \right\rceil \leq \left\lceil \log_2(R) ( \tfrac{\zeta_{\max}}{\epsilon'} + 1) \right\rceil $. Define $$\begin{aligned}
B_{\ve k}
:= B_{\ve k}\big(&[\ve c^0_1, \ldots, \ve c^0_m],
\\
&\ve c^1_1, \ldots, \ve c^{l_1}_1, \ldots, \ve c^1_m, \ldots, \ve c^{l_m}_m, \epsilon'/ \zeta_{\max}\big).\end{aligned}$$ In particular, ${\mathbb{Z}}^n \subseteq \bigcup_{\ve k \in {\mathbb{Z}}^{l_1 + \ldots + l_m}} B_{\ve k}$. Then for $K = \{-N, \ldots, N\}^{\sum_{j=1}^m l_j}$ we have $P \cap {\mathbb{Z}}^n \subseteq \bigcup_{\ve k \in K} B_{\ve k}$. Notice that many choices of $\ve k \in K$ are redundant since each non-empty set $B_{\ve k}$ is a cell or facet of a cell of the hyperplane arrangement in ${\mathbb{R}}^n$ given by the hyperplanes $\ve c^i_j \cdot \ve x + c^0_j = 0$ and $\ve c^i_j \cdot \ve x + c^0_j = \pm(1+\epsilon'/\zeta_{\max})^k$ for $k \in \{0, \ldots, N\}$, $i=1, \ldots, l_j$, and $j=1, \ldots, m$. By [@sleumer_output_1998], we can enumerate the $O \left( (2N + 3)^n \left( \sum_{j=1}^m l_j\right)^n \right)$ cells of the hyperplane arrangement in polynomial time with $n$ fixed. Hence, we can find a subset $K' \subseteq K$ with cardinality of polynomial size such that $P \subseteq \bigcup_{\ve k \in K'} B_{\ve k}$.
Let $\ve x_\epsilon$ be as in Lemma \[lem:main-tool\]. For $\ve k \in K'$, we determine a point $\ve x^{\ve k} \in P \cap B_{\ve k} \cap {\mathbb{Z}}^n$ using [@lenstra_integer_1983]. If no such point exists, then we remove $\ve k$ from $K'$. Set $L_j^{\ve k} := \frac{1}{1 + \epsilon'} s_j(\ve x^{\ve k})$, computed using the evaluation oracle for $s_j$, for $j=1, \ldots, m$. Note that $L_j^{\ve k}$ satisfies the assumptions of Lemma \[lem:main-tool\] because $s_j$ is sliceable and $\zeta_j \leq \zeta_{\max}$. Then, by Lemma \[lem:main-tool\], $\ve x_\epsilon$ is an $\epsilon$-approximate solution to Problem .
We show that we can compute $\ve x_\epsilon$ in polynomial time. For each $\ve k \in K'$, an $\epsilon'$-approximate solution $\ve{x}^{\ve k}_{\epsilon'}$ to $$\min \left\{ \sum_{j=1}^m g_j(\ve x) L_j^{\ve k} : \ve x \in P \cap B_{\ve k} \cap {\mathbb{Z}}^n \right\}$$ can be computed in polynomial time because $\Omega_{\mathcal P}$ is closed under addition and multiplication by positive scalars, and therefore $\sum_{j=1}^m g_j(\ve x) L_j^{\ve k} \in \Omega_{\mathcal P}$. Finally, $\arg \min \{ f(\ve{x}^{\ve k}_{\epsilon'}) : \ve k \in K' \}$ can be computed in polynomial time using the evaluation oracle of $f$ because $|K'|$ is polynomial in the input size.
\[rem:main-m=1\] The assumptions of Theorem \[thm:sliceable-product-sum\] can be slightly relaxed if we only consider functions of the form $f(\ve x) = g_1(\ve x) s_1(\ve x)$, i.e., $\Omega_{\mathcal P} = \{ \lambda g_1 : \lambda \ge 0 \}$ for some $g_1 \colon {\mathbb{R}}^n \to {\mathbb{R}}$ and $s_1$ sliceable. In fact, in this case no evaluation oracle for $s_1$ is needed, because an $\epsilon'$-approximate solution to $$\min \left\{ g_1(\ve x) : \ve x \in P \cap B_{\ve k} \cap {\mathbb{Z}}^n \right\}$$ is also an $\epsilon'$-approximate solution to $$\min \left\{ g_1(\ve x) L^{\ve k}_1 : \ve x \in P \cap B_{\ve k} \cap {\mathbb{Z}}^n \right\}$$ for any $L^{\ve k}_1 \ge 0$. Thus, $L^{\ve k}_1$ does not need to be computed explicitly.
Theorem \[thm:sliceable-product-sum\] can be generalized from bounded polyhedra $P$ to more general sets $C$, provided that for any bounded polyhedron $\bar P$ the feasibility and minimization problems on $C \cap \bar P \cap {\mathbb{Z}}^n$ and the feasibility problem on $C \cap {\mathbb{Z}}^n$ can be solved in polynomial time in the size of $\bar P$ and the bit size required to represent $C$. The class of convex semi-algebraic sets in fixed dimension is an example of such a class, because for fixed dimension the stated feasibility problem can be solved in polynomial time by [@khackiyan_integer_2000].
The size of a sliceable function $s$ is defined as the sum of the sizes of its slicing parameters, where unary encoding is assumed for the parameter $\zeta$. With this definition, a polynomial bound on the cardinality of $K'$ in terms of the sizes of the sliceable functions is derived in the proof of Theorem \[thm:sliceable-product-sum\]. This bound is not polynomial if binary encoding is assumed for the $\zeta$ parameters. However, in our main application of Theorem \[thm:sliceable-product-sum\], namely the proof of Theorem \[thm:main\], those parameters can be chosen to be constant.
FPTAS for Quadratic Forms {#sec:quadratics}
=========================
For positive definite matrices, it is well known that a Cholesky decomposition $Q = L D L^T$ can be easily computed, where $L$ is a lower-triangular matrix and $D$ is a diagonal matrix. Such a decomposition does not always exist for indefinite symmetric matrices [@dax_pivoting_1977]. However, there is a similar decomposition that serves our purpose, as described in the following remark.
\[rem:indefinite-decomposition\] For any symmetric matrix $Q \in {\mathbb{Z}}^{n \times n}$, we can find an invertible matrix $S \in {\mathbb{Q}}^{n \times n}$ and a diagonal matrix $D \in {\mathbb{Q}}^{n \times n}$ such that $Q = S D S^T$ in polynomial time in the size of $Q$ for fixed $n$.
We thus have $$\ve x^T Q \ve x = \sum_{i=1}^n d_i (S_i \cdot \ve x)^2$$ where $D = \mathrm{diag}(\ve d)$ is the diagonal matrix with $D_{ii} = d_i$, and $S_i$ denotes the $i$-th row of $S$.
\[rem:FPTAS-scaling\] Since an FPTAS for $c^3 f(\ve x)$, $c \in {\mathbb{Q}}_{> 0}$, is also an FPTAS for $f(\ve x)$ as long as $c$ is of polynomial size in the input size, $S$ and $D$ can be scaled to be in ${\mathbb{Z}}^{n \times n}$. In fact, let $c$ be the least common multiple of the denominators of all entries of $S$ and $D$, which is of polynomial size in the size of $Q$, because $S$ and $D$ can be computed in polynomial time in the size of $Q$. Then $c^3 f(\ve x) = \ve x^T (cS) (cD) (cS)^T \ve x$ and $cS, cD \in {\mathbb{Z}}^{n \times n}$.
Since $S$ is non-singular, Sylvester’s law of inertia (see, for example [@golub_matrix_1996] or [@gill_numerical_1991]) states that the inertia of $Q$ is the same as the inertia of $D$. Therefore $D$ has exactly as many positive, negative and zero diagonal entries as $Q$ has positive, negative and zero eigenvalues.
The following lemmas will be useful for determining the accuracy necessary for the numerical approximations.
\[thm:polynomial\_root\_bounds\] Let $m < n$ be nonnegative integers, $a_m, \ldots, a_n \in {\mathbb{R}}$, $a_m \ne 0$, $a_n \ne 0$, and let $\bar x \in {\mathbb{R}}$ be a nonzero root of the polynomial $f(x) := \sum_{i=m}^n a_i x^i$. Then $|\bar x| > \min \{ |a_m|/(|a_m| + |a_i|) \, : \, i = m+1, \ldots, n \}$.
The proof of the next lemma uses a standard technique for converting an equation with rational powers to one with integer powers.
\[lem:difference\] Let $p,q,p',q' \in {\mathbb{Z}}$ with $|p|, |q|, |p'|, |q'| \leq R$, $p,p' \geq 0$, and let $\delta = (\sqrt{p} + q) - (\sqrt{p'} + q')$. If $\delta \neq 0$, then $|\delta| > \tfrac{1}{5\cdot 48} R^{-4}$.
We show that any solution $\delta$ is a solution to a polynomial equation in terms of $\delta$. Rewriting the equation for $\delta$ by solving for $\sqrt{p}$ and squaring both sides yields $p = \delta^2 + p' + q^2 - 2 q q' + q'^2 + \delta (-2 q + 2 q') + \sqrt{p'} (2 \delta - 2 q + 2 q')$. Solving for $\sqrt{p'}(2 \delta - 2 q + 2 q')$ and squaring both sides yields a degree 4 polynomial in $\delta$ with integer coefficients that are polynomial in $R$. In particular, it can be shown the the absolute values of the coefficients are upper bounded by $48 R^4$. Since $\delta$ is a solution to this polynomial equation, by Lemma \[thm:polynomial\_root\_bounds\], it follows that if $\delta \neq 0$, then $|\delta| > \tfrac{1}{5\cdot 48} R^{-4}$.
With those technical tools, we can now prove Theorem \[thm:main\].
Given $Q \in {\mathbb{Z}}^{n \times n}$, it is possible to compute $S$ and $D = \mathrm{diag}(\ve d)$ such that $Q = S D S^T$ in polynomial time by Remark \[rem:indefinite-decomposition\]. Moreover, $S$ and $D$ can be assumed to be integral by Remark \[rem:FPTAS-scaling\]. Suppose that $Q$ has at most one negative eigenvalue, which implies that $-Q$ has at most one positive eigenvalue. By reordering variables, we may assume that $d_i \ge 0$ for $i = 1, \ldots, n-1$. Note that if $d_n \ge 0$, then $f_1(\ve x) = \ve x^T Q \ve x$ is convex and $f_2(\ve x) = \ve x^T (-Q) \ve x$ is concave, so Problem with $f = f_1$ and $f = f_2$ can be solved in polynomial time by Theorem \[thm:convex-concave\] (\[itm:quasi-convex-semi-algebraic\]) and (\[itm:quasi-concave\]) respectively. Thus assume henceforth that $d_n < 0$.
Let $$g(\ve x) := \sqrt{\sum_{i=1}^{n-1} d_i (S_i \cdot \ve x)^2} - \sqrt{- d_n} |S_n \cdot \ve x|$$ and $$s(\ve x) := \sqrt{\sum_{i=1}^{n-1} d_i (S_i \cdot \ve x)^2} + \sqrt{- d_n} |S_n \cdot \ve x| .$$ Notice that $f_1(\ve x) = g(\ve x) s(\ve x)$ and $f_2(\ve x) = (-g(\ve x)) s(\ve x)$. Let $\mathcal P := \{ \bar P \text{ polytope } : \bar P \subseteq P\}$ and $\Omega_{\mathcal P} := \{ \lambda g : \lambda\in {\mathbb{R}}\}$. We will show that $s$ is sliceable with size bounded by a polynomial in the size of $Q$, and that $\Omega_{\mathcal P}$ is a class of functions that we can minimize over the integers in polyhedra $\bar P \in \mathcal P$. Equivalently, we show that we can minimize $g$ and $-g$ over $\bar P \cap {\mathbb{Z}}^n$ for $\bar P \in \mathcal P$. The result then follows from Theorem \[thm:main\] and Remark \[rem:main-m=1\].
First, we define the slicing parameters $\ve c^j = S_j$ for $j = 1, \ldots, n$, $\ve c^0 = 0$, and $\zeta = 1$. Then for any $\ve k \in {\mathbb{Z}}^n$, $\epsilon > 0$, and any $\ve x, \ve y \in B_{\ve k}(\ve c^0, \ve c^1, \ldots, \ve c^n, \epsilon)$, we have $$\begin{aligned}
s(\ve x)
&= \sqrt{\sum_{i=1}^{n-1} d_i (S_i \cdot \ve x)^2} + \sqrt{- d_n} |S_n \cdot \ve x|
\\
&\leq \sqrt{\sum_{i=1}^{n-1} d_i ((1+\epsilon) S_i \cdot \ve y)^2} + \sqrt{- d_n} |(1+\epsilon) S_n \cdot \ve y|
\\
&\leq (1+\epsilon) s(\ve y) .\end{aligned}$$ Hence, $s$ is sliceable with size polynomial in the size of $Q$.
Second, we show that we can solve $\min\{g(\ve x) : \ve x \in \bar P \cap {\mathbb{Z}}^n\}$ exactly for any $\bar P \in \mathcal P$. The function $g$ is convex on the halfspaces $S_n \cdot \ve x \ge 0$ and $S_n \cdot \ve x \le 0$, because it is the sum of a norm with a linear function. Even though $g$ is convex, an exact first order oracle is not available because the function values can be irrational. We show the minimization problem can be solved by reformulating sublevel sets as convex semi-algebraic sets.
We separately consider the problem on the halfspaces $S_n \cdot \ve x \ge 0$ and $S_n \cdot \ve x \le 0$. The analysis for both cases is similar. Without loss of generality, assume that $S_n \cdot \ve x \geq 0$.
For $\beta \in {\mathbb{R}}$, the inequality $g(\ve x) \le \tfrac{\beta}{\sqrt{-d_n}}$ is equivalent to $$\sqrt{|d_n| \sum_{i=1}^{n-1} d_i (S_i \cdot \ve x)^2} \le \beta + |d_n| S_n \cdot \ve x.$$ This is a second order cone constraint, which is well known to define a convex set [@Boyd2004]. It can be reformulated by recognizing that the left hand side is nonnegative and adding the condition that the right hand side is nonnegative too. With this, we can square both sides to get the equivalent relation $$\begin{aligned}
\label{eq:convex-set}
\begin{split}
|d_n| \sum_{i=1}^{n-1} d_i (S_i \cdot \ve x)^2 - (\beta + |d_n| S_n \cdot \ve x)^2 &\le 0,
\\
\beta + |d_n| S_n \cdot \ve x &\ge 0 .
\end{split}\end{aligned}$$ In this representation, the set $\{\ve x \in {\mathbb{R}}^n : g(\ve x) \leq \tfrac{\beta}{\sqrt{-d_n}}, \ S_n \cdot \ve x \geq 0\}$ is a convex semi-algebraic set. It is thus possible to test integer feasibility of this set for $\beta \in {\mathbb{Q}}$ in polynomial time using Theorem \[thm:convex-concave\] (\[itm:quasi-convex-semi-algebraic\]). Since $\bar P$ is bounded, and $S$ and $D$ are of polynomial size, it follows from [@schrijver_theory_1986] that we can compute $R \in {\mathbb{Z}}$, $R \ge \|\ve d\|_\infty$ of polynomial size such that $\| S \ve x \|_\infty \leq R$ for all $\ve x \in \bar P$. Note that $|\sqrt{-d_n} \, g(\ve x)| \leq 2nR^3$ for all $\ve x \in \bar P$. We now perform binary search on $\beta$ in the interval $[-2nR^3, 2nR^3]$ for integer feasibility of with precision $\mu := \tfrac{1}{5\cdot 48} n^{-4}R^{-16}$. Since these numbers are of polynomial size, the binary search can be done in polynomial time. We show now that the choice of precision yields an optimal solution by showing that it distinguishes function values of $\sqrt{-d_n}\, g(\ve x)$ for $\ve x \in \bar P \cap{\mathbb{Z}}^n$.
Indeed, consider $\ve x,\ve y \in \bar P \cap {\mathbb{Z}}^n$ and define $\delta := \sqrt{-d_n}( g(\ve x) - g(\ve y))$. Setting $p = |d_n| \sum_{i=1}^{n-1} d_i (S_i \cdot \ve x)^2$, $q = - |d_n| S_n \cdot \ve x$, $p' = |d_n| \sum_{i=1}^{n-1} d_i (S_i \cdot \ve y)^2$, and $q' = -|d_n| S_n \cdot \ve y$, we see that $\delta = (\sqrt{p} + q) - (\sqrt{p'} + q')$. Since $p, p' \geq 0$ and $|p|,|p'|,|q|,|q'| \leq n R^4$, by Lemma \[lem:difference\], if $\delta \neq 0$, then $|\delta| > \mu$. Thus our choice of the precision for the binary search distinguishes function values of $\sqrt{-d_n} g(\ve x)$ for $\ve x \in \bar P$ and hence we arrive at an exact algorithm that runs in polynomial time. Note also, that this implies that the precision $\mu/R$ distinguishes function values of $g(\ve x)$ for $\ve x \in \bar P$.
Lastly, we show that we can solve $\min\{-g(\ve x) : \ve x \in \bar P \cap {\mathbb{Z}}^n\}$ for any $\bar P \in \mathcal P$. Since $g$ is convex, $-g$ is concave and hence is also quasi-concave. By Theorem \[thm:convex-concave\] (\[itm:quasi-concave\]), the minimization problem can be solved exactly provided that we have a comparison oracle for $-g$. Our comparison oracle is realized by approximating $-g$ to a precision of $\mu/R$, which, as stated above, distinguishes function values of $g$. This precision can be obtained by approximating the square roots in $g$ using, for instance binary search. See also [@mansour_complexity_1989] for approximation of square roots.
Thus, we have shown that the set $\Omega_{\mathcal P}$ is indeed a class of functions that we can minimize in polynomial time over polytopes $\bar P \in \mathcal P$.
In the proof of Theorem \[thm:main\], there is an alternative method that avoids computing solutions to $\min \{ -g(\ve x) \ :\ x \in \bar P \cap {\mathbb{Z}}^n \}$, thus avoiding potential numerical issues when approximating square roots. In fact, let $\ve{\bar x} \in \arg \min \{ f(\ve x) : \ve x \textrm{ is a vertex of } (\bar P \cap {\mathbb{Z}}^n)_I \}$ and $\ve{\tilde x} \in \arg \min \{ -g(\ve x) \ :\ x \in \bar P \cap {\mathbb{Z}}^n \}$. Since $-g$ is concave, its optimal value over $\bar P \cap {\mathbb{Z}}^n$ is attained at a vertex of $(\bar P \cap {\mathbb{Z}}^n)_I$. Therefore, because $\ve{\bar x}$ minimizes $f$ over the vertices of $(\bar P \cap {\mathbb{Z}}^n)_I$, we have $f(\ve{\bar x}) \le f(\ve{\tilde x})$.
Finally, we mention that for certain cases, we can obtain exact solutions to Problem for $f(\ve x) = \ve x^T Q \ve x$ where $Q$ has at most one negative or at most one positive eigenvalue. For this, we use the following lemma.
\[lem:f\_quasiconvex\] Let $S \in {\mathbb{Z}}^{n \times n}$ be non-singular and $D = \mathrm{diag}(\ve d)$ with $\ve d \in {\mathbb{Z}}^{n-1}_{\ge 0} \times {\mathbb{Z}}_{< 0}$. Then $f(\ve x) := \ve x^T S D S^T \ve x$ is quasi-convex on the sets $C_1 := \{ \ve x \in {\mathbb{R}}^n \,:\, f(\ve x) \le 0,\ S_n \cdot \ve x \ge 0 \}$ and $C_2 := \{ \ve x \in {\mathbb{R}}^n \,:\, f(\ve x) \le 0,\ S_n \cdot \ve x \le 0 \}$.
We show this for the set $C_1$; the proof for the set $C_2$ is similar. We only need to consider the case where $S$ is the identity matrix, because convexity is preserved under linear transformations.
$C_1 = \{ \ve x \in {\mathbb{R}}^n \,:\, \sqrt{\sum_{i=1}^{n-1} d_i (x_i)^2} \le \sqrt{- d_n} x_n,\ x_n \ge 0 \}$ is a convex because it is a second-order cone [@Boyd2004]. In particular, its interior, which is given by $\operatorname{int}(C_1) = \{ \ve x \in {\mathbb{R}}^n \,:\, f(\ve x) < 0,\ x_n \ge 0 \}$, is convex.
Note that if $d_i = 0$, then $f$ is quasi-convex on some set $C$ if and only if the restriction of $f$ to $x_i = 0$ is quasi-convex on $C \cap \{\ve x \in {\mathbb{R}}^n: x_i = 0 \}$. Therefore, by reduction of variables, assume without loss of generality that $d_i \ne 0$ for $i = 1, \ldots, n$.
Since $d_i \ne 0$ for $i = 1, \ldots, n$, $C_1$ is a pointed cone and therefore does not contain any lines. Let $\ve x, \ve y \in \operatorname{int}(C_1)$ and $\lambda \in (0, 1)$. We will show that $f(\lambda \ve x + (1-\lambda) \ve y) \le \max \{ f(\ve x), f(\ve y)\}$, thus proving quasi-convexity of $f$ on $\operatorname{int}(C_1)$. Let $$\begin{aligned}
g(\lambda)
&:= f(\lambda \ve x + (1-\lambda) \ve y)
\\
&\phantom{:}= \lambda^2 f(\ve x) + (1- \lambda)^2 f(\ve y) \\
&\phantom{:=} + 2 \lambda (1-\lambda) \sum_{i = 1}^n d_i (S_i \cdot \ve x) (S_i \cdot \ve y) ,\end{aligned}$$ and note that $g$ is a quadratic function in $\lambda$ and satisfies $g(0) = f(\ve x)$ and $g(1) = f(\ve y)$. If $g$ is convex, then the result follows by convexity. Otherwise, $g$ is strictly concave. Assume without loss of generality that $f(\ve x) \ge f(\ve y)$. We will show that $g$ is increasing on the interval $[0, 1]$, thus proving the result.
Assume for the sake of contradiction that $g$ is not increasing on the interval $[0, 1]$ and therefore has a local maximum on this interval, say for $\hat \lambda$. Since $g$ is quadratic and strictly concave, $\hat \lambda$ is also its global maximum. By convexity of $\operatorname{int}(C_1)$, $\hat \lambda \ve x + (1 - \hat \lambda) \ve y \in \operatorname{int}(C_1)$ and thus $g(\hat \lambda) < 0$. However, $\operatorname{int}(C_1)$ does not contain any line. Therefore, there is a $\bar \lambda \in {\mathbb{R}}$ such that $\bar \lambda \ve x + (1-\bar \lambda) \ve y \not \in \operatorname{int}(C_1)$, implying $f(\bar \lambda \ve x + (1-\bar \lambda) \ve y) \ge 0$ because $f$ takes the value 0 on the boundary of $C_1$. By maximality of $g(\hat \lambda)$, this implies that $g(\hat \lambda) \ge 0$, a contradiction. Therefore, $f$ is quasi-convex on $\operatorname{int}(C_1)$.
Finally, if a continuous function is continuous on the closure of a convex set $C$ and quasi-convex on its interior, it is quasi-convex on the closure of $C$ (see, for example, Theorem 2.2.12 in [@cambini_generalized_2009]). Therefore $f$ is quasi-convex in $C_1$.
\[prop:main\] Let $Q \in {\mathbb{Z}}^{n \times n}$ be a symmetric matrix and let $n$ be fixed. Then there is a polynomial time algorithm to solve Problem with $f(\ve x) = \ve x^T Q \ve x$ in the following cases:
(i) $Q$ has exactly one negative eigenvalue and $f(\ve x_{{{\min}}}) \leq 0$; \[itm:prop\_min\]
(ii) $Q$ has exactly one positive eigenvalue and $f(\ve x_{{{\min}}}) \geq 0$. \[itm:prop\_max\]
Furthermore, in polynomial time, we can detect if one of these cases occurs.
We suppose that $P \cap {\mathbb{Z}}^n \neq \emptyset$. As mentioned before, we can determine the inertia of $Q$ in polynomial time.
Suppose $Q$ has exactly one negative eigenvalue. Consider the sets $C_1$ and $C_2$ as in Lemma \[lem:f\_quasiconvex\]. Since $f$ is quasi-convex on each of those sets, we can minimize $f$ over $C_1 \cap P \cap {\mathbb{Z}}^n$ and $C_2 \cap P \cap {\mathbb{Z}}^n$ in polynomial time by Theorem \[thm:convex-concave\] (\[itm:quasi-convex-semi-algebraic\]). As $C_1 \cup C_2 = \{ \ve x \in {\mathbb{R}}^n : f(\ve x) \leq 0\}$, we have shown how to detect if $f(\ve x_{{{\min}}}) \leq 0$ and if so, determine an optimal solution.
Suppose instead that $Q$ has exactly one positive eigenvalue. We consider the decomposition in Lemma \[lem:f\_quasiconvex\] for $-Q = S (-D) S^T$, which has exactly one negative eigenvalue. Since we decomposed $-Q$ instead of $Q$, the sets $C_1$, $C_2$ are $C_1 := \{ \ve x \in {\mathbb{R}}^n \,:\, f(\ve x) \geq 0,\ S_n \cdot \ve x \ge 0 \}$ and $C_2 := \{ \ve x \in {\mathbb{R}}^n \,:\, f(\ve x) \ge 0,\ S_n \cdot \ve x \le 0 \}$. Now we compute the vertices of the integer hulls $P^1_I := (P \cap \{ \ve x \in {\mathbb{R}}^n : S_n \cdot \ve x \geq 0\})_I$ and $P^2_I := (P \cap \{ \ve x \in {\mathbb{R}}^n : S_n \cdot \ve x \leq 0\})_I$, which can be done in polynomial time by Theorem \[thm:convex-concave\] (\[itm:vertices-integer-hull\]). A standard convexity argument shows that $f(\ve x_{\min}) < 0$ if and only if there exists a vertex $\ve x_v$ of $P^1_I$ or $P^2_I$ such that $f(\ve x_v) < 0$. Otherwise, all vertices of $P^1_I$ and $P^2_I$ are contained in $C_1 \cup C_2$. This certifies that $P\cap {\mathbb{Z}}^n \subseteq C_1 \cup C_2$, and hence $f(\ve x_{{{\min}}}) \geq 0$. If this happens, $f$ is quasi-concave on both $C_1$ and $C_2$. By Theorem \[thm:convex-concave\] (\[itm:quasi-concave\]), we can compute an optimal solution in polynomial time.
[^1]: [email protected]
[^2]: [email protected]
[^3]: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In this paper, we mainly study one class of mixed-integer nonlinear programming problems (MINLPs) with vector conic constraint in Banach spaces. Duality theory of convex vector optimization problems applied to this class of MINLPs is deeply investigated. With the help of duality, we use the generalized Benders decomposition method to establish an algorithm for solving this MINLP. Several convergence theorems on the algorithm are also presented. The convergence theorems generalize and extend the existing results on MINLPs in finite dimension spaces.'
author:
- Zhou Wei
- 'M. Montaz Ali '
date:
-
-
title: Generalized Benders Decomposition for one Class of MINLPs with Vector Conic Constraint
---
[**Key words.**]{} [*Generalized Benders decomposition; MINLP; Duality theory; Vector optimization*]{}
[**AMS subject classifications.**]{} [*90C11, 90C25, 90C30*]{}
Introduction
============
In many optimization problems, decision variables appearing in objective and constraint functions are continuous and discrete. These optimization problems can be modelled as mixed-integer nonlinear programming problems (MINLPs). In general, MINLP is defined mathematically as follows: $$\label{1.1a}
{\rm \mathcal{P}} \left \{
\begin{array}l
\mathop{\rm minimize}\limits_{x,\, y} \ \ \ f(x, y)\\
{\rm subject\ to}\ \ g(x, y)\leq 0, \\
\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ x\in X, y\in Y\ {\rm discrete\ variable},
\end{array}
\right.$$ where $f: \mathbb{R}^n\times \mathbb{R}^p\rightarrow \mathbb{R}$ and $g: \mathbb{R}^n\times \mathbb{R}^p\rightarrow \mathbb{R}^m$ are nonlinear functions, $X\subset\mathbb{R}^n$, and $Y\subset\mathbb{R}^p$ is a polyhedral set of discrete points.
MINLP problem $\mathcal{P}$ is a natural approach to solve problems by simultaneously optimizing the system structure (discrete) and parameters (continuous). Over the past decades, MINLPs have been used in various applications such as the process industry, chemical engineering design, production planning and control, optimal design of gas or water transmission networks, finance and scheduling problems etc.(cf. [@CD; @FC; @GS1; @GS2; @TS; @TS1] and references therein). Note that two subclasses of mixed-integer linear programming (MILP) problem and nonlinear programming (NLP) problem are embedded in MINLP simultaneously, and thus MINLP problem $\mathcal{P}$ falls into the class of NP-hard problems and becomes one of the most difficult optimization problems. It is known that methods for solving MINLP problem $\mathcal{P}$ mainly fall in two broad classes. One class is heuristic methods which do not provide a guarantee that on termination the incumbent is a minimizer, while the other class is deterministic methods which terminate with a guaranteed solution or an indication that the problem has no integer solution. The deterministic methods for solving MINLP problem $\mathcal{P}$ with convex functions are mainly on NLP/LP based on branch-and-bound method (cf. [@L; @NV]), extended cutting-plane method (cf. [@EM1; @WP; @WP1]), outer approximation method (cf. [@FL; @WA1; @WA2]), variable and Lagrangean decompositions (cf. [@FR; @MM]), generalized Benders decomposition (cf. [@B; @B1]) etc.
Vector optimization relates to functional analysis and mathematical programming, and has been found to play many important roles in economics theory, engineering design, management science, multi-criteria decision making and so on. In recent years, the study on vector optimization has received increasing attentions in the literature (see [@B2; @F; @GJ; @J; @Lu; @M] and references therein). To the best of our knowledge, there is not much literature to study MINLPs in the framework of vector optimization, and from the theoretical viewpoint as well as for applications, it is of significance to continue studying MINLPs in general infinite dimension spaces. Motivated by this, in this paper, we mainly study one class of MINLPs with vector conic constraints in the context of Banach spaces, and aim to establish an appropriate algorithm for solving it. Let $E, Z$ be two Banach spaces, $D$ be a normed linear space and $K$ be a closed convex cone in $Z$ which specifies a partial order $\preceq_K$ on $Z$ as follows: $$\label{1.2a}
z_1\preceq_K z_2\Longleftrightarrow z_2-z_1\in K\ \ {\rm for\ all\ } z_1, z_2\in Z.$$ In this paper, we consider the following MINLP problem (VOP) with vector conic constraint: $$\label{1.3a}
{\rm (VOP)} \left \{
\begin{array}l
\mathop{\rm minimize}\limits_{x,\, y} \ \ \ f(x, y)\\
{\rm subject\ to}\ \ g(x, y)\preceq_K 0, \\
\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ x\in X, y\in Y\ {\rm discrete\ variable},
\end{array}
\right.$$ where $f: E\times D\rightarrow \mathbb{R}$ and $g: E\times D\rightarrow Z$, $X\subset E$ and $Y\subset D$ a set with discrete variables. When we take $E:=\mathbb{R}^n$, $Z:=\mathbb{R}^m$, $D:=\mathbb{R}^p$ and $K:=\mathbb{R}^m_+$, problem (VOP) reduces to the MINLP problem $\mathcal{P}$. Hence it is more general to study this class of MINLPs. With respect to solving problem (VOP), we are inspired by some ideas from generalized Benders decomposition and use this decomposition method to construct an appropriate algorithm for finding the optimal value of problem (VOP).
Benders decomposition was first introduced by Benders [@B] and has been applied to a variety of optimization problems such as mixed-integer linear programming, nonlinear programming and MINLPs. It is known that Benders decomposition is an approach for exploiting the structure of mathematical programming problems with complicating variables. Such variables, if temporarily fixed, may render the remaining optimization problem considerably more tractable. For the special class of problems studied by Benders [@B], fixing the complicating variables reduces the given problem to an ordinary linear programming, parameterized by the value of the complicating vectors. Along this line, Geoffrion [@Ge1] generalized the Benders decomposition to a broader class of problems where the parameterized subproblem need no longer be a linear programming. Rouhani et al. [@RL] and Floudas and Ciric [@FC] used the generalized Benders decomposition approach to solve MINLPs which are modelled from practical problems of reactive source planning in power systems and heat exchanger network synthesis respectively. It is noted that Hooker and Ottosson studied logice-based Benders decomposition, one extension of Benders decomposition, and applied this method to planning and scheduling. Readers are invited to consult references [@H1; @H2; @HO] for more details. For these reasons, the generalized Benders decomposition has been extensively studied by many authors over past decades (cf. [@B; @B1; @Ge; @Ge1; @Gr; @SG] and references therein).
Note that Geoffrion [@Ge1] employed the generalized Benders decomposition and nonlinear convex duality (cf. [@Ge]) to reformulate MINLP problem $\mathcal{P}$ and derive one equivalent master problem. The algorithm, presented through the generalized Benders decomposition procedure, alternates between solutions of relaxed master problems and nonlinear convex subproblems. Sahinidis and Grossmann [@SG] further discussed convergence properties of this generalized Benders decomposition procedure. Inspired by [@Ge1; @SG], in this paper, we mainly study the generalized Benders decomposition in vector optimization and use this approach to construct one corresponding algorithm for solving MINLP problem (VOP) of . To achieve this aim, along the line given by Geoffrion [@Ge1], it is necessary to separate problem (VOP) into many subproblems, establish an equivalent master problem of problem (VOP) and solve the relaxation of master problems. For the equivalence between problem (VOP) and its master problem, we are inspired by Geoffrion [@Ge] to study the duality of convex vector optimization problems and proved several duality results (see Section 3).
The paper is organized as follows. In Section 2, we give some definitions and preliminaries used in this paper. Section 3 is devoted to duality theory results on convex vector optimization problems. Several duality results generalize the corresponding ones obtained in [@Ge]. In Section 4, we use the generalized Benders decomposition to establish an algorithm for solving problem (VOP) of with the help of duality results given in Section 3. The convergence theorems on the algorithm are obtained therein. The conclusion of this paper is presented in Section 5.
Preliminaries
=============
Let $E$ be a Banach space (i.e. complete normed linear space) and $E^*$ denote the dual space of $E$ with dual pairing $\langle \cdot, \cdot\rangle$ between $E^*$ and $E$. Given a set $C\subset E$, let $\overline C$ and ${\rm int}(C)$ denote the [*norm closure*]{} and the [*interior*]{} of $C$, respectively. For $x\in E$ and $\delta>0$, denote $B(x, \delta)$ the [*open ball*]{} with center $x$ and radius $\delta$.
Let $S$ be a closed convex set of $E$ and $x\in S$. We denote $T(S, x)$ the [*contingent cone*]{} of $S$ at $x$; that is $v\in T(S, x)$ if and only if there exist a sequence $\{v_k\}$ in $E$ converging to $v$ and a sequence $t_k$ in $(0, +\infty)$ decreasing to $0$ such that $x+t_kv_k\in S$ for all $k\in\mathbb{N}$, where $\mathbb{N}$ denotes the set of all natural numbers. The [*normal cone*]{} of $S$ at $x$, denoted by $N(S, x)$, is defined as: $$\label{2.1}
N(S, x):=\{x^*\in E^*: \langle x^*, v\rangle\leq 0\ \ {\rm for\ all} \ v\in T(S, x)\}.$$ It is known that $N(S, x)$ and $T(S, x)$ are the dual cones of each other and one can verify that $$N(S, x)=\{x^*\in E^*: \langle x^*, y-x\rangle\leq 0\ \ {\rm for\ all} \ y\in S\}.$$
Let $\psi: E\rightarrow \mathbb{R}\cup\{+\infty\}$ be a convex function. We denote $${\rm dom}(\psi):=\{x\in E: \psi(x)\in\mathbb{R}\}\,\,\,{\rm and}\,\,\,{\rm epi}(\psi)=\{(x, \alpha)\in E\times \mathbb{R}: \psi(x)\leq \alpha\}$$ the [*domain*]{} and the [*epigraph*]{} of $\psi$, respectively. Recall that $\psi$ is said to be [*lower semicontinuous*]{} at $x\in E$, if $\liminf_{y\rightarrow x}\psi(y)\geq \psi(x)$. Let $x\in {\rm dom}(\psi)$. Recall that the [*subdifferential*]{} of $\psi$ at $x$ is defined by $$\partial\psi(x):=\{x^*\in E^*: (x^*, -1)\in N({\rm epi}(\psi), (x, \psi(x)))\}.$$ It is known that $x^*\in\partial\psi(x)$ if and only if $$\langle x^*, y-x\rangle\leq\psi(y)-\psi(x)\ \ {\rm for\ all} \ y\in E.$$
Let $F: E\rightrightarrows E^*$ be a set-valued mapping. We denote $
{\rm dom}(F):=\{x\in E: F(x)\not=\emptyset\}$ the [*domain*]{} of $F$. Let $x\in {\rm dom}(F)$. Recall that $F$ is said to be [*norm-to-weak$^*$ upper semicontinuous*]{} at $x$, if for every weak$^*$ open set $V$ containing $F(x)$ and every sequence $\{x_n\}\subset {\rm dom}(F)$ with $\|x_n-x\|\rightarrow 0$, one has $F(x_n)\subset V$ for all sufficiently large $n$. Equivalently, it is easy to verify that $F$ is norm-to-weak$^*$ upper semicontinuous at $x$ if and only if for any generalized sequences $\{x_k\}$ and $\{x_k^*\}$ satisfying $x_k\stackrel{\|\cdot\|}\longrightarrow x$, $x_k^*\stackrel{w^*}\longrightarrow x^*$ and $x_k^*\in F(x_k)$ for all $k$, one has $x^*\in F(x)$. Recall that $F$ is said to be [*locally bounded*]{} at $x$, if there exist constants $\delta, M\in (0, +\infty)$ such that $\|u^*\|\leq M$ holds for any $u\in B(x, \delta)$ and any $u^*\in F(u)$.
Let $Z$ be a Banach space and $K\subset Z$ be a closed convex cone. The [*partial order*]{} in $Z$ by $K$ is defined as . Let $\varphi: E\rightarrow Z$ be a function. Recall that $\varphi$ is said to be [*K-convex*]{}, if $$\varphi(\lambda x_1+(1-\lambda)x_2)\preceq_K\lambda \varphi(x_1)+(1-\lambda)\varphi(x_2)\ \ {\rm for \ any} \ x_1, x_2\in E \ {\rm and\ any} \ \lambda\in [0, 1].$$ If one takes $Z:=\mathbb{R}$ and $K:=[0, +\infty)$, then $K$-convexity of $\varphi$ reduces to the general convexity of real-valued function $\varphi$.
Duality for convex vector optimization problems
===============================================
In this section, we study one nonlinear convex primal vector optimization problem as well as its associate duality problem and pay main attention to duality theory for this problem. Note that Geoffrion [@Ge] investigated duality theory for nonlinear convex programming with convex function constraints in finite dimension spaces, and gave optimality and weak and strong duality theorems by virtue of the concept of perturbation function. Along the line in [@Ge], we apply this approach to the study on convex vector optimization primal problem in Banach spaces and endeavour to provide several duality results on this primal problem and its dual problem. These duality results on primal problem and its dual will play a key role in the construction of generalized Benders decomposition algorithm for solving problme (VOP) of . Furthermore the finite convergence of this algorithm is also mainly dependent on these duality results (see Section 4). We begin with this convex vector optimization primal problem.
Let $E, Z$ be two Banach spaces and $K\subset Z$ be a closed convex cone with a nonempty interior. We define the partial order in $Z$ by $K$ as follows: for any $z_1, z_2\in Z$, $$z_1\preceq_K z_2\Longleftrightarrow z_2-z_1\in K\ {\rm and} \ z_1\prec_K z_2\Longleftrightarrow z_2-z_1\in{\rm int}(K).$$ We consider the following convex [*primal*]{} programming problem: $$\label{3.1a}
{\rm (P)} \left \{
\begin{array}l
\mathop{\rm minimize}\limits_{x} \ \ \ f(x)\\
{\rm subject\ to}\ \ g(x)\preceq_K 0, \\
\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ x\in X,
\end{array}
\right.$$ where $X\subset E$ is convex, $f: X\rightarrow \mathbb{R}$ is convex and $g: X\rightarrow Z$ is $K$-convex.
The [*dual*]{} problem of (P) is taken to be: $$\label{3.2a}
{\rm (D)} \left \{
\begin{array}l
\mathop{\rm maximize}\limits_{u^*}\ \ \big[\inf\limits_{x\in X} \{f(x)+ \langle u^*, g(x)\rangle\}\big]\\
{\rm subject\ to}\ \ u^*\in K^+,
\end{array}
\right.$$ where $K^+:=\{z^*\in Z^*: \langle z^*, z\rangle\geq 0,\ \forall z\in K\}$ denotes the [*dual cone*]{} of $K$. If we take $E: =\mathbb{R}^n$, $Z:=\mathbb{R}^m$ and $K:=\mathbb{R}^m_+$, primal problem (P) and dual problem (D) reduce to the classic convex programming with convex function constraints and its associate dual programming in finite dimensional spaces, respectively.
Problems (P) and (D) are in close connection with each other and always have optimal values (possibly $\pm\infty$) provided we invoke the customary convention that an infimum (resp. supremum) taken over an empty set is $+\infty$ (resp. $-\infty$). To investigate the interrelationship between problems (P) and (D), we first recall some definitions on problems (P) and (D).
A linear continuous functional $u^*\in K^+$ is said to be [*essentially infeasible*]{} in problem [(D)]{}, if $$\inf\limits_{x\in X} \{f(x)+ \langle u^*, g(x)\rangle\}=-\infty.$$ Problem (D) is said to be [*essentially infeasible*]{}, if every $u^*\in K^+$ is essentially infeasible in problem [(D)]{}; otherwise, problem (D) is said to be [*essentially feasible*]{}.
A pair $(\bar x, \bar u^*)\in E\times Z^*$ is said to satisfy the [*optimality conditions*]{} for problem (P), if $$\label{3.3b}
\left\{\begin{array}l
{\rm (i)}\ f(\bar x)+\langle \bar u^*, g(\bar x)\rangle=\min\limits_{x\in X}\big\{f(x)+\langle \bar u^*, g(x)\rangle\big\},\\
{\rm (ii)}\ \langle \bar u^*, g(\bar x)\rangle=0,\\
{\rm (iii)}\ \bar u^*\in K^+, \\
{\rm (iv)}\ g(\bar x)\preceq_K0.
\end{array}
\right.$$ A linear continuous functional $\bar u^*\in Z^*$ is said to be an [*optimal Lagrange multiplier*]{} for problem (P), if $(\bar x, \bar u^*)$ satisfies the optimality conditions for some $\bar x\in X$.
[**Remark 3.1.**]{} (a) It is easy to verify that if $\bar u^*\in Z^*$ is an optimal Lagrange multiplier, then there exists $\bar x\in X$ such that $\bar x$ is an optimal solution to problem (P). This means that an optimal Lagrange multiplier presupposes the existence of an optimal solution to problem (P). Furthermore, if $\bar u^*\in Z^*$ is an optimal Lagrange multiplier, then $(\bar x, \bar u^*)$ satisfies the optimality conditions (i)-(iv) in for any optimal solution $\bar x$ to problem (P).
\(b) The optimality conditions are equivalent to a constrained [*saddle-point*]{} of the Lagrange function, that is, $(\bar x, \bar u^*)$ satisfies the optimality conditions (i)-(iv) in if and only if $(\bar x, \bar u^*)\in X\times K^+$ with $\langle \bar u^*, g(\bar x)\rangle=0$ and $$f(\bar x)+\langle u^*, g(\bar x)\rangle\leq f(\bar x)+\langle \bar u^*, g(\bar x)\rangle\leq f(x)+\langle \bar u^*, g(x)\rangle\ \ \forall (x, u^*)\in X\times K^+.$$
For the case of primal problem (P) when taking $E:=\mathbb{R}^n$, $Z:=\mathbb{R}^m$ and $K:=\mathbb{R}^m_+$, Geoffrion [@Ge] exploited the concept of [*perturbation function*]{} to study convex duality theory between primal problem (P) and its duality problem (D), and proved the existence of optimal Lagrange multipliers for this primal problem. To delve into the problems (P) and (D) in this section, we consider this notion of perturbation function in vector optimization and study its close interrelationship with optimal Lagrange multipliers for problem (P) of .
Recall that the [*perturbation function $v(\cdot)$*]{} associated with problem (P) is defined on $Z$ as follows: $$\label{3.1}
v: Z\rightarrow \mathbb{R}\cup\{\pm\infty\}, \ z\mapsto v(z):=\inf_{x\in X}\big\{f(x)\ \ {\rm subject \ to }\ g(x)\preceq_K z\big\},$$ where each $z\in Z$ is called the [*perturbation vector*]{} for $v(\cdot)$. We denote $$\label{3.2}
A:=\big\{z\in Z: {\rm there \ exists }\ x\in X\ {\rm such\ that }\ g(x)\preceq_Kz\big\}$$ the [*feasible set*]{} of the perturbed problem. Note that $v(z)=+\infty$ if and only if $z\not\in A$ by the customary convention.
[(i)]{} Let $A$ be defined as . Then $A$ is a convex set and $v(\cdot)$ is convex and monotone nonincreasing on $A$.
[(ii)]{} Suppose that $X$ is compact, $f$ is lower semicontinuous and that $g$ is continuous. Then $A={\rm dom}(v)$ is a closed subset and $v(\cdot)$ is lower semicontinuous. Furthermore, suppose that the Slater constraint qualification $$\label{3.6a}
g(\hat x)\prec_K0 \ \ {\it for \ some} \ \hat x\in X$$ holds. Then $v(\cdot)$ is continuous at $0_Z\in A$.
[**Proof.**]{} (i) The convexity of $A$ and $v(\cdot)$ as well as monotone nonincreasing of $v(\cdot)$ follows from the convexity of $f$ and the $K$-convexity of $g$.
\(ii) Since $X$ is compact, $f$ is lower semicontinuous and $g$ is continuous, it is easy to verify that $A$ is closed and for any $\bar z\in A$ there exists $\bar x\in X$ such that $v(\bar z)=f(\bar x)$. Then $A={\rm dom}(v)$ and $v(z)=+\infty$ for all $z\not\in A$. For the lower semicontinuity of $v(\cdot)$, it suffices to prove that $v(\cdot)$ is lower semicontinuous on $A$. Let $z\in A$ and $z_i\rightarrow z$ with $z_i\in A$ for all $i\in\mathbb{N}$. Then for any $i\in\mathbb{N}$, there exists $x_i\in X$ such that $g(x_i)\preceq_K z_i$ and $$v(z_i)\geq f(x_i)-\frac{1}{i}.$$ Since $X$ is compact, without loss of generality, we can assume that $x_i\rightarrow x\in X$ (considering subsequence if necessary). It follows that $g(x)\preceq_Kz$ as $K$ is closed and $g$ is continuous. This implies that $$\liminf_{i\rightarrow \infty}v(z_i)\geq\liminf_{i\rightarrow \infty}(f(x_i)-\frac{1}{i})\geq f(x)\geq v(z)$$ as $f$ is lower semicontinuous at $x$.
Now, suppose that Slater constraint qualification hold. Then $-g(\hat x)\in{\rm int}(K)$ and thus there exists $\delta>0$ such that $-g(\hat x)+B(0_Z, \delta)\subset K$. This implies that $B(0_Z, \delta)\subset A={\rm dom}(v)$. Hence $0_Z\in {\rm int}(A)={\rm int}({\rm dom}(v))$ and it follows from [@P Proposition 3.3] that $v(\cdot)$ is continuous at $0_Z$. The proof is complete. $\Box$
The following proposition provides an equivalent interpretation to optimal Lagrange multipliers and asserts precisely that the set of optimal Lagrange multipliers is essentially the negative of subdifferential of perturbation function at the origin.
Suppose that problem (P) has an optimal solution and denote $U$ the set of all optimal Lagrange multiplier for problem (P). Then $U=-\partial v(0)$.
[**Proof.**]{} The “$\subset$" part. Let $\bar u^*\in U$. Then there exists $\bar x\in X$ such that the pair $(\bar x, \bar u^*)$ satisfies the optimal conditions (i)-(iv) in . From the optimal conditions (i) and (ii), we have $$\label{3.3}
f(x)+\langle \bar u^*, g(x)\rangle\geq f(\bar x)+\langle \bar u^*, g(\bar x)\rangle=f(\bar x),\ \ \forall x\in X.$$ Let $z\in A$ and $x\in X$ with $g(x)\preceq_Kz$. Then $\langle \bar u^*, z\rangle\geq\langle \bar u^*, g(x)\rangle$ by the optimal condition (iii) and it follows from that $f(x)\geq f(\bar x)-\langle \bar u^*, z\rangle$. By taking the infimum of the left-hand side over the indicated values of $x$, one has $$v(z)\geq f(\bar x)-\langle \bar u^*, z\rangle\ \ \forall z\in A.$$ This implies that $-\bar u^*\in \partial v(0)$ since $f(\bar x)=v(0)$ and $v(z)=+\infty$ for any $z\not\in A$.
For the “$\supset$" part, let $-\bar u^*\in\partial v(0)$ and $\bar x$ be an optimal solution of problem (P). Then $g(\bar x)\preceq_K0$. We only need to prove that the pair $(\bar x, \bar u^*)$ satisfies the optimality conditions (i)-(iii). Noting that $-\bar u^*\in\partial v(0)$, it follows that $$\label{3.4}
v(z)\geq v(0)-\langle\bar u^*, z\rangle\ \ \forall z\in Z.$$ This implies that $\langle\bar u^*, z\rangle\geq v(0)-v(z)\geq 0$ holds for all $z\in K$ as $v(\cdot)$ is monotone nonincreasing and consequently $\bar u^*\in K^+$. Noting that decreasing the right-hand side of problem (P) to $g(\bar x)$ will not destroy the optimality of $\bar x$, it follows that $$v(g(\bar x))=v(0) \ \ {\rm and} \ \ \langle \bar u^*, g(\bar x)\rangle\geq 0.$$ On the other hand, $\langle \bar u^*, g(\bar x)\rangle\leq 0$ follows from $-g(\bar x)\in K$ and $\bar u^*\in K^+$. This means that $\langle \bar u^*, g(\bar x)\rangle=0$ and thus the optimality condition (ii) holds. To prove the optimality condition (i), for any $x\in X$, when taking $z:=g(x)$ in , we have $$v(g(x))\geq v(0)-\langle \bar u^*, g(x)\rangle.$$ Since $f(x)\geq v(g(x))$ for all $x\in X$ and $f(\bar x)=v(0)$, one has $$f(x)+\langle \bar u^*, g(x)\rangle\geq f(\bar x)=f(\bar x)+\langle \bar u^*, g(\bar x)\rangle\ \ \forall x\in X$$ (thanks to the optimality condition (ii)). Hence the optimality condition (i) holds. The proof is complete. $\Box$
It is known from Proposition 3.2 that optimal Lagrange multipliers can be determined from subdifferential $\partial v(0)$ and thus it is necessary to study equivalent conditions ensuring the nonempty of $\partial v(0)$. The following proposition provides a criterion for the existence of subdifferential of perturbation function $v(\cdot)$ at a point where it is finite.
Let $v(\cdot)$ associate with problem (P) be defined as and $\bar z\in {\rm dom}(v)$. Suppose that $X$ is compact, $f$ is lower semicontinuous and that $g$ is continuous. Then $\partial v(\bar z)\not=\emptyset$ if and only if there exists $M\in (0, +\infty)$ such that $$\frac{v(\bar z)-v(z)}{\|z-\bar z\|}\leq M\ \ \forall z\in Z\backslash\{\bar z\}.$$
[**Proof.**]{} By virtue of Proposition 3.1, one has $v(\cdot)$ is a lower semicontinuous convex function and $v(z)>-\infty$ for all $z\in Z$. Let $z^*\in \partial v(\bar z)$. Then the necessity part follows by taking $M:=\|z^*\|+1$. It suffices to prove the sufficiency part. Let $$\begin{aligned}
\Phi:&=&\{(z, r)\in Z\times \mathbb{R}: v(\bar z)-v(z)\geq r\}\ {\rm and} \\
\Psi: &=&\{(z, r)\in Z\times \mathbb{R}: M\|z-\bar z\|< r\}.\end{aligned}$$ Then $\Psi$ and $\Phi$ are convex sets, $\Psi\cap\Phi=\emptyset$ and $\Psi$ is open. By the seperation theorem (cf. [@R Theorem 3.4]), there exists $(z^*, \beta)\in (Z\times\mathbb{R})^*=Z^*\times\mathbb{R}$ with $(z^*, \beta)\not=(0, 0)$ such that $$\label{3.5}
\sup_{(z, r)\in \Psi}\big\{\langle z^*, z\rangle+\beta r\big\}<\alpha<\inf_{(z, r)\in \Phi}\big\{\langle z^*, z\rangle+\beta r\big\}.$$ Then $\beta<0$ (thanks to $(\bar z, 1)\in\Psi$ and $(\bar z, 0)\in\Phi$). Noting that $(\bar z, 0)\in \Phi$ and $(\bar z, \varepsilon)\in\Psi$ for all $\varepsilon>0$, it follows from that $$\begin{aligned}
\langle z^*, z-\bar z\rangle+\beta r&<&0\ \ \forall (z, r)\in \Psi \ \ {\rm and} \\
\langle z^*, z-\bar z\rangle+\beta r&\geq& 0\ \ \forall (z, r)\in \Phi.\end{aligned}$$ Noting that $(z, v(\bar z)-v(z))\in \Phi$ for all $z\in {\rm dom}(v)$, it follows that $$\langle \frac{z^*}{\beta}, z-\bar z\rangle\leq v(z)-v(\bar z),\ \ \forall z\in Z.$$ This implies that $\widetilde{z}^*:=\frac{z^*}{\beta}\in\partial v(\bar z)$. The proof is complete. $\Box$
[**Remark 3.2.**]{} (a) The proof of Proposition 3.3 is inspired by some ideas from [@Ge Lemma 2], and furthermore it is known from the proof that the conclusion is still valid for general proper extended-real-valued, but not taking negative infinity, convex function defined on $Z$.
\(b) Under the assumptions of Proposition 3.1(ii), the perturbation function $v(\cdot)$ is a lower semicontinuous convex function. A deep theorem referring to the subdifferential $\partial v$, proved by Br[ø]{}ndsted and Rockafellar, is that ${\rm dom}(\partial v)$ is dense in ${\rm dom}(v)$; that is for any $z\in Z$ with $v(z)\in\mathbb{R}$, there exists $z_n\rightarrow z$ such that $\partial v(z_n)\not=\emptyset$ for all $n\in \mathbb{N}$. Readers are invited to consult [@P Theorem 3.17] for Br[ø]{}ndsted-Rockafellar theorem and its proof in detail.
Using propositions 3.1, 3.2 and 3.3, we obtain the following theorem on characterizations for the existence of optimal Lagrange multipliers.
Suppose that problem (P) has an optimal solution. Then the following statements are equivalent:
[(i)]{} The set of all Lagrange multipliers for problem (P) is nonempty.
[(ii)]{} $\partial v(0)$ is a nonempty set.
[(iii)]{} $v(0)$ is finite and there exists $M\in (0, +\infty)$ such that $$\label{3.5a}
v(0)-v(z)\leq M\|z\|$$ holds for any $z\in Z$.
The following result provides a criterion for the essential feasibility of problem (D) and also gives one necessary condition for essential feasible problem (D).
[(i)]{} Suppose that problem (D) is essentially feasible. Then $v(z)>-\infty$ for all $z\in A$.
[(ii)]{} Suppose that $X$ is compact, $f$ is lower semicontinuous and that $g$ is continuous. Then problem (D) is essentially feasible.
[**Proof.**]{} (i) Suppose that problem (D) is essentially feasible. Then there exists $u^*\in K^+$ and $\beta\in \mathbb{R}$ such that $$\label{3.7}
f(x)+\langle u^*, g(x)\rangle\geq\beta,\ \ \forall x\in X.$$ Let $z\in A$ and $x\in X$ with $g(x)\preceq_Kz$. By , one has $$f(x)\geq f(x)+\langle u^*, g(x)-z\rangle\geq\beta-\langle u^*, z\rangle.$$ This implies that $v(z)\geq\beta-\langle u^*, z\rangle>-\infty$.
\(ii) By virtue of Proposition 3.1, one has $v(\cdot)$ is a lower semincontinuous convex function and $A={\rm dom}(v)$. Let $z\in A$. Using [@P Theorem 3.17], there exists $z_i\rightarrow z$ such that $v(z_i)\in\mathbb{R}$ and $\partial v(z_i)\not=\emptyset$ for all $i$. Then we can take $u_i^*\in\partial v(z_i)$ such that $$\label{3.6}
v(z)\geq v(z_i)+\langle u_i^*, z-z_i\rangle,\ \ \forall z\in Z,$$ and consequently $-u_i^*\in K^+$ due to the nonincreasing of $v(\cdot)$ and . Let $x\in X$. Then $g(x)\in A$ and it follows from that $$\langle u_i^*, g(x)-z_i\rangle\leq v(g(x))-v(z_i)\leq f(x)-v(z_i).$$ This implies that $f(x)+\langle-u_i^*, g(x)\rangle\geq v(z_i)+\langle-u_i^*, z_i\rangle$ and thus $$\inf_{x\in X}\big\{f(x)+\langle-u_i^*, g(x)\rangle\}\geq v(z_i)+\langle-u_i^*, z_i\rangle>-\infty.$$ Hence $-u_i^*$ is feasible to problem (D) and problem (D) is essentially feasible. The proof is complete. $\Box$
Clearly the customary weak duality result that the optimal value of primal problem (P) is not smaller than the optimal value of problem (D) holds. Furthermore, the next proposition is the strong duality result on problems (P) and (D) which demonstrates the close connection between optimal Lagrange multipliers and solutions to the dual problem (D).
Let $v(0)$ be finite. Then $\bar u^*\in Z^*$ is an optimal solution of problem (D) and the optimal values of problem (P) and problem (D) equal if and only if $-\bar u^*\in\partial v(0)$.
[**Proof.**]{} We first prove the sufficiency part. Suppose that $-\bar u^*\in\partial v(0)$. Then $$v(z)\geq v(0)+\langle -\bar u^*, z\rangle,\ \ \forall z\in Z.$$ Thus $\langle \bar u^*, z\rangle\geq v(0)-v(z)\geq 0$ for all $z\in K$ by the nonincreasing of $v(\cdot)$ and $\bar u^*\in K^+$. For any $x\in X$, $f(x)\geq v(g(x))$ and $f(x)+\langle \bar u^*, g(x)\rangle\geq v(0)$. This implies that $$\label{3.13a}
\inf_{x\in X}\{f(x)+\langle \bar u^*, g(x)\rangle\}\geq v(0)=\inf_{x\in X}\{f(x): g(x)\preceq_K 0\}.$$ Using the weak duality, one has $$\label{3.14b}
\max_{u^*\in K^+}\Big\{\inf_{x\in X}\{f(x)+\langle u^*, g(x)\rangle\}\Big\}\leq \inf_{x\in X}\{f(x): g(x)\preceq_K 0\}=v(0)$$ and it follows from and that $$\inf_{x\in X}\{f(x)+\langle\bar u^*, g(x)\rangle\}=\max_{u^*\in K^+}\Big\{\inf_{x\in X}\{f(x)+\langle u^*, g(x)\rangle\}\Big\}=v(0).$$ This means that $\bar u^*$ is an optimal solution of problem (D) and the optimal values of problem (P) and problem (D) equal.
The necessity part. Let $\bar u^*$ be an optimal solution of problem (D) and the optimal values of problems (P) and (D) equal. Then $\bar u^*\in K^+$ and $$\label{3.6b}
\inf_{x\in X}\{f(x)+\langle\bar u^*, g(x)\rangle\}=v(0).$$ For any $z\in A$ and any $x\in X$ with $g(x)\preceq_Kz$, by , one has $$f(x)+\langle\bar u^*, z\rangle\geq f(x)+\langle\bar u^*, g(x)\rangle\geq v(0).$$ This implies that $$v(z)\geq v(0)+\langle-\bar u^*, z\rangle \ \ \forall z\in A.$$ Hence $-\bar u^*\in \partial v(0)$ as $v(z)=+\infty$ for all $z\not\in A$. The proof is complete. $\Box$
We close this section with the following proposition which will be used in next section.
Suppose that the optimal value of problem (D) is finite. Then $0_Z\in \overline A$.
[**Proof.**]{} Suppose to the contrary that $0_Z\not\in \overline A$. By the seperation theorem, there exist $z^*\in Z^*$ with $z^*\not=0$ and $\alpha\in \mathbb{R}$ such that $$\label{3.14a}
\inf_{z\in A}\langle z^*, z\rangle>\alpha>0.$$ Let $u^*\in K^+$ such that $u^*$ is essentially feasible in problem (D). Then $$\inf_{x\in X}\{f(x)+\langle u^*, g(x)\}>-\infty.$$ This and imply that $u^*+tz^*$ is also essentially feasible in problem (D) for all $t>0$ as $g(x)\in A$. Hence $$\inf_{x\in X}\{f(x)+\langle u^*+tz^*, g(x)\rangle\}\geq\inf_{x\in X}\{f(x)+\langle u^*, g(x)\rangle\}+t\inf_{x\in X}\langle z^*, g(x)\rangle.$$ Letting $t\rightarrow +\infty$ and by virtue of , we obtain the contradiction that optimal value of (D) is $+\infty$. The proof is complete. $\Box$
Generalized Benders decomposition for MINLPs with vector conic constraint
=========================================================================
In this section, we pay main attention to one class of MINLPs with vector conic constraint, and use generalized Benders decomposition approach and duality results on convex vector optimization problems to construct an appropriate algorithm for solving this MINLP. We begin with this class of MINLPs.
Let $E, Z$ be two Banach spaces, $D$ be a normed linear space, and let $X\subset E$ be a closed convex set, $Y\subset D$ be a set with discrete variables and $K\subset Z$ be a closed convex cone with a nonempty interior. This MINLP problem (VOP) is defined as follows $$\label{4.1a}
{\rm (VOP)} \left \{
\begin{array}l
\mathop{\rm minimize}\limits_{x,\, y} \ \ \ f(x, y)\\
{\rm subject\ to}\ \ g(x, y)\preceq_K 0, \\
\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ x\in X, y\in Y\ {\rm discrete\ variable},
\end{array}
\right.$$ where $f: X\times Y\rightarrow \mathbb{R}$ and $g: X\times Y\rightarrow Z$ satisfy that $f(\cdot, y)$ is convex and $g(\cdot, y)$ is $K$-convex on $X$ for any fixed $y\in Y$.
As pointed out in [@Ge1], Geoffrion employed nonlinear duality theory and generalized Benders decomposition to project MINLP problem $\mathcal{P}$ in onto $y$-space, rather than $xy$-space, reformulate problem $\mathcal{P}$ as one equivalent master problem and construct the generalized Benders decomposition procedure for solving relaxation of master problems. To solve problem (VOP) along this line, we are inspired to separate problem (VOP) into many independent vector optimization problems by fixing discrete variables $y$.
Let $y\in Y$ be fixed. We consider the following vector optimization problem $P(y)$ $$P(y) \left \{
\begin{array}l
\mathop{\rm minimize}\limits_{x} \ \ \ f(x, y)\\
{\rm subject\ to}\ \ g(x, y)\preceq_K 0, \\
\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ x\in X,
\end{array}
\right.$$ and its associate dual is defined as follows: $$D(y) \left \{
\begin{array}l
\mathop{\rm maximize}\limits_{u^*}\ \ \big[\inf\limits_{x\in X} \{f(x, y)+ \langle u^*, g(x, y)\rangle\}\big]\\
{\rm subject\ to}\ \ u^*\in K^+.
\end{array}
\right.$$ The perturbation function $v_y(\cdot)$ associated with problem $P(y)$ is defined by $$\label{4.2a}
v_y(z):=\inf_{x\in X}\{f(x, y): g(x, y)\preceq_K z\}, \ \ \forall z\in Z.$$
We denote $$\label{4.1}
V:=\{y\in Y: g(x, y)\preceq_K 0\ \ {\rm for\ some}\ x\in X\}$$ the feasible set of all values of $y\in Y$ for which vector optimization problem $P(y)$ is feasible. Then problem (VOP) can be equivalently rewritten as $$\label{4.4a}
\left \{
\begin{array}l
\mathop{\rm minimize}\limits_{y} \ \ \ v_y(0)\\
{\rm subject\ to}\ \ y\in V.
\end{array}
\right.$$
After separating problem (VOP) into many problems $P(y)$, it is necessary to establish the master problem which is equivalent to problem (VOP). The key step for this equivalent reformulation is to provide dual equivalent representation for the optimal value $v_y(0)$ of problem $P(y)$ and feasible set $V$. To achieve it, we first give two propositions on the dual equivalent interpretations of $v_y(0)$ and $V$.
The first proposition shows that feasible set $V$ is represented in terms of the intersection of a collection of regions that contain it.
Suppose that $X$ is compact and $g(\cdot, y)$ is continuous for any $y\in Y$. Then $\bar y\in V$ if and only if $$\label{4.2}
\inf_{x\in X}\langle u^*, g(x, \bar y)\rangle\leq 0 \ \ \forall u^*\in K^+\ {\rm with}\ \|u^*\|=1.$$
[**Proof.**]{} The necessity part follows from the definition of $V$ in . We only need to prove the sufficiency part. Let $$\label{4.3}
A(y):=\{z\in Z: g(x, y)\preceq_K z\ {\rm for\ some}\ x\in X\},\ \ \forall y\in Y.$$ Since $X$ is compact and $g(\cdot, y)$ is continuous for any $y\in Y$, it follows that $A(y)$ is closed for any $y\in Y$. By , one has $$\sup_{u^*\in K^+, \|u^*\|=1}\inf_{x\in X}\langle u^*, g(x, \bar y)\rangle\leq 0.$$ This implies that $$\label{4.4}
\max_{u^*\in K^+}\Big\{\inf_{x\in X}\langle u^*, g(x, \bar y)\rangle\Big\}=0.$$ We consider the following vector optimization problem: $$\widetilde{P}(\bar y)\left \{
\begin{array}l
\mathop{\rm minimize}\limits_{x} \ \ \widetilde{f}_{\bar y}(x):=\langle 0_{E^*}, x\rangle\\
{\rm subject\ to}\ \ g(x, \bar y)\preceq_K 0_Z,\\
\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ x\in X.
\end{array}
\right.$$ Then the dual of problem $\widetilde{P}(\bar y)$ is as follows: $$\widetilde{D}(\bar y) \left \{
\begin{array}l
\mathop{\rm maximize}\limits_{u^*}\ \ \big[\inf\limits_{x\in X} \{\langle u^*, g(x, \bar y)\rangle\}\big]\\
{\rm subject\ to}\ \ u^*\in K^+,
\end{array}
\right.$$ and implies the optimal value of problem $\widetilde{D}(\bar y)$ equals $0$. Using the proof of Proposition 3.7, we obtain that $0_Z\in\overline{A(\bar y)}= A(\bar y)$. Then there exists $x\in X$ such that $g(x, \bar y)\preceq_K0_Z$ and therefore $\bar y\in V$. The proof is complete. $\Box$
By virtue of Theorem 3.4 and Proposition 3.6, we obtain the following proposition on the dual interpretation of $v_y(0)$ which is given by the pointwise supremum of a collection of functions.
Suppose that $v_y(0)$ is finite and problem $P(y)$ possesses a Lagrange multiplier for any $y\in V$. Then the optimal value of problem $P(y)$ equals that of its dual problem $D(y)$ for all $y\in V$; that is, $$v_y(0)=\max_{u^*\in K^+}\Big\{\inf_{x\in X}\big\{f(x, y)+\langle u^*, g(x, y)\rangle\big\}\Big\}$$ holds for all $y\in V$.
Under the assumptions that $X$ is compact, $g(\cdot, y)$ is continuous for any $y\in Y$ and problem $P(y)$ possesses an optimal Lagrange multiplier for any $y\in Y$ where problem $P(y)$ is feasible, by using Propositions 4.1 and 4.2, we obtain that problem (VOP) is equivalent to the following master problem: $$\label{4.9a}
\left\{
\begin{array}l
\mathop{\rm minimize}\limits_{y\in Y}\ \ \Big[\sup\limits_{u^*\in K^+}\big\{\inf\limits_{x\in X}\{f(x, y)+\langle u^*, g(x, y)\rangle\}\big\}\Big]\\
{\rm subject \ to}\ \ \inf\limits_{x\in X}\langle z^*, g(x, y)\rangle\leq 0, \ \ \forall z^*\in K^+\ {\rm with}\ \|z^*\|=1.
\end{array}
\right.$$ Using the definition of supremum as the smallest upper bound, the master problem is equivalent to the following master problem (MP): $$\label{4.7}
{\rm(MP)}\left\{
\begin{array}l
\mathop{\rm minimize}\limits_{y\in Y,\, \eta\in\mathbb{R}}\ \ \ \eta\\
{\rm subject\ to}\ \ \eta\geq\inf\limits_{x\in X}\{f(x, y)+\langle u^*, g(x, y)\rangle\}, \ \ \forall u^*\in K^+\\
\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \inf\limits_{x\in X}\langle z^*, g(x, y)\rangle\leq 0,\ \ \forall z^*\in K^+\ {\rm with}\ \|z^*\|=1.
\end{array}
\right.$$
It is known that one type of relaxation, in which not all constraints are included, is one natural strategy for solving master problem (MP) in . We begin to solve one relaxed version of master problem, not including all constraints in . If the obtained optimal solution does not satisfy constraints having not been considered, then we generate and add to the relaxed problem one or more violated constraints and solve it again. We continue this approach until a relaxed problem solution satisfies all constraints, or until a termination criterion demonstrates that a solution of acceptable accuracy has been obtained. Geoffrion [@Ge1] discussed in detail that a solution to a relaxed version of master problem can be tested for feasibility with respect to the ignored constraints and one violated constraint can be generated in case of infeasibility. This discussion given in [@Ge1] can also be applied to problem (MP) in and it enables (MP) of to be solved by this relaxation approach.
Now, we can formally state the generalized Benders decomposition procedure for solving problem (VOP). For the validity of equivalence between problems (VOP) and master problem (MP), we suppose that (VOP) satisfies the following assumption:
\(A) [*$X$ is compact, $g(\cdot, y)$ is continuous for any $y\in Y$ and the Slater constraint qualification $$\label{4.17}
g(\hat x, y)\prec_K0\ \ {\it for \ some} \ \hat x\in X$$ holds for any $y\in Y$ where problem $P(y)$ is feasible.*]{}
Using Propositions 3.1 and 3.2, it follows from the Slater constraint qualification that $P(y)$ possesses an optimal Lagrange multiplier for any $y\in Y$ where problem $P(y)$ is feasible.
The detailed algorithm, used to solve problem (VOP) by generalized Benders decomposition procedure, is stated as follows.
[**Algorithm 1**]{}(Generalized Benders Decomposition procedure for problem (VOP))
[**Step 1**]{}: Take $y_{1}\in V$ and $z_{1}^*\in K^+$ with $\|z^*_{1}\|=1$. Solve the primal problem $P(y_{1})$ and obtain an optimal Lagrange multiplier $u^*_{1}$ of $P(y_{1})$. Set $$T^1=S^1:=\{1\}\ {\rm and} \ {\rm UBD}^1:=v_{y_{1}}(0).$$ Select the convergence tolerance parameter $\varepsilon>0$ and let $k:=1$.
[**Step 2**]{}: Solve the following relaxed master problem ${\rm RMP}(T^k, S^k)$: $$\label{4.8}
{\rm RMP}(T^k, S^k)\left\{
\begin{array}l
\mathop{\rm minimize}\limits_{y\in Y,\, \eta\in\mathbb{R}} \ \ \ \eta\\ {\rm subject\ to}\ \ \eta\geq\inf\limits_{x\in X}\{f(x, y)+\langle u_{i}^*, g(x, y)\rangle\}, \ \ \forall i\in T^k,\\
\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \inf\limits_{x\in X}\langle z_{j}^*, g(x, y)\rangle\leq 0,\ \ \forall j\in S^k.
\end{array}
\right.$$ Denote $(y_{k+1}, \eta_{k+1})$ the optimal solution of ${\rm RMP}(T^k, S^k)$. If ${\rm UBD}^k\leq \eta_{k+1}+\varepsilon$, terminate; otherwise, go to [**Step 3**]{}.
[**Step 3**]{}: Solve the primal problem $P(y_{k+1})$. There must occur one of the following two cases:
[**(a)**]{} $v_{y_{k+1}}(0)<+\infty$. If $v_{y_{k+1}}(0)\leq\eta_{k+1}+\varepsilon$, terminate; otherwise, determine an optimal Lagrange multiplier $u_{k+1}^*$ of problem $P(y_{k+1})$, and let $$T^{k+1}:=T^k\cup\{k+1\}, S^{k+1}:=S^k\ {\rm and} \ {\rm UBD}^{k+1}:=\min\{{\rm UBD}^{k}, v_{y_{k+1}}(0)\}.$$ Set $k:=k+1$ and return to [**Step 2**]{}.
[**(b)**]{} $v_{y_{k+1}}(0)=+\infty$; that is, problem $P(y_{k+1})$ is infeasible. Take $z_{k+1}^*\in K^+$ with $\|z_{k+1}^*\|=1$ such that $$\inf_{x\in X}\langle z_{k+1}^*, g(x, y_{k+1})\rangle>0.$$ Let $$T^{k+1}:=T^k, S^{k+1}:=S^k\cup\{k+1\} \ \ {\rm and} \ \ {\rm UBD}^{k+1}:={\rm UBD}^{k}.$$ Set $k:=k+1$ and return to [**Step 2**]{}.
Now, we study the following example and demonstrate the generalized Benders decomposition procedure when solving MINLP problem by Algorithm 1.
[**Example 4.1.**]{} Consider the following MINLP problem: $$\label{4.13a}
\left \{
\begin{array}l
\mathop{\rm minimize}\limits_{x,\,y} \ \ \ f(x, y)=-x+\max\{y-1, -y+1\}\\
{\rm subject\ to}\ \ g(x, y)=x+\max\{-y, y-2\}\leq 0, \\
\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ x\in [-1, 1],\, y\in \{0, 1, 2\}.
\end{array}
\right.$$ Then $X=[-1, 1]$, $Y=\{0,1,2\}$ and $K=K^+=[0,+\infty)$. It is easy to verify that $(x^*, y^*)=(1, 1)$ and $f^*=f(1, 1)$ is the solution of problem . First, we take $y_1=0$ and $z_1^*=1\in K^+$. Solve primal problem $P(y_1)$ and its dual $D(y_1)$, and denote an optimal Lagrange multiplier $u_1^*=1$. Let $T^1=S^1:=\{1\}, {\rm UBD}^1=v_{y_1}(0)=1$ and $\varepsilon\in (0,1)$. By computing, the relaxed master problem ${\rm RMP}(T^1, S^1)$ is defined as follows: $${\rm RMP}(T^1, S^1)\left\{
\begin{array}l
\mathop{\rm minimize}\limits_{y\in Y,\, \eta\in\mathbb{R}} \ \ \ \eta\\ {\rm subject\ to}\ \ \max\{y-1, -y+1\}+\max\{-y, y-2\}\leq \eta,\\
\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ -1+\max\{-y, y-2\}\leq 0.
\end{array}
\right.$$ It is easy to verify that its solution is $(y_2,\eta_2)=(1,0)$. Since ${\rm UBD}^1>\eta_2+\varepsilon$, by [**Step 2**]{}, we go to [**Step 3**]{} and solve primal problem $P(y_2)$. Noting that $x_2=1$ is an optimal solution of $P(y_2)$ and $v_{y_2}(0)=-1<{\rm UBD}^1+\varepsilon$, then terminate the algorithm by [**Step 2**]{} and consequently $(x_2,y_2, v_{y_2}(0))=(1,1,-1)$ is an $\varepsilon$-tolerance optimal solution of problem .
Next, we focus on theoretical convergence of Algorithm 1 by generalized Benders decomposition procedure and prove convergence theorems with the help of some mild assumptions. We first need the following proposition which will be used in the proof of convergence theorems.
Suppose that $X$ is compact, and $f(\cdot, \cdot)$, $g(\cdot, \cdot)$ are continuous on $X\times Y$. Denote $U(y)$ the set of all optimal Lagrange multipliers of problem $P(y)$ for any $y\in Y$ where problem $P(y)$ is feasible and let $$\label{4.18}
L(y, u^*):=\inf_{x\in X}\big\{f(x, y)+\langle u^*, g(x, y)\rangle\big\},\ \ \forall (y, u^*)\in Y\times K^+.$$ Then $L(\cdot, \cdot)$ is $\|\cdot\|\times w^*$ continuous on $Y\times K^+$ and the set-valued mapping $U(\cdot)$ is norm-to-weak$^*$ upper semicontinuous on $V$.
[**Proof.**]{} Let $(\bar y, \bar u^*)\in Y\times K^+$, and take any generalized sequence $(y_{\alpha}, u^*_{\alpha})$ in $Y\times K^+$ such that $y_{\alpha}\rightarrow \bar y$ and $u^*_{\alpha}\stackrel{w^*}\longrightarrow \bar u^*$. Then for any $\alpha$, there exists $x_{\alpha}\in X$ such that $$\label{4.19}
f(x_{\alpha}, y_{\alpha})+\langle u^*_{\alpha}, g(x_{\alpha}, y_{\alpha})\rangle=L(y_{\alpha}, u^*_{\alpha})$$ as $X$ is compact and $f(\cdot, \cdot), g(\cdot, \cdot)$ are continuous. Noting that $X$ is compact, without loss of generality, we can assume that $x_{\alpha}\rightarrow \bar x\in X$ (considering generalized subsequence if necessary). By , one has $$\begin{aligned}
L(\bar y, \bar u^*)&\leq& f(\bar x, \bar y)+\langle \bar u^*, g(\bar x, \bar y)\rangle\\&\leq&\lim_{\alpha}(f(x_{\alpha}, y_{\alpha})+\langle u^*_{\alpha}, g(x_{\alpha}, y_{\alpha})\rangle)\\&=&\liminf_{\alpha}L(y_{\alpha}, u^*_{\alpha}).\end{aligned}$$ This implies that $L(\cdot, \cdot)$ is $\|\cdot\|\times w^*$ lower semicontinuity at $(\bar y, \bar u^*)$. It suffices to prove the $\|\cdot\|\times w^*$ upper semicontinuity of $L(\cdot, \cdot)$ at $(\bar y, \bar u^*)$. For any $x\in X$, one has $$\begin{aligned}
f(x, \bar y)+\langle \bar u^*, g(x, \bar y)\rangle&=&\lim_{\alpha}(f(x, y_{\alpha})+\langle u^*_{\alpha}, g(x, y_{\alpha})\rangle)\\ &\geq&\limsup_{\alpha} L(y_{\alpha}, u_{\alpha}^*),\end{aligned}$$ This yields that $$L(\bar y, \bar u^*)=\inf_{x\in X}(f(x, \bar y)+\langle \bar u^*, g(x, \bar y)\rangle)\geq\limsup_{\alpha} L(y_{\alpha}, u_{\alpha}^*)$$ and consequently $L(\cdot, \cdot)$ is $\|\cdot\|\times w^*$ upper semicontinuous at $(\bar y, \bar u^*)$. Thus $L(\cdot, \cdot)$ is $\|\cdot\|\times w^*$ continuous at $(\bar y, \bar u^*)$.
Next, we prove the norm-to-weak$^*$ upper semicontinuity of mapping $U(\cdot)$. Let $\bar y\in V$ and take any generalized sequence $(y_{\alpha}, u^*_{\alpha})$ in $V\times K^+$ such that $y_{\alpha}\rightarrow\bar y$ and $u^*_{\alpha}\stackrel{w^*}\longrightarrow \bar u^*$ with $u^*_{\alpha}\in U(y_{\alpha})$. We only need to show that $\bar u^*\in U(\bar y)$. By Definition 3.2, for any $\alpha$, there exists $x_{\alpha}\in X$ such that $g(x_{\alpha}, y_{\alpha})\preceq_K 0$, $u^*_{\alpha}\in K^+$ with $\langle u^*_{\alpha}, g(x_{\alpha}, y_{\alpha})\rangle=0$ and $$\label{4.20}
f(x_{\alpha}, y_{\alpha})+\langle u^*_{\alpha}, g(x_{\alpha}, y_{\alpha})\rangle=L(y_{\alpha}, u^*_{\alpha}).$$ Since $X$ is compact, without loss of generality, we can assume that $x_{\alpha}\rightarrow \bar x\in X$ (considering generalized subsequence if necessary) and it follows from the continuity of $g(\cdot, \cdot)$ that $g(\bar x, \bar y)\preceq_K0$, $\langle \bar u^*, g(\bar x, \bar y)\rangle=0$ and $\bar u^*\in K^+$. Using and the continuity of $L(\cdot, \cdot)$ and $f(\cdot, \cdot)$, one has $$f(\bar x, \bar y)+\langle \bar u^*, g(\bar x, \bar y)\rangle=L(\bar y, \bar u^*).$$ This implies that $(\bar u^*, \bar x)$ satisfies the optimality conditions (i)-(iv) in for $P(\bar y)$ and consequently $\bar u^*\in U(\bar y)$. The proof is complete. $\Box$
Suppose that $X, Y$ are compact and $f(\cdot, \cdot)$ and $g(\cdot, \cdot)$ are continuous on $X\times Y$. Denote $U(y)$ the set of all optimal Lagrange multipliers of $P(y)$ for any $y\in Y$ for which problem $P(y)$ is feasible and suppose that the set-valued mapping $y\mapsto U(y)$ is locally bounded on $V$. Then for any given $\varepsilon>0$, the algorithm by generalized Benders decomposition procedure terminates in a finite number of steps.
[**Proof.**]{} Suppose to the contrary that there exists $\varepsilon_0>0$ such that the procedure does not terminate in a finite number of steps. Then there exists a sequence $(y_k, \eta_k)$ in $V\times \mathbb{R}$ generated by [**Step 2**]{}. For any $k$, we take $u^*_k\in U(y_k)$. From [**Step 2**]{}, it is not hard to verify that $\{\eta_k\}$ is nondecreasing and bounded above. By taking a generalized subsequence if necessary, we can assume that $(y_k, \eta_k)\rightarrow (\bar y, \bar \eta)\in V\times \mathbb{R}$ since $X, Y$ are compact and $V$ is closed. Noting that mapping $U(\cdot)$ is locally bounded at $\bar y$, it follows that $\{u^*_k\}$ is bounded. Applying Banach-Alaoglu theorem (cf. [@R Theorem 3.15]), we can assume that $u_k^*\stackrel{w^*}\longrightarrow \bar u^*$ (considering the generalized subsequence if necessary). By the norm-to-weak$^*$ upper semicontinuity of $U(\cdot)$ in Proposition 4.3, one has $\bar u^*\in U(\bar y)$. Using [**Step 2**]{} and [**Step 3**]{}, we yield $$\eta_{k+1}\geq L(y_{k+1}, u_k^*)$$ and consequently $$\label{4.21}
\bar\eta\geq L(\bar y, \bar u^*)$$ by taking limits as $k$. Noting that $u^*_k\in U(y_k)$ and $\bar u^*\in U(\bar y)$, it follows from Propositions 3.2 and 3.6 that $$v_{y_k}(0)=L(y_k, u_k^*) \ \ {\rm and} \ \ v_{\bar y}(0)=L(\bar y, \bar u^*).$$ This and the $\|\cdot\|\times w^*$ continuity of $L(\cdot, \cdot)$ imply that $$\label{4.22}
\lim_{k}v_{y_k}(0)=v_{\bar y}(0).$$ Then, for $\varepsilon_0>0$, when $k$ is sufficiently large, one has $$\eta_k+\frac{\varepsilon_0}{2}>\bar\eta\geq v_{\bar y}(0)>v_{y_k}(0)-\frac{\varepsilon_0}{2}$$ (thanks to and ). Thus $$v_{y_k}(0)\leq\eta_k+\varepsilon_0\ \ {\rm for\ all}\ k \ {\rm sufficiently\ large},$$ which contradicts the termination criterion at [**Step 3(a)**]{}. The proof is complete. $\Box$
The following convergence theorem can be obtained from Theorem 4.1.
Suppose that $X$ is compact, the cardinality of $Y$ is finite and that $f(\cdot, y)$ is continuous on $X$ for any fixed $y\in Y$. Then for any given $\varepsilon>0$, the algorithm by generalized Benders decomposition procedure terminates in a finite number of steps.
[**Proof.**]{} Let $U(y)$ denote the set of all optimal Lagrange multipliers of problem $P(y)$ for which problem $P(y)$ is feasible. Let $y\in Y$ such that problem $P(y)$ is feasible. By virtue of the Slater constraint qualification and Proposition 3.1, one has $v_y(\cdot)$ is continuous at $0\in Z$ and it follows from Theorem 3.4 that $$\label{4.18a}
U(y)=-\partial v_y(0).$$ Since $v_y(\cdot)$ is continuous at $0$, by using [@P Proposition 1.11], one has $\partial v_y(0)$ is bounded. Noting that the cardinality of $Y$ is finite, it follows from that $U(\cdot)$ is bounded on $V$. Hence the termination criterion in Theorem 4.5 follows from Theorem 4.4. The proof is complete. $\Box$
Conclusions
===========
This paper is devoted to the study on one class of MINLPs with vector conic constraint in the context of Banach spaces. By using convex primal vector optimization programming and its associated duality results obtained in Section 3, the generalized Benders decomposition method has been used to study MINLP problem (VOP) and establish a corresponding algorithm for solving this problem (see Algorithm 1 in Section 4). With regards to the convergence of the algorithm, it is shown by Theorems 4.4 and 4.5 that the termination criterion in a finite number of steps follows with some mild assumptions. The algorithm extends the generalized Benders decomposition in the sense of solving MINLPs from problem $\mathcal{P}$ in finite dimension space to problem (VOP) in more general Banach space.\
[**Acknowledgement.**]{} The authors are indebted to two anonymous referees for their comments and suggestions which help us to improve our presentation and draw our attention to the work by Hooker on logic-based Benders decomposition.
[99]{}
J. F. Benders, Partitioning procedures for solving mixed-variable programming problems, Numer. Math., 4(1962), pp. 238-252.
P. Bonami, L. Biegler, A. R. Conn, G. Cornuéjols, I. E. Grossmann, C. Laird, J. Lee, A. Lodi, F. Margot, N. Sawaya and A. Wächter, An algorithmic framework for convex mixed integer nonlinear programs, Discrete. Optim., 5(2)(2008), pp. 186-204.
J. M. Borwein, On the existence of Pareto efficient points, Math. Oper. Res., 8 (1983), pp. 64-73.
S. Chandra and L. C. W. Dixon, Benders’ decomposition for the constraint $l_1$-problem, J. Optim. Theory. Appl., 68(1991), pp. 217-232.
V.-P. Eronen, M. M. Mäkelä and T. Westerlund, On the generalization of ECP and OA methods to nonsmooth convex MINLP problems, Optimization, 63(2014), pp. 1057-1073.
R. Fletcher and S. Leyffer, Solving mixed-integer nonlinear programs by outer approximation, Math. Program., 66(1994), pp. 327-349.
O. E. Flippo and A. H. G. Rinnoy Kan, Decomposition in general mathematical programming, Math. Program., 60(1993), pp. 361-382.
F. Flores-Bazan, Ideal, weakly efficient solutions for vector optimzation problems, Math. Program., 93 (2002), pp. 453-475.
C. A. Floudas and A. R. Ciric, Strategies for overcoming uncertainties in heat exchanger network synthesis. Comput. Chem. Eng., 13(1989), pp. 1133-1152.
A. M. Geoffrion, Duality in nonlinear programming: a simplified application-oriented development, SIAM Review, 13(1971), no. 1.
A. M. Geoffrion, Generalized Benders decomposition, J. Optim. Theory. Appl., 10(1972), no. 4, pp. 237-260.
I. E. Grossmann, Review of nonlinear mixed-integer and disjunctive programming techniques, Optim. Eng., 3(2002), pp. 227-252.
I. E. Grossmann and N. V. Sahinidis (eds), Special issue on mixed-integer programming and it Application to engineering, Part I, Optim. Eng., 3 (4), Kluwer Academic Publishers, Netherlands, 2002.
I. E. Grossmann and N. V. Sahinidis (eds), Special issue on mixed-integer programming and its Application to engineering, Part II, Optim. Eng., 4(1), Kluwer Academic Publishers, Netherlands, 2002.
C. Gutiérrez, B. Jimenez and V. Novo, A unified approach and optimality conditions for approximate solutions of vector optimization problems, SIAM J. Optim., 17 (2006), pp. 688-710.
J. N. Hooker, Logic-Based Methods for Optimization: Combining Optimization and Constraint Satisfaction. John Wiley and Sons, New York, 2000.
J. N. Hooker, Planning and scheduling by logic-based Benders decomposition, Oper. Res., 55(2007), pp. 588-602.
J. N. Hooker and G. Ottosson, Logic-based Benders decomposition, Math. Programram., 96(2003), pp. 33-60.
J. Jahn, Vector Optimization: Theory, Applications and Extensions, Springer-Verlag, New York, 2004.
S. Leyffer, Integrating SQP and branch-and-bound for mixed integer nonlinear programming, Comput. Optim. Appl., 18(2001), pp. 295-309.
D. T. Luc, Theory of Vector Optimization, Springer-Verlag, New York, 1989.
P. Michelon and N. Maculan, Lagrangean decomposition for integer nonlinear programming with linear constraints, Math. Program., 52(1991), pp. 303-313.
B. S. Mordukhovich, Variational Analysis and Generalized differentiation I/II, Springer-verlag, Berlin, Heidelberg, 2006.
I. Nowak and S. Vigerske, LaGO: a (heuristic) branch and cut algorithm for nonconvex MINLPs, Cent. Eur. J. Oper. Res, 16(2)(2008), pp.127-138.
R. R. Phelps, Convex functions, Monotone operators and Differentiability, Lecture Notes in Math. 1364, Springer, New York, 1989.
R. Rouhani, L. Lasdon, W. Lebow and A. D. Waren, A generalized Benders decomposition approach to reactive source planning in power systems, Math. Program. Studies., 25(1985), pp. 62-75.
W. Rudin, Functional Analysis, New York: McGraw-hill, 1973.
N. V. Sahinidis and I. E. Grossmann, Convergence properties of generalized Benders Decomposition, Comput. Chem. Eng., 15(1991), pp. 481-491.
M. Tawarmalani and N. V. Sahinidis, Global optimization of mixed-integer nonlinear programs: A theoretical and computational study, Math. Program., 99 (2004), pp. 563-591.
M. Tawarmalani and N. V. Sahinidis, Convexification and Global Optimization in Continuous and Mixed-Integer Nonlinear Programming: Theory, Algorithms, Software, and Applications, Kluwer Academic Publishers, 2002.
T. Westerlund and F. Pettersson, An extended cutting plane method for solving convex MINLP problems, Computer. Chem. Eng. 19(1995), pp. 131-136.
T. Westerlund and R. Pörn, Solving pseudo-convex mixed integer optimization problems by cutting plane techniques, Optim. Eng. 3(2002), pp. 253-280.
Z. Wei and M. M. Ali, Outer approximation algorithm for one class of convex mixed-integer nonlinear programming problems with partial differentiability, J. Optim. Theory. Appl., DOI 10.1007/s10957-015-0715-y.
Z. Wei and M. M. Ali, Convex mixed integer nonlinear programming problems and an outer approximation algorithm, J. Glob. Optim., DOI 10.1007/s10898-015-0284-5.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Employing a four-body classical model, various mechanisms responsible for the production of fast light particles in heavy ion collisions at low and intermediate energies have been studied. It has been shown that at energies lower than 50 A MeV, light particles of velocities of more than two times higher than the projectile velocities are produced due to the acceleration of the target light-particles by the mean field of the incident nucleus. It has also been shown that precision experimental reaction research in normal and inverse kinematics is likely to provide vital information about which mechanism is dominant in the production of fast light particles.'
address: '141980, Joint Institute for Nuclear Research, Dubna, Moscow region, Russia'
author:
- 'A.S. Denikin[^1], V.I. Zagrebaev'
date: 'August 21, 2002'
title: |
NEW MECHANISM FOR THE PRODUCTION OF THE EXTREMELY FAST LIGHT PARTICLES IN HEAVY-ION COLLISIONS\
IN THE FERMI ENERGY DOMAIN[^2]
---
Introduction
============
The production of pre-equilibrium light particles (n, p, t, $\alpha $) in nucleus-nucleus collisions depends on the way the nuclear system evolves at the reaction initial stage. For heavy ion collisions, the light particle cross section is known to be a noticeable fraction of the total reaction cross section even at low energies of the order of 10 A MeV, i.e. light particle production is a distinctive feature of all nuclear reactions involving heavy ions. This means that studying the production mechanisms for those particles is likely to provide direct information on both the reaction initial stage dynamics and the potential and dissipative forces of nucleus-nucleus interaction.
Vast bodies of data [@Awes; @Sackett; @Lanzano; @Penion; @Alba; @Sapienza] demonstrate that in heavy ion reactions at energies per nucleon of the order of the Fermi energy, light particles are produced in a wide angular range, their velocities being two and more times higher than the velocities of the beam particles. Fig. \[Fig1\] shows the measured differential proton cross section at $\theta _{lab} = 20^ \circ $ in the case of the ${}^{16}O$ (20 A MeV) + ${}^{197}Au$ collision [@Awes]. What is the mechanism of the production of these extremely fast light particles? The ultimate answer to that question has not been provided yet. Circumstantial evidence suggests that they are of a pre-equilibrium nature. This makes them more difficult to study by a direct experiment since there is currently no way of studying processes taking place in time intervals of the order of $10^{ - 20} $ s.
Attempts to interpret the experimental picture have resulted in the creation of a number of theoretical models and approaches. Among these in particular are the models of moving sources [@Awes2], Fermi-jet [@Bondorf; @Mohring; @Randrup], dissipative break-up and massive transfer [@Zagr] etc. A comprehensive survey of experimental and theoretical works relating to this problem is given in [@Zagr2]. Applying these approaches, the authors have succeeded in qualitatively describing the energy and angular dependencies of the spectra of emitted light particles as well as revealing some production mechanisms for them. Among the latter in particular is the mechanism of “splashing” nucleons out of a retarding projectile nucleus, the high velocities of the light particles being due to the addition of the velocity of Fermi motion within the projectile nucleus and the velocity of its centre of mass. The importance of taking this mechanism into account was shown in refs. [@Mohring; @Zagr], when describing the spectra of the $\alpha$-particles produced in projectile break-up and massive transfer at collision energies of the order of several tens of MeV per nucleon.
=7.5 cm
0.1 cm
Nucleon-nucleon collisions are also likely to result in a fast light particle being emitted due to scattering on nucleons of high velocities. The Fermian shape of nucleon momentum distributions explains the existence of such nucleons in a heated nuclear system. The first mechanism, capable of qualitatively describing the experimental results in a transparent way, fails to provide satisfactory quantitative agreement. The nucleon-nucleon collision model is adequate at high energies ($<$ 100 A MeV), but in the region of energies $E_0 <$ 50 A MeV the nucleon-nucleon interaction cross section for heavy ion collisions decreases due to Pauli’s principle. Therefore this mechanism influences the formation of the hard part of the spectrum of light particles to a smaller degree. Recent advances in instrumentation have made possible the precise measurement of the angular and energy spectra of light particles [@Penion; @Alba; @Sapienza], which is likely to shed light upon yet unresolved problems.
Mechanisms of Light-Particles Formation
=======================================
In ref. [@Denikin], we proposed a four-body classical model of nucleus-nucleus collisions, which permits establishing the role of mean nuclear fields and nucleon-nucleon collisions in the production of light-particles. This model considers the projectile and target nuclei to be bound two-body systems that are composed of a heavy core and a light fragment moving in its field. The interactions of the fragments, which follow classical trajectories, define six pair potentials. The interaction between the light fragments and the cores was described by the Woods-Saxon potential with compilation-based parameters [@Perey]. The interaction between the heavy cores was described by the proximity potential or the Woods-Saxon potential, the parameters of the latter being chosen so as to reproduce the height and position of the Coulomb barrier. The coupling with the reaction channels, which were not taken into account explicitly, was introduced by means of dissipative forces acting between the heavy cores of the target and projectile. The choice of friction coefficients and form-factors for the dissipative forces was based on ref. [@Gross]. To provide a more correct estimate of the channel differential cross sections, the description of the relative motion of the projectile (target) light particle and the target (projectile) core has an absorption probability for them has been introduced as $P_{ij}^{abs} = 1 - \exp \left( { -{{s_{ij}}
\mathord{\left/{\vphantom {{s_{ij} } {\lambda _{ij}}}} \right. \kern-\nulldelimiterspace}
{\lambda _{ij} }}} \right)$, where $s_{ij}$ is the distance traveled by particle i in the nucleus j, $\lambda _{ij} $ – the corresponding free path length, which is, as well known, related to the imaginary part of the optical potential, the parameters of which were chosen according to ref. [@Perey]. Specifying the relative distance vector ${\bf
r}_{ij} $ for the projectile components, the binding energy $E_{ij}^{sep} $ , known by experiment, as well as the orbital momentum $l_{ij} $ for their relative motion, completely defines the projectile inner spatial configuration. The components of the vector ${\bf r}_{ij} $ were chosen randomly and equiprobably in the classically allowed region. To have the relative motion momentum ${\bf p}_{ij} $ uniquely determined, one of the components of the vector ${\bf l}_{ij} $ must also be specified (by a random choice) in addition to the $r_{ij} $ and $E_{ij}^{sep} $ values. Repeating the same operations for the target nucleus and specifying the relative motion of the centers of mass of the nuclei depending on the reaction allow to determine the initial conditions that are required for solving the set of classical equations of motion.
=7.5 cm
0.1 cm
The model outlined above has 15 reaction channels with a different combination of particles in the final fragments. Eight of these channels contribute to the total cross section for pre-equilibrium light particles, which can be divided into two groups: particles emitted from the projectile and from the target. Taking into account all the pair potentials allows us to study the whole range of processes that result in light-particle emission as well as to establish the role played in them by one or another type of interaction. We have investigated heavy ion collisions at energies of several tens of MeV per nucleon and thoroughly studied the production mechanisms for fast light particles. It is remarkable and unexpected that the hardest part of the energy spectrum of emitted light particles corresponds to particles emitted from the target nucleus rather than from the projectile nucleus, as has been assumed up to now. There is a simple enough explanation for this phenomenon, which appears to be unusual at first sight. Let us assume for simplicity that the heavy cores have masses much larger than those of the light particles; the light particle moves in the nucleus at a velocity equal to the Fermi velocity $v_F $ ; the nuclei move relative to one another at the velocity $v_0 \approx
v_F $ ; the trajectories of the heavy fragments are taken to be undistorted; the binding energy of the light particle in the nucleus is assumed to be much smaller than the collision energy $E_{ij}^{sep} \ll E_0 = {{mv_0^2 } \mathord{\left/{\vphantom {{mv_0^2
} 2}} \right. \kern-\nulldelimiterspace} 2}$, here m is the mass of the light particle. Then, as has been shown in \[13\], for example, the neutrons emitted from the projectile (“splashing out”) will have a maximum laboratory system energy expressed as $$E_{n,\max }
\approx \frac{m}{2}\left( {v_0^2 + 2v_0 v_F } \right) = 3\frac{{mv_0^2 }}{2} = 3E_0.
\eqno{(1)}$$
In the framework of this model under the same assumptions, the maximum energy of the neutrons emitted both from the projectile and from the target in elastic nucleon-nucleon collisions is limited by the value $$E_{n,\max } \approx \frac{m}{2}\left( {v_0 + v_F
} \right)^2 = 4\frac{{mv_0^2 }}{2} = 4E_0. \eqno{(2)}$$ However, such a mechanism of emitting a fast nucleon suggests that the recoil nucleon is to impart the whole of its kinetic energy to the particle knocked out and, consequently, pass into an occupied lower energy state, which is forbidden by Pauli’s principle. Therefore, this mechanism does not manifest itself in full measure at low and intermediate collision energies and comes into play as the collision energy increases, when states more and more lower in energy become unoccupied in the process of the system getting excited.
The production mechanism for the fast light particles emitted form the target is more complicated. To begin with, let us consider the elastic scattering of a target light particle on a moving infinitely heavy repulsing wall. If before colliding, the light particle and the heavy wall have been moving collinearly in opposite directions with parallel velocities $v_F $ and $v_0 $ respectively, then the centre-of-mass velocity of the light particle has been $v_{cm} = - (v_0 + v_F )$ . After elastically colliding, the light particle will have the velocity $v'_{cm} = - v_{cm} = (v_0 + v_F )$ , the lab velocity being $v' = v'_{cm} + v_0 = 2v_0 + v_F $ . Thus the maximum velocity of a neutron emitted from the target will be $$E_{n,\max } \approx 4\frac{m}{2}\left( {v_0^2
+ v_0 v_F } \right) = 8\frac{{mv_0^2 }}{2} = 8E_0, \eqno{(3)}$$ i.e. the fastest neutrons will have a velocity of more than 2.5 times higher than that of the beam particles. Now, suffice it to say that an elastic collision with a repulsing wall is kinematically equivalent to elastic scattering at the angle $\theta _{cm} = - 180^ \circ$ (i.e. orbiting) in the attractive field of an incident nucleus. However, orbiting is possible at quite low relative motion energies. At higher energies, a light particle can be deflected only due to the field of the projectile at a certain centre-of-mass angle $\theta _R $ called an angle of rainbow scattering. Therefore the maximum energy of a pre-equilibrium light particle emitted will be a function of collision energy, light-particle-incident-nucleus interaction potential, particle binding energy and dissipative forces.
More accurate calculations of the maximum energy of the pre-equilibrium neutrons produced in the reaction ${}^{20}Ne$ + ${}^{165}Ho \to $ n + X are presented in Fig. \[Fig2\]. Given here is the $E_{n,\max } /E_0 $ ratio as a function of the beam energy $E_0 $ for the above mechanisms: the solid curve is for the energy of the neutrons emitted from the target; the dashed curve for the energy of the neutrons emitted from the projectile; the dash-dotted curve for the energy of neutrons produced in nucleon-nucleon collisions, which was calculated with regard to Pauli’s principle. So, it is seen that at collision energies per nucleon smaller than $E_F $ (indicated with an arrow in Fig. \[Fig2\]), the mechanism involving acceleration of neutrons by the moving mean field of the projectile is dominant in the formation of the high-energy portion of neutron spectra. In the high-energy region, the key role is played by nucleon- nucleon collisions. It should be also noted that the curves in Fig. \[Fig2\] do not represent $E_{n,\max }$ in a strict way since our calculations did not take into account the non-zero probability of there existing nucleons of velocities much higher than $v_F $. Therefore the boundaries indicated with the curves in Fig. \[Fig2\] will shift to higher energies.
=7.5 cm
0.1 cm
How should the predictions, made by our model be tested, and how should the yields of pre-equilibrium light particles emitted from a projectile and target be separated? There is currently no direct solution. However the above analysis of various processes of light-particle production in heavy ion collisions shows that the mechanism of emitting fast nucleons from the target is only slightly sensitive to the projectile nucleus mass, and, consequently, the maximum energy of light particles is not likely to change greatly with replacing one projectile by another. On the other hand, the maximum energy of the target nucleons accelerated by the attractive field of the incident nucleus is defined by the corresponding rainbow scattering angle, which depends on the dimensions of the deflecting mean field [@Knoll] $$\theta _R \approx {\left(V_C - 0.56 U_0
\sqrt{{R_U}\left/{a_U}\right.} \right)} \left/ {E_{cm}}\right., \eqno{(4)}$$ where $V_C $ is the height of the Coulomb barrier for the emitted light particle; $U_0 $, $R_U $, $a_U
$ – the depth, radius and diffusiveness of the light-particle-projectile interaction potential, and $R_U \approx 1.3A^{{1 \mathord{\left/ {\vphantom {1 3}} \right.
\kern-\nulldelimiterspace} 3}} $. It should be noted, the coefficient 0.56 in empirical equation (4), obtained in ref. [@Knoll], is not valid in case of nucleon scattering on heavy nucleus. We found that coefficient 0.7 enables to obtain more reasonable value of nucleon rainbow angle $\theta_R$.
Now it is evident that the energy distribution of emitted light particles must be more elongated for a reaction with a heavy projectile than for a reaction with a light one. This conclusion is supported by Fig. \[Fig3\], in which the calculated differential pre-equilibrium proton cross sections are shown for the angle $\theta _{lab} = 51^ \circ
$ for the ${}^{40}Ar$ + ${}^{51}V$ and ${}^{132}Xe$ + ${}^{51}V$ reactions (in normal and inverse kinematics) at the energy $E_0$ = 44 A MeV. It is seen, first, that the key role in the formation of the high-energy part of the spectrum is played by the protons emitted from the target nuclei (curves 1 and 2 in Fig \[Fig3\]). Second, the distribution of the protons emitted from projectiles of different masses (curves 3 and 4 in Fig. \[Fig3\]) only changes its shape to a large degree at small energies, i.e. a more massive ${}^{132}Xe$ projectile loses less energy in a collision with a ${}^{51}V$ target than does ${}^{40}Ar$ . Therefore these two reactions have projectile proton distributions with shifted maxima. Third, the ${}^{40}Ar$ induced reaction in normal kinematics has a pre-equilibrium-proton spectrum, which is several tens of MeV shorter than that for the reaction with the $^{132}Xe$ heavier projectile. So comparing precisely measured data on the distributions of the light particles produced in heavy ion collisions in reactions of normal and inverse kinematics will give us vital information about which mechanism is dominant in emitting a fast light particle.
T.C. Awes [*et al.*]{}, Phys. Rev. [**C**]{} 25 (1982) 2361. D. Sackett [*et al.*]{}, Phys. Rev. [**C**]{} 44 (1991) 384. G. Lanzano [*et al.*]{}, Phys. Rev. [**C**]{} 58 (1998) 281. Yu. E. Penionzhkevich [*et al.*]{}, Preprint JINR [E7-98-282]{} (1998). R. Alba [*et al.*]{}, Phys. Lett. [**B**]{} 322 (1994) 38. P. Sapienza [*et al.*]{}, Nucl. Phys. [**A**]{} 630 (1998) 215c. T.C. Awes [*et al.*]{}, Phys. Rev. [**C**]{} 24 (1981) 89. J.P. Bondorf [*et al.*]{}, Nucl. Phys. [**A**]{} 333 (1980) 285. K. Möhring, W.J. Swiatecki and M. Zielinska-Pfabe, Nucl. Phys. [**A**]{} 440 (1985) 89. J. Randrup and R. Vanderbosch, Nucl. Phys. [**A**]{} 474 (1987) 219. V.I. Zagrebaev, Ann. Phys. (N.Y.) 197 (1990) 33. V. Zagrebaev and Yu. Penionzhkevich, Prog. Part. Nucl. Phys. 35 (1995) 575. A.S. Denikin, V.I. Zagrebaev, Yad. Fiz. (Rus. Jour. Nucl. Phys.) 65 (2002) 1459. C.M. Perey and F.G. Perey, At. Data and Nucl. Data Tables 13 (1974) 293. D.H.E. Gross and H. Kallinowski, Phys. Rep. 45 (1978) 175. J. Knoll and R. Schaeffer, Ann. Phys. (N.Y.) 97 (1976) 307.
[^1]: E-mail: [email protected]
[^2]: Proceeding of VII International School-Seminar on Heavy-Ion Physics, May 27 – June 1, 2002, Dubna, Russia.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The one-dimensional XXZ model is studied in the presence of disorder in the Heisenberg Exchange Integral. Recent predictions obtained from renormalization group calculations are investigated numerically using a Lanczos algorithm on chains of up to 18 sites. It is found that in the presence of strong X-Y-symmetric random exchange couplings, a “random singlet” phase with quasi-long-range order in the spin-spin correlations persists. As the planar anisotropy is varied, the full zero-temperature phase diagram is obtained and compared with predictions of Doty and Fisher \[Phys. Rev. B [**45** ]{}, 2167 (1992)\].'
address: |
Department of Physics and Supercomputer Computations Research Institute,\
Florida State University, Tallahassee, Florida 32306
author:
- 'Stephan Haas, Jose Riera[@byline], and Elbio Dagotto'
title: |
Random Exchange Disorder\
in the Spin-1/2 XXZ Chain
---
\#1[[$\backslash$\#1]{}]{}
The study of quantum models in the presence of disorder is an emerging field onto which much attention has been focussed lately. Since all experimentally accessible systems [@wu] are to some extent affected by randomness in the form of impurities, fields, or couplings, a thorough understanding of disorder effects can help in comparing experimental observations and theoretical predictions. In particular, weakly disordered, low-dimensional quantum spin systems are of interest, since the interplay between randomness and strong quantum fluctuations can be observed.[@oldwork] At T=0, phase transitions in quantum spin models are driven by zero-point fluctuations, as opposed to thermal excitations in their classical counterparts. However, when a random potential is introduced, phase transitions can be also driven by random fluctuations. This mechanism is particularly interesting in the case of marginally ordered systems, where the long-range Néel order in the 2D isotropic Heisenberg model has been found to be unstable towards thermal fluctuations and random fields, but not towards randomness in the exchange couplings.[@murthy]
The anisotropic spin-1/2 Antiferromagnetic Heisenberg Chain is a generic model of strongly correlated electrons. It is described by the Hamiltonian, $$H_0 = J \sum_i (\lambda S^z_i S^z_{i+1} + S^x_i S^x_{i+1} + S^y_i S^y_{i+1}),$$ where the notation is standard. Due to the low dimensionality, quantum fluctuations destroy long-range order in the region $-1 < \lambda \leq 1$, and the spin-spin correlations decay spatially with a power-law. Beyond the Heisenberg point (i.e. $\lambda > 1$), a gap opens in the excitation spectrum and the system develops long-range Néel order with exponentially decaying correlation functions, while for $\lambda \leq -1$ there is a ferromagnetic region with Ising-type long-range order.
Let us now introduce disorder in the form of X-Y symmetric random exchange couplings, i.e. such that the planar symmetry of $H_0$ is not broken by the random potential, $$H_{random} = \sum_i \delta_i (S^x_i S^x_{i+1} + S^y_i S^y_{i+1}).$$ The random couplings $\delta_i$ are drawn from a uniform distribution $P(\delta_i) = \theta (\delta_i - \delta J_{xy})\theta (\delta J_{xy} -
\delta_i)$, where $\langle \delta_i \rangle = 0$ and $\langle (\delta_i)^2 \rangle = 2( \delta J_{xy} )^3/3$. The cut-off parameter $\delta J_{xy}$ serves as a measure for the strength of the random potential. The physical properties induced by this distribution are believed to be universal. However, in order to test this idea, we also studied random exchange couplings drawn from a Gaussian distribution, $P(\delta_i) = \frac{1}{\sqrt{2 \pi \sigma_{xy}}} \exp{(- \delta_i^2 / 2
\sigma_{xy})}$. Here, $\sigma_{xy}$ serves as a measure of the random strength.
The properties of XXZ chains in the presence of various random potentials have recently been studied by C. A. Doty and D. S. Fisher using renormalization group techniques.[@doty] It was found that, while random transverse fields destroy the (quasi)-long-range spin order, a power-law decay of the spin correlations may persist in the presence of random exchange couplings as long as the random Hamiltonian does not break the planar symmetry of $H_0$. In particular, it was predicted that a quasi-long-range-ordered phase extends from the X-Y regime ($-1 < \lambda \leq 1$), when $H_{random}$ is switched on.
In our study of the above system, we numerically diagonalized chains of up to 18 sites with periodic boundary conditions using a Lanczos algorithm. The observables were obtained from a quenched average, i.e. the ground state $|\phi_0 (j) \rangle$ of a chain was obtained for a given set of random couplings $j = \{ \delta_i \}$, and then the expectation value of some particular operators $\hat{O}$ were studied. This procedure was repeated for m $\simeq 500$ different sets of random couplings, and finally the algebraic average over all m random samples was taken. The quenched average of an operator $\hat{O}$ is thus defined by $$\langle \langle \hat{O} \rangle \rangle = \frac{1}{m} \sum_{j=1}^m \langle
\phi_0 (j)
|\hat{O} |\phi_0 (j) \rangle.$$
First, we would like to address the question of whether quasi-long-range order persists in the region $-1 < \lambda \leq 1$ when the disorder potential $H_{random}$ is switched on. The relevant observable is the normalized real-space spin-spin correlation function $$\omega^z (l) = \frac{3}{N} \sum_{i=1}^N \frac{\langle \langle S^z_i
S^z_{i+l} \rangle \rangle}
{S(S+1)},$$ where N denotes the number of sites, and $S=1/2$ in our study.
In Fig.1, the spin-spin correlations $\omega^z (N/2)$ at the maximum separation ($l=N/2$) are plotted as a function of the lattice size N at planar anisotropy $\lambda = 0.5$ for a couple of random strengths $\delta J_{xy}$. If the correlations decay with a power-law $|\omega^z(l) |
\propto l^{- \eta_z}$, we expect a straight line with negative slope $\eta_z$ in a double-logarithmic plot. It is found that for all random strengths, $\delta J_{xy}$, a power-law decay (solid line) fits the numerical data much better than an exponential decay (dashed line), e.g. the $\chi^2$-value obtained from least-square fits is typically two orders of magnitude larger when an exponential decay $|\omega^z(l) |
\propto \exp{(- \xi l)}$ is assumed. We observed a similar power-law behavior in a large region of parameter space.
Why does the random potential not destroy quasi-long-range order in this region? According to Doty and Fisher the “random singlet" phase which extends from the X-Y phase of the pure system ($H_0$) can be pictured in terms of randomly distributed tightly coupled singlet pairs of spins.[@doty] Those spins which are not bound in a singlet pair interact via virtual excitations. It turns out that these “almost-free" spins are anomalously strongly correlated. The probability that “almost-free" spins separated by a distance $R$ interact strongly is proportional to $1/R^2$. This gives rise to the observed power-law behavior in the spin-spin correlations. The decay exponent is found to be $\eta_z =2$. Note that also, in the exactly solvable X-Y limit ($\lambda = 0$) the system maps into a tight-binding model of free fermions with random nearest-neighbor hopping. In this limit the decay exponent is given by $\eta_z =2$ if a single characteristic localization length is assumed for the properties of the low-energy wave functions.[@doty]
In Fig.2, we show $\eta_z$ obtained in our numerical analysis, as a function of the disorder parameter $\delta J_{xy}$ for various anisotropies $\lambda$. The exponent has been extracted using chains of size N=6, 10, 14 and 18. [@scalinglaw] The inset of Fig.2(a) shows the decay exponent $\eta_z$ for the pure system $H_0$ as it has also been obtained in Ref. 6. The exact diagonalization results are in excellent agreement with predictions from conformal invariance,[@alcaraz] and in particular the Heisenberg limit ($|\omega^z(l) | \propto l^{-1}$) and the X-Y limit ($|\omega^z(l) |\propto l^{-2}$) are nicely recovered. For negative anisotropies, ($-1 < \lambda \leq 0$) conformal invariance predicts a constant exponent $\eta_z =2$, which is also in reasonable agreement with our data, showing that our techniques can reproduce known results very accurately.
On our finite chains and as we depart from the $\delta J_{xy} =0$ limit, three regions can be identified:
\(1) In the regime of small randomness ($\delta J_{xy} < J$) the exponent $\eta_z $ increases slightly as a function of the disorder parameter $\delta J_{xy} $, which is a sign of reduced order.
\(2) Around $\delta J_{xy} = J$, there is an area of high competition between the quantum fluctuations of the original Hamiltonian ($J \sum_i (S^x_i S^x_{i+1} + S^y_i S^y_{i+1}$) and $H_{random}$. Locally the random terms can compensate the zero-point fluctuations leading to an antiferromagnetic Ising-like behavior in the correlation functions. As a result, the decay exponent $\eta_z$ has a dent with onset at around $\delta J_{xy} = J$, indicating a crossover into a more ordered Ising-like regime, where correlations decay more slowly than for the uniform system.
\(3) For large disorder, ($\delta J_{xy} >> J$) $H_{random}$ is the dominant term. The dependence of the decay exponent on the planar anisotropy in $H_0$ becomes negligible, and it approaches $\eta_z =2$ for all values of $\lambda$, as it has been predicted by renormalization group arguments.[@doty]
In the vicinity of $\lambda = -0.75$ the exponent $\eta_z$ behaves anomalously for small disorder. The observed decay in $\eta_z$ for $\delta J_{xy}$ between $J$ and $2J$ is due to ferromagnetic behavior in the real space spin-spin correlations. This anomaly is observed specially for anisotropies $-1 < \lambda \leq -0.5$. The dent of $\eta_z$ around $\delta J_{xy} = J$ can be understood as a crossover into a phase of higher order. In particular, for $\lambda = -0.75$ we observed a transition into a partially polarized phase indicated by the change of sign in the energy difference $\delta E = E(S^z_{tot} = 0) - E(S^z_{tot} = 1)$, where $E(S^z_{tot} = n
)$ is the quenched ground state energy in the subspace with $S^z_{tot} = n$. The inset of Fig.2(b) shows $\delta E$ as a function of the disorder parameter $\delta J_{xy}$ at anisotropy $\lambda = -0.75$ for a 14-site chain. It can be nicely seen that the transition into the partially polarized phase ($0.55J \leq \delta J_{xy} \leq 3.05J$) corresponds to the dent in $\eta_z$ in the same regime of disorder.
In Fig.3(a), the dependence of the energy on the disorder parameters $\delta J_{xy}$ and $\sigma_{xy}$ at various anisotropies is shown for a 14-site chain. As the random potential becomes dominant, the system is allowed to relax into a ground state of higher entropy. The ground state energy drops proportionally to $\delta J_{xy}$ ($\sigma_{xy}$) in this region. In Fig.3(b), we show how the static structure factor ($S^{zz}(k) = \sum_j \exp (-i k j) \langle 0
| S^{z}_j S^{z}_{j+1} | 0 \rangle$) behaves as a function of the disorder parameters at antiferromagnetic momentum transfer $k = \pi$ for the 14-site chain. In analogy to Fig.2, three regions can be identified. At low disorder the structure factor remains approximately unchanged. In the region of competition, Néel order is favored for positive anisotropies ($0 \leq \lambda \leq 1$), resulting in an increase of the antiferromagnetic structure factor especially in the vicinity of $\lambda \sim 1$. For negative anisotropies ($-1 < \lambda \leq 0$), the ditch in $S^{zz}(\pi)$ indicates a crossover into a ferromagnetically polarized region. For large disorder, $S^{zz}(\pi)$ becomes independent of $\lambda$, and approaches the X-Y limit for all anisotropies.
The boundary between the long-range-ordered regime and the “random singlet" phase is obtained from the correlations $\omega^z(N/2)$. In the “random singlet" phase, the spin-spin correlations at distance N vanish in the bulk limit as ${\bf N \rightarrow \infty}$. However, as the anisotropy is tuned across the critical value $\lambda_c$, $\omega^z(N/2)$ becomes finite, approaching $| \omega^z(N/2) | =1 $ in the extreme Ising limit ($\lambda =
\infty$). At zero disorder the Heisenberg point $\lambda_c = 1$ is nicely recovered as the critical point (Fig.3(c)). In Fig.3(d), we see that the transition point between these two phases is $reduced$ to about $\lambda_c = 0.75$ at $\delta J_{xy} = J$. [@intermediate] As a result of the strong competition effects in the region $\delta J_{xy} \simeq J$, the antiferromagnetic phase bends into the random singlet regime in a $``reentrant"$ transition, indicating a stronger antiferromagnetic order in this region. The whole boundary between “random singlet" and Néel phase is plotted in the phase diagram given in Fig.4.
Both the “random singlet" and the Néel phase lie in the $S^z_{total} =0$ subspace. On the other hand, as the ferromagnetic limit is approached, there is a transition into a partially polarized phase, i.e. the ground state no longer has $S^z_{total} =0$. This phase boundary, as well as the transition from the partially into the fully polarized regime, is extracted from comparing the lowest energies of the various $S^z_{total}$ subspaces (averaged over the ensemble of random couplings). In the region of competition between quantum fluctuations and the disorder term, the partially polarized phase bends into the “random singlet" regime, in analogy to the effect at the phase boundary between the “random singlet" and the Néel phase, as shown in Fig.4.
For low disorder, our results agree qualitatively with those of Doty and Fisher.[@doty] However, their study predicts an X-Y-like “mole hill" phase in the region $-1 < \lambda \leq -0.5$, and for small disorder. Numerically, it is hard to distinguish this “mole-hill" from the “random singlet" regime, because both phases show power-law behavior in the correlation functions. However, from our exact diagonalization data we have observed a region (denoted with a question mark in Fig.4) which has power-law decay, and is a member of the $S^z_{total} =0$ subspace, but does not have any remnant antiferromagnetic correlations, as has been discussed above in the inset of Fig.2(b). We are currently investigating, whether this regime can be identified with the “mole hill" predicted by Doty and Fisher.
In summary, we have presented the first numerical study of the spin-1/2 XXZ chain in the presence of a random exchange potential ($H_{random}$). In contrast to a random field,[@zimanyi] quasi-long-range order of the zero-disorder X-Y regime $-1 < \lambda \leq 1$ is not destroyed by an X-Y symmetric random exchange. Also, Ising-type long-range order persists in the presence of small random exchange couplings. The power-law behavior in the “random singlet" phase may be due to virtual interactions of “almost-free" spins which are not bound in randomly strong singlet pairs. A complete phase diagram is provided. In addition, we have found an interesting reentrant transition of the ordered phases (in both the ferromagnetic and antiferromagnetic Heisenberg limits) when exchange disorder is included. Such a novel type of behavior (order induced by random couplings) deserves additional study.
We thank A. Moreo, F. Alcaraz, T. Barnes, G. Zimanyi, and K. Runge for useful discussions. J.R. has been supported in part by the U. S. Department of Energy (DOE) Office of Scientific Computing under the High Performance Computing and Communications Program (HPCC), and in part by DOE under contract No. DE-AC05-84OR21400 managed by Martin Marietta Energy Systems, Inc., and under contract No. DE-FG05-87ER40376 with Vanderbilt University. The work of E. D. is partially supported by the ONR grant N00014-93-1-0495, and by the donors of The Petroleum Research Fund. The support of the Supercomputer Computations Research Institute (SCRI) is acknowledged. The computer calculations were carried out on the CRAY-YMP at Florida State University.
Present address : Physics Division and Center for Computationally Intensive Physics, Oak Ridge National Laboratory, Oak Ridge, TN 37831-6373.
See, e.g., W. Wu et al., Phys. Rev. Lett. [**67**]{}, 2076 (1991); S. Nagler, et al., J. Phys. C : Solid State Phys. [**17**]{}, 4819 (1984); G. Theodorou, Phys. Rev. B [**16**]{}, 2273 (1977). G. Theodorou, Phys. Rev. B [**16**]{}, 2264 (1977); W.G. Clark and L.C. Tippie, Phys. Rev. B [**20**]{}, 2914 (1979); S. Ma, et al., Phys. Rev. Lett. [**43**]{}, 1434 (1979); S.R. Bondeson and Z.G. Soos, Phys. Rev. B [**22**]{}, 1793 (1980); C. Das Gupta and S. Ma, Phys. Rev. B [**22**]{}, 1305 (1980); J.E. Hirsch, Phys. Rev. B [**22**]{}, 5355 (1980); J.E. Hirsch and J.V. José, ibid., 5339 (1980); H.B. Schüttler, et al., Phys. Rev. B [**35**]{}, 3461 (1987); K. Runge, Phys. Rev. B [**45**]{}, 13136 (1992). G. Murthy, Phys. Rev. B [**38**]{}, 5162 (1988). C.A. Doty and D.S. Fisher, Phys. Rev. B [**45**]{}, 2167 (1992). The exponent $\eta_z$ was extracted from finite size data for the correlation function at maximum separation $\omega^z (N/2)$ using $\eta_z (N/2) \sim
\frac{\ln (A/\omega^z (N/2))}{\ln (N/2)}$ at large $N$. A. Moreo, Phys. Rev. B [**36**]{}, 8582 (1987); F.C. Alcaraz and A. Moreo, Phys. Rev. B [**46**]{}, 2896 (1992). Since the $\omega^z (N/2)$ were measured only at certain anisotropies ($\lambda = 0.0,0.5,1.0, ...$), the critical anisotropies $\lambda_c$ were roughly obtained from a linear extrapolation using $\omega^z (\infty )$ for $\lambda$ values closely above the point where the spin gap opens. K. Runge and G. Zimanyi (preprint).
|
{
"pile_set_name": "ArXiv"
}
|
0.3cm
[CERN-TH.6721/92]{}\
[ACT-22/92]{}\
[CTP-TAMU-75/92]{}\
[UMN-TH-1117/92]{}\
[**FLIPPED HEAVY NEUTRINOS:**]{}\
[**FROM THE SOLAR NEUTRINO PROBLEM TO BARYOGENESIS**]{}\
\
\
[*CH - 1211 Geneva 23*]{}\
and\
\
[*Center for Theoretical Physics, Department of Physics, Texas A&M University*]{}\
[*College Station, TX 77843-4242, USA*]{}\
[*and*]{}\
[*Astroparticle Physics Group, Houston Advanced Research Center (HARC)*]{}\
[*The Woodlands, TX 77382, USA*]{}\
and\
\
\
[*Minneapolis, MN 55455, USA*]{}\
\
We discuss baryogenesis using the flipped $SU(5)$ model for lepton mass matrices. We show that the generalized see-saw mechanism in this model can not only provide MSW neutrino mixing suitable for solving the solar neutrino problem, and supply a hot dark matter candidate ($\nu_\tau$) with mass $0(10)eV$ as indicated by recent COBE results, but can also naturally account for the baryon asymmetry of the universe. Heavy singlet neutrino decay generates a net lepton asymmetry which is subsequently reprocessed by nonperturbative electroweak interactions. We evaluate the baryon asymmetry so produced in light of the constraints that the COBE observations put on inflationary cosmologies, finding it comfortably consistent with observation.
CERN-TH.6721/92\
[ACT-22/92]{}\
[CTP-TAMU-75/92]{}\
UMN-TH-1117/92\
November 1992
[ [**Acknowledgements**]{} ]{}\
The work of DVN was supported in part by DOE grant DE-FG05-91-ER-40633 and by a grant from Conoco Inc. The work of KAO was supported in part by DOE grant DE-AC02-83ER-40105, and by a Presidential Young Investigator Award.
[99]{} G. ’t Hooft, Phys. Rev. Lett. [**37**]{}(1976) 8; V. Kuzmin, V. Rubakov and M. Shaposhnikov,[*Phys.Lett.*]{} [**B155**]{} (1985) 36 P. Arnold and L. McLerran, [*Phys.Rev.*]{} [**D36**]{} (1987) 581, [*Phys.Rev.*]{} [**D37**]{} (1988) 1020. M.E. Shaposhnikov, Nucl. Phys. [**B287**]{}(1987)757; [**B299**]{}(1988)797; A.I. Bochkarev, V. Kuzmin and M.E. Shaposhnikov, Phys. Lett. [**B244**]{}(1990)275; A.I. Bochkarev, S.Yu. Khlebnikov and M.E. Shaposhnikov, Nucl. Phys. [**B329**]{}(1990)490; L. McLerran, Phys. Rev. Lett. [**62**]{}(1989)1075; N. Turok and J. Zadrozny, Phys. Rev. Lett. [**65**]{}(1990)2331; Nucl. Phys. [**B358**]{}(1991)471; L. McLerran, M.E. Shaposhnikov, N. Turok and M. Voloshin, Phys. Lett. [**B256**]{}(1991)251; A. Cohen, D. Kaplan and A. Nelson, Nucl. Phys. [**B349**]{}(1991)727; Phys. Lett. [**B263**]{}(1991)86; G.W. Anderson and L.J. Hall, Phys. Rev. [**D45**]{}(1992)2685; M. Dine, P. Huet, R.S. Singleton Jr. and L. Susskind, Phys. Lett. [**B257**]{}(1991)351;M. Dine, P. Huet and R.S. Singleton Jr., Nucl. Phys. [**B375**]{}(1992)625; M. Dine, R.G. Leigh, P. Huet, A. Linde and D. Linde, SLAC preprints PUB-5740 and PUB-5741 (1992). J.N. Bahcall and R.N. Ulrich, Rev. Mod. Phys. [**60**]{}(1988) 297. S.D.M. White, C.S. Frenk and M. Davis, Ap.J. [**274**]{}(1983) 61. V.A. Kuzmin, V.A. Rubakov and M.E. Shaposhnikov, Phys. Lett. [**B191**]{} (1987) 171; A.E. Nelson and S.M. Barr, Phys. Lett. [**B246**]{} (1990)141; H. Dreiner and G.G. Ross, Oxford preprint OUTP-92-08P (1992). B. Campbell, S. Davidson, J. Ellis and K. A. Olive, CERN preprint TH.6642/92 (1992), Phys. Lett. B in press (1992). GALLEX Collaboration, P. Anselmann et al.,Phys. Lett. [**B285**]{} (1992) 390. F.L. Wright et al., Ap. J. Lett. [**396**]{} (1992) L13. S. Barr, Phys. Lett. [**B112**]{} (1982) 219; Phys. Rev. [**D40**]{} (1989) 2457; J.-P. Derendinger, J. Kim and D.V. Nanopoulos, Phys. Lett. [B139]{}(1984)170. I. Antoniadis, J. Ellis, J. Hagelin and D.V. Nanopoulos, Phys. Lett. [**B194**]{}(1987) 231. J. Ellis, J.L. Lopez and D.V. Nanopoulos, Phys. Lett. [**B292**]{} (1 189. L.Wolfenstein, Phys. Rev. [**D17**]{} (1978) 2369; [**D20**]{} (1979) 2634; S.P. Mikheyev and A. Yu. Smirnov, Yad. Fiz. [**42**]{} (1985) 14 Nuov. Cim. [**9C**]{} (1986) 17. M. Fukugita and T. Yanagida, [*Phys.Lett.*]{} [**B174**]{} (1986) 45. I. Antoniadis, J. Ellis, J. Hagelin and D.V. Nanopoulos, Phys. Lett. [**B208**]{}(1988) 209. G. Leontaris, Phys. Lett. [**B207**]{} (1988) 447; G. Leontaris and D.V. Nanopoulos, Phys. Lett. [**B212**]{} (1988) 327; G. Leontaris and K. Tamvakis, Phys. Lett. [**B224**]{} (1989) 319; S. Abel, Phys. Lett. [**B234**]{} (1990) 113; I. Antoniadis, J. Rizos and K. Tamvakis, Phys. Lett. [**B279**]{} (1992) 281. J.L. Lopez and D.V. Nanopoulos, Phys. Lett. [**B268**]{} (1991) 359. R. Barbieri, D.V. Nanopoulos and A. Masiero, Phys. Lett. [**B98**]{} (1981) 191; A. Masiero, D.V. Nanopoulos, K. Tamvakis and T. Yanagida, Z. Phys. [**C17**]{} (1983) 33. B. Campbell, J. Ellis, J. Hagelin, D.V. Nanopoulos and K.A. Oliv Phys. Lett. [**B200**]{} (1988) 483; J. Ellis, J.Hagelin, D.V. Nanopoulos, and K.A. Olive, Phys. Lett. [**B207**]{}(1988) 451. B. Campbell, S. Davi and K. A. Olive, University of Minnesota preprint UMN-TH-1114/92 (1992). B. Campbell, J. Ellis, J.Hagelin, D.V. Nanopoulos and K.A. Olive Phys. Lett. [**B197**]{} (1987) 355. D.V. Nanopoulos and S. Weinberg, Phys. Rev. [**D20**]{} (1979) 2484. K. Yamamoto, Phys. Lett. [**B168**]{} (1986) 341; K. Enqvist, D.V. Nanopoulos and M. Quiros, Phys. Lett. [**B169**]{} (1986) 343. J. Ellis, K. Enqvist, D.V. Nanopoulos and K.A. Olive, Phys. Lett. [**B188**]{} (1987) 415. S. Weinberg, Phys. Rev. Lett. [**42**]{} (1979) 850; D. Toussaint, S. B. Treiman, F. Wilczek and A. Zee, Phys. Rev. [**D19**]{} (1979) 1036;J. Ellis, M.K. Gaillard and D.V. Nanopoulos, Phys. Lett. [**B80**]{} (1979) 3 M. Fukugita and F. Yanagida, Phys.Rev. [**D42**]{} (1990) 1285; J.A. Harvey and M.S. Turner, Phys.Rev. [**D42**]{} (1990) 3344; B.A. Campbell, S. Davidson, J. Ellis and K.A. Olive, Phys.Lett. [**B256**]{} (1991 457; CERN preprint TH.6208/91 (1991), Astropart.Phys. [**1**]{} (1992) (in press) Fischler, G.F. Giudice, R.G. Leigh and S. Paban, Phys.Lett. [**B258**]{} (1991) 45 J. Ellis, K. Enqvist, D.V. Nanopoulos and K.A. Olive, Phys. Lett. [**B191**]{}(1987) 343. J. Ellis, G.L. Fogli and E. Lisi, CERN preprint TH.6568/92 (1992). R.H. Dalitz and G.R. Goldstein, Phys. Lett. [**B287**]{} (1992) 225. T. Walker, D.N. Schramm, G. Steigman, K.A. Olive and K. Kang, Ap. [**376**]{} (1991) 51.
[**Figure caption**]{} 0.5cm The tree and one-Higgs-loop diagrams whose interference gives the largest contribution to the lepton asymmetry in heavy singlet neutrino decay in flipped SU(5). 6.5cm 1.5 cm
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Flexion is a non-linear gravitational lensing effect that arises from gradients in the convergence and shear across an image. We derive a formalism that describes non-linear gravitational lensing by a circularly symmetric lens in the thin-lens approximation. This provides us with relatively simple expressions for first- and second-flexion in terms of only the surface density and projected mass distribution of the lens. We give details of exact lens models, in particular providing flexion calculations for a Sérsic-law profile, which has become increasingly popular over recent years. We further provide a single resource for the analytic forms of convergence, shear, first- and second-flexion for the following mass distributions: a point mass, singular isothermal sphere (SIS); Navarro-Frenk-White (NFW) profile; Sérsic-law profile. We quantitatively compare these mass distributions and show that the convergence and first-flexion are better indicators of the Sérsic shape parameter, while for the concentration of NFW profiles the shear and second-flexion terms are preferred.'
author:
- |
P. D. Lasky$^{1}$[^1] and C. J. Fluke$^{1}$[^2]\
$^{1}$Centre for Astrophysics & Supercomputing, Swinburne University of Technology, PO Box 218, Hawthorn, Victoria, 3122, Australia
bibliography:
- 'Flexion.bib'
title: 'Shape, shear & flexion: An analytic flexion formalism for realistic mass profiles [^3]'
---
\[firstpage\]
gravitational lensing – galaxies: halos – dark matter
Introduction
============
Quantifying the exact shape of baryonic and dark matter density profiles that form in expanding $\Lambda$CDM cosmologies is an ongoing issue. N-body simulations suggest that CDM halos are well fitted by either Navarro-Frenk-White (NFW) density profiles [@navarro97] or deprojected [*Sérsic-like*]{} laws in the form of [@einasto65] or [@prugniel97] density profiles [@navarro04; @merritt05; @merritt06; @graham06; @graham06a]. Numerous observations of early-type galaxies suggest their luminosity profiles, and hence stellar mass distributions, follow either Sérsic, core-Sérsic or Nuker-law models [e.g. @lauer95; @lauer05; @graham03a; @ferrarese06a; @cote06 and references therein], while gravitational lensing observations have suggested the [*total*]{} mass distribution (i.e. baryons plus dark matter) is consistently described by isothermal spheres [@treu02; @treu04; @rusin03; @rusin05; @koopmans06; @gavazzi07; @czoske08; @dye08; @tu09]. Meanwhile, debate about the mass distributions of galaxy clusters has been no less intense; only recently is the NFW profile [e.g. @carlberg97; @vandermarel00; @athreya02; @katgert04; @lin04; @hansen05; @lokas06; @rines06; @wojtak07; @okabe09] being favoured over the isothermal sphere [e.g. @athreya02; @ettori02; @katgert04]. As gravitational lensing traces total projected mass, it is an extremely powerful tool for determining the mass distributions of these systems without having to make assumptions about the dynamics or constitution of the lensing objects [for recent reviews see @schneider05; @hoekstra08].
Traditionally, the study of weak lensing has been limited to linear effects; convergence and shear. These fields have the effect that an elliptically shaped source galaxy gets mapped to an elliptical image. Therefore, to determine information about the lensing object using first-order quantities, assumptions must be made about the intrinsic ellipticity of the source galaxy, or a large number of source galaxies must be utilised to ensure statistically reasonable results can be inferred [see for e.g. @hoekstra04; @mandelbaum06]. Recently however, various authors have begun to consider higher-order lensing effects known as flexion [@goldberg02; @goldberg05; @bacon06][^4]. Flexion comes in two flavours that correspond to various spatial derivatives of the shear and convergence, implying they are due to gradients in the first-order fields across the extent of the source/image. Physically, one can think of first-flexion as a shift in the centroid of the image with respect to the source and second-flexion as creating an arc-like structure in the image [@bacon06]. That is, with the inclusion of flexion, an elliptical source galaxy gets mapped to a “jelly-bean” shaped image for a circularly symmetric lens. In this way, flexion provides a better observable than the first-order fields as only one reasonable assumption about the source galaxy is required – i.e. galaxies are not intrinsically flexed[^5].
The flexion of a lensed image is formally calculated using multipole moments [@goldberg02; @goldberg07; @okura07; @okura08], however in this work we choose to treat the gravitational lensing variables as field variables. That is, we determine the amount of convergence, shear and flexion one would measure as a function of the distance from the centre of the lensing mass and the angle in the sky, ignoring the overall shape and size of the source. Determining the change in shape between the source/image pair requires a more detailed mapping that calculates small changes in the position of numerous light rays in the image and source planes. This is a somewhat more difficult task that requires numerical methods, which is beyond the scope of the present work. However, treating the gravitational lensing terms as field variables is extremely useful and the benefits it purveys are three-fold:
1. It allows us to determine differences in the global gravitational lensing properties from various realistic density distributions (in particular see figure \[allallMpc\] below).
2. We can determine the relevant lensing terms (convergence, shear, first- or second-flexion) for discerning between individual shapes of density profiles. For example in section \[comparesec\] we show that convergence and first-flexion are good indicators of the Sérsic shape parameter, whereas the concentration of NFW profiles can be determined by looking at the shear and second-flexion.
3. The derivation of analytic solutions is a critical first step towards studying flexion through multiple lens planes with arbitrary mass distributions.
This paper is set out as follows; In section \[secone\] we systematically develop the two-dimensional thin-lens gravitational lens equation for an arbitrary, circularly symmetric matter distribution, deriving the first-order terms in section \[firstordersec\] and the flexion terms in section \[flexionsec\]. In section \[exactsolnssec\] we consider exact forms of the matter distributions, writing down analytic expressions for the convergence, shear and flexion for a point mass, singular isothermal sphere (SIS), NFW and Sérsic profiles in sections \[Schwsec\], \[SISsec\], \[NFWsec\] and \[Sersicsec\] respectively. In this way we are providing a single resource where the analytic forms for the convergence, shear, first- and second-flexion can be found for a range of useful density profiles. Finally, in section \[comparesec\] we compare the gravitational lensing effects of each of these profiles. We find that the shear and second-flexion for a Sérsic-law profile are systematically greater than for NFW and SIS profiles, whereas the convergence and first-flexion of each of the profiles are comparable. We further show that the convergence and first-flexion provide excellent tracers for the Sérsic shape parameter, whereas the shear and second-flexion are better indicators of the concentration parameter for the NFW profile. We make some concluding remarks in section \[conc\].
Analytic Lensing Formalism {#secone}
==========================
The Thin-Lens Equation {#thinlenssec}
----------------------
Given that flexion considers finite source sizes, we require a two-dimensional version of the thin-lens gravitational lens equation, whereby the mapping between a point on the source plane and the image plane are explicitly expressed. In this way the thin-lens gravitational lens equation is expressed as $$\eta_{i}=\frac{D_{S}}{D_{L}}\xi_{i}-D_{LS}\tilde{\alpha}_{i},\label{lens}$$ where $\xi_{i}$ is the impact parameter on the image plane, $\eta_{i}$ is the distance between the origin of the coordinate system and the source on the source plane, $\tilde{\alpha}_{i}$ is the deflection angle and $D_{S}$, $D_{L}$ and $D_{LS}$ are the angular diameter distances from the observer to the source, the observer to the lens and the lens to the source respectively (the lens configuration is shown in figure \[Diag\]). Equation (\[lens\]) can be put into a neater form by using angular coordinates, $\beta_{i}=\eta_{i}/D_{S}$ and $\theta_{i}=\xi_{i}/D_{L}$, such that $$\beta_{i}=\theta_{i}-\alpha_{i},\label{scalelens}$$ where $\alpha_{i}=\tilde{\alpha}_{i}D_{LS}/D_{S}$ is the scaled deflection angle.
Equation (\[scalelens\]) can be re-expressed as a coordinate map between the two angular coordinate systems allowing the mapping to be expressed as a linear transformation $$\beta_{i}=A_{ij}\theta_{j},\label{linear}$$ where $A_{ij}:={\partial}\beta_{i}/{\partial}\theta_{j}$ is the Jacobian of transformation and summation is assumed over repeated indices. Equation (\[linear\]) implicitly assumes there are no gradients in the components of $A_{ij}$ across the image. Physically, this implies that the convergence and shear are constant across the image, which is not necessarily a good approximation when the source has a finite spatial extent. This is a reasonable assumption if sources are assumed to be point-like, however for realistic cosmology one would like to be able to probe deviations of the shear and convergence across images. [@goldberg05] therefore generalized equation (\[linear\]) to include higher-order terms, coined [*flexion*]{}. These flexion terms account for an image’s “arciness”, and are therefore relevant if one is considering sources of finite spatial extent. The non-linear expansion of the thin-lens equation is $$\beta_{i}=A_{ij}\theta_{j}+\frac{1}{2}D_{ijk}\theta_{j}\theta_{k},\label{secondorder}$$ where $$D_{ijk}:=\frac{{\partial}A_{ij}}{{\partial}\theta_{k}},$$ is the gradient of the Jacobian. The components of the $D_{ijk}$ matrix make up the components of the first- and second-flexion, a point that we discuss in considerably more detail in section \[flexionsec\].
We are interested in explicit expressions for the first- and second-order lensing terms with respect to positions on the image and source planes. It is therefore instructive for us to work in the Cartesian $(\eta_{i},\xi_{i})$ system of coordinates rather than the angular $(\beta_{i},\theta_{i})$ coordinates (see figure \[Diag\]). In these coordinates, the Jacobian and its gradient are expressed as $$A_{ij}=\frac{D_{L}}{D_{S}}\frac{{\partial}\eta_{i}}{{\partial}\xi_{j}}\qquad{\rm and}\qquad D_{ijk}=\frac{D_{L}^{2}}{D_{S}}\frac{{\partial}^{2}\eta_{i}}{{\partial}\xi_{j}{\partial}\xi_{k}}.\label{6}$$
In this article, we restrict our attention to distributions of matter which are circularly symmetric when projected on to the lens plane. For such a profile, one can express the deflection angle as [see for e.g. @schneider06] $$\tilde{\alpha}_{i}=\frac{4G}{c^{2}}\frac{M(|{\xi}|)}{|\xi|^{2}}\xi_{i},\label{alpha}$$ where $|\xi|=\sqrt{\xi_{1}^{2}+\xi_{2}^{2}}$ and the projected mass, $M$, is defined as the area integral of the surface density, $\Sigma(|\xi|)$, which for circular symmetry is given by $$M(|\xi|)=2\pi\int_{0}^{|\xi|}\Sigma(\xi')\xi'{\rm d}\xi'.\label{massdef}$$ Equation (\[lens\]) can now be expressed in component form, also substituting equation (\[alpha\]), yielding $$\eta_{i}=\xi_{i}\frac{D_{S}}{D_{L}}\left[1-\frac{1}{\pi\Sigma_{cr}}\frac{M(|\xi|)}{|\xi|^{2}}\right],\label{eta}$$ where the critical surface density has been defined according to $$\Sigma_{cr}=\frac{c^{2}}{4\pi G}\frac{D_{S}}{D_{L}D_{LS}}.$$
First-Order Lensing {#firstordersec}
-------------------
The first-order Jacobian, $A_{ij}$, can now be calculated by differentiating equation (\[eta\]) which, after some algebra is $$\begin{aligned}
A_{11}&=&1-\frac{2\Sigma(|\xi|)\xi_{1}^{2}}{\Sigma_{cr}|\xi|^{2}}+\frac{M(|\xi|)}{\pi\Sigma_{cr}|\xi|^{4}}\left(\xi_{1}^{2}-\xi_{2}^{2}\right),\label{A11}\\
A_{22}&=&1-\frac{2\Sigma(|\xi|)\xi_{2}^{2}}{\Sigma_{cr}|\xi|^{2}}+\frac{M(|\xi|)}{\pi\Sigma_{cr}|\xi|^{4}}\left(\xi_{2}^{2}-\xi_{1}^{2}\right),\label{A22}\\
A_{12}&=&\frac{-2\xi_{1}\xi_{2}}{\pi\Sigma_{cr}|\xi|^{4}}{\cal Q}(|\xi|)=A_{21},\label{A12}\end{aligned}$$ where for convenience we have defined the function $${\cal Q}(\zeta):=\pi\Sigma(\zeta)\zeta^{2}-M(\zeta).$$
The convergence and two components of the shear are expressed in terms of the components of the Jacobian as follows $$\begin{aligned}
\gamma_{1}&=&-(A_{11}-A_{22})/2,\\
\gamma_{2}&=&-A_{12},\\
\kappa&=&1-(A_{11}+A_{22})/2.\label{kappa2}\end{aligned}$$ Evaluating the convergence by substituting equations (\[A11\]) and (\[A22\]) into (\[kappa2\]) implies $$\kappa=\frac{\Sigma(|\xi|)}{\Sigma_{cr}},\label{kappa}$$ which is the familiar version of this equation [see for e.g. @schneider06]. Evaluating the two components of the shear using equations (\[A11\]-\[A12\]) implies $$\begin{aligned}
\gamma_{1}&=&\frac{{\cal Q}(|\xi|)}{\pi\Sigma_{cr}|\xi|^{4}}\left(\xi_{1}^{2}-\xi_{2}^{2}\right),\label{gamma1}\\
\gamma_{2}&=&\frac{2{\cal Q}(|\xi|)}{\pi\Sigma_{cr}|\xi|^{4}}\xi_{1}\xi_{2}.\label{gamma2}\end{aligned}$$ To compare the equations for the shear with a more familiar form, we convert the above equations from Cartesian coordinates, $(\xi_{1},\xi_{2})$, into polar coordinates on the lens plane, $(R,\phi)$, defined by $\xi_{1}=R\cos\phi$ and $\xi_{2}=R\sin\phi$. This implies that the convergence and total shear, $\gamma=\gamma_{1}+i\gamma_{2}$, are $$\begin{aligned}
\kappa&=&\frac{\Sigma(R)}{\Sigma_{cr}},\label{kappaR}\\
\gamma&=&\frac{{\cal Q}(R)}{\pi\Sigma_{cr}R^{2}}{\rm e}^{2i\phi}=|\gamma|{\rm exp}\left(2i\phi\right).\label{gammaR}\end{aligned}$$ This is consistent with showing that the shear is a spin-two quantity, whereas the convergence is spin-zero.
Equations (\[kappaR\]) and (\[gammaR\]) give the convergence and shear distributions as a function of the mass profile of the galaxy. Whilst the convergence is only dependent on the surface density, the shear is also a function of the mass. This immediately gives the expected result for a point mass lens that the convergence is everywhere zero (except at the origin), whereas the shear is non-zero and goes as $M/R^{2}$ (see section \[Schwsec\]).
Flexion {#flexionsec}
-------
The real interest in this article lies in the flexion terms, which are given by the second-order Taylor series expansion, equation (\[secondorder\]). The three-tensor, $D_{ijk}$, is expressed in terms of the gradient of the linear Jacobian of transformation in equation (\[6\]) that, given equations (\[A11\]-\[A12\]), allows us to calculate the individual components of the three-tensor in terms of the coordinates $\xi_{1}$ and $\xi_{2}$. After much algebra, one finds $$\begin{aligned}
D_{111}&=&\frac{2D_{L}\xi_{1}\left(\xi_{1}^{2}-3\xi_{2}^{2}\right)}{\pi\Sigma_{cr}|\xi|^{6}}{\cal Q}-\frac{2D_{L}\xi_{1}^{3}}{\Sigma_{cr}|\xi|^{3}}\frac{d\Sigma}{d|\xi|},\label{D111}\\
D_{211}&=&\frac{2D_{L}\xi_{2}\left(3\xi_{1}^{2}-\xi_{2}^{2}\right)}{\pi\Sigma_{cr}|\xi|^{6}}{\cal Q}-\frac{2D_{L}\xi_{1}^{2}\xi_{2}}{\Sigma_{cr}|\xi|^{3}}\frac{d\Sigma}{d|\xi|},\label{D211}\\
D_{221}&=&\frac{-2D_{L}\xi_{1}\left(\xi_{1}^{2}-3\xi_{2}^{2}\right)}{\pi\Sigma_{cr}|\xi|^{6}}{\cal Q}-\frac{2D_{L}\xi_{1}\xi_{2}^{2}}{\Sigma_{cr}|\xi|^{3}}\frac{d\Sigma}{d|\xi|},\label{D221}\\
D_{222}&=&\frac{-2D_{L}\xi_{2}\left(3\xi_{1}^{2}-\xi_{2}^{2}\right)}{\pi\Sigma_{cr}|\xi|^{6}}{\cal Q}-\frac{2D_{L}\xi_{2}^{3}}{\Sigma_{cr}|\xi|^{3}}\frac{d\Sigma}{d|\xi|},\label{D222}\end{aligned}$$ where for the remainder of the article $\Sigma=\Sigma(|\xi|)$, ${\cal Q}={\cal Q}(|\xi|)$ and $M=M(|\xi|)$ unless otherwise explicitly stated. As shown by [@bacon06], the three-tensor, $D_{ijk}$, can be expressed as the sum of two other tensors, $D_{ijk}={\mathcal{F}}_{ijk}+{\mathcal{G}}_{ijk}$, which can be written component-wise as $$\begin{aligned}
{\mathcal{F}}_{ij1}&=&-\frac{1}{2}\left(\begin{array}{cc}
3{\mathcal{F}}_{1} & {\mathcal{F}}_{2} \\
{\mathcal{F}}_{2} & {\mathcal{F}}_{1}\end{array}\right),\\
{\mathcal{F}}_{ij2}&=&-\frac{1}{2}\left(\begin{array}{cc}
{\mathcal{F}}_{2} & {\mathcal{F}}_{1} \\
{\mathcal{F}}_{1} & 3{\mathcal{F}}_{2}\end{array}\right),\\
{\mathcal{G}}_{ij1}&=&-\frac{1}{2}\left(\begin{array}{cc}
{\mathcal{G}}_{1} & {\mathcal{G}}_{2} \\
{\mathcal{G}}_{2} & -{\mathcal{G}}_{1}\end{array}\right),\\
{\mathcal{G}}_{ij2}&=&-\frac{1}{2}\left(\begin{array}{cc}
{\mathcal{G}}_{2} & -{\mathcal{G}}_{1} \\
-{\mathcal{G}}_{1} & -{\mathcal{G}}_{2}\end{array}\right).\end{aligned}$$ In the above, ${\mathcal{F}}={\mathcal{F}}_{1}+i{\mathcal{F}}_{2}$ and ${\mathcal{G}}={\mathcal{G}}_{1}+i{\mathcal{G}}_{2}$ are known as first- and second-flexion respectively [@bacon06]. First-flexion is a spin-one quantity that measures the shift in the centroid of the image, and second-flexion is a spin-three quantity measuring the “arciness” of the image.
Inverting the above system of equations implies we can express the components of the first- and second-flexion in terms of the components of $D_{ijk}$; $$\begin{aligned}
{\mathcal{F}}_{1}&=&-\frac{1}{2}(D_{111}+D_{221}),\label{F1}\\
{\mathcal{F}}_{2}&=&-\frac{1}{2}(D_{211}+D_{222}),\label{F2}\\
{\mathcal{G}}_{1}&=&-\frac{1}{2}(D_{111}-3D_{221}),\label{G1}\\
{\mathcal{G}}_{2}&=&-\frac{1}{2}(3D_{211}-D_{222}).\label{G2}\end{aligned}$$ Therefore, utilizing equations (\[D111\]-\[D222\]), (\[F1\]) and (\[F2\]), one can show that the components of the first-flexion are $$\begin{aligned}
{\mathcal{F}}_{1}&
=&\frac{D_{L}}{\Sigma_{cr}}\frac{{\partial}\Sigma}{{\partial}\xi_{1}},\label{F1gen}\\
{\mathcal{F}}_{2}&
=&\frac{D_{L}}{\Sigma_{cr}}\frac{{\partial}\Sigma}{{\partial}\xi_{2}}.\label{F2gen}\end{aligned}$$ That is, the first and second components of first-flexion (i.e., ${\mathcal{F}}_{1}$ and ${\mathcal{F}}_{2}$ respectively) are the directional derivatives of the surface density. Note that the first-flexion terms do not include any functions of the mass, which is again consistent with the flexion due to a point mass being zero (see section \[Schwsec\]). The total first-flexion, ${\mathcal{F}}={\mathcal{F}}_{1}+i{\mathcal{F}}_{2}$, is given by $$\begin{aligned}
{\mathcal{F}}&=&\frac{D_{L}}{\Sigma_{cr}|\xi|}\frac{d\Sigma}{d|\xi|}\left(\xi_{1}+i\xi_{2}\right){\nonumber}\\
&=&\frac{D_{L}}{\Sigma_{cr}}\frac{d\Sigma(R)}{dR}{\rm e}^{i\phi},\label{1flexionR}\end{aligned}$$ implying the first-flexion is a spin-one field [@bacon06] which is the gradient of the surface density. Given that the convergence is proportional to the surface density, this is an equivalent way of saying that the first-flexion is the gradient of the convergence.
Following the same procedure outlined above, one can show that the components of the second-flexion are $$\begin{aligned}
{\mathcal{G}}_{1}&=&\frac{D_{L}\xi_{1}\left(\xi_{1}^{2}-3\xi_{2}^{2}\right)}{\pi\Sigma_{cr}|\xi|^{6}}\left[\pi\frac{d\Sigma}{d|\xi|}|\xi|^{3}-4{\cal Q}\right],\\
{\mathcal{G}}_{2}&=&\frac{D_{L}\xi_{2}\left(3\xi_{1}^{2}-\xi_{2}^{2}\right)}{\pi\Sigma_{cr}|\xi|^{6}}\left[\pi\frac{d\Sigma}{d|\xi|}|\xi|^{3}-4{\cal Q}\right].\end{aligned}$$ These form the total second-flexion term, ${\mathcal{G}}={\mathcal{G}}_{1}+i{\mathcal{G}}_{2}$, as $${\mathcal{G}}=\frac{D_{L}}{\pi\Sigma_{cr}|\xi|^{6}}\left[\pi\frac{d\Sigma}{d|\xi|}|\xi|^{3}-4{\cal Q}\right]\left(\xi_{1}+i\xi_{2}\right)^{3}.$$ Finally, expressing the second-flexion in polar coordinates and expanding ${\cal Q}$ to show the explicit dependence on the surface mass density and the projected mass one finds $${\mathcal{G}}=\frac{D_{L}}{\Sigma_{cr}}\left[\frac{d\Sigma(R)}{dR}-\frac{4\Sigma(R)}{R}+\frac{4M(R)}{\pi R^{3}}\right]{\rm e}^{3i\phi}.\label{2flexionR}$$ This expression is relatively simple and shows the dependence of second-flexion on the surface density, its divergence and the projected mass distribution. Moreover, one can explicitly see that this is a spin-three field, as shown in [@bacon06].
It is pertinent to note that [@schneider08] showed that first- and second-flexion are not observable quantities due to the mass-sheet degeneracy. Instead, they showed that the observable spin-one and spin-three fields are the [*reduced flexion*]{} terms; $$\frac{{\mathcal{F}}+g{\mathcal{F}}^{*}}{1-\kappa}\qquad{\rm and}\qquad\frac{{\mathcal{G}}+g{\mathcal{F}}}{1-\kappa},\label{42}$$ where $g=\gamma/(1-\kappa)$ is the reduced shear and ${\mathcal{F}}^{*}={\mathcal{F}}_{1}-i{\mathcal{F}}_{2}$ is the complex conjugate of first-flexion. Whilst it is beneficial to keep the mass-sheet degeneracy in mind, a majority of work on measuring flexion in real images using either the shapelets [@refregier03a; @refregier03b; @goldberg05] or HOLICs [@okura07; @okura08; @goldberg07] approaches have been based on decomposing the higher-order components of the images into ${\mathcal{F}}$ and ${\mathcal{G}}$. Therefore, for this article we continue to focus on the flexion quantities, ${\mathcal{F}}$ and ${\mathcal{G}}$, rather than the terms expressed in (\[42\]).
A Note On Units
---------------
The system of coordinates we are using in this article are somewhat unconventional in the sense that they use distance, rather than angular coordinates on the image and source planes. We do this as we believe these units are more conducive to numerical applications, and are also more descriptive to the reader. However, our approach implies it is worth spending some time elucidating the dimensions that we are using for each of the derived quantities.
The surface density is defined as the projection of the three-dimensional density distribution, implying it has units of mass per unit area. Integrating this according to equation (\[massdef\]) implies the two-dimensional (projected) mass distribution has units of mass (as one would expect). Therefore, the convergence and shear given by equations (\[kappa\]), (\[gamma1\]) and (\[gamma2\]) are dimensionless. This is not surprising, and indeed this is true when one uses angular coordinates. The flexion terms presented here are in dimensions of (angle)$^{-1}$, which is also consistent with those expressed in angular coordinates, however the angle is in units of radians. This can be seen as the expressions for the first- and second-flexion, equations (\[1flexionR\]) and (\[2flexionR\]) respectively, are proportional to the ratio of the angular diameter distance from the observer to the lens plane, $D_{L}$, and the impact parameter, $|\xi|$. As discussed in section \[thinlenssec\], this ratio is $\theta^{-1}$. This implies that to compare the results presented in this paper with those derived using angular coordinates [for example @bacon06], one must convert from units of radians to arcseconds. As an alternative, in appendix \[anglesapp\] we show the equations for convergence, shear and flexion expressed in angular coordinates.
Exact Solutions {#exactsolnssec}
===============
In this section we present analytic solutions of the equations expressed hitherto for various circularly symmetric matter distributions. For completeness, we first discuss the simplest case of a point mass, then move on to SIS, NFW and finally Sérsic-law profiles. These exact solutions are then used in section \[comparesec\] to investigate differences between the flexion of various lenses, and we discuss how these flexion terms can be used to constrain the density distributions.
Point Mass {#Schwsec}
----------
Consider a Schwarzschild lens (i.e. a point mass) situated at the origin of the coordinate system on the lens plane. This implies that $\Sigma=\delta(|\xi|)$, where $\delta$ is the Dirac delta function. Moreover, the projected mass distribution $M=M_{s}$ is a constant for all $|\xi|\neq0$. The Einstein radius for such a system in angular coordinates is $$\theta_{E}=\left(\frac{4GM_{s}}{c^{2}}\frac{D_{LS}}{D_{S}D_{L}}\right)^{1/2},$$ and the critical surface density and Einstein radius are related according to $$\Sigma_{cr}=\frac{M_{s}}{\pi D_{L}^{2}\theta_{E}^{2}}=\frac{M_{s}}{\pi\xi_{E}^{2}},$$ where we have defined $\xi_{E}:=D_{L}\theta_{E}$ to be the Einstein radius measured as a distance on the lens plane.
The convergence is linearly proportional to the surface density of the system, equation (\[kappa\]), implying it is trivially zero everywhere except $|\xi|=0$. The shear, however, includes terms involving the projected mass of the system. From equations (\[gamma1\]) and (\[gamma2\]) the first and second components of the shear are $$\gamma_{1}=\frac{\xi_{E}^{2}}{|\xi|^{4}}\left(\xi_{2}^{2}-\xi_{1}^{2}\right)\qquad{\rm and}\qquad\gamma_{2}=\frac{-2\xi_{E}^{2}}{|\xi|^{4}}\xi_{1}\xi_{2},$$ and the total shear is $$\gamma=\frac{-\xi_{E}^{2}}{|\xi|^{4}}\left(\xi_{1}+i\xi_{2}\right)^{2}=\frac{-\theta_{E}^{2}}{\theta^{2}}{\rm e}^{2i\phi}.\label{schgamma}$$ Note that the final equation in (\[schgamma\]) is the usual expression for the shear associated with a point mass, however with an additional negative sign. This is a direct result of the calculation, however we note that the negative sign is due to the choice of coordinates. That is, rotating our coordinate system by ninety degrees implies the negative sign vanishes \[i.e. $\phi\rightarrow\phi+\pi/2$ implies $\exp(2i\phi)\rightarrow-\exp(2i\phi)$\]. Therefore, one is free to scale away the negative sign in equation (\[schgamma\]) by rotating the coordinate system, which yields the familiar result for the shear induced by a point mass lens.
According to equations (\[F1gen\]) and (\[F2gen\]), the components of the first-flexion are the directional derivatives of the surface density, implying these vanish (for all $|\xi|\neq0$); $${\mathcal{F}}_{1}={\mathcal{F}}_{2}={\mathcal{F}}=0.$$ As mentioned above, first-flexion is associated with a shift of the centroid of the image with respect to the source. Therefore, ${\mathcal{F}}\equiv0$ for a point mass implies the centroid of the image is unchanged. Second-flexion, however, is associated with the “arciness” of the image, and one finds for the point mass lens that $$\begin{aligned}
{\mathcal{G}}_{1}&=&\frac{4D_{L}\xi_{E}^{2}}{|\xi|^{6}}\xi_{1}\left(\xi_{1}^{2}-3\xi_{2}^{2}\right),\\
{\mathcal{G}}_{2}&=&\frac{4D_{L}\xi_{E}^{2}}{|\xi|^{6}}\xi_{2}\left(3\xi_{1}^{2}-\xi_{2}^{2}\right).\end{aligned}$$ Combining these, the total second-flexion is given by $${\mathcal{G}}=\frac{4D_{L}\xi_{E}^{2}}{|\xi|^{6}}\left(\xi_{1}+i\xi_{2}\right)^{3}=\frac{4\theta_{E}^{2}}{\theta^{3}}{\rm e}^{3i\phi}.$$ Taking the magnitude of the above expression implies second-flexion decreases proportionally to the projected radius cubed, $|{\mathcal{G}}|\propto|\xi|^{-3}$, compared to the shear which decreases proportionally to the projected radius squared, $|\gamma|\propto|\xi|^{-2}$. Therefore, as one gets further from the source, the shear term will come to dominate over the second-flexion, and this effect will become more negligible as one looks further from the origin (see figure \[allallMpc\]).
It is interesting to note that while the point mass has vanishing first-flexion and non-zero second-flexion, it is relatively straightforward to also construct lens models where the second-flexion vanishes but the first-flexion is non-zero. Indeed by setting equation (\[2flexionR\]) equal to zero, one can show that the second-flexion vanishes for models with $\Sigma\propto |\xi|^{2}$. Whilst this is obviously unphysical, as the surface density distribution increases monotonically as a function of radius, one still finds that the first-flexion is non-zero everywhere (for all $|\xi|\neq0$) with $|{\mathcal{F}}|\propto|\xi|$.
Singular Isothermal Sphere {#SISsec}
--------------------------
Several observational studies of early-type galaxies using gravitational lensing have suggested that the total matter distribution (i.e. baryonic plus dark matter) is well described by a nearly isothermal density profile [@treu02; @treu04; @rusin03; @rusin05; @koopmans06; @czoske08; @dye08; @tu09]. Indeed the strongest evidence for this is from [@gavazzi07] who reported weak lensing studies of 22 early-type galaxies based on $HST$ imaging, concluding that an isothermal distribution is consistent out to 100 effective radii. On larger scales, isothermal spheres have been found to fit the density profile of galaxy clusters [@athreya02; @ettori02; @katgert04], although recent debate has suggested the NFW profile may provide a better fit (see section \[NFWsec\]). Analytically, gravitational lensing studies have compared SIS and NFW profiles at first- [@wright00] and higher-order [@bacon06], implying the SIS profile provides us with the simplest realistic profile to reproduce known results.
The SIS density profile is described in terms of the three-dimensional (deprojected) radius, $r$, as $$\rho(r)=\frac{\sigma^{2}}{2\pi Gr^{2}},\label{SISrho}$$ where $\sigma$ is the one-dimensional velocity dispersion. Projecting equation (\[SISrho\]) along the line-of-sight gives the surface density [for example see @binney87] $$\Sigma=\frac{\sigma^{2}}{2G|\xi|}.$$ We note that the SIS profile, like the point mass, is singular at the origin. This implies that the following analysis is valid for all $|\xi|\neq0$, however in reality one must quantify at what radius the system is a “weak lens” (see section \[comparesec\]). By defining the Einstein radius in units of distance as $\xi_{E}:=D_{L}\theta_{E}$, where $\theta_{E}$ is the standard Einstein radius for an SIS lens, one can show that the critical surface density, $\Sigma_{cr}$, is related to the Einstein radius as $$\Sigma_{cr}=\frac{\sigma^{2}}{G\xi_{E}}.$$ By substituting the above equations into (\[kappa\]), we can show that the convergence falls-off linearly with the distance from the center of the coordinate system; $$\kappa=\frac{\xi_{E}}{2|\xi|}=\frac{\theta_{E}}{2\theta}.$$ Moreover, the components of the shear have a similar behaviour with the extra angular dependence $$\begin{aligned}
\gamma_{1}&=&\frac{-\xi_{E}}{2|\xi|^{3}}\left(\xi_{1}^{2}-\xi_{2}^{2}\right),\\
\gamma_{2}&=&\frac{-\xi_{E}}{|\xi|^{3}}\xi_{1}\xi_{2}.\end{aligned}$$ As we know the shear is a spin-two field, we are essentially only interested in the strength of the respective fields. Therefore, for the remainder of the article, unless explicitly stated, we shall just be working with the magnitudes of these fields. Now, as $\gamma=|\gamma|\exp(2i\phi)$, one can show from the above that $$|\gamma|=\frac{\xi_{E}}{2|\xi|}=\frac{\theta_{E}}{2\theta}.$$ The first-flexion components are $${\mathcal{F}}_{1}=\frac{-D_{L}\xi_{E}}{2|\xi|^{3}}\xi_{1}\qquad{\rm and}\qquad
{\mathcal{F}}_{2}=\frac{-D_{L}\xi_{E}}{2|\xi|^{3}}\xi_{2}.\label{SISfirsttwo}$$ First-flexion is a spin-one field, implying ${\mathcal{F}}=|{\mathcal{F}}|\exp(i\phi)$. Equations (\[SISfirsttwo\]) imply the first-flexion falls-off proportionally to the distance squared; $$|{\mathcal{F}}|=\frac{-D_{L}\xi_{E}}{2|\xi|^{2}}=\frac{-\theta_{E}}{2\theta^{2}}.$$ The second-flexion components are $$\begin{aligned}
{\mathcal{G}}_{1}&=&\frac{3D_{L}\xi_{E}}{2|\xi|^{5}}\xi_{1}\left(\xi_{1}^{2}-3\xi_{2}^{2}\right),\\
{\mathcal{G}}_{2}&=&\frac{3D_{L}\xi_{E}}{2|\xi|^{5}}\xi_{2}\left(3\xi_{1}^{2}-\xi_{2}^{2}\right),\end{aligned}$$ which is a spin-three field, ${\mathcal{G}}=|{\mathcal{G}}|\exp(3i\phi)$, implying second-flexion also decreases proportionally to the radius squared $$|{\mathcal{G}}|=\frac{3D_{L}\xi_{E}}{2|\xi|^{2}}=\frac{3\theta_{E}}{2\theta^{2}}.$$ All of the above equations that are written in angular coordinates are consistent with those presented in [@bacon06]. These analytic expressions are plotted for a specific mass in figure \[allallMpc\] where they are compared with the NFW and Sérsic-law profiles.
Navarro-Frenk-White Profile {#NFWsec}
---------------------------
Compared to the SIS profile, the NFW profile introduces an extra parameter into the scaling of density distributions. The concentration, $c$, which is defined as the ratio of the three-dimensional virial radius to the three-dimensional scale radius, $c=r_{\Delta}/r_{s}$, is a function of the particular cosmology being used. In section \[comparesec\] we compare the NFW profile to the other profiles being analysed here, as well as comparing the effects of varying the concentration on the first- and higher-order lensing phenomena. The NFW profile is generally given in terms of the three-dimensional radii, $r$, $$\rho(r)=\frac{\delta_{c}\rho_{c}}{\left(r/r_{s}\right)\left(1+r/r_{s}\right)^{2}},$$ where $\rho_{c}$ is the critical density of the Universe and $$\delta_{c}=\frac{\Delta}{3}\frac{c^{3}}{\ln\left(1+c\right)-c/\left(1+c\right)}.$$ Projecting this onto the two-dimensional radius, $|\xi|$, gives the surface density [@bartelmann96] $$\Sigma=\frac{2\rho_{c}\delta_{c}r_{s}^{3}}{|\xi|^{2}-r_{s}^{2}}\left[1-\Xi(|\xi|)\right],\label{surfaceNFW}$$ where we have defined the following function $$\Xi(|\xi|):=\left\{
\begin{array}{lc}
\frac{2r_{s}}{\sqrt{r_{s}^{2}-|\xi|^{2}}}{\rm arctanh}\sqrt{\frac{r_{s}-|\xi|}{r_{s}+|\xi|}} &|\xi|<r_{s}\\
\frac{2r_{s}}{\sqrt{|\xi|^{2}-r_{s}^{2}}}{\rm arctan}\sqrt{\frac{|\xi|-r_{s}}{|\xi|+r_{s}}} & |\xi|>r_{s}
\end{array}\right..\label{fdef}$$ Integrating the surface density gives the projected mass distribution $$M=4\pi\rho_{c}\delta_{c}r_{s}^{3}\left[\ln\frac{|\xi|}{2r_{s}}+\Xi(|\xi|)\right].$$
The convergence and shear can now be expressed simply as functions of the above expressions. From equation (\[kappaR\]), the convergence is simply $\kappa=\Sigma/\Sigma_{cr}$, where the surface density is given by equation (\[surfaceNFW\]). The total shear is determined by equation (\[gammaR\]) as $$|\gamma|=\frac{2\rho_{c}\delta_{c}r_{s}^{3}}{\Sigma_{cr}\left(|\xi|^{2}-r_{s}^{2}\right)}\left[1-\Xi-2\left(1-\frac{r_{s}^{2}}{|\xi|^{2}}\right)\left(\ln\frac{|\xi|}{2r_{s}}+\Xi\right)\right],$$ where $\Xi=\Xi(|\xi|)$. Equation (\[1flexionR\]) implies first-flexion is found by differentiating the surface density, which can be shown to be $$|{\mathcal{F}}|=\frac{-2D_{L}\rho_{c}\delta_{c}r_{s}^{3}}{\Sigma_{cr}|\xi|\left(|\xi|^{2}-r_{s}^{2}\right)^{2}}\left(2|\xi|^{2}+r_{s}^{2}-3|\xi|^{2}\Xi\right),$$ and equation (\[2flexionR\]) implies the second-flexion for the NFW profile is given by the expression $$\begin{aligned}
|{\mathcal{G}}|&=&\frac{2D_{L}\rho_{c}\delta_{c}r_{s}^{3}}{\Sigma_{cr}|\xi|\left(|\xi|^{2}-r_{s}^{2}\right)^{2
}}\Bigg[8\left(1-\frac{r_{s}^{2}}{|\xi|}\right)^{2}\ln\frac{|\xi|}{2r^{s}}{\nonumber}\\
&&+3\left(r_{s}^{2}-2|\xi|^{2}\right)+\left(15|\xi|^{2}-20r_{s}^{2}+8\frac{r_{s}^{4}}{|\xi|^{2}}\right)\Xi\Bigg]\end{aligned}$$ Despite the mass of the system being infinite, one can show that the convergence, shear, first- and second-flexion all tend to zero as $|\xi|\rightarrow\infty$. These profiles are plotted against the SIS and Sérsic-law profiles in figure \[allallMpc\], and we also look at the dependence of the concentration in figure \[NFWvarycall\] and section \[comparesec\].
Sérsic Profile {#Sersicsec}
--------------
It has long been argued that a Sérsic-law [@sersic68] provides a remarkably good fit to luminosity profiles of early-type galaxies, ranging in size from dwarf galaxies to the largest elliptical galaxies [@caon93; @graham01; @graham03; @graham03a; @trujillo04]. For a concise reference to Sérsic quantities, see [@graham05]. Recently, a Sérsic-law has also been shown to provide a good fit to three-dimensional density profiles [@navarro04], and also to projected surface density profiles [@merritt05] of dark matter halos. In a series of papers [@merritt06; @graham06; @graham06a] it has further been shown that projected Sérsic surface density profiles provide the best fit to simulated galaxy- and cluster-sized dark matter halos.
[@cardone04] first analysed the Sérsic profile in the gravitational lensing context, showing that mass estimates using lens reconstructions is highly dependent on the choice of Sérsic parameter. [@eliasdottir07] compared gravitational lensing for Sérsic and NFW profiles, and found that mass estimates may differ by up to a factor of two, dependent on the choice of density profile and Sérsic index. In the weak lensing regime, they did this by looking at the shear of both profiles. We take this a step further by also analysing higher-order lensing terms. In this section, we provide the first explicit representation of flexion terms for the Sérsic-law profile, and in section \[comparesec\] we compare these results to those of the NFW and SIS profiles.
The Sérsic profile is defined in terms of the surface density $$\ln\left(\frac{\Sigma}{\Sigma_{e}}\right)=-b_{n}\left[\left(\frac{|\xi|}{\xi_{e}}\right)^{1/n}-1\right],\label{Sersic}$$ where $\Sigma_{e}$ is the surface density at the effective radius, $\xi_{e}$. The constant $n$ is the Sérsic shape parameter which describes the shape of the profile and $b_{n}$ is a function of $n$ that is chosen such that the effective radius contains half of the projected mass of the system. Analytically, this is given as the solution of $\Gamma(2n)=2\Gamma(2n,\,b_{n})$, where $$\Gamma(\alpha,\,x)=\int_{t=0}^{x}{\rm e}^{-t}t^{\alpha-1}dt,$$ is the lower incomplete gamma function and $\Gamma(\alpha):=\lim_{x\rightarrow\infty}\Gamma(\alpha,\,x)$ is the complete gamma function. As such, $b_{n}$ can be reasonably approximated to $b_{n}=2n-1/3+4/(405n)+{\cal O}(n^{-2})$ for $0.5<n<10$ [@ciotti99]. Integrating the surface density gives the projected mass $$M=2\pi n\frac{{\rm e}^{b_{n}}}{b_{n}^{2n}}\Sigma_{e}\xi_{e}^{2}\Gamma\left(2n,\,Z\right),$$ where $Z=b_{n}\left(|\xi|/\xi_{e}\right)^{1/n}$. One can see that the Sérsic-law has one more parameter than the NFW profile, a point we discuss in more detail in section \[comparesec\] and also appendix \[PSapp\].
After much algebra, one can show that the magnitudes of the convergence and shear for the Sérsic profile can be expressed as $$\begin{aligned}
\kappa&=&\frac{\Sigma_{e}}{\Sigma_{cr}}\exp\left\{b_{n}\left[1-\left(\frac{|\xi|}{\xi_{e}}\right)^{1/n}\right]\right\},\\
|\gamma|&=&\frac{\Sigma_{e}}{\Sigma_{cr}}\Bigg(\exp\left\{b_{n}\left[1-\left(\frac{|\xi|}{\xi_{e}}\right)^{1/n}\right]\right\}{\nonumber}\\
&&-\frac{2n{\rm e}^{b_{n}}\xi_{e}^{2}}{b_{n}^{2n}{|\xi|^{2}}}\Gamma\left(2n,\,Z\right)\Bigg).\end{aligned}$$ Furthermore, the magnitudes of the first- and second-flexion terms are $$\begin{aligned}
|{\mathcal{F}}|&=&\frac{-D_{L}\Sigma_{e}b_{n}}{n\Sigma_{cr}|\xi|}\left(\frac{|\xi|}{\xi_{e}}\right)^{1/n}\exp\left\{b_{n}\left[1-\left(\frac{|\xi|}{\xi_{e}}\right)^{1/n}\right]\right\},\\
|{\mathcal{G}}|&=&\frac{D_{L}\Sigma_{e}}{\Sigma_{cr}}\Bigg(\frac{8n{\rm e}^{b_{n}}\xi_{e}^{2}}{b_{n}^{2n}|\xi|^{3}}\Gamma\left(2n,\,Z\right)-\frac{1}{|\xi|}\Bigg[\frac{b_{n}}{n}\left(\frac{|\xi|}{\xi_{e}}\right)^{1/n}-4\Bigg]{\nonumber}\\
&&\times\exp\left\{b_{n}\left[1-\left(\frac{|\xi|}{\xi_{e}}\right)^{1/n}\right]\right\}\Bigg).\end{aligned}$$
Profile Comparisons {#comparesec}
===================
We compare gravitational lensing effects of the various density distributions by holding the virial mass of each system constant, which is the same method used by [@wright00] for comparing first-order lensing properties of the NFW and SIS profiles. The three-dimensional (i.e. deprojected) virial radius, $r_{\Delta}$, is defined as the radius inside which the average density of the halo is $\Delta$ times the critical density of the Universe, implying the three-dimensional virial mass is $M_{\Delta}=4\pi\Delta\rho_{c}r_{\Delta}^{3}/3$ (throughout the remainder of the article we use $\Delta=200$). Constructing the lens models then requires the three-dimensional (deprojected) density distribution for each profile, as well as the three-dimensional mass distribution. In general, these two equations can be inverted to find $r_{\Delta}$, and also the various parameters associated with the individual profiles (for example the velocity dispersion, $\sigma$, for the SIS profile). The Sérsic profile is a little more difficult to treat with this procedure as analytic forms of the deprojected density and mass functions do not exist. As such, we use the analytic approximations given by [@prugniel97], as well as empirical relations from [@graham06a]. Details of the way in which we construct Sérsic-law density distributions are provided in appendix \[PSapp\].
Following [@bacon06], we use a flat $\Lambda$CDM cosmology with $\Omega_{m}=0.3$, $\Omega_{\Lambda}=0.7$ and $h=0.72$. We place the lensing mass at a redshift of $z_{L}=0.35$ and the source at $z_{S}=0.8$ as these values correspond to $D_{LS}/D_{S}\simeq0.5$. To compare the four density profiles, we use a lens mass of $M_{200}=10^{12}h^{-1}M_{\odot}$. For the NFW profile, the concentration factor, $c$, which is the ratio of the virial radius to the scale radius (see section \[NFWsec\]), is a function of the cosmology and the redshift of the lens which, for the above system, is evaluated to be $c=7.20$. For the Sérsic model we use empirically derived relations between the Sérsic shape parameter, $n$, and the mass of the system: from [@graham06a], their equation (12), we find the Sérsic shape parameter for a galaxy of mass $M_{200}=10^{12}h^{-1}M_{\odot}$ is $n\simeq8.6$.
Figure \[allallMpc\] shows the lensing properties of the Schwarzschild lens (dotted red line), SIS (dashed green line), NFW (blue dashed-dot line) and Sérsic (thick black line) mass distributions. The Schwarzschild lens has an Einstein radius of $\xi_{E}\simeq11.5$ kpc, which implies that for impact parameters inside this radius the results presented are not in the weak lensing regime. The most striking feature in these plots is that at large distances, the convergence and first-flexion for the SIS are significantly larger than for the NFW and Sérsic profiles. Bearing in mind that the convergence is linearly proportional to the surface density, this implies that at large distances the surface density of the SIS is also significantly larger than the NFW and Sérsic profile. This is consistent with the fact that the outer-logarithmic slope for NFW profiles is $-3$, while the slope of an isothermal sphere is $-2$. Additionally, the slope of a Sérsic-law profile depends on the specific shape parameter, $n$, and varies as a function of the distance from the centre of the profile, i.e. $|\xi|$. Interestingly, whilst these features are evident in the convergence and the first-flexion, they are less apparent in the shear and second-flexion.
It is worth exploring the extent to which the above properties of the NFW and Sérsic profiles depend on the concentration and shape parameters respectively. The concentration parameter in the NFW profile, $c$, is defined as the ratio of the virial radius to the scale radius, which is a function of the specific cosmology. Figure \[NFWvarycall\] shows the effect of a varying concentration on the different lensing properties. A lensing mass of $M_{200}=10^{10}h^{-1}M_{\odot}$ is used, with values of $c=4,\,8,\,12,\,16,\,20,\,24$. For a Schwarzschild lens of this mass at these distances the Einstein radius is $\xi_{E}\simeq 1.2$ kpc, which again gives us a scale on which a weak lensing treatment is appropriate.
It is apparent from figure \[NFWvarycall\] that the lensing properties of an NFW profile are not linearly effected by the concentration parameter. Indeed as $c$ becomes larger, the lensing properties begin to converge, implying that at large $c$ the plots become indistinguishable. Moreover, the effect of the concentration parameter is greater at small distances from the centre of mass of the lensing galaxy. This is seen most pertinently in the convergence where at $|\xi|\sim10\xi_{E}$ the lines are indistinct. It is also interesting to note that variations in the concentration parameter cause the shear and second-flexion to change significantly more than the convergence and first-flexion. Therefore, given a series of lensed images, and assuming an NFW fit to the density profile, one can learn more about the specifics of the profile from the shear and the second-flexion than from the convergence and the first-flexion. There is likely to be a degeneracy between the mass of the lensing galaxy and the concentration parameter if there are only a handful of images. Our results imply that using both the shear and the second-flexion may be able to break this degeneracy so $c$ and $M$ can be obtained.
Figure \[Sersicvarynall\] shows the lensing properties of Sérsic profiles for various values of the Sérsic shape parameter, $n$. The lensing mass is again $M_{200}=10^{10}h^{-1}M_{\odot}$, implying the point mass lens has $\xi_{E}\simeq1.2$ kpc, and we vary $1\le n\le9$. The striking feature of these plots is that the shear and second-flexion have significantly less dependence on the shape, $n$, of the Sérsic profile than the convergence and the first-flexion. Indeed, zooming in on the vertical scale by an order of magnitude for the shear and second-flexion plots reveals that the dependence on $n$ can only be seen for $|\xi|\ll\xi_{E}$. Moreover, the effect of the shape parameter on the convergence and first-flexion increases as one moves further from the source, although this may be difficult to detect observationally due to the relative size of the signal being significantly weaker at large separations from the lensing mass. The fact that there is no dependence on the specific shape of the Sérsic profile on the shear and second-flexion implies that these two properties can be used to derive the mass of the lensing object, whilst the convergence and first-flexion can then be used to derive $n$. Weak gravitational lensing thus provides an independent method for deriving the masses of Sérsic-law galaxies and clusters which only weakly depends on the specific shape of the profile, provided the correct lensing properties are utilised, i.e. the convergence and second-flexion.
Conclusion {#conc}
==========
We have derived general equations governing the convergence, shear, first- and second-flexion for circularly symmetric gravitational lenses in terms of the surface density and projected mass of the lens. We have shown that the components of the first-flexion are simply the directional derivatives of the surface density, while the second-flexion is a slightly more complicated function of the surface density, its gradient and also the projected mass distribution. By applying the formalism to specific lens models, in particular a Schwarzschild lens, a singular isothermal sphere (SIS), Navarro-Frenk-White (NFW) profile and a Sérsic profile, we have compared the signature each profile has on each of the lensing terms as a function of the radial impact parameter. Whilst the NFW and SIS profiles have been compared previously, both at linear-order [@wright00] and also for flexion [@bacon06], to the best of our knowledge this is the first time flexion for Sérsic-law profiles have been presented. In particular, we showed that the shear and second-flexion effects for a Sérsic profile are systematically larger than for the NFW and SIS profiles (figure \[allallMpc\]). This implies that one must be careful about the specific matter distribution assumed when attempting to analytically reconstruct the total mass and mass profile of a circularly symmetric lens.
An SIS profile is uniquely determined by designating the mass at a certain radius (e.g. virial mass). However, both the NFW and Sérsic models have extra parameters that determine their exact shape. The concentration of the NFW profile is the ratio of the virial radius to the scale radius, which is a function of the particular cosmology being considered. We showed that the convergence and first-flexion are relatively weakly dependent on the concentration parameter used, compared with the shear and the second-flexion (see figure \[NFWvarycall\]). The Sérsic profile is parametrized by the Sérsic shape parameter, $n$. Counter to the NFW concentration parameter, the Sérsic shape parameter is more heavily dependent on the convergence and first-flexion, whereas the shear and second-flexion vary little as $n$ changes (figure \[Sersicvarynall\]). These properties therefore provide the opportunity to directly measure both the mass of lensing galaxies as well as the specific shape of their density profiles. Given a limited supply of images around a specific galaxy, higher-order lensing terms may provide the ability to give extra constraints on the individual profiles of the galaxies.
The gravitational lensing systems explored in this article are idealized in that the projection of their mass distributions are circularly symmetric. Moving beyond circular symmetry requires the numerical solution of the thin-lens gravitational lens equation. One is then free to study the effect of non-circular lens models by including anisotropies in the lensing galaxy, and also to look at the effect the size of the source has on the shape of the final image. Whilst these applications will be extremely useful for determining the mass distributions of various galaxies using next-generation gravitational lensing surveys, the higher-order gravitational lensing effects will also be useful in studying the dark matter and dark energy content of the Universe. Moreover, although the study of first-order gravitational lensing through N-body simulations has been around for some time [see for example the recent review by @munshi08], the study of flexion in these systems has scarcely been broached. In this way, one can study the expected probability distributions for first- and second-flexion as a function of the specific cosmology. The advantage of using flexion as opposed to linear gravitational lensing effects is that a systematic bias is taken out of the study as only one reasonable assumption about the source object is required – that it is not intrinsically flexed.
Convergence, shear and flexion in angular coordinates {#anglesapp}
=====================================================
It is instructive to show the main equations from sections \[firstordersec\] and \[flexionsec\] expressed in angular coordinates, $(\theta_{1},\,\theta_{2})$, where $\xi_{i}=D_{L}\theta_{i}$. From the definition of the projected mass function in terms of the surface density, equation (\[massdef\]), implies that the projected mass expressed in angular coordinates is related to the mass in distance coordinates according to $M(|\xi|)=D_{L}^{2}M(|\theta|)$, where $|\theta|=\sqrt{\theta_{1}^{2}+\theta_{2}^{2}}$. It is trivial to show that the expression for the convergence, equation (\[kappa\]) goes unchanged, however the components of the shear, equations (\[gamma1\]) and (\[gamma2\]), written in angular coordinates become $$\begin{aligned}
\gamma_{1}(\theta_{1},\theta_{2})&=&\frac{\theta_{1}^{2}-\theta_{2}^{2}}{\pi\Sigma_{cr}|\theta|^{4}}\left(\pi\Sigma|\theta|^{2}-M\right)\\
\gamma_{2}(\theta_{1},\theta_{2})&=&\frac{2\theta_{1}\theta_{2}}{\pi\Sigma_{cr}|\theta|^{4}}\left(\pi\Sigma|\theta|^{2}-M\right),\end{aligned}$$ where for the remainder of this appendix $\Sigma=\Sigma(|\theta|)$ and $M=M(|\theta|)$. The magnitude of the total shear is $$\left|\gamma(\theta_{1},\theta_{2})\right|=\frac{1}{\pi\Sigma_{cr}|\theta|^{2}}\left(\pi\Sigma|\theta|^{2}-M\right).$$ The components of the first-flexion, equations (\[F1gen\]) and (\[F2gen\]), are given in angular coordinates as $$\begin{aligned}
{\mathcal{F}}_{1}(\theta_{1},\theta_{2})&=&\frac{1}{\Sigma_{cr}}\frac{\partial\Sigma}{\partial\theta_{1}},\\
{\mathcal{F}}_{2}(\theta_{1},\theta_{2})&=&\frac{1}{\Sigma_{cr}}\frac{\partial\Sigma}{\partial\theta_{2}},\end{aligned}$$ implying the magnitude of the first-flexion is $$\left|{\mathcal{F}}\left(\theta_{1},\theta_{2}\right)\right|=\frac{1}{\Sigma_{cr}}\frac{{\partial}\Sigma}{{\partial}|\theta|}.$$ Where previously the flexion terms were expressed in units of distance, one can now see first-flexion has units of (angle)$^{-1}$. It is trivial to see that the following three expressions governing the second-flexion in terms of angular coordinates are also expressed in these same units. $$\begin{aligned}
{\mathcal{G}}_{1}\left(\theta_{1},\theta_{2}\right)&=&\frac{\theta_{1}\left(\theta_{1}^{2}-3\theta_{2}^{2}\right)}{\pi\Sigma_{cr}|\theta|^{6}}\left[\pi\frac{d\Sigma}{d|\theta|}|\theta|^{3}-4\pi\Sigma|\theta|^{2}+4M\right],\\
{\mathcal{G}}_{2}\left(\theta_{1},\theta_{2}\right)&=&\frac{\theta_{2}\left(3\theta_{1}^{2}-\theta_{2}^{2}\right)}{\pi\Sigma_{cr}|\theta|^{6}}\left[\pi\frac{d\Sigma}{d|\theta|}|\theta|^{3}-4\pi\Sigma|\theta|^{2}+4M\right],\\
\left|{\mathcal{G}}\left(\theta_{1},\theta_{2}\right)\right|&=&\frac{1}{\Sigma_{cr}}\left(\frac{d\Sigma}{d|\theta|}-\frac{4\Sigma}{|\theta|}+\frac{4M}{\pi|\theta|^{3}}\right).\end{aligned}$$
Creating Galaxies with Sérsic Profiles {#PSapp}
======================================
The Sérsic profile is defined in terms of the projected surface density, equation (\[Sersic\]). However, to compare the lensing effects of the Sérsic profile with other density distributions, one requires the deprojected form of the density and mass distributions (see the discussion in section \[comparesec\]). Whilst an analytic form of the deprojected Sérsic profile is not available, an analytic approximation has been provided by[@prugniel97], and has further been explored in detail by [@merritt06; @graham06; @graham06a].
[@prugniel97] showed that the three-dimensional density distribution associated with the surface density of the Sérsic profile, given by equation (\[Sersic\]), can be approximated as $$\rho(r)=\rho_{e}\left(\frac{r}{\xi_{e}}\right)^{-p}\exp\left\{-b_{n}\left[\left(\frac{r}{\xi_{e}}\right)^{1/n}-1\right]\right\}.\label{PS}$$ Here, $\rho_{e}$ is the three-dimensional density at the effective (projected) radius $\xi_{e}$. The function $p=p(n)$ is utilised to ensure that the projection of equation (\[PS\]) relates as closely as possible to the projected Sérsic profile, i.e. equation (\[Sersic\]), for the range $0.6\le n\le10$. This was first given by [@limaneto99] as $p=1.0-0.6097/n+0.05463/n^{2}$, and a goodness of fit is shown in [@merritt06]. Integrating the three-dimensional density distribution over the volume gives the three-dimensional mass distribution, $$M_{3D}(r)=4\pi n\xi_{e}^{3}\rho_{e}{\rm e}^{b_{n}}b_{n}^{-\left(3-p\right)n}\Gamma\left[\left(3-p\right)n,\,b_{n}\left(\frac{r}{\xi_{e}}\right)^{1/n}\right].\label{Sersic3dmass}$$
[@cardone04] first looked at gravitational lensing for a Sérsic profile. He discussed the need to reduce the parameter space of the system in order to build a graviational lens from a Sérsic model. To that end, he used an empirical relation governing the deprojected effective radius, the central surface brightness and the Sérsic shape parameter, $n$ (see equation A.4 and A.5 of [@cardone04]). It is possible for us to also use this relation, and subsequently convert the central surface brightness into a density by invoking more empirical relations and also assuming a mass-to-light ratio. Essentially, this procedure has already been completed for the Prugniel & Simien model by [@graham06a] \[their equations (13) and (14)\]; $$\log_{10}\rho_{e}=k-2.5\log_{10}\xi_{e}.\label{rhoRe}$$ Here, $\xi_{e}$ is in units of kiloparsecs, $\rho_{e}$ is in solar masses per cubic parsec and $k$ is a constant which is $0.5$ for luminous elliptical galaxies and galaxy-sized dark matter halos (with $\log R_{e}\gtrsim0.5$) and $2.5$ for cluster-sized dark matter halos (with $\log R_{e}\gtrsim1.5$). Finally, the three-dimensional density at the effective (projected) radius is related to the two-dimensional surface density at $\xi_{e}$ by $$\rho_{e}=\Sigma_{e}b_{n}^{\left(1-p\right)n}\frac{\Gamma\left(2n\right)}{2\xi_{e}\Gamma\left[\left(3-p\right)n\right]}.\label{rhoeSigmae}$$
As mentioned in section \[comparesec\], to compare profiles we specify the virial mass of the system, $M_{\Delta}$, implying we know the virial radius, $r_{\Delta}$. Substituting this into equation (\[Sersic3dmass\]), together with equation (\[rhoRe\]) implies we have an equation for $\xi_{e}$ as a function of the Sérsic shape parameter, $n$. This equation is not analytically invertible due to the presence of the incomplete gamma function, however it can be solved numerically for given values of $n$. Therefore, once this equation is solved, we know $\xi_{e}$ and hence $\rho_{e}$, which can both be substituted into equation (\[rhoeSigmae\]) to give $\Sigma_{e}$. The projected Sérsic profile (\[Sersic\]) can finally be evaluated, along with its various derivatives and also the two-dimensional mass, implying all of the lensing quantities can be evaluated.
acknowledgments {#acknowledgments .unnumbered}
===============
We thank Alister Graham for helpful comments regarding the Sérsic profile and the Prugniel & Simien model. We also thank the referee for their extremely thorough and insightful review of the original manuscript. This research was supported under the Australian Research Councils Discovery Projects funding scheme (project number DP0665574).
\[lastpage\]
[^1]: E-mail: [email protected]
[^2]: E-mail: [email protected]
[^3]: Research undertaken as part of the Commonwealth Cosmology Initiative (CCI:www.thecci.org), an international collaboration supported by the Australian Research Council
[^4]: [@irwin05; @irwin06] also consider higher-order lensing, which they term [*sextupole*]{} lensing with components [*sextupole*]{}, [*cardioid*]{} and [*displacement*]{}. In [@irwin07] the authors show that sextupole is equivalent to second-flexion and a combination of the cardioid and displacement terms is equivalent to first-flexion.
[^5]: Galaxies are not intrinsically flexed providing they are dynamically relaxed.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In this paper, we derive simple closed-form expressions for the $n$-queens problem and three related problems in terms of permanents of $(0,1)$ matrices. These formulas are the first of their kind. Moreover, they provide the first method for solving these problems with polynomial space that has a nontrivial time complexity bound. We then show how a closed form for the number of Latin squares of order $n$ follows from our method. Finally, we prove lower bounds. In particular, we show that the permanent of Schur’s complex-valued matrix is a lower bound for the toroidal semi-queens problem, or equivalently, the number of transversals in a cyclic Latin square.'
address: |
University of Connecticut, 324 Whitney Hall, Storrs, CT 06269\
<[email protected]>
author:
- Kevin Pratt
title: 'Closed-Form Expressions for the n-Queens Problem and Related Problems'
---
Introduction
============
The $n$-queens problem is to determine $Q(n)$, the number of arrangements of $n$ queens on an $n$-by-$n$ chessboard such that no two queens attack. It is a generalization of the eight queens puzzle posed in 1848 by Max Bezzel, a German chess player. The $n$-queens problem has been widely studied since then, attracting the attention of Pólya and Lucas. It is now best known as a toy problem in algorithm design [@survey].
Despite this rich history, little is known of the general behavior of $Q(n)$. Key results are that $Q(n) > 1$ for $n > 3$, and $Q(n) > 4^{n/5}$ when $\gcd(n,30) = 5$. See [@survey] for a comprehensive survey. The only closed-form expression[^1] we are aware of was given in [@closed1]. It is “very complicated" in the authors’ own words, however.
The variants of the $n$-queens problem we consider are the *toroidal* $n$-queens problem $T(n)$, the *semi-queens* problem $S(n)$, and the *toroidal semi-queens* problem $TS(n)$. As with $Q(n)$, the general behavior of these functions is not well understood; asymptotic lower bounds are only known for $TS(n)$ [@asymp].
In this paper, we derive closed-form expressions for $Q(n), T(n), S(n)$, and $TS(n)$ in terms of permanents of $(0,1)$ matrices. The method we use is general and proceeds as follows. First, we come up with an *obstruction matrix* for a problem. Each entry in this matrix is a multilinear monomial. We then prove a formula for the sum of the coefficients of the terms containing some number of distinct variables in a polynomial. This is then used to obtain closed-form expressions for our problems. The expressions we obtain are very similar to those for the number of Latin squares of order $n$, such as those given in [@latin]. In fact, we show that one such formula is an immediate corollary of our method.
The permanent was previously considered by Rivin and Zabih to compute $Q(n)$ and $T(n)$ [@nqp]. Similarly, in 1874 Gunther used the determinant to construct solutions to the $n$-queens problem for small values of $n$ [@survey]. As far as we can tell however, no one has previously attempted to obtain closed-form expressions with this approach. The expressions we obtain in doing so can be evaluated in nontrivial time (i.e., better than the $O(n!)$ brute-force approach) and with polynomial space. The only other algorithms for computing $Q(n)$ and $T(n)$ with nontrivial time complexity bounds were given in [@alg]; however, this approach requires exponential space. We are not aware of any previously known algorithms for computing $S(n)$ and $TS(n)$ with nontrivial complexity bounds.
Finally, we prove lower bounds for these problems in terms of determinants of $(0,1)$ matrices. As a consequence, we show that the permanent of Schur’s complex-valued matrix [@schur] provides a lower bound for the toroidal semi-queens problem.
Preliminary Definitions
=======================
The permanent of an $n$-by-$n$ matrix $\mathbf{A}= (a_{i,j})$ is given by $$\mathrm{per}(\mathbf{A}) = \sum_{\sigma \in S_n}\prod_{i=1}^{n}a_{i,\sigma(i)}$$ where $S_n$ is the symmetric group on $n$ elements. It is a well-known result in complexity theory that computing the permanent of a matrix is intractable, even when restricted to the set of $(0,1)$ matrices [@sharpp].
An *obstruction matrix* $\mathbf{A}$ is a square matrix whose entries are multilinear monomials. If $\mathbf{A}$ contains the variables $x_1, x_2, \ldots, x_m$ and $s = (s_i) \in \{0,1\}^m$, then $\mathbf{A}|s$ is the matrix obtained by substituting $x_i = s_i$ for all $i$.
An $n$-by-$n$ matrix $\mathbf{M} = (m_{i,j})$ is *diagonally constant* if each northwest-southeast diagonal is constant; that is, $m_{i,j} = m_{i+1,j+1}$. A *circulant matrix* is a diagonally constant matrix with the property that each row is obtained by rotating the preceding row one position to the right, i.e., $m_{i,j} = m_{i+1, j+1 {\,\operatorname{mod}}n}$.
$Q(n)$ is the number of arrangements of $n$ queens on an $n$-by-$n$ chessboard such that no two attack; that is, lie on the same row, column, or diagonal .
$S(n)$ is the number of arrangements of $n$ nonattacking *semi-queens* on an $n$-by-$n$ chessboard [@oeis Sequence A099152]. A semi-queen has the same moves as a queen except for the northeast-southwest diagonal moves. Note that $S(n) \ge Q(n)$.
$T(n)$ is the number of arrangements of $n$ nonattacking queens on a toroidal $n$-by-$n$ chessboard [@oeis Sequence A051906]. The toroidal board is obtained by identifying the edges of the board as if it were a torus. As a result, the diagonals a queen can move along wrap around the board. Note that $Q(n) \ge T(n)$.
$TS(n)$ is the number of arrangements of $n$ nonattacking semi-queens on an $n$-by-$n$ toroidal chessboard. $TS(n)$ is also the number of transversals in a cyclic Latin square [@oeis Sequence A006717]. Note that $S(n) \ge TS(n)$.
Derivation of the Main Results
==============================
We begin by introducing the $n$-by-$n$ obstruction matrices $\mathbf{Q}_n$, $\mathbf{T}_n$, $\mathbf{S}_n$, and $\mathbf{Z}_n$, which will be used to compute $Q(n)$, $T(n)$, $S(n)$, and $TS(n)$, respectively.
$\mathbf{Q}_n$ contains the variables $x_1, y_1, \ldots, x_{2n-1}, y_{2n-1}$. The variable $x_i$ corresponds to the $i$th northwest-southeast diagonal (indexed from bottom left to top right), and $y_i$ corresponds to the $i$th northeast-southwest diagonal (indexed from bottom right to top left). The $(i,j)^{\text{th}}$ entry of $\mathbf{Q}_n$ is $x_{n-i+j}y_{2n-i-j+1}$.
$\mathbf{T}_n$ contains the variables $x_1, y_1, \ldots, x_{2n}, y_{2n}$. The variable $x_i$ corresponds to the $i$th northwest-southeast broken diagonal, and $y_i$ corresponds to the $i$th northeast-southwest broken diagonal. The $(i,j)^{\text{th}}$ entry of $\mathbf{T}_n$ is $x_{(n-i+j) {\,\operatorname{mod}}n} y_{(2n-i-j+1) {\,\operatorname{mod}}n}$.
$\mathbf{S}_n$ contains the variables $x_1, x_2, \ldots, x_{2n-1}$, and $x_i$ corresponds to the $i$th northwest-southeast diagonal. The $(i,j)^{\text{th}}$ entry of $\mathbf{S}_n$ is $x_{n-i+j}$.
$\mathbf{Z}_n$ contains the variables $x_1, x_2, \ldots, x_n$, and $x_i$ corresponds to the $i$th northwest-southeast broken diagonal. The $(i,j)^{\text{th}}$ entry of $\mathbf{Z}_n$ is $x_{(n-i+j){\,\operatorname{mod}}n}$.
Obstruction matrices for $Q(n),T(n),S(n)$, and $TS(n)$. $$\begin{aligned}
&\mathbf{Q}_4 = \begin{bmatrix}
{\textcolor{blue}{x_4}}{\textcolor{teal}{y_7}} & {\textcolor{Cyan}{x_5}}{\textcolor{Plum}{y_6}} & {\textcolor{Plum}{x_6}}{\textcolor{Cyan}{y_5}} & {\textcolor{teal}{x_7}}{\textcolor{blue}{y_4}}\\
{\textcolor{red}{x_3}}{\textcolor{Plum}{y_6}} & {\textcolor{blue}{x_4}}{\textcolor{Cyan}{y_5}} & {\textcolor{Cyan}{x_5}}{\textcolor{blue}{y_4}} & {\textcolor{Plum}{x_6}}{\textcolor{red}{y_3}}\\
{\textcolor{green}{x_2}}{\textcolor{Cyan}{y_5}} & {\textcolor{red}{x_3}}{\textcolor{blue}{y_4}} & {\textcolor{blue}{x_4}}{\textcolor{red}{y_3}} & {\textcolor{Cyan}{x_5}}{\textcolor{green}{y_2}}\\
{\textcolor{black}{x_1}}{\textcolor{blue}{y_4}} & {\textcolor{green}{x_2}}{\textcolor{red}{y_3}} & {\textcolor{red}{x_3}}{\textcolor{green}{y_2}} & {\textcolor{blue}{x_4}}{\textcolor{black}{y_1}}
\end{bmatrix}
&&\mathbf{T}_4 = \begin{bmatrix}
{\textcolor{blue}{x_4}}{\textcolor{black}{y_1}} & {\textcolor{black}{x_1}}{\textcolor{green}{y_2}} & {\textcolor{green}{x_2}}{\textcolor{red}{y_3}} & {\textcolor{red}{x_3}}{\textcolor{blue}{y_4}}\\
{\textcolor{red}{x_3}}{\textcolor{green}{y_2}} & {\textcolor{blue}{x_4}}{\textcolor{red}{y_3}} & {\textcolor{black}{x_1}}{\textcolor{blue}{y_4}} & {\textcolor{green}{x_2}}{\textcolor{black}{y_1}}\\
{\textcolor{green}{x_2}}{\textcolor{red}{y_3}} & {\textcolor{red}{x_3}}{\textcolor{blue}{y_4}} & {\textcolor{blue}{x_4}}{\textcolor{black}{y_1}} & {\textcolor{black}{x_1}}{\textcolor{green}{y_2}}\\
{\textcolor{black}{x_1}}{\textcolor{blue}{y_4}} & {\textcolor{green}{x_2}}{\textcolor{black}{y_1}} & {\textcolor{red}{x_3}}{\textcolor{green}{y_2}} & {\textcolor{blue}{x_4}}{\textcolor{red}{y_3}}
\end{bmatrix}\\
&\mathbf{S}_4 = \begin{bmatrix}
{\textcolor{blue}{x_4}} & {\textcolor{Cyan}{x_5}} & {\textcolor{Plum}{x_6}} & {\textcolor{teal}{x_7}}\\
{\textcolor{red}{x_3}} & {\textcolor{blue}{x_4}} & {\textcolor{Cyan}{x_5}} &{\textcolor{Plum}{x_6}}\\
{\textcolor{green}{x_2}} &{\textcolor{red}{x_3}} & {\textcolor{blue}{x_4}} & {\textcolor{Cyan}{x_5}}\\
{\textcolor{black}{x_1}} & {\textcolor{green}{x_2}} & {\textcolor{red}{x_3}} & {\textcolor{blue}{x_4}}
\end{bmatrix}
&&\mathbf{Z}_4 = \begin{bmatrix}
{\textcolor{blue}{x_4}} & {\textcolor{black}{x_1}} & {\textcolor{green}{x_2}} & {\textcolor{red}{x_3}}\\
{\textcolor{red}{x_3}} & {\textcolor{blue}{x_4}} & {\textcolor{black}{x_1}} & {\textcolor{green}{x_2}}\\
{\textcolor{green}{x_2}} &{\textcolor{red}{x_3}} & {\textcolor{blue}{x_4}} & {\textcolor{black}{x_1}}\\
{\textcolor{black}{x_1}} & {\textcolor{green}{x_2}} & {\textcolor{red}{x_3}} & {\textcolor{blue}{x_4}}
\end{bmatrix}\end{aligned}$$
Let $P$ be a polynomial, and let $k \in \mathbb{N}$. Then $g(P, k)$ is defined to be the sum of the coefficients of the terms in $P$ that are a product of exactly $k$ distinct variables.
Note that when $k = \deg{P}$, the terms whose coefficients are summed by $g(P,k)$ are multilinear. This leads to the following fact:
\[grel\] $g(\mathrm{per}(\mathbf{Q}_n), 2n) = Q(n)$, $g(\mathrm{per}(\mathbf{T}_n), 2n) = T(n)$, $g(\mathrm{per}(\mathbf{S}_n), n) = S(n)$, and $g(\mathrm{per}(\mathbf{Z}_n), n) = TS(n)$.
This follows immediately from the definition of the permanent and the structure of $\mathbf{Q}_n, \mathbf{T}_n, \mathbf{S}_n,$ and $\mathbf{Z}_n$. Consider $\mathrm{per}(\mathbf{Q}_n)$ for instance. We can write this as a sum of $n!$ terms of degree $2n$. Each term in this polynomial corresponds to a permutation matrix. If a term is square-free, then from the definition of $\mathbf{Q}_n$ no two elements in the corresponding permutation matrix lie along the same diagonal. Since a permutation matrix has no two nonzero entries on the same row or column, it follows that this permutation matrix corresponds to a solution for the $n$-queens problem.
Suppose that $P$ is a polynomial in $m$ variables. Let $S_{m,k}$ be the subset of $\{0,1\}^m$ that consists of the tuples containing $k$ ones; that is, $$S_{m,k} = \{(s_1, \ldots , s_{m}) \in \{0,1\}^{m} : \sum_{i=1}^{m} s_i = k\}.$$ Define $$f(P, k) = {\sum_{(s_1, \ldots , s_m) \in S_{m,k}}} P(s_1, \ldots, s_m).$$ The following fact is now used to derive an expression for $g$ in terms of $f$.
Let $m \ge k \ge l \ge 0$. Assume $$\begin{aligned}
a(k) &= \sum_{i=l}^k b(i) \binom{m-i}{k-i}. \\ \intertext{Then}
b(k) &= \sum_{i=l}^k a(i) \binom{m-i}{k-i} (-1)^{k-i}.\end{aligned}$$
\[cf\] Let $P$ be a polynomial in $m$ variables, and let $1 \le k \le m$. Then $$g(P, k) = \sum_{i=1}^k (-1)^{i+k} f(P, i) \binom{m-i}{k-i}.$$
Consider a term in $P$ that is a product of $i$ distinct variables where $i \le k$. It follows from the definition of $f$ that the coefficient of this term is counted by $f(P,k)$ a total of $\binom{m-i}{k-i}$ times. Therefore $$f(P, k) = \sum_{i=1}^{k} g(P, i) \binom{m-i}{k-i}.$$ Then by applying Fact 3.4 with $a(k) = f(P,k)$, $b(k) = g(P,k)$, and $l = 1$, equation (1) follows.
The following expressions follow directly from Lemma \[grel\] and Theorem \[cf\], and the fact that $\mathrm{per}(\mathbf{Q}_n),$ $\mathrm{per}(\mathbf{T}_n),$ $\mathrm{per}(\mathbf{S}_n),$ and $\mathrm{per}(\mathbf{Z}_n)$ are polynomials in $4n-2$, $2n$, $2n-1$, and $n$ variables, respectively.
Let $S_{m,k}$ be the subset of $\{0,1\}^m$ that consists of the tuples containing $k$ ones, $U_n$ the set of all $n$-by-$n$ $(0,1)$ diagonally constant matrices, and $V_n$ the set of all $n$-by-$n$ $(0,1)$ circulant matrices. Then the following identities hold: $$\begin{aligned}
Q(n) &= \sum_{i=1}^{2n} (-1)^{i} \binom{4n-i-2}{2n-i} \sum_{s \in S_{4n-2, i}} \mathrm{per}(\mathbf{Q}_n |s), \\
T(n) &= \sum_{i=1}^{2n} (-1)^{i+n} \sum_{s \in S_{2n, i}} \mathrm{per}(\mathbf{T}_n |s), \\
S(n) &= \sum_{\mathbf{M} \in U_n} (-1)^{\gamma(\mathbf{M})+n} \mathrm{per}(\mathbf{M}) \binom{2n - \gamma(\mathbf{M}) -1}{n-\gamma(\mathbf{M})},\\
TS(n) &= \sum_{\mathbf{M} \in V_n} (-1)^{\sigma(\mathbf{M})+n} \mathrm{per}(\mathbf{M}),\end{aligned}$$ where $\gamma(\mathbf{M})$ is the number of nonzero diagonals in $\mathbf{M}$, and $\sigma(\mathbf{M})$ is the number of ones in the first row of $\mathbf{M}$.
Note that multiple $(0,1)$ variable assignments to $\mathbf{Q}_n$ and $\mathbf{T}_n$ can correspond to the same $(0,1)$ matrix. As a result, one can think of the formulas for $Q(n)$ and $T(n)$ as summing over multisets of $(0,1)$ matrices. In the cases of $\mathbf{S}_n$ and $\mathbf{Z}_n$, there is a one-to-one relationship between $(0,1)$ variable assignments and $(0,1)$ matrices, so we can write $S(n)$ and $TS(n)$ as sums over sets of $(0,1)$ matrices.
Complexity Analysis
-------------------
The above expressions are impractical to evaluate even for small values of $n$; however, they do provide nontrivial time complexity bounds.
$Q(n)$, $T(n)$, $S(n)$, and $TS(n)$ can be computed in quadratic space and in time $O(n 32^n)$, $O(n 8^n)$, $O(n 8^n),$ and $O(n 4^n)$, respectively.
We can compute $Q(n)$ as follows. There are $O(2^{4n})$ $(0,1)$-tuples to enumerate in the summation. For each such tuple $s$, we compute $\mathbf{Q}_n | s$ in $O(n^2)$ time and space, and compute the permanent of this matrix in $O(n2^n)$ time and with $O(n^2)$ space using Ryser’s formula [@permc], which states that $$\mathrm{per}(\mathbf{A}) = \sum_{S\subseteq\{1,\dots,n\}} (-1)^{|S| + n} \prod_{i=1}^n \sum_{j\in S} a_{ij}.$$ Thus $Q(n)$ can be computed in $O(n32^n)$ time using $O(n^2)$ space. The other bounds are obtained similarly.
The only other algorithms we know of for $Q(n)$ and $T(n)$ with nontrivial complexity bounds run in time $O(f(n)8^n)$ where $f(n)$ is a low-order polynomial [@alg]. However, these algorithms require $O(n^28^n)$ space, whereas we only require $O(n^2)$ space. We do not know of any algorithms with nontrivial complexity bounds for the other two problems.
Extension: Latin Squares
------------------------
A Latin square of order $n$ is an arrangement of $n$ copies of the integers $1,2, \ldots,n$ in an $n$-by-$n$ grid such that every integer appears exactly once in each row and column. We now show how an expression for $L_n$, the number of Latin squares of order $n$, follows naturally from the method used above.
\[easy\] Let $\mathbf{B}_n$ be the $n$-by-$n$ obstruction matrix containing the variables $(x_1, x_2, \ldots, x_{n^2})$ defined by $(\mathbf{B}_n)_{i,j} = x_{i + n(j-1)}$. Let $\mathbf{A}_n$ be the $n^2$-by-$n^2$ block diagonal matrix
$$\mathbf{\mathbf{A}_n} = \begin{bmatrix}
\mathbf{B}_n & 0 & \cdots & 0 \\ 0 & \mathbf{B}_{n} & \cdots & 0 \\
\vdots & \vdots & \ddots & \vdots \\
0 & 0 & \cdots & \mathbf{B}_{n}
\end{bmatrix}.$$ Then $L_n = g(\mathrm{per}(\mathbf{A}_n), n^2)$.
A Latin square of order $n$ can be thought of as an ordered set of $n$ disjoint permutation matrices of order $n$. On the other hand, a term in $\mathrm{per}(\mathbf{A}_n)$ can be thought of as an ordered set of $n$ permutation matrices of order $n$, one along each copy of $\mathbf{B}_n$. If this term contains $n^2$ distinct variables, these permutation matrices must be disjoint. Therefore the sum of the coefficients of the terms in $\mathrm{per}(\mathbf{A}_n)$ containing $n^2$ distinct variables is exactly $L_n$.
Let $L_n$ be the number of Latin squares of order $n$. Then $$L_n = \sum_{\mathbf{M} \in M_n} (-1)^{\sigma (\mathbf{M})+ n} \mathrm{per} (\mathbf{M})^n$$ where $M_n$ is the set of all $(0,1)$ $n$-by-$n$ matrices, and $\sigma (\mathbf{M})$ is the number of nonzero entries in $\mathbf{M}$.
From Lemma 3.8 and Theorem 3.5, it follows that $$\begin{aligned}
L_n &= \sum_{i=1}^{n^2} (-1)^{i+n^2} f(\mathrm{per}(\mathbf{A}_n), i) \\
&= \sum_{i=1}^{n^2} (-1)^{i+n} \sum_{s \in S_{n^2,i}} \mathrm{per} (\mathbf{A}_n |s)\\
&= \sum_{i=1}^{n^2} (-1)^{i+n} \sum_{s \in S_{n^2,i}} \mathrm{per} (\mathbf{B}_n |s)^n,\\
\intertext{where the last step follows from the fact that $\mathrm{per}(\mathbf{A}_n) = \mathrm{per}(\mathbf{B}_n)^n$. Because $\mathbf{B}_n | u \neq \mathbf{B}_n | v$ if $u \neq v$, we can rewrite this as}
L_n &= \sum_{\mathbf{M} \in M_n} (-1)^{\sigma (\mathbf{M})+ n} \mathrm{per} (\mathbf{M})^n. \tag*{\qedhere}\end{aligned}$$
This formula was first given in [@latin].
Lower Bounds
============
In the last section, we showed that sums of coefficients in the permanents of the obstruction matrices $\mathbf{Q}_n,\mathbf{T}_n,\mathbf{S}_n,$ and $\mathbf{Z}_n$ correspond to the values of $Q(n),T(n),S(n)$, and $TS(n)$, respectively. We then gave a closed-form expression for the function $g$ that computes these sums. More precisely, $g(P,k)$ was the sum of the coefficients of the terms in the polynomial $P$ containing $k$ distinct variables.
Now since each entry in $\mathbf{Q}_n$ is a monomial with coefficient $1$, the coefficient of a term in $\det(\mathbf{Q}_n)$ is at most the coefficient of the corresponding term in $\mathrm{per}(\mathbf{Q}_n)$. Therefore $|g(\det(\mathbf{Q}_n),2n)| \le g(\mathrm{per}(\mathbf{Q}_n),2n) = Q(n)$. The same argument applies to the other problems. As a result we have the following corollary:
\[lowerbounds\] Let $S_{m,k}$ be the subset of $\{0,1\}^m$ that consists of the tuples containing $k$ ones, $U_n$ the set of all $n$-by-$n$ $(0,1)$ diagonally constant matrices, and $V_n$ the set of all $n$-by-$n$ $(0,1)$ circulant matrices. Then the following inequalities hold: $$\begin{aligned}
Q_{\det}(n) & := \bigg | \sum_{i=1}^{2n} (-1)^{i} \binom{4n-i-2}{2n-i}\sum_{s \in S_{4n-2, i}} \det(\mathbf{Q}_n |s) \bigg | \le Q(n),\\
T_{\det}(n) & := \bigg | \sum_{i=1}^{2n} (-1)^{i+n} \sum_{s \in S_{2n, i}} \det(\mathbf{T}_n |s) \bigg | \le T(n), \\
S_{\det}(n) & := \bigg |\sum_{\mathbf{M} \in U_n} (-1)^{\gamma(\mathbf{M})} \det(\mathbf{M}) \binom{2n - \gamma(\mathbf{M}) -1}{n-\gamma(\mathbf{M})} \bigg | \le S(n),\\
TS_{\det}(n) & := \bigg |\sum_{\mathbf{M} \in V_n} (-1)^{\sigma(\mathbf{M})} \det(\mathbf{M}) \bigg | \le TS(n),\end{aligned}$$ where $\gamma(\mathbf{M})$ is the number of nonzero diagonals in $\mathbf{M}$, and $\sigma(\mathbf{M})$ is the number of ones in the first row of $\mathbf{M}$.
We now show that $TS_{\det}(n)$ is the permanent of Schur’s matrix of order $n$; see [@oeis Sequence A003112].
Let $\mathbf{M}_n = (\epsilon^{jk})$ be an $n$-by-$n$ matrix where $\epsilon$ is an $n$th root of unity, and let $P_n = \mathrm{per}(\mathbf{M}_n)$. The matrix $\mathbf{M}_n$ is known as Schur’s matrix of order $n$. It has been of interest in number theory, statistics, and coding theory. Its permanent is the topic of [@schur].
For all $n$, $|P_n| \le TS(n) \le S(n)$.
From Corollary \[lowerbounds\], it suffices to show that $TS_{\det}(n) = |P_n|$. This follows immediately from the fact that $P_n = g(\det(\mathbf{Z}_n),n)$ [@schur].
[8]{}
J. Bell and B. Stevens. A survey of known results and research areas for $n$-queens. [*Discrete Mathematics*]{}, 309(1):1–31, 2009.
S. Chaiken, C. Hanusa, and T. Zaslavsky. A $q$-Queens Problem. II. The square board. [*Journal of Algebraic Combinatorics*]{}, 41(3):619–642, 2015.
S. Eberhard, F. Manners, and R. Mrazovi[ć]{}. Additive triples of bijections, or the toroidal semiqueens problem. [[[`arXiv:1510.05987`](http://arxiv.org/abs/1510.05987)]{}]{}, 2016.
R. Graham and D. Lehmer. On the permanent of Schur’s matrix. [*Journal of the Australian Mathematical Society*]{}, 21(A):487–497, 1975.
J. Hsiang, F. Hsu, and Y. Shieh. On the hardness of counting problems of complete mappings. [*Discrete Mathematics*]{}, 277(1–3):87–100, 2004.
H. Minc. Permanents. [*Encyclopedia of Mathematics and its Applications*]{}, 6:122–126, 1978. Addison-Wesley.
I. Rivin, I. Vardi, and P. Zimmerman. The $n$-Queens Problem. [*American Mathematical Monthly*]{}, 101(7):629–639, 1994.
I. Rivin and R. Zabih. A dynamic programming solution to the $n$-queens problem. [*Information Processing Letters*]{}, 41(5):253–256, 1992.
J. Shao and W. Wei. A formula for the number of Latin squares. [*Discrete Mathematics*]{}, 110(1–3):293–296, 1992.
N. Sloane., published electronically at https://oeis.org, 2016.
L. Valiant. The complexity of computing the permanent. [*Theoretical Computer Science*]{}, 21(2):189–201, 1979.
[^1]: We would like to correct a misunderstanding in [@survey]. The authors state that there exists no closed-form expression for $Q(n)$ because it was shown to be beyond the $\#P$ complexity class. However, the result referenced only shows that the $n$-queens problem is beyond $\#P$ because $Q(n)$ can be more than polynomial in $n$ [@hard]. A function can clearly be beyond $\#P$ for this reason and still have a closed-form expression; consider $2^n$ for instance.
|
{
"pile_set_name": "ArXiv"
}
|
Soviet Physics JETP; volume 30, number 3, pp. 512-513; March 1970
**NONLINEAR EVOLUTION OF DISTURBANCES IN A (1+1)-DIMENSIONAL UNIVERSE**
E. A. Novikov
Institute of Atmospheric Physics, USSR Academy of Sciences
Submitted April 4, 1969
Zh. Exper. Teor Fiz. **57**, 938-940 (September, 1969).
General exact solution is obtained for the problem of the development of arbitrary disturbances of the density and velocity in a (1+1)-dimensional universe. This analytical solution may serve, particularly, as a test for numerical methods. For an illustration, the nonlinear distortion of a sinusoidal perturbation of the initial density is calculated.
**1.** The problem of the origin of galaxies and their distributions with respect to masses and angular momenta is, first of all, a problem of the nonlinear evolution of disturbances of the density. The spherically symmetric solution by R. Tolman \[1\] and the first-order perturbations to the Friedmann solution, which were obtained by E. M. Lifshitz \[2\], are well known. These results, and also the formulation of the problem about the origin of galaxies, are significantly simplified in the Newtonian approximation \[3,4\], and also the Newtonian approximation turns out to be adequate for the problem under consideration, at least for the epoch after recombination of the plasma. But even in the Newtonian approximation the problem reduces to a rather complicated system of nonlinear equations with partial derivatives \[4\], which must be investigated by approximate methods (using different orders of perturbation theory and numerical procedures). In this connection it is useful to have a simple model of the problem possessing an exact solution, which may serve as a test of the approximate methods.
**2.** Let us consider a (1+1)-dimensional universe filled with dust (we shell be interested in scales of distance which are larger than the Jeans wavelength). The basic equations (corresponding to a special case of equations in Ref. 4) have the form:
$$\frac{\partial v}{\partial t}+v\frac{\partial v}{\partial x}=-\frac{\partial
\varphi }{\partial x} \tag{1}$$
$$\frac{\partial \sigma }{\partial t}+\frac{\partial }{\partial x}(\sigma v)=0
\tag{2}$$
$$\frac{\partial ^{2}\varphi }{\partial x^{2}}=\sigma \tag{3}$$
where $v$ is the velocity, $\varphi $ is the gravitational potential and $\sigma $ is a quantity which is proportional to the product of the density times the gravitational constant.$^{1)}$ Differentiating (1) with respect to $x$, with (3) taken into account, we obtain:
$$\frac{dh}{dt}+h^{2}=-\sigma ,\;\frac{d\sigma }{dt}+\sigma h=0, \tag{4}$$
$$\frac{d}{dt}\equiv \frac{\partial }{\partial t}+v\frac{\partial }{\partial x},\;h\equiv \frac{\partial v}{\partial x}.$$
The system (4) reduces to a single equation of the second order:
$$\frac{d^{2}}{dt^{2}}(\sigma ^{-1})=-1 \tag{5}$$
In the Lagrangian description the solution has the form$$\sigma (t,x_{o})=\sigma _{o}(x_{o})[1+h_{o}(x_{o})-1/2\sigma
_{o}(x_{o})t^{2}]^{-1}, \tag{6}$$
$$h(t,x_{o})=[h_{o}(x_{o})-\sigma _{o}(x_{o})t][1+h_{o}(x_{o})-1/2\sigma
_{o}(x_{o})t^{2}]^{-1}, \tag{7}$$
where $x_{o}$ is the initial distance of a fluid particle from the fixed particle to which the reference system is attached, $\sigma _{o}(x_{o})$ and $h_{o}(x_{o})$ denote the initial distributions.
We see that in a (1+1)-dimensional universe a contraction to a singularity always occurs after a time
$$t_{\ast }(x_{o})=\sigma
_{o}^{-1}(x_{o})\{h_{o}(x_{o})+[h_{o}^{2}(x_{o})+2\sigma
_{o}(x_{o})]^{1/2}\},$$
although for $h_{o}>0$ at the beginning, of course, there will be an expansion. One can show that the same result will also hold in a (2+1)-dimensional universe.$^{2)}$
To formulas (6) and (7) it is necessary to add the time-dependence of the moving coordinate $x(t,x_{o})$ of the fluid particle. It is easy to obtain this dependence from the equation of continuity
$$\frac{\partial x}{\partial x_{o}}=\frac{\sigma }{\sigma _{o}},$$
which is another way of writing Eq. (2). We have
$$x(t,x_{o})=x_{o}+u_{o}(x_{o})t-1/2\mu _{o}(x_{o})t^{2}, \tag{8}$$
$$u_{o}(x_{o})=\int\limits_{0}^{x_{o}}h_{o}(z)dz,\;\mu
_{o}(x_{o})=\int\limits_{0}^{x_{o}}\sigma _{o}(z)dz \tag{9}$$
Formulas (6), (7), (8), and (9) completely solve the problem in the Lagrangian description. The same formulas in parametric form ($x_{o}$ is the parameter) give the solution in the Eulerian description.
Let us emphasize the basic physical fact which enable us to obtain a simple and exact solution: in a (1+1)-dimensional universe the relative acceleration of the fluid particles is constant and is proportional to the mass confined between the particles.
**3.** Let us consider the following example: a sinusoidal perturbation of the initial density
$$\sigma _{o}(x_{o})=<\sigma _{o}>(1+a_{o}\sin \frac{2\pi x_{o}}{L_{o}}),\;h_{o}=const,$$
where $<\sigma _{o}>$ denotes the average density, $a_{o}$ is the relative initial amplitude, and $L_{o}$ is the initial spatial period. Formulas (6) and (8) take the form
$$\sigma (\tau ,s_{o})=<\sigma _{o}>(1+a\sin 2\pi s_{o})[1+\theta _{o}\tau
-1/2\tau ^{2}(1+a_{o}\sin 2\pi s_{o})]^{-1},$$
$$s\equiv \frac{x}{L_{o}}=s_{o}+s_{o}\theta _{o}\tau -\frac{1}{2}\tau
^{2}[s_{o}+\frac{1}{\pi }a_{o}\sin ^{2}\pi s_{o}],$$
$$s_{o}=\frac{x_{o}}{L_{o}},\;\theta _{o}=h_{o}<\sigma _{o}>^{-1/2},\;\tau
=t<\sigma _{o}>^{1/2}.$$
The graphs are constructed in the Eulerian description for
$$\theta _{o}=1,\;a_{o}=1/4,\;\tau =0,1,2.$$
The value $\tau =2$ corresponds to that moment when the spatial period, after a temporary increase, is again equal to the initial period, but the shape of the perturbation is now substantially distorted in accordance with gravitational condensation.
The author wishes to thank Ya. B. Zel’dovich for a helpful discussion of this work.
\[1\] L. D Landau and E. M. Lifshitz, Teoriya polya, Fizmatgiz, M. 1962 \[The Classical Theory of Fields, Pergamon Press, 1975\].
\[2\] E. M. Lifshitz, Zh. Eksp. Teor. Fiz. **16**, 587 (1946).
\[3\] Ya. B. Zel’dovich and I. D. Novikov, Relyativistskaya astrofizika (Relativistic Astrophysics), Nauka, M. 1967.
\[4\] Ya. B. Zel’dovich, Usp. Mat. Nauk **23**, 171 (1968).
——————————–
$^{1)}$ One can show that Eqs. (1) - (3) are the nonrelativistic limit of Einstein’s equations for a (1+1)-dimensional universe.
$^{2)}$ We present here one fantastic conjecture. Perhaps the universe was not always (3+1)-dimensional. The dimensionality might change during a transition through the singular state with zero space-dimensionality. Only starting with space-dimensionality equal to three did the universe gain the possibility “to survive”.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Consider a closed lipid membrane (vesicle), modeled as a two-dimensional surface, described by a geometrical hamiltonian that depends on its extrinsic curvature. The vanishing of its first variation determines the equilibrium configurations for the system. In this paper, we examine the second variation of the hamiltonian about any given equilibrium, using an explicitly surface covariant geometrical approach. We identify the operator which determines the stability of equilibrium configurations.'
author:
- |
R. Capovilla${}^{(1)}$, J. Guven${}^{(2)}$ and J.A. Santiago${}^{(2)}$\
${}^{(1)}$[*Departamento de Física*]{}\
[*Centro de Investigación y de Estudios Avanzados del IPN*]{}\
[*Apdo. Postal 14-740,07000 México, DF, MEXICO*]{}\
${}^{(2)}$[*Instituto de Ciencias Nucleares*]{}\
[*Universidad Nacional Autónoma de México*]{}\
[*Apdo. Postal 70-543, 04510 México, DF, MEXICO*]{}
title: Deformations of the geometry of lipid vesicles
---
Introduction
============
Many of the physical properties of a lipid membrane (or vesicle) are captured by a hamiltonian which describes the degrees of freedom of an idealized two-dimensional membrane surface. Typically, this hamiltonian is a sum of terms each of which either penalizes or constrains some geometrical characteristic of this surface. A term quadratic in the mean extrinsic curvature is a measure of the energy penalty associated with the bending of the membrane [@Can:70; @Hel:73; @Eva:74]. This term is supplemented by various constraints: the area as well as the enclosed volume are usually fixed. A constraint or (a term in the energy) linear in the mean extrinsic curvature describes an asymmetry between the lipid bilayers which constitute the membrane [@Sve.Zek:89]. (For a review see [@Nel.Pir.Wei:89; @Saf:94; @Pel:94; @Sve.Zek:96; @Sei:97].)
In equilibrium, this hamiltonian will be stationary with respect to arbitrary infinitesimal deformations in the surface. This infinitesimal deformation has both a tangential part and a normal part. The former corresponds to a reparametrization of the surface, and so can contribute only in boundary terms. The remaining normal infinitesimal deformation will completely describe the physical state of deformation of the membrane in its bulk. As a consequence, the equilibrium condition is a single equation, known somewhat prosaically as ‘the shape equation’ [@Hel.OuY:87]. In the thirty or so years since the introduction of the model, a lot has been learnt about the solutions of this equation, both analytically and numerically (see, for example, [@Sei:97] and the references it contains).
In this paper, we will examine the second variation of the hamiltonian. From one point of view, the calculation of the second variation about an equilibrium configuration is necessary in order to assess its degree of stability; in addition, the fluctuations of the membrane due to its interaction with its environment, are encoded completely at lowest order in the second variation of the hamiltonian. This problem has, of course, been addressed before and in both contexts. In the former, for example, perturbations about spherical or cylindrical configurations have been examined analytically [@Pet:85; @Hel.OuY:89; @Pet:89; @BYW:96]. A decomposition of geometries with a convex hull into spherical harmonics has also proven very effective [@Boz.Sve.Zek:97]. In the statistical mechanical context, while arbitrary background geometries have been considered, the focus has been on the part of the second variation contributing at one-loop to the renormalization of the parameters of the theory [@For:86; @Kle:86; @Lei.Pel:87; @Dav:89]. In any case, there is a gap in perturbation theory which we would like to fill. We will write down the second variation of the hamiltonian about an arbitrary equilibrium configuration, approaching the problem from a manifestly geometrically covariant point of view. Two features of the approach we will adopt should be emphasized. At the level of the second variation, it is no longer justified to ignore tangential deformations of the membrane. This is because a finite tangential deformation unlike its infinitesimal counterpart is not a simple reparametrization of the surface. Indeed, it was by examining finite tangential deformations of a given surface of constant negative Gaussian curvature preserving the curvature that Backlund, Bianchi and others generated new surfaces [@Rog.Sch:02]. We show, however, that when the background geometry is an equilibrium, tangential deformations contribute only boundary terms to the second variation of the Hamiltonian, and therefore for the consideration of bulk fluctuations about equilibrium, they can be neglected without incurring any error. A second simple but subtle technical point is the exploitation of the fact that the variation of the divergence of a vector density is equal to the divergence of the variation. Though this is not so essential computationally at the quadratic level we will work at, by facilitating the isolation of boundary terms, it does make higher order expansions in the fluctuations about a non-trivial background feasible in practice.
The paper is organized as follows. In section 2, we introduce our conventions for the geometrical description of a hypersurface embedded in $R^{N+1}$. We extend our considerations to an $N$-dimensional hypersurface because of the little extra cost involved and because of potential applications in other geometrical problems in soft matter physics. In section 3, we examine how the geometry changes under an infinitesimal deformation of the hypersurface. This is used in section 3 to derive the first variation of a geometrical hamiltonian for lipid membranes, and to obtain its Euler-Lagrange derivative. In section 4 the the second variation is examined. In particular, we check that it vanishes at equilibrium when the deformation corresponds to a rigid normal translation. We end in section 6 with a few concluding remarks.
Geometry
========
We begin by describing briefly the geometry of a hypersurface embedded in Euclidean space $R^{N+1}$. This allows us to introduce our conventions. We will emphasize the peculiarities associated with a surface embedded in $R^3$. At the end of this section, we also address the issue of identifying the low order independent reparametrization invariants one can construct from the geometrical quantities that characterize the hypersurface.
Consider an orientable hypersurface $\Sigma$ embedded in $R^{N+1}$. This surface can be specified locally in parametric form by $N+1$ shape functions, $${\bf x} = {\bf X}(\xi^a )\,,$$ where ${\bf x} = x^\mu = (x^1,\cdots, x^{N+1})$ are coordinates for $R^{N+1}$, $\xi^a$ arbitrary coordinates on the surface $\Sigma$ ($a, b, \cdots = 1,\cdots, N$), and ${\bf X}= (X^1,\cdots,X^{N+1})$ are the shape functions.
The Euclidean metric on $R^{N+1}$ induces the metric $g_{ab}$ on $\Sigma$ defined by, $$g_{ab} := {\bf e}_a \cdot {\bf e}_b \,,
\label{im}$$ where the $N$ tangent vectors are defined by ${\bf e}_a (\xi^a ) = \partial_a {\bf X}$ ($\partial_a := \partial / \partial \xi^a $). Latin indices are lowered and raised with $g_{ab}$, and its inverse $g^{ab}$, respectively. The metric $g_{ab}$ determines the intrinsic geometry of the hypersurface $\Sigma$. It defines the unique torsionless covariant derivative $\nabla_a$ compatible with it, [*i.e.*]{} satisfying $\nabla_a g_{bc} = 0$ and $ ( \nabla_a \nabla_b - \nabla_b \nabla_a ) f (\xi^a ) = 0$ for some surface function $f(\xi^a )$. In terms of the Christoffel symbol $\Gamma_{ab}^c$, , acting on a hypersurface vector $V^a$, it reads $$\nabla_a V^b = \partial_a V^b + \Gamma_{ac}^b V^c \,,$$ where $$\Gamma_{ab}^c : = g^{cd} {\bf e}_d \cdot \partial_a {\bf e}_b
= {1 \over 2} g^{cd} \left(
\partial_a g_{bd} + \partial_b
g_{ad} - \partial_d g_{ab} \right)\,.
\label{eq:csy}$$ The intrinsic Riemann curvature ${\cal R}^a{}_{bcd}$ of $\nabla_a$ is defined by $$( \nabla_{a} \nabla_{b} - \nabla_b \nabla_a ) V^c
=: {\cal R}^c{}_{dab} V^d\,.$$ In terms of the Christoffel symbol, the Riemann curvature tensor is given by $${\cal R}^a{}_{bcd} = \partial_c \Gamma_{db}^a -
\partial_d \Gamma_{cb}^a
+ \Gamma_{ce}^a \Gamma_{db}^e - \Gamma_{de}^a \Gamma_{cb}^e
\,.
\label{eq:curv3}$$ The Ricci tensor is defined by contraction, ${\cal R}_{ab} := {\cal R}^{c}{}_{acb}$; the scalar curvature ${\cal R}$ is defined by ${\cal R} := g^{ab} {\cal R}_{ab} $.
When the hypersurface $\Sigma$ is two-dimensional, $N=2$, the Riemann curvature tensor is completely determined by the scalar curvature: $${\cal R}_{abcd} = {{\cal R} \over 2} ( g_{ac}
g_{bd} - g_{ad} g_{bc} )\,.
\label{eq:curv}$$ Note that, as a consequence, we then have that in two dimensions the Einstein tensor vanishes, $${\cal G}_{ab} := {\cal R}_{ab} - {1 \over 2} {\cal R} g_{ab} =0\,.
\label{eq:curv1}$$ The scalar curvature of a two-dimensional surface is related to the Gaussian curvature $G$ of the surface by ${\cal R} = 2 G$.
The simplest geometrical quantity invariant under reparameterizations of the surface one can construct out of the intrinsic geometry of the hypersurface $\Sigma$ is its area, $$A := \int dA = \int d^N \xi
\; \sqrt{g}\,,$$ where $g$ denotes the determinant of the metric $g_{ab}$. The next order invariant depending only on the intrinsic geometry of the surface is the average scalar curvature over the surface, $\int dA \; {\cal R}$. For a two-dimensional surface with no boundary, by the well-known Gauss-Bonnet theorem, this is not only a reparameterization invariant, it is also a topological invariant, with $$\int dA \; {\cal R } = 4 \pi (1 - {\rm g})\,,
\label{eq:gb}$$ where ${\rm g}$ is the genus of the surface.
Let us turn now to the extrinsic geometry of $\Sigma$. The single normal vector ${\bf n} (\xi^a ) $ to $\Sigma$ in $R^N$ can be defined in implicit form by, $${\bf e}_a\, \cdot {\bf n} = 0\,,
\label{eq:dn1}$$ with the normalization, $${\bf n}\cdot {\bf n} = 1\,.
\label{eq:dn2}$$ Note that these equations determine ${\bf n}$ only up to a sign: the normal can be inward or outward. Since the surface is assumed to be orientable, we can pick one sign consistently, and we choose the normal to be outward. The unit normal vector field can also be given explicitly as $$n_\mu = {1 \over N! \sqrt{g}} \;
\varepsilon_{\mu \rho_1 \cdots \rho_{N-1}} \; \varepsilon^{a_1\cdots
a_{N-1}} \;
(e^{\rho_1}{}_{a_1} ) \cdots (e^{\rho_{N-1}}{}_{a_{N-1}} )\,,
\label{eq:n2}$$ where $ \varepsilon_{\mu_1\cdots \mu_N}$ and $\varepsilon^{a_1\cdots
a_{N-1}}$ are the totally antisymmetric Levi-Civita symbols for $R^{N+1}$ and $\Sigma$, respectively ($ \varepsilon_{1\cdots N} = + 1$). The factor of $\sqrt{g}$ is necessary in order to make ${\bf n}$ a scalar under reparametrizations.
When $\Sigma$ is closed, as we are assuming in this paper, the total volume enclosed by the surface $\Sigma$ in $R^{N+1}$ is an invariant. The normal to $\Sigma$, as given by (\[eq:n2\]), allows us to provide an alternative definition of the volume occupied by the interior of $\Sigma$ as the surface integral $$V = {1 \over N+1} \int dA \; {\bf n}\cdot {\bf X}\,.
\label{eq:vola}$$
The space vectors $\{ {\bf n}, {\bf e}_a \}$ form a basis adapted to the hypersurface. Their gradients along the surface are themselves space vectors, and can be decomposed in turn with respect to this basis. These decompositions constitute the classical Gauss-Weingarten equations, $$\begin{aligned}
\partial_a {\bf e}_b &=&
\Gamma_{ab}^c \; {\bf e}_c -
K_{ab} \; {\bf n}\,, \label{eq:gw1} \\
\partial_a {\bf n} &=& K_{ab} \; g^{bc} \; {\bf e}_c\,.
\label{eq:gw2}\end{aligned}$$ Here $\Gamma_{ab}^c$ is the Christoffel symbol defined in (\[eq:csy\]). The extrinsic curvature of $\Sigma$ is given by the symmetric rank two surface tensor, $$K_{ab} := - {\bf n}\cdot \partial_a {\bf e}_b= K_{ba}\,.$$ (Note that many authors differ by a sign in this definition.) We define its trace with respect to the intrinsic metric, $$K := g^{ab} K_{ab}=: 2 H\,,$$ where $H$ represents the mean extrinsic curvature of the surface.
The intrinsic and the extrinsic geometries of $\Sigma$, determined respectively by $g_{ab}$ and by $K_{ab}$, cannot be specified independently. They are related by the well-known integrability conditions of Gauss-Codazzi, and Codazzi-Mainardi, given respectively by, $$\begin{aligned}
{\cal R}_{abcd} - K_{ac} K_{bd} + K_{ad} K_{bc} &=& 0 \,,
\label{eq:gauss}\\
\nabla_a K_{bc} - \nabla_b K_{ac} &=& 0 \,.
\label{eq:cm}\end{aligned}$$ These equations follow as integrability conditions from taking a gradient along the surface of of the Gauss-Weingarten equations (\[eq:gw1\]), (\[eq:gw2\]), and then the appropriate anti-symmetric part.
The fundamental theorem for surfaces states that, given $g_{ab}$ and $K_{ab}$, these equations are not only necessary, but also sufficient for the existence of an embedding with these quantities as intrinsic metric and extrinsic curvature. Furthermore, the embedding is unique, up to rigid motions in the ambient space (see [*e.g.*]{} [@Spivak]).
Contraction of the Gauss-Codazzi-Mainardi equations with the contravariant intrinsic metric $g^{ab}$ results in $$\begin{aligned}
{\cal R}_{ab} - K K_{ab} + K_{ac} K_b{}^c &=& 0 \,,
\label{eq:gausscc}\\
{\cal R} - K^2 + K_{ab} K^{ab} &=& 0 \,,
\label{eq:gaussc}\\
\nabla_a K_b{}^a - \nabla_b K &=& 0\,.
\label{eq:cmc}\end{aligned}$$ For a two-dimensional surface, the contracted equations (\[eq:gaussc\]) and (\[eq:cmc\]) possess the same content as the full Gauss-Codazzi-Mainardi equations, (\[eq:gauss\]) and (\[eq:cm\]).
The lowest order geometrical invariant which involves the extrinsic geometry of the surface is the trace of the extrinsic curvature integrated over of the surface, $$M := \int dA \; K\,.$$ This invariant, linear in ${\bf n}$, depends on the orientation of the surface. Note that $[ K ] = L^{-1} $, so that $[M] = L $. The next order reparameterization invariants are quadratic in the extrinsic curvature ($\approx L^0$ ), $$\int dA \; K^2\,, \quad
\int dA \; K^{ab} K_{ab}\,,$$ together with the Gaussian term, $\int dA \; {\cal R}$. In general, these three invariants are not independent, as follows from the contracted Gauss equation, (\[eq:gaussc\]). Furthermore, in two dimensions, the Gaussian term is the Gauss-Bonnet topological invariant, see (\[eq:gb\]). For a two-dimensional surface, at this order there is then only one independent local scalar associated with the embedding, and it is customary to choose $\int dA \; K^2$. The gradients of the extrinsic curvature $\nabla_a K_{ab}$ which appear in the Codazzi-Mainardi equations do not feature. Differential bending can be ignored up to this order.
Let us now construct explicitly the geometrically independent terms of order $L^{-2}$ and lower. In general, at order $L^{-1}$, we have three independent scalars. Of the set, $K^3, K K^{ab} K_{ab}, K^{ab} K_{bc} K^{c}{}_a$, $K {\cal R}$, and $K_{ab} {\cal G}^{ab}$, the Gauss-Codazzi equations can be exploited to produce three that are independent $$\int dA \; K^3 \,, \quad \int dA \; K \,{\cal R} \,,
\quad {\rm and}\quad
\int dA \; K^{ab} {\cal G}_{ab}\,.$$ The last of the three vanishes identically for two dimensional surfaces.
At order $L^{-2}$ using the Gauss-Codazzi equation (\[eq:gauss\]), it is easy to show that the independent scalars are ${\cal R}^{abcd} {\cal R}_{abcd}$, ${\cal R}^{ab}
{\cal R}_{ab}$, ${\cal R}^2$, $K^4$, ${\cal R} K^2$, and ${\cal G}_{ab}
K^{ab}$. Of these, for a two-dimensional surface, only ${\cal R}^2$, $K^4$, ${\cal R} K^2$ survive. Moreover, at this order gradients of the extrinsic geometry enter. We can reduce all terms quadratic in $\nabla_a K_{bc}$ to the form $\nabla_a K \nabla^a K$ plus terms that we have already considered. We exploit the uncontracted Codazzi-Mainardi integrability condition (\[eq:cm\]) to write $$%\fl
\int dA\; (\nabla_a K_{bc}) (\nabla^a K^{bc} ) =
\int dA \; (\nabla_a K_{bc})(\nabla_b K_{ac})
= - \int dA \; K_{bc} \nabla_a\nabla_b K^{ac} \,,$$ where we have integrated by parts. We now note that $$[\nabla_a,\nabla_b] K^{ac}= {\cal R}^a{}_{dab} K^{dc}
+ {\cal R}^c{}_{dab} K^{ad}\,,$$ so that $$\begin{aligned}
%\fl
\int dA \; (\nabla_a K_{bc}) (\nabla^a K^{bc} ) &=& -
\int dA
K_{c}{}^b
( \nabla_b \nabla_a K^{ac} +
{\cal R}_{db} K^{dc}
+ {\cal R}^c{}_{dab} K^{ad}) \nonumber
\\
&=& \int dA \left( K^{bc} \nabla_b \nabla_c K
+{\cal R}_{db} K^{dc} K^b{}_c
+ {\cal R}_{cdab} K^{ad} K^{bc}
\right)\,, \nonumber\end{aligned}$$ where we have used the contracted Codazzi-Mainardi equation (\[eq:cmc\]) in the first term. We now integrate by parts again the first term, and use the Gauss-Codazzi-Mainardi equations to obtain $$%\fl
\int dA \; (\nabla_a K_{bc}) (\nabla^a K^{bc} )
=
\int dA
(\nabla^a K) (\nabla_a K )
+ {\cal R}^{ab} {\cal R}_{ab} - K K^{ab} {\cal R}_{ab}
+ {1 \over 2} {\cal R}_{abcd} {\cal R}^{abcd}
\,.$$ We conclude that for a two-dimensional surface, at this order there are four independent scalars, $$\int dA \;\left[ {\cal R}^2\,, \quad \int dA \; {\cal R} K^2\,,
\quad \int dA \; K^4\,, \quad {\rm and}\quad
\int dA \; (\nabla^a K)( \nabla_a K) \right]\,.$$ These higher order terms appear in geometric models for the so called egg-carton membranes [@GH96], and for tubular structures [@FG]. Note that if $K$ is treated as a simple scalar field, $\phi$, then the model described at this order is a $\lambda\phi^4$ theory non-minimally coupled to the intrinsic curvature. The Euler-Lagrange equations of the two models are, of course, different.
Deformations
============
In this section, we consider infinitesimal deformations of a hypersurface $\Sigma$, and how the various geometrical quantities that characterize it change under deformation. A one-parameter deformation of the hypersurface is described by the functions ${\bf X}(\xi^a,u)$. An infinitesimal change of the embedding functions $${\bf X} \to {\bf X} (\xi) + \delta {\bf X} (\xi)\,,$$ is characterized by the infinitesimal vector $\delta {\bf X} (\xi)
= \partial_u {\bf X}(\xi^a,u)|_{u=0} \delta u $. This vector can be decomposed into its components normal and tangential to the hypersurface $\Sigma$: $$\delta {\bf X} =
\delta_\perp {\bf X} + \delta_\parallel {\bf X}
= \Phi {\bf n} + \Phi^a {\bf e}_a \,.
\label{eq:def}$$ The components $\Phi, \Phi^a$ have the dimensions of length. We will assume that they are much smaller than any characteristic length in the system, such as the curvature radii.
We first determine how the basis $\{ {\bf e}_a , {\bf n} \} $ changes under deformation. The key observation is that $\partial_u$ and $\partial_a$ commute, so that the change in the tangent vector satisfies $$\delta {\bf e}_a =
\partial_a ( \delta{\bf X}) \,.$$ We now decompose $\delta {\bf X}$ into its tangential and normal parts according to (\[eq:def\]) so that $$\delta {\bf e}_a
= (\nabla_a \Phi^b) {\bf e}_b
- K_{ab} \Phi^b {\bf n} + (\nabla_a \Phi ) {\bf n}
+ \Phi K_{ab} g^{bc} {\bf e}_c
\label{eq:dt}
\,,$$ where we have used the Gauss-Weingarten equations (\[eq:gw1\]) and (\[eq:gw2\]). We thus have for the deformations induced by $\delta{\bf X}_{||}$ and $\delta{\bf X}_\perp$, respectively, $$\delta_{\parallel} {\bf e}_a = (\nabla_a \Phi^b) {\bf e}_b
- K_{ab} \Phi^b {\bf n} \,, \quad \quad
\delta_{\perp} {\bf e}_a = (\nabla_a \Phi ) {\bf n}
+ \Phi K_{ab} g^{bc} {\bf e}_c \label{eq:dt1}\,.$$ For the deformation of the induced metric, it follows that $$\delta_\parallel g_{ab}
=
\nabla_a \Phi_b + \nabla_b \Phi_a \,,
\quad \quad
\delta_\perp g_{ab}
= 2 K_{ab} \Phi\,.$$ The tangential deformation is just the the Lie derivative along the surface vector field $\Phi^a$. The normal deformation summarizes the geometrical content of the extrinsic curvature as one half of Lie derivative of the intrinsic metric along the normal vector field. Note that for the inverse metric one has that $\delta_\perp g^{ab} = - 2 K^{ab} \Phi$. It follows from these relations that the first order deformation of the infinitesimal area element is $$\delta_\parallel dA = dA \; \nabla_a \Phi^a\,,
\quad \quad
\delta_\perp dA = dA \; K \; \Phi\,.
\label{eq:sqrt}$$ This last expression encodes the geometrical content of the trace of the extrinsic curvature as the relative change of area per unit normal deformation.
The variations of the normal vector, $$\delta_{\parallel} {\bf n} = K_{ab} \; \Phi^a \; g^{bc} {\bf e}_c\,,
\quad \quad
\delta_\perp {\bf n} = - (\nabla_a \Phi ) g^{ab} {\bf e}_b \,,
\label{eq:defn}$$ follow readily from the defining relations (\[eq:dn1\]) and (\[eq:dn2\]), together with (\[eq:dt1\]).
For the remainder of this section we will focus on normal deformations of the hypersurface. The reason is that infinitesimal tangential deformations correspond to reparametrizations of the surface, and, as shown in the next section, they contribute only boundary terms to the variation of global quantities.
It is possible to evaluate the first order normal variation of the intrinsic scalar curvature ${\cal R}$ either intrinsically or extrinsically, the latter using the Gauss-Codazzi equation. Let us consider the former approach. Using the definition of the Riemann tensor given by (\[eq:curv3\]), we have that the variation of the Riemann tensor is (see [*e.g.*]{} [@Wald]) $$\delta {\cal R}^a{}_{bcd} = \nabla_c (\delta \Gamma_{db}^a
)
- \nabla_d (\delta \Gamma_{cb}^a )\,,$$ which implies $$\delta {\cal R}_{ab} = \nabla_c (\delta \Gamma_{ab}^c )
- \nabla_b (\delta \Gamma_{ca}^c )\,.$$ For the scalar curvature we have then that its normal variation is $$\begin{aligned}
\delta_\perp {\cal R} &=& (\delta_\perp g^{ab} )
{\cal R}_{ab} + g ^{ab} (\delta_\perp {\cal R}_{ab} )
\nonumber \\
&=& - 2{\cal R}_{ab} K^{ab} \Phi
+ \nabla_c ( g^{ab} \delta_\perp \Gamma_{ab}^c )
- \nabla^a (\delta_\perp \Gamma_{ca}^c )\,.
\nonumber \end{aligned}$$ Now, an arbitrary variation of the metric induces a variation $$\delta \Gamma_{ab}^c = {1\over 2} g^{cd} (\nabla_b
\delta g_{ad} +
\nabla_a \delta g_{bd} -\nabla_d \delta g_{ab})$$ in the Christoffel symbols, so that using the Codazzi-Mainardi equations, we have that the normal variation of the Christoffel symbol is $$\delta_\perp \Gamma_{ab}^c =
K_a{}^c \nabla_b \Phi + K_b{}^c \nabla_a \Phi
- K_{ab} \nabla^c \Phi
+ (\nabla_a K_b{}^c ) \Phi
\,.\label{eq:delgam}$$ As a consequence of the Codazzi-Mainardi equations, the (apparently unsymmetric) last term on the right hand side is, in fact, symmetric under the interchange of tangent indices. We determine then that the scalar curvature varies according to $$\delta_\perp {\cal R} =
- 2{\cal R}_{ab} K^{ab} \Phi
+ 2 \nabla_a \left[ ( K^{ab}- g^{ab} K)
\nabla_b \Phi \right]\,.
\label{eq:vcurv1}$$ Note that for the densitized scalar curvature we have $$\delta_\perp \sqrt{g} {\cal R} =
\sqrt{g} ( - 2{\cal G}_{ab} K^{ab} \Phi
+ 2 \nabla_a \left[ ( K^{ab}- g^{ab} K)
\nabla_b \Phi \right]\,.
\label{eq:vcurv}$$ It follows that the variation of the integrated scalar curvature is a pure divergence in two dimensions, as expected from the Gauss-Bonnet theorem (\[eq:gb\]).
Let us consider now the first order normal deformation of the extrinsic geometry. We have that $$\begin{aligned}
\delta_\perp K_{ab} &=& - (\delta_\perp {\bf n})\cdot \partial_a
{\bf e}_b - {\bf n}\cdot \partial_a ( \delta_\perp {\bf e}_b
) \nonumber \\
&=&
(\nabla_c \Phi ) {\bf e}^c \cdot \partial_a
{\bf e}_b - {\bf n} \cdot \partial_a [(\nabla_b \Phi) {\bf n}
+ \Phi K_{bc} {\bf e}^c ]
\nonumber \\ &=&
\Gamma_{ab}^c \nabla_c \Phi
- \partial_a \nabla_b \Phi + K_{ac} K^c{}_b \Phi
\nonumber \\ &=&
- \nabla_a \nabla_b \Phi + K_{ac} K^c{}_b \Phi\,.\end{aligned}$$ This is a remarkably simple expression. Using the contracted Gauss equation (\[eq:gausscc\]) we can cast this expression in the alternative form, $$\delta_\perp K_{ab}
= - \nabla_a \nabla_b \Phi + (K K_{ab}
- {\cal R}_{ab} ) \Phi\,.
\label{eq:Kab1}$$ For the trace of the extrinsic curvature this gives (recall that $\delta_\perp g^{ab} = - 2 K^{ab} \Phi$), $$\delta_\perp K = - \Delta \Phi + ( {\cal R} - K^2 ) \Phi \,,
\label{eq:dpc}$$ where $\Delta := g^{ab} \nabla_a \nabla_b$ denotes the laplacian on $\Sigma$, and we have used the contracted Gauss equation (\[eq:gaussc\]).
As a check of the consistency of the expressions we have derived, one can verify that they imply the vanishing of the first order variation of the contracted Gauss equation, [*i.e.*]{} $\delta_\perp ( {\cal R} - K^2 + K_{ab} K^{ab}) = 0$.
In the calculations of higher order variations, an essential ingredient is the commutator of the deformation operator and the covariant derivative on $\Sigma$. This commutator allows one to express deformations of the covariant derivative of any geometrical quantity of interest in terms of the covariant derivative of the variation of such quantity. For instance, when acting on an arbitrary second rank tensor $A_a{}^b$ in $\Sigma$, it is given by $$[ \delta_\perp , \nabla_a ] A_b{}^c = - (\delta_\perp
\Gamma_{ab}^d ) \; A_d{}^c
+ (\delta_\perp \Gamma_{ad}^c )\; A_b{}^d \,,$$ where $\delta_\perp \Gamma_{ab}^c$ is defined in (\[eq:delgam\]). Also for this case, it is a useful check of the validity of these expressions, to verify that the first order variation of the Codazzi-Mainardi equation (\[eq:cm\]), which involves such commutator, vanishes identically.
A useful expression is the commutator of the deformation derivative with the Laplacian, acting on an arbitrary function $f$, $$%\fl
[ \delta_\perp , \Delta ] f =
- 2 K^{ab} \Phi \nabla_a \nabla_b f
- 2 K^{ab} \nabla_a f \nabla_b \Phi
+ K \nabla^a f \nabla_a \Phi - (\nabla_a K) (\nabla^a f ) \Phi\,.
\label{eq:lapla}$$
First variation
===============
In this section, we apply the formalism for deformations we have developed to the systematic determination of the Euler-Lagrange derivatives of the geometrical invariants which appear in the hamiltonian for the strict bilayer couple model $$F [ X] = \alpha \int dA\; K^2 + \beta M + \mu A + P V\,.
\label{eq:hami}$$ The first term is the bending energy, with $\alpha$ the bending rigidity, the constants $ \mu, P, \beta$ are lagrange multipliers that enforce the constraints of constant area, constant volume and constant area difference, respectively. For the sake of simplicity, we have not included the non-local bending rigidity term, necessary in a realistic description of lipid vesicles [@BSZ; @Miao].
We first comment on tangential deformations. For the surface area element we have that under a tangential deormation it transforms according to (\[eq:sqrt\]). In addition, any surface scalar satisfies $\delta_\parallel f (X) = \Phi^a \partial_a f (X)$. Thus, any reparameterization invariant functional of the form $$F [X] = \int dA f (X)\,,
\label{eq:Ff}$$ deforms tangentially as $$\delta_\parallel F [X] = \int dA \; \nabla_a [ \Phi^a f (X) ]\,.$$ Using Stokes theorem this becomes $$\delta_\parallel F [X] = \int
ds \; \eta_a \; \Phi^a \; f (X) \,,
\label{eq:bdry}$$ where the integral ranges over the boundary of $\Sigma$ and $\eta^a$ denotes the unit normal to the boundary into $\Sigma$. If this boundary is empty, the integral vanishes identically. It is worthwhile emphasizing that in the case of a surface with a boundary, this integral in general will be non-vanishing, and it is no longer correct to neglect the tangential variations of the surface. In fact, in the variational principle, the vanishing of such terms will determine the boundary conditions to be imposed on the shape functions (see [*e.g.*]{} [@Cap.Guv.San:02a]).
If the surface has no boundary, the tangential part of the variation can always be associated with a reparametrization. Since we are interested, to begin with, in quantities that are invariant under surface reparametrizations, we disregard this contribution and focus on normal deformations of the geometry of $\Sigma$. We write $$\delta_\perp F [ X ] = \int dA \; \left\{ {\cal E}[f] \; \Phi +
\nabla_a V^a_{(1)} [f] \right\}\,.
\label{eq:Ffv}$$ The divergence comes about when we integrate by parts to remove all derivatives from the normal deformation $\Phi$. The subscript in the vector appearing in the second term refers to the order of the variation.
We begin with the first order normal variation of $A$ and $V$. For the normal variation of the total area it follows immediately from (\[eq:sqrt\]) that $$\delta_\perp A = \int dA \; K \; \Phi \,.
\label{eq:d1A}$$ The difference in area between two surfaces separated by an infinitesimal constant normal distance $\Phi$ is proportional to the integrated mean curvature. The area of the surface is extremal for arbitrary normal deformations, $\delta_\perp A =0$, when $K=0$ at each point on the surface.
One way to determine the normal deformation of the volume enclosed by the surface is to exploit the definition given in (\[eq:vola\]). We have $$\begin{aligned}
\delta_\perp V &=& {1 \over N+1} \int d^N \xi \;
[ (\delta_\perp \sqrt{g} )\; {\bf n}\cdot {\bf X} + \sqrt{g} \; (\delta_\perp
{\bf n} ) \cdot {\bf X}
+ \sqrt{g} \; {\bf n}\cdot \delta_\perp {\bf X} ]
\nonumber \\
&=& {1 \over N+1} \int dA \; [ \Phi\; K \; {\bf n} \cdot {\bf X}
- (\nabla_a \Phi ) \; g^{ab} \; {\bf e}_b \cdot {\bf X}
+ \Phi ]\,.\end{aligned}$$ Integrating by parts the second term and dropping a total divergence gives $$\begin{aligned}
\delta_\perp V &=& {1 \over N+1} \int dA \; [ \Phi\; K \; {\bf n}\cdot
{\bf X}
+ \Phi \; g^{ab} \; \partial_a {\bf e}_b
\cdot {\bf X} + \Phi \; g^{ab} \; {\bf e}_b\cdot {\bf e}_a
+ \Phi ]\nonumber \\
&=& {1 \over N+1} \int_\Sigma dA \; [ \Phi \; K \; {\bf n} \cdot {\bf X}
- \Phi \; K \; {\bf n}\cdot
{\bf X} + N \Phi + \Phi ]\nonumber \\
&=& \int dA\; \Phi \,,
\label{eq:d1V}\end{aligned}$$ where we have used the Gauss-Weingarten equations (\[eq:gw1\]) to obtain the second term of the second line. This final expression should come as no surprise. The infinitesimal change in volume is simply proportional to the area of the surface times the normal displacement.
We now examine the integrated powers of the mean extrinsic curvature. The deformation of the density $\sqrt{g} K$ , using (\[eq:sqrt\]), (\[eq:dpc\]), is $$\delta_\perp \sqrt{g} \; K = - \Delta \Phi + {\cal R} \; \Phi \,.
\label{eq:d1m}$$ We thus find for the total mean curvature, up to a total divergence, $$\delta_\perp M = \int dA \; {\cal R} \; \Phi \,.
\label{eq:aK1}$$ It is interesting that this expression depends only on the intrinsic geometry of the surface. If the scalar curvature vanishes, so also does $\delta_\perp M $. This is to be expected in a two-dimensional surface where ${\cal R}=0$ implies that it is flat. It is, however, a non-trivial statement for higher dimensions.
For the variation of the integrated second power of the extrinsic curvature we have $$\delta_\perp \int dA \; K^2
= \int dA \; K \; [ - 2 \Delta \Phi
+ ( 2 {\cal R} - K^2) \Phi ] \,;
\label{eq:m2K1}$$ We now integrate by parts twice to obtain, again up to a total divergence, $$\delta_\perp \int dA \; K^2
= \int dA \; \left[ - 2 \Delta K
+ ( 2 {\cal R} - K^2) K \right] \Phi\,.
\label{eq:m2K1b}$$
We are now in the position to derive the Euler-Lagrange derivative for the model defined by the Helfrich hamiltonian (\[eq:hami\]). Using the expressions (\[eq:d1A\]), (\[eq:d1V\]), (\[eq:aK1\]), (\[eq:m2K1b\]), in (\[eq:Ffv\]), we find [@Hel.OuY:87] $${\cal E} = - 2
\alpha \Delta K
+ (2 \alpha K + \beta){\cal R}
- \alpha K^3 + \lambda K + P \,.
\label{eq:shape}$$ At equilibrium, ${\cal E} = 0$, it is known as the shape equation. Note that it involves four derivatives of the shape functions. It determines the equilibrium configurations of lipid membranes described by the hamiltonian (\[eq:hami\]). Recently it was shown how to cast the shape equation in the form of a conservation law [@Cap.Guv:02].
Second variation
================
In this section, we derive the second variation of the hamiltonian (\[eq:hami\]). Before we proceed, we note that because the shape functions ${\bf X}$ are the variables to be varied, we have that $\delta_\perp \Phi \ne 0\,,$ to second order, and similarly for the remaining three possibilities, $\delta_\perp \Phi^a$, $\delta_{||} \Phi$ and $\delta_{||} \Phi^a$. In fact, using (\[eq:dt1\]), (\[eq:defn\]), we have $$\delta_\perp \Phi = - \Phi^a \nabla_a \Phi\,,
\quad
\delta_\parallel \Phi = K_{ab} \Phi^a \Phi^b\,,$$ together with $$\delta_\perp \Phi^a = \Phi \nabla^a \Phi + K^a{}_b \Phi^b \Phi\,, \quad
\delta_\parallel \Phi^a = \Phi^b \nabla^a \Phi_b - K^a{}_b \Phi^b \Phi\,.$$ This appears to suggest that the computation of the second variation is a formidable task and that the decompostion of deformations into tangential and normal parts will not be so useful. As we will show, the formal apparatus we have developed simplifies the calculation enormously and all of the troublesome tangential terms are collected in a divergence when the Euler-Lagrange equations are satisfied: no actual error is incurred in setting $\delta_\perp \Phi=0$ and neglecting $\Phi^a$ in the sequel.
Let ${\cal F}=\sqrt{g} f$, where $f$ is a scalar, introduced in (\[eq:Ff\]). At first order, we have $$\delta F = \int d^N\xi \; \left[ \delta_{||} {\cal F} +
\delta_\perp {\cal F} \right] =
\int d^n\xi \; \left[\sqrt{g} {\cal E}\Phi +
\nabla_a ({\cal V}_{(1)}[{\cal F}] + {\cal F}\Phi^a)\right]\,,$$ where ${\cal V}_{(1)}^a [{\cal F}]= \sqrt{g} V_{(1)}^a[f]$ is the densitized vector defined by Eq.(\[eq:Ffv\]). Now the divergence of a vector density is independent of the surface affine connection $\Gamma_{ab}^c$, and thus variation of the divergence of a vector density ${\cal V}^a$ is equal to the divergence of the variation, $$\delta (\nabla_a {\cal V}^a)
= \nabla_a (\delta {\cal V}^a)\,.$$ We thus have at second order, $$\delta^2 F =
\int d^N\xi \; \delta \left[\sqrt{g} {\cal E}\Phi\right]
\,,$$ modulo another divergence. Thus, whereas for the second variation of ${\cal F}$ we cannot legitimately discard the divergence, we can for $F$. Furthermore, we have $$\delta \left[\sqrt{g} {\cal E}\Phi\right] =
\delta_\perp (\sqrt{g} {\cal E}) \Phi
+ \nabla_a (\sqrt{g} {\cal E}\Phi^a)\Phi + \sqrt{g}{\cal E}\delta\Phi\,.$$ Thus, modulo the Euler-Lagrange equation, $\delta \left[\sqrt{g} {\cal E}\Phi\right] =
\delta_\perp (\sqrt{g} {\cal E}) \Phi$. But $\delta_\perp (\sqrt{g} {\cal E})$ can always be expressed as ${\cal L}\Phi$, for some local differential operator ${\cal L}$, and therefore the second order variation of the hamiltonian can itself be written in the form $$\delta^2 F [X] = \int dA \;
\Phi \; {\cal L } \; \Phi \,.
\label{eq:op}$$ It must be remembered, however, that this is not true for the individual terms contributing to $F$, for which ${\cal E}\ne 0$. When the second variation of these terms is written down it, it should be understand that the expression refers to the second normal variation; off diagonal terms $\sim \Phi \Phi_a $ as well as purely tangential terms $\sim \Phi_a \Phi_b$ have been suppressed.
A second order variation can also be seen as a different deformation $ X^\mu \to X^\mu (\xi) + \Phi' {\bf n}
$ applied to the first order variations, and then letting $\Phi' = \Phi $. From a computational point of view, this is clearly equivalent to a repeated application of the normal deformation operator $\delta_\perp $.
For the second order variation of the enclosed volume no work is needed, since, using (\[eq:sqrt\]), it is simply related to the first variation of the area with $$\delta^2 V = \delta \int dA \; \Phi =
\int dA \; K \; \Phi^2\,.$$
In order to derive the second variation of the area, various strategies are possible. On one hand one can simply calculate the second variation of the metric. Using (\[eq:Kab1\]) one obtains, $$\delta_\perp^2 g_{ab}
= 2 \left( \delta_\perp K_{ab} \right) \; \Phi
= - 2 \Phi \;[
\nabla_a \nabla_b \Phi - K_a{}^c K_{cb} ]\; \Phi\,,$$ so for the area element this gives $$\delta_\perp^2 \sqrt{g} =
\sqrt{g} ( - \Phi \Delta \Phi + {\cal R } \Phi^2 )\,,
\label{eq:2varea}$$ and the second variation of the area is given by the well-known expression (see [*e.g.*]{} [@Lawson]) $$\delta^2 \, A = \int dA \;
\Phi \left\{- \Delta + {\cal R }
\right\}\Phi \,.$$
There is an alternative route to the second variation of the area, which becomes increasingly useful as one considers higher variations. The idea is to exploit the natural hierarchy in the variations illustrated by the relations $$\begin{aligned}
\delta_\perp \sqrt{g} &=& \sqrt{g} \; K \; \Phi \nonumber \\
\delta_\perp^2 \sqrt{g} &=& \delta_\perp (\sqrt{g} K ) \Phi\,.
\nonumber\end{aligned}$$ Using (\[eq:d1m\]), this gives directly (\[eq:2varea\]). The usefulness of this approach is apparent: the $n^{\rm th}$ variation of the surface volume element can be expressed in terms of the $(n-1)^{\rm th}$ variation of the densitized mean extrinsic curvature.
We now exploit this observation to derive the second variation of the densitized mean extrinsic curvature. We rewrite (\[eq:aK1\]) as $$\delta_\perp (\sqrt{g} K ) = \sqrt{g} \; {\cal R} \;
\Phi + \nabla_a {\cal V}^a_{(1)} [K]\,,
\label{eq:aK1a}$$ where we have defined the vector density ${\cal V}^a_{(1)} [K] := - \sqrt{g} \nabla^a \Phi$. Using the fact that the variation of the divergence of a vector density is equal to the divergence of the variation, we have $$\begin{aligned}
%\fl
\delta_\perp^2 (\sqrt{g} K ) &=&
\delta_\perp \{ \sqrt{g} {\cal R}
\Phi + \nabla_a {\cal V}^a_{(1)} [K] \}
\nonumber \\
&=& \Phi \delta_\perp ( \sqrt{g} {\cal R})
+ \nabla_a {\cal V}^a_{(2)} [K] \nonumber \\
&=& 2 \sqrt{g} \Phi \left[
( K^{ab} - K g^{ab} ) \nabla_a \nabla_b \Phi
- {\cal G}_{ab} K^{ab} \Phi \right]
+ \nabla_a {\cal V}^a_{(2)} [K])\,,
\label{eq:d2aKa}\end{aligned}$$ where we have used (\[eq:vcurv\]), and defined the vector density $${\cal V}^a_{(2)} [K] = \delta_\perp {\cal V}^a_{(1)} [K]
= \sqrt{g} \Phi ( 2 K^{ab} - g^{ab} K )
\nabla_b \Phi\,.
\label{eq:va2}$$ We have therefore for the second variation of $M$, $$\delta^2 M = \int dA \; \{
2 \Phi \; [ ( K^{ab} - K g^{ab} ) \nabla_a \nabla_b -
{\cal G}_{ab} K^{ab} ] \;\Phi \} \,.$$ Note that if the surface is flat then the second order variation, like the first, vanishes. Thus, in a gaussian approximation about a flat background geometry, this term is absent.
Our final task in this section is to compute the second variation of the densitized squared mean extrinsic curvature. Unfortunately, this quantity does not fit naturally in the hierarchy of variations we have been exploiting. One approach is to calculate the second variation of $K$ directly. From (\[eq:dpc\]), using (\[eq:vcurv1\]), (\[eq:lapla\]), we have that $$%\fl
\delta_\perp^2 K = 4 K^{ab} \Phi \nabla_a \nabla_b \Phi
+ (2 K^{ab} - K g^{ab} ) \nabla_a \Phi \nabla_b \Phi
+ 2 K (K^2 - {\cal R} ) \Phi^2 - 2 {\cal R}_{ab} K^{ab} \Phi^2\,.$$ We prefer to capitalize on the relative simplicity of the expressions for the variation of the densitized mean curvature and use the identity $$%\fl
\delta_\perp^2 (\sqrt{g} K^2 ) =
2K \;\delta_\perp^2 (\sqrt{g} K ) + 2 [ \delta_\perp (\sqrt{g} K) ]^2
g^{-1/2} - 5 K^2 \Phi \delta_\perp (\sqrt{g} K)
+ 2 \sqrt{g} K^4 \Phi^2\,.$$ A short calculation gives $$\begin{aligned}
%\fl
\delta_\perp^2 (\sqrt{g} K^2 ) &=&
\sqrt{g} [ 2 (\Delta \Phi )^2
+ 4 \Phi K K^{ab} \nabla_a \nabla_b \Phi
+ (K^2 - 4
{\cal R} ) \Phi \Delta \Phi
\nonumber \\
&+& ( 2 {\cal R}^2
+ 2 K^4 - 5 {\cal R} K^2 - 4 {\cal G}_{ab} K^{ab} K)
\Phi^2 ] + 2K \nabla_a {\cal V}^a_{(2)} [K] \,,\end{aligned}$$ where ${\cal V}^a_{(2)} [K]$ is defined by (\[eq:va2\]). We isolate a total divergence in the final term using $$2K \nabla_a {\cal V}^a_{(2)} [K]
= - 2 (\nabla_a K) {\cal V}^a_{(2)} [K]
+ \nabla_a {\cal V}^a_{(2)} [K^2]
\,,$$ where we define $${\cal V}_{(2)}^a [K^2] = 2 \sqrt{g} \left[ \Delta\Phi \nabla^a \Phi -
\Phi \nabla^a \Delta
\Phi
+ \Phi (2 K^{ab} - K g^{ab}) \nabla_b \Phi \right]\,,$$ to obtain $$\begin{aligned}
%\fl
\delta^2 \int dA \; K^2
&=& \int dA \; \Phi [
2 \Delta^2
+ 4 K K^{ab} \nabla_a \nabla_b
+ (K^2 - 4
{\cal R} ) \Delta +
2 {\cal R}^2
\nonumber \\
&-&
2 (\nabla_a K) ( 2 K^{ab} - g^{ab} K )
\nabla_b
+ 2 K^4 - 5 {\cal R} K^2 - 4 {\cal G}_{ab} K^{ab} K ] \Phi
\,,\end{aligned}$$ We can simplify this expression using $$%\fl
- 2 \int dA \Phi (\nabla_a K) ( 2 K^{ab} - g^{ab} K ) \nabla_b \Phi
= \int dA [
2 K^{ab} \nabla_a \nabla_b K - K \Delta K +
(\nabla^a K ) (\nabla_a K ) ] \Phi^2\,,$$ to arrive at the final expression $$\begin{aligned}
%\fl
\delta^2 \int dA \; K^2
&=& \int dA \; \Phi \; [
2 \Delta^2
+ 4 K K^{ab} \nabla_a \nabla_b
+ (K^2 - 4
{\cal R} ) \Delta
+2 K^{ab} \nabla_a \nabla_b K
\nonumber \\
&-&
K \Delta K +
(\nabla^a K ) (\nabla_a K )
+ 2 {\cal R}^2
+ 2 K^4 - 5 {\cal R} K^2 - 4 {\cal G}_{ab} K^{ab} K ] \Phi
\,.
\label{eq:expression}\end{aligned}$$
Therefore, for the second order variation of the Helfrich hamiltonian (\[eq:hami\]), as expressed in (\[eq:op\]), we can write the local differential operator ${\cal L}$ as $${\cal L} = 2\alpha \Delta^2 + 2 A \Delta
+ 2 A^{ab} \nabla_a \nabla_b + 2 B\,,
\label{eq:oper}$$ where $$\begin{aligned}
2 A &=& \alpha ( K^2 - 4 {\cal R} ) - 2 \beta K - \lambda\,,
\\
2 A^{ab} &=& 4 \alpha K K^{ab} +
2\beta K^{ab} \\
2B &=&
\alpha [ 2 {\cal R}^2
+ 2 K^4 - 5 {\cal R} K^2 - 4 {\cal G}_{ab} K^{ab} K
+2 K^{ab} \nabla_a \nabla_b K \nonumber \\
&-&
K \Delta K +
(\nabla^a K ) (\nabla_a K ) ]
- 2\beta {\cal G}_{ab} K^{ab} + \lambda {\cal R} + P K \,. \end{aligned}$$ We note that $A^{ab}$ is symmetric. Furthermore, the operator ${\cal L}$ is always self-adjoint, by which we mean that $$\int dA \,\Phi_1 {\cal L} \Phi_2 = \int dA \,\Phi_2 {\cal L} \Phi_1\,.$$ Thus, the eigenvalues of ${\cal L}$ are assured to be real valued. To see this, consider the terms that could potentially spoil self-adjointness originating in the contributions proportional to $\alpha$ with two derivatives of $\Phi$. Let $\tilde A^{ab} = A^{ab} + g^{ab} A$. Now, $$\begin{aligned}
\Phi \tilde A^{ab} \nabla_a \nabla_b \Phi &=& \Phi\nabla_a (\tilde A^{ab} \nabla_b \Phi)
- \Phi\nabla_a \tilde A^{ab}\, \nabla_b \Phi \nonumber \\
&=& \Phi\nabla_a (\tilde A^{ab}\, \nabla_b \Phi) +
{1\over 2} \nabla_a \nabla_b \tilde A^{ab}\,\, \Phi^2 \,,\end{aligned}$$ modulo a total divergence which we discard. In this form the operator $\tilde A^{ab} \nabla_a \nabla_b$ is manifestly self-adjoint: we have $$\Phi_1\nabla_a ( \tilde A^{ab}\, \nabla_b \,\Phi_2 )
= \Phi_2 \nabla_b ( \tilde A^{ab}\, \nabla_a\, \Phi_1)$$ modulo two total divergences. We can write $${\cal L} = 2\alpha \Delta^2 + 2 \nabla_a \tilde A^{ab} \nabla_b
+ 2 B_1\,,
\label{eq:oper1}$$ where $$2B_1 =
\alpha [ 2 {\cal R}^2
+ 2 K^4 - 5 {\cal R} K^2 - 4 {\cal G}_{ab} K^{ab} K
+2 \Delta (K^2- {\cal R})]
- 2\beta {\cal G}_{ab} K^{ab} + \lambda {\cal R} + P K \,.$$
We note that the expression Eq.(\[eq:expression\]) agrees with the one obtained in [@For:86; @Kle:86], in the terms containing derivatives of $\Phi$. These previous studies were focused on the effect of short wavelength fluctuations, where only these terms contribute. We also note that the expression we have obtained disagrees with the one obtained previously in [@Hel.OuY:89]. Agreement is restored in the special case of spherical and cylindrical configurations.
It is clear that a rigid translation of the surface should not alter the second variation, at equilibrium. This provides a non-trivial check of our expression (\[eq:oper\]). As it is not entirely obvious, we will outline the details. Throughout, we will use extensively the Gauss-Weingarten equations (\[eq:gw1\]), (\[eq:gw2\]), and the contracted Gauss-Codazzi-Mainardi equations (\[eq:gausscc\]), (\[eq:gaussc\]), (\[eq:cmc\]). Consider a constant normal deformation $\delta {\bf X} =
{\bf a}$, such that ${\bf a} \cdot {\bf e}_a = 0$. We have for the derivatives of $\Phi$ that appear in (\[eq:oper\]), $$\begin{aligned}
\nabla_a \nabla_b \Phi &=& - K_{ac} K^c{}_b ({\bf n \cdot a})\,,
\nonumber \\
%\Delta \Phi &=& - K_{ab} K^{ab} ({\bf n \cdot a})\,, \nonumber \\
\Delta^2 \Phi &=& [- 2 K^{ab} \nabla_a \nabla_b \Phi - (\nabla_a K ) (\nabla^a K )
- \Delta ( K_{ab} K^{ab} ) + (K_{ab} K^{ab})^2 ]({\bf n \cdot a})\,.
\nonumber \end{aligned}$$ If we now insert these expressions in the second variation, we find $$\begin{aligned}
\delta^2 F &=& \int dA\; \{ \alpha [ - 2 K^{ab} \nabla_a \nabla_b \Phi - (\nabla_a K ) (\nabla^a K )
- \Delta ( K_{ab} K^{ab} ) - K \Delta K + 2 {\cal R} K^2 - K^4 ]
\nonumber \\
&+& \beta {\cal R} K + \lambda K^2 + P K \}
( {\bf n \cdot a})^2\,.\end{aligned}$$ We now integrate by parts the terms that involve derivatives of the extrinsic curvature. The desired simplifications follow from the crucial identity, for any surface vector $V^a$, $$\int dA \; (\nabla_a V^a ) ({\bf n \cdot a})^2 = 0\,,$$ up to boundary terms. This is a consequence of $\nabla_a ({\bf n \cdot a}) = K_{ab} ({\bf e}_a \cdot {\bf a}) = 0$, by hypothesis. Using this identity, we easily arrive at $$\delta^2 F = \int dA \; K {\cal E} ({\bf n \cdot a})^2\,,$$ where ${\cal E}$ is the Euler-lagrange derivative defined by (\[eq:shape\]). Therefore at equilibrium the second variation vanishes for rigid translations.
Note that a similar check has been used for the second variation of particular configurations, as, for example, spheres in [@Pet:85; @Hel.OuY:87]. However, not all of the terms appearing in the second variation can be checked considering only symmetric configurations, since many terms simply vanish identically in these limits.
Concluding remarks
==================
In this paper we have presented a covariant geometric approach for examining the variation of geometric models of lipid membranes. For concreteness, we have restricted our attention to the first and second variation of the rigid bilayer couple hamiltonian (\[eq:hami\]). We note that it is straightforward to specialize the second variation in the form (\[eq:oper\]) to axisymmetric configurations. Moreover, our approach can be extended straightforwardly to other geometric models. However, computational difficulties are to be expected if the membrane has a free edge and it is no longer legitimate to throw away boundary terms. When considering higher order variations, a different difficulty arises. It is no longer justified to neglect tangential deformations. We will address this issue in a future publication.
[**Acknowledgements**]{}
RC is partially supported by CONACyT under grant 32187-E. JG and JAS are partially supported by DGAPA at UNAM under grant IN119799.
[99]{}
Canham P 1970 [*J. Theor. Biol.*]{} [**26**]{} 61
Helfrich W 1973 [*Z. Naturforsch.*]{} [**C28**]{} 693
Evans E 1974 [*Biophys. J.* ]{} [**14**]{} 923; 1983 [*ibid.*]{} [**43**]{} 27; 1985 [*ibid.*]{} [**48**]{} 175
Svetina S and Zeks B 1989 [*Eur. Biophys. J.*]{} [**17**]{} 101
Nelson D, Piran T and Weinberg S eds 1989 [*Statistical Mechanics of Membranes and Surfaces*]{} vol. 5 (Proceedings of the Jerusalem Winter School for Theoretical Physics) (Singapore: World Scientific)
Safran S 1994 [*Statistical Physics of Surfaces, Interfaces, and Membranes*]{} (Reading UK: Addison Wesley)
Peliti L 1996 in [*Fluctuating Geometries in Statistical Physics and Field Theory (Les Houches (1994)*]{}, eds. F. David, P. Ginsparg and J. Zinn-Justin (Elsevier Science B.V.)
Svetina S and Zeks B 1996 in [*Nonmedical Applications of Liposomes*]{}, eds. D.D. Lasic and Y. Barenholz (CRC: Boca Raton, FL)
Seifert U 1997 [*Adv. in Phys.*]{} [**46**]{} 13
Zhong-Can O Y and Helfrich W 1987 [*Phys. Rev. Lett.*]{} [**59**]{} 2486
Peterson M A 1985 [*Mol. Cryst. Liq. Cryst.*]{} [**127**]{} 257
Zhong-Can O Y and Helfrich W 1989 [*Phys. Rev. A*]{} [**39**]{} 5280
Peterson M A 1989 [*Phys. Rev A*]{} [**39**]{} 2643
Bukman D J, Yao Jian Hua, Wortis M 1996 [*Phys. Rev. E*]{} [**54**]{} 5463
Heinrich V, Sevsek F, Svetina S and Zeks B 1997 [*Phys. Rev. E*]{} [**55**]{} 1809
Förster D 1986 [*Phys. Lett. A*]{} [**114**]{} 115
Kleinert H 1986 [*Phys. Lett. A*]{} [**114**]{} 263
Peliti A and Leibler S 1985 [*Phys. Rev. Lett.*]{} [**54**]{} 1690
David F in [@Nel.Pir.Wei:89]
Rogers C and Schief W K 2002 [*Backlund and Darboux Transformations*]{} (Cambridge: Cambridge University Press)
Spivak M 1979 [*A Comprehensive Introduction to Differential Geometry. Vol. Four, Second Edition*]{} (Houston: Publish or Perish Inc.)
Goetz R and Helfrich W 1996 [*J. Phys. II*]{} [**6**]{} 215
Fournier J B and Galatola P 1997 [*Europhys. Lett.*]{} [**39**]{} 225
Wald R 1984 [*General Relativity*]{} (Chicago Univ. Press)
Bozic B, Svetina S, Zĕks B and Wuagh R 1992 [*Biophys. J.*]{} [**61**]{} 963
Miao L, Seifert U, Wortis M and Döbereiner H G 1993 [*Phys. Rev. E*]{} [**49**]{} 5389
Capovilla R, Guven J and Santiago J A 2002 [*Phys. Rev. E*]{} [**66**]{} 021607
Capovilla R and Guven J 2002 [*J. Math. Phys. A: Math. and General*]{} [**35**]{} 6233
Lawson B H 1980 [*Lectures on minimal submanifolds*]{} (Houston: Publish or Perish Inc.)
|
{
"pile_set_name": "ArXiv"
}
|
\#1[[**[\#1]{}**]{}]{} \#1[[**[\#1]{}**]{}]{} ==łł==\#1\#2 \#1\#2[[\^2[\#1]{}\^2]{}]{} \#1\#2
[**SHORTCUT METHOD OF SOLUTION OF\
GEODESIC EQUATIONS\
FOR SCHWARZSCHILD BLACK HOLE**]{}\
Département d’Astrophysique Relativiste et de Cosmologie\
(UPR 176 du C.N.R.S.)\
Observatoire de Paris - Section de Meudon\
F-92195 Meudon Cedex, France\
[*e-mail*]{} : [*[email protected]*]{} To appear in [*Classical and Quantum Gravity*]{}
**Abstract**
> It is shown how the use of the Kerr-Schild coordinate system can greatly simplify the formulation of the geodesic equation of the Schwarzschild solution. An application of this formulation to the numerical computation of the aspect of a non-rotating black hole is presented. The generalization to the case of the Kerr solution is presented too.
> PACS numbers: 0230H, 0470B
Introduction. {#intro}
=============
Since the publication of the Schwarzschild static solution solution [@sch16], it has been well known that the geodesic equation in this space-time can be solved analyticaly. The purpose of this note is to show that, by the mean of the Eddington coordinate system, the equation of motion of test particles in the Schwarzschild space-time can be greatly simplified so as to provide a more efficient method of solution, using ideas suggested by analysis of the less simple case of the Kerr rotating black hole solution.
The forms of the spherical Schwarzschild metric as expressed in term of the usual Schwarzschild coordinate system and in term of the less usual Eddington [@edi24] coordinate system (which is interpretable as a limiting case of the Kerr-Schild [@ks] coordinate system for a rotating black hole) are presented in § 2. Taking advantage of the obvious constants of motion which appear in the Schwarzschild space-time, we write down the geodesic equation in term of the Eddington coordinate system as second order equations with respect to some affine parameter.
These results will be useful for many problems for which geodesic motion is concerned. In particular, they can greatly simplify the numerical integration of the test particles trajectory. As an application, we present in § 3 the apparent shape of a non-rotating Black Hole surrounded by an accretion disc as seen by an observer who is travelling towards and ultimately through the event horizon.
The Schwarzschild metric and the geodesic equation.
===================================================
Using the standard Schwarzschild cordinate system $( t , r , \theta , \varphi )$, the line element of the metric takes the form
ds\^2 = - (1 -[2M r]{} ) dt\^2 + [1 (1 - ) ]{} dr\^2 + r\^2 ( d\^2 + \^2 d\^2) in units such that $c=G=1$, where $M$ is the mass of the Black Hole.
The first integrals of the equations of motion are well known to be expressible as
$$\begin{aligned}
\dot t &= &{E \over (1 - {2M \over r} ) } \\ \nonumber
r^4 \dot r^2 &= &E^2 r^4 - (r^2 - 2Mr) (\mu^2 r^2 + K) \\ \nonumber
r^4 \dot \theta^2 &= &K - {\Phi^2 \over \sin^2 \theta} \\ \nonumber
\dot \varphi &= &{\Phi \over r^2 \sin^2 \theta}\end{aligned}$$
where a dot denotes differenciation with respect to some affine parameter, $\tau$ say, and where the constants of motion $\mu^2$, $E$ and $\Phi$ are respectively the rest-mass, the energy and the angular momentum about the axis $\sin\theta = 0$ of the test-particle, $K$ being the Carter’s fourth constant of motion [@car68] which reduces in this simple spherical case to the square of the total angular momentum of the orbiting particle.
Introducing the Eddington coordinate system $( T , x , y , z )$, [*i.e.*]{} the Kerr-Schild coordinate system for the limiting case of a non rotating black hole, where $( x , y , z )$ are [*“cartesian-like”*]{} spatial coordinates and where $T$ is a retarded time, which are related to the Schwarzschild cordinates by means of
x = r ; y = r ; z = r and T = u-r ; du = dt + [1 1 - ]{} dr , the metric takes the (Kerr-Schild) form ds\^2 = - dT\^2 + dx\^2 + dy\^2 + dz\^2 + [2M r\^3]{} ( xdx + ydy + zdz + rdT )\^2 where r\^2 = x\^2 + y\^2 + z\^2 . At this stage, it is to be noticed that the above coordinate transformation (first published by Eddington in 1924 [@edi24]) solves the [*“Schwarzschild singularity”*]{} $(r=2M)$ problem in the sense of being regular accross the event horizon $r = 2 M$.
In terms of this system, it is straightforward to show that the geodesic equations reduce to
where $x^i := (x,y,z)$.
Keeping in mind that $\vert \dot r \vert = E$ when any test-particle reaches the event horizon, the above equations show clearly the irreversible nature of the Schwarzschild Black Hole. Moreover, unlike what is obtained in terms of the Schwarzschild cordinates system, for which one has to take care of the axis $\sin\theta = 0$, of the sign of $\dot r$ and $\dot \theta$ and of the presence of the event horizon, such differential equations can be numericaly solved in a straightforward way in order to determine the path of any test-particle.
Apparent shape of a non-rotating Black Hole.
============================================
Accretion disc are currently supposed to play an important role in several astrophysical situations, especially when high-energy phenomena are involved. Because of the astrophysical interest of such objects, several authors have computed the apparent shape of of black hole with thin accretion disc as seen from infinity ([@bar73] [@cun73] [@lum79]). The main purpose of these calculation was to simulate line profiles which are commonly observed from astrophysical sources which are generally interpreted as emission from an accretion disc around a compact object which may be a black hole. Our present purpose is quite different in the sense that, as an application of the previous formulæ, we will present the apparent shape of a black hole surrounded by a thin accretion disc as seen by an observer who is flying near the hole.
We will assume that the disc is a stationnary Keplerian one, orbiting in the plane $z=0$. Each particle of the disc then follows a circular orbit with angular velocity
:= [ddt]{} = .
Following Page and Thorne [@pag74], the flux of radiation from the surface of the disc is given by
\[f\_e\] F\_e = [3 M M 8]{} [1 (\^2 - 3) \^5]{} { - + [ 2]{} ( (3-2) [+ - ]{} ) } where $\dot M$ is the accretion rate and where we have introduced
= . As shown by Ellis [@ell71], the observed bolometric flux $F_o$ is given by
\[r1\] F\_o = [F\_e (1+z)\^4]{} where the redshift factor $(1+z)$ is given by the ratio of the two scalar products
\[r2\] (1+z) = [ [p\_w\^]{} \_[emission]{} \_[reception]{} ]{} , , [**w**]{} and [**u**]{} being the four-velocities of the photon, the emitting particle and the observer respectively. This redshift factor consists of a gravitational part due to the gravitational field of the black hole (which is measurable only in the closed vicinity of the hole), a Doppler part due to the rotation of the disc (the dominant one) and a Doppler part due to the motion of the observer.
It is now straightforward to compute how the black hole will look. Assuming that the observer takes a camera with him, the coordinates of each pixel of the plate of the camera allows to compute the components of the four-velocity of the photon that reaches that pixel (recall that the four-velocity of a photon has only two true degrees of freedom). The four velocity of each photon reaching the eyes of the observer can be obtained in terms of an orthonormal parallel-propagated frame $( \underline\lambda_0\ ,\ \underline\lambda_1\ ,\ \underline\lambda_2\ ,\
\underline\lambda_3 )$ along the observer’s trajectory as p\^= \_0\^ + \_2\^ + \_1\^ + \_3\^ , where $\vartheta$ and $\varphi$ are two angles describing the photographic plate in the rest frame of the observer.
A direct numerical integration of equations (\[geo1\]) (\[geo2\]) in the special case $\mu^2 = 0$ for negative values of the affine parameter $\tau$ gives the history of the photon and then the luminosity of the source of its origin. Finally, application of the correction factor $(1+z)^4$ as given by equations (\[r1\]) and (\[r2\]) gives the brightness of this fiducial pixel.
The figures represent eight simulated photographs obtained by this procedure. These “snapshots” have been computed at succesive steps during the flight of an observer who is on free fall along a parabolic orbit ($\mu^2=1\ ,\ E=1,\ K=12)$ of the Schwarzschild space-time (see figure 1). On the first snapshop, the observer is above the disc at a distant $r=1280 M$. The observer then crosses the disc (snapshot \# 4, $r=39 M$) and, finally, goes into the black hole (snapshot \#8, $r = 0.7 M$). During this trip, the observer directs his camera in the direction of his motion except on the last snapshot for which the observer directs his camera toward the exterior of the hole.
The brightness of the pictures has been computed by means of formul[æ]{} (\[f\_e\]) and (\[r1\]) while the coloration has been added artificially. The apparent position of distant stars has also been computed. This helps to show the gravitational lense effect which is more conspicuous on snapshot \# 2 ($r = 121 M$): some distant stars, which are more or less perfectly aligned with the observer–black hole system, appear as pieces of rings in the sky. The opacity of the disc has been extremely minimized in order to make more visible the second (and even third) image of the disc and the distant stars. Moreover, this allows to show both the apparence of the upper and down sides of the disc.
The trajectories of the photons and the link between these trajectories and the apparent shape of the hole as seen by the observer are schematically given in figure 2. Examination of figure 2 helps to understand the apparent shape of the disc. Some of the photons emitted by the disc are strongly deflected by the gravitational field (not the curvature) before reaching the eyes of the observer. Hence, the observer is able to see the upper rear side of the disc which appears like surrouding the top of the black hole ([*primary image*]{} on figure 2). In the same way, the observer can see the down rear side of the disc which seems to surround the bottom of the black hole. Now, some photons whose impact parameter is very close to the capturing one make one (even two) turn(s) around the black hole before reaching the eyes of the observer. This leads to the formation of a second (third) image of the disc.
The numerical scheme used in these calculation is of second order. The integration has been done where respect to the parameter $\lambda$ which is related to the affine parameter by d= [dr\^2]{} and which is well suited to the form of equations (\[geo1\]–\[geo2\]) in the sense that it exploits the fact that, far from the hole, null geodesics are straight lines. The code is fully parallelized. Vectorisation is also possible but less efficient since the number and the kind of operations needed to compute the value of each pixel of the screen vary from a pixel to another one.
Conclusion.
===========
We have shown that the use of the Eddington coordinates system in the Schwarzschild space-time facilitates the description of the geodesical motion in that space-time. This formulation is well adapted to direct numerical integration because it avoid the usual troubles tied to the [*spherical–type*]{} coordinates system and the trouble tied to the [*pseudo–singularity*]{} $r = 2M$.
Such a formulation can be extended to the non-spherical case of the Kerr black hole solution. One can show that, in the Kerr-Schild coordinate system, the equations of motion of a test particle of zero rest-mass can be written in the form:
$$\begin{aligned}
\Sigma^3 {\ddot x \over M} &= %% Xpp
&- 4ar {Q\over\Delta} \Sigma \dot y \\ \nonumber
& &+ (\Sigma - 4r^2) \sin\theta\cos\psi
\left\{K-\left(a{Q\over\Delta}\right)^2\right\} \\ \nonumber
& &- ar\sin\theta\sin\psi {Q\over\Delta}
\left\{4(E\Sigma-Q) + (4a^2-\Sigma){Q\over\Delta}\right\} \\
\Sigma^3 {\ddot y \over M} &= %% Ypp
&+ 4ar {Q\over\Delta} \Sigma \dot x \\ \nonumber
& &+ (\Sigma - 4r^2) \sin\theta\sin\psi
\left\{K-\left(a{Q\over\Delta}\right)^2\right\} \\ \nonumber
& &+ ar\sin\theta\cos\psi {Q\over\Delta}
\left\{4(E\Sigma-Q) + (4a^2-\Sigma){Q\over\Delta}\right\} \\
\Sigma^3 {\ddot z \over M} &= %% Zpp
& - K \cos\theta (3r^2 -a^2\cos^2\theta) \\
\dot T &= & { 2 M K r \over \Sigma ( {\cal E} - \Sigma \dot r)}
+ E %% Tp\end{aligned}$$
where we have introduced the quantities
$$\begin{aligned}
\Sigma &= &r^2 + a^2\cos^2\theta \ , \\
{\cal E} &= &E(r^2 + a^2)^2 - a\Phi \ , \\
Q &= &\Sigma \dot r + {\cal E}\end{aligned}$$
These results have been used in the study of profiles and shifts of lines emitted from Keplerian accretion disc around in X-ray binaries containing a neutron star or a black hole [@ham94] and to the study of microlensing effects in active galactic nuclei [@jar94].
**Acknowledgments**
I wish to thank B. Carter, for helpful discussions. I am also grateful to [*Silicon Graphics France*]{} for the loan of computers.
[22]{}
Bardeen, J.M., 1973, in [*Black Holes*]{}, Les Houches, ed. De Witt and De Witt, Gordon and Breach, New-York.
Carter, B., 1968, [*Phys. Rev.*]{}, [**174**]{}, 1559.
Cunningham, C.T., Bardeen, J., 1973, [*Ap. J.*]{}, [**183**]{}, 237.
Eddington, A.S., 1924, [*Nature*]{}, [**113**]{}, 192.
Ellis, G.F.R., 1971, Relativistic Cosmology in [*General Relativity and Cosmology*]{}, ed. R. Sachs, Academic Press, New-York.
Hameury, J.M., Marck, J.A., Pelat, D., 1994, [*A[&]{}A*]{}, [**287**]{}, 795.
Jaroszyński, M., Marck, J.A., 1994, [*A[&]{}A*]{}, [**291**]{}, 731.
Kerr, R.P., Schild, A., 1967, A new class of vacuum solutions of the Einstein field equations. In: Barbera G. (ed) Publicazioni del comitato nazionale per le manifestazioni celebrative del IV centenario della nascita di Galileo Galilei, Vol. II, tome 1, Firenze
Luminet, J.P., 1979, [*A[&]{}A*]{}, [**75**]{}, 228.
Page,D.N., Thorne,K.S., 1974, [*Ap. J.*]{}, [**191**]{}, 499.
Schwarzschild, K., 1916, [*Deut. Akad. Wiss. Berlin, Kl. Math.-Phys. Tech.*]{}, 189
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- '**Kimet Jusufi[^1]**'
title: 'Gravitational lensing by Reissner-Nordström black holes with topological defects '
---
*Department of Physics, State University of Tetovo, Ilinden Street nn, 1200, Macedonia*
0.5 truecm
Introduction
============
One of the most important predictions of Einstein’s General Theory of Relativity is the deflection of light rays from a distant source passing close to a massive body. This phenomena is crucial and played a key role as a first experimental verification of general relativity.
Besides the standard methods used for calculating the deflection angle, recently, a new spin forward was put by Gibbons and Werner [@gibbons] by introducing the optical geometry and applying the Gauss-Bonnet theorem to the optical geometry. Using this approach, one can show the importannce of topology in gravitational lensing, stated in other words the focusing of the light rays can be viewed globaly as topological effect of the spacetime. Along this line of research a number of static and spherically symmetric spacetime metrics have been studied [@gibbons; @gibbons1]. Recently, this method was extended to stationary metrics for the Kerr black hole solution [@werner].
Topological defects, such as domain walls, cosmic strings and monopoles, may have been produced by the phase transition in the early universe due to the breakdown of local or global gauge symmetries [@kible]. Cosmic strings are one dimensional object, characterized by the tension $G\mu$, where $G$ is the Newton’s gravitational constant and $\mu$ is the mass per unit length of the string. It is interesting to notice that athought globally the spacetime around a cosmic string is conical, localy the spacetime is just Minkowski spacetime. One way to think about this spacetime is to consider a Minkowski spacetime with a wedge removed given by the deficit angle $\delta=8\pi G \mu $. Note that, in the case of a global monopole the spacetime is not locally flat. However, globally the spacetime of the surface $\theta=\pi/2$, is conical, with deficit angle $\Delta=8\pi^{2}G\eta^{2}$ [@birola].
In this context, is important to ask whether there is a chance that may lead to a possible experimental detection of topological defects. There are however two classes of effects that may lead to indirect experimental evidence: the quantum effects and gravitational effects. Among these effects, a cosmic string can act as a gravitational lens [@gott], it can induce a finite electrostatic self-force on an electric charged particle [@linet], shifts in the energy levels of a hydrogen atom [@bezerra], they were also suggested as an explanation of the anisotropy of the cosmic microwave background radiation, and many others.
The aim of this paper is to use Gauss-Bonnet theorem and calculate the deflection angle around the spacetime of the Reissner-Nordström black holes in the background with a global monopole and a cosmic string. It is widely believed that charged black holes don’t exist in nature for a long time, since any initial charge will rather quickly be neutralized. However, as wa discussed here [@punsley], charged black holes may be produced during the gravitational collapse of massive magnetized rotating stars surrounded by a co-rotating magnetosphere of equal and opposite charge. The magnetosphere preserves the black hole from a neutralization due to a selective accretion of charge from the environment and the black hole can be quite stable in a typical astrophysical environment of low density. Gravitational lensing from charged black holes in the strong field scenario has been discussed by [@eiroa], in the case of the weak limit the deflection angle was also studied by [@sereno], using Fermat’s principle.
This paper is organized as follows. In Section 2, we briefly review the Gauss-Bonnet theorem and then calculate the corresponding Gaussian optical curvature for Reissner-Nordström black hole. In Section 3, we calculate the leading terms of the deflection angle by Reissner-Nordström black hole with topological defects. In Section 4, we comment on our results. We use metric signature $ (-,+,+,+) $ and $c=G=1$. The Latin indices are used to donate spatial coordinates and Greek indices for spacetime coordinates.
Reissner-Nordström optical metric with topological defects {#sec:1}
==========================================================
We can start by writing the simplest Lagrangian density, which describes a global monopole given by $$\mathcal{L}=\frac{1}{2}g^{\mu \nu}\partial_{\mu}\phi^{a}\partial_{\nu}\phi_{a}-\frac{1}{4}\lambda (\phi^{a}\phi_{a}-\eta^{2})^{2},$$ where $\lambda$ is the self-interaction term and $\eta$ is scale of gauge-symmetry breaking $\eta\sim 10^{16} $ GeV, $\phi^{a}$ is a triplet of scalar fields which transform under the group $O(3)$, whose symmetry is spontaneously broken to $U(1)$ given by $$\phi^{a}=\eta h(r)\frac{x^{a}}{r}$$ with $x^{a}x^{a}=r^{2}$. The most general static metric with spherical symmetry can be written as $$\mathrm{d}s^{2}=-A(r)\mathrm{d}t^{2}+A^{-1}(r)\mathrm{d}r^{2}+r^{2}\left(\mathrm{d}\theta^{2}+\sin^{2}\theta \,\mathrm{d}\varphi^{2} \right).\label{metric1}$$
Solving the Einstein field equations in spacetime with a global monopole [@mello], leads to the following expression for $A(r)$ $$A=1-8\pi \eta^{2}-\frac{2M}{r}+\frac{Q^{2}}{r^{2}},$$ where $M$ is the black hole mass and $Q$ black hole charge. By introducing the following coordinate transfromations $r\to(1-8\pi\eta^{2})^{-1/2}r$, $t\to(1-8\pi\eta^{2})^{1/2}t$ and $M\to(1-8 \pi \eta^{2})^{-3/2}M$, $Q\to (1-8\pi \eta^{2})Q$ [@vilenkin], and also by introducing a cosmic string in the metric , which can be done by using $\varphi\to(1-4\mu)\varphi$, we end up with the following Reissner-Nordström black hole metric in spherical coordinates given by $$\mathrm{d}s^{2}=-\left(1-\frac{2M}{r}+\frac{Q^{2}}{r^{2}}\right)\mathrm{d}t^{2}+\left(1-\frac{2M}{r}+\frac{Q^{2}}{r^{2}}\right)^{-1}\mathrm{d}r^{2}+a^{2}r^{2}\left(\mathrm{d}\theta^{2}+p^{2}\sin^{2}\theta \mathrm{d}\varphi^{2}\right),
\label{rn}$$ where the terms $a^{2}=1-8\pi\eta^{2}$ and $p^{2}=(1-4\mu)^{2}$, encodes the presence of a global monopole and a cosmic string, respectively. The metric , describes a static and noninteracting infinitely long cosmic string and a global monopole placed close to each other in Reissner-Nordström black hole spacetime. An infinitely long cosmic string, is aligned through $\theta=0$ and $\theta=\pi/2$, parallel to $z$-axes, passing through the Reissner-Nordström black hole spacetime. Without loss of generality, we can consider the equatorial plane $\theta=\pi/2$, in this way solving for null geodesics with $\mathrm{d}s^{2}=0$, the optical line element becomes $$\begin{aligned}
\mathrm{d}t^{2}=\frac{\mathrm{d}r^{2}}{\left(1-\frac{2M}{r}+\frac{Q^{2}}{r^{2}}\right)^{2}}+\frac{(1-8\pi \eta^{2})(1-4\mu)^{2}r^{2}\mathrm{d}\varphi^{2}}{1-\frac{2M}{r}+\frac{Q^{2}}{r^{2}}}.
\label{metric2}\end{aligned}$$
For convenience, we can write the optical metric $\tilde{g}_{ab}$ in terms of the new coordinate $r^{\star}$ as $$\mathrm{d}t^{2}= \tilde{g}_{ab}\,\mathrm{d}x^{a}\mathrm{d}x^{b}=\mathrm{d}{r^{\star}}^{2}+f^{2}(r^{\star})\mathrm{d}\varphi^{2},
\label{optical}$$ where the function $f(r^{\star})$ is given by $$f(r^{\star})=\frac{\sqrt{1-8\pi\eta^{2}}(1-4\mu)\,r}{\sqrt{1-\frac{2M}{r}+\frac{Q^{2}}{r^{2}}}}.\label{f}$$
The corresponding optical metric from , therefore reads $$\tilde{g}_{ab}=\begin{bmatrix}
1 & 0 \\
0 & f^{2}(r^{\star}) \\
\end{bmatrix},$$ with the determinatnt $\det \tilde{g}_{ab}=f^{2}(r^{\star})$. Using the optical metric , it is not difficult to find that the only nonvanishing Christoffel symboles are $\Gamma^{r}_{\varphi\varphi}=-f(r^{\star})f'(r^{\star})$ and $\Gamma^{\varphi}_{r\varphi}=f'(r^{\star})/f(r^{\star})$. One can now proceed to calculate the corresponding Gaussian optical curvature $K$ as an intrinsic measure of curvature. In the case of 2-dimensional surface, this is not difficult since the only nonvanishing component of the Riemann tensor for the optical curvature is given by [@gibbons] $$R_{r \varphi r \varphi}=K\left(\tilde{g}_{r \varphi}\tilde{g}_{\varphi r}-\tilde{g}_{rr}\tilde{g}_{\varphi \varphi}\right)=-K\det \tilde{g}_{r\varphi}$$ where $R_{r\varphi r \varphi}=\tilde{g}_{r r}\,R^{r}_{\varphi r \varphi}$. It follows that the Gaussian optical curvature is given by $$K=-\frac{R_{r\varphi r\varphi}}{\det \tilde{g}_{r \varphi}}=-\frac{1}{f(r^{\star})}\frac{\mathrm{d}^{2}f(r^{\star})}{\mathrm{d}{r^{\star}}^{2}}.$$
Using the last equation, one can show that the intrinsic Gaussian optical curvature $K$, can be expressed in terms of $r$ as [@bao] $$\begin{aligned}
\label{Gcurvature}
K&=&-\frac{1}{f(r^{\star})}\frac{\mathrm{d}^{2}f(r^{\star})}{\mathrm{d}{r^{\star}}^{2}}=-\frac{1}{f(r^{\star})}\left[\frac{\mathrm{d}r}{\mathrm{d}r^{\star}}\frac{\mathrm{d}}{\mathrm{d}r}\left(\frac{\mathrm{d}r}{\mathrm{d}r^{\star}}\right)\frac{\mathrm{d}f}{\mathrm{d}r}+\left(\frac{\mathrm{d}r}{\mathrm{d}r^{\star}}\right)^{2}\frac{\mathrm{d}^{2}f}{\mathrm{d}r^{2}}\right].\end{aligned}$$
Finally, we can now compute the corresponding Gaussian optical curvature for the Reissner-Nordström black hole in presence of topological defects by substituting the equation into , and find the following expression $$K=-\frac{2M}{r^{3}}\left(1-\frac{3M}{2r}\right)+\frac{3Q^{2}}{r^{4}}\left(1+\frac{2Q^{2}}{3r^{2}}\right)-\frac{6MQ^{2}}{r^{5}}.
\label{curvature}$$
Now it’s interesting to see that Gaussian optical curvature $K$ seems to be independent of topological defects, but as we will see in the next section, there is a contribution due to the topology of spacetime which is globally conical. The other interesting point about the last equation which may come as a surprise is that, $K$ is negative, which implies that locally light rays should diverge. As was pointed out by [@gibbons], the negative Gaussian curvature of the optical metric is a rather general feature of black hole metrics, and the interesting thing here is that, only globally, by considering the topology of spacetime, light rays can converge. As we will see, this can be done by using the Gauss-Bonnet theorem.
Deflection angle by Reissner-Nordström black hole with topological defects
==========================================================================
Having said that, let us now write the Gauss-Bonnet theorem, which relates the intrinsic geometry of the spacetime with its topology, for the region $D_ {R} $ in $M$, with boundary $\partial D_{R}=\gamma_{\tilde{g}}\cup C_ {R} $ which states that (see, e.g.,[@gibbons]) $$\int\limits_{D_{R}}K\,\mathrm{d}S+\oint\limits_{\partial D_{R}}\kappa\,\mathrm{d}t+\sum_{i}\epsilon_{i}=2\pi\chi(D_{R}),$$ where $K$ is the Gaussian curvature, $\kappa$ the geodesic curvature, given by $ \kappa=\tilde{g}\,(\nabla_{\dot{\gamma}}\dot{\gamma}, \ddot{\gamma})$, such that $\tilde{g}(\dot{\gamma}, \dot{\gamma})=1$, with the unit acceleration vector $\ddot{\gamma}$ and $\epsilon_{i}$ the corresponding exterior angle at the $i$th vertex. As $R\to \infty$, both jump angles become $\pi/2$, in other words $\theta_{O}+\theta_{S}\to \pi$. Since $D_ {R} $ is non-singular, than the Euler characteristic is $\chi(D_{R})=1$, finally we are left with $$\iint\limits_{D_{R}}K\,\mathrm{d}S+\oint\limits_{\partial D_{R}}\kappa\,\mathrm{d}t=2\pi\chi(D_{R})-(\theta_{O}+\theta_{S})=\pi.
\label{gaussbonnet}$$
Since $\gamma_{\tilde{g}}$ is geodesic, clearly than $\kappa(\gamma_{\tilde{g}})=0$, the only interesting part to be calculated remains $\kappa(C_{R})=|\nabla_{\dot{C}_{R}}\dot{C}_{R}|$ as $R\to \infty$. It should be noted that, unlike the case of optical Schwarzschild metric which is asymptotically Euclidean, i.e. $\kappa(C_{R})\mathrm{d}t/\mathrm{d}\varphi=1$, this is not valilid if topological defects are introduced. To see this, let’s calculate at very large $R$, given by $C_{R}:= r(\varphi)=R=const.$, the radial component of the geodesic curvature $$\left(\nabla_{\dot{C}_{R}}\dot{C}_{R}\right)^{r}=\dot{C}_{R}^{\varphi}\,\partial_{\varphi}\dot{C}_{R}^{r}+\Gamma^{r}_{\varphi \varphi}\left(\dot{C}_{R}^{\varphi}\right)^{2},$$ we can use the fact that $\tilde{g}_{\varphi \varphi}\,\dot{C}_{R}^{\varphi}\dot{C}_{R}^{\varphi}=1$, which implies $\dot{C}_{R}^{\varphi}=1/f^{2}(r^{\star})$, and recall that $\Gamma^{r}_{\varphi\varphi}=-f(r^{\star})f'(r^{\star})$, yielding $$\left(\nabla_{\dot{C}_{R}^{r}}\dot{C}_{R}^{r}\right)^{r}=-\frac{f'(r^{\star})}{f(r^{\star})}\to-\frac{1}{R}.$$
We can now use the last result and compute the geodesic curvature $$\kappa(C_{R})=|\nabla_{\dot{C}_{R}}\dot{C}_{R}|=\left(\tilde{g}_{rr}\dot{C}^{r}_{R}\dot{C}^{r}_{R}\right)^\frac{1}{2}\to\frac{1}{R}.$$
At very large $r(\varphi)=R=const.$, it follows that the geodesic curvature is indipendent of topological defects, $\kappa(C_{R})\to R^{-1}$, however from the optical metric , it’s not difficult to see that $\mathrm{d}t=(1-4\mu)\sqrt{1-8\pi\eta^{2}}\,R \,\mathrm{d}\,\varphi$, and hence $$\kappa(C_{R})\mathrm{d}t=\frac{1}{R}(1-4\mu)\sqrt{1-8\pi\eta^{2}}\,R \,\mathrm{d}\,\varphi.$$
Unlike the case of black hole configuration without topological defects, here the optical metric is not asymptotically Euclidian, since the last result leads to $\kappa(C_{R})\mathrm{d}t/\mathrm{d}\varphi=(1-4\mu)\sqrt{1-8\pi\eta^{2}}\neq 1$, which is satisfied only when $\mu=\eta=0$. Now, we could go back to , and substitute this result, yielding $$\begin{aligned}
\label{def1}
\iint\limits_{D_{R}}K\,\mathrm{d}S&+&\oint\limits_{C_{R}}\kappa\,\mathrm{d}t\overset{{R\to \infty}}{=}\iint\limits_{S_{\infty}}K\,\mathrm{d}S+(1-4\mu)\sqrt{1-8\pi\eta^{2}} \int\limits_{0}^{\pi+\hat{\alpha}}\mathrm{d}\varphi.\end{aligned}$$
In the weak deflection limit we may assume that the light ray is given by $r(t)=b/\sin\varphi $ at zeroth order, using and it follows that the deflection angle is given by $$\hat{\alpha}=\frac{\pi-\pi(1-4\mu)\sqrt{1-8\pi\eta^{2}}}{(1-4\mu)\sqrt{1-8\pi\eta^{2}}}-\frac{1}{(1-4\mu)\sqrt{1-8\pi\eta^{2}}}\int\limits_{0}^{\pi}\int\limits_{\frac{b}{\sin \varphi}}^{\infty}K\,\sqrt{\det \bar{g}}\,\mathrm{d}r\,\mathrm{d}\varphi .
\label{angle}$$ where $\mathrm{d}S=\sqrt{\det \tilde{g}}\,\mathrm{d}r\,\mathrm{d}\varphi$, is the optical surface area and $\sqrt{\det\tilde{g}}=(1-4\mu)\sqrt{1-8\pi\eta^{2}} \,r$. The first term can be approximated as $$\frac{\pi-\pi(1-4\mu)\sqrt{1-8\pi\eta^{2}}}{(1-4\mu)\sqrt{1-8\pi\eta^{2}}}\simeq 4\mu \pi+4\pi^{2}\eta^{2}.$$
Substituting the leading terms of the Gaussian curvature into the last equation we find $$\begin{aligned}
\nonumber
\int\limits_{0}^{\pi}\int\limits_{\frac{b}{\sin \varphi}}^{\infty}K\mathrm{d}S &\approx &\int\limits_{0}^{\pi}\int\limits_{\frac{b}{\sin \varphi}}^{\infty}\left(-\frac{2M}{r^{3}}+\frac{3Q^{2}}{r^{4}}\right)\sqrt{\det \tilde{g}}\mathrm{d}r\mathrm{d}\varphi\\\nonumber
&\approx &\sqrt{1-8\pi\eta^{2}}}{(1-4\mu)\left(-\frac{4M}{b}+\frac{3\pi Q^{2}}{4b^{2}}\right)
\label{angle}\end{aligned}$$ where the impact parameter $b>>2M$. Putting all together, we end up with the expression for the deflection angle given by $$\hat{\alpha}\simeq 4\mu \pi+4\pi^{2}\eta^{2}+\frac{4M}{b}-\frac{3\pi Q^{2}}{4b^{2}}.\label{deflection}$$
We have therefore found that, the total deflection angle by Reissner-Nordström black hole in background with topological defetcs, actually increses due to the presence of the first and second term. The first term gives the deflection angle due to the presence of a cosmic string, the second term gives the deflection angle due to the presence of a global monpole, the third term is the well known result for Schwarzschild black hole, and finally, the last term is coming due to the charge nature of the black hole. The light signal is propagating from the source $S$ to an observer $O$, such that both $S$ and $O$ lie on the same surface $\theta=\pi/2$. If $S$, $O$, global monopole and a cosmic string are perfectly aligned, it follows $$\hat{\alpha}\simeq\left( 4\mu \pi+4\pi^{2}\eta^{2}\right)l(d+l)^{-1}+\frac{4M}{b}-\frac{3\pi Q^{2}}{4b^{2}},$$ where $d$ and $l$ are the corresponding distances from the monopole/cosmic string to the observer $O$, and to the source $S$, respectively. It should be noted that the increased value of the deflecton angle is rather small, of the order of $10$ arcsec, since $G\mu \simeq 10^{-6}$ and $G\eta^{2}\simeq 10^{-5}$, which makes the possible detection difficult, however as pointed out by [@vilenkin], this effect is in the observable range. It is widely believed that global monopoles should move with some velocity $v$, in this particular case, we assume that a whole system consisted by a cosmic string and a global monopole is moving with some relativistic speed $v$, parallel to $z$-axes, with respect to the obsorver $O$. If this system pierces the Reissner-Nordström spacetime, than the deflection angle by using the same arguments as [@vilenkin], can be written as $$\hat{\alpha}_{1}=\gamma^{-1}\left(1-\textbf{n}\textbf{v}\right)^{-1}\hat{\alpha}_{0}+\frac{4M}{b}-\frac{3\pi Q^{2}}{4b^{2}}$$ here $\textbf{n}$ is the unit vector along the line of sight, $\gamma=(1-v^{2})^{-1/2}$ and $\hat{\alpha}_{0}=\left( 4\mu \pi+4\pi^{2}\eta^{2}\right)l(d+l)^{-1}$. Now let us consider and discuss some special cases. Setting $Q=0$, from we recover the Schwarzschild black hole with topological defects $$\mathrm{d}s^{2}=-\left(1-\frac{2M}{r}\right)\mathrm{d}t^{2}+\left(1-\frac{2M}{r}\right)^{-1}\mathrm{d}r^{2}+a^{2}r^{2}\left(\mathrm{d}\theta^{2}+p^{2}\sin^{2}\theta \mathrm{d}\varphi^{2}\right).$$
This metric has the same characteristics as metric , with the corresponding Gaussian optical curvature $K=-2M (1-3M/2r)/r^{3}$. It also follows directly from the deflection angle $\hat{\alpha}\simeq 4\mu \pi+4\pi^{2}\eta^{2}+4M/b$. On the other hand if $Q=M=\mu=0$, we recover the Minkowski spacetime pirced by a global monopole given by $$\mathrm{d}s^{2}=-\mathrm{d}t^{2}+\mathrm{d}r^{2}+(1-8\pi \eta^{2})r^{2}\left(\mathrm{d}\theta^{2}+\sin^{2}\theta \mathrm{d}\varphi^{2}\right).$$
The characteristics of this metric are well known, it describes a spacetime with a deficit solid angle, the area of a sphere of radius $r$ is not $4\pi r^{2}$, but $4\pi(1-8\pi \eta^{2})r^{2}$. In the particular case of $\pi/2$, the surface has the geometry of a cone with a deficit angle $\Delta=8\pi^{2}\eta^{2}$. The Gaussian curvature is zero, $K=0$, so we end up with a deflection angle $\hat{\alpha}\simeq 4\pi^{2}\eta^{2}$ [@birola]. On the other hand if $M=Q=\eta=0$, we only have the contribution coming from a cosmic string, the metric in cylindrical coordinates reads [@aryal] $$\mathrm{d}s^{2}=-\mathrm{d}t^{2}+\mathrm{d}z^{2}+\mathrm{d}r^{2}+(1-4\mu)^{2}r^{2}\mathrm{d}\varphi^{2}.$$
This metric describes a static straight string lying along the $z$-axis. It’s not difficult to see that for all surfaces $z=const.$, globally the topology is conical, with a deficit angle $\delta=8\pi \mu$. The Gaussian curvature is zero, $K=0$, with a deflection angle $\hat{\alpha}\simeq 4\mu \pi$ [@gibbons2]. And finally, in the spacetime without topological defects, $\eta=\mu=0$, the metric reduces to the Reissner-Nordström black hole $$\mathrm{d}s^{2}=-\left(1-\frac{2M}{r}+\frac{Q^{2}}{r^{2}}\right)\mathrm{d}t^{2}+\left(1-\frac{2M}{r}+\frac{Q^{2}}{r^{2}}\right)^{-1}\mathrm{d}r^{2}+r^{2}\left(\mathrm{d}\theta^{2}+\sin^{2}\theta \mathrm{d}\varphi^{2}\right)$$
The Gaussian curvature coincides with , the deflection angle $\hat{\alpha}\simeq 4M/b-3\pi Q^{2}/4b^{2}$, is in agreement with [@sereno]. It will be interesting to see if this method can be generalised for the case of image multiplicity in a more general black hole configuration.
Conclusion
==========
In this paper, we have investigated the deflection angle around Reissner-Nordström black holes in the background spacetimes with topological defects. By adopting an optical metric and applying the Gauss-Bonnet theorem to the optical metric, we have found that the total deflection can be expressed as the sum of four corresponding terms $4\mu \pi, 4 \pi^{2}\eta^{2}$, $4M/b$ and $-3\pi Q^{2}/4b^{2}$, respectively. In the particular case $M=Q=0$, and $\eta=0$, the Reissner-Nordström metric reduces to Minkowski spacetime pierced by an infinite static cosmic string with deflection angle $\hat{\alpha}=\delta/2=4\mu \pi$, and similary when $M=Q=\mu=0$, the metric reduces to Minkowski spacetime with a global monopole, with deflection angle $\hat{\alpha}=\Delta/2=4 \pi^{2}\eta^{2}$. Finally, setting $\eta=\mu=0$, we are left with the deflection angle around Reissner-Nordström black hole $\hat{\alpha}=4M/b-3\pi Q^{2}/4b^{2}$.
Acknowledgement {#acknowledgement .unnumbered}
===============
The author would like to thank the anonymous reviewers for the very useful comments and suggestions which help us improve the quality of this paper.
[99]{}
Aryal M., Ford L., Vilenkin A., Phys. Rev. D **34** 2263 (1986)
B. de Mello, Braz. J. Phys. **31** 211 (2011)
Bao D., Chern S., Shen Z., *An Introduction to Riemann-Finsler Geometry*. Springer, New York (2000)
Birrola M. and Vilenkin A. Phys. Rev. Lett. **63** 341 (1989)
Eiroa E.F., Romero G.E., Torres D.F., Phys. Rev. D **66** 024010, (2002).
G. W. Werner, Gen. Rel. Grav. **44** 3047-3057 (2012)
Gibbons and Werner, Class. and Quantum Grav. **25** 235009 (2008)
Gibbons G W, Phys. Lett. B **308** 237–39 (1993)
Gibbons G. W., Herdeiro C. A. R., Warnick C., Werner M. C., Phys. Rev. D **79** 044022 (2009)
Gott J.R. Astrophys. J. **288** 422 (1985)
Kibble T W B, J. Phys. A: Math. Gen. **9**, 1378 (1976)
Linet B., Phys. Rev. D **33** 1833 (1986)
Marques G.A., Bezerra V.B., Phys. Rev. D **66** 105011 (2002)
Punsley B., 1998, ApJ, 498, 640
Sereno, Phys. Rev. D **69** 023002 (2004)
Vilenkin A., Shelard S., *Cosmic Strings and Other Topological Defects*, Cambridge University Press, Cambridge (1994)
[^1]: E-mail: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The NA60 experiment studies muon pair production at the CERN SPS. In this letter we report on a precision measurement of [J/$\psi$]{} in collisions. We have studied the [J/$\psi$]{} centrality distribution, and we have compared it with the one expected if absorption in cold nuclear matter were the only active suppression mechanism. For collisions involving more than $\sim$80 participant nucleons, we find that an extra suppression is present. This result is in qualitative agreement with previous measurements by the NA50 experiment, but no theoretical explanation is presently able to coherently describe both results.'
author:
- |
R. Arnaldi$^{9}$, K. Banicz$^{2,4}$, J. Castor$^{3}$, B. Chaurand$^{6}$, C. Cicalò$^{1}$, A. Colla$^{9}$, P. Cortese$^{9}$, S. Damjanovic$^{2,4}$, A. David$^{2,5}$, A. de Falco$^{1}$, A. Devaux$^{3}$, L. Ducroux$^{10}$, H. En’yo$^{7}$, J. Fargeix$^{3}$, A. Ferretti$^{9}$, M. Floris$^{1}$, A. Förster$^{2}$, P. Force$^{3}$, N. Guettet$^{2,3}$, A. Guichard$^{10}$, H. Gulkanian$^{11}$, J. M. Heuser$^{7}$, M. Keil$^{2,5}$, L. Kluberg$^{2,6}$, C. Lourenço$^{2}$, J. Lozano$^{5}$, F. Manso$^{3}$, P. Martins$^{2,5}$, A. Masoni$^{1}$, A. Neves$^{5}$, H. Ohnishi$^{7}$, C. Oppedisano$^{9}$, P. Parracho$^{2}$, P. Pillot$^{10}$, T. Poghosyan$^{11}$, G. Puddu$^{1}$, E. Radermacher$^{2}$, P. Ramalhete$^{2}$, P. Rosinsky$^{2}$, E. Scomparin$^{9}$, J. Seixas$^{5}$, S. Serci$^{1}$, R. Shahoyan$^{2,5}$, P. Sonderegger$^{5}$, H. J. Specht$^{4}$, R. Tieulent$^{10}$, G. Usai$^{1}$, R. Veenhof$^{2,5}$, H. K. Wöhri$^{1,2,5}$\
(NA60 Collaboration)
title: 'J/$\psi$ production in Indium-Indium collisions at 158 GeV/nucleon'
---
The suppression of the charmonium states by color screening has been predicted as a signature of the phase transition from hadronic matter towards a Quark-Gluon Plasma [@Sat86]. Recently, it has been pointed out that the relatively loosely bound states $\psi'$ and $\chi_c$ should indeed melt for temperatures around the critical temperature $T_{\rm c}$, while the tightly bound [J/$\psi$]{} could survive, even if with strong medium-induced modifications, up to $\sim2 T_{\rm c}$ [@Dat04; @Asa04; @Sat06]. However, the description of the evolution of a $c\overline c$ pair formed by gluon fusion at early times in the history of the collision, which may eventually lead to the formation of a bound state, is still not fully explained by theory. The influence of the medium, a percolating partonic condensate [@Dig04], or a fully thermalized QGP [@Gra04], or even a dense hadronic gas [@Cap05; @Mai05] has been investigated in detail, but this physics topic is still largely data driven, and accurate experimental data are clearly needed.
At the CERN SPS, the NA38 and NA50 experiments have already studied [J/$\psi$]{} production in various colliding systems, including [@Ale04], [@Abr99] and [@Ale05]. Proton-nucleus data provide an important reference, describing the expected absorption of [J/$\psi$]{} in cold nuclear matter. By comparing the centrality dependence of the [J/$\psi$]{} yield observed in nucleus-nucleus collisions to this reference, one can look for suppression mechanisms connected with the formation of a strongly interacting medium. In particular, NA50 has observed, in collisions, that below a certain centrality threshold the [J/$\psi$]{} production is well described invoking nuclear absorption as the only suppression mechanism; on the contrary, above that threshold, an extra suppression (also known as “anomalous” suppression) sets in. Such an interesting observation needs to be complemented by further sets of accurate measurements obtained with different collision systems. In this way one can study in more detail the onset of the anomalous suppression, and understand which is the physics mechanism at its origin.
The NA60 experiment has studied [J/$\psi$]{} production in collisions at 158 GeV/nucleon at the CERN SPS, through its decay into two muons. The experimental apparatus is based on a muon spectrometer (MS), inherited from NA50 [@Abr97], used for triggering on muon pair production and for tracking purposes. A 12$\lambda_i$ thick hadron absorber, mostly made of graphite, separates the MS from the target region, equipped with a beam tracker (BT) and a vertex tracker (VT), placed inside a 2.5 T dipole magnet. Finally, a Zero Degree Calorimeter (ZDC) [@Col04] provides an estimate of the centrality of the collisions. A more detailed description of the apparatus can be found in [@Ban05; @Usa05]. The VT is a radiation-tolerant Si pixel detector, which tracks the charged particles produced in the collision (${\rm d}N_{\rm ch}/{\rm d}\eta \sim$ 200 at midrapidity for central interactions). By matching the tracks measured in the MS with the corresponding tracks in the VT, one can access the kinematical variables of the muons before their distortion due to multiple scattering and energy loss fluctuations in the hadron absorber [@Sha06; @Dav06].
The results presented in this letter refer to the full data sample collected by NA60, corresponding to $\sim$230 million events, taken at a beam intensity of 10$^7$s$^{-1}$. Only the events having at least one interaction vertex with $\ge$4 tracks attached to it are selected for the analysis. The distribution of the longitudinal coordinate of the vertices allows us to identify the target where the interaction took place with $\sim$200 $\mu$m accuracy. We require the interaction vertex to lie, within this tolerance, in one of the 7 In sub-targets, 1.5 mm thick each and spaced by 7.5 mm. Matching of tracks in the VT with the MS tracks is then carried out in coordinate and momentum space. The matching efficiency, of the order of $\sim$60% for events where a [J/$\psi$]{} has been detected in the MS, shows no significant centrality dependence. In principle, MS tracks could be wrongly matched to VT tracks. However, in the [J/$\psi$]{} mass region, this effect is negligible ($<$1%).
Two kinds of event selection have been applied, corresponding to the two different analysis techniques detailed hereafter. In the first, the matching of the muon tracks is not performed, retaining in this way a larger event sample. As a quality check, we perform a cut on the transverse distance between the extrapolated MS track at the target position and the beam axis, weighted by the momentum of the track itself. The level of this cut has been set at 10% of the $\chi^2$ probability for each muon, a value that allows us to efficiently reject muons produced downstream of the target. In the second selection, matching is applied and we require that the tracks matched to those detected in the MS originate from the reconstructed vertex or from the most upstream one, when more than one is found. In this way, we reject the small percentage ($\sim$4%) of events where the [J/$\psi$]{} originates from a downstream interaction of a nuclear fragment produced in a primary collision. For both selections, we apply a beam pile-up rejection cut, based on the BT, requiring two subsequent ions to be separated in time by at least 12 ns. In this way we avoid a superposition of signals in the read-out gate of the ZDC, that would bias the determination of the zero-degree energy $E_{\rm ZDC}$. Finally, in order to reject events at the edges of the MS acceptance we apply the kinematical cuts $0 < y_{cms} <1$ and $-0.5 < \cos\theta_{CS} < 0.5$, where $\theta_{CS}$ is the polar angle of the muons in the Collins-Soper reference frame. In the end, we are left with samples of $\sim$45000 [J/$\psi$]{} for the first event selection, and $\sim$29000 for the second.
The first analysis follows the approach used by the NA38/NA50 experiments. Namely, the [J/$\psi$]{} yield is normalized to the corresponding Drell-Yan (DY), a hard process unaffected by final-state interactions in the medium [@Ram03]. The ratio $\sigma_{\rm J/\psi}/\sigma_{\rm DY}$ has the further advantage of being free from systematic errors connected with the experimental inefficiencies and the integrated luminosity. It is extracted from a fit of the $\mu^+ \mu^-$ invariant mass spectrum (in the region $m_{\mu\mu}>$ 2.2 GeV) to a superposition of the expected contributions, namely the [J/$\psi$]{} and the [$\psi^\prime$]{} resonances, a continuum composed of Drell-Yan events and semi-muonic decays of D and $\overline{\rm D}$ mesons, and a combinatorial background from $\pi$ and K decays. The expected mass shapes of the signals and their acceptances are evaluated through a Monte-Carlo simulation based on PYTHIA [@Sjo01] with GRV94LO [@Glu95] parton distribution functions. The combinatorial background has been estimated from the measured sample of like-sign pairs and its contribution is negligible. Fig. \[fig:1\] shows the $\mu^+\mu^-$ invariant mass spectrum, together with the result of the fit.
The ratios of cross sections, $\sigma_{\rm J/\psi}/\sigma_{\rm DY}$, uncorrected for the [J/$\psi$]{} decay branching ratio, are given in Table \[tab:1\] for three centrality bins corresponding to different values of $\langle N_{\rm part}\rangle$, the average number of participant nucleons in the bin. $\langle N_{\rm part}\rangle$ has been obtained from [$E_{\rm ZDC}$]{} using the Glauber model and taking into account the smearing induced by the detector resolution. The Drell-Yan yield refers to the mass interval $2.9 < m_{\mu\mu} < 4.5$ GeV/c$^2$. Its low statistics (320 events for $m_{\mu\mu}>$ 4.2 GeV/c$^2$) limits the statistical significance of our result and prevents a finer binning in centrality.
Centrality bin $\langle N_{\rm{part}}\rangle$ $\sigma_{\rm J/\psi}/\sigma_{\rm DY}$
---------------------------- -------------------------------- ---------------------------------------
$ E_{\rm ZDC} >$ 11 TeV 63 26.8$\pm$3.2
7 $< E_{\rm ZDC} <$ 11 TeV 123 16.1$\pm$1.6
$E_{\rm ZDC} <$ 7 TeV 175 17.8$\pm$1.6
: \[tab:1\] Values of $\sigma_{\rm J/\psi}/\sigma_{\rm DY}$, uncorrected for the [J/$\psi$]{} decay branching ratio, as a function of centrality.
The values shown in Table \[tab:1\] indicate that for semi-central and central collisions the $\sigma_{\rm J/\psi}/\sigma_{\rm DY}$ ratio is significantly lower than for peripheral reactions. In order to understand how much of the observed reduction is due to cold nuclear matter effects, we have calculated, in the frame of the Glauber model, $\sigma_{\rm J/\psi}/\sigma_{\rm DY}$ as a function of centrality in a pure nuclear absorption scenario. This calculation requires as inputs the [J/$\psi$]{} absorption cross section in cold nuclear matter as well as the ratio of the [J/$\psi$]{} and DY elementary production cross sections at 158 GeV. Measurements performed by NA50 [@Ale05; @Bor05] in collisions at 450 GeV provide $\sigma^{\rm abs}_{\rm J/\psi}$=4.18 $\pm$ 0.35 mb and $(\sigma_{\rm J/\psi}/\sigma_{\rm DY})^{pp}_{450}$=57.5 $\pm$ 0.8. The latter quantity has been rescaled to 158 GeV/nucleon incident energy, using a procedure detailed in Ref. [@Ale05], obtaining $(\sigma_{\rm J/\psi}/\sigma_{\rm DY})^{pp}_{158}$ = 35.7$\pm$3.0. Fig. \[fig:2\] shows the measured $\sigma_{\rm J/\psi}/\sigma_{\rm DY}$, divided by the calculated reference and plotted as a function of $N_{\rm{part}}$. Even if statistical errors are large, a suppression signal of the [J/$\psi$]{} beyond nuclear absorption can be seen.
In the second analysis, we directly compare the measured [J/$\psi$]{} yield to the centrality distribution of [J/$\psi$]{} calculated for the case of pure nuclear absorption, using the input parameters detailed above. In this approach, the measured [J/$\psi$]{} yield has been obtained, in 1 TeV [$E_{\rm ZDC}$]{} bins, by means of a simple fitting procedure that allows us to subtract the small amount of Drell-Yan ($<$4%) and combinatorial background ($<$1%) under the resonance peak. In Fig. \[fig:3\] we compare the [J/$\psi$]{} distribution with the expectation from nuclear absorption. The relative normalization between data and the reference curve is not determined a priori; therefore we fix the ratio between data and the nuclear absorption curve, integrated over centrality, to the same value (0.87 $\pm$ 0.05) as obtained within the previous analysis. It must be noted that events where a heavy nuclear fragment reinteracts in a downstream target have a smaller [$E_{\rm ZDC}$]{} value, since the fragment reinteraction removes some nucleons that would have otherwise reached the ZDC. The measured [J/$\psi$]{} centrality distribution has been corrected for the small bias ($<$2%) induced by this effect. Finally, because of small inefficiencies of the BT, our data sample could be contaminated by high-[$E_{\rm ZDC}$]{} events, where a non-interacting pile-up ion is superimposed to the interacting one. A Monte-Carlo simulation shows that this effect is negligible in our analysis domain, i.e. [$E_{\rm ZDC}$]{} $< 16$ TeV (corresponding to $N_{\rm part} > 50$).
Fig. \[fig:2\] shows the ratio between the measured and expected [J/$\psi$]{} yield, rebinned in order to further reduce statistical fluctuations. Since statistical errors are very small ($<$2%), this analysis requires a careful estimate of systematic errors. We find that they are connected with the determination of the shape and normalization of the nuclear absorption reference, and with the calculation of $N_{\rm part}$ starting from [$E_{\rm ZDC}$]{}. In particular, the uncertainties on $(\sigma_{\rm J/\psi}/\sigma_{\rm DY})^{pp}_{158}$ and $\sigma^{\rm abs}_{\rm J/\psi}$ give an 8% and 4% systematic error on the normalization of the absorption curve, respectively. We then have a 6% error, originating from the centrality integrated value of $\sigma_{\rm J/\psi}/\sigma_{\rm DY}$ used for the normalization. Concerning centrality determination, by varying within errors the input parameters used in the Glauber model, we get a negligible influence on the nuclear absorption reference, except for very central events ([$E_{\rm ZDC}$]{}$<$3 TeV), where the size of the effect is $\sim$12%. Furthermore, the ZDC does not measure only spectator nucleons, but also a small amount of energy released by forward secondary particles emitted in the acceptance of the calorimeter ($\eta > 6.3$). This contribution, important only for central collisions, is taken into account when calculating $N_{\rm part}$ from [$E_{\rm ZDC}$]{}. By conservatively assuming a 10% uncertainty on this quantity we get, for events with [$E_{\rm ZDC}$]{}$<$3 TeV, a 9% error on the absorption curve. For more peripheral events the effect is negligible. Combining the various sources in quadrature, we end up with a $\sim$11% systematic error, independent of centrality. On top of that, the most central bins are affected by a further, sizeable systematic error relatively to the others. It must be noted that the systematic error plotted in Fig. \[fig:2\], except for the fraction due to the 6% normalization error quoted above, also affects the results on $\sigma_{\rm J/\psi}/\sigma_{\rm DY}$.
The results obtained by NA60 in collisions show that in the region $50 < N_{\rm part} < 100$ an anomalous suppression of the [J/$\psi$]{} sets in. Taking into account the $N_{\rm part}$ smearing due to the [$E_{\rm ZDC}$]{} resolution, the effect seen is compatible with the occurrence of a $\sim$15% drop of the [J/$\psi$]{} yield at $N_{\rm part} \sim 80$, followed by a more or less constant behaviour. When expressed in terms of the Bjorken energy density, the onset of the anomalous suppression roughly corresponds to 1.5 GeV/fm$^3$ (using $\tau_0$=1 fm/c and the VENUS [@Wer93] estimate for the charged multiplicity as a function of centrality). In Fig. \[fig:4\] we compare our result with the [J/$\psi$]{} suppression pattern obtained by NA50 in collisions [@Ale05]. The systematic errors on the determination of the nuclear absorption reference from the data sample amount to $\pm$9% and are not shown in this comparison plot since they affect and results in a similar way. Within errors, the two patterns look compatible in the $N_{\rm part}$ region explored by both systems, indicating that $N_{\rm part}$ might be, at SPS energy, a scaling variable for the anomalous suppression. A detailed investigation of the scaling properties of [J/$\psi$]{} suppression as a function of several centrality variables would give valuable insights into the origin of the observed effect. However, a meaningful comparison would require results with error bars similar to the ones obtained for the analysis.
Several theoretical predictions, tuned on already available results from NA50, were formulated for [J/$\psi$]{} suppression in collisions [@Dig04; @Cap05; @Rap05]. We find that none of them is able to quantitatively reproduce the suppression pattern measured by NA60 [@Sco07]. Recent results from the PHENIX Collaboration [@Ada06] have shown that also in collisions at $\sqrt{s}$=200 GeV/nucleon the [J/$\psi$]{} is suppressed beyond nuclear absorption, and that the suppression is larger at forward rapidities. A coherent interpretation of the results at SPS and RHIC energies is now mandatory in order to understand the physics mechanisms affecting charmonia in a dense partonic/hadronic environment. The results obtained by NA60 represent the most accurate measurement of [J/$\psi$]{} suppression in nuclear collisions available today and are a key element to strictly constrain theoretical models. Further studies on the $y$ and $p_{\rm T}$ dependence of the [J/$\psi$]{} suppression are underway and will be the subject of future publications.
In summary, we have measured [J/$\psi$]{} suppression in collisions at 158 GeV/nucleon. Comparing the [J/$\psi$]{} centrality distribution with the expectation from a pure nuclear absorption scenario, we find an anomalous suppression that sets in for $N_{\rm part} \sim 80$, and saturates for more central events. The statistical errors are negligible (of the order of 2%). Most of the systematic errors are centrality independent and therefore do not affect the measured shape of the [J/$\psi$]{} suppression pattern. None of the existing theoretical models, tuned on the measured [J/$\psi$]{} suppression in collisions, is able to quantitatively reproduce the results shown in this letter.
The YerPhI group was supported by the C. Gulbenkian Foundation, Lisbon and the Swiss Fund Kidagan.
[99]{}
T. Matsui and H. Satz, Phys. Lett. [**B178**]{}, 416 (1986). S. Datta [*et al.*]{}, Phys. Rev. [**D69**]{}, 094507 (2004). M. Asakawa and T. Hatsuda, Phys. Rev. Lett. [**92**]{}, 012001 (2004). H. Satz, J. Phys. [**G32**]{}, R25 (2006). S. Digal [*et al.*]{}, Eur. Phys. J. [**C32**]{}, 547 (2004). L. Grandchamp [*et al.*]{}, Phys. Rev. Lett. [**92**]{}, 212301 (2004). A. Capella and E.G. Ferreiro, Eur. Phys. J. [**C42**]{}, 419 (2005). L. Maiani [*et al.*]{}, Nucl. Phys. [**A748**]{}, 209 (2005). B. Alessandro [*et al.*]{}, (NA50), Eur. Phys. J. [**C33**]{}, 31 (2004). M.C. Abreu [*et al.*]{}, (NA38), Phys. Lett. [**B449**]{}, 128 (1999). B. Alessandro [*et al.*]{}, (NA50), Eur. Phys. J. [**C39**]{}, 335 (2005). M.C. Abreu [*et al.*]{}, (NA50), Phys. Lett. [**B410**]{}, 327 (1997). A. Colla, Ph.D. Thesis, Università di Torino, 2004. K. Banicz [*et al.*]{}, Nucl. Instr. Meth. [**A546**]{}, 51 (2005). G. Usai [*et al.*]{}, (NA60), Eur. Phys. J. [**C43**]{}, 415 (2005). R. Shahoyan [*et al.*]{}, (NA60), Nucl. Phys. [**A774**]{}, 677 (2006). A. David, Ph.D. Thesis, Instituto Superior Técnico, Lisbon, 2006, CERN-THESIS-2006-007. L. Ramello [*et al.*]{}, (NA50), Nucl. Phys. [**A715**]{}, 243 (2003). T. Sjöstrand [*et al.*]{}, Comp. Phys. Comm. [**135**]{}, 238 (2001). M. Glück [*et al.*]{}, Z. Phys. [**C67**]{}, 433 (1995). G. Borges [*et al.*]{}, (NA50), Eur. Phys. J. [**C43**]{}, 161 (2005). K. Werner, Phys. Rep. [**232**]{}, 87 (1993). R. Rapp, Eur. Phys. J. [**C43**]{}, 91 (2005). E. Scomparin [*et al.*]{}, (NA60), nucl-ex/0703030, proc. of QM2006 Int. Conf., to appear in J. Phys. G. A. Adare [*et al.*]{}, (PHENIX), nucl-ex/0611020.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We present the first *Chandra* and *Swift* X-ray study of the spiral galaxy NGC4088 and its ultraluminous X-ray source (ULX N4088–X1). We also report very long baseline interferometry (VLBI) observations at 1.6 and 5 GHz performed quasi-simultaneously with the *Swift* and *Chandra* observations, respectively. Fifteen X-ray sources are detected by *Chandra* within the D25 ellipse of NGC4088, from which we derive the X-ray luminosity function (XLF) of this galaxy. We find the XLF is very similar to those of star-forming galaxies and estimate a star-formation rate of 4.5 $M_{\odot}$ yr$^{-1}$. The *Chandra* detection of the ULX yields its most accurate X-ray position, which is spatially coincident with compact radio emission at 1.6 GHz. The ULX *Chandra* X-ray luminosity, $L_\mathrm{0.2-10.0 keV} = 3.4\ \times$ 10$^{39}$ erg s$^{-1}$, indicates that N4088–X1 could be located at the high-luminosity end of the high-mass X-ray binary (HMXB) population of NGC4088. The estimates of the black hole mass and ratio of radio to X-ray luminosity of N4088–X1 rule out a supermassive black hole nature. The *Swift* X-ray spectrum of N4088–X1 is best described by a thermal Comptonization model and presents a statistically significant high-energy cut-off. We conclude that N4088–X1 is most likely a stellar remnant black hole in a HMXB, probably fed by Roche lobe overflow, residing in a super-Eddington ultraluminous state. The 1.6 GHz VLBI source is consistent with radio emission from possible ballistic jet ejections in this state.'
author:
- 'M. Mezcua, G. Fabbiano, J.C. Gladstone, S.A. Farrell, and R. Soria'
bibliography:
- 'referencesALL.bib'
title: |
Revealing the nature of the ULX and X-ray population\
of the spiral galaxy NGC4088
---
Introduction
============
Early X-ray observations of nearby galaxies with the *Einstein Observatory*, and later with *ROSAT*, *ASCA*, *XMM-Newton*, and *Chandra*, revealed the presence of a population of off-nuclear X-ray point sources with X-ray luminosities above the Eddington limit for a stellar-mass black hole ($L_\mathrm{X} > 10^{39}$ erg s$^{-1}$; ). These high luminosities imply black holes (BHs) of masses $>$ 10 $M_{\odot}$, if the accretion is sub-Eddington and the radiation is isotropic. This suggested that ULXs could be intermediate mass BHs (IMBHs) with BH masses in the range $100 \leq M_\mathrm{BH} \leq 10^{5} M_{\odot}$ (e.g., @1999ApJ...519...89C). The alternative is stellar-remnant BHs ($M_\mathrm{BH} < 100 M_{\odot}$) accreting at around or above the Eddington limit (e.g., @2001ApJ...552L.109K; @2005Sci...307..533F; @2011NewAR..55..166F).
Studies of the X-ray luminosity function (XLF) of the X-ray populations of these nearby galaxies are a very useful tool for discerning between different types of X-ray populations and galaxy properties (e.g., morphological type, age of stellar population) and investigating the nature of the X-ray sources that populate them (e.g., X-ray binaries -XRBs; see review by ). The XLF of late-type star-forming galaxies is usually described by a straight power-law and is associated with a population of high-mass X-ray binaries (HMXBs) located in regions of star formation (e.g., @2002ApJ...577..726Z; @2004ApJ...602..231C; @2007ApJ...661..135Z). Early-type galaxies are best described by an XLF with a broken power-law and a break at a few times 10$^{38}$ erg s$^{-1}$ (attributed to the Eddington luminosity of neutron star XRBs; e.g., @2000ApJ...544L.101S [@2001ApJ...556..533S]), and are thought to be dominated by a population of low-mass XRBs (LMXBs, e.g., @2004ApJ...611..846K [@2010ApJ...721.1523K]; @2004MNRAS.349..146G). Studies to the XLF of different galaxies can also be used to gain insight into the nature of ULXs.
ULX population studies have revealed that many ULXs are associated with young star-forming regions and star-forming galaxies (e.g., @2004ApJS..154..519S [@2009ApJ...703..159S]). They also tend to be found in low-metallicity regions (e.g., @2009MNRAS.400..677Z; @2010ApJ...714.1217B; @2010MNRAS.408..234M). Such low metallicities are required in order to make more massive BHs (e.g., @2003ApJ...591..288H), which shows that it would be possible for ULXs to be massive stellar remnant BHs. Yet, population studies of the HMXB population of the local universe has also revealed an unbroken power-law slope in the XLF up to $\sim2 \times 10^{40}$ erg s$^{-1}$ (e.g., @2003MNRAS.339..793G; @2011ApJ...741...49S; @2012MNRAS.419.2095M). This continuation indicates that the majority of ULXs with $L_\mathrm{X} < 2 \times 10^{40}$ erg s$^{-1}$ are likely to be HMXBs with a stellar-mass BH accreting at around or above the Eddington limit. Many studies have been undertaken to obtain the mass of the putative BH contained in ULXs. Such methods include: studying the ULX optical counterpart (e.g., @2006IAUS..230..293P; @2009ApJ...697..950K; @2011ApJ...728L...5C; @2011ApJ...737...81T; @2013ApJS..206...14G); from X-ray analysis, using either spectral fitting (e.g., @2003ApJ...585L..37M; @2009MNRAS.397..124G; @2010MNRAS.402.2559C; @2011ApJ...734..111D; @2012ApJ...752...34G; @2013arXiv1310.0745B), the luminosity-temperature relation (e.g., @2003ApJ...585L..37M; @2009MNRAS.397..124G; @2009ApJ...692..443S; @2011ApJ...743....6S), quasi-periodic oscillations (e.g., @2003ApJ...586L..61S; @2007ApJ...660..580S; @2013MNRAS.tmp.2483C), or X-ray variability (; @2009MNRAS.397.1061H; @2013MNRAS.tmp.2522D). In those cases where candidate radio counterparts have been identified, attempts have been made to obtain the BH mass using the Fundamental Plane of accreting BHs (e.g., ; @2011IAUS..275..325C; @2011AN....332..379M; @2012Sci...337..554W; @2013MNRAS.436.1546Ma; @2013MNRAS.436.3128Mc). The Fundamental Plane is a correlation between 2–10 keV X-ray luminosity, 5 GHz radio luminosity, and BH mass that holds for sub-Eddington accreting BHs in the low/hard X-ray state and with steady jet emission (e.g., @2003MNRAS.345.1057M; ; @2012MNRAS.423..590G). The detection of compact core radio emission is required in order to locate a ULX in the Fundamental Plane, which can be achieved only by means of very long baseline interferometry (VLBI) radio observations. [@2011AN....332..379M] initiated a program with the European VLBI Network (EVN[^1]) aimed at detecting and studying the milliarcsecond-scale emission of ULXs. Such investigations have indicated that those ULXs with radio counterparts may be powered by IMBHs (e.g., @2011AN....332..379M; @2013MNRAS.436.1546Ma, [[email protected]]c). Here, we investigate one of these ULXs in NGC4088. NGC4088 is an asymmetric spiral galaxy ($D_\mathrm{L}$ = 13 Mpc, redshift = 0.002524; ) hosting a ULX (N4088–X1) offset $\sim$32 arcsec from the nucleus. N4088–X1 was first detected with the *ROSAT* satellite by [@2005ApJS..157...59L], who reported an X-ray luminosity $L_\mathrm{0.3-8.0 keV} \sim 6 \times 10^{39}$ erg s$^{-1}$. The ULX is located within the spiral arm of NGC4088, possibly within an HII region (e.g., @2005ApJS..157...59L; ). A cross-match of the VLA[^2] FIRST[^3] catalog with *ROSAT* ULX catalogs (e.g., @2005ApJS..157...59L) revealed a 1.4 GHz VLA radio counterpart for the ULX of 1.87 mJy, with an offset between the *ROSAT* and the radio position of 3.6 arcsec (). Later EVN observations at 1.6 GHz yielded the detection of a compact, unresolved component of 0.1 mJy centered at RA(J2000) = 12$^h$05$^m$31$^s$.7110 $\pm$ 0$^s$.0003, Dec.(J2000) = 50$^{\circ}$3246.729 $\pm$ 0.002 (@2011AN....332..379M). This provided an upper limit on the 5 GHz luminosity (assuming a radio spectral index $\alpha$ = 0.15), from which an upper limit on the ULX BH mass of $\sim10^{5} M_{\odot}$ was estimated using the Fundamental Plane of accretion (@2011AN....332..379M).
In this paper we present the first *Chandra* observations of the galaxy NGC4088 and the ULX N4088–X1, as well as quasi-simultaneous VLBI observations with the EVN at 5 GHz. We also report a reanalysis of the EVN data at 1.6 GHz, now imaged at the *Chandra* position reported in this paper, and the analysis of 23 *Swift* observations performed nearly at the same time (between April and September 2009) as the 1.6 GHz EVN observations (2009 June 1–2). With these data we investigate the proposed association between the X-ray and radio emission (for which a sub-arcsecond X-ray position is needed) of N4088–X1 and attempt to estimate the BH mass with the aim of revealing the nature of this ULX.
The paper is organized as follows: the observations and data analysis are presented in Section \[observations\], while the main results obtained are shown in Section \[results\] and discussed in Section \[discussion\]. Final conclusions are presented in Sect. \[conclusions\].
Through this paper we assume a $\Lambda$ CDM cosmology with parameters $H_\mathrm{0} = 73$ km s$^{-1}$ Mpc$^{-1}$, $\Omega_{\Lambda}=0.73$, and $\Omega_\mathrm{m}=0.27$.
Observations and data analysis {#observations}
==============================
*Chandra* X-ray observations {#chandraobservations}
----------------------------
The ULX N4088–X1 was observed on 2012 June 06 (Obs. ID: 14442; PI: Mezcua) with the *Chandra X-ray observatory* (Weisskopf et al. 2002). The observation was performed using the Advanced CCD Imaging Spectrometer detector (@1997AAS...190.3404G; ACIS-S) with an integration time of 19.8 ks. The data were reprocessed using CIAO version 4.5 and the corresponding calibration files, following the standard *Chandra* ACIS data analysis threads[^4]. The *chandra\_repro* reprocessing script was used to reprocess the data and generate a new level=2 event file. An image of the S3 chip (ccd\_id=7), a background image, and a PSF map that provides the size of the PSF at each pixel in the image were then produced using the tools *dmcopy*, *aconvolve*, and *mkpsfmap* and given to *wavdetect*, which performed source detection and extracted source net counts in the energy range 0.3–10 keV. The detected source count rates for all sources lying within the D25 ellipse of NGC 4088 were converted to source fluxes by applying a conversion factor calculated assuming a power-law spectrum of $\Gamma$ = 1.8 and line-of-sight Galactic absorption[^5] $N_\mathrm{H} = 2 \times 10^{20}$ cm$^{-2}$ using WebPIMMS[^6]. A $\Gamma$ = 1.4$\sim$2 (e.g., @2002ApJ...577..726Z) and $\Gamma$ = 1.4–1.8 (e.g., @2011ApJ...729...12B) was found for HMXBs and LMXBs, respectively. We note that the average value of $\Gamma$ = 1.8 used here may not be applicable to our case since those studies have a deeper detection limit that includes also XRBs in the low/hard state (dominated by a hard power-law component), while we can only see those in the high/soft state (dominated by a disk blackbody). To make sure that we are using a plausible slope, we used the [XSPEC]{} task [*fakeit*]{} to simulate a distribution of [*Chandra*]{} spectra for disk-blackbody models with column densities equal to the line-of-sight and twice the line-of-sight values, and inner-disk temperatures distributed between 0.6 and 1.0 keV (typical disk temperatures in the high/soft state; e.g., @2006csxs.book..157M). We then re-fitted the simulated spectra with a power-law model, fixing the column density to the line-of-sight value and leaving the photon index as a free parameter. We find that low signal-to-noise disk-blackbody spectra in that temperature range are indeed approximated in the [*Chandra*]{} band (at least for the purpose of converting from count rates to fluxes) by power-laws of photon index between $\sim$1.5 (for $T_\mathrm{in} \sim$ 1 keV) and $\sim$2 (for $T_\mathrm{in} \sim$ 0.6 keV), with the most likely value $\Gamma$ $\sim$ 1.8. The source spectrum of N4088–X1 was extracted using the *specextract* script and selecting a circular region of 2$\arcsec$ around the target source and of 15$\arcsec$ in a source-free area of the same chip for the background. The extracted spectrum was grouped to 15 counts per energy bin to allow for $\chi^{2}$ fitting using the tool *grppha*.
*Swift* X-ray Observations
--------------------------
NGC4088 was observed on 23 occasions in X-rays with the *Swift* X-ray Telescope (XRT) between 2009 April 15 and 2009 September 21 as part of a monitoring campaign targeting the supernova (SN) SN2009dd (see Table \[swiftlog\] for a log of the *Swift* observations). We generated images, light curves and spectra, including the background and ancillary response files, with the online XRT data product generator [^7] (@2009MNRAS.397.1177E). We downloaded suitable spectral response files for single and double events in photon-counting mode from the latest calibration database. N4088–X1 was clearly detected in all observations with an average XRT count rate of 0.0037 count s$^{-1}$. We extracted a light curve binned at the duration of each individual observation, finding evidence for small deviations from the average count rate in 5 of the 23 observations, though no evidence for significant spectral variability could be seen in the standard *Swift hardness ratio ((1.5–10 keV)/(0.3–1.5 keV))*. We next extracted a combined spectrum from the *Chandra* position of N4088–X1, excluding the 5 observations in which evidence for variability was seen, giving a total exposure time of 76 ks.
[cccc]{} Date & Obsid & Exp. time (s) & Spectrum\
2009-04-15 & 00031401001 & 1996 & Y\
2009-04-17 & 00031401002 & 830 & Y\
2009-04-19 & 00031401003 & 3946 & Y\
2009-04-21 & 00031401004 & 2670 & Y\
2009-04-25 & 00031401005 & 2712 & Y\
2009-05-10 & 00031401006 & 4306 & Y\
2009-05-16 & 00031401007 & 6045 & Y\
2009-07-04 & 00031401008 & 3824 & Y\
2009-07-05 & 00031401009 & 5957 & Y\
2009-07-15 & 00031401011 & 3930 & N\
2009-07-16 & 00031401012 & 8963 & N\
2009-07-19 & 00031401013 & 4863 & Y\
2009-07-19 & 00031401014 & 6049 & N\
2009-07-26 & 00031401015 & 4668 & N\
2009-07-27 & 00031401016 & 5181 & N\
2009-08-02 & 00031401017 & 5934 & Y\
2009-08-03 & 00031401018 & 4647 & Y\
2009-08-09 & 00031401019 & 6968 & Y\
2009-09-13 & 00031401020 & 33 & Y\
2009-09-14 & 00031401021 & 4593 & Y\
2009-09-16 & 00031401022 & 4587 & Y\
2009-09-20 & 00031401023 & 5858 & Y\
2009-09-21 & 00031401024 & 6135 & Y\
VLBI radio observations {#VLBI}
-----------------------
N4088–X1 was observed with the EVN at 5 GHz on 2012 June 01 (project code: EM095A; PI: Mezcua). Eight antennas participated in the observations: Effelsberg (Germany), Westerbork (The Netherlands), Jodrell Bank (United Kingdom), Onsala (Sweden), Medicina (Italy), Noto (Italy), Torun (Poland), and Yebes (Spain). The observations were performed using the phase-referencing technique, in which the target and a nearby, compact, bright source (the phase calibrator) are observed interleaving scans. A target-phase calibrator cycle of four minutes (3 minutes on the target, 1 min on the phase calibrator) was used. As a result, a total of 2.7 h was spent on the target source. The source J1203+4803 was used as phase calibrator, while the bright and compact radio source 4C+39.25 was used as fringe finder and bandpass calibrator.
The data were recorded in dual-circular polarization and at sample rate of 1024 megabit per second (Mbps). Eight intermediate frequency bands, each of 16 MHz each and 32 spectral channels, were used. The data correlation was performed at JIVE[^8] with an averaging time of 4 sec.
The calibration of the correlated data was performed using AIPS[^9]. Amplitudes were calibrated using the gains of the antennas and system temperatures. The data were then fringe-fitted using the phase calibrator. Delay, delay rate, and phase solutions derived from the phase calibrator were interpolated and applied to the target. The imaging was performed using CLEAN deconvolution with the AIPS task IMAGR. Natural weighting images were produced from non-channel-averaged data[^10] and imaging two fields of view (FOV) derived from the positional errors of the VLA and *Chandra* ULX counterparts: one of $\sim$1 $\times$ 1 arcsec$^{2}$ centered at the *Chandra* position, and another one of $\sim$5 $\times$ 5 arcsec$^{2}$ centered at the VLA position. The restoring beam size was 5.7 mas $\times$ 4.5 mas. To enhance the sensitivity, we also repeated the multi-field imaging without using the baselines longer than 20 M$\lambda$.
We did also recalibrate the 1.6 GHz EVN data previous analyzed by [@2011AN....332..379M] and reimaged the data without averaging the channels and using the same two FOVs as at 5 GHz. The imaging was also performed using only the baselines shorter than 15 M$\lambda$, which resulted in a beam size of 33 mas $\times$ 27 mas.
Results
=======
A total of thirty-one X-ray sources are detected by *wavdetect* in the 0.3–10 keV band, of which fifteen lay within the D25 ellipse of the host galaxy. The location of the fifteen detected sources is shown in Fig. \[fig1\], and their position, observed flux, and 0.3–10 keV band luminosity are provided in Table \[table1\].
![Digitized Sky Survey (@1990AJ.....99.2019L) image of NGC4088. The position of the fifteen sources detected by *wavdetect* in the 0.3–10 keV band within the D25 ellipse of NGC4088 are marked with orange circles of radius 3 arcsec. The position of the ULX is marked with a white circle.\[fig1\]](sourcesD25DSS_v2.pdf)
[lcccc]{} Source & RA & Dec. & Net counts & $L_\mathrm{0.3-10 keV}$\
& (J2000) & (J2000) & & ($\times 10^{38}$ erg s$^{-1}$ )\
N4088–X1& 12 05 32.33 $\pm$ 0.01 & 50 32 45.9 $\pm$ 0.1 & 221.2 $\pm$ 15.3 & 21.9 $\pm$ 1.5\
Src. 2 & 12 05 29.73 $\pm$ 0.01 & 50 32 28.3 $\pm$ 0.1 & 89.1 $\pm$ 9.8 & 8.8 $\pm$ 1.0\
Src. 3 & 12 05 31.92 $\pm$ 0.01 & 50 32 16.9 $\pm$ 0.2 & 60.5 $\pm$ 8.3 & 6.0 $\pm$ 0.8\
Src. 4 & 12 05 34.36 $\pm$ 0.02 & 50 32 19.9 $\pm$ 0.2 & 49.3 $\pm$ 7.8 & 4.9 $\pm$ 0.8\
Src. 5 & 12 05 35.41 $\pm$ 0.01 & 50 31 59.7 $\pm$ 0.1 & 49.3 $\pm$ 7.4 & 4.9 $\pm$ 0.7\
Src. 6 & 12 05 30.70 $\pm$ 0.02 & 50 32 24.9 $\pm$ 0.2 & 47.3 $\pm$ 7.4 & 4.7 $\pm$ 0.7\
Src. 7 & 12 05 34.18 $\pm$ 0.01 & 50 32 24.3 $\pm$ 0.1 & 27.6 $\pm$ 6.1 & 2.7 $\pm$ 0.6\
Src. 8 & 12 05 35.34 $\pm$ 0.00 & 50 31 47.6 $\pm$ 0.1 & 22.8 $\pm$ 5.1 & 2.3 $\pm$ 0.5\
Src. 9 & 12 05 30.83 $\pm$ 0.02 & 50 33 10.8 $\pm$ 0.2 & 20.3 $\pm$ 4.8 & 2.0 $\pm$ 0.5\
Src. 10 & 12 05 38.40 $\pm$ 0.01 & 50 32 43.2 $\pm$ 0.1 & 17.1 $\pm$ 4.5 & 1.7 $\pm$ 0.4\
Src. 11 & 12 05 37.29 $\pm$ 0.02 & 50 31 09.4 $\pm$ 0.4 & 15.1 $\pm$ 4.4 & 1.5 $\pm$ 0.4\
Src. 12 & 12 05 40.92 $\pm$ 0.05 & 50 32 58.6 $\pm$ 0.3 & 13.9 $\pm$ 4.4 & 1.4 $\pm$ 0.4\
Src. 13 & 12 05 35.80 $\pm$ 0.03 & 50 32 31.4 $\pm$ 0.3 & 11.6 $\pm$ 4.1 & 1.2 $\pm$ 0.4\
Src. 14 & 12 05 34.02 $\pm$ 0.01 & 50 32 12.9 $\pm$ 0.2 & 8.2 $\pm$ 3.5 & 0.8 $\pm$ 0.3\
Src. 15 & 12 05 30.13 $\pm$ 0.04 & 50 31 21.0 $\pm$ 0.2 & 8.0 $\pm$ 3.2 & 0.8 $\pm$ 0.3\
SN2009dd & 12 05 34.10 & 50 32 19.4 & 8.0 $\pm$ 3.2 & 0.8 $\pm$ 0.3\
X-ray properties of the ULX {#xrayprop}
---------------------------
For the ULX N4088–X1, we obtain an X-ray luminosity of $L_\mathrm{0.3-10 keV} = 2.2 \times 10^{39}$ erg s$^{-1}$ using the observed *Chandra* fluxes derived from the count rates assuming the power-law model described in Sect. \[chandraobservations\]. The *Chandra* detection of N4088–X1 shows that the source is located at RA(J2000) = 12$^h$05$^m$32$^s$.33 $\pm$ 0$^s$.01, Dec.(J2000) = 50$^{\circ}$3245.9 $\pm$ 0.1. The positional 1$\sigma$-errors correspond to the statistical uncertainties affecting the *wavdetect* centroid algorithm and the dispersion of photons due to the PSF. In addition, the *Chandra* absolute astrometry can be shifted by up to 0.6 arcsec due to pointing uncertainties.
### *Chandra* ULX spectral modeling
We fitted the *Chandra* spectrum using the X-ray spectral fitting package [XSPEC]{} (@1996ASPC..101...17A) v12.7.1 in the 0.2 to 10.0 keV energy range. Two models were used: an absorbed multicolor disk-blackbody model (*wabs\*diskbb*) and a power-law model (*wabs\*pow*). The fits were performed using the minimum $\chi^{2}$ method (i.e. Gehrels Chisq statistics). The fitting results of each model are shown in Table \[chandra\], while a plot of the power-law spectral fit is shown in Fig. \[chandrapow\].
Both the *pow* and the *diskbb* model provide acceptable statistical fits (i.e. null-hypothesis probability $>$ 5%, so rejection likelihood $<$ 95%), with values of $N_\mathrm{H}\sim$ 10–15 times larger than the Galactic column density (i.e. $N_\mathrm{H} = 2 \times 10^{20}$ cm$^{-2}$, @2003ApJ...588..805K). The data are insufficient to statistically distinguish between models. The disk-blackbody model provides an inner disk temperature $kT_\mathrm{in}$ = 2.5 keV, too high for a standard disk. The power-law fit provides a very hard photon index $\Gamma$ = 1.1 and unabsorbed flux = 1.7 $\times 10^{-13}$ erg s$^{-1}$ cm$^{2}$, from which we derive an X-ray luminosity $L_\mathrm{0.2-10.0 keV}$ = 3.4 $\times\ 10^{39}$ erg s$^{-1}$.
---------- ----------------------- ------------------------ ---------------------------------------- ------------------------------------------- -------------- ---------
Model $N_\mathrm{H}$ $\Gamma/T_\mathrm{in}$ Norm. Flux $\chi^2$/dof p-value
(10$^{22}$ cm$^{-2}$) ( /keV) (erg cm$^{-2}$ s$^{-1}$)
*pow* 0.3$^{+0.5}_{-0.3}$ 1.1$^{+0.6}_{-0.5}$ (1.6$^{+2}_{-0.7}$) $\times$ 10$^{-5}$ (1.7$^{+0.6}_{-1.5}$) $\times$ 10$^{-13}$ 16.9/10 0.08
*diskbb* 0.2$^{+0.3}_{-0.2}$ 2.5$^{+4}_{-0.9}$ 2.1 $\times$ 10$^{-4}$ (1.4$^{+0.2}_{-1.4}$) $\times$ 10$^{-13}$ 15.62/10 0.11
---------- ----------------------- ------------------------ ---------------------------------------- ------------------------------------------- -------------- ---------
### *Swift* ULX spectral modeling
We fitted the co-added *Swift* spectrum using [XSPEC]{} again, with an absorbed power-law model with absorption accounted for using the *wabs* model and the $N_\mathrm{H}$ column set to be greater than the Galactic absorption in the direction of N4088–X1 (i.e. 2 $\times$ 10$^{20}$ atoms cm$^{-2}$). The fit obtained, with a hard photon index of 1.3 (see Table \[specfit\]), does not satisfactorily describe the observed spectrum, as two data points above 5 keV fall below the power-law fit (see Fig. \[figSwift\], top). We thus attempted to fit this spectrum with *wabs\*diskbb* as with our *Chandra* data and an exponential cut-off *wabs\*pow\*highe*. Using the *wabs\*diskbb* model, a fit with $\chi^{2}$/dof=39.97/34 and an inner disk temperature of 2.1 keV (see Table \[specfit\]) is obtained. This temperature is again too high for optically thick emission from a typical optically thin accretion disk. A good fit ($\chi^{2}$/dof=36.89/32) is also obtained with the exponential cut-off model. The high-energy cut-off has been also observed in other ULXs (e.g., @2006ApJ...641L.125D, [[email protected]]; @2006MNRAS.368..397S; ; @2009MNRAS.397.1836G; @2013MNRAS.435.1758S) and is well described by thermal Comptonization of hot coronal electrons by soft photons (e.g., @1994ApJ...434..570T; @1995ApJ...449..188H). We thus test if a Comptonization model (*wabs\*comptt*) is able to fit the data (Fig. \[figSwift\], bottom). We find that the *Swift* spectrum is well described ($\chi^{2}$/dof=35.01/32) with a Comptonization model of input soft photon temperature $T_\mathrm{0}<$ 0.1 keV and plasma temperature $kT_\mathrm{e}$=1.2 keV (see Table \[specfit\]). The S/N is insufficient for us to detect any contribution from the disk, although the disk should still be present.
Both the *diskbb*, *pow\*highe*, and *comptt* models provide significant improved statistics with respect to the *pow* model, as indicated by the $\chi^{2}$/dof. The Bayesian Information Criterion (BIC; @schwarzBIC; @Kass1995) can be also used as a further indicator of the statistical improvement of one model over another. The BIC value can be calculated as BIC = 2log($L_\mathrm{1}$) - 2log($L_\mathrm{2}$) - ($k_\mathrm{1} - k_\mathrm{2}$)log(n), where $L$ = exp$(-\chi^{2}$/2) for models 1 and 2, respectively, $k$ = number of parameters in model, and $n$ = number of data points. We obtain BIC values in the range 2–6 for the three models compared to the *pow* one, which is a ‘positive’ result[^11]. When comparing the *comptt* to the *diskbb* and *pow\*highe* models we obtain BIC numbers between 0 and 2, which indicate that the *comptt* is ‘preferred’ over these models.
To quantify the significance of the spectral cut-off, we also try to fit a broken power-law with the two slopes tied together and then thaw one of them and use the F-test to compare the fits. The F-test assesses whether the improvement of the $\chi^{2}$ is due to chance or to the new component being significant (e.g., @1989sgtu.book.....B). We obtain a break energy of 4.7$^{+0.6}_{-0.9}$ keV, an F-statistic value of 6.04, and a probability of 0.006, which is $<<$1 and thus indicates that the cut-off is significantly there.
[lccccccccc]{} Model & $N_\mathrm{H}$ & $\Gamma/T_\mathrm{in}$ & Cut-off & $T_\mathrm{0}$ & $kT_\mathrm{e}$ & $\tau$ & Norm. & Flux & $\chi^{2}$/dof\
& (10$^{22}$ cm$^{-2}$) & ( /keV) & (keV) & (keV) & (keV) & & &(erg cm$^{-2}$ s$^{-1}$) &\
*pow* & 0.05$^{+0.06}_{-0.05}$ & 1.3$^{+0.2}_{-0.2}$ & – & – & – & – &(3.4$^{+0.8}_{-0.6}$) $\times$ 10$^{-5}$ & 3.7$^{+0.4}_{-0.3}$ $\times$ 10$^{-13}$ & 51.06/34\
*pow\*highe* & 0.02$^{+0.03}_{-0.02}$ & 0.9$^{+0.2}_{-0.3}$ & 3$^{+2}_{-1}$ & – & – & – & (2.9$^{+0.4}_{-0.3}$) $\times$ 10$^{-5}$ & (3.0$^{+0.4}_{-0.4}$) $\times$ 10$^{-13}$ & 36.89/32\
*diskbb* & 0.02$^{+0.02}_{-0.02}$ & 2.1$^{+0.3}_{-0.4}$ & – & – & – & –& (8$^{+7}_{-3}$) $\times$ 10$^{-4}$ & (3.0$^{+0.4}_{-0.3}$) $\times$ 10$^{-13}$ & 39.97/34\
*comptt* & 0.02$^{+0.03}_{-0.02}$ & – & – & 0.03$^{+0.1}_{-0.03}$ & 1.2$^{+0.2}_{-0.2}$ & 17$^{+3}_{-3}$ & (1.3$^{+1.0}_{-0.5}$) $\times$ 10$^{-4}$ & (2.9$^{+0.3}_{-0.3}$) $\times$ 10$^{-13}$ & 35.01/32\
![Fit to a power-law (top) and a thermal Comptonization model (bottom) to the *Swift* XRT combined spectrum of N4088–X1. \[figSwift\]](powplot3.pdf "fig:") ![Fit to a power-law (top) and a thermal Comptonization model (bottom) to the *Swift* XRT combined spectrum of N4088–X1. \[figSwift\]](comptt.pdf "fig:")
Radio counterpart of the ULX {#radiocounterpart}
----------------------------
A 1.6 GHz radio counterpart to N4088–X1 consistent with the *ROSAT* X-ray detection was reported by [@2011AN....332..379M]. The *Chandra* position of N4088–X1 obtained here reveals an offset of 6 arcsec between the X-ray position and the 1.6 GHz EVN radio source of [@2011AN....332..379M], which can now be ruled out as a candidate counterpart of the ULX.
The reanalysis of the 1.6 GHz using a FOV of 1 arcsec around the *Chandra* position yields the detection of compact radio emission consistent, within 0.3 arcsec, with the *Chandra* positional error (component A in Fig. \[radio\]). This compact component A is detected at a 5.2$\sigma$ level and is centered at RA(J2000) = 12$^h$05$^m$32$^s$.3048 $\pm$ 0.0004$^s$, Dec.(J2000) = 50$^{\circ}$3246.140 $\pm$ 0.004. It has an integrated flux density of 49 $\mu$Jy, from which we derive a 1.6 GHz luminosity $L_\mathrm{1.6 GHz}= 1.6 \times 10^{34}$ erg s$^{-1}$. The AIPS task JMFIT is used to fit an elliptical Gaussian to this peak of emission, which yields a lower limit on the brightness temperature T$_\mathrm{B}$ $>$ 3 $\times$ 10$^{4}$ K.
A second component (labeled B in Fig. \[radio\]) is detected at 5.4$\sigma$ level offset 0.6 arcsec from the *Chandra* X-ray position. Component B is centered at RA(J2000) = 12$^h$05$^m$32$^s$.2983 $\pm$ 0.0003$^s$, Dec.(J2000) = 50$^{\circ}$3245.398 $\pm$ 0.003, and has an integrated flux density of 55 $\mu$Jy, from which we derive a 1.6 GHz luminosity $L_\mathrm{1.6 GHz}= 1.8 \times 10^{34}$ erg s$^{-1}$ assuming it is in NGC4088. The fit of an elliptical Gaussian to this peak of emission yields a lower limit on the brightness temperature T$_\mathrm{B}$ $>$ 3 $\times$ 10$^{4}$ K.
![1.6 GHz EVN image at the *Chandra* position of N4088–X1. The contours are (3, 4, 5) times the off-source rms noise of 11 $\mu$Jy beam$^{-1}$. Two compact components labelled A and B are detected. A circle of radius 0.3 arcsec centered at the *Chandra* position is plotted to ease the visualization of the distance between the ULX and the radio components. The beam size shown at the bottom left corner is 33 mas $\times$ 27 mas oriented at a position angle of 86$^{\circ}$.4.\[radio\]](N4088casa2.pdf)
In the 5 GHz EVN observations, no radio emission is detected above a 5$\sigma$ level for N4088–X1 within 1 arcsec of the *Chandra* position nor within a FOV of 5 arcsec around the VLA position. An upper limit on the flux density of the ULX of 0.30 mJy beam$^{-1}$ is obtained by estimating the rms at the *Chandra* position, from which we derive an upper limit on the brightness temperature $T_\mathrm{B} < 6 \times 10^{5}$ K. Adopting a distance to NGC4088 of 13 Mpc, we derive an upper limit on the 5 GHz radio luminosity of $L_\mathrm{5 GHz} < 3.1 \times 10^{35}$ erg s$^{-1}$.
Combined with the 1.6 GHz detection, the upper limit on the flux density at 5 GHz can be used to constrain the spectral index of the source. Defining $F_{\nu}\propto\nu^{\alpha}$, we obtain $\alpha\leq1.6$, which is trivially satisfied by any physical class of radio spectra. It should be noted that this spectral index is derived from non-simultaneous observations and can thus be affected by variability effects (unless changes in the radio emission occur on timescales longer than years). Therefore, deeper 5 GHz observations simultaneous with new 1.6 GHz ones are needed before we can determine the nature of the radio emission.
X-ray luminosity function
-------------------------
### AGN fraction
In order to estimate the level of AGN contamination, we compared the flux distribution of the *Chandra* sources with the expected AGN flux distribution (@2012ApJ...752...46L). We estimate that about 2/3 of the thirty-one *Chandra* sources detected inside the full S3 chip (area = 0.0196 deg$^{2}$) are likely to be AGN. However, for the fifteen sources inside the D25 ellipse (area = 0.0028 deg$^{2}$), the AGN fraction is only $\sim$19% (i.e. $\leq$ 3 AGN). In particular, the probability of finding an AGN inside the D25 with the flux measured for the ULX is $\sim$15%. Although the probability of detecting a background AGN inside the D25 of NGC4088 is not very low, other considerations argue against this possibility for the ULX (see Section \[discussion\]).
### XLF fitting {#XLF}
The XLF of the X-ray source population of NGC4088 is constructed considering those sources inside the D25 ellipse of the galaxy. An apparent flattening of the XLF is observed at the low luminosities (see Fig. \[cumXLF\]), which may be caused by incompleteness effects. To correct for this, we eliminate the three lower points of the XLF that may be affected by incompleteness (it should be noted that the purpose of this paper is not to study the low-luminosity XLF of NGC4088). As a result, the XLF flattening disappears.
Since the errors in the cumulative XLF are not independent, from now on we use the differential XLF. This allows us to take into account the statistical uncertainties, which are large given the very small number of sources. These errors do not take into account the possibility that $\leq$ 3 sources inside the D25 are AGN.
![Cumulative XLF of NGC4088. The vertical dotted red line shows the cut performed in the XLF to correct for incompleteness.\[cumXLF\]](XLFbg_lum10_37_loglog_withoutfitLINE.pdf)
We fit the differential XLF with a power-law of the form $dN/dL_\mathrm{X}=BL_\mathrm{X}^{-\beta}$ normalized to $10^{37}$ erg s$^{-1}$, where $\beta=\alpha+1$ (Fig. \[diffXLF\]). The Cash statistic (@1979ApJ...228..939C) is used instead of the minimum $\chi^{2}$ method due to insufficient data points for the errors to be described by Gaussian statistics. The fit gives a slope of $\beta=1.5\pm 0.8$ (1$\sigma$ error), and thus $\alpha=0.5 \pm 0.8$, which is similar to the slope found in other spiral galaxies (e.g., $\alpha\sim0.4-0.5$; the Antennae galaxies, @2002ApJ...577..726Z; M82, @2007ApJ...661..135Z; M81, @2001ApJ...549L..43T). In these galaxies, the X-ray source population is mostly dominated by HMXBs. The ULX N4088–X1, with a luminosity $L_\mathrm{0.2-10.0 keV} = 3.4\ \times$ 10$^{39}$ erg s$^{-1}$, could thus be located at the high end of the HMXB distribution of NGC4088, which is consistent with the association of ULXs with young stellar populations (e.g., @2002ApJ...577..726Z; @2004ApJS..154..519S). However, the possibility that the ULX is a LMXB (e.g., @2012MNRAS.420.2969M; @2012ApJ...750..152S) cannot be ruled out.
To further study the star-formation rate (SFR) of NGC4088, we overplot the XLF of star-forming galaxies from [@2012MNRAS.419.2095M], fig. 4, black line, on the XLF of NGC4088. These authors normalize their XLF to 1 $M_{\odot}$ yr$^{-1}$. By rescaling this value to fit our data (see Fig. \[diffXLF\]), it is possible to obtain an estimate on the SFR for NGC 4088. This provides an estimated rate of 4.5 $M_{\odot}$ yr$^{-1}$, in good agreement with the range of SFRs 1.7–7.8 $M_{\odot}$ yr$^{-1}$ obtained by [@2006ApJ...643..173S] in the H$\alpha$, IR, and radio bands.
![Fit of the differential XLF inside the D25 ellipse of NGC4088 to a single power-law (red solid line). The best-fit slope is $\alpha=0.5 \pm 0.8$. The blue dotted line denotes the XLF of star-forming galaxies from [@2012MNRAS.419.2095M] scaled by a factor 4.5. \[diffXLF\]](differentialXLFbg_D25powerlawMineo.pdf)
The supernova SN2009dd
----------------------
Based on *Swift* XRT observations, reported the X-ray detection of a recent type II supernova (SN2009dd) in the galaxy NGC4088 at RA(J2000) = 12$^h$05$^m$34$^s$.10, Dec.(J2000) = 50$^{\circ}$3219.4. SN2009dd brightened from 8 $\times$ 10$^{38}$ erg s$^{-1}$ to 1.7 $\times$ 10$^{39}$ erg s$^{-1}$ in the 0.2–10 keV energy range. While *XMM-Newton* does not have enough resolution to resolve the SN from the nucleus, the *Chandra* observations of NGC4088 here presented can provide the most accurate X-ray position of SN2009dd.
In our *Chandra* observations, *wavdetect* fails to detect the SN2009dd. However, the source can be dimly seen in the 0.3–10 keV and 1.5–7.0 keV bands. Using CIAO Statistics, we obtain $\sim$8 counts in the 0.3–10 keV background subtracted image, which corresponds to an unabsorbed flux $F_\mathrm{0.3-10 keV}$ = 4.2 $\times$ 10$^{-15}$ erg cm$^{-2}$ s$^{-1}$ and luminosity $L_\mathrm{0.3-10 keV}$ = 8.4 $\times$ 10$^{37}$ erg s$^{-1}$ assuming $\Gamma = 1.8$ and Galactic column density $N_\mathrm{H} = 2 \times 10^{20}$ cm$^{-2}$. From the detected number of source counts and background counts in the source extraction area, we can estimate the 90% confidence limit to the true number of counts coming from SN2009dd from Poisson statistics (e.g., @1991ApJ...374..344K). We obtain a 90%–confidence level lower limit $\sim$3 and upper limit $\sim$13, which indicates that the source is real and not a background fluctuation. The chance probability of detecting $\geq$ 8 counts for the given background is $\sim10^{-5}$. The *Chandra* location of SN2009dd is RA(J2000) = 12$^h$05$^m$34$^s$.08, Dec.(J2000) = 50$^{\circ}$3219.0, which is consistent with the *Swift* position within the $\sim$0.6 arcsec *Chandra* absolute astrometry.
No radio emission was detected at the position of SN2009dd above a 3$\sigma$ upper limit of 0.35 mJy at 1.3 cm and of 0.15 mJy at 3.5 cm according to [@2009ATel.2016....1S], and there are no other recent observations of NGC4088 apart from the ones reported in 2009. The VLBI observations presented in this paper are centered too far from the position of SN2009dd; hence, no further upper limits to the radio flux density can be provided.
Discussion
==========
The nature of N4088–X1 {#the-nature-of-n4088x1 .unnumbered}
----------------------
The results of the XLF fitting (Section \[XLF\]) show that the XLF of NGC4088 is well described by a power-law of $\alpha$=0.5, which resembles the typical fits of HMXB XLFs and indicates that N4088–X1 could be located at the high-luminosity extension of the BH XRB distribution. These results are in agreement with studies of the XLF of star-forming galaxies (e.g., @2003MNRAS.339..793G; @2012MNRAS.419.2095M) and of the location of ULXs in stellar clusters (e.g., @2013MNRAS.432..506P), which conclude that ULXs are a high-luminosity end of the XRB population harboring most possibly stellar-mass BHs rather than IMBHs.
When analyzing the *Chandra* X-ray spectrum of N4088–X1 (Section \[xrayprop\]), we find that it is acceptably fitted both by a simple absorbed power-law continuum of $\Gamma$=1.1$^{+0.6}_{-0.5}$ and by a disk-blackbody model with $kT_\mathrm{in}$=2.5 keV. The hard power-law slope is consistent with the classification of N4088–X1 as a “hard ULX" (@2011AN....332..330S; @2013MNRAS.435.1758S). The physical interpretation of hard ULXs is still unclear: some authors (e.g., Winter et al. 2006) suggested that they are IMBHs in the low/hard state; others (e.g., @2009MNRAS.397.1836G; @2011AN....332..330S; @2013MNRAS.435.1758S) explain them instead as one possible variety of super-Eddington accretion (e.g., @2007MNRAS.377.1187P; @2009MNRAS.393L..41K). The photon index is much harder than the value of $\Gamma \sim$1.7 that is typical of the low/hard state (e.g., Remillard & McClintock 2006) but consistent with it within the 90% margin of error, which does thus not rule out the interpretation of N4088–X1 as an IMBH. A disk temperature $>$ 1 keV is however not consistent with an IMBH but favors the nature of N4088–X1 as an XRB with super-Eddington accretion. The low number of detected counts are insufficient to statistically distinguish between the models. In the spectral fitting performed on the *Swift* data, a disk-blackbody model with $kT_\mathrm{in}\sim$2 keV provided a better fit than the power-law model. A hint of a spectral cut-off at $\sim$5 keV seems to be observed in the residuals of the power-law model. This feature is similar to that observed in other ULXs (e.g., @2006ApJ...641L.125D, 2010; @2006MNRAS.368..397S; @2012MNRAS.422..990K; @2013MNRAS.435.1758S), suggesting that N4088–X1 could be in the ultraluminous state (e.g., ; @2009MNRAS.397.1836G; @2009MNRAS.398.1450K; @2013MNRAS.435.1758S) with a thermal spectrum described by Compton scattering of soft photons (e.g., @2012MNRAS.420.2969M; @2012ApJ...750..152S; ). The good fit provided by the thermal Comptonization model together with the statistically significant cut-off indicated by the F-test support this possibility, although the *Swift* data are not of high enough S/N to confirm it. Therefore, whilst the spectrum is merely quasi-thermal in shape, we are unable to confirm the accretion state at this time using the X-ray data at hand. However, we can place constraints on the nature of N4088–X1 when combining these fits with information from other wavebands.
In Section \[radiocounterpart\] we have reported the 1.6 GHz EVN detection of two compact components within the *Chandra* error circle of 0.6 arcsec. These could be either compact HII regions (e.g., see review by ), compact SNRs, or an accreting BH. Compact HII regions have typical sizes of 1–7 pc, a thermal X-ray spectrum with temperatures $>$ 2 keV, an inverted radio spectrum of spectral index $\sim$1, and $T_\mathrm{B} < 10^{4}$ K (e.g., @2001ApJ...559..864J; @2002MNRAS.334..912M; @2006ApJ...653..409T; @2007prpl.conf..181H). Compact SNRs have also typical sizes of a few pc and a steep radio spectrum (e.g., @2007AJ....133.2156L; @2013MNRAS.436.2454Mb). The limits on the brightness temperatures derived from the 1.6 GHz detections and the 5 GHz non-detection ($3 \times 10^{4} < T_\mathrm{B} < 6 \times 10^{5}$ K) together with the high X-ray luminosity of N4088–X1 make the presence of compact HII regions quite improbable and are more indicative of the presence of either compact SNRs or an accreting BH. The EVN beam size of $\sim$30 mas (which corresponds to $\sim$2 pc at the distance of the galaxy) is consistent with both a compact HII region, a compact SNR, and an accreting BH. Unfortunately, the non-detection at 5 GHz and the upper limit on the radio spectral index ($\alpha\leq$1.6) estimated for N4088–X1 does not clarify whether the radio counterpart of this ULX is steep, flat, inverted, or variable. We plan to obtain deeper radio observations to determine the spectral index and variability properties of this source and therefore constrain the physical interpretation.
In order to test if the two radio components are associated with the ULX, we derive the probability of a chance alignment between the *Chandra* counterpart and any compact radio source in the 5 arcsec field (positional error of the VLA) in that region of the galaxy. We use the number of sources detected above 5$\sigma$ in the imaged VLA field (see Section \[VLBI\]) and the *Chandra* error circle of 0.6 arcsec following the same approach as in @2013MNRAS.436.1546M (2013a). This gives a probability of chance alignment P(CA)=0.4, which is quite high and indicates that one or the two detected compact radio components (A and B, Fig. \[radio\]) could correspond to a random source (i.e. a compact SNR or BH) not associated with the ULX. Given the 0.6 arcsec offset between component B and the *Chandra* position, component B is most plausibly not related to N4088–X1 while the radio emission of component A (consistent with the *Chandra* position within 0.3 arcsec) could be coming from a BH associated with the ULX. In this case, the emission of component A could be due to flaring radio emission from a ballistic jet (e.g., @2012Sci...337..554W; @2013Natur.493..187M) or compact core emission if the source is in a low/hard state (albeit, as mentioned above, with a flatter photon spectral index than typical). The results of the X-ray spectral analysis suggest that N4088–X1 is in a thermal (i.e. Comptonized) ultraluminous state. Therefore, if the radio emission is associated with the ULX then the compact radio component is most likely associated with ballistic jet emission. If we assume for a moment that the source is residing in the low/hard state accreting at $L <$ 10% Eddington (e.g., @2003MNRAS.342.1041D), we are able to invoke the fundamental plane of accreting BHs (e.g., @2003MNRAS.345.1057M; ; @2012MNRAS.423..590G) and estimate a BH mass. For this, we use the 2–10 keV X-ray flux obtained from the *Chandra* spectral fitting and scale the 1.6 GHz flux density to 5 GHz using a radio spectral index $\alpha = 0.15$ (a typical spectral index for flat cores used to estimate the BH mass from the Fundamental Plane, e.g., ). Using the Fundamental Plane of that presents the least scatter, an upper limit on the BH mass of $3 \times 10^{5} M_{\odot}$ is obtained. This is consistent with this source being either an IMBH or an XRB, and rules out the nature of N4088–X1 as a SMBH.
We also derive the ratio $R_\mathrm{X}$ of 5 GHz radio emission to 2–10 keV X-ray emission defined by [@2003ApJ...583..145T]. Typical values of this ratio for XRBs are $R_\mathrm{X} < -5.3$ (e.g., @2011Natur.470...66R), while values of $R_\mathrm{X}=-2.7$ to $-2$ have been estimated in SNRs (e.g., @2003ApJ...599.1043N; @2013MNRAS.436.2454Mb). For low-luminosity AGN (LLAGN), Ho (2008) reports a range of values $-3.8 < R_\mathrm{X} < -2.8$ (see also @2013MNRAS.436.1546Ma, table 3). For N4088–X1, we obtain $R_\mathrm{X} < -4.7$ using the 2–10 keV X-ray flux obtained from the *Chandra* spectral fitting and the 5 GHz- scaled radio luminosity. This is in agreement with the previous results, ruling out both a SMBH and a SNR nature for this ULX.
The location of N4088–X1 in the spiral arm of the host galaxy, possibly within an HII region (e.g., ), and the low X-ray absorption seen in the *Chandra* and *Swift* data also argue very strongly against a LLAGN background scenario. On the other hand, the lack of a bright counterpart to N4088–X1 in the optical image makes it very unlikely to be a foreground star.
Conclusions
===========
We have presented the first *Chandra* and *Swift* X-ray observations of the galaxy NGC4088 and the ULX N4088–X1 that it hosts. EVN observations at 1.6 and 5 GHz of the ULX were performed almost simultaneously to the *Swift* and *Chandra* observations, respectively, which have allowed us to investigate the compact radio emission of a ULX radio counterpart previously proposed with the VLA.
The X-ray spectral analysis of N4088–X1 seems to favor a thermally Comptonized spectrum for this source, although the possibility that it is a hard ULX cannot be ruled out. The disk temperature ($kT_\mathrm{in}\sim$2 keV) obtained from the disk-blackbody model and the presence of a statistically significant spectral break at $\sim$5 keV are not consistent with N4088–X1 being an IMBH but suggest that the source could be an XRB in a super-Eddington ultraluminous state. Unfortunately, this cannot be confirmed with the present data due to the low S/N. If the source is in an ultraluminous state, the detection of compact radio emission at 1.6 GHz coincident with the *Chandra* counterpart could then correspond to ballistic jet emission from an accreting BH. Multi-epoch multi-wavelength observations are required to confirm this.
Finally, the detection of fifteen sources within the D25 ellipse of NGC4088 has allowed us to fit the XLF of this galaxy and estimate a SFR of 4.5 $M_{\odot}$ yr$^{-1}$. We find that the XLF resembles that of typical star-forming galaxies, where the ULX N4088–X1 could be at the high-luminosity end of the XRB population. We thus conclude that N4088–X1 is possibly a HMXB with a thermally Comptonized spectrum and either approaching the Eddington limit or in the ultraluminous state.
Acknowledgements
================
This work was partially supported by the Chandra X-Ray Center (CXC), which is operated by the Smithsonian Astrophysical Observatory (SAO) under NASA contract NAS8-03060, and by Chandra Director Discretionary Time grant DD2-13063X. MM acknowledges finantial support by PAYA2o11-25527
J.C.G. would like to acknowledge Avadh Bhatia Fellowship, the Alberta Ingenuity New Faculty Award, and the financial support from NSERC Discovery Grants. SAF is the recipient of an Australian Research Council Postdoctoral Fellowship, funded by grant DP110102889. RS acknowledges support from the Australian Research CouncilÕs Discovery Projects funding scheme (project number DP120102393).
[^1]: www.evlbi.org
[^2]: Very Large Array of the National Radio Astronomy Observatory (NRAO).
[^3]: Faint Images of the Radio Sky at Twenty-cm survey.
[^4]: http://cxc.harvard.edu/ciao/threads/
[^5]: $N_\mathrm{H}$ calculated using the COLDEN tool: http://cxc.harvard.edu/toolkit/colden.jsp
[^6]: http://heasarc.nasa.gov/Tools/w3pimms.html
[^7]: http://swift.ac.uk/user\_objects/
[^8]: Joint Institute for VLBI in Europe, Dwingeloo, the Netherlands.
[^9]: Astronomical Image Processing Software of NRAO.
[^10]: No channel averaging was applied to the calibrated data to avoid degradation of the synthesized beam away from the phase center (bandwidth smearing).
[^11]: The significance of a model over another is ‘preferred’ for BIC values between 0–2, ‘positive’ for BIC values between 2–6, ‘strong’ for values 6–10, and ‘very’ strong for BIC $>$ 10
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The algebra $H^\infty(D)$ of bounded holomorphic functions on $D\subset\mathbb C$ is projective free for a wide class of infinitely connected domains. In particular, for such $D$ every rectangular left-invertible matrix with entries in $H^\infty(D)$ can be extended in this class of matrices to an invertible square matrix (the generalization of the corona theorem for $H^\infty(D)$). This follows from a new result on the structure of the maximal ideal space of $H^\infty(D)$ asserting that its covering dimension is $2$ and the second Čech cohomology group is trivial.'
address: 'Department of Mathematics and Statistics University of Calgary Calgary, Alberta, Canada T2N 1N4'
author:
- Alexander Brudnyi
title: Projective Freeness of Algebras of Bounded Holomorphic Functions on Infinitely Connected Domains
---
[^1]
Formulation of the Main Results
===============================
Let $H^\infty(X)$ be the Banach algebra of bounded holomorphic functions on a complex manifold $X$ equipped with pointwise multiplication and supremum norm and let $\mathfrak M(H^\infty(X))$ be the [*maximal ideal space*]{} of $H^\infty(X)$, i.e., the set of nonzero homomorphisms $H^\infty(X)\rightarrow{{\mathbb C}}$ endowed with the weak$^*$ topology of the dual space $H^\infty(X)^*$ (the [*Gelfand topology*]{}). It is a compact Hausdorff space contained in the closed unit ball of $H^\infty(X)^*$. If $X$ is [*Caratheodory hyperbolic*]{} (i.e., $H^\infty(X)$ separates the points of $X$), it can be identified with an open subset of $\mathfrak M(H^\infty(X))$ formed by evaluation functionals at points of $X$. Then the [*corona problem*]{} asks whether $X$ is dense in $\mathfrak M(H^\infty(X))$. The most famous corona problem for $H^\infty$ on the unit disk ${\mathbb{D}}$ was posed by Kakutani in 1941 and solved positively by Carleson [@C] in 1962 (see the book [@DKSTW] and references therein for other significant results in this area). The general corona problem for algebras of bounded holomorphic functions on plane domains is still open as is the problem in several variables for the ball and polydisk.
Denseness of $X$ in $\mathfrak M(H^\infty(X))$ can be equivalently reformulated as follows, see, e.g., [@Ga Ch.V]:
For every family $f_1,\dots, f_n\in H^\infty(X)$, $n\in{ \mathbb{N}}$, satisfying the [*corona condition*]{} $$\label{eq1.2}
\sup_{x\in X}\max_{1\le i\le n}|f_i(x)|>0$$ there exist $g_1,\dots, g_n\in H^\infty(X)$ such that $$\label{eq1.3}
\sum_{i=1}^n\, f_i\!\cdot\! g_i=1.$$
A more general problem on extension of matrices over $H^\infty(X)$ is as follows:
Let $A$ be a $k\times n$ matrix, $k<n$, with entries in $H^\infty(X)$ such that the family of $k$-minors satisfies the corona condition. Is there a $n\times n$ matrix $B$ with entries in $H^\infty(X)$ and ${\rm det}(B)=1$ extending $A$ (i.e., containing $A$ as a submatrix)?
If this is true for all such matrices $A$ with $k,n\in{ \mathbb{N}}$, then $H^\infty(X)$ is called [*Hermite*]{}. Equivalently, $H^\infty(X)$ is Hermite if every finitely generated [*stably free*]{} $H^\infty(X)$-module is free.
It was first shown by Tolokonnikov that $H^\infty(X)$ is Hermite for $X$ being a Caratheodory hyperbolic Riemann surface of finite type [@To1 Thm.3] or a plain domain obtained by deleting from $\mathbb D^*:={\mathbb{D}}\setminus\{0\}$ a hyperbolically-rare sequence of closed disks converging to $\{0\}$, [@To2 Cor.]. Later in [@Br2], [@Br4] the author proved the following generalizations of [@To1 Thm.3].
Let $N$ be a bordered Riemann surface. A connected Riemann surface $U$ is called an [*$N$-domain*]{} if there is a holomorphic embedding of $U$ in an unbranched covering of $N$ inducing monomorphism from $\pi_1(U)$ to the fundamental group of the covering.
Let $\mathscr U=\{U_\alpha\}_{\alpha\in\Lambda}$ be a family of (not necessarily distinct) $N$-domains and $X_{\mathscr U}:=\sqcup_{\alpha\in\Lambda} U_\alpha$. Then the Banach algebra $H^\infty(X_{\mathscr U})$ is Hermite, see [@Br2 Thm.1.1].
This result was used by the author to prove a similar result for $H^\infty$ on unbranched coverings of Riemann surfaces of finite type. Specifically, let $N$ be a Caratheodory hyperbolic Riemann surface of finite type. Let $\mathscr U=\{U_\alpha\}_{\alpha\in\Lambda}$ be a family of (not necessarily distinct) unbranched coverings of $N$ and $X_{\mathscr U}:=\sqcup_{\alpha\in\Lambda} U_\alpha$. Then the Banach algebra $H^\infty(X_{\mathscr U})$ is Hermite, see [@Br4 Thm.1.2].
The concept of a Hermite ring is a weaker notion than that of a projective free ring. A commutative ring with identity $R$ is said to be [*projective free*]{} if every finitely generated projective $R$-module is free (i.e., if $M$ is an $R$-module such that $M \oplus N\cong R^{d}$ for an $R$-module $N$ and $d\in{\mathbb{Z} }_+$, then $M\cong R^{k}$ for some $k\in{\mathbb{Z} }_+$). Since every stably free module is projective, every projective free ring is Hermite. In terms of matrices, $R$ is projective free iff every nontrivial square idempotent matrix over $R$ is similar (by an invertible matrix over $R$) to a matrix of the form $$\textrm{diag}(I_m,0):=\left[ \begin{array}{cc} I_m & 0 \\ 0 & 0
\end{array} \right],\quad m\in{ \mathbb{N}},$$ where $I_m$ is the $m\times m$ identity matrix, see, e.g., [@Co Prop.2.6].
If $A$ is a projective free complex commutative unital Banach algebra, then its maximal ideal space $\mathfrak M(A)$ is connected, and the Čech cohomology group $H^2(\mathfrak M(A),{\mathbb{Z} })=0$, see, e.g., [@Br5 Sect.3] for the corresponding references.
In [@BS Thm.1.5] the authors proved that $H^\infty(U)$ is projective free for an $N$-domain $U$. The proof relies on the Lax–Halmos-type theorem [@Br2 Thm.1.7]. In fact, using uniform estimates of that theorem and arguing as in the proof of [@BS Thm.1.5] one obtains a more general statement generalizing [@Br2 Thm1.1]. Specifically, let $\mathscr U=\{U_\alpha\}_{\alpha\in\Lambda}$ be a family of (not necessarily distinct) $N$-domains and $X_{\mathscr U}:=\sqcup_{\alpha\in\Lambda} U_\alpha$. Let $F$ be a nontrivial matrix idempotent on $X_{\mathscr U}$ with entries in $H^\infty(X_{\mathscr U})$ such that $F(X_{\mathscr U})$ is connected. Then $F$ is similar (by an invertible matrix over $H^\infty(X_{\mathscr U})$) to a matrix $\textrm{diag}(I_m,0)$, $m\in{ \mathbb{N}}$.
In the present paper, continuing this line of research, we prove projective freeness of $H^\infty$ on a new wide class of plain domains, the so-called $\mathscr B$-domains, introduced and studied by Behrens [@Be]. By definition, $U$ is a $\mathscr B$-domain if it is obtained from a domain $V\subset{{\mathbb C}}$ by deleting a (possibly finite) [*hyperbolically-rare*]{} sequence of closed disks $\{\Delta_n\}\subset V$ with centers $\alpha_n$, i.e., such that there are disjoint closed disks $D_n$ with centers $\alpha_n$ satisfying $\Delta_n\Subset D_n\subset V$ and $$\label{equ1.1}
\sum\, \frac{{\rm rad}(\Delta_n)}{{\rm rad}(D_n)}<\infty.$$ Behrens [@Be Thm.6.1] proved that if a $\mathscr B$-domain $U$ is constructed from a domain $V$ for which the corona theorem is valid (i.e., $V$ is dense in $\mathfrak M(H^\infty(V))$), then it is valid for $U$ as well. In this case, he also described the topological structure of $\mathfrak M(H^\infty(U))$.
The main result of our paper reveals some additional topological properties of $\mathfrak M(H^\infty(U))$. For its formulation, recall that for a normal space $X$, ${\rm dim}\,X \le n$ if every open cover of $X$ can be refined by an open cover whose order $\le n + 1$. If ${\rm dim}\,X \le n$ and the statement ${\rm dim}\,X \le n-1$ is false, then ${\rm dim}\,X = n$.
\[te1.1\] Suppose $U$ is obtained from a domain $V\subset\mathbb C$ by deleting a (possibly finite) hyperbolically-rare sequence of closed disks such that
- $V$ is dense in $\mathfrak M(H^\infty(V))$;
- The covering dimension ${\rm dim}\, \mathfrak M(H^\infty(V)) =2$;
- The Čech cohomology group $H^2(\mathfrak M(H^\infty(V)),{\mathbb{Z} })=0$.
Then ${\rm dim}\, \mathfrak M(H^\infty(U)) =2$ and $H^2(\mathfrak M(H^\infty(U)),{\mathbb{Z} })=0$ as well.
As a direct corollary of Theorem \[te1.1\] we obtain the following.
\[cor1.2\] Under conditions of the theorem the Banach algebras $H^\infty(V)$ and $H^\infty(U)$ are projective free.
For instance, taking here $V:={\mathbb{D}}^*$ and $U$ constructed from $V$ by deleting a hyperbolically-rare sequence of closed disks converging to $\{0\}$ we obtain the generalization of [@To2 Cor.] because $\mathfrak M(H^\infty({\mathbb{D}}))$ satisfies the conditions of the theorem due to the classical work of Suárez [@S] (see also [@Br5] and references therein). Moreover, it is easily deduced from [@S] that $\mathfrak M(H^\infty(V))$ satisfies the conditions of the theorem for $V$ being a Caratheodory hyperbolic Riemann surface of finite type.
Up until now nothing was known about covering dimension and Čech cohomology groups of $\mathfrak M(H^\infty(V))$ for infinite type Riemann surfaces $V$. Theorem \[te1.1\] fills in this gap and provides many examples of domains satisfying conditions (i)–(iii).
Another class of examples of a different nature of Riemann surfaces satisfying these conditions is given by the following result.
\[appen\] Let $C$ be an unbranched covering of a bordered Riemann surface. Then
- $C$ is dense in $\mathfrak M(H^\infty(C))$;
- ${\rm dim}\,\mathfrak M(H^\infty(C))=2$;
- $H^2(\mathfrak M(H^\infty(C)),{\mathbb{Z} })=0$.
Thus the class of domains $V$ in Theorem \[te1.1\] includes plain unbranched coverings of bordered Riemann surfaces and domains obtained from them by deleting compact subsets of the analytic capacity zero (e.g., totally disconnected compact subsets), see Example \[ex1.3\]. Starting from such a $V$ one can construct a descending chain of $\mathscr B$-domains $V:=V_0\supset V_1\supset V_2\supset\dots\supset V_n$, $n\in{ \mathbb{N}}$, such that each $V_i$ is defined by deleting a hyperbolically-rare sequence of closed disks and then a compact subset of the analytic capacity zero from $V_{i-1}$. Then all $V_i$ satisfy assumptions of Theorem \[te1.1\] and, in particular, all algebras $H^\infty(V_i)$ are projective free (and so Hermite).
\[ex1.3\]
It is well known that every bordered Riemann surface $S$ is a domain in a compact Riemann surface $R$ such that $R\setminus S$ is the disjoint union of finitely many disks with analytic boundaries. Each $S$ is the quotient of a plane domain $\Omega$ by the discrete action of a Schottky group $G$ (the free group with $g$ generators, where $g$ is the genus of $S$) by Möbius transformations, see, e.g., [@M]. The corresponding quotient map $r: \Omega\rightarrow S$ determines the regular covering of $S$ with the deck transformation group $G$. Then $V:=r^{-1}(R)\subset \Omega$ is a regular covering of $S$ satisfying conditions of Theorem \[te1.1\]. By definition, $V$ is the complement in $\Omega$ of the finite disjoint union of $G$-orbits of compact simply connected domains with analytic boundaries biholomorphic by $r$ to the connected components of $R\setminus S$.
Further, if we consider the universal covering $r_u: X\rightarrow S$ of $S$ (where $X={\mathbb{D}}$ if $g\ge 2$, $X={{\mathbb C}}$ if $g=1$ and $X={{\mathbb C}}\mathbb P$ if $g=0$), then $V:=r_u^{-1}(R)\subset X$ satisfies conditions of Theorem \[te1.1\] as well. Here $V$ is the complement in $X$ of the finite disjoint union of orbits under the action by Möbius transformations of the fundamental group $\pi_1(R)$ of $R$ of compact simply connected domains with analytic boundaries biholomorphic by $r_u$ to the connected components of $R\setminus S$.
Using such $V$ we can define a descending chain of $\mathscr B$-domains satisfying conditions of Theorem \[te1.1\] $V:=V_0\supset V_1\supset V_2\supset\dots\supset V_n$, $n\in{ \mathbb{N}}$, such that each $V_i$ is constructed by deleting a hyperbolically-rare sequence of closed disks and then a compact subset of the analytic capacity zero from $V_{i-1}$.
Other examples of Riemann surfaces satisfying conditions of Theorem \[te1.1\] will be presented in a forthcoming paper.
In light of these results the following question seems plausible.
Are there plain domains $D$ for which either ${\rm dim}\,\mathfrak M(H^\infty(D))> 2$ or the Čech cohomology group $H^2(\mathfrak M(H^\infty(D)),{\mathbb{Z} })$ is not trivial?
Maximal Ideal Space of $\mathbf{H^\infty(\mathbb D\times{ \mathbb{N}})}$
========================================================================
The proof of Theorem \[te1.1\] relies on some properties of the maximal ideal space of the algebra $H^\infty(\mathbb D\times{ \mathbb{N}})$. Previously, some results on the structure of this space were obtained in [@Be Sect.4]. In particular, the corona theorem for $H^\infty(\mathbb D\times{ \mathbb{N}})$ is valid and follows from Carleson estimates for solutions of the corona problem for $H^\infty({\mathbb{D}})$, see, e.g., [@Be Thm.4.2]. The main result used in our proofs provides an additional information on the structure of this object.
\[te2.1\] We have
- ${\rm dim}\,\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}))=2$;
- $H^2(\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}),{\mathbb{Z} })=0$.
Part (a) of the theorem follows from a result of independent interest describing the maximal ideal space $\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}))$ by means of $\mathfrak M(H^\infty({\mathbb{D}}))$.
\[teo3.1\] $\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}))$ can be covered by interiors of two compact subsets homeomorphic to a subset of $\mathfrak M(H^\infty({\mathbb{D}}))$.
Since ${ \mathbb{N}}$ and ${\mathbb{Z} }$ are both countable, manifolds ${\mathbb{D}}\times{ \mathbb{N}}$ and ${\mathbb{D}}\times{\mathbb{Z} }$ are biholomorphic. Hence, $\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}))$ and $\mathfrak M(H^\infty({\mathbb{D}}\times{\mathbb{Z} }))$ are homeomorphic. It is more convenient for us to work with the latter space.
In what follows, ${\mathbb{D}}_r(c):=\{z\in{{\mathbb C}}\, :\, |z-c|<r\}$, $r>0$, $c\in{{\mathbb C}}$, i.e., ${\mathbb{D}}:={\mathbb{D}}_1(0)$. For a subset $S$ of a topological space we denote by $\bar S$ and $\mathring S$ its closure and interior. Also, we set $H^\infty:=H^\infty({\mathbb{D}})$.
Let $$\label{equ3.1}
\Omega:=\bigl({\mathbb{D}}\cup{\mathbb{D}}_1({\scriptstyle \frac 3 2})\bigr)\setminus\bigl\{\mbox{$ \frac 3 2$}\bigr\}.$$ The fundamental group $\pi_1(\Omega)$ of $\Omega$ is isomorphic to ${\mathbb{Z} }$, i.e., $\pi_1(\Omega)=\{a^n\}_{n\in{\mathbb{Z} }}$ for some $a\in \pi_1(\Omega)$. Let $r_u:{\mathbb{D}}\rightarrow\Omega$ be the universal covering of $\Omega$. The deck transformation group $\pi_1(\Omega)$ acts discretely on $\mathbb D$ by Möbius transformations. Since $r_u\in H^\infty$, it extends to a function $\hat r_u\in C(\mathfrak M(H^\infty))$ such that $\hat r_u(\mathfrak M(H^\infty))=\bar\Omega$. Let $U:=r_u^{-1}({\mathbb{D}})\subset{\mathbb{D}}$. Since each loop in ${\mathbb{D}}$ is contractible in $\Omega$, $$\label{equ3.1a}
U=\bigsqcup_{g\in\pi_1(\Omega)}g(U')$$ for some $U'\subset{\mathbb{D}}$ biholomorphic to ${\mathbb{D}}$ via $r_u$. In particular, the map $s: U\rightarrow {\mathbb{D}}\times{\mathbb{Z} }$, $$\label{equ3.2}
s(z):=(r_u(z),n),\quad z\in a^n(U'),\quad n\in{\mathbb{Z} },$$ is biholomorphic; hence, the pullback by $s$ defines an isomorphism of Banach algebras $s^*: H^\infty(\mathbb D\times{\mathbb{Z} })\rightarrow H^\infty(U)$.
We denote by $\tilde s: \mathfrak M(H^\infty(U))\rightarrow \mathfrak M(H^\infty({\mathbb{D}}\times{\mathbb{Z} }))$, $\tilde s|_U=s$, the homeomorphism of the maximal ideal spaces induced by the transpose $(s^*)^*$ of $s^*$.[^2]
Next, we consider the function $h:={\rm Re}\, r_u$ and its extension $\hat h:={\rm Re}\, \hat r_u\in C(\mathfrak M(H^\infty))$. By the definition of $\Omega$, the open set $$\label{equ3.3}
U_1:=\left\{z\in{\mathbb{D}}\, :\, h(z)<\mbox{$\frac 3 4$}\right\}\subset U$$ is the preimage under $r_u$ of the set $\{z\in{\mathbb{D}}\, :\, {\rm Re}(z)<\frac 3 4\}\subset{\mathbb{D}}$. By the corona theorem $U_1$ is dense in $$\label{equ3.4}
\widetilde U_1:=\left\{\xi\in \mathfrak M(H^\infty)\, :\, \hat h(\xi)<\mbox{$\frac 3 4$}\right\}.$$ Moreover, by [@S Thm.3.2], each $f\in H^\infty(U_1)$ extends to a (unique) $\hat f\in C(\widetilde U_1)$. In particular, this is valid for $f\in H^\infty(U)|_{U_1}$. This determines an isometric homomorphism of Banach algebras $e: H^\infty(U)\rightarrow C(U\cup \widetilde U_1)$ whose transpose $e^*$ induces a continuous injection of $U\cup\widetilde U_1$ into $\mathfrak M(H^\infty(U))$ such that $e^*|_U={\rm id}|_{U}$.
Further, $$\label{equ3.5}
\widetilde V:=\left\{\xi\in \mathfrak M(H^\infty)\, :\, \hat h(\xi)\le\mbox{$\frac 1 2 $}\right\}\subset\widetilde U_1.$$ Then $e^*$ maps the compact set $\widetilde V$ homeomorphically onto its image in $\mathfrak M(H^\infty(U))$.
We have $$(s\circ e^*)(U)={\mathbb{D}}\times{\mathbb{Z} }\quad {\rm and}\quad p\circ s\circ e^*|_U=r_u|_U,$$ where $p:{\mathbb{D}}\times{\mathbb{Z} }\rightarrow{\mathbb{D}}$ maps $(z,n)$ to $z$, $z\in {\mathbb{D}}$, $n\in{\mathbb{Z} }$.
Thus, $$\hat p\circ \tilde s\circ e^*|_{U\cup \widetilde U_1}=\hat r_u|_{U\cup \widetilde U_1};$$ here $\hat p: \mathfrak M(H^\infty ({\mathbb{D}}\times{\mathbb{Z} }))\rightarrow\bar{\mathbb{D}}$ is the extension of $p$ via the Gelfand transform. In particular, if $S:=\{z\in\bar {\mathbb{D}}\, :\, {\rm Re}(z)\le\frac 1 2\}$, then $\tilde s\circ e^*$ maps $\widetilde V$ homeomorphically onto $\hat p^{-1}(S)$. We set $S_o:=\{z\in\bar{\mathbb{D}}\, :\, -z\in S\}:=\{z\in\bar {\mathbb{D}}\, :\, {\rm Re}(z)\ge -\frac 1 2\}$. Then $\bar {\mathbb{D}}=\mathring S\cup \mathring S_o$ so that $$\mathfrak M(H^\infty ({\mathbb{D}}\times{\mathbb{Z} }))=\hat p^{-1}(\bar{\mathbb{D}})=\hat p^{-1}(\mathring S)\cup \hat p^{-1}(\mathring S_o)=\hat p^{-1}(S)\cup \hat p^{-1}(S_o),$$ where each term in the last expression is homeomorphic to $\widetilde V$.
\(a) Since $\mathfrak M(H^\infty ({\mathbb{D}}\times{\mathbb{Z} }))=\hat p^{-1}(S)\cup \hat p^{-1}(S_o)$, where each of the subsets is homeomorphic to the compact subset $\widetilde V\subset\mathfrak M(H^\infty)$, see above, and ${\rm dim}\, \mathfrak M(H^\infty)=2$, ${\rm dim}\,\mathfrak M(H^\infty ({\mathbb{D}}\times{\mathbb{Z} }))\le {\rm dim}\,\widetilde V\le 2$. It equals 2 as $\mathfrak M(H^\infty ({\mathbb{D}}\times{\mathbb{Z} }))$ contains the 2-dimensional subset ${\mathbb{D}}\times{\mathbb{Z} }$.
\(b) Since ${\rm dim}\,\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}))=2$, by the Hopf theorem, see, e.g., [@Hu], elements of the cohomology group $H^2(\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}})),{\mathbb{Z} })$ are in a one-to-one correspondence with elements of the set $[\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}})),\mathbb S^2]$ of homotopy classes of continuous maps of $\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}))$ to the two-dimensional unit sphere $\mathbb S^2$. In turn, according to the Novodvorskii-Taylor theory, there is a bijection of $[\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}})),\mathbb S^2]$ onto the set $[{\rm ID}_1(H^\infty({\mathbb{D}}\times{ \mathbb{N}})_2)]$ of connectivity components of the class of idempotents consisting of $2\times 2$ matrices with entries in $H^\infty({\mathbb{D}}\times{ \mathbb{N}})$ of constant rank $1$, see [@Ta Sec.5.3,page186]. Thus to show that $H^2(\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}})),{\mathbb{Z} })=0$ we must prove that each idempotent $F\in {\rm ID}_1(H^\infty({\mathbb{D}}\times{ \mathbb{N}})_2)$ is similar (by an invertible $2\times 2$ matrix with entries in $H^\infty({\mathbb{D}}\times{ \mathbb{N}})$) to a matrix of the form $$\label{eq7.1}
{\rm diag}(1,0):=\left[
\begin{array}{cc}
1&0\\
0&0
\end{array}
\right].$$ The proof follows the lines of the proof of the Theorem in [@Br5 Sect.5].
We set $$\label{eq7.2}
F_m:=F|_{{\mathbb{D}}\times\{m\}},\quad m\in{ \mathbb{N}}.$$ Then $N_{1m}=\ker(F_m)$ and $N_{2m}=\ker(I_2-F_m)$, where $I_2$ is the identity $2\times 2$ matrix, are weak$^*$ closed $H^\infty$-submodules of $H_2^\infty\, (:= H^\infty\oplus H^\infty)$ (i.e., if $\{f_n\}_{n\in{ \mathbb{N}}}\subset N_{im}$ is a bounded sequence pointwise converging to $f\in H_2^\infty$, then $f\in N_{im}$). Since $F_m$ is of constant rank $1$, the famous Beurling-Lax-Halmos theorem, see, e.g., [@Ni], [@To1 p.1025], implies that $N_{im}=H_{im}\cdot H^\infty$, where $H_{im}$ is a $2\times 1$ matrix with entries in $H^\infty$ of constant rank $1$ such that $$\label{eq7.3}
H_{im}^* \cdot H_{im}=1\quad {\rm a.e.\ on}\ \mathbb S:=\{z\in{{\mathbb C}}\, :\, |z|=1\}.$$ Hence, since the columns of $I_2-F_m$ and of $F_m$ belong to $N_{1m}$ and $N_{2m}$, respectively, $I_2-F_m=H_{1m}\cdot G_{1m}$ and $F_m=H_{2m}\cdot G_{2m}$ for some $1\times 2$ matrices $G_{im}$ with entries in $H^\infty$ of constant rank $1$. Then we have $$\label{eq7.4}
\begin{array}{l}
\displaystyle H_{1m}\cdot G_{1m}=(I_2-F_m)=(I_2-F_m)^2=H_{1m}\cdot (G_{1m}\cdot H_{1m})\cdot G_{1m},\\
\\
H_{2m}\cdot G_{2m}=F_m=F_m^2=H_{2m}\cdot (G_{2m}\cdot H_{2m})\cdot G_{2m},\\
\\
(H_{1m}\cdot G_{1m})(H_{2m}\cdot G_{2m})=(H_{2m}\cdot G_{2m})(H_{1m}\cdot G_{1m})=0.
\end{array}$$ These, and the maximum modulus principle for $H^\infty$ imply that $$\label{eq7.5}
G_{1m}\cdot H_{1m}=1,\quad G_{2m}\cdot H_{2m}=1,\quad G_{1m}\cdot H_{2m}=G_{2m}\cdot H_{1m}=0.$$ Let us define $2\times 2$ matrices with entries in $H^\infty$ $$H_m:=(H_{1m}\, H_{2m})\quad {\rm and}\quad G_m:=\left(\!\!\!
\begin{array}{c}
G_{1m}\\
G_{2m}
\end{array}
\!\!\!\right).$$ Then , the definitions of $G_{im}$ and imply that $$\label{eq7.6}
\begin{array}{l}
\displaystyle
G_m=H_m^{-1},\qquad H_{m}^{-1}\cdot F_m\cdot H_m={\rm diag}(1,0)\quad {\rm and}\\
\\
\displaystyle \|H_m\|_{{\scriptstyle H_{2}^\infty\rightarrow H_2^\infty}}\le 1,\qquad \|H_m^{-1}\|_{{\scriptstyle H_{2}^\infty\rightarrow H_2^\infty}}\le 1+\|F_m\|_{{\scriptstyle H_{2}^\infty\rightarrow H_2^\infty}}.
\end{array}$$ Finally, let us define a $2\times 2$ matrix $H$ on ${\mathbb{D}}\times{ \mathbb{N}}$ by the formula $$\label{eq7.7}
H|_{{\mathbb{D}}\times\{m\}}=H_m,\quad m\in{ \mathbb{N}}.$$ Since $\sup_{m\in{ \mathbb{N}}}\|F_m\|_{{\scriptstyle H_{2}^\infty\rightarrow H_2^\infty}}<\infty$, shows that $H$ is an invertible $2\times 2$ matrix with entires in $H^\infty({\mathbb{D}}\times{ \mathbb{N}})$ such that $H^{-1}|_{{\mathbb{D}}\times\{m\}}=H_m^{-1}$, $m\in{ \mathbb{N}}$, and $H^{-1}\cdot F\cdot H={\rm diag}(1,0)$.
This completes the proof of part (b) of the theorem.
Proofs of Theorem \[te1.1\] and Corollary \[cor1.2\]
====================================================
Structure of $\mathfrak M(H^\infty(U))$
---------------------------------------
Let $U$ be a $\mathscr B$-domain obtained from a domain $V\subset{{\mathbb C}}$ by deleting a (possibly finite) hyperbolically-rare sequence of closed disks $\{\Delta_n\}\subset V$ with centers $\alpha_n$ and let $V$ be dense in $\mathfrak M(H^\infty(V))$. Then by [@Be Thm.6.1] $U$ is dense in $\mathfrak M(H^\infty(U))$ and the latter space has the described below structure.
Let $R: \mathfrak M(H^\infty(U))\rightarrow \mathfrak M(H^\infty(V))$ be the continuous map transposed to the restriction homomorphism $H^\infty(V)\rightarrow H^\infty(U)$, $f\mapsto f|_U$. If the sequence $\{\alpha_n\}$ is infinite, we denote by $F:\beta{ \mathbb{N}}\rightarrow\mathfrak M(H^\infty(V))$ the continuous extension of the map ${ \mathbb{N}}\rightarrow V$, $n\mapsto\alpha_n$, to the Stone-Čech compactification of ${ \mathbb{N}}$ and set $$\label{Set}
S:=F(\beta{ \mathbb{N}}\setminus{ \mathbb{N}})\subset \mathfrak M(H^\infty(V))\setminus V.$$ Under these notations we have following properties.
- There is a continuous embedding $i:\bigl(\mathfrak M(H^\infty(V))\setminus V\bigr)\cup U\hookrightarrow\mathfrak M(H^\infty(U))$ such that $R\circ i={\rm id}$ and $i$ is invertible on $\bigl(\mathfrak M(H^\infty(V))\setminus (V\cup S)\bigr)\cup U$, [@Be Thm.3.1].
- There is a continuous mapping $G:\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}))\setminus ({\mathbb{D}}\times{ \mathbb{N}})\rightarrow R^{-1}(S)$ which maps $\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}))\setminus \bigl(({\mathbb{D}}\times{ \mathbb{N}})\cup (\{0\}\times (\beta{ \mathbb{N}}\setminus{ \mathbb{N}}))\bigr)$ homeomorphically onto $R^{-1}(S)\setminus i(S)$ and $\{0\}\times (\beta{ \mathbb{N}}\setminus{ \mathbb{N}})$ surjectively onto $i(S)$, [@Be Thm.6.2].
- $R^{-1}(\partial\Delta_n)$ is homeomorphic to $\mathfrak M(H^\infty({\mathbb{D}}))\setminus{\mathbb{D}}$ for all $n\in{ \mathbb{N}}$.
The last property easily follows from the decomposition of $H^\infty(U)$, see [@Be Sec.2].
In turn, if the sequence $\{\alpha_n\}$ is finite, then $\mathfrak M(H^\infty(U))$ satisfies property (1) with $S=\emptyset$ and property (2).
Proof of Theorem \[te1.1\]
--------------------------
We prove the theorem under the condition that $\{\alpha_n\}$ is infinite. If it is finite, then the proof is simpler but still repeats some of the arguments presented below. We leave the details to the reader.
First, we prove that under conditions of the theorem ${\rm dim}\,\mathfrak M(H^\infty(U))=2$.
To this end, it suffices to prove that the compact set $K_1:=R^{-1}((V\setminus U)\cup S)$ has dimension $\le 2$ and each compact subset of the open set $K_2:=\mathfrak M(H^\infty(U))\setminus K_1$ has dimension $\le 2$ as well, see, e.g., [@N Ch.2,Thm.9-11].
By this definition, $$K_2=R^{-1}\bigl(U\cup \bigl(\mathfrak M(H^\infty(V))\setminus (V\cup S)\bigr)\bigr).$$ Hence, since $R|_{U}={\rm id}$, due to property (1) each compact subset $Z\subset K_2$ is homeomorphic to a compact subset of $U\cup (\mathfrak M(H^\infty(V))\setminus (V\cup S))\subset \mathfrak M(H^\infty(V))$. In particular, ${\rm dim}\, Z\le 2$ because ${\rm dim}\,\mathfrak M(H^\infty(V))=2$ by the hypotheses of the theorem.
Next, $$\label{ka1}
K_1=\bigsqcup_n R^{-1}(\partial\Delta_n)\sqcup R^{-1}(S);$$ here $\partial \Delta_n$ stands for the boundary of $\Delta_n$.
Hence, due to [@N Ch.2,Thm.9-11] ${\rm dim}\, K_1\le 2$ iff ${\rm dim}\, R^{-1}(S)\le 2$ and ${\rm dim}\, Z\le 2$ for each compact subset $Z\subset\sqcup_n\, R^{-1}(\partial\Delta_i)$.
Since each $R^{-1}(\partial\Delta_i)$ is a relatively clopen subset of $K_1$, there is some $n=n(Z)$ such that $Z\subset \sqcup_{1\le k\le n}\, R^{-1}(\partial\Delta_k)$. Then according to property (3) and [@S Thm.4.5] $${\rm dim}\, Z\le\max_{1\le k\le n}{\rm dim}\,R^{-1}(\partial\Delta_k)={\rm dim}\,\bigl(\mathfrak M(H^\infty)\setminus{\mathbb{D}}\bigr)=2.$$
Thus it remains to show that ${\rm dim}\, R^{-1}(S)\le 2$. We have $$R^{-1}(S)=(R^{-1}(S)\setminus i(S))\sqcup i(S).$$ According to property (2) and Theorem \[te2.1\] for each compact subset $Z\subset R^{-1}(S)\setminus i(S)$ we obtain $${\rm dim}\,Z\le {\rm dim}\,\left(\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}))\setminus \bigl(({\mathbb{D}}\times{ \mathbb{N}})\cup (\{0\}\times (\beta{ \mathbb{N}}\setminus{ \mathbb{N}}))\bigr)\right)\le {\rm dim}\,\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}))=2.$$ Also, according to property (1) and the hypotheses of the theorem $${\rm dim}\,i(S)={\rm dim}\, S\le 2.$$ This and [@N Ch.2,Thm.9-11] imply that ${\rm dim}\, R^{-1}(S)\le 2$. Combining all preceding inequalities and using that ${\rm dim}\, U=2$ we obtain that ${\rm dim}\,\mathfrak M(H^\infty(U))=2$, as required.
Now, let us prove that under the hypotheses of the theorem $H^2(\mathfrak M(H^\infty(U)),{\mathbb{Z} })=0$.
To this end, using the natural one-to-one correspondence between isomorphism classes of rank 1 complex vector bundles on $\mathfrak M(H^\infty(U))$ and elements of $H^2(\mathfrak M(H^\infty(U)),{\mathbb{Z} })$ given by the first Chern classes of the bundles, it suffices to prove that each rank $1$ complex vector bundle $E$ on $\mathfrak M(H^\infty(U))$ is trivial, i.e., admits a nowhere vanishing continuous section. We prove this in several stages.
First, let us show that $E|_{i(S)}$ is trivial.
Indeed, since by the hypotheses ${\rm dim}\,\mathfrak M(H^\infty(V))=2$ and $H^2(\mathfrak M(H^\infty(V),{\mathbb{Z} })=0$, by property (1) employing the long cohomology sequence of the pair $\bigl(\mathfrak M(H^\infty(V)),i(S)\bigr)$ we obtain that $H^2(i(S),{\mathbb{Z} })=0$. Hence, the first Chern class of $E|_{i(S)}$ is 0, i.e., the bundle $E|_{i(S)}$ is trivial. By $s_1$ we denote a nowhere continuous section of $E|_{i(S)}$.
Next, we show that $E|_{R^{-1}(S)}$ is trivial.
Let $G^*(E|_{R^{-1}(S)})$ be the pullback by $G$ of the bundle $E|_{R^{-1}(S)}$ to the compact set $\mathfrak M(H^\infty({\mathbb{D}}\times{\mathbb{Z} }))\setminus ({\mathbb{D}}\times{ \mathbb{N}})$. Since $\mathfrak M(H^\infty({\mathbb{D}}\times{\mathbb{Z} }))=2$ and $H^2(\mathfrak M(H^\infty({\mathbb{D}}\times{\mathbb{Z} })),{\mathbb{Z} })=0$ by Theorem \[te2.1\], as above we obtain that $H^2\bigl(\mathfrak M(H^\infty({\mathbb{D}}\times{\mathbb{Z} }))\setminus ({\mathbb{D}}\times{ \mathbb{N}}),{\mathbb{Z} }\bigr)=0$. Hence, the bundle $G^*(E|_{R^{-1}(S)})$ is trivial, i.e. has a nowhere continuous section, say, $s_2$.
Due to property (2), $G^*s_1$ is a nowhere continuous section of the restriction of the bundle $G^*(E|_{R^{-1}(S)})$ to $G^{-1}(i(S))=\{0\}\times (\beta{ \mathbb{N}}\setminus{ \mathbb{N}})$. Hence, $f:=(G^*s_1)\cdot s_2^{-1}|_{\{0\}\times (\beta{ \mathbb{N}}\setminus{ \mathbb{N}})}$ is a nowhere vanishing continuous function on $\{0\}\times (\beta{ \mathbb{N}}\setminus{ \mathbb{N}})$.
Further, the pullback by means of the projection ${\mathbb{D}}\times{ \mathbb{N}}\rightarrow{ \mathbb{N}}$, $(z,n)\mapsto n$, determines a monomorphism of Banach algebras $\ell^\infty\hookrightarrow H^\infty({\mathbb{D}}\times{ \mathbb{N}})$ whose transpose induces a continuous surjection $P:\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}))\rightarrow\beta({ \mathbb{N}})$ such that $P^{-1}({ \mathbb{N}})={\mathbb{D}}\times{ \mathbb{N}}$. In particular, $P$ maps the compact set $\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}))\setminus ({\mathbb{D}}\times{ \mathbb{N}})$ onto $\{0\}\times (\beta{ \mathbb{N}}\setminus{ \mathbb{N}})$.
We set $$g:=P^*f\in C\bigl(M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}))\setminus ({\mathbb{D}}\times{ \mathbb{N}}),{{\mathbb C}}^*\bigr).$$ Then $g\cdot s_2$ is nowhere vanishing continuous section of $G^*(E|_{R^{-1}(S)})$ whose restriction to $G^{-1}(i(S))$ coincides with $G^*s_1$. In turn, according to property (2), there is a continuous nowhere vanishing section $s_3$ of $E|_{R^{-1}(S)}$ such that $s_3|_{i(S)}=s_1$ and $G^*s_3=g\cdot s_2$. This shows that the bundle $E|_{R^{-1}(S)}$ is trivial.
Now, let us show that $E|_{K_1}$ is trivial, see .
Indeed, since $E|_{R^{-1}(S)}$ is trivial, there is a relatively open neighbourhood $O\subset K_1$ of $R^{-1}(S)$ such that $E|_O$ is trivial (this follows from the standard extension property of sections of vector bundles). Then the compact set $K_1\setminus O$ is covered by relatively clopen pairwise disjoint sets $R^{-1}(\partial\Delta_n)$, $n\in{ \mathbb{N}}$. In particular, there is some $n_0\in{ \mathbb{N}}$ such that $$\label{embed}
K_1\setminus O\subset \bigsqcup_{1\le k\le n_0}R^{-1}(\partial\Delta_k).$$ According to property (3) and [@S Thm.4.5] $${\rm dim}\,\left(\bigsqcup_{1\le k\le n_0}R^{-1}(\partial\Delta_k)\right)=2\quad {\rm and}\quad H^2\left(\bigsqcup_{1\le k\le n_0}R^{-1}(\partial\Delta_k),{\mathbb{Z} }\right)=0.$$ Hence, the restriction of $E$ to $\sqcup_{1\le k\le n_0}R^{-1}(\partial\Delta_k)$ is trivial.
Then $\bigl(\{R^{-1}(\partial\Delta_k)\}_{1\le k\le n_0}, K_1\setminus\bigl(\sqcup_{1\le k\le n_0}R^{-1}(\partial\Delta_k)\bigr)\bigr)$ is an open cover of $K_1$ by pairwise disjoint relatively open sets and the restriction of $E$ to each of the sets is trivial, see . Thus, the bundle $E|_{K_1}$ is trivial.
Finally, we prove that the bundle $E$ is trivial.
Due to property (1) and the hypotheses of the theorem the restriction of $E$ to each compact subset of $i\bigl((\mathfrak M(H^\infty(V))\setminus V)\cup U\bigr)$ is trivial. On the other hand, due to the previous statement there is an open neighbourhood $O_1$ of $K_1$ such that $E|_{O_1}$ is trivial. Let $O_2$ be another open neighbourhood of $K_1$ such that $\overline{O}_2\subset O_1$. Then $\mathfrak M(H^\infty(U))\setminus O_2$ is a compact subset of $i\bigl((\mathfrak M(H^\infty(V))\setminus V)\cup U\bigr)$ and $\bigl(\mathfrak M(H^\infty(U))\setminus \overline{O}_2, O_1\bigr)$ is an open cover of $\mathfrak M(H^\infty(U))$. By the definition, $$Y_1:=i^{-1}\bigl(\mathfrak M(H^\infty(U))\setminus \overline{O}_2\bigr)
\quad {\rm
and}\quad Y_2:=i^{-1}\bigl(O_1\setminus K_1\bigr)$$ are open subsets of $(\mathfrak M(H^\infty(V))\setminus V)\cup U$ and $\bigl((\mathfrak M(H^\infty(V))\setminus V)\cup U\bigr)\setminus S$, see , such that $Y_1\cup Y_2$ covers $(\mathfrak M(H^\infty(V))\setminus V)\cup U$. Moreover, $$\label{exten}
Y_3:=Y_2\cup S\cup\left(\bigsqcup_i\,\overline{\Delta}_i\right)$$ is an open neighbourhood of the compact set $S\cup\bigl(\sqcup_{i}\,\overline{\Delta}_i\bigr)\subset \mathfrak M(H^\infty(V))$ such that $$R^{-1}(Y_3)=O_1.$$
Let $t_1$ and $t_2$ be continuous nowhere vanishing sections of the restrictions of $E$ to $O_1$ and $\mathfrak M(H^\infty(U))\setminus \overline{O}_2$, respectively (existing by the previous arguments). Then $$t_{12}:=t_1^{-1}\cdot t_2\in C\bigl((\mathfrak M(H^\infty(U))\setminus \overline{O}_2)\cap O_1,{{\mathbb C}}^*\bigr).$$ By our construction, $(\mathfrak M(H^\infty(U))\setminus \overline{O}_2)\cap O_1\subset i\bigl((\mathfrak M(H^\infty(V))\setminus V)\cup U\bigr)$ and $$i^{-1}\bigl((\mathfrak M(H^\infty(U))\setminus \overline{O}_2)\cap O_1\bigr) =Y_1\cap Y_3.$$ Hence, the pullback $(i^{-1})^* t_{12}$ by $i^{-1}$ is a $1$-cocycle on the cover $(Y_1,Y_3)$ of $\mathfrak M(H^\infty(V))$ with values in ${{\mathbb C}}^*$. Each such cocycle determines a complex rank 1 vector bundle on $\mathfrak M(H^\infty(V))$ which according to the hypotheses of the theorem is trivial. Hence, there exist $\tilde t_1\in C(Y_3,{{\mathbb C}}^*)$ and $\tilde t_2\in C(Y_1,{{\mathbb C}}^*)$ such that $$\tilde t_2^{-1}\cdot\tilde t_1=(i^{-1})^* t_{12}\quad {\rm on}\quad Y_1\cap Y_3.$$ The latter implies that $$(R^*\tilde t_2)^{-1}\cdot (R^*\tilde t_1)=R^*((i^{-1})^*t_{12})=t_1^{-1}\cdot t_2\quad {\rm on}\quad (\mathfrak M(H^\infty(U))\setminus \overline{O}_2)\cap O_1,$$ because $R^*(i^{-1})^*:=(i^{-1}\circ R)^*={\rm id}$ on $\bigl((\mathfrak M(H^\infty(V))\setminus V)\cup U\bigr)\setminus S$ by property (1). Here $R^*\tilde t_1\in C(O_1,{{\mathbb C}}^*)$ and $R^* \tilde t_2\in C\bigl(\mathfrak M(H^\infty(U))\setminus \overline{O}_2 ,{{\mathbb C}}^*\bigr)$.
Thus continuous nowhere vanishing sections $t_1\cdot R^*\tilde t_1$ of $E|_{O_1}$ and $t_2\cdot R^*\tilde t_2$ of $E|_{\mathfrak M(H^\infty(U))\setminus \overline{O}_2}$ coincide on $(\mathfrak M(H^\infty(U))\setminus \overline{O}_2)\cap O_1$ and so determine a continuous nowhere vanishing section of $E$, i.e., the bundle $E$ is trivial.
The proof of the theorem is complete.
Proof of Corollary \[cor1.2\]
-----------------------------
The result follows from Theorem \[te1.1\] by [@BS Cor.1.4].
Proof of Theorem \[appen\]
==========================
Let $r:S'\rightarrow S$ be an unbranched covering of a bordered Riemann surface $S$. We have to prove that
- $S'$ is dense in $\mathfrak M(H^\infty(S'))$;
- ${\rm dim}\,\mathfrak M(H^\infty(S'))=2$;
- $H^2(\mathfrak M(H^\infty(S')),{\mathbb{Z} })=0$.
In fact, part (a) was proved in [@Br1 Cor.1.6] and part (c) follows from the projective freeness of $H^\infty(S')$ established in [@BS Thm.1.5] (see also [@Br5 Sect.3]). Thus it remains to prove part (b) only. The proof is based on some results and constructions of the theory developed in [@Br1], [@Br2]. We refer to these papers for additional details.
Auxiliary Results
-----------------
For the facts presented in this section see, e.g., [@Br2 Sec.2.2], [@Br3] and references therein.
It is well known that $S$ can be regarded as a domain in a compact Riemann surface $R$ such that $$\label{eq4.1}
R\setminus \bar S=\bigsqcup_{i=1}^k D_i,$$ where $D_i$ are open disks with analytic boundaries.
Let $S_o\subset R$ be another bordered Riemann surface containing $\bar S$ as a deformation retract. By the covering homotopy theorem there is an unbranched covering $r: S_{o}'\rightarrow S_o$ such that $S'\subset S_{o}'$ is a domain and $r|_{S'}: S'\rightarrow S$ is the given unbranched covering of $S$.
The covering $r: S_{o}'\rightarrow S_o$ can be viewed as a fiber bundle over $S_o$ with a discrete fiber $F$. Let $E(S_o,\beta F)$ be the space obtained from $S_o'$ by taking the Stone-Čech compactifications of fibres under $r$. Then $E(S_o,\beta F)$ is a normal Hausdorff space of covering dimension $2$ and $r$ extends to a continuous map $r_E: E(S_o,\beta F)\rightarrow S_o$ such that $\bigl(E(S_o,\beta F),S_o,r_E,\beta F\bigr)$ is a fibre bundle over $S_o$ with fibre $\beta F$ and $S_o'$ embeds in $E(S_o,\beta F)$ as an open dense subbundle.
If $K\subset S_o$ is a compact set and $K':=r^{-1}(K)$, $K_E:=r_E^{-1}(K)$, then $K'$ is dense in $K_E$ and a bounded continuous function on $K'$ admits a continuous extension to the compact set $K_E$ if and only if it is uniformly continuous with respect to the path metric induced by a Riemannian metric pulled back by $r$ from $S_o$. In particular, this is valid for restrictions to $K'$ of bounded holomorphic functions defined on the preimage by $r$ of a neighbourhood of $K$. This implies that each function in $H^\infty(S')$ extends continuously to $S_E$. Moreover, the algebra of such extensions separates the points of $S_E$ so that there is an injective continuous map of $S_E$ into $\mathfrak M(H^\infty(S'))$. In what follows, we identify $S_E$ with its image under the embedding. Then $S_E$ is a dense subset of $\mathfrak M(H^\infty(S'))$. Similarly, we regard $E(S_o,\beta F)$ as a dense subset of $\mathfrak M(H^\infty(S_o'))$.
Proof of Theorem \[appen\]
--------------------------
We retain notations of the previous section.
Let $\hat r:\mathfrak M(H^\infty(S'))\rightarrow\bar S$ be the continuous surjective map induced by the transpose of the homomorphism $H^\infty(S_o)\rightarrow H^\infty(S')$, $f\mapsto f\circ r|_{S'}$. By definition, $\hat r|_{S'}=r$.
\[lem4.1\] The set $\mathfrak M(H^\infty(S'))\setminus \hat r^{-1}(\partial S)$ coincides with $S_E\, (:=r^{-1}(S))$.
Let $x\in \mathfrak M(H^\infty(S'))\setminus \hat r^{-1}(\partial S)$ and $\{x_\alpha\}\subset S'$ be a net converging to $x$. Since $\hat r(x)\in S$ and $\hat r(x)=\lim_\alpha r(x_\alpha)$, passing to a subnet, if necessary, without loss of generality we may assume that all points $r(x_\alpha)$ belong to a neighbourhood $U$ of $x$ such that $\bar U\subset S$. Then by the definition of $E(S_o,\beta F)$, $r_E^{-1}(\bar U)$ is a compact subset of $S_E$ in the original topology of $E(S_o,\beta F)$ containing all points $x_\alpha$. Since the image of a compact set under a continuous map is compact, $r_E^{-1}(\bar U)$ as a subset of $\mathfrak M(H^\infty(S'))$ is compact in the Gelfand topology. This implies that $x=\lim_\alpha x_\alpha\in r_E^{-1}(\bar U)\subset
S_E$, as required.
The lemma implies that $S_E$ is an open dense subspace of $\mathfrak M(H^\infty(S'))$ and $$\mathfrak M(H^\infty(S'))=S_E\sqcup \hat r^{-1}(\partial S).$$ Then, since ${\rm dim}\, S_E=2$ and for each compact subset $K\subset S_E$ the restriction of the Gelfand topology of $\mathfrak M(H^\infty(S'))$ to $K$ coincides with the topology induced from $E(S_o,\beta F)$, ${\rm dim}\, K\le 2$, where dimension is defined by open covers in the Gelfand topology. Thus, to show that ${\rm dim}\,\mathfrak M(H^\infty(S'))=2$ it suffices to show that ${\rm dim}\, \hat r^{-1}(\partial S)\le 2$, see [@N Ch.2,Thm.9-11]. To this end, we prove that $\hat r^{-1}(\partial S)$ can be covered by finitely many compact subsets homeomorphic to subsets of $\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}))$. This and Theorem \[te2.1\](a) will imply the required statement.
Let $p:\mathfrak M(H^\infty(S'))\rightarrow\mathfrak M(H^\infty(S_{o}'))$ be the continuous map induced by the transpose of the homomorphism $H^\infty(S_o')\rightarrow H^\infty(S')$, $f\mapsto f|_{S'}$. Then $p|_{S_E}={\rm id}$, the image of $p$ is $\bar S_E:=r_E^{-1}(\bar S)\, (\subset E(S_o,\beta F))$ and $$\label{e4.2}
\hat r=r_E\circ p.$$ Since $\bar S_E$ is a compact subset of $E(S_o,\beta F)$ in the original topology of $E(S_o,\beta F)$, it is a compact subset of $\mathfrak M(H^\infty(S_{o}'))$ as well. Therefore
(\*) [*if $f$ is a bounded continuous function on $S'$ which extends to a function $\tilde f$ on $\bar S_E$ continuous in the bundle topology of $E(S_o,\beta F)$, then $p^*\tilde f$ is an extension of $f\, (=p^*f)$ to $\mathfrak M(H^\infty(S'))$ continuous in the Gelfand topology.*]{}
This fact will be used in the proof.
Let us proceed with the proof of the statement ${\rm dim}\, \hat r^{-1}(\partial S)=2$. To this end, we choose some open disks $\widetilde D_i$ containing $D_i$ such that each $A_i:=\widetilde D_i\setminus D_i\subset S$ is biholomorphic up to the boundary to an annulus $A:=\{z\in{{\mathbb C}}\, :\, c<|z|<1\}$ with $\partial D_i$ homeomorphic to the outer boundary circle, see . We set $$A_i':=r^{-1}(A_i),\quad 1\le i\le k.$$ By definition, each connected component of $A_i'$ is biholomorphic either to an annulus or ${\mathbb{D}}$. (If the covering is not regular, then these components are not necessarily biholomorphic.) We cover $A_i$ by two open sets $A_{i1}$, $A_{i2}$ biholomorphic (by the above chosen biholomorphism of $A_i$ and $A$) to $$A_{1}=\left\{z:=re^{i\theta}\in A\, :\, -\pi<\theta<\frac{\pi}{2}\right\}\quad {\rm and}\quad A_2=\left\{z:=re^{i\theta}\in A\, :\, 0<\theta<\frac{3\pi}{2}\right\},$$ respectively, and set $$A_{ij}':=r^{-1}(A_{ij}).$$ By definition, each connected component of $A_{ij}'$ is biholomorphic by means of $r$ to $A_{ij}$; thus, $A_{ij}'$ is biholomorphic to $A_{ij}\times F$. By $\partial_o A_{ij}:=\overline{A}_{ij}\cap\partial D_i$ we denote the part of the ‘outer’ boundary of $A_{ij}$.
\[prop4.2\] Every holomorphic function from $H^\infty(A_{ij}')$ can be continuously extended to $\hat r^{-1}(A_{ij}\cup \partial_o A_{ij})$.
Note that $A_{ij}'$ is dense in $\hat r^{-1}(A_{ij}\cup \partial_o A_{ij})$ by the corona theorem for $H^\infty(S')$.
First, we prove the following result.
\[lem4.3\] Each $f\in H^\infty(A_{i}')$ admits an extension $\tilde f\in C(\hat r^{-1}(A_{i}\cup \partial D_i))$.
To avoid technicalities we use the convenient language of [@Br3 Sect.3.3].
Let $X:=S_o\times S_{o}'$. We embed $S_{o}'$ into $X$ by the formula $$\label{eq4.2}
e(z):=(r(z), z),\quad z\in S_{o}'.$$ Then $e(S_{o}')$ is a closed submanifold of the two-dimensional Stein manifold $X$. As follows from [@Br3 Thm.1.3] for each $f\in H^\infty(A_{i}')$ there is some $F\in H^\infty(A_{i}\times S_{o}')$ such that $$e^*F=f.$$ Such $F$ can be regarded as a $H^\infty$ function on $A_{i}$ with values in the Banach space $H^\infty(S_{o}')$.
Let $\rho$ be a $C^\infty$ function on $\bar S$ equal to $1$ in a neighbourhood of $\partial D_i$ with support in $A_i$. Then $\overline\partial(\rho F)$ is a $H^\infty(S_o')$-valued $C^\infty$ $(0,1)$-form on $\bar S$ equals 0 in a neighbourhood of $\partial D_i$. Hence, it is extended by $0$ to a $(0,1)$-form $\omega$ on $S_o$. According to the generalized H.Cartan theory, see [@Bu], equation $$\overline\partial G=\omega$$ is solvable on $S_o$. Its solution $G$ is a $C^\infty$ function on $S_o$ with values in $H^\infty(S_o')$ holomorphic (and bounded) in a neighbourhood of $\partial D_i$ such that $$G_1:=\rho F- G|_S\in H^\infty(S,H^\infty(S_o'))=H^\infty(S\times S_o').$$ Thus, $F=G_1+G$ in a neighbourhood $N$ of $\partial D_i$ in $S$. Considering $G_1$ and $G$ as functions on $S_o\times S_o'$ and restricting them to $e(A_i')$ we obtain $$\label{e4.4}
f=g_1+g\quad {\rm on}\quad N':=r^{-1}(N);$$ here $g_1:=e^*G_1\in H^\infty(S')$ and $g:=e^*G$ is such that $g|_{N'}\in H^\infty(N')$.
By definition, $g_1$ admits a continuous extension to $\mathfrak M(H^\infty(S'))$ by the Gelfand transform. Also, $g$ being bounded uniformly continuous on $r^{-1}(\bar S)$ with respect to the path metric induced by a Riemannian metric pulled back by $r$ from $S_o$, see [@Br3], admits a continuous extension to $\bar S_E\subset\mathfrak M(H^\infty(S_o'))$ and, hence, to $\mathfrak M(H^\infty(S'))$, see (\*). This and imply that $f|_{N'}$ can be extended to a function $\tilde f_1\in C(\hat r^{-1}(N\cup\partial D_i))$. On the other hand, $f$ admits a continuous extension to $S_E\subset \mathfrak M(H^\infty(S'))$ denoted by $\tilde f_2$, see Section 4.1.
We have $$\tilde f_1|_{N'}-\tilde f_2|_{N'}=0.$$ Since the open set $N'$ is dense in $\hat r^{-1}(N)$ by the corona theorem, the previous equation implies that $\tilde f_1=\tilde f_2$ on $\hat r^{-1}(N)$, i.e., these functions coincide on $\hat r^{-1}(N\cup\partial D_i)\cap S_E$ and so define a function $\tilde f\in C(\hat r^{-1}(S\cup\partial D_i))$ which extends $f$.
The proof of the lemma is complete.
Now, let us complete the proof of the proposition.
As before, for $f\in H^\infty(A_{ij}')$ there is some $F\in H^\infty(A_{ij}\times S_{o}')$ regarded as a $H^\infty$ function on $A_{ij}$ with values in the Banach space $H^\infty(S_{o}')$ such that $e^*F=f$, see [@Br3 Thm.1.3]. Let $\rho$ be a $C^\infty$ function on $\bar A_i$ with support in $\overline{A}_{ij}$ equal to 1 on a subset $Z_\rho\subset \overline{A}_{ij}$ which is the intersection of a closed sector with the origin at $0$ with $\overline{A}_{ij}$ (here we identify $\bar A_i$ with $\bar A$ by the above chosen biholomorphism). In particular, $\rho=1$ on the closed arc $L_\rho:=Z_\rho\cap\partial_o A_{ij}$. Then $G:=\frac{\partial(\rho F)}{\partial\bar z}$ is a bounded $H^\infty(S_o')$-valued $C^\infty$ function on $A_{i}\cup Z_\rho$ with support in $\overline{A}_{ij}$ equals 0 on $Z_\rho$.
Next, under the identification of $A_i$ with $A$ $$\label{eq4.4}
H(\xi)=\frac{1}{\pi}\iint_{A_{ij}}\,\frac{G(z)}{\xi-z}\,dx\,dy,\qquad \xi\in A_{i}\cup Z_\rho,\quad x={\rm Re}\, z,\quad y={\rm Im}\, z,$$ is a bounded $H^\infty(S_o')$-valued $C^\infty$ function on $A_{i}\cup Z_\rho$ holomorphic on $Z_\rho$ satisfying $$\frac{\partial H}{\partial\bar \xi}=G.$$ Hence, $$H_1:=\rho F- H|_{A_i}\in H^\infty(A_i,H^\infty(S_o'))=H^\infty(A_i\times S_o').$$ In turn, $F=H_1+H$ on $A_{ij}\cap Z_\rho$. Considering $H_1$ and $H$ as functions on $A_i\times S_o'$ and $(A_i\cup Z_\rho)\times S_o'$ and restricting them to $e(A_{ij}')$ we obtain $$f=h_1+h\quad {\rm on}\quad A_{ij}'\cap Z_\rho',\quad Z_\rho':=r^{-1}(Z_\rho);$$ here $h_1:=e^*H_1\in H^\infty(A_i')$ and $h:=e^*H\in C(A_i'\cup Z_\rho')$ is such that $h|_{Z_\rho'}$ is bounded uniformly continuous with respect to the path metric induced by a Riemannian metric pulled back by $r$ from $S_o$. According to Lemma \[lem4.3\], $h_1$ admits a continuous extension to $\hat r^{-1}(A_i\cup\partial D_i)$. Also, $h|_{Z_\rho'}$ admits a continuous extension to $r_E^{-1}(Z_\rho)\subset\mathfrak M(H^\infty(S_o'))$ and, hence, to $\hat r^{-1}( Z_\rho)$, see (\*). These imply that $f|_{A_{ij}'\cap Z_\rho'}$ admits a continuous extension to $\hat r^{-1}((A_{ij}\cap Z_\rho)\cup L_\rho)$. Since $$\bigcup_\rho\, Z_\rho=A_{ij},$$ where $\rho$ runs over all possible functions satisfying the above described conditions, the latter implies that $f$ admits an extension $\tilde f\in C(\hat r^{-1}(A_{ij}\cup \partial_o A_{ij}))$, as required.
Let $q_j:\mathfrak M(H^\infty(A_{ij}'))\rightarrow \mathfrak M(H^\infty(S'))$ be the continuous maps transposed to the embeddings $H^\infty(S')\hookrightarrow H^\infty(A_{ij}')$ determined by restrictions to $A_{ij}'$, $j=1,2$. Due to the corona theorem for $H^\infty(S')$ the image of $q_j$ coincides with $\hat r^{-1}(\overline{A}_{ij})$. Since $A_{ij}'$ is biholomorphic to ${\mathbb{D}}\times{ \mathbb{N}}$, the corona theorem is valid for $H^\infty(A_{ij}')$, i.e., $A_{ij}'$ is dense in $\mathfrak M(H^\infty(A_{ij}'))$.
As a consequence of Proposition \[prop4.2\] we obtain the following:
\[cor4.4\] The restriction of $q_j$ to $(\hat r\circ q_j)^{-1}(A_{ij}\cup \partial_o A_{ij})$ is a bijection onto the set $\hat r^{-1}(A_{ij}\cup \partial_o A_{ij})$.
The result implies that $q_j$ maps a compact subset $K\subset (\hat r\circ q_j)^{-1}(A_{ij}\cup \partial_o A_{ij})$ bijectively onto the compact set $q_j(K)\subset \hat r^{-1}(A_{ij}\cup \partial_o A_{ij})$. Hence, $q_j|_K$ is a homeomorphism.
Since the mapping $q_j|_{(\hat r\circ q_j)^{-1}(A_{ij}\cup \partial_o A_{ij})}:(\hat r\circ q_j)^{-1}(A_{ij}\cup \partial_o A_{ij})\rightarrow \hat r^{-1}(A_{ij}\cup \partial_o A_{ij})$ is surjective, it remains to show that it is injective.
Suppose that points $x_1,x_2\in (\hat r\circ q_j)^{-1}(A_{ij}\cup \partial_o A_{ij})$ are such that $q_j(x_1)=q_j(x_2)\in \hat r^{-1}(A_{ij}\cup \partial_o A_{ij})$. If $x_1\ne x_2$, then there is a function $f\in H^\infty(A_{ij}')$ whose Gelfand transform $\hat f\in C(\mathfrak M(H^\infty(A_{ij}')))$ satisfies $\hat f(x_1)\ne \hat f(x_2)$. Let $\{x_{k\alpha}\}_{\alpha\in\Lambda}\subset A_{ij}'$ be nets converging to $x_k$ in $\mathfrak M(H^\infty(A_{ij}'))$, $k=1,2$. Then, since $q_j|_{A_{ij}'}={\rm id}$, these nets converge to $q_j(x_k)$ in $\mathfrak M(H^\infty(S'))$, $k=1,2$. According to Proposition \[prop4.2\], $f$ admits an extension $\tilde f\in C(\hat r^{-1}(A_{ij}\cup \partial_o A_{ij}))$. Hence, due to the continuity of $\hat f$, $\tilde f$ and $q_j$ we obtain $$\begin{array}{r}
\displaystyle
0\ne \hat f(x_2)-\hat f(x_1)=\lim_\alpha f(x_{2\alpha})-\lim_\alpha f(x_{1\alpha})=\lim_\alpha f(q_j(x_{2\alpha}))-\lim_\alpha f(q_j(x_{1\alpha}))\medskip\\
\displaystyle
=\tilde f(q_j(x_2))-\tilde f(q_j(x_1))=0,\qquad
\end{array}$$ a contradiction proving the result.
Let us complete the proof of the theorem. Recall that we require to show only that ${\rm dim}\,\hat r^{-1}(\partial S)\le 2$ (see the explanation after the proof of Lemma \[lem4.1\]).
By definition, $\hat r^{-1}(\partial S)=\sqcup_{i=1}^k\, \hat r^{-1}(\partial D_i)$ and each $\hat r^{-1}(\partial D_i)$ is covered by two relatively open sets $\hat r^{-1}(\partial_o(A_{ij}))$, $j=1,2$. Due to Corollary \[cor4.4\] and Theorem \[te2.1\](a) we have for each compact subset $K\subset \hat r^{-1}(\partial_o(A_{ij}))$ $${\rm dim}\,K={\rm dim}\,q_j^{-1}(K)\le {\rm dim}\,\mathfrak M(H^\infty(A_{ij}'))={\rm dim}\,\mathfrak M(H^\infty({\mathbb{D}}\times{ \mathbb{N}}))=2.$$ From here and [@N Ch.2,Thm.9-11] we obtain that ${\rm dim}\,\hat r^{-1}(\partial D_i)\le 2$ for all $i$. This implies that ${\rm dim}\,\hat r^{-1}(\partial S)\le \max_{1\le i\le k}\bigl\{ {\rm dim}\,\hat r^{-1}(\partial D_i)\bigr\}\le 2$, as required.
Thus, we have proved that ${\rm dim}\,\mathfrak M(H^\infty(S'))\le 2$. In fact, since $\mathfrak M(H^\infty(S'))$ contains the dense open subset $S'$ of dimension two, ${\rm dim}\,\mathfrak M(H^\infty(S'))=2$ as well.
The proof of the theorem is complete.
M. Behrens, The maximal ideal space of algebras of bounded analytic functions on infinitely connected domains, Trans. Amer. Math. Soc. [**161**]{} (1971), 359–380.
A. Brudnyi, Projections in the space $H^\infty$ and the Corona Theorem for coverings of bordered Riemann surfaces, Ark. Mat. [**42**]{} (2004), no. 1, 31–59.
A. Brudnyi, Grauert- and Lax–Halmos-type theorems and extension of matrices with entries in $H^\infty$, J. Funct. Anal. [**206**]{} (2004), 87–108.
A. Brudnyi, Holomorphic functions of slow growth on coverings of pseudoconvex domains in Stein manifolds, Compositio Math. [**142**]{} (2006), 1018–1038.
A. Brudnyi, Extension of matrices with entries in $H^\infty$ on coverings of Riemann surfaces of finite type, St. Petersburg Math. J. [**21**]{} (2010), No. 3, 423–432.
A. Brudnyi, Topology of the maximal ideal space of $H^\infty$ revisited, Adv. Math. [**299**]{} (2016), 931–939.
L. Bungart, On analytic fibre bundles I. Holomorphic fibre bundles with infinite dimensional fibres, Topology [**7**]{} (1) (1968) 55–68.
A. Brudnyi and A. Sasane, Sufficient conditions for the projective freeness of Banach algebras, J. Funct. Anal. [**257**]{} (2009), 4003–4014.
L. Carleson, Interpolations by bounded analytic functions and the corona theorem, Ann. of Math. [**76**]{} (1962), 547–559.
P. M. Cohn, Free Rings and their Relations, London Math. Soc. Monographs [**19**]{}, Academic Press, London, 1985.
R. G. Douglas, S. G. Krantz, E. T. Sawyer, S. Treil, B. D. Wick (eds.), The Corona Problem: Connections Between Operator Theory, Function Theory, and Geometry, Fields Institute Communications [**72**]{}, Springer, New York, 2014.
J. B. Garnett, Bounded analytic functions, Academic Press, New York, 1981.
S.-T. Hu, Mappings of a normal space into an absolute neighborhood retract, Trans. Amer. Math. Soc. [**64**]{} (1948), 336–358.
B. Maskit, Kleinian groups, Grundlehren der Mathematischen Wissenschaften [**287**]{} Springer-Verlag, New York, 1988.
K. Nagami, Dimension Theory, Academic Press, New York, 1970.
N. K. Nikolskii, Treatise on the shift operator, Springer-Verlag, Berlin, 1986.
D. Suárez, Čech cohomology and covering dimension for the $H^\infty$ maximal ideal space, J. Funct. Anal. [**123**]{} (1994) 233–263.
J. L. Taylor, Topological invariants of the maximal ideal space of a Banach algebra, Adv. Math. [**19**]{} (1976), 149–206.
V. Tolokonnikov, Extension problem to an invertible matrix, Proc. Amer. Math. Soc. [**117**]{} (1993), 1023–1030.
V. Tolokonnikov, Stable rank of $H^\infty$ in multiply connected domains, Proc. Amer. Math. Soc. [**123**]{} 10 (1995), 3151–3156.
[^1]: Research is supported in part by NSERC
[^2]: Here and below we identify $U$ and ${\mathbb{D}}\times{\mathbb{Z} }$ with their images under the natural embeddings in the corresponding maximal ideal spaces.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We prove that if $X$ and $S$ are smooth varieties and $f\colon X\to S$ is an elliptic fibration with singular fibers curves of types I$_N$ with $N\geq 1$, II, III and IV, then the relative Jacobian $\hat{f}\colon \overline{M}_{X/S}\to S$ of $f$, defined as the relative moduli space of semistable pure dimension one sheaves of rank 1 and degree 0 on the fibers of $f$, is an elliptic fibration such that all its fibers are irreducible. This extends known results when fibers are integral or of type I$_2$.'
address: 'Departamento de Matemáticas, Universidad de Salamanca, Plaza de la Merced 1-4, 37008 Salamanca, Spain'
author:
- Ana Cristina López Martín
title: Relative Jacobians of elliptic fibrations with reducible fibers
---
Introduction
============
Let $f\colon X\to S$ be an elliptic fibration, that is, a proper flat morphism of schemes whose fibers are Gorenstein curves of arithmetic genus 1. If we have a relative ample sheaf $\mathcal{O}_X(1)$ on $X$, the relative Jacobian of $f$ is defined as the Simpson moduli space $\hat{f}\colon \overline{M}_{X/S}\to
S$ of semistable pure dimension one sheaves of rank 1 and degree 0 on the fibers of $f$ with respect to the polarization induced by $\mathcal{O}_X(1)$. Since every torsion free rank 1 sheaf on an integral curve is stable, when $f\colon X \to S$ is an integral fibration, that is, all its fibers are geometrically integral curves, its relative Jacobian is simply the relative moduli space of torsion free rank 1 sheaves of relative degree 0. In this case, it is known that for any closed point $s\in S$ we have an isomorphism $(\overline{M}_{X/S})_s\simeq X_s$ between the fibers of both fibrations over $s$. Then this relative Jacobian is again an integral elliptic fibration. Furthermore if the fibration $f\colon X\to S$ has a section, $\overline{M}_{X/S}$ is not only locally but globally isomorphic to $X$ over $S$ (see [@HM]).
On the other hand the geometry of such relative Jacobian $\hat{f}\colon \overline{M}_{X/S}\to S$ is not known for a general elliptic fibration. In [@Cal2] we find some examples showing that when $f\colon X\to S$ has reducible fibers (exactly $I_2$ fibers), it is no longer true that the two fibrations $f$ and $\hat{f}$ have isomorphic fibers. In fact, Căldăraru proves that if $X_s$ is a fiber of type $I_2$, then the fiber $(\overline{M}_{X/S})_s$ is isomorphic to a rational curve with one node.
One should point out that elliptic fibrations have been used in string theory, notably in connection with mirror symmetry on Calabi-Yau manifolds and D-branes. Some of the classic examples of families of Calabi-Yau manifolds for which there is a Greene-Plesser [@GP90] description of the mirror family produced by Candelas and others, are elliptic fibrations [@Can91]. Moreover there is a relative Fourier-Mukai transform for most elliptic fibrations ([@BBHM; @HM]) that can be understood in terms of duality in string theory ([@DoTani; @ACHY] or D-brane theory. The latter application is due to the interpretation of B-type D-brane states as objects of the derived category $D(X)$ of coherent sheaves [@Kon95; @As; @Dou] and to Kontsevich’s homological mirror symmetry proposal [@Kon95] that gives an equivalence between $D(X)$ and the Fukaya category [@Fuk93]. According to that proposal, the monodromies around special points of the known-models of local moduli spaces of Lagrangian submanifolds should correspond to Fourier-Mukai transforms [@ACHY; @Ho]; this explains the importance of elliptic Calabi-Yau manifolds in string theory.
The aim of this paper is to study the structure of the relative Jacobian of an elliptic fibration $f\colon X\to S$ such that $X$ and $S$ are smooth projective varieties and the singular curves appearing as fibers of $f$ are:
(I$_1$) : A rational curve with one node.
\(II) : A rational curve with one cusp.
\(III) : $C_1\cup C_2$ where $C_1$ and $C_2$ are rational smooth curves with $C_1\cdot C_2=2p$.
\(IV) : $C_1\cup C_2\cup C_3$, where $C_1, C_2, C_3$ are rational smooth curves and $C_1\cdot C_2=C_2\cdot C_3=C_3\cdot C_1=p$
(I$_N$) : $C_1\cup C_2\cup \hdots \cup C_N$, where $C_i$, $i=1,\hdots, N$, are rational smooth curves and $C_1\cdot
C_2=C_2\cdot C_3=\hdots=C_{N-1}\cdot C_N=C_N\cdot C_1=1$ if $N>2$ and $C_1\cdot C_2=p_1+p_2$ if $N=2$.
If $f\colon X\to S$ is an elliptic fibration of this type, we prove that the moduli space of semistable pure dimension one sheaves of rank 1 and degree 0 on a fiber $X_s$ is isomorphic to a smooth elliptic curve when $X_s$ is smooth, to a rational curve with one node when $X_s$ is I$_N$, $N\geq 1$ and to a rational curve with one cusp when $X_s$ is II, III or IV. The result is then that the relative Jacobian $\hat{f}\colon
\overline{M}_{X/S}\to S$ of $f$ is an integral elliptic fibration (Theorem \[teorema\]). In particular we deduce that, although $\overline{M}_{X/S}$ is irreducible, if the fibration $f$ has reducible fibers, it cannot be isomorphic to $X$, even assuming that $f$ has a section. For instance, if the variety $X$ has dimension 2 or 3, then the results about contractibility of curves on smooth surfaces [@A] and on smooth threefolds [@Z] allow us to to ensure the existence of singular points in the moduli space $\overline{M}_{X/S}$ that correspond to strictly semistable sheaves on the fibers of $f$. Then in these cases $\overline{M}_{X/S}$ is not isomorphic to the original variety $X$.
Since by Kodaira’s work [@K], we known that every elliptic surface $f\colon X\to S$ with reduced fibers is a fibration of this type, the theorem gives us in particular the structure of the relative Jacobian of any reduced elliptic surface.
I wish to express my gratitude to Prof. Hernández Ruipérez for his invaluable help and for his constant encouragement. I also thank A. Căldăraru for many useful suggestions.
Preliminares
============
All the schemes considered in this paper are of finite type over an algebraically closed field $\kappa$ of characteristic zero and all the sheaves are coherent.
Some elliptic fibrations with reducible fibers
----------------------------------------------
Let $f\colon X\to S$ be an elliptic fibration. By this we mean a proper flat morphism of schemes whose fibers are geometrically connected Gorenstein curves of arithmetic genus 1. We denote by $X_s$ the fibre of $f$ over $s\in S$ and by $\Sigma(f)$ the [*discriminant locus*]{} of $f$, that is, the closed subset of points $s\in S$ such that $X_s$ is not a smooth curve.
There are two important cases where the curves that can occur as singular fibers of an elliptic fibration $f\colon X\to S$ are known. When $X$ is an elliptically fibred surface over a smooth curve $S$, the singular fibers of $f$ are classified by Kodaira [@K]. And if $f\colon X\to S$ is an elliptic threefold with $X$ and $S$ smooth projective varieties and the map $f$ has a section, Miranda [@M] studies the kinds of degenerated fibers of $f$ that can appear. These two works allow us to ensure the existence of elliptic fibrations as in the following
\[d:fibracion\] An elliptic fibration of type ($*$) is an elliptic fibration $f\colon X\to S$, with $X$ and $S$ smooth projective varieties, without multiple fibers and such that if $s\in \Sigma(f)$, the fiber $X_s$ is one of the following curves:
(I$_1$) : A rational curve with one node.
\(II) : A rational curve with one cusp.
\(III) : $X_s=C_1\cup C_2$ where $C_1$ and $C_2$ are rational smooth curves with $C_1\cdot C_2=2p$.
\(IV) : $X_s=C_1\cup C_2\cup C_3$, where $C_1, C_2, C_3$ are rational smooth curves and $C_1\cdot C_2=C_2\cdot C_3=C_3\cdot
C_1=p$
(I$_N$) : $X_s=C_1\cup C_2\cup \hdots \cup C_N$, where $C_i$, $i=1,\hdots, N$, are rational smooth curves and $C_1\cdot
C_2=C_2\cdot C_3=\hdots=C_{N-1}\cdot C_N=C_N\cdot C_1=1$ if $N>2$ and $C_1\cdot C_2=p_1+p_2$ if $N=2$.
If $S$ is a smooth curve and $f\colon X\to S$ is an elliptic surface such that all fibers are reduced, by Kodaira’s classification, every singular fiber $X_s$ of $f$ is one of the above list. Then any smooth elliptic surface over $S$ with reduced fibers is a elliptic fibration of type ($*$).
If $f_0\colon X_0\to S_0$ is an elliptic fibration with $X_0$ and $S_0$ varieties of dimensions 3 and 2 respectively and $f_0$ has a section, Miranda constructs in [@M] a flat model $f\colon X\to
S$ with $X$ and $S$ smooth and such that the discriminant locus $\Sigma(f)$ is a curve with at worst ordinary double points as singularities. He proves that at a smooth point $s\in \Sigma(f)$ the singular fiber $X_s$ is one on Kodaira’s list and that the type of fiber is constant on the irreducible components of $\Sigma(f)^{\text{smooth}}$. Moreover he determines, case by case, the type of singular fiber over the collision points of $\Sigma(f)$. The results of Miranda imply that if the fibers of $f$ over the smooth points of $\Sigma(f)$ are reduced and all collisions are of type $I_{N_1}+I_{N_2}$, then $f\colon X\to S$ is an elliptic fibration of type ($*$).
The Jacobian of a projective curve
----------------------------------
Let $C$ be a projective curve. Let ${\mathcal L}$ be an ample invertible sheaf on $C$, let $H$ be the associated polarization and let $h$ denote the degree of $H$.
A sheaf $F$ on $C$ is [*pure dimension one*]{} if the support of any nonzero subsheaf of $F$ is of dimension one. The (polarized) [*rank*]{} and [*degree*]{} with respect to $H$ of $F$ are the rational numbers ${\operatorname{r}}_H(F)$ and ${\operatorname{d}}_H(F)$ determined by the Hilbert polynomial $$P(F,n,H)=\chi(F\otimes\mathcal{O}_C(nH))=h{\operatorname{r}}_H(F)n+{\operatorname{d}}_H(F)+{\operatorname{r}}_H(F)\chi(\mathcal{O}_C).$$ The [*slope*]{} of $F$ is defined by $$\mu_H(F)=\frac{{\operatorname{d}}_H(F)}{{\operatorname{r}}_H(F)}$$ The sheaf $F$ is [*stable*]{} (resp. [*semistable*]{}) with respect to $H$ if $F$ is pure of dimension one and for any proper subsheaf $F'\hookrightarrow F$ one has $$\mu_H(F')<\mu_H(F) \ (\text{resp.} \leq)$$ For every semistable sheaf $F$ with respect to $H$ there is a [*Jordan-Hölder filtration*]{} $$0=F_0\subset F_1\subset \hdots \subset F_n=F$$ with stable quotients $F_i/F_{i-1}$ and $\mu_H(F_i/F_{i-1})=\mu_H(F)$ for $i=1,\hdots,n$. This filtration need not be unique, but [*the graded object*]{} $Gr(F)=\textstyle{\bigoplus_{i}} F_i/F_{i-1}$ does not depend on the choice of the Jordan-Hölder filtration. Two semistable sheaves $F$ and $F'$ on $C$ are said to be [*$S$-equivalent*]{} if $Gr(F)\simeq Gr(F')$. Observe that two stable sheaves are $S$-equivalent only if they are isomorphic. If $F$ is a semistable sheaf on $C$, we will denote by $[F]$ its $S$-equivalence class.
By Simpson’s work [@Si], there exists a projective moduli space of semistable pure dimension one sheaves on $C$ of (polarized) rank 1 and degree 0.
If the curve is not integral, this moduli space can contain some components given by sheaves of (polarized) rank 1 whose restrictions to some irreducible components of $C$ are concentrated sheaves. These components correspond to moduli spaces of higher rank sheaves on reducible curves (see [@L] for some examples). Therefore, from here on by rank 1 sheaves we mean those sheaves having rank 1 on every irreducible component of $C$ and we define [*the Jacobian of* ]{}$C$ as the moduli space $\overline{M}(C)$ of pure dimension one sheaves of rank 1 and degree 0 on $C$ that are semistable with respect to the fixed polarization. Observe that if $C$ is a reduced curve, $\overline{M}(C)$ is also a projective scheme because it coincides with Seshadri’s compactification [@Se].
For certain projective curves, an explicit description of the structure of this Jacobian $\overline{M}(C)$ can be found in [@L] where the author studies not only the case of degree 0 sheaves, but also these moduli spaces for arbitrary degree $d$ sheaves.
The relative Jacobian
---------------------
Let $f\colon X\to S$ be an elliptic fibration and let ${\mathcal O}_X(1)$ be a relative ample sheaf on $X$. Let $\overline{{\mathcal M}}_{X/S}$ be the functor which to any $S$-scheme $T$ associates the space of $S$-equivalence classes of $T$-flat sheaves on $f_T\colon
X\times_S T \to T$ whose restrictions to the fibers of $f_T$ are semistable of rank 1 and degree 0 with respect to the induced polarization. Two such sheaves $F$ and $F'$ are said to be equivalent if $F'\simeq F\otimes f_T^*N$, where $N$ is a line bundle on $T$.
Again by [@Si], there is a projective scheme $\overline{M}_{X/S}\to S$ which universally corepresents the functor $\overline{{\mathcal M}}_{X/S}$. Moreover there is an open subscheme $\overline{M}^s_{X/S}\subseteq \overline{M}_{X/S}$ that universally corepresents the subfunctor $\overline{{\mathcal
M}}^s_{X/S}\subseteq \overline{{\mathcal M}}_{X/S}$ of families of stable sheaves.
The fibration $\hat{f}\colon \overline{M}_{X/S}\to S$ is defined to be [*the relative Jacobian*]{} of $f\colon X\to S$. Points of $\overline{M}_{X/S}$ represent semistable sheaves of rank 1 and degree 0 on the fibers of $f$ and the natural map $\hat{f}\colon
\overline{M}_{X/S}\to S$ sends a sheaf supported on the fiber $X_s$ to the corresponding point $s\in S$. In particular, for any closed point $s\in S$ one has that the fiber $\hat{f}^{-1}(s)$ is isomorphic to the (absolute) moduli space $\overline{M}(X_s)$ of semistable rank 1 degree 0 sheaves on $X_s$, that is, the Jacobian of the curve $X_s$.
In order to study the relative Jacobian $\hat{f}\colon
\overline{M}_{X/S}\to S$ we have then to know the structure of the corresponding Jacobians of the curves that can appear as fibers of the elliptic fibration $f\colon X\to S$. We do this in the following section when $f\colon X\to S$ is an elliptic fibration of type ($*$) (Definition \[d:fibracion\]).
The Jacobians of the fibers
===========================
Let $f\colon X\to S$ be an elliptic fibration of type ($*$) and let ${\mathcal O}_X(1)$ be a line bundle on $X$ ample relative to $S$.
Let $C$ denote a fiber of $f$ and let $H$ be the induced polarization on $C$. If $C$ is an integral curve (a smooth elliptic curve, a rational curve with one node or a rational curve with one cusp), it is well known that the moduli space $\overline{M}(C)$ is isomorphic to $C$. However if $C$ is a fiber of type $I_2$, that is, two projective lines meeting transversely at two points, Căldăraru [@Cal2] has proved that the moduli space $\overline{M}(C)$ is isomorphic to a rational curve with one node. Following his argument and as a consequence of the descriptions given in [@L], in this section we prove that for every reducible fiber $C$ of $f\colon X\to S$, the moduli space $\overline{M}(C)$ is isomorphic either to a rational curve with one node or to a rational curve with one cusp.
Let $C$ be any reducible fiber of $f\colon X\to S$, that is, a curve of type III, IV or I$_N$ with $N\geq 2$. Let us denote by $C_i$ the irreducible components of $C$. In the following lemma we collect some properties of rank 1 degree 0 sheaves on $C$ that we will use later (see [@L] for the proof).
\[l:propiedades\] If $C$ is a curve of type III, IV or I$_N$ with $N\geq 2$, it holds that:
1. The (semi)stability of a pure dimension one sheaf of rank 1 and degree 0 on $C$ does not depend on the polarization.
2. A degree 0 line bundle $L$ on $C$ is stable if and only if $L|_{C_i}\simeq \mathcal{O}_{\mathbb{P}^1}$ for all $i$.
3. If $F$ is a stable pure dimension one sheaf of rank 1 and degree 0 on $C$, then $F$ is a line bundle.
4. If $L$ is a line bundle on $C$ of degree 0, then $L$ is strictly semistable if and only if $L|_{C_i}\simeq
\mathcal{O}_{\mathbb{P}^1}(r)$ where $r=-1,0$ or $1$ in such a way that when we remove the components $C_i$ for which $r=0$ there are neither two consecutive $r=1$ nor two consecutive $r=-1$.
5. If $F$ is a strictly semistable pure dimension one sheaf of rank 1 and degree 0 on $C$, then its graded object is $Gr(F)=\oplus_{i}\mathcal{O}_{\mathbb{P}^1}(-1)$
Let $q$ be a fixed smooth point of $C$ and let us denote by $C_0$ the irreducible component of $C$ on which $q$ lies.
If $\Delta\subseteq C\times C$ denotes the diagonal and ${\mathcal
J}_\Delta$ is its ideal sheaf, define ${\mathcal O}_{C\times
C}(\Delta)= {\operatorname{\mathcal{H}om}}({\mathcal J}_{\Delta},{\mathcal O}_{C\times C})$ as the dual of $\mathcal{J}_\Delta$.
Consider the sheaf $$\mathcal{E}=\mathcal{O}_{C\times
C}(\Delta)\otimes \pi_1^*\mathcal{O}_C(-q)$$ where $\pi_1\colon
C\times C\to C$ is the projection on the first component. This sheaf is flat over $C$ via the projection $\pi_2\colon C\times
C\to C$ (see [@Cal1], for details) and we have the following
\[p:semiest\] For any point $p\in C$, the restriction $\mathcal{E}_p$ of $\mathcal{E}$ to $C\times \{p\}$ is a semistable pure dimension one sheaf of rank 1 and degree 0. Moreover, if $p$ is not a smooth point of $C_0$, then all sheaves $\mathcal{E}_p$ define the same point of the moduli space $\overline{M}(C)$.
Since $C$ is Gorenstein, we have that ${\operatorname{Ext}}^1(\mathcal{O}_P,\mathcal{O}_C(-q))=\kappa$, so that the restriction $\mathcal{E}_p$ is the unique non trivial extension $$0\to \mathcal{O}_C(-q)\to \mathcal{E}_p\to \mathcal{O}_p\to 0\,
.$$ Using this exact sequence one easily proves that $\mathcal{E}_p$, which is precisely $\mathcal{J}_p^*\otimes
\mathcal{O}_C(-q)$, is a pure dimension one sheaf of rank 1 and degree 0.
To prove that it is semistable, let us consider two cases: when $p$ is a smooth point and when it is a singular point of $C$. In the first case, $\mathcal{E}_p$ is the line bundle $\mathcal{O}_C(p-q)$ and we have the following:
1. If $p\in C_0$, the restrictions of $\mathcal{E}_p$ to all irreducible components of $C$ are of degree 0. Then, by (2) of the previous Lemma, the sheaf $\mathcal{E}_p$ is stable.
2. If $p\notin C_0$, let $C_1$ be the irreducible component of $C$ on which $p$ lies. Since the restriction of $\mathcal{E}_p$ to $C_0$ has degree -1, to $C_1$ degree 1 and to the others components degree 0, (4) in Lemma \[l:propiedades\] implies that the sheaf $\mathcal{E}_p$ is strictly semistable.
In the second case, since $p$ is a singular point of $C$, $\mathcal{E}_p$ is not an invertible sheaf and then, by (3) in Lemma \[l:propiedades\], it is not stable. Let us see that it is semistable. Let $\mathcal{G}\hookrightarrow \mathcal{E}_p$ be a proper subsheaf. We have the exact sequence $$0\to \mathcal{O}_C(-q)\to \mathcal{E}_p\to \mathcal{O}_p\to 0\,
.$$ Bearing in mind that every proper connected subcurve $D$ of $C$ has arithmetic genus 0 and $D\cdot \overline{D}=2$, it is easy to deduce from Lemma 3.4 in [@L] that the line bundle $\mathcal{O}_C(-q)$ is stable. Consider the composition map $g\colon \mathcal{G}\to \mathcal{O}_p$ that can be either zero or surjective. If $g$ is zero, $\mathcal{G}\hookrightarrow
\mathcal{O}_C(-q)$ and then $\mu_H(\mathcal{G})<-1$. If $g$ is surjective and we denote by $\mathcal{H}$ its kernel, we have that $\mathcal{H}$ is a subsheaf of $\mathcal{O}_C(-q)$ and then $\mu_H(\mathcal{H})<-1$. Since the degree of $\mathcal{H}$ is integer, from the exact sequence $$0\to \mathcal{H}\to \mathcal{G}\to \mathcal{O}_p\to 0\, ,$$ we conclude that $\mu_H(\mathcal{G})\leq 0$. Then $\mathcal{E}_p$ is a strictly semistable sheaf.
For the second part of the statement, it is enough to note that if $p$ is not a smooth point of $C_0$, $\mathcal{E}_p$ is a strictly semistable sheaf and then, by (5) in \[l:propiedades\], its graded object is isomorphic to $\oplus_i\mathcal{O}_{\mathbb{P}^1}(-1)$. Hence all these sheaves are in the same $S$-equivalence class and the proof is complete.
The restriction of the family $\mathcal{E}$ to $C\times C_0$ gives, by the universal property of $\overline{M}(C)$, a map $$\phi\colon C_0\to \overline{M}(C)$$ defined as $\phi(p)=[\mathcal{E}_p]$.
The same proof that Căldăraru gives in [@Cal2] when $C$ is a curve of type $I_2$ proves the following
If $C$ is any reducible fiber of a elliptic fibration of type ($*$), then the sheaves $\mathcal{E}_p$ satisfy $${\operatorname{Ext}}^i(\mathcal{E}_p,\mathcal{E}_{p'})=
\begin{cases}
\kappa \text{ \ \ if } p=p' \text{ and } i=0 \\
0 \text{ \ \ if } p \neq p' \text{ and all } i\notag
\end{cases}$$ In particular the moduli space $\overline{M}(C)$ has dimension 1 and is smooth at $[\mathcal{E}_p]$ for any smooth point $p\in C_0$.
Thus if $p$ and $p'$ are two different smooth points of $C_0$, since the sheaves $\mathcal{E}_p$ and $\mathcal{E}_{p'}$ are stable and ${\operatorname{Hom}}(\mathcal{E}_p,\mathcal{E}_{p'})=0$, we have that $\phi(p)\neq \phi(p')$. Then the map $\phi$ is injective in $C_0\setminus \overline{C_0}$ where $\overline{C_0}$ denotes the complementary subcurve of $C_0$ in $C$. Since $C_0$ is irreducible and $[\mathcal{O}_C]\in {\operatorname{Im}}\phi$, the map $\phi$ factors as $$\xymatrix{ C_0\ar[r]\ar[d] & \overline{M}(C)\\
M'(C)\ar@{^{(}->}[ur]}$$ where $M'(C)$ is an irreducible component of $\overline{M}(C)$ that contains the point $[\mathcal{O}_C]$. But since $M'(C)$ is of dimension 1 and smooth at $[\mathcal{O}_C]$, it is the unique irreducible component of $\overline{M}(C)$ containing $[\mathcal{O}_C]$ and the map $\phi\colon C_0\to M'(C)$ is also surjective.
In this point, we distinguish to cases:
1- If $C$ is a curve of type I$_N$ with $N\geq 2$, let $\{r_1,
r_2\}$ be the two intersection points between $C_0$ and $\overline{C_0}$. By Proposition \[p:semiest\], we have that $\phi(r_1)=\phi(r_2)$ and then $M'(C)$ is a rational curve with one node (Figure 1).
Figure 1.
In fact, since by Proposition 5.13 in [@L], stable line bundles on $C$ of degree 0 are given by the group exact sequence $$0\to \mathbb{G}_m\to \overline{M}^s(C)\to \prod_{i=1}^N
{\operatorname{Pic}}^0(C_i)\to 0$$ and, by Corollary 6.7 in [@L], there is only one extra point in $\overline{M}(C)$ corresponding to any strictly semistable sheaf, we can conclude that $\overline{M}(C)=M'(C)$. Thus the moduli space $\overline{M}(C)$ is in this case isomorphic to a rational curve with one node.
2- If $C$ is a curve of type III or IV and $r$ is the intersection point of $C_0$ and $\overline{C_0}$, we know that $\phi(r)$ is the unique singular point of $M'(C)$ and that it is a cusp (Figure 2).
Figure 2.
Moreover, Propositions 5.6 and 5.8 of [@L] imply that stable line bundles on $C$ of degree 0 are determined by the group exact sequence: $$0\to \mathbb{G}_a\to \overline{M}^s(C)\to \prod_{i=1}^N
{\operatorname{Pic}}^0(C_i)\to 0$$ and as above there is only one extra point in $\overline{M}(C)$. Then we conclude that $\overline{M}(C)=M'(C)$ is isomorphic to a cuspidal irreducible curve.
In this case we don’t know a priori the arithmetic genus of $\overline{M}(C)$. However in the next section we will prove that the relative Jacobian $\hat{f}\colon \overline{M}_{X/S}\to S$ of $f$ is a flat morphism, and then all its fibers have arithmetic genus 1. This allows us to conclude that the moduli space $\overline{M}(C)$ is isomorphic to a rational curve with one cusp.
The Flatness of the relative Jacobian
=====================================
Let $f\colon X\to
S$ be an elliptic fibration of type ($*$) and $\hat{f}\colon
\overline{M}_{X/S}\to S$ its relative Jacobian. From the previous section we know that this relative Jacobian is a projective morphism whose fibers are integral curves. We are now going to prove that it is a flat morphism.
Using Corollary 15.2.3 in [@EGA] about the flatness of universally open morphisms with reduced fibers, in order to prove that $\hat{f}$ is flat, we only need to see that it is universally open. Taking into account that $S$ is a smooth variety and then geometrically unibranch, if $\hat{f}$ is equidimensional we conclude thanks to Chevalley’s criterion (Corollary 14.4.4 in [@EGA]). Since $\hat{f}$ is surjective and all its fibers have the same dimension, to show that it is equidimensional it is enough to prove that $\overline{M}_{X/S}$ is irreducible.
Notice first that, since for all $s\in S$ the sheaf $\mathcal{O}_{X_s}$ is stable, the flat family $\mathcal{O}_X$ defines a natural section $\sigma\colon S\hookrightarrow
\overline{M}_{X/S}$ of $\hat{f}\colon \overline{M}_{X/S}\to S$ whose image is contained in only one irreducible component $M'$ of $\overline{M}_{X/S}$. Indeed, since $S$ is irreducible, ${\operatorname{Im}}\sigma$ is contained in some irreducible component of $
\overline{M}_{X/S}$. But for all $s\in S$, $\sigma(s)=[\mathcal{O}_{X_s}]$ is a smooth point of $\overline{M}_{X/S}$ so that this irreducible component $M'$ is unique.
Let us see that $\overline{M}_{X/S}=M'$. Actually, since $(\overline{M}_{X/S})_s$ is irreducible for every $s\in S$, it is contained in some irreducible component of $\overline{M}_{X/S}$, namely $\tilde{M}$. If $\tilde{M}\neq M'$, the point $[\mathcal{O}_{X_s}]$, which lies on $(\overline{M}_{X/S})_s$ and on the image of the section $\sigma$, belongs to $\tilde{M}\cap
M'$. But this is not possible because this is a smooth point of $\overline{M}_{X/S}$. Hence all fibers $(\overline{M}_{X/S})_s$ are contained in the irreducible component $M'$ and so it is the whole moduli space $\overline{M}_{X/S}$.
The final result is then the following
\[teorema\] If $f\colon X\to S$ is an elliptic fibration of type ($*$) (Definition \[d:fibracion\]), the moduli space of semistable pure dimension one sheaves of rank 1 and degree 0 on a fiber $X_s$ is isomorphic to:
1. A smooth elliptic curve if $X_s$ is smooth.
2. A rational curve with one node if $X_s$ is of type I$_N$ with $N\geq 1$.
3. A rational curve with one cusp if $X_s$ is of type II, III, or IV.
Thus the relative Jacobian $\hat{f}\colon \overline{M}_{X/S}\to S$ of an elliptic fibration of type ($*$) is an integral elliptic fibration which always has a global section even if the original fibration has no sections. Here the difference between the integral case and the case with reducible fibers is the following. If the integral elliptic fibration has a global section, we know that it is globally isomorphic to its relative Jacobian, in contrast if the original fibration has reducible fibers even if it has a section, we are not able to ensure that it is globally isomorphic to its relative Jacobian because, as we will see now, $\overline{M}_{X/S}$ can be a singular space.
In fact, this theorem shows that as long as the original fibration $f\colon X\to S$ has reducible fibres, to get its relative Jacobian $\hat{f}\colon \overline{M}_{X/S}\to S$ we have to contract to a point all but one irreducible components of every reducible fibre $X_s$ of $f$, that is, for every reducible fibre of $f$ we have to contract to a point $q\in\overline{M}_{X/S}$ a linear chain $\cup_i C_i$ of smooth rational curves.
When $f\colon X\to S$ is a smooth elliptic surface, being $C_i^2=-2$ for every $i$, we know by [@A] that the contraction of a such chain is a singular point. Since the discriminant locus of $f$ is a finite number of points, we conclude that the relative Jacobian $\hat{f}\colon
\overline{M}_{X/S}\to S$ is an integral elliptic surface with at worst a finite number of singular points. When $f\colon X\to S$ is a smooth elliptic threefold, by [@M] every irreducible component of these chains has length 1, and then using the results in [@Z] we have that the contraction point $q$ is also a singular point. However in this case, since the discriminant locus of $f$ has dimension one, the singular locus of the relative Jacobian $\overline{M}_{X/S}$ has dimension less or equal to 1.
[99]{}
B. Andreas, G. Curio, D. Hernández Ruipérez , S. Yau, [*Fibrewise T-Duality for D-Branes on Elliptic Calabi-Yau*]{}, J. High Energy Phys. [**3**]{} (2001), Paper 20, 13 pp.
M. Artin, [*Some numerical criteria for contractability of curves on algebraic surfaces*]{}, Amer. J. Math [**84**]{} (1962), 485-496.
P. S. Aspinwall, [*Some navigation rules for D-brane monodromy*]{}, J. Math. Phys. [**42**]{} (2001), 5534-5552 .
C. Bartocci, U. Bruzzo, D. Hernández Ruipérez, J.M. Muñoz Porras, [*Mirror symmetry on K3 surfaces via Fourier-Mukai transform*]{}, Comm. Math. Phys. [**195**]{} (1998), 79-93.
P. Candelas, X. de la Osa, P. S. Green, L. Parkes [*A pair of Calabi-Yau manifolds as an exactly soluble superconformal theory*]{}, Nuclear Phys. B [**359**]{} (1991), 21-74.
A. Căldăraru, [*Derived categories of twisted sheaves on Calabi-Yau manifolds*]{}, Ph. D. thesis, Cornell University (2000).
A. Căldăraru, [*Derived categories of twisted sheaves on elliptic threefolds*]{}, J. Reine. Angew. Math. [**544**]{} (2002), 161-179.
R. Donagi, [*Taniguchi lectures on principal bundles on elliptic fibrations*]{}, Integrable systems and algebraic geometry (Kobe/Kyoto, 1997) 33-46, World Sci. Publishing, River Edge, NJ, (1998).
M. R. Douglas, [*Gauge fields and D-branes*]{}, J. Geom. Phys. [**28**]{} (1998), 255-262.
K. Fukaya, [*Morse homotopy, $A^\infty$-category, and Floer homologies*]{}, in Proceedings of GARC Workshop on Geometry and Topology 93 (Seoul, 1993), vol. 18 of Lecture Notes Ser., Seoul, 1993, Seoul Nat. Univ., 1-102.
B. R. Greene, M. R. Plesser, [*Duality in Calabi-Yau moduli space*]{}, Nuclear Phys. B [**338**]{} (1990), 15-37.
A. Grothendieck, [*Éléments de Géométrie Algébrique IV. Étude locale des schémas et des morphismes de schémas*]{}, Publ. Math. Inst. Hautes Études Sci. (1967).
D. Hernández Ruipérez, J.M. Muñoz Porras [*Stable sheaves on elliptic fibrations*]{}, J. Geom. Phys [**43**]{} (2002), 163-183.
P. Horja, [*Derived category automorphisms from mirror symmetry*]{} to appear in Duke Mathematical Journal, math.AG/0103231.
K. Kodaira, [*On compact analitic surfaces II*]{}, Annals of Math. [**77**]{} (1963), 563-626.
M. Kontsevich, [*Homological algebra of mirror symmetry*]{}, in Proceedings of the International Congress of Mathematicians, Vol. 1, 2 (Z[ü]{}rich, 1994), Basel, 1995, Birkh[ä]{}user, 120-139.
A. C. López Martín, [*Simpson Jacobians of reducible curves*]{} to appear in J. Reine. Angew. Math.
R. Miranda, [*Smooth models for elliptic threefolds*]{}, Birational Geometry of Degenerations, Progress in Math. [**29**]{}, Birkhäuser, Boston (1983), 85-133.
C. S. Seshadri, [*Fibrés vectoriels sur les courbes algébriques*]{}, Asterisque [**96**]{} (1982).
C. T. Simpson, [*Moduli of Representations of Fundamental Group of a Smooth Projective Variety I*]{}, Publ. Math. Inst. Hautes Études Sci. [**79**]{} (1994), 47-129.
T. Zerger, [*Contraction criteria for reducible rational curves with components of length one in smooth complex threefolds*]{}, Pacific J. Math [**212**]{} (2003), 377-394.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Cold Dark Stars made of self-gravitating fermions in the degenerate limit are constructed in General Relativity and in R-square gravity, $f(R)=R+\alpha R^2$. The properties of the resulting Cold Dark Stars in both theories of gravity are studied. It is found that the same gravitational potential is generated for different election of the parameters of the model, such as the mass of the fermion, the self-interacting strength or the value of $\alpha$, thus, a possible confusion in the determination of the dark matter properties and the favored theory of gravity might arise.'
author:
- 'Armando A. Roque'
- 'J. Barranco'
bibliography:
- 'confusing.bib'
title: Confusing dark matter particle properties with modifications to General Relativity
---
Introduction
============
Current astrophysical observations favor General Relativity (GR) as the correct theory that describes the gravitational interaction [@Abbott:2018lct; @Sakstein:2017xjx]. Nevertheless, if GR is valid, in order to have a concordance model for the evolution of the universe, two dark components must be added to the energy density content of the universe i.e. Dark Matter (DM) and Dark Energy (DE), the latest in the form of a cosmological constant. This model that have as main ingredients GR, DM as a cold heavy non relativistic particle and a cosmological constant is known as the $\Lambda$-CDM model [@Peebles1993PrinciplesOP]. This model is consistent with most of the observational data (e.g. measurement of anisotropies in the temperature and polarization of the cosmic microwave background (CMB) [@Bennett:2013], fluctuations in the density of baryonic matter (BAO) [@Eisenstein:2005su; @Cole:2005sx; @Alam:2016hwk], observations of the magnitude-redshift relation for high redshift SNe Ia [@Riess:1998cb; @Scolnic:2017caz]) among many other observations. However, there are some unresolved problems: the Planck best-fit measurements of the current expansion rate $H_0$ [@Akrami:2018vks] is in tension with the value obtained by local measurements of $H_0$ [@Riess:2019cxk; @Wong:2019kwg; @Freedman:2019jwv; @Schoneberg:2019wmt]. Furthermore, the $\Lambda$-CDM model is not compatible with some astrophysical observations at galactic scales. Perhaps more important, $\Lambda$-CDM model has little to say about the nature of dark matter, except that it could be a heavy neutral particle that interacts very weakly with the rest of the particles of the Standard Model of particle physics (SM).
An unpleasant possibility, although favored by its simplicity, is that DM particles interact only through gravitational interactions, and thus, the only parameters to be determined are the spin and the mass of the dark matter particle, plus the coupling constants that parametrize the interactions among themselves.
If DM has no other interaction than gravitational, then DM particle properties will be obtained only through astrophysical observations of the dynamics of visible objects with the gravitational force generated by the DM.
In this scenario, another unpleasant possibility is that [*there is a degeneracy in the determination of the DM properties once variations to the theory of General Relativity are allowed*]{}. The objective of this work is to show one example of this possibility.
For definitiveness, we will consider as a model for DM particles, fermions of mass $m$ that have not any other interaction with the standard model of particles except the gravitational interaction. In particular this model of fermionic dark matter has been used to model dark matter halos [@Destri:2012yn; @Destri:2013pt; @Domcke:2014kla; @Randall:2016bqw; @Barranco:2018gjg; @Savchenko:2019qnn; @Gomez:2019mtl]. This dark matter fermions can interact themselves and thus a self-interaction term will be added [@PhysRevD.64.043005].
On the other hand, as for modification of GR, we will focus on a theory of gravity that is derived from the action $$S=-\int d^{4}x \sqrt{-g}(\text{R}+\alpha \text{R}^{2}), \label{action}$$ where $\text{R}$ is the Ricci scalar. This is known as R-square gravity or as the Starobinsky model [@Starobinsky:1980te; @Gottlober:1990um; @Cembranos:2008gj] This modification to GR is a particular case of the so-called $f(R)$ theories of gravity [@Capozziello:2011et]. Note that R-square gravity is not introduced to solve the low energy problems of GR such as to avoid the introduction of dark matter or dark energy. Instead, we have used it because R-square gravity it is the simplest non trivial four-derivative extension of GR that is free of ghosts [@Cembranos:2008gj; @Capozziello:2011et] and thus it is a natural extension to GR to be studied. We have been motivated to use this fermionic dark matter model and R-square gravity because the minimal number of parameters that are introduced. Even in this minimal scenario we will show there could be a confusion in the determination of fermions properties for different values of $\alpha$. In general, other modifications to GR have more than one free parameter, thus strengthening the confusing of DM properties. (e.g. Hu $\&$ Sawicki model [@PhysRevD.76.064004], Starobinsky (2007) [@Starobinsky:2007hu] , The exponential model [@Linder:2009jz; @Odintsov:2017qif]).
For a distant observed, the gravitational effects of compact objects depend only of their total mass and radius (i.e. the compactness). What we will show is that self-gravitating configurations made of DM fermions will have the same mass and radius although in one case one configuration is made of fermions with a specific value of self-interacting coupling constant in General relativity while the equivalent configuration is made for a different self-interacting constant but in R-square gravity. Thus, there is a possible confusion in the determination of the dark matter properties. Even in this very simplified scenario where few parameters are present. In the case where DM-SM interactions are introduced, a bigger indetermination will be expected.
Next, we solve the Tolman-Oppenheimer-Volkov equations in order to find the self-gravitating configurations made of this self-interacting fermions in both GR and R-square theories of gravity. We will call this DM self-gravitating compact objects as Cold Dark Stars (CDS). We study the general properties of CDS and we will find equivalent configurations for different elections of parameters.
The paper is organized as follows: In Section \[configurations\] we obtain the self-gravitating objects made of fermionic dark matter (CDS) both in General Relativity and in R-square gravity. Some previous results in GR are reproduced [@Narain:2006kx] and new results are obtained, specially those concerning the compactness of the resulting configurations. We present in section \[sec:results\] a comparison of CDS in GR and gravity and the equivalence of some configurations even for different values of the self-interacting coupling, i.e. a confusion on the determination of the DM properties. In Section \[sec:conclusions\] we give some concluding remarks.
Cold Dark Stars {#configurations}
===============
Dark matter over-densities might form small clumps that can evolve into Dark Stars, i.e self-gravitating objects made of dark matter [@Freese:2008hb; @Berezinsky:2014wya; @Freese:2015mta]. In is generally assumed that those Dark Stars will be powered by the heat from dark matter annihilation, rather than by fusion. In our case, since we are assuming that DM has no interaction with SM particles, thus, our Dark Stars will be Cold compact objects and in order to distinguish from dark stars we will call the resulting self-gravitating objects made of Fermionic DM in the degenerate limit as Cold Dark Stars.
CDS in General Relativity
-------------------------
For non interacting fermions in the degenerate limit, it is possible to establish a relationship between the pressure and the density, for a gas of free fermions. This relationship can be calculated via explicit expressions for the energy density $\rho$ and pressure $p$. For a completely degenerate gas of fermions $\rho$ and $p$ are given by [@Landau:1980mil; @Narain:2006kx] (in units where $c=\hbar=1$): $$\rho(z)=\frac{m^4}{8\pi^{2}}\left[ \left(2 z^{3}-3z\right)\left(1+z^{2}\right)^{1/2}+3\sinh^{-1}(z)\right]\,,
$$
$$p(z)=\frac{m^4}{24\pi^{2}}\left[ \left(2 z^{3}+z\right)\left(1+z^{2}\right)^{1/2}-\sinh^{-1}(z)\right]\,,
$$
where $z=k_{f}/m$ is the dimensionless Fermi momentum and $m$ the mass of the fermion. It is convenient to work in dimensionless variables so we define the dimensionless variables $$\bar p=\frac{p}{m^4},\;\; \bar \rho=\frac{\rho}{m^4},\;\;\bar{r}=r \frac{m^2}{m_p}\,, \label{dimensionless}$$ where, $m$ is the fermion mass, and $m_{p}$ is the Planck mass, defined as $m_{p}=\text{G}^{-1/2}$.
We will consider self-interacting dark matter since it may resolve some problems of $\Lambda$-CDM paradigm at galactic scales [@Spergel:1999mh]. Following [@Narain:2006kx], the interparticle interactions can be added effectively by adding a term proportional to the square of the density number of fermions: $$\begin{aligned}
\bar{\rho}(z)
&=&\frac{1}{8\pi^{2}}\left[ \left(2 z^{3}-3z\right)\left(1+z^{2}\right)^{1/2}+3\sinh^{-1}(z)\right]\nonumber\\ & &\;\;\;+\frac{1}{9\pi^{4}}y^{2}z^{6}. \label{eqrho}\end{aligned}$$
$$\begin{aligned}
\bar{p}(z)
&=&\frac{1}{24\pi^{2}}\left[ \left(2 z^{3}+z\right)\left(1+z^{2}\right)^{1/2}-\sinh^{-1}(z)\right]\nonumber\\ & &\;\;\;+\frac{1}{9\pi^{4}}y^{2}z^{6},\label{eqp}\end{aligned}$$
where $y$ is the interaction strength between fermions.
In order to obtain the self-gravitating object made of this gas of fermions, it is needed to solve Einstein’s equations and the conservation of the energy tensor.
Taking $\alpha=0$ in eq. , the action reduces to the General Relativity action. We consider a static, spherically symmetric ansatz for the line element, $$ds^{2}=-A^{2}(r)dt^{2}+B^{2}(r)dr^{2}+r^{2}d\theta^{2}+r^{2}\sin^2\theta d\varphi^{2}\,\label{metric}.$$ The energy momentum tensor for a perfect fluid is given by $$T^{\mu\nu}=(\rho+p)u^{\mu}u^{\nu}+g^{\mu\nu}p$$ and thus, Einstein-equations reduces to the Tolman-Oppenheimer-Volkoff equations (TOV): $$\begin{aligned}
B'&=&\frac{B^3 \left(8\pi\bar{r}^2 \bar{\rho}-1\right)+B}{2 \bar{r}},\\
2 A'&=&\frac{A \left(B^2 \left(8\pi\bar{r}^2 \bar{p}+1\right)-1\right)}{\bar{r}},\\
\bar{p}'&=&-\frac{A' (\bar{\rho}+\bar{p})}{A},\end{aligned}$$
We set as boundary conditions $A(0)=B(0)=1,\;\;\bar{p}(0)=\bar p_{0}$, with $\bar p_{0}$ a free parameter. As the equation of state depends on two parameters, $z$ and $y$, the $\bar{\rho}(\bar{p})$ relationship was obtained of interpolation of results of evaluate eq. and eq. for a range of values $0<z<20$ given a fixed interaction strength $y$. The range of explored for $y$ are between $0$ and $10$. And, for our analysis, the central pressure of the object $\bar{p}_{0}$, took values between $10^{-10}$ to $10$.
![Typical energy density profile and mass as a function of the radial coordinate $r$. For definitiveness it was chosen $\bar \rho_0=2.1\times10^{-6}$, and two values of the self-interacting strength: $y=1$ (dashed line) and $y=3$ (solid line). The introduction of self-interaction produces a more massive self-gravitating structure and with a stiffer density profile.[]{data-label="Fig01"}](Perfiles_RG.png){width=".5\textwidth"}
The mass of this objects was calculated using the ADM mass, $$M_{\text{ADM}}=\frac{1}{2} \bar{r} \left(1-\frac{1}{B^2}\right)
\label{eqmas}.$$
Typical configurations are shown in Fig. \[Fig01\], where the profile for the energy density as a function of $\bar{r}$ are shown for a fixed value of $\bar{p}_0$ (that is equivalent to a fixed value of $\bar{\rho}_0$) for two different values of the self-interacting strength $y$. For definitiveness it was chosen $\bar{\rho}_0=2.1\times10^{-6}$, and two values of the self-interacting strength: $y=1$ (dashed line) and $y=3$ (solid line). The introduction of self-interaction produces a more massive self-gravitating structure and with a stiffer density profile. The radius $\bar{R}$ is defined as the point where the ADM Mass starts to be constant ($\bar p_0 <0$). Thus, each CDS will have a fixed total mass $\bar{M}$ and a finite radius $\bar{R}$.
It is possible to obtain the full set of CDS for our parameter space $(\bar p_0,y) \to (\bar{R}, \bar{M}, y)$. These configurations are shown in Fig. \[Fig0\]. Note that, each value of $y$ imply a different equation of state. For small values of this and lower densities, the interaction terms can be ignored and the object was describe for an ideal Fermi gas. For higher $y$ values, the interaction terms become more and more important and the *transition point* from the relativistic curve to no relativistic move to lower densities. Similar behavior occur with the stability border of the dark star (we can defined the stability border from the maximum gravitational mass of the Cold Dark Star). It can be seen that the stability border move to right (lower central pressure) when is increasing the interaction strength parameter. As was expect, the mass ground for higher values of $y$, it is because the degeneration pressure is strengthened and it necessary more mass for counteract this, and arrived to hydrostatic equilibrium.
![ Black solid line correspond to $M$ [*vs*]{} $R$ for $y=0$ (General Relativity without auto-interaction) and the blues doted-lines corresponds to the dark star made of interacting fermions in General Relativity with coupling constant for a rang between $y=1$ and $y=5$.[]{data-label="Fig0"}](RG.png){width="49.00000%"}
Because the dimensionless definitions of the variables eq. , all values obtained in Figs. \[Fig01\]-\[Fig0\] are dimensionless as well. In order to obtain the physical values of the mass and energy density we have the following relations $$M_{phys}=\bar{M}_{\text{ADM}}\frac{m_p^3}{m^2},\; \rho_{phys}=\bar{\rho}\, m^4,\; R_{phys}=\bar{R}\frac{m_p}{m^2}\,,\label{physical}$$
Note that if $m=1~$KeV then $M_{phys}\sim10^{12}M_{\odot}$ and $R_{phys}\sim 0.1$pc, e.g. a possible super massive black hole mimicker.
We finish this section by enumerating some general properties of CDS. In GR the self gravitating configurations made of degenerate non interacting fermions (CDS) have the following properties:
1. The total mass of the CDS increases as $\bar \rho_0$ increases, then the mass reaches a maximum value $\bar{M}_{max}$ that defines the stability region.
2. The radius of the configuration $\bar{R}$ decreases as $\bar \rho_0$ increases.
3. The introduction of a self-interacting coupling between the fermions ($y>0$) make that the configuration increases the value of the maximum mass $\bar{M}_{max}$
Cold Dark Star in R square gravity
----------------------------------
Now we study CDS in the R-square gravity given by the action eq. .
The generic field equations for a theory $f(R)$ are given by, $$\begin{aligned}
f_{R} R_{\mu\nu}-\frac{1}{2} g_{\mu\nu} f(R)+ g_{\mu\nu}\Box f_{R} -
\nabla_{\nu}\nabla_{\mu} f_{R}= k T_{\mu\nu},\label{emod}\end{aligned}$$ where, $f_{R} :=\partial_{R} f(R)$, $\Box=g^{\mu\nu}\nabla_{\mu}\nabla_{\nu}$ .
Taking $f(\text{R})=\text{R}+\alpha \text{R}^2$, (where $\alpha$ is a constant with units of the inverse of the Ricci scalar $[\text{R}^{-1}]$) in eq. and using the 1-1, 2-2 components, and a spherically symmetric line element eq. and the conservation equation $\nabla_{\mu} T^{\mu \nu}=0$ with $T^{\mu\nu}$ a perfect fluid as before, it is possible to obtain the modified Tolman-Oppenheimer-Volkoff for R-square gravity. The R-square TOV system is given by:
$$\begin{aligned}
B'&=&\frac{B \left(2+B^2 \left(-2+\bar{\alpha} \bar{\text{R}} \left(-4+\bar{r}^2 \bar{\text{R}}\right)+16 \pi \bar{r}^2 \bar{\rho}\right)+4 \bar{\alpha}\left(\bar{\text{R}}+\bar{r}\left(2 \bar{\text{R}}'+\bar{r} \bar{\text{R}}''\right)\right)\right)}{4 \bar{r} \left(1+2 \bar{\alpha} \bar{\text{R}}+\bar{r} \bar{\alpha}\bar{\text{R}}'\right)}\label{eq0}\\
A'&=&\frac{A \left(-2-4 \bar{\alpha}\bar{\text{R}}+B^2 \left(2+16\pi\bar{r}^2\bar{p}+\bar{\alpha}\bar{\text{R}} \left(4-\bar{r}^2\bar{\text{R}}\right)\right)-8\bar{r}\bar{\alpha} \bar{\text{R}}'\right)}{4 \bar{r} \left(1+2 \bar{\alpha} \bar{\text{R}}+\bar{r} \bar{\alpha}\bar{\text{R}}'\right)}\label{eq01}\\
\bar{p}'&=&-\frac{A' (\bar{\rho}+\bar{p})}{A}\label{eq02}
\end{aligned}$$
For this case, we need a extra equation for describe the Ricci scalar behaviour because the theory have a extra degree of freedom. It is given by:
$$\begin{gathered}
\bar{\text{R}}''=\frac{6\bar{\alpha}\bar{\text{R}}'\left(-1-2\bar{\alpha} \bar{\text{R}}+2\bar{r}\bar{\alpha}\bar{\text{R}}'\right)+B^2 \left(\bar{r}(1+2\bar{\alpha}\bar{\text{R}})(24\pi \bar{p}+\bar{\text{R}}-8\pi\bar{\rho})+\bar{\alpha}\left(-6+\bar{\text{R}} \left(\bar{r}^2-12\bar{\alpha}+3\bar{r}^2 \bar{\alpha} \bar{\text{R}}\right)+16 \pi \bar{r}^2 \bar{\rho}\right) \bar{\text{R}}'\right)}{6\bar{r}\bar{\alpha}(1+2\bar{\alpha} \bar{\text{R}})}\label{eq10}
\end{gathered}$$
Note that eqs. - are written in dimensionless variables where we have used the extra variable change, $\alpha=\bar{\alpha}\, m_{p}^2/m^4$ and $\text{R}=\bar{\text{R}}\, m^{4}/m_{p}^{2}$. The value of the free parameter $\alpha$ of R-squared theory is constrained from observations in different scales. In the cosmological context, if one takes $\alpha<0$ then ghost modes instabilities arises [@Barrow_1983]. Furthermore, for negative values of $\alpha$, the Ricci scalar profile has a oscillating behaviour outside the star and this oscillating behaviour is in contradiction with our boundary conditions. Similar behaviour occurs in the neutron star context [@Astashenok_2017]. Then, we will work with $\alpha>0$ values.
From the strong gravity regime, $\left| \alpha \right| $ is constrained to be $\lesssim 10^{10} \;\text{cm}^{2}$ [@Arapo_lu_2011]. For weak-field limit, it is constrained by different experiments, the E$\ddot{\text{o}}$t-Wash provides the more stringent bound, $\left| \alpha \right| \lesssim 10^{-6}\;\text{cm}^{2}$, the Gravity Probe B, constrains $\left| \alpha \right|$ to have values $\lesssim 5\times 10^{15}\;\text{cm}^{2}$ and from measurements of the precession of the pulsar B in the PSR J0737-3039, $\left| \alpha \right| \lesssim 2.3\times10^{19} \; \text{cm}^{2}$. [@PhysRevD.81.104003] Although all the bounds has differences, they are still meaningful, because this type of theory present a chameleon effect and thus the $\alpha$ values could be different at different scales. In our case, to compare $\alpha$ with $\bar \alpha$ we recall that $$\alpha= 5.79 \times 10^{-5} \bar \alpha \left(\frac{m}{1\mbox{TeV}}\right)^{-2} \mbox{cm}^2\,.$$ For $m \sim 1$ TeV, values of $\bar \alpha \sim 10^{-1}$ will be in concordance with the strongest limit of $\left| \alpha \right| \lesssim 10^{-6}\;\text{cm}^{2}$. We will find the configurations for a specific value of $\bar \alpha=0.05$
{width="49.00000%"} {width="50.00000%"}
The next step for the numerical analysis is to define the boundary conditions compatible with solutions that are regular at the origin, localized, and asymptotically flat. We expand the system equation around to the origin in order to find the boundary conditions at $r=0$. It was found: $$\begin{aligned}
\label{boundayC}
\bar{\text{R}}(0)=\text{R}_{0},\;\;\; \bar{\text{R}}'(0)=0.\end{aligned}$$ where the “prime” indicate derivation with respect to $\bar{r}$. Here $\text{R}_{0}$ has to be chosen so that the Ricci scalar vanishes asymptotically (at $r \to \infty$). The correct value of $\text{R}_{0}$ such as the boundary conditions are fulfilled is found with a shooting algorithm. [@Numerical]
Since $\text{R}$ approaches asymptotically to zero, numerically we have to choose a particular value of $r$ such that the value of Ricci scalar is almost zero. Numerically we choose a value $\bar{r}=r_{\star}$ such as .
Because the Ricci scalar vanishes at infinity, the space-time metric approaches to the Schwarzschild space time and thus we can use the Schwarzschild metric to estimate the mass of object using with $\bar{r}=r_{\star}$. As the Ricci scalar does not strictly vanish for $r_{\star}$, we demand that $\bar{M}'(r_{\star})\approx 0$ must be satisfied, the “prime” indicate derivation respect to $\bar{r}$.
The system equations - were solved numerically with an equation of state given as in the previous section, i.e. a equation of state for a gas of self-interacting fermions in the degenerate limit. Typical selfgravitating configurations for two different values of the coupling strength ($y=1$ red solid line and $y=3$ blue line) are shown in Fig. \[Fig03\]. It is shown the density, $\bar{\rho}(\bar{r})$, and mass, $\bar{M}(\bar{r})$, profiles for the specific value of the central pressure $\bar \rho_0=2.1\times10^{-6}$. The profile for the Ricci scalar and their derivative are also shown in Fig. \[Fig03\].
We finish this section by constructing all configurations for a specific value of $\bar \alpha=0.05$. Solutions in the $M$ [*vs*]{} $R$ space for different values of the self-interacting strength $y$ are shown in Fig. \[Fig05\].
![ As in Fig. \[Fig0\], we obtained CDS configurations, but now in a R-square gravity with $\bar{\alpha} = 0.05$. Black solid line correspond to $y=0$ and the blues doted-lines corresponds to the dark star made of self-interacting fermions with coupling constant between $y=1$ and $y=5$. Grey line correspond to CDS in GR with $y=0$ in order to compare GR and R-square CDS configurations.[]{data-label="Fig05"}](M99_vsR99.png){width="49.00000%"}
In R-square gravity the CDS have the same behaviour as in GR, but the new parameter $\bar \alpha$ modifies the global structure of the configuration. In particular, in R-square gravity the maximum masses $M_{max}$ are smaller than the GR case for the same value of the self-interacting coupling constant $y$ (see Figs. 2-4). In order to understand this behaviour, let us take the Newtonian limit of R-square gravity.
In this case, the Newtonian gravitational potential for R-square gravity is [@Stelle:1977ry]: $$V(r)=-\frac{GM}{r}-\frac{GM}{3r}e^{-\beta r}\,,\label{newtonian}$$ where $\beta=\frac{1}{2}\sqrt{3 \alpha}$. Thus, the gravitational force is increased if $\beta>0$, and consequently, a lower amount of mass is needed to compensate the pressure produced by the fermions. This reduction in the mass of the configuration can be seen in Fig. \[Fig05\] where we have included in a grey solid line the corresponding CDS configurations for $y=0$ in GR. The black solid line that corresponds to R-square gravity is always below the grey line, meaning that all CDS in R-square gravity are smaller that the GR case.
Another important observation is that the density tends to zero faster that the Ricci scalar, and then, the ADM-mass can increases even if the star has negligible contribution of fermions. The same holds for the derivative of the Ricci scalar. Thus, we can define a stellar radius $r_s$ where the density is almost negligible $\rho(r_s)=0$. Recall that we have the configuration radius $r_*$ chosen by the shooting algorithm such as the boundary conditions are fulfilled. In general, $r_* > rs$.
Similarities and differences of CDS in GR and R-square gravity {#sec:results}
==============================================================
Compactness
-----------
![Compactness of Fermionic dark stars as a function of the self-interaction strength parameter, y. The black solid line correspond to the case $\bar{\alpha}=0$ (General Relativity), the orange dashes lines represent the compactness for two case, $\bar{\alpha}=0.05$, and $\bar{\alpha}=1$.[]{data-label="compactness"}](Compac_RG.png){width="49.00000%"}
The Cold Dark Stars are objects make of self-interacting fermionic dark matter that do not interact with SM particles. For this reason they can not be observed by standard electromagnetic probes. However, in principle, they can be observed by gravity probes, such as gravitational wave (GW) signals or by the effect that CDS can induce gravitationally to other stars (e.g., [@Palenzuela:2017kcg; @Bezares:2018qwa]). In the later case, an object that is far to the Cold Dark Star will feel the same gravitational potential both in GR and in R square gravity as long as the ratio $M/R$ is the same. The parameter $M/R$ is called the compactness of the star $C$ defined as: $$C=\frac{M_{ADM}}{R}=\frac{GM_{phys}}{R_{phys}c^2}\,,$$ where $G$ is the gravitational constant and $c$ the speed of light, and $M_{phys}$, $R_{phys}$ are defined in eq. \[physical\]. Note that $C$ is a dimensionless parameter. The compactness of CDS is affected when we considered a self-interacting fermionic dark matter: the larger the value of $y$ the bigger the mass, meanwhile the radius of the configuration is not severely affected by the value of $y$ and thus, the compactness of the star increases as $y$ increases. Changes in the compactness will imply changes in several properties such as the possible gravitational radiation emmitted by an asymmetric star, or a binary of CDS and of course a change in the compactness induces changes in the gravitational potential the stars produces in other objects.
With this in mind, next we study the change in the maximum compactness of CDS as a function of the self-interacting strength constant $y$ value. The compactness of the CDS in GR ($\bar \alpha=0$) are shown as a solid black line in Fig. \[compactness\]. In GR, for $y=0$ (no self-interacting fermions) the maximum compactness is $C=0.11$, and it increases as $y$ increases. To have an order of magnitude, let us recall that the Sun compactness is $\sim 10^{-5}$. Thus CDS can be very compact objects. As a comparison of other compact objects, in Fig. \[compactness\] we have included other important compactness values. For instance, the maximum compactness for a fluid star, which is given by the Buchdahl’s limit and that is given by $C=4/9$ [@PhysRev.116.1027] is shown as a blue dotted line. In addition, the compactness of a Schwarzschild black hole, i.e. $C=1/2$, is shown as a green dashed line. Finally, the maximum compactness for a Boson Star which is given by $C=0.158$ [@AmaroSeoane:2010qx] is plotted as a red dotted line. In summary: Fermionic CDS have a maximum stable compactness bigger that Boson Stars.
Now we can study Cold Dark Stars compactness in R-square gravity. It is important to recall that in R-square gravity one can distinguish between the asymptotic compactness $C = M_{ADM}/r_*$ - which captures contributions to the mass due to the non-vanishing value of the Ricci scalar. Nevertheless, $r_*$ is formally reached at infinity as in the case of Boson Stars, because the Ricci scalar decays as $\sim 1/r^2$ and thus $\text{R}=0$ can be reached only as $r_* \to \infty$. We used instead of the asymptotic compactness the following definition of compactness: $$C = \frac{M_{99}}{R_{99}}\, ,\label{c99}$$ where $M_{99}=0.99 M_{ADM}$ and $R_{99}$ is the radius where $M_{99}$ is reached.
We have plotted the maximum stable compactness (eq. \[c99\]) in R-square gravity as a function of the self-interacting coupling $y$ in orange dashed lines of Fig. \[compactness\]. Two values of $\bar \alpha$ were chosen: $\bar \alpha =0.05$ and $\bar \alpha=1$. For increasing values of $\bar \alpha$, the maximum compactness is decreasing, because the maximum mass in R-square gravity decreases meanwhile $R_{99}$ is not severely affected. Note that there are configurations where the compactness for CDS in GR intersects the compactness for CDS in R-square gravity for different values of $y$. This intersection leads us to the confusion between dark matter properties with modifications of GR as we discuss in the next section.
Confusing
---------
![ For every case the mass profiles correspond to $f(\text{R})$ theory are down to the mass profile obtained for the same case (same “y” value) but, using General Relativity. Is possible increase the mass obtained in R-squared in comparative we GR, if we choose for the first a higher value of ”y”. []{data-label="Fig04"}](MfR.png){width="49.00000%"}
The previous results show the linear relationship between the free parameter, $y$, $\alpha$ and the mass of the resulting CDS. In this section we will show that for a different selection of parameters $m$, $y$, $\alpha$ and $\bar p_0$ there are equivalent configurations, and thus, a confusing on the determination of the parameters is possible.
To illustrate this point, we have constructed three different configurations shown in Fig. \[Fig04\]:
- Configuration A: Solid line corresponds to $\bar \rho_0=4.1\times10^{-5}$, $y=2$ (with self-interaction) $\bar \alpha=0$ (General Relativity)
- Configuration B: Dotted line corresponds to $\bar \rho_0=4.1\times10^{-5}$, $y=2.2$ (with self-interaction) $\bar \alpha=0.05$ (R-square gravity)
- Configuration C: Dotted line corresponds to $\bar \rho_0=4.1\times10^{-5}$, $y=1$ (with self-interaction) $\bar \alpha=0.05$ (R-square gravity)
- Configuration D: Dashed line corresponds to $\bar \rho_0=3.4\times10^{-5}$, $y=5$ (with self-interaction) $\bar \alpha=0.05$ (R-square gravity)
Note that configuration B has the same central density as configuration A, but different interaction strength in R-square gravity. Nervertheless, the mass of the configuration and the radius are are almost indistinguishable. Thus, any test particle will follow the same trajectory in both configurations.
On the other hand, Configuration C has a smaller central density, but due to the increase of the interaction strength $y$ the total mass increases and it is above the RG value. Thus, it seems that a there will be a value of $\rho_0 \in [3,4]\times 10^{-5}$ and $y \in [1,5]$ for a R-square gravity with $\bar \alpha=0.05$ will coincide with the GR case for $y=2$ and $\bar \rho_0=4.1\times10^{-5}$. Thus any gravitational signal for this two objects will be almost identical for different values of the coupling constant.
{width="47.00000%"} {width="47.00000%"}
There are other possible confusing possibilities in the determination of the parameters. Let us for instance construct all configurations with $y=3$ in GR and all configurations for $y=0$ in R-square gravity with $\bar \alpha=0.05$. The plots $M_{99}$ [*vs*]{} $R_{99}$ for those configurations are shown in left panel of Fig. 7. Note that there are 2 points that intersect both plots. Those configurations corresponds to CDS with the same compactness. Thus, any other object will feel the same gravitational potential outside this CDS: no dynamical differences will be seen neither infall differences. Thus, if one luminous star is orbiting this cold Dark Star and this is the only observable we have to constraints the dark matter particle properties (remember that in this case that will be the mass of the dark matter particle and the self-interaction strength).
Finally, given the typical astrophysical uncertainties, small variations on the compactness or the $M_{99}$ [*vs*]{} $R_{99}$ relation will be difficult to be disentangled. On the top of that, we can have the already discussed confusing between GR and R-square gravity. Let us illustrate this final possibility by constructing all possible configurations with $y=0.5$ in GR and all configurations for $y=0$ in R-square gravity with $\bar \alpha=0.1$. The plots $M_{99}$ [*vs*]{} $R_{99}$ for those configurations are shown in right panel of Fig. \[Fig1\]. Note that in this case there is an almost identical region where both plots intersects, and thus a infinite number of configurations can be confused.
Conclusions {#sec:conclusions}
===========
Modifications to general Relativity have been usually invoked to replace the role of dark matter or dark energy. Nevertheless, it could be possible that even with the existence of dark matter, there are possible modifications to general relativity. This modifications to GR seems to be mandatory specially if one looks for renormalizability of the gravitational interactions at high energies. Thus, we have consider a scenario where both dark matter and modifications to GR are present. In particular, we have consider dark matter as a self -interacting fermion in the degenerate limit and R-square gravity as a possible modification of GR. In this scenario, we have constructed self-gravitating structures made of this fermionic dark matter in both theories of gravity and studied their properties such as the compactness of the configurations. We have called this configurations Cold Dark Stars.
In the context of GR, we have shown that is possible to obtain configurations of CDS more compact that Boson Stars. Depending on the election of the mass of the fermion, CDS can be as massive and compact to be considered as black hole mimickers.
We have shown that CDS in R-square gravity, the bigger the quadratic term in the Einstein-Hilbert action, i.e. the bigger the value of $\bar \alpha$, the smaller the maximum mass of the resulting CDS configurations are obtained. This reduction in the mass will produce less compact CDS in R-square gravity in comparison with GR.
Considering the possibility that dark matter might interact only through gravitational interactions, the determination of their properties will be only accessible by astrophysical observations. Thus, given that similar CDS can be obtained for different election of coupling constants $y$ or values of $\bar \alpha$ in R-square theories, thus, a possible confusion in the determination of the dark matter properties and possible modifications to general relativity could be possible.
Acknowledgements {#acknowledgements .unnumbered}
----------------
This work was partially support by CONACYT projects CB-259228 and CB- 286651 and Conacyt-SNI. We thank Alberto Diez-Tejedor and Gustavo Niz for very enlightening discussion regarding this work.
\[sec:citeref\] References {#secciteref-references .unnumbered}
--------------------------
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In 2011, we discovered a compact gas cloud (”G2”) with roughly three Earth masses that is falling on a near-radial orbit toward the massive black hole in the Galactic Center. The orbit is well constrained and pericenter passage is predicted for early 2014. Our data beautifully show that G2 gets tidally sheared apart due to the massive black hole’s force. During the next months, we expect that in addition to the tidal effects, hydrodynamics get important, when G2 collides with the hot ambient gas around Sgr A\*. Simulations show that ultimately, the cloud’s material might fall into the massive black hole. Predictions for the accretion rate and luminosity evolution, however, are very difficult due to the many unknowns. Nevertheless, this might be a unique opportunity in the next years to observe how gas feeds a massive black hole in a galactic nucleus.'
---
Introduction
============
In 2011, we made a surprising discovery: Our long-term Very Large Telescope-based, near-infrared observations of the Galactic Center (GC) showed a small gas cloud (G2) falling on a nearly radial orbit onto Sgr A\* ([@gil12 Gillessen et al. 2012]). In particular, we detected a temporally increasing velocity shear of G2’s line emission in deep integral field spectroscopy data. This is the unambiguous sign of the massive black hole’s (MBH) tidal field. We have followed up the evolution with similar observations in 2012 ([@gil13a Gillessen et al. 2013a]) and 2013 ([@gil13b Gillessen et al. 2013b]), spectacularly showing the onset of the disruption of G2. The case caught the immediate attention of a broad audience, since this might constitute the unique opportunity to watch in real-time, how a MBH is getting fed.
Here, we summarize our observations, constraining its orbit and properties. We present a sequence of position-velocity diagrams, showing the tidal disruption, which can be well-described by a simple test particle model. Adopting a gas cloud model, hydrodynamic simulations predict, that this description remains a good approximation until pericenter. Afterwards the further evolution is dominated by hydrodynamics ([@sch12 Schartmann et al. 2012], [@ann12 Anninos et al. 2012]). The motion remains of course Keplerian throughout for models with a stellar source inside. We try to summarize what is known about the nature of G2, i.e. its origin, and finally present ideas, what future observations across all wavebands might be able to detect.
The orbit of G2
===============
G2 caught our attention as a fast moving L-band ($\approx 4\,\mu$m) source that apparently was on a curved trajectory towards Sgr A\*. It did not show a K-band counterpart, but we were able to see strong Brackett-$\gamma$ emission spatially coincident in our SINFONI ([@eis03 Eisenhauer et al. 2003], [@bon04 Bonnet et al. 2004]) data. The line position changed consistently over the years. In total, we obtained eight dynamical quantities: the position on sky (2), the proper motion (2), the acceleration (1), the radial velocity $v_\mathrm{LSR}$ (1), its change with time (1), and even a second derivative of $v_\mathrm{LSR}$ (1). An orbit has six free parameters, so it was non-trivial that we were able to actually find an orbit describing the data. In turn, it means that it is very probable that astrometry and $v_\mathrm{LSR}$ data belong to the same object. The initial orbit estimate had an estimated time of pericenter passage of around mid 2013 ([@gil12 Gillessen et al. 2012]).
[@phi13] showed convincingly that there is a systematic offset in the L-band positions compared to the positions as derived from Brackett-$\gamma$. This is probably due to the underlying, faint and spatially variable dust emission. The main difference is that the pericenter date of their orbit is shifted by a few months to early 2014. These authors used the Keck-based OSIRIS instrument ([@lar06 Larkin et al. 2006]). It uses a lenslet array to achieve the integral field spectroscopy, as compared to SINFONI that uses an image slicer. The former intrinsically has a cleaner astrometric performance. Nevertheless, we were able to reproduce the same orbit when switching to SINFONI-based, Brackett-$\gamma$ astrometry ([@gil13b Gillessen et al. 2013b]). This orbit is to be preferred because it is less prone to biases. We cannot exclude currently that there might be even a source-intrinsic bias, i.e. that the dust emission of G2 does not exactly trace the gas. That would be an interesting finding in its own.
Now, the VLT- and Keck orbit estimates are in good agreement (see figure \[fig1\]). The orientation of the orbit is near to that of the inner part of the clockwise stellar disk ([@pau06 Paumard et al. 2006], [@lu09 Lu et al. 2009], [@bar09 Bartko et al. 2009]). The estimated pericenter distance is around $r_p \approx 2000 R_S$ (Schwarzschild radii), comparable to the pericenter distance of the famous star S2 on a 16-year orbit ([@gil09 Gillessen et al. 2009], $1400\,R_S$). The eccentricity is very high with $e\approx 0.98$, which puts strong constraints on the nature of G2. There is a remaining uncertainty on $e$, which is mainly owed to the difficulty of measuring the radial velocity. G2 shows a velocity gradient across the source, and a large intrinsic line width. The latter has grown dramatically in the last few years to around $600\,$km/s, hampering the measurement of the line position. This continued disruption might also mean that the 2013 data are the last trustworthy, and that the true orbit will not be known any better anymore.
![Compilation and comparison of the different orbits for G2 published so far. []{data-label="fig1"}](gillessen_s_f1.pdf){width="13.5cm"}
Properties of G2
================
From the Brackett-$\gamma$ emission we find that G2 is marginally spatially resolved in our SINFONI data. We find an intrinsic Gaussian FWHM size of $42\pm10\,$mas. The fact that we hardly can resolve G2 spatially, while showing a large $v_\mathrm{LSR}$ gradient across the source, is due to the orientation of the orbit. G2 mostly moves along the line-of-sight away from the observer.
The dust emission of G2 can be described by a black-body emission of $T\approx 550\,$K. This relies on the L-band flux, an M-band detection and the limits in K-band ([@gil12 Gillessen et al. 2012], [@phi13 Phifer et al. 2013]). The finite size of $r \approx 150\,$AU together with the temperature estimate and the observed brightness excludes that the emission comes from an optically thick surface, which would need to have a radius of $r\approx 0.5\,$AU only.
Using the Brackett-$\gamma$ luminosity of $L\approx 2\times 10^{-3}\, L_\odot$ and case B recombination yields an estimated mass of G2 of a few times Earth’s mass and a characteristic density of $10^5\, \mathrm{cm}^{-3}$. The density is around two orders of magnitude above the ambient density, and G2’s mass exceeds the mass in Sgr A\*’s accretion flow enclosed in $r_p$. The total kinetic energy at pericenter passage will be around $10^{45}\,$erg, and if the material falls down to $1\,R_S$, more than $10^{48}\,$erg are available.
Comparing with the UV radiation field in the central arcsecond from the numerous luminous, young, hot stars around Sgr A\*, G2 is plausibly fully ionized. The Brackett-$\gamma$ luminosity of G2 has remained constant over the whole time range spanned by spectroscopy from 2004 to 2013. Beyond Brackett-$\gamma$, we also detect G2 in Helium-I ($2.05\,\mu$m) and Paschen-$\alpha$. The latter line happens to be observable between atmospheric absorption bands only due to the high redshift of the emission. The line ratios He-I/Br-$\gamma$ and Pa-$\alpha$/Br-$\gamma$ also remained constant for those epochs, in which we were able to measure them (2008-2013).
Tidal disruption of G2
======================
A gas cloud with a radius of $150\,$AU in a distance to Sgr A\* of $2800\,$AU (as G2 had in 2008) would need to have a mass of $\approx 10^4\,M_\odot$ in order to be gravitationally bound against the tidal field of the MBH. It is unavoidable, that G2 will undergo tidal disruption, and indeed our data beautifully show the onset and continued evolution of that process. Our SINFONI data are presented best in the form of position-velocity diagrams, where the spatial axis is the line element along the curved trajectory of G2’s orbit. Note that it is only possible to perform such an analysis because of the integral field aspect of our spectroscopy. Also, we added up the three diagrams obtained for each epoch from the three emission lines Brackett-$\gamma$, Helium-I, and Paschen-$\alpha$. A few of the resulting diagrams are shown in figure \[fig2\].
![From top to bottom: Snapshots in time of the infall of the gas cloud G2 into Sgr A\*. First column: LÕ-band images obtained with NACO of the central arcsecond. The white arrows mark G2, the yellow asterisk Sgr A\*. Note that we donÕt detect G2 in 2013. Second column: Snapshots from a hydrodynamic simulation of the infall ([@sch12 Schartmann et al. 2012]). Third column: Position-velocity diagrams from the gas recombination lines of G2 (white arrows) observed with SINFONI, showing the beautiful tidal evolution. Fourth column: A test particle simulation for G2 can describe the 2008-2013 position-velocity diagrams very well. []{data-label="fig2"}](gillessen_s_f2.pdf){width="13.5cm"}
A simple test particle model can describe the evolution remarkably well. An initially spherical cloud with a Gaussian FWHM of $42\,$mas and a Gaussian FWHM of the velocity dispersion of $120\,$km/s starting at $t=2000.0$ captures not only the overall evolution of the velocity shear, but also some finer details:
- There appears to be gas overshooting in $v_\mathrm{LSR}$ the bulk of the emission in the 2012 data set. This is also seen in the test particle simulation.
- In the April 2013 data set, some of the gas is already detected at the blue shifted side, after pericenter (figure \[fig3\]). Also this is expected from the test particle simulation.
The emission on the blue-shifted side is at the SNR limit of our data set.
It is worth pointing out, that the test particle simulation also predicts that the pericenter flyby has a significant intrinsic duration of over one year, simply due to the tidal stretching. In this sense, there does not exist a well-defined pericenter date.
![Our highest quality position-velocity diagram of G2, extracted from the 2013 April SINFONI data set. The position axis is counted along the orbit projected into the cube (shaded region in the right panel). This diagram is a co-add around the lines Brackett-$\gamma$, Helium-I, and Paschen-$\alpha$ for the red part of the diagram ($v_\mathrm{LSR}>0$), and of the former two lines for the blue side ($v_\mathrm{LSR}<0$). The yellow line delineates the L’-band based orbit, the white line the Brackett-$\gamma$ based one from [@gil13b]. []{data-label="fig3"}](gillessen_s_f3.pdf){width="13.5cm"}
In the position-velocity diagrams there seems to be a tail of gas following the same orbit as G2. The whole structure might thus be a much more elongated gas feature, of which G2 appears to be the head. [@phi13] have questioned the physical connection of tail and head. The lower $v_\mathrm{LSR}$ of the tail means that it could also be a background feature in the general ambient gas. In addition, the tail emission in image space does only roughly follow the orbital trace. Nevertheless, our unparalleled, ultra-deep integral field spectroscopy reveals that there is a fainter ’bridge’ of emission between head and tail following exactly the orbit in the position-velocity diagram. It is visible in figure \[fig3\]. This argues in favor of a physical connection.
Interactions of G2
==================
The fact that the tail does not perfectly match the orbit in image space might also point towards an additional effect: It is expected that the gas of G2 interacts with the hot gas of the accretion flow around Sgr A\*. This is actually one of the most exciting aspects of the G2 -Sgr A\* encounter.
Accretion flow models for Sgr A\* have been designed to explain its extremely low (radio) luminosity. These models predict that there is an atmosphere of hot, thin gas around Sgr A\* that extends out to roughly the Bondi radius ($\approx 10^5 R_S$). The density profile of the atmosphere depends on the model type. For a radiatively inefficient accretion flow one has $\rho(r) \sim r^{-1}$ ([@yua03 Yuan et al. 2003]), while for an advection dominated accretion flow $\rho(r) \sim r^{-3/2}$ ([@nar95 Narayan & Yi 1995]), and for a convection dominated accretion flow the profile can be as flat as $\rho(r) \sim r^{-1/2}$ ([@qua00 Quataert & Gruzinov 2000]).
The observational constraints on the density profile are rather weak: The diffuse X-ray emission around Sgr A\* as resolved by Chandra is due to the accretion flow ([@bag03 Baganoff et al. 2003], [@wan03 Wang et al. 2013]) and yields a constraint roughly at the Bondi radius. The rotation measure of Sgr A\* obtained in the submm domain on the other hand allows estimating the density in the innermost $10 R_S$ ([@mar07 Marrone et al. 2007]).
G2 now might offer the unique opportunity to probe the accretion flow on scales from $10^4 R_S$ to $10^3 R_S$. [@gil12] estimated that detectable X-ray radiation might occur during the pericenter approach due to a shock front developing ([@mck75 McKee & Cowie 1975]). The two main unknowns for that are the volume filling factor of G2, and the density profile of the ambient gas. Predicting the full evolution of G2, however, is a more complex problem, because the hydrodynamic time scales all are around a few years. In particular, the Kelvin-Helmholtz and Rayleigh-Taylor instabilities might be important. Hence, hydrodynamic simulations are needed, and have been performed ([@bur12 Burkert et al. 2012], [@sch12 Schartmann et al. 2012], [@ann12 Anninos et al. 2012], [@aba13 Abarca et al. 2013]).
The simulations consistently show that the evolution of G2 up to pericenter is very similar to what one expects from tidal shearing only. After pericenter passage, the tidal description breaks down and hydrodynamic effects start to dominate the evolution. The details depend strongly on the assumptions made about G2 and the ambient gas. Turning the argument around, the combination of observing the line profile evolution and possible radiative reactions can be used to constrain the assumptions put into the simulations, and thus to learn about the structure of the accretion flow around the prototypical low-luminosity AGN, Sgr A\*.
The nature of G2
================
The nature of G2 is currently being discussed. While it was called ’gas cloud’ in the discovery paper ([@gil12 Gillessen et al. 2012]), other papers have proposed that actually there resides a (faint and undetected) star at the origin of G2. It is interesting to review the stellar and the gas cloud model in the following.
The original proposal of G2 being a gas cloud had a serious shortcoming. The orbit of G2 lies in the plane of the clockwise stellar disk and the apocenter as calculated from the orbit presented in [@gil12] is at the inner edge of that disk at $r\approx 1''$. Hence, a connection is likely, for example as a collisional product of stellar winds in the region. The shortcoming is that if G2 formed there in pressure equilibrium and started falling inward then, it should have been already tidally stretched into an almost linear feature by the time we detected it. This is not what is seen in the 2004 - 2008 data, where G2 appears rather compact. On that orbit, G2 must have formed relatively recently, somewhere in the years 1990 - 2000, and thus in a radial range well inside the inner edge of the stellar disk.
The updated orbit now has a larger apocenter distance of $r\approx 2''$. That opens a new possibility: G2 could have formed half way in on the orbit and still would originate from the disk. Giving it a significant inward ’birth’ velocity (such as to place it on the observed orbit) solves the issue of too early tidal disruption. Hence, on that orbit G2 can have formed in the stellar disk and remained compact most of the way in. This lifts the most serious shortcoming of the pure gas cloud model.
[@shc13] introduces the idea of a magnetically arrested cloud, takes into account the explicit positions of individual stars as ionizing sources, and models the dust emission. The model can explain the absolute values of both the line luminosities and the L-band and M-band magnitudes, as well as the velocity dispersions. It probably currently is the best ’pure gas cloud’ model.
In the gas cloud scenario, G2 could have formed from collisions of stellar winds. Such clumps, with few Earth masses, have already been seen in simulations of the stellar winds around Sgr A\* ([@cua06 Cuadra et al. 2006]), although these particular simulations were performed with an SPH-code, which is not well suited for the problem at hand. Follow-up work seems to indicate that indeed at any moment in time there are a several ten clumps that resemble G2 within the simulated volume of $r\approx10''$. G2 might thus be special only in the sense that it is on a very radial orbit. L-band images of the GC region show dusty sources all over, but the detection of G2 as a gas cloud was only possible in the central arcsecond, since there $v_\mathrm{LSR}$ is large enough to Doppler shift the emission away from the general mini spiral emission.
It is also worth noting that one can pinpoint the most likely progenitor stars for that scenario: First, there is the eclipsing binary IRS16SW ([@mar06 Martins et al. 2006]), which consists of two $50 M_\odot$ stars, and secondly the massive, young star S91. Both objects are part of the clockwise stellar disk, and their position on the disk is consistent with being the origin of G2.
Soon after the discovery of G2, [@mur12] proposed a completely different model for G2: It might be the evaporating protoplanetary disk around a fainter star. Such a disk is not tidally stable at the position of G2, and hence it gets disrupted, the closer the object gets to the MBH. The gas of the disk is then ionized by the UV radiation field of the surrounding stars and the recombination lines can be observed. One could call such an object a ’tidal comet’. This model is attractive mainly for two reasons: It can naturally explain the compactness of G2. And a priori, the presence of stars is more likely close to Sgr A\*, since gas clouds are short-lived. The implications of that model are interesting: The debris around a star would flag the star, which itself could be too faint to be observable. And one would need to speculate about planet formation in the GC.
This model also has two difficulties. The high eccentricity means that one needs a rather strong kick for the star to change its orbit from one, which does not disrupt the protoplanetary disk at pericenter passage, to the current one. Yet, the kick itself should not destroy the protoplanetary disk either. The authors estimate how likely that is, and conclude that “we are somewhat fortunate to observe” G2. Secondly, the observed constancy of the line luminosities does not match what one would expect for that model. Coming closer to the MBH, more and more mass of the protoplanetary disk is lifted, leading to an increase in luminosity.
Also the model of [@sco13] places a star at the origin of G2. They propose that the relatively slow wind of a T-Tauri star would create a shock front, the emission of which we see as G2. For this model, a scattering event to the high eccentricity orbit is unproblematic. Also the constancy of the luminosity is easier to reconcile with this model. The authors investigate the various sources of ionization and recombination. The inner, cold shock dominates the emission and is collisionally ionised from the stellar wind. In this model, the luminosity thus depends on the wind parameters and is independent of the orbit.
The explicit hydrodynamic simulations of [@bal13] for a stellar wind source plunging through the accretion flow of Sgr A\*, however, show that in the radial range through which G2 was observed, an increase of the luminosity by a factor of a few would still be expected - which renders thus the T-Tauri star model less likely. Overall, the constancy of the line emission appears to be difficult to explain in any stellar scenario.
The observational work by [@phi13] favors a stellar origin of G2, based, however, not on their actual data, but expressing the prior that it should be more likely to find stars around Sgr A\* than gas clouds. Their highest quality OSIRIS data set from 2006 confirms that G2 is an extended gas cloud, and is the earliest data set resolving G2. Later OSIRIS observations are mainly SNR-limited. The paper also presents the best upper limit on a potential K-band source at the position of G2: mag$_{K'} > 20$.
This limit is at odds with the results presented in [@eck13]. These authors claim to find a K-band counterpart of G2 in NACO/VLT-based images, and presented during this conference similar claims from a Keck data set. Given that the two other obervational groups did not see any significant evidence for a K-band counterpart from de facto identical data sets, the findings of [@eck13] are unexpected.
But even if a K-band source can be identified, one cannot firmly conclude on the existence of a stellar source inside G2. As shown by [@gil12] and [@eck13], a dust temperature of $550\,$K matches the L-band and M-band fluxes, and yields a K-band flux that would be compatible with the putative detection. [@eck13] note that a slightly cooler dust temperature ($450\,$K) together with an embedded, faint star can also reproduce the photometry. In other words, a K-band detection is still ambiguous. Only if one detected G2 in H-band, one would be able to conclude that G2 contains a star. So far, only upper limits have been reported for H-band.
The currently observed, beautiful tidal evolution of the gas is only weakly dependent on whether one places a star inside of G2 or not. At the observed distance from Sgr A\*, the tidal force of the MBH dominates over the gravity of the supposed star inside of G2. Differences only occur when the embedded object replenishes material. Then the resulting density profile for G2 is more centrally concentrated compared to a pure gas cloud. This in turn, leads to a different velocity gradient evolution in the position-velocity diagram ([@gil13b Gillessen et al. 2013b]). Comparing with the observed evolution, a pure gas cloud model seems to be the better match currently, but the parameter space for possible stellar scenarios certainly has not yet been systematically checked.
More models have been proposed for G2: [@sch12] noted that the head-tail geometry of G2 could be explained by an (originally) ring-like geometry of G2. [@mey12] propose that a nova is at the origin of G2. Both ideas suffer from the fact that in order to match the observations, one needs to finetune the moment when G2 started to expand. [@mir12] suggested that the debris of a collision between a low-mass star and a stellar black hole could create G2. The debris could have been on the same orbit for hundreds of revolutions, avoiding the finetuning problem.
The further evolution of G2 probably will tell about its true nature. The different models might vary dramatically on what will happen to G2 during and after pericenter passage. A pure gas cloud model predicts a complete disruption of G2, while the stellar models predict that the gas cloud should reform after pericenter passage. Observations during and after pericenter passage will probably shed decisive light on the nature of G2.
The future of G2
================
While it is clear that the gas currently observed cannot survive as compact cloud the upcoming pericenter passage, it is less clear what observable consequences the fly-by might have. Quite a number of observing proposals[^1] have been focusing on the G2 event. So far, no observation of any G2 related radiation increase has been reported in any waveband. Nevertheless, it is interesting to look at the various ideas.
The first observable sign of G2 plunging through the accretion flow might be from the shock front that is expected to form. [@gil12] estimated the amount of X-ray radiation originating from the shock-heated gas. The temperature might reach up to $10^7\,$K, and the luminosity in the observable $2-8\,$keV band would exceed by a small factor the current quiescent level of emission from Sgr A\*.
Shock accelerations of electrons in the bow shock might lead to radio emission ([@nar12 Narayan, Özel & Sironi 2012], [@sad13a Sadowski et al. 2013a], [@cru13 Crumley & Kumar 2013]). In the $0.1\,$GHz to $1\,$GHz band, the emission can be much higher than the source intrinsic radio flux. [@sad13b] showed that the peak of the radio emission might occur even nine months before the nominal pericenter date. The VLA is conducting currently a public monitoring porgram of Sgr A\*[^2], and so far no flux increase has been reported (but see the multi-year radio light curve of Sgr A\* in [@bea13 Beaklini & Abraham 2013]).
Sgr A\* is expected to host a cusp of stellar mass black holes around it ([@mor93 Morris 1993]). [@bar13] study the possibility that the G2-cloud collides with such an object. They conclude that under favorable circumstances the event might be detectable. Close to pericenter passage, multiple encounters are likely to occur.
The rotation measure of Sgr A\*, a measure for the amount of material between the source and the observer, has been stable for many years ([@mar07 Marrone et al. 2007]). Additional gas coming from G2 could manifest itself thus in a change of the quantity. A change occurring in the near future would be very likely to be associated with gas of G2 arriving at Sgr A\*.
More speculative are the ideas, what happens if material enters the innermost accretion zone. It is clear, that this might take much longer than the actual fly-by, since the orbit only deposits material at around $2000\,R_S$, and significant radiation is produced in the central $10\,R_S$. The spiral down is probably dominated by the viscous time scale, such that the process might even take years ([@mos12 Moscibrodzka et al. 2012]). An increase in the accretion rate onto Sgr A\* would lead to increased emission across all wavebands ([@yua04 Yuan et al. 2004]). This is different from the mechanism believed to create the NIR and X-ray flares of Sgr A\* ([@gen03 Genzel et al. 2003], [@bag01 Baganoff et al. 2001]). They are due to local heating of synchrotron emitting electrons ([@dod09 Dodds-Eden et al. 2009]), rather than a global change in accretion rate. Yet, additional material in the accretion flow might change the flaring characteristics - flares could happen more often, last longer, or shine brighter than currently ([@dod11 Dodds-Eden et al. 2011]).
Even more speculative are ideas that Sgr A\* could change its overall accretion state. Currently, the accretion flow is optically thin and geometrically thick. At higher rates, it might change to an optically thick, geometrically thin state, i.e. develop a pronounced accretion disk. In addition, Sgr A\* might exhibit visible jets then. Such structural changes might be observable in two ways: While the resolved mm-VLBI measurements ([@doe08 Doeleman et al. 2008]) do not yet allow reconstructing an image of Sgr A\*, the observed visibilities could show structural changes. The other way to detect these might be astrometry. The position of Sgr A\* can be determined to $\approx 100\,\mu$as ([@rei08 Reid et al. 2008]). The expected change in effective photocenter position of Sgr A\* due to jets developing might exceed that number, and thus VLBI astrometry has the power of detecting such a change.
Maybe it is not even surprising to observe variations in the accretion rate of Sgr A\*. The work of [@cua06] showed that it is expected to vary significantly on longer time scales, and scaling to shorter time scales then would predict still some level of variation. There are also observational hints for it: From the ISM surrounding Sgr A\* one can detect X-ray reflection radiation ([@mun07 Muno et al. 2007], [@cla13 Clavel et al. 2013]). The emission moves outward, away from Sgr A\*, and is most likely a light echo. If so, there must have been a source that reached $10^{39}\,$erg/s a few hundred years ago in the GC, Sgr A\* being the most likely candidate.
Final remarks
=============
The discovery of G2 has triggered large interest, among observers and theorists, as well as from the general public. It might constitute the unique opportunity of observing in real-time a MBH being fed, a process which takes place throughout cosmic time and the universe. G2 might deliver unexpected insights into MBH growth or accretion flows around MBHs. Continued observing might be highly rewarding, and even non-detections can be very telling. We wish to encourage further observations - and reporting thereof. Be it detections or non-detections.
2013, *arXiv/1309.2313*
2012, *ApJ*, 759, 132
2001, *Nature*, 413, 45
2003, *ApJ*, 591, 891
2013, *ApJ*, 776, 13
2009, *ApJ*, 697, 1741
2013, *Phys. Rev. Lett.*, 110, 221102
2013 *MNRAS*, 428, 2731
2004, *Messenger*, 117, 17
2012, *ApJ*, 750, 58
2013, *A&A*, 558, 32
2013, *MNRAS*, 436, 1955
2006, *MNRAS*, 366, 358
2009, *ApJ*, 698, 676
2011, *ApJ*, 728, 37
2008, *Nature*, 455, 78
2013, *A&A*, 551, A18
2003, *Proc. SPIE*, 4841, 1548
2003, *Nature*, 425, 934
2009, *ApJ*, 707, L114
2012, *Nature*, 481, 51
2013a, *ApJ*, 763, 78
2013b, *ApJ*, 774, 44
2006, *Proc. SPIE*, 6269, 151
2009, *ApJ*, 690, 1463
2007, *ApJ*, 654, L57
2006, *ApJ*, 649, L103
1975, *ApJ*, 195, 715
2012, *A&A*, 546, L2
2012, *ApJ*, 756, 86
1993, *ApJ*, 408, 496
2012, *ApJ*, 752, L1
2007, *ApJ*, 656, 69
2012, Nature Communications 3, 1049
2012, *ApJ*, 757, L20
1995, *ApJ*, 452, 710
2006, *ApJ*, 643, 1011
2013, *ApJ*, 773, L13
2008, *ApJ*, 682, 1041
2000, *ApJ*, 539, 809
2013a, *MNRAS*, 432, 478
2013b, *MNRAS*, 433, 2165
2012, *ApJ*, 750, 58
2013, *ApJ*, 768, 108
2013, *arXiv*/1309.2282
2013, *Science*, 341, 981
2003, *ApJ*, 598, 301
2004, *ApJ*, 606, 894
[^1]: For an overview of 2013 and 2014 G2 related observing proposals, see: `https://wiki.mpe.mpg.de/gascloud/ProposalList`. For additions to the list, please contact the authors.
[^2]: For details see: `https://science.nrao.edu/enews/5.10/index.shtml#g2_encounter`
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- Lev Glebsky and Luis Manuel Rivera
title: 'On low rank perturbations of complex matrices and some discrete metric spaces.'
---
Introduction
============
The article is devoted to different aspects of the question: “What can be done with a complex-valued matrix by a low rank perturbation?”[^1]
From the works of Thompson [@thompson] we know how the Jordan normal form can be changed by a rank $k$ perturbation, see Theorem \[th3\]. Particulary, it follows that one can do everything with a geometrically simple spectrum by a rank $1$ perturbation, see Corollary \[th1\]. But the situation is quite different if one restricts oneself to normal matrices, see Theorem \[th4\] and Corollary \[corol\_Norm1\]. We think that Corollary \[corol\_Norm1\] may be considered as a finite dimension analogue of the continuous spectrum conservation under compact perturbations in Hilbert spaces. For unitary and self-adjoint matrices the inequality of Corollary \[corol\_Norm1\] is the only restrictions on “what can be done with a spectrum by a rank $k$ perturbation”, see Theorem \[Th\_un\_ad\]. We don’t know if there is an analogue of Theorem \[Th\_un\_ad\] for normal matrices. It is worth to mention that Corollary \[corol\_Norm1\] for self-adjoint matrices follows from Cauchy interlacing theorem [@Cauchy]. Theorem \[Th\_un\_ad\] is related with the converse Cauchy interlacing theorem [@Fan].
The spectrum of $H_1+H_2$ with known spectra of self-adjoint matrices $H_1$ and $H_2$ is studied a lot, see [@Klyachko] and the bibliography therein. Although the complete set of restrictions on the spectrum $H_1+H_2$ known in this situation, we are not sure that there is an easy proof of Theorem \[Th\_un\_ad\] using results of [@Klyachko].
Although Theorem \[th3\] should be known (see, for example, [@SSV2], where Theorem \[th3\] formulated in one direction), we will give a proof here, manly because our proof falls in a general framework , which is also used in the proof of Theorem \[Th\_un\_ad\]. Let us describe the framework. The set $\C_{n\times n}$ of all complex $n\times n$-matrices (set of self-adjoint matrices) we equip with the arithmetic distance, $d(A,B)=\rank(A-B)$ (see [@Che]). The arithmetic distance is geodesic for these cases. The spectral properties of matrices, such as Weyr characteristics and spectra (multiset) also may be considered as a metric spaces with distance, related to the arithmetic distance on matrices, see Section \[sec\_dist\]. These distances also turn out to be geodesic. Then we prove Theorem \[th3\] (Theorem \[Th\_un\_ad\]) for $\rank(A-B)=1$ and the general results will follow from Proposition \[prop\_geod\].
\[prop\_geod\] Let $X$ and $Y$ be geodesic metric spaces, let $O^X_n(x)$ denote the closed ball of radius $n$ around $x$ in $X$, let $\phi:X\to Y$ be such that $\phi(O^X_1(x))=O^Y_1(\phi(x))$ for all $x\in X$. Then $\phi(O^X_n(x))=O^Y_n(\phi(x))$ for any $n\in {{\bf N}}$ and $x\in X$.
The proof is by induction. For $n=1$ there is nothing to prove. Step $n\to n+1$: It follows that $O^X_{n+1}(x)=\bigcup\limits_{z\in O^X_{n}(x)}O^X_1(z)$ ($X$ is geodesic), then $$\phi(O^X_{n+1}(x))=\bigcup\limits_{z\in O^X_{n}(x)}\phi(O^X_1(z))=
\bigcup\limits_{z\in \phi(O^X_{n}(x))}O^Y_1(z)=\bigcup\limits_{z\in O^Y_{n}(\phi(x))}O^Y_1(z)=
O^Y_{n+1}(\phi(x)).$$
In Section \[sec\_almost\] and Section \[sec\_com\] we use the normalized arithmetic distance $d_r(A,B)=\frac{\rank(A-B)}{n}$, where $n$ is the size of the matrices. We are interested in the following questions: “ Suppose that matrices almost satisfy some equations (in the sense of $d_r(\cdot,\cdot)$). If close to that matrices there exist matrices satisfying the equations (uniformly with respect to $n$)?” We manage to answer only the following: close to an almost unitary (self-adjoint) matrix there exists an unitary (self-adjoint) matrix. We do not know if the same is true for normal matrices. (This question has the affirmative answer for norm distance $dn(A,B)=\|A-B\|$, see [@Lin]. It is equivalent to the following: "close to any pair of almost commuting self-adjoint matrices there exists a pair of commuting self-adjoint matrices (with respect to the distance $dn(\cdot,\cdot)$). It is interesting that there are almost commuting (with respect to $dn(\cdot,\cdot)$) matrices, close to which there are no commuting matrices, [@Choi; @RuyTerry; @Dan]). The similar question have been studied for operators in Hilbert spaces (Calkin algebras, [@Farah]). In Hilbert spaces the operator $a$ is called to be essentially normal iff $aa^*-a^*a$ is a compact operator. In contrast with Theorem \[th6\], there exists an essentially unitary operator which is not a compact perturbation of an unitary operator (just infinite 0-Jordan cell). There is a complete characterization of compact perturbations of normal operators, see [@Farah] and the bibliography therein. Let us return to almost commuting matrices with respect normalized arithmetic distance $d_r$. In Section \[sec\_com\] we show that for any $A\in \C_{n\times n}$ with simple spectrum there exists an almost commuting matrix, which is far from each commuting with $A$ matrix. The similar problem for the pairs of almost commuting matrices, as far as we know, is open. Precisely, if for any $\epsilon>0$ there exists $\delta>0$ such that for any $A,B$, $d_r(AB,BA)<\epsilon$ there exists $(\tilde A,\tilde B)$ satisfying $\tilde A\tilde B=\tilde B\tilde A$ and $d_r(\tilde A,A),d_r(\tilde B,B)<\delta$ ($\delta$ does not depends on size of the matrices).
We think that low rank perturbations of matrices may related to sofic groups. The following question seems to be interesting from this point of view (although it seems to goes beyond the scope of the present article). One can show that all solutions of equation $C^{-1}A^{-1}CAC^{-1}AC=A^2$ in finite unitary matrices are trivial in $A$ ($A=E$). On the other hand, it is true that for any $\epsilon
>0$ there exist $A,C$, $d(A,E)=d(C,E)=1$ and $d(C^{-1}A^{-1}CAC^{-1}AC,A^2)<\epsilon$. If the above assertion is true with additional requirements $C^4=1$? If not, it gives an example of non-sofic group.
Note. All linear spaces are supposed to be finite dimensional in the rest of the article. $\C_{n\times n}$ will denote the set of all complex $n\times n$-matrices, ${{\bf N}}=\{0,1,2,...\}$.
Some discrete geodesic spaces. {#sec_dist}
==============================
Arithmetic distance on $C_{n\times n}$
--------------------------------------
\[lm\_arith\] The arithmetics distance $\rank(A-B)$ is geodesic on
- Set of all $n\times n$ matrices.
- Set of all self-adjoint $n\times n$ matrices.
- Set of all unitary $n\times n$ matrices.
It is clear that a rank $k$ matrix (self-adjoint matrix) may be represented as sum of $k$ matrices (self-adjoint matrices) of rank $1$. The first two items follow from the fact that set of matrices (self-adjoint matrices) is closed with respect to summation. For unitary matrices. Let $\rank(U_1-U_2)=k$, or, the same, $\rank(E-U_1^{-1}U_2)=k$. It means that, in a proper basis, $U_1^{-1}U_2=\diag(\lambda_1,\lambda_2,...,\lambda_k,1,1,...,1)$. Now the sequence $U_1,\;U_1\cdot\diag(\lambda_1,1,1,...,1),\;U_1\cdot
\diag(\lambda_1,\lambda_2,1,1,...,1)...
U_1\cdot\diag(\lambda_1,\lambda_2,...,\lambda_k,1,1,...,1)=U_2$ give us the geodesic needed.
The methods used in the above proof are not applied for normal matrices – the set of normal matrices is not closed with respect neither summation nor multiplication. In fact, an example from [@Fan] hints that arithmetic distance might be non geodesic on the set of normal matrices.
\[prop\_mob1\] Let $\phi(x)=(ax+b)^{-1}(cx+d)$ be a Möbius transformation of $\C_{n\times n}$, defined on $A$, $B$. Then $\rank(A-B)=\rank(\phi(A)-\phi(B))$.
A Möbius transformation is a composition of linear transformations $A\to aA+b$ ($a,b\in \C$) and taking inverse $A\to A^{-1}$. Those transformations (if defined) clearly conserve arithmetic distance, for example, $\rank(A^{-1}-B^{-1})=\rank(A^{-1}(B-A)B^{-1})=\rank(A-B)$. ($A^{-1}$ and $B^{-1}$ is of full rank.)
Distance on the spaces of the Weyr characteristics.
---------------------------------------------------
Having in mind the Weyr characteristics of complex matrices (see below), we introduce the spaces $\Im_n$ of the Weyr characteristics. Where $\Im_n$ is the space of functions $\Z^+\times\C\to {{\bf N}},\; (i,\la)\to\eta_i(\la)$ such that
- $\eta_i(\la)\neq 0$ for finitely many $(i,\la)$ only, and $\sum\limits_{\la\in\C}\sum\limits_{i\in{{\bf N}}}\eta_i(\la)=n$.
- $\eta_i(\la)\geq \eta_{i+1}(\la)$.
On $\Im_n$ define a metric $d(\eta,\mu)=\max\limits_{(i,\la)}\{|\eta_i(\la)-\mu_i(\la)|\}$. First of all let us note that $d(\cdot,\cdot)$ is indeed a metric. Trivially, $d(\eta,\mu)=0$ implies $\eta=\mu$ and $d(\cdot,\cdot)$ satisfies triangle inequality as supremum (maximum) of semimetrics. It is clear, that $d(\mu,\nu)$ is also well defined for $\mu$ and $\nu$ in different spaces of Weyr characteristics (for different $n$). We will need the following
\[prop\_weyr\] Let $\mu\in\Im_m$ and $n>m$. Then there exists $\nu\in\Im_n$ such that for any $\eta\in\Im_n$, the inequality $d(\mu,\eta)\geq
d(\nu,\eta)$ holds.
We can do as follows. Let $\mu_i(\lambda_0)\neq 0$ and $\mu_{i+1}(\lambda_0)=0$. We can take $\nu_{i+1}(\lambda_0)=\nu_{i+2}(\lambda_0)=...=\nu_{i+n-m}(\lambda_0)=1$ and $\mu_j(\lambda)=\nu_j(\lambda)$ for all other pairs $(j,\lambda)$.
$\Im_n$ are geodesic metric spaces.
Let $\eta,\mu\in \Im_n$ and $d(\eta,\mu)=k>1$ it is enough to find $\nu\in\Im_n$ such that either $d(\eta,\nu)=1$ and $d(\nu,\mu)=k-1$, or $d(\eta,\nu)=k-1$ and $d(\nu,\mu)=1$, moreover, by Proposition \[prop\_weyr\] it is enough to find $\nu\in\Im_m$ for $m\leq n$. Let $S_+=\{(j,\lambda)\in
\Z^+\times\C\;|\;\eta_j(\lambda)-\mu_j(\lambda)=k\}$ and $S_-=\{(j,\lambda)\in
\Z^+\times\C\;|\;\eta_j(\lambda)-\mu_j(\lambda)=-k\}$. Suppose, that $|S_+|\geq |S_-|$ (if not, we can change $\eta\leftrightarrow\mu$). Now let $$\nu_i(\lambda)=\left\{\begin{array}{lll}
\eta_i(\lambda)-1 & \mbox{if} & (i,\lambda)\in S_+\\
\eta_i(\lambda)+1 & \mbox{if} & (i,\lambda)\in S_-\\
\eta_i(\lambda) & \mbox{if} & (i,\lambda)\not\in S_+\cup S_-
\end{array}\right.$$ We have to show that $\nu\in\Im_m$ for $m=n-|S_+|+|S_-|$. It is enough to show that $\nu_{j+1}(\lambda)\leq \nu_j(\lambda)$. Suppose contrary, that $\nu_{j+1}(\lambda)> \nu_j(\lambda)$. There are three possibility:
a)
: $\eta_{j+1}(\lambda)=\eta_j(\lambda)$, $(j,\lambda)\in S_+$ and $(j+1,\lambda)\not\in S_+$, but then $k>\eta_{j+1}(\lambda)-\mu_{j+1}(\lambda)\geq \eta_{j+1}(\lambda)-\mu_j(\lambda)=
\eta_j(\lambda)-\mu_j(\lambda)=k$, a contradiction.
b)
: $\eta_{j+1}(\lambda)=\eta_j(\lambda)$, $(j+1,\lambda)\in S_-$ and $(j,\lambda)\not\in S_-$, but then $-k<\eta_{j}(\lambda)-\mu_{j}(\lambda)\leq
\eta_{j}(\lambda)-\mu_{j+1}(\lambda)=
\eta_{j+1}(\lambda)-\mu_{j+1}(\lambda)=-k$, a contradiction.
c)
: $\eta_{j+1}(\lambda)=\eta_j(\lambda)-1$, $(j,\lambda)\in S_+$ and $(j+1,\lambda)\in S_-$, but then $-k=\eta_{j+1}(\lambda)-\mu_{j+1}(\lambda)\geq \eta_{j+1}(\lambda)-\mu_j(\lambda)=
\eta_j(\lambda)-\mu_j(\lambda)-1=k-1$, so $-k\geq k-1$ and $1/2\geq k$, a contradiction with $k>1$.
Now, by construction, $d(\nu,\eta)=1$ and $d(\nu,\mu)=k-1$.
Distances $\dc$ and $\tilde\dc$ on finite multisets of the complex numbers.
---------------------------------------------------------------------------
Multisets and operations. The language of multisets is very convenient to deal with spectrums. We will need only finite multisets. For a multiset $\A$ let $\set(\A)$ denote the set of elements of $\A$, forgetting multiplicity. It is clear that a muiltiset maybe considered as the multiplicity function $\chi_\A:\set(A)\to{{\bf N}}$, for any $x\not\in\A$ we will suppose $\chi_\A(x)=0$. (For all cases, considered here, $\set(\A)\subset\C$, so we can consider $\chi_\A:\C\to{{\bf N}}=\{0,1,2...\}$.) As far as the authors aware, there are several generalizations of the set-theoretical operations to multisets. We will need the following operations:
- Difference of two multiset $\A\setminus\B$, $\chi_{\A\setminus\B}(x)=\max\{0,\chi(\A)-\chi(\B))\}$.
- Intersection $\A\cap X$ of a set $X$ and a multiset $\A$, $$\chi_{\A\cap X}(x)=\left\{\begin{array}{lll}
\chi_\A(x),&\mbox{if} & x\in X \\
0,&\mbox{if} & x\not\in X
\end{array}\right.$$
- Union $\A\uplus\B$, $\chi_{\A\uplus\B}(x)=\chi_\A(x)+\chi_\B(x)$.
Let $S(a,r)=\{x\in\C\;: |x-a|\leq r\}$ and $\S=\{S(a,r)\;|\;a\in\C,\;r\in \R^+\}$ be the set of circles. For $\A,\B\subset_M\C$ let $\dc(\A,\B)=\max\limits_{S\in\S}\{||\A\cap
S|-|\B\cap S||\}$. Let as extend $\S$ to $\tilde \S$ which include interior of complements of circles and semiplains: $\tilde S=S\cup
\{\{x\in\C\;:\;|x-a|\geq r\}\;|\;a\in\C,\;r\in\R^+\}\cup \{
\{x\in\C\;:\;\Imp(\frac{x-b}{a})\geq 0\}\;|\;a,b\in\C\}$; and introduce new metric $\tilde\dc$: $\tilde\dc(\A,\B)=\max\limits_{S\in\tilde\S}\{||\A\cap S|-|\B\cap
S||\}$.
- $\dc$ and $\tilde\dc$ are metrics on the finite multisubsets of $\C$.
- $\dc(\A,\B)=\dc(\A\setminus\B,\B\setminus\A)$, $\tilde\dc(\A,\B)=\tilde\dc(\A\setminus\B,\B\setminus\A)$.
- If $|\A|=|\B|$, then $\tilde\dc(\A,\B)=\dc(\A,\B)$.
<!-- -->
- The same as for the spaces of Weyr characteristics.
- Let $\sum_S(\A,\B)=|\A\cap S|-|\B\cap S|=
\sum\limits_{x\in S}(\chi_\A(x)-\chi_\B(x))$. Then $\sum_S(\A\setminus\B,\B\setminus\A)=
\sum\limits_{x\in S}(\max\{0,\chi_\A(x)-\chi_\B(x)\}-\max\{0,\chi_\B(x)-\chi_\A(x)\})=
\sum\limits_{x\in S}(\chi_\A(x)-\chi_\B(x))$. Now the item follows by definition of $\dc$ ($\tilde\dc$).
- First of all, due to $\A$ and $\B$ are finite multisets, for any semiplain $p$ we can find a circle $c$ such that $\sum_p(\A,\B)= \sum_c(\A,\B)$. Also for any closed circle $c_c$ there exists an open circle $c_o$ such that $\sum_{c_c}(\A,\B)= \sum_{c_o}(\A,\B)$. Now, under assumption of the item $\sum_S(\A,\B)=-\sum_{\C\setminus S}(\A,\B)$ and the result follows.
\[prop\_mob\] Let $\phi(x)=\frac{ax+b}{cx+d}$ be a Möbius transformation of $\C$, defined on $\set(\A)\cup\set(\B)$. Then $\tilde\dc(\A,\B)=\tilde\dc(\phi(\A),\phi(\B))$.
A Möbius transformation defines a bijection of $\tilde\S$.
We don’t know if the metric $\dc$ is geodesic on the multisets with fixed cardinality, but its restriction on any circle o line is:
\[lm\_geod1\] Let $l\subset\C$ be a circumference or a straight line. Let $\A,
\B\subset_M l$, $|\A|=|\B|=n$ and $\tilde\dc(\A,\B)=k\geq 2$. Then there exists ${\cal C}\subset_M l$, $|{\cal C}|=n$ such that $\tilde\dc(\A,{\cal C})=1$ and $\tilde\dc({\cal C},\B)=k-1$.
By Proposition \[prop\_mob\], it is enough to proof it for the unit circle. Let us start with the case when $\set(\A)\cap\set(\B)=\emptyset$. Let $\Gamma=\set(\A)\cup\set(\B)\subset C^1$. Let $|\Gamma|=r$. We will cyclically anticlockwise order $\Gamma=\{\gamma_0,\gamma_1,...,\gamma_{r-1}\}$ by elements of $Z_r$. To construct $\cal C$ we move each element of $\A$ to the next element in $\Gamma$, precisely, $\set({\cal C})\subseteq \Gamma$ and $$\chi_{\cal C} (\gamma_i)=\max\{0,\chi_\A(\gamma_i)-1\}+\chi_{\set(\A)}(\gamma_{i-1}),$$ the other words $$\chi_{\cal C} (\gamma_i)=\left\{\begin{array}{lll}
\chi_\A(\gamma_i)-1 &\mbox{if} &
\gamma_i\in\set(\A)\;\mbox{and}\;\gamma_{i-1}\not\in\set(\A) \\
1 &\mbox{if} & \gamma_i\not\in\set(\A)\;\mbox{and}\; \gamma_{i-1}\in\set(\A) \\
\chi_\A(\gamma_i) & \mbox{for} & \mbox{the other cases}
\end{array}\right.$$ We will check that $\cal C$ satisfies our needs. For $x,y\in \Gamma$ let $[x,y]$ denote the closed segment of $C^1$, starting from $x$ and going anticlockwise to $y$ (so $[x,y]\cup [y,x]=C^1$). It is clear that for $X,Y\subset_M\Gamma$ one has $\tilde\dc(X,Y)=\max\{||X\cap [\alpha,\beta]|-|Y\cap [\alpha,\beta]||\;:\;\alpha,\beta\in\set(X)\cup\set(Y)\}$. Denote by $\sum_{[\alpha,\beta]}(X,Y)=|X\cap [\alpha,\beta]|-|Y\cap [\alpha,\beta]|$.
Now, $\tilde\dc(\A,{\cal C})=\tilde\dc(\A\setminus{\cal C},{\cal C}\setminus\A)=1$, for $\A\setminus{\cal C}$ and ${\cal C}\setminus\A$ are interlacing sets on $\C^1$. Suppose further, $\tilde\dc({\cal C},\B)=\tilde\dc({\cal C}\setminus\B,\B\setminus{\cal C})\geq k$, then there exists $[\gamma_i,\gamma_j]$ such that either
1. $\sum_{[\gamma_i,\gamma_j]}({\cal C}\setminus\B,\B\setminus{\cal C})
=\tilde\dc({\cal C}\setminus\B,\B\setminus{\cal C})\geq k$\
or
2. $\sum_{[\gamma_i,\gamma_j]}({\cal C}\setminus\B,\B\setminus{\cal C})\leq -k$.
In the first case we may assume that $\gamma_i,\gamma_j\in{\cal C}\setminus\B$ and $\gamma_{i-1},\gamma_{j+1}\not\in{\cal C}\setminus\B$. Now, changing interval if necessary ($i^n=i-1$ and (or) $j^n=j-1$) we may, keeping $\sum_{[\gamma_i,\gamma_j]}({\cal C}\setminus\B,\B\setminus{\cal C})$, achieve that $\gamma_i,\gamma_j\in\A$ and $\gamma_{i-1},\gamma_{j+1}\not\in\A$. Then $\sum_{[\gamma_i,\gamma_j]}(\A,\B)=
\sum_{[\gamma_i,\gamma_j]}({\cal C}\setminus\B,\B\setminus{\cal C})+1\geq k+1$, a contradiction. The second case may be considered similarly.
If $\set(\A)\cap\set(\B)\neq\emptyset$ then we can find ${\cal C'}$ for $\A\setminus\B$ and $\B\setminus\A$ and then take ${\cal C}={\cal C'}\uplus X$, where $X=\A\setminus(\A\setminus\B)=\B\setminus(\B\setminus\A)$.
On spectrum of low rank perturbations.
======================================
\[tt\] Let $n \times n$ matrix $A$ over a field $F$ have similarity invariants $h_n(A) \mid h_{n-1}(A) \mid \ldots \mid h_1(A)$. Then: as column $n$ tuple $x$ and row $n$-tuple $y$ range over all vectors entries in $F$, the similarity invariants assumed by the matrix $$B=A+xy$$ are precisely the monic polynomials $h_n(B) \mid \ldots
\mid h_1(B)$ over $F$ for which $degree(h_1(B) \cdots h_n(B))=n$ and $$\begin{array}{l}
h_n(B) \mid h_{n-1}(A) \mid h_{n-2}(B) \mid h_{n-3}(A) \mid \ldots,\\
h_n(A) \mid h_{n-1}(B) \mid h_{n-2}(A) \mid h_{n-3}(B) \mid \ldots.
\end{array}$$
We are going to reformulate Theorem \[tt\] for the field $\C$ using Weyr characteristic.
Let $\eta_m (A, \lambda)$ denote the number of $\lambda$-Jordan blocks in $A$ of size greater or equal to $m$ ($m \in {{\bf N}}$). $$\eta_m(A,\la)=\mbox{dim}\hspace{0.2cm} Ker(\la E - A)^m-\mbox{dim}\hspace{0.2cm} Ker(\la E - A)^{m-1}$$ This sequence of numbers $\eta_1(A,\la), \ldots \eta_q(A,\la)$ is called the Weyr characteristic for the eigenvalue $\la$ of matrix $A$, see [@weyr].
\[th3\] Let $A\in \C_{n\times n}$ with Weyr invariants $\eta_m (A, \la)$. Then as $R$ ranges over all $n\times n$ complex matrices of rank less o equal $k$, the Weyr invariants assumed by the matrix $B=A+R$ are precisely those, that satisfy both of the following conditions:
- For any $\la\in\C$ and any $m\in{{\bf N}}$ $$\mid \eta_m (A, \la)-\eta_m (B, \la) \mid \leq k.$$
- $\sum\limits_{\la\in\C}\sum\limits_{m\in{{\bf N}}}\eta_m(B,\la)=n$.
It is enough to prove the theorem for $k=1$. Indeed, assume that the theorem is valid for $k=1$. Then we may consider the Weyr characteristics as a map $\psi:\C_{n\times n}\to \Im_n$, which satisfies Proposition \[prop\_geod\]. So, Theorem \[th3\] follows, though it states that $\psi(O_k(A))=O_k(\psi(A))$.
Let us prove Theorem \[th3\] for $k=1$. For a given eigenvalue $\la \in sp(A)$ the sequence of numbers $$q_1(A, \la) \geq q_2(A, \la) \geq ...,$$ corresponding to the sizes of the $\la$-Jordan blocks in the Jordan normal form of $A$ are know as the Segre characteristics of $A$ relative to $\la$, [@weyr].
The similarity invariant factors of $A \in \C_{n \times n}$ are sequence of monic polynomials in $x$, $h_n(A) \mid h_{n-1}(A) \mid h_{n-2}(A) \ldots \mid
h_{1}(A)$. It is known that $h_i(A)=\prod_{\lambda} (\lambda -x)^{q_i(A,
\la)}$, where $\la \in sp(A)$ and $q_i(A, \la)$ is a Segre characteristic corresponding to $\la$. So, if $\rank(A-B)=1$, by Thompson’s theorem \[tt\] one has $$\label{ineq1}
\begin{array}{l}
q_1(B, \la) \geq q_2(A, \la) \geq q_3(B, \la) \geq \ldots ,\\
q_1(A, \la) \geq q_2(B, \la) \geq q_3(A, \la) \geq \ldots .
\end{array}$$
As Weyr characteristic is the conjugate partition of Segre characteristic, we can use Ferrers diagram to compute $\eta_m(B, \la)$ (see, [@weyr]) as the number of points of column $m$ in Ferrer diagram of the Segre characteristic of $B$ relatively to $\la$ (by short the Ferrer diagram of $(B, \la)$). Precisely, the Ferrer diagram for $q(B,\la)$ is the set $F^\la_B=\{(i,j)\in\Z^+\times\Z^+\;|\;j\leq q_i(B,\la)\}$, see Figure \[figferrer\] (the numbering from top to bottom and left to right). The Weyr characteristics is related with Ferrers diagram by the formula $\eta_j(B,\la)=|\{(x,y)\in F^\la_B\;|\;x=j\}|$. From inequalities (\[ineq1\]) we have that $q_i(B, \la) \geq q_{i+1}(A, \la)$. This inequality is equivalent to the statement $\forall i\neq 1\; (i,j)\in F^\la_A \to (i-1,j)\in F^\la_B$ (Figure \[figferrer\]), which is equivalent to the fact that $\eta_j(B, \la) \geq \eta_j(A, \la)-1$. In similar form, from inequalities (\[ineq1\]) we can observe that $q_i(A, \la) \geq q_{i+1}(B, \la)$ with $1 \leq i \leq n-1$. Therefore, $\eta_j(A, \la) \geq \eta_j(B, \la)-1$, and the theorem follows.
$$\begin{array}{cccccccccccccccc}
\empty & \eta_1(B) & \eta_2 & \eta_3 & \eta_4 & \cdots & \eta_k & & &\eta_1(A) & \eta_2& \eta_3 & \eta_4 & \cdots & \eta_k \\
q_1(B, \la) & \bullet & \bullet & \bullet & \bullet & \cdots & \bullet &\hspace{1.cm} & q_2(A, \la ) & \bullet & \bullet & \bullet & \bullet & \cdots & \bullet\\
q_2(B, \la) & \bullet & \bullet & \bullet & \cdots & \bullet &
\hspace{1cm} & & q_3(A, \la ) & \bullet & \bullet & \bullet &\cdots & \bullet\\
q_3(B ,\la) & \bullet & \bullet & \cdots & \bullet & &
\hspace{1cm} & & q_4(A, \la ) & \bullet & \bullet & \cdots & \bullet\\
\vdots & & & & & & & & \vdots\\
q_{n-1}(B, \la) & \bullet & \cdots & \bullet & & &
\hspace{1cm} & & q_n(A, \la ) & \bullet & \cdots & \bullet\\
q_{n}(B, \la) & \bullet
\end{array}$$
\[th1\] If the geometric multiplicity of any eigenvalue $\la$ of $A$ (number of $\la$-Jordan cells)is $1$, then for any multiset $M$ of size $n$ there is a rank $1$ matrix $B$ such that $\spectr(A+B)=M$.
Case of normal matrices.
========================
We will say that the vector $x$ is an $\alpha$-eigenvector if $Ax=\alpha
x$. We will denote by $\RS(A,\lambda, \epsilon)$ the space generated by all the $\alpha$-eigenvectors of $A$ with $|\lambda-\alpha|\leq\epsilon$. For the case of normal matrices, the following theorem shows that the difference between the dimension of $\RS (A,\la,\e)$ and the dimension $\RS (B,\la,\e)$ is bounded by the rank of the difference matrix $A-B$.
\[th4\] If $A$ and $B$ are normal matrices, then for any $\la$, and for any $\e \geq 0$, $$\mid dim( \RS (A,\la,\e)) - dim ( \RS (B,\la,\e)) \mid \leq \rank(A-B)$$
Let $X^\perp$ be the orthogonal complement of subspace $X$ and $P_X$ be an orthogonal projection on $X$.
\[lm\_ort\] Let $N:L\to L$ be a normal operator, and $X$ be a subspace of $L$ such that $\|(N-\la)x\|\leq\e\|x\|$ for any $x\in X$, then (we will write $\RS (\la,\e)$ for $\RS (N,\la,\e)$)
1. $P_{\RS(\la,\e)}x\neq 0$ for any $x\in X$, $x\neq 0$.
2. $\|P_{\RS(\la,a\e)}x\|\geq \sqrt{1-\frac{1}{a^2}}\|x\|$ for any $x\in X$.
3. $\dim(R(\la,\e))\geq\dim(X)$
It is clear that (1) implies (3).
\(1) Let $e_1,e_2,...,e_n$ be a diagonal orthonormal basis for $N$, and $\la_1,\la_2,...\la_n$ corresponding eigenvalues ($Ne_i=\la_ie_i$). Let $x=\alpha_1e_1+\alpha_2e_2+...+\alpha_ne_n\in
X$ and $\|x\|=\sum\limits_{i=1}^n|\alpha_i|^2=1$. Let $x=x_1+x_2$ where $x_1\in \RS(\la,\e)$ and $x_2\in\RS^\perp(\la,\e)$. Now, $\|(N-\la)x\|^2=\sum\limits_{i=1}^n|\alpha_i|^2|\la_i-\la|^2\leq\e^2$ implies that $\sum\limits_{i\;|\;|\la_i-\la|>\e}|\alpha_i|^2<1$. So, $P_{\RS(\la,\e)}(x)=x_1=\sum\limits_{i\;|\;|\la_i-\la|\leq\e}\alpha_ie_i\neq
0$.
\(2) Similarly, $$\sum\limits_{i\;|\;|\la_i-\la|>a\e}|\alpha_i|^2<\frac{1}{a^2}\;\;\mbox{and}\;\;
\sum\limits_{i\;|\;|\la_i-\la|\leq a\e}|\alpha_i|^2\geq
(1-\frac{1}{a^2}),$$ so, $\|P_{\RS(\la,a\e)}x\|\geq \sqrt{1-\frac{1}{a^2}}\|x\|$
Now we a ready to prove Theorem \[th4\]. Let $\rank(A-B)=r$ then there exists $X= \RS(A,\la,\e)\cap\ker(A-B)$ with $\dim(X)\geq \dim(\RS(A,\la,\e))-r$ and $A|_X=B|_X$, so $\|(B-\la)|_X\|\leq\e$ and, by Lemma \[lm\_ort\], $\dim(\RS(B,\la,\e))\geq \dim(\RS(A,\la,\e))-r$. By symmetry, we get Theorem \[th4\].
Theorem \[th4\] implies that any circle in complex plain, containing $m$ spectral points of $A$ ($B$) should contain at least $m-k$ spectral points of $B$ ($A$). So, we have
\[corol\_Norm1\] If $A$ and $B$ are normal matrices then $\dc(\spectr(A),\spectr(B))\leq \rank(A-B)$.
It is just a reformulation of Theorem \[th4\].
If the condition of Corollary \[corol\_Norm1\] describes all accessible by rank $k$ perturbation spectra? We are going to show that the answer is “yes” for self-adjoint and unitary matrices.
\[Th\_un\_ad\] Let $A$ be a self-adjoint (unitary) $n\times n$-matrix. Let $\B\subset_M\R$ ($\B\subset_M C^1$), $|\B|=n$. Then there exists self-adjoint (unitary) matrix $B$ such that $\spectr(B)=\B$ and $\rank(A-B)=\dc(\spectr(A),\B)$.
In fact the following, more general result is valid:
\[Th\_un\_ad2\] Let $\l\subset\C$ be a circumference or straight line. Let $A$ be a normal $n\times n$-matrix, $\spectr(A)\subset_M\l$. Let $\B\subset_M\l$, $|\B|=n$. Then exists a normal matrix $B$ such that $\spectr(B)=\B$ and $\rank(A-B)=\dc(\spectr(A),\B)$.
- It is enough to prove Theorem \[Th\_un\_ad2\] for self adjoin matrices. Indeed, let $\spectr(A),\B\subset_M\l$, $|\B|=n$ for a circle (line) $l\subset\C$. Then there exists a Möbius transformation $\phi$, defined on $\spectr(A)\cup\B$, which map $l$ to the real line. Then $\phi(A)$ is a self-adjoint matrix and we can apply Theorem \[Th\_un\_ad\] to $\phi(A)$ and $\phi(\B)$ to find $\tilde B$ with $\spectr(\tilde B)=\phi(\B)$ and $\rank(\phi(A)-\tilde B)=\tilde\dc(\phi(\spectr(A)),\phi(\B))$. Now take $B=\phi^{-1}(\tilde B)$ and results follows, for the Möbius transformations conserve arithmetic distance on $C_{n\times n}$ and the distance $\tilde\dc$ on multisets (Proposition \[prop\_mob1\] and Proposition \[prop\_mob\]).
- It is enough to prove Theorem \[Th\_un\_ad2\] for $\dc(\spectr(A),\B)=1$ and the rest will follow from Proposition \[prop\_geod\], Proposition \[lm\_geod1\] and Lemma \[lm\_arith\].
- Also w.l.g. we may assume that $\set(\spectr(A))\cap\set(\B)=\emptyset$. For if $X=\spectr(A)\setminus(\spectr(A)\setminus\B)$ we can write $A=A_1\oplus A_2$ with $\spectr(A_1)=X$ and $\spectr(A_2)=\spectr(A)\setminus X$. We can find $B_2$ with $\spectr(B_2)=\B\setminus X$ and $\rank(A_2-B_2)=1$. Now, take $B=A_1\oplus B_2$.
- Let $\A,\B\subset_M\R$, $\set(\A)\cap\set(\B)=\emptyset$, $|\A|=|\B|$ and $\dc(\A,\B)=1$. Then, in fact, $\A$ and $\B$ are interlacing sets. It means that if $\A=\{\alpha_1,\alpha_2,...,\alpha_n\}$, $\B=\{\beta_1,\beta_2,...,\beta_n\}$ then $\alpha_1<\beta_1<\alpha_2<\beta_2<...$ or $\beta_1<\alpha_1<\beta_2<\alpha_2<...$.
So, we need to prove only
\[lm\_rank1\] Let $A\in\C_{n\times n}$ be a self-adjoint matrix with a simple spectrum. Let $\B\subset \R$ with $|\B|=n$. If $\spectr(A)$ and $\B$ are interlacing then there exists a self-adjoint matrix $B$ with $\spectr(B)=\B$ and $\rank(A-B)=1$.
Let $\spectr(A)=\A=\{\alpha_1,\alpha_2,...,\alpha_n\}$ and $\B=\{\beta_1,\beta_2,...,\beta_n\}$. As $A$ and $B$ can be put in diagonal normal form $\tilde A=\diag(\alpha_1,\alpha_2,...,\alpha_n)$ and $\tilde
B=\diag(\beta_1,\beta_2,...,\beta_n)$ by unitary transformations and unitary transformations map (by conjugation) self-adjoint matrices to self adjoint matrices, Lemma \[lm\_rank1\] is equivalent to the fact that under our assumptions on $\A$ and $\B$ the equation $$\label{eq_un}
\tilde A X- X\tilde B=R$$ has a solution in $(X,R)$ for unitary $X$ and $R$ of rank $1$. Before solving Eq.\[eq\_un\] let us introduce some notations and prove a proposition. For a finite $\A\subset\R$ let $P_\A(\lambda)=
\prod_{\alpha\in\A}(\lambda-\alpha)$. Let $\A$ and $\B$ be finite subsets of $\R$ of equal cardinality. It follows from interpolation that there exist unique $x:\A\to\A$, such that $$\label{eq_interpolation}
P_\B=P_\A-\sum_{\alpha\in\A}x_\alpha P_{\A\setminus\{\alpha\}}$$ (we write $x_\alpha$ not $x(\alpha)$). Moreover, $x_\alpha=P_\B(\alpha)/P_{\A\setminus\{\alpha\}}(\alpha)$. Studying signs of $P_\B(\alpha)$ and $P_{\A\setminus\{\alpha\}}(\alpha)$ we trivially get
\[prop\_interlace\] If $\A$ and $\B$ interlacing, then all $x_\alpha$ in Eq.\[eq\_interpolation\] have the same sign.
In fact inverse of this proposition is also valid, see Lemma 1.20 of [@Fisk].
Now let us go back to solutions of Eq.\[eq\_un\]. For $R=\{r_{ij}\}$ fixed the equation has the unique solution $X=\{x_{ij}\}$, with $x_{ij}=\frac{r_{ij}}{\alpha_i-\beta_j}$. Now, suppose that $\rank(R)=1$ or, the same $r_{ij}=y_iz_j$ for some $y,z\in\C^n$, $y,z\neq 0$. When the matrix $X$ is unitary? When its columns (rows) are orthonormal, or $$\label{eq_unit}
z_jz^*_k\sum_{i}\frac{|y_i|^2}{(\alpha_i-\beta_j)(\alpha_i-\beta_k)}=\delta_{jk}$$ It follows that $z_j\neq 0$ for all $j=1,...,n$, changing rows by columns we get the same for $y$. So, the difficult part is to guarantee that l.h.s. of Eq. \[eq\_unit\] is $0$ for $j\neq k$. Putting equality $$\frac{|y_i|^2}{(\alpha_i-\beta_j)(\alpha_i-\beta_k)}=\frac{1}{\beta_j-\beta_k}
(\frac{|y_i|^2}{\alpha_i-\beta_j}-\frac{|y_i|^2}{\alpha_i-\beta_k}).$$ into Eq.\[eq\_unit\] and multiplying it by $P_\A(\beta_k)P_\A(\beta_j)(\beta_j-\beta_k)/(z_jz^*_k)$ we get (after some elementary transformations): $$P_\A(\beta_k)\sum_{i=1}^n|y_i|^2P_{\A\setminus\{\alpha_i\}}(\beta_j)=
P_\A(\beta_j)\sum_{i=1}^n|y_i|^2P_{\A\setminus\{\alpha_i\}}(\beta_k),$$ which imply that there exists $c$, such that $$\sum_{i=1}^n|y_i|^2P_{\A\setminus\{\alpha_i\}}(\beta_j)=cP_\A(\beta_j),$$ for all $j$. Or the same, $\B$ is the set of roots of polynomial $$\Phi(\lambda)=\sum_{i=1}^n|y_i|^2P_{\A\setminus\{\alpha_i\}}(\beta_j)-cP_\A(\beta_j),$$ so $|y_i|^2=P_\B(\alpha_i)/cP_{\A\setminus\{\alpha_i\}}(\alpha_i)$. Choosing $c=1$ or $c=-1$, we get, by Proposition \[prop\_interlace\], that $|y_i|^2$ is well defined. Now, take, for example, $y_i=|y_i|$. From Eq.\[eq\_unit\] for $j=k$ we can find $|z_j|^2$. Then, taking $z_j=|z_j|$ we get needed solution of Eq.\[eq\_un\].
Almost unitary operators are near unitary operators with respect to normalized arithmetic distance {#sec_almost}
==================================================================================================
For $A,B \in C_{n \times n}$, let $d_r(A,B)$ be the normalized arithmetic distance: $$d_r(A,B)=\frac{\rank(A-B)}{n}$$
The matrix $A$ is called an $\alpha$-self-adjoint matrix if $d_r(A,
A^*)=\alpha$, where $A^*$ denotes the adjoint of $A$. The matrix $A$ is called an $\alpha$-unitary matrix if $d_r(A^*A,E) =\alpha$
The following theorems says that “near” to any $\alpha$-self-adjoint matrix there exists a self-adjoint matrix $S$, and that “near” to any $\alpha$-unitary matrix there exists an unitary matrix $U$ (for small $\alpha$).
\[th5\] For any $A \in \C_{n \times n}$ there exists a self-adjoint matrix $S$ ($S=S^*$) such that $d_r(A,S) \leq d_r(A,A^*)$.
Take $S=\frac{1}{2}(A+A^*)$.
\[th6\] For any $A \in \C_{n \times n}$ there exists a unitary matrix $U$ ($U^*U=E$), such that $d_r(A,U)\leq d_r(A^*A,E)$.
The good illustrations for this theorem are $0$-Jordan cells: $$\left(
\begin{array}{ccccccc}
0 & 0 & \cdots & 0 &0 & \cdots & 0\\
1 & 0 & \cdots & 0 & 0& \cdots &0\\
\multicolumn{7}{c}\dotfill\\
0 & 0 & \cdots & 0 & 0 & \cdots &0\\
0 & 0 & \cdots & 1 & 0 & \cdots &0\\
\multicolumn{7}{c}\dotfill\\
0 & 0 & \cdots & 0 & 0& \cdots & 0\\
\end{array}
\right)
\left(
\begin{array}{ccccccc}
0 & 1 & \cdots & 0 &0 & \cdots & 0\\
0 & 0 & \cdots & 0 & 0& \cdots &0\\
\multicolumn{7}{c}\dotfill\\
0 & 0 & \cdots & 0 & 1 & \cdots &0\\
0 & 0 & \cdots & 0& 0 & \cdots &0\\
\multicolumn{7}{c}\dotfill\\
0 & 0 & \cdots & 0 & 0& \cdots & 0\\
\end{array}
\right)=
\left(
\begin{array}{ccccccc}
0 & 0 & \cdots & 0 &0 & \cdots & 0\\
0 & 1 & \cdots & 0 & 0& \cdots &0\\
\multicolumn{7}{c}\dotfill\\
0 & 0 & \cdots & 1 & 0 & \cdots &0\\
0 & 0 & \cdots & 0 & 1 & \cdots &0\\
\multicolumn{7}{c}\dotfill\\
0 & 0 & \cdots & 0 & 0& \cdots & 1\\
\end{array}
\right),$$ but the matrix $$\left(
\begin{array}{ccccccc}
0 & 0 & \cdots & 0 &0 & \cdots & 1\\
1 & 0 & \cdots & 0 & 0& \cdots &0\\
\multicolumn{7}{c}\dotfill\\
0 & 0 & \cdots & 0 & 0 & \cdots &0\\
0 & 0 & \cdots & 1 & 0 & \cdots &0\\
\multicolumn{7}{c}\dotfill\\
0 & 0 & \cdots & 0 & 0& \cdots & 0\\
\end{array}
\right)$$ is unitary.
Let $\rank(A^*A-E)=r$, so there exist subspace $X\subset L$, $\dim(X)=n-r$ such that $A^*A|_X=E|_X$. Consider $A|_X \,:\, X\to
Y=A(X)$. Under assumptions of the theorem $A^*(Y)=X$, it follows that $(A|_X)^*=A^*|_Y:Y\to X$, so $A|_X\, :\,X\to Y$ is an unitary operator. Choose any unitary operator $B: X^\perp\to Y^\perp$ ($B^*B=E_{X^\perp}$). Then $U=A|_X\oplus B$ proves the theorem.
It is not clear if this proof could be adapted for normal matrices – unitary operator from an unitary space to another unitary space is well defined, but how to define normal operators between different unitary spaces...?
Question: If we define $\alpha$-normal matrices in similar form to self-adjoint and unitary matrices, the equivalent of theorems \[th5\] and \[th6\] are true for normal matrices?
Almost commuting matrices {#sec_com}
=========================
\[th\_com\]
For every $4 \leq n \in {{\bf N}}$ and every $A\in\C_{n \times n}$ with simple spectrum there exists $X\in \C_{n\times n}$ such that $d_r(AX,XA) < 2/n$ and for any matrix $B$, commuting with $A$, $d_r(B,X) \geq \frac{1}{2}$.
Before starting the proof of the theorem we need some facts.
\[propol\] Let $\{\la_1,...,\la_k\}$ and $\{\al_1,...,\al_k\}$ be two disjoint sets, then the matrix $M=[x_{ij}]$ with $x_{ij}=\frac{1}{\al_i-\la_j}$ is nonsingular
The matrix $M$ has the form $$M= \left(
\begin{array}{cccc}
\frac{1}{\al_1-\la_1} & \frac{1}{\al_1-\la_2} & \cdots & \frac{1}{\al_1-\la_k} \\
\frac{1}{\al_2-\la_1} & \frac{1}{\al_2-\la_2} & \cdots & \frac{1}{\al_2-\la_k} \\ \multicolumn{4}{c}\dotfill\\
\frac{1}{\al_k-\la_1} & \frac{1}{\al_k-\la_2} & \cdots & \frac{1}{\al_k-\la_k} \\
\end{array}
\right)$$ Let $P(\la)=(\la-\la_1)(\la-\la_2)\cdots (\la-\la_k)$. Multiply each row $j$ of the matrix $M$ by $P(\al_j)$ we obtain a matrix of the form $$\tilde M= \left(
\begin{array}{cccc}
P_1(\al_1) & P_2(\al_1) & \cdots & P_k(\al_1) \\
P_1(\al_2) & P_2(\al_2) & \cdots & P_k(\al_2) \\
\multicolumn{4}{c}\dotfill\\
P_1(\al_k) & P_2(\al_k) & \cdots & P_k(\al_k) \\
\end{array}
\right)$$ where $P_j(\la)=\frac{P(\la)}{\la-\la_j}$. This matrix will be nonsingular if and only if matrix $M$ is nonsingular. We will prove that matrix $\tilde M$ is nonsingular showing that the following system of linear equations has a unique solution: $$\left(
\begin{array}{cccc}
P_1(\al_1) & P_2(\al_1) & \cdots & P_k(\al_1) \\
P_1(\al_2) & P_2(\al_2) & \cdots & P_k(\al_2) \\
\multicolumn{4}{c}\dotfill\\
P_1(\al_k) & P_2(\al_k) & \cdots & P_k(\al_k) \\
\end{array}
\right)
\left(
\begin{array}{c}
a_1\\
a_2 \\
\vdots \\
a_k \\
\end{array}
\right)= \left(
\begin{array}{c}
b_1\\
b_2 \\
\vdots \\
b_k \\
\end{array}
\right)$$
So we have to solve the system $\sum_{i=1}^k a_iP_i(\al_j)=b_j$. Consider the polynomial $\Phi(\la)=\sum_{i=1}^k a_iP_i(\la)$, note that $\Phi(\la_i)=a_iP_i(\la_i)$ because $P_i(\la_j)=0$ for $i\neq j$. We have $k$ points $b_j$, therefore, we can use Lagrange interpolation to find the unique polynomial $\Phi(\la)$ of degree $k-1$ such that $\Phi(\al_j)=b_j$, and then we can compute the values $a_i=\frac{\Phi(\la_i)}{P_i(\la_i}$ ($P_i(\la_i)\neq 0$ for all $\la_i$ are different).
Now we are ready to prove Theorem \[th\_com\].
Consider the matrix equation $$\label{ax}
AX-XA=\{c_{ij}\},$$ with $c_{ij}=i+j \textsf{mod} 2$. This matrix has the following form
$$AX-XA= \left(
\begin{array}{cccccc}
0 & 1 & 0 & 1 & \cdots & 1+ n \hspace{0.2cm}\textsf{mod} 2 \\
1 & 0 & 1 & 0 & \cdots & n \hspace{0.2cm}\textsf{mod} 2 \\
0 & 1 & 0 & 1 & \cdots & 1+n \hspace{0.2cm}\textsf{mod} 2 \\
1 & 0 & 1 & 0 & \cdots & n \hspace{0.2cm}\textsf{mod} 2 \\
\multicolumn{6}{c}\dotfill\\
n +1 \hspace{0.2cm}\textsf{mod} 2 & \multicolumn{4}{c}\dotfill & 0
\end{array}
\right)$$
Let $A$ be a diagonal matrix with simple spectrum $A=diag(\la_1,\la_2,...,\la_n)$. One matrix $X$ that satisfies Eq. \[ax\] is $X = \{x_{ij}\}$ with\
$$x_{ij}=\begin{cases}
\frac{c_{ij}}{\la_i-\la_j} & \text{for $i \neq j$} \\
0 & \text{for $i=j$} \\
\end{cases}$$ Every matrix $B$ that commute with $A$ should be necessarily a diagonal matrix $B=diag(b_1,b_2,...,b_n)$, then $X-B=\{x^*_{ij}\}$ with $$x^*_{ij}=\begin{cases}
\frac{c_{ij}}{\la_i-\la_j} & \text{for $i \neq j$} \\
-b_i & \text{for $i=j$} \\
\end{cases}$$
If we delete from $X-B$ the odd columns and the even rows, we obtain a submatrix $X'$ of size $\lfloor \frac{n}{2} \rfloor \times \lfloor \frac{n}{2} \rfloor$ o f the form $$\{x'_{ij}\}= \frac{1}{\la_i^*-\la_j^*}$$ with $i^*=2i-1$, $j^*=2j$. By proposition \[propol\] this matrix is nonsingular. Therefore, $\rank(X-B)\geq \frac{n}{2}$, and we obtain $d_r(X,B)\geq \frac{1}{2}.$
[100]{}
An introduction to Numerical Analysus, 2nd ed. , 1989.
Sur l’éqution à l’aide de laquelle on détemine les inégalités séculaires,
World Scientific, 1996, ISBN 9810226381, 9789810226381
Almost commuting matrices need not be nearly commuting, , 102, (1988), N.3, pp. 529–533
Almost Commuting Unitary Matrices, , Vol. 106, No. 4. (Aug., 1989), pp. 913-915.
Imbedding conditions for hermitian and normal matrix , 9 (1957) pp. 298-304
All automorphisms of the calkin algebra are inner, Preprint ArXiv:0705.3085v3
Polynomials, roots, and interlacing, Preprint ArXiv: math/0612833
Vector Bundles, Linear Representations, and Spectral Problems, Preprint ArXiv: math/0304325 , Almost commuting selfadjoint matrices and applications, in [*Operator algebras and their applications*]{} (Waterloo, ON, 1994/1995), Fields Inst. Commun., 13, pp. 193–233, Amer. Math. Soc..
Low Rank Perturbation of Jordan Structure, , vol. 25, No 2, 2003, pp. 495–506.
On the change in the spectral properties of a matrix under a perturbation of a sufficiently low rank, , 38 (2004), no. 1, pp. 85–88; translation in Funct. Anal. Appl. 38 (2004), no. 1, pp 69–71
Weyr characteristic under low rank perturbation,
The Weyr characteristic. (1999), pp. 919–929.
Invariant factors under rank one perturbations. , XXXII, (1980), N.1. pp.240-245
, Asymptotically commuting finite rank unitary operators without commuting approximants, , 45, (1983), N.1-4, pp. 429–431,
[^1]: The authors would like to thank V.S.Savchenko, J. Moro and F. Dopico for useful comments and references. The work was partially supported by CONACyT grant SEP-25750, and PROMEP grant UASLP-CA-21.
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- Adam Harris and Krzysztof Wysocki
title: 'Branch Structure of $J$–holomorphic Curves near Periodic Orbits of a Contact Manifold'
---
[**Abstract:**]{} Let $M$ be a three–dimensional contact manifold, and $\tilde{\psi}:D\setminus\{0\}@>>>M\times{\Bbb R}$ a finite–energy pseudoholomorphic map from the punctured disc in ${\Bbb C}$ that is asymptotic to a periodic orbit of the contact form. This article examines conditions under which smooth coordinates may be defined in a tubular neighbourhood of the orbit such that $\tilde{\psi}$ resembles a holomorphic curve, invoking comparison with the theory of topological linking of plane complex algebroid curves near a singular point. Examples of this behaviour which are studied in some detail include pseudoholomorphic maps into ${\Bbb E}_{p,q}\times{\Bbb R}$, where ${\Bbb E}_{p,q}$ denotes a rational ellipsoid (contact structure induced by the standard complex structure on ${\Bbb C}^{2}$), as well as contact structures arising from non-standard circle–fibrations of the three–sphere.
[**math reviews subject classification:**]{} 32 Q65, 53 D10
Introduction
============
The local theory of pseudoholomorphic maps from a Riemann surface into an almost complex, symplectic $4$–manifold $({\cal M}, J, \omega)$ is developed largely around comparisons with the classical theory of plane algebroid curves. Following the initial investigations of Gromov, in work of McDuff \[12\], Micallef and White \[14\], and Sikorav \[15\], methods of construction of local diffeomorphisms between neighbourhoods of $p\in{\cal M}^{2k}$ and ${\bold 0}\in{\Bbb C}^{k}$ (specifically, $k=2$) were found in order to exhibit singular $J$–holomorphic curves $\tilde{\psi}:({\Bbb C},0)@>>>({\cal M},p)$ as being locally equivalent to holomorphic ones. As a result the local topological data associated with singularities of plane curves can be transferred to the pseudoholomorphic context. Recall that the germ of an algebraic curve $\Gamma$ with singularity at ${\bold 0}\in{\Bbb C}^{2}$ of multiplicity $n$ is represented by the vanishing locus of a function $f(w_{1},w_{2})
\in{\cal O}_{{\Bbb C}^{2}}({\cal U})$ for some neighbourhood ${\cal U}$ of the origin, such that the homogeneous polynomial $H_{n}(w_{1},w_{2})$ of terms of minimal degree $n$ in the Taylor expansion of $f$ at ${\bold 0}$, has $n\geq 2$. Each linear factor of $H_{n}$ describes a complex tangent line to $\Gamma$ at ${\bold 0}$, hence in order that the germ of $\Gamma$ be irreducible at the origin it is necessary that $H_{n}$ be the $n$–th power of a single linear term. If ${\Bbb S}^{3}_{
\varepsilon}$ bounds a ball of radius $\varepsilon$ in ${\cal U}$, then the geometric locus of this term (i.e., ignoring multiplicity) intersects ${\Bbb S}^{3}_{\varepsilon}$ in a great circle $C$ corresponding to the axis of a solid torus in which $\Gamma\cap{\Bbb S}^{3}_{\varepsilon}$ also describes the trajectory of an iterated torus knot $K_{\Gamma}$. The topological “linking” of $K_{\Gamma}$ with $C$ provides a set of numerical invariants in addition to the multiplicity, on which the classification of singular curve–germs is based (cf., e.g., \[2\]).
The existence of a local analytic [*parametrization*]{} of $\Gamma$ is a problem first studied systematically by Newton, and later developed rigorously by Weierstrass and Puiseux among others. Thus $\Gamma$ may be represented in standard form as the image of a holomorphic map $\Phi:(D,0)@>>>({\cal U},{\bold 0}) \ , \ D\subseteq{\Bbb C}$, such that $\Phi(z) = (z^{n},\Sigma_{i\geq n+1}a_{i}z^{i})$ (cf. \[2\]). This type of series expansion takes advantage of the natural splitting of ${\Bbb
C}^{2}$ into a product of coordinate planes. In order to see conveniently the relationship between certain essential exponents of the parametrization and the linking invariants of $K_{\Gamma}$ most treatments therefore replace the ball bounded by ${\Bbb S}^{3}_{\varepsilon}$ with a bidisc $\Delta\times
\Delta$ (this is after all homeomorphic to the ball, hence topological data are preserved). On the other hand, the totality of all complex $1$–dimensional subspaces of ${\Bbb C}^{2}$ induces a fibration of ${\Bbb
S}^{3}$ by great circles, discovered by Hopf. Given a vector ${\bold v}_{p}
\in{\Bbb C}^{2}$ corresponding to $p\in{\Bbb S}^{3}$, the standard complex structure $J_{0}$ defines a vector field $X(p):= J_{0}\cdot{\bold v}_{p}$ which is tangent to the great circle through each $p\in{\Bbb S}^{3}$. Via the Euclidean inner product on ${\Bbb R}^{4}\approx{\Bbb C}^{2}$, the orthogonal complement of $X$ in $T{\Bbb S}^{3}$ defines a non–integrable plane–field, or “contact structure”, on ${\Bbb S}^{3}$ corresponding to the kernel of the $1$–form $\lambda_{0}$ which is the metric dual of $X$. The splitting of $T{\Bbb C}^{2}\mid_{{\Bbb S}^{3}}$ induced by this structure provides an alternative natural frame for the relationship between $K_{\Gamma}$ and holomorphic parametrization of plane curves, though we will return to this matter only in the final section as the “classical model” of a more general study.
Contact structures are defined on manifolds $M$ of odd dimension, though for the purposes of this article we will always assume the dimension to be three. If $\lambda$ is a $1$–form representing such a structure, the extension of $\omega:= d\lambda$ to $M\times{\Bbb R}$ determines a “symplectisation” of $(M,\lambda)$. Conversely, compact symplectic manifolds $({\cal M},\omega)$ with ends of symplectisation type motivate the role of contact geometry within modern symplectic topology. Through the work of Hofer, Wysocki and Zehnder \[8\], \[9\] and Eliashberg \[3\], $J$–holomorphic curves have been adapted via symplectisation to contact geometry and topology, specifically as a tool for analysing the Weinstein conjecture \[6\] and conversely for explicit construction of moduli of pseudoholomorphic maps into manifolds ${\cal M}$ with symplectisation–type ends \[3\], \[9\]. Recall that the “Reeb vector field” $X$ associated with a given structure $\lambda$ is uniquely determined by the conditions $\lambda(X)\equiv
1 \ , \ {\cal L}_{X}\lambda = 0$ (Lie symmetry). The Weinstein conjecture asserts that the Reeb flow of a compact contact $3$–manifold always admits at least one closed characteristic (or “periodic orbit”) $\gamma:{\Bbb S}
^{1}@>>>M$ such that $\dot{\gamma}(t) = X(\gamma(t))$. If $\xi\subset TM$ denotes the plane–field corresponding to $ker(\lambda)$, then a contact structure on $M$ is said to be “pseudohermitian” when it is equipped with a partial almost complex structure $j\in C^{\infty}
(M,\xi\otimes\xi^{*})$ such that $d\lambda(*,j\cdot *)\mid_{\xi}$ is a positive definite and symmetric quadratic form. The splitting of $TM$ via $X$ and $\xi$ also provides a natural extension of $j$ to $T(M\times{\Bbb R})$ in relation to which the Cauchy–Riemann equation is defined for pseudoholomorphic maps into $M\times{\Bbb R}$ (cf., e.g., \[6\]).
Let $\tilde{\psi}:\Sigma@>>>M\times{\Bbb R}$ be one such map. For any smooth function $h:{\Bbb R}\rightarrow[0,1]$ we may extend the contact form to $M\times{\Bbb R}$ by defining $$\lambda_{h}(p,a) = h(a)\cdot\lambda(p)$$ to act on each $T_{(p,a)}(M\times{\Bbb R})$. In particular, if ${\cal F}_{+}^{1}$ denotes the space of all smooth $h$ such that the derivative $h'\geq 0$, then the “energy” is given as $$E(\tilde{\psi}) = \sup_{h\in{\cal F}_{+}^{1}}\int_{\Sigma}\tilde{\psi}^{*}d\lambda_{h} \ .$$ If $\psi$ denotes the projection of $\tilde{\psi}$ onto $M$ and $a$ its projection onto ${\Bbb R}$, recall that $$\psi^{*}d\lambda = (|\psi_{\eta}|^{2} + |\psi_{\zeta}|^{2})d\eta\wedge
d\zeta \ ,$$ where $z = \eta + {\bold i}\zeta$ is a local complex coordinate on $\Sigma$. $E(\tilde{\psi})$ consequently vanishes if and only if $\tilde{\psi}$ is constant – a condition that follows automatically from Stokes’ theorem when $\Sigma$ is compact, hence pseudoholomorphic maps into symplectisations are naturally defined on [*punctured*]{} Riemann surfaces. Specifically, let $\tilde{\psi}$ be a finite–energy map defined on a punctured disc $D\setminus\{0\}\subset{\Bbb C}$ and let $r = -\ln(|z|)$ and $\varphi =
\arg(z)\in{\Bbb S}^{1}\approx{\Bbb R}/2\pi{\Bbb Z}$. A theorem of Hofer, Wysocki and Zehnder \[8\] states that if $\tilde{\psi}$ has unbounded image in $M\times{\Bbb R}$ then there exists a real number $T\neq 0$ and a Reeb–periodic orbit $\gamma:{\Bbb S}^{1}@>>>M$ such that $$\lim_{r@>>>\infty}\psi(r,\varphi) = \gamma(T\varphi)\hspace{.1in};
\hspace{.1in} \lim_{r@>>>\infty}\frac{a(r,\varphi)}{r} = T\hspace{.1in}
\hbox{in}\hspace{.1in} C^{\infty}({\Bbb S}^{1}) \ .$$ $T$ is moreover an integer multiple of the minimal period $\tau$ of $\gamma$ and corresponds to the “charge” $\lim_{r@>>>\infty}\frac{1}{2\pi}\int_{{
\Bbb S}^{1}}\psi^{*}\lambda$ of $\tilde{\psi}$ at $0\in D$. Closed characteristics of the Reeb flow are thus realised asymptotically by cylinders mapped pseudoholomorphically into the corresponding symplectisation of $M$ (we pass over the very substantial theory devoted to [*existence*]{} of such mappings of cylinders in general; cf., however, \[6\],\[8\]).
The asymptotic relations established in this theorem and refined elsewhere \[10\] are fundamental to the present article in which we examine the topological behaviour of $\tilde{\psi}(D\setminus\{0\})$ within a tubular neighbourhood of ${\cal P}:=\gamma({\Bbb S}^{1})\subset M$. Before summarising our method and results, however, it should be mentioned that they require added technical hypotheses to be imposed locally upon the Reeb flow itself. Let $\Delta\times({\Bbb R}/ \tau{\Bbb Z}) \ , \ (\Delta\subseteq{\Bbb R}^{2})$ be the tubular domain on which a general system of ordinary differential equations $\dot{{\bold x}} = f({\bold x})$, for $f$ smooth and $\vartheta$–periodic (minimal period $\tau$) in the coordinates ${\bold x} = (x,y,\vartheta)\in
\Delta\times{\Bbb S}^{1}$, is defined. Suppose that $\{x = y = 0\}$ corresponds to a periodic solution $\gamma(\vartheta)$, i.e., if $\phi(x,y,\vartheta)$ denotes the associated $\vartheta$–parametric family of diffeomorphisms, depending on initial conditions $(x,y)\in\Delta$ suffciently small, then $\gamma(\vartheta) = \phi(0,0,\vartheta) \ , \ \gamma(\vartheta+\tau) =
\gamma(\vartheta)$. The linear variational equation for $\dot{{\bold x}} = f({\bold x})$ at $\gamma$ then has the form $$\frac{\partial}{\partial\vartheta}\phi_{*}(0,0,\vartheta) = f_{*}(0,0,
\vartheta)\cdot\phi_{*}(0,0,\vartheta) \hspace{.1in}(\dag),$$ noting that $f_{*}(0,0,0) = f_{*}(0,0,\tau)$ and $\phi_{*}(0,0,\tau) =
A\cdot\phi_{*}(0,0,0) = A$, where $A$ denotes the “holonomy matrix” associated with (). From the theory of ordinary differential equations (cf. e.g., \[5\]) it is well–known that the eigenvalues, or “characteristic multipliers” of $A$ are important in determining the stability of the flow $\phi$ near an orbit ${\cal P}$. In particular, $1$ is automatically a characteristic multiplier in the $\vartheta$–direction, while eigenvalues of complex modulus greater than or less than one imply that the flow will be either unstable or asymptotically stable along closed surfaces containing ${\cal P}$ (cf. \[5\]). Specifically when $f$ corresponds to the Reeb vector field of a contact structure, the flow $\phi$ is [*area–preserving*]{}, hence $det(A) = 1$ and the remaining pair of eigenvalues are mutually reciprocal (mutually conjugate if they are unimodular). Hofer, Wysocki and Zehnder actually use the asumption that $A$ is not the identity (cf. “non–degeneracy” of ${\cal P}$) as a technical hypothesis in obtaining their asymptotic formulae (cf. \[8\]). From the viewpoint of the present article this condition has the disadvantage that it does not hold for the Reeb flow of the standard contact structure on the $3$–sphere, corresponding to the Hopf fibration as mentioned above. In fact, it does not hold for a large sub–class of the contact structures which are examined below, although an extension of their methods does allow these authors to include some contact manifolds that are foliated by periodic orbits (cf. \[10\]). The asymptotics have been rederived in a recent thesis of F. Bourgeois \[1\], however, by means of the following alternative hypothesis.
(cf. \[1\]) A contact form $\lambda$ on $M$ is said to be of “Morse–Bott type” if, for every $T>0 \ , \ N_{T} := \{p\in M \
| \ \phi(p,T) = p\}$ is a smooth, closed, orientable submanifold of $M$ such that $d\lambda\mid_{N_{T}}$ has locally constant rank, and $T_{p}
N_{T} = ker(\phi_{*}(p,T) - Id)$.
For present purposes it will henceforth be assumed that all contact structures under consideration are of Morse–Bott type. Our aim is to understand the extent to which $J$–holomorphic cylinders in symplectisations actually [*link*]{} with the periodic orbits which they approach asymptotically. In this sense the case of algebroid knots $K_{\Gamma}$ linking with great circles in the $3$–sphere suggests a classical model, since ${\Bbb S}^{3}
\times{\Bbb R}$ is clearly diffeomorphic to ${\Bbb C}^{2}\setminus\{0\}$ (we will return to this matter below). By analogy with the work of Sikorav and others we will show that a suitably constructed diffeomorphism applied to a tubular neighbourhood of ${\cal P}$ allows $\tilde{\psi}$ to be represented locally by a holomorphic parametrization, and the sense in which this is possible will be the subject of sections two and three. The problem of “asymptotic similarity” addressed here has a global aspect, however, not found in the study of singularities of $J$–holomorphic curves. Because of this we have imposed two further conditions on the pseudohermitian structure :
$(1)$ in a sense to be made precise in section three, ${\cal P}$ will be said to be “locally recurrent” if a sufficiently small disc transverse to ${\cal
P}$ can be considered a surface of section of the Reeb flow (the diffeomorphism of the disc consequently induced by the return map will be denoted by $\alpha$). In particular this condition forces the characteristic multipliers of $A$ to be unimodular, but is a stronger assumption than orbital stability near ${\cal P}$;
$(2)$ it will be assumed that the Reeb flow is not only a Lie symmetry of the contact form, but also the almost complex structure, i.e., ${\cal L}_{X}j = 0$ (hence in particular $\alpha$ is diffeomorphically equivalent to a rotation).
It remains now to summarise our results.
[**Theorem 1**]{}(cf. section 3) [*Let $(\psi,a):D\setminus\{0\}@>>>M
\times{\Bbb R}$ be a $J$–holomorphic curve of finite energy and charge $n$ at $z=0$, asymptotic to a locally recurrent periodic orbit ${\cal P}$, near which ${\cal L}_{X_{\lambda}}j = 0$. Consider any tubular neighbourhood of ${\cal P}$ in $M$, diffeomorphic to $\Delta\times{\Bbb S}^{1}$ such that $\{0\}\times{\Bbb S}^{1}\approx{\cal P}$ and such that $\psi^{-1}(
\Delta\times\{\vartheta_{0}\})$ for some fixed $\vartheta_{0}\in{\Bbb S}^{1}$ divides $D\setminus\{0\}$ into $n$ “quasi–sectors” $Q_{k}$. Then there exists a diffeomorphic change of coordinates in $\Delta\times[0,2\pi)$ such that on each $Q_{k}\subset D\setminus\{0\}$ the map $(\psi,a)$ can be expressed in the form $$(F_{k}(z), H_{k}(z) - \frac{1}{2\pi{\bold i}}\hat{G}(z,\bar{z}))\ , \
\hspace{.1in} 0\leq k\leq n-1 ,$$ where $F_{k} \ , \ H_{k}$ are holomorphic on $Q_{k}$ and continuous on $\overline{Q}_{k}$, such that $$F_{k}\mid_{\overline{Q}_{k}\cap\overline{Q}_{k+1}} = \alpha\circ
F_{k+1}\mid_{\overline{Q}_{k}\cap\overline{Q}_{k+1}} \ ,$$ while each $H_{k}$ corresponds to an analytic branch of $\frac{1}{2\pi{\bold i}}\log(\rho)$ , for $\rho$ holomorphic such that $ord_{0}(\rho) = n$. Moreover, the function $\hat{G}$ belongs to $C^{2}(D)$ and is bounded by $K|z|$. Finally, if $\alpha = 1$, then each $F_{k}$ is the restriction of a single function $F$ holomorphic on $D$, $F(0) = 0$*]{}.
A representation corresponding to the classical local parametrization of plane algebroid curves is moreover easily derived when $\alpha = 1$. In this case the relationship between charge and algebraic multiplicity also becomes explicit.
In section four we construct a class of pseudohermitian structures on the $3$–sphere for which $\alpha = 1$, containing the standard structure as a distinct special case, and using results on characterisation of generalised circle fibrations due to Gluck and Warner \[4\]. Recent ideas of McKay \[13\] relating elliptic line congruences to “osculating” almost complex structures associated with a four–dimensional real vector space $V$ are also incorporated in the construction.
[**Theorem 2**]{}(cf. section 4) [*Let $J$ be the osculating complex structure of an elliptic line congruence $\Sigma\subset{\Bbb G}r_{2}(V)$, such that the associated skew–symmetric 2–form $\omega_{\Sigma}$ on $V\setminus\{0\}$ is closed. Then the 1–form $\lambda$, such that $\lambda_{{\bold v}}
:=i_{{\bold v}}\omega_{\Sigma}$, defines a (fillable, hence tight) contact structure on ${\Bbb S}^{3}$ for which ${\cal L}_{X_{\lambda}}
J\mid_{ker(\lambda)} = 0$ and $\alpha = 1$. Moreover, two such structures are equivalent via a diffeomorphism $\delta$ of ${\Bbb S}^{3}$ if and only if $\delta\in{\Bbb O}(4)$*]{}.
(The notions of “tight” and “over–twisted” contact structures are important in the study of global dynamics and topology (cf. \[3\],\[6\]) but are not referred to explicitly in this article.)
The final section is devoted to two examples, the first of which is the standard structure on ${\Bbb S}^{3}$. Holomorphic maps $\Phi:(D,0)@>>>
({\Bbb C}^{2},{\bold 0})$ corresponding to irreducible germs of a curve $\Gamma$ are translated explicitly into pseudoholomorphic maps $\Psi:D@>>>{\Bbb S}^{3}\times{\Bbb R}$ such that the tangent to $\Gamma$ at the origin becomes the locally recurrent periodic orbit ${\cal P}$ corresponding to a great circle. We note that $\alpha = 1$ in this case as a direct consequence of the Hopf fibration restricted to any tubular neighbourhood of ${\cal P}$. It should also be mentioned that the identification of algebraic curves with finite–energy pseudoholomorphic maps into $({\Bbb S}^{3}\times{\Bbb R}, \lambda_{0}, J_{0})$ has been addressed from a different perspective in an article of Hofer and Kriener \[7\]. The second example discussed in this section includes the well–known contact structures $\lambda_{0}$ more generally induced by $J_{0}$ and the Euclidean inner product on rational ellipsoids $${\Bbb E}_{p,q} = \{(w_{1},w_{2})\in{\Bbb C}^{2} \ | \ p|w_{1}|^{2} +
q|w_{2}|^{2} = 1\} \ ,$$ where $\frac{p}{q}\in{\Bbb Q}$. In this case the hypersurface is foliated by recurrent periodic orbits such that either $\alpha$ is the identity or a non–trivial rational rotation through either $2\pi\frac{p}{q}$ or $2\pi\frac{q}{p}$. Given a tubular neighbourhood $\Delta\times{\Bbb S}^{1}$ of one of the two periodic orbits ${\cal P}$ for which $\alpha$ is non–trivial, we construct a solid torus $\Delta'\times{\Bbb S}^{1}$ with contact form $\lambda$, and a smooth covering map $\beta:\Delta\times{\Bbb S}^{1}@>>>\Delta'\times{\Bbb S}^{1}$ such that $\lambda_{0} = \beta^{*}\lambda$ and $\beta_{*}\circ j_{0} =
j_{0}'\circ\beta_{*}$ (where $j_{0} = J_{0}\mid_{ker(\lambda_{0})}$ and $j_{0}' = J_{0}\mid_{ker(\lambda)}$). Pseudoholomorphic maps asymptotic to this orbit in $\Delta
\times{\Bbb S}^{1}$ then project pseudoholomorphically via $\beta$ into $\Delta'\times{\Bbb S}^{1}$, where the central axis is a recurrent orbit of the projected Reeb vector field such that $\alpha' = 1$. Working backwards from a holomorphic parametrization in $\Delta'\times{\Bbb S}^{1}\times
{\Bbb R}$, we then explicitly construct pseudoholomorphic maps $(\psi,a):D\setminus\{0\}@>>>\Delta\times{\Bbb S}^{1}\times{\Bbb R}$ according to the prescription of Theorem 1 (cf. section 5).
The authors would like to thank Dr S. Gadde and Dr Y. Tonegawa for helpful and stimulating discussions on occasions during the period of research for this article. The author (Harris) also gratefully acknowledges the support of an Australian Research Council grant during the initial phase of work undertaken at the University of Melbourne.
Local Model of parametrization
==============================
In this section we will first review the phenomenon of holomorphic similarity in the neighbourhood of a smoothly embedded point, using this as a model for the asymptotic case to follow. Let $M$ be a closed, oriented three–manifold with contact form $\lambda$ and associated Reeb vector field $X_{\lambda}$. If $\xi\subset TM$ denotes the sub-bundle corresponding to the kernel of $\lambda$, let $J$ be an almost complex structure on $\xi$ which is compatible with $\lambda$. Consider a Riemann surface $\Sigma$ and a pseudoholomorphic map $$(\psi,a):\Sigma@>>>M\times{\Bbb R}.$$ Let $p_{0}\in im(\psi)$ be a smoothly embedded point, and let $D$ be a neighbourhood of $\psi^{-1}(p_{0})$ with complex coordinate $z = \eta +
{\bold i}\zeta$ such that $z(\psi^{-1}(p_{0})) = 0$. For all $p\in M$, and all $v\in T_{p}M$, consider the projection map $\pi_{p}:T_{p}M@>>>\xi_{p}$ defined by $\pi_{p}(v) = v - \lambda_{p}(v)\cdot X_{\lambda}$. $\psi$ then satisfies the equations $$\pi(\psi_{\eta}) + J\pi(\psi_{\zeta}) = 0 \ \hspace{.1in}(*) \ ,$$ $$\lambda(\psi_{\zeta}) = -a_{\eta} \ \hspace{.1in} \
\lambda(\psi_{\eta}) = a_{\zeta} \ \hspace{.1in}(\dag) \ .$$ We recall as a consequence that $$\psi^{*}d\lambda = (|\pi(\psi_{\eta})|^{2} +
|\pi(\psi_{\zeta})|^{2})d\eta\wedge d\zeta \ ,$$ hence $X_{\lambda}(p_{0})$ does not belong to the image of $\psi_{*}$ at the origin of $D$ if and only if $|\pi(\psi_{\eta})|^{2} + |\pi(\psi_{\zeta})|^{2}$ does not vanish there – a fact which is in turn granted by the assumption that $p_{0}$ is smoothly embedded. Now choose a system of Darboux coordinates $(x_{1},x_{2},x_{3})$ neighbouring $p_{0}$ in $M$. Within this neighbourhood the image of $\psi$ may then be realized as the graph of a smooth function $x_{3} =
f(x_{1},x_{2})$. Noting that $\lambda = dx_{3}+x_{1}dx_{2}$ and $X_{\lambda} = \frac{\partial}{\partial x_{3}}$ in these coordinates, we remark that $$\pi_{p}(v_{1},v_{2},v_{3}) = (v_{1},v_{2},-x_{1}v_{2}) \ ,$$ hence the standard projection $(v_{1},v_{2},v_{3})\mapsto (v_{1},v_{2})$ defines a linear isomorphism $\mu$ between $\xi_{p}$ and ${\Bbb R}^{2}$ for all $p$ near the origin at $p_{0}$. It follows that $\mu\circ J\circ\mu^{-1}$ corresponds to a $2\times 2$ matrix $j$, the restriction of which to the graph of $im(\psi)$ depends only on the coordinates ${\bold x}:= (x_{1},x_{2})$. Under projection by $\mu$, equation (\*) then becomes $${\bold x}_{\eta}(\eta,\zeta) + j({\bold x})\cdot{\bold x}_{\zeta}
(\eta,\zeta) = {\bold 0} \ .$$ Letting ${\bold e}_{1}$ denote the vector $(1,0)$, consider the system of ordinary differential equations $$\frac{d{\bold x}}{dt} = j({\bold x})\cdot{\bold e}_{1} \ ,$$ and define a local diffeomorphism $\varphi$ near $p_{0}$, via existence and uniqueness of solutions, such that $$(\dot{x}_{1},\dot{x}_{2}) = j(x_{1},x_{2})\cdot{\bold e}_{1} \
\Rightarrow\varphi(x,t) := (x_{1}(t),x_{2}(t)) \ , \ \hspace{.2in}
\varphi(x,0) = (x,0) \ .$$ Hence $$\varphi_{*}^{-1}\circ j\circ\varphi_{*} = j_{0} =
\left(\begin{array}{cc}
0 & -1 \\
1 & 0
\end{array}\right) \ ,$$ and $\tilde{\bold x}_{\eta}:=\varphi_{*}^{-1}({\bold x}_{\eta}) \ , \ $ $\tilde{\bold x}_{\zeta}:=\varphi_{*}^{-1}({\bold x}_{\zeta})$ implies $$\tilde{\bold x}_{\eta}(\eta,\zeta) + j_{0}\cdot
\tilde{\bold x}_{\zeta}(\eta,\zeta) = {\bold 0} \ .$$ Or, more simply, writing $w = x+{\bold i}t$, we have $w = F(z)$ on a possibly smaller disc $D'\subseteq D\subset{\Bbb C}$, containing the origin, such that $F(0) = 0 \ , \ \frac{\partial F}
{\partial\bar{z}} = 0$.
Within the coordinate neighbourhood defined by $(x, t, x_{3})$, we now revisit the equations $(\dag)$, with respect to $$\lambda' = (\varphi\times 1)^{*}\lambda = dx_{3}+f_{1}(x,t)dx +
f_{2}(x,t)dt \ ,$$ in the form $$\lambda'(\Psi_{\eta}) = (x_{3})_{\eta} + f_{1}(x,t)x_{\eta}
+ f_{2}(x,t)t_{\eta} = a_{\zeta}$$ $$\lambda'(\Psi_{\zeta}) = (x_{3})_{\zeta} + f_{1}(x,t)x_{\zeta}
+ f_{2}(x,t)t_{\zeta} = -a_{\eta} \ ,$$ where $\Psi:=(\varphi\times 1)^{-1}\psi$. Hence $u = x_{3}+{\bold i}a$ implies $$\frac{\partial u}{\partial\bar{z}} = -\{f_{1}(x,t)\frac{\partial x}
{\partial\bar{z}} + f_{2}(x,t)\frac{\partial t}{\partial\bar{z}}\}$$ $$= -\{f_{1}(F(z),\overline{F}(z))\frac{\partial}{\partial\bar{z}}
\Re(F) + f_{2}(F(z),\overline{F}(z))\frac{\partial}{\partial\bar{z}}
\Im(F)\}$$ $$= -\frac{1}{2}[(f_{1}+{\bold i}f_{2})\circ F(z)]\cdot\overline{
F'(z)} := -F^{*}\omega \ ,$$ where $\omega :=
\frac{1}{2}(f_{1}(w,\bar{w})+{\bold i}f_{2}(w,\bar{w}))d\bar{w}$. Let $\Delta$ denote a small disc centred at the origin of the $(x,t)$–plane, and apply the Cauchy–Green formula to obtain a smooth function $g(w,\bar{w}) \ , \ g(0) = 0$ satisfying the equation $\bar{\partial}g = \omega$ on $\Delta$. Then we have $$F^{*}\omega = \bar{\partial}(F^{*}g) = \bar{\partial}(g\circ F) \ .$$ Hence for the parametrization of $\Psi$ we now have the system of equations $$w = F(z) \ ; \ \hspace{.1in} u = H(z) - g\circ F(z) \ ,$$ with respect to holomorphic functions $F$ and $H$, which simplifies under the coordinate transformation $$w' = w \ ; \ \hspace{.1in} u' = u + g(w,\bar{w}) - {\bold i}a(0)$$ to $u' = H(z) - {\bold i}a(0)$. Finally, $ord_{0}(H -{\bold i}a(0)) = n$ implies there exists a holomorphic function $h(z) \ , \ h(0) = 0 \ ,
\ h'(0)\neq 0$, on a possibly smaller disc $D'\subseteq D$, such that $H -{\bold i}a(0) = h^{n}$ , hence $z':=h(z)$ implies $$w' = F\circ h^{-1}(z') \ ; \ \hspace{.1in} u' = (z')^{n} \ ,$$ which corresponds to the classical local parametrization of algebroid curves in ${\Bbb C}^{2}$. In the following section we will examine a class of periodic orbits near which the analytic representation of the local model above can be achieved in a similar manner.
Asymptotic approximation near a periodic orbit
==============================================
Consider a periodic orbit of the Reeb flow, denoted ${\cal P}$, and a tubular neighbourhood $T_{\cal P}\subset M$. If $\Delta$ represents a disc centred at the origin in ${\Bbb R}^{2}$, let $\tilde{\Delta}\subset M$ be an embedded image such that the origin is mapped to the unique element $p_{0}$ of ${\cal P}\cap\tilde{\Delta}$, with $\tilde{\Delta}$ itself corresponding to a transverse slice of $T_{\cal P}$. The Reeb flow will be assumed moreover to be Lyapunov–stable near ${\cal P}$ in the sense that for all $p\in\tilde{\Delta'}$, where $\Delta'\subseteq\Delta$ is a sufficiently small disc centred at the origin, there exists a unique solution $\gamma_{p}:[0,\infty)@>>>M$ to the equation $$\frac{d\gamma_{p}}{dt} = X_{\lambda}(\gamma_{p}(t)) \ , \ \hspace{.2in}
\gamma_{p}(0) = p \ ,$$ which depends smoothly on both $t$ and $p$, and remains inside $T_{\cal P}$ for all $t\geq 0$. Given $p\in\tilde{\Delta'}$, we will define $(i)
\ \tau(p)$ to be the smallest $t>0$ such that $\gamma_{p}(t)\in
\tilde{\Delta} \ , \ (ii) \ \Gamma_{p} := \gamma_{p}((0,\tau(p)])$ and for each connected open neighbourhood of the origin $\Omega
\subseteq\tilde{\Delta}$ , $$(iii) \ \Gamma(\Omega) := \cup_{p\in\Omega\times\{\theta_{0}\}}
\Gamma_{p} \ .$$ We may now consider a recursively defined system of neighbourhoods $\{\Omega_{k}\}$, such that $\Omega_{0}:=\tilde{\Delta'}$, while $\Omega_{k}$ denotes the origin–component of $\Gamma(\Omega_{k-1})\cap\Omega_{k-1}$. The set $\Omega_{\infty}:=\cap_{k=0}^{\infty}\Omega_{k}$ measures an important dynamic aspect of the Reeb flow.
If $\Omega_{\infty}$ is open, then it is conformally equivalent to a disc.
Conformal equivalence to a disc will follow immediately if $\Omega_{\infty}$ can be shown to be simply connected, which we now prove by induction : $\Omega_{0} = \tilde{\Delta'}$. Suppose $\Omega_{k}$ is simply connected, and let $C\subset\Omega_{k+1}$ be a simple closed loop. By the existence and uniqueness of ordinary differential equations $C$ can be traced back under the Reeb flow to a simple closed loop $C'$ which bounds a contractible subdomain, say ${\cal U}\subset\Omega_{k}$. Once again, existence and uniqueness ensures that for all $p\in{\cal U}$ we have $\Gamma_{p}\cap\Gamma(C') = \emptyset$, while the Jordan Curve Theorem for plane domains implies that the domain bounded by a given closed simple loop is unique. It follows that $$C = \Gamma(\partial{\cal U})\cap\Omega_{k} = \partial\Gamma({\cal U})
\cap\Omega_{k} = \partial(\Gamma({\cal U})\cap\Omega_{k}) \ .$$ Moreover, $\Gamma({\cal U})\cap\Omega_{k}$ is a homeomorphic image of ${\cal U}$, and is therefore itself a contractible subdomain, hence $\Omega_{k+1}$ is simply connected.
Now consider a simple closed loop $C\subset\Omega_{\infty}$, i.e., $C\subset\Omega_{k}$ for all $k$, and hence there exists a contractible subdomain ${\cal U}_{k}$ in $\Omega_{k}$ such that $\partial{\cal U}_{k}
= C$. But uniqueness of the interior of $C$ of course implies that ${\cal U}_{k}\equiv{\cal U}$ for all $k$, and hence ${\cal U}\subset
\Omega_{\infty}$.
The Reeb flow will be said to be “locally recurrent” near a periodic orbit ${\cal P}$ if it is Lyapunov–stable within a tubular neighbourhood $T_{\cal P}$ and for any sufficiently small embedded disc $\tilde{\Delta}$, corresponding to a transversal slice through $T_{\cal P}$ at some point $p_{0}$, the limit set $\Omega_{\infty}\subseteq\tilde{\Delta'}\subseteq\tilde
{\Delta}$ is open. An orbit ${\cal P}$ itself may also be referred to as “locally recurrent” in this context.
As mentioned in section 1, the treatment of a finite–energy pseudoholomorphic map $(\psi,a):\Sigma@>>>M\times{\Bbb R}$ from a punctured Riemann surface, asymptotic to a periodic orbit, may be restricted to a map between $D\setminus\{0\}\subset{\Bbb C}$ and, for convenience, the Martinet neighbourhood $\Delta\times
{\Bbb S}^{1}\times[a_{0},\infty)$ of ${\cal P}$, with coordinates $(x,y,\theta,a)$ (cf. \[8\], \[10\], \[11\]). In particular, for $p_{0}$ corresponding to the origin of $\Delta$, we have $\gamma_{0}(t) =
(0,0,e^{{\bold i}(\theta_{0}+2\pi t/\tau_{0})})$, where $\tau_{0}$ denotes the minimal period of ${\cal P}$. In relation to the asymptotic results of \[8\], it will sometimes be convenient to use cylindrical coordinates on $D\setminus\{0\}$, viz $(r,\varphi) = (-\ln(|z|),
arg(z))\in [r_{0},\infty)\times{\Bbb S}^{1}$. In particular, the [*charge*]{} of the puncture is given by $$T:= \frac{1}{2\pi}\lim_{r@>>>\infty}\int_{{\Bbb S}^{1}}{\psi}^{*}
\lambda \ ,$$ and is a well–defined integer multiple of $\tau_{0}$ (cf. \[8\]). Within this coordinate system, the asymptotic relations of Hofer, Wysocki and Zehnder (mentioned in section one) may be interpreted as $$a(r,\varphi) = Tr + a_{0} + \varepsilon(r,\varphi) \ ,$$ $$\theta(r,\varphi) = T(\frac{\varphi}{\tau_{0}}) + \theta_{0} +
\delta(r,\varphi) \ ,$$ with $\varepsilon \ , \ \delta$ approaching zero uniformly in $\varphi$ as $r@>>>\infty$. For $|z|$ sufficiently small (i.e., $r$ sufficiently large) we may think of $\theta^{-1}(\theta_{0})$ as a union of radial arcs meeting at the origin and differing by only a small perturbation from the rays defined by $arg(z) = \frac{1}{n}(\theta_{0} + 2\pi k) \ ,
\ 0\leq k\leq n-1$, where $n = \frac{T}{\tau_{0}}$. Hence we will consider $D\setminus\{0\}$ as a union of $n$ [*quasi–sectors*]{} $Q_{k}$, bounded by these arcs, on which the quasi–analytic “branches” of $(\psi,a)$ are defined.
For the local model in section 2 it was sufficient to assume that $\psi$ is a local embedding in order to represent the image as a graph on which the restriction of an almost complex structure $J$ depends on just two of the coordinates of a Darboux chart. Under the assumption that the Reeb flow is locally recurrent near ${\cal P}$, we now select $\Omega_{\infty}\times\{\theta_{0}\}$ as coordinate disc within the initial Martinet tube (on which $\lambda
= f\cdot(d\vartheta +xdy)$, for a function $f$ such that $f(0,0,\vartheta)
\equiv\tau_{0}$ and $\nabla f(0,0,\vartheta)\equiv{\bold 0}$ as described in \[8\],\[11\], though these facts are not used here). Without loss of generality, let $\theta_{0}$ be zero and consider the cylinder $\Omega_{\infty}\times[0,2\pi]$, which maps to the tube via the obvious identification $mod(2\pi)$. The cylinder has ${\cal P}$ as its axis, $x=y=0$, and the Reeb vector field in Martinet coordinates already looks like $\frac{1}{\tau_{0}}\frac{\partial}{\partial\theta}$ when restricted to ${\cal P}$. There is no consequent loss of generality if we “normalise” $\lambda$ by the constant multiple $\frac{1}{\tau_{0}}$, so that the minimal period is effectively $1$, and hence the charge $T$ is an integer. By analogy with the standard construction of Darboux coordinates, the next step is to define $${\cal C}:=\{(p,t) \ | \ p=(x,y)\in\Omega_{\infty} \ , \ \hspace{
.1in} 0\leq t\leq\tau(p) \ \} \ ,$$ and a homeomorphism $$h:{\cal C}@>>>\Omega_{\infty}\times[0,2\pi] \ , \hspace{.1in}
h\mid_{\Omega_{\infty}\times\{0\}} = 1 \ ,$$ which is smooth for all $0<t<\tau(p)$, coming from solutions of the ordinary differential equation $$\frac{d\gamma_{p}}{dt} = X_{\lambda}(\gamma_{p}(t)) \ .$$ It follows that on the interior of ${\cal C}$, the standard contact form $\lambda_{0}$ and $\lambda':= h^{*}\lambda$ have the same Reeb vector field, corresponding to $\frac{\partial}{\partial t}$. We now consider the Cauchy–Riemann system $$\pi((h^{-1}\psi)_{\eta}) + J\pi((h^{-1}\psi)_{\zeta}) = 0 \
\hspace{.1in}(**) \ ,$$ $$\lambda'((h^{-1}\psi)_{\zeta}) = -a_{\eta} \ \hspace{.1in} \
\lambda'((h^{-1}\psi)_{\eta}) = a_{\zeta} \ \hspace{.1in}
(\dag\dag) \ .$$ As in the local model, for a sufficiently “thin” neighbourhood of ${\cal P}$, the standard projection $(v_{1},v_{2},v_{3})\mapsto (v_{1},
v_{2})$ determines a linear isomorphism $\mu$ between $\xi' :=
ker(\lambda')$ and ${\Bbb R}^{2}$. Hence we define a $2\times 2$ matrix–valued function $j(x,y) = \mu\circ J\circ\mu^{-1}$, such that ${\bold x}:=(x,y)$ implies (\*\*) can be written in the form $${\bold x}_{\eta}(z) + j{\bold x}_{\zeta}(z) = 0 \ .$$ Let $\alpha$ denote the diffeomorphism of $\Omega_{\infty}
\times\{\theta_{0}\}$ defined by the return map $\alpha(p):=
\gamma_{p}(\tau(p))$, hence $\alpha(0) = 0$.
If ${\cal L}_{X_{\lambda}}J = 0$, then in a neighbourhood of $0\in\Delta$, the smooth automorphism $\alpha$ is diffeomorphically equivalent to a rotation.
Letting ${\bold e}_{1}$ denote the vector $(1,0)$, consider the system of ordinary differential equations $$\frac{d{\bold x}}{ds} = j({\bold x})\cdot{\bold e}_{1} \ ,$$ and define a local diffeomorphism $\varphi:\Delta''@>>>U\subseteq
\Omega_{\infty}$ via existence and uniqueness of solutions, such that $$(\dot{x}_{1},\dot{x}_{2}) = j({\bold x})\cdot{\bold e}_{1} \
\Rightarrow\varphi(x,s) := (x_{1}(s),x_{2}(s)) \ ,$$ $\varphi(x,0) = (x,0) \ .$ Hence $$\varphi_{*}^{-1}\circ j\circ\varphi_{*} = j_{0} =
\left(\begin{array}{cc}
0 & -1 \\
1 & 0
\end{array}\right) \ .$$
Let $\Omega_{\infty}'$ denote the simply connected domain inside $U$ which is stabilised by the Reeb flow. The diffeomorphism $\hat{\alpha}:= \varphi^{-1}\circ\alpha
\circ\varphi$ then acts on $\varphi^{-1}(\Omega_{\infty}')\subseteq
\Delta''$ as a restricted automorphism such that $\hat{\alpha}(0) = 0$ under the assumption of local recurrence. The additional assumption ${\cal L}_{X_{\lambda}}J = 0$ implies that $\alpha^{*}j = j$, hence in particular $\hat{\alpha}j_{0} = j_{0}\hat{\alpha}$, i.e., $\hat{\alpha}$ is a conformal automorphism. Modulo a conformal transformation identifying $\varphi^{-1}(\Omega_{\infty}')$ with a disc, $\hat{\alpha}$ is then equivalent to a rotation.
Now $\tilde{\bold x}_{\zeta}:=\varphi_{*}^{-1}({\bold x}_{\zeta})$ and $\tilde{\bold x}_{\eta}:=\varphi_{*}^{-1}({\bold x}_{\eta})$ implies $$\tilde{\bold x}_{\eta}(\eta,\zeta) + j_{0}\cdot
\tilde{\bold x}_{\zeta}(\eta,\zeta) = {\bold 0} \ (\dag *) .$$ Note that each “branch” of $$\Psi:=(h\circ(\varphi\times 1))^{-1}\psi$$ is defined smoothly in the interior and continuously up to the boundaries of a quasi–sector $Q_{k}$ in $D\setminus\{0\}$, but discontinuities arise at points $z_{0}$ lying on the smooth arcs that bound adjacent sectors (in the usual way “$\pm$” will be used to denote opposite sides of the boundary). Discontinuities of the transverse projection of $\Psi$ are therefore described by the relations $$\lim_{z@>>>z_{0}^{\pm}}\tilde{{\bold x}}(z):= \tilde{{\bold x}}^{\pm}
(z_{0}) \ \Rightarrow \ \hat{\alpha}(\tilde{{\bold x}}^{-}(z_{0})) =
\tilde{{\bold x}}^{+}(z_{0}) \ .$$ Hence on each $Q_{k}\subset D\setminus\{0\} \ , \ (\dag *)$ defines a holomorphic function $w=F_{k}(z)$ which partially describes a branch of $\Psi$, such that $$F_{k}\mid_{\overline{Q}_{k}\cap\overline{Q}_{k+1}}= \hat{\alpha}
\circ F_{k+1}\mid_
{\overline{Q}_{k}\cap\overline{Q}_{k+1}} \ ,\ \hspace{.1in} 0\leq
k\leq n-1 \ .$$
If the Reeb flow determines a fibration of a tubular neighbourhood by periodic orbits over $\Omega_{\infty}$, i.e., $\alpha = 1$, then there is a single holomorphic function $F(z)$ on $D$ , $F(0) = 0$ , describing the transverse projection of $\Psi$.
$\alpha = 1$ implies the existence of a function $F(z)$, continuous on $D$ and holomorphic on the interior of each quasi–sector. The demonstration that $F$ is holomorphic on $D$ is a standard application of Morera’s Theorem (and the Removable Singularities Theorem at the origin). Specifically, let $z_{0}$ lie on one of the smooth arcs bounding a quasi–sector and let $z' = \eta' + {\bold i}\zeta'$ be a local complex coordinate with respect to which the arc is locally described as the graph of a function $\zeta' = \rho(\eta')$. Let $\Gamma$ be a simple loop inside the $z'$–coordinate neighbourhood of $z_{0}$. Clearly if $\Gamma$ does not intersect the arc, then $F$ must be holomorphic on a slightly larger domain containing $\Gamma$, hence $$\int_{\Gamma}F \ dz' = 0 \ .$$ If $\Gamma$ intersects the arc, consider a simple affine coordinate transformation so that the cord joining the endpoints $a,b$ of the intersection is now the axis corresponding to $\zeta' = 0$. Hence, without loss of generality, we may assume $\rho(a) = \rho(b) = 0$. Let $\gamma$ denote the arc corresponding to the graph between $a$ and $b$, and for $\varepsilon > 0$ let $\gamma_{\pm\varepsilon}$ denote the arcs corresponding to $\zeta' = \rho(\eta')\pm\varepsilon$, lying between $a\pm{\bold i}\varepsilon$ and $b\pm{\bold i}\varepsilon$ and on either side of $\gamma$. From the continuity of $F$ it follows that $$\int_{\gamma}F \ dz' = \int_{a}^{b}F(\eta' + {\bold i}\rho(\eta'))
d\eta' = \lim_{\varepsilon @>>> 0}\int_{a}^{b}F(\eta' + {\bold i}
(\rho(\eta') \pm \varepsilon)) \ d\eta'$$ $$\hspace{.4in} = \lim_{\varepsilon @>>> 0}
\int_{\gamma_{\varepsilon}}F \ dz' \ .$$ Now decomposing $\Gamma$ into two simple loops $\Gamma_{\pm}$ having $\gamma$ as their common boundary component, and applying the above limit, we see that $$\int_{\Gamma}F \ dz' = 0 \ .$$ Hence $F$ extends holomorphically to a neighbourhood of $z_{0}$.
Returning now to the particular form of the equations $(\dag\dag)$ in ${\cal C}$ note that any $\lambda'$ with $\frac{\partial}{\partial t}$ as its Reeb vector field must take the general form $$\lambda' = dt + f_{1}({\bold x})dx + f_{2}({\bold x})dy \ .$$ As was seen with respect to a Darboux chart of the local model, such a local presentation of the contact form allows decoupling of $(\dag\dag)$ into an inhomogeneous Cauchy–Riemann equation. This property is preserved under the diffeomorphism $\varphi$, however, if it is assumed that ${\cal L}_{X_{\lambda}}
J = 0$, hence in particular the matrix $j$ above is independent of $t$. In this case, letting $u = t+{\bold i}a$, $(\dag\dag)$ becomes
$$\frac{\partial}{\partial\bar{z}}[u\mid_{Q_{k}}] = -\{f_{1}(F_{k}(z))
\frac{\partial\Re(F_{k})}{\partial\bar{z}} + f_{2}(F_{k}(z))
\frac{\partial\Im(F_{k})}{\partial\bar{z}}\}$$ $$= -\frac{1}{2}[(f_{1} +{\bold i}f_{2})
\circ F_{k}(z)]\cdot\overline{F'_{k}(z)} \ , \$$ keeping in mind that this equation is defined smoothly only on the interior of each quasi–sector $Q_{k}$. Define $\omega := \frac{1}{2}(f_{1}+{\bold i}f_{2})(w,\bar{w})d\bar{w}$, so that $$\lambda' = dt + 2\Re(\omega) \ , \ \hbox{and}$$ $$\frac{1}{2}[(f_{1} +{\bold i}f_{2})\circ F_{k}(z)]\cdot\overline{F'_{k}(z)}
d\bar{z} = F_{k}^{*}\omega \ .$$ Now ${\cal L}_{X_{\lambda}}\lambda = 0$ implies $\hat{\alpha}^{*}
\Re(\omega) = \Re(\omega)$. In particular, ${\bf f}:= (f_{1},f_{2})$ implies $\Re(\omega) = ({\bf f},*)$ with respect to the standard inner product on ${\Bbb R}^{2}$, and hence $\hat{\alpha}_{*}^{t}{\bf f} =
{\bf f}$. Similarly $\Im(\omega) = (j_{0}{\bf f},*)$, while ${\cal L}_{X_{\lambda}}J = 0$ implies $\hat{\alpha}_{*}^{t}j_{0} =
j_{0}\hat{\alpha}_{*}^{t}$, so that $$\hat{\alpha}^{*}\Im(\omega) =
(\hat{\alpha}_{*}^{t}j_{0}{\bf f},*) = (j_{0}{\bf f},*) = \Im(\omega) \ .$$ It follows that $\hat{\alpha}^{*}\omega = \omega$, and hence $$F_{k}^{*}\omega\mid_{\overline{Q}_{k}\cap\overline{Q}_{k+1}} =
F_{k}^{*}(\hat{\alpha}^{*}\omega)\mid_{\overline{Q}_{k}\cap\overline
{Q}_{k+1}} =
(\hat{\alpha}\circ F_{k})^{*}\omega\mid_{\overline{Q}_{k}\cap
\overline{Q}_{k+1}}$$ $$= F_{k+1}^{*}\omega\mid_{\overline{Q}_{k}\cap\overline{Q}_{k+1}} \ .$$ There now exists a continuous function $G(z,\bar{z})$ on $D$ such that $$G(z,\bar{z})d\bar{z}\mid_{Q_{k}}:= F_{k}^{*}\omega \ , \ \hspace{.1in}
0\leq k\leq n-1 \ .$$ Letting $\Phi(w,\bar{w}) = f_{1}+{\bold i}f_{2}$ and assuming, from lemma 1, that $\hat{\alpha}$ is a rotation, we may write $$\hat{\alpha}^{*}\omega = \omega \ \Rightarrow \Phi\circ\hat{\alpha}
(w,\bar{w})
= \overline{\hat{\alpha}^{-1}}\Phi(w,\bar{w}) \ .$$ It is easily seen that $$\Phi_{w}\circ\hat{\alpha}(w,\bar{w}) = |\hat{\alpha}|^{-2}\Phi_{w} \ ,
\hspace{.1in}
\Phi_{\bar{w}}\circ\hat{\alpha}(w,\bar{w}) = \overline{\hat{\alpha}^{-2}}
\Phi_{\bar{w}} \ ,$$ and hence $$\frac{\partial G}{\partial z}\mid_{\overline{Q}_{k}\cap\overline{Q}_{k+1}}
= \Phi_{w}(F_{k+1})\cdot |F_{k+1}'|^{2}$$ $$= \Phi_{w}(\hat{\alpha}\cdot F_{k})
\cdot|\hat{\alpha}\cdot F_{k}'|^{2} = \Phi_{w}(F_{k})\cdot|F_{k}'|^{2} \ ,$$ while $$\frac{\partial G}{\partial \bar{z}}\mid_{\overline{Q}_{k}\cap\overline
{Q}_{k+1}}
= \Phi_{\bar{w}}(F_{k+1})\cdot(\bar{F}_{k+1}')^{2} + \Phi(F_{k+1})\cdot
\bar{F}_{k+1}''$$ $$= \Phi_{\bar{w}}(\hat{\alpha}\cdot F_{k})
\cdot(\overline{\hat{\alpha}\cdot F}_{k}')^{2} + \Phi(\hat{\alpha}\cdot
F_{k})\cdot\overline{\hat{\alpha}\cdot F}_{k}''$$ $$= \Phi_{\bar{w}}(F_{k})\cdot
(\bar{F}_{k}')^{2} + \Phi(F_{k})\cdot\bar{F}_{k}'' \ .$$ Thus $G$ is a continuously differentiable function, and $$\hat{G}(z,\bar{z}) := \int_{D}\frac{G(\mu,\bar{\mu})}{\mu-z}
d\mu\wedge d\bar{\mu}$$ is a twice-continuously differentiable function on $D$. We may therefore write the solutions $u = H_{k}(z) - \frac{1}{2\pi{\bold i}}\hat{G}(z,\bar{z})$ for a holomorphic function $H_{k}$ defined on each $Q_{k}$. Now $a(z) = \Im(u)$ is a smooth function on $D\setminus\{0\}$, while $\hat{G}$ belongs to $C^{2}(D)$, hence $\Im(H_{k}(z))$ must in fact correspond to a single function $h(z)$ for all $k$, which belongs to $C^{2}(D\setminus\{0\})$ and is harmonic inside each $Q_{k}$, therefore harmonic throughout $D\setminus\{0\}$. Moreover, the harmonic conjugate of $h$ is uniquely defined up to a constant, hence we have a single harmonic function $\hat{h}$ on the punctured disc such that $$\Re(H_{k}(z)) = \hat{h}(z) + c_{k} \ , \ \hspace{.1in} 0\leq k\leq
n-1 \ .$$
Recall that $\Re(u)$ corresponds to $t$ such that $0<t<\tau(p)$ for some $(p,t)$ belonging to the image of $\Psi$. Discontinuities of $t$ along the boundaries of each $Q_{k}$ are consequently determined by $\tau(p)$, i.e., $$\lim_{z@>>>z_{0}^{-}}\hat{h}_{k}(z) = \lim_{z@>>>z_{0}^{+}}
\hat{h}_{k+1}(z) + \tau(\tilde{{\bold x}}^{-}(z_{0}))$$ for $z_{0}$ lying in the boundary arc $\overline{Q}_{k}\cap\overline{
Q}_{k+1}$. From the discussion above it follows that $\tau(\tilde{{\bold x}}^{-}(z))$ must be constant for $\Psi$ restricted to a boundary arc. By continuity of these arcs as they radiate from the origin of $D$, this constant value must correspond to $\tau_{0}$, or $1$ if $\lambda$ is assumed to have been normalised. The arcs themselves were originally defined by the arbitrary choice of $\theta_{0}\in{\Bbb
S}^{1}$, hence we may conclude that $\tau(p) = 1$ for all $p\in\Omega_
{\infty}'$. Moreover, we have $$c_{k+1} - c_{k} = 1 \ , \ \hspace{.1in} 0\leq k\leq n-1 \ ,$$ where, without loss of generality, we may set $c_{0} = 0$, hence $c_{k} = k$. Recalling the asymptotic formulae $a(z) = -\ln|z|^{n} + \varepsilon(z)$ and $\theta(z) = arg(z^{n}) + \delta(z)$, one may notice also the approximation of the holomorphic part of $u$ by the branched analytic function $log(z^{n})$. Specifically, $$u = \frac{1}{2\pi{\bold i}}\{\log(\rho(z)e^{-\hat{G}(z,\bar{z})})\} \ ,$$ where the analytic function $\rho(z):= e^{2\pi H(z)}$ has order $n$ at $z = 0$, and $K:=sup_{D}|G(z,\bar{z})|$ implies $$\frac{1}{2\pi}|\hat{G}(z,\bar{z})| = \frac{1}{2\pi}|\int_{D}
\frac{G(\mu,\bar{\mu})}{\mu-z}d\mu\wedge d\bar{\mu}|\leq K|z|$$ is uniformly bounded in the parameter $arg(z)$, i.e., $$\lim_{r@>>>\infty}|\hat{G}(r,\varphi)| = 0\hspace{.1in}\hbox{in}
\hspace{.1in} C^{0}({\Bbb S}^{1}) \ .$$ In summary, we have the following
Let $(\psi,a):D\setminus\{0\}@>>>M\times{\Bbb R}$ be a $J$–holomorphic curve of finite energy and charge $n$ at $z=0$, asymptotic to a locally recurrent periodic orbit ${\cal P}$, near which ${\cal L}_{X_{\lambda}}J = 0$. Consider any tubular neighbourhood of ${\cal P}$ in $M$, diffeomorphic to $\Delta\times{\Bbb S}^{1}$ such that $\{0\}\times{\Bbb S}^{1}\approx{\cal P}$. There exists a diffeomorphic change of coordinates in $\Delta\times[0,2\pi)$ such that on each quasi–sector $Q_{k}\subset D\setminus\{0\}$ the map $(\psi,a)$ can be expressed in the form $$(F_{k}(z), H_{k}(z) - \frac{1}{2\pi{\bold i}}\hat{G}(z,\bar{z}))\ , \
\hspace{.1in} 0\leq k\leq n-1 ,$$ where $F_{k} \ , \ H_{k}$ are holomorphic on $Q_{k}$ and continuous on $\overline{Q}_{k}$, such that $$F_{k}\mid_{\overline{Q}_{k}\cap\overline{Q}_{k+1}} = \hat{\alpha}
\circ F_{k+1}\mid_{\overline{Q}_{k}\cap\overline{Q}_{k+1}} \ ,$$ while each $H_{k}$ corresponds to an analytic branch of $\frac{1}{2\pi{\bold i}}\log(\rho) \ , \
ord_{0}(\rho) = n$. Moreover, the function $\hat{G}$ belongs to $C^{2}(D)$ and is bounded by $K|z|$. Finally, if $\alpha = 1$, then each $F_{k}$ is the restriction of a single function $F$ holomorphic on $D$, $F(0) = 0$.
A representation corresponding to the classical local parametrization of plane algebroid curves is easily obtained as follows for the case $\alpha = 1$. Let $$g(w,\bar{w}) := \frac{1}{2\pi{\bold i}}\int_{\Delta'}\frac{(f_{1}
+{\bold i}f_{2})(\mu,\bar{\mu})d\mu\wedge d\bar{\mu}}{\mu - w} \ ,$$ so that $\bar{\partial}g = \omega$. Then $$G(z,\bar{z})d\bar{z} = F^{*}\omega = F^{*}(\bar{\partial}g)$$ $$= \bar{\partial}(F^{*}g) = \bar{\partial}(g\circ F) \ ,$$ and hence $\hat{G} = 2\pi{\bold i}(g\circ F + \hat{H})$, for some holomorphic function $\hat{H}$. Now define a coordinate $$v:= e^{2\pi{\bold i}u} = \rho(z)e^{-\hat{G}} = \rho(z)e^{-g\circ F
- \hat{H}}$$ and let $\rho(z)e^{-\hat{H}} = f(z)^{n}$ for some holomorphic function $f$ on $D'\subseteq D$, with $f'(0)\neq 0$. Now $\xi:=f(z)$ implies $w = F(z)
= F\circ f^{-1}(\xi)$, while $v = \xi^{n}e^{-g\circ F\circ f^{-1}(\xi)}$. From the final coordinate diffeomorphism $w' = w \ ; \ v' = ve^{g(w,
\bar{w})}$, it now follows that $$w' = F\circ f^{-1}(\xi) \ ; \ \hspace{.1in} v' = \xi^{n} \ .$$
$\alpha = 1$: Circle fibrations of ${\Bbb S}^{3}$
=================================================
A class of examples of tight contact structures for which $\alpha = 1$ and ${\cal L}_{X_{\lambda}}J = 0$ near a periodic orbit is provided by the circle fibrations of ${\Bbb S}^{3}$, of which the most elementary instances are the Hopf fibrations. Given a four–dimensional real vector space $V$, these fibrations correspond to families of invariant planes (i.e., “complex lines”) distinguished by linear endomorphisms $J_{0}$ determining standard complex structures on $V$, and are parametrised by ${\Bbb O}(4)/{\Bbb U}(2)$. The base space of each such fibration is a (Riemann) sphere inside the Grassmann manifold ${\Bbb G}r_{2}(V)$. More general fibrations correspond to families of planes (“line congruences”) parametrised by compact surfaces $\Sigma\subset{\Bbb G}
r_{2}(V)$. A line congruence is said to be [*elliptic*]{} if for all $P\in\Sigma$, there exists a 2–sphere ${\Bbb S}$ corresponding to some $J_{0}$ such that $T_{P}\Sigma = T_{P}{\Bbb S}$ inside ${\Bbb G}r_{2}(V)$, hence in particular $\Sigma$ is itself diffeomorphic to a sphere. Let $\bigwedge^{2} V$ denote the space of exterior 2-forms, on which the duality operator acts in the standard way. The spaces of “self-dual” and “anti–self–dual” forms then correspond to +1 and –1 eigenspaces of this operator, defining a direct sum decomposition $\bigwedge^{2} V\cong\bigwedge^{2}_{+}\bigoplus\bigwedge^{2}_{-} \ .$ If ${\Bbb S}_{+}$ and ${\Bbb S}_{-}$ denote the 2–spheres of radius $\frac{1}{\sqrt{2}}$ inside each of these eigenspaces, then the Grassmann manifold of oriented 2–planes of $V$ is well–known to correspond to ${\Bbb S}_{+}\times{\Bbb S}_{-}$. Moreover, it was shown by Gluck and Warner \[4\] that the surfaces $\Sigma$ of generalised Hopf fibrations are precisely the graphs of distance–decreasing smooth maps $f:{\Bbb S}_{-}@>>>{\Bbb S}_{+}$, with standard Hopf fibrations coresponding to constant maps. In \[13\], McKay also observed that they correspond to general elliptic line congruences and hence determine non–linear complex structures on $V$ which “osculate” with linear structures at each $P\in\Sigma$. We apply this idea to the explicit construction of contact structures on ${\Bbb S}^{3}$ as follows. Let ${\cal J}(V)$ denote the space of linear endomorphisms of $V$ corresponding to linear complex structures, then it was shown in \[13\] that each elliptic line congruence $\Sigma$ determines a map $$J:\Sigma @>>> {\cal J}(V)\subset V\otimes V^{*}$$ such that for each plane $P\in\Sigma$, $J(P)$ is linear and is the “osculating” structure to $\Sigma$ at $P$ in the sense that both $P$ and $P^{\perp}$ (with respect to a given inner product on $V$) are complex lines relative to $J(P)$. The family of planes determined by $\Sigma$ describes a rank–two vector bundle ${\cal P}
@>\pi>>\Sigma$ such that the total space, corresponding to the incidence manifold ${\cal P} = \{({\bold v},P)\in V\times\Sigma
\ | \ {\bold v}\in P\}$ also maps surjectively to $V$. In fact, there exists ${\cal P}@>\sigma>>V$ such that ${\cal P}\setminus\sigma^{-1}(0)
\cong V\setminus\{0\}$. Hence define $\varphi:=\pi\circ\sigma^{-1}:V\setminus
\{0\}@>>>\Sigma$, so that $\varphi^{-1}(P) = \{{\bold v}\in V\setminus
\{0\} \ | \ {\bold v}\in P\}$ .
Consider the pullback $\varphi^{*}J:V\setminus\{0\}@>>>V\otimes V^{*}$. Hence with respect to a designated orthonormal basis of $V$, noting that $J$ and $\varphi^{*}J$ are skew–symmetric matrix–valued functions, we may represent it in the form $$\varphi^{*}J =\Sigma_{\mu.\nu}J^{\mu\nu}\frac{\partial}{\partial x
_{\mu}}\wedge dx_{\nu} \ .$$ Note moreover that the isomorphism $V^{*}\cong V$ via the Euclidean inner product allows us to define a 2–form $\omega_{\Sigma}:= \Sigma_{\mu.\nu}J^{\mu\nu}dx_{\mu}\wedge dx_{\nu}$. In the following, let $i_{\bold v}$ denote contraction of a form by the position vector ${\bold v} = \frac{1}{2}\Sigma_{\eta}v_{\eta}\frac{
\partial}{\partial\eta}$ .
If $d\omega_{\Sigma} = 0$ then $\omega_{\Sigma}({\bold v})
= d(i_{\bold v}\omega_{\Sigma})$ .
$$i_{\bold v}\omega_{\Sigma} = \frac{1}{2}\Sigma_{\mu < \nu}J^{\mu\nu}
(x_{\mu}dx_{\nu} - x_{\nu}dx_{\mu})$$ $$= \frac{1}{2}\Sigma_{\beta\neq\alpha}(-1)^{\varepsilon}x_{\beta}
J^{\alpha\beta}dx_{\alpha} \ ,$$ where $$\varepsilon =
\left\{\begin{array}{l} 0 \hspace{.1in} \beta < \alpha \\
1 \hspace{.1in} \beta > \alpha \ . \\
\end{array} \right. \\$$
Therefore $$d(i_{\bold v}\omega_{\Sigma}) = \omega_{\Sigma} + \frac{1}{2}\Sigma_
{\gamma < \alpha\neq\beta}(-1)^{\varepsilon}x_{\beta}(\frac{\partial
J^{\alpha\beta}}{\partial x_{\gamma}} - \frac{\partial
J^{\gamma\beta}}{\partial x_{\alpha}})dx_{\gamma}\wedge dx_{\alpha}$$ $$= \omega_{\Sigma} + \frac{1}{2}\Sigma_{\gamma < \alpha\neq\beta}
x_{\beta}\frac{\partial
J^{\gamma\alpha}}{\partial x_{\beta}}dx_{\gamma}\wedge dx_{\alpha}$$ (using the relations provided by $d\omega_{\Sigma} = 0$) $$= \omega_{\Sigma} + \frac{1}{2}\Sigma_{\gamma < \alpha}
(\nabla_{\bold v}J^{\gamma\alpha})dx_{\gamma}\wedge dx_{\alpha} \ .$$ Note that the functions $J^{\gamma\alpha}$, obtained by pulling back $J$, are constant in the radial directions of $V$, hence the directional derivatives $\nabla_{\bold v}J^{\gamma\alpha} = 0$ for all ${\bold v}
\in V\setminus\{0\}$ .
Now define the 1–form $\lambda_{\Sigma}:= i_{\bold v}\omega_{\Sigma}$ for $\omega_{\Sigma}$ closed, noting that if $[J]$ denotes the matrix of $\varphi^{*}J$, then we may express $\lambda_{\Sigma}$ in terms of the inner product as $$\lambda_{\Sigma}({\bold w}) = - ({\bold w},[J]\cdot{\bold v})$$ for all ${\bold w}\in V$. Moreover, let $X_{\lambda}$ be the vector field defined by $X_{\lambda}({\bold v}) = -[J]\cdot{\bold v}$, so that
$(i) \ X_{\lambda}$ is tangent to ${\Bbb S}^{3}$, since $$({\bold v}, X_{\lambda}({\bold v})) = -({\bold v},[J]\cdot{\bold v})
= -([J]^{t}\cdot{\bold v},{\bold v}) = ([J]\cdot{\bold v},{\bold v})
\ ,$$ hence $({\bold v}, X_{\lambda}({\bold v})) = 0$ ,
$(ii) \ \lambda_{\Sigma}(X_{\lambda}) = |[J]\cdot{\bold v}|^{2} = 1$ , since $[J]_{\bold v}\in{\Bbb O}(4)$ for each ${\bold v}\in
V\setminus\{0\}$, and
$(iii)$ for all ${\bold w} \ , \ {\bold u}\in V$ we have $$\omega_{\Sigma}({\bold w},{\bold u}) = \Sigma_{\mu < \nu} J^{\mu\nu}
(w_{\mu}u_{\nu} - w_{\nu}u_{\mu}) = ({\bold w},[J]\cdot{\bold u}) \ ,$$ and hence $$i_{X_{\lambda}}\omega_{\Sigma} = -(*,[J]^{2}\cdot{\bold v}) = (*,{\bold v})
= 0 \$$ when restricted to $T{\Bbb S}^{3}$. Clearly $\varphi^{*}J$ is preserved by the Reeb flow, i.e., ${\cal L}_{X_{\lambda}}\varphi^{*}J
= 0$. Moreover, if $P_{\bold v}$ denotes the subspace spanned by $\{{\bold v} \ , \ [J]\cdot{\bold v}\}$, then $ker(\lambda)\cap T{\Bbb S}^{3} = P_{\bold v}^{\perp}$ is also an invariant subspace of $[J]_{\bold v}$, so we may write $j:= [J]\mid_{P_{\bold v}^
{\perp}}$ for all ${\bold v}\in{\Bbb S}^{3}$. In particular, for all $\xi\in ker(\lambda)$, we have $\omega_{\Sigma}(\xi,j\cdot\xi) =
-|\xi|^{2}$, which means that $\lambda_{\Sigma}$ is a contact structure compatible with the partial complex structure $j$.
It should be mentioned that all Hopf fibrations, including the non–linear ones, are smoothly equivalent as circle bundles (cf. \[4\]), and yet at the level of contact structures they are distinct. For suppose $\delta:
{\Bbb S}^{3}@>>>{\Bbb S}^{3}$ is a diffeomorphism that identifies the Reeb flows of a given structure $\lambda_{\Sigma}$ and that of the standard structure $\lambda_{0}$. In particular, suppose that $\delta_{*}
X_{\lambda} = X_{\lambda_{0}}$, and $\delta_{*}\circ \varphi^{*}J
\circ\delta_{*}^{-1} = J_{0}$, where $J$ denotes the osculating complex structure associated with $\lambda_{\Sigma}$. Now assume in addition that $\delta^{*}\lambda
_{0} = \lambda_{\Sigma}$, hence $$\delta^{*}\lambda_{0}({\bold u}) = \lambda_{0}(\delta_{*}{\bold u})
= (\delta_{*}{\bold u} \ , \ J_{0}\cdot{\bold v}) = \lambda_{\Sigma}
({\bold u}) = ({\bold u} \ , \ \varphi^{*}J\cdot\delta^{-1}_{*}
{\bold v}) \ ,$$ therefore $$(\delta_{*}{\bold u},J_{0}\cdot{\bold v}) = ({\bold u},\delta_{*}^{-1}
J_{0}\delta_{*}\delta_{*}^{-1}{\bold v}) = ({\bold u},\delta_{*}^{-1}
J_{0}\cdot{\bold v}) \ ,$$ that is, $$({\bold u},\delta_{*}^{t}J_{0}\cdot{\bold v}) = ({\bold u},\delta_{*}
^{-1}J_{0}\cdot{\bold v}) \ ,$$ and hence $\delta_{*}^{t}X_{\lambda_{0}} = \delta_{*}^{-1}X_{\lambda_{0}}$ (or, conversely, $\delta_{*}X_{\lambda} = (\delta_{*}^{t})^{-1}
X_{\lambda}$) . On the other hand, $(\delta_{*}{\bold u}, J_{0}\cdot
{\bold v}) = ((\delta_{*}^{-1})^{t}{\bold u}, J_{0}\cdot{\bold v})$ if and only if $$(\delta_{*}{\bold u}, \delta_{*}\varphi^{*}J\delta_{*}^{-1}{\bold v}) =
((\delta_{*}^{-1})^{t}{\bold u}, \delta_{*}\varphi^{*}J\delta_{*}^{-1}
{\bold v})$$ i.e., letting ${\bold v}':= \delta_{*}^{-1}{\bold v}$ , $$(\delta_{*}^{t}\delta_{*}{\bold u}, \varphi^{*}J\cdot{\bold v}') =
({\bold u}, \varphi^{*}J\cdot{\bold v}') \ .$$ In particular, we see that $ker(\lambda_{\Sigma})$ is an invariant subspace of $\delta_{*}^{t}\delta_{*}$. Note moreover that $$\delta_{*}^{t}\delta_{*}\varphi^{*}J\delta_{*}^{-1}(\delta_{*}^{t})^{-1}
= (\delta_{*}^{-1}J_{0}^{t}\delta_{*})^{t} = \varphi^{*}J \ ,$$ which implies that $\delta_{*}^{t}\delta_{*}\mid_{ker(\lambda)}$ is complex–linear. But since it is clearly symmetric, it follows that $\delta_{*}^{t}\delta_{*}\mid_{ker(\lambda)} = c\cdot I$ for some $c\in{\Bbb R}\setminus\{0\}$. Now $d(\delta^{*}\lambda_{0}) =
\delta^{*}(d\lambda_{0}) = d\lambda_{\Sigma}$ implies $$d\lambda_{\Sigma}({\bold u},{\bold w}) = d\lambda_{0}(\delta_{*}{\bold u},
\delta_{*}{\bold w}) = (\delta_{*}{\bold u}, J_{0}\cdot\delta_{*}
{\bold w})$$ $$= ({\bold u}, \delta_{*}^{t}J_{0}\delta_{*}{\bold w}) =
({\bold u}, \delta_{*}^{t}\delta_{*}\varphi^{*}J\cdot{\bold w}) =
({\bold u}, \varphi^{*}J\delta_{*}^{t}\delta_{*}{\bold w}) =
c({\bold u},\varphi^{*}J\cdot{\bold w})$$ (if ${\bold w}\in ker(\lambda_{\Sigma})$), and hence $$d\lambda_{\Sigma}({\bold u},{\bold w}) = c\cdot d\lambda_{\Sigma}
({\bold u}, {\bold w}) \ , \hspace{.1in} \hbox{i.e.} \ c = 1 \ .$$ We conclude that $\delta_{*}^{t} = \delta_{*}^{-1}$, and hence that $\delta\in{\Bbb O}(4)$, which restricts any such equivalence of contact structures to the family of linear Hopf fibrations. In summary:
Let $J$ be the osculating complex structure of an elliptic line congruence $\Sigma\subset{\Bbb G}r_{2}(V)$, such that the skew–symmetric 2–form $\omega_{\Sigma}$ on $V\setminus\{0\}$ is closed. Then the 1–form $\lambda$, such that $\lambda_{{\bold v}}
:=i_{{\bold v}}\omega_{\Sigma}$, defines a fillable, hence tight, contact structure on ${\Bbb S}^{3}$ for which ${\cal L}_{X_{\lambda}}J
= 0$ and $\alpha = 1$. Moreover, two such structures are equivalent via a diffeomorphism $\delta$ of ${\Bbb S}^{3}$ if and only if $\delta\in{\Bbb O}(4)$.
The classical models of the sphere and rational ellipsoids
==========================================================
As an explicit illustration of Theorem 1, we will first examine the classical case of analytic curves in ${\Bbb C}^{2}$. The realization of algebraic curves as finite-energy pseudoholomorphic maps has also been examined in \[7\], though from a slightly different point of view. Let $M = {\Bbb S}^{3}\subseteq{\Bbb C}^{2}$, with $$(\lambda_{0})_{{\bold v}} = (*,J_{0}\cdot{\bold v})\mid_{T{\Bbb S}^{3}}$$ the standard contact form defined with respect to the complex structure $J_{0}$ of ${\Bbb C}^{2}$ and the Euclidean inner product of ${\Bbb R}^{4}$. $M$ is then defined by the equation $|w_{1}|^{2}+|w_{2}|^{2} = 1$ with respect to complex coordinates in ${\Bbb C}^{2}$, while $$\lambda_{0} = \Re(-{\bold i}(\bar{w}_{1}dw_{1} + \bar{w}_{2}dw_{2})) \ .$$ Consider $\Phi(z):D@>>>{\Bbb C}^{2}$ a complex–analytic curve defined on a neighbourhood of the origin in ${\Bbb C}$ such that $\Phi(z) =
(z^{n},\rho(z))$ and $ord_{0}(\rho)\geq n+1$, hence $\Phi$ has a singularity of multiplicity $n$ at $(0,0)$. The corresponding map $(\psi, a):D@>>>{\Bbb
S}^{3}\times{\Bbb R}$ is given by $$\psi(z) = [|\rho|^{2}+|z|^{2n}]^{-\frac{1}{2}}(z^{n},\rho(z)) \ , \$$ $$a(z) = -\frac{1}{2}\ln(|z|^{2n}+|\rho|^{2}) = -n\ln|z| - \frac{1}{2}
\ln(1+|z|^{-2n}|\rho|^{2}) \ .$$ The periodic orbit ${\cal P}$ corresponds simply to the circle defined by $\{w_{2} = 0\}\cap{\Bbb S}^{3}$ and is associated with the degenerate tangent cone of $\Phi$ at $(0,0)$. It is, moreover, a simple calculation to verify that $charge_{z=0}(\psi) = n$. Consider the proper holomorphic map $\sigma:{\cal O}_{{\Bbb P}_{1}}(-1)@>>>{\Bbb C}^{2}$ , where ${\cal O}
_{{\Bbb P}_{1}}(-1)$ denotes the complex line bundle of Chern class equal to –1 on the Riemann Sphere ${\Bbb P}_{1}({\Bbb C}) = \sigma^{-1}({
\bold 0})$, with $\sigma:{\cal O}_{{\Bbb P}_{1}}(-1)\setminus\sigma^{-1}
({\bold 0})\cong{\Bbb C}^{2}\setminus\{0\}$, given in local coordinates by the quadratic transformation $w_{1} = \mu \ , \ w_{2} = \mu\nu$. Noting that $|\mu|^{2}(1+|\nu|^{2}) = 1$ on the chart of $\sigma^{-1}
({\Bbb S}^{3})$ corresponding to $\nu\neq\infty$, we have $$\sigma^{*}\lambda_{0} = \Re(-{\bold i}(\mu^{-1}d\mu + \bar{\nu}|\mu|^{2}
d\nu)) \ ;$$ moreover, for each $\nu\in{\Bbb P}_{1}({\Bbb C})\setminus\{\infty\}$, the Hopf fibration corresponding to $$\varpi:\sigma^{-1}({\Bbb S}^{3})@>>>{\Bbb P}_{1}({\Bbb C})
\approx{\Bbb S}^{2}$$ has fibres $\varpi^{-1}(\nu) = \{\mu\in{\Bbb C} \ | \
|\mu| = (1+|\nu|^{2})^{-\frac{1}{2}}\}$. Let $\nu = x+{\bold i}y$, then $\mu = |\mu|e^{{\bold i}\vartheta} = (1+x^{2}+y^{2})^{-\frac{1}{2}}
e^{{\bold i}\vartheta} \ , \ h:=\sigma\mid_{\varpi^{-1}({\Bbb P}_{1}
\setminus\{\infty\})}$ implies $$h^{*}\lambda_{0} = \Re(-{\bold i}\{\mu^{-1}d\mu + \bar{\nu}|\mu|^{2}
d\nu\})$$ $$= \Re(-{\bold i}\{(\sqrt{1+x^{2}+y^{2}} \ )e^{{\bold i}\vartheta}
\left(\frac{-(xdx + ydy)e^{{\bold i}\vartheta}}{\sqrt{(1+x^{2}+y^{2})^{3}}} +
\frac{{\bold i}e^{{\bold i}\vartheta}d\vartheta}{\sqrt{1+x^{2}+y^{2}}}
\right)$$ $$+ (x-{\bold i}y)\frac{dx+{\bold i}dy}{1+x^{2}+y^{2}}\})$$ $$= \Re(-{\bold i}\left\{\frac{-(xdx+ydy)}{1+x^{2}+y^{2}} + \frac{xdx+ydy}
{1+x^{2}+y^{2}} + {\bold i}(d\vartheta + \frac{xdy-ydx}{1+x^{2}+y^{2}})
\right\})$$ $$= d\vartheta + (1+x^{2}+y^{2})^{-1}(xdy - ydx) \ ,$$ while $h^{*}X_{\lambda_{0}} = \frac{\partial}{\partial\vartheta}$. Now $$\Psi(z):=h^{-1}\psi(z) = (\arg(z^{n}) \ ,
\ F(z)) \hspace{.1in}\hbox{where} \ F(z):=z^{-n}\rho(z) \ .$$ Clearly $\nu = F(z)$ is holomorphic, so it remains to show that $$h^{*}\lambda_{0}(\Psi_{\eta}) = a_{\zeta} \ ; \ h^{*}\lambda_{0}
(\Psi_{\zeta}) = -a_{\eta} \ .$$
Note that $a(z) = -\frac{1}{2}\ln((\eta^{2}+\zeta^{2})^{n}(1+x^{2}+y^{2}))$ implies $$a_{\zeta} = \frac{-n\zeta(\eta^{2}+\zeta^{2})^{n-1}(1+x^{2}+y^{2}) -
(\eta^{2}+\zeta^{2})^{n}(xx_{\zeta}+yy_{\zeta})}{(\eta^{2}+\zeta^{2})^{n}
(1+x^{2}+y^{2})}$$ $$= \frac{-n\zeta}{\eta^{2}+\zeta^{2}} -
\frac{xx_{\zeta} + yy_{\zeta}}{1+x^{2}+y^{2}} \ ,$$ while $\Psi_{\eta} = (x_{\eta},y_{\eta},\vartheta_{\eta}) = (
x_{\eta},y_{\eta},\frac{-n\zeta}{\eta^{2}+\zeta^{2}})$ implies $$h^{*}\lambda_{0}(\Psi_{\eta}) = \frac{-n\zeta}{\eta^{2}+\zeta^{2}} +
\frac{xy_{\eta} - yx_{\eta}}{1+x^{2}+y^{2}} \ .$$ But $\nu = x+{\bold i}y = F(z)$ is holomorphic, hence $x_{\eta} = y_{\zeta}
\ , \ x_{\zeta} = -y_{\eta}$ yields $a_{\zeta} = h^{*}\lambda_{0}
(\Psi_{\eta})$, and similarly for $a_{\eta}$.
Recalling the discussion of section 3, if we write $\lambda' = dt +
f_{1}(x,y)dx + f_{2}(x,y)dy$, then $h^{*}\lambda_{0} = 2\pi\lambda'$, where $$f_{1}(x,y) = \frac{-y}{2\pi(1+|\nu|^{2})} \hspace{.1in} ; \hspace{.1in}
f_{2}(x,y) = \frac{x}{2\pi(1+|\nu|^{2})} \ .$$ Now $$\frac{1}{2}(f_{1} + {\bold i}f_{2}) = \frac{{\bold i}\nu}
{4\pi(1+|\nu|^{2})} = \frac{{\bold i}}{4\pi}\bar{\partial}\ln(1+|\nu|^{2})
\ ,$$ and hence $g\circ F = \frac{{\bold i}}{2\pi}\ln(1+|F|^{2})^{\frac{1}{2}}$. Moreover, we can simply define $\hat{G} = 2\pi{\bold i}g\circ F
= -\ln(1+|F|^{2})^{\frac{1}{2}}$.
On the other hand, modulo rescaling by $2\pi$ so that $$\hat{a}(z) = \frac{-1}{2\pi}\ln(|z|^{n}(1+|F|^{2})^{\frac{1}{2}}) \ ,$$ we have $$u = t+{\bold i}\hat{a} = \frac{1}{2\pi}(arg(z^{n}) +
{\bold i}\{-\ln|z|^{n} - \ln(1+|F|^{2})^{\frac{1}{2}}\})$$ $$= \frac{1}{2\pi{\bold i}}(\log(z^{n}) - \hat{G}) \ ,$$ in accordance with the statement of Theorem 1.
Let $\Gamma\subset{\Bbb C}^{2}$ be the locus of the plane curve parametrized by $\Phi$. We remark in conclusion that although the multiplicity of the singular point of the strict transform $\overline{\sigma^{-1}(\Gamma)\setminus
\sigma^{-1}(0)}$ is less than $n$, the charge at $z=0$ of $\Psi$ is easily seen to be conserved by the diffeomorphism $h$. The singular plane curve corresponding to the strict transform (assuming the singularity has not been resolved by a single quadratic transformation) is in fact asymptotic (viewed locally as a $J$–holomorphic curve $\Psi'$) to a distinct periodic orbit within a new 3–sphere bounding a neighbourhood of the transformed singularity. However, the linking of the transform of $K_{\Gamma}$ with the original periodic orbit of $\sigma^{-1}({\Bbb S}^{3})$ is topologically unaffected.
Now let us turn to the ellipsoids $${\Bbb E}_{p,q} := \{(w_{1},w_{2})\in{\Bbb C}^{2} \ | \ p|w_{1}|^{2} +
q|w_{2}|^{2} = 1 \ ; \ (p,q)\in{\Bbb R}^{2}_{+}\} \ .$$ The restriction to ${\Bbb E}_{p,q}$ of $\lambda_{0}$ as defined above determines a different contact structure on $T{\Bbb E}_{p,q}$, and in particular a Reeb vector field $X_{P}:= A\cdot J_{0}\cdot{\bold v}_{P}$ for all $P\in{\Bbb E}_{p,q}$, where $$A = \left(\begin{array}{cc}
p & 0 \\
0 & q \\
\end{array}\right)$$ is viewed as an element of ${\bold G}{\bold L}(2,{\Bbb C})$. Solutions of the equation $\dot{\gamma}(t) = X(\gamma(t)) \ , \ \gamma(0) = P
= (z_{1},z_{2})$, then correspond to maps $t\mapsto(z_{1}e^{{\bold i}pt},
z_{2}e^{{\bold i}qt})$ (cf. e.g., \[7\]). Note that there are two periodic orbits corresponding to $w_{1} = 0$ and $w_{2} = 0$ separately. These are the [*only*]{} periodic orbits of the Reeb flow if $\frac{p}{q}$ is irrational, whereas ${\Bbb E}_{p,q}$ is foliated by periodic orbits if $\frac{p}{q}\in
{\Bbb Q}$. Moreover the periodic orbit corresponding to $w_{1} = 0$ has minimal period $\tau = \frac{2\pi}{q}$ and for $w_{2} = 0$ it is $\frac{2\pi}{p}$, while $\tau = \frac{2\pi k}{p} = \frac{2\pi l}{q}$ for all other orbits with respect to fixed relatively prime positive integers $k,l$. Without loss of generality, consider a tubular neighbourhood ${\cal U}$ of the orbit $w_{2} = 0$, and let $\nu:=\frac{z_{2}^{k}}{z_{1}^{l}}$ such that $l > k$. All periodic orbits of the Reeb vector field within this tubular neighbourhood then correspond to intersections of ${\Bbb E}_{p,q}$ with algebraic curves (uniquely determined by $\nu$) of the form $$w_{2}^{k} = \nu w_{1}^{l} \ , \ \hspace{.1in} |\nu| < \varepsilon \ ,$$ for some positive $\varepsilon$. Once again, we have $$\lambda_{0} = \Re(-{\bold i}(\bar{w}_{1}dw_{1} + \bar{w}_{2}dw_{2})) \ ,$$ but this time it will be convenient to introduce a formal coordinate transformation of the form $$w_{1} = \mu \ , \ \hspace{.1in} w_{2} = (\nu\mu^{l})^{\frac{1}{k}}$$ so that $$\bar{w}_{1}dw_{1} + \bar{w}_{2}dw_{2} = \bar{\mu}d\mu + \frac{1}{k}
|\nu|^{\frac{2}{k}}|\mu|^{\frac{2l}{k}}(\frac{1}{\nu}d\nu + \frac{l}{\mu}
d\mu) \ .$$ Moreover, $p|w_{1}|^{2} + q|w_{2}|^{2} = 1$ implies $$|\nu|^{\frac{2}{k}}|\mu|^{\frac{2l}{k}} = \frac{1}{q}(1 - p|\mu|^{2})
\ ,$$ hence $$\bar{w}_{1}dw_{1} + \bar{w}_{2}dw_{2} = \left(\bar{\mu} + \frac{(1-p|
\mu|^{2})l}{kq\mu}\right)d\mu + \left(\frac{1-p|\mu|^{2}}{kq\nu}\right)
d\nu \ .$$ Now $\mu = re^{{\bold i}\vartheta}$ implies $d\mu = e^{{\bold i}
\vartheta}(dr + {\bold i}rd\vartheta)$, so that $$\left(\bar{\mu} + \frac{(1-p|\mu|^{2})l}{kq\mu}\right)d\mu =
\frac{1}{p}(\frac{dr}{r} + {\bold i}d\vartheta) \ ,$$ while $\nu = x + {\bold i}y$ implies $$\frac{1-p|\mu|^{2}}{kq\nu}d\nu = \frac{1-p|\mu|^{2}}{kq}\left\{\frac{
xdx + ydy}{x^{2}+y^{2}} + {\bold i}\left(\frac{xdy - ydx}{x^{2}+y^{2}}\right)
\right\} \ ,$$ and hence $$\lambda_{0} = \frac{1}{p}\{d\vartheta + \frac{(1-p|\mu|^{2})(xdy-ydx)}
{l(x^{2}+y^{2})}\} \ (*).$$
Consider $f(r) := pr^2 + q|\nu|^{\frac{2}{k}}r^{\frac{2l}{k}} - 1$, so that $$f'(r) = 2pr(1 + \frac{q^2}{p^2}|\nu|^{\frac{2}{k}}r^{2(\frac{l}{k}-1)})
\ .$$ Hence $f'(r) = 0$ when $r=0$ or $(\frac{-p^2}{q^{2}|\nu|^{\frac{2}{k}}})^{
\frac{k}{2(l-k)}}$ . Moreover $f(0) = -1 \ , \ \lim_{r\rightarrow\infty}
f(r) = \infty$ implies that the equation $f(r) = 0$ has a unique positive real solution. In other words, the value of $|\mu|$ satisfying the equation $p|\mu|^{2} + q|\nu|^{\frac{2}{k}}|\mu|^{\frac{2l}{k}} = 1$ is uniquely determined by $p,q,|\nu|$. Hence write $|\mu| = \varphi(p,q,|\nu|)$ and recall that $$w_{2} = (\nu(\varphi(|\nu|)e^{{\bold i}\vartheta})^{l})^{\frac{1}{k}}
\hspace{.1in} \hbox{i.e.,}\hspace{.1in} w_{2}^{k}e^{-{\bold i}l\vartheta} =
\nu\varphi(|\nu|)^{l} \ .$$ Note also that $\varphi(p,q,0) = \frac{1}{\sqrt{p}}$ implies that the complex function $\chi(\nu):=\nu\varphi(|\nu|)^{l}$ admits a locally differentiable inverse, and we may write $$\nu = \chi^{-1}(w_{2}^{k}e^{-{\bold i}l\vartheta}) \ .$$ Let $\Delta = \{|w_{2}|<\varepsilon\} \ , \ \Delta' = \{|\nu|<
\varepsilon'\}$ and consider the $k$–fold covering map $\beta: {\cal U}\approx\Delta\times{\Bbb S}^{1}@>>>\Delta'\times{\Bbb S}
^{1}$ such that $$\beta(w_{2},\vartheta) = (\chi^{-1}(w_{2}^{k}e^{-{\bold i}l\vartheta}),
\vartheta) \ .$$ Alternatively, the map $\Theta:{\Bbb C}^{2}@>>>{\Bbb C}^{2}$ such that $(\mu,\nu) = \Theta(w_{1},w_{2}) = (w_{1}, w_{2}^{k}w_{1}^{-l})$ is holomorphic away from $\{w_{1} = 0\}$ such that $\beta = \Theta\mid_{{\Bbb E}_{p,q}}$. The equation (\*) above may then be written more precisely in the form $\lambda_{0} = \frac{1}{p}\beta^{*}\lambda$, where $$\lambda = d\vartheta + \frac{(1-p\varphi(p,q,|\nu|)^{2})(xdy-ydx)}
{l|\nu|^{2}}$$ (note also that $\lim_{|\nu|\rightarrow 0}\frac{1-p\varphi^2}{l|\nu|^2}
(xdy-ydx) = 0$). Consequently finite–energy pseudoholomorphic maps $$(\psi,a):D\setminus\{0\}@>>>\Delta\times{\Bbb S}^{1}\times{\Bbb R}$$ project onto finite–energy maps $(\beta\circ\psi,a)$, pseudoholomorphic with respect to $\frac{1}{p}\lambda$. This claim is easily verified if we note that $$a_{\zeta} = \frac{1}{p}\beta^{*}\lambda(\psi_{\eta}) = \frac{1}{p}
\lambda(\beta_{*}\circ\psi_{\eta}) = \frac{1}{p}\lambda((\beta\circ\psi)
_{\eta}) \ ,$$ and similarly for $a_{\eta}$. Moreover, $$\tilde{\pi}((\beta\circ\psi)_{\eta}) := \beta_{*}\psi_{\eta} - \frac{1}
{p}\lambda((\beta\circ\psi)_{\eta})\tilde{X} \ ,$$ where $\tilde{X}:=\beta_{*}X_{\lambda_{0}} = p\frac{\partial}{\partial
\vartheta}$ is a well–defined vector field under the above conditions. Hence $$\tilde{\pi}((\beta\circ\psi)_{\eta}) = \beta_{*}(\pi(\psi_{\eta})) \
,$$ and therefore $$0 = \beta_{*}(\pi(\psi_{\eta})+J_{0}\pi(\psi_{\zeta})) = \tilde{\pi}(
\beta\circ\psi)_{\eta} + \beta_{*}(J_{0}\pi(\psi)_{\zeta}) \ .$$ Let $J_{0}$ denote the standard complex structure on ${\Bbb C}^{2}$ as represented by both $(w_{1},w_{2})$ and $(\mu,\nu)$ (and its restriction to the contact planes of $\lambda_{0}$ and $\lambda$ respectively). Then $\beta = \Theta\mid_{{\Bbb E}_{p,q}}$ for $\Theta$ holomorphic implies $\beta_{*}\circ J_{0} = J_{0}\circ\beta_{*}$, and hence $$0 = \tilde{\pi}(\beta\circ\psi)_{\eta} + J_{0}\tilde{\pi}(\beta\circ
\psi)_{\zeta} \ .$$ Note that the tubular neighbourhood into which $\beta\circ\psi$ maps is fibred by Reeb orbits, hence the return map $\alpha = 1$. By comparison, the Reeb flow in a neighbourhood of the original orbit $\{w_{2}
= 0\}$ in ${\Bbb E}_{p,q}$ induces a return map that is equivalent to a rational rotation through $2\pi\frac{l}{k}$. In order to find a class of $J$–holomorphic curves asymptotic to the given periodic orbit within ${\Bbb
E}_{p,q}$, let us first make a harmless renormalization of the contact structure, i.e, $\lambda_{0}' := p\cdot\lambda_{0} \ , \ X_{\lambda_{0}'}
= \frac{1}{p}X_{\lambda_{0}}$. Now $$\lambda_{0}' = d\vartheta + \frac{(1-p\cdot\varphi(|\nu|)
^{2})(xdy-ydx)}{l\cdot|\nu|^{2}}$$ and we are ready to work backwards from a holomorphic parametrization of the form $$\mu = z^{n} \ ; \hspace{.1in} \nu = \Phi(z) \ ,$$ such that $ord_{0}(\Phi) = b\cdot l\geq -nl+1$ and $n = c\cdot k$ for some integers $b,c$. Now $$w_{1} = z^{n} \ ;\hspace{.1in} w_{2} = (z^{nl}\cdot\Phi(z))^{\frac{1}{k}}
= (z^{n+b}\cdot f_{0}(z))^{\frac{l}{k}};\hspace{.1in} f_{0}(0)\neq 0$$ implies that $w_{2}$ is a multi–valued function of $z$. Subdivide the disc $D$ into equal sectors $Q_{m} \ , \ 0\leq m\leq n-1$, and hence define on each $Q_{m}$ a holomorphic function $F_{m}(z)$, such that $F_{0}$ is the principal branch of $(z^{n+b}\cdot f_{0}(z))^{\frac{l}{k}}$, and $F_{m+1}(z) := e^{2\pi{\bold i}\frac{l}{k}}\cdot F_{m}(z)$.
As in the previous example, letting $\vartheta = 2\pi t$, we have $$\omega = \frac{1}{2}(f_{1}+{\bold i}f_{2})(\nu,\bar{\nu})d\bar{\nu} =
\frac{1-p\cdot\varphi(|\nu|)^{2}}{4\pi l\cdot|\nu|^{2}}\cdot{\bold i}
\nu d\bar{\nu} = \bar{\partial}g(\nu,\bar{\nu}) \ .$$ Let $$\gamma(s) := \frac{1-p\cdot\varphi(\sqrt{s})^{2}}{l\cdot s} \ ,$$ noting $\varphi(0) = \frac{1}{\sqrt{p}}$ implies that the improper integral $$\hat{\gamma}(s) := \int_{0}^{s}\gamma(\tau)d\tau$$ is convergent. It follows that we can set $g(\nu,\bar{\nu}) =
\frac{{\bold i}}{4\pi}\hat{\gamma}(|\nu|^{2})$. Hence $$\Phi^{*}\omega = \bar{\partial}(g\circ\Phi) = \frac{{\bold i}}{4\pi}
\bar{\partial}\hat{\gamma}(|\Phi|^{2})$$ is smoothly defined on $D$. Moreover $$\hat{G} = 2\pi{\bold i}g\circ\Phi = -\frac{1}{2}\hat{\gamma}(|\Phi|
^{2}) \ .$$ Setting $t = \frac{1}{2\pi}arg(z^{n})$, we then have $$t+{\bold i}\hat{a} = \frac{1}{2\pi{\bold i}}(\log(z^{n}) - \hat{G})
= \frac{1}{2\pi}arg(z^{n}) +\frac{{\bold i}}{2\pi}(\hat{G} - n\ln|z|) \ ,$$ with which we combine the statement of Theorem 1 to conclude $$a(z) = -n\ln(|z|e^{\frac{1}{2n}\hat{\gamma}(|\Phi|^{2})})$$ in order to define a pseudoholomorphic map $(\psi, a):D\setminus\{0\}
@>>>{\Bbb E}_{p,q}\times{\Bbb R}$ of charge $n$ at the origin, asymptotic to the orbit corresponding to $\{w_{2} = 0\}$.
references
==========
\[1\] Bourgeois, F. [*PhD Thesis*]{} (preprint) Stanford 2004
\[2\] Brieskorn, E. and Knörrer, H. [*Plane Algebraic Curves*]{}, Birkhäuser 1986
\[3\] Eliashberg, Y. [*Invariants in Contact Topology*]{}, Proceedings ICM, Berlin 1998, Volume II, Documenta Mathematica (1998) 327–338
\[4\] Gluck. H. and Warner, F., [*Great Circle Fibrations of the Three–Sphere*]{}, Duke Math. J. 50 No.1 (1983) 107–132
\[5\] Hale, J. [*Ordinary Differential Equations*]{}, Pure and Appl. Math. 21, Wiley 1969
\[6\] Hofer, H., [*Pseudoholomorphic Curves in Symplectisations with applications to the Weinstein Conjecture in dimension three*]{}, Invent. Math. 114 (1993) 515–563
\[7\] Hofer, H. and Kriener, M. [*Holomorphic Curves in Contact Dynamics*]{}, Proc. Symp. Pure Math. 65 (1999) 77–131
\[8\] Hofer, H., Wysocki, K. and Zehnder, E. [*Properties of Pseudoholomorphic Curves in Symplectisations I: Asymptotics*]{}, Ann. Inst. Henri Poincaré 13 (1996) 337–379
\[9\] Hofer, H., Wysocki, K. and Zehnder, E. [*Properties of Pseudoholomorphic Curves in Symplectisations II: Embedding controls and algebraic invariants*]{}, Geom. Funct. Anal. 5 (1995) 270–328
\[10\] Hofer, H., Wysocki, K. and Zehnder, E. [*Properties of Pseudoholomorphic Curves in Symplectisations IV: Asymptotics with degeneracies*]{}, in “Contact and Symplectic Geometry”, C.B Thomas ed., Cambridge (1996) 78–117
\[11\] Martinet, J. [*Formes de Contact sur les variétés de dimension 3*]{}, Springer Lecture Notes 207 (1971) 142–163
\[12\] McDuff, D. [*Singularities of $J$–holomorphic Curves in almost complex 4–manifolds*]{}, J. Geom. Anal. 2 (1992) 249–266
\[13\] McKay, B. [*Dual Curves and Pseudoholomorphic Curves*]{}, Selecta Math. 9 (2003) 251–311
\[14\] Micallef, M. and White, B. [*The structure of branch points in Minimal Surfaces and in Pseudoholomorphic Curves*]{}, Ann. Math. 139 (1994) 35–85
\[15\] Sikorav, J.–C. [*Singularities of $J$–holomorphic Curves*]{}, Math. Z. 226 (1997) 359–373
School of Mathematics, Statistics and Computer Science University of New England Armidale, NSW 2351 Australia adamh@@turing.une.edu.au
School of Mathematics and Statistics Melbourne University Parkville, VIC 3010 Australia wysocki@@ms.unimelb.edu.au
|
{
"pile_set_name": "ArXiv"
}
|
=23truept
[**Title: Critical curvature of large-$N$ nonlinear $O(N)$ sigma model on $S^2$**]{}\
[**Short title: Critical curvature on $S^2$**]{}
K.H. Kim and Dae-Yup Song[^1]
[*Department of Physics, Sunchon National University,\
Sunchon, Chonnam 540-742, Korea* ]{}
Abstract
We study the nonlinear $O(N)$ sigma model on $S^2$ with the gravitational coupling term, by evaluating the effective potential in the large-$N$ limit. It is shown that there is a critical curvature $R_c$ of $S^2$ for any positive gravitational coupling constant $\xi$, and the dynamical mass generation takes place only when $R<R_c$. The critical curvature is analytically found as a function of $\xi$ $(>0)$, which leads us to define a function looking like a natural generalization of Euler-Mascheroni constant.
PACS numbers: 04.62.+v, 11.15.Pg
Introduction
============
For a model described by boson fields $n_i$ $(n=1,2,\cdots,N)$ on a curved spacetime, a gravitational interaction term $\xi_i R n_i n_i$ may be added to the Lagrangian [@BiDa], where $\xi_i$ is a gravitational coupling constant and $R$ is the scalar curvature of the spacetime ( In this paper, we consider only Euclidean spacetime. ). Since this interaction term is a bilinear term of the fields, the change of mass through quantum loop corrections may be of interest.
Recently one of us [@Song] studied the three dimensional nonlinear $O(N)$ sigma model on $S^2\times R^1$ in the leading order of $1/N$ expansion, where $S^2$ is the two sphere with the canonical metric. For the $O(N)$ symmetry we assume that $\xi_i$ is the same $\xi$ for all $i$. On the three dimensional flat spacetime, there exists a critical coupling constant of the model and the dynamical mass generation takes place only when the coupling constant is larger than the critical one (strong coupling regime) [@Aref]. In the strong coupling regime, we showed that there is a critical curvature (radius) of $S^2$ for some value of $\xi$. Though the $O(N)$ model can be treated by the $1/N$ expansion in both two and three dimensions, the properties of the models are substantially different. One of them is that the dynamical mass generation takes place in two dimension no matter how small the coupling constant is, as it should be.
In this paper, we will study the two-dimensional $O(N)$ model [@PolBerZin] on $S^2$ with the gravitational coupling term in the large-$N$ limit, by evaluating the effective potential [@Colman]. We will show that there is a critical curvature $R_c(\xi)$ for every $\xi~(>0)$. As $\xi$ approaches 0, the $R_c$ goes to infinity and decreases monotonically as $\xi$ increases. By introducing a function which looks like a functional generalization of Euler-Mascheroni constant, the $R_c$ can be neatly expressed by the function.
Effective potential and critical curvature
==========================================
We assume that the metric of $S^2$ is the canonical one so that the curvature $R$ is $2/\rho^2$, where $\rho$ is the radius of $S^2$. Then the Lagrangian of our model is written as: $$\begin{aligned}
L=\int d\theta d\varphi ~\rho^2 \sin\theta
+(\rho^2\sin^2\theta)^{-1}\partial_\varphi n^i \partial_\varphi n^i
\nonumber\\
&&~+\xi R n^2 +\sigma(n^2-Ng^{-2})~],\end{aligned}$$ where $g_0$ is the bare coupling constant. For the evaluation of the effective potential we can write the Lagrangian density as $${\cal L}=n^i Dn_i-N\sigma/g_0^2.$$ In (2) $D$ is defined as $\rho^{-2}({\bf L}^2+2\xi)+\sigma$, where ${\bf L}$ is the quantum mechanical angular momentum operator. By performing the Gaussian functional integral [@Jackiw], the effective potential per unit volume is given as $$\frac{V}{N}=-\frac{\sigma}{g_0^2}+\frac{1}{4\pi\rho^2}({\rm Tr} \ln D+ C),$$ where $C$ is a constant which may be determined by requiring $V\mid_{\sigma=0}=0$. As is well-known, in the leading order of $1/N$ expansion the effective potential in (3) is written in terms of spacetime independent $\sigma$: $$\frac{V}{N}=-\frac{\sigma}{g_0}+\frac{1}{4\pi\rho^2}\sum_{l=0}^{I}
(2l+1)\ln(1+\frac{\sigma}{\frac{l(l+1)}{\rho^2}+\frac{2\xi}{\rho^2} }),$$ where we introduce an integer cut-off $I$ for the quantum number of the operator ${\bf L}$.
In order to compare effective potential in (4) with that on $R^2$, as in the previous paper [@Song], one can use the following formula: $$\sum_{l=0}^N f(l)=\frac{1}{2}f(0)+\int_0^{N+1}f(x)dx
+\sum_{l=0}^N\int_0^1 f'(x+l)(x-\frac{1}{2})dx
-\frac{1}{2}f(N+1),$$ which gives $$\begin{aligned}
\frac{V}{N}
&=&-\frac{\sigma}{g_0^2}
+\frac{1}{2\pi \rho} \int_0^I(x+\frac{1}{2})
\ln (1+\frac{\sigma \rho^2}{(x+\frac{1}{2})^2+2\xi-\frac{1}{4}})dx
+\frac{1}{8\pi\rho^2} \ln (1+\frac{\sigma \rho^2}{2\xi}) \nonumber \\
&&+\frac{1}{2\pi \rho^2}\sum_{l=0}^I\int_0^1 dx (x-\frac{1}{2})
\left[ \begin{array}{l}
\ln (1+\frac{\sigma\rho^2}{(x+l+\frac{1}{2})^2+2\xi-\frac{1}{4}})\\
+ \frac{2(x+l+\frac{1}{2})^2}{(x+l+\frac{1}{2})^2
+2\xi- \frac{1}{4} +\sigma \rho^2}
- \frac{2(x+l+\frac{1}{2})^2}{(x+l+\frac{1}{2})^2 +2\xi- \frac{1}{4}}
\end{array}\right].\end{aligned}$$ Since the integrals in (5) are well defined without infra-red problem for $\xi>0$, we will restrict our attention to these cases. By defining a cut-off $\Lambda=(I+\frac{1}{2})/\rho$ whose dimension is that of a momentum, we can write the potential as $$\begin{aligned}
\frac{V}{N}&=& -\frac{\sigma}{g_0^2}
+\frac{1}{2\pi}\int_{1/2\rho}^\Lambda y
\ln (1+\frac{\sigma}{y^2+\frac{2\xi-\frac{1}{4}}{\rho^2}})dy
+\frac{1}{8\pi \rho^2}\ln (1+\frac{\sigma \rho^2}{2\xi}) \nonumber\\
&&+\frac{1}{2\pi \rho^2}\sum_{l=1}^\infty
\int_{-\frac{1}{2}}^{\frac{1}{2}} dt~t
\left[\begin{array}{l}
\ln (1+\frac{\sigma\rho^2}{(t+l)^2+2\xi -\frac{1}{4}})\\
+\frac{2(t+l)^2}{(t+l)^2+2\xi-\frac{1}{4}+\sigma\rho^2}
-\frac{2(t+l)^2}{(t+l)^2+2\xi-\frac{1}{4}}
\end{array}\right] \nonumber\\
&&+O(1/\Lambda) \nonumber \\
&=&-\frac{\sigma}{g_0^2}-\frac{\sigma}{4\pi}
(\ln \frac{2\xi+\sigma\rho^2}{\rho^2\Lambda^2} -1)
+\frac{1-4\xi}{8\pi\rho^2}\ln (1+\frac{\sigma\rho^2}{2\xi})\nonumber\\
&&+\frac{1}{2\pi \rho^2}\sum_{l=1}^\infty
\int_{-\frac{1}{2}}^{\frac{1}{2}} dt~t
\left[\begin{array}{l}
\ln (1+\frac{\sigma\rho^2}{(t+l)^2+2\xi -\frac{1}{4}})\\
+\frac{2(t+l)^2}{(t+l)^2+2\xi-\frac{1}{4}+\sigma\rho^2}
-\frac{2(t+l)^2}{(t+l)^2+2\xi-\frac{1}{4}}
\end{array}\right] \nonumber\\
&&+O(1/\Lambda).\end{aligned}$$ The effective potential in (6) once again confirms the well-known fact that topological change of spacetime does not give rise to new counterterms, and thus we can use the renormalization relation of the model on $R^2$ [@Colman] $$-\frac{1}{g_0^2} =-\frac{1}{g^2} + \frac{1}{4\pi}\ln \frac{M^2}{\Lambda^2},$$ where $M$ is the renormalization mass. It is convenient to define $\sigma_0$ $$\sigma_0=M^2e^{-4\pi/g^2}$$ which denotes the square of the dynamically generated mass on $R^2$. Making use of the relations (7) and (8), one can find the renormalized effective potential $$\begin{aligned}
\frac{V}{N}&=& -\frac{\sigma}{4\pi}
(\ln\frac{2\xi +\sigma\rho^2}{ \sigma_0 \rho^2}-1)
+\frac{1-4\xi}{8\pi\rho^2}\ln (1+\frac{\sigma\rho^2}{2\xi})
\nonumber\\
&&+\frac{1}{2\pi \rho^2}\sum_{l=1}^\infty
\int_{-\frac{1}{2}}^{\frac{1}{2}} dt~t
\left[\begin{array}{l}
\ln (1+\frac{\sigma\rho^2}{(t+l)^2+2\xi -\frac{1}{4}})\\
+\frac{2(t+l)^2}{(t+l)^2+2\xi-\frac{1}{4}+\sigma\rho^2}
-\frac{2(t+l)^2}{(t+l)^2+2\xi-\frac{1}{4}}
\end{array}\right].\end{aligned}$$ To have a better understanding of the shape of $V$ in (9), we evaluate the first derivative of $V$ with respect to $\sigma$: $$\begin{aligned}
\frac{1}{N}\frac{\partial V}{\partial \sigma}\equiv\frac{V'(\sigma)}{N}
&=&-\frac{1}{4\pi}\ln\frac{2\xi+\sigma\rho^2}{\sigma_0\rho^2}
+\frac{1}{8\pi}\frac{1}{2\xi+\sigma\rho^2} \nonumber\\
& &+\frac{1}{2\pi}\sum_{l=1}^\infty \int_{-\frac{1}{2}}^\frac{1}{2}
dt~t\frac{-(t+l)^2+2\xi-\frac{1}{4}+\sigma\rho^2}
{[(t+l)^2+2\xi-\frac{1}{4}+\sigma\rho^2]^2}.\end{aligned}$$ As shown in the appendix in detail, one can find a simpler form of the first derivative in (10): $$\frac{1}{N}\frac{\partial V}{\partial \sigma}
=\frac{1}{4\pi}[\ln(\sigma_0\rho^2)+\frac{1}{2\xi+\sigma\rho^2}
+2\gamma(2\xi+\sigma\rho^2-\frac{1}{4})],$$ where the function $\gamma$ is defined by $$\gamma(\beta)=\lim_{N\rightarrow\infty}
[\sum_{n=1}^N\frac{n+\frac{1}{2}}{(n+\frac{1}{2})^2+\beta}
-\frac{1}{2}\ln(N^2+\beta)].$$ The $\gamma$ function looks like a generalization of Euler-Mascheroni constant $$\gamma_E=\lim_{N\rightarrow\infty}[\sum_{n=1}^N n^{-1} -\ln N]
=0.5772\cdots.$$ In fact, making use of the formulae e.g. in [@WhiWat], one can show that $\gamma(0)=\gamma_E-2-2\ln2$. Furthermore, it is easy to show that $\gamma(\beta)$ $(\beta>-\frac{1}{4})$ is a monotonically decreasing function of $\beta$, and it approaches to $-\infty$ as $\beta$ goes to $\infty$.
Now one can find that, for a fixed $\xi$, $V'(\sigma)$ monotonically decreases to $-\infty$ as $\sigma$ increases. That is, a global stationary point of the effective potential $V$ which denotes the dynamical mass generation appears only when $V'(0) >0$. Therefore, the critical radius of $S^2$ is given by $$\sigma_0\rho_c^2=\exp(-\frac{1}{2\xi}-2\gamma(2\xi-\frac{1}{4})).$$ Or, the critical curvature is given by $$R_c=2\sigma_0\exp(\frac{1}{2\xi}+2\gamma(2\xi-\frac{1}{4})).$$ The critical curvature which is $\infty$ in the limit $\xi\rightarrow 0$ decreases monotonically to 0 as $\xi$ increases to $\infty$, and the dynamical mass generation takes place only when $R<R_c$ (or $\rho>\rho_c$).
Discussion
==========
By evaluating the effective potential (9) in the leading order of $1/N$ expansion, we have shown that, the $O(N)$ nonlinear $\sigma$ model with the gravitational coupling term described by the Lagrangian (1), has a critical curvature ( given in (13,14) ) which decreases as $\xi$ increases.
For the conformal symmetry, $\xi$ must be $\xi_c=\frac{1}{4}[(d-2)/(d-1)]$ in a $d$-dimensional spacetime. That is, the conformal coupling constant $\xi_c$ is 0 in two dimension and $\frac{1}{8}$ in three dimension. One of the common features of the $O(N)$ nonlinear $\sigma$ model on $S^2$ and $S^2\times R^1$ [@Song] is that the dynamical mass generation takes place for any finite $R$ in the limit $\xi\rightarrow\xi_c$ while there exists a critical curvature for $\xi>\xi_c$.
[**Acknowledgments**]{}
One of us (D.Y.S) would like to thank Jeong Hyeong Park for the discussion on the $\gamma$ function. This work is supported in part by the Korea Research Foundation.
Appendix {#appendix .unnumbered}
========
Making use of the identity $$\begin{aligned}
\int\frac{t[-(t+l)^2+\beta]}{[(t+l)^2+\beta]^2}dt
&=&-\frac{l(t+l)+\beta}{(t+l)^2+\beta}-\frac{1}{2}\ln[(t+l)^2+\beta]
\nonumber\\
&&\mbox{ for}~~~ (t+l)^2+\beta>0,\end{aligned}$$ one can find the following equality which could be used to find (11) and (12) from (10); $$\begin{aligned}
&&\sum_{l=1}^\infty\int_{-\frac{1}{2}}^\frac{1}{2}
\frac{t[-(t+l)^2+x-\frac{1}{4}]}
{[(t+l)^2+x-\frac{1}{4}]^2}dt \nonumber\\
&&=\frac{1}{4x}+\frac{1}{2}\ln x
+\lim_{N\rightarrow\infty}
\{\sum_{n=1}^{N-1}\frac{n+\frac{1}{2}}{(n+\frac{1}{2})^2+x-\frac{1}{4}}
-\frac{1}{2}\ln[(N+\frac{1}{2})^2+x-\frac{1}{4}]\}.\end{aligned}$$
[\*\*]{} Birrell N D and Davies P C W 1982 [*Quantum Fields in Curved space*]{} (Cambridge:Cambridge Univ. Press) Song D Y 1995 [*Sunchon Nat’l Univ. preprint*]{}, hep-th/9505036 Aref’eva 1979 Ann. Phys. (N.Y.) [**117**]{} 393 Polyakov A M 1975 Phys. Lett. [**59B**]{} 79\
Berezin E and Zinn-Justin J 1976 Pys. Rev. B [**14**]{} 3110 Colman S 1985 [*Aspects of Symmetry*]{} (Cambridge:Cambridge Univ. Press) Jackiw R 1974 Phys. Rev. D [**9**]{} 1686 Whittacker E T and Watson G N 1927 [*A Course of Modern Analysis*]{} (Cambridge:Cambridge Univ. Press)
[^1]: Electronic address: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- 'Dongmin Gang,'
- 'Eunkyung Koh,'
- 'Sangmin Lee,'
- Jaemo Park
title: |
Superconformal Index and\
3d-3d Correspondence for Mapping Cylinder/Torus
---
Introduction and Summary {#sec : introduction}
========================
There is an interesting and fruitful approach of viewing lower dimensional superconformal field theories (SCFTs) from the vantage point of the $(2,0)$ theory in six dimensions. Though we do not fully understand the $(2,0)$ theory, this viewpoint leads to useful insight to understand SCFTs in lower dimensions. Perhaps the most well-known example would be the celebrated AGT conjecture [@AGT]. Heuristically, the conjecture can be motivated by considering a twisted compactification of the $(2,0)$ theory on $S^4 \times \Sigma_{g,h}$ where $\Sigma_{g,h}$ is a Riemann surface of genus $g$ with $h$ punctures. The compactification leads to interesting SCFTs with 8 supercharges in 4-dimensions [@GaiottoN=2]. A supersymmetric partition function of the $(2,0)$ theory on $S^4 \times \Sigma$ is expected to give the $S^4$-partition function of such 4d SCFTs. Pestun computed the $S^4$-partition function using localization techniques for theories whose Lagrangian is known [@Pestun]. Throughout this paper, we will mainly focus on the $A_1$ type of $(2,0)$ theory. In this case, the 4d theories, denoted by $T_\Sigma$, admit weakly coupled gauge theory descriptions with gauge group $SU(2)^{3g-3+h}$. On the other hand, the $(2,0)$ theory compactified on $S^4$ is expected to lead to a 2d conformal field theory. It turns out that for the $A_1$ $(2,0)$ theory this is Liouville theory (for a review of Liouville theory see [@Nakayama]). Hence, the $A_1$ $(2,0)$ theory on $S^4 \times \Sigma$ gives the partition function (correlation function) of the Liouville theory on $\Sigma$. Identifying the two partition functions obtained from two different regimes of the compactification, we obtain the AGT conjecture which relate $S^4$ partition function of $T_\Sigma$ theory with Liouville correlation function on $\Sigma$.
One might wonder if a similar relation may be found in 3-dimensions by compactifying the $(2,0)$ theory on some 3-manifold $M$. If so, it would lead to a plethora of $\CN=2$ SCFTs in 3 dimensions. In fact, in [@Dimofte:2011ju; @Dimofte:2011py], Dimofte, Gaiotto, Gukov (DGG) introduced an algorithm to construct the field theory $T_M$ associated with the 3-manifold $M$ using the ideal triangulation data of $M$. By specifying the gluing rules of the field theory corresponding to those of the triangulation, one can construct a huge class of 3d SCFTs. One interesting feature is that the same manifold with two different triangulations gives rise to two different descriptions of the same SCFT. Some simple mirror pairs of 3d were shown to be described in this way. Although it is difficult to see from their construction, the theory $T_M$ is believed to be the 3d theory obtained by compactifying the $A_1$ $(2,0)$ theory on $M$. Considering the $(2,0)$ theory on $ M \times S^2 \times_q S^1$ or $
M \times S_b^3$,[^1] we have the 3d-3d analogue of the AGT conjecture. If we first compacitify on $M$, we obtain the superconformal index ($S^2\times_q S^1$) or the sqaushed three-sphere partition function ($S_b^3$) for $T_M$. On the other hand, if we compactify on $ S^2 \times_q S^1$ or $
S_b^3$ first, the theory is expected to be a $SL(2,\IC)$ or $SL(2,\IR)$[^2] Chern-Simons (CS) theory on $M$ [@Dimofte:2011ju; @Dimofte:2011py; @Terashima:2011qi; @Yagi:2013fda], respectively. From this analysis, we obtain the following non-trivial prediction of the 3d/3d correspondence: $$\begin{aligned}
&\textrm{Superconformal index/$S^3_b$ partition function for $T_M$ } {\nonumber}\\
&= \textrm{ $SL(2,\IC)/SL(2,\IR)$ Chern-Simons partition function on $M$.} \label{3d/3d correspondence}
\end{aligned}$$
One interesting class of the 3d-3d correspondence arises from the 3d duality domain wall theory [@Gaiotto:2008ak; @Drukker:2010jp; @Hosomichi:2010vh; @Terashima:2011qi; @Terashima:2011xe; @Dimofte:2011jd; @Teschner:2012em; @Gang:2012ff] associated with 4d theory $T_\Sigma$ and a duality group element $\varphi$. The corresponding internal 3-manifold $M$ is the mapping cylinder $\Sigma \times_\varphi I$, where $I=[0,1]$ is the unit interval, equipped with the cobordism $\varphi: (x,0)
\rightarrow (\varphi(x),1)$. Here $\varphi$ is an element of the mapping class group for $\Sigma$, which can be identified with duality group for $T_\Sigma$. Further identifying the two ends of the interval by the cobordism $\varphi$, we obtain a mapping torus $\Sigma
\times_{\varphi} S^1$. Identifying the two ends of the interval corresponds to gluing two global $SU(2)^{3g-3+h}$ symmetries in the duality wall theory coupled to $SU(2)^{3g-3+h}$ gauge symmetry in $T_\Sigma$. On the other hand, the mapping torus admits an ideal triangulation and the corresponding 3d theory can be constructed by the DGG algorithm. Hence the mapping torus has two different realizations of the associated 3d SCFT. The one involving the duality wall theory has a clear origin from M5-brane physics but identifying the 3d SCFT for general $\Sigma$ is very non-trivial. In the other one using the DGG algorithm, the physical origin from M5-brane is unclear but generalization to arbitrary $\Sigma$ is quite straightforward. It boils down to the problem finding a triangulation of the mapping torus.
In this paper we are mainly interested in the 3d-3d correspondence between the superconformal index for $T_M$ and $SL(2,\mathbb{C})$ Chern-Simons theory on $M$, where $M$ is mapping cylinder or torus whose fiber is once-punctured torus, $\Sigma_{1,1}$. The mapping torus $\Sigma_{1,1}\times_\varphi S^1$ will be denoted by tori($\varphi$) for simplicity. The analysis of the 3d-3d correspondence for mapping torus was done at the semiclassical level using the $S^3_b$ partition function in [@Terashima:2011xe]; see also [@K.Nagao:2011; @Kashaev:2012; @Hikam:2012; @Terashima:2013; @Hikam:2013] for interesting generalizations. To check the 3d-3d correspondence at the full quantum level, we carefully define the Hilbert-space of $SL(2,\mathbb{C})$ CS theory on $\mathbb{R}
\times \Sigma_{1,1}$[^3] and construct quantum operators $\varphi \in SL(2,\mathbb{Z})$, which turn out to be unitary operators. Even though several basic ingredients of this construction were already given in references [@Dimofte:2011gm; @Terashima:2011xe; @Dimofte:2011jd; @Dimofte:2011py], working out the details of the Hilbert space turns out to be a non-trivial and worthwhile task. We are particularly interested in the case when the CS level is purely imaginary. In the case, the quantization is studied in a relatively recent paper [@Dimofte:2011py]. In the paper, the Hilbert-space is identified as $L^2 (\mathbb{Z}\times \mathbb{Z})$ which have the same structure with 3d index. We study the mapping-class group representation on the Hilbert-space which is a new and interesting object. We show that the superconformal index for the duality wall theory associated with $\varphi \in SL(2,\mathbb{Z})$ is indeed a matrix element of $\varphi$ in a suitable basis of the Hilbert-space. According to an axiom of topological quantum field theory, the matrix element is nothing but the $SL(2,\mathbb{C})$ CS partition function on the mapping cylinder, and thus it provides an evidence for the 3d-3d correspondence for the mapping cylinder.For mapping torus, tori($\varphi$), the CS partition function is given as a trace of an operator $\varphi \in
SL(2,\mathbb{Z})$. Depending on the choice of basis of the Hilbert-space, the expression for the $\Tr (\varphi)$ is equivalent to the expression of superconformal index for mapping torus theory obtained either using the duality wall theory or using the DGG algorithm. It confirms the equivalence of the two descriptions for mapping torus theory at the level of the superconformal index and also confirms the 3d/3d correspondence for $M=\textrm{tori}(\varphi)$. We also give some evidences for an isomorphism between the Hilbert-space of $SL(2,\mathbb{C})$ CS theory on $\Sigma_{1,1}$ and the Hilbert-space canonically associated to the boundary $S^2 \times S^1$ of 4d (twisted) $\CN=2^*$ theory on $B^3 \times S^1$.
The content of the paper is organized as follows. In section \[section : two routes\], we introduce the basic setup for the 3d geometry of the mapping torus and its ideal triangulation. We also explain the field theory realization, one as a ‘trace’ of the duality domain wall and the other as an outcome of the DGG algorithm based on the triangulation. In section \[quantum riemann surface\], we review the quantization of $SL(2,\mathbb{C})$ Chern-Simon theory on the Riemann surface $\Sigma_{1,1}$. For later purposes, we introduce several coordinate systems for the phase space and explain the relation between them. The A-polynomial for mapping torus is analyzed in two different ways. In section \[sec: SCI/SL(2,C)\], we show that the superconformal index of $T_M$ with $M$ being mapping cylinder/torus is the $SL(2,\mathbb{C})$ CS partition function on $M$. To calculate the CS partition functions, we construct a Hilbert-space for $SL(2,\mathbb{C})$ CS theory on $\Sigma_{1,1}$. We further show that the two computations of the mapping torus index are simply related by a basis change of the Hilbert space in taking trace of $\varphi\in SL(2,\mathbb{Z})$, thereby providing a consistency check for the duality of the two descriptions of the mapping torus theory. In section \[sec: SL(2,R)/squashed S3\], we make comments on the partition function on the squashed sphere for the theory on the mapping cylinder/torus. We indicate many parallels between the partition function and the superconformal index and argue that most of our findings in section \[sec: SCI/SL(2,C)\] can be carried over to the context of the squashed sphere partition function. Several computations are relegated to the appendices. For a technical reason, we mainly focus on general hyperbolic mapping torus which satisfies $|\Tr (\varphi)|>2$ [@Gueritaud]. Extension of our analysis to the non-hyperbolic case seems quite straightforward and some examples are given in section \[index-wall\].
When we were finishing this work, an interesting article [@Dimofte:2013lba] appeared on arXiv.org, which focuses on mapping cylinder and its triangulation. We expect that several expressions for the $SL(2,\mathbb{C})$ CS partition function on mapping cylinder in our paper can be directly derived from their construction.
Two routes to mapping torus field theories {#section : two routes}
==========================================
A mapping torus is specified by a Riemann surface $\Sigma_{g,h}$ of genus $g$ with $h$ punctures and an element $\varphi$ of the mapping class group of $\Sigma_{g,h}$. Topologically, it is a bundle with $\Sigma$ fibered over an interval $I=[0,1]$ with $\Sigma$ at one end of the interval identified with $\varphi(\Sigma)$ at the other end. In other words, $$\begin{aligned}
M=\Sigma \times_\varphi S^1 = \Sigma\times I/[(x,0)\sim (\varphi(x),1)]\,.\end{aligned}$$ In this paper, we only consider the mapping torus for the once punctured torus $\Sigma_{1,1}$ whose mapping class group is $SL(2,\mathbb{Z})$. The mapping torus associated with $\varphi \in SL(2,\mathbb{Z})$ will be denoted as $\textrm{tori}(\varphi)$. $$\begin{aligned}
\textrm{tori}(\varphi) := \Sigma_{1,1}\times_{\varphi}S^1\;.\end{aligned}$$ The 3d-3d correspondence [@Dimofte:2011ju; @Dimofte:2011py] states that one can associate a three-manifold $M$ with a 3d theory $T_M$.[^4] Physically, $T_M$ can be thought of as a dimensional reduction of the 6d (2,0) theory of $A_1$-type on $M$.[^5] The mapping torus theory, $T_M$ with $M= \textrm{tori}(\varphi)$, has two different realizations.
In the first approach, one compactifies the 6d (2,0) theory on $\Sigma_{1,1}$ to obtain the 4d $\mathcal{N}=2^*$ theory and reduces it on $S^1$ with a twist by $\varphi$ to arrive at $T_M$. The mapping class group $SL(2,\mathbb{Z})$ is the group of duality transformations in the sense of the 4d theory. From this viewpoint, $T_M$ can be obtained by taking a proper “trace” action on a 3d duality wall theory associated with $\varphi$.
In the other approach, one begins by triangulating the mapping torus using a finite number of tetrahedra. Dimofte, Gaiotto and Gukov (DGG) [@Dimofte:2011ju] proposed a systematic algorithm for constructing $T_M$ when the triangulation for $M$ is known. One can construct $T_M$ by applying the DGG algorithm to the known information on the triangulation of $M = \textrm{tori}(\varphi)$.
Duality wall theory {#duality wall theory}
-------------------
Following [@Gaiotto:2008ak; @Terashima:2011qi], we use the notation $T[SU(2),\varphi]$ to denote the 3d theory living on the duality wall between two copies of 4d $SU(2)$ $\mathcal{N}=2^*$ theory associated with an element $\varphi$ of the duality group $SL(2,\mathbb{Z})$.
$q_1\;\;\; q_2 \;\;\;q_3 \;\;\;q_4\;\;\;\phi_0$
-------------------- ---------------------------------------------------------- --
$U(1)_{\rm gauge}$ $1\;\;\;\; 1 \;-1\; -1\;\;\;0$
$U(1)_{\rm bot}$ $1\;-1 \;\;\;1\;-1\;\;\;0$
$U(1)_{\rm punct}$ $\;\half \;\;\;\;\half \;\;\;\;\half \;\;\;\;\half\;-1$
$U(1)_{\rm top}$ $ 0\;\;\;\;0\;\;\;\;0\;\;\;\;0\;\;\;\;0$
: $T[SU(2)]$ theory. $U(1)_{\rm top}$ denotes the topological $U(1)$ charge, $J_{\rm top} = *dA_{U(1)_{\rm gauge}}$.[]{data-label="tsu(2)-charge"}
We begin with the simplest case, $T[SU(2),S]$, often shortened to $T[SU(2)]$. It is the 3d $\mathcal{N}=4$ SQED with two fundamental hyper-multiplets. Let the four chiral fields in the two hyper-multiplets be $q_1 , q_2, q_3,q_4$ and the adjoint chiral field in the vector multiplet be $\phi_0$. The theory has global symmetry $SU(2)_{\rm bot}\times SU(2)_{\rm top}\times U(1)_{\rm punct}$ compatible with 3d $\mathcal{N}=2$ supersymmetries. The charge assignments for chiral fields under the Cartan subalgebras of the gauge and global symmetries are summarized in Table \[tsu(2)-charge\]. $U(1)_{\rm top}$ denotes the topological symmetry whose conserved charge is a monopole charge for the $U(1)_{\rm gauge}$. In the infrared (IR) limit, the $U(1)_{\rm top}$ is known to be enhanced to $SU(2)_{\rm top}$. The quiver diagram for $T[SU(2)]$ is presented in Figure \[fig:quiver\](a). To emphasize that there is an additional quantum $SU(2)$ symmetry, one sometimes draws the quiver diagram as in Figure \[fig:quiver\](b).
![Quiver diagrams for $T[SU(2)]$[]{data-label="fig:quiver"}](quiver.pdf)
Let us consider the generalization to $T[SU(2), \varphi]$ for an arbitrary $\varphi \in SL(2,\mathbb{Z})$. Firstly, multiplying $T^k$ to $S$ corresponds to adding Chern-Simons action of level $k$ for background gauge fields coupled to the $SU(2)$ global symmetries. Explicitly, one obtains the $T[SU(2),\varphi=T^{k}ST^{l}]$ theory by coupling the $T[SU(2)]$ theory with background gauge fields through the CS action of level $k$ for $SU(2)_{\rm top}$ and of level $l$ for $SU(2)_{\rm bot}$. Secondly, multiplication of two mapping class elements $\varphi_1$ and $ \varphi_2$ corresponds to ‘gluing’ $SU(2)_{\rm bot}$ in $T[SU(2),\varphi_1]$ with $SU(2)_{\rm top}$ in $T[SU(2),\varphi_2]$, where ‘gluing’ means gauging the diagonal subgroup. In ultraviolet (UV) region, no $SU(2)_{\rm top}$ symmetry is visible and the gluing procedure can’t be implemented. To make the gluing procedure sensible in UV region, one need to consider a dual description for the $T[SU(2)]$ theory which allows the $SU(2)$ symmetry visible in UV. Some examples of these dual description is given in appendix \[dual description for T\[SU(2)\]\]. Nevertheless, the gauging procedure for supersymmetric partition function can be implemented regardless of UV description choices since the partition function does not depend on the choice. Since $S$ and $T$ generate all elements of $SL(2,\mathbb{Z})$, one can construct all $T[SU(2),\varphi]$ theories by repeatedly using the field theory operations described above.
As a consistency check, we can examine the $SL(2,\mathbb{Z})$ structure of the $T[SU(2),\varphi]$ theory constructed above. $SL(2,\mathbb{Z})$ is generated by $S$ and $T$ subject to the two relations, $$\begin{aligned}
S^4 = (ST)^3 = I \;.
\end{aligned}$$ In the next sections, we will check the equivalence between $T[SU(2), S^4 \varphi]$ and $T[SU(2),\varphi]$ by computing supersymmetric quantities for two theories. On the other hand, the same computations indicate that $T[SU(2),(ST)^3 \varphi]$ can be identified with $T[SU(2),\varphi]$ only after an extra twist, namely,\
\
-3cm *$T[SU(2),(ST)^3 \varphi] \;\; = \;\; T[SU(2),\varphi]$ + $CS$ term with $k=\half$*
*for background gauge field coupled to $U(1)_{\rm punct}$*
\
\
In terms of the 4d $\mathcal{N}=2^*$ theory, the above relation says that $(ST)^3$ induces a $\theta$-term, $\Tr (F_{\rm punct}\wedge F_{\rm punct})$ for the background gauge field coupled to the $U(1)_{\rm punct}$ symmetry which rotates an adjoint hyper.
In the context of 3d-3d correspondence, the duality wall theory is associated to a 3-manifold $\Sigma_{1,1}\times_\varphi I $ called mapping cylinder [@Hosomichi:2010vh; @Terashima:2011qi]. Topologically, a mapping cylinder is a direct product of $\Sigma_{1,1}$ and interval $I=[0,1]$. At two ends of the interval (‘top’ and ‘bottom’), there are two boundary Riemann surfaces denoted as $\Sigma^{\textrm{top}}_{1,1}$ and $\Sigma^{\textrm{bot}}_{1,1}$. In 3d-3d correspondence, global symmetries of $T_M$ are related to the boundary phase space of $M$, $$\begin{aligned}
(\textrm{Rank of global symmetry in $T_M$}) = \half \textrm{dim}_\mathbb{C}\left[ \CM_{SL(2,\mathbb{C})}(\partial M) \right]\;.\end{aligned}$$ The two boundary phase spaces $\CM_{SL(2)}(\Sigma^{\textrm{top}}_{1,1})$ and $\CM_{SL(2)}(\Sigma^{\textrm{bot}}_{1,1})$ are related to $SU(2)_{\textrm{top}}$ and $SU(2)_{\textrm{bot}}$ symmetries, respectively. The phase space associated to the ‘cusp’ boundary made of the puncture on the Riemann surface is related to the $U(1)_{\textrm{punct}}$ symmetry.
![Mapping cylinder $\Sigma_{1,1}\times_\varphi I$. A global symmetry in the duality wall theory is associated to each component of boundary. []{data-label="fig:LR2"}](mappingcylinder.pdf)
The mapping torus, $\textrm{tori}(\varphi) := \Sigma_{1,1}\times_\varphi S^1$, can be obtained by gluing the two boundary Riemann surfaces, $\Sigma^{\textrm{top}}_{1,1}$ and $\Sigma^{\textrm{bot}}_{1,1}$. In the duality wall theory, the gluing amounts to gauging the diagonal subgroup of the two $SU(2)$ global symmetries. The theory obtained by gluing two $SU(2)$’s in a duality wall theory $T[SU(2),\varphi]$ will be denoted as $\Tr (T[SU(2),\varphi])$. From the above discussion, we found a concrete realization of $T_M$ with $M=\textrm{tori}(\varphi)$ in terms of the duality wall theory. The theory will be denoted as $T^{T[SU(2)]}_{\textrm{tori}(\varphi)}$ , $$\begin{aligned}
T^{T[SU(2)]}_{\textrm{tori}(\varphi)} = \Tr (T[SU(2),\varphi])\;.\end{aligned}$$
Tetrahedron decomposition
-------------------------
In [@Dimofte:2011ju], Dimofte, Gaiotto, Gukov (DGG) proposed a powerful algorithm to construct $T_M$ for a broad class of 3-manifolds $M$. We briefly review the DGG algorithm here. The basic building block of a hyperbolic 3-manifold is the ideal tetrahedron $\Delta$. The corresponding 3d theory, $T_{\Delta}$, is a theory of a free chiral field with a background CS action with level $- \frac{1}{2}$. If a 3-manifold $M$ can be triangulated by a finite number of tetrahedra, $T_M$ can be obtained by “gluing” copies of $T_\Delta$ accordingly. Schematically, $$\begin{aligned}
M=\left(\bigcup_{i=1}^N \Delta_i\right)/\sim \quad \Rightarrow \quad T_{M} = \left(\bigotimes_{i=1}^N T_{\Delta_i}\right)/\sim\;.\end{aligned}$$
#### Geometry of tetrahedra
An ideal tetrahedron has six edges and four vertices. To the three pairs of diagonally opposite edges, we assign edge parameters $(z,z',z'')$ which are the exponential of complexified dihedral angles $(Z,Z',Z'')$ of the edges. $$\begin{aligned}
z=\exp (Z)\; \;\textrm{with}\;\; Z= (\textrm{torsion})+i (\textrm{angle}) \,,
\quad {\rm etc.}\end{aligned}$$ Using the equivalence between equation of motion for hyperbolic metrics and $SL(2)$ flat connections on a 3-manifold, these edge variables can be understood in terms of either hyperbolic structure or $SL(2)$ flat connection on a tetrahedron. Although latter interpretation is more physically relevant, the former is more geometrically intuitive. The hyperbolic structure of an ideal tetrahedron is determined by the edge parameters $(Z,Z',Z'')$ subject to the conditions $$\begin{aligned}
Z+Z'+Z''=\pi i +\frac{\hbar}2 \,, \quad
e^{Z}+e^{-Z'} -1 = 0 \,.\end{aligned}$$ The first condition defines the so-called boundary phase space with the symplectic form $\O = \frac{1}{i\hbar}dZ \wedge dZ'$. The second condition defines a Lagrangian submanifold of the boundary phase space. Due to the first condition, the second condition is invariant under the cyclic permutation $(z\goto z' \goto z''\goto z)$.
The ideal triangulation requires that all faces and edges of the tetrahedra should be glued such that the resulting manifold is smooth everywhere except for the cusp due to the vertices of ideal tetrahedra. In particular, we have the smoothness condition at each internal edge, $$\begin{aligned}
C_I = \sum_{j=1}^{N} \left( c_{Ij} Z_i + c'_{Ij} Z'_j + c''_{Ij} Z''_j \right) = 2\pi i + \hbar \,.
\label{internal-edge}\end{aligned}$$ The coefficients $c_{Ij}$, $c'_{Ij}$, $c''_{Ij}$ take values in $\{ 0, 1, 2 \}$.
When all the edges of $\Delta_i$ are glued, all but one of the gluing condition give independent constraints, since the sum of all constraints, $\sum_I C_I=(2\pi i+\hbar )N$ trivially follows from $Z_i + Z'_i + Z''_i = \pi i +\frac{\hbar}2$. The resulting manifold $M$ has a cusp boundary, composed of the truncated ideal vertices, which is topologically a torus $\mathbb{T}^2$. The two cycles of the torus, ‘longitude’ and ‘meridian’, describe the boundary phase space of $M$. The logarithmic variables for the two cycles, $V = \log (\ell)$ and $U = \log m$, are some linear combinations of $(Z_i, Z'_i, Z''_i)$.
#### Ideal triangulation of the mapping torus
The mapping torus $\textrm{tori}(\varphi) =\Sigma_{1,1}\times_{\varphi} S^1$ is known to be hyperbolic when $|\textrm{Tr}(\varphi)|> 2$. The tetrahedron decomposition of these mapping torus is given explicitly in [@Gueritaud]. Any $\varphi \in SL(2,\mathbb{Z})$ satisfying $|\textrm{Tr}(\varphi)|>2$ admits a unique decomposition of the following form, $$\begin{aligned}
\varphi = F\left( \varphi_N \varphi_{N-1}\ldots \varphi_2 \varphi_1 \right) F^{-1} \,, \quad \varphi_i = L \textrm{ or } R\;.
\label{LR-decomp}\end{aligned}$$ where we use the following convention for the $SL(2,\IZ)$ generators, [^6] $$\begin{aligned}
L=
\begin{pmatrix}
1 & 1 \\ 0 & 1
\end{pmatrix} \,,
\;\;
R=
\begin{pmatrix}
1 & 0 \\ 1 & 1
\end{pmatrix} = T \,,
\;\;
S=
\begin{pmatrix}
0 & -1 \\ 1 & 0
\end{pmatrix} \,;
\;\;
L=S^{-1}T^{-1}S \,,
\;\;
S = L^{-1} R L^{-1} \,.
\label{sl2z-convention}\end{aligned}$$ The overall conjugation by $F$ is immaterial in the definition of the mapping torus and can be neglected.
According to [@Gueritaud], to each letter $L$ or $R$ appearing in one can associate a tetrahedron with edge parameters $(Z_i, Z^\prime_i, Z^{\prime\prime}_i)$, or equivalently, $(z_i, z'_i, z''_i) = (e^{Z_i}, e^{Z'_i}, e^{Z''_i})$. The index $i$ runs from 1 to $N$ with cyclic identification, $N+1 \sim 1$. $L$ and $R$ generate ‘flips’ on the triangulation of $\Sigma_{1,1}$. Each flip corresponds to a tetrahedron (see figure 2 in [@Gueritaud]). There are $N$ tetrahedra in total and $3N$ edge parameters. In the mapping torus, all the edges of tetrahedra are glued and there are $N-1$ independent internal edge conditions. How the internal edges are glued together is determined by the decomposition of $\varphi$. Taking account of the $N$ equations $Z_i +Z'_i+Z''_i = i \pi +\frac{\hbar}{2}$ for each $i$, there are in total $2N-1$ linear constraints on $3N$ edge parameters. These $2N-1$ constraints can be solved by parameterizing $3N$ edge parameters by $N+1$ variables ($W_i , V$) as shown in .
$$\begin{array}{c|cccc}
\; \varphi_{i} \varphi_{i-1} \; & Z_i & Z'_i & Z''_i & U_i
\\[5pt]
\hline
\\[-10pt]
LL & \; i \pi +\frac{\hbar}2-W_i \;
& \;W_i- \displaystyle{\frac{W_{i-1}+W_{i+1}}{2}} \;
& \;\displaystyle{\frac{W_{i-1}+W_{i+1}}{2}} \;
& \; 0 \;
\\[15pt]
RR & \;i \pi +\frac{\hbar}2-W_i \;
& \; \displaystyle{\frac{W_{i-1}+W_{i+1}}{2}} \;
& \; W_i- \displaystyle{\frac{W_{i-1}+W_{i+1}}{2}} \;
& \; 0 \;
\\[15pt]
LR & \; i \pi +\frac{\hbar}2-W_i \;
& \; \displaystyle{\frac{W_i +W_{i-1}-W_{i+1}-V-\pi i}{2}} \;
& \; \displaystyle{\frac{W_i -W_{i-1}+W_{i+1}+V+\pi i}{2}} \;
& \; -\displaystyle{\frac{W_i}{2}} \;
\\[15pt]
RL & \; i \pi +\frac{\hbar}2-W_i \;
& \; \displaystyle{\frac{W_i -W_{i-1}+W_{i+1}+V+\pi i}{2}} \;
& \; \displaystyle{\frac{W_i +W_{i-1}-W_{i+1}-V-\pi i}{2}} \;
& \; +\displaystyle{\frac{W_i}{2}} \;
\\[10pt]
\end{array}
\label{tetra-gluing}$$
The particular form of the linear combination depends on the ordering of letters in . The reparametrization is ‘local’ in the sense that the expressions for $(Z_i,Z'_i,Z''_i)$ involve $W_i$ and $W_{i\pm 1}$ only. Among the remaining $N+1$ variables, as was explained below eq. , two are identified as ‘longitude’ and ‘meridian’. In , $\ell=e^V$ is the longitude variable, while the meridian variable $m=e^{U}$ is the product of all $m_i=e^{U_i}$, $U = \sum U_i $. As an example, tetrahedron decomposition of a mapping torus with $\varphi=LR$ is given in figure \[fig:LR\]. In the case $\varphi = LR$, it is known that the mapping torus becomes the figure eight knot complement in $S^3$.
![Triangulation of boundary torus of tori($LR$). Four triangles for each letter, $L$ or $R$, come from ‘small’ boundary triangles in ideal tetrahedron associated to the letter. See figure 3 in [@Gueritaud].[]{data-label="fig:LR"}](figLR.pdf)
From the figure, the two internal edges are identified as $$\begin{aligned}
&C_1 = Z_1+Z_2+2 Z''_1+ 2 Z''_2 = 2\pi i +\hbar\;,{\nonumber}\\
&C_2 = Z_1+Z_2 +2 Z_1'+ 2 Z_2' = 2\pi i +\hbar \;. \label{internal edges for LR}\end{aligned}$$ Longitudinal (horizontal blue line) and meridian (vertical red line) variables from the figure are $$\begin{aligned}
V = Z_2''- Z_2'\;, \quad U= Z_1''-Z_2'\;. \label{long/merd for LR}\end{aligned}$$ Using , we will parameterize $$\begin{aligned}
&Z_1 = i \pi +\frac{\hbar}2 - W_1\;, \; Z_1' = \frac{W_1 +V+i\pi}{2}\;, \; Z_1''= \frac{W_1- V-i\pi}2\;, {\nonumber}\\
&Z_2 = i \pi +\frac{\hbar}2 - W_2\;, \; Z_2' = \frac{W_2 -V-i\pi}{2}\;, \; Z_2''= \frac{W_1+ V+i\pi}2\;. {\nonumber}\end{aligned}$$ In this parametrization, the internal edge conditions are automatically satisfied. From , the meridian variable $U$ is $\frac{W_1 - W_2}2$ which is the same as the meridian variable in via the above parametrization. But the longitudinal variable $V$ in become $V+i \pi $ via the parametrization. The discrepancy $i \pi$ is subtle and the factor can be absorbed by simple redefinition of $V$ in . As we will see in section \[sec: SCI/SL(2,C)\], however, the variable $V$ in has a more direct meaning in the duality wall theory.
#### Field Theory
We will give a very brief summary of the construction of $T_{M, \Pi}$ from the tetrahedron decomposition data for $M$; see [@Dimofte:2011ju; @Dimofte:2011py] for details. For each tetrahedron with a polarization choice $\Pi_\Delta$, we take a copy of the 3d theory $T_{\Delta, \Pi_\Delta}$. For the polarization $\Pi_{\Delta} = \Pi_Z$ in which we take $(Z, Z'')$ as (position, momentum), the theory (often called “the tetrahedron theory") is a free $\mathcal{N}=2$ chiral theory with a background CS term for the $U(1)$ global symmetry at CS level $-\half$. The 3d theory $T_{M,\Pi_M}$ associated a 3-manifold $M= \cup_{i=1}^N \Delta_i/\sim$ and its boundary polarization $\Pi_M$ can be constructed in three steps. First, we start with a direct product of $N$ tetrahedron theories, $$\begin{aligned}
T_{\{\Delta_i\},\{\Pi_{Z_i}\}} = \otimes_{i=1}^N T_{\Delta_i,\Pi_{Z_i}}\end{aligned}$$ Then, we perform a polarization transformation $\tilde{\Pi} = g \circ\{\Pi_{Z_i}\}$ such that all internal edges and positions in $\Pi_M$ become position variables in $\tilde{\Pi}$. In the field theory, the polarization transformation corresponds to an $Sp(2N,\IZ)$ action[^7] involving the $U(1)^N$ global symmetries in $T_{\{\Delta_i\},\{\Pi_{Z_i}\}}$. $$\begin{aligned}
T_{\{\Delta_i\},\tilde{\Pi}} = g\circ T_{\{\Delta_i\},\{\Pi_{Z_i}\}} \;.\end{aligned}$$ Finally, we impose the internal edges conditions by adding the superpotential $\CW= \sum \CO_I$ which breaks the global $\otimes_I U(1)_I$ symmetries of $T_{\{\Delta_i\},\tilde{\Pi}}$ associated to internal edges $C_I$. This completes the consturction of $T_{M,\Pi}$: $$\begin{aligned}
T_{M,\Pi_M} &= T_{\{\Delta_i\},\tilde{\Pi}} \textrm{ with superpotentail }\CW = \sum \CO_I \;.\end{aligned}$$ Applying this general algorithm to the case $M= \textrm{tori}(\varphi)$ using the tetrahedron decomposition data described above, we give a description for the mapping torus theory. We will denote the mapping torus theory by $T^{\Delta}_{\textrm{tori}(\varphi)}$.
Quantization of Chern-Simons theory on Riemann Surface {#quantum riemann surface}
======================================================
In this section, we will explain the classical phase space and its quantization that are relevant to a calculation of Chern-Simons (CS) partition function on a mapping torus/cylinder. Most parts of this section are reviews of known results but are included to make the paper self-contained. More details can be found, [*e.g.*]{}, in [@Dimofte:2011gm; @Dimofte:2011jd].
For a compact gauge group $G$, the Chern-Simons action on a 3-manifold $M$ is $$\begin{aligned}
I_{CS} = \frac{k}{4\pi }\int_M \Tr (A\wedge dA + \frac{2}3 A\wedge A \wedge A)\;,\end{aligned}$$ where $k$ is a quantized CS level. This is one of the most famous example of topological quantum field theory (TQFT). When $M$ is a mapping torus, the CS theory can be canonically quantized on $\Sigma_{1,1}$ regarding the $S^1$ direction as time. The phase space $\mathcal{M}_G$[^8] canonically associated to $\Sigma_{1,1}$ is [@Elitzur:1989] $$\begin{aligned}
\mathcal{M}_ G (\Sigma_{1,1})= \{\textrm{Flat $G$-connections on $\Sigma_{1,1}$ with fixed puncture holonomy $P$} \}/\sim\;. \label{flat connection moduli space}\end{aligned}$$ where $\sim$ denotes the gauge equivalence. The conjugacy class of gauge holonomy $P$ around a puncture is fixed as a boundary condition. The symplectic form $\O_G$ on $\CM_G$ derived from the CS action is $$\begin{aligned}
\Omega_{G} =\frac{k}{4\pi} \int_\Sigma \textrm{Tr}( \delta A \wedge \delta A)\;. \label{symplectic form for unitary}\end{aligned}$$ One can geometrically quantize the classical phase space and obtain a Hilbert-space $\mathcal{H}_G (\Sigma_{1,1})$. Following an axiom of general TQFTs (see, [*e.g.*]{}, [@Atiyah:1988]), the CS partition function on the mapping cylinder with gauge group $G$ can be computed as $$\begin{aligned}
Z_{\Sigma_{1,1}\times_\varphi I}(x_{\rm bot}, x_{\rm top}) = \langle x_{\rm bot} |\varphi | x_{\rm top}\rangle \;, \label{CS partition function as matrix element}\end{aligned}$$ which depends on the boundary conditions $\big{(}x_{\rm bot}, \varphi ( x_{\rm top})\big{)}$ on two boundary $\Sigma_{1,1}$’s . The CS partition function on the mapping torus with gauge group $G$ can be computed as $$\begin{aligned}
Z_{\textrm{tori}(\varphi)}(G) = \textrm{Tr} (\varphi)\; \textrm{over $\CH_G (\Sigma_{1,1})$}. \label{CS partition function as trace}\end{aligned}$$ Here $\varphi$ is an operator acting on the Hilbert space $\CH_G$ obtained from quantizing a mapping class group element $\varphi$ which generates a coordinate transformation on $\mathcal{M}_G$.
When the gauge group is $G= SL(2,\mathbb{C})$, the CS level becomes complex variables $t=k+i s$ and $\tilde{t} = k - i s$. $k$ should be an integer for consistency of the quantum theory and unitarity requires $s \in \mathbb{R}$ or $s \in i \mathbb{R}$ [@Witten:91]. $$\begin{aligned}
I_{CS} = \frac{t}{8\pi }\int_M \Tr (A\wedge dA + \frac{2}3 A\wedge A \wedge A)+\frac{\tilde{t}}{8\pi }\int_M \Tr (\bar{A}\wedge d\bar{A} + \frac{2}3 \bar{A}\wedge \bar{A} \wedge \bar{ A})\;.\end{aligned}$$ The induced symplectic form from the CS action is $$\begin{aligned}
\Omega_{SL(2,\mathbb{C})} =\frac{t}{8\pi }\int_\Sigma \textrm{Tr}( \delta A \wedge \delta A) + \frac{\tilde{t}}{8\pi} \int_\Sigma \textrm{Tr}(\delta \bar{A} \wedge \delta\bar{A})\;.\end{aligned}$$ In [@Dimofte:2011py], the superconformal index $I_{T_M}$ for the theory $T_M$ was claimed to be equivalent to the $SL(2,\mathbb{C})$ CS partition function on $M$ with $t=- \tilde{t} = i s \in i \mathbb{R}$. $$\begin{aligned}
I_{T_M} = Z_M(SL(2,\mathbb{C}))\;, \;\textrm{with $q:=e^{\hbar } :=e^{\frac{4\pi}s}$}\;. \label{3d-3d dictionary for SCI}\end{aligned}$$ Here, $q$ is a fugacity variable in the superconformal index to be explained in section \[sec: SCI/SL(2,C)\]. Using the above map, the symplectic form becomes $$\begin{aligned}
\Omega_{SL(2,\mathbb{C})} := \frac{i}{2\hbar}\int_\Sigma \textrm{Tr}(\delta A\wedge \delta A) - \frac{i}{2\hbar}\int_\Sigma \textrm{Tr}(\delta \bar{A}\wedge \delta \bar{A})\;. \label{SL(2,C) symplectic form}\end{aligned}$$
Classical phase space and its coordinates {#sec: Classical phase space}
-----------------------------------------
In this subsection, we will review coordinate systems for the phase space $(\mathcal{M},\O)_{SL(2,\mathbb{C})}$ on once-punctured torus $\Sigma_{1,1}$ and the action of $\varphi \in SL(2,\mathbb{Z})$ on these coordinates. We also consider a phase space $(\CM, \O)^{\rm knot}_{SL(2,\mathbb{C})}$ which is canonically associated to the cusp boundary $\mathbb{T}^2$ of mapping torus. For mapping cylinder, the boundary phase space is given by, at least locally, $\CM (\Sigma_{1,1})^2 \times \CM^{\rm knot}$.
#### Loop coordinates
Generally speaking, the moduli space of flat connections on manifold $M$ with gauge group $G$ is parametrized by holonomy variables up to conjugation. In other words, $$\begin{aligned}
\CM_G (M) = \textrm{Hom}(\pi_1 (M), G)/\textrm{conj}. \label{general flat moduli space}\end{aligned}$$ The fundamental group for $\Sigma_{1,1}$ is $$\begin{aligned}
\pi_1 (\Sigma_{1,1}) = \{ A , B,P | A B A^{-1}B^{-1} = P \}\;.\end{aligned}$$ Here $A,B$ are two cycles of the torus and $P$ denotes the loop around the puncture. Thus $\CM_{SL(2,\mathbb{C})}(\Sigma_{1,1})$ is given by $$\begin{aligned}
\CM_{SL(2,\mathbb{C})} = \{ A, B\in SL(2,\mathbb{C}) | ABA^{-1}B^{-1} = P \}/\textrm{conj}\;. \label{MSL(2,C) in terms of holonomies}\end{aligned}$$ Conjugacy class of the holonomy $P$ around the puncture is fixed by the following condition $$\begin{aligned}
\textrm{eigenvalues of $P$} = \{\ell , \ell^{-1} \} \;.\end{aligned}$$ Loop coordinates $(W,H,D)$ on $\CM_{SL(2,\mathbb{C})}$ are defined as trace of these holonomy variables $$\begin{aligned}
W = \Tr(A)\; , \quad H= \Tr(B)\; , \quad D = \Tr(AB)\;.\end{aligned}$$ They are not independent and subject to the follwoing constraint: $$\begin{aligned}
W^2 + H^2 + D^2 - WHD +\ell +\ell^{-1}-2=0 \;.
\end{aligned}$$ Anticipating close relations to gauge theory observables, we call the three loop coordinates Wilson loop ($W$), ‘t Hooft loop ($H$) and dyonic loop ($D$). [^9]
#### Shear coordinates
The shear coordinates $(\sqrt{t},\sqrt{t'},\sqrt{t''})$ are associated to the three edges appearing in the ideal triangulation of $\Sigma_{1,1}$ depicted in Figure \[fig:shear\].
![Triangulation of once-punctured torus, $\Sigma_{1,1}$, represented as $(\mathbb{R}^2 - \mathbb{Z}^2 )/\mathbb{Z}^2$. Each circle in the figure denotes an image of the puncture in the covering space $\IR^2$.[]{data-label="fig:shear"}](shear.pdf)
Naively, the shear coordinates represent the partial holonomy eigenvalues along a path crossing each edge. For more precise description of the shear coordinates, see, [*e.g.*]{}, [@Fock:1993pr; @Fock:1997]. Following the description in the references, the holonomy $A,B$ can be expressed in terms of the shear coordinates as follows, $$\begin{aligned}
&A(t,t',t'') = \CE(t)\CV \CE^{-1}(t')\CV^{-1} =\left(\begin{array}{cc}\sqrt{tt'} & \sqrt{t t'} \\ \sqrt{\frac{t'} t} & \frac{1}{\sqrt{tt'}}+\sqrt{\frac{t'}t}\end{array}\right) {\nonumber}\\
&B(t,t',t'') = \CE(t)\CV^{-1} \CE^{-1}(t'')\CV =\left(\begin{array}{cc}\sqrt{\frac{t}{t''}}+ \sqrt{tt''} & \sqrt{\frac{t}{t''}} \\ \frac{1}{\sqrt{t t''}} & \frac{1}{\sqrt{t t''}}\end{array}\right) \,,\label{A,B matrices}\end{aligned}$$ where $$\begin{aligned}
\CE(z):= \left(\begin{array}{cc}0 & z^{1/2} \\- z^{-1/2} & 0\end{array}\right)\;, \quad \CV:= \left(\begin{array}{cc}1 & 1 \\-1 & 0\end{array}\right)\;.\end{aligned}$$ The relation $ABA^{-1}B^{-1} = P$ in holds provided that $$\begin{aligned}
\sqrt{t}\sqrt{ t'}\sqrt{ t''} = \sqrt{-\ell} \;.
\label{conditions on shears}\end{aligned}$$ This relation states that products of partial holonomies around the three edges give the square root of holonomy around a puncture. The logarithmic shear variables $(T,T',T'')$ are defined as $$\begin{aligned}
(e^{T/2},e^{T'/2},e^{T''/2}) = (\sqrt{t},\sqrt{t'},\sqrt{t''})\;.\end{aligned}$$ Note that the matrices ($A, B$) in eq. are invariant under individual shifts of $T,T',T''$ by $4\pi i $.[^10] $$\begin{aligned}
T \sim T+4\pi i\;, \; T' \sim T'+4\pi i \;, \; T'' \sim T''+4\pi i \;. \label{periodicity in shear}\end{aligned}$$ In the logarithmic variables, the condition become $$\begin{aligned}
T+T'+T''= i \pi +V \;. \label{conditions on logarithmic shears}\end{aligned}$$ From this, we see $V$ is also periodic variable with periodicity $4 \pi i$. The symplectic form takes a simple form in the shear coordinates. $$\begin{aligned}
\Omega_{SL(2,\mathbb{C})} &=- \frac{i}{2\hbar } \frac{dt}{t}\wedge \frac{dt'}{t'} + \frac{i}{2\hbar } \frac{d\bar{t}}{\bar{t}}\wedge \frac{d\bar{t'}}{\bar{t'}} =- \frac{i}{2\hbar} dT\wedge dT' + \frac{i}{2\hbar} d\bar{T}\wedge d\bar{T'} {\nonumber}\\
&= (\textrm{cyclic permutation of } T\rightarrow T' \rightarrow T'')\;. \label{symplectic form in shear}\end{aligned}$$ The $SL(2,\IZ)$ generators act on the shear coordinates as follows. $$\begin{array}{c|ccc}
\;\;\varphi\;\; & \sqrt{t} \mapsto & \sqrt{t'} \mapsto & \;\; \sqrt{t''} \mapsto \;\;
\\[5pt]
\hline
\\[-10pt]
\;\;S\;\; & \;\; \displaystyle{\frac{1}{\sqrt{t}}} \;\; & \;\; \displaystyle{\frac{\sqrt{t''}}{1+ t^{-1}}} \;\; & \sqrt{t'}(1+t)
\\[12pt]
\;\;L\;\; & \;\; \displaystyle{\frac{1}{\sqrt{t''}}} \;\; & \;\; \displaystyle{\frac{\sqrt{t'}}{1+ t''^{-1}}} \;\; & \sqrt{t}(1+t'')
\\[12pt]
\;\;R\;\; & \;\; \displaystyle{\frac{1}{\sqrt{t'}}} \;\; & \;\; \displaystyle{\frac{ \sqrt{t}}{1+ t'^{-1}}} \;\; & \sqrt{t''}(1+t')
\end{array}
\label{c-shear-transf}$$
#### Fenchel-Nielson coordinates
We adopt the modified Fenchel-Nielson (FN) coordinates defined in [@Dimofte:2011jd]. Classically, the FN coordinates $(\lambda,\tau):=(\exp \Lambda, \exp \mathcal{T})$ and the shear coordinates are related by $$\begin{aligned}
\sqrt{t} = \frac{i(\tau^{-1/2} -\tau^{1/2})}{\l -\l^{-1}} \,, \;\;
\sqrt{t'} = \frac{i(\l -\l^{-1})}{\l^{-1}\t^{1/2} - \t^{-1/2} \l} \,, \;\;
\sqrt{t''} = \frac{i\sqrt{\ell}(\l^{-1}\t^{1/2} - \t^{-1/2} \l)}{\t^{-1/2} -\t^{1/2}} \,.
\label{shear2FN}
\end{aligned}$$ The FN coordinates are defined up to Weyl-reflection $\mathbb{Z}_2$, whose generator $\sigma$ acts as $$\begin{aligned}
\sigma \; : \; (\Lambda, \mathcal{T}) \; \rightarrow \; (-\Lambda,- \mathcal{T})\;. \label{Weyl-reflection}\end{aligned}$$ Note that the Weyl reflection leaves the shear coordinates invariant in the relation .
#### Phase space $(\CM,\Omega)^{\textrm{knot}}_{SL(2,\mathbb{C})}$
So far we have considered a phase space associated to the Riemann surface $\Sigma_{1,1}$ in the $\textrm{tori}(\varphi)$. There is another important phase space in the computation of the CS partition function on $\textrm{tori}(\varphi)$ which is associated to the cusp boundary $\mathbb{T}^2= \partial (\textrm{tori}(\varphi))$. We denote this phase space by $\CM^{\textrm{knot}}_{SL(2,\mathbb{C})}
:= \CM_{SL(2,\mathbb{C})}(\mathbb{T}^2)$ and parametrize it by the (logarithmic) holonomy variables $(U,V)$[^11] along the two cycles of $\mathbb{T}^2$. The symplectic form $\Omega^{\textrm{knot}}$ is given by $$\begin{aligned}
\Omega^{\textrm{knot}}_{SL(2,\mathbb{C})} = \frac{1}{i\hbar} dU \wedge dV-\frac{1}{i\hbar} d\bar{U} \wedge d\bar{V}\;. \label{Symplectic for knot}\end{aligned}$$
#### Boundary phase space of mapping cylinder
The boundary is the genus two Riemann surface without puncture, $\partial (\Sigma_{1,1} \times_\varphi I) = \Sigma_{2,0}$, which can be obtained by gluing punctures in two once-punctured tori. The fundamental group for $\Sigma_{2,0}$ is $$\begin{aligned}
\pi_1 (\Sigma_{2,0}) = \{ A_1, B_1, A_2, B_2 | A_1 B_1 A_1^{-1}B_1^{-1} A_2 B_2 A_2^{-1} B_2^{-1} = 1 \}\;.\end{aligned}$$ The non-trivial cycles $(A_i, B_i)$s on $\Sigma_{2,0}$ are depicted in Figure \[fig:genus2\].
![Generators of $\pi_1 (\Sigma_{2,0})$[]{data-label="fig:genus2"}](genus2.pdf)
\
Thus, by eq. $$\begin{aligned}
\mathcal{M}_{SL(2,\mathbb{C})} (\Sigma_{2,0})= \{ A_1, B_1, A_2, B_2 \in SL(2,\mathbb{C})| A_1 B_1 A_1^{-1}B_1^{-1} A_2 B_2 A_2^{-1} B_2^{-1} = 1 \}/{\rm conj} \;.\end{aligned}$$ The corresponding phase space $\CM(\Sigma_{2,0})$ can be sliced by a constant $P:=A_1 B_1 A_1^{-1}B_1^{-1}=(A_2 B_2 A_2^{-1}B_2^{-1})^{-1}$ surface. In the slice, the phase space locally looks like a $GL(1,\mathbb{C})$ bundle of two copies of $\CM(\Sigma_{1,1})$ with the same $\ell=e^{V}$, an eigenvalue of $P$. The $GL(1,\mathbb{C})$ fiber direction corresponds to opposite $GL(1,\mathbb{C})$ conjugation action on a representative elements $(A_1, B_1)$ and $(A_2, B_2)$ of the two $\CM(\Sigma_{1,1})$’s, where the $GL(1,\mathbb{C})\subset SL(2,\mathbb{C})$ is the stabilizer subgroup of $P$. Locally, the boundary phase space looks like $\CM(\Sigma_{1,1})^2 \times \CM^{\rm knot}$ where $\CM^{\rm knot}$ is parameterized by conjugacy class of $P$ (or equivalently $V$) and the $GL(1,\mathbb{C})$ fiber direction. In total, $\textrm{dim}_\mathbb{C} \left[\CM_{SL(2,\mathbb{C})} \big{(}\partial (\Sigma_{1,1}\times_\varphi I )\big{)}\right] = 6$. As is obvious from the construction of $\CM \big{(}\partial (\Sigma_{1,1}\times_\varphi I )\big{)}$, the $V$ variable in $\CM \big{(}\partial (\Sigma_{1,1}\times_\varphi I )\big{)}$ can be identified with the puncture variable $V$ in $\CM(\Sigma_{1,1})$. Considering the procedure of gluing two boundary $\Sigma_{1,1}$ components in $\Sigma_{1,1}\times_\varphi I$ to form a mapping torus $\Sigma_{1,1}\times_\varphi S^1$, the $V$ variable can also be identified with the longitudinal variable $V$ in $\CM^{\rm knot} = \CM \big{(}\partial (\Sigma_{1,1}\times_\varphi S^1) \big{)}$.[^12]
A-polynomial {#A-polynomial for mapping torus}
------------
Consider a 3-manifold $M$ with boundary $\partial M$. Obviously, the moduli space of $SL(2,\mathbb{C})$ flat connections on $M$ can be thought of as a submanifold of the moduli space on the boundary. $$\begin{aligned}
\CM(M) \subset \CM(\partial M)\;.\end{aligned}$$ In fact, the submanifold is Lagrangian with respect to the symplectic form . For a knot-complement $M$, the moduli space $\CM(M)$ is $(\mathbb{C}^*)^2/\mathbb{Z}_2$ parametrized by ‘longitude’ and ’meridian’ variable $(\ell, m)$ modulo a Weyl-reflection $(\ell , m)\sim (\ell^{-1},m^{-1})$. In this case the Lagrangian submanifold is given by the vanishing locus of the so-called “A-polynomial", $A(\ell, m)$ [@Gukov:2003na]. Mapping torus is one example of knot-complement. In this section we will analyze the A-polynomial for mapping torus from two different approaches and show their equivalence.
#### From tetrahedron decomposition
In , we presented a solution to the gluing conditions for the mapping torus by parametrizing the $3N$ edge parameters $(Z_i,Z'_i,Z''_i)$ for the $N$ tetrahedra in terms of $N+1$ parameters ($W_i,V$). The A-polynomial for the mapping torus can be obtained by imposing additional $N$ non-linear constraints, $e^{Z_i}+e^{-Z'_i} -1 = 0$, and eliminating all $W_i$’s in favor of $\ell= e^V$ and $m = \prod e^{U_i}$.
For instance, consider the simplest example, $\varphi= LR$ ($\varphi_2 = L, \varphi_1 = R$). From , we find ($w_i := e^{W_i}$) $$\begin{aligned}
&z_1 = -\frac{1}{w_1}\,, \quad
z'_1 = \sqrt{w_1 (-\ell)}\,, \quad
z''_1 = \sqrt{\frac{w_1}{(-\ell)}}\,,
\nonumber
\\
&z_2 = -\frac{1}{w_2}\,, \quad
z'_2 = \sqrt{\frac{w_2}{(-\ell)}}\,, \quad
z''_2 = \sqrt{w_2 (-\ell)}\,,
\qquad
m = \sqrt{\frac{w_2}{w_1}} \,.\end{aligned}$$ For the boundary phase spaces of the two tetrahedra, the equations for the Lagrangian submanifolds ($z'+ (z'')^{-1} -1 =0$) are $$\begin{aligned}
\sqrt{w_1(-\ell)} + \sqrt{\frac{(-\ell)}{w_1}} -1=0\,, \quad
\sqrt{\frac{w_2}{(-\ell)}} +\frac{1}{\sqrt{w_2(-\ell)}} -1=0\;.\end{aligned}$$ Eliminating $w_1$ and $w_2$ in favor of $\ell$ and $m$, we obtain the A-polynomial for the mapping torus with $\varphi=LR$: $$A(\ell,m) = \ell +\ell^{-1} -( m^{-2} - m^{-1} -2 - m + m^2) = 0 \; ,$$ which coincide with A-polynomial for figure eight knot complement; see [@Gukov:2003na].
#### From Lagrangian submanifold for mapping cylinder
The Lagrangian submanifold for a mapping torus is a ‘diagonal’ subspace of a Lagrangian submanifold for the corresponding mapping cylinder [@Dimofte:2011jd]. As explained above, the boundary phase space for the mapping cylinder contains a product of two phase spaces associated with two $\Sigma_{1,1}$’s at the two ends of the interval $I$, $$\begin{aligned}
\mathcal{M}^{\textrm{bot}}(\Sigma_{1,1})\times \mathcal{M}^{\textrm{top}}(\Sigma_{1,1})\subset \CM \big{(}\partial (\Sigma_{1,1}\times_\varphi I )\big{)} \;,\end{aligned}$$ where we labelled the two Riemann surface at the two ends of the interval $I$ by ‘top’ and ‘bot’(bottom). The phase space $\CM^{\textrm{bot,top}}$ can be parametrized by two copies of shear coordinates $(t,t',t'')_{\textrm{bot,top}}$ with common $\ell$. The Lagrangian submanifold $\CL_{\varphi}$ for mapping cylinder is [@Dimofte:2011jd] $$\begin{aligned}
\CL_\varphi = \{ t_{\textrm{top}} - \varphi_*(t)_{\textrm{bot}} =0 , t'_{\textrm{top}} - \varphi_*(t')_{\textrm{bot}} =0,t''_{\textrm{top}} - \varphi_*(t'')_{\textrm{bot}} =0\} \subset \mathcal{M}^{\textrm{bot}} \times \mathcal{M}^{\textrm{top}}\;. \label{Langrangian submanifold for mappying cylinder}\end{aligned}$$ Here $\varphi_*( O)$ is a coordinate for $\CM_{SL(2,\mathbb{C})}$ which is related to $O$ by a mapping class group element $\varphi \in SL(2,\mathbb{Z})$. For example, $\varphi_* (t) =1/t'' $ when $\varphi= L$ as one can see in . Among three equations between braces, only two equations are independent and the remaining one is automatically satisfied due to the relation $tt't'' = -\ell$. Then, the Lagrangian submanifold for mapping torus is $$\begin{aligned}
\CM ( \textrm{tori}(\varphi)) = \CL_{1} \cap \CL_{\varphi}\;.\end{aligned}$$ $ \CL_{1}$ denotes a diagonal subspace of $\mathcal{M}^{\textrm{bot}} \times \mathcal{M}^{\textrm{top}}$ which can be interpreted as the Langrangian submanifold for mapping cylinder with $\varphi=1$; see eq. .
However, the above construction for the Langrangian submanifold of $\textrm{tori}(\varphi)$ is incomplete since all the algebraic equations depend on $\ell$ but not on $m$. We need an additional algebraic relation involving $m$. As we saw above, $U=\log m$ is conjugate to $V = \log \ell$. Anticipating the consequences of quantization, which promotes $m$ to a shift operator ($m: \ell
\rightarrow q\ell$), we propose the following prescription for $m$; see . $$\begin{aligned}
&m = 1\; , \quad \textrm{for $\varphi=R$} {\nonumber}\\
&m = \frac{\sqrt{t}_{\textrm{top}}}{\sqrt{t}_{\textrm{bot}}}\;, \quad \textrm{for $\varphi=L$}\;. \label{equations involving m}\end{aligned}$$ Consider general $\varphi$ which can be written as product of $L$ and $R$, $$\begin{aligned}
\varphi = \varphi_{N}\cdots \varphi_1\;, \quad \varphi_i = L \textrm{ or }R\;.\end{aligned}$$ For each letter $\varphi_i$, we assign a mapping cylinder $\Sigma_{1,1}\times_{\varphi_i} I$ whose boundary phase space associated to two boundary $\Sigma_{1,1}$’s is parameterized by $(t_i, t'_i, t''_i)_{\textrm{bot},\textrm{top}}$. To glue $\Sigma^{\textrm{bot}}_{1,1}$ of $i$-th mapping cylinder with $\Sigma^{\textrm{top}}_{1,1}$ of $i+1$-th mapping cylinder, we parametrize $$\begin{aligned}
&(t_i, t'_i, t''_i)_{\textrm{bot}} =(t_{i+1}, t'_{i+1}, t''_{i+1})_{\textrm{top}} := (t_{i+1} , t_{i+1}', t_{i+1} '')\;.\end{aligned}$$ $i$ is a cyclic parameter running from 1 to $N$, $N+1 \sim 1$. In the parametrization, the Lagrangian in becomes $$\begin{aligned}
\CL_{\varphi_i} (t^*_i ,t^*_{i+1}) = \left\{\begin{array}{ll} \{ \sqrt{t_i} - \frac{1}{\sqrt{t''_{i+1}}}=0, \; \sqrt{t''_i} - \sqrt{t_{i+1}} (1+t''_{i+1}) =0\}\;, &\quad \varphi_i = L \\
\{ \sqrt{t_i } - \frac{1}{\sqrt{t'_{i+1}}}=0, \;\sqrt{t'_i} - \sqrt{t_{i+1}}/ (1+1/t'_{i+1}) =0\} \;, & \quad \varphi_i = R \label{Lag for L, R-1}
\end{array}\right.\end{aligned}$$ Equation involving $m$ for mapping cylinder $\Sigma_{1,1}\times_\varphi I $ is $$\begin{aligned}
&m= \prod_{i=1}^N m_i\;, \quad \textrm{where} {\nonumber}\\
&m_i = \left\{\begin{array}{ll}1 \;,& \quad \textrm{$\varphi_i$ = R}\\ \frac{\sqrt{t_i}}{\sqrt{t_{i+1}}} \;,& \quad \textrm{$\varphi_i$ = L}\end{array}\right. \label{Lag for L, R-2}\end{aligned}$$ Since $m$ act as a shift operator, $m$ for $\varphi_N \ldots \varphi_1$ is product of $m_i$ for each $\varphi_i$. Then, the A-polynomial for tori($\varphi$) is given by solving all equations in , in terms of $(\ell , m)$.
#### Equivalence of two approaches
In the above, we explained two ways of calculating the A-polynomial for the mapping torus. The equivalence of the two approaches can be explicitly shown by i) finding a map between $(w_i)$ variables in the first approach and $(t^*_i)$ variables in the second approach and ii) showing that the equations , in the second approach are either trivially satisfied or mapped into equations in the first approach. The map we found is $$\begin{array}{c|ccc}
\;\; \varphi_{i-1} \;\; & t_{i} & t'_{i} & t''_{i}
\\[5pt]
\hline
\\[-10pt]
\; L \; & \;\;\; \displaystyle{\frac{1}{w_{i}}} \;\;
& \; (-\ell) \displaystyle{\frac{w_{i}}{w_{i-1}}}\; & \; w_{i-1} \;
\\[12pt]
\; R\; & \;\;\; \displaystyle{\frac{1}{w_{i}}} \;\;
& \; w_{i-1} \; & \; (-\ell) \displaystyle{\frac{w_{i}}{w_{i-1}}}\;
\end{array}
\label{shear-w}$$ From the fact that a single flip ($\varphi_{i-1}$ =$L$ or $R$) generate a tetrahedron whose edges are originated from edges in a triangulation of $\Sigma_{1,1}$, identification of $\frac{1}{w_i}$ (which is $-z_i$) with $t_i$ is understandable. For $\varphi_i=L$, the map ensures that the transformation rule $\sqrt{t_{i}} = 1/\sqrt{t''_{i+1}}$ in is trivially satisfied. The other transformation rule, $\sqrt{t''_{i}} = \sqrt{t_{i+1}} (1+t_{i+1}'')$, is equivalent to the constraint $z''_i + (z_i)^{-1} - 1 = 0$. Similarly, for $\varphi_i=R$, the transformation rule $\sqrt{t_{i}} = 1/\sqrt{t'_{i+1}}$ in is trivially satisfied, while $\sqrt{t'_{i}} = \sqrt{t_{i+1}}/(1+1/t'_{i+1})$ is equivalent to $z''_i + (z_i)^{-1} - 1 = 0$. Finally, we note that the meridian variable $m$ can be written in terms of the shear coordinates as $$\begin{aligned}
m= \frac{ \prod_{\varphi_{i} \varphi_{i-1} =RL} \sqrt{w_i} }{ \prod_{\varphi_{i} \varphi_{i-1} = LR} \sqrt{w_i}}
&= \prod_{\varphi_i =L} \frac{\sqrt{w_{i+1}}}{\sqrt{w_{i}}}
= \prod_{\varphi_{i} =L} \frac{\sqrt{t_{i}}}{\sqrt{t_{i+1}}}\; ,\end{aligned}$$ which is the same as the last equation in .
Thus we have proved the classical equivalence of the two approaches using the A-polynomial. This classical equivalence was already observed in [@Terashima:2011xe]. In section \[Z(Tr)=Z(Delta)\], we will prove the equivalence at the quantum level by computing the $SL(2,\mathbb{C})$ CS partition function from the two approaches and confirming an exact agreement. The quantized $(\ell, m)$ variables, denoted as ($\fl, \sm$), act as difference operators on the CS partition function. The quantum A-polynomial $A(\fl ,\sm;q)$ annihilates the CS partition function. Taking the classical limit, $q\rightarrow 1$, we obtain the A-polynomial, $A(\ell, m)$ discussed in this section.
Quantization for $G=SL(2,\IC)$ \[qsl2c\]
-----------------------------------------
In this section, we will quantize the classical phase spaces $(\mathcal{M},\Omega)_{SL(2,\mathbb{C})}$ and $(\mathcal{M},\Omega)^{\rm knot}_{SL(2,\mathbb{C})}$. By quantization of the classical phase space $(\mathcal{M},\Omega)$, we mean finding the following maps: $$\begin{aligned}
&\textrm{Classical phase space }\mathcal{M} \quad \rightarrow \quad \textrm{Hilbert-space }\CH {\nonumber}\\
&\textrm{Observables (functions of coordinate $x_i$) $O(x_i)$ } \quad \rightarrow \quad \textrm{Operators $\sO(\sx_i)$ acting on } \CH {\nonumber}\\
&\textrm{Poisson bracket $\{O_1, O_2 \} = O_3 $} \quad \rightarrow \quad \textrm{Commutation relation $[\sO_1, \sO_2] = \sO_3$} \;.\end{aligned}$$ The two main ingredients of quantization are the Hilbert-space $\CH$ and operators $\{\sO \}$ acting on it. In this section, we will focus on the latter. Operators and their commutation relations can be considered before constructing a concrete Hilbert-space. The construction of the Hilbert-space will be given in section \[sec: SCI/SL(2,C)\] .
#### Extended shear operators
After quantization, the shear coordinates $T,T',T''$ (and its conjugations) become operators $$\begin{aligned}
&(T,T',T'')\rightarrow (\sT_+, \sT'_+,\sT''_+)\;, {\nonumber}\\
&(\bar{T},\bar{T}',\bar{T}'')\rightarrow (\sT_-, \sT'_-,\sT''_-)\;.\end{aligned}$$ The commutation relations for these shear operators follow from the symplectic form , $$\begin{aligned}
&[\sT_\pm, \sT'_\pm ] = [\sT'_\pm, \sT''_\pm ] = [\sT''_\pm, \sT_\pm ] =\pm 2\hbar \,.
\\
& [\sT^*_\pm , \sT^{**}_\mp] =0\;, \quad \textrm{for any }*,**\;.\end{aligned}$$ The exponentiated operators $(\sqrt{\st} = e^{\half \sT}, \sqrt{\st'} = e^{\half \sT'}, \sqrt{\st''} = e^{\half \sT''})$ satisfy $$\begin{aligned}
\sqrt{\st} \sqrt{\st'} = q^{ \half} \sqrt{\st'}\sqrt{\st} \,,
\quad
\sqrt{\st'}\sqrt{\st''} = q^\half \sqrt{\st''}\sqrt{\st'} \,,
\quad
\sqrt{\st''} \sqrt{\st'} = q^\half \sqrt{\st'}\sqrt{\st''} \, .\end{aligned}$$ Recall that the quantum parameter $q$ is defined as $q:=e^{\hbar}$. From here on, we will ignore the subscript $(\pm)$ and all expressions will be for $(+)$ operators unless otherwise stated. The same expressions hold for the $(-)$ operators upon replacing $q$ by $q^{-1}$.
Quantizing , shear operators are subject to the following central constraint. $$\begin{aligned}
\sqrt{\st} \sqrt{\st'} \sqrt{\st''} = q^{ \frac{1}4}\sqrt{-\ell}\;.\end{aligned}$$ In the logarithmic shear operators, the constraint becomes $$\begin{aligned}
\sT + \sT' + \sT'' = V +i\pi \;.\end{aligned}$$ In the literature, the variable $V$ is usually regarded as a central charge since they are focusing on the Riemann surface $\Sigma_{1,1}$ itself where $V$ is a fixed parameter. But, when considering mapping cylinder or torus, we need to elevate $V$ to a quantum operator $\sV$ and introduce its conjugate operator $\sU$ satisfying $$\begin{aligned}
[\sU, \sV] = \hbar\;, \label{uv-comm}\end{aligned}$$ since $V$ appears as a dynamical variable (a coordinate for the boundary phase space). Then, the central constraint is promoted to an operator relation $$\begin{aligned}
\sT + \sT' + \sT'' = \sV +i\pi \,.
\label{q-cent}\end{aligned}$$ Since $\sV$ originates from the central constraint, it is natural to assume that $$\begin{aligned}
[ \sV , \sT ] = [ \sV, \sT' ] = [\sV, \sT'' ] = 0\,.\end{aligned}$$ We cannot require that $\sU$ commute with all three shear coordinates; that would contradict with and . The best we can do is to demand that $\sU$ commutes with two of the shear coordinates and to determine the last commutator with and . For instance, $$\begin{aligned}
[ \sU, \sT ] = [ \sU, \sT']= 0 \quad \imp \quad [\sU,\sT''] = \hbar \,.\end{aligned}$$ Alternatively, we may choose $$\begin{aligned}
&[ \sU', \sT' ] = [ \sU', \sT'']= 0 \quad \imp \quad [\sU',\sT] = \hbar \,,
{\nonumber}\\
&[ \sU'', \sT'' ] = [ \sU'', \sT]= 0 \quad \imp \quad [\sU'',\sT'] = \hbar \,.\end{aligned}$$ The three choices are related by simple canonical transformations, $$\begin{aligned}
\sU' = \sU - \half \sT' \,, \quad
\sU'' = \sU + \half \sT \,.
\label{uuu}\end{aligned}$$ Among these choices, $\sU$ (instead of $\sU', \sU''$) is identified as quantum counterpart of the classical ‘meridian‘ variables $U$ in $\CM^{\textrm{knot}}$. $\sV$ is identified as a quantum counterpart of ‘longitudinal’ variable $\sV$ in $\CM^{\textrm{knot}}$.
Now, we can give a more precise meaning to the expression in , . If we consider $V$ as a fixed parameter, the trace of $\varphi$ will be a function on the parameter, $\textrm{Tr}(\varphi)(V) $. On the other hand, the CS partition function $Z_{\textrm{tori}}$ should be understood as a wave-function in the Hilbert space $\mathcal{H}^{\textrm{knot}}$ associated to a choice of polarization $\Pi = (\CX, \CP)$, $$\begin{aligned}
Z_{\textrm{tori}(\varphi)} (x) =\;_{\Pi}\langle \CX=x | Z_{\textrm{tori}(\varphi)}\rangle\;, \quad | Z_{\textrm{tori}(\varphi)}\rangle \in \CH^{\textrm{knot}}\;.
\end{aligned}$$ Here $_\Pi \langle \CX =x|$ denotes a position eigenstate in $\Pi$ polarization. A more precise statement of is that the function $\textrm{Tr}(\varphi) (V)$ is a wave-function in the polarization $\Pi = (\sV, \sU )$: $$\begin{aligned}
\textrm{Tr}(\varphi) (V) =\; _{\Pi}\langle \CX = V | Z_{\textrm{tori}(\varphi)}\rangle\;,\end{aligned}$$ Using the quantum operator $\sV$, the above can be written as (in the choice $\Pi = (\sV, \sU )$) $$\begin{aligned}
\textrm{Tr}(\varphi) (V) &= \; _{\Pi} \langle \CX = V | \Tr(\varphi)(\sV) | \CP =0 \rangle_\Pi {\nonumber}\\
& = \; _{\Pi} \langle \CX = V | \Tr(\varphi)(\sV) | \sU =0 \rangle\;.\end{aligned}$$ Thus, we find the following polarization-independent expression, $$\begin{aligned}
|Z_{\textrm{tori}(\varphi)}\rangle =\Tr(\varphi)| \sU =0 \rangle \in \CH^{\rm knot}\;.\end{aligned}$$ Similarly, for mapping cylinder, the precise meaning of is $$\begin{aligned}
|Z_{\Sigma_{1,1}\times_\varphi I} \rangle = \varphi |\sU =0\rangle \in \CH (\Sigma_{1,1}) \otimes \overline{\CH(\Sigma_{1,1})} \otimes \CH^{\rm knot}\;.\end{aligned}$$ Recall that the boundary phase space of mapping cylinder is locally $ \CM(\Sigma_{1,1})^2 \times \CM^{\rm knot} $ and quantization of the phase space gives a Hilbert-space $\CH (\Sigma_{1,1}) \otimes
\overline{\CH(\Sigma_{1,1})} \otimes \CH^{\rm knot}$. $\overline{\CH}$ denote a dual Hilbert-space and the structure $\CH
\otimes \overline{\CH}$ is due to the two oppositely oriented boundary Riemann surfaces.
Since $(\sqrt{t},\sqrt{t'},\sqrt{t''}, \ell)$ and $(\sqrt{\bar{t}},\sqrt{\bar{t}'},\sqrt{\bar{t}''}, \bar{\ell})$ are related by complex conjugation, it is natural to define the adjoint of their quantum counterparts as follows $$\begin{aligned}
(\sqrt{\st}, \sqrt{\st'} ,\sqrt{\st''}, \fl := e^\sV)_\pm^\dagger = (\sqrt{\st}, \sqrt{\st'} ,\sqrt{\st''}, \fl )_\mp\;. \label{adjoin of shear and ell}\end{aligned}$$
#### SL$(2,\IZ)$ action
Under the action of the generators of SL$(2,\IZ)$ in , the transformation rule for the quantum shear coordinates can be summarized as follows.
$$\begin{array}{c|ccc}
\;\;\varphi\;\; & \sqrt{\st} \mapsto & \sqrt{\st'} \mapsto & \;\; \sqrt{\st''} \mapsto \;\;
\\[5pt]
\hline
\\[-10pt]
\;\;S\;\; & \;\; \displaystyle{\frac{1}{\sqrt{\st}}} \;\; & \;\; \sqrt{\st''}\displaystyle{\frac{1}{1+q^{\half} \st^{-1}}} \;\; & \sqrt{\st'}(1+q^{\half}\st)
\\[12pt]
\;\;L\;\; & \;\; \displaystyle{\frac{1}{\sqrt{\st''}}} \;\; & \;\; \sqrt{\st'}\displaystyle{\frac{1}{1+q^{\half} \st''^{-1}}} \;\; & \sqrt{\st}(1+q^{\half}\st'')
\\[12pt]
\;\;R\;\; & \;\; \displaystyle{\frac{1}{\sqrt{\st'}}} \;\; & \;\; \sqrt{\st}\displaystyle{\frac{1}{1+q^{\half} \st'^{-1}}} \;\; & \sqrt{\st''}(1+q^{\half}\st')
\end{array}
\label{q-shear-transf}$$
After quantization, the $SL(2,\mathbb{Z})$ transformation $\varphi$ becomes an operator acting on Hilbert-space. Operator $\varphi$ can be expressed in terms of $(\sT,\sT',\sT'')_\pm$. For $\varphi=\sL$ and $\sR$, the operator is determined by the following conditions $$\begin{aligned}
& \sL \cdot\sqrt{\st_\pm} =\frac{1}{\sqrt{\st''_{\pm}}} \cdot \sL \;, \quad \sL\cdot \sqrt{\st''_\pm} - \sqrt{\st_\pm}(1+q^{\pm 1/2}\st^{\prime\prime}_\pm) \cdot \sL=0 \;, \nonumber
\\
& \sR \cdot\sqrt{\st_\pm} =\frac{1}{\sqrt{\st'_{\pm}}} \cdot \sR \;, \quad \sR\cdot \sqrt{\st''_\pm} - \sqrt{\st''_\pm}(1+q^{\pm 1/2}\st'_\pm) \cdot \sR=0 \;.
\label{conditions for L}\end{aligned}$$ The solution for the operator equation can be given as follows $$\begin{aligned}
&\sL =\sL_{+} \sL_{-} = \left( \prod_{r=1}^{ \infty} \frac{ 1+ q^{ r- \half} ( \st''_{+})^{-1} }{ 1+ q^{ r- \half} (\st''_{-})^{-1} } \right)
\exp \left[ - \frac{1}{ 4 \hbar} \left( (\sT''_{+} + \sT_{+})^2 - (\sT''_{-} + \sT_{-} )^2 \right) \right] \;, {\nonumber}\\
& \sR = \sR_{+} \sR_{-} = \left( \prod_{r=1}^{ \infty} \frac{ 1+ q^{r- \half} \st'_{-} }{ 1+ q^{ r- \half} \st'_{+} } \right) \exp \left[ \frac{1}{ 4 \hbar} \left( (\sT'_{+} + \sT_{+})^2 - (\sT'_{-} + \sT_{-} )^2 \right) \right] \;. \label{L,R operators}
\end{aligned}$$ From the solution, we find that ($\sm := e^{\sU} $) $$\begin{aligned}
& \sm_\pm \cdot \sL \cdot \sm_\pm^{-1} =\frac{1}{\sqrt{\st_\pm}}\cdot \sL \cdot \sqrt{\st_\pm} {\nonumber}\\
&\sm_\pm \cdot \sR \cdot \sm_\pm^{-1} = \sR \;. \label{[m, L] and [m,R]}\end{aligned}$$ We use that $\sm_\pm \cdot \st''_\pm \cdot \sm_\pm^{-1} = q^{\pm 1} \st''_\pm$. This give a derivation of . Note that these operators are all unitary; see . Since all $SL(2,\mathbb{Z})$ elements can be constructed by multiplying $\sL,\sR$ and their inverses, we can easily see that all $\varphi \in SL(2,\mathbb{Z})$ are unitary operators. As we will see in section \[sec: SCI/SL(2,C)\], this unitarity is closely related to $SL(2,\mathbb{Z})$ duality invariance of the supeconformal index for 4d $\mathcal{N}=2^*$ theory.
#### Shear vs Fenchel-Nielson
Quantization of the FN coordinates can be summarized as $$\begin{aligned}
\hat{\tau} = e^{\hat{\mathcal{T}}}\,,
\;\;
\hat{\lambda} = e^{\hat{\Lambda}}
\,, \quad
[ \hat{\mathcal{T}} , \hat{\Lambda} ] = \hbar
\,, \quad
\hat{\tau} \hat{\lambda} = q \hat{\lambda} \hat{\tau} \,.
\label{FN-op}\end{aligned}$$ The relation to quantum shear coordinates was given in [@Dimofte:2011jd]. $$\begin{aligned}
&\sqrt{\st} = \frac{i}{\hat{\l} -\hat{\l}^{-1}}(\hat{\tau}^{-\half} -\hat{\tau}^{\half}) \,,
{\nonumber}\\
&\sqrt{\st'} = \frac{i}{q^{-\frac{1}{4}}\hat{\l}^{-1}\hat{\t}^{\half} - q^{\frac{1}{4}}\hat{\t}^{-\half} \hat{\l}}(\hat{\l} -\hat{\l}^{-1}) \,,
{\nonumber}\\
&\sqrt{\st''} = \frac{i q^{\frac{1}4}\fl^{\half}}{\hat{\t}^{-\half} - \hat{\t}^{\half}}
(q^{-\frac{1}{4}}\hat{\l}^{-1}\hat{\t}^{\half} - q^{\frac{1}{4}}\hat{\t}^{-\half} \hat{\l}) \,.
\label{shear2FN-q}
\end{aligned}$$
#### Loop vs Fenchel-Nielson
Quantizing the loop coordinates $(W, H, D)$ yields [@Dimofte:2011jd] $$\begin{aligned}
&\sW = \hat{\lambda} +\hat{\lambda}^{-1} \;, {\nonumber}\\
&\sH= \frac{q^{-\frac{1}{4}} \fl^{\half} \hat{\lambda} - q^{ \frac{1}{4}} \fl^{-\half} \hat{\lambda}^{-1} }{\hat{\lambda}-\hat{\lambda}^{-1}} \hat{\tau}^{-\half}
+ \frac{q^{\frac{1}{4}} \fl^{-\half}\hat{\lambda}- q^{-\frac{1}{4}} \fl^{\half} \hat{\lambda}^{-1}}{\hat{\lambda} - \hat{\lambda}^{-1}} \hat{\tau}^{\half}\;. {\nonumber}\\
&\sD= \frac{q^{-\frac{1}{4}} \fl^{\half} \hat{\lambda} - q^{\frac{1}{4}} \fl^{-\half} \hat{\lambda}^{-1} }{\hat{\lambda}-\hat{\lambda}^{-1}} q^{-\frac{1}4} \hat{\lambda}\hat{\tau}^{-\half}
+ \frac{q^{\frac{1}{4}} \fl^{-\half}\hat{\lambda}- q^{-\frac{1}{4}} \fl^{\half} \hat{\lambda}^{-1}}{\hat{\lambda} - \hat{\lambda}^{-1}}q^{-\frac{1}4} \hat{\lambda}^{-1} \hat{\tau}^{\half}\;. \label{loop2FN-q}\end{aligned}$$
Superconformal index/$SL(2,\mathbb{C})$ CS partition function {#sec: SCI/SL(2,C)}
=============================================================
The superconformal index for 3d SCFTs with global symmetry $U(1)^N$ is defined as [@Kim:2009wb; @Imamura:2011su; @Krattenthaler:2011da; @Kapustin:2011jm] $$\begin{aligned}
I(q, m_i , u_i ) = \textrm{Tr} (-1)^F q^{\half R+j_3} \prod_{i=1}^N u_i^{H_i}\end{aligned}$$ where the trace is taken over Hilbert-space $\mathcal{H}_{\{m_i\}}$ on $S^2$, where background monopole fluxes $\{m_i\}$ coupled to global symmetries $U(1)^N$ are turned on. $R$ and $j_3$ denote $U(1)$ R-charge and spin on $S^2$ respectively. $\{u_i\}$ are fugacity variables for the $U(1)^N$ whose generators are denoted by $\{ H_i\}$. It is often useful to express the index in a charge basis $(m_i, e_i)$ instead of $(m_i, u_i)$, $$\begin{aligned}
I(m_i , u_i ) = \sum_{e_i} I(m_i, e_i) u_i^{e_i} \;.\end{aligned}$$ In the charge basis, the $Sp(2N,\mathbb{Z})$ transformation [@Witten:2003ya] on 3d SCFTs with $U(1)^N$ global symmetry acts linearly. For two 3d SCFTs, $\mathcal{T}$ and $g\cdot \mathcal{T}$, related by $g\in Sp(2N,\mathbb{Z})$, the generalized indices for the two theories are related as [@Dimofte:2011py] $$\begin{aligned}
I_{g\cdot \mathcal{T}} (m,e) = I_{\mathcal{T}} (g^{-1}\cdot (m,e))\;.\end{aligned}$$ In section \[section : two routes\], we gave two alternative descriptions for mapping torus theories which we denote by $T^{\textrm{T[SU(2)]}}_{\textrm{tori}(\varphi)}$ and $T^{\Delta}_{\textrm{tori}(\varphi)}$. The two descriptions give seemingly different expressions for the index. We will denote the index for $T^{\textrm{T[SU(2)]}}_{\textrm{tori}(\varphi)}$ and $T^{\Delta}_{\textrm{tori}(\varphi)}$ by $I^{\textrm{T[SU(2)]}}_{\textrm{tori}(
\varphi)}$ and $I^{\Delta}_{\textrm{tori}(\varphi)}$, respectively. By proving $$\begin{aligned}
I^{\textrm{T[SU(2)]}}_{\textrm{tori}(
\varphi)} = I^{\Delta}_{\textrm{tori}(\varphi)} \label{equality for two mapping tori index}\end{aligned}$$ for general $\varphi$ with $|\Tr (\varphi)|>2$, we will confirm the equivalence of two descriptions at the quantum level. The 3d-3d correspondence [@Dimofte:2011py] predicts that the index is the same as the $SL(2,\mathbb{C})$ CS partition function on the mapping torus, $Z_{\textrm{tori}(\varphi)}(SL(2,\mathbb{C}))$. $$\begin{aligned}
I_{\textrm{tori}(\varphi)} = Z_{\textrm{tori}(\varphi)}(SL(2,\mathbb{C}))\;.\end{aligned}$$ There are also two independent ways of calculating $
Z_{\textrm{tori}(\varphi)}(SL(2,\mathbb{C}))$ depending on the way of viewing the 3-manifold tori($\varphi$). Viewing tori($\varphi$) as a 3-manifold obtained by gluing tetrahedra, the CS partition can be calculated using a state integral model developed in [@Dimofte:2011gm]. Let’s denote the CS partition function obtained in this way by $Z^{\Delta}_{\textrm{tori}(\varphi )}
(SL(2,\mathbb{C}))$. It was shown in [@Dimofte:2011py] that the $SL(2,\mathbb{C})$ CS partition function on $M$ obtained from the state integral model is always the same as superconformal index for $T_M$ theory obtained from gluing tetrahedron theories, $\CT_{\Delta}$’s. Thus, it is already proven that $$\begin{aligned}
Z^{\Delta}_{\textrm{tori}(\varphi)} (SL(2,\mathbb{C})) = I^{\Delta}_{\textrm{tori}(\varphi)}\;. \label{ZDelta=IDelta}\end{aligned}$$ Another way of calculating the $SL(2,\mathbb{C})$ CS partition function is using the canonical quantization of the CS theory on $\Sigma_{1,1}$ viewing the $S^1$ direction in tori($\varphi$) as a time direction. The partition function obtained in this approach will be denoted as $Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)} (SL(2,\mathbb{C}))$. As mentioned in section \[quantum riemann surface\], $$\begin{aligned}
Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)} (SL(2,\mathbb{C})) = \textrm{Tr} (\varphi) \; \textrm{on $\CH_{SL(2,\mathbb{C})}$} \;. \label{CS partition from trace}\end{aligned}$$ We will show that two approaches are equivalent $$\begin{aligned}
Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)} (SL(2,\mathbb{C})) = Z^{\Delta}_{\textrm{tori}(\varphi)} (SL(2,\mathbb{C}))\;,\end{aligned}$$ by expressing the trace in using a basis of $\CH _{SL(2,\mathbb{C})}$ called ‘SR basis’. On the other hand, by expressing the trace in a basis called ‘FN basis’, we will show that $$\begin{aligned}
Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)} (SL(2,\mathbb{C})) = I^{\textrm{T[SU(2)]}}_{\textrm{tori}(
\varphi)} \;.\end{aligned}$$ Since the trace is independent of basis choice, the proof of now follows from the known proof of . Further, by showing that the matrix element of $\varphi$ in the FN basis is the same as the superconformal index for duality wall theory $T[SU(2),\varphi]$ we also confirm the 3d-3d dictionary for mapping cylinder.
Duality wall theory : $I^{T[SU(2)]}_{{\rm tori}(\varphi)}$ {#index-wall}
----------------------------------------------------------
In this section, we will calculate the superconformal indices for duality wall theories $T[SU(2),\varphi]$ and mapping torus theories $T^{T[SU(2)]}_{\textrm{tori}(\varphi)} =\Tr (T[SU(2),\varphi]) $. First, consider the case $\varphi= S$. The $T[SU(2)]\equiv T[SU(2),S]$ theory is explained in detail in section \[duality wall theory\] and summarized in table \[tsu(2)-charge\]. The generalized superconformal index for the theory can be obtained by the using general prescriptions in [@Kim:2009wb; @Imamura:2011su; @Kapustin:2011jm],[^13] $$\begin{aligned}
&I_{\varphi=S}(m_b,u_b,m_t, u_t;m_\eta,u_\eta)
=\sum_{m_s\in \mathbb{Z}+m_b+\half m_\eta}\oint \frac{du_s}{2\pi i u_s} I^{(0)} I^{(1)} \;. \label{index formula-1}\end{aligned}$$ Our notations for the fugacity and flux variables appearing in the index are summarized in Table \[tsu(2)-2\].
$q_1\;\;\; q_2 \;\;\;q_3 \;\;\;q_4\;\;\;\phi_0$ fugacity flux
-------------------- ---------------------------------------------------------- ---------- ---------- --
$U(1)_{\rm gauge}$ $1\;\;\;\; 1 \;-1\; -1\;\;\;0$ $u_s$ $m_s$
$U(1)_{\rm bot}$ $1\;-1 \;\;\;1\;-1\;\;\;0$ $u_b$ $m_b$
$U(1)_{\rm punct}$ $\;\half \;\;\;\;\half \;\;\;\;\half \;\;\;\;\half \;-1$ $u_\eta$ $m_\eta$
$U(1)_{\rm top}$ $ 0\;\;\;\;0\;\;\;\;0\;\;\;\;0\;\;\;\;0$ $ u_t$ $m_t$
: Fugacity, background monopole flux variables for symmetries in $T[SU(2)]$ theory []{data-label="tsu(2)-2"}
$I^{(1)}$ is the Plethystic exponential (PE) of the single letter indices from chiral-multiplets $$\begin{aligned}
I^{(1)} &=\textrm{PE}[\sum_{\epsilon_1,\epsilon_2 =\pm 1}\frac{ ( u_b^{\epsilon_1} u_\eta^\half u_s^{\epsilon_2} q^{1/4} -u_b^{-\epsilon_1} u_\eta^{-\half} u_s^{-\epsilon_2} q^{3/4})q^{\frac{1}{2}|\epsilon_1 m_b+\half m_\eta+ \epsilon_2 m_s|} }{1-q}+\frac{q^{\frac{1}{2}+\half |m_\eta|}}{1-q}(u_\eta^{-1}-u_\eta)]\;, \nonumber
\\
&=
\prod_{ l =0}^{ \infty}
\left(
\prod_{ \epsilon_1, \epsilon_2 = \pm 1}
\frac{ ( 1- ( u_b^{ \epsilon_1} u_\eta^\half u_s^{ \epsilon_2})^{-1} q^{ \frac{3}{4} + \frac{1}{2} | \epsilon_1 m_b + \half m_{ \eta} + \epsilon_2 m_s|+l} )}
{(1-( u_b^{ \epsilon_1} u_\eta^\half u_s^{ \epsilon_2}) q^{ \frac{1}{4} + \frac{1}{2} | \epsilon_1 m_b +\half m_{ \eta} + \epsilon_2 m_s|+l} )}
\right)
\frac{
( 1- u_\eta q^{ \frac{1}{2} + \half |m_{ \eta}| + l })
}{(1- u_\eta^{-1} q^{ \frac{1}{2} + \half |m_{ \eta}| + l } )}
\, ,
\nonumber\end{aligned}$$ where $$\begin{aligned}
\textrm{PE}[f(q, u_s, u_\eta,u_b)] = \exp \left[ \sum_{n=1}^\infty \frac{1}n f(q^n, u_s^n, u_\eta^n, u_b^n) \right] \,.\end{aligned}$$ The above index can be rewritten as in which is free from absolute values of magnetic fluxes. We assign conformal dimension $\Delta$ for chirals as follows $$\begin{aligned}
\Delta (q_i)= \frac{1}2\,, \quad \Delta(\phi_0) =1\,,\end{aligned}$$ which is canonical for 3d $\mathcal{N}=4$ SCFTs.[^14] $I^{(0)}$ collects all contributions from classical action and zero-point shifts $$I^{(0)} = u_t^{2m_s} u_s^{2m_t} q^{\epsilon_0} u_\eta^{F_{\eta,0}} u_b^{F_{b,0}} u_s^{F_{s,0}}(-1)^{{\rm sgn}}\;.\nonumber$$ The $u_t^{2s} u_s^{2m_t}$ term originates from the BF-term which couples background gauge field for $U(1)_{\rm top}$ to the field strength of $U(1)_{\rm gauge}$. The zero-point contributions, $ \epsilon_0$, $F_{\eta,0}, F_{b,0}, F_{ s,0}$ are given by [@Imamura:2011su] $$\begin{aligned}
&\epsilon_0 :=\frac{1}8 \sum_{\epsilon_1, \epsilon_2 = \pm} (|\epsilon_1 m_b+\half m_\eta+\epsilon_2 m_s|)\;,\nonumber
\\
&F_{\eta,0} := \half |m_\eta|-\frac{1}4 \sum_{\epsilon_1, \epsilon_2 = \pm} (|\epsilon_1 m_b+\half m_\eta+\epsilon_2 m_s|)\;,\nonumber
\\
&F_{b,0} :=-\frac{1}2 \sum_{\epsilon_1, \epsilon_2 = \pm} \epsilon_1 (|\epsilon_1 m_b+\half m_\eta+\epsilon_2 m_s|)\;,\nonumber
\\
&F_{s,0} :=-\frac{1}2 \sum_{\epsilon_1, \epsilon_2 = \pm} \epsilon_2 (|\epsilon_1 m_b+\half m_\eta+\epsilon_2 m_s|) \;.\nonumber\end{aligned}$$ The subtle sign factor $$\begin{aligned}
{\rm sgn} := 2m_b+\half (m_\eta+ |m_\eta|)+\sum_{\epsilon_1, \epsilon_2 = \pm}\frac{1}2 (|\epsilon_1 m_b+\half m_\eta+\epsilon_2 m_s|)\;,\end{aligned}$$ is chosen for the index to satisfy the so-called self-mirror property [@Tong:2000ky; @Hosomichi:2010vh; @Gang:2012ff].$$\begin{aligned}
I_{\varphi=S}(m_b,u_b,m_t,u_t;m_\eta,u_\eta) =I_{\varphi=S}(m_t,u_t,m_b,u_b;-m_\eta,u_\eta^{-1})\;.\end{aligned}$$ This sign factor (or more generally phase factor) always appears in the computation of 3d generalized index and lens space partition function [@Imamura:2012rq]. To the best our knowledge, a systematic method for fixing the subtlety has not been developed yet, though it has survived numerous tests.
In our normalization, the background monopole charges $(m_b, m_t)$ are half-integers and $(m_\eta)$ is an integer. $$\begin{aligned}
&2m_b, 2m_t, m_\eta \in \mathbb{Z} \;. \nonumber\end{aligned}$$ Note that the summation range, $ m_s \in {\mathbb Z} + m_b +\half m_{ \eta}$, is to satisfy the following Dirac quantization conditions, $$\begin{aligned}
& \pm m_b + \half m_{ \eta} \pm m_s \ \in {\mathbb Z}\; .\end{aligned}$$
Multiplying $\varphi \in SL(2, \mathbb{Z})$ by $T^{k}$ amounts to turning on a Chern-Simons term with level $k$ for the background gauge field of $U(1)_{\rm bot}$ or $U(1)_{\rm top}$. It affects the index as follows $$\begin{aligned}
I_{T^{k} \cdot \varphi} =( u_b)^{2km_b} I_{\varphi}\;, \quad I_{\varphi \cdot T^{k}} = (u_t)^{2k m_t} I_{\varphi}\;. \label{index formula-2}\end{aligned}$$ Here the phase factors $(u_b)^{2km_b}$ and $(u_t)^{2k m_t}$ come from the classical action for the added CS term. The theory $T[SU(2),\varphi_2\cdot \varphi_1]$ is obtained by gauging the diagonal subgroup of $SU(2)_{\textrm{top}}$ from $T[SU(2),\varphi_2]$ and $SU(2)_{\textrm{bot}}$ from $T[SU(2),\varphi_1]$. Accordingly, the index is glued by $\odot$ operation defined below under the $SL(2,\mathbb{Z})$ multiplication $$\begin{aligned}
&I_{\varphi_2 \cdot \varphi_1}(m_b,u_b,m_t,u_t;m_\eta,u_\eta) =(I_{\varphi_2} \odot I_{\varphi_1})(m_b,u_b,m_t,u_t;m_\eta,u_\eta) {\nonumber}\\
& := \sum_{n\in \frac{1}2 \mathbb{Z}} \oint [dv]_n I_{\varphi_2}(m_b,u_b,n,v;m_\eta,u_\eta) I_{\varphi_1}(n,v,m_t,u_t;m_\eta,u_\eta)\;. \label{odot gluing rule in T[SU(2)] theories}\end{aligned}$$ The integration measure $[dv]_m$ comes from the index for a $\mathcal{N}=2$ vector multiplet for the diagonal subgroup of the two $SU(2)$’s, $$\begin{aligned}
&\oint [dv]_n := \oint \frac{dv}{2\pi i v} \Delta(n,v)\;, \nonumber
\\
&\Delta(n,v):= \frac{1}2 (q^{\frac{n}2} v - q^{-\frac{n}2} v^{-1})(q^{\frac{n}2} v^{-1} - q^{-\frac{n}2} v) \;. \label{su(2) measure}\end{aligned}$$ In terms of the charge basis for $U(1)_{\rm punct}$, the operation $\odot$ is given by $$\begin{aligned}
&(I_{\varphi_2} \odot I_{\varphi_1})(m_b,u_b,m_t,u_t;m_\eta,e_\eta) {\nonumber}\\
& = \sum_{e_{\eta,1}e_{\eta,2}} \sum_{n} \oint [dv]_n \delta(e_\eta -\sum_{k=1}^2 e_{\eta,k})I_{\varphi_2}(m_b,u_b,n,v;m_\eta,e_{\eta,1}) I_{\varphi_1}(n,v,m_t,u_t;m_\eta,e_{\eta,2})\;.
\label{index formula-3}\end{aligned}$$ Using the $\odot$ operation, one can write $$\begin{aligned}
&I_{T^{k_1} \cdot \varphi \cdot T^{k_2}} =\overbrace{ I_{T} \odot \ldots I_{T}}^{k_1}\odot I_{\varphi} \odot \overbrace{ I_{T} \odot \ldots I_{T} }^{k_2}\;, \; \textrm{where} {\nonumber}\\
&I_{\varphi=T}(m_b, u_b, m_t, u_t ;m_\eta,u_\eta) = u_b^{2m_b}\frac{ \delta_{m_b, m_t}\delta(u_b-u_t)}{\Delta(m_b, u_b)} \;. \label{mapping cylinder index for T}\end{aligned}$$ Using prescriptions in eq. , and , one can calculate the index $I_{\varphi}$ for general $T[SU(2),\varphi]$. The $SL(2,\mathbb{Z})$ structure is encoded in the index. We find that $$\begin{aligned}
I_{S^2 \cdot \varphi} = (-1)^{ m_\eta} I_{\varphi} \;, \quad I_{(ST)^3 \cdot \varphi} = u_\eta^{\half m_\eta} I_{\varphi} \;, \label{SL(2,Z) in index}\end{aligned}$$ by calculating the index in $q$-series expansions. In appendix \[T\[SU(2)\] classical\], these $SL(2,\mathbb{Z})$ structure will be analyzed by studying classical difference equations for $I_{\varphi}$. The factor $u_\eta^{\half m_\eta}$ can be interpreted as a CS term for background gauge field coupled to $U(1)_{\rm punct}$.
Finally, the index $I^{T[SU(2)]}_{\textrm{tori}(\varphi)} (m_\eta, u_\eta)$ for the mapping torus theory $\Tr(T[SU(2),\varphi])$ is given by $$\begin{aligned}
I^{\textrm{T[SU(2)]}}_{\textrm{tori}(\varphi)} (m_\eta, u_\eta) =\sum_{n\in \frac{1}2 \mathbb{Z}} \oint [du]_n I_{\varphi}(m,u,m,u;m_\eta,u_\eta)\;. \label{mapping torus index from duality wall}\end{aligned}$$ For $\varphi = S$ the mapping torus index becomes extremely simple (checked in $q$ expansion) $$\begin{aligned}
I^{\textrm{T[SU(2)]}}_{\textrm{tori}(S)} (m_\eta, u_\eta) = \left\{\begin{array}{ll} (-1)^{\half m_\eta}\;, & \quad m_\eta\in 2\mathbb{Z} \\0 \;, & \quad m_\eta \in 2\mathbb{Z}+1\end{array}\right.\end{aligned}$$ This may imply that the corresponding $T_{\textrm{tori}(S)}$ theory is a topological theory. See [@Ganor:2010md; @Ganor:2012mu] for related discussion. The mapping torus index is also simple for $\varphi = R^{-1}L= -T^{-1}ST^{-1}S$, $$\begin{aligned}
I^{\textrm{T[SU(2)]}}_{\textrm{tori}(R^{-1}L)} (m_\eta, e_\eta) = (-1)^{e_\eta}\delta_{m_\eta ,-3 e_\eta}\;.\end{aligned}$$ Actually the mapping torus with $\varphi = R^{-1}L$ is trefoil knot complement in $S^3$ [@Garb:2013] and the above index is identical to the corresponding index in [@Dimofte:2011py] computed by gluing two tetrahedron indices, up to a polarization difference $(\CX,\CP)_{\rm here} = (-\CP + \pi i , \CX)_{\rm there}$. Refer to section \[index-tetra\] for how polarization change affects the index. For $\varphi= LR = ST^{-1}S^{-1}T$, the mapping torus index is $$\begin{aligned}
&I^{\textrm{T[SU(2)]}}_{\textrm{tori}(LR)}( m_\eta = 0,u_\eta)= 1-2q + 2 (u_\eta +\frac{1}{u_\eta}) q^{3/2} - 3q^2 + (2 +u_\eta^2+\frac{1}{u_\eta^2})q^3
+\cdots\;,
{\nonumber}\\
&I^{\textrm{T[SU(2)]}}_{\textrm{tori}(LR)}( m_\eta =\pm 1,u_\eta)= -(2-u_\eta - \frac{1}{u_\eta}) q -(2- u_\eta - \frac{1}{u_\eta}) q^2
+\cdots\;, {\nonumber}\\
&I^{\textrm{T[SU(2)]}}_{\textrm{tori}(LR)}(m_\eta =\pm 2,u_\eta)= (u_\eta + \frac{1}{u_\eta}) q^{1/2} -q -q^2 -(u_\eta + \frac{1}{u_\eta}) q^{5/2} \cdots\;, {\nonumber}\\
&I^{\textrm{T[SU(2)]}}_{\textrm{tori}(LR)}(m_\eta =\pm 3,u_\eta)= -(u_\eta +\frac{1}{u_\eta}-u_\eta^2 - \frac{1}{u_\eta^2})q^2 + \cdots\;, {\nonumber}\\
&I^{\textrm{T[SU(2)]}}_{\textrm{tori}(LR)}(m_\eta =\pm 4,u_\eta)= (u_\eta^2 +\frac{1}{u_\eta^2})q-(u_\eta +\frac{1}{u_\eta})q^{5/2} + q^3 + \cdots\;.
\label{LR index}\end{aligned}$$ Note that only integer powers of $u_\eta$ appear in the mapping torus indices. This is true for any mapping torus index with $\varphi$ being products of $L$ and $R$.[^15] On the other hand, for mapping cylinder indices, half-integer powers of $u_\eta$ may appear. For example, the index for $T[SU(2),\varphi = LR]$ is $$\begin{aligned}
\frac{u_b+ u_b u_t^2 }{u_t u^\half_\eta} q^{\frac{1}4} + \frac{u_b (1+u_t^2)(1+u_t^4)}{u_t^3 u_\eta^{\frac{3}2}} q^{\frac{3}4}+\ldots\end{aligned}$$ when $(m_b , m_t , m_\eta)= (\half, 0,0)$. The disappearance of $u_\eta^{\mathbb{Z}+\half}$ in mapping torus index is closely related to the fact that periodicity of longitudinal variable $V$ become half after making mapping torus from mapping cylinder, as mentioned in the last paragraph in section \[sec: Classical phase space\]. We will come back to this point in section \[index : Hilbert-space\] during the construction of $\CH^{\textrm{knot}}$.
Tetrahedron decomposition : $I^{\Delta}_{{\rm tori}(\varphi)}$ {#index-tetra}
--------------------------------------------------------------
In this section, we will explain how to calculate the superconformal index $I^{\Delta}_{{\rm tori}(\varphi)}$ for the theory $T^\Delta_{\textrm{tori}(\varphi)}$. In section \[section : two routes\], we briefly reviewed the construction of $T_M$ from the tetrahedron decomposition data of $M$. Using this construction and well-developed algorithms [@Kim:2009wb; @Imamura:2011su; @Kapustin:2011jm] for calculating the superconformal indices for general 3d theories, we can calculate the superconformal indices for $T_M$. The procedure of calculating indices from tetrahedron gluing is well explained in [@Dimofte:2011py] and the procedure is shown to be equivalent to the procedure of calculating $SL(2,\mathbb{C})$ CS partition function using the state integral model developed [@Dimofte:2011gm]. First, we will review the procedure of calculating the indices for general $T_M$ from the tetrahedron gluing data for $M$. Then, we will apply the general procedure to $M={\rm tori}(\varphi)$ with $|\Tr (\varphi)|>2$.
Suppose that $M$ can be decomposed into $N$ tetrahedra $\{ \Delta_i \}$ $(i=1,\ldots, N)$ with proper gluing conditions $\sim$, $M=(\bigcup_i \Delta_i)/\sim$. For each tetrahedron $\Delta_i$ we assign a “wave-function" (index) $\CI^{\Pi_i}_{\Delta}(m_i,e_i)$ which depends on the choice of polarization $\Pi_i =(\mathcal{X}_i , \mathcal{P}_i)$ of the tetrahedron’s boundary phase-space $\CM(\partial \Delta_i)$. Recall that the phase space $\CM(\partial \Delta)$ is a 2 dimensional space represented by three edge parameters $Z,Z',Z''$ with the constraint. $$\begin{aligned}
Z+Z'+Z'' = \pi i +\frac{\hbar}2 \;.\end{aligned}$$ The symplectic form on the phase space is $$\begin{aligned}
\frac{1}{i\hbar} dZ\wedge dZ' - \frac{1}{i\hbar} d\bar{Z}\wedge d\bar{Z}'= (\textrm{cyclic permutation in $Z, Z', Z''$})\;.\end{aligned}$$ For the choice of polarization $\Pi=\Pi_Z:=(Z,Z'')$, the index is given as [@Dimofte:2011py] (see also [@Garb:2012]) $$\begin{aligned}
\CI^{\Pi_Z}_{\Delta}(m, \zeta)= \sum_{e\in \mathbb{Z}} \CI^{\Pi_Z}_{\Delta}(m,e) \zeta^e =\prod_{r=0}^\infty \frac{1-q^{r-\frac{m}2+1} \zeta^{-1}}{1-q^{r-\frac{m}2}\zeta } \,.
\label{tetrahedron index}
\end{aligned}$$ The index can be understood as an element of $\mathbb{Z}[\zeta, \zeta^{-1}]((q^{\half}))$ by expanding the index in $q$. An element of $\mathbb{Z}[\zeta, \zeta^{-1}]((q^{\half}))$ contains only finitely many negative powers of $q$ and each coefficient is written as Laurent series in $\zeta$ . In the infinite product, there is an ambiguity when $r=m$ where a factor $\frac{1}{1-\zeta}$ appear. We formally interpret the factor as $$\begin{aligned}
\frac{1}{1-\zeta}= \sum_n \zeta^n \in \mathbb{Z}[\zeta, \zeta^{-1}]\;.\end{aligned}$$ The domain of $(m,e)$ in the tetrahedron index $\CI^{\Pi_Z}_\Delta(m,e)$ is $$\begin{aligned}
m,e\in \mathbb{Z}\;.\end{aligned}$$ For later use, we will extend the range of the function $\CI_{\Delta}$ to $\{ (m,e) \in \mathbb{Q} + i \pi \mathbb{Q}+ \frac{\hbar}2 \mathbb{Q}\} $. For $(m,e) \in \mathbb{Q}$, we define the function as $$\begin{aligned}
\CI_{\Delta} (m,e) :=\left\{\begin{array}{ll} \CI^{\Pi_Z}_{\Delta}(m,e) \;, & \quad (m,e) \in \mathbb{Z}^2\\ 0 \;, & \quad \textrm{$(m,e) \in \mathbb{Q}^2 - \IZ^2 $}\end{array}\right. \label{function IDelta}\end{aligned}$$ For general $(m+\alpha,e+\beta)$ with $(m,e) \in \mathbb{Q}$ and $(\alpha, \beta) \in i \pi \mathbb{Q}+ \frac{\hbar}2 \mathbb{Q}$, the function is determined by and the following additional relation $$\begin{aligned}
\CI_{\Delta} \big{(}m+\a, e+\b \big{)} :=e^{e \a - m \b} \CI_{\Delta}(m,e) \;.\label{affine shift in tetrahedron index}\end{aligned}$$ Under the polarization change from $\Pi=(\mathcal{X},\mathcal{P})$ to $\tilde{\Pi}=(\tilde{\mathcal{X}}, \tilde{\mathcal{P}})$, related by the following $SL(2,\mathbb{Q})$ and affine shifts[^16] $$\begin{aligned}
\left(\begin{array}{c} \tilde{\mathcal{X}} \\ \tilde{\mathcal{P}}\end{array}\right) = g\cdot \left(\begin{array}{c} \mathcal{X}\\ \mathcal{ P} \end{array}\right) + (\pi i + \frac{\hbar}2)\left(\begin{array}{c} \alpha_m \\ \alpha_e \end{array}\right) \;, \label{tetra polarization change}\end{aligned}$$ the tetrahedron index transforms as [@Dimofte:2011py] $$\begin{aligned}
&\CI^{\tilde{\Pi}}_{\Delta}(m,e) =(- q^{1/2})^{m \alpha_e - e\alpha_m} \CI^{\Pi}_{\Delta}(g^{-1} \cdot (m,e)) \;. \label{SL(2,Q)+affine transformation on tetra index}\end{aligned}$$ Under the $SL(2,\mathbb{Q})$ transformation, the domain of charge $(m,e)$ also should be transformed. The domain in the transformed polarization $\tilde{\Pi}$ is determined by demanding $g^{-1}\cdot (m,e)$ is in an allowed domain in the original polarization $\Pi$. The transformation rule can be written as $$\begin{aligned}
\CI^{\tilde{\Pi}}_{\Delta}(\tilde{m},\tilde{e}) = \CI^{\Pi}_\Delta (m,e)\;, \; \quad
\left(\begin{array}{c} \tilde{m} \\ \tilde{e}\end{array}\right) = g\cdot \left(\begin{array}{c} m\\ e \end{array}\right) + (\pi i + \frac{\hbar}2)\left(\begin{array}{c} \alpha_m \\ \alpha_e \end{array}\right) \;. \label{tetra polarization change-2}\end{aligned}$$ Shifts by a linear combination of $i \pi$ and $\frac{\hbar}2$ in the arguments of the function $\CI_{\Delta}$ is defined in eq. . Comparing with , we may identify $$\begin{aligned}
(m,e)\simeq (\CX, \CP)\;, \label{Polarization change equivalence}\end{aligned}$$ as far as transformation rules under polarization changes are concerned. Three choices of polarization, $\Pi_Z = (Z,Z''), \Pi_{Z'} = (Z' ,Z)$ and $\Pi_{Z''}=(Z'', Z')$ of a single tetrahedron are related to one another by discrete symmetries of tetrahedron. Demanding the conditions $\CI^{\Pi_Z}_\Delta = \CI^{\Pi_{Z'}}_\Delta = \CI^{\Pi_{Z''}}_\Delta$, we obtain the following triality relations on $\CI_\Delta$: $$\begin{aligned}
\CI_{\Delta} (m,e) =(-q^{1/2})^{-e} \CI_{\Delta}(e,-e-m) = (-q^{1/2})^m \CI_{\Delta}(-e-m,m) \;. \label{Triality}\end{aligned}$$ Another useful identity for the tetrahedron index is $$\begin{aligned}
\CI_{\Delta}(m,e) = \CI_{\Delta}(-e,-m)\;. \label{identity for tetrahedron index}\end{aligned}$$ The above identities on $\CI_{\Delta}(m,e)$ are valid only when $(m,e)\in \mathbb{Z}^2$.
The gluing conditions for $M=(\bigcup_i \Delta_i)/\sim$ can be specified by expressing linearly independent internal edges $C_I$ ($I=1,\ldots, k \leq N-1$) in terms of linear combination (and shifts) of $\mathcal{X}_i, \mathcal{P}_i$ variables. $$\begin{aligned}
&C_I = \sum_{j=1}^N (c^x_{Ij} \CX_j+c^{p}_{Ij} \CP_j ) + a_I = 2\pi i +\hbar \;\; \textrm{with coefficients $\{ c^{x}_{Ij} ,c^p_{Ij}, a_I\}$}. \label{internal edge conditions}\end{aligned}$$ The boundary phase space $\CM(\partial M)$ is given by a symplectic reduction $$\begin{aligned}
\CM(\partial M) = \prod_{i=1}^N \CM(\partial \Delta_i)//\{ C_I = 2\pi i +\hbar \}.\end{aligned}$$ The dimension of $\CM(\partial M)$ is $2d =2(N-k)$ and we choose a polarization $\Pi_{\partial M}=(\mathcal{X}_\alpha,\mathcal{P}_\alpha)|_{\alpha=1}^{d}$ for the boundary phase space as $$\begin{aligned}
\CX_\alpha = \sum_{j=1}^N (X^x_{\alpha j }\CX_j + X^p_{\alpha j} \CP_j )+ a_\alpha\;, \quad \CP_\alpha = \sum_{j=1}^N (P^x_{\alpha j }\CX_j + P^p_{\alpha j} \CP_j) + b_\alpha\;, \label{external polarization conditions}\end{aligned}$$ with coefficients $\{ X^x_{\alpha j} , X^p_{\alpha j}, P^{x}_{\alpha j }, P^p_{\alpha j} , a_\alpha, b_\alpha \}$ which guarantee that $[\CX_\alpha, \CP_\beta] = - \hbar \delta_{\alpha \beta}$ and $[\CX_\alpha, C_I ] =[\CP_\alpha, C_I] = 0 $ for all $\alpha, \beta, I$.
Now let’s explain how to calculate the index $I^\Delta_M$ for $T_M$, or equivalently the $SL(2,\mathbb{C})$ CS partition function $Z^{\Delta}_M (SL(2,\mathbb{C}))$, from the tetrahedron gluing data $\{C_I , \CX_\alpha, \CP_\alpha\}$ for $M$ explained above. In the polarization $\Pi_{\partial M}= (\CX_\a , \CP_\a)$, the index for $T_M$ theory with $M=\bigcup_{i=1}^{N} \Delta_i /\sim$ is given as $$\begin{aligned}
I_M(m_\alpha, e_\alpha) =\sum_{(m_*, e_*)\in \mathbb{Z}} \delta^{2d} (\ldots)\delta^{k}(\ldots)\prod_{i=1}^{N} \CI^{\Pi_i}_{\Delta}(m_i, e_i)\end{aligned}$$ Here, the Knocker delta functions $\delta^{2d}(\ldots)$ and $\delta^{k}(\ldots)$ come from external polarization choice and internal edge gluing conditions , respectively. In view of the identification , these constraints can be translated into constraints on charge variables $(m_i, e_i)$. $$\begin{aligned}
&\delta^{2d}(\ldots) = \prod_{\alpha=1}^d\delta (m_\alpha - \sum_{j=1}^N (X^{x}_{\alpha j} m_j +X^{p}_{\alpha j} e_j) -a_\alpha ) \delta (e_\alpha - \sum_{j=1}^N( P^{x}_{\alpha j} m_j +P^{p}_{\alpha j} e_j )-b_\alpha ) \;, {\nonumber}\\
&\delta^{k}(\ldots) = \prod_{I=1}^k \delta \big{(}\sum_{j=1}^N (c^{x}_{Ij} m_j +c^{p}_{Ij}e_j)+a_I -2\pi i - \hbar \big{)} \;.\end{aligned}$$ Solving the $2d+k$ Knonecker deltas on $2N$ variables $(m_i, e_i)$, we have remaining $k$ variables to be summed. Although the procedure described here looks different from the description in [@Dimofte:2011py], one can easily check that they are equivalent.
For each tetrahedron $\Delta_i$ in $\textrm{tori}(\varphi) = \bigcup \Delta_i/\sim $, we choose the following polarization $$\begin{aligned}
\Pi_i =\Pi_{Z'}\;, \; \textrm{for all $i$}\;.\label{polarization choice in tetrahedron gluing}\end{aligned}$$ Under this choice, tetrahedron gluing rule can be written in terms of $(\CX_i , \CP_i)$ as in Table \[tetra-gluing 2\].
$\varphi_i \varphi_{i-1}$ $(L,L)$ $(L,R)$ $(R,R)$ $(R,L)$
--------------------------- ---------------------------------- ------------------------------------------ ------------------------------ ------------------------------------------- --
$\mathcal{X}_i$ $\frac{-W_{i+1}+2W_i -W_{i-1}}2$ $\frac{-W_{i+1}+W_i +W_{i-1}-\sV-i\pi}2$ $\frac{W_{i+1}+ W_{i-1}}2$ $\frac{W_{i+1}+ W_i- W_{i-1}+\sV+i\pi}2$
$\mathcal{P}_i$ $i \pi +\frac{\hbar}2 - W_i$ $i \pi +\frac{\hbar}2 - W_i$ $i \pi +\frac{\hbar}2 - W_i$ $i \pi +\frac{\hbar}2 - W_i$
$\sU_i$ $0$ $-\frac{W_i}2 $ 0 $\frac{W_i}2$
: Tetrahedron gluing for tori($\varphi$) in the polarization .[]{data-label="tetra-gluing 2"}
In this polarization choice, the index for the $i$-th tetrahedron in the mapping torus is given by $\CI_{\Delta}\big{(}{\cal X}_i ,{\cal P}_i \big{)}$. The $i$-th tetrahedron’s position/momentum variables $(\CX_i , \CP_i)$ are thought of as magnetic/electric charge of $\CI_{\Delta_i}$ via . They are parametrized by the variables $W_*$. The index for mapping torus can be constructed by multiplying all the indices from each tetrahedron and summing over all $W_i$ variables modulo a ‘meridian’ condition $\sU=\sum_i \sU_i (W_*)$. The condition say that a particular linear combination of $W_*$ is fixed to be a meridian variable $\sU$. $$\begin{aligned}
I^{\Delta}_{\textrm{tori}(\varphi)} (\sV,\sU)= \sum_{ W_i \in \mathbb{Z} } \delta \big{(} \sU-\sum_i \sU_i(W_*) \big{)}\prod_{i=1}^{N} \CI_{\Delta}\big{(}\mathcal{X}_i (W_*), \mathcal{P}_i(W_*) \big{)} \label{mapping torus from tetra}\end{aligned}$$ The factors $(i \pi + \frac{\hbar}2)$ in the argument of $\CI_\Delta$ can be understood from . The cusp boundary variables $(\sU,\sV)$ are in $(\mathbb{Z}, \mathbb{Z})$. As an example, for $\varphi=LR$ $$\begin{aligned}
&I^{\Delta}_{\textrm{tori}(LR)} (\sV,\sU) \nonumber
\\
&= \sum_{W_1,W_2\in \mathbb{Z}} \delta(\sU-\frac{-W_1+ W_2}2)(q^{\half})^{-\half (W_1+W_2)} \CI_{\Delta}(\frac{W_1-\sV}2,-W_1 ) \CI_{\Delta}(\frac{W_2+\sV}2,-W_2) \;. \nonumber\end{aligned}$$ To list a few non-vanishing results, we have $$\begin{aligned}
&I^{\Delta}_{\textrm{tori}(LR)} (0,0)= 1-2q-3q^2+2q^3+\ldots \nonumber
\\
&I^{\Delta}_{\textrm{tori}(LR)} (0,\pm 1)= 2q^{3/2} - 4q^{7/2}+\ldots \nonumber
\\
&I^{\Delta}_{\textrm{tori}(LR)} (\pm 1,\pm 1)= q+q^2-2q^3+\ldots \nonumber
\\
&I^{\Delta}_{\textrm{tori}(LR)} (\pm 1,\pm 2)= q^3 +3q^4 \ldots \nonumber\end{aligned}$$ Comparing these indices with , we find the following non-trivial agreement, $$\begin{aligned}
I^{\Delta}_{\textrm{tori}(LR)} (\sV,\sU) := I^{\textrm{T[SU(2)]}}_{\textrm{tori}(LR)} (m_\eta = \sV, e_\eta = \sU)\;.\end{aligned}$$ In section \[equality between three SL(2,C)/index ptns\], we will show that $I^{\Delta}_{\textrm{tori}(\varphi)} = I^{T[SU(2)]}_{\textrm{tori}(\varphi)}$ for general $\varphi$ with $|\Tr(\varphi)|>2$.
Hilbert-spaces $\CH_{SL(2,\mathbb{C})}$ and $\CH^{\textrm{knot}}_{SL(2,\mathbb{C})}$ {#index : Hilbert-space}
-------------------------------------------------------------------------------------
In this section, we quantize the classical phase spaces $(\mathcal{M},\O)_{SL(2,\mathbb{C})}$ and $(\mathcal{M},\O)^{\rm knot}_{SL(2,\mathbb{C})}$ studied in section \[quantum riemann surface\] and construct the Hilbert-spaces $\mathcal{H}_{SL(2,\mathbb{C})}$ and $\mathcal{H}^{\rm knot}_{SL(2,\mathbb{C})}$. We show explicitly how the quantum operators introduced in section \[quantum riemann surface\] act on the Hilbert-spaces. Based on constructions in this section, we will calculate $SL(2,\mathbb{C})$ CS partition function on mapping cylinder/torus in section \[equality between three SL(2,C)/index ptns\].
#### Hilbert-space $\mathcal{H}_{SL(2,\mathbb{C})}$
As explained in section \[quantum riemann surface\], the phase space $\mathcal{M}(\Sigma_{1,1})_{SL(2,\mathbb{C})}$ can be parameterized by three shear coordinates $(\sT, \sT',\sT'')$ with one linear constraint. The symplectic form $\Omega_{SL(2,\mathbb{C})} $ on the phase space is given in . Rewriting the symplectic form in terms of real and imaginary parts of shear coordinates, $$\begin{aligned}
\Omega_{SL(2,\mathbb{C})} = -\frac{1}{\hbar}d \textrm{Im}(T)\wedge d \textrm{Re}(T') -\frac{1}{\hbar}d \textrm{Re}(T)\wedge d \textrm{Im}(T') \,.\end{aligned}$$ To obtain the Hilbert-space, we first need to specify a choice of ‘real’ polarization. We will choose the following polarization, $$\begin{aligned}
\big{(} X_1, X_2, P_1, P_2\big{)} = \big{(} \textrm{Re}(T'),\half \textrm{Re}(T), -\textrm{Im}(T), 2\textrm{Im}(T') \big{)}\;.\end{aligned}$$ In this choice of real polarization, as noticed in [@Dimofte:2011py], the momenta are periodic variables and thus their conjugate position variables should be quantized. Since the periods for $(P_1,P_2)$ are $(4\pi , 8\pi)$ respectively, the correct quantization condition for $X_1,X_2$ is $$\begin{aligned}
X_1 \in \frac{\hbar}{2}\mathbb{Z}\;, \quad X_2 \in \frac{\hbar}4 \mathbb{Z}\;.\end{aligned}$$ Thus position eigenstates $| X_1, X_2 \rangle$ are labelled by integers and we will introduce charge basis $| m,e \rangle$ as $$\begin{aligned}
| m, e \rangle := | X_1= m \frac{\hbar}2, X_2 = e\frac{\hbar}2 \rangle \;, \quad \textrm{with}\; m, 2e\in \mathbb{Z} \;.\end{aligned}$$ The shear operator $(\sT,\sT')_\pm=(2X_2 \mp i P_2, X_1
\pm \frac{i} 2 P_2) $ acts on the basis as $$\begin{aligned}
\sT_\pm =e\hbar \pm 2\partial_m\;, \quad \sT'_\pm = m \frac{\hbar}2 \mp \partial_e\;.\end{aligned}$$ The exponentiated operators act as $$\begin{aligned}
\langle m,e| \sqrt{\st}_\pm |I \rangle = q^{\half e} \langle m\pm 1, e| I \rangle \;, \quad \langle m,e| \sqrt{\st'}_\pm|I\rangle = q^{\frac{1}4 m} \langle m, e \mp \half | I\rangle \;. \label{t,t' action on (t,t') basis}\end{aligned}$$ Using the basis, the Hilbert-space $\CH_{SL(2,\mathbb{C})}$ can be constructed as $$\begin{aligned}
\CH_{SL(2,\mathbb{C})}= \textrm{Hilbert-space spanned by a basis } \{| m, e \rangle \}_{ m, 2e\in \mathbb{Z}}\;.\end{aligned}$$ One may introduce another basis called fugacity basis $\{ | m, u \rangle\}$ which is related to charge basis by Fourier expansion. $$\begin{aligned}
| m,e \rangle = \oint \frac{du}{2\pi i u} u^{e} | m,u \rangle \;. \label{fugacity basis}\end{aligned}$$ In $\{ | m,u \rangle \}$, $m$ is integer and $u^{1/2}$ is on a unit circle $|u^{1/2}|=1$ in complex plane. As explained in [@Dimofte:2011py], elements in this basis are position eigenstates under the following choice of real polarization $$\begin{aligned}
\big{(} X_1, X_2, P_1, P_2\big{)} = \big{(}\textrm{Re}(T'),\textrm{Im}(T') ,-\textrm{Im}(T), - \textrm{Re}(T)\big{)}\;.\end{aligned}$$
#### Inner product on $\mathcal{H}_{SL(2,\mathbb{C})}$
We defined the adjoint of shear operators in . Adjoint operation depends on the inner-product structure on $\mathcal{H}_{SL(2,\mathbb{C})}$. Requiring consistency between and , one can uniquely determine the inner-product on $\mathcal{H}_{SL(2,\mathbb{C})}$ up to an overall factor $\kappa$. $$\begin{aligned}
\langle m, e|m',e'\rangle = \kappa \delta (m-m')\delta (e-e')\;. \label{inner product on T,T' basis}\end{aligned}$$ For simplicity, we will set $\kappa =1$ by rescaling the charge basis.
#### Basis on $\CH_{SL(2,\mathbb{C})}$ associated to polarization $\Pi$
So far we have only considered two choices of basis, $\{| m,e \rangle \}$ and $\{| m,u \rangle\}$ for $\CH_{SL(2,\mathbb{C})}$. We will introduce more bases $\{| m,e\rangle_\Pi \}$ and $\{| m,u\rangle_\Pi\}$ for $\CH_{SL(2,\mathbb{C})}$, one for each polarization choice $\Pi$ of the phase space $\CM_{SL(2,\mathbb{C})}$. Polarization $\Pi=(\mathcal{X},\mathcal{P})$ is determined by identifying position variable $\mathcal{X}$ and its conjugate momentum variable $\mathcal{P}$ satisfying the canonical commutation relation $[\mathcal{X}_\pm ,\mathcal{P}_\pm] = \mp \hbar$. A simple choice of polarization is $\Pi_{\sT, \sT'} = (\sT',\half \sT)$. The basis $\{| m,e\rangle_{\Pi=(\mathcal{X},\mathcal{P})}\}$ is defined by following conditions $$\begin{aligned}
&_\Pi\langle m,e| e^{\mathcal{X}_\pm } = q^{\frac{m}2} \; _\Pi\langle m,e\mp 1| \;, \quad _\Pi\langle m,e| e^{\mathcal{P}_\pm} =q^{\frac{e}2}\;_\Pi\langle m\pm 1,e| \;.
\label{xp-operation}\end{aligned}$$ These conditions determine the basis $\{| m,e \rangle_\Pi\}$ up to an overall constant which is universal to all basis.[^17] In this notation, basis $| m,e\rangle$ in the above can be understood as $| m,e \rangle_{\Pi}$ with $\Pi = \Pi_{\sT, \sT'}:=(\sT', \half \sT)$. Similarly fugacity basis $| m,u\rangle_\Pi$ associated to a polarization $\Pi$ can be defied as Fourier transformation on $ | m,e\rangle_\Pi$. Under a linear transformation of the polarization $$\begin{aligned}
\left(\begin{array}{c} \tilde{\mathcal{X}}\\ \tilde{\mathcal{P}}\end{array}\right)_\pm =g\cdot \left(\begin{array}{c}\mathcal{ X}\\ \mathcal{P}\end{array}\right) \pm \left(\begin{array}{c} \a \\ \b \end{array}\right) \;, \quad g\in SL(2,Q)\end{aligned}$$ the basis transforms as $$\begin{aligned}
&_{\tilde{\Pi}} \langle m,e| =\;_\Pi \langle g^{-1}\cdot (m,e)| e^{ m \b - e \a} \;.\label{basis polarization transformation}\end{aligned}$$ Note that this transformation rule is equivalent to after identifying the index $\CI (m,e)$ as matrix element $\langle m,e |\CI \rangle$. Under the polarization transformation, the range of charge $(m,e)$ also should be transformed accordingly.
#### SR basis
We define ‘SR basis’ as a basis associated to a polarization $\Pi_{\SR}$, $$\begin{aligned}
\Pi_{\SR} := ({\cal S},{\cal R}) := \half (\sT+\sT', \sT -\sT') \;. \label{SR polarization}\end{aligned}$$ This basis will be denoted as $|m,e\rangle_{\SR}:=| m, e\rangle_{\Pi_{\SR}}$. From the basis transformation , the quantization condition for $(m,e)$ in the SR basis $|m,e\rangle_{\SR}$ is determined: $$\begin{aligned}
m,e \in \frac{\mathbb{Z}}2\;, \quad \textrm{with a condition } m+e\in \mathbb{Z}\;. \label{range of (m,e) for SR}\end{aligned}$$ The inner-product on SR basis takes the same form as , $$\begin{aligned}
_{\SR}\langle m,e|m',e'\rangle_{\SR}= \delta (m-m')\delta(e-e') \;. \label{Inner product on SR basis}\end{aligned}$$ Thus, the completeness relation in the SR basis is $$\begin{aligned}
\mathds{1}_{\CH_{SL(2,\mathbb{C})}} = \sum_{(2m,2e)\in \mathbb{Z}: m+e \in \mathbb{Z}} |m,e\rangle_{\SR} \langle m,e|\;. \label{completeness relation in SR basis}\end{aligned}$$ This SR basis will play a crucial role in section \[equality between three SL(2,C)/index ptns\] in proving $ Z^{\Delta}_{\textrm{tori}(\varphi)} = Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)} $.
#### FN basis
We will introduce yet another basis, called FN (Fenchel-Nielsen) basis, which will play important roles in section \[equality between three SL(2,C)/index ptns\] in proving $I^{T[SU(2)]}_{\textrm{tori}(\varphi)} = Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)}(SL(2,\mathbb{C})) $. The FN charge basis $| m,e \rangle _{\FN}$ is not defined on $\CH_{SL(2,\mathbb{C})}$ but on $\widetilde{\CH}_{SL(2,\mathbb{C})}$, which will be identified with a double cover of $\CH_{SL(2,\mathbb{C})}$. The Hilbert-space $\widetilde{\CH}_{SL(2,\mathbb{C})}$ is defined as $$\begin{aligned}
\widetilde{\CH}_{SL(2,\mathbb{C})}= \textrm{Hilbert-space whose basis are } \{| m, e \rangle_{\FN} \}_{ 2m, e\in \mathbb{Z}}\;.\end{aligned}$$ FN fugacity basis can be defined as Fourier expansion of FN charge basis $$\begin{aligned}
_{\FN}\langle m,e | = \oint \frac{du}{2\pi i u} u^{-e}\; _{\FN}\langle m,u| \;.\end{aligned}$$ In the FN basis, the FN operators $(\hat{\Lambda}, \hat{\mathcal{T}})$, introduced in , act like $(\mathcal{X},\mathcal{P})$, $$\begin{aligned}
&_{\FN}\langle m,e| \hat{\lambda}_\pm = q^{\frac{m}2}\; _{\FN}\langle m,e\mp 1|\;, \quad _{\FN}\langle m,e| \hat{\tau}_\pm = q^{\frac{e}2}\; _{\FN}\langle m \pm 1,e|\;, {\nonumber}\\
&_{\FN}\langle m,u| \hat{\lambda}_\pm = q^{\frac{m}2} u \; _{\FN}\langle m,u|\;, \quad _{\FN}\langle m,u| \hat{\tau}_\pm = e^{\frac{\hbar}2 u \partial_u}\;_{\FN}\langle m \pm 1, u |\;.\end{aligned}$$ In terms of the fugacity basis, the inner-product on $\widetilde{\mathcal{\CH}}_{SL(2,\mathbb{C})}$ is defined as $$\begin{aligned}
_{\FN}\langle m_1, u_1 | m_2, u_2 \rangle_{\FN} &=
\Delta(m_1,u_1)^{-1} \delta (m_1 - m_2)\delta(u_1 - u_2) \,. \label{inner product on FN basis}\end{aligned}$$ where $\Delta(m, u)$ is the measure factor appearing in the $\odot$ operation . The delta function $\delta(u_1,u_2)$ is defined by following condition $$\begin{aligned}
\oint \frac{du_1}{2\pi i u_1} \delta(u_1,u_2) f(u_2) = f(u_1)\;, \quad \textrm{for arbitrary $f(u)$.}
\label{delta-u-1}\end{aligned}$$ The inner product implies the completeness relation in $\widetilde{\CH}_{SL(2,\mathbb{C})}$ in the FN basis, $$\begin{aligned}
\mathds{1}_{\widetilde{\CH}(SL(2,\mathbb{C}))} = \sum_{m \in \mathbb{Z}/2} \oint \frac{du}{2\pi i u}\Delta(m,u) |m,u\rangle_{\FN}\langle m,u| \;. \label{completeness relation in FN basis}\end{aligned}$$ With respect to the inner product, the adjoint of FN operators are $$\begin{aligned}
(\hat{\lambda}_\pm )^\dagger = \hat{ \lambda}_\mp\;, \quad (\hat{\tau}_\pm)^\dagger = \frac{1}{\hat{\lambda}_\mp - \hat{\lambda}_\mp^{-1}} \hat{\tau}_\mp (\hat{\lambda}_\mp - \hat{\lambda}_\mp^{-1})\;. \label{adjoint of FN}\end{aligned}$$ To establish an isomorphism between $\mathcal{H}_{SL(2,\mathbb{C})}$ and a subspace of $\widetilde{\CH}_{SL(2,\mathbb{C})}$, we use the operator relation between FN and shear operators. Combining and , one can show that $$\begin{aligned}
(\sqrt{\st},\sqrt{\st'},\sqrt{\st''})_\pm^\dagger = (\sqrt{\st},\sqrt{\st'},\sqrt{\st''})_\mp\;, \label{adjoint of shear operator}\end{aligned}$$ with respect to the inner product , precisely as we anticipated in .
Using these relations, one can determine the action of the SR operators $(\ss,\sr)_\pm :=(\exp (\CS_\pm), \exp(\CR_\pm))$ in the FN basis. We will consider states $ | m,u\rangle_{\widetilde{\SR}}$ in $\widetilde{\CH}_{SL(2,\mathbb{C})}$ on which operator $(\CS,\CR)$ acts like $(\mathcal{X},\mathcal{P})$, $$\begin{aligned}
_{\widetilde{\SR}}\langle m,u| {\ss}_\pm = q^{\frac{m}2} u^{\pm 1} \;_{\widetilde{\SR}}\langle m,u|\;, \quad _{\widetilde{\SR}}\langle m, u| {\sr}_\pm =e^{\frac{\hbar}2 u \partial_u } \;_{\widetilde{\SR}}\langle m+1,u | \;. \label{def of tilde SR basis}\end{aligned}$$ As we will see in appendix \[basis from FN to shear\], from the above condition one can explicitly express the basis $|m,u\rangle_{\widetilde{\SR}}$ in terms of FN basis $|m,u\rangle_{\FN}$ up to overall constant. The explicit expression copied from appendix \[basis from FN to shear\] is $$\begin{aligned}
_{\widetilde{\SR}}\langle m,u|=\sum_{\tilde{m}} \oint \frac{d\tilde{u}}{2\pi i \tilde{u}} \Delta(\tilde{m},\tilde{u})(-q^{\half} u)^m \CI_{\Delta} (-m-\tilde{m}, u^{-1}\tilde{u}^{-1})\CI_{\Delta}(\tilde{m}-m, \tilde{u}/u) _{\FN}\langle \tilde{m},\tilde{u}| \;.
\label{SR basis in FN basis}\end{aligned}$$ As argued in appendix \[basis from FN to shear\], the range of charge $(m,e)$ for the charge basis $|m,e\rangle_{\widetilde{\SR}}$, which is related to $|m,u\rangle_{\widetilde{\SR}}$ by Fourier expansion, is the same as that of SR basis $|m,e\rangle_{\SR}$ and the inner product on $|m,e\rangle_{\widetilde{\SR}}$ is also the same as that of SR basis . Furthermore, by definition of $|m,e\rangle_{\widetilde{\SR}}$ in , the action of shear operators are the same on the two basis $|m,e\rangle_{\SR}$ and $|m,e\rangle_{\widetilde{\SR}}$. Thus one can naturally identify $$\begin{aligned}
|m,e\rangle_{\widetilde{\SR}} \in \widetilde{\CH}_{SL(2,\mathbb{C})} \textrm{ with } |m,e\rangle_{\SR} \in \CH_{SL(2,\mathbb{C})}\;.\end{aligned}$$ From the above identification, we can consider the SR basis $|m,e\rangle_{\SR}$ as an element in $\widetilde{\CH}_{SL(2,\mathbb{C})}$ and the Hilbert-space $\CH_{SL(2,\mathbb{C})}$ as a subspace of $\widetilde{\CH}_{SL(2,\mathbb{C})}$. The subspace is spanned by $\{ |m,e\rangle_{\widetilde{\SR}} \}$. The Weyl-reflection operator $\sigma$ in acts on $\widetilde{\CH}_{SL(2,\mathbb{C})}$ as $$\begin{aligned}
\sigma \; : \; &| m,u \rangle_{\FN} \rightarrow | -m, u^{-1}\rangle _{\FN}\;,\; \textrm{or equivalently} {\nonumber}\\
&| m,e\rangle _{\FN} \rightarrow | -m, -e \rangle _{\FN}\;.\end{aligned}$$ From the explicit expression , one can easily see that the SR basis $|m,u\rangle_{\SR}$ is Weyl-reflection invariant. Thus we see that $$\begin{aligned}
\mathcal{H}_{SL(2,\mathbb{C})} \subseteq \{ \textrm{$\sigma$-invariant subspace in } \widetilde{\CH}_{SL(2,\mathbb{C})} \}\;.\end{aligned}$$ Furthermore, it is argued in appendix \[basis from FN to shear\] that the equality holds. In other words, the Weyl-reflection invariant combination of the FN basis states $\{ |m,e\rangle^S_{\FN} \}$ form a complete basis for $\CH_{SL(2,\mathbb{C})}$. $$\begin{aligned}
|m,e\rangle^{S}_{\FN}:= \frac{1}2\left(|m,e\rangle_{\FN} + |-m,-e\rangle_{\FN}\right)\;. \label{S-FN basis}\end{aligned}$$
#### Hilbert-space $\CH^{\textrm{knot}}_{SL(2,\mathbb{C})}$
As we have seen in section \[sec: Classical phase space\], the phase space $\CM^{\rm knot}$ is parametrized by ‘longitude’ and ‘meridian’ variables, $\ell = e^{V}$ and $m= e^{U}$. The symplectic form is $$\begin{aligned}
\Omega^{\textrm{knot}}_{SL(2,\mathbb{C})} = \frac{1}{i\hbar} dU \wedge dV - \frac{1}{i\hbar} d\bar{U} \wedge d \bar{V}\end{aligned}$$ We choose the real polarization as $$\begin{aligned}
(X_1, X_2, P_1, P_2) = (\textrm{Re}(V), \textrm{Re}(U), -2 \textrm{Im}(U), 2 \textrm{Im}(V))\end{aligned}$$ Again, since the momenta are periodic variables, their conjugate position variables are quantized. Considering mapping torus, the periodicity of $U$ and $V$ are $4\pi $ and $2\pi $, respectively as we saw in the last paragraph in \[sec: Classical phase space\]. Thus the correct quantization for $X_1, X_2$ seems to be $$\begin{aligned}
X_1 \in \frac{\hbar}4 \mathbb{Z}\;, \quad X_2 \in \frac{\hbar}2 \mathbb{Z}\;.\end{aligned}$$ However, there is an additional quantum $\mathbb{Z}_2$ symmetry which shifts meridian variable $U$ by $2\pi i$ for CS theories on knot complement (see secion 4.2.5 and section 4.2.7 in [@Witten:2010cx][^18]). Taking account of this quantum $\mathbb{Z}_2$ effect, the quantization condition is modified as $$\begin{aligned}
X_1 \in \frac{\hbar}2 \mathbb{Z}\;, \quad X_2 \in \frac{\hbar}2 \mathbb{Z} \,.\end{aligned}$$ This is compatible with the quantization condition for $(m_\eta, e_\eta)\in \mathbb{Z}$ in the mapping torus index computation in section \[index-wall\]. When we consider mapping cylinder, as we already mentioned in section \[sec: Classical phase space\], the period for $V$ is doubled, and the correct quantization is $$\begin{aligned}
X_1 \in \frac{\hbar}2 \mathbb{Z}\;, \quad X_2 \in \frac{\hbar}4 \mathbb{Z}\;.\end{aligned}$$ This quantization is also compatible with the quantization conditions for $(m_\eta, 2 e_\eta) \in \mathbb{Z}$ in mapping cylinder index computation. Since we are also interested in the mapping cylinder index, we will use this quantization conditions in constructing $\CH^{\textrm{knot}}$. After making mapping torus by gluing two boundary $\Sigma_{1,1}$’s, the $SL(2,\mathbb{C})$ CS partition function vanishes automatically when $X_1 = \frac{\hbar}2 (\mathbb{Z}+\half)$ as we will see in section \[Z(Tr)=Z(Delta)\]. We introduce the charge basis $|m_\eta, e_\eta\rangle$ for $\CH^{\textrm{knot}}$ as $$\begin{aligned}
|m_\eta, e_\eta \rangle := |X_1 = \frac{\hbar}2 m_\eta, X_2 = \frac{\hbar}2 e_\eta \rangle\;, \quad (m_\eta, 2e_\eta)\in \mathbb{Z}\;.\end{aligned}$$ and we define $$\begin{aligned}
\CH^{\textrm{knot}}_{SL(2,\mathbb{C})} = \textrm{Hilbert-space spanned by $\{ |m_\eta, e_\eta\rangle \}$}\;.\end{aligned}$$ Using Fourier transformation, we can introduce fugacity basis $\{ |m_\eta, u_\eta\rangle \}$, $$\begin{aligned}
\{ |m_\eta, u_\eta\rangle \; :\; m_\eta \in \mathbb{Z}\;,\; u^{\half}_\eta=e^{i \theta}\; (0 \leq \theta <2\pi)\} \;.\end{aligned}$$ On the charge basis, the operators $(\sV_\pm, \sU_\pm)$, quantum counterparts of $(V, \bar{V}, U, \bar{U})$, act as $$\begin{aligned}
\sV_\pm = \frac{\hbar}2 m_\eta \mp \partial_{e_\eta}\;, \quad \sU_\pm = \frac{\hbar}2 e_\eta \pm \partial_{m_\eta}\;.\end{aligned}$$ In terms of the exponentiated operators $(\fl_\pm, \sm_\pm):=(e^{\sV_\pm}, e^{\sU_\pm})$, the action is given by $$\begin{aligned}
&\langle m_\eta, u_\eta| \fl_\pm = \langle m_\eta, u_\eta| q^{\half m_\eta}u_\eta^{\pm 1}\;,\quad \langle m_\eta, u_\eta| \sm_\pm =e^{\frac{\hbar}2 u_\eta\partial_{u_\eta}}\langle m_\eta\pm 1,e_\eta|\;, {\nonumber}\\
&\langle m_\eta, e_\eta| \fl_\pm = \langle m_\eta, e_\eta \mp 1| q^{\half m_\eta}\;,\quad \langle m_\eta, e_\eta| \sm_\pm =\langle m_\eta\pm 1,e_\eta|q^{\half e_\eta} \;.\end{aligned}$$ The inner-product on the Hilbert space is defined as $$\begin{aligned}
\langle m_\eta, e_\eta |m'_\eta, e'_\eta \rangle = \delta (m_\eta-m'_\eta)\delta (e_\eta - e'_\eta) \;,
\label{inner-product H-u(1)}\end{aligned}$$ which ensures that $$\begin{aligned}
\fl^\dagger_\pm = \fl_\mp\;,\quad \sm^\dagger_\pm = \sm_\pm\;.\end{aligned}$$ The completeness relation in $\CH^{\textrm{knot}}_{SL(2,\mathbb{C})}$ is $$\begin{aligned}
\mathds{1}_{\CH^{\textrm{knot}}_{SL(2,\mathbb{C})}} = \sum_{m_\eta} \oint \frac{d u_\eta}{2\pi i u_\eta} |m_\eta, u_\eta\rangle \langle m_\eta, u_\eta| = \sum_{m_\eta, e_\eta} |m_\eta,e_\eta\rangle \langle m_\eta, e_\eta|\;.\end{aligned}$$ Consider operators $\sO_i (\sV_\pm)$ constructed using only $\sV_\pm$ but not $\sU_\pm $. As already mentioned in section \[quantum riemann surface\], these operators $\sO_i$ can be understood as a state $|\sO_i \rangle$ in $\CH^{\textrm{knot}}$ through the following map, $$\begin{aligned}
&\sO (\sV_\pm) \Leftrightarrow |\sO_i \rangle := \sO_i |\sU=0\rangle \;,{\nonumber}\\
&\textrm{where } |\sU=0\rangle = |P_1 = 0 , X_2 =0\rangle = \sum_{m_\eta }|m_\eta, e_\eta= 0 \rangle\;.\end{aligned}$$ Using the basis $|m_\eta, u_\eta\rangle$, $\sO_i(\sV_\pm)$ can be further mapped to a “wave-function", $$\begin{aligned}
\sO_i(\sV_\pm ) \Leftrightarrow \sO_i (m_\eta, u_\eta) &:=\langle m_\eta, u_\eta| \sO_i \rangle = \langle m_\eta, u_\eta| \sO_i | \sU=0\rangle\;, {\nonumber}\\
&= \langle m_\eta, u_\eta |\sO_i (\sV_\pm) |m_\eta, e_\eta = 0\rangle\;.\end{aligned}$$ The function $\sO_i (m_\eta ,u_\eta)$ obtained in this way is nothing but $$\begin{aligned}
\sO_i (m_\eta, u_\eta) = \sO_i(\sV_\pm )_{\sV_\pm \rightarrow \frac{\hbar}2 m_\eta \pm \log_q u_\eta }\;.\end{aligned}$$ The multiplication of two operator $\sO_1 \cdot \sO_2$ is simply mapped to the multiplication of two functions. The following relations also hold $$\begin{aligned}
&\langle m_\eta, e_\eta | \sO_i |m_\eta, e_\eta =0\rangle = \textrm{Fourier transformation on $u_\eta$ of } \sO_i (m_\eta, u_\eta)\;, {\nonumber}\\
&\langle m_\eta, e_\eta |\sO_1 \ldots \sO_N |m_\eta, 0\rangle = \textrm{Fourier transformation on $u_\eta$ of } \sO_1 (m_\eta, u_\eta ) \ldots \sO_N (m_\eta, u_\eta) \; {\nonumber}\\
&=\sum_{e_{\eta_1},\ldots , e_{\eta_N}} \delta (e_\eta - \sum_{i=1}^N e_{\eta,i } ) \sO_1 (m_\eta, e_{\eta,1}) \ldots \sO_N (m_\eta, e_{\eta, N}) {\nonumber}\\
&= \sum_{e_{\eta_1},\ldots , e_{\eta_N}} \delta (e_\eta - \sum_{i=1}^N e_{\eta,i } ) \langle m_\eta, e_{\eta,1}| \sO_1 |m_\eta, 0 \rangle \ldots \langle m_\eta, e_{\eta, N}| \sO_N |m_\eta, 0 \rangle\;. \label{property of O(l)}\end{aligned}$$ These properties will be used in the below.
### The Hilbert-space from 4d gauge theory {#index: Hilbert-space from 4d}
The AGT relation [@AGT], which relates the $S^4$ partition function for a 4d theory $T_\Sigma$[^19] to a correlation function in 2d Liouville theory on $\Sigma$, can be recast as an isomorphism between the Hilbert-space $\CH(S^3)$ associated with the 4d theory on an omega-deformed four-ball $B^4$ (whose boundary is a squashed $S^3$) and the Hilbert-space $\CH^{Liouv}(\Sigma)$ on the 2d Liouville theory [@Nekrasov:2010ka; @Vartanov:2013ima]. Using dualities among 2d Liouville/Teichmuller/CS theory [@Chekhov:1999tn; @Kashaev:1998; @Teschner:2005; @Teschner:2003em; @Verlinde:1990] the Hilbert-space can be identified with $\CH_{SL(2,\mathbb{R})} (\Sigma)$, the Hilbert-space for $SL(2,\mathbb{R})$ CS theory on $\Sigma$. In this subsection, we try to make parallel stories in the ‘superconformal index version’ of AGT relation [@Gadde:2009kb; @Gadde:2011ik], which relates a superconformal index for 4d theory to a correlation function in 2d TQFT.
In the above we constructed the Hilbert-space $\CH_{SL(2,\mathbb{C})}$ on which operators studied in section \[quantum riemann surface\] act. There is an another space where the operators naturally act on. That is a space of half-indices for 4d $\CN=2^*$ theory which will be denoted as $\CH(S^2 \times S^1 )$. As first noticed in [@Dimofte:2011py], Schur superconformal index for 4d $SU(2)$ $\CN=2^*$ can be written in the following form $$\begin{aligned}
&I_{\mathcal{N}=2^*} (q, u_\eta) =\sum_{m\in \frac{ \mathbb{Z}}2} \oint \frac{du}{2\pi i u }\Delta(m,u) \Pi^{\dagger} (m,u; u_\eta)\Pi (m,u;u_\eta)\;, {\nonumber}\\
&\Pi(m,u; u_\eta) := \delta (m) \prod_{r=0}^\infty \frac{(1-q^{1+r} u^2)(1-q^{1+r})(1-q^{1+r} u^{-2})}{(1-q^{1/2+r}u_\eta u^2)(1-q^{1/2+r}u_\eta )(1-q^{1/2+r}u_\eta u^{-2})}\;.\end{aligned}$$ The Schur index is defined by [@Gadde:2011ik; @Gang:2012yr] $$\begin{aligned}
I_{\CN =2^*} (q, u_\eta) :=\Tr (-1)^{F} q^{\half R + j_3} u_\eta^{H_\eta}\;,\end{aligned}$$ where the trace is taken over a Hilbert-space of the $\CN=2^*$ theory on $S^3$($\times$ time). $j_3$ is a Cartan of the diagonal $SU(2)$ isometry of $SU(2) \times SU(2)$ of $S^3$ and $R$ is a Cartan of the $SU(2)$ R-symmetry. $H_\eta$ is a charge of global $U(1)_{\rm punct}$ which rotates the phase of an adjoint hypermultiplet. The half index $\Pi(m,u)$ can be understood as a (twisted) partition function on three ball $B^3$ (half of $S^3$) with supersymmetric boundary condition labelled by $(m,u)$ imposed on $SU(2)$ vector multiplet at the boundary $S^2 \times S^1$ ($=\partial B^3 \times S^1$). The half index can also be interpreted as a wave-function in a Hilbert-space canonically associated to the boundary. We will identify the half index $\Pi (m,u;u_\eta)$ as a coherent state $|0\rangle \in \CH_{SL(2,\mathbb{C})}$[^20] as follows $$\begin{aligned}
|0\rangle := \sum_{m\in \frac{\mathbb{Z}}2} \oint \frac{du}{2\pi i u} \Delta(m,u) \Pi(m,u; u_\eta ) |m,u\rangle_{\FN}\;.\end{aligned}$$ Then the 4d index is given by the norm of this state, $$\begin{aligned}
I_{\CN =2^*} (q,u_\eta) = \langle 0|0\rangle\;.\end{aligned}$$ One can ‘excite’ the vacuum state $|0\rangle$ by acting operators studied in section \[quantum riemann surface\]. For example, by acting loop operators $\sO_{L} = \sW,\sH, \sD $ on $|0\rangle$ one obtains a half-index $|\sO_L\rangle :=\sO_L |0\rangle $ with insertion of the loop operators. Taking the norm of the state, we could get the 4d superconformal index for $\CN=2^*$ theory with insertion of loop operators at both north and south poles of $S^3$. $$\begin{aligned}
I^{L}_{\CN=2^*}(q,u) = \langle \sO_L | \sO_L\rangle\;, \quad \textrm{4d index with loop operators $L$}\;.\end{aligned}$$ We will define the space of half-indices, $\CH(S^2 \times S^1 )$, as the set of all half-indices $|\sO\rangle$ obtained by acting all quantum operators $\sO (\sqrt{\st}, \sqrt{\st'},\sqrt{\st''})$ on $|0\rangle$. $$\begin{aligned}
\CH(S^2 \times S^1 ) := \{ |\sO\rangle:= \sO \cdot |0\rangle \;:\; \textrm{for all $\sO$} \}\;.\end{aligned}$$ It is obvious that $\CH(S^2 \times S^1)$ is a subspace of $\CH_{SL(2,\mathbb{C})}$. The $SL(2,\mathbb{Z})$ action is closed in the subspace $\CH(S^2 \times S^1)$. In [@Gang:2012ff], the following integral relation was found $$\begin{aligned}
&\sum_{m_t}\oint \frac{du_t}{2\pi i u_t}\Delta (m_t, u_t)I_{\varphi} (m_b, m_t, u_b, u_t;m_\eta =0 , u_\eta) \Pi(m_t, u_t;u_\eta) {\nonumber}\\
&=\Pi(m_b, u_b;u_\eta)\;\quad \textrm{for any $\varphi \in SL(2,\mathbb{Z})$}\;.\end{aligned}$$ In section \[Z(Tr)=Z(T\[SU(2)\])\] (see eq. ), we will identify the duality wall theory index $I_{\varphi}$ as a matrix element of an $SL(2,\mathbb{Z})$ operator $\varphi$ acting on $\CH_{SL(2,\mathbb{C})}$. In this interpretation, the above integral relation can be rewritten in the following simple form, $$\begin{aligned}
\varphi \cdot |0\rangle = |0 \rangle \;\quad \textrm{for anly $\varphi \in SL(2,\mathbb{Z})$}\;.\end{aligned}$$ For general element $|\sO \rangle \in \mathcal{H}(S^2\times S^1)$, $$\begin{aligned}
\varphi |\sO \rangle = \varphi \cdot \sO |0 \rangle = \varphi_*(\sO) \cdot\varphi |0\rangle = |\varphi_*(\sO)\rangle \in \CH (S^2 \times S^1)\;.\end{aligned}$$ In [@Gang:2012yr], it was argued that the 4d superconformal index for $\CN=2^*$ theory is invariant under $SL(2,\mathbb{Z})$ duality. As an example, it is checked in [@Gang:2012yr] that a superconformal index with Wilson line operators is the same as an index with ‘t Hooft line operators, which is $S$-dual of the Wilson line. This $SL(2,\mathbb{Z})$ invariance of the index implies that every $SL(2,\mathbb{Z})$ operator $\varphi$ are unitary operators in $\CH(S^2 \times S^1)$. $$\begin{aligned}
&\langle \varphi_*(\sO_1) | \varphi_*(\sO_2)\rangle = \langle \sO_1 | \varphi^\dagger \varphi |\sO_2\rangle = \langle \sO_1 |\sO_2\rangle\;, \quad\textrm{for all $\sO_1, \sO_2$} {\nonumber}\\
&\therefore \varphi^\dagger \varphi = 1 \; \textrm{in } \CH(S^2 \times S^1)\;.\end{aligned}$$ It is compatible with the observation in section \[quantum riemann surface\] that every $\varphi\in SL(2,\mathbb{Z})$ are unitary in $\CH_{SL(2,\mathbb{C})}$.
Turning off the puncture variable (setting $u_\eta \rightarrow q^\half$ as in [@Gadde:2011ik]) , the ‘vacuum state’ $|0\rangle$ be drastically simplified $$\begin{aligned}
|0\rangle_{\mathbb{T}^2} = |0\rangle_{u_\eta \rightarrow q^\half} = \oint \frac{du}{2\pi i u}\Delta (m=0,u)|m=0,u\rangle\;.\end{aligned}$$ Turning off the puncture, the once-puncture torus $\Sigma_{1,1}$ becomes a torus $\mathbb{T}^2$. Then, the phase space $(\CM,\Omega)_{SL(2,\mathbb{C})}$ becomes much simpler (for example, see [@Aganagic:2002wv]) and the corresponding Hilbert-space $\mathcal{H}_{SL(2,\mathbb{C})}$ also becomes simpler. Note that the above ‘vacuum state’ $|0\rangle_{\mathbb{T}^2}$ is the same as a vacuum state $|0_v\rangle$ in [@Aganagic:2002wv] obtained by quantizing the CS theory on $\mathbb{T}^2$.
So far we have only considered operators of the form $\sO(\sqrt{\st},\sqrt{\st'},\sqrt{\st''})$ which depends only on $\fl$ but not on $\sm$. One can excite $|0\rangle$ by an operator $\sO$ which depends on $\sm$. These operators corresponds to surface operators [@Gaiotto:2012xa] coupled to $U(1)_{\rm punct} $ in 4d $\CN =2^*$ theory. This interpretation is consistent with the results in [@Alday:2013kda], which relate surface operators in 4d $T_\Sigma$ theory with Wilson loop along $S^1$ direction in $\Sigma\times S^1$ in the context of 2d/4d correspondence. Recall that $\sm$ is obtained by quantizing the meridian variable $m = e^{U}$ which measures the holonomy along the $S^1$ direction in $\Sigma_{1,1}\times_\varphi S^1$.
$ Z^{\Delta}_{\textrm{tori}(\varphi)} =Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)} = I^{\textrm{T[SU(2)]}}_{\textrm{tori}(
\varphi)} $ {#equality between three SL(2,C)/index ptns}
--------------------------------------------------------------------------------------------------------------------------------
### $Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)}=Z^{\Delta}_{\textrm{tori}(
\varphi)} $ {#Z(Tr)=Z(Delta)}
In this section, we calculate the CS partition function on ${\rm tori}(\varphi)$ with $|\Tr(\varphi)|>2$ using the canonical quantization on $\Sigma_{1,1}$. As explained in section \[quantum riemann surface\], the partition function can be represented as a trace of a $SL(2,\mathbb{Z})$ operator $\varphi$ (see for $\varphi= \sL, \sR$) on the Hilbert-space $\mathcal{H}_{SL(2,\mathbb{C})}$ and the partition function will be denoted by $Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)}$. We compare the CS partition function with the partition function $Z^{\Delta}_{\textrm{tori}(\varphi)}$(=$I^{\Delta}_{\textrm{tori}(\varphi)}$) calculated in section \[index-tetra\] using tetrahedron decomposition and find an exact match. Classical equivalence of the two approaches was already proven in section \[A-polynomial for mapping torus\] by analyzing A-polynomial, see also [@Terashima:2011xe].[^21]
For a concrete computation of trace of $\varphi$ on $\CH_{SL(2,\mathbb{C})}$, we need to choose a basis of the Hilbert-space. In this section, we use the SR basis introduced in section \[index : Hilbert-space\]. In the SR basis, the matrix element for $\varphi = \sL$,$\sR$ is $$\begin{aligned}
&I^{\SR}_{\varphi=\sL} (m_2, e_2, m_1, e_1; m_\eta,u_\eta) := _{\SR}\langle m_2,e_2 |\sL(m_\eta, u_\eta)| m_1 , e_1\rangle_{\SR}\;, {\nonumber}\\
&I^{\SR}_{\varphi=\sR} (m_2, e_2, m_1, e_1; m_\eta,u_\eta) := _{\SR}\langle m_2, e_2 |\sR(m_\eta,u_\eta) | m_1,e_1\rangle_{\SR} \;.\end{aligned}$$ According to , the right hand sides are the $SL(2,\mathbb{C})$ CS partition functions for mapping cylinders $\Sigma\times_{\varphi=\sL,\sR} I$ in the polarization where positions are $ (\ss_{\rm bot}, \ss_{\rm top}, \fl)$ and momenta are $ (\sr_{\rm bot}, \sr_{\rm top}, \sm)$. Recall that the boundary phase space for the mapping cylinder is locally $\CM_{\rm bot}(\Sigma_{1,1}) \times \CM_{\rm top} (\Sigma_{1,1}) \times \CM^{\rm knot}$ and $(\ss, \sr)$ are shear coordinates for $\CM(\Sigma_{1,1})$ and $(\ell, m)$ are (longitude, meridian) variable for $\CM^{\rm knot}$. Since every operator $\varphi \in SL(2,\mathbb{Z})$ depends only on $\sV_\pm$ but not on $\sU_\pm$, $\varphi$ can be understood as function on $(m_\eta, u_\eta)$ as explained in the paragraph just above section \[index: Hilbert-space from 4d\]. Using properties in , the above indices in charge basis $(m_\eta,e_\eta)$ are $$\begin{aligned}
&I^{\SR}_{\varphi=\sL} (m_2, e_2, m_1, e_1; m_\eta,u_\eta) {\nonumber}\\
&= _{\SR}\langle (m_2, e_2),(m_\eta, e_{\eta})|\sL | (m_1,e_1),( m_\eta,e_\eta=0)\rangle_{\SR} {\nonumber}\\
&= (-1)^{e_2-e_1}\delta (-e_1+2m_2-m_1-m_\eta)\delta(e_2 -e_1+2e_\eta +m_2 -m_1) {\nonumber}\\
&\quad \times q^{\frac{1}4(-e_2+m_2+e_1-m_1)} \CI_{\Delta} (\half (-e_1+m_1+e_2-m_2),m_1+e_1 ) \;,\end{aligned}$$ and $$\begin{aligned}
&I^{\SR}_{\varphi=\sR} (m_2, e_2, m_1, e_1; m_\eta,e_\eta) {\nonumber}\\
&= _{\SR}\langle (m_2, e_2),(m_\eta, e_{\eta})| \sR | (m_1,e_1),( m_\eta,e_\eta=0)\rangle_{\SR} {\nonumber}\\
&=(-1)^{e_2-e_1}(-1)^{m_2 -e_2} \delta (m_2 + m_1 - e_2 + e_1)\delta(e_\eta) q^{ \half ( m_2 + e_1)} \CI_{\Delta} ( -e_1-m_2,m_1 +e_1 ) \;. \label{shear index for L,R}\end{aligned}$$ Here, the state $|(m,e),(m_\eta, e_\eta)\rangle_{\SR}$ denotes a basis state $|m,e\rangle_{\SR}\otimes |m_\eta,e_\eta \rangle$ in $\CH_{SL(2,\mathbb{C})} \otimes \CH^{\textrm{knot}}_{SL(2,\mathbb{C})} $. A derivation for the above formula is given in appendix \[derivation of SR indices\]. The SR basis charges $(m_i,e_i)$ are half-integers with an additional condition $m_i+e_i \in \mathbb{Z}$. The puncture variables $(m_\eta, e_\eta)$ are in $(\mathbb{Z},\mathbb{Z}/2 )$. For later use, we will express these indices in the following form $$\begin{aligned}
&I^{\SR}_{\varphi} (m_2, e_2, m_1, e_1; m_\eta,e_\eta)=(-1)^{e_2-e_1}\delta_{\varphi}(\ldots) \CI_{\Delta} (M_\varphi, E_\varphi) \;, \label{shear index for L,R 2}
\end{aligned}$$ For $\varphi=\sL, \sR$ $$\begin{aligned}
&(M, E)_{\varphi} = \big{(}\frac{-e_1+m_1+e_2-m_2}2,m_1+e_1 + \frac{\hbar}2 \big{)} \;,& {\nonumber}\\
&\delta_{\varphi}(\ldots) = \delta (-e_1+2m_2-m_1-m_\eta)\delta (e_2 -e_1+2e_\eta +m_2 -m_1)\;, &\textrm{for $\varphi=\sL$ } {\nonumber}\\
& \textrm{and}& {\nonumber}\\
&(M, E)_{\varphi} = \big{(}-e_1-m_2+i\pi, m_1+e_1 + \frac{\hbar}2 \big{)} \;, &{\nonumber}\\
&\delta_{\varphi} (\ldots) = \delta(e_\eta)\delta(m_2+m_1-e_2+e_1)\;,&\textrm{for $\varphi=\sR$} \;.
\label{shear index for L,R 3}
\end{aligned}$$ Factors like $(-1)^{\ldots}(q^{1/2})^{\ldots}$ in is reflected in a shift of $(M, E)$ by $i \pi \mathbb{Q}+\frac{\hbar}2 \mathbb{Q}$. Recall our definition of $\CI_\Delta(M,E)$ in . The $SL(2,\mathbb{C})$ CS partition function on $\textrm{tori}(\varphi)$ is given by $$\begin{aligned}
&Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)} (SL(2,\mathbb{C})) (m_\eta, u_\eta) = \textrm{Tr}_{\CH_{SL(2,\mathbb{C})}} \big{(} \varphi (m_\eta, u_\eta) \big{)}\;\; \textrm{in fugacity basis}\;, {\nonumber}\\
&Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)} (SL(2,\mathbb{C})) (m_\eta, e_\eta) = \langle m_\eta, e_\eta |\textrm{Tr}_{\CH_{SL(2,\mathbb{C})}} \varphi |m_\eta,0 \rangle\;\; \textrm{in charge basis}.\end{aligned}$$ Any element $\varphi \in SL(2,\mathbb{Z})$ with $|\Tr\varphi|>2$ can be written as (up to conjugation) $$\begin{aligned}
\varphi = \varphi_N \varphi_{N-1}\ldots \varphi_2 \varphi_1 \;, \quad \varphi_i = \sL \textrm{ or } \sR\;.\end{aligned}$$ Using the completeness relation , the partition function can be written as (the subscript $_{\SR}$ is omitted to avoid clutter) $$\begin{aligned}
&Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)} (SL(2,\mathbb{C})) (m_\eta, e_\eta) = \langle m_\eta, e_\eta |\textrm{Tr}_{\CH_{SL(2,\mathbb{C})}} \varphi |m_\eta,0 \rangle {\nonumber}\\
&= \sum_{\{ e_{\eta,*},m_{\eta,*},m_*,e_* \}} \langle (m_1, e_1),(m_{\eta},e_{\eta}) |\varphi_N |(m_N,e_N),(m_{\eta,N}, e_{\eta,N}) \rangle \ldots {\nonumber}\\
& \qquad \qquad \qquad \qquad \ldots \langle (m_2,e_2),(m_{\eta}, e_{\eta,2}) | \varphi_1 |(m_1, e_1),(m_\eta , 0)\rangle \;,{\nonumber}\\
& = \sum_{\{e_{\eta, *},m_*,e_* \}}\delta(e_\eta -\sum_{k=1}^N e_{\eta,k})\; \langle (m_1, e_1),(m_{\eta},e_{\eta,N}) |\varphi_N |(m_N,e_N),(m_{\eta}, 0) \rangle \ldots {\nonumber}\\
& \qquad \qquad \qquad \qquad \ldots \langle (m_2,e_2),(m_{\eta}, e_{\eta,1}) | \varphi_1 |(m_1,e_1),(m_\eta ,0)\rangle \;, {\nonumber}\\
&=\sum_{\{ e_{\eta,*}, m_*, e_* \}}\delta (e_\eta -\sum_{k=1}^N e_{\eta,k})\ldots I^{\SR}_{\varphi_{i+1}}(m_{i+2},e_{i+2}, m_{i+1},e_{i+1},m_\eta ,e_{\eta,i+1}) \nonumber
\\
&\quad \quad \times I^{\SR}_{\varphi_{i}}(m_{i+1}, e_{i+1},m_{i}, e_{i}, m_\eta ,e_{\eta,i}) I^{\SR}_{\varphi_{i-1}}(m_{i}, e_{i},m_{i-1}, e_{i-1},m_\eta,e_{\eta,i-1}) \ldots \;.
\label{general mapping tori}\end{aligned}$$ In the second line, we used the fact that $\varphi_i$ depends only on $\sV_\pm$ but not on $\sU_\pm$ and the property in eq. . $I^{\Pi_{\SR}}_{\varphi_i}$ in the third line can be written as[^22] $$\begin{aligned}
& I^{\SR}_{\varphi_i} =\delta_{\varphi_i}(\ldots) \CI_{\Delta}(M_{\varphi_i},E_{\varphi_i}) \,,\end{aligned}$$ where $(M_{\varphi_i},E_{\varphi_i}, \delta_{\varphi_i}(\ldots) )$ for $\varphi_i=\sL,\sR$ are given in with $(m_2,e_2, m_1,e_1, m_\eta, e_\eta)$ replaced by $(m_{i+1}, e_{i+1}, m_{i},e_i, m_\eta, e_{\eta,i})$. The index $i$ runs cyclically from $1$ to $N$. There are $2N+1$ Knoneker deltas in the above expression . Among them, $2N$ equations come from $\delta_{\varphi_i}(\ldots)|_{i=1,\ldots, N}$. These $2N$ equations can be solved by parametrizing $(M_{\varphi_i},E_{\varphi_i},e_{\eta,i})|_{i=1}^{N}$ variables ($3N$ in total) in terms of $N$ variables $\{ w_i \}$ $$\begin{aligned}
&M_{\varphi_i} = M_i (w_{i+1},w_i, w_{i-1})\;, \nonumber
\\
&E_{\varphi_i} = E_i (w_{i+1},w_i, w_{i-1}) \;, \nonumber
\\
&e_{\eta,i} = e_{\eta,i}(w_i)\;,
\end{aligned}$$ where the $M_i,E_i,e_{\eta,i}$ is given in Table 5.
$(\varphi_i ,\varphi_{i-1})$ $(L,L)$ $(L,R)$ $(R,R)$ $(R,L)$
------------------------------ ---------------------------------- --------------------------------------------- ----------------------------- ---------------------------------------------- --
$M_i$ $\frac{-w_{i+1}+2w_i -w_{i-1}}2$ $\frac{-w_{i+1}+w_i +w_{i-1}-m_\eta-i\pi}2$ $\frac{w_{i-1}+ w_{i+1}}2$ $\frac{w_{i+1}+ w_i- w_{i-1}+m_\eta+i\pi}2$
$E_i$ $i \pi +\frac{\hbar}2 -w_i$ $i \pi +\frac{\hbar}2 -w_i$ $i \pi +\frac{\hbar}2 -w_i$ $i \pi +\frac{\hbar}2 -w_i$
$e_{\eta,i}$ $\frac{1}2(w_{i+1}-w_i)$ $\frac{1}2(w_{i+1}-w_i)$ 0 0
: Solution for constraints from $2N$ Knonecker delta’s[]{data-label="tetra-gluing 3"}
From straightforward calculation, one can check that these parametrizations satisfy all equations from the $2N$ Kronecker deltas. Substituting this solution into eq. , the CS partition function can be written as $$\begin{aligned}
&Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)} (SL(2,\mathbb{C})) (m_\eta, e_\eta)= \sum_{w_*}\delta \big{(}e_\eta ,\frac{1}2\sum_{\varphi_k =L} (w_{k+1}-w_k) \big{)} \prod_{i=1}^{N} \CI_{\Delta} (M_i(w_*) ,E_i(w_*)) \;.
\label{final general index}\end{aligned}$$ Comparing this index with the index in and comparing Table \[tetra-gluing 2\] and Table \[tetra-gluing 3\], we see the following identification $$\begin{aligned}
&w_i \leftrightarrow W_i, \quad (M_i,E_i)\leftrightarrow (X_i, P_i),\quad m_\eta \leftrightarrow \sV\;.\end{aligned}$$ Under the identification, we see that $$\begin{aligned}
\prod_{i=1}^{N} \CI_{\Delta} (M_i(w_*) ,E_i(w_*)) = \prod_{i=1}^{N} \CI_{\Delta}\big{(}X_i (W_*), P_i(W_*)\;.\end{aligned}$$ The electric charge $e_{\eta}$ for $U(1)_\eta$ is related to $\sU$ in the following way $$\begin{aligned}
&\sU=\sum_i \sU_i = -\frac{1}2 \sum_{(\varphi_i,\varphi_{i-1}) = (L,R)}W_i +\frac{1}2 \sum_{(\varphi_i,\varphi_{i-1}) = (R,L)}W_i \;, \nonumber
\\
&e_\eta=\sum_i e_{\eta,i} =\frac{1}2 \sum_{\varphi_i = L} (w_{i+1}-w_i) =\sU\;,\quad \textrm{under the identification $w_i = W_i$} \;.\end{aligned}$$ From the above identifications, we see that $$\begin{aligned}
Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)} (SL(2,\mathbb{C})) (m_\eta, e_\eta)= I^{\Delta}_{\textrm{tori}(\varphi)} (\sV,\sU)|_{\sV=m_\eta, \sU=e_\eta}\;.\end{aligned}$$ for general $\varphi$ with $|\Tr (\varphi)|>2 $. One remarkable property of $Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)} (SL(2,\mathbb{C}))$ is that it always vanishes when $e_\eta \in \mathbb{Z}+\half$.
### $Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)}= I^{\textrm{T[SU(2)]}}_{\textrm{tori}(
\varphi)} $ {#Z(Tr)=Z(T[SU(2)])}
#### Duality wall index as matrix element in FN basis
We will argue that the mapping cylinder index $I_\varphi$ studied in section \[index-wall\] can be written as the matrix element of $\varphi \in SL(2,\mathbb{Z})$ in the FN basis. More explicitly, $$\begin{aligned}
&I_{\varphi} (m_b, u_b, m_t,u_t,m_\eta,u_\eta):=_{\FN}\langle m_b,u_b | \varphi (m_\eta, u_\eta)| m_t, u_t \rangle_{\FN} \;, \quad \textrm{or equivalently} \nonumber
\\
&I_{\varphi} (m_b, u_b , m_t,u_t,m_\eta,e_\eta):=_{\FN} \langle (m_b,u_b),(m_\eta,e_\eta)| \varphi | (m_t,u_t),(m_\eta, 0)\rangle_{\FN} \;. \label{mapping cylinder in hilbert}
\end{aligned}$$ for any operator $\varphi \in SL(2,\mathbb{Z})$ acting on a Hilbert space $ \CH_{SL(2,\mathbb{C})}$.[^23] The right hand side is the $SL(2,\mathbb{C})$ CS partition function for mapping cylinder in the polarization where positions are $ (\hat{\lambda}_{\rm bot}, \hat{\lambda}_{\rm top}, \fl)$ and momenta are $ (\hat{\tau}_{\rm bot}, \hat{\tau}_{\rm top}, \sm)$. Thus, the above statement is nothing but the 3d-3d dictionary in for $M= \Sigma_{1,1}\times_\varphi I$. Assuming eq. holds, the index for mapping torus theory can be represented as $$\begin{aligned}
I^{T[SU(2)]}_{\textrm{tori}(\varphi)} (m_\eta,u_\eta) &= \sum_m \oint \frac{du}{2\pi i u}\Delta(m,u) I_{\varphi}(m,u,m,u,m_\eta, u_\eta)\;, \quad \textrm{from }\eqref{mapping torus index from duality wall}{\nonumber}\\
&= \sum_m \oint \frac{du}{2\pi i u} \Delta (m,u) \;_{FN}\langle m_b, u_b| \varphi(m_\eta, u_\eta)|m_t, u_t\rangle_{FN}\;, \quad \textrm{from \eqref{mapping cylinder in hilbert}} {\nonumber}\\
&= \Tr_{\mathcal{H}_{SL(2,\mathbb{C})}} \varphi (m_\eta, u_\eta)\;, \quad \textrm{using } \eqref{completeness relation in FN basis}\;.\label{mapping torus in hilbert}
\end{aligned}$$ Note that the quantity in the last line is nothing but $Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)}(SL(2,\mathbb{C}))$. Thus the proposal automatically ensures that $Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)}(SL(2,\mathbb{C}))= I^{\textrm{T[SU(2)]}}_{\textrm{tori}(
\varphi)} $, which is the main result of this section. How can we justify the proposal in ? There are two steps in the argument for the proposal. First we will argue that the proposal holds for $\varphi=\sL, \sR$ by $i)$ showing the two sides in satisfy the same difference equations and $ii)$ by directly comparing the two sides in $q$-expansion. Then, we will prove that $$\begin{aligned}
\textrm{ If the proposal \eqref{mapping cylinder in hilbert} holds for $\varphi_1, \varphi_2$, then it also holds for $\varphi = \varphi_2 \cdot \varphi_1 $}. \label{proposition on gluing}
\end{aligned}$$ From the two arguments, we can claim that holds for general $\varphi\in SL(2,\mathbb{Z})$ which can be written as a product of $\sL$’s and $\sR$’s.
#### Proof of
Since the second argument is much simpler to prove, let’s prove it first. Suppose holds for $\varphi_1$ and $\varphi_2$, then $$\begin{aligned}
& _{\FN}\langle m_b,u_b | \varphi_2 \varphi_1 | m_t,u_t \rangle_{\FN} \nonumber
\\
&=\sum_{m'} \oint \frac{du'}{2\pi i u'} \Delta (m',u')\langle m_b,u_b| \varphi_2 | m',u' \rangle \langle m',u' | \varphi_1 | m_t, u_t \rangle \;, \quad \textrm{using } \eqref{completeness relation in FN basis} \nonumber
\\
&= \sum_{m'} \oint \frac{du'}{2\pi i u'} \Delta (m',u') I_{\varphi_2} (m_b, u_b, m', u'; m_\eta, u_\eta)I_{\varphi_1} (m', u',m_t, u_t;m_\eta, u_\eta) \;, \nonumber
\\
&= I_{\varphi_2 \varphi_1 }(m_b, u_b, m_t, u_t;m_\eta, e_\eta) \;, \quad \textrm{using } \eqref{odot gluing rule in T[SU(2)] theories}\;.\end{aligned}$$ Thus the proposal also holds $\varphi = \varphi_2 \cdot \varphi_1$.
#### Check of for $\varphi= \sL, \sR$ by difference equations
The index for $T[SU(2)]$ described in section \[index-wall\] satisfies the following difference equations,\
$$\begin{aligned}
\big{(}\sW_b - (\sH^{\rm T})_t \big{)}_\pm \cdot I_{\varphi=S} =0\;, \quad (\sH_b - (\sW^{\rm T})_t)_\pm \cdot I_{\varphi=S}=0\;, {\nonumber}\\
\bigg{(}p_\eta-\big{(}\frac{1}{p^\half-p^{-\half}}(x -x^{-1}) \big{)}_b \bigg{)}_\pm \cdot I_{\varphi=S} =0\;. \label{diff for S}\end{aligned}$$ The Wilson loop operator $\sW$ and ‘t Hooft operator $\sH$ are given by (cf. )[^24] $$\begin{aligned}
&\sW_\pm = x_\pm +x_\pm^{-1} \;, {\nonumber}\\
&\sH_\pm= \frac{q^{\mp 1/4} x_\pm x^{\half}_{\eta;\pm}- q^{\pm 1/4} x^{-1}_{\pm} x^{-\half}_{\eta,\pm} }{x_\pm- x^{-1}_\pm} p_\pm^{-\half} + \frac{q^{\pm 1/4} x_\pm x_{\eta,\pm}^{-\half}- q^{\mp 1/4} x_\pm^{-1} x^\half_{\eta,\pm}}{x_\pm- x_\pm^{-1}} p_\pm^{\half}\;.\end{aligned}$$ Basic operators $(x_\pm, p_\pm)$ act on the charge basis index as $$\begin{aligned}
& x_\pm := \exp ( \frac{\hbar}2 m \mp \partial_e)\;, \; p_\pm := \exp ( \frac{\hbar}2 e \pm \partial_m)\;.\end{aligned}$$ On the fugacity basis index, they act as $$\begin{aligned}
& x_\pm := \exp ( \frac{\hbar}2 m \pm \log u)\;, \; p_\pm := \exp ( \frac{\hbar}2 \partial_{\log u} \pm \partial_m) \;.\end{aligned}$$ Depending on the subscript $(b,t,\eta)$, they act on (‘bot’,‘top’,‘punct’) parameters, respectively. The notation $\sO^{\rm T}$ denotes a ‘transpose’ of $\sO$ to be defined for each operator. For $\sW$, $\sH$, the transposed operators are $$\begin{aligned}
\sW^{\rm T} = \sW\;, \quad \sH^{\rm T} =\sH/.\{q\rightarrow q^{-1}, p\rightarrow p^{-1}\} \;.\end{aligned}$$ The difference equations can be simplified using ‘shear’ operators. Shear operators $(\sqrt{\st},\sqrt{\st'},\sqrt{\st''})_{\pm}:= (\exp(\frac{1}2 \sT),\exp(\frac{1}2 \sT'),\exp(\frac{1}2 \sT''))_\pm$ are defined as (cf. ) $$\begin{aligned}
&(\sqrt{\st})_{\pm} = \frac{i}{x_{\pm} -x_{\pm}^{-1}}(p_{\pm}^{-1/2} -p_{\pm}^{1/2}) \;,\nonumber
\\
&(\sqrt{\st'})_{\pm} = \frac{i}{q^{\mp 1/4}x_{\pm}^{-1}p_{\pm}^{1/2} - q^{\pm1/4}p_{\pm}^{-1/2} x_{\pm}}(x_\pm -x_{\pm}^{-1}) \;, \nonumber
\\
&(\sqrt{\st''})_{\pm} = q^{\mp 1/4}x^\half_{\eta,\pm} \frac{i}{p^{-1/2}_\pm -p^{1/2}_\pm}(q^{\mp 1/4}x_{\pm}^{-1}p_{\pm}^{1/2} - q^{\pm1/4}p_{\pm}^{-1/2} x_{\pm}) \;. \label{shear operators}
\end{aligned}$$ In terms of the shear operators, the difference equations can be written as $$\begin{aligned}
\big{(}(\frac{1}{\sqrt{\st}})_b - (\sqrt{\st}^{\rm T})_t \big{)}_\pm \cdot I_{\varphi=S} =0\;, \quad \big{(}\sqrt{\st'} (1+q^{\half} \st)_b - (\sqrt{\st''}^{\rm T})_t \big{)}_\pm \cdot I_{\varphi=S} =0\;, {\nonumber}\\
\big{(} p_\eta- i (\frac{1}{\sqrt{\st}})_b \big{)}_\pm \cdot I_{\varphi=S} =0\;. \label{diff for S-1}\end{aligned}$$ For shear operators, the transposed operators are $$\begin{aligned}
(\sqrt{\st},\sqrt{\st'}, \sqrt{\st ''})^{\rm T} =(\sqrt{\st},\sqrt{\st'}, \sqrt{\st ''}) /. \{q\rightarrow q^{-1}, p\rightarrow p^{-1}\}\;. \label{Shear transpose}\end{aligned}$$ For $\varphi=\sL (=S^{-1}T^{-1}S) , \sR(=T)$, the corresponding duality wall theory indices $I_\varphi$ satisfy following difference equations. $$\begin{aligned}
&\big{(}(\frac{1}{\sqrt{\st''}})_b - (\sqrt{\st}^{\rm T})_t\big{)}_\pm \cdot I_{\varphi=\sL} =0\;, \quad \big{(} \sqrt{\st}_{\pm}(1+q^{\pm 1/2}\st''_\pm)_b-(\sqrt{\st''}^{\rm T}_\pm)_t\big{)} \cdot I_{\varphi=\sL} =0 \;, \nonumber
\\
&\big{(} p_\eta - \frac{(\sqrt{\st}^{\rm T})_t}{(\sqrt{\st})_b} \big{)}_\pm \cdot I_{\varphi=\sL} =0\;, \label{diff for L}
\\
&\big{(}(\frac{1}{\sqrt{\st'}})_b - (\sqrt{\st}^{\rm T})_t\big{)}_\pm \cdot I_{\varphi=\sR} =0\;, \quad \big{(}\sqrt{\st''}_{\pm}(1+q^{\pm 1/2}\st'_\pm)_b- (\sqrt{\st''}^{\rm T}_\pm)_t\big{)} \cdot I_{\varphi=\sR} =0 \;, \nonumber
\\
&\big{(} p_\eta - 1\big{)}_\pm \cdot I_{\varphi=\sR} =0\;. \label{diff for R}
\end{aligned}$$ One can check these difference equations by series expansion in $q$ at any desired order. For $\varphi=\sR$, we have a closed expression for $I_\varphi$, from which we can check that $$\begin{aligned}
&(x_t - x_b)_\pm \cdot I_{\varphi=\sR} = 0\;, \quad (p_{\eta,\pm} -1)\cdot I_{\varphi=\sR}=0\;, {\nonumber}\\
&\big{(} (\frac{1}{x_\pm-x_\pm^{-1}}p_\pm^{-\half} (x_\pm-x_\pm^{-1}))_t - (q^{\mp \frac{1}4}x^{-1}_\pm p^{1/2}_\pm)_b \big{)}\cdot I_{\varphi=\sR} =0\;, \label{diff for R-2}\end{aligned}$$ by a brute-force computation. Expressing these difference equations in terms of shear operators, we obtain the difference equations for $I_{\varphi=\sR}$ in .
Among the three difference equations in each of , and , two are of the form $\varphi_* (\sO)_b-\sO^{\rm T}_t \simeq 0$. From a purely 3d field theory point of view, there is no prior reason for that. As we will see below, this structure of the difference equations can be naturally understood from . Another interesting property of these difference equations is that they are always in $\pm$ pair. It is related to the factorization of 3d superconformal indices [@Beem:2012mb; @Hwang:2012jh] and this property is not restricted on duality wall theories. How can we guess these difference equations? Difference equations of the form $\varphi_* (\sO)_b-\sO^{\rm T}_t \simeq 0$ are largely motivated by the difference equations for $S^3_b$ partition function for $T[SU(2),\varphi]$ theory studied in [@Dimofte:2011jd]. From the works [@Dimofte:2011ju; @Dimofte:2011py], we know that the $S^3_b$ partition function and $S^2\times S^1$ superconformal index satisfy the same form of difference equations. A direct way of obtaining the difference equations is expressing the mapping cylinder indices in terms of tetrahedron indices and using the gluing rules for difference equations explained in [@Dimofte:2011gm] (see also [@Dimofte:2011py]). As we will see in appendix \[T\[SU(2)\] classical\], $I_{\varphi=S}$ can be expressed by gluing 5 tetrahedron indices with two internal edges. However, the corresponding operator equations for difference equation gluing is too complicated to solve. In appendix \[T\[SU(2)\] classical\], we consider classical Lagrangian (set of difference equations in the limit $q\rightarrow 1$) for $I_{\varphi=S}$. In the classical limit, operator equations become equations for ordinary commuting variables that are relatively easy to solve. In this way, we obtain the difference equations for $I_{\varphi=S}$ in the classical limit and check these exactly matches the difference equations in eq. with $q=1$. We want to emphasize that a pair of difference equations involving $p_\eta$ is obtained from quantization of a classical equation involving $p_\eta$ in the classical Lagrangian obtained in appendix \[T\[SU(2)\] classical\]. The ordering ambiguity is fixed by checking corresponding difference equation in $q$ expansion.
Now let us consider difference equations satisfied by the matrix element in right-hand side of . From the operator equations for $\varphi=\sL, \sR$ and the following observations, $$\begin{aligned}
&_{\FN}\langle m, e| \sO(\hat{\lambda}_\pm, \hat{\tau}_\pm)| I\rangle = \sO(x_\pm, p_\pm)\cdot \;_{\FN}\langle m, e|I\rangle\;, \nonumber
\\
&\langle I | \sO(\hat{\lambda}_\pm, \hat{\tau}_\pm)| m,- e\rangle_{\FN} = \sO^{\rm T}(x_\pm, p_{\pm})\cdot \langle I|m,-e\rangle_{\FN}\;, {\nonumber}\\
&\langle m_\eta, e_\eta|\sm_\pm \cdot \sO \cdot \sm_\pm^{-1} |m_\eta, e_\eta =0\rangle = p_{\eta,\pm} \cdot \langle m_\eta, e_\eta| \sO |m_\eta, e_\eta =0\rangle\;,\end{aligned}$$ one can check that the matrix element in satisfies the same difference equations in and for $\varphi=\sL,\sR$. The transposed operator $\sO^{\rm T}$ is defined by $\sO^{\rm T} = (\sO^\dagger)^*$, where the complex conjugation $*$ is given as $$\begin{aligned}
(c\; \hat{\lambda}_\pm^{m_1} \hat{\tau}_\pm^{m_2} \fl_\pm^{m_3} q^{m_4})^* = c^* \hat{ \lambda}_\mp ^{m_1} \hat{\tau}_\mp^{-m_2} \fl_\mp^{m_3} q^{m_4}\;, \quad \textrm{$c$ : $c$-number}\end{aligned}$$ Transpose of shear operators are (using eq. ) $$\begin{aligned}
(\sqrt{\st},\sqrt{\st'},\sqrt{\st''})^{\rm T} =(\sqrt{\st},\sqrt{\st'},\sqrt{\st''})/.\{ q\rightarrow q^{-1}, \hat{\tau} \rightarrow \hat{\tau}^{-1}\}\;.\end{aligned}$$ It is compatible with .
#### Check of for $\varphi= \sL, \sR$ by direct computation in $q$ expansion
A more direct evidence for the proposal in is an explicit comparison of both sides in $q$-expansion. Plugging the completeness relation in the SR basis into , $$\begin{aligned}
&(\textrm{right-hand side in \eqref{mapping cylinder in hilbert}}) {\nonumber}\\
&=
\sum_{(m_1,e_1)} \sum_{(m_2, e_2)} \;_{\FN}\langle m_b, u_b | m_2, e_2 \rangle_{\SR} \langle m_2, e_2| \varphi | m_1 , e_1 \rangle _{\SR}
\langle m_1, e_1 | m_t, u_t \rangle_{\FN} \, ,
\label{equiv}\end{aligned}$$ and using the following relation between SR and FN basis in [^25] $$_{\SR}\langle m_1, e_1 | m_t, u_t \rangle_{\FN}
= \sum_{ e \in \mathbb{Z} } (-1)^{m_1} q^{ \half m_1} u_t^{ - 2e - e_1+m_1}\CI_{ \Delta} (-m_1 -m_t , e) \CI_{ \Delta}( -m_1+m_t , - e_1+m_1-e) \, ,
{\nonumber}$$ one obtains the following expression $$\begin{aligned}
&_{\FN} \langle (m_b,u_b),(m_\eta,e_\eta)| \varphi | (m_t,u_t),(m_\eta, 0)\rangle_{\FN} {\nonumber}\\
&=
\sum_{(m_1, e_1)} \sum_{(m_2, e_2)}
\sum_{e,e' \in {\mathbb Z}}
(-1)^{m_1-m_2} q^{ \half (m_1+m_2)} u_b^{ 2e+e_2-m_2} u_t^{ -2e'-e_1+m_1} {\nonumber}\\
& \times \CI_{ \Delta} (-m_2-m_b,e) \CI_{ \Delta}(-m_2+m_b, -e_2+m_2-e) \CI_{ \Delta}(-m_1-m_t, e') {\nonumber}\\
&\times \CI_{ \Delta}(-m_1+m_t, -e_1+m_1-e')
_{\SR}\langle (m_2,e_2),(m_{ \eta}, e_{ \eta}) | \varphi | (m_1, e_1), (m_{ \eta}, 0) \rangle_{\SR} \, .
\label{result-fn}\end{aligned}$$ The summation ranges are over $m_i,e_i \in \frac{\mathbb{Z}}2$ such that $e_i+m_i \in \mathbb{Z}$ due to the completeness relation in SR basis. Plugging the matrix elements in eq. into eq. , one obtains $$\begin{aligned}
&_{\FN} \langle (m_b,u_b),(m_\eta,e_\eta)| \sL | (m_t,u_t),(m_\eta, 0)\rangle_{\FN} {\nonumber}\\
&=
\sum_{m_1, e_1, m_2, e_2, e,e' }
(-1)^{m_1-m_2+e_2-e_1} q^{ \frac{1}{4} (m_1+e_1+3m_2-e_2)} u_b^{ 2e+e_2-m_2} u_t^{ -2e'-e_1+m_1} {\nonumber}\\
& \times \delta (-e_1+2m_2-m_1-m_\eta)\delta(e_2-e_1+2e_{\eta} + m_2 - m_1 ) {\nonumber}\\
& \times \CI_{ \Delta} (-m_2-m_b,e) \CI_{ \Delta}(-m_2+m_b, -e_2+m_2-e) \CI_{ \Delta}(-m_1-m_t, e') {\nonumber}\\
& \times \CI_{ \Delta}(-m_1+m_t, -e_1+m_1-e') \CI_{ \Delta} ( \half(-e_1+m_1+e_2-m_2), m_1+e_1)
\label{L-fn} \;.\end{aligned}$$ We show some examples of explicit evaluation of the above formula in $q$-expansion $$\begin{aligned}
&_{\FN} \langle m_b,u_b| \sL (m_\eta, u_\eta)| m_t,u_t\rangle_{\FN} \qquad \textrm{for }(m_b, m_t, m_{ \eta}) = (0,0,0) {\nonumber}\\
&= \sum_{e_{ \eta} \in \mathbb{Z}/2} {_{\FN} \langle (m_b,u_b),(m_\eta,e_\eta)| \sL | (m_t,u_t),(m_\eta, 0)\rangle_{\FN}}\; u_{ \eta}^{ e_{ \eta}} \qquad \textrm{for }(m_b, m_t, m_{ \eta}) = (0,0,0) {\nonumber}\\
& = 1 + \big{(} \frac{1}{u_{ \eta}} \chi_1 (u_t)+ u_{ \eta} \chi_1 (u_b) \big{)} q^{ \half} + \big{(}-1- \chi_1 (u_b) - \chi_1(u_t) + u_{ \eta}^{-2} \chi_2 (u_t) + u_{ \eta}^2 \chi_{2} (u_b) \big{)} q + O (q^{ \frac{3}{2}})
{\nonumber}\\
&_{\FN} \langle m_b,u_b| \sL (m_\eta, u_\eta)| m_t,u_t\rangle_{\FN} \qquad \textrm{for }(m_b, m_t, m_{ \eta}) = (0,0,1) {\nonumber}\\
& = \left( u_{ \eta}^{-1} \chi_{ \half} (u_b) \chi_{ \half} (u_t) - \chi_{ \half} (u_b) \chi_{ \half} (u_t) \right) q + O(q^{ \frac{3}{2}})
{\nonumber}\\
& _{\FN} \langle m_b,u_b| \sL (m_\eta, u_\eta)| m_t,u_t\rangle_{\FN} \qquad \textrm{for }(m_b, m_t, m_{ \eta}) = (1,0,1) {\nonumber}\\
&= \frac{u_b}{ u_{ \eta}} \chi_{ \half} (u_t) q^{ \half} + \left( u_b^{-1} \chi_{ \half} (u_t) - u_b (-u_{ \eta}^{-2} + u_{ \eta}^{-1}) \chi_{ \frac{3}{2}}(u_t) \right) q^{ \frac{3}{2} }+ O(q^2)
{\nonumber}\end{aligned}$$ where $ \chi_j (u) $ is the character for $2j+1$ dimensional representation of $SU(2)$, $ \chi_j (u) := \sum_{l=-j}^{j} u^{ 2j}$. The result agrees with the index $I_{\sL}(m_b, u_b, m_t, u_t; m_\eta, e_\eta)$ obtained using the duality domain wall theory in section \[index-wall\].
For $\varphi=\sR$, we will start from the index in the FN basis. In eq. , we found that $$\begin{aligned}
I_{\varphi = \sR} = _{\FN}\langle m_{b},u_b |\sR |m_t ,u_t\rangle_{\FN} = u_b^{2m_b}\delta (m_b- m_t) \frac{\delta(u_b - u_t)}{\Delta(m_t, u_t )}\;.
\end{aligned}$$ Performing the basis change from FN to SR, we find $$\begin{aligned}
_{\SR}\langle m_2, e_2 | \sR |m_1, e_1\rangle_{\SR} = \sum_{m_b} \oint \frac{du_b}{2\pi i u_b} u_b^{2m_b}\Delta(m_b, u_b) _{\SR}\langle m_2 ,e_2| m_b, u_b\rangle_{\FN} \langle m_b, u_b | m_1, e_1\rangle_{\SR} \;.{\nonumber}\end{aligned}$$ Using , the above formula can be explicitly evaluated in $q$-expansion, which matches with $I^{\SR}_{\varphi=\sR}$ in .
Squashed sphere partition function/$SL(2,\mathbb{R})$ CS partition function {#sec: SL(2,R)/squashed S3}
===========================================================================
The squashed three sphere partition function of $T[SU(2), \varphi]$ has been discussed extensively in recent literature [@Dimofte:2011ju; @Dimofte:2011jd; @Terashima:2011qi; @Terashima:2011xe; @Vartanov:2013ima], where its relation to the $SL(2,\IR)$ CS partition function and quantum Teichmüller theory was pointed out. In this section, we review some salient features of the these work to help clarify the similarities and differences between the superconformal index of the previous section and the three sphere partition function.
#### Quantum dilogarithm identities
Before we proceed, let us take a brief digression to review some properties of the non-compact quantum dilogarithm (QDL) function [@Faddeev:1995nb; @Kashaev] which plays a fundamental role throughout this section and in comparison with section \[sec: SCI/SL(2,C)\].
1. Definition $(q_\pm =e^{2\pi i b^{\pm 2}})$: $$\begin{aligned}
e_b(x)
= \prod_{r=1}^{\infty} \frac{1+ (q_+)^{r-\half} e^{2\pi bx}}{1+(q_-)^{\half-r}e^{2\pi x/b}}
= \exp\left( \frac{1}{4} \int_{\IR+i\e} \frac{dw}{w} \frac{e^{-2ixw}}{\sinh(wb)\sinh(w/b)} \right) \,.
\label{qd-def}\end{aligned}$$
2. The zeros $(z_{mn})$ and poles $(p_{mn})$ of $e_b(x)$ are located at $$\begin{aligned}
z_{mn} = -c_b -i( m b + n b^{-1} ) \,,
\quad
p_{mn} = + c_b + i( m b + n b^{-1} ) \,,
\quad
(m,n\in \IZ_{\ge 0})\,.
\label{zero-pole}\end{aligned}$$ with $c_b = i(b+ b^{-1})/2$.
3. Inversion formula: $$\begin{aligned}
e_b(x) e_b(-x) = e^{\frac{\pi i}{12} (b^2+b^{-2})} e^{\pi i x^2} \,.
\label{q-inv}\end{aligned}$$
4. Quasi-periodicity and difference equation: $$\begin{aligned}
&e_b(x+ib) = (1+q_+^{\half}e^{2\pi bx})^{-1} \,e_b(x) \,,
\quad
e_b(x-ib) = (1+q_+^{-\half}e^{2\pi bx}) \,e_b(x) \,,
\label{shift-b}
\\
&e_b(x+i/b) = (1+q_-^{\half}e^{2\pi x/b})^{-1} \,e_b(x) \,,
\quad
e_b(x-i/b) = (1+q_-^{-\half}e^{2\pi x/b}) \,e_b(x) \,,
\label{shift-1/b}\end{aligned}$$ The second identity in can be rewritten as $$\begin{aligned}
\left(e^{-ib\partial_x} - q_+^{-\half} e^{2\pi b x} -1 \right) e_b(x) =
(\hat{z}''-\hat{z}^{-1} -1 ) e_b(x) = 0 \,,\end{aligned}$$ with $\hat{z}'' \equiv e^{-ib\partial_x}$, $\hat{z} \equiv q^{\half}e^{-2\pi bx}$ satisfying $\hat{z}''\hat{z} = q \hat{z} \hat{z}''$.
5. Generalized Fourier transform: $$\begin{aligned}
\Psi_n(\a_1,\ldots,\a_n;\b_1,\ldots,\b_{n-1};w)
\equiv \int_{\IR} dx \, e^{2\pi i x(w-c_b)}
\prod_{j=1}^n \frac{e_b(x+\a_j)}{e_b(x+\b_j-c_b)} \,
\label{Psi_n}\end{aligned}$$ with $\b_n=i0$ satisfy a number of identities, the simplest of which include $$\begin{aligned}
&\Psi_1(\a;w) = e^{\pi i (b^2+b^{-2}+3)/12} \frac{e_b(\a)e_b(w)}{e_b(\a+w-c_b)} \,,
\\
&\Psi_2(\a_1,\a_2;\b;w) = \frac{e_b(\a_1)}{e_b(\b-\a_2)}\Psi_2(\b-\a_2;w;\a_1+w;\a_2) \,.
\label{Psi_2}\end{aligned}$$
Duality wall theory \[Z-wall\]
-------------------------------
The partition function on the squashed three sphere, $S_b^3$, is obtained in [@Hama:2011ea] for general 3d ${\cal N}=2$ gauge theories. Here $b$ is the dimensionless squashing parameter normalized such that $b=1$ corresponds to the round sphere.
Let us first consider the partition function for the mass-deformed $T[SU(2), \varphi = S]$ $$\begin{aligned}
\CZ_S( \mu, \zeta, m )
&= s_b ( - m) \int d \sigma
\frac{ s_{b} ( \mu + \sigma + \frac{m}{2} + \frac{c_b}{2} ) s_{b} ( \mu - \sigma + \frac{m}{2} + \frac{c_b}{2} ) }
{s_{b} ( \mu + \sigma - \frac{m}{2} - \frac{c_b}{2} ) s_{b} ( \mu - \sigma - \frac{m}{2} - \frac{c_b}{2} ) }
e^{ 4 \pi i \sigma \zeta } \, . \label{ftn:tsu2}\end{aligned}$$ Here, $\mu$ denotes the mass for fundamental hyper-multiplets and $ \zeta$ the FI parameter. The phase factor $e^{ 4 \pi i \sigma \zeta}$ originates from the FI term. The double sine function $s_b(x)$ is defined as $$s_b \left( c_b ( 1-r) - \sigma \right) = \prod_{m, n \geq 0} \left( \frac{mb+nb^{-1}+ i \sigma -ic_b (2-r) } {mb+nb^{-1}- i \sigma-ic_b r } \right) \,,
\label{double-sine}$$ where $c_b = i(b+ b^{-1})/2$. This function is related to the QDL function $e_b(x)$ by $$\begin{aligned}
e_b(x) = e^{\frac{\pi i}{24} (b^2+b^{-2})} e^{\frac{\pi i}{2} x^2}\,.\end{aligned}$$ To simplify , we used an identity $s_b (x) s_b (-x)=1$ which is equivalent to .
The double sine function in is a contribution to the one-loop determinant from a free chiral multiplet with R-charge $r$ (for the scalar field) which is coupled to a background $U(1)$ . Thus $s_b (-m)$ in eq. originates from the adjoint chiral multiplet of $T[SU(2)]$, and the other four $s_b$ functions are from the four fundamental chiral multiplets; see Table \[tsu(2)-charge\].
Let us first generalize to the partition function of $T[SU(2), T^k ST^{l}]$. Recall that the multiplication of $T^k$ and $T^l$ elements add background CS terms with level $k$ and $l$ for the two $SU(2)$’s to the theory. The classical contributions from the CS terms shall be multiplied to the partition function as follows, $$\CZ_{T^l ST^k} ( \mu, \zeta , m)
= e^{ -2 \pi i l \mu^2 } e^{ - 2 \pi i k \zeta^2 } \CZ_S ( \mu, \zeta , m) \, .
{\nonumber}$$ For the multiplication of $SL(2,\IZ)$ elements, $ \varphi = \varphi_2 \cdot \varphi_1 $, the partition function can be obtained by ‘gluing’ $$\begin{aligned}
\CZ_{\varphi_2 \cdot \varphi_1} ( \mu, \zeta , m)
&= \int [ d \nu] \CZ_{ \varphi_2 } ( \mu, \nu ,m ) \CZ_{ \varphi_1 }( \nu, \zeta , m )
\,,
{\nonumber}\end{aligned}$$ where $[d \nu ] = d \nu \sinh (2 \pi b \nu) \sinh ( 2 \pi b^{-1} \nu) $ is the measure with the contribution from a vector multiplet of the gauged $SU(2)$ global symmetry.
As an application of the QDL identities, we prove the ‘self-mirror’ property of $\CZ_S$: $$\begin{aligned}
\CZ_S(\mu,\z,m) = \CZ_S(\z,\mu,-m) \,.\end{aligned}$$ For $b=1$, this property was proved earlier in [@Hosomichi:2010vh]. We begin with replacing $s_b$ in by $e_b$. Up to an overall normalization that may depend on $b$ but no other parameters, we find $$\begin{aligned}
\CZ_S(\mu, \zeta, m) =
e^{\pi i (2\mu +m -c_b)(2\z-m-c_b)-\frac{\pi i}{2}m^2}
e_b(-m) \Psi_2(m,2\mu+m;2\mu;2\z-m) \,,
\label{ftn:tsu2b}\end{aligned}$$ where the function $\Psi_n$ was defined in . The self-mirror property follows easily from the identity and .
Tetrahedron decomposition
-------------------------
The computation of the $S^3_b$ partition function of $T_M$ using the tetrahedron decomposition of $M$ was explained in [@Dimofte:2011ju], which parallels the computation of the index reviewed in section \[index-tetra\]. In the polarization $\Pi_Z$, the partition function is given by $$\begin{aligned}
\CZ_\Delta(x) = e_b(c_b-x) =
\prod_{r=1}^{\infty} \frac{1-(q_+)^{r} e^{-2\pi b x}}{1-(q_-)^{1-r} e^{-2\pi x /b}},\end{aligned}$$ where the real and imaginary part of the complex parameter $x$ correspond to the twisted mass and the R-charge of the elementary chiral multiplet $\phi_Z$.
The $SL(2,\IZ)$ polarization change acts on $\CZ_\Delta$ as follows, $$\begin{aligned}
&T: \CZ_\Delta(x) \;\; \goto \;\; \CZ'(x) = e^{-\pi i x^2} \CZ_\Delta(x) \,,
{\nonumber}\\
&S: \CZ_\Delta(x) \;\; \goto \;\; \CZ'(x') = \int dx e^{-2\pi i x x'} \CZ_\Delta(x) \,.
\label{TS-ZD}\end{aligned}$$ The computation of the partition function for $T_M$ can be done in three steps (see section 6.2 of [@Dimofte:2011ju]). First, one takes $\CZ_{\Delta_i}(x_i)$, for each tetrahedron $\Delta_i$ in the triangulation and multiplies them all. Second, act with $Sp(2N,\IZ)$ to transform to a polarization in which all internal edges are “positions". Third, set the parameters $x_I$ corresponding to the internal edges equal to $2c_b$.
Thus, the procedure is conceptually identical to that of the superconformal index. In practice, the partition function is slightly more difficult to deal with because the $S$ operation in involves Fourier transformation, whereas the $S$ operation for the index can be treated as a linear transformation on the lattice of basis states of the Hilbert space.
Quantization for $G= SL(2,\mathbb{R})$
--------------------------------------
#### An (approximate) isomorphism of operator algebra
In the previous sections, we saw how the CS theory with the non-compact gauge group $G=SL(2,\mathbb{C})$ is related to the superconformal index of 3d field theories. For the squashed sphere partition functions being discussed in this section, the relevant gauge group is $G=SL(2,\mathbb{R})$.
Since $SL(2,\mathbb{R})$ is a real slice of $SL(2,\mathbb{C})$, the phase space $\CM_{SL(2,\mathbb{R})}$ is also a real slice of $\CM_{SL(2,\mathbb{C})}$ in a suitable sense. Recall that we obtained the $(\pm)$ pair of operators after quantization because the coordinates for $\CM_{SL(2,\mathbb{C})}$ are complex variables, and that the $(+)$ operators commute with the $(-)$ operators. In contrast, the coordinates are real variables for $\CM_{SL(2,\mathbb{R})}$. So, at first sight, the splitting into $(+)$ and $(-)$ operators seem unlikely.
Remarkably, as first noted in [@Faddeev:1995nb; @Chekhov:1999tn] and brought into the present context in [@Dimofte:2011jd], the algebra of exponentiated operators for $G=SL(2,\mathbb{R})$ does factorizes into two mutually commuting subalgebras. To be explicit, we begin by introducing rescaled (logarithmic) shear operators as follows, $$\begin{aligned}
(\sT, \sT',\sT'') = 2 \pi b (\hat{\sT} , \hat{\sT'} , \hat{\sT''})\;,
\qquad
[\hat{\sT}, \hat{ \sT'}] = [\hat{\sT'}, \hat{ \sT''}]= [\hat{\sT''}, \hat{ \sT}]=\frac{i}{\pi}\;.\end{aligned}$$ The $(\pm)$ pair of exponentiated shear operators are defined by $$\begin{aligned}
\st_\pm := \exp (2\pi b^{\pm 1} \hat{\sT})\;, \quad \textrm{similarly for $\st'_\pm , \st''_\pm$}\;.\end{aligned}$$ Note that $$\begin{aligned}
[\st^{*}_+, \st^{**}_-]=0\;, \quad \textrm{for any $^*$, $^{**}$}\;.\end{aligned}$$ The $(+)$ shear operators are nothing but the original shear operators. The $(-)$ shear operators satisfy formally the same commutation relations but with the original quantum parameter $q_+:= q=e^{2\pi i b^2}$ replaced by $q_- := \exp (2\pi i b^{-2})$.
If we consider composite operators made of integer powers of $(\st_\pm, \st'_\pm, \st''_\pm)$ only, there exists an isomorphism between the operator algebra for $G=SL(2,\mathbb{R})$ and that for $G=SL(2,\mathbb{C})$ with the understanding that $(q_\pm)_{SL(2,\mathbb{R})}$ are mapped to $(q^{\pm 1})_{SL(2,\mathbb{C})}$. However, this isomorphism breaks down slightly if the square-root operators $(\sqrt{\st}, \sqrt{\st'}, \sqrt{\st''})$ are included. For instance, $\sqrt{\st}_\pm \sqrt{\st'}_\mp = \sqrt{\st}_\mp \sqrt{\st'}_\pm$ in the $SL(2,\mathbb{C})$ case whereas $\sqrt{\st}_\pm \sqrt{\st'}_\mp =- \sqrt{\st}_\mp \sqrt{\st'}_\pm$ in the $SL(2,\mathbb{R})$ case.
The (approximate) operator isomorphism will be the key to understanding a large degree of similarity between the computations in section \[sec: SCI/SL(2,C)\] and those in this section.
#### Shear/SR basis for the Hilbert-space
The Hilbert space $\CH_{SL(2,\IR)}$ is the familiar $L^2(\mathbb{R})$ for the quantum mechanics on a real line [@Kashaev]. To see this, we recombine two independent shear coordinates, $(\sT, \sT')$, to form a position-momentum pair. [^26] $$\begin{aligned}
[\sT,\sT'] = 2\hbar = 4\pi i b^2 \;\;
\imp \;\;
(2\pi b) {\cal S} \equiv \frac{\sT+\sT'}{2} \,, \;\;
(2\pi b) {\cal R} \equiv \frac{\sT-\sT'}{2} \,, \;\;
[{\cal R},{\cal S}] = \frac{i}{2\pi} \,.
\label{Z-shear-SR}\end{aligned}$$ Unlike in section \[sec: SCI/SL(2,C)\], we take $\sT$, $\sT'$ to be Hermitian. Then, $\CS$ and $\CR$, also Hermitian, can be identified with the position/momentum operators for quantum Teichmller theory originally introduced in [@Kashaev]. We define the ‘position’ and ‘momentum’ basis in the usual manner, [^27] $$\begin{aligned}
&\langle x | \CS = x \langle x | \,,
\quad
\langle x | \CR =
\frac{i}{2\pi} \frac{\partial}{\partial x} \langle x |\,,
\quad
\langle x | x' \rangle = \d(x-x') \,,
\quad
\int_{-\infty}^{\infty} | x \rangle \langle x | =1 \,.
{\nonumber}\\
&\langle p | \CR = p \langle p | \,,
\quad
\langle p | \CS =
\frac{1}{2\pi i} \frac{\partial}{\partial p} \langle p |\,,
\quad
\langle p | p' \rangle = \d(p-p') \,,
\quad
\int_{-\infty}^{\infty} | p \rangle \langle p | =1 \,.\end{aligned}$$ The transformation between the two bases can be performed as usual, $$\begin{aligned}
\langle x | p \rangle = e^{-2\pi i p x} \,, \quad
\int e^{-2\pi i p (x-x')} dp = \d(x-x') \,.\end{aligned}$$ The exponentiated SR operators, $$\begin{aligned}
\mathsf{s}_\pm \equiv e^{2\pi b^{\pm} \CS} \,,
\quad
\mathsf{r}_\pm \equiv e^{2\pi b^{\pm} \CR} \,.\end{aligned}$$ satisfy the following commutation relations, $$\begin{aligned}
&\mathsf{r}_\pm \mathsf{s}_\pm = q_\pm \mathsf{s}_\pm \mathsf{r}_\pm \,, \quad
[\mathsf{r}_\pm , \mathsf{r}_\mp] = 0 \,, \quad
[ \mathsf{r}_+ , \mathsf{r}_-] = 0 = [ \mathsf{s}_+ , \mathsf{s}_- ] \,.
\label{def-XP-Z}\end{aligned}$$ If we only needed operators of the type $\mathsf{r}^a \mathsf{s}^b$ with $(a,b)\in \IZ^2$, the isomorphism between the operator algebra of this section and that of section \[sec: SCI/SL(2,C)\] would have been exact, with minor modifications in the hermiticity condition and the definition of $q_\pm$. But, the operators $\sqrt{\st}$, $\sqrt{\st'}$ forces us to include ‘half-integer points’ $(a,b) \in \IZ/2$ with $a+b\in \IZ$ in the lattice of operators, which induces subtleties such as $\sqrt{\st}_\pm \sqrt{\st'}_\mp =- \sqrt{\st}_\mp \sqrt{\st'}_\pm$.
In the previous section, we encountered lattices of states such as $|m,e \rangle$ as well as lattices of operators such as $\mathsf{r}^a \mathsf{s}^b$. Here, while it is not clear how to organize the states of the Hilbert space on a lattice, we can still untilize the lattice structure of operators. For instance, it is useful to consider the generators of linear $SL(2,\mathbb{Z})$ polarization changes (not to be confused with the $SL(2,\mathbb{Z})$ of QTT which acts non-linearly on the shear coordinates) $$\begin{aligned}
e^{\pm \pi i \mathcal{S}^2}
\begin{pmatrix}
\mathcal{S} \\ \mathcal{R}
\end{pmatrix}
e^{\mp \pi i \mathcal{S}^2}
= R^{\pm 1}
\begin{pmatrix}
\mathcal{S} \\ \mathcal{R}
\end{pmatrix} \,,
\qquad
e^{\pm \pi i \mathcal{R}^2}
\begin{pmatrix}
\mathcal{S} \\ \mathcal{R}
\end{pmatrix}
e^{\mp \pi i \mathcal{R}^2}
= L^{\mp 1}
\begin{pmatrix}
\mathcal{S} \\ \mathcal{R}
\end{pmatrix} \,,
\label{pol-LR}\end{aligned}$$ where $R$ and $L$ are as in .
#### QDL and isomorphism of operator algebra revisited
We can promote the QDL function $e_b(x)$ to an operator by substituting $\sx$ for $x$, $$\begin{aligned}
e_b(\sx) = \prod_{r=1}^{\infty} \frac{1+(q_+)^{r-\thalf} e^{2\pi b \sx}}{1+(q_-)^{\thalf-r} e^{2\pi \sx /b}} = \prod_{r=1}^{\infty} \frac{1+(q_+)^{r-\thalf} e^{\CX_+}}{1+(q_-)^{\thalf-r} e^{\CX_-}} \equiv \CE_{q_+,q_-}(\CX_+,\CX_-) \,.
\label{QDL-pm-split}\end{aligned}$$ We will often use the short-hand notation $\CE(\CX)$ for $\CE_{q_+,q_-}(\CX_+,\CX_-)$. Similar operators can be defined in other polarizations of $L^2(\mathbb{R})$ by replacing $\CX_\pm$ by linear combinations of $\CX_\pm$ and $\CP_\pm = 2\pi b^{\pm 1} \mathsf{p}$ with integer coefficients. Here, we assume $[\mathsf{x},\mathsf{p}]=(2\pi i)^{-1}$.
This quantum version of the QDL function underlies essentially all non-trivial identities among various constructions. In addition, through the isomorphism of the operator algebra, the QDL function can also be used to illuminate the parallel between this section and section \[sec: SCI/SL(2,C)\]. To see this, note that $\CZ_\Delta(x)$ of the tetrahedron decomposition can be regarded as a collection of eigenvalues, $$\begin{aligned}
\langle x | \CE(\CX_+, \CX_- )
= \langle x | \CZ_\Delta(c_b-x) \,.\end{aligned}$$ Similarly, we can define the QDL operator $\CE(\CX)$ for the index computation by $$\begin{aligned}
\CE(\CX) = \prod_{r=1}^{\infty} \frac{1+(q_+)^{r-\thalf} e^{\CX_+}}{1+(q_-)^{\thalf-r} e^{\CX_-}} \,,\end{aligned}$$ where $e^{\CX_\pm}$ operators are defined as in and $q_\pm$ is to be understood as $q_\pm = q^{\pm 1}$. By construction, $\CE(\CX)$ is diagonal in the fugacity basis, $$\begin{aligned}
\langle m, u| \CE(\CX) = \CI_\Delta(-m,(-q^{\thalf})u^{-1}) \langle m, u| \,,\end{aligned}$$ where $\CI_\Delta$ on the right-hand-side is precisely the tetrahedron index . Now, let us see how the QDL identities imply identities for $\CI_{\Delta}$. From the operator version of the inversion identity and the $SL(2,\IZ)$ polarization change , we find $$\begin{aligned}
\sum_{e_1} (-q^{\thalf})^{e+2e_1} \CI(-m,e_1) \CI(m,e+e_1)
= \delta_{e,-m} \,.\end{aligned}$$ This identity appeared in [@Dimofte:2011py] in the computation of the index for the trefoil knot complement. Similarly, by taking the matrix elements of the quantum pentagon relation, $$\begin{aligned}
\CE(\CX) \CE(\CP) = \CE(\CP) \CE(\CX+\CP) \CE(\CX) \,,
\label{q-pentagon-2-copy}\end{aligned}$$ and reshuflling the indices using the parity and triality , we find $$\begin{aligned}
&\CI_\Delta(m_1-e_2,e_1) \CI_\Delta(m_2-e_1,e_2)
{\nonumber}\\
&\qquad = \sum_{e_3} q^{e_3} \CI_\Delta(m_1,e_1+e_3) \CI_\Delta(m_2,e_2+e_3)
\CI_\Delta(m_1+m_2,e_3) \,.\end{aligned}$$ This identity was used in [@Dimofte:2011py] to show the equivalence bewteen the two mirror descriptions of the bipyramid theory.
#### FN basis
The isomophism of the operator algebras allows us to use the same relation between the shear and FN operators . In [@Kashaev], what we might call the FN basis, in which $\hat{\lambda}+\hat{\lambda}^{-1}$ becomes diagonal, was defined through its relation to the SR basis. Using the loop-FN-shear triality explained in section \[qsl2c\], we take the Wilson loop operators $$\begin{aligned}
\sW^{\pm} = (\mathsf{r}_\pm)^{-1} + \mathsf{s}_\pm + (\mathsf{s}_\pm)^{-1} \,.\end{aligned}$$ In addition, we introduce the Dehn twist operator [@Kashaev], $$\begin{aligned}
\sD = e^{2\pi i(\mathcal{S}^2-c_b^2)} e_b(\mathcal{S}-\mathcal{R}) \,.
\label{dehnx}\end{aligned}$$ The loop operators and the Dehn twist operator commute with each other, $$\begin{aligned}
[ \sW^+ , \sW^-] = 0\,, \quad
[ \sD, \sW^\pm ] = 0 \,.\end{aligned}$$ Next, following [@Kashaev], but slightly modifying the normalization to conform to the convetions of [@Dimofte:2011jd], we introduce the states $| \mu)$ $(\mu \in \IR_+)$ by specifying the matrix elements, $$\begin{aligned}
\langle x | \mu) &= \frac{e_b(\mu+x+c_b-i0)}{e_b(\mu-x-c_b+ i0)} e^{-2\pi i(x+c_b)\mu + \pi i \mu^2}
{\nonumber}\\
&= (q_+ q_-)^{-\frac{1}{24}} \CZ_\Delta(-x-\mu) \CZ_\Delta(-x+\mu) e^{-\pi i(x+c_b)^2}
\,.
\label{SR-FN-DG}\end{aligned}$$ As proved in [@Kashaev], the FN basis vectors are simultaneous eigenstates of the loop and Dehn twist operators, $$\begin{aligned}
\sO^\pm | \mu ) = 2\cosh(2\pi b^\pm \mu) | \mu ) \equiv (\hat{\lambda}_\pm+(\hat{\lambda}_\pm)^{-1}) | \mu )
\,, \quad
\sD | \mu ) = e^{2\pi i (\mu^2 -c_b^2)} | \mu ) \,,
\label{LD-eigenx}\end{aligned}$$ The second expression in makes it clear that $\langle x |{\rm -}\mu) = \langle x |\mu)$ for all $x, \mu \in \mathbb{R}$. To avoid double-counting, we restrict the range of $\mu$ to $\mathbb{R}_+$, which reflects the $\IZ_2$ Weyl symmetry of the $T[SU(2)]$ theory. The FN basis vectors satisfy the orthogonality and completeness relations compatible with the results of section \[Z-wall\]. $$\begin{aligned}
(\mu|\nu) = \frac{\d(\mu - \nu)}{4\sinh(2\pi\mu b)\sinh(2\pi\mu/b)} \,,
\quad
\int_0^\infty 4\sinh(2\pi\mu b)\sinh(2\pi\mu/b) |\mu)(\mu| d\mu= 1\,.
\label{FN-Z-ortho-compl}\end{aligned}$$ We close this subsection by noting that the following hermiticity of the FN operators, $$\begin{aligned}
(\lambda_\pm)^\dagger = \lambda_\pm \,,
\quad
(\tau_\pm)^\dagger =
\frac{1}{\lambda_\pm - \lambda_\pm^{-1}} \tau_\pm (\lambda_\pm - \lambda_\pm^{-1}) \,,
\label{Z-FN-hermi}\end{aligned}$$ is compatible with the measure in and the hermiticity of the shear operators.
#### Proving the equivalence: $ Z^{\Delta}_{\textrm{tori}(\varphi)} =Z^{\textrm{Tr}(\varphi)}_{\textrm{tori}(\varphi)} =\CZ^{\textrm{T[SU(2)]}}_{\textrm{tori}(
\varphi)} $
Given the close parallel between the computation of the index and that of the partition function, it is natural to expect that the equality of three quantities explained in \[equality between three SL(2,C)/index ptns\] can be carried over to this section. The second equality is essentially a basis change between the shear basis and the FN basis. Since an explicit form of the basis change is known , the proof in \[equality between three SL(2,C)/index ptns\] can be repeated with little modification by using the isomorphism of operator algebra. The first equality is somewhat less trivial. The Hilbert space for the partition function does not exhibit a lattice structure on which the $SL(2,\IZ)$ polarization change linearly. Nevertheless, we expect that the proof of section \[Z(Tr)=Z(Delta)\] can be adapted to the current context by a suitable combination of $SL(2,\IZ)$ actions in .
We are grateful to Yoon Pyo Hong, Kazuo Hosomichi, Hee-Cheol Kim, Sung-Soo Kim, Bum-Hoon Lee, Kimyeong Lee, Yunseok Seo, Sang-Jin Sin, Yang Zhou and especially Sergei Gukov for helpful discussions. The work of SL is supported in part by the National Research Foundation of Korea (NRF) Grants NRF-2012R1A1B3001085 and NRF-2012R1A2A2A02046739. J.P. was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MEST) with the Grants No. 2012-009117, 2012-046278 and 2005-0049409 through the Center for Quantum Spacetime (CQUeST) of Sogang University. JP is also supported by the POSTECH Basic Science Research Institute Grant and appreciates APCTP for its stimulating environment for research.
**Appendix**
Alternative descriptions of $T[SU(2)]$ {#dual description for T[SU(2)]}
======================================
In this section, we will discuss other possible descriptions of $T[SU(2)]$.
The first dual theory is ${\cal N}=2$ $SU(2)$ Chern-Simons theory of level $k=1$ with four fundamental and three neutral chiral multiplets, found in [@Teschner:2012em]. The authors of the paper found that the squashed three sphere partition function of the mass-deformed $T[SU(2)]$ theory can be interpreted as a partition function of the dual theory. The 3d superconformal index for the dual theory also has been shown to coincide with the index for $T[SU(2)]$ [@Gang:2012ff]. The dual theory has the following advantage. The $SO(4) \simeq SU(2) \times SU(2)$ flavor symmetry of the dual theory, which corresponds to $SU(2)_{\rm top} \times SU(2)_{\rm bot}$ global symmetry of $T[SU(2)]$, is manifest in the Lagrangian. Thus the operations on the superconformal index described in section \[index-wall\] (i.e., adding Chern-Simons action and/or gluing) can be incorporated at the Lagrangian level. Accordingly, the resulting theory dual to $T[SU(2), \varphi]$ has a Lagrangian description.
Other dual theories of $T[SU(2)]$ can be found from the brane set-up for $T[SU(2)]$ theory given in Figure \[fig:brane\]-(a). Taking a limit where the length between NS5-branes is very small, the theory on a D3-brane becomes a 3d $U(1)$ theory. Two D5-branes give rise to two fundamental hyper-multiplets. Taking T-dual transformation of $SL(2, \mathbb{Z})$ of type IIB theory on (a) results in (b), i.e., two NS5-branes are mapped to (NS5,D5)=(1,1)-branes while D5-branes are invariant. The IR limit of (b) corresponds to $U(1)_1 \times U(1)_0 \times U(1)_{-1}$ Chern-Simons theory where subscripts denote the CS levels. The theory has two bi-fundamental hyper-multiplets with charges $(1,-1,0)$ and $(0,1,-1)$ under the gauge group. Crossing a D5-brane over the left NS5-brane in (b), a D3-brane is created between the 5-branes due to Hanany-Witten effect, as depicted in (c). It corresponds to $U(1)_{-1} \times U(1)_1 \times U(1)_{-1}$ theory with bi-fundamental hyper-multiplets. Crossing the other D5-brane over NS5-brane in turn results in the brane set-up in (d), which corresponds to $U(1)_0 \times U(2)_{-1} \times U(1)_0 $ theory. Since all four theories have Lagrangian descriptions, we can use the prescription in [@Imamura:2011su] to write down index formulas. We checked that the superconformal indices as a function of fugacity for $ \epsilon +j$ coincide with one aother to several orders in fugacity.
![Brane setups for dual theories of $T[SU(2)]$. Horizontal/vertical/tilted lines represent D3/D5/(1,1)-branes. Crossed circles represent NS5-branes. []{data-label="fig:brane"}](brane.pdf)
Difference equations for $T[SU(2),\varphi]$ at classical limit {#T[SU(2)] classical}
==============================================================
The index of $T[SU(2)]$ can be constructed from tetrahedron indices $\CI_{ \Delta}$ as follows $$\begin{aligned}
& I_{\varphi=S} (m_b, m_t, m_{ \eta}; u_b, u_t, u_\eta ) \nonumber \\
& = (-1)^{ 2m_b} \CI_{ \Delta} ( - m_{ \eta} , u_\eta^{-1} q^{ \frac{1}{2}}) u_ \eta^{ m_{ \eta}}
\sum_{m_s} \oint \frac{ d u_s}{ 2 \pi i u_s} u_s^{ 2m_t + 2m_s} u_t^{2m_s} u_b^{2 m_b} q^{ - \frac{1}{4} m_{ \eta}} {\nonumber}\\
& \quad \times \prod_{ \epsilon_1, \epsilon_2 = \pm1}
\CI_{ \Delta} ( \epsilon_1 m_b +\half m_{ \eta} + \epsilon_2 m_s , u_b^{ \epsilon_1} u_\eta^\half u_s^{ \epsilon_2} q^{ \frac{1}{4}} ) \; . \label{TSU2 index using tetrahedron index}\end{aligned}$$ In the charge basis, the index become $$\begin{aligned}
&I_{\varphi =S} (m_b, m_t, m_{ \eta};e_b, e_t, e_\eta )=
\sum_{e_1, e_2 \in \mathbb{Z}} \prod_{i=1}^5 \CI_\Delta( \tilde{m}_i , \tilde{e}_i)\;, \;\textrm{where} {\nonumber}\\
&(\tilde{m}_1, \tilde{e}_1)= (-m_{ \eta}+\frac{ \hbar}{2} , e_1 - \frac{ \hbar}{4}) \;, \;(\tilde{m}_2, \tilde{e}_2)= (m_b+\half m_{ \eta}+\frac{e_t}{2} + \frac{ \hbar}{4} , e_2 - i \pi ) \;, \nonumber \\
&(\tilde{m}_3 ,\tilde{e}_3) = ( -m_b+\half m_{ \eta}- \frac{e_t}{2}+ \frac{ \hbar}{4} , e_2-\frac{e_b}{2}+\frac{e_t}{2}+m_b+m_t )\;,
\nonumber \\
& (\tilde{m}_4,\tilde{ e}_4)= (-m_b+\half m_{ \eta}+\frac{e_t}{2} + \frac{ \hbar}{4} , e_1 - e_2 +e_{ \eta} - \frac{e_t}{2} - m_{ \eta} - m_t+ i \pi ) \;,
\nonumber \\
&(\tilde{m}_5, \tilde{e}_5)= (m_b+\half m_{ \eta}- \frac{e_t}{2}+ \frac{ \hbar}{4} , e_1-e_2+ \frac{e_b}{2} +e_{ \eta} - m_b-m_{ \eta})
\; .
\label{tetra-tsu2}
\end{aligned}$$ We have five difference equations for the index $\prod_{i=1}^5 \CI_{ \Delta}(\tilde{m}_i, \tilde{e}_i)$, $$\tilde{x}_i^{-1} + \tilde{p}_i -1 =0 \, , \qquad \mbox{for } i = 1, 2,\ldots , 5\;.
\nonumber$$ where $ \tilde{x}_i, \tilde{p}_i$ can be written in terms of $(p_1, p_2, x_b, x_t, x_\eta, p_b, p_t, p_\eta)$ from eq. ; see also ). We will focus on the classical limit, $ \hbar \to 0$ from here on. In the classical limit, $$\begin{aligned}
& \tilde{x}_1 = \frac{1}{x_{ \eta}}, \quad \tilde{x}_2 = p_t^{ \frac{1}{2}} x_b x^\half_{ \eta} , \quad \tilde{x}_3 = p_t^{ - \half} x^{-1}_b x^{\half}_{ \eta},
\quad \tilde{x}_4 = p_t^{ \frac{1}{2}} x^{-1}_b x^\half_{ \eta} , \quad \tilde{x}_5 = p_t^{ - \frac{1}{2}} x_b x^\half_{ \eta} \, ,
\nonumber \\
& \tilde{p}_1 = p_1, \quad \tilde{p}_2 = - p_2, \quad \tilde{p}_3 = \frac{p_t^{ \frac{1}{2}} p_2 x_b x_t}{ p_b^{ \frac{1}{2}}},
\quad \tilde{p}_4 = - \frac{ p_{ \eta} p_1}{p_t^{ \frac{1}{2}} p_2 x_{ \eta} x_t},
\quad \tilde{p}_5 = \frac{ p_b^{ \frac{1}{2}} p_{ \eta} p_1}{ p_2 x_b x_{ \eta} } \, .
\nonumber\end{aligned}$$ The summations $ \sum_{e_1, e_2}$ corresponds to integrating out $p_1, p_2$ from difference equations. We use the first two equations, $ \tilde{x}_{1} + \tilde{p}_{1}^{-1}-1=0$ and $ \tilde{x}_{2} + \tilde{p}_{2}^{-1}-1=0$ to integrate out $p_1, p_2$ $$p_1 = 1 - x_{ \eta} , \quad p_2 = \frac{1- p_t^{ \frac{1}{2}} x_b x^{\half}_{ \eta}}{ p_t^{ \frac{1}{2} } x_b x_{ \eta}^\half}
\, .$$ Thus we are left with three equations of six variables, $ \tilde{x}_i^{-1} + \tilde{p}_i - 1=0$ for $i = 3,4,5$ where $p_1, p_2$ in $ \tilde{x}_i, \tilde{p}_i$ are replaced by the above conditions $$\{ \tilde{x}_i^{-1} + \tilde{p}_i - 1=0, \mbox{ for } i = 3,4,5. \} /. \{ p_1 \to 1 - x_{ \eta}, \ p_2 \to \frac{1- p_t^{ \frac{1}{2}} x_b x^\half_{ \eta}}{ p_t^{ \frac{1}{2} } x_b x^\half_{ \eta}} \} \, .$$ which are classical difference equations of $I_{ \varphi = S}$. After some algebraic manipulations, the equations can be written as $$\begin{aligned}
& x_b + x_b^{-1} = \frac{ x^\half_{ \eta} x_t^{-1} - x_{ \eta}^{-\half} x_t}{ x_t^{-1} - x_t} p_t^{- \frac{1}{2}} + \frac{ x^{-\half}_{ \eta} x_t^{-1} - x^\half_{ \eta} x_t}{ x_t^{-1} - x_t} p_t^{ \frac{1}{2}}
\nonumber \\
& \frac{x^\half_\eta x_b - x_{ \eta}^{-\half} x_b^{-1}}{x_b-x_b^{-1}} p_b^{ - \frac{1}{2}}
+\frac{x^{-\half}_\eta x_b - x^\half_{ \eta} x_b^{-1}}{x_b-x_b^{-1}} p_b^{ \frac{1}{2}} = x_t + x_t^{-1}
\nonumber \\
& p_{ \eta}= \frac{p_t^{-\half}- p_t^\half}{x_t^{-1} - x_t}= \frac{x_b -x_b^{-1}}{p_b^{-\half} -p_b^\half} \;.\end{aligned}$$ which are indeed the classical limit, $\hbar \to 0$ i.e. $q \to 1$, of the difference equations of $I_{ \varphi = S}$ given in eq. . These equations are the same as algebraic equations studied in [@Gaiotto:2013bwa] which define a moduli space of vacua for $T[SU(2)]$. In terms of shear operators , the classical difference equations are $$\begin{aligned}
\mathcal{L}_{\varphi=S} =\{ (\sqrt{\st}^{\rm T})_t - (\frac{1}{\sqrt{\st}})_b=0\;, \;(\sqrt{\st''}^{\rm T})_t - \sqrt{\st'}(1+ \st)_b=0\;, \; p_\eta = - \frac{i}{\sqrt{\st}_b}\}\;.\end{aligned}$$ For $\varphi=T$, classical difference equations are (see eq. ) $$\begin{aligned}
\mathcal{L}_{\varphi=T} =\{ x_t = x_b\;, \; p_t^{-\half } = p_b^{\half} x_b^{-1}\;, \; p_\eta =1 \}\;.\end{aligned}$$ In terms of shear coordinates, $$\begin{aligned}
\mathcal{L}_{\varphi=T} =\{ (\sqrt{\st}^{\rm T})_t- (\frac{1}{\sqrt{\st'}})_b=0\;, \; (\sqrt{\st''}^{\rm T})_t- \sqrt{\st''}(1+ \st')_b=0\;, \; p_\eta =1 \}\;.\end{aligned}$$ A general $\varphi\in SL(2,\mathbb{Z})$ can be written as a product of $S$ and $T$. Thus, to obtain the classical difference equation for general $\varphi$, we only need to know the gluing rules for the classical difference equations. For $\varphi=\varphi_2 \varphi_1$, the classical Lagrangian can be obtained by\
\
1. Identify $(\sqrt{\st}^{\rm T},\sqrt{\st'}^{\rm T},\sqrt{\st''}^{\rm T})_t$ of $\varphi_2$ with $(\sqrt{\st},\sqrt{\st'},\sqrt{\st''})_b$ of $\varphi_1$.\
2. If $p_\eta = p_{\eta,i} $ for $\varphi_{i=1,2}$, then $p_\eta = p_{\eta,2}p_{\eta,1}$ for $\varphi=\varphi_2\varphi_1$.\
3. Integrate out $(\sqrt{\st}^{\rm T},\sqrt{\st'}^{\rm T},\sqrt{\st''}^{\rm T})_t$ of $\varphi_2 $ (or equivalently $(\sqrt{\st},\sqrt{\st'},\sqrt{\st''})_b$ of $\varphi_1$).\
\
Using the gluing rules, the classical Lagrangian for $\varphi=S^2$ and $\varphi=(ST)^3$ become $$\begin{aligned}
&\mathcal{L}_{\varphi=S^2} = \{ (\sqrt{\st}^{\rm T})_t - (\sqrt{\st})_b =0 \;,\; (\sqrt{\st''}^{\rm T})_t - (\sqrt{\st''})_b =0\;, \;p_\eta = -1\; \} {\nonumber}\\
&
\mathcal{L}_{\varphi=(ST)^3} = \{ (\sqrt{\st}^{\rm T})_t - (\sqrt{\st})_b =0 \;,\; (\sqrt{\st''}^{\rm T})_t - (\sqrt{\st''})_b =0\;, \;p_\eta = \frac{-i}{\sqrt{\st_b \st'_b\st''_b }}=x_\eta^{-\half}\; \}\;.\end{aligned}$$ In the last equation, we used the fact $\sqrt{\st\st'\st''} = -i x^\half_\eta$ in the classical limit. These classical difference equations reflect the $SL(2,\mathbb{Z})$ structure for index $I_{\varphi}$.
Derivation of eq. {#derivation of SR indices}
==================
For $\varphi = \sL$, let us first define a convenient basis $\{ |(m,e)\rangle_{\Pi_{\sL}}\}$ where the polarization $\Pi_{\sL}$ is defined by $$\begin{aligned}
\Pi_{\sL} = \big{(}{\cal X}_{\sL} , {\cal P}_{\sL})_\pm = (-\sT''_\pm \pm i \pi +\log\fl_\pm \mp \frac{\hbar}2, \half (\sT_\pm +\sT''_\pm -\log \fl_\pm ) \big{)}\;.
\end{aligned}$$ $\Pi_{\sL}$ is related to the $\Pi_{\SR}$ in in the following way $$\begin{aligned}
\left(\begin{array}{c} {\cal X}_\sL \\ {\cal P}_\sL\end{array}\right)_\pm &= \left(\begin{array}{cc}2 & 0 \\ - \half & \half \end{array}\right)\left(\begin{array}{c} {\cal X}_{\SR} \\ {\cal P}_{\SR}\end{array}\right)_\pm \pm \left(\begin{array}{c} -\frac{\hbar}2\\ \frac{ i \pi }{2} \end{array}\right) \; \label{Pi(L) to Pi(RS)} \, .
\end{aligned}$$ In terms of ${\cal X}_{\sL \pm}, {\cal P}_{\sL \pm}$ (position, momentum) operators in $\Pi_{\sL}$, 솓 operator $\sL$ in can be written as $$\begin{aligned}
\sL = \left( \prod_{r=1}^\infty \frac{1-q^r e^{{\cal X}_{\sL+}} \fl^{-1}_+}{1-q^{r-1} e^{{\cal X}_{\sL-} }\fl^{-1}_-} \right) \exp \big{[} -\frac{1}{\hbar}\big{(}({\cal P}_{\sL+} + \half \log \fl_+ )^2-({\cal P}_{\sL-} +\half \log \fl_- )^2 \big{)}\big{]}\;.
\end{aligned}$$ Let us suppress the subscript $\sL$ of ${\cal X}_{\sL \pm}, {\cal P}_{\sL \pm}$ hereafter. For instance, eq. would be written as $$_\sL\langle m,e | e^{ {\cal X}_{ \pm}} =\; _\sL \langle m,e \mp 1 | q^{ \frac{m}{2}}, \qquad
_\sL\langle m,e | e^{ {\cal P}_{ \pm}} =\; _\sL \langle m \pm 1 ,e | q^{ \frac{e}{2}}
\, .$$ Using eq. , one can find a relation between $_\sL \langle(m,e)|$ and $ _{\SR}\langle (m,e)|$ such as [^28] $$\begin{aligned}
_{\SR}\langle m,e| &= _{\sL}\langle 2m,\half(e-m)| q^{\frac{1}4 (m-e)} (-1)^{- m} \, , \label{sr-l} \\
_\sL \langle m, e| &= _{\SR}\langle \half m, 2e+\half m| q^{\half e} (-1)^{\frac{m}{2}} \, .\end{aligned}$$ The second equation results in the following inner product of $ _\sL \langle m,e|$ basis $$_\sL \langle m_2, e_2| m_1,e_1 \rangle_\sL = q^{e_1} \delta_{m_1,m_2} \delta_{e_1,e_2} \;, \label{l-inner}$$ which follows from the inner product of SR basis given in eq. . Using eq. , The evaluation of $\sL$ operator in $_{\SR} \langle m,e |$ basis can be rewritten as follows $$\begin{aligned}
&_{\SR}\langle (m_2, e_2),(m_\eta, e_\eta) |\sL | (m_1,e_1), (m_\eta, 0)\rangle_{\SR}{\nonumber}\\
&= q^{\frac{1}4(m_2-e_2+m_1-e_1)} (-1)^{-m_2+m_1}\; _\sL \langle (2m_2, \frac{e_2-m_2}2),(m_\eta, e_\eta)|\sL| (2m_1, \frac{e_1-m_1}2),(m_\eta, 0)\rangle_\sL \label{l-sr}\end{aligned}$$ To evaluate the right hand side, we first note that the infinite product part of $ \sL$ operator acts on the bra as follows $$\begin{aligned}
&_\sL \langle (m_2,e_2),(m_\eta, e_\eta)|\prod_{r=1}^\infty \frac{1-q^r e^{{\cal X}_+} \fl^{-1}_+}{1-q^{r-1} e^{{\cal X}_- } \fl^{-1}_-} {\nonumber}\\
&= \oint \frac{du_2}{2\pi i u_2} \oint \frac{du_\eta}{2\pi i u_\eta} u^{-e_2} u_\eta^{-e_\eta} \;_\sL \langle (m_2,u_2),(m_\eta, u_\eta)|\prod_{r=1}^\infty \frac{1-q^r e^{{\cal X}_+}\fl^{-1}_+}{1-q^{r-1} e^{{\cal X}_- } \fl^{-1}_-}{\nonumber}\\
&=\sum_{e'} \;_\sL \langle (m_2,e_2+e') , (m_\eta, e_\eta -e') |\CI_\Delta (m_\eta - m_2, e')\;, \label{bra}
\end{aligned}$$ and that the exponential part of $\sL$ operator acts on the ket as follows $$\begin{aligned}
&\exp \big{[} -\frac{1}{\hbar}(({\cal P}_+ + \half \log \fl_+)^2-({\cal P}_-+ \half \log \fl_-)^2)\big{]}|(m_1,e_1),(m_\eta, e_\eta)\rangle_\sL {\nonumber}\\
&=(\;_\sL\langle (m_1,e_1),(m_\eta, e_\eta) | \exp \big{[} - \frac{1}{\hbar}(({\cal P}_-+ \half \log \fl_-)^2-({\cal P}_++ \half \log \fl_+)^2 )\big{]} )^\dagger {\nonumber}\\
&= |(m_1+2e_1 +m_\eta,e_1),(m_\eta, e_\eta-e_1- \half m_\eta)\rangle_\sL \, . \label{ket}
\end{aligned}$$ Here we used the fact that the adjoint of operators are given by $$( {\cal P}_{\pm, \sL} )^\dagger = {\cal P}_{\mp, \sL}, \quad (\fl_{ \pm})^\dagger =\fl_{\mp}\; ,$$ and that they can be written as ${\cal P}_{ \pm} = \pm \partial_m+ \frac{ \hbar}{2}e, \ln \fl_{ \pm} = \mp \partial_{e_{ \eta}} + \frac{ \hbar}{2}m_{ \eta} $. Using eq. , , and then , the evaluation of $\sL$ operator in eq. can be rewritten as $$\begin{aligned}
&_{\SR}\langle (m_2, e_2),(m_\eta, e_\eta) |\sL | (m_1,e_1), (m_\eta, 0)\rangle_{\SR} {\nonumber}\\
&=\delta_{\ldots} \delta_{\ldots} (-1)^{m_1-m_2} q^{\frac{1}4(e_1-e_2-m_1+m_2)} \CI_{\Delta} (m_{\eta} - 2 m_2, \half (e_1-e_2-m_1+m_2)) \; {\nonumber}\\
&
=\delta_{\ldots} \delta_{\ldots} (-1)^{e_2-e_1} q^{\frac{1}4(e_1-e_2-m_1+m_2)} \CI_{\Delta} (-e_1-m_1, \frac{1}{2}(e_1-e_2-m_1+m_2))
{\nonumber}\\
&= \delta_{ \ldots} \delta_{ \ldots} (-1)^{e_2-e_1} q^{\frac{1}4(e_1-e_2-m_1+m_2)} \CI_{\Delta} (\frac{-e_1+m_1+e_2-m_2}{2}, m_1+e_1)
{\nonumber}$$ where $ \delta_{ \ldots} \delta_{ \ldots}$ denotes the following combination of Kronecker delta functions, $$\begin{aligned}
\delta_{\ldots} \delta_{\ldots}
&= \delta (-e_1+2m_2-m_1-m_\eta)\delta(e_2-e_1+2e_{\eta} + m_2 - m_1 )\;.
\label{delta-abb}\end{aligned}$$ In the third line, we changed the first argument in the tetrahedron index and the power of $(-1)$ using Kronecker delta functions. In the last line, we used the identity of $\CI_{\Delta}$ in eq. . This completes the derivation of the first part of .
For $\varphi=\sR$, we basically repeat the previous derivation for $ \varphi = \sL$ using a polarization $\Pi_{\sR}$ instead of the polarization $ \Pi_{\sL}$. $\Pi_{\sR}$ is defined by $$\Pi_\sR =({\mathcal X}_\sR, {\mathcal P}_{\sR})_{ \pm} =( \sT'_{ \pm} \pm i \pi \pm \frac{ \hbar}{2} , \half ( \sT'_{ \pm}+\sT_{ \pm} ) ) \;,$$ thus it is related to $ \Pi_{\SR}$ as follows $$\begin{pmatrix}
{\mathcal X_{\sR}} \\
{\mathcal P_{\sR}}
\end{pmatrix}_{ \pm}
= \begin{pmatrix} 1&-1 \\ 1& 0 \end{pmatrix}
\cdot
\begin{pmatrix}
{\mathcal X}_{ \SR} \\
{\mathcal P}_{\SR}
\end{pmatrix}_{ \pm}
\pm
\begin{pmatrix} \frac{ \hbar}{2} + i \pi \\ 0 \end{pmatrix}$$ In terms of momentum and position operators in this polarization, $\sR$ operator in can be written as $$\sR = \left( \prod_{r=1}^{ \infty} \frac{1-q^r e^{{\mathcal X}_{\sR-}}}{1-q^{r-1} e^{{\mathcal X}_{\sR+}}} \right)
\exp [ \frac{1}{ \hbar} ( {\mathcal P}_{\sR+}^2-{\mathcal P}_{\sR-}^2 ) ]\;.$$ The basis change between SR- and $\sR$-basis can be obtained as $$\begin{aligned}
_{\SR} \langle m,e | = _\sR \langle m-e, m | (-1)^{m} q^{ \frac{m}{2}} \;,{\nonumber}\quad
_\sR \langle m, e | = _{\SR}\langle e,e-m | (-1)^{-e} q^{ - \frac{e}{2}} \;,{\nonumber}\end{aligned}$$ where the inner product of $\sR$-basis is $$\begin{aligned}
& _\sR \langle m_2, e_2 | m_1, e_1 \rangle_{\sR} =\delta_{m_1, m_2} \delta_{e_1,e_2} q^{-e_2} \;.{\nonumber}\end{aligned}$$ In the $_\sR \langle m,e|$ basis, the expectation value of $\sR$ can be evaluated similarly to eq. and , which results in $$_{\sR}\langle (m_2, e_2) | \sR | (m_1, e_1) \rangle_{\sR}
= \sum_{ e^{ \prime}}
\CI_{ \Delta} (-m_2, e^{ \prime}) _{\sR }\langle m_2, e_2 - e^{ \prime} | m_1 - 2 e_1, e_1 \rangle_\sR\;. {\nonumber}$$ To evaluate the charge shifts of the ket basis, we used the adjoint relation of ${\mathcal P_{\sR,\pm}}$ $$( {\mathcal P_{\sR, \pm }})^{ \dagger} = {\mathcal P_{\sR, \mp}} \,$$ which can be deduced from the adjoint relation of ${\mathcal P}_{\SR, \pm}$. Thus, we obtain $$\begin{aligned}
_{\SR}\langle m_2,e_2 | \sR | m_1, e_1 \rangle_{\SR}
&= (-1)^{m_2-m_1} q^{ \frac{m_1+m_2}{2}} _{\sR}\langle m_2-e_2 , m_2 | \sR |m_1 - e_1, m_1 \rangle_{\sR} {\nonumber}\\
&=(-1)^{m_2+e_1} q^{ \half (m_2+e_1)} \CI_{ \Delta} (-e_1 -m_2, m_1+e_1) \delta_{m_2+m_1-e_2+e_1,0} \,.
{\nonumber}\end{aligned}$$ In the second line, the Kronecker delta function was used to change the first argument of $\CI_{ \Delta}$, then the triality relation was used in turn. This gives a derivation for the second part in eq. .\
Basis change between SR and FN basis {#basis from FN to shear}
====================================
Recall that $\Pi_{\SR}= ({\cal S},{\cal R}) := (\frac{1}{2}(\sT+\sT'), \frac{1}2 (\sT-\sT'))$. Explicit expressions for $\ss=\exp(\mathcal{S}), \sr = \exp(\mathcal{R})$ in terms of Fenchel-Nielsen operators, $(\hat{\lambda}, \hat{\tau})$ are given by $$\begin{aligned}
&\ss_{\pm} =\exp (\CS_\pm) =q^{\pm 1/4} \frac{1}{(q^{\mp 1/4} \hat{ \lambda}^{-1}_{\pm} \hat{\tau}_\pm^{1/2}-q^{\pm 1/4} \hat{\tau}_\pm^{-1/2}\hat{\lambda}_\pm) }(\hat{\tau}_{\pm}^{1/2} - \hat{\tau}_{\pm}^{-1/2}) \;. \nonumber
\\
&\sr_{\pm} = \exp (\CR_\pm) = q^{\pm 1/4} \frac{1}{\hat{\lambda}_\pm -\hat{ \lambda}_\pm^{-1}} (\hat{\tau}^{-1/2}_\pm - \hat{\tau}^{1/2}_\pm)\frac{1}{\hat{\lambda}_\pm - \hat{\lambda}_\pm^{-1}} (q^{\mp 1/4} \hat{\lambda}^{-1}_{\pm} \hat{\tau}_\pm^{1/2}-q^{\pm 1/4} \hat{\tau}_\pm^{-1/2} \hat{\lambda}_\pm) \;.
\label{s,r variable }\end{aligned}$$ We will express SR fugacity basis $\langle m,u|:=_{\widetilde{\SR}}\langle m,u|$ in terms of FN fugacity basis $\langle \tilde{m},\tilde{u}|:=_{\FN}\langle \tilde{m},\tilde{u}|$, $$\begin{aligned}
&\langle m,u| = \oint \frac{d\tilde{u}}{2\pi i \tilde{u}}\Delta(\tilde{m},\tilde{u},q) \langle m,u|\tilde{m},\tilde{u}\rangle \langle \tilde{m},\tilde{u}|\;.\end{aligned}$$ By imposing the following conditions, we obtain difference equations for the basis change coefficients $\langle m,u|\tilde{m},\tilde{u}\rangle$. $$\begin{aligned}
\langle m,u|\ss_\pm |\tilde{m},\tilde{u}\rangle &= x_\pm \cdot \langle m,u| \tilde{m},\tilde{u}\rangle {\nonumber}\\
&=\langle m,u| \ss_\pm (\hat{\lambda}_\pm, \hat{\tau}_\pm )|\tilde{m},\tilde{u}\rangle = s_\pm^T(\tilde{x}_\pm ,\tilde{p}_\pm)\cdot \langle m,u|\tilde{m},\tilde{u}\rangle {\nonumber}\\
\langle m,u|\sr_\pm |\tilde{m},\tilde{u}\rangle &= p_\pm \cdot\langle m,u| \tilde{m},\tilde{u}\rangle {\nonumber}\\
&=\langle m,u| \sr_\pm (\hat{\lambda}_\pm, \hat{\tau}_\pm )|\tilde{m},\tilde{u}\rangle = r_\pm^T(\tilde{x}_\pm, \tilde{p}_\pm )\cdot \langle m,u|\tilde{m},\tilde{u}\rangle \;. \label{difference equation for basis change matrix}\end{aligned}$$ Here $\sO^{\rm T}$ denote the transpose of $\sO$, $\sO_\pm^{\rm T} := (\sO^\dagger_\pm)^*$. For SR operators, their transpose operators are $$\begin{aligned}
(\ss_\pm, \sr_\pm)^{\rm T} = (\ss_\pm , \sr_\pm)/.\{ q\rightarrow q^{-1} , \tau\rightarrow \tau^{-1}\} \;.\end{aligned}$$ The operators $(x,p)_\pm,(\tilde{x},\tilde{p})_\pm$ are given $$\begin{aligned}
x_\pm = q^{\frac{m}2} u^{\pm 1}\;,\quad p_\pm = e^{\pm \partial_{m} + \frac{1}2\hbar u\partial_{u}}\;, \quad \tilde{x}_\pm = q^{\frac{\tilde{m}}2} \tilde{u}^{\pm 1}\;,\quad \tilde{p}_\pm = e^{\pm \partial_{\tilde{m}} + \frac{1}2\hbar \tilde{u}\partial_{\tilde{u}}} \;.\end{aligned}$$ To make the action of FN, SR operators simple, we introduce new variables $(a,b),(s,t)$ defined as follow $$\begin{aligned}
q^a := q^{\frac{\tilde{m}}2}\tilde{ u} ,\quad q^b = q^{\frac{\tilde{m}}2}\tilde{u}^{-1}, \quad q^s := q^{\frac{m}2} u ,\quad q^t = q^{\frac{m}2}u^{-1}\;.\end{aligned}$$ Let’s denote $\langle m,u|\tilde{m},\tilde{u}\rangle$ in terms of these variables as $C(a,b|s,t)$. $$\begin{aligned}
C(a,b|s,t):=\langle m, u|\tilde{m},\tilde{u}\rangle /.\{\tilde{m}\rightarrow a+b, m\rightarrow s+t, \tilde{u}\rightarrow q^{\frac{a-b}2}, u \rightarrow q^{\frac{s-t}2}\} \;.\end{aligned}$$ One advantage of these variables is that $+$ ($-$) type operators act only on $a,s$ ($b,t$) variables. Thus the difference equations for $C(s,t|a,b)$ factorizes into $\pm$ parts and we can set $$\begin{aligned}
C (s,t|a,b) = C_+ (s|a) C_- (t|b) \;. \label{basis change coefficient}\end{aligned}$$ Then, the difference equations for $C_+$ are $$\begin{aligned}
&(1-q^{-a+s}) C_+(s|a) = (1-q^{s+a+1}) C_+(s|a+1) \;, {\nonumber}\\
&(q^a - q^{-a}) C_+(s+1|a) {\nonumber}\\
&=\frac{q^{-(a+1/2)} C_+(s|a)-q^{a+1/2} C_+(s|a+1)}{q^{a+1/2}-q^{-(a+1/2)}}- \frac{q^{-(a-1/2)} C_+(s|a-1)-q^{a-1/2} C_+(s|a)}{q^{a-1/2}-q^{-(a-1/2)}} \;. {\nonumber}\end{aligned}$$ For $C_-$, the difference equations are $$\begin{aligned}
&(1-q^{b+t}) C_-(t|b) = (1-q^{t-b-1}) C_-(t|b+1) \;, \nonumber
\\
&(q^b -q^{-b})C_-(t-1|b) \nonumber
\\
&=\frac{q^{-(b-1/2)} C_-(t|b)-q^{b-1/2} C_-(t|b-1)}{q^{b-1/2}-q^{-(b-1/2)}}- \frac{q^{-(b+1/2)} C_-(t|b+1)-q^{b+1/2} C_-(t|b)}{q^{b+1/2}-q^{-(b+1/2)}} \;.\end{aligned}$$ We use the fact that $$\begin{aligned}
(x_+,x_-, p_+, p_-) = (q^s, q^t, e^{\partial_s}, e^{-\partial_t})\;, \quad (\tilde{x}_+,\tilde{x}_-, \tilde{p}_+, \tilde{p}_-) = (q^a, q^b, e^{\partial_a}, e^{-\partial_b})\;.\end{aligned}$$ Solving the two difference equations, we find the following solutions $$\begin{aligned}
&C_+ (s|a) =(-1)^{a} q^{-a(a+1)/2-as}\prod_{r=0}^\infty \frac{1-q^{r+1}q^{-a+s}}{1-q^{r}q^{-a-s}}\;, {\nonumber}\\
&C_- (t|b) = (-1)^b q^{b(b-1)/2+bt}\prod_{r=0}^\infty \frac{1-q^{r+1}q^{-b-t}}{1-q^{r}q^{-b+t}} \;. {\nonumber}\end{aligned}$$ Therefore, $$\begin{aligned}
&C(s,t|a,b)= (-1)^{a+b}q^{-a(a+1)/2-as+b(b-1)/2+bt}\prod_{r=0}^\infty \frac{1-q^{r+1}q^{-a+s}}{1-q^{r}q^{-a-s}}\prod_{r=0}^\infty \frac{1-q^{r+1}q^{-b-t}}{1-q^{r}q^{-b+t}}\;. \label{sol for C(s,t|a,b)}\end{aligned}$$ In the original fugacity variables$(m,u), (\tilde{m},\tilde{u})$, the basis change matrix is given by $$\begin{aligned}
&\langle m,u| \tilde{m},\tilde{u}\rangle = C(s,t|a,b)|_{a\rightarrow \frac{\tilde{m}}2 +\log_q \tilde{u},b\rightarrow \frac{\tilde{m}}2 -\log_q \tilde{u},s \rightarrow \frac{m}2 +\log_q u, t\rightarrow \frac{m}2 +\log_q u} {\nonumber}\\
&= (- \tilde{u}q^{1/2})^{-\tilde{m}} \frac{\CI_{\Delta}(\tilde{m}-m, \tilde{u}/u )}{\CI_{\Delta}(\tilde{m}+m, u \tilde{u}q^{-1})} u^{-\tilde{m}} \tilde{u}^{-m} \;, {\nonumber}\\
&= (-q^{\half} u)^m \CI_{\Delta} (-m-\tilde{m}, u^{-1}\tilde{u}^{-1})\CI_{\Delta}(\tilde{m}-m, \tilde{u}/u)\;.
\label{eq:D}\end{aligned}$$ In the SR charge basis, we have $$\begin{aligned}
\langle m, e| \tilde{m}, \tilde{u}\rangle = \sum_{e_1 \in \mathbb{Z}} (- q^{\half})^m \tilde{u}^{-2e_1 -e+m} \CI_{\Delta}(-m-\tilde{m} ,e_1)\CI_{\Delta}(-m+\tilde{m}, -e+m-e_1) \;. \label{eq:D-charge basis}\end{aligned}$$ This basis change matrix element can be thought of as $SL(2,\mathbb{C})$ CS partition function on a mapping cylinder $\Sigma_{1,1}\times_{\varphi} I$ with $\varphi= \textrm{identity}$ in the polarization where positions are $(\ss_{\textrm{bot}}, \lambda_{\rm top}, \fl)$ and momenta are $(\sr_{\rm bot}, \tau_{\rm top}, \sm)$. Identifying FN operators $(\lambda, \tau)$ as UV operators and SR operator $(\ss, \sr)$ as IR operators, the mapping cylinder is called a RG manifold in [@Dimofte:2013lba].
Note that the SR basis is Weyl-reflection invariant and thus the states are in $\mathcal{H}_{SL(2,\mathbb{C})} \subset \widetilde{\CH}_{SL(2,\mathbb{C})}$. For the charge basis $| m,e\rangle$ to be an (non-zero) element in $\mathcal{H}_{SL(2,\mathbb{C})}$, we need to impose following conditions $$\begin{aligned}
&\langle m,e|\tilde{m},\tilde{u}\rangle \neq 0\;, \quad \textrm{for some $\tilde{m} \in \mathbb{Z}/2$}\;.\end{aligned}$$ This condition implies that $$\begin{aligned}
m,e\in \mathbb{Z}/2, \quad m+e \in \mathbb{Z}\;. \label{Range for SR charge}\end{aligned}$$ Furthermore, we claim that SR charge basis are complete basis for $\mathcal{H}_{SL(2,\mathbb{C})}$. How can we prove the completeness? One simple answer uses the fact that a operator $\ss_+\ss_-$ is self-adjoint. Since $\langle m,e|$ are eigenstates for the operator with eigenvalues $q^{m}$, they form a complete basis. Using the property $(\ss_\pm, \sr_\pm)^\dagger = (\ss_\mp, \sr_\mp)$ and $$\begin{aligned}
\langle m,e|\ss_+ = \langle m,e\mp 1| q^{\frac{m}2}\;, \quad \langle m,e|\sr_\pm = \langle m\pm 1, e| q^{\frac{e}2}\;,\end{aligned}$$ One can see that $$\begin{aligned}
\langle m,e|m',e'\rangle = \kappa \delta_{m,m'}\delta_{e,e'}\;.\end{aligned}$$ Here $\kappa$ is $(m,e)$-independent constant and it is 1 in . From this orthonormality and the fact the basis $\langle m,e|$ are complete basis in $\mathcal{H}_{SL(2,\mathbb{C})}$, one obtain following completeness relation $$\begin{aligned}
\mathds{1}_{\mathcal{H}_{SL(2,\mathbb{C})}} = \sum_{(m,e)}| m,e\rangle \langle m,e| \;.\end{aligned}$$ More directly, the completeness relation is equivalent to the following identity, $$\begin{aligned}
&\sum_{(m,e)}\langle \tilde{m}_b, \tilde{u}_b| m,e\rangle \langle m,e| \tilde{m}_t, \tilde{m}_t\rangle = \delta(\tilde{m}_b - \tilde{m}_t) \frac{\delta(\tilde{u}_b - \tilde{u}_t)}{\Delta(\tilde{m}_t,\tilde{u}_t)}\;.{\nonumber}\end{aligned}$$ using the explicit expression in . While we do not have an exact proof, we have confirmed it by series expansion in $q$. More precisely, we checked that $$\begin{aligned}
&\sum_{\tilde{m}\in \mathbb{Z}/2}\oint \frac{d\tilde{u}_t}{2\pi i \tilde{u}_t}\Delta(\tilde{m}_t, \tilde{m}_t)\big{(}\sum_{(m,e)}\langle \tilde{m}_b, \tilde{u}_b| m,e\rangle \langle m,e| \tilde{m}_t, \tilde{m}_t\rangle \big{)} f(\tilde{m}_t,\tilde{u}_t ) = f(\tilde{m}_b, \tilde{u}_b)\;, {\nonumber}\end{aligned}$$ by expansion in $q$ for various Weyl-reflection invariant trial function $f(m,u)$, i.e., $f(m,u)= f(-m, u^{-1})$.
[10]{}
L. Alday, D. Gaiotto, Y. Tachikawa “Liouville Correlation Functions from Four-dimensional Gauge Theories,” Lett.Math.Phys.[**91**]{} (2010) 167. arXiv:0906.3219 \[hep-th\]
D. Gaiotto “N=2 dualities,” arXiv:0904.2715 \[hep-th\].
V. Pestun “Localization of gauge theory on a four-sphere and supersymmetric Wilson loops,” Commun.Math.Phys. 313 (2012) 71-129 arXiv:0712.2824 \[hep-th\].
See, for example, the following review and references therein.
Yu Nakayama “Liouville Field Theory – A decade after the revolution, ” Int.J.Mod.Phys.A19: (2004) 2771, arXiv:hep-th/0402009 \[hep-th\].
C. Cordova and D. L. Jafferis, arXiv:1305.2891 \[hep-th\]. T. Dimofte, D. Gaiotto and S. Gukov, “Gauge Theories Labelled by Three-Manifolds,” arXiv:1108.4389 \[hep-th\]. T. Dimofte, D. Gaiotto, and S. Gukov, “[3-Manifolds and 3d Indices]{},” arXiv:1112.5179 \[hep-th\]. N. Hama, K. Hosomichi and S. Lee, “SUSY Gauge Theories on Squashed Three-Spheres,” JHEP [**1105**]{}, 014 (2011) \[arXiv:1102.4716 \[hep-th\]\]. Y. Terashima and M. Yamazaki, “$SL(2,R)$ Chern-Simons, Liouville, and Gauge Theory on Duality Walls,” JHEP [**1108**]{} (2011) 135 \[arXiv:1103.5748 \[hep-th\]\].
J. Yagi, “3d TQFT from 6d SCFT,” arXiv:1305.0291 \[hep-th\]. D. Gaiotto and E. Witten, “S-Duality of Boundary Conditions In N=4 Super Yang-Mills Theory,” Adv. Theor. Math. Phys. [**13**]{} (2009) \[arXiv:0807.3720 \[hep-th\]\]. N. Drukker, D. Gaiotto and J. Gomis, “The Virtue of Defects in 4D Gauge Theories and 2D CFTs,” JHEP [**1106**]{}, 025 (2011) \[arXiv:1003.1112 \[hep-th\]\]. K. Hosomichi, S. Lee and J. Park, “AGT on the S-duality Wall,” JHEP [**1012**]{}, 079 (2010) \[arXiv:1009.0340 \[hep-th\]\].
Y. Terashima and M. Yamazaki, “Semiclassical Analysis of the 3D/3D Relation,” arXiv:1106.3066 \[hep-th\]. T. Dimofte and S. Gukov, “Chern-Simons Theory and S-duality,” arXiv:1106.4550 \[hep-th\].
J. Teschner and G. S. Vartanov, “6j symbols for the modular double, quantum hyperbolic geometry, and supersymmetric gauge theories,” arXiv:1202.4698 \[math-ph\]. D. Gang, E. Koh and K. Lee, “Superconformal Index with Duality Domain Wall,” arXiv:1205.0069 \[hep-th\]. K. Nagao, Y. Terashima, and M. Yamazaki, “Hyperbolic geometry and cluster algebra,” preprint (2011), arXiv:1112.3106 \[math.GT\].
Rinat M. Kashaev, Feng Luo, and Grigory Vartanov, “A TQFT of turaev-viro type on shaped triangulations,” preprint (2012), arXiv:1210.8393 \[math.QA\].
K. Hikami and R. Inoue, “Cluster algebra and complex volume of once-punctured torus bundles and two-bridge knots,” preprint (2012), arXiv:1212.6042 \[math.GT\].
Y. Terashima and M. Yamazaki, “3d N=2 Theories from Cluster Algebras,” arXiv:1301.5902 \[hep-th\].
K. Hikami and R. Inoue, “Braids, Complex Volume, and Cluster Algebra,” arXiv:1304.4776 \[math.GT\]
T. Dimofte, “Quantum Riemann Surfaces in Chern-Simons Theory,” arXiv:1102.4847 \[hep-th\]. F. Guéritaud and D. Futer, “On canonical triangulations of once-punctured torus bundles and two-bridge link complements,” Geom. Topol. 10, 1239 (2006).
T. Dimofte, M. Gabella and A. B. Goncharov, “K-Decompositions and 3d Gauge Theories,” arXiv:1301.0192 \[hep-th\].
Stavros Garoufalidis, Craig D. Hodgson, Hyam Rubinstein, and Henry Segerman, “1-efficient triangulations and the index of a cusped hyperbolic 3-manifold,” arXiv:1303.5278 \[math.GT\].
E. Witten, “SL(2,Z) action on three-dimensional conformal field theories with Abelian symmetry,” In \*Shifman, M. (ed.) et al.: From fields to strings, vol. 2\* 1173-1200 \[hep-th/0307041\]. S. Elitzur, G. W. Moore, A. Schwimmer and N. Seiberg, “Remarks on the canonical quantization of the Chern-Simons-Witten theory,” Nucl. Phys. B 326 (1989) 108.
Atiyah, Michael. “Topological quantum field theories." Publications Mathmatiques de l’IHS 68.1 (1988): 175-186.
E. Witten, “Quantization of Chern-Simons Gauge Theory with Complex Gauge Group,” Commun. Math. Phys 137 (1991) 29–66.
V. V. Fock, “Description of moduli space of projective structures via fat graphs,” hep-th/9312193.
V. V. Fock, “Dual Teichmuller space,” dg-ga/9702018v3.
S. Gukov, “Three-dimensional quantum gravity, Chern-Simons theory, and the A polynomial,” Commun. Math. Phys. [**255**]{}, 577 (2005) \[hep-th/0306165\].
S. Kim, “The Complete superconformal index for N=6 Chern-Simons theory,” Nucl. Phys. B [**821**]{}, 241 (2009) \[arXiv:0903.4172 \[hep-th\]\]. Y. Imamura and S. Yokoyama, “Index for three dimensional superconformal field theories with general R-charge assignments,” JHEP [**1104**]{}, 007 (2011) \[arXiv:1101.0557 \[hep-th\]\]. C. Krattenthaler, V. P. Spiridonov and G. S. Vartanov, “Superconformal indices of three-dimensional theories related by mirror symmetry,” JHEP [**1106**]{}, 008 (2011) \[arXiv:1103.4075 \[hep-th\]\]. A. Kapustin and B. Willett, “Generalized Superconformal Index for Three Dimensional Field Theories,” arXiv:1106.2484 \[hep-th\].
D. Tong, “Dynamics of N=2 supersymmetric Chern-Simons theories,” JHEP [**0007**]{}, 019 (2000) \[hep-th/0005186\].
Y. Imamura and D. Yokoyama, “$S^3/Z_n$ partition function and dualities,” JHEP [**1211**]{}, 122 (2012) \[arXiv:1208.1404 \[hep-th\]\]. O. J. Ganor, Y. P. Hong and H. S. Tan, “Ground States of S-duality Twisted N=4 Super Yang-Mills Theory,” JHEP [**1103**]{}, 099 (2011) \[arXiv:1007.3749 \[hep-th\]\]. O. J. Ganor, Y. P. Hong, R. Markov and H. S. Tan, “Static Charges in the Low-Energy Theory of the S-Duality Twist,” JHEP [**1204**]{}, 041 (2012) \[arXiv:1201.2679 \[hep-th\]\].
Stavros Garoufalidis, “The 3D index of an ideal triangulation and angle structures,” arXiv:1208.1663 \[math.GT\].
E. Witten, “Analytic Continuation Of Chern-Simons Theory,” arXiv:1001.2933 \[hep-th\]. N. Nekrasov and E. Witten, “The Omega Deformation, Branes, Integrability, and Liouville Theory,” JHEP [**1009**]{}, 092 (2010) \[arXiv:1002.0888 \[hep-th\]\]. G. Vartanov and J. Teschner, “Supersymmetric gauge theories, quantization of moduli spaces of flat connections, and conformal field theory,” arXiv:1302.3778 \[hep-th\]. L. Chekhov and V. V. Fock, “Quantum Teichmuller Space,” Theor. Math. Phys. [**120**]{} (1999) 1245 \[Teor. Mat. Fiz. [**120**]{} (1999) 511\] \[math/9908165 \[math-qa\]\].
R. M. Kashaev, “Quantization of Teichmüller spaces and the quantum dilogarithm,” Lett. Math.Phys. 43 (1998), no. 2 105–115.
J. Teschner, “An Analog of a Modular Functor from Quantized Teichmüller Theory,” in Handbook of Teichmuller Theory, Vol. I (2007) 685–760, \[math/0510174v4\].
J. Teschner, “On the relation between quantum Liouville theory and the quantized Teichmuller spaces,” Int. J. Mod. Phys. A [**19S2**]{}, 459 (2004) \[hep-th/0303149\]. H. Verlinde, “Conformal field theory, two-dimensional quantum gravity and quantization of Teichmüller space,” Nucl. Phys. B 337 (1990) 652.
A. Gadde, E. Pomoni, L. Rastelli and S. S. Razamat, “S-duality and 2d Topological QFT,” JHEP [**1003**]{}, 032 (2010) \[arXiv:0910.2225 \[hep-th\]\]. A. Gadde, L. Rastelli, S. S. Razamat and W. Yan, “The 4d Superconformal Index from q-deformed 2d Yang-Mills,” Phys. Rev. Lett. [**106**]{}, 241602 (2011) \[arXiv:1104.3850 \[hep-th\]\]. D. Gang, E. Koh, K. Lee and , “Line Operator Index on $S^{1}\times S^{3}$,” JHEP [**1205**]{}, 007 (2012) \[arXiv:1201.5539 \[hep-th\]\]. M. Aganagic, A. Klemm, M. Marino, C. Vafa and , “Matrix model as a mirror of Chern-Simons theory,” JHEP [**0402**]{}, 010 (2004) \[hep-th/0211098\].
D. Gaiotto, L. Rastelli and S. S. Razamat, “Bootstrapping the superconformal index with surface defects,” arXiv:1207.3577 \[hep-th\]. L. F. Alday, M. Bullimore, M. Fluder and L. Hollands, “Surface defects, the superconformal index and q-deformed Yang-Mills,” arXiv:1303.4460 \[hep-th\].
C. Beem, T. Dimofte and S. Pasquetti, “Holomorphic Blocks in Three Dimensions,” arXiv:1211.1986 \[hep-th\]. C. Hwang, H. -C. Kim and J. Park, “Factorization of the 3d superconformal index,” arXiv:1211.6023 \[hep-th\].
R.M. Kashaev, In: S. Pakuliak and G. von Gehlen (eds.), Integrable structures of exactly solvable two-dimensional models of quantum field theory (Kiev, 2000), 211221, NATO Sci. Ser. II Math. Phys. Chem., 35, Kluwer Acad. Publ., Dordrecht, 2001.
L. D. Faddeev, “Discrete Heisenberg-Weyl Group and Modular Group,” Lett. Math. Phys. [**34**]{} (1995) 249 \[hep-th/9504111\].
D. Gaiotto and P. Koroteev, “On Three Dimensional Quiver Gauge Theories and Integrability,” arXiv:1304.0779 \[hep-th\].
T. Dimofte, D. Gaiotto and R. van der Veen, “RG Domain Walls and Hybrid Triangulations,” arXiv:1304.6721 \[hep-th\].
[^1]: $S^2 \times_q S^1$ denote $S^1$ bundle of two-sphere twisted with holonomy for a combination of $U(1)$ R-symmetry and space-time rotation symmetry. $S^3_b$ denote a squashed three-sphere (ellipsoid) [@Hama:2011ea].
[^2]: It is $SL(2,\mathbb{R})$ CS theory in sense that the boundary Hibert-space looks like a quantization of $SL(2,\mathbb{R})$ flat connections on the boundary. In a recent paper [@Cordova:2013cea], the 3d-3d relation is derived from the first principal and find that the $S^3_b$ partition function corresponds to $SL(2,\mathbb{C})$ CS theory with level $k=1$. We expect there’s an isomorphism between a Hilbert-space obtained by quantizing $SL(2,\mathbb{R})$ flat connections and one from $SL(2,\mathbb{C})$ with $k=1$.
[^3]: Or, simply we express “CS theory on $\Sigma_{1,1}$” ignoring manifestly existing time-coordinate
[^4]: When $M$ has boundary, the 3d theory $T_M$ also depends on the choice of polarization $\Pi$ for the boundary phase space $\CM_{SL(2)} (\partial M)$, the space of $SL(2)$ flat connections on $\partial M$. In a strict sense, the 3d theory should be labelled by $T_{M,\Pi}$.
[^5]: It can be generalized to (2,0) theory of general A,D,E type and the corresponding 3d theory $T[M,\mathbf{g}]$ is labelled by 3-manifold $M$ and Lie algebra $\mathbf{g}$ of gauge group [@Dimofte:2013iv].
[^6]: Our convention for the $SL(2,\IZ)$ generators is the same as in [@Dimofte:2011ju; @Dimofte:2011py] but is the opposite from [@Dimofte:2011jd; @Gueritaud; @Garb:2013].
[^7]: It generalizes Witten’s $SL(2,\mathbb{Z})$ action [@Witten:2003ya].
[^8]: Since we are mainly focusing on the case $\Sigma = \Sigma_{1,1}$ throughout the paper, we simply denote $\CM_{G} (\Sigma)$ (phase space associated Riemann surface $\Sigma$ and gauge group $G$) by $\CM_G $ when $\Sigma = \Sigma_{1,1}$. We also sometimes omit the subscript $G$ when it is obvious in the context.
[^9]: $(W,H,D)$ here are the same as $(-x, -y, -z)$ in [@Dimofte:2011jd]. See eq.(2.13) and Figure 3 of [@Dimofte:2011jd].
[^10]: Under shifts by $2\pi i$, $A, B$ remains invariant up to a sign. Thus for $G=PSL(2,\mathbb{C}) = SL(2,\mathbb{C})/\langle \pm 1 \rangle$, the periodicity for each of $(T,T',T'')$ is $2\pi i$.
[^11]: The eigenvalues for the longitudinal holonomy is given by $(\ell , \ell^{-1}) =(e^V,
e^{-V}). $ On the other hand, for meridian holonomy, the eigenvalues are $(m^{\half} ,m^{-\half})=(e^{\frac{U}2}, e^{-\frac{U} 2}) $.
[^12]: But there is a subtle difference between the two $V$ variables. The $V$ in $\CM^{\rm knot}$ is periodic with period $ 2 \pi i $, while the $V$ in $\CM(\Sigma_{1,1})$ has period $4\pi i $. This discrepancy may be due to an additional $\mathbb{Z}_2$ quotient on $V$ during gluing the two $\Sigma_{1,1}$’s. The $\mathbb{Z}_2$ action could be identified by carefully analyzing how flat connection moduli space changes during the gluing procedure.
[^13]: Throughout this paper, the Cantor integral $\oint \frac{du}{2\pi i u} I(u)$ will be interpreted as picking up the coefficient of $u^{0}$ by regarding $I(u)$ as an element in a ring $\mathbb{Z}[u^{1/p},u^{-1/p}]$ with a positive integer $p$.
[^14]: However, general $R$ charge assignments can be easily incorporated.
[^15]: Due to the second property in , we need to specify decomposition of $L, R$ in terms of $S,T$ for the index computation. If not, the index is defined only up to an overall factor $u_\eta^{\half m_\eta \mathbb{Z}}$. Throughout this paper we use the simplest decomposition, $R=T$ (instead of $T(ST)^{3n}$ with $n \neq 0$) and $L = S^{-1}T^{-1}S$, in the index computation. In this choice, $I^{T[SU(2)]}_{\textrm{tori}(\varphi)}$ and $I^{\Delta}_{\textrm{tori}(\varphi)}$ are the same without any polarization change as we will see in section \[Z(Tr)=Z(T\[SU(2)\])\].
[^16]: In [@Dimofte:2011py], the $SL(2,\mathbb{Z})$ polarization change in CS theory on a tetrahedron $\Delta$ is identified with Witten’s $SL(2,\mathbb{Z})$ action on the tetrahedron theory $T_\Delta$. Witten’s $SL(2,\mathbb{Z})$ action can be extended to $SL(2,\mathbb{Q})$ by including charge rescaling of $U(1)$ global symmetry.
[^17]: There is no guarantee that for given polarization $\Pi$ there exist a basis satisfying these conditions.
[^18]: The normalization of meridian variable in [@Witten:2010cx] is different from ours, $m_{ours} = \sm^2_{theirs}$. In the reference, the $\mathbb{Z}_2$ symmetry is shown for knot complements in $S^3$. We expect that the $\mathbb{Z}_2$ symmetry also exists for our mapping torus case. One evidence is that the A-polynomial analyzed in section \[A-polynomial for mapping torus\] is always polynomial in $m$ instead of $m^{1/2}$.
[^19]: As defined in section \[sec : introduction\], $T_\Sigma$ denotes a 4d theory of class S obtained from $A_1$ type of $(2,0)$ theory on a Riemann surface $\Sigma$.
[^20]: $|0\rangle$ can be viewed as a state in $\CH_{SL(2,\mathbb{C})} \otimes \CH^{\textrm{knot}}_{SL(2,\mathbb{C})}$ with $m_\eta =0$ by regarding $\Pi(u_\eta)$ as $\langle m_\eta=0, u_\eta |0\rangle$.
[^21]: In [@Terashima:2011xe], they consider the case $G= SL(2,\mathbb{R})$ instead of $SL(2,\mathbb{C})$. But the A-polynomial computation in section \[A-polynomial for mapping torus\] does not depends on weather $G=SL(2,\mathbb{C})$ or $SL(2,\mathbb{R})$.
[^22]: The factor $(-1)^{e_{i+1}-e_i}$ in $I^{\Pi_{\SR}}_{\varphi_i}$ is ignored in this expression since $\prod_{i=1}^N (-1)^{e_{i+1}-e_i}$ =1 and thus the sign factors do not appear in the final expression for $Z^{\Tr (\varphi)}_{\textrm{tori}(\varphi)}$.
[^23]: More precisely, $I_\varphi (m_b, u_b, m_t, u_t) = \;^S _{\FN}\langle m_b, u_b |\varphi |m_t, u_t\rangle^S_{\FN}$ where $| m,u \rangle^S_{\FN}$ is a Weyl-reflection invariant combination of FN basis . However, it does not matter since operator $\varphi$ is Weyl-reflection invariant, $_{\FN}\langle m_b, u_b|\varphi|m_t, u_t\rangle_{\FN} =\;^S _{\FN}\langle m_b,u_b|\varphi|m_t, u_t\rangle^S_{\FN} $, and we will not distinguish them.
[^24]: In , loop operators act on $\CH_{SL(2,\mathbb{C})}$. On the other hand, loop operators here are difference operators acting on a function $I(m_b, u_b, m_t, u_t; m_\eta, u_\eta)$.
[^25]: To compute $_{\FN}\langle m_b, u_b | m_2, e_2 \rangle_{\SR}$, we need to take the complex conjugation on the expression. In taking the conjugation, we regard $(-1)$ as $e^{ i \pi}$. Thus, $$_{\FN}\langle m_b, u_b | m_2, e_2 \rangle_{\SR}
=\sum_{e \in \mathbb{Z}} (-1)^{-m_2}q^{ \half m_2} u_b^{2e+e_2 -m_2} \CI_{ \Delta} (-m_2 -m_b, e) \CI_{ \Delta} (-m_2+m_b, -e_2+m_2 -e) \, .
{\nonumber}$$
[^26]: The logarithmic operators $({\cal S}, {\cal R})$ are rescaled by a factor $(2\pi b)$ from those in previous sections.
[^27]: We flipped the sign of the ‘momentum’ $\CR$ to match the convention of [@Dimofte:2011ju; @Dimofte:2011py].
[^28]: The $(-1)$ in eq. should be regarded as $(e^{ i \pi})$, thus the kets are related as follows, $$| m, e \rangle_{\SR} = q^{ \frac{1}{4}(m-e) }(-1)^m | m, e \rangle_{\sL}
\, .$$
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- 'N. Falstad'
- 'F. Hallqvist'
- 'S. Aalto'
- 'S. K[ö]{}nig'
- 'S. Muller'
- 'R. Aladro'
- 'F. Combes'
- 'A. S. Evans'
- 'G. A. Fuller'
- 'J. S. Gallagher'
- 'S. Garc[í]{}a-Burillo'
- 'E. Gonz[á]{}lez-Alfonso'
- 'T. R. Greve'
- 'C. Henkel'
- 'M. Imanishi'
- 'T. Izumi'
- 'J. G. Mangum'
- 'S. Mart[í]{}n'
- 'G. C. Privon'
- 'K. Sakamoto'
- 'S. Veilleux'
- 'P. P. van der Werf'
bibliography:
- 'ref.bib'
title: 'Hidden or missing outflows in highly obscured galaxy nuclei?[^1]'
---
[Understanding the nuclear growth and feedback processes in galaxies requires investigating their often obscured central regions. One way to do this is to use (sub)millimeter line emission from vibrationally excited HCN (HCN-vib), which is thought to trace warm and highly enshrouded galaxy nuclei. It has been suggested that the most intense HCN-vib emission from a galaxy is connected to a phase of nuclear growth that occurs before the nuclear feedback processes have been fully developed.]{} [We aim to investigate if there is a connection between the presence of strong HCN-vib emission and the development of feedback in (ultra) luminous infrared galaxies ((U)LIRGs).]{} [We collected literature and archival data in order to compare the luminosities of rotational lines of HCN-vib, normalized to the total infrared luminosity, to the median velocities of $119$ $\mu$m OH absorption lines, potentially indicating outflows, in a total of 17 (U)LIRGs.]{} [The most HCN-vib luminous systems all lack signatures of significant molecular outflows in the far-infrared OH absorption lines. However, at least some of the systems with bright HCN-vib emission do have fast and collimated outflows that can be seen in spectral lines at longer wavelengths, including in millimeter emission lines of CO and HCN (in its vibrational ground state) as well as in radio absorption lines of OH.]{} [We conclude that the galaxy nuclei with the highest $L_{\mathrm{HCN-vib}}/L_{\mathrm{IR}}$ do not drive wide-angle outflows detectable using the median velocities of far-infrared OH absorption lines. It is possible that this is due to an orientation effect where sources which are oriented in such a way that their outflows are not along our line of sight also radiate a smaller proportion of their infrared luminosity in our direction. It could also be that massive wide-angle outflows destroy the deeply embedded regions responsible for bright HCN-vib emission, so that the two phenomena cannot coexist. This would strengthen the idea that vibrationally excited HCN traces a heavily obscured stage of evolution before nuclear feedback mechanisms are fully developed.]{}
Introduction
============
Luminous and ultraluminous infrared galaxies ((U)LIRGs) are gas-rich systems that radiate intensely in the infrared portion of the electromagnetic spectrum, with infrared luminosities ($L_{\mathrm{IR}}=L(8\text{--}1000$ $\mathrm{\mu m})$) in excess of $10^{11}L_{\sun}$ and $10^{12}L_{\sun}$, respectively [e.g., @san96]. These large luminosities are due to dust-reprocessed radiation from intense star formation, an active galactic nucleus (AGN), or both. In the local Universe, (U)LIRGs are relatively rare compared to less luminous objects, but surveys at millimeter and submillimeter wavelengths have shown that they are much more numerous at high redshifts [e.g., @sma97; @hug98], indicating that they play an important role in the evolution of galaxies. Two important questions in this context are: how did the supermassive black holes (SMBHs) found in the nuclei of most galaxies grow, and how are they related to the evolution of their host galaxies? From observations, we know that tight relations between the mass of the central SMBH and various properties of the host galaxy exist in large ellipticals [e.g., @mag98; @fer00; @geb00; @kor13]. It has been suggested that these relations were partly established through a process, mostly taking place in the early Universe, in which galaxies of similar masses collide and merge [see, e.g., @kor13 for a review]. In such mergers, large amounts of gas are funneled into the new nucleus, giving rise to intense starbursts, AGN activity, or both. The exact physics of how the black hole-host relations would be established are still obscure, but several suggested galaxy merger scenarios include a starburst dominated phase followed by a phase of obscured AGN activity. Eventually, the AGN becomes strong enough to drive outflows that clear the surrounding material and reveal the central activity [e.g., @san88a; @hop06; @urr08; @sim12]. Thus, studies of the most dust-embedded phase of evolution and the early stages of feedback are essential in order to understand how SMBHs grow together with their host galaxies.
However, observations of the most enshrouded galaxy nuclei are hampered by the large amounts of obscuring material that surround them. One solution to this problem is to observe at millimeter and radio wavelengths, where the dust is less optically thick [e.g., @bar15]. Some common tracers of obscured galaxy nuclei at these wavelengths are the low-$J$ rotational lines of the HCN and HCO$^{+}$ molecules. Recently, however, it has been shown that, in the most obscured systems, even these dense-gas tracers are heavily affected by continuum- and/or self-absorption by cooler foreground gas [e.g., @sak09; @aal15b; @mar16]. A better suited tracer of the most obscured systems is offered by HCN in its first vibrationally excited state, $v_{2}=1$ (hereafter HCN-vib). Strong emission lines from rotational transitions within the $v_{2}=1$ state have been detected in external galaxies at (sub)millimeter wavelengths, first by @sak10 and then in many subsequent observations. With energy levels that lie more than $1000$ K above the ground state, HCN-vib is mainly excited by absorption of mid-infrared radiation. However, to be efficiently populated it requires brightness temperatures in excess of $100$ K at $14$ $\mu$m, the wavelength of the transition between the ground and first vibrationally excited states of HCN, translating into a H$_{2}$ column density larger than $2\times10^{23}$ cm$^{-2}$ for a dust temperature of $100$ K [@aal15b]. This makes bright HCN-vib emission an excellent unabsorbed [@mar16] tracer of high column density gas with high mid-infrared surface brightness. While the direct mid-infrared radiation from such regions is often hidden from us by the large quantities of obscuring material, the rotational transitions inside the vibrationally excited states occur at millimeter and submillimeter wavelengths where the dust opacity is lower. The low-J rotational lines of HCN-vib are thus useful probes of the nuclear mid-infrared source in heavily obscured objects. Since the first extragalactic (sub)millimeter detection of HCN-vib rotational line emission by @sak10, the field has developed rapidly [e.g., @ima13; @aal15a; @aal15b; @ima16a; @ima16b; @ima18; @mar16], notably thanks to the Atacama Large Millimeter/submillimeter Array (ALMA).
In their study of HCN-vib in obscured galaxies, @aal15b found that all galaxies with detected HCN-vib emission show evidence of inflows, outflows, or both. In addition, they found a tentative trend that galaxies with fast molecular outflows have fainter HCN-vib emission relative to their total infrared luminosity. The trend is especially striking when comparing the strength of the HCN-vib emission to the velocity of the wide-angle OH outflows studied by, for example, @vei13. Based on a relatively small sample of nine galaxies, @aal15b tentatively suggest that strong HCN-vib emission is connected to a rapid phase of nuclear growth that occurs right before the onset of strong feedback.
In this paper, we present the results of a study employing a sample of $19$ galaxies including data from the literature as well as previously unpublished data, in order to further investigate the connection, or lack thereof, between strong HCN-vib emission and molecular outflows. The sample used in this study is described in Sect. \[sec:sample\] and in Sect. \[sec:results\] we present the results which are then discussed in Sect. \[sec:discussion\]. Finally, the conclusions that we draw from the analysis are summarized in Sect. \[sec:conclusions\]. Throughout the paper, a $H_{0} =73$ kms$^{-1}$Mpc$^{-1}$, $\Omega_{\mathrm{m}} = 0.27$, and $\Omega_{\Lambda} = 0.73$ cosmology is adopted.
Sample {#sec:sample}
======
We have searched for (U)LIRGs with existing ALMA observations of HCN-vib that are either already published or that have the data publicly available in the the ALMA science archive[^2]. In addition, we have included published HCN-vib observations taken with the Plateau de Bure Interferometer (PdBI), the Northern Extended Millimeter Array (NOEMA), or the Submillimeter Array (SMA). We have not included non-detections, unless they have existing observations of the OH doublet at $119$ $\mu$m, which we use to determine the outflow velocity. Non-detections have also been excluded when a meaningful upper limit to the HCN-vib flux could not be determined, either due to low sensitivity spectra or due to significant flux from the HCO$^{+}$ line which is separated from HCN-vib by ${\sim}400$ kms$^{-1}$. This is a possible source of bias, as HCN-vib emission may be hidden by the HCO$^{+}$ line in sources with outflowing gas. In practice, most sources that have been excluded either do not have any outflow signatures in the far-infrared OH lines that we use to trace outflows (see below), or have not been observed in these lines. The two exceptions are IRAS 19254-7245 [@ima16a] and NGC 6240 (ALMA project 2015.1.01448.S, PI: R. Tunnard), in which the median velocities of the OH $119$ $\mu$m lines are ${\sim}-250$ and ${\sim}-200$ kms$^{-1}$, respectively. In total, $10$ sources have been excluded, most of them due to lack of OH measurements.
For consistency of the sample, we ensure that outflow velocities were measured using the same common tracers and following the same procedures. All sources with HCN-vib observations have therefore been checked for observations of the far-infrared OH doublet at $119$ $\mu$m with the *Herschel* Space Observatory [@pil10] as this is a known tracer of molecular outflows on subkiloparsec scales, which has been observed in a large number of (U)LIRGs [e.g., @fis10; @vei13; @gon17]. The OH molecule has high abundances in photodissociation regions as well as in X-ray dominated regions [e.g., @mal96; @goi11; @mei11]. Unlike other common outflow tracers like the millimeter rotational transitions of CO and HCN, the ground state OH $119$ $\mu$m doublet is often detected in absorption towards the far-infrared background [e.g., @vei13], providing an unambiguous tracer of the gas motion. The transitions between excited states (for example the doublets at $65$ and $84$ $\mu$m) are generally weaker and the line wings are not always detected [@gon17]. Recently, the $119$ $\mu$m doublet was also used to detect a fast outflow ($\lesssim800$ kms$^{-1}$) in a galaxy at a redshift of $5.3$ [@spi18]. Systematic searches of outflow signatures in the far-infrared OH lines have revealed evidence of molecular outflows in approximately two thirds of the observed (U)LIRG samples [@vei13; @spo13]. From this detection rate, @vei13 inferred a wide average opening angle of ${\sim}145\degr$, assuming that all objects in their sample have outflows. In fact, based on radiative transfer models as well as spatially resolved observations of two outflows, @gon17 argue that the $119$ $\mu$m OH absorptions are primarily sensitive to wide-angle outflows. On the other hand, unambiguous signatures of inflows are only found in one tenth of the sources in the sample of @vei13, suggesting that OH inflows are either rare or have, on average, smaller opening angles.
In particular, the 43 sources in the far-infrared OH sample of @vei13 have been checked for observations of the (sub)millimeter HCN-vib $J=3\text{--}2$ or $J=4\text{--}3$ lines. Fourteen of these had HCN-vib detections or data that allowed for meaningful upper limits to be estimated. One of them, IRAS 05189-2524 (see Appendix \[app:HCN\]), has previously unpublished data in the ALMA science archive while data for the remaining thirteen were taken from @sak10 [@aal15a; @aal15b; @ima16a; @ima16b; @mar16; @pri17; @ala18]; and K[ö]{}nig et al. (in prep.). In addition, five sources not included in the sample of @vei13 were found to have usable observations of HCN-vib. One, IRAS 12224-0624 (see Appendix \[app:HCN\]), has previously unpublished data in the ALMA science archive while the other four were taken from @aal15b and @ima16a. Three of these five sources, Zw 049.057, IRAS 20414-1651, and NGC 7469, also have existing observations of the OH $119$ $\mu$m doublet (see Appendix \[app:OH\]).
In total, 19 (U)LIRGs with existing HCN-vib observations were found. Of these, 17 also have observations of the OH $119$ $\mu$m doublet. The final sample, with the adopted redshift and distance of each source, is presented in Table \[tab:sample\].
----------------- ------ ---------- ---------------- ------------------------
Name Type $z$ $D_{L}$ $L_{\mathrm{IR}}$
(Mpc) ($10^{11}$ L$_{\sun}$)
NGC 4418 Sy 2 $0.0071$ $33.6\pm2.5$ $1.32\pm0.19$
IC 860 $0.0130$ $62.1\pm4.3$ $1.60\pm0.22$
IRAS 12224-0624 L $0.0264$ $119.5\pm8.2$ $1.98\pm0.27$
Zw 049.057 $0.0130$ $62.4\pm4.3$ $2.04\pm0.28$
NGC 7469 Sy 1 $0.0164$ $68.2\pm4.7$ $4.13\pm0.57$
I Zw 1 Sy 1 $0.0610$ $258.2\pm17.7$ $8.22\pm1.13$
UGC 5101 L $0.0394$ $169.7\pm11.6$ $9.50\pm1.30$
IRAS 20551-4250 $0.0430$ $185.2\pm12.7$ $10.65\pm1.46$
IRAS 15250+3609 L $0.0552$ $243.7\pm16.7$ $11.08\pm1.53$
IRAS 08572+3915 L $0.0584$ $253.1\pm17.3$ $13.38\pm1.84$
IRAS 05189-2524 Sy 2 $0.0428$ $180.2\pm12.4$ $13.50\pm1.85$
IRAS 22491-1808 $0.0778$ $337.0\pm23.1$ $13.88\pm1.92$
Mrk 273 Sy 2 $0.0378$ $165.6\pm11.3$ $14.81\pm2.03$
IRAS 20414-1651 $0.0871$ $383.4\pm26.3$ $15.12\pm2.18$
Arp 220 L $0.0181$ $84.1\pm5.8$ $17.25\pm2.36$
IRAS 13120-5453 Sy 2 $0.0308$ $137.8\pm9.5$ $19.20\pm2.65$
IRAS 12112+0305 L $0.0733$ $326.3\pm22.4$ $20.35\pm2.79$
IRAS 17208-0014 $0.0428$ $189.3\pm13.0$ $26.21\pm3.59$
Mrk 231 Sy 1 $0.0422$ $184.3\pm12.6$ $34.30\pm4.70$
----------------- ------ ---------- ---------------- ------------------------
Results {#sec:results}
=======
Vibrationally excited HCN luminosities, $L_{\mathrm{HCN-vib}}/L_{\mathrm{IR}}$ ratios, and OH median velocities are presented in Table \[tab:properties\]. We adopt the $L_{\mathrm{HCN-vib}}/L_{\mathrm{IR}}$ ratio as a parameter to describe the strength of HCN-vib in a normalized fashion. However, we are aware that some part of the total $L_{\mathrm{IR}}$ may be unrelated to the dusty nucleus, and instead come from, for example, an extended starburst. In particular, in systems with multiple nuclei, for example Arp 220 and IRAS 12112+0305, $L_{\mathrm{IR}}$ will have contributions from both nuclei. In our sample, we are not aware of any sources were these effects would affect the $L_{\mathrm{HCN-vib}}/L_{\mathrm{IR}}$ ratio enough to significantly change our results. We use the median velocity of the OH lines instead of the terminal outflow velocity as the former provides more robust values [@vei13]. Velocities are only given for those sources with absorption in the OH $119$ $\mu$m doublet, as this means that the gas is in the foreground and that a positive or negative velocity shift can be interpreted as evidence of gas moving towards or away from the background nucleus, respectively. We note that this does not necessarily indicate in- or outflowing gas in all cases as, for example, interactions between two nuclei can also affect the OH kinematics. In addition, it is important to take the uncertainties in the velocity determination into account. For example, @vei13 define an outflow as having an OH absorption feature with a median velocity more negative than $-50$ kms$^{-1}$, which is the typical uncertainty on the velocities. The requirement to detect the OH doublet in absorption is a possible source of bias as this requires a high enough column density of OH in front of a strong enough far-infrared background source. However, we note that all galaxies with an HCN-vib detection, if observed, have also been detected in absorption in the OH $119$ $\mu$m doublet. Therefore, no sources with a combination of high HCN-vib luminosity and a fast outflow have been missed due to this effect.
The line luminosities presented in Table \[tab:properties\] have been calculated following Eq. (1) in @sol05, applied to HCN-vib: $$L_{\mathrm{HCN-vib}}=1.04\times10^{-3}S_{\mathrm{HCN-vib}}\,\Delta v\,\nu_{\mathrm{rest}}(1+z)^{-1}D_{\mathrm{L}}^{2},$$ where $L_{\mathrm{HCN-vib}}$ is the HCN-vib luminosity measured in $L_{\odot}$, $S_{\mathrm{HCN-vib}}\,\Delta v$ is the velocity integrated flux in Jykms$^{-1}$, $\nu_{\mathrm{rest}}$ is the rest frequency in GHz, and $D_{\mathrm{L}}$ is the luminosity distance in Mpc. Due to differences in the adopted redshifts and distances, these values differ slightly from those given in the original reference for some of the sources. In case of non-detections, we have included the $3\sigma$ upper limit to the line luminosity. In the case of IRAS 15250+3609, the HCN-vib line is likely blended with a potential outflow signature in the nearby HCO$^{+}$ line. Here, we have attributed the entire emission feature to the HCN-vib line, and the value given should be considered an upper limit to the HCN-vib luminosity. A short discussion of this source as well as descriptions of the analysis of previously unpublished HCN-vib observations and new OH outflow measurements can be found in Appendices \[app:15250\], \[app:HCN\], and \[app:OH\].
-------------------- ------------------------ ---------------------------------------- ------------------------- ------------- ------------
Name $L_{\mathrm{HCN-vib}}$ $L_{\mathrm{HCN-vib}}/L_{\mathrm{IR}}$ $v_{\mathrm{50}}$(abs) Observatory References
($10^{3} L_{\sun}$) ($10^{-8}$) (kms$^{-1}$)
$J=3\text{--}2$
NGC 4418 $5.0\pm1.2$ $3.77\pm0.71$ $111$ SMA 1
IC 860 $4.2\pm0.8$ $2.65\pm0.37$ not obs. PdBI 2
Zw 049.057 $7.3\pm2.0$ $3.57\pm0.76$ $37$ PdBI 2
NGC 7469 $<0.1$ $<0.03$ emi. ALMA 3
I Zw 1 $<1.7$ $<0.21$ emi. ALMA 3
UGC 5101 $20.5\pm5.5$ $2.15\pm0.50$ $-9$ NOEMA 4
IRAS 20551-4250 $2.3\pm0.7$ $0.21\pm0.06$ $-381$ ALMA 5
IRAS 15250+3609 $<11.3$ $<1.02$ $189$ ALMA 3
IRAS 08572+3915 NW $<6.4$ $<0.48$ $-489$ ALMA 3
IRAS 22491-1808 E $13.2\pm3.9$ $0.95\pm0.25$ $99$ ALMA 3
Mrk 273 $<16.9$ $<1.14$ $-201$ NOEMA 6
IRAS 20414-1651 $<7.5$ $<0.50$ $-32$ ALMA 3
Arp 220 W $40.9\pm7.0$ $2.37\pm0.25$ $21$ ALMA 7
Arp 220 E $8.9\pm1.5$ $0.51\pm0.05$ $21$ ALMA 7
IRAS 12112+0305 NE $14.3\pm4.0$ $0.70\pm0.17$ $-117$ ALMA 3
Mrk 231 $14.5\pm3.1$ $0.42\pm0.07$ $-237$ PdBI 8
$J=4\text{--}3$
NGC 4418 $12.4\pm2.5$ $9.41\pm1.25$ $111$ SMA 1
IRAS 12224-0624 $16.9\pm4.5$ $8.52\pm1.95$ not obs. ALMA 9
IRAS 20551-4250 $4.8\pm1.2$ $0.45\pm0.09$ $-381$ ALMA 10
IRAS 05189-2524 $8.0\pm2.7$ $0.59\pm0.18$ $-387\tablefootmark{d}$ ALMA 9
Arp 220 W $108.1\pm19.9$ $6.27\pm0.77$ $21$ ALMA 7
Arp 220 E $25.8\pm4.4$ $1.49\pm0.15$ $21$ ALMA 7
IRAS 13120-5453 $<5.5$ $<0.29$ $-195$ ALMA 11
IRAS 17208-0014 $101.9\pm17.5$ $3.89\pm0.40$ $51$ ALMA 2
-------------------- ------------------------ ---------------------------------------- ------------------------- ------------- ------------
In Fig. \[fig:HCN\_vs\_OH\], the OH absorption median velocities are plotted against the luminosities of the HCN-vib line, normalized to the total infrared luminosity of each galaxy. When available, the $J=3\text{--}2$ line has been chosen. In the four sources where both transitions have been observed, the luminosity ratio between the $J=4\text{--}3$ and $J=3\text{--}2$ lines lies in the range $2\text{--}3$, with a mean of $2.5$. For sources where only the $J=4\text{--}3$ line has been observed, we have therefore divided its luminosity by $2.5$ and included it in the plot with an extra $20\%$ uncertainty added in quadrature. The two galaxies with only emission in the $119$ $\mu$m doublet, I Zw 1 and NGC 7469, are not included in the plot, but we note that neither of them have an HCN-vib detection.
In the plot in Fig. \[fig:HCN\_vs\_OH\] we see that, in our sample, the sources with high $L_{\mathrm{HCN-vib}}/L_{\mathrm{IR}}$ ratios, in excess of $10^{-8}$, all lack fast outflows as traced by the median velocity of the OH $119$ $\mu$m doublet. However, galaxies without fast outflows do not necessarily have high $L_{\mathrm{HCN-vib}}/L_{\mathrm{IR}}$ ratios. The lack of sources in the upper left portion of the diagram is likely due to selection effects; that area should be populated by less obscured sources of lower luminosity that are not able to drive outflows or provide the necessary conditions for efficient HCN-vib excitation.
Discussion {#sec:discussion}
==========
Our comparison of HCN-vib line luminosities and OH $119$ $\mu$m absorption line velocities in nearby (U)LIRGs shows a trend of positive, or low negative, velocities, indicating inflows or slow outflows, in sources with bright HCN-vib emission relative to the total infrared luminosity. This might indicate that regions with the high mid-infrared brightness temperatures required for efficient population of HCN-vib are not affected by inflowing gas, but are destroyed by, or cannot form in the presence of, strong outflows. In this section, we discuss possible reasons for the lack of outflow signatures in the OH $119$ $\mu$m absorptions in HCN-vib luminous sources, and how the HCN-vib lines relate to other tracers of heavily obscured regions.
Why do the HCN-vib luminous galaxies lack outflow signatures in the far-infrared OH lines?
------------------------------------------------------------------------------------------
There are several scenarios that could explain the lack of outflow signatures in the galaxies that have bright HCN-vib emission. One is that there are no outflows, another that the outflows are hidden from detection using the median velocity of the OH $119$ $\mu$m absorption lines, and a third is that the distribution in Fig. \[fig:HCN\_vs\_OH\] is due to orientation effects. These scenarios are discussed in the following sections. We also briefly discuss a possible evolutionary scenario in which the HCN-vib luminous galaxies are in a pre-feedback phase. Admittedly, our sample is quite heterogeneous in that it contains galaxies of different merger types, in different merger stages, of different optical classifications, and with infrared luminosities spanning more than an order of magnitude. It is therefore possible that there are several independent reasons for the lack of far infrared outflow signatures in HCN-vib luminous galaxies.
### No outflows {#sec:nooutflows}
If there are no outflows, it might be due to the HCN-vib luminous galaxies being intrinsically different from the galaxies with weaker HCN-vib emission. For example, @vei13 find a tendency for systems with dominant AGN to have faster OH outflows; it might then be that the HCN-vib luminous systems, which do not show fast OH outflows, are instead starburst dominated. Indeed, in the two sources with the highest absolute HCN-vib luminosities, IRAS 17208-0014 and Arp 220, less than $10\%$ of the bolometric luminosities are estimated to come from AGN activity, based on the $15$ to $30$ $\mu$m continuum ratio [@vei13]. However, this diagnostic can be misleading in galaxies where the nuclei are optically thick into the far infrared. Furthermore, at least two other sources with strong HCN-vib emission, NGC 4418 and UGC 5101, have AGN fractions $\gtrsim 50\%$. While the uncertainty on the AGN fraction determined in this way is estimated to be $20\%$ on average, it is likely much higher in the strongly buried sources discussed here.
Another possible explanation for a real lack of outflows in some sources is that we are witnessing a pre-feedback phase in young systems that will later evolve into objects with weak HCN-vib emission and fast outflows. This possibility is discussed further in Sect. \[sec:evolution\]. As discussed by @gon17, the lack of outflows could also be due to the fact that, in the most extremely buried sources, it is difficult to find paths with more moderate columns that can be efficiently accelerated to high velocities.
Interestingly, as @aal15b notes, fast molecular outflows have been found, with other tracers and at longer wavelengths, in some of the HCN-vib luminous systems. For example, a compact $v>800$ kms$^{-1}$ CO $J=2\text{--}1$ outflow has been found in the ULIRG IRAS 17208-0014 [@gar15] and an equally fast, compact, and collimated, HCN $J=1\text{--}0$ outflow has been detected in the ULIRG Arp 220 [@bar18]. Possible outflow signatures are also found in Zw 049.057 [@fal18] as well as in NGC 4418 and IRAS 22491-1808 [@flu18], although at lower velocities. If we also consider the fact that UGC 5101 has faint high-velocity wings in its OH $119$ $\mu$m absorption lines [@vei13], it turns out that all of the six most HCN-vib luminous sources have signatures of molecular outflows.
### Obscured, episodic, or collimated outflows {#sec:obscuredoutflows}
For the HCN-vib bright systems that do have outflows, there must be a reason why we do not see them in the far-infrared OH absorption lines. One possibility is that these objects have extreme optical depths in the far-infrared dust continuum, so that much of the outflowing gas cannot be seen due to obscuration by the dust. In at least some of the sources, high optical depths have indeed been inferred from radiative transfer modeling of *Herschel* observations [e.g., @gon12; @fal15]. If this is combined with young outflows that are still very compact, any signatures of them may be completely hidden in the far-infrared. In such a situation, inflows on larger scales, for example as the one seen in Arp 299A [@fal17], may still be detectable. In fact, at least three of the five most obscured sources [Arp 220, IRAS 17208-0014, and Zw 049.057; @baa89; @fal18] have outflows that are seen in the OH lines at centimeter wavelengths but not in the far-infrared. In Zw 049.057, infrared and optical images reveal multiple clouds in a polar dust structure, which might suggest that the outflow in this galaxy is not steady in time (Gallagher et al. in prep.). It is thus possible that, in some galaxies, we are witnessing an episodic process where the nuclear region is alternating between feedback and accretion dominated states.
Looking at the spectra presented by @vei13, UGC 5101 and IRAS 17208-0014 even exhibit weak blueshifted line wings in the OH $119$ $\mu$m absorption lines. This indicates the presence of gas moving towards us, but as the absorptions at negative velocities are shallow, they do not have a strong effect on the median velocities of the lines. As discussed by @gon17, this may suggest that we are witnessing collimated outflows in these sources, as opposed to the wide-angle outflows that the OH $119$ $\mu$m lines seem to be primarily sensitive to. With all this in mind, it seems that the regions responsible for the bright HCN-vib emission are not destroyed by (fast) outflows in general, but rather by the kind of wide-angle outflows from the nuclear infrared-emitting region that are able to shift the median velocity of the OH $119$ $\mu$m ground state lines by several hundreds of kms$^{-1}$.
Turning to the two sources that show pure emission in the OH $119$ $\mu$m doublet, I Zw 1 and NGC 7469, we note that they lack HCN-vib detections and that the upper limits to their $L_{\mathrm{HCN-vib}}/L_{\mathrm{IR}}$ ratios are among the lowest in our sample. Furthermore, NGC 7469 has also been observed in the OH $65$ $\mu$m doublet, another tracer of obscured nuclei, but no absorption was detected [@gon15]. This is consistent with the results of @vei13 that the sources with pure OH emission are the ones where an AGN dominates the luminosity and that these objects represent a phase where the feedback has subsided after clearing a path through the dusty surroundings.
### Orientation effects {#sec:orientation}
An alternative explanation to the distribution of galaxies in Fig. \[fig:HCN\_vs\_OH\] is that it is produced by a simple orientation effect. For example, using the OH $119$ $\mu$m doublet, @vei13 found unambiguous evidence of outflows in $70\%$ of their sources, a detection rate which is consistent with all sources having molecular outflows, with an average opening angle of ${\sim}145\degr$. The sources with no outflow signatures would then be those that are oriented approximately edge-on, with outflows perpendicular to our line of sight. It is harder to explain why these would also be the sources with bright HCN-vib emission, as the obscuration in a dusty edge-on disk would rather have the effect of lowering the visible HCN-vib flux, unless the line is masing as seen towards some circumstellar envelopes [e.g., @luc89; @bie01; @men18].
However, the comparison so far has been based on the HCN-vib luminosity divided by the total infrared luminosity which, due to the higher dust opacity in the infrared, is more affected by orientation effects. The infrared luminosity that we see will thus depend on the orientation, with higher values for face-on galaxies. This effect may indeed be significant; for example, @efs14 estimated that the true infrared luminosity of IRAS 08572+3915 may be as high as $1.1 \times 10^{13}$ L$_{\odot}$, more than five times higher than the value inferred assuming the luminosity is isotropic. Another possible example is Arp 220 where the total infrared luminosity estimated from the IRAS fluxes is ${\sim}1.7\times10^{12}$ L$_{\odot}$ while @sak17 estimate the bolometric luminosity from the western nucleus alone to be ${\sim}3\times10^{12}$ L$_{\odot}$.
If we do not divide $L_{\mathrm{HCN-vib}}$ by the total infrared luminosity, we see that Mrk 231 has an HCN-vib luminosity comparable to those of the HCN-vib bright galaxies. However, it is still less luminous than in the ULIRGs Arp 220 W and IRAS 17208-0014, and only a factor of two or three brighter than in the LIRGs Zw 049.057, IC 860, and NGC 4418. Furthermore, none of the other sources with $v_{50}\lesssim-200$ kms$^{-1}$ have strong HCN-vib emission. Clearly, a larger sample will be required to investigate this issue further. If the distribution of sources in Fig. \[fig:HCN\_vs\_OH\] is indeed due to an orientation effect, more sources with both strong HCN-vib emission and outflow signatures in the far-IR OH lines should show up in an extended sample.
### Evolution {#sec:evolution}
Following @aal15b, we present a possible evolutionary scenario in which bright HCN-vib emission is tracing extremely obscured nuclei in a phase that occurs before the nuclear feedback is able to drive wide-angle outflows, or at least before such outflows can be observed in the far-infrared OH lines. In this scenario, the deeply embedded regions which provide the necessary conditions for HCN-vib excitation can coexist with fast collimated outflows, but are disrupted once nuclear wide-angle outflows have developed. A schematic view of one possible version of the scenario is presented in Fig. \[fig:scenario\]. Here, the most obscured sources consist of a mid-infrared core, responsible for the HCN-vib emission, surrounded by an obscuring layer of cooler dust that accounts for the far-infrared emission. Outflows in these objects are collimated and often so compact that they are still embedded in the outer layer of dust, and thus obscured from view in the far-infrared. In the next phase, the outflows have broken through the cooler dust layer and widened considerably, making them clearly detectable in most galaxies. As gas and dust is transported from the regions around the nucleus, the conditions in the mid-infrared core change, becoming less favorable for the excitation of HCN-vib. Finally, when a path has been cleared through the surrounding dust, OH is no longer detectable in absorption and the conditions for HCN-vib excitation in the core are no longer met.
A similar evolutionary sequence has been suggested by @gon17b, and we note that, apart from UGC 5101, the HCN-vib luminous sources from our sample coincide with the sources in their extremely buried pre-feedback phase. It should be noted that, even if an evolutionary scenario like this applies in some cases, all sources in our sample may not be on the same evolutionary track. For example, a lot of nuclear power is required in order to push large columns of gas to high velocities, and LIRGs like NGC 4418 and Zw 049.057 may therefore never develop powerful wide-angle outflows detectable in the far-IR OH lines. Instead, they might evolve along the horizontal axis in Fig. \[fig:HCN\_vs\_OH\].
Finally, an interesting comparison can be made with the highly collimated molecular outflows that are found around low-mass protostars in the earliest (class 0) stages of their formation [e.g., @kon00; @cod14]. With time, gas at larger and larger angles from the outflow axis may be entrained, eventually sweeping away most of the envelope around the star [e.g., @arc06]. Furthermore, vibrationally excited HCN has been detected toward molecular hot cores around massive protostars using the direct $l$-type transitions, that is transitions between the $l$-type levels inside a rotational state, at cm-wavelengths [e.g., @rol11a].
![Schematic view of a possible evolutionary scenario. The upper left panel shows the most obscured phase where both the mid-infrared core and the outflow are still completely embedded in a layer of cooler dust. The upper right panel shows the conditions in the mid-infrared core in this phase. A central energy source is heating the dusty core through radiative trapping. Mid-infrared photons, which cannot escape the dusty environment, vibrationally excite the HCN molecules which may then undergo rotational transitions and emit (sub)millimeter photons that are able to penetrate the obscuring dust. In the lower left panel, the outflow has broken through the cooler layer of dust and widened, making it visible in the far-infrared. The lower right panel shows a third stage where the outflow has cleared a path through the surrounding dust layer. In the two final stages, conditions in the core are no longer sufficient for luminous HCN-vib emission to occur.[]{data-label="fig:scenario"}](fig/drawing_scenario_v3.eps){width="50.00000%"}
HCN-vib compared to other tracers of obscured nuclear regions.
--------------------------------------------------------------
As discussed in @aal15b, a prerequisite for the excitation of HCN-vib is the presence of warm and dusty regions, making it a good tracer of deeply buried nuclei. Following this logic, our results indicate that the most obscured nuclei lack evolved, wide-angle outflows. Interestingly, the results of @gon17, who use the equivalent width of the OH $65$ $\mu$m doublet as a measure of the obscuration, suggest that the fastest outflows arise in some of the most obscured nuclei. They do however point out that sources with high equivalent width in OH $65$ $\mu$m do not necessarily have fast outflows. By comparing our Fig. \[fig:HCN\_vs\_OH\] with their Fig. 4, we see that five of the six most HCN-vib luminous sources that are included in both samples also belong to the group of sources with high equivalent width but low outflow velocity, indicating that they are obscured also when using the OH $65$ $\mu$m equivalent width as a measure. On the other hand, five of the sources with low HCN-vib luminosity are also found among the sources with high OH $65$ $\mu$m equivalent width, indicating that they also contain obscured nuclei. It appears that the two measures trace different parts or physics of the galaxy nuclei. Indeed, @gon17 state that high equivalent widths require dust temperatures in excess of $60$ K, while dust temperatures above $100$ K are required for bright HCN-vib emission [@aal15b]. It is thus possible that bright HCN-vib emission exclusively traces more obscured parts of the nuclei, where the dust is hotter.
In a similar manner, @spo13 also find that the fastest outflows are found in those sources that are still deeply embedded as indicated by strong mid-infrared silicate absorptions. However, as already noted by @aal15b, @gon15 find that silicate absorptions are biased toward relatively unobscured mid-infrared emitting regions, again indicating that the HCN-vib is actually tracing a more extreme form of obscuration.
Conclusions {#sec:conclusions}
===========
We explore a possible correlation between the strength of molecular outflows in the far-infrared OH $119$ $\mu$m lines and the luminosity of rotational lines of vibrationally excited HCN. A simple comparison of HCN-vib line luminosities, normalized to the total infrared luminosity of the host galaxy, and the median velocities of OH $119$ $\mu$m absorption lines shows that galaxies with unusually bright HCN-vib emission tend to lack fast molecular outflows, but that galaxies without fast outflows do not necessarily have bright HCN-vib emission. This may be an orientation effect or something that reflects a true difference between the sources, for example that the most obscured sources cannot drive wide-angle outflows or that their outflows are young and have not yet dispersed the nuclear gas and dust concentrations. Following @aal15b, we present an evolutionary sequence, that may apply to some of the sources, in which bright HCN-vib emission is tracing extremely obscured nuclei in a phase that occurs before the onset of wide-angle outflows. Once these massive outflows have been launched, they quickly disrupt the deeply embedded regions responsible for the bright HCN-vib emission. We note, however, that more studies are needed in order to extend the sample and explore the relation further. As the *Herschel* Space Observatory, which was used for the OH outflow measurements, is no longer operational, such studies should concentrate on observations of HCN-vib in galaxies which have already been observed in the OH $119$ $\mu$m doublet. Another possibility is to search for hidden outflows using the radio lines of OH, either in its ground state [@baa89] or in rotationally excited states [@fal18].
This paper makes use of the following ALMA data: ADS/JAO.ALMA\#2015.1.00708.S, ADS/JAO.ALMA\#2016.1.00140.S. ALMA is a partnership of ESO (representing its member states), NSF (USA) and NINS (Japan), together with NRC (Canada), MOST and ASIAA (Taiwan), and KASI (Republic of Korea), in cooperation with the Republic of Chile. The Joint ALMA Observatory is operated by ESO, AUI/NRAO and NAOJ.
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.
Vibrationally excited HCN in IRAS 15250+3609 {#app:15250}
============================================
The ULIRG IRAS 15250+3609 was observed in the HCO$^{+}$ $J=3\text{--}2$ line by @ima16a. They interpreted a peak close to the expected position of the HCN-vib line as an outflow signature in the HCO$^{+}$ line. Their interpretation is supported by the fact that similar features are seen in the HCN $J=3\text{--}2$ and $J=4\text{--}3$ lines [@ima16a; @ima18]. However, relative to the main lines, the sub-peak close to the HCO$^{+}$ line is stronger than the one close to the HCN line. This could be due to differing abundances or excitation in the outflowing gas but it could also be a contribution from HCN-vib. Furthermore, IRAS 15250+3609 has a high equivalent width in the OH $65$ $\mu$m doublet [@gon15], indicating that the galaxy is highly obscured, as well as an HCN absorption at $14$ $\mu$m [@lah07]. Here, we have attributed the feature as completely due to HCN-vib emission, but it is likely that it is indeed blended with an HCO$^{+}$ outflow signature and the value given here should be considered an upper limit to the HCN-vib luminosity.
Previously unpublished HCN-vib detections {#app:HCN}
=========================================
We found two sources with previously unpublished detections of HCN-vib in the ALMA science archive, IRAS 12224-0624 (project 2015.1.00708.S, PI: L. Armus), and IRAS 05189-2524 (project 2016.1.00140.S, PI: D. Iono). The observations of IRAS 05189-2524 were conducted during two runs on 2016 November 12 and 15 with a total of $42$ antennas with baseline lengths ranging between $15.1$ m and $1.0$ km. A spectral window centered at $341.8$ GHz covered a bandwidth of $1.875$ GHz (${\sim}1650$ kms$^{-1}$ at the frequency of HCN-vib $J=4\text{--}3$), with a frequency resolution of $3.9$ MHz. During both runs was used as bandpass and flux calibrator, and was used as phase calibrator. The total on-source time of the observations was $3340$ s and the final sensitivity achieved was $0.5$ mJybeam$^{-1}$ per $20$ kms$^{-1}$ (${\sim}24$ MHz) channel. For the imaging we used Briggs weighting with a robustness factor of 0.5. The resulting beam size is $0.29\arcsec\,\times\,0.22\arcsec$ (PA ${\sim}63\degr$).
The observations of IRAS 12224-0624 were conducted during a single run on 2016 April 27 with a total of $38$ antennas with baseline lengths ranging between $15.1$ m and $452.8$ m. A spectral window centered at $347.4$ GHz covered a bandwidth of $1.875$ GHz (${\sim}1600$ kms$^{-1}$ at the frequency of HCN-vib $J=4\text{--}3$), with a frequency resolution of $3.9$ MHz. The calibrators used were , which was used for bandpass and flux calibration, and , which was used for phase calibration. The total on-source time of the observations was $395$ s and the final sensitivity achieved was $1.5$ mJybeam$^{-1}$ per $20$ kms$^{-1}$ (${\sim}24$ MHz) channel. For the imaging we used Briggs weighting with a robustness factor of 0.5. The resulting beam size is $0.51\arcsec\,\times\,0.44\arcsec$ (PA ${\sim}82\degr$).
For both sources, the Common Astronomy Software Applications [CASA; @mcm07] package was used to reapply the calibration and perform imaging. The line fluxes used to calculate the HCN-vib luminosities were then found by fitting Gaussians to the HCO$^{+}$ and HCN-vib lines in spectra that were spatially integrated over the sources. These spectra are presented in Fig. \[fig:new\_hcnvib\]. We note that the wavelengths of the HCO$^{+}$ and HCN-vib lines in IRAS 05189-2524 are offset from those expected when using the redshift of $0.0426$ adopted by @vei13. They are however consistent with the redshift of ${\sim}0.0428$ found from the CO $J=1\text{--}0$ observations reported by @san91. Using this redshift instead of the one adopted by @vei13 to determine the median OH absorption velocity results in an extra velocity shift of $-45$ kms$^{-1}$.
![Spectral fits to the HCO$^{+}$ and HCN-vib $J=4\text{--}3$ lines in the two galaxies IRAS 05189-2524 and IRAS 12224-0624. The solid black histograms represent the data and the solid magenta lines are the best fits to the data. The velocity scale is set relative to the frequency of the HCO$^{+}$ $J=4\text{--}3$ line. Dashed vertical lines indicate the expected positions of the HCO$^{+}$ and HCN-vib $J=4\text{--}3$ lines given the adopted redshifts. The dotted lines in the plot for IRAS 05189-2524 indicate the expected positions given the redshift adopted by @vei13.[]{data-label="fig:new_hcnvib"}](fig/iras05189_2.eps "fig:") ![Spectral fits to the HCO$^{+}$ and HCN-vib $J=4\text{--}3$ lines in the two galaxies IRAS 05189-2524 and IRAS 12224-0624. The solid black histograms represent the data and the solid magenta lines are the best fits to the data. The velocity scale is set relative to the frequency of the HCO$^{+}$ $J=4\text{--}3$ line. Dashed vertical lines indicate the expected positions of the HCO$^{+}$ and HCN-vib $J=4\text{--}3$ lines given the adopted redshifts. The dotted lines in the plot for IRAS 05189-2524 indicate the expected positions given the redshift adopted by @vei13.[]{data-label="fig:new_hcnvib"}](fig/iras12224_2.eps "fig:")
New OH outflow measurements {#app:OH}
===========================
Three of the sources with existing HCN-vib observations also had observations of the OH doublet at $119$ $\mu$m taken with the Photodetector Array Camera and Spectrometer [PACS; @pog10] on *Herschel* but were not included in the sample of @vei13. The observations of Zw 049.057 (OBSID: 1342248368, PI: E. González-Alfonso) were conducted on 2012 July 20 for a duration of $1478$ s, the observations of IRAS 20414-1651 (OBSID: 1342217908, PI: D. Farrah) were conducted on 2011 April 5 for a duration of $8879$ s, and the observations of NGC 7469 (OBSID: 1342235840, PI: E. González-Alfonso) were conducted on 2011 December 31 for a duration of $1086$ s. All observations were performed in high spectral sampling, range spectroscopy mode. The observations had been processed with version 14.2 of the standard pipeline and were not in need of reprocessing. In all sources, the nuclear far-IR emission is spatially unresolved in the central $9.4$ (${\sim}2$ kpc) spaxel of the PACS $5$ x $5$ spaxel array. As the central PACS spaxel is smaller than the point spread function of the spectrometer, the spectrum was extracted using the point source correction task in the *Herschel* interactive processing environment [HIPE; @ott10] version 14.0.1. Before analyzing the absorption lines, polynomials of order two were fitted to the continuum and then subtracted from the spectra. For consistency, the profiles of the $119$ $\mu$m OH doublets were modeled using the same procedure as in @vei13. Each line was fitted with two Gaussian components characterized by their amplitude, position, and width. The separation between the two lines of the doublet was fixed at $0.208$ $\mu$m in the rest frame and the amplitude and width of the two lines were the same for each component. The median velocities of the absorptions ($v_{\mathrm{50}}$(abs)) were then determined from these fits. The fitting procedure was carried out using the spectroscopic analysis toolkit *PySpecKit* [@gin11] and the continuum subtracted spectra with the fits overplotted are presented in Fig. \[fig:new\_oh\].
![Spectral fits to the OH $119$ $\mu$m absorption lines in the galaxies Zw 049.057, IRAS 20414-1651, and NGC 7469. The solid black histograms represent the data, the solid magenta lines are the best multi-component fits to the data, and the dashed blue lines are the individual components. The velocity scale is set relative to the frequency of the blue component of the doublet. Dashed vertical lines indicate the expected positions of the two absorption components given the adopted redshifts.[]{data-label="fig:new_oh"}](fig/zw049.eps "fig:") ![Spectral fits to the OH $119$ $\mu$m absorption lines in the galaxies Zw 049.057, IRAS 20414-1651, and NGC 7469. The solid black histograms represent the data, the solid magenta lines are the best multi-component fits to the data, and the dashed blue lines are the individual components. The velocity scale is set relative to the frequency of the blue component of the doublet. Dashed vertical lines indicate the expected positions of the two absorption components given the adopted redshifts.[]{data-label="fig:new_oh"}](fig/iras20414.eps "fig:") ![Spectral fits to the OH $119$ $\mu$m absorption lines in the galaxies Zw 049.057, IRAS 20414-1651, and NGC 7469. The solid black histograms represent the data, the solid magenta lines are the best multi-component fits to the data, and the dashed blue lines are the individual components. The velocity scale is set relative to the frequency of the blue component of the doublet. Dashed vertical lines indicate the expected positions of the two absorption components given the adopted redshifts.[]{data-label="fig:new_oh"}](fig/ngc7469.eps "fig:")
[^1]: [*Herschel*]{} is an ESA space observatory with science instruments provided by European-led Principal Investigator consortia and with important participation from NASA.
[^2]: The ALMA science archive is available at https://almascience.nrao.edu/alma-data/archive.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We analyze the magnetic oscillations (MO) in pristine graphene, under a perpendicular magnetic field, taking into account the Zeeman effect. We consider a constant Fermi energy, such that the valence band is always full and only the conduction band is available. At zero temperature the MO consist of two sawtooth peaks, one for each spin. Both peaks have the same frequency, but different amplitude and phase. We show that, in order to observe the spin splitting in the MO, Fermi energy of about 0.1 eV is required. At low temperatures we obtain that the MO can be expressed as the MO at zero temperature, plus small Fermi-Dirac like functions, each centered around the MO peaks. Using this expression, we show that the spin splitting is observable in the MO only when the thermal energy is smaller than the Zeeman energy. We also analyze the shift of the MO extrema as the temperature increases. We show that it depends on the magnetic field, which implies a broken periodicity at nonzero temperature. Finally, we obtain an analytical expression for the MO envelope.'
address: |
IFISUR, Departamento de Física (UNS-CONICET), Av. Alem 1253, B8000CPB Bahía Blanca, Argentina\
Instituto de Física del Sur (IFISUR, UNS-CONICET)\
Av. Alem 1253, B8000CPB Bahía Blanca, Argentina
author:
- 'F Escudero, J S Ardenghi, P Jasen'
bibliography:
- 'graphene.bib'
title: 'Influence of temperature on the magnetic oscillations in graphene with spin splitting: a new approach'
---
Introduction
============
Since its experimental isolation in 2004, graphene has become one of the most interesting and promising materials in condensed matter science [@Novoselov2005; @Geim2007; @Zhang2005]. Its unique features, like high heat and electrical conductivity, has make graphene an exciting material for technological applications [@Avouris2012; @Novoselov2012]. These properties are related to its 2D hexagonal structure, composed of two interpenetrating sublattices A and B [@Wallace1947]. Without impurities or defects, the conduction and valence bands touch at the Fermi energy, with the valence band full and the conduction band empty in the ground state [@Neto2009]. Furthermore, in pristine graphene the density of states at the Fermi energy is zero, making graphene a semiconductor with zero band gap, or a semi-metal [@Lu2013]. In the long wavelength approximation the dispersion relation is relativistic and the electrons behave as massless fermions, moving with a Fermi velocity $\sim 10^6$ m/s [@Hwang2012].
The magnetic properties of graphene and 2D systems have been investigated in recent works [@Goerbig2011; @Ardenghi2015; @Ardenghi2014; @Ardenghi2013; @Escudero2018]. Unlike conventional materials, the magnetization in graphene has unique features. This can be related to the Landau levels (LL) that appear when a magnetic field is applied [@Kuru2009]. In a classical system, these LL are equidistant due to the parabolic dispersion relation. But in graphene, at low energies the dispersion relation is relativistic [@Neto2009], causing the LL to be not equidistant, which in turn affects the oscillating behavior in the thermodynamics potentials [@Goerbig2011]. For instance, the magnetic oscillations (MO), the so called de Haas van Alphen effect [@deHaas], are sawtooth at zero temperature [@Sharapov2004; @Fu2011], with the peaks being caused by the change in the last occupied energy level [@Escudero2017].
The MO are altered by impurities and temperature. In each case the MO are broadened as a result of the modified density of states (DOS) and the Fermi-Dirac distribution. In classical metals this is described by the Lifshitz-Kosevich formula [@Shoenberg1984], which incorporates the effects of impurities and temperatures, as well as the spin. The essential feature of this formula is that the MO are written as an infinite series, where the damping effects are taken into account by the introduction of reduction factors. This formula has been extended to the case of graphene [@Sharapov2004], where the difference only lies in the form the reduction factors take. In principle, this formula would be sufficient to describe the MO in the most general case, but due to its complexity it is not very useful. Thus, in general, limit cases are considered. The most familiar one is zero temperature an energy independent impurities, where the only reduction factor is the Dingle factor, in which case the infinite series can be easily evaluated. Another interesting situation is the pristine case (i.e. without impurities) at nonzero temperature. In this case the general series cannot yet be evaluated, due to the complicated temperature reducing factor, so in general some approximations are considered. The most prominent one is the limit $B\rightarrow0$ (low magnetic field), where one can simplify the temperature reduction factor and easily evaluate the infinite series. Nevertheless, one could argue that this limit may not be very useful, since the lower the magnetic field the more difficult is to observe the MO.
On the other hand, it is well known that when the Zeeman effect is considered, a splitting of the Landau levels appears, which is relevant for several thermodynamical properties [@Escudero2017]. In turn, when a gate voltage is applied, the electron density can be changed, which could be useful for spin filtering [@Hanson2004], spin-polarized currents in 2D systems [@Potok2003] and spin conductivity [@Sinitsyn2004]. Moreover, when other effects are considered, such as spin-orbit coupling due to an external electric field, the MO changes drastically [@Escudero2017a] and the interplay between the degeneracy of each level, the disorder and spin-orbit coupling is not trivial [@Escudero2017b].
Motivated by this, we analyzed the MO in pristine graphene, in the presence of a perpendicular magnetic field, taking into account the Zeeman effect. We have organized this work as follow: in section 2 we study the MO in graphene at zero temperature, taking into the account the Zeeman effect. We then analyze the conditions to observe the spin splitting in the MO. In section 3 we study the MO at nonzero temperature, where we make an approximation to obtain an useful analytical formula for the magnetization. From this we analyze how the temperature affects the observation of the spin splitting in the MO. We end with a study of the maxima and minima shift in the MO as a function of the temperature, from which we obtain an expression for the oscillation envelope. Finally, our conclusions follow in section 4.
MO at zero temperature
======================
In order to study the MO, we will consider that the Fermi energy $\mu$ is held constant. Moreover, we shall always take $\mu>0$, so that the valence band is always full. Given that $\mu$ is fixed, while the number of electrons $N$ may change, it is convenient to work with the grand potential $\Omega$. We will consider the long wavelength approximation, with energies close to the Fermi level. This condition is satisfied as long as [@Neto2009] $\mu\ll\left|t\right|\sim3$ eV ($t$ is the nearest neighbor hopping amplitude). In this case the electrons in graphene behave as relativistic massless fermions, whose dynamics is given by the corresponding Dirac Hamiltonian. In the presence of a perpendicular magnetic field $B$ one obtains the discrete Landau level, as well as a spin splitting due to the Zeeman effect [@Zeeman1897]. Then it can be shown [@Escudero2017] that in graphene the energy levels are given by $\varepsilon_{\lambda,n,s}=\lambda\alpha\sqrt{nB}-s\mu_{\mathrm{B}}B$, where $\alpha=\upsilon_{F}\sqrt{2e\hbar}$ and $\upsilon_{F}\sim10^{6}$ m/s is the Fermi velocity, while $\lambda=\pm1$ for the conduction and valence bands, $n=0,\:1,\,2,\ldots$ for the Landau level (LL) and $s=\pm1$ for the spin (throughout this paper we will use +1 to indicate spin up and -1 to indicate spin down). Each energy level has a degeneracy given by $D=2\mathcal{A}B/\phi$, where $\mathcal{A}$ is the sheet area of graphene, $\phi=h/e$ is the magnetic unit flux and the factor of 2 takes into account the valley degeneracy [@Goerbig2011].
For a Fermi energy $\mu>0$ at zero temperature, the valence band is full while the conduction band is partially filled. We will write the conduction energy levels as $\varepsilon_{m}=\alpha\sqrt{n_{m}B}-s_{m}\mu_{\mathrm{B}}B,$ where we have introduced the decreasing energy sorting index $m=0,\:1,\,2,\ldots$, so $n_{m}$ gives the LL and $s_{m}$ the spin for the $m$ position. In general, the mixing of the LL depends on the magnetic field [@Chang1997], but for usual values there is no spin mixing so $n_{m}=\sqrt{m/2-\left[1-\left(-1\right)^{m}\right]/4}$ and $s_{m}=\left(-1\right)^{m}$. At a given $\mu>0$, all conduction energy levels $m=0,\:1,\,2,\ldots,\,f$ are filled, where $f$ is such that $\varepsilon_{f}<\mu\leq\varepsilon_{f+1}$. Then the grand potential at zero temperature is
$$\Omega=\Omega_{V}+\sum_{m=0}^{f}D\left(\varepsilon_{m}-\mu\right),\label{GP}$$
where $\Omega_{V}$ is the grand potential due to the filled valence band. From equation (\[GP\]) we shall expect the oscillatory contribution coming only from the last term, associated with the change in the last energy level as $B$ is modified. On the other hand, the first term $\Omega_{V}$ makes a non-oscillatory contribution since the valence band is always full if $\mu>0$. Therefore we omit for the moment $\Omega_{V}$ and consider only the conduction grand potential $\Omega_{C}=\sum_{m=0}^{f}D\left(\varepsilon_{m}-\mu\right)$. Separating $\varepsilon_{m}=\varepsilon_{m}^{0}-s_{m}\mu_{\mathrm{B}}B$, with $\varepsilon_{m}^{0}=\alpha\sqrt{n_{m}B}$, we get
$$\Omega_{C}=\Omega_{0}-D\mu_{\mathrm{B}}B\sum_{m=0}^{f}s_{m},\label{GP2}$$
where $\Omega_{0}=\sum_{m=0}^{f}D\left(\varepsilon_{m}^{0}-\mu\right)$. The last term in equation (\[GP2\]) is related to the Pauli paramagnetism associated with the spin population. Indeed, we have $D\sum_{m=0}^{f}s_{m}=N_{+}-N_{-}$, where $N_{+}$($N_{-}$) is the total number of spin up(down) states. Thus $\mu_{\mathrm{B}}D\sum_{m=0}^{f}s_{m}=\mu_{\mathrm{B}}D\left[1+\left(-1\right)^{f}\right]/2=\mu_{\mathrm{B}}\left(N_{+}-N_{-}\right)=M_{p}$, so equation (\[GP2\]) becomes
$$\Omega_{C}=\Omega_{0}-BM_{P}.\label{GP3}$$
The conduction magnetization is given by $M_{C}=-\mathcal{A}^{-1}\left(\partial\Omega_{C}/\partial B\right)_{\mu}$, where $\mathcal{A}$ is the graphene area. From equation (\[GP3\]) we get
$$M_{C}=M_{0}+\frac{1}{\mathcal{A}}\left(M_{P}+B\frac{\partial M_{P}}{\partial B}\right),\label{Mc-1}$$
where $M_{0}=-\mathcal{A}^{-1}\left(\partial\Omega_{0}/\partial B\right)_{\mu}$. Given that $\partial D/\partial B=D/B$, while $\partial\varepsilon_{m}^{0}/\partial B=\varepsilon_{m}^{0}/2B$, we can write
$$M_{C}=-\frac{1}{2B}\left(\frac{3\Omega_{C}}{\mathcal{A}}+n\mu\right)+\frac{1}{2}m_{P},\label{Mc}$$
where $n=N/\mathcal{A}=\sum_{m=0}^{f}D/\mathcal{A}=D\left(f-1\right)/\mathcal{A}$ is the density of conduction electrons and $m_{p}=M_{P}/\mathcal{A}$. Equation (\[Mc\]) shows that the MO have a sawtooth oscillation (SO) produced whenever $n$ or $m_{P}$ changes discontinuously, $\Omega_{C}$ being continuous always. The sawtooth peaks amplitude $\Delta M$ are given by
$$\Delta M=-\frac{\mu}{2B}\Delta n+\frac{1}{2}\Delta m_{P}.\label{DeltaM}$$
Each contribution $\Delta n$ and $\Delta m_{P}$ is determined by the discontinuous change in the parameters $n_{f}$ and $s_{f}$, which define the last energy level occupied. Considering the possible changes of LL and spin, we can write the MO as a sum of two sawtooth. In general the period of oscillation is given by $\Delta(1/B)=1/B_{2}-1/B_{1}$, where $B_{i}$ is such that $\varepsilon_{f_{i}}(B_{i})=\mu$. Therefore $\mu=\alpha\sqrt{B_{i}n_{i}}-s_{i}\mu_{\mathrm{B}}B_{i}$, where for simplicity we have noted $n_{f_{i}}=n_{i}$ and $s_{f_{i}}=s_{i}$. Using the approximation $\mu^{2}+2s_{i}\mu\mu_{\mathrm{B}}B_{i}+\mu_{\mathrm{B}}^{2}B_{i}^{2}\simeq\mu^{2}+2s_{i}\mu\mu_{\mathrm{B}}B_{i}$, which holds for typical values of magnetic field, we obtain the $B_{i}$ at which the peaks occur:
$$\frac{1}{B_{i}}=\frac{n_{i}\alpha^{2}}{\mu^{2}}-\frac{2s_{i}\mu_{\mathrm{B}}}{\mu}.\label{1/Bi}$$
We shall consider that the two peaks correspond to a fixed spin, with the oscillation being given when the LL changes by one. Thus we take $\Delta n=n_{2}-n_{1}=1$, while $s=s_1=s_2$. Therefore, from equation (\[1/Bi\]) we obtain the period $\Delta(1/B)$ and frequency $\omega=\left[\Delta(1/B)\right]^{-1}$
$$\omega_{s}=\frac{\mu^{2}}{\alpha^{2}}.\label{freq}$$
Then we can write equation (\[1/Bi\]) as $1/B_{i,s}=n_{i}/\omega_{s}+\Delta_{s}$, where $\Delta_s$ is the phase
$$\Delta_{s}=-\frac{2s_{i}\mu_{\mathrm{B}}}{\mu}.\label{phase}$$
From equations (\[freq\]) and (\[phase\]) we see that $\omega_{1}=\omega_{-1}$ and $\Delta_{1}=-\Delta_{-1}$. This means that the two sawtooth peaks have the same frequency but different phase. The peaks amplitude are obtained from equation (\[DeltaM\]). Suppose the magnetic field is increased so the last sorted position $f$ changes to $f-1$. For $\Delta n$ and $\Delta m_{P}$ we easily get $\Delta n=D/\mathcal{A}=2B/\phi$ and $\Delta m_{P}=D\mu_{\mathrm{B}}s_{f}/\mathcal{A}=2B\mu_{\mathrm{B}}s_{f}/\phi$, so in general
$$A_{s}=\frac{-\mu+s\mu_{\mathrm{B}}B}{\phi}.\label{amp}$$
We are now in position to express the two SO as an infinite series, whose amplitude, frequency and phase are given by equations (\[freq\]), (\[phase\]) and (\[amp\]). We shall note $\omega\equiv\omega_{1}=\omega_{-1}=\mu^{2}/\alpha^{2}$ and $\Delta\equiv-\Delta_{1}=\Delta_{-1}=2\mu_{\mathrm{B}}/\mu$. Then we have
$$M_{SO}=\sum_{s=\pm1}A_{s}\sum_{p=1}^{\infty}\frac{1}{\pi p}\sin\left[2\pi p\omega\left(\frac{1}{B}+s\Delta\right)\right].\label{M sawtooth}$$
Equation (\[M sawtooth\]) gives the SO contribution to the MO. There is still another oscillatory contribution, which comes from the continuous oscillation in $\Omega_{C}$. This oscillation is not a SO, but an expression for it can be straightforward obtained noticing that $M_{osc}=-\mathcal{A}^{-1}\left(\partial\Omega_{C}^{osc}/\partial B\right)_{\mu}$. From equation (\[M sawtooth\]) we get that $\Omega_{C}^{osc}$ should be of the form $\Omega_{C}^{osc}=\sum_{s=\pm1}C_{s}\sum_{p=1}^{\infty}\left(\pi p\right)^{-2}\cos\left[2\pi p\omega\left(\frac{1}{B}+s\Delta\right)\right]$, where $C_{s}$ satisfies $M_{osc}=-\mathcal{A}^{-1}\left(\partial\Omega_{C}^{osc}/\partial B\right)_{\mu}$, so $2\omega C_{s}/\mathcal{A}B^{2}=-A_{s}$. Therefore the MO is given by
$$\begin{aligned}
\fl M_{osc} = \sum_{s=\pm1}\left\{ \left[\frac{B}{\omega\phi}\left(-\mu+\frac{3}{2}s\mu_{\mathrm{B}}B\right)\right]\sum_{p=1}^{\infty}\frac{1}{\left(\pi p\right)^{2}}\cos\left[2\pi p\omega\left(\frac{1}{B}+s\Delta\right)\right]\right.\nonumber\\
\left.+A_{s}\sum_{p=1}^{\infty}\frac{1}{\pi p}\sin\left[2\pi p\omega\left(\frac{1}{B}+s\Delta\right)\right]\right\}.\label{Mosc}\end{aligned}$$
Equation (\[Mosc\]) is in agreement with the result found in [@Sharapov2004], where the oscillating part of the magnetization is expressed as an infinite series. To obtain the total magnetization we also have to add the non-oscillatory contribution from both the valence and conduction band. Nevertheless, it can be shown [@Sharapov2004] that in graphene, at $\mu>0$, this contribution cancels and the total magnetization is just given by equation (\[Mosc\]). This result can be intuitively understood by noticing that the non-oscillatory contribution to $M$ comes from the variation of the degeneracy and energy as $B$ is changed (which gives a variation proportional to $\sqrt{B}$, for $D\varepsilon\sim B^{3/2}$ and thus $\delta D\varepsilon\sim B^{1/2}$). Ignoring the Zeeman splitting, the valence and conduction band energy levels are equal with opposite sign, so their total contribution is canceled. Consequently, in order to study the magnetization in graphene at $\mu>0$ it is sufficient to work with equation (\[Mosc\]). It is worth noting that in this formalism, the spin splitting due to the Zeeman effect is already taken into account in equation (\[Mosc\]), so there is no need to introduce it as a reduction factor. The spin splitting is usually neglected in the MO calculations, but we will see that it can have a noticeable effect in $M$.
Spin splitting in the MO at zero temperature
--------------------------------------------
We shall now analyze equation (\[Mosc\]) in more detail. First, notice that each spin has a different amplitude $A_{s}$, although from equation (\[amp\]) we have $\left|A_{1}-A_{-1}\right|=\mu_{\mathrm{B}}B/\phi$, which usually is very small. On the other hand, the MO are periodic in $1/B$ (as in the classical case), each spin peak with the same frequency $\omega=\mu^{2}/\alpha^{2}$, but with a phase difference $2\Delta=4\mu_{\mathrm{B}}/\mu$. These features can be seen in figure \[fig1\], where the MO given by equation (\[Mosc\]) is plot for (a) low $\mu$ and (b) high $\mu$. We observe that for $\mu\sim10$ meV the MO are practically seen as one unique oscillation, while for $\mu\sim100$ meV the spin splitting becomes noticeable. Moreover, we observe that at low $\mu$ there is a small curvature in the MO, which disappears at high $\mu$. This behavior can be explained by taking the ratio of amplitudes of both series, given by $A_{s}^{cos}=B\left(-\mu+3s\mu_{\mathrm{B}}B/2\right)/\omega\phi$ and $A_{s}^{sin}=\left(-\mu+s\mu_{\mathrm{B}}B\right)/\phi$. Considering that usually $B\mu_{\mathrm{B}}/\mu\ll1$, we get $A_{s}^{cos}/A_{s}^{sin}\simeq B/\omega$. Thus, if $\mu>0.1$ eV we have $\omega=\mu^{2}/\alpha^{2}>10$ T and $A_{s}^{cos}/A_{s}^{sin}<0.1B\left[\unit{T}\right]$, so unless $B$ is very high we have $A_{s}^{cos}/A_{s}^{sin}\ll1$. For smaller values of $\mu$ this would not be case, and for big $B$ one may even have $A_{s}^{cos}/A_{s}^{sin}>1$. This is seen in figure \[fig1\](a), where at low $\mu$ the cosine series in equation (\[Mosc\]) produces the small curvature in the MO. Nevertheless, we will not take $\mu$ this low for then the Zeeman effect becomes unobservable. Therefore, given that we will be mainly interested with the spin splitting in the MO, which is seen at high $\mu$, we shall directly neglect the first term in equation (\[Mosc\]). Then the MO reduce to equation (\[M sawtooth\]) and the sine series can be evaluated to obtain
$$M=\sum_{s=\pm1}\frac{A_{s}}{\pi}\arctan\left\{ \cot\left[\pi\omega\left(\frac{1}{B}+s\Delta\right)\right]\right\} .\label{M0}$$
![\[fig1\]Magnetization given by equation (\[Mosc\]), for (a) low $\mu$ and (b) high $\mu$, where $\mu$ is the Fermi energy.](fig1.eps)
We can calculate the amplitude of the spin splitting effect in the MO by taking into account the low point of the first peak and the high point of the second peak, as schematically indicated in figure \[fig2\]. We consider the two peaks corresponding to a LL $l$, so that the first peak is located at $1/B_{1}=l/\omega-\Delta$, while the second peak at $1/B_{2}=l/\omega+\Delta$. Then, from equation (\[M0\]) we get that at the low point of the first peak we have $M_{1}=A_{1}/2-A_{-1}\arctan\left[\cot\left(2\pi\omega\Delta\right)\right]/\pi$, while at the high point of the second peak we have $M_{2}=A_{1}\arctan\left[\cot\left(2\pi\omega\Delta\right)\right]/\pi-A_{-1}/2$. Thus the amplitude of splitting $\Delta M_{S}$ is given by
![\[fig2\]Scheme of two peaks due to the Zeeman effect, located at $1/B_1=l/\omega-\Delta$ and $1/B_2=l/\omega+\Delta$. We define the amplitude of splitting $\Delta M_s$ as shown, which measures the observation of the Zeeman effect in the MO. Notice also that the peaks have a phase difference $2\Delta=4\mu_{\mathrm{B}}/\mu$.](fig2.eps)
![\[fig3\]Plot of the parameters $\Delta M_{s}/A_{+}$ and $2\Delta\omega$ as a function of the Fermi energy, where $\Delta M_{s}$ is given by equation (\[Delta Ms\]) and $A_{+}\equiv\left|A_{1}+A_{-1}\right|/2=\mu/\phi$. The phase $\Delta$ and frequency $\omega$ are given by equations (\[freq\]) and (\[phase\]).](fig3.eps)
$$\Delta M_{s}=A_{+}\left(1-\frac{2}{\pi}\arctan\left[\cot\left(2\pi\omega\Delta\right)\right]\right),\label{Delta Ms}$$
where $A_{+}\equiv\left|A_{1}+A_{-1}\right|/2=\mu/\phi$. A better measurement of the effect is to take the relation between $\Delta M_{s}$ and the amplitude of the peaks, which can be approximated as $A_{+}$ (because $A_1\simeq A_{-1}$). On the other hand, we can also analyze the splitting peak width, which is given by the phase difference $2\Delta=4\mu_{\mathrm{B}}/\mu$. This width also affect the visualization of the spin splitting in $M$, for at low values of Fermi energy, $2\Delta$ becomes insignificant. Then we can compare this width with the oscillation period, given by $1/\omega=\alpha^{2}/\mu^{2}.$ In figure \[fig3\] it is plotted $\Delta M_{s}/A_{+}$ and $2\Delta\omega$ as a function of the Fermi energy. We see that for $\mu\lesssim100$ meV we have $\Delta M_{s}/A_{+}\sim10^{-2}$, while $2\Delta\omega$ is also very small, which makes the spin splitting effect practically unobservable. But when we get to higher Fermi energies, about $\mu\gtrsim250$ meV, we have $\Delta M_{s}/A_{+}\sim0.1$ so the effect starts to become more evident. Consequently, in order to account for the Zeeman effect in $M$ we will take Fermi energies above 200 meV. Notice also that $\Delta M_{s}/A_{+}$ does not depend on $B$, so the effect of the Zeeman splitting in the magnetization at zero temperature is independent of the magnetic field. Nevertheless, we shall see that the temperature makes the observation of the Zeeman splitting in the magnetization be strongly dependent of $B$.
MO at nonzero temperature
=========================
The effects of nonzero temperature can be taken into account in equation (\[M sawtooth\]) by introducing the temperature reduction factor $R_{T}$, which for these systems has the form [@Sharapov2004]
$$R_{T}=\frac{4\pi^{2}\mu pkT/\alpha^{2}B}{\sinh\left(4\pi^{2}\mu pkT/\alpha^{2}B\right)},\label{Rt}$$
where $\alpha=\upsilon_{F}\sqrt{2e\hbar}$. Thus the MO at nonzero temperature becomes
$$M_{T}=\sum_{s=\pm1}A_{s}\sum_{p=1}^{\infty}\frac{R_{T}}{\pi p}\sin\left[2\pi p\omega\left(\frac{1}{B}+s\Delta\right)\right].\label{M temperature}$$
In general, the series given by equation (\[M temperature\]) cannot be evaluated, but it can be simplified under some approximations. The usual one is to take the limit $B\rightarrow0$, in which case $1/\sinh\left(4\pi^{2}\mu pkT/\alpha^{2}B\right)\sim\exp\left(-4\pi^{2}\mu pkT/\alpha^{2}B\right)$. Nevertheless, this limit implies low magnetic field, which in turn makes it difficult to observe the MO. Instead we will impose the low temperature limit, but such that the MO are still observable.
To obtain an approximation for $M_{T}$ at low temperatures, it is convenient to start from the expression of the grand potential $\Omega$ at $T\neq0$ and $\mu>0$. It can be shown that in the absence of impurities, one can use the non-relativistic grand potential to obtain the magnetization [@Tabert2014]. Hence we start with
$$\Omega_{T}=-kT\int_{-\infty}^{\infty}\rho_{0}(\varepsilon)\ln\left[1+e^{\beta\left(\mu-\varepsilon\right)}\right]d\varepsilon,$$
where $\beta=1/kT$ and $\rho_{0}(\varepsilon)=D\sum_{m}\delta\left(\varepsilon-\varepsilon_{m}\right)$ is the density of states (DOS) in the pristine case, where the summation is to be done considering the valence and conduction band. In this way we can separate the contribution from both bands, so that $\Omega_{T}=\Omega_{V,T}+\Omega_{C,T}$, where in general $\Omega_{T}=-kT\sum_{\varepsilon}D\ln\left[1+e^{\beta\left(\mu-\varepsilon\right)}\right]$. Now, for $\mu>0$ and low temperature we always have $\beta\left(\mu-\varepsilon\right)\gg1$ for the valence band, so $\Omega_{V,T}\simeq\Omega_{V}\left(T=0\right)$ and $\partial\Omega_{V,T}/\partial B\simeq\partial\Omega_{V}\left(T=0\right)/\partial B$. Hence the valence band magnetization is not affected by the temperature under this conditions. Therefore, we shall omit it for the moment and continue with the conduction band grand potential
$$\Omega_{C,T}=-kT\sum_{m=0}^{\infty}D\ln\left[1+e^{\beta\left(\mu-\varepsilon_{m}\right)}\right]=\sum_{m=0}^{\infty}\Omega_{T}^{m},$$
where $\varepsilon_{m}=\alpha\sqrt{n_{m}B}-s_{m}\mu_{\mathrm{B}}B$ are the conduction energy levels, and we defined $\Omega_{T}^{m}=-kTD\ln\left[1+e^{\beta\left(\mu-\varepsilon_{m}\right)}\right]$. We have $\left(\partial\Omega_{T}^{m}/\partial B\right)_{\mu}=\Omega_{T}^{m}/B+D\left(\partial\varepsilon_{m}/\partial B\right)\left[1+e^{-\beta\left(\mu-\varepsilon_{m}\right)}\right]^{-1}$, so the conduction magnetization $M_{C,T}=-\mathcal{A}^{-1}\left(\partial\Omega_{C,T}/\partial B\right)_{\mu}$ is given by
$$M_{C,T}=\frac{KTD}{\mathcal{A}B}\sum_{m=0}^{\infty}\ln\left[1+e^{\beta\left(\mu-\varepsilon_{m}\right)}\right]-\frac{D}{\mathcal{A}}\sum_{m=0}^{\infty}\frac{\partial\varepsilon_{m}}{\partial B}\frac{1}{1+e^{-\beta\left(\mu-\varepsilon_{m}\right)}}.\label{Mt}$$
We now assume a Fermi energy $\varepsilon_{f}<\mu\leq\varepsilon_{f+1}$, so the conduction magnetization at zero temperature is $M_{C}=-\mathcal{A}^{-1}\left(\partial\Omega_{C}/\partial B\right)_{\mu}=-\mathcal{A}^{-1}\sum_{m=0}^{f}\left[D\left(\varepsilon_{m}-\mu\right)/B+D\left(\partial\varepsilon_{m}/\partial B\right)\right]$. Thus we can rewrite equation (\[Mt\]) as
$$\begin{aligned}
\fl M_{C,T} = M_{C}+\frac{KTD}{\mathcal{A}B}\sum_{m=0}^{f}\ln\left[1+e^{-\beta\left(\mu-\varepsilon_{m}\right)}\right]+\frac{D}{\mathcal{A}}\sum_{m=0}^{f}\frac{\partial\varepsilon_{m}}{\partial B}\frac{1}{1+e^{\beta\left(\mu-\varepsilon_{m}\right)}}\nonumber\label{Mt2}\\
+\frac{KTD}{\mathcal{A}B}\sum_{m=f+1}^{\infty}\ln\left[1+e^{\beta\left(\mu-\varepsilon_{m}\right)}\right]-\frac{D}{\mathcal{A}}\sum_{m=f+1}^{\infty}\frac{\partial\varepsilon_{m}}{\partial B}\frac{1}{1+e^{-\beta\left(\mu-\varepsilon_{m}\right)}}.\end{aligned}$$
We can simplify equation (\[Mt2\]) by noticing that the term with the logarithm is in general very small. Indeed, we always have $\ln\left[1+e^{-\beta\left(\mu-\varepsilon_{m}\right)}\right]\leq\ln2<1$ if $m\leq f$ and likewise $\ln\left[1+e^{\beta\left(\mu-\varepsilon_{m}\right)}\right]<\ln2$ if $m>f$. Furthermore, we have $kTD/\mathcal{A}B=2kT/\phi=2kTe/h\sim10^{-7}T\left[\unit{K}\right]\;\textrm{meV/T}{\textrm{\AA}}^2$, which for low $T$ is much smaller than the magnetization at zero temperature (see figure \[fig1\]). On the other hand, at low temperatures each exponential term in equation (\[Mt2\]) is in general very small unless $\mu$ is close to $\varepsilon_{m}$. Thus we can expand each $\varepsilon_{m}$ around $B_{m}$, where $B^{-1}_m=n_m/\omega+s_m\Delta$ and $\varepsilon_{m}\left(B_{m}\right)=\mu$, so $\left(\mu-\varepsilon_{m}\right)\simeq-\mu\left(B-B_{m}\right)/2B_{m}$. Moreover, $\left(\partial\varepsilon_{m}/\partial B\right)\simeq\left[\mu-s_{m}\mu_{\mathrm{B}}B\right]/2B$, so from equation (\[amp\]) we get $\mathcal{A}^{-1}\left(\partial\varepsilon_{m}/\partial B\right)\simeq-A_{s_{m}}/D.$ Finally, if we take into account the valence contribution to the magnetization (which we show it is not altered by the temperature), we have the total magnetization $M_{T}=M_{V,T}+M_{C,T}=M_{V}+M_{C,T}$. Thus from equation (\[Mt2\]) we get
![\[fig4\]Magnetization given by equation (\[MT\]), for $\mu=250$ meV. The red dashed line corresponds to the zero temperature case, while the solid blue line corresponds to (a) $T=0.1$ K and (b) $T=1$ K.](fig4.eps)
$$M_{T} \simeq M-\sum_{m=0}^{f}\frac{A_{s_{m}}}{1+e^{-\beta\mu\left(B-B_{m}\right)/2B_{m}}}+\sum_{m=f+1}^{\infty}\frac{A_{s_{m}}}{1+e^{\beta\mu\left(B-B_{m}\right)/2B_{m}}}.\label{Mt aprox}$$
where $M=M_{V}+M_{C}$ is the net magnetization at zero temperature, given by equation (\[M0\]). Equation (\[Mt aprox\]) corresponds to the case $\varepsilon_{f}<\mu\leq\varepsilon_{f+1}$, which implies $B_{f+1}\leq B<B_{f}$. Thus the low temperature effect is to introduce a factor proportional to $\left[1+e^{-\beta\mu\left(B-B_{m}\right)/2B_{m}}\right]^{-1}$ if $B<B_{m}$, or proportional to $\left[1+e^{\beta\mu\left(B-B_{m}\right)/2B_{m}}\right]^{-1}$ if $B>B_{m}$. In this way we can generalize equation (\[Mt aprox\]) to all $B$ and get the result (see the appendix for details)
$$\fl M_{T}=\sum_{s}\frac{A_{s}}{\pi}\arctan\left\{ \cot\left[\pi\omega\left(\frac{1}{B}+s\Delta\right)+\sum_{n}\frac{\pi}{1+e^{-\beta\mu\left(B-B_{n,s}\right)/2B_{n,s}}}\right]\right\},\label{MT}$$
where $B_{n,s}^{-1}=n/\omega-s\Delta$. In the practice it is sufficient to take only the $B_{n}$ between the range of magnetic fields considered.
In figure \[fig4\] it is show the magnetization given by equation (\[MT\]) for $\mu=250$ meV and temperatures (a) $T=0.1$ K, (b) $T=1$ K. We can see that as the temperature increases, the magnetization at $T\neq0$ broadens, similar to the Fermi-Dirac distribution with the DOS. Of course, this is exactly what is expected from the result given by equation (\[MT\]). Moreover, in figure \[fig4\](a) we see that at very low temperature $T=0.1$ K one can still appreciate the effect of the spin splitting in the magnetization, which results in a small bump around the two peaks. Nevertheless, as the temperature is increased, this bump disappears and the MO behave as maxima and minima around the center of the two peaks at zero temperature. Thus in this regime there would be no noticeable effect of the spin splitting in the magnetization. This is strongly dependent not only on the temperature, but also on the magnetic field and Fermi energy. In other words, for high magnetic fields, the temperature at which the spin splitting is negligible increases.
Spin splitting in the MO at nonzero temperature
-----------------------------------------------
An estimation for the temperature at which the spin splitting would not be observable in the MO can be obtained from equation (\[MT\]). We shall define $x\equiv1/B$, and consider two peaks at $x_{1}$ and $x_{2}$, corresponding to the same LL, with different spin, so in general $x_{1}=l/\omega-\Delta$ for the spin up, and $x_{2}=l/\omega+\Delta$ for the spin down (see equation (\[1/Bi\])). Then, between the two peaks, only the exponential terms corresponding to $x_{1}$ and $x_{2}$ would be appreciable, so for $x_{1}<x<x_{2}$ equation (\[MT\]) becomes
$$M_{T}=M-\frac{A_{1}}{1+e^{\beta\mu\left(x-x_{1}\right)/2x}}+\frac{A_{-1}}{1+e^{-\beta\mu\left(x-x_{2}\right)/2x}},\label{Mt spin}$$
where $M$ is given by equation (\[M0\]), and we have rewrite $\left(B-B_{m}\right)/2B_{m}=-\left(x-x_{m}\right)/2x$.
![\[fig5\]Spin splitting effect in the MO. In the left it is shown magnetization at zero temperature (red dashed line) and at different temperatures (blue solid line). In the right it is shown the first (orange solid line) and second (green dashed line) exponential in equation (\[Mt spin\]). The four cases correspond to the exponentials width being (a) $<\Delta$, (b) $=\Delta$, (c) $>\Delta$, (d) $=2\Delta$, where $2\Delta$ is the spin peaks separation.](fig5.eps)
From equation (\[Mt spin\]) we can see that the temperature effect behaves like a Fermi-Dirac distribution. The difference is that in this case our parameter is $x=1/B$, instead of the energy, and each exponential is center around the position of the peak $x_{m}$. Thus we can write $\beta\mu\left(x-x_{m}\right)/2x=\tilde{\beta}\left(x-x_{m}\right)$, where $\tilde{\beta}=1/k\tilde{T}\equiv\beta\mu/2x$ can be considered the corresponding *temperature* parameter, with defines the broadening of each exponential term in equation (\[Mt spin\]). This shows that the broadening depends not only on the real temperature $T$, but also on the Fermi energy $\mu$ and the magnetic field $x=1/B$. In this way we can get an estimation to how the temperature affects the observation of the spin splitting in the magnetization, which is dictated by how the width of the exponentials relates to the spin peaks separation $x_{2}-x_{1}=2\Delta$. This can be seen in figure \[fig5\], where for different temperatures it is shown the magnetization (left) and the exponentials that appear in equation (\[Mt spin\]) (right). Then we can identify four situations. In the first case, figure \[fig5\](a), the width is $<\Delta$ and one can clearly appreciate the spin splitting in the magnetization at $T\neq0$. The second case, figure \[fig5\](b), corresponds to the case when the width is $\Delta$, so the exponentials began to overlap. Then, for a width $>\Delta$ (but $<2\Delta$) in figure \[fig5\](c), the spin splitting effect starts to disappear, although one could still notice a small change of curvature in $M_{T}$. Finally, when the width is $\geq2\Delta$ as in figure \[fig5\](d), the spin splitting becomes unobservable in $M_{T},$ and one is left with what appears as one unique oscillation around the center $\left(x_{1}+x_{2}\right)/2$.
Now, for a Fermi-Dirac distribution of the form $\left[1+e^{(y-y_{0})/a}\right]^{-1}$, the width $w$ around $y_{0}$ is determined by the condition $1/\left[1+e^{w/a}\right]=\sigma$, where $\sigma\ll1$ is the appreciation considered. From our experience, it is sufficient to take $\sigma\sim10^{-2}$, which implies a width $w\sim5a$. Therefore, following figure \[fig5\](d), the critical temperature for which the spin splitting becomes unobservable satisfies $2\Delta\simeq5k\tilde{T}$. Given that $k\tilde{T}=2kTx/\mu$, and $2\Delta=4\mu_{\mathrm{B}}/\mu$, we get the temperature $T_{s}=2\mu_{\mathrm{B}}/5kx$. This result basically means that, in order of magnitude, the thermal energy $kT$ equals the spin splitting energy $2\mu_{\mathrm{B}}/x$. One should replace in $x$ the value at which the broadening is computed. If the peak at $x_{1}$ is considered, the corresponding exponential factor is $\tilde{\beta}\left(x-x_{1}\right)$, and the condition $2\Delta\simeq5k\tilde{T}$ implies to take $x=x_{2}$. On the other hand, if we take the peak $x_{2}$, we should take $x=x_{1}$. This gives two different temperatures $T_{s,1}\simeq2\mu_{\mathrm{B}}/5 Kx_{2}$ and $T_{s,2}\simeq2\mu_{\mathrm{B}}/5 Kx_{1}$, but $\left|T_{s,1}-T_{s,2}\right|=T_{s,1}\left(2\Delta/x_{1}\right)=T_{s,2}\left(2\Delta/x_{2}\right)$ with $\Delta/x_{i}\ll1$ for both cases. Thus in general $T_{s,1}\simeq T_{s,2}$, and we can take the average temperature $T_{s}=\left(T_{s,1}+T_{s,2}\right)/2$, so
$$T_{s}\simeq\frac{2}{5}\frac{\mu_{\mathrm{B}}B}{k},\label{Ts}$$
where $B=\omega/l$ is the center of the spin peaks. Equation (\[Ts\]) gives the maximum temperature to observe the spin splitting in the MO. Given that $B=\omega/l=\mu^{2}/l\alpha^{2}$, then $T_{s}$ is proportional to $\mu^{2}$, so high Fermi energy favors the observation of the spin splitting in the magnetization. It should be noted that the condition given by equation (\[Ts\]) has to be taken into account alongside with the need of high $\mu$ (about $\mu\gtrsim100$ meV, see figure \[fig1\]) to observe the spin splitting at zero temperature. Therefore, for practical terms, equation (\[Ts\]) should only be applied when $\mu\gtrsim100$ meV.
![\[fig6\]Spin temperature $T_s$, given by equation (\[Ts\]), as a function of the magnetic field $B$ (solid line) and for $\mu=250$ meV (scatter), where $B=\omega/l=\mu^2/l\alpha^2$ at the center of the spin peaks corresponding to the LL $l$.](fig6.eps)
In figure \[fig6\] it is shown the plot of equation (\[Ts\]) as a function of the magnetic field $B$ (solid line) and for $\mu=250$ meV (scatter), where $B=\omega/l=\mu^2/l\alpha^2$. We see that, in general, $T_{s}$ is very low unless $B$ is very high. For instance, at $\mu=250$ meV, we have that $B\sim1$ T gives $T_{s}\sim0.3$ K. Thus, even for high magnetic field, it requires a very low temperature to observe the spin splitting in the MO.
MO extrema shift
----------------
Once the temperature is such that $T>T_{s}$ in equation (\[Ts\]), the spin splitting becomes unobservable. Then the MO behave as if the spin splitting is neglected, with the oscillation been around $B=\omega/l$. This is just the usual behavior when the MO are studied in graphene without considering the Zeeman effect. Thus, because we shall take $T>T_{s}$ from now on, we will neglect the spin splitting. Then, what one observe is, essentially, that the MO are broadened and reduced as the temperature increases. This reduction depends on the temperature, as well as the magnetic field and Fermi energy. The higher the magnetic field, the more temperature it takes to reduce the MO, and vice versa. From this pattern we can analyze different parameters in the MO, in which case equation (\[MT\]) will prove useful.
First of all, notice that as the MO are broadened, the maxima and minima of the oscillations shift from $x_{n}\equiv B_{n}^{-1}=n/\omega$ (see equation (\[1/Bi\]), neglecting the spin splitting) to $x_{n}-p\delta$, where $p=1(-1)$ for the maxima (minima) and $\delta=\delta\left(T,B,\mu\right)$ is the shift parameter which in general depends on the three variables $T,\,B$ and $\mu$. For a fixed $\mu$, we can obtain the function $\delta$ by the extrema condition $\partial M_{T}/\partial x=0$, where $x=1/B$. Now, neglecting the spin splitting, equation (\[MT\]) becomes
$$M_{T}=-\frac{2\mu}{\pi\phi}\arctan\left[\cot\left(\pi\omega x+\sum_{n}\pi f_{n}\right)\right],\label{MTSP}$$
where we defined $f_{n}\equiv\left[1+e^{\beta\mu\left(x-x_{n}\right)/2x}\right]^{-1}$. Then, the extrema condition $\partial M_{T}/\partial x=0$ can be written as
$$1=\frac{\beta\mu}{8\omega x^{2}}\sum_{n}x_{n}\textrm{sech}^{2}\left[\frac{\beta\mu\left(x-x_{n}\right)}{4x}\right].\label{x extreme}$$
The $x=1/B$ that satisfy equation (\[x extreme\]) give all the new maxima and minima in the MO at $T\neq0$. Notice that at zero temperature $\beta\rightarrow\infty$ and equation (\[x extreme\]) implies $x\rightarrow x_{n}$, as expected. In general, for a peak at $x_{l}=l/\omega$, to obtain its shift it is sufficient to consider only the $n$ close to $l$ (say $l-2<n<l+2$) in the summation of equation (\[x extreme\]). The new extrema will be located at $x=x_{l}-p\delta_{l}$, where $\delta_{l}$ is the corresponding shift and $p=1(-1)$ for the maxima (minima). In this way we can rewrite equation (\[x extreme\]) as a function of $\delta_{l}$ and the considered peak $x_{l}$:
$$1 =\frac{\beta\mu}{8\omega\left(x_{l}-p\delta_{l}\right)^{2}}
\sum_{n}x_{n}\textrm{sech}^{2}\left[\frac{\beta\mu\left(x_{l}-x_{n}-p\delta_{l}\right)}{4\left(x_{l}-p\delta_{l}\right)}\right].\label{delta extreme}$$
![\[fig7\]Graphical solution of equation (\[delta extreme\]) for different LL $l$ and temperatures (a) $T=5$ K, (b) $T=10$ K. All cases correspond to $\mu=250$ meV and maxima shift, so $p=1$ in equation (\[delta extreme\]).](fig7.eps)
In figure \[fig7\] it is show the graphical solution of equation (\[delta extreme\]) for different values of temperature and $l$, for $\mu=250$ meV and $p=1$ (maxima shift).
As we can see, the shift $\delta$ clearly depends on the LL $l$ and thus on the magnetic field. It increases with the temperature and decreases with the magnetic field. This means that for non zero temperature, the MO are not anymore periodic in $1/B$. Indeed, $\delta$ as defined is always measured from the position of the peaks at zero temperature, whose distance between one another is always the same and identical to $1/\omega$ (the period). Consequently, if at $T\neq0$ the shift $\delta_{l}$ is not equal for all $l$, then the extrema separation will not be constant and thus not periodic as a function of $1/B$. Nevertheless, it should be notice that this effect is very small, because for close $l_{1}$ and $l_{2}$ one usually has $\left|\delta_{1}-\delta_{2}\right|\ll1/\omega$. On the other hand, in figure \[fig7\](b) we see that for the cases $l=30$ and $l=40$, the shift $\delta$ reaches an steady value of about $\delta\sim5.3\times10^{-3}\unit{T^{-1}}$. This means that the extrema in the MO will tend to be located between the peaks at zero temperature and the zeros in the magnetization. Indeed, the zeros in the MO occur at $B_{0n}^{-1}=\left(2n+1\right)/2\omega$. Thus, the distance between the peaks and the zeros is $\left(B_{n}^{-1}-B_{0n}^{-1}\right)=1/2\omega$, which is about $\sim1.05\times10^{-2}\unit{T^{-1}}$ for $\mu=250$ meV (as considered in figure \[fig7\]), so the midpoint distance is about $\sim5.3\times10^{-3}\unit{T^{-1}}$.
![\[fig8\] Numerical solution of equation (\[delta extreme\]) for $\delta_l$ as a function of the temperature, where $\mu=250$ meV and $B=1/x_l=\omega/l=1.01$ T (corresponding to LL $l=47$).](fig8.eps)
We can also analyze the shift as a function of the temperature, for a particular LL. This can be seen in figure \[fig8\], where for $\mu=250$ meV and $B=1/x_l=\omega/l=1.01$ T (corresponding to LL $l=47$), it is shown the numerical solution of equation (\[delta extreme\]) for $\delta_l$ as a function of the temperature. As expected, we see that a low temperature $\delta_l$ tends to zero, and increases with $T$. This behavior could be useful to measure temperature changes from the MO.
MO envelope
-----------
If, at a given temperature, $\delta$ is know as a function of $B$, then the envelope of the MO can be easily obtained. We start considering $x_{1}<x<x_{2},$ where $x_{1}$ and $x_{2}$ are two adjacent peaks at zero temperature, so in general $x_{1}=\left(l-1\right)/\omega$ and $x_{2}=l/\omega$. Then, from equation (\[MTSP\]) we can write
$$M_{T} \simeq M+\frac{2\mu}{\phi}\left(\sum_{n=0}^{l-1}\frac{1}{1+e^{\beta\mu\left(x-n/\omega\right)/2x}}\right.
\left.-\sum_{n=l}^{\infty}\frac{1}{1+e^{-\beta\mu\left(x-n/\omega\right)/2x}}\right),\label{Mt env}$$
where $M$ is given by equation (\[M0\]), which without spin splitting is $M=-2\mu\arctan\left[\cot\left(\pi\omega x\right)\right]/\pi\phi$. To obtain the positive envelope corresponding to the maxima in the MO, we just have to eliminate the oscillatory part in (\[Mt env\]). For $M$, this can be done by replacing $x=l/\omega-\delta$, the position at which the maximum occurs, where $\delta$ is the shift obtained from equation (\[delta extreme\]). Thus from equation (\[M0\]) we get $$M_{\delta}=\frac{2\text{\ensuremath{\mu}}}{\pi\phi}\arctan\left[\cot\left(\pi\omega\delta\right)\right]\label{M delta},$$ where we defined $M_{\delta}\equiv M\left(l/\omega-\delta\right)$. Notice that $M_{\delta}$ is independent of $l$. On the other hand, for the exponentials in equation (\[Mt env\]), we need to replace $x=l/\omega-\delta$ only in the numerator. Thus, defining $m=l-n$ we get
$$\begin{aligned}
\fl \left[\sum_{m=1}^{l}\left(\frac{1}{1+e^{\beta\mu\left(m/\omega-\delta\right)/2x}}-\frac{1}{1+e^{\beta\mu\left(m/\omega+\delta\right)/2x}}\right)-\frac{1}{1+e^{\beta\mu\delta/2x}}\right.\nonumber\\
-\left.\sum_{m=l+1}^{\infty}\frac{1}{1+e^{\beta\mu\left(m/\omega+\delta\right)/2x}}\right].\label{exp up}\end{aligned}$$
The last term in equation (\[exp up\]) can be neglected because it is very small if $x_{1}<x<x_{2}$. In this way, regrouping the exponentials we get the following expression for the positive envelope
![\[fig9\] For $T=5$ K and $\mu=250$ meV: (a) Numerical solution of equation (\[delta extreme\]), where $\delta\left(x\right)$ is obtained by replacing $x=l/\omega$. The points were fitted with a Boltzmann function. (b) MO with the positive $E_{+}$ and negative $E_{-}=-E_{+}$ envelope, where $E_{+}$ is given by equation (\[E+\]). The shift function $\delta(B)$ used was obtained from the fitting of (a).](fig9.eps)
$$\begin{aligned}
\fl E_{+} =\frac{2\mu}{\pi\phi}\left\{ \arctan\left[\cot\left(\pi\omega\delta\right)\right]-\frac{\pi}{1+e^{\beta\mu\delta/2x}}\right.\nonumber\\
\left.+\sum_{m\geq1}\frac{\pi\sinh\left[\beta\mu\delta/2x\right]}{\cosh\left[\beta\mu\delta/2x\right]+\cosh\left[\beta\mu m/2\omega x\right]}\right\}.\label{E+} \end{aligned}$$
It should be notice that although this envelope was originally obtained for $x_{1}<x<x_{2}$, the expression given by equation (\[E+\]) is directly generalized and valid for all magnetic fields. The negative envelope is just $E_{-}=-E_{+}.$ Although there is a summation involved in $E_{+},$ in the practice it is sufficient to consider only the first few terms. In fact, the terms $m>1$ give corrections only if we are at high temperature and/or low magnetic field, at which the MO are small.
The envelope given by equation (\[E+\]) depends on the shift $\delta$, which in general depends on the magnetic field, as it was shown in the last section. A numerical solution for $\delta$ can be obtained from equation (\[delta extreme\]). This gives $\delta_{l}$ as a function of $l$, from which the function $\delta\left(x\right)$ can be obtained by replacing $x=l/\omega$ and fitting the points. In figure \[fig9\] it is shown this procedure for $T=5$ K and $\mu=250$ meV, where $\delta$ was fitted with a Boltzmann function. Notice that the figure \[fig9\](a) is in agreement with the figure \[fig7\], where $\delta$ increases with $l$ and thus with $1/B=x=l/\omega$, and it tends to the limit $\delta\sim5.3\times10^{-3}\unit{T^{-1}}$. Once obtained $\delta\left(x\right)$, the positive envelope as a function of $x$ is given by equation (\[E+\]), while the negative envelope is $E_{-}=-E_{+}$. This can be seen in figure \[fig9\](b), where the MO and the envelopes are shown. We consider only the terms $m=1,2$ in the sum of equation (\[E+\]), which shows that only the few first terms are needed.
Conclusions
===========
We analyzed the magnetic oscillations (MO) in pristine graphene, under a perpendicular magnetic field and taking into account the Zeeman effect. For a constant Fermi energy $\mu>0$ and zero temperature we showed that, due to the Zeeman effect, the MO consist of two sawtooth peaks, each corresponding to one spin. Both peaks have the same frequency, but different amplitude and phase. We then studied the condition to observe these spin peaks in the MO, where it is shown that it requires high Fermi energy, about $\mu\gtrsim0.1$ eV.
At $T\neq0$ K, the temperature effect is usually considered by introducing a reduction factor. Nevertheless, the resulting infinite series cannot be evaluated, and can only be approximated for special cases like low $B$, which in turn difficult the observation of the MO. Hence, we took another route by going back to the grand potential. From this we obtained that the MO at low $T$ can be expressed as the MO at zero temperature, plus small correction functions. These functions are Fermi-Dirac like, each centered around the MO peaks at zero temperature. Moreover, they are very small unless the magnetic field is close to the corresponding peaks, which can be useful from a practical sense. Using this expression we then analyzed how the temperature affects the observation of the spin splitting in the MO. We show that, in order of magnitude, the observation is possible when the thermal energy $kT$ is less than the spin splitting energy $2\mu_{\mathrm{B}}B$.
We also analyzed the shift of the MO extrema as the temperature increases. We obtained an expression which was solved numerically, showing that the shift increases with the temperature and decreases with the magnetic field, implying that the MO are not anymore periodic at $T\neq0$. This behavior could be useful to measure temperature changes from the MO. For instance, one possibility would be a graphene device which measures the MO around a particular magnetic field. Then, by analyzing the extrema shift from its known value at zero temperature, one can infer the temperature. Finally, we obtained an analytical expression for the MO envelope, which in turn depends on the shift of the extrema.
This paper was partially supported by grants of CONICET (Argentina National Research Council) and Universidad Nacional del Sur (UNS) and by ANPCyT through PICT 2014-1351. Res. N° 270/15. N°: 2014-1351, and PIP 2014-2016. Res. N° 5013/14. Código: 11220130100436CO research grant, as well as by SGCyT-UNS., J. S. A. and P. J. are members of CONICET., F. E. acknowledge research fellowship from this institution.
Appendix {#appendix .unnumbered}
========
We shall prove that equation (\[MT\]) is in agreement with equation (\[Mt aprox\]). First of all, we recall the following properties for $x\in\mathbb{R}$:
$$\eqalign{\cot x =\tan\left(\pi/2-x\right) \cr
\arctan\left[\tan\left(x\right)\right] =x-\pi\mathrm{floor}\left\{ \frac{x+\pi/2}{\pi}\right\}} ,\label{properties}$$
where $\mathrm{floor}\left\{ \right\} $ is the floor function. Thus, taking only one spin, we can write equation (\[MT\]) as
$$\fl \frac{M_{T,s}}{A_{s}} =\frac{1}{2}-\omega\left(\frac{1}{B}+s\Delta\right)-\sum_{m}g_{m}\label{Mt aprox 4}
-\mathrm{floor}\left\{ 1-\omega\left(\frac{1}{B}+s\Delta\right)-\sum_{m}g_{m}\right\},$$
where we have defined $g_{m}=\left[1+e^{-\beta\mu\left(B-B_{m}\right)/2B_{m}}\right]^{-1}$, with $B_{m}^{-1}=n_m/\omega-s_m\Delta$. Considering $\varepsilon_{f}<\mu\leq\varepsilon_{f+1}$, which implies $B_{f+1}\leq B<B_{f}$, we have
$$\label{cases}
g_m=\cases{< 1/2&for $m \leq f$\\
\geq 1/2&for $m\geq f+1$\\}$$
Furthermore, at low temperatures $g_{m}\ll1$ if $m<f$, while $g_{m}\rightarrow1$ if $m>f+1$. These considerations, along with the properties of the floor function, implies that for $B_{f+1}\leq B<B_{f}$, equation (\[Mt aprox 4\]) can be written as
$$\begin{aligned}
\fl\frac{M_{T,s}}{A_{s}} = \frac{1}{2}-\omega\left(\frac{1}{B}+s\Delta\right)-\mathrm{floor}\left\{ 1-\omega\left(\frac{1}{B}+s\Delta\right)\right\} \label{Mt aprox 5}\nonumber\\
-\sum_{m\leq f}\frac{1}{1+e^{-\beta\mu\left(B-B_{m}\right)/2B_{m}}}+\sum_{m\geq f+1}\frac{1}{1+e^{\beta\mu\left(B-B_{m}\right)/2B_{m}}}.\end{aligned}$$
Consequently, from equation (\[M0\]) and the properties given by equation (\[properties\]), equation (\[Mt aprox 5\]) becomes
$$M_{T,s}=M_{s}-\sum_{m\leq f}\frac{A_{s}}{1+e^{-\beta\mu\left(B-B_{m}\right)/2B_{m}}}
+\sum_{m\geq f+1}\frac{A_{s}}{1+e^{\beta\mu\left(B-B_{m}\right)/2B_{m}}},$$
so equation (\[MT\]) effectively reduces to equation (\[Mt aprox\]) if $\varepsilon_{f}<\mu\leq\varepsilon_{f+1}$.
References {#references .unnumbered}
==========
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
Suppose that $A$ is a semi-simple and commutative Banach algebra. In this paper we try to characterize the character space of the Banach algebra $C_{\rm{BSE}}(\Delta(A))$ consisting of all BSE-functions on $\Delta(A)$ where $\Delta(A)$ denotes the character space of $A$. Indeed, in the case that $A=C_0(X)$ where $X$ is a non-empty locally compact Hausdroff space, we give a complete characterization of $\Delta(C_{\rm{BSE}}(\Delta(A)))$ and in the general case we give a partial answer.
Also, using the Fourier algebra, we show that $C_{\rm{BSE}}(\Delta(A))$ is not a $C^*$-algebra in general. Finally for some subsets $E$ of $A^*$, we define the subspace of BSE-like functions on $\Delta(A)\cup E$ and give a nice application of this space related to Goldstine’s theorem.
address: ' $^{1}$ Department of Mathematics and Statistics, Faculty of Basic Sciences and Engineering, Gonbad Kavous University, P.O.Box 163, Gonbad Kavous, Iran.'
author:
- 'Mohammad Fozouni$^1$$^{*}$'
title: 'On character space of the algebra of BSE-functions'
---
[Sahand Communications in Mathematical Analysis (SCMA) Vol. $\cdots$ No. $\cdots$(2014), $\cdots$-$\cdots$\
<http://scma.maragheh.ac.ir>]{}\
------------------------------------------------------------------------
width
------------------------------------------------------------------------
width
Introduction and Preliminaries
==============================
Suppose that $A$ is a semi-simple commutative Banach algebra and $\Delta(A)$ is the character space of $A$, i.e., the space of all non-zero homomorphisms from $A$ into $\mathbb{C}$.
A bounded continuous function $\sigma$ on $\Delta(A)$ is called a BSE-function if there exists a constant $C>0$ such that for each $\phi_{1},...,\phi_{n}\in\Delta(A)$ and complex numbers $c_{1},...,c_{n}$, the inequality $$\left|\sum_{i=1}^{n}c_{i}\sigma(\phi_{i})\right|\leq C\left\|\sum_{i=1}^{n}c_{i}\phi_{i}\right\|_{A^{*}}$$ holds. For each $\sigma\in C_{\mathrm{BSE}}(\Delta(A))$ we denote by $\|\sigma\|_{\mathrm{BSE}}$ the infimum of such $C$. Let $C_{\mathrm{BSE}}(\Delta(A))$ be the set of all $\mathrm{BSE}$-functions. We have a good characterization of $C_{\mathrm{BSE}}(\Delta(A))$ as follows:
\[1\] $C_{\mathrm{BSE}}(\Delta(A))$ is equal to the set of all $\sigma\in C_{b}(\Delta(A))$ for which there exists a bounded net $\{x_{\lambda}\}$ in $A$ with $\lim_{\lambda}\phi(x_{\lambda})=\sigma(\phi)$ for all $\phi\in \Delta(A)$
see [@Takahasi Theorem 4 (i)].
Using the above characterization one can see that $C_{\rm{BSE}}(\Delta(A))$ is unital if and only if $A$ has a bounded weak approximate identity in the sense of Lahr and Jones. We recall that a net $\{x_{\alpha}\}$ in $A$ is called a bounded weak approximate identity (b.w.a.i) for $A$ if $\{x_{\alpha}\}$ is bounded in $A$ and $$\lim_{\alpha}\phi(x_{\alpha}a)=\phi(a)\hspace{0.5cm}(\phi\in \Delta(A), a\in A),$$ or equivalently, $\lim_{\alpha}\phi(x_{\alpha})=1$ for each $\phi\in \Delta(A)$.
Also, Theorem \[1\], gives the following definition of $\|\cdot\|_{\rm{BSE}}$: $$\begin{aligned}
\|\sigma\|_{\rm{BSE}}=\inf\{ & \beta>0 : \exists \{x_{\lambda}\} \text{ in A with } \|x_{\lambda}\|\leq \beta, \\
& \lim_{\lambda}\phi(x_{\lambda})=\sigma(\phi)\ (\phi\in \Delta(A)) \}.\end{aligned}$$
The theory of BSE-algebras for the first time introduced and investigated by Takahasi and Hatori; see [@Takahasi] and two other notable works [@Kaniuth; @Kamali2]. In [@Kamali2], the authors answered to a question raised in [@Takahasi]. Examples of BSE-algebras are the group algebra $L^1(G)$ of a locally compact abelian group $G$, the Fourier algebra $A(G)$ of a locally compact amenable group $G$, all commutative $C^*$-algebras, the disk algebra, and the Hardy algebra on the open unit disk. We recall that a commutative and without order Banach algebra $A$ is a type I-BSE algebra if $\widehat{M(A)}=C_{\rm{BSE}}(\Delta(A))=C_b(\Delta(A))$, where $M(A)$ denotes the multiplier algebra of $A$ and $\widehat{M(A)}$ denotes the space of all $\widehat{T}$ which defined by $\widehat{T}(\varphi)\varphi(x)=\widehat{T(x)}(\varphi)$ for all $\varphi\in \Delta(A)$. Note that $x\in A$ should satisfies $\varphi(x)\neq 0$.
In this paper, we give a partial characterization of the character space of $C_{\rm{BSE}}(\Delta(A))$ where $A$ is a semi-simple commutative Banach algebra. Indeed, we show that if $A$ has a b.w.a.i and $C_{\rm{BSE}}(\Delta(A))$ is an ideal in $C_b(\Delta(A))$, then $$\Delta(C_{\rm{BSE}}(\Delta(A)))=\overline{\Delta(A)}^{w^*}.$$
Also, we give a negative answer to this question;\
Whether $(C_{\rm{BSE}}(\Delta(A)), \|\cdot\|_{\rm{BSE}})$ is a $C^*$-algebra? At the final section of this paper we study the space of BSE-like functions on subsets of $A^*$ which containing $\Delta(A)$ and as an application of this space we give a nice relation with Goldstine’s theorem.
Character space of $C_{\mathrm{BSE}}(\Delta(A))$
================================================
In view of [@Takahasi Lemma 1], $C_{\rm{BSE}}(\Delta(A))$ is a semi-simple commutative Banach algebra. So, the character space of $C_{\rm{BSE}}(\Delta(A))$ should be non-empty and one may ask: Is there a characterization of $\Delta(C_{\rm{BSE}}(\Delta(A)))$ for an arbitrary Banach algebra $A$?
In the sequel of this section we give a partial answer to this question. Let $X$ be a non-empty locally compact Hausdroff space and put $$C_{\mathrm{BSE}}(X):=C_{\mathrm{BSE}}(\Delta(C_{0}(X))).$$
To proceed further we recall some notions. Let $X$ be a non-empty locally compact Hausdroff space. A function algebra (FA) on $X$ is a subalgebra $A$ of $C_b(X)$ that seperates strongly the points of $X$, that is, for each $x, y\in X$ with $x\neq y$, there exists $f\in A$ with $f(x)\neq f(y)$ and for each $x\in X$, there exists $f\in A$ with $f(x)\neq 0$. A Banach function algebra (BFA) on $X$ is a function algebra $A$ on $X$ with a norm $\|\cdot\|$ such that $(A, \|\cdot\|)$ is a Banach algebra. A topological space $X$ is completely regular if every non-empty closed set and every singleton disjoint from it can be separated by continuous functions.
Let $X$ be a non-empty locally compact Hausdroff space. Then $C_{\mathrm{BSE}}(X)$ is a unital [BFA]{} and its character space is homeomorphic to $\overline{X}^{w^{*}}$, that is, $\Delta(C_{\mathrm{BSE}}(X))=\overline{\{\phi_{x} : x\in X\}}^{w^*}$.
By [@Takahasi Lemma 1], $C_{\mathrm{BSE}}(X)$ is a subalgebra of $C_{b}(X)$ and $\|\cdot\|_{\rm{BSE}}$ is a complete algebra norm. Since $C_{0}(X)$ has a bounded approximate identity, $C_{\rm{BSE}}(X)$ is unital. So, for each $x\in X$, there exists $f\in C_{\mathrm{BSE}}(X)$ with $f(x)\neq 0$. On the other hand, using the Urysohn lemma, for each $x, y\in X$ with $x\neq y$ one can see that there exists $f\in C_{\mathrm{BSE}}(X)$ such that $f(x)\neq f(y)$.
Finally, since $X$ is a locally compact Hausdroff space, it is completely regular by [@AK Corollary 2.74]. On the other hand, by [@Takahasi Theorem 3], we know that $C_{0}(X)$ is a type I-BSE algebra. Therefore, $C_{\textrm{BSE}}(X)=C_{b}(\Delta(C_{0}(X)))=C_{b}(X)$. Also, for every $f\in C_{\rm{BSE}}(X)$, by the remark after Theorem 4 of [@Takahasi], we have $\|f\|_{X}\leq \|f\|_{\rm{BSE}}$. Also, by the Open mapping theorem there exists a positive constant $M$ such that $\|f\|_{\rm{BSE}}\leq M\|f\|_X$. So, $C_{\rm{BSE}}(X)$ and $C_b(X)$ are topologically isomorphic, and so $\Delta(C_{\textrm{BSE}}(X))$ and $\Delta(C_{b}(X))$ are homeomorphic. Now, by using [@Kaniuth2 Theorem 2.4.12], we have $$\Delta(C_{\textrm{BSE}}(X))=\Delta(C_{b}(X))=\overline{X}^{w^{*}}=\overline{\{\phi_{x} : x\in X\}}^{w^{*}}.$$
\[Rem1\] In general for a commutative Banach algebra $A$, we have the following conditions concerning the character space of $C_{\rm{BSE}}(\Delta(A))$:
1. If $C_{\rm{BSE}}(\Delta(A))=C_{b}(\Delta(A))$, then $$\Delta(C_{\rm{BSE}}(\Delta(A)))=\overline{\Delta(A)}^{w^{*}}.$$ Examples of Banach algebras $A$ satisfying $C_{\rm{BSE}}(\Delta(A))=C_{b}(\Delta(A))$ are finite dimensional Banach algebras and commutative $C^*$-algebras; see the remark on page 609 of [@TH2]. Also, see [@Takahasi Lemma 2] for a characterization of Banach algebras $A$ for which satisfying $C_{\rm{BSE}}(\Delta(A))=C_{b}(\Delta(A))$.
2. If $A$ has a b.w.a.i, then $C_{\rm{BSE}}(\Delta(A))$ is unital and so $\Delta(C_{\rm{BSE}}(\Delta(A)))$ is compact and hence it is $w^*$-closed. On the other hand, we know that $$\Delta(A)\subseteq \Delta(C_{\rm{BSE}}(\Delta(A))),$$ in the sense that for each $\varphi\in \Delta(A)$, $f_{\varphi}: C_{\rm{BSE}}(\Delta(A))\longrightarrow \mathbb{C}$ defined by $f_{\varphi}(\sigma)=\sigma(\varphi)$ is an element of $\Delta(C_{\rm{BSE}}(\Delta(A)))$. Note that $f_\varphi\neq 0$, since in this case $C_{\rm{BSE}}(\Delta(A))$ is unital and $f_\varphi(1)=1$. So $$\overline{\Delta(A)}^{w^{*}}\subseteq \overline{\Delta(C_{\rm{BSE}}(\Delta(A)))}^{w^*}=\Delta(C_{\rm{BSE}}(\Delta(A))).$$
3. On can see that if $(B, \|\cdot\|_B)$ is a Banach algebra which containing the Banach algebra $(C,\|\cdot\|_C)$ as a two-sided ideal, then every $\varphi\in \Delta(C)$ extends to one $\widetilde{\varphi}\in \Delta(B)$. Now, let $C=C_{\rm{BSE}}(\Delta(A))$ and $B=C_{b}(\Delta(A))$. If $C$ is an ideal in $B$, then $$\Delta(C_{\rm{BSE}}(\Delta(A)))=\Delta(C)\subseteq \Delta(B)=\overline{\Delta(A)}^{w^*}.$$
4. Suppose that $B$ is a commutative semi-simple Banach algebra such that $\Delta(B)$ is compact. Then $B$ is unital; see [@Kaniuth2 Theorem 3.5.5]. Now, If $A$ has no b.w.a.i, then $$\Delta(C_{\rm{BSE}}(\Delta(A)))\neq \overline{\Delta(A)}^{w^{*}}.$$ Because if $\Delta(C_{\rm{BSE}}(\Delta(A)))=\overline{\Delta(A)}^{w^{*}},$ by using the above assertion, $C_{\rm{BSE}}(\Delta(A))$ is unital, since $\overline{\Delta(A)}^{w^{*}}=\Delta(C_{b}(\Delta(A)))$ is compact and $C_{\rm{BSE}}(\Delta(A))$ is a semi-simple commutative Banach algebra. Therefore, $A$ has a b.w.a.i which is impossible.
5. If $A$ has a b.w.a.i and $C_{\rm{BSE}}(\Delta(A))$ is an ideal of $C_b(\Delta(A))$, then using parts (ii) and (iii), we have $$\Delta(C_{\rm{BSE}}(\Delta(A)))=\overline{\Delta(A)}^{w^*}.$$
$\left(C_{\rm{BSE}}(\Delta(A)), \|\cdot \|_{\rm{BSE}}\right)$ is not a $C^{*}$-algebra
======================================================================================
The theory of $C^*$-algebras is very fruitful and applied. As an advantage of this theory, especially in Harmonic Analysis, one can see the $C^*$-algebra approach for defining a locally compact quantum group; see [@KV00]. So, verifying the Banach algebras from a $C^*$-algebraic point of view is very helpful. In this section, using a result due to Kaniuth and Ülger in [@Kaniuth], we show that $\left(C_{\rm{BSE}}(\Delta(A)), \|\cdot \|_{\rm{BSE}}\right)$ is not a $C^{*}$-algebra in general. On the other hand, there is a question which left open that, under what conditions on $A$, $\left(C_{\rm{BSE}}(\Delta(A)), \|\cdot \|_{\rm{BSE}}\right)$ is a $C^{*}$-algebra?
In the sequel for each locally compact group $G$, let $A(G)$ denote the Fourier algebra and $B(G)$ denote the Fourier-Stieltjes algebra introduced by Eymard; see [@pier §19]. Also, let $\widehat{G}$ denote the dual group of $G$ and $M(G)$ denote the Measure algebra; see [@Dales §3.3]. For the convenience of reader we give the definitions of $A(G)$ and $B(G)$ as follows: Let $G$ be a locally compact group. Suppose that $A(G)$ denotes the subspace of $C_{0}(G)$ consisting of functions of the form $u=\sum_{i=1}^{\infty}f_{i}\ast \widetilde{g_{i}}$ where $f_{i}, g_i\in L^{2}(G)$, $\sum_{i=1}^{\infty}||f_{i}||_{2}||g_{i}||_{2}<\infty$ and $\widetilde{f}(x)=\overline{f(x^{-1})}$ for all $x\in G$. The space $A(G)$ with the pointwise operation and the following norm is a Banach algebra, $$||u||_{A(G)}=\inf\{\sum_{i=1}^{\infty}||f_{i}||_{2}||g_{i}||_{2}: u=\sum_{i=1}^{\infty}f_{i}\ast \widetilde{g_{i}}\},$$ which we call it the Fourier algebra. It is obvious that for each $u\in A(G)$, $||u||\leq ||u||_{A(G)}$ where $||u||$ is the norm of $u$ in $C_{0}(G)$.
Now let $\Sigma$ denote the equivalence class of all irreducible representations of $G$. Then $B(G)$ consisting of all functions $\phi$ of the form $\phi(x)=<\pi(x)\xi, \eta>$ where $\pi\in \Sigma$ and $\xi, \eta$ are elements of $H_{\pi}$, the Hilbert space associated to the representation $\pi$. It is well-known that $A(G)$ is a closed ideal of $B(G)$. Also, recall that an involutive Banach algebra $A$ is called a $C^*$-algebra if its norm satisfies $\|aa^*\|=\|a\|^2$ for each $a\in A$. We refer the reader to [@Murphy] to see a complete description of $C^*$-algebras.
In the following remark we give the main result of this section.
In general $\left(C_{\rm{BSE}}(\Delta(A)), \|\cdot \|_{\rm{BSE}}\right)$ is not a $C^{*}$-algebra, that is, there is not any involution “\*” on $C_{\rm{BSE}}(\Delta(A))$ such that $$\|\sigma^*\sigma \|_{\rm{BSE}}=\|\sigma \|_{\rm{BSE}}^2 \ \ \ \forall \sigma\in C_{\rm{BSE}}(\Delta(A)).$$
Because we know that every commutative $C^{*}$-algebra is a BSE-algebra. For a non-compact locally compact Abelian group $G$ take $A=A(G)$. By [@Kaniuth Theorem 5.1], we know that $C_{\rm{BSE}}(\Delta(A))=B(G)$ and for each $u\in B(G)$, $\|u\|_{B(G)}=\|u\|_{\rm{BSE}}$. But $B(G)=M(\widehat{G})$ and it is shown in [@Takahasi] that $M(\widehat{G})$ and hence $B(G)$ is not a BSE-algebra. Therefore, $\left(C_{\rm{BSE}}(\Delta(A)), \|\cdot \|_{\rm{BSE}}\right)$ is not a $C^{*}$-algebra.
As the second example, let $G$ be a locally compact Abelian group. It is well-known that $C_{\rm{BSE}}(\Delta(L^1(G)))$ is isometrically isomorphic to $M(G)$, where $L^1(G)$ denotes the group algebra; see the last remark on page 151 of [@Takahasi]. On the other hand, by the Gelfand-Nimark theorem we know that every commutative $C^*$-algebra should be symmetric. But in general $M(G)$ is not symmetric, i.e., the formula $\widehat{\mu^*}(\xi)=\overline{\widehat{\mu}}(\xi)$ does not hold for every $\xi\in \Delta(M(G))$. For example if $G$ is non-discrete then by [@Rudin Theorem 5.3.4], $M(G)$ is not symmetric and hence fails to be a $C^*$-algebra.
It is a good question to characterize Banach algebras $A$ for which $\left(C_{\rm{BSE}}(\Delta(A)), \|\cdot \|_{\rm{BSE}}\right)$ is a $C^*$-algebra.
BSE-functions on subsets of $A^*$
=================================
Suppose that $A$ is a Banach algebra and $E\subseteq A^*\setminus \Delta(A)$. A complex-valued bounded continuous function $\sigma$ on $\Delta(A)\cup E$ is called a BSE-like function if there exists an $M>0$ such that for each $f_1, f_2, f_3, \ldots f_n\in \Delta(A)\cup E$ and complex numbers $c_1, c_2, c_3, \ldots, c_n$, $$\label{eq1}
\left|\sum_{i=1}^{n}c_{i}\sigma(f_{i})\right|\leq M\left\|\sum_{i=1}^{n}c_{i}f_{i}\right\|_{A^*}.$$ We show the set of all the BSE-like functions on $\Delta(A)\cup E$ by $C_{\rm{BSE}}(\Delta(A), E)$ and let $\|\sigma\|_{\rm{BSE}}$ be the infimum of all $M$ satisfying relation \[eq1\]. Obviously, $C_{\rm{BSE}}(\Delta(A), E)$ is a linear subspace of $C_b(\Delta(A)\cup E)$ and we have $$\{\sigma_{|\Delta(A)} : \sigma\in C_{\rm{BSE}}(\Delta(A), E)\}\subseteq C_{\rm{BSE}}(\Delta(A)).$$
Clearly, $\iota_{A}(A)\subseteq C_{\rm{BSE}}(\Delta(A), E)$ where $\iota_A: A\longrightarrow A^{**}$ is the natural embedding. For $a\in A$, we let $\widehat{a}=\iota_A(a)$ and $\widehat{A}=\iota_A(A)$.
To proceed further, we recall the Helly theorem.
\[Helly\] ([Helly]{}) Let $(X, \|\cdot\|)$ be a normed linear space over $\mathbb{C}$ and let $M>0$. Suppose that $x_{1}^{*},\ldots, x_{n}^{*}$ are in $X^{*}$ and $c_{1}^{*},\ldots, c_{n}^{*}$ are in $\mathbb{C}$. Then the following are equivalent:
1. for all $\epsilon>0$, there exists $x_{\epsilon}\in X$ such that $\|x_{\epsilon}\|\leq M+\epsilon$ and $x_{k}^{*}(x_{\epsilon})=c_{k}$ for $k=1,\ldots, n$.
2. for all $a_{1},\ldots, a_{n}\in \mathbb{C}$, $$\left|\sum_{i=1}^{n}a_{i}c_{i}\right|\leq M\left\|\sum_{i=1}^{n}a_{i}x_{i}^{*}\right\|_{X^*}.$$
See [@Larsen2 Theorem 4.10.1].
As an application of Helly’s theorem, we give the following characterization which is similar to [@Takahasi Theorem 4 (i)].
\[Th:2\] $C_{\rm{BSE}}(\Delta(A), E)$ is equal to the set of all $\sigma\in C_{b}(\Delta(A)\cup E)$ for which there exists a bounded net $\{x_{\alpha}\}$ in $A$ with $\lim_{\alpha}f(x_{\alpha})=\sigma(f)$ for all $f\in \Delta(A)\cup E$.
Suppose that $\sigma\in C_{b}(\Delta(A)\cup E)$ is such that there exists $\beta<\infty$ and a net $\{x_{\alpha}\}\subseteq X$ with $\|x_{\alpha}\|<\beta$ for all $\alpha$ and $\lim_{\alpha}f(x_{\alpha})=\sigma(f)$ for all $f\in \Delta(A)\cup E$. Let $f_{1}, \ldots, f_{n}$ be in $\Delta(A)\cup E$ and $c_1, \ldots, c_n$ be complex numbers. Then we have $$\begin{aligned}
\left|\sum_{i=1}^{n}c_{i}\sigma(f_{i})\right|\leq & \left|\sum_{i=1}^{n}c_{i}f_{i}(x_{\alpha})\right|+ \left|\sum_{i=1}^{n}c_{i}(f_{i}(x_{\alpha})-\sigma(f_{i}))\right|\\
\leq & \beta\left\|\sum_{i=1}^{n}c_{i}f_{i}\right\|+\sum_{i=1}^{n}|c_{i}||f_{i}(x_{\alpha})-\sigma(f_{i})|\end{aligned}$$ Taking the limit with respect to $\alpha$, we conclude that $\sigma\in C_{\rm{BSE}}(\Delta(A), E)$.
Conversely, let $\sigma\in C_{\rm{BSE}}(\Delta(A), E)$. Suppose that $\Lambda$ is the net consisting of all finite subsets of $\Delta(A)\cup E$. By Helly’s theorem, for each $\epsilon>0$ and $\lambda\in \Lambda$, there exists $x_{(\lambda,\epsilon)}\in A$ with $\|x_{(\lambda,\epsilon)}\|\leq \|\sigma\|_{\rm{BSE}}+\epsilon$ and $f(x_{(\lambda,\epsilon)})=\sigma(f)$ for all $f\in \lambda$. Clearly, $\{(\lambda, \epsilon) : \lambda\in \Lambda, \epsilon>0\}$ is a directed set with $(\lambda_1,\epsilon_1)\preceq (\lambda_2,\epsilon_2)$ iff $\lambda_1\subseteq \lambda_2$ and $\epsilon_1\leq \epsilon_2$. Therefore, we have $$\lim_{(\lambda,\epsilon)}f(x_{(\lambda,\epsilon)})=\sigma(f)\ \ \ \ (f\in \Delta(A)\cup E).$$
As an application of Theorem \[Th:2\], if $E=A^*\setminus \Delta(A)$, then one can see that $\overline{\widehat{A}}^{w^*}=A^{**}$, i.e., we conclude Goldstine’s theorem. That is, $\widehat{A}$ with the $w^*$-topology of $A^{**}$ is dense in $A^{**}$.
We say that $A$ has a b.w.a.i respect to $E$ if there exists a bounded net $\{x_{\alpha}\}$ in $A$ with $$\lim_{\alpha}f(x_{\alpha})=1\ \ \ (f\in \Delta(A)\cup E).$$ Using Theorem \[Th:2\],one can check that $1\in C_{\rm{BSE}}(\Delta(A), E)$ if and only if $A$ has a b.w.a.i respect to $E$.
We conclude this section with the following question.\
[**Question**]{} Is $C_{\rm{BSE}}(\Delta(A), E)$ a commutative and semi-simple Banach algebra? If it is what is its character space?
[**Acknowledgment.**]{} The author wish to thank the referee for his$\backslash$her suggestions. The author partially supported by a grant from Gonbad Kavous University.
[XX]{} C. D. Aliprantis and K. C. Border, [*Infinite Dimensional Analysis*]{}, Springer-Verlag Berlin Heidelberg, edition 3, 2006.
H. G. Dales, [*Banach Algebras and Automatic Continuity*]{}, Clarendon Press, Oxford, 2000.
Z. Kamali and M. L. Bami, [*The [B]{}ochner-[S]{}choenberg-[E]{}berlein Property for ${L}^{1}(\mathbb{R}^{+})$*]{}, J. Fourier Anal. Appl., 2014, 20, 2, 225–233.
E. Kaniuth, [*A Course in Commutative [B]{}anach Algebras*]{}, Springer Verlag, Graduate texts in mathematics, 2009.
E. Kaniuth and A. Ülger, [*The [B]{}ochner-[S]{}choenberg-[E]{}berlein property for commutative [B]{}anach algebras, especially [F]{}ourier and [F]{}ourier-[S]{}tieltjes algebras*]{}, Trans. Amer. Math. Soc., 2010, 362, 4331–4356. J. Kustermans and S. Vaes, [*Locally compact quantum groups*]{} Ann. Sci. Ecole Norm. Sup. (4) 33 (2000), 837–934. R. Larsen, [*Functional Analysis: an introduction*]{}, Marcel Dekker, New York, 1973. G. J. Murphy, [*$C^*$-Algebras and Operator Theory*]{}, Academic Press Inc, 1990. J. P. Pier, [*Amenable Locally Compact Groups*]{}, Wiley Interscience, New York, 1984. W. Rudin, [*Fourier Analysis on Groups*]{}, Wiley-Interscience, New York, 1962. S.- E. Takahasi and O. Hatori, [*Commutative [B]{}anach algebras which satisfy a [B]{}ochner-[S]{}choenberg-[E]{}berlein-type theorem*]{}, Proc. Amer. Math. Soc., 1990, 110, 149–158.
S.- E. Takahasi and O. Hatori, [*Commutative Banach algebras and [BSE]{}-inequalities*]{}, Math. Japonica, 1992, 37, 4, 607–614.
------------------------------------------------------------------------
width
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Predictive models for software projects’ characteristics have been traditionally based on project-level metrics, employing only little developer-level information, or none at all. In this work we suggest novel metrics that capture temporal and semantic developer-level information collected on a per developer basis. To address the scalability challenges involved in computing these metrics for each and every developer for a large number of source code repositories, we have built a designated repository mining platform. This platform was used to create a metrics dataset based on processing nearly 1000 highly popular open source GitHub repositories, consisting of 147 million LOC, and maintained by 30,000 developers. The computed metrics were then employed to predict the corrective, perfective, and adaptive maintenance activity profiles identified in previous works. Our results show both strong correlation and promising predictive power with $R^2$ values of $0.83$, $0.64$, and $0.75$. We also show how these results may help project managers to detect anomalies in the development process and to build better development teams. In addition, the platform we built has the potential to yield further predictive models leveraging developer-level metrics at scale.'
author:
-
-
bibliography:
- 'bibliography.bib'
title: 'Using Temporal and Semantic Developer-Level Information to Predict Maintenance Activity Profiles'
---
Software Maintenance; Software Metrics; Mining Software Repositories; Predictive Models; Human Factors;
Introduction
============
Forecasting maintenance activities performed in a source code repository could help practitioners reduce uncertainty and improve cost-effectiveness [@swanson1976dimensions] by planning ahead and pre-allocating resources towards source code maintenance. Maintenance activity profiles of software projects have been therefore a subject of research in numerous works [@swanson1976dimensions; @mockus2000identifying; @meyers1988; @lientz1978characteristics]. In this work we adopt the maintenance activity categories **corrective**, **perfective**, and **adaptive** as defined by Mockus et al. [@mockus2000identifying]:
- Corrective: fault fixing.
- Perfective: code structure / system design improvements.
- Adaptive: new feature introduction.
and put forth previously unexplored **temporal and semantic developer-level metrics**, which we then utilize to study the corrective, perfective and adaptive maintenance activity profiles on a developer-level granularity.
We seek to gain better understating of how personal characteristics such as commit patterns, commit frequencies, project join and departures dates, etc., impact the maintenance activities of individual developers, as well as the projects they work on. Moreover, given a software project, we consider its maintenance activity profiles as an aggregation of the maintenance activity profiles of all developers’ that have taken part in its development. Therefore, predictions made on a developer-level (i.e., for all individual developers in the project), can be used to reason about, and derive project-level predictions.
Studying developer-level impact requires sufficient and sufficiently fine-grained data, as well as the computational power to process it. This work is therefore driven by two main factors that have been trending up for the past decade:
- Big Code [@bigCode]: the availability of large source code corpora via open source.
- Big Data [@diebold2012origin] ecosystem: the availability of tools capable of processing extremely large data volumes.
The combination of the two has created unprecedented opportunities to collect and process an enormous volume of source code, and provide insights that were previously exponentially harder, or even impossible to obtain [@raychev2015predicting].
Related Work
============
Maintenance activity profiles and the application thereof have been the subject of numerous works dealing with fault prediction models, commit classification, software change recommendations, and more [@schach2003determining; @lientz1978characteristics; @johnson1988designing; @mockus2000identifying; @hassan2009predicting; @zimmermann2005mining]. While the precise distribution of maintenance activities is inconclusive [@schach2003determining], their classification into the corrective, perfective and adaptive categories has been a common practice. Our work therefore seeks to explore how these maintenance activity categories relate to the semantic and temporal developer-level metrics we define, and whether such metrics can be used to build effective predictive models for the corrective, perfective and adaptive maintenance activity profiles.
Research Question
=================
***RQ: How do temporal and semantic developer-level metrics relate to developers maintenance activity profiles?***\[rq2\] {#rq-how-do-temporal-and-semantic-developer-level-metrics-relate-to-developers-maintenance-activity-profilesrq2 .unnumbered}
=========================================================================================================================
To the best of our knowledge, this work is the first to explore temporal and semantic *developer-level* metrics (see Table \[tab:developerMetrics\]), such as the number of distinct semantic changes performed by each developer, the mean number of distinct semantic changes in a given developer’s commits, mean time between commits and others. Moreover, we explore these metrics at large scale, and analyze nearly 1000 different repositories that consist of dozens of millions of commits in total. Our large scale study was conducted using a VCS mining platform we have built to enable large scale analysis of version control systems (VCS). Our platform leverages Spark [@zaharia2010spark], an industrial cutting edge data processing framework.
We used the GQM approach [@basili1992software] to derive the questions, and then the metrics that are needed in order to answer the research question in a measurable way. The developer-level metrics we measured are listed in Table \[tab:developerMetrics\].
[m[3.7cm]{}|m[4.3cm]{}]{} $\operatorname{Commits}_{repo}$ &\
$\operatorname{Muse}_{repo}$ &\
$\operatorname{DeveloperVersatility}_{repo}$ &\
$\operatorname{MeanCommitV}_{repo}$ &\
$\operatorname{VersatilityLevel}_{repo}$ &\
& Developer’s join date (first recorded commit) expressed as the number of days since project’s first observed commit\
& The duration in days, of the period between the developer’s first and last commit dates.\
$\operatorname{MTBC}_{repo}$ & ean ime etween developer’s ommits, in days.\
Developers who have committed changes (contributed) to multiple source code repositories are considered as if they were different individuals.
Background {#sec:dataCollection}
==========
Fluri et al. [@fluri2006classifying] put forth a taxonomy of semantic source code changes for object-oriented programming languages (OOPLs), and Java in particular. This taxonomy consists of 47 different change types, such as *statement\_delete*, *statement\_insert*, *statement\_update*, *removed\_class*, *additional\_class*, *return\_type\_change* and so on. The computation of these change types from source code files was later implemented by a tool named “ChangeDistiller” [@gall2009change].
We embrace this taxonomy and define the notion of versatility based measures for developers and commits as defined in Table \[lst:versatilityFormula\].
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Given a source code repository $repo$ and a developer $dev$ that has committed code to $repo$:
$\boldsymbol{\operatorname{Commits}_{repo}(dev)} = \{\text{all commits by developer $dev$ to repository $repo$}\}$
$\boldsymbol{\operatorname{CommitVSet}_{repo}(commit)}:= \{ \text{all semantic source code changes in } commit \}$
$\boldsymbol{\operatorname{DeveloperVSet}_{repo}(dev)} := \bigcup\limits_{commit \in \operatorname{Commits}_{repo}(dev) } \operatorname{CommitVSet}_{repo}(commit)$
$\boldsymbol{\operatorname{CommitVersatility}_{repo}(commit)} := \mid \operatorname{CommitVSet}_{repo}(commit) \mid $
$\boldsymbol{\operatorname{DeveloperVersatility}_{repo}(dev)} := \mid \operatorname{DeveloperVSet}_{repo}(dev) \mid$
$\boldsymbol{\operatorname{Muse}_{repo}(dev)} := \max\limits_{commit \in \operatorname{Commits}_{repo}(dev) } \operatorname{CommitVersatility}_{repo}(commit) $
$\boldsymbol{\overline{\operatorname{CommitVersatility}}_{repo}(dev)} :=
(\sum\limits_{commit \in \operatorname{Commits}_{repo}(dev) } \operatorname{CommitVersatilisy}_{repo}(commit) ) * \frac{1}{\mid \operatorname{Commits}_{repo}(dev)\mid } $
$\boldsymbol{\operatorname{VersatilityLevel}_{repo}(dev)} := \mid \bigcup\limits_{commit \in \operatorname{Commits}_{repo}(dev) } \{S | S = \operatorname{CommitVSet}_{repo}(commit) \} \mid$
Note that: $ \operatorname{DeveloperVersatility}_{repo}(dev) \geq
\operatorname{Muse}_{repo}(dev)$
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
For example, if developer $Alice$ performed 2 commits:
$commit_1$ 2 x *stmt\_insert*, 1 x *stmt\_update*
------------ ----------------------------------------
$commit_2$ 3 x *stmt\_delete*
\
The intuition behind these versatility based measures is capturing the number of different fine-grained semantic change types present in commits made by individual developers.
Metrics Computation {#eval}
===================
To provide reproducible results, we sought to base our empirical study on publicly (and freely) available data. We chose GitHub as the data source for this work due to its prevalence among the repository hosting services. Candidate repositories were selected according to the following criteria:
1. used the Java programming language
2. had more than 100 stars (i.e. more than 100 users had “liked” these repositories)
3. had more than 60 forks (i.e., more than 60 users had “copied” these repositories for their own use)
4. had their code updated since 2016-01-01 (i.e., these repositories were active)
5. were created before 2015-01-01 (i.e., these repositories had been around for at least $\sim$1.5 years)
6. had size over 2MB (i.e. these repositories were beefy)
The result set consists of 1000 repositories, the maximum limit as stated by GitHub’s documentation [@gitHubSerchLimit]. Due to various technical reasons our final dataset consisted of 979 unique Git repositories, where the average number of developers per project was 32, and the average project age was 4.2 years. These repositories were then cloned and processed by our VCS mining platform
First, to distill semantic source code changes as per the taxonomy defined by Fluri et al. (see also section \[sec:dataCollection\]), our VCS mining platform repeatedly applied the ChangeDistiller ([@gall2009change; @fluri2008discovering; @martinez2013automatically]) on every two consecutive revisions of every Java file in every repository in our result set. This stage yielded 30 million semantic source code change instances. Then, all the semantic source code changes were aggregated using the key (developer-id, repository-id), forming data bins from which the temporal and semantic developer-level metrics (see Table \[tab:developerMetrics\]) were computed. To classify developer’s commits into the corrective, perfective, and adaptive categories, our VCS mining platform used methods similar to [@mockus2000identifying; @fischer2003populating; @sliwerski2005changes], and searched for indicative keywords in the commit’s comment field. Keyword matching was boosted by using common techniques such as stemming and case-folding, the keywords are listed in table \[tab:classificationWords\].
Corrective *fix, esolv, clos, handl, issue, defect, bug, problem, ticket*
------------ ------------------------------------------------------------------------------------------------------------
Perfective *refactor, re-factor, reimplement, re-implement, design, replac, modify, updat, upgrad, cleanup, clean-up*
Adaptive *add, new, introduc, implement, implemented, extend, feature, support*
: Keywords for classifying maintenance activities[]{data-label="tab:classificationWords"}
Results {#sec:results}
=======
We use generalized regression modeling (GLM) [@mccullagh1989generalized] in the R statistical environment [@R] to explore our dataset and build predictive models. The predictive models were trained on randomly chosen $90\%$ of the repositories in the dataset, while the remaining $10\%$ were used for validation and measuring goodness of fit.
In the rest of this section we present the predictive models (see Table \[tab:regressionModels\]) for the developer-level corrective, perfective and adaptive maintenance activity profiles. To predict the profile of a maintenance activity category $\operatorname{MA}_c \in \{\operatorname{Corrective}, \operatorname{Perfective}, \operatorname{Adaptive} \}$ for a developer $dev$ the following formula was used: where $model_{\operatorname{MA}_c}$ is the regression model for $\operatorname{MA}_c$, $\operatorname{Const_{MA_c}}$ is the constant in $model_{\operatorname{MA}_c}$, and $\operatorname{coeff}_{\operatorname{MA}_c}^{P_i}$ is the coefficient of predictor $P_i$ in $model_{\operatorname{MA}_c}$ as specified in table \[tab:regressionModels\]. $P_i(dev)$ is the value of predictor $P_i$ for a given developer $dev$.
All predictors were log transformed to alleviate skewed data, a common practise when dealing with software metrics [@shihab2012exploration; @camargo2009towards]. The standard error is specified in parenthesis underneath the estimate for each predictor. Figure \[fig:predictionPlots\] presents predictions results for 150 developers randomly selected from the test dataset.
![Prediction graphs for corrective, perfective and adaptive (left to right) maintenance activity profiles for 150 developers randomly selected form the test dataset. Actual values in blue, predicted ones are in red.[]{data-label="fig:predictionPlots"}](prediction-plots.png){width="50.00000%" height="3cm"}
For each maintenance activity category $\operatorname{MA}_c$, the X axis is a running developer-id and the Y axis is the number of the developer’s commits of category $\operatorname{MA}_c$. For each developer, we plot in blue the number of her commits classified as $\operatorname{MA}_c$ by the commit message classification algorithm (see section \[eval\]), and overlay it in red with the number of commits predicted to be of category $\operatorname{MA}_c$ by the GLM (see Table \[tab:regressionModels\]).
For all three maintenance activity profiles, the vast majority of both temporal and versatility based developer metrics were statistically significant with *p-value* $<0.01$. Metrics that were not statistically significant were excluded from the predictive models (represented by the empty cells in table \[tab:regressionModels\]) to improve prediction quality.
\[1\][>p[\#1]{}]{}
[m[0.3cm]{}m[3.5cm]{}P[1cm]{}P[1cm]{}P[1cm]{}]{} & &\
\
& **Corrective** (1) & **Perfective** (2) & **Adaptive** (3)\
\
$(P_1)$ & $log(\operatorname{Commits}_{repo})$ & 0.797 & 0.572 & 0.503\
& & (0.010) & (0.020) & (0.015)\
\
$(P_2)$ &$log(\operatorname{Muse}_{repo})$ & 0.171 & $-$0.288 & $-$0.135\
& & (0.010) & (0.020) & (0.013)\
\
$(P_3)$ & $log(\operatorname{MTBC}_{repo})$ & 0.012 & $-$0.018 &\
& & (0.002) & (0.004) &\
\
$(P_4)$ & & 0.014 & & $-$0.021\
& & (0.001) & & (0.001)\
\
$(P_5)$ & & 0.028 & & 0.033\
& & (0.009) & & (0.013)\
\
$(P_6)$ & & 0.030 & $-$0.050 & $-$0.018\
& & (0.002) & (0.005) & (0.002)\
\
$(P_7)$ & $log(\operatorname{DeveloperVersatility}_{repo})$ & $-$0.205 & 0.394 & 0.243\
& & (0.010) & (0.025) & (0.013)\
\
$(P_8)$ & $log(\operatorname{VersatilityLevel}_{repo})$ & 0.181 & 0.483 & 0.437\
& & (0.012) & (0.025) & (0.017)\
\
& Constant & $-$0.986 & $-$3.092 & $-$1.462\
& & (0.019) & (0.048) & (0.020)\
& **$R^2$** & **0.832** & **0.640** & **0.759**\
& Observations & 27,850 & 27,850 & 27,850\
In all three models, $\operatorname{Commits}_{repo}$ (the total number of commits made by a developer to the given repository) was the most powerful predictor, and accounted for a great of deal the high $R^2$ values.
**Corrective profile** (table \[tab:regressionModels\], column 1): In contrast to other predictors, *Versatility${_d}$* has a negative coefficient indicating that developers with higher *Versatility${_d}$* values are likely to perform less corrective commits given that other predictors remain fixed. The fact that *Muse* and *Versatility${_d}$* are both versatility based metrics, yet have opposite signs in this model, supports our assumption that *Muse* and *Versatility$_d$* capture different kinds of information. In addition, it is also evident that developers who commit less frequently (higher MTBC), join the project later, remain active for longer, and have more commits with distinct change type patterns, are likely to have a higher corrective profile (i.e., perform more corrective commits).
**Perfective profile** (see Table \[tab:regressionModels\], column 2): Similarly to the corrective model, the signs of *Versatility${_d}$* and *Muse* are opposite, but in contrast to the former, it is now *Muse* that has a negative sign. This indicates that developers with higher *Muse* values are likely to perform less perfective commits given that other predictors remain fixed. Also, developers who commit more frequently (lower MTBC), and remain active for shorter time (lower ContribDuration), but have more commits with distinct change type patterns (i.e., higher values for DistinctChangeTypeSets) are likely to have a higher perfective profile (i.e., perform more perfective commits).
**Adaptive profile** (see Table \[tab:regressionModels\], column 3): The adaptive model is more similar to the perfective one than to the corrective one, with *Versatility* having a positive sign and *Muse* a negative sign. Adaptive commits are likely to favour developers with lower Muse, who join the project earlier (lower ContribStartRel), remain active for shorter time (lower ContribDuration), have more commits with distinct change type patterns, and have a greater versatility (as defined in Table \[lst:versatilityFormula\]).
Discussion & Applications {#sec:discussion}
=========================
**Identifying anomalies in development process.** The manager of a large software project should aim to control and manage its maintenance activity profiles. Monitoring for unexpected spikes in maintenance activity profiles and investigating the reasons (root cause) behind them would assist managers and other stakeholders to plan ahead and identify areas that require additional resource allocation. For example, lower corrective profiles could imply that developers are neglecting bug fixing. Higher corrective profiles could imply an excessive bug count. Finding the root cause in cases of significant deviations from predicted values may reveal essential issues whose removal can improve projects’ health. Similarly, exceptionally well performing projects can also be a good subject for investigation in order to identify positive patterns.
**Improving development team’s composition.** Building a successful software team is hardly a trivial task as it involves a delicate balance between technological and human aspects [@gorla2004should; @guinan1998enabling]. We believe that developers’ maintenance activity profiles could assist in composing a more balanced team. We conjecture that composing a team that heavily favors a particular maintenance activity over the others could lead to an unbalanced development process and adversely affect the team’s ability to meet typical requirements such as developing a sustainable number of product features, adhering to quality standards, and minimizing technical debt so as to facilitate future changes.
Threats to validity {#sec:threatsToValidity}
===================
***Threats to Statistical Conclusion Validity*** is the degree to which conclusions about the relationship among variables based on the data are reasonable. Our results are based on nearly 30,000 observations, and the predictors are statistically significant with *p-value* $<0.01$.
***Threats to Construct Validity*** consider the relationship between theory and observation, in case the measured variables do not measure the actual factors.
- . While the method we used is commonly practiced ([@mockus2000identifying; @fischer2003populating; @sliwerski2005changes]), our experiments show it may be sensitive to the choice of keywords used as indicative for the various maintenance activity types.
- . ChangeDistiller ([@gall2009change; @fluri2008discovering]) was used to extract semantic changes from the VCS. Unfortunately, like any other software, it may not be immune to bugs and malfunctions.
- . To compute developer-level metrics we used a novel VCS mining platform we had built to support this study. While we invested great effort into testing it to ensure its proper functionality, it may not be immune to bugs and malfunctions.
***Threats to External Validity*** consider the generalization of our findings.
- . All analyzed commits were in the Java programming language. It is possible that developers who use other programming languages, have different maintenance activity patterns which have not been explored in the scope of this work.
- . The repositories studied in this paper were all popular open source projects from GitHub ([@gitHub]). It may be the case that developers’ maintenance activity profiles are different in an open source environment when compared to other environments.
- . We intentionally selected the most popular, data rich repositories. This could limit our results to developers and repositories of high popularity, and potentially skew the perspective on characteristics found only in less popular repositories and their developers.
Conclusions and Future Work
===========================
We have demonstrated that the developer-level metrics we have defined are statistically significant, and can be successfully used to model and predict the corrective, perfective and adaptive maintenance activity profiles with promising $R^2$ values of $0.83$, $0.64$ and $0.75$ respectively. Our work is based on studying Big Code ([@bigCode]), and involved processing nearly 1000 highly popular open source GitHub repositories, comprising a corpus of 147 million LOC, maintained by 30,000 developers, spread over 2.5 million revisions and 30 million individual code change instances.
We believe that considering a project’s characteristics as an aggregation of the characteristics of all the developers’ that have taken part in its development, and modeling the former using the latter, is a useful technique. It might, for example, assist in predicting further project characteristics such as fault potential, which has traditionally relied on project-level metrics [@mccabe1976complexity; @menzies2007data; @bell2011does; @nagappan2005use; @hassan2009predicting; @ostrand2011predicting] or only limited developer-level metrics [@bell2013limited; @matsumoto2010analysis; @shihab2012industrial; @jiang2013personalized] (e.g., numeric representations of experience, number of developers that changed a particular file or method).
In light of the promising results, we are in the process of conducting further studies that involve developer-level metrics. For example, we are working on comparing project-level and developer-level maintenance activity profiles. Preliminary results indicate that the *variance* in developer-level maintenance activity profiles is much greater than it is in the project-level ones.
We believe that the combination of Big Code and fined grained developer-level metrics, can open the door to a new range of applications and empirical studies. A particularly compelling direction is fusing more data sources such as bug tracking systems, social Q&A sites (e.g., StackOverflow [@stackOverflow]) and others, to study how these types of data relate to developers’ characteristics such as commit patterns, temporal activity, fault potential, etc. Augmented with developer surveys to validate the relations, we hope these studies could shed new light on our understanding of software development and evolution.
Acknowledgements {#acknowledgements .unnumbered}
================
We thank Ms Ilana Gelernter and the Tel Aviv University’s statistical counseling service for the help with the statistical analysis, and Dr. Boris Levin for comments that greatly improved the manuscript. This research was supported by THE ISRAEL SCIENCE FOUNDATION, grant No. 476/11.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We summarize several semi-phenomenological approaches to estimate the internal energy of one-component-plasma (OCP) in two (2D) and three (3D) dimensions. Particular attention is given to a hybrid approach, which reproduces the Debye-H$\ddot{\text{u}}$ckel asymptote in the limit of weak coupling, the ion sphere (3D) and ion disc (2D) asymptotes in the limit of strong coupling, and provides reasonable interpolation between these two limits. More accurate ways to estimate the internal energy of 2D and 3D OCP are also discussed. The accuracy of these analytic results is quantified by comparison with existing data from numerical simulations. The relevance of the KTHNY theory in locating melting transition in 2D OCP is briefly discussed.'
author:
- 'Sergey A. Khrapak'
- 'Alexey G. Khrapak'
title: 'Internal energy of the classical two- and three-dimensional one-component-plasma'
---
Introduction
============
The one-component-plasma (OCP) is an idealized system of identical point-like charges immersed in a uniform (rigid) neutralizing background of opposite charge[@Brush66; @Baus80; @Fortov06]. This model is of considerable interest from the fundamental point of view and has wide interdisciplinary applications, including ionized matter in white dwarfs, interiors of heavy planets, alkali metals, colloidal suspensions, and complex (dusty) plasmas[@Fortov06; @Brilliantov98; @Fortov04; @Fortov05; @Ivlev12]. In addition, the OCP represents a very important example of classical systems of interacting particles with extremely soft interactions (the limit opposite to hard-sphere interactions) and as such it plays significant role in condensed matter research.
In the three-dimensional (3D) case the interaction between the charged particles is described by the conventional Coulomb potential $$\label{Coulomb}
V(r)=Q^2/r,$$ where $Q$ is the particle charge and $r$ is the distance between two particles. The system is then characterized by the coupling parameter $\Gamma = Q^2/aT$, where $T$ is the temperature (in energy units), $a = (4\pi n/3)^{-1/3}$ is the (3D) Wigner-Seitz radius, and $n$ is the particle density. Thermodynamic properties of this system have been extensively studied in numerical simulations [@Hansen73; @Slattery80; @Stringfellow90; @Farouki94; @Hamaguchi96; @Dubin99; @Caillol99; @Caillol10].
In the two-dimensional case (2D) two different systems are actually referred to as the OCP. The first is characterized by the conventional 3D Coulomb interaction potential (\[Coulomb\]), but the particle motion is restricted to a 2D surface. This system has been used as a first approximation for the description of electron layers bound to the surface of liquid dielectrics and of inversion layers in semi-conductor physics[@Baus80; @Fortov06]. It has also some relevance to colloidal and complex (dusty) plasma mono-layers in the regime of week screening[@Fortov06; @Fortov04; @Fortov05; @Ivlev12]. This system is characterized by the same coupling parameter as in 3D, except 2D Wigner-Seitz radius is used, $a=(\pi n)^{-1/2}$, where $n$ is now the 2D density. Thermodynamics of these systems has also been studied in the literature [@Totsuji78; @Gann79].
There is another systems also referred to as the 2D OCP, in which the interaction potential is defined via the 2D Poisson equation and scales logarithmically with distance. The logarithmic potential, corresponding to the interaction of infinite charged filaments, is often employed to model interactions between vortices in thin-film superconductors. The 2D OCP has received considerable attention [@Caillol82; @Leeuw82; @Choquard83; @Radloff84] because of various field theoretical models[@Baus80] and existence of exact analytic solutions for some special cases [@Jancovici81; @Alastuey81]. The interaction potential between two particles follows from the solution of the 2D Poisson equation around a central test particle and reads $$V(r)=-Q^2\ln(r/L),$$ where $L$ is an arbitrary scaling length. It is common [@Caillol82] to set $L=a$, where $a$ is the 2D Wigner-Seitz radius. The thermodynamic of this system depends on the coupling parameter, $\Gamma=Q^2/T$, which is [*density independent*]{} (in contrast to Coulomb interactions in 3D and 2D).
The qualitative dependence of the OCP properties on the coupling strength is identical in 3D and 2D. As $\Gamma$ increases, the OCP shows a transition from a weakly coupled gaseous regime ($\Gamma\ll 1$) to a strongly coupled fluid regime ($\Gamma\gg 1$) and crystallizes at some $\Gamma_{\rm m}$ (the subscript “m” refers to melting). In the 3D case the stable crystalline phase is formed by the body-centered-cubic (bcc) lattice. The transition occurs at $\Gamma_{\rm m}\simeq170 - 175$ [@Farouki94; @Hamaguchi96; @Dubin99; @Khrapak14]. In the 2D case with the Coulomb interaction, numerical simulations located the transition into triangular lattice near $\Gamma_{\rm m}\simeq 125 \pm 15$ [@Gann79]. Experiments with a classical two-dimensional sheet of electrons yielded $\Gamma_{\rm m}\simeq 137 \pm 15$ [@Grimes79]. For the logarithmic interaction in 2D, numerical simulations and theory predict that the triangular lattice is thermodynamically favorable for $\Gamma\gtrsim 130 - 140$ [@Caillol82; @Leeuw82; @Choquard83; @Radloff84]. The closeness of $\Gamma_{\rm m}$ values for these two different 2D systems is likely a coincidence, because the definitions of the coupling strength are different for logarithmic and Coulomb interactions. At even higher $\Gamma$, the glass transition has been predicted for 3D OCP [@TanakaPRA_1987; @YazdiPRE_2014]. This can also be a scenario for 2D OCP, but we are not aware of any work in this direction.
Thermodynamic properties of the OCP (in both 2D and 3D) have been extensively studied over decades and accurate numerical results as well as their fits are available in the literature. Nevertheless, there has also been considerable continuous interest in deriving physically motivated analytical estimates or bounds on the thermodynamic quantities (in particular, internal energy) of the OCP. For example, analytical approaches of various complexity and accuracy have been discussed in Refs. [@Brilliantov98; @Caillol82; @Leeuw82; @Khrapak14; @Mermin68; @Gryaznov73; @Lieb75; @Sari76; @Stroud76; @DeWitt79; @Totsuji79; @Deutsch79; @Vieillefosse81; @Rosenfeld82; @Nordholm84; @Kaklyugin85; @Ortner99; @Caillol99_2]. Below, we briefly remind some of the results particularly relevant to the present discussion.
Mermin [@Mermin68] demonstrated that the internal energy of the 3D OCP is bounded below by the Debye-Hückel (DH) value. This demonstration is quite general and should be applicable to the 2D case with logarithmic interactions, too. This bound is a reasonable measure of the actual OCP energy at weak coupling. Lieb and Narnhofer [@Lieb75] derived another exact lower bound on the reduced energy (energy per particle in units of system temperature) of the 3D OCP, which reads $u> -0.9\Gamma$. This result is often refereed to as the ion sphere model [@Baus80; @Ichimaru82] (ISM) and provides rather good estimate of the internal energy at strong coupling. Similar lower bound has been identified for the 2D OCP with logarithmic interactions by Sari and Merlini [@Sari76]. It reads $u> -0.375 \Gamma$ and is usually referred to as the ion disc model (IDM). Again, IDM is surprisingly accurate at strong coupling. Gryaznov and Iosilevskiy [@Gryaznov73], and later independently Nordholm [@Nordholm84], proposed a simple modification of the DH theory for 3D, called “DH plus hole” (DHH) approximation, based on the recognition that the exponential particle density must be truncated close to the particles so as not to become negative. It improves considerably the DH theory at moderate coupling, $\Gamma\lesssim 1$, but exhibits improper scaling ($\propto -0.75\Gamma$) in the high-$\Gamma$ limit. This approach can be extended to the 2D OCP with logarithmic interactions, as we demonstrate below. More recently, Caillol derived two other exact lower bounds for the internal energy[@Caillol99_2], which have been demonstrated to be in better agreement with the numerical results than those obtained previously in a wide range of $\Gamma$.
The purpose of this paper is threefold. First, we summarize yet another simple analytical scheme to estimate the internal energy of the OCP in 3D and 2D. The approach is based on the hybrid DHH + ISM/IDM consideration formulated below. Simple electrostatic consideration, involving the solution of the Poisson equation, is used and thus, in 2D case, the approach is limited to the logarithmic interaction. It produces expressions, which reduce to the DH result at weak coupling and to the ISM/IDM results at strong coupling and provide reasonable interpolation between these limits. Second, we briefly summarize simple and accurate fits for all three OCP systems discussed here. In particular, we demonstrate that in the 2D OCP the thermal component of the internal energy exhibits the same scaling for Coulomb and logarithmic interactions. Similar scaling also holds for Yukawa interactions near the OCP limit (long screening length) and this suggests that it is a universal property of soft repulsive particle systems in 2D. Based on these accurate scalings of the internal energy other thermodynamic properties can be easily calculated. Finally, we briefly compare the location of the fluid-crystal phase transition in 2D OCP with Coulomb and logarithmic interactions, as estimated using the Kosterlitz-Thouless-Halperin-Nelson-Young (KTHNY) theory.
Hybrid approach to the internal energy of the OCP in 3D and 2D
==============================================================
Linear Debye-Hückel approximation.
----------------------------------
The solution of the linear Poisson-Boltzmann equation, $\Delta\phi = k_{\rm D}^2\phi$, in 3D and 2D yields $$\label{phia}
\phi(r)=
\left\{\begin{aligned}
\frac{Q}{r}e^{-k_{\rm D}r},\qquad k_{\rm D}=\sqrt{4\pi n Q^2/T},\qquad (3\rm D) \\
Q K_0(rk_{\rm D}),\qquad k_{\rm D}=\sqrt{2\pi n Q^2/T},\qquad (2\rm D)
\end{aligned} \right.$$ where $K_0(x)$ is the zero-order modified Bessel function of the second kind and $k_{\rm D}$ is the inverse screening length. Note the relations $k_{\rm D}a=\sqrt{3\Gamma}$ in 3D and $k_{\rm D}a=\sqrt{2\Gamma}$ in 2D. The reduced excess (that over non-charged particles) energy of the systems, independently of dimensionality, can be evaluated from $$\label{energy}
u_{\rm ex}\equiv \frac{U_{\rm ex}}{NT}=\frac{\left[Q\phi(r)-V(r)\right]_{r\rightarrow 0}}{2T},$$ where $N$ is the number of the particles ($N\rightarrow \infty$ in the thermodynamic limit). This corresponds to the DH approximation for the weakly coupled ($\Gamma\ll 1$) limit: $$\label{uDH}
u_{\rm DH}(\Gamma)=
\left\{\begin{aligned}
-\frac{\sqrt{3}}{2}\Gamma^{3/2},\qquad (3\rm D) \\
-\frac{\Gamma}{4}\left(\ln \frac{\Gamma}{2}+2\gamma\right),\qquad (2\rm D)
\end{aligned} \right.$$ where $\gamma\simeq 0.57721$ is the Euler’s constant (we used the expansion $K_0(x)\simeq -\gamma +\ln2 -\ln x +{\mathcal O}(x^2)$ for $x\ll 1$). The DH approximation provides accurate results only in the limit of extremely weak coupling.
Debye-Hückel plus hole approximation
------------------------------------
To extend the applicability of the DH approach to the moderately coupled OCP, the simple phenomenological “Debye-Hückel plus hole” (DHH) approximation was proposed [@Nordholm84; @Gryaznov73]. The main idea behind the DHH approximation is that the exponential particle density must be truncated close to a test particle in order to avoid density to be negative upon linearization. The DHH approach was originally applied to the 3D OCP. Here we outline its application to the 2D case, but 3D results are also summarized for completeness.
The potential inside the hole (sphere in 3D and disk in 2D cases) of radius $h$ can be written as $$\label{in}
\phi_{\rm in}(r)=
\left\{\begin{aligned}
\frac{Q}{r}+{\mathcal A}_0+{\mathcal A}_2r^2,\qquad (3\rm D) \\
-Q\ln(r/a)+{\mathcal A}_0+{\mathcal A}_2r^2.\qquad (2\rm D)
\end{aligned} \right.$$ Outside the hole, the potential satisfies the linearized Poisson-Boltzmann equation, so that $$\label{out}
\phi_{\rm out}(r)=
\left\{\begin{aligned}
\frac{{\mathcal B}}{r}e^{-k_{\rm D}r},\qquad (3\rm D) \\
{\mathcal B}K_0(rk_{\rm D}).\qquad (2\rm D)
\end{aligned} \right.$$ The two solutions should be matched at $r=h$, requiring $ \phi_{\rm in}(h)=\phi_{\rm out}(h)=T/Q$ (the last condition ensures that particle density vanishes at the hole boundary in the linear approximation) and $\phi'_{\rm in}(h)=\phi'_{\rm out}(h)$. Using the identity $K_0'(x)=-K_1(x)$ we get the following equations for $z=h/a$ $$\label{za}
z^2=
\left\{\begin{aligned}
\frac{1}{3\Gamma}\left\{\left[1+(3\Gamma)^{3/2}\right]^{1/3}-1\right\}^2,\qquad (3\rm D) \\
1-z\sqrt{\frac{2}{\Gamma}}\frac{K_1(\sqrt{2\Gamma}z)}{K_0(\sqrt{2\Gamma}z)}.\qquad (2\rm D)
\end{aligned} \right.$$ Unlike the 3D case, where the hole radius is expressed explicitly in terms of $\Gamma$, in the 2D case transcendent equation should be solved numerically. This, however, does not represent a major difficulty. The dependence $z(\Gamma)$ is shown in Fig. \[h1\] for both 3D and 2D OCP. In both cases $h\rightarrow 0$ when $\Gamma\rightarrow 0$, and $h\rightarrow a$ when $\Gamma\rightarrow\infty$.
![Reduced radius of the hole, $h/a$, around the test particle as a function of the coupling parameter $\Gamma$ in the 3D and 2D OCP.[]{data-label="h1"}](Fig1.eps){width="7cm"}
The reduced excess energy can be evaluated using equation (\[energy\]), which yields $u_{\rm DHH}=(Q{\mathcal A}_0/2T)$. This results in $$\label{uexb}
u_{\rm DHH}(\Gamma)=
\left\{\begin{aligned}
-\frac{1}{4}\left\{\left[1+(3\Gamma)^{3/2}\right]^{2/3}-1\right\},\qquad (3\rm D) \\
\frac{1}{2}+\frac{\Gamma}{2}\ln z-\frac{\Gamma}{4}z^2. \qquad (2\rm D)
\end{aligned} \right.$$ In the limit $\Gamma\ll 1$, Eq. (\[uexb\]) reduces to the DH results of Eq. (\[uDH\]), but it remains adequate at higher $\Gamma$ than the DH approach does. For example, in the 2D OCP the exact result can be obtained analytically in the special case $\Gamma=2$ [@Jancovici81; @Alastuey81]. The exact excess energy at this point is $u_{\rm ex}(2)=-\gamma/2\simeq -0.28861$ [@Jancovici81]. The DHH value is very close to that, $u_{\rm DHH}(2)\simeq -0.29324$, while the DH value is considerably below the exact one, $u_{\rm DH}(2)\simeq - 0.57721$. In the strongly coupled regime $\Gamma\gg1$, the DHH approximation yields the correct scaling $u_{\rm ex} \propto \Gamma$, but the coefficient of proportionality is incorrect ($-0.75$ instead of $\simeq -0.9$ in 3D and $-0.25$ instead of $\simeq -0.375$ in 2D). In Figure \[u2\] we compare the energies obtained using the DHH approach with those obtained using Monte Carlo (MC) and molecular dynamics (MD) computer simulations. It is worth noting that the application of the DHH approach to 3D Yukawa systems has been recently discussed in Ref. [@DHH] in the context of complex (dusty) plasmas.
![Reduced excess energy $u_{\rm ex}/\Gamma$ versus the coupling parameter $\Gamma$ for the 3D OCP (a) and 2D OCP (b). For the 3D case, symbols are the results from numerical MC [@Caillol99; @Caillol10] and MD [@Farouki94; @Hamaguchi96] simulations. Similarly, for the 2D case, symbols are the results from MC [@Caillol82] and MD [@Leeuw82] simulations. Various curves correspond to the DH, DHH, ISM and IDM approximations, as indicated in the figures. The (red) solid curves in both figures show the result of the hybrid DHH+ISM (3D) and DHH+IDM (2D) approximation of Eq. (\[Hybrid\]).[]{data-label="u2"}](Fig2.eps){width="6.9cm"}
![Reduced excess energy $u_{\rm ex}/\Gamma$ versus the coupling parameter $\Gamma$ for the 3D OCP (a) and 2D OCP (b). For the 3D case, symbols are the results from numerical MC [@Caillol99; @Caillol10] and MD [@Farouki94; @Hamaguchi96] simulations. Similarly, for the 2D case, symbols are the results from MC [@Caillol82] and MD [@Leeuw82] simulations. Various curves correspond to the DH, DHH, ISM and IDM approximations, as indicated in the figures. The (red) solid curves in both figures show the result of the hybrid DHH+ISM (3D) and DHH+IDM (2D) approximation of Eq. (\[Hybrid\]).[]{data-label="u2"}](Fig3.eps){width="7.cm"}
Ion sphere and ion disc models
------------------------------
The main idea of the ion sphere (ISM) and ion disk (IDM) models is that in the regime of strong coupling, the particles repel each other and form a regular structure with the interparticle spacing of order $a$[@Baus80]. Each particle can be considered as restricted to the cell (sphere in 3D and disc in 2D) of radius $a$, filled with the neutralizing background. The cells are charged neutral and do not overlap, and hence the potential energy of the system is just the sum of potential energy of each cell. The latter is readily calculated from the pure electrostatic consideration [@Lieb75; @Sari76]. The result is $$\label{ISDM}
u_{\rm ISM/IDM}=
\left\{\begin{aligned}
-\frac{9}{10}\Gamma= -0.9\Gamma,\qquad (3\rm D) \\
-\frac{3}{8}\Gamma= -0.375\Gamma.\qquad (2\rm D)
\end{aligned} \right.$$ The results are very close to the static components of the actual excess energy of the 3D and 2D OCP in both strongly coupled fluid and solid phases. They can be compared with the Madelung constants of the OCP bcc lattice, $u_{\rm M}=-0.895929\Gamma$ (3D), and triangular lattice, $u_{\rm M}=-0.37438\Gamma$ (2D). The agreement is impressive. It was proven mathematically that Eqs. (\[ISDM\]) provides the lower bounds of the excess internal energy in the thermodynamic limit [@Lieb75; @Sari76]. The ISM and IDM asymptotes are shown in Fig. \[u2\]. The ISM model can be easily generalized to 3D Yukawa systems. It is worth to mention that the ISM result for the excess energy can also be obtained from the energy equation using Percus-Yevick (PY) radial distribution function for hard spheres at the unphysical packing fraction $\eta=1$, which provides some link between the ISM approximation and the integral equation theories (for details see Ref. [@ISM] and references therein).
Hybrid approximation
--------------------
Now we discuss the recently proposed hybrid approach to the excess energy of 3D and 2D OCP, which tends to reproduce the DH and ISM (IDM) results in the respective limits of weak and strong coupling, and provides reasonable interpolation between these limits [@Hybrid3D; @Hybrid2D].
Let us consider a test particle along with the piece of the neutralizing background charge (sphere or disc of radius $h$ in 3D or 2D, correspondingly) as a new compound particle. The internal energy of such a compound particle consists of two parts: energy of a uniformly charged cell of radius $h$ and charge $q=-Q(h/a)^D$ ($D$ is the system dimension) and the energy of a charge $Q$ placed in the center of such a cell. Solving the Poisson equation inside and outside the cell and matching the solutions we get for the energy of the uniformly charged cell of background charge $$\label{ub}
u_{\rm b}=
\left\{\begin{aligned}
\frac{3}{5}\frac{q^2}{Th},\qquad (3\rm D) \\
\frac{q^2}{T}\left(\frac{1}{8}-\frac{1}{2}\ln\frac{h}{a}\right). \qquad (2\rm D)
\end{aligned} \right.$$ The energy of a charge $Q$ placed in the center of such a cell is $$\label{up}
u_{\rm p}=
\left\{\begin{aligned}
\frac{3}{2}\frac{qQ}{Th},\qquad (3\rm D) \\
\frac{qQ}{T}\left(\frac{1}{2}-\ln\frac{h}{a}\right). \qquad (2\rm D)
\end{aligned} \right.$$ The energy of the compound particle is then $$\label{ucp}
u_{\rm cp}(\Gamma)=
\left\{\begin{aligned}
\Gamma z^2\left(\frac{3}{5}z^3-\frac{3}{2}\right),\qquad (3\rm D) \\
\Gamma z^2\left(\ln z-\frac{1}{2}\right)+\Gamma z^4\left(\frac{1}{8}-\frac{1}{2}\ln z\right).\qquad (2\rm D)
\end{aligned} \right.$$ In the limit of strong coupling, the effective charge of the compound particle tends to zero and, therefore, its internal energy should be an adequate measure of the excess energy of the whole system (per particle). We get in this limit $z\rightarrow 1$ and $u_{\rm cp}\simeq-0.9\Gamma$ (3D) or $u_{\rm cp}\simeq-0.375\Gamma$ (2D), which coincides with the ISM/IDM results.
The energy associated with the remaining interaction between the compound particles (they are not charge neutral in the general case) can be estimated from the energy equation $$\label{uppa}
u_{\rm pp}=(n/2T)\int_{r>h} V_{\rm eff}(r)[g(r)-1]d{\bf r},$$ where $V_{\rm eff}(r)$ is the Coulomb (3D) or logarithmic (2D) interaction potential between the compound particles with effective charge $Q_{\rm eff}=Q+q=Q[1-z^D]$ and $g(r)$ is the radial distribution function. Since the effective charge $Q_{\rm eff}$ is considerably reduced compared to the actual charge $Q$, especially in the strong coupling regime, it is not very unreasonable to use an expression originating from the linearized Boltzmann relation, $g(r) \simeq 1-Q_{\rm eff}\phi_{\rm out}(r)/T$, where $\phi_{\rm out}$ is given by Eq. (\[out\]) in the DHH approximation. This yields $$\label{uppb}
u_{\rm pp}(\Gamma)=
\left\{\begin{aligned}
-\frac{\sqrt{3\Gamma}}{2}z(1-z^3)^3,\; \qquad (3\rm D) \\
\frac{\Gamma(1-z^2)^3}{K_0(\sqrt{2\Gamma}z)}\int_{z}^{\infty}x\ln x K_0(\sqrt{2\Gamma}x)dx.\; \qquad (2\rm D)
\end{aligned} \right.$$ In the 3D case, we have at weak coupling $z\sim \Gamma$ and the DH result is immediately recovered. In the 2D case numerical integration is generally required in (\[uppb\]), but it can be also shown that the result reduces to the DH one in the weakly coupled limit ($\Gamma\ll 1$).
Our estimate for the OCP excess energy within the hybrid DHH+ISM/IDM approximation is then simply $$\label{Hybrid}
u_{\rm hyb}(\Gamma)=u_{\rm cp}(\Gamma)+u_{\rm pp}(\Gamma).$$ Equation (\[Hybrid\]) reduces to the DH and ISM/IDM asymptotes in respective limits of weak and strong coupling. The quality of the interpolation between these two limits is illustrated in Fig. \[u2\] (red solid curves). The agreement with the accurate numerical data from MC and MD simulations is better in the 3D case, but remains also acceptable in the 2D case, taking into account the simplicity of the model. However, it is also obvious that in many situations this accuracy is insufficient, and we summarize more accurate expressions in the next Section.
Accurate expressions for the internal energy of OCP fluids
==========================================================
For strongly coupled systems, the reduced excess energy can be conveniently divided into static and thermal components $$\label{uex}
u_{\rm ex}=u_{\rm st}+u_{\rm th}.$$ The static contribution corresponds to the value of internal energy when the particles are frozen in some regular configuration (e.g., crystalline lattice for solids), and the thermal corrections arise due the deviations from these fixed positions, associated with thermal fluctuations. When the value of the static component of the excess energy is specified, the thermal component determines the excess energy and other thermodynamic properties of the system. It is known that the thermal energy exhibits quasi-universal scaling for soft repulsive interactions in 3D (first proposed by Rosenfeld and Tarazona (RT) [@Rosenfeld98; @Rosenfeld00]) and two interesting questions arise: (i) How accurate is the RT scaling for 3D OCP; and (ii) Whether there is some analog of the RT scaling in 2D.
First, let us consider the 3D case. A relevant measure of the static energy component in 3D OCP fluids is the ISM energy (we remind that it is very close to the bcc lattice sum). Based on the accurate MC simulation results from Ref.[@Caillol99], a simple two-term expression for $u_{\rm th}$ has been proposed [@Khrapak14] $$\label{uth}
u_{\rm th}=0.5944\Gamma^{1/3}-0.2786.$$ This fit, along with the MC numerical data, is plotted in Fig. \[fig3\]. In addition, a variant of the RT scaling, $$\label{uRT}
u_{\rm th}\simeq3.2(\Gamma/\Gamma_{\rm m})^{2/5}-0.1,$$ is also plotted. This scaling has been successfully used to obtain practical expressions for the internal energy and pressure of 3D Yukawa fluids [@KhrapakPRE15; @KhrapakJCP15; @KhrapakPPCF15] in a relatively wide range of screening strength. These practical expressions can be for instance used to estimate the sound velocity of Yukawa fluids with applications to complex (dusty) plasmas [@KhrapakPRE_DA]. Figure \[fig3\] demonstrates that the RT scaling describes fairly well the numerical data, but the OCP scaling is somewhat more accurate, especially in the regime $1\leq\Gamma\leq100$. This is absolutely not surprising, since the OCP scaling is nothing but the best simple fit to the numerical data shown in Fig. \[fig3\]. The very fact that the exponent $s=1/3$ (or close to that) in Eq. (\[uth\]) provides particularly good agreement with the numerical data for the OCP has been documented in a number of previous studies [@Farouki94; @Hamaguchi96; @Caillol99; @Stringfellow90; @Dubin99]. It is worth noting that the OCP expression (\[uth\]), rewritten in the RT-like form (i.e., $u_{\rm th}$ expressed in terms of $\Gamma/\Gamma_{\rm m}$) is also superior to RT scaling for Yukawa fluids near the OCP limit (when screening length is longer or comparable to the inter-particle spacing). This provides us with a simple and accurate practical tool to estimate thermodynamic properties of weakly screened Yukawa fluids [@KhrapakPoP15].
![Thermal component of the reduced excess energy, $u_{\rm th}$, of the strongly coupled 3D and 2D OCP fluids versus the coupling parameter $\Gamma$. Symbols correspond to MC and MD simulations: Crosses are MC results for 3D OCP [@Caillol99], circles are MC results for 2D OCP with the Coulomb interaction [@Gann79], triangles and stars are MC and MD results for 2D OCP with logarithmic interaction, respectively [@Caillol82; @Leeuw82]. The red solid curve is the 3D OCP fit of Eq.(\[uth\]). The blue solid line corresponds to the 2D OCP scaling. The red dashed curve represents the RT scaling of Eq.(\[uRT\]).[]{data-label="fig3"}](Fig4.eps){width="8.3cm"}
Let us now consider the 2D case. The energy can be again divided into the static and thermal parts, according to Eq. (\[uex\]). For the static component of the energy in the 2D case, we now chose the triangular lattice sums (Madelung energies), which are $u_{\rm M}=-0.37438\Gamma$ for the logarithmic and $u_{\rm M}=-1.106103\Gamma$ for the Coulomb potential, respectively [@Caillol82; @Gann79] (the ion disc model can be constructed for logarithmic interactions as we discussed above, but we are not aware of any such construction for the Coulomb interaction in 2D). Then, subtracting the static component from the full excess energy, available from the previous numerical simulations, we can obtain the thermal energy component. The resulting dependence of $u_{\rm th}$ on the coupling parameter $\Gamma$ is shown in Fig. \[fig3\]. The numerical data points, for both Coulomb and logarithmic interactions, tend to collapse on a single quasi-universal curve. Some scattering of the data points is present, but no clear systematic trend is observed, indicating that this may simply reflect the level of accuracy of the simulation results (note that at large $\Gamma$, the thermal component is a tiny fraction of the total OCP excess energy). The dependence $u_{\rm th} (\Gamma)$ has a logarithmic character, the blue solid line corresponds to $$\label{uth_2D}
u_{\rm th}\simeq 0.231\ln(1+2.798\Gamma).$$ Since the values of $\Gamma_{\rm m}$ for the 2D OCP with Coulomb and logarithmic interactions are rather close, an analog of the RT scaling for soft repulsive particles in 2D emerges. The expression proposed in Ref. [@KhrapakPoP15] is $$\label{utha}
u_{\rm th}\simeq b_1\ln[1+b_2(\Gamma/\Gamma_{\rm m})],$$ with the coefficients $b_1=0.231$ and $b_2 =391.655$. Similarly to the 3D case, this scaling is well applicable to weakly screened Yukawa systems in 2D. Various thermodynamic functions of 2D Yukawa fluids can be easily estimated. Good agreement with the simulation results of Refs. [@Totsuji04; @Hartmann05] has been documented [@KhrapakPoP15]. Applications to estimate the sound velocity in two-dimensional Yukawa fluids have been discussed in Ref. [@2Dsound]
Free energy and pressure
========================
Having relatively accurate expressions for the excess internal energy of strongly coupled OCP in 3D and 2D, we can evaluate other thermodynamic quantities. Here we present expressions for the reduced Helmholtz free energy and pressure. The generic expression for the reduced excess (that over non-interacting particles) free energy of the strongly coupled OCP fluid is $$\label{free}
f_{\rm fluid}(\Gamma)=f(\Gamma_0)+\int_{\Gamma_0}^{\Gamma}d\Gamma' u_{\rm ex}(\Gamma')/\Gamma',$$ where $\Gamma_0$ corresponds to the weakly or moderately coupled regime, and $f(\Gamma_0)$ is known to a good accuracy. In some cases $\Gamma_0$ can be simply set zero, because the exact behavior of the excess energy at weak coupling normally has very little effect on the excess free energy at strong coupling. This is, however, not the case for the 3D OCP, because the integral in Eq. (\[free\]) diverges when expression (\[uth\]) is used. Therefore, we chose $\Gamma_0=1$, $f(\Gamma_0)=-0.4368$ [@Farouki94] to get $$f_{\rm fluid}=-0.9\Gamma+1.7832\Gamma^{1/3}- 0.2786\ln\Gamma-1.3200.$$ This expression has been previously derived in Ref. [@Khrapak14]. For the 2D OCP the scaling (\[uth\_2D\]) does not lead to the integral divergence at small $\Gamma$. Although it also does not reproduce the exact behavior of $u_{\rm ex}$ in the limit of weak coupling, for strongly coupled 2D OCP with Coulomb interactions, we find appropriate to put $\Gamma_0=0$, which yields $$f_{\rm fluid}=-1.106103\Gamma -0.231{\rm Li}_2(-2.798\Gamma),$$ where ${\rm Li}_2(z)=\int_z^0dt \ln(1-t)/t$ is dilogarithm. For the 2D OCP with logarithmic interaction we should use the fact that the reduced free energy at $\Gamma_0=2$ is known exactly, $f(2)= 1-\tfrac{1}{2}\ln(2\pi)\simeq 0.0811$ [@Alastuey81]. We have then $$f_{\rm fluid}=-0.37438\Gamma -0.231{\rm Li}_2(-2.798\Gamma)+0.1469.$$
To check the accuracy of these expressions we look for the intersection of the fluid and solid free energies of the considered systems. This is a very stringent test, since the free energies of fluid and solid are nearly parallel near the intersection. For the free energy of the 3D OCP forming the bcc lattice we take the expression from Ref. [@Dubin99], $$f_{\rm solid}=-0.895929\Gamma+\frac{3}{2}\ln\Gamma-1.1704-\frac{10.84}{\Gamma}-\frac{352.8}{2\Gamma^2}-\frac{179400}{3\Gamma^3},$$ where the last three terms represent anharmonic corrections. For the 2D OCP forming the triangular lattice we use the result from Ref. [@Gann79] $$f_{\rm solid}=-1.106103\Gamma+\ln\Gamma-\ln2+0.298-\frac{5}{\Gamma}-\frac{560}{2\Gamma^2},$$ where the last two terms are again anharmonic corrections. Finally, for the 2D OCP solid with the logarithmic interaction we use the available result of a simple harmonic approximation [@Alastuey81], $$f_{\rm solid}=-0.37438\Gamma+\ln\Gamma-0.262.$$
![Reduced excess free energy, in units of $\Gamma$ (i. e. $f_{\rm ex}/\Gamma$), as a function of the coupling parameter $\Gamma$ of 3D OCP (a), 2D OCP with Coulomb interaction (b), and 2D OCP with logarithmic interaction (c). The red curves correspond to the fluid phase, the blue curves correspond to the crystalline solid. Their intersection locates the point of the fluid-crystal phase transition. This figure gives yet another illustration that the free energies of the fluid and solid phases are nearly parallel in the vicinity of their intersection, indicating that very high accuracy is required to properly determine the location of the phase transition.[]{data-label="fig5"}](Fig5.eps){width="15.cm"}
The fluid and solid reduced excess free energies near their intersection are shown in Fig. \[fig5\]. From the location of the intersection point we can estimate the coupling parameter at the fluid-crystal phase transition. For the 3D OCP we get $\Gamma_{\rm m}\simeq 174$ in very good agreement with the result of Ref. [@Dubin99]. For the 2D OCP with the Coulomb interaction we get $\Gamma_{\rm m}\simeq 140$, which is consistent with the range predicted in earlier numerical simulations and experiments. For the 2D OCP with the logarithmic interaction, intersection occurs near $\Gamma_{\rm m}\simeq 154$, which is somewhat higher than obtained in Refs. [@Caillol82; @Leeuw82]. However, we remind that the free energy of the solid phase has been evaluated using a simple harmonic approximation. By analogy with the two other OCP systems, it can be expected that if the anharmonic corrections are properly accounted for, the coupling parameter corresponding to the phase transition can decrease to $\Gamma_{\rm m}\simeq 130 - 140$, in much better agreement with the results of previous studies.
Regarding the excess pressure, it can be trivially obtained using the virial (pressure) equation. For the OCP with Coulomb interactions we have, in reduced units, $p_{\rm ex}=\tfrac{1}{3}u_{\rm ex}$ in 3D and $p_{\rm ex}=\tfrac{1}{2}u_{\rm ex}$ in 2D. For the 2D OCP with the logarithmic potential, the virial equation combined with the charge neutrality condition immediately yields $p_{\rm ex}=-\tfrac{1}{4}\Gamma$. This simple exact result is a consequence of the observation that density is an irrelevant variable for 2D OCP with the logarithmic potential [@Caillol82].
Other thermodynamics quantities can be evaluated in a similar way.
Melting of the 2D OCP in the KTHNY model
========================================
The Kosterlitz-Thouless-Halperin-Nelson-Young (KTHNY) theory [@KTHNY] describes melting in classical 2D systems with arbitrary interaction between the particles. Recently, the KTHNY scenario has been confirmed for sufficiently soft interactions [@Kapfer15], indicating that it should be relevant to 2D OCP, both with Coulomb and logarithmic interactions. The KTHNY theory states that the melting transition occurs when $$\frac{4 \pi T}{b^2}= \frac{\mu(\mu+\lambda)}{2\mu+\lambda},$$ where $b$ is the lattice spacing and $\mu$, $\lambda$ are the Lamé coefficients of 2D solid, which can be expressed via the longitudinal and transverse sound velocities [@Peeters87]. For the OCP the longitudinal sound velocity is infinite and the melting temperature can be expressed via the transverse sound velocity [@Thouless78] $$T_{\rm m}=\frac{mnb^2c_{\rm T}^2}{4\pi},$$ where $mn$ is the mass per unit area and $c_{\rm T}$ is the transverse sound velocity. For the triangular lattice we have $b=(2/n\sqrt{3})^{1/2}$. Then, for the Coulomb interaction, using the zero-temperature limit $c_{\rm T}\simeq 0.513\sqrt{Q^2/mb}$ [@Peeters87] we get $\Gamma_{\rm m}\simeq 79$, as obtained originally by Thouless [@Thouless78]. This is clearly only a rough estimate of the melting location, considerable improvement can be achieved by taking into account the temperature dependence of the shear modulus arising from the phonon-phonon interaction and the polarizability of dislocation pairs [@Morf79]. Nevertheless, it is interesting to make similar estimation for the 2D OCP with the logarithmic interaction. We use the transverse sound velocity $c_{\rm T}=\sqrt{Q^2/8m}$ derived in Ref. [@Alastuey81] to get $$\Gamma_{\rm m}=16\pi\sqrt{3}\simeq 87.$$ This estimate demonstrates the same level of accuracy as in the Coulomb case, improvements seem necessary.
Conclusion
==========
The one-component-plasma is an old model with wide iterdisciplinary applications. It also represents an important example of classical systems with extremely soft interactions. In this paper we mainly discussed thermodynamic properties of the model (in terms of the internal energy). Of particular significance can be the observation that the OCP scaling of the thermal component of the excess energy exhibits scaling, which is quasi-universal and applies to other soft repulsive potentials, both in 2D and 3D cases. We pointed out applications to the weakly screened Yukawa systems, mostly in the context of complex (dusty) plasmas. It is likely that the discussed observation can be also useful for other classical systems with soft repulsive interactions.
SAK present position at the Aix-Marseille-University is supported by the A\*MIDEX grant (Nr. ANR-11-IDEX-0001-02) funded by the French Government “Investissements d’Avenir” program. The development of the hybrid approach for the OCP in 3D and 2D was supported by the Russian Science Foundation, Project No. 14-12-01235.
[\[1\]]{}
S.G. Brush, H.L. Sahlin and E. Teller, J. Chem. Phys. [**45**]{}, 2102 (1966). M. Baus and J.P. Hansen, Phys. Rep. [**59**]{}, 1 (1980). V.E Fortov, I.T. Iakubov and A.G. Khrapak, Physics of Strongly Coupled Plasma (Clarendon Press, Oxford, 2006). N.V. Brilliantov, Contrib. Plasma Phys. [**38**]{}, 489 (1998). V.E. Fortov, A.G. Khrapak, S.A. Khrapak, V.I. Molotkov, and O.F. Petrov, Phys. Usp. [**[47]{}**]{}, 447 (2004). V.E. Fortov, A.V. Ivlev, S.A. Khrapak, A.G. Khrapak, and G.E. Morfill, Phys. Rep. [**421**]{}, 1 (2005). A. Ivlev, H. Löwen, G. Morfill, and C.P. Royall, Complex Plasmas and Colloidal Dispersions: Particle-resolved Studies of Classical Liquids and Solids (World Scientific, Singapore, 2012).
J.P. Hansen, Phys. Rev. A [**8**]{}, 3096 (1973). W.L. Slattery, G.D. Doolen, and H.E. DeWitt, Phys. Rev. A [**21**]{}, 2087 (1980); [**26**]{}, 2255 (1982). G.S. Stringfellow, H.E. DeWitt, and W.L. Slattery, Phys. Rev. A [**41**]{}, 1105 (1990). R.T. Farouki and S. Hamaguchi, J. Chem. Phys. [**101**]{}, 9885 (1994). S. Hamaguchi, R.T. Farouki, and D.H.E. Dubin, J. Chem. Phys. [**105**]{}, 7641 (1996). D.H.E. Dubin and T.M. O’Neil, Rev. Mod. Phys. [**71**]{}, 87 (1999). J.M. Caillol, J. Chem. Phys. [**111**]{}, 6538 (1999). J.M. Caillol and D. Gilles, J. Phys. A: Math. Theor. [**43**]{}, 105501 (2010).
H. Totsuji, Phys. Rev. A [**17**]{}, 399 (1978). R.C. Gann, S. Chakravarty and G.V. Chester, Phys. Rev. B [**20**]{}, 326 (1979).
J.M. Caillol, D. Levesque, J.J. Weis and J.P. Hansen, J. Stat. Phys. [**28**]{}, 325 (1982). S.W. de Leeuw and J.W. Perram, Physica A [**113**]{}, 546 (1982). Ph. Choquard and J. Cleroin, Phys. Rev. Lett. [**50**]{}, 2086 (1983). P. L. Radloff, B. Bagchi, C. Cerjan, and S. A. Rice, J. Chem. Phys. [**81**]{}, 1406 (1984). B. Jankovici, Phys. Rev. Lett. [**46**]{}, 386 (1981). A. Alastuey and B. Jancovici, J. Phys. [**42**]{}, 1 (1981).
S.A. Khrapak and A.G. Khrapak, Phys. Plasmas [**21**]{} 104505 (2014).
C. C. Grimes and G. Adams, Phys. Rev. Lett. [**42**]{}, 795 (1979).
S. Tanaka and S. Ichimaru, Phys. Rev. A [**35**]{}, 4743 (1987). A. Yazdi, A. Ivlev, S. Khrapak, H. Thomas, G. E. Morfill, H. Löwen, A. Wysocki, and M. Sperl, Phys. Rev. E [**89**]{}, 063105 (2014).
N.D. Mermin, Phys. Rev. [**171**]{}, 272 (1968). V.K. Gryaznov and I.L. Iosilevskiy, Numerical methods in fluid mechanics [**4**]{}, 166 (1973); For english translation see e-print arXiv:0903.4913 (2009). E.H. Lieb and H. Narnhofer, J. Stat. Phys. [**12**]{}, 291 (1975). R.R. Sari and D. Merlini, J. Stat. Phys. [**14**]{}, 91 (1976). D. Stroud and N.W. Ashcroft, Phys. Rev. A [**13**]{}, 1660 (1976). H.E. DeWitt and Y. Rosenfeld, Phys. Lett. A [**75**]{}, 79 (1979). H. Totsuji, Phys. Rev. A [**19**]{}, 1712 (1979); [**19**]{}, 2433 (1979). C. Deutsch, H.E. DeWitt, and Y. Furutani, Phys. Rev. A [**20**]{}, 2631 (1979). P. Vieillefosse, J. Phys. [**42**]{}, 723 (1981). Y. Rosenfeld, Phys. Rev. A [**25**]{}, 1206 (1982). S. Nordholm, Chem. Phys. Lett. [**105**]{}, 302 (1984). A.S. Kaklyugin, High Temp. [**23**]{}, 169 (1985). J. Ortner, Phys. Rev. E [**59**]{}, 6312 (1999). J.M. Caillol, J. Chem. Phys. [**111**]{}, 9695 (1999); J. M. Caillol, J. Chem. Phys. [**112**]{}, 6940 (2000).
S. Ichimaru, Rev. Mod. Phys. [**54**]{}, 1017 (1982).
S. A. Khrapak, A. G. Khrapak, A. V. Ivlev, and G. E. Morfill, Phys. Rev. E [**89**]{}, 023102 (2014). S. A. Khrapak, A. G. Khrapak, A. V. Ivlev, and H. M. Thomas, Phys. Plasmas [**21**]{}, 123705 (2014).
S. A. Khrapak and A. G. Khrapak, Phys. Plasmas [**22**]{}, 044504 (2015). A. G. Khrapak and S. A. Khrapak, AIP Advances [**5**]{}, 087175 (2015).
Y. Rosenfeld and P. Tarazona, Mol. Phys. [**95**]{}, 141 (1998). Y. Rosenfeld, Phys. Rev. E [**62**]{}, 7524 (2000).
S.A. Khrapak and H. Thomas, Phys. Rev. E [**91**]{}, 023108 (2015). S.A. Khrapak, N.P. Kryuchkov, S.O. Yurchenko, and H.M. Thomas, J. Chem. Phys. [**142**]{}, 194903 (2015). S. A. Khrapak, Plasma Phys. Control. Fusion [**58**]{}, 014022 (2015).
S. A. Khrapak and H. M. Thomas, Phys. Rev. E [**91**]{}, 033110 (2015).
S.A. Khrapak, I.L. Semenov, L. Couedel, and H. Thomas, Phys. Plasmas [**22**]{}, 083706 (2015).
H. Totsuji, M. S. Liman, C. Totsuji, and K. Tsuruta, Phys. Rev. E [**70**]{}, 016405 (2004). P. Hartmann, G.J. Kalman, Z. Donko, and K. Kutasi, Phys. Rev. E [**72**]{}, 026409 (2005).
I. Semenov, S. Khrapak, and H. Thomas, Phys. Plasmas [**22**]{}, 114504 (2015).
M. Kosterlitz and D. Thouless, J. Phys. C [**6**]{}, 1181 (1973); B. Halperin and D. Nelson, Phys. Rev. Lett. [**41**]{}, 121 (1978); D. Nelson and B. Halperin, Phys. Rev. B [**19**]{}, 2457 (1979); P. Young, Phys. Rev. B [**19**]{}, 1855 (1979).
S. Kapfer and W. Krauth, Phys. Rev. Lett. [**114**]{}, 035702 (2015).
F. Peeters and X. Wu, Phys. Rev. A [**35**]{}, 3109 (1987).
D. J. Thouless, J. Phys. C: Solid State Phys. [**11**]{}, L189 (1978). R. H. Morf, Phys. Rev. Lett. [**43**]{}, 931 (1979).
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'It is demonstrated that the collapse of the wave function is equivalent to the continuity of measurement outcomes. The latter states that a second measurement has to result in the same outcome as the first measurement of the same observable for a vanishing time between both observations. In contrast to the exclusively quantum-physical collapse description, the equivalent continuity requirement also applies in classical physics, allowing for a comparison of both domains. In particular, it is found that quantum coherences are the single cause for measurable deviations in statistical properties due to the collapse. Therefore, the introduced approach renders it possible to characterize and quantify the unique features of the quantum-physical measurement problem within the framework of modern quantum resource theories and compare them to classical physics.'
author:
- 'J. Sperling'
title: Continuity of measurement outcomes
---
The collapse of the wave function (CWF) is a cornerstone of quantum physics and describes how a system responds to a measurement process [@H27; @N32]. The consequence that the detection process can instantaneously alter the quantum state is a counterintuitive mechanism genuine to quantum physics. Despite having such a surprising property, the CWF makes predictions which have been repeatedly confirmed in experiments; see Ref. [@FTZWF15] for a recent implementation. In addition, the remote manipulation of quantum states, which is caused by the CWF, is a useful feature for novel applications, such as measurement-based quantum computation [@BBDRN09]. Still, the discontinuous reduction of the quantum state, also known as the measurement problem, is not fully understood yet.
Independently from the CWF, the field of quantum coherence has been developed for the aim of studying and quantifying quantum phenomena as resources for practical applications [@SAP17]. Specifically, quantum superpositions have been identified as a unique foundation for quantum coherence [@A06; @A14; @SV15; @WY16; @TKEP17]. A comprehensive analysis of interferences and coherences in general theories can be found in Ref. [@BDW17]. Moreover, transformations between diverse notions of quantum coherence allow for a versatile utilization of different quantum effects, such as conversions between local nonclassicality and entanglement [@VS14; @KSP16; @CH16]. Furthermore, quantum correlations in bipartite systems—in combination with the influence of local measurements of one subsystem—have been studied and quantified, for instance, in relation to steering [@HF16] and quantum discord [@MYGVG16]; see also Ref. [@SRBL17]. Moreover, conditional quantum correlations in such scenarios have been experimentally investigated [@SBDBJDVW16; @ASCBZV17], and they connect to ancilla-assisted quantum protocols [@CSRBAL16; @MZFL16]. These recent advances, using the projective properties of the CWF, hint at a deeper interconnection between application-based measures of quantum coherence and the fundamental aspect of the CWF [@YDXLS17].
The instantaneous CWF is inherent to quantum systems and, therefore, the formulation of a classical analog to the CWF appears to be an illusive undertaking, challenging classical intuitions about the physical world [@S35; @W05]. While the Schrödinger equation describes the continuous evolution, the collapse model represents a discontinuous alteration of the quantum state. For this reason, some theoretical approaches, with decoherence being arguably the most prominent example [@Z03], attempt to replace the CWF with other continuous quantum mechanisms. Nevertheless, a classical correspondence is required to assess the quantumness of the measurement process, for instance, in the framework of quantum coherence, which is based on comparing quantum states with incoherent ones. This open problem of finding a classical counterpart to the CWF is addressed here; its resolution is shown to lead to new insights into a resource-theoretical interpretation of the measurement problem.
In this contribution, it is demonstrated that the discontinuous CWF can be replaced by a continuity requirement, the continuity of measurement outcomes (CMO), by showing the equivalence of both concepts. Besides the resulting reinterpretation of the CWF as a consequence of a more intuitive and accessible principle, the CMO also applies in classical physics and can be related to conditional probabilities. Still, the implications of the CMO in the classical and quantum realm are distinctively different. This renders it possible to formulate measurable criteria, based on non-commuting observables, to verify the CWF through quantum coherences. Therefore, a substitution of the CWF with the equivalent CMO as the prime axiom for quantum measurements provides an experimentally accessible connection between fundamental aspects of the CWF and modern resource theories, relevant for applications of quantum information technology.
To formulate the desired equivalence, an observable with the spectral decomposition $\hat x=\sum_n x_n |x_n\rangle\langle x_n|$ is considered. To avoid that technical difficulties obstruct the physical meaning, it is assumed that this observable is non-degenerate and acts on a finite-dimensional Hilbert space. Born’s rule states that the probability for the outcome $x_n$ is given by $P(x_n)=\langle x_n|\hat\rho|x_n\rangle$ when the system is in the quantum state $\hat\rho$ at the measurement time $t=0$. The collapsed state after this measurement is the eigenstate $\hat\rho'=|x_n\rangle\langle x_n|$ and further evolves according to the Schrödinger equation, labeled as $\hat\rho^{(t)}$ for $t>0$.
One implication of the CWF is that a second measurement of $\hat x$ becomes deterministic when the waiting time after the first measurements tends to zero, $$\begin{aligned}
\label{eq:CMO}
\lim_{t\to 0^{+}}P^{(t)}(x_m|x_n)=\delta_{m,n},
\end{aligned}$$ where $t\to0^{+}$ indicates a limit to zero for positive delays ($t>0$) and $\delta$ denotes the Kronecker symbol. Here, $P^{(t)}(x_m|x_n)$ denotes the probability that $x_m$ is measured under the constraint that the first measurement outcome is $x_n$. Equation defines the CMO and states that the measurement outcomes of two subsequent measurements of the same observables becomes identical when the waiting time between the measurements approaches zero [@comment1]. The CMO is a result of the CWF as the collapsed state $\hat \rho'=\lim_{t\to0^{+}}\hat\rho^{(t)}=|x_n\rangle\langle x_n|$ implies $P^{'}(x_m|x_n)=\lim_{t\to0^{+}}P^{(t)}(x_m|x_n)=\langle x_m|\hat\rho'|x_m\rangle=\delta_{m,n}$.
Interestingly, the inverse direction—the CMO \[Eq. \] implies the CWF—can be proven as well. Namely, the state $\hat\rho'$ instantaneously after the measurement can be expanded, in general, as $\hat\rho'=\sum_{k,l}\rho'_{k,l}|x_k\rangle\langle x_l|$. Then, the premise in Eq. implies that $\rho'_{n,n}=\langle x_n|\hat\rho'|x_n\rangle=1$ and $\rho'_{m,m}=0$ for $m\neq n$. Since $\hat\rho'$ is a positive semidefinite operator, this further implies that $\rho'_{k,l}=0$ for all $k\neq l$. Thus, the state after the first measurement has to have the expansion $\hat\rho'=|x_n\rangle\langle x_n|$, which is the one state postulated by the CWF model.
Therefore, the CMO is demonstrated to be equivalent to the CWF. This equivalence enables one to start with the continuity requirement posed by the CMO as the primary axiom and degrade the discontinuous CWF to a conclusion from it, which is the opposite direction typically followed. One advantage is that it seems more natural to require that the measurement outcome is conserved for the same observable and a waiting time approaching to zero because there is no time for the system to evolve such that a different outcome is possible. In contrast to the clear interpretation of the CMO, the discontinuity of the instantaneous CWF is a more challenging concept. Another benefit is that Eq. makes sense in the context of classical measurements too.
In classical statistical physics, the probability $P(X)$ of a system to be in a configuration in a set $X$ at a measurement time $t=0$ is described in terms of a classical probability distribution $P$. Specifically, the disjoint decomposition $\{X_m\}_m$ of the configuration space can be made such that the elements of $X_m$ lead to the outcome $x_m$ with the probability $P(X_m)$. After a waiting time $t$, the initial set $X$ is mapped to a new part of the configuration space, $X^{(t)}$. The probability that the system evolved into another set of configurations, $\tilde X$, is then given by the conditional probability $P(\tilde X|X^{(t)})=P(\tilde X\cap X^{(t)})/P(X^{(t)})$. As the continuous evolution implies $X=\lim_{t\to0^{+}} X^{(t)}$, it follows in the limit of zero delay for a second measurement that $$\begin{aligned}
\label{eq:classicalCMO}
\lim_{t\to 0^{+}}P(X_m|X_n^{(t)})
=\delta_{n,m}
\end{aligned}$$ for the individual measurement outcomes. This is a conclusion of the definition of conditional probabilities, the identity $P(X_n\cap X_m)=P(X_n)$ for $n=m$, and $P(X_n\cap X_m)=P(\emptyset)=0$ for $m\neq n$. Moreover, relation , reasoned by classical statistics, resembles the quantum version of the CMO given in Eq. and, thus, is the sought-after classical analog.
While the subsequent measurement of a single observable defines the CMO, it is known that the impact of th the CWF is most pronounced when combining two incompatible measurements, which, for example, manifests itself in Heisenberg’s uncertainty relation [@H27]. Thus, the second measurement may be replaced by the observable $\hat y=\sum_{m}y_m|y_m\rangle\langle y_m|$, which does not commute with $\hat x$. Again, a vanishing waiting time between the measurements is considered, $t\to 0^{+}$.
In the classical domain, the outcome $y_m$ is obtained for a subset $Y_m$ of the configuration space with a probability $P(Y_m)$ for the case of no prior measurement. If, however, a first measurement is conducted and yields the outcome $x_n$, then the conditional probability reads $P(Y_m|X_n)=P(Y_m\cap X_n)/P(X_n)$. Thus, the total probability to measure $y_m$ in this scenario is $P'(Y_m)=\sum_n P(X_m)P(Y_m|X_m)$. A highly relevant observation is that both expressions $P$ and $P'$ are identical in classical statistical physics as the law of total probability [@S95], $P(Y)=P'(Y)$, is always satisfied. This raises the question if the same law holds true in quantum domain.
If the first quantum measurement yields the outcome $x_n$, the quantum version of the CMO predicts the state $\hat\rho'=|x_n\rangle\langle x_n|$ directly after the first measurement. The conditional probabilities of the second measurement are consequently given by $P'(y_m|x_n)=\langle y_m|\hat\rho'|y_m\rangle=|\langle y_n|x_m\rangle|^2$. Thus, the total probabilities for the measurement outcomes of $\hat y$ read $$\begin{aligned}
\label{eq:preXprob}
P'(y_m)=\sum_{n} P(x_n)P'(y_m|x_n)=\langle y_m|\left(\sum_{n} P(x_n)|x_n\rangle\langle x_n|\right)|y_m\rangle.
\end{aligned}$$ By contrast, without the prior measurement of $\hat x$, the probabilities for measuring $\hat y$ are given by $$\begin{aligned}
\label{eq:notXprob}
P(y_m)=\langle y_m|\hat\rho|y_m\rangle.
\end{aligned}$$ Therefore, the probability distribution $P'(y_m)$ after the measurement of $\hat x$ differs from the distribution $P(y_m)$ obtained without a prior observation of $\hat x$ as long as the state exhibits quantum coherences, i.e., $\hat\rho\neq \sum_{n}P(x_n)|x_n\rangle\langle x_n|$ with $P(x_n)=\langle x_n|\hat\rho|x_n\rangle$, and the observables do not commute—implying a relation of the form $|\langle x_n|y_m\rangle|^2=\delta_{m,n}$ does not hold true.
In the quantum measurement framework, the notion of incoherent states, which are of the general diagonal form $\sum_n p_n|x_n\rangle\langle x_n|$ for the basis $\{|x_n\rangle\}_n$ [@SAP17], naturally occurs. In particular, the differences of the state $\hat\rho$ and the corresponding incoherent ensemble of collapsed states, $$\begin{aligned}
\label{eq:ClassicalEnsemble}
\hat\sigma=\sum_{n}|x_n\rangle\langle x_n|\hat\rho|x_n\rangle\langle x_n|=\sum_n P(x_n)|x_n\rangle\langle x_n|,
\end{aligned}$$ results in different probability distributions, $P(y_m)\neq P'(y_m)$, given in Eqs. and , respectively. In this context, it is also worth pointing out that Eq. relates to the application of a so-called strictly incoherent operation [@WY16; @YMGGV16].
In conclusion, the classical CMO implies that $P$ and $P'$ are identical, but the same does not hold true for quantum measurements. Consequently, the CWF has a measurable impact in the quantum domain and can be accessed through the notion of quantum coherence. The other way around, for any incoherent state, being invariant under CWF \[$\hat\rho=\hat\sigma$\], $P(y)=P'(y)$ holds true for all observables $\hat y$. Therefore, coherent states (with respect to the first measurement $\hat x$) are uniquely identified by their ability to produce an observable statistical difference, $P\neq P'$, for at least one second measurement $\hat y$.
An interesting connection between the CWF and quantum coherence was previously reported in Ref. [@YDXLS17]. Here, however, it is important to stress that the presented results are obtained from a purely classical perspective on the CMO. Further, the implications valid in the classical framework (i.e., the law of total probabilities) have been shown to be violated by quantum physics.
In order to formulate an experimentally friendly criterion, the law of total variances [@S95] can be additionally considered. This relation from the theory of classical conditional probabilities reads $$\begin{aligned}
\label{eq:totalVar}
\mathbb V_{P(Y)}[Y]=\mathbb E_{P(X)}[\mathbb V_{P(Y|X)}[Y]]+\mathbb V_{P(X)}[\mathbb E_{P(Y|X)}[Y]],
\end{aligned}$$ where $\mathbb E_{P}$ and $\mathbb V_P$ denote the expectation value and variance for a distribution $P$, respectively. However, inserting the corresponding quantum-physical distributions, one readily gets $$\begin{aligned}
\label{eq:VarPARTS}
&\mathbb V_{P(y)}[y]=\langle (\Delta\hat y)^2\rangle_{\hat\rho}
\text{ and}
\\\nonumber
&\mathbb E_{P(x)}[\mathbb V_{P'(y|x)}[y]]+\mathbb V_{P(x)}[\mathbb E_{P'(y|x)}[y]]
= \mathbb V_{P'(y)}[y]=
\langle (\Delta\hat y)^2\rangle_{\hat\sigma},
\end{aligned}$$ using the quantum-physical variances $\langle (\Delta\hat y)^2\rangle_{\hat\pi}=\langle \hat y^2\rangle_{\hat\pi}-\langle \hat y\rangle_{\hat\pi}^2$ for $\hat\pi\in\{\hat\rho,\hat\sigma\}$. Thus, if the quantum analog to the classical identity is not satisfied, $$\begin{aligned}
\label{eq:Condition}
\mathbb E_{P(x)}[\mathbb V_{P'(y|x)}[y]]+\mathbb V_{P(x)}[\mathbb E_{P'(y|x)}[y]]-\mathbb V_{P(y)}[y]\neq0,
\end{aligned}$$ then the presence of coherence is certified through the consequences of the CWF. In that case, the measured quantum fluctuations for $\hat y$ are distinctively different depending on whether or not $\hat x$ was previously measured, $\langle (\Delta\hat y)^2\rangle_{\hat\rho}\neq\langle (\Delta\hat y)^2\rangle_{\hat\sigma}$.
Since the qubit is of fundamental importance as the basic unit of quantum information [@KL98; @NC00], such a system can be used to demonstrate the general function of the introduced methods. Suppose the two observables are given as $$\begin{aligned}
\label{eq:exampleObservables}
\hat x=\begin{bmatrix}
-1 & 0 \\ 0 & 1
\end{bmatrix}
\text{ and }
\hat y=\cos\vartheta\begin{bmatrix}
-1 & 0 \\ 0 & 1
\end{bmatrix}
+\sin\vartheta\begin{bmatrix}
0 & e^{-i\varphi} \\ e^{i\varphi} & 0
\end{bmatrix},
\end{aligned}$$ which both have the possible outcomes $x,y\in\{+1,-1\}$. An arbitrary, mixed initial qubit state can be parametrized as $$\begin{aligned}
\label{eq:exampleState}
\hat\rho=\begin{bmatrix}
1-p & \sqrt{p(1-p)}\gamma
\\
\sqrt{p(1-p)}\gamma^\ast & p
\end{bmatrix},
\end{aligned}$$ for $p\in[0,1]$ and $|\gamma|\leq 1$. The probabilities for the measurement of $\hat x$ are $P(x=+1)=p$ and $P(x=-1)=1-p$, and non-zero off-diagonal elements ($p(1-p)|\gamma|^2\neq0$) define quantum coherences. From the spectral decomposition of $\hat y$, the conditional probabilities can be obtained, $P'(y=\pm 1|x=+1)=(1\pm\cos\vartheta)/2$ and $P'(y=\pm 1|x=-1)=(1\mp\cos\vartheta)/2$.
Applying the law of total probabilities from classical statistics, cf. Eq. , the probability distribution for the measurement of $\hat y$ after the measurement of $\hat x$ reads $$\begin{aligned}
\label{eq:exampleClassicalP}
P'(y=\pm 1)=\frac{1\pm(2p-1)\cos\vartheta}{2}.
\end{aligned}$$ In comparison, the probability distribution without the prior CWF can be put into the form $$\begin{aligned}
P(y=\pm 1)
=P'(y=\pm 1)
\pm\mathrm{Re}(\gamma e^{i\varphi})\sqrt{p(1-p)}\sin\vartheta,
\end{aligned}$$ where the second summand accounts for the quantum interferences. Clearly, $P(y)$ is identical to $P'(y)$ \[Eq. \] when the extra contributions vanish. This holds true iff $\sin\vartheta=0$, which is is true when $\hat x$ and $\hat y$ commute \[cf. Eq. \], or $\mathrm{Re}(\gamma e^{i\varphi})\sqrt{p(1-p)}=0$, which is satisfied for any $\varphi$ if the state in Eq. is incoherent (i.e., diagonal). Moreover, the trace-norm distance between the state $\hat \rho$ and its incoherent counterpart in Eq. , $\hat\sigma=\left[\begin{smallmatrix}1-p&0\\0&p\end{smallmatrix} \right]$, can be obtained by varying over $\varphi$ and $\vartheta$ to probe all possible qubit observables (modulo the addition of the identity), $$\begin{aligned}
\label{eq:TraceDistance}
\|\hat\rho-\hat\sigma\|_{1}
=\max_{\varphi,\vartheta}\sum_{y\in\{+1,-1\}}|P(y)-P'(y)|
=2|\gamma|\sqrt{p(1-p)},
\end{aligned}$$ which is a quantifier of quantum coherence and, here, obtained from the incompatible implication of the principle of CMO in the classical and quantum domain.
Furthermore, the condition in Eq. can be applied. For this purpose, the variances $\langle (\Delta\hat y)^2\rangle_{\hat\rho}$ and $\langle (\Delta\hat y)^2\rangle_{\hat\sigma}$ in Eq. are computed, where the variance for $\hat\rho$ is obtained without a prior measurement of $\hat x$ and the second variance is obtained for the incoherent ensemble of collapsed states $\hat\sigma$ \[Eq. \]. For example, the difference in Eq. for the parameters $\varphi=-\arg\gamma$ and $\vartheta=\pi/2$ is $$\begin{aligned}
\langle (\Delta\hat y)^2\rangle_{\hat\sigma}-\langle (\Delta\hat y)^2\rangle_{\hat\rho}
= 4|\gamma|^2p(1-p),
\end{aligned}$$ which is the square of the trace distance in Eq. . Consequently, the directly applicable criterion certifies the collapse of the qubit state as a result of its quantum coherences through measured fluctuations. It is worth stressing again that the applied condition is formulated in terms of variances that are necessarily identical in classical theories \[Eq. \].
In summary, an equivalent paradigm to the CWF was found, the CMO. However, in contrast to the CWF, the CMO is directly applicable in classical statistics as well. Using conditional probabilities and the law of total probabilities, an analysis of the CMO was conducted, leading to two incompatible results in the quantum domain which should, however, be identical in classical physics. In contrast to a previous work [@YDXLS17], here the concept of an incoherent state is derived from the analysis of the CMO in a purely classical framework. Specifically, incoherent state are uniquely obtained as those states which are indeed consistent with the classical prediction. Quantum coherence, on the other hand, leads to at least one collapse scenario which is incompatible with classical physics and verifiable with the derived technique. As an example, an experimentally accessible criterion was formulated in terms of variances. As a proof of concept, this method was then applied to an example to quantify coherence in connection with non-commuting measurements.
In general, the aim of this work is to providing a deeper understanding of the instantaneous CWF from a classical perspective and characterizing this quantum phenomenon on a quantitative and measurable basis.
Because the CMO is shown to be equivalent to the CWF, it is valid to confer to the CMO as the primary paradigm, whereby the CWF becomes the derived property. It is also noteworthy that the formulated equivalence combines the discontinuous collapse with a continuity requirement. To be clear, it is not proposed to replace the CWF with an alternative mechanism, which is the case for other approaches, e.g., in the context of decoherence [@Z03]. Rather, the CWF is substituted with the equivalent principle of the CMO. This change of perspective is preferable as the CMO has a correspondence in the classical domain, which cannot be simply found for the CWF. However, the classical and quantum versions of the CMO are only superficially identical. The resulting deviation renders it possible to discriminate the quantum measurement process from classical observations using quantum coherences, leading to another physical motivation of this resource which connects it to the measurement-induced collapse. This connection also achieves the desired quantitative assessment of the CWF. Specifically, this can be done by analyzing the uncertainties with and without the collapse of the state due to a prior measurement as derived in this work.
In addition, the introduced concepts can be extended to more general scenarios. For instance, imperfections in the measurement process can have an impact on the collapsed state and are likely to soften the classical-quantum boundary. Furthermore, applying the proposed framework to a subsystem of a multipartite system will remotely influence the remaining parts, similarly to effects reported in previous studies, e.g., in Ref. [@SRBL17]. Thus, future investigations may establish collapse-based relations to nonlocal forms of quantum coherence, such as entanglement, in noisy environments. Also, the classical form of the CMO and the violation of the law of total probabilities in quantum physics hints at connections to somewhat related problems, such as contextuality and causality.
Finally, it is also worth mentioning that the proposed change of perspective—from CWF to CMO—also redirects some interpretations of quantum physics, e.g., the seminal gedankenexperiment by Schrödinger [@S35]. That is, the important aspect of the CWF in the detection process from the CMO standpoint is that the first measurement of the state of the cat (say the outcome is “dead”) is consistent with a second observation and does not alter the state to “alive.” Or, in simple terms, the CMO implies that observations of “zombie cats” are excluded from both the classical and quantum realm.
This work has received funding from the European Union’s Horizon 2020 Research and Innovation Program under grant agreement No. 665148 (QCUMbER).
[99]{} W. Heisenberg, *Über den anschaulichen Inhalt der quantentheoretischen Kinematik und Mechanik*, [Z. Physik **43**, 172 (1927)](https://doi.org/10.1007/BF01397280). J. von Neumann, *Mathematische Grundlagen der Quantenmechanik* ([Springer, 1932](https://doi.org/10.1007/978-3-642-61409-5)). M. Fuwa, S. Takeda, M. Zwierz, H. M. Wiseman, A. Furusawa, *Experimental proof of nonlocal wavefunction collapse for a single particle using homodyne measurements*, [Nat. Commun. **6**, 6665 (2015)](https://10.1038/ncomms7665). H. J. Briegel, D. E. Browne, W. Dür, R. Raussendorf, and M. Van den Nest, *Measurement-based quantum computation*, [Nat. Phys. **5**, 19 (2009)](https://doi.org/10.1038/nphys1157). A. Streltsov, G. Adesso, and M. B. Plenio, *Quantum coherence as a resource*, [Rev. Mod. Phys. **89**, 041003 (2017)](https://doi.org/10.1103/RevModPhys.89.041003). J. Åberg, *Quantifying Superposition*, [arXiv:quant-ph/0612146](https://arxiv.org/abs/quant-ph/0612146). J. Åberg, *Catalytic Coherence*, [Phys. Rev. Lett. **113**, 150402 (2014)](https://doi.org/10.1103/PhysRevLett.113.150402). J. Sperling and W. Vogel, *Convex ordering and quantification of quantumness*, [Phys. Scr. **90**, 074024 (2015)](https://doi.org/10.1088/0031-8949/90/7/074024). A. Winter and D. Yang, *Operational Resource Theory of Coherence*, [Phys. Rev. Lett. **116**, 120404 (2016)](https://doi.org/10.1103/PhysRevLett.116.120404). T. Theurer, N. Killoran, D. Egloff, and M. B. Plenio, *Resource Theory of Superposition*, [Phys. Rev. Lett. **119**, 230401 (2017)](https://doi.org/10.1103/PhysRevLett.119.230401). T. Biswas, M. G. Díaz, and A. Winter, *Interferometric visibility and coherence*, [Proc. Roy. Soc. London A **473**, 20170170 (2017)](https://doi.org/10.1098/rspa.2017.0170). W. Vogel and J. Sperling, *Unified quantification of nonclassicality and entanglement*, [Phys. Rev. A **89**, 052302 (2014)](https://doi.org/10.1103/PhysRevA.89.052302). N. Killoran, F. E. S. Steinhoff, and M. B. Plenio, *Converting Nonclassicality into Entanglement*, [Phys. Rev. Lett. **116**, 080402 (2016)](https://doi.org/10.1103/PhysRevLett.116.080402). E. Chitambar and M.-H. Hsieh, *Relating the Resource Theories of Entanglement and Quantum Coherence*, [Phys. Rev. Lett. **117**, 020402 (2016)](https://doi.org/10.1103/PhysRevLett.117.020402). X. Hu and H. Fan, *Extracting quantum coherence via steering*, [Sci. Rep. **6**, 34380 (2016)](https://10.1038/srep34380). J. Ma, B. Yadin, D. Girolami, V. Vedral, and M. Gu, *Converting Coherence to Quantum Correlations*, [Phys. Rev. Lett. **116**, 160407 (2016)](https://doi.org/10.1103/PhysRevLett.116.160407). A. Streltsov, S. Rana, M. N. Bera, and M. Lewenstein, *Towards Resource Theory of Coherence in Distributed Scenarios*, [Phys. Rev. X **7**, 011024 (2017)](https://doi.org/10.1103/PhysRevX.7.011024). E. Agudelo, J. Sperling, L. S. Costanzo, M. Bellini, A. Zavatta, and W. Vogel, *Conditional Hybrid Nonclassicality*, [Phys. Rev. Lett. **119**, 120403 (2017)](https://doi.org/10.1103/PhysRevLett.119.120403). J. Sperling, T. J. Bartley, G. Donati, M. Barbieri, X.-M. Jin, A. Datta, W. Vogel, and I. A. Walmsley, *Quantum Correlations from the Conditional Statistics of Incomplete Data*, [Phys. Rev. Lett. **117**, 083601 (2016)](https://doi.org/10.1103/PhysRevLett.117.083601). E. Chitambar, A. Streltsov, S. Rana, M. N. Bera, G. Adesso, and M. Lewenstein, *Assisted Distillation of Quantum Coherence*, [Phys. Rev. Lett. **116**, 070402 (2016)](https://doi.org/10.1103/PhysRevLett.116.070402). T. Ma, M.-J. Zhao, S.-M. Fei, and G.-L. Long, *Remote creation of quantum coherence*, [Phys. Rev. A **94**, 042312 (2016)](https://doi.org/10.1103/PhysRevA.94.042312). Y. Yao, G. H. Dong, X. Xiao, M. Li, and C. P. Sun, *Interpreting quantum coherence through a quantum measurement process*, [Phys. Rev. A **96**, 052322 (2017)](https://doi.org/10.1103/PhysRevA.96.052322). E. Schrödinger, *Die gegenwärtige Situation in der Quantenmechanik*, [Naturwiss. **23**, 807 (1935)](https://doi.org/10.1007/BF01491891). S. Weinberg, *Einstein’s Mistakes*, [Physics Today **58**, 31 (2005)](https://doi.org/10.1063/1.2155755). W. H. Zurek, *Decoherence, einselection, and the quantum origins of the classical*, [Rev. Mod. Phys. **75**, 715 (2003)](https://doi.org/10.1103/RevModPhys.75.715). Note that in Ref. [@YDXLS17], the twofold measurement of a single observable was termed *repeatability hypothesis*. M. J. Schervish, *Theory of Statistics* ([Springer, 1995](https://doi.org/10.1007/978-1-4612-4250-5)). B. Yadin, J. Ma, D. Girolami, M. Gu, and V. Vedral, *Quantum Processes Which Do Not Use Coherence*, [Phys. Rev. X **6**, 041028 (2016)](https://doi.org/10.1103/PhysRevX.6.041028). E. Knill and R. Laflamme, *Power of One Bit of Quantum Information*, [Phys. Rev. Lett. **81**, 5672 (1998)](https://doi.org/10.1103/PhysRevLett.81.5672). M. A. Nielsen and I. L. Chuang, *Quantum Computation and Quantum Information* ([Cambridge University Press, 2000](https://doi.org/10.1017/CBO9780511976667)).
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We present a novel and efficient method for fitting dynamical models of stellar kinematic data in dwarf spheroidal galaxies (dSph). Our approach is based on Gaussian-process emulation (GPE), which is a sophisticated form of curve fitting that requires fewer training data than alternative methods. We use a set of validation tests and diagnostic criteria to assess the performance of the emulation procedure. We have implemented an algorithm in which both the GPE procedure and its validation are fully automated. Applying this method to synthetic data, with fewer than 100 model evaluations we are able to recover a robust confidence region for the three-dimensional parameter vector of a toy model of the phase-space distribution function of a dSph. Although the dynamical model presented in this paper is low-dimensional and static, we emphasize that the algorithm is applicable to any scheme that involves the evaluation of computationally expensive models. It therefore has the potential to render tractable previously intractable problems, for example, the modelling of individual dSphs using high-dimensional, time-dependent $N$-body simulations.'
author:
- |
Amery Gration$^{1}$[^1] and Mark I. Wilkinson$^{1}$\
$^{1}$Department of Physics & Astronomy, University of Leicester, University Road, Leicester LE1 7RH
bibliography:
- 'bibliography.bib'
date: 'Accepted XXX. Received YYY; in original form ZZZ'
title: 'Dynamical modelling of dwarf-spheroidal galaxies using Gaussian-process emulation'
---
\[firstpage\]
galaxies: dwarf – galaxies: kinematics and dynamics – methods: statistical
INTRODUCTION
============
In the $\Lambda$CDM model of cosmology, galaxies form by *hierachical growth*, with large galaxies being formed by the agglomeration of smaller ones. Of particular interest are the dwarf-spheroidal (dSph) galaxies that orbit the Milky Way, the dark-matter haloes of which are the smallest dark matter-dominated structures observed, and which are likely to be relics of the earliest stages of galaxy formation. These galaxies are still not fully understood. Although dark matter-only $N$-body simulations of halo-formation using $\Lambda$CDM cosmology predict cusps in the dark-matter density profile of the dSphs (i.e. they predict that density is inversely proportional to radius for small radii), the literature on the modelling of observed dSphs contains claims of both cusps and cores (i.e. it contains claims that density may be constant for small radii) ([@battaglia_kinematic_2008], [@strigari_kinematics_2010], [@breddels_model_2013], and [@read_case_2018]). Either the $\Lambda$CDM model fails at small scales and requires modification ([@ludlow_2016; @lovell_2014]), or dSphs evolve over their lifetimes, under the influence of baryonic feedback. For example, it has been shown that supernova-driven flattening may turn dark-matter cusps into cores ([@navarro_cores_1996], [@read_mass_2005], and [@mashchenko_stellar_2008]). In the latter case we wish better to understand the evolution of dSphs and as such require robust dynamical modelling of their end-states to act as targets for evolutionary simulations.
To date, dynamical modelling of dSphs has for the most part been based on very restrictive simplifying assumptions, namely that the dSphs are spherical and in equilibrium (see, for example, [@wilkinson_dark_2002], [@walker_universal_2009], [@strigari_kinematics_2010], and [@read_how_2017]). While some authors have considered more general models [@breddels_model_2013], relaxation of these assumptions results in models that are significantly more computationally expensive, often prohibitively so. However, it is possible to reduce this computational expense by borrowing the technique of *emulation* from machine learning.
Let us establish some terminology. A *simulation* is a model of some phenomenon. Formally, a model is an indexed set of functions $\{f(\cdot\,;a)\}_{a \in A}$ where $f(\cdot\,;a): X \longrightarrow Y$ is a function, $a$ is called a *parameter* and the set $A$ is called the *parameter space*. Specifically, in this paper, we assume some model of the phase-space distribution function, $f(\cdot\,;a)$, of the stars within a dSph, where $a$ parameterizes the physical model we are using. In this case, $X$ is the phase space and $Y$ is the set of non-negative, real numbers. A computer simulation is just a means of evaluating such a model, a simulation *run* being an evaluation for a single parameter. In galactic dynamics we are typically interested in simulating (i.e. modelling) observations of a galaxy.
When a model is computationally expensive to evaluate we may use a *meta-model*, i.e. a model of the model that is computationally cheaper. An *emulator* is a meta-model together with a measure of confidence in that meta-model’s output. In emulation we perform a small number of runs, each using a different parameter, and then use the resulting data to estimate the output for a parameter that we have not explicitly computed, along with a confidence interval for that estimate. We do this without needing to make an additional simulation run or model evaluation. In this respect, emulation is a matter of *prediction*. But, since we may make a prediction for all points in the parameter space, we can fully map out the function being emulated and in this respect emulation is an efficient means of *curve-fitting*.
One commonly used method of emulation is *Gaussian-process emulation* (GPE). In the astrophysical literature it has been used to fit exoplanetary transit and secondary-eclipse light curves ([@gibson_gaussian_2012] and [@evans_uniform_2015]), to map interstellar extinction within the Milky Way [@sale_three-dimensional_2014], and to fit semi-analytic models of galaxy formation [@bower_parameter_2010], while in the cosmological literature it has has been used to predict the nonlinear matter power spectrum in the Coyote Universe simulation [@heitmann_coyote_2009], and to fit Gravitational-wave models [@moore_improving_2016]. In the dynamical modelling of stellar systems, GPE may allow us the use of more-expensive static models or even full $N$-body models. In this paper we take a step towards this goal by introducing the method and demonstrating its use with a toy model, namely a single-component anisotropic Plummer sphere, which we fit to synthetic data drawn from the same model. We focus on distribution function-based models of the internal dynamics of a galaxy in order to illustrate the value of the GPE approach. However, we note that in principle the GPE approach is applicable to any dynamical modelling scheme.
To avoid confusion, a word of clarification is in order with respect to terminology. The term ‘Gaussian’ in ‘Gaussian-process emulation’ refers only to assumptions within the emulation scheme. It does not imply any assumptions regards the Gaussianity, or otherwise, of any properties of the models themselves. In particular, there is no assumption that the line-of-sight velocity distributions are Gaussian, as is sometimes assumed in the dynamical modelling of dSphs ([@battaglia_kinematic_2008], [@strigari_common_2008], and [@wolf_accurate_2010]).
Our principal interest is in what observed data can tell us about the distribution of dark matter in a dSph. Which dark-matter morphologies do the data rule out? Which best account for the observations? We therefore adopt the maximum-likelihood approach, which allows us to draw robust confidence regions within parameter space. We outline this procedure in section \[sec:likelihood\]. In section \[sec:gpe\] we discuss the use of GPE to estimate the likelihood when it too expensive to for us to perform a parameter sweep. In section \[sec:plummer\] we illustrate the method using our toy model. In section \[sec:conclusion\] we summarize our conclusions.
Likelihood {#sec:likelihood}
==========
In this paper, our representation of a stellar system (in this case a dSph) consists of a parameterized set of functions (i.e. the dynamical model) representing the phase-space probability density function (also called the distribution function). We treat the positions, $\bm{X}$, and the velocities, $\bm{V}$, as random vectors, meaning that the state of a star is represented by the random vector, $\bm{W} := (\bm{X}, \bm{V})$. The distribution function (DF) for a single star is denoted $f_{\bm{W}}(\bm{w})$ [for an extensive discussion, see @binney_galactic_2008].[^2] We assume that the DF is an element of the model $\{f(\bm{w};\bm{a})\}_{\bm{a} \in \bm{A}}$, where the parameters are $D$-dimensional real vectors, the elements of which represent the total galactic mass, galactic scale length, velocity anisotropy, etc. From the DF we may calculate the observable properties of the system. For a dSph these observables are typically the projected stellar positions, (represented by the random variables $X$ and $Y$) and the line-of-sight velocity (represented by the random variable $V_z$). The DF for these observables is given by the marginalization of the phase-space DF: $$\begin{aligned}
f_{\bm{W}_{\mathrm{p}}}(\bm{w}_{\mathrm{p}}; \bm{a}) = \int_{\mathbf{R}^3}
f_{\bm{W}}(\bm{w}; \bm{a}) {\operatorname{d}}z {\operatorname{d}}v_{x} {\operatorname{d}}v_{y}.\end{aligned}$$ We can reasonably assume that the states of stars are independent and identically distributed [@binney_galactic_2008] so that the joint marginalized PDF for $N$ stars is $$\begin{aligned}
f_{(\bm{W}_{\mathrm{p}, 1}, \ldots, \bm{W}_{\mathrm{p},
N})}(\bm{w}_{\mathrm{p}, 1}, \ldots, \bm{w}_{\mathrm{p}, N}; \bm{a}) =
\prod_{i = 1}^{N} f_{\bm{W}_{\mathrm{p}}}(\bm{w}_{\mathrm{p}, i}; \bm{a}).\end{aligned}$$ By comparing these observables with a data set, we can then optimize the parameters of the phase-space DF using the maximum likelihood method.
By definition, the likelihood of model parameter $\bm{a}$ is $$\begin{aligned}
L(\bm{a}; \bm{w}_{\mathrm{p}, 1}, \ldots, \bm{w}_{\mathrm{p}, N}) =
f_{(\bm{W}_{\mathrm{p}, 1}, \ldots, \bm{W}_{\mathrm{p},
N})}(\bm{w}_{\mathrm{p}, 1}, \ldots, \bm{w}_{\mathrm{p}, N}; \bm{a}).
\label{eq:likelihood}\end{aligned}$$ We recover the parameter by maximizing this function for given data, namely the observed values of $\bm{w}_{\mathrm{p}, 1}, \ldots, \bm{w}_{\mathrm{p},
N}$. We denote the maximum-likelihood estimate (MLE) by $\hat{\bm{a}}$. The MLE is itself the realization of a random variable, which we denote $\hat{\bm{A}}$. It is asymptotically normal (i.e. it is normally distributed in the infinite-data limit) with mean $\bm{a}_0$ (the true parameter) and variance $\bm{I}(\bm{a}_0)^{- 1}$, where the *expected Fisher information matrix* is $$\begin{aligned}
\bm{I}(\bm{a}_0) = - {\operatorname{E}}\left( \dfrac{\partial^2 \ln L(\bm{a}_0;
\bm{W}_{\mathrm{p}, 1}, \ldots, \bm{W}_{\mathrm{p}, N})}{\partial \bm{a}
\partial \bm{a}^{\mathrm{t}}} \right)\end{aligned}$$ i.e. the negative of the expectation of the Hessian of the log-likelihood [see any text-book on the subject, for example @wasserman_all_2007]. The true parameter $\bm{a}_0$ may be approximated by $\hat{\bm{a}}$, and the expected Fisher information matrix by the *observed Fisher information matrix*, $$\begin{aligned}
\bm{J}(\hat{\bm{a}}) = - \dfrac{\partial^2 \ln L(\hat{\bm{a}};
\bm{w}_{\mathrm{p}, 1}, \ldots, \bm{w}_{\mathrm{p}, N})}{\partial \bm{a}
\partial \bm{a}^{\mathrm{t}}}.
\label{eq:fisher_information}\end{aligned}$$ In summary, the MLE is distributed as $$\begin{aligned}
\hat{\bm{A}} \sim N(\hat{\bm{a}}, \bm{J}^{-1}(\hat{\bm{a}})).\end{aligned}$$ This allows us to determine the confidence region, $C$, defined by the boundary that is the solution to the equation $$(\bm{a} - \hat{\bm{a}})^\mathrm{t} \bm{J}^{-1}(\hat{\bm{a}}) (\bm{a} -
\hat{\bm{a}}) = \chi_D^2(1 - \alpha)
\label{eq:likelihood_confidence}$$ where $\chi_D^2$ is the quantile function (i.e. the inverse of the cumulative distribution function) for the chi-squared distribution for $D$ degrees of freedom and $\alpha$ is the critical value, i.e. the value such that $C$ traps $\bm{a}_0$ with probability $\alpha$. The Fisher information quantifies the curvature of the likelihood function at its maximum value and hence the breadth of the distribution’s peak. A narrow peak (i.e. large curvature and large Fisher information) indicates that the maximum is well constrained. A broad peak (i.e. small curvature and small Fisher information) indicates that the maximum is poorly constrained.
Once we have the MLE of the parameters and the observed Fisher information matrix, we may compute the distribution of the MLE of any function of the parameters using the delta method [@wasserman_all_2007]. This states that any real-valued function of the parameters, $g: \bm{A} \longrightarrow
\bm{R}$, is distributed normally with mean $g(\hat{\bm{a}})$ and variance $$\begin{aligned}
\dfrac{\partial g(\hat{\bm{a}})}{\partial \bm{a}^\mathrm{t}}
\bm{J}^{-1}(\hat{\bm{a}}) \dfrac{\partial g(\hat{\bm{a}})}{\partial
\bm{a}}.
\label{eq:delta_method}\end{aligned}$$ For us $g$ will be some function of the phase-space DF (for example the galactic dark-matter profile, or Binney’s anisotropy parameter).
The likelihood (equation \[eq:likelihood\]) will not usually be expressible in closed form. Both the likelihood and its derivative may be costly to evaluate. In such circumstances we may reduce the computational burden by constructing a cheaper approximation of the likelihood using GPE.
Gaussian-process emulation {#sec:gpe}
==========================
At the heart of GPE [@ohagan_curve_1978; @sacks_design_1989] are random processes, which are sets of random variables. A realization of a random process (i.e. a realization of each individual random variable) is a function, meaning that we may use Gaussian-processes in the context of curve-fitting. More rigorously, a random process, $\{Y(\bm{a})\}_{\bm{a} \in
\bm{A}}$, is an indexed set of random variables. Again, we call an index, $\bm{a}$, a *parameter* (or *parameter vector* if $\bm{A}$ is multidimensional), and we call the index set, $\bm{A}$, the *parameter space*. A random process is *Gaussian* if every finite subset of the process has a multivariate normal distribution. A process is completely specified by the joint PDF for every finite subset of the process. Since the joint PDF of a multivariate normal distribution is completely specified by its mean and covariance, a Gaussian process is completely specified by its mean and covariance. The mean is the function $m: A \longrightarrow \bm{R}$ such that $m(\bm{a}) = {\operatorname{E}}(Y(\bm{a}))$. The covariance is the function $k:
\bm{A}^2 \longrightarrow \bm{R}$ such that $k(\bm{a}, \bm{a}') = {\operatorname{cov}}(\bm{a},
\bm{a}')$. If a process, $Z$, is Gaussian we write $$\begin{aligned}
Z \sim \mathrm{GP}(m(\bm{a}), k(\bm{a}, \bm{a}')).\end{aligned}$$
Recall that in curve-fitting (or, *regression* as it is called in the statistical literature) we have a random variable $\bm{A}$ (called the independent variable) and a random variable $Y$ (called the dependent variable), and wish to find the relationship between them. Given *training data* $((\bm{a}_{i}, y(\bm{a}_{i})))_{i = 1}^N$ we may always write the relationship in the form $$\begin{aligned}
Y(\bm{a}_i) = r(\bm{a}_i) + E(\bm{a}_i)
\label{eq:regression_model}\end{aligned}$$ where the *regression function* is $r(\bm{a}) := {\operatorname{E}}(Y(\bm{a}))$ for all $\bm{a}$ [@wasserman_all_2007], and $E(\bm{a}_i)$ is a random variable (the *noise*.
We seek an estimator for $r$, denoted $\hat{r}$. In *parametric* regression we assume that this estimator is an element of some parameterized set of functions, $\hat{r} \in \{f_{\bm{\beta}} \,|\, \bm{\beta} \in
\bm{B}\}$. We call this set the *regression model* and the set $\bm{B}$ the set of *regression parameters*. For example, in linear regression we assume that this model is $\{\beta_{0} + \sum_{i = 1}^{D} \beta_{i} a_{i}
\,|\, \beta_{0}, \ldots, \beta_{D} \in \bm{R}\}$. Regression is *non-parametric* if we may not write the estimator in this way. Non-parametric regression is useful when we have no motivation for a parameterized expression for the regression function. In Gaussian-process regression we assume that $$\begin{aligned}
Y \sim {\operatorname{GP}}(r, \sigma^2)
\label{eq:gpe_assumption}\end{aligned}$$ whereupon $r = {\operatorname{E}}(Y)$, as required. Then $E(\bm{a}) = Y(\bm{a}) -
r(\bm{a})$, i.e. $E$ is a Gaussian process with zero mean. We will assume that its covariance is the sum of two terms, i.e. that $\sigma^{2}(\bm{a},
\bm{a}') = k(\bm{a}, \bm{a}') + \sigma^{2}_{\epsilon}(\bm{a})$, the first term representing the signal and the second term representing the noise (i.e. the error in the dependent variable). Note that the noise is not in general constant, but is a function of $\bm{a}$. In statistical parlance, we would say that the errors are *heteroscedastic* rather than *homoscedastic*.
Suppose we wish to predict the dependent variable $Y(\bm{a})$ for some value $\bm{a} \in \bm{A}$. By hypothesis, the distribution of this random variable is univariate normal, i.e.$$\begin{aligned}
Y(\bm{a}) \sim N(r(\bm{a}), \sigma^{2}(\bm{a}, \bm{a})),\end{aligned}$$ and the joint distribution of the sample is multivariate normal, i.e. $$\begin{aligned}
\bm{Y} \sim N(\bm{r}, \bm{K}),
\label{eq:sample_distrib}\end{aligned}$$ where $\bm{r} := (r(\bm{a}_{1}), r(\bm{a}_{2}), \ldots, r(\bm{a}_{N}))$, and $[\bm{K}]_{ij} = k(\bm{a}_{i}, \bm{a}_{j}) + \sigma_{\epsilon}^{2}(\bm{a}_{i})\delta_{ij}$. The estimator for our regression function is the mean of the conditional random variable $$\begin{aligned}
Y(\bm{a}) \,|\, (\bm{Y} = \bm{y}) \sim N(\hat{r}, \hat{\sigma}^2)
\label{eq:gpe_conditional}\end{aligned}$$ where $[\bm{y}]_{i} = y(\bm{a}_{i})$, i.e. the expected value of $Y(\bm{a})$ given the training data. It is a standard result [@ohagan_curve_1978] that $$\begin{aligned}
\hat{r}(\bm{a})
&= r(\bm{a}) + \bm{k}^\mathrm{t}(\bm{a})\bm{K}^{- 1} (\bm{y} - \bm{r}), \text{ and}
\label{eq:gpe_mean} \\
\hat{\sigma}^2(\bm{a})
&= k(\bm{a}, \bm{a}) - \bm{k}^\mathrm{t}(\bm{a})\bm{K}^{- 1} \bm{k}(\bm{a}) \label{eq:gpe_var}\end{aligned}$$ where $[\bm{k}(\bm{a})]_i = k(\bm{a}_i, \bm{a})$. These two equations are the principal results of GPE. The estimator, $\hat{r}$, is the sum of two terms. The first is the regression function, and the second a smoothing of the residuals, $(\bm{y} - \bm{r})$. In fact this smoothing is a weighted sum of the residuals, where the weights are the elements of the vector $\bm{k}^\mathrm{t}(\bm{a})\bm{K}^{- 1}$.
Optimizing the emulator {#sec:optimizing_hyperparams}
-----------------------
To evaluate equations \[eq:gpe\_mean\] and \[eq:gpe\_var\], we must assume a mean function, $r$, and a covariance function, $k$. We will choose these from a model, i.e. we will assume that $r \in \{r_{\bm{\mu}} \,|\, \bm{\mu} \in \bm{M}\}$ and $k \in \{k_{\bm{\nu}} \,|\, \bm{\nu} \in \bm{N}\}$, where we call $\bm{M}$ and $\bm{N}$ sets of *hyperparameters*. We optimize our choice of hyperparameters using their maximum likelihood. By equation \[eq:sample\_distrib\] the PDF is $$\begin{aligned}
\begin{split}
&f_{\bm{Y}}(\bm{y}; \bm{\theta})\\
&\quad = \dfrac{1}{\sqrt{(2 \pi)^{N} |\bm{K}|}} \exp \left(-\dfrac{1}{2} (\bm{y} - \bm{r})^{\mathrm{t}} \bm{K}^{-1} (\bm{y} - \bm{r}) \right)
\end{split}\end{aligned}$$ where the hyperparameter vector $\bm{\theta} := (\bm{\mu}, \bm{\nu})$. (Note that $\bm{K}$ and $\bm{r}$ depend on $\bm{\theta}$.) By definition, the likelihood is $$\begin{aligned}
L_{\bm{\Theta}}(\bm{\theta}; \bm{y}) = f_{\bm{Y}_{N}}(\bm{y}; \bm{\theta})\end{aligned}$$ and hence $$\begin{aligned}
&\ln L_{\bm{\Theta}}(\bm{\theta}; \bm{y}) = - \dfrac{1}{2} (\bm{y} - \bm{r})^{\mathrm{t}} \bm{K}^{-1} (\bm{y} - \bm{r}) - \dfrac{1}{2} \ln |\bm{K}| - \dfrac{N}{2} \ln 2 \pi
\label{eq:hyperparam_max_likelihood}\end{aligned}$$ [@rasmussen_gaussian_2006]. To find the maximum-likelihood estimate of $\bm{\theta}$, namely $\hat{\bm{\theta}} = (\hat{\bm{\mu}}, \hat{\bm{\nu}})$, we may maximize this function subject to the constraint that $\bm{K}$ is positive-semidefinite (positive-semidefiniteness being a necessary property of covariance matrices). The equation for $\ln L_{\Theta}(\bm{\theta}; \bm{y})$ consists of three terms. The first is a measure of fit quality, the second is a complexity penalty, and the third a normalization constant.
The complexity penalty is a function of $\bm{K}$ only, and quantifies the complexity of our attempted fit independent of the data. For a complicated fit, the covariance of any two points is low. Hence the determinant of the covariance matrix $\bm{K}$ is small, and $\ln |\bm{K}|$ diverges with complexity (i.e. as $|\bm{K}| \longrightarrow 0$ so $\ln |\bm{K}| \longrightarrow - \infty$). This strongly penalizes complex models. [For a full discussion see @rasmussen_gaussian_2006 section 5.4.1.] The function $L_{\Theta}$ will in general have multiple maxima, each maximum giving a different tradeoff between fit quality and complexity.
The function $r_{\bm{\mu}}$ might be a finite linear combination of basis functions, $(\xi_i)_{i = 1}^N$, and $\bm{\mu}$ the set of coefficients: $$\begin{aligned}
r(\bm{a}; \bm{\mu})
&= \sum_{i = 1}^{N} \mu_{i} \xi_{i}(\bm{a})\\
&= \bm{\mu}^{\mathrm{t}} \bm{\xi}(\bm{a})
\label{eq:gpe_basis}\end{aligned}$$ where $[\bm{\mu}]_{i} = \mu_{i}$ and $[\bm{\xi}(\bm{a})]_{i} = \xi_{i}(\bm{a})$. However, we are free to choose any function for $r$. Specifically we may choose the zero function, such that $r(\bm{a}) = 0$ for all $\bm{a} \in \bm{A}$, whereupon the Gaussian-process fit to the residuals (the second term in equation \[eq:gpe\_mean\]) does all the work of the regression. It is common practice to do this [@rasmussen_gaussian_2006 section 2.7], and we will do so for the rest of this paper. [^3]
The covariance function, $k_{\bm{\nu}}$, is usually chosen from one of a number of standard functions. It is a function of the parameter space $\bm{A}$ and we will need to impose some structure of this space. For our purposes, $\bm{A} = \bm{R}^D$, treated as linear vector space of dimension $D$, together with a pseudo-metric, $d: \bm{A}^2 \longrightarrow \bm{R}$, i.e. a symmetric positive-semidefinite function of two variables that obeys the triangle inequality. [^4] It is common to assume that the pseudo-metric may be written in quadratic form, i.e. it is common to assume that $$\begin{aligned}
d^2(\bm{a} - \bm{a}') = (\bm{a} - \bm{a}')^\mathrm{t} \bm{M} (\bm{a} - \bm{a}'),
\label{eq:metric}\end{aligned}$$ for positive-semidefinite matrix $\bm{M}$.
The most common covariance function is the *squared-exponential* (SE), $$k_{\mathrm{SE}}(\bm{a}, \bm{a}') = \sigma_{\mathrm{SE}}^2 \exp \left(-\dfrac{1}{2} d^2(\bm{a} - \bm{a}') \right),
\label{eq:cov_SE}$$ where $\sigma_{\mathrm{SE}}^{2}$ is the *signal variance* (the value of $k_\mathrm{SE}$ when $\bm{a}' = \bm{a}$). Furthermore, it is normally assumed that the pseudo-metric matrix is diagonal, i.e. that $\bm{M} = {\operatorname{diag}}(m_{1}, m_{2}, \ldots , m_{D})$. Using equations \[eq:metric\] and \[eq:cov\_SE\] we may show [see, for example, @loeppky_choosing_2009] that the mean squared gradient is $${\operatorname{E}}\left( \frac{\upartial Y(\bm{a})}{\upartial a_i} \right)^2 = 2 \sigma_\mathrm{SE}^2 m_i,$$ and therefore call $m_i$ the *sensitivity* of the model to the $i$-th parameter. Because $\bm{M}$ is positive-semidefinite it has a unique positive-semidefinite inverse, which in turn has a unique square root, i.e. there exists a unique matrix $\bm{L}$ (not to be confused with the likelihood, $L$) such that $\bm{M} = \bm{L}^{-2}$. If $\bm{M}$ is diagonal then so is $\bm{L}$ and $\bm{L} = {\operatorname{diag}}(l_1, l_2, \ldots , l_D)$ where $m_i = l_i^{-2}$ for all $i$. We can see that equation \[eq:cov\_SE\] is formally identical to a Gaussian with covariance $\bm{M}^{-1}$. We therefore identify $\bm{L}^{2}$ as a covariance matrix and call the element $l_i$ the *correlation length* (also *scale length*) for the $i$-th parameter. The elements of $\bm{M}$ together with $\sigma_{\mathrm{SE}}^2$ and the coefficients of our basis functions form the hyperparameter vector, $\bm{\theta} = \{\sigma_{\mathrm{SE}}^2, \mu_{1}, \ldots, \mu_{N}, m_{1}, \ldots , m_{D}\}$, of the Gaussian-process estimator. One advantage of assuming that $r$ is the zero function is that this hyperparameter vector becomes shorter: $\bm{\theta} = \{\sigma_{\mathrm{SE}}^2, m_{1}, \ldots , m_{D}\}$. This means that the dimension of the domain of the hyperparameter likelihood is smaller and hence the hyperparameter vector is easier to optimize.
If $r$ is expanded according to equation \[eq:gpe\_basis\] and we use the squared-exponential covariance function then the mean-square error of the estimator, i.e. the mean-square difference of $\hat{r}$ and $r$ (Sacks et al. 1989), is $$\begin{aligned}
\begin{split}
&\operatorname{MSE}(Y(\bm{a}) \,|\, (\bm{Y} = \bm{y})) = \\
&\sigma_\mathrm{SE}^2
\bigg(1 -
\begin{bmatrix}
\bm{\xi}^{\mathrm{t}}(\bm{a}) & \sigma_\mathrm{SE}^{-2} \bm{k}^{\mathrm{t}}(\bm{a})
\end{bmatrix}
\begin{bmatrix}
\bm{0} & \bm{\Xi}^{\mathrm{t}}\\
\bm{\Xi} & \sigma_\mathrm{SE}^{-2} \bm{K}
\end{bmatrix}^{\mathrm{-1}}
\begin{bmatrix}
\bm{\xi}(\bm{a})\\
\sigma_\mathrm{SE}^{-2} \bm{k}(\bm{a})
\end{bmatrix}
\bigg)
\end{split}
\label{eq:mse}\end{aligned}$$ where $[\bm{\Xi}]_{i} = \bm{\xi}^{\mathrm{t}}(\bm{a})$. In the case that $\bm{\xi}(\bm{a}) = \bm{0}$ (as we assume here) this expression reduces to the estimator variance (equation \[eq:gpe\_var\]). We may therefore construct a confidence interval for the estimator, $\hat{r}(\bm{a})$, using the root mean-square error (RMSE), namely the interval $$\begin{aligned}
C = \left(\hat{r}(\bm{a}) - \Phi^{-1}(1 - \alpha / 2) \hat{\sigma}(\bm{a}), \hat{r}(\bm{a}) + \Phi^{-1}(1 - \alpha / 2) \hat{\sigma}(\bm{a})\right)
\label{eq:confidence_interval}\end{aligned}$$ where $\Phi$ is the CDF for the univariate normal distribution and $\alpha$ is the critical value. If there are no errors associated with the training data then the RMSE of the estimator is zero at the training points and increases as the distance of the test point from a training point increases. If the errors on the sample are constant, then we expect the RMSE to be approximately constant (except at the boundaries where it will be greater on account of boundary bias).
Validating the emulator {#sec:validating_the_emulator}
-----------------------
The machinery of Gaussian-process regression assumes that the covariance function, $k$, is known. In practice, it never is. We must chose an approximation to the covariance, invariably from a list of standard covariance functions, as discussed above. We would like to know that this covariance function has been well chosen, i.e. we would like to assess the performance of the emulator given our choice of covariance function. We may do this using leave-one-out cross-validation (LOOCV) [@wasserman_all_2007]. We note that the estimator (equation \[eq:gpe\_mean\]) is distributed normally with variance equal to the MSE (equation \[eq:mse\]), and expect that for a well-specified covariance this distribution will be observed in our estimator. Ideally we would evaluate our dynamical model at a large number of test points, and compare the distribution of the residuals of our predictions with that of the MSE. This, of course, imposes an impractical computational burden. Instead, we omit the $i$-th pair, $(\bm{a}_i, y(\bm{a}_i))$, from our training data to give the reduced data, $(\bm{a}_i, y(\bm{a}_i))_{i \neq i}$. Using these data we then compute an estimate for $y$ at the omitted point, $\bm{a}_i$, finding that (see equation \[eq:gpe\_conditional\]) $Y(\bm{a}_i) | (\bm{Y}_{-i} = \bm{y}_{-i}) \sim N(\hat{y}_{-i}(\bm{a}_i), \hat{\sigma}_{-i}^2(\bm{a}_i))$. This gives us $N$ residuals, which we may compare with the MSE. We expect that the *standardized predicted error* [@wasserman_all_2007], $$e_{-i}(\bm{a}_i) := \frac{y(\bm{a}_i) - \hat{r}_{-i}(\bm{a}_i)}{\hat{\sigma}_{-i}(\bm{a}_i)},
\label{eq:predicted_error}$$ is normally distributed with zero mean and unit variance.[^5]
One way to compare two distributions is by means of a *quantile-quantile plot*, in which we plot the quantiles of one distribution against the quantiles of the other. If the two distributions are the same, then the points will lie on the diagonal. If either distribution is empirical we may substitute its ordered observed values for its quantiles. To check that the standardized residuals are distributed as required, we therefore plot the $N$ ordered residuals of our fit against the $N$-th quantiles of the normal distribution. If the covariance function has been well-specified we expect them to be evenly scattered along the diagonal. We also require that the residuals are small, and hence calculate the *mean squared predicted error* (also *leave-one-out cross-validation score*), $$R := \frac{1}{N} \sum_{i = 1}^{N} (y(\bm{a}_i) - \hat{y}_{-i}(\bm{a}_i))^2,
\label{eq:cv_score}$$ which tends to zero in the infinite-training data limit, as well as plot the predicted values against the true values. Furthermore, we require that the residuals exhibit no trend, and hence plot the standardized predicted errors against the associated predicted values for $y$. Finally, we require that there are no significant outliers (greater than three sigma, or $e_{-i} > 3$, say), which would indicate that the estimator is underperforming is certain regions of parameter space.
We expect poor accuracy in our estimator when neighbouring points are poorly correlated, i.e. when the scale of features in our function is approximately equal to or less than the point separation. We also expect the accuracy to be poor at the boundary of parameter space, where the model is constrained by data on one side only. We may consider a point to be near the boundary of parameter space if it is within one parameter length scale of it.
We may in fact use LOOCV to optimize the hyperparameter vector, as opposed to the maximum-likelihood method described in section \[sec:gpe\]. Note that the logarithm of the likelihood of the hyperparameter vector when leaving out the $i$-th pair of data (called the *leave-one-out likelihood*) is $$\begin{aligned}
\ln L_{-i, \bm{\Theta}}(\bm{\theta}; \bm{y}) = - \dfrac{1}{2} \ln \sigma_{-i}^2(\bm{a}_i) - \dfrac{1}{2} e_{-i}^2(\bm{a}_i) - \dfrac{1}{2} \ln 2 \pi\end{aligned}$$ [@rasmussen_gaussian_2006]. Hence the logarithm of the joint leave-one-out likelihood (LOO likelihood) is $$\begin{aligned}
\ln L_{\mathrm{LOO}, \bm{\Theta}} = \sum_{i = 1}^N \ln L_{\mathrm{pseudo}, \bm{\Theta}}(\bm{\theta}; \bm{y}),\end{aligned}$$ which we may maximize to give the *maximum LOO likelihood estimate* of the hyperparameter vector. This estimate is more robust to mispecification of the covariance function [@bachoc_cross_2013] than the maximum-likelihood estimate (equation \[eq:hyperparam\_max\_likelihood\]).
Conditioning the likelihood {#sec:conditioning}
---------------------------
If the estimator fails validation then the covariance function has been misspecified. In this case we may do one of two things: choose a different covariance function, or transform the data so that the function is better suited to its task. Here we restrict ourselves to using the squared-exponential covariance function (equation \[eq:cov\_SE\]), and hence consider the second approach.
We have assumed that our data is drawn from a Gaussian random process with covariance $k_{\mathrm{SE}}(\bm{a}, \bm{a}') + \sigma_\epsilon^{2}(\bm{a})$ (equation \[eq:gpe\_assumption\]). Crucially, the squared exponential covariance function is a function of the difference of its arguments only and the properties of the random process are therefore independent of the absolute values its arguments. In particular, the variance and autocovariance of the random process are constant. The fact that the variance is constant means that noise-free data are identically normal. If we do not observe this distribution in our training data we may transform it to ensure that we do. Such a transformation is said to be *variance stabilizing*.
One such variance-stabilizing transformation is the *Box-Cox transformation*. Let $\{y(\bm{a}_i)\}_{i = 1}^N$ be a set of observations. Then the Box-Cox transformation of the observations [@box_analysis_1964] is the function $g$ such that $$\begin{aligned}
g(y(\bm{a}_i); \lambda_1, \lambda_2) = \dfrac{(y(\bm{a}_i) + \lambda_2)^{\lambda_1} -
1}{\lambda_1}\end{aligned}$$ for some real $\lambda_1$, $\lambda_2$ such that $\lambda_{2} > - y(\bm{a}_{i})$ for all $i$. Note that this expression is just a scaled power law, with the scaling chosen such that $\lim_{\lambda_1 \longrightarrow 0} g(y(\bm{a}_i)) = \ln(y(\bm{a}_i) + \lambda_2)$.
Box and Cox give the following theorem concerning the choice of the parameters $\lambda_1$ and $\lambda_2$. Assume that each observation $y(\bm{a}_{i})$ is a realization of the random variable $G_i$, and that each transformed observation, $g(y(\bm{a}_{i}))$ is a realization of the random variable $G'_{i}$. Furthermore assume that $G'_{1}, \ldots, G'_{N}$ are independent and identically normal, i.e. assume that for all $n$, $$\begin{aligned}
G'_{i} \sim N(\mu_\mathrm{t}^{\phantom{2}}, \sigma_\mathrm{t}^2).
\label{Box_Cox_assumption}\end{aligned}$$ for some mean, $\mu_\mathrm{t}^{\phantom{2}}$, and variance $\sigma_\mathrm{t}^2$. Then the joint distribution of the (untransformed) observations has PDF $$\begin{aligned}
\begin{split}
&f_{(G_{1}, \ldots, G_{N})}(y(\bm{a}_{1}), \ldots, y(\bm{a}_{N}); \lambda_1, \lambda_2) \\
&= \prod_{i = 1}^{N} \dfrac{1}{\sqrt{2 \pi \sigma_\mathrm{t}^2}} \exp \left( \dfrac{\sum_{i = 1}^{N} (g(y(\bm{a}_{i})) - \mu_\mathrm{t})^2}{2 \sigma_\mathrm{t}^2} \right) J(y(\bm{a}_{i}); \lambda_1, \lambda_2),
\end{split}\end{aligned}$$ where the Jacobian $$\begin{aligned}
J(y(\bm{a}_{i}); \lambda_1, \lambda_2)
&= \left| \dfrac{{\operatorname{d}}g(y(\bm{a}_i))}{{\operatorname{d}}y(\bm{a}_i)} \right|\\
&= (y(\bm{a}_i) + \lambda_2)^{(\lambda_1 - 1)}.\end{aligned}$$ Thus the log-likelihood of the parameters $\lambda_1$ and $\lambda_2$ is given by $$\begin{aligned}
\begin{split}
&\ln L_{(\lambda_1, \lambda_2)}(\lambda_1, \lambda_2; y(\bm{a}_{1}), \ldots, y(\bm{a}_i)) = - \frac{N}{2} \ln \sigma_\mathrm{t}^2 - \frac{N}{2}\log 2 \pi \\
& \quad - \frac{1}{2\sigma_\mathrm{t}^2}\sum_{i = 1}^{N} (g(y(\bm{a}_i)) - \mu_\mathrm{t})^2 + (\lambda_1 - 1) \sum_{i = 1}^{N} \ln(y(\bm{a}_i) + \lambda_2)
\end{split}
\label{eq:Box_Cox_likelihood}\end{aligned}$$ where we may substitute for $\mu_\mathrm{t}^{\phantom{2}}$ and $\sigma_\mathrm{t}^2$ their maximum-likelihood estimates, $$\begin{aligned}
\hat{\mu}_{\mathrm{t}} &= \frac{1}{N} \sum_{i = 1}^N g(y(\bm{a}_i)) \text{ and}\\
\hat{\sigma}_\mathrm{t}^2 &= \dfrac{1}{N} \sum_{i = 1}^N (g(y(\bm{a}_i)) -
\hat{\mu}_{\mathrm{t}})^2.\end{aligned}$$ The MLE of the transformation parameters has an associ ated distribution, but it is common to consider it known, and not propagate this uncertainty through the subsequent analysis. It is also common to round off the value of $\lambda_1$ to the nearest half-integer, e.g. to use one of the values $2, 1, 1 / 2, 0, -1 / 2, -1,$ or $-2$ (the square, identity, square root, logarithm, reciprocal square root, reciprocal, or reciprocal square) which give the transformation a ready interpretation. The natural transformation of the likelihood (\[eq:likelihood\]) is to its logarithm together with an offset that ensures this logarithm is defined when the likelihood is zero. In this case the likelihood of the offset, $\lambda_2$, is strictly decreasing, and its optimization will fail. In such a case we may use some arbitrary small value for $\lambda_2$, say the smallest non-zero member of the sample.
We must be careful about the direction of the implication in this theorem. It is not the case that the maximum-likelihood estimates of $\lambda_1$ and $\lambda_2$ ensure that the transformed variables are normal. We must compute these maximum-likelihood estimates and then *check* that the assumption of normality is met by inspecting a histogram of the transformed values. In general the condition is not met, i.e. there exists no power transformation such that $G'_{1}, \ldots, G'_{N} \sim N(\mu_\mathrm{t}, \sigma_\mathrm{t}^2)$. However we may still use the power transformation to *regularize* $G$, i.e. to bring its distribution closer to normal [@draper_distributions_1969].
When using the zero regression function the residuals $\bm{y} - \bm{r}$ (equation \[eq:gpe\_mean\]) may be poorly behaved. They may all have the same sign, or may contain outliers. The use of a transformation overcomes these problems, and avoids the need to specify a regression function, of which we have very little prior knowledge.
Having stabilized the variance of the data, we must also ensure the autocovariance is approximately constant. We do this by transforming the parameter space (i.e. by reparameterizing the random process). This is an altogether more difficult problem [see, for example @sampson_nonparametric_1992] as the length scale $l_i$ (and hence sensitivity $m_i$) is a function of the parameter $a_i$. We will assume that for positive parameters the sensitivity $m_i$ is monotonically decreasing in $a_i$ and converges on zero, i.e. we will assume that the model is sensitive to changes in the parameters when they are small, but insensitive when they are large, or, equivalently, that the likelihood is slowly varying for arbitrarily large parameter values, when it is approximately zero, but quickly varying when the parameter values are small, when it is significantly non-zero. We therefore make a logarithmic transformation of the parameters. If our assumption is wrong, this transformation will not improve the accuracy of our predictions. We must therefore rely on the validation process to tell us if we have made a useful transformation. We must accept that the covariance function will always be misspecified to some degree. If the misspecification is so significant that the model fails validation, we have to choose a different covariance function, i.e. we have to choose a covariance function that is not the squared-exponential. We do not consider such a situation here, but note that there is a large literature on the subject [see, for example @rasmussen_gaussian_2006].
Training the emulator {#sec:training}
---------------------
Assuming we are able to choose some region of parameter space for which we wish emulate the likelihood, we must chose a *design* (i.e. an arrangement of points in parameter space at which we will sample the function). If *a priori* we know nothing about our function (beyond our assumptions of continuity and smoothness encoded in our choice of covariance function) we wish the design to be *space-filling*, i.e. to have uniform density throughout parameter space. We also wish all projections of the design onto lower-dimensional subspaces to be space-filling, as the model may have low sensitivity to some parameters. Lattices are a poor choice for such a design, as their size grows exponentially with the dimension of the parameter space. The most commonly-used designs satisfying the above space-filling requirements are *Latin hypercubes* [@mckay_comparison_1979]. In Latin hypercube sampling (LHS), a parameter space is partitioned into a hypergrid of $N^D$ cells and $N$ points are placed in these cells such that there is only one point in any hyper-row or hyper-column of cells. We may optimize the space-filling property of the design by maximizing the minimum separation of pairs of points in all projections of the design onto lower-dimensional subspaces [@santner_design_2003]. LHS designs are restricted to rectangular regions. It is possible to form space-filling designs on nonrectangular regions (e.g. [@draguljic_noncollapsing_2012]) but we do not consider these here.
We choose the size of our design, $N$, so that the GPE model has acceptable accuracy. This size depends on the difficulty of the problem, i.e. the complexity of the function we are emulating. The more difficult the problem, the greater $N$ will need to be. The question obviously arises: how do we choose an appropriate value of $N$ for a particular problem?
We note that $N$ is satisfactory if the MSE (equation \[eq:mse\]) is small, and that the MSE is a function of $\bm{M}$ (through the covariance function, $k$), $N$ (through the size of the matrix $\bm{K}$, and $D$ (the dimensional of the parameter space). We wish to understand the relationship between these quantities. To this end, Loeppky *et al.* (2008) introduce the *total sensitivity*, $$\label{eq:tau}
\tau := \sum_{i = 1}^D m_i,$$ and the *sparsity*, $$\label{eq:psi}
\psi := \sum_{i = 1}^D m_i^2,$$ where $\{m_i\}_{i = 1}^D$ is the set of elements of the metric matrix. Recall that the length scale $l_i$ is defined such that $m_i = l_i^{-2}$. Consider the squared separation of a pair of design points, $\bm{a}_i$ and $\bm{a}_j$, namely $d^2(\bm{a}_i, \bm{a}_j)$. For a random LHS design, this separation is the realization of a random variable, $H$. Loeppky *et al.* show that for such a design $H$ is distributed with mean $E(H) = \mu(N) \tau$ and variance $\operatorname{var}(H) = \nu(N) \psi$ where $\mu$ and $\nu$ are weak and strictly decreasing functions of $N$ that converge to a positive constant. The accuracy of our emulator will be good when $E(H)$ is small (i.e. when the mean separation of sample points is small, and hence mean correlation is good) and when $\operatorname{var}(H)$ is large (i.e. when many pairs of points have separations smaller than the mean and are hence even better correlated).
If we minimize $\psi$ whilst keeping $\tau$ constant (i.e. if we minimize $\psi$ subject to the constraint $\sum_i m_i = c$ for some real $c$) we find that $m_j = c / D$ for all $j$, i.e. we find that $\psi$ is a minimum (and hence the accuracy poor) when the parameters are equally active, and hence $\psi = c^2 / D$. On the other hand, if we maximize $\psi$ whilst keeping $\tau$ constant we find that $m_j = c$ for some $j$ and $m_i = 0$ for $i \neq j$, i.e. we find that $\psi$ is a maximum (and hence the accuracy good) when only one parameter is active, and hence $\psi = c$. For fixed $\tau$, therefore, $\psi$ quantifies the sparsity of the matrix $\bm{M}$. In the case that all parameters are equally active it is the case that $E(H) = D c$ and that $\operatorname{\psi} = D c^2$, i.e. that both the mean and the variance of the separation are proportional to the number of parameters, and that for a sufficiently large number, the accuracy will be poor.
The accuracy of our estimator depends on both the total sensitivity and the sparsity. It does not depend on the total number of parameters but rather on the number of active parameters. Suppose that the parameter space has been mapped to a hypercube of side $h$. Motivated by practical experiment, Loeppky *et al.* propose that if $\tau h^2 = 3$ then the problem is “easy”, and if $\tau h^2= 40$ the problem is “very difficult”. If $\tau h^2 = 10$ the the problem will be tractable if $\psi$ is small but intractable if $\psi$ is large. Moreover, for easy problems the convergence of $R$ to one is fast whereas for difficult problems the convergence is slow. As a rule of thumb, easy problems will have good accuracy for $N = 10D$, whereas difficult problems will require significantly greater $N$.
Training is therefore best done iteratively. We first take a sample of size $10D$ and validate the emulator. If the model accuracy is poor the covariance function is misspecified, in which case we must use a different covariance function, or an unduly large amount of training data. Due to the slow convergence of the MSE in this case (i.e. the case where a sample size of $10D$ is too small), we will need to resample the function in a smaller region of parameter space. If the model accuracy is good, we augment our data. To do this we require some figure of merit for choosing new design points. If we wish to emulate the function faithfully across the region, we might resample at points of maximum variance. If we wish to maximize the function, as we do here, we may use the *expected improvement*. In this case the procedure is known as *efficient global optimization* [@jones_efficient_1998].
Improving the emulator {#sec:improving_the_emulator}
----------------------
We follow the presentation of [@schonlau_global_1996], which we reproduce here in our own notation, for clarity. In the mathematical literature, optimization problems are couched in terms of minimization rather than maximization. We adopt this convention here, understanding of course that we may maximize a function by minimizing its negative.
Suppose that we are performing Gaussian-process regression (equations \[eq:regression\_model\], \[eq:gpe\_assumption\]) using training data $(\bm{a}_i, y(\bm{a}_i))_{i = 1}^{N}$. The minimum of our response sample, $y_\mathrm{min} = \min(y(\bm{a}_i))_{i = 1}^{N}$. We would like to know where to sample in order to improve the accuracy of this minimum. To this end we define the *improvement in the minimum*, $$\begin{aligned}
I_Y(Y) := \max(y_\mathrm{min} - Y, 0).\end{aligned}$$ This is a random variable, the PDF of which is $$\begin{aligned}
I_Y(y) &= \max(y_\mathrm{min} - y, 0)\\ &=
\begin{cases}
y_\mathrm{min} - y &\text{if $y < y_\mathrm{min}$,}\\ 0 &\text{otherwise}.
\end{cases}\end{aligned}$$ By definition the expected improvement is $$\begin{aligned}
{\operatorname{E}}(I_Y(Y)) := \int_{\bm{R}} I_Y(y) f_Y(y) {\operatorname{d}}y,\end{aligned}$$ where $f_Y$ is the PDF of $Y$. In the case of GPE we know that $Y \sim N(\hat{y}, \hat{\sigma}^2)$, i.e. $f_Y$ is the normal (i.e. Gaussian) PDF $\phi(y; \hat{y}, \hat{\sigma}^2)$. If $\hat{\sigma}^2 = 0$ then the value $y$ is known with certainty and we cannot expect any improvement, hence ${\operatorname{E}}(I_Y(Y)) = 0$. If $\hat{\sigma}^2 > 0$ then we may make the change of variables from $y$ to $u' = (y - \hat{y}) / \hat{\sigma}$ to find that the expected improvement is $$\begin{aligned}
{\operatorname{E}}(I_Y(Y)) &= \hat{\sigma} \int_{-\infty}^{u}(u - u') \phi(u'; 0, 1) {\operatorname{d}}u'\end{aligned}$$ where $u := (y_\mathrm{min} - \hat{y}) / \hat{\sigma}$. Thus, $$\begin{aligned}
{\operatorname{E}}(I_Y(Y)) =
\begin{cases}
\hat{\sigma} (u \Phi(u; 0, 1) + \phi(u; 0, 1))
\label{EI_expression} &\text{ if $0 < \hat{\sigma}$,}\\
0 &\text{ otherwise}
\end{cases}\end{aligned}$$ where $\Phi(u; 0, 1)$ is the normal cumulative distribution function. We augment our training data, with the pair $(\bm{a}_{N + 1}, f(\bm{a}_{N + 1}))$ where $\bm{a}_{N + 1} = {\operatorname{argmax}}({\operatorname{E}}(I_Y(Y)))$, and then iterate this procedure until ${\operatorname{E}}(I_Y(Y))$ is smaller than some threshold, $\epsilon$. Efficient global optimization (EGO) is implemented by Algorithm \[algorithm\].
objective function, $f$, sample of objective function, $D := \{(\bm{x}_i, y(\bm{x}_i))\}_{i = 1}^N$, and stopping threshold, $\epsilon$ global minimum of objective function $E_\mathrm{max} \leftarrow \max({\operatorname{E}}(I))$ $D \leftarrow D \cup ({\operatorname{argmax}}({\operatorname{E}}(I)), f({\operatorname{argmax}}({\operatorname{E}}(I)))$ $E_\mathrm{max} \leftarrow \max({\operatorname{E}}(I))$ $\bm{x}_i$ such that $y(\bm{x}_i) = \min (y(\bm{x}_i))_{i = 1}^N$.
The expected improvement for $0 < \hat{\sigma}^2$ is the sum of two terms in $u$. The first term dominates if $u$ is large, while the second term dominates if $u$ is small. For given $\hat{y}$ it is the case that $u$ is large if $\hat{\sigma}$ is small (which will be the case close to design points, including the current minimum) and $u$ is small if $\hat{\sigma}$ is large (which will be the case away from design points, including the current minimum). The expected improvement is therefore a tradeoff between probable small improvements (near to the current minimum) and improbable large improvements (remote from the current minimum), or between local and global search. The fact that the expected improvement is a trade off between local and global search makes multistart optimization a sensible choice if we initialize it at each design point. We may use gradient-based methods as the gradient of the expected improvement has closed form. The efficient global optimization algorithm reduces the problem of the prohibitively-expensive optimization of $y$ to the cheap optimization of the expected improvement. There is a convergence theorem [@vazquez_convergence_2010] that guarantees that the expected improvement produces a sequence of points that is dense in the parameter space under mild assumptions about the covariance function, so that the result is guaranteed to be a global minimum in the infinite-sample limit. However, we do not know of any theorems concerning the rate of convergence.
We illustrate EGO by reproducing an example given by @jones_efficient_1998, namely the minimization of the *Branin function*, a real-valued function of two variables used as a test for optimization. For the sake of completeness, we also produce figures equivalent to theirs. The Branin function, $$ y(a_1, a_2) = \alpha (a_2 - \beta^2 + \gamma a_1 - \delta)^2 + \zeta(1 - \eta) \cos a_1 + \eta,
\label{eq:branin}$$ where $\alpha = 1$, $\beta = 5.1 / (4 \pi^2)$, $\gamma = 5 / \pi$, $\delta =
6$, $\zeta = 10$, $\eta = 1 / (8 \pi)$. It has three global maxima, at $(a_1,
a_2) = (-\pi, 12.275)$, $(\pi, 2.275)$, and $(9.425, 2.475)$ where the function takes the value $0.398$. It is evaluated on the domain $a_1 \in [-5,
10]$, $a_2 \in [0, 15]$. We treat the function as a realization of a random process, $\{Y_{\bm{a}}\}_{\bm{a} \in \bm{A}}$, where the parameter space, $\bm{A} = [-5, 10] \times [0, 15]$. We create a LHS design for the parameter space, namely the set $\{\bm{a}_i\}_{i = 1}^N$, of size $N = 10D = 20$ and evaluate the function, $y$, at these points, giving the data $\{\bm{a}_i,
y(\bm{a}_i)\}_{i = 1}^N$. The function and the design are plotted in Figure \[fig:branin\_function\]. We assume a squared-exponential covariance function, $k_\mathrm{SE}(\bm{a}, \bm{a}') = \sigma_{\mathrm{SE}}^2 \exp (-
(\bm{a} - \bm{a}')^\mathrm{t} \bm{M} (\bm{a} - \bm{a}') / 2)$ where $\bm{M} =
{\operatorname{diag}}(m_1, m_2)$ (see equation \[eq:cov\_SE\]), and then optimize its hyperparameters, $\sigma_\mathrm{SE}^2$, $m_1$, and $m_2$, using the maximum likelihood method (equation \[eq:hyperparam\_max\_likelihood\]), finding that $\sigma^2 = 15\thinspace500$, $m_1 = 0.0689$, and $m_2 = 0.00520$, or equivalently that $l_1 = 3.81$, $l_2 = 13.9$. For the sake of illustration, we plot the likelihood of the hyperparameters in Figure \[fig:branin\_hyperparameter\_likelihood\] as well as the mean and variance of the Gaussian-process estimator for the entire parameter space in Figure \[fig:branin\_function\_gpe\] (see equations \[eq:gpe\_mean\], \[eq:gpe\_var\]). We also compute the standardized mean square predicted error, finding that $R = 2.01$ (equation \[eq:cv\_score\]), and the maximum standardized predicted error, finding that the most extreme value of $e_{i}$ is 2.63computational (see equation \[eq:predicted\_error\]). Diagnostic plots are shown in Figure \[fig:branin\_function\_diagnostics\]. We see that the standardized predicted errors are distributed normally and exhibit no trend across the parameter space. The total sensitivity, $\tau h^2 = 16.7$, is moderate, explaining this success. Satisfied that our model is accurate, we then iteratively augment the data set using the maximum expected improvement and a stopping criterion of $\epsilon = 0.001$. For illustration we plot the expected improvement for the first iteration (Figure \[fig:branin\_function\_expected\_improvement\]) and see that it has three maxima close to the three minima of the Branin function. The algorithm requires a total of eight iterations to find the maximum to an accuracy of $1.3$ per cent . We plot the augmented design in Figure \[fig:branin\_function\_augmented\].
![The Branin function (equation \[eq:branin\]) and the Latin square design (marked with filled circles) used in its emulation. Following [@jones_efficient_1998], we use it to illustrate the methods of Gaussian process emulation [@ohagan_curve_1978] and efficient global optimization [@jones_efficient_1998]. It has three global minima (marked with crosses).[]{data-label="fig:branin_function"}](branin.pdf){width="1.0\linewidth"}
![The likelihood (equation \[eq:hyperparam\_max\_likelihood\]) of the hyperparameters of the squared-exponential covariance function (equation \[eq:cov\_SE\]), used in the emulation of the Branin function (equation \[eq:branin\]). In each panel the marginal likelihood is shown (i.e. the likelihood has been integrated over the parameters not shown), scaled to the unit interval. The maximum likelihood is found at $(\sigma_\mathrm{SE}^2, m_, m_2, m) = (15\thinspace500, 0.0689, 0.00520)$, i.e. for length scales, $l_1 = 3.81$ and $l_2 = 13.9$.[]{data-label="fig:branin_hyperparameter_likelihood"}](branin_likelihood_hyperparameters.pdf){width="1.0\linewidth"}
![The mean (left) and variance (right) of GPE estimate of the Branin function, computed using the squared-exponential covariance function and the design shown (filled circles). The variance is high in regions of parameter space that have been poorly sampled, or near the boundary of parameter space, where the estimator is constrained by less data than elsewhere.[]{data-label="fig:branin_function_gpe"}](branin_gpe.pdf){width="1.0\linewidth"}
![Diagnostic plots for the emulation of the Branin function. Top-left: distribution of likelihood values. Top-right: quantile-quantile plot showing the ordered standardized predicted errors from a LOOCV analysis (equation \[eq:predicted\_error\]) against the equivalent quantiles of the normal distribution. Bottom-left: true values against predicted values from a LOOCV analysis. Bottom-right: residuals from LOOCV analysis.[]{data-label="fig:branin_function_diagnostics"}](branin_diagnostic_plots.pdf){width="1.0\linewidth"}
![The expected improvement in the minimum (left) of our sample of the Branin function, computed using the design shown (filled circles). There are three maxima, each close to a minimum of the Branin function. The Branin function and the augmented design (marked with filled triangles) determined by the EGO algorithm (right). The new data cluster about the function’s three minima. Once the initial design has been computed only 8 additional design points are required to find a global minimum with an accuracy of $1.3$ per cent.[]{data-label="fig:branin_function_augmented"}](branin_ego.pdf){width="1.0\linewidth"}
\[fig:branin\_function\_expected\_improvement\]
As with any global optimization method, it is sensible to polish the result, which may lack accuracy. We may do so by resampling the function in the neighbourhood of this result, and again performing regression. The length scales of the GPE fit provides a guide to the size of this neighbourhood. If $\bm{x}_0 = (x_{0, 1}, x_{0, 2}, x_{0, 3})$ is the result of the EGO algorithm then we resample the function in the region $\bm{A}' = [x_{0, 1} - \delta l_1, x_{0, 1} + \delta l_1] \times [x_{0, 2} - \delta l_2, x_{0, 2} + \delta l_2] \times [x_{0, 3} - \delta l_3, x_{0, 3} + \delta l_3]$, where $(l_0, l_1, l_2)$ is the vector of length scales found in the final iteration, and $\delta$ is some positive real number less than one. We may again use GPE to perform this regression. Our polished maximum is then the maximum of the GPE estimator. We can find this maximum using gradient-based optimization, or, as the GPE estimator is cheap, by brute force i.e. by searching over a fine lattice of test points covering the whole region $\bm{A}'$. (This brute force method returns more the maximum, of course. It maps out the function over the entirety of $\bm{A}'$. In general the cheapness of the GPE estimator will allow us to do just this. When working in very high-dimensional parameter spaces the inversion the covariance matrix, $\bm{K}$, may not be so cheap, and we may wish to map out the region using, for example, Markov chain Monte Carlo methods.) There are two additional benefits to this polishing step. First, we may use a high termination threshold, $\epsilon$, which reduces the number of iterations required by the EGO algorithm. Second, the Hessian of the GPE estimate is available in closed form, and provides an estimate of the Hessian of the function. If the function in question is a likelihood, this allows us to compute an estimate of the Fisher information matrix. The derivative of a Gaussian process is itself a Gaussian process [@adler_geometry_2010]: $$\begin{aligned}
\dfrac{\partial Y}{\partial \bm{a}} \sim {\operatorname{GP}}\left( \dfrac{\partial \hat{r}}{\partial \bm{a}}, \dfrac{\partial^2 k(\bm{a}, \bm{a}')}{\partial \bm{a} \partial \bm{a}^\mathrm{\prime t}} \right).
\label{eq:GPE_deriv}\end{aligned}$$ If $H = \partial^2 y / \partial \bm{a} \partial \bm{a}^\mathrm{t}$ is the Hessian of the function $y$, then an estimate for the Hessian is $$\begin{aligned}
\hat{H} = \dfrac{\partial^2 \hat{r}}{\partial \bm{a} \partial \bm{a}^\mathrm{t}}.
\label{eq:gpecompute_hessian}\end{aligned}$$ We compute the Hessian for the case of the squared-exponential covariance function in Appendix \[sec:se\_derives\].
Computational expense
---------------------
The computational complexity of equations \[eq:gpe\_mean\] and \[eq:gpe\_var\] is dominated by the inversion of the matrix $\bm{K}$, which is of order $O(N^3)$ or better. This inversion may be done indirectly by solving the systems $\bm{K} \bm{\alpha} = \bm{y}$ and $\bm{K} \bm{\beta} = \bm{k}(\bm{a})$ for $\bm{\alpha}$ and $\bm{\beta}$ respectively. Moreover, it must be performed only once, regardless of the number of test evaluations required. Once the inversion has been performed, evaluation of the mean involves one matrix-vector multiplication, with complexity $O(N^2)$, followed by one vector-vector multiplication for each evaluation, with complexity $O(N)$. Each evaluation of the variance involves one matrix-vector multiplication and one vector-vector multiplication. Thus the computational complexity is $O(N^3)$. The covariance matrix $\bm{K}$ must also be inverted for every step in the optimization of the hyperparameters, which must be done at each iteration of the EGO algorithm. Furthermore, it must be computed explicitly for the validation step. Nonetheless, the total computational expense of these inversions is negligible compared with the expense of any interesting astrophysical simulation. The expense of the method is largely in the computation of the training data, and its augmentation required when using the EGO method. The initial sampling is trivially parallel, and linear in the number of parameters, but the EGO method is necessarily sequential. (A batch-sequential extension of EGO is available, and makes it possible to perform up to 10 function evaluations at each iteration.) In general we cannot estimate in advance the number of iterations required for EGO without knowing the rate of convergence of the EGO algorithm, so we do not explore this further here.
A TOY APPLICATION {#sec:plummer}
=================
The anisotropic Plummer sphere
------------------------------
![The log-marginalized likelihood, $L$, for the anistropic Plummer model with parameters $\log M = 0$, $\log b = 0$, $\log r_\mathrm{a} = \log 2 = 0.301$ computed using data for 1000 stars generated by the same model (equation \[eq:OM\_line\_profile\]). In each panel the likelihood has been marginalized over the unshown parameters, scaled to the unit interval, and its natural logarithm plotted.[]{data-label="fig:plummer_log_likelihood"}](plummer_likelihood.pdf){width="1.0\linewidth"}
![The log-marginalized likelihood, $L$, for the anistropic Plummer model with parameters $\log M = 0$, $\log b = 0$, $\log r_\mathrm{a} = \log 2 = 0.301$ computed using data for 1000 stars generated by the same model (equation \[eq:OM\_line\_profile\]). In each panel the likelihood has been marginalized over the unshown parameters, scaled to the unit interval, and its natural logarithm plotted.[]{data-label="fig:plummer_log_likelihood"}](plummer_log_likelihood.pdf){width="1.0\linewidth"}
We illustrate the use of GPE for stellar dynamical modelling using the toy model of an anisotropic Plummer sphere of the Osipkov-Merritt type. The relative potential of the Plummer sphere, $$\begin{aligned}
\Psi(r) = \dfrac{G M}{\sqrt{r^2 + b^2}},\end{aligned}$$ and its density $$\begin{aligned}
\rho(r) = \dfrac{3 M}{4 \pi b^3}\left( 1 + \frac{r^2}{b^2} \right)^{- 5 / 2}
\label{eq:density}\end{aligned}$$ where $M$ is the galactic mass, $r$ the radius, and $b$ the galactic scale length [@plummer_problem_1911].
Following [@osipkov_spherical_1979] and [@merritt_spherical_1985], we define the variable $Q = \mathcal{E} - L^2 / 2 r_\mathrm{a}^2$ where $\mathcal{E}$ is the relative energy, $L$ is the magnitude of the angular momentum, and $r_\mathrm{a}$ is the anisotropy radius. By use of Eddington’s inversion formula we then find that the phase-space DF may be expressed as a function of $Q$: $$\begin{aligned}
f_Q(Q) = \dfrac{3 M b^2}{\pi^3 \sqrt{2} r_\mathrm{a}^2} \left( \dfrac{16(r_\mathrm{a}^2 - b^2)}{7}
Q^{7 / 2} + (G M)^2 Q^{3 / 2}\right).
\label{eq:OM_pdf}\end{aligned}$$
The PDF of the observables is given by the integral of $f_Q(Q)$ with respect to the line-of-sight position and proper-motion velocities. If we define the parameter vector $\bm{a} = (M, b, r_\mathrm{a})$ and work in cylindrical coordinates with the $z$-axis parallel to the line of sight, this PDF $$\begin{aligned}
f_{(R_{\mathrm{p}}, V_z)}(r_{\mathrm{p}}, v_z; \bm{a}) = 2 \pi \int_{\bm{R}}
\int_{\bm{R}} \int_{\bm{R}} f_Q(Q) {\operatorname{d}}v_{r_\mathrm{p}} {\operatorname{d}}v_{\phi} {\operatorname{d}}z.
\label{eq:OM_line_profile}\end{aligned}$$ We seek to maximize the likelihood $L(\bm{a}; r_\mathrm{p}, v_z) := f_{(R_{\mathrm{p}}, V_z)}(r_\mathrm{p}, v_z; \bm{a})$. The inner double integral may be computed analytically using the method given by [@carollo_velocity_1995]: $$\begin{aligned}
\int_{\bm{R}}
\int_{\bm{R}} f_{Q}(Q) {\operatorname{d}}v_{r_{\mathrm{p}}} {\operatorname{d}}v_\phi =
\begin{cases}
2 \pi g(r, r_{\mathrm{p}}) F(Q_{\mathrm{max}}) &\text{ if $0 < Q_{\mathrm{max}}$,}\\
0 &\text{ otherwise}
\end{cases}\end{aligned}$$ where $$\begin{aligned}
g(r_{r, \mathrm{p}}) &:= \dfrac{a^2}{\sqrt{(r_\mathrm{a}^2 + r^2)(r_\mathrm{a}^2 + r^2 - r_\mathrm{p}^2)}},\\
F(Q) &:= \dfrac{6 b^2}{\pi^3 \sqrt{2} r_\mathrm{a}^2 (G M)^5} \left(\dfrac{16(r_\mathrm{a}^2 - b^2)}{63} Q^{9 / 2} + \dfrac{(GM)^2}{5} Q^{5 /
2}\right),\end{aligned}$$ and $$\begin{aligned}
Q_{\mathrm{max}}(r_\mathrm{p}, z, v_{z}) = \Psi(r) - \left(\dfrac{r_\mathrm{a}^2 + r^2}{r_\mathrm{a}^2 + r^2 - r_\mathrm{p}^2}\right)
\dfrac{v_{z}^2}{2}.\end{aligned}$$ However, the outer integral in equation \[eq:OM\_line\_profile\] must be computed numerically.
Optimization of the likelihood
------------------------------
![The likelihood (equation \[eq:hyperparam\_max\_likelihood\]) of the hyperparameters of the squared-exponential covariance function (equation \[eq:cov\_SE\]), used in the emulation of the Plummer-model likelihood (equation \[eq:OM\_line\_profile\]). In each panel the marginal likelihood is shown (i.e. the likelihood has been integrated over the parameters not shown), scaled to the unit interval. The maximum likelihood is found at $(\sigma_\mathrm{SE}^2, m_{\log M}, m_{\log b}, m_{\log{r_\mathrm{a}}}) = (535\thinspace000, 23.5, 3.30, 3.43)$, i.e. for length scales, $l_{\log M} = 0.206$, $l_{\log b} = 0.550$, and $l_{\log{r_\mathrm{a}}} = 0.540$).[]{data-label="fig:hyperparameter_likelihood"}](plummer_pseudolikelihood_hyperparameters.pdf){width="1.0\linewidth"}
![Diagnostic plots for the emulation of the transformed Plummer model likelihood. Top-left: distribution of likelihood values. Top-right: quantile-quantile plot showing the ordered standardized predicted errors from a LOOCV analysis (equation \[eq:predicted\_error\]) against the equivalent quantiles of the normal distribution. Bottom-left: true values against predicted values from a LOOCV analysis. Bottom-right: residuals from LOOCV analysis.[]{data-label="fig:plummer_function_diagnostics"}](plummer_diagnostic_plots.pdf){width="1.0\linewidth"}
We work in mass units of $10^9 \text{M}_{\sun}$, and distance units of kpc (meaning that the gravitational constant $G = 4.302 \times 10^{3} \text{kpc} \text{M}_{\sun}^{- 1} \text{km}^{2} \text{s}^{- 2}$). We use synthetic data generated using the same model and parameters $M = 1$, $b = 1$, and $r_\mathrm{a} = 2$. The data consist of positions and line-of-sight velocities for 1000 stars, each with zero error. In the case of the anistropic Plummer model the likelihood is cheaply computed, and is shown in Figure \[fig:plummer\_likelihood\]. Suppose, however, that the likelihood were not cheaply computed. In this case we would proceed as follows.
First we choose the region of parameter space on which we wish to emulate. By the virial theorem we know that $3 \langle v_z^2 \rangle = G M_\mathrm{virial} / r_\textrm{g}$ where $\langle v_z^2 \rangle$ is the line-of-sight velocity dispersion, $M_\mathrm{virial}$ is the virial mass, and $r_\mathrm{g}$ is the gravitational radius [@binney_galactic_2008]. We may approximate the gravitational radius by $r_{1 / 2} / 0.45$ [@binney_galactic_2008], where $r_{1 / 2}$ is the half-light radius, and note that for the Plummer sphere, $r_{1 / 2} = b / \sqrt{2^{2 / 3} - 1}$. We might guess that the true value of $M$ is within a factor of three either side of $M_\mathrm{virial}$. Similarly, we might guess that the true value of $b$ is a factor of three either side of its estimate, and that the true value of $r_\mathrm{a}$ is within an order of magnitude either side of its estimate.
However, in the case that the observed data have no error, the feasible region is bounded below by the curve $Q_\mathrm{max} = 0$ (the minimum value $Q$ can take). For a given projected radius, the maximum line-of-sight velocity is therefore set by the condition $$(v_z)^2 = \frac{2 \Psi(r) (r_\mathrm{a}^2 + r^2 - r_\mathrm{p}^2)}{r_\mathrm{a}^2 + r^2}
\label{eq:terminal_velocity}$$ where $r_\mathrm{p} \leq r$. We must maximize this expression. The maximum value, $(v_z)_\mathrm{max}$, occurs at a radius determined by the equation $$\begin{aligned}
\frac{{\operatorname{d}}v_z}{{\operatorname{d}}r}\bigg|_{r = r_\mathrm{max}} = 0
\label{eq:los_vel_max}\end{aligned}$$ which, if it exists, is unique, or (if this equation has no solution on account of the constraint $r \geq r_\mathrm{p}$) at a radius $r = r_\mathrm{p}$.
In the isotropic limit, $r_\mathrm{a} = \infty$, we have $v_z = (v_z)_\mathrm{max}$ at $r = r_\mathrm{p}$, and therefore $$\frac{M^2}{(v_z^2 r_\mathrm{p} / 2G)^2} - \frac{b^2}{r_\mathrm{p}^2} = 1,$$ a hyperbola in $M$ and $b$. Each pair $(r_\mathrm{p}, v_z)$ defines such a hyperbola. In the anisotropic case, equation \[eq:los\_vel\_max\] gives $$r_\mathrm{max}^2 = \frac{(3 r_\mathrm{p}^2 - 2 a^2) - r_\mathrm{p} \sqrt{9 r_\mathrm{p}^2 - 8 r_\mathrm{a}^2 + 8 b^2}}{2}
\label{eq:radius_max}$$ if the discriminant and numerator are real and nonnegative, i.e. if $$\begin{aligned}
r_\mathrm{p} &\ge \frac{r_\mathrm{a}^2}{\sqrt{r_\mathrm{a}^2 + 2b^2}}, \text{ and}\\
r_\mathrm{a} &\ge b.\end{aligned}$$ Otherwise, $r_\mathrm{max} = r_\mathrm{p}$. In the point-mass limit, $b =0$, and upon substituting equation \[eq:terminal\_velocity\] into \[eq:los\_vel\_max\] we find an equation in $M$ and $r_\mathrm{a}$. Again, each pair $(r_\mathrm{p}, v_z)$ defines such an equation. A given parameter vector is forbidden if the observed line-of-sight velocity of any star is greater than this maximum allowed velocity.
For our data we find that $\langle v_z^2 \rangle = 504$ km$^2$ s$^{- 2}$, and $r_{1 / 2} = 0.944$ kpc. Thus $b = 0.723$ kpc and $M_\mathrm{virial} = 0.737
\times 10^9$ M$_{\sun}$. The total mass, $M$, is bounded below by the maximum value of $v_z^2 r_\mathrm{p} / 2 G$, namely $0.461 \times 10^9$ M$_{\sun}$. We therefore choose to emulate on the region of parameter space $\bm{A} = [0.461,
2.21] \times [0.241, 2.17] \times [0.241, 7.23]$. We make a logarithmic transformation of the parameter space (according to prescription given in section \[sec:conditioning\]), mapping a parameter vector $\bm{a} = (M, b,
r_\mathrm{a})$ to $\bm{x} = (\log M, \log b, \log r_\mathrm{a})$. The transformed parameter space is $\bm{X} = [-0.336, 0.345] \times [-0.618,
0.337] \times [-0.618, 0.860]$. We also transform the likelihood (again according to prescription given in section \[sec:conditioning\]) from $L(\bm{a})$ to $\ln(L_{\bm{X}}(\bm{x}) + \epsilon)$ where $\epsilon = \min
(L_{\bm{X}}(\bm{x}_i))_{i = 1}^N$. We make $10D = 30$ samples from this transformed parameter space (according the prescription given in section \[sec:training\]) giving the training data $(\bm{x}_i,
\ln(L_{\bm{X}}(\bm{x}_i) + \epsilon))_{i = 1}^N$. We then optimize the model hyperparameter vector, $\bm{\theta} = (\sigma_\mathrm{SE}^2, m_{\log M},
m_{\log b}, m_{\log{r_\mathrm{a}}})$, using the maximum LOO-likelihood method (described in section \[sec:validating\_the\_emulator\]), finding that $\bm{\theta}_* = (535\thinspace000, 23.5, 3.30, 3.43)$, i.e. that the length scales are $l_{\log M} = 0.206$, $l_{\log b} = 0.550$, and $l_{\log{r_\mathrm{a}}} = 0.540$. We find that the LOOCV score is $R = 0.801$, and that the extreme value of the LOOCV residuals is 2.33. Diagnostic plots (Figure \[fig:plummer\_function\_diagnostics\]) show that the standardized square errors are distributed normally and show no trend across the parameter space. The results of the validation are acceptable, meaning that we may proceed to maximize the transformed likelihood using EGO. Using a stopping threshold of $\epsilon = 0.001$, the EGO algorithm requires $33$ iterations to find the maximum at $\log M = 0.0356$, $\log b = -0.0288$, and $\log
r_\mathrm{a} = 0.328$. At the last iteration the maximum LOO-likelihood estimate of the hyperparameter vector is $\bm{\theta} = (203\thinspace000,
17.0, 30.8, 8.10)$, i.e. that the length scales are $l_{\log M} = 0.242$, $l_{\log b} = 0.180$, and $l_{\log{r_\mathrm{a}}} = 0.351$.
We then polish this result by resampling the likelihood in its neighbourhood, and again performing GPE. We choose the region that is within one quarter of a length scale in each parameter, namely $\bm{X}' = [-0.0249, 0.0961] \times [-0.0738, 0.0162] \times [0.240, 0.416]$. We again transform our sample of the likelihood, finding that the most-appropriate transformation is to $\ln L(\bm{x})$, where no offset is required as the likelihood is everywhere defined in this new region of parameter space. The maximum LOO-likelihood estimate of the parameter vector is $\bm{\theta}_* = (436, 24.1, 28.3, 1.42)$, i.e. the estimate for the length scales are $l_{\log M} = 0.204$, $l_{\log b} = 0.188$, and $l_{\log{r_\mathrm{a}}} = 0.839$. We find the maximum at $\log M = 0.0327$, $\log b = -0.0213$, and $\log r_\mathrm{a} = 0.305$. We note that for this three-dimensional model we have recovered the MLE with fewer than $100$ evaluations of the likelihood. The first and last sets of 30 evaluations may be each be made in parallel, effectively reducing this number to approximately $40$. Batch-sequential EGO (section \[sec:improving\_the\_emulator\]), would reduce the effective number of runs still further.
The total sensitivity is the initial step of emulation is $\tau = 21.4$, indicating that this problem is hard (section \[sec:training\]). We can see why this is the case by inspecting the plot of the likelihood (Figure \[fig:plummer\_likelihood\]). We note that the likelihood is very sharply peaked. Another way of putting this is to say that it has multiple length scales (the function is highly sensitive to changes in the parameter vector around its maximum, but insensitive to such changes away from its maximum). The squared-exponential covariance function, which assumes a single set of length scales is thus grossly misspecified. Indeed the average separation of the design points is smaller than the peak. The sharpness of this peak is due to several factors: (1) that our data are drawn from the same model we are fitting, (2) that the dimension of our parameter space is small, and (3) that there is no error associated with our synthetic observations. The dynamical model is well-specified and its parameters tightly constrained by the data. The problem of multiple lengths persists even in the transformed data, to which we see an approximation in Figure \[fig:plummer\_log\_likelihood\]. In this case there is a sharp cliff on the boundary of the permitted and forbidden regions of parameter space. Such forbidden regions exist only for data with zero errors. We thus expect the task of fitting this perfectly specified low-dimensional toy model to perfect data to be the maximally difficult case for emulation. We expect it to be considerably harder to than the task of fitting more-sophisticated models to imperfect data, the likelihoods of which will be less sharply peaked, and for which forbidden regions of parameter space do not exist.
Confidence region
-----------------
![The one- to five-sigma confidence regions for the maximum-likelihood estimate of the Plummer-model parameters, $M$, $b$, and $r_\mathrm{a}$, computed using the Fisher information matrix (equation \[eq:likelihood\_confidence\]).[]{data-label="fig:plummer_param_error"}](plummer_gpe_confidence.pdf){width="1.0\linewidth"}
![The maximum-likelihood estimates (solid lines) of the Plummer-model density (left) and anisotropy parameter (right), together with their one-sigma confidence intervals (shaded regions).[]{data-label="fig:plummer_density_error"}](plummer_density_anisotropy.pdf){width="1.0\linewidth"}
The Hessian of the log-likelihood is available to us as a consequence of the polishing step (equation \[eq:gpecompute\_hessian\]). Hence, we may compute an estimate of the Fisher information matrix (equation \[eq:fisher\_information\]) without further evaluation of the dynamical model. However, our estimator $\hat{r}$ is for the log-likelihood, $\ln L(\bm{x})$, expressed as a function of the transformed parameters, $\bm{x}$. Thus, the first derivative is $$\begin{aligned}
\dfrac{\partial \ln L}{\partial a_j}
&= \dfrac{\partial \hat{y}}{\partial a_j}\\
&= \dfrac{\partial \hat{y}}{\partial x_j} \dfrac{\partial x_j}{\partial a_j},\end{aligned}$$ and the second derivative is $$\begin{aligned}
\dfrac{\partial^2 \ln L}{\partial a_i \partial a_j}
&= \dfrac{\partial^2 \hat{y}}{\partial a_i \partial a_j}\\
&= \dfrac{\partial^2 \hat{y}}{\partial x_i \partial x_j} \dfrac{\partial x_i}{\partial a_i} \dfrac{\partial x_j}{\partial a_j} + \dfrac{\partial \hat{y}}{\partial x_j} \dfrac{\partial^2 x_j}{\partial a_i \partial a_j},\end{aligned}$$ where the second term vanishes at the maximum. Given that $x_i = \log a_i$ we have that $$\begin{aligned}
\dfrac{\partial x_i}{\partial a_i}
&= \dfrac{1}{a_i \ln 10}, \text{ and}\\
\dfrac{\partial^2 x_i}{\partial a_i \partial a_j}
&= - \dfrac{1}{a_i^2 \ln 10} \delta_{ij}.\end{aligned}$$ In Figure \[fig:plummer\_param\_error\] we plot the confidence regions for the maximum-likelihood estimate of the parameters. The galactic mass, $M$, and scale length, $b$, are well constrained but the anistropy parameter, $r_\mathrm{a}$, is less so. This is as we would expect. For a self-consistent model of this kind the mass and extent of a galaxy are functions of one another through Poisson’s equation. All of the observational data therefore contain information about the $M$ and $b$. In the anisotropic case, however, there is an additional length scale, $r_\mathrm{a}$, which we must determine using data at radii greater than this value. Stars at smaller radii do not constrain the length scale, meaning that only a subset of our data contain information about it.
Given the distribution of the MLE for the parameters we may also compute the distribution of the MLE for the density and for Binney’s anisotropy parameter using equation \[eq:delta\_method\]. The density is given by equation \[eq:density\] and hence the MLE for the density, $$\hat{P} \sim N(\rho(\hat{\bm{a}}), \sigma_{\rho}^2),
\label{eq:density_distribution}$$ where $$\sigma_{\rho}^2 = \left( \dfrac{\upartial \rho(\hat{\bm{a}})}{\upartial
\bm{a}}\right)^{\mathrm{t}} \bm{I}_A^{- 1}(\hat{\bm{a}}) \dfrac{\upartial
\rho(\hat{\bm{a}})}{\upartial \bm{a}}.
\label{eq:density_distribution}$$ For an Ossipkov-Merritt model, Binney’s anisotropy parameter [@binney_galactic_2008], $$\begin{aligned}
\beta(r) = \dfrac{1}{1 + r_\mathrm{a}^2 / r^2}.\end{aligned}$$ Hence, the MLE for Binney’s anisotropy parameter, $$\begin{aligned}
\hat{B} &\sim N(\beta(\hat{\bm{a}}), \sigma_{\beta}^2),
\label{eq:beta_distribution}\end{aligned}$$ where $$\sigma_{\beta}^2 = \left( \dfrac{\upartial \beta(\hat{\bm{a}})}{\upartial \bm{a}}
\right)^{\mathrm{t}} \bm{I}_A^{- 1}(\hat{\bm{a}}) \dfrac{\upartial
\beta(\hat{\bm{a}})}{\upartial \bm{a}}.
\label{eq:beta_error}$$ We plot the distributions of these quantities in Figure \[fig:plummer\_density\_error\]. These are the pricipal results of our work.
CONCLUSION {#sec:conclusion}
==========
We have presented a novel statistical algorithm for the efficient dynamical modelling of a stellar system. Throughout, our interest has been in what an observational data set can tell us about the stellar system from which it is drawn. In particular, for a dSph, we would like to know which dark-matter morphologies a kinematic data set rules out, and which dark-matter morphologies best account for the data.
We have adopted the maximum-likelihood approach, which allows us to draw robust confidence intervals within the parameter space of our dynamical model (section 2). The method of maximum likelihood requires us to know the maximum of the likelihood function, and the second derivative of this likelihood function at its maximum, in order that we may compute the Fisher information matrix. Typically, this likelihood function has no closed-form expression and is expensive to evaluate. We have therefore used GPE ([@ohagan_curve_1978] and [@sacks_design_1989]) and efficient global optimization [@jones_efficient_1998], which allow us to optimize the likelihood at significantly reduced computational expense, and to compute good approximations to the second derivatives (section \[sec:gpe\]).
The methods of GPE and EGO are well-established, but there are particular issues in applying them to the situation we have described. The likelihood function is difficult to emulate as it may be sharply peaked. This can cause the GPE to underperform. The solution to this problem is to transform the data so that they better fit the assumptions of the method. This amounts to a nonlinear scaling and reparameterization of the dynamical model followed by validation of the results. Each stage of the analysis may be automated and is implemented in a Python module called PyMimic, which we make publicly available (<https://github.com/AmeryGration/pymimic>). We have given an example of the analysis for the case of a toy model, namely the single-component anisotropic Plummer sphere, which is (counterintuitively) the maximally difficult case. We note this example requires fewer than 100 runs of the dynamical model, and that because the method is trivially parallelizable, the effective number of runs is approximately 40 (if we were to use a naive lattice-based search of the parameter space we might expect to need more than 1000 runs). The method is readily applicable to more-sophisiticated models, and we anticipate that it will allow us to fit a broader class of models than has been computationally tractable. In future work we plan to use it to fit two-component general-profile equilibrium models to observations of the classical dSphs, as well as to fit $N$-body models to observations of tidally-disturbed dSphs [@ural_low_2015].
Acknowledgements {#acknowledgements .unnumbered}
================
We would like to thank Carlos Frenk and Richard Bower for valuable and insightful discussions. Their work prompted this research. We would also like to thank Sylvy Anscombe for numerous discussions of the mathematical structure of GPE.
This work used the DiRAC Complexity system, operated by the University of Leicester IT Services, which forms part of the STFC DiRAC HPC Facility (www.dirac.ac.uk ). This equipment is funded by BIS National E-Infrastructure capital grant ST/K000373/1 and STFC DiRAC Operations grant ST/K0003259/1. DiRAC is part of the National E-Infrastructure.
Derivatives of Gaussian processes {#appendix1}
=================================
For convenience we introduce the vector $\bm{\alpha} := \bm {K}^{-1}\bm {y}$. From equation \[eq:gpe\_mean\] we find that the gradient of the mean, $$\begin{aligned}
\dfrac{\partial \hat{y}(\bm {a})}{\partial \bm {a}} = \dfrac{\partial \bm {k}(\bm {a})}{\partial \bm {a}} \bm{\alpha},
\label{eq:gradient}\end{aligned}$$ or, in component form, $$\begin{aligned}
\dfrac{\partial \hat{y}(\bm{a})}{\partial a_j} = \sum_{i = 1}^{N} \dfrac{\partial k_i(\bm{a})}{\partial a_j} \alpha_{i}.\end{aligned}$$ The Hessian of the mean is then $$\begin{aligned}
\dfrac{\partial^2 \hat{y}(\bm {a})}{\partial \bm {a} \partial \bm{a}^\mathrm{t}} = \dfrac{\partial^2 \bm {k}(\bm {a})}{\partial \bm{a} \partial \bm{a}^\mathrm{t}} \bm{\alpha},
\label{eq:hessian}\end{aligned}$$ or, in component form, $$\begin{aligned}
\dfrac{\partial^2 \hat{y}(\bm{a})}{\partial a_i \partial a_j} = \sum_{i = 1}^N \dfrac{\partial^2 k_i(\bm{a})}{\partial a_i \partial a_j} \alpha_i.\end{aligned}$$
Squared-exponential kernel {#sec:se_derives}
--------------------------
The squared-exponential kernel, $$\begin{aligned}
k_\mathrm{SE}(\bm {a}, \bm {a}') := \sigma_\mathrm{SE}^2 \exp \left( -\dfrac{1}{2} (\bm {a} - \bm {a}')^\mathrm{t} \bm {M} (\bm {a} - \bm {a}') \right).\end{aligned}$$ Hence $$\begin{aligned}
\dfrac{\partial k_\mathrm{SE}(\bm {a}, \bm {a}')}{\partial \bm {a}}
&= - k_\mathrm{SE}(\bm {a}, \bm {a}') \bm {M} (\bm {a} - \bm {a}'),\\
\dfrac{\partial k_\mathrm{SE}(\bm {a}, \bm {a}')}{\partial \bm {a}'}
&= - \dfrac{\partial k_\mathrm{SE}(\bm {a}, \bm {a}')}{\partial \bm{a}},\\
\dfrac{\partial^2 k_\mathrm{SE}(\bm {a}, \bm {a}')}{\partial \bm{a} \partial \bm{a}^\mathrm{t}}
&= k_\mathrm{SE}(\bm {a}, \bm {a}') \bm {M} \left((\bm{a} - \bm {a}') (\bm {a} - \bm {a}')^\mathrm{t} \bm {M} - \bm {I}\right),\\
\dfrac{\partial^2 k_\mathrm{SE}(\bm {a}, \bm {a}')}{\partial \bm{a}' \partial {\bm{a}'}^\mathrm{t}}
&= \dfrac{\partial^2 k_\mathrm{SE}(\bm {a}, \bm {a}')}{\partial \bm {a} \partial \bm{a}^\mathrm{t}},\end{aligned}$$ and $$\begin{aligned}
\dfrac{\partial^2 k_\mathrm{SE}(\bm {a}, \bm {a}')}{\partial \bm{a} \partial {\bm{a}'}^\mathrm{t}} &= - \dfrac{\partial^2 k_\mathrm{SE}(\bm {a}, \bm {a}')}{\partial \bm{a} \partial \bm{a}^\mathrm{t}}.\end{aligned}$$ Thus we may compute the derivatives given in equations \[eq:gradient\] and \[eq:hessian\]. \[lastpage\]
[^1]: E-mail: [email protected]
[^2]: We adopt the notational convention that a random variable $\Omega$ (always capitalized) has PDF $f_\Omega$. Given a realization, $\omega$ (always lower case), of $\Omega$, and a parameter of the model, $a \in A$, this PDF then takes the value $f_{\Omega}(\omega; a)$.
[^3]: We note that in [@bower_parameter_2010] $r$ is not taken to be the zero function, but instead is a sum of polynomials in the elements of $\bm{a}$.
[^4]: Parameter space is dimensional meaning that a naive definition of a metric on the space $\bm{R}^D$ has no meaning. However, we may impose a metric on the space of functions defined by the phase-space PDF, which then induces a pseudo-metric on the parameter space used to index these functions. A pseudo-metric is similar to a metric, but the condition of positive-definiteness is relaxed to become positive-semidefiniteness, i.e. two nonidentical elements of the space may be separated by zero distance.
[^5]: In practice, we do not need to calculate the LOOCV residuals directly. It can be shown ([@sundararajan_predictive_2001]) that the standardized predicted error, $e_{-i}(\bm{a}_i) = [\bm{K}^{-1} \bm{y}]_{ii} / \sqrt{[\bm{K}]_{ii}}$, which regrettably requires the explicit inversion of $\bm{K}$.
|
{
"pile_set_name": "ArXiv"
}
|
**SINCULAR DEGENERATE PROBLEMS AND APPLICATIONS**
**VELI B. SHAKHMUROV**
Okan University, Department of of Electronics and Communication, Akfirat, Tuzla 34959 Istanbul, Turkey, E-mail: [email protected]
**AMS: 34G10,35J25,35J70 **
**ABSTRACT**
The boundary value problems for linear and nonlinear singular degenerate differential-operator equations are studied. We prove a well-posedeness of linear problem and optimal regularity result for the nonlinear problem which occur in fluid mechanics, environmental engineering and in the atmospheric dispersion of pollutants.
**Key Words:** differential-operator equations, Semigroups of operators, Banach-valued function spaces, separability, fredholmness, interpolation of Banach spaces, Atmospheric dispersion of pollutants.
**1. Introduction, notations and background**
The maximal regularity properties for boundary value problems (BVPs) for linear differential-operator equations (DOEs) have been studied extensively by many researchers (see e.g. $\left[ 1-10\right] $ and the references therein). The main objective of the present paper is to discuss BVPs for the following nonlinear singular degenerate DOE $$-x^{2\alpha }\frac{\partial ^{2}u}{\partial x^{2}}-y^{2\beta }\frac{\partial
^{2}u}{\partial y^{2}}+a\left( x,u,u_{x},u_{y}\right) u=F\left(
x,u,u_{x},u_{y}\right)$$on the rectangular domain $G=\left( 0,a\right) \times \left( 0,b\right) .$
Several conditions for the uniform separability and the resolvent estimates for the corresponding linear problem are given in abstract $L_{p}$-spaces. Especially, we prove that the linear differential operator is positive and is a generator of an analytic semigroup. Moreover, the existence and uniqueness of maximal regular solution of the above nonlinear problem are obtained. One of the important characteristics of these DOEs are that the degeneration process are taking place at different speeds at boundary, in general. Maximal regularity properties of regular degenerated nonlinear DOEs are studied e.g. in $\left[ \text{1, 8, 10}\right] .$ Unlike to these we consider here the singular degenerate DOEs. In applications maximal regularity properties of infinite systems of singular degenerate PDE are studied.
Let $\gamma =\gamma \left( x\right) ,$ $x=\left(
x_{1},x_{2},...,x_{n}\right) $ be a positive measurable function on a domain $\Omega \subset R^{n}.$ Let $L_{p,\gamma }\left( \Omega ;E\right) $ denote the space of strongly measurable $E$-valued functions that are defined on $\Omega $ with the norm
$$\left\Vert f\right\Vert _{L_{p,\gamma }}=\left\Vert f\right\Vert
_{L_{p,\gamma }\left( \Omega ;E\right) }=\left( \int \left\Vert f\left(
x\right) \right\Vert _{E}^{p}\gamma \left( x\right) dx\right) ^{\frac{1}{p}},\text{ }1\leq p<\infty .$$
For $\gamma \left( x\right) \equiv 1$, $L_{p,\gamma }\left( \Omega ;E\right)
$ will be denoted by $L_{p}=L_{p}\left( \Omega ;E\right) $. Let $\mathbf{C}$ be the set of the complex numbers and $$S_{\varphi }=\left\{ \lambda ;\text{ \ }\lambda \in \mathbf{C}\text{, }\left\vert \arg \lambda \right\vert \leq \varphi \right\} \cup \left\{
0\right\} ,\text{ }0\leq \varphi <\pi .$$
A linear operator $A$ is said to be $\varphi $-positive in a Banach space $E$ with bound $M>0$ if $D\left( A\right) $ is dense on $E$ and $\left\Vert
\left( A+\lambda I\right) ^{-1}\right\Vert _{B\left( E\right) }\leq M\left(
1+\left\vert \lambda \right\vert \right) ^{-1}$ $\ $for any $\lambda \in
S_{\varphi },$ $0\leq \varphi <\pi ,$ where $I$ is the identity operator in $E$ and $B\left( E\right) $ is the space of bounded linear operators in $E.$
The $\varphi $-positive operator $A$ is said to be $R$-positive in a Banach space $E$ if the set $L_{A}=\left\{ \xi \left( A+\xi I\right) ^{-1}\text{: }\xi \in S_{\varphi }\right\} ,$ $0\leq \varphi <\pi $ is $R$-bounded (see e.g. $\left[ 5\right] $ ). Let $G$ be a domain in $R^{n}$. Let $W_{p,\gamma
}^{m}=W_{p,\gamma }^{m}\left( G;E\left( A\right) ,E\right) $ and $W_{p,\gamma }^{\left[ m\right] }=W_{p,\gamma }^{\left[ m\right] }\left(
0,a;E\left( A\right) ,E\right) $ are $E$-valued weighted function spaces defined in $\left[ 8\right] .$
**2. Linear degenerate DOEs**
Consider the BVP for the singular degenerate differential-operator equation $$-x^{2\alpha }\frac{\partial ^{2}u}{\partial x^{2}}-x^{2\beta }\frac{\partial
^{2}u}{\partial y^{2}}+Au+x^{\alpha }A_{1}\frac{\partial u}{\partial x}+y^{\beta }A_{2}\frac{\partial u}{\partial y}+\lambda u=f\left( x,y\right) ,$$
$$L_{1}u=\sum\limits_{i=0}^{m_{1}}\delta _{1i}u_{x}^{\left[ i\right] }\left(
a,y\right) =0,L_{2}u=\sum\limits_{i=0}^{m_{2}}\delta _{2i}u_{y}^{\left[ i\right] }\left( x,b\right) =0,\text{ }$$
on the domain $G=\left( 0,a\right) \times \left( 0,b\right) ,$ where $u=u\left( x,y\right) ,$ $u_{x}^{\left[ i\right] }=\left[ x^{\alpha }\frac{\partial }{\partial x}\right] ^{i}u\left( x,y\right) ,$ $u_{y}^{\left[ i\right] }=\left[ y^{\beta }\frac{\partial }{\partial y}\right] ^{i}u\left(
x,y\right) $, $m_{k}\in \left\{ 0,1\right\} $; $\delta _{jik}$ are complex numbers, $\lambda $ is a complex parameter, $A$ and $A_{i}=A_{i}\left(
x,y\right) $ are linear operators in a Banach space $E.$
Let $\delta _{1m_{k}}\neq 0,$ $k=1,2.$The main result is the following
**Theorem 1.** Let $E$ be an $UMD$ space space (see$\left[ 11\right] $ ), $A$ be an $R$-positive operator in $E,$ $A_{i}A^{-\left( \frac{1}{2}-\mu
\right) }\in L_{\infty }\left( G;B\left( E\right) \right) $ for $\mu \in
\left( 0,\frac{1}{2}\right) $ and $1+\frac{1}{p}<\alpha ,\beta <\frac{\left(
p-1\right) }{2}$. Then the problem $\left( 1\right) $ has a unique solution $u\in W_{p,\alpha ,\beta }^{\left[ 2\right] }\left( G;E\left( A\right)
,E\right) $ for all $f\in L_{p}\left( G;E\right) $ and $\left\vert \arg
\lambda \right\vert \leq \varphi $ with sufficently large $\left\vert
\lambda \right\vert $ and the following coercive uniform estimate holds
$$\sum\limits_{i=0}^{2}\left\vert \lambda \right\vert ^{1-\frac{i}{2}}\left[
\left\Vert x^{i\alpha }\frac{\partial ^{i}u}{\partial x^{i}}\right\Vert
_{L_{p}}+\left\Vert y^{i\beta }\frac{\partial ^{i}u}{\partial y^{i}}\right\Vert _{L_{p}}\right] +\left\Vert Au\right\Vert _{L_{p}}\leq
M\left\Vert f\right\Vert _{L_{p}}.$$
For proving the main theorem, consider at fist, BVPs for the singular degenerate DOE
$$\ \left( L+\lambda \right) u=-u^{\left[ 2\right] }\left( x\right) +\left(
A+\lambda \right) u\left( x\right) =f,$$
$$L_{1}u=\sum\limits_{i=0}^{m_{k}}\delta _{i}u^{\left[ i\right] }\left(
a\right) =0\text{, }$$
where $u^{\left[ i\right] }=\left[ x^{\alpha }\frac{d}{dx}\right]
^{i}u\left( x\right) $, $m_{k}\in \left\{ 0,1\right\} ;$ $\delta _{i}$ are complex numbers and $A$ is a linear operator in $E,$ $\delta _{m_{k}}\neq 0$.
In a similar way as in $\left[ \text{9, Theorem 5.1}\right] $ we obtain
**Theorem A**$_{1}$**.** Suppose** **$E$ is an $UMD$ space, $A$ is an $R$ positive in $E$, $1+\frac{1}{p}<\alpha <\frac{\left( p-1\right)
}{2}.$ Then the problem $\left( 3\right) $ has a unique solution $u\in
W_{p,\alpha }^{\left[ 2\right] }$ for all $f\in L_{p}\left( 0,a;E\right) ,$ $p\in \left( 1,\infty \right) .$ Moreover for $\left\vert \arg \lambda
\right\vert \leq \varphi $ and sufficiently large $\left\vert \lambda
\right\vert $ the following uniform coercive estimate holds
$$\sum\limits_{i=0}^{2}\left\vert \lambda \right\vert ^{1-\frac{i}{2}}\left\Vert u^{\left[ i\right] }\right\Vert _{L_{p}\left( 0,a;E\right)
}+\left\Vert Au\right\Vert _{L_{p}\left( 0,a;E\right) }\leq C\left\Vert
f\right\Vert _{L_{p}\left( 0,a;E\right) }.$$
Let $B$ denote the operator in $L_{p}\left( 0,a;E\right) $ generated by problem $\left( 2\right) $, i.e. $$D\left( B\right) =\left\{ u:u\in W_{p,\alpha }^{\left[ 2\right] },\text{ }L_{k}u=0\right\} ,\text{ }Bu=-u^{\left[ 2\right] }+Au.$$ In a similar way as in $\left[ \text{8, Theorem 3.1}\right] $ we obtain
**Theorem A**$_{2}.$ Let all conditions of Theorem A$_{1}$ are satisfied. Then, the operator $B$ is $R$-positive in $L_{p}\left(
0,a;E\right) .$
Theorem A$_{1}$ implies that the operator $B$ is positive and is a generator of analytic semigroups in $L_{p}\left( 0,a;E\right) $.
Consider now the following degenerate DOEs with the boundary conditions $\left( 3\right) :$$$\ -x^{2\alpha }u^{\left( 2\right) }\left( x\right) +\left( A+\lambda \right)
u\left( x\right) =f,\text{ }L_{k}u=0.$$
**Theorem A**$_{3}.$ Let all conditions of Theorem A$_{1}$ are satisfied. Then the problem $\left( 5\right) $ has a unique solution $u\in
W_{p,\alpha }^{2}$ for all $f\in L_{p}\left( 0,a;E\right) .$ Moreover for $\left\vert \arg \lambda \right\vert \leq \varphi $ and sufficiently large $\left\vert \lambda \right\vert $ the following coercive estimate holds
$$\sum\limits_{i=0}^{2}\left\vert \lambda \right\vert ^{1-\frac{i}{2}}\left\Vert x^{i\alpha }u^{\left( i\right) }\right\Vert _{L_{p}\left(
0,a;E\right) }+\left\Vert Au\right\Vert _{L_{p}\left( 0,a;E\right) }\leq
C\left\Vert f\right\Vert _{L_{p}\left( 0,a;E\right) }.$$
**Proof.** Since $\alpha >1,$ by $\left[ \text{9, Theorem 2.3}\right] $ we get that there is a small $\varepsilon >0$ and $C\left( \varepsilon
\right) $ such that $$\left\Vert \alpha x^{\alpha -1}u^{\left[ 1\right] }\right\Vert _{L_{p}\left(
0,a;E\right) }\leq \varepsilon \left\Vert u\right\Vert _{W_{p,\alpha }^{\left[ 2\right] }\left( 0,a;E\left( A\right) ,E\right) }+C\left( \varepsilon
\right) \left\Vert u\right\Vert _{L_{p}\left( 0,a;E\right) }.$$Then in view of $\left( 6\right) $, $\left( 7\right) $ and due to positivity of operator $B,$ we have the following estimate $$\left\Vert \alpha x^{\alpha -1}u^{\left[ 1\right] }\right\Vert _{L_{p}\left(
0,a;E\right) }\leq \varepsilon \left\Vert Bu\right\Vert _{L_{p}\left(
0,a;E\right) }.$$
Since $-x^{2\alpha }u^{\left( 2\right) }=-u^{\left[ 2\right] }+\alpha
x^{\alpha -1}u^{\left[ 1\right] },$ the assertion is obtained from Theorem A$_{1}$ and the estimate $\left( 8\right) .$
In this stage we can show the proof of Theorem1.
**Proof of Theorem 1.** Consider at first the principal part of the problem $\left( 1\right) $ i.e
$$-x^{2\alpha }\frac{\partial ^{2}u}{\partial x^{2}}-x^{2\beta }\frac{\partial
^{2}u}{\partial y^{2}}+Au+\lambda u=f\left( x,y\right) ,\text{ }L_{1}u=0\text{, }L_{2}u=0.$$
Since $L_{p}\left( 0,b;L_{p}\left( 0,a;E\right) \right) =$ $L_{p}\left(
G;E\right) $ then, this BVP can be express as:
$$-y^{2\beta }\frac{d^{2}u}{dy^{2}}+\left( B+\lambda \right) u\left( y\right)
=f\left( y\right) \text{, }L_{2k}u=0.$$
By virtue of $\left[ \text{1, Theorem 4.5.2}\right] $ $F=L_{p}\left(
0,b;E\right) \in UMD$ provided $E\in UMD$, $p\in \left( 1,\infty \right) $. By Theorem A$_{2},$ the operator $B\ $is $R$-positive in $F.$ Then by virtue of Theorem A$_{3}$, for $f\in L_{p}\left( 0,a;F\right) =L_{p}\left(
G;E\right) $ problem $\left( 9\right) $ has a unique solution $u\in $ $W_{p,\beta }^{2}\left( 0,a;D\left( S\right) ,F\right) $ and the operator $Q
$ generated by problem $\left( 9\right) $ has a bounded inverse from $L_{p}\left( G;E\right) $ to $W_{p,\alpha ,\beta }^{\left[ 2\right] }.$ Moreover by using embedding theorems in $W_{p,\alpha ,\beta }^{\left[ 2\right] }$ (see e.g. $\left[ \text{9, Theorem 2.3}\right] $ we get the following estimate$$\left\Vert x^{\alpha }A_{1}\frac{\partial u}{\partial x}\right\Vert
_{L_{p}\left( G;E\right) }+\left\Vert y^{\beta }A_{2}\frac{\partial u}{\partial y}\right\Vert _{L_{p}\left( G;E\right) }\leq \varepsilon \left\Vert
Qu\right\Vert _{L_{p}\left( G;E\right) }\text{, }\varepsilon <1.$$By virtue the above estimate and by using perturbation properties of linear operators we obtain the assertion.
**Remark 1.** Note that, by using the similar techniques similar to those applied in Theorems 1, 2, we can obtained the same results for differential-operator equations of the arbitrary order.
3\. **Singular degenerate BVPs with small parameters**
Consider the BVP for the parameter dependent degenerate differential-operator equation
$$\ Lu=-tu^{\left[ 2\right] }\left( x\right) +\left( A+\lambda \right) u\left(
x\right) =f,\text{ }x\in \left( 0,a\right)$$
$$L_{1}u=\sum\limits_{i=0}^{1}t^{\sigma _{i}}\alpha _{i}u^{\left[ i\right]
}\left( a\right) =f_{1},$$
where $u^{\left[ i\right] }=\left[ x^{\gamma }\frac{d}{dx}\right]
^{i}u\left( x\right) ,$ $\sigma _{i}=\frac{i}{2}+\frac{1}{2\left( 1-\gamma
\right) p},$ $\gamma >1+\frac{1}{p}$; $\alpha _{i}$ are complex numbers $t$ is a small positive and $\lambda $ is a complex parameter, $A$ is a linear operator in a Banach space $E$ and $f_{1}\in E_{1}=\left( \left( E\left(
A\right) ,E\right) _{\theta }\right) ,$ $\theta =\frac{1}{2}\left( 1+\frac{1}{\left( 1-\alpha \right) p}\right) .$
A function $u\in $ $W_{p,\gamma }^{\left[ 2\right] }\left( 0,a;E\left(
A\right) ,E\right) $ satisfying the equation $\left( 1\right) $ a.e. on $\left( 0,1\right) $ is said to be the solution of the equation $\left(
1\right) $ on $\left( 0,1\right) .$
**Remark 2.** Let $$y=\int\limits_{0}^{x}z^{-\gamma }dz.$$
Under the substitution $\left( 11\right) $ spaces $L_{p}\left( 0,1;E\right) $ and $W_{p,\gamma }^{\left[ 2\right] }\left( 0,a;E\left( A\right) ,E\right) $ are mapped isomorphically onto weighted spaces $L_{p,\tilde{\gamma}}\left(
-\infty ,0;E\right) $ and $$W_{p,\tilde{\gamma}}^{2}\left( -\infty ,0;E\left( A\right) ,E\right) ,\text{
}\tilde{\gamma}=\tilde{\gamma}\left( x\left( y\right) \right) .$$Moreover, under the substitution $\left( 10\right) $ the problem $\left(
2\right) $ is transformed into a non degenerate problem
$$Lu=-tu^{\left( 2\right) }\left( y\right) +Au\left( y\right) =f,\text{ }L_{1}u=\sum\limits_{i=0}^{1}t^{\sigma _{i}}\alpha _{i}u^{\left( i\right)
}\left( 0\right) =f_{1},$$
**Theorem 2.** Suppose $E$ is a UMD and $1+\frac{1}{p}<\gamma
,1<p<\infty .$ Then the problem $\left( 10\right) $ for all $f\in
L_{p}\left( 0,a;E\right) ,$ $f_{1}\in E_{1}$ has a unique solution $u\in
W_{p,\gamma }^{\left[ 2\right] }\left( 0,a;E\left( A\right) ,E\right) $ and for $\left\vert \arg \lambda \right\vert \leq \varphi $ and sufficiently large $\left\vert \lambda \right\vert $ the following uniform coercive estimate holds
$$\sum\limits_{i=0}^{2}\left\vert \lambda \right\vert ^{1-\frac{i}{2}}t^{\frac{i}{2}}\left\Vert u^{\left[ i\right] }\right\Vert _{L_{p}\left( 0,a;E\right)
}+\left\Vert Au\right\Vert _{L_{p}\left( 0,a;E\right) }\leq C\left[
\left\Vert f\right\Vert _{L_{p}\left( 0,a;E\right) }+\left\Vert
f_{1}\right\Vert _{E_{1}}+\left\vert \lambda \right\vert ^{1-\theta
}\left\Vert f_{1}\right\Vert _{E}.\right]$$
**Proof.** Consider the problem $\left( 12\right) $. In a similar way as in $\left[ \text{10, Theorem 3.2}\right] $ we obtain that the problem $\left( 12\right) $ has a unique solution $u\in W_{p,\tilde{\gamma}}^{\left(
2\right) }\left( -\infty ,0;E\left( A\right) ,E\right) $ for all $f\in
L_{p,\tilde{\gamma}}\left( -\infty ,0;E\right) ,$ $f_{1}\in E_{1}$ and $\left\vert \arg \lambda \right\vert \leq \varphi $ with sufficiently large $\left\vert \lambda \right\vert $ the following uniform coercive estimate holds
$$\sum\limits_{i=0}^{2}\left\vert \lambda \right\vert ^{1-\frac{i}{2}}t^{\frac{i}{2}}\left\Vert u^{\left( i\right) }\right\Vert _{L_{p,\tilde{\gamma}}\left( -\infty ,0;E\right) }+\left\Vert Au\right\Vert _{L_{p,\tilde{\gamma}}\left( -\infty ,0;E\right) }\leq C\left[ \left\Vert f\right\Vert _{L_{p,\tilde{\gamma}}\left( -\infty ,0;E\right) }+\left\Vert f_{1}\right\Vert
_{E_{1}}+\left\vert \lambda \right\vert ^{1-\theta }\left\Vert
f_{1}\right\Vert _{E}.\right]$$
Then in view of the Remark 2 we obtain the assertion.
Consider now the parameter dependent singular degenerate BVP$$-t_{1}x^{2\alpha }\frac{\partial ^{2}u}{\partial x^{2}}-t_{2}x^{2\beta }\frac{\partial ^{2}u}{\partial y^{2}}+Au+\lambda u=f\left( x,y\right) ,$$
$$L_{1}u=\sum\limits_{i=0}^{m_{1}}t_{1}^{\sigma _{1i}}\delta _{1i}u_{x}^{\left[
i\right] }\left( a,y\right) =0,L_{2}u=\sum\limits_{i=0}^{m_{2}}t_{2}^{\sigma
_{2i}}\delta _{2i}u_{y}^{\left[ i\right] }\left( x,b\right) =0,\text{ }$$
on the domain $G=\left( 0,a\right) \times \left( 0,b\right) $, where $\sigma
_{1i}=\frac{i}{2}+\frac{1}{2p\left( 1-\alpha \right) },$ $\sigma _{2i}=\frac{i}{2}+\frac{1}{2p\left( 1-\beta \right) }$, $A$ is a linear operator in a Banach space $E$ and $t_{k}$ are small parameters.
**Theorem 3.** Let $E$ be an $UMD$ space space, $A$ be an $R$-positive operator in and $1+\frac{1}{p}<\alpha ,\beta <\frac{\left( p-1\right) }{2}$. Then the problem $\left( 14\right) $ has a unique solution $u\in W_{p,\alpha
,\beta }^{\left[ 2\right] }\left( G;E\left( A\right) ,E\right) $ for all $f\in L_{p}\left( G;E\right) $ and $\left\vert \arg \lambda \right\vert \leq
\varphi $, with sufficiently large $\left\vert \lambda \right\vert $ and the following coercive uniform estimate holds
$$\sum\limits_{i=0}^{2}\left\vert \lambda \right\vert ^{1-\frac{i}{2}}\left[
t_{1}^{i}\left\Vert x^{i\alpha }\frac{\partial ^{i}u}{\partial x^{i}}\right\Vert _{L_{p}}+t_{2}^{i}\left\Vert y^{i\beta }\frac{\partial ^{i}u}{\partial y^{i}}\right\Vert _{L_{p}}\right] +\left\Vert Au\right\Vert
_{L_{p}}\leq M\left\Vert f\right\Vert _{L_{p}}.$$
**Proof.** By reasoning as in the proof of Theorem1, the problem $\left( 14\right) $ is reduced to the following BVP for ordinary equation $$\ -t_{2}u^{\left[ 2\right] }\left( y\right) +\left( B_{t_{1}}+\lambda
\right) u\left( y\right) =f,\text{ }y\in \left( 0,b\right)$$$$L_{2}u=\sum\limits_{i=0}^{1}t_{2}^{\sigma _{i}}\alpha _{i}u^{\left[ i\right]
}\left( b\right) =0,$$
where $B_{t_{2}}$ is the operator in $L_{p}\left( 0,b;E\right) $ generated by BVP $$\ Lu=-t_{1}u^{\left[ 2\right] }\left( x\right) +\left( A+\lambda \right)
u\left( x\right) =f,\text{ }x\in \left( 0,a\right)$$$$L_{1}u=\sum\limits_{i=0}^{1}t_{1}^{\sigma _{i}}\alpha _{i}u^{\left[ i\right]
}\left( a\right) =0.$$
Then by applying Theorem 2 to problem $\left( 16\right) $ in $L_{p}\left(
0,a;F\right) =L_{p}\left( G;E\right) ,$ $F=L_{p}\left( 0,b;E\right) $ we obtain the assertion.
4\. **Singular degenerate BVPs in moving domains**
Consider the linear BVPs in moving domain $G_{s}=\left( 0,a\left( s\right)
\right) \left( 0,b\left( s\right) \right) $$$-x^{2\alpha }\frac{\partial ^{2}u}{\partial x^{2}}-x^{2\beta }\frac{\partial
^{2}u}{\partial y^{2}}+Au+du=f\left( x,y\right) ,$$
$$L_{1}u=\sum\limits_{i=0}^{1}\delta _{1i}u_{x}^{\left[ i\right] }\left(
a\left( s\right) ,y\right) =0,L_{2}u=\sum\limits_{i=0}^{1}\delta _{2i}u_{y}^{\left[ i\right] }\left( x,b\left( s\right) \right) =0,\text{ }k=1,2,$$
where $a\left( s\right) $ and $b(s)$ are positive continues function depended on parameter $s$ and $A$ is a linear operator in a Banach space $E,$ $\delta _{k1}$ $\neq 0.$
**Theorem 4.** Let $E$ be an $UMD$ space space, $A$ be an $R$-positive operator in $E,$ $1+\frac{1}{p}<\alpha ,\beta <\frac{\left( p-1\right) }{2}$ and $f_{k}\in E_{k}$.
Then problem $\left( 10\right) $ for $f\in L_{p}\left( G\left( s\right)
;E\right) $, $f_{k}\in E_{k},$ $p\in \left( 1,\infty \right) $ and the sufficiently large $d>0$ has a unique solution $u\in $ $W_{p,\alpha ,\beta
}^{2}\left( G\left( s\right) ;E\left( A\right) ,E\right) $ and the following coercive uniform estimate holds
$$\left\Vert x^{2\alpha }\frac{\partial ^{2}u}{\partial x^{2}}\right\Vert
_{L_{p}\left( G\left( s\right) ;E\right) }+\left\Vert y^{2\beta }\frac{\partial ^{2}u}{\partial y^{2}}\right\Vert _{L_{p}\left( G\left( s\right)
;E\right) }+\left\Vert Au\right\Vert _{L_{p}\left( G\left( s\right)
;E\right) }\leq$$
$$C\left\Vert f\right\Vert _{L_{p}\left( G_{s};E\right) }.$$
**Proof.** Under the substitution $\tau =xb(s),$ $\sigma =yb\left(
s\right) $ by denoting $\tau $, $\sigma ,$ $u\left( x\left( \tau \right)
,y\left( \sigma \right) \right) ,$ $f\left( x\left( \tau \right) ,y\left(
\sigma \right) \right) $ again by $x,$ $y,$ $u\left( x,y\right) ,$ $f\left(
x,y\right) ,$ respectively we get the moving boundary problem $\left(
10\right) $ maps to the following BVP with parameter in the fixed domain $\left( 0,1\right) \times \left( 0,1\right) $$$-b^{2\left( 1-\alpha \right) }\left( s\right) x^{2\alpha }\frac{\partial
^{2}u}{\partial x^{2}}-b^{2\left( 1-\beta \right) }\left( s\right) x^{2\beta
}\frac{\partial ^{2}u}{\partial y^{2}}+Au+du=f\left( x,y\right) ,$$
$$L_{1}u=\sum\limits_{i=0}^{1}b^{i}\left( s\right) \delta _{1i}u_{x}^{\left[ i\right] }\left( 1,y\right) =0,L_{2}u=\sum\limits_{i=0}^{1}b^{i}\left(
s\right) \delta _{2i}u_{y}^{\left[ i\right] }\left( x,1\right) =0,\text{ }k=1,2.$$
Then by virtue of Theorem 3 we obtain the assertion.
**5**. **Nonlinear degenerate DOE**
Consider now the following nonlinear problem$$-x^{2\alpha }\frac{\partial ^{2}u}{\partial x^{2}}-y^{2\beta }\frac{\partial
^{2}u}{\partial y^{2}}+a\left( u,u_{x},u_{y}\right) u=F\left(
x,y,u,u_{x},u_{y}\right) ,\text{ }L_{1k}u=0,\text{ }L_{2k}u=0\text{ }$$on the domain $G_{0}=\left( 0,a_{0}\right) \times \left( 0,b_{0}\right) ,$ where $L_{jk}$ are boundary conditions defined by $\left( 1\right) .$ Let $$X_{1}=L_{p}\left( 0,a;E\right) \text{, }Y_{1}=W_{p,\alpha }^{\left[ 2\right]
}\left( 0,a;E\left( A\right) ,E\right) ,\text{ }X_{2}=L_{p}\left(
0,b;E\right) ,$$$$Y_{2}=W_{p,\beta }^{\left[ 2\right] }\left( 0,b;E\left( A\right) ,E\right) ,\text{ }E_{ki}=\left( X_{k},Y_{k}\right) _{\theta _{ki},p},\text{ }\theta
_{1i}=\frac{p\left( 1-\alpha \right) i+1}{2p\left( 1-\alpha \right) },\text{
}$$
$$\theta _{2i}=\frac{p\left( 1-\beta \right) i+1}{2p\left( 1-\beta \right) },\text{ }E_{0}=\prod\limits_{i,k}E_{ki},\text{ }i=0,1,\text{ }k=1,2.$$
**Condition 1.** Assume the following satisfied:
(1) $E$ is an $UMD$ space$,$ $a\left( x,y,U\right) =A\left( x,y\right) $ is a positive operator in $E$ for $x,y\in G_{0}$, $u_{i}\in E_{1i},$ $g_{i}\in
E_{1i}$, $D\left( a\left( x,y,U\right) \right) $ does not depend on $x,y,U$, where $U=\left\{ u_{0},u_{1},g_{0},g_{1}\right\} $ and $a:$ $G_{0}\times
E_{0}\rightarrow B\left( E\left( A\right) ,E\right) $ is continuous;
\(2) $F:G_{0}\times E_{0}\rightarrow E$ be a measurable function; $F\left(
x,y,.\right) $ is continuous with respect to $x,y\in G_{0}$ and $f\left(
x,y\right) =$ $F\left( x,y,0\right) \in X.$ Moreover, for each $R>0$ there exists $\mu _{R}$ such that $\left\Vert F\left( x,U\right) -F\left( x,\bar{U}\right) \right\Vert _{E}\leq \mu _{R}\left\Vert U-\bar{U}\right\Vert
_{E_{0}} $ for a.a. $x,y\in G_{0}$, $u_{j},$ $\bar{u}_{j}\in X_{j}$ and $\left\Vert U\right\Vert _{E_{0}}\leq R,\left\Vert \bar{U}\right\Vert
_{E_{0}}\leq R$, $1+\frac{1}{p}<\alpha ,\beta <\frac{\left( p-1\right) }{2}$;
\(3) there exist $v_{j}\in E_{1j}$, $\upsilon _{j}\in E_{2j}$ such that the operator $A\left( x,y,\Phi \right) $ for $\Phi =\left\{ v_{1},v_{2},\upsilon
_{1},\upsilon _{2}\right\} $ is $R$-positive in $E$ uniformly with respect to $x,y\in G_{0};$ $A\left( x,y,\Phi \right) A^{-1}\left( x^{0},y^{0},\Phi
\right) \in C\left( G_{0};B\left( E\right) \right) ;$
\(4) Moreover, for each $R>0$ there is a positive constant $L\left( R\right) $ such that
$\left\Vert \left[ A\left( x,y,U\right) -A\left( x,y,\bar{U}\right) \right]
\upsilon \right\Vert _{E}\leq M\left( R\right) \left\Vert U-\bar{U}\right\Vert _{E_{0}}\left\Vert A\upsilon \right\Vert _{E}$ for $x,y\in G_{0}$, $\left\Vert U\right\Vert _{E_{0}},\left\Vert \bar{U}\right\Vert
_{E_{0}}\leq R$ and $\upsilon \in D\left( A\left( x,y,U\right) \right) .$
In this section we prove the existence and uniqueness of maximal regular solution for the nonlinear problem $\left( 11\right) $.
**Theorem 3.** Let the Condition1 holds. Then there is $a\in \left(
0\right. \left. a_{0}\right] ,$ $b\in \left( 0\right. \left. b_{0}\right] $ such that problem $\left( 11\right) $ has a unique solution belongs to $W_{p,\alpha ,\beta }^{2}\left( (G;E\left( A\right) ,E\right) .$
**Proof.** By Theorem , the linear problem $$\ -x^{2\alpha }\frac{\partial ^{2}w}{\partial x^{2}}-y^{2\beta }\frac{\partial ^{2}w}{\partial y^{2}}+Aw\left( x,y\right) =f\left( x,y\right) ,$$$$L_{1k}u=0,\text{ }L_{2k}u=0,\text{ }k=1,2,\text{ }x,y\in \left( 0,a\right)
\times \left( 0,b\right)$$is maximal regular in $X$ uniformly with respect to $a\in \left( 0\right.
\left. a_{0}\right] $ and $b\in \left( 0\right. \left. b_{0}\right] $ i.e. for all $f\in X$ there is a unique solution $w\in Y$ of the problem $(31)$ and has a coercive estimate $$\left\Vert w\right\Vert _{Y}\leq C\left\Vert f\right\Vert _{X},$$where the constant $C$ does not depends on $a\in \left( 0\right. \left.
a_{0}\right] $ and $$f\left( x\right) =F\left( x,0\right) .$$We want to to solve the problem $\left( 30\right) $ locally by means of maximal regularity of the linear problem $(31)$ via the contraction mapping theorem. For this purpose let $w$ be a solution of the linear BVP $(31).$ Consider a ball $$B_{r}=\left\{ \upsilon \in Y,\text{ }\left\Vert \upsilon -w\right\Vert
_{Y}\leq r\right\} .$$
Given $\upsilon \in B_{r},$ solve the problem$$-tu^{\left( 2m\right) }\left( x\right) +Au\left( x\right) =F\left(
x,\upsilon ,\upsilon ^{\left( 1\right) },...,\upsilon ^{\left( 2m-1\right)
}\right) ,\text{ }$$$$\sum\limits_{i=0}^{m_{k}}t^{\eta _{i}}\left[ \alpha _{ki}u^{\left( i\right)
}\left( 0\right) +\beta _{ki}u^{\left( i\right) }\left( a\right)
+\sum\limits_{j=1}^{N_{k}}\delta _{kj}u^{\left( i\right) }\left(
x_{kj}\right) \right] =f_{k},\text{ }k=1,2,....,2m$$where $x\in \left( 0,a\right) .$ Define a map $Q$ on $B_{r}$ by $Q\upsilon
=u,$ where $u$ is a solution of the problem $\left( 38\right) .$ We want to show that $Q\left( B_{r}\right) \subset B_{r}$ and that $L$ is a contraction operator in $Y$, provided $a$ is sufficiently small, and $r$ is chosen properly. For this aim by using maximal regularity properties of the problem $\left( 37\right) $ for $V=\left\{ \upsilon ^{\left( m_{k}\right) }\left(
0\right) \right\} ,$ $k=1,2,...,2m$ we have $$\left\Vert Q\upsilon -w\right\Vert _{Y}=\left\Vert u-w\right\Vert _{Y}\leq
C_{0}\left\Vert F\left( x,V\right) -F\left( x,0\right) \right\Vert _{X}.$$
By assumption Condition1 and in view of Remark1 we have $$\left\Vert F\left( x,V\right) -F\left( x,0\right) \right\Vert _{E}\leq$$$$\left\Vert F\left( x,V\right) -F\left( x,W\right) \right\Vert
_{E}+\left\Vert F\left( x,W\right) -F\left( x,0\right) \right\Vert _{E}\leq$$
$$M_{R}\left[ \left\Vert V-W\right\Vert _{E_{0}}+\left\Vert W\right\Vert
_{E_{0}}\right]$$
$$M_{R}C_{1}\left[ \left\Vert \upsilon -w\right\Vert _{Y}+\left\Vert
w\right\Vert _{Y}\right] \leq MC_{1}\left[ r+\left\Vert w\right\Vert _{Y}\right] ,$$
where $R=M_{R}C_{1}\left[ r+\left\Vert w\right\Vert _{Y}\right] $ is a fixed number such that $R\leq \frac{r}{C_{0}}$. In view of Condition C and the above estimates for sufficiently small $a\in \lbrack 0;a_{0})$we have $$\left\Vert Q\upsilon -w\right\Vert _{Y}\leq C_{0}R\leq r$$i.e. $$Q\left( B_{r}\right) \subset B_{r}.$$In a similar way for $\bar{V}=\left\{ \bar{\upsilon}^{\left( m_{k}\right)
}\left( 0\right) \right\} $ we obtain $$\left\Vert Q\upsilon -Q\bar{\upsilon}\right\Vert _{Y}\leq C_{0}\left[
\left\Vert F\left( x,V\right) -F\left( x,\bar{V}\right) \right\Vert _{X}\right] \leq C_{0}M_{R}\left\Vert \left( \upsilon -\bar{\upsilon}\right)
\right\Vert _{Y}.$$Therefore for $C_{0}M_{R}<1$ the operator $Q$ becomes a contraction mapping. Eventually, the contraction mapping principle implies a unique fixed point of $Q$ in $B_{r}$ which is the unique strong solution $u\in
Y=W_{p}^{2m}\left( 0,a;E\left( A\right) ,E\right) .$
** 4. Singular degenerate boundary value problems for infinite systems of equations**
Consider the infinite system of BVPs$$-x^{2\alpha }\frac{\partial ^{2}u_{m}}{\partial x^{2}}-x^{2\beta }\frac{\partial ^{2}u_{m}}{\partial y^{2}}+d_{m}u_{m}+\dsum\limits_{j=1}^{\infty
}x^{\alpha }a_{mj}\left( x,y\right) \frac{\partial u_{j}}{\partial x}$$$$+\dsum\limits_{j=1}^{\infty }y^{\beta }b_{mj}\left( x,y\right) \frac{\partial u_{j}}{\partial y}+\lambda u=f_{m}\left( x,y\right) ,\text{ }L_{1k}u=0,\text{ }L_{2k}u=0,$$
where $L_{ik}$ are defined by $\left( 1\right) $. Let $D$ $$D=\left\{ d_{m}\right\} ,\text{ }d_{m}>0,\text{ }u=\left\{ u_{m}\right\} ,\text{ }Du=\left\{ d_{m}u_{m}\right\} ,\text{ }m=1,2,...,$$
$$\text{ }l_{q}\left( D\right) =\left\{ u\text{: }u\in l_{q},\right.
=\left\Vert u\right\Vert _{l_{q}\left( D\right) }=\left. \left(
\sum\limits_{m=1}^{\infty }\left\vert d_{m}u_{m}\right\vert ^{q}\right) ^{\frac{1}{q}}<\infty ,q\in \left( 1,\infty \right) \right\} .$$
From Theorem1we obtain
**Theorem 3.** Assume $a_{mj},b_{mj}\in L_{\infty }\left( G\right) $. For $0<\mu <\frac{1}{2}$ and for all $x,y\in \left( G\right) $ $$\text{ }\sup\limits_{m}\sum\limits_{j=1}^{\infty }a_{mj}\left( x\right)
d_{j}^{-\left( \frac{1}{2}-\mu \right) }<M,\text{ }\sup\limits_{m}\sum\limits_{j=1}^{\infty }b_{mj}\left( x\right) d_{j}^{-\left( \frac{1}{2}-\mu
\right) }.$$Then for all $f\left( x\right) =\left\{ f_{m}\left( x\right) \right\}
_{1}^{\infty }\in L_{p}\left( \left( G\right) ;l_{q}\right) ,$ $p,q\in
\left( 1,\infty \right) $, $\left\vert \arg \lambda \right\vert \leq \varphi
$, $0\leq \varphi <\pi $ and for sufficiently large $\left\vert \lambda
\right\vert $ problem $\left( 12\right) $ has a unique solution $u=\left\{
u_{m}\left( x\right) \right\} _{1}^{\infty }$ that belongs to space $W_{p,\alpha ,\beta }^{2}\left( G,l_{q}\left( D\right) ,l_{q}\right) $ and $$\begin{aligned}
&&\sum\limits_{i=0}^{2}\left\vert \lambda \right\vert ^{1-\frac{i}{2}}\left[
\left\Vert x^{i\alpha }\frac{\partial ^{i}u}{\partial x^{i}}\right\Vert
_{L_{p}\left( G;l_{q}\right) }+\left\Vert y^{i\beta }\frac{\partial ^{i}u}{\partial y^{i}}\right\Vert _{L_{p}\left( G;l_{q}\right) }\right] +\left\Vert
Du\right\Vert _{L_{p}\left( G;l_{q}\right) } \\
&\leq &M\left\Vert f\right\Vert _{L_{p}\left( G;l_{q}\right) }. \notag\end{aligned}$$
**References**
1. Amann H., Linear and quasi-linear equations,1, Birkhauser, Basel 1995.
2. Yakubov S. and Yakubov Ya., Differential-operator Equations. Ordinary and Partial Differential Equations , Chapman and Hall /CRC, Boca Raton, 2000.
3. Krein S. G., Linear differential equations in Banach space, American Mathematical Society, Providence, 1971.
4. Sobolevskii P. E., Coerciveness inequalities for abstract parabolic equations, Doklady Akademii Nauk SSSR, 57(1),(1964), 27-40.
5. Denk R., Hieber M., Prüss J., $R$-boundedness, Fourier multipliers and problems of elliptic and parabolic type, Mem. Amer. Math. Soc. 166 (2003), n.788.
6. Favini A., Shakhmurov V., Yakubov Y., Regular boundary value problems for complete second order elliptic differential-operator equations in UMD Banach spaces, Semigroup form, v. 79 (1), 2009, 22-54.
7. Ashyralyev A., Claudio Cuevas and Piskarev S., “On well-posedness of difference schemes for abstract elliptic problems in spaces”, Numerical Functional Analysis & Optimization, v. 29, No. 1-2, Jan. 2008, 43-65.
8. Shakhmurov V. B, Shahmurova A., Nonlinear abstract boundary value problems atmospheric dispersion of pollutants, Nonlinear Analysis, Wold Applications, v.11 (2) 2010, 932-951.
9. Shakhmurov V. B., Degenerate differential operators with parameters, Abstract and Applied Analysis, 2007, v. 2007, 1-27.
10. Shakhmurov V. B., Nonlinear abstract boundary value problems in vector-valued function spaces and applications, Nonlinear Analysis Series A: Theory, Method & Applications, v. 73, 2010, 2383-2397.
11. Burkholder D. L., A geometrical conditions that implies the existence certain singular integral of Banach space-valued functions, Proc. conf. Harmonic analysis in honor of Antonu Zigmund, Chicago, 1981,Wads Worth, Belmont, (1983), 270-286.
12. Triebel H., Interpolation theory, Function spaces, Differential operators., North-Holland, Amsterdam, 1978.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'I study a special type of canonical relations given by twisted conormal bundles, construct a “subcategory” of the symplectic “category” out of these canonical relations and quantize them into semi-classical Fourier integral operators. Furthermore, I give a description of the intrinsic line bundle of symbols of these operators and describe how the symbols compose when the operators compose.'
author:
- Zongrui Yang
bibliography:
- 'bibfile.bib'
title: Twisted Conormal Bundles and Canonical Relations
---
Introduction
============
For a submanifold $Z$ of a smooth manifold $X$, let $f\in C^{\infty}(Z)$ be a smooth function on $Z$. Define the $f$-twisted conormal bundle of $Z$ to be the set $$N^*_fZ=\left\{(x, \xi+df_x): (x, \xi)\in N^*Z \right\}$$ where we also use $f$ to denote any extension of $f$ to $X$. It is easy to see that this is independent of the choice of the extension and is a Lagrangian submanifold of $T^*X$ .
For a submanifold $Z\subset X\times Y$ and $f\in C^\infty(Z)$, we can act the map $$\zeta_X:T^*X\rightarrow T^*X, (x, \xi)\mapsto (x, -\xi)$$ to $$N^*_fZ\subset T^*(X\times Y)=T^*X\times T^*Y$$ and get the Lagrangian submanifold $$\Gamma_{Z, f}=(\zeta_X \times id)N^*_fZ\subset T^*X^{-}\times T^*Y,$$ where $T^*X^{-}$ is the manifold $T^*X$ with its reversed symplectic form and $\zeta_X$ is the symplectomorphism from $T^*X$ to $T^*X^{-}$. Denote $\Gamma_{Z, f}$ by $\Gamma_Z$ in the non-twisted case $f=0$. Note that as Lagrangian submanifolds of $T^*X\times T^*Y^{-}$, we can view $\Gamma_{Z, f}$ and $\Gamma_Z$ as canonical relations from $T^*X$ to $T^*Y$.
In section 2 we will show that, under certain conditions, the composition of canonical relations $\Gamma_{Z}$(or $\Gamma_{Z, f}$) in the symplectic “category” $\mathcal{S}$ is again of this form. Specifically, we will prove:
Suppose $X_1, X_2, X_3$ are smooth manifolds and $Z_1\subset X_1\times X_2, Z_2\subset X_2\times X_3$ are submanifolds. Suppose $\Gamma_{Z_1}=(\zeta_1 \times id)N^*Z_1 \in Morph(T^*X_1, T^*X_2)$, $\Gamma_{Z_2}=(\zeta_2 \times id)N^*Z_2 \in Morph(T^*X_2, T^*X_3)$ are cleanly composable canonical relations, $$Z=\left\{ (x_1,x_3)\in X_1\times X_3 : {\exists}x_2\in X_2, s.t. (x_1, x_2)\in Z_1, (x_2, x_3)\in Z_2 \right\}$$ is a submanifold of $X_1\times X_3$, and the projection $$\Gamma_{Z_2}\circ\Gamma_{Z_1}\rightarrow Z$$ is a submersion in the category of smooth manifolds. Then $$\Gamma_{Z_2}\circ\Gamma_{Z_1}=\Gamma_{Z}$$where $\Gamma_{Z}=(\zeta_1 \times id)N^*Z \in Morph(T^*X_1, T^*X_3)$.
Suppose $X_1, X_2, X_3$ are smooth manifolds, $Z_1\subset X_1\times X_2, Z_2\subset X_2\times X_3$ are submanifolds and $f_1\in C^{\infty}(Z_1), f_2\in C^{\infty}(Z_2) $ are smooth functions. Suppose $\Gamma_{Z_1, f_1}=(\zeta_1 \times id)N^*_{f_1}Z_1 \in Morph(T^*X_1, T^*X_2)$, $\Gamma_{Z_2, f_2}=(\zeta_2 \times id)N^*_{f_2}Z_2\in Morph(T^*X_2, T^*X_3)$ are cleanly composable canonical relations, $$Z=\left\{ (x_1,x_3)\in X_1\times X_3 : \exists x_2\in X_2, s.t. (x_1, x_2)\in Z_1, (x_2, x_3)\in Z_2 \right\}$$is a simply connected submanifold of $X_1\times X_3$, the projection $$\Gamma_{Z_2, f_2}\circ\Gamma_{Z_1, f_1}\rightarrow Z$$ is a submersion in the category of smooth manifolds, and $$d_2f_1(x_1, x_2)+d_2f_2(x_2, x_3)=0\in T^*X_2$$ for every $(x_1, x_2, x_3)\in X_1\times X_2\times X_3$. Then there exists $f\in C^{\infty}(Z)$ such that $$\Gamma_{Z_2, f_2}\circ\Gamma_{Z_1, f_1}=\Gamma_{Z, f}$$where $\Gamma_{Z, f}=(\zeta_1 \times id)N^*_{f}Z \in Morph(T^*X_1, T^*X_3)$.
We then have a natural picture of a functor projecting the “upatairs category” to the “downstairs category”. Specifically, we can define $\mathcal{C}_0$ and $\mathcal{C}$ to be “categories” with cotangent bundles $T^*X$ as objects and morphisms $$Morph_{\mathcal{C}_0}(T^*X, T^*Y)=\left\{ \Gamma_Z: Z\subset X\times Y\text{ is a submanifold} \right\}$$ $$Morph_{\mathcal{C}}(T^*X, T^*Y)=\left\{\Gamma_{Z, f}: Z\subset X\times Y\text{ is a submanifold and }f \in C^\infty(Z)\right\}.$$ Two morphisms are defined to be composable when they satisfy the conditions in Theorem 1 or 2. Note that $\mathcal{C}_0$ and $\mathcal{C}$ are “subcategories” of the symplectic “category” $\mathcal{S}$.
Define the ‘downstairs category’’ $\mathcal{D}$ to be the “category” with smooth manifolds as objects and morphisms $$Morph(X, Y)=\left\{\text{submanifolds }Z\subset X\times Y\right\}$$ where $Z_1\subset X_1\times X_2$ and $Z_2\subset X_2\times X_3$ are composable if and only if $$Z=\left\{ (x_1,x_3)\in X_1\times X_3 : \exists x_2\in X_2, s.t. (x_1, x_2)\in Z_1, (x_2, x_3)\in Z_2 \right\}$$ is a submanifold of $X_1\times X_3$, in which case we define their composition to be $Z$. We then have the natural “projection” functors $$\mathcal{C}_0\rightarrow\mathcal{D}$$and $$\mathcal{C}\rightarrow\mathcal{D}$$ sending the “upstairs objects” $T^*X$ to “downstairs objects” $X$ and “upstairs morphisms” $\Gamma_Z$ and $\Gamma_{Z, f}$ to “downstairs morphisms” $Z$.
Note that in order to guarantee $$Z=Z_2\circ Z_1=\left\{ (x_1,x_3)\in X_1\times X_3 : \exists x_2\in X_2, s.t. (x_1, x_2)\in Z_1, (x_2, x_3)\in Z_2 \right\}.$$ is a submanifold, we can naturally define clean composition conditions of submanifolds $Z_1$ and $Z_2$ imitating the clean composition of canonical relations. Namely we call $Z_1\subset X_1\times X_2$ and $Z_2\subset X_2\times X_3$ cleanly composable if $$Z_2\star Z_1=\left\{(x_1, x_2, x_3)\in X_1\times X_2\times X_3: (x_1, x_2)\in Z_1, (x_2, x_3)\in Z_2\right\}$$ is a submanifold of $X_1\times X_2\times X_3$, and the projection $$\kappa: Z_2\star Z_1\rightarrow Z_2\circ Z_1$$ to be proper of constant rank, with each fiber simply connected.
Now that since we have defined the clean composition conditions of submanifolds, we can rephrase Theorem 1 by saying that if the “downstairs objects” $Z_1$, $Z_2$ and “upstairs objects” $\Gamma_{Z_1}$ and $\Gamma_{Z_2}$ are both cleanly composable, and the projection $$\Gamma_{Z_2}\circ\Gamma_{Z_1}\rightarrow Z_2\circ Z_1$$ is a submersion, then the composition of the “upstairs objects” corresponds to the composition of “downstairs objects” by the simple formula $$\Gamma_{Z_2}\circ\Gamma_{Z_1}=\Gamma_{Z_2\circ Z_1}.$$
The second part of the paper will show how to quantize our type canonical relations via the techniques of Hörmander (see \[1\] and \[2\]). In section 3 we will give the following result useful in the quantization procedure.
Suppose $X$ is a smooth manifold of dimension n, $\Lambda$ is a Lagrangian submanifold of $T^*X$ and $p\in\Lambda$ such that $\pi\arrowvert_\Lambda$ locally at p is a submersion onto a (n-k)-dimensional submanifold $Z\subset X$ with every fiber $\pi^{-1}(z), z\in Z$ connected. Then $\Lambda$ can be described locally at p in Hörmander sense as a fiber bundle $W=X\times \mathbb{R}^{k}\rightarrow X$ and a generating function $\phi: W\rightarrow\mathbb{R}$.
Since twisted conormal bundles are exact Lagrangian submanifolds, we will quantize them by associating the space of oscillatory $\frac{1}{2}$-densities and semi-classical Fourier integral operators. Section 4 will compute these spaces explicitly. In secion 5 we talk about symbols of such operators. Note that since the Maslov bundle is always trivial, Hörmander’s “intrinsic line bundle of symbols” $$\mathbb{L}=\mathbb{L}_{Maslov}(\Lambda)\otimes\left|T\Lambda\right|^{\frac{1}{2}}$$ is generally isomorphic to the $\frac{1}{2}$-density bundle $\left|T\Lambda\right|^{\frac{1}{2}}$ . We will prove that there is a canonical isomorphism $\mathbb{L}\cong\left|T\Lambda\right|^{\frac{1}{2}}$ in the case when $\Lambda$ is a twisted conormal bundle. Specifically, we will prove
Suppose $Z\subset X$ is a submanifold and $f\in C^\infty(Z)$. Let $\Lambda=N^*_fZ\subset T^*X$ be the twisted conormal bundle. Then the intrinsic line bundle of symbols $$\mathbb{L}\rightarrow\Lambda$$ is canonically isomorphic to the $\frac{1}{2}$-density line bundle $\left|T\Lambda\right|^{\frac{1}{2}}$ of $\Lambda$.
We will further show that this isomorphism is functorial under the composition of canonical relations.
Under the assumptions of Theorem 2, we have a clean composition of (exact) canonical relations $$\Gamma_2\circ\Gamma_1=\Gamma$$ where $\Gamma_i=\Gamma_{Z_i, f_i}\in Morph(T^*X_i, T^*X_{i+1})$ for $i=1, 2 $ and $\Gamma=\Gamma_{Z, f}\in Morph(T^*X_1, T^*X_3)$. In this case the intrinsic line bundles of symbols $$\mathbb{L}_1\rightarrow\Gamma_1, \mathbb{L}_2\rightarrow\Gamma_2, \mathbb{L}\rightarrow\Gamma$$ can be identified with the $\frac{1}{2}$-density bundles $\left|T\Gamma_1\right|^{\frac{1}{2}}$, $\left|T\Gamma_2\right|^{\frac{1}{2}}$ and $\left|T\Gamma\right|^{\frac{1}{2}}$.
If $F_1\in\mathcal{F}^{m_1}(\Gamma_1)$ and $F_2\in\mathcal{F}^{m_2}(\Gamma_2)$ (so that $F_2\circ F_1\in\mathcal{F}^{m_1+m_2}(\Gamma)$), then the symbol $$\sigma(F_2\circ F_1)\in C^\infty(\left|T\Gamma\right|^{\frac{1}{2}})$$ is the composition of $\sigma(F_2)\in C^\infty(\left|T\Gamma_2\right|^{\frac{1}{2}})$ with $\sigma(F_1)\in C^\infty(\left|T\Gamma_1\right|^{\frac{1}{2}})$ in the enhanced symplectic “category”.
[**Acknowledgements:** I am grateful to Prof. Victor Guillemin giving me the topic. I am also grateful to the help from Jingwen Chen and Nick Strehlke during the research. ]{}
Twisted conormal bundles as canonical relations
===============================================
In \[2\], the authors defined the symplectic ”category”, which has symplectic manifolds as objects and morphisms $$Morph(M_1, M_2)=\left\{ \text{Lagrangians submanifolds of } M_1^-\times M_2 \right\}.$$The morphisms are also called canonical relations. For $\Gamma_1\in Morph(M_1, M_2)$ and $\Gamma_2\in Morph(M_2, M_3)$, we define their composition $$\Gamma_2\circ\Gamma_1=\left\{(x, z)\in M_1\times M_3: \exists y\in M_2, s.t. (x, y)\in \Gamma_1, (y, z)\in \Gamma_2\right\}.$$ Note that this is not necessarily an Lagrangian submanifold of $M_1^-\times M_3$, so $\Gamma_1$ and $\Gamma_2$ may not compose, and the symplectic ”category” is not strictly a category. Several conditions are imposed in \[2\] to make two morphisms compose. Define$$\Gamma_2\star\Gamma_1=\left\{ (m_1, m_2, m_3)\in M_1\times M_2\times M_3: (m_1, m_2)\in\Gamma_1, (m_2, m_3)\in\Gamma_2\right\}.$$We call $\Gamma_1, \Gamma_2$ cleanly composable if $\Gamma_2\star\Gamma_1$ is a submanifold; the natural projection $\kappa: \Gamma_2\star\Gamma_1\rightarrow\Gamma_2\circ\Gamma_1$ is a smooth fiber bundle with compact fibers; and for every point $(m_1, m_2, m_3)\in \Gamma_2\star\Gamma_1$, the tangent space $T_{(m_1, m_2, m_3)}\Gamma_2\star\Gamma_1$ equals $$\left\{(v_1, v_2, v_3)\in T_{m_1}M_1\times T_{m_2}M_2\times T_{m_3}M_3: (v_1, v_2)\in T_{(m_1, m_2)}\Gamma_1, (v_2, v_3)\in T_{(m_2, m_3)}\Gamma_2\right\}.$$ In this case $\Gamma_2\circ\Gamma_1$ is an (embedded) Lagrangian submanifold of $M_1\times M_3$, and the canonical relations $\Gamma_1$, $\Gamma_2$ are defined to be composable with composition $\Gamma_2\circ\Gamma_1$.
Before delving into the proof of Theorem 1 and 2, we first prove the following result given by Jingwen \[3\], which says that under certain regularity conditions, any Lagrangian submanifold of $T^*X$ is locally a twisted conormal bundle.
Let $X$ be a smooth manifold and $T^*X$ be its cotangent bundle. Denote the canonical projection by $\pi:T^*X\rightarrow X$. Suppose $\Lambda$ is a Lagrangian submanifold of $T^*X$ such that $\pi(\Lambda)=Z$ is a simply connected submanifold of $X$, and that the projection $\pi\arrowvert_\Lambda: \Lambda\rightarrow Z$ is a surjective submersion. Then there is a smooth function $f$ on $Z$ such that $\Lambda$ is an open submanifold of $N_f^*Z$.
Suppose $\alpha$ is the canonical 1-form on $T^*X$ and $\iota: \Lambda\hookrightarrow T^*X$ is the inclusion map, then $\iota^*\alpha$ is closed on $\Lambda$. Furthermore, $\iota^*\alpha$ is a horizontal form since, for any $p=(x, \xi)\in\Lambda$ and vertical vetor $v\in T_p\Lambda$ (i.e. $(d\pi)_pv=0$), we have $$\iota^*\alpha_p(v)=\iota^*(d\pi)^*_p\xi(v)=\iota^*(\xi, (d\pi)_pv)=0.$$ As a consequence, there exists a 1-form $\beta$ on $Z$ such that $\pi^*\beta=\iota^*\alpha$. Since $Z$ is simply connected, we can write $\beta=d\phi_0$ for some $\phi_0\in C^\infty(Z)$.
Denote an extension of $\phi_0$ to $X$ by $\phi$, and let $$\gamma_\phi: T^*X\rightarrow T^*X, (x, \xi)\mapsto(x, \xi+d\phi_x)$$ be the symplectomorphism with the property $\pi\circ\gamma_\phi=\pi$. By constructions we have $$((\gamma_\phi)^{-1})^*\alpha=((\gamma_\phi)^{-1})^*d\pi^*\xi=d\pi^*(\xi-d\phi)=\alpha-\pi^*d\phi$$ and$$\iota^*\alpha=\pi^*\beta=\pi^*d\phi_0=\iota^*\pi^*d\phi.$$ Thus $$\iota^*((\gamma_\phi)^{-1})^*\alpha=\iota^*(\alpha-\pi^*d\phi)=\iota^*\pi^*d\phi-\iota^*\pi^*d\phi=0,$$ which means that the restriction of $\alpha$ to the image $\Lambda^\prime=\gamma_\phi(\Lambda)$ is zero. Note that $\Lambda^\prime$ is also a Lagrangian submanifold of $T^*X$ with $\pi(\Lambda^\prime)=Z$ and each fiber $\Lambda^\prime\cap T^*_xX$ is a connected submanifold of $T^*X$.
For any $p=(x, \xi)\in\Lambda^\prime$, we have $\alpha_p=(d\pi)^*_p\xi$ by definition. Let $\iota_Z: Z\hookrightarrow X$ be the inclusion map, then we have $\iota_Z\circ\pi=\pi\circ\iota$. It follows that $$0=(\iota^*\alpha)_p=\iota^*(d\pi_p)^*\xi=(d\pi_p)^*\iota_Z^*\xi$$ Since $(d\pi_p)^*: T^*_xZ\rightarrow T^*_p\Lambda$ is injective, we get $\iota_Z^*\xi=0$, i.e. $\xi\in N^*_xZ$. We have proved that $\Lambda^\prime\subset N^*Z$. Thus $\Lambda=(\gamma_\phi)^{-1}\Lambda^\prime\subset N^*_{-\phi}Z$.
Note that in the step of finding 1-form $\beta$, we have implicitly used the $k=1$ case of the result from \[4\]: For a surjective submersion $\pi: M\rightarrow N$ between smooth manifolds such that each fiber is connected, a tangent vector $v\in T_pM$ is called vertical if $d\pi_p(v)=0$. Suppose $w\in\Omega^k(M)$, then there exists $\eta\in\Omega^k(N)$ such that $w=\pi^*\eta$ if and only if $i_vw_p=0$ and $i_vdw_p=0$ for all $p\in M$ and vertical vector $v\in T_pM$.
Now we begin the proof of Theorem 2. Calculate $$N^*Z=\left\{(x, y, \xi, \eta): (x, \xi)\in T^*X, (y, \eta)\in T^*Y, (\xi, \eta) \text{ vanishes on } T_{(x, y)}Z \right\}$$ and$$N^*_fZ=\left\{(x, y, \xi, \eta): (x, \xi)\in T^*X, (y, \eta)\in T^*Y, (\xi-d_Xf, \eta-d_Yf) \text{ vanishes on } T_{(x, y)}Z \right\};$$ so $$\Gamma_{Z, f}=\left\{(x, y, \xi, \eta): (x, \xi)\in T^*X, (y, \eta)\in T^*Y, (-\xi-d_Xf, \eta-d_Yf) \text{ vanishes on } T_{(x, y)}Z \right\}.$$
In the context of Theorem 2, we have $$\begin{aligned}
\begin{autobreak}
\Gamma_2\circ\Gamma_1=
\left \{(x_1, x_3, \xi_1, \xi_3): \exists(x_2, \xi_2)\in T^*X_2, s.t. (x_1, x_2)\in Z_1, (x_2, x_3)\in Z_2, (-\xi_1-\right .
\left .d_1f_1, \xi_2-d_2f_1) \text{ vanishes on } T_{(x_1, x_2)}Z_1, (-\xi_2-d_2f_2,\xi_3-d_3f_2) \text{ vanishes on } \right .
\left .T_{(x_2, x_3)}Z_2 \right \}
\end{autobreak}\end{aligned}$$ and $$\begin{aligned}
\begin{autobreak}
(\zeta_1\times id)(\Gamma_2\circ\Gamma_1)=
\left \{(x_1, x_3, \xi_1, \xi_3): \exists(x_2, \xi_2)\in T^*X_2, s.t. (x_1, x_2)\in Z_1, (x_2, x_3)\in Z_2,\right .
\left . (\xi_1-d_1f_1, \xi_2-d_2f_1) \text{ vanishes on } T_{(x_1, x_2)}Z_1, (-\xi_2-d_2f_2,\xi_3-\right .
\left .d_3f_2) \text{ vanishes on } T_{(x_2, x_3)}Z_2 \right \}.
\end{autobreak}\end{aligned}$$
Notice that $\Lambda=(\zeta_1\times id)(\Gamma_2\circ\Gamma_1)$ is a Lagrangian submanifold of $T(X_1\times X_3)$ whose projection on $X_1\times X_3$ is inside $$\left\{(x_1,x_3)\in X_1\times X_3 : \exists x_2\in X_2, s.t. (x_1, x_2)\in Z_1, (x_2, x_3)\in Z_2 \right\}=Z.$$ Conversely, under the condition $d_2f_1+d_2f_2=0$, we have at least one point $$(x_1, x_3, d_1f_1, d_3f_2)\in\Lambda\cap T_{(x_1, x_3)}(X_1\times X_3)$$ in the fiber of $\pi\arrowvert_{\Lambda}$ on each $(x_1, x_3)\in Z$. So $\pi\arrowvert_{\Lambda}$ is a surjective submersion onto Z. We further notice that all its fibers are connected since they are unions of some affine subspaces of $T_{(x_1, x_3)}(X_1\times X_3)$ containing the same point $(d_1f_1, d_3f_2)$. Since $Z$ is simply connected, from Lemma 1 we know that there exists $f\in C^\infty(Z)$ such that $\Lambda$ is an open submanifold of $N^*_{f}Z$. Furthermore, since $N^*_{f}Z$ has all fibers affine subspaces and $\Lambda$ has all fibers unions of affine subspaces containing the same point, we have $$\Lambda=N^*_{f}Z\subset T^*X_1\times T^*X_3$$ thus $$\Gamma_2\circ\Gamma_1=(\zeta_1\times id)N^*_{f}Z\subset T^*X_1^-\times T^*X_3.$$ Theorem 2 has been proved. In the case when $Z_1$ and $Z_2$ are not twisted we have each fiber $$\begin{aligned}
\begin{autobreak}
\Lambda\cap T^*_{(x_1, x_3)}(X_1\times X_3)=
\left \{ (\xi_1, \xi_3): \exists(x_2, \xi_2)\in T^*X_2, s.t. (x_1, x_2)\in Z_1, (x_2, x_3)\in Z_2, (\xi_1,\right .
\left . \xi_2)\text{ vanishes on }T_{(x_1, x_2)}Z_1, (-\xi_2,\xi_3) \right .
\left . \text{ vanishes on } T_{(x_2, x_3)}Z_2 \right \}
\end{autobreak}\end{aligned}$$ a union of linear subspaces of $T^*_{(x_1, x_3)}(X_1\times X_3)$. So every fiber of $\Lambda=N^*_{f}Z$ contains the origin and thus $N^*_{f}Z=N^*Z$, i.e. the composition is not twisted. The simply connected condition of $Z$ can be removed in this case because $Z$ as a submanifold is always locally simply connected. From our proof, $$\Gamma_{Z_2}\circ\Gamma_{Z_1}=\Gamma_{Z}$$ is true locally near any point of $Z$, which means that it is globally true. Thus Theorem 1 is also proved.
For smooth manifolds $X_1, X_2, X_3$, pick a point $*\in X_2$ and let $Z_1=X_1\times \left\{ * \right\}$ and $Z_2=\left\{ * \right\}\times X_3$ in Theorem 2. Suppose $f_1\in C^\infty(X_1)$ and $f_2\in C^\infty(X_3)$ viewed as functions on $Z_1$ and $Z_2$, then we can easily see$$\Gamma_{Z_2, f_2}\circ\Gamma_{Z_1, f_1}=\Gamma_{Z, f}$$for $Z=X_1\times X_3$ and $f=f_1+f_2\in C^\infty(Z).$
Consider a trivial type of twisted conormal bundles $$\Lambda_f=\left\{(x, df(x)): x\in X\right\}\subset T^*X$$ for $f \in C^\infty(X)$. For smooth manifolds $X_1, X_2, X_3$ and $f_1\in C^\infty (X_1\times X_2)$, $f_2\in C^\infty (X_2 \times X_3)$, we have canonical relations $\Gamma_{f_1}=(\zeta_1 \times id) \Lambda_{f_1}$ and $\Gamma_{f_2}=(\zeta_2 \times id) \Lambda_{f_2}$. Suppose $$d_2f_1(x_1, x_2)+d_2f_2(x_2, x_3)=0\in T^*X_2$$ for every $(x_1, x_2, x_3)\in X_1\times X_2\times X_3$. Note that this condition means $f_1(x_1, x_2)+f_2(x_2, x_3)$ is constant in the variable $x_2$ and we can write it $f(x_1, x_3)$. In this case we have $$\Gamma_{f_2}\circ\Gamma_{f_1}=\Gamma_{f}$$where $\Gamma_f=(\zeta_1 \times id) \Lambda_{f}$.
Consider twisted conormal bundles of graphs $$\Gamma_{g, f}: =(\zeta_X \times id)N^*_f (\text{graph}(g))\subset T^*X^-\times T^*Y$$ as canonical relations, where $g: X\rightarrow Y$ is a smooth map and $f\in C^\infty(X)$ deemed as a function on $\text{graph}(g)$. We can get a smaller subcategory (without quotation marks) of the symplectic “category” from this. Specifically, suppose $X_1, X_2, X_3$ are smooth manifolds, $g_1: X_1\rightarrow X_2$, $g_2: X_2\rightarrow X_3$ are maps and $f_1\in C^\infty(X_1)$, $f_2\in C^\infty(X_2)$. Then $\Gamma_{g_2, f_2}$ and $\Gamma_{g_1, f_1}$ are always composable, and we have $$\Gamma_{g_2, f_2}\circ\Gamma_{g_1, f_1}=\Gamma_{g, f}$$ where $g=g_2\circ g_1: X_1\rightarrow X_3$ and $f=f_1+f_2\circ g_2\in C^\infty(X_1)$.
In the composition $$\Gamma_{Z_2, f_2}\circ\Gamma_{Z_1, f_1}=\Gamma_{Z, f}$$ from Theorem 2, the submanifold $Z\subset X_1\times X_3$ is given explicitly by $$Z=\left\{ (x_1,x_3)\in X_1\times X_3 : \exists x_2\in X_2, s.t. (x_1, x_2)\in Z_1, (x_2, x_3)\in Z_2 \right\},$$ but the function $f\in C^\infty(Z)$ can be somewhat mysterious. From Example 2.3, we know that $f$ cannot be expressed explicitly by a calculation of $f_1$ and $f_2$, since it also depends on $g_1$ and $g_2$, which are datas representing the “shapes of submanifolds” $Z_1$ and $Z_2$.
Hörmander descriptions at regular points
========================================
Suppose $\pi: W\rightarrow X$ is a smooth fiber bundle and $\phi$ is a smooth function on $W$. View the Lagrangian submanifold $$\Lambda_{\phi}=\left\{ (z, d\phi(z)): z\in W \right\}$$ of $T^*W$ as a canonical relation in $Morph(pt, T^*W)$ and the graph $$\Gamma_{\pi}=\left\{ (z, \xi, x, \eta)\in T^*W\times T^*X: x=\pi(z), \xi=(d\pi_z)^*\eta \right\}$$ of $\pi: W\rightarrow X$ as a canonical relation in $Morph(T^*W, T^*X)$. According to \[2\], the assumption that $\Gamma_\pi$ and $\Lambda_\phi$ are transversally composable is equivalent to $$C_\phi :=\left\{ z\in W: (d_{vert}\phi)_z=0\right\}$$ being a submanifold of $W$ and $$C_\phi\rightarrow T^*X, z\mapsto(\pi(z), \eta)$$ being a Lagrangian embedding, where $\eta$ is the unique vector in $T^*_{\pi(z)}X$ such that $d\phi_z=(d\pi_z)^*\eta$. The image of this embedding is the Lagrangian submanifold $\Lambda=\Gamma_\pi(\Lambda_\phi)=\Gamma_\pi\circ\Lambda_\phi$ of $T^*X$.
When this happens we say $\phi$ is a (transverse) generating function of $\Lambda$ with respect to the fiber bundle $(W,\pi)$, and the data $(W, \pi, \phi)$ is a Hörmander description of the Lagrangian submanifold $\Lambda\subset T^*X$. There is a deep result called Hörmander’s Theorem, proved in \[2\], saying that every Lagrangian submanifold of $T^*X$ has a Hörmander description locally. However, the construction in this theroem can be somewhat abstract, so it would be nice to actually find the Hörmander description assuming that $\Lambda\subset T^* X$ is locally regular.
Suppose $\Lambda\subset T^*X$ is regular at $p\in\Lambda$ in the sense of Theorem 3. Then there is a neighborhood $U$ of $p$ such that $\pi\arrowvert_{U\cap\Lambda}$ is a submersion onto a $(n-k)$-dimensional submanifold $Z\cap\pi(U)\subset X$ with connected fibers, where $\pi: T^*X\rightarrow X$ is the canonical projection. By restricting to a smaller neighborhood one can assume further that $Z\cap\pi(U)$ is simply connected and $$Z=\left\{x\in X: u_1(x)=u_2(x)=\dots=u_k(x)=0 \right\}$$ where $u_1, \dots u_k\in C^{\infty}(X)$ are functionally independent near $p$. From Lemma 1 we know that $U\cap\Lambda=U\cap N^*_f Z$ for some $f\in C^{\infty}(Z)$. We show that the fiber bundle $W=X\times \mathbb{R}^{k}\rightarrow X$ and the generating function $$\phi=\sum_{i=1}^{k} s_iu_i+f \in C^{\infty}(W)$$ is a Hörmander description for $N^*_f Z$, thus for $\Lambda$ locally at $p$, where the $s_i$ for $i=1,2, \dots k$ are coordinates for $\mathbb{R}^k$. In fact in this case $$\begin{split}
C_\phi :&=\left\{ z\in X\times \mathbb{R}^k: (d_{vert}\phi)_z=0\right\}\\
&=\left\{ z\in X\times\mathbb{R}^k: \frac{\partial\phi}{\partial s_i}=0, i=1,2, \dots k\right\}\\
&=\left\{ z\in X\times\mathbb{R}^k: u_i=0, i=1,2, \dots k\right\}\\
&=Z\times \mathbb{R}^k.
\end{split}$$ The transversality assumption amounts to saying that the equations $\frac{\partial\phi}{\partial s_i}=u_i=0, i=1,2, \dots k$ are functionally independent, and the Lagrangian submanifold of $T^*X$ determined by $W=X\times \mathbb{R}^{k}\rightarrow X$ with $\phi: W\rightarrow\mathbb{R}$ is the image of the embedding $$C_\phi \rightarrow T^*X, (x, s)\mapsto(x, \frac{\partial\phi}{\partial x})=(x, \sum\limits_{i=1}^{k} s_i d_Xu_i+d_Xf)$$ Since the differentials $d_Xu_i$ span the conormal bundle of $Z$ in $X$, the image of this embedding is $N^*_f Z\subset T^*X$. So we have proved Theorem 3.
The quantization process
========================
In semi-classical analysis we quantize canonical relations by associating semi-classical Fourier integral operators to them. In this section we will get simple formulas of these operators when the canonical relation takes the form of a twisted conormal bundle in a phase space.
Recall that as in \[2\], for every exact Lagrangian submanifold $(\Lambda , \psi)$ of $T^*X$ and $r\in\mathbb{Z}$ we can associate them to a space $I^r(X,\Lambda)$ of rapidly oscillating $\frac{1}{2}$-densities on $X$. In the special case when $\Lambda\subset T^*X$ admits a global Hörmander description $(W, \pi, \phi)$, we fix the arbitrary constant in $\phi$ so that $$\psi(x, \xi)=\phi(z) \text{ if } d\phi_z=\pi^*_z\xi\text{ where }\pi(z)=x$$ and pick an enhancement $\sigma$ of $\pi$. We define the space of compactly supported $\frac{1}{2}$-densities on $X$ to be $$I_0^r(X, \Lambda)=\left\{\mu=\hbar^{r-\frac{k}{2}}\pi_*(ae^{i\frac{\phi}{\hbar}}\tau)\arrowvert a=a(z, \hbar)\in C_0^\infty(W\times\mathbb{R})\right\}$$ where $\tau$ is a nowhere vanishing $\frac{1}{2}$-density on $W$. We can check that the space $I_0^r(X, \Lambda)$ is independent of the Hörmander description $(W, \pi, \phi)$, the enhancement $\sigma$ of $\pi$ and the $\frac{1}{2}$-density $\tau$. In the general case when $(\Lambda, \psi)$ is an exact Lagrangian submanifold of $T^*X$, we can find a locally finite cover of $\Lambda$ by open sets $\Lambda_i$ such that each $\Lambda_i$ is defined by a generating function $\phi_i$ relative to a fiber bundle $\pi_i: W_i\rightarrow U_i$ where the $U_i$ are open subsets of $X$. Define $I^r_0(X,\Lambda)$ to be the space of $\frac{1}{2}$-densities $$\mu=\sum_{j=1}^N\mu_{i_j},\text{ } \mu_{i_j}\in I_0^r(U, \Lambda_{i_j})$$and $I^r(X, \Lambda)$ to consist of those $\frac{1}{2}$-densities $\mu$ on $X$ such that $\rho\mu\in I^r_0(X, \Lambda)$ for every $\rho\in C_0^\infty(X)$. It is easy to see that these definitions are independent of the choice of open cover and the local Hörmander descriptions.
Suppose $X_1, X_2$ are manifolds and $(\Gamma, \Psi)$ is an exact canonical relation from $T^*X_1$ to $T^*X_2$, then by the following definition we can associate $\Gamma$ to a space of semi-classical Fourier integral operators $\mathcal{F}_0^m(\Gamma)$. Let $X=X_1\times X_2$ and $$\zeta_1: T^*X_1\rightarrow T^*X_1^-, \text{ }\zeta_1(x_1, \xi_1)=(x_1, -\xi_1),$$ then $\Lambda=(\zeta_1\times id)(\Gamma)$ and $\psi=\Psi\circ(\zeta_1\times id)$ give an exact Lagrangian submanifold $(\Lambda, \psi)$ of $$T^*X=T^*X_1\times T^*X_2.$$ Associated with $(\Lambda, \psi)$ we have the space of compactly supported oscillatory $\frac{1}{2}$-densities $I_0^r(X, \Lambda)$. We can write a typical element $\mu\in I^r_0(X, \Lambda)$ as $$\mu=u(x_1, x_2, \hbar)dx_1^{\frac{1}{2}}dx_2^{\frac{1}{2}}$$where $dx_i^{\frac{1}{2}}$ is a nowhere vanishing $\frac{1}{2}$-density on $X_i$ and $u$ is a smooth function of compact support in all three “variables”. Define the Fourier integral operator $$F_{\mu, \hbar}: L^2(X_1)\rightarrow L^2(X_2), gdx_1^{\frac{1}{2}}\mapsto(\int_{X_1}g(x_1)u(x_1, x_2, \hbar)dx_1)dx_2^{\frac{1}{2}}$$ and denote the space of such operators by $\mathcal{F}_0^m(\Gamma)$ where $m=r+\frac{n_2}{2}$ and $n_2=$dim$X_2$.
For the rest of this section we focus on the case when $\Lambda=(\zeta_1\times id)(\Gamma)$ is a twisted conormal bundle. Suppose $X$ is a smooth manifold with a nowhere vanishing $\frac{1}{2}$-density $dx^\frac{1}{2}$ and $u_1, \dots u_k$ are functionally independent functions on $X$. Then$$Z=\left\{x\in X: u_1(x)=u_2(x)=\dots=u_k(x)=0 \right\}$$ is a submanifold of $X$. From section 3, $\Lambda=N^*_fZ$ has a Hörmander description of a fiber bundle $W=X\times \mathbb{R}^{k}\rightarrow X$ and a generating function $$\phi=\sum_{i=1}^{k} s_iu_i+f \in C^{\infty}(W).$$ In this case the space of compactly supported oscillating $\frac{1}{2}$-densities is $$\begin{split}
I_0^r(X, \Lambda)&=\left\{\mu=\hbar^{r-\frac{k}{2}}(\int_{\mathbb{R}^k}a(x, s, \hbar)e^{i\frac{\phi}{\hbar}}ds)dx^\frac{1}{2}: a(x, s, \hbar)\in C_0^\infty(X\times\mathbb{R}^k\times\mathbb{R})\right\}\\
&=\left\{\mu=\hbar^{r-\frac{k}{2}}e^f(\int_{\mathbb{R}^k}a(x, s, \hbar)e^{i\frac{\sum_{i=1}^{k} s_iu_i}{\hbar}}ds)dx^\frac{1}{2}: a(x, s, \hbar)\in C_0^\infty(X\times\mathbb{R}^k\times\mathbb{R})\right\}
\end{split}$$ for every $r\in\mathbb{Z}$. Note that in the special case when $a(x, s, \hbar)=a(x, \hbar)$ and $u_i(x)=x_i$, this integration is just the Fourier transform of $a(x, \hbar)$ in the variables $x_1, \dots x_k$.
Supppose $X=X_1\times X_2$ and $dx^\frac{1}{2}=dx_1^\frac{1}{2}dx_2^\frac{1}{2}$, where $dx_1^\frac{1}{2}$ and $dx_2^\frac{1}{2}$ are nowhere vanishing $\frac{1}{2}$-densities on $X_1$ and $X_2$. Then the space $\mathcal{F}_0^m(\Gamma)$ of compactly supported semi-classical Fourier integral operators associated to the canonical relation $$\Gamma=\Gamma_{Z, f}=(\zeta_1 \times id)N^*_{f}Z \in Morph(T^*X_1, T^*X_2)$$has elements $$F_{\mu, \hbar}: L^2(X_1)\rightarrow L^2(X_2), gdx_1^\frac{1}{2}\mapsto(\int_{X_1}g(x_1)u(x_1, x_2, \hbar)dx_1)dx_2^\frac{1}{2}$$for every $$\mu=u(x_1, x_2, \hbar)dx_1^\frac{1}{2}dx_2^\frac{1}{2}\in I_0^r(X, \Lambda),$$which in this case is $$F_{\mu, \hbar}(gdx_1^\frac{1}{2})=\hbar^{r-\frac{k}{2}}(\int_{X_1}g(x_1)e^\frac{f}{\hbar} dx_1\int_{\mathbb{R}^k}a(x_1, x_2, s, \hbar)e^{i\frac{\sum_{i=1}^{k} s_iu_i}{\hbar}}ds)dx_2^\frac{1}{2}$$for every $a(x_1, x_2, s, \hbar)\in C_0^\infty(X_1\times X_2\times \mathbb{R}^k\times\mathbb{R})$, where $m=r+\frac{n_2}{2}, n_2=\text{dim}X_2$. Note that in the special case when $a(x_1, x_2, s, \hbar)=a(x_1, x_2, \hbar)$ does not depend on the $s$-variables and $u_i$ are coordinate functions, the inner integraion is just the Fourier transform of some $x_i$-variables.
Next we talk about compositions of these operators. Suppose $$(X_1, dx_1^\frac{1}{2}), (X_2, dx_2^\frac{1}{2}), (X_3, dx_3^\frac{1}{2})$$ are smooth manifolds with nowhere vanishing $\frac{1}{2}$-densities on them, $$Z_1=\left\{u_1=\dots=u_{k_1}=0\right\}\subset X_1\times X_2 \text{ and } Z_2=\left\{v_1=\dots v_{k_2}=0 \right\}\subset X_2\times X_3$$ are submanifolds and $f_1\in C^\infty(Z_1)$ and $f_2\in C^\infty(Z_2)$. Denote the canonical relations by $$\Gamma_1=\Gamma_{Z_1, f_1}\in Morph(T^*X_1, T^*X_2) \text{ and }\Gamma_2=\Gamma_{Z_2, f_2}\in Morph(T^*X_2, T^*X_3).$$Suppose $$\Gamma_2\circ\Gamma_1=\Gamma\in Morph(T^*X_1, T^*X_3)$$ is a clean composition. According to a theorem in \[2\], the composition of Fourier integral operators in $\mathcal{F}_0^{m_2}(\Gamma_2)$ with operators in $\mathcal{F}_0^{m_1}(\Gamma_1)$ are the operators in $\mathcal{F}_0^{m_1+m_2-\frac{e}{2}}(\Gamma)$, where $e$ is the fiber dimension of projection $\kappa: \Gamma_1\star\Gamma_2\rightarrow\Gamma_1\circ\Gamma_2$. Since the Hörmander descriptions of $N^*_{f_1}Z_1$ and $N^*_{f_2}Z_2$ can induce a Hörmander description of $(\zeta_1\times id)\Gamma$, it is thus easily computable that the compositions of $\mathcal{F}_0^{m_2}(\Gamma_2)$ with $\mathcal{F}_0^{m_1}(\Gamma_1)$ are of the form $$F_{\mu, \hbar}: L^2(X_1)\rightarrow L^2(X_3),$$ $$F_{\mu, \hbar}(gdx_1^\frac{1}{2})=\hbar^{r-\frac{k}{2}}(\int_{X_1\times X_2\times\mathbb{R}^{k_1}\times\mathbb{R}^{k_2}}g(x_1)a(x_1, x_2, x_3, s, t, \hbar)e^{i\frac{\sum_{i=0}^{k_1} s_iu_i+\sum_{j=0}^{k_2} t_jv_j+f_1+f_2}{\hbar}}dx_1dx_2dsdt)dx_2^\frac{1}{2}$$ for some $a(x_1, x_2, x_3, s, t, \hbar)\in C^\infty(X_1\times X_2\times X_3\times\mathbb{R}^{k_1}\times\mathbb{R}^{k_2}\times\mathbb{R})$, where $$m_1+m_2-\frac{e}{2}=r+\frac{n_3}{2}\text{ and } k=n_2+k_1+k_2.$$
The intrinsic line bundle of symbols
====================================
Recall that for any Lagrangian submanifold $\Lambda\subset T^*X$, we can associate the Maslov line bundle to it. Specifically, suppose $\Lambda$ has a Hörmander description $(W, \pi, \phi)$. Let $z$ be any point in the critical set $C_{\phi}\subset W$ and $x=\pi(z)$, then $z$ must be the critical point of $\phi\arrowvert_{\pi^{-1}(x)}$. Denote the signature of $\phi$ at $z\in\pi^{-1}(x)$ to be $\text{sgn}^\sharp(z)$, which gives an integer valued function $\text{sgn}^\sharp$ on $C_{\phi}$. By pulling back the function $\text{sgn}^\sharp$ via the diffeomorphism $$\lambda_{\phi}: C_{\phi}\rightarrow\Lambda$$we get a $\mathbb{Z}$-valued function $\text{sgn}_\phi$ on $\Lambda$. Let $$s_\phi: \Lambda\rightarrow\mathbb{C}^*, s_\phi=e^{\frac{\pi i}{4}\text{sgn}_\phi}$$and define the Maslov bundle $\mathbb{L}_{Maslov}\rightarrow\Lambda$ to be the trivial flat bundle having $s_\phi$ as a flat section.
Suppose we have two Hörmander descriptions $(W_i, \pi_i, \phi_i)$ of $\Lambda$, i=1,2. By the Hörmander-Morse Lemma in \[2\], each two Hörmander descriptions of $\Lambda$ are related by the three types of moves. The functions $s_{\phi_1}$ and $s_{\phi_2}$ are equal in the type 1) and 2) move, and in the type 3) move the functions $\text{sgn}_1^\sharp$ and $\text{sgn}_2^\sharp$ are related by $$\text{sgn}_1^\sharp=\text{sgn}_2^\sharp+\text{signature of A},$$thus $s_{\phi_1}=s_{\phi_2}e^{\frac{\pi i}{4}\text{signature of A}}$, i.e. $s_{\phi_1}$ and $s_{\phi_2}$ differ by multiple of a constant of norm 1. Thus our definition of Maslov bundle is intrinsic.
For a general Lagrangian submanifold $\Lambda\subset T^*X$ we can cover $\Lambda$ by open sets $U_i$ admitting generating functions $\phi_i$. By the provious arguments we get functions $s_{\phi_i}: U_i\rightarrow\mathbb{C}$ such that each pair $s_{\phi_i}$ and $s_{\phi_j}$ differ by multiple of a constant of norm 1 on $U_i\cap U_j$. Thus we can patch these local definitions together to get the globally defined Maslov bundle $$\mathbb{L}_{Maslov}\rightarrow\Lambda.$$
According to Hörmander \[1\], we construct the intrinsic line bundle of symbols $$\mathbb{L}=\mathbb{L}_{Maslov}\otimes\left|T\Lambda\right|^{\frac{1}{2}}$$ for any exact Lagrangian submanifold $\Lambda=(\Lambda, \psi)\subset T^*X$. Suppose $\Lambda$ has a Hörmander description $(W, \pi, \phi)$ and we pick a nonwhere vanishing section of the vertical subbundle $\left|V\right|^{\frac{1}{2}}$ of the fiber bundle $\pi: W\rightarrow X$, which can be deemed as an enhancement of $\pi$. This enhancement does to things: it both gives us a nowhere vanishing $\frac{1}{2}$-density $\rho_\pi$ on the canonical relation $\Gamma_\pi$ and a fiber integration $$\pi_*: C_0^\infty(\left|TW\right|^{\frac{1}{2}})\rightarrow C_0^\infty(\left|TX\right|^{\frac{1}{2}}).$$By the definition in section 3, $$\Lambda=\Gamma_\pi\circ\Lambda_\phi$$ where $\Lambda_\phi$ is the Lagrangian submanifold $\left\{ (p, d\phi_p): p\in Z \right\}$ of $T^*Z$. Pick a $\frac{1}{2}$-density $\tau$ on $W$ and write a typical element in $I_0^r(X, \Lambda)$ as $$\mu=\hbar^{r-\frac{k}{2}}\pi_*\nu, \text{ }\nu=ae^{i\frac{\phi}{\hbar}}\tau\in I_0^0(W, \Lambda_\phi)$$where $a=a(z, \hbar)\in C_0^\infty(W\times\mathbb{R})$. Denote by $\mathcal{P}$ the projection of $\Lambda_\phi$ to $W$. We define the “symbol” of $\nu$ to be $\frac{1}{2}$-density $$\sigma(\nu)=\mathcal{P}^*(a(z, 0)\tau))$$on $\Lambda$ and the symbol of $\mu$ to be $$\sigma(\mu)=s_\phi\otimes(\rho_\pi\circ\sigma(\nu))\in C^\infty(\mathbb{L})$$where $s_\phi$ is the section of $\mathbb{L}_{Maslov}$ associated with $\phi$ and $\rho_\pi\circ\sigma(\nu)\in C^\infty(\left|T\Lambda\right|^{\frac{1}{2}} )$ is a composition in the enhanced symplectic “category”. Call $$\sigma: I_0^r(X, \Lambda)\rightarrow C^\infty(\mathbb{L})$$ the symbol map.
For a general exact Lagrangian submanifold $\Lambda\subset T^*X$ one can add together the symbol map on local Hörmander pieces to get the global symbol map $$\sigma: I^r(X, \Lambda)\rightarrow C^\infty(\mathbb{L})$$which is proved to be intrinsically defined and surjective with kernel $I^{k+1}(X, \Lambda)$. Hence $$I^k(X, \Lambda)/I^{k+1}(X, \Lambda)\cong\ C^\infty(\mathbb{L}).$$
For an exact canonical relation $(\Gamma, \Psi)\in Morph(T^*X_1, T^*X_2)$ we have the corresponding exact Lagrangian submanifold $(\Lambda, \psi)\subset T^*X_1\times T^*X_2$ where $$\Lambda=(\zeta_1\times id)(\Gamma),\text{ }\psi=\Psi\circ(\zeta_1\times id).$$Define the intrinsic line bundle of symbols $\mathbb{L}_\Gamma\rightarrow\Gamma$ to be the pullback of the line bundle $\mathbb{L}_\Lambda\rightarrow\Lambda$ via the diffeomorphism $\zeta_1\times id$, $$\mathbb{L}_\Gamma=(\zeta_1\times id)^*\mathbb{L}\arrowvert_\Lambda=\mathbb{L}_{Maslov}(\Gamma)\otimes\left|T\Gamma\right|^{\frac{1}{2}}.$$
For a semi-classical Fourier integral operator $F\in\mathcal{F}_0^m(\Gamma, \Psi)$ given by a oscillatory $\frac{1}{2}$-density $\mu\in I_0^r(X, \Lambda)$, we define its symbol to be $$\sigma(F)=(\zeta_1\times id)^*\sigma(\mu)\in C^\infty(\mathbb{L}_\Gamma).$$ Suppose $\Gamma_2$ and $\Gamma_1$ are cleanly composable exact canonical relations and $\Gamma=\Gamma_2\circ\Gamma_1$. Note that the Maslov bundle is functorial under the composition: $$\kappa^*\mathbb{L}_{Maslov}(\Gamma_2\circ\Gamma_1)\cong \text{pr}_1^*\mathbb{L}_{Maslov}(\Gamma_1)\otimes \text{pr}_2^*\mathbb{L}_{Maslov}(\Gamma_2)$$ where $\kappa: \Gamma_2\star\Gamma_1\rightarrow\Gamma_2\circ\Gamma_1$, $\text{pr}_1: \Gamma_2\star\Gamma_1\rightarrow\Gamma_1$ and $\text{pr}_2: \Gamma_2\star\Gamma_1\rightarrow\Gamma_2$ are canonical projections. This means that given a section of $\mathbb{L}_{Maslov}(\Gamma_2)$ and a section of $\mathbb{L}_{Maslov}(\Gamma_1)$, one can compose them to get a section of $\mathbb{L}_{Maslov}(\Gamma)$. Also we can compose sections of $\left|T\Gamma_2\right|^{\frac{1}{2}}$ with sections of $\left|T\Gamma_1\right|^{\frac{1}{2}}$ as morphisms in the enhanced symplectic “category” to get a section of $\left|T(\Gamma_2\circ\Gamma_1)\right|^{\frac{1}{2}}$. Thus by tensoring these two compositions we can compose sections of $\mathbb{L}_{\Gamma_2}$ with sections of $\mathbb{L}_{\Gamma_1}$. By a deeper examination of all the composition laws we have $$\sigma(F_2\circ F_1)=\sigma(F_2)\sigma(F_1)$$ which means that the composition of symbols defined above is consitent with the composition of Fourier integral operators.
Next we look at the special case when the Lagrangian submanifolds are twisted conormal bundles. Suppose $Z\subset X$ is a submanifold and $f\in C^\infty(Z)$, then the twisted conormal bundle $\Lambda=N^*_fZ$ is an exact Lagrangian submanifold of $T^*X$. As in section 3, on any open set $U\subset X$ where $Z\cap U$ is the common zero set of functionally independent functions $u_1, \dots, u_k\in C^\infty (U)$, we have a Hörmander description $(W, \pi, \phi)$ of $\Lambda$ on $U$, where $\pi: W\rightarrow U$ is the projection from $W=U\times\mathbb{R}^k$ to $U$ and $\phi=\sum_{i=1}^ks_iu_i+f\in C^\infty(W)$. Note that on any fiber $\pi^{-1}(x)$ of $x\in U$, the function $\phi$ is just a linear functional in the variables $s_i$, which has Hassian zero, so the functions $\text{sgn}^\sharp\equiv 0$ and $s_\phi\equiv 1$ on $U$. Patching the local data together we get a canonical (global) isomorphism between $\mathbb{L}_{Maslov}(\Lambda)$ and the trivial line bundle $\mathbb{C}$ on $\Lambda$. Theorem 4 follows. Theorem 5 is true because for a twisted conormal bundle $\Lambda=N^*_fZ$, the symbol map $$\sigma: I^r(X, \Lambda)\rightarrow C^\infty(\mathbb{L}_{Maslov}\otimes\left|T\Lambda\right|^{\frac{1}{2}})$$ always has first argument the 1-section, which means that the composition of the Maslov factor is always trivial in the composition of symbols for canonical relations of the form $\Gamma=(\zeta_1\times id)N^*_fZ$. Thus via the isomorphism given in Theorem 4, the composition of symbols of such operators only work as composition of $\frac{1}{2}$-densities in the enhanced symplectic “category”.
[99]{} Lars Hörmander, $The$ $Analysis$ $of$ $Linear$ $Partial$ $Differential$ $Operators$ : Publisher: Springer Press, 2007 Victor Guillemin and Shlomo Sternberg, $Semi\textrm{-}classical$ $Analysis$ Publisher: International Press of Boston, 2013 Jingwen Chen, $Classification$ $of$ $Lagrangian$ $submanifolds$ $with$ $applications$ $to$ $canonial$ $relations$ (to appear) John M. Lee, $Introduction$ $to$ $Smooth$ $Manifolds$ Publisher: Springer Press, 2012\
*Email address:* `[email protected]`
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We investigate the coherence properties of an atomic beam evaporatively cooled in a magnetic guide, assuming thermal equilibrium in the quantum degenerate regime. The gas experiences two-dimensional, transverse Bose-Einstein condensation rather than a full three-dimensional condensation because of the very elongated geometry of the magnetic guide. First order and second order correlation functions of the atomic field are used to characterize the coherence properties of the gas along the axis of the guide. The coherence length of the gas is found to be much larger than the thermal de Broglie wavelength in the strongly quantum degenerate regime. Large intensity fluctuations present in the ideal Bose gas model are found to be strongly reduced by repulsive atomic interactions; this conclusion is obtained with a one-dimensional classical field approximation valid when the temperature of the gas is much higher than its chemical potential, $k_B T\gg |\mu|$.'
author:
- |
Y. Castin[^1], R. Dum and E. Mandonnet\
Lab. Kastler Brossel de l’ENS, 24 rue Lhomond,\
75 231 Paris Cedex 5, France\
A. Minguzzi and I. Carusotto\
Scuola Normale Superiore and INFM, Piazza dei Cavalieri 7,\
56 126 Pisa , Italy
title: Coherence properties of a continuous atom laser
---
[**Pacs:**]{} 03.75.Fi, 42.50.-p
The first experimental achievements of Bose-Einstein condensates in atomic vapors [@BEC1; @BEC2; @BEC3; @BEC4] have opened promising perspectives for atom optics: condensates constitute indeed atomic waves sources of much better coherence properties than the usual ‘thermal’ sources like the standard magneto-optical trap. These coherence properties have already been demonstrated experimentally: interferences experiments between two condensates have been performed at MIT [@interf_MIT] and at JILA [@interf_JILA], the first order correlation function of the atomic field has been measured in Münich [@Esslinger], and suppression of density fluctuations (that is fluctuations in the intensity of the atomic field) has been revealed by a measurement of the mean-field energy [@meanf] and of three-body losses [@three_body]. By inducing a coherent leak of atoms out of trapped condensates several groups have succeeded in creating pulsed or quasi-continuous ‘atom-lasers’ [@laser_manip].
For future applications the already realized ‘atom-lasers’ may suffer from the handicap of a low mean flux of atoms: the condensates were not experiencing any continuous loading of atoms, so that the coherent output of atoms terminated once the $\sim 10^6$ atoms of the condensate were leaked out. As the repetition rate of the whole sequence is limited by the time required to form a condensate by evaporative cooling (on the order of seconds) the resulting mean flux of atoms is $< 10^6$ atoms/s. Several proposals have been made to refill the condensates with atoms in a continuous way [@laser_theory] but they have to our knowledge not been realized yet. Recently we proposed a different scheme, based on the evaporative cooling of an atomic beam [@cal].
The goal of the present article is to predict the essential features of our ‘continuous atom-laser’ proposal. The paper is organized as follows. In section \[evap\] we summarize the calculations of [@cal] performed to determine the required length of evaporative cooling of the beam to reach the quantum degenerate regime. In section \[ideal\] we discuss the coherence properties of the beam once quantum degeneracy has been obtained, assuming that the atomic interactions are negligible; we find that the beam has large intensity fluctuations incompatible with expected coherence properties of an atom-laser. In section \[inter\] we propose a model of a one-dimensional interacting Bose gas and we construct a classical field approximation to this model in the high temperature limit. We solve the classical field approximation using the formal analogy between functional integrals and quantum propagators: we find that the interactions between particles can dramatically reduce the intensity fluctuations of the beam. We conclude in section \[concl\].
Evaporative cooling of an atomic beam {#evap}
=====================================
A continuous injection of atoms in a magnetic guide {#sub:param}
---------------------------------------------------
In the experimental scenario considered at the École normale supérieure the continuous source of atoms is provided by standard laser cooling and trapping techniques taking place inside a cell. Atoms in the cell are captured, cooled and trapped in the $x-y$ plane using a two-dimensional magneto-optical trap, that is with laser beams in the $x-y$ plane and a dipolar magnetic field $\vec{B}\propto x\vec{e}_x-y\vec{e}_y$, where $\vec{e}_{x,y}$ are unit vectors along $x,y$ axes. The atomic motion along $z$ is controlled with a standard moving molasses technique: the two counter-propagating laser beams along $z$ have different frequencies so that the atoms are cooled around a non-zero mean velocity $\bar{v}_0$.
The already cold atomic beam emerging from this set-up is sent towards a magnetic guide of axis $z$. The magnetic guide is produced by the superposition of a uniform magnetic field along $z$ and a dipolar magnetic field $\propto x\vec{e}_x-y\vec{e}_y$. This provides a transverse confinement of the atoms being in the right Zeeman sublevels, the atoms in the wrong Zeeman sublevels being not trapped or even expelled. The magnetic guide provides a trapping potential in the $x-y$ plane with a harmonic bottom that we write as $$U(x,y) = \frac{1}{2} m\omega_\perp^2 (x^2+y^2)
\label{eq:pot}$$ where $m$ is the atomic mass and $\omega_\perp$ the transverse oscillation frequency of the atoms.
The following parameters of the injected atoms are expected to be realistic for $^{87}$Rb atoms [@cal]. The initial velocity dispersion of the atoms is $\Delta v_0= 20$ cm/s corresponding to an initial temperature of 400 $\mu$K; such a high temperature is obtained after spatial compression of the cloud usually performed to increase the collision rate in preparation of evaporative cooling. We take an injection velocity $\bar{v}_0=3\Delta v_0$ larger than the velocity spread so that the incoming atoms form a beam. Assuming an injected flux of atoms of $3\times 10^9$ atoms/s and an oscillation frequency $\omega_\perp
=2\pi\times 1$ kHz we find an initial on-axis thermal density of $8\times 10^{11}$ atoms/cm$^3$. The initial phase space density is $7\times 10^{-7}\ll 1$. The initial collision rate of atoms is related to the on-axis density $n_0$ and to the collisional cross section $\sigma$ by $\gamma_{\rm coll}^{(0)}=(2/\sqrt{\pi})\, n_0\, \sigma \, \Delta v_0$. For the $s$-wave collisional cross-section of rubidium ($\sigma=7.6 \times 10^{-16}$ m$^2$) we get $\gamma_{\rm coll}^{(0)}
\simeq$ 100 s$^{-1}$ which is much smaller than $\omega_\perp$: $$\frac{\gamma_{\rm coll}^{(0)}}{\omega_\perp} = 0.02 \ll 1.
\label{eq:coll}$$ The atoms have therefore the time to perform a full transverse oscillation in the trapping potential before experiencing a collision.
Modeling of evaporative cooling
-------------------------------
We assume that the atoms are subject to evaporative cooling in the magnetic guide, e.g. by application of a $z$-dependent radio-frequency flipping the atoms to untrapped or expelled Zeeman sublevels when they are too far from $z$ axis.
The dynamics of evaporative cooling can be described by the classical Boltzmann equation on the phase space atomic density $f(\vec{r},\vec{p})$ as long as the phase space density remains small. The collision terms in the Boltzmann equation are simplified by the assumption (justified for rubidium) that atomic interactions take place in the $s$-wave only and have a constant (momentum independent) total cross-section $\sigma=8\pi a_{3d}^2$ ($a_{3d}$ is the scattering length). The effect of evaporative cooling is modeled by setting $f(\vec{r},\vec{p})$ to zero in the domain $x^2+y^2>\Lambda^2(z)$ for two-dimensional evaporation, or in the domain $x^2 > \Lambda^2(z)$ for one-dimensional evaporation, where $\Lambda(z)$ is an adjustable cut in position space.
We have first looked for an approximate analytical solution of the Boltzmann equation, adapting to our geometry the truncated Gaussian ansatz put forward in [@Walraven]. As argued in [@cal] it is then more rigorous to restrict to a one-dimensional evaporation scheme. The ansatz then takes the following form: $$f(\vec{r},\vec{p})= f_0(z) e^{-(\epsilon_x+\epsilon_y)/(k_B T(z))} e^{-(p_z-\bar{p}(z))^2/(2mk_B T(z))}
Y(\epsilon_{\rm evap}(z)-\epsilon_x)
\label{eq:ansatz}$$ where $\epsilon_x, \epsilon_y$ are the sum of kinetic energy and harmonic trapping potential energy along $x$ and $y$ respectively and $Y$ is the Heaviside function. The ansatz assumes a local thermal equilibrium with temperature $T(z)$. The temperature depends on $z$ only, not on $x$ and $y$ as, due to Eq.(\[eq:coll\]), the mean free path of the particles $\sim \Delta v/\gamma_{\rm coll}$ is much larger than the spatial transverse extension of the gas $(k_B T/(m \omega_\perp^2))^{1/2}\sim \Delta v/\omega_\perp$: transversally the gas is in the so-called collisionless regime. For the same reason the truncation of $f$ in position space is replaced by a truncation in energy space, with $$\epsilon_{\rm evap} (z) = \frac{1}{2} m\omega_\perp^2 \Lambda^2(z).$$ Knowing the energy of the particle along $x$ allows to calculate the maximal excursion of the trajectory along $x$ (as atoms have in general the time to perform a full harmonic oscillation before experiencing a collision); if this maximal excursion exceeds $\Lambda(z)$ the particle is evaporated.
There are [*a priori*]{} three unknown functions of $z$ in the ansatz Eq.(\[eq:ansatz\]): (i) the normalization factor $f_0(z)$ or equivalently the linear density $\rho_{\rm lin}(z)$, (ii) the mean momentum $\bar{p}(z)$ of the gas along $z$, and (iii) the temperature of the gas $T(z)$.
By multiplying Boltzmann’s equation by (i) unity, (ii) the momentum $p_z$, and (iii) the kinetic energy along $z$, $p_z^2/(2m)$, and by integrating over $x,y,p_x,p_y,p_z$ one gets three hydrodynamic type equations for $\rho_{\rm lin}(z)$, $\bar{p}(z)$ and $k_B T(z)$. These three equations contain the usual equations expressing conservation of probability, of momentum and of energy, plus extra terms describing the loss of particles, the change of momentum and energy due to the evaporation.
We have solved numerically the hydrodynamic type equations in steady state, assuming that the $z$-dependence of the parameter $\Lambda(z)$ is adjusted to maintain a $z$-independent ratio $\eta=\epsilon_{\rm evap}(z)/(k_B T(z))$. For the specific set of parameters of §\[sub:param\] we have to gain seven orders of magnitude on the phase space density to reach quantum degeneracy. The smallest spatial length of evaporation required is obtained for $\eta\simeq 5$ as shown in figure \[fig:evap\]; it is equal to 7600 $d_0$ where $d_0=\sqrt{\pi}/(2n_0 \sigma)$ is the mean free path at the entrance of the magnetic guide, that is $\simeq 11$ meters for the considered parameters. After evaporative cooling along these 11 meters the flux of particles has been reduced by a factor 90 and the temperature has been decreased by a factor 4000.
We have also performed a numerical simulation of the full Boltzmann equation by a Monte Carlo algorithm using macro-atoms [@Bird]. The resulting numerical calculations take several days on a workstation. We have first simulated the one-dimensional evaporation scheme. In this way we have confirmed the accuracy of the predictions based on the ansatz Eq.(\[eq:ansatz\]). We have also performed simulations for the full two-dimensional evaporation. We have found typically that the length required to reach quantum degeneracy is reduced by a factor three as compared to one-dimensional evaporation, with the same loss of two orders of magnitude on the flux. For the specific set of parameters of §\[sub:param\] the required evaporation length reaches the experimentally reasonable value of 4 meters.
Coherence properties in the ideal Bose gas model {#ideal}
================================================
We now assume that evaporative cooling has allowed to reach the quantum degenerate regime at a certain distance from the entrance of the tube. We have not performed any kinetic study of the approach of quantum degeneracy, and we will here simply assume that the gas is at thermal equilibrium in the frame moving at the mean velocity of the gas. Such an assumption is reasonable if the temperature remains significantly larger than the quantum $\hbar\omega_\perp$ of transverse oscillation of the atoms of the guide. If the temperature was much smaller than $\hbar\omega_\perp/k_B$ the transverse degrees of freedom of the gas would be frozen in the ground state of the transverse harmonic oscillator; the gas would become a free one-dimensional Bose gas along $z$, that could not thermalize as colliding identical particles in one dimension simply exchange their momenta.
In this section we consider the model of the ideal Bose gas. The effects of atomic interactions are discussed in the next section.
Transverse Bose-Einstein condensation
-------------------------------------
Let us enclose the ideal Bose gas in a fictitious box of size $L$ along $z$, with periodic boundary conditions. Transversally the gas is confined by the harmonic potential Eq.(\[eq:pot\]). The one-particle eigenstates of the system are then labeled by three integers, the non-negative integers $l_x,l_y$ labeling the eigenstates of the harmonic oscillator along $x$ and $y$, and the integer $l_z$ labeling the momentum along $z$: $$\hbar k_z = \frac{2\pi}{L} l_z.
\label{eq:genial}$$ In the grand canonical ensemble the mean occupation number of the single particle state $\vec{l}=(l_x,l_y,l_z)$ is given by $$n(\vec{l},\mu) = \left\{\exp\left[\beta\left(\frac{\hbar^2k_z^2}{2m}+(l_x+l_y)\hbar\omega_\perp-\mu\right)\right]-1\right\}^{-1}.
\label{eq:Bose}$$ For convenience we have included the transverse zero-point motion energy in the chemical potential, so that $\mu$ varies from $-\infty$ to 0.
It turns out that in our trapping geometry there is no Bose-Einstein condensation in the thermodynamical limit defined as $L,N\rightarrow +\infty$ with a fixed linear density $N/L$, $N$ being the mean number of particles.
Let us consider indeed a fixed value of $L$ and let us define the maximal mean number of particles $N'_{\rm max}$ that can be put in all states but the ground state of the trap at a fixed temperature. A Bose-Einstein condensate forms in the ground state of the trap when $N$ exceeds $N'_{\rm max}$. As each $n(\vec{l},\mu)$ (for $\vec{l}\neq\vec{0}$) reaches its maximal accessible value for $\mu=0$, $N'_{\rm max}$ is given by $$N'_{\rm max} = \sum_{\vec{l}\neq\vec{0}} n(\vec{l},\mu=0).$$ A lower bound on $N'_{\rm max}$ is obtained by restricting the sum to the states transversally in the ground state of the harmonic oscillator: $$N'_{\rm max} \geq \sum_{l_z\neq 0} \left[\exp\left(\beta\frac{h^2 l_z^2}{2 m L^2}\right) -1\right]^{-1}.$$ When $L$ is large enough so that $k_B T \gg h^2/(2m L^2)$ the exponential in the denominator can be expanded to first order, leading to $$N'_{\rm max} \geq \frac{\pi}{3} \left(\frac{L}{\lambda}\right)^2$$ where we have introduced the thermal de Broglie wavelength $$\lambda = \frac{h}{(2\pi m k_B T)^{1/2}}.
\label{eq:lambda}$$ One then realizes that $N'_{\rm max}$ grows faster than $N$ in the thermodynamical limit $L\rightarrow +\infty$, with $N/L$ fixed. This is connected to the known fact that there is no Bose-Einstein condensation in a one-dimensional homogeneous Bose gas in the thermodynamical limit.
On the other hand, transverse Bose-Einstein condensation [@Ketterle_ibg] is taking place in our system. Let us calculate indeed the maximal number of atoms that can be put for a fixed temperature in the transversally excited single particle states: $$N^{\perp}_{\rm max} = \sum_{(l_x,l_y)\neq (0,0)} \sum_{l_z} n(\vec{l},\mu=0).$$ Replacing in the large $L$ limit the sum over $l_z$ by an integral and expanding $1/(\exp(x)-1)$ in powers of $\exp(-x)$ we obtain: $$N^{\perp}_{\rm max} \simeq \frac{L}{\lambda} \sum_{s\geq 1} s^{-1/2} \left\{
\left[1-\exp\left(-s\beta\hbar\omega_\perp\right)\right]^{-2} -1\right\} \simeq \frac{L}{\lambda} \left(\frac{k_B T}{\hbar\omega_\perp}\right)^2
\zeta(5/2).
\label{eq:inter}$$ The last equality is correct in the limit $k_B T \gg \hbar\omega_\perp$. The function $\zeta$ is the Zeta function of Riemann, and $\zeta(5/2)\simeq 1.341$. If the linear density of the gas $\rho_{\rm lin}=N/L$ exceeds the critical value $$\rho_{\rm lin}^{(c)} = N^{\perp}_{\rm max}/L \simeq \frac{1}{\lambda} \left(\frac{k_B T}{\hbar\omega_\perp}\right)^2\zeta(5/2)
\label{eq:crit}$$ the excess of density will accumulate in the transverse ground state of the trap. At linear densities much higher than $\rho_{\rm lin}^{(c)}$ the gas becomes almost monomode transversally: this is an interesting feature for atom optics applications as we have achieved in this way a Heisenberg limited transverse focalization of the beam.
We can also consider the maximal on-axis density of atoms in the excited transverse states. In this case it is more convenient to label the eigenstates of the two-dimensional harmonic oscillator by the angular momentum $M$ along $z$ and the radial quantum number $l_{\rm r}$. Then the wavefunctions on $z$ axis, that is in $x=y=0$, have a squared modulus equal to zero if $M\neq 0$ and equal to $m\omega_{\perp}/(\pi \hbar)$ for $M=0$. We recall that the states with $M=0$ have an energy above the zero-point energy given by $2 l_{\rm r} \hbar\omega_\perp$. We then obtain the maximal on-axis density of atoms in excited transverse states: $$\rho_{\rm axis}^{(c)} = \frac{m\omega_{\perp}}{\pi \hbar L} \sum_{l_{\rm r}\geq 1} \sum_{l_z}
\left\{\exp\left[\beta\left(\frac{h^2 l_z^2}{2m L^2}+2 l_{\rm r} \hbar\omega_\perp\right)-1\right]\right\}^{-1}.
\label{eq:onaxis1}$$ With the same algebraic transformations as for Eq.(\[eq:inter\]) we obtain $$\rho_{\rm axis}^{(c)} \simeq \frac{m\omega_{\perp}}{\pi \hbar \lambda} \sum_{s\geq 1} s^{-1/2}
\left[\frac{1}{1-\exp(-2 s\beta\hbar\omega_\perp)}-1\right]
\simeq \frac{\zeta(3/2)}{\lambda ^3}
\label{eq:onaxis2}$$ where $\zeta(3/2)\simeq 2.612$ and we have used $k_B T \gg \hbar\omega_\perp$. Transverse condensation therefore takes place when the usual Einstein’s condition $\rho \lambda^3 =
\zeta(3/2)$ is satisfied on the axis of the trap!
In the quantum degenerate regime
--------------------------------
We assume now that the gas is in the strongly degenerate regime, with a linear density $\rho_{\rm lin}$ larger than the critical value Eq.(\[eq:crit\]). The linear density of atoms in the excited transverse states has reached its saturated value $\rho_{\rm lin}^{(c)}$. This implies that $|\mu|$ is much smaller than $\hbar\omega_\perp$ ($\mu$ can be replaced by zero for the transversally excited states); as $k_B T \gg \hbar\omega_\perp$ one has also $$|\mu|\ll k_B T
\label{eq:crucial}$$ so that the occupation number of the absolute trap ground state $\vec{l}=\vec{0}$, though not of order $N$, is much larger than unity: $$n(\vec{l}=0) = \frac{1}{\exp(-\beta\mu) -1} \gg 1.$$ Let us calculate the linear density of atoms in the transverse ground state of the trap: $$\begin{aligned}
\rho_{\rm lin}^{(0)} &=& \frac{1}{L} \sum_{l_z} n(0,0,l_z;\mu) \\
&=& \frac{1}{L} \sum_{l_z} \left\{\exp[\beta(\hbar^2 k_z^2/(2m) -\mu)]-1\right\}^{-1}.\end{aligned}$$ As $\beta |\mu|\ll 1$ and the sum is one-dimensional the main contribution to the sum comes from states with kinetic energies on the order of $|\mu|$. Expanding the exponential in the Bose formula to first order we obtain a Lorentzian approximation for the occupation number as function of momentum: $$n(0,0,l_z;\mu) \simeq \frac{k_B T}{\hbar^2 k_z^2/(2m) + |\mu|}.
\label{eq:lorentz}$$ Replacing finally the sum by an integral we obtain $$\rho_{\rm lin}^{(0)} = \frac{1}{\lambda} \left(\frac{\pi k_B T}{|\mu|}\right)^{1/2}.
\label{eq:rho_lin_0}$$ This allows to express the chemical potential as function of density, when combined with the relation $$\rho_{\rm lin}=\rho_{\rm lin}^{(c)} + \rho_{\rm lin}^{(0)}.$$ Note that such a calculation would fail for a two-dimensional or three-dimensional free Bose gas, a Lorentzian momentum distribution being not normalizable in this case.
To characterize the coherence properties of the gas we use correlation functions for the atomic field operator $\hat{\Psi}(x,y,z)$ in direct analogy with the correlation functions considered in optics for the photonic field [@quantum_optics].
We define the first order correlation function as $$g_1(z) = \langle \hat{\Psi}^\dagger (0,0,z) \hat{\Psi}(0,0,0)\rangle$$ where the expectation value is taken in thermal equilibrium; this function is sensitive to the coherence of the atomic field between two points on the axis of the guide separated by a distance $|z|$. It can be written as the sum of the contributions of the transversally excited states and of the states in the transverse ground state. The two contributions behave in a very different way in the degenerate limit, that is in the limit $\mu\rightarrow 0$. This is illustrated in figure \[fig:g1\] for a moderately degenerate regime.
By calculations similar to the ones leading to Eqs.(\[eq:onaxis1\],\[eq:onaxis2\]) we find for the contribution of the transverse excited states to $g_1$ in the thermodynamical limit: $$g_1^{\perp}(z) = \frac{m\omega_\perp}{\pi\hbar\lambda} \sum_{s\geq 1}
\frac{e^{\beta \mu s}}{s^{1/2}}
\left[\frac{1}{1-\exp(-2s\beta\hbar\omega_\perp)}-1\right]
\exp\left(-\pi \frac{z^2}{s \lambda ^2}\right)\label{eq:exc}.$$ Using furthermore the fact that $|\mu|<\hbar\omega_{\perp}\ll k_B T$ we set $\mu=0$ and we expand the exponential function between square brackets to first order in $\beta\hbar\omega_{\perp}$: $$g_1^{\perp}(z) \simeq \frac{1}{\lambda^3} \sum_{s\geq 1} \frac{1}{s^{3/2}} \exp\left(-\pi \frac{z^2}{s \lambda ^2}\right).$$ The maximal value of $g_1^{\perp}$ is obtained in $z=0$ and is equal to $\rho_{\rm axis}^{(c)}$. The half-width of $g_1^{\perp}$ is on the order of $0.75 \lambda$.
The contribution to $g_1$ of the states in the transverse ground state is given by $$g_1^{(0)}(z) = \frac{m\omega_\perp}{\pi\hbar L} \sum_{l_z} n(0,0,l_z;\mu) e^{i k_z z}.$$ In the thermodynamical limit, we replace the sum by an integral. We use the Lorentzian approximation for the occupation numbers Eq.(\[eq:lorentz\]) and calculate its Fourier transform. This leads to a correlation function being an exponential function of $|z|$: $$g_1^{(0)}(z) \simeq \frac{m\omega_\perp}{\pi \hbar}\rho_{\rm lin}^{(0)} e^{-z/l_c}
\label{eq:avant}$$ with a coherence length $$l_c = \frac{\hbar}{(2m|\mu|)^{1/2} }
= \frac{\lambda^2}{2\pi}\rho_{\rm lin}^{(0)}.
\label{eq:plus_exacte}$$ The exponential decay of $g_1$ found here is to be contrasted with $g_1$ going to a finite value in the case of a three dimensional homogeneous Bose-Einstein condensate [@revue]. The coherence length $l_c$ can be larger than the thermal de Broglie wavelength $\lambda$. It can be expressed in terms of the on-axis density using the relation $g_1^{(0)}(z=0)=\rho_{\rm axis}-\rho_{\rm axis}^{(c)}
\equiv\rho_{\rm axis}^{(0)}$: $$l_c = \frac{\lambda^2 \hbar}{2m\omega_\perp}\rho_{\rm axis}^{(0)}.$$ A related issue is the expression connecting the on-axis and the linear density of the atoms in the transverse Bose-Einstein condensate: $$\frac{\rho_{\rm axis}^{(0)}}{\rho_{\rm axis}^{(c)}} \simeq \frac{2\zeta(5/2)}{\zeta(3/2)}
\frac{k_B T}{\hbar\omega_{\perp}}\
\frac{\rho_{\rm lin}^{(0)}}{\rho_{\rm lin}^{(c)}}.
\label{eq:rapport}$$ This expression holds in the regime $k_B T \gg \hbar\omega_{\perp}$ so that a modest value of $\rho_{\rm lin}$ above the critical value $\rho_{\rm lin}^{(c)}$ may actually correspond to a strongly degenerate regime $\rho_{\rm axis}\lambda^3\gg 2.612$.
We define the second order correlation function of the atomic field as $$g_2(z) = \langle \hat{\Psi}^\dagger(0,0,z)\hat{\Psi}^\dagger(0,0,0) \hat{\Psi}(0,0,0)
\hat{\Psi}(0,0,z)\rangle.$$ From a field point of view $g_2$ is the correlation function of the intensity of the field; it is a measure of the intensity fluctuations of the field. From a corpuscular point of view $g_2(z)$ is proportional to the pair distribution function of the atoms in the gas, that is the probability density to find a pair of atoms separated by a distance $|z|$ in the gas.
For the ideal Bose gas in the grand canonical ensemble the use of Wick’s theorem readily allows one to express $g_2$ in terms of $g_1$ by performing all the possible binary contractions of the field operators: $$g_2(z) = g_1^2(0) + g_1^2(z).
\label{eq:g2p}$$ We find the unpleasant result that the gas is subject to large intensity fluctuations over a length scale on the order of the coherence length $l_c$. In particular $$\frac{g_2(0)}{g_1^2(0)} = 2.$$ This value of two is typical of a bosonic bunching effect that manifests itself in an Hanbury-Brown and Twiss experiment in optics with thermal sources. It has to be contrasted with $g_2 \simeq g_1^2$ obtained with laser light for photons or with almost pure Bose-Einstein atomic condensates [@meanf]. The output of our magnetic guide cannot be termed an ‘atom-laser’ if $g_2(z)$ significantly differs from $g_1^2(0)$ over the coherence length of the field. Fortunately we shall see in the next sections that atomic interactions can improve the situation.
[**Remark:**]{}
A careful reader may argue that it is dangerous to use the grand canonical ensemble for the ideal Bose gas to calculate fluctuations of the number of particles, and therefore of the field intensity. This fear is justified when a condensate is formed, a well known problem in three-dimensions: the number of particles in the condensate has then unphysically large fluctuations. This problem does not take place here in the thermodynamical limit where no Bose-Einstein condensate is formed. More precisely one can deduce from $g_2$ the standard deviation of the total number of particles: $$\frac{\Delta N}{N} = \left(\frac{l_c}{L}\right)^{1/2}.$$ The relative fluctuations in the number of particles become small as soon as the length $L$ of the gas becomes much larger than the coherence length $l_c$.
To illustrate the strong intensity fluctuations of the ideal Bose gas in a dramatic way we introduce the Sudarshan-Glauber P representation of the many-body density operator [@quantum_optics]: $$\hat{\sigma} = \int {\cal D}\Psi
\, |\mbox{coh}:\Psi\rangle
\langle\mbox{coh}:\Psi|\, P(\{\Psi\},\{\Psi^*\}).
\label{eq:defP}$$ This expression is a functional integral $\int {\cal D}\Psi$ over the real part and the imaginary part of the c-number field $\Psi(\vec{r}\,)$. It involves the coherent or Glauber state of the atomic field associated to $\Psi$: $$|\mbox{coh}:\Psi\rangle \equiv \exp\left[-\frac{1}{2} \int d^3\vec{r}\, |\Psi(\vec{r}\,)|^2\right]
\exp\left[\int d^3\vec{r}\, \Psi(\vec{r}\,) \hat{\Psi}^\dagger(\vec{r}\,)\right] |\mbox{vacuum}\rangle.
\label{eq:coh}$$ Here the many-body density operator $\hat{\sigma}$ is the grand canonical thermal density operator $$\hat{\sigma} \propto \exp[-\beta(\hat{H}-\mu \hat{N})]$$ where $\hat{H}$ is the Hamiltonian of the gas containing the kinetic energy and the trapping potential energy, and $\hat{N}$ is the operator total number of particles. The Glauber distribution function $P$ can then be calculated exactly [@quantum_optics]. One expands the field $\Psi$ on the eigenmodes of the trap: $$\Psi(\vec{r}\,) \equiv \sum_{\vec{l}} \alpha_{\vec{l}}\; \phi_{l_x}(x)\, \phi_{l_y}(y)
\frac{1}{L^{1/2}}\,e^{i k_z z}$$ where $\phi_n$, $n=0,1,\ldots$ are the normalized eigenfunctions of the 1D harmonic oscillator of frequency $\omega_\perp$ and where the plane waves along $z$ have a wavevector given by Eq.(\[eq:genial\]). Then the Glauber distribution function is simply a product over all modes of Gaussian distributions with squared widths given by the occupation number of the modes: $$P(\{\Psi\},\{\Psi^*\}) \propto \exp\left[-\sum_{\vec{l}}
\frac{|\alpha_{\vec{l}}\, |^2}{n(\vec{l}\,)}\right].$$
As this distribution $P$ is positive the thermal equilibrium $\hat{\sigma}$ can be viewed exactly as a statistical mixture of coherent states. One can then [*imagine*]{} that a given experimental realization of the Bose gas is a coherent state characterized by a field $\Psi$. This field $\Psi$ is stochastic as it varies in an unpredictable way from one experimental realization to the other. As the coherent state $|\mbox{coh}:\Psi\rangle$ is an eigenstate of $\hat{\Psi}(\vec{r}\,)$ with the eigenvalue $\Psi(\vec{r}\,)$ one can check that the correlation functions $g_1$ and $g_2$ are equal to the following averages over all possible realizations of the c-number field: $$\begin{aligned}
\label{eq:sav1}
g_1 (z) &=& \langle \Psi^*(z) \Psi(0)\rangle_{\rm stoch} \\
g_2(z) &=& \langle |\Psi(z)|^2 |\Psi(0)|^2 \rangle_{\rm stoch}.
\label{eq:sav2}\end{aligned}$$
We have plotted in figure \[fig:glaub\] the intensity of the field as function of position for two numerically generated realizations of $\Psi$. Figure \[fig:glaub\]a corresponds to a non-degenerate situation; the only spatial scale for the intensity fluctuations is the thermal de Broglie wavelength $\lambda$. Figure \[fig:glaub\]b corresponds to a strongly degenerate regime; there are clearly two spatial scales for the intensity fluctuations, one on the order of $\lambda$ coming from the transversally non-condensed fraction and the other one on the order of $l_c$ due to the gas in the transverse ground state of the trap. The large intensity fluctuations at the scale of $l_c$ manifest themselves as “droplets" in the atomic density.
The interacting case: a one-dimensional classical field model {#inter}
=============================================================
Model Hamiltonian for the transversally Bose condensed gas
----------------------------------------------------------
We have seen on the ideal Bose gas model that transverse Bose-Einstein condensation takes place in the magnetic guide at sufficient high density. Although we have not performed any detailed analysis we expect the same phenomenon to occur for the interacting Bose gas in the weakly interacting regime.
At the presently considered low temperatures, interaction between the atoms takes place in $s$-wave mainly, and the relative wavevector of two colliding atoms is much smaller than the inverse of the scattering length $a_{3d}$ of the interaction potential. The interaction potential is then commonly replaced by an effective low energy interaction potential $g_{3d} \delta(\vec{r}_1-\vec{r}_2)$ [@Huang; @Houches] with a coupling constant $$g_{3d} =\frac{4\pi\hbar^2}{m} a_{3d}.$$ The resulting Hamiltonian is expressed in terms of the field operator $\hat{\Psi}(x,y,z)$ as $$\begin{aligned}
\hat{H}_{3d} = \int dx\int dy \int_0^L dz\, \left[
\frac{\hbar^2}{2m}|\vec{\mbox{grad}}\,\hat{\Psi}|^2+\frac{1}{2} g_{3d} \hat{\Psi}^\dagger \hat{\Psi}^\dagger
\hat{\Psi} \hat {\Psi} +(U(x,y)-\mu) \hat{\Psi}^\dagger \hat {\Psi}\right]\end{aligned}$$ with the trapping potential given in Eq.(\[eq:pot\]) and with periodic boundary conditions along $z$. We have included for convenience the term $-\mu \hat{N}$ where $\mu$ is the chemical potential so that the thermal equilibrium density operator is simply $\propto\exp[-\beta \hat{H}_{3d}]$ in the grand canonical ensemble.
In our magnetic guide geometry the situation is particularly simple if the typical interaction energy per particle $\rho_{\rm axis} g_{3d}$ is smaller than the quantum of transverse oscillation $\hbar\omega_\perp$. Transverse Bose-Einstein condensation will then take place in a transverse wavefunction $\phi_{\perp}(x,y)$ close to the ground state of the harmonic oscillator. If we wish to describe only the one-dimensional Bose gas of atoms in this transverse condensate, assuming that the remaining atoms have a much smaller density, we can neglect the contribution of the transverse modes to the field operator by setting $$\hat{\Psi}(x,y,z) \simeq \phi_{\perp}(x,y) \hat{\psi}(z).
\label{eq:approx}$$ This will eliminate the contribution of the atoms in the transverse excited states of the trap; we keep in mind however that these atoms are essential to ensure thermalization, with a temperature $k_B T > \hbar\omega_\perp$. The reduced field operator $\hat{\psi}$ obeys the usual bosonic commutation relations of a one-dimensional bosonic field. Inserting the approximate expression for $\hat{\Psi}$ in the Hamiltonian results in the following model Hamiltonian for the transversally condensed Bose gas: $$\hat{H} = \int_0^L dz\, \left[
\frac{\hbar^2}{2m}|\partial_z\hat{\psi}|^2+\frac{1}{2} g \hat{\psi}^\dagger \hat{\psi}^\dagger
\hat{\psi} \hat {\psi} -\mu \hat{\psi}^\dagger \hat {\psi}\right].
\label{eq:model}$$ It corresponds to a one-dimensional Bose gas with a contact interaction potential $g\delta(z_1-z_2)$ between particles with an effective coupling constant [@Maxim] $$g = g_{3d} \int\!\!\!\int dx\, dy \, |\phi_\perp(x,y)|^4.$$
Classical field approximation
-----------------------------
The model Hamiltonian Eq.(\[eq:model\]) leads to an exactly solvable $N$-body problem: exact eigenenergies and eigenfunctions are known [@Lieb; @Gaudin]. It is however not easy to extract information from the exact solution, even at zero temperature.
We use here a simpler approach, valid in a sufficiently high temperature regime. The idea is to write the thermal equilibrium density operator (up to a normalization factor) as the result of a fictitious time evolution: $$\frac{d}{d\tau} \hat{\sigma} = -\frac{1}{2} \left(\hat{H}\hat{\sigma} +\hat{\sigma}\hat{H}\right)
\label{eq:master}$$ with the initial condition $\hat{\sigma}(\tau=0)$ equal to the identity operator. This evolution corresponds to the so-called imaginary time evolution. It leads to a density operator at ‘time’ $\tau$ given by: $$\hat{\sigma} = e^{-\tau \hat{H}}$$ which is (up to a normalization factor) the thermal equilibrium density operator at temperature $k_B T =1/\tau$. High temperatures correspond to low values of $\tau$ that is to short ‘time’ evolution.
To take advantage of the fact that the ‘time’ evolution is short it is better to rewrite Eq.(\[eq:master\]) using some of the representations of the density operator introduced in quantum optics. We use here the Glauber P distribution already introduced in section \[ideal\]. It is defined as in Eq.(\[eq:defP\]) with the difference that $\psi$ is now a function of the coordinate $z$ only. So we apply the Glauber transform to both sides of Eq.(\[eq:master\]). The transform of the right-hand is performed with the following rules: $$\begin{aligned}
\mbox{Glaub}\,[\hat{\psi}(z)\hat{\nu}] &=& \psi(z) \mbox{Glaub}\,[\hat{\nu}]
\label{eq:rule1}\\
\mbox{Glaub}\,[\hat{\psi}^\dagger(z)\hat{\nu}] &=& \left(\psi^*(z)-\partial_{\psi(z)}\right) \mbox{Glaub}\,[\hat{\nu}]
\label{eq:rule2}\end{aligned}$$ where $\hat{\nu}$ is any operator and $\mbox{Glaub}\,[\hat{\nu}]$ stands for the Glauber P distribution of $\hat{\nu}$. The first rule Eq.(\[eq:rule1\]) comes from the fact that the coherent state $|\mbox{coh}:\psi\rangle$ is an eigenstate of $\hat{\psi}(z)$ with the eigenvalue $\psi(z)$. The second rule Eq.(\[eq:rule2\]) involves a functional derivative with respect to the field $\psi$, the fields $\psi$ and $\psi^*$ being formally considered as independent variables. Its derivation closely follows the one for single mode fields in [@quantum_optics]. One first uses the following identity: $$\partial_{\psi(z)} \left[|\mbox{coh}:\psi\rangle \langle\mbox{coh}:\psi|\right] =
(\hat{\psi}^\dagger(z)-\psi^*(z)) |\mbox{coh}:\psi\rangle \langle\mbox{coh}:\psi|$$ as can be checked from the definition Eq.(\[eq:coh\]) transposed to the one-dimensional case. Then one integrates by part in the functional integral over the field to convert the derivative over the dyadic $|\mbox{coh}:\psi\rangle \langle\mbox{coh}:\psi|$ into a derivative of the Glauber P distribution. We finally obtain the Fokker-Planck type equation for the fictitious time evolution of the Glauber distribution: $$\partial_{\tau} P = - E\, P -\int dz\, \left[\partial_{\psi} (F(z)\, P)
+\frac{g}{4} \partial_{\psi}^2(\psi^2 P) +\mbox{c.c.}\right].
\label{eq:FP}$$
The first term in Eq.(\[eq:FP\]) involves a multiplication of $P$ by a functional $E$ which is simply the Gross-Pitaevskii energy functional [@revue], obtained by replacing in the Hamiltonian the field operator by the c-number field $\psi$: $$E[\{\psi\},\{\psi^*\}] = \int_0^L dz\, \left[
\frac{\hbar^2}{2m}|\partial_z{\psi}|^2+\frac{1}{2} g |{\psi}|^4 -\mu |{\psi}|^2\right].
\label{eq:gpen}$$ If this term was alone in the evolution equation for $P$ we could readily integrate it to obtain: $$P_{\rm class}[\{\psi\},\{\psi^*\}] = e^{-\tau E[\{\psi\},\{\psi^*\}]}.
\label{eq:Pclass}$$ We have termed this solution $P_{\rm class}$ as it corresponds to the thermal Boltzmann distribution for a classical field $\psi$ with an energy given by $E[\{\psi\},\{\psi^*\}]$! This is nineteenth century equilibrium physics for fields.
The next term in Eq.(\[eq:FP\]) can be termed a force term by analogy with the Fokker-Planck equation, as it involves a first order derivative in $\psi$. The ‘force’ functional is given by: $$F[\{\psi\},\{\psi^*\}](z)=-\frac{1}{2}\partial_{\psi^*(z)} E[\{\psi\},\{\psi^*\}]=
-\frac{1}{2}\left[
-\frac{\hbar^2}{2m}\partial_z^2 + g|\psi|^2 -\mu\right]\psi(z).$$ At sufficiently short ‘time’ $\tau$, that is at high temperature, the effect of the force term during the evolution ‘time’ $\tau$ is to shift the field $\psi(z)$ from its initial (random) value by the amount $F(z) \tau$. One may hope that this shift is negligible for high enough temperature.
Let us calculate this shift for the ideal Bose gas. The field $\psi$ is then conveniently expanded on plane waves with momenta given by Eq.(\[eq:genial\]): $$\psi(z)=\sum_{k_z} \alpha_{k_z} \frac{e^{i k_z z}}{L^{1/2}}.$$ Taking now $k_z$ and $\alpha_{k_z}$ as coordinate and field variables (rather than $z$ and $\psi(z)$) we can write: $$E = \sum_{k_z}\varepsilon_{k_z} \alpha^*_{k_z} \alpha_{k_z} \ \ \ \mbox{and} \ \ \ \
F(k_z)=-\frac{1}{2}\partial_{\alpha^*_{k_z}} E = -\frac{1}{2} \varepsilon_{k_z} \alpha_{k_z}$$ where we have introduced the mode eigenenergy $\varepsilon_{k_z}=\hbar^2 k_z^2/(2m)-\mu$. The shift $F(k_z)\tau$ has a negligible effect on field mode $k_z$ if it is small as compared to $\alpha_{k_z}$; this leads to the condition $$k_B T \gg \varepsilon_{k_z}.
\label{eq:highT}$$ The classical field approximation $P_{\rm class}$ is therefore an acceptable approximation for the modes with an energy much smaller than $k_B T$. The occupation number of such modes can then be obtained from the classical energy equipartition formula: a mode of a complex field corresponds to a one-dimensional harmonic oscillator so it has a mean energy equal to $k_B T$, and $$\langle |\alpha_{k_z}|^2\rangle = \frac{k_B T}{\varepsilon_{k_z}}.$$ This formula coincides indeed with the quantum Bose formula Eq.(\[eq:Bose\]) in the limit Eq.(\[eq:highT\]), that is in the limit of a large occupation number of the mode.
So when can we use the classical field approximation Eq.(\[eq:Pclass\])? The answer depends on the observable quantity we wish to calculate.
For the calculation of the mean energy the classical field approximation is never acceptable: in the absence of energy cut it predicts an infinite mean energy, the well known blackbody radiation catastrophe. To save the situation one has to introduce an energy cut $\varepsilon_{\rm cut}$ on the order of $k_B T$ to reproduce ‘by hand’ the fact that the Bose formula gives an exponentially small occupation number to modes with eigenenergy much larger than $k_B T$. The mean energy is then finite but depends on the precise value of $\varepsilon_{\rm cut}$.
The conclusion is different for the calculation of the correlation functions $g_1$ and $g_2$. We have actually already used the classical field approximation in the derivation of $g_1$, see the approximation Eq.(\[eq:lorentz\])! This did not lead to any divergence, a fortunate feature peculiar to the free one-dimensional Bose gas. The classical field predictions for $g_1$ and $g_2$ therefore do not depend on the energy cut $\varepsilon_{\rm cut}$ provided that the energy cut is large enough. For the ideal Bose gas the condition is $\varepsilon_{\rm cut}\gg |\mu|$, as the kinetic energy width of the Lorentzian Eq.(\[eq:lorentz\]) is $|\mu|$; as $\varepsilon_{\rm cut} \sim k_B T$ the classical field calculation of $g_{1,2}$ requires $k_B T \gg |\mu|$: we recover Eq.(\[eq:crucial\]).
The validity conditions of the classical field approximation for the interacting case are more subtle to derive and will be discussed in §\[subsec:val\_cond\].
Finally, the last term in Eq.(\[eq:FP\]) can be termed a diffusion term by analogy with the Fokker-Planck equation, as it involves second order derivatives in $\psi$. The corresponding ‘diffusion’ matrix in point $z$, given by: $$D = -\frac{g}{4} \left(\begin{tabular}{cc} $0$ & $\psi^2$ \\ $\psi^{*2}$ & $0$ \end{tabular}\right)$$ is however a non positive matrix: one can check that the field quadrature along $\psi$ is squeezed by the ‘diffusion’ while the field quadrature orthogonal to $\psi$ gets anti-squeezed. This non-positivity of the ‘diffusion’ matrix makes it impossible to perform a stochastic, Brownian type simulation of Eq.(\[eq:FP\]), which would have provided an exact numerical solution to the problem.
How to calculate the correlation functions in the classical field approximation {#subsec:how}
-------------------------------------------------------------------------------
We take here as Glauber P distribution the classical approximation Eq.(\[eq:Pclass\]), without introducing any energy cut, and we wish to calculate the stochastic averages Eqs.(\[eq:sav1\],\[eq:sav2\]). This amounts to calculating ratios of functional integrals over paths parametrized by $z\in [0,L]$.. For example the first order correlation function of the field $\hat{\Psi}(z)$ in the approximation Eq.(\[eq:approx\]) is given by $$g_1(z) = |\phi_\perp(0,0)|^2\frac{\int d\psi(0)\oint_{\Gamma_{\psi(0)}}{\cal D}\psi\,
\psi^*(z)\psi(0) e^{-\beta E[\{\psi\},\{\psi^*\}]}}
{\int d\psi(0)\oint_{\Gamma_{\psi(0)}} {\cal D}\psi\,
e^{-\beta E[\{\psi\},\{\psi^*\}]}}.
\label{eq:hor}$$ In the above formula the functional integrals are performed over all possible closed paths, as the gas is subject to spatially periodic boundary conditions; we have split the functional integral as a regular integral over the value of the path $\psi(0)$ in $z=0$ and a functional integral over the set $\Gamma_{\psi(0)}$ of all paths starting with the value $\psi(0)$ in $z=0$ and ending with the same value in $z=L$. We explain here how to calculate these functional integrals. The reader not interested in technicalities may jump directly to §\[subsec:results\].
To calculate functional integrals like Eq.(\[eq:hor\]) it is of course possible to use a Monte Carlo method to sample the distribution of $\psi$, e.g. by representing $\psi$ on a finite spatial grid with step $dz$ and by evolving $\psi$ one time step $dt$ after the other according to the stochastic evolution $$d\psi(z) = -dt\left[-\frac{\hbar^2}{2m}\partial_z^2 + g|\psi|^2 -\mu\right]\psi(z) + \left(\frac{k_B T}{dz}\right)^{1/2} d\xi(z)$$ where the $d\xi(z)$’s are statistically independent complex noises of variance $2 dt$.
There exists however a more elegant and much faster solution. One can use the link between path integrals and quantum mechanics propagator put forward by Feynman. The functional integral over the classical complex field then corresponds to a propagator in imaginary time of the quantum mechanical problem of a particle in two dimensions. The point by point analogy between the two problems is specified in the translation table:
------------------------------------------------------------------------------------------ -------------------- ------------------------------------------------------------
[**classical field problem**]{} [**quantum mechanical analogy**]{}
path integral $\leftrightarrow $ quantum propagator
abscissa $z$ $\leftrightarrow $ time $t$
Re($\psi(z)$) $\leftrightarrow $ position $x(t)$
Im($\psi(z)$) $\leftrightarrow $ position $y(t)$
$\int_{\psi(0)=\psi_i}^{\psi(z)=\psi_f}{\cal D}\psi\; e^{-\beta E[\{\psi\},\{\psi^*\}]}$ $\leftrightarrow $ $\langle x_f,y_f| e^{-t {\cal H}/\hbar} |x_i,y_i\rangle$ .
------------------------------------------------------------------------------------------ -------------------- ------------------------------------------------------------
We have to identify the Hamiltonian ${\cal H}$ of the equivalent quantum mechanics problem. We postulate the following form: $${\cal H} = \frac{p_x^2 + p_y^2}{2M} + V(x,y).
\label{eq:hamile}$$ The imaginary time propagator is then expressed in terms of the path integral [@Feynman]: $$\langle x_f,y_f| e^{-t {\cal H}/\hbar} |x_i,y_i\rangle =
\int_{x(0)=x_i}^{x(t)=x_f} {\cal D}x(\tau) \int_{y(0)=y_i}^{y(t)=y_f} {\cal D}y(\tau)\,
e^{-S[\{x\},\{y\}]/\hbar}$$ where the action $S$ is a functional of the path $x(\tau),y(\tau)$: $$S[\{x\},\{y\}]=\int_0^t d\tau\, \left[\frac{1}{2} M \left(\frac{dx}{d\tau}\right)^2+\frac{1}{2}
M\left(\frac{dy}{d\tau}\right)^2
+V(x(\tau),y(\tau))\right].$$ One identifies this action with $\hbar \beta E$ and one uses the translation table to obtain the values of the parameters of the equivalent quantum mechanics problem: a mass $$M = \frac{\hbar^3}{m k_B T}$$ and a potential $$V(x,y) =\hbar\beta \left[ \frac{g}{2} (x^2+y^2)^2-\mu (x^2+y^2)\right].
\label{eq:poten}$$ This potential is rotationally invariant, as a consequence of the $U(1)$ symmetry of the Hamiltonian Eq.(\[eq:model\]). We can therefore classify the eigenstates of $V$ with two quantum numbers, an angular momentum $m$ and a radial quantum number $n=0,1,\ldots$. We call $\phi_n^{m}$ the corresponding normalized eigenvector with eigenvalue $\varepsilon_n^{m}$. As usual the absolute ground state of ${\cal H}$ is of angular momentum $m=0$ and radial quantum number $n=0$.
We translate the functional integrals of Eq.(\[eq:hor\]) into quantum propagators. In particular we note that the integral over $\psi(0)$ in Eq.(\[eq:hor\]) corresponds to an integral over all possible initial coordinates of the particle, that is to a trace over all possible initial quantum states of the particle. We finally obtain: $$g_1(z) =|\phi_\perp(0,0)|^2\frac{\mbox{Tr}\left[ e^{-(L-z){\cal H}/\hbar} (x-iy) e^{-z{\cal H}/\hbar}(x+iy)\right]}
{\mbox{Tr}\left[e^{-L{\cal H}/\hbar}\right]}$$ and a similar expression for $g_2$.
Physically, as there is no Bose-Einstein condensation along $z$, one expects that the length of magnetic guide $L$ in the experiment is much larger than any correlation length of the gas. One can then take the thermodynamical limit along $z$, putting $L$ to infinity while keeping a constant chemical potential $\mu$ (this ensures that the mean linear density $N/L$ tends to constant) [@math]. In this case $\exp[-L {\cal H}/\hbar]$ becomes proportional to the projector on the absolute ground state of ${\cal H}$: $$e^{-L {\cal H}/\hbar} \sim e^{-L \varepsilon_0^{m=0}/\hbar} |\phi_0^{m=0}\rangle
\langle\phi_0^{m=0}|.$$ The thermodynamical limit approximation greatly simplifies the expressions for the correlation functions, as the trace $\mbox{Tr}$ can be restricted to the ground state of ${\cal H}$: $$\begin{aligned}
\label{eq:g1f}
g_1(z) &=& |\phi_\perp(0,0)|^2 \langle \phi_0^{m=0}| (x-iy) e^{-z({\cal H}-\varepsilon_0^{m=0})/\hbar}(x+iy)|\phi_0^{m=0}\rangle \\
g_2(z) &=& |\phi_\perp(0,0)|^4\langle \phi_0^{m=0}| (x^2+y^2) e^{-z({\cal H}-\varepsilon_0^{m=0})/\hbar}(x^2+y^2)|\phi_0^{m=0}\rangle.
\label{eq:g2f}\end{aligned}$$ The operator $x+iy$ maps the absolute ground state to a state with angular momentum equal to unity. If we restrict for simplicity to the large $z$ limit, the operator $e^{-z{\cal H}}$ in Eq.(\[eq:g1f\]) becomes proportional to the projector on the ground state $\phi_0^{m=1}$ of $\cal H$ with angular momentum $m=1$ so that $$g_1(z)\simeq |\phi_\perp(0,0)|^2 a_1 \exp(-\kappa_1 z)$$ with $$\kappa_1 = (\varepsilon_0^{m=1}-\varepsilon_0^{m=0})/\hbar \ \ \ \mbox{and}
\ \ \ \ a_1 = |\langle \phi_0^{m=1}|x+i y|\phi_0^{m=0}\rangle|^2.
\label{eq:k1}$$ A similar analysis can be applied to the correlation function $g_2$, with the difference that the operator $x^2+y^2$ in Eq.(\[eq:g2f\]) maps the absolute ground state to a rotationally invariant state. In the large $z$ limit we keep the contributions of the first two eigenstates with $m=0$ to obtain $$g_2(z)\simeq g_1^2(0) + |\phi_{\perp}(0,0)|^4 a_2 \exp(-\kappa_2 z)$$ with $$\label{eq:k2}
\kappa_2 = (\varepsilon_1^{m=0}-\varepsilon_0^{m=0})/\hbar \ \ \ \ \mbox{and}\ \ \ \
a_2 = |\langle \phi_1^{m=0}|x^2+ y^2|\phi_0^{m=0}\rangle|^2.$$
Results of the classical field approximation {#subsec:results}
--------------------------------------------
We wish to calculate the correlation functions $g_1$ and $g_2$ using the formalism of §\[subsec:how\]. One has then to solve the quantum mechanics equivalent problem of a particle in two-dimension with the Hamiltonian Eq.(\[eq:hamile\]). This Hamiltonian can be diagonalized numerically. We wish to express the results in terms of the linear density of the transversally condensed Bose gas, rather than in terms of the chemical potential $\mu$. For a given linear density $\rho_{\rm lin}^{(0)}$ we therefore have to adjust $\mu$ in order to satisfy $$\rho_{\rm lin}^{(0)}=\langle \hat{\psi}^\dagger(0)\hat{\psi}(0)\rangle.$$ The problem can be simplified by an efficient parameterization. We express the coordinates $x$ and $y$ in the quantum mechanics analogy (which correspond to the real and imaginary part of $\psi$) in units of the square root of the linear density. We express the physical length $z$ in units of $\rho_{\rm lin}^{(0)} \lambda^2/(2\pi)$ where $\lambda$ is the thermal de Broglie wavelength Eq.(\[eq:lambda\]). We then find that once $\mu$ has been eliminated there is a single parameter left in the classical field theory: $$\chi = \frac{\rho_{\rm lin}^{(0)}g}{k_B T} \, \frac{\left(\rho_{\rm lin}^{(0)}\lambda\right)^2}{2\pi}.
\label{eq:chi}$$
We plot in figure \[fig:kappa\] as function of $\chi$ the coefficients $\kappa_{1,2}$ giving the long distance behavior of the correlation functions $g_{1,2}$ as defined in Eqs.(\[eq:k1\],\[eq:k2\]). In the limit of a vanishing $\chi$ we recover the results of the ideal Bose gas, Eq.(\[eq:plus\_exacte\]), with $\kappa_1=1/l_c$, and Eq.(\[eq:g2p\]) leading to $\kappa_2 = 2 \kappa_1$. For an increasing interaction strength, $\chi$ increases: the coherence length $1/\kappa_1$ has a modest increase by up to a factor two; the correlation length of the intensity fluctuations $1/\kappa_2$ is dramatically reduced by the atomic interactions and becomes much smaller than the coherence length, a positive point already !
What happens to the amplitude of the intensity fluctuations ? In an ideal ‘atom-laser’ there is no fluctuation of the field intensity; the deviation from this ideal situation can be measured by the ratio of the maximal to the minimal value of $g_2$, that is $$\frac{g_2(0)}{g_1^{2}(0)}=\frac{\langle \phi_0^{(m=0)}| (x^2+y^2)^2|\phi_0^{(m=0)}\rangle}
{\left(\langle \phi_0^{(m=0)}|x^2+y^2|\phi_0^{(m=0)}\rangle\right)^2}.
\label{eq:ratio}$$ We have plotted this quantity in figure \[fig:fluc\]. It is equal to two for the ideal Bose gas, as predicted by Eq.(\[eq:g2p\]). It is sharply reduced by the atomic interactions for low values of $\chi$ then it slowly goes to unity for large values of $\chi$.
Can we understand the origin of the reduction of intensity fluctuations using the quantum mechanics analogy ? According to Eq.(\[eq:ratio\]) this amounts to understanding the fluctuations of the distance of the quantum mechanical particle from the origin in the ground state of ${\cal H}$. By inspection of Eq.(\[eq:poten\]) giving the trapping potential seen by the quantum mechanical particle we realize that there are two situations depending on the sign of the chemical potential. For a negative chemical potential, as in the case of the ideal Bose gas, the potential $V$ has an absolute [*minimum*]{} in $x=y=0$ so that $|\phi_0^{m=0}|^2$ is localized around the origin (see figure \[fig:poten\]a): large fluctuations of the intensity of the field are expected. For a positive chemical potential, which is the case for strong enough repulsive interactions, the potential $V(x,y)$ has a Mexican hat shape: it has a local [*maximum*]{} at the origin and a global minimum on a finite circle (see figure \[fig:poten\]b). The ground state wavefunction then tends to be localized around the circle. The critical regime for the apparition of the Mexican hat potential is such that $\mu=0$; we find numerically that this corresponds to $$\chi_c \simeq 0.28.$$ This low value of $\chi$ explains why a sharp variation is obtained at the scale of figure \[fig:fluc\].
In the large $\chi$ limit the quantum mechanical particle will get more deeply bound at the bottom of the Mexican hat potential. Writing Schrödinger’s equation for $\phi_n^{m=0,1}$ in polar coordinates and treating perturbatively the deviation of the Mexican hat plus centrifugal potential from a harmonic approximation we obtain after some algebra the large $\chi$’s expansions:
$$\begin{matrix}
\kappa_1 & =& \displaystyle\frac{2\pi}{\rho_{\rm lin}^{(0)}\lambda^2} \left[\frac{1}{2}
+\frac{1}{2\chi^{1/2}}+\ldots\right]
&\ \ \ \ \ & \kappa_2 &=&\displaystyle\frac{2\pi}{\rho_{\rm lin}^{(0)}\lambda^2}
\left[2\chi^{1/2}+\ldots\right] \\
a_1 &=&\displaystyle \rho_{\rm lin}^{(0)}\left[1+\ldots\right] & & a_2 &=&
\displaystyle\left(\rho_{\rm lin}^{(0)}\right)^2
\left[\frac{1}{\chi^{1/2}}+\ldots\right] \\
\displaystyle\frac{g_2(0)}{g_1^2(0)} &=& \displaystyle1+ \frac{1}{\chi^{1/2}}+\ldots & &
\mu&=&\displaystyle\rho_{\rm lin}^{(0)} g\left[1+\frac{1}{2\chi^{1/2}}+\ldots\right]
\end{matrix}
\label{eq:large}$$
We note that the correlation length of the intensity fluctuations $1/\kappa_2$ becomes proportional in the large $\chi$ limit to the so-called healing length $\xi$ of the gas, a crucial parameter in the theory of Bose-Einstein condensates [@revue]: $$\kappa_2^{-1} \simeq \frac{1}{2}\left(\frac
{\hbar^2} {m\rho_{\rm lin}^{(0)}g} \right)^{1/2}=
\frac{\xi}{\sqrt{2}}.
\label{eq:healing}$$
=6.5cm =6.5cm
=12 cm
Validity conditions of the classical field approximation {#subsec:val_cond}
--------------------------------------------------------
The validity condition of the classical field approximation in general depends on the observable to be calculated. Here the relevant quantities are the first and second order correlation functions $g_{1,2}$.
Let us recall briefly what happens in the ideal Bose gas case. The correlations functions are then characterized by a single length, the coherence length $l_c=1/\kappa_1=2/\kappa_2$. The plane wave modes of the field contributing to $g_{1,2}$ have therefore wave vectors on the order of $\kappa_1$ or less; for these modes the validity condition Eq.(\[eq:highT\]) reads $k_B T \gg \varepsilon_{\kappa_1}$. The mode eigenenergy $\varepsilon_{\kappa_1}$ is equal to $\hbar^2 \kappa_1^2/(2m)-\mu$; using the value Eq.(\[eq:plus\_exacte\]) of the coherence length $l_c=1/\kappa_1$ we arrive at $$k_B T \gg |\mu|.$$ This result was already obtained in Eq.(\[eq:crucial\]). It can be rewritten with Eq.(\[eq:rho\_lin\_0\]) as a hierarchy among the three relevant lengths of the problem: $$\left(\rho_{\rm lin}^{(0)}\right)^{-1} < \lambda < l_c=\kappa_1^{-1}.$$
What happens in the interacting regime $\chi > 1$ ? A difficulty is that the Hamiltonian for the classical field $\psi$ in Eq.(\[eq:gpen\]) is no longer quadratic in $\psi$ so that it is not straightforward to calculate energy eigenmodes. Can we find some good quadratic approximation to it?
One could think to use the Bogolubov approach [@revue]. In this approach one identifies the field $\psi_0$ minimizing the energy Eq.(\[eq:gpen\]); in the present homogeneous case $\psi_0$ is $z$-independent, $\psi_0=(\mu/g)^{1/2}\exp(i\theta_0)$ where $\theta_0$ is an arbitrary constant phase. Then one splits the field as $\psi(z)=\psi_0+\delta\psi(z)$. Under the assumption of $|\delta\psi|\ll |\psi_0|$ one neglects in Eq.(\[eq:gpen\]) the terms cubic and quartic in $\delta\psi$ which leads to a quadratic Hamiltonian that can be diagonalized. This approach is not well suited to the present situation in the large $L$ limit; it predicts $$\frac{\langle \delta\psi^*(z)\delta\psi(z)\rangle}{|\psi_0|^2} \simeq
\frac{1}{6} \kappa_1 L.$$ Although the emergence of a coefficient proportional to $\kappa_1$ is promising, the Bogolubov quadratization procedure is not justified when $L$ exceeds the coherence length of the gas. This is physically not surprising: the finite range of first order field coherence in our one-dimensional geometry is precisely due to large fluctuations of $\psi(z)$ away from $\psi_0$.
Fortunately it is possible to adapt Bogolubov’s idea taking advantage of the weak intensity fluctuations of the field in the large $\chi$ regime. We split the field in a modulus and a phase factor $$\psi(z) = \rho^{1/2}(z)e^{i\theta(z)}.$$ We recall that the intensity $\rho(z)$ and the phase $\theta(z)$ of a field are Hamiltonian conjugate variables in the same way as $\psi(z)$ and $\psi^*(z)$ are. The intensity $\rho(z)$ has only small fluctuations away from the most probable value $\rho_0=|\psi_0|^2=\mu/g$: $$\rho(z) = \rho_0 + \delta\rho(z) \ \ \ \ \mbox{with}\ \ \ \ \
|\delta\rho(z)|\ll \rho_0.$$ The mean value of $\rho(z)$ is simply the linear density of atoms in the transverse ground state $\rho_{\rm lin}^{(0)}$; to lowest order, $\rho_0$ can be identified with this mean density. The phase $\theta$ has on the contrary large fluctuations away from any fixed constant phase $\theta_0$ over distances larger than a few coherence lengths. So contrarily to Bogolubov’s method we do not assume $\theta(z)-\theta_0$ to be small.
The potential energy density of the field $g|\psi|^4/2 - \mu |\psi|^2$ leads to a constant term and a term quadratic in $\delta\rho$. In the kinetic energy density of the field we replace $\rho(z)$ by its lowest order approximation giving a non-zero contribution. Up to a constant, this leads to a quadratic approximation for the Hamiltonian Eq.(\[eq:gpen\]): $$E_{\rm quad} = \int_0^L dz\, \left\{\frac{\hbar^2}{2m}\left[\rho_0
\left(\partial_z\theta\right)^2+\frac{1}{4\rho_0}
\left(\partial_z \delta\rho\right)^2
\right]+\frac{g}{2}\delta\rho^2(z)\right\}.$$ One has finally to diagonalize this quadratic Hamiltonian using canonically conjugate variables. An easy way is to expand the field variables on the eigenmodes of the linear time evolution equations: $$\begin{aligned}
\hbar\partial_t \delta\rho(z) &=& \partial_{\theta(z)} E_{\rm quad} = -\frac{\hbar^2}{m}\rho_0
\partial_z^2\theta(z) \\
-\hbar\partial_t \theta(z) &=& \partial_{\delta\rho(z)} E_{\rm quad} = -\frac{\hbar^2}{4m\rho_0}
\partial_z^2\delta\rho(z)+g\delta\rho(z).\end{aligned}$$ These well known equations have the form of linearized hydrodynamics equations for a superfluid [@PRL_stringari]. Eigenmodes are plane waves of wavevector $k_z$ obeying Eq.(\[eq:genial\]) and with eigenenergy $$\varepsilon_{k_z}= \left[\frac{\hbar^2 k_z^2}{2m}\left(\frac{\hbar^2 k_z^2}{2m}
+2\rho_0 g\right)\right]^{1/2}.$$ This is the famous Bogolubov spectrum; the field energy will then formally appear as a sum of decoupled harmonic oscillators, corresponding to an ideal Bose gas of quasi-particles [@revue].
We can finally reproduce the reasoning performed above in the ideal Bose gas case to identify the validity condition of the classical field approximation. The temperature must be much larger than the mode eigenenergies $\varepsilon_{k_z}$ at wavevectors $k_z=\kappa_1,\kappa_2$ relevant for the correlation functions $g_1,g_2$: $$k_B T \gg \varepsilon_{\kappa_1},\varepsilon_{\kappa_2}.$$ As $\kappa_2\gg\kappa_1$ in the large $\chi$ regime, the condition involving $\kappa_2$ is the most stringent one. From Eq.(\[eq:healing\]) we find formally the same condition as the ideal Bose gas case $$k_B T \gg \mu
\label{eq:voila}$$ with now a different expression for the chemical potential, $\mu \simeq \rho_{\rm lin}^{(0)} g$. We note that in the large $\chi$ regime and the high temperature regime Eq.(\[eq:voila\]) we have the following hierarchy among the various relevant scales of the problem: $$\left(\rho_{\rm lin}^{(0)}\right)^{-1} < \lambda <
\kappa_2^{-1}\simeq \xi < l_c=\kappa_1^{-1} < a$$ where the ‘one-dimensional scattering length’ $a$ is defined as $$g=\frac{\hbar^2}{ma} .$$ The property that the microscopic scale $a$ is [*larger*]{} than the other physical lengths of the problem (with the exception of $L$ of course!) ensures that the one-dimensional Bose gas is a weakly interacting Bose gas. The fact that this is a necessary condition to a classical field approximation (like the Gross-Pitaevskii equation) is known for three-dimensional Bose gases, with the difference that the three-dimensional scattering length $a_{3d}$ has then to be [*smaller*]{} than the macroscopic scales of the gas [@revue]. In particular the small gaseous parameter is $a_{3d}/\xi$ for three-dimensional Bose gases whereas it is $\xi/a$ for one-dimensional Bose gases [@Lieb]. This is not surprising if one realizes that the weak interaction limit $g\rightarrow 0$ leads to $a\rightarrow \infty$ in one-dimension.
Comparison with results in the literature
-----------------------------------------
The model of the one-dimensional interacting Bose gas has already been studied by several authors without performing the classical field approximation.
A first line of thought deals with the exact solvable model of a contact interaction potential [@Lieb; @Gaudin]. To our knowledge a full calculation of $g_1(z)$ and $g_2(z)$ at finite temperature has not been performed. The value of $g_2(0)$ can be found in [@Lieb] at zero temperature. Exact results have been obtained in the strongly interacting regime $g\rightarrow +\infty$: the correlation function $g_1(z)$ has been calculated in [@Tonk] at zero temperature; the correlation functions $g_1(z)$ and $g_2(z)$ at finite temperature have been obtained in [@Efetov]. These regimes are different from the finite temperature, weakly interacting Bose gas considered in the present paper. The extension of the calculation of $g_2(0)$ to finite temperature seams feasible but we do not know any reference.
A second line of thought is to consider the regime of low intensity fluctuations of the field (our large $\chi$ regime): one may take advantage of the weakness of the intensity fluctuations by a linearization of the equations of motion in the hydrodynamic point of view [@Kane; @Haldane; @Monien], a Bogolubov type approach [@Schwartz], a quadratization of action in a path integral formulation [@Efetov; @Popov].
We have checked that the classical field predictions in the asymptotic limit $\chi\gg 1$ reproduce the results of e.g. [@Schwartz] when $(k_B T)^{1/2}\gg
(\rho_{\rm lin}^{(0)}g)^{1/2}$. The advantage of the classical field approximation is that it is not restricted to the large $\chi$ regime so that the transition from the ideal Bose gas to the strongly interacting case can be studied.
In preparing this article we have discovered that a one-dimensional classical field model very similar to our model has been studied in [@escalope], with a different physical motivation.
Conclusion and perspectives {#concl}
===========================
In this paper we have discussed several aspects of our proposal for the production of a continuous ‘atom-laser’ source, consisting in evaporatively cooling an atomic beam in a long magnetic guide. From the classical Boltzmann equation and for expected typical parameters we have estimated the length required to reach the quantum degenerate regime, a few meters for two-dimensional evaporation; the corresponding loss on the atomic flux is only two orders of magnitude.
We have also characterized the coherence properties of the output beam once quantum degeneracy is reached. The gas is expected to experience transverse Bose-Einstein condensation in the guide, leading to a transversally monomode output beam. We have therefore introduced a one-dimensional model for the gas, that we have solved in a classical field approximation. The coherence length of the field along the axis of the magnetic guide can be much larger than the thermal de Broglie wavelength, typically by one order of magnitude. The intensity fluctuations of the beam are very large for an ideal Bose gas but are strongly reduced by repulsive atomic interactions when the healing length $\xi$ of the gas becomes smaller than the coherence length of the gas.
Possible extensions of this work are the discussion of superfluidity properties of the ‘atom-laser’ and a complete three-dimensional modeling of the interacting gas in the magnetic guide.
We acknowledge useful discussions with Gora Shlyapnikov, Philippe Grangier, Gordon Baym and Tony Leggett. We thank Alice Sinatra for helpful comments on the manuscript. Laboratoire Kastler Brossel is a unité de recherche de l’École normale supérieure et de l’Université Pierre et Marie Curie, associée au CNRS.
[99]{} M. H. Anderson, J. Ensher, M. Matthews, C. Wieman, and E. Cornell, 1995, [*Science*]{}, [**269**]{}, 198. C. C. Bradley, C. A. Sackett, and R. G. Hulet, 1997, [*Phys. Rev. Lett.*]{}, [**78**]{}, 985; see also C. C. Bradley et al., 1995, [*Phys. Rev. Lett.*]{}, [**75**]{}, 1687. K. B. Davis, M.O. Mewes, N. Van Druten, D. Durfee, D. Kurn, and W. Ketterle, 1995, [*Phys. Rev. Lett.*]{}, [**75**]{}, 3969. D. Fried, T. Killian, L. Willmann, D. Landhuis, S. Moss, D. Kleppner, and T. Greytak, 1998, [*Phys. Rev. Lett.*]{}, [**81**]{}, 3811. M.R. Andrews, C.G. Townsend, H.-J. Miesner, D.S. Durfee, D.M. Kurn, and W. Ketterle, 1997, [*Science*]{}, [**275**]{}, 637. D. Hall, M. R. Matthews, C. Wiemann and E. A. Cornell, 1998, [*Phys. Rev. Lett.*]{}, [**81**]{}, 1543. I. Bloch, T. Hänsch, T. Esslinger, 2000, [*Nature*]{}, [**403**]{}, 166-170. M.-O. Mewes, M. R. Andrews, N. J. van Druten, D. M. Kurn, D. S. Durfee, and W. Ketterle, 1996, [*Phys. Rev. Lett.*]{}, [**77**]{}, 416; Y. Castin and R. Dum, 1996, [*Phys. Rev. Lett.*]{}, [**77**]{}, 5315; M. Holland, D.S. Jin, M.L. Chiofalo, J. Cooper, 1997, [*Phys. Rev. Lett.*]{}, [**78**]{}, 3801. E.A. Burt, R.W. Ghrist, C.J. Myatt, M.J. Holland, E.A. Cornell, C.E. Wieman, 1997, [*Phys. Rev. Lett.*]{}, [**79**]{}, 337. For experiments on pulsed or quasi-continuous coherent atomic sources extracted from a 3D Bose-Einstein condensate, see: M.-O. Mewes, M. R. Andrews, D. M. Kurn, D. S. Durfee, C. G. Townsend, and W. Ketterle, 1997, [*Phys. Rev. Lett.*]{}, [**78** ]{}, 582; B. P. Anderson and M. Kasevich, 1998, [*Science*]{}, [**282**]{}, 1686 ; E. W. Hagley, L. Deng, M. Kozuma, J. Wen, K. Helmerson, S. L. Rolston, W. D. Phillips, 1999, [*Science*]{}, [**283**]{}, 1706; I. Bloch, T. W. Hänsch, and T. Esslinger, 1999, [*Phys. Rev. Lett.*]{}, [**82**]{}, 3008. For proposals for atom lasers based either on laser cooling, molecular dissociation, or collisions, see: R. J. C. Spreeuw, T. Pfau, U. Janicke, and M. Wilkens, 1995, [*Euro. Phys. Lett.*]{}, [**32**]{}, 469; M. Olshanii, Y. Castin, and J. Dalibard, 1996, A model for an atom laser, [*Proceedings of the XII Conference on Laser Spectroscopy*]{}, edited by M. Inguscio, M. Allegrini, and A. Sasso (Singapore: World Scientific), p.7; H. M. Wiseman and M. J. Collett, 1995, [*Phys. Lett.*]{} A, [**202**]{}, 246; C. Bordé, 1995, [*Phys. Lett.*]{} A, [**204**]{}, 217; J. I. Cirac and M. Lewenstein, 1996, [*Phys. Rev.*]{} A, [**53**]{}, 2466; M. Holland, K. Burnett, C. Gardiner, J. I. Cirac, and P. Zoller, 1996, [*Phys. Rev.*]{} A, [**54**]{}, R1757; H. M. Wiseman, 1997, [*Phys. Rev.*]{} A, [**56**]{}, 2068; B. Kneer, T. Wong, K. Vogel, W. P Schleich, and D. F. Walls, 1998, [*Phys. Rev. A*]{}, [**58**]{}, 4841. E. Mandonnet, A. Minguzzi, R. Dum, I. Carusotto, Y. Castin, J. Dalibard, 2000, [*Eur. Phys. J.*]{} D, [**10**]{}, 9-18. O. Luiten, M. Reynolds, J. Walraven, 1996, [*Phys. Rev.*]{} A, [**53**]{}, 381. A. Bird, 1994, [*Molecular Dynamics and the Direct Simulation of Gas Flows*]{} (Oxford: Clarendon Press). W. Ketterle, N.J. van Druten, 1996, [*Phys. Rev.*]{} A, [**54**]{}, 656. C.W. Gardiner, 1991, [*Quantum noise*]{} (Berlin: Springer-Verlag). See also D.F. Walls and G.J. Milburn, 1994, [*Quantum Optics*]{} (Berlin: Springer-Verlag). P. Nozières and D. Pines, 1990, [*The Theory of Quantum Liquids*]{}, vol. 2 (Redwood City: Addison-Wesley). See also F. Dalfovo, S. Giorgini, L.P. Pitaevskii, S. Stringari, 1999, [*Rev. Mod. Phys.*]{}, [**71**]{}, 463-512. Kerson Huang, 1963, [*Statistical Mechanics*]{} (New York: John Wiley & Sons), p. 275. Y. Castin, to be published, Lecture Notes of Les Houches Summer School 1999, chapter 3. This formula holds if the spatial width of the ground state of the transverse harmonic oscillator remains larger than the three-dimensional scattering length. For a more complete discussion, see M. Olshanii, 1998, [*Phys. Rev. Lett.*]{}, [**81**]{}, 938. E.H. Lieb and W. Liniger, 1963, [*Phys. Rev.*]{}, [**130**]{}, 1605. A review for $g>0$ is given in M. Gaudin, 1983, [*La fonction d’onde de Bethe*]{} (Paris: Masson). R.P. Feynman, and A.R. Hibbs, 1965, [*Quantum Mechanics and paths integrals*]{} (New York: McGraw-Hill). As the length $L$ appears in the quantum mechanical formulation only through $\exp[-L {\cal H}/\hbar]$ the previous intuitive assumption can be cast in the mathematical condition: $L (\varepsilon_{\rm exc } -\varepsilon_0^{m=0})/\hbar \gg 1$ where $\varepsilon_{\rm exc }$ is the energy of the first excited state of ${\cal H}$. As we shall see this results in $\kappa_1 L\gg 1$ where $1/\kappa_1$ is the coherence length of the gas. S. Stringari, 1996, [*Phys. Rev. Lett.*]{}, [**77**]{}, 2360. H.G. Vaidya and C.A. Tracy, 1979, [*Phys. Rev. Lett.*]{}, [**42**]{}, 3; a corrected version of the expansions can be found in M. Jimbo, T. Miwa, Y. Mori, and M. Sato, 1980, [*Physica (Amsterdam)*]{}, [**1D**]{}, 80. K.B. Efetov, A.I. Larkin, 1976, [*Sov. Phys. JETP*]{}, [**42**]{}, 390. J.W. Kane and L.P. Kadanoff, 1967, [*Phys. Rev.*]{}, [**155**]{}, 80. F.D.M. Haldane, 1981, [*Phys. Rev. Lett.*]{}, [**47**]{}, 1840. H. Monien, M. Linn, N. Elstner, 1998, [*Phys. Rev.*]{} A, [**58**]{}, R3395. M. Schwartz, 1977, [*Phys. Rev.*]{} B, [**15**]{}, 1399. V.N. Popov, 1983, [*Functional Integrals in Quantum Field Theory and Statistical Physics*]{} (Dordrecht: D. Reidel Pub.). D.J. Scalapino, M. Sears, R.A. Ferrell, 1972, [*Phys. Rev.*]{} B, [**6**]{}, 3409.
[^1]: E-mail: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We consider the quantum first detection problem for a particle evolving on a graph under repeated projective measurements with fixed rate $1/\tau$. A general formula for the mean first detected transition time is obtained for a quantum walk in a finite-dimensional Hilbert space where the initial state $|\psi_{\rm in}\rangle$ of the walker is orthogonal to the detected state $|\psi_{\rm d}\rangle$. We focus on diverging mean transition times, where the total detection probability exhibits a discontinuous drop of its value, by mapping the problem onto a theory of fields of classical charges located on the unit disk. Close to the critical parameter of the model, which exhibits a blow-up of the mean transition time, we get simple expressions for the mean transition time. Using previous results on the fluctuations of the return time, corresponding to $|\psi_{\rm in}\rangle = |\psi_{\rm d}\rangle$, we find close to these critical parameters that the mean transition time is proportional to the fluctuations of the return time, an expression reminiscent of the Einstein relation.'
author:
- 'Q. Liu'
- 'R. Yin'
- 'K. Ziegler'
- 'E. Barkai'
title: 'Quantum walks: the first detected transition time'
---
\[sec:level1\]Introduction
==========================
![Schematic plot of the first detected transition problem in quantum walks. The quantum particle is prepared at the initial state (black cat) at $t=0$ and evolves unitarily (gray cats) in the detection free interval $\tau$. And measurements (the magnifying glass) are performed every $\tau$ units of time. Here $-$/$+$ means before/after measurement. In the failed attempts $t=\tau$, the detector collapses the wave function at the detected state. We repeat this process until the quantum particle is successfully detected (for example, here $t=2\tau$). The question is how long it takes to find the quantum particle.[]{data-label="fig:quantumwalks"}](fig1.pdf){width="0.99\columnwidth"}
A closed quantum system is prepared in some initial state and evolves unitarily over time. Our aim is to monitor the evolution of this system by repeated projective measurements until a certain state is detected for the first time. A corresponding simple classical [@Redner2001; @ralf2014first] example would be to take a picture of a rare animal in the wilderness. For this purpose a remote camera takes pictures at a fixed rate, and the camera’s software checks immediately whether the rare animal is on the last picture or not. Once the animal is caught on the last snapshot the process stops. It is obvious that we may miss the first appearance of the animal in the process. But when we continue long enough we might be lucky. The theoretical question is, what would be “long enough" to detect the animal at a given measurement rate?
Quantum walks are well investigated both theoretically and experimentally [@Aharonov1993; @Farhi1998; @Childs2002; @Karski174; @PhysRevLett.104.100503; @Preiss1229]. Also the quantum first detection problem, for a quantum walk on a graph has been considered in detail [@BACH2004562; @Krovi2006; @PhysRevA.74.042334; @Varbanov2008; @Gruenbaum2013; @Grunbaum2014; @Krapivsky2014; @Dhar2015; @Dhar_2015; @Friedman_2016; @Friedman2017; @Thiel2018] , as part of a wider investigation of unitary evolution pierced by measurements [@PhysRevB.98.104309; @PhysRevE.98.022129; @belan2019optimal; @PhysRevA.99.062105; @benzion2019disentangling; @zabalo2019critical; @PhysRevX.9.031009; @roy2019measurementinduced]. The rate $1/\tau$ at which we detect the particle on a given site becomes a crucial parameter, for example, if we sample too fast the “animal” cannot be detected at all due to the Zeno effect. This implies that there exist special sampling times that are optimal, in the sense that the detection time attains a minimum. Indeed it was shown by Krovi and Brun [@Krovi2006; @PhysRevA.74.042334; @Varbanov2008] that on certain graphs, due to constructive interference, the quantum search problem is highly efficient. At the same time, these authors noted that in other cases, destructive interference may render the quantum search inefficient in the sense that the hitting time even for a small system can be infinity (unlike classical random walks on a finite graph). In this paper we use a recently proposed quantum renewal equation [@Friedman2017] to find the average time of a quantum walker starting on $|\psi_{\rm in}\rangle$ to be detected on $|\psi_{\rm d}\rangle$.
We employ stroboscopic sampling, which allows for considerable theoretical advance, with generating function technique. It is hoped that in the long run, this type of investigation will lead to advances in quantum search algorithms [@PhysRevLett.79.325; @PhysRevA.70.022314; @BACH2004562; @Kempe; @PhysRevLett.116.100501]. More importantly, in this work we map the problem of calculating the averaged transition time to a classical charge theory. We show how the mean quantum transition time is related to the stationary points of a set of classical charges positioned on the unit circle in the complex plane with locations $\exp(i E_j \tau)$. This charge picture was previously promoted in the context of the return problem [@Gruenbaum2013] ($|\psi_{\rm in}\rangle = |\psi_{\rm d}\rangle$), while here we use this method to solve the transition problem. These two problems exhibit vastly different behavior. For the return problem the mean return time is quantized, since it is given as a topological invariant which is the winding number of the wave function [@Gruenbaum2013; @Yin2019]. In our problem this is equal to the dimensionality of the underlying Hilbert space with non-degenerate eigenvalues of the back-folded spectrum. Thus, the average return time is independent of the sampling rate. In contrast, the transition time is very sensitive, for instance, to the sampling rate $1/\tau$, and its behaviors are highly non-trivial [@Friedman2017].
The rest of this paper is organized as follows: In Sec. \[model and formalism\] we define our model and the degeneracies caused by the sampling time $\tau$. Then we derive our first main result the mean first detected transition (FDT) time in Sec. \[First detected transition time\]. We find the general relation of the transition time and return time fluctuations in Sec. \[The detected transition time and return fluctuations\]. In Secs. \[Weak charge\],\[Two merging charges\],\[time and fluctuation relation\], \[Big charge theory\] we study some characteristic diverging transition times, where special relations for the transition time and the return fluctuations are found. This includes some examples to confirm our theory. We close the paper with discussions and a summary in Sec. \[discussion\]. Detailed calculations are presented in the appendices.
Model and Formalism\[model and formalism\]
==========================================
Stroboscopic Protocol
---------------------
We consider a quantum particle prepared in the state $|\psi_{\rm in}\rangle$, for instance on a node of the lattice or other graphs. The evolution of this quantum particle is described by the time-independent Hamiltonian $H$ according to the Schrödinger equation. As an example consider a one-dimensional tight-binding model in discrete position space with nearest neighbor hopping: $$H=-\gamma \sum_{x=-N}^{N}(|x\rangle\langle x+1|+|x+1\rangle\langle x|).$$ However, our general formalism does not rely on a specific Hamiltonian, as long as we are restricted to a finite-dimensional Hilbert space.
In a measurement the detector collapses the wave function at the detected state $|\psi_{\rm d}\rangle$ by the projection operator $D=|\psi_{\rm d}\rangle\langle\psi_{\rm d}|$. For simplicity one may assume that $|\psi_{\rm d}\rangle$ is yet another localized node state of the graph, however our theory is developed in generality. We perform the measurements with a discrete time sequence $\tau,2\tau,\cdots,n\tau$ until it is successfully detected for the first time. Then the result of the measurements is a string: “no, no, $\cdots$,no, yes”. In the failed measurements the wave function collapses to zero at the detected state, and we renormalize the wave function after each failed attempt. The event of detecting the state $| \psi_{\rm d}\rangle$ for the first time after $n$ attempts implies that $n-1$ previous attempts failed and this certainly must be taken into consideration. Namely the failed measurements back fire and influence the dynamics, by erasing the wave function at the detected state. Finally, the quantum state is detected and the experiment is concluded (see Fig. \[fig:quantumwalks\]). Hence the first detection time is $t= n \tau$.
The key ingredients of this process are the initial state $|\psi_{\rm in}\rangle$ and the detected state $|\psi_{\rm d}\rangle$, which characterize this repeated measurements problem. If the initial state is the same as the detected state, namely $\langle\psi_{\rm in}|\psi_{\rm d}\rangle=1$ we call this case the first detected return (FDR) problem, which has been well studied by a series of works [@Gruenbaum2013; @Stefanak2008; @Xue2015; @Dhar2015; @Yin2019]. In the following we investigate the FDT problem, where $\langle\psi_{\rm in}|\psi_{\rm d}\rangle=0$. This transition problem describes the transfer of the quantum state from $|\psi_{\rm in}\rangle$ to $|\psi_{\rm d}\rangle$ in Hilbert space. The time this process takes is of elementary importance. Since the results in each experiment are random, we focus on the expected FDT time $\langle n \rangle\tau$, which gives the average quantum transition time in the presence of the stroboscopic measurements.
During each time interval $\tau$ the evolution of the wave function is unitary $|\psi(n\tau^-)\rangle=U(\tau)|\psi[(n-1)\tau^+]\rangle$, where $U(\tau)=\exp{(-i H \tau)}$ (we set $\hbar=1$ in this paper) and $-$/$+$ means before/after measurement. Let $\phi_n$ be the FDT amplitude, the probability of the FDT in the $n$-th measurement is $F_n=|\phi_n|^2$. If the particle is detected with probability one (see further details [@Thiel2019]), which means $\sum_{n=1}^{\infty}|\phi_n|^2=1$, the mean FDT time is $\langle t \rangle=\tau\sum_{n=1}^{\infty}n|\phi_n|^2$. As we will soon recap, $\phi_n$ can be evaluated from a unitary evolution interrupted by projective measurements. However, there exist a deep relation between $\phi_n$ and the unitary evolution without interrupting measurement.
Brief summary of the main results
---------------------------------
Before we start with the general discussion of the evolution of a closed quantum system under repeated measurements, we would like to summarize the main results: Repeated measurements interrupt the unitary evolution by a projection after a time step $\tau$. This has a strong effect on the dynamical properties, which can be observed in the transition amplitude $\phi_{n}$ of Eq. (\[phi origen\]). The unitary evolution $\exp{(-i H \tau)}$ is controlled by the energy spectrum. The overlaps $\{p_k\}$ and $\{q_k\}$ in Eqs. (\[qk\],\[pk\]) are crucial in that they connect the eigenstates of $H$ and the initial and measured states. The non-unitary evolution is characterized by the zeros of the polynomial Eq. (\[D\]) and these overlap functions. Those zeros are formally related to a classical electrostatic problem [@Gruenbaum2013]; namely they are the stationary points of a test charge in a system with charges on the unit circle, which is defined in Eq. (\[Force field\]). After solving this electrostatic problem, the zeros are used to calculate, for instance, the first detection amplitude with Eq. (\[phi n\]), the divergent behavior of the mean FDT time near degenerate points in Eq. (\[mean n excat\]), and a generalized Einstein relation between the mean FDT time and the FDR variance in Eqs. (\[n and var\],\[einstein1\]). This leads us to the conclusion that the mean FDT time, i.e. the mean time to reach a certain quantum state, is very sensitive to the time step $\tau$ of the measurements. In particular, degeneracies of the back-folded spectrum in Eq. (\[back\_folded\_spectrum\]) can lead to extremely long times for the detection of certain quantum states. Based on this general approach, we have calculated the mean FDT for a two-level system in Eq. (\[n two-level\]), for a Y-shaped molecule in Eq. (\[n Y shape\]), and for a Benzene-type ring in Sec. \[sect:ring\].
Generating function
-------------------
The FDT amplitude $\phi_{t,n}$ for the evolution from $| \psi_{\rm in}\rangle$ to $| \psi_{\rm d}\rangle$ and the FDR amplitude $\phi_{r,n}$ for the evolution from $| \psi_{\rm d}\rangle$ to $| \psi_{\rm d}\rangle$ read [@Gruenbaum2013; @Dhar2015; @Friedman_2016; @Friedman2017] $$\phi_{t,n}=\langle \psi_{\rm d}| (e^{-i \tau H }P)^{n-1}e^{-i \tau H}| \psi_{\rm in}\rangle,
\label{phi origen}$$ $$\phi_{r,n}=\langle \psi_{\rm d}| (e^{-i \tau H }P)^{n-1}e^{-i \tau H}| \psi_{\rm d}\rangle,
\label{phi origen r}$$ with $P=1-D=1-|\psi_{\rm d}\rangle\langle\psi_{\rm d}|$. As the equations show, the unitary evolution in the detection free interval $\tau$ is interrupted by the operation $P$. The combined unitary evolution and the projection goes with the power $n-1$, corresponding to the $n-1$ prior failed measurements. Moreover, we define the unitary transition amplitude $v_n$ and the unitary return amplitude $u_n$ as $$v_n=\langle \psi_{\rm d} | e^{-i n H \tau} | \psi_{\rm in}\rangle ,
\label{vn}$$ $$u_n=\langle \psi_{\rm d}| e^{-i n H \tau}| \psi_{\rm d}\rangle .
\label{un}$$ These amplitudes describe transitions from the initial state to the detected state and from the detected state back to itself, free of any measurement. Using the $v_n$ and $u_n$, we expand Eq. (\[phi origen\]) and (\[phi origen r\]) in $P$ which leads to an iteration equation known as the quantum renewal equation [@Friedman_2016; @Friedman2017]: $$\begin{aligned}
\phi_{t,n} &=& v_n-\sum_{j=1}^{n-1}\phi_{t,j} u_{n-j},\label{quantum renewal}\\
\phi_{r,n} &=& u_n-\sum_{j=1}^{n-1}\phi_{r,j} u_{n-j}.\label{quantum renewal r}\end{aligned}$$ Note that the first terms $v_n$, $u_n$ on the right-hand side describe the unitary evolution between the initial state and the detected state and between the detected state to itself. The second terms describe all the former wave function returns to the detected state. These recursive equations, together with the exact function Eq. (\[mean n excat\]) for mean transition times, are used in the example section to find exact solutions of the problem. In order to solve the recursive equations a direct method is to transform the quantum renewal equation into the frequency (or $\omega$) space. Since the renewal equations consist of $\{ v_j\}$ and $\{ u_j\}$, we need to transform these quantities into $\omega$ space first. Using Eqs. (\[vn\],\[un\]) we have $$\hat{v}( \omega):=\sum_{n=1}^{\infty}e^{i n \omega}v_n=\langle \psi_{\rm d}|(e^{i \tau H-i \omega}-1)^{-1}|\psi_{\rm in}\rangle,$$
$$\hat{u}( \omega):=\sum_{n=1}^{\infty}e^{i n \omega}u_n=\langle \psi_{\rm d}|(e^{i \tau H-i \omega}-1)^{-1}|\psi_{\rm d}\rangle.$$
The analogous calculation for the amplitudes $\phi_{t,n}$, $\phi_{r,n}$ leads to $$\hat{\phi}_t(\omega)\equiv\sum_{n=1}^{\infty}e^{i \omega n}\phi_{t,n}=\langle \psi_{\rm d}| A_{\omega}| \psi_{\rm in}\rangle,$$ $$\hat{\phi}_r(\omega)\equiv\sum_{n=1}^{\infty}e^{i \omega n}\phi_{r,n}=\langle \psi_{\rm d}| A_{\omega}| \psi_{\rm d}\rangle.$$ where $A_{\omega}=(e^{i \tau H-i \omega}-P)^{-1}$. The initial state $|\psi_{\rm in}\rangle$ distinguishes the return and transition problem. $A_{\omega}$ is related to the Green’s function $(e^{i H\tau}/z-P)^{-1}$ of the non-unitary evolution [@Thiel2019a]. Its poles are the solutions of $\text{det}(\mathbb{1}/z-P U(\tau))=0$. We will see later that these poles are essential for the evaluation of the mean FDT time. This property implies that the repeated measurement protocol can be possibly related to open quantum systems, in the sense that the measurements acting on the system is equivalent to the interaction between environment and the system [@PhysRevLett.123.140403; @s2019complex]. Thus we believe that further research on this topic is worth while.
Using the identity $(1+B)^{-1}=1-B(1+B)^{-1}$, we obtain $$\langle \psi_{\rm d}| A_{\omega}| \psi_{\rm in}\rangle=\hat{v}( \omega)-\hat{u}( \omega)\langle \psi_{\rm d}| A_{\omega}| \psi_{\rm in}\rangle,$$ $$\langle \psi_{\rm d}| A_{\omega}| \psi_{\rm d}\rangle=\hat{u}( \omega)-\hat{u}( \omega)\langle \psi_{\rm d}| A_{\omega}| \psi_{\rm d}\rangle.$$ Then the generating functions for the amplitude $\phi_t$ and $\phi_r$ read $$\hat{\phi}_t(\omega)=\frac{\hat{v}(w)}{1+\hat{u}(w)},\quad \hat{\phi}_r(\omega)=\frac{\hat{u}(w)}{1+\hat{u}(w)}.$$ In the return problem, the initial state and detected state coincide, so the generating function only contains $\hat{u}(w)$. Whereas in the transition problem the symmetry is broken leading to the term $\hat{v}(w)$ in the numerator.
A continuation of the phase factor $\exp{(i\omega)}$ from the unit disk to the parameter $z$ in the complex plane is convenient for further calculations. This leads to [@Friedman2017] $$\hat{\phi}_t(z)=\frac{\langle \psi_{\rm d}| \hat{U}(z)| \psi_{\rm in}\rangle}{1+\langle \psi_{\rm d}| \hat{U}(z)| \psi_{\rm d}\rangle},
\label{generating}$$ $$\hat{\phi}_r(z)=\frac{\langle \psi_{\rm d}| \hat{U}(z)| \psi_{\rm d}\rangle}{1+\langle \psi_{\rm d}| \hat{U}(z)| \psi_{\rm d}\rangle},
\label{generating r}$$ where $\hat{U}(z)=\sum_{n=1}^{\infty}z^n U(n\tau)=z U(\tau) /(\mathbb{1}-z U(\tau))$ is the Z (or discrete Laplace) transform of $U(n\tau)$. The difference between Eq. (\[generating\]) and Eq. (\[generating r\]) is again only the numerator.
Pseudo Degeneracy {#sect:p_degeneracy}
-----------------
The degeneracy of the energy levels plays a crucial role in the problem. For instance, a geometric symmetry of the graph can introduce such degeneracies. What is special here is that the measurement period $\tau$ leads to a new type of degeneracy of the distinct energy levels. This degeneracy is rooted in the stroboscopic sampling under investigation.
For an arbitrary Hamiltonian $H$ which has $w$ non-degenerate energy levels, the eigenvalues $\{E_k\}_{k=0,...,w-1}$ of the Hamiltonian $H$ and the corresponding eigenstates $\{| E_{k l}\rangle\}_{k=0,...,w-1}$ with $1 \leqslant l \leqslant g_k$, where $g_k$ is the degeneracy, can be used to express the matrix elements of Eq. (\[vn\]) and Eq. (\[un\]) in spectral representation as $$\begin{aligned}
v_n &=& \sum_{k=0}^{w-1}\bigg\{\sum_{l=1}^{g_k}\langle \psi_{\rm d}| E_{k l}\rangle\langle E_{k l}| \psi_{\rm in}\rangle\bigg\} e^{-i n E_k \tau},\\
u_n &=& \sum_{k=0}^{w-1}\bigg\{\sum_{l=1}^{g_k}|\langle \psi_{\rm d}| E_{k l}\rangle|^2\bigg\} e^{-i n E_k \tau}.\end{aligned}$$ These expressions are invariant under the change $ E_k\tau \rightarrow E_k\tau+2\pi j $ for integer $ j $. Thus, the eigenvalues $E_k$, $ E_{k^{\prime}} $ are effectively degenerate if $E_k\tau= E_{k^{\prime}}\tau+2\pi j $. Therefore, rather than the scaled eigenvalues $\{E_k \tau\}$ (which will be called simply eigenvalues subsequently), the back-folded eigenvalues $\{\Bar{E_k}\tau\}$ $$\Bar{E_k}\tau=E_k\tau (\text{mod}\quad 2\pi) \quad -\pi \leqslant \Bar{E_k}\tau < \pi,
\label{back_folded_spectrum}$$ determine the dynamics at fixed $\tau$. This can also be understood as the mapping $E_k\tau\rightarrow e^{-i E_k\tau} $ from the real axis to the unit circle on the complex plane [@Gruenbaum2013] (see Fig. \[fig:energy mapping\]). Here it is possible to change the value of $\tau$ until $\tau=\tau_c$ which leads to [@Gruenbaum2013; @Friedman2017; @Thiel2019] $$| E_k-E_i| \tau_c=2\pi j ,
\label{excptional points}$$ where $j$ is an integer. Thus, there are degeneracies of the back-folds eigenvalues for this critical $\tau_c$. Since the back-folded spectrum is relevant for the FDR/FDT and not the spectrum of $H$ itself, these degeneracies affect the discrete dynamics, even if the eigenvalues $\{E_k\}$ of $H$ are non-degenerate.
The quantum problem has a classical counterpart known as the first passage problem. The two problems exhibit vastly different behaviors, as might be expected. Let $P_{\rm det} =\sum_{n=1}^{\infty}|\phi_n|^2$ be the total detection probability. Unlike classical random walks on finite graphs, here one can find the total detection probability less than unity. The quantum particle will go to some “dark states”, where they will never be detected [@Thiel2019; @Thiel2019a; @Thiel2019b].
In Ref. [@Thiel2019; @Thiel2019a] it was shown that $P_{\rm det}<1$ when the Hilbert space is split into two subspaces dark and bright. The dark states can arise either from degeneracies of the energy spectrum or from energy levels that have no overlap with the detected state. The main focus of this paper is on cases where the total detection probability is unity (otherwise the search is clearly not efficient). Thus In our system we have $P_{\rm det}=1$, except for special sampling times, given by Eq. (\[excptional points\]). On these sampling times the detection probability is sub-optimal. Close to these sampling times the mean time for detection will diverge, and one of our goals is to understand this behavior.
![Schematic behaviors of (A) the scaled Hamiltonian spectrum $E_k\tau$ and (B) the phase $e^{-i E_k\tau}$ under a change of the sampling time $\tau$. The arrows indicate the movements of the scaled energy levels ($E_k\tau$) when increasing $\tau$. The positive (blue dots) and the negative (red dots) energy levels are well separated in (A). After mapping to the unit circle $E_k\tau\rightarrow e^{-i E_k \tau}$ they are not separated all the time, moving on the unit circle making fusion of the phases possible. In particular, the case (right) can lead to degeneracies in the back-folded spectrum and to very large mean transition times.[]{data-label="fig:energy mapping"}](fig2.pdf){width="0.8\columnwidth"}
Zeros and Poles
---------------
From $\hat{\phi}(z)=\sum_{n=1}^{\infty}z^n \phi_n$ we extract the amplitude $\phi_n$ by the inverse transformation [@Friedman2017] $$\phi_n=\frac{1}{2\pi i}\oint_{\gamma}\hat{\phi}(z)z^{-n-1}dz,
\label{phi-n}$$ where $\gamma$ is a counterclockwise closed contour around the circle of the complex plane with $|z|<1$, where $\hat{\phi}(z)$ is analytic. To perform the integration, we must analyze $\hat{\phi}(z)$. In Eqs. (\[generating\],\[generating r\]) the denominators only contain the state $|\psi_{\rm d}\rangle$ and not the initial condition $|\psi_{\rm in}\rangle$, for both the FDR and FDT case. The poles outside the unit disc in turn will determine the relaxation pattern of $\phi_n$ (see below). To progress in our study of the transition problem we will use recent advances on the properties of the return problem [@Gruenbaum2013; @Yin2019]. For this purpose we study the connection between the return and the transition problem more explicitly. First, we define the overlap functions $p_k$ and $q_k$ of the initial/detected state as $$\begin{aligned}
q_k &=& \sum_{l=1}^{g_k}\langle \psi_{\rm d}| E_{kl}\rangle\langle E_{kl}| \psi_{\rm in}\rangle,\label{qk}\\
p_k &=& \sum_{l=1}^{g_k}|\langle \psi_{\rm d}| E_{kl}\rangle|^2 ,
\label{pk}\end{aligned}$$ which correspond to the distinct energy level $E_k$ with degeneracy $g_k$. $q_k$ contains both detected and initial states while $p_k$ is only related to $|\psi_{\rm d}\rangle$. These expressions indicate that $p_k$ is real and non-negative while $q_k$ is complex. The normalization of the energy eigenstates imply $\sum_{k=0}^{w-1}p_k=1$. On the other hand, $\sum_{k=0}^{w-1}q_k=0$, since the initial state and detected state are assumed to be orthogonal in the transition problem.
Next, we write the generating function in spectral representation as before, using eigenstates $|E_{kl}\rangle$ and the corresponding $g_k$-folded eigenvalues $E_k$. By multiplying both numerator and denominator $\prod_{k=0}^{w-1}(e^{i E_k \tau}-z)$, we express $\hat{\phi}_t(z)$ and $\hat{\phi}_r(z)$ as $$\hat{\phi}_t(z)=\frac{{\cal N}_t(z)}{{\cal D}(z)},\qquad \hat{\phi}_r(z)=\frac{{\cal N}_r(z)}{{\cal D}(z)}.$$ Using $q_k$ and $p_k$ we can express ${\cal N}_t(z)$, ${\cal N}_r(z)$ and ${\cal D}(z)$ as $$\begin{aligned}
{\cal N}_t(z) &=& z\sum_{i=0}^{w-1} q_i \bigg[\prod_{k=0, k\neq i}^{w-1}(e^{i E_k \tau}-z)\bigg]\label{Nt},\\
{\cal N}_r(z) &=& z\sum_{i=0}^{w-1} p_i \bigg[\prod_{k=0, k\neq i}^{w-1}(e^{i E_k \tau}-z)\bigg]\label{Nr},\\
{\cal D}(z) &=& \sum_{i=0}^{w-1} p_i e^{i E_i \tau} \bigg[\prod_{k=0, k\neq i}^{w-1}(e^{i E_k \tau}-z)\bigg].
\label{D}\end{aligned}$$ The only difference between the ${\cal N}_r(z)$ and ${\cal N}_t(z)$ is that the $q_i$ in the former is replaced with $p_i$ in the latter. So $p_i$ and $q_i$ characterize the generating function of the return and the transition problem. ${\cal N}_r(z)$ and ${\cal D}(z)$ share the same multiplication term, each depending on the same group of real numbers $p_i$ and $\{E_i\}$. A straightforward calculation shows that the two polynomials are related [@Friedman2017]: $${\cal D}(z)=(-1)^{w-1} e^{i\sum_j E_j\tau}z^w[{\cal N}_r(1/z^*)]^{\ast}.
\label{relation D N}$$
From Eqs. (\[generating\],\[generating r\]) the poles of the return and transition problem are identical. These poles, denoted by $Z_{i}$, are found from the solutions of ${\cal D}(Z)=0$. We also define the zeros of the generating function in the return problem, denoted by $z_{r,i}$. The latter are given by ${\cal N}_r(z)=0$. From Eq. (\[relation D N\]), ${\cal D}(z)=(-1)^{w-1} e^{i\sum_j E_j\tau}z^w[{\cal N}_r(1/z^*)]^{\ast}=0$ yields $Z_{i}=1/z_{r,i}^*$. Hence transition poles $Z_{i}$ are given by $$Z_{i}=\frac{1}{z_{r,i}^{\ast}},\quad z_{r,i}\neq 0.
\label{relation of zero and pole}$$ The key point is that the $\{Z_i\}$ describe both the transition problem investigated here and the return problem [@Gruenbaum2013]. Subsequently, we write $z_{r,i}$ as $z_{i}$ for simplicity. Eq. (\[relation of zero and pole\]) gives us a way to find the poles $Z_i$ which are essential for the amplitude $\phi_n$, namely using the return zeros $z_i$, which have been studied already in the return problem [@Gruenbaum2013; @Yin2019].
Charge Theory {#sect:charge_th}
-------------
As already discussed before, the central goal is to determine the zeros $\{ z_i\}$. A very helpful method in this regard was proposed by Grünbaum *et al.* [@Gruenbaum2013], who mapped the return problem to a classical charge theory. More importantly, the classical charge theory provides an intuitive physical picture from which we can understand the behavior of the poles. Using Eq. (\[Nr\]) for the zeros of ${\cal N}_r(z)$ with some rearrangement, we have $z\sum_{k=0}^{w-1}p_k/(e^{i E_k\tau}-z)=0$. Neglecting the trivial zero at the origin we must solve $${\cal F}(z)=\sum_{k=0}^{w-1}\frac{p_k}{e^{i E_k\tau}-z}=0.
\label{Force field}$$ ${\cal F}(z)$ can be considered as a force field in the complex plane, stemming from charges $p_k$ whose locations are $e^{i E_k\tau}$ on the unit circle. Then the zeros $\{z_i\}$ of ${\cal N}_r(z)$ are the stationary points of this force field. Since there are $w$ charges which corresponds to the number of the discrete energy levels, we get $w-1$ stationary points in this force field from Eq. (\[Force field\]). All the zeros are inside the unit disc, which is rather obvious since all the charges have the same sign ($p_k>0$). The physical significance of this is that the modes of the problem decay. More precisely, the zeros are within a convex hull, whose edge is given by the position of the charges, hence $|z_i|<1$. Then Eq. (\[relation of zero and pole\]) implies $|Z_i|>1$, i.e. the poles lie outside the unit circle.
FDT time\[First detected transition time\]
==========================================
In this section we focus on finding the general expression for the mean FDT time. We assume $\langle\psi_{\rm d}|\psi_{\rm in}\rangle=0$ which is the definition of “transition". Since $\langle t \rangle=\tau\langle n \rangle=\tau\sum_{n=1}^{\infty}n|\phi_n|^2$, the first step is to find the amplitudes $\phi_n$, describing the detection probability for the $n$-th attempt. We start from the generating function of the FDT problem Eq. (\[generating\]): $$\hat{\phi}_t(z)=\frac{z\sum_{i=0}^{w-1} q_i \bigg[\prod_{k=0, k\neq i}^{w-1}(e^{i E_k \tau}-z)\bigg]}
{\sum_{i=0}^{w-1} p_i e^{i E_i \tau} \bigg[\prod_{k=0, k\neq i}^{w-1}(e^{i E_k \tau}-z)\bigg]},$$ The numerator ${\cal N}_t(z)$ reads with the polynomial ${\cal G}(z)$ $${\cal N}_t(z)=z\sum_{i=0}^{w-1} q_i \bigg[\prod_{k=0, k\neq i}^{w-1}(e^{i E_k \tau}-z)\bigg]=z{\cal G}(z).$$ Using $\sum_i q_i = 0$, it is not difficult to show that $\text{deg}({\cal D}(z))>\text{deg}({\cal G}(z)) $ (see details in Appendix \[order\]). We rewrite the generating function by “general partial decomposition" for isolated poles of the denominator and a polynomial ${\cal G}(z)$ of order smaller than $w-1$. Using the $w-1$ poles $\{ Z_i\}$ we found before, we rewrite ${\cal D}(z)=\beta(z-Z_1)(z-Z_2)\cdots(z-Z_{w-1})$ ($\beta$ is the coefficient of $z^{w-1}$, see Appendix \[order\]). Then we obtain $$\frac{{\cal G}(z)}{\beta(z-Z_1)...(z-Z_{w-1})}=\sum_{i=1}^{w-1}\frac{C_i}{Z_i(z-Z_i)} ,
\label{C_i 32}$$ where $C_i$ is given by $$\begin{aligned}
C_i & =\frac{Z_i}{2\pi i}\oint_{\gamma_i}\frac{{\cal G}(z)}{\beta(z-Z_1)\cdots(z-Z_{w-1})}d z \nonumber \\
& =\frac{{\cal N}_t(Z_i)}{\beta}\prod_{k\neq i}\frac{1}{Z_i-Z_k}.\end{aligned}$$ The contours $\gamma_i$ enclose only $Z_i$ but not $\{Z_k\}_{k\neq i}$. Since $Z_i$ is the pole of $[{\cal D}(z)]^{-1}$, we can rewrite the multiplication as $\beta^{-1}\prod_{k\neq i}(Z_i-Z_k)^{-1}=[\partial_z {\cal D}(z)]^{-1}|_{z=Z_i}$, hence $$C_i=\frac{{\cal N}_t(Z_i)}{\partial_z {\cal D}(z)|_{z=Z_i}}.
\label{C_i general}$$ This allows us to rewrite the generating function as $\hat{\phi}_t(z)=\sum_{i=1}^{w-1}z C_i/[Z_i(z-Z_i)]$, where $\hat{\phi}_t(z)$ is decomposed into the summation of the $z C_i/[(z-Z_i)Z_i]$ in which there is only one pole in the denominator. With Eq. (\[phi-n\]) the first detection amplitude reads $$\phi_n=\sum_{i=1}^{w-1}\frac{C_i}{2\pi i}\oint_{\gamma}\frac{z^{-n}}{Z_i(z-Z_i)}dz=-\sum_{i=1}^{w-1}C_i Z_i^{-n-1}.
\label{phi n}$$ The probability of finding the quantum state $|\psi_d\rangle$ at the $n^{\rm th}$ attempt is $F_n=|\phi_n|^2$. Summing the geometric series the total detection probability $P_{\rm det}=\sum_{n=1}^{\infty}F_n$ is $$P_{det}=\sum_{i,j=1}^{w-1}\frac{C_i C_j^{\ast}}{(Z_i Z_j^{\ast}-1)Z_i Z_j^{\ast}}.
\label{P det excat}$$ As mentioned before, other methods for finding $P_{\rm det}$ were considered in Ref. [@Thiel2019]. For a finite system, it was shown that $P_\text{det}$ is independent of the measurement interval $\tau$ except for the special resonant points in Eq. (\[excptional points\]) where new degeneracy appears. In finite-dimensional Hilbert space, the total detection probability is $P_{\rm det}=1$ when all the energy levels have projection on the detected state and the back-folded spectrum is not degenerate.
If the total detection probability is one, the detection of the quantum state in an experiment is guaranteed. We can define the mean FDT time $\langle t \rangle=\langle n \rangle \tau$, where $\langle n \rangle$ is the mean of the number of detection attempts. For convenience, we call $\langle n \rangle$ the mean of FDT time in the rest of the paper due to the simple relation between the $\langle t \rangle$ and $\langle n \rangle$. From $\langle n \rangle=\sum_{n=1}^{\infty}n| \phi_n | ^2$, together with Eq. (\[phi n\]), we find $$\langle n \rangle=\sum_{i,j=1}^{w-1}\frac{C_i C_j^{\ast}}{(Z_i Z_j^{\ast}-1)^2}.
\label{mean n excat}$$ Eqs. (\[Force field\],\[C\_i general\],\[mean n excat\]) expose how the mean FDT time depends on the spectrum of $H$, the initial state $|\psi_{\rm in}\rangle$, the detected state $|\psi_{\rm d}\rangle$ and the sampling time $\tau$. Since in general the denominator of Eq. (\[mean n excat\]) is vanishing when some $Z_k$ is approaching the unit circle, we may have some critical scenarios, where the $\langle n \rangle$ can be asymptotically computed by neglecting non-diverging terms in the formal formula Eq. (\[mean n excat\]). This leads to simpler formulas but with more physical insights. We will investigate these cases in the following sections.
Relation of the mean FDT time and the FDR variance \[The detected transition time and return fluctuations\]
===========================================================================================================
There is a general relation between the mean FDT time $\langle n \rangle$ and the matrix $\{V_{i,j}\}$, describing the variance of the FDR problem. The relation is rather general, but becomes especially useful when both $\langle n \rangle$ and $V_r$ are large.
First we reformulate some of the main equations which we will use later. The variance of the FDR time is [@Gruenbaum2013] $$V_r=\langle n^2 \rangle_r -\langle n \rangle_r^2=\sum_{i,j=1}^{w-1}V_{i,j},
\label{V Gruen}$$ where $V_{i,j}=2/(Z_i Z_j^{\ast}-1)$. Also $P_{\rm det}$ can be written in terms of summations over matrix elements of $P_{i,j}$: $$P_{\rm det}=\sum_{i,j}P_{i,j},\qquad P_{i,j}=\frac{C_i C_j^{\ast}}{(Z_i Z_j^{\ast}-1)Z_i Z_j^{\ast}}.$$ Using Eq. (\[mean n excat\]), the matrices $P_{i,j}$ and $V_{i,j}$ give also the mean FDT time: $$\langle n \rangle=\frac{1}{2}\sum_{i,j=1}^{w-1}Z_i Z_j^{\ast}P_{i,j}V_{i,j}.
\label{n and var}$$ This equation relates the $\langle n \rangle$ and terms $V_{i,j}$ of the $V_r$, which indicates that the fluctuations of the FDR time reveal the characteristics of the mean FDT time. Below we show cases where one element of the summation is dominating $V_r \sim V_{s,s}$ and $|Z_s|\rightarrow 1$ (subscript $s$ stands for single.), such that
$$\langle n \rangle \sim \frac{P_{s,s}}{2} V_{s,s}\sim \frac{P_{s,s}}{2}V_r.
\label{einstein1}$$
This is similar to the Einstein relation in the sense that diffusivity (a measure of fluctuations) is related to mobility (a measure of the average response). In the Sec. \[time and fluctuation relation\] we will find the exact expression for the different scenario.
After obtaining the general results Eqs. (\[mean n excat\],\[n and var\]), we will focus on the diverging mean FDT time, where the asymptotic $\langle n \rangle$ and its relation to $V_r$ are developed. Eq. (\[mean n excat\]) implies a divergent mean FDT time when $|Z_s|\rightarrow 1$. Since $|Z_s|=1/|z_s|$, where $z_s$ is the stationary point on the electrostatics field, the question is whether a stationary point is close to the unit circle. Next we will investigate three scenarios where $|Z_s|\rightarrow 1$, using the electrostatic picture. We distinguish them into the following cases: 1) a weak charge scenario, 2) two charges merging picture, and finally 3) one big charge theory.
Weak charge\[Weak charge\] {#sect:weak}
==========================
{width="1.7\columnwidth"}
In electrostatics, when one charge becomes much smaller than all other charges, one of the stationary points will be close to the weak charge [@Gruenbaum2013] (see Fig. \[fig:poles for weak charge and two charges\], where the yellow charge indicates the weak charge, and its corresponding pole is $Z_0$). In analogy, the stationary point of the moon-earth system is much closer to the moon than to the earth. We denote this charge $p_0$ and the stationary point $z_0$. The corresponding energy level of this weak charge is $E_0$ and its location is $\exp{(i E_0\tau)}$ on the unit circle. Since $z_0 \rightarrow e^{i E_0 \tau}$, from Eq. (\[relation of zero and pole\]) the reciprocal pole $|Z_0|=1/|z_0|\rightarrow 1$. Using Eq. (\[mean n excat\]), the asymptotic mean of the mean FDT time is $$\langle n \rangle \thicksim \frac{| C_0 |^2 }{(| Z_0|^2-1)^2},
\label{weak 1}$$ when $p_0\rightarrow 0$ and $|q_0|/p_0\gg1$. Here we assume $|C_0|^2/(|Z_0|^2-1)$ is the dominating part of $\langle n \rangle$, and all other terms in Eq. (\[mean n excat\]) are negligible. To find the exact expression of $\langle n \rangle$, we first need to find the pole $Z_0$. Using Eq. (\[Force field\]) together with perturbation theory presented in the Appendix \[weak charge Appendix\], we get $$Z_0\thicksim e^{i\tau E_0}+\epsilon^{\ast}e^{2i \tau E_0},
\label{Z_0}$$ with $$\epsilon = \frac{p_0}{\sum_{k=1}^{w-1}p_k/(e^{i \tau E_0}-e^{i \tau E_k})}.$$ Since $p_0\ll 1$, $e^{i E_0\tau}$ is the leading part of $Z_0$. Hence the pole $Z_0$ is located very close to the weak charge as we expect from basic electrostatics. The other $w-1$ charges give a small disturbance to $Z_0$ if they are not close to the weak charge. Substituting $Z_0$ into Eq. (\[C\_i general\]), the coefficient $C_0$ reads (see Appendix \[weak charge Appendix\]): $$C_0\sim -\frac{q_0}{p_0}\epsilon^{\ast}e^{2i\tau E_0}.
\label{C_0}$$ $C_0$ is determined by the fraction of $q_0$ and $p_0$, the parameter $\epsilon$ and the phase $e^{2i \tau E_0}$ which comes from the location of the weak charge. The small parameter epsilon is the effect of the remaining charges in the system, excluding the weak charge, acting on a test charge, where the stationary point is found.
Finally, using the normalization condition $\sum_k p_k=1$ and $1/(1-\exp[i x])=1/2+i\cot{[x/2]}/2$, we get from Eq. (\[weak 1\]) the mean FDT time $$\langle n \rangle\thicksim \frac{|q_0|^2}{4p_0^2}\Bigg\{ 1+\bigg[\sum_{k=1}^{w-1} p_k \cot{[(E_k-E_0)\tau/2]}\bigg]^2\Bigg\} .
\label{n weak}$$ The prefactor $|q_0|^2/4p_0^2$ depends on $q_0$ and $p_0$ defined in Eqs. (\[qk\],\[pk\]), and they rely only on the stationary states with energy level $E_0$ the initial and final states, but not on the other energy states of the system. This prefactor is the envelope of the mean FDT time as the $\cot()$ solution is oscillating when we modify $\tau$. From our assumption $|q_0|/p_0\gg1$ the value of this envelope is large. The summation in the bracket shows that $\langle n \rangle$ depends on all charges as expected.
As mentioned when Eq. (\[excptional points\]) holds we get the merging of two phases on the unit circle a case we will study in detail in the next section. In the vicinity of this point the mean FDT time diverges. So what is the physics for this divergence? We have shown before when two energy levels coalesce, the total detection probability $P_{\rm det}$ is not unity, which means the quantum particle goes to “dark states" in the Hilbert space [@Thiel2019]. This divergence reflects that the total detection probability $P_{\rm det}$ deviates from $1$, indicating that one or more states are not accessible by the quantum walker. We will see this connection in some examples below.
Two merging charges\[Two merging charges\] {#sect:merging}
==========================================
Another case with a pole close to the unit circle is when the phases of two charges, denoted by $p_a$ and $p_b$, satisfy the resonance condition $\exp{(i E_a\tau)}\simeq \exp{(i E_b \tau)}$ (see Fig. \[fig:poles for weak charge and two charges\], the merging charges are colored green). As mentioned, this means that we are close to a degeneracy of the backfolded spectrum. It can be achieved by modifying $H$ or the sampling time $\tau$. Then the small parameter $\delta=(\Bar{E}_b-\Bar{E}_a)\tau/2$ measures the angular distance between the two phases. When the two charges merge, a related pole denoted $Z_p$ (subscript $p$ is for pair of merging charges), will approach the unit circle $|Z_p|\rightarrow 1$. Using Eq. (\[mean n excat\]), for the mean FDT time $\langle n \rangle$, we get $$\langle n \rangle \sim \frac{|C_p|^2}{(|Z_p|^2-1)^2},\quad \delta\rightarrow 0.
\label{two merging charges}$$ To find the pole $Z_p$, we first treat the charge field as a two-body system. Because by our assumption all other charges are far away from the two merging charges. Then we take the background charges into consideration. Using the two-body hypothesis together with Eq. (\[relation of zero and pole\]), we find in perturbation theory (see Appendix \[two charge pole\]) $$Z_p\thicksim Z_p^{(0)}+Z_p^{(1)},
\label{Z_p}$$ here $Z^{(0)}_p$ and $Z^{(1)}_p$ are defined in Appendix \[two charge pole\] in Eq. (\[appendix Z\_p\]). Plugging $Z_p$ into Eq. (\[C\_i general\]) yields for the coefficient $C_p$ $$|C_p|\sim 2\delta\frac{|q_a p_b -q_b p_a|}{(p_a +p_b )^2} ,
\label{C_p}$$ where $|C_p|$ is determined by the phase difference, charges and $q_k$. Since $\delta$ is a small parameter, $|C_p|$ also becomes small when two charges merge. Substituting $C_p$ and $Z_p$ into Eq. (\[two merging charges\]), the mean FDT time becomes $$\langle n \rangle \thicksim \frac{(p_a+p_b)^2|q_a p_b-q_b p_a|^2}{p_a^2p_b^2}\frac{1}{\tau^2(\Bar{E}_b-\Bar{E}_a)^2} .
\label{n two charges}$$ It should be noted that this formula does not include the background, which is quite different from the weak charge case. When two charges are merging, the expected transition time $\langle n \rangle$ diverges since $(\Bar{E}_b-\Bar{E}_a)^2\tau^2$ is small. The term $|q_a p_b-q_b p_a|^2$ comes from the interference. At the special case $|q_a p_b-q_b p_a|^2=0$ we have an elimination of the resonance, meaning that the effect of divergence might be suppresses.
Relation between mean FDT time and FDR fluctuations {#time and fluctuation relation}
===================================================
When there is only one pole dominating, simple relations between the mean FDT time and the fluctuations of the FDR time are found. We start from the general relation Eq. (\[n and var\]). When the pole $|Z_s|\rightarrow 1$ we have Eq. (\[einstein1\]). Here $Z_s$ is a single pole approaching the unit circle, it could be either $Z_p$ for two merging charges or $Z_0$ for one weak charge. $P_{s,s}$ is the diagonal term of the matrix $\{P_{i,j}\}$, which is real and positive. Based on Secs. \[Weak charge\],\[Two merging charges\] we can get exact expressions for $P_{s,s}$ under different circumstances.
In the weak charge regime, $P_{s,s}\sim |C_0|^2/(|Z_0|^2-1)$. Substituting the $C_0$ and $Z_0$ into $P_{s,s}$, the ratio of the mean FDT time and the FDR variance reads $$\frac{\langle n \rangle}{V_r} \sim \frac{|q_0|^2}{2p_0},
\label{n v weak q}$$ when energy level $E_0$ is not degenerate, we have: $$\frac{\langle n \rangle}{V_r} \sim \frac{|\langle \psi_{\rm in}|E_0\rangle|^2}{2}.
\label{n v weak}$$ From Eq. (\[n v weak q\]) and Eq. (\[n weak\]), we can get the expression of $V_r$, which confirms the result for $V_r$ in [@Yin2019]. The beauty of this simple relation is that it only depends on the overlap of the initial state $|\psi_{\rm in}\rangle$ and $|E_0\rangle$. So how we prepare the quantum particle is of great importance for the mean FDT time. The quantum particle will remember its history. Furthermore, $ |\langle \psi_{\rm in}| E_0\rangle|^2/2<1/2$ implies that the mean FDT time is bounded by one half of the FDR variance.
For the two merging charges scenario we have $P_{s,s}\sim |C_p|^2/(|Z_p|^2-1)$. Using Eqs. (\[Z\_p\],\[C\_p\]) gives us the ratio $$\frac{ \langle n \rangle}{V_r}\sim\frac{|q_a p_b-q_b p_a|^2}{2(p_a+p_b)p_a p_b} .
\label{n v two charge}$$ From Eqs. (\[n two charges\],\[n v two charge\]) we get an expression for $V_r$, which was also derived in [@Yin2019]. Here the initial state $|\psi_{\rm in}\rangle$ plays an important role because $q_a$ and $q_b$ are related to the initial state (unlike $p_a$ and $p_b$). Under some special symmetry of the system we can get $p_a/q_a=p_b/q_b$, such that $|q_a p_b-q_b p_a|^2=0$. As mentioned, this is reflects a elimination of the resonance because $\langle n \rangle$ will tend to some small values, while the mean FDR variance diverges.
**Remark:** We may start from Eqs. (\[mean n excat\],\[V Gruen\]), if one of the poles is denoted $Z_s$ and is close to the unit circle. Then we have roughly $\langle n \rangle \sim |C_s|^2/(|Z_s|^2-1)^2$ and $V_r\sim 2/(|Z_s|^2-1)$. The relation of the mean FDT time and the FDR variance is $\langle n \rangle \sim |C_s|^2V_r^2/4$, i.e., $\langle n\rangle$ is proportional to $V_r^2$. This intuition does not reveal the real physics, since for a divergent $V_r$ we get $|C_s|\rightarrow 0$.
Two-level System\[Two level system\]
------------------------------------
![Schematic models. We perform the calculations on different graphs. The quantum particle is prepared in the initial state $| \psi_{\rm in} \rangle$ and we set the detector to detect the state $|\psi_{\rm d}\rangle$. $U$ is the strength of potential well or potential barrier we set in the system. $(a)$ Two level model. $(b)$ Y-shaped molecule. $(c)$ Benzene-like ring. $(d)$ Linear five-site molecule. []{data-label="fig:schematic model"}](fig4.pdf){width="1\columnwidth"}
As an application of our general theory we consider tight-binding models on simple graphs. The first example is a quantum walk on a two-site graph (see Fig. \[fig:schematic model\]($a$)) (i.e. a two-level system). The Hamiltonian of this system reads $$H=-\gamma(| 0 \rangle\langle 1|+| 1\rangle\langle 0|+U| 0\rangle\langle 0|).
\label{two level H}$$ It describes a quantum particle hopping between two sites 0 and 1, where a potential $U$ is added at site 0. This model also presents a spin $1/2$ in a field.
We prepare the initial quantum state as $|0\rangle$, which means that the particle is on site 0. The detector is set to detect the particle at site 1; i.e. the detected state is $|1\rangle$. From Eq. (\[two level H\]) the energy spectrum of the system is (we set $\gamma=1$ subsequently): $E_0 = (-U-\sqrt{U^2+4})/2$ and $E_1 = (-U+\sqrt{U^2+4})/2$. In the large $U$ limit, where $E_0\rightarrow -U$ and $E_1\rightarrow 0$, the two energy levels $E_0$ and $E_1$ are separated. From Eq. (\[pk\]) the charge $p_0=1/(E_0^2+1)$ and from normalization $p_1=1-p_0$. When we increase the value of the potential $U$, the charge $p_0\rightarrow 0$, which represents a weak charge in the system. From Eq. (\[qk\]) we have $q_0=E_0/(E_0^2+1)$ and $q_1=E_1/(E_1^2+1)$. The ratio $|q_0|/p_0$ is $|E_0|$ ,which is our dimensionless variable growing with the potential $U$. From Eq. (\[n weak\]) the mean FDT time of this simple two-level system is $$\langle n \rangle \sim \frac{U^2}{4}[ 1+\cot^2{(U\tau/2)}].
\label{n two-level}$$ $\langle n \rangle$ becomes larger as we increase $U$, indicating the potential well blocks the propagation of the wave function, making it hard to find the particle at the detected state. In Eq. (\[n two-level\]), when $U\tau$ is close to $2\pi k, k=1,2,\cdots$ the mean FDT time diverges. Note that $U \tau = 2 \pi k$ is the condition for exceptional points (Eq. (\[excptional points\])), in the limit of large $U$. At these exceptional points, the total detection probability $P_{det}$ drops from $1$ to $0$.
Choosing the sampling frequency $1/\tau=1/3$, the exact $\langle n \rangle$ can be obtained either from the quantum renewal equation Eq. (\[quantum renewal\]) or our first main result Eq. (\[mean n excat\]). Here we use the latter formula, and the result is visualized in Fig. \[fig:two level system\] (left y axis). In the vicinity of the exceptional points the total detection probability drops from the unity and the mean FDT time diverges.
. Here we fix $\tau=3$. The exact mean FDT time (black dashed line) meets quite well with our theoretical result Eq. (\[n two-level\]) (cyan line). Close to the exceptional points $U\sim 2\pi k/\tau, k=1,2,\cdots$, where the back-fold energy levels are degenerate, the total detection probability (red line) drops to $P_{\rm det}=0$ and $\langle n \rangle$ diverges as we expected.[]{data-label="fig:two level system"}](fig5.pdf){width="0.99\columnwidth"}
Y-shaped Molecule
-----------------
The next example is the Y-shaped molecule, where the quantum particle can jump from states $|0\rangle, \quad |1\rangle, \quad |2\rangle$ to state $|3\rangle$ and vice versa (see schematics in Fig. \[fig:schematic model\] ($b$)). We add a potential $U$ at site 0. Then the Hamiltonian of the Y-shaped molecule reads $$H=-\gamma(U|0\rangle\langle0|+\sum_{i=1}^3|3\rangle\langle i|+\sum_{j=1}^3|j\rangle\langle 3|).$$ We prepare the quantum particle in the state $|\psi_{\rm in}\rangle=|0\rangle$ and the detection is performed in the state $|1\rangle$. Due to the mirror symmetry of Y-shaped molecule, the energy level $E_3=0$. Other energy levels $E_0$, $E_1$ and $E_2$ are given by the roots of the equation $E^3+U E^2 - 3E-2U = 0$. When $U$ is large, we have $E_0\sim -U$, $E_1\sim \sqrt{2}$ and $E_2\sim -\sqrt{2}$. From Eq. (\[pk\]) the charges are $p_0\rightarrow 0$, $p_1\rightarrow 1/4$, $p_2\rightarrow 1/4$ and $p_3\rightarrow 1/2$. The appearance of the weak charge $p_0$ is because one of the eigenstate is nearly localised on $|0\rangle$, more specifically $|E_0\rangle \simeq |0\rangle$. The exact numerical values of both energy levels $\{E_i\}$ and charges $\{p_i\}$ are shown in Appendix \[order\] in Fig. \[fig: energy level charge dnesity\]. Using Eq. (\[n weak\]) the mean FDT time of the Y-shaped molecule reads $$\langle n \rangle \sim \frac{|q_0|^2}{4 p_0^2}\Bigg\{1+\bigg[\sum_{i=1}^3 p_i\cot{[(E_i-E_0)\tau/2]}\bigg]^2\Bigg\}.
\label{n Y shape}$$ The initial site and detected site are not symmetric because of the potential $U$. This implies $|\langle E_0|0\rangle|\gg |\langle E_0| 1\rangle|$ and $|q_0|/p_0\gg 1$. When two energy levels are coalescing Eq. (\[n Y shape\]) diverges. The prefactor in Eq. (\[n Y shape\]) indicates the asymptotic tendency of the mean FDT time versus the potential $U$ (see Fig. \[fig:Y-shaped n n/v\]), which should be observed experimentally. We denote this prefactor as the weak charge envelope $\langle n \rangle_e$ $$\langle n \rangle_{e}\sim \frac{|q_0|^2}{4 p_0^2}= \frac{1}{4}\frac{|\langle 0| E_0\rangle|^2}{|\langle 1| E_0\rangle|^2}.
\label{n Y e}$$ The weak charge envelope is determined by the overlaps of the initial and detected state. From Eq. (\[n v weak\]) the relation between the mean FDT time and the FDR variance gives $$\frac{\langle n \rangle}{V} \sim \frac{1}{2}.
\label{n v Y-shaped}$$
To plot an example, we solve the quantum renewal equations exactly, as was done in Sec. \[Two level system\], here we choose the sampling period $\tau=3$. The value of potential well $U$ goes from $2$ to $12$. As shown in Fig. \[fig:Y-shaped n n/v\], Eqs. (\[n Y shape\], \[n Y e\], \[n v Y-shaped\]) work well in the weak charge regime where $U$ is large.
{width="1.99\columnwidth"}
Benzene-type ring {#Benzene-type ring}
-----------------
\[sect:ring\]
For the third model we consider the Benzene-type ring which has six spacial states $|0\rangle,|1\rangle,\cdots,|5\rangle$ (see Fig. \[fig:schematic model\]($c$)). We use periodic boundary conditions and thus from the site labeled $x=5$ the particle may hop either to the origin $x=0$ or to the site labeled $x=4$. Then the Hamiltonian of the ring reads $$H=-\gamma[\sum_{x=0}^5(| x \rangle\langle x+1|+| x+1\rangle\langle x|)].$$ We prepare our quantum particle in the state $|0\rangle$ and perform the detection in the state $|3\rangle$, which monitors the travel of the quantum particle from site $0$ to the opposing site. In this case $P_{\rm det}=1$ except for special sampling times.
The Hamiltonian of the benzene-type ring has the energy spectrum $E_k=-2\cos{(\theta_k)}$ and the eigenstates are $|E_k\rangle^T =(1,e^{i \theta_k},e^{2i \theta_k},e^{3i \theta_k},e^{4i \theta_k},e^{5i \theta_k})/\sqrt{6}$ with $\theta_k=2\pi k/6$ and $k=0,1,2,3,4,5$ (the superscript $T$ is the transpose). In this case we have four distinct energy levels so $w=4$. From Eqs. (\[qk\],\[pk\]) the charges and $q_k$ read $$\begin{aligned}
p_1 &=&\frac{1}{6},\quad p_2 &=& \frac{1}{6},\quad p_3 &=& \frac{1}{3}, \quad p_4 &=& \frac{1}{3};\\
\quad q_1 &=& \frac{1}{6},\quad q_2 &=& -\frac{1}{6},\quad q_3 &=& -\frac{1}{3}, \quad q_4 &=& \frac{1}{3}.
\end{aligned}$$ As mentioned, The energy spectrum of the ring is degenerate and the sampling time $\tau$ will introduce effective degeneracies to the problem. From Eq. (\[excptional points\]), the exceptional sampling times are $\tau=\pi/2,2\pi/3,\pi,4\pi/3, 2\pi$ in the time interval $\{\tau | 0\leqslant \tau\leqslant2\pi\}$. Close to these exceptional points we will have the scenario of two charges merging, where we can employ our equations to give the theoretical predictions (see Fig. \[fig:ring lacation n pure\]).
- When $\tau$ is close to $\pi/2$ or $3\pi/2$ we have $|E_1-E_2|\tau\sim 2\pi k$. The charges $p_1$ and $p_2$ coalesce (see Fig. \[fig:ring lacation n pure\](B)). For the mean transition time $\langle n \rangle$ and $\langle n \rangle/V_r$, using Eqs. (\[n two charges\],\[n v two charge\]) we have $$\langle n \rangle\sim\frac{1}{36}\frac{1}{(\tau-\pi/2)^2},\quad \frac{\langle n \rangle}{V_r}\sim\frac{1}{6}.
\label{n ring t=pi/2}$$
- When $\tau$ is close to the $2\pi/3$ or $4\pi/3$ we have $|E_1-E_4|\tau\sim 2\pi k$ and $|E_2-E_3|\tau\sim 2\pi k$. Two pairs of charges are merging separately (see Fig. \[fig:ring lacation n pure\](C) and (D)). From Eqs. (\[C\_p\],\[n two charges\]), due to the elimination $q_1 p_4-q_4 p_1=0$ and $q_2 p_3-q_3 p_2=0$ we have $$\langle n \rangle\sim O(1), \quad \langle n \rangle/V_r\rightarrow 0 .% V= 0.
\label{Bring 0}$$ The leading order of $\langle n \rangle$ vanishes, so $\langle n \rangle$ drops to some small values, leading to a small “discontinuity" on the graph. Close to these points we find that it takes less time for the walker to reach the detected state.
- When $\tau$ is close to $\pi$ we also have two groups of charges merging Fig. \[fig:ring lacation n pure\](E), i.e. $p_1$ is close to $p_2$, and $p_3$ is close to $p_4$. Eq. \[n two charges\] gives $$\langle n \rangle\sim \frac{1}{36}\frac{1}{(\tau-\pi)^2}+\frac{4}{9}\frac{1}{(\tau-\pi)^2}.
\label{ring 3}$$ For the ratio of $\langle n \rangle$ and $V_r$ there are two groups of charges which we treat separately. The first group we use Eq. (\[n v two charge\]) to obtain $\langle n \rangle_{1,2} /V_{1,2}=1/6$. Similarly, for the second group we have $\langle n \rangle_{3,4} /V_{3,4}=1/3$. The return variance $V_r=V_{1,2}+V_{3,4}$ and the mean FDT time $\langle n \rangle=\langle n \rangle_{1,2}+\langle n \rangle_{3,4}=V_{1,2}/6+V_{3,4}/3$. We can measure the fluctuations $V_r$ but not the terms $V_{1,2}$ and $V_{3,4}$. So here we do not have a direct relation between $\langle n \rangle$ and $V_r$. Using Eqs. (\[two merging charges\],\[n v two charge\]), we first calculate $V_{1,2}$ and $V_{3,4}$ (then $V_r=V_{1,2}+V_{3,4})$. Comparing $V_r$ and Eq. (\[ring 3\]), we have $\langle n \rangle/V_r=5/18$.
As shown in Fig. \[fig:ring lacation n pure\], we plot the exact results of the $\langle n \rangle$ for the $\tau$ from $0$ to $2\pi$. The theoretical predictions meet the exact values quite well close to the exceptional points where the total detection probability exhibits a sudden jump in its value.
{width="1.8\columnwidth"}
So far we deal with one zero close to unit circle, and now we switch to the more complicated cases where we have more than one pole in the vicinity of the unit circle. We find the mean FDT time, but an Einstein like relation is not achieved in such case (as an example, see part 3 of the Benzene-type ring).
Big charge theory\[Big charge theory\] {#sect:big}
======================================
![Schematic plot of the zeros $\{z_i\}$ in the complex plane for the big charge theory. Here $p_b\rightarrow 1$ is the big charge and hence from normalization all other charges are weak. The stationary points ($\{z_i\}$) are close to the weak charges, so as mentioned in the main text the poles $|Z_i|=1/|z_i|\rightarrow 1$, and they are all out side the unit circle.[]{data-label="fig:big charge charge poles"}](fig8.pdf){width="0.9\columnwidth"}
Another scenario which leads to divergences of the mean FDT time $\langle n \rangle$ is when all the poles are close to the unit circle. This comes from the fact that the detected state is close to one of the eigenstates of Hamiltonian $H$, leading to a big charge appearing in the theory (Eq. (\[pk\]). Using Eq. (\[mean n excat\]), the off-diagonal terms in $\langle n \rangle$ are negligible compared with the diagonal terms, then we get $$\langle n \rangle\sim\sum_{i=0,i\neq b}^{w-1}\frac{|C_i|^2}{(-1+|Z_i|^2)^2}\ ,\ \ |Z_i|\sim 1.$$ The big charge, denoted $p_b\sim 1$, associated to the energy level $E_b$, is large in comparison with the other charges. Since the sum of all the charges is unity $\sum_{k=1}^{w-1}p_k=1$ and each of them is positive we have $1-p_b=\sum_{k\neq b}p_k\sim0$. Hence there is one big charge $p_b$ and $w-1$ weak charges. Basic electrostatics indicates that the $w-1$ stationary points will lie close to the $w-1$ weak charges. From Eq. (\[relation of zero and pole\]) we have $|Z_i|=1/|z_i|$, such that all the poles $|Z_i|\rightarrow 1$ in this case. As visualized in Fig. \[fig:big charge charge poles\], the $w$ charges have $w-1$ poles and all of them are close to the weak charges.
Because all the charges are weak except for $p_b$, we find a stationary point $z_i$ when we consider only a pair of charges, i.e. $p_b$ and one of the $w-1$ weak charges $p_i$. This problem becomes a two-body problem (the charge $p_b$ and the weak charge $p_i$) for finding the stationary point between them, and all other charges are negligible. Using Eq. (\[Force field\]), the zeros are given by the root of $p_b/(e^{i E_b \tau}-z_i)+p_i/(e^{i E_i \tau}-z_i)=0$, which yields $z_i=(p_i e^{i E_b \tau}+p_b e^{i E_i \tau})/(p_i+p_b)$. From the relation between zeros and poles in Eq. (\[relation of zero and pole\]) we have $$Z_i\sim e^{i E_i \tau}+(e^{i E_i \tau}-e^{i( 2E_i-E_b) \tau})\frac{p_i}{p_b}.$$ The $i$ goes from $i=0$ to $w-1$ but $i\neq b$, so all $w-1$ poles are found. The first part of $Z_i$ is just the location of the charge $p_i$, the second part is small and comes from the net field of $p_i$ and $p_b$. We put the $Z_i$ into Eq. (\[C\_i general\]) to get the coefficient $$C_i\sim -\frac{q_i}{p_b}(1-e^{i (E_i-E_b) \tau})e^{i E_i \tau}.
\label{big Ci}$$ Here enters the ratio of $q_i$ and the big charge $q_i/p_b$, which is a small parameter. $e^{i (E_i-E_b)}$ measures the phase difference between them. Substituting both $Z_i$ and $C_i$ into Eq. \[C\_i general\], the mean FDT time for the big charge scenario reads $$\langle n \rangle\sim \sum_{i=0,i\neq b}^{w-1}\frac{|q_i|^2}{4p_i^2\sin^2{[(E_i-E_b)\tau/2]}} .
\label{n big charge}$$ It is very interesting to recall that in our weak charge theory (see Eq. (\[n weak\])) the envelope is given by $|q_0|^2/4p_0^2$, where $p_0$ is a weak charge. For the big charge we have $|q_i|^2/4p_i^2$, where $p_i$ is also small.
Localized wave function
-----------------------
A good example for the big charge theory is when the wave function is effectively localized at the detected state by a strong potential. For instance, we localize the wave function at one node of the graph, and then we set our detector at this node. To establish a specific example, we choose a five-site linear molecule put the detector at the site $x=0$ and prepare the initial state at $|4\rangle$. In order to localize the wave function at the detected state, we add a potential barrier $U$ at site $x=0$ (see Fig. \[fig:schematic model\]($d$)). Then the Hamiltonian of this five-site molecule reads $$H=-\gamma[\sum_{x=0}^4 (| x \rangle\langle x+1|+| x+1\rangle\langle x|)+U| 0\rangle\langle 0|].
\label{H five}$$ Here the boundary conditions are that from the site labeled $x=4$ one can only hop to the site labeled $x=3$, and from the site labeled $x=0$ one can only hop to the site labeled $x=1$.
For the energy spectrum we consider the regime where the wave function is effectively localized. As we increase the value of $U$, the energy level $E_0\rightarrow -U$. At the same time, this large potential well makes it difficult for the quantum particle to hop to the state $|0\rangle$. So the remaining four energy levels are given by the new Hamiltonian $H_l=-\gamma\sum_{x=1}^4(| x \rangle\langle x+1|+| x+1 \rangle\langle x|)$ with the same boundary condition as Eq. (\[H five\]). Hence the energy spectrum reads $E_0 \sim -U$, $E_1 \sim (1+\sqrt{5})/2$, $E_ 2\sim -(1+\sqrt{5})/2$, $E_3 \sim (-1+\sqrt{5})/2$ and $E_4 \sim (1-\sqrt{5})/2$. Notice that the energy levels are non-degenerate hence $w=5$. The exact values of the energy levels are calculated and depicted in Appendix \[order\] in Fig. \[fig: energy level charge dnesity\] $(C)$.
Next we prepare the quantum particle in the state $|4\rangle$, such that the system describes the movement of the particle from site $x=4$ to $x=0$ on a linear molecule. From Eq. (\[pk\]) it follows that the big charge $p_0\rightarrow 1$ and the remaining weak charges $p_{i\neq 0}\rightarrow 0$. With Eq. (\[n big charge\]) we get for the mean FDT time $$\langle n \rangle\sim \sum_{i=1}^4\frac{|q_i|^2}{4p_i^2\sin^2{[(E_i-E_0)\tau/2]}}.
\label{n 5 sites}$$
In Fig. \[fig:five sites n t n\] we compare the numerical result with our big charge theory, choosing the sampling time $\tau=1$ and the potential well from $0$ to $15$. In the limit of large $U$ the four weak charges are fixed on the unit circle, their positions are given by their $U$ independent phase $\exp{(i E_i\tau)}$. When we increase $U$ the strong charge $p_0$, is thus crossing the location of the other charges and in the range $0<U<15$ which happens twice ($15/2\pi \sim 2$). As shown in Fig. \[fig:five sites n t n\], we have two groups of divergencies each with four peaks. The number of peaks in each group is $w-1=4$.
![The mean FDT time $\langle n \rangle$ versus $U$ for the five-site molecule. Here the sampling time is $\tau=1$ and the quantum particle moves from $|4\rangle$ to $|0\rangle$, see Fig \[fig:schematic model\] ($d$). The exact values (black dashed line) are calculated from Eq. (\[mean n excat\]). The cyan line is our big charge theory result Eq. (\[n 5 sites\]). Close to the exceptional points given by Eq. (\[excptional points\]), the mean FDT times diverge. Since the strong charge $p_0$ rotates two laps on the unite circle, there are two clusters of peaks. In each cluster the big charge passes through the remaining four charges, leading to the four peaks in the graph.[]{data-label="fig:five sites n t n"}](fig9.pdf){width="0.99\columnwidth"}
discussion
==========
We have used the quantum renewal equation [@Friedman2017] to investigate the mean FDT time, for systems in a finite-dimensional Hilbert space. A general formula for mean FDT time is developed. Then we focus on the diverging mean FDT times and find a relation similar as the Einstein relation, which relates the mean FDT time and the fluctuations of the FDR time.
The problem of the mean FDR time was considered in [@Gruenbaum2013]. For quantum walks which are subject to repeated measurements, the return to the initial state and the transition to another state have quite different dynamical properties. First, both the return and the transition properties are very sensitive to the back-folded spectrum of Eq. (\[back\_folded\_spectrum\]). The mean FDR time is topologically protected and equal to the number of non-degenerate back-folded eigenvalues in [@Gruenbaum2013; @Yin2019]. We have not found such a topologically protected time scale for the FDT properties. The mean FDT time is divergent near the degeneracies, in the presence of 1) a very small overlap $p_0=|\langle E_0|\psi_d\rangle|^2$, 2) merging of two phases, and 3) the big charge theory. We note that other scenarios for diverging mean FDT times can be found for example in the Zeno limit [@Dhar2015; @Thiel2019quantization] and when three or more charges are merging [@Yin2019]. Another difference between the return and the transition problem is that, for instance, the total detection probability $P_{\rm det}$ for the return is always unity, while the transition probability to another state is sensitive, e.g. to the geometric symmetry of the underlying graph [@Thiel2019b]. The qualitative difference between return and transition properties originates in the fact that the return properties are based on the amplitude $u_n$ alone, whereas the transition properties depend on both amplitudes $u_n$ and $v_n$. This implies a more complex physical behavior for the transition properties. Although a unitary evolution without the projective measurements is already complex due to the different energy levels, the interruption by the measurement adds another timescale $\tau$ to the dynamics. This fact implies that the degeneracy of two or more of the phase factors affects the dynamics substantially and that the dynamics depends strongly on the back-folded spectrum. It also explains why a small coefficient $p_0$ has a similar effect: the effective dimensionality of the available Hilbert space is either reduced by the degeneracy of the phase factors or by the vanishing overlap $p_0$, since each phase factor carries the coefficient $p_0$ as $p_0\exp(-i E_0\tau)$. This observation and the results of the calculations in Secs.\[sect:weak\], \[sect:merging\] and \[sect:big\] can be summarized to the statement that divergent mean FDT times are caused by the proximity to a change of the effective Hilbert space dimensionality.
We have found that when a single stationary point, $z$ (or the pole $Z=1/z^*$) approaches the unit circle in the complex plane, we get a relation between the mean FDT time and the fluctuations of the FDR time, see Eqs. (\[n v weak\],\[n v two charge\]). This is because the slow relaxations of $\phi_n$, which are controlled by a single pole $Z_i$, making all others irrelevant.
Our results also indicate that a quantum walk, interrupted by repeated measurements, is quite different from classical diffusion. The divergent mean FDT time reflects the fact that the transition to certain states can be strongly suppressed. In this sense the dynamics is controllable by choosing the time step $\tau$. This could be important for applications, for instance, in a quantum search process: the search time for finding a certain quantum state depends significantly on the choice of the time step $\tau$. Trapping of the quantum state by an external potential also influences strongly the value of the mean FDT time $\langle n \rangle$, as we have seen in our examples.
Acknowledgements
================
We thank Felix Thiel and David Kessler, for many helpful discussions, which led to simplifications of some of the formulas of this paper. The support of Israel Science Foundation’s grant 1898/17 as well as the support by the Julian Schwinger Foundation (KZ) are acknowledged.
[41]{}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/10.1017/CBO9780511606014) (, ) [**](https://doi.org/10.1142/9104) (, ) [****, ()](https://doi.org/10.1103/PhysRevA.48.1687) [****, ()](https://doi.org/10.1103/PhysRevA.58.915) [****, ()](https://doi.org/10.1023/A:1019609420309) [****, ()](https://doi.org/10.1126/science.1174436), [****, ()](https://doi.org/10.1103/PhysRevLett.104.100503) [****, ()](https://doi.org/10.1126/science.1260364), [****, ()](https://doi.org/https://doi.org/10.1016/j.jcss.2004.03.005) [****, ()](https://doi.org/10.1103/PhysRevA.73.032341) [****, ()](https://doi.org/10.1103/PhysRevA.74.042334) [****, ()](https://doi.org/10.1103/PhysRevA.78.022324) [****, ()](https://doi.org/10.1007/s00220-012-1645-2) [****, ()](https://doi.org/10.1007/s00220-014-1929-9) [****, ()](https://doi.org/10.1007/s10955-014-0936-8) [****, ()](https://doi.org/10.1103/PhysRevA.91.062115) [****, ()](https://doi.org/10.1088/1751-8113/48/11/115304) [****, ()](https://doi.org/10.1088/1751-8121/aa5191) [****, ()](https://doi.org/10.1103/PhysRevE.95.032141) [****, ()](https://doi.org/10.1103/PhysRevLett.120.040502) [****, ()](https://doi.org/10.1103/PhysRevB.98.104309) [****, ()](https://doi.org/10.1103/PhysRevE.98.022129) @noop (), [****, ()](https://doi.org/10.1103/PhysRevA.99.062105) @noop (), @noop (), [****, ()](https://doi.org/10.1103/PhysRevX.9.031009) @noop (), [****, ()](https://doi.org/10.1103/PhysRevLett.79.325) [****, ()](https://doi.org/10.1103/PhysRevA.70.022314) [****, ()](https://doi.org/10.1007/s00440-004-0423-2) [****, ()](https://doi.org/10.1103/PhysRevLett.116.100501) [****, ()](https://doi.org/10.1103/PhysRevResearch.1.033086) [****, ()](https://doi.org/10.1103/PhysRevLett.100.020501) [****, ()](https://doi.org/10.1103/PhysRevLett.114.140502) @noop (), @noop (), [****, ()](https://doi.org/10.1103/PhysRevLett.123.140403) @noop (), @noop (), @noop (),
Order of ${\cal G}(z)$ and ${\cal D}(z)$ {#order}
=========================================
{width="1.6\columnwidth"}
In this section we proof $\text{deg}({\cal D}(z))>\text{deg}({\cal G}(z))$ used in the main text. Since $ {\cal G}(z) \simeq \sum q_i z^{w-1} + \cdots$, the highest order of ${\cal G}(z)$ is $(\sum_{i=0}^{w-1}q_i) z^{w-1}$. However, what is special in the transition problem is $\sum_{i=0}^{w-1}q_i=\langle \psi_{\rm d}|\psi_{\rm in}\rangle=0$, namely that the highest order vanishes, such that $\text{deg}({\cal G}) <w-1$ for the numerator.\
Using Eq. (\[D\]), ${\cal D}(z)\simeq \sum p_i e^{i E_i \tau} z^{w-1} + \cdots$ the leading order of $z$ is $\beta z^{w-1}=(\sum_{i=0}^{w-1}p_i e^{i E_i \tau}) z^{w-1}$. $$\sum_{i=0}^{w-1}p_i e^{i E_i \tau}=\langle \psi_{\rm d}|e^{i \hat{H}\tau}|\psi_{\rm d}\rangle\neq 0.$$ Hence $\text{deg}({\cal D}(z))>\text{deg}({\cal G}(z))$.
Weak charge {#weak charge Appendix}
===========
In this section we derive Eqs. (\[Z\_0\],\[C\_0\],\[n weak\]) of the main text. Following the same procedure, we can derive Eqs. (\[C\_p\],\[n two charges\]) in the Sec. \[Two merging charges\].
As we mentioned in the main text, the weak charge $p_0\sim 0$ and the corresponding energy level is $E_0$. Using Eq. (\[Force field\]), we have: $$0=\sum_{k=0}^{w-1}\frac{p_k}{e^{i\uptau E_k}-z}=\frac{p_0}{e^{i\uptau E_0}-z}+\sum_{k=1}^{w-1}\frac{p_k}{e^{i\uptau E_k}-z}.
\label{for zeros}$$ Assuming that $z_0=e^{i E_0\uptau}-\epsilon$. The $\epsilon$ is the first order approximation. Using Eq. (\[for zeros\]), we have: $$0\approx \frac{p_0}{\epsilon}+\sum_{k=1}^{w-1}\frac{p_k}{e^{i\uptau E_k}-e^{i\uptau E_1}},$$ hence $$\epsilon \sim \frac{p_0}{\sum_{k=1}^{w-1}p_k/(e^{i\uptau E_1}-e^{i\uptau E_k})}.
\label{epsilon}$$ Using Eq. (\[relation of zero and pole\]), the pole $Z_0$ in the main text reads: $$Z_0=\frac{1}{z_0^{\ast}}=\frac{1}{e^{-i\uptau E_0}-\epsilon^{\ast}}\sim e^{i\uptau E_0}(1+\epsilon^{\ast}e^{i\uptau E_0}) .$$
The index $C_i$ is defined in Eq. (\[C\_i general\]). Plugging the pole $Z_0$ into Eq. (\[C\_i general\]), we obtain
$$\begin{aligned}
{\cal N}(Z_0) & \sim -e^{i E_0\tau}\bigg[\frac{q_0}{-\epsilon^{\ast}e^{2i E_0\tau}}+\sum_{j=1}^{w-1}\frac{q_j}{(e^{i E_j\tau}-e^{i E_0\tau})}\bigg]\\
&\sim \frac{q_0}{\epsilon^{\ast}e^{i E_0\tau}},\end{aligned}$$
and $$\begin{aligned}
{\cal D}^{\prime}(Z_0) & \sim \frac{p_0 e^{i E_0\tau}}{\epsilon^{\ast2}e^{4i E_0\tau}}+\sum_{j=1}^{w-1}\frac{p_j e^{i E_j\tau}}{(e^{i E_j\tau}-e^{i E_0\tau})^2}\\
& \sim \frac{p_0 }{\epsilon^{\ast2}e^{3i E_0\tau}}.\end{aligned}$$ Hence the coefficient $C_0$ used in the main text reads: $$C_0\sim \frac{q_0}{p_0}\epsilon^{\ast}e^{2 i E_0 \tau}.$$ Substituting $C_0$ and $Z_0$ into Eq. (\[weak 1\]), the mean FDT time becomes $$\langle n \rangle \sim \frac{|C_0|^2}{(|Z_0|^2-1)^2} \sim \frac{|q_0|^2}{p_0^2}\frac{|\epsilon|^2}{(2Re[\epsilon* e^{-i E_0\tau}])^2}.
\label{B6}$$ Using the mathematical property $1/(1-\exp{[ix]})=1/2+i\cot{[x/2]}/2$ and the normalization condition $\sum_{k=1}^{w-1}p_k=1-p_0\sim 1$, we can simplify the parameter $\epsilon$. From Eq. (\[epsilon\]), we have: $$\begin{aligned}
\epsilon &\sim \frac{p_0}{\sum_{k=1}^{w-1}p_k/(e^{i\uptau E_0}-e^{i\uptau E_k})}\\
& = e^{i E_0\tau}\frac{p_0}{\sum_{k=1}^{w-1}p_k/(1-e^{i\uptau (E_k-E_0)})}\\
& = e^{i E_0\tau}\frac{2p_0}{\sum_{k=1}^{w-1}p_k(1+i\cot{[\uptau (E_k-E_0)/2]})}\\
& \sim e^{i E_0\tau}\frac{2p_0}{1+i \sum_{k=1}^{w-1} p_k\cot{[\uptau (E_k-E_0)/2]}}.\end{aligned}$$ Plugging $\epsilon$ into Eq. (\[B6\]), the mean FDT time reads $$\langle n \rangle \sim \frac{|q_0|^2}{4p_0^2}\Bigg\{ 1+\bigg[\sum_{k=1}^{w-1} p_k \cot{[(E_k-E_0)\tau/2]}\bigg]^2\Bigg\}.$$
Two-charge pole $Z_p$\[two charge pole\]
========================================
In this section we derive Eq. (\[Z\_p\]) of the main text. When a pair of charges is nearly merging, say $\exp{(i E_a \tau)} \simeq \exp{(i E_b \tau)}$, one of the zeros denoted $z_p$ will be close to the unit circle. We define $2\delta=(\Bar{E}_b-\Bar{E}_a)\uptau$, hence $\delta$ is a order parameter measuring this process. We first consider the two merging charges. Using Eq. (\[Force field\]) we have $$\frac{p_a}{e^{i E_a\uptau}-z}=-\frac{p_b}{e^{i E_b\uptau}-z},
\label{only two}$$ which yields $$z_p^{(0)}=\frac{p_a e^{i E_b\uptau}+p_b e^{i E_a\uptau}}{p_a+p_b}.$$ Now we take the background charges into consideration. $$z_p=z_p^{(0)}-z_p^{(1)}.$$ Plugging $z_p$ into Eq. \[Force field\], we have
$$0=
\sum_{k=0}^{w-1}\frac{p_k}{e^{i E_k\uptau}-z}\approx\frac{p_a}{e^{i E_a\uptau}-z_p^{(0)}
+z_p^{(1)}}+\frac{p_b}{e^{i E_b\uptau}-z_p^{(0)}
+z_p^{(1)}}+\sum_{k\neq a,b}^{w-1}\frac{p_k}{e^{i E_k\uptau}-z_p^{(0)}} .
\label{appendix two charges zero}$$
The third part on the right-hand side is the effect of the background charges. We define it as $B$. $$B=\sum_{k\neq a,b}^{w-1}\frac{p_k}{e^{i E_k\uptau}-z_p^{(0)}}
\approx \sum_{k\neq a,b}^{w-1}\frac{p_k}{e^{i E_k\uptau}-e^{i\uptau\frac{\Bar{E}_A+\Bar{E}_B}{2}}}.
\label{B}$$ Using Eqs. (\[appendix two charges zero\],\[B\]), we obtain $$z_p^{(1)}\sim\frac{B p_a p_b (e^{i E_a\uptau}-e^{i E_b\uptau})^2}{(p_a+p_b)^3+B(p_a^2-p_b^2)(e^{i E_a\uptau}-e^{i E_b\uptau})}\sim
\frac{B p_a p_b (e^{i E_a\uptau}-e^{i E_b\uptau})^2}{(p_a+p_b)^3} .$$ Since $e^{i E_B\uptau}-e^{i E_A\uptau}\sim \delta$, $z_p^{(1)}\sim \delta^2$. The background charges give only a second order effect $O(\delta^2)$ to the zero $z_p$ as we expected. Using Eq. (\[relation of zero and pole\]) we have $$Z_p=Z_p^{(0)}+Z_p^{(1)}=\frac{1}{z_p^{\ast}}\approx \frac{p_A+p_B}{p_A e^{-i E_B\uptau}+p_B e^{-i E_A\uptau}}+\frac{B^{\ast}p_A p_B(e^{-i E_A\uptau}- e^{-i E_B\uptau})^2}{(p_A+p_B)(p_A e^{-i E_B\uptau}+p_B e^{-i E_A\uptau})^2} .
\label{appendix Z_p}$$
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- 'J.M. Riedl \[inst\]'
- 'T. Van Doorsselaere \[inst\]'
- 'I. C. Santamaria \[inst\]'
bibliography:
- '../sources.bib'
date: 'Received 4 March 2019; Accepted 22 April 2019'
title: 'Wave modes excited by photospheric p-modes and mode conversion in a multi-loop system'
---
\[inst\]
Introduction {#sec:intro}
============
Ever since the discovery of the high temperatures of the solar corona by [@edlen_1943], scientists have been trying to find an explanation for this phenomenon. One possible coronal heating mechanism is alternating current (AC) heating, in which magnetic energy is dissipated by waves [@aschwanden; @priest_2014; @parnell_demoortel_2012; @arregui_2017]. AC heating mechanisms were widely ignored for a long time until it was observed that waves are indeed ubiquitous in the solar corona [@de_pontieu_etal_2007; @tomczyk_etal_2007; @krishna_etal_2012; @morton_etal_2012; @nistico_etal_2013].
[@tomczyk_mcintosh_2009] observed waves with the Coronal Multi-channel Polarimeter (CoMP) instrument and found that the spectrum of velocity perturbations peaks at the same frequency ($\sim$ 3 mHz) as solar p-modes. [@morton_etal_2016] and [@morton_etal_2019] confirmed that the power enhancement at this frequency is a global phenomenon. Therefore, it seems likely that the ubiquitous waves are at least partially driven by p-modes. However, it is not yet clearly understood how p-modes propagate into the higher atmosphere. Observations tracing p-modes in the chromosphere and above were done by, for example, [@centeno_etal_2006], [@marsh_walsh_2006], [@de_wijn_etal_2009], [@prasad_etal_2015], and [@zhao_etal_2016], while numerical modeling was done by, for example, [@khomenko_etal_2008], [@fedun_etal_2011], [@santamaria_etal_2015], and [@griffiths_2018]. [@de_pontieu_etal_2004] and [@de_pontieu_etal_2005] found that, although p-modes with periods above the cutoff period are usually evanescent in the chromosphere, these p-modes can propagate upward along inclined magnetic flux tubes because the effective cutoff period is increased in a non-vertical magnetic field.
It was shown by [@bogdan_etal_1996], [@hindman_jain_2008], and [@gascoyne_etal_2014] that p-modes lose energy to magnetic tube waves, such as sausage waves and kink waves; the propagation of these excited waves into the solar atmosphere was not considered in those studies. Observations show that tube waves are indeed excited in the solar atmosphere. While propagating kink waves have already been observed many years ago [@verwichte_etal_2005], [@grant_etal_2015] were more recently able to observe sausage modes propagating from the photosphere to the transition region for the first time. These authors find surprisingly strong damping for those sausage waves; this is not well understood and therefore indicates again our limited knowledge of wave propagation in that region.
We therefore stress the importance of understanding the propagation of p-modes through the chromosphere, as they might be the source of decayless transverse waves. Those waves could be connected to coronal heating [e.g., @karampelas_etal_2017]. They have been observed by [@wang_etal_2012], [@nistico_etal_2013], and [@anfinogentov_etal_2013], for example. Transverse waves are currently modeled in the corona as loops with a horizontal driver in one or both footpoints [@karampelas_vandoorsselaere_2018; @guo_etal_2019; @karampelas_etal_2019; @pagano_de_moortel_2019], but from where those horizontal plasma movements originate is usually not discussed. One possible mechanism could be a self-oscillatory process due to the interaction of the loops with quasi-steady flows, as discussed by [@nakariakov_etal_2016]. Another possibility, as mentioned above, could be photospheric p-modes, which are converted to kink waves.
In this work, we study the conversion of photospheric p-modes to sausage and kink waves in an atmosphere that is gravitationally stratified and has additionally structuring perpendicular to the magnetic field. The model we use is in magnetohydrostatic (MHS) equilibrium and contains four loops with different inclinations ranging from the photosphere to the lower corona. We perturb the equilibrium at the bottom with a p-mode driver in the form of an analytic solution for gravity-acoustic waves and simulate the propagation of the waves. The waves interact with the cylindrical structure of our model atmosphere and tube modes are excited. The MHS model is described in Sect. \[sec:model\], while we explain the numerical setup in Sect. \[sec:numerics\]. We present assisting methods for data interpretation in Sect. \[sec:methods\] and present and discuss our results, including what kind of wave modes are excited in the corona and how their basic properties change, in Sect. \[sec:results\]. Finally, we present the summary and conclusions of our work in Sect. \[sec:summary&conclusions\].
Magnetohydrostatic equilibrium model {#sec:model}
====================================
We built a 3D MHS equilibrium atmosphere from the photosphere to the lower corona, which has to fulfill the condition $$\label{eq:mhs_equilibrium}
\vec{\nabla} p_0 - \frac{1}{\mu_0} \left( \vec{\nabla} \times \vec{B_0} \right) \times \vec{B_0}- \rho_0 \vec{g}=0,$$ where $p_0$, $\vec{B_0}$, and $\rho_0$ are the equilibrium pressure, equilibrium magnetic field, and equilibrium density, respectively, and $\mu_0$ is the permeability of vacuum. For our models, the gravity vector $\vec{g}=(0,0,-g)$ is constant and points to the negative $z$-direction. In addition, we built the atmosphere as periodic in the horizontal directions. The domain has a size of $n_x \times n_y \times n_z =$ 140$\times$140$\times$840 points with a resolution of $\Delta x \times \Delta y \times \Delta z \approx $ 14.3$\times$14.3$\times$6.0 km, which results in a domain with the approximate measurements of 2$\times$2$\times$5 Mm. The bottom seven planes of grid cells of the domain are located below the photosphere ($z=0$) to make space for the driver (see Sect. \[sec:numerics\]).
In the first step we define a divergence-free magnetic field, where the total field strength forms several straight loops of reduced magnetic field with a gauss-shaped cross section $$\label{eq:bxby_vertical}
B_{0,x}=B_{0,y}=0,$$ $$\label{eq:bz_vertical}
B_{0,z}=a \left[ 1-\sum_i^n{\exp \left\lbrace - \frac{ \left( x-x_{0,i} \right) ^2+ \left( y-y_{0,i} \right) ^2}{\sigma_i^2} \right\rbrace } \right] +b.$$ The sum corresponds to a sum over all loops, where $x_{0,i}$ and $y_{0,i}$ describe the coordinates of the loop centers and $\sigma_i$ define the thickness of each loop. We place four loops evenly inside the domain with a distance of 1 Mm in the $x$- and the $y$-direction between the loops and use $\sigma=1/3$ Mm for all $i$. The constants $a=309$ G and $b=5$ G define the strength of the magnetic field. The magnetic field has its minimum in the loop centers with value $b$, whereas the theoretical maximum outside the loops is $a+b$. However, owing to the tight structuring of our loops, this maximum is never reached. The resulting magnetic field ranges from 5 G to 300 G and is shown in Fig. \[fig:atmosphere\] at the top left. Since we only have a magnetic field component in the $z$-direction and the magnetic field does not change with height, $\vec{\nabla} \cdot \vec{B_0}=0$ is automatically fulfilled.
To study the effect of an inclined magnetic field as well, we rotate the vertical loop system from Equations \[eq:bxby\_vertical\] and \[eq:bz\_vertical\] clockwise around the y-axis by an angle $\theta$, which modifies the equations to $$\begin{aligned}
B_{0,x}&=&\tilde{B} \sin(\theta), \\
B_{0,y}&=&0, \\
B_{0,z}&=&\tilde{B} \cos(\theta),\end{aligned}$$ with $$\tilde{B}= a \left[ 1- \sum_i^n \exp \left\lbrace - \phi_i \right\rbrace \right] +b,$$ and $$\phi_i= \frac{ \left( -\sin \left( \theta \right) z+\cos \left( \theta \right) x-x_{0,i} \right)^2 }{\sigma_i^2}+ \frac{ \left( y-y_{0,i} \right)^2}{\sigma_i^2}.$$ We note that to keep the domain periodic, we extend the it in the $x$-direction and therefore also slightly change the corresponding resolution. That leads to a number of points in the $x$-direction of $n_x=$145 for $\theta = 15^{\circ}$. The total magnetic field for the case of $\theta = 15^{\circ}$ is shown in Fig. \[fig:atmosphere\] at the bottom left.
In order to calculate the pressure and density we split Equation \[eq:mhs\_equilibrium\] into its three components and slightly reorder these components as follows: $$\begin{gathered}
\label{eq:I}
\mathbf{I}: B_{0,y} \partial_y B_{0,x}+B_{0,z} \partial_z B_{0,x} -B_{0,y} \partial_x B_{0,y}-B_{0,z} \partial_x B_{0,z} \\ = \mu_0 \partial _x p_0, \end{gathered}$$ $$\begin{gathered}
\label{eq:II}
\mathbf{II}: B_{0,x} \partial_x B_{0,y}+B_{0,z} \partial_z B_{0,y} -B_{0,x} \partial_y B_{0,x}-B_{0,z} \partial_y B_{0,z} \\ = \mu_0 \partial _y p_0, \end{gathered}$$ $$\begin{gathered}
\label{eq:III}
\mathbf{III}: B_{0,x} \partial_x B_{0,z}+B_{0,y} \partial_y B_{0,z} -B_{0,x} \partial_z B_{0,x}-B_{0,y} \partial_z B_{0,y} \\ = \mu_0 \partial _z p_0 + \rho_0 g \mu_0, \end{gathered}$$ where $\partial x_j=\partial/\partial x_j, j={1,2,3}$. Reforming and integrating the first two components gives us $$\label{eq:pressure_calc}
p_0=\tilde{p}_I+f_1(y,z)=\tilde{p}_{II}+f_2(x,z)=\tilde{p}_I+h(z),$$ where $\tilde{p}_I$ and $\tilde{p}_{II}$ are the pressures calculated from integrating Equation \[eq:I\] and Equation \[eq:II\], respectively, and $f_1$ and $f_2$ are functions resulting from the integral that have to be determined. For our magnetic field model $\tilde{p}_I=\tilde{p}_{II}$, which leads to the last equality of Equation \[eq:pressure\_calc\]. The resulting expression for the pressure is $$\label{eq:pressure_expression}
p_0=- \frac{1}{2 \mu_0} \tilde{B}^2+h(z).$$ The function $h(z)$ is arbitrary, as it has no influence on Equations \[eq:I\] and \[eq:II\], and represents the vertical pressure stratification, and the constant that has to be added to make Equation \[eq:pressure\_expression\] positive. However, we stress the significance of choosing $h(z)$ wisely, as this term essentially determines our vertical density profile. Therefore, we define the vertical pressure stratification according to the VAL-C model and add an exponential term to modify the stratification $$\begin{gathered}
\label{eq:pressure_strat}
h(z)=p_{\mathrm{bot}} \exp \left\lbrace - \int \frac{1}{H(z)} dz \right\rbrace+ \\ 150 \mathrm{[Pa]} \exp \left\lbrace - \frac{z}{600 \cdot 6046 \mathrm{[m]}} +0.015 \right\rbrace + 422\mathrm{[Pa]}.\end{gathered}$$ In this case, $p_{\mathrm{bot}}=14$ kPa is the pressure at the bottom of the domain for the first term of Equation \[eq:pressure\_strat\]. The scale height $H(z)$ is calculated with a temperature profile that follows the VAL-C model until the transition region and has a constant temperature for the corona. The two regions of the temperature profile are connected by a cosine-shaped transition region. Since this initial temperature profile is only used for calculating the vertical pressure stratification and is changed in the next step, we abstain from mentioning the exact expression. The constants in the exponential term of Equation \[eq:pressure\_strat\] are related to our practical implementation and their exact values have no deeper meaning. Figure \[fig:profiles\] shows the pressure profile as a function of height for $\theta=0\degr$ for pressure according to the VAL-C model (dashed lines) and for pressure with added exponential term, as used in our model (solid lines). Both profiles have the same start and end points, but in the modified version the slope in the upper part of our model is larger, which leads to higher density values in that region. This is necessary as a consequence of the constant magnetic field along the loops the density would be very low otherwise, which would lead to unreasonably high temperatures. However, the additional term also leads to a strong broadening of our transition region, which we deem a compromise that has to be made.
As soon as the magnetic field and the pressure are known it is straight forward to get an expression for $\rho$ from Equation \[eq:III\]. Finally, we calculate the temperature from the ideal gas law. Figure \[fig:profiles\] (right) shows the density (dashed) and temperature (solid) profiles as a function of $z$ for the vertical case, whereas Fig. \[fig:atmosphere\] shows a 3D plot of the pressure and temperature for both the vertical and the $15^{\circ}$ inclined case. We note that the pressure is vertically much more stratified than horizontally structured.
{width="33.00000%"} {width="33.00000%"} {width="33.00000%"} {width="33.00000%"} {width="33.00000%"} {width="33.00000%"}
Since the pressure is higher in the loop interior, where the magnetic field is lower, we get a sound speed profile that is much higher inside the loops than outside. On the contrary, the Alfvén speed has its maximum outside the loops, while being very low in their center. This would be expected for a coronal loop with higher density than its surroundings, however, to fulfill Equation \[eq:mhs\_equilibrium\] the density is horizontally constant in our model. This is easily visible in Equation \[eq:III\] considering the vertical case with $B_{0,x}=B_{0,y}=0$ and taking into account that $\partial_z p_0$ is independent of $x$ and $y$ for $\theta=0\degr$. The pressure and magnetic field distributions lead to a horizontally strongly structured plasma-$\beta$ profile. The $\beta=1$ contour is plotted in Fig. \[fig:beta\]. As we go up in the atmosphere, the plasma-$\beta$ decreases in the loop exterior already below the transition region to lower than unity, while it is always much higher than unity in the loop interior until the top of the domain.
{width="33.00000%"} {width="33.00000%"}
Numerical setup {#sec:numerics}
===============
For our simulations we used the MANCHA3D code [@khomenko_collados_2006; @felipe_etal_2010; @khomenko_etal_2018] developed at the Instituto de Astrofísica de Canarias in Tenerife, Spain. This code solves the fully nonlinear magnetohydrodynamics equations for perturbed variables, which is why we initially had to define a MHS equilibrium to perturb it in the simulations. For the present work we consider an adiabatic system and neglect changes of the mean molecular weight due to ionization.
On the vertical faces of the computational box we set periodic boundary conditions. Our system can therefore be viewed as an ensemble of thin loops (or loop strands) extending infinitely to all horizontal sides that reasonably represent groups of spicules. To allow waves to escape, we set a Neumann-type zero-gradient open boundary condition at the upper boundary. At the first seven layers of the lower boundary we applied a driver that follows an analytic solution for a vertical gravity-acoustic wave [@mihalas_mihalas_1984]. The driver is described in detail by [@santamaria_etal_2015] and in this work we only repeat the general form, $$v_{z,1}=V_0 \exp \big \lbrace \frac{z}{2H}+k_{zi}z \big \rbrace \sin(\omega t -k_{zr}z),$$ $$\frac{p_1}{p_0}=V_0 |P| \exp \big \lbrace \frac{z}{2H}+k_{zi}z \big \rbrace \sin(\omega t -k_{zr}z +\phi_P) ,$$ $$\frac{\rho_1}{\rho_0}=V_0 |R| \exp \big \lbrace \frac{z}{2H}+k_{zi}z \big \rbrace \sin(\omega t -k_{zr}z +\phi_R) ,$$ where $v_{z,1}$, $p_1$, and $\rho_1$ are the velocity perturbation in the $z$-direction, pressure perturbation, and density perturbation, respectively. The values $|P|$ and $|R|$ are the relative amplitudes, while $\phi_P$ and $\phi_R$ are the phase-shifts of pressure and density perturbation compared to the velocity perturbation. The value $H$ is the pressure scale height and $V_0$ is the amplitude of the velocity perturbation. The vertical wave number $k_z$ is either complex or real, depending on the frequency $\omega$ compared to the isothermal acoustic cutoff frequency $\omega_c$, $$k_z=k_{zr}+ik_{zi}=\frac{\sqrt{\omega^2-\omega_c^2}}{c_s}$$ with $$\omega_c=\frac{\gamma g}{2c_s},$$ where $c_s$ is the sound speed and $\gamma=5/3$ is the adiabatic index.
In this work we used a small perturbation of $V_0=10^{-2}$ m/s to stay in the linear regime. In addition, we used a period of 100 s, which leads to a frequency of $\omega\approx0.063$ rad/s. The main reason for choosing this small period is that the imperfect open boundary conditions at the top cause waves to reflect and propagate downward. The chosen period allows us to study at least half of a wave period before the reflected waves interfere. For this period, $\omega > \omega_c$ is valid for the whole domain, so the waves excited from the bottom of the domain never reach a cutoff region. To put it into a solar context, the p-modes in our simulations are no longer trapped within the solar interior and can propagate through the chromosphere. Since the waves are not trapped within a resonant cavity, their amplitude is not amplified by constructive interference, which validates our choice of a small $V_0$. We plan to use larger periods that allow a cutoff region in future work.
Methods for data interpretation {#sec:methods}
===============================
Decomposition into components {#subsec:decomposition}
-----------------------------
In order to distinguish the different wave modes, it is necessary to bring our simulation data into a form that allows us to visualize the characteristics of the expected modes. [@tarr_etal_2017] decomposed their data into kinetic, acoustic, and magnetic energy densities, which allowed these authors to decouple fast from slow waves and magnetic from acoustic waves. Another decomposition method was carried out by [@khomenko_etal_2018], who, following [@cally_2017], constructed three quantities based on the physical properties of the waves: (1) $f_\mathrm{alf}$ for the incompressible perturbation propagating along the magnetic field, (2) $f_\mathrm{long}$ for the compressible perturbation propagating along the magnetic, and (3) $f_\mathrm{fast}$ for the compressible perturbation perpendicular to the magnetic field. While $f_\mathrm{alf}$ is a quantity describing the Alfvén waves for all $\beta$, $f_\mathrm{long}$ and $f_\mathrm{fast}$ decouple the slow and the fast magneto-acoustic waves only for $\beta<1$.
However, since we have both $\beta<1$ and $\beta>1$ regions in our model and expect tube modes to be excited, we adopted the approach of [@mumford_etal_2015], who split the velocities and fluxes into three orthogonal components defined by the magnetic flux surfaces. These components are defined by the unit vectors longitudinal ($\vec{\hat{e}}_\parallel$), azimuthal ($\vec{\hat{e}}_a$), and normal ($\vec{\hat{e}}_\perp$) to those surfaces. Since we only used small perturbations, magnetic flux surfaces and therefore also the resulting unit vectors for the three components are constant in time. The unit vector for the longitudinal component $\vec{\hat{e}}_\parallel$ points into the direction of the magnetic field and is therefore easy to compute. Because all field lines in our model are straight, this vector is the same for all points of the domain.
Calculating the azimuthal unit vector $\vec{\hat{e}}_a$ proves to be more difficult. We solve it by calculating the 2D isocontour of the magnetic field for all horizontal layers and fitting a straight line to the contour for each pixel. The direction of the linear fit is then the direction of $\vec{\hat{e}}_a$. For the inclined case $\vec{\hat{e}}_a$ gets an appropriate vertical component to be normal to $\vec{\hat{e}}_\parallel$. However, with this method all azimuthal vectors point to the positive $x$-direction, which results in half of the vectors being clockwise (top half of the loops in respect to $y$), while the others are counterclockwise (bottom half of the loops in respect to $y$). Because of the regular structure of the loop system it is simple to distinguish those regions. We multiply all clockwise azimuthal unit vectors with -1 to have a consistent sense of direction. Finally, the normal unit vector $\vec{\hat{e}}_\perp$ is calculated by $\vec{\hat{e}}_\perp=\vec{\hat{e}}_a \times \vec{\hat{e}}_\parallel$. With this convention, $\vec{\hat{e}}_\parallel$ always points upward (and for the inclined case also into the positive $x$-direction), $\vec{\hat{e}}_a$ is parallel to the flux surfaces and points to the counterclockwise direction, and $\vec{\hat{e}}_\perp$ points away from the loop centers. Figure \[fig:pna\_cartoon\] sketches the vector directions for vertical and inclined loops.
![Orthogonal decomposition vectors parallel ($\vec{\hat{e}}_\parallel$), normal ($\vec{\hat{e}}_\perp$), and azimuthal ($\vec{\hat{e}}_a$) to the magnetic iso-surfaces (cylindrical shapes) for a vertical (left) and inclined (right) loop. \[fig:pna\_cartoon\]](pna_cartoon-eps-converted-to.pdf){width="50.00000%"}
Expected wave modes {#subsec:expected_modes}
-------------------
Since our atmosphere in not horizontally uniform but has a cylindrical shape, we would also expect the wave modes excited in our simulations to be those of a plasma with cylindrical shape, such as the $m=0$ sausage mode and $m=1$ kink mode, where $m$ is the azimuthal wave number. To approximately calculate what wave modes would appear in our setup, we assume a simple vertically constant cylinder with radius $R$ and internal values $f_i$ embedded in an external plasma with values $f_e$. Similar to our atmosphere, we assume that the magnetic field is parallel to the loop axis. We then followed the mathematical framework of [@moreels_vandoorsselaere_2013]. As external and internal values for this simplified model we used the average external and average internal values of our atmosphere at a height of 2 Mm, where the loop boundary is defined by the $\beta=1$ layer. If $c_{s,i}$ is the internal sound speed, this leads to an external sound speed of $c_{s,e}=0.745 c_{s,i}$, an internal Alfvén speed of $c_{A,i}=0.618 c_{s,i}$, and an external Alfvén speed of $c_{A,e}=0.994 c_{s,i}$. Figure \[fig:phase\_speed\_diagram\] shows the resulting phase speed diagram, where the internal and external sound and Alfvén speeds are indicated by horizontal gray lines. Also plotted are the internal and external tube speed $$c_{T,f}=\frac{c_{s,f}c_{A,f}}{\sqrt{c_{s,f}^2+c_{A,f}^2}}$$ and the kink speed $$\label{eq:kink_speed}
c_k=\sqrt{\frac{\rho_{0,i}c_{A,i}^2+\rho_{0,e}c_{A,e}^2}{\rho_{0,i}+\rho_{0,e}}}.$$ For both sausage and kink waves, we only get non-leaky solutions for fast surface waves and slow body waves. Slow surface waves are theoretically possible below $c_{T,i}$, but no solutions are found in this region. Fast body modes could occur above $c_{s,i}$, but they are leaky; we indeed find leaky solutions for a kink body mode there for higher $kR$. The phase speed line of the fast sausage surface mode (red dotted line) actually stops where $kR$ is approximately 0.8 as there are no solutions found for low $kR$, not even for leaky waves.
![Phase speed diagram of sausage modes and kink modes for conditions similar to the model atmosphere at a height of 2 Mm. The horizontal gray lines indicate various characteristic speeds. \[fig:phase\_speed\_diagram\]](phase_speed_diagram_height22-eps-converted-to.pdf)
It is now possible to calculate the magnitude of the ratio of longitudinal displacements to perpendicular displacements for the modes shown in Fig. \[fig:phase\_speed\_diagram\]. Unlike the normal component in the decomposition we described in Sect. \[subsec:decomposition\], perpendicular displacement describes plasma displacement in all directions perpendicular to the loop axis, so $\xi_{\mathrm{perp}}=\sqrt{\xi_a^2+\xi_\perp^2}$. This ratio depends on the distance to the loop axis $r$ and is shown in Fig. \[fig:displacement\_ratio\] at $r=0.69R$ (left) and at $r=R$ (right). From the figures it is clear that, although the curves slightly change, the general behavior of the displacement ratios stays the same regardless of the chosen point inside the loop or the loop surface. While the slow body modes and the fast sausage surface mode have much higher parallel displacements than perpendicular displacements for small $kR$ compared to larger $kR$, it is the opposite for the fast kink surface mode. This makes the fast sausage surface mode and the fast kink surface mode easily distinguishable from each other.
Results and discussion {#sec:results}
======================
Wave propagation for vertical flux tubes
----------------------------------------
We first study simulations with the vertical ($\theta=0^{\circ}$) case. Our goal is to investigate the conversion of p-modes that arrive at the corona. For that purpose we look at the evolution of a horizontal cut through the domain at a height of 2 Mm. We later show that we have some issues with wave reflection from the upper boundary, so investigating the wave behavior at 2 Mm instead of higher up allows us to analyze a longer time sequence before the reflections from the upper boundary intervene. This is possible because the wave behavior does not change much above the transition region after 2 Mm, which is the case even though the high $\beta$ regions inside the loops are still merged together at that height. The movie showing the time development of the horizontal cut before the reflections from the upper boundary arrive ($ t \le 170$ s) is available online. A screenshot of the movie is shown in Fig. \[fig:pna\_horizontal\_0deg\]. It shows the three velocity components at $t=106$ seconds together with the $\beta=1$ contour. For this time series the longitudinal velocity perturbation is always larger inside the loop than outside. The waves arrive at 2 Mm at approximately the same time; the waves inside the loop arrive slightly earlier. The maximum perturbations of the normal component are more than one order of magnitude smaller than the maximum perturbations of the parallel component. In the first part of the time series, the normal component has a positive sign everywhere and thus shows an expansion of the whole loop cross section with some normal velocity components outside the loops as well. In its first maximum at $t=106$ s (as indicated in Fig. \[fig:pna\_horizontal\_0deg\]) the expansion of the loop has similarities with a $m=4$ fluting mode. However, this mode would also require some plasma to flow into the loops at the top, bottom, and sides of the loops. At around $t=114$ s the normal component changes its sign and at $t=128$ s it looks the same as in Fig. \[fig:pna\_horizontal\_0deg\] but with changed sign (i.e., contraction instead of expansion). We therefore conclude that the wave the p-modes excited is in fact a $m=0$ sausage mode that is deformed by the tight packing of the grid-like positioned loops. The deformed sausage modes are very similar to a superposition of a $m=0$ sausage mode with a $m=4$ fluting mode.
Immediately apparent in the azimuthal component of Fig. \[fig:pna\_horizontal\_0deg\] is the ring-like structure of counter-flowing plasma velocities close to the $\beta=1$ layer. However, this ring structure propagates from outside of the loop inward and is only coincidentally at the position of the $\beta=1$ layer for this screenshot. This propagation inward is only apparent [@raes_etal_2017], since it results from a cone-shaped area of high $c_A$ propagating upward. Those waves are probably Alfvén waves that are excited by the first impulse of the driver. Other than that, stationary counter-streaming regions appear around the $\beta=1$ layer with the same periodicity as the normal component. To help visualize the direction of the velocity perturbations we overplot the color scale for the parallel component with vectors showing the horizontal velocity perturbation (Fig. \[fig:vp\_with\_vectors\]) for the same time and height as in Fig. \[fig:pna\_horizontal\_0deg\]. It is now obvious that the plasma expands from the loop centers toward the centers of the loop exteriors (and half of a period later vice versa). Where the loops are closest to each other, counter-streaming flows get deflected sideways toward the centers of the loop exteriors. This also solidifies our interpretation of a deformed sausage wave as mode identification.
{width="\textwidth"}
![Vectors of the horizontal velocity perturbations at 2 Mm at time $t=106$ s. The color scale shows the longitudinal velocity component in m/s and the black contours show the $\beta=1$ border. The spatial scales are given in Mm. \[fig:vp\_with\_vectors\]](image_v_paper_2Mm_0_100_gridgaus_0053-eps-converted-to.pdf)
In order to study the wave propagation through the system, we looked at the data of two vertical lines in the domain: one of the lines is located in the loop interior where there is a low magnetic field, while the other is close to the center of the loop exterior, which has the maximum magnetic field. The vertical line in the loop interior is at a distance of $0.69R$ from the loop center and therefore at the same position in our MHS atmosphere as the ratio of displacements in Fig. \[fig:displacement\_ratio\] (left) in the simplified model of Sect. \[subsec:expected\_modes\]. We refrain from using data from the exact center of the loops and loop exteriors because the azimuthal unity vector $\vec{\hat{e}}_a$, and therefore also for the normal unity vector $\vec{\hat{e}}_\perp$, are not defined there. The positions of the vertical lines are shown in Fig. \[fig:line\_location\], while the velocity components in these lines are plotted as a function of time in Fig. \[fig:wave\_propagation\_vertical\]. In the latter figure, four characteristic speeds are indicated: the local sound speed (dashed black line), local Alfvén speed (dotted black line), local tube speed (solid black line), and kink speed (dashed dotted line). As seen in Equation \[eq:kink\_speed\], the kink speed is calculated by external and internal Alfvén speed and density. Those values are the mean of the values inside and outside the loop for each height with the border at $\beta=1$. Since the loop width is very constant above the transition region, the kink speed is only plotted from the transition region upward.
The plots on the left side of Fig. \[fig:wave\_propagation\_vertical\] show the velocity components in the loop interior, where there is high $\beta$ for all heights. For the longitudinal component inside the loop (top left) the waves propagate smoothly with approximately the sound speed or kink speed and do not seem to be disturbed by the transition region (region between red dotted lines), except for some slight reflection, which is visible between 110 to 130 seconds. No features propagating with slower speeds are visible. However, if we look onto the same component but in the loop exterior (Fig. \[fig:wave\_propagation\_vertical\] top right), we see a different picture. There, the waves have to travel through the $\beta=1$ layer before going through the transition region. Below this border, where $\beta \gg 1$, the waves travel again with the sound speed as for the loop interior, but as soon as the first waves pass the $\beta=1$ layer there are suddenly wave features that travel with the Alfvén or tube speed. Following the definition for mode conversion and transition of [@cally_2005], this could be interpreted as a conversion from fast acoustic waves to fast magnetic waves. However, we have to be cautious when calling a wave acoustic or magnetic above the transition region in our model because we are dealing with tube waves with high $\beta$ inside the loops and low $\beta$ outside the loops. In addition, we also see features propagating with the sound speed or tube speed above the $\beta=1$ layer, which seems like a transition from fast acoustic waves to slow acoustic waves at first sight. Similar to the waves inside the loop, there is a sign of reflection from the transition region, which is best visible at a time of about 135 seconds. However, we also have unwanted reflections from the upper boundary, which distort the wave shapes coming from below.
Compared to the longitudinal velocity components, the normal velocity components (Fig. \[fig:wave\_propagation\_vertical\] middle) are about one order of magnitude smaller. This is no surprise, as only the longitudinal component is driven at the bottom of the domain, while the other components arise from mode conversion or coupling due to inhomogeneity. The general behavior is similar to the longitudinal component, where there are only waves propagating with the sound speed or kink speed in the loop interior and a combination of slow and fast waves in the loop exterior. However, apart from the much stronger reflections from the transition region compared to the general amplitude and the much less prominent reflections from the upper boundary, two striking effects appear. The first is the high velocity amplitude around the lower border of the transition region, which we do not investigate in this paper. From the first high amplitude wave, a wave is launched that travels faster than the fastest characteristic speed (i.e., sound speed for the loop interior and Alfvén speed for the loop exterior) until it vanishes. This could be a sign of a leaky sausage wave, as these waves can travel faster than the external Alfvén speed [@pascoe_etal_2007], which would be the Alfvén speed in the center of the loop exterior for our case. By looking closer at the simulation data we find indications that this is indeed the case. We did not find a leaky sausage mode with high phase speed for our simplified model in Sect. \[sec:methods\], however, it could still appear in our more complicated MHS model. The second effect are the vertical stripe patterns, which are more pronounced inside the loop than outside. These may either be interference patterns by partial reflection of the waves due to the temperature gradient (the vertical temperature gradient is stronger inside the loops than outside), or just artifacts due to the cylindrical structure within a Cartesian grid. Those patterns are not of interest for our study and are therefore not considered in the following.
The azimuthal velocity component (Fig. \[fig:wave\_propagation\_vertical\] bottom) closely resembles the normal component, except for the generally smaller amplitude and that we now also see waves with a phase speed of approximately the Alfvén speed inside the loop. In addition, this component is more strongly affected by the reflection from the upper boundary. The close relation between the normal and azimuthal component is no surprise, as Fig. \[fig:vp\_with\_vectors\] shows us that the azimuthal velocities arise from the deflection of expanding (normal) plasma movements toward the centers of the loop exteriors.
To roughly estimate how much of the wave energy is reflected at the transition region in Figure \[fig:wave\_propagation\_vertical\], we determine the wave energy flux parallel to the magnetic field and look at the ratio of maximum (positive, upward) to minimum (negative, downward) amplitude of the first wave front. This shows that about 3% of the flux is reflected in the loop interior and about 4% in the loop exterior. Because of this crude approximation there might be an error of several percent, but the reflected flux is still very low. For a steeper transition region, more reflection would be expected. We refrain from giving an estimate about the amount of flux reflected from the upper boundary, since this is a purely numerical issue that holds no physical value.
![Location of the vertical lines of Fig. \[fig:wave\_propagation\_vertical\]. The black lines show the $\beta=1$ contour at a height of 2 Mm, where there is high $\beta$ inside and low $\beta$ outside the “circles”. The red and blue crosses denote the location of the vertical line in the loop interior (Fig. \[fig:wave\_propagation\_vertical\] left) and the loop exterior (Fig. \[fig:wave\_propagation\_vertical\] right), respectively. \[fig:line\_location\]](line_location-eps-converted-to.pdf)
{width="50.00000%"} {width="50.00000%"} {width="50.00000%"} {width="50.00000%"} {width="50.00000%"} {width="50.00000%"}
From Fig. \[fig:pna\_horizontal\_0deg\] we get an approximate loop radius of $R \approx 0.5$ Mm and from \[fig:wave\_propagation\_vertical\] we can estimate that the wavelength lies between 5 and 10 Mm. From those values we determine that $kR$ lies between 0.31 and 0.63, which allows us to compare our simulation data to the wave mode solutions of the simplified model in Figs. \[fig:phase\_speed\_diagram\] and \[fig:displacement\_ratio\] for small $kR$. All our data suggests that the parallel displacement is larger than the perpendicular displacement, which immediately excludes the fast kink surface mode. We also exclude the slow kink body mode because there is no reason why a kink mode would be excited owing to the symmetry of our system. We therefore conclude that the fast waves we see in Fig. \[fig:wave\_propagation\_vertical\] are fast sausage surface waves, which is supported by the fact that the magnitude of the velocity perturbations is smaller in the loop center than in its surroundings, while the slow waves are slow sausage body waves. However, for small $kR$ there is no solution of fast sausage surface waves (Fig. \[fig:phase\_speed\_diagram\]), so the simple model we used to calculate the phase speed diagram does not describe those wave modes well in that region. In addition, we might still have a plane-like wave traveling through the whole domain, which mostly ignores the horizontal structuring because of its relatively slow changes. It is, however, difficult to distinguish such waves from the fast surface sausage waves.
Wave propagation for inclined flux tubes
----------------------------------------
We now study the case with the inclined magnetic field for an inclination of $\theta = 15^{\circ}$. The time development of a horizontal cut of the simulation box with $15^{\circ}$ inclined magnetic field at a height of 2 Mm is shown in the movie that is available online. A snapshot of this movie at $t=112$ s is shown in Fig. \[fig:pna\_horizontal\_15deg\]. The parallel velocity component in Fig. \[fig:pna\_horizontal\_15deg\] behaves very similarly to the vertical case (Fig. \[fig:pna\_horizontal\_0deg\]), where there are higher velocity perturbations within the loops than outside, which change signs over time. However, both the normal and azimuthal components show not only a similar stripe pattern, but also have the same magnitude. These effects arise from the whole plasma moving first to the left (like in the figure) and later to the right[^1], which corresponds to a kink wave. During the transition from plasma moving from one direction to the other, we have a short time span in which plasma flows both to the left and right. Another kind of fast wave was also excited because we expect the magnitude of perpendicular displacement to be larger than the magnitude of parallel displacement for fast kink surface waves (see Fig. \[fig:displacement\_ratio\]); however, we find the opposite in Fig. \[fig:pna\_horizontal\_15deg\]. This wave is either a fast sausage surface wave, such as in the case with vertical flux tubes or a plane wave that does not “feel” the loop structure.
We checked the kink wave assumption by plotting the velocity disturbance perpendicular to the magnetic field in the loop center as a function of the height, where we find upward propagating waves for the inclined cases with $\theta=15^{\circ}$ and $\theta=30^{\circ}$, whereas there is no kink wave for the vertical case ($\theta=0^{\circ}$). These kink waves are excited in the simulations with the inclined loops, as the driver is still purely vertical and therefore gives the loops a push perpendicular to the magnetic field. Since the magnetic field is only inclined in the $x$-direction, there is no significant perturbation of the loop centers in the $y$-direction. A snapshot of these waves is shown in Fig. \[fig:kink\_oscillation\].
{width="\textwidth"}
![Velocity perturbation perpendicular to the magnetic field in the loop center as a function of height at time $t=118$ s for three different magnetic field inclinations.\[fig:kink\_oscillation\]](kink_oscillation_0059-eps-converted-to.pdf)
Figure \[fig:wave\_propagation\_inclined\] shows the equivalent of Fig. \[fig:wave\_propagation\_vertical\] for the inclined case, for which we examine the wave propagation along two lines inclined $15^{\circ}$ from the vertical (parallel to the magnetic field) that lie within the loop interior and loop exterior, respectively, at equivalent locations as for the vertical case. The black lines in Fig. \[fig:wave\_propagation\_inclined\] show the local sound speed (dashed), local Alfvén speed (dotted), local tube speed (solid), and kink speed (dashed dotted) for wave propagation along the loop, i.e., along the magnetic field. Similarly, the brown lines show these speeds for vertical wave propagation, i.e., the direction of the driver. The reason for the strange S-shape for the vertical propagation speeds is that for different heights $z$ vertically propagating waves that reach that height also start from different horizontal locations $x$, where the characteristic speeds are different. Therefore, for some vertical lines, the waves would find more “favorable” conditions to propagate than for others, which causes them to arrive earlier at a larger height. The S-shape is more pronounced for the Alfvén speed (and therefore also the tube speed) because the Alfvén speed changes more between loop interior and exterior than the sound speed. Since vertically propagating waves would have to go through regions with very low Alfvén speed for the $y$ location of the loop-interior-line (left column), the S-shape is extreme in those plots. However, we do not see any signs in Fig. \[fig:wave\_propagation\_inclined\] of waves propagating along these strange (brown) lines. Therefore, we can conclude that the waves mainly propagate along the magnetic field (black lines) from the transition region onward.
{width="50.00000%"} {width="50.00000%"} {width="50.00000%"} {width="50.00000%"} {width="50.00000%"} {width="50.00000%"}
From Fig. \[fig:wave\_propagation\_inclined\] it is immediately apparent that the reflection problem from the upper boundary is much stronger for the inclined case. For this reason the plots for the parallel component (top row) and for the azimuthal component (bottom row) are saturated to allow a better visibility of the waves propagating upward. What is also noticeable is that the magnitudes of the normal and azimuthal components are much higher than before, as also seen in Fig. \[fig:pna\_horizontal\_15deg\].
As in the vertical case, the parallel component propagates with the sound speed or kink speed in the loop interior for all heights and splits at the $\beta=1$ layer into waves propagating with the sound speed or kink speed and waves propagating with the Alfvén speed or tube speed in the loop exterior. In the plots for the normal component we again see (small) wave signatures that propagate faster than the local fast speed in the transition region at about $t=180$ s, which could be a sign of leaky sausage waves. These wave signatures are also present in the azimuthal component, but not visible in the displayed data range. In both the normal and azimuthal component we only see wave propagation with the sound speed or kink speed for high $\beta$ and the Alfvén speed or kink speed for low $\beta$.
There is again some wave reflection from the transition region, which is about 3% of the energy flux in the loop interior and 6% in the loop exterior. Given that this is just an estimation, it is not possible to say if the reflection is dependent on the inclination angle, since these values are very similar to those obtained for the vertical case (3% and 4%).
Similar as before in the vertical case, we compare our simulation results with Figs. \[fig:phase\_speed\_diagram\] and \[fig:displacement\_ratio\]. We found from Figs. \[fig:pna\_horizontal\_15deg\] and \[fig:kink\_oscillation\] that definitely a kink wave is excited, and from Fig. \[fig:wave\_propagation\_inclined\] that it propagates fast. Therefore, we identify it as a fast kink surface wave. To explain the ratios of velocity perturbation components in Fig. \[fig:pna\_horizontal\_15deg\] there must also be another fast wave, which could be a fast sausage surface wave or a plane fast wave that ignores the cylindrical structuring. The slow waves appearing in Fig. \[fig:wave\_propagation\_inclined\] are symmetric around the center of the loop exterior and are therefore identified as slow sausage body waves.
To reiterate the results of this section, we inserted a fast (acoustic) wave at the bottom of our domain, which converted to different wave modes. This was also found by [@cally_2017], who injected a fast (magnetic) wave into a model with gravitationally stratified Alfvén speed profile with discrete (and also “touching”), inclined flux tubes, using the cold plasma ($\beta=0$) approximation. The initially $m=n=0$ fast waves scattered in Fourier space into other modes, i.e., essentially the $m=0,n=-1$ kink mode. In addition, there was also significant conversion to Alfvén waves, which decayed with height as they were also scattered into higher mode numbers as a consequence of mode mixing.
Mode conversion {#subsec:mode_conversion}
---------------
We would like to estimate how much the initially acoustically dominated waves, as excited by the driver, take on magnetic properties. This can be described by mode conversion from acoustic to magnetic waves. When speaking about mode conversion in the following, we mean conversion from acoustic to magnetic behavior, not conversion from fast to slow waves or vice versa. An often used conversion coefficient was given by Equation 26 of [@schunker_cally_2006] and we repeat for convenience, i.e., $$\label{eq:schunker_conversion}
C=1-T=1-\exp \Big [ - \frac{\pi k^2 k_\perp^2}{|k_z| (k^2+k_\perp^2)} \Big (\frac{d(c_A^2/c_s^2)}{dz}\Big ) ^{-1} \Big ] _{c_A=c_s} .$$ The accuracy of this expression was analytically tested by [@hansen_cally_2009]. It is defined by the portion of the wave energy flux that is converted from acoustic to magnetic, where $C=1$ (transmission coefficient $T=0$) describes full conversion from acoustic to magnetic waves (fast wave to fast wave) and $C=0$ ($T=1$) describes no conversion. In this equation, $k$ is the wave number with its components in the $z$-direction $k_z$ and the direction perpendicular to the magnetic field $k_\perp$. The equation is evaluated at the $c_A=c_s$ layer, where mode conversion is supposed to happen. In the following, we check Equation \[eq:schunker\_conversion\] in our simulation data.
We calculate the conversion coefficient at three different points at the $c_A=c_s$ layer. All three points lie in the loop exterior below the transition region and are located at different arbitrary distances from the center of the loop exterior. Since the excited acoustic waves propagate vertically when below the transition region, we can assume that $k_z=k$ and $k_\perp=k \sin(\theta)$. Furthermore, we assume that $k=\omega / c_{\mathrm{eq}}$, where $\omega$ is our driver frequency and $c_{\mathrm{eq}}$ is the sound or Alfvén speed at the equipartition layer. The resulting conversion coefficients are plotted in Fig. \[fig:conversion\] (top) for various inclination angles in 5$\degr$ intervals as solid red lines. Two of those lines are more similar because the points in which they were calculated lie closer to each other than to the third point. The mode conversion clearly increases with increasing angle, as we expected. At a field inclination of about 15$\degr$ the curvature changes from positive to negative.
![*Top:* Conversion coefficient describing mode conversion from acoustic to magnetic waves as a function of inclination angles. The full red lines are mode conversion according to [@schunker_cally_2006] (Equation \[eq:schunker\_conversion\]) in three different points within the loop exterior at the equipartition layer. The dotted black line is calculated according to Equation \[eq:conversion2\] at $z=2$ Mm and averaged over the whole horizontal plane. The dashed black line is the same, but only considering fluxes in regions with $\beta <1$ (see text). *Middle:* Absolute error of the red curves [@schunker_cally_2006] compared to the dashed curve (Eq. \[eq:conversion2\] for $\beta<1$). *Bottom:* Same, but relative error. \[fig:conversion\] ](mode_conversion-eps-converted-to.pdf)
To compare the outcome of Equation \[eq:schunker\_conversion\] with a reasonable quantity of our simulation results, we use the mean acoustic and mean magnetic energy flux defined by [@bray_loughhead_1974], $$\vec{F}_{ac}=\langle p_1 \vec{v}_1 \rangle,$$ $$\vec{F}_{mag}=\langle \vec{B}_1 \times (\vec{v}_1 \times \vec{B}_0) \rangle / \mu_0,$$ where $p_1$, $\vec{v}_1$, and $\vec{B}_1$ are the perturbations of pressure, velocity, and magnetic field, respectively, and $\vec{B}_0$ is the equilibrium magnetic field. The (outer) brackets denote the average over time, however, for the following we also average over space. We now define a quantity $$\label{eq:conversion2}
C_{f}=\frac{|\vec{F}_{mag}|}{|\vec{F}_{mag}|+|\vec{F}_{ac}|},$$ which should tell us how much of the energy flux, which is initially fully acoustic, was converted into magnetic energy flux. Since we want to know how much our waves are dominated by which kind of energy flux, without taking into account the direction, we avoid positive and negative fluxes to be canceled out by taking the absolute value before averaging the fluxes. The value $C_{f}$ has the same properties as $C$ given that it is 1 for full conversion and 0 for no conversion. The dotted black line in Fig. \[fig:conversion\] (top) shows $C_{f}$ averaged in time over half of a period at a height of 2 Mm, starting from the time when the first wave reaches that height, and averaged in space over the whole horizontal plane. The dashed black line shows the same, but only averaged in space over the areas where $\beta<1$. The latter line shows higher values, as there is more flux converted in the considered regions than in the rest of the plane because there the waves have to travel through the $\beta=1$ (and $c_A=c_s$) layer. That line is therefore more comparable to mode conversion in the three chosen points than the dotted line. Like before, we see more conversion from acoustic to magnetic waves with increasing magnetic field inclination $\theta$ and a change of curvature at around $\theta=15\degr$.
By comparing $C_{f}$ (Equation \[eq:conversion2\]) with $C$ ([@schunker_cally_2006], Equation \[eq:schunker\_conversion\]) it is apparent that the curve of the latter conversion coefficient shows a higher inclination for increasing $\theta$ than the former. Fig. \[fig:conversion\] shows the absolute (middle) and relative (bottom) errors of $C$ compared to $C_f$ (for $\beta<1$). Between inclination angles of around $10\degr$ and $30\degr$ the relative error stays within $\pm 40 \%$. Below $10\degr$ the relative error is due to the small values of $C$ and $C_f$ much higher, however, the absolute error stays within the interval \[-0.1,0\]. In general, both conversion coefficients show the same qualitative behavior and are much more similar than we expected. This result is interesting because in our simulations we do not just have simple plane waves, except perhaps right above the driver at low heights along with some additional plane-like waves higher up. Instead, we mainly have waves that are modified by the cylindrical structure of the atmosphere, in particular sausage waves and kink waves. The local analysis around a point at the $c_A=c_s$ layer of [@schunker_cally_2006] allows the use of the simple analytic formula in Equation \[eq:schunker\_conversion\] for those cases as well.
We note that another, less general conversion coefficient was given by [@cally_2005]. It yields the same results as in Fig. \[fig:conversion\] for $\theta \le 10\degr$, but deviates from Equation \[eq:schunker\_conversion\] for higher inclination angles, as the curve does not change its curvature.
Summary and conclusions {#sec:summary&conclusions}
=======================
In this paper we presented a simple method to calculate a 3D MHS equilibrium when a divergence-free magnetic field is given. We built an equilibrium model resembling the solar atmosphere from the photosphere to the lower corona, including four flux tubes of decreased Alfvén speed and increased sound speed with the inclination $\theta$ from the vertical. This led to a tube-like plasma-$\beta=1$ layer with $\beta>1$ inside the tubes and everywhere in the bottom layers, and $\beta<1$ outside the tubes starting from a certain height. We then perturbed the plasma at the bottom with vertically polarized gravity-acoustic waves according to an analytical solution. We investigated the resulting waves by studying the behavior of the velocity perturbations parallel to the magnetic field, perpendicular to the magnetic iso-surfaces and azimuthal to these surfaces in a horizontal plane above the transition region. In addition, we studied the propagating velocity disturbances in two different lines (inside and outside the loop) along the magnetic field. By comparing our results with a simplified model of waves in a cylindrical structure, we could classify the waves appearing in our simulations. For the vertical case ($\theta=0\degr$), where the magnetic field and flux tubes are oriented along the driver polarization direction, we identified deformed fast sausage surface waves and slow sausage body waves. There might have additionally been a plane-like wave excited, which is difficult to distinguish from the fast sausage surface mode. For the inclined flux tubes and magnetic field, where the driver polarization now has a component perpendicular to the tubes, a fast kink surface wave is excited in conjunction with either a (deformed) fast sausage surface wave or a plane-like wave. Moreover, we also find slow sausage body waves.
In addition, we investigated the mode conversion from the initially acoustic waves to magnetic waves. We compared the outcome of a simple formula for a mode conversion coefficient by [@schunker_cally_2006] with the ratio of the magnetic energy flux to the sum of the magnetic and acoustic energy flux. We find that both methods give similar results with a maximum absolute error of 0.1 for inclination angles from $\theta=0\degr$ to $10\degr$ and a maximum relative error of 40% for angles from $\theta=10\degr$ to $30\degr$. The deviation of the simple formula from the other method is remarkably small, given that we anticipated a large influence of the cross-field wave speed structuring. This validates the frequent use of the simple formula. We note, however, that the influence of a cutoff region was not tested in the present work.
According to our simulations, vertical gravity-acoustic waves from the photosphere are converted to waves with partial magnetic properties in areas with flux tubes (especially in between the tubes), if the magnetic field lines and the flux tubes are inclined from the vertical. In that case the initially vertically propagating plane waves changed direction to propagate along the magnetic field above the transition region and were transformed into kink and sausage modes. In the case with vertical magnetic field and flux tubes, we only observed sausage waves without any significant magnetic wave properties.
There are some important limitations of this work we would like to mention. First, because of our model containing straight flux tubes, the magnetic field does not change along the loops to satisfy $\nabla \cdot \vec{B}=0$. However, this leads to an unusually high magnetic field strength of 300 G in the corona. In addition, there are regions in our model with $\beta>1$ in the corona, which is a much higher value than expected for that height [see, e.g., @gary_2001]. Realistic flux tubes are expected to strongly expand between photosphere and lower corona. Such a model would not only allow the magnetic field to decrease with height, but the expansion would also affect waves guided along the flux tubes. We assume the biggest change would be that the wave fronts are refracted along the field lines and would broaden, which would lead to damping of the waves. This was also mentioned in the results of [@mumford_etal_2015]. In fact, we assume the simple geometry of our flux tubes to be the main reason why the damping of the waves in our simulations is far smaller than in the observations of [@grant_etal_2015]. Despite these drawbacks we decided to first study straight flux tubes as they simplify the analysis of the excited waves. A similar study with expanding flux tubes is currently in progress.
A second limitation of this study are the high frequency and low amplitude of our driver. While we do not think that a higher amplitude would change the core of our results, we assume that a lower frequency of the driver with a realistic period of about five minutes leads to less waves being transmitted into the corona. This is because of the acoustic cutoff region, which prohibits the propagation of acoustic waves below the cutoff frequency. Instead, many waves would be reflected from that region. Since the wavelength in the current study is already much larger than the flux tube radius, we do not expect other big changes by decreasing the frequency. We plan to use a driver period of approximately five minutes in future work.
We would like to thank Paul Cally for helpful comments and suggestions. Furthermore, we would like to thank the referee for helping us to improve the manuscript. This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No. 724326).
[^1]: The vector normal to the flux surface $\vec{\hat{e}_\perp}$ does also have a component in the $z$-direction for $\theta \neq 0$, so strictly speaking the plasma also moves up and down.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In this note, we prove some new entropy formulae for linear heat equation on static Riemannian manifold with nonnegative Ricci curvature. The results are simpler versions of Cao and Hamilton’s entropies for Ricci flow coupled with heat-type equations.'
address: |
Rowland Hall\
University of California\
Irvine, CA 92697
author:
- Yucheng Ji
date: 'May 1st, 2017'
title: Remarks on Entropy Formulae for Linear Heat Equation
---
Introduction
============
In G. Perelman’s remarkable paper [@P], he derived an entropy formula for Ricci flow coupled with conjugate heat equation. This formula played an important role in the proof of Poincaré conjecture. Inspired by Perelman’s result and the early work of P. Li and S.-T. Yau [@LY] on the Harnack inequality for linear heat equation, L. Ni [@N] proved an entropy formula for linear heat equation on Riemannian manifold with static metric, and also obtained some geometric applications. His result can be seen as ‘the linear version’ of Perelman’s entropy.
On the other hand, also motivated by Perelman’s work, X. Cao and R. Hamilton [@CH] and X. Cao [@C] proved some new entropy formulae for Ricci flow. To illustrate the difference bewteen their work and Perelman’s, we shall recall that Perelman’s entropy is for the fundamental type (heat-kernel type) solution of conjugate heat equation (under Ricci flow). In contrast, Cao and Hamilton considered both forward and backward heat equations with potentials, and general positive solutions instead of fundamental type solutions of those heat equations.
In this note, we will combine the ideas of Ni and of Cao and Hamilton. Namely, we are going to prove some entropy formulae for linear heat equation which are counterpart of entropy formulae for Ricci flow proved by Cao and Hamilton, just as Ni’s entropy is counterpart of Perelman’s entropy in the case of linear heat equation on static manifold. Namely, we will prove:
Let $(M,g)$ be a closed Riemannian manifold of dimension $n$ and with nonnegative Ricci curvature. Let $f$ be a positive solution to the heat equation $$\frac{\partial f}{\partial t}=\Delta f.$$ Here the Laplacian is defined as $g^{ij}\partial_{i}\partial_{j}$. Let $f=e^{-u}$, and we define $$F=\int_{M}(t^2{\left\lvert{\nabla}u\right\rvert}^2-2nt)e^{-u}dV_{g},$$ then for all time $t>0$, $$F\leq0$$ and $$\frac{\partial F}{\partial t}\leq0.$$
and
Let $(M,g)$ be a closed Riemannian manifold of dimension $n$ and with nonnegative Ricci curvature. Let $f$ be a positive solution to the heat equation (1.1).\
Let $f=\frac{e^{-v}}{{(4\pi t)}^\frac{n}{2}}$, and we define $$W=\int_{M}(t^2{\left\lvert{\nabla}v\right\rvert}^2-2nt)\frac{e^{-v}}{{(4\pi t)}^\frac{n}{2}}dV_{g},$$ then for all time $t>0$, $$W\leq0$$ and $$\frac{\partial W}{\partial t}\leq0.$$
We will reduce the proofs of above two entropy formulae to the proofs of following two differential Harnack inequalities:
Let $(M,g)$ be a closed Riemannian manifold of dimension $n$ with nonnegative Ricci curvature. Let $f$ be a positive solution to the heat equation (1.1), $u=-\ln f$ and $$H=2\Delta{u}-{{\left\lvert{\nabla}{u}\right\rvert}}^2-\frac{2n}{t}.$$ Then for all time $t>0$, $$H\leq0.$$
and
Let $(M.g)$ be a closed Riemannian manifold of dimension $n$ with nonnegative Ricci curvature. Let $f$ be a positive solution to the heat equation (1.1), $v=-\ln f -\frac{n}{2}\ln(4\pi t)$ and $$P=2\Delta{v}-{{\left\lvert{\nabla}{v}\right\rvert}}^2-\frac{2n}{t}.$$ Then for all time $t>0$, $$P\leq0.$$
Theorem 1.1 (and Proposition 1.3) is the ‘linear version’ of Theorem 4.1 (and Theorem 1.1) in [@CH].
The rest of this note is organized as follows. In section 2, we will give the proof of Proposition 1.3. An integral version of the Harnack inequality (Theorem 2.3) will also be given. In section 3, we will prove Proposition 1.4. We will also discuss about Ni’s entropy. In section 4, the proofs of Theorem 1.1 and 1.2 will be completed. Finally, in section 5, we discuss the backward heat equation.
The author would like to thank his advisor, Prof. Zhiqin Lu for his constant support and many helpful conversations.
Proof of Proposition 1.3.
=========================
In this section we will prove the differential Harnack estimate Proposition 1.3. The strategy is similar as in [@CH] and [@C]; we shall first derive a general evolution formula for function $H$. Let us consider positive solutions of the heat equation (1.1): $$\frac{\partial f}{\partial t}=\Delta f,$$ let $f=e^{-u}$, then $\ln f= -u$. We have $$\frac{\partial}{\partial t}\ln f=-\frac{\partial}{\partial t}u,$$ and $${\nabla}\ln f=-{\nabla}u, \Delta\ln f=-\Delta u.$$ Hence $u$ satisfies the following equation, $$\frac{\partial u}{\partial t}=\Delta u-{\left\lvert{\nabla}u\right\rvert}^2.$$
Let $(M,g)$ be a Riemannian manifold of dimension $n$ and $u$ be a solution of (2.1). Let $$H=\alpha\Delta u-\beta{\left\lvert{\nabla}u\right\rvert}^2-b\frac{u}{t}-c\frac{n}{t}$$ where $\alpha,\beta,b$ and $c$ are constants that we will pick later. Then $H$ satisfies the following evolution equation: $$\begin{aligned}
\frac{\partial}{\partial t}H=\Delta H&-2{\nabla}H\cdot{\nabla}u-2(\alpha-\beta){\left\lvert{\nabla}_{i}{\nabla}_{j}u-\frac{\lambda g_{ij}}{2t}\right\rvert}^2-2(\alpha-\beta)Ric({\nabla}u,{\nabla}u)
\\
&-\frac{2(\alpha-\beta)}{\alpha}\frac{\lambda}{t}H-\big(b+\frac{2(\alpha-\beta)\beta}{\alpha}\lambda\big)\frac{{{\left\lvert{\nabla}u\right\rvert}}^2}{t}+\big(1-\frac{2(\alpha-\beta)}{\alpha}\lambda\big)b\frac{u}{t^2}
\\
&+\big(1-\frac{2(\alpha-\beta)}{\alpha}\lambda\big)c\frac{n}{t^2}+(\alpha-\beta)\frac{n{\lambda}^2}{2{t}^2},\end{aligned}$$ where $\lambda$ is also a constant that we will pick later.
The proof follows from a direct computation. We calculate evolution equation of $H$ term by term. First, since $M$ has static metric, $$\begin{aligned}
\frac{\partial}{\partial t}(\Delta u)&=\Delta(\frac{\partial}{\partial t}u)
\\
&=\Delta(\Delta u)-\Delta({\left\lvert{\nabla}u\right\rvert}^2);\end{aligned}$$
Secondly, $$\begin{aligned}
\frac{\partial}{\partial t}{\left\lvert{\nabla}u\right\rvert}^2=&\frac{\partial}{\partial t}(g^{ij}{\nabla}_{i}u {\nabla}_{j}u)
\\
=&g^{ij}({\nabla}_{i}\frac{\partial u}{\partial t}){\nabla}_{j}u+g^{ij}{\nabla}_{i}u({\nabla}_{j}\frac{\partial u}{\partial t})
\\
=&2{\nabla}(\Delta u)\cdot{\nabla}u-2{\nabla}({\left\lvert{\nabla}u\right\rvert}^2)\cdot{\nabla}u;\end{aligned}$$
Since $$\begin{aligned}
\Delta({\left\lvert{\nabla}u\right\rvert}^2)=&g^{ij}{\nabla}_{i}{\nabla}_{j}(g^{\alpha\beta}{{\nabla}_{\alpha}u}{{\nabla}_{\beta}u})
\\
=&g^{\alpha\beta}\big(g^{ij}{{\nabla}}_{i}{{\nabla}}_{j}({{\nabla}}_{\alpha}u)\big){{\nabla}}_{\beta}u+g^{\alpha\beta}{{\nabla}}_{\alpha}u\big(g^{ij}{{\nabla}}_{i}{{\nabla}}_{j}({{\nabla}}_{\beta}u)\big)
\\
&+g^{ij}g^{\alpha\beta}({{\nabla}}_{i}{{\nabla}}_{\alpha}u)({{\nabla}}_{j}{{\nabla}}_{\beta}u)+
g^{ij}g^{\alpha\beta}({{\nabla}}_{j}{{\nabla}}_{\alpha}u)({{\nabla}}_{i}{{\nabla}}_{\beta}u)
\\
=&2\Delta({\nabla}u)\cdot{\nabla}u+2{\left\lvert{\nabla}{\nabla}u\right\rvert}^2,\end{aligned}$$
and by Ricci identity $$\Delta({\nabla}u)\cdot{\nabla}u={\nabla}(\Delta u)\cdot{\nabla}u+Ric({\nabla}u,{\nabla}u),$$
So we get that $$\begin{aligned}
\frac{\partial}{\partial t}{\left\lvert{\nabla}u\right\rvert}^2=&\Delta({\left\lvert{\nabla}u\right\rvert}^2)-2{\left\lvert{\nabla}{\nabla}u\right\rvert}^2-2{\nabla}({\left\lvert{\nabla}u\right\rvert}^2)\cdot{\nabla}u-2Ric({\nabla}u,{\nabla}u).\end{aligned}$$
Combine with (2.1), we arrive at $$\begin{aligned}
\frac{\partial}{\partial t}H=&\Delta H-\alpha\Delta({{\left\lvert{\nabla}u\right\rvert}}^2)+2\beta{\left\lvert{\nabla}{\nabla}u\right\rvert}^2+2\beta{\nabla}({\left\lvert{\nabla}u\right\rvert}^2)\cdot{\nabla}u
\\
&+2\beta Ric({\nabla}u,{\nabla}u)+b\frac{{\left\lvert{\nabla}u\right\rvert}^2}{t}+b\frac{u}{t^2}+c\frac{n}{t^2}
\\
=&\Delta H-2{\nabla}H\cdot{\nabla}u-2(\alpha-\beta){\left\lvert{\nabla}{\nabla}u\right\rvert}^2
\\
&-2(\alpha-\beta)Ric({\nabla}u,{\nabla}u)-b\frac{{\left\lvert{\nabla}u\right\rvert}^2}{t}+b\frac{u}{t^2}+c\frac{n}{t^2}
\\
=&\Delta H-2{\nabla}H\cdot{\nabla}u-2(\alpha-\beta){\left\lvert{\nabla}_{i}{\nabla}_{j}u-\frac{\lambda}{2t}g_{ij}\right\rvert}^2
\\
&-2(\alpha-\beta)Ric({\nabla}u,{\nabla}u)-2(\alpha-\beta)\frac{\lambda}{t}\Delta u
\\
&+(\alpha-\beta)\frac{n{\lambda}^2}{2t^2}-b\frac{{\left\lvert{\nabla}u\right\rvert}^2}{t}+b\frac{u}{t^2}+c\frac{n}{t^2}
\\
=&\Delta H-2{\nabla}H\cdot{\nabla}u-2(\alpha-\beta){\left\lvert{\nabla}_{i}{\nabla}_{j}u-\frac{\lambda}{2t}g_{ij}\right\rvert}^2
\\
&-2(\alpha-\beta)Ric({\nabla}u,{\nabla}u)-\frac{2(\alpha-\beta)}{\alpha}\frac{\lambda}{t}H
\\
&-\Big(b+\frac{2(\alpha-\beta)}{\alpha}\lambda\beta\Big)\frac{{\left\lvert{\nabla}u\right\rvert}^2}{t}+\Big(1-\frac{2(\alpha-\beta)}{\alpha}\lambda\Big)b\frac{u}{t^2}
\\
&+(\alpha-\beta)\frac{n{\lambda}^2}{2t^2}+\Big(1-\frac{2(\alpha-\beta)}{\alpha}\lambda\Big)c\frac{n}{t^2}. \end{aligned}$$
In the above lemma, let us take $\alpha=2$, $\beta=1$, $b=0$, $c=2$, $\lambda=2$. As a consequence, we have
Let $(M,g)$ be a Riemannian manifold of dimension $n$, $f$ be a positive solution of heat equation (1.1); $u=-\ln f$ and $$H=2\Delta u-{\left\lvert{\nabla}u\right\rvert}^2-\frac{2n}{t},$$ then we have $$\begin{aligned}
\frac{\partial}{\partial t}H=\Delta H-2{\nabla}H\cdot{\nabla}u-2{\left\lvert{\nabla}_{i}{\nabla}_{j}u-\frac{\lambda g_{ij}}{2t}\right\rvert}^2-2Ric({\nabla}u,{\nabla}u)-\frac{2H}{t}-2\frac{{{\left\lvert{\nabla}u\right\rvert}}^2}{t}.\end{aligned}$$
Now we can finish the proof of Proposition 1.3.
Since $M$ is a closed manifold, it is easy to see that for $t$ small enough $H(t)<0$. Then by Corollary 2.2 and the maximum principle of heat equation, $$H\leq0$$ for all time $t$, when $M$ has nonnegative Ricci curvature.
We can now integrate the inequality along a space-time path, and have the following:
Let $(M,g)$ be a Riemannian manifold of dimension $n$ and with nonnegative Ricci curvature. Let $f$ be a positive solution to the heat equation (1.1). Assume that $(x_{1},t_{1})$ and $(x_{2},t_{2})$, with $t_{2}>t_{1}>0 $, are two points in $M\times(0,\infty)$. Let $$\Gamma=\inf_{\gamma}\int^{t_{2}}_{t_{1}}{\left\lvert\dot \gamma\right\rvert}^2dt,$$ where $\gamma$ is any space-time path joining $(x_{1},t_{1})$ and $(x_{2},t_{2})$. Then we have $$f(x_{1}, t_{1})\leq f(x_{2}, t_{2})\Big(\frac{t_{2}}{t_{1}}\Big)^{n}e^{\frac{\Gamma}{2}}.$$
Since $H\leq0$ and $u$ satisfies $$\frac{\partial u}{\partial t}=\Delta u-{\left\lvert{\nabla}u\right\rvert}^2,$$ we have $$2\frac{\partial u}{\partial t}+{\left\lvert{\nabla}u\right\rvert}^2-\frac{2n}{t}\leq 0.$$ If we pick a space-time path $\gamma(x,t)$ joining $(x_{1},t_{1})$ and $(x_{2},t_{2})$ with $t_{2}>t_{1}> 0$, then along $\gamma$, we get $$\begin{aligned}
\frac{du}{dt}&=\frac{\partial u}{dt}+{\nabla}u\cdot\dot \gamma
\\
&\leq -\frac{1}{2}{\left\lvert{\nabla}u\right\rvert}^2+\frac{n}{t}+{\nabla}u\cdot\dot \gamma
\\
&\leq\frac{1}{2}{\left\lvert\dot \gamma\right\rvert}^2+\frac{n}{t},\end{aligned}$$ thus $$u(x_{2},t_{2})-u(x_{1},t_{1})\leq\frac{1}{2}\inf_{\gamma}\int^{t_{2}}_{t_{1}}{\left\lvert\dot \gamma\right\rvert}^2dt+n\ln\Big(\frac{t_{2}}{t_{1}}\Big).$$ By the definition of $u$ and $\Gamma$ we obtain $$f(x_{1}, t_{1})\leq f(x_{2}, t_{2})\Big(\frac{t_{2}}{t_{1}}\Big)^{n}e^{\frac{\Gamma}{2}}.$$
Proof of Proposition 1.4.
=========================
In this section, we consider fundamental solutions of heat equation (1.1). Let $f=(4\pi t)^{-\frac{n}{2}}e^{-v}$, then $v=-\ln f -\frac{n}{2}\ln(4\pi t)$. We have $$\frac{\partial}{\partial t}\ln f=-\frac{\partial}{\partial t}v-\frac{n}{2t},$$ and $${\nabla}\ln f=-{\nabla}v, \Delta\ln f=-\Delta v.$$ Hence $v$ satisfies the following equation: $$\frac{\partial v}{\partial t}=\Delta v-{\left\lvert{\nabla}v\right\rvert}^2-\frac{n}{2t}.$$
Let $(M,g)$ be a Riemannian manifold and $v$ be a solution of (3.1). Let $$P=\alpha\Delta u-\beta{\left\lvert{\nabla}u\right\rvert}^2-b\frac{u}{t}-c\frac{n}{t}$$ where $\alpha,\beta,a,b$ and $d$ are constants that we will pick later. Then $P$ satisfies the following evolution equation: $$\begin{aligned}
\frac{\partial}{\partial t}P=\Delta P&-2{\nabla}P\cdot{\nabla}v-2(\alpha-\beta){\left\lvert{\nabla}_{i}{\nabla}_{j}v-\frac{\lambda g_{ij}}{2t}\right\rvert}^2-2(\alpha-\beta)Ric({\nabla}v,{\nabla}v)
\\
&-\frac{2(\alpha-\beta)}{\alpha}\frac{\lambda}{t}P-\big(b+\frac{2(\alpha-\beta)\beta}{\alpha}\lambda\big)\frac{{{\left\lvert{\nabla}v\right\rvert}}^2}{t}+\big(1-\frac{2(\alpha-\beta)}{\alpha}\lambda\big)b\frac{v}{t^2}
\\
&+\big(1-\frac{2(\alpha-\beta)}{\alpha}\lambda\big)c\frac{n}{t^2}+(\alpha-\beta)\frac{n{\lambda}^2}{2{t}^2}+b\frac{n}{2t^2},\end{aligned}$$ where $\lambda$ is also a constant that we will pick later.
The proof again follows from the same direct computation as in the proof of Lemma 2.1. Notice that the only extra term $b\frac{n}{2t^2}$ comes from the evolution of $v$.
Now again we analyze all the terms in the equation here. To apply the maximum principle of linear heat operator, we must make sure all the terms after $\Delta P$ are either mutiple of $P$, first order variation of $P$, or some terms with definite negative sign. Obviously we can’t control the sign of the term $\big(1-\frac{2(\alpha-\beta)}{\alpha}\lambda\big)b\frac{v}{t^2}$ (because $v$ can have indefinite sign on $M$), so we must kill this term. Then we have two choices:\
\
(1) $1-\frac{2(\alpha-\beta)}{\alpha}\lambda=0$.\
\
In this case, we have the evolution equation becomes $$\begin{aligned}
\frac{\partial}{\partial t}P=\Delta P&-2{\nabla}P\cdot{\nabla}v-2(\alpha-\beta){\left\lvert{\nabla}_{i}{\nabla}_{j}v-\frac{\lambda g_{ij}}{2t}\right\rvert}^2-2(\alpha-\beta)Ric({\nabla}v,{\nabla}v)
\\
&-\frac{P}{t}-(b+\beta)\frac{{{\left\lvert{\nabla}v\right\rvert}}^2}{t}+\frac{{\alpha}^2}{4(\alpha-\beta)}\frac{n}{2{t}^2}+b\frac{n}{2t^2}.\end{aligned}$$\
So obviously we must have $\alpha-\beta\geq 0$, $b+\beta\geq 0$ and $\frac{{\alpha}^2}{4(\alpha-\beta)}+b\leq 0$. Then $\frac{{\alpha}^2}{4(\alpha-\beta)}-\beta\leq 0$, which is just $\frac{(\alpha-2\beta)^2}{4(\alpha-\beta)}\leq 0$. This forces us to choose $\alpha=2\beta=-2b>0$, $\lambda=1$. Now $c$ is only the constant needs to be determined. Since $P$ should be negative at $t=0$, any strictly positive number would work for $c$. Let’s just set $c=-b$. Hence after a rescaling, we obtain the entropy of L. Ni (see [@N], Theorem 1.2): $$P=2\Delta v-{\left\lvert{\nabla}v\right\rvert}^2+\frac{v}{t}-\frac{n}{t}.$$ From the argument above, we can see that Ni’s entropy is the unique entropy of heat equation in this case. In fact, it is not only the counterpart of Perelman’s entropy, but also the counterpart of one of Cao and Hamilton’s entropies (see [@CH],Theorem 1.2), on the static manifold.\
\
(2) $b=0$.\
\
Here the evolution equation reads $$\begin{aligned}
\frac{\partial}{\partial t}P=\Delta P&-2{\nabla}P\cdot{\nabla}v-2(\alpha-\beta){\left\lvert{\nabla}_{i}{\nabla}_{j}v-\frac{\lambda g_{ij}}{2t}\right\rvert}^2-2(\alpha-\beta)Ric({\nabla}v,{\nabla}v)
\\
&-\frac{2(\alpha-\beta)}{\alpha}\frac{\lambda}{t}P-\frac{2(\alpha-\beta)\beta}{\alpha}\lambda\frac{{{\left\lvert{\nabla}v\right\rvert}}^2}{t}+\big(1-\frac{2(\alpha-\beta)}{\alpha}\lambda\big)c\frac{n}{t^2}
\\
&+(\alpha-\beta)\frac{n{\lambda}^2}{2{t}^2}.\end{aligned}$$
In this case, we may have various differential inequalities. If we take $\alpha=2$, $\beta=1$, $c=2$, $\lambda=2$, then
Let $(M,g)$ be a Riemannian manifold of dimension $n$, $f$ be a positive fundamental solution of heat equation; $v=-\ln f-\frac{n}{2}\ln(4\pi t)$ and $$P=2\Delta v-{\left\lvert{\nabla}v\right\rvert}^2-\frac{2n}{t},$$ then we have $$\begin{aligned}
\frac{\partial}{\partial t}P=\Delta P&-2{\nabla}P\cdot{\nabla}v-2{\left\lvert{\nabla}_{i}{\nabla}_{j}v-\frac{\lambda g_{ij}}{2t}\right\rvert}^2-2Ric({\nabla}v,{\nabla}v)-\frac{2P}{t}-2\frac{{{\left\lvert{\nabla}v\right\rvert}}^2}{t}.\end{aligned}$$
Here comes the proof of Proposition 1.3.
Again since $M$ is a closed manifold, then for $t$ small enough $P(t)<0$. By Corollary 3.2 and the maximum principle of heat equation, $$P\leq0$$ for all time $t$, when $M$ has nonnegative Ricci curvature.
On the other hand, we can also take $\alpha=2$, $\beta=0$, $c=1$, $\lambda=1$, then we indeed get the classical Li-Yau differential Harnack inequality (see [@LY]):
Let $M$ be a Riemannian manifold of dimension $n$ and with nonnegative Ricci curvature, $f$ be a positive fundamental solution of heat equation; $v=-\ln f-\frac{n}{2}\ln(4\pi t)$ and $$P=2\Delta v-\frac{n}{t},$$ then we have $$\begin{aligned}
\frac{\partial}{\partial t}P=\Delta P-2{\nabla}P\cdot{\nabla}v-4{\left\lvert{\nabla}_{i}{\nabla}_{j}v-\frac{\lambda g_{ij}}{2t}\right\rvert}^2-4Ric({\nabla}v,{\nabla}v)-\frac{4P}{t}.\end{aligned}$$ So by the maximum principle, $P\leq0$ for all $t>0$.
Proofs of Theorem 1.1 and Theorem 1.2.
======================================
In this section, we complete the proof of monotonicities of the two entropies given in Theorem 1.1 and Theorem 1.2.
First, let $(M,g)$ be the closed Riemannian manifold, and $f$ be a positive solution of heat equation (1.1). Let $u=-\ln f$ and $H$ as in (2.3), we define $$F=\int_{M}t^2He^{-u}dV_{g},$$ it is easy to observe that $$\begin{aligned}
F&=\int_{M}t^2He^{-u}dV_{g}=\int_{M}t^2(2\Delta u-{\left\lvert{\nabla}u\right\rvert}^2-\frac{2n}{t})e^{-u}dV_{g}
\\
&=\int_{M}t^2({\left\lvert{\nabla}u\right\rvert}^2-\frac{2n}{t})e^{-u}dV_{g}+\int_{M}t^2(2\Delta u-2{\left\lvert{\nabla}u\right\rvert}^2)e^{-u}dV_{g}
\\
&=\int_{M}t^2({\left\lvert{\nabla}u\right\rvert}^2-\frac{2n}{t})e^{-u}dV_{g}-2t^2\int_{M}\Delta(e^{-u})dV_{g}
\\
&=\int_{M}(t^2{\left\lvert{\nabla}u\right\rvert}^2-2nt)e^{-u}dV_{g}.\end{aligned}$$ Here we used Stokes theorem for the last equality. So $F$ is indeed the entropy we have in Theorem 1.1.
$F\leq0$ follows directly from $H\leq0$. For time derivative of $F$, using (2.1) and Corollary 2.2 we have $$\begin{aligned}
\frac{d}{dt}F=\int_{M}&t^2He^{-u}dV_{g}
\\
=\int_{M}&\Big(2tHe^{-u}+t^2e^{-u}\frac{\partial}{\partial t}H+t^2H\frac{\partial}{\partial t}e^{-u}\Big)dV_{g}
\\
=\int_{M}&\Big(t^2e^{-u}\big(\Delta H-2{\nabla}H\cdot{\nabla}u-2{\left\lvert{\nabla}_{i}{\nabla}_{j}u-\frac{\lambda g_{ij}}{2t}\right\rvert}^2-2Ric({\nabla}u,{\nabla}u)-2\frac{{{\left\lvert{\nabla}u\right\rvert}}^2}{t}\big)
\\
&+t^2H\Delta e^{-u}\Big)dV_{g}
\\
=\int_{M}&\Big(\Delta(t^2He^{-u})-2t^2e^{-u}\big({\left\lvert{\nabla}_{i}{\nabla}_{j}u-\frac{\lambda g_{ij}}{2t}\right\rvert}^2+Ric({\nabla}u,{\nabla}u)+\frac{{{\left\lvert{\nabla}u\right\rvert}}^2}{t}\big)\Big)dV_{g}
\\
=-2&t^2\int_{M}\Big(e^{-u}\big({\left\lvert{\nabla}_{i}{\nabla}_{j}u-\frac{\lambda g_{ij}}{2t}\right\rvert}^2+Ric({\nabla}u,{\nabla}u)+\frac{{{\left\lvert{\nabla}u\right\rvert}}^2}{t}\big)\Big)dV_{g}\leq0.\end{aligned}$$ Since $M$ has nonnegative Ricci curvature.
Now we prove Theorem 1.2. In this case $v=-\ln f-\frac{n}{2}\ln(4\pi t)$ and $P$ is as in (3.2).\
We set $$\begin{aligned}
W&=\int_{M}t^2P\frac{e^{-v}}{{(4\pi t)}^\frac{n}{2}}dV_{g}=\int_{M}t^2(2\Delta v-{\left\lvert{\nabla}v\right\rvert}^2-\frac{2n}{t})\frac{e^{-v}}{{(4\pi t)}^\frac{n}{2}}dV_{g}
\\
&=\int_{M}t^2({\left\lvert{\nabla}v\right\rvert}^2-\frac{2n}{t})\frac{e^{-v}}{{(4\pi t)}^\frac{n}{2}}dV_{g}+\int_{M}t^2(2\Delta v-2{\left\lvert{\nabla}v\right\rvert}^2)\frac{e^{-v}}{{(4\pi t)}^\frac{n}{2}}dV_{g}
\\
&=\int_{M}t^2({\left\lvert{\nabla}v\right\rvert}^2-\frac{2n}{t})\frac{e^{-v}}{{(4\pi t)}^\frac{n}{2}}dV_{g}-2t^2\int_{M}\Delta\big(\frac{e^{-v}}{{(4\pi t)}^\frac{n}{2}}\big)dV_{g}
\\
&=\int_{M}(t^2{\left\lvert{\nabla}v\right\rvert}^2-2nt)\frac{e^{-v}}{{(4\pi t)}^\frac{n}{2}}dV_{g},\end{aligned}$$ and
Using (3.1) and Corollary 3.2, $$\begin{aligned}
\frac{d}{dt}W=\int_{M}&\Big(2tP(4\pi t)^{-\frac{n}{2}}e^{-v}+t^2(4\pi t)^{-\frac{n}{2}}e^{-v}\frac{\partial}{\partial t}P+t^2P\frac{\partial}{\partial t}\big((4\pi t)^{-\frac{n}{2}}e^{-v}\big)\Big)dV_{g}
\\
=\int_{M}&\Big(\Delta(t^2P(4\pi t)^{-\frac{n}{2}}e^{-v})
\\
&-2t^2(4\pi t)^{-\frac{n}{2}}e^{-v}\big({\left\lvert{\nabla}_{i}{\nabla}_{j}v-\frac{\lambda g_{ij}}{2t}\right\rvert}^2+Ric({\nabla}v,{\nabla}v)+\frac{{{\left\lvert{\nabla}v\right\rvert}}^2}{t}\big)\Big)dV_{g}\leq0.\end{aligned}$$
On the Backward Heat Equation
=============================
In this section, let’s consider the backward heat equation $$\frac{\partial f}{\partial t}=-\Delta f,$$ if we use $\tau$ to denote a variable $\tau(t)$ satisfying $\frac{\partial\tau}{\partial t}=-1$, then we shall have $$\frac{\partial f}{\partial\tau}=\Delta f.$$
Then by the same proof as in section 2, 3, 4, we directly get
Let $(M,g)$ be a closed Riemannian manifold of dimension $n$ and with nonnegative Ricci curvature. Let $f$ be a positive solution to the backward heat equation (5.1); $u=-\ln f$, $\tau=\tau(t)$ with $\frac{\partial\tau}{\partial t}=-1$ and $$H=2\Delta{u}-{{\left\lvert{\nabla}{u}\right\rvert}}^2-\frac{2n}{\tau}.$$ Then for all time $\tau>0$, $$H\leq0.$$
and
Let $(M,g)$ be a closed Riemannian manifold of dimension $n$ and with nonnegative Ricci curvature. Let $f$ be a positive solution to the backward heat equation (5.1); $v=-\ln f -\frac{n}{2}\ln(4\pi \tau)$, $\tau=\tau(t)$ with $\frac{\partial\tau}{\partial t}=-1$ and $$P=2\Delta{v}-{{\left\lvert{\nabla}{v}\right\rvert}}^2-\frac{2n}{\tau}.$$ Then for all time $\tau>0$, $$P\leq0.$$
Let $(M,g)$ be a closed Riemannian manifold of dimension $n$ and with nonnegative Ricci curvature. Let $f$ be a positive solution to the backward heat equation (5.1);\
Let $f=e^{-u}$, $\tau=\tau(t)$ with $\frac{\partial\tau}{\partial t}=-1$ and we define $$F=\int_{M}({\tau}^2{\left\lvert{\nabla}u\right\rvert}^2-2n\tau)e^{-u}dV_{g},$$ then for all time $\tau>0$, $$F\leq0$$ and $$\frac{\partial F}{\partial t}\geq0.$$
and
Let $(M.g)$ be a closed Riemannian manifold of dimension $n$ and with nonnegative Ricci curvature. Let $f$ be a positive solution to the backward heat equation (5.1);\
Let $f=\frac{e^{-v}}{{(4\pi \tau)}^\frac{n}{2}}$, $\tau=\tau(t)$ with $\frac{\partial\tau}{\partial t}=-1$ and we define $$W=\int_{M}({\tau}^2{\left\lvert{\nabla}v\right\rvert}^2-2n\tau)\frac{e^{-v}}{{(4\pi\tau)}^\frac{n}{2}}dV_{g},$$ Then for all time $\tau>0$, $$W\leq0$$ and $$\frac{\partial W}{\partial t}\geq0.$$
Theorem 5.3 (and Proposition 5.1) is the ‘linear version’ of Theorem 4.1, Theorem 4.2 (and Theorem 1.1, Theorem 1.3) in [@C].
[1]{}
Xiaodong Cao. Differential [H]{}arnack estimates for backward heat equations with potentials under the [R]{}icci flow. , 255(4):1024–1038, 2008.
Xiaodong Cao and Richard S. Hamilton. Differential [H]{}arnack estimates for time-dependent heat equations with potentials. , 19(4):989–1000, 2009.
Peter Li and Shing-Tung Yau. On the parabolic kernel of the [S]{}chrödinger operator. , 156(3-4):153–201, 1986.
Lei Ni. The entropy formula for linear heat equation. , 14(1):87–100, 2004.
G. [Perelman]{}. . , November 2002.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Links in a practical network may have different functions, which makes the original network a combination of some functional subnetworks. Here, by a model of coupled oscillators, we investigate how such functional subnetworks are evolved and developed according to the network structure and dynamics. In particular, we study the case of evolutionary clustered networks in which the function of each link (either attractive or repulsive coupling) is updated by the local dynamics. It is found that, during the process of system evolution, the network is gradually stabilized into a particular form in which the attractive (repulsive) subnetwork consists only the intralinks (interlinks). Based on the properties of subnetwork evolution, we also propose a new algorithm for network partition which is distinguished by the convenient operation and fast computing speed.'
author:
- Menghui Li
- Xingang Wang
- 'Choy-Heng Lai'
title: Evolutionary Subnetworks in Complex Systems
---
The past decade has witnessed the blooming of network science, in which one important issue is to explore the interplay between the network structure and dynamics [@CN:REV; @SYN:REV]. While the influences of the network structure on dynamics have been intensively studied in the past [@SYN:NET], recently attentions have also been paid to the influences of the network dynamics on structure, i.e. the evolution of complex networks driven by dynamics [@BBH:2004; @GL:2004; @ADP:2006; @ZK:2006; @BILPR:2007; @OCKK:2008; @LANBHB:2008]. In Ref. [@GL:2004] it has been shown that, rewiring network links according to the node synchronization, a random network can be gradually developed to a small-world network. In Ref. [@ZK:2006] it has been shown that, driven by node synchronization, the weight of the network links can be developed to a particular form in favor of global network synchronization. Besides network evolution, dynamics has been also used for network detection, e.g., detecting the modular structures in clustered complex networks [@ADP:2006; @ZK:2006; @BILPR:2007; @OCKK:2008; @LANBHB:2008; @ZZZHK:2006].
It has been well recognized that links in a practical network are usually different from each other. In previous studies, this has been mainly reflected in the variation of the weight of the network links, i.e., the weighted network [@CN:REV]. Weighted network, however, describes only the case of single-function networks, i.e. all links in the network have the same function, but failing to describe the situation of multi-function networks in which the network links have the diverse functions. A type of commonly seen multi-function networks in practice is the cooperation-competition network (CCN) [@NERVOUS:BOOK; @PDG], in which the network links are divided into two groups of opposite functions. For instance, in the nervous network of the human brain, the synapses are roughly divided into two groups, excitatory and inhibitory, which play the contrary roles to the neuron activities [@NERVOUS:BOOK]. Another typical example of CCN is the relationship network shown in the prisoner’s dilemma game, in which each suspect may either cooperate with (remain silent) or defect from (betray) the other suspects [@PDG].
For multi-function networks like CCN, to facilitate the analysis, it will be convenient if we treat the different groups of links separately. That is, we pick out links serving the same function and, together with their associated nodes, construct a small single-function network. In this way, a multi-function network can be decomposed into a number of functional subnetworks, while each supports a unique function to the system behaviors. Here an interesting question is: How do these functional subnetworks co-evolve with each other and develope into their “adult" forms according to the system properties, e.g., the network structure and dynamics?
To mimic the evolution of the functional subnetworks, we propose the following model of coupled phase oscillators, $$\dot{\theta}_i = \omega_i + \varepsilon \sum_{j=1}^{N}a_{ij} [ \sin
(\theta_j - \theta_i)e_{ij} + \cos(\theta_j - \theta_i)(1-e_{ij}) ].
\label{model}$$ Here, $i,j=1,2,\ldots,N$ are the node indices, $\varepsilon$ is the uniform coupling strength. $\theta_i$ and $\omega_i$ are the instant phase and intrinsic frequency of the $i$th oscillator, respectively. The network structure is represented by the adjacency matrix $A=\{
a_{ij} \}$, in which $a_{ij}=1$ if nodes $i$ and $j$ are directly connected, and $a_{ij}=0$ otherwise. $E(t)=\{ e_{ij}(t) \}$ is a time-dependent binary matrix whose elements are defined as follows. Let $\psi_i(t)$ be the instant phase of the local order parameter defined by the equation [@ROH:2005] $$r_i e^{i\psi_i(t)}=\sum_{j=1}^N a_{ij}e^{i\theta_j(t)}.$$ We set $e_{ij}(t)=1$ if the difference between $\psi_i$ and $\psi_j$ is smaller than a threshold $D$, otherwise we set $e_{ij}(t)=0$.
Different from the traditional models of coupled phase oscillators, in Eq. (\[model\]) the coupling term is made up of two parts of the opposite functions. While the attractive coupling, $H_A=\sin(\theta_j-\theta_i)$, is going to synchronize the connected nodes, the repulsive coupling, $H_R=\cos(\theta_j-\theta_i)$, will work against this tendency. These opposite functions, however, cannot coexist. That is, at any time instant each link can only take on one type of coupling function, either attractive ($e_{ij}=1$) or repulsive ($e_{ij}=0$). The attractive links, together with their associated nodes, constitute the attractive subnetwork, which is represented by the matrix $B=A\circ E$ ($``\circ"$ is the entry-wise product and $I$ is the identity matrix). Similarly, we can construct the repulsive subnetwork, and represent it by the matrix $R=A\circ
(I-E)$. Because $e_{ij}(t)$ is being updated with the system dynamics, the two subnetworks, therefore, will also be changing with time. It should be noted that, despite the evolution of the subnetworks, the global network structure is kept unchanged, i.e., $B(t)+R(t)\equiv A$. Imagine a complex network that is weakly coupled and there is no synchronization between any pair of nodes. It can be expected that, as the system evolves, the two subnetworks will be continuously updated in a random fashion. The question we are interested here is: What happens to the evolution of the subnetworks if the coupling strength is stronger?
We start our investigation by considering the evolution of clustered networks (CN) [@CN:REV]. A typical model of CN is the *ad hoc* network introduced in Ref. [@FOOTBALL], which consists of $4$ clusters, each contains $32$ nodes. In this model, each node on average has $\left< k \right>=16$ links, among which $\left< k_l
\right>$ links are connected to nodes within the same cluster, i.e. the intralinks, and $\left< k_p \right>$ links are connected to nodes from different clusters, i.e, the interlinks. Since we are interested in the case of strongly coupled clustered networks, we use, without loss of generality, in our simulations the parameters $\varepsilon =5$ and $\left< k_p \right>=1$. Meanwhile, to generate the matrix $E(t)$, we use the threshold $D=0.45$. (The influences of these parameters to the evolution will be discussed later.) The natural frequencies and initial conditions of the oscillators are chosen randomly from the ranges $[0,1]$ and $[0, 2\pi)$, respectively. To monitor the evolution, we keep a record of the instant states of the oscillators ($\{\theta_i(t)\}$) and the instant subnetwork matrices ($B(t)$ and $(R(t)$).
![The evolution of functional subnetworks in the *ad hoc* network. (a)-(c) The evolution of the attractive matrix $B(t)$. (d)-(f) The evolution of the repulsive matrix $R(t)$. From time $t\approx50$, all intralinks (interlinks) are contained in the attractive (repulsive) subnetwork, and the structures of the subnetworks will be stabilized. Nodes are rearranged according to the topological clusters.[]{data-label="Fig_1"}](fig1.eps){width="\linewidth"}
The evolution of the structures of the subnetworks can be described as follows. At the beginning, the two subnetworks have similar configuration, i.e. both are an abbreviated version of the original network \[Figs. 1(a) and (d)\]. This is because in a very short time the oscillators have not reached any synchronization, and therefore the matrices $B$ and $R$ are mainly determined by the initial conditions of the oscillators. But, due to the small value of $D$, the repulsive subnetwork has more links than the attractive subnetwork. Then, as time increases, the *interlinks* are gradually excluded from the attractive subnetwork; meanwhile, the *intralinks* are excluded from the repulsive subnetwork \[Figs. 1(b) and (e)\]. The separation of the two subnetworks, however, is not an even process, as some links may jump between the subnetworks repeatedly before settling down. Finally, at the time about $T= 50$, the subnetworks are stabilized into fixed structures and the evolution is complete. In this final stationary state, all intralinks (intralinks) of the network are included in the attractive (repulsive) subnetwork \[Figs. 1(c) and (f)\].
![(Color online) For the same network as in Fig. 1, (a) the time evolution of the oscillator states and (b) a snapshot of the oscillator states at time $t=100$.[]{data-label="Fig_2"}](fig2.eps){width="\linewidth"}
With the evolution of the subnetworks, the system dynamics is also changed. As shown in Fig. 2(a), with the increase of time, the oscillators are gradually organized into $4$ synchronous clusters. The pattern of the synchronous clusters is more evident in Fig. 2(b), where a snapshot of the oscillator states is taken at time $t=100$. After a check of the node indices of the pattern, it is found that the organization of the dynamical clusters obey precisely the topological clusters. Moreover, in forming the synchronous clusters, it is found that the overlapping nodes, i,e. nodes which have interlinks, are more difficult to be synchronized than those internal nodes. This is clearly evidenced in the synchronizing process of the first cluster, where the few overlapping nodes are traveling among the synchronous clusters for an extra period before settling down (the black curves in Fig. 1(a)). Another interesting finding is that, in the stabilized pattern, the states of the synchronous clusters are well separated from each other, which, of course, is attributed to the repulsive coupling on the interlinks.
A prominent feature of our evolutionary model is that the link functions are updated only by the local network information, i.e., the phase of the local order parameter. While it has been commonly believed that the identification of the link attribute, i.e. intralink or interlink, relies on the knowledge of the global network information, e.g., the betweenness centrality of the network links, it is somewhat surprising to see that here the identification can be accomplished by only the local network information. This interesting phenomenon can be explained by a local mean-field theory, as follows. Let $a_{ij}$ be an intralink of a clustered network. Since nodes inside a cluster are densely connected, nodes $i$ and $j$ thus are surrounded by a similar set of neighboring nodes. Because of the large overlap of their neighboring sets, the average phases $\psi_i$ and $\psi_j$ will have small difference, leading to the attractive coupling on the intralink, i.e., $e_{ij}=1$. In contrast, if $a_{ij}$ is an interlink, the nodes $i$ and $j$ will be surrounded by very different neighboring sets, which will generate a larger difference in the average phase, finally leading to the repulsive coupling on the interlink.
We next discuss the influences of the network structure on the evolution. Having understood the critical role of the overlapping neighbors in the evolution, we are able to predict that the above phenomena of subnetwork formation can be observed in any network of clear modular structures. To verify this, we have studied the evolution of the other two typical network models. The first one is the overlapping clustered network studied in Ref. [@LANBHB:2008], in which two larger clusters (each has $96$ nodes) are mediated by two smaller complete clusters (each has $4$ nodes). The smaller clusters have no direct connection, but each is connected to the two larger clusters by an equal number of links. By analyzing their neighbor sets, the network nodes are immediately classified into four groups: two for the larger clusters and two for the smaller clusters. Correspondingly, the oscillators are expected to be synchronized into $4$ dynamical clusters. This is indeed what we have observed in the simulations \[Fig. 3(a) and (b)\]. The second model we have simulated is an ER network [@CN:REV]. Since an ER network has no topological cluster, the neighboring sets of the network nodes thus are different from each other. According to the neighbor-set analysis, this will lead to the repulsive couplings on the links, and generating the turbulent system dynamics. This is indeed what we have found at the beginning of the evolution \[Fig. 3(c) and (d)\]. The repulsive network and turbulent dynamics, however, are unstable. As shown in Fig. 3, after a transient period, the repulsive couplings are quickly switched to the attractive couplings and the turbulent state is changed to the state of global synchronization \[Figs. 3(c) and (d)\]. The switching of the link functions and system dynamics suggest the dual properties of the ER network, i.e., it can be regarded either as containing no module or as containing one unified module.
![(Color online) For the overlapped clustered network, (a) the evolution of the system dynamics and (b) the stabilized link functions. In (b), the intralinks of the larger and smaller clusters are marked by black squares and red dots, respectively. The interlinks are marked by blue triangles. For a ER network of size $N=128$ and average degree $\left< k \right >=16$, (c) the evolution of the system dynamics and (d) the stabilized link functions. The switching of the link functions and system dynamics are started at about $t=30$. In (d), all network links are granted with the attractive coupling function.[]{data-label="Fig_3"}](fig3.eps){width="\linewidth"}
We go on to study the influences of other system parameters on the evolution, including the coupling strength $\varepsilon$, the threshold $D$, and the local dynamics. The numerical results show that, given the network has a clear modular structure, the link attribute can always be successfully detected by the local dynamics, despite the changes of $\varepsilon$ and $D$. Specifically, for the *ad hoc* network of Fig. 1, the system will always develop to the same functional subnetworks \[Fig. 1\] and dynamical pattern \[Fig. 2\] in the parameter space constructed by $\varepsilon \in [0.5,8]$ and $D \in [0.45,\pi/2]$. Furthermore, the main feature of the evolution is independent of the specific form of the local dynamics, as has been verified by other nonlinear oscillators, such as the Logistic map and Lorenz oscillator. However, it should be pointed out that, by changing these parameters, the transient process of the network evolution could be strongly affected, e.g., the transient states of the evolution.
Finally, we discuss the possible application of the evolutionary model in network partition [@FOOTBALL; @GDGGA:2003; @RB:2004; @CNM:2004; @WL:2008]. In network partition, the performance of an algorithm is mainly measured by the following three factors: ease of implementation, accurate detection and fast computation. All these factors are well met in the evolutionary subnetwork model (ESM). Firstly, on the aspect of ease of implementation, the model employs only the local network information, and the partition is accomplished automatically by the system dynamics. In particular, at the end of the evolution, based on the states of the synchronous clusters, the topological clusters can be readily identified. Automatic detection and local network information are the main features of the ESM algorithm, which are also the major difference to the other dynamics-based algorithms [@BILPR:2007; @OCKK:2008; @RB:2004]. Secondly, the ESM algorithm is fast in computing speed. The computational cost of the ESM method is estimated to be $O(NT)$, with $N$ the network size and $T$ the transient time of the evolution. To estimate the computational cost further, it is necessary to characterize the relationship between $T$ and $N$. Numerically, we have checked this relationship by increasing the size of the *ad hoc* network in the following two approaches. In the first approach, the number of the clusters are kept unchanged, but the size of the $4$th cluster is gradually increased from $32$ to $864$. In the second approach, the size of each cluster is kept unchanged, but the number of the clusters is increased from $4$ to $30$. The variations of $T$ as a function of $N$ are plotted in Fig. 4(a), together with the synchronizing time, $T'$ , of the largest cluster in the network. Very interestingly, it is found that for both approaches we have $T\propto T'$. That is, the transient time of the whole network is proportional to that of the largest cluster. Particularly, for the first approach we even have $T\sim T'$. Previous studies have indicated that $T'$ is mainly determined by the intrinsic properties of the cluster, e.g. the cluster size, instead of the global network properties [@WHLL:2007]. This implies that $T'\sim M_{max}$, with $M_{max}$ the size of the largest cluster in the network. Therefore, the computational cost of ESM is estimated to be proportional to $O(NM_{max})$. Since for practical networks we generally have $M_{max} \ll N$, the computational cost of ESM thus is estimated to increase linearly with the network size. Finally, on the aspect of detecting accuracy, the ESM algorithm works very well for clustered networks and reasonably well for fuzzy networks \[Fig. 4(b)\].
![(Color online) For the *ad hoc* network of Fig. 1, (a) the transient time of the whole network, $T$, and the synchronizing time of the largest cluster in the nework, $T'$, as a function of the network size, and (b) the variation of the mutual information (see Ref. [@DGDA:2005] for details about mutual information) as a function of $\left<k_p\right>$ generated by the partition algorithms “Potts" [@RB:2004], “ESM", “GKM" [@OCKK:2008], “OCR" [@BILPR:2007], and “CNM" [@CNM:2004]. The coappearance matrix generated by ESM for (c) the Zachary’s karate club and (d) the football club. The overlapping nodes are marked by the squares. In (c) and (d), each data is averaged over $100$ realizations. []{data-label="Fig_4"}](fig4.eps){width="\linewidth"}
As applications of the new algorithm, we have tested the partitions of two empirical clustered networks. The first one is the Zachary’s karate club network [@KARATE], which contains $34$ nodes and $78$ links. The numerical result is plotted in Fig. 4(c), in which the network is clearly divided into $4$ clusters. Moreover, the overlapping node, i.e. the $10$th node in the network, is also well characterized. These results coincide with that of Ref. [@KARATE]. The second example we have tested is the football network [@FOOTBALL], which contains $115$ nodes (teams) and $613$ (matches) links. The numerical result is plotted in Fig. 4(d), in which the network is clearly divided into a number of conferences (clusters). Again, the independent teams, which have equal number of games (links) with multiple conferences, are well identified by the overlapping nodes. These results coincide with that of Ref. [@FOOTBALL].
In summary, we have studied the evolution of functional subnetworks in clustered networks and proposed a new algorithm for partitioning networks. Hopefully, the finding that the network function is jointly determined by the network structure and system dynamics could be helpful to the study of complex behaviors in functional networks.
[99]{}
R. Albert and A.-L. Barabási, Rev. Mod. Phys. **74**, 47 (2002); M.E.J. Newman, SIAM Rev. **45**, 167 (2003).
S. Boccaletti, *et.al.*, Phys. Rep. **424**, 175 (2006); A. Arenas, *et.al.*, Phys. Rep. **469**, 93 (2008).
X. F. Wang and G. Chen, Int. J. Bifurcation Chaos Appl. Sci. Eng. **12**, 187 (2002); M. Barahona and L. M. Pecora, Phys. Rev. Lett. **89**, 054101 (2002); T. Nishikawa, *et.al.*, Phys. Rev. Lett. **91**, 014101 (2003); X. G. Wang, *et.al.*, Phys. Rev. E **75**, 056205 (2007); A. Arenas, Phys. Rev. Lett. **98**, 034101 (2007).
I.V. Belykh, *et.al.*, Physica D **195**, 188 (2004); W. Li, *et.al.*, Phys. Rev. E **76**, 045102(R) (2007). P. Gong, C.V. Leeuwen, Europhys. Lett. **67**, 328 (2004).
A. Arenas, *et.al.*, Phys. Rev. Lett. **96**, 114102, (2006)
C. Zhou and J. Kurths, Phys. Rev. Lett. **96**, 164102 (2006).
S. Boccaletti, *et.al.*, Phys. Rev. E **75**, 045102(R) (2007).
E. Oh, *et.al.*, Europhys. Lett. **83**, 68003 (2008).
D. Li, *et.al.*, Phys. Rev. Lett. **101**, 168701 (2008).
C.S. Zhou, *et.al.*, Phys. Rev. Lett. **97**, 238103 (2006).
D. Purves and J. Lichtman, *Principles of Neural Development* (Sinauer Associates, Sunderland, MA, 1985).
J. Hofbauer and K. Sigmund, *Evolutionary Games and Population Dynamics* (Cambridge University Press, Cambridge, 1998).
J. G. Restrepo, *et.al.*, Phys. Rev. E **71**, 036151 (2005); X.G. Wang, *et.al.*, Chaos **18**, 037117 (2008).
M. Girvan and M.E.J. Newman, Proc. Natl. Acad. Sci. USA **99**, 7821 (2002).
R. Guimerá, *et.al.*, Phys. Rev. E **68**, 065103(R), (2003); J. Duch and A. Arenas, Phys. Rev. E **72**, 027104, (2005);
J. Reichardt and S. Bornholdt, Phys. Rev. Lett. **93**, 218701 (2004).
A. Clauset, *et.al.*, Phys. Rev. E **70**, 066111 (2004).
J. Wang, C.-H. Lai, New J. Phys. **10**, 123023 (2008).
L. Danon, *et.al.*, J. Stat. Mech. P09008 (2005).
X.G. Wang, *et.al.*, Phys. Rev. E **76**, 056113 (2007).
W.W. Zachary, J. Anthropol. Res. **33**,452 (1977).
|
{
"pile_set_name": "ArXiv"
}
|
---
bibliography:
- './lensingbib\_final.bib'
title: 'Constraining the neutrino emission of gravitationally lensed Flat-Spectrum Radio Quasars with ANTARES data'
---
Introduction
============
Active Galactic Nuclei (AGNs) are longstanding candidate sources for the highest-energy cosmic rays, the origin of which is still unknown. The blazar subclass is comprised of BLLac objects and the more luminous Flat-Spectrum Radio Quasars (FSRQs), the relativistic jets of which are pointed at a small angle to the line of sight [@1995PASP..107..803U]. The matter content of AGN jets is still heavily debated, as both leptonic and hadronic models can in principle explain the broad band spectral energy distribution of blazars [@boettcher07]. As of now, for electromagnetic radiation there is no simple observable that allows us to distinguish between both types of models. The detection of neutrinos from such jets might therefore help solve this long-standing issue.
In hadronic models for AGN jets, neutrinos are predicted to be emitted along with gamma-rays by processes involving the interaction of accelerated hadrons with the surrounding medium and radiation fields, and the subsequent production and decay of pions and kaons (see e.g. [@gaisser95; @learned00] and references therein). The neutrino spectrum should therefore be closely related to the hadron spectrum, usually assumed to be a power-law with spectral index $\gamma \simeq -2$ as suggested by the theory of diffuse shock acceleration [@Bell:1978zc; @Bell:1978fj]. The IceCube neutrino telescope has recently provided evidence for a cosmic component in the diffuse high-energy neutrino flux [@Aartsen:2013jdh], part of which could originate from a population of unresolved extragalactic sources, possibly blazars [@Krauss:2014tna]. Yet, no individual neutrino source has so far been identified, and, given the low expected neutrino fluxes, the published upper limits do not allow different jet models to be distinguished. BLLacs are interesting targets to derive generic constraints on the jet emission mechanisms as they can be found at relatively small redshifts. For blazar populations which are typically distributed at much larger distances, such as the FSRQs, we argue that gravitational lensing can help enhance the sensitivity to neutrino emission.
Gravitational lensing of electromagnetic radiation from distant astrophysical sources is a well-known prediction of Einstein’s theory of general relativity [@Einstein; @Zwicky]. Since the first detection of multiple images of a gravitationally lensed quasar, QSO 0957+561 [@1979Nature; @1979ApJ], hundreds of lens systems have been discovered and studied, opening new perspectives both in astrophysics and cosmology (see [@Bartelmann:2010fz] for a recent review). Due to their very low masses ($m_\nu$ do not exceed $\sim$1 eV [@Beringer:1900zz]), neutrinos are expected to undergo the same phenomenon of gravitational lensing as photons. Possible configurations for neutrino lensing by massive astrophysical objects have been theoretically studied in the literature, e.g.[@Gerver; @Elewyck2; @Mena; @Eiroa]. For distant neutrino sources gravitationally lensed by an interposed galaxy (or galaxy cluster), the magnification of the neutrino flux is expected to be equal to that of the photon flux.
Here, we make use of this similarity between the lensing of photons and of neutrinos to significantly lower the upper limits on neutrino emission from FSRQs, by using the luminosity boost provided by gravitational lenses that magnify some distant radio-loud AGN. We illustrate this method by performing a search for an excess of neutrino events in the direction of distant lensed FSRQs in the field of view of the ANTARES neutrino telescope. The remainder of this paper is organized as follows. Section \[sec:antares\] describes the ANTARES neutrino telescope, which is used to detect neutrino events from gravitationally lensed sources, followed in Section \[sec:addobjects\] by a description of the candidate sample. We present and discuss our results in Section \[sec:results\]. Specifically, we show that this strategy leads to an improvement in the upper limit on the intrinsic neutrino luminosity of FSRQs.
The ANTARES Neutrino Telescope and Data Sample {#sec:antares}
==============================================
The ANTARES neutrino telescope [@Ant2011] is located in the Mediterranean Sea, about 40km offshore from Toulon, France ($42^\circ 48'$N, $6^\circ 10'$E). The detector consists of 885 photomultiplier tubes (PMTs) mounted on twelve vertical lines anchored to the seabed at a depth of 2475m, with a typical inter-line separation of $\sim$65m. The operation principle is based on the detection of Cherenkov light induced by relativistic muons produced in the charged-current interactions of high-energy neutrinos with matter surrounding the detector. The knowledge of the time and amplitude of the light pulses recorded by the PMTs allows the reconstruction of the trajectory of the muon and to infer the arrival direction of the incident neutrino. To limit the background from down-going atmospheric muons, the design of ANTARES is optimised for the detection of up-going muons produced by neutrinos which have traversed the Earth. Its instantaneous field of view is $\sim 2\pi\,\mathrm{sr}$ for neutrino energies $E_\nu \gtrsim
100\,\mathrm{GeV}$. Further detail on the above can be found in refs. [@Ant2011; @Ant2002; @DAQ].
The ANTARES collaboration has developed several strategies to search for cosmic neutrino candidates, also in association with other cosmic messengers such as cosmic-, X-, and gamma-rays, and gravitational waves (e.g. [@Aguilar:2010ab; @Ageron:2011pe; @2012APh....36..204A; @AdrianMartinez:2012tf; @Adrian-Martinez:2013sga]). A search for neutrino point-like sources was conducted with the data sample corresponding to the first four years of operation of the detector, 2007–2010 [@AdrianMartinez:2012rp]. This search has recently been extended to the years 2011–2012, for a total detector livetime of 1338 days [@Adrian-Martinez:2014wzf]. The selection criteria have been optimised to search for $E^{-2}$ neutrino fluxes from point-like astrophysical sources. Upgoing events have been selected to reject the bulk of background atmospheric muons. Additional cuts on the quality of the muon track reconstruction have also been applied to eliminate events that correspond to downgoing atmospheric muons which are misreconstructed as upgoing. Most of the remaining events are atmospheric muon neutrinos which constitute the primary background for cosmic neutrino searches. The final sample contains 5516 neutrino candidates with a predicted atmospheric muon neutrino purity of around 90% and an estimated median angular resolution of $0.38^\circ$ [@Adrian-Martinez:2014wzf].
The above-mentioned sample was searched for an excess neutrino flux in the direction of 51 candidate neutrino sources (including the Galactic Centre). This list of sources was established mainly on the basis of the intensity of their gamma-ray emission as observed by *Fermi* [@Atwood:2009ez] and H.E.S.S. [@Bernloehr:2003vd] and includes five non-lensed FSRQs. No statistically significant excess has been found in the direction of any of the candidate sources. The corresponding 90% confidence level (C.L.) upper limits on the neutrino flux, assuming an $E^{-2}$ spectrum, have been derived; they are the most restrictive to date in a significant fraction of the Southern Sky. In the next section, we show how these limits can be improved upon by using lensed sources.
Search for Neutrino Emission from Distant, Lensed Blazars {#sec:addobjects}
=========================================================
Determination of the Magnification Factor {#sec:mu}
-----------------------------------------
The strategy presented here relies on an estimation of the magnification factor for each lensed system, as obtained from photon observations. To estimate the magnification of a given lensed image, a model is required for the mass distribution of the lens. Here, we utilise the simplest model able to account for the image morphology and magnification ratios in each lens system: the singular isothermal ellipsoid (SIE). SIE models are also a reasonable description of the mass distributions of elliptical galaxies (e.g., [@Sonnenfeld:2013]). We adjust the parameters of SIEs with the “gravlens” modeling software [@Keeton] to fit image positions and flux ratios. We use image positions from infrared H-band HST observations from the CASTLES project[^1], which have uncertainties of $0.003$ arcsec. We assume the centre of mass of each lens galaxy to be its centre of brightness. The centre for each SIE model is then the centre of brightness of each lens galaxy. The positions of the SIEs have uncertainties between $0.01$ and $0.05$ arcsec (CASTLES). Because the flux ratios may be influenced by microlensing, the SIEs cannot reproduce them to better than a 10% uncertainty, which we adopt for our flux uncertainties. The resulting uncertainties for the estimates of magnifications for the lensed images in our sample are about 15% (10% for quadruple-image systems), excluding systematic uncertainties.
Candidate Distant Lensed Blazars {#sec:cand}
--------------------------------
The most interesting lensed blazar candidates are B0218$+$357 [@1972AJ.....77..797P] and PKS 1830$-$211 [@1988MNRAS]: they are visible to ANTARES and have also been detected by *Fermi*. B0218$+$357 is a double-image lens system with an extended jet at redshift $z=0.96$, while the intervening lensing spiral galaxy has $z=0.68$ [@1992AJ....104.1320O]. We obtain magnification values of 12.3 and 8.5 for the two images based on the SIE model. These estimates are consistent with the observed flux ratios at radio wavelengths [@Biggs:1998ez]. B0218$+$357 is also the first lensed system for which a clear $\gamma$-ray measurement of a time delay for two images has been reported [@Cheung:2014dma]. The measurement utilised flares detected with the *Fermi* Large Area Telescope (LAT). Although the LAT is unable to resolve the two images, the flares were of sufficient amplitude (peaking at $\sim$ 20$-$50 times the quiescent flux) to conduct an autocorrelation analysis. The result was a delay of $11.46\pm 0.16$ days, generally consistent with previous measurements at radio wavelengths. Some of the uneven structures of the flares implied that microlensing may be occurring in this system. To establish the effect of microlensing on the magnifications of the images that we use in this paper, models of the distribution of stars in the lens galaxy and their kinematics are necessary. Further analysis is underway to model the behavior of the $\gamma$-ray flares (Falco et al., in preparation). Flares in $\gamma$-rays from *Fermi*-LAT were also reported for PKS 1830$-$211, a double-image lens system with a separation of . This radio-loud blazar is at $z=2.51$ and is gravitationally lensed by a spiral galaxy at $z=0.89$. Based on the SIE model, the magnification values that we obtain for the two images are 3.7 and 1.5. A time delay of $27.1 \pm 0.6$ days between the images was estimated at radio wavelengths [@Lovell:1998ka]. Evidence for a time delay in the *Fermi* LAT $\gamma$-ray data, consistent with the radio measurement, has also been reported for this source . Observations with ALMA that overlapped with some of those of *Fermi* suggested microlensing at sub-mm wavelengths [@Marti-Vidal:2013vva]. Their measurements also revealed chromatic variations in the flux densities of the images at these wavelengths, which may arise from microlensing or from intrinsic variability of the blazar jet. Our estimates for the magnifications of the images of PKS 1830$-$211 yield a magnification ratio of $\sim 2.4$, compared with the estimated magnification ratio of $\sim 1.5$ at radio wavelengths [@Lovell:1998ka]. Because this source is highly variable (both extrinsically and intrinsically), the estimates of magnification ratios are affected and also variable. We adopt our model magnifications as fiducial values.
We also include in our study B1422$+$231 [@Raychaudhury:2003cf] and B1030$+$074 [@Xanthopoulos:1998zm] which are two lensed blazars in the ANTARES field of view, although with no associated *Fermi* detection so far. B1422$+$231 is a four-image quasar at $z=3.62$ lensed by a galaxy at $z=0.34$. Its optical spectrum shows broad emission lines, and it is categorized as an FSRQ in the Multifrequency Catalogue of Blazars [@Massaro:2010si]. The magnification values of the four images as obtained from the SIE model are 16.0, 15.0, 11.1 and 0.9. B1030$+$074 is a two-image system with a variable source exhibiting jet structure at $z=1.54$, and a lens galaxy at $z=0.60$. It is a blazar of uncertain type in [@Massaro:2010si]. We obtain magnification values of 2.4 and 0.27 for the two images.
[lrrrrrrr]{}\
\
Name & $z$ & $d_L$ & $\phi_{\nu}^{90CL}$ & $\mu$ (n) & $L^{90}$\
\
\
3C 279 & 0.54 & 3.12 & 3.45 & 1 & 6.44\
PKS 1454$-$354 &1.40 &10.2 & 1.70 & 1 & 33.8\
3C 454.3 &0.86 & 5.54 & 2.39 & 1 &14.1\
PKS 1502$+$106 &1.84 &14.3 & 2.31 & 1 &90.5\
PKS 0727$-$11 &1.60 & 12.0 & 2.01 & 1 &55.3\
PKS 1830$-$211 & 2.51 & 21.0 & 1.89 & 5.20 (2) &30.8\
B0218$+$357 & 0.96 & 6.35 & 2.91 & 20.8 (2) & 1.08\
B1422$+$231 & 3.62 & 32.7 & 2.46 & 43.0 (4)&11.7\
B1030$+$074 &1.54 & 11.5 & 2.26 & 2.67 (2)& 21.5\
QSO 0957$+$561 &1.41 & 10.3 & - & 2.80 (2) & -\
\
\
Search for Neutrino Emission {#sec:search}
----------------------------
We search for neutrino emission from the above sources following the same procedure as described in [@AdrianMartinez:2012rp; @Adrian-Martinez:2014wzf]. Neutrino events within 20$^\circ$ of the source are used to construct an unbinned likelihood function including both signal and atmospheric background components. For each source, this likelihood is then maximised with respect to the number of signal events. The sensitivity of the analysis is evaluated through the generation of large numbers of pseudo-experiments simulating background and signal. No significant excess of neutrinos above the expected background has been found for the four candidate sources. The corresponding 90% C.L. upper limits on the neutrino flux have been derived using the Feldman & Cousins approach [@Feldman:1997qc]; they are given in Table \[tab:fsrqs\] along with the limits on the five non-lensed FSRQs obtained in the previous ANTARES analysis.
![Limit on the instrinsic neutrino luminosity $L^{90}$ of blazars viewed by ANTARES as a function of their luminosity distance. The black diamonds (resp. blue triangles) correspond to non-lensed FSRQs (resp. BLLacs) in the list of candidate sources of [@Adrian-Martinez:2014wzf]. The red squares correspond to the two lensed FSRQs also seen by *Fermi* (B0218$+$357 and PKS 1830$-$211) and the green circles correspond to the two other lensed blazars considered in this paper (B1422$+$231 and B1030$+$074). The error bars on the lensed sources represent the uncertainty in the determination of the magnification factor.[]{data-label="fig"}](luminosity_vs_distance_final-eps-converted-to.pdf){width="\hsize"}
Results and Discussion {#sec:results}
======================
From the upper limits on the neutrino flux at Earth $\phi_{\nu}^{90CL}$ (given in Table \[tab:fsrqs\]), we derive limits on the intrinsic neutrino luminosities of the sources. The total isotropic emitted power in high-energy neutrinos ($\gtrsim 1$TeV) for a blazar at luminosity distance $d_L$ is $$L^{90} = \frac{1}{\mu}\, 4\pi d_L^2\, \phi_{\nu}^{90CL},$$ where $\mu$ is the magnification factor ($\mu = 1$ for non-lensed sources). The corresponding values are given in Table \[tab:fsrqs\] for both lensed and non-lensed FSRQs considered in this study. Note that the value $L^{90}$ itself is not a source-intrinsic quantity. It depends on the viewing angle via the Doppler factor. In lack of well-constrained viewing angles for most blazars, however, conclusions about the intrinsic luminosities can be derived by applying statistical methods [@Cohen2007].
The results are summarised in Figure \[fig\] where the intrinsic luminosity limit is plotted against the luminosity distance of the source. For the lensed systems, the error bar account for the uncertainty in the determination of the magnification factor as discussed in Sect. \[sec:mu\]. We have added for comparison the luminosity limits obtained for the BLLacs in the ANTARES list of sources, most of which are located at a much closer distance [@Adrian-Martinez:2014wzf]. The isotropic power limits obtained for the non-lensed FSRQs are between $6 \times 10^{46}$ and $9 \times 10^{47}\,\mathrm{erg}\,\mathrm{s}^{-1}$. These powers can be compared with the bolometric luminosities of AGNs, which are typically in the range $10^{44}$–$10^{47}\,\mathrm{erg}\,\mathrm{s}^{-1}$ [@Woo:2002un; @MNR:MNR21513; @boettcher07], but can rise up to $10^{49}\,\mathrm{erg}\,\mathrm{s}^{-1}$ for some hadronic jet models such as in the synchrotron proton blazar interpretation of 3C 279 [@Boettcher:2008tq].
One directly sees from Figure \[fig\] that the limits derived from lensed FSRQs are stronger than those corresponding to non-lensed sources of the same class at comparable distances, and that lensing can be efficiently used to improve the constraints on neutrino emission from FSRQs. The strongest limit is obtained for the *Fermi*-detected B0218$+$357; it corresponds to a total neutrino power of $1.1\times 10^{46}\,\mathrm{erg}\,\mathrm{s}^{-1}$, about one order of magnitude lower than the lowest limit achieved with non-lensed FSRQs. This limit is expected to improve in the future, in particular when the multi-km$^3$ neutrino telescope KM3NeT [@KM3NeTTDR], with an instrumented volume about 100 times bigger than that of ANTARES, becomes operational in the Mediterranean.
A similar study could in fact be performed with the lensed quasar QSO 0957$+$561, a doubly-imaged, wide-separation system with the source at $z = 1.41$ and the lensing galaxy at $z=0.36$ [@1979Nature; @1979ApJ]. This quasar is not in the ANTARES field of view, but its neutrino emission could be constrained by IceCube; the information for this source is therefore also included in Table \[tab:fsrqs\]. Assuming a typical ANTARES sensitivity for a similar source in the ANTARES field of view, as given in [@Adrian-Martinez:2014wzf], the 90% C.L. upper limit on the neutrino luminosity for QSO 0957$+$561 would be $L^{90}\simeq 1.8\times 10^{47}\,\mathrm{erg}\,\mathrm{s}^{-1}$. Based on the sensitivities presented in [@Abbasi:2010rd], the limit set by IceCube would approximately be 10 times stronger, and thus comparable to the limit obtained by ANTARES on B0218$+$357.
In conclusion, we suggest that neutrino telescopes include the lensed FSRQs discussed above in their future searches for steady point-source neutrino emission.
The authors would like to thank D. Allard, O. Mena, J. A. Muñoz and G. E. Romero for enlightening discussions during the preparation of this manuscript.
They also acknowledge the financial support of the funding agencies: Centre National de la Recherche Scientifique (CNRS), Commissariat à l’énergie atomique et aux énergies alternatives (CEA), Commission Européenne (FEDER fund and Marie Curie Program), Région Alsace (contrat CPER), Région Provence-Alpes-Côte d’Azur, Département du Var and Ville de La Seyne-sur-Mer, France; Bundesministerium für Bildung und Forschung (BMBF), Germany; Istituto Nazionale di Fisica Nucleare (INFN), Italy; Stichting voor Fundamenteel Onderzoek der Materie (FOM), Nederlandse organisatie voor Wetenschappelijk Onderzoek (NWO), the Netherlands; Council of the President of the Russian Federation for young scientists and leading scientific schools supporting grants, Russia; National Authority for Scientific Research (UEFISCDI), Romania; Servicio Público de Empleo Estatal (SEPE), Ministerio de Ciencia e Innovación (MICINN), Prometeo of Generalitat Valenciana and MultiDark, Spain; Agence de l’Oriental and CNRST, Morocco. We also acknowledge the technical support of Ifremer, AIM and Foselev Marine for the sea operation and the CC-IN2P3 for the computing facilities.
[^1]: www.cfa.harvard.edu/castles/
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- 'Giorgio Busoni,'
- 'Andrea De Simone,'
- 'Thomas Jacques,'
- 'Enrico Morgante,'
- Antonio Riotto
bibliography:
- 'relicLHC.bib'
title: Making the Most of the Relic Density for Dark Matter Searches at the LHC 14 TeV Run
---
SISSA 54/2014/FISI
Introduction
============
After the recent discovery of a Standard Model (SM)-like Higgs boson [@Aad:2012tfa; @Chatrchyan:2012ufa], the focus of Run II at the Large Hadron Collider (LHC), with $\sqrt{s}=(13-14)$ TeV and a luminosity of about $10^{34}$ cm$^{-2}$ s$^{-1}$, is to search for new physics at the TeV scale. In particular, one of the most commonly publicized goals is to find evidence of a new fundamental particle which might compose the so-called Dark Matter (DM), a non-relativistic degree of freedom contributing to about 30% of the energy in our universe.
If DM consists of particles whose mass and interactions with SM particles are dictated by physics in the electroweak energy range, there are two bonuses. First, the DM particles might be produced at the LHC and subsequently escape the detectors. This is the main reason why a great deal of effort by the ATLAS and CMS collaborations is dedicated to searching for events where large missing energy is the dominant discriminating signature of DM [@monojetCMS2; @monojetATLAS2; @Abdallah:2014dma]. Secondly, the DM abundance in our universe is likely to be fixed by the thermal freeze-out phenomenon: DM particles, initially present in our universe in thermal equilibrium abundance, annihilate with one another till chemical equilibrium is lost due to the expansion of the universe [@Srednicki:1988ce; @Scherrer:1985zt; @Gondolo:1990dk; @Bertone:2004pz]. The present-day relic density of these particles is predictable and, in the simple case of $s$-wave self-annihilation of DM in the early universe, it comes out to be (in units of the critical energy density of the universe) \_[DM]{}h\^2, \[simplerelic\] where $\langle\sigma v\rangle_{\rm ann}$ is the total thermally-averaged annihilation cross section, and the factor of 2 in the numerator is made explicit to emphasize the fact that we are assuming a non-self-conjugate DM particle. This abundance must match the one recently measured by the Planck collaboration, $\Omega_{\rm DM}^{\rm obs}h^2=0.1199\pm0.0027$ [@Ade:2013zuv].
Now, a fundamental question which one should ask is the following: under the optimistic hypothesis that the next LHC run gives evidence for a new weakly interacting particle with a lifetime that exceeds about a microsecond, how confident can we be in claiming we have finally revealed the true nature of the DM?
To answer this question, one needs to operate within a given framework and identify the parameter space which is compatible with a positive LHC signal and possibly with the current (and future) bounds (or signals) from direct and indirect searches. Last, but not least, it is imperative to check if the properties of the new particle are compatible with the observed DM abundance.
The goal of this paper is to investigate whether a new stable particle, within a given set of models, may be assigned the label of thermal relic DM by comparing the regions of the parameter space where the right abundance is attained with the exclusion regions for the forthcoming Run II at the LHC. The latter are a useful benchmark for evaluating the sensitivity of the analysis at 14 TeV. However, if DM is within the reach of the LHC, it is also useful to make the comparison with the 5$\sigma$ discovery potential regions. Of course, one can also reverse the logic of this exercise and identify the regions of the parameter space of a given model where the DM abundance fits the observed one. This might be useful to set priorities for the LHC collaborations when comparing the future data with the plethora of models.
This is not to say that this analysis can exclude the possibility that a new stable particle can be DM. Rather, if the new particle is inconsistent with thermal-relic DM under our assumptions and in a particular model, then we learn that either: [**1)**]{} the model is not the correct model of DM, or [**2)**]{} one of the assumptions enumerated in Section \[general\] do not hold.
This paper is structured as follows. In Section \[general\] we provide some general considerations and state our assumptions, along with a description of the model we consider. In Section \[EFTresults\] we compare ATLAS 14TeV sensitivity with the region of parameter space consistent with thermal relic DM. In Section \[SIMPresults\] we extend this analysis to simplified models. Finally, we collect our concluding remarks in Section \[conclusion\].
Working assumptions\[general\]
==============================
The goal of this Section is to provide some general considerations about the DM abundance and its link with collider searches and, above all, to list as clearly as possible the set of assumptions we are working with.
DM Abundance Considerations {#subsec:DM-Ab-Cons}
---------------------------
Consider the general scenario where a DM candidate $\chi$ will eventually be efficiently pair-produced at the LHC. This implies that $\chi$ must interact with first-generation quarks, therefore one can define the thermally averaged DM annihilation cross section [^1] v\_\* \_[ |u |u ]{} + \_[|d |d]{}, which also sets a reference for DM production at the LHC. In the early universe, besides annihilations into quarks, there can be additional annihilation channels, so that the total DM annihilation cross section which is relevant for the relic abundance is \[f\] v\_[ann]{}v\_[\*]{}. So, by requiring that the particles $\chi$ and $\bar\chi$ compose the DM abundance, we find \_[DM]{}\^[obs]{} h\^2 , or \[first\] v\_[\*]{}4.010\^[-9]{}[GeV]{}\^[-2]{}. On the other hand, one can make the reasonable assumption that the dominant DM annihilation channel is to SM fermions and the coupling to the first generation of quarks is not less than the coupling to other SM fermions. This hypothesis follows from the requirement that the would-be DM particles are efficiently produced in the next Run II. We are the first to admit that this assumption is debatable, but we consider it as a working hypothesis. We will show later how weakening this assumption affects our results. In this case, we get \[s\] v\_[ann]{}\_[quarkgen.]{} v\_[\*]{} + \_[leptongen.]{} v\_[\*]{} 4v\_[\*]{}, and therefore
\_[DM]{}\^[obs]{} h\^2 , or
\[second\] v\_[\*]{}1.010\^[-9]{}[GeV]{}\^[-2]{}. Let us illustrate the relevance of these inequalities with a simple example. Assume that the interactions between DM and SM quarks are described within an Effective Field Theory (EFT), where the basic parameters are the DM mass $m_{\rm DM}$ and the UV scale $\Lambda$. Let us also imagine that the annihilation controlling the thermal abundance takes place in the $s$-wave. One therefore expects roughly that $\langle\sigma v\rangle_{*}\simeq 10^{-1}m_{\rm DM}^2/\Lambda^4$. We then obtain, from Eqs. (\[first\]) and (\[second\]), \[a\] 0.7 ()\^[1/2]{} [TeV]{}1.0 ()\^[1/2]{} [TeV]{}. This value of the UV scale needs to be compatible with the one needed to explain the positive DM signature at the LHC. For instance, if $\Lambda$ turns out to be larger than the value of the lower bound, one concludes that the would-be DM particle has to annihilate in other channels which we do not have control of and therefore it would be difficult, if not impossible, to assign it the label “dark matter".
Curves corresponding to the correct relic abundance have been used as a benchmark or comparison for EFT constraints since the early usage of EFTs [@Goodman:2010ku; @Goodman:2010yf]. However, these relic density constraints on thermal DM are usually considered not to be robust: for a given set of parameters, the relic density can be smaller if the cross section is enhanced by inclusion of other annihilation channels, such as annihilation to leptons; conversely, the true relic density can be larger if there is a larger dark sector including other types of DM. However, under a modest set of assumptions, these constraints can become substantially more powerful. Throughout this analysis, we will assume:
1. the DM candidate $\chi$ makes up 100% of the DM of the universe;
2. the DM annihilation rate is related to the observed density today via the standard thermal production mechanism;
3. the dominant annihilation channel is to SM fermions, via one dark mediator;
4. the DM couples to $u,d$ quarks, so that it can be produced at the LHC;
5. the coupling to the first generation of quarks is no less than the coupling to other SM fermions.
In this situation, the relic density constraint gives a range within which the dark sector parameters should lie. It is clear that assumption 5 is by no means a certainty, and so we will show how our results are sensitive to relaxing this assumption. In the event of a signal, this assumption can instead be used to learn about the flavour structure of a thermal relic model that attempts to explain the signal. If the signal falls into the region where DM would be overproduced, then there must be enhanced couplings to other SM particles relative to $u,d$ quarks in order to avoid overproduction, or alternatively, the DM is produced by some mechanism other than thermal production.
Assumption 2 can break down if either the DM was not produced thermally in the early universe, or if some other effect breaks the relationship between the DM density and annihilation rate. For example, unusual cosmologies between freezeout and today can influence the relic density of DM [@Gelmini:2006pw].
To summarize, under our generic assumptions 1-5 the DM production cross section must satisfy the bounds 1.010\^[-9]{}[GeV]{}\^[-2]{}v\_[ann]{}v\_[\*]{}v\_[ann]{}4.010\^[-9]{}[GeV]{}\^[-2]{}, where the value of the annihilation cross section is dictated by ensuring the correct relic abundance.
These tidy inequalities break down when we include the effect of the top quark mass, mediator widths, and a more accurate expression for the relic density later in the text, although the principle behind them remains the same. The two limits on the cross section describe two contours in the parameter space: if $\sigvmin$ is too large, then DM will be underproduced, we call this the *underproduction line*; if $\sigvmin$ is too small, then DM will be overproduced; this is called the *overproduction line*. This information is summarised in the table below, where $g_{({\rm DM}, f)}$ generically indicate the mediator couplings to DM and SM fermions, respectively.
------------------------------- ------------------------------------- -------------------------------------------------------------
[ **Overproduction line**]{} $\sigv_{\rm ann} \simeq 4 \sigvmin$ **EFT:** Max $\Lambda$, min $\mx$.
**Simp. model:** Max $M$, min $g_{({\rm DM},f)}$ and $\mx$.
[ **Underproduction line**]{} $\sigv_{\rm ann} = \sigvmin$ **EFT:** Min $\Lambda$, max $\mx$
**Simp. model:** Min $M$, max $g_{({\rm DM},f)}$ and $\mx$.
------------------------------- ------------------------------------- -------------------------------------------------------------
Models and cross sections\[SigmaCalcs\]
---------------------------------------
To illustrate our point, we focus on a class of simplified models where the DM is a Dirac particle annihilating to SM fermions in the $s$-channel via a $Z'$-type mediator. This popular scenario has seen much attention, including searches by CDF [@Aaltonen:2008dn], ATLAS [@Aad:2012hf] and CMS [@Chatrchyan:2013qha].
Working with simplified models is more timely than ever. For some years, the use of effective operators has been popular as a way to place general constraints on the dark sector [@Cao:2009uw; @Beltran:2010ww; @Bai:2010hh; @Fan:2010gt; @Goodman:2010ku; @Cheung:2010ua; @Zheng:2010js; @Cheung:2011nt; @Rajaraman:2011wf; @Yu:2011by; @Goodman:2010yf; @Fox:2011pm; @Lowette:2014yta]. However, there has always been concern that this approximation breaks down at some mediator mass scale and it is now clear that the effective operator assumption is not a good approximation at LHC energies unless the DM-SM coupling is very large [@Goodman:2010yf; @Fox:2011pm; @Fox:2011fx; @Goodman:2011jq; @Shoemaker:2011vi; @Fox:2012ee; @Weiner:2012cb; @Busoni:2013lha; @Busoni:2014sya; @Busoni:2014haa; @Buchmueller:2013dya]. On the other end of the spectrum, studies of specific well-motivated models such as supersymmetry [@Chung:2003fi] or extra dimensions [@ArkaniHamed:1998rs] continue to play an important role, but the broad parameter space and specific assumptions required in these models make it difficult to draw general conclusions about the dark sector. Hence, simplified models have become the best way to constrain the DM parameter space [@Abdallah:2014dma; @Malik:2014ggr; @Buchmueller:2014yoa; @Alves:2011wf; @Harris:2014hga; @Buckley:2014fba]. However, this parameter space is still broad, and it is usually unfeasible to constrain the entire space in just one analysis. This necessitates a specific choice for one or more parameters – for example the coupling-strength and mediator-mass may be constrained for a specific choice of the DM mass. Clearly this is sub-optimal, since we do not want our constraints to be valid only for one arbitrary choice of an unknown parameter. It is important to remember that the search for new neutral particles with electroweak couplings is motivated by the existence of dark matter, and so the requirement that these particles are a viable thermal relic DM candidate can be a powerful motivator for these arbitrary choices.
There are many other simplified models to choose from. For example, one could consider a model where dark matter couples to the standard model via $s$-channel exchange of a scalar mediator. In these models, the dark sector usually couples to the standard model via mixing between the new dark mediator and the Higgs. This leads to a Yukawa-type mediator-SM coupling, proportional to the SM fermion mass. This suppresses the production rate via $u$ and $d$ quarks at the LHC relative to top quark loop-induced production via gluon initial states. This suppression also applies to the annihilation rate, especially if annihilation to top quarks is kinematically (or otherwise) unavailable, resulting in very large DM masses and couplings and small mediator masses in order to reach the correct relic density. Hence we do not consider this model here. Another alternative is DM coupling to SM particles via exchange of a scalar mediator in the $t$-channel, as studied in e.g. [@Bell:2012rg; @Chang:2013oia; @An:2013xka; @Bai:2013iqa; @DiFranzo:2013vra]. The phenomenology is a little different here, for example in the t-channel model the colored mediator can decay into a quark-DM pair [@Papucci:2014iwa]. Whilst this is an interesting model, we choose to study a $Z'$-type model as it has the best prospective LHC Run-II constraints with which to compare.
We consider the general interaction term in the Lagrangian for a vector mediator $Z'$, $$\mathcal{L} = -\sum_f Z'_\mu [\bar f \gamma^\mu (g^V_f - g^A_f \gamma_5) f] - Z'_\mu \
[\bar \chi \gamma^\mu(\gx^V - \gx^A \gamma_5)\chi],
\label{lagr}$$ where $f$ is a generic SM fermion, the kinetic and gauge terms have been omitted, and the sum is over the quark and lepton flavours of choice (see e.g. Ref. [@Buchmueller:2014yoa]).
The LHC searches are only mildly sensitive to the ratios $g^V_f/g^A_f$ and $\gx^V/\gx^A$, however the distinction is important for relic density calculations, and so we consider a pure vector coupling ($g^A_{f, \rm{DM}}=0$). In the EFT limit, we also consider pure axial ($g^V_{f, \rm{DM}}=0$) interactions. In the low-energy limit, the Lagrangian (\[lagr\]) leads to the effective operators \_V&=&\[|\^\]\[|f\_f\],\
[O]{}\_A&=&\[|\^\^5\]\[|f\_\^5 f\]. The effective operators ${\cal O}_V$ and ${\cal O}_A$ correspond to the usual D5 and D8 operators respectively, defined in Ref. [@Goodman:2010ku].
The process relevant for relic density calculations is the annihilation of DM particles of mass $\mx$ into SM fermions of mass $m_f$ |f|f. \[process\] In the effective operator limit, the relative cross sections per SM fermion flavour, expanded up to order $v^2$, are $$\begin{aligned}
\label{sigv-EFT-V}(\sigma v)_{*}^V &\simeq& \frac{N_C \mx^2 }{2 \pi \Lambda^4} \left(\sqrt{1-\frac{m_f^2}{\mx^2}}\left(\frac{m_f^2}{\mx^2}+2\right)+v^2\frac{11m_f^4/\mx^4+2m_f^2/\mx^2-4}{24\sqrt{1-m_f^2/\mx^2}}\right),\\
(\sigma v)_{*}^A &\simeq& \frac{N_C}{2 \pi \Lambda^4} \left(m_f^2\sqrt{1-\frac{m_f^2}{\mx^2}}+v^2\frac{23m_f^4/\mx^2-28m_f^2+8\mx^2}{24\sqrt{1-m_f^2/\mx^2}}\right)\label{sigv-EFT-A}.\end{aligned}$$ where the colour factor $N_C$ is equal to 3 for quarks and 1 for colourless fermions. The full expressions relative to the process (\[process\]) with $Z'$ exchange, and the corresponding mediator widths, are reported in Appendix \[app:cross-sections\].
Results: Effective operator limit\[EFTresults\]
===============================================
In the extreme EFT limit, for massless SM annihilation products, the annihilation cross section for a dimension-6 operator goes like $\gx^2 g_f^2 \mx^2 / M^4\equiv \mx^2 / \Lambda^4$, where $M$ is the mediator mass, and $g_f$ is its coupling with fermion species $f$. Thus, in general, the underproduction contour is a contour of maximum $\gx$, $g_f$, and $\mx$, and of minimum $M$, and vice-versa for the overproduction contour. Here we compare a range of constraints in the effective operator scenario, where the momentum carried by the mediator is assumed to be small relative to its mass and we define [^2] . The LHC constraints in this scenario are generally valid in the range $\pi \lesssim \sqrt{\gx g_f} \lesssim 4\pi$ [@ATL-PHYS-PUB-2014-007]. Since the annihilations relevant to relic density calculations take place when the DM is non-relativistic, the effective operator approximation is valid as long as $M \gg 2\mx$, or $\sqrt{\gx g_f} \gg 2\mx/ \Lambda$, while direct detection constraints are valid across the entire parameter space of interest.
Our results in this limit are summarized in Fig. \[fig:EFT\], where we compare the projected exclusion and discovery reach by ATLAS with the *under*- and *over*-production lines defined in the previous Section for the vector and axial-vector operator. In the following subsections we describe all the elements appearing in Fig. \[fig:EFT\].
ATLAS reach {#subsec:ATLreach}
-----------
We use simulations of the exclusion and discovery reach of ATLAS at 14 TeV from Ref. [@ATL-PHYS-PUB-2014-007]. This reference estimates the sensitivity of ATLAS to DM in the missing energy + jets channel. This is a powerful general-purpose channel which has led to strong constraints on DM by both ATLAS and CMS at 7 and 8 TeV [@monojetATLAS1; @Chatrchyan:2012me]. Searches for other final states such as mono-W/Z [@Bell:2012rg; @Carpenter:2012rg; @ATLASWZ; @Aad:2014vka], mono-photon [@monogammaATLAS1; @monogammaCMS1], mono-higgs [@Petrov:2013nia; @Carpenter:2013xra] and mono-top [@Andrea:2011ws] can play a complementary role, especially when combined into mono-all searches. However, the monojet searches still give the strongest constraints [@Zhou:2013fla], and thus represent a good choice for sensitivity studies.
The limits from Ref. [@ATL-PHYS-PUB-2014-007] are only given for two DM masses, $\mx = \{ 50, 400 \}$ GeV, however there is minimal variation in the constraint between the two masses, so we interpolate constraints on $\Lambda$ between these two points.[^3] These limits are determined for the vector operator, but are expected to be the same for the axial-vector operator [@monojetATLAS1].
The 1% and 5% labels indicate projected limits assuming a 1% or 5% systematic uncertainty in the SM background, respectively. Achieving 1% systematics may be overly optimistic, and can be considered a “best-case scenario”. Other labels indicate the results at a given collision energy and integrated luminosity. The red bands indicate the potential significance of an observed signal, from 3$\sigma$ to 5$\sigma$.
![ Blue and orange lines show the under- and over-production lines respectively, defined in the text, for the Vector (D5) (left) and Axial-Vector (D8) (right) operators. The black lines show prospective ATLAS exclusion limits for various energies and systematic uncertainties, and for luminosities of (3000, 300, 20) fb$^{-1}$ from top to bottom. The red bands show the 3 to 5 $\sigma$ discovery potential [@ATL-PHYS-PUB-2014-007]. EFT approximation is valid for $\pi \lesssim \sqrt{\gx g_f} < 4\pi$ for ATLAS prospects, and $\sqrt{\gx g_f} \gg 2\mx/ \Lambda$ for the relic density constraints. See text for more details. Direct detection constraints are not shown, but for the vector operator D5 they would rule out the entire visible space (cf. sect. \[subsec:dd\]). \[fig:EFT\]](D5 "fig:"){width="49.00000%"} ![ Blue and orange lines show the under- and over-production lines respectively, defined in the text, for the Vector (D5) (left) and Axial-Vector (D8) (right) operators. The black lines show prospective ATLAS exclusion limits for various energies and systematic uncertainties, and for luminosities of (3000, 300, 20) fb$^{-1}$ from top to bottom. The red bands show the 3 to 5 $\sigma$ discovery potential [@ATL-PHYS-PUB-2014-007]. EFT approximation is valid for $\pi \lesssim \sqrt{\gx g_f} < 4\pi$ for ATLAS prospects, and $\sqrt{\gx g_f} \gg 2\mx/ \Lambda$ for the relic density constraints. See text for more details. Direct detection constraints are not shown, but for the vector operator D5 they would rule out the entire visible space (cf. sect. \[subsec:dd\]). \[fig:EFT\]](D8 "fig:"){width="49.00000%"}
Direct Detection constraints {#subsec:dd}
----------------------------
We use corrected versions of the equations from Ref. [@Goodman:2010ku] to translate limits on the spin-dependent (SD) and spin-independent (SI) cross sections into limits on the effective operator parameter $\Lambda$. In this mass range, the strongest limits are currently from LUX [@Akerib:2013tjd] (SI cross section) and Xenon100 [@Aprile:2013doa] (SD cross section). For our simplified models, constraints on $\Lambda$ correspond to a constraint on $M/\sqrt{\gx g_f}$.
The vector operator ${\cal O}_V$ is subject to constraints on the spin-independent scattering cross section. These constraints are significantly stronger than prospective LHC bounds on this operator, ruling out the entire region displayed in Fig. \[fig:EFT\] (left). However, the strength of direct detection constraints falls of quickly below $\mx \simeq 10$ GeV, while LHC constraints are expected to be relatively flat below $\mx = 50$ GeV. If the prospective LHC constraints in Fig. \[fig:EFT\] (left) can be extrapolated down, they will become stronger than direct detection constraints at around $\mx = 10$ GeV. Conversely, the axial-vector operator ${\cal O}_A$ is subject to much weaker constraints on the spin-dependent scattering cross section. In this range they are barely distinguishable from the $\Lambda=0$ line and thus are not shown.
Relic Density Bounds
--------------------
In Fig. \[fig:EFT\], we show the *under*- and *over*-production lines defined in the previous Section, for the vector (${\cal O}_V$, D5) and axial-vector (${\cal O}_A$, D8) operator, under the assumptions 1-5 of Sect. \[subsec:DM-Ab-Cons\]. The range between the orange and blue lines shows the region of parameter space in which any observed $\chi$ can also be thermal relic DM. This marks a good starting point for WIMP searches. For example, we can see that pure vector DM will be difficult to observe for larger DM masses, and in any case it is ruled out by direct detection constraints. Conversely, axial-vector DM is unconstrained by direct detection, but it is already heavily constrained by 8 TeV collider bounds, and it is accessible to the 14 TeV searches even for DM masses above 500 GeV. The jump in the orange line is the point where annihilation into top quarks becomes kinematically allowed.
The overproduction lines in Fig. \[fig:EFT\] rely on the assumption that the DM coupling to the first generation of quarks is not less than the coupling to other SM fermions ($g_f\leq g_{u,d}$), while the underproduction line only depends on the couplings $g_{u,d}$ to the first-generation quarks. Relaxing/strengthening the assumption 5 of Sect. \[subsec:DM-Ab-Cons\] means allowing the couplings to other SM fermions to span over a wider/smaller range and correspondingly the upper limit is Eq. (\[second\]) is changed. The effect on the overproduction lines is shown in Fig. \[fig:EFT\_scan\]. We see that if the constraint on $g_f$ is relaxed, the orange line of Fig. \[fig:EFT\] gradually becomes too strong, and correspondingly the region in which to search for DM becomes broader (green curves of Fig. \[fig:EFT\_scan\]). In the event that a signal compatible with DM is observed, the region where it falls on the plot can be used to infer something about its nature. To be a credible DM candidate, it must either have a production mechanism aside from the usual thermal production, or its couplings to other SM particles can be inferred from where its parameters fall on this plot.
![ Analogues of the orange overproduction line in Fig. \[fig:EFT\_scan\] (corresponding to $g_f\leq g_{u,d}$), changing the relative value of the coupling between $u,d$ quarks and other SM fermions. \[fig:EFT\_scan\]](D5_scan "fig:"){width="49.00000%"} ![ Analogues of the orange overproduction line in Fig. \[fig:EFT\_scan\] (corresponding to $g_f\leq g_{u,d}$), changing the relative value of the coupling between $u,d$ quarks and other SM fermions. \[fig:EFT\_scan\]](D8_scan "fig:"){width="49.00000%"}
It is also interesting to note that in the EFT limit there exist simple expressions relating the DM parameters to the correct relic density. In fact, one can trade the annihilation cross section for the relic abundance (see e.g. Eq. (\[simplerelic\])). Then, using the Eqs. (\[sigv-EFT-V\]) and (\[sigv-EFT-A\]) for the annihilation cross section in the EFT limit, we can find a simple expression for a combination of the effective parameters in the low-energy theory $$\begin{aligned}
\frac{\mx^2}{\Lambda^4} \times\sum_i N^C_i&\simeq& 1\times 10^{-8} \,{\rm GeV}^{-2}\quad (V), \\
\frac{m_f^2 + \mx^2 v^2 / 3}{\Lambda^4}\times\sum_i N^C_i &\simeq& 2.5\times 10^{-8} \,{\rm GeV}^{-2}\quad (A),\end{aligned}$$ where the sum is over the fermionic annihilation products and the colour factor $N^C_i$ is 3 for coloured fermions and 1 for colourless fermions, and this equation assumes $\Lambda$ is the same for all channels.
Results: Simplified models\[SIMPresults\]
=========================================
We again consider the two scenarios discussed in the previous section: DM coupling to the minimum and maximum number of SM particles. Now we relax the assumptions leading to the effective operator approximation, and consider the simple UV-complete model described by the Lagrangian (\[lagr\]). This expands the relevant parameter space from just two parameters, $\mx$ and $\Lambda$, to the set of parameters $\{\mx, M, \gx^V, \gx^A, g_f^V, g_f^A\}$, where $f$ runs over all SM fermions which the mediator can decay into. As already anticipated, we restrict our attention to the case of pure vector couplings, for which ATLAS projected limits exist [@ATL-PHYS-PUB-2014-007]. Thus we consider $\gx^A = g_f^A = 0$, and we define $\gx^V \equiv \gx$, $g_f^V \equiv g_f$. The annihilation rates and mediator decay widths have been computed and are shown in Appendix \[app:cross-sections\].
For the *overproduction line*, any change in parameters which decreases the cross section will lead to overproduction of DM. Similarly, for the *underproduction line*, any change in parameters which increases the cross section will lead to underproduction of DM.
In order to compare directly with prospective ATLAS constraints, in Figs. \[fig:Mvsmx\]-\[fig:gvsM\] we show lines for specific choices of $\sqrt{\gx g_f}=$0.5, 1, $\pi$ and $\mx=$50, 400 GeV respectively. The ATLAS constraints are again from Ref. [@ATL-PHYS-PUB-2014-007] and refer to a vector mediator model. These constraints have some degeneracy in $M$ for low values of $\sqrt{\gx g_f}$, and so we do not show a line corresponding to $\sqrt{\gx g_f}=0.5$ In order to compare with their prospective constraints, the relic density constraints assume the same (arbitrary) widths as ATLAS.
While the annihilation rate of DM particles only depends on the product $\gx g_f$, the mediator decay widths depend on each coupling individually. So we are forced to fix the ratio $g_f / \gx$, in addition to keeping the product $\gx g_f$ as a parameter,. For fixed values of the mediator width, a bound on the product $\sqrt{\gx g_f}$ can be recast into a bound on the ratio $g_f / \gx$. The arbitrary widths used in Figs. \[fig:Mvsmx\]-\[fig:gvsM\] can be compared to the physical widths to fix the ratio $g_f / \gx$. This is shown in Fig. \[fig:ratio\]. In some regions there is no solution, and the width used by ATLAS is in fact not physical. For this reason we recommend to avoid the use of arbitrary mediator widths, and suggest instead that the widths are fixed to their minimal value given by the decay channels to SM particles and to DM particles.
![ Over- (orange) and under- (blue) production boundary lines for thermal relic dark matter, for three different choices of the coupling strengths, and a $Z'$-type mediator with pure vector couplings. Black lines are ATLAS projected 95% lower bounds after 25 fb$^{-1}$ at 14TeV, assuming 5% systematic uncertainties. \[fig:Mvsmx\]](Mvsmx-V-Mon3 "fig:"){width="49.00000%"} ![ Over- (orange) and under- (blue) production boundary lines for thermal relic dark matter, for three different choices of the coupling strengths, and a $Z'$-type mediator with pure vector couplings. Black lines are ATLAS projected 95% lower bounds after 25 fb$^{-1}$ at 14TeV, assuming 5% systematic uncertainties. \[fig:Mvsmx\]](Mvsmx-V-Mon8pi "fig:"){width="49.00000%"}
![ Over- (orange) and under- (blue) production boundary lines for thermal relic dark matter, compared with projected ATLAS reach (black), for two values of the dark matter mass, and a $Z'$-type mediator with pure vector couplings. Black lines are ATLAS projected 95% upper bounds after 25 fb$^{-1}$ at 14TeV, assuming 5% systematic uncertainties. \[fig:gvsM\]](gvsM-V-Mon3 "fig:"){width="49.00000%"} ![ Over- (orange) and under- (blue) production boundary lines for thermal relic dark matter, compared with projected ATLAS reach (black), for two values of the dark matter mass, and a $Z'$-type mediator with pure vector couplings. Black lines are ATLAS projected 95% upper bounds after 25 fb$^{-1}$ at 14TeV, assuming 5% systematic uncertainties. \[fig:gvsM\]](gvsM-V-Mon8pi "fig:"){width="49.00000%"}
![ The solution to the ratio $g_f / \gx$ corresponding to the bounds on the product $g_f\cdot \gx$ combined with fixed mediator widths (as represented in Fig. \[fig:gvsM\]). At large mediator masses, no solution exists and the widths are unphysical for the coupling strengths in Fig. \[fig:gvsM\]. \[fig:ratio\]](ratio-V-Mon3 "fig:"){width="49.00000%"} ![ The solution to the ratio $g_f / \gx$ corresponding to the bounds on the product $g_f\cdot \gx$ combined with fixed mediator widths (as represented in Fig. \[fig:gvsM\]). At large mediator masses, no solution exists and the widths are unphysical for the coupling strengths in Fig. \[fig:gvsM\]. \[fig:ratio\]](ratio-V-Mon8pi "fig:"){width="49.00000%"}
Conclusion\[conclusion\]
========================
The upcoming LHC searches for new weakly interacting stable particles may indeed provide some positive signal in the near future. Then, how confident can we be in claiming that the new particle actually accounts for the DM of the universe?
In this paper we have stressed the importance of using relic density considerations in the searches for DM at the next LHC Run, not only regarded as a mere constraint but also used as a powerful search tool. In fact, in order to reveal the true nature of DM, any future signal of a new weakly interacting particle possibly produced in the collider must be confronted with the requirement the new particle has a relic abundance compatible with observations before assigning it the label of thermal DM.
We have followed both the approach of effective operators (in terms of which most experimental analyses are carried out) and the approach of simplified models, for a reference case of a vector mediator. We have found that, in both situations, the forthcoming Run II of LHC has the potential to explore a large portion of the parameter space of thermal-relic DM, either in terms of claiming discovery or in terms ruling out models.
The results of this paper are twofold. One the one hand, they can be used by LHC collaborations as a guidance into the parameter space of DM models; in fact, simple relic density considerations help to set priorities and parameter choices when analysing future data in terms of DM.
On the other hand, our results provide clear messages in case of observation of a new stable particle: if the new particle is not compatible with our thermal relic curves, either it is not the DM or one of our working assumptions is not valid. In any case, very interesting lessons about the nature of DM will be learned from LHC data.
Acknowledgments {#acknowledgments .unnumbered}
===============
We thank A. Boveia, C. Doglioni and S. Schramm for many interesting conversations. ADS acknowledges partial support from the European Union FP7 ITN INVISIBLES (Marie Curie Actions, PITN-GA-2011-289442).
Cross Sections and Widths\[app:cross-sections\]
===============================================
In this Appendix we collect the results of cross sections calculations for the process of DM annihilation into SM fermions |f|f We performed the calculation at zero temperature in the lab frame where $\chi$ is at rest, and the center of mass energy $s = 2 \mx^2\left(\frac{1}{\sqrt{1-v^2}}+1\right)$. This is equivalent to performing the calculation in the Moeller frame, and is the correct frame for the relic density calculations [@Gondolo:1990dk].
Full expressions
----------------
$$\begin{aligned}
(\sigma v)_V &=& \frac{N_C (g_f^V)^2 (\gx^V)^2}{2\pi} \frac{\sqrt{1-m_f^2/\mx^2}}{(M^2 - 4\mx^2)^2 + \Gamma^2 M^2} \Big[(m_f^2+2\mx^2) +
\nonumber\\
&& v^2\Big(\frac{11m_f^4+2m_f^2\mx^2-4\mx^4}{24\mx^2(1-\frac{m_f^2}{\mx^2})}+2\frac{\mx^2(m_f^2+2\mx^2)(M^2-4\mx^2)}{(M^2 - 4\mx^2)^2 + \Gamma^2 M^2}\Big)\Big],\\
(\sigma v)_A &=& \frac{N_C (g_f^A)^2 (\gx^A)^2}{2 \pi}
\frac{\sqrt{1- m_f ^2/ m_{\rm DM} ^2}}{(M^2 - 4 m_{\rm DM} ^2)^2 + \Gamma^2 M^2}
\bigg[\bigg(1 - 4\frac{ m_{\rm DM} ^2}{M^2}\bigg)^2 m_f ^2 +
\nonumber\\
&& v^2\bigg(\Big
(\frac{2m_f^2 m_{\rm DM}^2}{M^4}+\frac{2m_{\rm DM}^2}{M^2}-\frac{3m_f^2}{M^2}+\frac{23m_f^2}{24m_{\rm DM}^2}-\frac{7}{6}\Big)\frac{ m_f ^2}
{(1- m_f ^2/ m_{\rm DM} ^2)}
+\frac{ m_{\rm DM} ^2}
{3(1- m_f ^2/ m_{\rm DM} ^2)}
\nonumber\\
&&+
\frac{2 (M^2 - 4 m_{\rm DM} ^2)^3 m_{\rm DM} ^2 m_f ^2}
{M^4 \left( (M^2 - 4 m_{\rm DM} ^2)^2 + \Gamma^2 M^2 \right)}
\bigg)\bigg].\end{aligned}$$
Limit $m_f \rightarrow 0$
-------------------------
$$\begin{aligned}
(\sigma v)_V &=& \frac{N_C (g_f^V)^2 (\gx^V)^2}{\pi} \frac{\mx^2}{(M^2 - 4\mx^2)^2 + \Gamma^2 M^2} \left[1 +v^2\left(-\frac{1}{12}+\frac{2\mx^2(M^2-4\mx^2)}{(M^2-4\mx^2)^2+\Gamma^2M^2}\right) \right],\nn\\
&&\\
(\sigma v)_A &=& \frac{N_C (g_f^A)^2 (\gx^A)^2}{6 \pi} \frac{\mx^2}{(M^2 - 4\mx^2)^2 + \Gamma^2 M^2} v^2.\end{aligned}$$
Effective Operator Approximation
--------------------------------
$$\begin{aligned}
(\sigma v)_V &=& \frac{N_C \mx^2 }{2 \pi \Lambda^4} \sqrt{1-\frac{m_f^2}{\mx^2}}\left[\left(\frac{m_f^2}{\mx^2}+2\right)+v^2\frac{11m_f^4/\mx^4+2m_f^2/\mx^2-4}{24(1-m_f^2/\mx^2)}\right],\\
(\sigma v)_A &=& \frac{N_C}{2 \pi \Lambda^4}\sqrt{1-\frac{m_f^2}{\mx^2}} \left[m_f^2+v^2\frac{23m_f^4/\mx^2-28m_f^2+8\mx^2}{24(1-m_f^2/\mx^2)}\right].\end{aligned}$$
Widths
------
The widths for the vector mediator decay to a pair of fermions are given by $$\begin{aligned}
\Gamma_V & = & \frac{N_C (g_f^V)^2 (M^2 + 2m_f^2) \sqrt{1 - 4m_f^2/M^2}}{12 \pi M}, \\
\Gamma_A & = & \frac{N_C (g_f^A)^2 M (1 - 4m_f^2/M^2)^{3/2}}{12 \pi}.\end{aligned}$$
Relic density general formalism
===============================
Our technique to compute the abundance and notation follow Refs. [@Gondolo:1990dk] and [@Bertone:2004pz]. First we find the freezeout temperature by solving
$$e^{x_F} = \frac{\sqrt{\frac{45}{8}} g_{\rm DoF} \mx M_{\rm Pl} c(c+2) \sigv}{2 \pi^3 g_\star^{1/2} \sqrt{x_F}},$$
where $x = \mx / T$ and subscript $F$ denotes the value at freezeout, $g_{\rm DoF} = 2$ is the number of degrees of freedom of the DM particle, $c$ is a matching constant usually taken to be $1/2$, $g_\star$ is the number of relativistic degrees of freedom, $M_{\rm Pl} = 1/\sqrt{G_N}$ is the Planck mass. Usually, it is safe to expand in powers of the velocity and use the approximation $$\sigv = a + b \langle v^2 \rangle + {\cal O}(\langle v^4 \rangle) \simeq a + 6b/x_F.$$ However, when the mediator width is small, this approximation can down near the $s$-channel resonance in the annihilation rate at $M \simeq 2\mx$ [@Griest:1989wd; @Griest:1990kh; @Gondolo:1990dk] if the width is small. Around this point it becomes more accurate to use the full expression, $$\sigv = \frac{x}{8 \mx^5 K_2^2[x]}\int_{4\mx^2}^\infty {\rm d}s\, \sigma (s-4\mx^2) \sqrt{s} K_1[\sqrt{s} \,x/\mx].$$ With this information, one can calculate the relic abundance, $$\Omega_{\rm DM}h^2 = \Omega_\chi h^2 + \Omega_{\bar\chi}h^2= \frac{2\times 1.04\times 10^9 \, {\rm GeV}^{-1} \mx}{M_{\rm Pl} \int_{T_0}^{T_F} g_\star^{1/2} \sigv {\rm d}T},\label{relic-full}$$ where the factor of 2 is for Dirac DM. When the non-relativistic approximation to the annihilation rate holds, this simplifies to $$\Omega_{\rm DM} h^2 = \frac{2\times 1.04\times 10^9 \, {\rm GeV}^{-1} x_F}{\overline{g_\star}^{1/2} \, M_{\rm Pl} \,(a + 3 b / x_F)}
\label{relic-simple}$$ where $\overline{g_\star}^{1/2}$ is a typical value of $g_\star^{1/2}(T)$ in the range $T_0\leq T \leq T_F$. We have tested the validity of this approximation and find that there is a negligible difference to the full relativistic calculation, since the widths we consider are relatively large. If the physical widths are used, then care should be taken that this approximation still holds when the width becomes small, especially when the annihilation rate has a larger $p$-wave component.
[^1]: Gluons and other quarks can of course contribute to DM production at the LHC, so the $_*$ subscript defines a reference channel rather than all possible channels of DM production at the LHC.
[^2]: The parameter $\Lambda$ is sometimes called $M_\star$.
[^3]: We thank Steven Schramm for discussions on this point.
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- |
Yoshua Bengio\
Université de Montréal, MILA
bibliography:
- 'strings.bib'
- 'aigaion.bib'
- 'ml.bib'
title: The Consciousness Prior
---
Introduction
============
We propose here a new kind of prior for top-level abstract representations, inspired by our understanding of consciousness as a form of awareness [@vanGulick-2004], i.e., as defined by Locke, consciousness is “the perception of what passes in a man’s own mind”, or awareness of an external object or something within oneself (Wikipedia definition). This proposal is based on a regularization term which encourages the top-level representation (meant to be at the most abstract level) to be such that when a sparse attention mechanism focuses on a few elements of the state representation (factors, variables or concepts, i.e. a few axes or dimensions in the representation space), that small set of variables of which the agent is aware at a given moment can be combined to make a useful statement about reality or usefully condition an action or policy. We do not refer here to more elusive meanings that have been attributed to the word “consciousness” (like qualia [@Kriegel-2014]), sticking instead to the notion of attentive awareness in the moment, our ability to focus on information in our minds which is accessible for verbal report, reasoning, and the control of behaviour.
Consciousness Prior Theory
==========================
The following points can be derived from the basic idea introduced above, in the context of a learning agent, where we refer the reader to standard notions [@Sutton+Barto-98] of reinforcement learning (RL).
Subnetworks
-----------
Let $s_t$ be the [**observed state**]{} at time $t$ and let $h_t$ be the high-level representation derived from $s_t$ (and from past observed values $s_{t-k}$ in the partially observable case). For example, $h_t$ could be the output of some kind of RNN (with whatever architecture is appropriate) that reads the sequence of $s_t$ as input and produces an output $h_t$ at each time step: $$\begin{aligned}
h_t = F(s_t, h_{t-1})\end{aligned}$$ where we call $F$ the [**representation RNN**]{} or encoder and $h_t$ the [**representation state**]{}. A core objective is to learn good representations in $h_t$, which disentangles abstract explanatory factors, in the sense that there exist a simple transformation of $h_t$ which can select the information about a single factor (its value or uncertainty about it).
We can think of the representation RNN as the content of almost the whole brain at time $t$, i.e., the representation state $h_t$ is a very high-dimensional vector (and probably sparse if we want to imitate biology), which is an abstract representation of the full current information available to the agent (beyond what is stored in the weights), thus summarizing the current and recent past observations.
In contrast, we will define the [**conscious state**]{} $c_t$ as a very low-dimensional vector which is derived from $h_t$ by a form of attention mechanism applied on $h_t$, taking into account the previous conscious state as context. $$\begin{aligned}
c_t = C(h_t, c_{t-1}, z_t)\end{aligned}$$ where $z_t$ is a random noise source. The cognitive interpretation is that the value of $c_t$ corresponds to the content of a thought, a very small subset of all the information available to us unconsciously, but which has been brought to our awareness by a particular form of attention which picks several elements or projections from $h_t$. The function $C$ is the [**consciousness RNN**]{} and because of its random noise inputs, produces a random choice of the elements on which the attention gets focused. This is useful if we think of the consciousness RNN as a tool for exploring interpretations or plans or to sample predictions about the future. We can also think of the consciousness RNN as the tool to isolate a particular high-level abstraction and extract the information about it (its value, uncertainty about it or even the fact that it is unobserved). This would happen if we think about a single factor, but in general $C$ will aggregate a few (e.g. a handful) of such factors into a more complex and composed thought.
Training Objectives
-------------------
To capture the assumption that a conscious thought can encapsulate a statement about the future, we introduce a [**verifier network**]{} which can match a current representation state $h_t$ with a past conscious state $c_{t-k}$: $$\begin{aligned}
V(h_t, c_{t-k}) \in {\mathbb R}\end{aligned}$$ which should be structured so that $V(h_t, c_{t-k})$ indicates the consistency of $c_{t-k}$ with $h_t$, e.g., estimating the probability of the corresponding statement being true, given $h_t$.
More generally, we would like to define an objective (or reward) function which embodies the idea that the attended (conscious) elements are used in some way whose value can be quantified and optimized, i.e., that the representation RNN is trained to optimize this objective function, as well as possibly other objectives such as being able to reconstruct the raw input or any other supervised, RL, or unsupervised objective which we want to throw in, such as the independently controllable factors prior [@bengioe+al-2017-control-arxiv].
There are two distinct mechanisms at play which contribute to map the high-level state representation to the objective function: (1) the attention mechanism (e.g. the consciousness RNN) which selects and combines a few elements from the high-level state representation into a low-dimensional “conscious sub-state” object (the current content of our consciousness), and (2) the predictions or actions which are derived from the sequence of these conscious sub-states. The second mechanism is easy to grasp and frame in standard ML practice, either in deep learning or RL, e.g. for supervised or unsupervised or RL tasks. For example, the attention mechanism could select elements $B$ from the current representation state and choose to make a prediction about future elements $A$. Then to improve the quality of the prediction mechanism we may just want to maximize $log P(A | B)$ or some proxy for it, e.g., using a variational auto-encoder [@Kingma+Welling-ICLR2014] objective or a a conditional GAN [@mirza2014conditional] if one wants to sample accurately an $A$ from $B$. Note again that such an objective function is not just used to learn the mapping from $B$ to $A$ (or to probabilities over the space of $A$ values), but also drives the learning of the representation function itself, i.e., is back-propagated into the representation RNN). However, this part of the objective function (e.g. predictive value, computed by $V$ above) is not sufficient and in fact is not appropriate to train the attention mechanism itself (which variables $A$ and $B$ should be selected?). Indeed, if that was the driving objective for attention, the learner would always pick a $B$ which is trivially predictable (and there are such aspects of reality which are trivially predictable yet do not help us to further understand the world and make sense of it or achieve our goals). It remains an open question what other objectives would be appropriate for learning how to attend to the most useful elements, but ultimately we should be able to use the actual RL reward of the learning agent for that purpose (though some shorter-term proxy might be welcome). Some form of entropy or diversity may be needed so that the attention mechanism is stochastic and can choose a very diverse set of possible attended elements, so as to cover widely the possible variables $A$ on which a prediction is made.
Naming Variables and Indirection
--------------------------------
It would be very convenient for the consciousness attention mechanism and for the verifier network to be able to refer to the “names” of variables on which a prediction is made. In some models, we already distinguish keys and values in variations of memory augmented neural networks [@Weston2014; @Graves2014]. The conscious state must indirectly [*refer*]{} to some of the aspects or dimensions computed in the representation $h$. Wether this should be done explicitly or implicitly remains to be determined. A key-value mechanism also makes it easier for the verifier network to do its job because it must match just the [*key*]{} of the predicted variable with its instances in a future representation state (with that variable becomes observed). If the key and value are mixed up and the predicted value differs substantially from the observed value, a simple associative process might miss the opportunity to match these and thus provide a strong training signal (to correct the predictor).
Connection to Language and Symbolic Knowledge Representation
------------------------------------------------------------
Linked to this is the interesting property of the conscious state that there is a fairly simple transformation of it into a natural language sentence (possibly taking into account previously uttered sentences). An externally provided sentence could also elicit an associated conscious state, although we postulate that the conscious state is generally a richer object than the uttered sentence, i.e., mapping from conscious states to sentences loses information (think about visual imagery, or artistic expression, which are difficult to put in words), and the same sentence could thus be interpreted differently depending on context and the particulars of the agent who reads that sentence. Formally, we could use another RNN to map a conscious state to an utterance $u_t$: $$\begin{aligned}
u_t = U(c_t, u_{t-1}).\end{aligned}$$ A learning agent which uses language could thus benefit from an additional regularization term: the set of currently consciously attended elements can often be mapped to something like a sentence in natural language which may be uttered by another agent, such as a human teacher. A sentence focuses on just a handful of elements and concepts, unlike our full internal state. This imposes further constraints on the representation function in that its individual elements or dimensions are more likely to correspond to concepts which can typically be expressed by a single word or phrase. Based on these arguments, it is reasonable to hypothesize that language may actually help humans build sharper internal representations (which are better disentangled) as well as facilitate learning – see the arguments around curriculum learning [@Bengio+al-2009-small] and cultural learning [@bengio-2014-cultural] – and enable collaborative task-solving.
Along the same line, this research opens the door to the possibility of better connecting deep learning with classical symbolic AI and cognitive science, and move deep learning from perception (where it shines) to higher-level cognition and knowledge representation (where many questions remain open). For example, knowledge is classically represented by facts and rules: each of them is a very sharp statement about reality involving just a few concepts. Such a nugget of information or knowledge seems to fit well as a conscious state. Combining such conscious states sequentially in order to make more complex predictions and inferences or actions is basically what reasoning is about. However, I am not implying that we should return to the symbolic forms of knowledge representation, which have their well-known limitations. Instead, we may consider a form of regularization on the representations captured by deep learning agents, which could have many of attributes of classical AI facts and rules, while keeping a richer representation needed for inference and planning in the presence of uncertainty and non-discrete aspects of the world. Progress in this direction would also address the often expressed concern about obtaining explanations from deep nets, since the approach proposed here would make it easier for a trained agent to communicate verbally its high-level state.
Considerations for Experimenting with the Consciousness Prior
=============================================================
Because this is a novel theory which may be developped in many different ways, it is important to start with simple toy experiments allowing one to test and evaluate qualitatively different approaches, such that the turnaround time for each experiment is very short and the analysis of the representations learned very easy (because we already have a preconceived idea of what concepts would be the most appropriate to disentangle).
Although working with natural language input would be likely to help the agent learn better and more abstract representations, it would be better to start with experiments with no linguistic input, to make sure that it is the training objective and the training framework alone which are leading to the discovery of the appropriate high-level features. For example, learning some form of intuitive physics is done by babies without the need for linguistic guidance. Similarly, although the consciousness prior could be used in supervised learning or task-oriented RL, testing its ability alone to discover high-level abstractions would be best done in the context of unsupervised RL, e.g., using an intrinsic reward which favours the discovery of how the environment works.
It would be more interesting for the learning task to involve meaningful abstractions which have a high predictive power. For example, consider predicting whether a pile of blocks will fall on or off a table. It involves a high-level discrete outcome which can be predicted easily, even if the details of where the blocks will fall is very difficult even for humans to predict. In that case, predicting the future at the pixel level would be extremely difficult because future states have high entropy, with a highly multi-modal distribution. However, some aspects of the future may have low entropy. If in addition, these aspects have a big impact on predicting what will come next (or on taking the right decisions now), then the consciousness prior should be very useful.
In terms of experimental comparisons, it would be good to compare systems based on the consciousness prior with systems based on more common RL approaches such as policy gradient deep RL on one hand, or model-based RL on the other hand (still with neural nets to learn the transition operator in sensory space, as well as the reward function). Even better, in toy problems we can compute the oracle solution, so we can get an upper bound on the best achievable performance.
Acknowledgements {#acknowledgements .unnumbered}
================
The author wants to thank Philippe Beaudoin, Gerry (Tong) Che, William Fedus, Devon Hjelm and Anirudh Goyal for preliminary discussions about the consciousness prior, as well as funding from NSERC, CIFAR, the Canada Research Chairs, and the Open Philanthropy Project.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Weyl semimetals harbor unusual surface states known as Fermi arcs, which are essentially disjoint segments of a two dimensional Fermi surface. We describe a prescription for obtaining Fermi arcs of arbitrary shape and connectivity by stacking alternate two dimensional electron and hole Fermi surfaces and adding suitable interlayer coupling. Using this prescription, we compute the local density of states – a quantity directly relevant to scanning tunneling microscopy – on a Weyl semimetal surface in the presence of a point scatterer and present results for a particular model that is expected to apply to pyrochlore iridate Weyl semimetals. For thin samples, Fermi arcs on opposite surfaces conspire to allow nested backscattering, resulting in strong Friedel oscillations on the surface. These oscillations die out as the sample thickness is increased and Fermi arcs from the bottom surface retreat and weak oscillations, due to scattering between the top surface Fermi arcs alone, survive. The surface spectral function – accessible to photoemission experiments – is also computed. In the thermodynamic limit, this calculation can be done analytically and separate contributions from the Fermi arcs and the bulk states can be seen.'
address: 'Department of Physics, University of California at Berkeley, Berkeley, CA 94720, USA'
author:
- Pavan Hosur
bibliography:
- 'references.bib'
title: Friedel oscillations due to Fermi arcs in Weyl semimetals
---
*Weyl semimetals* (WSMs) are rapidly gaining popularity [@VolovikBook; @PyrochloreWeyl; @KrempaWeyl] as a new, gapless topological phase of matter, as opposed to topological insulators, which are gapped. A WSM is defined as a phase that has a pair of non-degenerate bands touching at a certain number of points in its Brillouin zone. Each such point or Weyl node has a *chirality* or a *handedness*; very general conditions constrain the right- and the left-handed Weyl nodes to be equal in number [@NielsenFermionDoubling1; @NielsenFermionDoubling2]. Near the nodes, the Hamiltonian resembles that of Weyl fermions well-known in high-energy physics. These nodes are topologically stable as long as translational symmetry is conserved, and can only be destroyed by annihilating them in pairs. Several theoretical proposals for realizing WSMs now exist in the literature [@PyrochloreWeyl; @KrempaWeyl; @WeylMultiLayer; @BernevigDoubleWeyl; @CarpentierWeyl; @ChoTItoWeyl; @HalaszWeyl; @JiangWeyl; @PhotonicCrystalWeyl]. WSMs have already been predicted to exhibit several interesting bulk properties, ranging from unusual quantum hall effects [@RanQHWeyl; @FangChernSemimetal] to various effects that rely on a 3D chiral anomaly present in this phase [@NielsenABJ; @AjiABJAnomaly; @QiWeylAnomaly; @RanQHWeyl; @SonSpivakWeylAnomaly; @ZyuninBurkovWeylTheta; @WeylCDW]. Preliminary bulk transport studies of WSMs have been performed both theoretically [@HosurWeylTransport; @BurkovNodalSemimetal; @WeylMultiLayer], as well as experimentally in some candidate materials [@WeylResistivityMaeno; @EuIridateExperiments].
A remarkable feature of WSMs is the existence of unconventional surface states known as *Fermi arcs* (FAs). These FAs are of a different origin from the FAs that exist in cuprate superconductors. A FA on a WSM is essentially a segment of a 2D Fermi surface (FS) that connects the projections of a pair of bulk Weyl nodes of opposite chiralities onto the surface Brillouin zone [@PyrochloreWeyl]. Although FAs always connect Weyl nodes of opposite chiralities, their exact shapes and connectivities depend strongly on the local boundary conditions. Such disconnected segments of zero energy states cannot exist in isolated 2D systems, which must necessarily have a well-defined FS. A WSM in a slab geometry, however, *is* an isolated 2D system and indeed, FAs on opposite surfaces together do form a well-defined 2D FS. A natural question to therefore ask is, “what signatures does this unusual Fermi surface have in scanning tunneling microscopy (STM) and angle-resolved photoemission spectroscopy (ARPES) – two common techniques that can probe surface states directly?”
In this work, we answer this question by computing the local density of states (LDOS) on the surface of WSM, $\rho_{\mbox{surf}}(\boldsymbol{r},E)$, in the presence of a point scatterer on the surface as well as the surface spectral function for a clean system, $A_{\mbox{surf}}^{0}(\boldsymbol{k},E)$. We apply our results to the iridates, A$_{2}$Ir$_{2}$O$_{7}$, A$=$Y, Eu, which are predicted to be WSMs [@PyrochloreWeyl; @KrempaWeyl]. Both $\rho_{\mbox{surf}}$ and $A_{\mbox{surf}}^{0}$ evolve as the sample thickness is increased, and the evolution is explained in terms of the amplitude of the FAs on the far surface diminishing on the near surface. The calculation is done using a prescription that can give FAs of arbitrary shape and connectivity and simultaneously generate the corresponding Weyl nodes in the bulk. The procedure, in a nutshell, entails stacking electron and hole FSs alternately, and gapping them out pairwise via interlayer couplings that are designed to leave the desired FAs on the end layers. The resulting Hamiltonian is of a simple tight-binding form, which allows us to calculate $A_{\mbox{surf}}^{0}$ analytically in the thermodynamic limit. This quantity can be directly measured by ARPES.
FAs appear in two qualitatively distinct ways: (a) either FAs on opposite surfaces overlap, resulting in a gapless semimetal, or, (b) FAs on opposite surface do not overlap, resulting in a 2D metal with a FS. The 2D particle density in this metal is proportional to the FS area according to Luttinger’s theorem, and lives predominantly on the surface. In general, however, some particles will leak into the bulk, but the bulk filling will typically be $\mathcal{O}(1/L)$, where $L$ is the slab thickness, and will thus vanish in the thermodynamic limit: $L\to\infty$. In the model presented here, (a) results when equal numbers of electron and hole FSs are stacked while (b) is obtained when the total number of 2D FSs is odd. This is consistent with the statement made earlier that the FA structure depends strongly on the boundary conditions, since peeling off a single layer interchanges (a) and (b).
FAs, in principle, can be generated by: (i) starting with a bulk model with the desired number of Weyl nodes, (ii) discretizing it in real space in the finite direction, and (iii) applying suitable boundary conditions to obtain FAs of the desired structure. While this approach works in principle, it has several associated complications. For example, determining the boundary conditions that result in the desired connectivity of the FAs is non-trivial. For instance, a WSM with four Weyl nodes $W_{1,2}^{\chi}$ at momenta $\chi\boldsymbol{Q}_{1,2}$, where $\chi=\pm$ denotes the chirality of the Weyl node, has two pairs of FAs on any surface on which the projections of the Weyl points are distinct. These FAs can pair up the Weyl points in two qualitatively different ways: as $(W_{1}^{+}W_{1}^{-})$ and $(W_{2}^{+}W_{2}^{-})$ on each surface, which is an (a)-type connectivity, or as $(W_{1}^{+}W_{1}^{-})$ and $(W_{2}^{+}W_{2}^{-})$ on the top surface and as $(W_{1}^{+}W_{2}^{-})$ and $(W_{1}^{-}W_{2}^{+})$ on the bottom surface, which falls in the (b) category. However, there is currently no general prescription for determining the boundary conditions that give one or the other connectivity. Moreover, to our knowledge there is also no general prescription for deriving lattice models with arbitrary numbers and locations of Weyl nodes. These gaps in working methods are filled by our top-down approach for generating FAs directly. Our approach should be useful to model FAs in real systems, where surface effects can bend the FAs and change their connectivity unpredictably.
**Layering prescription:** We describe the prescription by considering the simplest WSM, which has just two Weyl nodes, at $(k_{x},k_{y},k_{z})\equiv(\boldsymbol{k},k_{z})=(\boldsymbol{K}_{1,2},0)$ and the FAs connect $\boldsymbol{K}_{1}$ and $\boldsymbol{K}_{2}$ along a segment $S$ ($S'$) on the $z=1$ ($z=L$) surface in the surface Brillouin zone, as shown in Fig \[fig:layering picture\]. $S=S'$ and $S\neq S'$ correspond to the two qualitatively different situations (a) and (b) mentioned earlier, and will be obtained by distinct boundary conditions. Generalization to more Weyl points and Weyl points away from the $k_{z}=0$ plane is straightforward.
![Above: Layering prescription for obtaining FAs of arbitrary shape. Dotted (dashed) ellipses represent electron (hole) FSs, and solid red segments are the residual FAs on adding interlayer hoppings $t_{\boldsymbol{k}}$ and $\Delta_{\boldsymbol{k}}$. The horizontal black dashed line on the topmost layer separates regions with $\Delta_{\boldsymbol{k}}>t_{\boldsymbol{k}}$ and $\Delta_{\boldsymbol{k}}<t_{\boldsymbol{k}}$. An even (odd) number of total layers gives identical (non-identical) FAs on the two surfaces, as shown on the left (right). Below: 1D systems at fixed $\boldsymbol{k}\in C$ under the influence of $\Delta$ and $t$ in the extreme cases where the smaller hopping vanishes, for even (left) and odd (right) $L$. Filled (empty) circles denote a state on an electron (a hole) Fermi surface in the limit of decoupled layers. The ellipses enclose the states which get mutually gapped out by the hoppings.\[fig:layering picture\]](evenlayers "fig:"){width="0.5\columnwidth"}![Above: Layering prescription for obtaining FAs of arbitrary shape. Dotted (dashed) ellipses represent electron (hole) FSs, and solid red segments are the residual FAs on adding interlayer hoppings $t_{\boldsymbol{k}}$ and $\Delta_{\boldsymbol{k}}$. The horizontal black dashed line on the topmost layer separates regions with $\Delta_{\boldsymbol{k}}>t_{\boldsymbol{k}}$ and $\Delta_{\boldsymbol{k}}<t_{\boldsymbol{k}}$. An even (odd) number of total layers gives identical (non-identical) FAs on the two surfaces, as shown on the left (right). Below: 1D systems at fixed $\boldsymbol{k}\in C$ under the influence of $\Delta$ and $t$ in the extreme cases where the smaller hopping vanishes, for even (left) and odd (right) $L$. Filled (empty) circles denote a state on an electron (a hole) Fermi surface in the limit of decoupled layers. The ellipses enclose the states which get mutually gapped out by the hoppings.\[fig:layering picture\]](oddlayers "fig:"){width="0.5\columnwidth"}
------------------------------------------------------------------------
![Above: Layering prescription for obtaining FAs of arbitrary shape. Dotted (dashed) ellipses represent electron (hole) FSs, and solid red segments are the residual FAs on adding interlayer hoppings $t_{\boldsymbol{k}}$ and $\Delta_{\boldsymbol{k}}$. The horizontal black dashed line on the topmost layer separates regions with $\Delta_{\boldsymbol{k}}>t_{\boldsymbol{k}}$ and $\Delta_{\boldsymbol{k}}<t_{\boldsymbol{k}}$. An even (odd) number of total layers gives identical (non-identical) FAs on the two surfaces, as shown on the left (right). Below: 1D systems at fixed $\boldsymbol{k}\in C$ under the influence of $\Delta$ and $t$ in the extreme cases where the smaller hopping vanishes, for even (left) and odd (right) $L$. Filled (empty) circles denote a state on an electron (a hole) Fermi surface in the limit of decoupled layers. The ellipses enclose the states which get mutually gapped out by the hoppings.\[fig:layering picture\]](chainhoppingeven "fig:"){height="1.5cm"} ![Above: Layering prescription for obtaining FAs of arbitrary shape. Dotted (dashed) ellipses represent electron (hole) FSs, and solid red segments are the residual FAs on adding interlayer hoppings $t_{\boldsymbol{k}}$ and $\Delta_{\boldsymbol{k}}$. The horizontal black dashed line on the topmost layer separates regions with $\Delta_{\boldsymbol{k}}>t_{\boldsymbol{k}}$ and $\Delta_{\boldsymbol{k}}<t_{\boldsymbol{k}}$. An even (odd) number of total layers gives identical (non-identical) FAs on the two surfaces, as shown on the left (right). Below: 1D systems at fixed $\boldsymbol{k}\in C$ under the influence of $\Delta$ and $t$ in the extreme cases where the smaller hopping vanishes, for even (left) and odd (right) $L$. Filled (empty) circles denote a state on an electron (a hole) Fermi surface in the limit of decoupled layers. The ellipses enclose the states which get mutually gapped out by the hoppings.\[fig:layering picture\]](chainhoppingodd "fig:"){height="1.5cm"}
We claim that this WSM is generated by the following Bloch Hamiltonian: $$\begin{aligned}
H_{\boldsymbol{k}} & = & \sum_{z=1}^{L}\psi_{z,\boldsymbol{k}}^{\dagger}(-1)^{z}\mathcal{E}_{\boldsymbol{k}}\psi_{z,\boldsymbol{k}}\label{eq:main H}\\
& & +\sum_{z=1}^{L-1}\psi_{z,\boldsymbol{k}}^{\dagger}h_{z,\boldsymbol{k}}\psi_{z+1,\boldsymbol{k}}+\mbox{h.c.}\nonumber \end{aligned}$$ where even (odd) $L$ generates $S=S'$ ($S\neq S'$), $\mathcal{E}_{\boldsymbol{k}}$ is a phenomenological function that vanishes along a contour $C$ given by $$C\begin{cases}
\supset S & S=S'\\
=S\cup S' & S\neq S'
\end{cases}\label{eq:contour}$$ and the interlayer coupling $h_{z,\boldsymbol{k}}=-t_{\boldsymbol{k}}(\Delta_{\boldsymbol{k}})$ if $z$ is even (odd). If $S=S'$, $C$ can be chosen arbitrarily as long as it contains the entire segment $S$. The functions $t_{\boldsymbol{k}}$ and $\Delta_{\boldsymbol{k}}$ are real, non-negative phenomenological functions that satisfy: $$t_{\boldsymbol{k}}\begin{cases}
>\Delta_{\boldsymbol{k}} & \boldsymbol{k}\in S\\
<\Delta_{\boldsymbol{k}} & \boldsymbol{k}\in C/S(=S'\mbox{ if }C=S\cup S')
\end{cases}\label{eq:tDelta condition}$$ (\[eq:tDelta condition\]) dictates that $t_{\boldsymbol{k}}=\Delta_{\boldsymbol{k}}$ exactly at $\boldsymbol{k}=\boldsymbol{K}_{1,2}$. The $\boldsymbol{k}$-dependence of $t$ and $\Delta$ away from $C$ is unimportant for our purposes, and will be assumed to be negligible henceforth. We now justify the above claim.
*Bulk:* ****** If the interlayer couplings $t_{\boldsymbol{k}}=\Delta_{\boldsymbol{k}}=0$, then $\mathbb{H}=\sum_{\boldsymbol{k}}H_{\boldsymbol{k}}$ describes a stack of alternate *non-degenerate* electron and hole FSs. When $t_{\boldsymbol{k}}$ and $\Delta_{\boldsymbol{k}}$ are turned on, these FSs get gapped out in pairs in the bulk. Indeed, the bulk Hamiltonian is $$H_{\boldsymbol{k},k_{z}}^{\mbox{bulk}}=\mathcal{E}_{\boldsymbol{k}}\sigma_{z}+\left(\Delta_{\boldsymbol{k}}-t_{\boldsymbol{k}}\cos k_{z}\right)\sigma_{x}+t_{\boldsymbol{k}}\sin k_{z}\sigma_{y}$$ where $z$ is the layering direction, and is gapped everywhere except at the desired Weyl points: $(\boldsymbol{k},k_{z})=\left(\boldsymbol{K}_{1,2},0\right)$, due to (\[eq:tDelta condition\]). Allowing $t_{\boldsymbol{k}}$ and $\Delta_{\boldsymbol{k}}$ to be negative or complex simply moves the Weyl points off the $k_{z}=0$ plane, but this does not affect the shape of the FAs. Near the gapless points, $H^{\mbox{bulk}}$ realizes the Weyl Hamiltonian: $$\begin{aligned}
& H_{\boldsymbol{K}_{i}+\boldsymbol{p},0+p_{z}}^{\mbox{bulk}}\nonumber \\
& \,\,\,\approx\left[\boldsymbol{p}\cdotp\boldsymbol{\nabla_{k}}\mathcal{E}_{\boldsymbol{K}_{i}}\right]\sigma_{z}+\left[\boldsymbol{p}\cdot\boldsymbol{\nabla_{k}}\left(\Delta_{\boldsymbol{K}_{i}}-t_{\boldsymbol{K}_{i}}\right)\right]\sigma_{x}+\left[t_{\boldsymbol{K}_{i}}p_{z}\right]\sigma_{y}\nonumber \\
& \,\,\,\equiv p_{\perp}v_{F}(\boldsymbol{K}_{i})\sigma_{z}+p_{\parallel}v_{i}\sigma_{x}+\Delta_{0}p_{z}\sigma_{y}\end{aligned}$$ where $p_{\perp}=\boldsymbol{p}\cdot\hat{\boldsymbol{e}}_{r}(\boldsymbol{K}_{i})$ and $p_{\parallel}=\boldsymbol{p}\cdot\hat{\boldsymbol{e}}_{t}(\boldsymbol{K}_{i})$ are momenta perpendicular and parallel to $C$ ($\hat{e}_{r}(\boldsymbol{k})$ and $\hat{\boldsymbol{e}}_{t}(\boldsymbol{k})$ are 2D unit vectors normal and tangential to $C$), $v_{F}$ is the Fermi velocity of the 2D FSs and $v_{i}=\boldsymbol{\nabla_{k}}\left(\Delta_{\boldsymbol{K}_{i}}-t_{\boldsymbol{K}_{i}}\right)\cdot\hat{e_{t}}(\boldsymbol{K}_{i})$. In going to the second line, the variation of $t_{\boldsymbol{k}}$ and $\Delta_{\boldsymbol{k}}$ perpendicular to $C$ has been assumed to be negligible, since it does not affect the shape of the FAs. $v_{i}$ has opposite signs at $\boldsymbol{K}_{1}$ and $\boldsymbol{K}_{2}$, ensuring that the Weyl nodes have opposite chirality. $H_{\mbox{bulk}}$ is obviously unaffected by the boundary conditions at $z=1$ and $z=L$ for large $L$.
*Surface:* The surface, however, strongly depends on the boundary conditions; in particular, it is qualitatively different for odd and even $L$. If $L$ is odd, at each $\boldsymbol{k}\in C$, a state remains unpaired and hence, gapless, at $z=1$ ($z=L$) whenever $\Delta_{\boldsymbol{k}}<t_{\boldsymbol{k}}$ ($\Delta_{\boldsymbol{k}}>t_{\boldsymbol{k}}$). The gapless states at $z=1$ ($z=L$) thus, trace out $S$ ($S'$). On the other hand, when $L$ is even, both ends of a the chain at fixed $\boldsymbol{k}$ carry a gapless state when $\Delta_{\boldsymbol{k}}<t_{\boldsymbol{k}}$ and neither end has gapless states when $\Delta_{\boldsymbol{k}}>t_{\boldsymbol{k}}$. In this case, the gapless states on both surfaces of the slab trace out $S$.
Viewed differently, the 1D system at fixed $\boldsymbol{k}\in C$ and $\left|\Delta\right|\neq\left|t\right|$ is an insulator in the CII symmetry class, which is known to have a $\mathbb{Z}$ topological classification [@SFRLClassification; @KitaevClassification]. While $\left|\Delta\right|>\left|t\right|$ gives a trivial phase, $\left|\Delta\right|<\left|t\right|$ is topologically non-trivial with a zero mode at each end protected by a chiral symmetry, if the 1D lattice has a whole number of unit cells. These end states are nothing but the FA states at that $\boldsymbol{k}$ when $L$ is even. As $\boldsymbol{k}$ is varied along $C$, the 1D system undergoes a topological phase transition at $\boldsymbol{K}_{1}$ and $\boldsymbol{K}_{2}$. For odd $L$, there is always a state at one end of the chain, as show in Fig \[fig:layering picture\]. This prescription is similar in spirit to node-pairing picture of Ref [@HosurRyuChiralTISC] for chiral topological insulators in three dimensions. Note that it is necessary to start with two sets of FSs, since a single FS cannot be destroyed perturbatively.
*Symmetry analysis:* WSMs can only exist in systems in which at least one symmetry out of time-reversal symmetry ($\mathcal{T}$) and inversion symmetry ($\mathcal{I}$) is broken; the presence of both would make each band doubly degenerate and give Dirac semi-metals instead with four-component fermions near at each node. Moreover, $\mathcal{I}$-symmetric, $\mathcal{T}$-breaking ($\mathcal{T}$-symmetric, $\mathcal{I}$-breaking) WSMs have an odd (even) number of pairs of Weyl nodes. In the current picture, the breaking of these symmetries can be understood as follows. Let us assume $\mathcal{E}_{\boldsymbol{k}}=\mathcal{E}_{-\boldsymbol{k}}$; if this weren’t true, both symmetries would broken from the outset. In general, $t_{\boldsymbol{k}}$ and $\Delta_{\boldsymbol{k}}$ are unrelated to $t_{-\boldsymbol{k}}$ and $\Delta_{-\boldsymbol{k}}$, in which case both symmetries would again be broken. However, $\mathcal{T}$ is preserved if $t_{\boldsymbol{k}}=t_{-\boldsymbol{k}}$ and $\Delta_{\boldsymbol{k}}=\Delta_{-\boldsymbol{k}}$, which can only happen if the number of points on $C$ at which $t_{\boldsymbol{k}}=\Delta_{\boldsymbol{k}}$, and thus the number of Weyl nodes, is an integer multiple of four. On the other hand, inversion about a particular layer interchanges $t$ and $\Delta$; thus, $t_{\boldsymbol{k}}=\Delta_{-\boldsymbol{k}}$ preserves this inversion symmetry. This condition requires $t_{\boldsymbol{k}}-\Delta_{\boldsymbol{k}}$ to change sign twice an odd number of times along $C$, giving an odd number of Weyl node pairs.
![(a) FAs in the surface Brillouin zone on the 111 surface of iridate WSMs. Solid (dashed) lines denote FAs on the near (far) surface. Filled (empty) circles denote projections of Weyl nodes of positive (negative) chirality in all the figures above. (b and c) Surface LDOS in arbitrary units due to the six FAs near the $L$ point in the presence of a point scatterer on the surface (brown cross) for a thin sample (b) and a thick sample (c). Insets show the numerically computed surface spectral function at $E=0$ for the clean system, with darker colors representing larger values. The computation is done for the model, described in the text, which generates the six FAs near the $L$-point but not the remaining eighteen FAs near the Brillouin zone edges.\[fig:iridate STM\]](iridateFA "fig:"){width="0.16\columnwidth"}![(a) FAs in the surface Brillouin zone on the 111 surface of iridate WSMs. Solid (dashed) lines denote FAs on the near (far) surface. Filled (empty) circles denote projections of Weyl nodes of positive (negative) chirality in all the figures above. (b and c) Surface LDOS in arbitrary units due to the six FAs near the $L$ point in the presence of a point scatterer on the surface (brown cross) for a thin sample (b) and a thick sample (c). Insets show the numerically computed surface spectral function at $E=0$ for the clean system, with darker colors representing larger values. The computation is done for the model, described in the text, which generates the six FAs near the $L$-point but not the remaining eighteen FAs near the Brillouin zone edges.\[fig:iridate STM\]](L3circular "fig:"){width="0.42\columnwidth"}![(a) FAs in the surface Brillouin zone on the 111 surface of iridate WSMs. Solid (dashed) lines denote FAs on the near (far) surface. Filled (empty) circles denote projections of Weyl nodes of positive (negative) chirality in all the figures above. (b and c) Surface LDOS in arbitrary units due to the six FAs near the $L$ point in the presence of a point scatterer on the surface (brown cross) for a thin sample (b) and a thick sample (c). Insets show the numerically computed surface spectral function at $E=0$ for the clean system, with darker colors representing larger values. The computation is done for the model, described in the text, which generates the six FAs near the $L$-point but not the remaining eighteen FAs near the Brillouin zone edges.\[fig:iridate STM\]](L23circular "fig:"){width="0.42\columnwidth"}
(a) (b) (c)
**LDOS results:** Having described the procedure for obtaining FAs from a 2D limit, we demonstrate its utility by calculating the surface spectral function for a clean system and the surface LDOS in the presence of a point surface scatterer within a model that should be relevant to the pyrochlore iridates A$_{2}$Ir$_{2}$O$_{7}$, A$=$Y, Eu, which are purported WSMs with 24 Weyl nodes [@PyrochloreWeyl]. The lattice in the WSM phase has inversion symmetry as well as a threefold rotation symmetry $R_{3}$ about the cubic 111-axis, and there are six Weyl nodes related by these symmetries near each of the four $L$ points in the FCC Brillouin zone. Additionally, the lattice also has a $D_{6}$ symmetry, i.e., $\pi/3$ rotation about [\[]{}111[\]]{} followed by reflection in the perpendicular plane. This symmetry has an important implication for the FAs: if it is preserved in a slab geometry, then the FAs will be as shown in Fig \[fig:iridate STM\] (a). In particular, the six FAs near the center of the surface Brillouin zone enclose an area, while the remaining eighteen overlap in pairs on the opposite surfaces. We note that Ref [@KrempaWeyl] also predicted Weyl semimetallic behavior in the above iridates, but with only 8 Weyl nodes. In this case, the hexagonal figure around the $L$ point would collapse to a point. As we argue below, LDOS oscillations on the surface stem predominantly from the hexagonal figure; hence, the proposal of Ref [@KrempaWeyl], if true, would imply no strong LDOS oscillations.
We compute the surface LDOS for a model that has six FAs like the ones around the $L$ point, as a function of the the sample thickness. The remaining eighteen FAs in the iridates are expected to be destroyed by finite size effects for thin samples, while for thick samples backscattering occurs across the Brillouin zone and hence can give rise to LDOS oscillations only on the lattice scale. These oscillations are unlikely to be distinguishable from the electron density variations on this scale already present.
The LDOS is calculated via the standard $T$-matrix formalism. Given the time-ordered Green’s function for the clean system: $G_{0}(\boldsymbol{k},E)=\left(E-H_{\boldsymbol{k}}\right)^{-1}$ and a scattering potential: $U_{z,z'}(x,y)=u\delta(x)\delta(y)\delta_{z,1}\delta_{z',1}$, the $T$-matrix is given by $T(\omega)=\left(1-U\sum_{\boldsymbol{k}}G_{0}(\boldsymbol{k},E)\right)^{-1}U$, independent of momentum, since the scattering potential in momentum independent. Here, $G_{0}$, $U$ and $T$ are all $L\times L$ matrices indexed by $z$. The full Green’s function in the presence of the impurity is $G(\boldsymbol{k},\boldsymbol{k}',E)=\delta_{\boldsymbol{k},\boldsymbol{k}'}G_{0}(\boldsymbol{k},E)+G_{0}(\boldsymbol{k},E)T(\omega)G_{0}(\boldsymbol{k}',E)$, and the LDOS on the $z=1$ surface is related to the $(1,1)$ element of its retarded cousin: $\rho_{\mbox{surf}}(\boldsymbol{r},E)=-\frac{1}{\pi}\mathrm{Im}G^{11}(\boldsymbol{r},\boldsymbol{r},E+i\delta)$, where $G(\boldsymbol{r},\boldsymbol{r}',E)=\int_{\boldsymbol{k},\boldsymbol{k}'}e^{i(\boldsymbol{k}\cdot\boldsymbol{r}-\boldsymbol{k}'\cdot\boldsymbol{r}')}G(\boldsymbol{k},\boldsymbol{k}',E)$. For the calculation, we use $\mathcal{E}(\boldsymbol{k})=\sqrt{k^{2}+2k^{6}\cos^{2}3\theta_{\boldsymbol{k}}}-1$ to generate the hexagonal figure and $t_{\boldsymbol{k}}\equiv t_{0}=0.5$, $\Delta_{\boldsymbol{k}}=t_{0}(1-\cos3\theta_{\boldsymbol{k}})$ to obtain the FAs. Here, $(k,\theta_{\boldsymbol{k}})$ the polar coordinates of $\boldsymbol{k}$.
The results are presented in Fig \[fig:iridate STM\] (b) and (c) for $E=0$. For thin samples, clear LDOS oscillations are seen in the horizontal direction as well as along the two equivalent directions related by $\pi/3$ rotation. The origin of these oscillations becomes clear if one looks at $A_{\mbox{surf}}^{0}(\boldsymbol{k},E=0)$, displayed inset. Since the sample is thin, FA wavefunctions from the far surface have significant amplitude on the near surface, which allows backscattering to occur diametrically across the hexagon. The dominant backscattering processes are the ones involving the midpoints of the FAs, since the Fermi surface is nested here. On the other hand, as the sample thickness is increased, three of the six FAs retreat to the far surface and backscattering is exponentially suppressed. The result is small variations in the LDOS arising from scattering between FAs solely on the top surface. Thus, the STM map has a distinct evolution with sample thickness which is characteristic of the FA structure in the iridate WSMs.
**Surface spectral function in clean thermodynamic limit:** $A_{\mbox{surf}}^{0}(\boldsymbol{k},E)$ was computed numerically in order to generate the insets of Fig \[fig:iridate STM\]. In the thermodynamic limit, however, this calculation can be done analytically. Denoting the $(1,1)$ element of the clean Green’s function for a $L$-layer slab by $G_{0(L)}^{11}$, it is straightforward to show, by explicitly evaluating the $(1,1)$ cofactor of $E-H_{\boldsymbol{k}}$ and using $\mathrm{det}\left(\begin{array}{cc}
A & B\\
C & D
\end{array}\right)=\mathrm{det}(A)\mathrm{det}(D-CA^{-1}B)$, that $$\begin{aligned}
& (E-\mathcal{E}_{\boldsymbol{k}})G_{0(L)}^{11}(\boldsymbol{k},E)\nonumber \\
& =1+\frac{\Delta_{\boldsymbol{k}}^{2}}{E^{2}-\mathcal{E}_{\boldsymbol{k}}^{2}-\Delta_{\boldsymbol{k}}^{2}-t_{\boldsymbol{k}}^{2}(E-\mathcal{E}_{\boldsymbol{k}})G_{0(L-2)}^{11}(\boldsymbol{k},E)}\end{aligned}$$ In the thermodynamic limit: $L\to\infty$, $G_{0(L)}^{11}(\boldsymbol{k},E)\approx G_{0(L-2)}^{11}(\boldsymbol{k},E)\equiv g(\boldsymbol{k},E)$. Thus,
$$g(\boldsymbol{k},E)=\frac{1}{2t_{\boldsymbol{k}}^{2}(E-\mathcal{E}_{\boldsymbol{k}})}\left[\left(E^{2}-\mathcal{E}_{\boldsymbol{k}}^{2}+t_{\boldsymbol{k}}^{2}-\Delta_{\boldsymbol{k}}^{2}\right)\pm\sqrt{\left(E^{2}-\mathcal{E}_{\boldsymbol{k}}^{2}+t_{\boldsymbol{k}}^{2}-\Delta_{\boldsymbol{k}}^{2}\right)^{2}-4t_{\boldsymbol{k}}^{2}(E^{2}-\mathcal{E}_{\boldsymbol{k}}^{2})}\right]\label{eq:gtherm}$$
The physical condition $A_{\mbox{surf}}^{0}(\boldsymbol{k},E)=-\frac{1}{\pi}\mathrm{Im}\left[g(\boldsymbol{k},E+i\delta)\right]\ge0$ fixes the sign in front of the square root. Clearly, $$\begin{gathered}
A_{\mbox{surf}}^{0}(\boldsymbol{k},E)=\delta(E-\mathcal{E}_{\boldsymbol{k}})\frac{t_{\boldsymbol{k}}^{2}-\Delta_{\boldsymbol{k}}^{2}+\left|t_{\boldsymbol{k}}^{2}-\Delta_{\boldsymbol{k}}^{2}\right|}{2t_{\boldsymbol{k}}^{2}}+\nonumber \\
\frac{1}{2t_{\boldsymbol{k}}^{2}(E-\mathcal{E}_{\boldsymbol{k}})}\mathrm{Im}\sqrt{\left(E^{2}-\mathcal{E}_{\boldsymbol{k}}^{2}+t_{\boldsymbol{k}}^{2}-\Delta_{\boldsymbol{k}}^{2}\right)^{2}-4t_{\boldsymbol{k}}^{2}(E^{2}-\mathcal{E}_{\boldsymbol{k}}^{2})}\label{eq:asurf}\end{gathered}$$ The first line is non-zero only when $t_{\boldsymbol{k}}^{2}>\Delta_{\boldsymbol{k}}^{2}$ and has a sharp peak at $E=\mathcal{E}_{\boldsymbol{k}}$. Clearly, this represents the contribution to $A_{\mbox{surf}}^{0}$ from the FA. Whereas, the second line is non-vanishing when $|E|>|\mathcal{E}_{\boldsymbol{k}}|$ and $\left|t_{\boldsymbol{k}}-\sqrt{E^{2}-\mathcal{E}_{\boldsymbol{k}}^{2}}\right|<|\Delta_{\boldsymbol{k}}|<\left|t_{\boldsymbol{k}}+\sqrt{E^{2}-\mathcal{E}_{\boldsymbol{k}}^{2}}\right|$. These inequalities are satisfied in the region near the projection of the Weyl points onto the surface. Moreover, this contribution to $A_{\mbox{surf}}^{0}$ has no delta-function peak. Thus, it represents contributions form the bulk states near the Weyl nodes. The quantity $A_{\mbox{surf}}^{0}$ can be directly measured by ARPES experiments.
In conclusion, we have studied impurity-induced Friedel oscillations due to FAs in WSMs, focusing on the FA structure of the purported iridate WSMs, and observed their dependence on sample thickness. For thin samples, FAs on both surfaces collude to allow nested backscattering and hence produce strong LDOS oscillations, whereas for thick samples, the FAs on the far surface do not reach the near surface and such backscattering and the consequent LDOS oscillations are suppressed. The calculation is done by building the desired WSM and FA structure by stacking electron and hole Fermi surfaces and adding suitable interlayer hopping. Within this prescription, the surface spectral function for a clean system can be calculated analytically in the thermodynamic limit.
P.H. would like to thank Ashvin Vishwanath, Tarun Grover, Siddharth Parameswaran, Xiaoliang Qi and Shou-Cheng Zhang for insightful discussions, Ashvin Vishwanath and Tarun Grover for useful comments on the manuscript, and DOE grant DE-AC02-05CH11231 for financial support during most of this work. During the final stages of writing this manuscript, P.H. was supported by Packard Fellowship grant 1149927-100-UAQII.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
####
A system comprising a $\Lambda$-type or V-type atom interacting with two radiation fields exhibits, during its dynamical evolution, interesting optical phenomena such as electromagnetically-induced transparency (EIT) and a variety of nonclassical effects. Signatures of the latter are seen in the entanglement dynamics of the atomic subsystem and in appropriate field observables. Some of these effects have been experimentally detected, and have even been used to change the nonlinear optical properties of certain atomic media. It is therefore useful to investigate the roles played by specific initial states of the radiation fields, detuning parameters, field nonlinearities and the nature of field-atom couplings on EIT and on the entanglement between subsystems. We investigate these aspects in the framework of a simple model that captures the salient features of such tripartite entangled systems. Entanglement dynamics is shown to be very sensitive to the intensity-dependent atom-field couplings. Unexpected interesting features pertaining to the collapses and revivals of the atomic subsystem von Neumann entropy appear. These features could, in principle, be useful in enabling entanglement.
address: 'Department of Physics, Indian Institute of Technology Madras, Chennai 600 036, India'
author:
- |
Pradip Laha, B. Sudarsan, S. Lakshmibala and\
V. Balakrishnan
title: Entanglement dynamics in a model tripartite quantum system
---
Introduction
============
####
Interacting quantum systems exhibit many interesting features during temporal evolution. These include diverse nonclassical effects such as quantum entanglement, revival phenomena [@robi; @aver1; @milburn; @kita], collapse of the measure of entanglement to a constant value over certain time intervals, and so on. Atom optics provides a convenient framework for examining these effects. In tripartite entangled systems comprising an atom interacting with two radiation fields, further effects can occur, such as electromagnetically-induced transparency (EIT): the appearance, under suitable conditions, of a transparency window within the absorption spectrum of the atomic system. Apart from the change in the transmission coefficient, atomic media can also exhibit interesting dispersive properties as a consequence of EIT. This feature has been exploited to create materials that demonstrate slow light-pulse propagation and enhanced nonlinear optical properties (for a review, see [@marangos]).
Extensive experimental investigations on EIT have been carried out on three-level atoms interacting with a probe field and a coupling field. Following the report on the occurrence of EIT in optically opaque Sr$^{+}$ vapour in 1991 [@boller], several detailed experiments have been performed on the nature of this optical phenomenon in various atom-field configurations (see, e.g., [@Li; @entin]). In the absence of the coupling field, the intensity of the probe field (equivalently, the corresponding mean photon number) will remain nearly constant over a small range of values of the detuning parameter about zero. The transparency window created by the coupling field in this absorption spectrum is signalled by the appearance of a peak in the probe field intensity. It is therefore reasonable to expect that, if the entangled tripartite system displays collapses and revivals of the probe intensity during time evolution, this peak should be seen at any instant lying in a time interval in which the mean photon number of the probe field collapses to a constant value in the absence of the coupling field.
In tripartite systems comprising a V-type or $\Lambda$-type atom and two radiation fields, the inclusion of Kerr-type nonlinearities in the field subsystem opens up the possibility of collapses and revivals of the mean photon number corresponding to either field, under specific conditions. EIT can therefore be investigated, for instance, at an instant in a time interval when the mean photon number collapses for the first time, independent of whether or not further collapses occur. On longer time scales the subsystem von Neumann entropy (SVNE) corresponding to the atom can collapse to a constant value over a sufficiently long time interval and this feature could possibly be mirrored in the temporal evolution of the mean photon number of a field subsystem. Further, the manner in which different initial field states, detuning parameters and interaction strengths affect both EIT and the entanglement dynamics on long time scales needs to be examined. Both $\Lambda$ and V atoms interacting with a probe field and a coupling field are good candidates for theoretical and experimental investigations of all the foregoing aspects.
The revival phenomenon is exhibited even by a single-mode radiation field propagating in a Kerr-like medium, governed by an effective Hamiltonian of the form ${a^{\dagger}}^{2} a^{2}$, where $a$ and $a^{\dagger}$ are the photon annihilation and creation operators[@milburn; @kita]. Fractional revivals occur at specific instants between two successive revivals of the initial wave packet, when the initial wave packet splits into two or more similar copies of itself [@robi]. Signatures of revivals and fractional revivals are captured by appropriate quadrature observables and their higher moments [@sudh1; @sudh2]. Collapses and revivals are of greater interest, however, in bipartite and multipartite systems that involve interaction between the subsystems, entailing nontrivial entanglement dynamics. A bipartite model[@puri] of a multi-level atom interacting with a single-mode radiation field predicts[@sudhjphysb] collapses and revivals of the state of the field (also reflected, in this case, in the temporal behaviour of the field SVNE), when the nonlinearity in the atomic medium is weak compared to the strength of atom-field interaction. For stronger nonlinearity, however, the revival phenomenon is absent; a detailed time-series analysis of the mean photon number reveals a gamut of ergodicity properties displayed by this observable, ranging from regular to chaotic behaviour[@sudhergodicity; @sudhrecurrence].
Entanglement dynamics in the foregoing bipartite model turns out to depend significantly on the degree to which the initial state of the field departs from perfect coherence. This has been deduced by selecting, as initial field states, the family of photon-added coherent states (PACS)[@tara] denoted by ${\left|{\alpha,m}\right\rangle}$, where $\alpha \in \mathbb{C}$ and $m = 0, 1, 2, \ldots\,$. The standard oscillator coherent state (CS) $${\left|{\alpha}\right\rangle}=e^{-|\alpha|^2/2}\,\sum_{n=0}^{\infty}\,
\alpha^{n}\,{\left|{n}\right\rangle} /\sqrt{n!}
\label{eqn:cs}$$ corresponds to the case $m = 0$, while the $m$-photon-added coherent state ${\left|{\alpha,m}\right\rangle}$ is obtained by normalising the state $(a^{\dagger})^{m}{\left|{\alpha}\right\rangle}$ to unity. The set $\{{\left|{\alpha,m}\right\rangle}\}$ provides a family of states whose departure from coherence is precisely quantifiable. Experimental realisation of the single photon-added coherent state using quantum state tomography [@zavatta] has added to the relevance of these studies.
If the atom has only a very small number of energy levels, the dynamics of entanglement differs both qualitatively and quantitatively from that of the bipartite model discussed above. This aspect has been studied[@athreya1; @athreya] in the framework of a three-level V or $\Lambda$ atom interacting with a single-mode radiation field. A striking feature is that, relatively independent of the degree of coherence of the initial state of the radiation field, the photon number statistics and the degree of entanglement are affected strongly by the low dimensionality of the atomic Hilbert space. This feature continues to hold good for tripartite extensions of the model in which the atom interacts with two radiation fields.
This tripartite model is most suitable for our present purposes, as it provides a convenient framework for examining EIT, both on short time scales when the mean photon number records its first collapse, and on longer time scales where the measure of entanglement between the atom and fields remains constant over a significant time interval [@athreya1], for appropriate choices of the field state and parameter values. Another important aspect we study is the role played by intensity-dependent couplings in entanglement dynamics. It has been found [@buck] that, for an intensity-dependent coupling of the form $(a^{\dagger}a)^{1/2}$ between an initial CS and a Jaynes-Cummings atom, the mean photon number can be evaluated in closed form, and the mean photon energy undergoes periodic collapses and revivals in this case. Subsequent studies have been carried out on the dynamics of the Jaynes-Cummings atom interacting through an intensity-dependent coupling with other initial states of the radiation field such as the squeezed vacuum and the $SU(1,1)$ coherent state [@buzek1; @buzek2]. A coupling proportional to $1/(a^{\dagger}a)^{1/2}$ has also been used[@zait] in a tripartite model, motivated by the fact that this form arises naturally in the context of diagonal-state representations of the density matrix in a restricted Hilbert space where the zero-photon state is absent[@sudarshan]. An intensity-dependent coupling of the form $(1 + \kappa\,a^{\dagger}\,a)^{1/2}, \; 0 \leq \kappa \leq 1$ has been shown[@siva1] to lead to a closed-form expression for the mean photon energy. Here $\kappa = 0$ reduces to the Heisenberg-Weyl algebra for the field operators, while $\kappa = 1$ leads to the $SU(1,1)$ algebra for (nonlinear combinations of) these operators[@siva2]. Intermediate values of $\kappa$ corresponds to a deformed $SU(1,1)$ operator algebra. While the revival phenomenon has been examined in several models with intensity-dependent couplings, including the tripartite system of an atom interacting with two radiation fields [@tav1; @tav2], the effect of such a coupling on the entanglement dynamics and on EIT has not been investigated. In this paper, we report on the effects of a general intensity-dependent coupling of the form $(1 + \kappa \:{a^{\dagger}}a)^{1/2}$ on these two phenomena in the tripartite model of a $\Lambda$ atom interacting with two radiation modes. We have also investigated the corresponding case of a V atom. We do not present those results, as they are not qualitatively different in any significant manner from those for the $\Lambda$ atom.
$\Lambda$ atom interacting with two radiation modes {#model}
===================================================
####
The tripartite model has two radiation fields: a ‘probe field’ $F_{1}$ and a ‘coupling field’ $F_{2}$, of respective frequencies $\Omega_{1}$ and $\Omega_{2}$, with annihilation and creation operators $a_{i}$ and $a^{\dagger}_{i}\,(i = 1,2)$. The highest energy state of the $\Lambda$ atom is denoted by ${\left|{3}\right\rangle}$, and ${\left|{1}\right\rangle}$ and ${\left|{2}\right\rangle}$ are the two lower energy states. $F_{1}$ and $F_{2}$ induce, respectively, the ${\left|{1}\right\rangle}\leftrightarrow {\left|{3}\right\rangle}$ and ${\left|{2}\right\rangle}\leftrightarrow {\left|{3}\right\rangle}$ transitions, while the transition ${\left|{1}\right\rangle}\leftrightarrow {\left|{2}\right\rangle}$ is dipole-forbidden. The general Hamiltonian that incorporates field nonlinearities and intensity-dependent couplings is given (setting $\hbar = 1)$ by $$\begin{aligned}
\fl {H} = \sum\limits_{j=1}^{3} \omega_{j} \sigma_{jj} + \Omega_{1} \,a_{1}^{\dagger} a_{1} + \chi_{1} \,a_{1}^{\dagger 2} a_{1}^{2} + \Omega_{2} \,a_{2}^{\dagger} a_{2} + \chi_{2} \,a_{2}^{\dagger 2} a_{2}^{2} \nonumber \\
+ \lambda_{1} ( R_{1} \,\sigma_{31} + R^{\dagger}_{1} \,\sigma_{13} ) + \lambda_{2} ( R_{2} \,\sigma_{32} + R^{\dagger}_{2}\,\sigma_{23} ).
\label{eqn:two_mode_lambda_hamiltonian}\end{aligned}$$ Here, $\sigma_{ij} = {\left|{i}\right\rangle}{\left\langle{j}\right|}$ where ${\left|{j}\right\rangle}$ is an atomic state, $\{\omega_{j}\}$ are positive constants, $\chi_{i}$ represents the strength of the nonlinearity in the field $F_{i}$, and $\lambda_{1}, \lambda_{2}$ are the respective atom-field coupling strengths corresponding to the ${\left|{3}\right\rangle}\leftrightarrow {\left|{1}\right\rangle}$ and ${\left|{3}\right\rangle}\leftrightarrow {\left|{2}\right\rangle}$ transitions. Further, $$\fl R_{i} = a_{i} \,f(N_{i}),
\label{defnofR}$$ where $f(N_{i})$ is a real-valued function of $N_{i} = a^{\dagger}_{i} a_{i}$ that serves to incorporate a possible intensity-dependent coupling. As stated in the Introduction, we consider the functional form $f(N_{i} )= (1 + \kappa_{i} N_{i})^{1/2}$, where $\kappa_{i}$ takes values in the range $[0,1]$. Wherever relevant, a comparison will be made with results pertinent to the cases $f(N_{i}) = 1$ and $f(N_{i}) = N_{i}^{1/2}$. $H$ can be written as the sum $H_{0} + H_{1}$, where $$\begin{aligned}
\fl {H}_{0} = \sum\limits_{i=1}^{2} \Omega_{i} N_{i}^{\rm tot} + \omega_{3} I,
\label{eqn:two_mode_lambda_h0}\end{aligned}$$ $$\begin{aligned}
\fl {H}_{1} = \sum\limits_{i=1}^{2} \chi_{i} a_{i}^{\dagger 2} a_{i}^{2} - \Delta_{i} \sigma_{ii} + \lambda_{i} ( R_{i} \sigma_{3i} + R^{\dagger}_{i} \sigma_{i3} ).
\label{eqn:two_mode_lambda_h1}\end{aligned}$$ Here $I = \sum_{j=1}^{3} \sigma_{jj}$, $N_{i}^{\rm tot} = a_{i}^{\dagger} a_{i} - \sigma_{ii} \;(i=1,2)$ are constants of the motion, and the two detuning parameters $\Delta_{i}$ are given by $$\Delta_{i} = \omega_{3} - \omega_{i} - \Omega_{i}.
\label{eqn:two_mode_v_delta1}$$ $H_{0}$ merely introduces a phase factor in the time evolution of the state. It is therefore convenient to work in an appropriate interaction picture to eliminate this trivial dependence.
Earlier work [@tav2] pertaining to a $\Lambda$ atom interacting with two radiation fields involved an intensity-dependent coupling specified by $f(N_{i}) =
N_{i}^{1/2}$, and a cross-Kerr nonlinearity between the fields, but not individual field nonlinearities as in the present instance. Entanglement collapse to a constant value over a significant time interval has not been reported hitherto, nor has EIT been examined in this case. The model we consider here provides a natural setting for examining the effects of individual field nonlinearities, detuning parameters, a family of intensity-dependent couplings and specific classes of initial states on EIT and entanglement dynamics during temporal evolution.
The quantities we require for our study are the matrix elements of the time-dependent reduced density matrices of the tripartite system. We consider $F_{1}$ and $F_{2}$ to be initially in general superpositions of Fock states, $\sum_{0}^{\infty} q_{n} {\left|{n}\right\rangle}$ and $\sum_{0}^{\infty} r_{m} {\left|{m}\right\rangle}$, respectively. For definiteness, the initial state of the atom is taken to be ${\left|{1}\right\rangle}$ throughout. The initial state of the full system is thus $$\fl {\left|{\psi (0)}\right\rangle} = \sum\limits_{n=0}^{\infty} \sum\limits_{m=0}^{\infty} q_{n} r_{m}
{\left|{1; n; m}\right\rangle},
\label{eqn:two_mode_lambda_initial_state}$$ where ${\left|{1; n; m}\right\rangle}$ denotes a state with the atom in ${\left|{1}\right\rangle}$ and the fields $F_{1}$ and $F_{2}$ in states with $n$ and $m$ photons, respectively. (An analogous notation will be used for other field states as well, such as coherent and photon-added coherent states.) Proceeding on lines similar to those for the V atom[@athreya], the interaction picture state vector at time $t$ is found to be $$\begin{aligned}
\fl {\left|{\psi (t)}\right\rangle}_{\rm int} = \sum\limits_{n=0}^{\infty}\sum\limits_{m=0}^{\infty} q_{n} r_{m} \biggl\{ A_{nm}(t) e^{i \Delta_{1}t} {\left|{1; n; m}\right\rangle} \nonumber \\
+ \,B_{nm}(t) e^{i \Delta_{2}t} {\left|{2; n-1; m+1}\right\rangle} \nonumber \\
+ \,C_{nm}(t) {\left|{3; n-1; m}\right\rangle} \biggr\},
\label{eqn:two_mode_lambda_interaction_state}\end{aligned}$$ where the functions $A_{nm}(t), B_{nm}(t), C_{nm}(t)$ are as follows. Since the atom is taken to be initially in the state ${\left|{1}\right\rangle}$, it cannot make a transition to ${\left|{3}\right\rangle}$ if $n = 0$. Hence $$\fl
A_{0m}(t) = 1, \;B_{0m}(t) =
C_{0m}(t) = 0
\label{eqn:two_mode_v_a_n0}$$ for all $m$. When $n, m \geq 1$, we find $$\begin{aligned}
\fl A_{nm}(t) = \frac{e^{i(\Delta_{2} - \Delta_{1}) t}} {f_{1}f_{2}} \sum\limits_{j=1}^{3}
b_{j} \bigg\{ (\Delta_{2} + \mu_{j} + V_{12} + V_{22}) \nonumber \\
+ (\mu_{j} + V_{12} + V_{21}) - f_{2}^{2} \bigg\} e^{i \mu_{j} t},
\label{eqn:two_mode_lambda_a}\end{aligned}$$ $$\fl B_{nm}(t) = \sum\limits_{j=1}^{3} b_{j} e^{i \mu_{j} t},
\label{eqn:two_mode_lambda_b}$$ $$\fl C_{nm}(t) = - \frac{e^{i \Delta_{2} t}} {f_{2}} \sum\limits_{j=1}^{3} b_{j} (\mu_{j} + V_{12} + V_{21}) e^{i \mu_{j} t},
\label{eqn:two_mode_lambda_c}$$ where $$\fl V_{11} = \chi_{1} n (n-1), \quad V_{12} = \chi_{1} (n-1)(n-2),
\label{eqn:two_mode_lambda_v11v12}$$ $$\fl V_{21} = \chi_{2} m(m+1), \quad V_{22} = \chi_{2} m (m-1),
\label{eqn:two_mode_lambda_v21v22}$$ $$\fl f_{1} = \lambda_{1} n^{1/2} f(n), \quad f_{2} = \lambda_{2}
(m+1)^{1/2} f(m+1).
\label{eqn:two_mode_lambda_f1f2}$$ The effect of the operator $f(N)$ in the Hamiltonian on the state of the system is captured in the quantities $f_{1}$ and $f_{2}$. Further, $$\fl \mu_{j} = - {\textstyle\frac{1}{3}}\,x_{1} +
{\textstyle\frac{2}{3}} \,(x_{1}^{2} - 3 x_{2})^{1/2}
\cos \,\Big\{\theta +
{\textstyle\frac{2}{3}} (j-1) \pi \Big\}
\label{eqn:mu_j}$$ where $j = 1, 2, 3$, and $$\fl \theta = {\textstyle\frac{1}{3}}
\cos^{-1} \Big\{ [9 x_{1} x_{2} - 2 x_{1}^{3} - 27 x_{3}]
\big/[2 ( x_{1}^{2} - 3 x_{2} )^{3/2}]
\Big\},
\label{eqn:theta}$$ $$\begin{aligned}
\fl x_{1} = V_{11} + 2 V_{12} + V_{21} + 2 V_{22} - \Delta_{1} + 2\Delta_{2},
\label{eqn:two_mode_lambda_x1}\end{aligned}$$ $$\begin{aligned}
\fl x_{2} = (V_{12} + V_{21} + \Delta_{2}) (V_{11} + V_{12} + 2 V_{22} - \Delta_{1}) \nonumber \\
+ (V_{12} + V_{22}) (V_{11} + V_{22} - \Delta_{1}) + 2\Delta_{2}(V_{12} + V_{21}) \nonumber \\
+ \Delta_{2}^{2} - f_{1}^{2} - f_{2}^{2},
\label{eqn:two_mode_lambda_x2}\end{aligned}$$ $$\begin{aligned}
\fl x_{3} = \Delta_{2}(V_{12} + V_{21})(V_{11} + V_{12} + 2V_{22} - \Delta_{1}) \nonumber \\
- f_{2}^{2} (V_{11} + V_{22} - \Delta_{1} + \Delta_{2})+ \Delta_{2}^{2}(V_{12} + V_{21}) \nonumber \\
+ (V_{12} + V_{21}) \left\{ (V_{12} + V_{22}) (V_{11} + V_{22} - \Delta_{1}) - f_{1}^{2} \right\},
\label{eqn:two_mode_lambda_x3}\end{aligned}$$ and $$\fl b_{j} = f_{1} f_{2}/[(\mu_{j}-\mu_{k}) (\mu_{j}-\mu_{l})], \;
j \neq k \neq l.
\label{eqn:b_j}$$ Finally, for $m=0$ we find $$\fl A_{n0}(t) = \sum\limits_{j=1}^{2} c_{j} e^{i \alpha_{j} t},
\label{eqn:two_mode_v_a_m0}$$ $$\fl B_{n0}(t) = 0,
\label{eqn:two_mode_v_b_m0}$$ $$\fl C_{n0}(t) = - \frac{e^{i \Delta_{1}}} {f_{1}} \sum\limits_{j=1}^{2} c_{j} (\alpha_{j} + V_{11}) e^{i \alpha_{j} t},
\label{eqn:two_mode_v_c_m0}$$ where $$\fl
c_{1} = \frac{V_{11} + \alpha_{2}} {\alpha_{2} - \alpha_{1}},\quad c_{2} = \frac{V_{11} + \alpha_{1}} {\alpha_{1} - \alpha_{2}}
\label{eqn:two_mode_v_a1}$$ and $$\fl \alpha_{1,2} = {\textstyle\frac{1}{2}}
[-y_{1} \pm (y_{1}^{2} - 4y_{2})^{1/2}],
\label{eqn:two_mode_v_a1}$$ where $$\fl y_{1} = V_{11} + V_{12},\quad y_{2} = V_{12} V_{11} - f_{1}^{2}.
\label{eqn:two_mode_v_y1}$$ In the expressions for $B_{nm}(t)$ in the foregoing, the contribution from spontaneous emission has not been included, as it can be shown to be negligible.
With these results at hand, the general matrix elements of the reduced density matrices $\rho_{1}(t), \rho_{2}(t)$ and $\rho_{\rm a}(t)$ for the field subsystems $F_{1}, F_{2}$ and the atom subsystem, respectively, can be computed by tracing over the other two subsystems. We obtain, finally, $$\begin{aligned}
\fl {\left\langle{n}\right|} \rho_{1} (t) {\left|{n^{\prime}}\right\rangle} = \sum_{l=0}^{\infty} \bigg[ q_{n} q_{n^{\prime}}^{*} r_{l} r_{l}^{*} A_{n,l} A_{n^{\prime},l}^{*} + + q_{n+1} q_{n^{\prime}+1}^{*} r_{l} r_{l}^{*} C_{n+1,l} C_{n^{\prime}+1,l}^{*}\nonumber \\
+ (1-\delta_{l,0})\big( q_{n+1} q_{n^{\prime}+1}^{*} r_{l-1} r_{l-1}^{*} B_{n+1,l-1} B_{n^{\prime}+1,l-1}^{*} ) \bigg],
\label{eqn:two_mode_lambda_rho_f1_matrix_elts}\end{aligned}$$ $$\begin{aligned}
\fl {\left\langle{l}\right|} \rho_{2} (t) {\left|{l^{\prime}}\right\rangle} = \sum\limits_{n=0}^{\infty} \bigg[ q_{n} q_{n}^{*} r_{l} r_{l^{\prime}}^{*} A_{n,l} A_{n,l^{\prime}}^{*} + q_{n+1} q_{n+1}^{*} r_{l} r_{l^{\prime}}^{*} C_{n+1,l} C_{n+1,l^{\prime}}^{*} \nonumber \\
+ (1-\delta_{l,0}) (1-\delta_{l^{\prime},0})\big( q_{n+1} q_{n+1}^{*} r_{l-1} r_{l^{\prime}-1}^{*} B_{n+1,l-1} B_{n+1,l^{\prime}-1}^{*} ) \bigg],
\label{eqn:two_mode_lambda_rho_f2_matrix_elts}\end{aligned}$$ and $$\begin{aligned}
\label{eqn:two_mode_lambda_rho_at_matrix_elts_a}
\fl {\left\langle{1}\right|} \rho_{\rm a} (t) {\left|{1}\right\rangle} &=& \sum\limits_{k=0}^{\infty} \sum\limits_{l=0}^{\infty} q_{k} q_{k}^{*} r_{l} r_{l}^{*} A_{k,l} A_{k,l}^{*}, \\
\label{eqn:two_mode_lambda_rho_at_matrix_elts_b}
\fl {\left\langle{2}\right|} \rho_{\rm a} (t) {\left|{2}\right\rangle} &=& \sum\limits_{k=0}^{\infty} \sum\limits_{l=1}^{\infty} q_{k+1} q_{k^+1}^{*} r_{l-1} r_{l-1}^{*} B_{k+1,l-1} B_{k+1,l-1}^{*}, \\
\label{eqn:two_mode_lambda_rho_at_matrix_elts_c}
\fl {\left\langle{3}\right|} \rho_{\rm a} (t) {\left|{3}\right\rangle} &=& \sum\limits_{k=0}^{\infty} \sum\limits_{l=0}^{\infty} q_{k+1} q_{k+1}^{*} r_{l} r_{l}^{*} C_{k+1,l} C_{k+1,l}^{*}, \\
\label{eqn:two_mode_lambda_rho_at_matrix_elts_d}
\fl {\left\langle{1}\right|} \rho_{\rm a} (t) {\left|{2}\right\rangle} &=& \sum\limits_{k=0}^{\infty} \sum\limits_{l=1}^{\infty} q_{k} q_{k+1}^{*} r_{l} r_{l-1}^{*} A_{k,l} B_{k+1,l-1}^{*}, \\
\label{eqn:two_mode_lambda_rho_at_matrix_elts_e}
\fl {\left\langle{1}\right|} \rho_{\rm a} (t) {\left|{3}\right\rangle} &=& \sum\limits_{k=0}^{\infty} \sum\limits_{l=0}^{\infty} q_{k} q_{k+1}^{*} r_{l} r_{l}^{*} A_{k,l} C_{k+1,l}^{*}, \\
\label{eqn:two_mode_lambda_rho_at_matrix_elts_f}
\fl {\left\langle{2}\right|} \rho_{\rm a} (t) {\left|{3}\right\rangle} &=& \sum\limits_{k=0}^{\infty} \sum\limits_{l=1}^{\infty} q_{k+1} q_{k+1}^{*} r_{l-1} r_{l}^{*} B_{k+1,l-1} C_{k+1,l}^{*}.\end{aligned}$$ The quantity of interest in the context of EIT (to be analysed in the next section) is the expectation value of the photon number of the probe field $F_{1}$, given by $$\fl {\left\langle{N_{1}(t)}\right\rangle} = \Tr\,[\rho_{1}(t) N_{1}],
\label{}$$ while entanglement dynamics (examined in section \[entanglement\]) is characterised by the SVNE of the atomic subsystem, defined as $$\fl S_{\rm a}(t) = - \Tr\, [\rho_{\rm a}(t) \,\ln\,\rho_{\rm a}(t)].
\label{}$$
Electromagnetically-induced transparency (EIT) {#eit}
==============================================
####
We are ready, now, to investigate EIT for a $\Lambda$ atom interacting with the probe field $F_{1}$ and the coupling field $F_{2}$. In order to have a reference point from which to understand the general behaviour, we consider first the relatively simple case in which the detuning parameter $\Delta_{2}=0$, the couplings $\lambda_{1} = \lambda_{2} = \lambda = 1$, and the field nonlinearity parameters $\chi_{1} = \chi_{2} = 0$. As mentioned earlier, we consider situations where ${\left\langle{N_{1}}\right\rangle}$ exhibits collapses and revivals during the temporal evolution of the system, and examine features of EIT during that particular interval of time when ${\left\langle{N_{1}}\right\rangle}$ first collapses to a constant value in the absence of $F_2$.
![${\left\langle{N_{1}(t)}\right\rangle}$ in the case of initial coherent states ${\left|{\alpha_{1}}\right\rangle}$ and ${\left|{\alpha_{2}}\right\rangle}$ of the field modes. $|\alpha_1|^2=10$ and (a) $|\alpha_2|^2=0$, (b) $|\alpha_2|^2=10$, (c) $|\alpha_2|^2=18$. (d) ${\left\langle{N_{1}}\right\rangle}$ as a function of the detuning parameter $\Delta_{1}$, for different values of $|\alpha_{2}|^{2}$.[]{data-label="fig:EIT_1"}](plots/sudarsan/coh-coh/cohcoh-timeseries-nu1=10nu2=0.eps "fig:") ![${\left\langle{N_{1}(t)}\right\rangle}$ in the case of initial coherent states ${\left|{\alpha_{1}}\right\rangle}$ and ${\left|{\alpha_{2}}\right\rangle}$ of the field modes. $|\alpha_1|^2=10$ and (a) $|\alpha_2|^2=0$, (b) $|\alpha_2|^2=10$, (c) $|\alpha_2|^2=18$. (d) ${\left\langle{N_{1}}\right\rangle}$ as a function of the detuning parameter $\Delta_{1}$, for different values of $|\alpha_{2}|^{2}$.[]{data-label="fig:EIT_1"}](plots/sudarsan/coh-coh/cohcoh-nu1=10nu2=10-timeseries.eps "fig:") ![${\left\langle{N_{1}(t)}\right\rangle}$ in the case of initial coherent states ${\left|{\alpha_{1}}\right\rangle}$ and ${\left|{\alpha_{2}}\right\rangle}$ of the field modes. $|\alpha_1|^2=10$ and (a) $|\alpha_2|^2=0$, (b) $|\alpha_2|^2=10$, (c) $|\alpha_2|^2=18$. (d) ${\left\langle{N_{1}}\right\rangle}$ as a function of the detuning parameter $\Delta_{1}$, for different values of $|\alpha_{2}|^{2}$.[]{data-label="fig:EIT_1"}](plots/sudarsan/coh-coh/coh-coh-nu1=10nu2=18-timeseries.eps "fig:") ![${\left\langle{N_{1}(t)}\right\rangle}$ in the case of initial coherent states ${\left|{\alpha_{1}}\right\rangle}$ and ${\left|{\alpha_{2}}\right\rangle}$ of the field modes. $|\alpha_1|^2=10$ and (a) $|\alpha_2|^2=0$, (b) $|\alpha_2|^2=10$, (c) $|\alpha_2|^2=18$. (d) ${\left\langle{N_{1}}\right\rangle}$ as a function of the detuning parameter $\Delta_{1}$, for different values of $|\alpha_{2}|^{2}$.[]{data-label="fig:EIT_1"}](plots/sudarsan/coh-coh/cohcoh-NF1-sweep-over-nuF2.eps "fig:")
When the initial states of $F_1$ and $F_2$ are coherent states ${\left|{\alpha_{1}}\right\rangle} $ and ${\left|{\alpha_{2}}\right\rangle}$, ${\left\langle{N_{1}(t)}\right\rangle}$ undergoes collapses and revivals in time for a range of values of $\Delta_1$ and $|\alpha_2|^{2}$, for a given value of $|\alpha_1|^{2}$ (figures \[fig:EIT\_1\](a)-(c)). The time interval over which ${\left\langle{N_{1}}\right\rangle}$ exhibits collapse is sensitive to the values of both $\Delta_{1}$ and $|\alpha_1|^{2}$. Since the intervals of first collapse for different values of $\Delta_{1}$ overlap with each other, we can capture the appearance of EIT in a plot of ${\left\langle{N_{1}(t)}\right\rangle}$ versus $\Delta_{1}$ at any specific instant of time in this overlap interval (figure \[fig:EIT\_1\](d)). In particular, the generation of a transparency window around $\Delta_1=0$ for $|\alpha_{2}|^2 \gg |\alpha_{1}|^2$ is manifest. We have also verified that the occupation probabilities of the atomic states $|1\rangle$ and $|3\rangle$ appropriately complement the behaviour of ${\left\langle{N_{1}}\right\rangle}$ in this time interval.
![${\left\langle{N_{1}(t)}\right\rangle}$ versus $\Delta_{1}$, when the initial state of $F_{2}$ is (a) an even CS (b) an odd CS, (c) a Yurke-Stoler state (d) an $m$-photon-added CS.[]{data-label="fig:EIT_catpacs"}](plots/sudarsan/coh-cat/coh-ecs-sweepovernuF2.eps "fig:") ![${\left\langle{N_{1}(t)}\right\rangle}$ versus $\Delta_{1}$, when the initial state of $F_{2}$ is (a) an even CS (b) an odd CS, (c) a Yurke-Stoler state (d) an $m$-photon-added CS.[]{data-label="fig:EIT_catpacs"}](plots/sudarsan/coh-cat/coh-ocs-sweepovernuF2.eps "fig:") ![${\left\langle{N_{1}(t)}\right\rangle}$ versus $\Delta_{1}$, when the initial state of $F_{2}$ is (a) an even CS (b) an odd CS, (c) a Yurke-Stoler state (d) an $m$-photon-added CS.[]{data-label="fig:EIT_catpacs"}](plots/sudarsan/coh-cat/coh-ys-sweepovernuF2.eps "fig:") ![${\left\langle{N_{1}(t)}\right\rangle}$ versus $\Delta_{1}$, when the initial state of $F_{2}$ is (a) an even CS (b) an odd CS, (c) a Yurke-Stoler state (d) an $m$-photon-added CS.[]{data-label="fig:EIT_catpacs"}](plots/sudarsan/coh-cat/coh-pacs-sweepovernuF2.eps "fig:")
We now proceed to examine the effect of field nonlinearities, intensity-dependent couplings and different initial states of the coupling field on EIT. As before, we take $F_1$ to be initially in a CS with $|\alpha_{1}|^2=10$, and set $\chi_1 = \chi_2 = 0$, $\Delta_2=0$, $\lambda_1=\lambda_2=\lambda=1$, while the initial state of $F_2$ is chosen to be an $m$-PACS or a Schrödinger cat state. Once again, EIT occurs during the first interval of collapse of ${\left\langle{N_{1}(t)}\right\rangle}$ (figures \[fig:EIT\_catpacs\](a)-(c)). The height of the transparency window in the absorption spectrum now depends directly on ${\left\langle{N_{2}}\right\rangle}$, consistent with the experimental finding [@boller]. When the initial state of $F_2$ is an $m$-PACS (as opposed to a CS), then, with increasing $m$, the field mode undergoes incomplete collapses which have Rabi-like oscillations about a mean value. This is reflected in the jaggedness of the transparency peak for large values of $m$ (figure \[fig:EIT\_catpacs\](d)). For very large values of $m$ the intervals of collapse corresponding to different values of $\Delta_1$ no longer overlap significantly, consistent with what happens when $F_2$ is initially in a photon number state.
The presence of nonlinearities in $F_1$ mitigates collapses of ${\left\langle{N_{1}(t)}\right\rangle}$. On the other hand, if $F_2$ has a Kerr-like nonlinearity, complete collapses in ${\left\langle{N_{1}(t)}\right\rangle}$ occur. We investigate the nature of EIT during the time interval of the first collapse by choosing initial coherent states with $|\alpha_{1}|^2 = 10$ and $|\alpha_{2}|^2 = 18$ for $F_{1}$ and $F_{2}$, setting $\lambda_1=\lambda_2=\lambda=1$, $\chi_1=0$, and varying $\chi_2$. The EIT spectrum (a plot of ${\left\langle{N_{1}(t)}\right\rangle}$ versus $\Delta_1$) is no longer symmetric about $\Delta_{1}=0$, in contrast to the earlier case, and the asymmetry becomes more pronounced with increasing $\chi_2$. A noteworthy feature is that this behaviour is similar to that obtained by varying $\Delta_2$ over a small range of values about zero (figures \[fig:EIT\_kerrsweep2\](a), (b)).
![EIT spectrum for different values of (a) $\Delta_2$ and (b) $\chi_2$.[]{data-label="fig:EIT_kerrsweep2"}](plots/sudarsan/coh-cohkerr/sweep/cohcoh-NF1-sweep-over-weakdelta2.eps "fig:") ![EIT spectrum for different values of (a) $\Delta_2$ and (b) $\chi_2$.[]{data-label="fig:EIT_kerrsweep2"}](plots/sudarsan/coh-cohkerr/sweep/cohcoh-NF1-sweep-over-weakX2.eps "fig:")
![EIT spectrum for different values of $\kappa$.[]{data-label="fig:EIT_idc"}](plots/sudarsan/idc/cohcoh-NF1-sweep-over-k-idc.eps)
Finally, if the coupling between the atom and either of the two fields has an intensity dependence of the form $f(N) =
(1+\kappa N)^{1/2}$, the EIT spectrum stretches and flattens out with increasing $\kappa$. Figure \[fig:EIT\_idc\] illustrates this feature in the case of initial coherent states of $F_{1}$ and $F_{2}$ with $|\alpha_{1}|^2=10, |\alpha_{2}|^2=18$. Unlike what happens for non-zero values of $\chi_{2}$ or $\Delta_{2}$, however, the spectrum remains symmetric about $\Delta_{1} = 0$ as $\kappa$ is increased, while the minima in ${\left\langle{N_{1}(t)}\right\rangle}$ that indicate EIT become less pronounced and move out to larger and larger values of $|\Delta_{1}|$.
Entanglement dynamics of a $\Lambda$ atom coupled to two field modes {#entanglement}
====================================================================
####
We turn, next, to the entanglement dynamics of our tripartite system. We start with the simple case of an intensity-independent field-atom coupling, and then go on to consider the case of an intensity-dependent coupling proportional to $(1+\kappa N)^{1/2}$. In order to avoid inessential complications, we set $\Delta_{1}=\Delta_{2} =\Delta$, $\lambda_{1}=\lambda_{2}=\lambda$, $\chi_{1}=\chi_{2}=\chi$, and $\kappa_{1}=\kappa_{2}=\kappa$.
Field-atom interactions with constant coupling strengths
---------------------------------------------------------
####
Earlier work[@athreya1] has indicated an interesting feature in the behaviour of the entanglement in the system as measured by $S_{\rm a}(t)$, the SVNE of the atomic subsystem, when the nonlinearity parameter $\chi$ is large compared to the field-atom interaction strength $\lambda$. In this regime, $S_{\rm a}$ displays a collapse (to a steady value) when the initial field state is a PACS, in contrast to what happens when it is a CS. This feature has been verified in detail for the system at hand, and figures \[fig:svne\_1\](a) and (b) depict some representative results in this regard. With an increase in the number of added photons in the initial field states, there is a systematic increase of the interval during which $S_{\rm a}$ remains at a steady value. We have verified that these features are also reflected in the Mandel Q parameter and in the mean and variance of the photon number $N_{1}$.
![$S_{\rm a}(t)$ for a $\Lambda$ atom interacting with two field modes, for $\chi/\lambda = 5$, $\Delta = 0$. Initial state (a) ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}$ and (b) ${\left|{1;\alpha, 10; \alpha, 10}\right\rangle}$, with ${|\alpha|}^{2} =10$.[]{data-label="fig:svne_1"}](plots/pradip/svne_idc_off/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_kappa_0_0_chi_5_5.eps "fig:") ![$S_{\rm a}(t)$ for a $\Lambda$ atom interacting with two field modes, for $\chi/\lambda = 5$, $\Delta = 0$. Initial state (a) ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}$ and (b) ${\left|{1;\alpha, 10; \alpha, 10}\right\rangle}$, with ${|\alpha|}^{2} =10$.[]{data-label="fig:svne_1"}](plots/pradip/svne_idc_off/svne_l_atom_alpha2_10_10_pacs_10_10_delta_0_0_kappa_0_0_chi_5_5.eps "fig:")
The time interval during which the SVNE holds at a steady value is enhanced when ${|\alpha |}^{2}$ is increased (compare figures \[fig:svne\_1\](a) and \[fig:svne\_2\](a)), or the ratio ${\chi}/{\lambda}$ is increased (see, e.g., figures \[fig:svne\_1\](a) and \[fig:svne\_2\](b)), or both, all other parameters remaining unchanged. This sort of entanglement collapse is absent in the case of initial coherent states, and also for initial PACS provided the nonlinearity parameter $\chi$ is sufficiently small [@athreya1]. The route to entanglement collapse with increase in nonlinearity is evident by comparing figures \[fig:svne\_3\](a) and (b) with figure \[fig:svne\_1\](a). As one might expect, there is a shrinkage of the time interval over which the entanglement collapses as one moves away from exact resonance to non-zero values of the detuning parameters $\Delta_{i}$.
![$S_{\rm a}(t)$ for the initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}$, where (a) ${|\alpha|}^{2} = 20$ and $\chi/\lambda = 5$; (b) ${|\alpha|}^{2} =10$ and $\chi/\lambda = 10$ (strong nonlinearity). In both cases, $\Delta = 0$.[]{data-label="fig:svne_2"}](plots/pradip/svne_idc_off/svne_l_atom_alpha2_20_20_pacs_5_5_delta_0_0_chi_5_5_kappa_0_0.eps "fig:") ![$S_{\rm a}(t)$ for the initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}$, where (a) ${|\alpha|}^{2} = 20$ and $\chi/\lambda = 5$; (b) ${|\alpha|}^{2} =10$ and $\chi/\lambda = 10$ (strong nonlinearity). In both cases, $\Delta = 0$.[]{data-label="fig:svne_2"}](plots/pradip/svne_idc_off/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_kappa_0_0_chi_10_10.eps "fig:")
![$S_{\rm a}(t)$ for the initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}$ with ${|\alpha|}^{2} = 10$ and (a) $\chi = 0$ and (b) $\chi/\lambda = 1$ (weak nonlinearity). In both cases, $\Delta = 0$.[]{data-label="fig:svne_3"}](plots/pradip/svne_weak_non/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_chi_0_0.eps "fig:") ![$S_{\rm a}(t)$ for the initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}$ with ${|\alpha|}^{2} = 10$ and (a) $\chi = 0$ and (b) $\chi/\lambda = 1$ (weak nonlinearity). In both cases, $\Delta = 0$.[]{data-label="fig:svne_3"}](plots/pradip/svne_weak_non/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_chi_1_1.eps "fig:")
For completeness, we have also investigated the role of squeezing on entanglement collapse in this model. No collapse is exhibited by $S_{\rm a}$ when $F_{1}$ is a standard squeezed vacuum state (labelled by the complex parameter $\xi$) and $F_{2}$ is either a PACS or a squeezed vacuum state (with the same parameter $\xi$). Figures \[fig:svne\_4\](a) and (b) illustrate these conclusions. Moreover, we have verified that, for those initial states for which a collapse of $S_{\rm a}$ does occur, the field states do not exhibit squeezing or second-order squeezing during the time interval of the collapse. It would seem, therefore, that it is the extent of [*coherence*]{}, above all else, that is the primary determining factor in the occurrence of entanglement collapses in our model system.
![$S_{\rm a}(t)$ in the presence of squeezing. Initial states (a) ${\left|{1;\xi;\alpha,5}\right\rangle}$, (b) ${\left|{1;\xi; \alpha, 10}\right\rangle}$ and (c) ${\left|{1;\xi;\xi}\right\rangle}$, with ${|\alpha|}^{2} =10$ and squeezing parameter $\xi = 2$. In all cases, $\Delta = 0$ and $\chi/\lambda = 5$.[]{data-label="fig:svne_4"}](plots/pradip/squeeze/sq_svne_l_atom_data_chi_5_5_pacs_5_r_2_theta_0_alpha2_10.eps "fig:") ![$S_{\rm a}(t)$ in the presence of squeezing. Initial states (a) ${\left|{1;\xi;\alpha,5}\right\rangle}$, (b) ${\left|{1;\xi; \alpha, 10}\right\rangle}$ and (c) ${\left|{1;\xi;\xi}\right\rangle}$, with ${|\alpha|}^{2} =10$ and squeezing parameter $\xi = 2$. In all cases, $\Delta = 0$ and $\chi/\lambda = 5$.[]{data-label="fig:svne_4"}](plots/pradip/squeeze/sq_svne_l_atom_data_chi_5_5_pacs_10_r_2_theta_0_alpha2_10.eps "fig:") ![$S_{\rm a}(t)$ in the presence of squeezing. Initial states (a) ${\left|{1;\xi;\alpha,5}\right\rangle}$, (b) ${\left|{1;\xi; \alpha, 10}\right\rangle}$ and (c) ${\left|{1;\xi;\xi}\right\rangle}$, with ${|\alpha|}^{2} =10$ and squeezing parameter $\xi = 2$. In all cases, $\Delta = 0$ and $\chi/\lambda = 5$.[]{data-label="fig:svne_4"}](plots/pradip/squeeze/sq_sq_svne_l_atom_data_chi_5_5_pacs_5_r_2_theta_0_alpha2_10_10.eps "fig:")
Intensity-dependent couplings: From SVNE collapse to the revival phenomenon
---------------------------------------------------------------------------
####
We turn,now, to the effect of an intensity-dependent coupling of the atom with the field modes, as characterised by a non-constant function $f(N_{i})$ in (\[defnofR\]).
![$S_{\rm a}(t)$ for intensity-dependent coupling $f(N) = N^{1/2}$. Initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}$, ${|\alpha|}^{2} = 10$, $\Delta = 0$. (a) $\chi = 0$, (b) $\chi/\lambda = 6$, (c) $\chi/\lambda = 10$.[]{data-label="fig:svne_6"}](plots/pradip/idc_bs/bs_svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_chi_0_0.eps "fig:") ![$S_{\rm a}(t)$ for intensity-dependent coupling $f(N) = N^{1/2}$. Initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}$, ${|\alpha|}^{2} = 10$, $\Delta = 0$. (a) $\chi = 0$, (b) $\chi/\lambda = 6$, (c) $\chi/\lambda = 10$.[]{data-label="fig:svne_6"}](plots/pradip/idc_bs/bs_svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_chi_6_6.eps "fig:") ![$S_{\rm a}(t)$ for intensity-dependent coupling $f(N) = N^{1/2}$. Initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}$, ${|\alpha|}^{2} = 10$, $\Delta = 0$. (a) $\chi = 0$, (b) $\chi/\lambda = 6$, (c) $\chi/\lambda = 10$.[]{data-label="fig:svne_6"}](plots/pradip/idc_bs/bs_svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_chi_10_10.eps "fig:")
In order to facilitate ready comparison, we consider the same initial state, ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}$ with $|\alpha|^{2} = 10$ and $\Delta = 0$, as in figures \[fig:svne\_1\](a), \[fig:svne\_3\](a) and \[fig:svne\_3\](b). These correspond to intensity-independent coupling, i.e., $f(N_{i}) = 1$, for a range of values of $\chi/\lambda$. Consider, now, the functional form[@buck] $f(N_{i}) = N_{i}^{1/2}$. Figures \[fig:svne\_6\](a)-(c) depict the behaviour of $S_{\rm a}(t)$ in this case. It is evident that SVNE collapse is absent now, even for strong nonlinearity. The behaviour in these two limiting cases motivates an examination of the effect of the intensity-dependent coupling $f(N_{i}) = (1+\kappa N_{i})^{1/2}$ over a range of values of the parameter $\kappa$, running from $0$ to $1$. Remarkably diverse features emerge, including SVNE collapse as well as recurrent collapses and revivals, as shown in figures \[fig:svne\_5\](a)-(l). The behaviour is sensitively dependent on the value of $\kappa$.
![$S_{\rm a}(t)$ for intensity-dependent coupling $f(N) = (1+\kappa N)^{1/2}$. Initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}, \, |\alpha|^{2} = 10, \,\Delta = 0, \,\chi/\lambda = 5$. The value of $\kappa$ is (a) 0, (b) 0.0012, (c) 0.002, (d) 0.0034, (e) 0.005, (f) 0.006, (g) 0.01, (h) 0.02, (i) 0.03, (j) 0.1, (k) 0.3 and (l) 1.[]{data-label="fig:svne_5"}](plots/pradip/idc_siva/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_kappa_0_0_chi_5_5.eps "fig:") ![$S_{\rm a}(t)$ for intensity-dependent coupling $f(N) = (1+\kappa N)^{1/2}$. Initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}, \, |\alpha|^{2} = 10, \,\Delta = 0, \,\chi/\lambda = 5$. The value of $\kappa$ is (a) 0, (b) 0.0012, (c) 0.002, (d) 0.0034, (e) 0.005, (f) 0.006, (g) 0.01, (h) 0.02, (i) 0.03, (j) 0.1, (k) 0.3 and (l) 1.[]{data-label="fig:svne_5"}](plots/pradip/idc_siva/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_kappa_0_0012_0_0012_chi_5_5.eps "fig:") ![$S_{\rm a}(t)$ for intensity-dependent coupling $f(N) = (1+\kappa N)^{1/2}$. Initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}, \, |\alpha|^{2} = 10, \,\Delta = 0, \,\chi/\lambda = 5$. The value of $\kappa$ is (a) 0, (b) 0.0012, (c) 0.002, (d) 0.0034, (e) 0.005, (f) 0.006, (g) 0.01, (h) 0.02, (i) 0.03, (j) 0.1, (k) 0.3 and (l) 1.[]{data-label="fig:svne_5"}](plots/pradip/idc_siva/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_kappa_0_002_0_002_chi_5_5.eps "fig:") ![$S_{\rm a}(t)$ for intensity-dependent coupling $f(N) = (1+\kappa N)^{1/2}$. Initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}, \, |\alpha|^{2} = 10, \,\Delta = 0, \,\chi/\lambda = 5$. The value of $\kappa$ is (a) 0, (b) 0.0012, (c) 0.002, (d) 0.0034, (e) 0.005, (f) 0.006, (g) 0.01, (h) 0.02, (i) 0.03, (j) 0.1, (k) 0.3 and (l) 1.[]{data-label="fig:svne_5"}](plots/pradip/idc_siva/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_kappa_0_0034_0_0034_chi_5_5.eps "fig:") ![$S_{\rm a}(t)$ for intensity-dependent coupling $f(N) = (1+\kappa N)^{1/2}$. Initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}, \, |\alpha|^{2} = 10, \,\Delta = 0, \,\chi/\lambda = 5$. The value of $\kappa$ is (a) 0, (b) 0.0012, (c) 0.002, (d) 0.0034, (e) 0.005, (f) 0.006, (g) 0.01, (h) 0.02, (i) 0.03, (j) 0.1, (k) 0.3 and (l) 1.[]{data-label="fig:svne_5"}](plots/pradip/idc_siva/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_kappa_0_005_0_005_chi_5_5.eps "fig:") ![$S_{\rm a}(t)$ for intensity-dependent coupling $f(N) = (1+\kappa N)^{1/2}$. Initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}, \, |\alpha|^{2} = 10, \,\Delta = 0, \,\chi/\lambda = 5$. The value of $\kappa$ is (a) 0, (b) 0.0012, (c) 0.002, (d) 0.0034, (e) 0.005, (f) 0.006, (g) 0.01, (h) 0.02, (i) 0.03, (j) 0.1, (k) 0.3 and (l) 1.[]{data-label="fig:svne_5"}](plots/pradip/idc_siva/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_kappa_0_006_0_006_chi_5_5.eps "fig:") ![$S_{\rm a}(t)$ for intensity-dependent coupling $f(N) = (1+\kappa N)^{1/2}$. Initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}, \, |\alpha|^{2} = 10, \,\Delta = 0, \,\chi/\lambda = 5$. The value of $\kappa$ is (a) 0, (b) 0.0012, (c) 0.002, (d) 0.0034, (e) 0.005, (f) 0.006, (g) 0.01, (h) 0.02, (i) 0.03, (j) 0.1, (k) 0.3 and (l) 1.[]{data-label="fig:svne_5"}](plots/pradip/idc_siva/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_kappa_0_01_0_01_chi_5_5.eps "fig:") ![$S_{\rm a}(t)$ for intensity-dependent coupling $f(N) = (1+\kappa N)^{1/2}$. Initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}, \, |\alpha|^{2} = 10, \,\Delta = 0, \,\chi/\lambda = 5$. The value of $\kappa$ is (a) 0, (b) 0.0012, (c) 0.002, (d) 0.0034, (e) 0.005, (f) 0.006, (g) 0.01, (h) 0.02, (i) 0.03, (j) 0.1, (k) 0.3 and (l) 1.[]{data-label="fig:svne_5"}](plots/pradip/idc_siva/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_kappa_0_02_0_02_chi_5_5.eps "fig:") ![$S_{\rm a}(t)$ for intensity-dependent coupling $f(N) = (1+\kappa N)^{1/2}$. Initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}, \, |\alpha|^{2} = 10, \,\Delta = 0, \,\chi/\lambda = 5$. The value of $\kappa$ is (a) 0, (b) 0.0012, (c) 0.002, (d) 0.0034, (e) 0.005, (f) 0.006, (g) 0.01, (h) 0.02, (i) 0.03, (j) 0.1, (k) 0.3 and (l) 1.[]{data-label="fig:svne_5"}](plots/pradip/idc_siva/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_kappa_0_03_0_03_chi_5_5.eps "fig:") ![$S_{\rm a}(t)$ for intensity-dependent coupling $f(N) = (1+\kappa N)^{1/2}$. Initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}, \, |\alpha|^{2} = 10, \,\Delta = 0, \,\chi/\lambda = 5$. The value of $\kappa$ is (a) 0, (b) 0.0012, (c) 0.002, (d) 0.0034, (e) 0.005, (f) 0.006, (g) 0.01, (h) 0.02, (i) 0.03, (j) 0.1, (k) 0.3 and (l) 1.[]{data-label="fig:svne_5"}](plots/pradip/idc_siva/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_kappa_0_1_0_1_chi_5_5.eps "fig:") ![$S_{\rm a}(t)$ for intensity-dependent coupling $f(N) = (1+\kappa N)^{1/2}$. Initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}, \, |\alpha|^{2} = 10, \,\Delta = 0, \,\chi/\lambda = 5$. The value of $\kappa$ is (a) 0, (b) 0.0012, (c) 0.002, (d) 0.0034, (e) 0.005, (f) 0.006, (g) 0.01, (h) 0.02, (i) 0.03, (j) 0.1, (k) 0.3 and (l) 1.[]{data-label="fig:svne_5"}](plots/pradip/idc_siva/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_kappa_0_3_0_3_chi_5_5.eps "fig:") ![$S_{\rm a}(t)$ for intensity-dependent coupling $f(N) = (1+\kappa N)^{1/2}$. Initial state ${\left|{1;\alpha, 5; \alpha, 5}\right\rangle}, \, |\alpha|^{2} = 10, \,\Delta = 0, \,\chi/\lambda = 5$. The value of $\kappa$ is (a) 0, (b) 0.0012, (c) 0.002, (d) 0.0034, (e) 0.005, (f) 0.006, (g) 0.01, (h) 0.02, (i) 0.03, (j) 0.1, (k) 0.3 and (l) 1.[]{data-label="fig:svne_5"}](plots/pradip/idc_siva/svne_l_atom_alpha2_10_10_pacs_5_5_delta_0_0_kappa_1_1_chi_5_5.eps "fig:")
The collapse of the SVNE over a considerable time interval that occurs for $\kappa = 0$ (figure \[fig:svne\_5\](a)) is gradually lost (figures \[fig:svne\_5\](b) to (e)) as $\kappa$ increases to a value slightly above $0. 005$ . An apparent precursor to collapse again appears as $\kappa$ is further increased slightly (figure \[fig:svne\_5\](f), $\kappa = 0.006$), but what happens is that there is a shorter-duration collapse, followed by a revival, and then a second incipient collapse and revival (figure \[fig:svne\_5\](g), $\kappa = 0.01$), within the original time interval of collapse. A further small increase in $\kappa$ produces a distinctive sequence of collapses and revivals (figures \[fig:svne\_5\](h), (i), (j)), which remains clear-cut till $\kappa$ becomes a little larger than $0.1$. Subsequently, the intervals between successive revivals become too short to be discernible on the scale used in figure \[fig:svne\_5\], and, moreover, fractional revivals start filling these small intervals (figure \[fig:svne\_5\](k), $\kappa = 0.3$). As $\kappa$ is increased further, the entanglement collapse and revival phenomenon is no longer discernible. Figure \[fig:svne\_5\](l) depicts $S_{\rm a}(t)$ for $\kappa = 1$. It is thus evident that a series of qualitative changes is exhibited by the entanglement entropy in the system under study as the value of $\kappa$ is changed in a relatively small range, signalling very sensitive dependence on this parameter in a manner reminiscent of bifurcation cascades preceding the onset of chaos in nonlinear classical dynamical systems.
Finally, as stated in the Introduction, we have verified that these interesting features are also exhibited in the case of a V atom interacting with two radiation modes.
Concluding remarks {#concluding remarks}
==================
####
We have considered a tripartite system comprising a $\Lambda$-type or V-type atom interacting with two radiation fields. The mean photon number corresponding to the probe field is seen to display collapses and revivals for specific initial field states and parameters in the Hamiltonian, in the absence of the coupling field. With the coupling field turned on, a window of electromagnetically-induced transparency appears during the collapse interval in the absorption spectrum. On a longer time scale, interesting dynamical effects are observed in the time evolution of the entanglement. This includes a collapse of the subsystem von Neumann entropy of the atom over a considerable time interval. Both these features are sensitive to the nature of the initial states of the fields, and seem to reflect the extent of the departure from coherence of those states.
We have attempted to identify the roles played by field nonlinearities, detuning parameters and the departure from coherence of the initial states of the radiation fields on EIT and on the extent of entanglement between subsystems during temporal evolution, since these two aspects are suitable for potential experimental investigations. Detailed experiments on EIT, the identification of the photon-added coherent state in the laboratory, and the necessity of retaining the extent of entanglement between subsystems from the point of view of quantum information processing add impetus to our investigations. Reconstruction of the state of the system during SVNE collapse could possibly be attempted through continuous-variable quantum state tomography. The unanticipated behaviour of the SVNE as the strength of the intensity-dependent field-atom coupling is varied would correspondingly take the system through a spectrum of nonclassical states which are worth identifying through state reconstruction procedures. The mean photon number is seen to reflect the long time dynamics of the SVNE corresponding to the atomic subsystem. This is an observable which lends itself to experimental observation, and hence one that can be used to examine the dynamical features predicted for the tripartite model considered.
References {#references .unnumbered}
==========
[50]{} Robinett R W 2004 [*Phys. Rep.*]{} [**392**]{} 1. Milburn G J 1986 [*Phys. Rev.*]{} A [**33**]{} 674. Kitagawa M and Yamamoto Y 1986 [*Phys. Rev.*]{} A [**34**]{} 3974. Averbukh I Sh and Perelman N F 1989 [*Phys. Lett.*]{} A [**139**]{} 449. Marangos J P 1998 [*J. Mod. Opt.*]{} [**45**]{} 471. Boller K J, Imamo[ğ]{}lu A and Harris S E 1991 [*Phys. Rev. Lett.*]{} [**66**]{} 2593. Li Y and Xiao M 1995 [*Phys. Rev.*]{} A [**51**]{} R2703. Éntin V M, Ryabtsev I I, Boguslavskii A E and Beterov I M 2000 [*JETP Lett.*]{} [**71**]{} 175. Sudheesh C, Lakshmibala S and Balakrishnan V 2004 [*Phys. Lett.*]{} A [**329**]{} 14. Sudheesh C, Lakshmibala S and Balakrishnan V 2005 [*Europhys. Lett.*]{} [**71**]{} 744. Agarwal G S and Puri R R 1989 [*Phys. Rev.*]{} A [**39**]{} 2969. Sudheesh C, Lakshmibala S and Balakrishnan V 2006 [*J. Phys.*]{} B [**39**]{} 3345. Sudheesh C, Lakshmibala S and Balakrishnan V 2009 [*Phys. Lett.*]{} A [**373**]{} 2814. Sudheesh C, Lakshmibala S and Balakrishnan V 2010 [*EPL*]{} [**90**]{} 50001. Tara K, Agarwal G S and Chaturvedi S 1993 [*Phys. Rev.*]{} A [**47**]{} 5024. Zavatta A, Viciani S and Bellini M 2004 [*Science*]{} [**306**]{} 660. Athreya Shankar 2014 [*Dynamics of field-atom interactions*]{}, B. Tech. dissertation, IIT Madras (unpublished). Athreya Shankar, Lakshmibala S and Balakrishnan V 2014 [*J. Phys.*]{} B [**47**]{} 215505. Buck B and Sukumar C V 1981 [*Phys. Lett.*]{} A [**81**]{} 132. Bu[z]{}ek V 1989 [*Phys. Lett.*]{} A [**139**]{} 231. Bu[z]{}ek V 1989 [*Phys. Rev.*]{} A [**39**]{} 3196. Zait R A 2003 [*Phys. Lett.*]{} A [**319**]{} 461. Sudarshan E C G 1993 [*Int. J. Th. Phys.*]{} [**32**]{} 1069. Sivakumar S 2004 [*Int. J. Th. Phys.*]{} [**43**]{} 2405. Sivakumar S 2002 [*J. Phys. A: Math. Gen.*]{} [**35**]{} 6755. Faghihi M J, Tavassoly M K and Harouni M B 2014 [*Laser Phys.*]{} [**24**]{} 045202. Faghihi M J, Tavassoly M K and Hooshmandasl M R 2013 [*J. Opt. Soc. Am.*]{} B [**30**]{} 1109.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The effects of mass-varying neutrinos on cosmic microwave background (CMB) anisotropies and large scale structures (LSS) are studied. In these models, dark energy and neutrinos are coupled such that the neutrino masses are functions of the scalar field playing the role of dark energy. We begin by describing the cosmological background evolution of such a system. It is pointed out that, similar to models with a dark matter/dark energy interaction, the apparent equation of state measured with SNIa can be smaller than -1. We then discuss the effect of mass-varying neutrinos on the CMB anisotropies and the matter power spectrum. A suppression of power in the CMB power spectrum at large angular scales is usually observed. We give an explanation for this behaviour and discuss different couplings and quintessence potentials to show the generality of the results obtained. We perform a likelihood analysis using wide-ranging SNIa, CMB and LSS observations to assess whether such theories are viable. Treating the neutrino mass as a free parameter we find that the constraints on the coupling are weak, since CMB and LSS surveys give only upper bounds on the neutrino mass. However, fixing a priori the neutrino masses, we find that there is some evidence that the existence of such a coupling is actually preferred by current cosmological data over the standard $\Lambda$CDM cosmology.'
author:
- 'A. W. Brookfield'
- 'C. van de Bruck'
- 'D. F. Mota'
- 'D. Tocchini-Valentini'
date: 14 December 2005
title: 'Cosmology of Mass-Varying Neutrinos Driven by Quintessence: Theory and Observations'
---
Introduction
============
Recent cosmological observations indicate that the expansion of the universe is accelerating [@observation1]-[@observation3]. It follows from General Relativity that the dominant energy component today must have negative pressure. Many candidates have been proposed over the years, including scalar field models, which are well motivated from the point of view of particle physics theories, see e.g. [@wetterich]-[@wetterich2]. The main prediction of these types of models is that the dark energy equation of state becomes a dynamical quantity, and can vary from the usual value of $w=-1$ for a cosmological constant. Although such models are very attractive, they are plagued with several theoretical difficulties, such as the stability of the potential under quantum corrections [@doran] or why the dark energy scalar field seems not to mediate a force between normal matter particles [@carroll]. In addition, the energy scale of the scalar field is put in by hand and usually not connected to a more fundamental energy scale. However, attempts have been made to address these problems, such as models with ultralight pseudo-Nambu-Goldstone bosons (see, for example, [@frieman], [@nilles] and [@kaloperdark]; for a review, see [@pecceireview]).
It is expected that any explanation for dark energy will involve physics beyond the standard model of particle physics. Recently, a new class of models have been proposed, which entertain the idea of a possible connection between neutrinos and dark energy. Their theoretical and observational consequences have already been studied very extensively [@neutrinobeg]-[@neutrinoend]. The main motivation for a connection between dark energy and neutrinos is that the energy scale of dark energy (${\cal O}(10^{-3})$ eV) is of the order of the neutrino mass scale. In these models the neutrino mass scale and the dark energy scale are linked to each other, and hence the observed non-zero neutrino masses (see [@massiveneutrinos1]-[@massiveneutrinos3]) cannot be understood without an understanding of dark energy. Also, one may hope that these models might provide an explanation for the coincidence problem [@fardon1].
In this paper we investigate the cosmology of neutrino models of dark energy. We take into account the [*full*]{} evolution of the neutrinos, i.e. studying the relativistic and non-relativistic regimes and the transition in between.
Armed with a complete numerical model for the evolution of the coupled neutrinos, we compare the background evolution with Supernova data and study how the modified perturbations affect the cosmic microwave background radiation (CMB) temperature anisotropies and large scale structures (LSS) matter power spectrum. We thereby present the details of the results outlined in [@us] and discuss other forms of coupling between dark energy and neutrinos and potentials for the dark energy field.
The paper is organized as follows: In Section II we discuss the background evolution of the coupled dark energy-neutrino system in the context of a typical quintessential potential. In Section III we derive the evolution equations for cosmological perturbations in neutrino models of dark energy, and present the modified CMB and matter power spectra. In Section IV we discuss other couplings and potentials, such as inverse power-law potentials and field–dependent couplings. In Section V we compare our theory with data, using a public Markov-Chain Monte-Carlo data analysis program. We conclude in Section VI.
The Cosmological Background Evolution
=====================================
In a flat, homogeneous, Friedmann–Robertson–Walker universe with line-element $$ds^2 = a^2(\tau)\left(-d\tau^2 + \delta_{ij}dx^i dx^j \right),$$ the Einstein equations describe the evolution of the scale factor $a(\tau)$: $$\begin{aligned}
H^2\equiv\left(\frac{\dot{a}}{a}\right)^2 &=& \frac{8\pi}{3}Ga^2\rho,\\
\frac{d}{d\tau}\left(\frac{\dot{a}}{a}\right)&=&-\frac{4\pi}{3}Ga^2(\rho+3p).\end{aligned}$$ In these equations, $\rho(\tau)$ and $p(\tau)$ are the total energy density and pressure respectively and the dot refers to the derivative with respect to conformal time $\tau$. Defining $\Omega_i=\rho_i/\rho_c$, where $\rho_c$ is the critical energy density for a flat universe and $\rho_i$ are the energy densities of the individual matter species, the equations above require that $\Omega=\sum_i \Omega_i =1$. In the following we will set $8\pi G \equiv 1$.
In our model we consider a universe with the usual energy–matter composition. At early times, the energy density is dominated by the relativistic species – radiation and highly relativistic neutrinos. As the universe expands the energy density in radiation decays, and the universe becomes matter dominated. The dominant matter species is assumed to be Cold Dark Matter (CDM), which is non–relativistic, weakly interacting and behaves like a perfect pressureless fluid. At this time there are also contributions to $\Omega$ from baryons and neutrinos (which having cooled behave in a manner similar to CDM). At late times the matter energy densities also decay away, and we enter the dark energy dominated epoch. In common with standard quintessence models we describe the dark energy sector using a dynamical scalar field with potential $V(\phi)$, where the form of the potential is chosen (and fine–tuned) to produce the necessary late time acceleration. The energy density and pressure of the scalar field are defined by the usual expressions,
$$\begin{aligned}
\rho_\phi&=&\frac{1}{2a^2}\dot{\phi}^2+V(\phi) \label{rhop}\\
p_\phi&=&\frac{1}{2a^2}\dot{\phi}^2-V(\phi). \label{pp}\end{aligned}$$
In this paper, we consider the consequences of a coupling between neutrinos and dark energy. To describe this coupling, we follow [@fardon1]: the coupling of dark energy to the neutrinos results in the neutrino mass becoming a function of the scalar field, i.e. $m_\nu=m_\nu(\phi)$, and so the mass of the neutrinos changes as the scalar field evolves. For our purposes it does not matter if the neutrinos are Majorana or Dirac particles, and for simplicity we assume three species of neutrinos with degenerate mass[^1] It is well known [@doran] that the light mass of the quintessence potential results in it being highly unstable to radiative corrections, and that the addition of a coupling between the dark energy and other matter species only serves to further exacerbate this problem. In this regard it is important that both the quintessence potential and the neutrino mass are regarded as classical, effective quantities, which already include radiative corrections.
It is also important to note that our theory differs significantly in one key aspect from the work of [@fardon1]. In our models, the dark energy sector is described by a *light* scalar field, with a mass which is at most of order $H$. The potential chosen by Fardon et al. was such that the mass of the scalar field is much larger than $H$ for most of its history, and this can have significant implications upon the behaviour of the neutrino background and the growth of perturbations [@zalda] as we will discuss later.
To fully describe the evolution of cosmological neutrinos, we must calculate their distribution function $f\left(x^i,p^i,\tau\right)$ in phase space. An important fact to note is that even though the neutrinos interact with dark energy, we treat the interaction classically and, as will be shown in eq. (\[eq:action\]), they can be thought as free-falling in a metric given by $$g_{\alpha \beta}^{\nu} = m_{\nu}(\phi)^{2}g_{\alpha \beta}.$$ Thus, the theory we are going to consider is a special type of scalar-tensor theory, in which the scalar degree of freedom couples only to neutrinos. It follows that the neutrino phase-space density is incompressible and we can treat the neutrinos as collisionless particles throughout the period of interest as long as we keep track of the evolution of the neutrino mass. We therefore need to solve the Boltzmann equation in collisionless form simultaneously with the scalar field evolution equations. Once the distribution function is known, the pressure and energy density of the neutrinos can be calculated. In this Section we will discuss the background evolution only; in the next Section we will discuss cosmological perturbations in these models.
The energy density stored in the neutrinos is given by $$\label{eq:density}
\rho_\nu=\frac{1}{a^4}\int q^2 dq\, d\Omega \epsilon f_0(q),$$ and the pressure by $$\label{eq:pressure}
p_\nu=\frac{1}{3a^4}\int q^2 dq\, d\Omega f_0(q) \frac{q^2}{\epsilon},$$ where $f_0(q)$ is the usual unperturbed background neutrino Fermi-Dirac distribution function $$f_0(\epsilon)=\frac{g_s}{h_P^3}\frac{1}{e^{\epsilon/k_B T_0}+1},$$ and $\epsilon^2 = q^2 + m_\nu^2(\phi)a^2$ ($q$ denotes the comoving momentum). As usual, $g_s$, $h_P$ and $k_B$ stand for the number of spin degrees of freedom, Planck’s constant and Boltzmann’s constant respectively. In the following we will assume that the neutrinos decouple whilst they are still relativistic, and therefore the phase-space density only depends upon the comoving momentum. Taking the time-derivative of eq. (\[eq:density\]), it can be easily shown that $$\label{eq:nuenergy}
\dot{\rho}_\nu +3H\left(\rho_\nu+p_\nu\right) = \frac{d \ln m_\nu}{d
\phi} \dot{\phi} \left(\rho_\nu - 3 p_\nu\right).$$
We describe the dark energy sector using a scalar field with potential $V(\phi)$. Taking into account the energy conservation of the coupled neutrino–dark energy system, one can immediately find that the evolution of the scalar field is described by a modified Klein-Gordon equation $$\label{eq:kleingordon}
\ddot{\phi}+2H\dot{\phi}+a^2\frac{dV}{d\phi}=-a^2\frac{d \ln m_\nu}{d
\phi}\left(\rho_\nu-3p_\nu\right).$$ This equation contains an extra source term with respect to the uncoupled case, which accounts for the energy exchange between the neutrinos and the scalar field.
For the remainder of this Section and the next, we consider a typical exponential form for the dark energy potential, namely $$V(\phi)=V_0 e^{-\sigma \phi}$$ and define $\sigma \equiv\sqrt{\frac{3}{2}}\lambda$. We also choose to take $$m_\nu(\phi)=M_0e^{\beta \phi}$$
The exponential potential can produce a non-scaling solution that may give late time acceleration, depending upon the steepness of the potential, $\sigma$ (see e.g. [@copeland; @ferreira; @amendola1; @domenico1; @kallosh]). In an uncoupled system with $\sigma<\sqrt{6}$ there exists a critical point that is stable for $\sigma^2<3(1+w)$, where $w$ stands for the equation of state of matter or radiation, and in which $\Omega_\phi=1$. This solution will lead to acceleration provided that $\sigma<\sqrt{2}$. The existence of scaling solutions depends upon the equation of state of the other components present in the universe. Choosing $\sigma^2>3(1+w)$ leads to a scaling solution with $\Omega_\phi=3(1+w)/\sigma^2$ [@copeland]. (See also [@weller2], who use the exponential potential as a dark energy model.) The requirement that the present day dark energy density is $\Omega_\phi \sim 0.7$ is hard to reconcile with the scaling solution at early times, since in this case it follows that $\Omega_\phi=4/\sigma^2$, whilst big bang nucleosynthesis requires that the dark energy density in the early universe is very small [@bean].
In this Section we focus our discussion on models with $\sigma<\sqrt{2}$, which with an appropriate choice of $V_0$, can provide late-time acceleration with $\Omega_\phi \sim 0.7$ today (note that the late-time attractor $\Omega_\phi=1$ lies in the future). This choice of $\sigma$ also ensures that the energy density in the form of dark energy at the time of BBN is very small, because for early times the quintessence field is frozen and acts like a cosmological constant with an energy density similar to the observed dark energy density today.
The presence of neutrino coupling can potentially affect this result, as the coupled field begins to evolve at earlier redshifts ($z \sim 10^7$), however the fraction of the total energy stored in the scalar field at early times remains insignificant. Note that this choice of potential reduces to the cosmological constant case for a perfectly flat potential with zero coupling.
From the neutrino energy conservation equation (\[eq:nuenergy\]), and for our choice of $m_\nu(\phi)$ and $V(\phi)$, one can see that the dynamics of the scalar field can be described by an [*effective potential*]{} $$V_{\rm eff}=V(\phi) +
\left(\tilde{\rho}_\nu-3\tilde{p}_\nu\right)e^{\beta \phi},$$ where $\tilde{\rho}_\nu \equiv \rho_\nu e^{-\beta \phi}$ and $\tilde{p}_\nu \equiv p_\nu e^{-\beta \phi}$ are independent of $\phi$. It can be shown that the effective potential will only have a minimum when $\beta \sigma>0$. For the neutrinos we have numerically evaluated the integrals (\[eq:density\]) and (\[eq:pressure\]), which then have been used in the Klein–Gordon eq. (\[eq:kleingordon\]) to find the evolution of the scalar field.
Figure 1 shows some typical examples of how the coupling of the neutrinos to the scalar field causes the mass of the neutrinos to evolve with time. Deep within the radiation dominated epoch, at times when the neutrinos are highly relativistic, the scalar field is Hubble damped and therefore the neutrino mass is (almost) constant. For quintessence models, $\dot \phi$ is at most of order $H$, and therefore for relativistic species the coupling term in equation (\[eq:nuenergy\]) is clearly suppressed relative to the Hubble damping term. As the universe expands the neutrinos cool and become non–relativistic at a temperature corresponding to the neutrino mass. Hence, the extra coupling terms in equations (\[eq:nuenergy\]) and (\[eq:kleingordon\]) become more and more important, allowing energy to be exchanged between the neutrinos and the scalar field. This interaction causes the scalar field, and hence the neutrino mass, to evolve.
It is important to note the two different types of behaviour seen in Figure 1 for the evolution of the neutrino mass. For models which have $\beta \sigma>0$ the effective potential possesses a minimum, and after some time the field passes through this minimum, slows down, stops and eventually rolls back towards the minimum. For models which do not posses an effective minimum $\dot{\phi}$ is always negative, and the scalar field will continue to roll down the effective potential unimpeded. We compare the behaviour of our light scalar field with the heavy acceleron field used in [@fardon1] - in their model the scalar field sits in the effective minimum of its potential for most of the time during the cosmic history, and it is the evolution of the effective minimum which drives the dynamics of the neutrino mass. As discussed in [@fardon1], the mass of their neutrinos increases as the universe expands, whereas in our model the neutrinos are heavier in the past and become lighter (although as we will discuss in Section IV, suitable choices of coupling and potential can realize coupled neutrino–quintessence models with neutrinos which are lighter in the past).
For the model described in this section the coupled neutrinos are heavier in the past than uncoupled neutrinos, which implies that the energy density stored in the neutrinos is larger than would normally be expected. This means that the evolution of the neutrino density parameter $\Omega_\nu$ depends on the evolution of the neutrino mass, which in turn depends on the choice of the coupling $\beta$ and the slope of the potential $\sigma$. This can be seen from Figure 2. The coupling of neutrinos to dark energy significantly alters the evolution of the cosmological background. In particular it can be seen that the extra energy stored in the neutrinos in the past can alter the redshift of matter–radiation equality.
Typically one expects non–relativistic neutrinos to behave in a similar manner to CDM, however the interaction between the neutrinos and the scalar field modifies the scaling behaviour of the non-relativistic neutrinos. It can be seen that the neutrino energy density dilutes away faster than that of CDM, which is especially notable for large values of coupling. The evolution of $\phi$ caused by the transfer of energy between the coupled neutrinos and scalar field also results in the energy density of the quintessence field becoming dominated by kinetic energy. Finally at a redshift of the order of unity the potential energy of the scalar field begins to dominate, and all other matter species decay away.
A final point we would like to raise is the fact the [*apparent*]{} equation of state measured by an observer is not given by $$w_\phi = \frac{p_\phi}{\rho_\phi},$$ where $\rho_\phi$ and $p_\phi$ are defined in eqns (\[rhop\]) and (\[pp\]). One of the usual assumptions made in the measurements of the dark energy equation of state using supernovae is that matter (dark, baryonic or neutrinos) is decoupled from dark energy. At low redshifts, all these components are assumed to scale like $a^{-3}$. This is clearly not the case with the coupled neutrinos here. It was pointed out in [@khoury], that the apparent equation of state is given by[^2] $$\label{eq:wapp}
w_{\rm ap} = \frac{w_\phi}{1-x},$$ with $$x = -
\frac{\rho_{\nu,0}}{a^3\rho_\phi}\left[\frac{m_\nu(\phi)}{m_\nu(\phi_0)}
- 1\right].$$ In this equation, the subscript $0$ denotes the quantities at the present epoch. We emphasise that this quantity is [*not*]{} the effective equation of state of dark energy, which is defined as $$\dot \rho_\phi + 3H \rho_\phi\left(1+ w_{\rm eff}\right) = 0,$$ while assuming that the neutrino density and neutrino pressure evolve according to Eq. (\[eq:nuenergy\]). Using the Klein-Gordon equation, one finds that the effective equation of state can be written as $$w_{\rm eff} = w_\phi +
\frac{\beta\dot\phi}{3H}\frac{\rho_\nu}{\rho_\phi}.$$
In Fig. 3, we plot the apparent equation of state $w_{\rm ap}$ as a function of redshift. Note that $w_{\rm ap}$ can be less than $-1$, as was pointed out in [@wandelt], [@amendola] and [@khoury] in the context of models with dark matter/dark energy interaction.
To conclude this part, even if dark energy couples only to a subdominant component such as neutrinos, the apparent equation of state can be less than $-1$, without introducing phantom fields. As it can be seen from Figure 3, the apparent equation of state might even cross the boundary $w=-1$.
Perturbation evolution
======================
Let us now turn our attention to the evolution of cosmological perturbations in our model. We adopt the conventions of Ma and Bertschinger [@Ma] and work in the synchronous gauge, taking the line element to be $$ds^2=-a^2d\tau^2+a^2\left(\delta_{ij}+h_{ij}\right)dx^idx^j.$$ (For a review of cosmological perturbation theory, see [@feldman], [@durrer] or [@giovan].)
As already mentioned in the last section, to fully describe the evolution of cosmological neutrinos, we must calculate their distribution function $f\left(x^i,p^i,\tau\right)$ in phase space. We can treat the neutrinos as collisionless particles throughout the period of interest, and hence we can find the neutrino distribution function by solving the collisionless Boltzmann equation [@Ma]
$$\label{eq:cbe}
\frac{\partial f}{\partial\tau} + \frac{d x^i}{d\tau }\frac{\partial
f}{\partial x^i} + \frac{dq}{d\tau }\frac{\partial f}{\partial q} +
\frac{d n^i}{d\tau }\frac{\partial f}{\partial n^i} = 0,$$
where the comoving momentum is $q_i = ap_i$. It is convenient to rewrite the comoving momentum in terms of its magnitude and direction: $q^i = qn^i$. The last term in equation (\[eq:cbe\]) is a second order quantity and will be neglected in the following linear perturbation formalism.
The path of a neutrino in spacetime is governed by the general action $$\label{eq:action}
S = - \int m_\nu(\phi) \sqrt{-ds^2},$$ which can be minimised to derive the neutrino geodesic equation[^3] $$P^0\frac{\partial P^\rho}{\partial \tau}+\Gamma^\rho_{\alpha\beta}P^\alpha
P^\beta = -m_\nu^2 \frac{d\ln m_\nu}{d\phi}\frac{\partial \phi}{\partial
x_\rho},$$ where $P^\mu$ is the proper momentum of the neutrino. Taking the zeroth component of this equation and using the relation $P^0 =
\epsilon a^{-2}$, one finds that the comoving three-momentum of the neutrinos is given by$^{3}$ $$\frac{dq}{d\tau}=-\frac{1}{2}q\dot{h}_{ij}n_i n_j.$$ This equation does not depend explicitly on the coupling or the scalar field perturbations. Following [@Ma], we write the phase space distribution of the neutrinos as a zeroth order distribution plus a small perturbation $$f(x^i,p_j,\tau)=f_0(q)\left[1+\Psi\left(x^i,q,n_j,\tau\right)\right].$$ Substituting this expression into the Boltzmann equation and performing a Fourier transformation, we find$^{3}$ $$\begin{aligned}
\label{eq:ftcbe}
\frac{\partial \Psi}{\partial \tau} &+& i\frac{q}{\epsilon}\left({\bf
k\cdot n}\right)\Psi \nonumber \\ &+& \frac{d\ln f_0}{d\ln q}
\left[\dot\eta-\frac{\dot h+6\dot \eta}{2}\left({\bf k\cdot
n}\right)^2\right]=0\end{aligned}$$ In this equation (and in eqs. (\[pertkg\]) and (\[denscont\]), given later), $\eta$ and $h$ are the standard scalar parts of the metric perturbation $h_{ij}$. It is clear that equation (\[eq:ftcbe\]) does not contain terms proportional to $d\ln
m_\nu/d\phi$. Therefore, the equations for the neutrino hierarchy derived in [@Ma] do not change$^{3}$. However, the expressions for the perturbed neutrino energy density and neutrino pressure, which will be calculated using $f$, are modified. The perturbed energy density is given by $$\delta\rho_\nu = \frac{1}{a^4}\int q^2dq\,d\Omega f_0 \left(\epsilon\Psi + \frac{d\ln
m_\nu}{d\phi}\frac{m_\nu^2a^2}{\epsilon}\delta\phi\right)$$ which can be written as $$\delta\rho_\nu=\frac{1}{a^4}\int q^2dq\,d\Omega\, \epsilon f_0(q)\Psi
+ \delta\phi \frac{d\ln m_\nu}{d\phi}(\rho_\nu-3p_\nu).$$ Similarly, the expression for the perturbation in the neutrino pressure is given by $$\delta p_\nu = \frac{1}{3a^4}\int q^2 dq\,d\Omega
f_0(q)\left(\frac{q^2}{\epsilon}\Psi - \delta \phi \frac{d\ln m_\nu}{d
\phi} \frac{q^2 m_\nu^2 a^2}{\epsilon^3} \right)\nonumber.$$ The expressions for the neutrino shear and energy flux remain unchanged as they do not depend explicitly upon $m_\nu$. Finally, the perturbed Klein-Gordon equation is given by: $$\begin{aligned}
\label{pertkg}
\lefteqn{\ddot{\delta \phi}+ 2H \dot{\delta \phi}+\left(k^{2} +
a^{2}\frac{d^{2}V}{d\phi^{2}}\right)\delta \phi+\frac{1}{2}\dot{h}
\dot{\phi}=} \\ \nonumber & &-a^2 \left[\frac{d\ln
m_\nu}{d\phi}(\delta\rho_\nu-3\delta p_\nu)+\frac{d^{2} \ln
m_\nu}{d\phi^{2}}\delta\phi(\rho_\nu-3 p_\nu)\right].\end{aligned}$$
To calculate the temperature anisotropy spectrum and matter power spectrum we apply these modifications to CAMB [@camb]. This code calculates the linear cosmic background anisotropy spectra by solving the Boltzmann equation which governs the evolution of the density perturbations, and integrating the sources along the photon past light cone. To ensure the accuracy of our calculations, we directly integrate the neutrino distribution function, rather than using the standard velocity weighted series approximation scheme. We do not consider lensing effects, nor tensor contributions.
The results of the neutrino-dark energy coupling on the temperature anisotropy spectrum can be seen in Figure 4. The most obvious modifications to the anisotropy spectrum occur for large angular scales, with $\ell<100$, although for some choices of parameters the positions and relative heights of the peaks are also affected. We generally observe an increase in power for $10<\ell<100$, whilst for $\ell<10$ we find either an excess or reduction in power depending upon our choice of parameters. Note that this is in marked contrast to models of coupled CDM, where an increase in power on large scales is usually observed [@domenico]. For the models where the neutrinos were much heavier in the past than today, we also observe a slight shift in the acoustic peaks and a change in their relative amplitudes.
For scales larger than a degree ($\ell<100$), the dominant contribution to the anisotropy spectrum is the Integrated Sachs-Wolfe Effect (IWS). This arises due to the evolution of the gravitational potentials along the photon path from the surface of last scattering. The modification to the cosmological background arising from the neutrino coupling can have a significant effect upon the evolution of the perturbations. In particular there is a larger energy density in neutrinos in coupled models during the transition period when the neutrinos become non–relativistic. As a result, the intermediate regime between radiation and matter domination is prolonged, and so the evolution of the gravitational potentials are significantly modified. The evolution of the sum of the metric perturbations $\Phi$ and $\Psi$ is shown in Figure 5. The modifications to the behaviour of the metric perturbations for the different models is immediately apparent.
For very large scales ($\ell\leq20$) anisotropies arise primarily from the late time Integrated Sachs-Wolfe effect (ISW), which is caused by the evolution of the metric perturbations for redshifts in the range $0<z<2$. In particular, $\rho_\phi$ and $\rho_\nu$ as well as the equation of state of dark energy affect the late time behaviour of cosmological perturbations. As mentioned above, the evolution of the scalar field is influenced by the presence of the coupling to the neutrinos and hence the equation of state of dark energy depends upon $\beta$. Likewise, the clustering properties of dark energy depends on the neutrino coupling (see [@weller] for a discussion on the clustering of dark energy and its impact on the CMB). The neutrinos will generally tend to fall into the potential wells of dark matter, although at a rate slightly dependent on the coupling to the scalar field. The scalar field itself will cluster together with the neutrinos and thereby affecting the gravitational potential.
Let us turn our discussion to the evolution of neutrino perturbations. Figure 6 shows the effects of neutrino coupling on the matter power spectrum. Here we typically observe damping, and our results appear similar to standard models of CDM and hot dark matter, where a similar reduction in power could be achieved with a heavier neutrino mass.
We can use the perturbed part of the energy momentum conservation equation for the coupled neutrinos $$T^\mu_{~~\gamma;\mu} =\frac{d \ln m_\nu}{d\phi}
\phi_{,\gamma}T^{\alpha}_{~\alpha}$$ to calculate the evolution equations for the neutrino perturbations ($T^{\alpha}_{~\alpha}$ stands for the trace of the neutrino energy momentum tensor and the semicolon denotes the covariant derivative). Taking $\gamma=0$ we derive the equation governing the evolution of the neutrino density contrast , $\delta_\nu \equiv \frac{\delta
\rho_\nu}{\rho_\nu}$ whilst taking $\gamma=i$ (spatial index) yields the velocity perturbation equation $\theta_\nu \equiv
ik_{i}v^{i}_{\nu}$, with the coordinate velocity $v^{i}_{\nu} \equiv
dx^{i}/d {\tau}$: $$\begin{aligned}
\label{denscont}
\dot{\delta}_\nu &=& 3\left(H+\beta
\dot{\phi}\right)\left(w_\nu-\frac{\delta p_\nu}{\delta
\rho_\nu}\right)\delta_\nu - \left(1+w_\nu\right)\left(\theta_\nu +
\frac{\dot{h}}{2}\right)\nonumber \\ &+&
\beta\left(1-3w_\nu\right)\dot{\delta
\phi}+\frac{d\beta}{d\phi}\dot{\phi}\delta\phi\left(1-3w_\nu\right),\end{aligned}$$
$$\begin{aligned}
\dot{\theta_\nu} &=& -H(1-3w_\nu)\theta_\nu -
\frac{\dot{w_\nu}}{1+w_\nu}\theta_\nu+\frac{\delta p_\nu / \delta
\rho_\nu}{1+w_\nu}k^{2}\delta_\nu \nonumber \\ &+&
\beta\frac{1-3w_\nu}{1+w_\nu} k^{2} \delta
\phi-\beta(1-3w_\nu)\dot{\phi}\theta_\nu - k^{2} \sigma_\nu.\end{aligned}$$
The variable $\sigma_\nu$ represents the neutrino anisotropic stress and we have used the more general definition $\beta = d\ln
m_\nu/d\phi$, which in general might be not constant. Furthermore the neutrino equation of state is given by $w_\nu \equiv p_\nu /
\rho_\nu$.
It is the presence of the additional coupling terms in these expressions for the growth of the neutrino density and velocity perturbations, as well as the modifications to the evolution of the cosmological background, which alters the behaviour of the neutrino perturbations in comparison with the standard uncoupled case.
\[fig:denconk0p1\]
Figure 7 shows the evolution of the neutrino and CDM density contrasts, comparing the uncoupled model with an extreme coupled case with $\beta=-0.79$, for which the mass of the neutrinos is $m_\nu
\approx 2.5$ eV at $z\ge 1400$ but $m_\nu=0.3$ eV today. Deep inside the radiation dominated epoch the neutrinos are highly relativistic and their density contrast grows in a similar manner to radiation. In the uncoupled models the growth of the density perturbations of the neutrinos makes a transition to matter–like behaviour once the neutrinos become non–relativistic. The neutrinos will fall into the CDM potential wells, which is the dominant component at recombination. At small wavelengths (large wavenumbers), the neutrinos undergo freestreaming, which prevents neutrinos from clustering at an arbitrary small scale. The freestreaming length scale after the neutrinos become non-relativistic can be estimated to be [@Ma] (reinstating Newton’s constant $G$) $$k_{\rm fs}(a) = \frac{4\pi G \rho a^2}{v_{\rm med}^{2}}$$ where $\rho$ is the background total density. The median neutrino speed is given by: $$v_{\rm med}=15 a^{-1} \left( \frac{m_\nu(a)}{10 \rm{~eV}} \right)
\rm{km~s^{-1}}.$$ Since the neutrino momentum decays like $a^{-1}$, the neutrino velocity behaves like $\left(a m_\nu(a) \right)^{-1}$, taking into account that the neutrino mass evolves with time. Freestreaming stops as soon as $k<k_{\rm fs}$, allowing the neutrino density contrast to grow. This behaviour can clearly be seen in Figure 7 for both the uncoupled and coupled cases. At around $z\approx 10^4$ the neutrinos become non–relativistic and start to freestream immediately, as can be seen from the oscillating behaviour of $\delta_\nu$. At this stage $k_{\rm fs}<k$. However, as soon as $k_{\rm fs}=k$ freestreaming stops, and $\delta_\nu$ can grow unimpeded. The case with neutrino-dark energy coupling differs from the uncoupled case since in the result shown the neutrinos are heavier in the past, so for a given redshift $k_{\rm fs}$ is larger. This means that freestreaming stops earlier than in the uncoupled case. This behaviour is apparent in Figure 7 (dashed lines), where we see that $\delta_\nu$ starts to grow earlier than in the uncoupled case (solid lines). The neutrino-coupling also has an effect on the growth of the density contrast itself since we observe that $\delta_\nu$ grows slower than in the uncoupled case. This is probably because the rate of gravitational infall of the neutrinos tends to be reduced by the presence of the much less clustered dark energy. Also, the coupling has a slight effect on the growth of the dark matter density contrast, which arises from the fact that the background evolution is modified.
Another choice of coupling and potential
========================================
So far we have restricted our discussion to one choice of quintessence–neutrino coupling and one form for the dark energy potential. At this stage, the reader might wonder whether the results obtained so far are simply due to our choice of potential and coupling. For a scalar field with standard kinetic term, the exponential potential is not a favored model for a quintessential potential, since the initial value of the scalar field has to be fine–tuned to obtain scalar field domination today [@copelandwands]. The interesting alternative possibility of a global attractor unfortunately does not seem to be viable due to the large perturbation growth [@domenico]. The coupling of the scalar field to neutrinos does not cure the fine-tuning problem of the exponential potential.
For our second form of coupling we choose $$m_\nu (\phi) = M_0 e^{\beta\phi^2},$$ which was also recently used in a model with dark matter/dark energy coupling in [@massimo]. The effect of this choice is that the coupling function $d\ln m_\nu/d\phi$ becomes [*field–dependent*]{}, whereas it has been constant so far. Depending on how the field evolves with time, the coupling can either grow or become smaller during the cosmic history. Field–dependent couplings are not uncommon in higher–dimensional theories and can appear in brane–world theories, see for example [@brax1] or [@brax2].
For the potential, we choose an inverse power-law potential, which is a well–motivated candidate for a quintessence field (see e.g. [@steinhardt] and [@binetruy]). To be concrete, we use $$V(\phi) = \frac{M^6}{\phi^2}.$$ With these choices for the potential and coupling, the effective minimum will exist if $\beta>0$.
The results for the neutrino–mass evolution, the apparent equation of state and the CMB anisotropy power-spectrum are shown in Figures 8, 9 and 10.
The biggest difference to the case of a purely exponential coupling is that for positive $\beta$ the neutrinos are lighter in the past, as can be seen from Figure 8. Thus, with a convenient choice of potential and coupling, the neutrinos can become heavier as the universe expands. In the case of a negative $\beta$, the effective potential does not have a minimum and the neutrinos become lighter as the universe expands. The results for the apparent equation of state are shown in Figure 9. The results are similar to the ones found in Section II: $w_{\rm ap}$ can be smaller than $-1$ and can cross the boundary of the cosmological constant with $w=-1$. As it can be seen in Figure 9, the apparent equation of state varies substantially in the redshift range $z=0-2$ if $\beta$ is non-zero. A strongly, however, varying equation of state is not preferred by the data. Finally, the effects on the CMB anisotropies are similar to the ones found in Section III, as can be seen from Figure 10. The only visible deviation from $\beta=0$ is the case with negative $\beta$, in which a reduction of power at low multipoles can be observed. The cases with positive $\beta$ can not be distinguished from the uncoupled case. The reason is that the neutrino density is smaller in the past than in the uncoupled case for this choice of potential and coupling. Hence, neutrinos are less important for the dynamics of the universe, and their imprint upon the CMB is correspondingly reduced.
In essence, the physical explanations of the model presented in our earlier paper [@us] and in Sections II and III remain valid even for other choices of the potential and couplings, since they show how to relate the general behaviour of a dynamical neutrino mass to the cosmological evolution.
Parameter Constraints
=====================
In the earlier sections we demonstrated that models of coupled dark energy and neutrinos could produce a detectable signature in cosmological surveys. Indeed, the modifications to the background evolution (mainly to the dark energy equation of state), temperature anisotropy spectrum and matter power spectrum should allow us to constrain our model using current data sets.
We perform our likelihood analysis using CosmoMC [@cosmomc]. This program uses a Markov–Chain Monte–Carlo (MCMC) engine to efficiently explore the cosmological parameter space. Typically we run five chains for each simulation, with no less than 35,000 samples per chain. We perform the usual convergence checks on the individual chains to ensure that the chains have fully sampled our parameter space. As well as visually confirming that the individual chains converge, we check the Gelman and Rubin R statistic (variance of chain means/mean of chain variances) for each parameter, and ensure that the Raftery and Lewis convergence diagnostic is satisfied.
We take advantage of the wide range of cosmological data which is currently available to constrain our model. The CMB temperature anisotropy spectrum is constrained using WMAP [@wmap1] [@wmap2], CBI [@cbi], ACBAR [@acbar] and VSA [@vsa] datasets. The neutrino–dark energy coupling can also affect the formation of large scale structure which is sensitive to the neutrino mass, and so we use data from the Sloan Digital Sky Survey [@sdss] to further constrain our model. Data from the Supernova Cosmology Project [@supernova] can also be used to constrain the equation of state of dark energy, and thus place further constraints on our model.
We choose to perform the data analysis using our usual choice of potential and coupling, namely $V(\phi)=V_0e^{-\sigma\phi}$ and $m_\nu(\phi)=M_0e^{\beta\phi}$. We choose to focus on these potentials because they embody the typical behaviour observed for most models of coupled neutrinos, and they easily reduce to the standard $\Lambda$CDM case ($\beta=\sigma=0$). These potentials also have the advantage that the initial choice of $\phi_i$ does not affect the evolution of the cosmology, as changes to the initial choice of $\phi$ are equivalent to re-scalings of the mass parameters $M_0$ and $V_0$. For general choices of potentials and couplings this useful degeneracy does not exist, as the neutrino–dark energy coupling can severely restrict the range of attractor solutions. Consequently the increased number of fine-tuned free parameters required for these models would compromise the goodness of fit compared to simpler models.
Throughout our analysis we assume a flat universe, with $\Omega=1$. Initially we use the standard parameterization for our cosmological model, and vary the following parameters: $\Omega_{b} h^2$, $\Omega_{\rm CDM}h^2$, $h$, $z_{\rm re}$ (the redshift of reionisation), $\Omega_\nu h^2$, $n_s$ (the spectral index), $10^{10}A_s$ (the initial scalar perturbation amplitude) and the dark energy parameters $\sigma$ and $\beta$. We show the results from this initial analysis is Figure \[betasigmafnu\].
Clearly Figure \[betasigmafnu\] shows that current cosmological data places no constraints on our new coupling parameters. It is well known that the current best fit analysis of cosmological data can only place an upper limit on the mass of the neutrino, and that massless neutrinos are not excluded by most cosmological data sets. However the results presented so far in this paper require that the neutrinos having a significant mass; indeed for low mass neutrinos the effects observed in this paper become largely insignificant. Despite this, as we will show later, the strength of the neutrino coupling can still be constrained by the requirement for later time acceleration.
Current cosmological data requires that the universe contains approximately 70% of dark energy, 30% of dark matter and some minor quantities of baryons and neutrinos. In the dynamics of our model there could exist *only* one critical point able to guarantee such proportions in its vicinity (see e.g. [@domenico]). When exactly reached, this point is characterized by the total domination of the scalar field and exists only for $\lvert \sigma \rvert < \sqrt{6} \approx
2.5$. Indeed our computed likelihood will be shown to be contained in such boundaries. A preferred range for the parameter $\beta$ is more difficult to predict, however, because of the previously described effects induced on the background and perturbation at different stages of the evolution, although it is clear that very large values of $\beta$ will not be favoured by the data. But it is very important to emphasize that values of $\beta$ of the order of unity are perfectly acceptable.
Another consideration when choosing the scalar field parameters is the BBN constraint resulting from the early time modification of the scale factor evolution due to the presence of dark energy [@copeland; @bean]. As discussed earlier, the neutrino coupling to dark energy in our model does not modify the energy density of the dark energy at the time of BBN (when the neutrinos are highly relativistic, and the coupling terms are negligible). For a quintessence potential with a scalar field dominated late time attractor the BBN constraints of [@bean] are easily satisfied. We have confirmed this numerically for our coupled models (for instance, in the case of $\beta=1$, $\Omega_{\phi}\sim 10^{-25}$ at $z\sim 10^8$). Note that, in principle, our parameter space includes both the late-time attractor and scaling solutions (that evolve like radiation or matter and which would provide a non-negligible contribution to the energy density at the time of BBN). We find, however, that the scaling solution for the exponential potential is already strongly disfavored by the observations that we have used for our analysis and therefore additional BBN constraints would not modify our findings.
Recent works [@bridlexray] have used data from X-ray clusters to reduce the uncertainties on the lower bound for the neutrino mass, finding a value for the neutrino masses of $\sum
m_\nu=0.56^{+0.30}_{-0.26}$ eV. Notice though that the upper bounds could in fact go up to $\sum m_{\nu}=2$ eV, depending on the datasets used and the assumed priors [@elgaroy]. Measurements of atmospheric neutrino oscillations suggest that there is at least one neutrino species with $m_\nu>0.05$ eV [@massiveneutrinos1]. The Heidelberg-Moscow experiment, searching for neutrino-less double beta decay claim detection of electron neutrino mass $m_{\nu_e}$ between $0.2$ eV and $0.6$ eV, with best fit $m_{\nu_e}=0.36$ eV [@moscow].
We therefore choose to perform our analysis using two values of the neutrino mass today, $m_\nu=0.2$ eV and $m_\nu=0.3$ eV, to investigate whether models of neutrino–dark energy coupling could in principle be constrained if neutrinos were independently confirmed to have a significant ($m_\nu
\gtrsim 0.1$ eV) mass, consistent with current experiments measuring the neutrino mass.
By choosing to fix the value of the neutrino mass today, we are required to specify the current value for the energy density stored in neutrinos and value of the Hubble constant as the neutrino mass, the critical energy density in neutrinos and the Hubble constant are related via the usual formula $$\Omega_\nu h^2 = \frac{\sum m_\nu}{93.2\text{~eV}}.$$
We also choose to fix the value of $\Omega_b h^2$ as we do not expect the behaviour and constraints of the baryon energy density to be significantly modified by our neutrino–dark energy coupling, as the observed effects on the anisotropy spectrum are largely limited to relatively low multipoles.
The values for $H_0$ and $\Omega_b h^2$ can be determined independently from the cosmological data used in our MCMC analysis. The value of $H_0=72~\rm{km~s^{-1}~Mpc^{-1}}$ can be obtained from the best fit of the Hubble Space Telescope Key Project [@hst], whilst the baryon density parameter $\Omega_b h^2= 0.022$ is favoured by Big Bang Nucleosynthesis models [@bbn]. We are therefore left with a cosmological model requiring 6 parameters: $\beta$, $\sigma$, $\Omega_\text{CDM}h^2$, $z_{\rm re}$, $A_s$ and $n_s$.
The results of the MCMC analysis for neutrinos with a mass today of $m_\nu=0.2$ eV can be found in Table \[tab:lightnu\], whilst the results for $m_\nu=0.3$ eV are given in Table \[tab:heavynu\]. We quote the marginalised probability distributions and confidence intervals. Figures \[fig:betasigmaslow\] & \[fig:betasigma\] show the 2D probability distributions for the $m_\nu=0.2$ eV and $m_\nu=0.3$ eV models respectively.
------------------------- ------------------ --------------- ---------------
Parameter Mean 68% 95%
likelihood interval interval
$\Omega_\text{CDM} h^2$ $0.102\pm 0.004$ $0.099-0.106$ $0.096-0.110$
$z_\text{re}$ $17.6\pm 3.7$ $16.0-19.8$ $10.7-23.2$
$\sigma$ $0.43\pm 0.32$ $0.29-0.60$ $0.13-0.95$
$\beta$ $0.75\pm 0.64$ $0.64-0.98$ $0.11-1.18$
$n_s$ $0.96\pm 0.01$ $0.95-0.97$ $0.93-0.99$
------------------------- ------------------ --------------- ---------------
: Marginalised parameter constraints for our 6 parameter model with fixed $m_\nu=0.2$ eV, $\Omega_b h^2=0.022$ and $h=0.72$. For this model we find $\chi^2/\text{dof} = 1570.1/1459$. This compares with a $\chi^2/\text{dof} = 1610.1/1461$ for a best-fit $\Lambda$CDM model using the same parameter set with $\sigma=\beta=0$.[]{data-label="tab:lightnu"}
------------------------- ------------------- --------------- ---------------
Parameter Mean 68% 95%
likelihood interval interval
$\Omega_\text{CDM} h^2$ $0.100 \pm 0.003$ $0.097-0.104$ $0.094-0.107$
$z_\text{re}$ $20.5\pm 3.1$ $19.4-22.0$ $15.0-25.1$
$\sigma$ $0.52\pm 0.29$ $0.40-0.67$ $0.00-0.97$
$\beta$ $0.62\pm 0.21$ $0.58-0.74$ $0.15-0.86$
$n_s$ $0.97\pm 0.01$ $0.96-0.99$ $0.95-1.00$
------------------------- ------------------- --------------- ---------------
: Marginalised parameter constraints for our 6 parameter model with fixed $m_\nu=0.3$ eV, $\Omega_b h^2=0.022$ and $h=0.72$. In this case we find $\chi^2/\text{dof} = 1593.7/1459$. This compares with a $\chi^2/\text{dof} = 1636.8/1461$ for a best-fit $\Lambda$CDM model using the same parameter set with $\sigma=\beta=0$.[]{data-label="tab:heavynu"}
As expected, the neutrino coupling has little effect on the value of $\Omega_{CDM} h^2$ as the peak structure of the temperature anisotropy spectrum is largely unaffected. The value found for $n_s$ is also within the usual range for parameter analysis of cosmological models which neglect tensor contributions.
For both the $m_\nu=0.2$ eV and $m_\nu=0.3$ eV models we find that non-zero values of neutrino coupling strengths are preferred by the data. We also see that for these models a non–zero value for $\sigma$ is preferred over the usual cosmological constant, although $\sigma=0$ is not excluded at the 68% level. This is not surprising since we have seen that the equation of state of the dark energy for some choices of parameters in our coupled models is entirely consistent with the preferred value of $w_{ap}\sim -1$ found from supernova surveys. It is clear that models with heavier neutrinos allow stronger constraints to be placed upon the strength of the coupling. Indeed, for the 0.3 eV neutrinos we find that neutrino–dark energy coupling is preferred at the 1 sigma confidence level. This is to be expected as a larger neutrino mass today is equivalent to a higher energy density in neutrinos, and so any modification to the neutrino evolution will have a larger impact on the CMB and LSS for models with greater densities of neutrinos. In particular we have seen that there exist a range of non-zero $\beta$ values capable of reducing power at low CMB multipoles. This last effect is most probably the cause for the relative peak in the likelihood for $\beta$ of order unity. Furthermore a sharper drop at large values of $\beta$ is observed in the likelihood most likely to limit the excessive growth at multipoles $10 < \ell <100$.
It is however important to make clear that these constraints rely upon the assumption that the neutrino mass is known, and that the neutrinos have a mass $m_\nu \gtrsim 0.1$ eV. Although this assumption is consistent with current neutrino experiments, we can only make the statement that should the mass of the neutrino be found to be greater than $0.1$ eV, then cosmological data can be used to constrain the strength of any neutrino–dark energy coupling; indeed we find that there is some evidence that the existence of such a coupling is actually preferred by current cosmological data over the standard $\Lambda$CDM cosmology.
Conclusions
===========
We have investigated models of dark energy which couple a quintessence scalar field to massive neutrinos. In these models, dark energy and neutrinos are coupled such that the neutrino masses become functions of the scalar field. The effects of such models on the cosmological background evolution, on the cosmic microwave background anisotropies, and on the formation of large scale structures were analyzed. Additionally, we have also performed a likelihood analysis on the parameter space of such theories.
We have focused on two specific models: In the first, the coupling between neutrinos and dark energy is constant and the quintessential potential is an exponential. The second model, which is better motivated from the particle physics point of view, has a neutrino–coupling which depends on the quintessence field (hence changes with time), whilst the scalar field has a power-law potential. In spite of some specific differences between these two models (such as the energy density stored in the scalar field at recombination for example), the effects of the coupling on the CMB anisotropies and on the matter power spectrum are nevertheless explainable by the basic mechanisms that we have identified earlier. Namely, the coupling modifies the background history and induces an ISW contribution to the CMB spectrum; the matter power spectrum is modified by the magnitude of the neutrino mass during structure formation. Given the generality of these explanations, the conclusions drawn from this investigation could probably be applied to any similar model with a neutrino-dark energy coupling.
It is important to note that in our models, the dark energy sector is described by a *light* scalar field, with a mass which is at most of order $H$. This is in contrast to previous models [@fardon1] in which the mass of the scalar field is much larger than $H$ for most of its history. The latter can have significant effects upon the behaviour of the neutrinos and the growth of their perturbations, and which is difficult to reconcile with current astronomical data [@zalda].
Solving the collisionless Boltzmann equation for the neutrinos, we have investigated the relativistic and non-relativistic regimes and the transition period in between. Initially the neutrinos are highly relativistic, and during this period the quintessence field is frozen. The mass of the neutrinos therefore remains constant. As the neutrinos become non-relativistic they begin to exchange energy with the quintessence field via the coupling term. At a temperature scale comparable to the neutrino mass, the neutrinos become non-relativistic, whilst the quintessence field is dominated by kinetic energy. It is at this point that the neutrino mass begins to evolve significantly. The details of this behaviour and evolution depends on the choice of the coupling $\beta$ and the potential parameter $\sigma$. In fact, the masses of the neutrinos can be heavier or lighter in the past depending on the choice of potential and coupling parameters.
The coupling of neutrinos to dark energy slightly alters the evolution of the cosmological background. It was found that similarly to models with a dark matter/dark energy interaction, the apparent equation of state measured with Type Ia Supernovae at high redshift can be smaller than $-1$, without introducing phantom fields, and might even cross the boundary $w=-1$.
The most obvious modifications to the CMB anisotropy spectrum occur for large angular scales, with $\ell<100$, where the dominant contribution to the anisotropies is generated by the Integrated Sachs-Wolfe Effect (IWS). This arises due to the evolution of the gravitational potentials along the photon path from the surface of last scattering. The modification to the cosmological background arising from the neutrino coupling can also have a significant effect upon the evolution of the perturbations. We generally observe an increase in power for $10<\ell<100$, whilst for $\ell<10$ we find either an excess or reduction in power depending upon our choice of parameters. For the models where the neutrinos were much heavier in the past than today, we also observe a slight shift in the peaks and a modification in their relative amplitude.
The matter power spectrum exhibits free-streaming damping even in the presence of dark energy–neutrino coupling. However, since the damping scale is mainly dependent on the value of the neutrino mass at the end of their relativistic stage, our results appear similar to the standard models with CDM and hot dark matter in which the mass is fixed at the relativistic plateau. It is obvious that the mass infered from the damping of the matter power spectrum is, in general, different from the neutrino mass measured with experiments in the laboratory.
We performed a likelihood analysis using SNIa, CMB and LSS datasets. Initially, we used the standard parameterization for our cosmological model, characterized by exponential dependence of the dark energy potential and neutrino mass on the scalar field. For a flat universe we varied all of the matter parameters, the Hubble constant, the initial power spectrum spectral index and amplitude and the instantaneous reionization parameter $z_{\rm re}$. As expected, the cosmological data did not place strong constraints on our new coupling parameters. This is no surprise, since it is well known that the current best fit analysis of cosmological data can only place an upper limit on the mass of the neutrino, and a zero neutrino mass is not excluded by most cosmological data sets. An interesting outcome was that couplings of order unity are perfectly acceptable with the actual data.
To proceed, we chose to perform an analysis using two values of the neutrino mass today, $m_\nu=0.2$ eV and $m_\nu=0.3$ eV, to investigate whether models of neutrino–dark energy coupling could in principle be constrained if neutrinos were independently confirmed to have a significant mass ($m_\nu \gtrsim 0.1$ eV), consistent with current experiments.
For both the $m_\nu=0.2$ eV and $m_\nu=0.3$ eV models we found that non-zero values of neutrino coupling strengths of order unity are preferred by the data. We also saw that for these models a non–zero value for $\sigma$ is preferred over the usual cosmological constant, although $\sigma=0$ is not excluded at the $68\%$ level. Models with heavier neutrinos allow stronger constraints to be placed upon the strength of the coupling. Indeed, for the $0.3$ eV neutrinos we found that neutrino–dark energy coupling is preferred at the $1$ sigma confidence level.
One should note that these constraints rely upon the assumption that the neutrino mass is known, and that the neutrinos have a mass $m_\nu
\gtrsim 0.1$ eV. Although this assumption is consistent with current neutrino experiments, we can only make the statement that should the mass of the neutrino be found to be greater than $0.1$ eV, then current cosmological data can be used to constrain the strength of any neutrino–dark energy coupling.
We are grateful to S. Bridle, O. Elgaroy, H. K. Eriksen, F.K. Hansen, D. Hooper, A. Melchiorri, J. Silk, C. Skordis and J. Weller for useful discussions. We also thank A. Lewis for allowing us to use his CAMB quintessence module. AWB is supported by PPARC. DFM acknowledge support from the Research Council of Norway through project number 159637/V30. DTV acknowledges a Scatcherd Scholarship.
Erratum {#sec:erratum .unnumbered}
=======
The original version of this paper contained a typo and a mistake, as noted by [@keum]. This section contains the corrected equations and Figures, as published in our Erratum [@erratum].
The geodesic equation (23) contains a typo and should read $$P^0\frac{d P^\rho}{d \tau}+\Gamma^\rho_{\alpha\beta}P^\alpha
P^\beta = -m_\nu^2 \frac{d\ln m_\nu}{d\phi}\frac{\partial \phi}{\partial
x_\rho},$$ A subtle error occurred in eq. (24), which should read $$\frac{dq}{d\tau}=-\frac{1}{2}q\dot{h}_{ij}n_i n_j - a^2 \frac{m^2_\nu}{q}\frac{\partial\ln m_\nu}{\partial \phi}\frac{\partial\phi}{\partial x^i}\frac{\partial x^i}{d\tau}.$$ It follows that a scalar field dependent term should be included in the Boltzmann equation (26), giving: $$\begin{aligned}
\frac{\partial \Psi}{\partial \tau} &+& i\frac{q}{\epsilon}\left({\bf
k\cdot n}\right)\Psi \nonumber + \frac{d\ln f_0}{d\ln q}
\left[\dot\eta-\frac{\dot h+6\dot \eta}{2}\left({\bf k\cdot
n}\right)^2\right]\\ &=&i\frac{q}{\epsilon}\left({\bf k\cdot n}\right)k\frac{a^2 m^2_\nu}{q^2} \frac{\partial \ln m_\nu}{\partial \phi}\frac{d\ln f_0}{d\ln q}\delta\phi\end{aligned}$$ Therefore, the dipole equation for the neutrino hierarchy derived in [@Ma] is subject to a change represented by a new term once again dependent on the scalar field: $$\dot \Psi_1=\frac{1}{3}\frac{q}{\epsilon}k\left( \Psi_0 - 2\Psi_2 \right)
-\frac{1}{3}\frac{q}{\epsilon}k\frac{a^2 m^2_\nu}{q^2}\frac{\partial \ln m_\nu}{\partial \phi}\frac{d\ln f_0}{d\ln q}\delta\phi$$
This modification will have an effect on the ISW effect, which is less pronounced than reported in our paper. The corrected evolution of the metric variables $\Phi+\Psi$ is shown in Figure \[fig:fixphipsi\].
This will effect the anisotropies in the CMB, which are shown in Figure \[fig:clsfixed\].
On the other hand, we do not find changes to the matter power spectrum or the evolution of the neutrino density contrast at the scale given in Fig. 7 in our original paper. At smaller scales we register small differences in the neutrino and scalar field density contrasts, which leads to the mentioned changes in the ISW effect. Note that the background evolution reported in [@us; @usprd] is not affected.
We do not attempt to redo the comparison of our model with data, as the new WMAP 3-year data have been published since [@wmap3], and these models where analysed in [@keum].
We are grateful to K. Ichicki for correspondence. The correct equations (2)-(4) have been derived in [@keum].
[99]{} D. Spergel et al., Astrophys.J.Suppl. [**148**]{}, 175 (2003) A. Riess et al., Astron. J. [**116**]{}, 1009 (1998) S. Perlmutter et al., Astrophys. J. [**517**]{}, 565 (1999) C. Wetterich, Nucl. Phys. B [**302**]{}, 668 (1988) B. Ratra and J. Peebles, Astrophys. Journ. Lett. [**325**]{}, 17 (1988) A. Albrecht and C. Skordis, Phys. Rev. Lett. [**84**]{}, 2076 (2000) C. Wetterich, Astron.Astrophys. [**301**]{}, 321 (1995) M. Doran and J. Jaeckel, Phys.Rev.D [**66**]{}, 043519 (2002) S.M. Carroll, Phys.Rev.Lett. [**81**]{}, 3067 (1998) J.A. Frieman, C.T. Hill, A. Stebbins and I. Waga, Phys.Rev.Lett. [**75**]{}, 2077 (1995) J.E. Kim and H.-P. Nilles, Phys.Lett.B [**533**]{}, 1 (2003) N. Kaloper and L. Sorbo, astro-ph/0511543 R. Peccei in: [*Sources and detection of dark matter and dark energy in the universe*]{}, ed. David B. Cline, Springer Verlag (2001) P.Q. Hung, hep-ph/0010126 M. Li, X. Wang, B. Feng and X. Zhang, Phys.Rev.D [**65**]{}, 103511 (2002) M. Li and X. Zhang, Phys.Lett. B[**573**]{}, 20 (2003) P. Gu, X. Wang and X. Zhang, Phys.Rev D [**68**]{}, 087301 (2003) R. Fardon, A.E. Nelson and N. Weiner, JCAP [**0410**]{}, 005 (2004) D.B. Kaplan, A.E. Nelson and N. Weiner, Phys.Rev.Lett. [**93**]{}, 091801 (2004) P. Gu and X.-J. Bi, Phys.Rev.D [**70**]{}, 063511 (2004) X. Bi, P. Gu, X. Wang and X. Zhang, Phys.Rev.D [**69**]{}, 113007 (2004) H. Bi, B. Feng, H. Li and X. Zhang, hep-ph/0412002 E.I. Guendelman and A.B. Kaganovich, hep-th/0411188 H. Li, Z. Dai and X. Zhang, Phys. Rev. D[**71**]{}, 113003 (2005) R.D. Peccei, Phys.Rev.D [**71**]{}, 023527 (2005) P.Q. Hung and H. Pas, Mod.Phys.Lett. A[**20**]{}, 1209 (2005) X.-F. Zhang, H. Li, Y. Piao and X.-M. Zhang, astro-ph/0501652 A.W. Brookfield, C. van de Bruck, D.F. Mota and D. Tocchini-Valentini, Phys. Rev. Lett. 96:061301,2006. V. Barger, P. Huber and D. Marfatia, hep-ph/0502196 M. Cirelli, M.C. Gonzalez-Garzia and C. Pena-Garay, Nucl.Phys.B [**719**]{}, 219 (2005) R. Horvat, astro-ph/0505507 R. Barbieri, L.J. Hall, S.J. Oliver and A. Strumia, hep-ph/0505124 N. Afshordi, M. Zaldarriaga and K. Kohri, Phys.Rev.D [**72**]{}, 065024 (2005)
E. J. Copeland, A. R. Liddle and D. Wands, Phys. Rev. D [**57**]{} (1998) 4686 L. Amendola, Phys. Rev. D [**62**]{} (2000) 043511 L. Amendola and D. Tocchini-Valentini, Phys. Rev. D [**64**]{} (2001) 043509 P. G. Ferreira and M. Joyce, Phys. Rev. D [**58**]{} (1998) 023503 R. Bean, S. H. Hansen and A. Melchiorri, Phys. Rev. D [**64**]{} (2001) 103508
R. Takahashi and M. Tanimoto, hep-ph/0507142 R. Fardon, A. Nelson and N. Weiner, hep-ph/0507235 M. Honda, R. Takahashi and M. Tanimoto, hep-ph/0510018 N. Weiner and K. Zurek, hep-ph/0509201 V. Barger, D. Marfatia and K. Whisnant, hep-ph/0509163 H. Li, B. Feng, J.-Q. Xia and X. Zhang, astro-ph/0509272 P.-H. Gu, X.-J. Bi and X. Zhang, hep-ph/0511027 T. Schwetz and W. Winter, hep-ph/0511177 M.C. Gonzalez-Garcia, P.C. de Holanda, R.Z. Funchal, hep-ph/0511093 P.-H. Gu, X.-J. Bi, B.-L. Young and X. Zhang, hep-ph/0512076 Y. Fukuda et al., Phys.Rev.Lett.[**81**]{}, 1158 (1998) Y. Ashi et al., Phys.Rev.Lett. [**93**]{}, 101801 (2004) E. Aliu et al., Phys.Rev.Lett. [**94**]{}, 081802 (2005) R. Kallosh, A. Linde, S. Prokushkin and M. Shmakova, Phys.Rev.D [**66**]{}, 123503 (2002) J. Weller and A. Albrecht, Phys.Rev.D [**65**]{}, 103512 (2002) S. Das, P.S. Corasaniti and J. Khoury, astro-ph/0510628 G. Huey and B.D. Wandelt, astro-ph/0407196 L. Amendola, M. Gasperini and F. Piazza, JCAP09(2004)014 C.-P. Ma and E. Bertschinger, Astrophys. Journ. [**455**]{}, 7 (1995) V.F. Mukhanov, H.A. Feldman and R.H. Brandenberger, Phys.Rep.[**215**]{}, 203 (1992) R. Durrer, J.Phys.Stud. [**5**]{}, 177 (2001) M. Giovannini, Int.J.Mod.Phys.D [**14**]{}, 363 (2005) A. Lewis, A. Challinor and A. Lasenby, Astrophys. Journ. [**538**]{}, 473 (2000) D. Tocchini-Valentini and L. Amendola, Phys.Rev.D [**65**]{}, 063508 (2002) J. Weller and A. Lewis, MNRAS [**346**]{}, 987 (2003); R. Bean and O. Doré, Phys.Rev.D [**69**]{}, 083503 (2004) E.J. Copeland, A.R. Liddle and D. Wands, Phys.Rev.D [**57**]{}, 4686 (1998) M. Pietroni, Phys.Rev.D [**72**]{}, 043535 (2005) Ph. Brax, C. van de Bruck, A.-C. Davis and C.S. Rhodes, Phys.Rev.D [**67**]{}, 023512 (2003) C.S. Rhodes, C. van de Bruck, Ph. Brax and A.-C. Davis, Phys.Rev.D [**68**]{}, 083511 (2003) P.J. Steinhardt, L.-M. Wang and I. Zlatev, Phys.Rev.D [**59**]{}, 123504 (1999) P. Binetruy, Phys.Rev.D [**60**]{}, 063502 (1999)
A. Lewis and S. Bridle, Phys.Rev.D [**66**]{}, 103511 (2002) L. Verde et al., Astrophys. J. Suppl. [**148**]{}, 195 (2003) G. Hinshaw et al., Astrophys. J. Suppl. [**148**]{} 135 (2003) A. C. S. Readhead et al., Astrophys. J. [**609**]{} 498 (2004) C-l. Kuo et al., Astrophys. J. [**600**]{} 32 (2004) C. Dickinson et al., astro-ph/0402498 M. Tegmark et al., Astrophys. J. [**606**]{} 702 (2004) S. Perlmutter et al, Astrophys. J. [**517**]{} 565 (1999) S.W. Allen, R.W. Schmidt and S.L. Bridle, Mon. Not. Roy. Astron. Soc. [**346**]{}, 593 (2003) O. Elgaroy and O. Lahav, JCAP [**0304**]{}, 004 (2003); O. Elgaroy [*et al.*]{}, Phys. Rev. Lett. [**89**]{}, 061301 (2002). L. Baudis [*et al.*]{}, Phys. Lett. B [**407**]{}, 219 (1997). W.L. Freedman et al, Astrophys. J. [**553**]{} 47 (2001) R.H. Cyburt et al, Phys.Lett. B[**567**]{}, 227 (2003) K. Ichiki and Y.-Y. Keum, arXiv:0705.2134
A. W. Brookfield, C. van de Bruck, D. F. Mota and D. Tocchini-Valentini, Phys. Rev. D [**76**]{} 049901 (2007) A. W. Brookfield, C. van de Bruck, D. F. Mota and D. Tocchini-Valentini, Phys. Rev. D [**73**]{}, 083515 (2006)
D. N. Spergel [*et al.*]{} \[WMAP Collaboration\], arXiv:astro-ph/0603449.
[^1]: In fact, such an assumption is quite natural and has no strong consequences to this work. In the mass regions detectable in astronomical observations, the three neutrino masses are nearly degenerate. Adding to that, cosmology is in leading order sensitive to $\sum m_{\nu}$.
[^2]: The authors of [@khoury] called this quantity an effective equation of state. However, we will define an effective equation of state below.
[^3]: Please note Errata at the end of this paper.
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- 'C. Lujan-Peschard,'
- 'G. Pagliaroli,'
- 'F. Vissani.'
title: 'Spectrum of Supernova Neutrinos in Ultra-pure Scintillators'
---
Introduction {#sec:intro}
============
A Core Collapse Supernova (SN) releases 99% of its total energy by emitting neutrinos of the six flavors. The capability to observe the electronic antineutrino component of this emission has already been proven by the detection of SN1987A neutrinos [@KII; @IMB; @Baksan]. The very large statistics that we will collect from the next galactic supernova will allow us to study the time dependence of the spectrum, specific features of the $\bar{\nu}_e$ luminosity and of its average energy [@SK; @IceCube]. The detection of the other neutrinos flavors, however, requires specific detectors and interactions, typically with smaller cross sections. This is true, in particular, for the non electronic component of the spectrum, that can be observed only through Neutral Current (NC) interactions.
During the last years a new generation of ultra-pure liquid scintillators, Borexino (BRX) [@2009BorexinoColl] and KamLAND (KAM) [@2003KamlandColl], have been operated, obtaining excellent results thanks to the unprecedented low background levels reached and the new sensitivity in the very low energy range, below 1 MeV. They have a particularly good physics potential for the detection supernova NC channels, and quite remarkably, the Elastic Scattering (ES) of (anti)neutrinos on protons [@beato]. It has been argued that the high statistics from this reaction should suffice to constrain the spectra of the non electronic component for a SN emission, already with the existing detectors [@dasg]. In view of the importance of this conclusion, we would like to reconsider it in this work.
The outline of this paper is as follows. First of all, we summarize the available information regarding SN neutrino detection in the existing ultra-pure scintillators, and calculate for each of them the total number of expected events as well as their spectral features. We consider the contributions of all neutrino interaction channels and obtain in this way the spectrum of events for a galactic supernova. In this way, we are in the position to evaluate which are the capabilities of the present generation of ultra-pure scintillators to identify and measure the different neutrino flavors.
Emission from a Standard Core Collapse Supernova
================================================
The aim of this work is to discuss an important question: what we can really see with the existing ultra-pure scintillators and to which extent we can distinguish the different neutrino flavors. With this purpose in mind, we will use very conservative assumptions on the emission model. We suppose that the energy radiated in neutrinos is $\mathcal{E}=3\times 10^{53}$ erg, which is a typical theoretical value that does not contradict what is found in the most complete analyses of SN1987A events [@ll; @paglia]. We also assume that the energy is partitioned in equal amount among the six types of neutrinos, that should be true within a factor of 2 [@Keil:2002in].
In agreement with the recent studies, e.g., [@Tamborra:2012ac], we consider quasi-thermal neutrinos, each species being characterized by an average energy $\langle E_i\rangle$ and including a mild deviation from a thermal distribution described by the parameter $\alpha=3$ for all flavors. Thus, the neutrino fluence differential in the neutrino energy $E$ is $$\Phi_i=\frac{\mathcal{E}_i}{4\pi D^2}\times \frac{E^\alpha e^{-E/T_i}}{ T_i^{\alpha+2} \Gamma(\alpha+2)} \ \ \
i=\nu_e,\nu_\mu,\nu_\tau,\bar{\nu}_e,\bar{\nu}_\mu,\bar{\nu}_\tau$$ where the energy radiated in each specie is $\mathcal{E}_i=\mathcal{E} f_i$, with $f_i=1/6$ in the case of equipartition, and the ‘temperature’ is $T_i=\langle E_i\rangle/(\alpha+1)$. In particular, the neutrino and antineutrino fluences relevant to NC interactions $$\Phi^{\mbox{\tiny SN}}_\nu=\Phi_{\nu_e}+2 \Phi_{\nu_\mu} \mbox{ and }
\Phi^{\mbox{\tiny SN}}_{\bar\nu}=\Phi_{\bar{\nu}_e}+2 \Phi_{\bar{\nu}_\mu}$$ since we suppose that the distribution of the 4 non-electronic species is identical.
The average energies are fixed by the following considerations: consistent with the simulations in [@Tamborra:2012ac] and with the findings from SN1987A [@ll; @paglia], we set the electron antineutrino average energy to $\langle E_{\bar{\nu}_e}\rangle=$12 MeV. For the average energy of the non-electronic species, that cannot be seriously probed with SN1987A [@paglia], we suppose that the non-electronic temperature is 30% higher than the one of $\bar{\nu}_e$: $\langle
E_x \rangle=15.6$ MeV, this is in the upper range of values, but still compatible with what is found in [@Keil:2002in]. For a comparison we will consider also the worst case in which the energies of the non electronic component is equal to the one of the $\bar{\nu}_e$, namely $\langle
E_x \rangle=12$ MeV as showed in very recent simulation [@Mueller:2014rna]. We calculate the electron neutrino average energy by the condition that the proton (or electron) fraction of the iron core in the neutron star forming is 0.4: this gives $\langle E_{\nu_e}\rangle=9.5$ MeV.
Note that the NC reactions are independent from neutrino oscillations, while for the CC interactions also considering only the standard oscillation scenario, the choice of the mass hierarchy has an important impact on the expectations. Thanks to the fact that $\theta_{13}$ is large (say, larger than about 1 degree) this means that, in normal mass hierarchy, the survival probability of electron neutrinos and antineutrinos are $|U_{e3}^2|$ and $|U_{e1}^2|$ respectively, whereas for inverted mass hierarchy, the two values become $|U_{e2}^2|$ and $|U_{e3}^2|$. Thus, the approximate numerical values that we will assume in the calculations are
$P_{\bar\nu_e\to \bar\nu_e}$ $P_{\nu_e\to \nu_e}$
---------- ------------------------------ ----------------------
Normal 0.7 0.0
Inverted 0.0 0.3
The value of $P_{\bar\nu_e\to \bar\nu_e}$ in the case of inverted hierachy means that what we measure as electronic antineutrinos in terrestrial detectors, are non-electronic antineutrinos at the emission in fact; thus, it has a particularly important impact on the interpretation of the data. We note also that these numerical values would imply that there are only little chances to probe the emission of electron neutrinos, which are, from the astrophysical point of view, the most important type of neutrinos emitted by a supernova.
In the following we will consider only the case of the normal mass hierarchy for definiteness and adding a bit of theoretical bias; recall however that this hypothesis is immaterial for the discussion of the neutral current events. The total fluences expected to reach the Earth under these assumptions are shown in Fig.\[fig1\] for a Supernova exploding at $10$ kpc from us.
{width="48.00000%"} {width="48.00000%"}
Interaction Channels
====================
In the scintillators and at SN energies, we need consider the several interaction processes:
#### **CC processes.**
Those involving electronic antineutrinos are
- Inverse Beta Decay (IBD), i.e. $\bar{\nu}_e+p\to n+e^+$;
- $\bar{\nu}_e + ^{12}C \to ^{12}B + e^+$;
while those involving electronic neutrinos are
- Proton Knockout $^{12}$C$(\nu, p e^-) ^{11}$C;
- $\nu_e + ^{12}C \to ^{12}N + e^-$.
#### NC processes.
We considered the following channels,
- ES on protons, $\stackrel{(-)}\nu p\to \stackrel{(-)}\nu p$;
- The 15.11 MeV de-excitation line of the $^{12}C$ nucleus, $\nu ^{12}C \to \nu ^{12}C^*$;
- The Proton Knockout $\stackrel{(-)}\nu+ ^{12}C\to \stackrel{(-)}\nu+ ^{11}B$.
Moreover, we consider the ES on electrons, that receives a contribution from both CC and NC. Let us discuss these reactions in detail.
Detailed description of the cross sections {#detailed-description-of-the-cross-sections .unnumbered}
------------------------------------------
The [**IBD**]{}, i.e., $\bar{\nu}_e+p\to n+e^+$, represents the main signal not only in water Cherenkov and also in scintillator detectors. It produces a continuous spectrum due to the positrons energy release. The approximated kinematic of this reaction connects the neutrinos energy with the detected energy through $E_\nu=E_d+Q-m_e$ where $Q\simeq1.3$ MeV is the $Q$ value of the reaction and $m_e$ is the electron mass. For the calculations, the IBD cross section reported in [@strumiavissa] was used. The [**delayed neutron capture**]{} on a proton is characterized by a monochromatic $\gamma_{2.2 \text{ MeV}}$ emission. The coincidence in a typical time window of about 250 $\mu$s between the latter and the prompt signal from the $e^+$ gives a clear signature of an IBD event. This means that, in the time integrated events spectrum, there will be a very high peak around $2.2$ MeV that integrates the same number of events expected for IBD, reduced by the efficiency of the tag. The spectral shape of this peak is due to both the energy resolution of the detector and the quenching of the gamma ray energy in the scintillator. In this work, due to lack of information, we neglect the last effect and consider the optimistic case in which the width of this peak is only due to the energy resolution.
The [**superallowed CC reactions**]{} $\nu_e+^{12}$C$\to e^-
+^{12}$N and $\bar{\nu}_e+^{12}$C$\to e^+
+^{12}$B present physical thresholds of $E_{\nu_{e}}>\!17.3\,$ MeV and $E_{\bar{\nu}_{e}}\!>\!14.4\,$ MeV respectively. They are detectable through the prompt leptons $e^{-}$ ($e^{+}$), which give a continuous spectrum. Moreover the nucleus of both reactions in the final state, $^{12}$N and $^{12}$B, are unstable. The former will decay $\beta^+$ to $^{12}$C with a half life of $\sim 11$ ms. The latter will decay $\beta^-$ to $^{12}$C with a half life of $\sim 20$ ms [@N12]. The high energy positrons and electrons emitted in these beta decays can be observed, giving the possibility to tag these events. The cross sections used for the evaluation are those reported in [@Fukugita].
In the NC channels all neutrino flavors are involved potentially increasing the number of signal events detected.
For the [**ES on protons**]{} channel the cross section in [@ahrens; @xxsec] was used, with a proton strangeness of $\eta=0.12$. However it is important to stress that the uncertainty on the number of events expected for this channel is not negligible due to the proton structure and amounts to about $20\%$ [@cyclotron]. To understand the spectral shape of this class of events it is necessary to model the quenching factor for protons in the scintillators; this accounts for the proton light output and depends on the liquid scintillator composition. A detailed description of this factor is given in the next section.
The cross section for the [**superallowed NC reaction**]{} $\nu +^{12}$C $\to \nu+
^{12}$C$^*$ followed by the emission of a monochromatic $\gamma$ at $15.11$ MeV is reasonably well known. It was measured in KARMEN [@Karmen], confirming the correctness of the calculations as reported in [@Fukugita] within an accuracy of 20%. Future measurements, most remarkably in OscSNS [@oscsns], claim the possibility of measuring more than 1,000 events in one year with a systematic estimated at 5% level or better. The prominent spectral feature of this channel can permit the identification of these events, as a sharp peak around $15$ MeV, standing out from the main signal due to IBD. The total cross section for [**NC proton knockout** ]{} $\nu+{}^{12}\mbox{C}\to \nu + \mbox{p}+^{11}$B has been calculated in [@yoshida], as a part of a network of reactions needed to describe the nucleosynthesis of light elements. However, the calculation of [@haxton] finds a cross section about 30% larger, which suggests an error of at least this order. The neutrino energy has to exceed a pretty high threshold, i.e. $E\!>\!\left[(M_B+m_p)^2-M_C^2\right]/(2 M_C)\!\simeq\!15.9$ MeV (where we use obvious symbols for the masses of the carbon nucleus, of the boron nucleus and of the proton). The initial neutrino energy (minus the activation energy, quantified by the threshold) is shared by the neutrino and the proton in the final state, $E+M_C\approx E'+T_p+M_B+m_p$ so that the maximum kinetic proton energy $T_p^{\mbox{\tiny max}}$ is obtained when the final state neutrino is almost at rest, $E'\approx
0$. The expression for the maximum of the proton kinetic energy is $$T_p^{\mbox{\tiny max}}=\left[(M^*-m_p)^2-M_B^2\right]/(2M^*),$$ with $M^*=\sqrt{M_C^2+2M_C E}$. In view of the smallness of this sample of events, we adopted a very simple procedure to describe the distribution in the kinetic energy of the proton $T_p$, namely, we resorted to the pure phase space, that gives ${d\sigma}/{dT_p}\propto
{d\Phi}/{dT_p}\propto\sqrt{T_p}(M^*-m_p-m_B-T_p)^2$. We checked that the integral in the proton kinetic energy of this expression agrees at the level of few percent with the theoretical behaviour of the total cross sections as reported in [@yoshida].[^1]
In the [**CC knockout proton**]{} reaction on $^{12}$C the outgoing kinetic energy is shared between the electron and the proton. The maximum kinetic proton energy $T_p^{\mbox{\tiny max}}$ is obtained when the final electron is at rest. Similarly to the previous case, this is given by $$T_p^{\mbox{\tiny
max}}=\left[(M^*-m_p-m_e)^2-M_{C11}^2\right]/(2M^*-2m_e),$$ where again $M^*=\sqrt{M_C^2+2M_C E}$. In this case the phase space is ${d\sigma}/{dT_p} \propto {d\Phi}/{dT_p}
\propto\sqrt{T_p}(M^*-m_p-m_{C11}-T_p)^2$. The theoretical cross sections reported in [@yoshida] and the value estimated from pure phase space agree at the level of $\sim 20$%.
In the [**Elastic Scattering on electrons**]{} all the flavors participate, but the cross section is slightly different for the different flavors. The current best measurement of this interaction arises in a sample of 191 events [@ESe; @ESe2], and quotes 17% of total error. The error that we estimate in the standard model is instead absolutely negligible for our purposes.
Numerical formulae {#numerical-formulae .unnumbered}
------------------
Let us conclude this section by giving two numerical formulas to evaluate easily the main neutral current cross sections:\
An easy-to-implement effective formula for the $\nu +^{12}C \to \nu+^{12}C^*$ cross section, that agrees with [@Fukugita] results at better than 1% in the region below 100 MeV, is (\_+\_[|]{})= (E-15.11 )\^2 10\^[p(E)]{}, with $p(E)=\sum_{n=0}^3 c_n \left(E/100\mbox{ MeV}\right)^n$ where $E$ is the incoming neutrino energy, and the numerical coefficients are $c_0=-0.146$, $c_1=-0.184$, $c_2=-0.884$, $c_3=+0.233$.\
A simple parametrization of the cross section for the ES scattering, $\nu p \to \nu p$, assuming that the proton strangeness is $\eta=0.12$, is simply (\_+\_[|]{})=G\_F\^2 E\^2 10\^[q(E)]{}, where $q(E)=-0.333-0.16 (E/100\mbox{ MeV})$.
Description of the Ultra-pure Scintillating Detectors
=====================================================
We consider the ultrapure liquid scintillators detectors that are running or under construction, namely the following three: Borexino (BRX) [@2009BorexinoColl] (0.3 kt of $C_9 H_{12}$) in Gran Sasso National Laboratory, Italy, KamLAND (KAM) in Kamioka Observatory, Japan [@2003KamlandColl] (1 kt of mixture of $C_{12}H_{20}$(80%) and $C_9H_{12}$(20%)) and SNO+ (0.8 kt of $C_6H_5C_{12}H_{25}$) currently under construction in the SNOLAB facility, located approximately 2 km underground in Sudbury, Ontario, Canada [@SNO].
We assume that the energy resolution for each detector is a Gaussian with an error described by $\sigma(E_d)=A\times \sqrt{E_d/\mbox{MeV}}$ and a different value of the constant $A$ for each detector. The trigger threshold in Borexino is as low as about 200 keV, reaching full efficiency at $E_d=250$ keV [@BorexPRL]. The overall light collection in Borexino is $\simeq$ 500 photoelectrons (p.e.)/MeV of deposited energy. The resolution is $\simeq$ 5% at 1 MeV (namely $A=50$ keV). The trigger efficiency in KamLAND currently reaches 100% at 350 keV. The energy resolution of the KamLAND detector can be expressed in terms of the deposited energy as $\sim\,6.9\%/\sqrt{E_d(\text{MeV})}$ (i.e., $A=69$ keV) [@dasg]. The energy threshold expected for SNO+ is the optimistic one of $200$ keV and the energy resolution is supposed to be the same as in Borexino.
$a_1$ $a_2$ $a_3[\mbox{\small MeV$^{-1}$}]$
------ ------- ----------- ---------------------------------
BRX 0.624 $-0.175$ $-0.154$
KAM 0.581 $-0.0335$ $-0.207$
SNO+ 0.629 $ -0.286$ $-0.163$
: Constants appearing in the parametrized formula of the quenching function here adopted.
\[const\]
As we mentioned earlier when the detected particle is a proton, the visible energy is only a fraction of the kinetic energy $T_p$, as described by the ‘quenching function’. Each detector has its own quenching function, that depends on its chemical composition; for Borexino detector we consider the quenching function discussed in [@cyclotron], for KamLAND the one recently discussed in [@KamQuen] and finally for SNO+ the response to proton in LAB scintillator as measured in [@vonKrosigk:2013sa].
Following [@Madey] a simple parametrization of the quenching function is $$E_d=a_1[1-\exp(a_2+a_3\cdot T_p)]\cdot T_p
\label{quench}$$ The values for the constants $a_1$, $a_2$ and $a_3$ that should be used for the different detectors are reported in Tab. \[const\] and the resulting functions are shown in Fig.\[fig3\].
At this point, we obtain the following important conclusion:
> in ultrapure scintillators, the observation of protons from the NC elastic scattering reaction allows us to observe only the high energy part of the neutrino spectra.
In fact, due to the thresholds and to the quenching, the protons below a minimum kinetic energy cannot be detected; this is $0.9$ MeV for SNO+, $1.8$ MeV for Kamland and $1.3$ MeV for Borexino. Thus, taking into account the kinematical relation between the proton kinetic energy and the one of incoming neutrinos, we find that the elastic scattering on protons is sensitive to neutrino energies above a threshold of $22$ MeV in the best situation of SNO+, it becomes 25 MeV for Borexino, and raises to $30$ MeV in the case of Kamland. In view of these considerations, one concludes that the exploration of the low energy of the spectrum via neutral currents is not possible with the existing ultrapure scientillators.
Results
=======
For each detection channel, we estimate the number of expected events and report them in Table \[tab\]. Moreover, we plot the energy distributions of the events, considering the specific features of the ultrapure scintillating detectors in Figure \[fig2\].
The IBD channel (red line) starts to dominate the global signal at $5$ MeV and reaches the maximum around $14$ MeV. The total number of interactions expected for a supernova located at $10$ kpc is of about $54$ events for Borexino, $257$ for Kamland and $176$ for SNO+. These results are reported in the first row of Table \[tab\]. The subsequent gamma from neutron capture gives the peak at 2.2 MeV, shown by a purple line. The efficiency of the neutron tag is (85$\pm$1)% in Borexino (see [@2010Borexino]), (78$\pm$2)% in KamLAND (see [@2003KamlandColl]) and also in SNO+. The condition for a successful IBD tag [@2010Borexino] is that no more than one interaction occurs during the time between the IBD interaction and the neutron capture inside a specific volume, namely $$R_{IBD}\times \Delta t \Delta V \rho \leq 1$$ where $R_{IBD}$ is the rate of IBD events per second and per unit mass, $\Delta t$ is the temporal window of the tag, that we assume to be $\Delta t=2\tau = 512 \mu s$, $\Delta V$ is the volume of a sphere with 1 meter of radius, $\rho$ is the density of the scintillator. This is related to the detector mass and to the distance of the supernova by $$R_{IBD}=\frac{256.5}{T}\cdot\left(\frac{10 \text{kpc}}{D}\right)^2\cdot
\left(\frac{M}{1 \text{kton}} \right),$$ where $M$ is the mass of the detector, $D$ is the distance of the SN and $T$ is the duration of the emission. For example to allow the IBD tag in a detector with the density of Borexino and 1 kton of mass, considering that $50$% of the total emission is expected during the first second [@paglia], then the minimum distance of a SN is $D\geq 0.16$ kpc, that is not a severe limitation.[^2]
Channel Color code Signal BRX KAM SNO+
---------------------------------------------- ------------ ---------- ------------- --------------- ---------------
$\bar{\nu}_e+p\to n+e^+$ red $e^+$ 54.1 (49.6) 256.5 (235.3) 175.8 (161.2)
$n+p\to D +\gamma_{2.2 \text{ MeV}}$ purple $\gamma$ 46.0 (42.1) 200.1(183.5) 137.1 (125.8)
$\nu +p\to \nu+p$ blue $p$ 12.7 (3.8) 29.0 (6.2) 74.9 (29.2)
$\nu +^{12}C \to \nu+^{12}C^*$ orange $\gamma$ 4.7 (2.1) 15.0 (6.7) 12.3 (5.5)
$\nu +e^-\to \nu+e^-$ green $e^-$ 4.4 (4.5) 14.8 (15.5) 12.0 (12.4)
$\nu_e+^{12}C\to e^- +^{12}N$ magenta $e^-$ 2.0 (0.7) 6.4 (2.1) 5.3 (1.7)
$\bar{\nu}_e+^{12}C\to e^+ +^{12}B$ black thin $e^+$ 1.2 (0.8) 3.7 (2.6) 3.0 (2.1)
$\nu +^{12}C \rightarrow \nu + p + ^{11}B$ yellow $p$ 0.7 (0.2) 2.4 (0.6) 2.1 (0.6)
$\nu_e +^{12}C \rightarrow e^- + p + ^{11}C$ red dashed $p$ 0.5 (0.1) 1.5 (0.3) 1.3 (0.2)
Let us discuss now the NC elastic proton scattering. This channel dominates the low energy part of the spectrum, represented with the blue line in Figure \[fig2\], even if as discussed previously, this reaction probes only the high energy part of the supernova neutrinos. It is evident from Table \[tab\] and Fig. \[fig2\] that the event spectrum depends strongly on the quenching of the proton signal. The detector threshold used in the case of KamLAND is an optimistic value, since we assumed a threshold of $350$ keV, lower than the one that can be obtained with the current radioactivity level, i.e., $600$ keV [@Tolich]. If this higher threshold is assumed the ES with protons on KamLAND gives only $17$ signal events.
It is important to remark that the number of events due to this channel is also very sensitive to the SN emission parameters; in fact, as discussed in the end of the previous section, this interaction is sensitive only to the high energy tail of the SN neutrinos spectra. In particular, the average energy of the different neutrino flavors have gradually been changing in recent years, moving toward lower mean values [@Janka:2012wk] and toward minor differences between the average energies of the different components [@Ott:2012mr; @Tamborra:2012ac]. For comparison we have considered the new paradigm of emission, where the average energy of non-electronic flavors is the same as of the electronic antineutrinos, namely $\langle E_x \rangle=12$ MeV [@Janka:2012wk; @Tamborra:2012ac] and have investigated the two different cases to outline the impact on this and the other NC process. As shown by the values in brackets in Table \[tab\], the expectations in this case are quite meager.
The NC neutral current reaction $\nu +^{12}C \to \nu+
^{12}C^*$ followed by the emission of a monochromatic $\gamma$ is shown in orange in Figure \[fig2\]. This channel does not require the low energy threshold and the efficiency for its detection is taken 100% for all the detectors considered. However, the possibility of a successful identification is affected by the quality of the energy resolution of the detector and by the effectiveness to tag the IBD signal. These events can be observed if the IBD events are identified through the correlated neutron capture signal, since they are expected to occur in the same energy region. With the assumed energy resolutions we have that this neutral current reaction can be observed in the energy range (14-16) MeV. For Borexino, the number of events due to the IBD signal in the same range is $5.7$; thus, more than the 50% of the total signal collected in this energy window is due to the IBD channel, while for KamLAND and SNO+ the IBD signal is 26.9 and 18.4, representing about 60% of the total one. In the case of Borexino, if the tagging efficiency is of 85% as assumed in the plot, we expect only 1 event due to IBD not identified, so the uncertainty on the $\gamma_{15.11 \text{MeV}}$ signal is reduced to 14%.
The ESe involves all the flavors of neutrinos and we expect to collect about $4$ events for Borexino, $15 $ for KamLAND and $12 $ for SNO+. Their spectrum is reported with a dark green line in the spectra of Figure \[fig2\], and dominates in the energy region between the ES on protons and the IBD signals. As we mentioned the cross section for the different flavors are slightly different; the $\nu_e$ contribution produces half of the events.
{width="45.00000%"} {width="45.00000%"}\
{width="45.00000%"} {width="45.00000%"}\
{width="45.00000%"} {width="45.00000%"}
The rest of the detection channels have a low signal. All of them show a continuos spectrum, being from $e^{-}$, $e^{+}$ or protons. The two CC superallowed reactions are indicated by a magenta line for the $^{12}N$ final state nucleus and by a black thin line for the $^{12}B$ one. The decay products of the unstable nucleus are not considered in this plot. For both knockout channels, besides the high energy thresholds, the quenching has to be considered, so the total number of events collected for them is pretty small. The one due to NC is shown in yellow, while the one due to CC is shown with the dashed red line.
Conclusions
===========
In this paper, we have obtained and discussed the spectrum of supernova neutrino events in ultrapure scintillators for a supernova exploding at 10 kpc from the Earth. We have examined the capability to distinguish the various detection channels and we have quantified the uncertainties in this type of detectors.
As discussed in the introduction, a major reason of specific interest for a future supernova is the possibility to observe neutral current interactions of neutrinos. We have investigated the three possible reactions of detection in ultrapure scintillators, namely: 1) the elastic scattering with protons, 2) the 15.11 MeV $\gamma$ de-excitation line, 3) the proton knockout channel. Our conclusions are as follows:
The first reaction is characterized by the larger number of expected events in all the detectors; however the number of detectable events is strongly limited by the energy thresholds. The uncertainty on the total number of elastic scattering on protons, due to the proton structure, amounts to the 20%; moreover in the same energy region where this reaction can be observed there are also the indistinguishable events due to the elastic scattering with electrons and those due to NC and CC proton knockout. In other words, all we can observe is the total number of events collected in the energy region from the detector threshold to the threshold of the IBD signal, about $1.8$ MeV. In this detection window, we have found that a fraction of 8% (Borexino), the 7% (KamLAND), the 4% (SNO+) of the signal is due to the other channels and this uncertainty is small but irreducible. We have also seen that, in the case that the the energy of non-electronic neutrinos is low, the number of events due to this reaction is too small to permit the investigation of the $\nu_x$ spectrum at the level discussed in [@dasg].
The gamma line due to neutrino-induced $^{12}$C de-excitation is in principle easier, giving a signal at a high energies; its detection does not require the extreme performances at very low energies are not needed. However, in the same region of the spectrum where this line is visible we will have also positrons due to IBD reaction; thus, the efficiency to tag the concomitant neutron will be of crucial importance to identify cleanly a sample of this NC reaction. While this concern is not a severe issue for the type of detectors we have considered in this work, it is much more relevant for future scintillators with a much larger mass and limited performances at low energies.
Finally, we have shown that the proton knockout will give a comparably small number of NC events.
[99]{}
K. Hirata, et al., *Observation of a Neutrino Burst from the Supernova SN 1987a*, *Phys. Rev. Lett.* [**58**]{} (1987) 1490.
R. M. Bionta, et al., *Observation of a Neutrino Burst in Coincidence with Supernova SN 1987a in the Large Magellanic Cloud*, *Phys. Rev. Lett.* [**58**]{} (1987) 1494.
E. N. Alekseev, et al., *Possible Detection of a Neutrino Signal on 23 February 1987 at the Baksan Underground Scintillation Telescope of the Institute of Nuclear Research*, *JETP Lett.* [**45**]{} (1987) 589.
M. Ikeda, et al. \[Super-Kamiokande Collaboration\], *Search for Supernova Neutrino Bursts at Super-Kamiokande*, *Astrophys. J.* [**669**]{} (2007) 519 \[arXiv:0706.2283 \[astro-ph\]\].
R. Abbasi, et al. \[IceCube Collaboration\], *IceCube Sensitivity for Low-Energy Neutrinos from Nearby Supernovae*, *Astron. Astrophys.* [**535**]{} (2011) A109 \[arXiv:1108.0171 \[astro-ph.HE\]\].
G. Alimonti, et al. \[Borexino Collaboration\], *The Borexino detector at the Laboratori Nazionali del Gran Sasso*, *Nucl. Instrum. Meth.* [**A 600**]{} (2009) 568
K. Eguchi, et al. \[KamLAND Collaboration\], *First results from KamLAND: Evidence for reactoranti-neutrino disappearance* *Phys. Rev. Lett.* [**90**]{} (2003) 021802
J. F. Beacom, W. M. Farr and P. Vogel, *Detection of supernova neutrinos by neutrino proton elastic scattering*, Phys. Rev. D [**66**]{} (2002) 033001
B. Dasgupta and J. .F. Beacom, *Reconstruction of supernova $\nu_\mu$, $\nu_\tau$, anti-$\nu_\mu$, and anti-$\nu_\tau$ neutrino spectra at scintillator detectors*, *Phys. Rev.* [**D 83**]{} (2011) 113006 T. J. Loredo and D. Q. Lamb, *Bayesian analysis of neutrinos observed from supernova SN-1987A*, *Phys. Rev.* [**D 65**]{} (2002) 063002 \[astro-ph/0107260\]. G. Pagliaroli, F. Vissani, M. L. Costantini and A. Ianni, *Improved analysis of SN1987A antineutrino events*, *Astropart. Phys.* [**31**]{} (2009) 163 \[arXiv:0810.0466 \[astro-ph\]\]. M. T. .Keil, G. G. Raffelt and H. -T. Janka, *Monte Carlo study of supernova neutrino spectra formation*, *Astrophys. J.* [**590**]{} (2003) 971 \[astro-ph/0208035\]. I. Tamborra, B. Muller, L. Hudepohl, H. -T. Janka and G. Raffelt, *High-resolution supernova neutrino spectra represented by a simple fit*, *Phys. Rev.* [**D 86**]{} (2012) 125031 \[arXiv:1211.3920 \[astro-ph.SR\]\]. B. Mueller and H. -T. .Janka, arXiv:1402.3415 \[astro-ph.SR\].
A. Strumia, F. Vissani, *Precise quasielastic neutrino/nucleon cross-section* *Phys. Lett.* [**B 564**]{} (2003) 42. F. Ajzenberg-Selove, *Energy levels of light nuclei A = 11-12*, *Nucl. Phys.* [**A 506**]{} (1990) 1
M. Fukugita, Y. Kohyama, K. Kubodera, *Neutrino Reaction Cross-Sections on C-12 Target* *Phys. Lett.* [**B 212**]{} (1988) 139.
L.A. Ahrens, et al. \[BNL 734\] *Measurement of Neutrino - Proton and anti-neutrino - Proton Elastic Scattering*, *Phys. Rev.* [**D 35**]{} (1987) 785.
G. Pagliaroli, C. Lujan-Peschard, M. Mitra and F. Vissani, *Using Low-Energy Neutrinos from Pion Decay at Rest to Probe the Proton Strangeness*, *Phys. Rev. Lett.* [**111**]{} (2013) 022001.
B. E. Bodmann, et al. \[KARMEN. Collaboration\], *Neutrino interactions with carbon: Recent measurements and a new test of electron-neutrino, anti-muon-neutrino universality*, *Phys. Lett.* [**B 332**]{} (1994) 251.
H. Ray \[OscSNS Collaboration\], *OscSNS: Precision Neutrino Measurements at the Spallation Neutron Source*, *J. Phys. Conf. Ser.* [**136**]{} (2008) 022029 \[arXiv:0810.3175 \[hep-ex\]\].
T. Yoshida, et al. *Neutrino-Nucleus Reaction Cross Sections for Light Element Synthesis in Supernova Explosions* *Astrophys. J.* [**686**]{} (2008) 448
A. Heger, E. Kolbe, W. C. Haxton, K. Langanke, G. Martinez-Pinedo and S. E. Woosley, *Neutrino nucleosynthesis*, *Phys. Lett.* [**B 606**]{} (2005) 258
L. B. Auerbach, et al., *Measurement of electron - neutrino - electron elastic scattering*, *Phys. Rev.* [**D 63**]{} (2001) 112001.
R. C. Allen, et al., *Study of electron-neutrino electron elastic scattering at LAMPF* *Phys. Rev.* [**D 47**]{} (1992) 11.
H. M. O’Keeffe, E. O’Sullivan, M. C. Chen, *Scintillation decay time and pulse shape discrimination in oxygenated and deoxygenated solutions of linear alkylbenzene for the SNO+ experiment*, *Nucl. Instrum. Meth.* [**A 640**]{} (2011) 119.
C. Arpesella et al. \[Borexino Collaboration\], *Direct Measurement of the Be-7 Solar Neutrino Flux with 192 Days of Borexino Data*, *Phys. Rev. Lett.* [**101**]{} (2008) 091302. R. Madey, et al., *Nucl. Instrum. Meth.* [**151**]{} (1978) 445.
S. Yoshida et al. *Light output response of KamLAND liquid scintillator for protons and 12 C nuclei*, *Nucl. Instrum. Meth.* [**A 622**]{} (2010) 574.
B. von Krosigk, L. Neumann, R. Nolte, S. Rottger and K. Zuber, *Measurement of the proton light response of various LAB based scintillators and its implication for supernova neutrino detection via neutrino-proton scattering*, arXiv:1301.6403 \[astro-ph.IM\].
D. K. Nadyozhin, *The neutrino radiation for the hot neutron star formation and the envelope outburst problem*, *Astrophys. Space Sci.* [**53**]{} (1978) 131.
H. A. Bethe, J. R. Wilson, *Revival of a stalled supernova shock by neutrino heating*, *Astrophys. J.* [**295**]{} (1985) 14.
H. Duan, G. M. Fuller and Y. -Z. Qian, *Collective Neutrino Oscillations*, *Ann. Rev. Nucl. Part. Sci.* [**60**]{} (2010) 569 \[arXiv:1001.2799 \[hep-ph\]\].
G. Badino, et al., *The 90 ton liquid scintillator detector in the mont blanc laboratory*, *Nuovo Cimento* [**C 7**]{} (1984) 573.
N. Yu. Agafonova, et al., *Study of the effect of neutrino oscillations on the supernova neutrino signal in the LVD detector*, *Astropart. Phys.* [**27**]{} (2007) 254.
P. Antonioli, W. Fulgione, P. Galeotti and L. Panaro *Simulation of low-energy neutrino interactions in liquid scintillation counters*, *Nucl. Instrum. Meth.* [**A 309**]{} (1991) 569.
T. Totani, K. Sato, H. E. Dalhed and J. R. Wilson, *Future detection of supernova neutrino burst and explosion mechanism*, *Astrophys. J.* [**496**]{} (1998) 216 \[astro-ph/9710203\].
G. Bellini, et al. \[Borexino Collaboration\], *Observation of Geo-Neutrinos*, *Phys. Lett.* [**B 687**]{} (2010) 299.
K. Tolich \[KamLAND Collaboration\], *Supernova detection with KamLAND* *Nucl. Phys. Proc. Suppl.* [**221**]{} (2011) 355.
H. -T. Janka, *Explosion Mechanisms of Core-Collapse Supernovae*, *Ann. Rev. Nucl. Part. Sci.* [**62**]{} (2012) 407 \[arXiv:1206.2503 \[astro-ph.SR\]\].
C. D. Ott, et al. *General-Relativistic Simulations of Three-Dimensional Core-Collapse Supernovae*, *Astrophys. J.* [**768**]{} (2013) 115 \[arXiv:1210.6674 \[astro-ph.HE\]\].
N. Smith, K. H. Hinkle, and N. Ryde, *Red Supergiants as Potential Type IIn Supernova Progenitors: Spatially Resolved 4.6 $\mu$m CO Emission Around VY CMa and Betelgeuse*, *The Astronomical Journal* [**137**]{} (2009) 3558.
[^1]: However, a word of caution is in order; while the above considerations on phase space are suggestive, they are just a reasonable way to explore of the consequences of this reaction in scintillator detectors: in fact, the distribution in $T_p$ of [@yoshida] is not available. Certainly, it would be better to have a true calculation of the distribution in $T_p$ of this reaction (or possibly its parameterization) along with an assessment of the theoretical error in the relevant energy range.
[^2]: If instead, a similar detector but with a $50$ kton mass is considered the distance becomes $D\geq 1.13$ kpc, that includes several known potential Core-Collapse SNe as Betelgeuse and VY Canis Majoris [@bet].
|
{
"pile_set_name": "ArXiv"
}
|
[**Impact of the economic crisis\
0.3cm on the Italian public healthcare expenditure**]{}\
[<span style="font-variant:small-caps;">Carlo Castellana</span>]{}\
[ *Management School in Clinical Engineering,\
University of Trieste, Italy*]{}
**Abstract**
The global financial crisis, beginning in 2008, took an historic toll on national economies around the world. Following equity market crashes, unemployment rates rose significantly in many countries: Italy was among those. What will be the impact of such large shocks on Italian *healthcare* finances? An empirical model for estimating the impact of the crisis on Italian public healthcare expenditure is presented. Based on data from epidemiological studies related to past economic crisis, the financial impact is estimated to be comparable to the healthcare deficit of Italian Regions (EUR 3-5 bn). According to current agreements between the Italian State and its Regions, public funding of regional National Health Services (NHSs) is limited to the amount of regional deficit and is subject to previous assessment of strict adherence to constraint on regional healthcare balance-sheet. Those Regions that will fail to comply to balance-sheet constraints will suffer cuts on their public NHS financing with foreseeable bad consequences for the health of their regional population. The current crisis could be a good timing for a large-scale re-engineering of the Italian NHS, probably the only way for self-sustainability of the public system.
Introduction {#Introduction}
============
Following the tech bubble and the events of September 11, the Federal Reserve stimulated a struggling economy by cutting interest rates to historically low levels. As a result, a housing bull market was created. In turn, investors sought higher returns through riskier investments. Lenders took on greater risks and approved subprime mortgage loans to borrowers with poor credit. Consumer demand drove the housing bubble to all-time highs. Interest rates climbed back up which caused default on many subprime mortgages. This left mortgage lenders with property that was worth less than the loan value due to a weakening housing market. Defaults increased, the problem snowballed and several lenders went bankrupt.
The crisis was named [ “]{}Subprime Crisis“, which was fine at first, but it’s now totally inadequate. It started as a phenomenon affecting one country (the USA) and one market sub-sector (the subprime mortgage market) only. In few months the contagion widespread to the overall financial system where it took many facets: turmoil reached the global mortgage markets and later on credit markets where it became clear that this was not just a painless [ “]{}infection” but a true [ “]{}epidemic". A bank crisis ensued, thus hitting the heart of the overall financial system: the lack of liquidity forced exceptional interventions of the European Central Bank and Federal Reserve. By leaking into the real economic tissue, the crisis entered the real life of common people. The effects of the economic crisis in the USA are currently being felt worldwide and Italy is not an exception.
The crisis will likely have a long-lasting impact on Italy’s economic potential [@IMFCountryReport]. Indeed, innovation and investment opportunities may weaken because demand prospects are likely to be poor and the real cost of borrowing remains high. In addition, some of the increase in unemployment may be structural given that very difficultly displaced workers will be able to return to the labor market as industrial restructuring takes hold.
Italy’s structural problems turned in a poor productivity performance and a reduced growth over the last fifteen years [@Fachin]. The global financial crisis has exacerbated these long-standing weaknesses, taking a heavy toll on Italy’s economy. Italy has suffered from chronically low economic growth, even before the global financial crisis. Real Gross Domestic Product (GDP) growth averaged 1.6% during the period 1995-2007, down from over 2% in the earlier decade [^1]. In the first quarter of 2009, growth witnessed a decline four times as large as the one experienced during the European Monetary System (EMS) crisis of 1992-93 [@Intro_Bassanetti].
While most forecasters were expecting the beginnings of recovery in late 2009 or early 2010, and even if many financial indicators point for a gradual general improvement in the main economies, the current situation has not recovered yet to the pre-crisis one and the actual duration will depend on the effectiveness of policies in the developed countries.\
Financial markets crashes, the economy shrinks, people loose their jobs: how can this affect the Italian NHS?
The relation (fig. \[fig: Intro\_3\] and \[fig: Intro\_4\]) is that job losses mean a reduction of household income and this has serious impacts on people’s health [@HealthEconStatus1; @HealthEconStatus2; @HealthEconStatus3; @HealthEconStatus4; @HealthEconStatus5; @HealthEconStatus6; @HealthEconStatus7; @HealthEconStatus8].
Past economic crisis were associated with an increase in morbidity and mortality [@PastCrisisConseq1], and this one is not going to be an exception [@SubpCrisisConseq1; @SubpCrisisConseq2; @SubpCrisisConseq3]. Understanding the routes of potential impact on health and healthcare services of the reduction of income for Italian individuals as a consequence of the crisis, provides the framework for action to mitigate against this threat becoming a reality.\
Methods
=======
\[sec:Methods\] The healthcare implications of the economic crisis can be numerous and serious. Before choosing how to act in order to cope for such bad consequences, one need to quantify their financial impact: any health plan to re-engineer the NHS will depend on the availability of enough financial resources to cover that plan.\
[**What will be the impact of the current economic crisis on the Italian public NHS expenditure?**]{}\
To answer this question a financial model has been implemented. The model has three types of inputs [@Polder]:
1. [**demographics**]{}: population, GDP, unemployment rate, mortality rates, life-expectancy;
2. [**healthcare**]{}: healthcare expenditure, ratio between the healthcare expenditure of a person during its last year of life and that of a person with the same age but in good health conditions (*Deceased/Survivor Ratio*);
3. [**econometrics**]{}: hypothesis on the *main driver* of healthcare expenditure (GDP, healthcare personnel, *death-related costs*).
Predicting the future evolution of healthcare expenditure is one of the crucial challenges that the EU and its Member States are facing in the context of the demographic and social changes that are currently taking place in Europe. It is a hard and yet unknown subject which is constantly threatening the main economic panels worldwide [@EP1; @EP2; @EP3] due to the complexity of the systems and multiplicity of factors affecting both total and public spending. To tackle this issue a major project was undertaken by the European Commission and Economic Policy Committee which aimed at projecting future public healthcare expenditure in twenty-seven Member States of the EU and Norway over the period 2007- 2060 [@RE_Przywara].
Two types of uncertainties make projecting the future evolution of the *absolute* value of healthcare expenditure a very complex issue: one at model level (the multifactorial nature of healthcare spending) and another at the level of model parameters (the intrinsic volatility of variables underlying the model).
However, within this paper we are only interested in calculating the impact of the economic crisis on the healthcare expenditure which, thanks to its *differential* nature [^2], has a lower level of uncertainty. The basic idea is that [ “]{}noise“ affecting two estimates of a variable calculated with the same model would net out when calculating their difference. As a consequence, the differential estimate will be much more reliable than the absolute estimates: within this level of approximation, the details of the underlying econometric model (EM) become less relevant. Of course this is even truer if the differential quantity we are calculating is a [ “]{}small” one compared to the other two entering the difference. The impact of the economic crisis on Italian healthcare expenditure falls within these assumptions: as it will be shown throughout this paper, its estimate is EUR 0.5-9.5 bn, far less than the total Italian NHS expenditure of EUR 109.7 bn [^3].
Remarkably the actual financial impact of the crisis, even if being a small portion of the total healthcare expenditure, it is not negligible compared to the EUR 2.368 bn [^4] of *healthcare deficit*. The knowledge of such an estimate becomes very important for those Institutions in charge of regional Health Planning, especially during the current difficult conditions of the Italian NHS that finds many Regions enforced to implement recovery plans, [**[ “]{}Piani di Rientro"**]{}, aimed at reducing the issue of the inflating regional healthcare deficit.\
Epidemiological analysis {#subs: prevstu}
========================
How can we measure what the current economic crisis will mean for the health of Italian population? How much distress will the impact on health cause to the already fragile financial [ “]{}health" of Italian regional balances?
We will answer by reviewing the epidemiological evidence supporting an association between unemployment and health outcomes in the population.
In order to find some parameters that can help [ “]{}measuring" the impact of the crisis on healthcare, we review the experience of two major economic crises of the 20th century in addition to various epidemiological studies on specific cohorts:
1. Post-communist Depression (early 1990s) [@Impact_Men] and [@Impact_Sachs; @Impact_Wedel; @Impact_Murrell; @Impact_Kontorovich];
2. East Asian crisis (late 1990s) [@Impact_Kim; @Impact_Khang; @Impact_Waters] and [@Moreno; @Intereconomics];
3. studies on specific cohorts:
1. Canadian recession (1985) [@Impact_Siddique]
2. British cohorts (1981-1983 and 1971-1981) [@Impact_Moser1; @Impact_Moser2; @Impact_Beale1; @Impact_Beale2; @Impact_Yuen];
3. Finnish cohort (1981-1985) [@Impact_Martikainen];
4. Danish cohort (1970-1980) [@Impact_Iversen];
5. Italian cohorts (1976-1985 and 1981-1985) [@Impact_Costa] ;
6. US veterans cohort [@Impact_Linn]
Available evidence suggests that health is at risk in times of rapid economic change, in both booms and busts. However the impact on mortality is exacerbated where people have easy access to the means to harm themselves and is ameliorated by the presence of strong social cohesion and social protection systems.
We can look to experiences of the past to guide our expectations of the public health effects of this crisis.
Reported results of studies on specific cohorts are based on the review of [@Impact_Jin] where the authors selected articles published in 1980s and 1990s supporting an association between unemployment and adverse health outcomes. Those include time-series analysis, cross-sectional, case-control and cohort studies.
### Approximation n.1 {#sec: Approx1 .unnumbered}
We translate the effect of the crisis into a perturbation of some parameters entering our model of public healthcare expenditure. As a first approximation the impact of the crisis on healthcare has been assumed to depend on (see fig. \[fig: ReE\_1\]):
1. mortality rates (demographic measure)
2. utilization rate of healthcare services (healthcare measure)
### Approximation n.2 {#approximation-n.2 .unnumbered}
The second type of approximation has been used to address the way the increased death relative risk (RR) translates into increased mortality rates. Independently of the term-structure of mortality rates, we adopt the following simplifying relationship: $$\label{eq: Impact_1}
PD' = \frac{UEP \cdot PD \cdot RR + EP \cdot PD}{WAP}$$ where PD and PD’ are the probability of death in 5 years time [^5] that should be expected under normal circumstances and, respectively, as a consequence of the economic crisis; *UEP* and *EP* are the number of people that are unemployed and, respectively, employed; *WAP* is the active population, i.e. the number of people in working-age cohorts.
Eq. \[eq: Impact\_1\] can be rearranged as: $$\label{eq: Impact_2}
PD' = PD \cdot \frac{UEP \cdot RR + EP }{WAP}$$ that, together with *WAP*=*UEP* + *EP*, yields: $$\label{eq: Impact_3}
PD' = PD \times RR'$$ and based on the definition of unemployment rate $\omega$, the effective RR can be calculated as: $$\label{eq: Impact_4}
RR' = 1 + \frac{UEP}{WAP} \cdot (RR-1) = 1 + \omega \cdot (RR-1)$$
Based on what stated so far we can now answer to the following two questions.\
[**Q.1** ]{} *What number will we use to quantify the mortality increase due to the economic crisis?*\
As a first step, we need to make sure data on which we extrapolate the mortality increase to be homogeneous. As explained in fig. \[fig: Impact\_3\] the Russian data are diluted to the overall population and hence are not directly comparable to the [ “]{}undiluted" ones of the other studies.\
By means of eq. \[eq: Impact\_4\], assuming a 10% current level of unemployment rate [^6], we are able to compare RRs values of cohort studies that applied to the subpopulation of unemployeds only (data in fig. \[fig: Impact\_3\]), to RRs values reported for the Russian crisis.
Results of this process are reported in fig. \[fig: Impact\_5\] where the selected range for each age-cohort is shown.\
[**Q.2** ]{} *What number will we use to quantify the increased usage of healthcare services due to the economic crisis?*\
The use of healthcare services is strongly dependent on the type of healthcare system. Indeed the association of unemployment with increased use of healthcare services depends on services being universally available and free of charge at point of use, as they are in Canada and Britain. In the USA, in contrast, hard economic times may mean [ “]{}nearly empty waiting rooms" because jobless people often lack health insurance and the ability to pay [@Impact_Frey]. The same applies to the East Asia: as reported in fig. \[fig: Impact\_2\], the effect of the East Asian crisis on countries as Korea and Indonesia was a decline in usage of healthcare services. This is not surprising given that Korean healthcare system can be described as a privately controlled delivery system in combination with a publicly regulated financing system [@Impact_Kim] and, as of 1998, health insurance was an income-based contributory insurance program. Indonesia was lacking of social protection schemes.\
The Italian NHS is much more similar to the UK, Swedish and Canadian ones than it is to the US or East Asian ones. Due to this important difference averaging numbers of cohort studies - which, as reported in fig. \[fig: Impact\_4\], refer to Canada, UK and US veterans - with those coming from studies on East Asian crisis can turn into meaningless results.
Henceforth we will discard data of fig. \[fig: Impact\_2\] and we will look at data in fig. \[fig: Impact\_4\] only. Several studies [@Impact_Siddique; @Impact_Beale1; @Impact_Beale2; @Impact_Linn; @Impact_Yuen] reported increased use of general healthcare services in association to an increased unemployment rate. These studies report increases in visits to physicians, hospital inpatient or outpatient admissions and use of prescription medication.
Limited by the available statistics reported in the aforementioned studies, we selected the following ranges to quantify the usage of healthcare services:
- we take 1.20 - 1.57 as range for the increase in usage of general practitioners (GP) services [^7];
- for the increase in hospital utilization rates we use the only two values available: 1.33 - 2.00;
- for the Outpatient visits we use the only data available: 1.63.
We translate the above ranges according to eq. \[eq: Impact\_4\] and get: 1.02 - 1.057 for the Visits to GPs, 1.033 - 1.10 for Admissions to hospitals and 1.063 for the Outpatient visits (see fig. \[fig: ReE\_1\]).
The financial model {#subs: finmod}
===================
Given the wide range of underlying factors and channels through which they affect spending, rather than attempting to construct an all-encompassing projection methodology to capture all demographic and non-demographic factors, two projection scenarios have been run in order to tackle the issue from different perspectives:
1. pure demographic (PD);
2. constant health (CH);
A discussion of the mathematical details of each one of those econometric projection scenarios can be found in [@RE_Przywara].\
Population projections have been modeled according to the six scenarios of fig. \[fig: ReE\_4\]:
1. four scenarios (PopMV, PopHV, PopLV and PopCFV) are available on the European Statistical Agency (ESA) website and are based on ESA assumptions. Indeed, the four scenarios differ by the distributions of persons among age-cohorts, PopLV representing a relatively [ “]{}older“ population, PopHV a [ “]{}younger” one and PopMV an [ “]{}average“ between the two. PopCFV is a [ “]{}constant fertility variant”. Please note that these four scenarios are projected independently of the mortality rates scenarios [^8];
2. two additional scenarios (PopSV br: 1.7% and PopSV br: 1.3%) have been simulated starting from current mortality rates [^9]. These two scenarios are calculated based on current population structure by age-cohorts, mortality tables and a crude birth rate of 1.7% (for the HIGH birth-rate scenario) and 1.3% (for the LOW birth-rate scenario).
As far as spot age-related mortality rates [^10], Expected Life and Deceased/Survivor ratio are concerned we referenced to [@RE_Przywara] (see fig. \[fig: ReE\_6\] and fig. \[fig: Impact\_8\]).
The general improvement of health reflects into longer expected life and lower mortality rates. Future expectations for mortality rates and life expectancy have been drawn starting from historical trends and adjusted according to a 3 months over year general health improvement [@RE_Przywara].
Scenarios of age-related public healthcare costs, as shown in fig.\[fig: ReE\_7\], have been adapted from various sources [@RE_Przywara; @Impact_EC; @Impact_Aprile].
The impact of the crisis-related increased mortality (CRIMI) has been calculated assuming a functional relationship between healthcare expenditure (hcEXP) and the socio-economic-healthcare environment, i.e. a series of demographic ($d_1$, $d_2$, .., $d_M$) and healthcare related ($h_1$, $h_2$, .., $h_N$) parameters. For a given EM: $$\label{eq: Impact_5}
hcEXP = \Lambda^{[EM]}(d_1, d_2, .., d_M; h_1, h_2, .., h_N)$$
As previously discussed, the only two variables that we have used to parametrize the crisis impact on the public healthcare expenditure are (1) the increase in mortality rates and (2) the change in utilization of healthcare services. Both effects are simplified as a rescaling of the corresponding base scenario according to the relative risk in fig. \[fig: ReE\_1\].
Let’s group all the parameters but the two above as $\mathcal{P}$. Let’s call the mortality rates as $m^{(a)}_i$ where $a$ is the age group (e.g. 15-19 years) and $i$ is a date in the future (e.g. 2035) [^11]. Be named $u^{(a)}$ the utilization rate at age cohort $a$. Eq. \[eq: Impact\_5\] is rearranged as: $$\label{eq: Impact_7}
hcEXP = \Lambda^{[EM]}(m^{(a)}_i, u^{(a)};\mathcal{P})$$ Within this framework, the CRIMI can be expressed as: $$\begin{aligned}
\label{eq_CRIMI}
CRIMI & = & \Lambda^{[EM]}(RR_{a, i}'^{MR}\cdot m^{(a)}_{i, BS}, u^{(a)}_{BS};\mathcal{P}_{BS}) |_{t=2015} \nonumber \\
& - & \Lambda^{[EM]}(m^{(a)}_{i,BS}, u^{(a)}_{BS};\mathcal{P}_{BS}) |_{t=2015}\end{aligned}$$ and the crisis-related increased utilization impact (CRIUI) as: $$\begin{aligned}
\label{eq_CRIUI}
CRIUI & = & \Lambda^{[EM]}(m^{(a)}_{i,BS}, RR_{a, i}^{UR}\cdot u^{(a)}_{BS};\mathcal{P}_{BS})|_{t=2015} \nonumber \\
& - & \Lambda^{[EM]}(m^{(a)}_{i,BS}, u^{(a)}_{BS};\mathcal{P}_{BS})|_{t=2015}\end{aligned}$$ where values with BS refer to the base scenario, i.e. the one without any crisis effect. $RR_{a, i}'^{MR}$ and $RR_{a}^{UR}$ are the relative risks that should be applied to mortality rates and, respectively, utilization rate (fig. \[fig: ReE\_1\]).
Projections have been run over a discrete set of 5-years time values, hence the mortality rescaling $RR_{a, i}'^{MR}$ has been taken as (fig. \[fig: Impact\_7\]): $$\label{eq: Impact_6}
RR_{a, i}'^{MR} =
\left \{
\begin{array} {rl}
RR_{a}'^{MR} & \mbox{, if } t = 2015 \\
1 & \mbox{, otherwise}
\end{array}
\right .$$
As far as $RR_{a, i}^{UR}$ is concerned, all age-cohorts are assumed to have the same relative risk $RR^{UR}$: $$\label{eq: Impact_UR}
RR_{a, i}'^{UR} =
\left \{
\begin{array} {rl}
RR'^{UR} & \mbox{, if } t = 2015 \\
1 & \mbox{, otherwise}
\end{array}
\right .$$
The total impact of the economic crisis on healthcare expenditure is calculated as: $$\begin{aligned}
CRI = CRIMI + CRIUI\end{aligned}$$
Results
=======
CRIMI estimates
---------------
Calculations for the impact of the economic crisis on 2015 healthcare public expenditure are reported in fig. \[fig: ReE\_10\]. In addition to calculating an estimate based on the impact of historical mortality and utilization rates, a sensitivity analysis to the estimates of mortality relative risk and utilization rate factor has been run. Estimates based on CH and DC models differ by an order of magnitude which is not surprising since mortality rates are treated differently by the two models.
CRIUI estimates
---------------
Once we have chosen (i) the EM, (ii) a reference scenario (BS) and (iii) a mapping between RR and mortality rates $m^{(a)}_{i,BS}$, the calculation of the impact of an increased relative risk of death is straightforward. In relation to the calculation of the impact of an increased relative risk of utilization rates, another assumption is needed: while mortality rates enter explicitly within the dynamics of future healthcare costs, the same does not hold for the utilization rate of healthcare services and a model assumption becomes essential.[^12]
### Assumption n.3 {#assumption-n.3 .unnumbered}
By definition of age-related healthcare costs pro-capita $c_i^{(a)}$, eq. \[eq: Impact\_5\] can be written as: $$hcEXP = \sum_a N_i^{(a)} \cdot c_i^{(a)}$$ where $N_i^{(a)}$ is the number of persons falling within age-cohort $a$. Total healthcare expenditure can be divided into four typologies: (i) hospital (i.e. in-patients), *H*; (ii) pharmaceutical, *P*; (iii) specialistic (i.e. out-patients), *S*; (iv) general practitioners, *GP*; (v) rehabilitation, *R*; (vi) other minors, *m* [^13]. Thus we can also write: $$hcEXP = H + P + S + GP + R + m$$ Data available from epidemiological studies show that each one of these cost item is increased as a consequence of unemployment by its own RR. Unfortunately epidemiological RR are available for $H$, $P$ and $S$ only (see fig. \[fig: ReE\_1\]) [^14]. Nevertheless this should not cause any major concern since $R$ and $m$ represent a minor amount of the Italian public healthcare expenditure. The rescaling factor (RF) that should be applied to the age-related costs in order to model the increase in healthcare utilization rates can be calculated as: $$\label{eq: Impact_8}
RF = RR_{H} \cdot \%H + RR_{P} \cdot \%P + RR_{S} \cdot \%S + RR_{GP} \cdot \%GP + RR_{R} \cdot \%R + RR_{m} \cdot \%m$$ Based on Italian Treasury 2008 data, we have %H=71%, %P=10%, %S=4%, %GP=6%, %R=2% and %m=7%. Accordingly, RF falls within the range 1.045 - 1.095. CRIUI is thus calculated as: $$\begin{aligned}
\label{eq: Impact_9}
CRIUI & = & \Lambda^{[EM]}(m^{(a)}_{i, BS}, RF \cdot c^{(a)}_{BS};\mathcal{P}_{BS}) |_{t=2015} \nonumber \\
& - & \Lambda^{[EM]}(m^{(a)}_{i,BS}, c^{(a)}_{BS};\mathcal{P}_{BS}) |_{t=2015}\end{aligned}$$
and results are reported in fig. \[fig: ReE\_10\]. By combining the ranges calculated for CRIMI and CRIUI we get as [**estimate for the total crisis-related impact on public healthcare expenditure**]{}, the range [****0.03 - 0.62 %]{} of Italian GDP. Using GDP level for 2009 [^15] we get an estimate for CRI: [**EUR 0.5 - 9.5 bn**]{} (fig. \[fig: Impact\_9\]).
Limits {#sec: Limits}
======
To the best of our knowledge there are no conclusive and comprehensive studies in the scientific literature on the effect of an economic crisis in terms of mortality rates and utilization rates of healthcare services. The epidemiological studies that we reported show an ample range of variability when the same parameters are considered. Moreover the effects of an economic crisis are highly dependent on the type of healthcare system in place. As a general example one could compare the totally opposite results that has been found for two countries involved in the same East Asian crisis, i.e. Indonesia and Thailand [@Impact_Waters].
Our estimates have privileged epidemiological studies related to European countries and, among these, we focused on those healthcare systems similar to the Italian one.
If on the mortality relative risk we have been able to average on a set of ten studies, much less information was available for the utilization rates of healthcare services. As a consequence the estimates on CRIUI have a higher uncertainty than those on CRIMI.
Estimated model parameters could be a weak point of our calculations. A wider set of data would need to be collected from other epidemiological studies. Unfortunately this is a difficult task since there is no database reporting the sought-after effects. Additional complications come from the lack of standardization among epidemiological studies in terms of reported parameters.
Two different approaches (CH and DC) have been used to assess the impact of the crisis on potential public healthcare expenditure. None of them is deemed to be perfect or superior, but each offers some insight into this difficult issue. There is no evidence at all that one of the two is preferable to the other and indeed they are both taken into account when elaborating demographic projections at the European Community level [@RE_Przywara]. Further insight on the actual dynamics between unemployment, mortality and usage of healthcare services could help improve our results.
Conclusions {#Conclusions}
===========
One of the main issues affecting the Italian NHS is the healthcare deficit: according to current agreements between the Italian State and its Regions [@ConferenzaStatoRegioni] public funding of regional NHS is now limited to the amount of regional deficit and is subject to previous assessment of strict adherence to constraint on regional healthcare balance-sheet. Many Regions with previously uncontrolled healthcare deficit have now to plan their [ “]{}Piano di Rientro" (PdR) and submit it for the approval of the Italian Ministry of Economy and Finance. Those Regions that will fail to comply to deficit constraints will suffer cuts on their public NHS financing.
The estimated impact of the economic crisis is comparable to the healthcare deficit of Italian Regions (EUR 3-5 bn), hence any serious Health Planning over the next five years needs to take the problem of possible negative financial implications of current economic crisis into account.
There is much that can be done to mitigate the impact of the financial crisis. Clearly, even in times of crisis, the first goal of the Italian NHS should be to guarantee that health is protected. A smart Health Planning can make sure health-spending is managed appropriately. Italy should recognize current crisis as an opportunity to undertake financial and sectoral reforms. It should pursue a multi-sectorial response, with an emphasis in the health sector. Although short-term measures to mitigate negative consequences are urgent, a longer term perspective aimed at making the Italian health sector more resilient in the future is needed.
Figures {#sec:Figures}
=======
![The economic crisis induces a waterfall of events in the economic, healthcare, social and Government sectors. The main driver of fiscal policies is the funding pressure on Government balance sheet. Two opposite actions are possible: either increase or decrease healthcare spending. The two have different consequences that feed-back to Government funding pressure by three very different routes (*Bad feedback 1, Good feedback 2 and Bad feedback 3 lines*).[]{data-label="fig: Intro_3"}](Intro_3)
![Relationship between Health Planning and Economic crisis. Through a re-engineering of the public NHS (*RE-ENGINEERING line*), Health Planning avoids increased funding pressures allowing the *Good feedback 2* route to positively feed through health improvement and increased labour productivity. Compare with fig. \[fig: Intro\_3\].[]{data-label="fig: Intro_4"}](Intro_4)
![Impact of the economic crisis on healthcare expenditure. The economic crisis causes (i) worsening of health conditions, that shows through an increased mortality associated to cardiovascular diseases and to suicides; (ii) an increased demand for healthcare services (more visits to GPs and specialists as well as increasing hospital admissions) that translates into higher costs for the NHS. The reader should note that while Type II effects *directly* support an increased usage of healthcare services (hence, higher healthcare costs), Type I effects *indirectly* translate into higher healthcare expenditure through worsening health conditions. *n/a*: not assessed.[]{data-label="fig: ReE_1"}](paper2)
![Results of longitudinal studies of rates of overall mortality among unemployed people due to general and specific causes. Please note that SMR are expressed in hundreds, i.e. to compare SMR with RR one should firstly divide the shown SMR number by 100. Numbers in brackets are 95% confidence range. *RR*: relative risk. *SMR*: standardized mortality ratio. *Source: mod. from [@Impact_Jin].*[]{data-label="fig: Impact_3"}](Impact_3)
![[**(Top)**]{} Distribution of RRs for *increased mortality* (see fig.\[fig: ReE\_1\]). RRs have been taken from longitudinal cohort studies only as reported in fig. \[fig: Impact\_3\]. A basic statistical characterization of the distribution is reported in Box 1. By means of eq. \[eq: Impact\_4\] we have calculated the equivalent RR spread over the total population aged 15-64 (Box 2). [**(Middle)**]{} Comparison between the RR variability inferred from the longitudinal cohort studies and those published for the first (1991-94) and second (1998-2001) economic downturns (see fig. \[fig: Impact\_3\]). The picture shows that the two different groups of studies overlap for middle age-cohorts. For middle age-cohorts, the overlapping ranges are selected. For age-cohorts with no overlap, the Russian levels have been selected. [**(Bottom)**]{}. Values named [ “]{}Lower and Upper boundary" will be used to estimate CRIMI.[]{data-label="fig: Impact_5"}](Impact_5ENG)
![Summary of the main measured health impacts of the two late 20th century’s crisis. Numbers in brackets are 95% confidence range. *RR:* relative risk. *OR:* odds ratio. *Source: [@Impact_Men; @Impact_Kim; @Impact_Khang; @Impact_Waters]* []{data-label="fig: Impact_2"}](Impact_2)
![Results of studies of the use of healthcare services by unemployed people. Studies of visits to physicians (top) and of visits or admissions to hospitals (bottom). Numbers in brackets are 95% confidence range. *Source: mod. from [@Impact_Jin]*.[]{data-label="fig: Impact_4"}](Impact_4)
![Population projections. Four scenarios (PopMV, PopHV, PopLV, PopCFV) are shown based on different assumptions on mortality and birth rates. Value shown are in thousands. PopSV br: 1.7% (resp. PopSV br: 1.3%) is the HIGH (resp. LOW) birth-rate scenario. *Source: http://esa.un.org/unpp*.[]{data-label="fig: ReE_4"}](ReE_4)
![Age-related Mortality rate, Expected Life and Deceased/Survivor ratio. *Source: Mortality rate and life expectancy, Eurostat 2009. Decedent/Survivor ratio, mod. from [@RE_Przywara; @Impact_Raitano]*.[]{data-label="fig: ReE_6"}](ReE_6)
![Scenarios on Deceased/Survivor ratio. *Source: mod. from [@RE_Przywara]*.[]{data-label="fig: Impact_8"}](Impact_8)
![Scenarios on age-related healthcare costs (ARC). *Source: ARC1 mod. from [@Impact_EC]. ARC2, ARC3 and ARC4 mod. from [@RE_Przywara]*.[]{data-label="fig: ReE_7"}](ReE_7)
![CRIMI calculations: assumption on applied perturbation.[]{data-label="fig: Impact_7"}](Impact_7)
![Simulated CRIMI and CRIUI are shown for different econometric assumptions for a list of relative risks (RR) related to mortality rate and, respectively, healthcare utilization. The crisis effect is supposed to be taking place over a 5 years time, as pictured in fig. \[fig: Impact\_7\]. Please note that the results in the top table are not directly comparable to the ones showed in the CRIMI table on bottom left: the first ones are calculated by applying a RR which is *uniform* throughout age-cohorts, while the second ones have been calculated based on *age-dependent* RR tables reported in fig. \[fig: Impact\_5\]. CH: Constant Health, DC: Death-related Costs. Numbers in round brackets are negative numbers.[]{data-label="fig: ReE_10"}](ReE_10)
![Estimated impact of current economic crisis on public NHS healthcare expenditure for year 2015.[]{data-label="fig: Impact_9"}](Impact_9)
[99]{}
“Changes in morbidity and medical care utilization after the recent economic crisis in the Republic of Korea." International Monetary Fund October 2010 Country Report. [****]{} (2010).
S. Fachin and A. Gavosto, “The decline in Italian Productivity: a study in the estimation of long-run trends in total factor productivity with panel cointegration methods." LLEE Working Document [**50**]{}, (2007)
A. Bassanetti, M. Iommi, C. Jona-Lasinio, and F. Zolin, “La crescita dell’economia italiana negli anni novanta tra ritardo tecnologico e rallentamento della produtività." Temi di discussione della Banca d’Italia, n. 539. [****]{} (2004).
J. Adda, T. Chandola and M. Marmot, “Socio-economic status and health: causality and pathways." Journal of Econometrics [**112**]{}, 57-63 (2003)
D. E. Bloom, D. Canning and J. Sevilla, “The effect of health on economic growth: a production function approach." World Development [**32**]{}, 1-13 (2004)
J. Bound, M. Schoenbaum, T. R. Stinebrickner and T. Waidmann, “The dynamic effects of health on the labour force transitions of older workers." Labour Economics, 179-202 (1999)
P. Contoyannis and A. M. Jones, “Socio-economic status, health and lifestyle." Journal of Health Economics, [**23**]{} 965-995 (2004)
J. Currie and R. Hyson, “Is the impact of health shocks cushioned by socioeconomic status? The case of low birthweight." NBER Working Paper (1999)
M. A. Winkleby, D. E. Jatulis, E. Frank and S. P. Fortmann, “Socioeconomic status and health: how education, income and occupation contribute to risk factors for cardiovascular disease." American Journal of Public Health, [**82**]{} 816-820 (1992)
M. G. Marmot, M. Kogevinas and M. A. Elston, “Social-economic status and disease." Annual Review of Public Health, [**8**]{} 111-135 (1987)
N. Adler, T. Boyce and M. A. Chesney, “Socioeconomic status and health." American Psychologist, [****]{} 15-22 (1994)
H. Kim, W. J. Chung and Y. J. Song, “Changes in morbidity and medical care utilization after the recent economic crisis in the Republic of Korea." Bulletin of the World Health Organization, [**81**]{} 8 (2003)
S. Kwon and Y. Jung, “The impact of the global recession on the health of the people in Asia." 3rd China-ASEAN Forum on Social Development and Poverty Reduction, [****]{} (2009)
M. Landler, “Financial Chill May Hit Developing Countries." New York Times, September 25, [****]{} (2008)
D. Roodman, “History says financial crisis will suppress aid." Center for Global Development, 13 Oct 2008. [****]{} (2008)
For an introduction to Deceased/Survivor Ratio and Death-related costs see:\
A. Wong, P. H. M. Van Baal, H. C.vBoshuizen and J. J. Polder, “Exploring the influence of proximity to death on disease-specific hospital expenditures: a carpaccio of red herrings." Health Economics [**20**]{}(4) 379-400 (2011); A. Wong, “Describing, explaining and predicting health care expenditures with statistical methods.". [\[PhD thesis at Tilburg University\]](http://arno.uvt.nl/show.cgi?fid=122354); M. Raitano, “The Impact of Death-Related Costs on Health Care Expenditure: A Survey" [****]{} (2006); S. Gabriele, C. Cislaghi, F. Costantini, F. Innocenti, V. Lepore, F. Tediosi, M. Valerio and C. Zocchetti, “Demographic factors and health expenditure profiles by age: the case of Italy" [\[ENEPRI Report WP7\]](http://www.enepri.org/files/AHEAD/Reports/);
M. Seshamani and A. Gray, “Time to death and health expenditure: an improved model for the impact of demographic change on health care costs." Age and Ageing [**33**]{} 556-561 (2004)
T. T. Dang, P. Antolin and H. Oxley, Economics Department Working Papers No. 305. Organization for Economic Co- operation and Development [****]{} (2001)
T. Miller, “Increasing longevity and Medicare expenditures." Demography [**38**]{} 215-226 (2001)
B. Przywara, “Projecting future health care expenditure at European level: drivers, methodology and main result." EUROPEAN ECONOMY, Economic Papers [**417**]{} (2010)
T. Men, P. Brennan, P. Boffetta and D. Zaridze, “Russian mortality trends for 1991-2001: analysis by cause and region." BMJ [**327**]{}, 964 (2003)
J. Sachs, “Understanding [ “]{}shock therapy“.” Occasional paper. London: Social Market Foundation [****]{} (1994)
J. Wedel, “Collision and Collusion: The strange case of Western aid to Eastern Europe." New York: St. Martin’s [****]{} (2001)
P. Murrell, “What is shock-therapy? What did it do in Poland and Russia?." Post-Soviet Affairs [**9**]{}(2) 111-140 (1993)
V. Kontorovich, “The Russian health crisis and the economy." Communist and Post-Communist Studies [**34**]{} 221-240 (2001)
WHO, “Changes in morbidity and medical care utilization after the recent economic crisis in the Republic of Korea." Bulletin of the World Health Organization [**81**]{} (2003)
Y. Khang, J. W. Lynch and G. A. Kaplan, “Impact of economic crisis on cause-specific mortality in South Korea." International Journal Epidemiology [**34**]{} 1291-1301 (2005)
H. Waters, F. Saadah and M. Pradhan, “The impact of the 1997 East Asian economic crisis on health and health care in Indonesia." Health policy and planning [**18**]{} 172-181 (2003)
R. Moreno, G. Pasadilla and E. Remolona, “Asia’s financial crisis: lessons and policy responses." Pacific Basin Working Paper Series [****]{}PB98-02 (1998)
G. Aschinger, “An Economic Analysis of the East Asia Crisis." Pacific Basin Working Paper Series [****]{}March/April 55-63 (1998)
C. M. Siddique and C. D’Arcy, “Unemployment and health: an analysis of the Canada health survey." International Journal Health Services [**15**]{} 609-635 (1985)
D. Jones, K. Moser and A. Fox, “Unemployment and mortality in the OPCS longitudinal study." The Lancet [**2**]{} 8415, 1324-1329 (1984)
K. Moser, A. Fox, D. Jones and P. Goldblatt, “Unemployment and mortality: further evidence from the OPCS longitudinal study 1971-81." The Lancet [**327**]{} 8477, 365-367 (1986)
N. Beale and S. Nethercott, “The nature of unemployment morbidity. Recognition." Journal of the Royal College of General Practitioners [**38**]{} 310, 197-199 (1988)
N. Beale and S. Nethercott, “The nature of unemployment morbidity. Description." Journal of the Royal College of General Practitioners [**38**]{} 310, 200-202 (1988)
P. Yuen and R. Balarajan, “Unemployment and patterns of consultation with the general practitioner." BMJ [**298**]{} 1212-1214 (1989)
P. Martikainen, “Unemployment and mortality among Finnish men, 1981-1985." BMJ [**301**]{} 407-411 (1990)
L. Iversen, O. Andersen and P. Andersen, “Unemployment and mortality in Denmark, 1970-80." BMJ [**295**]{} 879-884 (1987)
C. Costa and N. Segman, “Unemployment and mortality." BMJ [**294**]{} 1550-1551 (1987)
M. W. Linn, R. S. Sandifer and S. Stein, “Effects of unemployment on mental and physical health." Am J Public Health [**75**]{} 502-506 (1985)
R. L. Jin, C. P. Shah, T. J. Svoboda, “The impact of unemployment on health: a review of the evidence." Canadian Medical Association Journal [**153**]{}(5) 529-540 (1985)
J. Frey, “Unemployment and health in the US." BMJ [**284**]{} 1112 (1982)
European Commission, Special Report n.4/2005 - European Economy (Directorate-General for Economic and Financial Affairs).
R. Aprile, “The impact of ageing on health and long-term care: the case of Italy." Bank of Italy, Workshop on Fiscal Sustainability: Analytical Developments and Emerging Policy Issues. [****]{} (2008) [\[The impact of ageing on health and long-term care: the case of Italy.\]](http://www.bancaditalia.it/studiricerche/convegni/atti/fiscal_sustainability/session_3/Aprile.pdf)
M. Raitano, ENEPRI Research Report N. 17 [****]{} (2006) [\[ENEPRI Research Report N. 17\]](http://www.enepri.org)
ConferenzaStato-Regioni, Rep. n. 243/CSR del 3 dicembre 2009 [\[http://www.governo.it/GovernoInforma/Dossier/patto salute/\]](http://www.governo.it/GovernoInforma/Dossier/patto salute/)
[^1]: Eurostat, data publicly available at *http://epp.eurostat.ec.europa.eu*.
[^2]: See eq. \[eq\_CRIMI\] and eq. \[eq\_CRIUI\].
[^3]: Source: Italian Ministry of Health, 2009.
[^4]: Source: Italian Ministry of Health, 2009.
[^5]: Based on previous studies [@Impact_Kim; @Impact_Jin], the impact of other economic crisis on mortality has been reported to last from 2 to 10 years. We consider a 5 years period as a reasonable [ “]{}average" duration.
[^6]: According to Eurostat, 2011Q4 unemployment rate amounts to 9.1% in Italy and to 10.6% in Europe (17 Countries). [\[Eurostat\]](http://epp.eurostat.ec.europa.eu/portal/page/portal/statistics/search_database)
[^7]: Note: data on the first line of the Yuen et al. study in fig. \[fig: Impact\_4\] include those with chronic illnesses. In order to minimize health-selection bias we only accounted for the value 1.53 reported on the second line.
[^8]: However, as it will be demonstrated hereafter, this is not a big issue: being interested in a differential effect, the detailed assumptions on the future projections of the different variables (population, D/S ratio, etc.) would affect the result as a second order correction (i.e. for what we are concerned in this study, they are negligible).
[^9]: As opposed to the other four scenario, this time we are making sure consistency between population and mortality rate projections is preserved.
[^10]: A set of mortality rates distinguished by age-cohort. Each one of them is applicable *today* to persons within the same age-cohort.
[^11]: The vector $m^{(a)}_{2010}$ corresponds to the second column in fig.\[fig: ReE\_6\]
[^12]: Within the CH model, the improvement in the age-related healthcare costs is a function of the change in the mortality rates; within the DC model, the healthcare expenditure depends on the number of expected deaths.
[^13]: Definitions for $H, P, S, GP, R$ and $m$ available at 2008 Italian Treasury economic statement, *www.salute.gov.it*.
[^14]: An RR of 1.00 has been assumed for P, R and m.
[^15]: GDP: EUR 1,520,346 mln; total NHS expenditure: EUR 109.7 bn; total regional Deficit: EUR 2,368 mln. Source: ISTAT, year 2009.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'PageRank is a well-known centrality measure for the web used in search engines, representing the importance of each web page. In this paper, we follow the line of recent research on the development of distributed algorithms for computation of PageRank, where each page computes its own PageRank value by interacting with pages connected over hyperlinks. Our approach is novel in that it is based on a reinterpretation of PageRank, which leads us to a set of algorithms with exponential convergence rates. We first employ gossip-type randomization for the page selections in the update iterations. Then, the algorithms are generalized to deterministic ones, allowing simultaneous updates by multiple pages. Finally, based on these algorithms, we propose a clustering-based scheme, in which groups of pages make updates by locally interacting among themselves many times to expedite the convergence. In comparison with other existing techniques, significant advantages can be exhibited in their convergence performance, as demonstrated via numerical examples using real web data, and also in the limited amount of communication required among pages.'
author:
- 'Atsushi Suzuki and Hideaki Ishii[^1][^2]'
bibliography:
- 'BibDataBase\_jun18.bib'
title: |
**Efficient PageRank Computation via Distributed Algorithms\
with Web Clustering**
---
Introduction
============
For search engines at Google, one of the many measures used for ranking the web pages in search results is the so-called PageRank. For each web page, the PageRank value provides a measure of its importance or popularity, which is based on the network structure of the web in terms of the hyperlinks. A page is considered more important and popular if it receives more hyperlinks from other pages and especially those that are important themselves. PageRank has received a great deal of interest in the context of complex networks as it is an effective measure of centrality; see, e.g., [@Gleich:15; @IshTem:14; @LanMey:06] and the references therein.
The problem of computing PageRank has been a subject of studies over the years. Despite the simple nature of the problem, because of the problem size involving billions of pages in the web, its efficient computation remains a difficult task. For centralized computation, the simple power method has been the realistic option for this reason. Alternative methods have been studied based on Monte Carlo simulations of the underlying Markov chain (e.g., [@AvrLitNem:07]) and distributed algorithms (e.g., [@SarMolPan:15; @ShiYuYan:03]).
This paper follows the line of recent research in systems and control, where PageRank has gained much attention from the viewpoint of distributed algorithms. The approach is to view each web page as an agent which computes its own PageRank value iteratively by communicating with neighbors connected via hyperlinks. In [@IshTem:10; @IshTem:14], it was pointed out that the problem shares similarities with the multi-agent consensus problem [@Bullo:18; @MesEge:10], and randomized distributed algorithms were developed. To cope with the network size, the pages determine to initiate updates randomly, which is called gossipping; for the use of randomization techniques in the systems control literature, see [@TemCalDab_book]. The method is guaranteed to converge in the mean-square sense. However, it involves the time averaging of the state values, resulting in the convergence rate of order $1/k$ with respect to the updating time $k$.
This approach has been further extended in different directions. An efficient computation scheme based on aggregation of pages is presented in [@IshTemBai:tac12], which provides another motivation of our study as we will discuss below. In [@ZhaCheFan:tac13], an alternative analysis of the algorithms was carried out based on methods in stochastic approximation. Moreover, in [@ChaHadRab:16; @IshTemBai:scl12; @LeiChe:15], different probability distributions are employed for the randomization. Related studies on distributed computation of PageRank include [@NazPol:11; @PolTre:12; @RavFraTem:15]. More in general, distributed computation of other network centrality measures is studied in, e.g., [@MonOliGas:18; @WanTan:15]. Other works considered the problem of optimizing PageRank for pages of interest by changing the link structure [@CsaJunBlo:14; @FerAkiBou:tac12] and a game theoretic analysis for enhancing PageRank via page aggregation [@MaeIshAlg:17].
More recently, distributed algorithms for PageRank demonstrating exponential convergence speeds were proposed. In [@YouTemQiu:17], the PageRank problem is formulated as a least squares problem and then a gradient-based distributed algorithm is applied. The algorithm in [@LagZacDab:17] introduces an additional feature to maintain the state to be a probability vector throughout the iterations. The work [@DaiFre:17] employs techniques from matching pursuit algorithms and presents a randomized version. In this paper, we introduce a new approach for interpreting the PageRank problem by reexamining its definition. The idea is quite simple, but as a consequence, we arrive at a set of very efficient distributed algorithms. We propose algorithms for both synchronous and asynchronous cases in the communication among the linked pages and fully analyze their convergence properties, which are shown to be exponential. In particular, for the asynchronous case, we first employ randomization-based gossipping, but then extend the approach to deterministic gossipping, where multiple pages may be selected to simultaneously make updates. As long as each page updates its state infinitely often, convergence to its corresponding PageRank value is guaranteed.
The highlight of this work is that through this development, we become able to construct an efficient algorithm based on clustering of web pages for distributed computation of PageRank. As the web inherently has a hierarchical structure, clustering can be easily carried out, for example, by grouping pages in the same domains or subdomains. In this context, instead of the pages, it is the groups that initiate updates for their member pages. When a group determines to do so, the pages in the group make calculations by interacting among themselves, which is equivalent to iterating infinitely many times. Such updates can actually be performed in one step, expressed as matrix operations involving only local states. Part of the computation can be completed offline based on the information of the link structure within the group. Hence, the additional requirement for computation should be limited. We demonstrate the fast convergence performance in numerical examples using real web data.
The novel aspects of our approach can be summarized as follows. First, the reformulation idea is simple and its advantage may not be immediately clear. This is partly because additional states are introduced for the pages, which may increase the computational burden. In fact, in the synchronous case, the convergence is not necessarily faster than the power method. Second, in the proposed algorithms, the states are guaranteed to reach the true PageRank values from below in a monotonic fashion. Hence, even if randomization is adopted, the responses of the states are smooth, which may explain the efficiency of the approach. Third, the pages communicate only over their outgoing hyperlinks and do not require the knowledge of the incoming ones; this is another advantage of the schemes in comparison with conventional methods.
In the clustering-based algorithm, pages within each group collectively update their values and the exact values of PageRank can be obtained. It is emphasized that this approach relies on the properties of the specific schemes developed in this paper. In gossipping for page selections, no specific randomization is required especially for obtaining the true PageRank. Moreover, multiple pages can make updates at the same time partly due to the simple communication scheme. It is remarked that for large-scale computation, a related, but slightly different approach based on web aggregation has been studied as well. There, aggregated PageRank values representing the groups are computed first, and then a more local computation takes place within groups to assign values to individual pages, which typically results in approximation in the final values. Such studies can be found, for example, in [@LanMey:06; @ZhuYeLi:05] using classical methods in Markov chains and in [@BroLem_infret:06] via extensive simulations. The work [@IshTemBai:tac12] developed a method motivated by the studies on large-scale systems based on singular perturbation analyses for Markov chains (e.g., [@AldKha:91]) and consensus networks (e.g., [@BiyArc:08]).
This paper is organized as follows: In Section \[sec:pagerank\], we first give a brief overview of the PageRank problem and then introduce an alternative formulation. In Section \[sec:dist\], a novel distributed algorithm based on randomized gossipping is presented along with an analysis on their convergence properties. In Section \[sec:generalized\], we extend our approach and develop a generalized distributed algorithm. This is then further exploited to deal with clustering-based calculations in Section \[sec:group\]. Illustrative numerical examples are provided in Section \[sec:example\]. The paper is finally concluded in Section \[sec:concl\]. Preliminary versions of this paper have appeared as [@SuzIsh:acc18; @SuzIsh:cdc19]. The current paper provides the full proofs of the results and extended discussions along with a numerical example of larger scale.
[*Notation*]{}: For vectors and matrices, inequalities are used to denote entry-wise inequalities: For $X,Y\in\mathbb{R}^{n\times m}$, $X\leq Y$ implies $x_{ij}\leq y_{ij}$ for all $i,j$; in particular, we say that the matrix $X$ is nonnegative if $X\geq 0$ and positive if $X> 0$. A probability vector is a nonnegative vector $v\in\mathbb{R}^n$ such that $\sum_{i=1}^n v_i = 1$. A matrix $X\in\mathbb{R}^{n\times n}$ is said to be (column) stochastic if it is nonnegative and each column sum equals 1, i.e., $\sum_{i=1}^n x_{ij}=1$ for each $j$. Let $\mathbf{1}_n\in\mathbb{R}^n$ be the vector whose entries are all $1$ as $\mathbf{1}_n:=[1\,\cdots\,1]^T$. For a discrete set $\mathcal{D}$, its cardinality is given by $\bigl|\mathcal{D}\bigr|$.
A Novel Approach Towards PageRank {#sec:pagerank}
=================================
In this section, we briefly introduce the notion of PageRank and its interpretation commonly employed for its computation. Then, we discuss an alternative formulation of the problem, which will lead us to a novel class of distributed algorithms.
The PageRank Problem
--------------------
The computation of PageRank proposed by Brin and Page [@BriPag:98] starts with regarding the entire web as a directed graph. Let $n$ be the number of pages; we assume $n\geq 2$ to avoid the trivial case. The web graph is given by $\mathcal{G}:=\left( \mathcal{V}, \mathcal{E}\right)$ where $\mathcal{V}:=\left\{ 1,2,\ldots,n \right\}$ is the set of vertices representing the web pages, and $\mathcal{E}$ is the set of hyperlinks connecting the pages. Here, $(i,j)\in \mathcal{E}$ holds if and only if page $i$ has a hyperlink to page $j$. Hyperlinks are not always mutual, so this graph is generally a directed graph. For node $i$, let the set of outgoing neighbors and that of incoming neighbors be given, respectively, by $\mathcal{L}_{i}^{\text{out}}:=\left\{ j:\,(i,j)\in\mathcal{E} \right\}$ and $\mathcal{L}_{i}^{\text{in}}:=\left\{ j:\,(j,i)\in\mathcal{E} \right\}$.
When a node does not have any outgoing link, it is referred to as a dangling node. Here, to simplify the discussion, we assume that all pages have at least one outgoing hyperlink. This is commonly done by slightly modifying the structure of the web, specifically by adding hyperlinks from such dangling nodes, which correspond to the use of back buttons; see, e.g., [@LanMey:06; @LeiChe:15] for more details.
Next, we define the hyperlink matrix $A=\left( a_{ij} \right)\in \mathbb{R}^{n\times n}$ of this graph by $$a_{ij}
:= \begin{cases}
\frac{1}{n_j} & \text{if $i\in\mathcal{L}_j^{\text{out}}$},\\
0 & \text{otherwise},
\end{cases}
% \label{def-of-A}$$ where $\mathcal{L}_{i}^{\text{out}}$ is the set of outgoing neighbors of page $i$ and $n_i$ is its cardinality. By the assumption that all pages have one or more hyperlinks, this matrix $A$ is stochastic. For the web consisting of $n$ pages, the PageRank vector $x^*\in \mathbb{R}^n$ is defined as $$x^{*}
= (1-m)Ax^{*} + \frac{m}{n}\mathbf{1}_n,
~~\mathbf{1}_n^{T}x^{*}=1,
\label{def-normal}$$ where the parameter is chosen as $m\in(0,1)$; in this paper, we take the commonly used value $m=0.15$. The definition in can be rewritten as $$x^{*}=M x^{*},~~\mathbf{1}_n^{T}x^{*}=1,
\label{def-normal-mod}
% \label{def-sum}$$ where the modified link matrix $M$ is given by $M=(1-m)A + (m/n) \mathbf{1}_n\mathbf{1}_n^{T}$. Since $M$ is a convex combination of two stochastic matrices $A$ and $(1/n)\mathbf{1}_n\mathbf{1}_n^{T}$, it is stochastic as well. It is now clear that $x^*$ is the eigenvector corresponding the eigenvalue 1 of the link matrix $M$.
For its computation, the PageRank vector $x^*$ can be obtained by solving the linear equation or . However, due to its large dimension, the computation must rely on simple algorithms. It is common to use the power method given by the iteration of the form $$x(k+1) = (1-m)Ax(k) + \frac{m}{n}\mathbf{1}_n,
\label{power}$$ where $x(k)\in\mathbb{R}^n$ is the state whose initial value $x(0)$ can be taken as any probability vector. By Perron’s theorem [@HorJoh:85], it follows that $x(k)\to x^*$ as $k\to \infty$.
![An example graph with seven nodes[]{data-label="example1"}](0.eps){width="6cm"}
\[example2\]Consider the web consisting of seven pages depicted in Fig. \[example1\]. We can calculate the PageRank vector of this graph as $$\begin{aligned}
x^{*}&=\bigl[
0.316 ~
0.259 ~
0.156 ~
0.132 ~%\\
% &\hspace*{2cm}
0.0951 ~
0.0214 ~
0.0214
\bigl]^T.
\end{aligned}$$ Note that the indices of the pages are given according to the order of their PageRank values. Pages 1 and 2 rank the first and second, which can be due to having, respectively, 4 and 3 incoming links. Pages 3 and 4 have only 1 incoming link, but take better rankings than page 5, which has 3 links. This is because the ranks depend not only on the number of incoming links, but also on the values of the pages from which the links originate. In this respect, pages 3 and 4 are clearly advantageous, being neighbors of pages 1 and 2. Pages 6 and 7 have no incoming hyperlink and, as a result, take the lowest possible value, which is equal to $m/n=0.15/7= 0.0214$.
Reformulation of the PageRank Problem {#kasaneawase}
-------------------------------------
Now, we present a new formulation of PageRank by transforming its original definition. This formulation becomes the key for developing novel distributed algorithms. The idea itself is simple, but its advantage in the context of distributed computation of PageRank will become clear.
The formula of PageRank in (\[def-normal\]) can be transformed as $$\begin{aligned}
& x^{*}
= (1-m)Ax^{*} + \frac{m}{n}\mathbf{1}_n
\nonumber\\
& \Longleftrightarrow ~
x^{*}
= \left( I-(1-m)A \right)^{-1}\frac{m}{n}\mathbf{1}_n
\nonumber\\
& \Longleftrightarrow ~
x^{*}
= \displaystyle
\sum_{t=0}^{\infty}\left( (1-m)A \right)^{t}
\frac{m}{n}\mathbf{1}_n.
\label{eq_kasaneawase}\end{aligned}$$ In the last transformation, the Neumann series (e.g., [@HorJoh:85]) is applied. Notice that $(1-m)A$ is a Schur stable matrix because the link matrix $A$ is stochastic with spectral radius equal to 1.
The formula in implies that the PageRank computation can be carried out iteratively in several ways. It is immediate to write down an equation for the state $x(k)\in\mathbb{R}^n$ given by $$x(k) = \displaystyle \sum_{t=0}^{k}\left( (1-m)A \right)^{t}
\frac{m}{n}\mathbf{1}_n.
\label{eqn:alt_formula}$$ Clearly, the power method in is a compact way to realize this using only $x(k)$ as the state.
Another approach is to use a slightly redundant iteration by using an additional state. This is denoted by $z(k)\in\mathbb{R}^n$. Set the initial states as $x(0)=z(0)=(m/n)\mathbf{1}_n$. Then, the update scheme of the two states is given as follows: $$\begin{split}
x(k+1) &= x(k) + (1-m)Az(k),\\
z(k+1) &= (1-m)Az(k).
\end{split}
\label{eqn:synch}$$ Through this alternative algorithm, we can obtain the PageRank vector $x^*$. We formally state this along with its other properties as a lemma in the following. Similar properties will appear in our development of distributed algorithms later.
\[lemm:synch\] In the update scheme in , the states $x(k)$ and $z(k)$ satisfy the following:
1. $z(k)\rightarrow 0$ as $k\rightarrow\infty$.
2. $x(k)\leq x(k+1)\leq x^*$ for $k$.
3. $x(k)\rightarrow x^*$ as $k\rightarrow\infty$.
*Proof:* (i) As the link matrix $A$ is stochastic, its spectral radius equals 1, and thus $(1-m)A$ is a Schur stable matrix. This implies that $z(k)$ converges to zero. (ii) Note that $z(k)\geq 0$ because $A$ is stochastic and $z(0)>0$. Furthermore, we have $x(0)>0$. Thus, it is clear that $x(k)$ is nondecreasing as a function of $k$. The fact that it is upper bounded by $x^*$ follows from (iii), which is shown next.
(iii) From , we can write $x(k)$ as $$\begin{aligned}
x(k)
&= x(0) + \displaystyle \sum_{t=1}^{k} z(t)
= x(0) + \displaystyle \sum_{t=1}^{k}
\left( (1-m)A \right)^{t} z(0)
\nonumber\\
&= \displaystyle \sum_{t=0}^{k}
\left( (1-m)A \right)^{t} \frac{m}{n}\mathbf{1}_n.
\label{eqn:prop:synch}\end{aligned}$$ This and indicate that $x(k)$ converges to $x^{*}$ as $k\to \infty$.
We have a few remarks on the alternative approach introduced above in comparison with the power method in . First, the computation requires the second state $z(k)$ in addition to $x(k)$. As seen in , this state $z(k)$ is integrated over time to compute $x(k)$ in . Second, the initial values of $x(k)$ and $z(k)$ are fixed to $(m/n)\mathbf{1}_n$, and there is no freedom in these choices. Hence, each time the computation takes place through the update scheme , the algorithm cannot make use of the PageRank values computed in the past as initial guesses. This point may be a limitation of this approach. Also, the initial states are not probability vectors as in the power method. In fact, $x(k)$ becomes a probability vector only after converging to $x^*$. Third, notice that $m/n$ is the minimum PageRank value, which will be assigned to pages having no incoming links. For such pages, the states will not change during the updates.
Distributed Algorithm for Synchronous Updates
---------------------------------------------
One interpretation of from the perspective of distributed algorithms can be given as follows:
1. At time $0$, all pages start with the value ${m}/{n}$.
2. At time $k$, each page attenuates its current value by $1-m$ and then sends it to its linked pages after equally dividing it. At that time, page $i$ computes the weighted sum of the values received from the neighbors having links to the page.
Though we do not discuss in this paper, there is a generalized PageRank definition which uses a probability vector $v\in\mathbb{R}^n$ instead of $(1/{n})\mathbf{1}_n$, that is, $x^*=(1-m)Ax^*+mv$. In such a case, the proposed algorithm can be modified by replacing the initial states with $x(0)=z(0)=mv$.
We finally present a distributed algorithm based on .
\[alg:1\]
1. For page $i$, set the initial values as $x_{i}(0)=z_{i}(0)=m/n$.
2. At time $k$, page $i$ transmits its value $z_i(k)$ to its neighbors along its outgoing links and then updates its states to obtain $x_i(k+1)$ and $z_i(k+1)$ according to $$\begin{aligned}
x_{i}(k+1)
&= x_i(k) + \displaystyle \sum_{j:\,i\in \mathcal{L}_j^{\text{out}}}\frac{1-m}{n_j}z_{j}(k),\\
z_{i}(k+1)
&= \displaystyle \sum_{j:\,i\in \mathcal{L}_j^{\text{out}}}\frac{1-m}{n_j}z_{j}(k).\end{aligned}$$
As we show through simulations in Section \[sec:example\], this synchronized algorithm may not be particularly fast, especially in comparison with the power method. Moreover, due to the additional state $z(k)$, the algorithm requires more memory. The advantage of the proposed reformulation however becomes evident in the asynchronous versions of this distributed algorithm, which will be presented in the next section.
Gossip-Type Distributed Algorithms {#sec:dist}
==================================
We now develop asynchronous versions of the distributed algorithm. They are based on randomized communication among the pages, which is referred to as gossipping.
In the asynchronous update schemes, at each time $k$, one page is randomly chosen, which transmits its current state value to the linked pages. We present two algorithms which differ in their probability distributions for selecting the updating pages. One uses the uniform distribution and the other is more general. In both cases, the distributions remain fixed throughout the execution of the algorithms; thus, the updating pages are chosen in an independently identically distributed (i.i.d.) manner. Denote by $\theta(k)\in\mathcal{V}$ the selected page at time $k$. In the section on numerical examples, it will be shown that nonuniform distribution may be beneficial from the perspective of convergence speed.
Algorithm Based on the Uniform Distribution {#unif-algo}
-------------------------------------------
We consider the case where the selection of the updating pages follows the uniform distribution. The proposed distributed algorithm for this case is provided below.
\[alg:2\]
1. For page $i\in\mathcal{V}$, set the initial values as $x_{i}(0)=z_{i}(0)={m}/{n}$.
2. At time $k$, select one page $\theta(k)$ based on the uniform distribution: $$\text{Prob}\bigl\{
\theta(k)=i
\bigr\} = \frac{1}{n}~~\text{for $i\in\mathcal{V}$}.
\label{uniform}$$
3. Page $\theta(k)$ transmits its value $z_{\theta(k)}(k)$ over its outgoing links to pages in $\mathcal{L}^{\text{out}}_{\theta(k)}$.
4. Each page $i\in\mathcal{V}$ updates its states to obtain $x_i(k+1)$ and $z_i(k+1)$ as $$\begin{aligned}
x_{i}(k+1)
&= \begin{cases}
x_i(k)+\frac{1-m}{n_{\theta(k)}}z_{\theta(k)}(k)
& \text{if $i\in\mathcal{L}^{\text{out}}_{\theta(k)}$},\\
x_i(k) & \text{otherwise,}
\end{cases}\notag\\%\label{eqn:gdra1}\\
z_{i}(k+1)
&= \begin{cases}
0 & \text{if $i=\theta (k)$},\\
z_i(k) + \frac{1-m}{n_{\theta(k)}} z_{\theta(k)}(k)
& \text{if $i\in\mathcal{L}^{\text{out}}_{\theta(k)}$},\\
z_i(k) & \text{otherwise.}
\end{cases}
\label{eqn:gdra1}
\end{aligned}$$
The resemblance of this algorithm to Algorithm \[alg:1\] is obvious. The two states $x_i(k)$ and $z_i(k)$ play similar roles in both algorithms. The differences are that in the asynchronous case, the updates are made with one neighbor at a time, and also both $x_i(k)$ and $z_i(k)$ are integrated over time. For $z_i(k)$, this was not the case in Algorithm \[alg:1\]. The two variables are updated differently when page $i$ is the selected page $\theta(k)$ at time $k$ as in such cases, its own $z_i(k)$ is set to zero. By contrast, in Algorithm \[alg:1\], $z_i(k)$ is zero only in the case where page $i$ has no incoming link.
We now rewrite this algorithm in a vector form. First, let $$Q := (1-m)A.
\label{eqn:Q}$$ Denote by $e_i$ and $q_i$, respectively, the $i$th columns of the $(n \times n)$-identity matrix and $Q$ for $i\in\mathcal{V}$. Then, we define the matrices $Q_i,R_i,S_i\in\mathbb{R}^{n\times n}$, $i\in\mathcal{V}$, by $$\begin{split}
Q_i &:= \begin{bmatrix}
e_1 & e_2 & \cdots & e_{i-1} & q_i & e_{i+1} &\cdots& e_{n}
\end{bmatrix},\\
R_i &:= \begin{bmatrix}
{0}_n & {0}_n & \cdots& {0}_n & q_i
& {0}_n & \cdots & {0}_n
\end{bmatrix},\\
S_i
&:= \begin{bmatrix}
e_1 & e_2 & \cdots & e_{i-1} & {0}_n
& e_{i+1} & \cdots & e_{n}
\end{bmatrix},
\end{split}
\label{eqn:QiRi}$$ where in both $Q_i$ and $R_i$, it is the $i$th column that is equal to $q_i$, and in $S_i$, only the $i$th column is zero. Note that the matrices $Q$, $Q_i$, $R_i$, and $S_i$ are all nonnegative matrices for $i\in\mathcal{V}$. Moreover, by definition, it holds $Q_i = R_i+S_i$.
As in the synchronous case, the initial states are taken as $x(0)=z(0)=({m}/{n})\mathbf{1}_n$. The update schemes in for the two states can be written in the compact form as $$\begin{aligned}
\begin{split}
x(k+1) &= x(k) + R_{\theta(k)} z(k),\\
z(k+1) &= Q_{\theta(k)} z(k),
\end{split}
\label{eqn:gdra1v}\end{aligned}$$ where $\theta(k)$ is the page selected for updating at time $k$ in step 2) of the algorithm.
We are now ready to present the main result for this distributed algorithm for PageRank computation. It shows that the true PageRank values can be obtained almost surely.
\[unif\_thm\] Under Algorithm \[alg:2\], the PageRank vector $x^*$ is computed with $x(k)\to x^{*}$ as $k\to \infty$ with probability $1$. In particular, the following two properties hold:
1. $x(k)\leq x(k+1)\leq x^{*}$ holds for $k\geq 0$.
2. $\mathbb{E}\left[ x(k)\right]\to x^{*}$ as $k\to \infty$, and the convergence speed is exponential.
We first show the following lemma regarding the synchronous update scheme . It is a simple result, but will be useful in the proofs of different results in this paper. Denote by $x_{\rm{s}}(k)$ and $z_{\rm{s}}(k)$ the states in . Let $$\widetilde{Q}_{\rm{s}}(k)
:= Q^{k+1}.
\label{eqn:Qtilde}$$ Also, let $q_{ij}:=\left[ Q \right]_{ij}=(1-m)a_{ij}$ for $i,j\in\mathcal{V}$. Clearly, $q_{ij}\geq 0$ holds for any $i,j$.
\[lem:synch\]The state $x_{\rm{s}}(k)$ of the synchronous update scheme can be expressed as $$x_{\rm{s}}(k)
= \frac{m}{n} \mathbf{1}_n
+ \displaystyle \sum_{t=0}^{k-1}
\widetilde{Q}_{\rm{s}}(t) \frac{m}{n} \mathbf{1}_n,
~~k\geq 0.
\label{eqn:x_s}$$ Moreover, the $(i,j)$th element of the matrix $\widetilde{Q}_{\rm{s}}(t)$ in can be written as $$% \displaystyle \sum_{t=0}^{k-1}
\bigl[ \widetilde{Q}_{\rm{s}}(t) \bigr]_{ij}
= %\sum_{t=0}^{k-1}
\sum_{m_1,\ldots,m_{t}\in\mathcal{V}}
q_{i m_t}q_{m_t m_{t-1}}\cdots q_{m_{1}j}
~~\text{for $i,j$}.
\label{eqn:Qs_ij}$$
In the summation in , it is clear that the term $q_{i m_t}q_{m_t m_{t-1}}\cdots q_{m_{1}j}$ is nonzero if and only if there exists a sequence of hyperlinks $(j,m_1)$, $(m_1,m_2),\ldots$, $(m_t,m_{t-1})$, $(i,m_t)\in\mathcal{E}$ in the web graph. It is however noted that such a property will not be explicitly used in our analysis.
*Proof:* From the update scheme in and the definition of $Q$ in , we have $$\begin{aligned}
x_{\rm{s}}(k+1) &= x_{\rm{s}}(k) + Q z_{\rm{s}}(k),~~
z_{\rm{s}}(k+1) = Q z_{\rm{s}}(k).\end{aligned}$$ Thus, it follows that $z_{\rm{s}}(k)
= Q^k z_{\rm{s}}(0)$. Furthermore, $$x_{\rm{s}}(k)
= x_{\rm{s}}(0)
+ \sum_{t=0}^{k-1} Q z_{\rm{s}}(t)
= x_{\rm{s}}(0)
+ \sum_{t=0}^{k-1} Q^{t+1} z_{\rm{s}}(0).
% = \frac{m}{n} \mathbf{1}_n
% + \displaystyle \sum_{t=0}^{k-1}
% \widetilde{Q}_{\rm{s}}(t) \frac{m}{n} \mathbf{1}_n.$$ Since the initial states are set as $x_{\rm{s}}(0)= z_{\rm{s}}(0)=(m/n)\mathbf{1}_n$, we obtain .
Because $\widetilde{Q}_{\rm{s}}(t)=Q^{t+1}$, we have that $\bigl[ \widetilde{Q}_{\rm{s}}(t) \bigr]_{ij}$ is the summation of all terms that can be expressed as the product $q_{i m_t}q_{m_t m_{t-1}}\cdots q_{m_{1}j}$ of $t+1$ elements in the matrix $Q$, where $m_1,m_2,\ldots,m_{t}$ are all taken from $\mathcal{V}$. Thus, we have .
*Proof of Theorem \[unif\_thm\]:* (i) Since both $Q_{\theta(k)}$ and $R_{\theta(k)}$ are nonnegative, by , $x(k)$ and $z(k)$ are nonnegative at all $k$. In particular, $x(k)$ is a nondecreasing function of time, i.e., $x(k)\leq x(k+1)$ holds.
Next, we show $x(k)\leq x^*$. This is done by proving $$x(k)\leq x_{\rm{s}}(k)~~\text{for $k\geq 0$},
\label{eqn:unif_thm:2a}$$ where $x_{\rm{s}}(k)$ is the state of the synchronous update scheme in . Then, by Lemma \[lemm:synch\](ii), we obtain $x(k)\leq x_{\rm{s}}(k)\leq x^{*}$.
For $k=0$, we have $x(0)= x_{\rm{s}}(0)$ and thus holds. For $k\geq 1$, the state $z(k)$ in can be written as $$z(k)
= Q_{\theta (k-1)}Q_{\theta (k-2)}\cdots Q_{\theta (0)}z(0). \nonumber$$ Thus, we can express $x(k)$ as $$\begin{aligned}
&x(k)
= x(0) + \displaystyle \sum_{t=0}^{k-1} R_{\theta (t)}z(t)\nonumber\\
&~~= \displaystyle\frac{m}{n}\mathbf{1}_n
+ \displaystyle \sum_{t=0}^{k-1}
R_{\theta (t)}Q_{\theta (t-1)}Q_{\theta (t-2)}\cdots Q_{\theta (0)}
\frac{m}{n}\mathbf{1}_n.
\label{eqn:xk}\end{aligned}$$ For $k\geq 0$, define the nonnegative matrix $\widetilde{Q}_{\theta}(k)\in\mathbb{R}^{n\times n}$ by $$\widetilde{Q}_{\theta}(k)
:= R_{\theta (k)}Q_{\theta (k-1)}Q_{\theta (k-2)}
\cdots Q_{\theta (0)}.
\label{eqn:Qthtea}$$ Then, we have from $$\begin{aligned}
x(k)
&= \frac{m}{n} \mathbf{1}_n
+ \displaystyle \sum_{t=0}^{k-1} \widetilde{Q}_{\theta}(t) \frac{m}{n}
\mathbf{1}_n.
\label{eqn:x_unif_q}\end{aligned}$$
By comparing with in Lemma \[lem:synch\], we observe that for establishing , it suffices to show the inequality below: $$\displaystyle \sum_{t=0}^{k-1} \widetilde{Q}_{\theta}(t)
\leq \displaystyle \sum_{t=0}^{k-1} \widetilde{Q}_{\rm{s}}(t)~~\text{for $k\geq 1$}.
\label{eqn:QQ_ineq}$$ In particular, we should show $$%\displaystyle \sum_{t=0}^{k-1}
\bigl[
\widetilde{Q}_{\theta}(t)
\bigr]_{ij}
\leq %\displaystyle \sum_{t=0}^{k-1}
\bigl[
\widetilde{Q}_{\rm{s}}(t)
\bigr]_{ij}
~\text{for $i,j$ and $t=0,\ldots,k-1$}.
\label{qneq}$$ The approach for its proof is to establish that any term appearing on the left-hand side in always appears in the right-hand side. Since all terms are nonnegative, the inequality implies . The right-hand side of is written out in of Lemma \[lem:synch\].
In what follows, we obtain the formula for the left-hand side of , that is, $\bigl[ \widetilde{Q}_{\theta}(t) \bigr]_{ij}$. Recall that by , we have $Q_i = R_i+S_i$. Hence, by using $R_i$ and $S_i$, we can write $\widetilde{Q}_{\theta}(t)$ as $$\begin{aligned}
\widetilde{Q}_{\theta}(t)
&= R_{\theta (t)}Q_{\theta (t-1)} Q_{\theta (t-2)}
\cdots Q_{\theta (0)}\notag\\
&= R_{\theta (t)}\left( R_{\theta (t-1)} + S_{\theta(t-1)} \right)
\cdots\left( R_{\theta (0)} + S_{\theta(0)} \right).
% \label{eqn:Qt1}\end{aligned}$$ We must compute the products of $R_i$ and $S_i$ appearing on the far right-hand side above.
To this end, we derive a formula for the product where $R_i$ appears $\ell\leq t+1$ times and $S_i$ appears $k_{\ell}-{\ell}\leq t$ times: $$R_{\theta (k_{\ell})}\underbrace{\cdots}_{\text{$0$ or more $S_i$}}
R_{\theta (k_{{\ell}-1})}\underbrace{\cdots}_{\text{$0$ or more $S_i$}}\cdots
R_{\theta(k_1)}\underbrace{\cdots}_{\text{$0$ or more $S_i$}},
\label{eqn:prod1}$$ where $0\leq k_1< \cdots< k_{\ell}=t\leq k-1$. By , in $R_i$, all elements except the $i$th column are $0$ while $S_i$ is equal to the identity matrix except for the $i$th column, which is a zero vector. These facts lead us to the following relation for arbitrary $i,j$: $$\begin{aligned}
R_{i} S_{j}&=
\begin{cases}
0 & \text{if $i=j$},\\
R_{i} & \text{otherwise},
\end{cases}\end{aligned}$$ Thus, the product in becomes zero if one of the following conditions holds:
- $\theta(k_{\ell})$ is equal to one of $\theta(k_{\ell}-1),\ldots,\theta(k_{{\ell}-1}+1)$;
- $\theta(k_{{\ell}-1})$ is equal to one of $\theta(k_{{\ell}-1}-1),\ldots,\theta(k_{{\ell}-2}+1)$;
- $\cdots$
- $\theta(k_1)$ is equal to one of $\theta(k_1-1),\ldots,\theta(0)$.
Otherwise, the term in reduces to $$\begin{aligned}
&R_{\theta (k_{\ell})}\underbrace{\cdots}_{\text{$0$ or more $S_i$}}
R_{\theta (k_{{\ell}-1})}\underbrace{\cdots}_{\text{$0$ or more $S_i$}}\cdots
R_{\theta(k_1)}\underbrace{\cdots}_{\text{$0$ or more $S_i$}}\notag\\
&~~= R_{\theta (k_{\ell})} R_{\theta (k_{{\ell}-1})} \cdots R_{\theta(k_1)}.
\label{eqn:prod2}\end{aligned}$$ For the product of $R_i$ above, we can use the formula $$\begin{aligned}
R_{i} R_{j}
&= q_{ij} R_{j}^{(i)},\end{aligned}$$ where the matrix $R_{j}^{(i)}\in\mathbb{R}^{n\times n}$ is nonzero only in the $j$th column as $$R_{j}^{(i)}
:= \begin{bmatrix}
{0}_n & {0}_n & \cdots& {0}_n & q_i
& {0}_n & \cdots & {0}_n
\end{bmatrix}.
% \label{eqn:R_ji}$$ We also need another formula that holds for arbitrary $i,j,m$: $$R_{j}^{(i)} R_m = q_{jm} R_{m}^{(i)}.$$ Then, applying these formulae to the product in repeatedly yields $$\begin{aligned}
&R_{\theta (k_{\ell})} R_{\theta (k_{{\ell}-1})} \cdots R_{\theta(k_1)} \notag\\
&~= q_{\theta(k_{\ell})\theta(k_{{\ell}-1})}q_{\theta(k_{\ell}-1)\theta(k_{{\ell}-2})}\cdots
q_{\theta(k_2)\theta(k_1)} R_{\theta(k_1)}^{(\theta(k_{\ell}))}.
% \label{eqn:prod3}\end{aligned}$$ Finally, the $(i,j)$th element is obtained as $$\begin{aligned}
&\left[
R_{\theta (k_{\ell})} R_{\theta (k_{{\ell}-1})} \cdots R_{\theta(k_1)}
\right]_{ij} \notag\\
&~= q_{\theta(k_{\ell})\theta(k_{{\ell}-1})}
q_{\theta(k_{\ell}-1)\theta(k_{{\ell}-2})}\cdots
q_{\theta(k_2)\theta(k_1)}
\left[
R_{\theta(k_1)}^{(\theta(k_{\ell}))}
\right]_{ij} \notag\\
&~= \begin{cases}
q_{i\,\theta(k_{\ell})}
q_{\theta(k_{\ell})\theta(k_{{\ell}-1})}
q_{\theta(k_{\ell}-1)\theta(k_{{\ell}-2})}\cdots
q_{\theta(k_2)\,j}\\
~~~~~~\text{if $\theta(k_1)=j$},\\
0~~~~\text{otherwise}.
\end{cases}
% \label{eqn:prod4}\end{aligned}$$
To prove , it remains to show the following: Given the sequence $\{\theta(t)\}_{t=0}^{k-1}$ and time $k$, for any sequence of nodes $(m_2,\ldots,m_{\ell})\in\mathcal{V}^{\ell-1}$ with ${\ell}\leq k-1$, there exists at most one sequence of time $0\leq k_1<\cdots<k_{\ell}\leq k-1$ such that $$\left[
R_{\theta (k_{\ell})} R_{\theta (k_{{\ell}-1})} \cdots R_{\theta(k_1)}
\right]_{ij}
= q_{i m_{\ell}} q_{m_{\ell} m_{{\ell}-1}}\cdots q_{m_{2}j}%\notag\\
% &\hspace*{6cm}~\text{for $i\in\mathcal{V}$},
\label{eqn:C_and_q}$$ for $i\in\mathcal{V}$, where in the left-hand side, the product of $R_i$ is interpreted as being obtained from . This is shown by establishing that if holds, then the sequence of time $k_1,\ldots,k_{\ell}$ is uniquely determined from $\{\theta(t)\}_{t=0}^{k-1}$. This can be done through the procedure below:
- $k_1$ is the smallest $t\geq 0$ such that $\theta(t)=j$;
- $k_2$ is the smallest $t> k_1$ such that $\theta(t)=m_2$;
- $\cdots$
- $k_{\ell}$ is the smallest $t> k_{{\ell}-1}$ such that $\theta(t)=m_{\ell}$.
We have now proven that (\[qneq\]) holds for arbitrary $i,j,k$.
(ii) Here, we study the average dynamics of the randomized update scheme . To this end, let the average matrices be given by $\overline{Q}:= \mathbb{E}\left[ Q_{\theta(k)} \right]$ and $\overline{R}:=\mathbb{E}\left[ R_{\theta(k)} \right]$. Since the updated pages are selected in an i.i.d. manner from the uniform distribution, we have $$\overline{Q}
=\frac{n-1}{n}I+\frac{1}{n}Q,~~
\overline{R}
=\frac{1}{n}Q.
\label{ex_a1}$$ Here, by , the expectation of $x(k)$ is obtained as $$\begin{aligned}
&\mathbb{E}\left[ x(k) \right]% \nonumber\\
= \mathbb{E}\biggl[ \displaystyle
\frac{m}{n}\mathbf{1}_n
+ \sum_{t=0}^{k-1}
R_{\theta(t)} Q_{\theta(t-1)}\cdots Q_{\theta(0)}
\frac{m}{n}\mathbf{1}_n
\biggr] \nonumber\\
&= \frac{m}{n}\mathbf{1}_n
+ \displaystyle
\sum_{t=0}^{k-1}\mathbb{E}\left[ R_{\theta(t)}\right]
\mathbb{E}\left[Q_{\theta(t-1)}\right]\cdots
\mathbb{E}\left[Q_{\theta(0)}\right]
\frac{m}{n}\mathbf{1}_n \nonumber\\
&= \frac{m}{n}\mathbf{1}_n
+ \displaystyle
\sum_{t=0}^{k-1}\overline{R} \overline{Q}^{t}
\frac{m}{n}\mathbf{1}_n.
\label{tmp-a1}\end{aligned}$$ Notice that in the second term in the far right-hand side above, the average link matrix $\overline{Q}$ in is Schur stable since it is a nonnegative matrix whose column sums are equal to ${(n-1)}/{n} + {(1-m)}/{n} = 1-{m}/{n} < 1$. Thus, taking the limit $k\to \infty$, we can apply the Neumann series as $$\displaystyle\lim_{k\to \infty}
\sum_{t=0}^{k-1}\overline {R}\overline{Q}^{t}
= \overline {R}
\left(I-\overline {Q}\right)^{-1}.$$ Moreover, by , we have $$\begin{aligned}
&\overline{R}
\left(
I-\overline {Q}
\right)^{-1}
\nonumber\\
&= \overline {R}
\biggl[
I-\biggl(
\frac{n-1}{n}I+\frac{1}{n}Q
\biggr)
\biggr]^{-1}
= n\overline {R}\left(I-Q \right)^{-1}
\nonumber\\
&= n\overline {R} \displaystyle
\lim_{k\to \infty}\sum_{t=0}^{k}Q^{t}
= \displaystyle\lim_{k\to \infty}Q\sum_{t=0}^{k} Q^{t}
= \displaystyle\lim_{k\to \infty}\sum_{t=1}^{k}Q^{t}.
\label{used_nounif}\end{aligned}$$ Substituting this into (\[tmp-a1\]) as $k\to\infty$ and by , we obtain $$\begin{aligned}
& \displaystyle\lim_{k\to \infty}
\mathbb{E}\left[ x(k) \right]
= \frac{m}{n}\mathbf{1}_n
+ \displaystyle\lim_{k\to \infty}
\sum_{t=0}^{k-1}\overline {R}\overline {Q}^{t}
\frac{m}{n}\mathbf{1}_n
\nonumber\\
&= \frac{m}{n}\mathbf{1}_n
+ \displaystyle\lim_{k\to \infty}
\sum_{t=1}^{k} Q^{t} \frac{m}{n}\mathbf{1}_n
= \sum_{t=0}^{\infty} Q^{t}\frac{m}{n} \mathbf{1}_n
= x^{*}.\end{aligned}$$
From (\[tmp-a1\]), it is clear that $\mathbb{E}\left[ x(k) \right]$ can be written as a step response of a stable linear time-invariant system. This implies that it converges to $x^*$ exponentially fast. This completes the proof of (ii). Finally, by property (i) above, $x(k)$ is monotonically nondecreasing and has an upper bound $x^*$, so $x(k)$ converges with probability 1. Then, due to property (ii), the convergence value for $x(k)$ is $x^*$.
Theorem \[unif\_thm\] guarantees that the proposed gossip-based algorithm computes the true PageRank almost surely in a fully distributed fashion. Each page keeps track of its states $x_i(k)$ and $z_i(k)$ and when chosen by $\theta(k)$, it transmits $z_i(k)$ to its neighboring pages along its hyperlinks. Such hyperlinks are clearly known to the pages and the necessary communication is limited with only one value at a time. Other pages not linked by page $\theta(k)$ will simply keep their states unchanged.
The convergence is shown to be exponential in the mean, that is, $\mathbb{E}[x(k)]$ goes to $x^*$ exponentially fast. It is interesting to note that in in the proof, we have not shown $\sum_{t=0}^{k-1}\overline {R}\overline {Q}^{t}=
\sum_{t=1}^{k}Q^{t}$, which would indicate that in the mean the system is the same as the synchronous one. Indeed, this equality holds only in the limit as $k\to\infty$. Our method is based on a simple reinterpretation of the definition of PageRank from the systems viewpoint, and it seems well suited for the PageRank computation in terms of convergence. The distributed algorithms proposed in [@DaiFre:17; @LagZacDab:17; @YouTemQiu:17] also have exponential convergence speeds (under different notions). The approaches there rely on techniques for distributed optimization. The work [@YouTemQiu:17] views the PageRank problem as a least-squares problem while [@DaiFre:17] employs a randomized version of the so-called matching pursuit algorithms. On the other hand, in [@LagZacDab:17], a modified gradient-descent algorithm is constructed so that the states of all pages remain to have the total equal to one throughout its execution.
It is highlighted that our approach has an advantage in terms of the communication loads for each node. As seen earlier, in the update scheme, the nodes need to transmit their values only over their outgoing links, and no further communication is necessary. The same type of communication scheme is adopted in [@FraIshRav:15], which is an extension of those in [@IshTem:10], and thus the algorithms there do not exhibit exponential convergence. Similarly, in the algorithm of [@DaiFre:17], the nodes utilize only the outgoing links, but there is a difference in that the nodes must also receive the values from the linked pages during the same time step, and hence communication is always bidirectional. Meanwhile, the knowledge of the pages connected by the incoming links is necessary in [@YouTemQiu:17]. The scheme in [@LagZacDab:17] requires communication along both incoming and outgoing links. In this respect, among the different approaches, our Algorithm \[alg:2\] is superior in requiring the least amount of communication per update. Further discussions on comparisons of the methods can be found in [@IshSuz:18].
Generalization to Non-Uniform Distributions {#nounif-algo}
-------------------------------------------
We next generalize the gossip-type distributed algorithm to the case where the pages will be chosen from distributions not limited to the uniform one. This extension is an interesting feature of the proposed approach and makes the algorithm more suitable for its use in distributed environments. For example, depending on the computational and communication resources, the pages or the servers that carry out the PageRank computation may adjust to update at different frequencies [@ChaHadRab:16]. The update scheme here follows Algorithm \[alg:2\]. Consider an i.i.d. random sequence $\{\theta(k)\}$ for the page selections. Let $p_i$ be the probability of page $i$ to be chosen. Assume that $p_i>0$ for $i$ and $\sum_{i=1}^n p_i = 1$. In Algorithm \[alg:2\], step 2) should be replaced with the following
1. Select one page $\theta(k)$ based on the distribution $p_i$: $$\text{Prob}\bigl\{
\theta(k)=i
\bigr\} = p_i~~\text{for $i\in\mathcal{V}$}.
\label{eqn:p_i}$$
For this algorithm, we now state the main result.
\[prop:nounif\_thm\]Under Algorithm \[alg:2\] using step 2)$'$ introduced above, the PageRank vector $x^*$ is computed with $x(k)\to x^{*}$ as $k\to \infty$ with probability $1$.
*Proof:* This proposition can be established similarly to Theorem \[unif\_thm\] by showing the properties (i) and (ii), where the main difference is in (ii). Let the matrix $P\in\mathbb{R}^{n\times n}$ be the diagonal matrix whose $i$th diagonal entry is $p_i$, i.e., $P:=\text{diag}(p_1,\ldots,p_n)$. By $p_i>0$, $P$ is nonsingular. Next, let the average matrices of $Q_{\theta(k)}$ and $R_{\theta(k)}$ be respectively $$\begin{aligned}
\overline{Q}^{\prime}
&:= \mathbb{E}\left[ Q_{\theta(k)} \right]=(I-P)+ QP,\\
\overline{R}^{\prime}
&:=\mathbb{E}\left[ R_{\theta(k)} \right]= QP.\end{aligned}$$ These matrices are nonnegative. Moreover, for $\overline{Q}^{\prime}$, the sum of its $i$th column is equal to $1-m p_i$; this means that it has the spectral radius $\max_i 1-m p_i <1$ and thus is Schur stable. Now, as in the discussion around (\[tmp-a1\]), we can establish $$\mathbb{E}\left[ x(k) \right]
= \frac{m}{n}\mathbf{1}_n
+ \displaystyle
\sum_{t=0}^{k-1}{\overline{R}^{\prime}}
\bigl({\overline{Q}^{\prime}}\bigr)^t
\frac{m}{n}\mathbf{1}_n.$$ For the summation in the right-hand side, take the limit $k\to \infty$ and then apply the Neumann series to obtain $$\begin{aligned}
&\displaystyle\lim_{k\to\infty}
\sum_{t=0}^{k-1}
{\overline{R}^{\prime}}
\bigl({\overline{Q}^{\prime}}\bigr)^t
= {\overline{R}^{\prime}}
\bigl(I-{\overline{Q}^{\prime}}\bigr)^{-1}
\nonumber\\
&~~= {\overline{R}^{\prime}}
\left[I-\left( (I-P)+QP \right)\right]^{-1}
= {\overline{R}^{\prime}}P^{-1}
\left(I-Q \right)^{-1}
\nonumber\\
&~~= {\overline{R}^{\prime}}P^{-1}
\displaystyle\lim_{k\to \infty}
\sum_{t=0}^{k} Q^{t}
= \displaystyle\lim_{k\to \infty}
\left( QP \right)P^{-1}\sum_{t=0}^{k} Q^{t}
\nonumber\\
&~~= \displaystyle\lim_{k\to \infty}
\sum_{t=1}^{k} Q^{t}. \end{aligned}$$ This expression is the same as (\[used\_nounif\]). The rest of the proof follows similarly to the proof of Theorem \[unif\_thm\](ii).
This gossip-type distributed algorithm can be carried out even if the probability distribution for the page selection is not uniform. Though other algorithms may be able to deal with non-uniform selections [@ChaHadRab:16; @IshTemBai:scl12; @LeiChe:15], in those cases, additional computations and adjustments are often required. In contrast, in our algorithm, no change is necessary, and the update scheme performed by each page is exactly the same. We have seen that the state values increase monotonically to reach the true PageRank. This might indicate that increasing the selection probability of a page with a large value may lead to faster convergence. We will examine this idea in the numerical example later.
Generalization of the Approach {#sec:generalized}
==============================
We extend the randomization-based distributed algorithms developed in the previous section in two directions to enhance their convergence performance and also the flexibility in implementation.
First, while Algorithm \[alg:2\] is restricted to allowing only one page to initiate an update at a time, here we realize simultaneous updates by multiple nodes in distributed computation. The other extension is to incorporate update times which are deterministic so that no randomization is necessary. It will be shown that an algorithm with these novel features possess similar convergence properties.
In the proposed algorithm, we denote the set of updating pages chosen at time $k$ by $\phi(k)\subset\mathcal{V}$. This set need not be randomly determined and may contain arbitrary number of page indices. We now introduce the algorithm in the following.
\[generalized\_algo\]
1. For each page $i\in\mathcal{V}$, set the initial states as $x_{i}(0)=z_{i}(0)=m/n$.
2. At time $k$, each page $i$ decides whether to make an update or not. Let $\phi(k)\subset\mathcal{V}$ be the set of indices of all pages that decided to make an update.
3. Each page $i\in\phi(k)$ transmits its value $z_{i}(k)$ over its outgoing links to pages in $\mathcal{L}^{\text{out}}_{i}$.
4. Each page $i\in\mathcal{V}$ makes an update in its states to obtain $x_i(k+1)$ and $z_i(k+1)$ as $$\begin{aligned}
&x_{i}(k+1)
= x_i(k)
+ \sum_{j\in \mathcal{L}_i^{\rm in}\cap\phi(k)}
\frac{1-m}{n_j}z_{j}(k),
\nonumber\\
&z_{i}(k+1) \nonumber\\
&~~~= \begin{cases}
\displaystyle
\sum_{j\in \mathcal{L}_i^{\rm in}\cap\phi(k)}
\frac{1-m}{n_j}z_{j}(k)
& \text{if $i\in\phi (k)$},\\
\displaystyle
z_i(k) + \sum_{j\in \mathcal{L}_i^{\rm in}\cap\phi(k)}
\frac{1-m}{n_j}z_{j}(k)
& \text{otherwise}.
\end{cases}
\label{eqn:update:alg3}\end{aligned}$$
This algorithm has a structure similar to that of Algorithm \[alg:2\]. The communication load is minimal since the pages that initiate updates in step 2) transmit only their states $z_i(k)$ and not $x_i(k)$, and this is done over their outgoing hyperlinks. In step 4), the update scheme for the states is a generalized version of the one in from Algorithm \[alg:2\]. The pages receiving state values over their incoming edges are characterized by having a nonempty set $\mathcal{L}_i^{\rm in}\cap\phi(k)$, and only these pages make changes in their states.
For the set $\phi\subset\mathcal{V}$ of chosen pages, we introduce three nonnegative matrices $Q_{\phi}$, $R_{\phi}$, and $S_{\phi}$ as $$\begin{aligned}
Q_{\phi}
&=\begin{bmatrix}
q_{1}(\phi) & \cdots & q_{n}(\phi)
\end{bmatrix},\notag\\
R_{\phi}
&=\begin{bmatrix}
r_{1}(\phi) & \cdots & r_{n}(\phi)
\end{bmatrix},
\label{eqn:alg4:QRS}\\
S_{\phi}
&=\begin{bmatrix}
s_{1}(\phi) & \cdots & s_{n}(\phi)
\end{bmatrix},\notag\end{aligned}$$ where the component vectors $q_i(\phi)$, $r_i(\phi)$, and $s_i(\phi)$, $i=1,\ldots,n$, are given by $$\begin{aligned}
q_{i}(\phi)
&= \begin{cases}
q_i & \text{if $i\in\phi$},\\
e_i & \text{otherwise},
\end{cases}~~
r_{i}(\phi)
= \begin{cases}
q_i & \text{if $i\in\phi$},\\
0_n & \text{otherwise},
\end{cases}\\
s_{i}(\phi)
&= \begin{cases}
0_n & \text{if $i\in\phi$},\\
e_i & \text{otherwise}.
\end{cases}\end{aligned}$$ We note that the following three relations hold: $$\begin{aligned}
R_{\phi}
&= \sum_{i\in\phi} R_i,
\label{eqn:alg4:relation1}\\
Q_{\phi}
&= R_{\phi} + S_{\phi},
\label{eqn:alg4:relation2}\\
R_i S_{\phi}
&= \begin{cases}
0 & \text{if $i\in\phi$},\\
R_{i} & \text{otherwise}.
\end{cases}
\label{eqn:alg4:relation3}\end{aligned}$$
With these matrices, we can write the update scheme of Algorithm \[generalized\_algo\] in a vector form as $$\begin{split}
x(k+1)
&= x(k) + R_{\phi(k)}z(k),\\
z(k+1)
&= Q_{\phi(k)}z(k),
\end{split}
\label{general_q}$$ where the initial states are $z(0)=x(0)=(m/n)\mathbf{1}_n$.
Regarding the choice of the sequence $\{\phi(k)\}$, we make the following assumption. It says that each page must initiate the updates of its states infinitely often over time.
\[generalized\_asm\]Each page $i\in\mathcal{V}$ is contained in infinitely many sets $\phi(0),\phi(1),\ldots,\phi(k),\ldots$.
We are now in the position to state the main result for the distributed algorithm with multiple updates.
\[geranalized\_conv\] Under Assumption \[generalized\_asm\], in the distributed algorithm with simultaneous updates of Algorithm \[generalized\_algo\], the state $x(k)$ converges to the true PageRank vector $x^*$, that is, $x(k)\to x^*$ as $k\to\infty$. If, in addition, for some $T>0$, each page updates at least once in every $T$ steps, then the convergence to $x^*$ is exponential.
The proof of this theorem is a generalization of that for Theorem \[unif\_thm\] for Algorithm \[alg:2\]. It consists of showing two properties similar to (i) and (ii) in Theorem \[unif\_thm\]. Here, we state the first property as a lemma. Its proof is given in Appendix \[appendix:1\], which follows similar lines as that of Theorem \[unif\_thm\], but is more technical and involved.
\[generalized\_monot\]Under Assumption \[generalized\_asm\], in the distributed algorithm with simultaneous updates of Algorithm \[generalized\_algo\], it holds $x(k)\leq x(k+1)\leq x^*$ for $k\in\mathbb{Z}_+$.
*Proof of Theorem \[geranalized\_conv\]*: By Lemma \[generalized\_monot\], the state $x(k)$ of Algorithm \[generalized\_algo\] converges to some vector $x'\leq x^*$. We must show that this $x'$ is always equal to $x^*$. Take an arbitrary $\varepsilon>0$. Let $k_{\text{s}}(\varepsilon)$ be the step number $k\geq 1$ such that under the synchronous update scheme , whose states are denoted by $x_{\text{s}}(k)$ and $z_{\text{s}}(k)$, the error bound of $$\|x_{\text{s}}(k)-x^*\|_1\leq\varepsilon$$ is achieved for the first time. By Lemma \[lemm:synch\], in the synchronous algorithm, the error $x_{\text{s}}(k)-x^*$ asymptotically converges to zero. Thus, for any given $\varepsilon$, a finite value for $k_{\text{s}}$ always exists.
Next, for a given sequence $\{\phi(k)\}_{k=0}^{\infty}$, we recursively define the time sequence $\{\delta(k)\}_{k=0}^{\infty}$ by $\delta(0)=-1$ and $$\delta(k)
=\min\bigg\{
\ell> \delta(k-1):\
\bigcup_{t=\delta(k-1)+1}^{\ell} \phi(t)
= \mathcal{V} %\left\{ 1,2,\cdots,n \right\}
\bigg\}
\label{eqn:delta}$$ for $k\geq 1$. By definition, during the time interval from $\delta(k-1)$ to $\delta(k)$, all nodes $1,2,\ldots,n$ are chosen at least once. By Assumption \[generalized\_asm\], this sequence $\{\delta(k)\}_{k=0}^{\infty}$ is well defined.
To establish that the state $x(k)$ converges to $x^*$, we must show that for arbitrary $\varepsilon$, the following inequality holds: $$\begin{aligned}
x^*
&> x(\delta(k_{\text{s}}(\varepsilon)))
\geq x_{\text{s}}(k_{\text{s}}(\varepsilon)).
\label{eqn:thm2:1}\end{aligned}$$ This relation indicates that as the state $x_{\text{s}}(k)$ of the synchronous algorithm converges to the PageRank vector $x^*$, it will be followed by the state $x(k)$ of Algorithm \[generalized\_algo\] at a slower speed governed by $\delta(k)$.
In what follows, we prove the relation $$\sum_{t=0}^{\delta(k_s(\varepsilon))}
\left[\widetilde Q_{\phi}(t) \right]_{ij}
\geq \sum_{t=0}^{k_s(\varepsilon)}
\left[\widetilde Q_{s}(t) \right]_{ij}~~
\text{for $i,j$}.
\label{general_lower}$$ Note that this relation is similar to in the proof of Lemma \[generalized\_monot\], but with the difference in the direction of the inequality and the times over which the summations are taken. By of Lemma \[lem:synch\], we can rewrite the right-hand side of as $$\sum_{t=0}^{k_s(\varepsilon)}
\left[ \widetilde{Q}_{\rm{s}}(t) \right]_{ij}
= \sum_{t=0}^{k_s(\varepsilon)}
\sum_{m_1,\ldots,m_t\in \mathcal V}
q_{im_t}q_{m_tm_{t-1}} \cdots q_{m_1j}.$$
We now focus on the left-hand side of . Note that similarly to the proof of Lemma \[generalized\_monot\], it is the sum of the terms appearing in at most once. For each $t\leq \{0,1,\ldots,k_s(\varepsilon)\}$, consider the ordered set of nodes, $(m_1,m_2,\ldots,m_t)\in \mathcal{V}^{t}$. We show that for each $i,j$, the term $q_{im_t}q_{m_tm_{t-1}}\cdots q_{m_1j}$ corresponding to this set of nodes appears once in the left-hand side of .
Consider the sequence $\left( k_{t+1},k_{t},\cdots,k_1 \right)$ of time, where
- $k_1$ is the smallest $\ell\geq0$ such that $m_1\in\phi(\ell)$;
- $k_2$ is the smallest $\ell>k_1$ such that $m_2\in\phi(\ell)$;
- $\ldots$
- $k_t$ is the smallest $\ell>k_{t-1}$ such that $m_t\in\phi(k)$.
The sequence $\left( k_{t+1},k_{t},\ldots,k_1 \right)$ exists. This is because by the choice of $\delta(\cdot)$ in , it holds $\delta(t-1)<k_t\leq \delta(t)$. Therefore, it follows that the left-hand side of is a sum of nonnegative terms, and moreover, it contains all terms appearing in the right-hand side. We therefore conclude that the relations and thus hold.
Since this holds for arbitrary $\varepsilon$, it finally follows that $x(k)$ converges to $x^*$. Under the additional assumption that each page updates at least once in every $T$ steps, it holds $\delta(k+1)-\delta(k)\leq T$. Since $x_{\text{s}}(k)$ exponentially converges to $x^*$, $x(k)$ does as well.
Clustering-Based Distributed Algorithm {#sec:group}
======================================
In this section, we develop a novel approach based on web clustering for the computation of PageRank.
In this context, web clustering means the following: (i) Prior to running the algorithm, we group the pages, preferably, having strong dependence on each other through hyperlinks. (ii) During the computation, we allow the pages within groups to communicate with each other for updating their states together. That is, in this case, the states are updated not by the individual pages but by the groups. In doing so, we assume that extra computation resources are available locally within the group, which will be exploited to expedite the convergence speed. This scheme is especially suitable in view of the structure of the web since pages belong to domains and subdomains, which can be directly considered as groups.
Even through the grouping, our approach is able to compute the true PageRank values and, more important, this can be realized much more efficiently in terms of computation speed. This advantage is realized by introducing extra local computation in the group-wise updates, where multiplications based on submatrices of the link matrices are performed. In previous research, grouping of pages often arises as part of an aggregation process, where PageRank values representing the groups are computed for reducing the size of the problem; see, e.g., [@BroLem_infret:06; @IshTemBai:tac12; @LanMey:06; @ZhuYeLi:05].
Grouping of Pages
-----------------
We partition the web consisting of $n$ pages into $N\leq n$ groups, denoted by $\mathcal{V}_{1},\mathcal{V}_{2},\ldots,\mathcal{V}_{N}\subset\mathcal{V}$. Here, let $l_h:=|\mathcal{V}_{h}|$ be the size of group $h=1,2,\dots,N$. Then, the constraints on the grouping are $l_h \geq 1$, $\bigcup_{h=1}^{N}\mathcal{V}_{h}=\mathcal{V}$, and $\mathcal{V}_{h_1}\cap\mathcal{V}_{h_2}=\emptyset$ for $h_1\neq h_2$. It is however expected that the convergence of the computation would be faster if groups are dense in the links among the group member pages and if the numbers of links going outside the groups are small.
The order of the indices can be changed without loss of generality and is done according to the grouping as follows: $$\begin{aligned}
\mathcal{V}_{1}&=\left\{ 1,2,\ldots,l_1 \right\},\\
\mathcal{V}_{2}&=\left\{ l_1+1,l_1+2,\ldots,l_1+l_2 \right\},\\
&\vdots\\
\mathcal{V}_{N}&=\left\{ n-l_N+1,n-l_N+2,\ldots,n \right\}.\end{aligned}$$ After this renaming of the pages, let $A$ be the link matrix of the web. Recall that $Q=(1-m)A$. We partition this matrix $Q$ according to the groups as $$Q = \begin{bmatrix}
\check{Q}_{11} & \check{Q}_{12} & \cdots & \check{Q}_{1N} \\
\check{Q}_{21} & \check{Q}_{22} & \cdots & \check{Q}_{2N} \\
\vdots & \vdots & \ddots & \vdots \\
\check{Q}_{N1} & \check{Q}_{N2} & \cdots & \check{Q}_{NN}
\end{bmatrix},
\label{eqn:Qpartition}$$ with the submatrices $\check{Q}_{ij}\in\mathbb{R}^{l_{i}\times l_{j}}$, $i,j=1,2,\ldots,N$. Similarly, the states $x(k)$ and $z(k)$ are partitioned as $$\begin{split}
x(k) &= \begin{bmatrix}
\check{x}_{1}(k)^T & \check{x}_{2}(k)^T &
\cdots & \check{x}_{N}(k)^T
\end{bmatrix}^{T},\\
z(k) &= \begin{bmatrix}
\check{z}_{1}(k)^T & \check{z}_{2}(k)^T &
\cdots & \check{z}_{N}(k)^T
\end{bmatrix}^{T},
\end{split}
\label{eqn:xz_group1}$$ where $\check{x}_{h}(k),\check{z}_{h}(k)\in\mathbb{R}^{l_h}$ denote the states for group $h$.
In the clustering-based algorithm to be presented, one group at each time makes an update. The group making updates at time $k$ is denoted by $\psi(k)\in\{1,2,\ldots,N\}$. In Algorithm \[generalized\_algo\] from Section \[sec:generalized\], this can be expressed as $\phi(k)=\mathcal{V}_{\psi(k)}$ at each time $k$. In the next subsection, we introduce a novel method for accelerating the convergence in the state updates.
Group-Based Update Scheme
-------------------------
The idea behind our approach for clustering-based computation is that at time $k$, the pages within the chosen group $\psi(k)$ make updates at once as in Algorithm \[generalized\_algo\]. The major difference however is that when they do so, they make a large number of updates by exploiting the information locally available within the group. More concretely, at time $k$, based on their present states $\check{x}_{\psi(k)}(k)$ and $\check{z}_{\psi(k)}(k)$, the members of the group $\psi(k)$ update their states infinitely many times. Then, the asymptotic values will be set as the next states $\check{x}_{\psi(k)}(k+1)$ and $\check{z}_{\psi(k)}(k+1)$. Note that the infinite updates are based only on intra-group communications, and those with pages in other groups will be performed only after the updates in the group are completed. We show that the infinite updates within the group can be done in one step, by small-scale matrix operations.
We first derive the update scheme for the states. At time $k$, assume that group $\psi(k)$ is chosen to make updates for the rest of the time $t\geq k$. We write the update scheme in with auxiliary states $x'(t|k)$ and $z'(t|k)$ whose initial values are set as $x'(k|k)=x(k)$ and $z'(k|k)=z(k)$. For time $t\geq k$, it holds $$\begin{aligned}
x'(t+1|k)
&= x'(t|k) + R_{\psi(k)} z'(t|k),\\
z'(t+1|k)
&= Q_{\psi(k)} z'(t|k).\end{aligned}$$ By using the partition of $Q$ in , and also by partitioning $x'(t|k)$ and $z'(t|k)$ as in , the states in the updating group $\psi(k)$ are given as $$\begin{split}
\check{x}'_{\psi(k)}(t+1|k)
&= \check{x}'_{\psi(k)}(t|k)
+ \check{Q}_{\psi(k)\psi(k)}
\check{z}'_{\psi(k)}(t|k),\\
\check{z}'_{\psi(k)}(t+1|k)
&= \check{Q}_{\psi(k)\psi(k)}
\check{z}'_{\psi(k)}(t|k).
\end{split}
\label{eqn:xz_psi1}$$ Moreover, the states in any remaining group $h\neq\psi(k)$ are updated by $$\begin{split}
\check{x}'_{h}(t+1|k)
&= \check{x}'_{h}(t|k)
+ \check{Q}_{h \psi(k)}
\check{z}'_{\psi(k)}(t|k),\\
\check{z}'_{h}(t+1|k)
&= \check{z}'_{h}(t|k)
+ \check{Q}_{h \psi(k)}
\check{z}'_{\psi(k)}(t|k).
\end{split}
\label{eqn:xz_h1}$$
As mentioned above, in this algorithm, the states $x(k+1)$ and $z(k+1)$ are taken as the limits of the states $x'(t|k)$ and $z'(t|k)$ with $t\to\infty$ while the same group $\psi(k)$ is continuously chosen infinitely many times. First, the state $\check{z}_{\psi(k)}(k+1)$ of the chosen group $\psi(k)$ is set to zero because by , we have $$\begin{aligned}
\check{z}_{\psi(k)}(k+1)
&= \lim_{t\to\infty}
\check{z}'_{\psi(k)}(t|k)\notag\\
&= \lim_{t\to\infty}
\check{Q}_{\psi(k)\psi(k)}^{t-k}
\check{z}_{\psi(k)}(k|k)
= 0.
\label{eqn:z_V_k1}\end{aligned}$$ Note that the submatrix $\check{Q}_{\psi(k)\psi(k)}$ is Schur stable since it is a submatrix of $Q$, which is nonnegative and Schur stable.
Second, from , the state $\check{z}_{h}(k+1)$ of group $h\neq \psi(k)$ can be obtained by using the relation in and by the Schur stability of the matrix $\check{Q}_{\psi(k)\psi(k)}$ as follows: $$\begin{aligned}
&\check{z}_{h}(k+1)
= \lim_{t\to\infty}
\check{z}'_{h}(t|k)\notag\\
&~= \check{z}'_{h}(k|k)
+ \lim_{t\to\infty}
\check{Q}_{h \psi(k)}
\sum_{l=k}^{t-1}
\check{z}'_{\psi(k)}(l|k)
\notag\\
&~= \check{z}'_{h}(k|k)
+ \lim_{t\to\infty}
\check{Q}_{h \psi(k)}
\sum_{l=0}^{t-k-1}
\check{Q}^{l}_{\psi(k)\psi(k)}
\check{z}_{\psi(k)}(k|k)\notag\\
&~= \check{z}_{h}(k)
+ \check{Q}_{h \psi(k)}
\Big(
I - \check{Q}_{\psi(k)\psi(k)}
\Big)^{-1}
\check{z}_{\psi(k)}(k),
\label{eqn:z_Vh2}\end{aligned}$$ where in the last equality, the Neumann series is used. Third, observe in and that the state $\check{x}_{h}(k+1)$ takes similar forms for both $h=\psi(k)$ and $h\neq\psi(k)$. Thus, by a derivation similar to , we have $$\begin{aligned}
&\check{x}_{h}(k+1)
= \lim_{t\to\infty}
\check{x}'_{h}(t|k)
\notag\\
&~= \check{x}'_{h}(k|k)
+ \lim_{t\to\infty}
\check{Q}_{h\psi(k)}
\sum_{l=k}^{t-1}
\check{z}'_{\psi(k)}(l|k)\notag\\
&~= \check{x}_{h}(k)
+ \check{Q}_{h\psi(k)}
\Big(
I - \check{Q}_{\psi(k)\psi(k)}
\Big)^{-1}
\check{z}_{\psi(k)}(k).
\label{eqn:x_Vpsi2}\end{aligned}$$
To summarize the discussion above, by –, we arrive at the following distributed algorithm for the PageRank computation based on web clustering.
\[alg:aggregation\]
1. For each group $h\in\{1,2,\ldots,N\}$, set the initial values of the states as $\check{x}_{h}(0)=\check{z}_{h}(0)=m/n\mathbf{1}_{l_h}$.
2. At time $k$, one group $\psi(k)\in\{1,2,\ldots,N\}$ is chosen for making updates in the states.
3. Using its own state $\check{z}_{\psi(k)}(k)$, group $\psi(k)$ computes the auxiliary vector $$\overline{\check{z}}_{\psi(k)}(k)
:= \big(
I-\check{Q}_{\psi(k)\psi(k)}
\big)^{-1}
\check{z}_{\psi(k)}(k)
\label{eqn:z_check_bar}$$ and transmits it over outgoing links to groups containing pages in $\mathcal{L}^{\text{out}}_{i}$ for $i\in\psi(k)$.
4. Each group $h$ updates its states to obtain $\check{x}_{h}(k+1)$ and $\check{z}_{h}(k+1)$ as follows: $$\begin{split}
\check{x}_{h}(k+1)
&= \check{x}_{h}(k)
+ \check{Q}_{h\psi(k)}
\overline{\check{z}}_{\psi(k)}(k),\\
% \Big(
% I-\check{Q}_{\psi(k)\psi(k)}
% \Big)^{-1}
% \check{z}_{\psi(k)}(k),\\
\check{z}_{h}(k+1)
&= \begin{cases}
0 & \text{if $h=\psi (k)$},\\
\check{z}_{h}(k)
% ~\mbox{}
+ \check{Q}_{h\psi(k)}
\overline{\check{z}}_{\psi(k)}(k)
% \Big(
% I-\check{Q}_{\psi(k)\psi(k)}
% \Big)^{-1}
% \check{z}_{\psi(k)}(k)\\
& \text{otherwise}.
\end{cases}
\end{split}
\label{eqn:alg:aggregation}$$
We interpret the updates made by one group $h=\mathcal{V}_{\psi(k)}$ at each time $k$ to be those made by the member pages in the group simultaneously. In this way, the argument in Section \[sec:generalized\] can be similarly applied to the clustering-based case.
This clustering-based algorithm has several advantageous features in terms of computation speed and distributed implementation. To be more specific, in this algorithm, one update by a group $h$ corresponds to an infinite number of updates by the pages in the group in Algorithm \[generalized\_algo\]. As seen in , it involves matrix operations of the size of the group. This can greatly accelerate the convergence in one step compared to the previous algorithms. The performance would likely improve especially by grouping the pages so that more groups consist of dense subgraphs in the web, and/or each group has a limited number of links going outside.
Furthermore, in comparison to Algorithm \[generalized\_algo\], the main additional computation in the iteration is step 3) for obtaining $\overline{\check{z}}_{\psi(k)}(k)$ in . Note however that this is done locally within the group, and $\overline{\check{z}}_{\psi(k)}(k)$ need not be stored for the next step. Also, the matrix inversion for computing $\left( I-\check{Q}_{hh} \right)^{-1}$ only once within each group $h$ can be performed offline prior to running the algorithm. We should note that the matrix $\check{Q}_{hh}$ may be a sparse matrix in general for PageRank, but the matrix $\left( I-\check{Q}_{hh} \right)^{-1}$ may have a dense structure. We express the algorithm above in a vector form. It is useful to notice that by the definition of $R_{\mathcal{V}_{\psi}(k)}$, we have $$\begin{aligned}
R_{\mathcal{V}_{\psi}(k)}^{t+1}
&= \begin{bmatrix}
0 & \check{Q}_{1\psi(k)}
\check{Q}_{\psi(k)\psi(k)}^{t} & 0\\
0 & \check{Q}_{2{\psi}(k)}
\check{Q}_{\psi(k)\psi(k)}^{t} & 0\\
\vdots & \vdots & \vdots\\
0 & \check{Q}_{N\psi(k)}
\check{Q}_{\psi(k)\psi(k)}^{t} & 0
\end{bmatrix}\end{aligned}$$ for $t\geq 0$, where only the columns corresponding to the chosen group $\mathcal{V}_{\psi}(k)$ are nonzero. Due to the matrix $\check{Q}_{{\psi}(k){\psi}(k)}$ being Schur stable, by applying the Neumann series, we obtain $$\begin{aligned}
\sum_{t=0}^{\infty}
R_{\mathcal{V}_{\psi}(k)}^{t+1}
% &= \sum_{t=0}^{\infty}
% \begin{bmatrix}
% 0 & \check{Q}_{1\mathcal{V}_{\psi}(k)}
% \check{Q}_{\mathcal{V}_{\psi}(k)\mathcal{V}_{\psi}(k)}^{t} & 0\\
% 0 & \check{Q}_{2\mathcal{V}_{\psi}(k)}
% \check{Q}_{\mathcal{V}_{\psi}(k)\mathcal{V}_{\psi}(k)}^{t} & 0\\
% \vdots & \vdots & \vdots\\
% 0 & \check{Q}_{N\mathcal{V}_{\psi}(k)}
% \check{Q}_{\mathcal{V}_{\psi}(k)\mathcal{V}_{\psi}(k)}^{t} & 0
% \end{bmatrix}\notag\\
&= \begin{bmatrix}
0 & \check{Q}_{1{\psi}(k)}
\big(
I-\check{Q}_{{\psi}(k){\psi}(k)}
\big)^{-1}
& 0\\
0 & \check{Q}_{2{\psi}(k)}
\big(
I - \check{Q}_{{\psi}(k){\psi}(k)}
\big)^{-1}
& 0\\
\vdots & \vdots & \vdots\\
0 & \check{Q}_{N{\psi}(k)}
\big(
I-\check{Q}_{\psi(k)\psi(k)}
\big)^{-1}
& 0
\end{bmatrix}.
\label{eqn:Q_h_psi2}\end{aligned}$$ Denote this matrix by $\widehat{R}_{\mathcal{V}_{\psi}(k)}$. Since the matrix $R_{\mathcal{V}_{\psi}(k)}$ is also Schur stable, it follows that $$\begin{aligned}
\widehat{R}_{\mathcal{V}_{\psi}(k)}
:=\sum_{t=0}^{\infty}
R_{\mathcal{V}_{\psi}(k)}^{t+1}
= R_{\mathcal{V}_{\psi}(k)}
\big(
I - R_{\mathcal{V}_{\psi}(k)}
\big)^{-1}.
\label{eqn:Q_h_psi3}\end{aligned}$$ Now, by and , the updates of $x(k)$ in of Algorithm \[alg:aggregation\] can be written as $$\begin{aligned}
x(k+1)
&= x(k)
+ %R_{\mathcal{V}_{\psi}(k)}
%\big(
% I - R_{\mathcal{V}_{\psi}(k)}
%\big)^{-1}
\widehat{R}_{\mathcal{V}_{\psi}(k)}
z(k).\end{aligned}$$ Further, based on and the definition of $S_{\mathcal{V}_{\psi(k)}}$ in , we can write the updates of $z(k)$ in as follows: $$\begin{aligned}
z(k+1)
&= S_{\mathcal{V}_{\psi(k)}}
\big(
I + \widehat{R}_{\mathcal{V}_{\psi}(k)}
% R_{\mathcal{V}_{\psi}(k)}
% \big(
% I - R_{\mathcal{V}_{\psi}(k)}
% \big)^{-1}
\big)z(k).\end{aligned}$$ The initial states are set as $x(0)=z(0)=(m/n)\mathbf{1}_n$.
We introduce an assumption regarding $\psi(k)$.
\[aggregation\_asm\]Each group $i$ is chosen infinitely many times in $\psi(0),\psi(1),\ldots,\psi(k),\ldots$.
The main result for this algorithm is stated as follows:
\[aggregation\_conv\] Under Assumption \[aggregation\_asm\], in Algorithm \[alg:aggregation\], the state $x(k)$ converges to the true PageRank vector $x^*$, that is, $x(k)\to x^*$ as $k\to\infty$. If, in addition, for some $T>0$, each group is chosen at least once in every $T$ steps, then the convergence to $x^*$ is exponential.
![Time responses of the synchronous algorithms for the small graph: The power method and Algorithm \[alg:1\][]{data-label="simple-cent"}](simple_cent.eps){width="9cm"}
The proof of this proposition relies on the following lemma, whose proof is presented in Appendix \[appendix:2\].
\[aggregation\_monot\]In the distributed algorithm based on clustering of Algorithm \[alg:aggregation\], it holds $x(k)\leq x(k+1)\leq x^*$ for $k\in\mathbb{Z}_+$.
With this lemma, the proof of Proposition \[aggregation\_conv\] itself follows similarly to that of Theorem \[geranalized\_conv\] and is hence omitted. The crucial difference however is due to the infinite updates made within groups in Algorithm \[alg:aggregation\]. This aspect becomes evident by comparing Lemma \[generalized\_monot\] (for Theorem \[geranalized\_conv\]) and Lemma \[aggregation\_monot\] (for Proposition \[aggregation\_conv\]). Specifically, the finite summation in in the proof of the former result will be replaced with an infinite one, which is found in in the proof of the latter.
Numerical Examples {#sec:example}
==================
In this section, we illustrate the performance of the proposed algorithms by numerical simulations and compare them with conventional methods. Our update schemes are applied to two graphs, a simple one and one from actual web data.
Small Graph {#case-simple}
-----------
We first use the simple graph with seven pages in Fig. \[example1\].
![Time responses of the asynchronous algorithms for the small graph: You et al. [@YouTemQiu:17], Dai and Freris [@DaiFre:17], Ishii and Tempo [@IshTem:10], and Algorithm \[alg:2\][]{data-label="simple-dist"}](simple_dist.eps){width="9cm"}
### Synchronous Algorithms
We compare two synchronous algorithms, the power method and our proposed Algorithm \[alg:1\]. The two algorithms differ in their initial states. The proposed algorithm requires $x(0)$ to be $(m/n)\mathbf{1}_n$ while the power method can take any initial state as long as it is a probabilistic vector; in this simulation for the latter, we also used uniform values, i.e., $(1/n)\mathbf{1}_n$. On the other hand, these two algorithms are both deterministic and, as a consequence, the responses of pages $6$ and $7$ become exactly the same. The time responses of the PageRank values for the seven pages are shown in Fig. \[simple-cent\]. We observe that the power method converges faster in most nodes. In the responses of the proposed algorithm, it is noticeable that they are nondecreasing with respect to time, a property shown in Lemma \[lemm:synch\](i). Also, recall that for pages 6 and 7, in the proposed algorithm, the PageRank values are equal to the assigned initial values $m/n$. Hence, for these pages, the proposed algorithm is faster.
### Gossip-Type Distributed Algorithms {#randomized_compare}
Next, we discuss the results for the proposed randomized distributed algorithm, Algorithm \[alg:2\], based on the uniform distribution for the gossip-based communication. We make comparisons of its performance with several randomized algorithms in the literature. Specifically, we implemented those of Ishii and Tempo [@IshTem:14], Dai and Freris [@DaiFre:17], and You, Tempo, and Qiu [@YouTemQiu:17]. In the latter two algorithms, the total number $n$ of pages in the web is considered unknown though it is of course needed for the calculation of the PageRank as defined in ; here, we assume that $n$ is known by all pages.
All four algorithms select one page at each time $k$ based on the uniform distribution, and we used the same sequence $\{\theta(k)\}$. Concerning the initial states, our proposed algorithm and that of [@DaiFre:17] require that the pages take a fixed value, respectively, equal to $m/n$ and $0$. Other algorithms have some freedom in the choices. Here, however, we set them so that all pages are given the same initial values and took $1/n$.
The time responses of the calculated PageRank values of the nodes are plotted in Fig. \[simple-dist\]. We omit the result for page $7$ as its behavior is similar to that of page 6. It is clear that the proposed algorithm is the fastest in terms of convergence time for all pages in comparison with other distributed randomized algorithms. The responses of the proposed algorithm are characteristic in that despite the randomization due to gossipping, they are very smooth and again nondecreasing as in Fig. \[simple-cent\].
![Time responses of the sums of errors for Algorithm \[alg:2\] in the small graph: Comparison between two distributions, uniform and non-uniform[]{data-label="nounif-dist"}](unif-nounif2.eps){width="8cm"}
### Comparison of Distributions in Page Selection
Here, we illustrate how the convergence speed can be improved by employing Algorithm \[alg:2\] with a non-uniform distribution for $\theta(k)$. As discussed earlier, it seems reasonable to increase the selection probability of pages expected to take larger PageRank values. We adjusted the probabilities so that pages having more incoming links are more likely to be selected. In particular, we assigned each page the probability proportional to its in-degree plus 1.
We made Monte Carlo simulations of 1,000 runs by executing Algorithm \[alg:2\] for two cases: One under the uniform distribution and the other under the non-uniform distribution. The time responses of the sample averages of the sum of the errors, i.e., $\|x(k)-x^*\|_1$, are shown in Fig. \[nounif-dist\]. The non-uniform distribution slightly accelerates the convergence. It remains to be investigated what kind of distribution can be beneficial.
![Web graph of the large network[]{data-label="fig:Lincoln_graph"}](Lincoln_graph2){width="9.9cm"}
![PageRank values of the pages in the large network: Markers are colored for groups larger than 20 nodes.[]{data-label="fig:PageRank_agg"}](PageRank_agg2){width="8.4cm"}
Clustering-Based Algorithm Using Real Web Data
----------------------------------------------
In this subsection, we apply the proposed algorithms including the one based on clustering to real data of the web and demonstrate their performance.
### Web Data and Clustering
As the web graph, we used data from the database [@StaCybRes:06] collected from Lincoln University in New Zealand in 2006. This data has been used as a benchmark for testing different algorithms related to PageRank in, e.g., [@FerAkiBou:tac12; @IshTem:14; @YouTemQiu:17]. In particular, we adopted the data from [@IshTem:14], which is slightly modified to remove unlinked pages and to add additional links for dangling nodes. It consists of $n=3,754$ pages with 40,646 hyperlinks. Further details regarding the data can be found in [@IshTem:14].
We ordered the pages alphabetically according to their address names and then grouped them. First, for the 2,891 pages in the university domain (www.lincoln.ac.nz), they were grouped based on the first subdomain names. Then, the remaining pages outside the university were grouped based on their domain names. In total, the number of groups is 718, numerically indexed starting with group 1 containing page 1.
The larger groups contain 1,502, 346, 282, 221 pages and so on, but then there were 594 groups with only one page and 78 groups with two pages. The graph structure is shown in Fig. \[fig:Lincoln\_graph\], where the dots indicate the nonzero entries of the hyperlink matrix $A$. Here, the colored columns correspond to some of the larger groups, containing more than 20 member pages; there are 10 such groups. Also, in Fig. \[fig:PageRank\_agg\], the PageRank values of the pages are plotted, with the same coloring scheme.
Observe that the pages of the group colored in blue (group 282, around page index 700, with 346 pages) and those in the group shown in cyan color (group 290, around page index 1,100, with 282 pages) take especially high PageRank values. As seen in Fig. \[fig:Lincoln\_graph\], these groups have fairly dense link structures within their groups and many incoming links from outside. The pages taking the two highest PageRank values (page indices 991 and 992) are the university search page, which form a group by themselves; they receive about 270 incoming links.
![Time responses of errors in asynchronous randomized algorithms: Algorithm \[alg:2\], Ishii and Tempo [@IshTem:10], You et al. [@YouTemQiu:17], Dai and Freris [@DaiFre:17], and Lagoa et al. [@LagZacDab:17][]{data-label="fig:single_update"}](random_result_1){width="8.3cm"}
### Comparison among Distributed Algorithms
We discuss the results for the asynchronous distributed algorithm, Algorithm \[alg:2\]. As in the previous subsection, we make comparisons of its performance with the algorithms from [@DaiFre:17], [@IshTem:14], [@LagZacDab:17], and [@YouTemQiu:17], which are all randomization based. All five algorithms select one page at each time $k$ based on the uniform distribution, and we used the same sequence $\{\theta(k)\}$. As initial states, all pages were given the same values. In our algorithms, this is $m/n$. For the algorithm of [@DaiFre:17], it was set to $0$, and in the remaining three algorithms, we took $1/n$.
In Fig. \[fig:single\_update\], the error $\|x(k)-x^*\|_1$ from the true PageRank vector $x^*$ is plotted in the logarithmic scale for all five algorithms. We must highlight that while most of them decrease exponentially fast, our proposed method is by far the fastest. The plot is cut at the error level of $10^{-2}$, but in fact, the decrease in error continues at this rate.
### Influence of Initial States
It may appear that the fast convergence of the proposed Algorithm \[alg:2\] is due to the restricted choice in the initial values. Since many pages in the web take very small values, assigning the smallest possible value $m/n$ as the initial values may be advantageous. To check this point, we also ran simulations of other methods by assigning $m/n$ as the initial states to pages having especially small PageRanks; other pages received values by equally dividing the remaining PageRank. However, the results did not exhibit major changes in the responses, at least at the scale visible in plots similar to Fig. \[fig:single\_update\]. Thus, we conclude that at least for this example, the influence of the initial states seems very limited.
### Clustering-Based Method
Finally, in Fig. \[fig:agg\_sim1\], we compare the performance of the two proposed algorithms and the (centralized) power method. To make the comparison fair, the horizontal axis is taken as the number of updated nodes. The page selection for Algorithm \[alg:2\] is by randomization and that for Algorithm \[alg:aggregation\] is periodic (i.e., selecting groups as $1,2,\ldots,n,1,2\ldots$). Though the power method is faster than Algorithm \[alg:2\], and also comparable with Algorithm \[alg:aggregation\] in the very beginning, the clustering-based method shortly catches up and shows faster convergence.
In Fig. \[fig:agg\_sim2\], we show an enlarged version of Fig. \[fig:agg\_sim1\] with markers $\times$ put at the times when updates by groups took place. It shows how for Algorithm \[alg:aggregation\], the error decreases when certain groups make updates in their state values. In fact, groups 282 and 290 mentioned above have major contributions here. This is likely because their member pages take large values.
Conclusion {#sec:concl}
==========
In this paper, we have developed a new class of distributed algorithms for the computation of PageRank based on a reinterpretation of its definition. We first have introduced two types of distributed algorithms, synchronous and asynchronous based on gossipping. Their exponential convergence properties have been established, and the relation of the proposed algorithms to those in the literature has been discussed. The second part of the paper has been devoted to their extensions to the case with multiple updates and grouping. We have shown that our algorithms exhibit superior performance through simulations using real web data. In future research, we will further analyze the convergence speeds of the algorithms and employ other schemes for page selections.
Proof of Lemma \[generalized\_monot\] {#appendix:1}
-------------------------------------
By and , we can easily show $z(k)\geq0$ and $0\leq x(k)\leq x(k+1)$. Thus, in the remaining, we must show $x(k)\leq x^*$. By using the state $x_{\rm{s}}(k)$ of the synchronous update scheme , it suffices to show $$x(k)\leq x_{\rm{s}}(k)~~\text{for $k\geq 0$}.
\label{eqn:unif_thm:2}$$ Then, by Lemma \[lemm:synch\](ii), we obtain $x(k)\leq x_{\rm{s}}(k)\leq x^{*}$.
For $k=0$, we have $x(0)= x_{\rm{s}}(0)$, and thus holds. For $k\geq 1$, by , the closed-form solution of $z(k)$ is given by $$z(k)
= Q_{\phi(k-1)} Q_{\phi(k-2)} \cdots Q_{\phi(0)} z(0)$$ and that of $x(k)$ as $$\begin{aligned}
&x(k)
= z(0)
+ \displaystyle
\sum_{t=0}^{k-1}
R_{\phi (t)}z(t)\notag\\
&~~= \frac{m}{n}\mathbf{1}_n
+ \sum_{t=0}^{k-1}
R_{\phi (t)}Q_{\phi (t-1)}Q_{\phi (t-2)}
\cdots Q_{\phi (0)}\frac{m}{n}\mathbf{1}_n .%\notag\\
% &~~= \frac{m}{n}\mathbf{1}_n
% + \sum_{t=0}^{k-1}
% \widetilde Q_{\phi}(t)\frac{m}{n}\mathbf{1}_n,
%\label{general_x}\end{aligned}$$ For $k\geq 0$, let $$\widetilde{Q}_{\phi}(k)
:= R_{\phi(k)} Q_{\phi(k-1)} Q_{\phi (k-2)}
\cdots Q_{\phi(0)}.$$ Then, we have $$x(k)
= \frac{m}{n}\mathbf{1}_n
+ \sum_{t=0}^{k-1}
\widetilde Q_{\phi}(t)
\frac{m}{n}\mathbf{1}_n.
\label{general_x}$$
Thus, for establishing , by of Lemma \[lem:synch\], we must show for each $k>0$ $$\sum_{t=0}^{k-1}
\bigl[
\widetilde Q_{\phi}(t)
\bigr]_{ij}
\leq \sum_{t=0}^{k-1}
\bigl[
\widetilde Q_{s}(t)
\bigr]_{ij}.
% ~~\text{for $i,j$}.
\label{general_q_neq1}$$ Since the matrices $\widetilde Q_{\phi}(t)$ and $\widetilde Q_{s}(t)$ are nonnegative, more specifically, it suffices to show that the following inequality holds element-wise: $$% \sum_{t=0}^{k-1}
\bigl[\widetilde Q_{\phi}(t) \bigr]_{ij}
\leq %\sum_{t=0}^{k-1}
\bigl[\widetilde Q_{s}(t) \bigr]_{ij}
~~\text{for $i,j$ and
$t=0,1,\ldots,k-1$}.
\label{general_q_neq}$$ The terms in the right-hand side of are written out in of Lemma \[lem:synch\]. Thus, we will obtain the expressions for $\bigl[ \widetilde{Q}_{\phi}(t) \bigr]_{ij}$ on the left-hand side. Here, due to the relations in and , we have $$\begin{aligned}
\widetilde Q_{\phi}(t)
&= R_{\phi (t)}Q_{\phi (t-1)}Q_{\phi (t-2)}
\cdots Q_{\phi (0)}\\
&= R_{\phi(t)}
\left(
S_{\phi(t-1)}+R_{\phi(t-1)}
\right)\cdots
\left(
S_{\phi(0)} + R_{\phi(0)}
\right)\\
&= \bigg(
\sum_{i\in\phi(t)}R_i
\bigg)
\bigg(
S_{\phi(t-1)}
+ \sum_{i\in\phi(t-1)}R_i
\bigg)\\
&\hspace*{2.5cm}\mbox{}
\cdots
\bigg(
S_{\phi(0)} + \sum_{i\in\phi(0)}R_i
\bigg).\end{aligned}$$ This indicates that $\widetilde Q_{\phi}(t)$ can be written as the sum of all matrices which are products of $t+1$ matrices of the form $$\begin{aligned}
&R_{m_l}S_{\phi(k_l-1)}
\cdots S_{\phi(k_{l-1}+1)}\nonumber\\
&\hspace*{.5cm}\cdot R_{m_{l-1}} S_{\phi(k_{l-1}-1)}
\cdots S_{\phi(k_{l-2}+1)}\nonumber\\
&\hspace*{1cm}\cdots
R_{m_{1}} S_{\phi(k_{1}-1)}
\cdots S_{\phi(0)},
\label{general_tenkai}\end{aligned}$$ where the indices $k_1,\ldots,k_l$ are taken such that $k_l=t$ and $0\leq k_1<k_2<\cdots <k_l$; moreover, $m_i$ are taken such that $m_i\in \phi(k_i)$, $i=1,2,\ldots,l$.
From , we have that the matrix product in is equal to either a nonzero matrix of the form $R_{m_l}R_{m_{l-1}}\cdots R_{m_{1}}\geq 0$ or a zero matrix. It becomes a zero matrix if the chosen sequence $\phi$ and the indices $m_{1},\ldots,m_{l}$ satisfy at least one of the following conditions:
- $m_{1}$ is contained in one of the sets $\phi(k_{1}-1),\ldots,\phi(0)$;
- $m_{2}$ is contained in one of the sets $\phi(k_{2}-1),\ldots,\phi(k_1+1)$;
- $\ldots$
- $m_l$ is contained in one of the sets $\phi(k_l-1),\ldots,\phi(k_{l-1}+1)$.
If none of the conditions holds, then the matrix product in becomes equal to $R_{m_{l}}R_{m_{l-1}}\cdots R_{m_{1}}$.
Next, we reduce the expression of $R_{m_{l}}R_{m_{l-1}}\cdots R_{m_{1}}$ to a product of scalars and matrices. Here, we can use the formula $$\begin{aligned}
R_{i} R_{j}
&= q_{ij} R_{j}^{(i)},\end{aligned}$$ where $R_{j}^{(i)}$ is nonzero only in the $j$th column as $$R_{j}^{(i)}
:= \begin{bmatrix}
{0}_n & {0}_n & \cdots& {0}_n & q_i
& {0}_n & \cdots & {0}_n
\end{bmatrix}.
\label{eqn:R_ji}$$ We also need another formula that holds for arbitrary $i,j,m$: $$R_{j}^{(i)} R_m = q_{jm} R_{m}^{(i)}.$$ By repeatedly applying these formulae, we obtain $$\begin{aligned}
&R_{m_{l}}R_{m_{l-1}}\cdots R_{m_{1}} \notag\\
&\hspace*{1cm}
= q_{m_{l}m_{l-1}}q_{m_{l-1}m_{l-2}}\cdots
q_{m_{2}m_{1}}
R_{m_{1}}^{(m_{l})}.
\label{generalized_rr}\end{aligned}$$ Here, the $(i,j)$th element can be expressed as $$\begin{aligned}
&\left[
R_{m_{l}}R_{m_{l-1}}\cdots R_{m_{1}}
\right]_{ij}\\
&~~= q_{m_{l}m_{l-1}}q_{m_{l-1}m_{l-2}}\cdots
q_{m_{2}m_{1}}\left[R_{m_{1}}^{(m_{l})}\right]_{ij}\\
&~~= \begin{cases}
q_{im_{l}}q_{m_{l}m_{l-1}}\cdots q_{m_{2}j}
& \text{if $m_1= j$},\\
0 & \text{otherwise}.
\end{cases}\end{aligned}$$
So far, we have shown that the summation $\sum_{t=0}^{k-1}\left[ \widetilde{Q}_{\phi}(t) \right]_{ij}$ in the left-hand side of can be described as a sum of the terms $q_{im_l}q_{m_lm_{l-1}}\cdots q_{m_2j}$ using the set of nodes, $\left( m_2,\ldots,m_l \right)\in\mathcal{V}^{l-1}$, $2\leq l\leq k-1$.
To establish the inequality in , we must prove that for each $i,j,k$ and each node sequence $\left( m_2,\ldots,m_{\ell} \right)\in\mathcal{V}^{\ell-1}$, the number of sequences of time $0\leq k_1<\cdots<k_{\ell}\leq k-1$ satisfying the following conditions is at most one: $$\left[
R_{\phi (k_{\ell})}R_{\phi (k_{\ell-1})}\cdots R_{\phi(k_1)}
\right]_{ij}
= q_{im_{\ell}}q_{m_{\ell} m_{\ell-1}}\cdots q_{m_2j}.
\label{general_meidai}$$ We can find the times $k_1,\ldots,k_{\ell}$ so that the left-hand side of becomes nonzero in a unique manner by the given sequence of $\phi(k)$ through the following procedure:
- $k_1$ is the smallest $t\geq0$ such that $m_1=j\in\phi(t)$;
- $k_2$ is the smallest $t>k_1$ such that $m_2\in\phi(t)$;
- $\ldots$
- $k_{\ell}$ is the smallest $t>k_{\ell-1}$ such that $m_{\ell}\in\phi(k)$.
If there are $k_1,\ldots,k_{\ell}$ such that $j\in\phi(k_1)$, $m_2\in\phi(k_2)$, $\ldots$, $m_{\ell}\in\phi(k_{\ell})$, then the corresponding term in becomes zero. In conclusion, there is at most one combination of times $k_1,\ldots,k_{\ell}$ for which holds.
To summarize, we have that for each $2\leq \ell\leq k-1$, the left-hand side of contains at most one term expressed as $q_{i m_{\ell}}q_{m_{\ell}m_{\ell-1}}\cdots q_{m_2j}\geq0$. As shown in , the right-hand side always contains one term $q_{im_{\ell}}q_{m_{\ell}m_{\ell-1}}\cdots q_{m_2j}\geq0$. Therefore, for each $i,j,t$, the inequality holds. This implies $x(k)\leq x^*$, and consequently, we arrive at $x(k)\leq x(k+1)\leq x^*$.
Proof of Lemma \[aggregation\_monot\] {#appendix:2}
-------------------------------------
We can write $z(k)$ as $$\begin{aligned}
z(k)
&= S_{\mathcal{V}_{\psi(k-1)}}
\left(
I + \widehat{R}_{\mathcal{V}_{\psi}(k-1)}
\right)
S_{\mathcal{V}_{\psi(k-2)}}%\nonumber\\
\left(
I+ \widehat{R}_{\mathcal{V}_{\psi(k-2)}}
\right)\notag\\
&\hspace*{2cm}\mbox{}
\cdots
S_{\mathcal{V}_{\psi(0)}}
\left(
I+\widehat{R}_{\mathcal{V}_{\psi(0)}}
\right)z(0).
\label{aggr_z}\end{aligned}$$ Moreover, $x(k)$ can be written as $$\begin{aligned}
x(k)
&= z(0)
+ \sum_{t=0}^{k-1}
\widehat{R}_{\mathcal{V}_{\psi(t)}} z(t)
\nonumber\\
&= x(0)
+ \sum_{t=0}^{k-1}
\widehat{R}_{\mathcal{V}_{\psi(t)}}
S_{\mathcal{V}_{\psi(t-1)}}
\left(
I+\widehat{R}_{\mathcal{V}_{\psi(t-1)}}
\right)\notag\\
&\hspace*{2.2cm}\cdots
S_{\mathcal{V}_{\psi(0)}}
\left(
I+\widehat{R}_{\mathcal{V}_{\psi(0)}}
\right)z(0)\notag\\
&=: \frac{m}{n}\mathbf{1}_n
+ \sum_{t=0}^{k-1}
\widehat{Q}_{\psi(t)} \frac{m}{n}\mathbf{1}_n.
\label{aggr_x}\end{aligned}$$ Since $Q_{\mathcal{V}_{\psi(t)}}$, $R_{\mathcal{V}_{\psi(t)}}$, $S_{\mathcal{V}_{\psi(t)}}$, and $z(0)$ are all nonnegative, $z(k)$ is nonnegative as well. Therefore, the term $\widehat{R}_{\mathcal{V}_{\psi(k)}}z(t)$ appearing in $x(k)$ is nonnegative. Hence, it holds $x(k)\leq x(k+1)$.
It thus suffices to show that for each $k$ $$\sum_{t=0}^{k-1}
\left[
\widehat{Q}_{\psi(t)}
\right]_{ij}
\leq \sum_{t=0}^{\infty}
\left[\widetilde Q_{s}(t) \right]_{ij}
~~\text{for $i,j$}.
\label{aggr_proof_1}$$ Note that the summation on the right-hand side has infinite terms, which is different from the relations used in other proofs such as that of Theorem \[unif\_thm\]. We can however still use the expression for $\widetilde Q_{s}(t)$ given in of Lemma \[lem:synch\].
In what follows, we show that the left-hand side of is the sum of terms written only in the form of $q_{im_{\ell}}q_{m_{\ell}m_{\ell-1}}\cdots q_{m_1j}\geq 0$ and, moreover, these terms are all distinct in that each term is different from others. That is, the terms appearing in the left-hand side form a subset of those in the right-hand side of , confirming the inequality.
In , observe that $\widehat{Q}_{\psi(t)}$ consists of terms only of the form $$R_{\mathcal{V}_{\psi(t-1)}}^{\alpha_t + 1}
S_{\mathcal{V}_{\psi(t-2)}}
R_{\mathcal{V}_{\psi(t-2)}}^{\alpha_{t-1}}\cdots
S_{\mathcal{V}_{\psi(0)}}
R_{\mathcal{V}_{\psi(0)}}^{\alpha_0},
\label{aggr_q_elem}$$ where $\alpha_{t},\ldots, \alpha_{0}$ are nonnegative integers. Here, note that by the definitions of $R_{\mathcal{V}_h}$ and $S_{\mathcal{V}_h}$, this product is either a zero matrix or a product of matrices $R_{\mathcal{V}_h}$. Moreover, the $(i,j)$th element of the matrix in is nonnegative and is a summation of terms only expressed as $q_{i m_{\ell}}q_{m_{\ell} m_{\ell-1}}\cdots q_{m_1j}$ for $\ell> 0$.
Based on this observation, for establishing , we must show that for each term expressed as $q_{im_{\ell}}q_{m_{\ell}m_{\ell-1}}\cdots q_{m_1j}$, there is only one matrix product in the form of whose $(i,j)$th element contains the term.
For each page $m$, denote by $g(m)$ the index of the group to which it belongs. Then, it can be confirmed that a term expressed as $q_{im_{\ell}}q_{m_{\ell}m_{\ell-1}}\cdots q_{m_1j}$ is contained in $$\begin{aligned}
\left[
R_{\mathcal{V}_{ g\left( m_{\ell} \right) }}
R_{\mathcal{V}_{ g\left( m_{\ell-1} \right) }}\ldots
R_{\mathcal{V}_{ g\left( j \right) }}
\right]_{ij}.
\label{aggr_qq}\end{aligned}$$ To find the product of the form that contains the term $q_{im_{\ell}}q_{m_{\ell}m_{\ell-1}}\cdots q_{m_1j}$, we consider inserting matrices $R_{\mathcal{V}_h}$ and $S_{\mathcal{V}_h}$ between the matrices in anywhere except on the left side of $R_{\mathcal{V}_{ g\left( m_{\ell} \right) }}$.
First, if we insert any matrix $R_{\mathcal{V}_h}$ in , then the term $q_{im_{\ell}}q_{m_{\ell}m_{\ell-1}}\cdots q_{m_1j}$ will not be present any more. Next, we consider inserting $S_{\mathcal{V}_h}$ in . By definition, it holds $$\begin{aligned}
R_{\mathcal{V}_{h_1}}S_{\mathcal{V}_{h_2}}
&= \begin{cases}
0 & \text{if $h_1=h_2$},\\
R_{\mathcal{V}_{h_1}} & \text{if $h_1\neq h_2$}.
\end{cases}\end{aligned}$$ Thus, if $S_{\mathcal{V}_{h_1}}$ is inserted on the right side of a matrix $R_{\mathcal{V}_{h_2}}$ with $h_1\neq h_2$, then the term $q_{im_{\ell}}q_{m_{\ell}m_{\ell-1}}\cdots q_{m_1j}$ will remain in the product. This fact indicates that $S_{\mathcal{V}_{ g\left(m_{u} \right) }}$ cannot be inserted between matrices $R_{\mathcal{V}_{g(m_u)}}$ and $R_{\mathcal{V}_{g(m_{u-1})}}$ whose indices satisfy $g(m_{u})=g(m_{u-1})$.
Now, we can find the specific matrix product of the form containing the term $q_{im_{\ell}}q_{m_{\ell}m_{\ell-1}}\cdots q_{m_1j}$. First, the product should contain $\ell+1$ matrices of the type $R_{\mathcal{V}_h}$. Thus, the product would be written as $$\begin{aligned}
&R_{\mathcal{V}_{ g\left( m_{\ell} \right) }}
\underbrace{\cdots}_{0 \rm{\;or\; more\;}
S_{\psi}}R_{\mathcal{V}_{ g\left( m_{\ell-1} \right) }}
\underbrace{\cdots}_{0 \rm{\;or\; more\;}
S_{\psi}}\cdots
\nonumber\\
&R_{\mathcal{V}_{ g\left( m_{1} \right) }}
\underbrace{\cdots}_{0 \rm{\;or\; more\;}
S_{\psi}}R_{\mathcal{V}_{ g\left( j \right) }}
\underbrace{\cdots}_{0 \rm{\;or\; more\;}
S_{\psi}}.
\label{aggr_rs}\end{aligned}$$ For the given sequence $\psi(0),\ldots,\psi(k-1)$, this product is a nonzero matrix if there exists a sequence $0\leq i_0 \leq \cdots \leq i_{\ell}\leq k-1$ of time instants satisfying the following:
- $i_0$ is the smallest $k\geq 0$ such that $g\left( j \right)\in\psi(k)$;
- $i_1$ is equal to $i_0$ if $g\left( m_1 \right)=g\left( j \right)$; otherwise, it is the smallest $k>i_0$ such that $g\left( m_1 \right)\in\psi(k)$;
- $\cdots$
- $i_{\ell}$ is equal to $i_{\ell-1}$ if $g\left( m_{\ell}\right)=g\left( m_{\ell-1} \right)$; otherwise, it is the smallest $k>i_{\ell-1}$ such that $g\left( m_{\ell} \right)\in\psi(k)$.
It is clear that if such a sequence $i_0,\ldots, i_{\ell}$ exists, then it is unique.
Therefore, we conclude that the left-hand side of consists of terms only in the form $q_{im_{\ell}}q_{m_{\ell}m_{\ell-1}}\cdots q_{m_1j}$ and they are all distinct.
[^1]: A. Suzuki and H. Ishii are with the Department of Computer Science, Tokyo Institute of Technology, Yokohama, 226-8502, Japan. E-mails: [email protected], [email protected]
[^2]: This work was supported in part by the JST CREST Grant No. JPMJCR15K3 and by JSPS under Grant-in-Aid for Scientific Research Grant No. 15H04020 and No. 18H01460.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'A biexciton in a semiconductor quantum dot is a source of polarization-entangled photons with high potential for implementation in scalable systems. Several approaches for non-resonant, resonant and quasi-resonant biexciton preparation exist, but all have their own disadvantages, for instance low fidelity, timing jitter, incoherence or sensitivity to experimental parameters. We demonstrate a coherent and robust technique to generate a biexciton in an InGaAs quantum dot with a fidelity close to one. The main concept is the application of rapid adiabatic passage to the ground state-exciton-biexciton system. We reinforce our experimental results with simulations which include a microscopic coupling to phonons.'
author:
- Timo Kaldewey
- Sebastian Lüker
- 'Andreas V. Kuhlmann'
- 'Sascha R. Valentin'
- Arne Ludwig
- 'Andreas D. Wieck'
- 'Doris E. Reiter'
- Tilmann Kuhn
- 'Richard J. Warburton'
title: 'Coherent and robust high-fidelity generation of a biexciton in a quantum dot by rapid adiabatic passage'
---
Entangled photon pairs are a powerful resource, especially for quantum teleportation and quantum key distribution protocols. Spontaneous parametric down-conversion in non-linear optics is a source of entangled photon pairs [@Kwiat1995], but success is not guaranteed – the emission is a probabilistic process – and the error rate is high. In contrast, semiconductor quantum dots (QDs) are bright, on-demand sources of both single photons [@He2013] and entangled photon pairs and hence have enormous potential in quantum computing and quantum cryptography [@Gisin2002].
A biexciton in a QD is the starting point for a two photon cascade: when perfectly prepared, biexciton decay leads to the subsequent emission of two photons, Fig. \[fig:fig1\](f). In a QD without a significant fine structure splitting (FSS), the two photons are polarization-entangled [@Muller2014]. The majority of InGaAs QDs show a FSS due to a reduced symmetry [@Gammon1996; @Li2000_JAP; @Juska2013]. However, sophisticated techniques were developed to compensate for the FSS with strain [@Trotta2014], electric [@Kowalik2005] or magnetic fields [@Bayer2002; @Stevenson2006] and with special growth conditions [@Juska2013].
Several approaches for biexciton preparation have been proposed [@Glassl2013; @Debnath2013; @Gawarecki2012; @Hui2008] and demonstrated [@Brunner1994; @Stufler2006; @Jayakumar2013; @Muller2014; @Gotoh2013; @Bounouar2015]. Resonant two-photon schemes involving Rabi rotations [@Stufler2006; @Jayakumar2013; @Muller2014] are sensitive to fluctuations in both laser power and QD optical frequency. They are likely to suffer from an imperfect biexciton preparation resulting in undesired exciton photons unrelated to the cascade process.
A more robust scheme using phonon-assisted excitation was reported by several groups recently [@Ardelt2014; @Gotoh2013; @Quilter2015; @Bounouar2015; @Jayakumar2013]. An impressively high biexciton occupation of up to 95% was demonstrated using this quasi-resonant scheme [@Bounouar2015]. But the strength here is also a weakness. The scheme relies on the coupling to the phonon bath in the semiconductor environment: it is an inherently incoherent process. Also, a dependence on relaxation processes in the state preparation results in a timing jitter. In some cases, charge carrier relaxation times can reach values of up to a nanosecond [@Reithmaier2014].
We present here a coherent technique to create a biexciton with high probability, low jitter and weak dependence on the excitation and system parameters. The technique is based on rapid adiabatic passage (RAP). RAP allows the robust creation of an exciton [@Simon2011; @Wu2011; @Mathew2014] via a process requiring two-levels. RAP is applied here to the ground state-exciton-biexciton system, $\ket{0} - \ket{\mathrm{X}^{0}} -\ket{\mathrm{2X}^{0}}$, a three-level system [@Hui2008; @Glassl2013], and allows biexciton creation without significant exciton creation. In the implementation here, we use the full bandwidth of ultra-short pulses allowing us to access spectrally both the ground state-exciton and exciton-biexction transitions within one laser pulse. The broad bandwidth pulses have the advantage over the narrow bandwidth pulses suggested in [@Glassl2013] of enhanced robustness owing to stronger avoided crossings and a decoupling of phonons even for negative chirped pulses. We describe this process theoretically and demonstrate excellent agreement with the experimental results. Moreover, we analyze the influence of phonons on the preparation protocol. For RAP-based exciton creation, the influence of phonons depends sensitively on the sign of the chirp [@Luker2012; @Mathew2014; @glassl2011lon]. We find also in RAP-based biexciton creation that we can choose the chirp such that the phonons are unimportant at low temperature.
![ **Resonance fluorescence (RF) on a single quantum dot excited with a narrowband continuous wave laser and with broadband laser pulses.** **(a)** Structure of the n-i-p diode with epitaxial gate. **(b)-(c)** RF with narrowband excitation. **(b)** $\mathrm{X}^{1-}$ RF signal versus excitation power. **(c)** RF versus laser detuning at an excitation power of (points). A Lorentzian fit (red solid line) determined the FWHM linewidth, . **(d)** RF with broadband pulsed excitation as a function of gate voltage. The laser polarization was linear, the chirp positive and the center frequency resonant with the neutral exciton $\mathrm{X}^{0}$. **(e)** Scheme of a folded $4f$ pulse-shaper controlling the chirp introduced into an ultra-short, transform-limited laser pulse. **(f)** Generation of a biexciton with broadband, chirped excitation. Scheme of the two-photon cascade (left) after chirped excitation of a QD (right). []{data-label="fig:fig1"}](biexciton_paper_fig1_QD_structure_comp.eps){width="1\columnwidth"}
We study self-assembled InGaAs QDs at a temperature of . The QDs are grown by molecular beam epitaxy and embedded in an n-i-p or n-i-Schottky structure, Fig. \[fig:fig1\](a) (with more details in the supplementary information (SI) [@SI]). The bias voltage allows control over the QD charge via Coulomb blockade; and within a charging plateau, control over the optical resonance frequency via the DC Stark shift, Fig. \[fig:fig1\](d). The biexciton binding energy is positive and in the order of a few meV, a typical feature for InGaAs QDs. On driving the optical resonance with a narrowband continuous wave laser and detecting the resonance fluorescence (RF), we find that QDs in both samples have linewidths below , Fig. \[fig:fig1\](c), close to the transform limit [@Kuhlmann2013_NatPhys]. Above saturation on a QD in the n-i-p sample, Fig. \[fig:fig1\](a), we detect a RF single photon count-rate of , Fig. \[fig:fig1\](b). In the RAP experiments, we excite single QDs with the full bandwidth of pulses with a center wavelength of around and linear polarization. The spectral full-width-at-half-maximum (FWHM) of the pulses is $\Delta \lambda=\SI{10}{\nano\meter}$. This allows us to address the exciton and biexciton optical transitions (but not the transitions involving higher shells in the QD) with just one laser pulse. The transform-limited pulses (repetition rate of ) from a passively mode-locked laser were manipulated in a folded $4f$ pulse-shaper [@Martinez1987], Fig. \[fig:fig1\](e), in order to introduce chirp [@SI]. We control precisely the sign and magnitude of the chirp [@SI]. The FWHM of the pulse duration in intensity is stretched up to $\Delta t=\SI{15}{\pico\second}$ covering chirp coefficients up to $|\alpha|=\SI{0.70}{\pico\second\squared}$ For more details about the set-up, we refer to the SI [@SI].
The electronic structure in QDs is in general complex. For example in II-VI semiconductor colloidal QDs the electronic structure of multi exciton complexes was studied [@Kambhampati2012] and several biexciton states were identified [@Sewall2009]. However, in the III-V semiconductor QDs studied here, only the s-shells are populated such that the system can be described by four states, namely the ground state $\ket{0}$, the two bright exciton states $\ket{\mathrm{X}^{0}_{H/V}}$ and a single biexciton $\ket{\mathrm{2X}^{0}}$, Fig. \[fig:fig2\](a). The p-shells are energetically far removed even when using the full bandwidth of the ultrafast laser pulses. The energy of the biexciton state $\ket{\mathrm{2X}^{0}}$ is $\hbar\omega_{2\mathrm{X}^{0}} = 2\hbar\omega_{\mathrm{X}^{0}} - \Delta_B$ where $\hbar\omega_{\mathrm{X}^0}$ is the energy of the neutral exciton and $\Delta_B$ the biexciton binding energy. These energies are determined experimentally from the emission spectrum. Considering only one linear polarization, H or V, and assuming that the laser pulses are much faster than the fine structure-induced quantum beat, a three-level system with only one exciton state, $\ket{\mathrm{X}^{0}}$, is sufficient.
The exciton and biexciton state are coupled to longitudinal acoustic (LA) phonons in the pure dephasing regime via the deformation potential [@SI]. We use standard GaAs parameters and take the excitation parameters from the experiment leaving the QD size as the only fitting parameter. This model has been successfully used to describe the phonon influence on Rabi rotations [@Reiter2014] and on the population inversion via RAP [@Luker2012; @Glassl2013]. For the occupations of the exciton and biexciton states as well as for the coherences between all involved states a set of equations of motion is derived within a well-established fourth-order correlation expansion method [@krugel2005the]. The equations are then solved numerically.
![**Biexciton generation on QD1.** **(a)** Energy level scheme. **(b)** Instantaneous eigenenergies, the dressed states. **(c)** $\mathrm{2X}^{0}$ emission and **(d)** $\mathrm{X}^{0}$ emission as a function of the time-averaged square-root excitation power for different chirp parameters. **(e)** and **(f)** Simulation of the biexciton occupation and the sum of exciton and biexciton occupations as a function of the pulse area. **(g)** Ratio of the experimental data: $\mathrm{2X}^{0}/\mathrm{X}^{0}$. []{data-label="fig:fig2"}](biexciton_paper_fig2_niSchottky.eps){width="1\columnwidth"}
The goal of our work is high fidelity biexciton generation using RAP. Ideally, a vanishing intermediate occupation of the exciton is required. Less critical is a residual occupation of the ground state: this reduces only the rate of entangled photon emission but not the entanglement. A figure of merit is the ratio of biexciton photons to exciton photons. The ideal case is a ratio of one where every exciton photon is part of the two photon cascade on biexciton decay, Fig. \[fig:fig1\](f). In determining a biexciton creation fidelity we assume that biexciton and exciton photons are created and detected with equal probability. This is likely to be a very good assumption. The biexciton could in principle decay by an Auger process but in practice the probability of an Auger process is very small (probability of 0.23% [@Kurzmann2016]); and a re-excitation process during the cascade decay is also very unlikely on account of the hierarchy of times (the pulse duration (up to ) is much shorter than the radiative lifetime () which is much shorter than the time separating successive pulses ()). Therefore, we define here the fidelity as the ratio of detected photons from the transition $\left|\mathrm{2X}^0\right>\rightarrow\left|\mathrm{X}^0\right>$ ($2\mathrm{X}^0$ photon) and from the transition $\left|\mathrm{X}^0\right>\rightarrow\left|0\right>$ ($\mathrm{X}^0$ photons).
Fig. \[fig:fig2\] shows data from a single QD in the n-i-Schottky device. The laser pulses are centered at the two-photon biexciton resonance at . The $\mathrm{2X}^0$ and $\mathrm{X}^0$ emission intensities and their ratio are shown as a function of the square-root of the excitation power in Fig. \[fig:fig2\](c), (d) and (g), respectively. For close-to-zero chirp (black curves in Fig. \[fig:fig2\]), Rabi rotations are observed. At the first maximum, the $\mathrm{X}^0$ and $\mathrm{2X}^0$ emission intensities are equal to within an error of 5%, then both curves go down nearly to zero. At the second maximum, the $\mathrm{2X}^0$ emission reduces to 75% of the $\mathrm{2X}^0$ emission. For higher excitation power, we observe mainly $\mathrm{X}^0$ emission with little emission from the $\mathrm{2X}^0$. At the first maximum it is clear that ultrafast pulses with close-to-zero chirp enable high fidelity preparation of the biexciton by a Rabi rotation. However, this $\pi$-pulse excitation is very sensitive to variations in the detuning or excitation power. We turn to RAP which potentially offers a more robust scheme.
To create the biexciton using RAP, we first concentrate on positive chirp. Introducing a chirp of + stretches the pulse to an intensity FWHM of . In the experiment, the $\mathrm{2X}^0$ emission rises more slowly with pulse power than in the Rabi rotation experiment (red curve in Fig. \[fig:fig2\]) but then reaches a very broad maximum. Both $\mathrm{2X}^0$ and $\mathrm{X}^0$ signals reach a common maximum where the signals correspond closely to the maximum achieved in the Rabi rotation experiment. These are the main signatures of RAP. In terms of $\mathrm{2X}^0$ preparation, the ratio of $\mathrm{2X}^0$ to $\mathrm{X}^0$ emission reaches the ideal case of 100% with an error of 5%, proving that we can achieve a high fidelity biexciton preparation with RAP, robust against power and detuning fluctuations.
We now focus on the response of the system to a negative chirp of (pulse duration of , blue curves in Fig. \[fig:fig2\]). In this case the $\mathrm{2X}^0$ signal rises at smaller pulse areas than for positive chirp and in fact at even smaller pulse areas than the Rabi rotations. The maximum of the $\mathrm{2X}^0$ signal is considerably less than the signal following a $\pi$ Rabi pulse. The maximum $\mathrm{X}^0$ signal is likewise reduced. The highest $\mathrm{2X}^0$ to $\mathrm{X}^0$ ratio is 94% decreasing rapidly at high pulse areas. A negative chirp works less well at $\mathrm{2X}^0$ creation than a positive chirp. However, at higher excitation powers the difference between positive and negative chirp becomes less and less pronounced.
An analysis of the calculated instantaneous eigenenergies, the dressed states Fig. \[fig:fig2\](b), allows a qualitative understanding of the physics. Starting in the ground state $\ket{\mathrm{0}}$, the system evolves along the lower (red) branch for positive chirp. Provided that the pulse area is above the RAP threshold, the system remains in the lower branch at the $\ket{\mathrm{0}}$ and $\ket{\mathrm{2X}^{0}}$ avoided crossing such that the system ends up in the biexciton state $\ket{\mathrm{2X}^{0}}$. Although there is no direct coupling between the $\ket{0}$ and $\ket{\mathrm{2X}^{0}}$ states, the avoided crossing arises as $\ket{0}$ and $\ket{\mathrm{2X}^{0}}$ have a common coupling to the $\ket{\mathrm{X}^{0}}$ state. Phonons can interrupt the adiabatic transfer by causing a jump from one branch to the other one [@Glassl2013]. For positive chirp, the system starts in the lowest energy dressed state and only phonon absorption is possible. At however, phonons at the relevant energy scale are frozen out and the probability for absorption is small. Hence the process with positive chirp is barely affected by phonons at low temperature. Conversely, for negative chirp the system starts out in the upper-most branch (blue curve in Fig. \[fig:fig2\](b) with time running from “right" to “left"); the system can now jump to lower branches by phonon emission, a process which is possible even at low temperature. This leads to a significant probability of exciton and ground state population, reducing the fidelity of the RAP process. All these observations correspond well to the experimental data.
The theory provides a quantitative account of three-level RAP in the presence of phonon coupling. We calculate the $\mathrm{2X}^0$ and $\mathrm{X}^0$ occupations as a function of the pulse area: the $\mathrm{2X}^0$ signal is proportional to the $\mathrm{2X}^0$ occupation; the $\mathrm{X}^0$ signal is proportional to the sum of the $\mathrm{2X}^0$ and $\mathrm{X}^0$ occupations. The simulation results are also shown in Fig. \[fig:fig2\] and reproduce the main features of the experimental data extremely well, notably the exact form of the Rabi oscillations for close-to-zero chirp (a broad minimum at pulse area $\sim 3 \pi$ in the $\mathrm{2X}^0$ signal, a broad maximum at $\sim 1.5\pi$ in the $\mathrm{X}^0$ signal); the “delayed" (“accelerated") rise of the $\mathrm{2X}^0$ signal for positive (negative) chirp; the relative signal strengths; and the better RAP performance for positive chirp. In the theory, the Rabi rotations are not influenced significantly by phonons because the Rabi dynamics are too fast for the phonons to follow. The theoretical results show clearly that $\mathrm{2X}^0$ generation with RAP is strongly (weakly) influenced by phonons for negative (positive) chirp for small to modest pulse areas. At the highest pulse areas, however, the theory predicts essentially perfect RAP independent of the sign of the chirp. The interpretation is that the splittings between the branches at the avoided crossings are energetically so large that they lie well above the energy range of the phonons which are efficiently coupled to the exciton system [@Reiter2014]. In other words, the wavelength of the phonons required for scattering between the dressed states becomes much smaller than the size of the QD at high pulse area. There is evidence that this reduction of the phonon efficiency is also seen in the experiment: the RAP signals for positive and negative chirp approach each other at the largest pulse areas. However, in addition the experimental signals exhibit a decay at high pulse areas which cannot be explained by the phonon coupling model. Instead, we tentatively attribute this decay to an occupation of higher energy levels by multi-photon absorption [@Patton2005].
![**Biexciton generation on QD2.** **(a)** $\mathrm{2X}^{0}$ emission and **(b)** $\mathrm{X}^{0}$ emission as a function of the time-averaged square-root excitation power for different chirp parameters. **(c)** and **(d)** Simulation of the biexciton occupation and the sum of exciton and biexciton occupations as a function of the pulse area. **(e)** Ratio of the experimental data: $\mathrm{2X}^{0}/\mathrm{X}^{0}$. []{data-label="fig:fig3"}](biexciton_paper_fig3_nip.eps){width="1\columnwidth"}
To stress-test biexciton generation with RAP, we probe a second QD with much larger biexciton binding energy. RAP is more difficult in this case: the pulse area required for efficient $\mathrm{2X}^0$ generation increases, potentially entering the regime in which the additional decay process is active. The second QD, embedded in the n-i-p sample Fig. \[fig:fig1\](a), with biexciton resonance at wavelength , has a biexciton binding energy of (the first QD has a binding energy of ). The $\mathrm{2X}^0$ emission intensity of the RAP experiment is shown in Fig. \[fig:fig3\]. Despite the large biexciton binding energy, the ratio of the $\mathrm{2X}^0$ to $\mathrm{X}^0$ signals reaches 90%, Fig. \[fig:fig3\](e). The general behavior is the same as for the first QD: a Rabi rotation-like behavior for minimal chirp and a distinct difference between positive and negative chirp. Again, apart from the decay at high pulse areas, the theory accounts extremely well for the experimental data. Notably, we find that for stronger negative chirp values, the phonons also become more effective resulting in a broader and deeper minimum of biexciton occupation which shifts to higher pulse areas, Fig. \[fig:fig4\]. Further, the RAP onset occurs at smaller pulse areas for negative chirp than for positive chirp, Fig. \[fig:fig4\], a feature of the experimental data in Fig. \[fig:fig3\](a).
![**Calculated biexciton generation on QD2.** Simulation of the biexciton occupation as a function of the chirp parameter and the pulse area. The inset shows the top-view with the same color-scale. []{data-label="fig:fig4"}](biexciton_paper_fig4_3d_and_flat_comp.eps){width="1\columnwidth"}
In conclusion, we have demonstrated the coherent generation of a biexciton in a semiconductor quantum dot using a single chirped laser pulse. The state preparation has a very high fidelity over a broad range of excitation powers. The sign of the chirp is important: the scheme is robust with respect to phonon scattering at low temperature for positive chirp. A negative chirp results in damping due to phonon scattering. Theoretical calculations including a microscopic coupling to phonons reproduce all the experimental features apart from a damping in the experiment at high pulse areas.
We acknowledge financial support from EU FP7 ITN S$^{3}$NANO, NCCR QSIT and SNF project 200020\_156637. AL and ADW acknowledge gratefully support from DFH/UFA CDFA05-06, DFG TRR160 and BMBF Q.com-H 16KIS0109.
[37]{}ifxundefined \[1\][ ifx[\#1]{} ]{}ifnum \[1\][ \#1firstoftwo secondoftwo ]{}ifx \[1\][ \#1firstoftwo secondoftwo ]{}““\#1””@noop \[0\][secondoftwo]{}sanitize@url \[0\][‘\
12‘\$12 ‘&12‘\#12‘12‘\_12‘%12]{}@startlink\[1\]@endlink\[0\]@bib@innerbibempty [****, ()](\doibase
10.1103/PhysRevLett.75.4337) [****, ()](\doibase 10.1038/nnano.2012.262) [****, ()](\doibase
10.1103/RevModPhys.74.145) [****, ()](\doibase 10.1038/nphoton.2013.377) [****, ()](\doibase
10.1103/PhysRevLett.76.3005) [****, ()](\doibase
10.1063/1.371842) [****, ()](\doibase 10.1038/nphoton.2013.128) [****, ()](\doibase 10.1021/nl500968k) [****, ()](\doibase
10.1063/1.1855409) [****, ()](\doibase 10.1103/PhysRevB.65.195315) [****, ()](\doibase 10.1103/PhysRevB.73.033306) [****, ()](\doibase
10.1103/PhysRevB.87.085303) [****, ()](\doibase
10.1103/PhysRevB.88.201305) [****, ()](\doibase
10.1103/PhysRevB.86.235301) [****, ()](\doibase 10.1103/PhysRevB.78.155315) [****, ()](\doibase
10.1103/PhysRevLett.73.1138) [****, ()](\doibase
10.1103/PhysRevB.73.125304) [****, ()](\doibase 10.1103/PhysRevLett.110.135505) [****, ()](\doibase 10.7567/JJAP.52.120202) [****, ()](\doibase 10.1103/PhysRevB.91.161302) [****, ()](\doibase
10.1103/PhysRevB.90.241404) [****, ()](\doibase 10.1103/PhysRevLett.114.137401) [****, ()](\doibase 10.1063/1.4894239) [****, ()](\doibase 10.1103/PhysRevLett.106.166801) [****, ()](\doibase
10.1103/PhysRevLett.106.067401) [****, ()](\doibase 10.1103/PhysRevB.90.035316) [****, ()](\doibase 10.1103/PhysRevB.85.121302) [****, ()](\doibase
10.1103/PhysRevB.84.195311) @noop [****, ()](\doibase 10.1038/nphys2688) [****, ()](\doibase 10.1109/JQE.1987.1073201) [****, ()](\doibase 10.1021/jz300239j) [****, ()](\doibase 10.1103/PhysRevB.80.081310) [****, ()](\doibase 10.1088/0953-8984/26/42/423203) [****, ()](\doibase
10.1007/s00340-005-1984-1) [****, ()](\doibase
10.1021/acs.nanolett.6b01082) [****, ()](\doibase 10.1103/PhysRevLett.95.266401)
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We present optical long-slit spectroscopy of the nucleus of the nearby radio galaxy M84 (NGC 4374 = 3C 272.1) obtained with the Space Telescope Imaging Spectrograph (STIS) aboard the [*Hubble Space Telescope*]{} (HST). Our spectra reveal that the nuclear gas disk seen in WFPC2 imaging by Bower et al. (1997, ApJ, 483, L33) is rotating rapidly. The velocity curve has an S-shape with a peak amplitude of 400 km s$^{-1}$ at $0\farcs1 = 8$ pc from the nucleus. To model the observed gas kinematics, we construct a thin Keplerian disk model that fits the data well if the rotation axis of the gas disk is aligned with the radio jet axis. These models indicate that the gas dynamics are driven by a nuclear compact mass of $1.5 \times 10^9 \ M_{\odot}$ with an uncertainty range of $(0.9 - 2.6) \times 10^9 \ M_{\odot}$ and that the inclination of the disk with respect to the plane of the sky is $75\arcdeg - 85\arcdeg$. Of this nuclear mass, only $\le 2 \times 10^7 \ M_{\odot}$ can possibly be attributed to luminous mass. Thus, we conclude that a dark compact mass (most likely a supermassive black hole) resides in the nucleus of M84.'
author:
- 'G. A. Bower, R. F. Green, A. Danks, T. Gull, S. Heap, J. Hutchings, C. Joseph, M. E. Kaiser, R. Kimble, S. Kraemer, D. Weistrop, B. Woodgate, D. Lindler, R. S. Hill, E. M. Malumuth, S. Baum, V. Sarajedini, T. M. Heckman, A. S. Wilson, and D. O. Richstone'
title: 'Kinematics of the Nuclear Ionized Gas in the Radio Galaxy M84 (NGC 4374)'
---
36truept
To appear in the Astrophysical Journal [*Letters*]{}
Introduction
============
M84 is an E1 galaxy in the Virgo Cluster with an active galactic nucleus and hosts the F-R I (Fanaroff & Riley 1974) radio source 3C 272.1. Bower et al. (1997; hereafter Paper I) obtained images of M84 with the Wide Field/Planetary Camera 2 (WFPC2) aboard HST, showing that the ionized gas within the central kpc has three components: a nuclear gas disk, outer filaments, and an ‘ionization cone’. The nuclear gas disk has diameter $\approx 1''$ (82 pc) and a major axis P.A. $\approx 58\arcdeg$ that is tilted by $\approx 25\arcdeg$ with respect to the major axis P.A. of the outer filamentary emission. This outer filamentary emission had been seen in ground-based imaging (e.g., Hansen et al. 1985; Baum et al. 1988). Its major axis is approximately perpendicular to the axis of the radio jets (Laing & Bridle 1987; Jones et al. 1981).
The presence of a nuclear gas disk in M84 is especially interesting. If the gas exhibits Keplerian motion about the nucleus, then a straightfoward application of Newton’s laws to the dynamics of this gas disk would provide an estimate of the mass of the putative supermassive black hole (BH) in M84’s nucleus. It is plausible that M84 contains a BH, since it is a radio galaxy and the rotation gradient of the ionized gas is spatially unresolved (i.e., $> 100$ km s$^{-1}$ arcsec$^{-1}$) in ground-based observations (Baum et al. 1990, 1992). Previous HST observations using the Faint Object Spectrograph have found gas-dynamical evidence for BHs in other galaxies containing nuclear gas disks, such as M87 and NGC 4261 (Harms et al. 1994; Ferrarese et al. 1996). STIS (through the use of a CCD in a long-slit spectrograph) provides a significant improvement in HST’s efficiency for measuring the nuclear dynamics of galaxies. We chose M84 as a target for a demonstration.
In this paper, we analyze the dynamics of the nuclear gas disk in M84 to probe the nuclear gravitational potential. The analysis of gas dynamics in galactic nuclei is complementary to the method of analyzing the stellar dynamics (e.g., Kormendy et al. 1996). However, measuring the stellar kinematics in M84 using HST would be challenging, since the stellar surface brightness at the nucleus (determined from the WFPC2 F547M image in Paper I) is rather modest ($\mu_V
\approx$ 16 mag arcsec$^{-2}$). For M84, the gas kinematics are far easier to measure than the stellar kinematics given the high surface brightness in the emission lines.
Throughout this paper, we adopt a distance to M84 of 17 Mpc (Mould et al. 1995). At this distance, $1''$ corresponds to 82 pc. The Galactic extinction along the line of sight is $A_B = 0\fm13$ (Burstein & Heiles 1984).
Observations and Data Calibration
=================================
Long-slit spectroscopy of M84’s nuclear region was obtained with the STIS CCD, which has a pixel scale of $0\farcs05$/pixel (Baum et al. 1996), aboard HST on 1997 April 14 and 17 with the telescope tracking in fine lock with one FGS probe (nominal jitter $\approx 0\farcs007$). Since M84’s nucleus contains a bright optical point source (Paper I), the nucleus was acquired easily to an accuracy of $0\farcs05$ by the ACQ mode using two iterations of two 10 sec imaging exposures through the F28X50LP optical long-pass filter. The ACQ/PEAK mode (Baum et al. 1996) was not available during these observations since they were obtained early during Servicing Mission Orbital Verification (SMOV). The $52'' \times 0\farcs2$ slit was aligned at a position angle (P.A.) of $104\arcdeg$. This was the closest that the slit could be aligned with the gas disk’s major axis (P.A. $= 58\arcdeg$; Paper I) because of HST scheduling constraints during SMOV. To allow for the centering accuracy of only $0\farcs05$ and for the offset between the slit P.A. and the gas disk’s major axis, we planned to obtain spectra at four different slit positions offset from the nucleus by $-0\farcs3$, $-0\farcs1$, $+0\farcs1$, and $+0\farcs3$, where the offsets were perpendicular to the slit and negative spatial offsets moved the slit toward a P.A. of $14\arcdeg$ on the sky. However, below we discuss an empirical determination of the slit positions using our STIS data and the WFPC2 images from Paper I, showing that the actual offsets were $-0\farcs2$, $0\farcs0$, $+0\farcs2$, and $0\farcs0$. For the fourth offset position, the discrepancy between the planned and actual positions occurred because this last spectrum was obtained during the second visit of M84, with an erroneously commanded offset. This error was fortuitous because from our analysis below it is apparent that the kinematic signature of the nuclear gas disk is readily detectable only within $\sim 0\farcs3$ of the nucleus, beyond which the kinematics of the outer filamentary emission (which do not necessarily provide good leverage on the nuclear gravitational potential) dominates the spectrum.
At each slit position, we obtained spectra with the G750M grating, which has a dispersion of 0.56 Å/pixel. This grating was set to cover the wavelength range of 6295 Å to 6867 Å, which includes the emission lines of H$\alpha$, \[N II\] $\lambda\lambda
6548, 6583$, and \[S II\] $\lambda\lambda 6717, 6731$. The spectral resolution of our instrumental configuration was 2.2 Å $\approx 100$ km s$^{-1}$ (FWHM), assuming uniform illumination of the slit. However, Paper I shows that at the nucleus, there is a point source in the optical continuum, and the H$\alpha$ + \[N II\] emission is very compact. Thus, our spectral resolution at the nucleus was better than 100 km s$^{-1}$. We integrated for two HST orbits at each slit position, which was equivalent to $4500 - 5100$ sec per slit position depending on the occurrence of instrumental overheads. Spectra of the internal wavelength calibration source (wavecals) were interspersed among the galaxy spectra to allow for correction of thermal drifts during data reduction.
The data were calibrated using the CALSTIS pipeline to perform the steps of bias subtraction, dark subtraction, applying the flatfield, and combining the two sub-exposures to reject cosmic-ray events. The accuracy of the flatfield calibration was 1%. To reject hot pixels from the data, we employed dark frames obtained immediately before and after the M84 observations. We examined the input data, the flagged hot pixels, and the cleaned output data to ensure that only hot pixels were rejected. The data were wavelength calibrated and rectified by tracing the wavecals (using the Ne emission lines for the dispersion axis and the shadows of the two occulting bars for the spatial axis) and then applying these solutions for the geometric distortions to the data. The largest offset that we found between the actual and nominal dispersion solutions was 0.30 pixels, which is significant given that the dispersion solutions were accurate to 0.06 pixels (0.03 Å = 1 km s$^{-1}$). The data were then rebinned onto a log $\lambda$ scale with a reciprocal dispersion of 25 km s$^{-1}$ pixel$^{-1}$.
At this point, we measured the slit positions implied by comparing the flux along the slit with the WFPC2 images from Paper I. Since the on-orbit flux calibration for our grating tilt was not known accurately at the time of our analysis, we normalized the spectra by the observed continuum intensity distribution. The predicted continuum normalized WFPC2 F658N fluxes were then determined by using the STSDAS task “synphot” on the normalized STIS spectra. These predictions were compared with the WFPC2 F658N image normalized by the synthetic 6590 Å continuum image (constructed from the F547M and F814W images; see Paper I for details), indicating that the actual slit positions were offset by $-0\farcs2$, $0\farcs0$, $+0\farcs2$, and $0\farcs0$ (shown in Fig. 1). However, the uncertainty is as high as $\pm 0\farcs1$ since the nuclear gas disk is very compact (Paper I) and the slit width was rather large. These empirical positions are favored because the gas kinematics (measured in §3) are symmetric with respect to the nucleus for these empirical positions. The relative steps between slit positions of $0\farcs2$ are assumed to be much more accurate than the absolute placement.
Measurement of the Gas Kinematics
=================================
Fig. 2 shows the last offset = $0\farcs0$ spectrum centered on the \[N II\] $\lambda 6583$ emission line. The first iteration at measuring the radial velocities involved cross correlating each spectral row from the four long-slit spectra with a synthetic emission-line spectrum, which included only the five emission lines that we detected in M84 (see Fig. 3) with flux ratios set at values typically found in the data. We then compared the velocities measured by the cross correlation technique with those measured manually from the emission-line peaks. These measurements agree very well for distances $> 0\farcs3$ from the nucleus. For rows closer to the nucleus than this, the emission line profiles usually exhibit two kinematic components rather than a single component. These two components are readily seen in the strong \[N II\] $\lambda 6583$ profile (as shown in Fig. 4), especially for the two spectra with offset = $0\farcs0$. Since the velocity measured by our cross correlation technique (using our synthetic template spectrum) coincides with the flux-weighted centroid over a given emission-line profile, these measurements will be distorted when more than one kinematic component is present. We determined which of the \[N II\] $\lambda 6583$ profiles in Fig. 4 have two kinematic components by fitting model profiles to a few examples. Separate models with one or two Gaussians were fit to the profiles. If the improvement in $\chi^2$ was significant, then the profile was classified as having two kinematic components. Although these models were not good fits to the observed profiles (which have broader wings than a Gaussian), this procedure was sufficient for objectively determining which profiles have two components. The velocities for the separate components were then measured by finding the centroid of each component peak.
Based on the H$\alpha$ + \[N II\] image (see Fig. 1), it is not surprising that two kinematic components are seen within $\sim 0\farcs3$ of the nucleus in the STIS spectroscopy. Paper I identified three spatial components in the ionized gas, including the nuclear gas disk, the outer filaments, and an ionization cone. Given these spatial components, the line of sight to the nuclear gas disk should also intersect the outer filamentary gas (and perhaps the ionization cone) lying in the foreground. Since the outer filamentary gas rotates about the nucleus at $\approx \pm 100$ km s$^{-1}$ (Baum et al. 1990, 1992), one expects to see this low-velocity component superposed onto the high-velocity kinematics of the nuclear gas disk. Fig. 5 shows our velocity measurements along the slit for each of the four slits. Our measurements of the low-velocity component agree very well with those of Baum et al. (1990). This measurement of the high-velocity component is the first time that the kinematics of M84’s nuclear gas disk have been resolved.
Interpretation
==============
The high amplitude and S-shape of the velocity curve in the center of M84 strongly suggest that the emitting gas is in organized motion around a massive central object. To estimate the central mass, numerical models were constructed of a thin Keplerian disk. The continuous velocity sampling along the slit and the two offset slit positions give important constraints on the model parameters.
The models were calculated as follows. By construction, $V$ (the velocity of gas at any point on the disk) is proportional to $r^{-1/2}$, where $r$ is the distance from the center in the disk plane. The observed velocities were matched by adopting a systemic velocity for the emitting gas. The relative weighting as a function of radius was assumed to follow the major axis isophotal intensity map $I(r)$ derived in Paper I (i.e., $I(r) \propto r^{-1}$). To avoid divergence, the velocity was truncated interior to a selected physical radius at its value at that radius; the intensity weighting was truncated interior to a chosen projected radius at its corresponding value. The disk is inclined to the plane of the sky by an angle $i$, and the slit is rotated with respect to the major axis of the disk by an angle $\theta$. The slit can also be offset from the nucleus by a specified distance in a direction perpendicular to the slit rotation angle. The contribution to the velocity peak at any sample pixel is weighted by both the projected disk intensity distribution and by the PSF of HST + STIS. The PSF was approximated as a Gaussian with dispersion $\sigma$, centered at the midpoint of each sample pixel and integrated to a distance of $4\sigma$. Since the $0\farcs2$ slit is substantially wider than the FWHM of the delivered PSF, the measured velocity centroid of each pixel sample should be dominated by the effects of the PSF.
The most important free parameter is the mass of the central object. The range of the other parameters is interesting primarily for the extent to which they bound the acceptable range of masses, with one exception to be discussed below. After many realizations of the model, several of the parameters were fixed, either because they were tightly or externally constrained, or because the results were insensitive to reasonable choices. The physical inner radius and the intensity weighting inner radius are examples of the latter case. The PSF is known from both modeling and measurement of standard stars. A Gaussian fit to a model encircled energy curve for HST + STIS + guiding jitter suggested $\sigma = 0\farcs06$. A fit to a short exposure of a standard star observed with the G750M grating produced $\sigma = 0\farcs04$. Those values bracketed the actual width of the velocity inflection of the S-shaped curve. A $\sigma = 0\farcs05$ was adopted, corresponding to a FWHM of $0\farcs12$. The fact that the PSF for a longer exposure guided with just one FGS probe appears to be somewhat wider than that for a brief standard star exposure is not unreasonable. The systemic velocity required to symmetrize the peak velocities of the model with respect to the data is 1125 km s$^{-1}$. A variation of 25 km s$^{-1}$ in either direction leads to a noticeable degradation in the quality of the fit.
The first approach to determining the central mass was to assume that the isophotal fit of the emission-line intensity in Paper I gave directly the ellipticity and P.A. of the gas disk within $0\farcs5$ of the nucleus. Those values are an ellipticity of 0.17, corresponding to $i = 34\arcdeg$, and a P.A. of $58\arcdeg$, leading to a relative slit angle of $46\arcdeg$. No acceptable fit of the zero-offset positions could be produced at $i = 34\arcdeg$. The modeled velocities of the “Keplerian” portion of the curve, which are particularly well represented in the data in the negative direction as plotted in Fig. 5, returned to zero much more slowly than the data. An acceptable fit to the zero-offset positions could be obtained by assuming that the ellipticity of the isophotes did not represent the inclination angle of the gas disk. A much higher inclination near $75\arcdeg$ was required in that case, and no corresponding fit to the offset positions at $+0\farcs2$ and $-0\farcs2$ could be produced. The velocity amplitudes at the point of slit closest approach to the nucleus were very much higher than observed and the shapes were not at all congruent.
This result led us to question whether the abrupt isophote twist in the inner arcsecond actually represented the major axis of the gas disk. An alternative approach is to assume that the major axis of the larger scale emission-line structure defines the P.A. of the kinematic major axis of the inner gas disk. Baum et al. (1988) measured that angle on the larger scale to be $83\arcdeg$. That direction is nearly perpendicular to the measured P.A.’s of the radio jets of 3C 272.1, which are $-5\arcdeg$ and $+170\arcdeg$ (Laing & Bridle 1987). That choice fixes the relative slit angle at $21\arcdeg$. The free parameters are then the central mass and inclination angle. The best fit model is plotted over the data in Fig. 5, while its parameters (with uncertainty ranges) are given in Table 1. Fig. 5 shows reasonable agreement with the velocity amplitude and Keplerian shape for the zero-offset curves, and qualitative correspondence with the off-nuclear curves. Our Keplerian disk model fits the data very well, although there may be minor non-disk contributors to the emission-line profile (as suggested by the complexities of the inner emission-line isophotes \[Paper I\] and spectral profiles \[Fig. 4\]).
The acceptable range of the model parameters should be determined quantitatively. However, a proper derivation of $\chi^2$ per degree of freedom is complicated by two factors: the data from adjacent pixels are strongly correlated through the PSF, and the uncertainties in velocity points are dominated by systematic effects in disentangling the multiple broad components. In addition, the derived central mass and disk inclination angle are themselves correlated. As the disk becomes more edge-on, the observed radial velocities more nearly represent the total velocity, implying a smaller mass for a given amplitude. That effect is overwhelmed by a competing effect in a thin disk model, which is the foreshortening in the transverse direction. A given pixel samples to much larger radii as the disk becomes edge-on, requiring larger mass to produce a given velocity amplitude. Increasing inclination angle therefore requires increasing mass. To set bounds on the acceptable range of mass, we sampled the two corners of the error ellipse. These represent approximately factors of 2 increase in $\chi^2$, but we did not attempt to optimize a data weighting scheme that fairly represented the systematic uncertainties. We did move to inclination angles that produced just discernibly poorer fits to the velocity curves, then varied the central mass to depart from the velocity amplitude by about 1 $\sigma$ at each extremum. A more realistic disk model with finite thickness would project more low radial velocity component into a given pixel, suggesting that the mass estimated from a thin disk is systematically too low.
How much of this nuclear mass can be attributed to luminous mass at the nucleus? The WFPC2 continuum images in Paper I show a bright nuclear point source with V = $19\fm9$ (i.e., $L_V = 4.3 \times 10^6 \ L_{\odot}$ if $A_V$ to the nucleus is $0\fm54$; Paper I) surrounded by the outer stellar distribution. Is the emission from this continuum point source stellar or non-stellar? Additional spectroscopic observations of M84 are needed to address this issue. If we assume that the emission arises entirely from a stellar population with $M/L_V \approx 5$, the upper limit on the luminous mass at the nucleus is $M_* \le 2 \times 10^7 \ M_{\odot}$. This is much less than the nuclear mass $\approx 1.5 \times 10^9 \ M_{\odot}$ required to explain the gas disk dynamics (even given the uncertainty in the extinction). Thus, M84 contains a dark compact mass (most likely a BH given the presence of powerful radio jets).
Although it was surprising that the emission-line isophotes did not describe the geometry of the inner gas disk, the resulting picture is consistent with physical expectations. The angular momentum axis of the disk is now aligned with the axes of the dual radio jets. A gas disk nearly perpendicular to the plane of the sky would suggest that these bipolar radio jets are nearly in the plane of the sky, consistent with the roughly comparable power in the jets and lobes on either side of the nucleus. The more extended emission-line structure can now represent a natural source of gas for an inner accretion region, without the need for a strong, abrupt warp at $\sim 40$ pc from the center.
The BH mass estimate of $1.5 \times 10^9 \ M_{\odot}$ can be compared with the expectation from Kormendy & Richstone’s (1995) correlation between black hole mass and bulge mass. An update to that relationship by Kormendy et al. (1997) including 12 BH mass determinations predicts that the BH mass will be $0.0022^{+0.0014}_{-0.0009} \ \times$ the bulge mass. For the total mass of M84, we use the structural parameter estimate of Bender, Burstein, & Faber (1992) of $M = 5G^{-1}\sigma_0^2r_e$. We adopt their value of effective radius (scaled to a distance of 17 Mpc) of 4.5 kpc, and a central velocity dispersion of $\sim 310$ km s$^{-1}$ from Davies & Birkinshaw (1988). The resulting total galaxy mass is $5.1 \times 10^{11} \ M_{\odot}$, with a predicted BH mass of $1.1 \times 10^9 \ M_{\odot}$ within a range of $(0.7-1.8) \times 10^9
\ M_{\odot}$. Our best fit value of $1.5 \times 10^9 \ M_{\odot}$ is very close to the ridge line of the Kormendy et al. correlation. At an $M_B = -21.0$, M84 is similar to M87 in total absolute magnitude (therefore, total implied mass) and central BH mass.
Our measurement of the BH mass in M84 (when combined with those for M87 and NGC 4261; see §1) has interesting implications for the physics of radio galaxies. A comparison of these three measured BH masses with the total radio luminosities of their associated radio sources (e.g., Roberts et al. 1991) suggests tentatively that these two quantities might not be directly correlated.
Several interpretive loose ends make M84 a galaxy meriting continuing study. The shapes of the inner isophotes are not now explained straightforwardly as a manifestation of an inner accretion disk. What produces the change in apparent P.A. and ellipticity? Are there unresolved near-nuclear H II regions that add to the complexity of the structure? Is there any relation between the sharp change in direction of the central dust lanes at $\sim 100$ pc from the nucleus, as noted by Jaffe et al. (1994) and in Paper I, and the dynamical environment produced by the central BH? Although the best fitting systemic velocity for the gas disk is consistent with the systemic velocity of the outer filamentary gas derived from the two offset = $0\farcs0$ spectra, it is $\sim 60$ km s$^{-1}$ higher than the measured velocity for the stars (e.g., Davies & Birkinshaw 1988). Stiavelli & Setti (1993) noted that galaxies with a high contrast central potential could produce gravitational redshifts between central and outer samplings of the stellar velocity field of several tens of km s$^{-1}$ in extreme cases. They found a $1.4\sigma$ effect in the Davies & Birkinshaw measurements of the stellar field. If the dust lane obscuration of the central stellar light is significant, it is possible that the emission from the gas within 50 pc of the nucleus is coming from much deeper in the central potential well, accounting for some of the discrepancy. Resolution of this velocity difference is necessary to add confidence to the interpretation of the gas motion as that of a Keplerian disk around a large unseen mass at the dynamical center of M84.
We acknowledge useful comments from Eric Emsellem and the referee Ralf Bender and the assistance of P. Hall and C. Liu in planning these observations. Support for this work was provided to the STIS Investigation Definition Team by NASA.
[**References**]{}
Baum, S., et al. 1996, STIS Instrument Handbook, Version 1.0 (Baltimore: STScI)
Baum, S. A., et al. 1988, , 68, 643
Baum, S. A., Heckman, T., & van Breugel, W. 1990, , 74, 389
Baum, S. A., Heckman, T., & van Breugel, W. 1992, , 389, 208
Bender, R., Burstein, D., & Faber, S. M. 1992, , 399, 462
Bower, G. A., Heckman, T. M., Wilson, A. S., & Richstone, D. O. 1997, , 483, L33
(Paper I)
Burstein, D., & Heiles, C. 1984, , 54, 33
Davies, R. L., & Birkinshaw, M. 1988, , 68, 409
Fanaroff, B. L., & Riley, J. M. 1974, , 167, 31P
Ferrarese, L., Ford, H. C., & Jaffe, W. 1996, , 470, 444
Hansen, L., Nørgaard-Nielsen, H. U., & Jørgensen, H. E. 1985, , 149, 442
Harms, R. J., et al. 1994, , 435, L35
Jaffe, W., et al. 1994, , 108, 1567
Jones, D. L., Sramek, R. A., & Terzian, Y. 1981, ApJ, 246, 28
Kormendy, J., Bender, R., Evans, A. S., & Richstone, D. 1997, , submitted
Kormendy, J., & Richstone, D. 1995, , 33, 581
Kormendy, J., et al. 1996, ApJ, 459, L57
Laing, R. A., & Bridle, A. H. 1987, , 228, 557
Mould, J., et al. 1995, ApJ, 449, 413
Roberts, M. S., Hogg, D. E., Bregman, J. N., Forman, W. R., & Jones, C. 1991, , 75,
751
Stiavelli, M., & Setti, G. 1993, , 262, L51
[lcc]{} Black Hole Mass ($M_{\odot}$) & $1.5 \times 10^9$ & $(0.9-2.6) \times 10^9$ Disk Inclination ($\arcdeg$) & 80 & $75-85$ Disk P.A. ($\arcdeg$) & 83 & $80-85$ Gas systemic velocity (km s$^{-1}$) & 1125 & $1100-1150$ Intensity law & $I(r) \propto r^{-1}$ $I(r)$ inner radius (pc) & 1 & $0.3-3$ $V(r)$ inner radius (pc) & 0.03 & $0.01-0.1$ PSF $\sigma \ ('')$ & $0.05$ & $0.04 - 0.06$
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Azimuthal distributions of $\pi^+$, $K^+$ and $K^-$ mesons have been measured in Au+Au reactions at 1.5 and Ni+Ni reactions at 1.93 . In semi-central collisions at midrapidity, $\pi^+$ and $K^+$ mesons are emitted preferentially perpendicular to the reaction plane in both collision systems. In contrast for $K^-$ mesons in Ni+Ni reactions an in-plane elliptic flow was observed for the first time at these incident energies.'
author:
- |
F. Uhlig$^b$, A. Förster$^{b,+}$, I. Böttcher$^d$, M. Dȩbowski$^{e,f}$, F. Dohrmann$^f$, E. Grosse$^{f,g}$, P. Koczoń$^a$, B. Kohlmeyer$^d$, F. Laue$^{a,*}$, M. Menzel$^d$, L. Naumann$^f$, H. Oeschler$^b$, W. Scheinast$^f$, E. Schwab$^a$, P. Senger$^a$, Y. Shin$^c$, H. Ströbele$^c$, C. Sturm$^{b,a}$, G. Surówka$^{a,e}$, A. Wagner$^f$, W. Waluś$^e$\
(KaoS Collaboration)\
$^a$ Gesellschaft für Schwerionenforschung, D-64220 Darmstadt, Germany\
$^b$ Technische Universität Darmstadt, D-64289 Darmstadt, Germany\
$^c$ Johann Wolfgang Goethe-Universität, D-60325 Frankfurt am Main, Germany\
$^d$ Phillips Universität, D-35037 Marburg, Germany\
$^e$ Jagiellonian University, PL-30059 Kraków, Poland\
$^f$ Forschungszentrum Rossendorf, D-01314 Dresden, Germany\
$^g$ Technische Universität Dresden, D-01062 Dresden, Germany\
$^*$ Present address: Brookhaven National Laboratory, USA\
$^+$ Present address: CERN, CH-1211 Geneva 23, Switzerland
title: 'Observation of different azimuthal emission patterns for $K^+$ and of $K^-$ mesons in Heavy Ion Collisions at 1-2 '
---
version of
Relativistic heavy ion collisions provide a unique opportunity to study both the behavior of nuclear matter at high densities as well as the properties of hadrons in dense nuclear matter. In particular, strange mesons are considered to be sensitive to in-medium modifications. Theory predicts a repulsive $K^+N$ potential and an attractive $K^-N$ potential in dense matter [@schaf]. It is suggested that the latter effect leads to a condensate of $K^-$ mesons in the interior of the neutron stars, causing dramatic consequences for the neutron star stability [@brobet].
Microscopic transport calculations simulating heavy-ion collisions predict measurable consequences of the in-medium modifications of strange mesons. The $KN$ potentials reduce the $K^+$ yield and enhance the $K^-$ yield, resulting in an increase of the $K^-/K^+$ ratio.
First experimental evidence for in-medium modifications of $K^-$ mesons in dense nuclear matter was the observation that the $K^-/K^+$ ratio was enhanced in Ni+Ni collisions as compared to nucleon-nucleon collisions [@Barth; @Menzel]. A large $K^{-}/K^{+}$ ratio was also found in C+C [@laue] and in Au+Au collisions [@AF]. In heavy-ion collisions, however, strangeness-exchange reactions like $\pi\Lambda\rightarrow K^- N$ contribute significantly to the production of $K^-$ mesons. This process, although taken into account by transport calculations, reduces the sensitivity of the $K^-$ meson yield to the in-medium $KN$ potential [@ko84; @Hart02; @HO; @AF].
Another observable affected by in-medium effects is the azimuthal emission pattern of $K^+$ and $K^-$ mesons in heavy-ion collisions. $K^+$ and $K^-$ mesons experience different potentials in nuclear matter: While the scalar potential acts attractively on both kaon species, the vector potential repels $K^+$ mesons and attracts $K^-$ mesons. For the $K^+$ mesons these two contributions almost cancel each other leading to a small repulsive $K^+N$ interaction. The superposition of the two attractive interactions results in a strongly attractive potential for $K^-$ mesons [@schaf].
A repulsive $K^+N$ potential would repel the $K^+$ mesons from the bulk of the nucleons and therefore cause an preferred out-of-plane emission of the $K^+$ mesons at midrapidity and a directed flow opposite to the nucleons at target and projectile rapidity. These effects were found in experiments [@shin; @crochet] and interpreted as evidence for a repulsive $K^+N$ potential [@Li; @Fuchs2].
The propagation of $K^-$ mesons in nuclear matter is governed by a large $K^-p$ cross section of up to 100 mb which is dominated by inelastic scattering via the strangeness-exchange reaction $\pi
Y\leftrightarrow K^- N$ with $Y=\Lambda, \Sigma$. Therefore, one would expect a pronounced azimuthal anisotropy of the $K^-$ meson emission in heavy-ion collisions due to the interaction of $K^-$ mesons with spectator matter. However, when taking into account the strongly attractive in-medium $K^-N$ potential, transport calculations predict an almost isotropic azimuthal emission pattern at midrapidity [@Fuchs2].
In this Letter we present experimental data on the azimuthal distributions of both $K^+$ and $K^-$ mesons in nucleus-nucleus collisions. We have measured two systems: Ni+Ni at 1.93 (both for $K^+$ and $K^-$ mesons) and Au+Au at 1.5 (only $K^+$ mesons). For comparison also results from an analysis of $\pi^+$ mesons are given. An azimuthal distribution of $K^-$ mesons emitted in heavy-ion collisions at subthreshold beam energies is shown for the first time.
The experiments were performed with the Kaon Spectrometer (KaoS) at the heavy-ion synchrotron (SIS) at GSI in Darmstadt [@senger], using an Au beam of 1.5 $A\cdot$GeV impinging on an Au target (0.96 g/cm$^2$) and a Ni beam of 1.93 $A\cdot$GeV on a Ni target (0.68 g/cm$^2$). The particles were identified using the momentum and time-of-flight information of the magnetic spectrometer, and two hodoscopes were used for event characterization [@Brill]. The Large Angle Hodoscope is used to derive the centrality of the collision from the multiplicity of charged particles measured in the polar angle range 12$^{\circ} <
\theta_{lab} < 48^{\circ}$. The orientation of the event plane was reconstructed from the azimuthal emission angle of the charged projectile spectators using the transverse momentum method [@daniel]. These particles were identified (up to $Z$ = 8) by their energy loss and their time of flight as measured with the Small Angle Hodoscope located about 7 m downstream from the target covering polar angles between 0.5$^\circ$ and 11$^\circ$. The resolution in the determination of reaction plane [@Brill] is $\langle\Delta\Phi^2\rangle^{1/2}=37^\circ$ for the Au-system and $\langle\Delta\Phi^2\rangle^{1/2}=61^\circ$ for the Ni-system.
Figure \[Au\_midrap\] shows the azimuthal distributions of $K^+$ and $\pi^+$ mesons for semi-central Au+Au collisions at 1.5 . The distributions are corrected for the angular resolution of the reaction plane determination [@Brill]. The data are fitted using the first two components of a Fourier series $$\frac{dN}{d\Phi}\sim 2\, v_1 \cos(\phi) \, + \, 2\, v_2 \cos (2\phi)
\label{v1_v2}$$ resulting in values for $v_1$ and $v_2$, as given in the figures together with the statistical errors. The determination of the coefficient $v_1$ is subject to an additional systematic error of 0.04.
Both $\pi^+$ and $K^+$ mesons exhibit a pronounced enhancement at $\phi = \pm 90^o$, i.e. perpendicular to the reaction plane. For $\pi^+$ mesons this effect can be interpreted as rescattering and absorption at the spectator fragments. The data are in agreement with previous observations [@shin; @Brill].
The study of Ni+Ni collisions was performed at a higher incident energy of 1.93 . The resulting higher production cross section for $K^-$ mesons provides an opportunity to study both charged kaon species. The data are shown in Fig. \[Ni\_midrap\] along with $\pi^+$ mesons for semi-central Ni+Ni collisions. Both $\pi^+$ and $K^+$ mesons follow the same trend already observed in Au+Au collisions. The values for $v_2$ are smaller than in Au+Au as one might expect for the smaller system. In contrast to the $\pi^+$ and $K^+$ mesons, the $K^-$ mesons show an in-plane enhancement.
This “positive” (in-plane) elliptic flow of particles is observed for the first time in heavy-ion collisions at SIS energies. In contrast to this observation one would expect a preferential out-of-plane emission (negative elliptic flow) of $K^-$ mesons due to their large absorption cross section in spectator matter.
A depletion of the expected out-of-plane emission pattern of $K^-$ mesons might be due to the fact that they are produced via strangeness-exchange reactions. This causes a delay in the freeze-out of the $K^-$ mesons [@AF; @Hart02; @HO], and, hence, a reduced shadowing effect by the spectator fragments which have moved further away. The observed in-plane emission of $K^-$ mesons, however, cannot be easily explained with this scenario.
In order to quantitatively explain the $K^+$ and $K^-$ meson azimuthal distributions we compare in Fig. \[iqmd\_calculation\] our data to recent results of the Isospin Quantum Molecular Dynamics (IQMD) model [@hart01]. This transport calculation takes into account both the space-time evolution of the reaction system and the in-medium properties of the strange mesons. The dashed and solid lines represent results of calculations without and with in-medium potentials, respectively. In the case of the $K^+$ mesons (upper panel of Fig. \[iqmd\_calculation\]) the effect of the repulsive $K^+N$ potential is small in this model. A large fraction of the observed out-of-plane enhancement, in contrast to other models [@Fuchs2; @Li; @Mishra], is caused by the scattering of $K^+$ mesons. Another transport code (HSD) [@Mishra] predicts a dominant influence of the potential on the emission pattern of the $K^+$ mesons. In the system Au+Au at 1 where both size and life time of the fireball are larger than in the Ni+Ni case, the effect of the repulsive $K^+N$ potential was studied using the RBUU code and was found to be very important [@shin; @Li; @Fuchs2].
In the lower part of Fig. \[iqmd\_calculation\] we compare a calculation without (dashed) and with (solid) $K^-N$ potential. When neglecting the $K^-N$ potential, the calculation predicts a weak in-plane elliptic flow caused by shadowing. This effect is rather small because of the late emission of $K^-$ mesons. When taking into account the attractive in-medium $K^-N$ potential the model is able to describe the experimental in-plane elliptic flow pattern much better. Model calculations with the HSD code [@Mishra] predict a flat azimuthal distribution both with and without a $K^-N$ potential and, hence cannot explain the observed in-plane flow of $K^-$ mesons.
In summary, we have measured the azimuthal emission patterns of $\pi$ and K mesons in heavy-ion collisions at threshold beam energies. We found a pronounced out-of-plane emission (negative elliptic flow) for the $K^+$ mesons confirming previous results. We presented new data on the azimuthal angle distribution of $K^-$ mesons which exhibit a positive (in-plane) elliptic flow pattern, which is in contrast to all other measured particles exhibiting a preferred out-of-plane emission.
This observation can be explained by a transport model (IQMD) by a late emission of $K^-$ mesons and assuming an attractive in-medium $K^-N$ potential. Hence, the distribution of strange mesons in space and their multiplicity which has been used so far, are independent probes to extract information on in-medium properties at high densities.
[50]{}
J. Schaffner et al., Nucl. Phys. A [**625**]{} (1997) 325. G.E. Brown and H.A. Bethe, Astrophys. Jour. [**423**]{} (1994) 659 and Nucl. Phys. [**A 567**]{} (1994) 937 R. Barth et al., (KaoS Collaboration), Phys. Rev. Lett. 78 (1997) 4007. M. Menzel et al., (KaoS Collaboration), Phys. Lett. [**B495**]{} (2000) 26. F. Laue, C. Sturm et al., (KaoS Collaboration), Phys. Rev. Lett. [**82**]{} (1999) 1640. A. Förster, F. Uhlig et al., (KaoS Collaboration), Phys. Rev. Lett. [**91**]{} (2003) 152301. C. M. Ko, Phys. Lett. B [**138**]{} (1984) 361. C. Hartnack, H. Oeschler and J. Aichelin, Phys. Rev. Lett. [**90**]{} (2003) 102302. H. Oeschler, J. Phys. G: Nucl. Part. Phys. [**27**]{} (2001) 257. Y. Shin et al., (KaoS Collaboration), Phys. Rev. Lett. [**81**]{} (1998) 1576. P. Crochet et al., (FOPI Collaboration), Phys. Lett. [**B486**]{} (2000) 6. Z.S. Wang et al., Eur. Phys. J. A [**5**]{} (1999) 275. G. Q. Li et al., Phys. Lett. B. [**381**]{} (1996) 17. P. Senger et al., (KaoS Collaboration), Nucl. Instr. Meth. A [**327**]{} (1993) 393. D. Brill et al., (KaoS Collaboration), Phys. Rev. Lett. [**71**]{} (1993) 336; D. Brill et al., (KaoS Collaboration), Z. Phys. A [**355**]{} (1996) 61; D. Brill et al., (KaoS Collaboration), Z. Phys. A [**357**]{} (1997) 207. P. Danielewicz and G. Odyniec, Phys. Lett. [**157B** ]{} (1984) 147 C. Hartnack et al., Eur. Phys. J. A1 (1998) 151 and to be published. A. Mishra et al., nucl-th/0402062, Phys. Rev. C (in print).
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- 'N.A. Abdulov$^{1}$, A.V. Lipatov$^{2,\,3}$'
title: 'Bottomonia production and polarization in the NRQCD with $k_T$-factorization. I: $\Upsilon(3S)$ and $\chi_b(3P)$ mesons'
---
[*$^1$Faculty of Physics, Lomonosov Moscow State University, 119991 Moscow, Russia*]{}\
[*$^2$Skobeltsyn Institute of Nuclear Physics, Lomonosov Moscow State University, 119991 Moscow, Russia*]{}\
[*$^3$Joint Institute for Nuclear Research, 141980 Dubna, Moscow Region, Russia*]{}
[**Abstract** ]{}
The $\Upsilon(3S)$ production and polarization at high energies is studied in the framework of $k_T$-factorization approach. Our consideration is based on the non-relativistic QCD formalism for bound states formation and off-shell production amplitudes for hard partonic subprocesses. The transverse momentum dependent (TMD, or unintegrated) gluon densities in a proton were derived from the Ciafaloni-Catani-Fiorani-Marchesini (CCFM) evolution equation as well as from the Kimber-Martin-Ryskin (KMR) prescription. Treating the non-perturbative color octet transitions in terms of the multipole radiation theory and taking into account feed-down contributions from radiative $\chi_b(3P)$ decays, we extract the corresponding non-perturbative matrix elements for $\Upsilon(3S)$ and $\chi_b(3P)$ mesons from a combined fit to $\Upsilon(3S)$ transverse momenta distributions measured by the CMS and ATLAS Collaborations at the LHC energies $\sqrt s = 7$ and $13$ TeV and central rapidities. Then we apply the extracted values to describe the CDF and LHCb data on $\Upsilon(3S)$ production and to investigate the polarization parameters $\lambda_\theta$, $\lambda_\phi$ and $\lambda_{\theta\phi}$, which determine the $\Upsilon(3S)$ spin density matrix. Our predictions have a good agreement with the currently available data within the theoretical and experimental uncertainties.
Introduction
============
Since it was first observed, the production of charmonia and bottomonia in hadronic collisions remains a subject of considerable theoretical and experimental studies. A theoretical framework for the description of heavy quarkonia production and decays is provided by the non-relativistic QCD (NRQCD) factorization[@1; @2]. This formalism implies a separation of perturbatively calculated short-distance cross-sections for the production of $Q\bar Q$ pair in an intermediate Fock state ${}^{2S+1}L_J^{(a)}$ with spin $S$, orbital angular momentum $L$, total angular momentum $J$ and color representation $a$ from long-distance non-perturbative matrix elements (NMEs), which describe the transition of that intermediate $Q\bar Q$ state into a physical quarkonium via soft gluon radiation. The NMEs are assumed to be universal (process- and energy-independent), not dependent on the quarkonium momentum and obeying certain hierarchy in powers of the relative heavy quark velocity $v_Q \sim \log^{-1} m_Q/\Lambda_{\rm QCD}$ with $m_Q$ being the heavy quark mass. The color octet (CO) NMEs are not calculable within the theory and have to be only extracted from the data.
At present, the cross sections of prompt $S$- and $P$-wave charmonia production ($\psi^\prime$, $\chi_c$, $J/\psi$ and $\eta_c$) in $pp$ collisions are known at the next-to-leading order (NLO NRQCD)[@3; @4; @5; @6; @7; @8; @9; @10; @11; @12; @13; @14; @15]. The dominant tree-level next-to-next-to-leading order (NNLO$^*$) corrections to the color-singlet (CS) production mechanism have been calculated[@16]. With properly adjusted values of NMEs, one can achieve a good agreement between the NLO NRQCD predictions and the experimental data on the $\psi^\prime$, $\chi_c$ and $J/\psi$ transverse momenta distributions[@3; @4; @5; @6; @7; @8; @9; @10]. However, the extracted NMEs strongly depend on the minimal charmonia transverse momentum $p_T$ used in the fits and are almost incompatible with each other when obtained from fitting different data sets. Moreover, none of the fits can reasonably describe the $\psi^\prime$ and $J/\psi$ polarization data (the so-called “polarization puzzle”). The fits involving low-$p_T$ data result in the conclusion that the $\psi^\prime$ and $J/\psi$ production at large transverse momenta is dominated by color-octet ${}^3S_1^{(8)}$ contributions with strong transverse polarization, that contradicts to the unpolarized production seen at the Tevatron and LHC. To obtain an unpolarized $\psi^\prime$ and $J/\psi$ mesons, it is necessary to assume that the production is dominated by the scalar ${}^1S_0^{(8)}$ intermediate state[@4]. However, such assumption immediately contradicts recent $\eta_c$ production data since the respective $\eta_c$ and $J/\psi$ NMEs are related by the heavy quark spin symmetry (HQSS) principle[@1; @2]. The HQSS requires that the $\eta_c$ and $J/\psi$ NMEs have to be determined from the simultaneous fit for the entire charmonia family, that turned out to be impossible in the NLO NRQCD[^1] (see also discussions[@17; @18; @19]). The overall complicated situation is still far from understanding and has been even called “challenging”[@13].
A possible solution to the problem above has been proposed recently[@20] in the framework of a model that interprets the soft final state gluon radiation (which transforms an unbound $Q\bar Q$ pair into a physical quarkonium state) as a series of color-electric dipole transitions. In this way the NMEs are represented in an explicit form inspired by the classical multipole radiation theory, that has dramatic consequences for the polarization of the final state mesons since the spin structure of the transition amplitudes is specified. The proposed approach leads to unpolarized or only weakly polarized charmonia either because of the cancellation between the ${}^3P_1^{(8)}$ and ${}^3P_2^{(8)}$ contributions or as a result of two successive color-electric $E1$ dipole transitions in the chain ${}^3S_1^{(8)} \rightarrow {}^3P_J^{(8)} \rightarrow {}^3S_1^{(1)}$ giving us the possibility to simultaneously solve the polarization puzzle for $J/\psi$ mesons and production puzzle for $\eta_c$ mesons[@21; @22].
An alternative laboratory for understanding the physics of the hadronization of heavy quark pairs is provided by the $\Upsilon(nS)$ and $\chi_b(mP)$ production, which has been measured recently by the CMS[@23; @24], ATLAS[@25] and LHCb[@26; @27] Collaborations at the LHC. Polarization of $\Upsilon(nS)$ mesons has been also investigated by the CMS[@28] and LHCb[@29] Collaborations. Due to heavier masses of bottomonia and smaller relative velocity $v_b$ of $b$ quarks in the bottomonium rest frame ($v_b \simeq 0.08$ against $v_c \simeq 0.23$), these processes could be even a more suitable case to apply the NRQCD factorization because of a more faster convergence of the double NRQCD expansion in strong coupling $\alpha_s$ and $v_Q$. The complete NLO NRQCD predictions for $\Upsilon(nS)$ and $\chi_b(mP)$ production in $pp$ and $p\bar p$ collisions were presented[@30; @31; @32; @33; @34]. As it was shown, one can reasonably explain the LHC data, both on $\Upsilon(nS)$ or $\chi_b(mP)$ yield and $\Upsilon(nS)$ polarization, by taking into account latest measurements on the $\chi_b(mP)$ production. In particular, the polarization puzzle for $\Upsilon(3S)$ meson can be solved by considering the $\chi_b(3P)$ feed-down contributions[@33; @34]. The latter have been observed recently by the LHCb Collaboration for the first time[@35] and found to be rather significant (up to 40%).
However, it is important to investigate the $S$- and $P$-wave bottomonia production and polarization within the same framework which has been already successfully applied for charmonia[@21; @22]. Here we start with a short series of papers dedicated to the $\Upsilon(nS)$ and $\chi_b(mP)$ production in $pp$ and $p\bar p$ collisions at the high energies, that continues line of our previous studies. In the present note we concentrate on the $\Upsilon(3S)$ production with a consistent treatment for large $\chi_b(3P)$ feed-down contribution. The $\Upsilon(1S), \Upsilon(2S), \chi_b(1P)$ and $\chi_b(2P)$ production requires a dedicated study which will be the subject of our forthcoming papers. To describe the perturbative production of the $b\bar b$ pair in the hard scattering subprocess we apply the $k_T$-factorization approach[@36; @37]. This approach is based on the Balitsky-Fadin-Kuraev-Lipatov (BFKL)[@38] or Ciafaloni-Catani-Fiorani-Marchesini (CCFM)[@39] evolution equations, which resum large logarithmic terms proportional to $\ln s \sim \ln 1/x$, important at high energies (or, equivalently, at low longitudinal momentum fraction $x$ of proton carried by gluon). The $k_T$-factorization approach has certain technical advantages in the ease of including higher-order radiative corrections (namely, leading part of NLO + NNLO + $\dots$ terms corresponding to real gluon emissions in initial state) in the form of transverse momentum dependent (TMD, or unintegrated) gluon density function in a proton[^2]. To describe the non-perturbative transition of an unbound $b\bar b$ pair into physical bottomonia we employ the model[@20]. We determine the NMEs for $\Upsilon(3S)$ and $\chi_b(3P)$ mesons from the $\Upsilon(3S)$ transverse momentum distributions measured by the CMS[@23; @24] and ATLAS[@25] Collaborations in the central rapidity region at $\sqrt s = 7$ and $13$ TeV (where the $k_T$-factorization approach is expected to be mostly relevant) and from the relative production ratio $R^{\chi_b(3P)}_{\Upsilon(3S)}$ measured recently by the LHCb Collaboration at $\sqrt s = 7$ and $8$ TeV[@35]. Then, we examine the extracted NMEs on the Tevatron and LHC data taken by the CDF[@41] and LHCb[@26; @27] Collaborations and make predictions for polarization parameters $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta\phi}$ (and frame-independent parameter $\tilde\lambda$), which determine the $\Upsilon(3S)$ spin density matrix and compare them to the currently available data[@28; @42].
The outline of our paper is the following. In Section 2 we briefly recall the basic steps of our calculations. In Section 3 we perform a numerical fit and extract the NMEs from the LHC data. Then we check the compatibility of the extracted NMEs with the available data on $\Upsilon(3S)$ yeild and polarization. Our conclusions are collected in Section 4.
Theoretical framework
=====================
In the present note we follow the approach described in the earlier publications[@43; @44; @45]. For the reader’s convenience, we briefly recall here main points of the theoretical scheme. Our consideration is based on the off-shell gluon-gluon fusion subprocesses that represent the true leading order (LO) in QCD: $$g^*(k_1) + g^*(k_2) \rightarrow \Upsilon[{}^3S_1^{(1)}](p) + g(k),$$ $$g^*(k_1) + g^*(k_2) \rightarrow \Upsilon[{}^1S_0^{(8)},{}^3S_1^{(8)},{}^3P_J^{(8)}](p).$$ $$g^*(k_1) + g^*(k_2) \rightarrow \chi_{bJ}(p)[{}^3P_J^{(1)},{}^3S_1^{(8)}] \rightarrow \Upsilon(p_1) + \gamma(p_2),$$
where $J = 0, 1$ or $2$ and the four-momenta of all particles are given in the parentheses. The color states taken into account are directly indicated. To obtain the production amplitudes for $b\bar b$ states with required quantum numbers from the ones for an unspecified $b\bar b$ state we use the appropriate projection operators. These operators for the spin-singlet and spin-triplet states read[@46]: $$\Pi_0 = (\hat p_{\bar b} - m_b)\gamma_5(\hat p_{b} + m_b)/m^{3/2},$$ $$\Pi_1 = (\hat p_{\bar b} - m_b)\hat\epsilon(S_z)(\hat p_{b} + m_b)/m^{3/2},$$
where $m = 2m_b$, $p_b = p/2 + q$ and $p_{\bar b} = p/2 - q$ are the four-momenta of the quark and anti-quark and $q$ is the four-momentum of quarks in the bound state, which is associated with the orbital angular momentum $L$. States with various projections of the spin momentum onto the $z$ axis are represented by the polarization four-vector $\hat\epsilon(S_z)$. Then, to calculate off-shell production amplitudes (1) — (3), one has to integrate the product of the hard scattering amplitude $A(q)$ expanded in a series around $q = 0$ and meson bound state wave function $\Psi^{(a)}(q)$ with respect to $q$: $$A(q)\Psi^{(a)}(q) = A|_{q=0}\Psi^{(a)}(q) + q^\alpha(\partial A/\partial q^\alpha)|_{q=0}\Psi^{(a)}(q) + \dots$$
A term-by-term integration of this series employs the identities[@46]: $$\int{{{d^3q}\over{(2\pi)^3}}\Psi^{(a)}(q)} = {{1}\over{\sqrt{4\pi}}}\mathcal{R}^{(a)}(0),$$ $$\int{{{d^3q}\over{(2\pi)^3}}q^\alpha\Psi^{(a)}(q)} = -i\epsilon^\alpha(L_z){{\sqrt{3}}\over{\sqrt{4\pi}}}\mathcal{R'}^{(a)}(0),$$
where $\mathcal{R}^{(a)}(x)$ is the radial wave function in the coordinate representation. The first term in (6) contributes to $S$-waves only and vanishes for $P$-wave. In contrast, the second term contributes only to $P$-waves and vanishes for $S$-wave. States with various projections of the orbital angular momentum onto the $z$ axis are represented by the polarization four-vector $\epsilon_\mu(L_z)$. The corresponding NMEs are directly related to the wave functions $\mathcal{R}^{(a)}(x)$ and their derivatives[@1; @2]: $$\langle\mathcal{O}^{\cal Q}[{}^{2S+1}L_J^{(a)}]\rangle = 2N_c(2J+1)|\mathcal{R}^{(a)}(0)|^2/4\pi,$$ $$\langle\mathcal{O}^{\cal Q}[{}^{2S+1}L_J^{(a)}]\rangle = 6N_c(2J+1)|\mathcal{R'}^{(a)}(0)|^2/4\pi$$
for $S$- and $P$-wave quarkonium $\cal Q$ respectively, where $N_c = 3$. Additionally, the NMEs obey the multiplicity relations coming from HQSS at LO: $$\langle\mathcal{O}^{\cal Q}[{}^{3}P_J^{(a)}]\rangle = (2J+1)\langle\mathcal{O}^{\cal Q}[{}^{3}P_0^{(a)}]\rangle.$$
A similar relation holds for color-octet $^3S_1^{(8)}$ states if $P$-wave quarkonia are considered. The color-singlet wave functions and their derivatives can be obtained from the potential model calculation[@47; @48] or extracted from the measured quarkonia decay widths. Further evaluation of partonic amplitudes is straightforward and was done in our previous papers[@43; @44; @45]. We only mention here that the summation over polarizations of the incoming off-shell gluons is performed according the BFKL prescription ${\sum {\epsilon^\mu\epsilon^{*\nu}}} = {\bm k}_T^\mu{\bm k}_T^\nu/{\bm k}_T^2$, where ${\bm k}_T$ is the gluon transverse momentum orthogonal to the beam axis[@36; @37]. The spin density matrix of the $S$-wave quarkonia is expressed in terms of the momenta $l_1$ and $l_2$ of the decay leptons and reads $${\sum {\epsilon^\mu\epsilon^{*\nu}}} = 3(l_1^\mu l_2^\nu + l_1^\nu l_2^\mu - {{m^2}\over{2}}g^{\mu\nu})/m^2.$$
This expression is equivalent to the standard expression $\sum{\epsilon^\mu\epsilon^{*\nu}} = - g^{\mu\nu} + p^\mu p^\nu/m^2$, but more suitable for determining the polarization observables. In all other respects the evaluation follows the standard QCD Feynman rules. The obtained results have been explicitly tested for gauge invariance by substituting the gluon momenta for corresponding polarization vectors. We have observed their gauge invariance even with off-shell initial gluons[^3].
As it was done for the prompt charmonia production[@21; @22], to describe the transition of an unbound octet $b\bar b$ quark pair to an observed singlet state we employ the mechanism proposed in[@20]. In this approach, a soft gluon with a small energy $E\sim\Lambda_{\rm QCD}$ is emitted after the hard interaction is over, bringing away the unwanted color and changing other quantum numbers of the produced CO system. In the conventional NRQCD calculations the emitted final state gluons are regarded as carrying no energy-momentum, that is in obvious contradiction with confinement, which prohibits the emission of infinitely soft colored quanta. In reality, the $b\bar b$ system must undergo a kind of final state interaction, where the energy-momentum exchange must be larger than the typical confinement scale. Thus, having small energy of the emitted gluons gives us the confidence that we do not enter the confinement or perturbative domains[^4]. This is not the matter of only kinematical corrections since one cannot organize transition amplitudes with correct spin properties without some finite energy-momentum transfer. In our calculations such soft gluon emission is described by a classical multipole expansion, in which the electric dipole ($E1$) transition dominates[@50]. Only a single $E1$ transition is needed to transform a $P$-wave state into an $S$-wave state and the structure of the respective ${^3P_J^{(8)}}\to {^3S_1^{(1)}}+g$ amplitudes is given by[@50]: $$A({}^3P_0^{(8)} \rightarrow \Upsilon + g) \sim k_\mu^{(g)}p^{\rm (CO)\mu}\epsilon_\nu^{(\Upsilon)}\epsilon^{(g)\nu},$$ $$A({}^3P_1^{(8)} \rightarrow \Upsilon + g) \sim e^{\mu\nu\alpha\beta}k_\mu^{(g)}\epsilon_\nu^{\rm (CO)}\epsilon_\alpha^{(\Upsilon)}\epsilon_\beta^{(g)},$$ $$A({}^3P_2^{(8)} \rightarrow \Upsilon + g) \sim p_\mu^{\rm (CO)}\epsilon_{\alpha\beta}^{\rm (CO)}\epsilon_\alpha^{(\Upsilon)} \left[ k_\mu^{(g)}\epsilon_\beta^{(g)} - k_\beta^{(g)}\epsilon_\mu^{(g)} \right],$$
where $p^{\rm (CO)}_\mu$, $k^{(g)}_\mu$, $\epsilon^{(\Upsilon)}_\mu$, $\epsilon^{(g)}_\mu$, $\epsilon^{\rm (CO)}_\mu$ and $\epsilon^{\rm (CO)}_\mu$ are the momenta and polarization vectors of corresponding particles and $e^{\mu\nu\alpha\beta}$ is the fully antisymmetric Levi-Civita tensor. The transformation of color-octet $S$-wave state into the color-singlet $S$-wave state is treated as two successive $E1$ transitions ${}^3S_1^{(8)} \rightarrow {}^3P_J^{(8)} + g$, ${}^3P_J^{(8)} \rightarrow {}^3S_1^{(1)} + g$ proceeding via either of three intermediate ${}^3P_J^{(8)}$ states with $J = 0,1,2$. For each of these transitions we apply the same expressions (13) — (15). Of course, all the expressions above are the same for gluons and photons (up to an overall color factor) and therefore can be used to calculate the polarization variables in radiative decays in feed-down process $\chi_b(3P) \rightarrow \Upsilon(3S) + \gamma$. Thus, the polarization of the outgoing $\Upsilon(3S)$ meson can then be calculated without any ambiguity.
The approach[@20] contrasts to conventional NRQCD calculations which show that heavy quarkonia produced from high-$p_T$ gluons as $^3S_1^{(8)}$ states carry strong transverse polarization. With our completely different view on the heavy quarkonia depolarization mechanism, we finally arrive at a completely different set of the fitted NMEs (see Section 3). The squares of the matrix elements, as being too lengthy, are not presented here but implemented into the newly developed parton-level Monte-Carlo event generator <span style="font-variant:small-caps;">pegasus</span>[@51].
The cross sections of $\Upsilon(3S)$ and $\chi_{b}(3P)$ production in the $k_T$-factorization approach are calculated as a convolution of the off-shell partonic cross sections and TMD gluon densities $f_g(x,\bm{k}^2_T,\mu^2)$ in a proton. The cross section for $2 \rightarrow 2$ and $2 \rightarrow 1$ subprocesses (1) — (3) can be written as: $$\begin{gathered}
\sigma = \int{{1}\over{8\pi(x_1x_2s) F}}f_g(x_1,\bm{k}^2_{1T},\mu^2)f_g(x_2,\bm{k}^2_{2T},\mu^2)\times\\
\times \overline{|A(g^*+g^*\rightarrow Q\bar Q + g)|^2}d\bm{p}^2_{T}d\bm{k}^2_{1T}d\bm{k}^2_{2T}dydy_g{{d\phi_1}\over{2\pi}}{{d\phi_2}\over{2\pi}},\end{gathered}$$ $$\sigma=\int{{{2\pi}\over{x_1x_2sF}}f_g(x_1,\bm{k}^2_{1T},\mu^2)f_g(x_2,\bm{k}^2_{2T},\mu^2)\overline{|A(g^*+g^*\rightarrow Q\bar Q)|^2}}d\bm{k}^2_{1T}d\bm{k}^2_{2T}dy{{d\phi_1}\over{2\pi}}{{d\phi_2}\over{2\pi}},
\label{sigma}$$
where $\phi_1$ and $\phi_2$ are the azimuthal angles of the initial off-shell gluons having the fractions of the momentum $x_1$ and $x_2$ and non-zero transverse momenta $\bm{k}^2_{1T}$ and $\bm{k}^2_{2T}$, $\bm{p}^2_{T}$ and $y$ are the transverse momentum and rapidity of produced mesons, $y_g$ is the rapidity of the outgoing gluon and $\sqrt{s}$ is the $pp$ center-of-mass energy. According to the general definition[@52], the off-shell gluon flux factor $F$ is defined as $F = 2\lambda^{1/2}(\hat s, k_1^2, k_2^2)$, where $\hat s = (k_1 + k_2)^2$ and $\lambda(x,y,z)$ is the known kinematic function. Note that for $2 \to 2$ subprocesses one can use the approximation $\lambda^{1/2}(\hat s, k_1^2, k_2^2) \simeq \hat s \simeq x_1 x_2 s$. However, it is not suitable for the $2 \to 1$ kinematics because the difference between $\hat s \simeq m^2_\Upsilon$ and $x_1 x_2 s = m_\Upsilon^2 + p_T^2$ can make pronounced effect on the $p_T$ spectrum. This effect is specially discussed in Section 3.
In the present paper we have tested a few sets of the TMD gluon densities in a proton, namely, A0[@53], JH’2013 set 1[@54] and KMR[@55] ones[^5]. First two of them were obtained from the numerical solutions of the CCFM gluon evolution equation. The CCFM equation provides a suitable tool since it converges to the BFKL equation in the region of small $x$ and to the DGLAP equation at large $x$ (see[@39] for more details). The typical values of the variable $x$ probed in the considered processes are of order $x \sim (m_\Upsilon^2 + p_T^2)^{1/2}/\sqrt s$ at central rapidities, that corresponds to $x \sim 10^{-3} \, ... \, 10^{-2}$ in the kinematical conditions of the CMS and ATLAS experiments[@23; @24; @25]. Thus, the CCFM evolution can be used in the whole $p_T$ range. The input parameters of these gluon distributions were determined from the best description of the precision DIS data on the proton structure functions $F_2(x, Q^2)$. Additionally, we have used a set obtained with Kimber-Martin-Ryskin (KMR) prescription[@55], which provides a method to evaluate the TMD parton densities from the conventional (collinear) ones. For the input, we have used recent LO NNPDF3.1 set[@57]. The A0, JH’2013 and KMR gluon densities are shown in Fig. 1 as a function of ${\mathbf k}_T^2$ for different values of $x$ and $\mu^2$. One can observe a difference in the absolute normalization and shape between all these TMD gluon distributions. Below we discuss the corresponding phenomenological consequences.
The renormalization $\mu_R$ and factorization $\mu_F$ scales were set to $\mu_R^2 = m_\Upsilon^2 + p_T^2$ and $\mu_F^2 = \hat s + {\mathbf Q}_T^2$ for CCFM-evolved gluon densities, where ${\mathbf Q}_T$ is the transverse momentum of the initial off-shell gluon pair. The choice of $\mu_R$ is a standard for bottomonia production, while the special choice of $\mu_F$ is connected with the CCFM evolution (see[@53; @54]). In the KMR calculations, we used standard choice $\mu_R^2 = \mu_F^2 = m_\Upsilon^2 + p_T^2$.
The parton level calculations were performed using the Monte-Carlo event generator <span style="font-variant:small-caps;">pegasus</span>[@51].
Numerical results
=================
As it was mentioned above, in the present paper we concentrate on the inclusive $\Upsilon(3S)$ and $\chi_b(3P)$ production, leaving other bottomonia states for forthcoming studies. Below we set the masses $m_{\Upsilon(3S)} = 10.3552$ GeV, $m_{\chi_{b1}(3P)} = 10.512$ GeV and $m_{\chi_{b2}(3P)} = 10.522$ GeV [@58] and adopt the usual non-relativistic approximation $m_b = m_{\cal Q}/2$ for the beauty quark mass, where $m_{\cal Q}$ is the mass of bottomonium $\cal Q$. We set the branching ratios $B(\Upsilon(3S) \rightarrow \mu^+\mu^-) = 0.0218$[@58], $B(\chi_{b1}(3P)\rightarrow \Upsilon(3S) + \gamma) = 0.1044$ and $B(\chi_{b2}(3P) \rightarrow \Upsilon(3S) + \gamma) = 0.0611$[@34]. Note that there are no experimental data for branching ratios of $\chi_b(3P)$, so the values above are the results of an assumption[@34] that the total decay widths of $\chi_b(mP)$ are approximately independent on $m$. Following experimental analysis[@35], we neglected the $\chi_{b0}(3P)$ contribution as it is almost zero. We use the one-loop formula for the coupling $\alpha_s$ with $n_f = 4(5)$ quark flavours at $\Lambda_{\rm QCD} = 250(167)$ MeV for A0 (KMR) gluon density and two-loop expression for $\alpha_s$ with $n_f = 4$ and $\Lambda_{\rm QCD} = 200$ MeV for JH’2013 set 1 gluon. As a commonly adopted choice, we set CS NMEs $\langle\mathcal{O}(\Upsilon[{}^{3}S_1^{(1)}])\rangle = 3.54$ GeV$^3$ and $\langle\mathcal{O}(\chi[{}^{3}P_0^{(1)}])\rangle = 2.83$ GeV$^5$. These values were obtained in the potential model calculations[@47].
Fit of color octet NMEs
-----------------------
We have performed a global fit to the $\Upsilon(3S)$ production data at the LHC and determined the corresponding NMEs for both $\Upsilon(3S)$ and $\chi_b(3P)$ mesons. We have included in the fitting procedure the $\Upsilon(3S)$ transverse momentum distributions measured by the CMS[@23; @24] and ATLAS[@25] Collaborations at $\sqrt s = 7$ and $13$ TeV and central rapidities, where our $k_T$-factorization calculations are most relevant due to essentially low-$x$ region probed. To determine NMEs for $\chi_b(3P)$ mesons, we also included into the fit the recent LHCb data[@35] on the radiative $\chi_b(3P) \to \Upsilon(3S) + \gamma$ decays taken at $\sqrt s = 7$ and $8$ TeV. We have excluded from our fit low $p_T$ region and consider only the data at $p_T > p_T^{\rm cut} = 10$ GeV, where the NRQCD formalism is believed to be mostly reliable. As it was already mentioned above, the double NRQCD expansion in $\alpha_s$ and $v_Q$ is not good at low $p_T$, where a more accurate treatment of large logarithms $\sim \ln m_{\Upsilon}^2/p_T^2$ and other nonperturbative effects becomes necessary[^6].
Before we proceed with the numerical fit, we would like to point out a few points. First of them is connected with the importance of proper definition of the off-shell flux factor for $2 \to 1$ subprocesses (2) and (3). The definition of the flux, which is the velocity of the off-shell interacting partons, is not clear and can be disputable. As it was mentioned above, we use the “$\lambda^{1/2}$” prescription $F = 2\lambda^{1/2}(\hat s, k_1^2, k_2^2)$ in factorization formula (\[sigma\]). Our choice is based on the toy simulation[@59] of $\chi_c$ meson production in $e^+e^-$ collisions. It was argued[@59] that such definition leads to a good agreement of calculations based on Equivalent Photon Approximation and exact ${\cal O}(\alpha^4)$ results. Contrary, the calculations performed with using conventional (collinear) $2 \to 1$ flux treatment $\lambda^{1/2}(\hat s, k_1^2, k_2^2) \simeq x_1 x_2 s$ did not reproduce the latter and therefore, in our opinion, seems to be rather doubtful[^7].
Our calculation shows that the “$\lambda^{1/2}$” prescription results in different $p_T$ shapes of color-octet $^1S_0^{(8)}$ and $^3P_J^{(8)}$ contributions to the $\Upsilon(3S)$ production. Let us consider the ratio $R$ defined as $$R = { m_{\Upsilon(3S)}^2 \sum\limits_{J = 0}^{2} (2J+1) \, d\sigma[\Upsilon(3S), {}^3P_J^{(8)}]/dp_T \over d\sigma [\Upsilon(3S), {}^1S_0^{(8)}]/dp_T}
\label{eqrf}$$
as a function of $\Upsilon(3S)$ meson transverse momentum. While the calculations with collinear treatment of the flux factor $F$ show a flat behavior of this ratio in a wide $p_T$ region $10 < p_T < 100$ GeV (see Fig. 2, left panel) the calculations performed with using the “$\lambda^{1/2}$” prescription demonstrate the strong rise of the ratio $R$ with increasing $p_T$ giving us a possibility to separately extract the values of $\langle\mathcal{O}^{\Upsilon(3S)}[{}^{1}S_0^{(8)}]\rangle$ and $\langle\mathcal{O}^{\Upsilon(3S)}[{}^{3}P_0^{(8)}]\rangle$ from the experimental data. The latter turns out to be impossible when one inconsistently uses the collinear treatment of flux factor in the $k_T$-factorization calculations.
Our next point is connected with the correct treatment of feed-down contributions from the radiative decays of $\chi_b(3P)$ mesons, observed recently by the LHCb Collaboration[@35]. We found that the $p_T$ shape of the direct $\Upsilon[^3S_1^{(8)}]$ and feed-down $\chi_b[^3S_1^{(8)}]$ contributions is almost the same in all kinematical regions probed by the LHC and Tevatron experiments. Thus, the ratio $$r = { \sum\limits_{J = 0}^{2} (2J+1) \, B(\chi_{bJ}(3P) \to \Upsilon(3S) + \gamma) d\sigma[\chi_{bJ}(3P), {}^3S_1^{(8)}]/dp_T \over d\sigma [\Upsilon(3S), {}^3S_1^{(8)}]/dp_T }
\label{eqr}$$
can be well approximated by a constant for a wide $\Upsilon(3S)$ transverse momentum $p_T$ and rapidity $y$ ranges at different energies, as it is demonstrated in Fig. 2 (right panel). We estimate the mean-square average $r = 0.654 \pm 0.005$, which is practically independent on the TMD gluon density in a proton. Since up to now there are no experimental data on the $\chi_b(3P)$ transverse momentum distributions, we cannot separately determine the values of $\langle\mathcal{O}^{\Upsilon(3S)}[{}^{3}S_1^{(8)}]\rangle$ and $\langle\mathcal{O}^{\chi_{b0}(3P)}[{}^{3}S_1^{(8)}]\rangle$ from the available $\Upsilon(3S)$ data[@23; @24; @25]. Instead, we introduce the linear combination $$M_r = \langle\mathcal{O}^{\Upsilon(3S)}[{}^{3}S_1^{(8)}]\rangle + r \langle\mathcal{O}^{\chi_{b0}(3P)}[{}^{3}S_1^{(8)}]\rangle,$$ which can be extracted from the measured $\Upsilon(3S)$ transverse momentum distributions. Then we use recent LHCb data[@35] on the fraction of $\Upsilon(3S)$ mesons originating from the $\chi_b(3P)$ radiative decays measured at $\sqrt s = 7$ and $8$ TeV. To be precise, the LHCb Collaboration reported the ratio $$R^{\chi_b(3P)}_{\Upsilon(3S)} = \sum\limits_{J = 1}^{2} {\sigma(pp\rightarrow \chi_{bJ}(3P)+X) \over \sigma(pp \rightarrow \Upsilon(3S) + X)} \times B(\chi_{bJ} \to \Upsilon(3S) + \gamma),
\label{eqrUp}$$
where the possible contributions from $\chi_{b0}(3P)$ decays are neglected because of the small branching fraction. From the known $M_r$ and $R^{\chi_b(3P)}_{\Upsilon(3S)}$ values one can separately determine the $\langle\mathcal{O}^{\Upsilon(3S)}[{}^{3}S_1^{(8)}]\rangle$ and $\langle\mathcal{O}^{\chi_{b0}(3P)}[{}^{3}S_1^{(8)}]\rangle$, thus reconstructing full map of color octet NMEs for both $\Upsilon(3S)$ and $\chi_b(3P)$ mesons.
Using the strategy described above, we performed a numerical fit of $\Upsilon(3S)$ and $\chi_b(3P)$ NMEs. Nowhere we impose any kinematic restrictions but the experimental acceptance. The fitting procedure was separately done in each of the rapidity subdivisions (using the fitting algorithm as implemented in the commonly used <span style="font-variant:small-caps;">gnuplot</span> package[@61]) under the requirement that all the NMEs are strictly positive. Then, the mean-square average of the fitted values was taken. The corresponding uncertainties are estimated in the conventional way using Student’s t-distribution at the confidence level $P = 80$%. The results of our fits are collected in Table 1. For comparison, we also presented there the NMEs obtained in the conventional NLO NRQCD by other authors[@33]. We have found that extracted values of $\langle\mathcal{O}^{\Upsilon(3S)}[{}^{1}S_0^{(8)}]\rangle$ are compatible with zero for all the TMD gluon densities. However, other color octet NMEs strongly depend on the latter, although JH’2013 set 1 and KMR gluons result in the more or less close values. The dependence of the fitted NMEs values on the TMD gluon densities reflects their different $x$ and ${\mathbf k}_T^2$ behavior, that is the consequence of different approaches to evaluate them. The corresponding $\chi^2/d.o.f.$ are listed in Table 2, where we additionally show their dependence on the $p_T^{\rm cut}$. As one can see, the $\chi^2/d.o.f.$ decreases when $p_T^{\rm cut}$ grows up and the best fit of the data is achieved with A0 gluon. We note that the returned relatively large (but still reasonable) $\chi^2/d.o.f.$ values are connected with the recent precision CMS data[@24] included into the fit. So, as an exercise, we have excluded these data and repeated the fit procedure using the ATLAS data[@25] only. In this way, the $\chi^2/d.o.f. \sim 1$ was obtained for all the considered TMD gluon densities.
All the data used in the fits are compared with our predictions in Figs. 3 — 5. Note that the data at $p_T > p_T^{\rm cut} = 10$ GeV are only shown. The shaded areas represent the theoretical uncertainties of our calculations, which include the scale uncertainties, uncertainties coming from the NME fitting procedure and uncertainties connected with the choice of the intermediate color-octet mass, added in quadrature. To estimate the scale uncertainties the standard variations in the scale $\mu_R\to 2\mu_R$ or $\mu_R\to\mu_R/2$ were introduced through replacing the gluon densities A0 and JH’2013 set 1 with A0$+$ and JH’2013 set 1$+$, or with A0$-$ and JH’2013 set 1$-$. This was done to preserve the intrinsic correspondence between the TMD set and the scale used in the evolution equation (see[@53; @54]). To estimate the uncertainties connected with the intermediate color-octet mass we have varied amount of energy $E$ emitted in the course of transition of an unbound color octet $b\bar b$ pair into the observed bottomonium by a factor of $2$ around its default value $E = \Lambda_{\rm QCD}$. We find that the main effect here is only in changing the overall normalization with almost no changes in the shape of the $p_T$ spectrum (see also[@62]). These uncertainties are about of $20$% and therefore comparable with the scale uncertainties. One can see that we have achieved a reasonably good description of the CMS[@23; @24] and ATLAS[@25] data in the whole $p_T$ range within the experimental and theoretical uncertainties for the $\Upsilon(3S)$ transverse momentum distributions. The ratio $R^{\chi_b(3P)}_{\Upsilon(3S)}$ measured by the LHCb Collaboration[@35] is well described also. At large $p_T$, the JH’2013 set 1 and KMR gluons tend to overestimate the latest CMS data[@24] taken at $\sqrt s = 13$ TeV, but agree well with other measurements. This result shows a dependence of our predictions on the TMD gluon densities in certain kinematical regions.
With obtained NMEs for $\Upsilon(3S)$ and $\chi_b(3P)$ mesons, we achieved reasonably good description (of course, at $p_T > p_T^{\rm cut}$) of the earlier CDF data[@41] taken at the $\sqrt s = 1.8$ TeV and recent data[@26; @27] taken by the LHCb Collaboration at $\sqrt s = 7$, $8$ and $13$ TeV and forward rapidities, see Fig. 6. We find that the KMR gluon distribution is able to describe well the CDF data even at low $p_T$ region, $p_T < 10$ GeV. Some discrepancy between the LHCb data and our predictions observed in very forward region $4 < y < 4.5$ at $\sqrt s = 7$ TeV can be easily understood since here one can probe the essentially large-$x$ region, there the $k_T$-factorization becomes less applicable.
The consequence of our fit for $\Upsilon(3S)$ polarization is discussed in the next Section.
$\Upsilon(3S)$ polarization
---------------------------
As it is well known, the polarization of any vector meson can be described with three parameters $\lambda_\theta$, $\lambda_\phi$ and $\lambda_{\theta\phi}$, which determine the spin density matrix of a meson decaying into a lepton pair and can be measured experimentally. The double differential angular distribution of the decay leptons can be written as[@63]: $${{d\sigma}\over{d\cos\theta^*d\phi^*}} \sim {{1}\over{3+\lambda_\theta}}(1 + \lambda_\theta\cos^2\theta^* + \lambda_\phi\sin^2\theta^*\cos2\phi^* + \lambda_{\theta\phi}\sin2\theta^*\cos\phi^*),
\label{eqlam}$$
where $\theta^*$ and $\phi^*$ are the polar and azimuthal angles of the decay lepton measured in the meson rest frame. The case of $(\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta \phi}) = (0,0,0)$ corresponds to unpolarized state, while $(\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta \phi}) = (1,0,0)$ and $(\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta \phi}) = (-1,0,0)$ refer to fully transverse and fully longitudinal polarizations.
The CMS Collaboration has measured all of these parameters as functions of $\Upsilon(3S)$ transverse momentum in three complementary frames: the Collins-Soper, helicity and perpendicular helicity ones at $\sqrt{s} = 7$ TeV[@28]. The CDF Collaboration also measured these parameters in the helicity frame at $\sqrt{s} = 1.96$ TeV[@42]. In the Collins-Soper frame the polarization axis $z$ bisects the two beam directions whereas the polarization axis in the helicity frame coincides with the $\Upsilon(3S)$ direction in the laboratory frame. In the perpendicular helicity frame the $z$ axis is orthogonal to that in the Collins-Soper frame and lies in the plane spanned by the two beam ($P_1$ and $P_2$) momenta. In all cases, the $y$ axis is taken to be in the direction of the vector product of the two beam directions in the $\Upsilon(3S)$ rest frame, $\vec P_1 \times \vec P_2$ and $\vec P_2 \times \vec P_1$ for positive and negative rapidities, respectively. Additionally, the frame-independent parameter $\tilde \lambda = (\lambda_\theta + 3\lambda_\phi)/(1 - \lambda_\phi)$ has been studied[@28; @42]. Below we estimate the polarization parameters $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta \phi}$ and $\tilde \lambda$ for the CMS and CDF conditions[^8]. As it was done earlier[@22; @43; @44; @45], our calculation generally follows the experimental procedure. We collect the simulated events in the kinematical region defined by the CMS and CDF experiments, generate the decay lepton angular distributions according to the production and decay matrix elements and then apply a three-parametric fit based on (22). Of course, we took into account the polarization of $\Upsilon(3S)$ mesons originated from radiative $\chi_b(3P)$ decays, that is in full agreement with the experimental setup.
Our results are presented in Figs. 7 — 10. These calculations were done using the A0 gluon density which provides the best description of the measured $\Upsilon(3S)$ transverse momenta distributions. The obtained predictions for the $\Upsilon(3S)$ polarization parameters have a reasonable agreement with the CMS and CDF data. In all the kinematical regions we find only weak or zero polarization, which coincides with the measurements within the uncertainties. These predictions are practically independent of the $\Upsilon(3S)$ rapidity. The absence of strong polarization is not connected with parameter tuning, but seems to be a natural and rather general feature of the scenario[@20]. Thus, one can conclude that treating the soft gluon emissions within the NRQCD as a series of color-electric dipole transitions does not contradicts the available Tevatron and LHC data on the $\Upsilon(3S)$ production. The same conclusion was done for charmonia family[@21; @22].
Finally, we would like to note that the qualitative predictions for the $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta\phi}$ and $\tilde \lambda$ are stable with respect to variations in the model parameters. In fact, there is practically no dependence on the strong coupling constant and TMD gluon densities, i.e. two of important sources of theoretical uncertainties cancel out. So, the proposed way, in our opinion, can provide an easy and natural solution to the quarkonia production and polarization puzzle.
Conclusion
==========
We have considered the $\Upsilon(3S)$ production at the Tevatron and LHC in the framework of $k_T$-factorization approach. Our consideration was based on the off-shell production amplitudes for hard partonic subprocesses (including both color-singlet and color-octet contributions), NRQCD formalism for the formation of bound states and TMD gluon densities in a proton (derived from the CCFM evolution equation and KMR scheme as well). Treating the nonperturbative color octet transitions in terms of multipole radiation theory and taking into account feed-down contributions from the radiative $\chi_b(3P)$ decays, we extracted $\Upsilon(3S)$ and $\chi_b(3P)$ NMEs in a fit to $\Upsilon(3S)$ transverse momentum distributions measured by the CMS and ATLAS Collaborations at $\sqrt s = 7$ and $13$ TeV. We have inspected the extracted NMEs with the available Tevatron and LHC data taken in different kinematical regions and demostrated that these NMEs do not contradict the data. We found that the best description is achieved with the CCFM-evolved A0 gluon density, although the KMR one is able to describe the data even at low transverse momenta. Then we estimated polarization parameters $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta \phi}$ and frame-independent parameter $\tilde \lambda$ which determine the $\Upsilon(3S)$ spin density matrix. We show that treating the soft gluon emission as a series of explicit color-electric dipole transitions within the NRQCD leads to unpolarized $\Upsilon(3S)$ production at moderate and large transverse momenta, that is in agreement with the Tevatron and LHC data.
Acknowledgements {#acknowledgements .unnumbered}
================
The authors thank S.P. Baranov, M.A. Malyshev and H. Jung for their interest, useful discussions and important remarks. N.A.A. is supported by the Foundation for the Advancement of Theoretical Physics and Mathematics “Basis” (grant No.18-1-5-33-1) and by the RFBR grant 19-32-90096. A.V.L. is grateful the DESY Directorate for the support in the framework of Cooperation Agreement between MSU and DESY on phenomenology of the LHC processes and TMD parton densities.
[99]{} G. Bodwin, E. Braaten, G. Lepage, Phys. Rev. D [**51**]{}, 1125 (1995). P. Cho, A.K. Leibovich, Phys. Rev. D [**53**]{}, 150 (1996); Phys. Rev. D [**53**]{}, 6203 (1996). B. Gong, X.Q. Li, J.-X. Wang, Phys. Lett. B [**673**]{}, 197 (2009). Y.-Q. Ma, K. Wang, K.-T. Chao, Phys. Rev. Lett. [**106**]{}, 042002 (2011). M. Butenschön, B.A. Kniehl, Phys. Rev. Lett. [**108**]{}, 172002 (2012). K.-T. Chao, Y.-Q. Ma, H.-S. Shao, K. Wang, Y.-J. Zhang, Phys. Rev. Lett. [**108**]{}, 242004 (2012). B. Gong, L.-P. Wan, J.-X. Wang, H.-F. Zhang, Phys. Rev. Lett. [**110**]{}, 042002 (2013). Y.-Q. Ma, K. Wang, K.-T. Chao, H.-F. Zhang, Phys. Rev. D [**83**]{}, 111503 (2011). A.K. Likhoded, A.V. Luchinsky, S.V. Poslavsky, Phys. Rev. D [**90**]{}, 074021 (2014). H.-F. Zhang, L. Yu, S.-X. Zhang, L. Jia, Phys. Rev. D [**93**]{}, 054033 (2016). H. Han, Y.-Q. Ma, C. Meng, H.-S. Shao, K.-T. Chao, Phys. Rev. Lett. [**114**]{}, 092005 (2015). H.-F. Zhang, Z. Sun, W.-L. Sang, R. Li, Phys. Rev. Lett. [**114**]{}, 092006 (2015). M. Butenschön, Z. G. He, B.A. Kniehl, Phys. Rev. Lett. [**114**]{}, 092004 (2015). S.S. Biswal, K. Sridhar, J. Phys. G: Nucl. Part. Phys. [**39**]{}, 015008 (2012). A.K. Likhoded, A.V. Luchinsky, S.V. Poslavsky, Mod. Phys. Lett. A [**30**]{}, 1550032 (2015). P. Artoisenet, J. Campbell, J.P. Lansberg, F. Maltoni, F. Tramontano, Phys. Rev. Lett. [**101**]{}, 152001 (2008). J.-P. Lansberg, H.-S. Shao, H.-F. Zhang, Phys. Lett. B [**786**]{} 342 (2018). Y. Feng, J. He, J.-P. Lansberg, H.-S. Shao, A. Usachov, H.-F. Zhang, arXiv:1901.09766 \[hep-ph\]. J.-P. Lansberg, arXiv:1903.09185 \[hep-ph\]. S.P. Baranov, Phys. Rev. D [**93**]{}, 054037 (2016). S.P. Baranov, A.V. Lipatov, Eur. Phys. J. C [**79**]{}, 621 (2019). S.P. Baranov, A.V. Lipatov, arXiv:1906.07182 \[hep-ph\]. CMS Collaboration, Phys. Lett. B [**749**]{}, 14 (2015). CMS Collaboration, Phys. Lett. B [**780**]{}, 251 (2018). ATLAS Collaboration, Phys. Rev. D [**87**]{}, 052004 (2013). LHCb Collaboration, JHEP [**1511**]{}, 103 (2015). LHCb Collaboration, JHEP [**1807**]{}, 134 (2018). CMS Collaboration, Phys. Rev. Lett. [**110**]{}, 081802 (2013). LHCb Collaboration, JHEP [**1217**]{}, 110 (2017). B. Gong, J.-X. Wang, H.-F. Zhang, Phys. Rev. D [**83**]{}, 114021 (2011). K. Wang, Y.-Q. Ma, K.-T. Chao, Phys. Rev. D [**85**]{}, 114003 (2012). B. Gong, L.-P. Wan, J.-X. Wang, H.-F. Zhang, Phys. Rev. Lett. [**112**]{}, 032001 (2014). Y. Feng, B. Gong, L.-P. Wan, J.-X. Wang, H.-F. Zhang, Chin. Phys. C [**39**]{}, 123102 (2015). H. Han, Y.-Q. Ma, C. Meng, H.-S. Shao, Y.-J. Zhang, K.-T. Chao, Phys. Rev. D [**94**]{}, 014028 (2016). LHCb Collaboration, Eur. Phys. J. C [**74**]{}, 3092 (2014). S. Catani, M. Ciafaloni, F. Hautmann, Nucl. Phys. B [**366**]{}, 135 (1991);\
J.C. Collins, R.K. Ellis, Nucl. Phys. B [**360**]{}, 3 (1991). L.V. Gribov, E.M. Levin, M.G. Ryskin, Phys. Rep. [**100**]{}, 1 (1983);\
E.M. Levin, M.G. Ryskin, Yu.M. Shabelsky, A.G. Shuvaev, Sov. J. Nucl. Phys. [**53**]{}, 657 (1991). E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Sov. Phys. JETP [**44**]{}, 443 (1976);\
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Sov. Phys. JETP [**45**]{}, 199 (1977);\
I.I. Balitsky, L.N. Lipatov, Sov. J. Nucl. Phys. [**28**]{}, 822 (1978). M. Ciafaloni, Nucl. Phys. B [**296**]{}, 49 (1988);\
S. Catani, F. Fiorani, G. Marchesini, Phys. Lett. B [**234**]{}, 339 (1990);\
S. Catani, F. Fiorani, G. Marchesini, Nucl. Phys. B [**336**]{}, 18 (1990);\
G. Marchesini, Nucl. Phys. B [**445**]{}, 49 (1995). R. Angeles-Martinez et al., Acta Phys. Polon. B [**46**]{}, 2501 (2015). CDF Collaboration, Phys. Rev. Lett. [**88**]{}, 161802 (2002). CDF Collaboration, Phys. Rev. Lett. [**108**]{}, 151802 (2012). S.P. Baranov, A.V. Lipatov, N.P. Zotov, Eur. Phys. J. C [**75**]{}, 455 (2015). S.P. Baranov, A.V. Lipatov, N.P. Zotov, Phys. Rev. D [**93**]{}, 094012 (2016). S.P. Baranov, A.V. Lipatov, Phys. Rev. D [**96**]{}, 034019 (2017). C.-H. Chang, Nucl. Phys. B [**172**]{}, 425 (1980);\
E.L. Berger, D.L. Jones, Phys. Rev. D [**23**]{}, 1521 (1981);\
R. Baier, R. Rückl, Phys. Lett. B [**102**]{}, 364 (1981);\
S.S. Gershtein, A.K. Likhoded, S.R. Slabospitsky, Sov. J. Nucl. Phys. [**34**]{}, 128 (1981). E.J. Eichten, C. Quigg, arXiv:1904.11542 \[hep-ph\]. E.J. Eichten, C. Quigg, Phys. Rev. D [**52**]{}, 1726 (1995). B.A. Kniehl, D.V. Vasin, V.A. Saleev, Phys. Rev. D [**73**]{}, 074022 (2006). A.V. Batunin, S.R. Slabospitsky, Phys. Lett B [**188**]{}, 269 (1987);\
P. Cho, M. Wise, S. Trivedi, Phys. Rev. D [**51**]{}, R2039 (1995). S.P. Baranov, A.V. Lipatov, M.A. Malyshev, in preparation. E. Bycling, K. Kajantie, Particle Kinematics, John Wiley and Sons (1973). H. Jung, arXiv:hep-ph/0411287. F. Hautmann, H. Jung, Nucl. Phys. B [**883**]{}, 1 (2014). M.A. Kimber, A.D. Martin, M.G. Ryskin, Phys. Rev. D [**63**]{}, 114027 (2001);\
A.D. Martin, M.G. Ryskin, G. Watt, Eur. Phys. J. C [**31**]{}, 73 (2003);\
A.D. Martin, M.G. Ryskin, G. Watt, Eur. Phys. J. C [**66**]{}, 163 (2010). http://tmd.hepforge.org NNPDF Collaboration, Eur. Phys. J. C [**77**]{}, 663 (2017). PDG Collaboration, Phys. Rev. D [**98**]{}, 030001 (2018). S.P. Baranov, A. Szcurek, Phys. Rev. D [**77**]{}, 054016 (2008). B.A. Kniehl, M.A. Nefedov, V.A. Saleev, Phys. Rev. D 94, 054007 (2016). www.gnuplot.info S.P. Baranov, Phys. Rev. D [**91**]{} 034011 (2015). M. Beneke, M. Krämer, M. Vänttinen, Phys. Rev. D [**57**]{}, 4258 (1998).
[lcccc]{}\
& A0 & JH’2013 set 1 & KMR & NLO NRQCD[@33]\
\
\
$\langle\mathcal{O}^{\Upsilon(3S)}[{}^{3}S_1^{(1)}]\rangle$/GeV$^{3}$ & $3.54$ & $3.54$ & $3.54$ & $3.54$\
\
$\langle\mathcal{O}^{\Upsilon(3S)}[{}^{1}S_0^{(8)}]\rangle$/GeV$^3$ & $0.0$ & $0.0$ & $0.0$ & $-0.0107 \pm 0.0107$\
\
$\langle\mathcal{O}^{\Upsilon(3S)}[{}^{3}S_1^{(8)}]\rangle$/GeV$^3$ & $0.018 \pm 0.001$ & $0.007 \pm 0.002$ & $0.006 \pm 0.001$ & $0.0271 \pm 0.0013$\
\
$\langle\mathcal{O}^{\Upsilon(3S)}[{}^{3}P_0^{(8)}]\rangle$/GeV$^{5}$ & $0.0$ & $0.09 \pm 0.03$ & $0.073 \pm 0.006$ & $0.0039 \pm 0.0023$\
\
$\langle\mathcal{O}^{\chi_{b0}(3P)}[{}^{3}P_0^{(1)}]\rangle$/GeV$^{5}$ & $2.83$ & $2.83$ & $2.83$ & $2.83$\
\
$\langle\mathcal{O}^{\chi_{b0}(3P)}[{}^{3}S_1^{(8)}]\rangle$/GeV$^{3}$ & $0.016 \pm 0.003$ & $0.009 \pm 0.001$ & $0.005 \pm 0.001$ & —\
\
[lcccc]{}\
& $p_T^{\rm cut} = 10$ GeV & $p_T^{\rm cut} = 12$ GeV & $p_T^{\rm cut} = 15$ GeV & $p_T^{\rm cut} = 17$ GeV\
\
\
A0 & $2.35$ & $1.99$ & $1.79$ & $1.72$\
\
JH’2013 set 1 & $4.22$ & $3.59$ & $3.28$ & $3.21$\
\
KMR & $2.59$ & $2.43$ & $2.38$ & $2.37$\
\
![The TMD gluon densities in the proton calculated as a function of the gluon transverse momentum ${\mathbf k}_T^2$ at different longitudinal momentum fractions $x = 10^{-2}$ (left panel) or $x = 10^{-3}$ (right panel) and $\mu^2 = 10^{4}$ GeV$^2$.[]{data-label="fig1"}](fg-x-001.png "fig:"){width="7.0cm"} ![The TMD gluon densities in the proton calculated as a function of the gluon transverse momentum ${\mathbf k}_T^2$ at different longitudinal momentum fractions $x = 10^{-2}$ (left panel) or $x = 10^{-3}$ (right panel) and $\mu^2 = 10^{4}$ GeV$^2$.[]{data-label="fig1"}](fg-x-0001.png "fig:"){width="7.0cm"}
![The production ratios $R$ (left panel) and $r$ (right panel) calculated as a function of $\Upsilon(3S)$ transverse momentum $p_T$ in the different kinematical regions.[]{data-label="fig2"}](Rf.png "fig:"){width="7.0cm"} ![The production ratios $R$ (left panel) and $r$ (right panel) calculated as a function of $\Upsilon(3S)$ transverse momentum $p_T$ in the different kinematical regions.[]{data-label="fig2"}](r.png "fig:"){width="6.85cm"}
![Transverse momentum distribution of inclusive $\Upsilon(3S)$ production calculated at $\sqrt s = 7$ TeV in the different rapidity regions. The red, green and blue histograms correspond to the predictions obtained with A0, KMR and JH’2013 set 1 gluon densities. Shaded bands represent the total uncertainties of our calculations, as it is described in text. The experimental data are from ATLAS[@25].[]{data-label="fig3"}](pt-ATLAS-I.png "fig:"){width="7.0cm"} ![Transverse momentum distribution of inclusive $\Upsilon(3S)$ production calculated at $\sqrt s = 7$ TeV in the different rapidity regions. The red, green and blue histograms correspond to the predictions obtained with A0, KMR and JH’2013 set 1 gluon densities. Shaded bands represent the total uncertainties of our calculations, as it is described in text. The experimental data are from ATLAS[@25].[]{data-label="fig3"}](pt-ATLAS-II.png "fig:"){width="7.0cm"}
![Transverse momentum distribution of inclusive $\Upsilon(3S)$ production calculated at $\sqrt s = 7$ TeV (upper histograms) and $\sqrt s = 13$ TeV (lower histograms, divided by $100$) in the different rapidity regions. Notation of all histograms is the same as in Fig. 3. The experimental data are from CMS[@23; @24].[]{data-label="fig4"}](pt-CMS-I.png "fig:"){width="7.0cm"} ![Transverse momentum distribution of inclusive $\Upsilon(3S)$ production calculated at $\sqrt s = 7$ TeV (upper histograms) and $\sqrt s = 13$ TeV (lower histograms, divided by $100$) in the different rapidity regions. Notation of all histograms is the same as in Fig. 3. The experimental data are from CMS[@23; @24].[]{data-label="fig4"}](pt-CMS-II.png "fig:"){width="7.0cm"} ![Transverse momentum distribution of inclusive $\Upsilon(3S)$ production calculated at $\sqrt s = 7$ TeV (upper histograms) and $\sqrt s = 13$ TeV (lower histograms, divided by $100$) in the different rapidity regions. Notation of all histograms is the same as in Fig. 3. The experimental data are from CMS[@23; @24].[]{data-label="fig4"}](pt-CMS-III.png "fig:"){width="7.0cm"}
![The ratio $R^{\chi_b(3P)}_{\Upsilon(3S)}$ calculated as function of $\Upsilon(3S)$ transverse momentum at $\sqrt{s} = 7$ TeV (left panel) and $\sqrt{s} = 8$ TeV (right panel). Notation of all histograms is the same as in Fig. 3. The experimental data are from LHCb[@35].[]{data-label="fig5"}](R-7.png "fig:"){width="7.0cm"} ![The ratio $R^{\chi_b(3P)}_{\Upsilon(3S)}$ calculated as function of $\Upsilon(3S)$ transverse momentum at $\sqrt{s} = 7$ TeV (left panel) and $\sqrt{s} = 8$ TeV (right panel). Notation of all histograms is the same as in Fig. 3. The experimental data are from LHCb[@35].[]{data-label="fig5"}](R-8.png "fig:"){width="7.0cm"}
![Transverse momentum distribution of inclusive $\Upsilon(3S)$ production calculated at $\sqrt s = 1.8$, $7$, $8$ and $13$ TeV in the different rapidity regions. Notation of all histograms is the same as in Fig. 3. The experimental data are from CDF[@41] and LHCb[@26; @27].[]{data-label="fig6"}](pt-LHCb-I.png "fig:"){width="7.0cm"} ![Transverse momentum distribution of inclusive $\Upsilon(3S)$ production calculated at $\sqrt s = 1.8$, $7$, $8$ and $13$ TeV in the different rapidity regions. Notation of all histograms is the same as in Fig. 3. The experimental data are from CDF[@41] and LHCb[@26; @27].[]{data-label="fig6"}](pt-LHCb-II.png "fig:"){width="7.0cm"} ![Transverse momentum distribution of inclusive $\Upsilon(3S)$ production calculated at $\sqrt s = 1.8$, $7$, $8$ and $13$ TeV in the different rapidity regions. Notation of all histograms is the same as in Fig. 3. The experimental data are from CDF[@41] and LHCb[@26; @27].[]{data-label="fig6"}](pt-LHCb-III.png "fig:"){width="7.0cm"} ![Transverse momentum distribution of inclusive $\Upsilon(3S)$ production calculated at $\sqrt s = 1.8$, $7$, $8$ and $13$ TeV in the different rapidity regions. Notation of all histograms is the same as in Fig. 3. The experimental data are from CDF[@41] and LHCb[@26; @27].[]{data-label="fig6"}](pt-LHCb-IV.png "fig:"){width="7.0cm"} ![Transverse momentum distribution of inclusive $\Upsilon(3S)$ production calculated at $\sqrt s = 1.8$, $7$, $8$ and $13$ TeV in the different rapidity regions. Notation of all histograms is the same as in Fig. 3. The experimental data are from CDF[@41] and LHCb[@26; @27].[]{data-label="fig6"}](pt-LHCb-V.png "fig:"){width="7.0cm"} ![Transverse momentum distribution of inclusive $\Upsilon(3S)$ production calculated at $\sqrt s = 1.8$, $7$, $8$ and $13$ TeV in the different rapidity regions. Notation of all histograms is the same as in Fig. 3. The experimental data are from CDF[@41] and LHCb[@26; @27].[]{data-label="fig6"}](pt-CDF.png "fig:"){width="7.0cm"}
![The polarization parameters $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta\phi}$ and $\tilde\lambda$ of $\Upsilon(3S)$ mesons calculated in the CS frame as function of its transverse momentum at $\sqrt{s} = 7$ TeV. The A0 gluon density is used. The blue and red histograms correspond to the predictions obtained at $|y|<0.6$ and $0.6<|y|<1.2$, respectively. The experimental data are from CMS[@28].[]{data-label="fig7"}](lt-cs-CMS "fig:"){width="7.0cm"} ![The polarization parameters $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta\phi}$ and $\tilde\lambda$ of $\Upsilon(3S)$ mesons calculated in the CS frame as function of its transverse momentum at $\sqrt{s} = 7$ TeV. The A0 gluon density is used. The blue and red histograms correspond to the predictions obtained at $|y|<0.6$ and $0.6<|y|<1.2$, respectively. The experimental data are from CMS[@28].[]{data-label="fig7"}](lf-cs-CMS "fig:"){width="7.0cm"} ![The polarization parameters $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta\phi}$ and $\tilde\lambda$ of $\Upsilon(3S)$ mesons calculated in the CS frame as function of its transverse momentum at $\sqrt{s} = 7$ TeV. The A0 gluon density is used. The blue and red histograms correspond to the predictions obtained at $|y|<0.6$ and $0.6<|y|<1.2$, respectively. The experimental data are from CMS[@28].[]{data-label="fig7"}](ltf-cs-CMS "fig:"){width="7.0cm"} ![The polarization parameters $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta\phi}$ and $\tilde\lambda$ of $\Upsilon(3S)$ mesons calculated in the CS frame as function of its transverse momentum at $\sqrt{s} = 7$ TeV. The A0 gluon density is used. The blue and red histograms correspond to the predictions obtained at $|y|<0.6$ and $0.6<|y|<1.2$, respectively. The experimental data are from CMS[@28].[]{data-label="fig7"}](lti-cs-CMS "fig:"){width="7.0cm"}
![The polarization parameters $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta\phi}$ and $\tilde\lambda$ of $\Upsilon(3S)$ mesons calculated in the helicity frame as function of its transverse momentum at $\sqrt{s} = 7$ TeV. Notation of all histograms is the same as in Fig. 7. The experimental data are from CMS[@28].[]{data-label="fig8"}](lt-hl-CMS "fig:"){width="7.0cm"} ![The polarization parameters $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta\phi}$ and $\tilde\lambda$ of $\Upsilon(3S)$ mesons calculated in the helicity frame as function of its transverse momentum at $\sqrt{s} = 7$ TeV. Notation of all histograms is the same as in Fig. 7. The experimental data are from CMS[@28].[]{data-label="fig8"}](lf-hl-CMS "fig:"){width="7.0cm"} ![The polarization parameters $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta\phi}$ and $\tilde\lambda$ of $\Upsilon(3S)$ mesons calculated in the helicity frame as function of its transverse momentum at $\sqrt{s} = 7$ TeV. Notation of all histograms is the same as in Fig. 7. The experimental data are from CMS[@28].[]{data-label="fig8"}](ltf-hl-CMS "fig:"){width="7.0cm"} ![The polarization parameters $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta\phi}$ and $\tilde\lambda$ of $\Upsilon(3S)$ mesons calculated in the helicity frame as function of its transverse momentum at $\sqrt{s} = 7$ TeV. Notation of all histograms is the same as in Fig. 7. The experimental data are from CMS[@28].[]{data-label="fig8"}](lti-hl-CMS "fig:"){width="7.0cm"}
![The polarization parameters $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta\phi}$ and $\tilde\lambda$ of $\Upsilon(3S)$ mesons calculated in the perpendicular helicity frame as function of its transverse momentum at $\sqrt{s} = 7$ TeV. Notation of all histograms is the same as in Fig. 7. The experimental data are from CMS[@28].[]{data-label="fig9"}](lt-pl-CMS "fig:"){width="7.0cm"} ![The polarization parameters $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta\phi}$ and $\tilde\lambda$ of $\Upsilon(3S)$ mesons calculated in the perpendicular helicity frame as function of its transverse momentum at $\sqrt{s} = 7$ TeV. Notation of all histograms is the same as in Fig. 7. The experimental data are from CMS[@28].[]{data-label="fig9"}](lf-pl-CMS "fig:"){width="7.0cm"} ![The polarization parameters $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta\phi}$ and $\tilde\lambda$ of $\Upsilon(3S)$ mesons calculated in the perpendicular helicity frame as function of its transverse momentum at $\sqrt{s} = 7$ TeV. Notation of all histograms is the same as in Fig. 7. The experimental data are from CMS[@28].[]{data-label="fig9"}](ltf-pl-CMS "fig:"){width="7.0cm"} ![The polarization parameters $\lambda_\theta$, $\lambda_\phi$, $\lambda_{\theta\phi}$ and $\tilde\lambda$ of $\Upsilon(3S)$ mesons calculated in the perpendicular helicity frame as function of its transverse momentum at $\sqrt{s} = 7$ TeV. Notation of all histograms is the same as in Fig. 7. The experimental data are from CMS[@28].[]{data-label="fig9"}](lti-pl-CMS "fig:"){width="7.0cm"}
![The polarization parameters $\lambda_\theta$ and $\tilde\lambda$ of $\Upsilon(3S)$ mesons calculated in the helicity frame as function of its transverse momentum at $\sqrt{s} = 1.96$ TeV. Notation of all histograms is the same as in Fig. 7. The experimental data are from CDF[@42].[]{data-label="fig10"}](lt-hl-CDF "fig:"){width="7.0cm"} ![The polarization parameters $\lambda_\theta$ and $\tilde\lambda$ of $\Upsilon(3S)$ mesons calculated in the helicity frame as function of its transverse momentum at $\sqrt{s} = 1.96$ TeV. Notation of all histograms is the same as in Fig. 7. The experimental data are from CDF[@42].[]{data-label="fig10"}](lti-hl-CDF "fig:"){width="7.0cm"}
[^1]: The impact of the $\eta_c$ data on charmonia production and polarization was investigated[@12].
[^2]: For different aspects of using the $k_T$-factorization approach the reader may consult the review[@40].
[^3]: Our results for perturbative production amplitudes squared and summed over polarization states agree with ones[@49].
[^4]: The dependence of the numerical results on the emitted energy $E$ is discussed in Section 3.
[^5]: A comprehensive collection of the TMD gluon distributions can be found in the <span style="font-variant:small-caps;">tmdlib</span> package[@56], which is a C++ library providing a framework and an interface to the different parametrizations.
[^6]: By this reason, we have also excluded from the fit the earlier CDF data[@41], which mostly refer to the low $p_T$ region.
[^7]: Such calculations were done[@60].
[^8]: The LHCb Collaboration has also measured $\Upsilon(3S)$ polarization[@29]. However, these data were obtained at rather low transverse momenta and, therefore, we will not analyze them here.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We construct the quantum group $GL_q(2)$ as the semi-infinite cohomology of the tensor product of two braided vertex operator algebras based on the algebra $W_2$ with complementary central charges $c+\bar{c}=28$. The conformal field theory version of the Laplace operator on the quantum group is also obtained.'
author:
- |
Igor B. Frenkel[^1] Anton M. Zeitlin[^2]\
Department of Mathematics,\
Yale University,\
442 Dunham Lab, 10 Hillhouse Avenue,\
New Haven, CT 06511
title: |
**[Quantum Group $GL_q(2)$ and Quantum Laplace Operator via Semi-infinite Cohomology\
]{}**
---
Introduction
============
We have shown in [@fz] that the quantum group $SL_q(2)$ admits a realization as the semi-infinite cohomology of the Virasoro algebra with coefficients in the tensor product of two braided vertex operator algebras (VOA) with complementary central charges $c+\bar{c}=26$. At the end of the paper we have discussed extensions of our results to higher ranks and more general $W$-algebras (see e.g. [@FFr] and references therein). In the present paper we consider a simplest generalization of this type, namely the algebra $W_2$, which is the semi-direct sum of the Virasoro and the Heisenberg algebra. We prove that the semi-infinite cohomology of $W_2$ in the tensor product of two appropriate braided vertex operator algebras yields the quantum group $GL_q(2)$. The central charges of the two braided VOAs add up to the critical value for $W_2$, namely $c+\bar{c}=28$.
It is well known that one can develop the calculus of differential forms on $GL_q(2)$ and define the $q$-analogues of partial derivatives and the quantum version of Laplace operator. Since $GL_q(2)$ is realized as semi-infinite cohomology of $W_2$, one can look for a “lift” of various structures on the quantum group to the corresponding braided VOA. In this paper we do find a simple operator on the braided VOA that commutes with the action of $W_2$ and induces the quantum Laplace operator on the cohomology. This suggests that other structures of noncommutative geometry might admit equally simple and natural realization in terms of chiral two-dimensional conformal field theory.
The structure of the paper is as follows. In Section 2 we recall the basic facts about representation theory of $U_q(gl(2))$. In Section 3 we discuss $W_2$ and related semi-infinite cohomology complex. In Section 4 we construct the intertwining operators for $W_2$ and related braided VOA (see [@styrkas], [@fz]). In Section 5 we find realization of $GL_q(2)$ as the semi-infinite cohomology of certain braided VOA and write a simple formula for the quantum Laplace operator using a Fock space realization of $W_2$ modules.
$\mathbf{U_q(gl(2))}$, its representations and intertwining operators
=====================================================================
Let $U_q(gl(2))$ be the Hopf algebra over $\mathbb{C}(q)$ with generators $E, F, q^{\pm H}, I$ and commutation relations: $$\begin{aligned}
q^HE&=&q^2Eq^H,\nonumber\\
q^HF&=&q^{-2}Fq^H,\nonumber \\
{[}E,F] &=&\frac{q^H-q^{-H}}{q-q^{-1}},\nonumber\end{aligned}$$ and $I$ is a central element. The comultiplication is given by $$\begin{aligned}
\Delta(I)&=&I\otimes 1+1\otimes I,\nonumber\\
\Delta(q^H)&=&q^{H}\otimes q^{H}\nonumber\\
\Delta(E)&=&E\otimes q^H+1\otimes E,\nonumber\\
\Delta(F)&=&F\otimes 1 + q^{-H}\otimes F.\end{aligned}$$ The universal R-matrix for $U_q(gl(2))$, which is an element of a certain completion of $U_q(gl(2))\otimes U_q(gl(2))$, is given by: $$\begin{aligned}
R&=&C\Theta, \qquad C=q^{{I\otimes I}}q^{\frac{H\otimes H}{2}},\nonumber\\
\Theta&=&\sum_{k\geqslant 0}q{^{k(k-1)/2}\frac{(q-q^{-1})^k}{[k]!}E^k\otimes F^k},\end{aligned}$$ where $[n]=\frac{q^n-q^{-n}}{q-q^{-1}}$ and $[n]!=[1][2]\ldots[n]$.
For any given pair $V,W$ of representations, R-matrix gives the following commutativily isomorphism: $\check{R}=PR: V\otimes W \rightarrow W\otimes V$, where $P$ is a permutation: $P(v\otimes w)=w\otimes v$.
Let $\lambda\in \mathbb{Z}_+$ and $k\in \mathbb{Z}$. We denote by $V_{\lambda,k}$ the finite dimensional irreducible representation of $U_q(gl(2))$, such that $V_{\lambda,k}$ is the highest weight representation with the highest weight $\lambda$ for the subalgebra $U_q(sl(2))$ of $U_q(gl(2))$, and $k$ is the eigenvalue of central element $I$.
One can construct the intertwining operators for finite dimensional representations, i.e. elements of $Hom (V_{\lambda , l}\otimes V_{\mu ,m}, V_{\nu ,n})$ and $Hom (V_{\nu ,n}, V_{\lambda , l}\otimes V_{\mu ,m})$. The following Proposition holds:
Let $\lambda, \mu, \nu\in\mathbb{Z}_+$ and $l,m,n\in \mathbb{Z}$. Then $dim Hom (V_{\lambda, l}\otimes V_{\mu,m}, V_{\nu,n})=1$ iff $\lambda+\mu\ge\nu\ge|\lambda-\mu|$ and $n=m+l$. Otherwise $dim Hom (V_{\lambda, l}\otimes V_{\mu,m}, V_{\nu,n})=0$.
[**Proof.**]{} Proof follows from a similar fact from $U_q(sl(2))$ representation theory.
$\blacksquare$
The same statement holds for intertwiners from $Hom (V_{\nu ,n}, V_{\lambda , l}\otimes V_{\mu ,m})$. The next proposition gives quadratic relations between intertwiners and will be very crucial in the following.
Let $\lambda_i\in \mathbb{Z}_+$, $l_i\in \mathbb{Z}$ $(i=0,1,2,3)$ . Then there exists an invertible operator $$B\left[\begin{array}{cc}
\lambda_0, l_0 ; \lambda_1, l_1\\
\lambda_2, l_2 ; \lambda_3,l_3
\end{array}
\right]$$ such that the following diagram is commutative: $$\xymatrixcolsep{50pt}
\xymatrixrowsep{10pt}
\xymatrix{
{\begin{array}{l}
\oplus_\rho \big(Hom(V_{\rho, r},V_{\lambda_1, l_1}\otimes V_{\lambda_2,l_2})\\
\otimes Hom(V_{\lambda_0, l_0}, V_{\rho, r}\otimes V_{\lambda_3, l_3})\big)
\end{array}
\ar[d]_{i}}
\ar[r]^{
B\left[\begin{array}{cc}
\lambda_0,l_0; \lambda_1,l_1\\
\lambda_2,l_2 ;\lambda_3,l_3
\end{array}
\right]
}&
{\begin{array}{r}
\oplus_{\xi, k} \big(Hom(V_{\xi,k},V_{\lambda_1, l_1}\otimes V_{\lambda_3, l_3})\\
\otimes Hom(V_{\lambda_0, l_0}, V_{\xi, k}\otimes V_{\lambda_2, l_2})\big)
\end{array}
\ar[d]_{i}
}
\\
Hom(V_{\lambda_0, l_0}, V_{\lambda_1, l_1}\otimes V_{\lambda_2, l_2}\otimes V_{\lambda_3, l_3})
\ar[r]^{PR}&
Hom(V_{\lambda_0, l_0}, V_{\lambda_1, l_1}\otimes V_{\lambda_3, l_3}\otimes V_{\lambda_2, l_2}),
}$$
where $|\lambda_1+\lambda_2|\ge \rho\ge |\lambda_1-\lambda_2|$, $|\lambda_3+\rho|\ge \lambda_0\ge |\lambda_3-\rho|$, $|\lambda_1+\lambda_3|\ge \xi\ge |\lambda_1-\lambda_3|$, $|\lambda_2+\xi|\ge \lambda_0\ge |\lambda_2-\xi|$, and $i$ is an isomorphism.
This fact follows from a similar fact from the representation theory of $U_q(sl(2))$. Moreover, when the braiding matrix $$\begin{aligned}
B\left[\begin{array}{cc}
\lambda_0,l_0; \lambda_1,l_1\\
\lambda_2,l_2 ;\lambda_3,l_3
\end{array}
\right]\end{aligned}$$ is not equal to zero, it is equal to $$\begin{aligned}
q^{l_2l_3}B^V\left[\begin{array}{cc}
\lambda_0; \lambda_1\\
\lambda_2 ;\lambda_3
\end{array}
\right],\end{aligned}$$ where $B^V$ is the braiding matrix of $U_q(sl(2))$. Let’s now denote by $\phi^{\nu,n}_{\lambda,l;\mu,m}$ the generating element of $Hom (V_{\lambda , l}\otimes V_{\mu ,m}, V_{\nu ,n})$ and by $\phi_{\nu,n}^{\lambda,l;\mu,m}$ the generating element of $Hom (V_{\nu ,n}, V_{\lambda , l}\otimes V_{\mu ,m})$. Then Proposition 2.2. gives the following quadratic relations between intertwining operators: $$\begin{aligned}
\label{intfd}
&&(1\otimes PR)\phi_\rho^{\lambda_1,l_1;\lambda_2,l_2}\phi_{\lambda_0,l_0}
^{\rho,r;\lambda_3,l_3}=\nonumber\\
&&\sum_{\xi}B_{\rho,r;\xi,k}\left[\begin{array}{cc}
\lambda_0,l_0 & \lambda_1,l_1\\
\lambda_2,l_2 & \lambda_3,l_3
\end{array}
\right]
\phi_{\xi,k}^{\lambda_1,l_1;\lambda_3,l_3}\phi_{\lambda_0,l_0}^{\xi,k;\lambda_2,l_2},\\
&&\label{intfdnew}
\phi_{\rho,r;\lambda_3,l_3}^{\lambda_0,l_0}\phi_{\lambda_1,l_1\lambda_2,l_2}^\rho(1\otimes PR)=\nonumber\\
&&\sum_{\xi,k}B_{\xi,k;\rho,r}\left[\begin{array}{cc}
\lambda_0,l_0 & \lambda_1,l_1\\
\lambda_2,l_2 & \lambda_3,l_3
\end{array}
\right]
\phi_{\xi,k;\lambda_2,l_2}^{\lambda_0,l_0}\phi_{\lambda_1l_1,\lambda_3,l_3}^{\xi,k}.\end{aligned}$$
Virasoro, $W_2$ algebras and semi-infinite cohomology: basic facts
==================================================================
[**3.1. Virasoro algebra and Feigin-Fuks representation.**]{} The Virasoro algebra $$[L_n,L_m]=(n-m)L_{m+n}+\frac{\hat c}{12}(n^3-n)\delta_{n,-m}$$ has been extensively studied for many years. Here we need only basic facts. Let us denote by $M_{\hat c,h}$ and $V_{\hat c,h}$ the Verma module and irreducible module (with highest weight $h$), correspondingly. Throughout the paper we will consider only generic values of $\hat c$. This means that the central charge $\hat{c}$ is parametrized in the following way: $$\hat c=13-6(\varkappa+\frac{1}{\varkappa}),$$ where parameter $\varkappa\in \mathbb{R}\backslash \mathbb{Q}$. Then we have the following proposition (see e.g. [@fb] and references therein).
For generic value of $\hat c$, Verma module $M_{\hat c,h}$ has a unique singular vector in the case if $h=h_{m,n}$, where $$\label{virsing}
h_{m,n}=\frac{1}{4}(m^2-1)\varkappa+\frac{1}{4}(n^2-1)\varkappa^{-1}-\frac{1}{2}(mn-1).$$ This singular vector occurs on the level mn, i.e. the value of $L_0$ is $h_{m,n}+mn$.
In the following we will be interested in the modules with $h=h_{1,n}=\Delta(\lambda)$, where $\lambda=n-1$, $\Delta(\lambda)=-\frac{\lambda}{2}+\frac{\lambda(\lambda+2)}{4\varkappa}$.
Let $\hat c$ be generic and $\lambda \ge 0$, then $V_{\Delta(\lambda),\hat c}=M_{\Delta(\lambda),\hat c}/M_{\Delta(\lambda)+\lambda+1,\hat c}$, where $V_{\Delta(\lambda),\hat c}$ is the irreducible Virasoro module with the highest weight $\Delta(\lambda)$. For $\lambda < 0$ and generic values of $c$ the irreducible module is isomorphic to Verma one, namely, $V_{\Delta(\lambda),\hat c}=M_{\Delta(\lambda),\hat c}$.
Let us consider the Heisenberg algebra $$[\alpha_n,\alpha_m]=2\varkappa m\delta_{n+m,0}$$ and denote by $F_{\lambda,\varkappa}$ the Fock module associated to this algebra. Namely, $F_{\lambda,\varkappa}=S(\alpha_{-1},\alpha_{-2},\dots)\otimes\mathbf{1}_\lambda$, such that $a_n\mathbf{1}_\lambda=0$ if $n>0$ and $a_0\mathbf{1}_\lambda=\lambda\mathbf{1}_\lambda (\lambda\in\mathbb{C})$. It is well known (see e.g. [@fb], [@FHL] and references therein) that $F_{0,\varkappa}$ gives rise to the vertex operator algebra, generated by the field $\alpha(z)=\sum a_nz^{-n-1}$, such that $\deg(a(z))=1$ which has the following operator product expansion (OPE): $$\alpha(z)\alpha(w)\sim \frac{2\varkappa}{(z-w)^2}.$$ We will denote this vertex algebra as $F_{0,\varkappa}(\alpha)$. The following Proposition holds.
\[feiginfuks\] Vertex algebra $F_{0,\varkappa}(a)$ has a vertex operator algebra structure, where the vertex operator, corresponding to the Virasoro element, is given by the following formula: $$L(z)=\frac{1}{4\varkappa}:\alpha(z)^2:+\frac{\varkappa-1}{2\varkappa}{\partial}\alpha(z),$$ such that $L(z)=\sum_nL_nz^{-n-2}$ and $L_n$ satisfy Virasoro algebra relations with the central charge $\hat c=13-6(\varkappa+\frac{1}{\varkappa})$.
[**3.2. $W_2$ as an extension of Virasoro algebra.**]{} $W_2$ is the semi-direct sum of Virasoro and Heisenberg algebras. The commutation relations are: $$\begin{aligned}
\label{w2}
&&[{\cal L}_n,{\cal L}_m]=(n-m){{\cal L}}_{m+n}+\frac{c}{12}(n^3-n)\delta_{n,-m},\nonumber\\
&&[a_n,a_m]=2\eta m\delta_{n+m,0},\nonumber\\
&&[{{\cal L}}_n,a_m]=-ma_{n+m}.\end{aligned}$$ The pair of central charges $(c,\eta)$ determine the algebra. Throughout this paper we will use the following parameterization of the central charge: $$\begin{aligned}
c=14-6(\varkappa+\frac{1}{\varkappa}).
\end{aligned}$$ Moreover, we will require, that $\varkappa\in \mathbb{R}\backslash \mathbb{Q}$. However, for any representation of $W_2$ algebra one can find a $direct$ sum of Virasoro and Heisenberg algebras acting in the same representation. Let us define generators $$\begin{aligned}
L_n\equiv{{\cal L}}_n-\frac{1}{2\eta}\sum^{\infty}_{m=-\infty} :a_{n-m}a_m:,\end{aligned}$$ where symbol $::$ stands for standard Fock space normal ordering. Then the following Proposition holds.
Let ${{\cal L}}_n, a_m$ generate algebra $W_2$ with central charges $(c,\eta)$, then $L_n$, $a_n$ satisfy the following commutation relations: $$\begin{aligned}
&&[L_n,L_m]=(n-m)L_{m+n}+\frac{\hat c}{12}(n^3-n)\delta_{n,-m},\nonumber\\
&&[a_n,a_m]=2\eta m\delta_{n+m,0},\nonumber\\
&&[L_n,a_m]=0,\end{aligned}$$ where $\hat{c}=c-1$.
We will be interested in the highest weight modules for $W_2$ algebra $$\begin{aligned}
W_{\Delta(\lambda),k}^{\varkappa,\eta}=V_{\Delta(\lambda),\varkappa}\otimes F_{k,\eta}\end{aligned}$$ which are the tensor products of the irreducible highest weight representation of Virasoro algebra $V_{\Delta(\lambda),\varkappa}$, generated by $L_n$ and Fock module $F_{k,\eta}$ for the Heisenberg algebra, generated by $a_n$.\
[**3.3. Ghost VOAs and semi-infinite cohomology for $W_2$.**]{} In this section we will show how to reduce the semi-infinite cohomology of $W_2$ to the semi-infinite cohomology of the Virasoro algebra [@fgz].
In the case of $W_2$, semi-infinite forms can be realized by means of the following Heisenberg superalgebras: $$\begin{aligned}
&&\{\psi_n,\chi_m\}=\delta_{n+m,0}, \nonumber\\
&&\{b_n, c_m\}=\delta_{n+m,0}, \quad n,m \in \mathbb{Z}.\end{aligned}$$ One can construct Fock modules $\Lambda$, $\Lambda'$ in the following way: $$\begin{aligned}
\Lambda&=&\mathbb{C}\{b_{-n_1}\dots b_{-n_k}c_{-m_1}\dots c_{-m_\ell} \mathbf{1};\nonumber\\
&& c_k \mathbf{1}=0,\ k\geqslant 2;\quad b_k\mathbf{1}=0, \ k\geqslant -1\}\nonumber\\
\Lambda'&=&\mathbb{C}\{\chi_{-n_1}\dots \chi_{-n_k}\psi_{-m_1}\dots \psi_{-m_\ell} \mathbf{1};\nonumber\\
&& \psi_k \mathbf{1}=0,\ k\geqslant 1;\quad \chi_k\mathbf{1}=0, \ k\geqslant 0\}\end{aligned}$$ Let’s denote $M=\Lambda\otimes \Lambda'$. Each of $\Lambda, \Lambda'$ and therefore $M$ has a VOA structure on it, namely, one can define four quantum fields: $$\begin{aligned}
&&b(z)=\sum_m b_mz^{-m-2}, \qquad c(z)=\sum_n c_nz^{-n+1},\nonumber\\
&&\psi(z)=\sum_m \psi_m z^{-m},\qquad \chi(z)=\sum_m b_mz^{-m-1}\end{aligned}$$ which according to the commutation relations between modes have the following operator products: $$b(z)c(w)\sim\frac{1}{z-w},\quad \chi(z)\psi(w)\sim\frac{1}{z-w},$$ such that all other operator products do not contain singular terms. The Virasoro element is given by the following expression: $$L^{M}(z)=2:\partial b(z)c(z):+:b(z)\partial c(z):+:\partial \psi(z)\chi(z):,$$ such that $b(z)$, $c(z)$ have conformal weights $2$,$-1$, and $\psi(z)$, $\chi(z)$ have conformal weights $0$,$1$ correspondingly. The central charge of the corresponding Virasoro algebra is equal to -28. One can define the following operator: $$N_g(z)=:c(z)b(z):+:\psi(z)\chi(z):,$$ which is known as ghost number current.
One can show that the module $W_{\Delta(0),0}^{\varkappa,\eta}$ has the structure of the VOA generate by the quantum fields ${{\cal L}}(z)=\sum_n{{\cal L}}_n z^{-n-2}$, and $a(z)=\sum_n a_nz^{-n-1}$, which have the following operator products (which are equivalent to commutation relations (\[w2\])): $$\begin{aligned}
&&{{\cal L}}(z){{\cal L}}(w)\sim \frac{c}{2(z-w)^4}+\frac{2{{\cal L}}(w)}{(z-w)^2}+\frac{\partial {{\cal L}}(w)}{z-w},\nonumber\\
&&{{\cal L}}(z)a(w)\sim \frac{a(w)}{(z-w)^2}+\frac{\partial a(w)}{z-w},\nonumber\\
&& a(z)a(w)\sim \frac{2\eta}{(z-w)^2}.\end{aligned}$$ Let the space $\mathcal{W}$ be such that $\eta=0, c=28$. Then the following Proposition is true.
The operator of ghost number 1 $$\begin{aligned}
&&Q=\int \frac{{\mathrm{d}}z}{2\pi i}J(z), \\
&& J(z)=:c(z){{\cal L}}(z):+:c{\partial}cb(z):
\frac{3}{2}\partial^2c(z)+
\psi(z)a(z)+:c\partial\psi\chi(z):\nonumber\end{aligned}$$ is nilpotent: $Q^2=0$ on $\mathcal{W}\otimes M$.
The space $\mathcal{W}\otimes M$ is known as a semi-infinite cohomology complex, where the differential is $Q$, which is sometimes called the BRST operator. The grading in the complex is given by ghost number operator $N_g$. The $k$-th cohomology group is usually denoted as $H^{\frac{\infty}{2}+k}(W_2,\mathbb{C}{\bf{c}}, \mathcal{W})$.
In this article we will be interested in computing semi-infinite cohomology for $W_2$-modules which have the following form: ${\cal W}=W\otimes
\bar {W}$. Here each of $W,\bar{W}$ are representations of $W_2$, with central charges $(c,\eta)$ $(\bar{c},\bar{\eta})$ correspondingly, such that the following relation is satisfied: $c+\bar c=28$ and $\eta+\bar \eta=0$. Let us denote the quantum fields for Virasoro and Heisenberg algebras in $W$ and $\bar {W}$ as $\mathcal{L}(z), a(z)$ and $\bar {\mathcal{L}}(z), {\bar a}(z)$ correspondingly. Then the BRST operator on $\cal W$ has the following form: $$\begin{aligned}
&&Q_{\cal W}=\int \frac{{\mathrm{d}}z}{2\pi i}J_{\cal W}(z), \\
&& J_{\cal W}(z)=:c(z)({{\cal L}}(z)+\bar{{{\cal L}}}(z)):+:c{\partial}cb(z):+
\frac{3}{2}\partial^2c(z)+\nonumber\\
&&\psi(z)a^{+}(z)+:c\partial\psi\chi(z):\nonumber\end{aligned}$$ where $a^{+}(z)=a(z)+\bar{a}(z)$. It makes sense to define also the field $a^{-}(z)=a(z)-\bar{a}(z)$, which is crucial in the following statement.
The operator $Q_{\cal W}$ can be rewritten in the following form: $$\begin{aligned}
&&Q_{\cal W}=\int \frac{{\mathrm{d}}z}{2\pi i}J^{\oplus}_{\cal W}(z), \\
&& J^{\oplus}_{\cal W}(z)=:\tilde c(z)(L(z)+\bar{L}(z)):+:\tilde c{\partial}\tilde c\tilde b(z):+
\frac{3}{2}\partial^2\tilde c(z)+
\tilde \psi(z)\tilde a^{+}(z),\nonumber\end{aligned}$$ Here $L(z)={{\cal L}}(z)-\frac{1}{4\eta}:a(z)^2:$, $\bar L(z)={{\cal L}}(z)-\frac{1}{4\eta}:\bar a(z)^2:$ and $\tilde A(z)=e^{R}A(z)e^{-R}$ (A is any quantum field), where $R=\frac{1}{8\eta\pi i }\int dz c(z) a^-(z)\chi(z)$.
[**Proof.**]{} Let us write explicitly the action of the transformation $\tilde A(z)=e^{R}A(z)e^{-R}$ for every quantum field: $$\begin{aligned}
&&\tilde a^{+}(z)=a^{+}(z)+{\partial}(c(z)\chi(z))(z), \quad \tilde b(z)=b(z)-\frac{1}{4\eta}a^-(z)\chi(z),\nonumber\\
&&\tilde \psi(z)=\psi(z) +\frac{1}{4\eta}c(z) a^-(z), \quad \tilde a^{-}=a^{-}, \quad
\tilde c=c, \quad \tilde \chi=\chi.\end{aligned}$$ To prove this Proposition we just expand the tilded quantum fields in terms of usual ones: $$\begin{aligned}
&&J^{\oplus}_{\cal W}(z)=: c(z)(L(z)+\bar{L}(z)):+:c{\partial}c\tilde b(z):+
\frac{3}{2}\partial^2 c(z)+
\tilde \psi(z)\tilde a^{+}(z)=\nonumber\\
&&: c(z)(L(z)+\bar{L}(z)):+:c{\partial}c(b(z)-\frac{1}{4\eta}a^-(z))\chi(z):+\frac{3}{2}\partial^2 c(z)+\nonumber\\
&&:(\psi(z)+\frac{1}{4\eta}c(z) a^-(z))(a^{+}(z)+{\partial}(c(z)\chi(z))(z)):=\nonumber\\
&&: c(z)(L(z)+\bar{L}(z)+\frac{1}{4\eta}(a^2(z)-\bar a^2(z))
):+:c{\partial}c\tilde b(z):+
\frac{3}{2}\partial^2 c(z)+\nonumber\\
&&\tilde \psi(z)\tilde a^{+}(z)+:\psi(z){\partial}(c(z)\chi(z):\end{aligned}$$
$\blacksquare$
In other words, by means of similarity transformation, one can transform the BRST operator associated with semidirect sum of Virasoro and Heisenberg algebras to the one associated with the direct sum of those algebras. This Proposition has the following Corollary, which will be crucial for computing the semi-infinite cohomology.
All nontrival cycles of the semi-infinite cohomology belong to the kernels of the following operators: $(i) L_0+\bar L_0+\tilde L^M_0$, $(ii) a_0+\bar{a}_0$, $(iii) T_0$, where $T(z)=\frac{1}{4\eta}:\tilde a^{+}(z) \tilde a^{-}(z):+:{\partial}\tilde \psi(z) \tilde\chi(z)$.
[**Proof.**]{} Let us prove $(i)$ first. We know that $[Q_{\cal W},\tilde b_0]=L_0+\bar L_0+\tilde L^M_0$, therefore, $$\begin{aligned}
[Q_{\cal W},\tilde b_0]\Phi=Q_{\cal W}\tilde b_0\Phi=\Delta \Phi.\end{aligned}$$ Hence, for $\Delta\neq 0$, $\Psi=\Delta^{-1}b_0\Phi$. The proofs of (ii) and (iii) are identical to (i), one just needs to use the conditions that $[Q_{\cal W},\tilde \chi_0]=a_0+\bar{a}_0$ and $[Q_{\cal W},S_0]=T_0$, where $S_0$ is the zero mode of an operator $S(z)=\frac{1}{4 \eta}a^-(z)\chi(z)$. $\blacksquare$
Intertwiners and braided VOAs
==============================
In this section we will study braided VOAs. This will provide a necessary framework for the next section, where we will study semi-infinite cohomology and the Lian-Zuckerman multiplication.\
[**4.1. Definition of a braided VOA and the simplest example.**]{} First, we give a definition of braided VOAs.\
[**Definition.**]{}[@fz]
*Let $\mathbb{V}=\oplus_{\lambda\in I}\mathbb{V}_{\lambda}$ be a direct sum of graded complex vector spaces, called sectors: $\mathbb{V}_{\lambda}=\oplus_{n\in \mathbb{Z}_+}\mathbb{V}_{\lambda}[n]$, indexed by some set $I$. Let $\Delta_{\lambda}$, $\lambda\in I$ be complex numbers, which we will call conformal weights of the corresponding sectors. We say that $\mathbb{V}$ is a braided vertex algebra, if there are distinguished elements $0\in I$ such that $\Delta_0=0$, $\mathbf{1}\in \mathbb{V}_0[0]$, linear maps $D:\mathbb{V}\to\mathbb{V}$, $\mathcal{R}:\mathbb{V}\otimes\mathbb{V}\to \mathbb{V}\otimes\mathbb{V}$ and the linear correspondence $$\begin{aligned}
\mathbb{Y}(\cdot,z)\cdot:\mathbb{V}\otimes\mathbb{V}\to \mathbb{V}\{z\},\quad \mathbb{Y}=\sum_{\lambda,\lambda_1,\lambda_2}\mathbb{Y}^{\lambda_1\lambda_2}_{\lambda}(z),\end{aligned}$$ where $$\begin{aligned}
\mathbb{Y}^{\lambda_1\lambda_2}_{\lambda}(z)\in Hom(\mathbb{V}_{\lambda_1}\otimes\mathbb{V}_{\lambda_2},\mathbb{V}_{\lambda})\otimes z^{\Delta_{\lambda}-\Delta_{\lambda_1}-\Delta_{\lambda_2}}
\mathbb{C}[[z,z^{-1}]],\end{aligned}$$ such that the following properties are satisfied:\
i)Vacuum property: $\mathbb{Y}(\mathbf{1},z)v=v$, $\mathbb{Y}(v,z)\mathbf{1}|_{z=0}=v$.\
ii) Complex analyticity: for any $v_i\in \mathbb{V}_{\lambda_i}$, $(i=1,2,3,4)$ the matrix elements $\langle v_4^*, \mathbb{Y}(v_3,z_2)\mathbb{Y}(v_2,z_1)v_1 \rangle$ regarded as formal Laurent series in $z_1,z_2$, converge in the domain $|z_2|> |z_1|$ to a complex analytic function $r(z_1,z_2)\in z_1^{h_1}z_2^{h_2}
(z_1-z_2)^{h_3} \mathbb{C}[z_1^{\pm 1},z_2^{\pm 1}, (z_1-z_2)^{-1}]$, where $h_1, h_2,h_3\in \mathbb{C}$.\
iii) Derivation property: $\mathbb{Y}(Dv,z)\mathbf{1}=\frac{d}{dz}\mathbb{Y}(v,z).$\
iv) Braided commutativity (understood in the weak sense [^3]): $$\begin{aligned}
\mathscr{A}_{z,w}(\mathbb{Y}(v,z)\mathbb{Y}(u,w))=\sum_i\mathbb{Y}(u_i,w)\mathbb{Y}(v_i,z),\end{aligned}$$ where $\mathcal{R}(u\otimes v)=\sum_i u_i\otimes v_i$ and $\mathscr{A}_{z,\omega}$ denote the monodromy around the path $$\begin{aligned}
w(t)&=&\frac{1}{2}\big((z+w)+(w-z)e^{\pi it}\big),\nonumber\\
z(t)&=&\frac{1}{2}\big((z+w)+(z-w)e^{\pi it}\big),\end{aligned}$$ as shown on the picture: $$\begin{aligned}
\xymatrix{
z\ \bullet\ \ar@/_1pc/[r] &
\bullet \ \ar@/_1pc/[l] w
}\end{aligned}$$*
v\) There exists an element $\omega\in \mathbb{V}_0$, such that $$Y(\omega,z)=\sum_{n\in\mathbb{Z}}L_nz^{-n-2}$$ and $L_n$ satisfy the relations of Virasoro algebra with $L_{-1}=D$.\
vi) Associativity (understood in the weak sense): $$\begin{aligned}
\mathbb{Y}(\mathbb{Y}(u,z-w)v, w)=\mathbb{Y}(u,z)\mathbb{Y}(v,w).\end{aligned}$$
The associativity condition puts a restriction on the operator $\mathcal{R}$, namely one can show that matrix $R$ satisfies the Yang-Baxter equation (see [@fz], [@styrkas]).
The simplest example of braided VOA (which does not reduce to usual VOA) is a natural extension of the VOA $F_{\eta}(a)$ generated by Heisenberg algebra, considered in the previous section. Let us consider the following space: $$\hat{F}_\eta=\oplus_{\lambda\in\mathbb{Z\eta}}F_{\lambda,\eta}.$$ Below we will show that $\hat{F}_\eta$ carries a structure of braided VOA. The operators $$\label{xz}
\mathbb{X}(\lambda,z)=\mathbf{1}_\lambda z^{\frac{\lambda a_0}{2\eta}}
e^{\big(\frac{\lambda}{2\varkappa}\sum_{n>0}\frac{a_{-n}}{n}z^n\big)}
e^{-\big(\frac{\lambda}{2\varkappa}\sum_{n>0}\frac{a_n}{n}z^{-n}\big)},$$ where $\lambda\in\mathbb{Z}$, generate vacuum vectors in Fock modules. It is clear that $\mathbb{X}(\lambda,z)\mathbf{1}_0|_{z=0}=\mathbf{1}_\lambda$. Denoting $$\mathbb{X}_{n_1,\dots,n_k}(\lambda,z)\equiv :a^{(n_1)}(z)\dots a^{(n_k)}(z)\mathbb{X}(\lambda,z):,$$ where $a^{(n)}(z)=\frac{1}{n!}\big(\frac{{\mathrm{d}}}{{\mathrm{d}}z}\big)^na(z)$, one can see that $$\mathbb{X}_{n_1,\dots,n_k}(\lambda,z)\mathbf{1}_0|_{z=0}=a_{-n_1},\dots,a_{-n_k}\mathbf{1}_\lambda.$$ In such a way, we build the correspondence $$i:v\rightarrow Y(v,z)=\sum_{n\in \mathbb{Z}}v_{(n)}z^{-n-1},$$ such that $v\in\hat{F}_\eta$ and $v_{(n)}\in \mathrm{End}(\hat{F}_\eta)$. To see the connection with definition of braided VOA, we note that in this case the set $I=\mathbb{Z}$, the sectors are Fock spaces $F_{\lambda,\eta}$, $\Delta(\lambda)$ is the conformal weight of the vacuum.
Let $|z|>|w|$, then $$\mathbb{X}(\lambda,z)\mathbb{X}(\mu,w)=(z-w)^{\frac{\lambda\mu\eta}{2}}
(\mathbb{X}(\lambda+\mu,w)+\dots),$$ where dots stand for the terms regular in $(z-w)$, hence $$\begin{aligned}
\label{acon}
\mathscr{A}_{z,w}\big(\mathbb{X}(\lambda,z)\mathbb{X}(\mu,w)\big)=
p^{\frac{\lambda\mu}{2}}\mathbb{X}(\mu,w)\mathbb{X}(\lambda,z),\end{aligned}$$ where $p=e^{\pi i\eta}$. Here, we underline that the expression above should be understood in a weak sense, i.e. the analytical continuation is performed for the matrix elements of the corresponding operator products. Moreover, the matrix elements of operator product expansion $\mathbb{X}(\lambda,z)\mathbb{X}(\mu,w)$ exist in the domain $|z|>|w|$ and the analytical continuation relates it to the matrix elements of operator product expansion $\mathbb{X}(\mu,w)\mathbb{X}(\lambda,z)$, which converge in the domain $|w|>|z|$. The relation (\[acon\]) is a simplified case of associativity condition, since in our case $\mathbb{Y}^{\lambda}_{\lambda_1\lambda_2}$ is nonzero only for $\lambda=\lambda_1+\lambda_2$ and the $\mathcal{R}$-operator is therefore reduced to the multiplication on some power of $p$.
In the remainig part of this section we will discuss examples which are more involved. In order to construct them we need to use additional constructions.\
[**4.3. Intertwiners for Virasoro and $W_2$ algebras.**]{} In [@fz] we have constructed an intertwiner for Virasoro algebra. This is a map $$\label{intvir}
\Phi_{\lambda\mu}^\nu(z):V_{\Delta(\lambda),\varkappa}\otimes V_{\Delta(\mu),\varkappa}\rightarrow
V_{\Delta(\nu),\varkappa}[[z,z^{-1}]]z^{\Delta(\nu)-\Delta(\mu)-\Delta(\lambda)},$$ which has the following property [@ms]: $$\label{propint}
L_n\cdot\Phi_{\lambda\mu}^\nu(z)=\Phi_{\lambda\mu}^\nu(z)\Delta_{z,0}(L_n),$$ where $$\Delta_{z,0}(L_n)=\oint_z\frac{{\mathrm{d}}\xi}{2\pi i}\xi^{n+1}\Big(\sum_m(\xi-z)^{-m-2}L_m\Big)\otimes 1+1\otimes L_n.$$ Moreover, we have found that they satisfy the relation involving the braiding matrix $B^V$ of $U_q(sl(2))$.
[[@fz]]{} Let $z_1,z_2\in \mathbb{C}$ such that $0<|z_1|<|z_2|$, $\lambda_i\ge 0$ $(i=0,1,2,3)$. Then the following relation holds: $$\begin{aligned}
\label{fockint}
&&\mathscr{A}_{z_1,z_2}\big(\Phi_{\lambda_3\rho}^{\lambda_0}(z_2)\Phi_{\lambda_2\lambda_1}^\rho(z_1)\big)(P\otimes 1)=\nonumber\\
&&\sum_{\xi}B^V_{\rho\xi}
\left[\begin{array}{cc}
\lambda_0 &\lambda_1 \\
\lambda_2 &\lambda_3
\end{array}
\right]
\Phi_{\lambda_2\xi}^{\lambda_0}(z_1)\Phi_{\lambda_3\lambda_1}^\xi(z_2), \end{aligned}$$ where $P$ is an interchange operator, namely $P(v_1\otimes v_2)=v_2\otimes v_1$ and $q=e^{\frac{\pi i}{\varkappa}}$.
Now we show that there exists an intertwining operator between the corresponding irreducible highest weight representations of $W_2$ algebra.
i)There exists a map $$\begin{aligned}
\label{intw2}
&&\Phi_{\lambda,l;\mu,m}^{\nu,n}(z):W^{\eta,\kappa}_{\Delta(\lambda),l}\otimes W^{\eta,\kappa}_{\Delta(\mu),m}\rightarrow \nonumber\\
&&W^{\eta,\kappa}_{\Delta(\nu),n}[[z,z^{-1}]]z^{\eta(n^2-m^2-l^2)}z^{\Delta(\nu)-\Delta(\mu)-\Delta(\lambda)},\end{aligned}$$ such that $$\begin{aligned}
&&{{\cal L}}_n\cdot\Phi_{\lambda,l;\mu,m}^{\nu,n}(z)=\Phi_{\lambda,l;\mu,m}^{\nu,n}(z)\Delta_{z,0}({{\cal L}}_n),\nonumber\\
&&a_k\cdot\Phi_{\lambda,l;\mu,m}^{\nu,n}(z)=\Phi_{\lambda,l;\mu,m}^{\nu,n}(z)\Delta_{z,0}(a_k),\end{aligned}$$ where $$\begin{aligned}
&&\Delta_{z,0}({{\cal L}}_n)=\oint_z\frac{{\mathrm{d}}\xi}{2\pi i}\xi^{n+1}\Big(\sum_m(\xi-z)^{-m-2}{{\cal L}}_m\Big)\otimes 1+1\otimes {{\cal L}}_n ,\nonumber\\
&&\Delta_{z,0}(a_k)=\oint_z\frac{{\mathrm{d}}\xi}{2\pi i}\xi^{k}\Big(\sum_m(\xi-z)^{-m-1}a_m\Big)\otimes 1+1\otimes a_k.\end{aligned}$$ ii) The intertwining operators $\Phi_{\lambda,l;\mu,m}^{\nu,n}(z)$ satisfy the following quadratic relation: $$\begin{aligned}
\label{w2intrel}
&&\mathscr{A}_{z_1,z_2}\big(\Phi_{\lambda_3,l_3;\rho,r}^{\lambda_0,l_0}(z_2)\Phi_{\lambda_2,l_2;\lambda_1,l_1}^{\rho,r}(z_1)\big)(P\otimes 1)=\nonumber\\
&&\Big(\frac{p}{q}\Big)^{l_2l_3}\sum_{\xi,k}B_{\rho,r;\xi,k}
\left[\begin{array}{cc}
\lambda_0,l_0 &\lambda_1,l_1 \\
\lambda_2,l_2 &\lambda_3,l_3
\end{array}
\right]
\Phi_{\lambda_2,l_2;\xi,k}^{\lambda_0,l_0}(z_1)\Phi_{\lambda_3,l_3;\lambda_1,l_1}^{\xi,k}(z_2), \end{aligned}$$ where $p=e^{\pi i\eta}$, $q=e^{\frac{\pi i}{\varkappa}}$.
[**Proof.**]{} i) First, we construct an intertwining operator $$\begin{aligned}
\Psi^n_{lm}:F_{l,\eta}\otimes F_{m,\eta}\to F_{n,\eta}[[z,z^{-1}]]z^{\eta(n^2-m^2-l^2)},\end{aligned}$$ which obeys the rule $$\begin{aligned}
\label{aint}
a_k\cdot\Psi_{lm}^{n}(z)=\Phi_{lm}^{n}(z)\Delta_{z,0}(a_k).\end{aligned}$$ However, we already constructed one, when studied the first example of braided algebra in subsection 4.1. The operator $$\begin{aligned}
\delta_{l+m,n}Y(\cdot, z)\cdot:F_{l,\eta}\otimes F_{m,\eta}\to F_{n,\eta}[[z,z^{-1}]]z^{\eta(n^2-m^2-l^2)}\end{aligned}$$ satisfies the property (\[aint\]). Now we introduce an operator $$\begin{aligned}
\Phi_{\lambda,l;\mu,m}^{\nu,n}(z)=\Phi_{\lambda,\mu}^{\nu}(z)\otimes \Psi_{lm}^{n}(z),\end{aligned}$$ where $\Phi_{\lambda,\mu}^{\nu}(z)$ is the intertwining operator for irreducible highest weight modules of Virasoro algebra, generated by $L_n\equiv {{\cal L}}_n-\frac{1}{4\eta}\sum_m:a_{n-m}a_m:$ operators, and therefore is the intertwining operator for the irreducible highest weight representations of $W_2$ algebra.
The second statement of the Proposition follows from the commutativity condition of $Y$ and similar statement for intertwining operators for Virasoro algebra. $\blacksquare$
[**4.4. From $W_2$ and $U_q(gl(2))$ to braided VOA.**]{} Now we have all necessary tools to build a more sophisticated example of a braided vertex algebra than the one in section 4.1. In [@fz] we have constructed the braided VOA on the space $$\mathbb{F}_\varkappa =\bigoplus_{\lambda\in\mathbb{Z}_+} (V_{\Delta(\lambda),\varkappa}\otimes V_\lambda),$$ where $V_{\lambda}$ is an irreducible representation of $U_q(sl(2))$ with highest weight $\lambda$ and $V_{\Delta(\lambda)}$ are highest weight Virasoro modules discussed in Subsection 3.1., such that the $\mathcal{R}$-operator from braided commutativity relation of $\mathbb{F}_{\varkappa}$ is related to the universal $R$-matrix for $U_q(sl(2))$.
Here we modify the construction of [@fz] in order to build braided VOA compatible with the $W_2$ and $U_q(gl(2))$ structures, namely, it will be defined on the space $$\mathbb{G}_{\varkappa,\eta}= \bigoplus_{\lambda\in\mathbb{Z}_+, k\in\mathbb{Z}}(W^{\varkappa,\eta}_{\Delta(\lambda), k}\otimes V_{\lambda,k}).$$ We define a map $$Y(\cdot, z)\cdot: \mathbb{G}_{\varkappa,\eta}\otimes \mathbb{G}_{\varkappa,\eta}\to \mathbb{G}_{\varkappa,\eta}\{z\}$$ in such a way that $$\label{bvoa}
Y:v\otimes a\rightarrow Y(v\otimes a,z)=
\sum_{\nu,\mu} \Phi^{\nu,n}_{\lambda,l;\mu,m}(z)(v\otimes \cdot)\otimes \phi^{\nu,n}_{\mu,m;\lambda,l}(\cdot \otimes a),$$ where $v\in W^{\varkappa,\eta}_{\Delta(\lambda),l}$ and $a\in V_{\lambda,l}$.
Let us denote $\mathbb{G}_{\varkappa,\eta}(l,\lambda)\equiv W^{\varkappa,\eta}_{\Delta(\lambda), l}\otimes V_{\lambda,l}$. One can see that the sectors for the map $Y$ are given by the spaces $\mathbb{G}_{\varkappa,\eta}(l,\lambda)$ and the conformal weight of the sector is $\Delta(\lambda)+\eta k^2$. Let us prove that $Y$ satisfies the braided commutativity relation and compute an explicit expression for the $\mathcal{R}$-operator. Let $v_i\otimes a_i\in W^{\varkappa,\eta}_{\Delta(\lambda_i), l_i}\otimes V_{\lambda_i,l_i}$ $(i=1,2,3)$, then $$\begin{aligned}
&&\mathscr{A}_{z_2,z_1}\big(Y(v_1\otimes a_1,z_2)Y(v_2\otimes a_2,z_1)\big)(v_3\otimes a_3)=\nonumber\\
&&\mathscr{A}_{z_2,z_1}\Big(\sum_{\lambda_1,\lambda_2,\nu,\rho}
\Phi_{\lambda_1,l_1\rho,r}^{\nu,n}(z_2)\Phi_{\lambda_2,l_2;\lambda_3,l_3}^{\rho,r}(z_1)\otimes
\phi_{\rho,r;\lambda_1,l_1}^{\nu,n}\phi_{\lambda_3,l_3;\lambda_2,l_2}^{\rho,r}\Big)\cdot\nonumber\\
&&(v_1\otimes v_2\otimes v_3)\otimes(a_3\otimes a_2\otimes a_1)=\nonumber\\
&&\Big(\sum_{\lambda_1,l_1,\lambda_2,l_2,\nu,n,\rho,r,\xi,k}\Phi_{\lambda_2,l_2,\xi,k}^{\nu,n}(z_1)\Phi_{\lambda_1,l_1;\lambda_3,l_3}^{\xi,k}(z_2)\nonumber\\
&&
\Big(\frac{p}{q}\Big)^{l_1l_2}
B_{\rho,r;\xi,k}
\left[\begin{array}{cc}
\nu,n &\lambda_3,l_3 \\
\lambda_2,l_2 &\lambda_1,l_1
\end{array}
\right] \otimes
\phi_{\rho,r;\lambda_1,l_1}^{\nu,n}\phi_{\lambda_3,l_3;\lambda_2,l_2}^{\rho,r}\Big)\cdot\nonumber\\
&&(v_2\otimes v_1\otimes v_3)(a_3\otimes a_2\otimes a_1)=\nonumber\\
&&\Big(\sum_{\lambda_1,l_1,\lambda_2,l_2,\nu,n,\xi,k}
\Phi_{\lambda_2,l_2;\xi,k}^{\nu,n}(z_1)\Phi_{\lambda_2,l_2;\lambda_3,l_3}^{\xi,k}(z_2)\otimes
\phi_{\xi,k;\lambda_2,l_2}^{\nu,n}\phi_{\lambda_3,l_3;\lambda_1,l_1}^{\xi,k}\Big)\cdot\nonumber\\
&&(v_2\otimes v_1\otimes v_3)\otimes(a_3\otimes \sum_ir_i^{(2)}a_1\otimes r_i^{(1)}a_2)=\nonumber\\
&&\sum_i Y(v_2\otimes r_i^{(1)}a_2,z_1)Y(v_1\otimes r_i^{(2)}a_1,z_1)(v_3\otimes a_3),\end{aligned}$$ where $$\mathcal{R}\equiv r^{(1)}_i\otimes r_i^{(2)}=\Big(\frac{p}{q}\Big)^{I\otimes I}R,$$ and $R$ is a universal R-matrix for $U_q(gl(2))$. Therefore $Y$ satisfies the braided commutativity relation. One can prove that $Y$ satisfies all other necessary properties along the lines of [@fz] and we arrive to the following Proposition.
The map $Y$ defines a structure of braided VOA on $\mathbb{G}_{\varkappa,\eta}$.
The braided VOA $\mathbb{G}_{\varkappa,\eta}$ possesses the following remarkable subalgebras.
$\mathbb{G}_{\varkappa,\eta}$ possesses braided vertex subalgebras $\hat{\mathbb{G}}_{\varkappa,\eta}$, $\hat{\mathbb{G}^{+}}_{\varkappa,\eta}$ with the spaces $$\begin{aligned}
&&\hat{\mathbb{G}}_{\varkappa,\eta}=\bigoplus_{\lambda\in\mathbb{Z}_+, r\in\mathbb{Z}}(W^{\varkappa,\eta}_{\Delta(\lambda), \lambda+2r}\otimes V_{\lambda,\lambda+2r}),\nonumber\\
&&\hat{\mathbb{G}^{+}}_{\varkappa,\eta}=\bigoplus_{\lambda,r\in\mathbb{Z}_+}(W^{\varkappa,\eta}_{\Delta(\lambda), \lambda+2r}\otimes V_{\lambda,\lambda+2r}).\end{aligned}$$ Here $\hat{\mathbb{G}}^+_{\varkappa,\eta}$ is the minimal braided subalgebra of $\mathbb{G}_{\varkappa,\eta}$, containing subspace $\mathbb{G}_{\varkappa,\eta}(1,1)$, and $\hat{\mathbb{G}}_{\varkappa,\eta}$ can be obtained from it by extending via subspace $\mathbb{G}_{\varkappa,\eta}(-2,-2)$.
Finally, we mention the following important property. There is a natural action of $U_q(gl(2))$ on ${\mathbb{G}}_{\varkappa,\eta},\ \hat{\mathbb{G}}_{\varkappa,\eta}, \hat{\mathbb{G}}^+_{\varkappa,\eta}$. The next Proposition shows that it is compatible with the braided VOA structure.
There is a natural $U_q(gl(2))$ action on the vertex algebras ${\mathbb{G}}_{\varkappa,\eta},\ \hat{\mathbb{G}}_{\varkappa,\eta}, \hat{\mathbb{G}}^+_{\varkappa,\eta}$, such that $gY=Y\Delta(g),$ where $g\in U_q(gl(2))$.
[**Proof.**]{} This is a direct consequence of the definition of $Y$ and the properties of the intertwining operator on $U_q(gl(2))$. $\blacksquare$
$GL_q(2)$ as semi-infinite cohomology and Laplace operator.
============================================================
[**5.1. Computation of the semi-infinite cohomology.**]{} We are interested in computation of the semi-infinite cohomology of the tensor product of the following modules of $W_2$: $$W^{\varkappa,\eta}_{\Delta(\lambda),l}\otimes W^{-\varkappa,-\eta}_{\Delta(\mu),m}.$$ For $H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}{\bf c},\cdot)$ the answer is given in the following proposition.
The 0th semi-infinite cohomology group for the tensor product of two irreducible highest weight $W_2$ modules is given by $$H^{\frac{\infty}{2}+0}
(W_2,\mathbb{C}{\bf c},W^{\varkappa,\eta}_{\Delta(\lambda),l}\otimes W^{-\varkappa,-\eta}_{\Delta(\mu),m})=\mathbb{C}\delta_{\lambda,\mu}\delta_{l,m}.$$
[**Proof.**]{} First, to simplify the semi-infinite cohomology operator we use Proposition 3.5. As well as there we have tensor product of $W_2$-modules with complimentary central charges. Moreover, the associated semi-infinite complexes formed out of “tilded” variables and original ones are isomorphic to each other. Therefore, we can compute the semi-infinite cohomology using the operator $Q^{\oplus}$. Part (ii) of Corollary 3.2. says that $H^{\frac{\infty}{2}+0}$ is nontrivial iff $l=m$. Part (i) says that $\lambda$ should be equal to $\mu$, since $\varkappa$ is generic. Part (iii) leads to the fact that only the highest weight vectors of the tensor product of Fock modules $F_{\varkappa,\eta}\otimes F_{-\varkappa,-\eta}\otimes M_{\psi,\chi}$ (where $M_{\psi,\chi}$ is the part of $M$ generated by $\psi_m,\chi_n$ modes) contribute to nontrivial cohomology classes. Hence, the problem is reduced to the computation of the 0th semi-infinite cohomology group of Virasoro modules, i.e. $H^{\frac{\infty}{2}+0}(Vir,\mathbb{C}{\bf c},V_{\Delta(\lambda),\varkappa}\otimes V_{\Delta(\lambda), -\varkappa})$, which is known to be equal to $\mathbb{C}$ (see [@lz1], [@lz2], [@fs]). Hence, the proposition is proven. $\blacksquare$
Let us introduce the following spaces with the structure of braided VOA: $$\begin{aligned}
&&\mathbb{G}=\mathbb{G}_{\varkappa,\eta}\otimes \mathbb{G}_{-\varkappa,-\eta}, \quad \hat{\mathbb{G}}=\hat{\mathbb{G}}_{\varkappa,\eta}\otimes \hat{\mathbb{G}}_{-\varkappa,-\eta}\nonumber\\
&&\hat{\mathbb{G}}^+=\hat{\mathbb{G}}^+_{\varkappa,\eta}\otimes \hat{\mathbb{G}}^+_{-\varkappa,-\eta}\end{aligned}$$ We see that $\hat{\mathbb{G}}$ is a braided VOA subalgebra in $\mathbb{G}$. An immediate consequence of Proposition 5.1. is the following Theorem.
The 0th cohomology groups of the spaces of braided vertex algebras $\mathbb{G}$, $\hat{\mathbb{G}}$, $\hat{\mathbb{G}}^+$ are given by $$\begin{aligned}
&& H^{\frac{\infty}{2}+0}(Vir,\mathbb{C}{\bf c},\mathbb{G})=
\oplus_{\lambda,k\in\mathbb{Z}}V_{\lambda,k}\otimes \bar{V}_{\lambda,k},\nonumber\\
&& H^{\frac{\infty}{2}+0}(Vir,\mathbb{C}{\bf c},\hat{\mathbb{G}})=
\oplus_{\lambda\in\mathbb{Z}_+,r\in\mathbb{Z}}V_{\lambda,\lambda+2r}\otimes \bar{V}_{\lambda,\lambda+2r},\nonumber\\
&& H^{\frac{\infty}{2}+0}(Vir,\mathbb{C}{\bf c},\hat{\mathbb{G}}^+)=
\oplus_{\lambda,r\in\mathbb{Z}_+}V_{\lambda,\lambda+2r}\otimes \bar{V}_{\lambda,\lambda+2r}\end{aligned}$$ where $\bar{V}_{\lambda,k}$ stands for irreducible representation of $U_q(gl(2))$ after involution $q \to q^{-1}$.
Now we will define a product structure on these cohomology spaces using the natural product structure on VOA.\
[**5.2. Ring structure on the semi-infinite cohomology spaces.**]{} The Lian-Zuckerman associative product structure is defined on the representatives $U$, $V$ of the cohomology classes of VOA as follows [@lz2]: $$\label{fzprod}
\mu(U,V)=\mathrm{Res}_z \Big(\frac{U(z)V}{z}\Big).$$ In [@fz] we have shown that it is also gives an associative algebra structure on the space $\mathbb{F}=\mathbb{F}_{\varkappa}\otimes\mathbb{F}_{\varkappa}$. The same reasoning we used in that case applies to $\mathbb{G}$. Hence, we obtain the following Proposition.
The 0th semi-infinite cohomology space of the braided VOAs $\mathbb{G}$, $\hat {\mathbb{G}}$, $\hat {\mathbb{G}^+}$ possess an associative product given by the formula (\[fzprod\]) on the representatives of cohomology classes, such that it has the following subalgebras: $$\begin{aligned}
&&(H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\hat{\mathbb{G}}^+),\mu)\subset\nonumber\\
&&(H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\hat{\mathbb{G}}),\mu)
\subset (H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\mathbb{G}),\mu).\end{aligned}$$
This proposition has an immediate corollary, which follows from the braided commutativity property.
The operation $\mu$ being considered on $H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\mathbb{G})$ is associative and satisfies the following commutativity relation: $$\mu(U,V)=\mu(\hat{r}_i^{(1)}V,\hat{r}_i^{(2)}U).$$ Here $\hat{\mathcal{R}}=\sum_i \hat{r}_i^{(1)}\otimes\hat{r}_i^{(2)}=\mathcal{R}\bar{\mathcal{R}}$, where $\mathcal{R}$, $\bar{\mathcal{R}}$ are the braiding operators on $\mathbb{G}_{\varkappa}, \mathbb{G}_{-\varkappa}$ correspondingly.
Now we want to find the generating set for both $H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\hat{\mathbb{G}})$ and $H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\mathbb{G})$ under the multiplication $\mu$.
We recall that we denoted $\mathbb{G}_{\varkappa,\eta}(l,\lambda)\equiv W^{\varkappa,\eta}_{\Delta(\lambda), l}\otimes V_{\lambda,l}$. Let $\mathbb{G}(\mu,k)=\mathbb{G}_{\varkappa,\eta}(\mu,k)\otimes\mathbb{G}_{-\varkappa,-\eta}(\mu,k)$. Due to the structure of the braided VOA on $\mathbb{G}$ and $\hat{\mathbb{G}}$ we have the following Proposition.
i\) Algebra $(H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\hat{\mathbb{G}}^+),\mu)$ is generated by\
$H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},{\mathbb{G}}(1,1))$.\
ii) Algebra $(H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\hat{\mathbb{G}}),\mu)$ is generated by\
$H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},{\mathbb{G}}(1,1))$ and $H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},{\mathbb{G}}(0,-2))$.\
iii)Algebra $(H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},{\mathbb{G}}),\mu)$ is generated by $H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},{\mathbb{G}}(1,1))$ and $H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},{\mathbb{G}}(0,-1))$.
We note that $H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\mathbb{G}(1,1))$ $\cong
V_{1,1}\otimes \bar{V}_{1,1}$ and $H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},{\mathbb{G}}(0,-2))$ $\cong\mathbb{C}$, $H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\hat{\mathbb{G}}(0,-1))\cong\mathbb{C}$.
Since $\mathcal{R}=p^{I\otimes I}\mathcal{R}'$, where $\mathcal{R}'$ is the universal R-matrix for $U_q(sl(2))$ (i.e. the braiding operator from braided commutativity relation for $\mathbb{F}_{\varkappa}$), we have $\hat{\mathcal{R}}=R'\bar{R'}$. Now we calculate the commutation relations between the elements from the generating set. There are only two vectors in each of $V_{1,1},\bar V_{1,1}$, i.e. the highest weight and the lowest weight vectors. We denote them as $a_+, a_-$ and $\bar{a}_+,\bar{a}_-$ correspondingly. Let us make the following notation: $$\begin{aligned}
\label{abcd}
v\otimes a_-\otimes\bar{a}_+=A,\quad
v\otimes a_+\otimes\bar{a}_-=D,\nonumber\\
v\otimes a_+\otimes\bar{a}_+=B,\quad
v\otimes a_-\otimes\bar{a}_-=C,\end{aligned}$$ where $v\in H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},W^{\varkappa,\eta}_{1,1}\otimes W^{-\varkappa,-\eta}_{1,1})$. Then the following Proposition holds, which is a consequence of Proposition 6.9 of [@fz].
i)The generators $A,B,C,D$ satisfy the following relations: $$\begin{aligned}
&&AB=BA{q^{-1}}, \quad CB=BC,\quad DB=BDq,\quad CA=ACq,\nonumber\\
&&AD-DA=(q^{-1}-q)BC,\quad CD=DCq^{-1},\end{aligned}$$ and $det_q\equiv AD-q^{-1}BC$ is a nonzero element in $H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\hat{\mathbb{G}}(0,2))$.\
ii) The element $det_q$ has inverse, which belongs to $H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\hat{\mathbb{G}}(0,-2))$, i.e. $(H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\hat{\mathbb{G}}),\mu)\cong GL_q(2)$.
[**Remark.** ]{} One can see that the algebra $(H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\hat{\mathbb{G}}^+),\mu)$ is a subalgebra of $GL_q(2)$ generated by the elements $A,B,C,D$ only, and $(H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},{\mathbb{G}}),\mu)$ corresponds to $GL_q(2)$ extended by element $t\in H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\hat{\mathbb{G}}(0,1))$, such that $t^2=det_q$.\
[**5.3. Quantum Laplacian.**]{} Let us denote $GL_q(2)^+\equiv (H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\hat{\mathbb{G}}^+),\mu)$. In [@fj] it was defined an extension of this algebra by the invertible element $\delta$, obeying the following commutation relations with the generating elements: $$\begin{aligned}
\delta A=A\delta, \quad \delta B=qB\delta, \nonumber\\
\delta C=q^{-1}C\delta, \quad \delta D=D\delta.\end{aligned}$$ In such a way one can define elements $x_{11}, x_{12}, x_{21}, x_{22}$: $$\begin{aligned}
x_{11} = \delta A, \quad x_{12} = q^{-1/2}\delta B ,\nonumber\\
x_{21} = q^{1/2}\delta C ,\quad x_{22} = \delta D\end{aligned}$$ satisfying commutation relations: $$\begin{aligned}
&&x_{11}x_{12} = x_{12}x_{11} ,\quad x_{21}x_{22} = x_{22}x_{21} , \nonumber\\
&&[x_{11} , x_{22} ] + [x_{21} , x_{12} ] = 0 , \nonumber\\
&&x_{11}x_{21} = q^{-2}x_{21}x_{11} , \quad x_{12}x_{22} = q^{-2}x_{22}x_{12} , \nonumber\\
&&x_{21}x_{12} = q^2x_{12}x_{21}.\end{aligned}$$ The space generated by $x_{11},x_{12}, x_{21},x_{22}$ will be called, according to [@fj], the $quantum$ $Minkowski$ $space-time$. One can define the quantum Laplacian operator on this space by the formula: $$\begin{aligned}
\square_x={\partial}_{11}{\partial}_{22}-{\partial}_{12}{\partial}_{21},
\end{aligned}$$ where ${\partial}_{11}{\partial}_{22},{\partial}_{12},{\partial}_{21}$ obey the following commutation relations: $$\begin{aligned}
&&{\partial}_{11}{\partial}_{21} = {\partial}_{21}{\partial}_{11} ,\quad {\partial}_{12}{\partial}_{22} = {\partial}_{22}{\partial}_{21} , \nonumber\\
&&[{\partial}_{11} , {\partial}_{22} ] + [{\partial}_{21} , {\partial}_{12} ] = 0 , \nonumber\\
&&{\partial}_{11}{\partial}_{12} = q^{-2}{\partial}_{12}x_{11} , \quad {\partial}_{21}{\partial}_{22} = q^{-2}{\partial}_{22}{\partial}_{21},
\nonumber\\
&&{\partial}_{12}{\partial}_{21} = q^2{\partial}_{21}{\partial}_{12}.
\end{aligned}$$ It was shown in [@fj] that the kernel of quantum Laplacian operator is spanned by elements of the following kind: $$\begin{aligned}
X^{\lambda}_
{rk} =\frac{1}{2\pi i}
\oint (x_{11}s + x_{21})^{\frac{\lambda-r}{2}}(x_{12}s + x_{22} )^{\frac{\lambda+r}{2}}s^{\frac{k-\lambda}{2}-1} ds,\end{aligned}$$ where $$\begin{aligned}
-\lambda \le r,k \le \lambda , \lambda\in{\mathbb{Z}_+},\quad r,k \equiv \lambda(\rm{mod} 2).\end{aligned}$$ Moreover, the operator $$\begin{aligned}
\tilde \square_x\equiv det(x)\square_x, \quad{\rm where} \quad det(x)=\delta^2det_q=x_{11}x_{22}-x_{12}x_{21}\end{aligned}$$ is diagonalizable on the quantum Minkowski space-time.
[[@fj]]{} The eigenvectors of the operator $\tilde \square_x$ have the form $$\begin{aligned}
det(x)^jX^{\lambda}_
{rk}, \end{aligned}$$ such that the eigenvalues are $$\begin{aligned}
\label{eig}
(j)_{q}(j+\lambda+1)_{q},\end{aligned}$$ where $(n)_q=\frac{q^{2n}-1}{q^2-1}$.
Let us denote $$\begin{aligned}
Y^{\lambda,j}_{r,k}(A,B,C,D)\equiv \delta^{-2j-\lambda}det(x)^jX^{\lambda}_
{rk}. \end{aligned}$$ The following Proposition holds.
The elements $Y^{\lambda,j}_{r,k}(A,B,C,D)$ form a basis in the subspace $V_{\lambda,\lambda+2j}\otimes \bar{V}_{\lambda,\lambda+2j}$ of $GL^+_q(2)$.
[**Proof.**]{} It is enough to prove that $Y^{\lambda,0}_{r,k}(A,B,C,D)$ form a basis in $V_{\lambda,\lambda}\otimes \bar{V}_{\lambda,\lambda}$. We will prove this by induction in $\lambda$. The assertion is obvious in the case $\lambda=1$. Then $V_{1,1}\otimes \bar{V}_{1,1}$ has a basis $Y^{1,0}_{r,k}(A,B,C,D)$, where $r,k$ are equal to $0$ or $2$. Suppose that this statement holds for $\lambda-k$, where $\lambda\ge k\ge 0$. Let’s prove it for $\lambda+1$. First of all, we know that the monomials of $\lambda$ elements form a basis in the space $U_{\lambda}=\oplus_{\lambda\ge 2m\ge 0}V_{\lambda-2m,\lambda}\otimes \bar{V}_{\lambda-2m,\lambda}$. By induction, we know that $Y^{\lambda-2m}_{r,k}(A,B,C,D)$ form a basis in all $U_{\mu}$, where $\mu\le \lambda$.
As a consequence of Proposition 4.4. we have two natural actions of $U_q(gl(2))$ on $GL^+_q(2)$ inherited from vertex algebras. Let’s consider the action of $U_q(gl(2))$, obtained from $\mathbb{G}_{-\varkappa,-\eta}$ braided VOA. The action of the corresponding $I, q^H,E,F$-generators on $A,B,C, D$ is as follows: $$\begin{aligned}
&&IA=A\quad, IB=B,\quad IC=C,\quad ID=D,\nonumber\\
&&FB=A,\quad FD=C, \quad FC=0,\quad FA=0,\nonumber\\
&&q^HB=qB, \quad q^HD=qD,\quad q^HC=q^{-1}C, \quad q^HA=q^{-1}A,\nonumber\\
&&EA=B,\quad EC=D, \quad EB=0, \quad ED=0.\end{aligned}$$ One can see that action of $E,F$ interchanges the commuting elements $(x_{11}s + x_{21}) \leftrightarrow (x_{12}s + x_{22} )$ and therefore the space generated by $Y^{\lambda,0}_{r,k}(A,B,C,D)$ is invariant under $U_q(gl(2))$ action. Moreover, for fixed $k$, $Y^{\lambda,0}_{r,k}(A,B,C,D)$ span the $\lambda+1$ dimensional representation of $U_q(gl(2))$. At the same time, the elements of $U_{\lambda}$ which belong to $U_q(gl(2))$ irreducible representations with highest weight $\lambda$, should belong to the space $V_{\lambda,\lambda}\otimes \bar{V}_{\lambda,\lambda}$. Therefore, $Y^{\lambda,0}_{r,k}(A,B,C,D)\in
V_{\lambda,\lambda}\otimes \bar{V}_{\lambda,\lambda}$ and hence $Y^{\lambda,0}_{r,k}(A,B,C,D)$ form a basis in $V_{\lambda,\lambda}\otimes \bar{V}_{\lambda,\lambda}$, since they are linearly independent. Thus the Proposition is proven. $\blacksquare$
Now we give meaning to the operator $\tilde\square_x$ in vertex algebra setting. Consider an operator $\tilde\square^{CFT}_x$ on the braided VOA $\hat{ \mathbb{G}}^+_{\varkappa,\eta}$ of the following form: $$\begin{aligned}
\tilde\square^{CFT}_x=(\hat{a}_0-\alpha_0)_q(\hat{a}_0+1)_q,\end{aligned}$$ where $\alpha_0$ is zero-mode of the quantum field $\alpha(z)$ from Feigin-Fuks realization of Virasoro algebra with generators $\{L_n\}$ and $\hat{a}_0=\eta^{-1}a_0$.
Being constructed by means of zero-modes, this operator commutes with the action of $W_2$ algebra and semi-infinite cohomology operator. Comparing the eigenvalues of $\tilde\square^{CFT}_x$ and $\tilde\square_x$ (see (\[eig\])) we obtain the following theorem.
The operator $\tilde\square^{CFT}_x$ induces the operator $\tilde\square_x$ on\
$(H^{\frac{\infty}{2}+0}(W_2,\mathbb{C}\mathbf{c},\hat{\mathbb{G}}^+),\mu)\cong
GL^+_q(2)$.
Acknowledgements {#acknowledgements .unnumbered}
================
We are indebted to P.I. Etingof, M. Jardim, A.A. Kirillov Jr., K. Styrkas, G.J. Zuckerman for fruitful discussions. The research of I.B.F. was supported by NSF grant DMS-0457444. A.M.Z. would like to thank the organizers of the Simons Workshop 2011, where this work was partly done.
[10]{} B. Feigin, E. Frenkel, [*Integrals of Motion and Quantum Groups*]{}, Lect. Notes in Math., 1620, Springer, Berlin 1996. arXiv:hep-th/9310022. E. Frenkel, D. Ben-Zvi, [*Vertex algebras and algebraic curves*]{}, AMS, Providence, USA (2004). I.B. Frenkel, H. Garland, G.J. Zuckerman, [*Semi-infinite cohomology and string theory*]{}, Proc. Nat. Acad. Sci. [**83**]{} (1986) 8442-8446. I.B. Frenkel, M. Jardim, [*Complex ADHM equations, sheaves on $\mathbb{P}^3$ and quantum instantons*]{}, arXiv:math/0408027. I.B. Frenkel, Y.-Z. Huang, J. Lepowsky, [*On axiomatic approach to vertex operator algebras and modules*]{}, Memoirs of AMS [**494**]{} (1993). I.B. Frenkel, K. Styrkas, [*Modified regular representations of affine and Virasoro algebras, VOA structure and semi-infinite cohomology*]{}, Adv. Math. 206 (2006) 57-111. I.B. Frenkel, A.M. Zeitlin, [*Quantum group as semi-infinite cohomology*]{}, Comm. Math. Phys. [**297**]{} (2010) 687-732, arXiv:0812.1620. B. H. Lian, G.J. Zuckerman, [*2D gravity with c=1 matter*]{}, Phys. Lett. [**B266**]{} (1991) 21-28. B. H. Lian, G.J. Zuckerman, [*Semi-infinite cohomology and 2D Gravity. I*]{}, Commun. Math. Phys. [**145**]{} (1992) 561-593. B. H. Lian, G.J. Zuckerman, [*New Perspectives on the BRST-algebraic structure of string theory*]{}, Commun. Math. Phys. [**154**]{} (1993) 613-646. G. Moore, N. Seiberg, [*Classical and quantum conformal field theory*]{}, Commun. Math. Phys. [**123**]{} (1989) 177-254. K. Styrkas, [*Quantum groups, conformal field theories, and duality of tensor categories*]{}, PhD Thesis, Yale, 1998.
[^1]: [email protected]
[^2]: [email protected], http://math.yale.edu/$\sim$az84 http://www.ipme.ru/zam.html
[^3]: By the $weak$ sense we mean that the relation holds for the matrix elements of the corresponding operator products.
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- 'N. Christlieb'
- 'P.J. Green'
- 'L. Wisotzki'
- 'D. Reimers'
bibliography:
- 'carbonstars.bib'
- 'datanaly.bib'
- 'imageprocessing.bib'
- 'mphs.bib'
- 'ncastro.bib'
- 'ncpublications.bib'
- 'quasar.bib'
- 'HES.bib'
date: 'Received 26-April-2001; accepted 11-June-2001'
subtitle: 'II. A large, homogeneously-selected sample of high latitude carbon stars'
title: 'The stellar content of the Hamburg/ESO survey[^1] '
---
Introduction
============
Models of the chemical and dynamical properties of the Galactic spheroid (the ‘halo’) are still rather weakly constrained. In the grand scheme, did a monolithic protogalaxy undergo rapid collapse and enrichment [@Eggenetal:1962], or did many smaller dwarf galaxies merge together [@Searle/Zinn:1978]? Both processes probably contribute, since there is solid recent evidence of ongoing mergers [@Ibataetal:1994; @Majewskietal:2000]. Stars and gas that are tidally stripped from accreting dwarf galaxies remain aligned with the orbit of the satellite for timescales comparable to the age of the Galaxy. Thus, a number of tidal streams exist today whose phase-space signature can constrain the stripping and merging events that contribute to the present-day galactic halo [@Johnstonetal:1999].
An important goal of astronomy in this century is to measure and model the potential of the Milky Way using halo stars as tracers. To simultaneously disentangle the remnants of disrupted satellites requires full knowledge of the angular positions, proper motions, radial velocities, and distances of a large number of such stars. But first, a large sample of distant halo stars must be amassed. Intrinsically bright stars visible to large galactocentric distances ($10$–$100$ kpc) provide the best opportunity. Because they are readily recognizable from their strong C$_2$ and CN absorption bands, and because they were thought to be giants without exception, faint C stars have been sought as excellent tracers of the outer halo.
Faint high galactic latitude carbon (FHLC) stars have been sought in prior objective prism surveys [e.g., @Sanduleak/Pesch:1988; @MacAlpine/Lewis:1978] and in the CCD survey of [@Greenetal:1994]. Objective-prism photography with wide-field Schmidt telescopes has yielded low-dispersion spectra for thousands of objects over substantial portions of the sky, but not a large number of carbon stars. Fewer than 1% of the $6\,000$ stars in Stephenson’s (1989) catalogue are the faint, high-latitude carbon (FHLC) stars ($V>13,~|b|>40^{\circ}$) most useful as dynamical probes of the outer halo. The two most prolific sources of published FHLC stars, the Case low-dispersion survey [CLS; @Sanduleak/Pesch:1988] and the University of Michigan – Cerro Tololo survey [UM; @MacAlpine/Williams:1981] appear to probe to about $V=16$ and have provided about 30 FHLC stars. Emission-line objects, not FHLC stars, were the primary goal of these photographic surveys, and known FHLC stars were not examined to help predefine selection criteria or estimate completeness. The surface density of FHLC stars from objective-prism surveys is low, about one per 50 deg$^2$ to $V\approx16$. At high galactic latitudes, mostly warm carbon stars are found – CH stars, and possibly some R stars. However, color selection of very red stars at high latitude also reveals a small number (one per 200 deg$^2$ to $R\sim 16$) of classical intermediate age AGB carbon (AGBC) stars [@Totten/Irwin:1998]. [@Margonetal:2000] recently reported the discovery of more than 30 new FHLC stars in the commissioning data of the Sloan Digitized Sky Survey (SDSS), which may eventually provide the majority of known FHLC stars.
In this paper, we describe our use of the Hamburg/ESO survey [HES; @hespaperI; @heshighlights; @hespaperIII] to greatly augment the number of known FHLC stars. The HES is an objective-prism survey designed to select bright ($12.5 \gtrsim B_J \gtrsim 17.5$) quasars in the southern extragalactic sky ($\delta<+2.5^\circ$; $|b|\gtrsim 30^\circ$). It is based on IIIa-J plates taken with the 1m ESO Schmidt telescope and its 4$^{\circ}$ prism, yielding a wavelength range of $3200\,\mbox{\AA} < \lambda
< 5200\,\mbox{\AA}$ and a seeing-limited spectral resolution of typically 15[Å]{} at H$\gamma$. This resolution makes possible the identification of carbon stars with high confidence without follow-up slit spectroscopy, based on their strong C$_2$ and CN molecular bands (cf. Fig. \[fig:Cbands\]).
Since carbon can reach the surface of an isolated star only in late evolutionary stages, it has long been assumed that all carbon (C) stars are giants. Due to their high luminosity ($M_R\sim -3.5$), it is possible to detect the red AGBC stars at large distances: [@Breweretal:1996] have identified C stars even in the local group galaxy M31. The more typical FHLC stars such as CH giants, with $0< M_V < -2.5$, can be detected to $\sim
60$kpc in sensitive photographic surveys. However, the long-held assumption that all C stars are giants has fallen. Trigonometric parallax measurements for the carbon star [@Dahnetal:1977] showed that this star lies close to the main sequence ($M_V\sim+10$). For years, the dwarf carbon (dC) stars phenomenon was assumed to be extremely rare until many new dCs were discovered in the early 1990s [@Greenetal:1991; @Greenetal:1992; @Warrenetal:1993; @Heberetal:1993; @Liebertetal:1994]. Discovery of so many dCs, and the remarkable similarity of their spectra to those of C giants means that care must be taken to distinguish dwarfs from giants in FHLC star samples intended for distant halo studies [@Greenetal:1992]. We are therefore undertaking a two-part investigation. In the current paper, we describe our automated selection of C stars in the HES, and we present a large, uniformly selected, and flux-limited sample of FHLC stars. We complement this sample in upcoming work with recent epoch astrometry, to measure proper motions for as many objects as possible, and thereby separate the dCs from C giants.
We note that carbon-enhanced, metal-poor stars may be among the FHLC stars presented here. It was recognized by several authors that the fraction of such stars among metal-poor stars rises with decreasing metallicity, reaching $\sim 25$% for stars with $\mbox{[Fe/H]}<-3.0$, and that carbon overabundances are as high as $\mbox{[C/Fe]}=+2.0$dex [e.g., @Norrisetal:1997; @Rossietal:1999]. $^{12}$C/$^{13}$C isotope measurements of a larger sample of such stars would help to identify the carbon production site(s) at work.
Carbon Star Selection {#selection}
=====================
The full HES database consists of $\sim 10$ million extracted, wavelength calibrated spectra. The input catalog for extraction of objective-prism spectra is generated by using the Digitized Sky Survey I (DSS I). An astrometric transformation between DSS I plates and HES plates yields, for each object in the input catalog, the location of its spectrum on the relevant HES plate, and provides a wavelength calibration zero point [@hespaperIII].
Carbon stars can be identified in the HES data base by their strong C$_2$ and CN bands. We select carbon star candidates when the mean signal-to-noise ratio ($S/N$) in the relevant wavelength range is $>5$ per pixel and both of the C$_2$ bands $\lambda\lambda\,5165$, $4737$, or both of the CN bands $\lambda\lambda\,4216$, $3883$ are stronger than a selection threshold. Band strengths are measured by means of line indices – ratios of the mean photographic densities in the carbon molecular absorption features and the continuum bandpasses shown in Fig. \[fig:Cbands\], and listed in Tab. \[tab:Cbands\]. The use of [*pairs*]{} of indices prevents confusion with plate artifacts, e.g., scratches. It is very unlikely that two such artifacts are present at the positions of two molecular bands. Selection boxes in the $I\,(\mbox{C}_2\;\lambda\,5165)$ versus $I\,(\mbox{C}_2\;\lambda\,4737)$ and $I\,(\mbox{CN}\;\lambda\,4216)$ versus $I\,(\mbox{CN}\;\lambda\,3883)$ planes were chosen well-separated from the dense locus of “normal” stars (see Fig. \[C2selection\]). The selection criteria are listed in Tab. \[tab:selectcrit\].
[lcccc]{}
------------------------------------------------------------------------
&\
[\[-1.5ex\][Passband]{}]{} & C$_2$ 5165 & C$_2$ 4737 & CN 4216 & CN 3883\
5190–5240[Å]{} & cont & & &
------------------------------------------------------------------------
\
5060–5150[Å]{} & flux & & &\
4800–4970[Å]{} & cont & cont & &\
4620–4730[Å]{} & & flux & &\
4460–4560[Å]{} & & cont & &\
4210–4270[Å]{} & & & cont &\
4130–4180[Å]{} & & & flux &\
3830–3890[Å]{} & & & & flux\
3610–3740[Å]{} & & & & cont\
[lc]{} Feature & Index range \[[Å]{}\]
------------------------------------------------------------------------
\
$\mbox{C}_2\;\lambda\,5165$ & \[10,91\]
------------------------------------------------------------------------
\
$\mbox{C}_2\;\lambda\,4737$ & \[15,114\]\
$\mbox{CN}\;\lambda\,4216$ & \[2,56\]\
$\mbox{CN}\;\lambda\,3883$ & \[13,55\]\
Carbon stars can be distinguished reliably from other late type stars, e.g. M or S stars, even if only weak C bands are present in their spectra (cf. Fig. \[CstarMstar\_compare\]).
Other potential sample contaminators are white dwarfs of type DQ, which show carbon molecular bands. However, since the latter usually have a much bluer continuum (see Fig. \[Cstar\_templates\]), they can easily be recognized by visual inspection of the spectra, and by their $U-B$ color. [@McCook/Sion:1999] list 49 DQs, of which 30 have an available $U-B$ measurement. The average $U-B$ of those is $-0.58$, i.e., $\sim 1.5$mag away from the average $U-B$ of the HES C star sample. Our $U-B$ colors are measured directly from the HES spectra with a mean accuracy of $\sigma_{U-B}=0.09$mag [@HESStarsI herafter Paper I]. The average $U-B$ of HES C stars is $\sim 0.9$, more than 90% have $U-B>0.5$, and there is *no* C star of $U-B<0$ in the HES sample. While 4 (i.e., 13%) of the 30 DQs with $U-B$ in [@McCook/Sion:1999] have $U-B>0.0$, the pressure-broadened features of DQs are easily distinguished by visual inspection of the carbon bands (see Fig. \[Cstar\_templates\]).
With a rough estimate of their surface density, we can quantify an upper limit for the contamination of the HES C star sample by “red” ($U-B>0.0$) DQs. First of all, we have to take into account that the ratio of northern hemisphere to southern hemisphere DQs is unbalanced in [@McCook/Sion:1999], as much as the *total* catalog is. This is because the southern hemisphere so far has been surveyed less extensively for white dwarfs. Assuming that the northern hemisphere sample of DQs is complete, we derive a surface density of 9 DQs brighter than $V=16.5$ in $20\,000$deg$^{2}$, i.e. $4.5\cdot
10^{-4}$deg$^{-2}$. Hence, the surface density of $U-B>0.0$ DQs is $5.9\cdot 10^{-5}$deg$^{-2}$, and we expect $0.44$ DQs to be present on all 329 plates currently used for the exploitation of the stellar content of the HES. Therefore, even if we assume that the sample of DQs known so far is incomplete by a factor of 2, we statistically expect less than 1 DQ to be present in the HES C star sample.
On the 329 HES plates (effective area $6\,400$deg$^2$) we found 403 FHLCs. 90 of them were selected by C$_2$ band indices only, 171 by CN band indices only, and 144 by C$_2$ and CN indices. The $V$ and $B-V$ distributions are displayed in Fig. \[FHLC\_VBminVhisto\]. The faintest objects have $V\sim 16.5$, and the most distant objects reach $\sim 35$kpc (cf. Fig. \[HESFHLCsdist\]), assuming they are all giants with $M_V=-1$mag.
Testing the Automated Selection
===============================
We tested the automated selection extensively and by various methods. In Sect. \[sect:sel\_eff\] we investigate the selection efficiency. In Sect. \[sect:dC\_selfunct\] we derive plate-by-plate selection probabilities for halo dCs on HES plates by simulations. The results of tests with “real” objects are given in Sect. \[Cstar\_testobjects\].
Selection Efficiency {#sect:sel_eff}
--------------------
An important criterion for the evaluation of the quality of a selection algorithm is the *selection efficiency*, i.e. the fraction of desired stars in the raw candidate sample. Tab. \[tab:sel\_eff\] summarizes the results.
[lrr]{} Raw candidate &
------------------------------------------------------------------------
\
reduction factor & [\[-1.5ex\][1/2900]{}]{}\
C stars & 31.6%\
UNID & 7.0%\
OVL & 29.2%\
ART & 8.7%\
NOIS & 3.8%\
SAT & 15.6%\
Our selection is very efficient. The low fraction of artifacts demonstrates that the usage of *pairs* of C$_2$ bands and CN bands indeed very reliably excludes artifacts from selection. However, a considerable number of overlapping spectra (overlaps) are selected. Overlaps are detected by an automatic overlap detection algorithm in the HES, using the direct plate data of the DSS I. It appears that our carbon star selection technique is very sensitive in finding the small number of overlaps not detected by the automatic algorithm.
Decrease of Selection Probability for Halo dCs {#sect:dC_selfunct}
----------------------------------------------
In the HES, some care must be taken when objects with large proper motion are selected. This is because the input catalog for extraction of objective prism spectra is generated by using the DSS I. The dispersion direction of the HES spectra is along declination. Therefore, large proper motions and/or large epoch differences between HES and DSS I plates (13.5 years on average) may result in an offset of the wavelength calibration zero point, leading to smaller C band values, and/or non-detection of objects in the HES, if $\mu_{\alpha}\Delta\,t_{\mbox{\scriptsize HES-DSS~I}}\gtrsim 4''$, i.e., $>$ 3 pixels. Offsets of $\mu_{\alpha}\Delta\,t_{\mbox{\scriptsize HES-DSS~I}}< 4''$ can be recovered by the spectrum extraction algorithm. Note however that proper motions of a typical halo object ($<\!\!u\!\!>=<\!\!w\!\!>=0$km/s; $<\!\!v\!\!>\sim 200$km/s) result in $\sim 2\times$ larger offsets along declination than in R.A., since the galactic plane is tilted by $62.6^{\circ}$ with respect to the equatorial coordinate system.
In order to estimate how many dCs are expected to be missed in our survey due to the epoch difference problem, we carried out a simulation study in which the plate-by-plate selection function for halo dCs was determined. The simulation is similar to that described in [@Greenetal:1992]. We employ a sample of simulated dCs with halo kinematics, as given by [@Norris:1986]. For the solar neighborhood, he gives $$\label{eq:vrot}
v_{\mbox{\scriptsize rot}} = 37\pm 10\,\mbox{km/s}
\Longleftrightarrow <\!\!v\!\!> = -187\,\mbox{km/s},$$ and he determined the velocity ellipsoid to be $$\begin{aligned}
\sigma_u&=&131\pm 6\,\mbox{km/s}\label{eq:sigma_u}\\
\sigma_v&=&106\pm 6\,\mbox{km/s}\label{eq:sigma_v}\\
\sigma_w&=& 85\pm 6\,\mbox{km/s}\label{eq:sigma_w}.\end{aligned}$$ In each simulation we constructed 100 random velocity vectors $(u,v,w)$, with components following Gaussian distributions according to the above parameters. These velocity vectors were each applied to stars located at the center of the plate under investigation, and converted to proper motions assuming distances $d$. These were computed from the apparent $V$ magnitude distribution of a sample of 86 C stars without significant p.m. (see Fig. \[FHLC\_VBminVhisto\]), and assuming $M_V=+10$ for dwarf carbon stars. This yields $86\cdot 100=8\,600$ simulated stars. We compute the position of the star after the time $\Delta t_{\mbox{\scriptsize HES-DSS~I}}$, the epoch difference between DSS I and HES plate, and derive proper motions $\mu_\alpha$, $\mu_\delta$ from the position differences. We then select the subsample of the 8600 stars with $\mu_{\alpha}\Delta\,t_{\mbox{\scriptsize
HES-DSS~I}}< 4''$.
As a test sample for an investigation of the dependence of the selection probability on $\mu_\delta\Delta t$, we used a sample of 78 C stars from 44 HES plates *without* significant p.m., as measured in an follow-up campaign carried out in April at ESO, using the Wide Field Imager attached to the MPG/ESO 2.2m telescope (Christlieb et al. 2001, in preparation). The stars were shifted in 1 pixel ($={1\mbox{$''\mskip-7.6mu.\,$}35}$) steps through the range $-700\,\mu\,\mbox{m}<x<+700\,\mu\,\mbox{m}$, corresponding to $-{47\mbox{$''\mskip-7.6mu.\,$}25}<\mu_\delta\Delta t<{47\mbox{$''\mskip-7.6mu.\,$}25}$. At each shift step, the selection algorithms were applied.
By applying the selection probability (a function of $\mu_\delta\Delta t$) to the subsample of the 8600 stars with $\mu_{\alpha}\Delta\,t_{\mbox{\scriptsize HES-DSS~I}}< 4''$, we determine the fraction of stars which would be detected in the HES *and* selected by our selection algorithm. On the 329 stellar HES plates, $21.4$% of the simulated halo dCs are detected and selected.
[@Greenetal:1992] found that 13% of their C stars are dwarfs. Applying this estimate to our sample, and taking into account that we find only $\sim
20$% of the dCs detectable on the HES plates, we estimate that 10–15 out of our 403 FHLCs are dCs. However, since our sample is biased to low-p.m., it is likely that not all of these can be *proven* to be dCs by their large transverse velocity. Based on our simulations we estimate that additional $\sim 40$ dCs would be detectable on the HES plates, but are currently missed due to the epoch difference problem. This incompleteness will be addressed in a later paper focusing on dC stars in the HES.
Tests with Known C Stars {#Cstar_testobjects}
------------------------
We also compiled a test sample of known dwarf and giant C stars present on HES plates (see Tab. \[tab:seltest\]). We took all three dCs in the southern hemisphere listed by [@Deutsch:1994], i.e. LHS 1075, G77–61, and KA 2. The (possible) dCs of [@Warrenetal:1993], having $B_J>20$, unfortunately are by far too faint to be detectable on HES plates. Cross-identification with the C star lists of [@Slettebaketal:1969a], [@Stephenson:1989], [@Bothunetal:1991], and [@Totten/Irwin:1998], yielded 21 stars. Another 6 spectra were produced from slit spectra with the procedures described in Paper I.
[llccrrcccl]{} & & & & & &\
[\[-1.5ex\][Name]{}]{} & [\[-1.5ex\][HE Name]{}]{} & [\[-1.5ex\][$B_J$]{}]{} & [\[-1.5ex\][$B-V$]{}]{} & [\[-1.5ex\][$\mu_{\alpha}\Delta\,t$]{}]{} & [\[-1.5ex\][$\mu_{\delta}\Delta\,t$]{}]{} & C$_2$ & CN & All & Source
------------------------------------------------------------------------
\
CGCS 39 & HE 0017$+$0055 &(sat.)& & & & 1 & 1 & 1 & S89
------------------------------------------------------------------------
\
SKB 2 & HE 0039$-$2635 & 13.1 & 1.1 & & & 1 & 1 & 1 & SKB69\
BEM91 23 & HE 0100$-$1619 & 15.9 & 1.5 & & & 1 & 0 & 1 & BEM91\
CGCS 177 & HE 0106$-$2837 & 13.8 & 2.1 & & & 1 & 0 & 1 & S89\
SKB 5 & HE 0111$-$1346 & 13.3 & 1.4 & & & 1 & 1 & 1 & SKB69\
0207$-$0211 & HE 0207$-$0211 & 15.5 & 2.2 & 0.0 & 0.0 & 1 & 0 & 1 & TI98\
BEM91 08 & HE 0228$-$0256 & 16.2 & 2.0 & & & 1 & 0 & 1 & BEM91\
CGCS 525 & HE 0330$-$2815 & 13.8 & 1.5 & & & 0 & 1 & 1 & S89\
CGCS 935 & HE 0521$-$3425 & 13.0 & 1.3 & & & 1 & 1 & 1 & S89\
0915$-$0327 & HE 0915$-$0327 & 14.5 & 2.3 & 0.0 & 0.0 & 1 & 0 & 1 & TI98\
1019$-$1136 & HE 1019$-$1136 & 15.2 & 1.8 & 0.0 & 0.0 & 1 & 0 & 1 & TI98\
CGCS 2954 & HE 1104$-$0957 &(sat.)& & & & 1 & 1 & 1 & S89\
CGCS 3180 & HE 1207$-$3156 & 12.8 & 1.2 & & & 0 & 1 & 1 & S89\
CGCS 3274 & HE 1238$-$0836 &(sat.)& & & & 1 & 1 & 1 & S89\
1254$-$1130 & HE 1254$-$1130 & 16.1 & 2.2 & 0.0 & 0.0 & 1 & 0 & 1 & TI98\
1339$-$0700 & HE 1339$-$0700 & 15.0 & 1.7 & 0.0 & 0.0 & 1 & 0 & 1 & TI98\
1442$-$0058 & HE 1442$-$0058 & 17.8 & 2.2 & 0.0 & 0.0 & 1 & 0 & 1 & TI98\
CGCS 5435 & HE 2144$-$1832 & 12.6 & 1.4 & & & 0 & 1 & 1 & S89\
CGCS 5549 & HE 2200$-$1652 & 12.3 & 0.9 & & & 1 & 1 & 1 & S89\
2213$-$0017 & HE 2213$-$0017 & 16.4 & 2.4 & 0.0 & 0.0 & 1 & 0 & 1 & TI98\
2225$-$1401 & HE 2225$-$1401 & 16.5 & 2.9 & 0.0 & 0.0 & 1 & 0 & 1 & TI98\
CLS 50 & & & & 0.0 & 0.0 & 1 & 0 & – & Simul.\
CLS 31 & & & & 0.0 & 0.0 & 1 & 1 & – & Simul.\
CLS 54 & & & & 0.0 & 0.0 & 1 & 1 & – & Simul.\
KA 2 & & & & 0.0 & 0.0 & 1 & 1 & – & Simul.\
B1509$-$0902 & & & & 0.0 & 0.0 & 1 & 1 & – & Simul.\
UM 515 & & & & 0.0 & 0.0 & 1 & 0 & – & Simul.\
LHS 1075 & HE 0023$-$1935 & 16.1 & 1.4 & $-$[$0\mbox{$''\mskip-7.6mu.\,$}24$]{} & $-$[$10\mbox{$''\mskip-7.6mu.\,$}0$]{} & 0 & 0 & 0 & D94\
KA 2 & HE 1116$-$1628 & 16.6 & 1.3 & $-$[$0\mbox{$''\mskip-7.6mu.\,$}21$]{} & [$0\mbox{$''\mskip-7.6mu.\,$}24$]{} & 1 & 0 & 1 & D94\
G77-61 & HE 0330$+$0148 & 15.0 & 1.4 & [$1\mbox{$''\mskip-7.6mu.\,$}9$]{} & $-$[$7\mbox{$''\mskip-7.6mu.\,$}5$]{} & 0 & 0 & 0 & D94\
In our test, *all* 21 stars not known as dwarfs were selected either by their strong C$_2$ bands, or their CN bands. The simulated spectra were also *all* selected. Of the three dCs, one (KA 2) was selected, and the other two (G77–61, LHS 1075) not. From these results we conclude that our sample of giant C stars and dwarfs with low p.m. (e.g. dCs belonging to the disk population) is highly complete. From the small number of dCs in our test sample we are not able to draw any definitive conclusions, but our results suggest that only a minor fraction of the halo dCs are detected in the HES. This is consistent with $\sim 20$% of the simulated halo dCs being found (see Sect. \[sect:dC\_selfunct\]).
The Surface Density of C Stars
==============================
In an effective area of $6\,400\,\mbox{deg}^2$ [329 of 380 the HES plates; for a description how the effective area is estimated see @hespaperIII], we have isolated a total of 403 C stars. A straightforward estimate of the surface density of FHLC stars we detect with the HES is hence obtained from the ratio of these two numbers, yielding $0.063$deg$^{-2}$. However, one has to take into account that the effective area accessible on average for each object depends on its brightness, since the HES limiting magnitude varies from plate to plate. For example, the effective area for an object as faint as $B=17.0$ is only $73$% of the overall survey area, mainly because only 254 of the contributing plates reach this magnitude [see Fig. 2 in @hespaperIII]. An additional brightness dependence is caused by the fact that faint objects are more easily affected by overlapping spectra than bright objects. We therefore determine the FHLC surface density as follows: $$\mbox{surface density}=\sum_{i=1}^{403}\frac{1}{\mbox{effarea}(B_J)_i},$$ where $\mbox{effarea}(B_J)$ was determined as described in [@hespaperIII]. We obtain a FHLC surface density of $0.072\pm 0.005$deg$^{-2}$ on the 329 HES plates we used.
Discussion and Conclusions {#conclusions}
==========================
In an effective area of $6\,400\,\mbox{deg}^2$ we have isolated a total of 403 C stars. Our efforts have thus already increased the number of known FHLC stars by a factor of nearly five.
We find almost quadruple the surface density of carbon stars compared to the surveys summarized by [@Greenetal:1994]. Since those previous surveys claimed limiting magnitudes about $1.5$mag fainter than the HES, this highlights the greatly enhanced selection sensitivity of FHLC stars in the HES, which is more sensitive to a variety of C$_2$ or CN molecular absorption band strengths. Automated selection techniques may be superior to visible inspection of objective-prism spectra with binocular microscopes, as done e.g. in the survey of [@Sanduleak/Pesch:1988]. Photometric surveys for C stars have generally selected red objects only, which preferentially selects mostly the much less common high latitude AGB stars. [@Margonetal:2000] report a FHLC star surface density of “at least” $0.04$deg$^{-2}$ in the SDSS. This is still almost a factor of 2 below our value, and again, the SDSS is much deeper than the HES ($r'<19.5$).
Due to an average epoch difference of $13.5$ years between DSS I and HES plates, we expect to detect and select only $\sim 20$% of the halo dCs that we could detect and select if direct plates had been taken simultaneously with the HES plates. Our simulations indicate that 10–15 out of the 403 FHLCs published in this paper are dCs. Note, however, that this number is uncertain, because the kinematics of halo dCs is not precisely known. We estimate that additional $\sim 40$ dCs are detectable on the HES plates, but are currently missed due to the epoch difference problem. We are extending the current sample to include proper-motion corrected input catalogs for the extraction of HES spectra, to find *all* dCs, and other objects that can have large proper motions, like halo white dwarfs.
We thank D. Koester for providing model spectra of DQs, and C. Fechner for technical support in preparing this article. This work was partly supported by Deutsche Forschungsgemeinschaft under grant Re 353/40. P.J.G. acknowledges support through NASA Contract NAS8-39073 (ASC).
HES example spectra of C stars {#sect:Cstarexamples}
==============================
The HES FHLC sample
===================
In Tab. B.1 we list the sample of 403 HES FHLC stars described in this paper. The table is made available only electronically. It contains the following columns:
--------- ---------------------------------------------------
hename HE designation
ra2000 R.A. at equinox 2000.0, derived from DSS I
dec2000 Declination at equinox 2000.0, derived from DSS I
field ESO-SERC field number
plate HES plate number
q Plate quarter
objtyp Object type (stars/bright/ext)
B\_J $B_J$ magnitude
$V$ $V$ magnitude
$B-V$ $B-V$ magnitude, derived from HES spectra
$U-B$ $U-B$ magnitude, derived from HES spectra
C2idx1 Band index of C$_2$ 5165[Å]{}
C2idx2 Band index of C$_2$ 4737[Å]{}
CNidx1 Band index of CN 4216[Å]{}
CNidx3 Band index of CN 3883[Å]{}
selC2 C$_2$ band index selection flag
selCN CN band index selection flag
--------- ---------------------------------------------------
$B_J$ magnitudes are accurate to better than $\pm 0.2$mag, including zero point errors [@hespaperIII]. $V$ magnitudes were derived by the procedures described in Paper I. The object types “stars”, “bright” and “ext” refer to point sources, sources above a saturation threshold, and sources detected as extended in DSS I images, respectively. We do not list $V$, $B-V$ and $U-B$ for saturated objects, because our color calibrations are not valid for them.
[^1]: Based on observations collected at the European Southern Observatory, Chile (Proposal IDs 145.B-0009 and 63.L-0148). Table B.1 is only available in electronic form at the CDS via anonymous ftp to cdsarc.u-strasbg.fr (130.79.125.5) or via http://cdsweg.u-strasbg.fr/Abstract.html.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Waves excited on the surface of deep water decay in time and/or space due to the fluid viscosity, and the momentum associated with the wave motion is transferred from the waves to Eulerian slow currents by the action of the virtual wave stress. Here, based on the conservation of the total momentum, we found the virtual wave stress produced by calm gravity waves under the assumption that the slow Eulerian currents are weak in the sense that the Froude number is small and the scattering of surface waves by slow flow inhomogeneities can be neglected. In particular, we calculated the virtual wave stress generated by a propagating wave and two orthogonal standing waves. The obtained results possess Euler invariance, are consistent with previously known ones, and generalize them to the case of the excitation of almost monochromatic waves propagating in arbitrary directions.'
author:
- 'Vladimir M. Parfenyev'
- 'Sergey S. Vergeles'
title: 'Virtual wave stress in deep-water crossed surface waves'
---
Introduction
============
It was shown by Stokes that in a surface wave excited in an ideal fluid the Lagrangian particles possess a second-order drift velocity, which is usually called the Stokes drift [@stokes1847theory]. Later, Longuet-Higgins found that the fluid viscosity substantially modifies the Stokes prediction [@longuet1953mass]. The correction is associated with Eulerian slow current that corresponds to the mean velocity of fluid inside the bulk. The Stokes drift and the Eulerian slow current are very different. The Stokes drift is the result of nonlinear Lagrangian dynamics during one time period of oscillations and it does not produce any contribution into the mean velocity of the fluid (in the Euler description), while the slow current is excited by a force, which is localized in the narrow viscous sublayer near the fluid surface and is produced due to hydrodynamic nonlinearity (it is also known as the virtual wave stress; see Ref. [@longuet1969nonlinear]). Therefore, the dynamics of the slow current is relatively slow and it is determined by the fluid viscosity and inertia. In the stationary regime, the slow current is independent of fluid viscosity, even though it originates from the viscosity.
The origin of the force which excites the Eulerian slow current is similar to that of the force which produces the acoustic streaming in fluid during the propagation of a sound wave [@boluriaan2003acoustic]. Both forces are of the second-order in the wave amplitude and linear in the fluid viscosity. The acoustic streaming finds numerous applications in microfluidics [@friend2011microscale; @wiklund2012acoustofluidics], as it enables remote flow excitation and object manipulation. The standard theoretical approach to derive an equation describing the acoustic streaming flow is to average hydrodynamic equations over fast wave oscillations. The acoustic flow is excited near the boundaries inside the viscous sublayer, where the viscosity reduces the amplitude of the sound wave. Thus, the approach needs to resolve the viscous sublayer which is parametrically thinner compared to the typical spatial scale of the acoustic flow.
The same applies to the existing theoretical treatment of the excitation of a slow current by waves propagating on the free surface of a fluid [@longuet1953mass; @filatov2016nonlinear]. Nevertheless, there is an important difference between these phenomena. The boundaries that confine the fluid in acoustic experiments produce stresses acting on the fluid, whereas there is no external force acting on the free surface of the fluid. Thus, in the latter case, one can reformulate the hydrodynamic equations in the form of conservation laws and then utilize them to find the virtual wave stress that excites the current [@longuet1969nonlinear; @weber2001virtual].
In this paper, we propose the treatment of the virtual wave stress produced by calm gravity waves in the deep water approximation based on the momentum conservation law, which binds together the damping of surface waves due to the viscous dissipation and the nonlinear generation of the mean fluid velocity. Roughly saying, the momentum associated with the wave motion decreases together with the wave amplitude during the propagation of the surface wave due to the fluid viscosity. However, the total momentum must be conserved and it means that the wave attenuation gives rise to the force, which excites the additional fluid flow. We demonstrate that this force is applied near the fluid surface within the crest-trough layer, which includes the oscillating boundary of the fluid and the viscous sublayer under the boundary. The thickness of the crest-trough layer is small as compared to the scale of the slow current, so the force can be treated as surface stress. We obtain the explicit expression for it in terms of the excited wave motion. Our approach avoids the fine resolution of the viscous sublayer and allows us to obtain simple equations describing the slow currents under the assumption that they are weak in the sense that the Froude number is small enough and the scattering of surface waves by slow flow inhomogeneities can be neglected. In the case of a progressive wave excited on the surface of deep water, we can reproduce the Longuet-Higgins’ result [@longuet1953mass].
The initial interest in this problem was inspired by the recently observed phenomenon of nonlinear vorticity generation by crossed surface waves [@filatov2016nonlinear; @francois2017wave]. Using the established general expression for the virtual wave stress, we check that its curl corresponds to the boundary condition for the vertical vorticity used in Ref. [@filatov2016nonlinear]. On the whole, the developed approach allows one to look at the generation of slow currents by crossed surface waves from a new angle, reveals the physical nature of this phenomenon, and generalizes the previous results to the case of the excitation of almost monochromatic waves propagating in arbitrary directions.
Problem Formulation
===================
We consider an incompressible flow with a free surface that corresponds to the surface gravity waves excited against a background of a slow current. The wave motion has a characteristic frequency $\omega$ and its spectral width $\Delta \omega$ is assumed to be small, $\Delta \omega \ll \omega$. The axis $Z$ is directed vertically, opposite to the gravitational acceleration $\bm g$, and the fluid surface is determined by the equation $z=h(t,x,y)$ (it coincides with the plane $z=0$ at rest). The wave breaking is absent, the wave steepness is small, $|\nabla h| \ll 1$, and the deep water assumption for the wave motion is implied. We also assume that the fluid kinematic viscosity $\nu$ is small, $\gamma = \sqrt{\nu k^2/\omega} \ll 1$, where $k = \omega^2/g$ is a characteristic wave number. The viscosity of the fluid results in the fact that the wave motion ceases to be potential in a thin viscous sublayer of thickness $\delta \sim \gamma /k$ near the fluid surface. We represent the wave velocity as a sum of potential and vortical terms, $\bm u = \bm u^{\phi} + \bm u^{\psi}$, where $\bm u^{\phi} = \nabla \phi$ corresponds to the potential term and the vortical term $\bm u^{\psi}$ is parametrically smaller near the surface, $|\bm u^{\psi}| \sim \gamma |\bm u^{\phi}|$, and it is absent in the fluid bulk below the viscous sublayer [@lamb1975hydrodynamics]. Note that the wave amplitude can be either larger or smaller than the thickness $\delta$ of the viscous sublayer.
![Schematic of the slow current generation by a progressive surface wave in a slightly viscous fluid.[]{data-label="fig:0"}](p1.png){width="0.7\linewidth"}
Next, we denote by $\bm V$ the slow current, which is different from the slow potential contribution to the wave motion (it appears as a result of hydrodynamic nonlinearity and its amplitude is proportional to $\Delta \omega$, see Ref. [@longuet1963effect]), and so the velocity of fluid is $\bm v = \bm u + \bm V$. The characteristic time scale $T$ of the slow current $\bm V$ is much larger than the inverse wave frequency, i.e. $\omega T \gg 1$. Concerning the characteristic length scale $L$ of the slow flow $\bm V$, it does not always far exceed the wavelength, they can be of the same order, see, e.g., Ref. [@parfenyev2019formation]. Also, we assume that the fluid surface remains approximately flat if only the slow current is excited. This means that the Froude number $\mathrm{Fr}=V^2/(gL)$ for the slow current is small.
Our goal is to establish the influence of the wave motion on the slow current $\bm V$. There are two fundamentally different ways for this effect. The first way is due to the fluid viscosity and therefore it is forbidden in an ideal fluid. The wave motion attenuates and its momentum is transferred to Eulerian slow current by the action of the virtual wave stress [@longuet1969nonlinear; @weber2001virtual]. This stress $\bm \tau$ slowly changes in time, of the second-order in the wave amplitude $h$ and linear in the fluid viscosity $\nu$. Its amplitude can be estimated as the rate of decay of the surface density of momentum $|\bm \pi| = \rho \omega \langle h^2 \rangle$ in progressive wave (which is equal to the Stokes drift integrated over the fluid depth and multiplied by the fluid density $\rho$, see Ref. [@longuet1969nonlinear]), i.e. $|\bm \tau| \sim \nu k^2 |\bm \pi|$. Here and below angle brackets $\langle \cdots \rangle$ mean the extraction of slow harmonics with frequencies of the order of or less than $\Delta \omega$. The virtual wave stress $\bm \tau$ is applied near the fluid surface and therefore its action produces the slow current with non-zero vorticity, see Fig. \[fig:0\]. One can say that this vorticity is created in the viscous sublayer due to the fluid viscosity and hydrodynamic nonlinearity, and then it spreads downward in the fluid bulk through the viscous diffusion [@longuet1953mass; @parfenyev2019formation]. The concept of the virtual wave stress is inevitably associated with the fluid viscosity, since the generation of vorticity in an ideal fluid by the potential wave motion is forbidden due to Kelvin’s theorem [@falkovich2011fluid]. In this paper, we focus on finding the virtual wave stress $\bm \tau$ in the case of excitation of almost monochromatic waves propagating in arbitrary directions.
The second way of the influence is amplification of the vorticity associated with the already excited slow current $\bm V$ due to its interaction with the wave motion (see, e.g., paper [@ardhuin2017comments] and references therein, as well as recent experimental works [@savelyev2012turbulence; @d2014quantifying]). The fluid viscosity is not important here. The interaction between the slow current and the wave motion is localized at the wave penetration depth $\sim 1/k$, where the wave motion is potential. The effect is associated with the wave scattering on the inhomogeneities of the slow vortical flow $\bm V$ due to hydrodynamic nonlinearities [@phillips1959scattering]. In this paper, we assume that the effect is negligible as compared to the action of the virtual wave stress. The scattering in the fluid bulk is negligible if the condition $V/L \ll \nu k^2$ is satisfied, see Sec. \[sec:WCI\] below, and the scattering on a curved surface can be neglected under the additional condition ${\mathrm{Fr}}\ll \gamma^2$, see the explanation after equation (\[corr\]) in Sec. \[sec:WVS\].
Theoretical analysis of the problem is based on the use of integral forms of the continuity equation and the Navier-Stokes equation, which are respectively the laws of mass and momentum conservation. We introduce the momentum flux density tensor, $$\label{eq:Pi}
\Pi_{ij} = p \delta_{ij} + \rho v_i v_j - \rho \nu (\partial_i v_j + \partial_j v_i),$$ which is the $i$-th component of the amount of momentum flowing in unit time through unit area perpendicular to the $j$-axis [@landau1987course]. Here $p$ is the pressure, $\delta_{ij}$ is the Kronecker delta and the fluid is assumed to be incompressible, $\mathrm{div}\, \bm v = 0$. Now, we can write the mass conservation law $$\label{eq:incompress}
\partial_t[\theta(h-z) \rho] + \partial_j [\theta(h-z) \rho v_j] = 0,$$ and the Navier-Stokes equation $$\label{eq:Navier-Stokes}
\partial_t[\theta(h-z) \rho v_i] + \partial_j [\theta(h-z) \Pi_{ij}] = -\delta_{iz} \theta(h-z) \rho g,$$ where $\theta(h-z)$ is the Heaviside step function and we sum over the repeated Latin indices that run through the values $\{x,y,z\}$. Note that both these equations are applicable in the whole space and contain exact boundary conditions, which should be found from the requirement that the coefficients before the Dirac delta function $\delta(h-z) = \theta'(h-z)$ are equal to zero, see Appendix \[sec:A\] for detail.
In the fluid bulk, below the viscous sublayer, the wave motion is potential, and since we have neglected its scattering on the inhomogeneities of the slow current, the wave motion cannot change the vorticity of the flow. It means that the slow vortical flow $\bm V$ is described by the usual Navier-Stokes equation in the fluid bulk $$\label{eq:Navier-Stokes-V}
\partial_t \bm V + (\bm V \nabla) \bm V + \nabla P/\rho - \nu \nabla^2 \bm V = 0,$$ supplemented by the incompressibility condition $\mathrm{div}\,\bm V = 0$, which follows from equation (\[eq:incompress\]). Here $P$ is contribution to the pressure, associated with the slow current $\bm V$. Note that due to the incompressibility condition $\nabla^2 P = - \rho (\partial_i V_j) (\partial_j V_i)$.
The Navier-Stokes equation (\[eq:Navier-Stokes-V\]) for the slow current $\bm V$ should also be supplemented by the boundary conditions. Since we have assumed that the fluid surface remains flat if only the slow current is excited, these conditions can be posed at fixed virtual boundary $z = 0$ corresponding to the unperturbed fluid surface. As was explained earlier, due to the wave motion and the fluid viscosity, the virtual wave stress $\bm \tau$ is applied to this boundary and so it is not free. The presence of full divergency in equation (\[eq:Navier-Stokes\]) allows one to exploit the momentum conservation law in integral form for crest-trough layer which is only partially filled with the fluid, see Fig. \[fig:0\], and obtain the explicit expression for the virtual wave stress in terms of the excited wave motion. Details of the calculation are presented in Sec. \[sec:WVS\].
Wave-Current Interaction in the Fluid Bulk {#sec:WCI}
==========================================
Before proceeding to the calculation of the virtual wave stress, we discuss the condition when the wave-current interaction can be neglected. The condition is equivalent to the requirement that the scattering of wave motion by the slow flow inhomogeneities due to the hydrodynamic nonlinearity is small as compared with the viscous wave damping. Here we consider the influence of nonlinearity in the fluid volume. If the scattering is weak, the wave motion is potential below the viscous sublayer and the velocity of fluid is equal to $\bm v^0 = \bm u^{\phi} + \bm V$. In general case, the Navier-Stokes equation (\[eq:Navier-Stokes\]) in the fluid bulk has the form $$\label{eq:Navier-Stokes-uV}
\rho \partial_t v_i = -\partial_j (\Pi_{ij}^0 + \delta \Pi_{ij}) - \delta_{iz} \rho g,$$ where the momentum flux $\Pi_{ij}^0$ corresponds to the velocity field $\bm v^0$ and $\delta \Pi_{ij} = \Pi_{ij} - \Pi_{ij}^0$. First, we consider the terms in equation (\[eq:Navier-Stokes-uV\]) that have a characteristic frequency of $\omega$. The scattering of the wave motion on the inhomogeneities of the slow current $\bm V$ corresponds to the term $\rho u_j^{\phi} \partial_j V_i$ in $\partial_j \Pi_{ij}^0$, which leads to the deviation of the wave flow $\bm u$ from the potential flow $\bm u^{\phi}$ at depth of the order of $1/k$. This difference can be estimated as $\bm u^{scat} \sim h(V/L)$, and let us stress that the vortical correction $\bm u^{scat}$ is localized at the depth of $1/k$ and has nothing common with the vortical correction $\bm u^{\psi}$ localized in the viscous boundary sublayer near the fluid surface.
Next, the vortical correction $\bm u^{scat}$ produces contribution in the average value of the momentum flux difference $\langle \delta \Pi_{ij} \rangle$, which can be estimated as $\rho \langle u^{\phi} u^{scat} \rangle \sim \rho \omega \langle h^2 \rangle (V/L)$. The influence of this additional term on the flow can be neglected, if it is less than the virtual wave stress $|\bm \tau| \sim \rho \nu \omega k^2 \langle h^2 \rangle$. Thus, we obtain the condition for the slow current gradient $V/L \ll \nu k^2$, which is assumed to be fulfilled in this paper. The condition is equivalent to the requirement that the wave scattering length on slow current inhomogeneities is greater than the propagation length of the wave $l_\nu \sim \omega/(\nu k^3)$, which is determined by viscous damping. Indeed, one has $|\nabla {\bm V}| l_\nu \ll c_g$, where $c_g=\omega/(2k)$ is the group velocity of the waves, that is the variation of velocity ${\bm V}$ is negligible for a propagating wave.
Virtual Wave Stress {#sec:WVS}
===================
The bulk equation (\[eq:Navier-Stokes-V\]) has to be supplemented by three boundary conditions for the velocity field $\bm V$ posed at a fixed virtual boundary $z = 0$ corresponding to the unperturbed fluid surface. The vertical velocity can be estimated as divergence of Stokes mass transport for progressive wave, $V_z\vert_{z=0} \sim \nu k^3 \langle h^2 \rangle$, and then in the leading order the first boundary condition is $V_z\vert_{z=0} \approx 0$ (see also Sec. \[sec:Ex\] for examples and Sec. \[sec:discussion\] for a more thorough analysis). To obtain other boundary conditions, we introduce the Heaviside step function for a fixed virtual boundary $\theta^0 \equiv \theta(-z)$ and for the real boundary $\theta \equiv \theta(h-z)$, and consider the horizontal component of the Navier-Stokes equation (\[eq:Navier-Stokes\]), which can be rewritten as $$\label{eq:VWS_eq}
\partial_t(\theta^0 \rho v_{\alpha}^0) + \partial_j ( \theta^0 \Pi_{\alpha j}^0 ) = -\rho \partial_t (\delta v_{\alpha}) - \partial_j (\delta \Pi_{\alpha j}).$$ Here and below Latin indices take the values $\{x,y,z\}$ and Greek indices take only $\{x,y\}$, we sum over the repeated indices, $\delta {\bm v} = \theta {\bm v}-\theta^0{\bm v}^0$, $\delta \Pi_{ij} = \theta \Pi_{ij} - \theta^0\Pi^0_{ij}$ and the momentum flux $\Pi_{ij}^{0}$ corresponds to velocity field ${\bm v}^0 = \bm u^{\phi} + \bm V$. The left-hand side of this equation corresponds to the Navier-Stokes equation with the fixed flat virtual boundary and a purely potential flow in surface waves. The virtual boundary partially extends beyond the fluid, therefore we do an analytic continuation of the velocity ${\bm v}^0={\bm V}+{\bm u}^\phi$ toward the boundary. Since the boundary $z=0$ is virtual and not real, the left-hand side of equation (\[eq:VWS\_eq\]) is not zero. It is equal to $-\delta(z) \Pi_{\alpha z}^0$ and this imbalance is compensated by the right-hand side, which is non-zero only near the fluid surface, inside the region $|z|<\varepsilon$, see Fig. \[fig:1\]. The constant $\varepsilon$ is much less than the wavelength, $k \varepsilon \ll 1$, but the plane $z=-\varepsilon$ is always below the fluid surface and the vortical part of velocity associated with waves $\bm u^{\psi}$ is always negligible at the depth $z=-\varepsilon$. Such a constant exists because we assumed that $kh \ll 1$ and $k \delta \ll 1$.
Next, we average equation (\[eq:VWS\_eq\]) over the wave oscillations and approximate the right-hand side as $\delta(z) F_\alpha$. The effective boundary conditions imposed on the virtual boundary should be obtained by equating the overall coefficient before $\delta(z)$ in relation (\[eq:VWS\_eq\]) to zero, i.e. $\langle\Pi^0_{\alpha z}\rangle\big\vert_{z=0}=-F_\alpha$. To simplify calculations, we choose an inertial reference frame in which the horizontal components of the slow current are equal to zero near the surface at a given position and time, $V_\alpha\vert_{z=0}=0$. In this reference frame, not only the gradient of the slow current is small, but also its absolute value, and therefore its interaction with the wave motion and with itself can be neglected. Below, in Sec. \[sec:discussion\], we will discuss how to take into account additional terms that correspond to advection by a constant horizontal velocity associated with a moving reference frame, and thereby restore the Euler invariance inherent in the original equations. Note also that the considered case corresponds to the initial stage of the slow current generation by surface waves, if it was initially absent.
![A change in the horizontal component of the wave motion momentum generates a surface force $\tau_{\alpha}$ (virtual wave stress) that excites a slow flow.[]{data-label="fig:1"}](p2.png){width="0.7\linewidth"}
To find $F_\alpha$, one needs to integrate the right-hand side of time-averaged equation (\[eq:VWS\_eq\]) in $Z$-direction, after which we obtain $$\label{BC-project}
\langle \Pi_{\alpha z}^V + \Pi^\phi_{\alpha z}\rangle \big\vert_{z=0} = \partial_\beta \int\limits_{-\infty}^{+\infty}\mathrm{d}z \left\langle\delta \Pi_{\alpha \beta}\right\rangle + \partial_t\int\limits_{-\infty}^{+\infty}\mathrm{d}z\langle\rho\,\delta v_\alpha\rangle,$$ where $\Pi_{\alpha z}^0 = \Pi_{\alpha z}^V + \Pi^\phi_{\alpha z}$, and $\Pi^V_{\alpha z}$ and $\Pi^\phi_{\alpha z}$ are momentum fluxes corresponding to velocity fields ${\bm V}$ and ${\bm u}^\phi$ respectively. Note that during subsequent calculations, we should keep only linear contributions in viscosity, which are leading in the parameter $\gamma \ll 1$, and should limit our analysis to second-order in wave amplitude, following the mentioned estimates for the virtual wave stress. All higher-order corrections should be neglected. We can also skip all linear in the wave amplitude contributions, because the wave velocity field satisfies the linear equations and these terms will cancel each other. Also note, that we may not think about terms at a double frequency $2 \omega$ produced due to hydrodynamic nonlinearity, since they are separated from all equations and boundary conditions and form a closed subsystem of equations. Accordingly, we can put $\bm u^{\phi} = \nabla (\phi^{(1)}+\phi^{(2)})$, where $\phi^{(1)}$ is the wave potential in the linear approximation and $\phi^{(2)}$ is the slow second-order correction proportional to $\Delta \omega$, see Ref. [@longuet1963effect]. Some useful well-known expressions for the wave motion are summarized in Appendix \[sec:B\].
Now we proceed to the analysis of the left-hand side of equation (\[BC-project\]). In our reference frame $V_\alpha\vert_{z=0}=0$, and therefore we obtain $\Pi^V_{\alpha z}\vert_{z=0} = -\rho \nu \partial_z V_\alpha\vert_{z=0}$. We have neglected $\rho \nu\partial_\alpha V_z\vert_{z=0}$ because $V_z\vert_{z=0} \sim \nu k^3 \langle h^2 \rangle$ and the term contains an additional smallness in parameter $\gamma^2 \ll 1$. Next, the mean value of the momentum flux associated with the potential wave motion is equal to $$\begin{aligned}
\label{Pi_through_phi}
\left\langle
\Pi^{\phi}_{\alpha z}
\right\rangle\big\vert_{z=0}
=
\rho
\left\langle
\partial_\alpha\phi^{\scriptscriptstyle (1)} \,\partial_z\phi^{\scriptscriptstyle (1)}
\right\rangle\big\vert_{z=0}
-2\rho\nu \partial_{\alpha z}\langle \phi^{\scriptscriptstyle (2)}\rangle \big\vert_{z=0}.\end{aligned}$$ Since the last term contains an explicit factor $\nu$, the velocity potential $\langle \phi^{\scriptscriptstyle (2)}\rangle$ should be found in the limit of an ideal fluid without the viscous correction. According to Ref. [@longuet1963effect], we can estimate $\partial_{\alpha z}\langle \phi^{\scriptscriptstyle (2)}\rangle \sim \Delta \omega k^2 \langle h \rangle^2$ and therefore $\rho\nu\partial_{\alpha z}\langle \phi^{\scriptscriptstyle (2)}\rangle/|{\boldsymbol \tau}| \sim \Delta\omega/\omega \ll 1$ and the term should be neglected. Note also, that the potential $\phi^{\scriptscriptstyle (2)}$ satisfies Laplace equation and boundary conditions $\nabla \phi^{\scriptscriptstyle (2)} \to 0$ when $z\to-\infty$ and $(\partial_t^2+g\hat k)\phi^{\scriptscriptstyle (2)}= -\partial_t (\nabla \phi^{\scriptscriptstyle (1)})^2$ at $z=0$, where $\hat k = \sqrt{-\partial_{\alpha} \partial_{\alpha}}$ is the wave number operator. The presence of the full-time derivative in the last boundary condition leads to the fact that $\partial_{\alpha z}\langle \phi^{\scriptscriptstyle (2)}\rangle \vert_{z=0}$ itself is the full-time derivative of limited in time quantity. This means that the time integral of this quantity is also limited. Thus, the last term in equation (\[Pi\_through\_phi\]) cannot lead to the excitation of a significant slow current at long times. Finally, let us verify that the cross-contribution to the momentum flux $\Pi^{V\!\text{-}\phi}_{\alpha z}$ can be neglected. The contribution is $\rho V_z\langle u^{\scriptscriptstyle (2)}_\alpha\rangle\vert_{z=0} \sim \rho\nu \Delta\omega k^4 \langle h^2\rangle^2$ and it is of the fourth-order in wave amplitude.
Next, we turn out to the analysis of the right-hand side of equation (\[BC-project\]). The last term without $\partial_t$ is the horizontal momentum surface density $\pi_\alpha$ for the wave motion $$\label{pi_a}
\pi_\alpha
=
\rho\left\langle\int\limits^{h}_{0} \mathrm{d}z\,u^\phi_\alpha + \int\limits^{h}_{-\infty} \mathrm{d}z\,u^\psi_\alpha\right\rangle
=
\rho\langle h \,\partial_\alpha \phi^{\scriptscriptstyle (1)}\rangle \big\vert_{z=0}.$$ The vortical part of the velocity $u^\psi_\alpha$ produces $\rho \langle \psi_\alpha^{\scriptscriptstyle (1)} \vert_{z=h}\rangle$, which should be neglected due to time averaging, see equation (\[psi\]). Note that in our analysis, the second-order slow vortex contribution $\psi_\alpha^{\scriptscriptstyle (2)}$ generated by waves due to hydrodynamic nonlinearity is included in the definition of the slow flow $\bm V$, so there are no additional terms in the equation. In the case of plane wave in an ideal fluid, $\pi_{\alpha}$ corresponds to the mass-transport through the wave total cross-section, since the corresponding integral is accumulated at the fluid surface in the Euler description, see Ref. [@longuet1969nonlinear]. The time derivative takes into account the possible time decay of this quantity due to the fluid viscosity.
The first term in the right-hand side of equation (\[BC-project\]) without $\partial_\beta$ is surface density of the horizontal momentum flux $\pi_{\alpha\beta}$, and the spatial derivative takes into account the possible spatial decay of this quantity due to the fluid viscosity, see also Ref. [@weber2001virtual], $$\label{pi_alpha-beta}
\pi_{\alpha\beta}
=
\left\langle \int_0^{h}
\big(p_u^{\scriptscriptstyle (1)}\delta_{\alpha\beta}
-
\rho \nu (\partial_\alpha u^\phi_\beta + \partial_\beta u^\phi_\alpha)\big)
\mathrm{d}z\right\rangle
+
\left\langle
\int_{-\infty}^{h}
\Big(\rho(u^\psi_\alpha u^\phi_\beta + u^\psi_\beta u^\phi_\alpha)
+
p_u^\psi \delta_{\alpha\beta}\Big)
\mathrm{d}z
\right\rangle.$$ Here, we neglected the second-order contributions in the wave amplitude that contain the small factor $\gamma^3$ and denoted by $p_u^\psi $ the pressure associated with the vortex flow $\bm u^\psi$, which is non-zero only inside the viscous sublayer. As it turns out, the second term in equation (\[pi\_alpha-beta\]) should also be neglected, and the first term gives the result $$\label{pi_ab_found}
\pi_{\alpha\beta}
=
-\rho\left\langle h\,\partial_t \phi^{\scriptscriptstyle (1)}\vert_{z=0} + gh^2/2\right\rangle\delta_{\alpha\beta}
-
2\rho \nu \left\langle h\, \partial_{\alpha\beta} \phi^{\scriptscriptstyle (1)}\right\rangle\vert_{z=0},$$ where we have used Bernoulli equation (\[Bernoulli\]) in the linear approximation, which is valid for the viscous fluid as well, see Ref. [@lamb1975hydrodynamics §349].
To justify the neglect of the second term in equation (\[pi\_alpha-beta\]), we first consider the term containing the pressure $p_u^\psi$. The contribution satisfies $\nabla^2 p_u^\psi = -\rho [\partial_i u_k \partial_k u_i - (\partial_{ik}\phi)^2]$, and the right-hand side of this equation can be estimated as $\rho (\omega k h)^2 $ and it is localized in the viscous sublayer. Below it, the wave motion is potential and satisfies the Euler equation, which can be integrated, leading to the Bernoulli equation (\[Bernoulli\]) in the fluid bulk. Thus, the correction to pressure $p_u^{\psi}$ is non-zero only inside the thin viscous sublayer and it can be estimated as $\langle p_u^\psi\rangle \sim \gamma^2 \rho \omega^2 \langle h \rangle^2$. The integration across the viscous sublayer in relation (\[pi\_alpha-beta\]) produces one more factor $\gamma$, which makes the contribution negligible. Second, let us analyze the remaining contribution to the second term of equation (\[pi\_alpha-beta\]), which is equal to $\delta\pi_{\alpha\beta} = \rho\int^h\mathrm{d}z \langle\partial_{z}\psi_\alpha^{\scriptscriptstyle (1)}\,\partial_\beta \phi^{\scriptscriptstyle (1)} + \partial_{z}\psi_\beta^{\scriptscriptstyle (1)}\,\partial_\alpha\phi^{\scriptscriptstyle (1)}\rangle = \rho \langle \psi_\alpha^{\scriptscriptstyle (1)} \, \partial_\beta\phi^{\scriptscriptstyle (1)} + \psi_\beta^{\scriptscriptstyle (1)} \, \partial_\alpha\phi^{\scriptscriptstyle (1)}\rangle\big\vert_{z=h}$. It follows from equation (\[psi\]) and the wave equation (\[wave\_equation\]) that $\psi_\alpha^{\scriptscriptstyle (1)}\vert_{z=h} = (2\nu/g)\partial_{t\alpha}\phi^{\scriptscriptstyle (1)}\vert_{z=0}$ in the limit of small viscosity, and hence the contribution is the full-time derivative, $\delta\pi_{\alpha\beta}= (2\rho\nu/g) \partial_t \langle \partial_\alpha\phi^{\scriptscriptstyle (1)} \,\partial_\beta \phi^{\scriptscriptstyle (1)} \rangle\vert_{z=0} \sim \rho \nu \, \Delta\omega\,k\langle h^2\rangle $. This expression is small as $\Delta \omega/\omega \ll 1$ compared to the last term in equation (\[pi\_ab\_found\]), and it should be neglected as it was done for the last term in relation (\[Pi\_through\_phi\]).
Now we consider the remaining terms in the right-hand side of equation (\[BC-project\]) that have not yet been discussed and show that they can also be neglected. These terms are the result of a separate averaging of either the upper integration limit $h$ or the integrand. Because the corrections arising from the vortical part $\bm u^{\psi}$ of wave flow inside the viscous sublayer have already been taken into account and neglected, we can replace $ \Pi_{ij} $ with $\Pi^0_{ij}$ in the right-hand side of equation (\[BC-project\]) and then the integrand in it becomes equal to $$\begin{aligned}
\nonumber
(\theta-\theta^0)(\rho \partial_t v^0_\alpha + \partial_\beta\Pi^0_{\alpha \beta} )
+
\delta(z-h) (\Pi_{\alpha\beta}^0 \partial_\beta h + \rho v^0_\alpha \partial_t h)
=
\\[5pt]\label{corr0}
=
-(\theta-\theta^0)\partial_z\Pi^0_{\alpha z}
+
\delta(z-h) (\Pi_{\alpha\beta}^0 \partial_\beta h + \rho v^0_\alpha \partial_t h).
\hskip20pt\end{aligned}$$ The equality in equation (\[corr0\]) is valid since nonlinear interaction between waves and slow current can be neglected in the fluid bulk, and we remind that under the assumption $\langle \Pi^0_{\alpha z}\rangle = \Pi^V_{\alpha z} + \langle \Pi^\phi_{\alpha z}\rangle$. Since we have already taken into account the correlations between the upper limit of integration $h$ and integrable in the right-hand side of equation (\[BC-project\]), the remaining terms are $$\begin{aligned}
\nonumber
&-\langle h\rangle \partial_z \Pi^V_{\alpha z}\vert_{z=0}
-
h^V \partial_z\langle \Pi^\phi_{\alpha z}\rangle\vert_{z=0}
+
\langle \Pi^\phi_{\alpha\beta}\vert_{z=h}\rangle \partial_\beta h^V+&
\\[5pt]\label{corr}
&+ \Pi^V_{\alpha\beta}\vert_{z=h^V}\partial_\beta \langle h^{(2)}\rangle
+ \rho \langle u_\alpha^\phi\vert_{z=h} \rangle \partial_t h^V.&\end{aligned}$$ Here $\langle h \rangle \approx h^V + \langle h^{\scriptscriptstyle (2)} \rangle$, where $h^V$ is the surface elevation corresponding to the slow current in the absence of waves and $\langle h^{\scriptscriptstyle (2)} \rangle$ is produced by the slow second-order wave motion $\phi^{\scriptscriptstyle (2)}$. We kept only the part $h^V$ from $\langle h\rangle$ in several terms in the right-hand side of relation (\[corr\]) to omit all contributions proportional to $h^4$. The first term can be estimated as $(\langle h\rangle/L)\Pi^V_{\alpha z}$, and it is small as compared to the term $\Pi^V_{\alpha z}$, which is already taken into account in the left-hand side of equation (\[BC-project\]). The third term in relation (\[corr\]) is small as compared to the virtual wave stress if the condition $\mathrm{Fr}\ll \gamma^2$ is fulfilled. Physically this means that the wave scattering on the curved surface is negligible. Under the condition, $h^V$ can be considered constant in the second term, so the sum of this term with the second term in the left-hand side of equation (\[BC-project\]) means that $h^V$ simply changes the zero elevation point for fluid when describing the surface wave motion. Considering the fourth term, one can estimate $\Pi^V_{\alpha\beta}\sim \rho \nu V/L$, including the contribution from pressure which can be estimated from the boundary condition $P = 2 \rho \nu \partial_z V_z$. Therefore, the term can be neglected as compared to the left-hand side of equation (\[BC-project\]) due to $\partial_\beta \langle h^{(2)}\rangle \ll 1$. The last term has a similar nature to the second one. It takes into account that the fluid surface can move upwards as a whole with the local velocity $\partial_t h^V$, and it can be eliminated if we choose the corresponding reference frame.
Finally, collecting all the contributions together, we can write equation (\[BC-project\]) in the form $$\label{BC}
\rho \nu \partial_z V_\alpha\vert_{z=0}
=
-\partial_\beta\pi_{\alpha\beta}
+\langle \Pi^\phi_{\alpha z}\rangle\vert_{z=h^V}
-\partial_t \pi_\alpha
\equiv
\tau_\alpha,$$ which is illustrated in Fig. \[fig:1\]. It is convenient to express the final result in terms of the wave elevation, and by substituting relations (\[Pi\_through\_phi\],\[pi\_a\],\[pi\_ab\_found\]) in equation (\[BC\]) and using wave equation (\[wave\_equation\]) and relation (\[flow\_linear\]), we obtain $$\label{eq:tau}
\tau_\alpha = -2 \rho \nu \left\langle k^{-1} (\partial_\beta \partial_t h) (\partial_{\alpha} \partial_{\beta} h) + k (\partial_{\alpha} h) (\partial_t h) \right\rangle.$$ As we originally expected, the expression for the virtual wave stress is proportional to the viscosity and the square of the amplitude of the wave motion. Note that the term $-\partial_t\pi_\alpha$ in equation (\[BC\]) is the full-time derivative. And although its amplitude is not small as compared to $|{\boldsymbol \tau}|$ in the general case, it can be neglected for long times, since it cannot generate a significant slow current. However, taking this term into account, we can clearly show that the virtual wave stress (\[eq:tau\]) is proportional to the viscosity.
To summarize, the bulk equation (\[eq:Navier-Stokes-V\]) has to be supplemented by the boundary conditions $$\label{eq:BCBC}
\rho \nu \partial_z V_{\alpha} \vert_{z=0} = \tau_{\alpha}, \quad V_z\vert_{z=0} = 0,$$ posed at the unpertubed fluid surface $z=0$, where the virtual wave stress $\tau_{\alpha}$ is defined by equation (\[eq:tau\]). These relations imply the fulfillment of two inequalities: $V/L \ll \nu k^2$ and $V^2/(gL) \ll \gamma^2$, which mean the weakness of the slow flow. Note that the divergence of the virtual wave stress is zero, $\partial_\alpha \tau_\alpha = 0$, and the curl of both sides of equation (\[eq:BCBC\]) gives the boundary condition for $Z$-derivative of the vertical vorticity (the part which spreads outside the viscous sublayer) that was previously obtained in Ref. [@filatov2016nonlinear Eq. (12)].
Examples {#sec:Ex}
========
To illustrate our result let us find the slow flow generated by a propagating wave [@longuet1953mass] and two orthogonal standing waves [@parfenyev2019formation]. Following these references, we assume that the wave motion is stationary in time due to external pumping and the resulting slow current $\bm V$ is rather weak, so the nonlinear term in the Navier-Stokes equation (\[eq:Navier-Stokes-V\]) can be neglected. In the first case, the wave elevation is $$\label{eq:h1}
h(t,x) = H \cos(kx-\omega t),$$ where $H=H_0 \exp(-4 \gamma^2 kx)$ decays in space due to the fluid viscosity. By using equation (\[eq:tau\]), we find that the surface force density exciting the slow current is equal to $\tau_x = 2 \rho \nu \omega (kH)^2$. Therefore, we obtain that at the virtual boundary $$\label{bc1}
\partial_z V_x\vert_{z=0} = 2 \omega (kH)^2, \quad V_z\vert_{z=0} = 0.$$ Next, the slow flow penetrates the fluid bulk due to viscous diffusion and ultimately reaches the bottom of the system, which is located at $z=-d$. To find the stationary solution we need to impose here the usual no-slip boundary condition $$\label{bc2}
V_x\vert_{z=-d}=0, \quad V_z\vert_{z=-d}=0.$$ Note that $d \gg 1/k$ to satisfy the deep-water approximation for surface waves. Nevertheless, the slow current reaches the bottom in the stationary regime and therefore its position plays an important role.
To simplify the problem, we suppose that the fluid depth $d$ is much less than the wave propagation length $l_{\nu} = 1/(4 \gamma^2 k)$. In this case, in the main approximation with respect to parameter $d/l_{\nu} \ll 1$, one can assume that the virtual wave stress $\tau_x$ acting on the fluid surface is constant. This means that the considered system is homogeneous along the direction of wave propagation and therefore nothing depends on the $x$-coordinate. Following Ref. [@longuet1953mass], we also assume that total horizontal transport is zero, $$\label{eq:int}
\int_{-d}^{0} dz \, V_x(z) = 0.$$ As one can see later, the correction to this expression associated with the Stokes drift can be neglected since $d \gg 1/k$.
Now, let us proceed to calculations. The incompressibility condition leads to $\partial_z V_z = 0$ and due to the condition $V_z(-d)=0$ this means that $V_z = 0$. Next, from $z$-component of equation (\[eq:Navier-Stokes-V\]) we obtain $\partial_z P = 0$, which means that pressure $P$ does not depend on $z$-coordinate. Considering $x$-component of equation (\[eq:Navier-Stokes-V\]) and using boundary conditions (\[bc1\]) and (\[bc2\]) together with expression (\[eq:int\]), we find the stationary solution $$\label{eq:result}
V_x(z) = \dfrac{\omega (kH)^2 d}{2} \left( \frac{3z^2}{d^2} + \frac{4z}{d} +1 \right).$$ The presented derivation leads to the same result as in Ref. [@longuet1953mass Eq.(305)]. Note that the applicability condition $V/L \ll \nu k^2$ is equivalent to $H \ll \delta$, i.e. the wave amplitude should be much smaller than the thickness of the viscous sublayer. In the opposite case, the expression (\[bc1\]) is valid only at the initial stage. In the course of further evolution, the slow flow becomes so strong that it is necessary to take into account its interaction with the waves.
In the second example, we consider the slow current generated by two orthogonal standing surface waves [@parfenyev2019formation]. The surface elevation is given by $$\label{eq:h2}
h = H_1 \cos(\omega t) \cos (kx) + H_2 \cos(\omega t + \vartheta) \cos (ky),$$ where $H_1$ and $H_2$ are the amplitudes of the waves, and $\vartheta$ is the phase shift between them. As in the previous case, one does not need to take into account viscous corrections to this expression corresponding to the wave spatial decay, since they will produce parametrically smaller contribution to the generated slow flow.
Following Ref. [@parfenyev2019formation], we will describe the corresponding slow current in terms of the vertical vorticity, $\Omega_E = \partial_x V_y - \partial_y V_x$. Using relations (\[eq:Navier-Stokes-V\]) and (\[eq:BCBC\]) and neglecting the nonlinear term in the Navier-Stokes equation, one finds the bulk equation $$\label{eq:vorticity}
\partial_t \Omega_E - \nu \nabla^2 \Omega_E = 0,$$ which has to be supplemented by the boundary conditions $$\label{eq:bc_crossed}
\rho \nu \partial_z \Omega_E\vert_{z=0} = \epsilon_{\alpha \beta} \partial_{\alpha} \tau_{\beta}, \quad \Omega_E\vert_{z \rightarrow -\infty}=0,$$ where $\epsilon_{\alpha \beta}$ is unit antisymmetric tensor and the Greek indices run over $\{x,y\}$. By substituting the wave elevation (\[eq:h2\]) to equation (\[eq:tau\]), we obtain $$\epsilon_{\alpha \beta} \partial_{\alpha} \tau_{\beta} = -2 \rho \nu \omega k^3 H_1 H_2 \sin (kx) \sin(ky) \sin \vartheta.$$ Therefore, equation (\[eq:vorticity\]) with boundary conditions (\[eq:bc\_crossed\]) literally coincides with the equations (7) and (8) in Ref. [@parfenyev2019formation] (under the assumption of zero compression modulus of the surface film) and has exactly the same solution. In the stationary regime, one finds $$\Omega_E = - \sqrt{2} e^{kz\sqrt{2}} H_1 H_2 \omega k^2 \sin (kx) \sin (ky) \sin \vartheta.$$ The applicability condition $\Omega_E \ll \nu k^2$ leads to the same restriction $H \ll \delta$ for the wave amplitude in the stationary regime.
The advantage of our method is its relative simplicity due to the lack of the need to resolve the fine details of the viscous sublayer, since it does not produce any contribution during calculations. Moreover, our approach has a clear physical meaning because it is based directly on the momentum conservation law, which makes the phenomenon of generation of slow current by surface waves similar to, for example, the radiation pressure of light.
Discussion {#sec:discussion}
==========
In this section, we would like to discuss some additional issues. The first issue is related to the boundary condition for the vertical velocity $V_z$ of slow current under the assumption that the fluid surface remains approximately flat if only the slow current is excited. The second issue is related to the fact that up to this point we have solved the problem in a special reference frame in which the slow current is small. The initial equations possess Euler invariance, and therefore, our theoretical scheme can be generalized to an arbitrary reference frame moving with some horizontal velocity.
To obtain the boundary condition at the virtual boundary $z=0$ for the vertical velocity $V_z$, we consider the continuity equation (\[eq:incompress\]) in the framework of the same approach that we used earlier to obtain and analyze the equation (\[eq:VWS\_eq\]). First, we rewrite the continuity equation in the form $$\label{continuity0}
\partial_t(\theta^0\rho) + \partial_i(\theta^0\rho v^0_i)
=
-\partial_t\big((\theta-\theta^0)\rho\big)
-\partial_i(\rho \,\delta v_i),$$ and let us remind that we are still working in the reference frame where $V_\alpha\vert_{z=0}=0$ at a given position and time. Next, we proceed according to the scheme that led to the boundary condition (\[eq:tau\]). We average equation (\[continuity0\]) over fast oscillations. Then we approximate the right-hand side of this equation as $\delta(z)Q$ and equate the total coefficient before $\delta(z)$ to zero. As a result, we obtain equation $\rho \langle v_z^0\rangle\vert_{z=0} = -Q$, which leads to $$\label{VzEq}
V_z\vert_{z=0}
=
\left(\partial_t\langle h^{(2)}\rangle
-
\langle u_z^{(2)}\rangle \vert_{z=h^V}
+
\frac{\partial_\alpha\pi_\alpha}{\rho}\right)
+
\partial_t h^V.$$ The round bracket is equal to zero in the case of an ideal fluid. One can check this using the results for $ h^{(2)}$ and $u_z^{(2)}=\partial_z \phi^{(2)}$ obtained in Ref. [@longuet1963effect]. However, the bracket becomes non-zero and proportional to the viscosity for real fluid. We assume that $\phi^{(2)}$ is the full-time derivative not only for an ideal but also for a viscous fluid as well. In particular, $\phi^{(2)}$ should vanish in the limit $\Delta\omega\to0$. Then the role of the first and the second terms in the round bracket is analogous to the role of term $\partial_t\pi_\alpha$ in relation (\[BC\]): these terms compensate full-time derivative which is contained in the third term. We do not need to calculate these terms explicitly since only the last term in the round bracket contains the contribution that is not a full-time derivative, and therefore only this contribution is of interest. We use relations (\[wave\_equation\],\[flow\_linear\]) to calculate the contribution and obtain the boundary condition $$\label{BCz}
V_z\big\vert_{z=0} - \partial_t h^V = -2\nu k \left\langle (\partial_\alpha h)^2 + 3 (kh)^2 \right \rangle.$$
Our consideration implies the slow change in the space of the current $\bm V$, while the absolute value of $h^V$ does not have to be small. From the wave’s point of view, $h^V$ remains flat and it determines the level of the unperturbed surface for the wave motion. In the general case, this level can change over time and its changes $\partial_t h^V$ are not necessarily small. For example, one can imagine a vessel filled with water into which fluid is constantly added, so that the average water level rises. The estimate $V_z \vert_{z=0} \sim \nu k^3 \langle h^2 \rangle$ used earlier in the text implicitly assumed that there was no such movement. Now we show that in the general case $V_z \vert_{z=0} - \partial_t h^V \sim \nu k^3 \langle h^2 \rangle$. Then, the cross-contribution to the momentum flux $\rho V_z\langle u^{\scriptscriptstyle (2)}_\alpha\rangle \vert_{z=0}$ from the left-hand side of equation (\[BC-project\]) can be combined with the last term in equation (\[corr\]), and we obtain $\rho \langle u^{\scriptscriptstyle (2)}_\alpha\rangle (V_z - \partial_t h^V)\vert_{z=0}$ that is small and can be neglected. Thus, the aforementioned changes in the average fluid level do not modify expression (\[eq:tau\]) for the virtual wave stress obtained previously in the reference frame which moves upwards with the local velocity $\partial_t h^V$.
Next, analyzing examples in Sec. \[sec:Ex\], we suggested that $V_z\vert_{z=0} = 0$ in the leading approximation. This is justified because the viscosity does not enter in the boundary condition for the horizontal velocity, see, e.g., expression (\[bc1\]), and the viscous correction (\[BCz\]) to the leading approximation will produce the parametrically smaller slow current. Note also that the already obtained result (\[eq:result\]) does not satisfy the incompressibility condition, if one reminds that the wave amplitude decays in space due to the fluid viscosity, $H=H_0 \exp(-4 \gamma^2 kx)$. In particular, the discussed viscous correction participates in the resolution of this discrepancy, but it is small and beyond the scope of the present paper.
To restore the Euler invariance for boundary conditions (\[BC\]) and (\[BCz\]), we now return to the laboratory reference frame, where $V_\alpha \vert_{z=0}$ is not equal to zero. However, due to assumed restriction $V/L\ll\nu k^2$, the horizontal velocity $V_\alpha$ should be considered almost constant in space. The Euler invariance implies that the partial time derivatives should be replaced with the material derivatives $$\label{EulerINV}
\partial_t \to \partial_t + V_\beta \partial_\beta$$ in boundary conditions (\[BC\]) and (\[BCz\]). In particular, the replacement (\[EulerINV\]) should be implemented in wave equations (\[wave\_equation\]) and (\[flow\_linear\]) in order to take into account the Doppler effect [@stewart1974hf]. This entails the corresponding replacement in equation (\[eq:tau\]).
Let us demonstrate that our theoretical scheme indeed inherits Euler invariance. If horizontal velocity $V_\alpha$ is not equal to zero, then in our calculations we should replace $$\begin{aligned}
\label{pi_alpha_gen}
&\pi_\alpha
\to \pi_\alpha + \rho V_\alpha \langle h^{(2)}\rangle \vert_{z=0},&
\\[5pt]\label{pi_alpha-beta_gen}
&\pi_{\alpha\beta}
\to
\pi_{\alpha\beta} + \left( V_\alpha \pi_\beta + V_\beta \pi_\alpha + \rho V_\alpha V_\beta \langle h^{(2)}\rangle \right)\vert_{z=0},&
\\[5pt] \label{Pi_alpha-z_gen}
&\langle \Pi_{\alpha z}^0\rangle
\to
-\rho \nu \partial_z V_\alpha\vert_{z=0} + \langle \Pi^\phi_{\alpha z}\rangle\vert_{z=0} + \rho V_\alpha \big(V_z + \langle u^{(2)}_z\rangle\big)\vert_{z=0},&\end{aligned}$$ and here we have assumed that $h^V =0$ and $\partial_t h^V=0$. The first rule (\[pi\_alpha\_gen\]) results in a change (\[EulerINV\]) in expression (\[VzEq\]). Rules (\[pi\_alpha\_gen\],\[pi\_alpha-beta\_gen\],\[Pi\_alpha-z\_gen\]) together with relation (\[VzEq\]) results in a change (\[EulerINV\]) in equation (\[BC\]), as it was expected.
Conclusion
==========
The attenuation of surface waves due to the fluid viscosity inevitably leads to the excitation of a slow flow $\bm V$. If it turns out to be weak, $V/L \ll \nu k^2$ and $V^2/(gL) \ll \nu k^2/\omega$, for example, at the initial stage of evolution, then one can neglect the scattering of waves by the slow flow inhomogeneities. In this case, the influence of waves on the slow flow is reduced to the action of a virtual wave stress applied at the fluid surface. As a result, the horizontal momentum associated with the wave motion is transferred from the waves to the slow current.
Here, based on the momentum conservation law, we found the explicit expression (\[eq:tau\]) for the virtual wave stress in terms of the excited wave motion in the deep-water limit. To demonstrate the validity of our approach, we analyzed the slow currents generated by a propagating wave and two orthogonal standing waves. These cases we extensively studied earlier, see Refs. [@longuet1953mass; @parfenyev2019formation], and we were able to reproduce the previously known results, see Sec. \[sec:Ex\].
The main result of this work is the generalization of the expression for the virtual wave stress to the case of excitation of arbitrary wave motion having a narrow spectrum, $\Delta \omega \ll \omega$. It can be used in the numerical simulation to effectively take into account the effect of fast wave motion on a slow flow. Also, our results allow studying the dynamics of slow currents, if the statistics of surface waves is known.
This work was supported by the Project 075-15-2019-1893 funded by the Ministry of Science and Higher Education of the Russian Federation. V.M.P. acknowledges support from the Foundation for the Advancement of Theoretical Physics and Mathematics ”BASIS”.
General Equations {#sec:A}
=================
In this section we demonstrate how to obtain hydrodynamic equations and boundary conditions in the usual form based on relations (\[eq:incompress\]) and (\[eq:Navier-Stokes\]). By using $\delta(h-z) = \theta'(h-z)$, one can rewrite equation (\[eq:incompress\]) in the form: $$\rho \delta(h-z)\left[\partial_t h + v_{\alpha} \partial_{\alpha} h - v_z\right] +
\theta(h-z)\left[\partial_t \rho + \partial_j(\rho v_j)\right] = 0,$$ where Latin indices take the values $\{x,y,z\}$, Greek indices take only $\{x,y\}$, and we sum over the repeated indices. The first term corresponds to the kinematic boundary condition posed at the fluid surface, and the second term — to the mass conservation law inside the fluid. Since we assume $\rho = const$, then from the second term we find the incompressibility condition $\mathrm{div}\, \bm v = 0$.
Similarly, let us consider relation (\[eq:Navier-Stokes\]). After straightforward calculations, we obtain $$\delta(h-z)\left[\rho v_i \partial_t h + \Pi_{i\alpha} \partial_{\alpha} h - \Pi_{iz} \right] +
\theta(h-z)\left[\rho \partial_t v_i + \partial_j \Pi_{ij} + \delta_{iz} \rho g \right] = 0.$$ Here the second term corresponds to the Navier-Stokes equation inside the fluid, $$\partial_t \bm v + (\bm v \nabla )\bm v = -\nabla p/\rho + \nu \nabla^2 \bm v + \bm g,$$ where we have used expression (\[eq:Pi\]). To simplify the first term, one needs to utilize the kinematic boundary condition, $$\label{kinematicBC}
\partial_t h = \left( v_z - v_{\alpha} \partial_{\alpha} h \right)\vert_{z=h},$$ which was obtained above, and then we find that $-p \partial_i (z-h) + \sigma_{ij}' \partial_j (z-h) = 0$ at the fluid surface, where $\sigma_{ij}' = \rho \nu (\partial_i v_j + \partial_j v_i)$ is the viscous stress tensor. Introducing the unit vector normal to the fluid surface $\bm n (t,x,y) = (-\partial_x h, -\partial_y h, 1)/\sqrt{1 + (\nabla h)^2}$, we finally obtain the dynamic boundary condition in the usual form $$\label{dynamicBC}
(-p n_i + \sigma_{ij}' n_j)\vert_{z=h} = 0.$$
Linear Waves {#sec:B}
============
Here we discuss some properties of the wave motion itself, assuming that there is no slow current, i.e. $\bm V = 0$. First, we consider an irrotational waves in an ideal fluid of infinite depth. Due to Kelvin’s theorem the velocity field is always potential, $\bm u = \nabla \phi$, and due to incompressibility condition the velocity potential $\phi$ satisfies the Laplace equation $\nabla^2 \phi = 0$. The pressure $p_u$ can be found from Bernoulli equation [@lamb1975hydrodynamics] $$\label{Bernoulli}
\partial_t \phi + \frac{1}{2}\big(\nabla \phi\big)^2 + \frac{p_u}{\rho} + gz = 0,$$ and one has to supplement these equations by the condition of the absence of the wave motion $\nabla \phi \rightarrow 0$ at infinite depth $z \rightarrow -\infty$, and by the boundary conditions posed at the fluid surface $z=h(x,y,t)$. This is the kinematic boundary condition (\[kinematicBC\]) and the dynamic boundary condition (\[dynamicBC\]), which is simply equal to $p_u\vert_{z=h}=0$ in an ideal fluid.
Since the potential satisfies Laplace equation $(\partial_z^2+\hat k^2)\phi=0$ and decreases downward, one finds $\partial_z \phi = \hat k \phi$, where we introduced the wave number operator $\hat k = (-\partial_x^2 - \partial_y^2)^{1/2}$. In the linear approximation, the boundary conditions have a form $\partial_t\phi^{\scriptscriptstyle (1)}\vert_{z=0}+gh=0$ and $\partial_th=\partial_z\phi^{\scriptscriptstyle (1)}\vert_{z=0}$, and therefore one can obtain the dispersion law $\omega^2 = gk$, where $k>0$ is the wave number.
Next, we begin to describe surface waves in a fluid with low kinematic viscosity $\nu$. Besides the change in potential $\phi$ due to modified boundary conditions, the viscosity produces small vortical correction which is nonzero inside the thin viscous sublayer near the fluid surface and is described by the vector stream function $\psi_{\alpha}$. Now the velocity of the wave motion is ${\bm u} = {\bm u}^\phi + {\bm u}^\psi$, where $u^\phi_i = \partial_i \phi$ and $u_\alpha^\psi = \partial_{z} \psi_\alpha$, $u^\psi_z= -\partial_{\alpha} \psi_\alpha$. The imaginary part of the wave frequency describes the decay of surface waves due to the fluid viscosity, $\omega''=-2 \nu k^2$, and this means that the waveform in the linear approximation satisfies the wave equation [@lamb1975hydrodynamics] $$\label{wave_equation}
\partial_t^2 h + g\hat k h + 4\nu \hat k^2 \partial_t h = 0.$$ To be self-consistent, all further calculations should be implemented up to the relative accuracy of ${\mathcal O}(\gamma^2)$. Note that the same equation (\[wave\_equation\]) is valid for the velocity potential $\phi^{\scriptscriptstyle (1)}$.
The vortical part of the velocity field ${\bm u}^\psi$ is small in viscosity, so it should be neglected for the normal component of the dynamic boundary condition $p_u-2\rho \nu n_in_j\partial_iu_j=0$, see equation (\[dynamicBC\]). Using this condition, the wave equation (\[wave\_equation\]) and the relation $p_u^{\scriptscriptstyle (1)} = - \rho g z - \rho \partial_t \phi^{\scriptscriptstyle (1)}$, which is found from the linearized Bernoulli equation (\[Bernoulli\]) that is valid in the viscous fluid as well, see Ref. [@lamb1975hydrodynamics §349], we conclude that the velocity potential in the linear approximation is equal to $$\label{flow_linear}
\phi^{\scriptscriptstyle (1)}
=
\frac{\partial_t+2\nu\hat k^2}{\hat k}\exp(z\hat k)h.$$ Considering the solenoidal part ${\bm u}^\psi$, we should take into account that the wave amplitude $h$ can be larger than the thickness of the viscous sublayer $\delta = \sqrt{2} \gamma/k$. This situation was studied systematically, e.g., in Ref. [@longuet1953mass]. In this case, the viscous boundary layer cannot be described by a simple linear theory. For our purposes it is sufficient to note that the vector stream function decreases when moving downward from the surface as $\exp((z-h)/\delta)$ and it is equal to $$\label{psi}
\psi^{\scriptscriptstyle (1)}_\alpha\vert_{z=h} = -2\nu \partial_\alpha h$$ at the fluid surface, see Ref. [@longuet1992theory].
[20]{}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 [**]{} (, ) in @noop [**]{}, Vol. (, ) pp. @noop [****, ()]{}
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We consider an Einstein-Yang-Mills Lagrangian in a five dimensional space-time including a cosmological constant. Assuming all fields to be independent of the extra coordinate, a dimensional reduction leads to an effective (3+1)-dimensional Einstein-Yang-Mills-Higgs-dilaton model where the cosmological constant induces a Liouville potential in the dilaton field. We construct spherically symmetric solutions analytically in specific limits and study the generic solutions for vanishing dilaton coupling numerically. We find that in this latter case the solutions bifurcate with the branch of (Anti-) de Sitter-Reissner-Nordström ((A)dSRN) solutions.'
author:
- |
[Betti Hartmann[^1]]{}\
[Department of Mathematical Sciences, University of Durham,]{}\
[Durham DH1 3LE, United Kingdom]{}\
[ ]{}\
[Yves Brihaye[^2]]{}\
[Faculté des Sciences, Université de Mons-Hainaut, ]{}\
[B-7000 Mons, Belgium ]{}\
[ ]{}\
[Bruno Bertrand]{}\
[Faculté des Sciences, Université de Mons-Hainaut, ]{}\
[B-7000 Mons, Belgium ]{}\
title: '**Spherically symmetric Yang-Mills solutions in a 5-dimensional (Anti-) de Sitter space-time**'
---
Introduction
============
The scalar dilaton field arised as companion of the metric tensor in (super)string theories and is associated with the scale invariance of these theories [@maeda]. Thus it is interesting to study classical field theory solutions coupled to a dilaton. In most studies, the dilaton was assumed to be massless while, however, from the viewpoint of a realistic theory the dilaton should be massive in order to avoid long-range scalar forces. In [@chm] a dilaton potenial of Liouville type was introduced to take into account the effects of a specific symmetry-breaking mechanism which gives mass to the dilaton. This type of potential has a constant prefactor which in the limit of vanishing dilaton coupling reduces to a cosmological constant. It was found that there exist no asymptotically flat/ de Sitter/ Anti-de Sitter solutions for non-vanishing potential [@polet]. Rotating generalisations of the black hole solutions found in [@chm] have been constructed in [@mitra].
Volkov argued recently [@volkov] that if $\frac{\partial}{\partial x_4}$ is a symmetry of the Einstein-Yang-Mills (EYM) system in ($4+1$) dimensions, where $x_4$ is the coordinate associated with the $5$th dimensions, than the ($4+1$)-dimensional EYM system reduces effectively to a ($3+1$)-dimensional EYMHD system with a specific coupling between the dilaton field and the Higgs field. The generalisation of this ($3+1$)- dimensional EYMHD model was consequently studied in [@bh].
In this paper, we study spherically symmetric solutions of the ($3+1$)-dimensional EYMHD model deduced from the ($4+1$)-dimensional EYM system including a cosmological constant. The dimensional reduction then leads to a Liouville-type potential in the $(3+1)$-dimensional model. In Section 2, we present both the five-dimensional model and the from this deduced and then generalised $(3+1)$-dimensional EYMHD model. In Section 3, we discuss the solutions for the case of vanishing dilaton coupling, especially, we present our numerical results for the generic solutions in this case. In section 4, we discuss possible solutions for the generic case of non-vanishing dilaton coupling. Our conclusions are presented in Section 5.
The model
=========
We start with the Einstein-Yang-Mills Lagrangian in five dimensions including a cosmological constant given by:
$$S = \int \Biggl(
\frac{1}{16 \pi G_5}( R - 2 \Lambda_{(5)})
- \frac{1}{4 \tilde{e}^2}F^a_{M N}F^{a M N}
\Biggr) \sqrt{g^{(5)}} d^5 x$$
with the SU(2) Yang-Mills field strength $F^a_{M N} = \partial_M A^a_N -
\partial_N A^a_M + \epsilon_{a b c} A^b_M A^c_N$ , the gauge index $a=1,2,3$ and the space-time index $M=0,1,2,3,4$. $G_5$ and $\tilde{e}$ denote respectively the 5-dimensional Newton’s constant and the coupling constant of the gauge field theory. $G_5$ is related to the 5-dimensional Planck scale $M_{Pl(5)}$ by $G_5=M^{-3}_{Pl(5)}$. $\Lambda_{(5)}$ is the 5-dimensional cosmological constant.
If both the matter functions and the metric functions are independent on $x_4$, the $5$-dimensional fields can be parametrized as follows [@volkov]: $$g^{(5)}_{MN}dx^M dx^N = e^{-\zeta}g^{(4)}_{\mu\nu}dx^{\mu}dx^{\nu}
+e^{2\zeta} (dx^4)^2 \ , \ \mu , \nu=0, 1, 2, 3$$ and $$A_M^{a}dx^M=A_{\mu}^a dx^{\mu}+\Phi^a dx^4 \ , $$ where $g^{(4)}$ is the $4$-dimensional metric tensor and $\zeta$ plays the role of the dilaton.
In [@volkov] it was shown that for $\Lambda_{(5)}=0$ the classical equations are equivalent to those of a four-dimensional Einstein-Yang-Mills-Higgs dilaton theory. In this paper, we consider the case with a cosmological constant. We then choose the generalised $(3+1)$-dimensional action to be: $$S=S_{G}+S_{M}=\int L_{G}\sqrt{-g^{(4)}}d^{4}x+
\int L_{M}\sqrt{-g^{(4)}}d^{4}x
\ . \label{action}$$ with the gravity Lagrangian: $$L_{G}=\frac{1}{16\pi G}R
\ ,$$ and $G$ denoting the 4-dimensional Newton‘s constant. The matter Lagrangian $L_M$ reads: $$L_{M}=-\frac{1}{4} e^{2\kappa\Psi}F_{\mu\nu}^{a}F^{\mu\nu,a}
-\frac{1}{2}\partial_{\mu}\Psi\partial^{\mu}\Psi
-\frac{1}{2}e^{-4\kappa\Psi}D_{\mu}\Phi^{a}
D^{\mu}\Phi^{a}-e^{-2\kappa\Psi}V(\Phi^{a})
- \frac{\tilde{\Lambda}}{2}e^{-2\kappa\Psi}
\ , \label{lag}$$ with the Higgs potential $$V(\Phi^{a})=\frac{\lambda}{4}(\Phi^{a}\Phi^{a}-v^2)^2
\ ,$$ the non-abelian field strength tensor $$F_{\mu\nu}^{a}=\partial_{\mu}A_{\nu}^{a}-\partial_{\nu}A_{\mu}^{a}+
e\varepsilon_{abc}A_{\mu}^{b}A_{\nu}^{c}
\ ,$$ and the covariant derivative of the Higgs field in the adjoint representation $$D_{\mu}\Phi^{a}=\partial_{\mu}\Phi^{a}+
e\varepsilon_{abc}A_{\mu}^{b}\Phi^{c}
\ .$$ The gauge field coupling constant is denoted $e$, $\lambda$ is the Higgs field coupling constant and $v$ the vacuum expectation value of the Higgs field.
Note that we have introduced a coupling $\kappa$ specific to the dilaton field by setting $\zeta=2\kappa\Psi$. This will allow to study the influence of the dilaton systematically. We remark that the $5$-dimensional cosmological constant has through dimensional reduction led to a Liouville potential in the dilaton field with coupling constant $\tilde{\Lambda}$. For $\kappa=0$, $\tilde{\Lambda}$ is proportional to the four-dimensional cosmological constant.
The Ansatz
----------
For the metric the spherically symmetric Ansatz in Schwarzschild-like coordinates reads [@bfm; @weinberg]: $$ds^{2}=g^{(4)}_{\mu\nu}dx^{\mu}dx^{\nu}=
-A^{2}(r)N(r)dt^2+N^{-1}(r)dr^2+r^2 d\theta^2+r^2\sin^2\theta
d^2\varphi
\label{metric}$$ with $$N(r)=1-\frac{2m(r)}{r}
\ .$$ In these coordinates, $m(\infty)$ denotes the (dimensionful) mass of the field configuration. For the gauge and Higgs fields, we use the purely magnetic hedgehog ansatz [@thooft] $${A_r}^a={A_t}^a=0
\ ,$$ $${A_{\theta}}^a= \frac{1-K(r)}{e} {e_{\varphi}}^a
\ , \ \ \ \
{A_{\varphi}}^a=- \frac{1-K(r)}{e}\sin\theta {e_{\theta}}^a
\ ,$$ $${\Phi}^a=v H(r) {e_r}^a
\ .$$ The dilaton is a scalar field depending only on $r$ $$\Psi=\Psi(r)
\ .$$ Inserting the Ansatz into the Lagrangian and varying with respect to the matter fields yields the Euler-Lagrange equations, while variation with respect to the metric yields the Einstein equations.
Classical field equations
-------------------------
With the introduction of dimensionless coordinates and fields $$x=evr \ , \ \ \mu=evm \ ,\ \ \phi=\frac{\Phi}{v}\ , \ \
\psi=\frac{\Psi}{v}
\label{scale}$$ the Lagrangian and the resulting set of differential equations depend on the following coupling constants: $$\alpha =\sqrt{G}v =\frac{M_W}{eM_{\rm Pl}} \ , \ \
\beta=
\frac{\sqrt{\lambda}}{e} = \frac{M_H}{\sqrt{2}M_W} \ , \ \
\gamma =\kappa v =\frac{\kappa M_W}{e}
\label{coupling} \ , \ \ \Lambda=2\alpha^2 \tilde{\Lambda} \ ,$$ where $M_W=e v$, $M_H= \sqrt{2\lambda} v$ and $M_{\rm Pl}=1/\sqrt{G}$. With the rescalings (\[scale\]) and (\[coupling\]), the dimensionless mass of the solution is given by $\frac{\mu(\infty)}{\alpha^2}$. Note that we have rescaled the cosmological constant in order to obtain the equations of a conventional (3+1)-dimensional Einstein-Yang-Mills-Higgs model including a cosmological constant in the limit of vanishing dilaton coupling.
With (\[scale\]) and (\[coupling\]) the Euler-Lagrange equations read: $$(e^{2\gamma\psi}ANK')'=A(e^{2\gamma\psi}\frac{K(K^2-1)}{x^2}+
e^{-4\gamma\psi}H^2
K)
\ , \label{dgl1}$$ $$(e^{-4\gamma\psi}x^2 ANH')'=AH(2e^{-4\gamma\psi}K^2+ \beta^2 x^2
e^{-2\gamma\psi}(H^2-1))
\ , \label{dgl2}$$ $$\begin{aligned}
(x^2 A N\psi')' &=& 2\gamma A [e^{2\gamma\psi}(N(K')^2+\frac{(K^2-1)^2}{2
x^2}) -\frac{\Lambda}{4\alpha^2} x^2 e^{-2\gamma\psi}
\nonumber \\
&-& e^{-2\gamma\psi}\frac{\beta^2 x^2}{4}(H^2-1)^2-2
e^{-4\gamma\psi}(\frac{1}{2}
N (H')^2 x^2+H^2 K^2) ]
\ , \label{dgl3} \end{aligned}$$ where the prime denotes the derivative with respect to $x$, while we use the following combination of the Einstein equations $$G_{tt}=2\alpha^2 T_{tt}=-2\alpha^2 A^2 N L_{M}
\ ,$$ $$g^{xx}G_{xx}-g^{tt}G_{tt}=-4\alpha^2 N \frac{\partial L_{M}}{\partial N}$$ to obtain two differential equations for the two metric functions: $$\mu ' = \alpha^2 \left(
e^{2\gamma\psi}N(K')^2 + \frac{1}{2}N x^2(H')^2
e^{-4\gamma\psi}+
\frac{1}{2x^2}(K^2-1)^{2} e^{2\gamma\psi}+K^2 H^2 e^{-4\gamma\psi}\right.
$$ $$+ \left. \frac{\beta^{2}}{4}x^2
(H^2-1)^2 e^{-2\gamma\psi}+\frac{1}{2}Nx^{2}(\psi ')^2\right)
+ \frac{\Lambda}{4} x^2 e^{-2\gamma\psi}
\ , \label{dgl4}$$ $$A'=\alpha^2 x A \left(\frac{2(K')^2}{x^2}e^{2\gamma\psi}+
e^{-4\gamma\psi}(H')^2+(\psi ')^2
\right)
\ . \label{dgl5}$$
Note that the equations of the original five dimensional theory are recovered by using the following specific choice of the coupling constants: $$\label{5dlimit}
\alpha^2 = 3 \gamma^2 \ \ , \ \
\Lambda = \Lambda_{(5)} \ \ , \ \
\beta = 0 \ \ .$$
The case $\Lambda=0$ was previously studied in [@volkov; @bh]. If in addition $\gamma=0$, the equations of the Einstein-Yang-Mills-Higgs equations are recovered [@bfm; @weinberg]. Choosing $\Lambda=\alpha=0$ (assuming $\Lambda/\alpha^2=0$ as well), the model reduces to the Yang-Mills-Higgs-dilaton system studied in [@forgacs1].
Spherically symmetric solutions for $\gamma=0$
===============================================
We will first discuss the solutions in the case $\gamma = 0$. The equation of the dilaton field can then be decoupled and $\psi(x)\equiv 0$. We will study solutions of this system which are regular at the origin this implies the following conditions $$K(0)=1 \ , \ \ H(0)=0 \ , \ \ \mu(0)=0
\ . \label{bc1}$$ Finiteness of the ADM mass requires that the fields approach particular values asymptotically, namely: $$K(\infty)=0 \ , \ \ H(\infty)=1 \ , \ \ A(\infty)=1
\ . \label{bc2}$$
For $\Lambda > 0$ the metric function $N(x)$ has a zero at a finite value of $x$, say $x=x_c$. This is the so-called “cosmological horizon”. The value $x_c$ depends on the actual values of the coupling constants.
(Anti-) De Sitter-Reissner-Nordström ((A)dSRN) solutions
--------------------------------------------------------
Setting $\gamma = 0$ the system admits embedded abelian solutions, the so-called (Anti-) de Sitter-Reissner-Nordström solutions: $$K(x) = 0 \ \ , \ \ H(x) = 1 \ \ , \ \ \psi(x) = 0 \ \ , \ \ A(x) = 1 \ \ , \ \
N(x)= 1-\frac{1}{6} \Lambda x^2 -\frac{2\mu_{\infty}}{x}+\frac{\alpha^2}{x^2} \ .$$ The metric function $N(x)$ has a physical singularity at the origin $x=0$ which is evident from the Kretschmann scalar $K=R_{\alpha\beta\gamma\delta}
R^{\alpha\beta\gamma\delta}$: $$K=\frac{2}{3x^8}\left(\Lambda^2x^8+72 \mu_{\infty}^2x^2-144 \mu_{\infty}
\alpha^2 x+84 \alpha^2\right) \ .$$ Depending on the choice of the sign of the cosmological constant, up to $4$ zeros of $N(x)$ can exist. $3$ of the $4$ zeros correspond to horizons since the first zero has always negative value and thus has no physical meaning. The two inner horizons $x_-$, $x_+$ with $x_- \le x_+$ correspond to the well known Cauchy, respectively event horizon of the Reissner-Nordström solution, while the third outer horizon $x_c > x_+$ exists only for positive cosmological constant.
Extremal black hole solutions - like in the asymptotically flat space - are possible. Then, we have $x_-=x_+=x_h$ with $N(x_h)=N'|_{x=x_h}=0$. This leads to the equation: $$\Lambda x_h^4 -2x_h^2+2\alpha^2=0 \ .$$ This is solved by: $$\label{xhds}
x_{h/c}=\frac{1}{\sqrt{\Lambda}}\sqrt{1\pm\sqrt{1-2\alpha^2\Lambda}} \ \
\ \ \ {\rm for} \ \ \ \frac{1}{2\alpha^2} \ge \Lambda > 0$$ and $$x_{h}=\frac{1}{\sqrt{|\Lambda|}}\sqrt{-1+\sqrt{1-2\alpha^2\Lambda}} \ \
\ \ \ {\rm for} \ \ \ \Lambda < 0 \ .$$ For $\Lambda > 0$, the solution with the plus sign is the outer, cosmological horizon $x_c$, while the inner, event horizon $x_h$ is the solution with the minus sign. Obviously, the appearance of horizons in dS space is restricted by $\alpha^2 \le \frac{1}{2\Lambda}$. The corresponding mass of the extremal solution is given by: $$\mu_{\infty}=\frac{2}{3}\frac{\alpha^2}{x_h}+\frac{x_h}{3} \ .$$ Apparently, the $\Lambda=0$ limit is ill-defined. However, for $0 < \Lambda \ll 1$ we find $$\mu_{\infty} = \alpha - \frac{\alpha^3}{9} \Lambda + O(\Lambda^2) \ \ , \ \
x_h = \alpha + \frac{\alpha^3}{3} \Lambda + O(\Lambda^2)$$ which for $\Lambda\rightarrow 0$ obviously leads to the corresponding values of the well-known asymptotically flat Reissner-Nordström solution.
de Sitter (dS) gravitating monopoles
------------------------------------
Since gravitating monopoles in Anti-de Sitter space have been studied previously [@adsmono], we concentrate here on monopoles in de Sitter space. To our knowledge, these type of solutions have not been studied previously.
In the absence of a cosmological constant, the flat space magnetic monopole [@thooft] is deformed by gravity and exist up to a critical value of $\alpha=\alpha_{cr}$ where the solution bifurcates with the branch of extremal Reissner-Nordström solutions [@bfm]. For instance in the BPS limit ($\beta = 0$) the gravitating monopole bifurcates with this branch at $\alpha_{cr} \approx 1.386$.
Now analysing the equations in the presence of a cosmological constant, we were able to construct dS-gravitating monopoles. They are characterised by a cosmological horizon at $x = x_c$ with $N(x=x_c)=0$. The behaviour of the function $N(x)$ is illustrated in Fig. 1 for $\alpha = 0.8$ and different values of $\Lambda$. We find that $x_c$ is decreasing with the increase of $\Lambda$: $x_c \sim 108$ for $\Lambda \sim 0.0005$ and $x_c \sim 77$ for $\Lambda = 0.001$. As is obvious from the figure, the solutions have a local minimum at some value of the radial coordinate $x=x_{min}(\Lambda)$.
The main aim of this study was to determine the domain of coupling constants in which dS-gravitating monopoles exist. Fixing $\beta$ and $\Lambda$ our analysis demonstrates that dS-gravitating monopoles bifurcate with the branch of extremal dSRN solutions described in the previous section at a critical value of $\alpha$. Since we limited our analysis to small values of $\Lambda$ the critical value of $\alpha$ where the bifurcation occurs hardly differs from the corresponding one in the asymptotically flat case.
The way how the extremal dSRN solution is approached is illustrated in Fig. 2 for $\Lambda = 0.001$ and $\beta=0.1$. This clearly shows that the value of the local minimum of the function $N(x)$ decreases while $\alpha$ increases. We find that solutions exist up to a maximal value of the gravitational coupling $\alpha=\alpha_{max}\approx 1.382$. There another branch of non abelian solutions exist which bifurcates with the branch of dSRN solutions at a critical value of $\alpha=\alpha_{cr}\approx
1.378$. At this point, a degenerate horizon forms at $x=x_h$. The critical solution can be described by the dS-RN solution with horizons (\[xhds\]) for $x \ge x_h$, while for $x_h > x \ge 0$, it is non-singular and non-trivial. Compared to the case $\Lambda=0$ [@bfm], the values of $\alpha_{max}$ and $\alpha_{cr}$ are smaller when $\Lambda > 0$. Moreover, the interval of $\alpha$ on which two solutions exist decreases. This can be related to the increased cosmological expansion for $\Lambda > 0$.
Spherically symmetric solutions for $\gamma\neq 0$
===================================================
In the case of Einstein-Maxwell-dilaton theory, the Liouville potential leads to the fact that the solutions are neither asymptotically flat nor de Sitter nor Anti-de Sitter [@chm]. As far as our numerical simulations suggest, this holds also true for the case of non-abelian gauge fields, since we were not able to construct asymptotically flat/ de Sitter/ Anti- de Sitter solutions. However, in a specific limit, namely the embedded abelian case, analytic solutions are available.
The case $H(x)\equiv 1$, $K(x)\equiv 0$
---------------------------------------
Setting $H\equiv 1$ and $K\equiv 0$ for all $x$, we find the following solutions of the system of equations: $$A(x)=a_0 x^{\frac{\alpha^2}{\gamma^2}} \ , \ \ \psi(x)=\psi_0+\frac{1}{\gamma}\ln(x)$$ and $$N(x)=n_0-n_1 x^{-\frac{(\gamma^2+\alpha^2)}{\gamma^2}} \ \ {\rm with} \ \
n_0=\frac{\gamma^4}{\alpha^2+\gamma^2}\left(e^{2\gamma\psi_0}-
\frac{\Lambda}{2\alpha^2} e^{-2\gamma\psi_0}\right) \ \ .$$ The cosmological constant is given by: $$\Lambda=2\alpha^2\left(\frac{1}{\alpha^2-\gamma^2}e^{2\gamma\psi_0}-\frac{\gamma^2+\alpha^2}
{\alpha^2-\gamma^2} e^{4\gamma\psi_0}\right) \ .$$ This solution has a single event horizon for $n_1 > 0$. Moreover, it can be seen, that this solution is ill-defined for $\alpha=\gamma$. Note that these are generalisations of the solutions constructed in [@chm]. For $\alpha=1$, the above solution corresponds to one of the solutions found in [@chm]. In Fig.3, we show qualitative profiles of the functions for the choice of parameters which corresponds to the $5$-dimensional limit (\[5dlimit\]). In addition, we choose $\gamma=\psi_0=a_0=n_1=1$. It is obvious from this figure that the solution has a horizon (here at $x=x_h \approx 0.655$) and thus represents a black hole.
If we choose instead the limit $\alpha=0$, the function $A(x)$ becomes constant $=a_0$. The metric function $N(x)=1-n_1 x^{-1}$ in this limit.
Conclusions
===========
In a previous paper [@volkov], it was shown that a Einstein-Yang-Mills model in $5$ dimensions can be reduced to an effective $(3+1)$-dimensional Einstein-Yang-Mills-Higgs-dilaton model under certain symmetry conditions -spherical symmetry and independence on the coordinate associated with the $5$th dimension. One of the main results of the present paper shows that the reduction of a $5$ dimensional de Sitter (dS)/Anti- de Sitter (AdS) Einstein-Yang-Mills system to an effective $(3+1)$-dimensional action (with the same symmetry assumptions as in [@volkov]) leads to a self-interaction of the dilaton field via a Liouville potential.
Previous considerations of an Einstein-Maxwell-dilaton model including a Liouville potential [@chm] have revealed that no asymptotically flat/ de Sitter/ Anti- de Sitter solutions can be constructed [@mitra]. All our attempts to construct numerically solutions of the non-abelian counterpart have failed. Thus, we believe that the absence of asymptotically flat/ de Sitter/ Anti- de Sitter solutions holds also true in the case of non-abelian gauge fields. However, considering the limit of vanishing dilaton coupling, we were able to recover the AdS gravitating monopoles studied previously [@adsmono] and to produce previously not studied solutions, namely the dS gravitating monopoles. We show for the latter solutions that they bifurcate with the branch of dS-Reissner-Nordström (dSRN) solutions at a critical value of the gravitational coupling. Finally, considering the limit $K(x)\equiv 0$ and $H(x)\equiv 1$ for non-vanishing Liouville potential, we were able to construct generalisations of the solutions found in [@chm].\
\
\
[**Acknowledgements**]{} Y. B. gratefully acknowledges the Belgian F.N.R.S. for financial support. B. H. was supported by an EPSRC grant.
[99]{} G. Gibbons and K. Maeda, Nucl. Phys. [**B298**]{} (1988), 741; D. Garfinkle, G. Horowitz and A. Strominger, Phys. Rev. [**D43**]{} (1991), 371. K. C. K. Chan, J. H. Horne and R. B. Mann, Nucl. Phys. [**B447**]{} (1995), 441. S. Poletti and D. Wiltshire, Phys. Rev. [**D50**]{} (1994) 7260. T. Ghosh and P. Mitra, Class. Quantum Grav. [**20**]{} (2003), 1403. M. S. Volkov, Phys. Lett. [**B524**]{} (2002), 369. Y. Brihaye and B. Hartmann , Phys. Lett. [B 534]{} (2002) 137. P. Breitenlohner, P. Forgacs and D. Maison, Nucl. Phys. [**B383**]{} (1992), 357;\
P. Breitenlohner, P. Forgacs and D. Maison, Nucl. Phys. [**B442**]{} (1995), 126. K. Lee, V. P. Nair and E. J. Weinberg, Phys. Rev. [**D45**]{} (1992) 2751. P. Forgacs and J. Gyueruesi, Phys. Lett. [**B366**]{} (1996), 205. A. R. Lugo, E. F. Moreno and F. A. Shaposnik, Phys. Lett. [**B 473**]{} (2000) 35. G. ‘t Hooft, Nucl. Phys. [**B79**]{} (1974), 276;\
A. M. Polyakov, JETP Lett. [**20**]{} (1974), 194.
[^1]: [email protected]
[^2]: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
0.7cm
[Eun Kyung Park[^1] and Pyung Seong Kwon[^2]\
0.2cm]{}
0.05cm
*$^1$Department of Materials Physics, Dong-A University,\
Busan 604-714, Korea\
0.1cm*
$^2$Department of Energy Science, Kyungsung University,\
Busan 608-736, Korea
2.0cm
[**Abstract**]{} 0.1cm
We apply a new self-tuning mechanism to the well-known Kachru-Kallosh-Linde-Trivedi (KKLT) model to address the cosmological constant problem. In this mechanism the cosmological constant $\lambda$ contains a supersymmetry breaking term ${\mathcal E}_{\rm SB}$ besides the usual scalar potential ${\mathcal V}_{\rm scalar}$ of the $N=1$ supergravity, which is distinguished from the usual theories where $\lambda$ is directly identified with ${\mathcal V}_{\rm scalar}$ alone. Also in this mechanism, whether $\lambda$ vanishes or not is basically determined by the tensor structure of the scalar potential density, not by the zero or nonzero values of the scalar potential itself. As a result of this application we find that the natural scenario for the vanishing $\lambda$ of the present universe is to take one of the AdS (rather than dS) vacua of KKLT as the background vacuum of our present universe. This AdS vacuum scenario has more nice properties as compared with dS vacua of the usual flux compctifications. The background vacuum is stable both classically and quantum mechanically (no tunneling instabilities), and the value $\lambda =0$ is also stable against quantum corrections because in this scenario the perturbative corrections of ${\mathcal V}_{\rm scalar}$ and quantum fluctuations $\delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ on the branes are all gauged away by an automatic cancelation between ${\mathcal V}_{\rm scalar} + \delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ and ${\mathcal E}_{\rm SB}$.
0.25cm
[PACS number: 11.25.-w, 11.25.Uv]{}\
0.2cm [*Keywords*]{}: cosmological constant problem, KKLT, supersymmetry breaking, self-tuning
6.0mm
0.5cm
One of the most mysterious problems in the area of high energy physics including cosmology can be summarized as why the vacuum energy (or the cosmological constant) of our present universe is so small despite that the supersymmetry of our universe is considerably broken. Recently there has been proposed a new mechanism to address this cosmological constant problem in the framework of type IIB supergravity [@1], where the four-dimensional cosmological constant $\lambda$ is forced to vanish by six-dimensional Einstein equation of the transverse sector, and therefore tunes itself to zero as a result. This mechanism is based on the viewpoint that our three-dimensional space is a stack of BPS (visible sector) $D3$-branes located at the conifold singularity of the Calabi-Yau threefold, and in this setup $\lambda$ generally appears as $$\lambda = \frac{\kappa^2}{2} \Big( {\mathcal V}_{\rm scalar} + \delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)} - {\mathcal E}_{\rm SB} \Big) \,\,.$$ See Sec. 3.5 of this paper for the details.
In (1.1), ${\mathcal V}_{\rm scalar}$ is the usual scalar potential for the moduli of the $N=1$ supergravity, and $\delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ are NS-NS and R-R vacuum energies arising from quantum fluctuations (of the gravitational and standard model degrees of freedom with support) on the $D3$-branes. $\delta_Q {\hat I}_{\rm brane}^{(NS)}$ and $\delta_Q {\hat I}_{\rm brane}^{(R)}$ are expected to cancel out when supersymmetry of the brane region is unbroken. (The cancelation at one-loop order on the BPS $D3$-branes has been proven explicitly in Sec. VIIA of [@1] for the case where the three-form fluxes of type IIB theory are turned off.) The last term ${\mathcal E}_{\rm SB}$ is a supersymmetry breaking term which originates from a gauge symmetry breaking of the R-R four-form $A_{(4)}$ arising at the quantum level in the brane region. So ${\mathcal E}_{\rm SB}$ is an energy scale of the supersymmetry breaking of the brane region, and at the same time it is also an energy scale of the gauge symmetry breaking (or an anomaly) generated by quantum fluctuations. Finally in (1.1), $\kappa^2 = 1/2 M_{pl}^2$ where $M_{pl}$ is the four-dimensional Planck scale. Eq. (1.1) is distinguished from the corresponding equation of the usual flux compactifications where $\lambda$ is simply given by ${\mathcal V}_{\rm scalar}$ alone. According to (1.1), $\lambda =0$ does not necessarily imply ${\mathcal V}_{\rm scalar} =0$ unlike in the usual $N=1$ supergravity, for instance, in [@1-1; @2].
${\mathcal E}_{\rm SB}$ is kind of an anomaly generated in the brane region, and for the $D3$-branes located at the conifold singularity of the Calabi-Yau threefold it takes (at one-loop order) the form $${\mathcal E}_{\rm SB}^{(1)} = -\delta_0 \int r^5 dr \epsilon_5 \rho_{\rm T}^{(1)} \,\,,~~~~~\big(\delta_0 = {\rm constant} \big) \,\,,$$ (Compare (1.2) with (3.46), where $\rho_{\rm T}^{(1)}$ and $\delta \mu_{\rm T}^{m} (\phi)$ are given by (1.4) and (8.10), respectively) where $\epsilon_5 = \sqrt{det |{\hat h}_{mn}|} \, d\psi \wedge d\theta_1 \wedge d\phi_1 \wedge d\theta_2 \wedge d\phi_2$ is the volume-form of the base of the cone in the conifold metric $ds^2 = dr^2 + r^2 d \Sigma_{1,1}^2$ with $$d \Sigma_{1,1}^2 = \frac{1}{9} \Big( d \psi + \sum_{i=1}^{2} \cos \theta_i d \phi_i \Big)^2 + \sum_{i=1}^{2} \frac{1}{6} \Big( d\theta_i^2 + \sin^2 \theta_i d\phi_i^2 \Big) \equiv {\hat h}_{mn} dy^m dy^n \,\,,$$ so the volume of the base of the cone with unit radius is given by ${\rm Vol(B)}= \int \epsilon_5$. Also the integration $\int r^5 dr \epsilon_5$ in (1.2) is taken over the brane region, $0<r<r_B$, where $r_B$ being the thickness of the brane, and the constant $\delta_0$ is given by $\delta_0 = {6}/[{r_B^6 \,{\rm Vol(B)}}]$. (See Secs. VIB and VIIA of [@1] for the notations.) Finally $\rho_{\rm T}^{(1)}$ sources the supersymmetry breaking of the brane region (which is why ${\mathcal E}_{\rm SB}$ is called supersymmetry breaking term. See (8.11).), and it takes the form $$\rho_{\rm T}^{(1)} (y)= \nu_{(1)}^m f_m(y) \,\,.$$
In (1.4), $f_m (y)$’s are arbitrary gauge parameters and $\nu_{(1)}^m$’s represent (one-loop order) quantum excitations on the brane with components along the transverse directions of the $D3$-branes. Since $\rho_{\rm T}^{(1)}$ contains arbitrary gauge parameters, ${\mathcal E}_{\rm SB}$ in (1.2) also has gauge arbitrariness. In (1.1), ${\mathcal V}_{\rm scalar}$ takes nonzero values at the quantum level because it receives generically both perturbative and nonperturbative corrections. Also, $\delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ does not vanish if the brane supersummetry is broken. But $\lambda$ in (1.1) contains ${\mathcal E}_{\rm SB}$. So any nonzero ${\mathcal V}_{\rm scalar}$ and $\delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ can be gauged away by this ${\mathcal E}_{\rm SB}$ so that $\lambda$ vanishes as a result. Such a cancelation between ${\mathcal V}_{\rm scalar} + \delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ and ${\mathcal E}_{\rm SB}$ really occurs in (1.1), forced by a self-tuning equation (eq (3.41)) which imposes a constraint on $\lambda$. See Sec. 3.5 for this.
In the self-tuning mechanism of this paper, whether $\lambda$ vanishes or not is basically determined - in the six-dimensional internal space - by the tensor structure of the scalar potential density, not by the zero or nonzero values of the scalar potential ${\mathcal V}_{\rm scalar}$ itself. Thus in our self-tuning mechanism, whether ${\mathcal V}_{\rm scalar}$ vanishes or not is not important unlike in the usual theories where $\lambda=0$ is equivalent to ${\mathcal V}_{\rm scalar}=0$. In this paper we will apply this mechanism to the well-known scenario of KKLT [@2] to address the cosmological constant problem, especially aiming at explaining the (cause of the) vanishing cosmological constant of our present universe.
As a result of this application we find - basically in the framework of the type IIB $N=1$ supergravity - that the natural scenario for the vanishing $\lambda$ of our present universe is to take one of the AdS (rather than dS) vacua of KKLT as the background vacuum of the present universe. This AdS vacuum scenario has more nice properties as compared with dS vacua of the usual flux compactifications. The background vacuum is stable both classically and quantum mechanically (i.e., no tunneling instabilities), and the value $\lambda =0$ is perturbatively (radiatively) stable unlike in the usual theories because in our self-tuning mechanism of this paper the perturbative and nonperturbative corrections of ${\mathcal V}_{\rm scalar}$ are all gauged away by an automatic cancelation between ${\mathcal V}_{\rm scalar} + \delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ and ${\mathcal E}_{\rm SB}$.
0.5cm
Kachru $et~al.$ have shown in the framework of the Klebanov-Strassler (KS) compactifications [@3] that one can construct a de Sitter (dS) vacuum (of type IIB theory) with broken supersymmetry if we allow for nonperturbative corrections and anti-$D3$-branes. They first obtained a supersymmetric anti-de Sitter (AdS) vacuum from the superpotential of the form $$W=W_0 + A e^{i a \rho} \,\,,$$ where $W_0$ is a tree level contribution arising from the fluxes and does not contain the K$\ddot{\rm a}$hler modulus $\rho$. The second term is a nonperturbative correction coming from Euclidean $D3$-branes (instantons) [@4], or the gaugino condensation in the $N=1$ supersymmetric $SU(N_c)$ gauge theory generated by the stack of $N_c$ coincident $D7$-branes wrapping four-cycles in the Calabi-Yau threefold [@5]. Since $W$ contains K$\ddot{\rm a}$hler modulus the no-scale structure of the Lagrangian has been broken and the supersymmetric vacuum is now described by $$D W=0 \,\,,$$ but not necessarily $W=0$, where the covariant derivative $D_a W$ is defined by $D_a W= \partial_a W + (\partial_a {\mathcal K}) W$, and where the K${\ddot{\rm a}}$hler potential ${\mathcal K}$ is given at the tree level of type IIB by (see [@6]) $${\mathcal K}=-3 \ln \big[-i (\rho - {\bar \rho}) \big] - \ln \big[-i (\tau - {\bar \tau}) \big] - \ln \big[-i \int_{\mathcal M_6} \Omega \wedge {\bar \Omega} \,\big]\,\,.$$ In (2.3), $\tau$ is type IIB axion/dilaton and $\Omega$ is holomorphic three-form of the Calabi-Yau threefold ${\mathcal M}_6$.
From the superpotential $W$ and the K${\ddot{\rm a}}$hler potential $\mathcal K$ one can construct the scalar potential of the ${\mathcal N}=1$ supergravity [@1-1; @6]: $${\mathcal V}_{\rm scalar} = \frac{1}{2 \kappa^2_{10}} e^{\mathcal K} \Big(\, {\mathcal G}^{a {\bar b}} D_a W \,\overline{D_b W} -3 |W|^2 \,\Big) \,\,,$$ where ${\mathcal G}_{a {\bar b}}=\partial_a \partial_{\bar b} {\mathcal K}$, and $a$, $b$ are summed all over the complex structure moduli $\tau^I$, the axion/dilaton $\tau$ and the K${\ddot{\rm a}}$hler modulus $\rho$. For the no-scale structure [@6; @7] in which $W=W_0$, (2.4) reduces to $${\mathcal V}_{\rm scalar} = \frac{1}{2 \kappa^2_{10}} e^{\mathcal K} \Big(\, {\mathcal G}^{i {\bar j}} D_i W_0 \,\overline{D_j W_0} \Big) \,\,,$$ where $i$, $j$ are now summed over $\tau^I$ and $\tau$, and the superpotential $W_0$ is given by $$W_0 = \int_{\mathcal M_6} G_{(3)} \wedge \Omega \,\,, ~~~~~~~~~~ \big( G_{(3)} = F_{(3)} - \tau H_{(3)} \big)\,\,,$$ where $F_{(3)}$ and $H_{(3)}$ are R-R resp. NS-NS three-form field strengths. If we take $F_{(3)}$ and $H_{(3)}$ as $F_{(3)}$, $H_{(3)}$ $\in H^3 ({\mathcal M}_6 , {\mathcal Z})$, then the potential (2.5) fixes the moduli at values for which $G_{(3)}$ is imaginary self-dual (ISD) at the tree level [@2]. But once the nonperturbative term comes in, $G_{(3)}$ will not be ISD anymore. Concerning this point a little more explanation may be necessary as follows.
In the original KKLT the authors used a two step procedure in which they first fix the complex structure moduli (and also the dilaton moduli as well) at values where $G_{(3)}$ becomes ISD, and then fix the K$\ddot{\rm a}$hler modulus by introducing nonperturbative correction to the superpotential (see (2.1)) in such a way that $G_{(3)}$ still remains ISD. This is possible if the masses of the complex structure moduli and the dilaton moduli are much larger than the mass of the K$\ddot{\rm a}$hler modulus. Indeed in KKLT the complex structure moduli are fixed at string scale and they are integrated out. Hence in KKLT the instanton determinant $A$ in (2.1) is effectively a constant and the shifts of the complex structure moduli from their classical ISD positions are consequently negligible.
But after this original KKLT, there also came out some other articles in which the KKLT mechanism of moduli stabilization is extended to more general cases where the complex structure (and the dilaton) moduli are not integrated out anymore and hence they appear explicitly in the effective theory [@8; @9]. In these theories the instanton determinant depends on the complex structure moduli and $G_{(3)}$ now acquires imaginary anti self-dual (IASD) components by the nonperturbative corrections. In our present paper we want to extend our discussions as much as possible so that the self-tuning mechanism of this paper can be applied even to these generalized theories as well. Hence in our paper we will include these IASD components when we investigate the whole possible contributions to the scalar potentials of the background vacua. We will be back to this point later. (See for instance the paragraph below eq. (4.3).)
Turning back to the superpotential (2.1), one can concentrate only on the K${\ddot{\rm a}}$hler modulus $\rho$ if we neglect the no-scale part (2.5) (for a moment) as in the original KKLT. (But see the first paragraph of Sec. 4.1.) The scalar potential is therefore given by $${\mathcal V}_{\rm scalar} = \frac{1}{2 \kappa^2_{10}} e^{\mathcal K} \Big(\, {\mathcal G}^{\rho {\bar \rho}} D_{\rho} W \,\overline{D_{\rho} W} -3 |W|^2 \,\Big) \,\,,$$ and using (2.1) one obtains [@2] $${\mathcal V}_{\rm scalar} = \frac{1}{2 \kappa^2_{10}}\, e^{{\mathcal K}_{\tau}+{\mathcal K}_{\rm cs}}\Bigg[ \frac{a A e^{-a \sigma}}{2 \sigma^2} \Big(\, \frac{1}{3} a A \sigma e^{-a \sigma} + W_0 + A e^{-a \sigma} \Big) \Bigg] \,\,,$$ where the axion in $\rho$ has been set to zero and $\sigma$ is defined by $\rho = i\sigma$. From (2.7) one finds that the minimum of ${\mathcal V}_{\rm scalar}$ takes negative values for the superpotential $D_a W =0$, and therefore it describes supersymmetric AdS vacua because ${\mathcal V}_{\rm scalar}$ is identified with the four-dimensional cosmological constant $\lambda$ in the usual flux compactifications including KKLT.
At the final step of KKLT the AdS minimum is uplifted to a dS minimum by the anti-${D3}$-branes introduced at the tip of the KS throat where the introduction of anti-${D3}$-branes does not violate the tadpole condition. By this process ${\mathcal V}_{\rm scalar}$ acquires an additional term[^3] $$\delta {\mathcal V}_{\rm scalar} = \frac{D}{\sigma^2} \,\,,$$ where $D$ is a positive constant proportional to the number of anti-${D3}$-branes. So in [@2] Kachru $et~al.$ obtain dS vacua by fine-tuning the constant $D$ so that the minimum of the resulting ${\mathcal V}_{\rm scalar}$ becomes very close to zero.
0.5cm
In general the dS vacua uplifted by anti-$D3$-branes can have two different kinds of tunneling instabilities (see Sec. 5.2), one of which is related to the fact that $\lambda$ is directly given by ${\mathcal V}_{\rm scalar}$ alone in these theories. The scalar potential ${\mathcal V}_{\rm dS}$ of the dS minimum at $\sigma =\sigma_m$ takes positive (though it is very small) values, while ${\mathcal V}_{\rm scalar}$ asymptotically vanishes, ${\mathcal V}_{\rm scalar}|_{\sigma \rightarrow \infty}=0$. So these dS vacua are only local minima of the potential which eventually decay into the run away vacuum at $\sigma =\infty$ which corresponds to a Minkowski space with a large (or a decompactified) internal Calabi-Yau volume.
In the original KKLT, however, it was shown that the lifetime of the dS vacua is larger than the cosmological time scale of $10^{10}$ years in certain approximations. So KKLT does not suffer from this tunneling instability problem. Besides this, the elegance of KKLT is that all stringy corrections are very small. Both $g_s$- and $\alpha^{\prime}$-corrections are small in the part of moduli space their vacuum lives and hence the quantum corrections are only subleading.
But still, though the KKLT is an attractive scenario for the late-time cosmology with a small positive cosmological constant, it has some difficulties as for being a realistic model of our universe, especially when looking at from a standpoint of the cosmological constant problem. In this paper we propose a new self-tuning mechanism in which the fine-tuning $\lambda=0$ is automatically achieved by a certain constraint (or a self-tuning) equation. As shown in (1.1) $\lambda$ contains an extraordinary term ${\mathcal E}_{\rm SB}$ which possesses gauge arbitrariness, and the whole quantum fluctuations $\delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ on the branes and nonzero contributions to ${\mathcal V}_{\rm scalar}$ coming from perturbative and nonperturbative corrections are all gauged away by ${\mathcal E}_{\rm SB}$ (and by a self-tuning equation as mentioned above) and as a result the fine-tuning $\lambda =0$ is always preserved.
Beside this, the instabilities of the background vacua are innately absent in our case. As described above $\lambda$ contains an additional term ${\mathcal E}_{\rm SB}$, and hence in (1.1) we are allowed to take negative values for ${\mathcal V}_{\rm scalar}$ at the minimum $\sigma =\sigma_m$ because in our case ${\mathcal V}_{\rm scalar} <0$ does not directly imply a negative $\lambda$ due to the presence of this ${\mathcal E}_{\rm SB}$. Indeed in our self-tuning mechanism the background state of our present universe will be identified with one of the AdS (rather than dS) vacua of KKLT. (See the AdS vacuum scenario proposed in Sec. 5.2.) So the instabilities of dS vacua described above are essentially irrelevant to our case. In this section we will discuss about the basic principle of our self-tuning mechanism described above, together with brief reviews of some formulas and ideas presented in [@1] if necessary for reader’s convenience.
0.3cm 0.15cm
In the string frame the type IIB action is given by $$I_{\rm IIB} = \frac{1}{2 \kappa_{10}^2} \int d^{10} x \sqrt{-G} \Big\{ e^{-2 \phi} \big[ {\mathcal R}_{10} + 4 (\nabla \phi)^2 \big] - \frac{1}{2} F_{(1)}^2 - \frac{1}{2 \cdot 3!} G_{(3)} \cdot {\bar G}_{(3)} - \frac{1}{4 \cdot 5!} {\tilde F}_{(5)}^2 \Big\} \nonumber$$ $$+ \frac{1}{8 i \kappa_{10}^2} \int e^{\phi} A_{(4)} \wedge G_{(3)} \wedge {\bar G}_{(3)} \,\,,$$ where $\phi$ is the dilaton with $e^{\phi} = g_s e^{\hat{\phi}}$, and $F_{(1)} =dA_{(0)}$, ${\tilde F}_{(5)} =F_{(5)} -\frac{1}{2} A_{(2)} \wedge H_{(3)} + \frac{1}{2} B_{(2)} \wedge F_{(3)}$ with $F_{(n+1)} =dA_{(n)}$. Among these field strengths, ${\tilde F}_{(5)}$ is self-dual and the ansatz is given by $${\tilde F}_{(5)} = (1+ \ast_{10}) d \xi (y) \wedge \sqrt{-g_4}\,\, dx^0 \wedge dx^1 \wedge dx^2 \wedge dx^3 \,\,.$$ In addition to this we have the local terms $$I_{\rm brane} = - \int d^4 x \sqrt{-det \,(G_{\mu\nu})}\,\, T(\phi) + \mu (\phi) \int A_{(4)} \,\,,$$ where $G_{\mu\nu}$ is a pullback of the target space metric $G_{MN}$ to the four-dimensional brane world. Also $T(\phi)$ represents the tension of the $D3$-brane and at the tree level it is given by $T(\phi) = T_3 e^{-\phi}$. But at the quantum level it becomes $T(\phi) = T_3 e^{-\phi} + \rho_{\rm vac} (\phi)$, where $\rho_{\rm vac} (\phi)$ represents quantum correction terms (see for instance ref. [@10]) and it is identified with NS-NS sector vacuum energy density of the three-dimensional space. Similarly, $\mu (\phi)$ is simply $\mu (\phi) = \mu_3$ at the tree level. But it turns into $\mu (\phi)=\mu_3 + \delta \mu (\phi)$ at the quantum level, where $\delta \mu (\phi)$ is an R-R counterpart of $\rho_{\rm vac} (\phi)$ representing R-R sector vacuum energy density of the three-dimensional space.
Upon reduction (see Sec. IV of [@1]) $$ds_{10}^2 = e^{B(y)} g_{\mu\nu}(x) dx^{\mu} dx^{\nu} + e^{{\hat \phi}(y) - B(y)} h_{mn}(y) dy^m dy^n \,\,,$$ where $\mu,\nu =(0,1,2,3)$, $m,n =(5, \cdots , 10)$, the type IIB action (3.1) reduces into $$I_{\rm IIB} = \frac{1}{2 \kappa_{10}^2 g_s^2} \Big( \int d^4 x \sqrt{-g_4} {\mathcal R}_4 (g_{\mu\nu}) \, \Big)\Big(\int d^6 y \,\sqrt{h_6} e^{{\hat \phi} - 2B} \,\Big) + \frac{1}{2 \kappa_{10}^2 g_s^2} \Big( \int d^4 x \sqrt{-g_4} \,\Big) \nonumber$$ $$\times \Big( \int d^6 y \,\sqrt{h_6} \, \big( {\mathcal R}_6 (h_{mn}) - {\mathcal L}_F \big) \Big) + {\rm topological~\,\, term} \,\,,~~~~~~~~~$$ where ${\mathcal L}_F$ is given by $${\mathcal L}_F = (\partial \hat{\phi})^2 - 2(\partial \hat{\phi})(\partial B) + 2 (\partial B)^2 + \frac{g_s^2}{2} e^{2 \hat{\phi}} (\partial A_{(0)} )^2 - \frac{g_s^2}{2} e^{2 \hat{\phi} -4B} (\partial {\xi}\,)^2 \nonumber$$ $$+ \frac{g_s^2}{2 \cdot 3!} e^{2B} G_{mnp} \bar{G}^{mnp} \,\,,~~~~~~~~~~~~~~~~~~~~$$ (But see also the sentences below eq. (3.17). At the quantum level the Lagrangian ${\mathcal L}_F$ can also include off-shell contributions coming from perturbative and nonperturbative corrections. See Sec. VII as an example.) and the topological term comes from the Chern-Simons term $\int e^{\phi} A_{(4)} \wedge G_{(3)} \wedge \bar{G}_{(3)}$, which does not involve any moduli (except the dilaton $\tau$) or the metric. From (3.5) the six-dimensional action defined on the internal space can be written as $$I_{\rm IIB}/\Big(\int d^4 x \sqrt{-g_4}\Big) = \frac{1}{2 \kappa_{10}^2 g_s^2} \int d^6 y \,\sqrt{h_6} \, \Big( {\mathcal R}_6 (h_{mn}) - {\mathcal L}_F + \beta e^{{\hat \phi} - 2B} \,\Big) + {\rm topological~term}\,\,,$$ where $\beta$ is defined by $$\beta = \frac{\int d^4 x \sqrt{- g_4} {\mathcal R}_4 (g_{\mu\nu})}{\int d^4 x \sqrt{- g_4} } \,\,,$$ and hence on the brane, $\beta= 4 \lambda$ for the maximally symmetric spacetime. Varying (3.7) with respect to $\delta h^{mn}$ one obtains $$\mathcal R_{mn} - \frac{1}{2} h_{mn} \mathcal R_6 - \frac{1}{2} T_{mn} - \frac{\beta}{2} e^{{\hat\phi}-2B} h_{mn}=0 \,\,,$$ where the energy momentum tensor $T_{mn}$ is defined by $$T_{mn} = \frac{2}{\sqrt{h_6}} \frac{\delta I_F}{\delta h^{mn}}\,\,, ~~~~~~\big( I_F \equiv \int d^6 y \sqrt{h_6} {\mathcal L}_F \big) \,\,.$$ (3.9) does not involve local terms arising from (3.3) because $D3$-branes do not couple to the unwarped metric $h_{mn}$ in the action (3.3).
In (3.9), $\mathcal R_{mn}$ and $\mathcal R_6$ vanish at the classical level because the internal Calabi-Yau is Ricci-flat. But at the quantum level, $h_{mn}$ acquires correction terms in the equations of motion, $$h_{mn} = h_{mn}^{(0)} + h_{mn}^{(1)} + h_{mn}^{(2)} + \cdots \,= h_{mn}^{(0)} + \delta_Q h_{mn} \,\,,$$ in our perturbation scheme (see (7.13)). Hence in (3.9) (and also in what follows) we can not take $\mathcal R_{mn} (h_{mn}) = \mathcal R_6 (h_{mn})=0$ at the quantum level though we have $\mathcal R_{mn} (h_{mn}^{(0)}) = \mathcal R_6 (h_{mn}^{(0)}) = 0$, because they do not vanish at off-shell. Besides the perturbations, there are also backreactions of the fluxes and local sources like $D3$-branes which carry standard model fields etc. These backreactions on the internal geometry also could yield $R_{mn} \neq 0$ and $\mathcal R_6 \neq 0$. In this paper such deformations of internal geometry caused by perturbations and backreactions are all under consideration because we never set $\mathcal R_{mn} = \mathcal R_6 =0$ in the whole procedure of our discussions as mentioned above.[^4] The nonzero $\mathcal R_{mn}$ and $\mathcal R_6$ cancel out themselves during the process of obtaining the self-tuning equation (3.30).
0.3cm 0.15cm
The four-dimensional action defined on the external space can be obtained by rewriting (3.5) as $$I_{\rm IIB} = \frac{1}{2\kappa^2} \int d^4 x \sqrt{-g_4} {\mathcal R}_4 (g_{\mu\nu}) + \int d^4 x \sqrt{-g_4} {\hat I}_{\rm bulk} + {\rm topological~term}\,\,,$$ where ${2\kappa^2} \equiv 2 \kappa_{10}^2 g_s^2 / \big( \int d^6 y \,\sqrt{h_6} e^{{\hat \phi} - 2B}\big)$ and ${\hat I}_{\rm bulk}$ is defined by $${\hat I}_{\rm bulk} = \frac{1}{2 \kappa_{10}^2 g_s^2} \int d^6 y \,\sqrt{h_6} \big( {\mathcal R}_6 (h_{mn}) - {\mathcal L}_F \big) \,\,.$$ Adding (3.3) to (3.12) one can show that the total action $I_{\rm IIB} + I_{\rm brane}$ can be written in the form $$I_{\rm total} = \frac{1}{2\kappa^2} \int d^4 x \sqrt{-g_4} \big( {\mathcal R}_4 (g_{\mu\nu}) - 2 \lambda \big) + {\rm topological~term}\,\,,$$ where the cosmological constant $\lambda$ is defined by $$\lambda = - \kappa^2 [ {\hat I}_{\rm bulk} + {\hat I}_{\rm brane}] \,\,,$$ where ${\hat I}_{\rm brane} \equiv I_{\rm brane}/ \Big(\int d^4 x \sqrt{-g_4}\Big)$.
Turning back to (3.6) we see that the Lagrangian ${\mathcal L}_F$ can be written as $${\mathcal L}_F = K - V\,\,, ~~~~~~~~~~(\,K=h^{mn}K_{mn}\,) \,\,,$$ where $K_{mn}$ and $V$, the kinetic and potential parts of the Lagrangian, take respectively the forms $K_{mn} = \sum_{A,B} F_{AB} (\phi_C ) \partial_{m} \phi_A \partial_{n} \phi_B$ and $V=V(\phi_A , h^{mn})$, where $\phi_A$’s represent the six-dimensional scalar fields such as $\hat{\phi}$, $B$, $A_{(0)}$ and $\xi$ etc. Namely in (3.16), while $V$ involves $h^{mn}$, $K_{mn}$ does not. Also in (3.16), $V$ is related to ${\mathcal V}_{\rm scalar}$ by the equation $${\mathcal V}_{\rm scalar} = \frac{1}{2 \kappa_{10}^2 g_s^2} \int d^6 y \sqrt{h_6} V \,\,,$$ and thus for the no-scale structure the potential density $V$ arising from the fluxes is identified with $-\frac{g_s^2}{3!} G_{(3)}^{\rm IASD} \cdot {\bar G}_{(3)}^{\rm IASD}$ (see (4.3)) in the case of type IIB action. But in general, $V$ also includes the off-shell contributions coming from perturbative and nonperturbative corrections including those, for instance the $D3$-brane potential induced by IASD fluxes in Sec VII etc.
Now we substitute (3.10) - with ${\mathcal L}_F$ given by (3.16) - into (3.9) and contract the indices $m$ and $n$. Then we obtain $${\mathcal R}_6 - {\mathcal L}_F - \frac{1}{2} ({\mathcal N}-1) V +\frac{3}{2} \beta e^{{\hat \phi} - 2B} =0 \,\,,$$ where ${\mathcal N}$ is defined by ${\mathcal N}\equiv h^{mn} \frac{\partial}{\partial h^{mn}}$. Again, we do not take ${\mathcal R}_6 =0$ in (3.18) because $h_{mn}$ in ${\mathcal R}_6$ (and other fields in (3.18) as well) contains correction terms coming from perturbatons. But integrating (3.18) and using (3.13) one finds that $${\hat I}_{\rm bulk}= - \frac{3 \beta}{4 \kappa^2} + \frac{1}{4\kappa_{10}^2 g_s^2} \int d^6 y \,\sqrt{h_6} \big( {\mathcal N}-1 \big) V \,\,,$$ and substituting (3.19) into (3.15) (and using $\beta = 4 \lambda$) one finally obtains $$\lambda = \frac{\kappa^2}{8\kappa_{10}^2 g_s^2} \int d^6 y \,\sqrt{h_6} \big( {\mathcal N}-1 \big) V + \frac{\kappa^2}{2} {\hat I}_{\rm brane} \,\,,$$ which is now independent of ${\mathcal R}_6 (h_{mn})$.
0.3cm 0.15cm
Now we proceed to obtain a self-tuning equation for $\lambda$, which is one of the main points of this paper. First, we substitute ${\mathcal L}_F$ in (3.18) into (3.10) to get $$T_{mn} = 2 ({\mathcal R}_{mn} - \frac{1}{2} h_{mn} {\mathcal R}_{6}) + \frac{1}{2} h_{mn} ({\mathcal N} - 1) V - \frac{\partial}{\partial h^{mn}}({\mathcal N} - 1) V - \frac{3}{2} \beta e^{{\hat \phi}-2B} h_{mn} \,\,.$$ Next, substitute (3.21) into (3.9) and contract $m$ and $n$. Then we obtain $$\beta = - \frac{1}{3} \chi^{1/2} ({\mathcal N} - 1)({\mathcal N} -3) V \,\,,~~~~~~\big( \chi^{1/2} \equiv e^{2B -{\hat \phi}} \big)\,\,.$$ Let us repeat the same procedure again. Substitute (3.22) into (3.18) to obtain $${\mathcal L}_F = {\mathcal R}_6 - \frac{1}{2} ({\mathcal N} - 1 )({\mathcal N} -2) V \,\,.$$ Next, substitute (3.23) into (3.10) to obtain $$T_{mn} = 2 \big( \mathcal R_{mn} - \frac{1}{2} h_{mn} \mathcal R_6 \big) + \frac{1}{2} h_{mn} ({\mathcal N} - 1)({\mathcal N}- 2)V
- \frac{\partial}{\partial h^{mn}} ({\mathcal N} - 1)({\mathcal N} - 2) V \,\,.$$ Finally, substitute (3.24) back into (3.9) and contract $m$ and $n$. We obtain $$\beta = \frac{1}{6} \, \chi^{1/2} ({\mathcal N} - 1)({\mathcal N} -2)({\mathcal N} -3) V \,\,.$$
(3.22) and (3.25) suggest that $\beta$ always contains the operators $({\mathcal N} -1)$ and $({\mathcal N} -3)$ in common. We prove this as follows. First, we observe that $\beta$’s in (3.22) and (3.25) both take the form $$\beta = b_0 \chi^{1/2} ({\mathcal N} -1) \Pi ({\mathcal N}) V \,\,,$$ where $b_0$ is a constant and $\Pi ({\mathcal N})$ is an operator of the form $$\Pi ({\mathcal N}) = \sum_{k} c_k ({\mathcal N} -n_1 ) \cdots ({\mathcal N} -n_k ) \,\,,$$ where $n_i$ are integers. So we start by assuming that $\beta$ always appears in the form (3.26). Now we substitute (3.26) into (3.18) to obtain $${\mathcal L}_F = {\mathcal R}_6 - \frac{1}{2} ({\mathcal N} - 1) (1-3 b_0 \Pi ({\mathcal N})) V \,\,.$$ Next, substitute (3.28) into (3.10) to obtain $$T_{mn} = 2 \big({\mathcal R}_{mn} -\frac{1}{2} h_{mn} {\mathcal R}_6 \big) + \frac{1}{2} h_{mn} ({\mathcal N} - 1) (1-3 b_0 \Pi ({\mathcal N})) V - \frac{\partial}{\partial h^{mn}} ({\mathcal N} - 1) (1-3 b_0 \Pi ({\mathcal N})) V \,\,.$$ Finally, substitute (3.29) back into (3.9) and contract $m$ and $n$. Then we obtain $$\beta = \frac{1}{6} \chi^{1/2} ({\mathcal N} -1) ({\mathcal N} -3) (1-3 b_0 \Pi ({\mathcal N})) V \,\,.$$ (3.30) takes the form (3.26) again, which ensures that the prerequisite assumption (3.26) on $\beta$ is valid. Also (3.30) shows that $\beta$ always contains $({\mathcal N} -1)$ and $({\mathcal N} -3)$ acting on $V$, which proves the proposition.
We obtained (3.30) starting from the Einstein equation (3.9). But (3.30) does not contain ${\mathcal R}_6$ of the perturbed $h_{mn}$ because it has canceled out during the process of obtaining (3.30). This suggests that $\lambda$ (Recall that $\beta = 4\lambda$) is not affected by the deformations of internal geometry caused by quantum fluctuations (and backreactions which also could yield ${\mathcal R}_{mn} \neq 0$ and ${\mathcal R}_6 \neq 0$) at least in the supergravity framework. Besides this, (3.30) suggests a very important fact. According to (3.30), whether $\lambda$ vanishes or not is entirely determined by the tensor structure of $V$, not by any other factors like zero or nonzero values of the scalar potential ${\mathcal V}_{\rm scalar}$ etc. We will be back to this point in Sec. 3.5.
0.3cm 0.15cm
0.3cm 0.15cm
(3.3) shows that $I_{\rm brane}$ consists of two (NS-NS and R-R) parts. Among these two, the second part represents an electric coupling of $D3$-branes to the R-R four-form $A_{(4)}$ and it can be rewritten as $$I_{\rm brane}^{(R)} = \frac{1}{4!} \int d^4 x A_{\mu_0 \mu_1 \mu_2 \mu_3} J^{\mu_0 \mu_1 \mu_2 \mu_3} \,\,,$$ where $J^{\mu_0 \mu_1 \mu_2 \mu_3}$ is the world volume current density of the $D3$-brane, $$J^{\mu_0 \mu_1 \mu_2 \mu_3} = \mu_3 \epsilon^{\alpha_0 \alpha_1 \alpha_2 \alpha_3} \big(\frac{\partial X^{\mu_0}}{\partial x^{\alpha_0}} \big) \cdots \big(\frac{\partial X^{\mu_3}}{\partial x^{\alpha_3}} \big) \,\,.$$ At the classical level $J^{\mu_0 \mu_1 \mu_2 \mu_3}$ is just a solitonic current density, $J^{\mu_0 \mu_1 \mu_2 \mu_3}_{\rm sol}$, representing classical world volume dynamics of the $D3$-brane. In that case $X^{\mu} (x)$’s in (3.32) stand for the classical fields, $X^{\mu}_{\rm cl} (x)$, defined on the world volume of the $D3$-brane, and for the embedding $X^{\mu}_{\rm cl} (x) = x^{\mu}$, $J^{0123}_{\rm sol}$ is simply $\mu_3$. At the quantum level, however, $X^{\mu}(x)$’s include fluctuations $X^{\mu^{\prime}}$, $X^{\mu} = X^{\mu}_{\rm cl} + X^{\mu^{\prime}}$.
Since $X^{\mu^{\prime}}$’s are fluctuations of the open string degrees of freedom, they correspond to the fluctuations of the standard model fields with support on the $D3$-brane. Due to these fluctuations $J^{\mu_0 \mu_1 \mu_2 \mu_3}$ acquires an additional term, $J^{\mu_0 \mu_1 \mu_2 \mu_3} = J^{\mu_0 \mu_1 \mu_2 \mu_3}_{\rm sol} + <\chi^{\mu_0 \mu_1 \mu_2 \mu_3}_{\rm vac}>$, where $<\chi^{\mu_0 \mu_1 \mu_2 \mu_3}_{\rm vac}>$ represents quantum corrections corresponding to the fluctuations (of the standard model degrees of freedom with support) on the $D3$-brane. Denoting $J^{0 1 2 3}_{\rm sol}$ and $<\chi^{0 1 2 3}_{\rm vac}>$, respectively, by $\mu_3$ and $\delta \mu (\phi)$, one can rewrite (3.31) as $$I_{\rm brane}^{(R)} = \big[ \int d^4 x \sqrt{-g_4} \,\big] \, \int d^6 y \sqrt{h_6} \,\, \mu (\phi) \xi(y) \delta^6 (y) \,\,,$$ where we have used $$A_{(4)} = \xi(y) \sqrt{-g_4} \,\, dx^0 \wedge dx^1 \wedge dx^2 \wedge dx^3 \,\,,$$ and the normalization convention $\int d^6 y \sqrt{h_6} \,\, \delta^6 (y) =1$ of the six-dimensional delta function. (3.33) coincides with the second term of (3.3), and where $\mu(\phi) = \mu_3 + \delta \mu (\phi)$ as before.
0.3cm 0.15cm
Going back to the classical level, the second term of (3.3) is invariant under the gauge transformation $A_{(4)} \rightarrow A_{(4)} + \delta A_{(4)}$ with $\delta A_{(4)} = d \Lambda_{(3)}$, where $\Lambda_{(3)}$ is an arbitrary three-form. Indeed $\delta_G I_{\rm brane}^{(R)}$ vanishes for $\delta A_{(4)} = d \Lambda_{(3)}$ : $\delta_G I_{\rm brane}^{(R)} = \mu_3 \int_{\partial \Sigma} \Lambda_{(3)} =0$ because $\Lambda_{(3)}$ is assumed to vanish at the boundary $\partial \Sigma$ of the four-dimensional spacetime. But once we go up to quantum level, $I_{\rm brane}^{(R)}$ is not gauge invariant anymore. The reason is because while $J^{\mu_0 \mu_1 \mu_2 \mu_3}_{\rm sol}$ satisfies $\partial_{\mu_0} J^{\mu_0 \mu_1 \mu_2 \mu_3}_{\rm sol}=0$, the off-shell quantity $<\chi^{\mu_0 \mu_1 \mu_2 \mu_3}_{\rm vac}>$ does not necessarily satisfy $<\partial_{\mu_0} \chi^{\mu_0 \mu_1 \mu_2 \mu_3}_{\rm vac}> =0$. So the total $J^{\mu_0 \mu_1 \mu_2 \mu_3}$ is not locally conserved at the quantum level, and the gauge transformation $\delta A_{\mu_0 \mu_1 \mu_2 \mu_3} = 4 \partial_{[\mu_0}\Lambda_{\mu_1 \mu_2 \mu_3 ]}$ generally induces a nonzero variation of $I_{\rm brane}^{(R)}$. Integrating by part one obtains from (3.31) that $$\delta_G I_{\rm brane}^{(R)} = - \frac{1}{3!} \int d^4 x \Lambda_{\mu_1 \mu_2 \mu_3} <\partial_{\mu_0} \chi^{\mu_0 \mu_1 \mu_2 \mu_3}_{\rm vac}> \,\,,$$ which generally takes nonzero values because so does $<\partial_{\mu_0} \chi^{\mu_0 \mu_1 \mu_2 \mu_3}_{\rm vac}>$.
In addition to (3.35), there is another important variation of $I_{\rm brane}^{(R)}$ which plays a crucial role in our self-tuning mechanism. To find its explicit form, rewrite $\delta_G I_{\rm brane}^{(R)}$ as $\delta_G I_{\rm brane}^{(R)} = \mu_3 \int d \Lambda_{(3)}$ and take an ansatz [@1] $$\Lambda_{(3)} = F(y) \sqrt{-g_4} dx^1 \wedge dx^2 \wedge dx^3 \,\,,$$ where $F(y)$ is an arbitrary function of the internal coordinates $y^m$. (3.36) is the most appropriate ansatz which accords with (3.34) and therefore respects the Poincar$\acute{\rm e}$ symmetry of our four-dimensional spacetime. Once we take $\Lambda_{(3)}$ as in (3.36), $\delta_G I_{\rm brane}^{(R)}$ in (3.35) vanishes because $\partial_{[\mu_0} \Lambda_{\mu_1 \mu_2 \mu_3 ]} =0$ for a constant $\sqrt{-g_4}$. (Indeed $\sqrt{-g_4}$ is constant when $\lambda =0$. See below.) However, $\Lambda_{(3)}$ in (3.36) generates another type of $\delta_G I_{\rm brane}^{(R)}$ as shown below.
Taking derivative to $\Lambda_{(3)}$ one obtains $$\delta_G I_{\rm brane}^{(R)} = \int d^4 x \sqrt{-g_4} f_m (y) J^{m123} + \frac{3}{2} \int d^4 x \sqrt{-g_4} H\,F(y) J^{0123} \,\,,$$ where $f_m (y)$($\equiv \partial_m F(y)$) represents $\delta A_{m123} / \sqrt{-g_4}$, and $H$($\equiv (2/3) \partial_0 \ln \sqrt{-g_4}$) is the Hubble constant of the four-dimensional spacetime $ds_4^2 = -dt^2 + e^{Ht} d {\vec x}_3$, which therefore vanishes for $\lambda=0$ because $\lambda \propto H^2$. In (3.37), $J^{m123}$ is defined by $$J^{m123} = \mu_3 \epsilon^{\alpha_0 \alpha_1 \alpha_2 \alpha_3} \big(\frac{\partial Y^m}{\partial x^{\alpha_0}} \big) \wedge \big(\frac{\partial X^1}{\partial x^{\alpha_1}} \big) \wedge \big(\frac{\partial X^2}{\partial x^{\alpha_2}} \big) \wedge \big(\frac{\partial X^3}{\partial x^{\alpha_3}} \big)\,\,,$$ which, at the classical level, vanishes for the embedding $X^{\mu}_{\rm cl} (x) = x^{\mu}$ because ${\partial Y^m_{\rm cl}}/{\partial x^{\alpha_0}}=0$. So the nonzero contribution to $J^{m123}$ comes from the quantum excitations $<\chi^{m123}_{\rm vac}>$. Denoting $<\chi^{m123}_{\rm vac}>$ by $\delta \mu_{\rm T}^m (\phi)$ (and omitting the second term) one can rewrite (3.37) as $$\delta_G I_{\rm brane}^{(R)} = \big(\int d^4 x \sqrt{-g_4}\big) \int dy \sqrt{h_6} \delta \mu_{\rm T}^m (\phi) f_m (y) \delta^6 (y) \,\,,$$ where $f_m (y)$’s are arbitrary functions of $y^m$, representing (derivatives of) local gauge parameters.
0.3cm 0.15cm
From (3.3) and (3.39) (or from (8.1)) one finds that at the quantum level the brane action consists of various parts, $$I_{\rm brane} = \Big( I_{\rm brane}^{(NS)}({\rm tree}) + I_{\rm brane}^{(R)}({\rm tree}) \Big) + \Big( \delta_Q I_{\rm brane}^{(NS)} + \delta_Q I_{\rm brane}^{(R)} \Big) + \delta_G I_{\rm brane}^{(R)}\,\,.$$ Among these terms, $I_{\rm brane}^{(NS)}(\rm tree)$ and $I_{\rm brane}^{(R)}(\rm tree)$ are the tree level actions and they always cancel out by field equations for the BPS $D3$-branes. (See, for instance, Sec VI.C of [@1] for this.) The correction terms $\delta_Q I_{\rm brane}^{(NS)}$ and $\delta_Q I_{\rm brane}^{(R)}$ arise from $\rho_{\rm vac} (\phi)$ and $\delta \mu (\phi)$, and they represent quantum fluctuations (of the gravitational and standard model fields with support) on the $D3$-brane. So $\delta_Q I_{\rm brane}^{(NS)} + \delta_Q I_{\rm brane}^{(R)}$ correspond to the gravitational plus electroweak and QCD vacuum energies of the standard model configurations of the brane region. These two terms are conjectured to cancel out to all orders of perturbations in supersymmetric theories, but they do not when supersymmtry (of the brane region) is broken. (The cancelation at one-loop order on the BPS $D3$-branes has been proven explicitly for the case $G_{(3)} =0$ in Sec. VIIA of [@1].) In our self-tuning mechanism, however, it is not important whether such a cancelation occurs or not, as we will see in what follows.
Using $\beta = 4 \lambda$, one can rewrite (3.30) as $$\lambda = \frac{1}{24} \chi^{1/2} ({\mathcal N} -1)({\mathcal N} -3)\big(1- 3 b_0 \Pi({\mathcal N})\big)V \,\,.$$ (3.41) requires that $\lambda$ must vanish once $V$ belongs to $V_n$ ($V \in V_n$) with $n=1$ or 3, where $V_n$ represents a class of potential densities satisfying $${\mathcal N} V_n = n V_n \,\,.$$ Aside from this, one also finds that if $V \in V_n$, (3.20) becomes $$\lambda = \frac{(n-1)}{4} \kappa^2 {\mathcal V}_{\rm scalar} + \frac{\kappa^2}{2} {\hat I}_{\rm brane} \,\,$$ by (3.42) and (3.17). So if $V \in V_1$, $\lambda$ is simply $\lambda = \frac{\kappa^2}{2} {\hat I}_{\rm brane}$. But if $V \in V_3$, then $\lambda$ becomes $$\lambda = \frac{\kappa^2}{2} \Big({\mathcal V}_{\rm scalar} + {\hat I}_{\rm brane} \Big) \,\,,$$ and in both cases $\lambda$ vanishes by (3.41). In our scenario proposed in Sec. 5.2, the background vacuum of our present universe is identified with one of the AdS vacua of KKLT, and in Secs. IV and VI it will be shown that these AdS vacua all belong to $V_3$, $V_{\rm AdS} \in V_3$. So in our case $\lambda$ is basically given by (3.44) and it must vanish by the self-tuning equation (3.41). (But in the next paragraphs we will show that (3.44) becomes (1.1) by (3.40). So in our AdS vacuum scenario $\lambda$ is basically given by (1.1) and it must vanish by the self-tuning equation (3.41).)
Let us go back to (3.40). We have seen in Sec. 3.4 that the last term $\delta_G {\hat I}_{\rm brane}^{(R)}$ represents the magnitude of gauge symmetry breaking of ${\hat I}_{\rm brane}^{(R)}$ (caused by an anomaly $<\partial_m \chi_{\rm vac}^{m123}> \neq 0$) arising at the quantum level, where $<\chi_{\rm vac}^{m123}>$ are quantum excitations on the branes with components along the transverse directions of the $D3$-branes. In [@1], it was shown that $\delta_G {\hat I}_{\rm brane}^{(R)}$ is closely related to the supersymmetry breaking of the brane region. It plays the role of a supersymmetry breaking term. (The supersymmetry breaking caused by $\delta_G {\hat I}_{\rm brane}^{(R)}$ is also discussed in detail in Sec. VIII of this paper.) So $\delta_G {\hat I}_{\rm brane}^{(R)}$ is an energy scale of the gauge symmetry breaking (or an anomaly) of the action ${\hat I}_{\rm brane}^{(R)}$, and at the same time it is also an energy scale of the supersymmetry breaking induced by this gauge symmetry breaking of ${\hat I}_{\rm brane}^{(R)}$.
After all, renaming $\delta_G {\hat I}_{\rm brane}^{(R)}$ as $$\delta_G {\hat I}_{\rm brane}^{(R)} \equiv -{\mathcal E}_{\rm SB} \,\,,$$ one obtains (1.1) from (3.40) and (3.44) (Recall that the tree level actions ${\hat I}_{\rm brane}^{(NS)} (\rm tree) + {\hat I}_{\rm brane}^{(\it R)} (\rm tree)$ cancel out for the BPS $D3$-branes.), where ${\mathcal E}_{\rm SB}$ is now $${\mathcal E}_{\rm SB} = - \int d^6 y \sqrt{h_6} \delta \mu_{\rm T}^{m} (\phi) f_m (y) \delta^6 (y) \,\,$$ from (3.39) and (3.45). Note that ${\mathcal E}_{\rm SB}$ contains arbitrary gauge parameters $f_m (y)$. This implies that ${\mathcal E}_{\rm SB}$ possesses gauge arbitrariness. Due to this property of ${\mathcal E}_{\rm SB}$, any nonzero values of ${\mathcal V}_{\rm scalar}$ and $\delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ in (1.1) can be gauged away by this ${\mathcal E}_{\rm SB}$, and as a result $\lambda$ vanishes (by (3.41)) as long as the potential density $V$ satisfies $V \in V_n$ with $n=1$ or 3.
The above self-tuning mechanism is distinguished from the usual theories where $\lambda$ is directly identified with ${\mathcal V}_{\rm scalar}$. In those theories, $\lambda$ is generally unstable under perturbative (radiative) corrections because so is ${\mathcal V}_{\rm scalar}$. Also the dS vacua necessarily imply ${\mathcal V}_{\rm scalar} > 0$, which can lead to a tunneling instability as mentioned in the opening paragraph of this section. But in our self-tuning mechanism described above, these are not to be the cases anymore. $\lambda$ can vanish by (3.41) regardless of whether ${\mathcal V}_{\rm scalar}$ in (1.1) vanishes or not. So we can take ${\mathcal V}_{\rm scalar} < 0$ (while maintaining $\lambda =0$) to avoid the tunneling instability, and the value $\lambda =0$ is always stable against radiative corrections. Any nonzero contributions to ${\mathcal V}_{\rm scalar}$ and quantum fluctuations on the branes are forced to be gauged away by (3.41) as long as $V$ satisfies $V \in V_n$ with $n=1$ or 3, and $\lambda =0$ is automatically achieved in our self-tuning mechanism of this paper. Hence in the following sections we will mainly check if our background configurations really satisfy $V \in V_n$ with $n=1$ or 3.
0.5cm
In no-scale structure (and in the ISD background) $\lambda$ trivially vanishes from (3.41) because potential densities arising from ISD fluxes all vanish. But once the no-scale structure is broken by nonperturbative effects as in AdS vacua of KKLT, the potential density does not vanish anymore because $G_{(3)}$ now acquires IASD components due to the presence of nonperturbative terms in the superpotential $W$. Besides this, the scalar potential (2.4) receives nontrivial contributions from both perturbative and nonperturbative corrections of the superpotential and the K${\ddot{\rm a}}$hler potential. More explicitly, while the superpotential receives only nonperturbative corrections $W=W_{\rm tree} +W_{\rm np}$ as in (2.1) [@11], the K${\ddot{\rm a}}$hler potential receives both perturbative and nonperturbative corrections ${\mathcal K}={\mathcal K}_{\rm tree} + {\mathcal K}_{\rm p} +{\mathcal K}_{\rm np}$. So in order to maintain $\lambda=0$, the potential density of our background vacuum must remain to satisfy (3.42) with $n=1,3$ even under these corrections. In this section we want to show that the AdS vacua of KKLT satisfy the above property. Namely the potential densities of AdS vacua of KKLT belong to $V_3$, and this result does not change under $W=W_{\rm tree} +W_{\rm np}$ and ${\mathcal K}={\mathcal K}_{\rm tree} + {\mathcal K}_{\rm p} +{\mathcal K}_{\rm np}$.
0.3cm 0.15cm
0.3cm 0.15cm
The scalar potential arising from the fluxes can be obtained from the $G_{mnp}{\bar G}^{mnp}$ term of the action (see (3.6)). Rewrite the $G_{mnp}{\bar G}^{mnp}$ term as[^5] [@6] $$-\frac{1}{24 \kappa_{10}^2} \int d^6 y \sqrt{h_6} \,e^{2B} G_{mnp}{\bar G}^{mnp} =\frac{i}{4 \kappa_{10}^2 Im \tau} \int \frac{\chi^{1/2}}{g_s} G_{(3)} \wedge {\bar G}_{(3)} ~~~~~~~~~~~~~~~~~~~~\nonumber$$ $$~~~~~~~~~~- \frac{1}{12 \kappa_{10}^2 Im \tau} \int d^6 y \sqrt{h_6} \, \frac{\chi^{1/2}}{g_s} G_{mnp}^{+}{\bar G}^{+mnp} \,\,,$$ where $$G_{(3)}^{\pm} = \frac{1}{2} ( G_{(3)} \pm i \ast_6 G_{(3)})\,\,, ~~~~~~ \ast_6 G_{(3)}^{\pm} =\mp i G_{(3)}^{\pm} \,\,,$$ are the IASD/ISD parts of $G_{(3)}$, $G_{(3)}^+ = G_{(3)}^{\rm IASD}$ and $G_{(3)}^- = G_{(3)}^{\rm ISD}$. The scalar potential (arising from the fluxes) is defined by the second term of (4.1) as $${\mathcal V}_{\rm no-scale} = \frac{1}{12 \kappa_{10}^2 Im \tau} \int d^6 y \sqrt{h_6} \, \frac{\chi^{1/2}}{g_s} G_{mnp}^{+}{\bar G}^{+mnp} \,\,,$$ which is identified with (2.5) of the four-dimensional effective theory (see Sec. A.2 of [@6]). Since ${\mathcal V}_{\rm no-scale}$ is given as to be $\propto \int G_{(3)}^{\rm IASD} \cdot {\bar G}_{(3)}^{\rm IASD}$, it (and its density as well) vanishes in the ISD compactifications where the superpotential is simply given by (2.6). But once the nonperturbative term is added as in (2.1), $G_{(3)}$ can not remain ISD anymore. The unbroken supersymmetry $DW =0$ requires that $G_{(3)}$ must also contain (1,2) and (3,0) components (see for instance [@9]) in addition to the ISD components. Hence in this case (4.3) receives nonzero contributions from these fluxes.
The nonzero density of (4.3), however, satisfies $V_{\rm no-scale} \in V_3$, so it does not contribute to $\lambda$ in (3.41). But (4.3) is only referred to the no-scale type potential (2.5). In the AdS vacua of KKLT there is another important contribution to ${\mathcal V}_{\rm scalar}$ coming from the nonperturbative superpotential (2.1). Namely from (2.7) one obtains $${\mathcal V}_{\rm AdS} = - \frac{3}{2 \kappa_{10}^2} e^{\mathcal K} |W|^2 \,\,$$ under $D_\rho W=0$, which reduces to (2.8) by (2.1). (4.4) includes nonperturbative correction because $W$ in (4.4) contains the term $A e^{-a \sigma}$. The nonperturbative term can arise for instance from the gaugino condensation on $D7$-branes wrapping a four-cycle of the internal space [@5]. In the heterotic string theory the three-form structure of the potential density with a gaugino condensation $<{\rm tr}{\bar \lambda}\Gamma^{mnp}\lambda>$ is manifest in the action [@12], $$I_{\rm het}= - \frac{1}{2 \kappa_{10}^2} \int e^{-\phi} \big( H_{(3)} - \frac{\alpha^{\prime}}{16} e^{\phi/2} {\rm tr}{\bar \lambda}\Gamma_{(3)}\lambda \big)^2 \,\,.$$ So the potential density associated with (4.5) obviously belongs to $V_3$. In the case of type IIB theory, however, the tensor structure of (4.4) is not quite obvious at this point and we need some procedure to find it out.
0.3cm 0.15cm
Since $G_{(3)}$ generally contains both ISD and IASD components, we decompose $G_{(3)}$ as $$G_{(3)} = \alpha_0 \Omega + {\bar \alpha}_0 {\bar \Omega} + \beta^I \chi_I + {\bar \beta}^{\bar I} {\bar \chi}_{\bar I} \,\,,$$ where $\Omega$ is the holomorphic (3,0)-form and $\chi_I$ denotes the basis of $H^{(2,1)}$. Then using (4.6) one can express $W_0$ in (2.6) as $$W_0 = - {\bar \alpha}_0 \int \Omega \wedge {\bar \Omega} \,\,.$$ Apart from this, on the other hand, the total superpotential (2.1) satisfies $$D_I W=0$$ at the supersymmetric (AdS) minimum $\sigma = \sigma_m$ of KKLT, where the index $I$ labels the complex structure moduli. From (2.3) and the definition of the covariant derivative $D_I$ one finds that (4.8) requires $W$ to take the form $$W= -{\bar \alpha} \int \Omega \wedge {\bar \Omega} \,\,,$$ where ${\bar \alpha}$ can depend on $\tau$, but not on the complex structure moduli. The above result suggests that the $\int \Omega \wedge \bar \Omega$ structure of the superpotential $W_0$ remains unchanged under nonperturbative corrections. To understand this more explicitly, we consider the following discussions.
0.3cm 0.15cm
When the nonperturbative term $A e^{-a \sigma}$ is absent the complex structure moduli of the superpotential $\int_{{\mathcal M}_6} G_{(3)} \wedge \Omega(\tau^I )$ ($\equiv W(\tau^I )$) are stabilized at values $\tau^I = \tau_0^I$ for which $G_{(3)}$ is ISD at the tree level. Hence in this case the superpotential of the stabilized minimum is just $W(\tau_0^I )$, which is identified with $W_0$ in (2.6). Let us now introduce the nonperturbative term $A e^{-a \sigma}$. Once we introduce this term, the point $\tau^I = \tau_0^I$ in the moduli space would not be the stabilized point anymore. It deviates from $\tau^I = \tau_0^I$ along the complex structure moduli direction by the same amount of the nonperturbative corrections. So the new stabilized point becomes $\tau^I = \tau_0^I + \delta \tau^I$ and the superpotential of the supersymmetric minimum also changes from $W(\tau_0^I )$ to $W(\tau_0^I + \delta \tau^I )$, where $W(\tau_0^I + \delta \tau^I )$ now includes the nonperturbative correction and the deviation $\delta \tau^I$ will be determined by the nonperturbative term because the former is generated by the latter. To see this more explicitly, in the followings we will decompose $W(\tau_0^I + \delta \tau^I )$ into $W_0 + \delta W(\tau_0^I )$ and identify $\delta W (\tau_0^I )$ with the nonperturbative term $Ae^{-a \sigma_m}$.
0.3cm 0.15cm
Since $W(\tau^I )$ is given by $\int_{{\mathcal M}_6} G_{(3)} \wedge \Omega(\tau^I )$, we have $$W(\tau_0^I + \delta \tau^I ) = \int_{{\mathcal M}_6} G_{(3)}^{\rm NEW} \wedge \Omega (\tau_0^I ) + \int_{{\mathcal M}_6} G_{(3)}^{\rm NEW} \wedge \delta \Omega (\tau_0^I )\,\,,$$ where $$\delta \Omega (\tau_0^I )= \delta \tau^J \partial_J \Omega (\tau_0^I ) + O(({\delta \tau^J})^2 ) \,\,.$$ In (4.10), $G_{(3)}$ has been replaced by the new three-form $G_{(3)}^{\rm NEW}$ ($\equiv G_{(3)} + \delta G_{(3)}$) because the stabilized point has been changed into a new one by the nonperturbative correction, and the complex structure moduli of the new stabilized point cannot be fixed by the original $G_{(3)}$ in $W_0$. So in this section we temporarily use $G_{(3)}^{\rm NEW}$ to denote the three-form fluxes in $W(\tau^I )$, to distinguish it from the original $G_{(3)}$ in $W_0$. Now the first term of (4.10) can be decomposed into $W_0$ plus $\int_{{\mathcal M}_6} \delta G_{(3)} \wedge \Omega(\tau_0^I )$ because $\int_{{\mathcal M}_6} G_{(3)} \wedge \Omega(\tau_0^I )$ is just $W_0$ as mentioned in the subsection (3). Similarly, the second term of (4.10) can be also decomposed into $\int_{{\mathcal M}_6} G_{(3)} \wedge \delta \Omega (\tau_0^I )$ plus $\int_{{\mathcal M}_6} \delta G_{(3)} \wedge \delta \Omega (\tau_0^I )$, but where $\int_{{\mathcal M}_6} \delta G_{(3)} \wedge \delta \Omega (\tau_0^I )$ can be neglected in the leading order approximation because it is already of the second order in $\delta \tau^I$.
In (4.11) $\partial_J \Omega (\tau_0^I )$ represent the values of $\partial_J \Omega (\tau^I )$ at $\tau^I = \tau_0^I$, and if we use the well-known formula $\partial_J \Omega = (- \partial_J {\mathcal K}) \Omega + \chi_J$, one can rewrite (4.11) as $$\delta \Omega (\tau_0^I )= (-\delta {\mathcal K}) \Omega (\tau_0^I ) + \delta \tau^J \chi_J (\tau_0^I) + O(({\delta \tau^J})^2 ) \,\,,$$ where $\delta {\mathcal K}$ ($\equiv \delta \tau^J \partial_J {\mathcal K}$) is the variation of the K$\ddot{\rm a}$hler potential ${\mathcal K}$ caused by the nonperturbative correction $A e^{-a \sigma_m}$. Using (4.12) one can rewrite (4.10) as $$W(\tau_0^I + \delta \tau^I )= W_0 +\delta W(\tau_0^I ) \,\,,$$ where $\delta W(\tau_0^I )$ is now given by $$\delta W(\tau_0^I ) = \int_{{\mathcal M}_6} \delta G_{(3)} \wedge \Omega (\tau_0^I ) +(-\delta {\mathcal K}) \int_{{\mathcal M}_6} G_{(3)} \wedge \Omega (\tau_0^I ) + {\delta \tau^J} \int_{{\mathcal M}_6} G_{(3)} \wedge \chi_J (\tau_0^I )+ O(({\delta \tau^J})^2 ) \,\,.$$
0.3cm 0.15cm
Turning back to the nonperturbative term $A(\tau^I )e^{-a \sigma_m}$, let us consider the properties of $A(\tau^I)$. In ten-dimensional pictures the complex structure moduli of the Calabi-Yau threefolds are encoded in the harmonic three-form basis $\Omega$ and $\chi_I$. So in the ten-dimensional picture the scalar functions of the complex structure moduli, such as $A(\tau^I)$, in four-dimensional effective theory must appear essentially in terms of (or as linear combinations of) the nonzero six-dimensional integrals $\int \Omega \wedge \bar \Omega$ and $\int \chi_I \wedge {\bar \chi}_{\bar J}$ as in (4.14) because the four-dimensional effective theory is obtained by a dimensional reduction of the ten-dimensional theory.
Now we identify $\delta W(\tau_0^I )$ in (4.14) with $A(\tau^I )e^{-a \sigma_m}$ as mentioned in the subsection (3) : $$\int_{{\mathcal M}_6} \delta G_{(3)} \wedge \Omega (\tau_0^I )+ (-\delta {\mathcal K}) \int_{{\mathcal M}_6} G_{(3)} \wedge \Omega (\tau_0^I )+ {\delta \tau^J} \int_{{\mathcal M}_6} G_{(3)} \wedge \chi_J (\tau_0^I ) + O(({\delta \tau^J})^2 ) \nonumber$$ $$\equiv A(\tau_0^I + \delta \tau^I )e^{-a \sigma_m} \,\,.$$ The above identification may be achieved by adjusting ${\delta \tau^J}$ and $\delta G_{(3)}$ properly, which means that ${\delta \tau^J}$ and $\delta G_{(3)}$ are entirely determined by the nonperturbative term $A e^{-a \sigma_m}$. As an example, we take ${\delta \tau^J} = e^{-a \sigma_m} {\delta \tilde{\tau}^J}$, ${\delta G_{(3)}} = e^{-a \sigma_m} {\delta \tilde{G}_{(3)}}$, and adjust ${\delta \tilde{\tau}^J}$ and ${\delta \tilde{G}_{(3)}}$ properly so that (4.15) is satisfied. Then in the leading order approximation ${\delta \tilde{\tau}^J}$ and ${\delta \tilde{G}_{(3)}}$ are determined from the coefficients of the Hodge decompositions $\int \Omega \wedge \bar{\Omega}$ and $\int \chi_I \wedge \bar{\chi}_{\bar J}$ of $A(\tau_0^I )$. In this way, from (4.13) and (4.15) we finally have $$W(\tau_0^I + \delta \tau^I ) = W_0 + A(\tau^I )e^{-a \sigma_m} \,\,,$$ where $W(\tau_0^I + \delta \tau^I )$ is given by $\int_{{\mathcal M}_6} G_{(3)}^{\rm NEW} \wedge \Omega (\tau_0^I + \delta \tau^I )$ (see (4.10)).
The above result shows that the combined (total) superpotential $W_0 + A e^{-a \sigma_m}$ in (2.1) can be written in the form $$W (\tau^I ) = \int_{{\mathcal M}_6} G_{(3)}^{\rm NEW} \wedge \Omega (\tau^I ) \,\,,$$ where $\tau^I$ denotes $\tau_0^I + \delta \tau^I$. $W(\tau^I )$ in (4.17) has the $\int \Omega \wedge \bar \Omega$ structure just like $W_0$ in (2.6) despite that $A e^{-a \sigma_m}$ in $W(\tau^I )$ includes both $(3,0)\bigotimes(0,3)$ and $(2,1)\bigotimes(1,2)$ terms (see (4.15)). Indeed the last term ${\delta \tau^J} \int_{{\mathcal M}_6} G_{(3)} \wedge \chi_J (\tau_0^I )$ of (4.15) vanishes if $W_0$ satisfies $D_I W_0 =0$. But the F-term condition of the stabilized point has now been changed into $D_I W (\tau^I )=0$ from $D_I W_0 =0$. But still if we neglect the higher order terms, the last term of (4.15) can be neglected because $D_I W (\tau^I )=0$ approximately requires that $\delta \tau^J \int_{{\mathcal M}_6} G_{(3)} \wedge \chi_{J} (\tau_0^I )$ must vanish. So in this approximation the total superpotential $W (\tau^I )$ can be written as $$W (\tau^I ) = \int_{{\mathcal M}_6} G_{(3)}^{\rm EFF} \wedge \Omega (\tau_0^I )\,\,,$$ where $G_{(3)}^{\rm EFF}$ is defined by $G_{(3)}^{\rm EFF} \equiv (1- \delta {\mathcal K}) G_{(3)} + \delta G_{(3)}$.
The above $W (\tau^I )$ is of the same form as $W_0$ in (2.6) only except that $G_{(3)}$ in (2.6) is replaced by $G_{(3)}^{\rm EFF}$. According to (4.18), $W(\tau^I )$ may be regarded as a superpotential generated by an effective three-form flux $G_{(3)}^{\rm EFF}$, where the nonperturbative effects are merged with $G_{(3)}$ to form $G_{(3)}^{\rm EFF}$. So at least at the supersymmetric minimum the effect of the nonperturbative term $A e^{-a \sigma_m}$ in $W(\tau^I )$ is to change $G_{(3)}$ into a new flux $G_{(3)}^{\rm EFF}$ which also has the three-form structure like the original $G_{(3)}$.
0.3cm 0.15cm
We have just seen that the total superpotential $W(\tau^I )$ has the $(3,0)\bigotimes(0,3)$ structure in the leading order approximation where the terms of order higher than $(\delta \tau^{I})^2$ are neglected. In our case, however, we don’t have to use this approximation to find the Hodge structure of $W (\tau^I )$. We already know from (4.17) that the superpotential $W(\tau^I )$ has the $\propto \int \Omega \wedge {\bar \Omega}$ structure just like $W_0$. The only difference between $W_0$ and $W(\tau^I )$ is that $\Omega (\tau_0^I )$ and $G_{(3)}$ in $W_0$ are now replaced by $\Omega (\tau^I )$ and $G_{(3)}^{\rm NEW}$ in (4.17), and hence also in (4.7) and (4.9), $\Omega$ in $W$ is $\Omega (\tau^I )$, while $\Omega$ in $W_0$ is $\Omega (\tau_0^I )$. (The fact that $\Omega$ in (4.9) is $\Omega (\tau^I )$ means that the holomorphic three-forms contained in (2.3) also change from $\Omega (\tau_0^I )$ to $\Omega (\tau^I )$ under the nonperturbative correction because (4.9) is obtained from (2.3) by $D_I W=0$.) Despite these differences, however, the Hodge structures of $W$ and $W_0$ are entirely identical. They are both $\propto \int \Omega \wedge {\bar \Omega}$ (or $\int_{{\mathcal M}_6} G_{(3)} \wedge \Omega$), though the complex structures of each $\Omega$’s in $W$ and $W_0$ are different from one another.
Now using (4.17) one can determine the tensor structure of ${\mathcal V}_{\rm AdS}$ in (4.4). (4.17) shows that $W$ acquires nonzero values from the $(0,3)$ component of $G_{(3)}^{\rm NEW}$ ($\equiv G_{(0,3)}^{\rm NEW}$). Writing $G_{(0,3)}^{\rm NEW}$ as $G_{(0,3)}^{\rm NEW} = {\bar \alpha}{\bar \Omega}$, one obtains $$\int {\bar G}_{(0,3)}^{\rm NEW} \wedge G_{(0,3)}^{\rm NEW} = - \frac{\int G_{(0,3)}^{\rm NEW} \wedge \Omega \int {\bar G}_{(0,3)}^{\rm NEW} \wedge {\bar \Omega}}{\int \Omega \wedge {\bar \Omega}} \,\,,$$ and using $\ast_6 G_{(0,3)}^{\rm NEW}= i G_{(0,3)}^{\rm NEW}$ (note that $G_{(0,3)}^{\rm NEW}$ is ISD) one finds that ${\mathcal V}_{\rm AdS}$ in (4.4) becomes $${\mathcal V}_{\rm AdS} = \frac{1}{4 \kappa_{10}^2} \, e^{\mathcal K_{\tau} + {\mathcal K}_{\rho} } \int d^6 y \sqrt{h_6} \big( G_{(0,3)}^{\rm NEW} \big)_{mnp} \big({\bar G}_{(0,3)}^{\rm NEW}\big)^{mnp}\,\,.$$ (4.20) shows that the density of ${\mathcal V}_{\rm AdS}$ clearly belongs to $V_3$.
In (4.20) ${\mathcal V}_{\rm AdS}$ receives nonzero contribution from the ISD $(0,3)$ component of $G_{(3)}^{\rm NEW}$ as opposed to the case of (4.3) where ${\mathcal V}_{\rm no-scale}$ receives nonzero contributions only from the IASD components of $G_{(3)}$. Indeed $G_{(3)}^{\rm NEW}$ in (4.17) contains only $(3,0)$ component as an IASD piece because (4.8) requires that $(1,2)$ component of $G_{(3)}^{\rm NEW}$ should vanish. In any case, both ${\mathcal V}_{\rm no-scale}$ and ${\mathcal V}_{\rm AdS}$ in (4.3) and (4.20) take nonzero values in the AdS vacua of KKLT even at the tree level. However, they never contribute to $\lambda$ in (3.41) because their densities $V_{\rm no-scale}$ and $V_{\rm AdS}$ both belong to $V_3$. Hence $\lambda$ must be self-tuned to vanish in the AdS vacua of KKLT.
0.3cm 0.15cm
The scalar potential arising from (2.1) does not vanish at the supersymmetric minimum of the potential. In general it is proportional to $|W_0 |^2$, or more precisely, $${\mathcal V}_{\rm scalar} \propto e^{\mathcal K} |W_0 |^2 \,\,,$$ at the extremum of the potential [@13; @14]. Indeed at the AdS minimum $D_{\rho} W=0$ of KKLT, the coefficient $A$ is given by $$A = - W_0 \, e^{a \sigma_{m}} \big(1+\frac{2}{3}a \sigma_{m} \big)^{-1} \,\,,$$ and therefore the scalar potential (2.8) becomes proportional to $e^{\mathcal K} |W_0 |^2$ there, $${\mathcal V}_{\rm AdS} = \frac{1}{2 \kappa_{10}^2} \Big( -\frac{a^2}{6 \sigma_m} \big( 1+ \frac{2}{3} a \sigma_{m}\big)^{-2} \,\Big)\, e^{\mathcal K_{\tau} + {\mathcal K}_{\rm cs} } |W_0 |^2 \,\,,$$ where ${\mathcal K}_{\rm cs}$, $\mathcal K_{\tau}$ are the K${\ddot{\rm a}}$hler potentials for the complex structure moduli and the axion/dilaton, respectively.
(4.22) shows that $W_0$ necessarily takes nonzero values in the presence of the nonperturbative correction $A e^{-a \sigma_{m}}$. The nonzeroness of $W_0$ implies that $G_{(3)}$ must contain $(0,3)$ component, and in the presence of this component the gravitino generally acquires nonzero mass $m_{3/2}$ from the $G_{(3)}$ flux. The gravitino mass term of the reduced action for the type IIB theory can be obtained through the decomposition $$\Psi_\mu = \psi_\mu \otimes e^{\frac{B}{4}} \eta \,\,,$$ where $\Psi_\mu / \psi_\mu$ are the ten$/$four-dimensional gravitini, respectively, and $\eta$ is a six-dimensional killing spinor satisfying $\gamma^{\bar{i}} \eta =0$, where $\gamma^{\bar{i}}$ is the six-dimensional Dirac matrix represented in the complex basis. In the real basis of the Calabi-Yau one obtains [@14] $$I_{3/2} = \frac{1}{\kappa_{10}^2} \int d^4 x \sqrt{-g_4} \frac{1}{(Im \rho)^{3/2}}\, \Big\{ \big(\bar{\psi}_\mu \gamma^{\mu \nu} \psi_{\nu}^{\ast} \big) \big(\frac{i}{48} \int d^6 y \sqrt{h_6} \frac{1}{(Im \tau)^{1/2}} \eta^{+} \gamma^{mnp} \eta^{\ast} G_{mnp} \big) \nonumber$$ $$+\, {\rm hermitian \,~conjugate\,~ term} \Big\}\,\,,$$ where $m_{3/2}$ is identified as $$m_{3/2} = \frac{\kappa^2}{\kappa_{10}^2} \frac{1}{(Im \rho)^{3/2}} \frac{1}{(Im \tau)^{1/2}} \Big(\frac{1}{24} \int d^6 y \sqrt{h_6} \,\eta^{+} \gamma^{mnp} \eta^{\ast} G_{mnp} \Big)\,\,.$$ Since all components except $\eta^{+} \gamma^{\bar{i}\bar{j}\bar{k}} \eta^{\ast}$ ($= \Omega^{\bar{i}\bar{j}\bar{k}}/ \|\Omega \|$) of $\eta^{+} \gamma^{mnp} \eta^{\ast}$ vanish by $\gamma^{\bar{i}} \eta =0$ in the complex basis, only the $(0,3)$ piece of $G_{(3)}$ contributes to $m_{3/2}$.
(4.25) shows that the density $V_{3/2}$ of $I_{3/2}$ is proportional to $\eta^{+} \gamma^{mnp} \eta^{\ast} G_{mnp}$. So $V_{3/2} \in V_3$, and the gravitino mass term $I_{3/2}$ arising from $G_{(0,3)}$ does not contribute to $\lambda$ just like ${\mathcal V}_{\rm AdS}$ of KKLT. Indeed, the gravitino mass $m_{3/2}$ is closely related to the ${\mathcal V}_{\rm AdS}$ of KKLT. Since $m_{3/2}^2$ is identified with $<e^{\mathcal K} |W_{0}|^2 >$ (see [@11] or [@14]), and $<e^{\mathcal K} |W_{0}|^2 >$ is a constant times ${\mathcal V}_{\rm AdS}$ at the AdS minimum (see (4.21) or (4.23)), $m_{3/2}^2$ is proportional to ${\mathcal V}_{\rm AdS}$ of KKLT. So one of the ways of ascertaining whether ${V}_{\rm AdS} \in V_3$ is really true is to check whether $m_{3/2}$ in (4.26) satisfies ${\hat m}_{3/2}^2 \in V_3$ or not, where ${\hat m}_{3/2}^2$ is the density of $m_{3/2}^2$ defined by $m_{3/2}^2 \equiv (1/2 \kappa_{10}^{2} g_s^2 \,) \int d^6 y \sqrt{h_6} \,{\hat m}_{3/2}^2$. The fact that ${\hat m}_{3/2}^2$ satisfies ${\hat m}_{3/2}^2 \in V_3$ can be proved easily as follows. Using $\ast_6 G_{(0,3)} = i G_{(0,3)}$ one can show that the square of (4.26), $m_{3/2}^2$, is proportional to $\big| \int G_{(0,3)} \wedge {\hat \Omega} \big|^2$, where ${\hat \Omega} \equiv \Omega / \|\Omega\|$. Next, using (4.19) one can show that $\big| \int G_{(0,3)} \wedge {\hat \Omega} \big|^2$ is proportional to $\int {\bar G}_{(0,3)} \wedge G_{(0,3)}$. Finally, using $\ast_6 G_{(0,3)} = i G_{(0,3)}$ again one can show that $\int {\bar G}_{(0,3)} \wedge G_{(0,3)}$ is proportional to $\int d^6 y \sqrt{h_6} (G_{(0,3)})_{mnp} ({\bar G}_{(0,3)})^{mnp}$. After all, one finds that ${\hat m}_{3/2}^2 \in V_3$ because $m_{3/2}^2$ is proportional to $\int d^6 y \sqrt{h_6} (G_{(0,3)})_{mnp} ({\bar G}_{(0,3)})^{mnp}$, which confirms the result of Sec. 4.1 that $V_{\rm AdS} \in V_3$.
The result $V_{\rm AdS} \in V_3$ is not affected by the perturbative and nonperturbative corrections ${\mathcal K} = {\mathcal K}_{\rm tree} + {\mathcal K}_{\rm p} +{\mathcal K}_{\rm np}$ and $W=W_{\rm tree} + W_{\rm np}$. The corrections ${\mathcal K}_{\rm p}$, ${\mathcal K}_{\rm np}$ in ${\mathcal K}$ act only as multiplicative factors $e^{{\mathcal K}_{\rm p}}$, $e^{{\mathcal K}_{\rm np}}$ in (4.20) (or in (4.21)), and on the other hand $W_{\rm np}$ in $W$ has already been considered in our discussions (namely in (2.1) and (6.3)). So the structure (4.20) of ${\mathcal V}_{\rm AdS}$, and consequently the result $V_{\rm AdS} \in V_3$ does not change by ${\mathcal K} = {\mathcal K}_{\rm tree} + {\mathcal K}_{\rm p}+ {\mathcal K}_{\rm np}$ and $W=W_{\rm tree} + W_{\rm np}$.
0.5cm
The next step of KKLT is to introduce $\overline{D3}$-branes (anti-$D3$-branes) at the end of the KS throat to obtain dS vacua. Introduction of $\overline{D3}$-branes induces an additional term $\delta {\mathcal V}_{\rm scalar}$ (see (2.9)) to the scalar potential as anticipated from the analysis of [@15]. Thus the scalar potential after introducing $\overline{D3}$-branes must be the sum of (2.8) and (2.9), where ${\mathcal V}_{\rm scalar}$ in (2.8), which is ${\mathcal V}_{\rm AdS}$ in fact, has already been verified to respects (3.42) with $n=3$ at the supersymmetric minimum. So the next procedure will be to check what happens to the structure of the potential density after adding (2.9) to the nonperturbative potential (2.8). Does the sum of these two potentials still respects (3.42) with $n=3$ at the dS minima? As an answer to this question, we will first show in Sec. 5.1 that $\delta {\mathcal V}_{\rm scalar}$, and consequently the sum of (2.8) and (2.9) does not respect (3.42). This means that the density of $\delta {\mathcal V}_{\rm scalar}$( $\equiv V_{\overline{D3}}$) caused by $\overline{D3}$-branes makes a nonzero contribution to $\lambda$ in (3.41), and consequently $\lambda$ of the dS vacua described by (2.8) plus (2.9) may not be fine-tuned to vanish unlike in the scenario of the original KKLT. Hence in the second part of this section (Sec. 5.2) we will propose an alternative scenario for the vanishing $\lambda$ of our present universe. This alternative scenario uses AdS, instead of dS, vacua of KKLT, and it has more nice properties as compared with those dS vacua uplifted by anti-$D3$-branes.
0.3cm 0.15cm
In [@15] the dynamics of $\overline{D3}$-branes is described by a Dirac-Born-Infeld (DBI) plus Chern-Simons (CS) world volume action for the $NS5$-brane due to technical difficulties in obtaining DBI action for the pure $\overline{D3}$-branes in the KS background geometry. In this S-dual description the $\overline{D3}$-branes are described by $NS5$-branes wrapping $S_2$ inside the $A$-cycle of the conifold geometry. At the apex of the conifold the metric becomes $$ds^2 = a_0^2 dx_{\mu} dx^{\mu} + R_0^2 \big( d \psi^2 + \sin^2 \psi d \Omega_2^2 \big)\,\,,$$ where $a_0$ and $R_0$ are constants, and the world volume action for the $NS5$-brane of type IIB theory takes the form (see [@15] or [@16]) $$I_{NS5} = \frac{\mu_5}{g_s^2} \int d^6 \xi \big[ -det (g_{\mu\nu}) \cdot det(h_{{\hat m}{\hat n}} + 2 \pi g_s {\mathcal F}_{(2)} ) \big]^{1/2} + \mu_5 \int B_{(6)} \,\,,$$ where $h_{{\hat m}{\hat n}}$ is a two-dimensional metric induced along $S_2$ of the $A$-cycle and $2 \pi {\mathcal F}_{(2)} = 2\pi F_{(2)} -A_{(2)}$ with $F_{(2)} =dA$ a two-form field strength of the world volume gauge field of the $NS5$-brane. In (5.2) $F_{(2)}$ is assumed to satisfy $$2 \pi \int_{S_2} F_{(2)} = 4 \pi^2 p \,\,,$$ so that the $NS5$-brane carries $\overline{D3}$ charge $p$. R-R two-form $A_{(2)}$ is also assumed to satisfy $$\int_{S_2} A_{(2)} = 4 \pi M \Big(\psi - \frac{1}{2} \sin (2\psi) \Big) \,\,,$$ which follows from the well-known R-R flux quantization $\int_{A} F_{(3)} = 4 \pi^2 M$.
The DBI part of (5.2) contains an internal metric $h_{{\hat m}{\hat n}}$ because (5.2) is an world volume action for the $NS5$-brane rather than genuine $\overline{D3}$-brane. But using (5.3) and (5.4), one finds that (5.2) turns into $$I_{\overline{D3}} = \int d^4 x \sqrt{- det (g_{\mu\nu})} \,\, \mathcal{L}_{\overline{D3}} (\psi)\,\,,$$ which is typical of the world volume action for the $D3/\overline{D3}$-branes. In (5.5), $\mathcal{L}_{\overline{D3}} (\psi)$ can be written, upon taking ${\dot \psi}=0$, in the form $$\mathcal{L}_{\overline{D3}} (\psi) = \frac{4 \pi^2 \mu_5 M}{g_s} {\hat V}(\psi)\,\,,$$ where $M$ is related with $h_{{\hat m}{\hat n}}$ by the integral $$\int_{S_2} d^2 y \sqrt{det (h_{{\hat m}{\hat n}} + 2 \pi g_s {\mathcal F}_{(2)})\,} \, = 4 \pi^2 M g_s {\hat V}(\psi)\,\,,$$ and ${\hat V}(\psi) \simeq p/M$ for $\psi \ll 1$ (see \[16\]). The scalar potential for the $\overline{D3}$-branes can be read from $\mathcal{L}_{\overline{D3}} (\psi)$ in (5.6) and it turns out to take the form (2.9).[^6]
Turning back to (5.5), $\mathcal{L}_{\overline{D3}} (\psi)$ is given as a function of $\psi$ and where $\overline{D3}$-branes correspond to $\psi =0$. But in the KS geometry $\psi =0$ is not a stable, nor a metastable point of the potential, and hence in the S-dual description the $\overline{D3}$-branes are necessarily described by the $NS5$-branes which occupy $S_2$ of the $A$-cycle in the internal space. So $\mathcal{L}_{\overline{D3}} (\psi)$ in (5.5) necessarily contains the two-dimensional internal metric $h_{{\hat m}{\hat n}}$ implicitly in the form $\int d^2 y \sqrt{det (h_{{\hat m}{\hat n}} + 2 \pi g_s {\mathcal F}_{(2)})\,}$ (see (5.6) and (5.7)), and we infer that the potential density $V_{\overline{D3}}$ will be of the form $$V_{\overline{D3}} \sim \frac{\mu_5}{g_s} \frac{\sqrt{det (h_{{\hat m}{\hat n}} + 2 \pi g_s {\mathcal F}_{(2)})\,}}{\sqrt{det (h_{{\hat m}{\hat n}})}} \,\delta^4 (y) \,\,,$$ where $\delta^4 (y)$ is defined by $\int d^4 y \sqrt{h_4} \, \delta^4 (y) =1$ with $\sqrt{h_4} \equiv \sqrt{h_6}/ {\sqrt{det (h_{{\hat m}{\hat n}}) }}$. After all, we find that $V_{\overline{D3}} \notin V_n$ because $V_{\overline{D3}}$ in (5.8) does not satisfy (3.42).
0.3cm 0.15cm
0.3cm 0.15cm
In KKLT, the scalar potential is so adjusted that the constituents (2.8) and (2.9) cancel out at the dS minima. So the scalar potentials of the dS vacua almost vanish at their dS minima $\sigma =\sigma_m$, and we can write $${\mathcal V}_{\rm dS} = \epsilon |{\mathcal V}_{\rm AdS}|\,\,,$$ where $\epsilon$ is an arbitrarily small positive constant of order $\sim 10^{-120}/ O({\mathcal V}_{\rm AdS})$. If ${\mathcal V}_{\rm dS}$ in (5.9) can take sufficiently small values, the corresponding $\lambda$ will also be very small, and we may take one of the dS vacua of KKLT as the background vacuum of our present universe. However, this is true only in the traditional theories. According to our discussions in Sec. 5.1 it is very unlikely that such a fine-tuning is really possible.
The superpotentials of dS vacua do not satisfy $DW =0$ at the dS minimum because the introduction of anti-${D3}$-branes breaks the supersymmetry slightly. Thus the superpotentials for the dS vacua do not have the structure (4.9) or (4.17), and consequently ${\mathcal V}_{\rm dS}$ of dS vacua may not be able to be written in the form (4.20), which suggests that the corresponding $V_{\rm dS}$ necessarily makes a nonzero contribution to $\lambda$ in (3.41). Indeed in Sec. 5.1, we have shown that the density of $\delta {\mathcal V}_{\rm scalar}$ in (2.9) ($= V_{\overline{D3}}$) does not satisfy $V_{\overline{D3}} \in V_3$ and hence $\delta {\mathcal V}_{\rm scalar}$ arising from anti-$D3$-branes necessarily makes a nonzero contribution to $\lambda$ in the equation (3.41).
In the case of AdS vacua, however, it was shown that $V_{\rm AdS}$ belongs to $V_3$ (See Sec. IV.) and therefore it does not contribute to $\lambda$ as opposed to the case of $V_{\overline{D3}}$. So these things make us to doubt that $\lambda$ of dS vacua can be really fine-tuned to vanish by adding ${\mathcal V}_{\rm AdS}$ in (2.8) and $\delta {\mathcal V}_{\rm scalar}$ in (2.9). Indeed, even when we accept the possibility of this fine-tuning, it is preserved only at the tree level. The problem is that once the perturbations enter, there is no way to make $\lambda$ remain to be of order $\sim 10^{-120}$ in the units of Planck density. For instance if we take the quantum fluctuations on the $D3$-branes into account, the fine-tuning $\lambda =0$ will be severely disturbed. For these reasons it seems that we may need to introduce an alternative scenario which can substitute for the dS vacua of KKLT type models. In this section we propose a new vacuum scenario for the background state of our present universe, as a substitute for the original dS vacuum scenario of KKLT.
0.3cm 0.15cm
As mentioned above, in Sec. IV we have shown that the potential density of the AdS vacua belongs to $n=3$ ($V_{\rm AdS} \in V_3$) and consequently it does not make any nonzero contributions to $\lambda$. So the simplest, and perhaps the most natural scenario using KKLT is to take one of these AdS vacua of KKLT to identify it as the background vacuum of our present universe. (Recall that in our self-tuning mechanism ${\mathcal V}_{\rm scalar} <0$ does not necessarily imply $\lambda <0$ due to ${\mathcal E}_{\rm SB}$ in (1.1).) This AdS vacuum configuration with certain numbers of $D3$-branes may be identified with the supersymmetric (stable) minimum at $\psi =\pi$ of the brane/flux annihilation description in [@15]. Namely the nonsupersymmetric configuration with $p$ anti-${D3}$-branes (the dS vacua) rolls down (via tunneling and a classical process at some early stage during or after inflation) the potential to the north pole $\psi = \pi$ to form a supersymmetric configuration with $M-p$ $D3$-branes which is now identified with the present stage of our universe. In this scenario the supersymmetry breaking of the brane region is basically generated by ${\mathcal E}_{\rm SB}$, not by anti-${D3}$-branes (see Sec.VIII).
The above AdS vacuum scenario can substitute for the dS vacua of KKLT in the framework where $\lambda$ is given by (1.1), and AdS vacua of this scenario have more nice properties as compared with the dS vacua, as listed below.
0.6cm
- In general dS vacua of the usual flux compactifications have a tunneling instability since these dS vacua are only local minima of the potential and they eventually decay into run away vacuum at $\sigma = \infty$. Hence in the theories using these dS vacua the authors need to show that their lifetimes are huge enough to describe our present universe as in KKLT. In the AdS vacuum scenario, however, the background (AdS) vacua describing our present universe are stable both classically and quantum mechanically and such a tunneling instability is inherently absent.
- The dS vacua uplifted by anti-${D3}$-branes also suffer from another kind of tunneling instability. As mentioned above, nonsupersymmetric configurations with anti-${D3}$-branes (the dS vacua) correspond to the metastable states in the brane/flux annihilation descriptions in [@15], and these metastable states decay, via tunneling and classical process, into supersymmetric configurations with ${D3}$-branes (the AdS vacua) which correspond to the stable minima of the brane/flux annihilation description. Since these AdS vacua correspond to the stable minima, there is no other minimum (or minima) to decay into.
- There is no any parameter, nor coefficient to be fine-tuned in the AdS vacuum scenario. $\lambda =0$ is automatically achieved by the cancelation between ${\mathcal V}_{\rm scalar} + \delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ and ${\mathcal E}_{\rm SB}$, forced by (3.41).
- Most of all, in the AdS vacuum scenario of our self-tuning mechanism the fine-tuning $\lambda=0$ is radiatively stable. Any nonzero contributions to ${\mathcal V}_{\rm scalar}$ coming from $g_s$ - perturbations and quantum fluctuations (vacuum energies) on the visible sector $D3$-branes are all gauged away by ${\mathcal E}_{\rm SB}$ and as a result $\lambda =0$ is always preserved.
The dS vacua with anti-${D3}$-branes might be suitable for the description of the early universe including inflation, rather than the present universe with vanishing $\lambda$. The anti-${D3}$-branes are indispensable in the brane-antibrane inflation scenario [@17; @18] because the potential for the inflation (inflaton potential) is generated by the brane-antibrane interaction. Also in the inflationary era the coefficient $D$ in (2.9) (and therefore $\epsilon$ in (5.9)) does not have to be fine-tuned. Entire ${\mathcal V}_{\rm dS}$ of the dS vacua can contribute, together with the potential generated by the brane-antibrane interaction, to $\lambda$ to make it positive. But these nonsupersymmetric dS vacua with anti-$D3$-branes are only metastable, hence they eventually decay into the supersymmetric AdS vacua describing our present universe in the AdS vacuum scenario.
0.5cm
In the AdS vacuum scenario the supersymmetric configuration at $\psi = \pi$ contains $D3$-branes in the KS throat. Introduction of $D3/\overline{D3}$-branes generally induces a scalar potential coming from the DBI plus CS action. For instance in KKLT, an introduction of anti-$D3$-branes induces an additional term (2.9) to the scalar potential as we have already seen. Also the potential for the $D3$-branes, which vanishes in the ISD compactifications, acquires nonzero contributions once the background turns into IASD because in this background the IASD fluxes become a source for the scalar potential of the $D3$-branes. Besides this, the presence of $D3/\overline{D3}$-branes also yields open string moduli such as locations of the branes in the compact space. Thus we may need to check if all these contributions to the scalar potential also respect (3.42) with $n=1,3$ to make $\lambda$ vanish. In this section we want to check the contributions coming from the open string moduli of the $D3$-branes, and then in the next section we will consider the $D3$-brane potential sourced by IASD fluxes. In our discussions of this section we will consider the general case where the nonperturbative vacua are basically given by the AdS type vacua, rather than dS, of KKLT according to the discussions of the previous section. So we do not have anti-${D3}$-branes in our configurations.
Suppose that we have a single (or a stack of) $D3$-brane(s) in the six-dimensional compact space for simplicity. In the presence of a $D3$-brane the K$\ddot{\rm a}$hler modulus[^7] $\rho$ acquires an additional term $k (Y, {\bar Y})$ [@19]: $$\rho = \frac{b}{\sqrt 2} + i e^{4u} + \frac{i}{2} k (Y, {\bar Y}) \,\,,$$ where the three complex scalars $Y^{\alpha}$, $\alpha =1,2,3$, in $k (Y, {\bar Y})$ represent the location of the $D3$-brane.[^8] The K$\ddot{\rm a}$hler potential for this K$\ddot{\rm a}$hler modulus is therefore $${\mathcal K}_{\rho} = -3 \ln e^{4u} = -3 \ln \big[ -i ( \rho - {\bar \rho}) - k (Y, {\bar Y}) \big] \,\,.$$ Besides this, the nonperturbative superpotential (2.1) also changes in the presence of $D3$-brane into the form [@19] $$W=W_0 +A e^{i a \rho -\zeta (Y)} \,\,.$$ So the supersymmetric vacua must satisfy $$D_{\rho} W = i a A e^{i a \rho -\zeta (Y)} + \frac{3iW}{\big[ -i ( \rho - {\bar \rho}) - k (Y, {\bar Y}) \big]} =0 \,\,,$$ $$D_{\alpha} W = - A \partial_\alpha \zeta (Y) e^{i a \rho -\zeta (Y)} + \frac{3 (\partial_\alpha k)W}{\big[ -i ( \rho - {\bar \rho}) - k (Y, {\bar Y}) \big]} =0 \,\,,$$ and from these two equations one obtains $$\partial_\alpha \zeta (Y) + a \partial_\alpha k(Y, {\bar Y}) =0 \,\,.$$ (6.6) guarantees that (6.4) and (6.5) are not inconsistent with each other as far as it admits a solution.
Now we can show that the potential density $V_{\rm AdS}$ associated with the superpotential (6.3) still belongs to $V_3$. $W$ in (6.3) differs from $W$ in (2.1) only in that $e^{ia\rho}$ is replaced by $e^{ia\rho \, -\zeta(Y)}$, and in (6.3) the complex structure moduli are only contained in $W_0$ and $A$ as before. Also since the K$\ddot{\rm a}$hler potential for the complex structure moduli is still given by $\propto \ln [\,-i \int_{{\mathcal M}_6} \Omega \wedge {\bar \Omega}\,]$ (where $\Omega$ represents $\Omega(\tau^I )$), the F-term condition (4.8) requires $W$ to take the form (4.9) again except that $\bar \alpha$ may now depend on both $\tau$ and $Y^{\alpha}$, instead of $\tau$ alone. (But see the footnote 6.) Indeed, repeating the same procedure from eq. (4.10) to (4.16) one obtains (4.17) again. The only difference is that $\delta \tau^J$’s in $\tau_0^J + \delta \tau^J$ now also depend on $Y^{\alpha}$ in addition to $\tau_0^I$ and $\sigma_m$. So we finally obtain (4.20) again for ${\mathcal V}_{\rm AdS}$, implying that $V_{\rm AdS} \in V_3$ and therefore $V_{\rm AdS}$ does not contribute to $\lambda$ even in the presence of the open string moduli.
One can reaffirm the above result as follows. Substituting (6.3) into (6.4) gives $$-3 \frac{W_0}{A} = \big[ 3- i a ( \rho - {\bar \rho}) - a k (Y, {\bar Y}) \big] e^{i a \rho -\zeta (Y)} \,\,.$$ But since $$- i ( \rho - {\bar \rho}) - k (Y, {\bar Y}) = 2 e^{4u} \,\,$$ from (6.1), one obtains $$A = -W_0 e^{-i a \rho +\zeta (Y)} (1+\frac{2}{3} a e^{4u} )^{-1} \big|_m \,\,.$$ (6.9) coincides with (4.22) except $e^{a \sigma_m}$ is replaced by $e^{-i a \rho +\zeta (Y)}|_{m}$. So ${\mathcal V}_{\rm AdS}$ obtained from (6.3), which will be identical with (2.8) only except that $e^{-a \sigma}$ is replaced by $e^{-a \sigma + \xi(Y)}$, will take the same form as (4.23) by (6.9) at the supersymmetric minimum, and by repeating the same discussions as in Sec. 4.2 one finds that $V_{\rm AdS}$ associated with (6.3) also belongs to $V_3$ as before.
0.5cm
In the ISD compactifications $-$ and in the absence of branes $-$ $\lambda$ trivially vanishes from (3.41) because potential density arising from the fluxes vanishes in the ISD background. But once the perturbations come into the theory, ${\mathcal V}_{\rm scalar}$ does not vanish anymore because in this case $G_{(3)}$ aquires IASD components. Besides this, the IASD fluxes also induce a potential for the $D3$-branes because they become a dominant source in the equation of motion for the D3-brane potential. In [@20], it was shown that there exist three distinct types of closed, IASD three-form fluxes which induce the $D3$-brane potential.
Among these fluxes the simplest one is the type I flux which contains only $G_{(1,2)}$, the IASD $G_{(3)}$ of Hodge type $(1,2)$. Compared with other two types of fluxes, the type I flux is of particular importance because the other two contain non-primitive $(2,1)$ which is forbidden in a compact Calabi-Yau space. Besides this, it was also shown in [@20] that there is a holographic correspondence between perturbations of supergravity solution by the type I flux and superpotential perturbations of the conformal field theory. In this correspondence the scalar potential for a probe D3-brane in the conifold geometry precisely matches the scalar potential computed in the gauge theory with superpotential $W$, and the scalar potential for a $D3$-brane in the conifold geometry is reproduced by the $G_{(1,2)}$ flux.
0.3cm 0.15cm
The $D3/\overline{D3}$-brane potentials follow from the DBI plus CS action (3.3) with $T(\phi) \backslash \mu(\phi)$ replaced by $T_3 e^{-\phi} \backslash \mu_3$. In string frame it is given by $$I_{D3/\overline{D3}} = - T_3 \int d^4 x \, e^{-\phi} \sqrt{-det (g_{\mu\nu})} + \mu_3 \int A_{(4)} \,\,,$$ where $T_3 = |\mu_3 | = (2\pi)^{-3} (\alpha^{\prime})^{-2}$ and $$A_{(4)}= \xi(y) \sqrt{-g_4} \, dx^0 \wedge dx^1 \wedge dx^2 \wedge dx^3 \,\,.$$ For the given compactification (3.4), $I_{D3/\overline{D3}}$ becomes $$I_{D3/\overline{D3}} = - T_3 \int d^4 x \sqrt{-g_4} \, \frac{1}{g_s} \Phi_{\mp}\,\,,$$ where $\Phi_{\pm}$ are defined by $$\frac{1}{g_s} \Phi_{\pm} = \frac{\chi^{1/2}}{g_s} \pm \xi \,\,.$$ Here we ignored the kinetic terms of the $D3/\overline{D3}$-brane actions because we assumed that the $D3/\overline{D3}$-branes are all fixed at some certain points of the compact space. According to (7.4), $\Phi_{-}$ vanishes in the ISD background if the (bulk) supersymmetry is unbroken (see Sec. V of [@1]), and therefore $D3$-branes feel no potential in this case. But once the higher order perturbations come into the theory, the situation changes. Because higher order terms of $G_{(3)}$ generally contain IASD components, and these components become a dominant source in the equation of motion for $\Phi_{-}(y)$ [@19; @20], the $D3$-branes certainly feel a potential arising from the higher order terms of $\Phi_{-}(y)$. In this section we will show that this $D3$-brane potential arising from the IASD flux perturbations also respects the condition (3.42), but this time not with $n=3$, but with $n=1$. Namely $V_{D3} \in V_1$ for the type I flux.
0.3cm 0.15cm
The equation of motion for $\Phi_{-}$ may be obtained from the field equations for $\chi^{1/2}$ and $\xi$, among which the latter follows from the field equation for $A_{(4)}$. The field equation for $A_{(4)}$ can be obtained from the three terms $$\frac{1}{8 \kappa_{10}^2} \int {\tilde F}_{(5)} \wedge \ast {\tilde F}_{(5)} + \frac{1}{8i \kappa_{10}^2} \int e^{\phi} A_{(4)} \wedge G_{(3)} \wedge {\bar G}_{(3)} + \frac{\mu_3}{2} \int A_{(4)}$$ in the actions (3.1) and (3.3), where we have rewritten the ${\tilde F}_{(5)}^2$ term in (3.1) as $\frac{1}{8 \kappa_{10}^2} \int {\tilde F}_{(5)} \wedge \ast {\tilde F}_{(5)}$ for convenience, and replaced $\mu_3 \rightarrow \frac{\mu_3}{2}$ which is necessary to obtain correct equation for the self-dual field $A_{(4)}$ (see for instance [@14] or [@20-1] for this). We obtain from (7.5) $$d \ast {\tilde F}_{(5)} = \frac{G_3 \wedge {\bar G}_3}{2 i Im \tau} + 2 \kappa_{10}^2 \,\mu_3 \,\rho_3^{\rm loc} \,\,,$$ which, by (3.2), reduces to $$\nabla^2 \xi = \frac{i}{12 Im \tau}\, \chi \, G_{mnp} \ast_6 {\bar G}^{mnp} + 2 \chi^{-1/2} (\partial \chi^{1/2}) (\partial \xi) + 2 \kappa_{10}^2 \mu_3 \, \chi \, \rho_3^{\rm loc} \,\,.$$
The field equation for $\chi^{1/2}$, on the other hand, can be obtained from (3.6) plus the topological term $$\frac{1}{8i \kappa_{10}^2} \int e^{\phi} A_{(4)} \wedge G_{(3)} \wedge {\bar G}_{(3)}=\frac{1}{2 \kappa_{10}^2 g_s^2} \big[\int d^4 x \sqrt{-g_4}\,\big] \Big(\frac{i g_s^2}{24} \int d^6 y \sqrt{h_6} \,e^{\phi} \xi \,G_{mnp} \ast_6 {\bar G}^{mnp}\Big)\,\,.$$ Varying the action with respect to $B$ we obtain $$\nabla^2 \Big( \frac{\chi^{1/2}}{g_s} \Big) = \frac{i}{12 Im \tau}\, \chi \, G_{mnp} \ast_6 {\bar G}^{mnp} +
\frac{1}{6 Im \tau} \, \chi \, G_{mnp}^{+}{\bar G}^{+ mnp} + \Big( \frac{\chi^{1/2}}{g_s} \Big)^{-1} \Big[ \partial \Big( \frac{\chi^{1/2}}{g_s} \Big) \Big]^2 \nonumber$$ $$+ \Big( \frac{\chi^{1/2}}{g_s} \Big)^{-1} (\partial \xi)^2 + \frac{\beta}{g_s} + 2 \kappa_{10}^2 T_3 \, \chi \, \rho_3^{\rm loc} \,\,.$$ Finally the equation of motion for $\Phi_{-}$ can be obtained by subtracting (7.7) from (7.9). Upon setting $\mu_3 =T_3$, we obtain $$\nabla^2 \Phi_{-} = \frac{g_s}{6 Im \tau} \, \chi \, |G_{(3)}^{+}|^2 + {\chi^{-1/2}} |\partial \Phi_{-}|^2 + \beta \,\,.$$ (7.10) is the string frame version of Eq. (2.8) of [@20], and they coincide if we replace $\chi^{1/2}$ by $e^{4A}$, and $h^{mn}$ by $e^{-\phi /2}h^{mn}$.
0.3cm 0.15cm
(7.10) shows that the IASD fluxes $G_{(3)}^{+}$ become a source for the potential $\Phi_{-}$. The explicit forms of the IASD fluxes can be found systematically by solving the equation of motion [@6; @20] $$d\Lambda + \frac{i}{Im \tau} d\tau \wedge Re \Lambda =0 \,\,$$ perturbatively around ISD solutions. (7.11) can be obtained from a linear combination of the field equations for $A_{(2)}$ and $B_{(2)}$, and where $\Lambda$ is defined by $$\Lambda = \Phi_{+} G_{-} + \Phi_{-} G_{+}\,\,, ~~~~~~ \big( G_{\pm} \equiv \pm i G_{(3)}^{\mp} \big) \,\,.$$ To solve (7.10) perturbatively we expand all fields as [@20] $$X = X_{(0)} + X_{(1)} + X_{(2)} + \cdots \,\,,$$ where $X_{(0)}$ represents the background fields and in particular $\Phi_{-}$ and $G_{-}$ both vanish in the ISD background $$\Phi_{-}^{(0)} = G_{-}^{(0)}=0 \,\,.$$ Since $\Lambda_{(0)}=0$ by (7.14), we need to solve (7.11) for $\Lambda_{(1)}$ which is now given by $$\Lambda_{(1)} = \Phi_{+}^{(0)} G_{-}^{(1)} + \Phi_{-}^{(1)} G_{+}^{(0)}$$ from (7.12) and (7.14).
At first order, (7.11) reduces to [@20] $$d \Lambda_{(1)} = 0\,\,,$$ which requires that $\Lambda_{(1)}$ should be a closed three-form. Also in (7.10) the flux-induced $\Phi_{-}$ should be of the second order because the smallest order of nonvanishing $G_{-}$ is already first order by (7.14). Hence we put $\Phi_{-}^{(1)} =0$ and therefore $\Lambda_{(1)}=\Phi_{+}^{(0)} G_{-}^{(1)}$ from (7.15), which shows that $\Lambda_{(1)}$ is IASD [@20], $$\ast_{6}^{(0)} \Lambda_{(1)} = -i \Lambda_{(1)} \,\,,$$ in the background metric. Finally for $\Phi_{-}^{(1)} =0$, (7.10) reduces to[^9] [@20] $$\nabla^2 \Phi_{-} = \frac{g_s^2}{24} \big| \Lambda \big|^2 \,\,,$$ where $\Phi_{-}$, $\Lambda$ and $ \nabla^2$ are $\Phi_{-} = \Phi_{-}^{(2)}$, $\Lambda = \Lambda_{(1)}$ and $\nabla^2 = \nabla_{(0)}^2$, respectively. So the potential $\Phi_{-}$ arising from the IASD fluxes can be obtained from (7.18) if we know the explicit forms of $\Lambda$ which is any closed, IASD three-form allowed on the Calabi-Yau cones.
Fortunately, the explicit solutions for the flux perturbations on arbitrary Calabi-Yau cones have been thoroughly studied in [@20]. According to the computations of [@20] there exist three distinct types of closed, IASD three-forms. See Sec. 3.3.2 of [@20] for these three types of IASD three-forms. Among these fluxes the type I flux is of particular importance since its contribution to $\Phi_{-}$ is dominant over the other two in the neighborhood of $y=0$ where the visible sector $D3$-branes are located. (We will see this soon.) Also the type II and III fluxes contain non-primitive $(2,1)$ which is forbidden in a compact Calabi-Yau space.[^10]
The potential $\Phi_{-}$ due to the type I flux is found to be [@20] $$\Phi_{-} = \frac{g_s^2}{8} h^{\alpha {\bar \alpha}} \nabla_{\alpha}{f}_1 {\overline{\nabla_{\alpha}{f}_1}} \,\,,$$ which is an $F$-term potential due to the superpotential perturbations of the form $\int d^2 \theta \, \triangle W$ with $\triangle W \sim f_1$. (7.19) suggests that the potential density $V_{D3}$ induced by the type I flux belongs to $V_1$. Indeed from (7.3) and (3.17) $V_{D3}$ can be written as $$V_{D3} = 2 \kappa_{10}^2 g_s T_3 \Phi_{-}(0) \delta^6 (y) \,\,,$$ and since $\Phi_{-}$ in (7.19) contains a single $h^{mn}$ in the real basis, (7.20) shows that $V_{D3} \in V_1$. So $V_{D3}$ induced by the type I flux does not contribute to ${\mathcal V}_{\rm scalar}$ in $\lambda$ (see (3.43)).
0.3cm 0.15cm
In (7.20) $V_{D3}$ is proportional to $\Phi_{-} (0)$ instead of $\Phi_{-} (y)$, which is due to the fact that in our AdS vacuum scenario the $D3$-branes are not the mobile branes anymore because we are considering the present (not inflationary) stage of our universe. In our descriptions of the present universe (a stack of) visible sector $D3$-branes are assumed to be fixed at the apex $y=0$ of the Calabi-Yau cones, and consequently we have the delta-function $\delta^6 (y)$ in (7.20), and also $\Phi_{-} (0)$ instead of $\Phi_{-} (y)$. The presence of delta-function, or having $\Phi_{-} (0)$ instead of $\Phi_{-} (y)$ in $V_{D3}$ enables us to ignore the whole (not just only the type I) contributions to $V_{D3}$ arising from the above three types of flux perturbations. This can be shown as follows.
The field equation (7.18) can be solved by using the green function method. Again in [@20] it was found that the resulting spectrum of $\Phi_{-}$ can be written as[^11] $$\Phi_{-} (y) = \sum_{\delta_{i}, \delta_{j}} r^{\triangle ({\delta_{i}, \delta_{j}})} h_{({\delta_{i}, \delta_{j}})} (\Psi) \,\,,$$ where $h_{({\delta_{i}, \delta_{j}})} (\Psi)$ are angular wave functions which are related to the harmonics $Y_{LM} (\Psi)$ of the unperturbed Laplacian and $\triangle ({\delta_{i}, \delta_{j}})$ are radial scaling dimensions defined by $$\triangle = \delta_{i}+ \delta_{j} -4 \,\,,$$ where $\delta_{i}$ and $\delta_{j}$ are the scaling dimensions of the fluxes $\Lambda_i$ and $\Lambda_j$. The smallest value of $\triangle$ is obtained from a square of $\delta = \frac{5}{2}$ chiral mode of the type I flux, for which $\Phi_{-}$ is linear in $r$, $\Phi_{-} \propto r$. The other smallest scaling dimensions (including the above $\triangle =1$ of $\delta_1 = \delta_2 = \frac{5}{2}$) of the flux-induced potential are (See Sec. 4.1.3 of [@20].) $$\triangle = 1, \, 2, \, \frac{5}{2}, \, \sqrt{28}-\frac{5}{2}, \, \cdots,$$ which shows that the contribution of the type I flux to $\Phi_{-}$ is dominant over the other two as $r \rightarrow 0$. In any case, every term in (7.21) vanishes at $r=0$ for any $\triangle ({\delta_{i}, \delta_{j}})$, and so does $\Phi_{-} (0)$ in (7.20) as well. This suggests that the contributions of the other two can be also ignored $-$ despite that they do not belong to $V_n$ with $n=1,3$ $-$ since $\Phi_{-} (0)$, and therefore $V_{D3}$ itself vanishes for the $D3$-branes fixed at $r=0$ of the Calabi-Yau cones.
Apart from this, the potential $\Phi_{-}$ can also include harmonic functions on the cones as the homogeneous solutions to (7.18). The contributions of these harmonic functions, however, can be also ignored for the $D3$-branes fixed at $r=0$. The harmonic expansion performed on the conifold takes the form [@21] $$f(r, \Psi) = \sum_{L,M} c_{_{LM}} \Big( \frac{r}{r_{_{UV}}} \Big)^{\triangle_f (L)} Y_{LM} (\Psi) +c.c. \,\,, ~~~~~(r < r_{_{UV}})\,\,,$$ where $c_{_{LM}}$ are constant coefficients and the radial scaling dimensions $\triangle_f (L)$ take the values of $$\triangle_f (L) = \frac{3}{2}, \, 2, \, 2, \, 3, \, \sqrt{28}-2, \, \cdots \,\,.$$ Since $\triangle_f (L)$ are all positive, all terms in (7.24) vanish at $r=0$, and therefore we can also ignore these contributions of the harmonic functions to $V_{\rm D3}$ as well.
In addition to these terms there might be a constant term, which is the trivial solution to the Laplace equation $\nabla^2 f=0$. This constant term does not vanish at $r=0$. However, it might be irrelevant to our configurations which do not involve $\overline{D3}$-branes. The constant term appears in the perturbative expansion of the $\overline{D3}$-brane potential $T_3 \Phi_{+} (r;r_{0})$ ($\equiv V_{D3/\overline{D3}} (r)$) (see [@18]). Since mobile $D3$-branes affect $\Phi_{+}$ perturbatively, $V_{D3/\overline{D3}}$ depends on the $D3$-brane position $r$, and it serves as a potential for the $D3$-brane. In this expansion of $V_{D3/\overline{D3}} (r)$ the constant term appears as the unperturbed potential energy of the $\overline{D3}$-branes fixed at $r=r_0$, and therefore it must vanish for the configurations which do not contain $\overline{D3}$-branes. After all, those terms (including harmonic functions) arising from the Coulomb interaction $V_{D3/\overline{D3}}$ between $D3$-branes and $\overline{D3}$-branes must all be excluded from $V_{D3}$ since they are irrelevant to our AdS vacuum scenario which does not involve the $\overline{D3}$-branes at all. See Sec. 5.2.
Besides all this, we finally observe that (3.41) contains the factor $\chi^{1/2}$. In the simple compactifications with $F_{(3)}=H_{(3)}=0$, $\chi^{1/2}$ takes the form (see eq. (5.8) of [@1]) $$\chi^{1/2} (r) = \Big( 1+ \frac{Q_0}{r^4} \Big)^{-1} \,\,,~~~~~\Big( Q_0 \equiv \frac{2 \kappa_{10}^2 g_s \mu_3}{4 \rm Vol(B)}\, \Big)\,\,,$$ and in the neighborhood of $y=0$ it becomes $\chi^{1/2} (r) \sim {r^4} / {Q_0}$.[^12] Hence the densities $V$ which survive the projection $\Pi_\lambda ({\mathcal N}) \equiv \frac{1}{24} \, \chi^{1/2} ({\mathcal N} - 1)({\mathcal N} -3) (1-3 b_0 \Pi ({\mathcal N}))$ in (3.41) (i.e. those $V$’s that do not respect (3.42) with $n=1,3$ just like $V_{D3}$ due to the type II and III IASD fluxes for instance) are highly suppressed again because they all have an extra factor $\chi^{1/2} (r)$ which strongly vanishes at $y=0$ ($r=0$) in the approximation $F_{(3)}=H_{(3)}=0$.[^13]
0.5cm
0.3cm 0.15cm
In an attempt to address the cosmological constant problem (especially aiming at explaining the fine-tuning $\lambda=0$ of our present universe) we have considered a new type of self-tuning mechanism whose basic principle has been partially presented in [@1]. The main point of this self-tuning mechanism can be summarized as
- Whether $\lambda$ vanishes or not is basically determined (in the six-dimensional internal space) by the tensor structure of the scalar potential density $V$, not by the zero or nonzero values of the scalar potential ${\mathcal V}_{\rm scalar}$ itself. If the density of ${\mathcal V}_{\rm scalar}$ belongs to one of $V_n$ with $n=1,3$, then $\lambda$ is forced to be fine-tuned to vanish regardless of whether ${\mathcal V}_{\rm scalar}$ vanishes or not.
- In the new self-tuning mechanism $\lambda$ contains an exceptional term ${\mathcal E}_{\rm SB}$, and this ${\mathcal E}_{\rm SB}$ has its own gauge arbitrariness. So any nonzero ${\mathcal V}_{\rm scalar}$ and quantum fluctuations $\delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ on the branes can be gauged away by this ${\mathcal E}_{\rm SB}$ so that $\lambda$ vanishes as a result. The cancelation between ${\mathcal V}_{\rm scalar} +\delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ and ${\mathcal E}_{\rm SB}$ is automatically achieved by a self-tuning equation (3.41) once the density of ${\mathcal V}_{\rm scalar}$ satisfies $V \in V_n$ with $n=1,3$ as stated above.
- Hence in the new self-tuning mechanism the self-tuning $\lambda=0$ is radiatively stable. Any contributions to ${\mathcal V}_{\rm scalar}$ coming from $g_s$-perturbation and quantum fluctuations on the $D3$-branes are all gauged away by ${\mathcal E}_{\rm SB}$ (and by a self-tuning equation), and as a result $\lambda=0$ is always preserved as mentioned above.
We applied the above self-tuning mechanism to the well-known scenario of KKLT to obtain a realistic model of our present universe with nearly vanishing cosmological constant. As a result of this application we found that the simplest, and perhaps the most natural scenario using KKLT is to take one of the AdS, instead of dS, vacua of KKLT as the background vacuum of our present universe. These AdS vacua are stable both classically and quantum mechanically. They do not have the tunneling instabilities of the dS vacua uplifted by anti-${D3}$-branes. The AdS vacuum scenario suggests that the F-term upliftings in the literature [@2; @22] are basically unnecessary in obtaining a vanishing (or a nearly-vanishing) cosmological constant. The vanishing $\lambda$ is automatically achieved by the self-tuning equation (3.41), and by the gauge arbitrariness of ${\mathcal E}_{\rm SB}$ contained in (1.1). Namely the AdS vacuum scenario, or the self-tuning mechanism of this paper is basically realized by the two unusual equations (1.1) and (3.41).
The first equation (1.1) suggests that the cosmological constant $\lambda$ is not simply given by a scalar potential ${\mathcal V}_{\rm scalar}$ alone. According to (1.1), $\lambda$ contains an additional term, the supersymmetry breaking term ${\mathcal E}_{\rm SB}$, which possesses its own gauge arbitrariness. Hence in our case $\lambda=0$ does not necessarily imply ${\mathcal V}_{\rm scalar}=0$, and AdS vacua with ${\mathcal V}_{\rm scalar}<0$ are not inconsistent with $\lambda =0$ unlike in the theories where $\lambda$ is directly identified with ${\mathcal V}_{\rm scalar}$. In our self-tuning mechanism $\lambda$ is generally given by (3.20). But in the AdS vacuum scenario proposed in Sec. 5.2, $V$ is basically given by $V_{\rm AdS}$, and in this case (3.20) reduces (upon using (3.17)) back to (1.1) by (3.40) and (3.45) because $V_{\rm AdS} \in V_3$ and therefore $({\mathcal N}-1) V_{\rm AdS} = 2V_{\rm AdS}$. This result does not change even when we add $V_1$ (for instance, $V_{\rm D3}$ due to $\Phi_{-}$ in (7.19)) to $V$ because $({\mathcal N}-1) V _1$ simply vanishes.
Together with (1.1), the second equation (3.41) suggests that whether $\lambda$ vanishes or not is basically determined by the tensor structure of the potential density $V$, not by the zero or nonzero values of ${\mathcal V}_{\rm scalar}$ itself. (3.41) leads to the self-tuning $\lambda=0$ once our $V$ belongs to one of the class $V_n$ with $n=1,3$. We have shown that the AdS vacua of KKLT (including open string moduli of $D3$-branes) belong to $V_3$. So $\lambda$ of our present universe must tune itself to zero in the AdS vacuum scenario of Sec. 5.2. The negative values of ${\mathcal V}_{\rm scalar}$ of the AdS vacua are gauged away by ${\mathcal E}_{\rm SB}$ in (1.1), and $\lambda =0$ is automatically achieved by (3.41). This self-tuning process is not affected by the perturbations ${\mathcal K}= {\mathcal K}_{\rm tree} + {\mathcal K}_{\rm p} + {\mathcal K}_{\rm np}$ and $W= W_{\rm tree} + W_{\rm np}$ because these perturbations do not change the tensor structure of $V$. (See the last paragraph of Sec. IV.) Thus the whole radiative corrections of ${\mathcal V}_{\rm scalar}$ are also gauged away by (3.41), and the fine-tuning $\lambda =0$ remains stable against these corrections in the self-tuning mechanism of this paper.
The background vacua of the AdS vacuum scenario are supersymmetric, and therefore stable unlike the dS vacua uplifted by anti-${D3}$-branes. In the descriptions in [@15] the dS vacua necessarily involve the anti-$D3$-branes. So they are not supersymmetric and they are stable only classically at most. The dS vacua must eventually decay into the supersymmetric configurations of AdS vacua by the brane/flux annihilation process of [@15], and this also suggests that the AdS vacuum scenario is more natural description of our present universe as compared with the dS vacua uplifted by anti-${D3}$-branes.
0.3cm 0.15cm
In the AdS vacuum scenario the supersymmetry is basically broken by ${\mathcal E}_{\rm SB}$ in (1.1) (and by IASD components of the three-form fluxes arising from the perturbations), not by anti-$D3$-branes. In order to see it we rewrite (3.3) plus (3.39) as $$I_{\rm brane}= \Big[ \int d^4 x \sqrt{-g_4} \, \Big] \int r^5 dr \epsilon_5 \Big( - e^{2B} T(\phi) + \mu (\phi) \xi (r) + \delta \mu_{T}^m (\phi) f_m (y) \Big) \delta^6 (r) \,\,,$$ where $T(\phi)$ and $\mu (\phi)$ are given by $$T(\phi) = T_3 e^{-\phi} + \rho_{\rm vac} (\phi)\,\,,~~~~~\mu (\phi) = \mu_3 + \delta \mu (\phi)\,\,,$$ because we are now taking quantum fluctuations on the branes into account. The last term of (8.1) occurs as a result of the gauge symmetry breaking of $A_{(4)}$ arising at the quantum level. The substance of this term is a vacuum energy density of the brane region arising from the quantum excitations with components along the transverse directions to the $D3$-branes, and it plays very important roles in the supersymmetry breaking of the brane region, and in the process of self-tuning $\lambda =0$.
0.3cm 0.15cm
The supersymmetry transformations of the fermi fields of type IIB supergravity are [@23] $$\delta \chi_{\phi} = \frac{1}{2} \Gamma^m (\partial_m \phi) \eta + \frac{i}{4} e^{\phi}\, {\bar {\mathbf G}}^{(3)} \eta^{\ast} \,\,,$$ $$\delta \psi_m = \nabla_m \eta + \frac{i}{16} e^{\phi} \, {\tilde {\mathbf F}}^{(5)} \Gamma_m \eta - \frac{1}{8} \big(2 {\mathbf H}_m^{(3)} +i e^{\phi} \,{\mathbf F}^{(3)} \Gamma_m \big) \eta^{\ast} \,\,,$$ where ${\mathbf F}^{(n)}$, ${\mathbf F}_m^{(n)}$ are defined by $${\mathbf F}^{(n)} \equiv \frac{1}{n!} \,\Gamma^{M_{1} \cdots M_{n}} F_{M_{1} \cdots M_{n}}\,\,,~~~~~{\mathbf F}_m^{(n)}\equiv \frac{1}{(n-1)!} \, \Gamma^{M_{1} \cdots M_{n-1}} F_{m M_{1} \cdots M_{n-1}}\,\,.$$ In (8.3) and (8.4), the last terms represent supersymmetry transformations generated by the three-form fluxes $F_{(3)}$ and $H_{(3)}$. But if we want an easy understanding of the supersymmetry breaking of the AdS background, it is useful to consider a simple situation where the three-form fluxes are turned off, $F_{(3)}=H_{(3)}=0$ or $G_{(3)}=0$. Note that such compactification is a good approximation in our AdS vacuum scenario because in KKLT the superpotential $W_0$ (and therefore $G_{(3)}$) is only of an order $\sim 10^{-4}$ (see footnote 11). So the supersymmetry breaking generated by the three-form fluxes $-$ regardless of whether ISD or IASD $-$ could be neglected for a moment in the simplified analysis for the core principle. For $F_{(3)}=H_{(3)}=0$, (8.3) and (8.4) reduce to $$\delta \chi_{\phi} = \frac{1}{2} \Gamma^m (\partial_m \phi) \eta \,\,, ~~~~~~\delta \psi_m = \nabla_m \eta + \frac{i}{16} e^{\phi} \, {\tilde {\mathbf F}}^{(5)} \Gamma_m \eta\,\,,$$ and these $\delta \chi_{\phi}$ and $\delta \psi_m$ vanish for (7.26) and constant $\phi$ [@24]. Hence in the approximation $F_{(3)}=H_{(3)}=0$, the supersymmetry is unbroken when $\phi$ is constant [@1].
Now consider the field equation for $\phi$. Using (8.1) we obtain $$\nabla^2 \phi - \frac{i g_s^2}{12} \Big( \frac{\Phi_{-}}{g_s} \Big) e^{\phi} G_{mnp} \ast_6 {\bar G}^{mnp} - \frac{g_s^2}{6} \Big( \frac{\chi^{1/2}}{g_s} \Big) e^{\phi} G^{+}_{mnp}{\bar G}^{+mnp} ~~~~~~~~~~~~~~~~~~~~\nonumber$$ $$~~~~~= 2 \kappa_{10}^2 g_s^2 \Bigg[ e^{2B} \Big( T(\phi) + \frac{\partial T(\phi)}{\partial \phi} \Big) - \frac{\partial \mu(\phi)}{\partial \phi} \xi (r)- \frac{\partial \delta \mu_{T}^{m}(\phi)}{\partial \phi} f_m (y) \Bigg] \delta^6 ({\vec r})$$ from a linear combination of the field equations for $\hat{\phi}$ and $B$. In the given approximation (8.7) reduces to $$\nabla^2 \phi =0$$ in the bulk region, and therefore the bulk supersymmetry remains unbroken because (8.8) admits constant solutions. In the brane region, however, (8.7) reduces to $$\nabla^2 \phi = c_0 \chi^{1/2} e^{\phi} \Big( \rho_{\rm vac} + \frac{\partial \rho_{\rm vac}}{\partial \phi} \Big) - c_0 g_s \Big( \frac{\partial \delta \mu} {\partial \phi} \xi + \frac{\partial \delta \mu_{T}^{m}} {\partial \phi} f_m \Big) \,\,,$$ where $c_0 = 2 \kappa_{10}^2 g_s \delta_0$ (see eq. (6.12) of [@1]), and $\rho_{\rm vac}$, $\delta \mu$, $\delta \mu_T^m$ are expanded respectively as $$\rho_{\rm vac} (\phi) = \sum_{n=0}^{\infty} \rho_{(n)}\, e^{n \phi} \,\,,~~~~~\delta \mu (\phi) = \sum_{n=1}^{\infty} \mu_{(n)}\, e^{n \phi} \,\,, ~~~~~\delta \mu_T^m (\phi) = \sum_{n=1}^{\infty} \nu_{(n)}^m \, e^{n \phi} \,\,.$$ In [@1] it was shown (up to one-loop level) that all but the last term in (8.9) cancel out for $\mu_3 =T_3$ and $\mu_{(1)} = \rho_{(0)}$ which are required by consistency equations (see Sec. VIC of [@1]), and we are left with $$\nabla^2 \phi =- c_0 \rho_{T}^{(1)} \,\,,~~~~~~ \big( \rho_{T}^{(1)} = \nu_{(1)}^m f_m \big) \,\,,$$ and similarly we obtain $${\hat I}_{\rm brane} = \delta_0 \int r^5 dr \epsilon_5 \rho_{T}^{(1)}$$ from (8.1) (see eq. (7.2) of [@1]). Since (8.12) comes from the last term of (8.1), it is identified (at one-loop level) with $\delta_G {\hat I}_{\rm brane}^{(R)}$, or equivalently with $-{\mathcal E}_{\rm SB}$ by (3.45) (see (1.2)). In (8.11), $\rho_{T}^{(1)}$ sources the supersymmetry breaking of the brane region, and ${\mathcal E}_{\rm SB}$ in (1.2) plays the role of a supersymmetry breaking term because the term $-c_0 \rho_{T}^{(1)}$ in (8.11) is obtained from the last term $\delta_G {\hat I}_{\rm brane}^{(R)} (=-{\mathcal E}_{\rm SB})$ of the action ${\hat I}_{\rm brane}$. Hence in the brane region the supersymmetry is broken by $\rho_{T}^{(1)}$ even in the absence of three-form fluxes, while in the bulk region it remains unbroken in that approximation.
Turning back to (1.1), ${\mathcal V}_{\rm scalar}$ of the AdS vacua vanishes for $G_{(3)}=0$ because the supersymmetric AdS vacua are defined by $DW =0$, which then implies ${\mathcal V}_{\rm AdS} \propto |W_{0}|^2$ from (2.4) and (4.22) (see (4.23)), and therefore ${\mathcal V}_{\rm AdS}$ vanishes for $G_{(3)}=0$ because so does $W_0$. Hence in the absence of the three-form fluxes (1.1) becomes $$\lambda = \frac{\kappa^2}{2} \big( \delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)} -{\mathcal E}_{\rm SB} \big)\,\,,$$ and from (8.13) the self-tuning $\lambda=0$ requires that the energy scale ${\mathcal E}_{\rm SB}$ of the supersymmetry breaking must be equal to the magnitude of the non-vanishing fluctuations $\delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ on the branes. Also if the branes are BPS $D3$-branes with unbroken supersymmetry (which is the simplest, but not realistic case), $\delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ is expected to cancel out and $\lambda$ is simply given by $$\lambda = \frac{\kappa^2}{2} Q_{\rm total}^T \,\,,$$ where $$Q_{\rm total}^T \equiv -{\mathcal E}_{\rm SB}^{(1)} = \delta_0 \int_{r=0}^{r_B} r^5 dr \epsilon_5 \rho_{T}^{(1)} \,\,.$$ $Q_{\rm total}^T $ represents the total vacuum energy (per unit volume of the four-dimensional spacetime) of the brane region which originated from the excitations with components along the transverse directions to the D3-branes. Now in this case the point of the cosmological constant problem can be summarized as whether we can find a nonzero function $\rho_{T}^{(1)}$ satisfying $Q_{\rm total}^T =0$. The existence of such functions implies nonsupersymmetric configurations with vanishing $\lambda$, and important examples of such functions have been found in [@1] (see Sec. VII).
0.3cm 0.15cm
Let us now turn on the three-form fluxes $G_{(3)}$ to obtain a full description of the supersymmetry breaking of our AdS vacuum scenario. In the ISD compactifications ($\Phi_{-} = G^{+}_{mnp}=0$), the dilaton $\phi$ still satisfies (8.8) and (8.11) even in the presence of nonzero $G_{(3)}$. However, these ISD compactifications are not appropriate to the general cases of our AdS vacuum scenario because in the AdS minimum, the unbroken supersymmetry $DW=0$ requires that $G_{(3)}$ must also contain IASD $(1,2)$ and $(3,0)$ in addition to the ISD $(2,1)$ and $(0,3)$ (see [@9]). These IASD components of the AdS background are entirely due to the nonperturbative corrections of the superpotential and they have nothing to do with the perturbative corrections which also give rise to the IASD components of $G_{(3)}$ and $\Phi_{-}$. In any case, the IASD terms with $G^{+}_{mnp} {\bar G}^{+mnp}$ or $\Phi_{-}$ acquire nonzero values from both perturbative and nonperturbative corrections, and they are now involved $-$ together with those terms caused by $\delta_G {\hat I}_{\rm brane}^{(R)} (=-{\mathcal E}_{\rm SB})$ $-$ in the supersymmetry transformations in some complicated manner. But still, if we ignore the perturbative corrections and supersymmetry breaking generated by $\rho_T^{(1)}$ for a moment, then we expect the solution to the equations of motion becomes a supersymmetric solution satisfying $\delta \chi_{\phi}=\delta \psi_m =0$.[^14] Namely the supersymmetry of the AdS background is simply given by $\delta \chi_{\phi}=\delta \psi_m =0$ at the tree level.
Now we finally turn to the situation where the perturbative corrections and supersymmetry breaking generated by ${\mathcal E}_{\rm SB}$ are both taken into account. In this case $\delta \chi_{\phi}$ and $\delta \psi_{m}$ fail to vanish since they now acquire the terms coming from the perturbations and supersymmetry breaking, and consequently the supersymmetries of the brane and bulk regions are both broken. In the case of ${\mathcal V}_{\rm scalar}$, however, the situation is a little different. In the AdS minima of KKLT the scalar potential ${\mathcal V}_{\rm scalar}$ receives contributions both from perturbative and nonperturbative corrections. Hence in this case, ${\mathcal V}_{\rm scalar}$ already takes nonzero values even when perturbations and supersymmetry breakings are not taken into account in the theory yet. However, in our self-tuning mechanism any nonzero contributions to ${\mathcal V}_{\rm scalar}$ coming from perturbative and nonperturbative corrections, and also the contributions coming from the IASD fluxes described above are all gauged away by ${\mathcal E}_{\rm SB}$ in (1.1), and $\lambda =0$ is always preserved even when supersymmetry of the system is broken by the perturbations and supersymmetry breaking term ${\mathcal E}_{\rm SB}$.
In (1.2), we decompose $\rho_T^{(1)}$ into ${\tilde \rho}_T^{(1)} + \delta \rho_T^{(1)}$ to get ${\mathcal E}_{\rm SB} \rightarrow {\tilde {\mathcal E}_{\rm SB}} +\delta {\mathcal E}_{\rm SB}$, where ${\tilde {\mathcal E}_{\rm SB}}$ and $\delta {\mathcal E}_{\rm SB}$ are $${\tilde {\mathcal E}_{\rm SB}} = - \delta_0 \int r^5 dr \epsilon_5 {\tilde \rho}_T^{(1)} \,\,, ~~~~~\delta {\mathcal E}_{\rm SB} = -\delta_0 \int r^5 dr \epsilon_5 \delta \rho_T^{(1)} \,\,.$$ ${\tilde \rho}_T^{(1)}$ and $\delta \rho_T^{(1)}$ in ${\mathcal E}_{\rm SB}$ are arbitrary because they contain six arbitrary gauge parameters $f_m^{(0)}(y)$. Hence if we adjust $\delta \rho_T^{(1)}$ such that $\delta {\mathcal E}_{\rm SB}$ cancels nonzero deviations of ${\mathcal V}_{\rm scalar}$ plus $\delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ on the brane, then (1.1) reduces to $$\lambda = \frac{\kappa^2}{2} {\tilde Q}_{\rm total}^T \,\,,$$ where ${\tilde Q}_{\rm total}^T$ is the generalized version of (8.15), $${\tilde Q}_{\rm total}^T \equiv \delta_0 \int_{r=0}^{r_B} r^5 dr \epsilon_5 {\tilde \rho}_T^{(1)} \,\,.$$ The adjustment of $\delta \rho_T^{(1)}$, or the cancelation between $\delta {\mathcal E}_{\rm SB}$ and ${\mathcal V}_{\rm scalar} + \delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)}$ in (1.1) is automatic by the self-tuning $\lambda =0$ as required by (3.41), and ${\tilde Q}_{\rm total}^T$ now plays the role of $Q_{\rm total}^T$ as one can see from (8.14) and (8.17). So if we want a nonsupersymmetric theory with $\lambda=0$, we may need to find a nonzero function ${\tilde \rho}_T^{(1)}$ satisfying ${\tilde Q}_{\rm total}^T =0$ as in the case of $G_{(3)} =0$. But still, it may also be possible to take simply $${\mathcal E}_{\rm SB} = {\mathcal V}_{\rm scalar} + \delta_Q {\hat I}_{\rm brane}^{(NS)} + \delta_Q {\hat I}_{\rm brane}^{(R)} \,\,,$$ because this ${\mathcal E}_{\rm SB}$ would be large enough to break the supersymmetry of the system sufficiently.
0.15cm
So far we have considered a new type of self-tuning mechanism to address the cosmological constant problem, especially aiming at explaining the fine-tuning $\lambda =0$ of our present universe. But more precisely, $\lambda$ of our present universe is known to take a positive value though it is very small. So the next step of the project would be this issue of identifying small positive $\lambda$ of our present universe.
In this paper we have considered a theory based on the type IIB supergravity, and from this supergravity action we obtained a result that $\lambda$ must vanish precisely if the density of ${\mathcal V}_{\rm scalar}$ satisfies $V \in V_n$ with $n=1,3$ as in the AdS vacua of our self-tuning mechanism. We have also shown that the result $\lambda=0$ of our AdS vacua is stable against $g_s$-perturbations. But full string theory requires the action to admit $\alpha^{\prime}$-corrections which are usually higher order in derivatives, and due to these corrections the self-tuning equation (3.41) may be modified into the corrected form.
Besides this, in the case of the type I or the heterotic theory the $\alpha^{\prime}$-corrections contain extra terms which do not satisfy $V \in V_n$ with $n=1$ or 3, and these terms also require that $\lambda$ must take nonzero values. Namely if we take the stringy (or any other) effects which have not been considered in this paper into account, we may expect a result with nonvanishing $\lambda$. But still, once $\lambda$ is determined by (the modified) (3.41), these nonzero values of $\lambda$ will be stable against quantum corrections as in the case $\lambda=0$ of this paper because (3.41) is based on the self-tuning mechanism where the perturbative corrections of ${\mathcal V}_{\rm scalar}$ and quantum fluctuations on the branes are always gauged away by ${\mathcal E}_{\rm SB}$ in (1.1). So the result obtained from (3.41) needs to be distinguished from the result of nonvanishing $\lambda$ due to $\alpha^{\prime}$-corrections in the literature [@25] in this sense.
In any case, if some convincing values of $\lambda$ is obtained from (3.41) modified by $\alpha^{\prime}$-corrections, then we may say that the nonzeroness of $\lambda$ of our present universe is essentially due to the stringy effect of the string theory, because $\lambda$ vanishes in the absence of $\alpha^{\prime}$-corrections and this result was not affected by the $g_s$-perturbations in our self-tuning mechanism of this paper. But any nonzero values of $\lambda$ suggested by (3.41) will be highly suppressed again by the factor $\chi^{1/2}$ as stated in the last paragraph of Sec. 7.4, and hence $\lambda$ obtained from (3.41) would be very small anyway.
[**Acknowledgement**]{}
This work was supported by the National Research Foundation of Korea (NRF), under Grant No. 353-2009-2-C00045, funded by the Korean Government.
[999]{}
E. K. Park, and P. S. Kwon, [*Remark on Calabi-Yau vacua of the string theory and the cosmological constant problem*]{}, Phys. Rev. [**D88**]{} (2013) 046007 \[arXiv:1301.1783\].
E. Cremmer, S. Ferrara, C. Kounnas, and D. V. Nanopoulos, [*Naturally vanishing cosmological constant in $N=1$ supergravity*]{}, Phys. Lett. [**B133**]{} (1983) 61.
S. Kachru, R. Kallosh, A. Linde, and S. P. Trivedi, [*de Sitter Vacua in String Theory*]{}, Phys. Rev. [**D68**]{} (2003) 046005 \[arXiv:hep-th/0301240\].
I. R. Klebanov, and M.J. Strassler, [*Supergravity and a confining gauge theory: duality cascades and $\chi$SB-resolution of naked singularities*]{}, J. High Energy Phys. [**08**]{} (2000) 052 \[arXiv:hep-th/0007191\].
E. Witten, [*Nonperturbative superpotentials in string theory*]{}, Nucl. Phys. [**B474**]{} (1996) 343 \[arXiv:hep-th/9604030\].
P. Tripathy, and S. P. Trivedi, [*Copmactification with flux on $K3$ and Tori*]{}, \[arXiv:hep-th/0301139\].
S. B. Giddings, S. Kachru, and J. Polchinski, [*Hierarchies from Fluxes in String Compactifications*]{}, Phys. Rev. [**D66**]{} (2002) 106006 \[arXiv:hep-th/0105097\].
J. Ellis, A.B. Lahanas, D.V. Nanopoulos, and K. Tamvakis, [*No-scale Supersymmetric Standard Model*]{}, Phys. Lett. [**B134**]{} (1984) 429.
K. Choi, A. Falkowski, H. P. Nilles, M. Olechowski, and S. Pokorski, [*Stability of flux compactifications and the pattern of supersymmetry breaking*]{}, J. High Energy Phys. [**11**]{} (2004) 076 \[arXiv:hep-th/0411066\]; R. Brustein, and S. P. de Alwis, [*Moduli potentials in string compactifications with fluxes: mapping the Discretuum*]{}, Phys. Rev. [**D69**]{} (2004) 126006 \[arXiv:hep-th/0402088\].
G. Curio, A. Krause, and D. L$\ddot{\rm u}$st, [*Moduli Stabilization in the Heterotic/IIB Discretuum*]{}, Fortsch. Phys. [**54**]{} (2006) 225-245 \[arXiv:hep-th/0502168\].
S. Kachru, M. Schulz, and E. Silverstein, [*Self-tuning flat domain walls in 5d gravity and string theory*]{}, Phys. Rev. [**D62**]{} (2000) 045021 \[arXiv:hep-th/0001206\].
M. Dine, and N. Seiberg, [*Nonrenormalization Theorems in Superstring Theory*]{}, Phys. Rev. Lett. [**57**]{} (1986) 2625 ; C. P. Burgess, C. Escoda, and F. Quevedo, [*Nonrenormalization of flux superpotentials in string theory*]{}, J. High Energy Phys. [**06**]{} (2006) 044 \[arXiv:hep-th/0510213\] ; K. Becker, M. Becker, C. Vafa, and J. Walcher, [*Moduli stabilization in non-geometric backgrounds*]{}, Nucl. Phys. [**B770**]{} (2007) 1 \[arXiv:hep-th/0611001\] ; M. Cicoli, J. P. Conlon, and F. Quevedo, [*Systematics of String Loop Corrections in Type IIB Calabi-Yau Flux Compactifications*]{}, J. High Energy Phys. [**01**]{} (2008) 052 \[arXiv:0708.1873\].
M. Dine, R. Rohm, N. Seiberg, and E. Witten, [*Gluino condensation in superstring models*]{}, Phys. Lett. [**B156**]{} (1985) 55; E. A. Bergshoeff and M. de Roo, [*The quartic effective action of the heterotic string and supersymmetry*]{}, Nucl. Phys. [**B328**]{} (1989) 439; A. Krause, [*Supersymmetry Breaking with Zero Vacuum Energy in M-Theory Flux Compactifications*]{}, Phys. Rev. Lett. [**98**]{} (2007) 241601 \[arXiv:hep-th/0701009\].
D. L$\ddot{\rm u}$st, S. Reffert, E. Scheidegger, W. Schulgin, and S. Stieberger, [*Moduli Stabilization in Type IIB Orientifolds (II)*]{} Nucl. Phys. [**B766**]{} (2007) 178 \[arXiv:hep-th/0609013\].
O. DeWolfe, and S. B. Giddings, [*Scales and hierarchies in warped compactifications and brane worlds*]{}, Phys. Rev. [**D67**]{} (2003) 066008 \[hep-th/0208123\]
S. Kachru, John Pearson, and H. Verlinde, [*Brane/Flux Annihilation and the String Dual of a Non-Supersymmetric Field Theory*]{}, J. High Energy Phys. [**06**]{} (2002) 021 \[arXiv:hep-th/0112197\].
J. Polchinski, and M. J. Strassler, [*The String Dual of a Confining Four-Dimensional Gauge Theory*]{}, \[arXiv:hep-th/0003136\].
S. Kachru, R. Kallosh, A. Linde, J. Maldacena, L. McAllister, and S. P. Trivedi, [*Towards Inflation in String Theory*]{}, JCAP [**10**]{} (2003) 013 \[arXiv:hep-th/0308055\] ; D. Baumann, A. Dymarsky, I. R. Klebanov, and L. McAllister, [*Towards an Explicit Model of D-brane Inflation*]{}, JCAP [**01**]{} (2008) 024 \[arXiv:0706.0360\].
D. Baumann, A. Dymarsky, S. Kachru, I. R. Klebanov, and L. McAllister, [*Holographic Systematics of D-brane Inflation*]{}, J. High Energy Phys. [**03**]{} (2009) 093 \[arXiv:0808.2811\].
O. DeWolfe, L. McAllister, G. Shiu, and B. Underwood, [*D3-brane Vacua in Stabilized Compactifications*]{}, J. High Energy Phys. [**09**]{} (2007) 121 \[arXiv:hep-th/0703088\].
D. Baumann, A. Dymarsky, S. Kachru, I. R. Klebanov, and L. McAllister, [*D3-brane Potentials from Fluxes in AdS/CFT*]{}, \[arXiv:1001.5028\].
K. Becker, Y. C. Chung, and G. Guo, [*Metastable Flux Configurations and de Sitter Spaces*]{}, Nucl. Phys. [**B790**]{} (2008) 240 \[arXiv:0706.2502\].
A. Ceresole, G. Dall’Agata, R. D’Auria, and S. Ferrara, [*Spectrum of Type IIB Supergravity on $AdS_5 \times T^{1,1}$: Predictions on $N=1$ SCFT’s*]{}, Phys. Rev. [**D61**]{} (2000) 066001 \[arXiv:hep-th/9905226\].
O. Lebedev, H. P. Nilles, and M. Ratz, [*De Sitter Vacua from Matter Superpotentials*]{}, Phys. Lett. [**B636**]{} (2006) 126 \[arXiv:hep-th/0603047\] ; E. Dudas, C. Papineau, and S. Pokorski, [*Moduli stabilization and uplifting with dynamically generated F-terms*]{}, J. High Energy Phys. [**02**]{} (2007) 028 \[arXiv:hep-th/0610297\] ; H. Abe, T. Higaki, T. Kobayashi, and Y. Omura, [*Moduli stabilization, F-term uplifting and soft supersymmetry breaking terms*]{}, Phys. Rev. [**D75**]{} (2007) 025019 \[arXiv:hep-th/0611024\] ; R. Kallosh, and A. Linde, [*O’KKLT*]{}, J. High Energy Phys. [**02**]{} (2007) 002 \[arXiv:hep-th/0611183\].
S. S. Gubser, [*Supersymmetry and F-theory realization of the deformed conifold with three-form flux*]{}, \[arXiv:hep-th/0010010\] ; K. Becker, M. Becker and J. H. Schwarz, [*String theory and M-Theory*]{}, Cambridge University Press (2007).
M. J. Duff, and J. X. Lu, [*The self-dual type IIB superthreebrane*]{}, Phys. Lett. [**B273**]{} (1991) 409.
See, for instance, K. Becker, M. Becker, M. Haack, and J. Louis, [*Supersymmetry breaking and $\alpha^{\prime}$-corrections to flux induced potentials*]{}, J. High Energy Phys. [**06**]{} (2002) 060 \[arXiv:hep-th/0204254\] ; V. Balasubramanian, and P. Berglund, [*Stringy corrections to Kahler potentials, SUSY breaking, and the cosmological constant problem*]{}, J. High Energy Phys. [**11**]{} (2004) 085 \[arXiv:hep-th/0408054\] ; F. F. Gautason, D. Junghans, and M. Zagermann, [*On Cosmological Constants from $\alpha^{\prime}$-Corrections*]{}, J. High Energy Phys. [**06**]{} (2012) 029 \[arXiv:1204.0807\]. Also see the references therein.
[^1]: E-mail: [email protected]
[^2]: E-mail: [email protected]
[^3]: Our convention is $\rho = \frac{b}{\sqrt 2} + i e^{4u}$ and the prefactor $\frac{1}{2 \kappa_{10}^2} e^{{\mathcal K}_{\tau}+{\mathcal K}_{\rm cs}}$ of (2.8) has been omitted in (2.9).
[^4]: There is another viewpoint on this backreaction problem. For instance in Sec. III of ref. \[1\] (also see ref. \[6\] therein) it was argued that the Calabi-Yau three-folds may be thought of as NS-NS solitons whose ADM masses are proportional to $1/ g_s^2$. Hence in the limit $g_s \rightarrow 0$ these Calabi-Yau three-folds are very heavy and rigid, and consequently deformations of internal geometry due to backreactions are highly suppressed.
[^5]: In obtaining (4.1) we have used the identity $G_{(3)} \wedge \ast_6 {\bar G}_{(3)}= - i G_{(3)} \wedge {\bar G}_{(3)} + 2i G_{(3)}^+ \wedge {\bar G}_{(3)}^{+}$.
[^6]: To obtain (2.9) one should insert the scale factor $e^{2u(x)}$ of the internal space in the metric in advance.
[^7]: For simplicity we consider the configuration which only has a single K$\ddot{\rm a}$hler modulus as in KKLT.
[^8]: In our AdS vacuum scenario $D3$-branes are fixed at the apex of the Calabi-Yau cone, so in our case $Y^{\alpha}$ is simply $Y^{\alpha}=0$.
[^9]: In (7.18) we have set $\beta =0$ because we are considering the present stage (with $\lambda=0$) of our universe.
[^10]: But for chiral perturbations, each flux becomes of pure Hodge type, and the type II and III fluxes do not contain non-primitive $(2,1)$ anymore.
[^11]: We would like to thank the authors of [@20] for presenting very useful results of the complete studies on the issue under discussion.
[^12]: $Q_0$ in $\chi^{1/2}$ will cancel with $2 \kappa_{10}^2 g_s T_3$ in $V_{\rm D3}$ (see (7.20)) in the self-tuning equation (3.41).
[^13]: The compactifications with $F_{(3)}=H_{(3)}=0$ are good approximations in the AdS vacuum scenario because in KKLT the superpotential $W_0$ (and therefore $G_{(3)}$) is only of an order $\sim 10^{-4}$, instead of $\sim O(1)$.
[^14]: Supersymmetric solutions of type II theories have been discussed, for instance, in [@23]. (Also see the last paper in [@12].)
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'One-dimensional topological phases can host localized zero-energy modes that enable high-fidelity storage and manipulation of quantum information. Majorana fermion chains support a classic example of such a phase, having zero modes that guarantee two-fold degeneracy in all eigenstates up to exponentially small finite-size corrections. Chains of ‘parafermions’—generalized Majorana fermions—also support localized zero modes, but, curiously, only under much more restricted circumstances. We shed light on the enigmatic zero-mode stability in parafermion chains by analytically and numerically studying the spectrum and developing an intuitive physical picture in terms of domain-wall dynamics. Specifically, we show that even if the system resides in a gapped topological phase with an exponentially accurate ground-state degeneracy, higher-energy states can exhibit a splitting that scales as a *power law* with system size—categorically ruling out exact localized zero modes. The transition to power-law behavior is described by critical behavior appearing exclusively within excited states.'
author:
- 'Adam S. Jermyn'
- 'Roger S. K. Mong'
- Jason Alicea
- Paul Fendley
bibliography:
- 'zeroModes.bib'
title: Stability of zero modes in parafermion chains
---
Introduction
============
Topological quantum computing represents a promising and conceptually elegant route to scalable quantum computation.[@kitaev; @TQCreview] Underlying this approach are topological phases of matter that harbor emergent particles known as non-Abelian anyons. Such particles exhibit two defining features: $(i)$ they generate a ground-state degeneracy that scales exponentially with the number of anyons present in the system,[@MooreRead:Nonabelion:1991; @NayakWilczek:2nStatesQHPf96] and $(ii)$ braiding the anyons around one another non-commutatively rotates the system’s quantum state within the ground-state manifold.[@MooreSeiberg:89; @Witten:JonesPolynomial:1989] The advantage that topological quantum computation offers over more traditional quantum computing schemes is that information is encoded and processed *non-locally* in the braiding history of non-Abelian anyons. Local environmental perturbations that ordinarily cause decoherence are thereby expected to be relatively benign.
One conceptually simple realization of non-Abelian anyons are quasiparticles (or defects) that bind exponentially localized, topologically protected zero-energy modes.[@ReadGreen:p+ipFQHE:00; @KitaevWireMajorana:01] These modes are described by operators with appreciable weight on some finite length scale $\xi$ and that commute with the system’s Hamiltonian up to exponentially small corrections $\sim e^{-R/\xi}$, where $R$ is the separation between adjacent anyons. The localized character of the zero modes ensures well-defined braiding relations for the anyons that bind them, while the fact that they carry no energy guarantees the ground-state degeneracy necessary for non-Abelian statistics. It is worth emphasizing, however, that zero modes so defined make an extremely strong statement about the system’s spectrum: they imply an exponentially accurate degeneracy not just for ground states, but in fact for *all* eigenstates.
![Schematic illustration of either the Majorana or parafermion chain. Adjacent sites couple with strength $f$ or $J$ as labeled above. []{data-label="fig:jfCouplingKitaev"}](chain_fig){width="\columnwidth"}
The Kitaev chain [@KitaevWireMajorana:01] provides an illuminating example. The Hamiltonian reads $$\begin{aligned}
H=if\sum_x \gamma_{2x-1}\gamma_{2x}+iJ\sum_x \gamma_{2x}\gamma_{2x+1},\end{aligned}$$ where $\gamma_x$ denotes a Hermitian Majorana fermion operator satisfying the commutation relation $\{\gamma_x,\gamma_{x'}\} = 2\delta_{x,x'}$. As Fig. \[fig:jfCouplingKitaev\] illustrates, the couplings $f$ and $J$ favor competing Majorana dimerization patterns. In the special case with $f=0$, the outermost Majorana operators completely decouple from the rest of the system, commute with $H$, and thus form exact localized zero modes. Consequently, every eigenstate assumes at least two-fold degeneracy. Turning on finite $f$ preserves the zero modes—and with it the degeneracy in the spectrum—provided that $|f| < |J|$.[@KitaevWireMajorana:01] Rather than localizing to one site the zero modes then simply decay exponentially into the bulk on the scale of the correlation length (which diverges at $|f| = |J|$). The survival of localized zero modes indicates that the chain resides in the same topologically nontrivial phase for any $|f| < |J|$. Throughout this phase the ends of the chain behave as non-Abelian anyons whose non-trivial exchange statistics can be meaningfully harvested in networks.[@AliceaBraiding; @ClarkeBraiding; @HalperinBraiding; @BondersonBraiding] Braiding this type of anyon, however, enables only rather limited (i.e., non-universal) fault-tolerant quantum information processing.[@BravyiKitaev]
In the pursuit of non-Abelian anyons with greater utility for quantum computation, a variation of the Kitaev chain due to Fendley[@Fendley] has proven influential. The Hamiltonian for this ‘parafermion chain’ (which Sec. \[Model\] discusses in depth) is given by $$\begin{aligned}
H=-f\sum_x\alpha_{2x-1}^\dagger\alpha_{2x}-J\sum_x \alpha_{2x}^\dagger\alpha_{2x+1}+{{h.c.}}\label{Hpara}\end{aligned}$$ Here $\alpha_x$ denotes a parafermion operator satisfying a $\mathbb{Z}_3$ generalization of the Majorana fermion algebra: $$\alpha_x^3 = 1,~~~~ \alpha_x^\dagger = \alpha^2_x,~~~~ \alpha_x \alpha_{x'>x} = e^{i2\pi /3} \alpha_{x'}\alpha_x.
\label{alpha_properties}$$ Figure \[fig:jfCouplingKitaev\] still illustrates the structure of the couplings, which we assume are real and non-negative throughout. With $f=0$ the outermost operators drop out from the Hamiltonian—precisely as in the Kitaev chain—and represent localized parafermion zero modes[^1] that guarantee a three-fold degeneracy of every eigenstate.
Similarities with the Majorana case, however, largely end here. Most strikingly, there is strong evidence that localized zero modes disappear entirely upon introducing *arbitrarily small* $f$![@Fendley] Such dramatic behavior defies intuition given that for $f< J$ the system resides in a gapped topological phase[@Fendley; @Motruk; @Bondesan] where one would naturally expect $f$ to yield only perturbative effects. Stable localized zero modes were instead found only in a ‘chiral’ deformation of the Hamiltonian wherein $J \rightarrow e^{i \phi} J$ with non-zero $\phi$ \[more precisely, Fendley constructed localized zero modes when $f \ll J |\sin (3\phi)|$\].
Understanding the stark differences from the Kitaev chain and diagnosing implications for quantum information applications seem particularly pressing given the growing literature devoted to realizing parafermion zero modes (see, e.g., Refs. ). Despite all this work, the curious state of affairs regarding the stability of zero modes in parafermion chains has remained largely unexplored. The purpose of this paper therefore is to explain the generic absence of localized zero modes in Eq. as well as their resurrection in the chiral case [@Fendley].
Based on various complementary analytical and numerical methods, our work paints the following picture: In the topological phase with $f\neq 0$ the ground states, as expected, remain three-fold degenerate up to corrections that vanish exponentially as one approaches the thermodynamic limit. Surprisingly, however, even the lowest-lying excited states that would otherwise be exactly degenerate at $f = 0$ exhibit a *power-law* splitting with system size for $f \neq 0$, implying the destruction of localized zero modes. We show that the onset of power-law splitting can be understood via domain-wall tunneling processes that simply have no analogue in the Kitaev chain. We further demonstrate that chirally deforming the Hamiltonian frustrates these domain-wall tunneling events, eventually restoring exponential splitting of the states (at least in part of the spectrum) consistent with zero-mode revival.
One noteworthy implication of our work is that the disappearance of localized zero modes should *not* be conflated with a demise of non-Abelian anyons. On the contrary, throughout the topological phase exhibited by Eq. the parafermion chain still allows one to demonstrate non-Abelian statistics since the all-important ground-state degeneracy persists modulo exponentially small corrections. For such cases it should be possible to define weaker zero mode operators that are localized and commute with a *projected* Hamiltonian.[^2] As another interesting corollary, we show that the transition between power-law and exponential splitting noted above can be associated with chirality-tuned critical behavior in the *excited states*, even though the ground state sector remains regular.
For completeness we note that parafermion braiding, while carrying some advantages over the Majorana case, remains non-universal.[@ClarkeParafendleyons; @LindnerParafendleyons; @Hastings] One can, nevertheless, leverage parafermionic systems to generate new two-dimensional phases that do permit universal topological quantum computation.[@Mong] Interestingly, similar physics can even appear in local bosonic two-dimensional systems.[@QiSlaveGenon].
To flesh out the above results, we begin in Sec. \[Model\] by describing basic properties of the parafermion chain model—in particular explaining a non-local mapping to ‘spins’ of the three-state Potts model—and introduce the criteria used for evaluating the existence of zero modes. Sections \[PerturbativeRegime\] and \[sec:nonperturb\] then explore the ground states and excited states of the Hamiltonian using perturbation theory, exact diagonalization of a truncated Hilbert space model, and density matrix renormalization group (DMRG) simulations. Finally, we conclude in Sec. \[Conclusions\] by highlighting additional implications and extensions of this study.
The Model and Zero-Mode Criterion {#Model}
=================================
The most general parafermion chain Hamiltonian studied in this paper is given by $$\begin{aligned}
H=-f\sum_{x = 1}^{L}\alpha_{2x-1}^\dagger\alpha_{2x}-Je^{i\phi}\sum_{x = 1}^{L-1} \alpha_{2x}^\dagger\alpha_{2x+1}+{{h.c.}},
\label{eq:Halpha}\end{aligned}$$ where again $\alpha_x$ satisfies the properties in Eq. . Note that in total the system consists of $2L$ parafermion sites (to define a sensible Hilbert space this number is necessarily even). Without loss of generality we will restrict the chiral phase $\phi$ appearing in the second term to the range $\phi \in [0,\pi/3]$, since symmetry relates Hamiltonians with $\phi \rightarrow \phi + 2\pi/3$ and $\phi \rightarrow -\phi$ to Eq. .
As noted above, in the limit $f = 0$ the existence of zero modes at the ends of the chain is obvious since then $[H, \alpha_1] = [H,\alpha_{2L}] = 0$ for any $J, \phi$. To appreciate the implications of these zero modes it is useful to define a ‘triality’ operator $$\hat{Q} = \prod_{x = 1}^L \alpha_{2x-1}^\dagger\alpha_{2x}$$ akin to the total fermion parity in the Kitaev chain. Since $\hat{Q}^3=1$, $\hat{Q}$ admits eigenvalues $1$, $\omega$, or $\omega^2$, where $$\begin{aligned}
\omega = e^{i2\pi/3}.\end{aligned}$$ For any choice of couplings $\hat{Q}$ commutes with the Hamiltonian. Crucially, however, the zero-mode operators $\alpha_{1}$ and $\alpha_{2L}$ do *not* commute with $\hat Q$—they cycle the triality by $\omega$. It follows that the entire spectrum can be grouped into triplets of energy eigenstates with trialities $Q = 1$, $\omega$, and $\omega^2$ that are exactly degenerate at $f = 0$. The spectrum of the Kitaev chain in the analogous limit consists of degenerate doublets with opposite fermion parity.
Our goal is to explore the fate of these localized zero-mode operators in the generic situation with $f \neq 0$. We now define the precise criteria used in evaluating whether or not exact edge zero modes exist. Finite-size effects at nonzero $f$ generically split the exact degeneracy between different triality states except with fine-tuning. Let $E_{a,Q}$ denote the system’s energies, where $Q$ labels the triality and $a = 1,2,3,\ldots$ indexes the levels such that $E_{1,Q} \leq E_{2,Q} \leq E_{3,Q} \cdots$. The existence of exponentially localized zero modes implies that $|E_{a,Q}-E_{a,Q'}|=\mathcal{O}\left(e^{-L/\xi}\right)$ holds for every $a,Q,Q'$ with some length scale $\xi>0$. An exceedingly useful corollary is that the existence of such modes may be categorically ruled out in large swaths of parameter space merely by demonstrating sub-exponential (e.g., power-law) splitting of a single triplet of energy levels $E_{a,Q = 1,\omega,\omega^2}$. Demonstrating the presence of exact zero modes, by contrast, poses a much more difficult problem, as doing so requires proving a global property of all energy levels. In this paper we will content ourselves with identifying regimes where zero modes are definitely absent.
One can obtain a great deal of intuition by exploiting an exact mapping between the parafermion chain Hamiltonian in Eq. and the chiral three-state Potts model. This mapping is analogous to that between the Kitaev chain and the transverse-field Ising model, and accordingly is implemented with a variation of the Jordan-Wigner transformation introduced by Fradkin and Kadanoff[@FradkinKadanoff]. In particular, writing $$\begin{aligned}
\alpha_{2x-1} &= \sigma_x\prod_{j<x}\tau_j,
& \alpha_{2x} &= \sigma_x\prod_{j\leq x}\tau_j.
\label{PottsMapping}\end{aligned}$$ decomposes the parafermions via strings of Potts model ‘spin’ operators that satisfy $$\begin{aligned}
\sigma_x^3&=\tau_x^3=1, &\sigma_x\tau_x&= \omega\tau_x\sigma_x.
\label{PottsAlgebra}\end{aligned}$$ All other commutators amongst $\sigma_x$ and $\tau_x$ vanish, a straightforward result of Eq. . Under this non-local change of variables the Hamiltonian becomes[^3] $$\begin{aligned}
H=-Je^{i \phi}\sum_{x = 1}^{L-1} \sigma_x^\dagger\sigma_{x+1}-f\sum_{x = 1}^L\tau_x+{{h.c.}},
\label{Hpotts}\end{aligned}$$ which defines a local bosonic model whose states are much easier to analyze than those in the parafermion chain. We stress, however, that both models are equivalent and exhibit precisely the same spectra. Equation can be obtained from the anisotropic limit of a well-known classical lattice model. With $\phi=0$ and $J > 0$ this is the ferromagnetic three-state Potts model. When $\phi\ne 0$, this is typically called the chiral clock or sometimes the chiral Potts model.
Using a basis of $\sigma$ eigenstates, denoted $|s = 0,1,2\rangle$, the Potts operators on a given site can be represented as $$\begin{aligned}
\sigma=\begin{pmatrix}
1 & 0 & 0 \\
0 & \omega & 0 \\
0 & 0 & \omega^2 \end{pmatrix} ,\quad
\tau=\begin{pmatrix}
0 & 1 & 0 \\
0 & 0 & 1 \\
1 & 0 & 0 \end{pmatrix} .\end{aligned}$$ We then have $$\begin{aligned}
\sigma\ket{s} = \omega^s\ket{s} ,
\quad \tau\ket{s} = \ket{s+1\bmod3} ,\end{aligned}$$ from which it follows that $\tau$ cycles the ‘spin’ measured by $\sigma$. Notice that $\sigma$ and $ \tau$ represent a straightforward generalization of anticommuting Pauli matrices in the Ising model. The triality defined earlier is simply the generator of a global $\mathbb{Z}_3$ symmetry exhibited by the Potts Hamiltonian, winding every spin: $$\begin{aligned}
\hat{Q} = \prod_x \tau_x \ .
\label{Qpotts}\end{aligned}$$
Hereafter we find it much more illuminating to work in the Potts model representation. Thus it is worth translating into Potts language the consequences of localized zero modes in the parafermion chain. For simplicity, consider $\phi = 0$ and suppose for the moment that $f = 0$—where parafermion zero modes definitely exist. In this case Eq. reduces to the ferromagnetic Potts model with a vanishing ‘transverse field’. The Hamiltonian has three broken-symmetry ground states that transform into one another under the action of $\hat{Q}$ in Eq. ; we label these by $|A\rangle = |00\cdots0\rangle$, $|B\rangle = |11\cdots1\rangle$, and $|C\rangle = |22\cdots2\rangle$. One can of course also define a basis of Schrödinger-cat-like ground states with definite triality via $$\begin{aligned}
\begin{split}
|Q = 1\rangle &= |A\rangle + |B\rangle + |C\rangle ,
\\ |Q = \omega\rangle &= |A\rangle + \omega |B\rangle + \omega^2|C\rangle ,
\\ |Q = \omega^2\rangle &= |A\rangle + \omega^2|B\rangle + \omega|C\rangle .
\label{Qgdstates}
\end{split}\end{aligned}$$ The zero-mode operators $\alpha_1$ and $\alpha_{2L}$ cycle the system amongst the degenerate triplet defined in Eqs. \[to see this recall Eq. \]. Similar conclusions hold for the excited states, which one can fruitfully view in terms of domain walls (i.e., kinks and anti-kinks) separating different ferromagnetically ordered regions. All such excited states may also be grouped into degenerate triplets of triality eigenstates that transform into one another under the action of the localized parafermion zero-mode operators. Analogous results apply to the chiral case with $\phi \neq 0$ despite the fact that chirality can nontrivially rearrange the spectrum (see the next section).
Restoring non-zero $f$ lifts the degeneracy among the ground-state and excited triplets via a transparent physical mechanism. Namely, the $f$ term in Eq. creates mobile domain walls that can tunnel the system between states related by a global $\mathbb{Z}_3$ transformation. Repeated action of the $f$ term can, for instance, send $|A\rangle \rightarrow |B\rangle$, $|B\rangle\rightarrow |C\rangle$, and $|C\rangle \rightarrow |A\rangle$, thereby splitting the three-fold ground-state degeneracy in the ferromagnetic case. The question we address in the following sections is thus: [*how*]{} do the degeneracies split?
Perturbative Regime {#PerturbativeRegime}
===================
With the insights of the previous section we now distill our earlier criterion for the existence of localized zero modes to the following question: Do domain-wall hopping processes preserve the degeneracy amongst *all* triplets of energy eigenstates, up to corrections that decay to zero exponentially with system size? If the answer is ‘no’ then localized zero modes in the parafermion chain are ruled out. In this section we address this question by studying both the splitting of ground states and of low-lying excited states in the limit $f/J \ll 1$. Section \[sec:nonperturb\] then explores the complementary regime where $f/J$ is of order one.
As described above the entire spectrum consists of triplets of exactly degenerate states when $f = 0$. With $f/J \ll 1$, degenerate perturbation theory allows one to quantify the effect of non-zero $f$ on these triplets. In what follows we discuss the ferromagnetic model with $\phi = 0$ from this perturbative perspective, then attack the chiral case with $\phi \neq 0$, and finally address the antiferromagnetic limit $\phi = \pi/3$. In both the ferromagnetic and antiferromagnetic limits we will show that localized zero modes certainly do not exist even for arbitrarily small but finite $f$, consistent with Ref. . Rigorous conclusions are harder to obtain for the chiral case, though in our analysis we definitively rule out zero modes over ranges of $\phi$ and $f$ in the vicinity of $\phi = 0$ and $\pi/3$; we argue that outside of these regimes chirality restores zero modes in the perturbative regime. Figure \[fig:smallFphases\] summarizes our results for this section.
![ Summary of results for the perturbative regime $f\ll J$. The two limits $\phi = 0$ and $\phi = \pi/3$ respectively correspond to the ferromagnetic and antiferromagnetic points of the chiral Potts model \[Eq. \]. Note that the critical values $\phi_{c1}$ and $\phi_{c2}$ depend on $f/J$. []{data-label="fig:smallFphases"}](phasesSmallF){width="67mm"}
Ferromagnetic limit, phi = 0 {#FMpotts}
----------------------------
{width="85.00000%"}
Figure \[fig:spectrum\] (left side) illustrates the $f = 0$ spectrum for the ferromagnetic case. Let us now discuss these levels in greater detail. There are three ground states $|A\rangle, |B\rangle$, and $|C\rangle$ corresponding to all spins uniformly polarized along one of three possible directions. As in the Ising model every excited state can be viewed in terms of domain walls between differently polarized regions, although here six flavors exist rather than two because of the larger ground-state degeneracy. The energy cost of a domain wall comes from the single link connecting different values of the spins, giving $$E_{\rm wall} \Big|_{f=0} = 2J\left[1-\cos\left(2\pi/3\right)\right] = 3J.$$ The lowest excited levels of Fig. \[fig:spectrum\] are single-domain-wall states. The six flavors are denoted as $|A|B\rangle,$ $|B|C\rangle$, $|C|A\rangle$, $|B|A\rangle, |C|B\rangle$, and $|A|C\rangle$, where for example $|A|B\rangle$ indicates the presence of a domain wall where spins on the left are in ground state $|A\rangle$ and on the right are in $|B\rangle$. Each flavor can sit at any of $L-1$ positions, so a total of $6(L-1)$ such states exist—all exactly degenerate at $f = 0$. There exists a larger set of $\mathcal{O}(L^2)$ degenerate two-domain-wall states corresponding to $|A|B|C\rangle, |A|B|A\rangle, |C|B|A\rangle$, etc., and so on up the spectrum. Crucially, the three ground states and each set of degenerate domain-wall states form subspaces that are energetically well-separated from one another by a gap $3J$.
Including infinitesimal $f$ enables domain-wall creation, annihilation, and hopping. The splitting of the levels within a given subspace can then be computed using standard degenerate perturbation theory methods. We are particularly interested in processes that take one $f = 0$ eigenstate to another related by a global $\mathbb{Z}_3$ transformation, e.g., $|A\rangle \rightarrow |B\rangle$ or $|A|B\rangle \rightarrow |B|C\rangle$. These are precisely the processes that produce finite-size splitting of the degeneracy encoded by any localized zero modes.
The splitting of the ground-state energies due to non-zero $f$ follows from standard arguments. These imply that mixing between the three ground states $|A\rangle, |B\rangle, |C\rangle$ is suppressed *exponentially* in system size. Namely, the ‘cheapest’ way to evolve from, say, $|A\rangle$ to $|B\rangle$ is to $(i)$ create an $|A|B\rangle$ domain wall at one of the end chain—leaving the system in an excited state, $(ii)$ tunnel the wall over to the opposite end, and $(iii)$ annihilate the domain wall to re-enter the ground-state manifold. One can visualize the process graphically via Fig. \[fig:diag2\]. Here the horizontal axis denotes the position along the chain, the vertical axis represents the perturbation step (which can also be interpreted as time), and the diagonal line indicates the domain-wall trajectory. Such a process necessitates exiting the ground-state subspace for a macroscopic number of steps, resulting in a ground-state splitting $$\Delta E_{\rm g.s.} \sim f\left(\frac{f}{3J}\right)^{L-1}.
\label{Egs}$$ In short, an energy barrier prevents efficient mixing of these states through local spin flips. Thus the *ground-state* degeneracy encoded by the exact zero modes of the $f = 0$ limit survives the introduction of non-zero $f$, up to exponentially small corrections in system size.
Qualitatively different behavior arises in the single-domain-wall sector. In fact here the energy barrier that suppresses ground-state mixing disappears altogether so that the dominant contribution to the splitting that we seek arises already at *first order* in degenerate perturbation theory. To see this consider the process illustrated in Fig. \[fig:oneWallDiag1\] which takes $|A|B\rangle \rightarrow |B|C\rangle$ without leaving the single-wall subspace. This process proceeds by first hopping the $|A|B\rangle$ domain wall all the way to the right end of the chain. If we were dealing with an Ising model, then the only way to remain in the original subspace would be to subsequently tunnel the domain wall leftward—which fails to accomplish the domain-wall change required. The Potts chain, however, offers an alternative: the rightmost spin can wind, converting $|A|B\rangle$ to $|A|C\rangle$. This newly formed $|A|C\rangle$ domain wall can then tunnel all the way to the chain’s left end where it can similarly convert to $|B|C\rangle$. Moving the domain wall back to its original location completes the process.
The splitting amongst triplets of single-domain-wall states resulting from such processes certainly depends sub-exponentially on system size. Indeed, the energy denominators responsible for the exponential dependence in Eq. are entirely absent since the system never leaves the degenerate single-wall subspace (again, all the action takes place at first order in perturbation theory). We can explore the splitting more quantitatively by examining the Hamiltonian projected onto the single-domain-wall subspace, which resembles a tight-binding model for six flavors of particles reflecting the domain-wall types; for details see Appendix \[ProjectedH\]. If one were to ignore the fact that these particles are domain walls and give them periodic boundary conditions, then the energy would be $$E(k) = 3J-2f\cos k,
\label{Ek}$$ where $k \in 2\pi \mathbb{Z}/L$ labels the momentum. These energy levels are trivially six-fold degenerate since the domain-wall flavor is conserved with periodic boundary conditions. With open boundaries, however, the system’s edges mediate backscattering processes that *do* convert domain-wall flavors into one another as in Fig. \[fig:diag2\]. Since the backscattering happens within a continuum of extended levels, power-law splitting generically arises for all triplets of single-wall states even with arbitrarily small non-zero $f$.
We can compute analytically the power-law splitting arising with open boundaries in the single-domain-wall projection. The key observation is that there is only one way for domain walls to change flavor at each end, e.g., $|A|B\rangle$ to $|A|C\rangle$ at the right boundary. As a consequence, we can ‘unfold’ the system into a *periodic* chain of size $6(L-1)$ with a single flavor. In this unfolded picture each flavor of the original model corresponds to a region of size $L-1$ so that a domain wall bouncing from an edge is replaced by a particle moving into a different region. Cycling through all six flavors corresponds to going around the periodic chain.
We can then obtain the spectrum using Fourier analysis; details of the calculation appear in Appendix \[ProjectedH\]. The solution exhibits an interesting form that can be anticipated from previously known results for the scaling limit of the three-state Potts model: the amplitude for changing flavor when a kink scatters off the boundary is of magnitude 1, while the amplitude for not changing (i.e., just bouncing back as is) vanishes [@Chim]. Indeed we find that the eigenstates take the form of simple right- or left-moving plane waves that propagate unreflected around the unfolded periodic chain. More quantitatively, the results are as follows. Labeling the energies of a single domain-wall triplet as $E_{a,Q = 1,\omega,\omega^2}$ for integer $a=1\dots L-1$, we define $${\rm Splitting}[a] \equiv \sqrt{\sum_{Q' < Q} (E_{a,Q}-E_{a,Q'})^2}.
\label{Splitting}$$ In the large-$L$ limit this becomes $${\rm Splitting}[a] \approx \frac{2\sqrt{2} \pi f}{3L}\sin\left(\frac{\pi a}{L}\right)\ .
\label{eq:splittEq}$$ Near the band edges a power-law of the form $1/L^2$ emerges; elsewhere a $1/L$ scaling takes over. We verified the splittings extracted above by numerically simulating the projected Hamiltonian with open ends; very large system sizes can be readily simulated since the truncated Hilbert-space dimension scales only linearly with $L$. Figure \[fig:oneWallED\] (red circles) illustrates our simulation results. The vertical axis denotes the natural log of the splitting of the three lowest-energy domain-wall eigenstates while the horizontal axis represents $L$. Clear $1/L^2$ splitting indeed appears as shown by the solid line.
![ Exact diagonalization results for the Hamiltonian projected into the single-domain-wall subspace with various values of $\tilde{\phi} = \sqrt{3} J\sin\phi/f$. The vertical axis denotes the natural log of the splitting for a given triplet, defined as in Eq. . All data correspond to the lowest excited triplet except for the diamonds, which represent a triplet at the center of the lowest single-domain-wall band. The $\tilde{\phi}=10$ data has been rescaled by a factor of $1/10$ for visual clarity. With $\tilde \phi = 0$, power-law splitting arises that fits extremely well to a $1/L^2$ curve (solid line through circles). Similarly, the $\tilde\phi = 0.01$ data corresponding to the band middle exhibit power-law splitting consistent with a $1/L$ scaling (dotted line through diamonds). Other finite-$\tilde\phi$ data sets exhibit exponential decay as shown by the fits (dashed and dotted lines through squares and triangles). []{data-label="fig:oneWallED"}](edOneWall){width="\columnwidth"}
The power-law behavior captured here immediately precludes the existence of exponentially localized zero modes in the ferromagnetic limit for *any* small but finite $f/J$. We stress that this result is quite robust as processes that involve neglected subspaces are higher-order in $f$ and in the perturbative regime cannot remove the power-law splitting arising at first order.
Chiral case, 0 < phi < pi/3 {#Chiralpotts}
---------------------------------
Having ruled out exact localized zero modes in the ferromagnetic case, we turn now to the chiral regime where $\phi$ lies between 0 and $\pi/3$ non-inclusive. It is instructive to first discuss the $f = 0$ spectrum illustrated schematically on the right side of Fig. \[fig:spectrum\]. For any $\phi$ in this range $|A\rangle, |B\rangle, |C\rangle$ remain the unique exact ground states. Chirality does, however, alter the excited states in an important way. More precisely, the $|A|B\rangle$, $|B|C\rangle$, and $|C|A\rangle$ domain walls now carry energy $E^+_{\rm wall}(\phi)$ that differs from the energy $E^-_{\rm wall}(\phi)$ carried by their mirror counterparts $|B|A\rangle$, $|C|B\rangle$, and $|A|C\rangle$. Explicitly, $$E^\pm_{\rm wall}(\phi) = 2J\left[\cos\phi-\cos\left(2\pi/3\pm\phi\right)\right].
\label{Echiral}$$ The single-domain-wall sector therefore splits into two degenerate subspaces, the two-wall sector splits into three ($|A|B|C\rangle$, $|C|B|A\rangle$ and $|A|B|A\rangle$ all yield different energy), etc. Note that Fig. \[fig:spectrum\] illustrates the levels only for small chiral phases $\phi$; larger $\phi$ re-orders the excited states as we discuss below.
Chirality imparts only minor quantitative effects on the ground-state splitting induced by non-zero $f$. Our arguments from the previous subsection indeed carry over straightforwardly since a finite gap $\Delta(\phi) = E^-_{\rm wall}(\phi)$ to the first excited state persists whenever $0<\phi<\pi/3$. Exponentially small splitting is thus again guaranteed over a range of $f$, the primary difference being that the denominator $3J$ in Eq. should be replaced by the gap $\Delta(\phi)$.
For the excited states, by contrast, chirality yields more dramatic consequences. This originates from the reduced degeneracy of the $f = 0$ domain-wall eigenstates relative to the ferromagnetic case, which tends to suppress the tunneling processes that previously led to power-law splitting.
It is simplest to examine small chiral phases $\phi$ where the system remains close to the ferromagnetic point; values of $\phi$ near the antiferromagnetic point $\pi/3$ will be discussed separately below. Following Sec. \[FMpotts\] we can capture the leading effects of $f$ on the low-lying excited states by projecting the Hamiltonian onto the single-domain-wall subspace. The effective Hamiltonian (see Appendix \[ProjectedH\]) again resembles a tight-binding model for six flavors of particles, half of which now experience different on-site energies resulting in two types of excitation branches. With periodic boundary conditions one obtains band energies $$E_\pm(k,\phi) = E^\pm_{\rm wall}(\phi)-2f\cos k
\label{bands}$$ that exhibit three-fold degeneracy for each momentum $k$. The upper branch $E^+(k,\phi)$ represents the bands formed by $\{|A|B\rangle, |B|C\rangle, |C|A\rangle\}$ states while the lower branch similarly corresponds to $\{|B|A\rangle, |C|B\rangle, |A|C\rangle\}$.
Our aim is to now understand how these branches mix at the ends of a system with open boundaries. As Fig. \[ChiralBands\] illustrates, there are three distinct regimes distinguished by the degree to which the bands overlap. Implications for zero modes depend sensitively on the ratio of $f$ to the energy difference $E^+_{\rm wall}(\phi)-E^-_{\rm wall}(\phi) = 2\sqrt{3}J\sin\phi$. It is thus often convenient to utilize the ratio $$\tilde \phi \equiv \sqrt{3}J\sin\phi/f
\label{phitilde}$$ when analyzing the spectrum.
In the limit $\tilde \phi \gg 1$, the bands exhibit no overlap as shown in Fig. \[ChiralBands\](a). The hybridization between the upper and lower branches at the edges of the chain thus can be treated perturbatively. If we revisit the process sketched in Fig. \[fig:oneWallDiag1\], a clear qualitative difference from the ferromagnetic case appears. Now, when the $|A|B\rangle$ domain wall (originating from the upper branch) tunnels to the right end of the chain, conversion to $|A|C\rangle$ necessitates moving to the energetically well-separated lower band. Returning the $|A|C\rangle$ domain wall to the left end—where it can transition back to a $|B|C\rangle$ wall in the upper branch—requires the system to remain in the lower-energy space for a macroscopic number of steps. The key physics is that chirality precludes resonantly switching domain-wall flavors at the system’s boundaries. Thus such processes produce exponential rather than power-law splitting among all triplets of single-wall eigenstates. Numerical simulations of the projected Hamiltonian confirm this picture. The yellow diamonds in Fig. \[fig:oneWallED\] illustrate the natural log of the splitting for the lowest three single-domain states versus system size assuming $\tilde{\phi} = 10$; the splitting is well-captured by an exponential.
The obvious indications for the absence of localized zero modes that we uncovered in the ferromagnetic limit thus disappear upon introducing a small chiral phase $\phi$ in the $\tilde \phi \gg 1$ regime. This finding is in harmony with the construction in Ref. of edge zero modes at $\phi\ne 0$ and with $f/J$ sufficiently small. Our results here do not prove, however, that zero modes persist to finite $f/J$, as sub-exponential splitting could arise from various other sources including tunneling within multi-domain-wall subspaces and higher-order processes in perturbation theory.
Further observations do nevertheless suggest that localized zero modes indeed persist. First, we have studied numerically the effective Hamiltonian projected onto the two-domain-wall subspace and again found only exponential splitting when $\tilde \phi \gg 1$. Second, it is possible to address higher-order processes rather efficiently by developing a set of Feynman-diagram-like rules governing domain wall dynamics. Appendix \[Diagrams\] describes the methodology. There we briefly sketch a calculation indicating that through second order in $f/J$ the splitting amongst single-wall states remains exponential. This is a result of the cancellation of contributions from various diagrams related by symmetries of the allowed tunneling processes. Though we have not proven that the cancellation occurs at all orders, the symmetries by which it works at low orders suggest that this is the case.
Lowering $\tilde \phi$ brings the upper and lower branches closer to one another in energy until they eventually begin to overlap at a critical $\tilde \phi_{c1}$. Indirect gapless inter-band excitations are then permitted as shown in Fig. \[ChiralBands\](b). The critical value follows from the condition that $E_+(k = 0,\phi) = E_-(k = \pi,\phi)$, yielding $\tilde \phi_{c1} = 2$. This value corresponds to an $f$-dependent critical chiral phase $$\phi_{c1} = \sin^{-1}\left(\frac{2f}{\sqrt{3}J}\right).
\label{phic1}$$ With open boundaries the chain’s edges provide the momentum transfer needed here to *resonantly* scatter $k = 0$ domain walls in the upper branch into $k = \pi$ walls in the lower branch. Processes like that of Fig. \[fig:oneWallDiag1\] consequently yield power-law splitting among the triplets at the bottom of the upper band and top of the lower band.
With $\tilde \phi < \tilde\phi_{c1}$ the bands overlap in a finite energy window \[see Fig. \[ChiralBands\](c)\] within which all triplets similarly admit power-law splitting. This picture is supported by our numerics in Fig. \[fig:oneWallED\] for $\tilde \phi = 0.01$, which reveal splitting that is exponential for the lowest excited triplet (squares) yet power-law for triplets at the middle of the lower branch (diamonds). In the perturbative regime, we can therefore conclusively rule out localized zero modes not just in the ferromagnetic limit, but in fact along the finite interval $0 \leq \phi \leq \phi_{c1}$.
Mapping the problem to an ‘unfolded’ system as described in the previous subsection and Appendix \[ProjectedH\] provides an enlightening perspective. The projected single-wall Hamiltonian for the open chain again effectively describes a particle hopping in a periodic system of length $6(L-1)$. Due to chirality, however, the unfolded Hamiltonian now includes a square-well potential that alternates between $+\sqrt{3}J \sin\phi$ and $-\sqrt{3}J \sin\phi$ every $L-1$ sites. For simplicity let us consider small chiralities and focus on states near the bottom of the upper and lower branches so that the particle’s kinetic energy can be approximated with a free-particle dispersion $\propto k^2$. (Analogous arguments follow for states near the top of the bands.) Given sufficient kinetic energy to overcome the square-well barrier, the particle will efficiently sample the entire extended periodic chain. Translating back to Potts model language, this means that domain walls can resonantly convert between different flavors at the edges of the system—resulting in power-law splitting at such energies. With insufficient kinetic energy the particle will remain predominantly within the square-well minima, decaying evanescently into the regions corresponding to square-well maxima. The evanescent decay translates into inefficient domain wall conversion and a splitting of triplets that decays exponentially with system size.
Suppose next that the chiral phase $\phi$ is slightly below $\pi/3$ so that the system resides close to the antiferromagnetic point. Because $E^-_{\rm wall}(\phi) \ll E^+_{\rm wall}(\phi)$ in this regime, the ordering of domain-wall subspaces changes dramatically compared to the cases examined above. At $f = 0$ the lowest-energy subspace consists of $|B|A\rangle, |C|B\rangle, |A|C\rangle$ states while the second-lowest subspace arises from the two-domain wall states $|C|B|A\rangle, |A|C|B\rangle, |B|A|C\rangle$. By contrast the states $|A|B\rangle$, $|A|C\rangle, |B|A\rangle$ that we often invoked earlier lie much higher up in the spectrum so that conversion processes sketched in Fig. \[fig:oneWallDiag1\] are now irrelevant.
We can, however, identity a new series of domain-wall processes that yield power-law splitting over a range of $\phi$ values. As an example, one can convert $|B|A\rangle$ to an $|A|C\rangle$ domain wall related by a global $\mathbb{Z}_3$ transformation by utilizing the lowest-lying two-wall subspace. This is accomplished by moving the original domain wall to the system’s right end and then inserting a second domain wall, i.e., $|B|A\rangle \rightarrow |B|A|C\rangle$. Upon hopping both domain walls to the other end one can remove the leftmost wall, sending $|B|A|C\rangle \rightarrow |A|C\rangle$. The remaining domain wall, having switched flavors, finally returns to its original location. Such processes generically split the degeneracy among single-wall triplets in the lowest excited subspace.
To quantify the induced splitting it is once again useful to project onto the relevant subspaces. For periodic boundary conditions on the resulting effective Hamiltonian, the single-wall states broaden into a band with energies $E_-(k,\phi)$ given in Eq. . The two-wall states are similar, although the energies for the latter are nontrivial since domain walls interact. Introducing open boundaries enables transitions between these subspaces at the system’s edges, which can supply any momentum necessary. Power-law splitting arises for triplets in the lower band *if* they overlap in energy with the upper band. The critical $\phi_{c2}$ at which power-law behavior first appears follows from $E_-(k = \pi,\phi) \approx E_-(k = 0,\phi) + E^-_{\rm wall}(\phi)$, where the right side approximates the energy needed to insert an extra domain wall into a zero-momentum single-wall state. (Note that this estimate neglects domain wall interactions; this is reasonable since they are dilute.) We then obtain $$\phi_{c2} \approx \pi/3-\sin^{-1}\left(\frac{2 f}{\sqrt{3}J }\right).
\label{phic2}$$ Interestingly, the $f$ dependence is remarkably similar to Eq. despite the rather different processes involved. In the interval $\phi_{c2} \leq \phi < \pi/3$ at least one excited triplet exhibits power-law splitting in the perturbative regime, ruling out localized zero modes here too.
Scaling behavior {#Scaling}
----------------
In the previous subsections we gained quite a bit of mileage by studying an effective Hamiltonian projected onto the single-domain-wall subspace. This truncated model is expected to capture the dominant contribution to the splitting amongst triplets of low-lying excited states in the limit $f, J\sin\phi \ll J$. Through analytical arguments corroborated by numerics we concluded that chirality induces a sharp change in the level spacing among triplets of excited states.
More precisely, at $\phi = 0$ we showed that all single-domain-wall triplets exhibit power-law splitting with system size for arbitrary non-zero $f/J$. Introducing small non-zero $\phi$ immediately restores exponential splitting for a range of the lowest- and highest-energy triplets but preserves power-law splitting for all intermediate single-wall states. Further increasing $\phi$ shrinks the window of power-law-split states until exponential splitting fully returns beyond a critical value $\phi_{c1}$ given in Eq. —presumably restoring exact localized zero modes. It is tempting to view this behavior as indicating chirality-tuned critical behavior in the excited-state spectrum, despite the fact that no singular behavior arises in the ground-state sector. We now provide evidence for such a scenario by showing that our numerical results exhibit data collapse consistent with a simple scaling ansatz.
Provided $f, J\sin\phi \ll J$, the single-domain-wall splitting $\Delta E_\textrm{1-wall}$ of a given triplet should depend only on $\tilde \phi$ from Eq. and the effective system size $\tilde L \equiv L-1$ for domain walls. Let us consider a particular triplet that transitions from exponential to power-law splitting at some (state-dependent) critical value $\tilde \phi_c$. We stress that $\tilde \phi_c$ is distinct from $\tilde \phi_{c1,2}$ discussed in the previous subsection; the former applies to just one particular multiplet. Under rescaling of the length by $b$ we postulate the following scaling form, $$\Delta E_\textrm{1-wall}(b^{1/\nu} \delta \tilde \phi,\tilde L/b) = b^{\alpha} \Delta E_\textrm{1-wall}(\delta\tilde \phi,\tilde L)$$ where $\delta \tilde \phi$ denotes the deviation from $\tilde \phi_c$ and $\nu,\alpha$ are critical exponents to be determined. Setting $b = \tilde L$ then allows us to write $$\Delta E_\textrm{1-wall}(\delta \tilde \phi,\tilde L) = \tilde L^{-\alpha}\mathcal{E}(\tilde L^{1/\nu} \delta\tilde \phi)
\label{Collapse}$$ for some function $\mathcal{E}$.
As a concrete example consider the lowest-lying excited triplet, for which $\tilde \phi_c = 0$. Our results for the ferromagnetic case imply that in the perturbative regime $\alpha = 2$ while $\mathcal{E}(0)$ is some non-zero constant.[^4] The remaining exponent $\nu$ follows from our simulations. In the inset of Fig. \[fig:oneWallED3\] we show the natural log of the splitting for the lowest single-wall triplet versus system size for a variety of $\tilde \phi$ values, illustrating the change from power-law to exponential behavior. The same data appear in the main plot, but with the splitting scaled by $\tilde L^2$ and with the horizontal axis representing $\tilde L^2\tilde \phi$. Very clean data collapse is evident, confirming that our scaling ansatz holds with an exponent $\nu = 1/2$. Figure \[fig:oneWallED2\] displays analogous scaling results for a triplet at the middle of the spectrum and with $\tilde\phi_c$ close to $\phi_{c1}$ \[recall Fig. \[ChiralBands\](b)\]. Different exponents appear here, with $\alpha = 1$ and $\nu \approx 0.31$, but we again see evidence of critical behavior in the excited states.
Antiferromagnetic limit, phi = pi/3 {#AFMpotts}
-----------------------------------
Finally, we consider the antiferromagnetic limit where $\phi = \pi/3$. One can anticipate that this case is rather special from the $f = 0 $ excitation energy quoted in Eq. ; in particular, with $\phi = \pi/3$ half of the domain-wall flavors cost no energy, resulting in a macroscopic ground-state degeneracy in the $f = 0$ limit. Here localized zero modes disappear for *any* non-zero $f$ in the perturbative regime, just as in the ferromagnetic case (and for similar reasons). Indeed, the construction of edge zero modes in Ref. works only for $f<J\sin(3\phi)$, and so fails at the antiferromagnetic point.
To demonstrate this result directly we need only examine the $f = 0$ ground-state manifold perturbed by infinitesimal $f$. It is convenient here to perform a gauge transformation $\sigma_x\rightarrow \omega^x \sigma_x$, which effectively transforms the Potts Hamiltonian to one with $\phi = \pi$. In the transformed model the $J$ term is minimized by any state for which no two adjacent spins align with one another (hence the ‘antiferromagnetic’ nomenclature). Thus the $f = 0$ limit supports $3\times2^{L-1}$ ground states. Consider, for instance, a ground state of the form $|0101\cdots1010\rangle$. As usual we are interested in tunneling processes that connect states related by a global $\mathbb{Z}_3$ transformation, i.e., $|1212\cdots2121\rangle$, since such events split the degeneracy encoded by the exact zero modes present at $f = 0$. One possible pathway is for repeated action of the $\tau_x$ operators to first wind all of the even spins, sending $|0101\cdots1010\rangle \rightarrow |0202\cdots2020\rangle$, followed by all odd spins, sending $ |0202\cdots2020\rangle \rightarrow |1212\cdots2121\rangle$. The system remains in the ground-state manifold throughout such processes, so that a splitting amongst ground-state triplets arises at first order in degenerate perturbation theory.
This splitting is thus expected to scale as a power-law with system size—similar to what we observed previously for the ferromagnetic case—precluding localized zero modes even for arbitrarily small non-zero $f$ as claimed. Interference effects that might alter this conclusion are absent since all non-zero matrix elements in the ground-state manifold are unity. Higher-order processes that require exiting the ground-state manifold are also negligible in the limit considered here.
There is a complementary way of arguing for power-law splitting in the antiferromagnetic case. In the perturbative regime one can obtain an effective low-energy Hamiltonian for the chain by projecting the operator $\tau_i+\tau_i^\dagger$ appearing in the $f$ term onto the $f = 0$ ground-state manifold. (The $J$ term of course projects trivially.) To do so it is useful to define bond operators that measure whether the spin at a given site is wound ‘clockwise’ or ‘counterclockwise’ relative to its neighbor. More specifically, we can introduce Pauli matrices $\eta^{z}_{i+1/2}$ such that $P \sigma_i^\dagger \sigma_{i+1}P = \omega^{\eta^z_{i+1/2}}$ with $P$ the ground-state projector. As an example $\eta^z_{i+1/2} = +1$ for the spin pair $|0_i 1_{i+1}\rangle$ while $\eta^z_{i+1/2} = -1$ for the mirrored configuration $|1_i 0_{i+1}\rangle$. With this notation one can see that $\tau_i + \tau_i^\dagger$ flips both $\eta^z_{i-1/2}$ and $\eta^z_{i+1/2}$ if they have opposite signs, and otherwise projects trivially. More formally, we obtain $$\begin{aligned}
\begin{split}
P(\tau_i + \tau_i^\dagger)P &= \frac{1}{2}\eta_{i-1/2}^x\eta_{i+1/2}^x\left(1-\eta_{i-1/2}^z\eta_{i+1/2}^z\right)
\\ &= \frac{1}{2}\left(\eta^x_{i-1/2}\eta^x_{i+1/2}+\eta^y_{i-1/2}\eta^y_{i+1/2}\right).
\end{split}\end{aligned}$$ Thus the bulk maps onto an XY model[^5]—which is gapless, making the power-law splitting argued above extremely natural. The vanishing bulk excitation gap strongly suggests the generic absence of localized zero modes in the antiferromagnetic limit.
Non-Perturbative Regime {#sec:nonperturb}
=======================
Both the analytical arguments and numerical diagonalization exploited in Sec. \[PerturbativeRegime\] break down when $f/J$ becomes of order unity. Here we complement our earlier perturbative analysis using extensive DMRG simulations. To understand the nature of the spectrum in the non-perturbative regime, we computed the splitting among the ground states and first-excited triplet for a wide range of $\phi$ and $f/J < 1$ over which DMRG exhibits good convergence. The simulations turn out to perform particularly well at larger $f/J$.[@White-1992; @Schollwoeck:Review05; @McCulloch-2007] System sizes were taken to be at least $L = 10$ to minimize finite-size effects but sufficiently small to avoid the splitting falling below machine precision; the maximum $L$ considered varies from $19$ to $34$ depending on the Hamiltonian parameters. More details of our simulations are described in Appendix \[app:DMRG\].
To quantify the splittings obtained from the DMRG, our data were fit to an exponential form $\propto e^{-\kappa L}$. The color scales in Figs. \[fig:phases1\] and \[fig:phases2\] show the optimal $\kappa$ value versus $f/J$ (radius) and $\phi$ (angle) for the ground-state and first excited splittings, respectively. The ground states exhibit an exponential splitting characterized by a non-zero $\kappa$ throughout the regime of convergence, which notably does not include the antiferromagnetic limit because of its gaplessness. As in the perturbative regime, however, richer physics arises for the first excited triplet. In the ferromagnetic limit we obtain $\kappa = 0$, indeed consistent with the power-law splitting found in the perturbative regime. For the chiral cases, by contrast, $\kappa$ takes on finite values indicating exponential splitting with system size, also in agreement with our perturbative analysis.
We showed earlier that power-law splitting for excited states survives over a *finite* range of chiral phases $\phi$. However, accessing these levels within DMRG rapidly becomes prohibitive as one increases the energy. Thus the finite-$\phi$ transitions are difficult to capture numerically. Instead we crudely estimate the global zero-mode stability regions by naively extrapolating our results for the critical $\phi$ values obtained for the perturbative regime in Sec. \[Chiralpotts\]. Figure \[fig:phases3\] displays the results of our extrapolation. The shaded regions indicate the parameter space over which zero modes are expected to survive based on our perturbative criterion. While of course this scheme is not quantitatively reliable, we do expect to capture the qualitative trends. \[One feature that is *not* expected to be robust is the ‘accidental’ symmetry of Fig. \[fig:phases3\] with respect to sending $\phi \rightarrow \pi/6-\phi$, as is clear from the numerics in Fig. \[fig:phases2\]. This pathological property should be absent even in the perturbative regime upon including interactions between domain walls, which were neglected in our estimate of $\phi_{c2}$.\]
It is interesting to discuss these findings in relation to Ref. , where using an iterative method localized zero modes were constructed explicitly in the limit where the control parameter $$r \equiv \frac{f}{J\sin\left(3\phi\right)}$$ is much less than one. In this case corrections to the $f = 0$ zero modes could be arranged in a series that clearly decays exponentially into the bulk of the chain. Remarkably, our extrapolation in Fig. \[fig:phases3\] follows the lobe-like form of the control parameter $r$—strongly suggesting that we have correctly identified the essential physics that determines the robustness of exact zero modes in parafermion chains.
Conclusions {#Conclusions}
===========
In this paper we employed a variety of techniques to diagnose the puzzling stability of local parafermion zero modes—which differs markedly from the Majorana case—identified by Ref. . Viewing the physics from the mathematically equivalent lens of the chiral three-state Potts model proved particularly illuminating. In Potts language physically intuitive domain-wall hopping/conversion processes produce power-law splitting among low-lying excited states in the ferromagnetic and antiferromagnetic limits of the model, ensuring the generic demise of localized parafermion zero modes in both cases. Chirally deforming the Hamiltonian tends to suppress these processes, and for sufficiently large deviations from the ferromagnetic/antiferromagnetic limit restores exponential splitting of all low-lying excited states that we examined. We speculate that the restoration of exponential splitting coincides with the emergence of localized zero modes; although a general proof remains unavailable such a scenario gels nicely with the results from Ref. .
We also showed that the transition from power-law to exponential behavior reflects a subtle type of chirality-tuned quantum criticality that, interestingly, emerges only at energies above the ground states. Indeed, except in the antiferromagnetic limit the ground states remain degenerate up to exponentially small corrections in all cases and do not exhibit any singular behavior as a function of chirality.
Although we focused on $\mathbb{Z}_3$ parafermion chains for simplicity, many of our results extend straightforwardly to $\mathbb{Z}_{N>3}$ systems (where similar stability issues arise[@Fendley]). For instance, it is clear that here too domain-wall tunneling events generically preclude exponentially localized modes in the ferromagnetic and antiferromagnetic limits. Our findings also naturally explain the comparative robustness of localized Majorana zero modes in the Kitaev chain, since the relevant domain-wall processes that we invoked for the Potts model have absolutely no analogue in that context.
Since we motivated this work from the vantage point of topological quantum computation, it is worth re-emphasizing that the existence of localized zero modes as defined here is most certainly overkill for this application. Harnessing non-Abelian statistics in parafermion chains (and related systems) merely requires degenerate *ground states* of a topological phase to within exponential accuracy; this weaker condition appears much more broadly as noted above, even when exact zero mode operators are definitely absent. Systems supporting exact zero modes do, nevertheless provide the appealing possibility of performing topological quantum computation *at finite energy density*—a possibility first proposed in the framework of many-body localization[@HuseMBL] (see also Ref. ). We suggest that parafermion chains offer particularly interesting platforms to explore in this regard. Apart from chirality, localization via quenched disorder should provide another means of suppressing the domain-wall processes that produced power-law splitting in the ferromagnetic limit. It would be quite interesting to perform large-scale exact-diagonalization studies to explore this scenario further in future work.
Path cancellations at higher orders in perturbation theory {#Diagrams}
==========================================================
Section \[PerturbativeRegime\] performed a perturbative analysis that incorporated the lowest-order processes that produced a splitting among eigenstates with different trialities. For excited states it is possible to capture very similar transformations to that of Fig. \[fig:oneWallDiag1\] in far fewer perturbation steps if one considers terms higher order in $f/J$. One thus might worry that such processes, while parametrically smaller in $f/J$, nevertheless produce the dominant scaling with system size. In the chiral case with $\phi_{c1}<\phi<\phi_{c2}$, for instance, if power-law splitting were to appear at higher orders then that would immediately imply a further reduction in stability window for localized zero modes. However, our DMRG results—which are non-perturbative and hence include *all* orders—suggest that this is not the case since our simulations captured exponential splitting among the lowest excited triplet for any non-zero chirality. In this appendix we provide more support for this conclusion by examining higher-order perturbative computations.
While cumbersome, higher-order calculations can be greatly facilitated by using of a diagrammatic representation of domain-wall events similar to Fig. \[fig:oneWallDiag1\]. For this purpose it is useful to attach an upward-pointing arrow to lines indicating domain walls of the type $|A|B\rangle$, $|B|C\rangle$, and $|C|A\rangle$ and a downward arrow for the other three flavors. The following vertex rules then arise:
1. The boundary of the system acts as a source and sink for domain walls. This is the one-line vertex.
2. Two lines may be created or destroyed at a point if and only if they have opposite arrow directions. This is the two-line vertex.
3. A three-line vertex may exist if and only if all of the arrows are incoming or outgoing.
4. In a given perturbation step each line may either remain fixed or move one spatial unit.
Since all matrix elements in the perturbation theory are either 0 or 1 the assignment of weights to diagrams conforming to these rules is straightforward. Any step that moves the system out of the original subspace gets penalized by a factor of $\pm f/|\delta E|$ where $|\delta E|$ is the magnitude of the energy change incurred. The $+$ sign arises if the system moves to a higher-energy state, while the $-$ sign arises if a lower-energy state results.
Consider first the diagram in Fig. \[fig:cancel1\] depicting a second-order process wherein $|A|B\rangle \rightarrow |B\rangle \rightarrow |B|C\rangle$. By themselves, events like this yield power-law splitting $-c f^2/[L^2 E^+_{\textrm{1-wall}}(\phi)]$ with $c > 0$, *even in the chiral case*. There is, however, a compensating process shown in Fig. \[fig:cancel2\] where $|A|B\rangle \rightarrow |A|B|C\rangle \rightarrow |B|C\rangle$. Since here the system enters a higher-energy sector these events produce the exact opposite contribution $+c f^2/[L^2 E^+_{\textrm{1-wall}}(\phi)]$. The existence of such cancellations is the main message of this appendix. Preliminary calculations at third order point to a similar outcome, and we expect that they are a generic feature of higher-order perturbation theory in certain regions of parameter space. Again, this conclusion is supported by our DMRG results at intermediate $f/J$.
Projection domain-wall Hamiltonians {#ProjectedH}
===================================
In the limit $f,J\sin\phi \ll J$ one can to a good approximation neglect tunneling processes that mix sectors with different numbers of domain walls. Formally this is achieved by projecting $H$ in Eq. onto a subspace with fixed domain wall number. The procedure is generally straightforward though some care is necessary at the boundaries. In the one-wall sector the resulting effective Hamiltonian admits a particularly clean block form: $$\begin{aligned}
H_{\textrm{1-wall}} =\begin{pmatrix}
(V-Q) & -f I\\
-f I & -Q & -f I\\
& -f I & -Q & \ddots\\
& & \ddots & \ddots & -f I\\
& & & -f I & -Q & -f I\\
& & & & -f I & (W-Q)
\end{pmatrix},\end{aligned}$$ with $I$ the $6\times6$ identity matrix, $$\begin{aligned}
V= -f \begin{pmatrix}
0 & 0 & 0 & 0 & 0 & 1\\
0 & 0 & 1 & 0 & 0 & 0\\
0 & 1 & 0 & 0 & 0 & 0\\
0 & 0 & 0 & 0 & 1 & 0\\
0 & 0 & 0 & 1 & 0 & 0\\
1 & 0 & 0 & 0 & 0 & 0
\end{pmatrix} , \quad
W= -f \begin{pmatrix}
0 & 1 & 0 & 0 & 0 & 0\\
1 & 0 & 0 & 0 & 0 & 0\\
0 & 0 & 0 & 1 & 0 & 0\\
0 & 0 & 1 & 0 & 0 & 0\\
0 & 0 & 0 & 0 & 0 & 1\\
0 & 0 & 0 & 0 & 1 & 0
\end{pmatrix} , \quad\textrm{and}\quad
Q=\sqrt{3}J\sin\phi\begin{pmatrix}
1 & 0 & 0 & 0 & 0 & 0\\
0 & -1 & 0 & 0 & 0 & 0\\
0 & 0 & 1 & 0 & 0 & 0\\
0 & 0 & 0 & -1 & 0 & 0\\
0 & 0 & 0 & 0 & 1 & 0\\
0 & 0 & 0 & 0 & 0 & -1
\end{pmatrix}.\end{aligned}$$
As noted in Sec. \[FMpotts\], because of the form of the boundary terms described by $V$ and $W$ it is possible to recast $H_{\textrm{1-wall}}$ in terms an enlarged periodic chain of size $6(L-1)$. In this mapping the domain wall becomes a particle living on sites labeled by position $x$, with the domain-wall flavor corresponding to $j \equiv {\rm ceiling}\left(\frac{x}{L-1}\right)$. We will take $j = 1,3,5$ to represent $|A|B\rangle$, $|B|C\rangle$, and $|C|A\rangle$, and $j = 2,4,6$ to represent $|A|C\rangle$, $|B|A\rangle$, and $|C|B\rangle$, respectively. Let us now specialize to the ferromagnetic limit, $\phi = 0$. Denoting the position modulo $L-1$ by $y$, it is straightforward to verify that the eigenstates of the projected Hamiltonian are given by $$|\psi_k\rangle = \sum_{j=1}^6 \sum_{y=1}^{L-1} e^{ik\left((L-1)j+y\right)}|j,y\rangle,
\label{eq:eigs}$$ with eigenvalues $$E(k)=-2f\cos(k).
\label{Ek2}$$ These energies are the same as those in Eq. up to a constant; however, because of the enlarged system size in this description the allowed momenta are now $k\in \frac{2\pi}{6(L-1)}\mathbb{Z}$. Note that the wavefunctions specified in Eq. are also triality eigenstates with eigenvalue $$Q = e^{2ik(L-1)}.$$ \[The action of $\hat{Q}$ simply maps $j\rightarrow(j+2) \bmod 6$.\]
For extracting the splittings of interest, it is convenient to label the energies and momenta according to both the triplet $a$ to which they belong and their triality $Q$, i.e., as $E_{a,Q}$ and $k_{a,Q}$. We need only specify that $k_{a,Q=\omega}=-k_{a,Q=\omega^2}>0$ and require that $\left|k_{a,Q=\omega}-k_{a,Q=1}\right|=\frac{\pi}{3(L-1)}$. Using Eqs. and , we then find $$\begin{aligned}
{\rm Splitting}[a] = 2\sqrt{2}f\left|\cos(k_{a,Q=1})-\cos(k_{a,Q=\omega})\right|,\end{aligned}$$ which in the large-$L$ limit converges to Eq. .
For our numerical simulations of the truncated single-wall model, we explicitly split the spectrum into the three triality sectors by simultaneously diagonalizing the triality operator and the effective Hamiltonian. In practice this may be done by adding the two operators together with random coefficients, determining the diagonalizing unitary operator for the combination, and then using it to diagonalize the two operators individually. For almost all choices of random numbers this diagonalizes both operators. Pathological cases may be handled separately, or by re-running the calculation.
To handle the two-wall sector, a Python+Cython+NumPy code was written to explicitly enumerate all states of interest and compute the relevant matrix elements of $H$ between them. In either sector the effective Hamiltonians can be diagonalized numerically via standard routines to obtain splittings for any triplet of single-domain-wall states.
DMRG methods {#app:DMRG}
============
All DMRG computations in this paper utilized the Developer Branch of ITensor (<http://itensor.org>) commit 475352f76c6209db865ea4405cb86f665f40fae5. A control file, a Hamiltonian file, and a model file were created based on existing ITensor code and with the assistance of Miles Stoudenmire, one of the authors of ITensor. These files extended ITensor to perform calculations on the $\mathbb{Z}_3$ Potts model, which is not a native function of the code. The Eigen C++ library version 3.0 was also used in the main control file for diagonalizing the states that ITensor produced. This generates excited states with less ground-state overlap than DMRG alone gives.
The results from ITensor were verified for small system sizes with the open-source Quantum Chains package (<https://github.com/adamjermyn/QuantumChains/>) written by the first author. Any of the code used in this paper is available upon request from the first author.
10000
[^1]: To distinguish from parafermions in conformal field theory, which are related but distinct, these zero modes are sometimes referred to as ‘parafendleyons’.
[^2]: For an explicit construction see A. Alexandradinata, C. Fang, N. Regnault, and B. A. Bernevig, unpublished.
[^3]: In writing the Potts representation of the parafermion chain, we dropped a factor of $e^{-i2\pi/3}$ in the $J$ term; this factor can always be absorbed into the chiral phase $\phi$.
[^4]: Of course in the extreme limit $f=0$ we have $\mathcal{E}(0)=0$. Note also that the scaling forms provided here assume $f/J \ll 1$. For larger transverse fields mixing between other sectors matters, and $\Delta E_\textrm{1-wall}$ hence becomes a function of three parameters—$\tilde \phi, L$, and $f/J$.
[^5]: We are grateful to David Clarke for pointing out to us this XY model mapping.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'A theorem of Lawson and Simons states that the only stable minimal submanifolds in $\mathbb{CP}^{n}$ are complex submanifolds. We generalize their result to the cases of $\mathbb{HP}^{n}$ and $\mathbb{OP}^{2}$. Our approach gives a unified viewpoint towards conformal and projective geometries.'
author:
- 'Siu-Cheong Lau and Naichung Conan Leung'
bibliography:
- 'geometry.bib'
title: Conformal geometry and special holonomy
---
Introduction
============
Riemannian holonomy group $hol\left(M,g\right)$ measures the richness of algebraic structure on a Riemannian manifold[^1]. For a generic metric, the holonomy group equals $SO\left( m\right) $ with $m=\dim_{\mathbb{R}}M$. Manifolds with special holonomy include Kähler manifolds with $hol\left(
M,g\right) =U\left( n\right) $ and Calabi-Yau manifolds with $hol\left( M,g\right) =SU\left( n\right) $ where $m=2n$. They play very important roles in geometry and mathematical physics such as string theory and M-theory. Riemannian holonomy groups were completely classified by Berger [@Berger_holonomy] and all these geometries have been given a unified description in terms of real, complex, quaternionic and octonionic structures (that is, normed division algebras) and orientability in [@boss_Huang] for symmetric spaces and [@boss_div_alg] for non-symmetric ones.
Another important branch in Riemannian geometry is the conformal geometry where one allows the Riemannian metric to be scaled by a conformal factor, i.e. $g\sim e^{u}g$ for any function $u$. In this article, we explain how one integrates conformal geometry with real, complex, quaternionic and octonionic geometries. In particular we give a uniform proof to the following theorem on rigidity of calibrated cycles in projective spaces, which is a generalization of the results of Lawson and Simons from conformal and complex geometries to quaternionic and octonionic geometries. After we have discovered this, we were informed that this result has been proved earlier by [@Ohnita_stable]. We hope that our approach from Jordan algebra provides a unified viewpoint on all these seemingly different kinds of geometries.
> **Main Theorem:** In $\mathbb{AP}^{n}$, where $\mathbb{A\in
> }\left\{ \mathbb{R},\mathbb{C},\mathbb{H},\mathbb{O},\mathbb{R}^{m}\right\}
> $, any stable minimal submanifold $S$ (or more generally rectifiable current) must be complex, by which we means $T_{x}S$ is invariant under all the linear complex structures at $x$ for almost every $x\in S$.
There is an $\mathbb{S}^{2}$-family of linear complex structures at every point of $\mathbb{H}\mathbb{P}^{n}$, and also an $\mathbb{S}^{6}$-family of linear complex structures at each point of $\mathbb{O}\mathbb{P}^{2}$.
When $\mathbb{A=O}$, we only allow $n\leq2$; When $\mathbb{A=R}^{m}$, only $n\mathbb{=}1$ is admitted, and $\mathbb{R}^{m}\mathbb{P}^{1}=\mathbb{S}^{m}$. We will explain this notation in the next section.
$\mathbb{R}$, $\mathbb{C}$, $\mathbb{H}$, $\mathbb{O}$ and conformal geometry
=============================================================================
In [@boss_div_alg] the second author gave a unified description of geometries of each holonomy group by first defining the group $G_{\mathbb{A}
}\left( n\right) $ of twisted automorphisms of $\mathbb{A}^{n}$ and its subgroup $H_{\mathbb{A}}\left( n\right) $ of special twisted automorphisms, where $\mathbb{A\in}\left\{ \mathbb{R},\mathbb{C},\mathbb{H}
,\mathbb{O}\right\} $ is a normed division algebra and $n$ equals one when $\mathbb{A}=\mathbb{O}$. They are given explicitly in the following table:
\[c\][|r||c|c|c|c|]{}$\mathbb{A}$ & $\mathbb{R}$ & $\mathbb{C}$ & $\mathbb{H}$ & $\mathbb{O}
$\
$G_{\mathbb{A}}\left( n\right) $ & $O\left( n\right) $ & $U\left(
n\right) $ & $Sp\left( n\right) Sp\left( 1\right) $ & $\mathrm{Spin}
\left( 7\right) $\
$H_{\mathbb{A}}\left( n\right) $ & $SO\left( n\right) $ & $SU\left(
n\right) $ & $Sp\left( n\right) $ & $G_{2}$\
Their corresponding geometries are as follows.
\[c\][|r||c|c|c|c|]{}$\mathbb{A}$ & $\mathbb{R}$ & $\mathbb{C}$ & $\mathbb{H}$ & $\mathbb{O}
$\
$G_{\mathbb{A}}\left( n\right) $ & Riemannian & Kähler & Quaternionic-Kähler & $\mathrm{Spin}\left( 7\right) $\
$H_{\mathbb{A}}\left( n\right) $ & Volume & Calabi-Yau & Hyperkähler & $G_{2}$\
Due to the nonassociativity of the octonion, there are obvious difficulties to define its modules $\mathbb{O}^{n}$ and their automorphism groups $H_{\mathbb{O}}\left( n\right) $. Nonetheless, for $n\leq3$, this problem can be resolved by considering the space of self-adjoint operators, leading to the notion of Jordan algebra which we shall describe below.
On $\mathbb{R}^{n}$, the space of self-adjoint operators is simply the space of symmetric $n\times n$ matrices, denoted by $S_{n}\left( \mathbb{R}\right)
$. The symmetrization of ordinary matrix multiplication $$A\circ B=\left( AB+BA\right) /2$$ makes $S_{n}\left( \mathbb{R}\right) $ into a formally real Jordan algebra. Namely it is an algebra over $\mathbb{R}$ whose multiplication $\circ$ is commutative and power associative (that is, $(a\circ a)\circ a=a\circ(a\circ
a)$), together with $$a_{1}\circ a_{1}+\ldots+a_{n}\circ a_{n}=0\ \Rightarrow\ a_{1}=\ldots
=a_{n}=0\text{.}$$
The same product also makes the space $S_{n}\left( \mathbb{A}\right) $ of Hermitian symmetric matrices with entries in $\mathbb{A\in}\left\{
\mathbb{R},\mathbb{C},\mathbb{H}\right\} $ into a Jordan algebra. When $n=3$, an analog of the product can still be defined for $\mathbb{A}=\mathbb{O}$, making $S_{3}\left( \mathbb{O}\right) $ into an *exceptional Jordan algebra* (see e.g. [@Baez_octonions]) even though $\mathbb{O}$ lacks of associativity.
Inside $S_{n}\left( \mathbb{A}\right) $ we may collect all rank one projections, which are matrices $p$ with $p\circ p=p$ and $\mathrm{tr}\,p=1$, to form the projective space $\mathbb{AP}^{n-1}$. For instance, while the module $\mathbb{O}^{3}$ does not exist, the concept of octonion lines in $\mathbb{O}^{3}$ can be replaced by rank one projection operators in $S_{3}\left( \mathbb{O}\right)$, and the space of them forms the *octonion projective plane* $\mathbb{OP}^{2}$, which can be identified as the symmetric space $F_{4}/Spin\left( 9\right) $.
Since $S_{n}\left( \mathbb{A}\right) $ and $\mathbb{AP}^{n-1}$ are spaces of self-adjoint operators on $\mathbb{A}^{n}$, they should share the same automorphism group $H_{\mathbb{A}}\left( n\right) $ as $\mathbb{A}^{n}$. This is indeed true in the classical cases when $\mathbb{A}\in\left\{
\mathbb{R},\mathbb{C},\mathbb{H}\right\} $ and continues to have such an interpretation in the exceptional case $\mathbb{A}=\mathbb{O}$. The following gives a complete list of simple formally real Jordan algebras [@Jordan_alg] and their automorphism groups (The center has removed for simplicity):
\[c\][|r||c|c|c|c|c|]{}$\mathbb{A}$ & $\mathbb{R}$ & $\mathbb{C}$ & $\mathbb{H}$ & $\mathbb{O}$ & $\mathbb{R}^{m}$\
$S_{n}\left( \mathbb{A}\right) $ & $S_{n}\left( \mathbb{R}\right) $ & $S_{n}\left( \mathbb{C}\right) $ & $S_{n}\left( \mathbb{H}\right) $ & $S_{3}\left( \mathbb{O}\right) $ & $S_{2}\left( \mathbb{R}^{m}\right)
\simeq\mathbb{R}^{m}\oplus\mathbb{R}^{1,1}$\
$\mathbb{AP}^{n-1}$ & $\mathbb{RP}^{n-1}$ & $\mathbb{CP}^{n-1}$ & $\mathbb{HP}^{n-1}$ & $\mathbb{OP}^{2}$ & $\mathbb{AP}^{1}=\mathbb{S}^{m}
$\
$H_{\mathbb{A}}\left( n\right) $ & $SO\left( n\right) $ & $SU\left(
n\right) $ & $Sp\left( n\right) $ & $F_{4}$ & $SO\left( m+1\right)
$\
Amazingly there is one more item in the list of Jordan algebras besides those coming from normed division algebras, namely the *spin factor* $S_{2}\left( \mathbb{R}^{m}\right) \simeq\mathbb{R}^{m}\oplus\mathbb{R}^{1,1}$. It consists of $2\times2$ matrices of the form $$\left(
\begin{array}
[c]{cc}a-b & v\\
v & a+b
\end{array}
\right) \leftrightarrow\left(
\begin{array}
[c]{c}v\\
b\\
a
\end{array}
\right)$$ where $v\in\mathbb{R}^{m}$ and $a,b\in\mathbb{R}$, and we set $v\cdot
w=v^{t}w$ for $v,w\in\mathbb{R}^{m}$ to carry out matrix multiplication. The embedded projective space is$$\left\{ \left(
\begin{array}
[c]{c}v\\
b\\
\frac{1}{2}\end{array}
\right) :\left\Vert v\right\Vert ^{2}+b^{2}=\frac{1}{4}\right\}
\cong\mathbb{S}^{m}\newline\text{.}$$
Notice that the automorphism group $SO\left( m+1\right) $ of $S_{2}\left(
\mathbb{R}^{m}\right) $ is also the isometry group of $\mathbb{S}^{m}$, and it is contained as a maximal compact subgroup in the non-compact group $\mathrm{Conf}(\mathbb{S}^{m})=SO\left( m+1,1\right) $. A natural question arises: For $\mathbb{A\in}\left\{ \mathbb{R},\mathbb{C},\mathbb{H},\mathbb{O}\right\} $, is there a symmetry group of $\mathbb{AP}^{n-1}$ which gives an analog to the conformal symmetry $SO\left( m+1,1\right) $ of $\mathbb{S}^{m}$?
To answer this question, one identifies $\mathbb{S}^{m}$ as the conformal boundary of the hyperbolic ball $$B^{m+1}:=\{M\in S_{2}\left( \mathbb{R}^{m}\right) :\mathrm{\det}M=1\} \cong
SO(m+1,1)/SO(m+1)$$ on which $SO\left( m+1,1\right) $ acts as isometries. Under this identification, one has $\mathrm{Conf}(\mathbb{S}^{m})\cong\mathrm{Isom}
(B^{m+1})=SO\left( m+1,1\right) $ which preserves collinearity in the sense that $\mathrm{Conf}(\mathbb{S}^{m})$ maps circles to circles in $\mathbb{S}
^{m}$.
Now for $\mathbb{A}\in\left\{ \mathbb{R},\mathbb{C},\mathbb{H}\right\} $, if we collect the symmetries of $\mathbb{AP}^{n-1}$ which is linear but not necessarily isometries, we obtain the group $SL\left( n,\mathbb{A}\right) $ [@Salzmann_proj_planes]. Analogously $\mathbb{AP}^{n-1}$ can be identified as a part of the conformal boundary of $\{M\in S_{n}\left( \mathbb{A}\right)
:\mathrm{\det}M=1\}\cong$ $SL\left( n,\mathbb{A}\right) /SU(n,\mathbb{A)}$ on which $SL\left( n,\mathbb{A}\right) ~$acts as isometries. We get the answer for $\mathbb{A}\in\left\{ \mathbb{R},\mathbb{C},\mathbb{H}\right\} $: $SL\left( n,\mathbb{A}\right) $ can be regarded as the *conformal symmetry* of $\mathbb{AP}^{n-1}$, which plays the same role as $SO\left(
m+1,1\right) $ acting on $\mathbb{S}^{m}$. In general, let’s denote these non-compact symmetry groups as $N_{\mathbb{A}}\left( n\right) $ which are listed below. Notice that $H_{\mathbb{A}}\left( n\right) $ sits inside $N_{\mathbb{A}}\left( n\right) $ as a maximal compact subgroup, and $N_{\mathbb{A}}\left( n\right) /H_{\mathbb{A}}\left( n\right) $ can be identified with the space of symmetric matrices with determinant one.
\[c\][|r||c|c|c|c|c|]{}$\mathbb{A}$ & $\mathbb{R}$ & $\mathbb{C}$ & $\mathbb{H}$ & $\mathbb{O}$ & $\mathbb{R}^{m}$\
$H_{\mathbb{A}}\left( n\right) $ & $SO\left( n\right) $ & $SU\left(
n\right) $ & $Sp\left( n\right) $ & $F_{4}$ & $SO\left( m+1\right)
$\
$N_{\mathbb{A}}\left( n\right) $ & $SL\left( n,\mathbb{R}\right) $ & $SL\left( n,\mathbb{C}\right) $ & $SL\left( n,\mathbb{H}\right) $ & $E_{6}^{-26}$ & $SO\left( m+1,1\right) $\
We may observe that when $m=1,2,4$ and $8$, $N_{\mathbb{R}^{m}}\left(
2\right) =SL\left( 2,\mathbb{A}\right) $ with $\mathbb{A}$ being real, complex, quaternion and octonion respectively. Hence, $sl\left(
2,\mathbb{R}\right) =so\left( 2,1\right) $, $sl\left( 2,\mathbb{C}\right)
=so\left( 3,1\right) $, $sl\left( 2,\mathbb{H}\right) =so\left(
5,1\right) $, $sl\left( 2,\mathbb{O}\right) =so\left( 9,1\right) $. In general we have $sl\left( 2,\mathbb{A}\right) =so\left( \mathbb{A}
\oplus\mathbb{R}^{1,1}\right) $ [@Baez_octonions].
The above point of view integrates conformal geometry with real, complex, quaternionic and octonionic geometries. In the next section we will illustrate this viewpoint by studying the variation of volume of cycles under the conformal symmetry $N_{\mathbb{A}}\left( n\right) $ of $\mathbb{AP}^{n-1}$ in a unified manner.
In [@Atiyah_Proj], Atiyah and Berndt studied the complexified version of $\mathbb{AP}^{n-1}$ with $\mathbb{A\in}\left\{ \mathbb{R},\mathbb{C},\mathbb{H},\mathbb{O}\right\} $. We can extend these descriptions to $\mathbb{A=R}^{m}$ as in the following table:
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
$\mathbb{A}$ $\mathbb{R}$ $\mathbb{C}$ $\mathbb{H}$ $\mathbb{O}$ $\mathbb{R}^{m}$
---------------------------------------------------------------- ------------------------- ------------------------------------------ ------------------------------------------ ---------------------------------- ---------------------------------------
$\left( \mathbb{A}\otimes\mathbb{C}\right) \mathbb{P}^{n-1}$ $\mathbb{CP}^{n-1}$ $\left( \mathbb{CP}^{n-1}\right) ^{2}$ $Gr_{\mathbb{C}}\left( 2,2n-2\right) $ $\frac{E_{6}}{Spin\left( $\frac{O\left( m+2\right) }{O\left(
10\right) U\left( 1\right) }$ m\right) O\left( 2\right) }$
$H_{\mathbb{A\otimes C}}\left( n\right) $ $SU\left( n\right) $ $SU\left( n\right) ^{2}$ $SU\left( 2n\right) $ $E_{6}$ $SO\left(
m+2\right) $
$N_{\mathbb{A}\otimes\mathbb{C}}\left( n\right) $ $Sp\left( $SU\left( n,n\right) $ $O^{\ast}\left( $E_{7}^{-25}$ $SO\left( m+2,2\right) $
2n,\mathbb{R}\right) $ 4n\right) $
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Notice that the maximal compact subgroup of $N_{\mathbb{A}\otimes\mathbb{C}}\left( n\right) $ is the product of $H_{\mathbb{A\otimes C}}\left(
n\right) $ with $U(1)$. Furthermore, $$\frac{N_{\mathbb{A}\otimes\mathbb{C}}\left( n\right) }{H_{\mathbb{A\otimes
C}}\left( n\right) U(1)}=S_{n}^{+}\left( \mathbb{A}\right) +iS_{n}(\mathbb{A)}$$ is a tube domain (see for example [@Gross_tube]). This gives a complete list of tube domains.
They also have a quaternionic analog:
\[c\][|r||c|c|c|c|c|]{}$\mathbb{A}$ & $\mathbb{R}$ & $\mathbb{C}$ & $\mathbb{H}$ & $\mathbb{O}$ & $\mathbb{R}^{m}$\
$\left( \mathbb{A}\otimes\mathbb{H}\right) \mathbb{P}^{n-1}$ & $\mathbb{HP}^{n-1}$ & $Gr_{\mathbb{C}}(2,2n-2)$ & $Gr_{\mathbb{R}}(4,4n-4)$ & $\frac{E_{7}}{\mathrm{Spin}(12)O(4)}$ & $\frac{O(m+4)}{O(m)O(4)}$\
$H_{\mathbb{A\otimes H}}\left( n\right) $ & $Sp(n)$ & $SU(2n)$ & $SO(4n)$ & $E_{7}$ & $SO(m+4)$\
$N_{\mathbb{A}\otimes\mathbb{H}}\left( n\right) $ & $Sp(n,1)$ & $SU(2n,1)$ & $SO(4n,4)$ & $E_{8}^{-24}$ & $SO(m+4,4)$\
Cycles under conformal symmetries
=================================
In the last section, we regard $N_{\mathbb{A}}\left( n+1\right) $ as the conformal symmetry group of $\mathbb{AP}^{n}$. Its Lie algebra $$\mathfrak{n}_{\mathbb{A}}\left( n+1\right) =\mathfrak{h}_{\mathbb{A}}\left(
n+1\right) \oplus S_{n+1}^{\prime}\left( \mathbb{A}\right)$$ induces vector fields which acts infinitestimally on $\mathbb{AP}^{n}$. Here the Lie algebra $\mathfrak{h}_{\mathbb{A}}\left( n+1\right) $ of $H_{\mathbb{A}}\left( n+1\right) $ induces Killing vector fields, and $S_{n+1}^{\prime}\left( \mathbb{A}\right) $ consists of trace-free symmetric matrices, which can be regarded as constant vector fields in $S_{n+1}^{\prime
}\left( \mathbb{A}\right) $, projecting to conformal vector fields on $\mathbb{AP}^{n}\subset S_{n+1}^{\prime}(\mathbb{A})$. We are adopting the metric $$\left\langle A,B\right\rangle :=2\,\mathrm{Re}(\mathrm{tr}\,AB)=2\,\mathrm{Re}
(\mathrm{tr}\,A\circ B)$$ on $S_{n+1}^{\prime}(\mathbb{A})$ which induces the standard metric on $\mathbb{AP}^{n}$.
We would like to compute the average second variation of the volume of a cycle in $\mathbb{AP}^{n}$ under the action of $\mathfrak{n}_{\mathbb{A}}\left(
n+1\right) $. First, Let us quickly review the terminology and set up some notations.
Terminology and notations
-------------------------
For a global vector field $V$ on a Riemannian manifold $M$, the second variation $\mathcal{Q}_{S}(V)$ of the volume $\mathbf{M}$ of a rectifiable current $S$ under $V$ is defined as $$\mathcal{Q}_{S}(V):=\left. \frac{\mathrm{d}^{2}}{\mathrm{d}t^{2}}\right\vert
_{t=0}\mathbf{M}((\phi_{t})_{\ast}S)=\int_{M}\,\left. \frac{\mathrm{d}^{2}
}{\mathrm{d}t^{2}}\right\vert _{t=0}||(\phi_{t})_{\ast}S_{x}||\mathrm{d}
\nu_{S}(x)$$ where $\phi_{t}$ is the flow induced by $V$, $S_{x}$ denotes the unit simple vector representing the oriented tangent space of $S$ at $x$, and $\nu_{S}$ denotes the Borel measure associated with $S$. $S$ is said to be stable if $\mathcal{Q}_{S}(V)\leq0$ for all vector fields $V$ on $M$. We will denote the integrand $\left. \frac{\mathrm{d}^{2}}{\mathrm{d}t^{2}}\right\vert
_{t=0}||(\phi_{t})_{\ast}\xi||$ by $\mathcal{Q}_{\xi}\left( V\right) $, the second variation of an oriented orthonormal $p$-frame $\xi$ under $V$. One has the following second variation formula for a gradient vector field $V$ [@Lawson_stable]: $$\begin{aligned}
\label{sec_var}\mathcal{Q}_{\xi}\left( V\right) & =\left\langle
\mathcal{A}_{V,V}\xi,\xi\right\rangle +2\Vert\mathcal{A}_{V}\xi\Vert
^{2}-(\left\langle \mathcal{A}_{V}\xi,\xi\right\rangle )^{2}\nonumber\\
& =\left( \sum_{j=1}^{p}\left\langle \mathcal{A}_{V}e_{j},e_{j}\right\rangle
\right) ^{2}+2\sum_{j=1}^{p}{\sum\limits_{k=1}^{q}}\left( \left\langle
\mathcal{A}_{V}e_{j},n_{k}\right\rangle \right) ^{2}+\sum_{j=1}
^{p}\left\langle \mathcal{A}_{V,V}e_{j},e_{j}\right\rangle\end{aligned}$$ where $\xi=e_{1}\wedge\ldots\wedge e_{p}$, which is extended to an orthonormal basis $\{e_{1},\ldots,e_{p},n_{1},\ldots,n_{q}\}$ of $TM$. Here for any smooth vector fields $V$ and $W$, $\mathcal{A}_{V}(u)$, $\mathcal{A}
_{V,W}$ are endomorphisms of $TM$ defined by $$\begin{aligned}
\mathcal{A}_{V}X:= & \nabla_{X}V;\nonumber\label{D^2}\\
\mathcal{A}_{V,W}X:= & (\nabla_{V}\mathcal{A}_{W})X=\nabla_{V}\nabla
_{\tilde{X}}W-\nabla_{\nabla_{V}\tilde{X}}W\end{aligned}$$ where $\nabla$ is the Levi-Civita connection, $\tilde{X}$ is a smooth local extension of $X\in TM$. An endomorphism $L$ of $TM$ is extended to operate on $\bigwedge^{p}TM$ by Leibniz rule: $$L(e_{1}\wedge\ldots\wedge e_{p})=\sum_{j=1}^{p}e_{1}\wedge\ldots\wedge
Le_{j}\wedge\ldots\wedge e_{p}\text{.}$$
From the above second variation formula, we see that $\mathcal{Q}_{\xi}$, and hence $\mathcal{Q}_{S}$, is a quadratic form on the space of smooth vector fields on $M$, and we may restrict it to a finite-dimensional subspace $F$ of vector fields and take the trace $(\mathrm{tr}\,\mathcal{Q}_{\xi}|_{F}
)=\sum\mathcal{Q}_{\xi}(V)$, where $V$ runs through an orthonormal basis of $F$.
Main theorem
------------
Coming back to our situation $M=\mathbb{AP}^{n}$, since vector fields induced by $\mathfrak{h}_{\mathbb{A}}\left( n+1\right) $ preserve metric and does not contribute to the second variation, we have $$\mathrm{tr}\,\mathcal{Q}_{\xi}|_{\mathfrak{n}_{\mathbb{A}}\left( n+1\right)
}=\mathrm{tr}\,\mathcal{Q} _{\xi}|_{S_{n+1}^{\prime}\left( \mathbb{A}\right)
}$$ and so we may concentrate on $F=S_{n+1}^{\prime}\left( \mathbb{A}\right) $.
Moreover, notice that $\mathbb{AP}^{n}$ is an orbit of the group $H_{\mathbb{A}}\left( n+1\right) $ acting on $S_{n+1}^{\prime}\left(
\mathbb{A}\right) $. This symmetry helps to reduce a lot of calculations, as illustrated by the following lemma:
\[inv of secvar under G\] Let $G$ act isometrically on an inner product space $\mathbb{V}$, and $M\subset\mathbb{V}$ be a $G$-invariant submanifold. The projection of each $u\in\mathbb{V}$ gives a vector field $V_{u}$ on $M$, and the space of all these vector fields is denoted by $F$. Then $$\mathrm{tr}\,\mathcal{Q}_{\xi}|_{F}=\mathrm{tr}\,\mathcal{Q}_{g\cdot\xi}|_{F}$$ for all $g\in G$.
Since the metric on $M$ is $G$-invariant, the Levi-Civita connection $\nabla$ is $G$-equivariant, that is, $$\nabla_{g_{\ast}\cdot X}(g_{\ast}\cdot V)=g_{\ast}\cdot(\nabla_{X}V)\text{.}$$ Hence one has $$\mathcal{A}_{V}(g\cdot\xi)=g\cdot(\mathcal{A}_{g_{\ast}^{-1}V}\cdot
\xi);\,\mathcal{A}_{V,W}(g\cdot\xi)=g\cdot(\mathcal{A}_{g_{\ast}
^{-1}V,\,g_{\ast}^{-1}W}\cdot\xi)\text{.}$$ Applying to the second variation formula, we get $$\mathcal{Q}_{g\cdot\xi}(V_{u})=\mathcal{Q}_{\xi}(g_{\ast}^{-1}V_{u}
)=\mathcal{Q}_{\xi}(V_{g_{\ast}^{-1}u})$$ where the last equality is due to $G$-invariance of metric. And so $$\mathrm{tr}\,\mathcal{Q}_{\eta}=\sum_{u}\mathcal{Q}_{\eta}(u)=\sum
_{u}\mathcal{Q}_{\xi}(g_{\ast}^{-1}u)=\mathrm{tr}\,\mathcal{Q}_{\xi}$$ where $u$, and hence $g_{\ast}^{-1}u$, runs through an orthonormal basis of $\mathbb{V}$. The last equality follows from the fact that trace is independent of choice of orthonormal basis.
By the above lemma, where we take $M=\mathbb{AP}^{n},\mathbb{V=}
S_{n+1}^{\prime}\left( \mathbb{A}\right) $ and $G=H_{\mathbb{A}}\left(
n+1\right) $, it suffices to consider average second variation of a $p$-frame $\xi=e_{1}\wedge\ldots\wedge e_{p}$ at a particular point $x\in\mathbb{AP}
^{n}$, because $p$-frames at another point can be moved to $x$ by some $g\in
H_{\mathbb{A}}\left( n+1\right) $. Let’s fix $x=\mathbb{E}_{n+1,n+1}
\in\mathbb{AP}^{n}$, which is the matrix with value $1$ at the $(n+1,n+1)$ position and all other entries zero.
We shall need the following formula, whose proof is given in the appendix:
\[secvarorbit\] Assume that $M=G/K\subset$ $\mathbb{V}$ is a compact symmetric space which is a $G$-orbit of an orthogonal representation $\mathbb{V}$ of $G$. The projection of each $u\in\mathbb{V}$ gives a vector field $V_{u}$ on $M$. The average second variation of an oriented orthonormal $p$-frame $\xi=e_{1}\wedge\ldots\wedge e_{p}$ at $x\in M$ under all such vector fields is given by $$\mathrm{tr}\,\mathcal{Q}_{\xi}=\sum_{j,k=1}^{p,q}\left( 2\,\Vert
\,\mathrm{I}\mathrm{I}(e_{j},n_{k})\Vert^{2}-\left\langle \,\mathrm{I}
\mathrm{I}(e_{j},e_{j}),\,\mathrm{I}\mathrm{I}(n_{k},n_{k})\right\rangle
\right)$$ where $\mathrm{I}\mathrm{I}$ is the second fundamental form of $M\subset\mathbb{V}$ at $x$, and $\{e_{j}\}_{j=1}^{p}\cup\{n_{k}\}_{k=1}^{q}$ is an orthonormal basis of $TM$.
With the above formula, it remains to compute the second fundamental form of $\mathbb{AP}^{n}$. Let’s take the following coordinates around $x$: $$\begin{aligned}
\mathbb{A}^{n} & \rightarrow\mathbb{A}\mathbb{P}^{n}\subset S_{n+1}^{\prime
}(\mathbb{A})\\
Q & \mapsto\frac{1}{1+\Vert Q\Vert^{2}}\left(
\begin{array}
[c]{c}Q\\
1
\end{array}
\right) \left(
\begin{array}
[c]{cc}Q^{\ast} & 1
\end{array}
\right)\end{aligned}$$
Here we adopt the following notations: $$Q=\sum_{l=0}^{\Lambda}\mathbf{i}_{l}X_{l}$$ where $X_{l}$ are column $n$-vectors, $\mathbf{i}_{0}:=1$, and for $1\leq
l\leq\Lambda$, $\mathbf{i}_{l}$ are the linearly independent imaginary square roots of unity in $\mathbb{A}$. Recall that for the case $\mathbb{A}
=\mathbb{R}^{m}$, $n=1$, $\Lambda=0$, $Q=X_{0}$ is an element in $\mathbb{R}^{m}$ with $Q^{\ast}=Q$ and $Q\cdot Q:=\left\langle
Q,Q\right\rangle $. For the other four cases, the entries of $X_{l}$ are real numbers.
The basis of coordinate tangent vector fields is $\{\frac{\partial}{\partial
x^{j}_{l}}: 0 \leq l \leq\Lambda, 1 \leq j \leq N\}$, where $\frac{\partial
}{\partial x^{j}_{l}}$ denote the $\mathbf{i}_{l}$-directions. $N = m$ in the case of $\mathbb{A} = \mathbb{R}^{m}$, and $N = n$ for all the other four cases. Using product rule (which is valid for multiplication in $\mathbb{A}$), $$\begin{aligned}
\left. \frac{\partial}{\partial x_{l}^{j}}\right\vert _{Q} & =\frac
{1}{1+\Vert Q\Vert^{2}}\left(
\begin{array}
[c]{c}\mathbf{i}_{l}w_{j}\\
0
\end{array}
\right) \left(
\begin{array}
[c]{cc}Q^{\ast} & 1
\end{array}
\right) \\
& +\frac{1}{1+\Vert Q\Vert^{2}}\left(
\begin{array}
[c]{c}Q\\
1
\end{array}
\right) \left(
\begin{array}
[c]{cc}\overline{\mathbf{i}_{l}}w_{j}^{T} & 0
\end{array}
\right) \\
& -\frac{2X_{l}^{T}w_{j}}{(1+\Vert Q\Vert)^{2}}\left(
\begin{array}
[c]{c}Q\\
1
\end{array}
\right) \left(
\begin{array}
[c]{cc}Q^{\ast} & 1
\end{array}
\right)\end{aligned}$$ where $w_{j}$ stands for the column $n$-vector with $j$-th coordinate $1$ and other coordinates zero, and $T$ stands for transpose. Recall that when $\mathbb{A}=\mathbb{R}^{m}$, $n$ equals $1$, and so transpose of an element is just itself. Differentiating both sides along $\frac{\partial}{\partial
x_{r}^{k}}$ at $0\in\mathbb{A}^{n}$, $$\begin{aligned}
& \left. \frac{\partial}{\partial x_{r}^{k}}\right\vert _{0}\left(
\frac{\partial}{\partial x_{l}^{j}}\right) \\
& =\left\{
\begin{array}
[c]{ll}\left(
\begin{array}
[c]{cc}2\delta_{jk} & 0\\
0 & -2\delta_{jk}\end{array}
\right) & \text{ for $\mathbb{A}=\mathbb{R}^{m}$}\\
& \\
\left(
\begin{array}
[c]{cc}\mathbf{i}_{r}\overline{\mathbf{i}_{l}}\mathbb{E}_{kj}+\mathbf{i}_{l}
\overline{\mathbf{i}_{r}}\mathbb{E}_{jk} & 0\\
0 & -(\mathbf{i}_{r}\overline{\mathbf{i}_{l}}+\mathbf{i}_{l}\overline
{\mathbf{i}_{r}})\delta_{jk}\end{array}
\right) & \text{ for }\mathbb{A}=\mathbb{R},\mathbb{C},\mathbb{H},\mathbb{O}\end{array}
\right.\end{aligned}$$ which is already perpendicular to $T_{x}\mathbb{A}\mathbb{P}^{n}$, because $$\left. \frac{\partial}{\partial x_{l}^{j}}\right\vert _{0}=\left(
\begin{array}
[c]{cc}0 & \mathbf{i}_{l}w_{j}\\
\overline{\mathbf{i}_{l}}w_{j}^{T} & 0
\end{array}
\right) \text{.}$$ Under the metric $\left\langle A,B\right\rangle =2\,\mathrm{Re}\,\mathrm{tr}
\,(AB)$, our coordinate vectors are pairwise orthogonal, each has length $2$. We scale them to get an orthonormal basis $\{\frac{1}{2}\frac{\partial
}{\partial x_{l}^{j}}:1\leq j\leq n,0\leq l\leq\Lambda\}$.
We conclude that
\[secform\]The second fundamental form $\mathrm{I}\mathrm{I}(\frac{1}
{2}\frac{\partial}{\partial x_{l}^{j}},\frac{1}{2}\frac{\partial}{\partial
x_{r}^{k}})$ of $\mathbb{AP}^{n}\subset S_{n+1}^{\prime}(\mathbb{A})$ at $x$ is given by $$\left\{
\begin{array}
[c]{ll}\frac{1}{2}\left(
\begin{array}
[c]{cc}\delta_{jk} & 0\\
0 & -\delta_{jk}\end{array}
\right) & \text{ for $\mathbb{A}=\mathbb{R}^{m}$}\\
\frac{1}{4}\left(
\begin{array}
[c]{cc}\mathbf{i}_{r}\overline{\mathbf{i}_{l}}\mathbb{E}_{kj}+\mathbf{i}_{l}
\overline{\mathbf{i}_{r}}\mathbb{E}_{jk} & 0\\
0 & -(\mathbf{i}_{r}\overline{\mathbf{i}_{l}}+\mathbf{i}_{l}\overline
{\mathbf{i}_{r}})\delta_{jk}\end{array}
\right) & \text{ for }\mathbb{A}\in\{\mathbb{R},\mathbb{C},\mathbb{H}
,\mathbb{O}\}\text{.}\end{array}
\right.$$
Now we are ready to compute $\mathrm{tr}\,\mathcal{Q}_{\xi}$ for an orthonormal $p$-frame $\xi=e_{1}\wedge\ldots\wedge e_{p}$ at $x\in\mathbb{A}\mathbb{P}^{n}$. Complete $B=\{e_{j}\}_{j=1}^{p}$ to an orthonormal basis $\{e_{j},n_{k}\}$ in the form $$\left\{
\begin{array}
[c]{cccc}v_{1}, & \mathbb{J}_{1}v_{1}, & \ldots & \mathbb{J}_{\Lambda}v_{1}\\
\vdots & \vdots & & \vdots\\
v_{N}, & \mathbb{J}_{1}v_{N}, & \ldots & \mathbb{J}_{\Lambda}v_{N}\end{array}
\right\}$$ where $\mathbb{J}_{l}:T_{x}\mathbb{A}\mathbb{P}^{n}\rightarrow T_{x}
\mathbb{A}\mathbb{P}^{n}$ is the differential of left multiplication of $\mathbf{i}_{l}$ on $\mathbb{A}^{n}\subset\mathbb{A}\mathbb{P}^{n}$.
Such an orthonormal basis can be brought to the basis of normalized coordinate vectors by the action of the isotropy group $K<G$. This is easy for $\mathbb{R}\mathbb{P}^{n}$, $\mathbb{C}\mathbb{P}^{n}$ and $\mathbb{H}\mathbb{P}^{n}$: $SO(n)$, $SU(n)$ and $Sp(n)$ acts transitively on orthonormal frames, unitary frames and quaternionic unitary frames respectively. For $\mathbb{O}\mathbb{P}^{2}$, $K=\mathrm{Spin}(9)<\mathrm{F}_{4}$, we argue as follows: $T_{x}\mathbb{O}\mathbb{P}^{2}$ is the spinor representation of $\mathrm{Spin}(9)$. Under this action $$T_{x}\mathbb{O}\mathbb{P}^{2}\supset\mathbb{S}^{15}\cong\mathrm{Spin}(9)/\mathrm{Spin}(7)$$ (see P.283 of [@Harvey_spinors]). Hence we can use $\sigma\in
\mathrm{Spin}(9)$ to bring $\frac{1}{2}\frac{\partial}{\partial x_{0}^{1}}$ to $v_{1}$. $\mathrm{Spin}(7)$ fixes $v_{1}$ and hence acts on $T_{v_{1}}\mathbb{S}^{15}$, which splits into the vector representation $V_{7}$ and the spinor representation of $\mathrm{Spin}(7)$. $\left\{ \sigma\left( \frac
{1}{2}\frac{\partial}{\partial x_{l}^{1}}\right) \right\} _{l=1}^{7}$ and $\{\mathbb{J}_{l}v_{1}\}_{l=1}^{7}$ form two bases of $V_{7}$ having the same orientation. Then we can bring $\left\{ \sigma\left( \frac{1}{2}\frac{\partial}{\partial x_{l}^{1}}\right) \right\} _{l=1}^{7}$ to $\{\mathbb{J}_{l}v_{1}\}_{l=1}^{7}$ by an element in $\mathrm{Spin(7)}$. $\left\{ \sigma\left( \frac{1}{2}\frac{\partial}{\partial x_{l}^{2}}\right)
\right\} _{l=1}^{7}$ can be brought to $\{\mathbb{J}_{l}v_{2}\}_{l=0}^{7}$ by $\mathrm{Spin(7)}$ using similar reasoning, because $$\mathrm{Spin(7)}/G_{2}\cong\mathbb{S}^{7}\text{ and }\mathrm{G_{2}}/\mathrm{SU}(3)\cong\mathbb{S}^{6}$$ and $\mathrm{SU}(3)$ acts transitively on the collection of unitary bases of $\mathbb{C}^{3}$.
By Lemma \[inv of secvar under G\], average second variations of $\xi$ and $g\cdot$ $\xi$ are the same for all $g\in G$, and hence we may assume $$\mathbb{J}_{l}v_{j}=\frac{1}{2}\frac{\partial}{\partial x_{l}^{j}}$$ so that we can apply Lemma \[secform\] directly.
For the case $\mathbb{A}=\mathbb{R}^{m}$ in which $\mathbb{A}\mathbb{P}
^{1}=\mathbb{S}^{m}$, Lemma \[secform\] gives $$\left\Vert \,\mathrm{I}\mathrm{I}(\frac{1}{2}\frac{\partial}{\partial x^{j}
},\frac{1}{2}\frac{\partial}{\partial x^{k}})\right\Vert ^{2}=\delta_{jk}$$ which is the usual formula for the second fundamental form of $\mathbb{S}
^{m}\subset\mathbb{R}^{m+1}$. Together with Theorem \[secvarorbit\], the result of Lawson and Simons [@Lawson_stable] is reproduced: $$\mathrm{tr}\,\mathcal{Q}_{\xi}=\sum_{j,k=1}^{p,q}(-1)=-pq\leq0$$ where $p+q=m$, implying that the average second variation of a rectifiable current of non-zero volume in $\mathbb{S}^{n}$ is negative for $0<p<m$, and hence cannot be stable.
Now let’s turn to the other four cases. Lemma \[secform\] gives $$\Vert\,\mathrm{I}\mathrm{I}(e_{j},n_{k})\Vert^{2}=\left\{
\begin{array}
[c]{ll}0 & \text{for }n_{k}=\pm\mathbb{J}_{l}e_{j}\text{ for some $1\leq l\leq
\Lambda$}\\
\frac{1}{4} & \text{otherwise}\end{array}
\right.$$ and $$\left\langle \,\mathrm{I}\mathrm{I}(e_{j},e_{j}),\,\mathrm{I}\mathrm{I}
(n_{k},n_{k})\right\rangle =\left\{
\begin{array}
[c]{ll}1 & \text{for }n_{k}=\pm\mathbb{J}_{l}e_{j}\text{ for some $1\leq l\leq
\Lambda$}\\
\frac{1}{2} & \text{otherwise}\end{array}
\right.$$ so the summand appeared in Theorem \[secvarorbit\] is $$2\Vert\,\mathrm{I}\mathrm{I}(e_{j},n_{k})\Vert^{2}-\left\langle \,\mathrm{I}
\mathrm{I}(e_{j},e_{j}),\,\mathrm{I}\mathrm{I}(n_{k},n_{k})\right\rangle
=\left\{
\begin{array}
[c]{ll}-1 & \text{for }n_{k}=\pm\mathbb{J}_{l}e_{j}\text{ for some $1\leq
l\leq\Lambda$}\\
0 & \text{otherwise}\end{array}
\right.$$ meaning that for each $e_{j}$, every $\mathbb{J}_{l}e_{j}$-direction normal to $\xi$ contributes $-1$ to $\mathrm{tr}\,\mathcal{Q}_{\xi}$, and all other normal directions have no effect. Hence $$\begin{aligned}
\mathrm{tr}\,\mathcal{Q}_{\xi} & =-\sum_{j=1}^{p}\,(\text{number of $l$ such
that $\pm\mathbb{J}_{l}e_{j}\not \in B$})\\
& =-\sum_{j=1}^{p}\sum_{l=1}^{\Lambda}\Vert e_{1}\wedge\ldots\wedge
\mathbb{J}_{l}e_{j}\wedge\ldots\wedge e_{p}\Vert^{2}\\
& =-\sum_{l=1}^{\Lambda}\Vert\mathbb{J}_{l}\cdot\xi\Vert^{2}\leq0\text{.}$$ (Here $\mathbb{J}$ acts on $\xi$ by Leibniz rule.) Equality holds if and only if $\Vert\mathbb{J}_{l}\cdot\xi\Vert^{2}=0$ for all $1\leq l\leq\Lambda$, meaning that $\xi$ is invariant under each $\mathbb{J}_{l}$, and hence invariant under the $\mathbb{S}^{\Lambda-1}$-family of complex structures. Hence we obtain the following theorem:
\[main\]In $\mathbb{AP}^{n}$, where $\mathbb{A\in}\left\{ \mathbb{R},\mathbb{C},\mathbb{H},\mathbb{O},\mathbb{R}^{m}\right\} $, any stable minimal submanifold $S$ (or more generally rectifiable current) must be complex, by which we means $T_{x}S$ is invariant under all the linear complex structures at $x$ for almost every $x\in S$.
We remark that in $\mathbb{H}\mathbb{P}^{n}$, a quaternionic submanifold must be totally geodesic.
Appendix: Average second variation in symmetric orbits
======================================================
Our aim is to prove the following theorem, which we have used in the last section to compute the average second variation of the volume of a cycle in $\mathbb{A}\mathbb{P}^{n}$ along directions in $\mathfrak{h}
_{\mathbb{A}}\left( n+1\right) $:
**Theorem:** [*Assume that $M=G/K$ is a compact symmetric space which is a $G$-orbit of an orthogonal representation $\mathbb{V}$ of $G$. The projection of each $u\in\mathbb{V}$ determines a vector field $V_{u}$, or simply $V$, on $M$. The average second variation of an oriented orthonormal $p$-frame $\xi=e_{1}\wedge\ldots\wedge e_{p}$ at $x\in M$ under all such vector fields is given by $$\mathrm{tr}\,\mathcal{Q}_{\xi}=\sum_{j,k=1}^{p,q}\left( 2\,\Vert
\,\mathrm{I}\mathrm{I}(e_{j},n_{k})\Vert^{2}-\left\langle \,\mathrm{I}\mathrm{I}(e_{j},e_{j}),\,\mathrm{I}\mathrm{I}(n_{k},n_{k})\right\rangle
\right)$$ where $\,\mathrm{I}\mathrm{I}$ is the second fundamental form of $M\subset\mathbb{V}$ at $x$, and $\{e_{j}\}_{j=1}^{p}\cup\{n_{k}\}_{k=1}^{q}$ is an orthonormal basis of $T_{x}M$.*]{}
The method of proof is similar to [@Lawson_stable]. The Lie algebra $\mathfrak{g}$ of $G$ decomposes: $$\mathfrak{g}=\mathfrak{k}\oplus\mathfrak{m}$$ where $\mathfrak{m}:=\mathfrak{k}^{\perp}$. On $G$ we have a natural $G$-invariant metric given by negative of the Killing form, which can be scaled such that $\mathfrak{m}$ is isometric to $T_{x}M$. We shall use the same symbol to denote an element of $\mathfrak{g}$, its induced vector field on $\mathbb{V}$, and the restricted Killing vector field on $M$. Recall that $$\lbrack g_{1},g_{2}]_{M}=-[g_{1},g_{2}] \label{bracket}$$ where $[\cdot,\cdot]_{M}$ is the Lie bracket for vector fields on $M$, and $[\cdot,\cdot]$ is the Lie bracket on $\mathfrak{g}$. On the right hand side $g_{1},g_{2}$ denote elements in $\mathfrak{g}$, while on the left hand side they denote their induced Killing vector fields on $M$.
Let’s complete $\xi=e_{1}\wedge\ldots\wedge e_{p}$ to an orthonormal basis $\{e_{1},\ldots,e_{p},n_{1},\ldots,n_{q}\}$ of $T_{x}M\cong\mathfrak{m}$, and further take an orthonormal basis $\{\beta_{1},\ldots,\beta_{r}\}$ of $\mathfrak{k}$, so that $\{\beta_{1},\ldots,\beta_{r},e_{1},\ldots,e_{p}
,n_{1},\ldots,n_{q}\}$ forms an orthonormal basis of $\mathfrak{g}$.
We now express the projection $V=V_{u}$ of $u\in\mathbb{V}$ in terms of Killing vector fields induced by $\mathfrak{g}$ on $M$.
\[exp for V\] $$V=\sum_{\mu=1}^{r}\left\langle u,\beta_{\mu}\right\rangle \beta_{\mu}
+\sum_{\nu=1}^{p}\left\langle u,e_{\nu}\right\rangle e_{\nu}+\sum_{\gamma
=1}^{q}\left\langle u,n_{\gamma}\right\rangle n_{\gamma}\text{.}$$
Denote the basis $\{\beta_{1},\ldots,\beta_{r},e_{1},\ldots,e_{p},n_{1}
,\ldots,n_{q}\}$ of $\mathfrak{g}$ by $A$.
At $x\in M$ the above equation is obvious, because $\beta_{\mu}(x)=0$,and $\{e_{1},\ldots,e_{p},n_{1},\ldots,n_{q}\}$ forms an orthonormal basis of $T_{x}M$.
At another point $y\in M$, let $\{\tilde{e_{1}},\ldots,\tilde{e_{p}},\tilde{n_{1}},\ldots,\tilde{n_{q}}\}$ be an orthonormal basis of $T_{y}M\cong\mathfrak{m}$, and we complete it to an orthonormal basis $$B=\{\tilde{\beta}_{1},\ldots,\tilde{\beta}_{r},\tilde{e}_{1},\ldots,\tilde
{e}_{p},\tilde{n}_{1},\ldots,\tilde{n}_{q}\}$$ of $\mathfrak{g}$. Both $A,B$ are orthonormal basis of $\mathfrak{g}$, so $B=AT$, where $T$ is an orthogonal matrix. $$V(x)=\sum_{j}\left\langle u,B_{j}\right\rangle B_{j}=\sum_{j}\left\langle
u,A_{k}T_{j}^{k}\right\rangle A_{i}T_{j}^{i}=\sum_{j}\left\langle
u,A_{j}\right\rangle A_{j}$$ since $\sum_{j}T_{j}^{k}T_{j}^{i}=\delta^{ki}$.
**Proof to Theorem \[secvarorbit\]:** From the second variation formula , the average second variation is given by $$\mathrm{tr}\,\mathcal{Q}_{\xi}=\sum_{u}\left( \sum_{j=1}^{p}\left\langle
\mathcal{A}_{V}e_{j},e_{j}\right\rangle \right) ^{2}+2\sum_{u}\sum
_{j=1,k=1}^{p,q}\left( \left\langle \mathcal{A}_{V}e_{j},n_{k}\right\rangle
\right) ^{2}+\sum_{u}\sum_{j=1}^{p}\left\langle \mathcal{A}_{V,V}e_{j},e_{j}\right\rangle$$ where $u$ runs through an orthonormal basis of $\mathbb{V}$, each gives a vector field $V=V_{u}$ on $M$ by projection. We compute term by term for the three terms appeared in the above expression.
Recall [@Helgason_symmetric] that for a symmetric space, $$\nabla_{K_{1}}K_{2}=\frac{1}{2}\,[K_{1},K_{2}]_{M}$$ for Killing vector fields $K_{1}$ and $K_{2}$ on $M$. Applying this to the expression of $V$ given in Lemma \[exp for V\], $$\begin{aligned}
\nabla_{e_{j}}V & =\left\langle u,\partial_{e_{j}}\beta_{\mu}\right\rangle
\beta_{\mu}+\frac{1}{2}\,\left\langle u,\beta_{\mu}\right\rangle [e_{j}
,\beta_{\mu}]_{M}+\left\langle u,\partial_{e_{j}}e_{\nu}\right\rangle e_{\nu
}+\frac{1}{2}\,\left\langle u,e_{\nu}\right\rangle [e_{j},e_{\nu}
]_{M}\nonumber\label{dV}\\
& +\left\langle u,\partial_{e_{j}}n_{\gamma}\right\rangle n_{\gamma}+\frac
{1}{2}\,\left\langle u,n_{\gamma}\right\rangle [e_{j},n_{\gamma}]_{M}$$ where $\partial$ is the trivial connection of $\mathbb{V}$, and so $\partial_{v}$ is the usual directional derivative along $v\in T_{x}
\mathbb{V}\cong\mathbb{V}$. (Recall that $\beta_{\mu}$, $e_{\nu}$, $n_{\gamma
}$ can be regarded as vector fields on $\mathbb{V}$, and so the above directional derivatives make sense.)
To simplify the above expression at $x$, notice that $\mathfrak{k}$ induces zero vectors at $x$, and hence $\beta_{\mu}\in\mathfrak{k}$ vanishes at $x $. Together with equation and the fact that $$\lbrack\mathfrak{k},\mathfrak{k}]\subset\mathfrak{k},[\mathfrak{k}
,\mathfrak{m}]\subset\mathfrak{m},[\mathfrak{m},\mathfrak{m}]\subset
\mathfrak{k} \label{t_and_p}$$ we have $$\nabla_{e_{j}}V(x)=\left\langle u,\partial_{e_{j}}e_{\nu}\right\rangle e_{\nu
}+\left\langle u,\partial_{e_{j}}n_{\gamma}\right\rangle n_{\gamma}$$ and hence $$\begin{array}
[c]{lllll}\left\langle \mathcal{A}_{V}e_{j},e_{j}\right\rangle & = & \left\langle
\nabla_{e_{j}}V(x),e_{j}\right\rangle & = & \left\langle u,\partial_{e_{j}
}e_{j}\right\rangle \text{;}\\
\left\langle \mathcal{A}_{V}e_{j},n_{k}\right\rangle & = & \left\langle
\nabla_{e_{j}}V(x),n_{k}\right\rangle & = & \left\langle u,\partial_{e_{j}
}n_{k}\right\rangle \text{.}\end{array}$$
The first term $\sum_{u}\left( \sum_{j=1}^{p}\left\langle \mathcal{A}
_{V}e_{j},e_{j}\right\rangle \right) ^{2}$ is $$\begin{aligned}
\sum_{u}\left( \sum_{j=1}^{p}\left\langle \mathcal{A}_{V}e_{j},e_{j}
\right\rangle \right) ^{2} & =\sum_{u}\sum_{j,k=1}^{p}\left\langle
u,\partial_{e_{j}}e_{j}\right\rangle \left\langle u,\partial_{e_{k}}
e_{k}\right\rangle \\
& =\sum_{j,k=1}^{p}\left\langle \partial_{e_{j}}e_{j},\partial_{e_{k}}
e_{k}\right\rangle \\
& =\left\Vert \sum_{j=1}^{p}\,\mathrm{I}\mathrm{I}(e_{j},e_{j})\right\Vert
^{2}$$ where $\partial_{e_{j}}e_{j}=\,\mathrm{I}\mathrm{I}(e_{j},e_{j})$ because $\nabla_{e_{j}}e_{j}=[e_{j},e_{j}]_{M}/2=0$.
The second term $2\sum_{u}\sum_{j=1,k=1}^{p,q}\left( \left\langle
\mathcal{A}_{V}e_{j},n_{k}\right\rangle \right) ^{2}$ is $$\begin{aligned}
2\sum_{u}\sum_{j=1,k=1}^{p,q}\left( \left\langle \mathcal{A}_{V}e_{j}
,n_{k}\right\rangle \right) ^{2} & =2\sum_{u}\sum_{j=1,k=1}^{p,q}\left(
\left\langle u,\partial_{e_{j}}n_{k}\right\rangle \right) ^{2}\\
& =2\sum_{j,k=1}^{p,q}\Vert\partial_{e_{j}}n_{k}\Vert^{2}\\
& =2\sum_{j,k=1}^{p,q}\Vert\,\mathrm{I}\mathrm{I}(e_{j},n_{k})\Vert^{2}$$ where $\partial_{e_{j}}n_{k}=\,\mathrm{I}\mathrm{I}(e_{j},n_{k})$ at $x$ because $\nabla_{e_{j}}n_{k}(x)=[e_{j},n_{k}]_{M}/2=0$.
Now we turn to compute the third term $\sum_{u}\sum_{j=1}^{p}\left\langle
\mathcal{A}_{V,V}e_{j},e_{j}\right\rangle $, which is more complicated. At $x
$, $$\begin{aligned}
\left\langle \mathcal{A}_{V,V}e_{j},e_{j}\right\rangle & =\left\langle
\nabla_{V}\nabla_{e_{j}}V-\nabla_{\nabla_{V}e_{j}}V,e_{j}\right\rangle \\
& =\left\langle \nabla_{V}\nabla_{e_{j}}V,e_{j}\right\rangle \\
& =\sum_{\nu=1}^{p}\left\langle u,e_{\nu}\right\rangle \left\langle
\nabla_{e_{\nu}}\nabla_{e_{j}}V,e_{j}\right\rangle +\sum_{\gamma=1}
^{q}\left\langle u,n_{\gamma}\right\rangle \left\langle \nabla_{n_{\gamma}
}\nabla_{e_{j}}V,e_{j}\right\rangle\end{aligned}$$ where $\nabla_{\nabla_{V}e_{j}}V(x)=0$ because $$\nabla_{V}e_{j}(x)=\sum_{\nu=1}^{p}\left\langle u,e_{\nu}\right\rangle
\frac{[e_{\nu},e_{j}]_{M}}{2}+\sum_{\gamma=1}^{q}\left\langle u,n_{\gamma
}\right\rangle \frac{[n_{\gamma},e_{j}]_{M}}{2}=0\text{.}$$
We now compute the first part $\sum\left\langle u,e_{\nu}\right\rangle
\left\langle \nabla_{e_{\nu}}\nabla_{e_{j}}V,e_{j}\right\rangle $ of the third term. Differentiating equation along $e_{\nu}$, we get $$\begin{aligned}
\nabla_{e_{\nu}}\nabla_{e_{j}}V(x) & =\frac{1}{2}\,\left\langle
u,\partial_{e_{j}}\beta_{\mu}\right\rangle [e_{\nu},\beta_{\mu}]_{M}+\frac
{1}{2}\,\left\langle u,\partial_{e_{\nu}}\beta_{\mu}\right\rangle [e_{j}
,\beta_{\mu}]_{M}\\
& +\left\langle u,\partial_{e_{\nu}}\partial_{e_{j}}e_{\alpha}\right\rangle
e_{\alpha}+\frac{1}{4}\,\left\langle u,e_{\alpha}\right\rangle [e_{\nu}
,[e_{j},e_{\alpha}]_{M}]_{M}\\
& +\left\langle u,\partial_{e_{\nu}}\partial_{e_{j}}n_{\gamma}\right\rangle
n_{\gamma}+\frac{1}{4}\,\left\langle u,n_{\gamma}\right\rangle [e_{\nu}
,[e_{j},n_{\gamma}]_{M}]_{M}\text{.}$$
Using the identity $\left\langle [X,Y]_{M},Z\right\rangle =-\left\langle
Y,[X,Z]_{M}\right\rangle $ for Killing vector fields $X,Y,Z$, together with the relation repeatedly, we get $$\left\langle \nabla_{e_{\nu}}\nabla_{e_{j}}V(x),e_{j}\right\rangle
=\left\langle u,\partial_{e_{\nu}}\partial_{e_{j}}e_{j}\right\rangle$$ and so $$\begin{aligned}
& \sum_{u}\sum_{j=1}^{p}\sum_{\nu=1}^{p}\left\langle u,e_{\nu}\right\rangle
\left\langle \nabla_{e_{\nu}}\nabla_{e_{j}}V,e_{j}\right\rangle
\nonumber\label{part1}\\
& =\sum_{u}\sum_{j=1}^{p}\sum_{\nu=1}^{p}\left\langle u,e_{\nu}\right\rangle
\left\langle u,\partial_{e_{\nu}}\partial_{e_{j}}e_{j}\right\rangle
\nonumber\\
& =\sum_{j,\nu=1}^{p}\left\langle \partial_{e_{\nu}}\partial_{e_{j}}
e_{j},e_{\nu}\right\rangle \nonumber\\
& =-\left\Vert \sum_{j=1}^{p}\,\mathrm{I}\mathrm{I}(e_{j},e_{j})\right\Vert
^{2}\text{.}$$
Now proceed to compute the second part $\sum\left\langle u,n_{\gamma
}\right\rangle \left\langle \nabla_{n_{\gamma}}\nabla_{e_{j}}V,e_{j}
\right\rangle $ of the third term. Differentiating the equation along $n_{\gamma}$, we get $$\begin{aligned}
\nabla_{n_{\gamma}}\nabla_{e_{j}}V(x) & =\frac{1}{2}\,\left\langle
u,\partial_{e_{j}}\beta_{\mu}\right\rangle [n_{\gamma},\beta_{\mu}]_{M}
+\frac{1}{2}\,\left\langle u,\partial_{n_{\gamma}}\beta_{\mu}\right\rangle
[e_{j},\beta_{\mu}]_{M}\\
& +\left\langle u,\partial_{n_{\gamma}}\partial_{e_{j}}e_{\nu}\right\rangle
e_{\nu}+\frac{1}{4}\,\left\langle u,e_{\nu}\right\rangle [n_{\gamma}
,[e_{j},e_{\nu}]_{M}]_{M}\\
& +\left\langle u,\partial_{n_{\gamma}}\partial_{e_{j}}n_{\alpha
}\right\rangle n_{\alpha}+\frac{1}{4}\,\left\langle u,n_{\alpha}\right\rangle
[n_{\gamma},[e_{j},n_{\alpha}]_{M}]_{M}$$ and so $$\left\langle \nabla_{n_{\gamma}}\nabla_{e_{j}}V(x),e_{j}\right\rangle
=\left\langle u,\partial_{n_{\gamma}}\partial_{e_{j}}e_{j}\right\rangle
\text{.}$$ $$\begin{aligned}
& \sum_{u}\sum_{j=1}^{p}\sum_{\gamma=1}^{q}\left\langle u,n_{\gamma
}\right\rangle \left\langle \nabla_{n_{\gamma}}\nabla_{e_{j}}V,e_{j}
\right\rangle \nonumber\label{part2}\\
& =\sum_{j,\gamma=1}^{p,q}\left\langle \partial_{n_{\gamma}}\partial_{e_{j}
}e_{j},n_{\gamma}\right\rangle \nonumber\\
& =-\sum_{j,\gamma=1}^{p,q}\left\langle \,\mathrm{I}\mathrm{I}(e_{j}
,e_{j}),\,\mathrm{I}\mathrm{I}(n_{\gamma},n_{\gamma})\right\rangle \text{.}$$
Adding up equations and , we get the third term $$-\left\Vert \sum_{j=1}^{p}\,\mathrm{I}\mathrm{I}(e_{j},e_{j})\right\Vert
^{2}-\sum_{j,\gamma=1}^{p,q}\left\langle \,\mathrm{I}\mathrm{I}(e_{j}
,e_{j}),\,\mathrm{I}\mathrm{I}(n_{\gamma},n_{\gamma})\right\rangle \text{.}$$
Adding up all the three terms, the average second variation is $$\sum_{j,k=1}^{p,q}\left( 2\,\Vert\,\mathrm{I}\mathrm{I}(e_{j},n_{k})\Vert
^{2}-\left\langle \,\mathrm{I}\mathrm{I}(e_{j},e_{j}),\,\mathrm{I}
\mathrm{I}(n_{k},n_{k})\right\rangle \right) \text{.\ \rule{5pt}{5pt}}$$
> **Acknowledgement:** The second author is partially supported by an RGC grant from the Hong Kong Government.
[^1]: All manifolds are connected compact oriented smooth manifolds.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In recent works [@BHP] (BHP), a generalized universality has been proposed, linking phenomena as dissimilar as 2D magnetism and turbulence. To test these ideas, we performed a MC study of the $2D XY$-model. We found that the shape of the probability distribution function for the magnetization $M$ is non Gaussian and independent of the system size –in the range of the lattice sizes studied– below the Kosterlitz-Thoules temperature. However, the shape of these distributions does depend on the temperature, contrarily to the BHP’s claim. This behavior is successfully explained by using an extended finite-size scaling analysis and the existence of bounds for $M$.'
address: |
Departamento de Física, Universidad de Santiago de Chile\
Casilla 307, Correo 2, Santiago, Chile
author:
- 'G. Palma, T. Meyer and R. Labbé'
date: Received
title: 'Finite size scaling in the 2D XY-model and generalized universality'
---
The study of critical phenomena is of great interest not only because it allows the understanding of a large number of very different physical systems, like the super fluid Helium three, low temperature superconductors, ferromagnetic-paramagnetic systems, turbulent fluids and plasmas, polymers, snow flakes and earthquakes, but also due to the existence of scale independence of the fluctuations at the critical temperature. In fact, although the underlying inter-molecular forces, responsible for the existence of phase transitions, have a well-defined length scale, the structures they give rise do not. This leads, very close to the critical temperature, to the power-law behavior of physical quantities, which characterizes universality. The main challenge of the theory of critical phenomena is to explain how dissimilar systems exhibit the same critical behavior. Renewed interest in this subject has been raised, because in a seminal paper [@BHP] (BHP), it was argued that turbulence experiments can be explained in terms of a self-similar structure of fluctuations, just as in a finite critical system like the harmonic finite $2D XY$-model ($2D HXY$-model). The starting point of this conjecture was the observation that the probability distribution function (PDF) of the injected power fluctuations in a confined shear turbulent flow [@Labbe] has the same shape as the PDF of the magnetization in the $2D HXY$-model. It was also proposed that this analogy should provide a new application of finite size scaling in critical systems with experimental consequences.
In this paper, we report the results of a high precision Monte Carlo study of the full $2D XY$-model. This computation was carried out over the whole physical range of temperatures. The magnetic susceptibility was computed and the lattice-shifted critical temperature was obtained for different lattice sizes. Scaling laws for the magnetization-temperature ratio were tested. Our results agree with the rigorous findings of Chung [@Chung]. We also found that, below the Berezinskii-Kosterlitz-Thouless temperature $T_{BKT}$, the shape of the PDF of the magnetization is non-Gaussian and independent of the lattice size, in agreement with previous results [@ABH]. However, we found that the shape of these distributions do depend on the temperature, contrarily to the generalized universality claimed in (BHP), who stated that the PDF of magnetization is independent of both, system size and temperature. Our results can be seen as a powerful extension of finite-size scaling and phenomenological renormalization of the PDF, suggested originally by Binder in the context of the Ising model [@Binder], with sligth modifications introduced by the bounds of $M$. This allows in particular to understand the scaling form of the PDF of the order parameter in the $2D XY$-model as well as in the turbulent system.
The universality proposed by (BHP) might go beyond the idea of equivalence classes in Wilson’s renormalization group approach [@Wilson], by including into a generalized universality class systems sharing the properties of finite size, strong correlations and self-similarity, even if their space-dimensions are different.
In [@ABH], the two-dimensional probability distribution for the magnetization is calculated by means of a Monte Carlo simulation in the context of the $2D HXY$-model. This model is a further simplification of the Villain model [@Villain], where the vortex variable $n$ is not a thermodynamical quantity, but it is constrained to the values $n = -1, 0, 1$. By using diagrammatic techniques, they showed that this asymmetry could be the result of three-spin interactions and higher order corrections.
Here, we consider the $2D XY$-model, which describes classical planar spins with nearest neighbor interactions, with a Hamiltonian given by
$$H=-J\sum_{\langle i,j\rangle }\cos (\theta _{i}-\theta _{j})
\label{Hamilton}$$
where $J$ is the ferromagnetic coupling constant and $\theta _{i}$ is the angle of orientation of the unitary spin vector $\overrightarrow{s}_{i}$. The summation $\langle i,j\rangle$ is over nearest neighbors and the spins are defined on the sites of a square lattice of lattice size $L$, with periodic boundary conditions. From hereon the ratio $k_{B}/J$ is set equal to unity throughout the paper. This model undergoes a remarkable binding-unbinding topological phase transition, such that the free energy and all its derivatives remain continuous [@BKT], and no long-range order at low temperatures exists, as stated by the Mermin-Wagner theorem [@MWagner]. This model has been extensively studied through both numerical and analytical methods [@XY-m].
Our simulation was performed on a square lattice of lattice sizes L=10, 12, 16, 22, and 32 respectively. We estimate the MC sweeps needed for thermalization by plotting some observables like magnetization and energy. Typically $10^{5}$ MC sweeps were used to reach thermal equilibrium. For thermal averages we used $5\times
10^{5}$ spin configurations $\alpha _{j}$. Because the $2D XY$-model has a continuous line of critical points below $T_{BKT}$, special care was taken to choose statistical independent configurations to evaluate thermal averages of physical observables $X$. This was achieved by computing its normalized autocorrelation function [@Binney]
$$C(K)=\frac{<X_{\alpha _{i}}X_{\alpha _{i+K}}>-<X_{\alpha _{i}}><X_{\alpha
_{K}}>}{<X_{\alpha _{i}}^{2}>-<X_{\alpha _{i}}>^{2}} \label{corr}$$
where $X_{\alpha _{i}}$ is the value of $X$ in the configuration $\alpha _{i}$ at the $i-th$ step along the MC-path through the configuration space, and the average $<\dots>$ was taken over this particular path of configurations separated by $K$ steps from each other. $C(K)=1$ for $K=0$, but for large enough $K,C(K)$ drops to zero, which means that these configurations become totally uncorrelated. We choose $K$ so that $C(K)$ was less that the recommended value $0.05$ [@Sokal]. It is well known that as a critical system approaches the critical temperature, the decorrelation time $\tau $ diverges with the power law $\tau \sim \xi ^{z}$, where $\xi $ is the (divergent) correlation length of the system and $z$ is known as the dynamical critical exponent, which is approximately two for local-flip algorithms like the Metropolis algorithm. This phenomenon is known as critical slowing down [@swang]. This means in practice a serious limitation to numerical simulations of critical systems close to a critical point.
In fig. 1 we show MC data for the susceptibility for $L^{2}=256$ spins, as a function of the temperature. The peak occurs at the value $T_{C}(16)=1.15$, and corresponds to the temperature at which the correlation length equals L, which is the standard definition of the critical temperature of a finite system. We compute also the errors (standard deviations), which become larger as the critical temperature is approached. Another interesting feature of these errors is that they are larger below $T_{C}$. This can probably be explained because of the comparative larger correlations lengths in this region, which corresponds to a continuous line of critical points with temperature dependent exponents [@ID] in the infinite volume limit.
We computed the critical temperature for the lattice sizes L=10, 12, 16, 22, 32. For L=32 we found an effective transition temperature $T_{C}=1.08$, in agreement with the value obtained by [@BH93], where the linearized RG equations for the finite size scaling were used. Fig. 2 shows $T_{C}(L)$ as a function of $(\ln (L))^{-2}$. The values can be described by the finite size scaling formula [@Chung]
$$T_{c}(L) \approx T_{\infty }+\frac{\pi ^{2}}{4c(\ln L)^{2}} \label{scaltcri}$$
where $T_{\infty }$ is the extrapolated value of the critical temperature for infinite volume. Within a few percents of error, we found that the value of $T_{\infty }$ agrees with the seemingly exact value $0.892$ of the critical temperature $T_{BKT}$ of the Berezinskii-Kosterlitz-Thouless phase transition.
The ratio of the mean magnetization to critical temperature is plotted in fig. 3 as a function of $\ln (L)$. These values are compatible with a negative straight line, as suggested by [@ABH] in the context of the harmonic XY-model. The values closer to the origin have larger statistical errors probably because of the finite size effects, which are proportional to $\ln (L)$ [@Chung; @Pinn].
Finally it should be emphasized that we did not make use of scaling relations to define physical quantities, like the shifted BKT-temperature $T^{\ast }(L)$, or temperature at which the renormalized spin-wave stiffness becomes the universal value $2/\pi $ of the infinite system. This is because the use of the BKT theory beyond its confirmed validity needs at least justification, (the scaling region is defined by the inequality $\mid T-T_{BKT})/T_{BKT} \mid <10^{-2}$, where the renormalization group equations confidently apply [@Cardy]). In fact, we were able to obtain accurate values for thermal averages and test the scaling equations (see eqn. (\[scaltcri\]) and Figs. 2 and 3), in spite of the difficulties of numerical simulations due to the essential finite-size effects present in this model [@Chung] ($\ln (L)$ dependence of physical quantities), and the very narrow critical region.
The renewed interest in the PDF of fluctuations of magnetization, $M$, is a consequence of the observation that similarly shaped PDF arise in completely different systems. For instance, in [@Labbe] it was found that fluctuations of the injected power in a confined turbulent flow show the same behavior. In Fig. 4 a plot of $\sigma Q(M)$ as a function of $(M-<M>)/\sigma$ can be seen, for lattice sizes $L=16$ and $L=32$ at the same temperature $T=0.70$. Here, $Q$ is the PDF of $M$ and corresponds to $P_{L}(M)$ in the language of Binder (see discussion below eqn. (\[PDF-B\])). These curves have similar shapes like those found in turbulence experiments, but only within a reduced range of temperatures below $T_{BKT}$. These PDFs can be conveniently compared with the universal form $\Pi(y)=K(e^{b(y-s)-e^{b(y-s)}})$ proposed in [@BHP], by plotting the ratio $\sigma Q(M;T)/\Pi(M)$ vs. $(M - \langle M \rangle)/\sigma$. In Fig. 5, four of such plots are displayed for $T=0.40, 0.80, 0.95$ and $1.05$. The upper curves are successively multiplied by factors of 10 for clarity. As can be seen, when $T$ is increased, these ratios consistently change, showing the dependence of $Q(M;T)$ on the temperature. The rightmost part of the lower curves is raised, probably due in part to the upper bound $M=1$. On the other hand, for $T/=0.95$ we can see that the opposite occurs. This temperature is slightly greater than $T^{*}$, above which the population of spin vortex pairs begins to increase. This happens because the system volume occupied by these vortices no longer contributes to the magnetization, which leads to a depleted probability density. This can be appreciated in the central and rightmost parts of the two upper curves. The hills are due to the bounds in the magnetization $0<M<1$. At $T=1.05$ this effect is greatly enhanced, and the leftmost part of the curve shows even more clearly the effect of the lower bound $M=0$. Concerning turbulent flows, we do not expect this type of bounding effects in the statistics of injected power. In principle, there are no limits to the fluctuations of such a quantity, and negative values are not excluded, meaning that the flow is delivering power to the driving system. Although this type of events are expected to be very unlike, they are not forbidden.
The temperature dependence of the PDF is not a surprising result. In fact, the use of the probability distribution of the order parameter to study finite size scaling and phenomenological renormalization, has been discussed by Binder in the context of the Ising model [@Binder2]. For the region $\xi \sim L$ he proposed that the probability distribution function $P_{L}(M)$ does not depend separately on the three variables $\xi,L,M$, but only on the two scaled combinations, $L/\xi$ and $M\xi ^{\beta/\nu }$:
$$\label{PDF-B}
P_{L}(M)=\xi ^{\beta /\nu }\widetilde{P}(L/\xi ,M\xi ^{\beta /\nu
})=L^{\beta /\nu }P(L/\xi ,ML^{\beta /\nu }).$$
He also argued that in the critical region $\xi \gg L$, $P_{L}(M)$ is no longer Gaussian. In the scaling region, it is a good approximation to take $P_{L}(M)$ equal to the PDF proposed by Bramwell [*et al.*]{} (The standard deviation $\sigma $ plays the role of $L^{-\beta /\nu }$ in the BHP distribution; this can be seen by using the standard definitions of the critical exponents and the relation $\sigma =\sqrt{\frac{T}{L^{2}}\chi }$ ). Nevertheless, and away from the region defined by $\mid T-T_{BKT} \mid /T_{BKT}<10^{-2}$, there is a temperature dependence in expresion (4) via the correlation length $\xi$ for finite size. It turned out that this dependence is rather weak in the range $0.5<T<T^{*}(L)$ ($T^{*}(L=16) \sim 0.94 $), but out of this range this dependence becomes stronger due to the presence of vortices and/or bounds.
In conclusion, we found that the probability distribution function for the magnetization is indeed independent of the system size, but its shape happens to vary with the temperature of the system, contrarily to the generalized universality proposed by Bramwell [*et al.*]{} [@BHP]. This effect comes from the intrinsic temperature dependence on the first scaled variable $L/\xi$ in the distribution function proposed by Binder for the order parameter $M$. Also, there is a contribution coming from the constrained character of the magnetization.
We thank L. Vergara for helpful suggestions. This work was supported in part through projects FONDECYT 1980608 and 1990169, and DICYT 04-9631PA and 04-9631LM.
S.T. Bramwell, P.C.W. Holdsworth and J.-F. Pinton, Nature [**396**]{} (1998) 552. S.T.Bramwell [*et al.*]{}, Phys. Rev. Lett. [**84**]{} (2000) 3744.
R. Labbé, J.-F. Pinton and S. Fauve,Journal de Physique II France, [**6**]{} (1996) 1099.
S.G. Chung, Phys. Rev. B [**60**]{}, (1999), 11761.
P. Archambault, S.T. Bramwell, P.C.W. Holdsworth, J. Phys. A:Math. Gen. [**30**]{} (1997) 8363.\
P. Archambault,S.T. Bramwell,J.-Y. Fortin, P.C.W. Holdsworth, S. Peysson and J.-F. Pinton, J. Appl. Phys. [**83**]{} (1998) 7234.
K. Binder, [*Computational Methods in Field Theory*]{}, eds. H. Gauslever and C.B. Lang, Springer (Berlin), (1992).
K.G. Wilson and J. Kogut, Phys. Report [**12C**]{}, (1974) 75.\
K.G. Wilson, Rev. Mod. Phys. [**47**]{}, (1975) 773.
J. Villain, J. Physique [**36**]{}, (1975) 581.
V.L. Berezinskii, Sov. Phys. JETP [**32**]{} (1971) 493.\
J.M.Kosterlitz and D.J.Thouless, J. Phys. C: Solid State Phys. [**6**]{} (1973) 1181-1203.
N. D. Mermin and H. Wagner, Phys. Rev. Lett. [**17**]{} (1966) 1133.
J. Tobochnik and G.V. Chester, Phys. Rev. [**B20**]{} (1979) 3761.\
R. Gupta and C.F. Baillie, Phys. Rev. [**B45**]{} (1992) 2883.\
J.R. Lee and S. Teitel, Phys. Rev. [**B46**]{} (1992) 3247.\
M. Campostrini, A. Pelisetto, P. Rossi and E. Vicari, Phys. Rev. B [**54**]{} (1996) 7301.\
Z. Gulácsi and M. Gulácsi, Advances in Physics [**47**]{}, (1998) 1-89.\
J.V. José,L.P. Kadanoff, S. Kirkpatrick and D.R. Nelson, Phys. Rev. [**B16**]{}, (1977) 1217.
J. J. Binney, N. J. Dowrick, A. J. Fisher and M. E. J. Newman,[*The Theory of Critical Phenomena*]{}, Oxford Science Pub. (1993).
A. Sokal, Nucl. Phys. B [**20**]{} (Proc. Suppl.) (1991), 55.
R. Swendsen and J.-S. Wang, Phys. Rev. Lett., [**58**]{}, (1987) 86.\
U. Wolff, Nucl. Phys. Proc. Suppl. 17 (1990) 93.\
A. Sokal, Nucl. Phys. Proc. Suppl. 20 (1991) 55.
J. Cardy, [*Scaling and Renormalization in statistical physics*]{}, Cambridge University Press, (1996).\
C. Itzikson and J-M.Drouffe, [*Statistical Field Theory*]{}, Vol.1,Cambridge Monographs on Mathematical Physics, (1992).
S.T. Bramwell and P.C.W. Holdsworth, J. Phys. : Condens. Matter [**5**]{}, (1993) L53.
M.Hasenbusch and K.Pinn, J.Phys. A: Math Gen. [**30**]{} (1997) 63.\
R. Kenna and A. C. Irving, Phys. Lett., B 351 (1995) 273.\
J.-K. Kim, Adauto F. De Souza, D. P. Landau, F. Phys. Rev. E 54 (1996) 2291.\
W. Janke, Phys. Lett. A [**173**]{} (1993) 8.
J. Cardy, Phys. Rev. [**B26**]{}, (1982) 6311.
K. Binder, Z. Phys. [**V43**]{} (1981) 119.\
K. Binder, Phys. Rev. Lett. [**47**]{} (1981) 693.
[**Figure captions**]{}
[**Figure 1**]{}:Susceptibility for $L=16$ in the range $0 < T \leq 3$. The peak at $1.15$ corresponds to the shifted critical temperature.
[**Figure 2**]{}: The shifted critical temperature for different lattice sizes is plotted as a function of the system size.
[**Figure 3**]{}: Scaling relation for the magnetization-temperature ratio as a function of system size.
[**Figure 4**]{}: Plots of $\sigma Q(M)$ vs. $(M-<M>)/\sigma $ at $T=0.70$ for lattice sizes $L=16$ (+) and $L=32$ (\*).
[**Figure 5**]{}: $\sigma Q(M)/\Pi(M)$ ratios for four values of temperature (see text).
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We study the non-equilibrium dynamics of [the Luttinger model]{} after a quantum quench, when the initial state is a finite temperature thermal equilibrium state. [The diagonal elements of the density matrix in the steady state show thermal features for high temperature initial states only, otherwise retain highly non-thermal character. The time evolution of Uhlmann fidelity, which measures the distance between the time evolved and initial states, is evaluated for arbitrary initial temperatures and quench protocols. ]{} In the long time limit, the overlap between the time evolved and initial system decreases [exponentially with the temperature with a universal prefactor.]{} Within perturbation theory, the statistics of final total energy and work are numerically evaluated in the case of a sudden quench[, which yield identical distributions at zero temperature]{}. In both statistics, temperature effects are more significant in small systems. [The Dirac-delta peak at the adiabatic ground state energy remains present in the probability distribution of the total energy, but disappears from the work distribution at non-zero initial temperatures.]{}'
author:
- Ádám Bácsi
- Balázs Dóra
bibliography:
- 'fintempstatworklutt4.bib'
title: 'Quantum quench in [the Luttinger model]{} with finite temperature initial state'
---
Introduction
============
Quantum quenches have been attracting lots of interest due to their experimentally controllable realizations in cold atomic systems [@greinernat; @toshiyanat; @hofferberthnat]. Recent experiments allow one to investigate quantum dynamics of closed systems and to perform quantum quenches by modulating the parameters of the system [@blochrmp; @dziarmagareview]. The quench drives the system out of equilibrium, raising interesting questions about how the closed system equilibrates after the quench, if at all, and how the long-time behaviour, i. e. the steady state, can be described.[@rigoltherm2; @polkovnikovrmp; @rigoltherm; @rigoleth]
[The characterization of the steady state can be given by determining all the diagonal density matrix elements in the representation of eigenstates of the final Hamiltonian. Off-diagonal elements do not contribute to the expectation value of physical observables in the steady state due to dephasing. This is the concept of the diagonal ensemble[@rigolgge; @rigolnat; @rigoltherm2; @ggecaux], which also enables us to determine the probability distribution of any constants of motion in the steady state.]{}
[One-dimensional strongly correlated systems often form a Luttinger liquid (LL), made of bosonic sound-like collective excitations, regardless to the statistics of the original system. Such phases have already been created out of cold atoms[@hofferberthnat; @cazalillarmp]. However, it is not entirely clear whether the LL universality class can be extended to a nonequilibrium situation, though combined numerical and analytical studies indicate that this is indeed the case [@medenprl; @balazslattnum; @doraloschmidt]. The non-equilibrium dynamics of the Luttinger model (LM), describing the low energy physics in LLs, has been studied extensively [@cazalillaprl; @cazalillapra; @balazsprl; @perfetto; @mitra; @mitra1; @nessiprb].]{} In our previous work, we have also investigated the statistics of work done on a LL – prepared initially in the ground state – during quantum quenches with different duration [@ztwork].
Finite temperature effects in thermalization[@rigolfintemp], correlation functions and the momentum distribution function [@cazalillaprl; @cazalillapra; @tylutki] have already been investigated. In this paper, we study how the finite temperature modifies the time evolved density matrix and the diagonal ensemble if the [[system]{}]{} is initially at thermal equilibrium, described by a canonical ensemble.
The time evolved state can be characterized by calculating the fidelity (or the Loschmidt echo) which measures the overlap with the initial thermal equilibrium state [@zanardipra; @zanardi; @damskiprl; @zanardifintemp]. [[The fidelity is an important quantity in various fields of physics ranging from nuclear physics to quantum information theory [@Gorin200633; @goussevscholar] and also provides direct insight to dynamical properties of the quantum system without reference to any particular physical quantity.]{}]{} In our model the final Hamiltonian does not commute with the initial density operator, therefore, the fidelity is expected to have explicit time-dependence [[with equilibration in the steady state. How the time evolution and the long time behaviour depend on the initial temperature is one of the major concerns of the present work.]{}]{}
With zero initial temperature, i. e., if the initial state is the pure ground state of the Hamiltonian, the statistics of total energy in the final state and the statistics of work done on the system are basically the same [@silvaqcp; @ztwork]. At finite temperature, however, these distributions differ from each other because the initial energy is not well defined. Moreover, the initial energy may be arbitrarily large, therefore, the probability distribution function (PDF) of work has no lower bound[@hanggirmp; @jarzynski]. In this paper our goal is to explore finite temperature effects in the statistics of total energy and work after a sudden quench (SQ).
The present article is organized as follows. After introducing the model and the time evolution during a quantum quench, we determine the diagonal ensemble for an arbitrary quench protocol and temperature in Section \[sec:de\]. We derive exact analytical expressions for the Loschmidt echo in Section \[sec:fidmain\]. The long time limit of the Loschmidt echo is numerically evaluated for SQs. In Section \[sec:tepdf\] we study the statistics of the total energy in the SQ limit and within perturbation theory for weak interaction strength. The generating function of the distribution is obtained analytically while the PDFs are evaluated numerically with low initial temperature. In Section \[sec:wpdf\] the statistics of work is determined in the SQ limit.
Time evolution during the quench
================================
[[We study the time evolution of the LM described by the time-dependent bosonic Hamiltonian]{}]{} $$\begin{gathered}
\hat{H}(t)=\hat{H}_{0}+Q(t)\hat{V}, \label{hamilton}\\
\hat{H}_{0}=\sum_{q>0}\omega_{0}(q)\left(b^{+}_{q}b^{}_{q}+b^{+}_{-q}b^{}_{-q}\right),\\
\hat{V}=\sum_{q>0}\delta\omega(q)\left(b^{+}_{q}b^{}_{q}+b^{+}_{-q}b^{}_{-q}\right)+g(q)\left(b^{+}_{q}b^{+}_{-q}+b^{}_{q}b^{}_{-q}\right),\end{gathered}$$ where $\omega_{0}(q)=vq$ is the non-interacting dispersion of bosons, $\delta\omega(q)=\delta v q$ comes from velocity renormalization and $g(q)=g_{2}q\exp{(-R_{0}q)}$ is the interaction strength with $R_{0}$ characterizing the finite range of the interaction[@giamarchi; @nersesyan]. In the following we use the notation $\omega(q,t)=\omega_{0}(q)+Q(t)\delta\omega(q)$ and $g(q,t)=Q(t)g(q)$. The quench protocol $Q(t)$ vanishes for $t<0$ and equals $1$ for $t>\tau$ with $\tau$ denoting the quench duration.
[In equilibrium, the LM describes successfully the low energy dynamics of LLs. In a non-equilibrium situation, additional processes, which are termed irrelevant in equilibrium, are inevitably present in lattice models and can still play an important role. To understand the applicability of the LM in non-equilibrium situation, several lattice models have been tested and investigated by comparing numerically exact calculations with analytical results using bosonization [@medenprl; @balazslattnum; @doraloschmidt; @kennes]. These exhibit convincing agreement in all the examined cases.]{}
After the quench, the Hamiltonian $\hat{H}(\tau)=\hat{H}_{0}+\hat{V}$ can be diagonalized by standard Bogoliubov transformation, leading to $$\begin{gathered}
\hat{H}(\tau)=E_{\mathrm{ad}}+\sum_{q>0}\Omega_{q}\left(d^{+}_{q}d_{q}+d^{+}_{-q}d_{-q}\right),\end{gathered}$$ where $\Omega(q)=\sqrt{\omega(q,\tau)^{2}-g(q)^{2}}<\omega(q,\tau)$ is the quasi-particle dispersion and $E_{\mathrm{ad}}=\sum_{q>0}\left(\Omega(q)-\omega(q,\tau)\right)<0$ is the ground state energy of the final Hamiltonian. The annihilation operators $b_{\pm q}$ is expressed with the new bosonic quasi-particle operators as $$\begin{gathered}
b_{\pm q}=\frac{\sqrt{\omega(q,\tau)+\Omega(q)}\,d_{\pm q}-\sqrt{\omega(q,\tau)-\Omega(q)}\,d_{\mp q}^{+}}{\sqrt{2\Omega(q)}}.\end{gathered}$$ We now focus on the time evolution of the density operator. The initial state is considered the finite temperature equilibrium state, $\hat{\rho}_{0}=\exp{(-\beta \hat {H}_{0})}/Z_{0}$, where $\beta=1/T$ is the inverse temperature.
The coupling between the system and reservoir is assumed to be so small, that the relaxation time of thermalization is much longer than the time-scale of the experiment. Energy exchange between the system and the environment is neglected apart from the energy change due to the quench. Therefore, the time evolution is driven by the time dependent Schrödinger equation and can be transferred to the Bogoliubov coefficients defined through the time dependent creation and annihilation operators $$\label{eq:btime}
b_{\pm q}(t)=u_{q}(t)b_{\pm q}+v_{q}^{*}(t)b_{\mp q}^{+}\,,$$ where the $b$ bosons on the r.h.s. are those before the quench. The coefficients are determined from Heisenberg’s equation of motion and obey $$\label{eq:tdschr}
i\partial_{t}\left[\begin{array}{c} u_{q}(t) \\ v_{q}(t) \end{array}\right]=\left[\begin{array}{cc} \omega(q,t) & g(q,t) \\ -g(q,t) & -\omega(q,t) \end{array}\right]\left[\begin{array}{c} u_{q}(t) \\ v_{q}(t) \end{array}\right]$$ with the initial conditions $u_{q}(0)=1$ and $v_{q}(0)=0$ and $|u_{q}(t)|^{2}-|v_{q}(t)|^{2}=1$. In the next section, we discuss the result of a general quench protocol on the diagonal ensemble, while in Sections IV and V we will focus on the SQ limit for the sake of simplicity.
Density operator and diagonal ensemble {#sec:de}
======================================
The density matrix $\hat{\rho}(\tau)$ after the quench can be given exactly in second quantized formalism by means of $u_{q}(\tau)$ and $v_{q}(\tau)$. Using the exact expression (given in Appendix \[sec:Fxi\]), we derive the diagonal elements of the density matrix which are essential to describe the steady state in the long time limit.
Since the various $q>0$ momentum modes are completely decoupled, as seen in Eq. , the density operator is block diagonal in momentum representation, and we consider only a single $q>0$ channel in this section. The resulting Hamiltonian reduces to two coupled harmonic oscillators with the same frequency $\omega_{0}$. Our results can easily be generalized to all channels by taking the product of the density operators of all modes. We also drop the subscript $q$ and the indices $q$ and $-q$ are replaced by $+$ and $-$, respectively (e.g. $d_{q}$ will henceforth be denoted with $d_{+}$).
After the quench the integrals of motion are $\hat{n}_{+}=d^{+}_{+}d_{+}$ and $\hat{n}_{-}=d^{+}_{-}d_{-}$. Of course, their products and linear combinations are also preserved. The Hilbert space is spanned by the occupation number eigenstates $|n_{+},n_{-}\rangle$ in which the number of $d_{\pm}$ bosons is $n_{\pm}$.
In order to determine the diagonal ensemble, we calculate all the diagonal matrix elements $$\label{eq:densme}
\rho(n_{+},n_{-}):=\langle n_{+},n_{-}|\hat{\rho}(\tau)|n_{+},n_{-}\rangle\,,$$ which give the probability distribution of occupation numbers.
Off-diagonal elements, such as $\langle n'_{+},n'_{-}|\hat{\rho}(\tau)|n_{+},n_{-}\rangle$ which are non-zero only if $n_{+}-n_{-}=n'_{+}-n'_{-}$, are important only if the time evolution of a non-preserved quantity is studied. In the next section, for example, the fidelity will be such a quantity since $\hat{H}_{0}$ is not an integral of motion after the quench. In the long time limit, however, contributions from off-diagonal elements vanish due to dephasing which is a consequence of the continuous spectrum $\Omega(q)$[@barthel].
The generating function of the occupation number distribution, Eq. is obtained as $$\label{eq:fdef}
f(\xi_{+},\xi_{-})=\mathrm{Tr}\,\left[\hat{\rho}(\tau)e^{i(\xi_{+}\hat{n}_{+}+\xi_{-}\hat{n}_{-})}\right]$$ for all $\xi_{+}$ and $\xi_{-}$. The expectation value of all integrals of motion can be calculated by taking the derivatives of $f(\xi_{+},\xi_{-})$ with respect to $\xi_{+}$ and $\xi_{-}$. For instance, $\langle
\hat{n}_{+}\rangle=-i\partial_{\xi_{+}}\left.f(\xi_{+},\xi_{-})\right|_{\xi_{+}=0,\xi_{-}=0}$ in the steady state. We will see in Section \[sec:tepdf\] that Eq. is very useful in determining the generating function of the probability distribution of the total energy as well. It is important to realize that both the expectation value and the whole distribution of any integrals of motion can be deduced from $f(\xi_{+},\xi_{-})$.
a\) ![Diagonal matrix elements of the density operator $\langle n_{+},n_{-}|\hat{\rho}(\tau)|n_{+},n_{-}\rangle$. In the figures $n_{+}$ and $n_{-}$ are measured on the horizontal and the vertical axes. The quantity $a(\tau)$, which characterizes the quench protocol and is independent from the temperature, is $2$ for all figures. The temperature varies such that the initial occupation number is a) $n_{0}=0$ zero temperature b) $n_{0}=1$ c) $n_{0}=30$ high temperature. Colors do not represent the same values in different figures and only illustrate the structure of diagonal matrix elements.[]{data-label="fig:denm"}](densmp0.eps "fig:") b) ![Diagonal matrix elements of the density operator $\langle n_{+},n_{-}|\hat{\rho}(\tau)|n_{+},n_{-}\rangle$. In the figures $n_{+}$ and $n_{-}$ are measured on the horizontal and the vertical axes. The quantity $a(\tau)$, which characterizes the quench protocol and is independent from the temperature, is $2$ for all figures. The temperature varies such that the initial occupation number is a) $n_{0}=0$ zero temperature b) $n_{0}=1$ c) $n_{0}=30$ high temperature. Colors do not represent the same values in different figures and only illustrate the structure of diagonal matrix elements.[]{data-label="fig:denm"}](densmp2.eps "fig:") c)![Diagonal matrix elements of the density operator $\langle n_{+},n_{-}|\hat{\rho}(\tau)|n_{+},n_{-}\rangle$. In the figures $n_{+}$ and $n_{-}$ are measured on the horizontal and the vertical axes. The quantity $a(\tau)$, which characterizes the quench protocol and is independent from the temperature, is $2$ for all figures. The temperature varies such that the initial occupation number is a) $n_{0}=0$ zero temperature b) $n_{0}=1$ c) $n_{0}=30$ high temperature. Colors do not represent the same values in different figures and only illustrate the structure of diagonal matrix elements.[]{data-label="fig:denm"}](densmp3.eps "fig:")
The function $f(\xi_{+},\xi_{-})$ is obtained for an arbitrary quench protocol and arbitrary temperature analytically. The detailed calculation is given in Appendix A and results in $$\begin{gathered}
f(\xi_{+},\xi_{-})=\Big[1+n(\tau)\left(1-e^{i(\xi_{+}+\xi_{-})}\right)+\nonumber \\
\label{eq:Fpm}
+\left(n_{0}+n_{0}^{2}\right)\left(e^{i\xi_{+}}-1\right)\left(e^{i\xi_{-}}-1\right)\Big]^{-1}\end{gathered}$$ where $n_{0}=(e^{\beta\omega_{0}}-1)^{-1}$ is the expectation value of the occupation number in the initial state and $$\begin{gathered}
n(\tau):=\mathrm{Tr}\left[\hat{\rho}(\tau)\hat{n}_{\pm}\right]=a(\tau)n_{0}+\frac{a(\tau)-1}{2}\end{gathered}$$ is the expectation value of the occupation number after the quench[@tylutki]. The real, [temperature independent]{} quantity $$\label{eq:atau}
a(\tau)=\frac{\omega(\tau)}{\Omega}\left(1+2|v(\tau)|^{2}\right)+2\frac{g}{\Omega}\textmd{Re}\left(u(\tau)^{*}v(\tau)\right)$$ characterizes the quench protocol and does not depend on time for $t>\tau$ since it is related to $n(\tau)$, being the expectation value of the preserved quantity $\hat{n}_{\pm}$. [We note that, since $a(\tau)\geq 1$, the average occupation number after the quench is larger than before, i.e. during the time evolution more bosons are created than annihilated on average. Moreover, the difference $n(\tau)-n_{0}$ grows as the initial temperature increases.]{}
The diagonal matrix elements are obtained by Fourier transforming Eq. . This is carried out analytically by introducing the complex variables $z_{\pm}=e^{-i\xi_{\pm}}$. Then, complex integrals provide the matrix elements $$\begin{aligned}
\label{eq:temp}
\rho(n_{+},n_{-})=\frac{(n_{0}+n_{0}^{2})^{n_{+}-n_{-}}\left(n(\tau)-n_{0}-n_{0}^{2}\right)^{n_{-}}}{\left(1+n(\tau)+n_{0}+n_{0}^{2}\right)^{n_{+}+1}}\nonumber\\
\times\sum_{l=0}^{n_{-}}\frac{\left(\begin{array}{c} n_{-} \\ l \end{array}\right)\left(\begin{array}{c} n_{+}+l \\ n_{-} \end{array}\right)\left(n_{0}+n_{0}^{2}\right)^{2l}}
{\left(n(\tau)-n_{0}-n_{0}^{2}\right)^{l}\left(1+n(\tau)+n_{0}+n_{0}^{2}\right)^{l}}\end{aligned}$$ for $n_{+}\geq n_{-}$. The opposite case is obtained from $\rho(n_{+},n_{-})=\rho(n_{-},n_{+})$. [[We emphasize again that Eq. is exact for arbitrary quench protocol and initial temperature. All information about the quench is incorporated into the expectation value of the occupation number $n(\tau)$.]{}]{}
[Let’s start to analyze our results at zero temperatures first.]{} Only the $l=0$ term is finite in the sum in Eq. and only the $n_{+}=n_{-}$ matrix element survives, meaning that the number of bosons in the $+q$ and $-q$ modes are the same. This behaviour stems from the fact that the difference $\hat{n}_{+}-\hat{n}_{-}$ is preserved during the time evolution and its expectation value is zero in the initial state[@ztwork]. [The density matrix is highly non-thermal.]{}
In the finite temperature initial state, however, this difference may be nonzero and, therefore, $\rho(n_{+},n_{-}\neq n_{+})$ elements show up in the final state. Diagonal elements of the density matrix are illustrated in Fig. \[fig:denm\] with different initial temperatures.
At low temperature or for high frequencies ($\beta\omega_{0}\gg 1$) the occupation number $n_{0}$ is exponentially small, $n_{0}\approx e^{-\beta\omega_{0}}$ to leading order. Matrix elements up to first order in $n_{0}$ are non-vanishing only if $n_{+}=n_{-}$ or $n_{+}=n_{-}\pm 1$, [and the corrections to the zero temperature case are given by]{} $$\begin{gathered}
\rho(n_{+},n_{-})=\frac{2(a(\tau)-1)^{n_{+}}}{(a(\tau)+1)^{n_{+}+1}}
\begin{cases} 1-2n_{0} & \mathrm{if}\, n_{+}=n_{-} \\ {\displaystyle\frac{2n_{0} n_{\pm}}{a(\tau)\mp 1}} & \mathrm{if}\,\, n_{+}=n_{-}\pm 1 \end{cases},\end{gathered}$$ which is highly non-thermal again.
[Finally,]{} at high temperature or for low frequencies ($\beta\omega_{0}\ll 1$) the initial occupation number is large ($n_{0}\gg 1$). If $n_{0}\gg a(\tau)$ also holds, the elements of the density matrix are written as $$\begin{gathered}
\rho(n_{+},n_{-})\approx\frac{1}{n_{0}^{2}} \exp{\left(-(n_{+}+n_{-})\frac{a(\tau)}{n_{0}}\right)},\end{gathered}$$ which resemble to a thermal density matrix. Note that the trace of this approximate matrix does not yield $1$ due to the high temperature approximation.
We emphasize again that only a single $q>0$ mode was considered in this section. All the modes should be taken into account when certain physical quantities, e.g. total energy, are evaluated.
Loschmidt echo {#sec:fidmain}
==============
Here we investigate the question of how much the time evolved state described by $\hat{\rho}(t)$ differs from the initial state $\hat{\rho}_{0}$ for $t>\tau$. The physical quantity measuring the “similarity” of these states, i.e. the overlap of the two density operators, is the fidelity or sometimes called Loschmidt echo [@zanardi; @zanardifintemp]. Since the initial Hamiltonian is not a constant of motion after the quench, the fidelity has explicit time-dependence [@aperes].
[The distinguishability of quantum states is measured by means of the so-called Uhlmann fidelity [@jozsauhlmann; @zanardi; @nielsen], which is defined as $$\begin{gathered}
F_{U}(t)=\mathrm{Tr}\left[\sqrt{\hat{\rho}^{1/2}_{0} \hat{\rho}(t)\hat{\rho}^{1/2}_{0}}\right]\,.
\label{uhlmann}\end{gathered}$$ The fidelity is symmetric with respect to its arguments and $0\leq F_{U}(t)\leq 1$ always holds where the latter relation becomes equality in the case of identical density operators. The Uhlmann fidelity is related to the Bures metric in which the angle between the two density matrices is given by the angle $\arccos F_{U}(t)$. Since the density matrices are normalized to unity, their angle can be used to quantify their distance. In the case of pure states, the Uhlmann fidelity simplifies to the absolute value of the overlap between the wavefunctions. For instance, with zero initial temperature the fidelity yields $F_{U}(t)=|\langle\Psi(t)|\Psi_{0}\rangle|$ where $|\Psi_{0}\rangle$ is the ground state of the initial Hamiltonian and $|\Psi(t)\rangle$ is the time evolved wavefunction. At finite initial temperature, evaluation of the trace provides $$\label{eq:uhlmannfid}
\ln F_{U}(t)=\sum_{q>0}\ln\frac{\cosh(\beta\omega_{0}(q))-1}{\sqrt{1+|u_{q}(t)|^{2}\sinh^{2}(\beta\omega_{0}(q))}-1}$$ where $u_{q}(t)$ is the Bogoliubov coefficient defined in Eq. . For technical details, see Appendix \[sec:fid\]. The resulting expression in Eq. shows that the Loschmidt echo depends remarkably on the initial temperature. This property seems to be natural but if we choose the Frobenius norm instead of the Bures metric, no initial temperature dependence is found. ]{}
[Let us briefly mention that Eq. represents the trace norm of the operator $\hat{\rho}^{1/2}_{0} \hat{\rho}^{1/2}(t)$. However, one can also use the Frobenius norm [@zanardi] instead, which also coincides with the finite temperature generalization of the Loschmidt echo as given in the pioneering paper by A. Peres[@aperes]. In this case the overlap of the time-evolved and initial states is given by $$\begin{gathered}
\label{eq:ffdef}
F_{F}(t)=\sqrt{\mathrm{Tr}\left[\hat{\rho}(t)\hat{\rho}_{0}\right]}\end{gathered}$$ and $$\label{eq:frobfid}\ln F_{F}(t)=\sum_{q>0}\ln\frac{\cosh(\beta\omega_{0}(q))-1}{|u_{q}(t)|\sinh(\beta\omega_{0}(q))}\,.$$ Eq. does not necessarily yield $1$ in the case of identical operators. Therefore, the Loschmidt echo is normalized by the square root of the so called effective dimension[@zanardi] $d_{\mathrm{eff}}=1/\mathrm{Tr}[\rho_{0}^{2}]$. This normalization leads to $$\label{eq:ffnorm}
\ln\big(\sqrt{d_{\mathrm{eff}}}F_{F}(t)\big)=-\sum_{q>0}\ln\left|u_{q}(t)\right|$$ which leads to the rather counterintuitive result that the Loschmidt echo using the Frobenius norm does not depend on the initial temperature. This means that the Uhlmann fidelity, as used in quantum information theory, enables us to distinguish the time-evolved and initial states with finite temperature in a more delicate way than using the Frobenius norm.]{}
Eqs. , and fulfill the inequality $$\begin{gathered}
F_{F}(t)\leq F_{U}(t)\leq \sqrt{d_{\mathrm{eff}}}F_{F}(t),\end{gathered}$$ where the first relations holds true in general, while the second inequality is specific to bosonic systems, and is reversed for fermions [@zanardi]. The zero temperature limit of the Uhlmann fidelity yields the normalized Frobenius fidelity.
Eqs. , and are the main results for the finite temperature Loschmidt echo, valid for arbitrary temperature, quench protocol and interaction strength for quadratic bosonic Hamiltonians. Previous studies of the fidelity at finite temperatures focused on fermionic systems[@zanardi], though the fidelity susceptibility for bosons was also considered[@sirker].
In the following of this section, we investigate the Uhlmann fidelity in special cases. In the SQ limit ($\tau\rightarrow 0$), the Bogoliubov coefficient is obtained as $$\begin{gathered}
u_{q}(t)=\cos(\Omega(q)t)-i\frac{\omega(q,\tau)}{\Omega(q)}\sin(\Omega(q)t)\end{gathered}$$ for $t>0$. The time dependence of the Uhlmann fidelity is evaluated numerically, and the resulting function decreases monotonically but saturates to a non-zero value as shown in Fig. \[fig:fid\].a. Here, the velocity renormalization is neglected in this calculation because we are interested in interaction effects coming from a finite $g_{2}$. Figure \[fig:fid\].b shows the long time limit of the Uhlmann fidelity as a function of $g_{2}/v$. The final state deviates from the initial one with increasing temperature in the long time limit.
a)![a) Time evolution of the logarithm of the Loschmidt echo using the Bures metric for different temperatures following a sudden quench in units of $L/(4\pi R_{0})$. The interaction strength was chosen $g_{2}=0.4 v$ for the plot. b) Logarithm of the long time limit of the Loschmidt echo, in units of $L/(4\pi R_{0})$, as a function of $g_{2}/v$ (numerical results). In both figures, the blue dashed line corresponds to the zero temperature case. The yellow and red curves correspond to finite temperatures $\beta=10\tau_{0}$ and $\beta=4\tau_{0}$, respectively, where $\tau_{0}=R_{0}/v$ is related to the finite range of the interaction.[]{data-label="fig:fid"}](tdfid2.eps "fig:")\
\
b)![a) Time evolution of the logarithm of the Loschmidt echo using the Bures metric for different temperatures following a sudden quench in units of $L/(4\pi R_{0})$. The interaction strength was chosen $g_{2}=0.4 v$ for the plot. b) Logarithm of the long time limit of the Loschmidt echo, in units of $L/(4\pi R_{0})$, as a function of $g_{2}/v$ (numerical results). In both figures, the blue dashed line corresponds to the zero temperature case. The yellow and red curves correspond to finite temperatures $\beta=10\tau_{0}$ and $\beta=4\tau_{0}$, respectively, where $\tau_{0}=R_{0}/v$ is related to the finite range of the interaction.[]{data-label="fig:fid"}](ltlfid2.eps "fig:")
[Analytical results are obtained only within perturbation theory for small values of $g_{2}/v$, when Eq. is expanded in $|v_q(t)|$. At low temperatures and for a SQ, its long time value ($t\gg\beta\gg\tau_{0}$) is obtained as $$\begin{gathered}
\label{eq:fusq}
F_{U}(t\rightarrow\infty)=\exp\left(-\alpha-\frac{1}{16}\left(\frac{g_2}{v}\right)^2\frac{LT}{v}\right),\end{gathered}$$ where $\tau_{0}=R_{0}/v$ is the time scale corresponding to the finite range of the interaction, and the temperature dependent term possesses a universal prefactor in the exponent in the LL sense, namely that it is independent of the high energy cutoff, $1/\tau_0$. This universality is reminiscent of the universal term in the partition function and consequently in the specific heat of 1D critical quantum systems[@giamarchi]. Similar behaviour of the fidelity was reported in Ref. , where the finite temperature fidelity susceptibility was investigated between different LM ground states. Our result also shows that increasing temperature results in less fidelity. This behaviour is related to the fact that more bosons are created during the quench for higher initial temperatures. In Eq. ]{}, $\alpha=|E_{\mathrm{ad}}|\tau_{0}$ is the orthogonality exponent with $E_{\mathrm{ad}}=-Lg_{2}^{2}/(16\pi\tau_{0}^{2}v^{3})$ being the ground state energy of the final Hamiltonian within perturbation theory, $L$ is the length of the sample[@ztwork]. [In Ref. , it was shown that the long time limit of the zero temperature Loschmidt echo obtained from the LM describes that of the XXZ Heisenberg spin chain. We believe that this agreement can be extended to finite temperatures, given the fact, that finite temperatures mostly affect states with energy smaller than $T$. Therefore, as long as $T\ll 1/\tau_0$, these corrections are expected to be universal, as demonstrated in Eq. .]{} Even in the long time limit, we require $t\ll L/v$. [For larger times, comparable to $L/v$, quantum revival occurs similarly to other cases[@zanardi], which is beyond the scope of the present paper.]{}
Probability distribution of total energy after quantum quench {#sec:tepdf}
=============================================================
In this section we analyze the statistics of the total final energy, which, as opposed to work statistics, requires only one energy measurement. Therefore, repeating the procedure of releasing the LL from the trap and measuring its energy many times is expected to lead to the probability distribution function of the total energy.
[Since the total energy is preserved after the quench, its distribution does not change while the steady state is reached. Therefore, it is sufficient to determine the PDF right at $t=\tau$.]{} Its generating function is defined as $$\label{eq:gdef}G(\lambda)=\mathrm{Tr}\left[\hat{\rho}(\tau)e^{i\lambda \hat{H}(\tau)}\right]\,.$$ Since the Hamiltonian $\hat{H}(\tau)$ is a linear combination of $\hat{n}_{q}$ and $\hat{n}_{-q}$, the generating function can be given by means of $f(\xi_{+},\xi_{-})$ defined in Eq. . From this, we get
$$\label{eq:lngfull}
\ln G(\lambda)=i\lambda E_{\mathrm{ad}}+\sum_{q>0}\ln f(\lambda\Omega(q),\lambda\Omega(q))=
i\lambda E_{\mathrm{ad}}-\sum_{q>0}\ln\left[1+n(\tau,q)\left(1-e^{2i\lambda\Omega(q)}\right)-\frac{\partial n_{0}(q)}{\partial(\beta\omega_{0}(q))}\left(e^{i\lambda\Omega(q)}-1\right)^{2}\right]$$
where $n(\tau,q)$ and $n_{0}(q)$ are the expectation value of the occupation number after and before the quench, respectively. The summation over $q$ cannot be performed analytically, therefore to make progress, we assume a small $g_{2}/v$ and consider the generating function within perturbation theory for the case of SQ. We also disregard the velocity renormalization. The occupation number in the final state is $$\begin{gathered}
n(\tau=0,q)=n_{0}(q)+\frac{g(q)^{2}}{4\omega_{0}(q)^{2}}\left(n_{0}(q)+\frac{1}{2}\right)\end{gathered}$$ up to second order in $g_{2}/v$. Within perturbation theory, the generating function is obtained as
$$\label{eq:lng}\ln G(\lambda)=\frac{i\lambda}{\beta-i\lambda}\ln Z_{0}(\beta)+i\lambda E_{\mathrm{ad}}\left[1+\left(\frac{2\tau_{0}}{\beta-i\lambda}\right)^{2}\zeta\left(2,1+\frac{2\tau_{0}}{\beta-i\lambda}\right)\right]+E_{\mathrm{ad}}\tau_{0} h\left(\frac{\beta}{\tau_{0}},\frac{\lambda}{\tau_{0}}\right)$$
$$\label{eq:hf}h\left(\frac{\beta}{\tau_{0}},\frac{\lambda}{\tau_{0}}\right)=1-\frac{1}{(z^{*})^{2}}\left[(1-z)\psi\left(\frac{1-z}{z^{*}} \right)+(1+z)\psi\left(\frac{1+z}{z^{*}}\right)-2\psi\left(\frac{1}{z^{*}} \right)\right]\qquad z=\frac{\beta+i\lambda}{2\tau_{0}}\,.$$
In Eqs. and , $\zeta(x)$ is the generalized zeta function and $\psi(x)$ is the digamma function. The partition function of the initial state is $\ln Z_{0}(\beta)=L\pi/(6\beta v)$. We note that $\ln G(\lambda)$ has poles on the lower complex semiplane only. It follows that the PDF is identically zero for energies lower than the ground state energy $E_{\mathrm{ad}}$, which meets physical expectations as well. Before presenting results on the PDF of total energy, we investigate two simple cases when the PDF can be calculated analytically and will also play important role later.
[At zero temperature, Eq. simplifies to $$\begin{gathered}
\ln G(\lambda;\beta\rightarrow\infty)=i\lambda E_{ad}-\frac{\lambda}{i\tau_0+\lambda}\alpha
\label{eq:gzt}\end{gathered}$$ which reproduces the results of Ref. , leading to a noncentral chi-squared distribution for the PDF.]{}
[With finite initial temperature,]{} the behaviour of the unquenched case ($g_{2}=0$) is also interesting. The final state is the same as the initial thermal equilibrium state and the generating function reads as $$\label{eq:lngunpert}
\ln G(\lambda;g_{2}=0)=-\frac{\lambda}{i\beta+\lambda}\ln Z_{0}$$ which leads to another noncentral chi-squared distribution with the noncentrality parameter $2\ln Z_{0}$. The PDF is $$\begin{gathered}
P(E;g_{2}=0)\label{eq:pdffintemp}
=\frac{e^{-\beta E}}{Z_{0}}\left[\delta(E)+\sqrt{\frac{L\pi}{6E v}}
I_{1}\left(2\sqrt{\frac{L\pi}{6 v} E}\right) \right],\end{gathered}$$ which is equal to the Boltzmann factor $\exp{(-\beta E)}/Z_{0}$ multiplied by the total energy density of states of a one-dimensional Bose gas with linear dispersion. The modified Bessel function behaves as an exponential function in the thermodynamic system limit and almost all the spectral weight is carried by a [non-Gaussian sharp peak centered at $\langle E\rangle=(\ln Z_{0})/\beta$ and of width $\Delta E=\sqrt{2\ln Z_{0}}/\beta$ as $$\begin{gathered}
P(E;g_2=0)\approx\frac{\beta(\ln Z_{0})^{\frac{1}{4}}}{2\sqrt{\pi}(\beta E)^{\frac{3}{4}}}\exp\left(-\left(\sqrt{\beta E}-\sqrt{\ln Z_{0}} \right)^2\right)\end{gathered}$$ whose high energy tail decays according to the Gamma distribution as $\exp(-\beta E)/(\beta E)^{3/4}$.]{} [In the strict thermodynamic limit, $L\rightarrow \infty$, the peak becomes infinitely narrow since $\Delta E/\langle E\rangle\rightarrow 0$, as universally expected, and]{} the Dirac delta part of Eq. is exponentially suppressed. For small systems, however, the Dirac-delta carries most of the probability weight and the continuous part contributes an [exponentially decaying tail only as $$\begin{gathered}
P(E;g_{2}=0)\approx\frac{1}{Z_{0}}\delta(E)+\frac{\beta\ln Z_{0}}{Z_{0}}\exp(-\beta E) \,.\end{gathered}$$]{}
Now we go on calculating the PDF of total energy within perturbation theory and at finite temperature. At low temperature $\beta\gg\tau_{0}$, we obtain $$\label{eq:sqlowT}
\ln G(\lambda;\beta \gg \tau_{0})=-\frac{\lambda}{i\beta+\lambda}\ln Z_{0}+i\lambda E_{\mathrm{ad}}-\frac{\lambda}{i\tau_{0}+\lambda}\alpha$$ which is the sum of Eqs. and . This means that the PDF is the convolution of Eq. and the zero temperature PDF both described by a noncentral chi-squared distribution. These consist of a Dirac delta part and a continuous part and so does their convolution. The weight of the Dirac delta after convolution is $\exp(-\alpha)/Z_{0}$ [which is the probability of the vacuum-to-vacuum process]{}.
We have numerically checked that at low temperature the convolution of the two abovementioned PDFs equals the exact PDF calculated by Fourier transforming the generating function Eq. . The results of the numerical convolution are plotted in Fig. \[fig:numpdfs\] for different system sizes. For small systems ($\alpha$ is small, see Fig. \[fig:numpdfs\].a), most of the spectral weight is carried by the Dirac-delta at zero temperature. With increasing temperature, one part of the Dirac delta stays at $E_{\mathrm{ad}}$ while another part evolves to a finite-width peak with the expectation value of $(\ln Z_{0})/\beta$.
In the thermodynamic limit (large system with $\alpha\gg 1$, see Fig. \[fig:numpdfs\].c) the zero temperature PDF consists of a [ broadened peak at about $E=0$ and a Dirac delta at the adiabatic ground state energy difference]{} with a small probability weight. At finite temperature the broadened peak is shifted to $\langle E\rangle(T)=(\ln Z_{0})/\beta$ and its width changes as $\Delta E(T)=\sqrt{2\alpha/\tau^{2}_{0}+2(\ln Z_{0})/\beta^{2}}$. At low temperatures $$\begin{gathered}
\frac{\Delta E(T)-\Delta E(0)}{\Delta E(0)}\sim T^{3}\,.\end{gathered}$$ This broadening is, however, not as spectacular as for small systems because its ratio with the shift of the peak scales as $(\Delta E(T)-\Delta E(0))/\langle E\rangle(T)\approx
\sqrt{2}\tau_{0}/(\sqrt{\alpha}\beta)$, being almost negligible in the thermodynamic limit.
a) b) c)
Statistics of work {#sec:wpdf}
==================
Measuring the work statistics requires in principle two energy measurement[@hanggirmp; @jarzynski], one before and one after the time dependent protocol, though in the zero temperature case, the first one could be omitted[@ztwork]. At finite temperature, however, this problem can be circumvented by coupling the system of interest to a qubit[@mazzola; @dorner], whose interferometry or spectroscopy would yield the desired correlation function. In the case of a LL, a hybrid system containing cold atoms and a flux qubit[@rmpnori] coupled to a Feshbach resonance was proposed at zero temperature to measure the Loschmidt echo using rf spectroscopy or Ramsey interferometry[@doraloschmidt], and this can readily be extended for finite temperatures as well. This setting can also be useful to measure the statistics of work in other systems [@greekguy; @yulia].
In this section we investigate the PDF of work $\tilde{P}(W)$. At zero temperature this distribution coincides with the total energy distribution since the energy of the initial state is well-defined. At finite temperature, however, this is not the case. In the initial thermal equilibrium state the system can have arbitrary positive energy. It follows that the PDF of work differs from the PDF of total energy and has no lower bound.
The generating function of the distribution of work is defined as [@notanobservable] $$\begin{gathered}
\tilde{G}(\lambda)=\mathrm{Tr}\left[\hat{\rho}_{0}e^{-i\lambda\hat{H}_{0}}e^{i\lambda\hat{H}_{H}(\tau)}\right],
\label{genfuncwork}\end{gathered}$$ where $\hat{H}_{H}(\tau)$ is the final Hamiltonian in Heisenberg picture. We note that with finite temperature initial state the work statistics cannot be derived from the time dependent fidelity unlike the zero temperature case where the fidelity as a function of time and the generating function of work are basically the same in the case of a SQ[@silvaqcp].
The generating function can still be rewritten by means of the generating function of the total energy $G(\lambda,\beta)$ as $$\begin{gathered}
\tilde{G}(\lambda,\beta)=\frac{Z_{0}(\beta+i\lambda)}{Z_{0}(\beta)}G(\lambda,\beta+i\lambda).\end{gathered}$$ Using Eq. , we obtain $$\begin{gathered}
\ln \tilde{G}(\lambda)=i\lambda E_{\mathrm{ad}}\left(1+8\frac{\tau_{0}^{2}}{\beta^{2}}\zeta\left(2,1+\frac{2\tau_{0}}{\beta}\right) \right)+\nonumber \\
\label{eq:work}
+E_{\mathrm{ad}}\tau_{0} h\left(\frac{\beta+i\lambda}{\tau_{0}},\frac{\lambda}{\tau_{0}}\right)\end{gathered}$$ up to second order in $g_{2}/v$ where the function $h$ was defined in Eq. . This result is valid only within perturbation theory and in the SQ limit. We note that $\ln\tilde{G}(\lambda)$ has poles on both complex semiplanes. This is the mathematical reason for the absence of lower bound in the PDF. It can be proven that Eq. satisfies the Jarzynski equality [@jarzynski], i.e. $\langle \exp{(-\beta W)}\rangle= \tilde{G}(\lambda=i\beta)= Z(\tau)/Z_{0}$ where $Z(\tau)=\mathrm{Tr}\left[\exp{(-\beta \hat{H}(\tau))}\right]$ is the partition function of the final Hamiltonian.
The Fourier transform of the generating function is evaluated numerically, and is shown in Fig. \[fig:workstat\] for different system sizes and initial temperatures.
a)b)c)
The expectation value and the variance of the PDF of work is calculated analytically by taking the derivatives of the generating function Eq. at $\lambda=0$, yielding $\langle W\rangle_{SQ}=0$ for arbitrary temperature and only the form of the distribution changes as the temperature increases.
In the small system limit, the zero temperature PDF consists of a Dirac delta with large probability weight and an exponentially decaying tail (see Fig. \[fig:workstat\].a). At finite temperature the Dirac delta disappears and deforms to a broadened, negatively skewed peak. This deformation is a significant modification of the PDF.
In the thermodynamic limit the only change is that the zero temperature peak slightly broadens (see Fig. \[fig:workstat\].c). The width of the peak varies as $$\label{eq:workwidth}\frac{\Delta W(T)-\Delta W(0)}{\Delta W(0)}=\frac{4\pi^{2}}{3}(\tau_{0}T)^{3},$$ which means that significant broadening could be noticeable at higher temperatures only.
Conclusion
==========
We have theoretically studied quantum quenches [in the Luttinger model with finite temperature equilibrium initial state]{}. The steady state has been described by the diagonal ensemble, i.e. by determining the diagonal elements of the time evolved density matrix. At finite temperature, the boson numbers in the $+q$ and $-q$ modes can differ from each other and the difference may be larger for higher initial temperature.
The long time limit of the Uhlmann fidelity shows that the time evolved state deviates from the initial state with increasing temperature. How this relation depends on the quench duration is an interesting and open question.
Finite temperature effects in the statistics of final total energy and work done on the system during the quench have also been investigated in the SQ limit. It is worth mentioning again that these two distributions differ from each other, unlike in the zero temperature case, because in the finite temperature initial state the energy is not well defined. Within perturbation theory and at low temperature the PDF of total energy is found to be the convolution of the PDF of initial energy and the PDF of zero temperature work statistics. Numerical results show that the distribution is shifted and broadened due to finite temperature for both the thermodynamic and small system limits. In small systems, however, the broadening is more robust.
The finite temperature effects in the statistics of work depend remarkably on the system size. In the small system limit, significant rearrangement may be observed with increasing temperature, while in the thermodynamic limit, the peak of the PDF slightly broadens only. We believe that our results related to the distribution function of total energy and work done can be observed experimentally, using the setups suggested in Ref. [@dorner; @mazzola; @doraloschmidt].
We thank M. Rigol and G. Zaránd for stimulating comments. This work was supported by the Hungarian Scientific Research Fund under Grants No. OTKA K101244, K105149, CNK80991 and by the ERC Grant No. 259374-Sylo and by the Bolyai Program of the Hungarian Academy of Sciences.
Derivation of the generating function of occupation number distribution {#sec:Fxi}
=======================================================================
The generating function of the occupation number probability distribution is defined as $$\begin{gathered}
\label{eq:fdef2}
f(\xi_{+},\xi_{-})=\mathrm{Tr}\,\left[\hat{\rho}(\tau)e^{i(\xi_{+}\hat{n}_{+}+\xi_{-}\hat{n}_{-})}\right]\end{gathered}$$ where $\hat{\rho}(\tau)$ is the exact time evolved density operator after the quench, see Eq. . In this section we consider only a single $q>0$ mode. The initial state is $$\begin{gathered}
\hat{\rho}_{0}=\frac{e^{-\beta\omega_{0}(b_{+}^{+}b^{}_{+}+b_{-}^{+}b^{}_{-})}}{z_{0}}\qquad\qquad z_{0}=(1-e^{-\beta\omega_{0}})^{-2}\end{gathered}$$ describing a canonical ensemble. We introduce the operators $$\begin{gathered}
\hat{K}_{0}=\frac{d_{+}^{+}d_{+}+d_{-}d_{-}^{+}}{2}\end{gathered}$$ $$\begin{gathered}
\hat{K}_{+}=d^{+}_{+}d^{+}_{-}\qquad \hat{K}_{-}=d_{+}d_{-}=\hat{K}_{+}^{+}\end{gathered}$$ where $d_{\pm}$ is the annihilation operator of quasiparticles diagonalizing the final Hamiltonian. The operator $\hat{K}_{0}$ does not change the number of bosons while $\hat{K}_{+}$ ($\hat{K}_{-}$) creates (annihilates) a pair of $d_{+}$ and $d_{-}$ bosons. The operators obey the commutation relations of $su(1,1)$ algebra, $\left[\hat{K}_{0},\hat{K}_{\pm}\right]=\pm\hat{K}_{\pm}$ and $\left[\hat{K}_{+},\hat{K}_{-}\right]=2\hat{K}_{0}$. The time evolved density operator is expressed as $$\begin{gathered}
\hat{\rho}(\tau)=\frac{1}{z_{0}}\exp{\left(\beta\omega_{0}\left(1-a(\tau)2\hat{K}_{0}-c(\tau)\hat{K}_{-}-c(\tau)^{*}\hat{K}_{+}\right)\right)},\end{gathered}$$ where $a(\tau)$ is given in Eq. and $$\begin{aligned}
c(\tau)=-\frac{g}{\Omega}(1+2|v(\tau)|^{2})+\nonumber \\
+2i\mathrm{Im}(u(\tau)v(\tau)^{*})-\frac{2\omega(\tau)}{\Omega}\mathrm{Re}(u(\tau)v(\tau)^{*})
\label{eq:c2}\end{aligned}$$ where $u(\tau)$ and $v(\tau)$ are the Bogoliubov coefficients defined in Eq. ). In Eqs. and , $\omega(\tau)$ is the non-interacting energy with renormalized velocity, $g$ is the interaction strength and $\Omega=\sqrt{\omega(\tau)^{2}-g^{2}}$ is the eigenenergy of the final Hamiltonian. [For arbitrary quench protocol, $a(\tau)^{2}-|c(\tau)|^{2}=1$ holds true and, hence, $a(\tau)\geq 1$.]{} The other exponential under the trace in Eq. is rewritten as $$\begin{gathered}
e^{i(\xi_{+}\hat{n}_{+}+\xi_{-}\hat{n}_{-})}=e^{-i\xi+i\Delta\xi\Delta\hat{n}}e^{i\xi\,2\hat{K}_{0}},\end{gathered}$$ where we have introduced $\xi=(\xi_{+}+\xi_{-})/2$ and $\Delta\xi=(\xi_{+}-\xi_{-})/2$. The operator $\Delta\hat{n}=\hat{n}_{+}-\hat{n}_{-}$ commutes with both $\hat{K}_{0}$ and $\hat{K}_{\pm}$. Using a faithful representation of $su(1,1)$ algebra [@solomon; @gilmore], we derive a single exponential which equals the product $\hat{\rho}(\tau)e^{i(\xi_{+}\hat{n}_{+}+\xi_{-}\hat{n}_{-})}$. The generators of the $su(1,1)$ algebra may be faithfully represented by $2\hat{K}_{0}\rightarrow \sigma_{z}$ and $\hat{K}_{\pm}\rightarrow (\pm\sigma_{x}+i\sigma_{y})/2$ where $\sigma_{x}$, $\sigma_{y}$ and $\sigma_{z}$ are the $2\times 2$ Pauli matrices. The single exponent then can be diagonalized by standard Bogoliubov transformation. After diagonalization, $$\begin{gathered}
\hat{\rho}(\tau)e^{i(\xi_{+}\hat{n}_{+}+\xi_{-}\hat{n}_{-})}= \frac{e^{-i\xi+\beta\omega_{0}}}{z_{0}} \nonumber \\
\times \exp{\left(i\Delta\xi\Delta\hat{\bar{n}}-\ln(B+\sqrt{B^{2}-1})(1+\hat{\bar{n}}_{+}+\hat{\bar{n}}_{-})\right)},
\label{eq:sexp}\end{gathered}$$ where $$\begin{gathered}
B=\cos\xi\cosh(\beta\omega_{0})-ia(\tau)\sin\xi\sinh(\beta\omega_{0}),\end{gathered}$$ and $\hat{\bar{n}}_{\pm}=\bar{d}^{+}_{\pm}\bar{d}_{\pm}$ is the occupation number operator after Bogoliubov transformation and $\Delta \hat{\bar{n}}=\hat{\bar{n}}_{+}-\hat{\bar{n}}_{-}$. The annihilation operator is expressed with the new annihilation and creation operators as $$\begin{gathered}
d_{\pm}=\frac{\bar{d}_{\pm}+\gamma\bar{d}^{+}_{\mp}}{\sqrt{1-|\gamma|^{2}}}, \\
\gamma=\frac{\sqrt{B^{2}-1}+i\sin\xi\cosh(\beta\omega_{0})-a(\tau)\cos\xi\sinh(\beta\omega_{0})}{c(\tau)e^{-i\xi}\sinh(\beta\omega_{0})}.\end{gathered}$$ Substituting Eq. into Eq. , the generating function is obtained as $$\begin{aligned}
f(\xi_{+},\xi_{-})=\Big[1+n(\tau)\left(1-e^{i(\xi_{+}+\xi_{-})}\right)+ \nonumber \\
+(n_{0}+n_{0}^{2})\left(e^{i\xi_{+}}-1\right)\left(e^{i\xi_{-}}-1\right)\Big]^{-1}\end{aligned}$$ where $n_{0}=(e^{\beta\omega_{0}}-1)^{-1}$ and $n(\tau)=\mathrm{Tr}\left[\hat{\rho}(\tau)\hat{n}_{\pm}\right]=\left(a(\tau)(2n_{0}+1)-1\right)/2$ are the expectation value of the occupation number before and after the quench. Note that $f(\xi_{+},\xi_{-})$ is a $2\pi$-periodic function of its variables and $f(\xi_{+},\xi_{-})=f(\xi_{-},\xi_{+})$. The latter implies $\rho(n_{+},n_{-})=\rho(n_{-},n_{+})$.
Fidelity {#sec:fid}
========
[[The technical difficulty in computing the fidelity is evaluating the trace of some exponentials, the exponent of which are expressed in terms of $\hat{K}_{0}$, $\hat{K}_{\pm}$ and $\Delta \hat{n}$, irrespectively of the norm chosen on the set of density operators.]{}]{} Using again the faithful representation of the $su(1,1)$ algebra, the product of exponentials can be transformed into a single exponential in the same way as in Appendix \[sec:Fxi\]. In order to calculate the square root of an exponential, which we need in the case of the Uhlmann fidelity, we diagonalize the exponent and halve the eigenvalues. The trace of the single exponential is evaluated after diagonalizing the exponent. Using this procedure, we obtain the fidelity using both the Frobenius and the Bures metric[@zanardi].
The Uhlmann fidelity is evaluated as $$\begin{gathered}
\ln F_{U}(t)=\ln\mathrm{Tr}\left[\sqrt{\hat{\rho}^{1/2}_{0}\hat{\rho}(t)\hat{\rho}^{1/2}_{0}}\right]=\nonumber \\
=\sum_{q>0}\ln\frac{\cosh(\beta\omega_{0}(q))-1}{\sqrt{1+|u_{q}(t)|^{2}\sinh^{2}(\beta\omega_{0}(q))}-1}\,.\end{gathered}$$ In Eq. , $u_{q}(t)$ is the Bogoliubov coefficient defined in Eq. . Up to second order in $g_{2}/v$ and for a SQ, $$\begin{gathered}
|u_{q}(t)|\approx 1+\frac{g(q)^{2}}{2\omega_{0}(q)^{2}}\sin^{2}(\omega_{0}(q)t),\end{gathered}$$ and
$$\begin{gathered}
\ln F_{U}(t)=-{\displaystyle\sum_{q>0}}\dfrac{\frac{g^{2}}{\omega_{0}^{2}}\sin^{2}(\omega_{0}t)}{1+\tanh^{2}\left(\frac{\beta\omega_{0}}{2}\right)}=\nonumber \\
=-\alpha\left[1+\dfrac{\tau_{0}}{\beta}\left(\psi\left(\dfrac{3}{4}+\dfrac{\tau_{0}}{2\beta}\right)-\psi\left(\dfrac{1}{4}+\dfrac{\tau_{0}}{2\beta}\right)\right)+\mathrm{Re}\left(\frac{i\tau_{0}}{t-i\tau_{0}}-\frac{\tau_{0}}{\beta}\left(\psi\left(\dfrac{3}{4}+\dfrac{\tau_{0}}{2\beta}-\frac{it}{2\beta}\right)-\psi\left(\dfrac{1}{4}+\dfrac{\tau_{0}}{2\beta}-\frac{it}{2\beta}\right) \right)\right)\right]
\label{eq:fut}\end{gathered}$$
where $\psi(x)$ is the digamma function. In the long time limit the last term in Eq. converges to zero. At low temperatures, $$\begin{gathered}
\ln F_{U}(t\gg\beta\gg\tau_{0})=-\alpha\big(1+\pi\tau_{0}T\big)\end{gathered}$$ where we used $\psi(3/4)-\psi(1/4)=\pi$.
Using the Frobenius norm, the overlap of the time-evolved and initial states is derived as $$\begin{gathered}
\ln F_{F}(t)=\frac{1}{2}\ln\mathrm{Tr}\left[\hat{\rho}(t>\tau)\hat{\rho}_{0}\right]=\nonumber \\
=\sum_{q>0}\ln\frac{\cosh(\beta\omega_{0}(q))-1}{|u_{q}(t)|\sinh(\beta\omega_{0}(q))}\,.\end{gathered}$$ The effective dimension, $$\begin{gathered}
\ln d_{\mathrm{eff}}=-\ln \mathrm{Tr}\left[\rho_{0}^{2}\right]=2\ln Z_{0}(\beta)-\ln Z_{0}(2\beta)=\nonumber \\
=2\sum_{q>0}\ln\frac{1-e^{-2\beta\omega_{0}(q)}}{(1-e^{-\beta\omega_{0}(q)})^{2}}\end{gathered}$$ leads to the temperature independent normalized fidelity $$\begin{gathered}
\ln \left(\sqrt{d_{\mathrm{eff}}}F_{F}(t)\right)=-\sum_{q>0}\ln|u_{q}(t)|\label{eq:fff}\end{gathered}$$ for arbitrary quench protocol. Within perturbation theory and for a SQ, we obtain $$\begin{gathered}
\ln \left(\sqrt{d_{\mathrm{eff}}}F_{F}(t)\right)=
-\sum_{q>0}\frac{g(q)^{2}}{2\omega_{0}(q)^{2}}\sin^{2}(\omega_{0}(q)t)=\nonumber\\ =-\alpha\frac{t^{2}}{\tau_{0}^{2}+t^{2}}.\end{gathered}$$ For large times ($t\gg\tau_{0}$) the normalized fidelity saturates at $e^{-\alpha}$.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We employ the Bayesian framework to define a cointegration measure aimed to represent long term relationships between time series. For visualization of these relationships we introduce a dissimilarity matrix and a map based on the Sorting Points Into Neighborhoods (SPIN) technique, which has been previously used to analyze large data sets from DNA arrays. We exemplify the technique in three data sets: US interest rates, monthly inflation rates and gross domestic product growth rates.'
address:
- 'GRIFE, Escola de Artes, Ci[ê]{}ncias e Humanidades, Universidade de São Paulo, Parque Ecológico do Tietê, 03828-020, São Paulo-SP, Brazil'
- 'Dep. de F[í]{}sica, IBILCE, Universidade Estadual Paulista, 15054-000 São José do Rio Preto - SP, Brazil'
- 'Dep. de F[í]{}sica Geral, Instituto de F[í]{}sica, Universidade de São Paulo, Caixa Postal 66318, 05315-970 São Paulo - SP, Brazil'
author:
- Renato Vicente
- ', Carlos de B. Pereira'
- ', Vitor B.P. Leite'
- and Nestor Caticha
title: Long Term Economic Relationships from Cointegration Maps
---
complex systems,econophysics ,cointegration ,clustering ,Bayesian inference 89.65.-s ,89.65.Gh ,02.50.Sk
Introduction
============
Correlations are a central topic in the study of the collective properties of complex systems, being of particularly practical importance when systems of economic interest are concerned [@bouchaud]. Unlike correlation, the idea of cointegration [@engle; @hamilton] brings in a relationship measure that is long term in nature being somewhat related to the concept of damage spreading in a pair of spin models [@damage]. However, cointegration has up to now been rather absent from the description of physical systems and, in particular, from economic systems studied from a physical perspective. A set of non-stationary time series cointegrate if there exists a linear combination of them that is mean reverting. Plainly speaking, two appropriately scaled time series cointegrate if in the long term they either tend to move together or as mirror images.
Bayesian methods provide a unifying approach to statistics [@jaynes]. They help to establish, from clear first principles, the methods, assumptions and approximations made in a particular statistical analysis. A major issue in the study of cointegration is the detection of cointegrated sets, a problem that has been extensively dealt with in the econometrics literature both from classical [@hubrich] and Bayesian [@koop] perspectives.
Dealing with extensive volumes of data is a common trend in several areas of science. The need to sort, cluster, organize, categorize, mine or visualize large data sets brings a perspective that unifies distant fields, if not at all in aims, at least in methods. Cross fertilization may promptly provide candidate solutions to problems, avoiding the need of rediscovery or worst, just plain non-discovery. Bioinformatics presents a good example, where the availability of genome, protein and DNA array data has prompted the proposal by several groups of new methods. From this repertoire we borrow a method, SPIN [@tsafrir], previously developed for automated discovery of cancer associated genes.
Our first goal in this paper is to devise a cointegration measure for time series of economic interest that is both physically meaningful and reasonably simple to compute. Our second goal is, by employing the SPIN method, to emphasize the importance of visual organization and presentation of relationship pictures (or maps) that emerge when complex systems are analysed.
This paper is organized as follows. In the next section we derive a cointegration measure employing Bayesian statistics and briefly discuss the relation between cointegration and correlation and between the proposed measure and usual unit-root statistics. In section 3 we use the SPIN method to introduce the cointegration heat map as a visualization tool. In section 4 we exemplify the proposed technique in three macroeconomic time series: US interest rates (USIR), inflation rates (IFR) and gross domestic products (GDP). Conclusions are presented in section 5.
Cointegration measure
=====================
A pair of time series ${\bm x}_1$ and ${\bm x}_2$ cointegrates [@hamilton] if there exists a linear combination $$\begin{aligned}
a_1 x_{1,t} + a_2 x_{2,t} +b &=& \epsilon_t \label{eq:cointegra}\end{aligned}$$ such that the residues ${\bm \epsilon}$ satisfy the following stationarity condition: $$\begin{aligned}
\epsilon_{t+1}&=&\gamma\epsilon_t + \eta_t, \label{eq:stationary}\end{aligned}$$ where $\langle \eta_t \rangle =0$, $\langle \eta_t^2 \rangle=\sigma^2$ and $\gamma<1$. If $\gamma=1$ the residues are non-stationary and if $\gamma>1$ the system is unstable. Notice that $\gamma$ is related to a time scale $\tau=1/(1-\gamma)$ for relaxation of $\epsilon_t$ to its long term mean.
We also assume a budget constraint taking the form $$\begin{aligned}
a_1^2+a_2^2&=&1.
\label{eq:budget}\end{aligned}$$ Since eq. \[eq:cointegra\] is linear, we can impose this constraint by assuming that $a_1=\sin(\theta)$ and $a_2=\cos(\theta)$ without loss of generality. Note that the above system has still more freedom arising from the following symmetry group: $$\begin{aligned}
x_i&\rightarrow& x_i'= \alpha x_i+y_i,\nonumber\\
b &\rightarrow& b'=\alpha b -a_1 y_1 -a_2 y_2,\nonumber\\
\sigma &\rightarrow&\sigma'=\alpha \sigma,\nonumber\\
\gamma&\rightarrow&\gamma'=\gamma, \label{eq:symmetries}\end{aligned}$$ which means that we can change the units in which quantities are measured and add constants $y_i$ without interfering with the cointegration property. We, therefore, can partially fix the [*gauge*]{} so that $x_i \rightarrow x_i'=x_i - \overline{x_i}$, such that the empiric time series averages are zero. This forces a choice of $b=0$.
That cointegration and correlation in time series fluctuations are quite distinct properties can be easily seen with the aid of an elementary example. Suppose two time series $x_t$ and $y_t$ that orbit the same random walk $w_t$ as follows: $$\begin{aligned}
x_t&=&w_t+\epsilon^x_t \\
y_t&=&w_t+\epsilon^y_t\nonumber\\
w_t&=&w_{t-1}+\eta_t \nonumber,\end{aligned}$$ where $\epsilon^x_t$, $\epsilon^y_t$ and $\eta$ are random i.i.d shocks with zero mean and variances $\sigma^2_x$, $\sigma^2_y$ and $\sigma^2_\eta$, respectively. The correlation $\rho$ between the first differences $\Delta x=x_t-x_{t-1}$ and $\Delta y=y_t-y_{t-1}$ can be easily computed yielding: $$\begin{aligned}
\rho&=&\frac{\langle\Delta x \Delta y \rangle}{\sigma_x\sigma_y}=
\frac{\sigma^2_\eta}{\sigma_x\sigma_y}.\end{aligned}$$ Clearly $x$ and $y$ strongly cointegrate ($\gamma=0$). However the choice $\sigma_\eta\ll\sigma_x,\sigma_y$ would imply that the linear correlation in their fluctuations are at the same time very low.
The main ingredients in a Bayesian approach are three. First we need a model as given by eqs. \[eq:cointegra\], \[eq:stationary\] and \[eq:budget\]. Then a noise model to build the likelihood and finally the priors. The interesting consequence of a group of invariance as the one described by eq. \[eq:symmetries\] is that it, together with the budget and stability conditions, constrains [@jaynes] the form of the priors to: $$\begin{aligned}
p(\gamma)&=&\Theta(\gamma)\Theta(1-\gamma), \label{eq:priorgama}\end{aligned}$$ where $\Theta(\cdot)$ is the Heaviside step function, and $$\begin{aligned}
p(\sigma)&\propto& \frac{1}{\sigma^2}. \label{eq:priorsigma}\end{aligned}$$
With these ingredients we can calculate the posterior probability of $\gamma$ given the residues as: $$\begin{aligned}
p(\gamma\mid{\bm \epsilon})&\propto&\int_{\sigma_{\mbox{\tiny min}}}^{\infty}d\sigma\;
p({\bm \epsilon}\mid \gamma,\sigma)p(\sigma)p(\gamma), \label{eq:posterior}\end{aligned}$$ where $\sigma_{\mbox{\tiny min}}>0$ can be made arbitrarily small without changing the main results to follow.
Equations \[eq:cointegra\] and \[eq:stationary\] combined give the following likelihood function for the residues: $$\begin{aligned}
p(\bm{\epsilon}\mid \gamma,\sigma)&\propto&
\prod_{t=1}^{T-1}\frac{1}{\sigma}\exp\left[-\frac{\left(\epsilon_{t+1}-
\gamma\epsilon_{t}\right)^2}{2\sigma^2}\right].
\label{eq:likelihood}\end{aligned}$$ Performing the integral in eq. \[eq:posterior\] yields: $$\begin{aligned}
p(\gamma\mid{\bm
\epsilon})\propto\Theta(\gamma)\Theta(1-\gamma)\left[\sum_{t=1}^{T-1}
\left(\epsilon_{t+1}-\gamma\epsilon_{t}\right)^2\right]^{-\frac{T-2}{2}}.
\label{eq:posterior_integral} \end{aligned}$$
![Left: Posterior probabilities for two synthetic pairs ${\bm
x}_1,{\bm x}_2$ of time series. Cointegrating pair, characterized by the maximum a-[*posteriori*]{} estimate $\hat{\gamma}\approx 0.5$ (top). Non-cointegrating pair characterized by $\hat{\gamma}\approx 1$ (bottom). Right: Examples of the cointegration measure in US interest rate data. []{data-label="fig1"}](posterior_sinteticos.eps "fig:"){width="50.00000%"} ![Left: Posterior probabilities for two synthetic pairs ${\bm
x}_1,{\bm x}_2$ of time series. Cointegrating pair, characterized by the maximum a-[*posteriori*]{} estimate $\hat{\gamma}\approx 0.5$ (top). Non-cointegrating pair characterized by $\hat{\gamma}\approx 1$ (bottom). Right: Examples of the cointegration measure in US interest rate data. []{data-label="fig1"}](usir_coint_example.eps "fig:"){width="50.00000%"}
For large $T$, the distribution of residues, given the data, can be approximated by
$$\begin{aligned}
p(\bm{\epsilon}\mid {\bm x}^{\prime}_1,{\bm
x^{\prime}}_2)&\approx&\delta(\epsilon_t-x^{\prime}_{1,t}\sin\hat{\theta}
+x^{\prime}_{2,t}\cos\hat{\theta})\end{aligned}$$
with $\hat{\theta}$ estimated by minimizing the variance $\langle\epsilon^2\rangle$ to find: $$\begin{aligned}
\hat{\theta}&=&
\frac{1}{2}\arctan\left[2\frac{\langle
x^{\prime}_1x^{\prime}_2\rangle}{\langle {x^{\prime}_1}^2\rangle
-\langle {x^{\prime}_2}^2\rangle}\right].
\label{eq:theta}\end{aligned}$$
The maximum of the posterior distribution gives an estimate for the relaxation time as $$\begin{aligned}
\hat{\gamma}&=&\mbox{argmax}
\log p(\gamma\mid {\bm x^{\prime}}_1,{\bm x^{\prime}}_2)\\
&=&\mbox{argmax}\log\int d\bm{\epsilon}\,
p(\gamma\mid{\bm \epsilon})p(\bm{\epsilon}
\mid {\bm x}^{\prime}_1,{\bm x^{\prime}}_2)\nonumber. \end{aligned}$$ Finally, we define a family of cointegration $\alpha$-measures as $$\begin{aligned}
d_\alpha({\bm x}_1,{\bm x}_2)\equiv \hat{\gamma}^\alpha.\end{aligned}$$ These measures are symmetric, non-negative and agree with the usual augmented Dickey-Fuller unit-root tests (ADF) [@hamilton] in the sense that lower p-value t-statistics imply higher degrees of similarity as measured by the cointegration property (see Table \[table\]).
The value of $\alpha$ controls the quality of visualizations generated and has been chosen to be $\alpha=1$ (IFR,GDP) and $\alpha=2$ (USIR) in the datasets we have analyzed. In Fig.\[fig1\] (left) we show the log-posteriors obtained for synthetic time series generated with $T=1000$ and $\gamma=0.5$ and $\gamma=1.0$. In Fig. \[fig1\](right) we illustrate the cointegration measure with time series from the USIR dataset. Notice that it can be easily verified by a Taylor expansion of the logarithm of the posterior density (eq. \[eq:posterior\_integral\]) around its maximum that the error bar for the estimate $\hat{\gamma}$ is proportional to $T^{-1/2}$.
Cointegration heat map
======================
*Pair* *$\hat{\gamma}$* *ADF t-stat*
---------- ------------------ --------------
SLB-FP3 0.99 -2.21
CP6-TC1Y 0.95 -4.70
CP1-CP6 0.92 -6.92
CP1-CP3 0.87 -8.72
FED-CP6 0.77 -6.97
FED-CP3 0.64 -8.47
FED-CP1 0.50 -10.36
: ADF test and cointegration measure: Using the USIR dataset as an example, it can be seen that the measure estimated are consistent with the ADF tests in the sense that more improbable t-statistics imply stronger cointegration. The critical value at $1\%$ is $t=-3.88$ in this case.\[table\]
Given a set of time series we are interested in discovering low dimensional structures embedded in an appropriate dissimilarity matrix $D$. The way we define this dissimilarity matrix is conditioned by the use we intend for the data. In principle, we can define $D_{jk}=d_\alpha({\bm x}_j,{\bm x}_k)$ meaning that shorter relaxation times $\tau$ imply more similarity between two time series. Alternatively, inspired by the expression matrices employed in bioinformatics [@tsafrir], we can define vectors $\mathbf{d}_j=(d_{1j},...,d_{Nj})$ representing the cointegration profile between time series $j$ and each one of the $N$ series composing the system with an arbitrary but fixed ordering. A dissimilarity matrix can be then defined along this lines as: $$D_{jk}=\sqrt{\sum_{l=1}^N \left(d_{lj}-d_{lk}\right)^2}.
\label{dissimilarity}$$ In this case two time series are similar if they interact with each system component alike. We have observed that the latter choice yields the same basic structures with clearer and smoother visualization. The use of primitive measures to build second order dissimilarity matrices is a basic idea behind the spectral clustering techniques [@spectral]. However, to our knowledge, the particular construction described by eq. \[dissimilarity\] has not appeared in the literature to date.
![Left: Correlation heat map for the USIR dataset. Rectangles represent the classification yielded by the SPC technique. The general pattern is comparable with figure 3b on [@dimatteo_usir]. Right: Cointegration heat map for the USIR dataset with SPC classification again represented by rectangles. The general emergent pattern clusters short term Treasury instruments around $n4$, financial companies related instruments around $n3$ and long term instruments around $n1$.[]{data-label="fig2"}](usir_corr_spin_spc_expression.eps "fig:"){width="50.00000%"}![Left: Correlation heat map for the USIR dataset. Rectangles represent the classification yielded by the SPC technique. The general pattern is comparable with figure 3b on [@dimatteo_usir]. Right: Cointegration heat map for the USIR dataset with SPC classification again represented by rectangles. The general emergent pattern clusters short term Treasury instruments around $n4$, financial companies related instruments around $n3$ and long term instruments around $n1$.[]{data-label="fig2"}](usir_spin_spc_expression.eps "fig:"){width="50.00000%"}
There are several possible aims behind unsupervised segmentation based on a dissimilarity matrix $D$. Categorization from clustering algorithms has been used for market segmentation [@dimatteo; @tumminello]. For example, the Superparamagnetic Clustering (SPC) algorithm [@SPC; @kullmann] has been particularly useful since the number of clusters is not [*a priori*]{} known and the scale of resolution of the categories can be tuned by a temperature like parameter. Sometimes the data might not have a clear discrete class structure and here the SPIN algorithm provides a difference with its capability of helping identify low dimensional structures in a high dimensional space. Without knowing in advance what type of segmentation will emerge, the clustering and SPIN algorithms should be thought of as complementary. The aim of SPINing a similarity matrix is to obtain a permutation such that points close in distance are brought, by the permutation to places in the matrix that are also close. Since the space of permutations is factorially large this can easily be seen to be a potentially hard problem. The permutations are sequentially chosen, for example to minimize a cost function that penalizes large distances and puts them far from the diagonal or alternatively seek permutations that bring pairs with small distances near to the diagonal. Unless the structure can be ordered in one dimension, these requirements can lead to frustration. The class of cost functions proposed in [@tsafrir] is of the form $\mathcal{F}(P)=$[**Tr**]{}$(PDP^T W)$, with $P$ being a permutation of matrix indices and $W$ a weight matrix which defines the algorithm. For their choices, namely, [*Side-to-Side (STS)*]{} defined as $W=XX^T$, with $X_i>X_j$ if $i>j$ and [*Neighborhood*]{} defined as $W_{ij}=exp(|i-j|²\sigma²)$, the minimization was shown to be NP-complete. The way out is to be satisfied with non optimal solutions that can be obtained in fast times ($\mathcal{O} (n^{2-3})$) and that turn out to be just as informative. The problem of sorting into categories is ill posed and therefore there will not be something like ‘the answer’. The reduction to an optimization problem, using either [*STS*]{} or [*Neighborhood*]{} leads to a NP-complete problem. It is fair to expect that any reasonable weight function will share that characteristic. So we have found that it is adequate to play around with the algorithms and apply them for different subsets, try optimizing the whole matrix, then choose a relevant cointegrating subset, optimize the subset, go up optimize the whole set, intercalate different algorithms. The result will tend to be better as measured by the cost function. This heuristics helps escape from local minima, of course it does not cure the fundamental problem that there might be frustration in a general sorting problem. This is not really a problem, good albeit not optimal solutions are just as informative as a perfect solution would be for all practical purposes. In the following analysis we have found that the best visualizations were simply achieved by employing the [*STS*]{} solution as an initial condition to the [*Neighborhood*]{} variant iterated with a schedule for reducing the scale parameter $\sigma$ in a simulated annealing fashion.
![Left: SPINed cointegration map for the IFR dataset. The clusters that emerge correspond to economies with highly volatile prices (n1), developed economies with stable prices (n2), economies with events of hyperinflation in the period observed (n3) and economies with highly stable prices (n4). Right: Examples in each group. Group n1: Benin (1a), Central African Republic (1b) and Syria (1c). Group n2: USA (2a), Sweden (2b), Japan (2c). Group n3: Brazil (3a), Russia (3b) and Indonesia (3c). Group n4: Australia (4a), New Zealand (4b) and Tuvalu (4c). []{data-label="fig3"}](ifr_alfa1_spin_expression.eps "fig:"){width="50.00000%"}![Left: SPINed cointegration map for the IFR dataset. The clusters that emerge correspond to economies with highly volatile prices (n1), developed economies with stable prices (n2), economies with events of hyperinflation in the period observed (n3) and economies with highly stable prices (n4). Right: Examples in each group. Group n1: Benin (1a), Central African Republic (1b) and Syria (1c). Group n2: USA (2a), Sweden (2b), Japan (2c). Group n3: Brazil (3a), Russia (3b) and Indonesia (3c). Group n4: Australia (4a), New Zealand (4b) and Tuvalu (4c). []{data-label="fig3"}](ifr_series.eps "fig:"){width="50.00000%"}
Application examples
====================
We exemplify the method by calculating cointegration maps for three data sets: (USIR) weekly US interest rates for 34 instruments from January 8, 1982 to August 29, 1997 ($T=817$ weeks) [@usinterest]; (IFR) monthly inflation rates for 179 countries from August, 1993 to December 2004 ($T=137$ months) [@inflation]; (GDP) yearly gross domestic product growth rates for 71 countries from 1980 to 2004 ($T=25$ years) [@gdp].
Measurement in soft sciences is itself a challenging activity [@boumans]. Socio-economic systems are self-aware, there are severe limits to the accuracy of statistical data that can be gathered and even the definition of several macroeconomic quantities is still debatable [@swanson; @morgenstern]. An exception to these data quality constraints are the organized financial markets like those of interest rate instruments in dataset USIR.
In figure \[fig2\] (left) we show the SPINed heat map for correlation coefficients of time series fluctuations. Pseudocolors are assigned according to dissimilarities calculated with eq. \[dissimilarity\] by replacing the cointegration measure by correlation coefficients. In the same figure we show as rectangles identified by $n1,...,n6$ the hierarchical grouping structure generated by the SPC technique ($K=7$, see [@SPC]). Characteristic of unsupervised classification techniques is the reliance of the results on the dissimilarity measure adopted. Despite the differences, the general patterns revealed in figure \[fig2\] compare well with those of figure 3b on [@dimatteo_usir], which employs a classical agglomerative clustering with a metric distance based on linear correlation coefficients. Notice that the SPINed heat map is capable of showing nuances in the relationship structure that are absent in the traditional or SPC approaches. For example, the Treasure bill rates with maturities 3 and 6 months (TBA3M and TBA6M) and other instruments of the same maturity, in particular, Treasure securities at constant maturity (TC3M and TC6M) correlate alike. However, this sort of information is lost both in the SPC classification and in [@dimatteo_usir] with the former classifying these instruments accordingly with their maturities and the latter grouping TBAs in one group and TCs in another.
![Left: SPINed cointegration map for the GDP dataset. The clusters that emerge correspond to countries with accelerating or decelerating economies (Group $n1$); developed countries with stable and accelerating economies (Group $n2$); volatile economies including major oil producers (Group $n3$) and stable economies (Group $n4$). Right: Examples in each group are. Group $n1$: Bangladesh (1a), Tanzania (1b) and Pakistan (1c). Groups $n2$: Norway (2a), United Kingdom (2b) and Japan (2c). Group $n3$: Kuwait (3a), Venezuela (3b) and Saudi Arabia (3c). Group $n4$: US (4a), Australia (4b) and Italy (4c).[]{data-label="fig4"}](imf_gdp_redux_spin_alfa1_expression.eps "fig:"){width="50.00000%"}![Left: SPINed cointegration map for the GDP dataset. The clusters that emerge correspond to countries with accelerating or decelerating economies (Group $n1$); developed countries with stable and accelerating economies (Group $n2$); volatile economies including major oil producers (Group $n3$) and stable economies (Group $n4$). Right: Examples in each group are. Group $n1$: Bangladesh (1a), Tanzania (1b) and Pakistan (1c). Groups $n2$: Norway (2a), United Kingdom (2b) and Japan (2c). Group $n3$: Kuwait (3a), Venezuela (3b) and Saudi Arabia (3c). Group $n4$: US (4a), Australia (4b) and Italy (4c).[]{data-label="fig4"}](imf_gdp_series.eps "fig:"){width="50.00000%"}
Figure \[fig2\] (right) shows the cointegration map for USIR. Considering the reliability of the estimates ($\sigma_{\gamma}\approx
0.035$ for USIR), the map produced allows a direct visualization of relationships through the whole set of time series without imposing any [*ad hoc*]{} classification criteria. The classification provided by SPC ($K=7$) is also shown as rectangles identified by $n1,...,n6$. As we have already discussed, correlation of the fluctuations and cointegration are different relationship measures. Two time series will pertain to the same cointegration group if they tend to orbit a common time series. The general pattern groups short term Treasury bills in $n4$, long term instruments, both treasury and corporate around $n1$ and finance company related instruments (interbank eurodollar (EDs), certificates of deposit (CDs) and finance company papers (FPs)) around $n3$.
Figure \[fig3\] (left) shows the SPINed cointegration map for monthly inflation data (IFR). Even though the estimates are less reliable in this case ($\sigma_{\gamma}\approx 0.085$) it is possible to identify groups by inspecting their mutual relationships represented by the color map. Figure \[fig3\] (right) allows a direct interpretations of segmentation provided. Group $n1$ consists of countries that exhibit volatile inflation profiles with both high inflation and high deflation periods (Benin (1a), Central African Republic (1b) and Syria (1c)). Group $n2$ is mainly composed by advanced economies with stable inflation patterns (USA (2a), Sweden (2b) and Japan (2c)) and countries that are very closely related to them (e.g. Martinique, Singapore and Bahamas). Group $n3$ consists of countries that have experienced hyperinflation in the period observed (Brazil (3a), Russia (3b) and Indonesia (3c)). Group $n4$ contains countries with very stable and low inflation profiles, among them New Zealand (4b), that has adopted inflation targeting as early as 1988, Australia (4a), that also has adopted inflation targeting in 1993 and Tuvalu (4c) that adopts the Australian dollar as currency. However, apart from Australia and New Zealand, all the other countries that have adopted inflation targeting before the period observed (1993-2004) (Canada, Finland, Korea, Sweden and United Kingdom) have been classified in the Group $n2$.
The cointegration map for GDP data (Fig. \[fig4\] (left)) must be dealt with care as this data set is smaller ($T=25$) and, therefore, statistically less reliable than the previous two sets ($\sigma_{\gamma}\approx 0.2$). To minimize interpretation problems due to GDP measurement issues we have selected from the IMF database $71$ countries that have had market economies in the period observed (1980-2004). As a criterion to classify different groups, we have looked at general interaction patterns compatible with the limited reliability of the estimates. The SPINned matrix shows that there are four distinguishable classes, but that their boundaries are not sharp. This illustrates again the difference between SPIN and traditional clustering techniques. For the latter either sharp boundaries (e.g. for hierarchical and K-means techniques) or some sort of low dimensional structure (e.g. fuzzy clustering) must be imposed even when there are none [@hastie]. We, therefore, have [*defined*]{} Group $n1$ as being composed by countries that interact with countries in Group $n2$. Group $n2$ consists of countries that interact with Group $n4$ and less strongly with Group $n3$. Group $n3$ is characterized by countries that do not interact with Group $n1$, interact strongly with Group $n4$ and less strongly with Group $n2$. Finally, Group $n4$ interacts with Groups $n2$ and $n3$ but not with Group $n1$. This procedure results in accelerating (Fig. \[fig4\] (right) Bangladesh (1a) and Tanzania (1b) ) or decelerating economies (Pakistan (1c)) in Group $n1$; low volatility economies both stable (Norway (2a) and United Kingdom (2b)) and decelerating (Japan (2c)) in Group $n2$; Group $n3$ concentrates highly volatile unstable economies including developing countries and all major oil producers (Kuwait (3a), Venezuela (3b) and Saudi Arabia (3c)); stable economies in Group $n4$ (US (4a), Australia (4b) and Italy (4c)). Notice that the difference between Groups $n4$ and $n2$ is their relation with Group $n1$, to say, the presence of some countries with accelerating or decelerating growth rates in $n2$.
Conclusion
==========
In this paper we have developed a simple measure for long term pairwise relationships in sets of time series by introducing a Bayesian estimate for a cointegration distance. For visualization of the relationships, with a minimum introduction of [*ad hoc*]{} structures, we have borrowed from the repertoire of Bioinformatics the SPIN ordering technique to produce [ *cointegration heat maps*]{}. We have exemplified the technique in three sets of time series of financial and economic interest and have been capable of identifying low-dimensional structures of economic sense emerging from the procedure.
Our aim in this work has been the development of tools that may be useful for discovering collective long term structures in economic time series. We think that a thorough understanding of the economic phenomena behind the observed patterns depends on our capability of describing the system interactions in some detail, what is out of the scope of the present work. Considering that socio-economic systems belong to a class of complex systems with unreliably known interactions and dynamics we regard pattern recognition tasks as hereby described a first important step towards a deeper quantitative understanding of such systems.
A previous version of this work has been presented at APFA5 in Torino. We wish to thank Eytan Domany and his collaborators for valuable comments on the manuscript and for gently providing a shareware license of the Analyzer and Sorter (SPIN) package we have employed to build SPINed cointegration heat maps. RV would like to thank the hospitality and financial support of the École de Physique Les Houches where part of this work has been completed. VBPL has been funded by [**FAPESP**]{} under research grant 05/58474-1. A detailed description of the clusters obtained as well as the scripts used for computing the cointegration measure will be available at the webpage of one of the authors (RV).
[30]{} J.-P. Bouchaud, M. Potters, Theory of Financial Risk and Derivative Pricing, Cambridge University Press (2003). R.F. Engle, C.W.J. Granger, Econometrica [**55**]{} (1987), 251-276. J.G. Hamilton, Time Series Analysis, Princeton University Press (1994). H. Hinrichsen, J. S. Weitz and E. Domany, J. Stat. Phys. 88, 617-636 (1997).
E.T. Jaynes, (G.L. Bretthorst ed.), Probability theory: the logic of science, Cambridge University Press (2003). K. Hubrich, H. Lütkepohl, P. Saikkonen, Econometrics Reviews [**20**]{} 3 (2001), 247-318. G. Koop et al., Bayesian Approaches to Cointegration in Palgrave Handbook of Econometrics: Econometric Theory Volume 1 (K. Patterson ed.), Palgrave Macmillan (2006). D. Tsafrir et al., Bioinformatics [**21**]{} 102005 (2005), 2301-2308. A.Y. Ng, M. I. Jordan, Y. Weiss, On Spectral Clustering: Analysis of an Algorithm, NIPS 2001. T. Di Matteo et al., Physica A [**355**]{} (2005), 21-33. M. Tumminello et al. PNAS [**102**]{} 30 (2005) 10421-10426. M. Blatt, S. Wiseman, E. Domany, Phys. Rev. Lett. [**76**]{} (1996) 3251-3254. L. Kullmann, J. Kertész, R.N. Mantegna, Physica A [**287**]{} (2000), 412-419. Available from http://www.federalreserve.gov. Available from http://www.clevelandfed.org. Available from http://www.imf.org. M. Boumans, Measurement [**38**]{} (2005) 275-284. N.R. Swanson, D. Van Dijk, Journal of Business and Economic Statistics [**24**]{}(1) (2006) 24-42. O. Morgenstern, On the Accuracy of Economic Observations, 2nd Ed., Princeton University Press (1991). T. Di Matteo, T. Aste, R.N. Mantegna, Physica A [**339**]{} (2004), 181-188. T. Hastie, R. Tibshirani, J.H. Friedman, The Elements of Statistical Learning, Springer (2003).
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
In this article, we focus on investigating the focusing Kundu-Eckhaus equation with nonzero boundary condition. A appropriate two-sheeted Riemann surface is introduced to map the spectral parameter $k$ into a single-valued parameter $z$. Starting from the Lax pair of Kundu-Eckhaus equation, two kind of Jost solutions are construed. Further their asymptotic, analyticity, symmetries as well as spectral matrix are detailed analyzed. It is shown that the solution of Kundu-Eckhaus equation with nonzero boundary condition can characterized with a matrix Riemann-Hilbert problem. Then a formula of $N$-soliton solutions is derived by solving Riemann-Hilbert problem. As applications, the first-order explicit soliton solution is obtained.\
\
\
Keywors: the focusing Kundu-Eckhaus equation; nonzero boundary conditions; Riemann-Hilbert problem; soliton solution.
author:
- 'Li-Li Wen$^1$, En-Gui Fan$^1$[^1]'
title: ' The Riemann-Hilbert approach to focusing Kundu-Eckhaus equation with nonzero boundary conditions '
---
Introduction
============
It is well-known that the nonlinear Schr$\ddot{o}$dinger equation plays an important role and achieved great success in physical fields such as nonlinear optics, nonlinear water waves, plasma physics and so on. But Schr$\ddot{o}$dinger equation with higher-order nonlinear terms, such as the self-steepening and self-frequency shift, has an significant effect in specific physical background, for example, optic fiber communication and Bose-Einstein condensates [@ky1985; @kl2006].
The modified Gross-Pitaevskii equation [@kvr2010] $$iu_{t}+u_{xx}+2f(t)|u|^{2}u+\frac{1}{2}\delta^{2}x^{2}u+\kappa_{1}|u|^{4}u+i\kappa_{2}(|u|^{2})_{x}u=0 \label{mgp}$$ was firstly proposed by Wadati for describing the interactions between two-body and three-body on the condensates. The variable parameters $\kappa_{1}$ and $\kappa_{2}$ are constants. The two-body scattering lengths $f(t)$ can be adjusted by Feshbach resonance. Neither $\kappa_{1}=0$ nor $\kappa_{2}=0$, equation (\[mgp\]) is suitable for Bose-Einstein condensates with higher densities. Reversely, if $\kappa_{1}=0$ and $\kappa_{2}=0$, equation (\[gp\]) only for lower densities in Bose-Einstein condensates. This situation can be described by Gross-Pitaevskii equation [@pcj2004] $$iu_{t}+u_{xx}+2f(t)|u|^{2}u+\frac{1}{2}\delta^{2}x^{2}u=0.\label{gp}$$ If $f(t)\in\mathbb{R}$ and independent of $t$, the equation (\[mgp\]) reduce to the Kundu-Eckhaus (KE) equation [@ka1984] $$iu_{t}+u_{xx}-2\sigma|u|^{2}u+4\beta^{2}|u|^{4}u+4i\beta\sigma(|u|^{2})_{x}u=0, \label{KE1}$$ with $u(x,t)$ being the complex potential function of spatial $x$ and temporal $t$ $u(x,t): \mathbb{R}^{2}\mapsto\mathbb{C}$. This equation is called the defocusing KE equation as $\sigma=1$ and focusing KE equation as $\sigma=-1$. The KE equation (\[KE1\]) was firstly put forward by Kundu to investigate the Landau-Lifshitz equations and derivative nonlinear Schr$\ddot{o}$dinger type equations [@ka1984]. For the special case $\beta=0$, the KE equation (\[KE1\]) reduces to nonlinear Schr$\ddot{o}$dinger equation. In recent years, the KE equation (\[KE1\]) has been investigated via different methods, for example, the long time asymptotic [@zqz2018; @wds2018; @wds2019; @gbl2018], the higher-order rogue wave solutions [@wx2014], rogue waves in a chaotic wave field [@bc2016], the Darboux transformation [@qd2015; @gbl2012], integrable discretizations [@dl2009]. Recently, the KE equation with zero boundary conditions was investigated by using Riemann-Hilbert method [@wds2018].
To our knowledge, there is still no work on investigating KE equation (\[KE1\]) with nonzero boundary conditions by using inverse scattering transformation or Riemannn-Hilbert method. In this paper, we would like to investigate the soliton solution of focusing KE equation (\[KE1\]) with the following nonzero boundary conditions $$\lim_{x\rightarrow\pm\infty}u(x,t)=u_{\pm}e^{2it(q_{0}^{2}-2\beta^{2}q_{0}^{4})+i\theta_{\pm}},\label{condition}$$ where $|u_{\pm}|=u_{0}>0$ and $u_{0}$ and $\theta_{\pm}$ are constants, $\theta_{\pm}$ is the arguments of $q_{\pm}$.
As we all known, the inverse scattering transform method plays an important role for finding the exact solutions of completely integrable systems [@csg1967]. The Riemann-Hilbert method as a new version of inverse scattering transform method streamline the research process and preferred by researchers. And Riemann-Hilbert method can be used to investigate the soliton solutions [@yjk2010] and the long-time asymptotic of integrable systems [@zqz2018]. Especially, in recent years, it has become a hot topic to investigate integrable systems with nonzero boundary conditions [@pb2006; @df2013; @kdk2016; @kd2015; @pb2015; @bp2015; @cv2015; @gb2015; @gb2016; @gb2014; @pm2017; @ls2018].
This work is organized as follows. In section 2, we analyze the spectral problem via introduce a transformation. And we introduced a appropriate Riemann surface for the single-valued function of the spectral parameter, that is $k$-plane mapped into $z$-plane. Section 3-5, we obtained the asymptotic, analyticity and symmetries of Jost solution and scattering matrix, which are used to get a Riemann-Hilbert problem in section 6. Section 7, we analyze the discrete spectrum and residue conditions which are used to solve the Riemann-Hilbert problem. Section 8, we establish the connection between the solution of KE equation and the solution of Rimann-Hilbert problem. A formula of the $N$-soliton solution of KE equation is obtained by using the Rimann-Hilbert problem. A conclusion is given in section 9.
Spectral Analysis
=================
It is well-known that the focusing KE equation ($\sigma=-1$) admits Lax pair $$(\partial_{x}-\mathcal{U})\phi=0,\ \ (\partial_{t}-\mathcal{V})\phi=0,\label{laxpair}$$ where $$\begin{aligned}
&&\mathcal{U}=-ik\sigma_{3}+i\beta |q|^{2}\sigma_{3}+U, \hspace{0.5cm} U=\left(
\begin{array}{cc}
0& u \\
-\overline{u} & 0
\end{array}\right), \ \ \ \sigma_{3}=\left(
\begin{array}{cc}
1 & 0 \\
0 & -1
\end{array}\right)\nonumber\\
&&\begin{split}\mathcal{V}=&-2ik^{2}\sigma_{3}+2kU+2\beta|u|^{2}U+i\left(|u|^{2}+4\beta^{2}|u|^{4}\right)\sigma_{3}-iU_{x}\sigma_{3}\\
&-\beta\left(UU_{x}-U_{x}U\right),
\end{split}
\nonumber\end{aligned}$$ where $k$ is the spectral parameter and $\overline{u}$ denotes the complex conjugation of $u$.
For convenience of using Riemann-Hilbert method, we first deal with Lax par (\[laxpair\]) and the boundary condition (\[condition\]). By making transformation $$q(x,t)=u(x,t)e^{-2it(q_{0}^{2}-2\beta^{2}q_{0}^{4})},\nonumber$$ the focusing KE equation can be reduced to the form $$iq_{t}+q_{xx}+2(|q|^{2}+2\beta^{2}|q|^{4}-q_{0}^{2}-2\beta^{2}q_{0}^{4})q-4i\beta(|q|^{2})_{x}q=0, \quad \beta\in\mathbb{R},\label{KE}$$ and corresponding boundary condition (\[condition\]) becomes $$\lim_{x\rightarrow\pm\infty}q(x,t)=q_{\pm}=q_{0}e^{i\theta_{\pm}},\nonumber$$ where we have denote $u_{\pm}$ and $u_{0}$ as $q_{\pm}$ and $ q_0$ respectively.
The equation (\[KE\]) admits the following Lax pair
$$\begin{aligned}
&(\partial_{x}-\mathcal{U})\phi=0,\label{LaxU}\\
&(\partial_{t}-\mathcal{V})\phi=0,\label{LaxV}\end{aligned}$$
\[Lax1\]
where $$\begin{aligned}
&&\mathcal{U}=-ik\sigma_{3}+i\beta |q|^{2}\sigma_{3}+Q, \hspace{0.5cm} Q=\left(
\begin{array}{cc}
0& q \\
-\overline{q} & 0
\end{array}\right),\nonumber\\
&&\begin{split}\mathcal{V}=&-2ik^{2}\sigma_{3}+2kQ+2\beta|q|^{2}Q+i\left(|q|^{2}+4\beta^{2}|q|^{4}-q_{0}^{2}-2\beta^{2}q_{0}^{4}\right)\sigma_{3}\\
&-iQ_{x}\sigma_{3}-\beta\left(QQ_{x}-Q_{x}Q\right).
\end{split}
\nonumber\end{aligned}$$
Let $x\rightarrow\pm\infty$ and by using boundary condition (\[condition\]), we obtain the following asymptotic Lax pair
$$\begin{aligned}
&(\partial_{x}-\mathcal{U}_{\pm})\phi=0, \quad \mathcal{U}_{\pm}=\lim_{x\rightarrow\pm\infty}\mathcal{U}=\left(-ik+i\beta q_{0}^{2}\right)\sigma_{3}+Q_{\pm},\label{LaxU2}\\
&(\partial_{t}-\mathcal{V}_{\pm})\phi=0, \quad \mathcal{V}_{\pm}=\lim_{x\rightarrow\pm\infty}\mathcal{V}=
\left(-2ik^{2}+2i\beta^{2}q_{0}^{4}\right)\sigma_{3}+\left(2k+2\beta q_{0}^{2}\right)Q_{\pm},\label{LaxV2}\end{aligned}$$
\[lax2\]
where $$\begin{aligned}
Q_{\pm}=\lim_{x\rightarrow\pm\infty}Q=\left(
\begin{array}{cc}
0& q_{\pm} \\
-\overline{q}_{\pm} & 0
\end{array}\right).
\nonumber\end{aligned}$$ it is obvious that $\mathcal{V}_{\pm}$ and $\mathcal{U}_{\pm}$ possess the linear relationship $$\mathcal{V}_{\pm}=\left(2k+2\beta q_{0}^{2}\right)\mathcal{U}_{\pm}.\nonumber$$
The fundamental matrix solution of asymptotic spectral problem (\[lax2\]) can be obtained as $$\phi_{\pm}(x,t,k)=\Xi_{\pm}(k)e^{-i\theta(x,t,k)\sigma_{3}}, \quad\quad k\neq\beta q_{0}^{2}\pm iq_{0},\label{phipm}$$ where $$\theta(x,t,k)=\lambda\left[x+(2k+2\beta q_{0}^{2})t\right], \quad \lambda=\sqrt{(k-\beta q_{0}^{2})^{2}+q_{0}^{2}},\nonumber$$ $$\begin{aligned}
\Xi_{\pm}=\left(
\begin{array}{cc}
1& \frac{-iq_{\pm}}{ \lambda+(k-\beta q_{0}^{2})} \\
\frac{-i\bar{q}_{\pm}}{\lambda+(k-\beta q_{0}^{2})} & 1
\end{array}\right).
\nonumber\end{aligned}$$
Since $\lambda$ is doubly branched with branch points are $k=\beta q_{0}^{2}\pm iq_{0}$, it is necessary to introduce a two-sheeted Riemann surface to such that $\lambda$ is a single-valued function on each sheet.
Denote $h=k-\beta q_{0}^{2}$ and let $$h+iq_{0}=\rho_{1}e^{i\theta_{1}}, \quad h-iq_{0}=\rho_{2}e^{i\theta_{2}}, \ \ -\frac{\pi}{2}\leq\theta_{1},\theta_{2}\leq\frac{3\pi}{2}, \nonumber$$ one can rewrite $\lambda$ on each sheet as $$\lambda_{\mathrm{I}}=\sqrt{\rho_{1}\rho_{2}}e^{i\frac{\theta_{1}+\theta_{2}}{2}},
\quad \lambda_{\mathrm{II}}=-\lambda_{\mathrm{I}}=\sqrt{\rho_{1}\rho_{2}}e^{i(\frac{\theta_{1}+\theta_{2}}{2})+i\pi}.\nonumber$$ And the branch cut of the Riemann surface is the segment $[-iq_{0},iq_{0}]$ in the complex $h$-plane.
Now we introduce a uniformization variable $$z=\lambda+h,$$ then its inverse transformation gives $$\lambda=\frac{1}{2}\big(z+ q_{0}^2/z\big), \quad
h=\frac{1}{2}\big(z-q_{0}^2/z\big), \quad
k=\frac{1}{2}\big(z-q_{0}^2/z\big)+\beta q_{0}^{2}.$$ Further we can show the following relations between the Riemann surface, the $h$-plane and the $z$-plane.
- [The region where $\mathrm{Im}\lambda>0$ come from the upper-half plane of the sheet-$\mathrm{I}$ and the lower-half plane of the sheet-$\mathrm{II}$. The region where $\mathrm{Im}\lambda<0$ come from the upper-half plane of the sheet-$\mathrm{II}$ and the lower-half plane of the sheet-$\mathrm{I}$.]{}
- [On the sheet-$\mathrm{I}$, $z\rightarrow\infty$ as $h\rightarrow\infty$, and on the sheet-$\mathrm{II}$, $z\rightarrow 0$ as $h\rightarrow\infty$.]{}
- [The real $\lambda$ (real $k$) axes is mapped into the real $z$ axes.]{}
- [The branch cut $[-iq_{0},iq_{0}]$ is mapped into the circle $C_{0}$ of the radius $q_{0}$ in $z$-plane.]{}
- [The sheet-$\mathrm{I}$ and sheet-$\mathrm{II}$, except for the branch cut, are mapped into the exterior and the interior of $C_{0}$, respectively.]{}
The jump contour in the complex $z$-plane is denoted by $\Sigma=\mathbb{R}\cup C_{0}$. The gray and white regions in Fig.1 denote $D^{+}$ and $D^{-}$, respectively $$\begin{split}
&D^{+}=\{z\in \mathbb{C}|\mathrm{Im}\lambda=\left(|z|^{2}-q_{0}^{2}\right)\mathrm{Im}z>0\},\\
&D^{-}=\{z\in \mathbb{C}|\mathrm{Im}\lambda=\left(|z|^{2}-q_{0}^{2}\right)\mathrm{Im}z<0\}.
\end{split}\nonumber$$
[]{}\
\
[]{}\
\
Based on the above results, we can rewrite the fundamental matrix solution (\[phipm\]) as $$\phi_{\pm}(x,t,z)=\Xi_{\pm}(z)e^{-i\theta(x,t,z)\sigma_{3}}, \quad z\neq iq_{0}\nonumber$$ where $$\begin{aligned}
\begin{split}
&\Xi_{\pm}(z)=\left(
\begin{array}{cc}
1& \frac{-iq_{\pm}}{z} \\
\frac{-i\bar{q}_{\pm}}{z} & 1
\end{array}\right)=\mathrm{I}-\frac{i}{z}\sigma_{3}Q_{\pm},
\nonumber\\
&\theta(x,t,z)=\frac{1}{2}\big(z+\frac{q_{0}^{2}}{z}\big)\left\{x+\Big[\big(z-\frac{q_{0}^{2}}{z}\big)+4\beta q_{0}^{2}\Big]t\right\}.\nonumber\\
\end{split}\end{aligned}$$ Direct calculation shows that $$\begin{split}
&\det\Xi_{\pm}=1+\frac{q_{0}^{2}}{z^{2}}\triangleq\gamma\neq0, \ \ \
\Xi_{\pm}^{-1}=\frac{1}{\gamma}\big(\mathrm{I}+\frac{i}{z}\sigma_{3}Q_{\pm}\big).\nonumber\\
\end{split}$$
Jost Solutions
==============
The Lax pair (\[Lax1\]) can be rewrite as
$$\begin{aligned}
&(\partial_{x}-\mathcal{U}_{\pm}-\Delta\hat{Q}_{\pm})\phi=0,\label{LaxU2}\\
&(\partial_{t}-\mathcal{V}_{\pm}-\Delta\hat{R}_{\pm})\phi=0,\label{LaxV2}\end{aligned}$$
\[Lax2\]
where $$\begin{aligned}
\begin{split}
&\Delta\hat{Q}_{\pm}=i\beta\left(|q|^{2}-q_{0}^{2}\right)\sigma_{3}+\Delta Q_{\pm}, \quad \Delta\hat{R}_{\pm}=\hat{R}-\hat{R}_{\pm},\nonumber\\
&\Delta Q_{\pm}=Q-Q_{\pm}, \quad \hat{R}_{\pm}=2kQ_{\pm}+2\beta q_{0}^{2}Q_{\pm}+2i\beta^{2}q_{0}^{4}\sigma_{3},\nonumber\\
&\begin{split}\hat{R}=&2kQ+2\beta|q|^{2}Q+i\left(|q|^{2}+4\beta^{2}|q|^{4}-q_{0}^{2}-2\beta^{2}q_{0}^{4}\right)\sigma_{3}
-iQ_{x}\sigma_{3}\\
&-\beta\left(QQ_{x}-Q_{x}Q\right).\end{split}\nonumber
\end{split}\end{aligned}$$ Now, one can define the Jost solutions as the simultaneous solutions of Lax pair (\[Lax1\]) such that $$\Psi_{\pm}(x,t,z)\sim\Xi_{\pm}(z)e^{-i\theta(x,t,z)\sigma_{3}}, \quad z\in\Sigma, \quad x\rightarrow\pm\infty,$$ and the modified Jost solutions $$\begin{aligned}
\begin{split}
e^{\int_{-\infty}^{x}i\beta\left(|q|^{2}-q_{0}^{2}\right) dy\hat{\sigma}_{3}}\mu_{\pm}(x,t,z)e^{-\int_{x}^{+\infty}i\beta\left(|q|^{2}-q_{0}^{2}\right) dy\sigma_{3}}=&\Psi_{\pm}(x,t,z)e^{i\theta(x,t,z)\sigma_{3}}\\
=&\eta_{\pm}(x,t,z),\label{mupsi}
\end{split}\end{aligned}$$ Then the Lax pair (\[Lax2\]) can be rewritten as $$\begin{aligned}
\begin{split}
&\left(\Xi_{\pm}^{-1}\eta_{\pm}(x,t,z)\right)_{x}=-i\lambda\left[\sigma_{3},\Xi_{\pm}^{-1}\eta_{\pm}(x,t,z)\right]
+\Xi_{\pm}^{-1}\Delta\hat{Q}_{\pm}\eta_{\pm}(x,t,z),\\
&\begin{split}\left(\Xi_{\pm}^{-1}\eta_{\pm}(x,t,z)\right)_{t}
=&-i\lambda\left(2k+2\beta q_{0}^{2}\right)\left[\sigma_{3},\Xi_{\pm}^{-1}\eta_{\pm}(x,t,z)\right]\\
&+\Xi_{\pm}^{-1}\Delta\hat{R}_{\pm}\eta_{\pm}(x,t,z).\end{split}
\end{split}\label{Lax3}\end{aligned}$$ It is easily known that Lax pair (\[Lax3\]) can be written in full derivative form $$d\left(e^{i\theta(x,t,z)\hat{\sigma}_{3}}\Xi_{\pm}^{-1}\eta_{\pm}(x,t,z)\right)
=e^{i\theta(x,t,z)\hat{\sigma}_{3}}\left(V_{1}dx+V_{2}dt\right),$$ where $$\begin{aligned}
\begin{split}
&V_{1}=\Xi_{\pm}^{-1}\Delta\hat{Q}_{\pm}\eta_{\pm}(x,t,z),\\
&V_{2}=\Xi_{\pm}^{-1}\Delta\hat{R}_{\pm}\eta_{\pm}(x,t,z).
\end{split}\nonumber\end{aligned}$$ We can obtain the Volterra integral equations $$\begin{aligned}
\begin{split}
&\begin{split}\mu_{-}(x,t,z)=\Xi_{-}+\int_{-\infty}^{x}&e^{-\int_{-\infty}^{x}i\beta\left(|q|^{2}-q_{0}^{2}\right) dy\sigma_{3}}\Xi_{-}e^{i\lambda(x'-x)\hat{\sigma}_{3}}\\
&\Xi_{-}^{-1}\Delta\hat{Q}_{-}e^{\int_{-\infty}^{x'}i\beta\left(|q|^{2}-q_{0}^{2}\right) dy\sigma_{3}}\mu_{-}dx',
\end{split}\\
&\begin{split}\mu_{+}(x,t,z)=\Xi_{+}-\int_{x}^{+\infty}&e^{-\int_{-\infty}^{x}i\beta\left(|q|^{2}-q_{0}^{2}\right) dy\sigma_{3}}\Xi_{+}e^{i\lambda(x'-x)\hat{\sigma}_{3}}\\
&\Xi_{+}^{-1}\Delta\hat{Q}_{+}e^{\int_{-\infty}^{x'}i\beta\left(|q|^{2}-q_{0}^{2}\right) dy\sigma_{3}}\mu_{+}dx'.\end{split}
\end{split}\end{aligned}$$
It can be shown that the first column of $\mu_{-}$ is analytically extended to $D^{+}$ and the second column of $\mu_{-}$ is analytically extended to $D^{-}$. Similarly, the first column of $\mu_{+}$ can be analytically extended to $D^{-}$ and the second column of $\mu_{+}$ can be analytically extended to $D^{+}$. It can be summarized as follows $$\begin{split}
D^{+}: \quad \mu_{-,1}, \quad \mu_{+,2};\\
D^{-}: \quad \mu_{-,2}, \quad \mu_{+,1},
\end{split}\nonumber$$ where the subscripts ‘1’ and ‘2’ identify the columns of matrix.
Consider the Laurent expansion of $\Xi_{\pm}^{-1}\eta_{\pm}(x,t,z)$ in system (\[Lax3\])
$$\begin{aligned}
&\Xi_{\pm}^{-1}\eta_{\pm}(x,t,z)=\alpha^{(0)}+\frac{\alpha^{(1)}}{z}+\frac{\alpha^{(2)}}{z^{2}}+O\big(\frac{1}{z^{3}}\big),\quad\quad z\rightarrow\infty,\\
&\Xi_{\pm}^{-1}\eta_{\pm}(x,t,z)=\eta^{(0)}+\eta^{(1)}z+\eta^{(2)}z^{2}+O\left(z^{3}\right), \quad z\rightarrow 0.\end{aligned}$$
where $\alpha^{(0)},\alpha^{(1)},\alpha^{(2)},\cdots$ and $\eta^{(0)},\eta^{(1)},\eta^{(2)},\cdots$are independent of $z$. Substituting the above expansion into (\[Lax3\]) and comparing the coefficients of $z^{n} (n=0,\pm1,\pm2,\cdots)$, we obtain the following equations $$\begin{aligned}
\begin{split}
&\left[\sigma_{3},\alpha^{(0)}\right]=0,\nonumber\\
&\alpha^{(0)}_{x}=-\frac{i}{2} \left[\sigma_{3},\alpha^{(1)}\right]+i\beta\left(|q|^{2}-q_{0}^{2}\right)\sigma_{3}\alpha^{(0)}+\Delta Q_{\pm}\alpha^{(0)},\nonumber\\
&\begin{split}\alpha^{(1)}_{x}=&-\frac{i}{2}q_{0}^{2} \left[\sigma_{3},\alpha^{(0)}\right]-\frac{i}{2}\left[\sigma_{3},\alpha^{(2)}\right]
+\beta\left(|q|^{2}-q_{0}^{2}\right)\left[\sigma_{3},\sigma_{3}Q_{\pm}\right]\alpha^{(0)}\\
&+i\left[\sigma_{3}Q_{\pm},\Delta Q_{\pm}\right]\alpha^{(0)}+i\beta\left(|q|^{2}-q_{0}^{2}\right)\sigma_{3}\alpha^{(1)}+\Delta Q_{\pm}\alpha^{(1)},
\end{split}\nonumber
\end{split}\end{aligned}$$ $$\begin{aligned}
\begin{split}
&\left[\sigma_{3},\eta^{(0)}\right]=0,\nonumber\\
&\eta^{(0)}_{x}=-\frac{i}{2}q_{0}^{2} \left[\sigma_{3},\eta^{(1)}\right]+\frac{1}{q_{0}^{2}}i\beta\left(|q|^{2}-q_{0}^{2}\right)\sigma_{3}\eta^{(0)}+\frac{1}{q_{0}^{2}}\sigma_{3}Q_{\pm}\Delta Q_{\pm}\sigma_{3}Q_{\pm}\eta^{(0)},\nonumber\\
&\begin{split}\eta^{(1)}_{x}=&-\frac{i}{2} \left[\sigma_{3},\eta^{(0)}\right]-\frac{i}{2}q_{0}^{2}\left[\sigma_{3},\eta^{(2)}\right]
+\frac{1}{q_{0}^{2}}\beta\left(|q|^{2}-q_{0}^{2}\right)\left[\sigma_{3},\sigma_{3}Q_{\pm}\right]\eta^{(0)}\\
&+\frac{i}{q_{0}^{2}}\left[\sigma_{3}Q_{\pm},\Delta Q_{\pm}\right]\eta^{(0)}+\frac{1}{q_{0}^{2}}i\beta\left(|q|^{2}-q_{0}^{2}\right)\sigma_{3}Q_{\pm}^{2}\eta^{(1)}\\
&+\frac{1}{q_{0}^{2}}\sigma_{3}Q_{\pm}\Delta Q_{\pm}\sigma_{3}Q_{\pm}\eta^{(1)},
\end{split}\nonumber
\end{split}\end{aligned}$$ from which, we can derive the following results $$\begin{aligned}
\begin{split}
&\alpha^{(0)}=e^{-\int_{x}^{+\infty}i\beta\left(|q|^{2}-q_{0}^{2}\right)dy\sigma_{3}},\\
&\alpha^{(1)}_{o}=-i\sigma_{3}\Delta Q_{\pm}\alpha^{(0)}_{d},\\
&\alpha^{(1)}_{d,x}=i\left[\sigma_{3}Q_{\pm},\Delta Q_{\pm}\right]\alpha^{(0)}_{d}+i\beta\left(|q|^{2}-q_{0}^{2}\right)\sigma_{3}\alpha^{(1)}_{d}-i\Delta Q_{\pm}\sigma_{3}\Delta Q_{\pm}\alpha^{(0)}_{d}.
\end{split}\nonumber\end{aligned}$$ and $$\begin{aligned}
\begin{split}
&\eta^{(0)}=e^{\int_{x}^{+\infty}i\beta\left(|q|^{2}-q_{0}^{2}\right)dy\sigma_{3}},\\
&\begin{split}\eta^{(1)}_{d,x}=&\frac{i}{q_{0}^{2}}\left[\sigma_{3}Q_{\pm},\Delta Q_{\pm}\right]\eta^{(0)}_{d}-i\beta\left(|q|^{2}-q_{0}^{2}\right)\sigma_{3}\eta^{(1)}_{d}\\
&+\frac{1}{q_{0}^{4}}\sigma_{3}Q_{\pm}\Delta Q_{\pm}\sigma_{3}\Delta Q_{\pm}\sigma_{3}Q_{\pm}\eta^{(0)}_{d}.
\end{split}
\end{split}\nonumber\end{aligned}$$ And the asymptotic of the modified Jost solutions for $z\rightarrow\infty$ and $z\rightarrow 0$ are respectively derived as $$\begin{aligned}
\begin{split}
&\begin{split}\mu_{\pm}=&I-\frac{i}{z}e^{-\int_{-\infty}^{x}i\beta\left(|q|^{2}-q_{0}^{2}\right)dy\hat{\sigma}_{3}}\sigma_{3}Q_{\pm}\\
&+\frac{1}{z}e^{-\int_{-\infty}^{x}i\beta\left(|q|^{2}-q_{0}^{2}\right)dy\hat{\sigma}_{3}}\alpha^{(1)}e^{\int_{x}^{+\infty}i\beta\left(|q|^{2}-q_{0}^{2}\right)dy\sigma_{3}}+O\big(\frac{1}{z^{2}}\big), \quad z\rightarrow\infty,\end{split}\\
&\mu_{\pm}=-\frac{i}{z}\sigma_{3}Q_{\pm}+O\left(1\right), \quad z\rightarrow 0.
\end{split}\label{muexp}\end{aligned}$$
Scattering Matrix and Asymptotic
================================
It is easy to check that $\mathrm{tr}\mathcal{U}=\mathrm{tr}\mathcal{V}=0$, then by using Abel formula, we have $$(\det\Psi_{\pm})_{x}=(\det\Psi_{\pm})_{t}=0,$$ which implies that $$\begin{aligned}
&\det\Psi_{\pm}=\det Y_{\pm}=\gamma\label{ppo}\end{aligned}$$ by using asymptotic at $x\rightarrow \pm \infty$. Since $\Psi_{+}$ and $\Psi_{-}$ are the fundamental solutions of the spectral problem (\[Lax1\]), they satisfy the following linear relationship $$\Psi_{+}(z)=\Psi_{-}(z)S(z), \quad z\in\Sigma\setminus\{\pm iq_{0}\},\label{S}$$ where $S(z)=(s_{ij}(z))_{2\times 2}$ is called spectral matrix. From (\[ppo\]), we know that $\det S=1$. The relation formula (\[S\]) can be expanded as $$\begin{aligned}
\Psi_{+,1}=s_{11}\Psi_{-,1}+s_{21}\Psi_{-,2}, \quad
\Psi_{+,2}=s_{12}\Psi_{-,1}+s_{22}\Psi_{-,2}.\label{psis}\end{aligned}$$ The reflection coefficients are defined as $$\rho(z)=\frac{s_{21}}{s_{11}}, \quad \tilde{\rho}(z)=\frac{s_{12}}{s_{22}}.\label{rho}$$ According to (\[psis\]) the scattering coefficients have the following Wronskian representations
$$\begin{aligned}
&s_{11}=\frac{\mathrm{Wr}\left(\Psi_{+,1},\Psi_{-,2}\right)}{\gamma}, \quad s_{12}=\frac{\mathrm{Wr}\left(\Psi_{+,2},\Psi_{-,2}\right)}{\gamma},\\
&s_{21}=\frac{\mathrm{Wr}\left(\Psi_{-,1},\Psi_{+,1}\right)}{\gamma}, \quad s_{22}=\frac{\mathrm{Wr}\left(\Psi_{-,1},\Psi_{+,2}\right)}{\gamma}.\end{aligned}$$
\[sij\]
The equation (\[sij\]) implies that $s_{11}$ is analytic in $D^{-}$ and $s_{22}$ is analytic in $D^{+}$. However, $s_{12}$ and $s_{22}$ are just continuous on $\Sigma$.
**Proposition 1.** The asymptotic behaviors of the scattering matrix $S(z)$ are given as follows
$$\begin{aligned}
&S(z)=I+O\big(\frac{1}{z}\big),\quad z\rightarrow\infty,\label{s1}\\
&S(z)=\mathrm{diag}\left( q_{-}/q_{+}, q_{+}/q_{-}\right)+O\left(z\right),\quad z\rightarrow 0.\label{s2}\end{aligned}$$
\[s\]
By using (\[mupsi\]), (\[sij\]) and the asymptotic behaviors of $\mu_{\pm}$, we can prove that
As $z\rightarrow\infty$, $$\begin{aligned}
\begin{split}
s_{11}=&\frac{Wr\left(\Psi_{+,1},\Psi_{-,2}\right)}{\gamma} = \frac{1 }{1+\frac{q_{0}^{2}}{z^{2}}} \det
\left(
\begin{array}{cc}
\mu_{+,11}& \mu_{-,12}\\
\mu_{+,21} & \mu_{-,22}
\end{array}\right)\nonumber\\
=&\det
\left(
\begin{array}{cc}
1+O(\frac{1}{z})& O(\frac{1}{z})\\
O(\frac{1}{z}) & 1+O(\frac{1}{z})
\end{array}\right)\big(1-\frac{q_{0}^{2}}{z^{2}}+\frac{q_{0}^{4}}{z^{4}}-\cdots\big)\\
=&\big(1+O\big(\frac{1}{z}\big)\big)\big(1-\frac{q_{0}^{2}}{z^{2}}+\cdots\big)\\
=&1+O\big(\frac{1}{z}\big).
\end{split}\end{aligned}$$
As $z\rightarrow 0$, $$\begin{aligned}
\begin{split}
s_{11}=&\frac{Wr\left(\Psi_{+,1},\Psi_{-,2}\right)}{\gamma}=\frac{ 1}{1+\frac{q_{0}^{2}}{z^{2}}} \det
\left(
\begin{array}{cc}
\mu_{+,11}& \mu_{-,12} \\
\mu_{+,21} & \mu_{-,22}
\end{array}\right)\nonumber\\
=&\det
\left(
\begin{array}{cc}
O(1)& -\frac{i}{z}q_{-}\\
\frac{i}{z}\overline{q}_{+} & O(1)
\end{array}\right)\frac{z^{2}}{q_{0}^{2}}\big(1-\frac{z^{2}}{q_{0}^{2}}+\cdots\big)\\
=&\big(O(1)+\frac{1}{z^{2}}q_{-}\overline{q}_{+}\big)\big(\frac{z^{2}}{q_{0}^{2}}-\frac{z^{4}}{q_{0}^{4}}+\cdots\big)\\
=&\frac{q_{-}\overline{q}_{+}}{q_{0}^{2}}+O\left(z\right)=\frac{q_{-}}{q_{+}}+O(z).
\end{split}\end{aligned}$$ The asymptotic behaviors of $s_{22}$, $s_{12}$, and $s_{21}$ can also be derived by the similar way.
Symmetries
==========
For the focusing KE equation with nonzero boundary conditions, the Jost functions $\Psi_{\pm}(z)$ and spectral matrix $S(z)$ possess two kinds of symmetries.
First Symmetry
--------------
Here we consider the symmetries between two points $z\mapsto \overline{z}$ (upper/lower half plane).\
**Proposition 2.** For $z\in \Sigma$,\
(1) The Jost solutions satisfy the symmetries
$$\begin{aligned}
&\Psi_{\pm}(z)=\sigma_{2}\overline{\Psi_{\pm}(\overline{z})}\sigma_{2},\\ &\Psi_{\pm,1}(z)=i\sigma_{2}\overline{\Psi_{\pm,2}(\overline{z})}, \\ &\Psi_{\pm,2}=-i\sigma_{2}\overline{\Psi_{\pm}(\overline{z})}.\end{aligned}$$
\[psisym1\]
\(2) The scattering matrix satisfy the symmetries
$$\begin{aligned}
&S(z)=\sigma_{2}\overline{S(\overline{z})}\sigma_{2},\\
&s_{11}(z)=\overline{s_{22}(\overline{z})}, \\
&s_{12}(z)=-\overline{s_{21}(\overline{z})}.\label{ssym1}\end{aligned}$$
\(3) The reflection coefficient satisfy the symmetries $$\begin{aligned}
\rho(z)=-\overline{\tilde{\rho}(\overline{z})}.\label{rhosym1}\end{aligned}$$
\(1) The $\mathcal{U}$ and $\mathcal{V}$ in the Lax pair (\[Lax1\]) with the following symmetries on $z$-plane $$\overline{\mathcal{U}(\overline{z})}=\sigma_{2}\mathcal{U}(z)\sigma_{2}, \quad \overline{\mathcal{V}(\overline{z})}=\sigma_{2}\mathcal{V}(z)\sigma_{2},\nonumber$$ by which, we can show that $$\begin{aligned}
\begin{split}
&\big(\sigma_{2}\overline{\Psi_{\pm}(\overline{z})}\sigma_{2}\big)_{x}=\mathcal{U}(z)\big(\sigma_{2}\overline{\Psi_{\pm}(\overline{z})}\sigma_{2}\big),\\
&\big(\sigma_{2}\overline{\Psi_{\pm}(\overline{z})}\sigma_{2}\big)_{t}=\mathcal{V}(z)\big(\sigma_{2}\overline{\Psi_{\pm}(\overline{z})}\sigma_{2}\big),
\end{split}\nonumber\end{aligned}$$ Further, by suing asymptotic $$\Psi_{\pm}(z), \ \sigma_{2}\overline{\Psi_{\pm}(\overline{z})}\sigma_{2}\sim\Xi_{\pm}(z)e^{-i\lambda\left[x+(2k+2\beta q_{0}^{2})t\right]\sigma_{3}}, \ \ x\rightarrow \infty,\nonumber$$ we obtain the symmetry (\[psisym1\]).\
(2) On the basis of (\[S\]), $$\begin{aligned}
\begin{split}
S(z)=&\Psi_{-}^{-1}(z)\Psi_{+}(z)\\
=&\sigma_{2}\overline{\Psi_{-}(\overline{z})}^{-1}\sigma_{2}\sigma_{2}\overline{\Psi_{+}(\overline{z})}\sigma_{2}\\
=&\sigma_{2}\overline{\Psi_{-}(\overline{z})}^{-1}\overline{\Psi_{+}(\overline{z})}\sigma_{2}\\
=&\sigma_{2}\overline{S(\overline{z})}\sigma_{2}.
\end{split}\nonumber\end{aligned}$$ Then (\[ssym1\]) can be obtained.\
(3) $$\rho(z)=\frac{s_{21}(z)}{s_{11}(z)}=-\frac{\overline{s_{12}(\overline{z})}}{\overline{s_{22}(\overline{z})}}=-\overline{\tilde{\rho}(\overline{z})}.\nonumber$$
Second Symmetry
---------------
Here we consider the symmetries between two points $z\mapsto -\frac{q_{0}^{2}}{z}$ (outside/inside of the circle $C_{0})$.\
**Proposition 3.** For $z\in \Sigma$,\
(1) The Jost solutions satisfy the following symmetries
$$\begin{aligned}
&\Psi_{\pm}(z)=-\frac{i}{z}\Psi_{\pm}\big(-\frac{q_{0}^{2}}{z}\big)\sigma_{3}Q_{\pm},\\ &\Psi_{\pm,1}(z)=-\frac{i}{z}\overline{q}_{\pm}\Psi_{\pm,2}\big(-\frac{q_{0}^{2}}{z}\big), \\ &\Psi_{\pm,2}(z)=-\frac{i}{z}q_{\pm}\Psi_{\pm,1}\big(-\frac{q_{0}^{2}}{z}\big),\end{aligned}$$
\[psisym2\]
\(2) The scattering matrix satisfies the symmetries
$$\begin{aligned}
&S(z)=\left(\sigma_{3}Q_{-}\right)^{-1}S\big(-\frac{q_{0}^{2}}{z}\big)\sigma_{3}Q_{+}, \\
&s_{11}\big(-\frac{q_{0}^{2}}{z}\big)=\frac{q_{-}}{q_{+}}s_{22}(z), \\
&s_{12}\big(-\frac{q_{0}^{2}}{z}\big)=\frac{q_{+}}{\overline{q}_{-}}s_{21}(z).\end{aligned}$$
\[ssym1\]
\(3) The reflection coefficient satisfies the symmetries $$\rho(-\frac{q_{0}^{2}}{z})=-\frac{\overline{q}_{-}}{q_{-}}\overline{\rho(\overline{z})}.\label{rhosym1}$$
The process of proof is similar with the Proposition 2.
Riemann-Hilbert Problem
=======================
Based on the analytical and asymmetry properties of eigenfunctions $\mu_\pm$ and $S(z)$, we derive Riemann-Hilbert problem associated with KE equation with nonzero boundary conditions.\
**Proposition 4.** Define sectionally meromorphic matrix $$\begin{aligned}
M(x,t,z)=\begin{cases}
\begin{split}
&M^{+}(x,t,z)=\left(
\begin{array}{cc}
\mu_{-,1}& \frac{\mu_{+,2}}{s_{22}}
\end{array}\right), \quad z\in D^{+},\\
&M^{-}(x,t,z)=\left(
\begin{array}{cc}
\frac{\mu_{+,1}}{s_{11}}& \mu_{-,2}
\end{array}\right), \quad z\in D^{-},
\end{split}
\end{cases}\label{M}\end{aligned}$$ then we have the following Riemann-Hilbert problem
- [Analyticity: $M(x,t,z)$ is meromorphic in $D^{+}\cup D^{-}$ and has simple poles.]{}
- [Jump condition: $$M^{-}(x,t,z)=M^{+}(x,t,z)\left(I-G(x,t,z)\right), \quad z\in\Sigma,\nonumber$$ where $$\begin{aligned}
G(x,t,z)=e^{-i\theta\sigma_{3}}\left(
\begin{array}{cc}
\rho\tilde{\rho}& \tilde{\rho} \\
-\rho & 0
\end{array}\right)e^{i\theta\sigma_{3}}.\nonumber\end{aligned}$$]{}
- [Asymptotic behavior: $$\begin{aligned}
\begin{split}
&M^{\pm}(x,t,z)=I+O\big(\frac{1}{z}\big),\quad z\rightarrow\infty\\
&M^{\pm}(x,t,z)=-\frac{i}{z}\sigma_{3}Q_{-}+O(1).\quad z\rightarrow 0
\end{split}\nonumber\end{aligned}$$]{}
The analyticity can be derived from (\[psis\]) and the analyticity of $\mu_{\pm}$.
From (\[psis\]), we can known that $$\begin{aligned}
\begin{split}
&\frac{\Psi_{+,1}}{s_{11}}=\left(1-\rho\tilde{\rho}\right)\Psi_{-,1}+\rho\frac{\Psi_{+,2}}{s_{22}},\\
&\Psi_{-,2}=-\tilde{\rho}\Psi_{-,1}+\frac{\Psi_{+,2}}{s_{22}},
\end{split}\nonumber\end{aligned}$$ which lead to $$\begin{aligned}
\left(
\begin{array}{cc}
\frac{\Psi_{+,1}}{s_{11}}& \Psi_{-,2}
\end{array}\right)=\left(
\begin{array}{cc}
\Psi_{-,1}&\frac{\Psi_{+,2}}{s_{22}}
\end{array}\right)\left(
\begin{array}{cc}
1-\rho\tilde{\rho}&-\tilde{\rho}\\
\rho&1
\end{array}\right).\nonumber\end{aligned}$$ Then the jump condition can be derived as $$\begin{aligned}
\begin{split}M^{-}(x,t,z)=&\left(
\begin{array}{cc}
\frac{\mu_{+,1}}{s_{11}}& \mu_{-,2}
\end{array}\right)\\
=&\left(
\begin{array}{cc}
\mu_{-,1}&\frac{\mu_{+,2}}{s_{22}}
\end{array}\right)e^{-i\theta\sigma_{3}}\left(
\begin{array}{cc}
1-\rho\tilde{\rho}&-\tilde{\rho}\\
\rho&1
\end{array}\right)e^{i\theta\sigma_{3}}\\
=&M^{+}(x,t,z)\left(I-G(x,t,z)\right).
\end{split}\nonumber\end{aligned}$$ Now we proof the asymptotic behavior. From (\[M\]), $M^{+}(x,t,z)$ with the following form $$\begin{aligned}
\begin{split}
M^{+}(x,t,z)=&\left(
\begin{array}{cc}
\mu_{-,1}&\mu_{+,2}
\end{array}\right)\left(
\begin{array}{cc}
1&0\\
0&\frac{1}{s_{22}}
\end{array}\right).
\end{split}\nonumber\end{aligned}$$ As $z\rightarrow\infty$, $$\begin{aligned}
\begin{split}
M^{+}(x,t,z)=&\left(I+O\big(\frac{1}{z}\big)\right)\left(
\begin{array}{cc}
1&0\\
0&\frac{1}{1+O\left(\frac{1}{z}\right)}
\end{array}\right)\\
=&I+O\big(\frac{1}{z}\big).
\end{split}\nonumber\end{aligned}$$ As $z\rightarrow 0$, $$\begin{aligned}
\begin{split}
M^{+}(x,t,z)=&\left(
\begin{array}{cc}
O(1)&-\frac{i}{z}q_{+}+O(1)\\
-\frac{i}{z}\overline{q}_{-}+O(1)&O(1)
\end{array}\right)\left(
\begin{array}{cc}
1&0\\
0&\frac{1}{\frac{q_{+}}{q_{-}}+O(z)}
\end{array}\right)\\
=&-\frac{i}{z}\sigma_{3}Q_{-}+O(1).
\end{split}\nonumber\end{aligned}$$ The asymptotic behavior of $M^{-}(x,t,z)$ can be derived in the similar way.
Discrete Spectrum and Residue Conditions
========================================
If $s_{22}(z_{n})=0$, we can derive from the forth equation of (\[sij\]) that the eigenfunctions $\Psi_{+,2}(x,t,z_{n})$ and $\Psi_{-,1}(x,t,z_{n})$ must be proportional $$\Psi_{+,2}(x,t,z_{n})=b_{n}\Psi_{-,1}(x,t,z_{n}),\label{psib}$$ where $b_{n}$ is independent of $x$, $t$ and $z$. Let $s_{22}$ has a finite number of simple zeros $z_{1}, z_{2},\cdots, z_{N}$ in $D^{+}\cap\{z\in\mathbb{C}|\mathrm{Im}z>0\}$. According to the symmetry properties of $S(z)$, we have that $$s_{22}(z_{n})=0\Longleftrightarrow s_{11}(\overline{z}_{n})=0\Longleftrightarrow s_{11}(-\frac{q_{0}^{2}}{z_{n}})=0\Longleftrightarrow s_{22}(-\frac{q_{0}^{2}}{\overline{z}_{n}})=0.\nonumber$$ So the discrete spectrum accumulate the set $$\mathcal{Z}=\{z_{n},\overline{z}_{n},-\frac{q_{0}^{2}}{z_{n}},-\frac{q_{0}^{2}}{\overline{z}_{n}}\}_{n=1}^{N}.\nonumber$$
Now we study the residue condition that is very important to solve the Riemann-Hilbert problem. The equation (\[psib\]) can be rewritten as $$\mu_{+,2}(z_{n})=b_{n}\mu_{-,1}(z_{n})e^{-2i\theta(z_{n})}.\nonumber$$ Then $$\begin{aligned}
\begin{split}
\mathop{\mathrm{Res}}\limits_{z=z_{n}}\left[\frac{\mu_{+,2}(z)}{s_{22}(z)}\right]=&\frac{b_{n}\mu_{-,1}(z_{n})e^{-2i\theta(z_{n})}}{s_{22}^{'}(z_{n})}=A_{n}\mu_{-,1}(z_{n})e^{-2i\theta(z_{n})},
\end{split}\end{aligned}$$ where $A_{n}=\frac{b_{n}}{s_{22}^{'}(z_{n})}$.
If $s_{11}(\overline{z}_{n})=0$, similarly, the first equation of (\[sij\]) implies that the eigenfunctions $\Psi_{+,1}(x,t,\overline{z}_{n})$ and $\Psi_{-,2}(x,t,\overline{z}_{n})$ must be proportional $$\Psi_{+,1}(x,t,\overline{z}_{n})=\tilde{b}_{n}\Psi_{-,2}(x,t,\overline{z}_{n}),\label{psib1}$$ (\[psib1\]) also can be rewritten as $$\mu_{+,1}(\overline{z}_{n})=\tilde{b}_{n}\mu_{-,2}(\overline{z}_{n})e^{2i\theta(\overline{z}_{n})}.\nonumber$$ And derived that $$\begin{aligned}
\begin{split}
\mathop{\mathrm{Res}}\limits_{z=\overline{z}_{n}}\left[\frac{\mu_{+,1}(z)}{s_{11}(z)}\right]=&\frac{\tilde{b}_{n}\mu_{-,2}
(\overline{z}_{n})e^{-2i\theta(\overline{z}_{n})}}{s_{11}^{'}(\overline{z}_{n})}=\tilde{A}_{n}\mu_{-,2}(\overline{z}_{n})e^{2i\theta(\overline{z}_{n})},
\end{split}\end{aligned}$$ where $\tilde{A}_{n}=\frac{\tilde{b}_{n}}{s_{11}^{'}(\overline{z}_{n})}$. And it is easy to show that $$\tilde{A}_{n}=-\overline{A}_{n}.$$ Combining (\[psisym2\]), (\[psib\]) and (\[psib1\]) we have the following relations
$$\begin{aligned}
&\Psi_{+,2}\big(-\frac{q_{0}^{2}}{\overline{z}_{n}}\big)=\frac{q_{-}}{\overline{q}_{+}}\tilde{b}_{n}\Psi_{-,1}\big(-\frac{q_{0}^{2}}{\overline{z}_{n}}\big),\\
&\Psi_{+,1}\big(-\frac{q_{0}^{2}}{z_{n}}\big)=\frac{\overline{q}_{-}}{q_{+}}b_{n}\Psi_{-,2}\big(-\frac{q_{0}^{2}}{z_{n}}\big).\end{aligned}$$
Using the symmetries of $s_{ij}$, we have results
$$\begin{aligned}
&\mathop{\mathrm{Res}}\limits_{z=-\frac{q_{0}^{2}}{\overline{z}_{n}}}\left[\frac{\mu_{+,2}(z)}{s_{22}(z)}\right]
=A_{N+n}\mu_{-,1}(-\frac{q_{0}^{2}}{\overline{z}_{n}})e^{-2i\theta(-\frac{q_{0}^{2}}{\overline{z}_{n}})},\\
&\mathop{\mathrm{Res}}\limits_{z=-\frac{q_{0}^{2}}{z_{n}}}\left[\frac{\mu_{+,1}(z)}{s_{11}(z)}\right]
=\tilde{A}_{N+n}\mu_{-,2}(-\frac{q_{0}^{2}}{z_{n}})e^{2i\theta(-\frac{q_{0}^{2}}{z_{n}})}.\end{aligned}$$
where $$A_{N+n}=\frac{q_{-}}{\overline{q}_{-}}\frac{q_{0}^{2}}{\overline{z}_{n}^{2}}\tilde{A}_{n}, \quad \tilde{A}_{N+n}=\frac{\overline{q}_{-}}{q_{-}}\frac{q_{0}^{2}}{z_{n}^{2}}A_{n},$$ and $\tilde{A}_{N+n}=-\overline{A}_{N+n}$.
N-soliton solutions of the KE equation
======================================
To obtain solution of KE equation with nonzero boundary conditions, we should establish the connection between the solution of KE equation and the Riemann-Hilbert problem.
Reconstruction formula
----------------------
To solve the above Riemann-Hilbert problem, it is necessary to regularize it by subtracting out the asymptotic and the pole contributions. For convenient, we define $\zeta_{n}=z_{n}$ and $\zeta_{N+n}=-\frac{q_{0}^{2}}{\overline{z}_{n}}$, $(n=1,\cdots,N)$ and $$\begin{aligned}
\begin{split}
&M^{-}-I+\frac{i}{z}\sigma_{3}Q_{-}-\mathop{\mathrm{\sum}}\limits_{n=1}^{2N}\frac{\mathop{\mathrm{Res}}\limits_{\overline{\zeta}_{n}}M^{-}}{z-\overline{\zeta}_{n}}
-\mathop{\mathrm{\sum}}\limits_{n=1}^{2N}\frac{\mathop{\mathrm{Res}}\limits_{\zeta_{n}}M^{+}}{z-\zeta_{n}}\\
=&M^{+}-I+\frac{i}{z}\sigma_{3}Q_{-}-\mathop{\mathrm{\sum}}\limits_{n=1}^{2N}\frac{\mathop{\mathrm{Res}}\limits_{\overline{\zeta}_{n}}M^{-}}{z-\overline{\zeta}_{n}}
-\mathop{\mathrm{\sum}}\limits_{n=1}^{2N}\frac{\mathop{\mathrm{Res}}\limits_{\zeta_{n}}M^{+}}{z-\zeta_{n}}-M^{+}G
\end{split}\label{Msub}\end{aligned}$$ The left-hand side of this equation is analytic in $D^{-}$ and is $O\left(\frac{1}{z}\right)$ as $z\rightarrow\infty$, and the sum of the first four terms of the right-hand side is analytic in $D^{+}$ and is $O\left(\frac{1}{z}\right)$ as $z\rightarrow\infty$. The asymptotic behavior of the off-diagonal scattering coefficients implies that $G(x,t,z)$ is $O\left(\frac{1}{z}\right)$ as $z\rightarrow\pm\infty$ and $O(z)$ as $z\rightarrow 0$ along the real axis.
Now we introduce the Cauchy projectors $P_{\pm}$ over $\Sigma$ $$P_{\pm}[f](z)=\frac{1}{2i\pi}\int_{\Sigma}\frac{f(\zeta)}{\zeta-(z\pm i0)}d\zeta,\nonumber$$ where $\int_{\Sigma}$ denotes the integral along the oriented contour shown in Fig.1, and the notation $z\pm i0$ indicates that when $z\in\Sigma$, the limit is taken from the left(right) of it. Now recall Plemelj’s formulae: if $f^{\pm}$ are analytic in $D^{\pm}$ and are $O\left(\frac{1}{z}\right)$ as $z\rightarrow\infty$, one has $P^{\pm}f^{\pm}=\pm f^{\pm}$ and $P^{+}f^{-}=P^{-}f^{+}=0$. Applying $P^{+}$ and $P^{-}$ to (\[Msub\]), then we have $$\begin{aligned}
\begin{split}
M(x,t,z)=&I-\frac{i}{z}\sigma_{3}Q_{-}+\mathop{\mathrm{\sum}}\limits_{n=1}^{2N}\frac{\mathop{\mathrm{Res}}\limits_{\overline{\zeta}_{n}}M^{-}}{z-\overline{\zeta}_{n}}
+\mathop{\mathrm{\sum}}\limits_{n=1}^{2N}\frac{\mathop{\mathrm{Res}}\limits_{\zeta_{n}}M^{+}}{z-\zeta_{n}}\\
&+\frac{1}{2i\pi}\int_{\Sigma}\frac{M^{+}(x,t,\zeta)G(x,t,\zeta)}{\zeta-z}d\zeta, \quad z\in\mathbb{C}\backslash \Sigma.
\end{split}\label{Msolu}\end{aligned}$$ From (\[Msolu\]), we deriving the asymptotic behaviors of $M^{\pm}(x,t,z)$ as $z\rightarrow\infty$ as $$\begin{aligned}
\begin{split}
M(x,t,z)=&I+\frac{i}{z}\Big\{-\sigma_{3}Q_{-}+\mathop{\mathrm{\sum}}\limits_{n=1}^{2N}\mathop{\mathrm{Res}}\limits_{\overline{\zeta}_{n}}M^{-}
+\mathop{\mathrm{\sum}}\limits_{n=1}^{2N}\mathop{\mathrm{Res}}\limits_{\zeta_{n}}M^{+}\\
&-\frac{1}{2i\pi}\int_{\Sigma}M^{+}(x,t,\zeta)G(x,t,\zeta)d\zeta\Big\}+O\big(\frac{1}{z^{2}}\big),
\end{split}\label{Mexp}\end{aligned}$$ Comparing the (1,2) element of (\[Mexp\]) with (\[muexp\]). Then we obtain the formula of the potential $q(x,t)$ $$q(x,t)=e^{2\int_{-\infty}^{x}i\beta(|q|^{2}-q_{0}^{2})dy}\left[q_{-}+i\Big(\mathop{\mathrm{\sum}}\limits_{n=1}^{2N}\mathop{\mathrm{Res}}\limits_{\zeta_{n}}M^{+}\Big)_{1,2}\\
-\frac{1}{2\pi}\int_{\Sigma}(M^{+}(x,t,\zeta)G(x,t,\zeta))_{1,2}d\zeta\right].\label{poten}$$
N-soliton solutions
-------------------
In this subsection we consider the reflectionless potentials. In this case, the reflection coefficient $\rho(z)=0$, the potential formula (\[poten\]) reduced as $$q(x,t)=e^{2\int_{-\infty}^{x}i\beta(|q|^{2}-q_{0}^{2})dy}\left[q_{-}+i\mathop{\mathrm{\sum}}\limits_{n=1}^{2N}A_{n}e^{-2i\theta(\zeta_{n})}\mu_{-,11}(\zeta_{n})\right].\label{ppp}$$ For convenient, we introduce the quantities $$\begin{aligned}
a_{j}(x,t,z)=\frac{\tilde{A}_{j}e^{2i\theta(x,t,\overline{\zeta}_{j})}}{z-\overline{\zeta}_{j}}, \quad j=1,\cdots,2N.\nonumber\end{aligned}$$ Then from (\[Msolu\]), we can derive the results
$$\begin{aligned}
&\mu_{-,12}(\overline{\zeta}_{j})=-\frac{iq_{-}}{\overline{\zeta}_{j}}+\mathop{\mathrm{\sum}}\limits_{k=1}^{2N}\frac{A_{k}e^{-2i\theta(\zeta_{k})}}{\overline{\zeta}_{j}-\zeta_{k}}\mu_{-,11}(\zeta_{k})
=-\frac{iq_{-}}{\overline{\zeta}_{j}}-\mathop{\mathrm{\sum}}\limits_{k=1}^{2N}\overline{a}_{k}(\zeta_{j})\mu_{-,11}(\zeta_{k})\\
&\mu_{-,11}(\zeta_{n})=1+\mathop{\mathrm{\sum}}\limits_{j=1}^{2N}\frac{\tilde{A}_{j}e^{2i\theta(\overline{\zeta}_{j})}}{\zeta_{n}-\overline{\zeta}_{j}}\mu_{-,12}(\overline{\zeta}_{j})
=1+\mathop{\mathrm{\sum}}\limits_{j=1}^{2N}a_{j}(\zeta_{n})\mu_{-,12}(\overline{\zeta}_{j}).\end{aligned}$$
Substituting the equation into the second one gives $$\mu_{-,11}(\zeta_{n})=1-iq_{-}\mathop{\mathrm{\sum}}\limits_{j=1}^{2N}\frac{a_{j}(\zeta_{n})}{\overline{\zeta}_{j}}
-\mathop{\mathrm{\sum}}\limits_{j=1}^{2N}\mathop{\mathrm{\sum}}\limits_{k=1}^{2N}a_{j}(\zeta_{n})\overline{a}_{k}(\zeta_{j})\mu_{-,11}(\zeta_{k}).$$
We write this system in matrix form. Let $$\mathbf{X}=(X_{1},\cdots,X_{2N})^{T}, \quad \mathbf{B}=(B_{1},\cdots,B_{2N})^{T},\nonumber$$ where $$X_{N}=\mu_{-,11}(\zeta_{n}), \quad B_{n}=1-iq_{-}\mathop{\mathrm{\sum}}\limits_{j=1}^{2N}\frac{a_{j}(\zeta_{n})}{\overline{\zeta}_{j}}, \quad n=1,\cdots,2N,\nonumber$$ and defining the $2N\times2N$ matrix $A=(A_{n,k})$, where $$A_{n,k}=\mathop{\mathrm{\sum}}\limits_{j=1}^{2N}a_{j}(\zeta_{n})\overline{a}_{k}(\zeta_{j}), \quad n,k=1,\cdots,2N,$$ the system can be rewritten as $M\mathbf{X}=\mathbf{B}$, where $M=I+A=(\mathbf{M}_{1},\cdots,\mathrm{M}_{2N})$. The system is simply $$X_{n}=\frac{\det{M_{n}^{ext}}}{\det{M}}, \quad n=1,\cdots,2N,\nonumber$$ where $$M_{n}^{ext}=(\mathbf{M}_{1},\cdots,\mathbf{M}_{n-1},\mathbf{B},\mathbf{M}_{n+1},\cdots,\mathbf{M}_{2N}).$$ Finally, upon substituting $X_{1},\cdots,X_{2N}$ into the reconstruction formula (\[ppp\]), the $N$-soliton solution of KE equation can be written compactly as $$q(x,t)=e^{2\int_{-\infty}^{x}i\beta(|r|^{2}-q_{0}^{2})dy}r(x,t),$$ where $$r(x,t)=q_{-}-i\frac{\det M^{aug}}{\det M},$$ and the augmented $(2N+1)\times(2N+1)$ matrix $M^{aug}$ is given by $$\begin{aligned}
M^{aug}=\left(
\begin{array}{cc}
0&\mathbf{Y}^{t}\\
\mathbf{B}&M
\end{array}\right), \quad \mathbf{Y}=(Y_{1},\cdots,Y_{2N})^{T},
\nonumber\end{aligned}$$ $$Y_{n}=A_{n}e^{-2i\theta(x,t,\zeta_{n})}, \quad n=1,\cdots,2N.$$
one-Soliton solutions
---------------------
In this subsection we mainly consider the one-soliton solution for which the reflection coefficient $\rho=0$. Take $q_{0}=1$, eigenvalue $z_{1}=i\chi$ ($\chi$=constant value and $\chi>1$) and $\tilde{A}_{1}=e^{\alpha-i\gamma}$ ($\alpha,\gamma\in\mathbb{R}$). From the $N$-soliton solutions formula, we can obtain the soliton solution $$q_{1}(x,t)=e^{2\int_{-\infty}^{x}i\beta(|r_{1}|^{2}-q_{0}^{2})dy}r_{1}(x,t),\label{qr}$$ where $r_{1}(x,t)$ with the following trigonometric function form $$r_{1}(x,t)=\frac{\cosh\nu+\frac{1}{2}a_{1}(1+\frac{a_{2}^{2}}{a_{1}^{2}})\sin(\gamma+a_{1}a_{2}t)-ia_{2}\cos(\gamma+a_{1}a_{2}t)}
{\cosh\nu+\frac{2}{a_{1}}\sin(\gamma+a_{1}a_{2}t)},\label{q1}$$ and $$\begin{aligned}
\begin{split}
&\nu=(x+4\beta t)a_{2}+\alpha+a_{0}, \\
&e^{a_{0}}=\frac{a_{1}}{2\chi a_{2}},\\
&e^{-a_{0}}=\frac{2\chi a_{2}}{a_{1}}.\nonumber
\end{split}\end{aligned}$$ For $\beta=0$, soliton solution (\[qr\]) convert into the one-soliton solution of focusing nonlinear Schr$\ddot{o}$dinger equation Fig.2 (a). Fig.2 (b) displays the $\beta\neq 0$ case and Fig.2 (c) displays the $\beta\neq 0$ and $\chi\rightarrow 1$ case. The Fig.2 (d) displays the more general situation that is the zero point $z_{1}=i\chi e^{i\alpha}$ where $\chi=2$ and $\alpha=\frac{\pi}{4}$.
[]{}\
\
[]{}\
\
Conclusion
==========
In this work, we investigated the focusing Kundu-Eckhaus equation with nonzero boundary condition at infinity. For $\beta=0$, the soliton solutions are reduced as the soliton solutions of focusing NLS with nonzero boundary conditions. We introduced a transformation, such that the asymptotic spectral problem with the linear relationship $\mathcal{V}_{\pm}=(2k+2\beta q_{0}^{2})\mathcal{U}_{\pm}$. A appropriate Riemann surface for the single-valued function of the spectral parameter was introduced. Then, the complex $k$-plane transformed into the complex $z$-plane. Unlike the focusing nonlinear Schr$\ddot{o}$dinger equation, the branch cut not on the $\mathrm{Im}k$ axis, but shift $\beta q_{0}^{2}$ along the $x$ axis. The orientation is up to $\beta$. The nonzero boundary conditions is different from zero boundary conditions mainly reflected in the analytic region and the zero points of $s_{11}$ and $s_{22}$. The analytic region not only involved with the upper-half/lower-half plane, but also involved with the inside/outside of the circle $C_{0}$.
Acknowledgments {#acknowledgments .unnumbered}
===============
This work is supported by the National Science Foundation of China (Grant No. 11671095, 51879045).
[99]{} Kodama Y, Optical solitons in a monomode fiber, J Stat Phys, **39**, 597-614 (1985).
Khaykovich L, Malomed BA, Deviation from one dimensionality in stationary properties and collisional dynamics of matter-wave solitons, Phys Rev E, **74**, 023607 (2006).
Kumar VR, Radha R, Wadati M, Phase engineering and solitons of Bose-Einstein condensates with two- and three-body interactions, J Phys Soc Japan, **79**, 074005 (2010).
Pethick CJ, Smith H, Bose-Einstein Condensation in Dilute Gases, Cambridge University Press, Cambridge, 2004.
Kundu A, Landau-Lifshitz and higher-order nonlinear systems gauge generated from nonlinear Schr$\ddot{o}$dinger-type equations, J Math Phys, **25**, 3433-3438 (1984).
Wang DS, Guo BL, Wang XL, Long-time asymptotics of the focusing Kundu-Eckhaus equation with nonzero boundary conditions, J Diff Equ, **266**, 5209-5253 (2019).
Zhu QZ, Xu J and Fan EG, The Riemann-Hilbert problem and long-time asymptotics for the Kundu-Eckhaus equation with decaying initial value, Appl Math Lett, **76**, 81-89 (2018).
Wang DS, Wang XL, Long-time asymptotics and the bright N-soliton solutions of the Kundu-Eckhaus equation via the Riemann-Hilbert approach, Nonlinear Analysis: Real World Applications, **41**, 334-361 (2018).
Guo BL, Liu N, Long-time asymptotics for the Kundu-Eckhaus equation on the half-line, J Math Phys, **59**, 061505 (2018). Wang X, Yang B, Chen Y, Yang YQ, Higher-order rogue wave solutions of the Kundu-Eckhaus equation, Phys Scr, **89**, 095210 (2014).
Bayindir C, Rogue waves of the Kundu-Eckhaus equation in a chaotic wave field, Phys Rev E **93**, 032201 (2016).
Qiu D, He JS, Zhang YS, Porsezian K, The Darboux transformation of the Kundu-Eckhaus equation, Proc R Soc Lond Ser A, **471** 20150236 (2015).
Guo BL, Ling LM, Liu QP, Nonlinear Schr$\ddot{o}$dinger equation: Generalized Darboux transformation and rogue wave solutions, Phys Rev E, **85**, 026607 (2012).
Levi D, Scimiterna C, The Kundu-Eckhaus equation and its discretizations, J Phys A, **42**, 465203-465210 (2009).
Gardner CS, Greene JM, Kruskal MD, Miura RM, Method for solving the Korteweg-deVries equation, Phys Rev Lett, **19**, 1095 (1967).
Yang JK, Nonlinear waves in intergrable and nonintergrable systems, Philadelphia: Soc Indus Appl Math, (2010).
Prinari B, Ablowitz MJ, Biondini G, Inverse scattering transform for the vector nonlinear Schr$\ddot{o}$dinger equation with nonvanishing boundary conditions, J Math Phys, **47**, 063508 (2006).
Demontis F, Prinari B, van der Mee C, Vitale F, The inverse scattering transform for the defocusing nonlinear Schr$\ddot{o}$dinger equations with nonzero boundary conditions, Stud Appl Math Phys, **131**, 1-40 (2013).
Biondini G, Kraus DK, Prinari B, The three-component defocusing nonlinear Schr$\ddot{o}$dinger equation with nonzero boundary conditions, Commun Math Phys, **348**, 475-533 (2016).
Kraus D, Biondini G, Kova$\check{c}$i$\check{c}$, The focusing Manakov system with nonzero boundary conditions, Nonlinearity, **28**, 3101-3151 (2015).
Prinari B, Vitale F, Biondini G, Dark-bright soliton solutions with nontrivial polarization interactions for the three-component defocusing nonlinear Schr$\ddot{o}$dinger equation with nonzero boundary conditions, J Math Phys, **56**, 071505 (2015).
Prinari B, Vitale F, Inverse scattering transform for the focusing nonlinear Schr$\ddot{o}$dinger equation with one-sided nonzero boundary condition, Cont Math, **651**, 157-194 (2015).
van der Mee C, Inverse scattering transform for the discrete focusing nonlinear Schr$\ddot{o}$dinger equation with nonvanishing boundary conditions, J Nonlinear Math Phys, **22**, 233-264 (2015).
Biondini G, Kraus D, Inverse scattering transform for the defocusing Manakov system with nonzero boundary conditions, SIAM J Math Anal, **47**, 706-757 (2015).
Biondini G, Fagerstrom E, Prinari B, Inverse scattering transform for the defocusing nonlinear Schr$\ddot{o}$dinger equation with fully asymmetric non-zero boundary conditions, Physica D, **333**, 117-136 (2016).
Biondini G, Kova$\check{c}$i$\check{c}$ G, Inverse scattering transformation for the focusing nonlinear Schr$\ddot{o}$dinger equation with nonzero boundary conditions, J math phys, **55**, 031506 (2014).
Pichler M, Biondini G, On the focusing non-linear Schr$\ddot{o}$dinger equation with non-zero boundary conditions and double poles, IMA J Appl Math, **82**, 131-151 (2017).
Li S, Biondini G, Soliton interactions and degenerate soliton complexes for the focusing nonlinear Schr$\ddot{o}$dinger equation with nonzero background, Eur Phys J Plus, **133**, 400 (2018).
[^1]: Corresponding author and email address: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Turbulent state of spectrally stable shear flows may be developed and sustained according to the bypass scenario of transition. If it works in non-magnetised boundless and homogeneous quasi-Keplerian flow, transiently growing shearing vortices should supply turbulence with energy. Employing the large shearing box approximation, as well as a set of global disc models, we study the optimal growth of the shearing vortices in such a flow in the whole range of azimuthal length-scales, $\lambda_y$, as compared to the flow scale-height, $H$. It is shown that with the account of the viscosity the highest possible amplification of shearing vortices, $G_{max}$, attains maximum at $\lambda_y\lesssim H$ and declines towards both the large scales $\lambda_y\gg H$ and the small scales $\lambda_y\ll H$ in a good agreement with analytical estimations based on balanced solutions. We pay main attention to the large-scale vortices $\lambda_y\gg H$, which produce $G_{max}\propto (\Omega/\kappa)^4$, where $\Omega$ and $\kappa$ denote local rotational and epicyclic frequencies, respectively. It is demonstrated that the large-scale vortices acquire high density perturbation as they approach the instant of swing. At the same time, their growth is not affected by bulk viscosity. We check that $G_{max}$ obtained globally is comparable to its local counterpart and the shape and localisation of global optimal vortices can be explained in terms of the local approach. The obtained results allow us to suggest that the critical Reynolds number of subcritical transition to turbulence in quasi-Keplerian flow, as well as the corresponding turbulent effective azimuthal stress should substantially depend on shear rate.'
author:
- 'D. N. Razdoburdin$^{1,2}$ and V. V. Zhuravlev$^{1}$[^1]'
date: |
$^{1}$Sternberg Astronomical Institute, Moscow M.V. Lomonosov State University, Universitetskij pr., 13, Moscow 119992, Russia\
$^{2}$Department of Physics, Moscow M.V. Lomonosov State University, Moscow, 119992, Russia
title: Transient growth of perturbations on scales beyond the accretion disc thickness
---
hydrodynamics — accretion, accretion discs — instabilities — turbulence — protoplanetary discs
Subcase of incompressible dynamics {#incompr}
==================================
Shearing vortex solution {#subsect1}
------------------------
Let us consider the incompressible limit of transient dynamics also studied by @yecko-2004 and in the case of a flow confined between two walls separated from one another by distance $2L$ across the shear. On the contrary, here we are going to deal with an unbounded shear. As $a_*\to \infty$, the continuity equation (\[sys3\]) yields the following restriction on the velocity components $$\frac{\partial u_x}{\partial x} = - \frac{\partial u_y}{\partial y},$$ and we are left with two projections of Euler equation (\[sys1\]-\[sys2\]).
We change to dimensionless comoving Cartesian coordinates in (\[sys1\]-\[sys3\]), $$\label{incompr_coords}
x^\prime = x/L,\, y^\prime = (y+q\Omega_0 xt)/L,\, t^\prime=\Omega_0 t,$$ where the quantity $L$ is an auxiliary radial scale, which does not have any physical meaning. The coordinates (\[incompr\_coords\]) lead us to a spatially homogeneous set of equations which have partial solutions in the form of SFH corresponding to the dimensionless wavenumbers $k_x$ and $k_y$ expressed in units of arbitrary $L^{-1}$. Similarly to §\[subsect2\], we omit the primes and introduce the shearing radial wavenumber $\tilde k_x \equiv k_x + q k_y t$ and the full wavenumber squared $k^2 \equiv \tilde k_x^2 +k_y^2$, arriving at the following dimensionless equations for a single SFH:
$$\label{SFH_sys_1}
\frac{d \hat u_x}{d t} = 2\hat u_y - {\rm i}\tilde k_x \hat W - R^{-1} k^2 \hat u_x,$$
$$\label{SFH_sys_2}
\frac{d \hat u_y}{d t} = -(2-q)\hat u_x - {\rm i}\tilde k_y \hat W - R^{-1} k^2 \hat u_y,$$
provided that $$\label{SFH_sys_3}
\tilde k_x \hat u_x = -k_y \hat u_y,$$ and it is assumed that $\hat u_x,\hat u_y$ are expressed in units of $U \equiv \Omega_0 L$. $\hat W$ is expressed in units of $U^2$. Correspondingly, $R \equiv U L / \nu$ is Reynolds number. Note that used Reynolds number $$\label{R_05}
R_{05} = q R.$$
Using (\[SFH\_sys\_1\]-\[SFH\_sys\_3\]), we obtain equation for $\hat u_x$ which reads $$\label{Eq_incompr}
\frac{d\hat u_x}{dt} = -2q \frac{k_y \tilde k_x}{k^2} \hat u_x -R^{-1} k^2 \hat u_x.$$
It can be checked that (\[Eq\_incompr\]) is reproduced from equation (16) of and has the following analytical solution: $$\label{sol_incompr}
\hat u_x (t) = \hat u_x(0) \frac{k_x^2+k_y^2}{k^2} \exp\left ( -\frac{\tilde k^3-k_x^3}{3qk_y R} - \frac{k_y^2 t}{R} \right ).$$
The solution (\[sol\_incompr\]) is used to obtain the transient growth factor of incompressible perturbations for particular values of $k_x, k_y$, $R$ and $q$, see §\[subsect3\] for further consideration.
Transient growth factor {#subsect3}
-----------------------
As usual, the size of incompressible perturbations in an unbounded shear flow is measured by a surface density of their kinetic energy, $E_k$, which leads to simple expression for a single SFH: $$\label{incompr_norm}
E_k = \frac{1}{2} ( |\hat u_x|^2 + |\hat u_y|^2) = \frac{1}{2} \frac{k^2}{k_y^2} |\hat u_x|^2.$$ Substituting equation (\[sol\_incompr\]) into (\[incompr\_norm\]) yields the following growth factor of SFH: $$\label{g_incompr}
g \equiv \frac{E_k(t)}{E_k(0)} = \frac{k_x^2+k_y^2}{k^2} \exp\left ( -2\frac{\tilde k^3-k_x^3}{3qk_y R} - 2\frac{k_y^2 t}{R} \right ).$$
Given the particular values of $k_y$, $R$ and $q$, (\[g\_incompr\]) has a global maximum over all $k_x$ and $t$ which we are going to denote here as $$\label{G_max_incompr}
G_{max} \equiv \max_{\forall k_x, \forall t}\, g,$$ since this quantity is analogous to the maximum optimal growth defined in §\[subsect4\], see equation (\[G\_max\_compr\]).
Analytical estimations\
of optimal growth {#prelim}
=======================
![Curves of $G_{max}$ in the subcase of incompressible dynamics of 2D perturbations for $R_{05}=2000$ and $q=3/2$: solid curve is obtained numerically using the definition (\[G\_max\_incompr\]), dashed line represents equation (\[anal\_G\_max\]), dot-dashed curve corresponds to equation (77) of taken with $k_{X,min}L=1.7$ in their notations. The dotted curve represents approximate $G_{max}$ given by equation (\[anal\_G\_max\_compr\]) for the case of finite disc thickness and $k_y$ normalised by $H$ rather than by $L$.[]{data-label="fig_1"}](fig_1){width="9cm"}
By means of the analytical supplement to their numerical optimisation results, demonstrate that in the case of incompressible dynamics $G_{max}(k_y)$ acquires a maximum owing to the presence of rigid boundaries (see also the results of @yecko-2004 in its Fig. 3b). In Fig. \[fig\_1\] we take the Keplerian shear, $q=3/2$, $R_{05}=2000$ and plot equation (77) of (dot-dashed line) together with our quantity (\[G\_max\_incompr\]) obtained numerically (solid line). It can also be checked that equation (77) of taken for a boundless shear, $$\label{anal_G_max}
G_{max} = \left ( \frac{qR}{k_y^2} \right )^{2/3} \, {\rm e}^{-2/3},$$ virtually coincides with our numerical curve in Fig. \[fig\_1\]. Equation (\[anal\_G\_max\]) can be easily obtained from (\[g\_incompr\]) setting $\tilde k_x=0$ (the swing instant of SFH) and taking the limit $|k_x| \gg k_y$. Thus, in general, $G_{max}\to \infty$ as $k_y\to 0$ in the absence of boundaries. Physically, this takes place due to a simple fact that for a tightly wound SFH with $|k_x|\gg k_y$ both viscous dissipation time-scale, $t_\nu$, and transient growth duration, $t_{tg}$, become longer as we proceed to small azimuthal wavelengths, whereas the (inviscid) growth factor increases monotonically with time, see equation (\[g\_incompr\]). As it is discussed, e.g., by , equation (\[anal\_G\_max\]) can be approximately derived suggesting that $G_{max}$ corresponds to equality $t_\nu=t_{tg}$, which gives the time when the maximum growth occurs as $$\label{t_max}
t_{max} \approx R^{1/3} (qk_y)^{-2/3}.$$ Note that estimation (\[t\_max\]) provides exactly the same value which one gets looking for a maximum of (\[g\_incompr\]) in order to derive an approximate $G_{max}$ given by equation (\[anal\_G\_max\]). Then, using the inviscid form of $g\approx (qt)^2$, we recover equation (\[anal\_G\_max\]) up to the damping factor $\exp(-2/3)$.
Of course, the growth factor (\[g\_incompr\]) and so $G_{max}$ do not depend on the scaling factor $L$. As we see from equation (\[anal\_G\_max\]), the transient growth value is controlled by the ratio $R/k_y^2 \sim (\lambda_y/l_p)^2$, where $\lambda_y$ is characteristic azimuthal length of perturbations and $l_p$ is mean free path of gas particles which determine microscopic viscosity. In fact, $\lambda_y^{-1} \sim k_y L^{-1}$ of SFH exhibiting $G_{max}$ cannot be too low since $|k_x|\sim (qk_y R)^{1/3}$ also decreases: but $|k_x| L^{-1}$ cannot be less than $H^{-1}$, since in the last case we start dealing with patch of disc comparable to its thickness in the direction of shear what forces us to take into account the effects of compressibility, see e.g. the discussion about SFH swing interval in §3.2.1 of . Thus, we of course cannot reach an arbitrary high $G_{max}$ at the expense of $\lambda_y$ value. It turns out, see §2.3 of , that changing to a model with the account of finite disc thickness, we can derive an approximate expression for $G_{max}$ in the opposite case of $\lambda_y\gg H$ (i.e. $ k_y\ll 1$) employing the balanced solutions to equations (\[SFH\_sys\_4\]-\[SFH\_sys\_6\]), see @heinemann-papaloizou-2009a. The inviscid result reads $g_{opt} \approx 4\kappa^{-4}(k_y q t)^2$, where $\kappa^2=2(2-q)$ is the dimensionless epicyclic frequency squared expressed in units of $\Omega_0^2$. Substituting $t_{max}$ given by equation (\[t\_max\]) into this result we obtain $$\label{anal_G_max_compr}
G_{max} \approx \frac{1}{(2-q)^2} (k_y q R)^{2/3} \, {\rm e}^{-2/3},$$ where damping factor $\exp(-2/3)$ has been added since we expect a similar action of viscous forces on shearing vortices having both $k_y\gg 1$ and $k_y \ll 1$, see a detailed justification of this assumption in §\[bulk\_visc\]. As we see, one gets an opposite dependence of the maximum optimal growth on $k_y$, i.e. $G_{max}$ decreases as we approach larger azimuthal scales. Thus, there is no problem of divergence of transient growth factor towards $k_y \to 0$, see Fig. \[fig\_1\]. At the same time, as it is seen in Fig. [\[fig\_1\]]{}, one expects much larger transient growth of columnar perturbations (i.e. perturbations independent of vertical direction) in the domain of $\lambda_y \gtrsim H$, than it might seem within the model of @yecko-2004 and .
Suppression of shearing density waves for tightly wound spirals {#bulk_app}
===============================================================
In the limit $|\tilde k_x|\gg 1$[^2] the inviscid shearing vortices are approximately given by balanced solutions obtained from wave equations for $\hat u_{x,y}$ and $\hat W$, see e.g. equations (27-29) of . Their explicit form is given by equations (35-37) of . The non-zero viscosity leads to the following wave equations for $\hat u_y$ and $\hat u_x$:
$$\begin{gathered}
\label{wave_visc_u_x}
\ddot{\hat u}_x = - k_y I_\nu - K \hat u_x - 2{\rm i}qk_y \hat W - \\ R^{-1} k^2 \left [ \dot{\hat u}_x + \hat u_y + 2q \frac{\tilde k_x k_y}{k^2} \hat u_x \right ] - \\
(R^{-1}/3 + R_b^{-1}) \left [ \frac{d}{dt}\left (\tilde k_x \, {\rm i} \dot{\hat W} \right ) + 2k_y \, {\rm i} \dot{\hat W} \right ],\end{gathered}$$
$$\begin{gathered}
\label{wave_visc_u_y}
\ddot{\hat u}_y = \tilde k_x I_\nu - K \hat u_y - \\ R^{-1} k^2 \left [ \dot{\hat u}_y - (2-q)\hat u_x + 2q \frac{\tilde k_x k_y}{k^2} \hat u_y \right ] - \\
(R^{-1}/3 + R_b^{-1}) [k_y \, {\rm i} \ddot{\hat W} - (2-q)\tilde k_x \, {\rm i} \dot{\hat W}],\end{gathered}$$
where $K\equiv k^2 + \kappa^2$ and $I_\nu\equiv \tilde k_x \hat u_y - k_y \hat u_x + {\rm i}(2-q)\hat W$ is SFH of the potential vorticity perturbation, see corresponding expressions e.g. of @bodo-2005 or @tevzadze-2003 for the case of baratropic perturbations. $I_\nu$ does not depend on time in the inviscid limit, however, in presence of non-zero viscosity this is not true anymore. To emphasise this fact, we add a subscript ’$\nu$’ to the notation of the potential vorticity perturbation. It is not difficult to construct equation for $I_\nu$ using the set of equations (\[SFH\_sys\_4\]-\[SFH\_sys\_6\]): $$\label{I_eq}
\dot I_\nu = - R^{-1} k^2 I_\nu,$$ where the second term in the right-hand side of equation (\[I\_eq\]), explicitly ${\rm i} (2-q) R^{-1} k^2 \hat W$, has been omitted for the following reason. In the problem we consider, the viscosity is small, $R\gg 1$, $R_b \gg 1$. Thus, in the leading orders in $R^{-1}$ and $R_b^{-1}$ one can use an inviscid balanced solution for $\hat W$ (see equation (37) of ) which gives that $|\hat W/I_\nu|\ll 1$ provided that $|\tilde k_x| \gg 1$. This means that the omitted term is much less than the main term in the right-hand side of equation (\[I\_eq\]). Similarly, in the leading orders in $R^{-1}$ and $R_b^{-1}$, it is sufficient to take the slow varying solution of inhomogeneous equation (\[wave\_visc\_u\_y\]) in its inviscid form, $$\label{u_y_sol}
\hat u_y = \frac{\tilde k_x}{K} I_\nu,$$ but with $I_\nu$ decaying as described by equation (\[I\_eq\]). This is the decay of $I_\nu$ which yields an additional factor $\exp (-2/3)$ in equation (\[anal\_G\_max\_compr\]) being evaluated at the $t_s=-k_x/(qk_y)$ with $t_s$ additionally equated to (\[t\_max\]). Correspondingly, $\hat u_x$ is given by $$\label{u_x_sol}
\hat u_x = -\frac{K+4q}{K^2+4q^2k_y^2} k_y I_\nu,$$ see equation (35) by . We see that as far as $|\tilde k_x|$ is large, the radial velocity perturbation is much less than the azimuthal one in the case of tightly wound vortices, i.e. when $|\tilde k_x| \gg k_y$, and the velocity perturbation divergence is small: $$\label{vort_property}
{\rm i}\dot{\hat W} \sim \frac{k_y}{\tilde k_x^2} \hat u_y.$$ The tightly wound vortical spirals are of primary interest here, since we are dealing with high $R$ and focus on $k_y \lesssim 1$ which leads to $t_{max}\gg 1$, see equation (\[t\_max\]). Further, since $t_{max}$ approximately equals to $t_s$ of an optimal SFH exhibiting $G_{max}$, at the initial moment $|k_x| \gg k_y$ and the same is about $|\tilde k_x|$ until $t>0$ is not too close to the instant of swing.
Equation (\[vort\_property\]) implies that terms in the second square brackets in the right-hand sides of equations (\[wave\_visc\_u\_x\]) and (\[wave\_visc\_u\_y\]) are smaller at least by a factor $|\tilde k_x|^{-3}$ than the first and the second terms therein. Taking into account that terms proportional to $R_b^{-1}$ are absent in equation (\[I\_eq\]), we come to the conclusion that bulk viscosity cannot affect the dynamics of vortical SFH even if $R_b\sim 1$.
The opposite situation takes place with density waves. These are rapidly oscillating solutions to the homogeneous equations (27-29) of . That is why, as long as $|\tilde k_x|\gg 1$, $\dot{\hat u}_x \sim |\tilde k_x| \hat u_x$ and so is about $\dot{\hat u}_y \sim |\tilde k_x| \hat u_y$ in the inviscid case. We see that in this situation at least the term $R_b^{-1} \tilde k_x^2 \dot{\hat u}_x$ becomes the leading one among all viscous terms in equation (\[wave\_visc\_u\_x\]) and it exceeds the inviscid terms as soon as $R_b \lesssim |\tilde k_x|^2$.
Supplementary material to the global problem
============================================
Viscous terms {#visc_app}
-------------
Viscous terms entering the dynamical equations (\[direct1\]-\[direct3\]) explicitly are
$$\label{direct4}
N_r=\frac{\nu}{r}\frac{\partial}{\partial r}\left(r\frac{\partial \delta v_r}{\partial r}\right)-\frac{\nu\left(m^2-1\right)}{r^2}\delta v_r+\frac{2}{\Sigma r}\frac{\partial\delta v_r}{\partial r}\frac{d}{dr}\left(\nu \Sigma r\right),$$
$$\label{direct5}
\begin{aligned}
&N_{\varphi}=-\frac{\nu r\Omega^{\prime}}{\Sigma a_*^2}r^{-\frac{2n}{n+1}}\frac{d}{dr}\left(\Sigma r^{\frac{2n}{n+1}}\right)\delta h+\\
&+\frac{1}{\Sigma r^2}\frac{\partial}{\partial r}\left(\nu \Sigma r^3 \frac{\partial}{\partial r}\frac{\delta v_{\varphi}}{r}\right)+\frac{{\rm i}m}{\Sigma r^3}\frac{d}{d r}\left(\nu \Sigma r^2\right)\delta v_r-\frac{\nu m^2}{r^2}\delta v_{\varphi},
\end{aligned}$$
$$\label{direct7}
B_r=\frac{1}{\Sigma}\frac{\partial }{\partial r}\left(\Sigma\left(\nu_b+\frac{\nu}{3}\right){\rm div}~\delta \mathbf{v}\right)-\frac{1}{\Sigma r^2}\frac{d}{d r}\left(\nu \Sigma r^2\right) {\rm div}~\delta \mathbf{v},$$
$$\label{direct8}
B_{\varphi}=\frac{{\rm i}m}{r}\left(\nu_b+\frac{\nu}{3}\right){\rm div}~\delta \mathbf{v}.$$
The norm (\[glob\_norm\]) yields the following viscous terms in the adjoint equations (\[adjoint1\]-\[adjoint3\]):
$$\label{adjoint4}
\tilde{N_r}=
-\frac{\nu}{r}\frac{\partial}{\partial r}\left(r\frac{\partial\delta\tilde{v}_r}{\partial r}\right)
+\frac{\nu\left(m^2-1\right)}{r^2}\delta \tilde{v}_r
-\frac{2}{\Sigma r}\frac{\partial \delta \tilde{v}_r}{\partial r}\frac{d}{dr}\left(\nu \Sigma r\right)$$
$$\label{adjoint5}
\tilde{N}_{\varphi}=-\frac{1}{\Sigma r^2}\frac{\partial}{\partial r}\left(\nu \Sigma r^3 \frac{\partial}{\partial r}\frac{\delta \tilde{v}_{\varphi}}{r}\right)-\frac{{\rm i}m}{\Sigma r^3}\frac{d}{dr}\left(\nu \Sigma r^2\right)\delta \tilde{v}_r+\frac{\nu m^2}{r^2}\delta \tilde{v}_{\varphi}$$
$$\label{adjoint6}
\tilde{N}_h=\frac{\nu r \Omega^{\prime}}{\Sigma}r^{-\frac{2n}{n+1}}\frac{d}{dr}\left(\Sigma r^{\frac{2n}{n+1}}\right) \delta \tilde{v}_{\varphi}$$
$$\label{adjoint7}
\tilde{B}_r=-\frac{1}{\Sigma}\frac{\partial}{\partial r}\left(\Sigma \left(\nu_b+\frac{\nu}{3}\right) {\rm div}~\delta\tilde{\mathbf{v}}\right)
+\frac{1}{\Sigma r^2}\frac{d}{dr}\left(\nu \Sigma r^2\right) {\rm div}~\delta\tilde{\mathbf{v}}$$
$$\label{adjoint8}
\tilde{B}_{\varphi}=-\frac{{\rm i}m}{r}\left(\nu_b+\frac{\nu}{3}\right) {\rm div}~\delta\tilde{\mathbf{v}}$$
Note that boundary terms that emerge in the course of derivation of equations (\[adjoint1\]-\[adjoint3\]) are used to formulate boundary conditions for the adjoint variables, see the Appendix \[boundary\_app\]. Also, similar to what we have in the local problem, the terms in equations (\[adjoint4\]), (\[adjoint5\]), (\[adjoint7\]) and (\[adjoint8\]) acquire opposite signs comparing to corresponding terms in equations (\[direct4\]-\[direct8\]). Additionally, the term reciprocal to term $\propto \delta h$ in equation (\[direct5\]) appears in the adjoint equation for $\delta \tilde h$, see equation (\[adjoint3\]). Note that the last term vanishes as one turns to local perturbations.
Boundary conditions {#boundary_app}
-------------------
Let us impose the boundary conditions for perturbations necessary to advance equations (\[direct1\]) – (\[direct3\]) forward in time and equations (\[adjoint1\]) – (\[adjoint3\]) backward in time. As the disc is considered to be radially infinite, only a condition at the inner boundary $r=r_i$ is relevant for the dynamics at a finite timespan.
### Models N1 and P1 {#N1_P1}
At first, let us consider the set of equations for projections of the displacement vector $\mathbf{\xi}$. It can be done using a general relationship between $\mathbf{\xi}$ and the velocity Lagrangian perturbation : $$\frac{\partial \xi_r}{\partial t}+{\rm i}m\Omega\xi_r=\delta v_r+\frac{\partial}{\partial r}\left(v_r \xi_r\right)$$
$$\frac{\partial \xi_{\varphi}}{\partial t}+{\rm i}m\Omega\xi_{\varphi}=\delta v_{\varphi}+r\Omega^{\prime}\xi_r+r v_r \frac{\partial}{\partial r}\left(\frac{\xi_{\varphi}}{r}\right)$$
Since $v_r$ tends to infinity at $r_i$ (see equation (\[v\_r2\])), $\xi_r$ and $\xi_{\varphi}$ must vanish at $r_i$. This implies that the boundary remains unperturbed, which simplifies the boundary conditions. Now, as $v_r \propto D^{-3}$ in the vicinity of the boundary, the components of the displacement vector must behave at least as $$\label{xi_r_prop}
\xi_r \propto D^3,$$ $$\label{xi_phi_prop}
\xi_{\varphi} \propto D^4$$ close to $r=r_i$.
Boundary conditions to the set (\[direct1\]) – (\[direct3\]) must be that the Lagrangian perturbation of surface force acting on the inner boundary of the flow is equal to zero. A relation between this force and the full stress tensor, $\sigma_{ik}$, can be found, e.g., in @landau-lifshitz-1987 [§15]. Since $\xi_r=0$, the Lagrangian perturbation of the normal to the boundary is equal to zero. Thus, the boundary condition turns into two independent conditions on the Lagrangian perturbation of $\sigma_{ik}$: $$\label{sigm_rr}
\Delta \sigma_{rr}=0,$$ $$\label{sigm_rph}
\Delta \sigma_{r\varphi}=0.$$
Integration of equation (\[sigm\_rph\]) over $z$ yields $$\begin{aligned}
\label{boundary1}
&\Sigma \nu \left[r\frac{\partial}{\partial r}\left(\frac{\delta v_{\varphi}}{r}\right)+\frac{{\rm i}m}{r}\delta v_r\right]+
\xi_r \frac{d}{dr}\left(\Sigma\nu r \Omega^{\prime}\right)-\\
&-2\xi_r\frac{dH}{dr}\nu r \Omega^{\prime} \rho \big{|}_{z=H}=0
\end{aligned}$$ In equation (\[boundary1\]) all variables are taken at the inner boundary of disc. However, because of the relation (\[v\_r3\_integrated\]) and limitations on the displacement vector (\[xi\_r\_prop\]), (\[xi\_phi\_prop\]) equation (\[boundary1\]) is satisfied for arbitrary $\delta v_r$ and $\delta v_{\varphi}$. Thus, equation (\[boundary1\]) does not provide us with any restriction on perturbations at $r_i$.
At the same time, equation (\[sigm\_rr\]) results in the following equality: $$\begin{aligned}
\label{boundary2}
&2\rho \nu \frac{\partial \delta v_r}{\partial r}+\rho\left(\nu_{b}-\frac{2}{3}\nu\right)\left(\frac{1}{r}\frac{\partial}{\partial r}\left(r \delta v_r\right)+\frac{{\rm i}m}{r}\delta v_{\varphi}\right)-\\
&-2\rho \nu \Omega^{\prime} \xi_{\varphi}-\delta p -\frac{\partial p}{\partial r}\xi_r=0
\end{aligned}$$ In equation (\[boundary2\]) all variables are taken at the inner boundary of disc. Equation (\[boundary2\]) allows us to formulate the necessary restrictions on $\delta v_r$ and $\delta v_{\varphi}$ at $r_i$, provided that it is supplied by the regularity condition for $\delta h=\delta p / \rho$ at the boundary. Indeed, equation (\[boundary2\]) yields $$\begin{aligned}
\label{boundary3}
&\delta h=
\frac{\delta p}{\rho}=
\left(\nu_b+\frac{4}{3}\nu\right)\left(\frac{\nu_b-\frac{2}{3}\nu}{\nu_b+\frac{4}{3}\nu}\frac{\delta v_r}{r}+\frac{\partial \delta v_r}{\partial r}\right)+\\
&+\left(\nu_b-\frac{2}{3}\nu\right)\frac{{\rm i}m}{r}\delta v_{\varphi}-
2\nu \Omega^{\prime} \xi_{\varphi}-
\frac{\partial p}{\partial r}\frac{\xi_r}{\rho}.
\end{aligned}$$ The last two terms vanish due to limitations (\[xi\_r\_prop\]), (\[xi\_phi\_prop\]). Since $\nu$ diverges as $r\to r_i$, $\delta h$ is regular at $r_i$ as soon as $$\label{boundary4}
-\frac{1}{2}\frac{\delta v_r}{r}+\frac{\partial \delta v_r}{\partial r}=0,$$ $$\label{boundary5}
\delta v_{\varphi}=0.$$
As it is mentioned in §\[adj\_eqs\], in order to get the boundary conditions for the adjoint variables, it is necessary to put the boundary terms emerging in derivation of ${\bf A}^\dag$ equal to zero for each adjoint equation independently. This results in the following set of the adjoint boundary conditions: $$\label{boundary_adj1}
\delta\tilde{v}_r=0$$
$$\label{boundary_adj2}
\delta \tilde{v}_{\varphi}=0$$
$$\label{boundary_adj3}
\delta \tilde{h}=0$$
### Models N2 and P2 {#N2_P2}
In this case we impose the no-slip boundary conditions at $r=r_i$: $$\label{boundary_simple1}
\delta v_r=0,$$
$$\label{boundary_simple2}
\delta v_{\varphi}=0.$$
For a disc of uniform surface density the substitutions in derivation of the adjoint equations vanish under the conditions $$\label{boundary_adj_simple1}
\delta\tilde{v}_r=0,$$
$$\label{boundary_adj_simple2}
\delta \tilde{v}_{\varphi}=0$$
at $r=r_i$. Note that for models N2 and P2 the boundary condition onto $\delta \tilde{h}$ is not required, since the advective terms are absent in this case, see explanation in §\[num\_meth\].
[^1]: E-mail: [email protected]
[^2]: or, equivalently, outside of the swing interval of SFH, see .
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- 'S. Bethke'
- 'S. Kluth'
- 'C. Pahl'
- 'J. Schieck'
- 'the JADE Collaboration[^1]'
date: 'Received: date / Revised version: date'
title: 'Determination of the Strong Coupling from hadronic Event Shapes with ${\cal O}({\ensuremath{\alpha_{\mathrm{S}}}}^3)$ and resummed QCD predictions using JADE Data'
---
Introduction
============
Analyses of events originating from [$\mathrm{e^+e^-}$]{} annihilation into hadrons allow studies [@biebel01a; @dasgupta03; @dissertori03; @kluth06] of Quantum Chromodynamics (QCD), the theory of the strong interaction [@fritzsch73; @gross73a; @gross73b; @politzer73]. Comparison of observables like jet production rates or event shapes with theoretical predictions provides access to the determination of the strong coupling [$\alpha_{\mathrm{S}}$]{}. Recently significant progress in the theoretical calculations of event shape observables has been made and next-to-next-to-leading order (NNLO) predictions are now available [@gehrmannderidder07b] as well as matching with resummed calculations in the next-to-leading-log-approximation (NLLA) [@gehrmann08]. As a first application, measurements of [$\alpha_{\mathrm{S}}$]{} at centre-of-mass-system (cms) energies between ${\ensuremath{\sqrt{s}}}=91$ GeV and ${\ensuremath{\sqrt{s}}}=206$ GeV were presented [@dissertori07]. The same theoretical predictions are used in this paper to determine the strong coupling [$\alpha_{\mathrm{S}}$]{} from JADE[^2] data recorded at lower cms energies. As in [@dissertori07] and the previous standard LEP and JADE analyses [@kluth06] we use Monte Carlo simulations to treat hadronisation effects. In [@davison08] data for thrust at cms energies $14 \le {\ensuremath{\sqrt{s}}}\le 207$ GeV are analysed with combined NNLO+NLLA calculations and an analytic model for non-perturbative physics.
The JADE experiment operated at the PETRA[^3] [$\mathrm{e^+e^-}$]{} collider at DESY[^4], Hamburg, Germany. The data taken in the years from 1979 to 1986 cover cms energies between 12 and 46.4 GeV.
The JADE Detector
=================
A detailed description of the JADE detector can be found in [@naroska87]. For this analysis, tracks from charged particles and energy deposits in the electromagnetic calorimeter are used. The main tracking device, a large volume jet chamber was located in a 0.48 T solenoidal magnetic field. The electromagnetic calorimeter consisted of 2520 lead glass blocks in the barrel and 192 lead glass blocks in both endcaps with radiation length varying between 9.6 in the endcaps and up to 15.7 in the barrel.
Data and Monte Carlo Samples
============================
The Data and Monte Carlos samples utilised in this analysis are identical to those used in a previous determination of [$\alpha_{\mathrm{S}}$]{} [@jader4]. The data correspond to a total integrated luminosity of 195/pb taken at average cms energies between 14.0 GeV and 43.8 GeV. The breakdown of the data samples, including the average cms energies, the energy ranges, data taking periods, integrated luminosities and the overall numbers of selected hadronic events are summarised in table \[TableNumberEvents\].
[ccrcr]{} [$\sqrt{s}$]{}& energy & year & $L$ & selected\
GeV & range in GeV & & (1/pb) & events\
14.0 & 13.0–15.0 & 1981 & 1.46 & 1783\
22.0 & 21.0–23.0 & 1981 & 2.41 & 1403\
34.6 & 33.8–36.0 & 1981–1982 & 61.7 & 14313\
35.0 & 34.0–36.0 & 1986 & 92.3 & 20876\
38.3 & 37.3–39.3 & 1985 & 8.28 & 1585\
43.8 & 43.4–46.4 & 1984–1985 & 28.8 & 4374\
Monte Carlo events are generated in large numbers to correct the data for experimental acceptance, resolution effects and background. Events are simulated using PYTHIA 5.7 [@jetset3], and for systematic studies with HERWIG 5.9 [@herwig]. Subsequently the generated events are processed through a full simulation of the JADE detector and are reconstructed in the same way and with the same program chain as the data. For comparison with the corrected data and for the correction of hadronisation effects large samples of Monte Carlo events have been produced using PYTHIA 6.158, HERWIG 6.2 and ARIADNE 4.11 [@ariadne3]. We use the model parameters as determined at [$\sqrt{s}$]{}=91 GeV by the OPAL experiment at the LEP [$\mathrm{e^+e^-}$]{}collider [@OPALPR141; @OPALPR379].
Experimental Procedure
======================
Event Selection
---------------
The selection of identified and well measured hadronic event candidates follows the procedure outlined in [@jader4]. Events with a large momentum imbalance due to photons emitted in the initial state are rejected. The event selection is based on minimal requirements for charged particle multiplicity, visible energy and longitudinal momentum imbalance. The dominating backgrounds from hadronic $\tau$ decays and two-photon interactions with hadronic final states are supressed to negligble levels.
Event Shape Distributions
-------------------------
The properties of hadronic events can be described by event shape observables. Event shape observables used for this analysis are thrust ([$1-T$]{}) [@thrust1; @thrust2], heavy jet mass ([$M_{\mathrm{H}}$]{}) [@jetmass], wide and total jet broadening ([$B_{\mathrm{W}}$]{} and [$B_{\mathrm{T}}$]{}) [@nllabtbw], C-Parameter ([$C$]{}) [@parisi78; @donoghue79; @ert] and the transition value between 2 and 3 jet configurations [@OPALPR003; @komamiya90] defined by the Durham jet algorithm ([$y_{23}^\mathrm{D}$]{}) [@durham]. Whenever we refer to a generic event shape observable [$1-T$]{}, [$M_{\mathrm{H}}$]{}, [$B_{\mathrm{T}}$]{}, [$B_{\mathrm{W}}$]{}, [$C$]{} or [$y_{23}^\mathrm{D}$]{} we use the symbol $y$.
The event shape observables are calculated from selected charged particle tracks and calorimeter clusters after correcting for double counting of energy as described in [@jader4]. We compared the data with the predictions from Monte Carlo simulations as described above and found good agreement at all energy points. Similar observations were made in [@jader4] for related observables.
Corrections to the Data {#detcorr}
-----------------------
The corrections to the data for limited experimental resolution, acceptance and [$\mathrm{b\overline{b}}$]{} background follows exactly the treatment described in [@jader4]. Selected charged particle tracks as well as electromagnetic clusters are used to calculate the event shape observables.
From simulated events two different distributions are built: the [*detector-level*]{} distribution and the [*hadron-level distribution*]{}. The detector-level distributions are calculated exactly in the same way as for data using measured charged particle tracks and calorimeter clusters. The hadron-level distributions use the true four-momenta of the stable particles[^5] in events where the centre-of-mass energy is reduced due to initial state radiation (ISR) by less than 0.15 GeV. The bin-by-bin corrections for the data distributions are derived from the ratio of hadron-level to detector-level distributions for simulated events with u, d, s or c primary quarks. Contributions from B hadron decays bias the measurement of event shape observables and therefore the expected contribution from ${\ensuremath{\mathrm{e^+e^-}}}\rightarrow{\ensuremath{\mathrm{b\overline{b}}}}$ events is subtracted from the detector-level distributions before the corrections are applied. The simulations were optimised by OPAL to describe production and decays of B hadrons [@OPALPR141; @OPALPR379]. The good description of our uncorrected data by the simulations confirms that using the simulations to subtract the ${\ensuremath{\mathrm{e^+e^-}}}\rightarrow{\ensuremath{\mathrm{b\overline{b}}}}$ background is justified.
In order to study systematic uncertainties the selection and correction procedures are modified and the whole analysis is repeated. The evaluation of the systematic uncertainties follows identically the procedure described in [@jader4].
Determinations of
==================
QCD Calculations {#qcdprediction}
----------------
The distributions of the event shape observables are predicted by [$\mathcal{O}({\ensuremath{\alpha_{\mathrm{S}}}}^3)$]{} (NNLO) perturbative QCD calculations [@gehrmannderidder07b]: $$\frac{1}{\sigma}\frac{d\sigma}{dy}_{{\ensuremath{\mathcal{O}({\ensuremath{\alpha_{\mathrm{S}}}}^3)}}}
= \frac{dA}{dy}{\ensuremath{\hat{\alpha}_{\mathrm{S}}}}+ \frac{dB}{dy}{\ensuremath{\hat{\alpha}_{\mathrm{S}}}}^2 + \frac{dC}{dy}{\ensuremath{\hat{\alpha}_{\mathrm{S}}}}^3
\label{NNLOcalc}$$ with ${\ensuremath{\hat{\alpha}_{\mathrm{S}}}}={\ensuremath{\alpha_{\mathrm{S}}}}(\mu)/(2\pi)$. Equation (\[NNLOcalc\]) is shown for renormalisation scale $\mu=Q$, where $Q$ is the physical scale usually identified with the cms energy [$\sqrt{s}$]{} for hadron production in [$\mathrm{e^+e^-}$]{}annihilation. The coefficient distributions for leading order (LO) $dA/dy$, next-to-leading order (NLO) $dB/dy$ and NNLO $dC/dy$ were kindly provided by the authors of [@gehrmannderidder07b]. In [@weinzierl08] a problem at small values of $y$ with the NNLO terms calculated in [@gehrmannderidder07b] was shown, but it does not affect the kinematic regions selected in our fits. The normalisation to the total hadronic cross section and the terms generated by variation of the renormalisation scale parameter ${\ensuremath{x_{\mu}}}=\mu/Q$ are implemented according to [@gehrmannderidder07b]. The prediction in equation (\[NNLOcalc\]) may be combined with resummed NLLA calculations [@gehrmann08] using the $\ln
R$-matching scheme; we refer to these predictions as NNLO+NLLA. The $\ln R$-matching procedure ensures that in the combination of the fixed order NNLO calculations and the resummed NLLA calculations no double counting of common terms occurs. The NNLO+NLLA predictions are here compared with experimental data for the first time.
The theoretical predictions provide distributions at the level of quarks and gluons, the so-called [*parton-level*]{}. The distributions calculated using the final state partons after termination of the parton showering in the models are also said to be at the [*parton-level*]{}. To compare the QCD predictions with measured hadron-level event shape distributions the predictions are corrected for hadronisation effects. These corrections are obtained by calculating in the Monte Carlo models the ratio of the cumulative distributions at hadron-level and parton-level. The corrections are applied to the cumulative prediction $R(y)=\int_0^y 1/\sigma
d\sigma/dy'dy'$ as in [@OPALPR158].
It was shown in [@dissertori07] that the event shape observable distributions derived from the parton-level of the Monte Carlo generators are described reasonably well by the NNLO calculation in their fit ranges. We compared the parton-level predictions of the Monte Carlo generators with the QCD predictions in NNLO or NNLO+NLLA with ${\ensuremath{{\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})}}=0.118$ and ${\ensuremath{x_{\mu}}}=0.5, 1.0, 2.0$ at ${\ensuremath{\sqrt{s}}}=14$, 22, 35 and 44 GeV. We study the quantity $r(y)_{theory,MC}=d\sigma/dy_{theory} /
d\sigma/dy_{MC}-1$. In addition we compute the corresponding quantities $r(y)_{MC_i,MC_j}$ for any pair $i,j$ of Monte Carlo predictions and $r(y)_{{\ensuremath{x_{\mu}}}=0.5;1;2,{\ensuremath{x_{\mu}}}=0.5;1;2}$ for theory predictions with different renormalisation scale values. The largest values of the $\mathrm{abs}(r(y)_{MC_i,MC_j})$ and the $\mathrm{abs}(r(y)_{{\ensuremath{x_{\mu}}}=0.5;1;2,{\ensuremath{x_{\mu}}}=0.5;1;2})$ at each $y$ are added in quadrature to define the uncertainty $\Delta r(y)$ of $r(y)_{theory,MC}$. The average values $\bar{r}$ of the $\mathrm{abs}(r(y))$ over the fit ranges (see below) are taken as a measure of the consistency between between theory and Monte Carlo predictions. The ratios of $\bar{r}_{theory,MC}$ with the average error $\Delta r(y)$ are generally smaller than or about equal to unity and reach values of about two for [$C$]{} at $\sqrt{s}= 44$ GeV. The model dependence of the hadronisation correction and the renormalisation scale dependence of the theory will be studied as systematic uncertainties below. Our studies show that systematic uncertainties introduced by discrepancies between the theory predictions and the Monte Carlo parton-level predictions will be covered by the combined hadronisation and theory systematic variations.
Measurements of
----------------
The strong coupling [$\alpha_{\mathrm{S}}$]{} is determined by a [$\chi^2$]{} fit to each of the measured event shape distributions at the hadron-level, i.e.corrected for experimental effects. A [$\chi^2$]{} value is calculated at each cms energy: $${\ensuremath{\chi^2}}= \sum_{i,j}^{n} (d_i-t_i({\ensuremath{\alpha_{\mathrm{S}}}})) (V^{-1})_{ij} (d_j-t_j({\ensuremath{\alpha_{\mathrm{S}}}}))$$ where $i,j$ count the bins within the fit range of the event shape distribution, $d_i$ is the measured value in the $i$th bin, $t_i({\ensuremath{\alpha_{\mathrm{S}}}})$ is the QCD prediction for the $i$th bin corrected for hadronisation effects, and $V_{ij}$ is the covariance matrix of the $d_i$. The final prediction is obtained by integrating the QCD predictions in equation (\[NNLOcalc\]) over the bin width after application of the hadronisation correction as explained above in section \[qcdprediction\]. The [$\chi^2$]{} value is minimised with respect to [$\alpha_{\mathrm{S}}$]{} while the renormalisation scale factor is set to ${\ensuremath{x_{\mu}}}=1$.
The evolution of the strong coupling [$\alpha_{\mathrm{S}}$]{} as a function of the renormalisation scale is implemented in three loops as shown in [@kluth06]. Since the evolution of [$\alpha_{\mathrm{S}}$]{} in the cms energy range considered here does not involve the crossing of flavour thresholds it does not introduce significant uncertainties. In order to quantify the uncertainty from the evolution procedure we evolve ${\ensuremath{{\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})}}=0.118$ from [$m_{{\ensuremath{\mathrm{Z^0}}}}$]{} to 14 GeV in three loops and two loops and find a relative difference of 0.1%.
In order to take the correlations between different bins into account the covariance matrix $V_{ij}$ is computed following the approach described in [@OPALPR404]: $$\begin{aligned}
V_{ij} & = & \sum_k \frac{\partial P_i}{\partial N_k}
\frac{\partial P_j}{\partial N_k} N_k \\ \nonumber
& = & \frac{1}{N^4} \sum_k \alpha_k^2 N_k
\left(N\delta_{ik} - \widetilde{N}_i\right)
\left(N\delta_{jk} - \widetilde{N}_j\right) \;.\end{aligned}$$ $N_i$ is the number of data events, $\widetilde{N}_i=
\alpha_i(N_i-b_i)$ the number of events after subtraction of background $b_i$ from [$\mathrm{b\overline{b}}$]{} events and multiplying by a correction $\alpha_i$ for detector effects, $P_i$ is the normalised hadron-level distribution at bin $i$ and $N=\sum_k \widetilde{N}_k$.
The fit ranges are determined by several considerations. We require the leading log terms to be less than unity, because we also use a fixed order expansion without resummation of log-enhanced terms, see e.g. [@ellis96], and for consistency we use the same fit ranges in the NNLO and NNLO+NLLA analyses. The leading log term of $dA/dy$ is $\ln y/y$ and we require ${\ensuremath{\hat{\alpha}_{\mathrm{S}}}}\ln y/y<1$ for ${\ensuremath{{\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})}}=0.118$, $14\le{\ensuremath{\sqrt{s}}}\le200$ GeV and $y>0.1$. The upper limit is given by the requirement that all three orders of the NNLO calculations contribute, i.e. the fit range extends to the kinematic limit of the LO coefficients $dA/dy$. The fit range for [$M_{\mathrm{H}}$]{} should be compared with the related fit range for [$1-T$]{} after squaring, because ${\ensuremath{M_{\mathrm{H}}}}^2\sim{\ensuremath{1-T}}$ in LO. The fit ranges for [$C$]{} and [$1-T$]{} are related by a factor of $(\ln6)/6\simeq0.3$. The resulting fit ranges are shown in table \[fitranges\].
[ ccc ]{} [$1-T$]{}& [$M_{\mathrm{H}}$]{}& [$B_{\mathrm{T}}$]{}\
0.10–0.27 & 0.26–0.50 & 0.16–0.30\
[$B_{\mathrm{W}}$]{}& [$C$]{}& [$y_{23}^\mathrm{D}$]{}\
0.10–0.23 & 0.34-0.72 & 0.01–0.20\
The detector corrections (see section \[detcorr\]) are generally $\pm20$% or less within the fit ranges. The hadronisation correction factors are maximally 3.5 for [$B_{\mathrm{T}}$]{} at ${\ensuremath{\sqrt{s}}}=14$ GeV but are generally smaller than 2 for the larger cms energy points. The hadronisation corrections are smallest and have the least variations over the fit ranges for [$y_{23}^\mathrm{D}$]{}.
The evaluation of the systematic errors of the [$\alpha_{\mathrm{S}}$]{} measurements takes into account experimental effects, the hadronisation correction procedure and uncertainties of the theory. The three sources of systematic uncertainty are added in quadrature to the statistical error taken from the fits to obtain the total errors. Below we describe how we find the systematic uncertainties:
Experimental Uncertainties
: The analysis is repeated with slightly varied event and track selection cuts and a systematic uncertainty from variation of the fit ranges is studied [@jader4]. The cross section used in the subtraction of [$\mathrm{b\overline{b}}$]{} events is varied by $\pm5\%$ which takes account of possible differences in the efficiency determination using the simulations of ${\ensuremath{\mathrm{e^+e^-}}}\rightarrow{\ensuremath{\mathrm{q\overline{q}}}}$ (q=u,d,s,c) and ${\ensuremath{\mathrm{e^+e^-}}}\rightarrow{\ensuremath{\mathrm{b\overline{b}}}}$ events. For each experimental variation the value of [$\alpha_{\mathrm{S}}$]{} is determined and compared to the central (default) value. The quadratic sum of the differences and the fitrange uncertainty is taken as the experimental systematic uncertainty.
Hadronisation
: For the default analysis, PYTHIA is used to estimate the corrections originating from hadronisation effects (section \[qcdprediction\]). As a systematic variation HERWIG and ARIADNE are used to evaluate the effects of hadronisation. The larger of the deviations is taken as systematic hadronisation uncertainty. It was observed in [@jones03; @pedrophd] that systematic uncertainties between the PYTHIA, HERWIG and ARIADNE models are generally much larger than systematic uncertainties from varying the parameters of a given model.
Theoretical Uncertainties
: The theoretical prediction of event shape observables is a finite power series in [$\alpha_{\mathrm{S}}$]{}. The uncertainties originating from missing higher order terms are assessed by changing the renormalisation scale factor to ${\ensuremath{x_{\mu}}}=0.5$ and ${\ensuremath{x_{\mu}}}=2.0$. The larger deviation from the default value of [$\alpha_{\mathrm{S}}$]{} is taken as the systematic uncertainty.
Results from NNLO Fits {#sec_NLLOresults}
----------------------
The results of the NNLO fits are summarised in table \[asresultsnnlo\]. In figure \[esdistributions\] the [$1-T$]{}event shape distributions together with the NNLO fit results for the six energy points are shown. The [$\chi^2/\mathrm{d.o.f.}$]{} values are between 0.7 for ${\ensuremath{\sqrt{s}}}=14$ GeV and 2.5 for ${\ensuremath{\sqrt{s}}}=34.6$ GeV. The fit results for the other event shape observables return similar results with $0.3<{\ensuremath{\chi^2/\mathrm{d.o.f.}}}<3.8$. We note that at cms energies where we have big data samples the [$\chi^2/\mathrm{d.o.f.}$]{} value tend to be larger. The [$\chi^2/\mathrm{d.o.f.}$]{}values are based on statistical errors only while the combined experimental and hadronisation uncertainties are at least a factor of two larger than the statistical errors leading to a reduction of [$\chi^2/\mathrm{d.o.f.}$]{} by a factor of at least 4 if these uncertainties were taken into account in the fits. We conclude that there is no significant disagreement between the event shape data and the QCD fits.
The results at each cms energy are remarkably consistent with each other, we find root-mean-square (rms) values for ${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{\sqrt{s}}})$ between 0.003 at 44 GeV and 0.008 at 22 GeV. The hadronisation uncertainties at 14 GeV dominate the total errors (except for [$M_{\mathrm{H}}$]{} and [$C$]{}), because at 14 GeV the hadronisation corrections are largest. The statistical errors are sizeable at ${\ensuremath{\sqrt{s}}}=14$, 22, 38 and 44 GeV, where there is only limited statistics and quite small at ${\ensuremath{\sqrt{s}}}=34.6$ and 35 GeV where we have large data samples. The experimental uncertainties depend somewhat on the cms energy with smaller values at higher [$\sqrt{s}$]{} where we have larger data samples.
The hadronisation uncertainties for [$C$]{} are the largest except at ${\ensuremath{\sqrt{s}}}=14$ GeV. This as been observed before [@jadec; @pedrophd; @OPALPR425].
[ ccrrrrrc ]{} [$\sqrt{s}$]{} \[GeV\] & Obs. & ${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{\sqrt{s}}})$ & $\pm$stat. & $\pm$exp. & $\pm$had. & $\pm$theo. & [$\chi^2/\mathrm{d.o.f.}$]{}\
14.0 & [$1-T$]{}& 0.1587 & 0.0098 & 0.0213 & 0.0366 & 0.0100 & $ 4.0/ 6$\
14.0 & [$M_{\mathrm{H}}$]{}& 0.1759 & 0.0080 & 0.0133 & 0.0093 & 0.0099 & $ 9.0/ 5$\
14.0 & [$B_{\mathrm{T}}$]{}& 0.1687 & 0.0086 & 0.0098 & 0.0337 & 0.0132 & $ 1.6/ 5$\
14.0 & [$B_{\mathrm{W}}$]{}& 0.1730 & 0.0053 & 0.0088 & 0.0188 & 0.0088 & $ 3.8/ 5$\
14.0 & [$C$]{}& 0.1583 & 0.0150 & 0.0169 & 0.0113 & 0.0089 & $ 9.2/ 5$\
14.0 & [$y_{23}^\mathrm{D}$]{}& 0.1671 & 0.0039 & 0.0054 & 0.0101 & 0.0063 & $11.5/ 8$\
22.0 & [$1-T$]{}& 0.1410 & 0.0075 & 0.0054 & 0.0195 & 0.0070 & $ 6.8/ 6$\
22.0 & [$M_{\mathrm{H}}$]{}& 0.1555 & 0.0070 & 0.0105 & 0.0064 & 0.0061 & $ 5.3/ 5$\
22.0 & [$B_{\mathrm{T}}$]{}& 0.1399 & 0.0070 & 0.0050 & 0.0178 & 0.0076 & $ 4.0/ 5$\
22.0 & [$B_{\mathrm{W}}$]{}& 0.1551 & 0.0046 & 0.0046 & 0.0106 & 0.0060 & $10.2/ 5$\
22.0 & [$C$]{}& 0.1385 & 0.0084 & 0.0073 & 0.0332 & 0.0062 & $ 3.7/ 5$\
22.0 & [$y_{23}^\mathrm{D}$]{}& 0.1545 & 0.0033 & 0.0025 & 0.0084 & 0.0049 & $10.6/ 8$\
34.6 & [$1-T$]{}& 0.1396 & 0.0017 & 0.0040 & 0.0100 & 0.0069 & $14.8/ 6$\
34.6 & [$M_{\mathrm{H}}$]{}& 0.1477 & 0.0017 & 0.0070 & 0.0025 & 0.0053 & $18.9/ 5$\
34.6 & [$B_{\mathrm{T}}$]{}& 0.1392 & 0.0016 & 0.0025 & 0.0063 & 0.0076 & $ 6.4/ 5$\
34.6 & [$B_{\mathrm{W}}$]{}& 0.1457 & 0.0013 & 0.0047 & 0.0049 & 0.0049 & $ 3.5/ 5$\
34.6 & [$C$]{}& 0.1374 & 0.0017 & 0.0041 & 0.0126 & 0.0062 & $12.7/ 5$\
34.6 & [$y_{23}^\mathrm{D}$]{}& 0.1404 & 0.0009 & 0.0012 & 0.0061 & 0.0035 & $14.3/ 8$\
35.0 & [$1-T$]{}& 0.1430 & 0.0014 & 0.0040 & 0.0094 & 0.0074 & $14.0/ 6$\
35.0 & [$M_{\mathrm{H}}$]{}& 0.1532 & 0.0014 & 0.0065 & 0.0023 & 0.0059 & $ 7.0/ 5$\
35.0 & [$B_{\mathrm{T}}$]{}& 0.1432 & 0.0013 & 0.0042 & 0.0064 & 0.0082 & $13.8/ 5$\
35.0 & [$B_{\mathrm{W}}$]{}& 0.1496 & 0.0011 & 0.0063 & 0.0046 & 0.0054 & $10.7/ 5$\
35.0 & [$C$]{}& 0.1427 & 0.0014 & 0.0036 & 0.0118 & 0.0069 & $ 9.3/ 5$\
35.0 & [$y_{23}^\mathrm{D}$]{}& 0.1451 & 0.0008 & 0.0021 & 0.0061 & 0.0039 & $15.0/ 8$\
38.3 & [$1-T$]{}& 0.1427 & 0.0049 & 0.0064 & 0.0081 & 0.0073 & $11.3/ 6$\
38.3 & [$M_{\mathrm{H}}$]{}& 0.1578 & 0.0048 & 0.0067 & 0.0022 & 0.0066 & $ 2.0/ 5$\
38.3 & [$B_{\mathrm{T}}$]{}& 0.1447 & 0.0045 & 0.0068 & 0.0048 & 0.0085 & $ 1.8/ 5$\
38.3 & [$B_{\mathrm{W}}$]{}& 0.1488 & 0.0039 & 0.0069 & 0.0034 & 0.0053 & $ 3.1/ 5$\
38.3 & [$C$]{}& 0.1369 & 0.0049 & 0.0036 & 0.0101 & 0.0061 & $ 9.7/ 5$\
38.3 & [$y_{23}^\mathrm{D}$]{}& 0.1380 & 0.0028 & 0.0038 & 0.0071 & 0.0032 & $23.5/ 8$\
43.8 & [$1-T$]{}& 0.1341 & 0.0029 & 0.0034 & 0.0057 & 0.0061 & $11.3/ 6$\
43.8 & [$M_{\mathrm{H}}$]{}& 0.1403 & 0.0029 & 0.0062 & 0.0014 & 0.0046 & $11.5/ 5$\
43.8 & [$B_{\mathrm{T}}$]{}& 0.1312 & 0.0027 & 0.0027 & 0.0043 & 0.0064 & $ 9.8/ 5$\
43.8 & [$B_{\mathrm{W}}$]{}& 0.1373 & 0.0023 & 0.0050 & 0.0028 & 0.0041 & $10.7/ 5$\
43.8 & [$C$]{}& 0.1342 & 0.0028 & 0.0042 & 0.0083 & 0.0058 & $ 4.3/ 5$\
43.8 & [$y_{23}^\mathrm{D}$]{}& 0.1333 & 0.0017 & 0.0026 & 0.0054 & 0.0029 & $21.0/ 8$\
![The plots show as points with statistical error bars the [$1-T$]{} distributions at hadron level at ${\ensuremath{\sqrt{s}}}=14$ to 43.8 GeV. Some error bars are smaller than the data points. Superimposed as histograms are the NNLO+NLLA predictions combined with hadronisation effects using the corresponding fit results for ${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{\sqrt{s}}})$ shown in table \[asresultsnnlonlla\]. The arrows indicate the fit ranges.[]{data-label="esdistributions"}](eventshape_fit_1.eps){width="1.\columnwidth"}
Results from NNLO+NLLA Fits
---------------------------
The results of the NNLO+NLLA fits are given in table \[asresultsnnlonlla\]. The values of [$\chi^2/\mathrm{d.o.f.}$]{} are slightly smaller than the corresponding values of the NNLO fits in most cases, indicating a somewhat better consistency of the NNLO+NLLA calculations with the fitted data points. The rms values of ${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{\sqrt{s}}})$ are between 0.007 at ${\ensuremath{\sqrt{s}}}=22$ GeV and 0.003 at ${\ensuremath{\sqrt{s}}}=44$ GeV, i.e.the scatter of individual results is essentially the same as for the NNLO analysis. The pattern of statistical errors and experimental and hadronisation uncertainties is the same as for the NNLO fits discussed above. Compared with the NNLO analysis the theoretical uncertainties are reduced by $10-20$% and the values of [$\alpha_{\mathrm{S}}$]{} are lower by 4% on average. The hadronisation uncertainties of the NNLO+NLLA fits are also smaller in most cases. The difference in [$\alpha_{\mathrm{S}}$]{} between NNLO and NNLO+NLLA calculations is smaller than the difference in [$\alpha_{\mathrm{S}}$]{} between NLO and NLO+NLLA calculations as expected in [@gehrmann08]. As discussed above in section \[sec\_NLLOresults\] for NNLO fits the sometimes large [$\chi^2/\mathrm{d.o.f.}$]{} values can be explained by the small statistical errors in some data sets.
[ ccrrrrrc ]{} [$\sqrt{s}$]{} \[GeV\] & Obs. & ${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{\sqrt{s}}})$ & $\pm$stat. & $\pm$exp. & $\pm$had. & $\pm$theo. & [$\chi^2/\mathrm{d.o.f.}$]{}\
14.0 & [$1-T$]{}& 0.1543 & 0.0092 & 0.0197 & 0.0362 & 0.0089 & $ 4.3/ 6$\
14.0 & [$M_{\mathrm{H}}$]{}& 0.1641 & 0.0065 & 0.0105 & 0.0126 & 0.0095 & $ 9.2/ 5$\
14.0 & [$B_{\mathrm{T}}$]{}& 0.1620 & 0.0078 & 0.0083 & 0.0343 & 0.0122 & $ 1.5/ 5$\
14.0 & [$B_{\mathrm{W}}$]{}& 0.1540 & 0.0038 & 0.0060 & 0.0157 & 0.0072 & $ 1.4/ 5$\
14.0 & [$C$]{}& 0.1466 & 0.0128 & 0.0151 & 0.0131 & 0.0063 & $11.1/ 5$\
14.0 & [$y_{23}^\mathrm{D}$]{}& 0.1661 & 0.0038 & 0.0057 & 0.0129 & 0.0060 & $ 8.0/ 8$\
22.0 & [$1-T$]{}& 0.1383 & 0.0070 & 0.0051 & 0.0185 & 0.0061 & $ 6.1/ 6$\
22.0 & [$M_{\mathrm{H}}$]{}& 0.1464 & 0.0058 & 0.0089 & 0.0040 & 0.0049 & $ 4.8/ 5$\
22.0 & [$B_{\mathrm{T}}$]{}& 0.1350 & 0.0064 & 0.0044 & 0.0159 & 0.0062 & $ 3.4/ 5$\
22.0 & [$B_{\mathrm{W}}$]{}& 0.1408 & 0.0034 & 0.0035 & 0.0074 & 0.0047 & $ 7.2/ 5$\
22.0 & [$C$]{}& 0.1337 & 0.0078 & 0.0064 & 0.0323 & 0.0061 & $ 4.3/ 5$\
22.0 & [$y_{23}^\mathrm{D}$]{}& 0.1538 & 0.0033 & 0.0022 & 0.0090 & 0.0045 & $ 8.5/ 8$\
34.6 & [$1-T$]{}& 0.1365 & 0.0016 & 0.0036 & 0.0092 & 0.0057 & $13.7/ 6$\
34.6 & [$M_{\mathrm{H}}$]{}& 0.1399 & 0.0014 & 0.0055 & 0.0019 & 0.0041 & $15.9/ 5$\
34.6 & [$B_{\mathrm{T}}$]{}& 0.1338 & 0.0014 & 0.0021 & 0.0055 & 0.0058 & $ 6.3/ 5$\
34.6 & [$B_{\mathrm{W}}$]{}& 0.1332 & 0.0010 & 0.0034 & 0.0037 & 0.0035 & $ 6.2/ 5$\
34.6 & [$C$]{}& 0.1326 & 0.0016 & 0.0037 & 0.0111 & 0.0061 & $ 6.1/ 5$\
34.6 & [$y_{23}^\mathrm{D}$]{}& 0.1401 & 0.0009 & 0.0011 & 0.0059 & 0.0030 & $ 5.5/ 8$\
35.0 & [$1-T$]{}& 0.1399 & 0.0013 & 0.0037 & 0.0086 & 0.0062 & $10.5/ 6$\
35.0 & [$M_{\mathrm{H}}$]{}& 0.1444 & 0.0012 & 0.0050 & 0.0015 & 0.0047 & $ 8.1/ 5$\
35.0 & [$B_{\mathrm{T}}$]{}& 0.1376 & 0.0012 & 0.0038 & 0.0055 & 0.0065 & $11.9/ 5$\
35.0 & [$B_{\mathrm{W}}$]{}& 0.1363 & 0.0008 & 0.0048 & 0.0035 & 0.0040 & $12.8/ 5$\
35.0 & [$C$]{}& 0.1373 & 0.0013 & 0.0032 & 0.0104 & 0.0068 & $ 4.6/ 5$\
35.0 & [$y_{23}^\mathrm{D}$]{}& 0.1447 & 0.0008 & 0.0022 & 0.0059 & 0.0035 & $ 9.6/ 8$\
38.3 & [$1-T$]{}& 0.1400 & 0.0046 & 0.0060 & 0.0073 & 0.0064 & $10.2/ 6$\
38.3 & [$M_{\mathrm{H}}$]{}& 0.1484 & 0.0040 & 0.0050 & 0.0015 & 0.0056 & $ 1.9/ 5$\
38.3 & [$B_{\mathrm{T}}$]{}& 0.1390 & 0.0041 & 0.0060 & 0.0043 & 0.0068 & $ 1.3/ 5$\
38.3 & [$B_{\mathrm{W}}$]{}& 0.1360 & 0.0030 & 0.0052 & 0.0026 & 0.0042 & $ 2.5/ 5$\
38.3 & [$C$]{}& 0.1329 & 0.0045 & 0.0034 & 0.0085 & 0.0064 & $ 7.4/ 5$\
38.3 & [$y_{23}^\mathrm{D}$]{}& 0.1385 & 0.0028 & 0.0040 & 0.0063 & 0.0034 & $18.1/ 8$\
43.8 & [$1-T$]{}& 0.1313 & 0.0027 & 0.0031 & 0.0052 & 0.0048 & $11.6/ 6$\
43.8 & [$M_{\mathrm{H}}$]{}& 0.1337 & 0.0025 & 0.0049 & 0.0011 & 0.0034 & $10.8/ 5$\
43.8 & [$B_{\mathrm{T}}$]{}& 0.1265 & 0.0025 & 0.0023 & 0.0039 & 0.0047 & $10.0/ 5$\
43.8 & [$B_{\mathrm{W}}$]{}& 0.1269 & 0.0018 & 0.0037 & 0.0022 & 0.0027 & $11.1/ 5$\
43.8 & [$C$]{}& 0.1295 & 0.0025 & 0.0037 & 0.0072 & 0.0054 & $ 3.5/ 5$\
43.8 & [$y_{23}^\mathrm{D}$]{}& 0.1331 & 0.0017 & 0.0024 & 0.0050 & 0.0024 & $18.6/ 8$\
Combination of Results
----------------------
The results obtained at each energy point for the six event shape observables are combined using error weighted averaging as in [@jones04; @OPALPR404; @aleph265; @kluth06]. The statistical correlations between the six event shape observables are estimated at each energy point from fits to hadron-level distributions derived from 50 statistically independent Monte Carlo samples. The experimental uncertainties are determined assuming that the smaller of a pair of correlated experimental errors gives the size of the fully correlated error (partial correlation). The hadronisation and theory systematic uncertainties are found by repeating the combination with changed input values, i.e. using a different hadronisation model or a different value of [$x_{\mu}$]{}. The results are given in table \[asecmcomb\] and shown for the NNLO analysis in figure \[asvscme\], because this allows a direct comparison with the results of the NNLO analysis of ALEPH event shape data [@dissertori07].
The statistical uncertainties of the combined results are reduced as expected. The systematic uncertainties of the combined results tend to be close to the best values from individual observables, because the systematic uncertainties are not completely correlated and because observables with smaller uncertainties have larger weights in the combination procedure.
[ crrrrr ]{} [$\sqrt{s}$]{} \[GeV\] & ${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{\sqrt{s}}})$ & $\pm$stat. & $\pm$exp. & $\pm$had. & $\pm$theo.\
14.0 & 0.1690 & 0.0046 & 0.0065 & 0.0124 & 0.0076\
22.0 & 0.1527 & 0.0040 & 0.0036 & 0.0090 & 0.0056\
34.6 & 0.1420 & 0.0012 & 0.0025 & 0.0058 & 0.0050\
35.0 & 0.1463 & 0.0010 & 0.0032 & 0.0059 & 0.0055\
38.3 & 0.1428 & 0.0033 & 0.0045 & 0.0060 & 0.0051\
43.8 & 0.1345 & 0.0021 & 0.0031 & 0.0043 & 0.0045\
14.0 & 0.1605 & 0.0044 & 0.0065 & 0.0148 & 0.0073\
22.0 & 0.1456 & 0.0036 & 0.0033 & 0.0077 & 0.0048\
34.6 & 0.1367 & 0.0011 & 0.0023 & 0.0046 & 0.0040\
35.0 & 0.1412 & 0.0009 & 0.0032 & 0.0049 & 0.0047\
38.3 & 0.1388 & 0.0030 & 0.0043 & 0.0042 & 0.0048\
43.8 & 0.1297 & 0.0019 & 0.0028 & 0.0033 & 0.0034\
![The values for [$\alpha_{\mathrm{S}}$]{} at the JADE energy points. The inner error bars correspond to the combined statistical and experimental errors and the outer error bars show the total errors. The results from ${\ensuremath{\sqrt{s}}}=34.6$ and 35 GeV have been combined for clarity. The full and dashed lines indicate the result from our JADE NNLO analysis as shown on the figure. The results from the NNLO analysis of ALEPH data [@dissertori07] are shown as well.[]{data-label="asvscme"}](asplotjade.eps){width="1.\columnwidth"}
A combination of the combined results at the six JADE energy points shown in table \[asecmcomb\] after running to a common reference scale [$m_{{\ensuremath{\mathrm{Z^0}}}}$]{} using the combination procedure described above results in $$\begin{aligned}
\label{equresultnnlo}
{\ensuremath{{\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})}}& = 0.1210 & \pm 0.0007{\ensuremath{\mathrm{(stat.)}}}\pm 0.0021{\ensuremath{\mathrm{(exp.)}}}\\ \nonumber
& & \pm 0.0044{\ensuremath{\mathrm{(had.)}}}\pm 0.0036{\ensuremath{\mathrm{(theo.)}}}\end{aligned}$$ (${\ensuremath{{\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})}}= 0.1210\pm0.0061$) for the NNLO analysis and $$\begin{aligned}
\label{equresultnnlonlla}
{\ensuremath{{\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})}}& = 0.1172 & \pm 0.0006{\ensuremath{\mathrm{(stat.)}}}\pm 0.0020{\ensuremath{\mathrm{(exp.)}}}\\ \nonumber
& & \pm 0.0035{\ensuremath{\mathrm{(had.)}}}\pm 0.0030{\ensuremath{\mathrm{(theo.)}}}\end{aligned}$$ (${\ensuremath{{\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})}}=0.1172\pm0.0051$) for the NNLO+NLLA analysis. The NNLO+NLLA result has smaller hadronisation and theory uncertainties compared with the values in the NNLA analysis. We choose the latter result from NNLO+NLLA fits as our final result, because it is based on the most complete theory predictions and it has smaller theory uncertainties. It is consistent with the world average of ${\ensuremath{{\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})}}=0.119\pm0.001$ [@bethke06], the recent NNLO analysis of event shape data from the ALEPH experiment ${\ensuremath{{\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})}}=0.1240\pm0.0033$ [@dissertori07] as well as with the related average of ${\ensuremath{{\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})}}=0.120\pm0.005$ from the analyses of the LEP experiments using NLO+NLLA QCD predictions [@kluth06]. The total error for [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} of 4% is among the most precise determinations of [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} currently available.
After running the fit results for ${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{\sqrt{s}}})$ for each observable to the common reference scale [$m_{{\ensuremath{\mathrm{Z^0}}}}$]{} we combine the results for a given observable to a single value. We use the same method as above and obtain the results for [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} shown in table \[asvarcomb\]. The rms values of the results for [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} are 0.0029 for the NNLO analysis and 0.0026 for the NNLO+NLLA analysis; both values are consistent with the errors of the corresponding combined results shown in equations (\[equresultnnlo\]) and (\[equresultnnlonlla\]). Figure \[asobsscatter\] shows the combined results of [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} for each observable together with results from alternative analyses discussed below. Combining the combined results for each observable or combining all individual results after evolution to the common scale [$m_{{\ensuremath{\mathrm{Z^0}}}}$]{} yields results consistent with equation (\[equresultnnlonlla\]) within $\Delta{\ensuremath{{\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})}}=0.0004$ and the uncertainties also agree.
The hadronisation uncertainty of [$M_{\mathrm{H}}$]{} at each energy point and in the combinations shown in table \[asvarcomb\] is the smallest. We have repeated the combinations without [$M_{\mathrm{H}}$]{} and found results for [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{}consistent within 0.6% with our main results with hadronisation uncertainties increased by 14% (NNLO) or 20% (NNLO+NLLA).
[ crrrrr ]{} Obs. & [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{}& $\pm$stat. & $\pm$exp. & $\pm$had. & $\pm$theo.\
[$1-T$]{}& 0.1196 & 0.0011 & 0.0028 & 0.0067 & 0.0049\
[$M_{\mathrm{H}}$]{}& 0.1266 & 0.0009 & 0.0047 & 0.0014 & 0.0040\
[$B_{\mathrm{T}}$]{}& 0.1190 & 0.0009 & 0.0023 & 0.0047 & 0.0055\
[$B_{\mathrm{W}}$]{}& 0.1232 & 0.0008 & 0.0034 & 0.0037 & 0.0035\
[$C$]{}& 0.1184 & 0.0013 & 0.0029 & 0.0081 & 0.0045\
[$y_{23}^\mathrm{D}$]{}& 0.1201 & 0.0005 & 0.0014 & 0.0046 & 0.0026\
[$1-T$]{}& 0.1175 & 0.0010 & 0.0026 & 0.0061 & 0.0041\
[$M_{\mathrm{H}}$]{}& 0.1210 & 0.0008 & 0.0037 & 0.0011 & 0.0032\
[$B_{\mathrm{T}}$]{}& 0.1151 & 0.0009 & 0.0019 & 0.0039 & 0.0042\
[$B_{\mathrm{W}}$]{}& 0.1143 & 0.0006 & 0.0026 & 0.0028 & 0.0026\
[$C$]{}& 0.1148 & 0.0011 & 0.0027 & 0.0073 & 0.0044\
[$y_{23}^\mathrm{D}$]{}& 0.1199 & 0.0005 & 0.0013 & 0.0046 & 0.0023\
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --
![The combined results for [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} for each type of analysis as indicated on the figure. The shaded bands and dashed lines show the combined values of [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} with total errors. The inner and outer error bars show the combined statistical and experimental and the total errors. []{data-label="asobsscatter"}](obsscatterplot2.eps "fig:"){width="1.\columnwidth"}
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --
In order to study the compatibility of our data with the QCD prediction for the evolution of the strong coupling with cms energy we repeat the combinations with or without evolution of the combined results to the common scale. We set the theory uncertainties to zero since these uncertainties are highly correlated between energy points. We conservatively assume the hadronisation uncertainties to be partially correlated, because these uncertainties depend strongly on the cms energy. The [$\chi^2$]{} probabilities of the averages for running (not running) with NNLO+NLLA fits then become 0.39 ($9.9\cdot
10^{-3}$). With the NNLO fits the [$\chi^2$]{} probabilities for running (not running) are 0.48 ($1.2\cdot 10^{-3}$). We interpret this as strong evidence for the dependence of the strong coupling on cms energy as predicted by QCD from JADE data alone.
Comparison with NLO and NLO+NLLA {#compNLONLLA}
--------------------------------
For a comparison of our results with previous [$\alpha_{\mathrm{S}}$]{} measurements the fits to the event shape distributions are repeated with NLO predictions and with NLO predictions combined with resummed NLLA with the modified $\ln R$-matching scheme (NLO+NLLA), both with ${\ensuremath{x_{\mu}}}=1$. The NLO+NLLA predictions with the modified $\ln R$-matching scheme were the standard of the final analysis of the LEP experiments [@l3290; @aleph265; @OPALPR404; @delphi327]. The fit ranges as well as the procedures for evaluation of the systematic uncertainties are identical to the ones in our NNLO and NNLO+NLLA analyses.
The combination of the fits using NLO predictions returns ${\ensuremath{{\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})}}= 0.1301\pm 0.0009{\ensuremath{\mathrm{(stat.)}}}\pm 0.0029{\ensuremath{\mathrm{(exp.)}}}\pm 0.0054{\ensuremath{\mathrm{(had.)}}}\pm 0.0086{\ensuremath{\mathrm{(theo.)}}}$, the fits using combined NLO+ NLLA predictions yield ${\ensuremath{{\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})}}= 0.1172\pm 0.0007{\ensuremath{\mathrm{(stat.)}}}\pm 0.0022{\ensuremath{\mathrm{(exp.)}}}\pm 0.0039{\ensuremath{\mathrm{(had.)}}}\pm 0.0054{\ensuremath{\mathrm{(theo.)}}}$ and these results are shown in figure \[asobsscatter\]. The result obtained with the NLO+NLLA prediction is consistent with our NNLO and NNLO+NLLA analyses, but the theory uncertainties are larger by about 60%. The analysis using NLO predictions gives theoretical uncertainties larger by a factor of 2.6 and the value for [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} is larger compared to the NNLO or NNLO+NLLA results. It has been observed previously that values for [$\alpha_{\mathrm{S}}$]{} from NLO analysis with ${\ensuremath{x_{\mu}}}=1$ are large in comparison with most other analyses [@OPALPR054]. Both the NLO+NLLA or NNLO+NLLA analyses yield a smaller value of [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} compared to the respective NLO or NNLO results. The difference between NNLO+NLLA and NNLO is smaller than the difference between NLO+NLLA and NLO, since a larger part of the NLLA terms is included in the NNLO predictions.
Renormalisation Scale Dependence
--------------------------------
The theoretical uncertainty due to missing higher order terms is evaluated by setting the renormalisation scale parameter [$x_{\mu}$]{} to 0.5 or 2.0. In order to assess the dependence of [$\alpha_{\mathrm{S}}$]{} on the renormalisation scale the fits are repeated using NNLO, NNLO+NLLA, NLO and NLO+NLLA predictions with $0.1<{\ensuremath{x_{\mu}}}<10$. The strong coupling [$\alpha_{\mathrm{S}}$]{} as well as the [$\chi^2/\mathrm{d.o.f.}$]{} as a function of [$x_{\mu}$]{} for [$1-T$]{} at ${\ensuremath{\sqrt{s}}}=35$ GeV are shown in figure \[FigureScaleDep\]. The [$\chi^2/\mathrm{d.o.f.}$]{} curves for the NLO+NLLA and NNLO+NLLA fits show no local minimum in the [$x_{\mu}$]{} range studied. The [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} values from NLO predictions are the largest for ${\ensuremath{x_{\mu}}}>0.2$. The [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} values using NLO+NLLA and NNLO calculation almost cross at the natural choice of the renormalisation scale ${\ensuremath{x_{\mu}}}=1$ while the [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} value from the NNLO+NLLA fit is slightly lower. The NLLA terms at ${\ensuremath{x_{\mu}}}=1$ are almost identical to the [$\mathcal{O}({\ensuremath{\alpha_{\mathrm{S}}}}^3)$]{}-terms in the NNLO calculation. A similar behaviour can be observed for [$B_{\mathrm{T}}$]{} and [$y_{23}^\mathrm{D}$]{}. The slopes of the [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} curves of the NNLO and NNLO+NLLA fits around the default choice ${\ensuremath{x_{\mu}}}=1$ are smaller than the slopes for the NLO and NLO+NLLA fits leading to the decreased theoretical uncertainties in our analyses.
![The plot shows the result of [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} and $\chi^{2}$/d.o.f. of the fit to the thrust event shape distribution for ${\ensuremath{\sqrt{s}}}=35$ GeV.[]{data-label="FigureScaleDep"}](asvsxmu_-3_1-T.eps){width="0.75\columnwidth"}
Conclusion
==========
In this paper we present measurements of the strong coupling [$\alpha_{\mathrm{S}}$]{}using event shape observable distributions at cms energies $14<{\ensuremath{\sqrt{s}}}<44$ GeV. To determine [$\alpha_{\mathrm{S}}$]{} fits using NNLO and combined NNLO+NLLA predictions were used. Combining the results from NNLO+NLLA fits to the six event shape observables [$1-T$]{}, [$M_{\mathrm{H}}$]{}, [$B_{\mathrm{W}}$]{}, [$B_{\mathrm{T}}$]{}, [$C$]{} and [$y_{23}^\mathrm{D}$]{}at the six JADE energy points returns ${\ensuremath{{\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})}}= 0.1172\pm 0.0006{\ensuremath{\mathrm{(stat.)}}}\pm 0.0020{\ensuremath{\mathrm{(exp.)}}}\pm 0.0035{\ensuremath{\mathrm{(had.)}}}\pm 0.0030{\ensuremath{\mathrm{(theo.)}}}$, with a total error on [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} of 4%. The investigation of the renormalisation scale dependence of [${\ensuremath{\alpha_{\mathrm{S}}}}({\ensuremath{m_{{\ensuremath{\mathrm{Z^0}}}}}})$]{} shows a reduced dependence on [$x_{\mu}$]{} when NNLO or NNLO+NLLA predictions are used, compared to analyses with NLO or NLO+NLLA predictions. The more complete NNLO or NNLO+NLLA QCD predictions thus lead to smaller theoretical uncertainties in our analysis. The combined results for [$\alpha_{\mathrm{S}}$]{} at each cms energy are consistent with the running of [$\alpha_{\mathrm{S}}$]{} as predicted by QCD and exclude absence of running with a confidence level of 99%.
[10]{}
J. Allison$^e$, K. Ambrus$^c$, J. Armitage$^e$, J. Baines$^e$, A.H. Ball$^e$, G. Bamford$^e$, R. Barlow$^e$, W. Bartel$^a$, L. Becker$^a$, A. Bell$^d$, S. Bethke$^c$, C. Bowdery$^d$, T. Canzler$^a$, S.L. Cartwright$^g$, J. Chrin$^e$, D. Clarke$^g$, D. Cords$^a$, D.C. Darvill$^d$, A. Dieckmann$^c$, G. Dietrich$^b$, P. Dittmann$^a$, H. Drumm$^c$, I.P. Duerdoth$^e$, G. Eckerlin$^c$, R. Eichler$^a$, E. Elsen$^b$, R. Felst$^a$, A. Finch$^d$, F. Foster$^d$, R.G. Glasser$^f$, I. Glendinning$^e$, M.C. Goddard$^g$, T. Greenshaw$^e$, J. Hagemann$^b$, D. Haidt$^a$, J.F. Hassard$^e$, R. Hedgecock$^g$, J. Heintze$^c$, G. Heinzelmann$^b$, G. Heinzelmann$^b$, K.H. Hellenbrand$^c$, M. Helm$^b$, R. Heuer$^c$, P. Hill$^f$, G. Hughes$^d$, M. Imori$^h$, H. Junge$^a$, H. Kado$^b$, J. Kanzaki$^h$, S. Kawabata$^a$, K. Kawagoe$^b$, T. Kawamoto$^h$, B.T. King$^e$, C. Kleinwort$^b$, G. Knies$^a$, T. Kobayashi$^h$, S. Komamiya$^c$, M. Koshiba$^h$, H. Krehbiel$^a$, M. Kuhlen$^b$, P. Laurikainen$^a$, P. Lennert$^c$, F.K. Loebinger$^e$, A.A. Macbeth$^e$, N. Magnussen$^b$, R. Marshall$^g$, T. Mashimo$^h$, H. Matsumura$^c$, H. McCann$^e$, K. Meier$^b$, R. Meinke$^a$, R.P. Middleton$^g$, H.E. Mills$^e$, M. Minowa$^h$, P.G. Murphy$^e$, B. Naroska$^a$, T. Nozaki$^d$, M. Nozaki$^h$, J. Nye$^d$, S. Odaka$^h$, T. Oest$^b$, J. Olsson$^a$, L.H. O’Neill$^a$, S. Orito$^h$, F. Ould-Saada$^b$, G.F. Pearce$^g$, A. Petersen$^b$, E. Pietarienen$^a$, D. Pitzl$^b$, H. Prosper$^e$, J.J. Pryce$^d$, R. Pust$^b$, R. Ramcke$^b$, H. Rieseberg$^c$, P. Rowe$^e$, A. Sato$^h$, D. Schmidt$^a$, U. Schneekloth$^b$, B. Sechi-Zorn$^f$, J.A. Skard$^f$, L. Smolik$^c$, J. Spitzer$^c$, P. Steffen$^a$, K. Stephens$^e$, T. Suda$^h$, H. Takeda$^h$, T. Takeshita$^h$, Y. Totsuka$^h$, H. v.d.Schmitt$^c$, J. von Krogh$^c$, A. Wagner$^c$, S.R. Wagner$^f$, I. Walker$^d$, P. Warming$^b$, Y. Watanabe$^h$, G. Weber$^b$, A. Wegner$^b$, H. Wenninger$^a$, J.B. Whittaker$^g$, H. Wriedt$^d$, S. Yamada$^h$, C. Yanagisawa$^h$, W.L. Yen$^a$, M. Zachara$^a$, Y. Zhang$^a$, M. Zimmer$^c$, G.T. Zorn$^f$\
$^a$ Deutsches Elektronen-Synchrotron DESY, Hamburg, Germany\
$^b$ II. Institut für Experimentalphysik der Universität Hamburg, Germany\
$^c$ Physikalisches Institut der Universität Heidelberg, Germany\
$^d$ University of Lancaster, England\
$^e$ University of Manchester, England\
$^f$ University of Maryland, College Park, MD, USA\
$^g$ Rutherford Appleton Laboratory, Chilton, Didcot, England\
$^h$ International Center for Elementary Particle Physics, ICEPP, University of Tokyo, Tokyo, Japan
O. Biebel, Phys. Rep. [**340**]{}, 165 (2001)
M. Dasgupta, G.P. Salam, J. Phys. G [**30**]{}, R143 (2004)
G. Dissertori, I.G. Knowles, M. Schmelling, [Quantum Chromodynamics]{}. International Series of Monographs on Physics number 115, Clarendon Press (2003), Oxford
S. Kluth, Rept. Prog. Phys. [**69**]{}, 1771 (2006)
H. Fritzsch, Murray Gell-Mann, H. Leutwyler, Phys. Lett. B [**47**]{}, 365 (1973)
D.J. Gross, Frank Wilczek, Phys. Rev. Lett. [**30**]{}, 1343 (1973)
D.J. Gross, Frank Wilczek, Phys. Rev. D [**8**]{}, 3633 (1973)
H.David Politzer, Phys. Rev. Lett. [**30**]{}, 1346 (1973)
A. Gehrmann-De Ridder, T. Gehrmann, E.W.N. Glover, G. Heinrich, J. High Energy Phys. [**12**]{}, 094 (2007)
T. Gehrmann, G. Luisoni, H. Stenzel, Phys. Lett. B [**664**]{}, 265 (2008)
G. Dissertori et al., J. High Energy Phys. [**02**]{}, 040 (2008)
R.A. Davison, B.R. Webber, Eur. Phys. J. C [**59**]{}, 13 (2009)
B. Naroska, Phys. Rep. [**148**]{}, 67 (1987)
JADE Coll., J. Schieck et al., Eur. Phys. J. C [**48**]{}, 3 (2006), Erratum-ibid.C50:769,2007
T. Sj[ö]{}strand, Comput. Phys. Commun. [**82**]{}, 74 (1994)
G. Marchesini et al., Comput. Phys. Commun. [**67**]{}, 465 (1992)
L. L[ö]{}nnblad, Comput. Phys. Commun. [**71**]{}, 15 (1992)
OPAL Coll., G. Alexander et al., Z. Phys. C [**69**]{}, 543 (1996)
OPAL Coll., G. Abbiendi et al., Eur. Phys. J. C [**35**]{}, 293 (2004)
S. Brandt, Ch. Peyrou, R. Sosnowski, A. Wroblewski, Phys. Lett. [**12**]{}, 57 (1964)
E. Fahri, Phys. Rev. Lett. [**39**]{}, 1587 (1977)
T. Chandramohan, L. Clavelli, Nucl. Phys. B [**184**]{}, 365 (1981)
S. Catani, G. Turnock, B.R. Webber, Phys. Lett. B [**295**]{}, 269 (1992)
G. Parisi, Phys. Lett. B [**74**]{}, 65 (1978)
J.F. Donoghue, F.E. Low, S.Y. Pi, Phys. Rev. D [**20**]{}, 2759 (1979)
R.K. Ellis, D.A. Ross, A.E. Terrano, Nucl. Phys. B [**178**]{}, 421 (1981)
OPAL Coll., M.Z. Akrawy et al., Phys. Lett. B [**235**]{}, 389 (1990)
Coll., S. Komamiya et al., Phys. Rev. Lett. [**64**]{}, 987 (1990)
S. Catani et al., Phys. Lett. B [**269**]{}, 432 (1991)
S. Weinzierl, MZ-TH-08-22 (2008), 0807.3241
OPAL Coll., G. Alexander et al., Z. Phys. C [**72**]{}, 191 (1996)
OPAL Coll., G. Abbiendi et al., Eur. Phys. J. C [**40**]{}, 287 (2005)
R.K. Ellis, W.J. Stirling, B.R. Webber, [QCD and Collider Physics]{}. Vol. 8 of Cambridge Monographs on Particle Physics, Nuclear Physics and Cosmology, Cambridge University Press (1996)
R.W.L. Jones, M. Ford, G.P. Salam, H. Stenzel, D. Wicke, J. High Energy Phys. [**12**]{}, 007 (2003)
P.A. [Movilla Fernández]{}, [Ph.D.]{} thesis, RWTH Aachen, 2003, PITHA 03/01
JADE Coll., O. Biebel, P.A. [Movilla Fern[á]{}ndez]{}, S. Bethke et al., Phys. Lett. B [**459**]{}, 326 (1999)
OPAL Coll., G. Abbiendi et al., Eur. Phys. J. C [**53**]{}, 21 (2008)
R.W.L. Jones, Nucl. Phys. Proc. Suppl. [**133**]{}, 13 (2004)
ALEPH Coll., A. Heister et al., Eur. Phys. J. C [**35**]{}, 457 (2004)
S. Bethke, Prog. Part. Nucl. Phys. [**58**]{}, 351 (2007)
L3 Coll., P. Achard et al., Phys. Rep. [**399**]{}, 71 (2004)
DELPHI Coll., J. Abdallah et al., Eur. Phys. J. C [**37**]{}, 1 (2004)
OPAL Coll., P.D. Acton et al., Z. Phys. C [**55**]{}, 1 (1992)
[^1]: The members of the JADE collaboration are listed in [@authors]
[^2]: JApan-Deutschland-England
[^3]: Positron-Elektron-Tandem-Ring-Anlage
[^4]: Deutsches Elektronen SYnchrotron
[^5]: All particles with lifetimes greater than 300 ps are considered stable.
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- 'Josef Schicho, JKU Linz, Austria'
title: |
And Yet It Moves:\
Paradoxically Moving Linkages in Kinematics
---
Look at Figure \[fig:ell\]: you see a mechanism that is able to draw an ellipse. If you press gently on the green bar (connected to the right endpoint of the grey segment which is fixed), then the whole vehicle will start to move and bounce so that the red point traces the ellipse. Historically, it was a famous challenge in the 19th-century to find a mechanism that draws a straight line segment. Mathematicians even tried to prove the non-existence of an exact solution. But then the French engineer Peaucellier and the Russian mathematician Lipkin independently found an exact solution. Starting from the mechanism in Figure \[fig:ell\], we can do the same thing as well (even though this was not the solution of Peaucellier/Lipkin): you can change some of the lengths so that the ellipse degenerates into a line segment traced twice in a full round.
![A mechanism which is able to draw an ellipse. The short gray horizontal bar is fixed on the x-axis, whereas all the other bars are allowed to move, according to the rotational joints which link them one to another. []{data-label="fig:ell"}](Ellipse.jpg){width="7cm"}
#### Kempe’s Universality Theorem.
A few years after the invention of the “straight line mechanism” by Peaucellier and Lipkin, Kempe [@Kempe] proved that every plane algebraic curve can be drawn by a mechanism moving with one degree of freedom! His construction uses the implicit equation of the algebraic curve, and the linkage draws a bounded subset of the curve. Kempe himself admits that the mechanisms constructed by his general construction are quite complicated. One of the objectives in this article is to explain how to construct a mechanism that draws a given rational curve, i.e., a curve that it is given by a parameterization by rational functions. Compared with Kempe, this construction gives simpler results when it applies (not every algebraic curve is rational).
#### Unexpected Mobility.
Most of the mechanisms in this paper will be [*paradoxical*]{}, in the following sense: by a systematic counting of degrees of freedom and constraints, one can estimate if a given mechanism moves. For a paradoxical mechanism, this estimate predicts that the mechanism is rigid: there are sufficiently many constraints so that there should be no freedom left for motion, except moving the mechanism as a whole like a rigid body. Still, the mechanism does move non-trivially. We discuss five mathematical tools that somehow “explain” the unexpected mobility:
- edge colorings of graphs;
- factorization of polynomials over skew coefficient rings;
- symmetry as a rule changer for counting variables and constraints;
- a projective duality relating a set of relative positions to a set of geometric parameters;
- compactification, i.e., a closer analysis of “limit configurations at infinity”.
#### Links and Joints.
We need to introduce a few concepts from kinematics (please do not worry, we will keep the amount of definitions at a minimal level). A [*linkage*]{} (or mechanism) in 3-space is composed of rigid bodies called links (or bars, rods) that are connected by joints (e.g., hinges or spherical joints); examples occur in mechanical engineering and robotics, but also in sports medicine – the human skeleton may be considered as a quite complex linkage – and in chemistry, at a microscopic scale. If two links are connected by a joint, then the type of joint determines a set of possible relative positions of one link with respect to the other. A [*revolute joint*]{} (or R-joint or hinge) allows a one-dimensional set of rotations around an axis which is fixed in both links; this set is a copy of ${\mathrm{SO}}_2$. This type of joint appears most frequently, for example in doors and windows or in connection with wheels (see also Figure \[fig:joints\], left). A [*spherical joint*]{} (or S-joint) allows a three-dimensional set of rotations around a point which is fixed in both links; this set of motions is a copy of ${\mathrm{SO}}_3$. An example is the hip joint of the human skeleton (see Figure \[fig:joints\], middle). And a [*prismatic joint*]{} (or P-joint) allows a one-dimensional set of translations in a fixed direction; this set is theoretically a copy of ${\mathbb R}$, but in reality, it is a bounded interval. Teachers and students in mathematics often operate such a joint when moving a blackboard up and down (see Figure \[fig:joints\], right, for a different example).
![ A hinge, the hip joint (spherical), and a prismatic joint on a crane. []{data-label="fig:joints"}](hinge.jpg "fig:"){height="3cm"} ![ A hinge, the hip joint (spherical), and a prismatic joint on a crane. []{data-label="fig:joints"}](spherical.jpg "fig:"){height="3cm"} ![ A hinge, the hip joint (spherical), and a prismatic joint on a crane. []{data-label="fig:joints"}](crane.jpg "fig:"){height="3cm"}
#### Configurations.
If two links are not directly connected by a joint, then the set of possible relative positions of one with respect to the other is determined by other links and joints forming chains that connect the two given links. In general, the description is more complicated, and it is one of the main tasks of kinematics to determine these sets. In any case, they are subsets of the group ${\mathrm{SE}}_3$ of direct isometries, also known as Euclidean displacements. The set of all possible relative positions of any pair of rigid bodies of a linkage $L$ is called the [*configuration space*]{} of $L$. It is possible to express the constraints coming from the joints by algebraic equations in the joint parameters. Therefore, the configuration space is an algebraic variety. Its dimension is called the [*mobility*]{} of $L$.
A linkage is given by combinatoric data, namely the graph indicating which rigid bodies are connected by joints and the type of joints such as revolute, spherical, prismatic; and by geometric parameters determining the fixed position of the joint axis in each of the two links attached to any R-joint and the fixed position of the anchor point in each of the two links attached to any S-joint. The computation of the configuration space of a given linkage can be reduced to solving a system of algebraic equations with parameters, with the size of the system determined by the combinatorics. These systems form a rich source of computational problems in computer algebra and polynomial system solving (see [@SommeseWampler:11] and the references cited there).
#### Structure of the Paper.
The paper has 6 sections. In Section \[sec:gk\], we discuss combinatoric methods for estimating the dimension of the configuration space, based on counting variables and equational conditions; this is necessary to make precise what “paradoxical” means. Section \[sec:over\] deals with planar linkages whose links are line segments joined by revolute joints, also known as moving graphs; we discuss graphs that should be rigid but actually move. Section \[sec:rl\] deals with spatial linkages in the plane with revolute joints, and uses dual quaternions to construct examples of simply closed linkages that are paradoxically movable. Section \[sec:sym\] deals with symmetries and explains how they can change the counting rules. Section \[sec:pod\] deals with a particular type of linkage called multipods or Stewart platforms; here, projective duality is a powerful mathematical tool that allows us to construct paradoxical examples. Section \[sec:bond\] is concerned with the problem of finding necessary conditions for mobility, based on the idea to analyze the “configurations at infinity” of a mobile linkage. In the three subsections of Section \[sec:bond\], moving graphs, simply closed loops with revolute joints, and multipods are revisited from this point of view what happens at infinity.
#### Acknowledgements.
Matteo Gallet, Georg Grasegger, Christoph Koutschan, Jan Legersky, Zijia Li, Georg Nawratil, and Hans-Peter Schröcker are coauthors of papers of which I took pictures - thanks for allowing me to use their work. I also would like to thank Matteo Gallet, Zijia Li, and Jiayue Qi for helping to improve the narration. This work has been supported by the Austrian Science Fund (FWF): P31061.
Predicting Mobility {#sec:gk}
===================
Given the combinatorics of a linkage, i.e., the number of its rigid bodies and the information which of them are connected by joints, it is possible to estimate the mobility by counting free variables and equational constraints. In kinematics, this is called the [*Chebychev/Grübler/Kutzbach (CGK) formula*]{}.
#### Moving Graphs.
In this section, we start with the two-dimensional situation. Every link is a line segment in the plane ${\mathbb R}^2$. In the plane, it does not make sense to distinguish revolute joints and spherical joints, and we do not consider prismatic joints. All joints in the linkages we consider allowing rotations around a fixed point. The combinatorics of the linkage is conveniently described by a graph $G=(V,E)$, with vertices corresponding to joints and edges corresponding to links. If a line segment has three or more (say $k$) joints connecting to other links, then we have to “split it up” into several edges: we get $k$ vertices corresponding to joints and we connect them by ${k\choose 2}$ edges. For instance, the green link in Figure \[fig:ell\] will correspond to a triangle in the graph, which is geometrically degenerate because its three vertices are collinear. We assume that the linkage has no “dangling links”, i.e., no vertices of degree 1, because they would obviously rotate around the connected vertex.
For a graph $G=(V,E)$, an “edge length assignment” is a vector $\lambda\in{\mathbb R}^E$ indexed by the edges with positive real coordinates $\lambda_e$, $e\in E$. A configuration of $(V,E,\lambda)$ is a collection $(\rho_v)_{v\in V}$ with $\rho_v\in{\mathbb R}^2$, such that for any edge $e=(u,v)$, we have $||\rho_u-\rho_v||=\lambda_e$. Two configurations $\rho,\rho'$ are equivalent if there is a direct isometry $\sigma:{\mathbb R}^2\to{\mathbb R}^2$ of the plane such that $\sigma(\rho_v)=\rho'_v$ for all $v$. If we choose two vertices $v,w\in V$ such that $\rho_v\ne\rho_w$, then there is a unique representative $\rho'$ in the equivalence class of $\rho$ such $\rho_v=(0,0)$ and $\rho_w=(0,c)$ for some $c>0$; we then say that $\rho'$ is a [*normalized*]{} configuration.
For a given graph $G=(V,E)$ with edge length assignment $\lambda$, its normalized configurations are the solutions of a system of algebraic equations of the form $$(x_a-x_b)^2+(y_a-y_b)^2 = \lambda_{ab}^2$$ for each edge $\{a,b\}\in E$, and the normalization conditions $$x_v=y_v=x_w=0, y_w>0 .$$ The number of nonzero variables is $2|V|-3$, and the number of equations is $|E|$. We leave out the inequality, because it is inessential for the dimension count. Now the CGK formula predicts that the linkage is rigid if $2|V|-3=|E|$. If this number is nonnegative, then we call $2|V|-3-|E|$ the CGK estimate for the dimension of equivalence classes of configurations. In kinematics, this dimension is called the [*mobility*]{} of the linkage.
\(a) at (0,0) ; (b) at (1,0) ; (c) at (0.5,0.5) ; (d) at (0,1.5) ; (e) at (1,1.5) ; (f) at (0.5,1) ;
(a)edge(b) (b)edge(c) (c)edge(a) (a)edge(d) (d)edge(e) (e)edge(f) (f)edge(d) (b)edge(e) (c)edge(f);
\(a) at (0,0) ; (b) at (1,0) ; (c) at (0.5,0.5) ; (d) at (0,1) ; (e) at (1,1) ; (f) at (0.5,1.5) ;
(a)edge(b) (b)edge(c) (c)edge(a) (a)edge(d) (d)edge(e) (e)edge(f) (f)edge(d) (b)edge(e) (c)edge(f);
(d2) at (0.6,0.8) ; (e2) at (1.6,0.8) ; (f2) at (1.1,1.3) ;
(a)edge(b) (b)edge(c) (c)edge(a) (a)edge(d2) (d2)edge(e2) (e2)edge(f2) (f2)edge(d2) (b)edge(e2) (c)edge(f2);
#### Generic Mobility.
For a concrete instance, the CGK estimate comes without any warranties. But we can say something definite for the “generic case”. Here we use the word “generic” in the following sense. Assume that a certain statement depends on instances parametrized by an open subset of an irreducible algebraic variety $P$ (in most cases, $P$ is an open subset of a vector space). Then we say that the statement is generically true if the subset of instances such that the statement is false is contained in an algebraic subvariety of $P$ of strictly smaller dimension.
\[prop:cgk\] Let $G=(V,E)$ be a graph. Let $\lambda\in{\mathbb R}^E$ be a generic length assignment. Let $X_\lambda\in{\mathbb R}^{2|V|-3}$ be set of normalized configurations of $(V,E,\lambda)$. If $2|V|-3-|E|\ge 0$, then $X_\lambda$ is either empty or a real manifold of dimension $2|V|-3-|E|$. In particular, if $2|V|-3-|E|= 0$, then a generic length assignment allows only finitely many normalized configurations.
Let $f:{\mathbb R}^{2|V|-3}\to{\mathbb R}^{|E|}$ be the map $(x_a,y_a)_{a\in V}\mapsto ((x_a-x_b)^2+(y_a-y_b)^2)_{\{a,b\}\in E}$ (in the domain, remove the three coordinates known to be zero). This is a differential map, which assigns to each normalized configuration of points in ${\mathbb R}^2$ the square of the lengths of edges. Therefore $X_\lambda = f^{-1}(\lambda)$.
If the image of $f$ does not contain an open neighborhood of $\lambda$, then it also does not contain $\lambda$ because $\lambda$ is chosen generically. Hence $X_\lambda$ is empty and there is nothing left to prove.
Otherwise, let $U$ be an open neighborhood of $\lambda$ and apply Sard’s theorem to the map $f|_{f^{-1}(U)}$. It implies that the set of critical values does not contain $\lambda$. Hence the Jacobian of $f$ has rank $E$ at every point of $f^{-1}(U)$, and this shows the claim.
#### Generic Rigidity.
If $|E|=2|V|-3$, then two cases are possible: either the image of the map $f:{\mathbb R}^{2|V|-3}\to{\mathbb R}^{|E|}$ in the proof contains an open subset. Then the graph is rigid: a generic configuration cannot move continuously, by Proposition \[prop:cgk\]. Or the image of the map is contained in a subset of lower dimension. The following theorem determines which of the two cases holds.
\[thm:laman\] Let $G=(V,E)$ be a graph such that $|E|=2|V|-3$. Then there is an open set of edge assignments $\lambda$ with a finite and positive number of configurations if and only if $|E'|\le 2|V'|-3$ for every subgraph $G'=(V',E')$ of $G$.
This theorem was proven by Pollaczek-Geiringer [@Geiringer:27] and rediscovered 40 years later by Laman [@Laman:70]. The graph that satisfy the necessary and sufficient condition above are called [*Laman graphs*]{}. The necessity is easy to see: if there is a subgraph $G'=(V',E')$ with $|E'|> 2|V'|-3$, then the algebraic system describing normalized configurations of the subgraph is overdetermined. So, for generic edge length assignments, there is no configuration for the subgraph, and therefore also no configuration for the graph $G$ itself.
In dimension 3, the CGK estimate for the mobility of a graph $G=(V,E)$ is equal to $3|V|-6-|E|$. Proposition \[prop:cgk\] holds with that bound: if $\lambda\in{\mathbb R}^{|E|}$ is a generic edge assignment, and the normalized configuration space $X_\lambda$ is not empty, then it has dimension $3|V|-6-|E|$. The condition $|E'|\le 3|V'|-6$ for every subgraph $(V',E')$ is still necessary for the statement that $X_\lambda$ is generically not empty, but it is not sufficient: Figure \[fig:l3d\] shows the “double banana”, a graph with 8 vertices and 18 edges, such that a generic assignment of its vertices to points in ${\mathbb R}^3$ is flexible. The Jacobi matrix of the map $f$ mapping normalized configurations to edge assignments (see Proposition \[prop:cgk\]) is quadratic and singular. So the 3-dimensional analogue of Theorem \[thm:laman\] is not true, and the search for another combinatoric analogue is an active research topic in rigidity theory (see [@Meera]).
\(a) at (0,0) [1]{}; (b) at (0,1.5) [2]{}; (c) at (-0.6,0.5) [3]{}; (d) at (-0.5,0.6) [4]{}; (e) at (-0.36,0.55) [5]{}; (f) at (0.6,0.6) [6]{}; (g) at (0.5,0.4) [7]{}; (h) at (0.36,0.55) [8]{};
(a)edge(c) (a)edge(d) (a)edge(e) (b)edge(c) (b)edge(d) (b)edge(e) (c)edge(d) (c)edge(e) (d)edge(e); (a)edge(f) (a)edge(g) (a)edge(h) (b)edge(f) (b)edge(g) (b)edge(h) (f)edge(g) (f)edge(h) (g)edge(h);
#### Molecules.
For some classes of graphs, the 3-dimensional analogue of Theorem \[thm:laman\] is true. The most interesting class appears in a statement which used to be called the “Molecular Conjecture”, until it was proven in [@KatohTanigawa:11]. It is of special interest because it makes a statement on linkages that appear as models of molecules: atoms are modeled as balls with cylinders attached. A molecular joint is a cylinder who is joined to an atom at both of its ends (see Figure \[fig:molecule\]). From a kinematic point of view, a molecule model is a linkage with R-joints, such that for each link, all axis of joints attached to this link meet in a fixed point (the center of the atom).
![A kinematic model of the Methoxyethanol molecule $\mathrm{C}_3\mathrm{H}_6(\mathrm{OH})_2$. The cylinders are joints allowing a rotation around the central axis of the cylinder. Note that the axis always passes through the centers of the joined atoms.[]{data-label="fig:molecule"}](molecule.png){height="4cm"}
The following equivalent re-formulation appear in [@Jackson_Jordan:07]. For any graph, we can define its [*square*]{} by drawing an edge between any two vertices of graph distance two. A graph is called a [*square graph*]{} if it can be obtained as the square of a subgraph.
Assume that $G=(V,E)$ is a square graph such that $|E|=3|V|-6$. Assume that $|E'|\le 3|V'|-6$ for every subgraph of $G$. Then a generic assignment of the vertices by points in ${\mathbb R}^3$ defines a rigid embedding.
To see the equivalence with the molecule conjecture, start with a molecule and draw a graph $G$ with vertices corresponding to atoms and edges corresponding to cylinders in the molecular model. It is clear that every motion of the molecule fixes the length of each edge. However, every such motion also fixes the angle between two cylinders attached to the same atom. But this is equivalent to the statement that the motion fixes the length between the two atoms that are on the other end of the two cylinders. If you add an edge for any two such atoms, then you get exactly the square of $G$.
Overconstrained Linkages {#sec:over}
========================
Let us call a linkage [*paradoxical*]{} if a generic linkage with the same combinatoric structure is rigid, but the linkage itself is moving. For instance, an instance of a Laman graph which is mobile in the plane is paradoxical.
#### Should we Expect Paradoxical Linkages?
Let us do a simple variable counting, as in the CGK formula, to see if we should be surprised by the existence of paradoxical linkages. Fix a combinatorial structure, for instance a Laman graph $G=(V,E)$. For a generic instance, the number of non-equivalent configurations is finite. These configurations are real solutions of a system of algebraic equations; let $N_G$ be the number of complex solutions of these system. Note that the number of complex solutions does not depend on the choice of the generic instance, as long as the choice is generic, in contrast to the number of real solutions, which would depend on the choice of a generic instance.
For any system of equations that has finitely many solutions, it is possible to compute a single univariate polynomial, such that the solutions of the system are in bijection with the zeroes of the polynomial. In theory, it is possible to compute such a polynomial by introducing a new variable together with a generic linear equation between the new variable and the old variables, and then by eliminating all old variables. (In practice, it turns out that the elimination is quite costly.) The process can even be carried out in the presence of parameters, which will then also appear in the coefficients of the univariate polynomial. Let us therefore assume that we have now, for each graph $G=(V,E)$, such a polynomial $F_G$, with coefficients depending on an edge length assignment $\lambda$. The degree of $F_G$ would then have to be equal to $N_G$, because it has $N_G$ complex solutions and we may assume that $F_G$ is squarefree.
Now, a labeled graph $(V,E,\lambda)$ is mobile if and only of all $N_G+1$ coefficients of the polynomial are zero, i.e. the polynomial $F_G$ vanishes identically and there are infinitely many configurations. (We have to take non-real configurations into account, but let us ignore this point for the moment.) The instances of the graph form a family of dimension $|E|$ parametrized by the edge lengths. In order to find a paradoxical linkage, we need to find a solution of a system in $|E|$ variables with $N_G+1$ equations. So we need to compare these two numbers. If the number $|E|$ of variables is bigger than or equal to the number $N_G+1$ of equations, then we should not be surprised by the existence of paradoxical linkages.
Currently, we do not know any lower bounds for $N_G$, but there are conjectured lower bounds which are exponential in $|E|$, so the system of equations that would have to be fulfilled for the parameters of a paradoxical linkage would be highly overdetermined. This is also true for small graphs: for $5\le |V|\le 12$, the numbers $N_G$ are all known [@Schicho:17c], and we always have $|E|<N_G+1$. Consequently, the very existence of paradoxical linkages is itself paradoxical! At least, this is so for the type of linkages we considered in this counting, namely moving graphs in the plane.
#### Bipartite Graphs.
The smallest mobile Laman graphs have 6 vertices. One is the complete bipartite graph $K_{3,3}$. In [@Dixon:99], Dixon describes a construction to make arbitrary bipartite graphs mobile. The set $V$ of vertices is partitioned into two disjoint subsets $V_1,V_2$. Put all vertices in $V_1$ on the $x$-axis and all vertices of $V_2$ on the $y$-axis. An easy exercise using Pythagoras’ Theorem shows that the linkage is actually moving.
(a1) at (1.573,1.490) ; (a2) at (1.573,-1.490) ; (a3) at (-1.573,1.490) ; (a4) at (-1.573,-1.490) ; (b1) at (0.636,0.949) ; (b2) at (0.636,-0.949) ; (b3) at (-0.636,0.949) ; (b4) at (-0.636,-0.949) ;
(a1)edge(b1) (a1)edge(b2) (a1)edge(b3) (a1)edge(b4) (a2)edge(b1) (a2)edge(b2) (a2)edge(b3) (a2)edge(b4) (a3)edge(b1) (a3)edge(b2) (a3)edge(b3) (a3)edge(b4) (a4)edge(b1) (a4)edge(b2) (a4)edge(b3) (a4)edge(b4) ;
(a1) at (1.307,1.747) ; (a2) at (1.307,-1.747) ; (a3) at (-1.307,1.747) ; (a4) at (-1.307,-1.747) ; (b1) at (0.765,0.810) ; (b2) at (0.765,-0.810) ; (b3) at (-0.765,0.810) ; (b4) at (-0.765,-0.810) ;
(a1)edge(b1) (a1)edge(b2) (a1)edge(b3) (a1)edge(b4) (a2)edge(b1) (a2)edge(b2) (a2)edge(b3) (a2)edge(b4) (a3)edge(b1) (a3)edge(b2) (a3)edge(b3) (a3)edge(b4) (a4)edge(b1) (a4)edge(b2) (a4)edge(b3) (a4)edge(b4) ;
(a1) at (1.307,0.810) ; (a2) at (1.307,-0.810) ; (a3) at (-1.307,0.810) ; (a4) at (-1.307,-0.810) ; (b1) at (0.765,1.747) ; (b2) at (0.765,-1.747) ; (b3) at (-0.765,1.747) ; (b4) at (-0.765,-1.747) ;
(a1)edge(b1) (a1)edge(b2) (a1)edge(b3) (a1)edge(b4) (a2)edge(b1) (a2)edge(b2) (a2)edge(b3) (a2)edge(b4) (a3)edge(b1) (a3)edge(b2) (a3)edge(b3) (a3)edge(b4) (a4)edge(b1) (a4)edge(b2) (a4)edge(b3) (a4)edge(b4) ;
Using computer algebra, Husty/Walter[@HustyWalter:07] proved that Dixon’s construction is one of two possible mobile $K_{3,3}$’s; in all other cases, $K_{3,3}$ is rigid. The second mobile $K_{3,3}$, also found in [@Dixon:99], is a mobile $K_{4,4}$ with two points removed – see Figure \[fig:dixon2\]. The configuration has a finite symmetry group, namely the symmetry of a rectangle. Indeed, the points form two rectangles sharing their symmetry axes.
Note that Dixon I applies to arbitrary bipartite graphs. In contrast, the symmetric construction Dixon II does not scale, it just applies to $K_{4,4}$ and to its subgraphs.
#### NAC colorings.
Another construction that does scale is based on the possibility of partitioning the set $E$ of edges into two non-empty subsets $E_r,E_b$ of red and blue edges. We assume that every cycle in $G$ is either unicolored or has at least two edges of both colors; especially, triangles are always unicolored. Such a partition is called a NAC – “no almost (unicolored) cycle” – coloring. For each connected component of the subgraph $R_i$ of $(V,E_r)$ we assign a complex number $z_i$, and for each vertex of the subgraph $B_j$ of $(V,E_b)$, we assign a complex number $w_j$. Then we choose a real parameter $t$ parametrizing a periodic motion, as follows: map any vertex in $R_i\cap B_j$ to the point $z_i+e^{it}w_j\in{\mathbb C}$. But ${\mathbb C}$ is a model for the plane ${\mathbb R}^2$. Hence we have constructed, for any real value of $t$, a configuration of the graph in ${\mathbb R}^2$. The construction is continuous in $t$, so we may call it a motion. The blue edges always keep their orientation while the red edges are rotated with uniform speed, as in Figure \[fig:nac\].
![A mobile graph with a NAC coloring. The blue edges remain parallel to the original orientation, and the orientation of the red edges rotates with speed that is independent of the edge, as long as it is red.[]{data-label="fig:nac"}](nac.pdf){width="12cm"}
A partition of $E$ into $E_r\cup E_b$ is a NAC coloring if and only if we can map the vertices into the plane so that all red edge are parallel to the first coordinate axis and all blue edges are parallel to the second coordinate exists. It is obvious that this map defines a flexible embedding. Such a moving graph can be constructed by taking a very small moving graph, with three vertices and two edges, and enlarging it by parallel copies of edges. But wait - we can do the same with other graphs as well! Let us start with a moving quadrilateral. Then we add more edges that are parallel to one of the four edges of the quadrilateral. We get a bigger graph with the property that every motion of the quadrilateral induces a motion of the bigger graph – see Figure \[fig:oj\] for an example.
\(a) at (0,0) ; (b) at (5,0) ; (c) at (-2,1) ; (d) at (3,1) ; (e) at (2.411,3.766) ; (f) at (7.411,3.766) ; (g) at (0.411,4.766) ; (h) at (5.411,4.766) ;
(a)edge(b) (b)edge(d) (d)edge(e) (a)edge(e); (a)edge(c) (b)edge(f) (c)edge(d) (e)edge(f) (f)edge(h) (d)edge(h) (c)edge(g) (e) edge (g) (g)edge(h);
\(a) at (0,0) ; (b) at (5,0) ; (c) at (-1,2) ; (d) at (4,2) ; (e) at (2,4) ; (f) at (7,4) ; (g) at (1,6) ; (h) at (6,6) ;
(a)edge(b) (b)edge(d) (d)edge(e) (a)edge(e); (a)edge(c) (b)edge(f) (c)edge(d) (e)edge(f) (f)edge(h) (d)edge(h) (c)edge(g) (e) edge (g) (g)edge(h);
In Section \[sec:bond\], we will see that the existence of a NAC coloring is not only sufficient, but also necessary for the existence of a length assignment that makes a given graph mobile in ${\mathbb R}^2$. This result requires a few tools from algebraic geometry. More examples of graphs moving in the plane and NAC-colorings can be found in <https://jan.legersky.cz/project/movablegraphs/>.
Revolute Loops and Dual Quaternions {#sec:rl}
===================================
Let $n\ge 4$. An $n$R chain is a linkage consisting of $n+1$ links connected by $n$ revolute joints. In robotics, the first link is called the [*base*]{} and the last link is called the [*hand*]{} or [*end effector*]{}. Each joint can is controlled by an electric motor in such a way that the end effector performs a particular task.
If we firmly connect the first and the last link of an $n$R chain, then we get an $n$R loop: a linkage with $n$ links connected cyclically by $n$ revolute joints. According to the CGK formula, the mobility is $\max(0,n-6)$. If $n\ge 7$, then a generic $n$R loop is generically mobile. A generic 6R linkage is rigid; the number of configurations, including complex solutions, is $16$ (see [@selig05 11.5.1]). For $n=5$ and $n=4$, we obtain an overdetermined system of equations.
![ A thumbnail movie of a mobile 6R loop. Each of the 6 link is realized as a tetrahedron. Each tetrahedron has two edges, opposite to each other, playing the role of R-joints connecting the link to its to two neighbors. []{data-label="fig:thumb"}](dpendel1-30.png "fig:"){height="4.5cm"} ![ A thumbnail movie of a mobile 6R loop. Each of the 6 link is realized as a tetrahedron. Each tetrahedron has two edges, opposite to each other, playing the role of R-joints connecting the link to its to two neighbors. []{data-label="fig:thumb"}](dpendel1-32.png "fig:"){height="4.5cm"} ![ A thumbnail movie of a mobile 6R loop. Each of the 6 link is realized as a tetrahedron. Each tetrahedron has two edges, opposite to each other, playing the role of R-joints connecting the link to its to two neighbors. []{data-label="fig:thumb"}](dpendel1-34.png "fig:"){height="4.5cm"} ![ A thumbnail movie of a mobile 6R loop. Each of the 6 link is realized as a tetrahedron. Each tetrahedron has two edges, opposite to each other, playing the role of R-joints connecting the link to its to two neighbors. []{data-label="fig:thumb"}](dpendel1-36.png "fig:"){height="4.5cm"}\
![ A thumbnail movie of a mobile 6R loop. Each of the 6 link is realized as a tetrahedron. Each tetrahedron has two edges, opposite to each other, playing the role of R-joints connecting the link to its to two neighbors. []{data-label="fig:thumb"}](dpendel1-38.png "fig:"){height="4.5cm"} ![ A thumbnail movie of a mobile 6R loop. Each of the 6 link is realized as a tetrahedron. Each tetrahedron has two edges, opposite to each other, playing the role of R-joints connecting the link to its to two neighbors. []{data-label="fig:thumb"}](dpendel1-40.png "fig:"){height="4.5cm"} ![ A thumbnail movie of a mobile 6R loop. Each of the 6 link is realized as a tetrahedron. Each tetrahedron has two edges, opposite to each other, playing the role of R-joints connecting the link to its to two neighbors. []{data-label="fig:thumb"}](dpendel1-42.png "fig:"){height="4.5cm"} ![ A thumbnail movie of a mobile 6R loop. Each of the 6 link is realized as a tetrahedron. Each tetrahedron has two edges, opposite to each other, playing the role of R-joints connecting the link to its to two neighbors. []{data-label="fig:thumb"}](dpendel1-44.png "fig:"){height="4.5cm"}
\[rem:red\] Revolute loops may be considered as special cases of linkages of graph type, in the following way: we pick two points on each joint axis and connect them by an edge. For each link, we draw 4 additional edges connecting the points on the two axes that belong to the link, so that every link carries a complete graph $K_4$, which is geometrically a tetrahedron. This graph has $2n$ vertices and $5n$ edges, and it is apparent that the linkage of graph type has exactly the same mobility as the revolute loop. See Figure \[fig:thumb\] for an example of a tetrahedral 6R loop.
But even though revolute loops may be considered as a subclass of linkages of graph type, it is advantageous to introduce new techniques especially suited for them.
#### 4R Loops.
The classification of mobile 4R loops is due to Delassus [@Delassus]. He proved that there are three types of mobile 4R linkages:
planar:
: all rotation axes are parallel. Essentially, this is a quadrilateral moving in the plane. The third coordinate is not changed in any of the moving links.
spherical:
: all rotation axes pass through a single point. Essentially, this is a moving spherical quadrilateral. The planar case may be considered as a limit case of the spherical case.
skew isogram:
: Bennett [@Bennett:14] discovered a mobile 4R linkage such that the axes of joints attached to the same link are skew, for all four links. We describe it below in more detail.
Let $L_1,\dots,L_n=L_0$ be the rotation axes of in some configuration of an $n$R loop. For $i=0,\dots,n-1$, we assume that the lines $L_i$ and $L_{i+1}$ belong to the $i$-th link. Since the link is assumed to be a rigid body, the normal distance $d_i$ and the angle $\alpha_i$ between $L_i$ and $L_{i+1}$ does not change as the linkage moves: they are invariant parameters. Assume that none of the angles is zero, i.e., $L_i$ and $L_{i+1}$ are not parallel. Then there is a unique line $N_i$ intersecting both $L_i$ and $L_{i+1}$ at a right angle. The distance $s_i$ between $N_i\cap L_i$ and $N_i\cap L_{i+1}$ is called offset. The angles, normal distances, and offsets are $3n$ invariant geometric parameters of the linkage; in robotics, they are called the [*invariant Denavit-Hartenberg parameters*]{} [@DeHa]. A configuration is determined by $n$ angles, and the $3n$ invariant Denavit-Hartenberg parameters together with the $n$ configuration parameters determine the positions of the $n$ rotation axes and the position of the links uniquely up to ${\mathrm{SE}}_3$. These $4n$ parameters fulfill a condition of codimension 6, called the [*closure equation*]{}: we attach an internal coordinate system to each link, with the axis $L_i$ being the $x$ and the common normal $N_i$ being the $z$-axis. Then the transformation of the $i$-th coordinate system to the $(i+1)$-th coordinate system is the composition of the translation by a vector of length $d_i$ parallel to the $z$-axis, the rotation around the $z$-axis by the angle $\alpha_i$, the translation by a vector of length $s_i$ parallel to the $x$ axis, and a rotation around the $x$-axis determined by the $i$-th configuration parameter. The product of all these $4n$ direct isometries is equal to the identity, and this statement gives the closure equation.
![The [*skew isogram*]{} is a mobile linkage of type 4R-loop with four rotation axes, so that axes in the same link are always skew. It is the only mobile 4R-loop which is neither planar (all axes are parallel) nor spherical (all axes are concurrent). []{data-label="fig:skewiso"}](bennett.pdf){height="4cm"}
A skew isogram is a 4R linkage such that the invariant Denavit-Hartenberg parameters $d_0,\dots,s_3$ satisfy the conditions $$\label{eq:bennett}
d_1=d_3,d_0=d_2,\alpha_1=\alpha_3,\alpha_0=\alpha_2,\frac{d_1}{\sin(\alpha_1)}=\frac{d_0}{\sin(\alpha_0)}, s_0=s_1=s_2=s_3=0 .$$
#### Dual Quaternions.
In order to prove that the skew isogram is mobile, we use an algebraic way suggested by [@Study:03] to parametrize ${\mathrm{SE}}_3$. The algebra ${\mathbb D}{\mathbb H}$ of dual quaternions is the 8-dimensional real vector space generated by $1,{\mathbf i},{\mathbf j},{\mathbf k},{\epsilon},{\epsilon}{\mathbf i},{\epsilon}{\mathbf j},{\epsilon}{\mathbf k}$. Its multiplication is ${\mathbb R}$-linear, associative, the element ${\epsilon}$ – the dual unit – is central and satisfies ${\epsilon}^2=0$. The symbols ${\mathbf i},{\mathbf j},{\mathbf k}$ are Hamiltonian quaternions: ${\mathbf i}^2={\mathbf j}^2={\mathbf k}^2={\mathbf i}{\mathbf j}{\mathbf k}=-1$. The center, generated by $1$ and ${\epsilon}$, is called the algebra of dual numbers. Conjugation is a ${\mathbb D}$-linear antihomomorphism from ${\mathbb D}{\mathbb H}$ to itself: it maps $1$ to itself, ${\mathbf i}$ to $-{\mathbf i}$, ${\mathbf j}$ to $-{\mathbf j}$, and ${\mathbf k}$ to $-{\mathbf k}$. For any dual quaternion $h\in{\mathbb D}{\mathbb H}$, the element $N(h):=h\overline{h}=\overline{h}h$ is a dual number, called the norm of $h$. The norm is a semigroup homomorphism with respect to multiplication; its image is the subsemigroup consisting of all dual numbers with nonnegative primal part.
The set ${\mathbb S}$ of dual quaternions with norm in ${\mathbb R}^\ast$ is a multiplicative group. Its center is ${\mathbb R}^\ast$. The quotient group ${\mathbb S}/{\mathbb R}^\ast$ happens to be isomorphic to ${\mathrm{SE}}_3$. The isomorphism is determined by the action of ${\mathbb S}/{\mathbb R}^\ast$ on ${\mathbb R}^3$. We may regard ${\mathbb R}^3$ as the abelian normal subgroup $T$ of ${\mathbb S}/{\mathbb R}^\ast$ of classes represented by dual quaternions of the form $1+x{\epsilon}{\mathbf i}+y{\epsilon}{\mathbf j}+z{\epsilon}{\mathbf k}$ (this subgroup is going to be the subgroup of translations in ${\mathrm{SE}}_3$). The substitution of ${\epsilon}$ by $-{\epsilon}$ is an outer automorphism of ${\mathbb S}/{\mathbb R}^\ast$ of order 2 – lets call it $\tau$ – which fulfills the following property: if $h\in {\mathbb S}/{\mathbb R}^\ast$, then $h^{-1}\tau(h)\in T$. This implies that for all $h\in {\mathbb S}/{\mathbb R}^\ast$ and $v\in T$, the element $h^{-1}v\tau(h)=(h^{-1}vh)(h^{-1}\tau(h))$ is in $T$, and this defines a right action of ${\mathbb S}/{\mathbb R}^\ast$ on $T$. The bijections of $T$ in the image of this action are direct isometries, and this defines a group isomorphism ${\mathbb S}/{\mathbb R}^\ast\cong{\mathrm{SE}}_3$. At the same time, we have constructed an embedding of ${\mathrm{SE}}_3$ into the projective space ${\mathbb P}({\mathbb D}{\mathbb H})\cong{\mathbb P}^7$, as the subset defined by a quadratic form $S=0$, namely the dual part of the norm, and by a quadratic inequation $N\ne 0$, namely the primal part of the norm.
There is a bijection between elements of order 2 in ${\mathrm{SE}}_3$ and lines in ${\mathbb R}^3$: every line corresponds to a [*half turn*]{} round that line (a rotation by the angle $\pi$). A point in ${\mathrm{SE}}_3\subset{\mathbb P}({\mathbb D}{\mathbb H})$ has order 2 if and only if its scalar part is zero. Here we have two linear equations, namely the coefficient of $1$ and the coefficient of ${\epsilon}$, defining a ${\mathbb P}^5$ in ${\mathbb P}({\mathbb D}{\mathbb H})$. The intersection of this ${\mathbb P}^5$ with the quadric hypersurface defined by $S$ (a.k.a. the [*Study quadric*]{}) is isomorphic to the Plücker quadric, and the remaining six coefficients are the Plücker coordinates of lines.
Let $l\in{\mathbb D}{\mathbb H}$ be a dual quaternion representing an element of order 2 in ${\mathrm{SE}}_3$. Then $l^2=-N(l)$ is a negative real number; without loss of generality, we may assume $l^2=-1$. The line connecting $[1]$ and $[l]$ is contained in the Study quadric: its elements are the rotations around the line $L$ corresponding to $l$. (Note that $[1]$ denotes the equivalence class of the dual quaternion $1$ in ${\mathbb P}^7$ and does not indicate a reference to the bibliography.) These elements form a group; indeed, the vector space generated by $1$ and $l$ is a subalgebra isomorphic to ${\mathbb C}$ over ${\mathbb R}$, and the projectivization of this two-dimensional real algebra is a Lie group isomorphic to $\mathrm{SO}_2$. We call this group the [*revolution*]{} with axes $L$. A parametric representation of the revolution is $(t+l)_t$, where the parameter $t$ ranges over the real projective line; the parameter $t=0$ corresponds to $[l]$, and the parameter $t=\infty$ corresponds to $[1]$. In general, the parameter $t$ corresponds to the cotangent of half of the rotation angle.
\[rem:revolution\] Conversely, assume that we have a line in $S$ passing through $[1]$. Then we can parametrize it by a linear polynomial in $t$ with leading coefficient $1$, i.e., by a polynomial $(t+h)$ with $h\in S$. Because $N(t+h)=t^2+(h+\bar{h})t+N(h)$ has to be real for all $t\in{\mathbb R}$, it follows that $h+\bar{h}\in{\mathbb R}$: the scalar part of $h$ is real (its dual part is zero). Then a reparametrization of the line is $(s+\frac{h-\bar{h}}{2})_s$, setting $s=t+\frac{h+\bar{h}}{2}$. This reparametrization shows that the line parametrizes a revolution with axis corresponding to $[h-\bar{h}]$, except in the case when $N(h-\bar{h})=0$. In the exceptional case, the line will parametrize a translation along a fixed direction.
Let us now study conics passing through $[1]$ and contained in the Study quadric. Any such conic has a quadratic parametrization $(t^2+at+b)_t$ where $a,b\in{\mathbb D}{\mathbb H}$. Does this quadric polynomial factor into two linear polynomials? And if yes, do the linear polynomials parametrize revolutions? To answer these questions, we study ${\mathbb D}{\mathbb H}[t]$, the non-commutative algebra of univariate polynomials with coefficients in ${\mathbb D}{\mathbb H}$, where the variable $t$ is supposed to be central, i.e., it commutes with the coefficients.
#### Quaternion Polynomials.
As a preparation, let us ask the analogous question for the non-commutative algebra ${\mathbb H}[t]$. We will show that here, every polynomial can be written as a product of linear factors; in other words, the skew field of quaternions is algebraically closed! The proof is taken from [@GordonMotzkin:65].
\[lem:euclid\] Let $A,B\in{\mathbb H}[t]$, $B\ne 0$. Then there exist unique polynomials $Q,R\in{\mathbb H}[t]$ such that $A=QB+R$ and either $\deg(R)<\deg(B)$ or $R=0$.
We start with uniqueness. Assume $Q_1B+R_1=Q_2B+R_2$ for $\deg(R_1)<\deg(B)$ and $\deg(B_2)<\deg(R)$. We obtain $(Q_1-Q_2)B=R_2-R_1$. If the left side of this equation is not zero, then its degree is at least $\deg(B)$. If the right side s not zero, then its degree is less than $\deg(B)$. Hence both sides must be zero.
For the existence, we proceed by induction on the degree of $A$: if $\deg(A)<\deg(B)$, then we set $Q:=0$ and $R:=A$. If $\deg(A)\ge\deg(B)$, then we can write $A=ht^{\deg(A)-\deg(B)}B+A'$ for a suitable $h\in{\mathbb H}$ and $A'\in{\mathbb H}[t]$ with $\deg(A')<\deg(A)$. By induction, we get $A'=Q'B+R'$. But then we can set $Q:=Q+ht^{\deg(A)-\deg(B)}$ and $R:=R'$.
If $\deg(B)=1$ in the Lemma \[lem:euclid\], say $B=t-h$, then $R$ is a constant in ${\mathbb H}$. The constant is zero if and only if $(t-h)$ is a right factor of $A$. If this is true, then we also say “$h$ is a right zero of $A$”. So, the questions is: does every polynomial $A$ of positive degree have a right zero? And maybe we are also interested in the question how to find it.
A right zero of $A$ is also a right zero of the norm polynomial $N(A)=\bar{A}A$. We know that that the norm polynomial is in ${\mathbb R}[t]$. It is also the sum of four squares – if $A=A_0+A_1{\mathbf i}+A_2{\mathbf j}+A_3q_k$, then $N(A)=A_0^2+A_1^2+A_2^2+A_4^2$. If $N(A)$ has a real zero $r$, then this real zero is also a zero of $A_0,A_2,A_2,A_3$; hence it is a zero of $A$, and we have found what we wanted to find.
What do we do if $N(A)$ has no real zeroes? In this case, we choose a quadratic irreducible factor $M\in{\mathbb R}[t]$. By Lemma \[lem:euclid\], there are $Q,R\in{\mathbb H}[t]$, with $\deg(R)<2$ or $R=0$, such that $A=QM+R$. We distinguish three cases.
1. If $R=0$, then $M$ is a right factor of $A$. Every right zero of $M$ is also a right zero of $A$. So it suffices to show that $M$ has a right zero. But we know that $M$ has a complex zero. So, assume that $z=a+{\mathrm i}b$ is a complex zero of $M$, for some $a,b\in{\mathbb R}$, $b\ne 0$. Then we have the equation $M=(t-a-{\mathrm i}b)(t-a+{\mathrm i}b)$ between complex polynomials. But now we can replace the complex number ${\mathrm i}$ by the dual quaternion ${\mathbf i}$, which also fulfills the equation ${\mathbf i}^2+1=0$ It follows that $M=(t-a-{\mathbf i}b)(t-a+{\mathbf i}b)=0$, and $a-{\mathbf i}b$ is a right zero of $M$ and also a right zero of $A$.
2. If $\deg(R)=1$, say $R=ut+v$ for suitable $u,v\in{\mathbb H}$, $u\ne 0$, then $h:=u^{-1}v$ is a right zero of $R$. Since $$\label{eq:r}
\overline{R}{R}=(\overline{P}-\overline{Q}M)(P-QM)=N(P)+M(-\overline{Q}P-\overline{P}Q+\overline{Q}QM)$$ is a multiple of $M$, and $\deg(\overline{R}{R})=\deg(M)=2$, it follows that $M$ is a left multiple of $R$. It follows that $h$ is right zero of $M$. Hence it is also a right zero of $A=QM+R$.
3. If $\deg(R)=0$, then Equation \[eq:r\] is self-contradictory: the right side is a multiple of $M$, and the left side is a nonzero constant. So, this case cannot occur.
Every polynomial in ${\mathbb H}[t]$ can be written as a product of linear polynomials.
The proof is already clear: given $A$ of positive degree, we can find a right $h$, write $A=A'(t-h)$, and iterate.
How many distinct factorizations do there exist? Starting with one factorization, we may get infinitely many distinct factorizations by multiplying with constants and their inverses in between the linear factors. In order to get rid of these “essentially same” factorizations, it suffices to assume that the polynomial $A$ and the linear factors are normed, i.e., they have leading coefficient 1.
If $A$ is a multiple of an irreducible real quadric $M$ (the first case in the above case distinction), then $A$ has infinitely many right zeroes (see [@GordonMotzkin:65]). But if not, then the number of distinct factorizations is finite. Indeed, the only non-deterministic step in the iterative procedure sketched above is the choice of the sequence of irreducible factors used for factoring out the right zeroes. In particular, we have:
A normed polynomial of degree $d$ with generic coefficients has exactly $d!$ distinct factorizations into normed linear factors.
The comparison with polynomial factorization in ${\mathbb C}[t]$ is illuminating: there, the factorization is unique. But if we consider two factorizations which differ only by the order of the factors as being distinct, then we have again $d!$ distinct factorizations. In the case of ${\mathbb H}[t]$, permutation of factors would not lead to the same product, because ${\mathbb H}[t]$ is not commutative; hence permutation is not a method to get more factorizations, and all $d!$ factorizations are different.
#### Mobility of the Skew Isogram.
Feeling well prepared? Then, let us go back to polynomials over the dual quaternions. Can we write every polynomial in ${\mathbb D}{\mathbb H}[t]$ that parametrizes a curve in the Study quadric into a linear factors parametrizing revolutions?
Let us assume that we have given such a polynomial $P\in{\mathbb D}{\mathbb H}[t]$. We can try to copy the factorization strategy that worked in ${\mathbb H}[t]$: factorize the norm polynomial $N(P)$, choose a quadratic irreducible factor $M$ (lets assume that $N(P)$ has no real zeroes for now), compute the remainder of $P$ modulo $M$; if this remainder is a linear polynomial $R=ut+v$ for some $u,v\in{\mathbb D}{\mathbb H}$, compute a right zero $h:=u^{-1}v$, factor out $(t-h)$ from the right, and iterate. This is going to work for generic coefficients. Moreover, since $N(P(t_0))$ is in ${\mathbb R}$ (and not in ${\mathbb D}\setminus{\mathbb R}$) for all $t_0\in{\mathbb R}$, the norm polynomial $N(P)$ is in ${\mathbb R}[t]$. Therefore it has a factorization into irreducible factors in $M_r\in{\mathbb R}[t]$, $r=1,\dots,\deg(P)$. The right factors $(t-h_r)$ produced by our strategy satisfy the equation $(t-h_r)(t+h_r)=M_r$, so by Remark \[rem:revolution\], the linear factor will generically parametrize a revolution. So, at least generically, everything is fine!
The application of our strategy leads to the following characterization of skew isograms. It was first found in [@BHS] by different methods.
\[thm:bhs\] For a generic conic in the Study quadric passing through $[1]$, there is a skew isogram such that the conic parametrizes the motion of the second link. (In particular, this skew isogram is mobile.)
Let $P=t^2+at+b$ be a quadratic parametrization of the conic, with $a,b\in{\mathbb D}{\mathbb H}$. The norm $N(P)$ is a real polynomial that has only nonnegative values. By genericity, it has no double zeroes, and can be written as a product $M_1M_2$ of two distinct quadratic irreducible factors. For $i=1,2$, we construct as above a factorization $P=(t-r_i)(t-w_i)$ such that $N(t-w_i)=M_i$. (It follows that $N(t-r_i)=N(t-w_{2-i})$, for $i=1,2$.)
The linear polynomials $t-r_1,t-w_1,t-r_2,t-w_2$ parametrize lines on the Study quadric. Each of them corresponds to a subgroup of rotations around a line in ${\mathbb R}^3$. Let $L_1,K_1,L_2,K_2$ be these four lines, respectively. We construct a mobile 4R loop as follows: the base link contains the lines $L_1$ and $L_2$, the first link contains the lines $L_1$ and $K_1$, the second link contains the lines $K_1$ and $K_2$, and the third link contains the lines $L_2$ and $K_2$. For each $t\in ({\mathbb R}\cup\{\infty\})$, we get a configuration of the 4R loop: the relative displacement of the first link with respect to the base link is the rotation $t-r_1$, the relative displacement of the third link with respect to the base link is the rotation $t-r_2$, the relative motion of the second link with respect to the first link is the rotation $t-w_1$, and relative motion of the second link with respect to the third link is the rotation $t-w_2$. The relative position of the second link with respect to the base link can be computed in two ways, via the first link or via the third link. In both ways, the result is $(t-r_1)(t-w_1)=(t-r_2)(t-w_2)=P$.
Once the lines are constructed, it is straightforward to compute the invariant Denavit-Hartenberg parameters of the 4R loop – we omit this calculation. The result are exactly the equations \[eq:bennett\]. It follows that the 4R loop is a skew isogram.
The paper [@BHS] also contains the converse statement: for any skew isogram, the relative motion of two links that are not connected by a joint is parametrized by a conic curve on the Study quadric that passes through $[1]$. In [@hss1], factorizations of cubic polynomials in ${\mathbb D}{\mathbb H}[t]$ are used to construct paradoxically moving 5R loops and 6R loops.
#### Drawing Rational Curves.
It is time to lift the veil of the mystery about the ellipse circle shown in Figure \[fig:ell\]. This example is taken from [@Schicho:16c], which contains a construction of a linkage that draws a rational plane curve. In [@Schicho:18a], the construction is extended to rational space curves. An online illustration with several examples can be found in <http://www.koutschan.de/data/link/>.
The ellipse with implicit equations $\frac{(x+a)^2}{a^2}+\frac{y^2}{b^2}=z=0$ has a rational parametrization $$(x,y,z) = p(t) := \left( \frac{-2a}{t^2+1},\frac{2bt}{t^2+1},0 \right) .$$ For any $t\in{\mathbb R}$, the dual quaternion $1+{\epsilon}(\frac{-a}{t^2+1}{\mathbf i}+\frac{bt}{t^2+1}{\mathbf j})$ represents a translation that maps the origin to $p(t)$. The class of a dual quaternion is not changed when we multiply it with $t^2+1$. So we set $P:=t^2+1+{\epsilon}(-a{\mathbf i}+bt{\mathbf j})$ and try to factorize. The norm polynomial is $(t^2+1)^2$, hence our only choice of an irreducible factor is $M=t^2+1$. The remainder of $P$ modulo $M$ is $R={\epsilon}(-a{\mathbf i}+bt{\mathbf j})$. But now something is wrong: even though $R$ has a right zero, namely $h=-\frac{b}{a}{\mathbf k}$, there is no common zero $R$ and $M$ except in the case $a=\pm b$. (If $a=\pm b$, then the ellipse is a circle, and we are not interested.) The argument we used in the quaternion case fails because $N(R)=0$.
There is a way out: instead of factorizing $P$, we can factorize $Q:=(t-{\mathbf i})P$. The displacement $[t-{\mathbf i}]$ fixes the origin, hence the displacement $[Q(t)]$ maps the origin to the point $p(t)$, just like the translation $[P(t)]$. The remainder of $Q$ modulo $M$ is ${\epsilon}(b-a)({\mathbf i}t-{\mathbf j})$, and this time we do have a common right zero of $M$ and $R$! Any dual quaternion of the form $-k-\epsilon(c{\mathbf j}+d{\mathbf j})$ is fine. For simplicity, we set $d=0$. Now we can factor $(t+k+\epsilon c{\mathbf j})$ from the right and proceed. The final result is $$Q = (t-{\mathbf k}+(a/2+b/2){\epsilon}{\mathbf j})(t-{\mathbf k}+(-a/2+b/2-c){\epsilon}{\mathbf j})(t+{\mathbf k}+\epsilon c{\mathbf j})=(t-h_1)(t-h_2)(t-h_3)$$ (we leave the remaining steps as an exercise – they are not problematic and give a unique result).
In order to construct a linkage with mobility one, we could use another factorization with a different linear factor on the left. But such a factorization does not exist: the norm polynomial of $Q$ is $(t^2+1)^3$, so there is no choice of choosing different factors of the norm polynomial. We need to mix a different quadratic irreducible polynomial into our soup.
Let $d\in{\mathbb R}$ and define $h_0:=2{\mathbf k}+d{\epsilon}{\mathbf j}$. The polynomial $(t-h_0)(t-h_1)$ has exactly two factorizations – one we know already, the second one is $(t-h_4)(t-h_5)$, for some $h_4,h_5\in{\mathbb D}{\mathbb H}$. Then the polynomial $(t-h_5)(t-h_2)$ also has exactly two factorizations, and we can define two more dual quaternions such that the second factorization is $(t-h_6)(t-h_7)$. Finally, let $h_8,h_9\in{\mathbb D}{\mathbb H}$ such that $(t-h_7)(t-h_3)=(t-h_8)(t-h_9)$. The different factorizations giving the same result correspond to paths in the directed graph $G$ in Figure \[fig:dia\] with equal starting and ending vertex.
$$\xymatrix{
{\tikz[baseline=(char.base)]{
\node[shape=circle,draw,inner sep=2pt] (char) {1};}} \ar^{t-h_1}[rr] & & {\tikz[baseline=(char.base)]{
\node[shape=circle,draw,inner sep=2pt] (char) {2};}} \ar^{t-h_2}[rr] & & {\tikz[baseline=(char.base)]{
\node[shape=circle,draw,inner sep=2pt] (char) {3};}} \ar^{t-h_3}[rr] && {\tikz[baseline=(char.base)]{
\node[shape=circle,draw,inner sep=2pt] (char) {4};}} \\
{\tikz[baseline=(char.base)]{
\node[shape=circle,draw,inner sep=2pt] (char) {5};}} \ar^{t-h_4}[rr] \ar^{t-h_0}[u] && {\tikz[baseline=(char.base)]{
\node[shape=circle,draw,inner sep=2pt] (char) {6};}} \ar^{t-h_6}[rr] \ar^{t-h_5}[u] & &
{\tikz[baseline=(char.base)]{
\node[shape=circle,draw,inner sep=2pt] (char) {7};}} \ar^{t-h_8}[rr] \ar^{t-h_7}[u] && {\tikz[baseline=(char.base)]{
\node[shape=circle,draw,inner sep=2pt] (char) {8};}} \ar^{t-h_9}[u]
}$$
The ellipse circle consists now of eight links corresponding to the eight vertices of $G$. Two links are connected by a joint if and only if the vertices are connected by an edge. The label of the edge – a linear polynmial in ${\mathbb D}{\mathbb H}$ – parametrizes the relative position of the target link with respect to the source link. As $t$ varies, the linear polynomials parametrize a revolution. Therefore, the two links are connected by an R-joint. Now we fix the link corresponding to vertex 4. Then the relative motion of the link corresponding to vertex 1 maps the origin to the point $p(t)$ on the ellipse.
Note that $b=0$ is allowed; in this case, the ellipse degenerates to a line segment traced twice, and we have constructed a linkage that draws this line.
Symmetry {#sec:sym}
========
The second construction by Dixon of a moving $K_{4,4}$ is symmetric. Indeed, symmetry may change the counting rules and can sometimes be the explanation of paradoxical mobility. We discuss here two cases in more detail: line symmetry and plane symmetry. Both cases appeared in Bricard’s families of moving octahedra in [@Bricard:97]. Schulze [@Schulze:10] was the first to describe paradoxical moving symmetric graphs systematically, in every dimension.
#### Line Symmetry. {#ss:linesym}
We assume that we have a graph $G=(V,E)$ such that $|E|=3|V|-6$, and an assignment $(\lambda)_{e\in E}$ of a positive real number for each edge. Generically, the configuration set, i.e., the set of all maps $V\to{\mathbb R}^3$ respecting edge lengths modulo ${\mathrm{SE}}_3$, is finite: we have $3|V|-6$ variables and $|E|$ equations. Let us now assume that we have a graph automorphism $\tau:V\to V$ that preserves the edge assignment. Assume also that $\tau$ has order 2, does not fix a vertex, and does not fix an edge – a priori, an edge could be fixed if $\tau$ permutes its two vertices. Then $|V|$ consists of $n:=\frac{|V|}{2}$ pairs of conjugated vertices, and $E$ consists of $3n-3$ pairs of conjugated edges. In order to construct line symmetric configurations, we fix a line $L\subset{\mathbb R}^3$; let $\sigma:{\mathbb R}^3\to{\mathbb R}^3$ be the rotation around $L$ by $\pi$. For any conjugated pair $(v,\tau(v))$ of vertices, we pick one point $p_v$ anywhere in ${\mathbb R}^3$; the second point is determined by $p_{\tau(v)}:=\sigma(p_v)$. The number of variables to specify all points is $3v$. There is also a two-dimensional subgroup of ${\mathrm{SE}}_3$ fixing $L$, generated by rotations around $L$ and translations into the direction of $L$. We use two of the variables to get a canonical representative. Hence the number of variables to specify an equivalence class of configurations is $3n-2$. The number of equations is equal to the number of pairs of conjugated edges, which is §$3n-3$, because conjugated edges always have the same length. Hence the expected mobility is one.
The smallest line symmetric moving graph is the 1-skeleton of an octahedron, with 6 vertices and 12 edges. The group of graph automorphisms is isomorphic to the Euclidean symmetry group of a regular octahedron, which has 48 elements. There is a unique automorphism of order 2 without fixed vertex and fixed edge, corresponding to the point reflection of the regular octahedron. The construction applies, and we get a moving line symmetric octahedron (see Figure \[fig:oct\] left side). Bricard [@Bricard] proved that there are three types of moving octahedra, and the line symmetric is one of the three.
![Left side: a flexible octahedron that is symmetric by a line reflection; right side: a flexible octahedron that is symmetric by a plane reflection. Corresponding edges are shown in the same color.[]{data-label="fig:oct"}](typeI.png "fig:"){width="5cm"} ![Left side: a flexible octahedron that is symmetric by a line reflection; right side: a flexible octahedron that is symmetric by a plane reflection. Corresponding edges are shown in the same color.[]{data-label="fig:oct"}](typeII.png "fig:"){width="5cm"}
More generally, we can take any centrally symmetric convex polyhedron $\Gamma$ with only triangular faces and choose as a graph $G=(V,E)$ its 1-skeleton. By Euler’s formula, the number of edges is $3|V|-6$. The point reflection acting on $\Gamma$ defines an automorphism of the graph which satisfies the required properties: order 2, no fixed vertex, no fixed edge. The construction applies, and we get, for instance, a line symmetric moving icosahedron with 12 vertices and 30 edges.
\[rem:sym\] Be careful: the point symmetry defines only the graph automorphism! It is geometrically different from the line symmetry in all configurations we allow. Point symmetric configurations do also exist, but only finitely many.
Another classical example is Bricard’s line symmetric 6R loop. Any 6R loop consists of 6 links, cyclically connected by revolute joints that allow rotations around an axes which is common to the two attached links; generically, a 6R loop is rigid. In the line symmetric case, we assume that the 18 invariant Denavit-Hartenberg parameters $d_0,\dots,s_5$ satisfy $$d_i=d_{i+3}, \alpha_i=\alpha_{i+3}, s_i=s_{i+3} \mbox{ for } i=0,1,2 ,$$ and we are only looking for configurations such that there exists a half turn mapping the $i$-th link to the $(i+3)$-rd link.
Recall that configurations can be found by solving the closure equation (see Section \[sec:rl\]): we attach an internal coordinate system to each link and parametrize the transformation $T_i$ from the $i$-th link to the $(i+1)$-th link (where the 6-th link is the 0-th link) by the $i$-th configuration parameter $\phi_i$. As mentioned above, $T_i(\phi_i)$ is the composition of the translation by a vector of length $d_i$ parallel to the $z$-axis, the rotation around the $z$-axis by the angle $\alpha_i$, the translation by a vector of length $s_i$ parallel to the $x$ axis, and a rotation around the $x$-axis determined by the $i$-th configuration parameter $\phi_i$. The configuration set is the set of solutions $(\phi_0,\dots,\phi_5)$ of the closure equation $$T_0(\phi_0)T_1(\phi_1)T_2(\phi_2)T_3(\phi_3)T_4(\phi_4)T_5(\phi_5) = e ,$$ where $e$ is the identity of the group ${\mathrm{SE}}_3$. The functions $T_0,\dots,T_5$ depend on the invariant Denavit-Hartenberg parameters, and as a consequence we have $T_0=T_3$, $T_1=T_4$, and $T_2=T_5$. Recall that the closure equation is a codimension 6 condition, because ${\mathrm{SE}}_3$ is a six-dimensional group, hence the CGK-formula estimates that there are only finitely many solutions.
Since we are only interested in line symmetric configurations, we assume $\phi_0=\phi_3$, $\phi_1=\phi_4$, and $\phi_2=\phi_5$. The closure equation reduces to $$\left(T_0(\phi_0)T_1(\phi_1)T_2(\phi_2)\right)^2 = e .$$ We ignore the solutions of $T_0(\phi_0)T_1(\phi_1)T_2(\phi_2)=e$ (these are at most finitely many). This means, we search for configuration parameters such that the transformation of the coordinate system of the 0-th link to the coordinate system of the 3rd link is a half turn. This is a codimension 2 condition: as we have already mentioned in Section \[sec:rl\], the set of involutions in ${\mathrm{SE}}_3$ is a 4-dimensional manifold. Hence there is a one-dimensional set of line symmetric configurations generically.
\[rem:ls\] Is there a good reason to explain the mobility of a line symmetric linkage by the closure equation, instead of just considering them as special cases of line symmetric linkages of graph type, as in Remark \[rem:red\]. Here is one: we may replace some of the revolute joints by other types of joints, like prismatic joints, as in hydraulic telescopes, or helical joints, as commonly seen in the form of screws. In both cases, such a joint allows a one-parameter subgroup of displacements of the connected links, and exactly the same proof of mobility is valid. On the other hand, a loop with helical joints cannot be considered as a linkage of graph type, because its closure equation is not even algebraic.
Yet another classical example, the line symmetric Stewart platform, will be explained in Section \[sec:pod\].
#### Plane Symmetry.
Plane reflections are involutions in the group $\mathrm{E}_3$ of isometries reversing the orientation. They are of course not direct isometries, but they still may be responsible for paradoxical mobility of various types of linkages, similar to half turns in the case of line symmetric linkages. Let us start with 6R loops. In a plane symmetric configuration of a 6R loop, there exists a plane reflection mapping link 0 to link 5, link 1 to link 4, and link 2 to link 3. The existence of a plane symmetric configuration has the following implications on the invariant Denavit-Hartenberg parameters: $$d_0=d_5, d_1=d_4, d_2=d_3, \alpha_0=\alpha_5, \alpha_1=\alpha_4, \alpha_2=\alpha_3, s_1=-s_0, s_2=-s_5, s_0=s_3=0 .$$ The relations between the functions in the closure equations are the following: $$RT_0(\phi_0)R=T_0(-\phi_0), RT_1(\phi_1)R=T_5(-\phi_1), RT_2(\phi_2)R=T_4(-\phi_2), RT_3(\phi_3)R=T_3(-\phi_3),$$ where $R$ is the reflection by the coordinate plane $\Pi$ spanned by the first and second axes. Instead of solving the closure equation, we find all quadruples $(\phi_0,\phi_1,\phi_2,\phi_3)$ such that $RXR=X$, where $X:=T_0(\phi_0)T_1(\phi_1)T_2(\phi_2)T_3(\phi_3)$. An element $X\in{\mathrm{SE}}_3$ fulfills the equation $RXR=X$ if and only if it is a rotation with an axis orthogonal to $\Pi$ or a translation by a vector in $\Pi$. These rotations and translations form a manifold of dimension 3 (isomorphic to ${\mathrm{SE}}_2$), hence the condition above is a codimension $6-3=3$ condition. In general, there is a one-dimensional set of solutions.
For every solution $(\phi_0,\phi_1,\phi_2,\phi_3)$ of $RXR=X$, the six-tuple $(2\phi_0,\phi_1,\phi_2,2\phi_3,\phi_2,\phi_1)$ is a solution of the closure equation: $$T_0(2\phi_0)T_1(\phi_1)T_2(\phi_2)T_3(2\phi_3)T_4(-\phi_2)T_5(-\phi_1) =T_0(\phi_0)XT_3(\phi_3)RT_2(-\phi_2)T_1(-\phi_1)R =$$ $$RT_0(-\phi_0)RXRT_3(-\phi_3)T_2(-\phi_2)T_1(-\phi_1)R = RT_0(-\phi_0)RXRX^{-1}T_0(\phi_0)R = e .$$ Hence we again get a mobile 6R loop, also known as “Bricard’s plane symmetric 6R linkage”.
\[rem:ps\] As in Remark \[rem:ls\], we may replace some of the revolute joints by prismatic or helical joints – see [@Baker:97]. Care has to be taken for the special role of the 0-th joint and the 3rd joint, because these two joints are supposed to be mapped to their own inverse by the plane symmetry. This is not possible at all for helical joints. Prismatic joints are fine, but the direction vector has to be perpendicular to the symmetry plane and not parallel to it.
For linkages of graph type, there is also a construction of plane symmetric linkages that are paradoxically mobile. We assume that we have a graph $(V,E)$ such that is generically rigid and satisfies $|E|=3|V|-6$, for instance the 1-skeleton of a convex polyhedron with triangular faces. Assume that we have a graph automorphism $\tau:V\to V$ of order 2 that fixes $2m$ vertices and $2m-2$ edges, for some $m\ge 1$. Choose a generic edge assignment that respects the involutive symmetry. Fix a plane $\Pi$ in ${\mathbb R}^3$, and let $R:{\mathbb R}^3\to{\mathbb R}^3$ be the reflection at $\Pi$. A configuration $(p_v)_{v\in V}$ is symmetric with respect to the plane $\Pi$ if and only if $R(p_v)=p_{\tau(v)}$ holds for all $v\in V$. The number of indeterminates is $3\frac{|V|-2m}{2}+4m-3=\frac{3}{2}|V|+m-3$: for each 2-orbit in $V$, the realization is determined by 3 indeterminates, and for each fixed point, we have two indeterminates because the point must lie in $\Pi$. The symmetry group of the plane has dimension 3, which reduces the number of indeterminates of equivalence classes by 3. The number of equations is $\frac{|E|-2m+2}{2}+2m-2=\frac{3|V|-2m-4}{2}+2m-2=\frac{3}{2}|V|+m-4$. Again, we obtain a paradoxically mobile graph.
So, how do we find graphs with an automorphism of order 2 fixing $2m$ vertices and $2m-2$ edges? Say, the graph is the 1-skeleton of a convex polyhedron $\Gamma$ with triangular faces. If $\Gamma$ is symmetric with respect to the half turn around a line passing through 2 vertices, then we get a involution with 2 fixed points and no fixed edge, so that $m=1$. This works, for example, for the octahedron – see Figure \[fig:oct\] right side – and for the icosahedron.
In the construction above, there are two geometric symmetries playing entirely different roles: The line symmetry of the convex polyhedron defines a graph automorphism of order 2 with the right properties; the plane symmetry defines a condition on the configurations that we consider. See also Remark \[rem:sym\].
Here is an example of a generically rigid graph with 12 vertices and 30 edges and with an automorphism of order 2 that fixes $4$ vertices and $2$ edges: take a 6R loop and construct a graph as in Remark \[rem:red\], by putting two vertices on each of the four rotation axes. In this case, the plane symmetric construction just reconstructs plane symmetric 6R loops, which we already did in another way.
Multipods and Group-Leg Duality {#sec:pod}
===============================
The Prix Vaillant 1904 asked for curves in the Lie group $\mathrm{SE}_3$ of direct isometries such that “many” points in ${\mathbb R}^3$ move on spheres. Connecting the moving points by sticks with the midpoints of these spheres, we obtain a [*multipod*]{}, also known as [*Stewart platform*]{}, which is a linkage consisting of a fixed base and a moving platform that are connected by [*legs*]{} of fixed length that are attached to platform and base by spherical joints (see Figure \[fig:planar\]). Flight simulators or other linkages that are supposed to make irregular motions are often manufactured as hexapods with additional prismatic joints at each leg that change its length; in this section, as already stated, the leg lengths remain constant. Each leg gives a codimension 1 condition on the displacement of the platform with respect to the base, hence the CGK-formula gives the estimate $6-n$ for the mobility an $n$-pod. Strictly speaking, each leg may be considered as a link that may also revolve around the line connecting its two anchor points, but we disregard this component of the motion. So, pentapods are generically mobile, and hexapods are generically rigid.
A displacement ${\mathbb R}^3\to{\mathbb R}^3$ of the platform relative to the base is given by an orthogonal matrix $M\in\mathrm{SO}_3$ with determinant $1$ and the image $y\in{\mathbb R}^3$ of the origin of the base. We set $x:=-M^ty=-M^{-1}y$ to be the preimage of the origin of the platform and $r:=\langle x,x\rangle = \langle y,y\rangle$, where $\langle \cdot, \cdot \rangle$ is the Euclidean scalar product. If we take coordinates $m_{11}, \dots, m_{33}$, $x_1, x_2, x_3$, $y_1, y_2, y_3$ and $r$, together with a homogenizing variable $h$, in ${\mathbb P}^{16}$, then a direct isometry defines a point in projective space satisfying $h \neq 0$ and $$\label{eq:compactification}
\begin{gathered}
M M^t \; = \; M^t M \; = \; h^2 \cdot \mathrm{id}_{{\mathbb R}^3}, \quad \mathrm{adj}(M) \; =hM^t , \\
M^t y + h x \; = \; 0, \quad M x + h y \; = \; 0 , \\
\langle x,x \rangle \; = \; \langle y,y \rangle \; = \; r h,
\end{gathered}$$ where $\mathrm{adj}(M)$ is the adjugate matrix. Recall that $A\cdot\mathrm{adj}(A)=\mathrm{adj}(A)\cdot A=\det(A)\cdot\mathrm{id}_{{\mathbb R}^3}$ for any $A\in{\mathbb R}^{3\times 3}$, therefore the above equations imply $\det(M)=h^3$. The equations define a variety $X$ of dimension 6 and degree 40 in ${\mathbb P}^{16}$, whose real points satisfying $h \neq 0$ are in one to one correspondence with the elements of ${\mathrm{SE}}_3$. We call it the [*group variety*]{}; its projective space ${\mathbb P}^{16}$ containing $X$ is called [*group space*]{}.
Mathematically, a leg is a triple $(a,b,d)$, where $a\in{\mathbb R}^3$ is a point of the base, $b\in{\mathbb R}^3$ is a point of the platform, and $d\in{\mathbb R}$ is a positive number, the length of the leg. We define the [*leg variety*]{} $Y$ as the cone over the Segre variety $\Sigma_{3,3} \cong {\mathbb P}^3\times{\mathbb P}^3$ in the projective space $\check{{\mathbb P}}^{16}$; recall that the Segre variety is a subvariety of a projective space of dimension 15 and degree ${3+3\choose 3}=20$, hence $Y$ has dimension $7$ and degree $20$. The values of projective coordinates of a leg $(a,b,d)$ are $u:=1$, $a_i$, $b_j$ and $z_{ij}:=a_ib_j$ for $i=1,2,3$, and the [*corrected leg length*]{} $l := \langle a, a \rangle + \langle b, b \rangle - d^2$. The projective space $\check{{\mathbb P}}^{16}$ containing $Y$ is called [*leg space*]{}.
The reason for this very specific choice of coordinates is the following. The algebraic condition $\langle Ma+y-b,Ma+y-b\rangle = d^2$ is bilinear in these coordinates: $$\begin{gathered}
\label{eq:bilinear_sphere_condition}
l h + u r - 2 \sum_{i=1}^3 a_ix_i - 2 \sum_{j=1}^3 b_jy_j
- 2 \sum_{i, j = 1}^{3} z_{ij} m_{ij} \; = \; 0.
\end{gathered}$$ Hence it defines a duality between group space and leg space. Every point in group space, in particular every group element, corresponds to a hyperplane in leg space; every point in leg space, in particular every leg, corresponds to a hyperplane in group space. More generally, to every $k$-plane in group space there is a dual $(15-k)$-plane in leg space, for $k=0,\dots,15$.
The duality has various implications for multipods, whether they are paradoxical or not. To start with, choose 6 generic legs. They span a generic 5-plane in leg space. The dual 10-plane in group space is also generic and, since it has codimension 6, intersects $X$ in $\deg(X)=40$ points (real or complex). Hence a generic hexapod has 40 configurations, possibly complex.
Now, we choose 5 generic legs. They span a generic 4-plane in leg space, dual to a generic 11-plane in group space, which intersects $X$ in a curve $C$ of degree 40: the configuration curve of a generic pentapod. We can compute its genus. We first compute the Hilbert series of $X$ from a generating set of its ideal: $H(X)(t)=\frac{1+10t+18t^2+10t^3+t^4}{(1-t)^7}$. Because $C$ is a codimension 5 subvariety of $X$ defined by 5 linear forms, we may compute the Hilbert series of $C$ from the Hilbert series of $X$: $$H_C(t) = H_X(t)(1-t)^5 = \frac{1+10t+18t^2+10t^3+t^4}{(1-t)^2} = 1+12t+41t^2+80t^3+120t^4+\dots .$$ This implies that $C$ is a curve of genus 41, and its embedding in ${\mathbb P}^{11}$ is half canonical.
#### The Bricard-Borel Infinity-Pod.
Here is the infinity-pod that has won the Prix Vallaint to Borel [@Borel:08] and Bricard [@Bricard:07]. We intersect $X$ with the 3-space defined by $$r+\beta h-2\alpha m_{11} = m_{11}-m_{22} = m_{12}+m_{21} = m_{33}-h = x_{3}+y_{3} =$$ $$m_{13} = m_{23} = m_{31} = m_{32} = x_1 = x_2 = y_1 = y_3 = 0,$$ where $\alpha,\beta\in{\mathbb R}$ are real parameters such that $\alpha\ne 0$. The result is a quartic curve defined by the equations $$m_{11}^2+m_{12}^2-h^2=x_3^2-2\alpha m_{11}h+\beta h^2 = 0$$ and by the linear equations above. It parametrizes a motion $C$ contained in the two-dimensional stabilizer of the third axes $L$, generated by rotations around $L$ and translations in the direction of $L$. The dual 12-plane in leg space is defined by $$z_{12}-z_{21} = z_{11}+z_{22}-\alpha u = a_3-b_3 = l-2z_{33}-\beta u = 0 .$$ A leg $(a,b,d)$ in the intersection with $Y$ if and only if $$a_1b_2-a_2b_1 = a_1b_1+a_2b_2-\alpha = a_3-b_3 = a_1^2+a_2^2+b_1^2+b_2^2-d^2-\beta = 0 .$$ For any point $(a_1,a_2,a_3)$ in the base such that $(a_1,a_2)\ne (0,0)$, there is a unique point $(b_1,b_2,b_3)$ in the platform and a length such that the motion $C$ keeps the distance of base and platform point equal to $d$. To get the platform point corresponding to a given base point $(a_1,a_2,a_3)$, we invert its projection $(a_1,a_2)$ on the circle with radius $\sqrt{|\alpha|}$ and keep the third coordinate; if $\alpha<0$, then we also have to rotate the projection by an angle of $\pi$.
In the degenerate case $\alpha=\beta=0$, one of the equations of the quartic curve is a perfect square, and the reduce equations $m_{11}^2+m_{12}^2-h^2=x_3$ define a conic in a 2-space. In leg space, we have one less linear equation: $z_{12}-z_{21}=z_{11}+z_{22}=l-2z_{33}=0$, or equivalently $$a_1b_2-a_2b_1 = a_1b_1+a_2b_2 = a_1^2+a_2^2+b_1^2+b_2^2-d^2 = 0 .$$ Here we get a four-dimensional set of possible legs with two components, namely the set of legs where the platform point or the base point lies on the $z$-axis. The motion is just a revolution around the $z$-axis.
#### Planar Multipods.
We consider now the linear subspace $L_p\subset\check{{\mathbb P}}^{16}$ of dimension 9 in the leg space defined by the equations $$a_3=b_3=z_{13}=z_{23}=z_{31}=z_{32} = z_{33}= 0 .$$ Its intersection $Y_p$ with the leg variety consists of all legs such that the two anchor points lie on a fixed plane. The variety $Y_p$ is the Segre variety $\Sigma_{2,2} \cong {\mathbb P}^2\times{\mathbb P}^2$; let us call its elements informally [*planar legs*]{}. The degree of $Y_p$ is ${2+2\choose 2}=6$.
A multipod such that all its base points are coplanar and all its platform points are coplanar is called a [*planar multipod*]{} (see Figure \[fig:planar\]). To obtain the configuration of a planar multipod, one has to intersect the dual space of the linear span of all legs with the group variety $X$. The linear span of the legs is contained in $L_p$, hence the dual space of the linear span contains the dual space $L_p^\bot$. This linear space does not intersect the group variety, otherwise we would have a displacement that preserves the length of all legs in $Y_p$, which is impossible. What we can say is that the projection ${\mathbb P}^{16}\dashrightarrow {\mathbb P}^9$ with center $L_p^\bot$ projects the group variety to a subvariety $X_p\in{\mathbb P}^9$ of dimension 6 and degree 20 by a map that is generically 2:1. Hence the configuration of a planar multipod come in pairs: for every configuration, there is a conjugated configuration. It can be obtained by an outer automorphism of ${\mathrm{SE}}_3$, namely the conjugation by the reflection with respect to the plane containing the anchor points.
[hexapod]{}
(-4,0)[$a_1$]{} (50,-4)[$a_2$]{} (100,2)[$a_3$]{} (78,19)[$a_4$]{} (34,24)[$a_5$]{} (20,13)[$a_6$]{} (-6,73)[$b_1$]{} (50,56)[$b_2$]{} (70,69.5)[$b_3$]{} (98,80)[$b_4$]{} (45,85)[$b_5$]{} (12,86)[$b_6$]{}
It is surprisingly easy to construct paradoxically moving planar hexapods. Here is the reason.
Let $y_1,\dots,y_5\in Y_p$ be five generic planar legs. Then there exists a planar leg $y_6\in Y_p$ such that the configuration space of the pentapod defined by $(y_1,y_2,y_3,y_4,y_5)$ is equal to the configuration space of the hexapod defined by $(y_1,y_2,y_3,y_4,y_5,y_6)$.
Let $V\subset \check{{\mathbb P}}^{16}$ be the linear span of $y_1,\dots,y_5$. Its dimension is $4$. The dimension of $Y_p$ is $5$. Both $V$ and $Y_p$ are contained in $L_p\cong{\mathbb P}^9$, hence the intersection $Y_p\cap V$ is finite. Its cardinality is equal to the degree of $Y_p$, which is 6. We know already 5 points; we choose $y_6$ to be the 6-th.
For both, the pentapod and the hexapod, the configuration set of the pentapod is the intersection of the group variety $X$ with the dual space $V^\bot$. The linear condition imposed by the 6-leg does not impose an independent condition because it lies in the linear span of the other five.
#### Line-symmetric Multipods.
Another class of paradoxically moving hexapods is the class of line symmetric hexapods. They can be obtained as special cases of line symmetric moving graphs (see Section \[ss:linesym\]). The graph consists of two octahedra $G_1,G_2$ together with six edges each joining one point of $G_1$ to one point of $G_2$, so that these six edges provide a graph symmetry between $G_1$ and $G_2$. The automorphism $\tau$ of the whole graph $G$ maps each vertex $v_1$ of $G_1$ to the vertex in $G_2$ connected with the unique vertex in $G_1$ that is not connected with $v_1$ (see Figure \[fig:graphaut\]). This graph automorphism does not fix any vertex or any edge. We fix a line $L$ of symmetry and embed $G$ so that the half turn around $L$ maps each vertex $v$ to the image of $\tau(v)$, generically with respect to this condition. By the count in Section \[ss:linesym\], the configurations are solutions of an algebraic system in 16 unknowns and 15 equations, implying mobility.
(a1) at (16,0) [1]{}; (a2) at (8,12) [2]{}; (a3) at (-8,12) [3]{}; (a4) at (-16,0) [4]{}; (a5) at (-8,-12) [5]{}; (a6) at (8,-12) [6]{};
(b1) at (4,0) [4]{}; (b2) at (2,3) [5]{}; (b3) at (-2,3) [6]{}; (b4) at (-4,0) [1]{}; (b5) at (-2,-3) [2]{}; (b6) at (2,-3) [3]{};
(a1)edge(a2) (a2)edge(a3) (a3)edge(a4) (a4)edge(a5) (a5)edge(a6) (a6)edge(a1) (a1)edge(a3) (a2)edge(a4) (a3)edge(a5) (a4)edge(a6) (a5)edge(a1) (a6)edge(a2) (b1)edge(b2) (b2)edge(b3) (b3)edge(b4) (b4)edge(b5) (b5)edge(b6) (b6)edge(b1) (b1)edge(b3) (b2)edge(b4) (b3)edge(b5) (b4)edge(b6) (b5)edge(b1) (b6)edge(b2) (b1)edge(a1) (b2)edge(a2) (b3)edge(a3) (b4)edge(a4) (b5)edge(a5) (b6)edge(a6);
It pays off to analyze the situation again by group-leg duality, following an analysis from [@Borel:08]. Let $L_i\subset{\mathbb P}^{16}$ be the linear subspace in group space defined by the linear equations $M=M^t$ and $x=y$; it intersects $X$ in the subset $X_i$ of all displacement of order 2 or 1. Note that the order 2 elements in ${\mathrm{SE}}_3$ are exactly the rotations around lines by an angle of $\pi$. These are six equations, hence $\dim(L_i)=10$. The dual subspace $L_i^\bot$ in leg space has dimension 5 and is defined by the equations $l=u=z_ii=a_i+b_j=z_{ij}=0$ for $i,j=1,\dots,3$, $i\ne j$. We have a situation that mirror the planar hexapod case: the subspace $L_i^\bot$ does not intersect the leg variety, otherwise there would be a leg which does not change length in all involutions. But the projection $\check{{\mathbb P}}^{16}\dashrightarrow \check{{\mathbb P}}^{10}$ with center $L_i^\bot$ projects the leg variety $Y$ to a subvariety $Y_i\in\check{{\mathbb P}}^{10}$ of dimension 7 and degree 10 by a map that is generically 2:1. Hence the legs of a multipod with involutive displacements come in pairs: if $(a,b,d)$ is a leg, then $(b,a,d)$ is also a leg. This can also be shown directly: if $\sigma\in{\mathrm{SE}}_3$ has order 2, then $$||\sigma(a)-b|| = ||\sigma^2(a)-\sigma(b)|| = ||\sigma(b)-a|| .$$ Group-leg duality induces a duality between the projective subspace $L_i$ of dimension 10 that contains $X_i$ and the projective image space $\check{{\mathbb P}}^{10}$ that contains $Y_i$. Let us call the elements in $Y_i$ [*twin pairs of legs*]{}; each such pair of legs is constituted by a leg $(a,b,d)$ and by its conjugated leg $(b,a,d)$. Generically, three twin pairs in $Y_p$ correspond to three hypersurfaces in $L_i$. Since $\dim(X_i)=4$, the intersection of these three hypersurfaces and $X_p$ is a curve. So, we have explained again the paradoxical mobility.
But there is more. We have not just constructed a moving hexapod, we have even constructed, at the same time, a moving icosapod! Here is the precise statement.
\[thm:bls\] Let $p_1,p_2,p_3$ be three generic twin pairs of legs. Let $C\subset X_s$ be the configuration curve of the hexapod defined by all six legs. Then there exist seven additional twin pairs, maybe complex, such that $C$ is the set of all order 2 displacements compatible with all 20 legs.
The three twin pairs span a generic 2-plane in $V\subset\check{{\mathbb P}}^{10}$. The subvariety $Y_i\subset\check{{\mathbb P}}^{10}$ has dimension 7, hence $V$ and $Y$ intersect in $\deg(Y_i)=10$ points. Three of them correspond to $p_1,p_2,p_3$, and the remaining seven are the additional pairs we require. The linear span of all 10 points is equal to the linear span of $p_1,p_2,p_3$, namely $V$, hence the conditions for displacements do not change.
In [@Schicho:16d], it is shown that there exist examples where all twenty legs are real. The proof is based on a result on quartic spectahedra in [@DegtyarevItenberg:11; @Ottem:15].
Compactification {#sec:bond}
================
In enumerative algebraic geometry, for instance for the problem of counting rational curves on a projective variety, compactifications of moduli spaces are known as a powerful tool. Here, we compactify the algebraic varieties in which the configuration spaces are naturally embedded: products of subgroups of ${\mathrm{SE}}_3$ in the case of linkages with revolute joints, ${\mathrm{SE}}_3$ itself in the case of multipods, and products of the plane in the case of moving graphs.
Moving Graphs
-------------
The main theorem in [@Schicho:19a] states that for a given graph, the existence of a flexible labeling is equivalent to the existence of a NAC coloring. The construction of a flexible linkage from a given NAC-coloring was already explained in Section \[sec:over\]. For a construction proving the other implication, we need a compactification.
Let $(V,E,\lambda)$ be a graph with an edge assigment. We would like to projectivize in order to compactify; for this purpose, it is convenient to slightly change the notion of a configuration slightly. A [*homogeneous configuration*]{} is an assignment of vertices by points in ${\mathbb R}^2$ such that for any two edges $e=(i,j)$, $f=(k,l)$, the equality $$\lambda_e ||p_k-p_l||^2 = \lambda_f ||p_i-p_j||^2$$ holds. For each vertex $i\in V$ with assigned point $p_k$, we write $p_k=(x_k,y_k)$ and $z_k:=x_k+{\mathrm i}y_k$, $w_k:=x_k-{\mathrm i}y_k$. In other words, the complex numbers $z_1,\dots,z_{|V|}$ represent the vertices in the Gaussian plane of complex numbers. In order to normalize, we require $p_1=(0,0)$.
The homogeneous configuration $p$ defines a point in ${\mathbb P}^{|V|-2}\times{\mathbb P}^{|V|-2}$ as follows: its first component has projective coordinates $(z_2:\dots:z_{|V|})$, and its second component has coordinates $(w_2:\dots:w_{|V|})$. The equality above reads $$\label{eq:cf}
\lambda_e (z_k-z_l)(w_k-w_l) - \lambda_f (z_i-z_j)(w_i-w_j)$$ in these projective coordinates. This is a bihomogeneous equation of bidegree $(1,1)$. The set of all solutions of (\[eq:cf\]) is a projective subvariety of ${\mathbb P}^{|V|-2}\times{\mathbb P}^{|V|-2}$, the [*configuration variety*]{} of $(V,E,\lambda)$. Equivalent homogeneous configurations define the same point in the configuration variety: since we fixed $p_1=(0,0)$, equivalent configurations are related by a rotation or a scaling; but such a transformation just multiplies all $z$-coordinates by a complex nonzero constant and all $w$-coordinates by a different complex nonzero constant, hence does not change the points in ${\mathbb P}^{|V|-2}$.
A point $\alpha\in{\mathbb P}^{|V|-2}\times{\mathbb P}^{|V|-2}$ corresponds to a homogeneous configuration if and only if it fulfills two extra conditions. First, the conjugate has to coincide with the flip of the first and second component; if this condition fails, then some of the corresponding points in the plane have nonreal coordinates. Second, for some edge $e=(i,j)$, we have $(z_i-z_j)(w_i-w_j)\ne 0$. By (\[eq:cf\]), the choice of the edge has no influence on the validity of this extra condition.
The boundary of the configuration set is defined as the set of points in the configuration variety that fail to satisfy the two extra conditions. In particular, for some edge $e=(i,j)$, or equivalently for all edges, we have $(z_i-z_j)(w_i-w_j)=0$. For each point $\beta$ in the boundary, we define a coloring of the edges of the graph in the following way: the edge $(i,j)$ is colored red if $z_i-z_j$ vanishes at $\beta$, and blue otherwise.
\[lem:NAC\] For any point $\beta$ in the boundary of the configuration variety, the coloring defined by it is a NAC-coloring.
Assume, indirectly, that all edges are red. Then the first projection of $\beta$ to ${\mathbb P}^{|V|-2}$ has only zero coordinates, which is impossible.
Assume, indirectly, that all edges are blue. For any edge $(i,j)$, we have $(z_i-z_j)(w_i-w_j)=0$ and $z_i-z_j\ne 0$. It follows that the second projection of $\beta$ to ${\mathbb P}^{|V|-2}$ has only zero coordinates, which is impossible.
Assume, indirectly, that $(i_1,\dots,i_k,i_1)$ is cycle such that $(i_r,i_{r+1})$ is red for all $r=1,\dots,k-1$, and $(i_k,i_1)$ is blue. Then $z_{i_1}=\dots=z_{i_k}$ and $z_{i_k}\ne z_{i_1}$, which is impossible.
Assume, indirectly, that $(i_1,\dots,i_k,i_1)$ is cycle such that $(i_r,i_{r+1})$ is blue for all $r=1,\dots,k-1$, and $(i_k,i_1)$ is red. Then $w_{i_1}=\dots=w_{i_k}$, hence $w_{i_k}=w_{i_1}$. In addition, we also have $z_{i_k}=z_{i_1}$ as $(i_k,i_1)$ is red. Therefore the form $(z_{i_1}-z_{i_k})(w_{i_1}-w_{i_k})$ vanishes with order $m\ge 2$ at $\beta$. The order of this form is the same for every edge, and because $(i_r,i_{r+1})$ is blue, the forms $z_r-z_{r+1}$ have order zero for $r=1,\dots,k-1$. Hence the order of the forms $w_r-w_{r+1}$ is at least $m$, for all $r$. Then the form $w_{i_1}-w_{i_k}$ vanishes with order at least $m$, and this is a contradiction.
\[thm:NAC\] A $(V,E)$ has a flexible labeling $\lambda$ if and only if it has a NAC-coloring.
If $(V,E,\lambda)$ is flexible, then its configuration set is a projective variety $K$ of positive degree in ${\mathbb P}^{|V|-2}\times{\mathbb P}^{|V|-2}$. For any edge $(i,j)\in E$, the form $(z_i-z_j)(w_i-w_j)$ has to vanish somewhere in $K$. Therefore, $K$ meets the boundary. By Lemma \[lem:NAC\], it follows that $(V,E)$ has a NAC-coloring.
Conversely, assume that we have a NAC-coloring of the edges. Then we make the graph moving by a construction already given in Section \[sec:gk\]: the red edges always keep their direction and move by translations only, while the blue edges rotate with uniform speed.
For example, the graph in Figure \[fig:nonac\] has no NAC-coloring and therefore never moves for any labeling $\lambda$.
\(a) at (-0.5,-0.75) ; (b) at (0.5,0.5) ; (c) at (1.5,0.5) ; (d) at (2.5,-0.75) ; (e) at (0.5,1.5) ; (f) at (1.5,1.5) ; (g) at (1,-0.25) ; (a)edge(d) (a)edge(e) (b)edge(c) (b)edge(e) (c)edge(d) (c)edge(e) (c)edge(f) (d)edge(f) (e)edge(f) (g)edge(a) (g)edge(b) (g)edge(d);
A weakness of Theorem \[thm:NAC\] is that its constructive part – the construction of flexible labelings – produces only a particular type of motions that we my call “uniform speed motions”. Also, these motions sometimes map different non-adjacent vertices to the same point in the plane. For example, in the case of the complete bipartite graph $K_{3,3}$, all uniform speed motions map at least two pairs of vertices to the same point in the plane, and the moving graph looks like a moving parallelogram. Deciding if a given graph has labeling with a generically injective motion is much harder than deciding the existence of a flexible labeling; see [@Schicho:20b].
Revolute Loops
--------------
The complete classification of mobile 4R loops was given by Delassus (see Section \[sec:rl\]). The complete classification of mobile 5R loops was given in [@Karger:98] with the help of computer algebra. For 6R loops, the classification is still open; the difficult part is to come up with necessary conditions for mobility. In this subsection, we explain a method for deriving necessary criteria, which applies to $n$R loops for $n=4,5,6$.
We start with the closure equation expressed in algebraic way. Let $d_1,\dots,d_n$ (normal distances), $s_1,\dots,s_n$ (offsets), and $w_1,\dots,w_n$ (cotangents of half angles) the invariant Denavit-Hartenberg parameters. For $r=0,\dots,n-1$, the dual quaternion $g_r := (1-s_r{\epsilon}{\mathbf i})(w_r-{\mathbf k})(1-d_i{\mathbf k})$ is the displacement that transforms the internal coordinate system of link $r$ to the internal coordinate system of link $r+1$ (modulo $n$), if the configuration parameter is zero. The closure equation is an equation in the variables $t_0,\dots,t_{n-1}$, which denote the cotangents of the half configuration angles: the dual quaternion $$x(t_0,\dots,x_{n-1}) := (t_0-{\mathbf i})g_0(t_1-{\mathbf i})g_1\cdots (t_{n-1}-{\mathbf i})g_{n-1}$$ is a multiple of $1$, hence 7 of its 8 coefficients are zero. The variables $t_0,\dots,t_{n-1}$ may also assume the value $\infty$; in this case, the corresponding factor $(t_r-{\mathbf i})$ is replaced by the scalar 1, or is simply omitted. In this section, we will avoid this technicality.
We focus on solutions on the boundary, but this time we do not consider $t_r=\infty$ as boundary. Instead, we define the boundary of $({\mathbb P}^1)^n$ as the set of $n$-tuples $(t_0,\dots,t_{n-1})$ such that $t_r^2+1=0$ for at least one $r$. Indeed, if we remove the boundary, then we get a group variety isomorphic to $({\mathrm{SO}}_2)^n$, with an isomorphism respecting real structures. The statement that $t_r^2+1=0$ for at least one $r$ is equivalent to the statement $N(x(t_0,\dots,t_n))=0$, by the multiplicativity of the norm. Boundary solutions can never be real, at least one of the variable must be equal to $\pm{\mathrm i}$.
[**Note:**]{} Throughout this paper, we use ${\mathbf i}$ for the first quaternion unit in ${\mathbb H}$, ${\mathrm i}$ for the imaginary unit in ${\mathbb C}$, and $i$ for a running integer. In this sections, both ${\mathbf i}$ and ${\mathrm i}$ will appear, sometimes in the same expression; but we will try to avoid using $i$ for an integer.
Unfortunately, the closure equation often has many solutions that are not of interest. But we can obtain more equations by cyclic permutation of its factors, or by using quaternion conjugation to bring some factors to the other side, as in $$\lambda (t_0-{\mathbf i})g_0= \nu \overline{g_{n-1}}(t_{n-1}+{\mathbf i})\overline{g_{n-2}}\cdots \overline{g_1}(t_1+{\mathbf i})$$ for some scalars $\lambda,\nu$ that are not both equal to zero. This condition can be expressed by polynomial equations, namely the $2$-minors of the $2\times 8$ matrix whose rows are the coordinates of $(t_0-{\mathbf i})g_0$ and of $\overline{g_{n-1}}(t_{n-1}+{\mathbf i})\overline{g_{n-2}}\cdots \overline{g_1}(t_1+{\mathbf i})$. After having added all these reformulation of the closure equations to our system of equations, we look for solutions on the boundary. These are called [*bonds*]{}.
It is an easy exercise to prove that at least two of $t_0,\dots,t_{n-1}$ must be $\pm{\mathrm i}$. [*Hint:*]{} use a formulation of the closure equation with factors on both sides, and then take the norm on both sides. There are many examples with exactly two of $t_0,\dots,t_{n-1}$ being $\pm{\mathrm i}$. If, say, $t_1^2+1=t_k^2+1=0$ for some $k<n$, and $t_i^2+1\ne 0$ for $i\ne 1,k$, then we say that the first joint and the $k$-th joint are [*entangled*]{} in the respective bond. We can then prove the following equations: $$\label{eq:bond}
\begin{aligned}
(t_1-{\mathbf i})g_1(t_2-{\mathbf i})g_2\cdots (t_k-{\mathbf i}) = 0, \\
(t_k-{\mathbf i})g_k(t_{k+1}-{\mathbf i})g_{k+1}\cdots (t_0-{\mathbf i})g_n(t_1-{\mathbf i}) = 0.
\end{aligned}$$ If the number of coordinates $t_r$ with $t_r^2+1=0$ is bigger than two, then Equation \[eq:bond\] also holds form some $k$, up to cyclic permutation by [@hss2 Lemma 2 and Theorem 3].
Equation \[eq:bond\] together with $t_1^2+1=t_k^2+1=0$ is quite restrictive and often has implications on the invariant parameters that are hard-coded in $g_0,\dots,g_{n-1}$. The case $k=2$ is easy to analyze: assume $$({\mathrm i}-{\mathbf i})g_1({\mathrm i}-{\mathbf i}) = 0 .$$ Then it follows that $w_1=d_1=0$; geometrically this means that the first two rotation axes are equal, except that they have opposite orientation in the closure equation. We may exclude this degenerate case, and then we always have $k>2$ (and modulo $n$, this also excludes $k=0$).
If $k=3$, then we get the equation $$({\mathrm i}-{\mathbf i})g_1(t_2-{\mathbf i})g_2({\mathrm i}-{\mathbf i}) = 0,$$ up to orientation of the first and/or third axis. This is a system of inhomogeneous linear equations for $t_2$. It has a solution in three cases: either the three axes are parallel, or the three axes are concurrent, or the equations $$s_2=0, \frac{d_1}{\sin(\alpha_1)}=\frac{d_2}{\sin(\alpha_2)}$$ are true. This should be compared with Bennett’s condition for a 4R loop to be mobile in Section \[sec:rl\]: it is exactly the condition that has to be fulfilled for three axis that is true if and only if there exists a 4th line that supplements the three lines to a mobile 4R loop. The “Benett condition for three lines” mysteriously appears in Dietmaier’s collection [@Dietmaier:95] of known families of 5R loops and 6R loops. The bond equation explains why: in a mobile linkage, bonds have to be present, and for each bond there must be two non-neighboring joints entangled in a bond. In a 5R loop, we then have $k=3$ up to a cyclic permutation. In a 6R loop, we have either $k=3$ – entanglement of diagonal joints –, or $k=4$, entanglement of opposite joints. Many known families have some bond that entangles diagonal joints.
The analysis of the case $k=4$ is more involved; however, it is necessary in order to explain mobility of 6R linkages that have no three consecutive axis fulfilling the Bennett condition for three lines. Assume that $n=6$, and we have a bond $\vec{t}$ that entangles the first and the fourth joint. Without loss of generality, we may assume $t_1=t_4={\mathrm i}$. Then we obtain the equations $$\begin{aligned}
({\mathrm i}-{\mathbf i})g_1(t_2-{\mathbf i})g_2(t_3-{\mathbf i})g_3({\mathrm i}-{\mathbf i}) = 0, \\
({\mathrm i}-{\mathbf i})g_4(t_5-{\mathbf i})g_5(t_0-{\mathbf i})g_0({\mathrm i}-{\mathbf i}) = 0.
\end{aligned}$$ Excluding some degenerate cases (4 parallel lines, or 4 lines meeting in a point), the first equation allows two solutions for $(t_2,t_3)$, while the second equation allows two solutions for $(t_0,t_5)$. These partial solutions are not independent. They have to satisfy another reformulation of the closure equation: $$\lambda({\mathrm i}-{\mathbf i})g_1(t_2-{\mathbf i})g_2(t_3-{\mathbf i})g_3 = \nu\overline{g_0}(t_0+{\mathbf i})\overline{g_5}(t_5+{\mathbf i})\overline{g_4}({\mathrm i}+{\mathbf i}) ,$$ for some complex numbers $\lambda,\nu$ that are not both equal to zero. By resultants, we can eliminate the variables $t_0,t_2,t_3,t_5$ and obtain an equivalent formulation without these variables: the two quadratic polynomials $$Q_1^+(x) = \left(x+\frac{b_{3}c_3-b_{1}c_1}{2}-\frac{s_{1}}{2}{\mathrm i}\right)^2 +$$ $$\frac{{\mathrm i}}{2}\left(b_1 s_{2}+b_{3} s_{3}+s_{2} b_{3} c_{2}+s_{3} b_1 c_{2}\right) -$$ $$\frac{b_1 b_{3} c_{2}-s_{2} s_{3} c_{2}}{2}
+ \frac{s_{2}^2+s_{3}^2-b_1^2+b_{2}^2-b_{3}^2-b_{2}^2 c_{2}^2}{4},$$ $$Q_4^+(x) = \left(x+\frac{b_0c_0-b_{4}c_4}{2}-\frac{s_{4}}{2}{\mathrm i}\right)^2 +$$ $$\frac{{\mathrm i}}{2}\left(b_4 s_{5}+b_0 s_0+s_{5} b_0 c_{5}+s_0 b_4 c_{5}\right) -$$ $$\frac{b_4 b_0 c_{5}-s_{5} s_0 c_{5}}{2}
+ \frac{s_{5}^2+s_0^2-b_4^2+b_{5}^2-b_0^2-b_{5}^2 c_{5}^2}{4}$$ in ${\mathbb C}[x]$ have a common zero; here, $c_i:=\cos(\alpha_i)$ and $b_i:=\frac{d_i}{\sin(\alpha_i)}$ for $i=1,\dots,6$. For the details this elimination of variables, we refer to [@Schicho:15e].
If one of the coordinates of $t_1$ and $t_4$, or both, are equal to $-{\mathrm i}$, then there are two quadratic polynomials that are similarily defined, having a common zero. In total, the number of bonds entangling the first joint and fourth joint is even (because bonds always appear in complex conjugate pairs) and at most 8. It is equal to 8 if and only if the two polynomials are equal in each of the four pairs of quadratic polynomials.
Suppose that we have the maximal number of 8 bonds entangling opposite axes, for all three pairs of opposite axes. This assumption leads to a system of algebraic equations in the invariant Denavit/Hartenberg parameters (18 variables). Using computer algebra, we can compute the solution set (see [@Schicho:15e]). It turns out that there are two components $F_1$ and $F_2$, of dimension 6 and 7, respectively. Both are families of mobile 6R loops that have not been known before bonds were used in kinematics. But the family $F_1$ (the one of dimension 6) has a 5-dimensional subfamily which is classical: Bricard’s orthogonal 6R loops, characterized by the vanishing of $c_0,\dots,c_5$ (i.e., all angles are right angles) and $s_0,\dots,s_5$ (i.e., all offsets are zero), and the single equation $b_0^2-b_1^2+b_2^2-b_3^2+b_4^2-b_5^2=0$.
Multipods
---------
The two varieties that play a role in the analysis of multipods, namely the group variety $X\in{\mathbb P}^{16}$ and the leg variety $Y\in\check{P}^{16}$, both come with a natural definition of a boundary: the boundary of $X$ is defined by the linear equation $h=0$ and the boundary of $Y$ is defined by the linear equation $u=0$, with the variable setting as in Section \[sec:pod\]. The group variety is more interesting, because the configuration set of a mobile multipod – defined as the intersection of $X$ with hyperplanes dual to the legs – will always intersect the boundary $h=0$. The leg set of a mobile multipod, on the other hand, might be disjoint from the boundary $u=0$.
Let us have a closer look at the boundary $B:=X\cap H$, where $H$ is the hyperplane $H:h=0$. We refer to [@Schicho:14c] for the calculation; here we report on only the facts we will use later. First, $B$ is a variety of dimension 5 and degree 20. The variety $X$ – which has degree 20 – and the hyperplane $H$ intersect tangentially along $B$, with intersection multiplicity 2. The boundary $B$ has a natural decomposition into five locally closed subsets, which we denote by $Z_i$, $Z_b$, $Z_s$, $Z_c$, and $Z_v$. The stratum $Z_i$ has dimension 5 and consists of all points in $B$ which are smooth points of $X$ such that at least one of the $m_{ij}$-coordinates is not zero. The stratum $Z_i$ has dimension 4 and consists of all points in $B$ which are singular points of $X$ such that at least one of the $m_{ij}$-coordinates is not zero. The stratum $Z_s$ has dimension 3 and consists of all remaining boundary points such that one of the coordinates $x_1,x_2,x_3$ is not zero and one of the coordinates $y_1,y_2,y_3$ is not zero. The stratum $Z_c$ has dimension 2, and here one of the previous three triples of coordinates has values all zero. Finally, the stratum $Z_v$ consists of a single point: all coordinates except $r$ are zero. It is the only point on $B$ defined over the reals, all other boundary points are complex.
For a multipod given by a set of legs, we have a set of hyperplanes in ${\mathbb P}^{16}$ dual to the legs. We now define the set of bonds of the multipod as the set of intersection points of all these hyperplanes with the boundary $B$. The main point of the analysis of the boundary is that the presence of bonds in particular strata has geometric implications for the geometry of the legs. Let us show this for the stratum $Z_s$. Here, the projections $x:=(x_1:x_2:x_3)$ and $y:=(y_1:y_2:y_3)$ both exist, because there is at least one in both triples of coordinates that does not vanish. From the equations, it is easy to derive that both $x$ and $y$ have to lie on the absolute conic $x_1^2+x_2^2+x_3^2=0$, which clearly has no real points.
\[thm:simbond\] Let $\{ (a_l,b_l,d_l) \mid l\in L\}$ be the leg set of a multipod, where $L$ is an index set parametrizing the legs. Assume that this multipod has a bond in $Z_s$. Then there exist orthogonal projections $p_a:{\mathbb R}^3\to{\mathbb R}^2$ and $p_b:{\mathbb R}^3\to{\mathbb R}^2$ and a similarity transformation $s:{\mathbb R}^2\to{\mathbb R}^2$ such that $s(p(a_l))=p(b_l)$ for all $l\in L$.
The variety $X$ has an automorphism group of dimension 12, by left and right multiplication of group elements. The statement of the theorem is invariant under these automorphism. We use suitable automorphisms to transform the bond in $Z_s$ to a point with coordinates $(x_1,x_2,x_3)=(1,{\mathrm i},0)$, $(y_1,y_2,y_3)=(\lambda,\lambda{\mathrm i},0)$ and all remaining coordinates zero; it is important that this transformation can be achieved by real transformations (see [@Schicho:14c] for the calculation). The corresponding hyperplane in leg space has equation $$a_1+a_2{\mathrm i}+\lambda b_1+\lambda b_2{\mathrm i}= 0 .$$ For all $l\in L$, the leg $(a_l,b_l,d_l)$ must lie on this hyperplane. The real part and the imaginary part of this equation must both be zero: $a_{l,1}+\lambda b_{l,1}=a_{l,2}+\lambda b_{l,2}=0$. Therefore the claim follows.
The stratum $Z_s\subset B$ is also called the [*similarity stratum*]{}, and the any bond in this stratum is called a [*similarity bond*]{}. The theorem above states, in more informal language, that the presence of a similarity bond implies the existence of two similar projections of base and platform. If a linkage has an infinite number of similarity bonds, then it can be shown that for all projections of the platform points, there is a similar projection of the base points. This implies that the platform points and the base points are related by a similarity transformation of ${\mathbb R}^3$. This geometric observation plays a crucial role in the classification of multipods of mobility 2 in [@Schicho:15c].
There is the analoguous statement for the stratum $Z_i$; also the proof is analoguous.
\[thm:invbond\] Assume that the multipod above has a bond in $Z_i$. Then there exist orthogonal projections $p_a:{\mathbb R}^3\to{\mathbb R}^2$ and $p_b:{\mathbb R}^3\to{\mathbb R}^2$ and an inversion $i:{\mathbb R}^2\to{\mathbb R}^2$ such that $i(p(a_l))=p(b_l)$ for all $l\in L$.
Recall the Bricard-Borel multipod with infinitely many legs, described i Section \[sec:pod\]: all its legs $(a,b,d)$ satisfy the condition $$a_1b_2-a_2b_1 = a_1b_1+a_2b_2-\alpha = 0$$ for some fixed $\alpha\in{\mathbb R}$, $\alpha\ne 0$. As we already saw, this is an inversion relation between the projections of base and platform.
If a multipod has a bond in $Z_b$, then there are two lines $G_a,G_b\subset{\mathbb R}^3$ such that for any leg $(a,b,d)$ in the dual hyperplane in leg space, either $a$ lies in $G_a$ or $b$ lies in $G_b$. The presence of a bond in $Z_b$ implies the existence of two lines and a partition of the set of legs into two subsets, with the first subset having collinear anchor points in the base and the second subset having collinear anchor points in the platform. Let us called such a configuration a [*combined collineation*]{}. The existence of such a combined collineation already implies mobility for a suitable choice of leg lengths. To see this, we start with a configuration such that the lines $G_a$ and $G_b$ coincide – the leg lengths have to chosen so that such a configuration exists. Then we can rotate the platform around this line (similar as the “double banana” Figure \[fig:l3d\]). The stratum $Z_c$ has two irreducible components. For one of these components, the projection $(x_1:x_2:x_3)$ is defined. The geometric implication is stronger than the implication from a bond in $Z_b$: all anchor points in the base have to be collinear. For the second component, all anchor points in the platform have to be collinear. If one of these two conditions is fulfilled, then a rotation motion is possible from any starting position.
The hyperplane corresponding to the one point in $Z_v$ is the hyperplane at infinity. Hence there is no multipod with a finite leg that has a bond in $Z_v$.
In summary, any boundary point implies [*some*]{} geometric condition on the two configurations of platform points and of base points. Hence the compactification gives necessary conditions for mobility: if a multipod is mobile, then it must have some bonds, therefore one of the above geometric conditions hold.
Many mobile multipods have more than just one pair of complex conjugate bonds, since the number of bonds is related to the degree of the configuration curve embedded in ${\mathbb P}^{16}$. The correlation between the degree of the mobility curve of a hexapod and the number of special geometric events – similar projections, inverse projections, or combined collineations – motives the question on the maximal number of such events. Here are the answers.
Assume that the six-tuple of points in the base and the six-tuple of points in the platform are not similar, and that neither the base nor the platform consist of coplanar points.
a\) The number of combined collineations is at most 16. If every anchor point appears in at most one leg, for both base and platform, then the maximal number of combined collineations is 4.
b\) The number of projections related by a similarity transformations is at most 2. The maximum is reached if and only if the two six-tuples are affine equivalent.
c\) The number of projections related by an inversion is at most 7.
The proofs of (a) and (b) are left as exercises. For the proof of (c), we refer to [@Schicho:17b].
It is conjectured in [@Schicho:17b] that for a generic choice of six points in ${\mathbb R}^3$, there exists a second six-tuple of points, such that the maximal number of 7 projections related by an inversion is reached; such a six-tuple would then be unique up to similarity. The conjecture continues to state that there is a unique scaling and choice of leg length such that the so constructed hexapod is mobile, with a mobility curve of maximal degree 28. For a numeric random choice, the conjecture can be tested by a construction taking about 300 seconds using computer algebra. Using this construction, the conjecture has been verified for 50 random choices. Theoretically, it is still theoretically possible that these 50 random choices were picked on some unknown subvariety with non-generic behavior, but it is quite unlikely.
[10]{}
J. E. Baker. The single screw reciprocal to the general plane-symmetric six-screw linkage. , 1:5–12, 1997.
G. T. Bennett. The skew isogramm-mechanism. , 13(2nd Series):151–173, 1913–1914.
E. Borel. Mémoire sur les déplacements à trajectoires sphériques. , 33(1):1–128, 1908.
R. Bricard. Mémoire sur la théorie de l’octaèdre articulé. , 3:113–148, 1897.
R. Bricard. Mémoire sur les déplacements à trajectoires sphériques. , 11:1–96, 1906.
R. Bricard. . Gauthier-Villars, 1927.
K. Brunnthaler, H.-P. Schröcker, and M. Husty. A new method for the synthesis of [Bennett]{} mechanisms. In [*Proceedings of CK 2005, International Workshop on Computational Kinematics*]{}, Cassino, 2005.
J. Capco, M. Gallet, G. Grasegger, C. Koutschan, N. Lubbes, and J. Schicho. The number of realizations of a [Laman]{} graph. , 2:94–125, 2018.
A. Degtyarev and I. Itenberg. On real determinantal quartics. In [*Proceedings of the [G]{}ökova [G]{}eometry-[T]{}opology [C]{}onference 2010*]{}, pages 110–128. Int. Press, Somerville, MA, 2011.
E. Delassus. The closed and deformable linkage chains with four bars. , 46:283–304, 1922.
J. Denavit and R. S. Hartenberg. A kinematic notation for lower-pair mechanisms based on matrices. , 22:215–221, 1955.
P. Dietmaier. . Habilitation thesis, Graz University of Technology, 1995.
A. C. Dixon. . , 29(2):1–21, 1899.
M. Gallet, Nawratil G, J. Schicho, and J. Selig. Mobile icosapods. , 88:1–25, 2017.
M. Gallet, C. Koutschan, Z. Li, G. Regensburger, J. Schicho, and N. Villamizar. Planar linkages following a prescribed motion. , 86:473–506, 2017.
M. Gallet, G. Nawratil, and J. Schicho. Bond theory for pentapods and hexapods. , 2014.
M. Gallet, G. Nawratil, and J. Schicho. Liaison linkages. , 79:65–98, 2017.
B. Gordon and T. S. Motzkin. On the zeros of polynomials over division rings. , 116:218–226, 1965.
G. Grasegger, J. Legersky, and J. Schicho. Graphs with flexible labelings. , 62:461–480, 2019.
G. Grasegger, J. Legersky, and J. Schicho. Graphs with flexible labelings allowing injectve realizations. , pages 1–14, 2020. to appear.
G. Heged[ü]{}s, J. Schicho, and H.-P. Schröcker. Factorization of rational curves in the [Study]{} quadric and revolute linkages. , 69(1):142–152, 2013.
G. Hegedüs, J. Schicho, and H.-P. Schröcker. The theory of bonds: A new method for the analysis of linkages. , 70(0):407–424, 2013.
B. Jackson and T. Jordán. Rigid components in molecular graphs. , 48(4):399–412, 2007.
C. Jialong and M. Sitharam. Maxwell-independence: a new rank estimate for the 3-dimensional generic rigidity matroid. , 105:26–43, 2014.
A. Karger. Classification of [5R]{} closed kinematic chains with self mobility. , pages 213–222, 1998.
N. Katoh and S. i. Tanigawa. A proof of the molecular conjecture. , 45(4):647–700, 2011.
A. Kempe. On a general method of describing plane curves of the nth degree by linkwork. , S1-7(1):213, 1876.
G. Laman. On graphs and rigidity of plane skeletal structures. , 4:331–340, 1970.
Z. Li and J. Schicho. A technique for deriving equational conditions on the denavit-hartenberg parameters of a 6r linkage that are necessary for movability. , 94:1–8, 2015.
Z. Li, J. Schicho, and H.-P. Schr[ö]{}cker. empe’s universality theorem for rational space curves. , 18:509–536, 2018.
G. Nawratil and J. Schicho. Pentapods with mobility 2. , 7, 2015.
J. C. Ottem, K. Ranestad, B. Sturmfels, and C. Vinzant. Quartic spectrahedra. , 151(2, Ser. B):585–612, 2015.
H. [Pollaczek-Geiringer]{}. . , 7:58–72, 1927.
B. Schulze. Symmetry as a sufficient condition for a finite flex. , 24(4):1291–1312, 2010.
J. Selig. . Monographs in Computer Science. Springer, 2 edition, 2005.
A. J. Sommese and C. W. Wampler. Numerical algebraic geometry and algebraic kinematics. , 20:469–567, 2011.
E. Study. . Teubner, Leipzig, 1903.
D. Walter and M. L. Husty. On a nine-bar linkage, its possible configurations and conditions for paradoxical mobility. In [*12th World Congress on Mechanism and Machine Science, IFToMM 2007*]{}, 2007.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We construct a metrizable semitopological semilattice $X$ whose partial order $P=\{(x,y)\in X\times X:xy=x\}$ is a non-closed dense subset of $X\times X$. As a by-product we find necessary and sufficient conditions for the existence of a (metrizable) Hausdorff topology on a set, act, semigroup or semilattice, having a prescribed countable family of convergent sequences.'
address:
- 'T.Banakh: Ivan Franko National University of Lviv (Ukraine) and Jan Kochanowski University in Kielce (Poland)'
- 'S. Bardyla: Institute of Mathematics, Kurt Gödel Research Center, Vienna (Austria)'
- 'A.Ravsky: Department of Analysis, Geometry and Topology, Pidstryhach Institute for Applied Problems of Mechanics and Mathematics National Academy of Sciences of Ukraine'
author:
- 'Taras Banakh, Serhii Bardyla, and Alex Ravsky'
title: |
A metrizable semitopological semilattice\
with non-closed partial order
---
Introduction
============
In this paper we shall construct an example of a metrizable semitopological semilattice with non-closed partial order.
A [*semilattice*]{} is a commutative semigroup $X$ whose any element $x\in X$ is an [*idempotent*]{} in the sense that $xx=x$. A typical example of a semilattice is any partially ordered set $X$ in which any finite non-empty set $F\subset X$ has the greatest lower bound $\inf(F)$. In this case the binary operation $X\times X\to X$, $(xy)\mapsto\inf\{x,y\}$, turns $X$ into a semilattice.
Each semilattice $X$ carries a partial order $\le$ defined by $x\le y$ iff $xy=x$. For this partial order we have $xy=\inf\{x,y\}$.
A ([*semi*]{})[*topological semilattice*]{} is a semilattice $X$ endowed with a topology such that the binary operation $X\times X\to X$, $xy\mapsto xy$, is (separately) continuous.
The continuity of the semilattice operation in a Hausdorff topological semilattice implies the following well-known fact, see [@Bible VI-1.14].
For any Hausdorff topological semilattice $X$ the partial order $P=\{(x,y)\in X\times X:xy=x\}$ is a closed subset of $X\times X$.
It is natural to ask whether this proposition remains true for Hausdorff semitopological semigroups. The following example answers this question in negative.
There exists a metrizable countable semitopological semilattice $X$ whose partial order is dense and non-closed in $X\times X$.
This example will be constructed in Section \[s:Ex\] after some preliminary work, made in Sections \[s:T\]–\[s:SL\]. In Section \[s:T\] we establish necessary and sufficient conditions on a set $X$ and function $\ell:{\mathrm{dom}}(\ell)\to X$ defined on a subset ${\mathrm{dom}}(\ell)\subset X^{\omega}$ ensuring that $X$ admits a (metrizable) Hausdorff topology in which every sequence $s\in{\mathrm{dom}}(\ell)$ converges to the point $\ell(s)$. In Section \[s:TA\] we study the analogous problem for acts, i.e. sets endowed with monoids of self-maps and in Section \[s:SG\], \[s:SL\] we apply the obtained results about acts to constructing topologies with prescribed convergent sequences on semigroups and semilattices. More information on the closedness of the partial order in semitopological semilattices can be found in [@BB §7].
Convergent sequences in topological spaces {#s:T}
==========================================
Let $X$ be a set and $X^{\omega}$ be its countable power. Elements of $X^{\omega}$ are sequences $s=(s_n)_{n\in{\omega}}$. Let $\ell:{\mathrm{dom}}(\ell)\to X$ be a function defined on a subset ${\mathrm{dom}}(\ell)\subset X^{\omega}$.
A topology $\tau$ on the set $X$ is called [*$\ell$-admissible*]{} if each sequence $s\in{\mathrm{dom}}(\ell)$ converges to the point $\ell(s)$ in the topological space $(X,\tau)$.
Observe that the indiscrete topology $\{\emptyset,X\}$ on $X$ is $\ell$-admissible. So, the family of $\ell$-admissible topologies is not empty. This family has the largest element. This is the topology $\tau_\ell$ consisting of all subsets $U\subset X$ such that for any sequence $s=(s_n)_{n\in{\omega}}\in{\mathrm{dom}}(\ell)$ with $\ell(s)\in U$ the set $\{n\in{\omega}:s_n\notin U\}$ is finite. The topology $\tau_\ell$ will be referred to as the [*largest $\ell$-admissible topology*]{} on $X$. In this section we discuss the following problem.
Under which conditions the largest $\ell$-admissible topology $\tau_\ell$ on $X$ is Hausdorff?
Below we define two necessary conditions of the Haudorffness of the topology $\tau_\ell$.
The function $\ell:{\mathrm{dom}}(\ell)\to X$ is defined to be
- [*$T_1$-separating*]{} if for any sequence $s\in{\mathrm{dom}}(\ell)$ any any point $x\in X$ with $x\ne\ell(s)$ the set $\{n\in{\omega}:s_n=x\}$ is finite;
- [*$T_2$-separating*]{} if $\ell$ is $T_1$-separating and for any sequences $s,t\in{\mathrm{dom}}(\ell)$ with $\ell(s)\ne\ell(t)$ there exists a finite set $F\subset{\omega}$ such that $s_n\ne t_m$ for any $n,m\in{\omega}\setminus F$.
We say that a topology $\tau$ on a set $T$ [*satisfies the separation axiom*]{} $T_1$ if each finite subset of $T$ is $\tau$-closed in $T$. In this case we say that $(T,\tau)$ is a [*$T_1$-space*]{}.
\[l:T1\] The function $\ell$ is $T_1$-separating if and only if the topology $\tau_\ell$ satisfies the separation axiom $T_1$.
Assume that $\ell$ is $T_1$-separating and take any finite set $F\subset X$. To show that the set $U:=X\setminus F$ belongs to the topology $\tau_\ell$, it suffices to check that for every $s\in {\mathrm{dom}}(\ell)$ with $\ell(s)\in U$ the set $\{n\in{\omega}:s(n)\notin U\}$ is finite. By the $T_1$-separating property of $\ell$, for every $x\in F$ the set $\Omega_x=\{n\in{\omega}:s_n=x\}$ is finite and so is the set $\{n\in{\omega}:s_n\notin U\}=\bigcup_{x\in F}\Omega_x$.
Now assuming that each finite subset $F\subset X$ is closed in the topology $\tau_\ell$, we shall prove that the function $\ell$ is $T_1$-separating. Given any point $x\in X\setminus\{\ell(s)\}$, observe that $\ell(s)\in X\setminus\{x\}\in\tau_\ell$ implies that the set $\{n\in{\omega}:s_n\notin X\setminus\{x\}\}=\{n\in{\omega}:s_n=x\}$ is finite, which means that $\ell$ is $T_1$-separating.
\[l:T2a\] If the topology $\tau_\ell$ is Hausdorff, then the function $\ell$ is $T_2$-separating.
By Lemma \[l:T1\], the function $\ell$ is $T_1$-separating. To prove that $\ell$ is $T_2$-separating, take two sequences $s,t\in {\mathrm{dom}}(\ell)$ with $\ell(s)\ne\ell(t)$. By the Hausdorff property of the topology $\tau_\ell$, there are disjoint open sets $U,V\in\tau_\ell$ such that $\ell(s)\in U$ and $\ell(t)\in V$. By the definition of the topology $\tau_\lambda$, the sets $F:=\{n\in{\omega}:s_n\notin U\}$ and $E:=\{m\in{\omega}:t_m\notin V\}$ are finite. Then $s_n\ne t_m$ for any $n,m\in{\omega}\setminus(F\cup E)$.
Now we shall prove that the largest $\ell$-admissible topology $\tau_\ell$ is Hausdorff if the function $\ell$ is $T_2$-separating and ${\mathrm{dom}}(\ell)$ is at most countable.
For a sequence $s\in{\mathrm{dom}}(\ell)$ and a subset $I\subset{\omega}$ let $s[I]:=\{s_n:n\in I\}$ and $s[I]^*:=s[I]\cup\{\ell(s)\}$.
\[l:T2b\] Assume that the function $\ell:{\mathrm{dom}}(\ell)\to X$ is $T_2$-separating. Let $s\in{\mathrm{dom}}(\ell)$ and $I\subset{\omega}$.
1. the set $s[I]^*$ is closed in $(X,\tau_\ell)$;
2. each point $x\in s[I]^*\setminus\{\ell(s)\}$ is isolated in $s[I]^*$;
3. the subspace $s[I]^*$ of $(X,\tau_\ell)$ is compact and Hausdorff.
1\. The inclusion $X\setminus s[I]^*\in \tau_\ell$ will follow as soon as we show that for any sequence $t\in {\mathrm{dom}}(\ell)$ with $\ell(t)\notin s[I]^*$ the set $\{n\in{\omega}:t_n\in s[I]^*\}$ is finite. By the $T_2$-separating property of $\ell$, there exists a finite set $\Omega\subset{\omega}$ such that $s_n\ne t_m$ for any $n,m\in{\omega}\setminus\Omega$. Consider the finite set $E=\{\ell(s)\}\cup\{s_n:n\in\Omega\}\setminus\{\ell(t)\}$. By the $T_1$-separating property of $\ell$, the set $\Lambda=\Omega\cup\{n\in{\omega}:t_n\in E\}$ is finite. Then the set $\{n\in{\omega}:t_n\in s[I]^*\}\subset \Lambda$ is finite, too.
2\. Given any point $x\in s[I]^*\setminus\{\ell(s)\}$, observe that $s[I]^*\setminus\{x\}=s[J]^*$ where $J=I\setminus s^{-1}(x)$.
By Lemma \[l:T2b\](1), the subspace $s[J]^*$ is closed in $(X,\tau_\ell)$ and then the singleton $\{x\}=s[I]^*\setminus s[J]^*$ is open in $s[I]^*$.
3\. The compactness of $s[I]^*$ follows from the fact that each neighborhood $U\in\tau_\ell$ of $\ell(s)$ contains all but finitely many points of the set $s[I]$. To see that $s[I]^*$ is Hausdorff, take any two distinct points $x,y\in s[I]^*$. One of these points is distinct from the limit point $\ell(s)$ of the sequence $s$ and we lose no generality assuming that $x\ne\ell(s)$. By Lemmas \[l:T1\] and \[l:T2b\](2), the singleton $U_x=\{x\}$ closed-and-open in $s[I]^*$ and so is its complement $U_y=s[I]^*\setminus U_x$. Then $U_x,U_y$ are disjoint neighborhoods of the points $x,y$ in $s[I]^*$, witnessing that the subspace $s[I]^*$ of $(X,\tau_\ell)$ is Hausdorff.
\[c:T2\] If the function $\ell:{\mathrm{dom}}(\ell)\to X$ is $T_2$-separating, then for every finite subset $S\subset {\mathrm{dom}}(\lambda)$ the subspace $S[{\omega}]^*=\bigcup_{s\in S}s[{\omega}]^*$ of $(X,\tau_\ell)$ is compact, Hausdorff, and closed in $(X,\tau_\ell)$.
This corollary follows from Lemma \[l:T2b\] and the following known fact.
The union $A\cup B$ of two closed Hausdorff subspaces of a topological space $T$ is Hausdorff.
We lose no generality assuming that $T=A\cup B$. The Hausdorff property of the space $T=A\cup B$ will follow as soon as we check that its diagonal $\Delta_T=\{(x,x):x\in T\}$ is closed in $T\times T$. For this observe that $\Delta_T=\Delta_A\cup\Delta_B$. Since the space $A$ is Hausdorff and closed in $T$, its diagonal $\Delta_A$ is closed in $A\times A$ and in $T\times T$. By analogy, the diagonal $\Delta_B$ is closed in $B\times B$ and in $T\times T$. Then the union $\Delta_T=\Delta_A\cup\Delta_B$ is closed in $T\times T$ and the space $T$ is Hausdorff.
We say that the topology $\tau$ of a topological space $X$ is [*generated*]{} by a family ${\mathcal K}$ of subspaces of $X$ if a set $U\subset X$ is open in $X$ if any only if for every $K\in{\mathcal K}$ the intersection $U\cap K$ is open in the subspace topology of $K$. A topology $\tau$ on a set $X$ is called [*a $k_{\omega}$-topology*]{} if it is generated by a countable family of compact subsets of the topological space $(X,\tau)$.
\[l:T2n\] If the function $\ell:{\mathrm{dom}}(\ell)\to X$ is $T_2$-separating and ${\mathrm{dom}}(\ell)$ is at most countable, then the topology $\tau_\ell$ is Hausdorff and normal. Moreover, $\tau_\lambda$ is a $k_{\omega}$-topology, generated by the countable family ${\mathcal K}=\{s[{\omega}]^*:s\in{\mathrm{dom}}(\ell)\}$ of compact sets in $(X,\tau_\ell)$.
The definition of the topology $\tau_\ell$ ensures that it is generated by the countable family ${\mathcal K}$. Now we show that the topology $\tau_\ell$ is Hausdorff and normal. By Lemma \[l:T1\], the topology $\tau_\ell$ satisfies the separation axiom $T_1$. Now it suffices to check that this topology is normal. From now on, we consider $X$ as a topological space endowed with the topology $\tau_\ell$.
Given two disjoint closed sets $A,B\subset X$ we should find two disjoint open sets $V,W\subset X$ such that $A\subset V$ and $B\subset W$.
Let ${\mathrm{dom}}(\ell)=\{s_n\}_{n\in{\omega}}$ be an enumeration of the countable set ${\mathrm{dom}}(\ell)$. By Corollary \[c:T2\], for every $n\in{\omega}$ the subspace $K_n:=\bigcup_{i\le n}s_i[{\omega}]^*$ of $(X,\tau_\ell)$ is compact, Hausdorff, and closed in $(X,\tau_\ell)$. Let $A_0:=A\cap K_0$ and $B_0:=B\cap K_0$. By induction we shall construct sequences $(A_n)_{n\in{\omega}}$, $(B_n)_{n\in{\omega}}$, $(V_n)_{n\in{\omega}}$, $(W_n)_{n\in{\omega}}$ of subsets in $X$ such that for every $n\in{\omega}$ the following conditions are satisfied:
1. the sets $A_n,B_n$ are disjoint and closed in $K_n$;
2. $A_n\subset V_n\subset K_n$ and $B\subset W_n\subset K_n$;
3. the sets $V_n,W_n$ are open in $K_n$ and $\overline V_n\cap\overline W_n=\emptyset$;
4. $A_{n+1}=\overline{V}_n\cup(K_{n+1}\cap A)$ and $B_{n+1}=\overline W_n\cup(K_{n+1}\cap B)$;
5. $A_{n+1}\cap B_{n+1}=\emptyset$.
Assume that for some $n\in{\omega}$ disjoint closed sets $A_n,B_n\subset K_n$ with $K_n\cap A\subset A_n$ and $K_n\cap B\subset B_n$ have been constructed. By the normality of the compact Hausdorff space $K_n$, there are open sets $V_n,W_n\subset K_n$ satisfying the conditions (2),(3). Define the sets $A_{n+1},B_{n+1}$ by the formula (4) and observe that $$A_{n+1}\cap B_{n+1}=(A_{n+1}\cap B_{n+1}\cap K_n)\cup(A_{n+1}\cap B_{n+1}\cap(K_{n+1}\setminus K_n)\subset(\overline V_n\cap\overline W_n)\cup(A\cap B)=\emptyset.$$ After completing the inductive construction, observe that $V:=\bigcup_{n\in{\omega}}V_n$ and $W=\bigcup_{n\in{\omega}}W_n$ are disjoint open sets in $X$ such that $A\subset V$ and $B\subset W$.
The following theorem is the main result of this section.
\[t:T\] For a set $X$ and function $\ell:{\mathrm{dom}}(\ell)\to X$ defined on a countable subset ${\mathrm{dom}}(\ell)\subset X^{\omega}$ the following conditions are equivalent:
1. $X$ admits a metrizable topology $\tau$ in which every sequence $s\in{\mathrm{dom}}(\ell)$ converges to the point $\ell(s)$.
2. $X$ admits a Hausdorff topology $\tau$ in which every sequence $s\in{\mathrm{dom}}(\ell)$ converges to the point $\ell(s)$.
3. The following two properties are satisfied:
- for any $s\in{\mathrm{dom}}(\ell)$ and $x\in X$ with $s\ne\ell(s)$ the set $\{n\in{\omega}:s_n=x\}$ is finite;
- for any sequences $s,t\in{\mathrm{dom}}(\ell)$ with $\ell(s)\ne\ell(t)$ there exists a finite set $F\subset{\omega}$ such that $s_n\ne t_m$ for any $n,m\in{\omega}\setminus F$.
The implications $(1){\Rightarrow}(2){\Rightarrow}(3)$ are trivial.
To prove that $(3){\Rightarrow}(1)$, assume the the condition (3) is satisfied. Then the function $\ell$ is $T_2$-separating and by Lemma \[l:T2n\], the largest $\ell$-admissible topology $\tau_\ell$ is Hausdorff and normal. Consider the countable subset $D=\bigcup_{s\in{\mathrm{dom}}(\ell)}s[{\omega}]^*$ of $X$ and observe that $X\setminus D$ is a closed-and-open discrete subspace of the topological space $X$ endowed with the topology $\tau_\ell$. Being countable and Tychonoff, the closed-and-open subspace $D$ of $X$ is zero-dimensional. Then for any distinct points $x,y\in D$ we can choose a closed-and-open subset $U_{x,y}\subset X$ such that $x\in U_{x,y}$ and $y\notin U_{x,y}$. Let $\tau$ be the topology on $D$, generated by the countable subbase $\{U_{x,y},X\setminus U_{x,y}:x,y\in D,\;x\ne y\}$. It is clear that the topology $\tau$ is second-countable, Hausdorff, zero-dimensional and hence regular. By Urysohn Metrization Theorem [@Eng 4.2.9], the topological space $D_\tau=(D,\tau)$ is metrizable. Then the topology of the topological sum on $(X\setminus D)\oplus D_\tau$ is also metrizable. Since $\tau\subset\tau_\ell$, the topology $\tau$ is $\ell$-admissible, which means that each sequence $s\in{\mathrm{dom}}(\ell)$ converges to the point $\ell(s)$.
The countability of the domain ${\mathrm{dom}}(\ell)$ in Theorems \[t:T\] is essential as shown by the following example.
There exists a $T_2$-separating function $\ell:{\mathrm{dom}}(\ell)\to \{0,1\}\subset {\omega}$ defined on a subset ${\mathrm{dom}}(\ell)\subset[{\omega}]^{\omega}$ of cardinality $|{\mathrm{dom}}(\ell)|={\omega}_1$ such that the largest $\ell$-admissible topology $\tau_\ell$ is not Hausdorff.
To construct such function $\ell$, take any Hausdorff $({\omega}_1,{\omega}_1)$-gap on ${\omega}$, which is a pair $\big((A_i)_{i\in{\omega}_1},(B_i)_{i\in{\omega}_1}\big)$ of families of infinite subsets of ${\omega}$ satisfying the following two conditions:
- for any $i<j<{\omega}_1$ we have $A_i\subset^*A_j$ and $B_i\subset^* B_j$;
- $A_i\cap B_j$ is finite for any $i,j\in{\omega}_1$;
- for any set $C\subset{\omega}$ one of the sets $\{i\in{\omega}_1:A_i\subset^* C\}$ or $\{i\in{\omega}_1:B_i\subset^*{\omega}\setminus C\}$ is at most countable.
Here the notation $A\subset^* B$ means that the complement $A\setminus B$ is finite.
It is well-known [@JW2 Ch.20] that Hausdorff $({\omega}_1,{\omega}_1)$-gaps do exist in ZFC.
For every $i\in{\omega}_1$ choose any bijective functions $\alpha_i:{\omega}\to A_i$ and $\beta_i:{\omega}\to B_i$, and put ${\mathrm{dom}}(\ell)=\{\alpha_i,\beta_i:i\in{\omega}_1\}$. Let $\ell:{\mathrm{dom}}(\ell)\to\{0,1\}\subset{\omega}$ be the function such that $\ell^{-1}(0)=\{\alpha_i\}_{i\in{\omega}_1}$ and $\ell^{-1}(1)=\{\beta_i\}_{i\in{\omega}}$. The injectivity of the functions $\alpha_i,\beta_i$ and the condition (H2) ensure that the function $\ell$ is $T_2$-separating. Assuming that the $\ell$-admissble topology on ${\omega}$ is Hausdorff, we could find two disjoint open sets $U_0,U_1\in\tau_\ell$ such that $0\in U_0$ and $1\in U_1$. By condition (H3), there exists $i\in{\omega}_1$ such that $A_i\not\subset^* U_0$ or $B_i\not\subset^* U_1$. In the first case the set $\{n\in{\omega}:\alpha_i(n)\notin U_0\}$ is infinite, which contradicts $U_0\in\tau_\ell$. In the second case the set $\{n\in{\omega}:\beta_i(n)\notin U_1\}$ is infinite, which contradicts $U_1\in\tau_\ell$.
Convergent sequences in topological acts {#s:TA}
========================================
For a set $X$ denote by $X^X$ the set of all self-maps $X\to X$. The set $X^X$ endowed with the operation of composition is a monoid whose unit is the identity map ${\mathrm{id}}_X$ of $X$.
An [*act*]{} is a pair $(X,{\mathcal A})$ consisting of a set $X$ and a submonoid ${\mathcal A}\subset X^X$. Elements of the set ${\mathcal A}$ are called the [*shifts*]{} of the act $(X,{\mathcal A})$.
A topology $\tau$ on the underlying set $X$ of an act $(X,{\mathcal A})$ is called an [*shift-continuous*]{} if each shift $\alpha\in {\mathcal A}$ is a continuous self-map of the topological space $(X,\tau)$.
\[t:TA\] For an act $(X,{\mathcal A})$ with countable set ${\mathcal A}$ of shifts and a function $\ell:{\mathrm{dom}}(\ell)\to X$ defined on a countable subset ${\mathrm{dom}}(\ell)\subset X^{\omega}$ the following conditions are equivalent:
1. $X$ admits a shift-continuous metrizable topology $\tau$ in which every sequence $s\in{\mathrm{dom}}(\ell)$ converges to the point $\ell(s)$;
2. $X$ admits a shift-continuous Hausdorff topology $\tau$ in which every sequence $s\in{\mathrm{dom}}(\ell)$ converges to the point $\ell(s)$;
3. the following two properties hold:
- for any $s\in{\mathrm{dom}}(\ell)$, $\alpha\in {\mathcal A}$ and $x\in X$ with $x\ne \alpha\circ \ell(s)$ the set $\{n\in{\omega}:\alpha\circ s_n=x\}$ is finite;
- for any sequences $s,t\in{\mathrm{dom}}(\ell)$ and any shifts $\alpha,\beta\in{\mathcal A}$ with $\alpha{\circ}\ell(s)\ne\beta{\circ}\ell(t)$ there exists a finite set $F\subset {\omega}$ such that $\alpha{\circ}s_n\ne \beta{\circ}t_m$ for any $n,m\in{\omega}\setminus F$.
The implications $(1){\Rightarrow}(2){\Rightarrow}(3)$ are trivial. To prove that $(3){\Rightarrow}(1)$, assume that the condition (3) is satisfied. Consider the set ${\mathrm{dom}}({\mathcal A}\ell)=\{\alpha\circ s:\alpha\in {\mathcal A},\;s\in {\mathrm{dom}}(\ell)\}\subset X^{\omega}$ and the function ${\mathcal A}\ell:{\mathrm{dom}}({\mathcal A}\ell)\to X$ defined by the formula ${\mathcal A}\ell(\alpha\circ s)=\alpha(\ell(s))$. The conditions (3a) and (3b) ensure that the function ${\mathcal A}\ell$ is well-defined and is $T_2$-separating. By Lemma \[l:T2n\], the largest ${\mathcal A}\ell$-admissible topology $\tau_{{\mathcal A}\ell}$ on $X$ is Hausdorff and normal. Since the monoid ${\mathcal A}$ contains the identity map of $X$, the topology $\tau_{{\mathcal A}\ell}$ is $\ell$-admissible, which implies that each sequence $s\in{\mathrm{dom}}(\ell)$ converges to $\ell(s)$ in the topological space $(X,\tau_{{\mathcal A}\ell})$.
We claim that the topology $\tau_{{\mathcal A}\ell}$ is shift-continuous. Given any shift $\alpha\in{\mathcal A}$ and open set $U\in\tau_{{\mathcal A}\ell}$, we need to check that $\alpha^{-1}(U)\in\tau_{{\mathcal A}\ell}$. The latter inclusion holds if any only if for any sequence $s\in{\mathrm{dom}}({\mathcal A}\ell)$ with ${\mathcal A}\ell(s)\in \alpha^{-1}(U)$ the set $\{n\in{\omega}:s_n\notin \alpha^{-1}(U)\}=\{n\in{\omega}:\alpha\circ s_n\notin U\}$ is finite. Since ${\mathcal A}$ is a monoid, the sequence $\alpha\circ s$ belongs to ${\mathrm{dom}}({\mathcal A}\ell)$. Since $U\in\tau_{{\mathcal A}\ell}$ and ${\mathcal A}\ell(\alpha\circ s)=\alpha({\mathcal A}\ell(s))\in U$, the set $\{n\in{\omega}:\alpha\circ s_n\notin U\}=\{n\in{\omega}:s_n\notin \alpha^{-1}(U)\}$ is finite and we are done.
The countability of the sets ${\mathrm{dom}}(\ell)$ and ${\mathcal A}$ imply the countability of the sets ${\mathrm{dom}}({\mathcal A}\ell)$ and $D=\{s[{\omega}]^*:s\in {\mathrm{dom}}({\mathcal A}\ell)\}$. Observe that for every $\alpha\in {\mathcal A}$ we have $\alpha[D]\subset D$.
By the definition of the topology $\tau_{{\mathcal A}\ell}$, the set $D$ is open-and-closed in $(X,\tau_{{\mathcal A}\ell})$ and the complement $X\setminus D$ is discrete. Being Tychonoff, the countable subspace $D$ of $(X,\tau_{{\mathcal A}\ell})$ is zero-dimensional. This allows us to choose a countable family $\mathcal B\subset\tau_{{\mathcal A}\ell}$ of open-and-closed sets that separate points of the countable set $D$ in the sense that for any distinct points $x,y\in D$ there exists a set $B\in\mathcal B$ such that $x\in B$ and $y\notin B$. Since $\tau_{{\mathcal A}\ell}$ is an act topology on $(X,{\mathcal A})$, for every $\alpha\in {\mathcal A}$ and $B\in\mathcal B$ the set $\alpha^{-1}(B)$ is closed-and-open. Then the topology $\tau_D$ on $D$ generated by the subbase $\{D\cap \alpha^{-1}(B), D\setminus \alpha^{-1}(B):\alpha\in{\mathcal A},\;B\in\mathcal B\}$ is second-countable, Hausdorff and zero-dimensional. By the Urysohn metrization Theorem [@Eng 4.2.9], the topological space $D_\tau=(D,\tau_D)$ is metrizable. Then the topology $\tau$ of topological sum $D_\tau\oplus(X\setminus D)$ of $D_\tau$ and the discrete topological space $X\setminus D$ is metrizable. By the definition of the topology $\tau_D$, for every $\alpha\in{\mathcal A}$ the restriction $\alpha{\restriction}D$ is a continuous self-map of the topological space $D_\tau$. Since $X\setminus D$ is a closed-and-open discrete subspace of $(X,\tau)$, the continuity of $\alpha{\restriction}D$ implies that $\alpha$ is a continuous self-map of the metrizable topological space $(X,\tau)$. This means that the topology $\tau$ is shift-continuous. Since $\tau\subset\tau_{{\mathcal A}\ell}$, the metrizable topology $\tau$ is $\ell$-admissible.
Theorem \[t:TA\] can be compared with the following result proved in [@BPS 3.4].
Let $\kappa$ be an infinite cardinal, $(X,{\mathcal A})$ be an act, $x\in X$, and $(x_i)_{i\in \kappa}$ be a transfinite sequence of points in $X$. Assume that there exists a (not necessarily bijective) enumeration ${\mathcal A}=\{\alpha_i\}_{i\in\kappa}$ of the set ${\mathcal A}$ such that for each ordinal $m\in\kappa$ and ordinals $i,j,k<m$ the following conditions are satisfied:
1. if $\alpha_i(x)\ne \alpha_j(x)$, then $\alpha_i(x_m)\ne \alpha_j(x_m)$;
2. if $\alpha_i(x)\ne \alpha_j(x_k)$, then $\alpha_i(x_m)\ne \alpha_j(x_k)$.
Then $X$ admits a shift-continuous hereditarily normal topology $\tau$ in which the transfinite sequence $(x_i)_{i\in\lambda}$ converges to the point $x$ in the sense that for every neighborhood $O_x\in\tau$ of $x$ there exists $n\in\kappa$ such that $x_i\in O_x$ for all $i\ge n$ in $\kappa$.
Convergent sequences in semigroups {#s:SG}
==================================
Let $X$ be a semigroup and $X^1$ be the semigroup $X$ with attached unit. A topology $\tau$ on $X$ is called [*shift-continuous*]{} if for every $a,b\in X^1$ the two-sided shift $$X\to X,\;\;x\mapsto axb,$$is a continuous self-map of the topological space $(X,\tau)$.
Each semigroup $X$ has the structure of an act $(X,{\mathcal A})$ endowed with the family of shifts ${\mathcal A}=\{s_{a,b}:a,b\in X^1\}$. Applying Theorem \[t:TA\] to this act, we obtain the following theorem, which is a main result of this section.
\[t:TS\] For a countable semigroup $X$ and a function $\ell:{\mathrm{dom}}(\ell)\to X$ defined on a countable subset ${\mathrm{dom}}(\ell)\subset X^{\omega}$, the following conditions are equivalent:
1. The semigroup $X$ admits a shift-continuous metrizable topology $\tau$ in which every sequence $s\in{\mathrm{dom}}(\ell)$ converges to the point $\ell(s)$;
2. The semigroup $X$ admits a shift-continuous Hausdorff topology $\tau$ in which every sequence $s\in{\mathrm{dom}}(\ell)$ converges to the point $\ell(s)$;
3. the following two properties hold:
- for any $s\in{\mathrm{dom}}(\ell)$, $a,b\in X^1$ and $x\in X$ with $x\ne a{\cdot}\ell(s){\cdot}b$ the set $\{n\in{\omega}:a{\cdot}s_n{\cdot}b=x\}$ is finite;
- for any $s,t\in{\mathrm{dom}}(\ell)$ and $a,b,c,d\in X^1$ with $a{\cdot}\ell(s){\cdot}b\ne c{\cdot}\ell(t){\cdot}d$ there exists a finite set $F\subset{\omega}$ such that $a{\cdot}s_n{\cdot}b\ne c{\cdot}t_m{\cdot}d$ for all $n,m\in{\omega}\setminus F$.
For commutative semigroups, Theorem \[t:TS\] has a bit simpler form.
\[t:TAS\] For a countable commutative semigroup $X$ and a function $\ell:{\mathrm{dom}}(\ell)\to X$ defined on a countable subset ${\mathrm{dom}}(\ell)\subset X^{\omega}$, the following conditions are equivalent:
1. The semigroup $X$ admits a shift-continuous metrizable topology $\tau$ in which every sequence $s\in{\mathrm{dom}}(\ell)$ converges to the point $\ell(s)$.
2. The semigroup $X$ admits a shift-continuous Hausdorff topology $\tau$ in which every sequence $s\in{\mathrm{dom}}(\ell)$ converges to the point $\ell(s)$.
3. The following two properties hold:
- for any $s\in{\mathrm{dom}}(\ell)$, $a\in X^1$ and $x\in X$ with $x\ne a{\cdot}\ell(s)$ the set $\{n\in{\omega}:a{\cdot}s_n=x\}$ is finite;
- for any $s,t\in{\mathrm{dom}}(\ell)$ and $a,b\in X^1$ with $a{\cdot}\ell(s)\ne b{\cdot}\ell(t)$ there exists a finite subset $F\subset{\omega}$ such that $a{\cdot}s_n\ne b{\cdot}t_m$ for any $n,m\in{\omega}\setminus F$.
Convergent sequences in semilattices {#s:SL}
====================================
Applying Theorem \[t:TAS\] to semilattices we obtain the following characterization.
\[t:TL\] For a countable semilattice $X$ and a function $\ell:{\mathrm{dom}}(\ell)\to X$ defined on a countable subset ${\mathrm{dom}}(\ell)\subset X^{\omega}$ the following conditions are equivalent:
1. The semilattice $X$ admits a shift-continuous metrizable topology $\tau$ in which every sequence $s\in{\mathrm{dom}}(\ell)$ converges to the point $\ell(s)$.
2. The semilattice $X$ admits a shift-continuous Hausdorff topology $\tau$ in which every sequence $s\in{\mathrm{dom}}(\ell)$ converges to the point $\ell(s)$.
3. The following two conditions hold:
- for any $s\in{\mathrm{dom}}(\ell)$, $a\in X$ and $x\in X$ with $x\ne a{\cdot}\ell(s)$ the set $\{n\in{\omega}:a{\cdot}s_n=x\}$ is finite;
- for any $s,t\in{\mathrm{dom}}(\ell)$ and $a,b\in X$ with $a{\cdot}\ell(s)\ne b{\cdot}\ell(t)$ there exists a finite set $F\subset{\omega}$ such that $a{\cdot}s_n\ne b{\cdot}t_m$ for any $n,m\in{\omega}\setminus F$.
The implications $(1){\Rightarrow}(2){\Rightarrow}(3)$ are trivial. The implication $(3){\Rightarrow}(1)$ will follow from Theorem \[t:TAS\] as soon as we check that the condition $(3)$ of Theorem \[t:TL\] implies condition (3) of Theorem \[t:TAS\]. So, assume that condition (3) of Theorem \[t:TL\] is satisfied.
To check the condition (3a) of Theorem \[t:TAS\], take any sequence $s\in {\mathrm{dom}}(\ell)$ and element $a\in X^1$ and $x\in X$ with $x\ne a{\cdot}\ell(s)$. If $a\in X$, then the set $\{n\in{\omega}:a{\cdot}s_n=x\}$ is finite by the condition (3a) of Theorem \[t:TL\]. So, we assume that $a$ is an external unit for $X$. In this case $\{n\in{\omega}:a{\cdot}s_n=x\}=\{n\in{\omega}:s_n=x\}\subset\{n\in{\omega}:x{\cdot}s_n=x{\cdot}x=x\}$. Assuming that the set $\{n\in{\omega}:s_n=x\}$ is infinite, we conclude that the set $\{n\in{\omega}:x{\cdot}s_n=x\}$ is infinite, which implies that $x{\cdot}\ell(s)=x$. Since $\ell(s){\cdot}\ell(s)=\ell(s)=a{\cdot}\ell(s)\ne x=\ell(s){\cdot}x$, the set $$\{n\in{\omega}:\ell(s){\cdot}s_n=x\}=\{n\in{\omega}:\ell(s){\cdot}s_n=\ell(s){\cdot}x\}\supset \{n\in{\omega}:s_n=x\}$$is finite, which contradicts our assumption.
Next, we check the condition (3b) of Theorem \[t:TAS\]. Given any sequences $s,t\in{\mathrm{dom}}(\ell)$ and elements $a,b\in X^1$ with $a{\cdot}\ell(s)\ne b{\cdot}\ell(t)$, we need to find a finite set $F\subset{\omega}$ such that $a{\cdot}s_n\ne b{\cdot}t_m$ for any $n,m\in{\omega}\setminus F$. The condition (3a) of Theorem \[t:TL\] ensures that $a$ or $b$ does not belong to $X$. We lose no generality assuming that $a\notin X$ and hence $a$ is the external unit to $X$. In this case the inequality $a{\cdot}\ell(s)\ne b{\cdot}\ell(t)$ transforms into the inequality $\ell(s)\ne b{\cdot}\ell(t)$. We claim that there exists an element $c\in X$ such that $c{\cdot}\ell(s)\ne cb{\cdot}\ell(t)$. If $\ell(s)\ne \ell(s){\cdot}b{\cdot}\ell(t)$, then put $c=\ell(s)$. If $\ell(s)=\ell(s){\cdot}b{\cdot}\ell(t)$, then put $c=b{\cdot}\ell(t)$ and conclude that $c{\cdot}\ell(s)=\ell(s)\ne b{\cdot}\ell(t)=cb{\cdot}\ell(t)$.
In both cases we get $c{\cdot}\ell(s)\ne cb{\cdot}\ell(t)$. By condiction (3b) of Theorem \[t:TL\], there exists a finite set $F\subset {\omega}$ such that $c{\cdot}s_n\ne cb{\cdot}t_m$ and hence $s_n\ne b{\cdot}t_m$ for any $n,m\in{\omega}\setminus F$.
A topology on a semilattice $X$ is called [*Lawson*]{} if it has a base consisting of open subsemilattices.
There exists a countable semilattice $X$ and a function $\ell:\{s,t\}\to X$ defined on a subset $\{s,t\}\subset X^{\omega}$ such that
1. The semilattice $X$ admits a shift-continuous metrizable topology $\tau$ in which the sequence $s$ converges to $\ell(s)$ and the sequence $t$ converges to $\ell(t)$.
2. The semilattice $X$ admits no Lawson Hausdorff topology $\tau$ in which the sequence $s$ converges to $\ell(s)$ and the sequence $t$ converges to $\ell(t)$.
By [@CHK 2.21], there exists a compact metrizable topological semilattice $K$ containing two points $x,y\in K$ such that for any neighborhoods $O_x,O_y\subset K$ of the points $x,y$ there exists a finite subset $F\subset O_x$ such that $\inf F\in O_y$. Fix countable neighborhood bases $\{V_n\}_{n\in{\omega}}$ and $\{W_n\}_{n\in{\omega}}$ at the points $x,y$, respectively. For every $n\in{\omega}$ choose a finite subset $F_n\subset \bigcap_{k\le n}V_k$ such that $\inf F_n\in \bigcap_{k\le n}W_k$. Let $s\in X^{\omega}$ be a sequence such that $s_n=\inf F_n$ for every $n\in{\omega}$ and $t\in X^{\omega}$ be a sequence such that $F_n=\{t_k:\sum_{i<n}|F_i|<k\le\sum_{i\le n}|F_i|\}$ for every $n\in{\omega}$. Let $\ell(s)=y$ and $\ell(t)=x$.
Let $X$ be the countable semilattice generated by the countable set $\{x,y\}\cup s[{\omega}]\cup t[{\omega}]$. The metrizable topology on $X$ inherited from $K$ witnesses that the condition (1) is satisfied.
It remains to prove that $X$ admits no Lawson Hausdorff topology $\tau$ in which the sequence $s$ converges to $\ell(s)=y$ and the sequence $t$ converges to $\ell(t)=x$. To derive a contradiction, assume that such topology $\tau$ exists. Then the points $x,y$ have disjoint open neighborhoods $O_x,O_y\in\tau$ such that $O_x$ is a subsemilattice of $X$. Since the sequences $s$ and $t$ converge to $y$ and $x$, respectively, there exists $n\in{\omega}$ such that $s_k\in O_y$ and $t_k\in O_x$ for all $k\ge n$. Then $F_n\subset\{t_k\}_{k\ge n}\subset O_x$ and hence $\inf F_n\in O_x$ (as $O_x$ is a subsemilattice of $X$). On the other hand, $\inf F_n=s_n\in O_y$. Then $\inf F_n\in O_x\cap O_y$, which contradicts the choice of the neighborhoods $O_x$ and $O_y$.
The example {#s:Ex}
===========
In this section we shall apply Theorem \[t:TL\] to construct an example of a metrizable semitopological semilattice with dense non-closed partial order.
Consider the semilattice $\{0,1,2\}$ endowed with the operation of taking minimum. In the semilattice $\{0,1,2\}^{\omega}$ consider the countable subsemilattice $X$ consisting of functions $f:{\omega}\to\{0,1,2\}$ having non-empty finite support ${\mathrm{supp}}(f):=f^{-1}(\{0,1\})$.
It is easy to see that the partial order on $\{0,1,2\}$ induced by the semilattice operation (of minimum) coincides with the usual linear order on $\{0,1,2\}$. Then the semilattice operation (of coordinatewise minimum) on $X\subset \{0,1,2\}^{\omega}$ induces the natural partial order on $X$.
For every $n\in{\omega}$ consider the functions $\mathbf 0_n,\mathbf 1_n\in X$ defined by $$\mathbf 0_n(i)=\begin{cases}0&\mbox{if $i=n$}\\
2&\mbox{otherwise}
\end{cases}\quad \mbox{ and }\quad
\mathbf 1_n(i)=\begin{cases}1&\mbox{if $i=n$}\\
2&\mbox{otherwise}.
\end{cases}$$ It is clear that $\mathbf 0_n{\cdot}\mathbf 1_n=\mathbf 0_n$ and hence $\mathbf 0_n\le \mathbf 1_n$ for all $n\in{\omega}$.
\[t:main\] The semilattice $X$ admits a metrizable shift-continuous topology $\tau$ such that the set $\{(\mathbf 0_n,\mathbf 1_n):n\in\mathbb N\}$ is dense in the square $X\times X$ of the semitopological semilattice $(X,\tau)$. Since $$\{(\mathbf 0_n,\mathbf 1_n)\}_{n\in{\omega}}\subset P:=\{(x,y)\in X\times X:xy=x\}\ne X\times X$$ the partial order $P$ of $X$ is dense and non-closed in $X\times X$.
Write the countable set $X\times X$ as $X\times X=\{(x_k,y_k):k\in{\omega}\}$. For every $k,n\in{\omega}$ consider the elements $z_{k,n}=\mathbf 0_{2^k3^n}$ and $u_{k,n}:=\mathbf 1_{2^k3^n}$ of the set $X$. These elements form sequences $\vec z_k=(z_{k,n})_{n\in{\omega}}$ and $\vec u_k=(u_{k,n})_{n\in{\omega}}$, which are elements of the set $X^{\omega}$. Let ${\mathrm{dom}}(\ell)=\{\vec z_k,\vec u_k:k\in{\omega}\}$ and $\ell:{\mathrm{dom}}(\ell)\to X$ be the function defined by $\ell(\vec z_k)=x_k$ and $\ell(\vec u_k)=y_k$ for $k\in{\omega}$.
We claim that for the function $\ell$ the condition (3) of Theorem \[t:TL\] is satisfied. In fact, the condition (3a) is satisfied in the stronger form: for any $s\in {\mathrm{dom}}(\ell)$ and $a,b\in X$ the set $\{n\in{\omega}:a{\cdot} s_n=b\}\subset \{n\in{\omega}:\exists k\in{\omega}$ with $2^k3^n\in {\mathrm{supp}}(b)\}$ is finite.
Now we check the condition (3b). Fix elements $a,b\in X$ and sequences $s,t\in {\mathrm{dom}}(\lambda)$ such that $a{\cdot}\ell(s)\ne b{\cdot}\ell(t)$. It is easy to see that $a{\cdot}s_n\ne b{\cdot}t_m$ for any $n,m\in{\omega}\setminus F$ where $F=\{n\in{\omega}:\exists k\in{\omega}$ such that $2^k3^n\in {\mathrm{supp}}(a)\cup{\mathrm{supp}}(b)\}$.
By Theorem \[t:TL\], the semilattice $X$ admits a metrizable shift-continuous topology $\tau$ in which every sequence $s\in{\mathrm{dom}}(\ell)$ converges to $\ell(s)$. In particular, for every $k\in{\omega}$ the sequence $(\mathbf 0_{2^k3^n})_{n\in{\omega}}=\vec z_k$ converges to $x_k=\ell(\vec z_k)$ and the sequence $(\mathbf 1_{2^k3^n})_{n\in{\omega}}=\vec z_k$ converges to $y_k=\ell(\vec u_k)$. Consequently, the set $\{(\mathbf 0_{2^k3^n},\mathbf 1_{2^k3^n}):k,n\in{\omega}\}\subset \{(\mathbf 0_m,\mathbf 1_m):m\in{\omega}\}$ is dense in $X\times X=\{(x_k,y_k):k\in{\omega}\}$.
Since $$\{(\mathbf 0_n,\mathbf 1_n)\}_{n\in{\omega}}\subset P:=\{(x,y)\in X\times X:xy=x\}\ne X\times X,$$ the partial order $P$ is a dense non-closed subset of $X\times X$.
Does the semilattice $X$ admit a Lawson Hausdorff shift-continuous topology such that the partial order $P=\{(x,y)\in X\times X:xy=x\}$ is not closed (and dense) in $X\times X$?
T. Banakh, S. Bardyla, [*Complete topologized posets and semilattices*]{}, preprint(https://arxiv.org/abs/1806.02869).
T. Banakh, I. Protasov, O. Sipacheva, [*Topologization of sets endowed with an action of a monoid*]{}, Topology Appl. [**169**]{} (2014) 161–174.
J.H. Carruth, J.A. Hildebrant, R.J. Koch, [*The theory of topological semigroups*]{}, Vol. 2. Monographs and Textbooks in Pure and Applied Mathematics, 100. Marcel Dekker, Inc., New York, 1986.
R. Engelking, [*General topology*]{}, Heldermann Verlag, Berlin, 1989.
G. Gierz, K.H. Hofmann, K. Keimel, J.D. Lawson, M. Mislove, D.S. Scott, [*Continuous lattices and domains*]{}, Cambridge University Press, Cambridge, 2003.
W. Just, M. Weese, [*Discovering modern set theory. II. Set-theoretic tools for every mathematician*]{}, Graduate Studies in Mathematics, 18. American Mathematical Society, Providence, RI, 1997.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Scaling properties of Yang-Mills fields are used to show that fractal structures are expected to be present in system described by those theories. We show that the fractal structure leads to recurrence formulas that allow the determination of non perturbative effective coupling. Fractal structures also cause the emergence of non extensivity in the system, which can be described by Tsallis statistics. The entropic index present in this statistics is obtained in terms of the field theory parameters. We apply the theory for QCD, and obtain the entropic index value, which is in good agreement with values obtained from experimental data. The Haussdorf dimension is calculated in terms of the entropic index, and the result for hadronic systems is in good agreement with the fractal dimension accessed by intermittency analysis of high energy collision data. The fractal dimension allow us to calculate the behavior of the particle multiplicity with the collision energy, showing again good agreement with data.'
author:
- 'Airton Deppman$^{1,2}$, Eugenio Megías$^{2}$, Débora P. Menezes$^{3}$'
date: |
1- Instituto de Física - Universidade de São Paulo\
email: [email protected] ;\
2- Departamento de F[í]{}sica Atómica, Molecular y Nuclear and Instituto Carlos I de Física Teórica y Computacional - Universidad de Granada;\
3- Departamento de Física - Universidade Federal de Santa Catarina.
title: 'Fractal structure in Yang-Mills fields and non extensivity'
---
Main
====
Fractals are complex systems with internal fine structure featuring scale invariance and self-similarity. Fractal measures, contrary to more conventional quantities for which an increase in resolution results in the same measured value with increased precision, yield different values for different resolutions. A classical example is the lengths of coastlines [@Mandelbrot], which increase significantly if measured with smaller scales: if measured in centimeters, the value is quite different from the one obtained if measure in kilometers. The concept of fractal has found applications in Mathematics, Biology and Physics, among other areas. It has been associated to socioeconomics evolution of cities, as well as to the actual configuration of urban area infrastructure [@West]. It can emerge from the system structure or from its dynamical evolution. In geometry, it introduces a new kind of topology and renders beautiful pictures. An interesting introduction to the subject can be found in Ref. [@Falconer].
Scale invariance is a fundamental aspect of the Yang-Mills field (YMF) theory, playing an important role in the renormalization of the theory after divergences are cut [@Dyson1; @Ward; @Gell-Mann_Low]. This is a prototype theory for the description of three among the four known interactions: electromagnetic, weak and strong. The Standard Model is the most complete implementation of the theory so far, encompassing strong, weak and electromagnetic forces found in Ref. [@YMF-60y]. In the case of Quantum Chromodynamics (QCD), used to describe the strong interaction, the scale invariance is expressed through the Callan-Symanzik equation [@Callan; @Symanzik1; @Symanzik2], which is a basic tool to show that QCD is an asymptotic free interaction [@Politzer1; @Politzer2; @GrossWilczek1; @GrossWilczek2]. This aspect of QCD is of fundamental importance in particle physics, since it means that for processes with high momentum transfer, as in deep inelastic scattering [@DIS], quarks are almost free and the scattering can be studied in low-order perturbative calculation. Thus, scale invariance means that the coupling constant, which represents the strength of the interaction, depends on the energy scale for the interacting particles. Scale dependent couplings are called effective couplings or running couplings.
In this work we show that fractal structures can be formed by systems described by the Yang-Mills field theory. The presence of fractal structures lead to a recurrence formula that allows the determination of the effective coupling even in high perturbative orders. A consequence of the fractal structure is that the proper thermodynamical theory for describing the interacting system is the non extensive Tsallis statistics [@Tsallis1988; @TsallisBook], rather than the traditional Boltzmann-Gibbs statistics, where the entropic index, q, is a measure of the non additivity of the entropy. Here $q$ is obtained, for the first time, in terms of the field theory parameters. The fractal dimension is determined as a function of the entropic index. When applied to QCD in the asymptotic approximation, the theoretical valued obtained for $q$ is in good agreement with the values found in the analyses of experimental data. We show that various experimental features can be explained by the theoretical results derived here, in particular the behavior of particle multiplicity as a function of the collision energy, which depends on the fractal dimension.
Although fractal structure could be perceived in any field theory of Yang-Mills type, in QCD its effects are more evident. No wonder that hints of fractal structure were present in theories proposed already in the 1960’s. About fifty years ago, Rolf Hagedorn proposed a thermodynamical description of high energy collisions based on a self-consistent principle [@Hagedorn0; @Hagedorn65] that states that fireballs, the transient hot system formed in those collisions, is made of ... fireballs! Although the definition of fireballs is based on the concept of fireball itself, Hagedorn was able to obtain a complete thermodynamic physical description for this system. The resemblance between the definition of fireball and that of fractals is clear, and we suppose that, were the concept of fractals already known at that time, Hagedorn would have used it for his definition of fireball. A little later, Chew and Frautschi proposed the bootstrap model for hadrons [@Frautschi; @Chew], where it is supposed that hadrons are made of hadrons, showing again the similitude to fractals. Among the predictions of these theories are the existence of a limiting temperature, known as Hagedorn temperature; a formula for the mass spectrum of hadrons ; an exponential distribution of energy and momentum of the particles. These phenomenological theories obtained great success in the first years after their proposals. An entire new line of research emerged from these results, and many Hadron Resonance Gas Models are still used today [@Petreczky_HRG; @Megias_HRG; @Venugopalan_HRG]. Despite the initial success, after the energy available for collisions in accelerators increased, it was soon realized that the exponential distribution predicted by the theory was in disagreement with experiments. With the success of QCD, those phenomenological approaches were, to some extent, dismissed. As is shown in the present work, our results reconcile Hagedorn theory with QCD.
Our starting point is the scaling property of YMF, which is expressed in terms of amplitudes or vertex functions, $\Gamma(p,m,g)$, as $$\Gamma(p,m,g)=\lambda^{-D} \Gamma(p,\bar{m},\bar{g}) \,,$$ where $p,m,g$ are, respectively, the momentum, mass and coupling constant associated to a state in a non interacting configuration, as for instance, for systems that are so apart of each other that interaction is null in all practical aspects. The effective mass and effective coupling, $\bar{m}$ and $\bar{g}$ respectively, are obtained when interaction is considered. A direct consequence of the equation above is the renormalization group equation $$\left[M\frac{\partial}{\partial M} + \beta_g \frac{\partial}{\partial \bar{g}} + \gamma \right]\Gamma=0 \label{CallanSymanzik} \,,$$ where $$\beta_g=M\frac{\partial \bar{g}}{\partial M}$$ is known as beta function and gives the logarithmic ratio in which the coupling varies with the scale, and $\gamma$ is related to the change of scale of the interacting fields.
Consider an initial free state $\ket{\Psi_o}$. Its time dependent evolution is given by $$\ket{\Psi}= e^{-iHt} \ket{\Psi_o}\, \label{Gamma}$$ where $H$ is the Hamiltonian operator. In the irreducible representation, each interaction is represented graphically by a proper vertex, while lines represent effective partons for which self-interaction is included through an effective mass. In the following, parton means effective parton, and interaction means proper vertex.
We introduce new states, $\ket{\Psi_n}$, such that $$\ket{\Psi_n} = (-i)^n \int dt_n \dots dt_1 g e^{-iH_o(t_n-t_{n-1})} g e^{-iH_o(t_{n-1}-t_{n-2})} g \dots e^{-iH_o(t_1-t_o)} \ket{\Psi_o}$$ and $$\ket{\Psi}=\sum_{\{n\}} \braket{\Psi_n|\Psi} \ket{\Psi_n}\,.$$ These new states are those with a well defined number of vertexes, or interactions. The sum in equation above is taken over all possible configurations with $n$ vertexes. The number of particles created is not necessarily equal to $n$, since some of the particles are created at high orders of perturbative calculation. So we introduce the states $\ket{\psi_N}$ with well defined number of particles created, $N$. These states are $$\ket{\Psi_n}=\sum_{N} \braket{\psi_N|\Psi_n} \ket{\psi_N}\,.$$ The states $\ket{\psi_N}$ are autovectors of $H_o$ with fixed number of particles, $N$. Of course $\braket{\Psi_n|\psi_N}=0$ whenever $N > M(n)$, and $\braket{\Psi_{N'}|\Psi_N}=\delta_{N'N}$. Since the number of partons is fixed and they do not interact but by contact interaction, the states $\ket{\psi_N}$ can be understood as the states of an ideal gas of $N$ partons. Therefore $$\ket{\psi_N}={\cal S} \ket{\gamma_1,m_1,p_1, \dots , \gamma_N,m_N,p_N}\,,$$ where $m_i$ and $p_i$ are the mass and momentum of the $i$ partonic state, and $\gamma_i$ represents all relevant quantum numbers necessary to completely characterize the partonic state. ${\cal S}$ is the symmetrization operator acting over fermions and bosons. In the case of hot systems the momentum of the effective partons varies continuously, ${\cal S}$ gives a negligible modification of the single parton states, so mass and momentum of each parton can vary independently, as far as the total energy is conserved.
As a consequence of the large number of possible ways to obtain $N$ particles in the final state, the actual process that leads to a particular state $\ket{\psi_N}$ is not relevant, and statistical methods can be applied. For instance, the probability to find a state where at least one parton has mass between $m_o$ and $m_o+dm_o$, and momentum coordinates between $p_{oi}$ and $p_{oi}+dp_{oi}$, is given by matrix elements of the type $$P(\varepsilon_o)=\braket{\gamma_o,m_o,p_o, \dots|\Psi(t)}\,,$$ where $\varepsilon_o=p_o^0$ is the energy of one parton in a system with an arbitrary number of partons. Initially, we assume the particles form an ideal gas and ignore that they have a fractal structure. We also assume, for reasons that will become clear below, that the energy of the $N$ particle system may fluctuate according to a probability density $P(E)$. In this case we can show that $$ P(\varepsilon_o)= \sum_n \sum_{N} G^n
\left(\frac{N}{n(\tilde{N}-1)}\right)^4 \left(1+\frac{\varepsilon_o }{E}\right)^{-(4N-5)} P(E) \,, \label{probdensideal}$$
The assumption that the particles have no internal structure and that the system behaves like an ideal gas is not strictly valid, since in the actual case particles present a fine structure with scaling and self-similarity, that is, they have a [*fractal structure*]{} (see Figure \[fractalstructure\]). The internal degrees of freedom share part of the total energy available, and thus distort the distribution of energy. Our next step is to include the effects of the fractal structure into the distribution obtained above. The probability density $P_N$, when it is written in terms of scale independent quantities, must be identical for all effective partons, since these systems are self-similar objects, therefore $P_N$ must be independent of the number of particles, $N$, and on the order of perturbative calculation, $n$. The scaling property also imposes that the ratio between the energy $\varepsilon_o$ of a parton and that of its parent parton, $E$, must be the same for all partons, independently of $N$ and $n$. Therefore we can write $$\frac{\varepsilon_o}{E} \sim \frac{E}{{\cal M}}\equiv\frac{\varepsilon}{\Lambda} \,,$$ with $\varepsilon/\Lambda$ being independent of the level in the fractal structure. The self-similarity among the partons implies that the probability that the parent parton with mass $E$ inside a larger system with mass ${\cal M} $ is similar to the probability given in Eq. (\[probdensideal\]), as represented in Figure \[fractalstructure\](a), so we write $$P\left(E\right)=\left(1+\frac{E}{{\cal M}}\right)^{-\alpha} \,, \label{power-law}$$ where, $\alpha$ is the number of degrees of freedom of the fractal structure in the system with energy ${\cal M}$. Since part of the fractal structure corresponds to the $N$ particles we are considering apart, only a fraction $\nu$ of the degrees of freedom that remains to be added to the fractal structure , as schematically shown in Fig \[fractalstructure\](b). Then we can write for the probability density $$ P(\varepsilon_o)= \sum_n \sum_{N} G^n
\left(\frac{N}{n(\tilde{N}-1)}\right)^4 \left(1+\frac{\varepsilon_o }{E}\right)^{-(4N-5)} \left[P(E)\right]^{\nu} \,, \label{probdensfractal}$$ which includes some characteristics of the ideal gas behavior, but takes into account fractal degrees of freedom.
![(a) Scaling property of Yang-Mills fields as observed in diagrammatic representation. Loops and partons at different perturbative orders are equal, provided the appropriate scales are considered. (b) Schematic view of the fractal structure. The initial parton (first black line form the left) may be considered as a constituent of another parton (first red line from the left). At the vertexes highlighted by blue circles, the distribution of energy and momentum of the generated partons are determined. The region in black lines was considered in the text as an ideal gas with $N$ particles.[]{data-label="fractalstructure"}](ScalingDiags2.pdf)
![(a) Scaling property of Yang-Mills fields as observed in diagrammatic representation. Loops and partons at different perturbative orders are equal, provided the appropriate scales are considered. (b) Schematic view of the fractal structure. The initial parton (first black line form the left) may be considered as a constituent of another parton (first red line from the left). At the vertexes highlighted by blue circles, the distribution of energy and momentum of the generated partons are determined. The region in black lines was considered in the text as an ideal gas with $N$ particles.[]{data-label="fractalstructure"}](VertexScaling.pdf)
Substituting Eq (\[power-law\]) into Eq. (\[probdensideal\]), and using self-similarity of fractals to identify $P(\varepsilon_o)=P(E)$, it results $$(4N-5)+\alpha \nu=\alpha \,,$$ where the parameter $\nu \leq 1$ represents the fraction of the total number of degrees of freedom of the state $\ket{\psi_N}$, when it is considered as a constituent of a larger system. We can introduce the parameters $q$ such that $$\frac{1}{q-1}=\frac{4N-5}{1-\nu}\,, \label{qvalue}$$ with $q>1$, and $\Lambda$ such that $$\Lambda=\frac{4N-5}{1-\nu} \lambda\,, \label{resolution}$$ where $\lambda$ is a reduced scale, independent of the number of degrees of freedom relevant to the system, which gives the energy per degree of freedom and therefore can be interpreted as a kind of temperature. In terms of the new parameters we obtain $$P(\varepsilon/\lambda)=\left[1+(q-1)\frac{\varepsilon}{\lambda}\right]^{-\frac{1}{q-1}}\,. \label{Tdistribution}$$
This result shows that the distribution of parton energy in a system governed by Yang-Mills fields depends only on the ratio between the parton energy, $\varepsilon$, and the energy scale per degree of freedom $\lambda$. Furthermore, the right hand side of Eq. (\[Tdistribution\]) is exactly the q-exponential distribution commonly found in Tsallis non extensive statistics. The non additive entropy as the basis for a non extensive statistics was proposed by Tsallis in the late 80’s [@Tsallis1988; @TsallisBook] and its effects have been explored since then, but they are not completely understood yet.
The $q$-deformed entropy functional that underlines non extensive statistics depends on a real parameter, $q$, that determines the degree of nonadditivity of the functional, and in the limit $q \rightarrow 1$ it becomes additive and the standard Boltzmann-Gibbs entropy is recovered. Similar results have been obtained through a different approach using the concept of thermofractals, introduced in Ref. [@Deppman2016] and studied in details in Ref. [@DFMM]. There it is shown that the fractal structure leads to the non extensive statistics, and it is discussed the relations between thermofractals and Hagedorn’s self-consistent thermodynamics developed to study high energy collisions [@Hagedorn0; @Hagedorn65], and that was extended to non extensive statistics [@Deppman2012]. Thus the fact that the non extensive statistics is obtained from considerations of the scaling properties and self-similarity in Yang-Mills fields, shows that Tsallis statistics should have been used in Hagedorn’s theory, instead of Boltzmann-Gibbs statistics, and this is the reason Hagedorn’s theory fails when confronted with high energy colisions data. Indeed, when the self-consistent thermodynamics is obtained by using the non extensive statistics it can describe more accurately the experimental observations of hadronic systems [@Deppman2012]. Our results show that Tsallis statistics emerges in YMF from the fact that the fields can be quantized into partons only if these partons present a complex internal structure similar to fractals. The meaning of the entropic index, in this case, is clear: Eq (\[qvalue\]) shows that $(q-1)^{-1}$ gives the number of degrees of freedom relevant in the system. Observe that as the number of degrees of freedom increases, $q \rightarrow 1$ and Boltzmann statistics is recovered as a limiting case. However, we will see that $q$ can be calculated from the field theory parameters, and that for QCD the number of degrees of freedom is finite, so Tsallis statistics must be used.
The role played by the q-exponential in Eq. (\[Tdistribution\]) is that of determining, stochastically, the values of effective masses and momenta of the particles generated at each vertex (see Fig \[fractalstructure\]b). Therefore, the effective vertex should contain a term which is given by that q-exponential function, that is $$g=\prod_{i=1}^{\tilde{N}}G\left[1+(q-1)\frac{\varepsilon_o}{\lambda}\right]^{-1/(q-1)}\,, \label{coupling}$$ we can compare the behavior of the effective coupling given above with the expected behavior of the QCD coupling as calculated in 1-loop approximation, where perturbative methods can be used to obtain the $\beta$-function. From Eq. (\[coupling\]) we get $$\beta_{\bar{g}}=-\frac{1}{16\pi^2} \frac{1}{q-1} g^{\tilde{N}+1}\,, \label{betaFractal}$$ and we emphasize that $\tilde{N}=2$. The beta-function for QCD is [@Politzer1] $$\beta_{{{\textrm{\scriptsize QCD}}}}=- \frac{g^3}{16\pi^2} \left[\frac{11}{3}c_1-\frac{4}{3}c_2\right]\,, \label{betaQCD}$$ where the parameters $c_1$ and $c_2$ are related to the number of colors and flavors by $c_1=N_c$ and $c_2=N_f/2$. Using $N_c=N_f/2=3$ it results $$\frac{11}{3}c_1-\frac{4}{3}c_2=7\,,$$ which, by comparison of Eqs. (\[betaFractal\]) and (\[betaQCD\]), leads to $q=1.14$. From experimental data analysis[@Cleymans; @WilkWlodarczyk; @Lucas1; @Lucas2; @Sena] $q=1.14 \pm 0.01$, showing good agreement between theory and experiments. The agreement found between experimental and theoretical values for $q$ shows that the effective coupling obtained from considerations on the fractal structure of QCD is good agreement with the accumulated knowledge about strong interaction. The advantage of the effective coupling described by Eq. (\[coupling\]) is that it may be used in non perturbative calculations. Furthermore, it represents a reconciliation between QCD and Hagedorn’s self-consistent approach.
The results obtained here have shown that a system with fractal structure, similar to the thermofractals [@Deppman2016], can be understood as a natural consequence of the scale invariance of gauge field theories. We give solid grounds for phenomenological approaches that have been used to describe hadron mass spectrum [@Lucas1] and multiparticle production with non extensive statistics [@Cleymans; @WW2018; @Lucas1; @Lucas2; @Sena; @WilkWlodarczyk; @DeBhaskar; @Parvan; @Ishihara; @Tawfik; @Rybczynski], which can explain the long tail distribution observed in multiparticle production. From the recurrence method used here we understand the reason why, even at small order of perturbative QCD calculation, it is possible to describe correctly the transversal momentum distributions measured at high energies by adopting Tsallis distribution [@Wong]. The fact that an statistical interpretation of the field theory is possible, allows us to understand why the non extensive self-consistent theory is in good agreement with Lattice QCD calculations [@Deppman2014].
It is possible to understand, from the considerations made here, that the fractal structure of YMF is the basis for investigations of hadron properties [@PedroCardoso], phase-transition in hot hadronic matter [@Megias], neutron stars [@Debora] and cosmic-ray [@BeckCosmicRay]. These phenomenological approaches are, in fact, implementations of scaling symmetries observed in Yang-Mills fields. The fractal structure also allows the understanding of the self-similarity [@WWselfsymmetry; @Tokarev; @Zborovsky] and scaling properties observed in high energy experimental data. In fact, these findings are direct consequences of the scaling properties of YMF, as discussed here. Moreover, the fact that the entropic index, $q$, is obtained from well-known field-theoretical parameters, the results we have obtained allows a new interpretation of Tsallis statistics in terms of fractal structure in the same lines it was obtained in thermofractals approach [@Deppman_Universe].
The fractal structure presents at least one fractal dimension, the Haussdorf dimension, which was calculated for the case of thermofractals in Ref. [@Deppman2016]. The Haussdorf dimension can be calculated by using the box-counting technique [@Falconer], where the dimension $D$ is related to the number of boxes, ${\cal N}$, necessary to completely cover all possible values for the measured quantity, whatever it is, and $D_t$ is the topological dimension, that is, the dimension expected for some quantity when there is no fractal dimension. At some scale $r$ these quantities are related by[@Falconer] $${\cal N} r^{-D} \propto r^{-D_t}\,.$$ In our case, $D_t=1$ is the dimension that describes how the total energy varies when the energy unit, $r$ is modified, and ${\cal N}=\tilde{N}^n$, where $n$ is the layer in the fractal structure where the partons have energies of the order of $\lambda$. $D$ is how the energy of the system components varies with $r$. We determine $D$ by noticing that the average parton energy at a scale $\lambda$ is $$\braket{\varepsilon}=\frac{\lambda}{2q-1}\,.$$ Then, it follows that the fractal dimension is $$D-1=\frac{\log \tilde{N}}{\log R}\,,$$ where $R=(q-1)/(2q-1)$ is the ratio between the average energy of the constituent fractal and the energy of its parent system. Using $q=1.14$ and $\tilde{N}=2$, it results $D=0.69$, in good agreement with findings from intermittency analysis of high energy collision data [@Bialas_Peschanski; @Bialas_Peschanski2; @Hwa; @HwaPan; @Hegyi1; @DreminHwa; @Hegyi2; @Antoniou]. The fractal dimension gives the behavior of the parton energy with the energy scale, $r$, that is, while the total energy goes as $E \propto r^{-1}$, the partons observed at scale $\lambda$ have energies that depend on the scale as $\varepsilon \propto r^{-D}$. A more direct way to access the fractal dimension is the particle multiplicity. In fact, being $\cal{M}$ the particle multiplicity, we have $${\cal M} \braket{\varepsilon}=E \,.$$ From the dimensional behavior obtained above, we get $${\cal M}=E^{1-D}\,.$$ For the case of hadrons, as we have seen, $q=1.14$ and $D=0.69$, so we obtain ${\cal M}\propto E^{0.31}$, which is in excellent agreement with the result obtained for $pp$ collision at high energy [@Sarkisyan_Multiplicity], which gives, for a power-law fit, an exponent corresponding to $1-D=0.302$.
In conclusion, we used scaling properties of Yang-Mills fields to show that fractal structures are expected to be formed in systems described by that theory. These structures lead to a thermodynamic description of the fields that follows Tsallis statistics, with the entropic index, $q$, being for the first completely determined in terms of the field theory parameters. We discuss that Hagedorn’s self-consistent theory fails because the non extensive effects were not considered, and verify that our results confirm the phenomenological extension of Hagedorn theory by adopting Tsallis statistics, which was proposed in previous works in the context of thermofractals.
Another consequence of the fractal structure is that we obtain a recurrence formula which allows the determination of the effective coupling in terms of particle momentum and of a scale. The fractal dimension is also obtained in terms of $q$, and therefore also in terms of the field theory parameters. We use QCD as an exemple of application of the theory. The beta function describing the behavior of the effective coupling can be obtained for QCD in 1-loop approximation, due to asymptotic freedom, and in such approximation we compare the results of our theory with the begavior expected for perturbative QCD, obtaining a good agreement. As a result, we obtain $q=1.14$, which is in good agreement with the value obtained from the analysis of experimental data from high energy collisions.
Haussdorff fractal dimension is calculated in terms of $q$, and for QCD results in $D=0.69$, which is a value in agreement with those obtained from intermittency analyses for $pp$ collisions. The fractal dimension is a manufestation of non extensivity which appears clearly in the behavior of the particle multiplicity as a function of the collisions energy. We show that multiplicity increases as a power-law of the collision enegly with exponent $1-D$, which gives 0.31 for QCD, again in good agreement with experimental data.
With the theory developed here, we explain self-similarity effects observed in multi-particle production in high energy collisions, as well as scaling properties of distributions obtained experimentally. Our results give a strong basis for phenomenological studies that have been used so far to explain experimental data from high energy collisions, as well as for phenomenological models that have been used to describe hadronic systems, such as nucleon structure and neutron stars. It also allows for a better understanding of the origins of non extensivity in high energy collisions and on the role of the entropic parameter, which is related here to the number of relevant degrees of freedom of the fractal structure.
Method
======
Consider that at an initial instant $t_0$ we have the system in an initial state $\ket{\Psi_o}$ which can be considered [*free*]{} in the sense that no interaction can take place. The time-dependent states $\Psi$ can be written as $$\ket{\Psi}=e^{-iHt}\ket{\Psi_o}\,,$$ where $H$ is the Hamiltonian operator. If $H_o$ is the Hamiltonian for free effective partons, for which self-interaction is already included by considering that the parton has an effective mass and interacts according to an effective coupling [@Dyson1; @Gell-Mann_Low]. The state $\ket{\Psi}$ can be written as $$\ket{\Psi}=\sum_{\{n\}} \braket{\Psi_n|\Psi} \ket{\Psi_n}\,,$$ where $\sum_{\{n\}}$ represents the sum over all possible irreducible graphs with $n$ interactions with the states $\ket{\Psi_n}$ corresponding to a fixed number, $n$, of interactions in the vertex function, such that $$\ket{\Psi_n} = (-i)^n \int dt_n \dots dt_1 g e^{-iH_o(t_n-t_{n-1})} g e^{-iH_o(t_{n-1}-t_{n-2})} \dots g e^{-iH_o(t_1-t_o)} \ket{\Psi_o}$$ with $t_n>t_{n-1}>\dots>t_1$. Of course these states satisfy the relation $\braket{\Psi_{n'}|\Psi_n}=\delta_{n'n}$.
The number of particles in the state $\ket{\Psi_n}$ is not directly related to $n$, since high order contributions to the $N$ particles states can be important, but certainly $N \le M(n):=n(\tilde{N}-1)+1$, where $\tilde{N}$ is the number of particles created or annihilated at each interaction. In Yang-Mills field theory $\tilde{N}=2$.[^1] We can introduce states of well defined number of effective partons, $\ket{\psi_N}$, so that $$\ket{\Psi_n}=\sum_{N} \braket{\psi_N|\Psi_n} \ket{\psi_N}\,.$$ The states $\ket{\psi_N}$ are autovectors of $H_o$ with fixed number of particles, $N$. Of course $\braket{\Psi_n|\psi_N}=0$ whenever $N > M(n)$, and $\braket{\Psi_{N'}|\Psi_N}=\delta_{N'N}$. Since the number of partons is fixed and they do not interact but by contact interaction, the states $\ket{\psi_N}$ can be understood as the states of an ideal gas of $N$ partons. Therefore $$\ket{\psi_N}={\cal S} \ket{\gamma_1,m_1,p_1, \dots , \gamma_N,m_N,p_N}\,,$$ where $m_i$ and $p_i$ are the mass and momentum of the $i$ partonic state, and $\gamma_i$ represents all relevant quantum numbers necessary to completely characterize the partonic state. ${\cal S}$ is the symmetrization operator acting over fermions and bosons. Since the mass of the effective partons varies continuously, ${\cal S}$ gives a negligible modification of the single parton states, so mass and momentum of each parton can vary independently, as far as the total energy is conserved.
Assuming that a statistical approach can be used, the probability to find a particle with quantum numbers, mass and momentum given, respectivelly, by $\gamma_o$, $m_o$ and $p_o$ is $$\braket{\gamma_o,m_o,p_o, \dots|\Psi(t)}=\sum_n \sum_{N} \braket{\Psi_n|\Psi(t)} \braket{\psi_N|\Psi_n} \braket{\gamma_o,m_o,p_o,\dots|\psi_N}\,. \label{1partonprob1}$$
The first bracket depends on the intensity of the interaction, determined by a coupling constant $G$, so $$\braket{\Psi_n|\Psi}=G^n P(E) dE\,, \label{braket1}$$ where $P(E) dE$ is the probability to have the $N$ particle system with energy between $E$ and $E+dE$. The second bracket depends on the relative number of possibilities to get the configuration with $N$ particles, so $$\braket{\psi_N|\Psi_n} = C_N(n)\, \label{braket2}$$ with $$\sum_{n} C_N(n)=1 \,,$$ with $n \ge N/(\tilde{N}-1)$. The last bracket in the expression above is calculated statistically by supposing that all possible configuration of the system have the same probability, and by counting the number of configurations with energy between $E$ and $E+dE$ and one particle with quantum numbers $\gamma_o$, effective mass $m_o$ and momentum $p_o$. . The result is $$f(p_o) d^4p_o = A(N) P_N\left( \frac{\varepsilon_o}{E} \right) d^4\left( \frac{p_o}{M} \right) \,,$$ where $$P_N\left( \frac{\varepsilon_o}{E} \right) = \left( 1 - \frac{\varepsilon_o}{E} \right)^{4N-5}\,, \label{braket3}$$ with $$A(N)=\frac{\Gamma(4N)}{8\pi\Gamma(4(N-1))}\,, \label{normalconst}$$ and $\varepsilon_o=p_o^0$ being the the particle energy. In most cases, including the cases of interest here, $\varepsilon \ll E$. Noticing that $$\left(1-\frac{\varepsilon_o }{E}\right)^{4N-5}= \left(1+\frac{\varepsilon_o }{E}\right)^{-(4N-5)} \left(1-\frac{\varepsilon_o^2 }{E^2}\right)^{4N-5}\,,$$ and observing that the last term in the right-hand side is approximately the unit, we can write $$\left(1-\frac{\varepsilon_o }{E}\right)^{4N-5}= \left(1+\frac{\varepsilon_o }{E}\right)^{-(4N-5)}\,. \label{chgsignals}$$
Observe that $A(N) \propto (4N)^4$ for $N$ sufficiently larger than unit, so the number of possible configurations increase fast with $N$. The maximum number of particle for a given number of interactions, $n$, is $M(n)=n(\tilde{N}-1)$, so we can consider the probability for $N$ particles generated in $n$ interactions as $$C_N(n) \propto \left(\frac{N}{n(\tilde{N}-1)}\right)^4\,. \label{Ndependence}$$
Substituting Eqs. (\[braket1\]), (\[braket2\]), (\[braket3\]), (\[chgsignals\]) and (\[Ndependence\]) into Eq. (\[1partonprob1\]), we obtain Eq. (\[probdensideal\]). This result shows that for an ideal gas with finite number of particles, the probability depends on a power-law function of the ratio $\varepsilon_j/E$.
[**Acknowledgments:**]{} A.D. would like to thank the University of Granada, where part of this work has been done, for the hospitality and financial support under a grant of the Visiting Scholars Program of the Plan Propio de Investigación of the University of Granada. He also acknowledges the hospitality at Carmen de la Victoria. A.D. and D.P.M. are partially supported by the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq-Brazil) and by Project INCT-FNA Proc. No. 464898/2014-5. The work of E.M. is supported by the Spanish MINEICO and European FEDER funds under Grants FIS2014-59386-P and FIS2017-85053-C2-1-P, by the Junta de Andalucía under Grant FQM-225, and by the Consejería de Conocimiento, Investigación y Universidad of the Junta de Andalucía and European Regional Development Fund (ERDF) Grant SOMM17/6105/UGR. The research of E.M. is also supported by the Ramón y Cajal Program of the Spanish MINEICO under Grant RYC-2016-20678. The authors thank Dr. Tobias Frederico for reading the manuscript and for his suggestions and criticisms. A.D. is thankful to Dr. José Ademir de Lima for discussions about the statistics of small systems.
[1]{} Mandelbrot, B.B. The Fractal Geometry of Nature; WH Freeman: New York, NY, USA, 1983. G. West, [*Scale, The Universal Laws of Life, Growth, and Death in Organisms, Cities, and Companies*]{}, Penguin Books, 2018 - New York. K. Falconer, [*Fractals, a short introduction*]{}, Oxford Univ. Press, 2013 - Oxford. F.J. Dyson, Phys. Rev. 75 (1949) 1736. M. Gell-Mann and F.E. Low, Phys. Rev. 95 (1954) 1300. J.C. Ward, Proc. Phys. Soc. (London) A64 (1951) 54. L. Brink and K. K. Phua (edts). in 60 Years of Yang-Mills Gauge Field Theories, C. N. Yang’s Contributions to Physics, Nanyang Tech. Univ., Singapore, 2015. C. G. Callan, Jr., Phys. Rev. D 2 (1970) 1541. K. Symanzik, Commun. math. Phys. 18 (1970) 227. K. Symanzik, Commun. math. Phys. 23 (1971) 49. H. D. Politzer, Phys. Reports 14 (1974) 129. H. Georgi and H. D. Politzer, Phys. Rev. D9 (1974) 416. D. Gross and F. Wilczek, Phys. Rev. D 8 (1973) 3633. D. Gross and F. Wilczek, Phys. Rev. D9 (1974) 980. J. Blümleim, Prog. Part. Nucl. Phys. [**69**]{}, 28 (2013). C. Tsallis, J. Stat. Phys. 52 (1988) 479. M. Gell-Mann, C. Tsallis, [*Nonextensive Entropy: Interdisciplinary Applications*]{}, Oxfor University Press, USA, 2004. R. Hagedorn, CERN - TH. 520 65/166/5 (1965). R. Hagedorn, Nuovo Cimento Suppl. 3 (1965) 147. S. Frautschi, Phys. Rev. D 3 (1973) 2821. G. F. Chew, Phys. Rev. D 4 (1971) 2330. P. Huovinen and P. Petreczky, Nucl. Phys. A 837 (2010) 23. E. Megias, E.R. Arriola and L. Salcedo, Phys. Rev. D 99 (2019) 074020 R. Venugopalan and M. Prakash, Nucl. Phys. A 546 (1992) 718. A. Deppman, Phys. Rev. D 93 (2016) 054001. A. Deppman, T. Frederico, E. Megias and D. P. Menezes, Entropy 2018, 20, 633. A. Deppman, Physica A 391 (2012) 6380. J. Cleymans, D. Worku, J. Phys. G: Nucl. Part. Phys. 39 (2012) 025006. G. Wilk and Z. Wlodarczyk, J. Phys. G: Nucl. Part. Phys. 39, 095004 (2012) 98. L. Marques, E. Andrade-II and A. Deppman, Phys. Rev. D 87 (2013) 114022. L. Marques, J. Cleymans and A. Deppman, Phys. Rev. D 91 (2015) 054025. I. Sena and a. Deppman, Eur. Phys. J. A 49 (2013) 17. G. Wilk and Z. Włodarczk, Int. J. Mod. Phys. A 33 (2018) 1830008. B. De, Eur. Phys J. A 50 (2014) 70. A.S. Parvan, Eur. Phys. J. A 2017, 53. M. Ishihara,Int. J. Mod. Phys. E 26 (2017) 1750071. A.N. Tawfik, Phys. Part. Nucl. Lett. 33 (2018) 1830008. M. Rybczynski, Z. Włodarczyk and G. Wilk, J. Phys. G 39, 095004 (2012). C-Y. Wong, G. Wilk, J.L. Cirto, et al., Phys. Rev. D 91 (2015) 114027. A. Deppman, J. Phys. G: Nucl. Part. Phys. 41 (2014) 055108. P. H. G. Cardoso, T. Nunes da Silva, A. Deppman and D. P. Menezes, Eur. Phys. J. A 53, no. 10, 191 (2017). E. Megías, D. P. Menezes, and A. Deppman, Physica A 421, 15 (2015). D. P. Menezes, A. Deppman, E. Megías and L. B. Castro, Eur. Phys. J. A [**51**]{}, no. 12, 155 (2015). G.C. Yalcin and C. Beck, Sci. Rep. 8 (2018) 1764. G. Wilk and Z. Włodarczyk, Phys. Lett. B (2013) 163. I. Zborovský, M.V. Tokarev, Phys. Rev. D 75 (2007) 094008. I. Zborovsky, Int. J. Mod. Phys. A 33 (2018) 1850057. A. Deppman, Universe 3 (2017) 62.
A. Bialas and R. Peschanki, Nucl. Phys. B273 (1986) 703. A. Bialas and R. Peschanki, Nucl. Phys. B308 (1988) 857. R. C. Hwa, Phys. Rev. D 41 (1990) 1456. R.C. Hwa and J. Pan, Phys. Rev. D 45 (1992) 1476. S. Hegyi, Phys. Lett. B 318 (1993) 642. I.M. Dremin and R.C. Hwa, Phys. Rev. D 49 (1994) 5805. S. Hegyi and T. Csörgö, Phys. Lett. B 296 (1992) 256. N.G. Antoniou, N. Davis and F.K. Diakonos, Phys. Rev. C 93 (2016) 014908. Edward K.G. Sarkisyan, Aditya Nath Mishra, Raghunath Sahoo and Alexander S. Sakharov, Phys. Rev. D 93 (2016) 054046.
[^1]: Diagrams with four external lines represent contact interaction and are not considered, since they give a non renormalizable contribution. However, when all diagrams are summed up, the contribution of the contact interaction is null.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We consider a generalised complex Monge-Ampère equation on a compact Kähler manifold and treat it using the method of continuity. For complex surfaces, we prove an easy existence result. We also prove that (for three-folds and a related real PDE in a ball in $\mathbb{R}^3$), as long as the Hessian is bounded below by a pre-determined constant (whilst moving along the method of continuity path), a smooth solution exists. Finally, we prove existence for another real PDE in a 3-ball, which is a local, real version of a conjecture of X.X.Chen.'
address: |
Department of Mathematics\
Stony Brook University\
Stony Brook, NY 11794, USA
author:
- 'Vamsi P. Pingali'
title: 'A generalised Monge-Ampère equation'
---
Introduction
============
Let $(X,\omega)$ be an $n$-dimensional, compact, Kähler manifold. Here, we consider a generalised complex Monge-Ampère PDE (to be solved for a smooth function $\phi$) $$\begin{gathered}
\alpha_0(\omega + {d d^{c}}\phi)^n + \alpha _1 \wedge (\omega + {d d^{c}}\phi)^{n-1} + \ldots + \alpha _{n-1} \wedge (\omega + {d d^{c}}\phi) = \eta
\label{maineq}\end{gathered}$$ where $\eta$, $\alpha _i$ are smooth, closed forms satisfying the obvious necessary condition $\int _X \eta = \int _X (\alpha _0 \omega ^n + \alpha _1 \wedge \omega ^{n-1} +\ldots)$.\
When $\eta >0$, $\alpha_i=0 \ \forall \ i \neq 0$ and $\alpha _0=1$, equation \[maineq\] is the one introduced by Calabi [@Calabi] and solved by Aubin [@Aub] and Yau [@Yau]. Equations of this type are ubiquitous in geometry. A version of this generalised one appeared in [@XX]. The geometric applications of this equation shall be explored elsewhere.\
**Acknowledgements**: The author thanks his adviser Leon A. Takhtajan for suggesting this direction of study and for sparing time generously to discuss the same. We also thank Xiu Xiong Chen, Dror Varolin, Yanir Rubinstein, and, Marcus Khuri for fruitful discussions.
Statements of results
=====================
We state a somewhat general theorem about uniqueness, openness and $C^0$ estimates. The proof is quite standard (adapted largely from [@Tian] which is in turn based on [@Yau]). Although the theorem is folklore, we haven’t found the precise statement (in this level of generality) in the literature on the subject. In what follows, positivity of $(p,p)$ forms is strong positivity. Let $\mathcal{B}$ be the product of Banach submanifolds of forms wherein, an element of $\mathcal{B}$ is of the form $(\alpha _0, \alpha_1, \ldots, \alpha _{n-1}, \phi)$ where $\alpha_i$ are $C^{1,\beta}$ $(i,i)$, closed forms and $\phi$ is a $C^{3,\beta}$ function satisfying $\displaystyle \int _M \phi =0$, $n\alpha _0 (\omega + {d d^{c}}\phi)^{n-1} + (n-1) \alpha _1 \wedge (\omega + {d d^{c}}\phi)^{n-2} + \ldots + \alpha _{n-1}>0$ and, $\displaystyle \int _X (\sum _i \alpha _i \wedge \omega^{n-i}) \neq 0$. Also, let $\mathcal{\tilde{B}}$ be the Banach submanifold of $C^{1,\beta}$ top forms $\gamma$ with $\displaystyle \int _X \gamma =1$ and $\gamma >0$.
If $\alpha _0 \omega ^n + \alpha _1 \omega ^{n-1} + \ldots >0$, $\eta >0$ and, $d\alpha _i =0$, then, any smooth solution $\phi$ of \[maineq\] satisfying $\int _X \phi \omega ^n= 0$ and, $\kappa \geq K \omega ^{n-1} $ where $K>0$ and $\displaystyle \sum _k (\alpha _k (\omega + {d d^{c}}\phi)^{n-k}-\alpha _ k\omega ^{n-k}) = \kappa \wedge {d d^{c}}\phi$, is bounded a priori: $\Vert \phi \Vert _{C^0} \leq C_{\eta}$. Also, if $\alpha _i >0 \ \forall i$ and, if there exists a smooth solution $\phi$ such that $\omega + {d d^{c}}\phi >0$, it is unique (upto a constant) among all such solutions; In addition, the mixed derivatives of $\phi$ are bounded a priori $\Vert \phi \Vert _{C^{1,1}} \leq C_{\eta}$.\
The map $T : \mathcal{B} \rightarrow \mathcal{\tilde{B}}$ defined by $T(\alpha _0, \alpha _1, \ldots, \phi) = \frac{\sum _i \alpha _i \wedge (\omega + {d d^{c}}\phi)^{n-i}}{\int _X (\sum _i \alpha _i \wedge \omega^{n-i})}$ is open and, so is the restriction of $T$ to a subspace defined by fixing the $\alpha_i$. Also, a level set of this map is locally a graph with $\phi$ being a function of the $\alpha _i$.\
When $n=2$, and, $\alpha _0=1$, $\eta -\alpha _2 +\frac{\alpha _1 ^2}{4} >0$, there exists a unique, smooth solution to \[maineq\] satisfying $\omega +{d d^{c}}\phi + \frac{\alpha _1}{2}>0$ .
\[estimate\]
In particular, if $\alpha _i = \omega ^i$ for some $i$ and all the other $\alpha _j$ are small enough, then, by the solution of the $k$-Hessian equations [@Hess], [@Kol] we have a smooth solution of equation \[maineq\].\
One may formulate a version of the same problem locally as a Dirichlet problem on a pseudoconvex domain in $\mathbb{C}^n$. In this context, we note that viscosity solutions to the Dirichlet problem exist by [@LawsHarv] and [@Caff]. We also have the following result for a real version of the PDE :
The following Dirichlet problem on the ball $B$ of radius $1$ centred at the origin $$\begin{aligned}
\det (D^2u) + \Delta u &=& tf + (1-t)36 \nonumber \\
u\vert _{\partial B} &=& 0 \nonumber \\
f&>& 36 \nonumber\end{aligned}$$ has a unique smooth solution at $t=T$ if $f$ is smooth and, for all $t\in [0, T)$, smooth solutions $u_t$ exist and satisfy $D^{2}u_t > 3$. \[realv\]
A similar result holds for complex three-folds.
If $\alpha >0$, $\omega >0$ are smooth Kähler forms on a compact Kähler manifold $(X,\omega _0)$, then, there exists a constant $C>0$ depending only on $\alpha$ and $\omega _0$ such that, the equation $$\begin{aligned}
(\omega + {d d^{c}}u_t)^3 + \alpha^2 (\omega + {d d^{c}}u_t) &=& \frac{e^{tf}\int (\omega ^3 + \alpha ^2 \omega)}{\int e^{tf} (\omega ^3 + \alpha ^2 \omega)}(\omega ^3 + \alpha ^2 \omega)
\label{threeeq}\end{aligned}$$ has a unique smooth solution at $t=T$, if for all $t \in [0,T)$, smooth solutions exist and satisfy $\omega + {d d^{c}}u > C \omega _0$. \[threefolds\]
Finally, we present a local, real version of a conjecture of X.X. Chen (conjecture 4 in [@XX] made in the compact complex manifold case). Some progress has been made in a few special cases [@Lai]. However, in all these cases, the problem was reduced to an inverse Hessian equation. We prove existence in a special case here, using the method of continuity. Actually, a far more general result was proven in [@Kryl], but, results on the Bellman equation were used (as opposed to a direct method of continuity). Such results may not carry over in an obvious way to the manifold case and hence our proof of this “toy model”.
If $f>0$ is a smooth function on $\bar{B}(0,1)$ (the closed unit ball), then, the following Dirichlet problem has a unique, smooth, convex solution. $$\begin{aligned}
\det (D^2u) - \Delta u &=& f \nonumber \\
u\vert _{\partial B} &=& 0 \nonumber \\
f&>& 0 \nonumber\end{aligned}$$
\[Chen\]
Standard results used
=====================
For the convenience of the reader, we have included statements of some standard results in the form that we use in the proofs.\
Our principal tool to study fully nonlinear PDE like equation \[maineq\], is the method of continuity (It is like a flow technique. In fact this analogy was exploited more seriously, to great advantage, in [@Yanir]). To solve $Lu =f$ where $L$ is a nonlinear operator, one considers the family of equations $Lu_t = \gamma (t)$ where, $\gamma (1)=f$ and $\gamma (0)=g$ such that, at $t=0$, one has a solution $Lu_0=g_0$. Then, one proves that the set of $t\in [0,1]$ for which the equation has a solution is both, open and closed (and clearly non-empty). In order to prove openness, one considers $L$ to be a map between appropriate Banach spaces. Then, the implicit function theorem of Banach spaces proves openness. However, while dealing with equations like Monge-Ampère equations, one has to verify that certain conditions like ellipticity are preserved along the “continuity path". This is crucial because, in order to solve the linearised equation and, to prove that indeed one has a solution in an appropriate Banach space, one needs ellipticity in these cases. In fact, in a few of the cases we shall consider, ellipticity is not preserved and hence, the best we can do is a “short-time" existence result. In order to prove closedness, one needs to prove uniform (i.e. independent of $t$) *a priori* estimates for $u$. In our case, we shall need these estimates in $C^{2, \alpha}$ in order to use the Arzela-Ascoli theorem to conclude closedness. These estimates are usually proved by improving on lower order estimates. Once, one produces a $C^{2,\alpha}$ solution, one “bootstraps" the regularity (at teach $t \in [0,1]$) using the Schauder estimates. The Schauder estimates on a compact manifold (without boundary) are (they can be derived easily using similar interior and boundary ones in domains in $\mathbb{R}^n$ [@Trud]).
*Schauder $\emph{a priori}$ estimates on a Riemannian manifold*: If $Lu = f$, where $L$ is a second-order, uniformly elliptic operator with smooth coefficients, and, $u$ is a $C^{2,\alpha}$ and $f$ is a $C^{0,\alpha}$ function, $$\begin{gathered}
\Vert u \Vert _{C^{2, \alpha}} \leq C(\Vert u \Vert _{C^{0}} + \Vert f \Vert _{C^{0, \alpha}}) \nonumber\end{gathered}$$
In order to derive *a priori* estimates, we shall use standard techniques as in [@Yau], [@Tian] for the manifold case, and, [@Caff] for the Euclidean case. The main blackbox is the Evans-Krylov-Safanov theory for proving $C^{2, \alpha}$ estimates from $C^{2}$ ones. This requires (apart from uniform ellipticity) concavity of the equation. There is a similar version for the complex case. The real version is :
Let $u$ be a smooth function on the unit ball satisfying, $$\begin{gathered}
F(D^2 u, x, Du) = g \nonumber\end{gathered}$$ on the unit ball in $\mathbb{R}^n$ centred at the origin $B(0,1)$ with $u=0$ on the boundary of the ball. Here, $F$ is a smooth function defined on a convex open set of symmetric $n \times n$ matrices $\times \mathbb{R} \times \mathbb{R}^n$ which satisfies,\
a) *Uniform ellipticity on solutions* : There exist positive constants $\lambda$ and $\Lambda$ so that $0 < \lambda \vert \xi \vert^2 \leq F_{ij} (D^2 u ,x , Du) \xi _i \xi _j \leq \Lambda \vert \xi \vert ^2$ for all vectors $\xi$ and all $u$ satisfying the equation.\
b) *Concavity on a convex open set* : $F$ is a concave function on a convex open set of symmetric matrices (containing $D^2 u$ for all solutions $u$).\
Then, $\Vert u \Vert _{C^{2, \alpha} (\bar{B(0,1)})} \leq C $ where $C$ and $\alpha$ depend on the first and second derivatives of $F$, $\Vert u \Vert _{C^2(\bar{B})}$, $\Vert g \Vert _{C^2(\bar{B})}$, $n, \lambda$ and $\Lambda$. \[EvansKrylov\]
The complex, interior version (that we need) is :
Let $u$ be a $C^4$ function on the unit ball in $\mathbb{C}^n$ satisfying $$\begin{gathered}
F(u _{i\bar{j}}, z,\bar{z}) = 0 \nonumber \end{gathered}$$ for a $C^{2,\beta}$ function $F (x, p, \bar{p})$ satisfying,\
a) *Uniform ellipticity on solutions* : There exist positive constants $\lambda$ and $\Lambda$ so that $0 < \lambda \vert \xi \vert^2 \leq F_{i\bar{j}} ({d d^{c}}u, z, \bar{z}) \xi _i \xi _j \leq \Lambda \vert \xi \vert ^2$ for all vectors $\xi$ and all $u$ satisfying the equation.\
b) *Concavity on a convex open set* : $F$ is a concave function on a convex open set of hermitian matrices (containing $u_{i\bar{j}}$ for all solutions $u$).\
Then, $\Vert u \Vert _{C^{2,\alpha}(B(0,\frac{1}{2}))} \leq C$ where, $C$ and $\alpha$ depend on $\lambda$, $\Lambda$, $n$, and $\Vert u_{i\bar{j}} \Vert _{C^0 (\bar{B})}$ and uniform bounds on the first and second derivatives of $F$ evaluated at $u$. \[CompEvans\]
The proofs are standard [@Trud], [@Siu], [@Blocki], [@Kaz]. Usually, one proves these estimates when the PDE is concave on all symmetric matrices. In Monge-Ampère equations, one needs a weaker requirement of being concave on a convex open set of symmetric matrices [@Caff] (the proofs go through easily with this requirement).\
To conclude, we add a few words about uniqueness. The usual technique for demonstrating uniqueness (due to Calabi) of $Lu=f$ is to assume two solutions $u_1$ and $u_2$, and, to write $0=Lu_1 - Lu_2 = \int _{0} ^{1} \frac{dL}{dt} (tu_2+(1-t)u_1) dt$. If the integrand is an elliptic operator, by the maximum principle, $u_1 = u_2$.
Proofs of the Theorems
======================
Proof of theorem \[estimate\]
-----------------------------
This proof is similar to the one for the usual Monge-Ampère equation [@Tian].\
*The $C^0$ estimate* : As usual, without loss of generality, we may change the normalisation to $\sup \phi = -1$ i.e. we may add $-1-\sup \phi$ to $\phi$. Indeed, if the new $\phi$ has a $C^0$ estimate, then, $\int _X \phi =0$ yields the desired $C^0$ estimate. This means, we just have to find a lower bound on $\phi$. Certainly $\phi$ has an $L^1$ bound [@Tian]. Let $\phi = -\phi _{-}$ (so that $\phi _{-} \geq 1$). Subtracting $\Theta = \displaystyle \sum_k \alpha _k \wedge \omega ^{n-k}$ and then, multiplying the equation by $\phi _{-} ^p$ and integrating, we have (here $\eta = e^f \Theta $), $$\begin{aligned}
-\int \phi _{-}^p {d d^{c}}\phi_{-} \wedge \kappa &=& \int \phi _{-}^p (e^f-1)\Theta \nonumber \\
\int \phi _{-}^p (e^f-1)\Theta &\leq& c \Vert \phi_{-} \Vert _{L^p} ^p \nonumber \\
-\int \phi _{-}^p {d d^{c}}\phi_{-} \wedge \kappa &=& \int d(\phi _{-} ^p) \wedge d^c \phi_{-} \wedge \kappa \nonumber \\
&=& c\int d(\phi _{-} ^{\frac{p+1}{2}}) \wedge d^c (\phi _{-} ^{\frac{p+1}{2}}) \wedge \kappa \nonumber \\
&\geq& C\Vert \nabla (\phi _{-} ^{\frac{p+1}{2}}) \Vert _{L^2} ^2 \nonumber \\
&\geq& C_1 \left ( \int \phi _{-} ^{\frac{(p+1)n}{n-1}} \right )^{\frac{n-1}{n}} -C_2 \int \phi _{-} ^{p+1} \nonumber \end{aligned}$$ where the last inequality follows from the Sobolev embedding theorem. Upon rearranging, we have $$\begin{aligned}
\Vert \phi _{-} \Vert _{L^{(p+1)(n)/(n-1)}} \leq (C (p+1))^{\frac{1}{p+1}} \Vert \phi _{-} \Vert _{L^{p+1}} \nonumber \end{aligned}$$ The Moser-iteration procedure gives $\sup \vert \phi \vert \leq C \Vert \phi \Vert _{L^2}$. If we prove that the right hand side is controlled by the $L^1$ norm of $\phi$, we will be done. Indeed, $$\begin{aligned}
C\Vert \phi \Vert _{L^1} &\geq& \int \phi (1-e^f) \Theta \nonumber \\
&\geq& C \Vert \nabla \phi \Vert _{L^2} ^2 \nonumber \\
&\geq& C \Vert \phi - \langle \phi \rangle \Vert _{L^2} ^2 \nonumber \\
\Vert \phi \Vert _{L^2} &\leq& C(\Vert \phi \Vert _{L^1} +1) \nonumber \end{aligned}$$ where, we have used the Poincar' e inequality. Hence, proved.\
*Uniqueness* : If $\phi _1$ and $\phi _2$ are two solutions, upon subtraction we have, $$\begin{gathered}
\sum _i \alpha_i \wedge ((\omega + {d d^{c}}\phi _1)^{n-i} - (\omega + {d d^{c}}\phi_2)^{n-i}) = 0 \nonumber \\
\Rightarrow \int _{0}^{1} \displaystyle \sum _ k k\alpha_k \wedge (\omega + {d d^{c}}\phi_1+ t{d d^{c}}(\phi _2-\phi_1))^{n-k-1}dt \wedge {d d^{c}}(\phi_2 - \phi_1) = 0 \nonumber \end{gathered}$$ Thus, by the maximum principle, $\phi_2 - \phi_1$ is a constant.\
*The mixed derivatives estimate*: When $\alpha _i >0$, $$\begin{gathered}
\eta \geq \alpha _{n-1} \wedge (\omega + {d d^{c}}\phi)
> C(\mathrm{tr}(\omega + {d d^{c}}\phi)) \nonumber \end{gathered}$$ where $C>0$. Since $0<\omega + {d d^{c}}\phi$, the eigenvalues of ${d d^{c}}\phi$ are bounded above. Thus, the mixed second derivatives of $\phi$ are bounded. Note that, by the Schauder estimate [@Morr], the first derivatives are bounded as well.\
*Openness* : The map $T$ is smooth. Its Gâteaux derivative is $DT(0, 0, \ldots, 0, \chi) = (n\alpha _0 (\omega + {d d^{c}}\phi)^{n-1} + (n-1) \alpha _1 \wedge (\omega + {d d^{c}}\phi)^{n-2} + \ldots + \alpha _{n-1}) \wedge {d d^{c}}\chi$. It is clearly a bounded surjection (by the Schauder theory) onto its image if $n\alpha _0 (\omega + {d d^{c}}\phi)^{n-1} + (n-1) \alpha _1 \wedge (\omega + {d d^{c}}\phi)^{n-2} + \ldots + \alpha _{n-1} >0$. If $DT$ is restricted to vectors of the form $(0,0,\ldots,0,\chi)$, then, it is a Banach space isomorphism. Hence, by the implicit function theorem of Banach manifolds, openness is guaranteed. In fact, it also guarantees that, on a level set, $\phi$ can be solved for (locally), in terms of $\alpha_i$.\
*The n=2 case* : The equation we have is equivalent to $$\begin{gathered}
(\omega + {d d^{c}}\phi + \frac{\alpha_1}{2}) ^2 = \eta - \alpha _2 + \frac{\alpha _1 ^2}{4} \nonumber \end{gathered}$$ This is just the usual Monge-Ampère equation and hence we are done.
Proof of theorem \[realv\]
--------------------------
Uniqueness is proved as before. We shall only prove existence. Let $Lu = \det (D^2 u) + \Delta u$. To this end, we use the method of continuity. Consider the equation $$\begin{aligned}
Lu_t &=& tf +(1-t)L\phi \nonumber \\
u_t \vert _{\partial B} &=& 0 \nonumber \\
\phi &=& \frac{3}{2} \sum x_i^2 -\frac{3}{2}
\label{tequa}\end{aligned}$$ When $t=0$, it has a smooth solution, namely, $\phi$.\
*Openness*: Let $\Omega \subset C_{0}^{2,\alpha} (\bar{B})$ be the set of $u$ such that $D^2 u > 3$ (where the subscript $0$ indicates vanishing on the boundary). This is an open subset. Define $T: \Omega \rightarrow C_{0} ^{0,\alpha}$ to be $T(u_t) = \det (D^2 u_t) + \Delta u_t$. If $u_s$ is a solution of \[tequa\], then, it is easy to see that $DT_{u_s}$ is a linear isomorphism. Hence, by the inverse function theorem of Banach manifolds, we see that the set of $t$ for which there is a solution is open.\
*Closedness*: Suppose there is a sequence $t_i \rightarrow t$ such that there are smooth solutions $u_{t_i}$ satisfying $D^2 u > 3$. Then, we wish to prove that a subsequence of the $u_{t_i}$ converges to a smooth solution $u_t$ in the $C^{2,\beta}$ topology. This requires *apriori* estimates (the convergence following from the Arzela-Ascoli theorem). We shall prove the same for the equation \[realv\]. We just have to prove the $C^{2,\alpha}$ estimate in order to ensure smoothness (by the Schauder theory).\
*$C^0$ estimate*: Note that $\Delta u \leq f$. Hence, for $A >>1$, $0> f-\Delta(A \sum x_i^2) = \Delta (u-A\sum x_i ^2)$. The minimum principle implies that $u \geq A\sum x_i ^2- A$. Since, $\Delta u > 9$, $u\leq 0$ by the maximum principle. thus, $\Vert u \Vert _{C^0} \leq C$.\
*$C^1$ estimate*: Differentiating both sides using the operator $D$, $\mathrm{tr}((\mathrm{Hess} u)^{-1} D^2w) + \Delta w = Df$ where $w=Du$. Just as before, by adding or subtracting a large multiple of $\sum x_i ^2$ to $w$ and using the maximum principle, we see that $\Vert Du \Vert _{C^0}$ is controlled by its supremum on the boundary. The tangential boundary derivatives are $0$. Since, $A \sum x_i ^2 - A \leq u \leq 0$, $ \vert \frac{\partial u}{\partial n} \vert \leq 2A $. Hence, $\Vert u \Vert _{C^1} \leq C$.\
*$C^2$ estimate*: Since $\Delta u \leq Lu \leq f$ and $\Delta u>0$, $\Vert u_{ij} \Vert _{C^0} \leq C$. Hence, $\Vert u \Vert _{C^2} \leq C$.\
*$C^{2,\alpha}$ estimate*: So far, we haven’t used anything about the sequence except that $D^2 u_{t_i} >0$. This will change presently. For any function $F: \mathbb{R} \rightarrow \mathbb{R}$, $F(\det (D^2 u) + \Delta u) = F(f)$. If we choose the function appropriately, then the resulting equation will be a concave, uniformly elliptic Monge Ampère PDE to which we may apply the Evans-Krylov theory to extract a $C^{2, \alpha}$ estimate.\
We claim that, the function $F(x) = \int _{36} ^{x} e^{-\frac{t^2}{2}} dt$ is such that, $g(\lambda _1, \lambda _2, \lambda _3) = F(\sum \lambda _i + \lambda _1 \lambda _2 \lambda _3)$ has a uniformly positive gradient and is concave if $\lambda _i > 3$. By using the $C^2$ estimate and theorem \[EvansKrylov\], we have the desired estimate.\
We shall prove the aforementioned fact: Let $x=\sum \lambda _i + \lambda _1 \lambda _2 \lambda _3$. We see that $\frac{\partial g}{\partial \lambda _i}\vert_{D^2 u} = e^{-\frac{x^2}{2}}(1+\frac{\lambda _1 \lambda _2 \lambda _3}{\lambda _i}) > e^{-\frac{f^2}{2}}$ and is less than $1+3f$ where we have evaluated the derivative at the eigenvalues of the Hessian of a solution of equation \[realv\]. Hence, it is uniformly elliptic.\
If $(v_1 , v_2 , v_3 ) \in \mathbb{R}^3 $, then $ -v_i v_j\frac{\partial ^2 g}{\partial x_i \partial x_j} = e^{-\frac{x^2}{2}}(x(v_1(1+\lambda _2 \lambda _3)+v_2 (1+\lambda _3 \lambda _1) + v_3 (1+\lambda _1 \lambda _2))^2)-2e^{-\frac{x^2}{2}}(v_1 v_2 \lambda_3 + v_2 v_3 \lambda _1 + v_3 v_1 \lambda _2)$, which is in turn equal to $ e^{-\frac{x^2}{2}} (v_1 ^2 \alpha + \beta v_1 + \gamma) \geq 0 \Leftrightarrow \beta ^2 - 4 \alpha \gamma \leq 0$ and, $$\begin{aligned}
\frac{\beta ^2- 4\alpha \gamma}{4} &=& (v_2(x(1+\lambda _2 \lambda _3)(1+\lambda _1 \lambda _3)-\lambda _3)+v_3 (x(1+\lambda _1 \lambda _2)(1+\lambda _2 \lambda _3)-\lambda _2))^2 \nonumber \\
&-& x(1+\lambda _2 \lambda _3)^2(v_2 ^2 x (1+\lambda _1 \lambda _3)^2 + v_3 ^2 x(1+\lambda _1 \lambda _2)^2
+ 2v_2 v_3 (x(1+\lambda _1 \lambda _3)(1+\lambda _1 \lambda _2)-\lambda _1)) \nonumber \\
&=& \tilde{\alpha}v_2 ^2 + \tilde{\beta} v_2 + \tilde{\gamma} \leq 0 \nonumber \end{aligned}$$ with the last inequality holding if and only if $\tilde{\alpha} \leq 0$ and $\tilde{\beta}^2 - 4\tilde{\alpha} \tilde{\gamma}\leq 0$. Let us assume (without loss of generality) that $v_3 \neq 0$ and that $\lambda _1 < \lambda _2 <\lambda _3$. $$\begin{aligned}
\tilde{\alpha} &=& (x(1+\lambda _2 \lambda _3)(1+\lambda _1 \lambda _3)-\lambda _3)^2-x^2(1+\lambda _2 \lambda _3)^2(1+\lambda _1 \lambda _3)^2 \nonumber \\
&=& \lambda _3 ^2 - 2\lambda _3 x (1+\lambda _2 \lambda _3)(1+\lambda _1 \lambda _3) \nonumber \\
&\leq& -2x \lambda _3 ^2 (\lambda _1 +\lambda _2 + \lambda _1 \lambda _2 \lambda _3)
\leq -2x \lambda _3 ^2 \frac{2x}{3}\nonumber \end{aligned}$$ $$\begin{aligned}
\frac{\tilde{\gamma}}{v_3 ^2} &=&(x(1+\lambda _2 \lambda _1)(1+\lambda _2 \lambda _3)-\lambda _2)^2-x^2(1+\lambda _2 \lambda _3)^2(1+\lambda _1 \lambda _2)^2 \nonumber \\
&\leq& -\frac{4x^2\lambda _2^2}{3}\nonumber \end{aligned}$$ $$\begin{aligned}
\frac{\tilde{\beta}}{2v_3} &=& (x(1+\lambda _2 \lambda _3)(1+\lambda _1 \lambda _2)-\lambda _2)(x(1+\lambda _2 \lambda _3)(1+\lambda _1 \lambda _3)-\lambda _3)\nonumber \\
&-&x(1+\lambda _2 \lambda _3)^2(x(1+\lambda _1 \lambda _3)(1+\lambda _1 \lambda _2)-\lambda _1)\nonumber \end{aligned}$$ $$\begin{aligned}
\left (\frac{\tilde{\beta}}{2v_3} \right ) ^2 &=& (x(1+\lambda _2 \lambda _3)(\lambda _2 + \lambda _3 - \lambda _1 + \lambda _1 \lambda _2 \lambda _3)-\lambda _2 \lambda _3)^2 \nonumber \\
&\leq& x^4 (1+\lambda _2 \lambda _3) ^2 \nonumber \\
\frac{\tilde{\beta} ^2 - 4 \tilde{\alpha}\tilde{\gamma}}{4v_3 ^2} &\leq& 0 \nonumber \end{aligned}$$
Hence proved.\
*Remark* : Writing equation \[maineq\] for $n=3$ and $\alpha_0 =1$ we have, $$\begin{gathered}
(\omega + {d d^{c}}\phi)^3 + \alpha _1 (\omega + {d d^{c}}\phi) ^2 + \alpha _2 (\omega + {d d^{c}}\phi) = \eta \nonumber \\
\Rightarrow (\omega + \frac{\alpha _1}{3} + {d d^{c}}\phi)^3 + (\alpha _2 - \frac{\alpha _1 ^2}{3})(\omega + \frac{\alpha _1}{3} + {d d^{c}}\phi) = \eta - \frac{2\alpha _1 ^3}{27} + \frac{\alpha _1 \alpha _2}{3} \nonumber \end{gathered}$$ A local, real version of a special case of the above is equation \[realv\].
Proof of theorem \[threefolds\]
-------------------------------
Once again, we apply the method of continuity. We shall impose several conditions on $C$ (as we go along). It should be large enough so that, whenever $\beta > C\omega _0$, $\beta ^3 > 3\alpha ^2 \beta$ (Indeed, if $K>0$ and $B>0$ are given, $\det(A) > K\mathrm{tr}(BA)$ for sufficiently large $A>0$). Obviously, at $t=0$, $u=0$ solves the equation. Openness and uniqueness, follow from theorem \[estimate\]. As before, if $t_i \rightarrow t$ is a sequence such that there exist smooth solutions $u_i$ satisfying $\omega + {d d^{c}}u_i > C \omega _0$, then, we shall prove that a subsequence converges to a smooth solution $u$ in the $C^{2,\beta}$ topology. As usual, we need *apriori* estimates for this.\
The $C^{0}$ and the mixed derivative estimates follow directly from theorem \[estimate\]. We have to prove the $C^{2,\alpha}$ estimate (thus proving existence and smoothness as before). It suffices to prove a local (interior) estimate. We shall accomplish this via the complex version of the (interior) Evans-Krylov theory done in [@Blocki] and [@Siu].\
The local (in a ball) version of the equation is $$\begin{aligned}
\det (\phi _{i \bar{j}}) + \mathrm{tr}(B^{-1}[\phi _{i\bar{j}}]) &=& f \nonumber \\
\phi _{i \bar{j}} &>& C>1 \nonumber \\
f &>& C^3+9\Vert B ^{-1} \Vert^2 C \nonumber \\
\label{locver} \end{aligned}$$ where $B^{-1}_{i\bar{j}} = \det(\alpha) [\alpha]^{-1} _{i\bar{j}}$. We claim that the function $g (A) = F(\det (A) + \mathrm{tr}(B^{-1}A))$ from hermitian matrices satisfying $A > CId $ to $\mathbb{R}$ (where $F(x) = \int _{c} ^{x} e^{-\frac{t^2}{2}}dt$) is concave and uniformly elliptic. Let the eigenvalues of $A$ be $\lambda _1, \lambda _2$ and $\lambda _3$. The uniform ellipticity is trivial (as in the proof of theorem \[realv\]). The concavity is also somewhat similar to theorem \[realv\], but requires some modification. Indeed (here $V$ is an arbitrary hermitian matrix and $x=\det (A) + \mathrm{tr}(B^{-1}A)$), $$\begin{aligned}
g^{''}(V,V) &=& g''(x) (\det (A) \mathrm{tr}(A^{-1}V)+\mathrm{tr}(B^{-1}V))^2 \nonumber \\
&+& g'(x) (-\det(A)\mathrm{tr}((A^{-1}V)^2)+\det(A)(\mathrm{tr}(A^{-1}V))^2) \nonumber \end{aligned}$$ We wish to prove that $g^{''}(V,V)<0$ for every hermitian $V$. Let’s diagonalise the positive-definite form $B^{-1}$, i.e. $PB^{-1}P^{\dag} = I$ for some matrix $P$. Define $\tilde{A} = (P^{\dag})^{-1} A P^{-1}$ and $\tilde{V} = (P^{\dag})^{-1} V P^{-1}$. Now, using a unitary matrix $U$, we may diagonalise $\tilde{A}$ i.e. $\tilde{\tilde{A}} = U \tilde{A} U^{\dag} = \mathrm{diag} (a_1, a_2, a_3)$ where $a_1 \leq a_2 \leq a_3$ and $\tilde{\tilde{V}} = U \tilde{V} U^{\dag}$. This implies that $\det (\tilde{A}) \det(B) = \det(A)$ and $\mathrm{tr} ( \tilde{A}) = \mathrm{tr}(B^{-1}A)$. Let $\tilde{\tilde{V}}_{ii} = v_i$. Hence, $$\begin{aligned}
g^{''}(V,V) &=& -xe^{-\frac{x^2}{2}} (\det(B)a_1 a_2 a_3 (\sum \frac{ v_i}{a_i}) + \sum v_i) ^2 \nonumber \\
&+& e^{-\frac{x^2}{2}} ((\sum \frac{ v_i}{a_i})^2-(\sum \frac{v_i ^2}{a_i ^2} + 2 (\frac{\vert v_{12} \vert^2}{a_1 a_2}+\frac{\vert v_{23} \vert^2}{a_2 a_3}+\frac{\vert v_{13} \vert^2}{a_1 a_3}))) \nonumber \\
&\leq& -xe^{-\frac{x^2}{2}} (\sum v_i (\det(B)\frac{a_1 a_2 a_3}{a_i} + 1)) ^2
+ 2\det(B)e^{-\frac{x^2}{2}} (v_1 v_2 a_3 + v_2 v_3 a_1 + v_3 v_1 a_2) \nonumber \\
&=& e^{-\frac{x^2}{2}}(Pv_1 ^2 + Q v_1 +R) \nonumber \end{aligned}$$ where, $$\begin{aligned}
P &=& -x (\det(B) a_2 a_3 +1)^2 \leq 0\nonumber \\
Q &=& 2(\det(B)(v_2 a_3 + v_3 a_2)-x (\det(B)a_2 a_3 + 1)(v_2 (\det(B) a_1 a_3 +1) + v_3 (\det(B) a_1 a_2 + 1))))\nonumber \\
R &=& 2 \det(B) v_2 v_3 a_1 - x (v_2 (\det(B) a_1 a_3 + 1) + v_3 (\det(B) a_1 a_2 + 1))^2\nonumber \end{aligned}$$ as before, we want $Q^2 - 4 PR < 0$. Assume (without loss of generality) that $v_3=1$. $$\begin{aligned}
\frac{Q^2-4PR}{4} &=& Jv_2 ^2 + K v_2 + L \nonumber \end{aligned}$$ where, $$\begin{aligned}
J &=& \det(B) a_3 (\det(B)a_3 - 2x(\det(B)a_2 a_3 + 1)(\det(B) a_1 a_3 +1))\nonumber \\
&<& \det(B) ^2 a_3 (1-2a_1 a_2 \det (B)) \nonumber \\
&<& 0 \nonumber \\
K &=& 2 (a_2 a_3 (\det(B))^2 -x(\det(B)a_2 a_3 +1)((\det(B)) ^2 a_1 a_2 a_3 + \det(B) (a_2+a_3-a_1)))\nonumber \\
L &=& \det(B)a_2 (\det(B)a_2 - 2x(\det(B)a_2 a_3 + 1)(\det(B) a_1 a_2 +1)) \nonumber \end{aligned}$$ where the first inequality follows from the assumption that $\det(A) =\det(B) a_1 a_2 a_3 > 3\mathrm{tr}(B^{-1}A) > 3\sum a_i$ . Hence, $$\begin{aligned}
\frac{K^2}{4} &=& (a_2 a_3 (\det(B))^2 -x\det(B)(\det(B)a_2 a_3 +1)(\det(B) a_1 a_2 a_3 + (a_2+a_3-a_1)))^2\nonumber \\
&\leq& x^4(\det(B) a_2 a_3 + 1)^2 \det(B) ^2 \nonumber \\
J &\leq& -2x \det(B) ^2 a_3 ^2 (\det(B) a_1 a_2 a_3 + a_1 + a_2) \nonumber \\
&\leq& -2x \det(B) ^2 a_3 ^2 \frac{2x}{3} \nonumber \\
K &\leq& -2x \det(B) ^2 a_2 ^2\frac{2x}{3} \nonumber \end{aligned}$$ Thus, $K^2-4JL < 0$ implying that $g$ is concave. The $C^{2, \alpha}$ estimate follows from theorem \[CompEvans\].
Proof of theorem \[Chen\]
-------------------------
We use the method of continuity again. As before, openness follows easily using the Implicit function theorem on Banach spaces. Here, we prove only the *a priori* estimates. Smoothness follows by bootstrapping, as indicated earlier. Lastly, we shall also prove the uniqueness of convex solutions.\
*$C^0$ estimate*: Since $u$ is convex, its maximum is attained on the boundary and hence, $u\leq 0$. Let $\phi = \frac{\mu}{2} r^2 - \frac{\mu}{2}$ where $\mu >0$, and, $\mu^3-3\mu > \max f$; then, subtracting $\det(D^2 u) - \Delta u$ from $\det(D^2 \phi) - \Delta \phi$, we have (assume that the eigenvalues of $D^2 u$ are $\lambda _i$), $$\begin{gathered}
L(\phi - u) = \det(D^2 \phi) - \det (D^2 u) - \Delta (\phi - u) \nonumber \\
= (\mu - \lambda _1)(\frac{\mu ^2 + \frac{\mu}{2} (\lambda _2 + \lambda _3)+\lambda _2 \lambda _3}{3}-1) + (\mu - \lambda _2)\ldots \nonumber \\
= \mu ^3 - \mu - f > 0 \nonumber \end{gathered}$$ We see that, since $\mu^2 >3$, hence, $L$ is an elliptic operator acting on $\phi -u$ with $L (\phi - u)>0$. So, by the maximum principle, $\phi < u$. This gives us a $C^0$ estimate on $u$.\
*$C^1$ estimate*: Follows from ellipticity as before.\
*$C^2$ estimate*: For future use, notice that, atleast two of the eigenvalues of $D^2 u$ are larger than $1$. Taking derivatives of the equation we have (let $u_0$ be the minimum of $u$), $$\begin{gathered}
\det(D^2 u) \mathrm{tr}((D^2 u)^{-1} D^2 u_i) - \Delta u_i = f_i \nonumber \\
\det(D^2 u) \mathrm{tr}((D^2 u)^{-1} D^2 \Delta u) - \Delta \Delta u = \Delta f + \sum _i \det (D^2 u) \mathrm{tr}(((D^2 u)^{-1} D^2 u_i)^2 ) \nonumber \\
- \det (D^2 u) \sum _i (\mathrm{tr}((D^2 u)^{-1} D^2 u_i))^2 \nonumber \end{gathered}$$ Let $A = \det (D^2 u) (D^2 u)^{-1} - I$. Consider $g = \Delta u + \mu (u-u_0)>0$ (we shall choose the constant $\mu >0$ later. It can depend on $\Vert u \Vert _{C^1} $ and other constants). Notice that, if $g$ is bounded, then, so is $\Delta u$ and thus, $D^2 u$ is bounded. At the maximum of $g$ (if it occurs in the interior), $(\Delta u)_i = -\mu u_i$ and $\mathrm{tr}(A D^2 g) \leq 0$. This implies, $$\begin{aligned}
0 &\geq& \Delta f + \mu\mathrm{tr} (A D^2 u) - \frac{\vert -\mu \nabla u + \nabla f \vert^2}{\Delta u + f} \nonumber \\
&\geq& C_1 (\mu) - \frac{C_2 (\mu)}{\Delta u + f} + (2\Delta u +3f)\mu \nonumber \end{aligned}$$ Hence, $\Delta u$ is bounded at that point. Thus, $g$ is bounded at that point. This implies that $\Delta u$ is bounded everywhere. If the maximum of $g$ occurs on the boundary (call the max $g_0$), we shall have to analyse it separately. Let $\tilde{g} = g + g_0 (1-2r^2)$. Clearly, the maximum of $\tilde{g}$ has to occur in the interior. There, $D\tilde{g}=0$ and $\mathrm{tr}(AD^2\tilde{g}) \leq 0$. Hence (here, we assume that, $\mathrm{tr}(A) = \sum (\lambda _i \lambda _j - 1) $ and, that $g_0$ are sufficiently large compared to constants; If not, we are done), $$\begin{aligned}
0 &\geq& -C_1 (\mu) + \mu \mathrm{tr}(A) g_0 -C_2 g_0 - (C_4 g_0 + C_6 (\mu)) \mathrm{tr}(A) - C_5 \frac{g_0 ^2}{\Delta u +f} \nonumber \\
&\geq& -\tilde{C}_1 (\mu) + \mu \mathrm{tr}(A) g_0 -\tilde{C}_2 g_0 - \tilde{C}_4 g_0 \mathrm{tr}(A) - \tilde{C}_5 g_0 \nonumber \\
&\geq& -E_1 (\mu) + (\mu g_0 - E g_0) \mathrm{tr}(A) \nonumber \end{aligned}$$ Choosing $\mu >E$, we see that, $g_0$ is bounded. Notice that, this also implies a lower bound on $D^2 u$. This is because, $M\lambda_i > \lambda _1 \lambda _2 \lambda _3 > f$.\
*$C^{2, \alpha}$ estimate* : Notice that, the set $Y$ of positive, symmetric matrices satisfying $\det(A)-\mathrm{tr}(A)>0$ is a convex open set (lemma $4.16$ of [@Kryl]). Also, our equation maybe written as $-1=-\frac{\Delta u}{\det (D^2 u)} - \frac{f}{\det(D^2 u)} = F(D^2 u, x)$ which is certainly concave on $Y$ by the same lemma in [@Kryl]. It is uniformly elliptic on solutions as long as the eigenvalues of the Hessian are bounded below and above (which they are, by the $C^2$ estimates). Theorem \[EvansKrylov\] yields the desired estimates.\
*Uniqueness* : If $u_1$ and $u_2$ are two convex solutions of the equation $F(u)=-1$ (as above), then, upon subtraction, $ 0 = \int _{0} ^{1} \mathrm{tr}((-I+\det (D^2 u_t) (D^2 u_t)^{-1})D^2 (u_2-u_1)) dt = L(u_2-u_1)$ where, $L$ is elliptic. By the maximum principle, $u_1 = u_2$.
[99]{} D. Gilbarg, N.S. Trudinger, Elliptic partial differential equations of second order. *Springer*, $1998$. T. Aubin, M' etriques Riemanniennes et courbure. *J. Diff. Geom*, $1970$. E. Calabi, The space of Kähler metrics. *Proc. Intern. Congress Math. Amsterdam*, $1954$. S.T. Yau, On the ricci curvature of a compact kähler manifold and the complex monge-ampère equation, I. *Comm. Pure and App. Math.*, $1978$. F.R. Harvey and B. Lawson, Dirichlet duality and the nonlinear Dirichlet problem on riemannian manifolds. *J. Diff. Geom.*, $2011$. G. Tian, Canonical metrics in Kähler geometry. *Birkhäuser*, $2000$. L. Caffarelli, J.J. Kohn, L. Nirenberg and J. Spruck, The dirichlet problem for nonlinear second-order elliptic equations. II. Complex monge-ampère and uniformly elliptic equations. *Comm. Pure and App. Math.*,38, $1985$. Y. A. Rubinstein, Geometric quantization and Dynamical constructions on the space of Kähler metrics. *PhD thesis, M.I.T*, $2008$. Z. Hou, X.N. Ma, D. Wu, A second order estimate for complex Hessian equations on a compact Kähler manifold. *Math. Res. Lett.*, $2010$. S. Dinew, S. Kolodziej, Liouville and Calabi type theorems for complex Hessian equations. arXiv:1203.3995v1, $2012$. C.B. Morrey, Multiple Integrals in the Calculus of Variation. *Springer-Verlag*, $1966$. J.L. Kazdan, Prescribing the curvature of a Riemannian manifold. *Am. Math. Soc*, $1985$. Z. Blocki, The Calabi-Yau theorem. *Lecture Notes in Mathematics 2038*, $2005$ (also on http://gamma.im.uj.edu.pl/ blocki/publ/). Y.T. Siu, Lectures on Hermitian-Einstein metrics for stable bundles and Kähler-Einstein metrics. *Birkhäuser*, $1987$. X.X. Chen, On the lower bound of the Mabuchi energy and its application, *Int. Math. Res. Notices*, $2000$. M. Lai, Fully nonlinear flows and Hessian equations on compact Kahler manifolds, *PhD thesis, U. Iowa*, $2011$. N.V. Krylov, On a general notion of fully nonlinear second-order elliptic equations, *Trans. Am. Math. Soc*, $1995$.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
We present observations of the Type Ia supernova 2003du obtained with the [Hobby$^*$Eberly Telescope]{} (HET) and report the detection of a high-velocity component in the infrared triplet near 8000 Å, similar to features previously observed in SN 2000cx and SN 2001el. This feature exhibits a large expansion velocity ($\approx$ 18,000 km s$^{-1}$) which is nearly constant between $-7$ and $+2$ days relative to maximum light, and disappears shortly thereafter. Other than this feature, the spectral evolution and light curve of SN 2003du resemble those of a normal SN Ia.
We consider a possible origin for this high-velocity line in the context of a self-consistent spherical delayed-detonation model for the supernova. We find that the feature can be caused by a dense shell formed when circumstellar material of solar abundance is overrun by the rapidly expanding outermost layers of the SN ejecta. Model calculations show that the optical and infrared spectra are remarkably unaffected by the circumstellar interaction and the resulting shell. In particular, no hydrogen lines are detectable in either absorption or emission after the phase of dynamic interaction. The only qualitatively different features in the model spectra are the strong, high velocity feature in the IR-triplet around 8,000 Å, and a somewhat weaker feature near 7,300 Å. The Doppler shift and time evolution of these features provides an estimate for the amount of accumulated matter (decreasing Doppler shift with increasing shell mass) and also an indication of the mixing within the dense shell. For high shell masses ($\approx 5 \times 10^{-2} M_\odot$), the high-velocity component of the line merges with the photospheric line forming a broad feature. A cut-off of the blue wings of strong, un-blended lines (particularly the feature at about 6,000 Å) may also be observable for larger shell-masses. The model SN Ia light curves are little effected except at very early times when the shell is partially optically thick due to Thomson scattering, resulting in larger $(B - V)$ colors by up to $0.3^m$. We apply these diagnostic tools to SN 2003du and infer that about $2 \times 10^{-2}M_\odot$ of solar abundance material may have accumulated in a shell prior to the observations. Furthermore, in this interpretation, the early light curve data imply that the circumstellar material was originally very close to the progenitor system, perhaps from an accretion disk, Roche lobe, or common envelope. Because of the observed confinement of in velocity space and the lack of ongoing interaction inferred from the light curve, the matter cannot be placed in the outer layers of the exploding white dwarf star or related to a recent period of high mass loss in the progenitor system prior to the explosion. We note that the signatures of circumstellar interaction could be rather common in SNe Ia and may have eluded discovery because optical spectra often do not extend significantly beyond 7500 Å.
author:
- 'Christopher L. Gerardy, Peter Höflich, Robert A. Fesen, G. H. Marion, Ken’ichi Nomoto, Robert Quimby, Bradley E. Schaefer, Lifan Wang, J. Craig Wheeler'
title: 'SN 2003du: Signatures of the Circumstellar Environment in a Normal Type Ia Supernova? '
---
Introduction
============
There is general agreement that Type Ia Supernovae (SNe Ia) result from some process involving the combustion of a degenerate C/O white dwarf (WD) [@hf60]. Within this general picture, two classes of models are most likely: (1) An explosion of a C/O-WD with a mass close to the Chandrasekhar limit ($M_{Ch}$), which accretes matter through Roche-lobe overflow from an evolved companion star [@Whelan73]. In this case, the explosion is triggered by compressional heating near the WD center. Alternatively, (2) the SN could be an explosion of a rotating configuration formed from the merging of two low-mass WDs, after the loss of angular momentum due to gravitational radiation [@webbink94; @it84; @pac85]. Candidate progenitor systems have been observed for both scenarios: WD binary systems with the correct period to merge in a Hubble time and an appropriate total mass [@max2000]; and supersoft X-ray sources [@greiner91; @vdh92; @rap94; @kah97] showing accretion onto the WD from an evolved companion. However, there are still open questions about the details of both the merging and accretion processes [e. g. @nomoto82; @benz90; @piersanti00; @nomoto00; @nomoto03].
From the observed spectral and light curve properties, the first scenario appears to be the most likely candidate for the majority of normal SNe Ia. In particular, delayed detonation (DD) models [@k91; @yamaoka92; @ww94] have been found to reproduce the majority of the observed optical/infrared light curves (LC) and spectra of SNe Ia reasonably well [@h95; @fisher95; @hk96; @wheeler98; @lentz01; @h02]. In the DD scenario, a slow deflagration front turns into a detonation. We note, however, that detailed analyses of the observed spectra and light curves indicate that mergers may still contribute to the supernova population [@hk96; @hatano2000]. For recent reviews see @branch99 [@hn00; @hetal03].
In both of these scenarios, a certain amount of loosely bound material associated with the mass transfer is likely to remain in the system at the time of the explosion. In the binary mass transfer scenario, the donor star itself may be the source of such material. Based on hydrodynamic calculations, @mb00 found that the SN ejecta wraps around the donor star and, depending on the donor’s evolutionary phase, may strip off up to several tenths of a solar mass of H/He rich gas. As a result, H or He might be observed with expansion velocities of a few hundred km s$^{-1}$. To date, however, no convincing evidence for this kind of stripped material has been observed.
Indeed, a number of possible signatures of interaction with a circumstellar environment have been studied, including X-rays [@schlegel93; @schlegel95], radio [@bb95] and, most often, narrow absorption and emission lines due to , , and lines [@chugai86; @ltw92; @branch95]. However, most SNe Ia show no observational evidence for any of these indicators. The most comprehensive search for emission lines been done for SN 1994D by @cumming96 who found an upper limit for the progenitor mass loss of $1 \times 10^{-5} M_\odot$.
Similarly in the case of WD mergers, the exploding star is expected to be surrounded by debris from the merging process which will not undergo thermonuclear burning [e. g. @benz90]. @k93 showed that the interaction of the supernova with this material may lead to a shell structure in the ejecta, and the observable consequences for the SN light curve and spectra have been studied subsequently [@hk96]. Although a few events (e. g. SN 1990N; @hk96) show some evidence for this kind of structure, a strong case for such a shell has yet to be found.
Thus, despite significant progress in our understanding of Type Ia supernovae, we still have few observational constraints on the progenitor environment. Likewise, little is known about the variety of the progenitor systems [@wheeler91].
However, the recent discovery of strong hydrogen lines in SN 2002ic [@hamuy2003] has drawn new interest in this subject, and emphasized the importance of detecting circumstellar material as a tool for understanding the progenitor system. Whereas most of the optical spectrum of SN 2003ic closely resembles that of a normal SN Ia, it also exhibits H lines similar to those seen in SNe IIn with both broad and narrow components. Significantly more than $0.1 M_\odot$ of H-rich material is required to explain the features in SN 2003ic, with the H-rich gas at distances between $10^{16}$–$10^{17}$cm. This matter might be attributed to a short period of high mass loss in a binary system or during a planetary nebula phase several thousand years before the explosion [@hamuy2003; @wang2003b; @livio2003].
In this work, we revisit the question of circumstellar interaction in SNe Ia in the context of a high-velocity component of the infrared triplet feature, which we observe in the optical spectra of SN 2003du. Similar strong, high-velocity components have been observed in SN 2001el, an otherwise normal SN Ia [@kris03], and also in the unusual supernova SN 2000cx [@li01; @thomas03]. In SN 2001el, this high-velocity feature was well separated in velocity space from the photospheric calcium, and was strongly polarized [@wang2003a]. @wang2003a ([-@wang2003a]; see also @kasen2003) suggest that this feature in SN 2001el could be a consequence of nuclear burning in the WD (perhaps during the deflagration to detonation transition) which causes the ejection of a high-velocity, Ca-rich filament. Alternatively, they suggest that it might be attributed to the surrounding accretion disk likely having undergone nuclear burning to increase the Ca abundance.
We note that observations of some other SNe Ia have also shown a high-velocity component of the IR triplet (e.g. SN 1994D; @hatano99 [@fisher00]), which may be understood as a transient ionization effect when recombines to [@hwt98]. However, given the steep density gradient expected in the outer regions of the SN ejecta, it may be difficult to create a high-velocity feature as persistent as those seen in SN 2000cx, and SN 2001el (or in SN 2003du).
Through a quantitative study of the formation of this IR Ca II feature and related spectral properties, we examine a possible signature of solar abundance circumstellar matter. In § 2, we discuss the observations and data reduction. In § 3, we use detailed models for the explosion, light curves and spectra to study the signatures of the accumulation of hydrogen-rich matter by a SNe Ia. We then develop these signatures as diagnostic tools for probing the circumstellar environment of SNe Ia. In § 4, these tools are applied to the observations of SN 2003du, and we show that the spectral features are consistent with a H-rich shell formed by the interaction with matter related to the mass transfer in the progenitor system. Finally, in § 5 we discuss the results in the general context and address the limits of our study.
Observations and Data Reduction
===============================
SN 2003du was discovered in UGC 9391 by LOTOSS [@schwartz03] on 22 April 2003 (UT) at about $15.9^m$. It was classified as Type Ia by @kotak03 on 24 April 2003 and resembled SN 2002bo about two weeks before maximum light.
We obtained low-resolution (${\rm R} \approx 300$) optical spectra of SN 2003du using the Marcario Low Resolution Spectrograph (LRS; @hill98) on the Hobby$^*$Eberly Telescope (HET; @ramsey98). For each epoch, two different setups were used giving effective wavelength ranges of 4100–7800 Å and 5150–10 000 Å. The data for each setup were reduced separately using standard IRAF routines, and then flux-matched in the overlapping region and combined to create a single 4100–10 000 Å spectrum for each night. Relative spectrophotometric calibration was accomplished by observing standards from @massey88, @massey90, and @oke90. We are not able to obtain accurate absolute fluxes with LRS due to the time-variable effective aperture of the HET, and so the absolute flux levels of these spectra are only approximate.
In addition to the HET spectroscopic observations, we also imaged SN 2003du with the 0.45 m robotic ROTSE IIIb telescope [@akerlof03] to obtain an un-filtered broad-band light-curve. ROTSE IIIb observed SN 2003du at roughly 1 hour intervals during each night when conditions were safe to open the telescope enclosure. The data were processed using the ROTSE automated reduction pipeline software, which delivers magnitudes for every source identified, relative to the red magnitudes of USNO A2.0 stars in the field. The pipeline photometry typically show a scatter of about 0.1 – 0.2 mag for all well-detected sources. (These data have been acquired during the commissioning phase of ROTSE and test images suggest that the reduced CCD frames are significantly more accurate than the results of the current ROTSE pipeline.) For this work, we averaged the output of the pipeline data for each night and the resulting light curve is presented in Figure \[sn93du\]. The arrows indicate the timing of the HET observations, as well as the timing of one epoch, near maximum-light, of [*UBVRI*]{} observations obtained using the 0.8 m telescope at McDonald Observatory.
The observed spectroscopic evolution of SN 2003du is presented in Figure \[spectra\]. These spectra exhibit a high-velocity blueshifted component in the feature near 8000 Å. The Doppler shift of this feature ($\approx 18,000$ km s$^{-1}$) remains well above the photospheric velocity ($\approx$ 11,000 km s$^{-1}$) until the feature fades shortly after maximum light. Note that although the centroid of the absorption feature shifts somewhat to the red as it evolves (Fig. \[spectra\_ca\]), this is due to the blue side fading earlier than the red, rather than a shift of the entire feature. The high-velocity absorption remains constrained in velocity space, and in particular, the red edge of the absorption remains constant at about -13,000 km s$^{-1}$. In all other respects, the observed optical spectral evolution closely resembles that of a normal Type Ia supernova such as SN 1994D [@branch96; @branch99]. The near-maximum [*UBVRI*]{} colors (Table \[colors\]) of SN 2003du are also consistent with the expected colors of a normal SN Ia [@phillips99].
To provide further constraint on the color evolution, we convolved the reduced HET spectra with the [*BVRI*]{} filter functions of @bessel90 to obtain effective photometry for each of the HET observations. Zero-points for the effective photometry were obtained by convolving the same filter functions with the synthetic spectrum of Vega presented by @castelli94. While the absolute fluxing of the HET data is poor, the relative spectrophotometry is accurate and thus the derived colors are meaningful. (Note that for the [*B*]{}-band we performed the convolution over only part of the Bessel bandpass, since the data cut off below 4200 Å. For this reason, the HET $(B-V)$ colors are somewhat less accurate than the $(V-R)$ and $(R-I)$ colors.) The resulting colors are presented in Table \[colors\], along with the real [*UBVRI*]{} colors measured on 07 May 2003. Comparison of the HET derived colors on 06 May and 08 May with the [*UBVRI*]{} measurements suggest that the HET derived colors are probably accurate to about 0.1 mag. When compared to the @riess96 templates, the observed SN 2003du colors lie within the “1-$\sigma$” region for normal SNe Ia.
Model Calculations
==================
To examine the effects of interaction with circumstellar material, we begin with a 1-D delayed detonation model for the supernova. The chemical and density structure of the outer regions of the SN ejecta are modified to model the hydrodynamic effect of the circumstellar interaction. The result is then input into a radiation transport code to calculate synthetic non-LTE light curves and spectra (see Appendix). Previous analyses of the feature for SN 2000cx and SN 2001el (@kasen2003 [@thomas03]) are based on 3D parameterized density structures and assuming LTE population numbers, leaving a large number of free parameters. In contrast, we try to minimize the number of free parameters by using more realistic physics but a spherical configuration. Our study presented here is based on detailed explosion models and NLTE light curves and spectra. For example, in our analysis the velocity of the shell is linked directly to the mass of the shell, and non-LTE provides a proper treatment of the ionization balance and detailed atomic models, needed since the level populations and opacities in supernovae are are very different from LTE. In the conclusions, we discuss the limitations of our models and how asphericity will influence the results.
Delayed-Detonation SN model
---------------------------
Our study is based on the delayed detonation scenario [@k91] which has been found to reproduce the optical and infrared light curves and spectra of typical SNe Ia reasonably well [e.g. @h95; @hk96; @nugent97; @fisher95; @wheeler98; @lentz01; @h02]. It also provides a natural explanation for the brightness decline relation [@phillips93; @hamuy96a; @hamuy96b] as a consequence of opacity effects in combination with nearly constant explosion energies for SNe Ia [@h96; @maeda03].
Hydrodynamic explosions, light curves, and synthetic spectra are all calculated self-consistently using only physically motivated connections between the different calculations. Given an initial structure for the progenitor and a description of the nuclear burning front, the light curves and spectra are calculated directly from the explosion model without any additional tunable parameters. This methodology forges a strong link between the physical processes being modeled and the predicted observables. As a result, comparing the model results with observations can provide a great deal of insight, at the expense of having to perform rather more difficult calculations.
For this study, a single SN model (Fig. \[model\]) was chosen, which roughly matches the observed properties of normal Type Ia supernovae. However, no attempt has been made to fine-tune the SN model to “fit” the observations of SN 2003du. The details of the numerical methods and the resulting SN model used for these calculations are presented in Appendix A.
Circumstellar Interaction Model
-------------------------------
We consider the case of homologously expanding SN envelope running into a stationary circumstellar medium of solar-abundance gas. We model the region of circumstellar interaction in a manner similar to @chevalier82 and the density profiles of the interaction region are based on his self-similar solution. The collision of the SN ejecta and environment sets up a forward- and reverse-shock structure, separated by a contact discontinuity at a distance $R_c$. This produces a high-density region with shocked ejecta and swept up matter, subsequently referred to as the shell, and a low-density precursor region of ambient gas. The inner edge of the shell is given by the location ($R_{sh}$) of the reverse, adiabatic shock. Because the shock front is Rayleigh-Taylor unstable [@chevalier82; @dwarkadas98] we assume that the reverse shocked region consists of a mixture of SN and circumstellar matter. We use a rather generic profile for shell because the details of the structure depend on the exact location, origin and morphology of the environment. Moreover, we apply parameters, which result in a maximum density contrast of four between SN ejecta and the shocked region as expected for adiabatic shocks whereas other conditions, such as nearby interaction (Höflich & Khokhlov 1996) or more sophisticated environments [@dwarkadas98] produce shells with higher contrast. In such models, the strength of the individual features may be enhanced but we do not expect significant changes to our deceleration of the ejecta and the resulting position of the shell (in velocity) as a function of circumstellar mass. These parameters depend on energy and momentum conservation and are thus relatively insensitive to such details. For the calculation of synthetic spectra, we assume that the circumstellar matter originates close to the supernova and has been overrun by the SN ejecta well before the observations. As a consequence, the structure of the interaction region undergoes free homologous expansion in our model structure, with little or no appreciable emission from the forward shock. Furthermore, at the time of the observations, rapid expansion and radiative cooling will have erased the thermal signature from the shock interaction itself. The temperature structure of the expanding shell is calculated by taking into account radiative processes, gamma-ray heating, and adiabatic cooling. Thus the interaction will not significantly contribute to the luminosity if the interaction takes place early on, i.e. at distances small compared to the SN photosphere at the time of the observation. (As we discuss below, the light-curve of SN 2003du indeed suggests that any strong circumstellar interaction occurred before its discovery around two weeks before maximum light.)
We note that this model makes certain implicit simplifying assumptions, such as spherical symmetry, power law density profiles for the ejecta and the surrounding medium, and pure adiabatic shocks. However, the qualitative results are relatively robust. While deviations away from these assumptions will tend to change the quantitative details of the resulting shell structure somewhat, the models are sufficient for the qualitative and order-of-magnitude quantitative analysis presented here.
Model Results
-------------
### General Considerations
The basic result of the interaction is the accumulation of circumstellar matter (of mass $M_{acc}$) in a shell (of mass $M_{sh}$). As the highest velocity ejecta runs into the reverse shock, it is decelerated down to the shell velocity $v_{sh}$. Some of the basic quantities of the shell are shown in Figure \[shell\]. Both the shell velocity and the total amount of kinetic energy converted by the interaction depend on $M_{acc}$. Typically, the shell consists of about 1/3 accumulated solar-abundance circumstellar matter and 2/3 SN ejecta with a composition depending on the shock velocity (compare Figs. \[model\] & \[shell\]).
For a given SN explosion model, the relation between the shell mass and its velocity is fixed. Within the delayed-detonation scenario, the energy generation is nearly independent of changes in the explosion model because for $M_{Ch}$ models, the mass and density structure of the SN envelope hardly changes, although the chemistry of the outer envelope is different [@h02]. Conversely, only part of the C/O WD is burned in a pure deflagration scenario, reducing the kinetic energy of the ejecta. As a consequence, we expect lower shell velocities for those models.
As example cases, we consider shells which are produced by running into (case I) a stellar wind with a velocity $v_{RSG}$, (case II) a combination of a nearby mass and a stellar wind, or (case III) a constant density environment. For the first two cases, the mass accumulated in the shell at time $t$ can be obtained by scaling the relation for a constant mass loss rate $$M_{acc}(t)= M_{acc}(t=0)+ \dot{M} {v_{sh}(t) \over v_{RSG}} \times t$$ where $v_{sh}$, $v_{RSG}$ and $\dot{M} $ are the shell velocity, the wind velocity of the progenitor system and the mass loss rate, respectively. For an environment of constant particle density $N$ (case III), we have $$M_{acc} = {4 \pi \over 3 N_{av}} \mu _e * N \times R_c^3$$ where $\mu _e$ and $N_{av}$ are the mean molecule mass and Avogadro’s number, respectively.
For this discussion, we consider H-rich shells produced by the accumulation of $M_{acc} (t=20d) = 2.\times 10^{-2}$M$_\odot$. The properties of the system are $\dot M = 2.\times 10^{-4} $M$_\odot$ yr$^{-1}$ and $v_{RSG}=10$ km s$^{-1}$ (case I), $\dot M = 1.\times 10^{-5} $M$_\odot$ yr$^{-1}$ plus an early accumulation of $1.98 \times 10^{-2}
$M$_\odot$, (case II), and a constant particle density $N=1.2 \times 10^{8}$g cm$^{-3}$ (case III).
### Conversion of Kinetic Energy
A general study of light curves for interactions between the SN shell and the environment is well beyond the scope of this paper. Therefore, we want to present a qualitative discussion which let us to select our setup for the detailed calculation. The arguments are based on energy and momentum conservation and should apply rather generally but, as a consequence, the conclusions are rather qualitative, and should be regarded as a check for consistency rather than than quantitative. In the upper right of Figure \[shell\], we plot the rate of kinetic energy conversion in the shock. Though some of this energy will go into turbulent motion, and ionization of the gas, we expect significant modification of the X-Ray/UV/optical radiation [@fransson96]. To decelerate the highest velocity ejecta to about about 20,000 km s$^{-1}$ as observed in SN2003du, the energy gain by continuous accretion exceeds those of the underlying SN light curve (powered by radioactive heating via $^{56}$Ni and $^{56}$Co decay) by an order of magnitude. Thus, we expect significant modification of the light curves for case I and III, and even case II unless the wind component is very small.
For SNe Ia, such as SN 2003du, that exhibit essentially normal light curves, we can therefore rule out that the accumulated matter originates predominantly from mass loss over an extended period of time or from a constant density environment. In our light curve (Fig. \[sn93du\]), we do not see any significant additional energy input 60 days after the explosion (about 40 days after maximum) which, in the context of this analysis, limits $\dot M$ to about $ \leq 1. \times 10^{-5} $M$_\odot$ yr$^{-1}$. We note that this limit is consistent with upper limits for the mass loss in SN 1994D by Cumming et al. (1996) based on the lack of observed $H_\alpha $ emission. However, since we have not calculated detailed self-consistent LCs with ongoing interaction it is not clear exactly how much ongoing interaction could be actually hidden in normal SNe Ia. More detailed analysis may require correction factors for this simple analysis although LCs and spectra of SNe are rather sensitive to a change in the energetics. In any case, it is worth noting that the additional energy will likely dominate the LC at some point in time and decrease its slope. However, late time light curves of typical SNe Ia do not show evidence for any additional energy source even after several years [e. g. @wells94; @schmidt94].
### Light Curve and Spectral Signatures
In light of this constraint, we will consider shells that are formed early on in absence of continuous mass loss (case II with $\dot M = 0$). Such a shell might originate, for example, from an accretion disk, the Roche-lobe of a companion donor star, or a common envelope of the progenitor system. To construct the density structure of the expanding shell, we assume the ejecta freely expand out to $10^{13}cm$ before running into circumstellar matter of constant density with a thickness of $R_C-R_{sh}$. To construct the shell, we assume power law densities for the outer SN ejecta ($\rho_{SN} \propto r^{-n}$), with $n$ given by the explosion model (close to $n \approx 7$; see Fig. \[model\])
Light curves and detailed spectra have been calculated for a SN Ia without a shell, and for cases with shell masses of $2,$ and $ 5 \times 10^{-2} M_\odot $ at several epochs. In Figure \[model\_lc\], the evolution of $V$ and $(B-V)$ is given. Overall, even the high mass shell has comparably little effect on the light curves because it hardly influences the diffusion time scales and the lack of energy generation without ongoing interaction (see Figs. \[shell\] & \[model\_spec\]). Because the Thomson optical depth of the shell scales with about $t^{-2}$, it is partially optically thick up to about 10 days after the explosion. As a result, the model photosphere becomes somewhat cooler and redder ($\Delta (B-V) \leq 0.3^m$) but still within the observed variations for normal SNe Ia. For illustration, optical and near IR spectra are shown in Figure \[model\_spec\] for day 15 (about 3 days before maximum light). We chose to exhibit this epoch of the spectral evolution as an example because the high-velocity feature is well developed (the photosphere is far inside the shell region). Furthermore, at this epoch, the transient high-velocity feature due to the recombination of to is at its most pronounced [@hwt98]. Thus thus epoch represents a “worst-case” for distinguishing between the shell and ionization interpretations of the high-velocity calcium feature. We will discuss this potential complication in detail in § \[HVCa\_section\].
Overall, the model spectra with the three different shell masses are remarkably similar, although large electron cross sections and backscattering cause some “smearing out” of the line features in the higher shell mass models. This insensitivity of the overall spectrum to the shell mass is a result of the large distance between the shell and the line forming region. Nearly all of the spectral features are formed in regions well inside the shell, and are therefore largely unaffected by the interaction. There are a few places where and lines form subtly different features in the model spectra. However, these are highly blended with lines of singly and doubly ionized iron-peak elements. Furthermore, these weak features are below the accuracy of our numerical models (primarily due to the depth discretization; see Appendix). Models with an analytic density structure are more suitable for investigation of these features. (See Branch et al. 2003). In most optical and IR features, the main difference is the cutoff at the blue edge of strong lines due to the deceleration of the high-velocity ejecta by the reverse shock (see Fig. \[model\_si\]). Still, for the most part, the shell has little effect on the spectra beyond quantitative changes at the level of the intrinsic variability in normal SNe Ia. Even shells with about $5 \times 10 ^{-2}M_\odot$ might elude discovery.
However, there are two features that exhibit clear qualitative rather than quantitative changes with increasing shell mass. In the presence of a shell, high-velocity components appear in the IR triplet at about 8000 Å and in a feature between 7300 and 7500 Å due to a blend of / (Fig. \[model\_ca\]). Because these features are qualitatively different in the interaction scenario, they can be used to probe for such an interaction without having to rely on the detailed quantitative accuracy of model spectra.
With a strong high-velocity feature forming in the IR-triplet, a similar feature may be expected in the H&K lines. However, in our models the spectral region below 4000 Å, containing the H&K lines, is highly contaminated with other features. At early times lines of Fe/Co/Ni III populate this spectral region, and beginning about 7 days before maximum light a strong appears at the expected wavelength of a high-velocity component for the H&K features. In contrast, the IR triplet lies in a comparatively empty spectral region, making it much better suited as a spectral diagnostic. For further discussion of high-velocity components in the H&K lines, see Branch et al. 2003.
### High-Velocity IR Triplet Feature\[HVCa\_section\]
In the models which include a shell from circumstellar interaction, a rather strong high-velocity component appears in the IR triplet (Fig. \[model\_ca\]), with a Doppler shift corresponding to the expansion velocity of the shell (Fig. \[shell\]) rather than the photospheric velocity. In combination with explosion models (but actually relatively insensitive to the model details within a delayed-detonation scenario), this provides a good measure for the mass of circumstellar material accumulated in the shell. For example, with $M_{acc} = 2 \times 10^{-2}M_\odot$, and $ 5 \times 10^{-2}M_\odot$ the high-velocity Doppler shift corresponds to about 19,000 km s$^{-1}$ and 14,000 km s$^{-1}$, respectively. In the model with the higher shell mass the deceleration of the outermost SN ejecta is sufficient to cause a partial merging of the high velocity component with the photospheric line. The resulting feature looks more like a very broad feature rather than two distinct components.
However, Figure \[model\_ca\] also illustrates a complication such an analysis. At the epoch shown in the top panel of Figure \[model\_ca\] (about three days before maximum light), the model without an accreted shell also shows a strong high-velocity component to the IR triplet feature. As we noted in the introduction, observations of some SNe Ia (e.g. SN 1994D) show a short-lived high-velocity component of IR triplet shortly before maximum light. In SNe Ia where the high-velocity feature is short-lived, or observed at only a single epoch near maximum light, this feature can also be interpreted as a transient ionization effect as recombines to [@h95; @hwt98; @lentz01]. During this recombination phase, models for normal SNe Ia show an outer and inner region of separated by . The recombination from to occurs around maximum light over the course of a couple of days. As a result, a two-component feature due only to this recombination effect is rather short-lived, and the Doppler shift of the high-velocity absorption component recedes rapidly and merges with the photospheric feature. Because of the steeply declining density gradient in the ejecta, a high-velocity component formed in this way is rather weak, only about 20% relative to the continuum in our model.
In contrast to this transient ionization effect, the high-velocity component formed by a dense shell evolves much more slowly, and has a Doppler shift that is nearly constant if the mass of the shell is not significantly changing. The time evolution of features due to these two effects are compared in the lower panels of Figure \[model\_ca\]. The high-velocity feature in the shell model can be seen as early as 5 days after the explosion, while the feature due to the ionization effect is visible only in the day 15 spectrum. In both cases, the feature disappears around maximum light.
Note that even in a SN with a statically evolving shell, a slight shift in the centroid of the high-velocity absorption is to be expected. Since the shell is also undergoing free homologous expansion, the outer regions of the shell will undergo geometric dilution faster than the inner edge. As a result, the feature fades faster at the blue, high-velocity end than the red end, shifting the feature’s centroid to the red.
### High-Velocity Feature
The other new feature in the models which include a shell is an absorption between 7300 and 7400 Å due to an blend (7775, 7774 & 7772 Å). This feature shows the same Doppler shift and evolution with time as the feature (i.e. related to the dynamics of the shell and not the photosphere). However, unlike the Ca feature, the oxygen is not primordial but produced during explosion. Burned matter contributes about 2/3 of the shell mass (see Figs. \[model\] & \[shell\]). Solar abundances of oxygen do not provide sufficient optical depth in the shell to cause a strong feature. The appearance of this feature would imply that a significant amount of burnt SN ejecta has piled up in the dense shell.
### Hydrogen, Helium, & Carbon
Although 20 % of the shell consists of hydrogen, the model spectra do not exhibit any significant Balmer or Paschen lines, even for $M_{acc} $ of $5 \times 10^{-2} M_\odot$. In the absence of heating by an ongoing interaction between the SN and its environment, the temperature in the shell is low ($T\approx$ 4500 – 5000 K). At these temperatures, excitation of metal lines is strongly favored over hydrogen, due to the much lower excitation energies. (e.g. 10.2 eV for vs. 2.22 eV ($1D \rightarrow 1S$), 9.0 eV ($3P \rightarrow 3S$), 1.57 eV ($5S
\rightarrow 5P$) for the singlet, triplet, and quintet state of calcium and, 1.9 eV for collisional coupling between the singlet and triplet state). The optical depth of the shell is small for $\gamma $ rays ($\leq 0.01$ in all cases), keeping the non-thermal excitation of hydrogen low, and thus the hydrogen is mainly neutral. Since most atoms are in the ground state, the optical depth of the shell is very small for Balmer and Paschen lines and no absorption features are produced in the models. Furthermore, the models do not predict any significant emission component due to hydrogen. In the presence of heavy elements, strong charge exchange reactions between hydrogen and metals (with a lower ionization potential) are the preferred path for H recombination rather than the radiative process. These same arguments hold even more true for . Note that while strong charge exchange reactions are expected to suppress H and He features, they are not really needed because, even without microscopic mixing, detection of the broad, weak H emission features ($\approx$ 1 to 2 % of continuum in case II) would hardly be possible.
On the other hand, because carbon and oxygen have excitation energies and line cross sections closer to those of Ca, we could expect strong features due to transitions at 9405 Å, and 10691 Å (as was observed in the subluminous SN 1999by; @h02). In normal-bright delayed detonation models, these lines do not show up strongly because nearly all of the WD is incinerated, leaving little carbon in the ejecta. However, in alternative scenarios for SNe Ia such as deflagration and merger models, more than 0.1 $M_\odot$ of the WD remains unburned. For example, in the W7 model [@nomoto84], about $0.17 M_\odot$ does not undergo nuclear burning and, as a consequence, approximately equal amounts of O and C would be seen in the shell. A clear detection of this high-velocity feature in would indicate that the outermost ejecta had been trapped by the circumstellar interaction and piled up in a dense shell. Any significant ammount of unburned carbon in these outer layers would also necessarily result in strong high-velocity features of . A lack of high-velocity in the presence of high-velocity would therefore argue for nearly complete burning of the WD progenitor.
Comparison with Observations
============================
What follows is a comparison of the above model predictions with our observations of SN 2003du, to determine whether such an interaction scenario might plausibly reproduce the observed phenomenon. However, we will not attempt to make a detailed comparison between theory of the underlying SN explosion and these observations as has been done, for example, for SN 1994D and SN 1999by; [@h95; @h02]. Such an analysis would involve fine tuning of the supernova model parameters: the initial progenitor, properties of the burning front and the central density of the exploding WD. Instead, we have simply used a delayed-detonation model which roughly reproduces the features of a normal Type Ia SN, without attempting to tune this model to match the specific features of SN 2003du. As an example, in Figure \[obsthe\], the theoretical spectra have been plotted at about 3 days before maximum light, along with the May 06 data. The continuum slope of the model roughly matches the spectrum and the Doppler shifts of lines generally agree to within about $ \approx 1000$ km s$^{-1}$. (Note that the vertical offset in the red is an artifact due to a poor match of the calibration between the two spectroscopic setups.) Thus our model, although not tuned in any manner, is a decent match for SN 2003du.
Figure \[obsthe\_ca\] shows a blow-up of the comparison of Figure \[obsthe\], showing the region of the IR triplet feature. The strength and Doppler shift of the high-velocity component of the is roughly consistent with $M_{acc} \approx 2 \times 10^{-2}M_\odot$. The observed line profile is broader than the model profile, which is an indication that the absorbing material is somewhat less confined than in the model. As seen in the observations, the model predicts that the high-velocity starts to become weak around maximum light, and disappears a few days later. The higher mass model can clearly be ruled out, as the high-velocity feature in this model is essentially completely blended with the photospheric absorption, and does not appear as a well separated feature at all.
The model without a shell does exhibit a high-velocity component to the feature, although the total absorption is somewhat weaker than the observations (at least with this SN model). However, this epoch (three days before maximum light) was specifically chosen as a “worst case” for distinguishing our interaction scenario from the short lived ionization effect discussed above. The high-velocity feature in the model without a shell exhibits an entirely different evolution than is seen in SN 2003du. As the SN ejecta expands and cools, the inner and outer regions of the region begin to recombine to . Thus the observed high-velocity feature would shift rapidly to the red, merging with the photospheric line, and fading over the course of 2 – 3 days.
In contrast, the observed evolution of the high-velocity feature more closely resembles the expected evolution for a shell that is fixed in velocity space. The line centroid shifts slightly to the red as the feature ages, but the absorption never moves beyond its upper or lower boundaries. In the shell model, the slight shift of the minimum in the high-velocity component occurs because the inner layers of the shell contribute more to the opacity as it expands homologously, and also because the absorption depression is formed on the steep blue edge of the main component of , which gains considerable strength during the observations. Note that the observed red edge of the high-velocity feature, corresponding to the inner edge of the shell, remains constant. This is consistent with the notion that the kinematics of the shell are unchanged, and only the relative density of the inner and outer regions of this finite thickness shell are evolving.
Such kinematic stability suggest a shell in free expansion, which implies that most of the circumstellar matter must have been accumulated prior to the first spectroscopic observations. This, in turn, imposes an upper limit for the distance between the circumstellar material and the WD progenitor of $D_{matter} \leq v_{shell}* t \approx 1.5 \times 10^{15}$cm. Such a distance constrains the origin of the material, which might originate from an accretion disk around the white dwarf, mass filling the Roche lobe of the donor star, a common envelope in which the WD is embedded, or perhaps a period of very high mass loss immediately prior to the explosion. The last is rather unlikely, as the period of mass loss must be shorter than 50 years prior to the explosion, even if we assume a wind of 10 km s$^{-1}$ which is at the low end for red supergiant winds. Moreover, the interaction with such a wind would convert bulk flow kinetic energy at a rate of $2 \times 10^{44}$ erg s$^{-1}$ (Fig. \[shell\]), a significant fraction of which should be visible as additional luminosity, but we see no evidence for any significant excess in the observed light curve.
A further constraint on the source of the circumstellar material can be inferred from the early light curve. SN 2003du was discovered on April 22, about 3 days after the explosion in the timeline of our SN model. At this epoch, SN 2003du was about $2.5^m$ below its maximum light brightness, which is typical for normal SNe Ia (e.g. Phillips et al. 1999). A strong interaction with the surrounding CSM at such an epoch would likely dramatically alter the observed light curve as some fraction of the kinetic energy of the ejecta is converted into light. Since the light curve of SN 2003du does not appear to have been highly unusual, we can infer that the bulk of any extended interaction occurred prior to the supernova’s discovery. This leads to an estimate for $D_{matter} \leq v_{shell}* t \leq 6 \times 10^{14}cm$. Furthermore, since the kinetic energy conversion is proportional to the mass accumulation rate in the shell, to produce the same shell by interaction with a stellar wind would either result in very unrealistic luminosities early on, or the deposited energy has to go into expansion work rather than luminosity.
Indeed, the lack of any observed interaction in the light curve argues that the bulk of the interaction likely occurred close to the progenitor system, so that adiabatic expansion of the freely expanding shell dissipated most of the shock energy. These constraints tend to argue that the circumstellar material is most likely directly related to the progenitor system as in the accretion disk or Roche lobe scenarios, rather than an effect of the random environment around the progenitor at the time of the explosion.
Note that for small $D_{matter}$, a shell may not stay as confined as assumed in our models since there will have been more time for internal dynamics to “smear-out” the sharp edges of the shell. Such an effect would tend to broaden the shell features, and produce an absorption more akin to the high-velocity Ca feature observed in SN 2003du. If the shell were significantly non-spherical (as in the case of an accretion disk, for example) the kinematics would also likely be significantly affected.
On the other hand, because of the observed confinement of in velocity space, the matter cannot be attributed to the outer layers of the WD itself. Redistribution of energy during the hydrodynamical phase would produce a smooth velocity profile very similar to the freely expanding ejecta (Fig. \[model\]). Since no dense shell would be formed in the outer layers, any material from the surface of the WD would have a very low optical depth and the observed spectrum would look essentially identical to a normal SN Ia.
Final Discussion
================
[**Summary:**]{}\
We have presented a series of optical spectra and broad band photometry of SN 2003du showing that its spectral evolution and light curve resemble that of a normal bright SNe Ia, except for a high-velocity component to the IR triplet near 8000Å. This high-velocity component exhibits an expansion velocity of about 18,000 km s$^{-1}$, significantly larger than the photospheric velocity implied from the other spectral features. We propose that this high-velocity Ca feature may be the spectroscopic signature of a shell formed by the interaction of the supernova ejecta with the circumstellar environment.
Based on model calculations for the explosion, light curves and spectra, the observable effects of such a shell have been examined. The model shell is formed by interaction between SN ejecta and circumstellar material of solar abundance, with the assumption that the circumstellar material has been overtaken by the expanding envelope prior to the time of our observations, with little or no significant ongoing interaction. We find that the high-velocity component of the IR triplet can indeed be understood as being caused by solar abundance material piled up in a dense shell behind the shock. This material is likely mixed via Rayleigh-Taylor instabilities with the reverse-shocked outer layers of the SN ejecta. Overall, the SN light curves are little effected except at very early times when the shell is partially optically thick due to Thomson scattering resulting in larger $(B-V)$ colors by up to $0.3^m$. Similarly, we find that the optical and IR spectra are otherwize little effected by a shell with an accumulated mass of up to a few hundredths of a solar mass. In particular, hydrogen and helium lines are strongly suppressed due to the low temperatures in the shells and charge exchange with heavy elements being the preferred method of hydrogen recombination.
The main signatures of the shell in our model spectra are the high-velocity component of the IR triplet, and a weaker line near 7300 to 7400 Å. Although similar in nature, these features probe different effects. The Doppler shift of the high-velocity component of is a sensitive measure of the amount of accumulated matter in the circumstellar shock, whereas the Doppler shift of the can be used as a test for mixing of shell and envelope material. Unlike the feature, the feature is only formed with oxygen abundances which exceed the solar value by about 2 orders of magnitude.
Related to the interaction process and the deceleration of high-velocity ejecta is the conversion of kinetic energy. Assuming reasonable factors of efficiency for the conversion to photons, the lack of evidence for any additional associated luminosity strongly argues in favor of a very low density environment for SN 2003du (outside of the immediate vicinity of the progenitor system) and provides additional constraints for the origin of any circumstellar material.
For SN 2003du, we find that the observations can be understood as a result of a shell formed by the accumulation of about $2 \times 10^{-2}M_\odot$ of solar composition material. The early light curve data strongly suggest that the matter originates from close to or within the progenitor system, as an accretion disk, the Roche lobe of the companion, or a common envelope. An episode of strong mass loss would require unrealistically short durations, very high mass loss rates and probably very high luminosities early on.
[**Relation to Observations of Other SNe Ia:**]{}\
Our models indicate that moderate mass shells of around $10^{-2}M_\odot$ show no significant signatures in the optical and NIR except the high-velocity feature near 8000 Å and the feature at about 7300 Å. In the past, both could easily elude detection because observations often did not extend far enough to the red, and even when they did, the increasing noise longwards of 7000 Å and contamination by atmospheric absorption bands confuse the issue. SN 2003du may not, in fact, be all that unusual an event. Other SNe Ia (SN 2002cx, SN 2001el) have shown similar features, and a systematic search for this phenomenon in SNe Ia may provide significant constraints on the environment and progenitor systems of these objects.
In light of these results, the interpretation of the high-velocity IR feature seen in SN 2001el and SN 2000cx should be revisited. Our results here suggest that calcium enrichment may not be required to reproduce the effect seen in SN 2001el. The stronger feature and the observed polarization provide somewhat stronger constraints, however, and will require a more detailed calculation which is beyond the scope of this work. We note, however, that the circumstellar phenomena we suggest here for SN2003du are unlike those for SN 2003ci which require a large H mass at large distances ($\approx 10^{16}$–$10^{17}$ cm; @wang2003b)
[**Limitations and Future Work:**]{}\
The analysis presented here, while suggestive, does not provide a definitive conclusion as to the nature of the high-velocity Ca absorption, as alternative explanations may still be viable. In particular, we have not ruled out that the Ca feature is produced by processed Ca-rich material originating from burning under very special conditions in a disk or as a consequence of nuclear burning in the WD which causes ejection of a high-velocity, Ca-rich filament. Rather we have shown that interaction with solar abundance circumstellar material is a plausible explanation for this phenomenon.
In examining the implications for such an interpretation, we have developed observational tools which could be used to probe the circumstellar environment. However, these tools still need significant refinement. Our study has only covered a small parameter space and, for the most part, has been limited to the case without ongoing interaction. For the formation of the shell, we assumed adiabatic shocks. Depending on the origin of the accreted matter, some of the shell material may be accumulated before the phase of homologous expansion has been established for the SN ejecta, and the shell structure might therefore be modified. The larger width of the high-velocity component of SN 2003du (compared to the models) may already be an indication of such an effect.
Also, we have not accounted for the likely 3-D nature of either the SN ejecta or the circumstellar material. Indeed, the polarization measurements for SN 2002el [@wang2003a] indicate that the (and thus, in this interpretation, the shell) is not spherical but may be toroidal, or a large scale clump with a significant scale height which selectively blocks light from the underlying photosphere and causes polarization [@h95; @h02; @kasen2003]. Moreover, the polarization confirms that the high-velocity in SN 2001el is related to a region morphologically separated from the overall SN ejecta. The 3-D nature of the absorbing gas will also affect the implied mass of circumstellar material. The high-velocity component of the IR measures a column depth along the line of sight and the the masses quoted here assume spherical symmetry. Thus the actual mass may be lower by a factor of a few depending on the covering factor of the absorbing material. Finally, asymmetry may also change the early light curve because the luminosity becomes directionally dependent (Höflich 1991), with higher or lower luminosity depending on whether we observe the shell from the top or ’equator on’ (the latter showing the strong high-velocity Ca II feature). Within our configuration, the latter would produce a decrease in the *V* luminosity and even redder $(B - V)$ color at early times. To address these questions, detailed 3D calculations for the interaction of ejecta and environment should be performed similar to those for the interaction of the donor star by @mb00
We would like to thank the staff of the Hobby$^*$Eberly Telescope and McDonald Observatory for their support. We would like to thank E. Robinson for helpful discussion. PAH would like to thank David Branch for very useful discussions in Trento. We would like to acknowledge Carl Akerlof, Don Smith, Eli Rykoff, and the members of the ROTSE collaboration for their enormous (and continuing) work on the design, construction, and implementation of the ROTSE telescopes and software. The Hobby-Eberly Telescope (HET) is a joint project of the University of Texas at Austin, the Pennsylvania State University, Stanford University, Ludwig-Maximillians-Universitaet Muenchen, and Georg-August-Universitaet Goettingen. The HET is named in honor of its principal benefactors, William P. Hobby and Robert E. Eberly. The Marcario Low Resolution Spectrograph is a joint project of the Hobby-Eberly Telescope partnership and the Instituto de Astronomia de la Universidad Nacional Autonoma de México.
This research is supported, in part, by NASA grant NAG 5-7937 (PH) and NSF grants AST0307312(PAH) and AST0098644 (JCW).
Details of the SN Model
=======================
Numerical Methods
-----------------
Our calculations take into account detailed hydrodynamics, nuclear networks, radiation transport for low and high energy photons, opacities and include solvers to calculate the atomic level populations in full (albeit simplified) NLTE for both light curves and synthetic spectra utilizing accelerated Lambda Iteration and level merging as commonly used in stellar atmospheres [also known as superlevels, e.g. @hilier03; @hugeny03; @werner03]. These computational tools have been used to carry out several previous analyses of SN Ia [e.g. @h95; @h02]. For more details, see @hoeflich03a [@hoeflich03b], and references therein.
Spectra are computed using the chemical, density, and luminosity structure as well as the $\gamma$-ray deposition resulting from the light curve code. For the detailed atomic models, typically between 27 and 137 bound levels are taken into account for the main ionization stages. For each of these detailed atomic models, neighboring ionization stages have been approximated by simplified atomic models restricted to just LTE levels with a few NLTE levels. The energy levels and cross sections for bound-bound transitions are taken from @kurucz93 starting at the ground state. Here, we use (14/31/91), (15/28/46), (27/123/242), (23/31/57), (43/129/431), (28/43/75), (20/60/153), (35/212/506), (41/195/742), (62/75/592), (137/3120/7293), (84/1355/5396), and (71/865/3064), where the first, second and third numbers in brackets denote the number of levels, the number of bound-bound transitions, and the number of discrete lines used for the radiation transport. The third number is larger than the second because nearby levels within multiplets have been merged for the rates. In addition, 404508 LTE-lines are taken into account using an equivalent-two level approach.
For calculating detailed spectra, the explosion models have been remapped from 912 to about 200 radial grid points for the atmospheres with the zones concentrated in the line forming region. $7.4 \times 10^{4}$ frequency points have been used, oversampling the synthetic spectra. In an expanding atmosphere, the frequency and velocity space are coupled and thus the effective spectral resolution of the synthetic spectra is about 1000. The noise in the spectra (see Fig. \[model\_si\]) below this effective resolution is a direct result of the oversampling in the frequency space and can be used as an estimate for the internal numerical accuracy.
SN Model Results
----------------
The structure of the initial model of the C/O white dwarf is based on a star with 5 solar masses at the main sequence and solar metallicity which, at the end of its evolution, has lost all of its H and He-rich layers. By accretion, its core has grown close to the Chandrasekhar limit. At the time of the explosion of the WD, the central density is 2.0$\times 10^9$ g cm$^{-3}$ and its mass is close to 1.37$M_\odot$. The deflagration-detonation transition density $\rho_{tr}$ is $2.5 \times 10^7$ g cm$^{-3}$. During the explosion, about 0.6 $M_\odot$ of radioactive $^{56}Ni$ are produced. The density and chemical structure are shown in Figure \[model\]. The maximum brightness in *V*, the $(B-V)$ color near maximum, the rise time to maximum light, and decline ratio in *V* over 15 days are $-19.29^m$, $-0.02^m$, 18.6 days, and $1.0^m$, respectively; typical for normal SNe Ia. Further details for this reference model, including the progenitor evolution, explosion, and light curves can be found in @dominguez01 and @h02, where it is named [*5p0z22.25*]{}.
references
[lcccccr]{} 01 May & & $-0.10$ & $0.01$ & $-0.16$ & HET\
02 May & & $-0.02$ & $0.04$ & $-0.07$ & HET\
06 May & & $-0.12$ & $-0.08$ & $-0.37$ & HET\
07 May & $-0.35 \pm 0.06$ & $0.04 \pm 0.05$ & $0.03 \pm -0.05$ & $-0.39 \pm
0.04$ & 0.8 m\
08 May & & $0.00$ & $-0.02$ & $-0.33$ & HET\
12 May & & $0.19$ & $0.09$ & $-0.14$ & HET\
30 May & & $0.58$ & $0.11$ & $0.06$ & HET\
{width="10.7cm"}
{width="13.7cm"}
{width="10.7cm"}
{width="7.7cm"} {width="6.7cm"}
{width="12.7cm"}
{width="12.7cm"}
{width="12.7cm"}
{width="8.7cm"} {width="8.7cm"}
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The goal of *temporal alignment* is to establish time correspondence between two sequences, which has many applications in a variety of areas such as speech processing, bioinformatics, computer vision, and computer graphics. In this paper, we propose a novel temporal alignment method called *least-squares dynamic time warping* (LSDTW). LSDTW finds an alignment that maximizes statistical dependency between sequences, measured by a squared-loss variant of mutual information. The benefit of this novel information-theoretic formulation is that LSDTW can align sequences with different lengths, different dimensionality, high non-linearity, and non-Gaussianity in a computationally efficient manner. In addition, model parameters such as an initial alignment matrix can be systematically optimized by cross-validation. We demonstrate the usefulness of LSDTW through experiments on synthetic and real-world *Kinect* action recognition datasets.'
author:
- |
Makoto Yamada$^1$, Leonid Sigal$^2$, Michalis Raptis$^2$, and Masashi Sugiyama$^{1}$\
$^1$Tokyo Institute of Technology, 2-12-1 O-okayama, Meguro-ku, Tokyo 152-8552, Japan\
$^2$Disney Research Pittsburgh, 4720 Forbes Ave., Pittsburgh, PA 15213\
`{yamada@sg. sugi@}cs.titech.ac.jp`
bibliography:
- 'main.bib'
title: |
Dependence Maximizing Temporal Alignment\
via Squared-Loss Mutual Information
---
Introduction
============
Temporal alignment of sequences is an important problem with many practical applications such as speech recognition [@saoke78; @Rabiner:1993dq], activity recognition [@DBLP:journals/pami/JunejoDLP11; @DBLP:conf/iccv/GongM11], temporal segmentation [@DBLP:conf/fgr/ZhouTH08], curve matching [@Sebastian03b.b.:on], chromatographic and micro-array data analysis [@Listgarten05multiplealignment], synthesis of human motion [@Hsu05styletranslation], and temporal alignment of human motion [@NIPS2009_0760; @CVPR:Feng:2012a]. *Dynamic time warping* (DTW) is a classical temporal alignment method that aligns two sequences by minimizing the pairwise squared Euclidean distance [@saoke78; @Rabiner:1993dq]. An advantage of DTW is that the minimization can be efficiently carried out by *dynamic programming* (DP) [@bellman52]. However, due to the Euclidean formulation, DTW may not be able to find a good alignment when the characteristics of the two sequences are substantially different (e.g., sequences have different amplitudes). Moreover, DTW cannot handle sequences with different dimensions (e.g., image to audio alignment), which limits the range of applications significantly.
To overcome the weaknesses of DTW, *canonical time warping* (CTW) was introduced [@NIPS2009_0760]. CTW performs sequence alignment in a common latent space found by canonical correlation analysis (CCA) [@hotelling36cca]. Thus, CTW can naturally handle sequences with different dimensions. However, CTW can only deal with linear projections, and it is difficult to optimize model parameters such as the initial alignment matrix, the regularization parameter used in CCA, and the dimensionality of the common latent space. To handle non-linearity, *dynamic manifold temporal warping* (DMTW) was recently proposed in [@DBLP:conf/iccv/GongM11]. DMTW first transforms original data onto a one-dimensional non-linear manifold and then finds an alignment on this manifold using DTW. Although DMTW is highly flexible by construction, its performance depends heavily on the choice of the non-linear transformation and, moreover, it implicitly assumes the smoothness of sequences. For this reason, DMTW has limited applicability. In this paper, we propose a novel information-theoretic temporal alignment method based on statistical dependence maximization. Our method, which we call *least-squares dynamic time warping* (LSDTW), employs a squared-loss variant of mutual information called *squared-loss mutual information* (SMI) as a dependency measure. SMI is estimated by the method of *least-squares mutual information* (LSMI) [@BMCBio:Suzuki+etal:2009a], which is consistent estimator achieving the optimal non-parametric convergence rate to the true SMI. An advantage of the proposed LSDTW over existing methods is that it can naturally deal with non-linearity and non-Gaussianity in data through SMI. Moreover, cross-validation (CV) with respect to the LSMI criterion is possible, which allows selection of model parameters such as the initial alignment matrix, the Gaussian kernel width, and the regularization parameter. Furthermore, the formulation of LSDTW is quite general and does not require strong assumptions on the topology of the latent manifold (e.g., smoothness). Thus, LSDTW is expected to perform well in a broader range of applications. Indeed, through experiments on synthetic and real-world *Kinect* action recognition tasks, LSDTW is shown to be a promising alternative to existing temporal alignment methods.
Dependence Maximizing Temporal Alignment via SMI
================================================
In this section, we first formulate the problem of *dependence maximizing temporal alignment* (DMTA) and then develop a DMTA method based on *squared-loss mutual information* (SMI) [@BMCBio:Suzuki+etal:2009a].
Formulation of Dependence Maximizing Temporal Alignment (DMTA) {#sec:prob_formulate}
--------------------------------------------------------------
\[sec:Problem\]
Given two sequences represented by a set of samples (ordered in time), $$\{\boldx_i\;|\; \boldx_i \in \mathbbR^{d_{\mathrm x}}\}_{i=1}^{n_{\mathrm x}}
\mbox{ and }
\{\boldy_j\;|\; \boldy_j \in \mathbbR^{d_{\mathrm y}}\}_{j=1}^{n_{\mathrm y}},$$ the goal of DMTA is to find a temporal alignment such that the statistical dependency between two sets of samples is maximized. Note that $n_{\mathrm x}$ and $d_{\mathrm x}$ can, in general, be different from $n_{\mathrm y}$ and $d_{\mathrm x}$.
Let $\boldpi^{\mathrm x}$ and $\boldpi^{\mathrm y}$ be alignment functions over $\{1,\ldots,n_{\mathrm x}\}$ and $\{1,\ldots,n_{\mathrm y}\}$, and let $\boldPi$ be the corresponding alignment matrix: $$\begin{aligned}
\boldPi &:= [\boldpi^{\mathrm x}~\boldpi^{\mathrm y}]^{\top} \in \mathbbR^{2 \times m},\\
\boldpi^{\mathrm x}&:= [\pix_1, \ldots, \pix_{m}]^\top \in \{1,\ldots,n_{\mathrm x}\}^{m \times 1}, \\
\boldpi^{\mathrm y}&:= [\piy_1, \ldots, \piy_{m}]^\top \in \{1,\ldots,n_{\mathrm y}\}^{m \times 1},\end{aligned}$$ where $m$ is the number of indexes needed to align the sequences and $^\top$ denotes the transpose. $\boldPi$ needs to satisfy the following three additional constraints:\
[**$\bullet$ Boundary condition:**]{} $[\pix_1~ \piy_1]^\top = [1~1]^\top$ and $[\pix_{m}~ \piy_{m}]^\top = [n_{\mathrm x}~n_{\mathrm y}]^\top$.\
[**$\bullet$ Continuity condition:**]{} $0 \leq \pix_t - \pix_{t-1} \leq 1$ and $0 \leq \piy_t - \piy_{t-1} \leq 1$.\
[**$\bullet$ Monotonicity condition:**]{} $t_1 \geq t_2 \rightarrow \pix_{t_1} \geq \pix_{t_2} , \piy_{t_1} \geq \piy_{t_2}$.
Let us denote the paired samples aligned by $\boldpi^{\mathrm x}$ and $\boldpi^{\mathrm y}$ as $$Z(\boldPi) := \{(\boldx_{\pix_i},\boldy_{\piy_i})\}_{i=1}^m.$$ Then, the optimal alignment, denoted by $\boldPi^\ast$, is defined as the maximum of a certain statistical dependence measure $D$ between the two sets $\{\boldx_i\}_{i=1}^{n_{\mathrm x}}$ and $\{\boldy_j\}_{j=1}^{n_{\mathrm y}}$: $$\begin{aligned}
\boldPi^* := \mathop{\text{argmax}}_{\boldPi} D(Z(\boldPi)).\end{aligned}$$
Least-Squares Dynamic Time Warping (LSDTW)
------------------------------------------
A popular measure of statistical dependence is *mutual information* (MI) [@Bell:Shannon:1948], and its estimation has been studied thoroughly [@IEEE-IT:Wang+etal:2005; @AISM:Sugiyama+etal:2008; @ISIT:Perez-Cruz:2008; @JSPI:Silva+Narayanan:2010; @IEEE-IT:Nguyen+etal:2010]. However, these MI approximations are computationally expensive, mainly due to the non-linearity introduced by the “log” function. In this paper, we propose to use a squared-loss variant of MI called *squared-loss MI* (SMI), which results in a simple and computationally efficient estimation algorithm called *least-squares dynamic time warping* (LSDTW).
### Overview
The optimization problem of LSDTW is defined as$$\begin{aligned}
\boldPi^* := \mathop{\text{argmax}}_{\boldPi} \textnormal{SMI}(Z(\boldPi)),
\label{eq:LSDTW-Optimization-Problem}\end{aligned}$$ where SMI is defined and expressed as $$\begin{aligned}
\textnormal{SMI}(Z) &= \frac{1}{2}\iint \left(\frac{\pxy(\boldx,\boldy)}{\px(\boldx)\py(\boldy)} - 1\right)^2 \px(\boldx) \py(\boldy) \textnormal{d}\boldx \textnormal{d}\boldy,
\nonumber\\
&= \frac{1}{2}\iint \left(\frac{\pxy(\boldx,\boldy)}{\px(\boldx)\py(\boldy)}\right)\pxy(\boldx,\boldy)\textnormal{d}\boldx \textnormal{d}\boldy - \frac{1}{2},
\label{eq:SMIdef}\end{aligned}$$ where $\pxy(\boldx,\boldy)$ is the joint density of $\boldx$ and $\boldy$, and $\px(\boldx)$ and $\py(\boldy)$ are the marginal densities of $\boldx$ and $\boldy$, respectively. SMI is the *Pearson divergence* [@PhMag:Pearson:1900] from $\pxy(\boldx,\boldy)$ to $\px(\boldx)\py(\boldy)$, while the ordinary MI is the *Kullback-Leibler divergence* [@Annals-Math-Stat:Kullback+Leibler:1951] from $\pxy(\boldx,\boldy)$ to $\px(\boldx)\py(\boldy)$. SMI is non-negative and is zero if and only if $\boldx$ and $\boldy$ are statistically independent, as the ordinary MI. Based on Eq., we develop the following iterative algorithm for estimating $\boldPi$:
(i) Initialization:
: Initialize the alignment matrix $\boldPi$.
(ii) Dependence estimation:
: For the current $\boldPi$, obtain an SMI estimator $\widehat{\textnormal{SMI}}(Z(\boldPi))$.
(iii) Dependence maximization:
: Given an SMI estimator $\widehat{\textnormal{SMI}}(Z(\boldPi))$, obtain the maximum alignment $\boldPi$.
(iv) Convergence check:
: The above (ii) and (iii) are repeated until $\boldPi$ fulfills a convergence criterion.
### Dependence Estimation {#subsec:dependence-estimation}
In the dependence estimation step, we utilize a non-parametric SMI estimator called *least-squares mutual information* (LSMI) [@BMCBio:Suzuki+etal:2009a], which was shown to possess a superior convergence property [@AISTATS:Suzuki+Sugiyama:2010]. Here, we briefly review LSMI.
[**Basic Idea:**]{} The key idea of LSMI is to directly estimate the *density ratio* in Eq. [@book:Sugiyama+etal:2012], $$r(\boldx, \boldy) := \frac{\pxy(\boldx, \boldy)}{\px(\boldx)\py(\boldy)},$$ from paired samples $Z(\boldPi) = \{(\boldx_{\pix_i},\boldy_{\piy_i})\}_{i=1}^m$ without going through density estimation of $\pxy(\boldx, \boldy)$, $\px(\boldx)$, and $\py(\boldy)$. Here, the density-ratio function $r(\boldx, \boldy)$ is directly modeled as $$\begin{aligned}
r_{\boldalpha}(\boldx, \boldy)= \sum_{\ell = 1}^m \alpha_{\ell} K(\boldx,\boldx_{\pix_\ell})L(\boldy,\boldy_{\piy_{\ell}}),
\label{ratio-model}\end{aligned}$$ where $K(\boldx,\boldx')$ and $L(\boldy,\boldy')$ are kernel functions (e.g., Gaussian kernels) for $\boldx$ and $\boldy$, respectively.
Then, the parameter $\boldalpha=(\alpha_1,\ldots,\alpha_m)^\top$ is learned so that the squared error to the true density ratio is minimized: $$\begin{aligned}
J_0(\boldalpha):=\frac{1}{2}\iint (r_{\boldalpha}(\boldx,\boldy) - r(\boldx,\boldy))^2 \px(\boldx) \py(\boldy) \textnormal{d}\boldx \textnormal{d}\boldy.
$$ After a few lines of calculation, $J_0$ can be expressed as $$\begin{aligned}
J_0(\boldalpha) = J(\boldalpha)+\textnormal{SMI}(Z(\boldPi))+\frac{1}{2},\end{aligned}$$ where $$\begin{aligned}
J(\boldalpha) &:=
\frac{1}{2}\boldalpha^\top \boldH_{\boldPi} \boldalpha -\boldh_{\boldPi}^\top\boldalpha, \\
H_{\boldPi,\ell,\ell'} &:= \iint K(\boldx,\boldx_{\pix_\ell})L(\boldy,\boldy_{\piy_{\ell}}) K(\boldx,\boldx_{\pix_{\ell'}})L(\boldy,\boldy_{\piy_{\ell'}}) \px(\boldx) \py(\boldy) \textnormal{d}\boldx \textnormal{d}\boldy, \\
h_{\boldPi,\ell} &:= \iint K(\boldx,\boldx_{\pix_\ell})L(\boldy,\boldy_{\piy_{\ell}}) \pxy(\boldx,\boldy)\textnormal{d}\boldx \textnormal{d}\boldy.
$$ Since $\textnormal{SMI}(Z(\boldPi))$ is constant with respect to $\boldalpha$, minimizing $J_0$ is equivalent to minimizing $J$.
[**Computing the Solution:**]{} Approximating the expectations in $\boldH_\boldPi$ and $\boldh_\boldPi$ included in $J$ by empirical averages, we have the following optimization problem: $$\begin{aligned}
\min_{\boldalpha} \left[
\frac{1}{2}\boldalpha^\top \widehat{\boldH}_\boldPi\boldalpha
-\widehat{\boldh}_\boldPi^\top\boldalpha
+ \frac{\lambda}{2} \boldalpha^\top\boldalpha \right],
\label{eq:SMI_cost_func}\end{aligned}$$ where $\lambda\boldalpha^\top \boldalpha/2$ is the regularization term to avoid overfitting, $\lambda$ ($\ge0$) is the regularization parameter, and $$\begin{aligned}
\widehat{H}_{\boldPi,\ell,\ell'} &:= \frac{1}{m^2}
\sum_{i,j = 1}^mK(\boldx_{\pix_i},\boldx_{\pix_\ell})L(\boldy_{\piy_j},\boldy_{\piy_{\ell}}) K(\boldx_{\pix_i},\boldx_{\pix_{\ell'}})L(\boldy_{\piy_j},\boldy_{\piy(\ell')}),\\
\widehat{h}_{\boldPi,\ell} &:= \frac{1}{m}\sum_{i = 1}^m
K(\boldx_{\pix_i},\boldx_{\pix_\ell})L(\boldy_{\piy_i},\boldy_{\piy_{\ell}}).\end{aligned}$$ Differentiating Eq. with respect to $\boldalpha$ and equating it to zero, we can obtain the optimal solution $\widehat{\boldalpha}_\boldPi$ analytically as $$\begin{aligned}
\label{eq:LSMI_Solution}
\widehat{\boldalpha}_\boldPi= (\widehat{\boldH}_\boldPi + \lambda \boldI)^{-1}\widehat{\boldh}_\boldPi,\end{aligned}$$ where $\boldI$ is the $m \times m$ identity matrix. Note that, LSMI has time complexity $O(m^3)$ due to the matrix inversion. However, when the number of training data is large, we can reduce the number of kernels in Eq. to $l (<m)$ by sub-sampling. With this approximation, the inverse matrix in Eq. can be computed with time complexity $O(l^3)$.
Finally, the following SMI estimator can be obtained by taking the empirical average of Eq. as $$\begin{aligned}
\widehat{\mathrm{SMI}}(Z(\boldPi))&= \frac{1}{2m} \sum_{i = 1}^m r_{\widehat{\boldalpha}_\boldPi}(\boldx_{\pix_i}, \boldy_{\piy_i}) - \frac{1}{2}.\label{SMIhat}\end{aligned}$$
[**Model Selection:**]{} \[subsubsec:LSMI-CV\] Hyper-parameters included in the kernel functions and the regularization parameter can be optimized by cross-validation with respect to $J$ [@BMCBio:Suzuki+etal:2009a], which is described below.
First, samples $\calZ=\{(\boldx_i, \boldy_i)\}_{i=1}^n$ are divided into $K$ disjoint subsets $\{\calZ_k\}_{k=1}^K$ of (approximately) the same size. Then, an estimator $\widehat{\boldalpha}_{\calZ_k}$ is obtained using $\calZ\backslash\calZ_k$ (i.e., all samples without $\calZ_k$), and the approximation error for hold-out samples $\calZ_k$ is computed as $$J_{\calZ_k}^{(K \text{-CV})} := \frac{1}{2}\widehat{\boldalpha}_{\calZ_k}^\top \widehat{\boldH}_{\calZ_k}\widehat{\boldalpha}_{\calZ_k} - \widehat{\boldh}_{\calZ_k}^\top\widehat{\boldalpha}_{\calZ_k},$$ where, for $|\calZ_k|$ being the number of samples in subset $\calZ_k$, $$\begin{aligned}
[\widehat{H}_{\calZ_k}]_{\ell,\ell'} &:= \frac{1}{|\calZ_k|^2}
\sum_{\boldx\in\calZ_k}
\sum_{\boldy\in\calZ_k}
K(\boldx,\boldx_\ell)L(\boldy,\boldy_\ell) K(\boldx,\boldx_{\ell'})L(\boldy,\boldy_{\ell'}),\\
[\widehat{h}_{\calZ_k}]_\ell &:= \frac{1}{|\calZ_k|}\sum_{(\boldx,\boldy)\in\calZ_k}
K(\boldx,\boldx_\ell)L(\boldy,\boldy_\ell).\end{aligned}$$
This procedure is repeated for $k = 1, \ldots, K$, and its average $J^{(K\text{-CV})}$ is taken as $$\begin{aligned}
J^{(K\text{-CV})} := \frac{1}{K}\sum_{k = 1}^K J_{\calZ_k}^{(K\text{-CV})}.
$$ Finally, we compute $J^{(K\text{-CV})}$ for all model candidates, and choose the one with minimum $J^{(K\text{-CV})}$.
Dependence Maximization
-----------------------
Based on the empirical estimate of SMI, the dependence maximization problem is given as $$\begin{aligned}
\max_{\boldPi}~~ & \widehat{\mathrm{SMI}}(Z(\boldPi)).
$$ We here provide a computationally efficient approximation algorithm based on *dynamic programming* (DP) [@bellman52]. Let us rewrite the empirical SMI, Eq., as $$\begin{aligned}
\widehat{\mathrm{SMI}}(Z(\boldPi))&= \frac{1}{2m}\sum_{i = 1}^{n_{\mathrm x}}\sum_{j = 1}^{n_{\mathrm y}} \delta(\pi_i^{\mathrm x}, \pi_j^{\mathrm y}) r_{\widehat{\boldalpha}_{\boldPi}}(\boldx_i, \boldy_j) - \frac{1}{2},$$ where $$\begin{aligned}
\delta(\pi_i^{\mathrm x}, \pi_j^{\mathrm y}) = \left\{ \begin{array}{ll}
1 & \textnormal{if}~\boldx_{\pi_i^{\mathrm x}}~\textnormal{and}~\boldy_{\pi_j^{\mathrm y}}~\textnormal{are paired}, \\
0 & \textnormal{otherwise}. \\
\end{array} \right.\end{aligned}$$
Then, the solution is updated with the current $\boldPi^{\textnormal{old}}$ as $$\begin{aligned}
\label{eq:DP-LSDTW}
\boldPi^{\textnormal{new}} = \argmax_{\boldPi} \sum_{i = 1}^{n_{\mathrm x}}\sum_{j = 1}^{n_{\mathrm y}} \delta(\pi_i^{\mathrm x}, \pi_j^{\mathrm y}) r_{\widehat{\boldalpha}_{\boldPi^{\textnormal{old}}}}(\boldx_i, \boldy_j).\end{aligned}$$ This problem can be efficiently solved by DP with time complexity $O(n_{\mathrm x}n_{\mathrm y})$ (see Appendix). Note, however, that the solution to Eq. does not always increase the empirical SMI, Eq.; we update the alignment matrix $\boldPi$ only if the SMI score increases after the update.
Related Methods {#sec:related}
===============
In this section, we review existing temporal alignment methods which are based on pairwise distance minimization (not dependence maximization) and point out their potential weaknesses.
Dynamic Time Warping (DTW)
--------------------------
The goal of *dynamic time warping* (DTW) is, given two sequences of the *same* dimensionality and the *different* number of samples, $\{\boldx_i\;|\; \boldx_i \in \mathbbR^{d}\}_{i=1}^{n_{\mathrm x}}$ and $\{\boldy_j\;|\; \boldy_j \in \mathbbR^{d}\}_{j=1}^{n_{\mathrm y}}$, to find an alignment such that the sum of pairwise distances between two sets is minimized [@saoke78; @Rabiner:1993dq]: $$\begin{aligned}
\min_{\boldpi^{\mathrm x}, \boldpi^{\mathrm y}} & \sum_{i = 1}^m \| \boldx_{\pi^{\mathrm x}_i} - \boldy_{\pi^{\mathrm y}_i} \|^2,\end{aligned}$$ where $m$ is the number of indices needed to align the sequences. $\boldpi^{\mathrm x}$ and $\boldpi^{\mathrm y}$ need to satisfy the boundary, continuity, and monotonicity conditions (see Section \[sec:Problem\]). The above DTW optimization problem can be efficiently solved by DP with time complexity $O(n_{\mathrm x}n_{\mathrm y})$.
A potential weakness of DTW is that it cannot handle sequences with different dimensions such as image to audio alignment. Moreover, even when the dimensionality of sequences is the same, DTW may not be able to find a good alignment of sequences with different characteristics such as sequences with different amplitudes. These drawbacks highly limit the applicability of DTW.
Canonical Time Warping (CTW)
----------------------------
*Canonical time warping* (CTW) can align sequences with different dimensions in a common latent space [@NIPS2009_0760; @CVPR:Feng:2012a]. The CTW optimization problem is given as $$\begin{aligned}
\label{eq:CTW}
\min_{\boldW_{\mathrm x}, \boldW_{\mathrm y},\boldV_{\mathrm x}, \boldV_{y}} & \| \boldV_{\mathrm x}^\top \boldX \boldW_{\mathrm x}^\top - \boldV_{\mathrm y}^\top \boldY \boldW_{\mathrm y}^\top\|^2_{\mathrm{Frob}},\end{aligned}$$ where $\|\cdot\|_{\mathrm{Frob}}$ is the Frobenius norm, $\boldX = [\boldx_1, \ldots, \boldx_{n_{\mathrm x}}] \in \mathbbR^{d_{\mathrm x} \times n_{\mathrm x}}$, $\boldY = [\boldy_1, \ldots, \boldy_{n_{\mathrm y}}] \in \mathbbR^{d_{\mathrm y} \times n_{\mathrm y}}$, $\boldW_{\mathrm x} \in \{0,1\}^{m \times n_{\mathrm x}}$ and $\boldW_{\mathrm y} \in \{0,1\}^{m \times n_{\mathrm y}}$ are binary selection matrices that need to be estimated to align $\boldX$ and $\boldY$, and $\boldV_{\mathrm x} \in \mathbbR^{d_{\mathrm x} \times b}$ and $\boldV_{\mathrm y} \in \mathbbR^{d_{\mathrm y} \times b}$ ($b \leq \min(d_{\mathrm x}, d_{\mathrm y}))$ are linear projection matrices of $\boldx$ and $\boldy$ onto a common latent space, respectively. The above optimization problem can be efficiently solved by alternately solving CCA and DTW, where the alignment matrix obtained using DTW is usually used as initialization (initial alignment matrix).
However, since CTW finds a common latent space using CCA, it can only deal with linear and Gaussian temporal alignment problems. Thus, CTW cannot properly deal with multi-modal and non-Gaussian data. Another limitation of CTW is that comparing the alignment quality over different model parameters is not straightforward. This is because, for different model parameters, a common latent space found by CCA is generally different and thus the metric of the pairwise distance Eq. is also different. For this reason, a systematic model selection method for the regularization parameter, dimensionality of the common latent space, and the initial alignment matrix has not been developed so far, to the best of our knowledge.
Dynamic Manifold Temporal Warping (DMTW)
----------------------------------------
*Dynamic manifold temporal warping* (DMTW) is a non-linear extension of CTW [@DBLP:conf/iccv/GongM11]. The DMTW optimization problem is defined as $$\begin{aligned}
\min_{\boldW_{\mathrm x}, \boldW_{\mathrm y},\calF_{\mathrm x}, \calF_{\mathrm y}} & \| \calF_{\mathrm x}(\boldX) \boldW_{\mathrm x}^\top - \calF_{\mathrm y}(\boldY) \boldW_{\mathrm y}^\top\|^2_{\mathrm{Frob}},\end{aligned}$$ where $\calF_{\mathrm x}: \mathbbR^{d_{\mathrm x} \times n_{\mathrm x}} \rightarrow \mathbbR^{b \times n_{\mathrm x}}$ and $\calF_{\mathrm y}: \mathbbR^{d_{\mathrm y} \times n_{\mathrm y}} \rightarrow \mathbbR^{b \times n_{\mathrm y}}$ are non-linear mapping functions that map $\boldx$ and $\boldy$ to a common latent subspace. DMTW first maps $\boldX$ and $\boldY$ to a one-dimensional smooth manifold (i.e., $b = 1$) by the *tensor voting* method [@DBLP:journals/jmlr/MordohaiM10] and then align sequences on the manifold.
DMTW highly depends on a specific non-linear transformations and requires the smooth manifold assumption. Thus, the usage of DMTW is limited to specific applications. On the other hand, CTW and LSDTW do not require the latter strong assumption and thus can be useful for a broader range of applications. This assertion will be experimentally validated in the next section.
Experiments {#sec:experiments}
===========
In this section, we experimentally evaluate our proposed LSDTW method on synthetic and real-world *Kinect* action recognition tasks.
Setup
-----
In LSDTW, we use the Gaussian kernels: $$\begin{aligned}
K(\boldx, \boldx') = \exp \left(-\frac{\|\boldx - \boldx'\|^2}{2\sigma_{\mathrm{x}}^2} \right),~
L(\boldy, \boldy') = \exp \left(-\frac{\|\boldy - \boldy'\|^2}{2\sigma_{\mathrm{y}}^2} \right),\end{aligned}$$ where $\sigma_\mathrm{x}$, $\sigma_\mathrm{y}$, and $\lambda$ are chosen by 3-fold CV from $$(\sigma_\mathrm{x}, \sigma_\mathrm{y}) = c\times(m_\mathrm{x}, m_\mathrm{y}),~c = 2^{-1/2}, 1.8^{-1/2}, \ldots, 0.2^{-1/2},
~~\lambda =10^{-1}, 10^{-2},$$ and $$m_\mathrm{x} = 2^{-1/2}\textnormal{median}(\{\|\boldx_i - \boldx_j\|\}_{i,j=1}^{n_{\mathrm x}}),~m_\mathrm{y} = 2^{-1/2}\textnormal{median}(\{\|\boldy_i - \boldy_j\|\}_{i,j=1}^{n_{\mathrm y}}).$$
Due to non-convex nature of the objective, setting a good initial alignment is an important issue for LSDTW. Here, from the alignment obtained using CTW and the simple uniform initialization, $$\begin{aligned}
\boldpi^{\mathrm x} &= [1, \lfloor 1 + n_{\mathrm x}/m \rfloor, \lfloor 1+2n_{\mathrm x}/m \rfloor, \ldots, n_{\mathrm x}]^\top \in \mathbbR^{m \times 1}, \\
\boldpi^{\mathrm y} &= [1, \lfloor 1 + n_{\mathrm y}/m \rfloor, \lfloor 1+2n_{\mathrm y}/m \rfloor, \ldots, n_{\mathrm y}]^\top \in \mathbbR^{m \times 1}, \end{aligned}$$ where $m = \min(n_{\mathrm x}, n_{\mathrm y})$, we choose the one with the largest SMI score as the initial alignment for LSDTW.
We compare the performance of LSDTW with DTW and CTW. For CTW, we choose the dimensionality of CCA to preserve 90% of the total correlation, and we fix the regularization parameter at $0.01$. We use the alignment given by DTW as the initial alignment of CTW. To evaluate the alignment results, we use the following standard alignment error [@CVPR:Feng:2012a]: $$\begin{aligned}
Error = \frac{\mathrm{dist}(\boldPi^\ast, \widehat{\boldPi})
+ \mathrm{dist}(\widehat{\boldPi},\boldPi^\ast)}{m^\ast + \widehat{m}},
~~~\mathrm{dist}(\boldPi_1, \boldPi_2) = \sum_{i = 1}^{m_1} \min(\{\|\boldpi_1^{(i)} - \boldpi_2^{(j)}\|\}_{j = 1}^{m_2}),\end{aligned}$$ where $\boldPi^\ast$ and $\widehat{\boldPi}$ are true and estimated alignment matrices and $\boldpi_1^{(i)}, \boldpi_2^{(j)} \in \mathbbR^{2\times1}$ are the $i$-th and $j$-th row of $\boldPi_1$ and $\boldPi_2$, respectively.
\
(a-1) Multi-modal data.
\
(a-2) Alignment path.
\
(a-3) SMI score.
\
\
(b-1) Non-Gaussian data.
\
(b-2) Alignment path.
\
(b-3) Mean alignment error.
Synthetic Dataset
-----------------
First, we illustrate the behavior of the proposed LSDTW method for non-linear and non-Gaussian data using synthetic datasets.
[**Non-linear (multi-modal) data[^1]:**]{} $$\begin{aligned}
x_i &= z_i + 0.4\sin(2\pi z_i),~i = 1,\ldots,1000, \\
y_j &= z_{(j -1)\times 2 + 1},~j=1,\ldots,500,\end{aligned}$$ where $z_i = i/1000$ (see Figure \[fig:result\_toy\]-(a-1)).
Figure \[fig:result\_toy\]-(a-2) shows the alignment path obtained by LSDTW, CTW, and DTW, respectively. In this experiment, we initialize LSDTW and CTW with the true alignment matrix and check whether those methods perform well. As can be observed, LSDTW can find a better alignment in the middle region (i.e., a multi-modal region) than DTW and CTW. This shows that LSDTW objective is much better than alternatives when it comes to multi-modal data. Figure \[fig:result\_toy\]-(a-3) depicts the SMI score with respect to the number of iterations in LSDTW, showing that LSDTW converges in 10 iterations.
[**Non-Gaussian data:**]{} $$\begin{aligned}
\boldX = \boldU_{\mathrm x}^\top \boldZ \boldM_{\mathrm x}^\top + \eta \boldE_{\mathrm x},
~~
\boldY = \boldU_{\mathrm y}^\top \boldZ \boldM_{\mathrm y}^\top + \eta \boldE_{\mathrm y},\end{aligned}$$ where $\boldU_{\mathrm x}, \boldU_{\mathrm y} \in \mathbbR^{2 \times 2}$ are randomly generated affine transformation matrices, $\boldZ \in \mathbbR^{2 \times m}$ is a trajectory in two dimensions, $\boldM_{\mathrm x} \in \mathbbR^{n_{\mathrm x} \times m}$ and $\boldM_{\mathrm y} \in \mathbbR^{n_{\mathrm x} \times m}$ are randomly generated matrices for time warping, $\boldE_{\mathrm x} \in \mathbbR^{2 \times n_{\mathrm x}}$ and $\boldE_{\mathrm y} \in \mathbbR^{2 \times n_{\mathrm y}}$ are randomly generated additive exponential noise with rate parameter 1 (and its mean is adjusted to be zero), and $\eta = \{0, 0.3, 0.6,\ldots, 3.0\}$ is the noise level. Note that larger noise level $\eta$ means stronger non-Gaussianity in the data.
Figures \[fig:result\_toy\]-(b-1) and (b-2) show an example of synthetic data with additive exponential noise ($\eta = 2.4$) and corresponding alignment paths obtained by LSDTW, CTW, and DTW. As can be seen, only the proposed method can find a good alignment. Figure \[fig:result\_toy\]-(b-3) shows the mean alignment error over 100 runs, from which we can confirm that the proposed method tends to outperform the existing methods for larger noise levels.
Real-world Kinect Action Recognition Data
-----------------------------------------
Next, we evaluate the proposed LSDTW method on the publicly available *Kinect* action recognition dataset[^2] [@DBLP:conf/iccvw/MasoodNKZT11]. This dataset consists of the human skeleton data (15 joints) obtained using a *Kinect* sensor, and there are 16 subjects and 16 actions with 5 runs. Instead of using the raw skeleton data, we here use the 105-dimensional feature vector, where each element of the feature vector is the Euclidean distance between joint pairs. In evaluation, we carry out *unsupervised* action recognition experiments and evaluate the performance of alignment methods by classification accuracy. More specifically, we first divide the action recognition dataset into two disjoint subsets: 8 subjects (\#1-\#8) with all actions for testing (in total 640 sequences) and the remaining subjects (\#9-\#16) with all actions for “training" database (in total 640 sequences). Then, we retrieve $N=10$ similar actions for each test action from the database by DTW, CTW, and LSDTW. Here, we use the pairwise Euclidean distance based on an estimated alignment to measure the similarity between sequences. Finally, if there is at least one correct action in the retrieved sequences, we regard the action to be correctly retrieved.
Figure \[fig:result\_kinect\] shows the mean classification accuracy as functions of the number of retrieved actions, $N$, where three different database sizes are tested. The graphs clearly show that LSDTW compares favorably with existing methods in terms of classification accuracy.
\
(a)
\
(b)
\
(c)
Conclusions
===========
In this paper, we proposed a novel temporal alignment framework called the *dependence maximization temporal alignment* (DMTA) and developed a DMTA method called the *least-squares dynamic time warping* (LSDTW). LSDTW adopts *squared-loss mutual information* as a dependence measure, which is efficiently estimated by the method of *least-squares mutual information*. Notable advantages of LSDTW are that it can naturally deal with non-linear and non-Gaussian sequences and it can optimize model parameters such as the Gaussian kernel width and the regularization parameter by cross-validation. We applied the proposed method on the *Kinect* action recognition task, and experimentally showed that LSDTW is the promising alternative to the compared methods.
Acknowledgments {#acknowledgments .unnumbered}
===============
We thank Prof. Fernando De la Torre, Mr. Feng Zhou, and Dr. Akisato Kimura for their valuable comments. MY acknowledges the JST PRESTO and PLIP programs, and MS acknowledges the JST PRESTO program and AOARD for financial support.
Appendix {#appendix .unnumbered}
========
Given the empirical estimate of SMI computed at the dependence estimation step (Sect. 2.2.2), the depence maximization problem is given as $$\begin{aligned}
\begin{split}
\max_{\boldPi}~~ \widehat{\mathrm{SMI}}(Z(\boldPi)) &~~=~ \max_{\boldPi}~~ \frac{1}{2m} \sum_{i = 1}^{m} r_{\widehat{\boldalpha}_{\boldPi_{\mathrm{old}}}}(\boldx_{\pix_i}, \boldy_{\piy_i}) - \frac{1}{2} \\
&\overset{\text{Eq. (3)}}= \max_{\boldPi}~~ \frac{1}{2 m} \sum_{i = 1}^{m} \sum_{\ell = 1}^{m_{\mathrm{old}}} {\widehat{\alpha}_{\ell}} K(\boldx_{\pix_i},\boldx_{{\pix_\ell}_{\mathrm{old}}} ) L(\boldy_{\piy_i},\boldy_{{\piy_\ell}_{\mathrm{old}}} ) - \frac{1}{2}\\
& ~~\approx~ \frac{1}{2( n_{\mathrm x} + n_{\mathrm y})} \max_{\boldPi}~ \sum_{i = 1}^{m} \sum_{\ell = 1}^{m_{\mathrm{old}}} {\widehat{\alpha}_{\ell}} K(\boldx_{\pix_i},\boldx_{{\pix_\ell}_{\mathrm{old}}} ) L(\boldy_{\piy_i},\boldy_{{\piy_\ell}_{\mathrm{old}}} ) - \frac{1}{2}.
\end{split}\end{aligned}$$ Based on the constraints on the alignment functions $\boldPi$ described in Sect. 2.1, this optimal alignement can be computed by dynamic programming (DP) [@bellman52]. In order to verify this, we define the prefix sequences $\boldX_n := \{\boldx_i\;|\; \boldx_i \in \mathbbR^{d_{\mathrm x}}\}_{i=1}^{n} \text{ and }
\boldY_{n'} := \{\boldy_j\;|\; \boldy_j \in \mathbbR^{d_{\mathrm y}}\}_{j=1}^{n'} $, with $n \leq n_{\mathrm x} \text{ and } n' \leq n_{\mathrm y}$, and set $A(n,n') := \widehat{\mathrm{SMI}}(\boldX_{n}, \boldY_{n'})$ denoting the optimal SMI for the aligned prefix sequences $\boldX_n$ and $\boldY_{n'}$.
Following the boundary conditions of the alignement functions, we have: $$A(1,1) = r_{\widehat{\boldalpha}_{\boldPi_{\mathrm{old}}}}(\boldx_{1}, \boldy_{1}).$$ Based on the continuity and monotonicity conditions, the DP-equation is given as $$A(n,n') = \max \{ A(n-1, n'-1), A(n-1, n'), A(n,n'-1) \}+ r_{\widehat{\boldalpha}_{\boldPi_{\mathrm{old}}}}(\boldx_{n}, \boldy_{n'}),$$ for $1 < n \leq n_{\mathrm x}$ and $1< n' \leq n_{\mathrm y}$. Therefore, the optimal $\widehat{\mathrm{SMI}}(Z(\boldPi)) = \frac{1}{2( n_{\mathrm x} + n_{\mathrm y})} A(n_{\mathrm x} ,n_{\mathrm y}) - \frac{1}{2}$ can be computed in $ O(n_{\mathrm x} n_{\mathrm y})$. Given the accumulated cost matrix $A$, we can compute the optimal alignment $\boldPi$ using backtracking.
[^1]: A distribution of a multi-modal data has two or more modes.
[^2]: [www.cs.ucf.edu/\~smasood/datasets/UCFKinect.zip](www.cs.ucf.edu/~smasood/datasets/UCFKinect.zip)
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- Francesco Campaioli
title: ' ****'
---
[ ]{}
Appendix
========
[245]{}ifxundefined \[1\][ ifx[\#1]{} ]{}ifnum \[1\][ \#1firstoftwo secondoftwo ]{}ifx \[1\][ \#1firstoftwo secondoftwo ]{}““\#1””@noop \[0\][secondoftwo]{}sanitize@url \[0\][‘\
12‘\$12 ‘&12‘\#12‘12‘\_12‘%12]{}@startlink\[1\]@endlink\[0\]@bib@innerbibempty [****, ()](\doibase 10.1103/PhysRevLett.120.060409) [****, ()](\doibase 10.22331/q-2019-08-05-168) [****, ()](\doibase
10.1103/PhysRevA.100.062328) [****, ()](\doibase
10.1103/PhysRevLett.118.150601) “,” in [**](\doibase 10.1007/978-3-319-99046-0_8), (, ) pp. [****, ()](\doibase 10.1007/BF01397280) [****, ()](\doibase 10.1103/PhysRevLett.60.385) [****, ()](\doibase https://doi.org/10.1016/j.physleta.2003.07.025) [****, ()](\doibase 10.1103/PhysRev.34.163) [****, ()](\doibase https://doi.org/10.1016/j.physleta.2003.12.001) [****, ()](http://dl.acm.org/citation.cfm?id=2011593.2011606) [****, ()](\doibase 10.1088/1464-4266/7/12/033) [****, ()](\doibase 10.1063/1.4871444) [****, ()](\doibase https://doi.org/10.1016/S0034-4877(74)80004-2) [****, ()](\doibase 10.1007/BF02813438) [****, ()](\doibase 10.1119/1.1430697) [****, ()](\doibase 10.1103/PhysRevA.56.3425) [ ()](http://arxiv.org/abs/1810.12869) [****, ()](\doibase https://doi.org/10.1016/0375-9601(87)90363-X) [****, ()](\doibase 10.1088/1751-8113/41/25/255301) [****, ()](\doibase
10.1103/PhysRevA.93.062127) [****, ()](\doibase 10.1088/1751-8121/aad50f) [****, ()](\doibase 10.1103/PhysRevLett.113.260401) [****, ()](\doibase
10.1103/PhysRevLett.122.100401) in [**](\doibase 10.1007/978-3-642-74626-0_8) (, , ) pp. [****, ()](\doibase 10.1088/1751-8121/aa86c6) [****, ()](\doibase 10.1103/PhysRev.122.1649) [****, ()](\doibase 10.1103/PhysRevD.7.359) [****, ()](\doibase 10.1016/0003-4916(78)90223-3) [****, ()](\doibase 10.1007/BF00739034) [****, ()](\doibase 10.1103/PhysRevA.31.2078) [****, ()](\doibase 10.1103/PhysRevLett.65.1697) [****, ()](\doibase 10.1119/1.16940) [****, ()](\doibase 10.1016/S0167-2789(98)00054-2) [****, ()](\doibase 10.1103/PhysRevA.67.052109) in [**](\doibase 10.1117/12.507486), Vol. () [****, ()](\doibase 10.1103/PhysRevLett.120.070402) [****, ()](\doibase 10.1007/BF02819419) [****, ()](\doibase 10.1119/1.17368) [****, ()](\doibase 10.1016/0375-9601(92)90555-Z) [****, ()](\doibase 10.1088/1751-8113/46/33/335302) [****, ()](\doibase 10.1038/srep04890) [****, ()](\doibase 10.1016/J.PHYSLETA.2015.12.015) [****, ()](\doibase 10.1103/PhysRevLett.110.050403) [****, ()](\doibase 10.1103/PhysRevLett.110.050402) [****, ()](\doibase 10.1088/1464-4266/6/8/028) [****, ()](\doibase 10.1088/1751-8113/40/11/020) [****, ()](\doibase
10.1209/0295-5075/81/30007) [****, ()](\doibase 10.1103/PhysRevA.72.032337) [****, ()](\doibase 10.1103/PhysRevA.73.049904) [****, ()](\doibase 10.1103/PhysRevLett.46.623) [****, ()](https://doi.org/10.1103/PhysRevA.82.022318) [****, ()](\doibase 10.1038/35023282) [****, ()](\doibase 10.1103/PhysRevLett.112.120405) [****, ()](\doibase 10.1038/nphoton.2011.35) [****, ()](\doibase 10.1103/PhysRevLett.109.233601) [****, ()](\doibase
10.1038/ncomms2067) [****, ()](\doibase
10.1103/PhysRevLett.118.140403) [****, ()](\doibase 10.1088/0953-4075/44/15/154011) [****, ()](\doibase 10.1063/1.3691827) [****, ()](\doibase
10.1103/PhysRevLett.103.240501) [****, ()](\doibase 10.1103/PhysRevLett.109.115703) [****, ()](\doibase 10.1103/PhysRevLett.111.260501) [****, ()](\doibase
10.1038/ncomms12999) [****, ()](\doibase 10.1103/PhysRevLett.118.100601) [****, ()](\doibase
10.1103/PhysRevLett.118.100602) [****, ()](\doibase 10.1088/1367-2630/aa96f8) [****, ()](\doibase 10.1103/PhysRevLett.105.170402) [ ()](https://arxiv.org/abs/1910.06811) [****, ()](\doibase 10.1088/1367-2630/17/7/075015) **, [Ph.D. thesis](https://ora.ox.ac.uk/objects/uuid:279871ea-3b2e-4baf-975c-1bd42b4961c3), () [****, ()](\doibase
10.1103/PhysRevB.99.205437) [****, ()](\doibase 10.1103/PhysRevA.78.042305) [****, ()](\doibase 10.1088/1751-8113/45/41/415304) [****, ()](\doibase 10.1038/srep15775) [****, ()](\doibase 10.1103/PhysRevA.93.012311) [****, ()](\doibase
10.1088/1751-8113/49/14/143001) [****, ()](\doibase 10.1209/0295-5075/115/40003) [****, ()](\doibase 10.1016/J.PHYSLETA.2016.02.018) [****, ()](\doibase 10.1103/PhysRevA.94.052125) [****, ()](\doibase 10.1103/PhysRevX.6.021031) [****, ()](\doibase 10.1103/PhysRevA.93.052331) [****, ()](\doibase 10.1103/PhysRevE.93.042135) [****, ()](\doibase 10.1103/PhysRevA.95.042314) [****, ()](\doibase 10.1103/PhysRevA.95.022115) [****, ()](\doibase 10.1142/S0129054117500204) [****, ()](\doibase 10.1088/1367-2630/ab099e) [****, ()](\doibase 10.1016/J.PHYSE.2017.08.020) [****, ()](\doibase 10.1088/1751-8121/aa96f1) [****, ()](\doibase
10.1364/OL.43.003136) [****, ()](\doibase 10.22331/q-2018-10-01-96) [****, ()](\doibase 10.1103/PhysRevLett.120.070401) [**](http://chaos.if.uj.edu.pl/~karol/geometry.htm) (, ) [****, ()](\doibase 10.1103/PhysRevLett.70.3365) @noop [****, ()]{} @noop [****, ()]{} [**](\doibase
DOI:10.1093/acprof:oso/9780199213900.001.0001) (, ) p. [****, ()](https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.103.160502) [**](https://books.google.com.au/books/about/Quantum{_}Computation{_}and{_}Quantum{_}Informat.html?id=65FqEKQOfP8C{&}redir{_}esc=y) (, ) [****, ()](\doibase https://doi.org/10.1016/0034-4877(76)90060-4) [****, ()](\doibase 10.1103/PhysRevD.23.357) in [**](\doibase 10.1007/978-94-011-2801-8_23), (, , ) pp. [****, ()](\doibase 10.1088/1367-2630/aa83dc) [****, ()](\doibase https://doi.org/10.1016/S0375-9601(03)00941-1) [****, ()](\doibase 10.1103/PhysRevA.68.062322) [ ()](http://arxiv.org/abs/1608.01443) [****, ()](\doibase 10.1137/090757058) [****, ()](\doibase 10.1017/S0308210511001648) [****, ()](\doibase
10.1103/PhysRevLett.88.217901) [****, ()](\doibase 10.1103/PhysRevA.64.052311) [****, ()](\doibase https://doi.org/10.1016/j.cpc.2011.08.002) [****, ()](\doibase 10.1142/S1230161217400169) [****, ()](\doibase 10.1103/PhysRevE.90.022103) [****, ()](\doibase 10.1103/PhysRevA.55.2290) [****, ()](\doibase 10.1103/PhysRevA.81.023816) [****, ()](\doibase 10.1021/jz201154t) [****, ()](\doibase 10.1103/PhysRevB.77.115403) [****, ()](\doibase 10.1103/PhysRevA.65.010101) [****, ()](\doibase
10.1103/PhysRevLett.122.140401) [****, ()](\doibase 10.1103/PhysRevLett.120.040405) [****, ()](\doibase 10.1103/PhysRevA.41.2295) [****, ()](\doibase 10.1103/PhysRevA.50.4582) [****, ()](\doibase
10.1103/PhysRevLett.101.180402) [****, ()](\doibase 10.1103/PhysRevLett.100.090503) @noop [**]{}, , Vol. (, ) [****, ()](\doibase 10.1103/PhysRevLett.111.010402) [****, ()](\doibase
10.1038/srep08444) [****, ()](\doibase 10.1142/s0129054114400073) [****, ()](\doibase 10.1103/PhysRevLett.72.3439) [****, ()](\doibase 10.1063/1.2218675) [****, ()](\doibase 10.1103/PhysRevA.86.034101) [****, ()](\doibase 10.1103/PhysRevLett.91.097906) [****, ()](\doibase
10.1103/PhysRevLett.106.050403) [****, ()](\doibase 10.22331/q-2018-07-11-76) [****, ()](\doibase 10.1103/PhysRevA.69.032106) [****, ()](\doibase
10.1016/J.PHYSLETA.2010.10.005) [**** ()](https://arxiv.org/abs/0805.2037) [****, ()](http://www.jmlr.org/papers/v10/abernethy09a.html) [****, ()](\doibase
10.1103/PhysRevLett.114.170501) [****, ()](\doibase
10.1103/PhysRevLett.117.170501) [ ()](https://arxiv.org/abs/1906.04766) [****, ()](\doibase 10.1088/1367-2630/16/6/065023) [****, ()](\doibase
10.1088/1367-2630/aac6f3) [****, ()](\doibase 10.1088/1367-2630/aa8242) [****, ()](\doibase 10.1080/00029890.1995.11990579) [****, ()](\doibase 10.1103/PhysRevLett.98.040403) [****, ()](\doibase 10.1088/1751-8113/41/24/244002) [**** ()](https://doi.org/10.1103/PhysRevLett.99.130502) [****, ()](\doibase 10.1103/PhysRevLett.107.177204) [****, ()](\doibase 10.1088/1751-8113/48/11/115303) [****, ()](\doibase 10.1088/1367-2630/17/3/033048) [****, ()](\doibase 10.1088/0953-4075/40/18/R01) [****, ()](\doibase 10.1103/PhysRevA.90.033628) [****, ()](\doibase 10.1103/PhysRevLett.112.080801) [****, ()](\doibase 10.1103/PhysRevA.90.012318) [****, ()](\doibase
10.1103/PhysRevA.72.042331) [****, ()](\doibase 10.1103/RevModPhys.76.1037) [****, ()](\doibase 10.1038/srep34187) [****, ()](\doibase 10.1038/ncomms1374) [****, ()](\doibase 10.1103/PhysRevX.5.021026) [****, ()](\doibase
10.1038/nature13461) [****, ()](\doibase 10.1007/s11433-016-5789-5) [****, ()](\doibase
10.1103/PhysRevLett.108.153601) [****, ()](\doibase 10.1103/PhysRevLett.112.150802) [****, ()](\doibase 10.1126/science.1104149) [****, ()](\doibase 10.1103/PhysRevB.60.11404) [****, ()](\doibase
10.1103/PhysRevLett.103.080502) [****, ()](\doibase 10.1103/PhysRevLett.111.100502) [****, ()](\doibase
10.1103/PhysRevLett.96.060503) [****, ()](https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.103.160502) [****, ()](\doibase
10.1088/1751-8113/44/14/145302) [****, ()](\doibase 10.1103/PhysRevA.86.054302) [****, ()](\doibase 10.1088/1751-8113/46/4/045307) “,” in [**](\doibase 10.1007/978-1-4612-0349-0_3), (, , ) pp. [****, ()](\doibase
10.1103/PhysRevA.75.042308) [**** ()](http://iopscience.iop.org/article/10.1088/1751-8113/41/4/045303/meta) [**](\doibase https://doi.org/10.1002/zamm.19870670330) (, ) [**](\doibase 10.26180/5d7edea2b178e) (, ) [****, ()](\doibase 10.1103/PhysRevLett.79.325) [****, ()](\doibase 10.1038/s41567-018-0124-x) [****, ()](\doibase 10.1103/PhysRevA.86.062309) [****, ()](\doibase 10.1093/biomet/57.1.97) [****, ()](\doibase 10.1016/0021-9991(84)90095-0) [**](\doibase https://doi.org/10.1002/bbpc.198800128) (, ) [****, ()](\doibase 10.1103/PhysRevX.5.031044) [****, ()](\doibase 10.1088/1367-2630/13/11/113017) [****, ()](\doibase 10.1038/ncomms12479) [****, ()](\doibase 10.1088/1751-8113/42/36/365303) [****, ()](\doibase 10.1103/PhysRevLett.109.115703) [****, ()](\doibase 10.1103/PhysRevA.93.012311) [****, ()](\doibase 10.1103/PhysRevLett.118.100601) [ ()](https://arxiv.org/abs/1803.07119) [****, ()](http://stacks.iop.org/1751-8121/49/i=14/a=143001) [****, ()](\doibase 10.1088/1367-2630/18/1/011002) [****, ()](\doibase 10.1080/00107514.2016.1201896) [****, ()](\doibase 10.1103/PhysRevE.92.042126) [****, ()](\doibase 10.1088/0305-4470/33/24/302) [****, ()](\doibase 10.1126/science.1078955) [****, ()](\doibase
10.1103/PhysRevE.89.032115) [****, ()](\doibase 10.1103/PhysRevE.87.042123) [****, ()](\doibase 10.1103/PhysRevLett.111.240401) [****, ()](\doibase 10.1088/0953-4075/48/3/035501) [****, ()](\doibase 10.1103/PhysRevE.91.032118) [****, ()](\doibase 10.1103/PhysRevE.93.042135) [****, ()](\doibase 10.1088/1367-2630/17/6/065008) [****, ()](\doibase 10.1088/1367-2630/17/7/075015) [****, ()](\doibase
10.1103/PhysRevE.92.042147) **, [](https://ora.ox.ac.uk/objects/uuid:279871ea-3b2e-4baf-975c-1bd42b4961c3), () in [**](\doibase 10.1016/bs.aamop.2015.07.002), Vol. (, ) [****, ()](\doibase 10.1103/PhysRevLett.111.240401) [****, ()](\doibase 10.1103/PhysRevE.87.042123) [****, ()](\doibase
10.1103/PhysRevLett.122.047702) [****, ()](\doibase 10.1209/epl/i2004-10101-2) [****, ()](\doibase 10.1038/s41534-017-0012-8) [****, ()](\doibase 10.1007/BF01614224) [****, ()](\doibase 10.1007/BF01011769) [****, ()](\doibase 10.1103/PhysRevLett.110.157207) [****, ()](\doibase 10.1103/PhysRevLett.111.260501) [****, ()](\doibase 10.1103/PhysRevA.90.032110) [****, ()](\doibase 10.1103/PhysRevE.91.032119) [****, ()](\doibase 10.1088/1367-2630/18/7/073005) [****, ()](\doibase 10.1103/PhysRevA.85.052127) [****, ()](\doibase 10.1088/1751-8113/47/42/424006) [****, ()](\doibase 10.1103/PhysRevA.66.062311) [****, ()](\doibase 10.1103/PhysRevA.72.032322) [****, ()](\doibase 10.1103/PhysRevA.73.022109) [****, ()](\doibase
10.1103/PhysRevLett.106.170501) [****, ()](http://dx.doi.org/10.1038/nphys678) [****, ()](\doibase 10.1103/PhysRevLett.98.090401) [****, ()](\doibase
10.1038/nature09778) [****, ()](\doibase 10.1103/PhysRev.93.99) [****, ()](\doibase 10.1016/0029-5582(65)90862-X) [****, ()](\doibase 10.1103/PhysRevLett.82.1835) [****, ()](\doibase 10.1103/PhysRevX.1.021022) [****, ()](\doibase 10.1103/PhysRevX.9.011034) [ ()](https://arxiv.org/abs/1910.10567) [****, ()](\doibase
10.1103/PhysRevA.97.022106) [****, ()](\doibase
10.1103/PhysRevLett.120.117702) [ ()](http://arxiv.org/abs/1812.10139) [****, ()](\doibase
10.1103/PhysRevB.98.205423) [****, ()](\doibase
10.1103/PhysRevE.99.052106) [****, ()](\doibase 10.22331/q-2018-04-23-61) [****, ()](\doibase 10.1103/PhysRevLett.122.210601) [ ()](https://arxiv.org/abs/1909.03558) [****, ()](\doibase 10.1103/PhysRevResearch.2.013095)
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Recent neuroimaging studies have shown that functional connectomes are unique to individuals, i.e., two distinct fMRIs taken over different sessions of the same subject are more similar in terms of their connectomes than those from two different subjects. In this study, we present significant new results that identify, for the first time, specific parts of resting state and task-specific connectomes that code the unique signatures. We show that a very small part of the connectome codes the signatures. A network of these features is shown to achieve excellent training and test accuracy in matching imaging datasets. We show that these features are statistically significant, robust to perturbations, invariant across populations, and are localized to a small number of structural regions of the brain. Furthermore, we show that for task-specific connectomes, the regions identified by our method are consistent with their known functional characterization. We present a new matrix sampling technique to derive computationally efficient and accurate methods for identifying the discriminating sub-connectome and support all of our claims using state-of-the-art statistical tests and computational techniques.'
author:
- |
Vikram Ravindra\
Department of Computer Science\
Purdue University\
W Lafayette, IN 47906\
`[email protected]`\
Petros Drineas\
Department of Computer Science\
Purdue University\
W Lafayette, IN 47906\
`[email protected]`\
Ananth Grama\
Department of Computer Science\
Purdue University\
W Lafayette, IN 47906\
`[email protected]`\
bibliography:
- 'bibliography.bib'
title: Constructing compact signatures for individual fingerprinting of brain connectomes
---
Results {#results .unnumbered}
=======
Discussion {#discussion .unnumbered}
==========
Methods and Materials {#methods-and-materials .unnumbered}
=====================
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Fast and inexpensive characterization of materials properties is a key element to discover novel functional materials. In this work, we suggest an approach employing three classes of Bayesian machine learning (ML) models to correlate electronic absorption spectra of nanoaggregates with the strength of intermolecular electronic couplings in organic conducting and semiconducting materials. As a specific model system, we consider PEDOT:PSS, a cornerstone material for organic electronic applications, and so analyze the couplings between charged dimers of closely packed PEDOT oligomers that are at the heart of the material’s unrivaled conductivity. We demonstrate that ML algorithms can identify correlations between the coupling strengths and the electronic absorption spectra. We also show that ML models can be trained to be transferable across a broad range of spectral resolutions, and that the electronic couplings can be predicted from the simulated spectra with an 88% accuracy when ML models are used as classifiers. Although the ML models employed in this study were trained on data generated by a multi-scale computational workflow, they were able to leverage leverage experimental data.'
author:
- 'Loïc M. Roch'
- 'Semion K. Saikin'
- Florian Häse
- Pascal Friederich
- 'Randall H. Goldsmith'
- Salvador León
- 'Alán Aspuru-Guzik'
bibliography:
- 'main.bib'
title: From absorption spectra to charge transfer in PEDOT nanoaggregates with machine learning
---
Introduction
============
Organic-based materials are attractive for optoelectronic device applications, notably due to their low fabrication cost and their relative ease to produce and characterize.[@Prineted_book] Not only can the structural properties of these materials be tuned through the functionalization of molecules,[@Cheng_ChemRev2009] but they are also composed of elements which are Earth-abundant. In contrast to conventional inorganic electronic materials, organic compounds bring in flexibility,[@LOGOTHETIDIS200896] biocompatibility[@Rivnay-NatComm:2016] and biodegradability,[@Lei5107] as well as self-healing properties.[@Oh2016; @Ocheje2017] Organic conducting and semiconducting materials hold promises for several application niches, including next-generation wearable and printed photovoltaics,[@Facchetti_ChemMat2011; @Lipomi_AdvMat2011] fuel cells, [@guo2012self; @winther2008high] thermoelectrics, [@Dubey_JPolSci2011; @Wang_AdvEnMat2015; @Wei_Materials2015; @Bubnova_NatMat2011] and other optoelectronics applications.[@sun2015review; @Roch2017]\
One of the fundamental challenges for the design of organic optoelectronics lies in the intrinsic structural disorder of the materials. This disorder occurs on multiple length scales starting from the conformations of single molecules and the nearest-neighbor packing to the formation of multi-molecule domains and nanocrystals. The electronic properties of organic materials are highly sensitive to the packing of composing molecules hence dependent on the processing conditions.[@Hinton-JACS:2018] Fast optical probing of local electronic couplings can benefit both applied and fundamental research. On one hand, such a method brings the possibility to combine continuous testing of devices with roll-to-roll device manufacturing technology.[@Tabor:2018; @MI-report] On the other hand, optical characterization techniques can advance our understanding of charge transport in organic structures. In particular, UV-Vis, XPS, and Raman scattering measurements of thin films of conductive polymers can provide insight on composition and electronic structure, including the nature of charge carriers.[@Luo-JMCA:2013; @Ouyang-polymer:2004]\
The microscopic structure of molecule and polymer packing is difficult to measure directly. In contrast, obtaining optical spectra such as IR absorption, Raman scattering, electronic absorption, and fluorescence is more straightforward and requires sufficiently less experimental effort. Therefore, indirect characterization methods play a key role in evaluating the level of disorder. Both optical and electronic transport properties are influenced by the microscopic molecular packing. For example, in the simplest qualitative picture, the close proximity of two molecules yields an overlap of electronic clouds which leads to charge transfer. This proximity also leads to a F[ö]{}rster coupling between electronic excitations, which can be observed as changes in the lines in the electronic absorption spectra.[@Saikin-Nanophotonics:2013] Moreover, weak charge-transfer excitations are observed if the electronic coupling between molecules becomes sufficient. Because both effects are caused by the molecular interactions, in principle it is possible to find a machine learning model that correlates both of them. The conventional computational approach involves three steps: (i) building physical models that describe both properties of interests; (ii) fitting the parameters of the models to experimental data, e.g., absorption spectra; and (iii) using the fitted models to describe the other property, e.g., conductivity. Such an approach might be tedious, since the interrelations between these properties can be too complex to derive a simple or tractable physical model.\
Herein, we report an alternative approach, where the aforementioned physical model is replaced by machine learning (ML) models. To this end, we design a multi-scale computational workflow where the first three steps – force-field calculations, molecular dynamic simulations, and quantum-based approaches – generate the data for the ML algorithms. It is of course possible to employ experimental data in addition if it is available. In this work, and to begin with, we employed ML algorithms to identify correlations between the two properties of interests i.e., strength of intermolecular coupling and electronic absorption spectra. Then, we used the trained ML models as a relative classifier of the coupling strength of a given spectrum with respect to a reference coupling, which is to be defined by a scientist for the application at hand. As a model system demonstrating the reliability of the classifier to identify structures with strong electronic couplings from their electronic absorption spectra, we study pairs of poly(3,4-ethylenedioxythiophene) (PEDOT) oligomers. PEDOT is one of the most technologically-developed conducting polymers. Owing to its high hole conductivity and optical transparency in a doped state[@Groenendaal_AdvMat2000] it is has been used as transparent contacts in photovoltaic devices, touch screens, and light-emitting diodes. [@Lovenich_PolSci2014] PEDOT-based materials are frequently used as a mixture with polystyrene sulfonate polymers (PEDOT:PSS). In this mixture, PEDOT oligomers transfer the charges while PSS chains play the role of a solid electrolyte. This material becomes conductive at high concentrations of dopant.[@PEDOT_book]\
Although multiple experimental studies have addressed the molecular organization of PEDOT-based materials,[@Martin-PR:2010; @Groenendaal_AdvMat2000; @Takano_MacroM2012; @Nardes-AM:2007; @Crispin-JPS:2003; @Kemerink-JPCB:2004; @VanDeRuit-AFM:2013; @Crispin-ChemMater:2006; @Luo-JMCA:2013; @Ouyang-polymer:2004] the microscopic electronic states that lead to high conductance and the interplay between these states, optical properties, and the material structure have yet to be determined. As a matter of fact, the key factor for practical applications of PEDOT-based materials lies in understanding the relations between their solid-state packing and their unique electronic properties. Consequently, the main obstacle to elucidating this relationship is the strong structural disorder that appears on multiple length scales and is highly sensitive to the thin film preparation procedure.[@Groenendaal_AdvMat2000]\
Hereafter, we demonstrate that our ML models confirm the existence of correlation between the coupling strengths and the electronic absorption spectra. We also show the robustness of our ML models with respect to potential spurious statistical correlations to capture the relevant physical correlations. Finally, we use the ML models as classifiers to determine whether a given electronic absorption spectrum of interest relates to a coupling strength above or below an *a priori* selected reference coupling strength. Such an approach has proven to be reliable and robust, reaching an average error rate of only 12% when employing a Bayesian convolutional neural network. The importance of such a classifier becomes apparent in the context of the self-driving laboratories,[@SciRobot18:Roch; @Tabor:2018; @Hase-2019; @MI-report; @Dimitrov-ACS:2019] where the goal of the experimentation process to identify fabrication procedures for aggregates yielding high conductivities is embodied as an optimization procedure.\
Methods {#sec:methods}
=======
This section details the computational workflow designed to generate the data and to correlate electrical and optical properties using ML algorithms. The workflow is depicted in Fig. \[fig:concept\]. Each of the five composing steps (i.e., initial structures, refinement, MD simulation, physical models, ML models in Fig. \[fig:concept\]A) are described in their corresponding subsections.\
Hereafter, we assume that the packing of conjugated oligomers in the solid PEDOT:PSS mixture depends on the initial preparation procedure and post-processing steps. For PEDOT:PSS, such post-processing steps, typically evaluated via trial and error, are critical to achieve peak performance. Furthermore, it is hypothesized that PEDOT:PSS solid films consist of grains with a hydrophobic and highly conductive PEDOT-rich core and a hydrophilic insulating PSS-rich shell.[@David-2016; @Takano_MacroM2012] This phase segregation of PEDOT and PSS occurs on a scale beyond current computational capabilities and, thus, is not captured by our model. Nonethless, our computational workflow allows to study the disorder within each of these grains.\
![The computational pipeline used, from structure generation to correlating $J$ and electronic absorption spectra. (A) General workflow highlighting the steps and summarizing the methods involved. (B) Structures of PEDOT and PSS, represented in the top and bottom panels, respectively. (C) One of the ten supercells of PEDOT:PSS blends. (D) Two distinct example pairs of PEDOT oligomers extracted from the PEDOT:PSS bulk. $J_1$ and $J_2$ are the coupling strengths for each of the pairs. (E) Associated simulated electronic absorption spectra. (F) State, and (G) orbital representation of the monomers and dimers involved in the calculation of the coupling strength, $J$.[]{data-label="fig:concept"}](computational_pipeline_v7.png){width="90.00000%"}
#### Initial structures. {#initial-structures. .unnumbered}
To screen the orientation stability of the PEDOT:PSS complex within the grains, 100 starting structures were generated using the Packmol software package.[@Packmol2009] Each of the complexes consisted of one PEDOT chain with eight 3,4-ethylenedioxythiophene units ($n=8$, Fig. \[fig:concept\]B) carrying two positive charges, and two PSS chains consisting of three poylstyrene sulfonate units ($m=3$, Fig. \[fig:concept\]B) with one negative SO$_3^-$ and two SO$_3$H groups per chain. In the generation of the initial structures, we constrained the SO$_3^-$ group of the PSS chain to point towards the positive PEDOT chain. These initial structures were then optimized using a classical force field (FF) approach.
#### Refinement. {#refinement. .unnumbered}
The 20 energetically most stable PEDOT:PSS complexes obtained from the initial structure search were relaxed at the B97-D/6-31G(d,p) level of theory in the gas phase, using the Gaussian software package.[@g09] Note that the influence of the solvent was found to be negligible, and that the performance of the B97-D functional on geometries has already been assessed in previous work.[@Roch:jctc-2017] Single point energy calculations at the HF/6-31G(d,p)//B97-D/6-31G(d) level were performed on the 20 relaxed complexes to parametrize the charges of the molecular dynamics simulation, as customary with GAFF.[@GAFF2004]
#### Molecular dynamics simulation. {#molecular-dynamics-simulation. .unnumbered}
Ten molecular dynamics simulations were carried out on PEDOT:PSS model systems in periodic cubic boxes of size[^1] $\times$ $\times$ for a density of *ca.* , with the LAMMPS software package.[@PLIMPTON19951] The GAFF force-field was chosen to describe the systems. PEDOT oligomers with eight repeat units ($n=8$, Fig. \[fig:concept\]B) and a +2 charge were considered, while the PSS atactic chains consisted of 20 repeat units ($m=20$, Fig. \[fig:concept\]B), with four deprotonated units randomly distributed in the sequence of each chain. Note that the PSS chain length was increased from $m=3$ to $m=20$ to better represent experimental blends. Additional details can be found in the Supporting Information (see section \[sec:si\_md\]).
#### Physical models to compute $J$, and simulate the electronic absorption spectra. {#physical-models-to-compute-j-and-simulate-the-electronic-absorption-spectra. .unnumbered}
For two interacting PEDOT monomers, denoted $A$ and $B$ from hereon, the strength of the charge transfer integral $J$ can be determined in several ways.[@Bredas_JACS1983; @Bredas_CPL2002; @Bredas_JACS2008; @martin2000monodisperse; @bredas:jcp-2007; @Apperloo_ChemEurJ2002; @Beljonne_AdvFM2001; @Li-JCP:2007; @deibel:prb-2011; @Kubas-JCP:2014; @Cave-JCP:97] Bi-polaron charge transport models have previously been discussed in the case of PEDOT systems.[@Bredas_AChR1985] For the sake of simplicity, we assume a single-polaron transport model. Nonetheless, the ML models used for electronic coupling prediction are agnostic to the type of transport and will learn correlations between $J$ and electronic absorption spectra independently of the type of the charge transport model. To calculate the coupling $J$, we used the framework of a tight-binding formalism[@MolCrys-Book; @Bassani-Book] as well as a Kohn-Sham orbital based method.[@bredas:jcp-2007; @deibel:prb-2011] In both cases, orbital energies were obtained at the B3LYP/def2-SV(P) level of theory. The choice of basis set and functional balances computational cost and accuracy. Detailed results on the performance of the def2-SV(P) results can be found in the Supporting Information (see section \[sec:basis\_set\]).\
The nearest-neighbor PEDOT dimers were extracted across the ten simulation boxes. These pairs were selected according to a distance criterion; any pair of PEDOT molecules having at least two heavy (i.e., non hydrogen) atoms at a distance closer than is selected. The cutoff distance was taken to be comparable to the sum of the van der Waals radii of these heavy atoms. This procedure lead to the selection of 1,420 PEDOT pairs, ulteriorly used to model the strength of the charge transfer, and to simulate the electronic absorption spectra.\
Tight-binding formalism is based on the change of orbital energies when going from isolated monomers to dimer systems [@Bredas-ChemRev:2004]
$$\centering
J=\sqrt[]{\left(\Delta\epsilon_{\left|AB\right>,+5}^{H,L}\right)^2 - \frac{1}{4}\left[ \left(\epsilon_{\left|A\right>,+3}^{H,H-1} + \epsilon_{\left|B\right>,+2}^{H,H-1}\right) - \left(\epsilon_{\left|A\right>,+2}^{H,H-1} + \epsilon_{\left|B\right>,+3}^{H,H-1}\right) \right]^2},
\label{eq:j}$$
where $\Delta\epsilon_{\left|AB\right>,+5}^{H,L}$ is the splitting between the HOMO and the HOMO-1 level of the dimer $AB$ with the charge $q=+5$, and $\epsilon_{\left|i\right>,+q}^H$ is the HOMO energy of monomer $i$, carrying charge $+q$. Note that the correction due to the offset between the HOMOs of the monomers is negligible; hence, $J$ is mostly governed by the splitting $\Delta\epsilon_{\left|AB\right>,+5}^{H,H-1}$. This formalism considers frontier orbitals assuming that only highest occupied orbitals are hybridized due to the electronic coupling between the oligomers (see Fig. \[fig:concept\]F-G). Both the presence of non-equilibrated charges and non-zero spin makes the model rather complicated. Nonetheless, our interpretation of eq. \[eq:j\] is a lower estimate for the electronic coupling between the oligomers. Note that this model can also be used to describe bi-polaron transport.\
The second approach to calculate charge transfer integrals uses the Kohn-Sham orbitals of isolated monomers as well as the Fock matrix and the overlap matrix of the dimer systems:[@bredas:jcp-2007; @deibel:prb-2011]
$$J_{\text{AB}}=\frac{F_{\text{AB}}-\frac{1}{2}(F_{\text{AA}}+F_{\text{BB}})S_{\text{AB}}}{1-S_{\text{AB}}^2}
\label{eq:j2}$$
The matrix elements $F_{\text{AB}}=\left<\text{A}|F_{\text{dimer}}|\text{B}\right>$ and $S_{\text{AB}}=\left<\text{A}|S_{\text{dimer}}|\text{B}\right>$ are calculated using the Fock and overlap matrices of a molecular dimer system with a charge of $+4$. The states $\left.|\text{A}\right>$ and $\left.|\text{B}\right>$ are the highest occupied molecular orbitals of the doubly positively charged monomers. The distributions of electronic couplings obtained with eq. \[eq:j\], and with eq. \[eq:j2\] are depicted in Fig. \[sec:si\_coupling\_strength\_distributions\].\
The electronic absorption spectra of the 1,420 PEDOT pairs were simulated using the computed TD-CAM-B3LYP/def2-SV(P) transitions, in the gas phase. Note that for the simulation of the spectra, the influence of solvation was found to be negligible. We employed a Lorentzian distribution to broaden the TD-DFT transitions. This translates the discrete oscillator strength, $f$, and transition energies, $\omega$, to a continuous spectra to resemble experimental outcomes. The broadening was chosen to be .[@Randall:NL-2018]\
Some insights about correlations between electronic absorption spectra and intermolecular charge transfer can be obtained only for the case when the coupling is weak. A completely relaxed doubly charged oligomer composed of eight to ten units would have a strong electronic transition at about 0.9 – .[@David-2016] This transition is predominately composed of HOMO and LUMO orbitals. For an oligomer with an odd number of charges, additional HOMO-1 $\rightarrow$ HOMO transition appears at lower frequencies, at *ca.* . Therefore, the low frequency part of the spectra of dimers without the interaction should be composed of three lines – a low-frequency, weak transition and a strong doublet. Electronic or excitonic interaction between the molecules modulate the spectra. Specifically, weak charge transfer transitions should appear at the low frequency tail of the spectra. The intensities of these transitions should be sensitive to the coupling strength, while their frequencies should be more stable as determined by the alignment of the molecular energy levels. However, this intuitive picture fails for intermediate and strong intermolecular couplings. In the latter case, the intramolecular states hybridize, which in turn leads to a realignment of the energy levels and a redistribution of the oscillator strengths among multiple transitions. Yet another advantage of our ML approach is that it allows to capture correlations between the electronic interaction and optical spectra independently of the coupling regime used as our ML independent variable.
#### Machine learning models to identify correlations between electronic absorption spectra and coupling strengths. {#machine-learning-models-to-identify-correlations-between-electronic-absorption-spectra-and-coupling-strengths. .unnumbered}
Correlations between electronic absorption spectra and coupling strengths were identified with ML models at different levels of complexity. To mimic experimental conditions, we encoded the Lorentzian broadened electronic absorption spectra based on their intensities at specific frequencies, using a binning on the considered frequency domain ( to ). A total of 170 () of the 1,420 spectra-coupling pairs were randomly selected to construct a test set. The remaining 1,250 () of the dataset were used for ten-fold cross-validation. The size of the dataset motivates the use of Bayesian models for a robust and transferable identification of relevant physical correlations.\
Specifically, we employed three different classes of Bayesian models (see Fig. \[fig:ml\_models\]): (i) Bayesian linear regression models assume a linear dependence of coupling strengths on electronic spectra and thus present the simplest approximation; (ii) Bayesian multi-layer perceptrons (MLPs) are Bayesian generalizations of conventional deterministic MLPs with similar flexibility to model non-linear relations while retaining the robustness to overfitting of Bayesian methods; and (iii) Bayesian one-dimensional convolutional neural networks (CNNs) are special cases of MLPs which have the potential to efficiently exploit spatial correlations in the presented features due to their sparse nature. All models were set up to predict coupling strengths directly from the intensities of the associated electronic absorption spectra at different frequencies. Additionally, we constructed Bayesian MLPs which are trained on a compressed representation of the electronic absorption spectra obtained from principle component analysis (PCA). All models were trained based on an early-stopping criterion. Hyperparameters for all four models are optimized in a random grid search, and the details are provided in the Supporting Information (see section \[sec:si\_hyperparameter\_optimization\]).
![Schematic representation of (A) the data processing and (B) the three classes of ML models, depicting Bayesian linear regression models, Bayesian multi-layer perceptrons (MLPs), and Bayesian one-dimensional convolutional neural networks (CNN). Note that the MLPs are trained either on the raw intensities (blue array) or on the compressed representation after a principle component analysis (green array).[]{data-label="fig:ml_models"}](ML_models_shadow-effect_4.png){width="75.00000%"}
#### Machine learning models as relative classifiers. {#machine-learning-models-as-relative-classifiers. .unnumbered}
The aforementioned ML models trained for predicting absolute values of coupling strengths from electronic absorption spectra can be used to classify the conductivity performance associated with the electronic absorption spectra of the materials. Instead of asking for the absolute value of the coupling strength, the model provides an estimate for whether the considered coupling is above or below a reference. This reference is a hyperparameter defined by a scientist as a threshold for high and low values of $J$.\
Results and discussion
======================
We begin by discussing our results with the performance of the ML models to identify correlations and to predict absolute values of coupling strengths from electronic absorption spectra. Fig. \[fig:test\_set\_predictions\] illustrates the accuracies of all four models to predict coupling strengths computed from the tight-binding formalism with all four models after a full hyperparameter optimization. Then, we detail the test designed to assess the performance of the ML models to capture relevant physical correlation. We continue our discussion with the results obtained when the ML models are used as relative classifier, where associated error rates are reported in Fig. \[fig:contingency\_plot\]. We also highlight the practicality of such an approach in discovery applications with the self-driving laboratories. Finally, we discuss the robustness of our ML models upon variations of the peak broadening, which experimentally translates to noise. Fig. \[fig:predictions\_for\_broadenings\] illustrates the performance of our ML models at different broadening.
Correlation between electronic spectra and coupling strengths
-------------------------------------------------------------
All constructed ML models are able to predict coupling strengths at positive coefficients of determination ($r^2$), which indicate that the coupling strengths correlate with the electronic absorption spectra of the PEDOT dimers and that the models are capable of identifying this correlation. Bayesian CNNs provide the most accurate predictions based on all computed comparative metrics ($r^2 = 0.313$, RMSD $= 0.107$ eV, MAD $= 0.082$ eV)[^2] and Bayesian linear regression yields the least accurate predictions ($r^2 = 0.163$, RMSD $= 0.118$ eV, MAD $= 0.092$ eV). We further observe an improved prediction accuracy when compressing the electronic absorption spectra via PCA for the Bayesian MLP models.\
Despite the relatively small size of the dataset, we found that the studied ML models, most notably the Bayesian CNN, present an efficient approach to identify the relevant correlations. Estimations of the sampling efficiency of the Bayesian CNN model suggest that it can be trained to reach similar prediction accuracies with only 850 instead of 1,250 training points. No significant improvement in the prediction accuracy is observed when increasing the size of the training set from beyond 850 to 1,250 examples (see Supporting Information, section \[sec:si\_sampling\_efficiency\]). This observation, in conjunction with the generalization of the models observed for the test set predictions, indicates that the Bayesian CNN exploits all identifiable correlations to their full extent.\
To ensure that our ML models did not capture spurious correlations that could arise from the methods and formalisms employed to compute the electronic absorption spectra and model the coupling strengths, we tested for the nature of the identified correlations by training the Bayesian CNNs to predict couplings strengths obtained with the Kohn-Sham orbital formalism from the same electronic absorption spectra. Details are provided in the Supporting Information (see section \[sec:si\_cross\_predictions\]). The trained Bayesian CNNs achieve prediction accuracies of $r^2 = 0.264$ on the same test set. In addition, we constructed a hybrid dataset, where half of the couplings are randomly chosen from the tight-binding formalism and the other half from the Kohn-Sham orbital formalism. Again, the trained Bayesian CNNs achieve prediction accuracies of $r^2 = 0.280$ indicating that the presented ML models do not capture potential spurious statistical correlations but extract the relevant physical correlations.\
Machine learning models as relative classifiers
-----------------------------------------------
The practicality of the presented trained ML models becomes more apparent when weakening the requirement of accurate absolute predictions of the coupling strengths to accurate relative predictions, which are of interest in discovery applications. Rather than requesting an estimate for the actual numeric value of the coupling strength, the trained model is used to determine if a given electronic absorption spectrum of interest relates to a coupling strength above or below an *a priori* selected reference coupling strength. For such scenarios, the prediction accuracy of the model can be assessed by treating it as a binary classifier to determine true positive and true negative rates for different reference coupling strengths.\
We assess the prediction accuracies of such relative classifiers by estimating the probability of the model to make a correct prediction, i.e., predicting the coupling to be above the reference when it is above or predicting the coupling to be below the reference when it is below, versus the probability of the model to make an incorrect prediction, i.e., predicting the coupling to be above the reference when it is below and vice versa (see Fig. \[fig:contingency\_plot\]). These probabilities are estimated for different reference coupling strengths, spanning the entire range of coupling strengths computed with the tight-binding formalism.\
Fig. \[fig:contingency\_plot\] illustrates the error rates, i.e., wrongly predicting a coupling strength to be above or below the considered reference coupling strength, for the trained ML models along with two naïve models for comparison. The most naïve model draws random samples from a uniform distribution to predict the coupling strength for a given electronic absorption spectrum (model-1, depicted in light grey in Fig. \[fig:contingency\_plot\]). A slightly more sophisticated naïve model predicts by drawing random samples from the distribution of coupling strengths (model-2, depicted in dark grey in Fig. \[fig:contingency\_plot\]).\
We find that the error rates for the two naïve models yield the largest error rates: and . Bayesian linear regression scored an average error rate of , which, despite its simplicity, already provides an advantage over simple models and captures some of the relevant correlation in the dataset. The lowest error rate is observed for the Bayesian CNN with a average error for coupling strengths chosen within the range of smallest and largest computed coupling strengths. Additionally, it is noted that Bayesian CNN never exceeds an error of for any chosen reference coupling. In fact, if the focus of discovery process is to identify fabrication procedures and post-processing steps leading to large coupling strengths (above ), the Bayesian CNN yields error rates of less than . We suggest that the trained ML models can be applied to classify the coupling strengths with respect to a reference coupling strength with reasonable confidence.\
Robustness of the machine learning models
-----------------------------------------
Finally, we estimate the dependence of the model performances on the particular choice of the peak broadening. While we demonstrated that for this particular choice, the trained ML models are indeed capable of identifying the relevant correlations between the electronic absorption spectra and the coupling strengths, experimentally obtained electronic absorption spectra might be noisy and feature peaks at slightly varying broadenings. The robustness of the model predictions for different broadenings is tested by predicting coupling strengths from electronic absorption spectra at different broadenings, ranging from to . Note that while a broadening is too small for a room temperature measurement, a broadening would correspond to an unphysically fast dephasing rate. Fig. \[fig:predictions\_for\_broadenings\] summarizes the coefficients of determination for predictions of coupling strengths from electronic absorption spectra generated at different broadenings. Note that all coupling strengths are predicted by models, which were trained on electronic absorption spectra at a broadening.
We find that prediction accuracies of the Bayesian linear regression model are mostly insensitive to the particular broadening value (yellow trace). Only for very small broadenings below and very large broadenings above degradations in the predictive power can be observed. Bayesian MLPs (turquoise trace) show faster degradations in their prediction accuracy for small and large broadenings, but maintain comparative predictive powers across broadenings of to . Bayesian CNNs (blue trace) are the least robust with respect to changes in the broadening, with accurate predictions only within the to interval. Nevertheless, they demonstrate their predictive power for varying broadenings despite having been trained on peaks of one particular broadening, indicating that ML models can indeed be trained to identify relevant correlations without being overly sensitive to the broadening of the peaks.\
Associating ML models with experimental studies
-----------------------------------------------
A long-standing goal of experimental materials characterization of organic optoelectronic materials is a map of how electronic properties are distributed in space as a result of different instantaneous molecular configurations. Scanning probe[@ODea-MRS:2012; @Groves-ACR:2010; @Eisfeld-2018] and super-resolution optical measurements[@Penwell-NatMater:2017] can provide readouts of electronic properties on length scales below the diffraction limit. Simultaneously, single-particle measurements[@Barbara-ACR:2005; @Thiessen-PNAS:2013] can provide a bottom-up understanding of how optoelectronic properties evolve from molecular precursors. For conductive polymers like PEDOT:PSS, single-particle measurements have been particularly difficult to employ due to the lack of emission in these materials because of rapid quenching. Simultaneously, single-particle measurements have tremendous spectroscopic utility due to reduced inhomogeneous broadening. Use of high quality-factor optical microresonators as the readout for ultrasensitive photothermal spectroscopy[@Heylman-NatPhoton:2016] has allowed the first single-particle optical measurements to be performed on PEDOT:PSS,[@Randall:NL-2018] even down to a single or a small number of polymer strands. This study provided an experimental bound for the line broadening used in the above simulations. More recently, optical microresonator spectroscopy has been used to show how annealing processing act on single PEDOT:PSS polymer particles.[@Rea-InReview] A means of directly connecting spectral measurements on single PEDOT:PSS polymer strands and particles to electronic couplings would significantly amplify the information content of these experiments.\
Conclusion
==========
Our findings suggest that ML models can identify physical correlations between the measurable electronic absorption spectra and the strength of intermolecular electronic couplings, which in turn determine the charge transport. While the presented ML models provide coupling strength estimates with limited accuracy, relative estimates with respect to reference coupling strengths show promising error rates. Using the trained Bayesian CNN model to classify given electronic absorption spectra above or below an *a priori* selected reference coupling strength displays an error rate of only , and as low as at high coupling regime. With such a promising error rate, we suggest to use the trained models as classifiers to evaluate performance of fabrications procedure and post-processing steps. Further investigations towards the construction of reliable and transferable ML models, notably the usage of ensemble methods such as adaboost,[@Freund:1997; @Freund:1996] or mixture density networks,[@Bishop:1994; @Richter:2003] might allow for more detailed insights into the relation between couplings and electronic absorption spectra. Another important venue for improvement of our approach is the incorporation of features, such as structural information, which would introduce the notion of similarity between complexes.\
We believe that the combination of the developed approach with spectroscopy techniques and its integration with the self-driving laboratories[@SciRobot18:Roch; @Tabor:2018; @Hase-2019; @MI-report; @Dimitrov-ACS:2019] has the potential to enhance characterization and accelerate optimization of organic materials. As experimental approaches for providing optical readouts improve in sensitivity, spatial resolution, and access to different spectral features, growth in theoretical treatments will allow one to draw deeper connections between these measurements and the underlying molecular structure. We also envision the use spectroscopic methods to measure spectra of nanoaggregates with high-finesse toroidal optical cavities.
The computations were done on the Arran cluster supported by the Health Sciences Platform (HSP) at Tianjin University, P.R. China, on the Odyssey cluster supported by the FAS Division of Science, Research Computing Group at Harvard University, USA, and on the computational resource ForHLR II funded by the Ministry of Science, Research and the Arts Baden-W[ü]{}rttemberg and DFG (“Deutsche Forschungsgemeinschaft”). L. M. R. and A. A.-G. acknowledge Natural Resources Canada (EIP2-MAT-001) for their financial support. F. H. acknowledges support from the Herchel Smith Graduate Fellowship and the Jacques-Emile Dubois Student Dissertation Fellowship. P. F. has received funding from the European Unions Horizon 2020 research and innovation program under the Marie Skodowska-Curie grant agreement MolDesign No 795206. L. M. R., S. K. S., F. H., P. F., and A. A.-G. thank Dr. Anders Fr[ø]{}seth for generous support. R. H. G. acknowledges the National Science Foundation (NSF: DMR-1610345). S. L. acknowledges the Real Colegio Complutense in Harvard for a Research Grant, and to the Spanish Ministerio de Ciencia e Innovaci[ó]{}n for a Fellowship through the Salvador de Madariaga Program.
The following files are available free of charge.
- SupportingInformation.pdf: Supporting information to the main manuscript
\
Assessment of the basis set {#sec:basis_set}
===========================
$\epsilon_\mathrm{HOMO}$ $\epsilon_\mathrm{HOMO-1}$ $\Delta \epsilon$
-- -------------------------- ---------------------------- ------------------- -----
Def2-SVP -9.134 -10.018 885
Def2-SVPD -9.254 -10.141 887
Def2-SV(P) -9.144 -10.030 885
Def2-TZVP -9.164 -10.051 887
Def2-SVP -11.850 -12.420 570
Def2-SVPD -11.934 -12.505 571
Def2-SV(P) -11.863 -12.434 571
Def2-TZVP -11.842 -12.411 569
Def2-SVP -9.139 -10.067 928
Def2-SVPD -9.263 -10.189 926
Def2-SV(P) -9.149 -10.077 929
Def2-TZVP -9.171 -10.096 925
Def2-SVP -11.932 -12.518 586
Def2-SVPD -12.017 -12.605 588
Def2-SV(P) -11.944 -12.531 586
Def2-TZVP -11.921 -12.510 589
Def2-SVP -11.721 -12.594 873
Def2-SVPD -11.860 -12.733 873
Def2-SV(P) -12.319 -12.809 489
Def2-TZVP -12.358 -12.739 381
: Energies of the HOMO and HOMO-1 molecular orbitals of pair \#1, in \[eV\], at the CAM-B3LYP level of theory. $\Delta \epsilon$ is the energy difference between the HOMO and the HOMO-1, in \[meV\].[]{data-label="table:pair1_BSet"}
$\epsilon_\mathrm{HOMO}$ $\epsilon_\mathrm{HOMO-1}$ $\Delta \epsilon$
-- -------------------------- ---------------------------- ------------------- ------
Def2-SVP -9.126 -10.047 921
Def2-SVPD -9.246 -10.166 921
Def2-SV(P) -9.137 -10.058 922
Def2-TZVP -9.156 -10.074 919
Def2-SVP -12.031 -12.551 520
Def2-SVPD -12.119 -12.645 525
Def2-SV(P) -12.044 -12.565 521
Def2-TZVP -11.653 -12.445 792
Def2-SVP -9.190 -10.206 1017
Def2-SVPD -9.301 -10.321 1020
Def2-SV(P) -9.200 -10.218 1017
Def2-TZVP -9.208 -10.227 1018
Def2-SVP -12.031 -12.668 637
Def2-SVPD -12.114 -12.749 635
Def2-SV(P) -12.044 -12.682 638
Def2-TZVP -12.021 -12.654 633
Def2-SVP -11.783 -12.744 961
Def2-SVPD – – –
Def2-SV(P) -11.793 -12.755 962
Def2-TZVP -11.842 -12.789 947
: Energies of the HOMO and HOMO-1 molecular orbitals of pair \#2, in \[eV\], at the CAM-B3LYP level of theory. $\Delta \epsilon$ is the energy difference between the HOMO and the HOMO-1, in \[meV\].[]{data-label="table:pair2_BSet"}
J Pair \#1 Pair \#2
------------ ---------- ----------
Def2-SVP 870 959
Def2-SVPD 870 –
Def2-SV(P) 483 960
Def2-TZVP 374 893
: J in \[meV\].[]{data-label="table:J_BSet"}
Molecular dynamics {#sec:si_md}
==================
To ensure equilibration of the systems, the following approach has been adopted. First, random starting configurations have been generated with the packmol program, and each one has been allowed to relax with an energy minimization followed by a molecular dynamics simulation of within the NVT ensemble at . After that, a NPT simulation of at and to relax the dimensions of the box. Then, the temperature of the system has been subjected to a heating up to in a run, a NVT run at for another , and a cooling down to for . After that, the production run has been carried out in the NPT ensemble at and for . To more thoroughly explore the configurational space, a total of ten different starting configurations generated by Packmol have been simulated through this procedure.
Hyperparameter optimization of correlation models {#sec:si_hyperparameter_optimization}
=================================================
Well performing model architectures were determined from a random grid search of selected hyperparameter for each model. Tab. \[tab:hyperparam\_ranges\] summarizes the hyperparameters which were varied for each model and their respective ranges. Multi-layer perceptrons (applied to the full spectra and PCA contracted spectra) were set up with three hidden layers but varying number of neurons per layer. Activations for all model architectures were chosen to be either a leaky version of the ReLU function ($\alpha = 0.2$) or the softplus function.
$$\begin{aligned}
\text{leaky\_ReLU}(x) &= \begin{cases} x, & \text{if } x > 0 \\ \alpha x, & \text{otherwise} \end{cases} \\
\text{softplus}(x) &= \log(1 + \exp(x))
\end{aligned}$$
Tab. \[tab:hyperparam\_ranges\] also reports the set of hyperparameters for which the trained models achieved the highest prediction accuracies as determined from a 10-fold cross-validation protocol. Best performing sets of hyperparameters were determined from a random grid search with a total of 512 different models constructed for each model class. Training of individual models was aborted based on an early stopping criterion.
Model Hyperparameter optimal lower bound upper bound
------------------------------------ ------------------- --------------- ------------- -------------
[Bayesian Linear Regression ]{} Regularization $10^{-3.875}$ $10^{-4}$ $10^1$
Learning rate $10^{-4.75}$ $10^{-5}$ $10^{-1}$
[Convolutional Neural Network ]{} Regularization $10^0$ $10^{-4}$ $10^1$
Learning rate $10^{-1.875}$ $10^{-5}$ $10^{-1}$
Filters $24$
Filter sizes $20, 50$
Filter activation softplus
Neurons / layer $2$
Dense activation leaky ReLU
[Multi-Layer Perceptron (full) ]{} Regularization $10^{-2}$ $10^{-4}$ $10^1$
Learning rate $10^{-3.5}$ $10^{-5}$ $10^{-1}$
Neurons / layer $91$ $1$ $100$
Activations softplus
[Multi-Layer Perceptron (pca) ]{} Regularization $10^{-3.5}$ $10^{-4}$ $10^1$
Learning rate $10^{-1.5}$ $10^{-5}$ $10^{-1}$
Neurons / layer $41$ $1$ $100$
Activations leaky ReLU
: Hyperparameter ranges and sets of hyperparameters for which trained models achieved the highest prediction accuracies. Best performing sets of hyperparameters were determined from a random grid search. []{data-label="tab:hyperparam_ranges"}
Sampling efficiency {#sec:si_sampling_efficiency}
===================
The predictive power of a regression model depends on the how representative the training set is for the underlying (unknown) physical correlations. Typically, larger training sets with more examples of absorption spectra and associated couplings yield higher prediction accuracies. To asses how many examples are needed for sufficient prediction accuracies we train the introduced regression models on training sets of different sizes. More specifically, we employ the [Convolutional Neural Network ]{}model (see main text for details) with the determined best performing set of hyperparameters. The training set of 1,250 examples is reduced by randomly selected examples. Following this procedure, we generate a total of $20$ training subsets of different sizes. [Convolutional Neural Network ]{}models are then fully trained on their respective training set and their predictive powers are assessed via test set predictions. Note, that the test set is identical for all predictions. Results are reported in Fig. \[fig:sampling\_efficiency\]
We observe a steady increase in the average prediction accuracy of the [Convolutional Neural Network ]{}model when increasing the size of the training set from $100$ examples up to about $850$ examples. From there on, no significant change in the average prediction accuracy can be observed with a further increase of the training set size.
Prediction accuracies for coupling strengths computed with the Kohn-Sham orbital formalism {#sec:si_cross_predictions}
==========================================================================================
ML models can, at best, identify statistical correlations between the features and the targets to which they are exposed during the training process. As such, the positive correlation between predicted and computed coupling strengths obtained from the tight-binding formalism does not necessarily indicate that the ML models identified relevant physical correlations, and might just be due to statistical correlations based on the impreciseness of the methods used to compute Vis/NIR spectra and/or coupling strengths.\
We suggest to test if the ML models in the main text only identified statistical correlations (and not the relevant physical correlations) by constructing additional models trained to predict coupling strengths computed with the Kohn-Sham orbital formalism. Similar to the ML models reported in the main text, we run a full hyperparameter optimization for the Bayesian CNN models to predict Kohn-Sham orbital formalism coupling strengths. Test set predictions of the best performing model as identified from the hyperparameter search are reported in Fig. \[fig:sampling\_efficiency\]. We observe, that the trained Bayesian CNN is capable of identifying correlations between the Vis/NIR spectra and the coupling strengths even for this case where coupling strengths are computed based on the Kohn-Sham orbital formalism. Moreover, we observe similar prediction accuracies despite significantly different coupling strength distributions (see section \[sec:si\_coupling\_strength\_distributions\]).
If the predictive power of the Bayesian CNN reported in the main text relied on spurious statistical correlations between the Vis/NIR spectra and the tight-binding coupling strengths, the model presented in this section no longer has the opportunity to exploit these spurious correlations due to the change in formalism for computing coupling strengths. However, we still observe relatively high prediction accuracies for Kohn-Sham orbital couplings despite the change in the range of the couplings and the change in their distribution. We conclude that the predictive power of the Bayesian CNN model presented in Fig. \[fig:kohn\_sham\_predictions\] must either be due to the model identifying the relevant physical correlations, or other statistical correlations arising from the Kohn-Sham orbital couplings.\
To rule out the possibility that the Bayesian CNN presented in Fig. \[fig:kohn\_sham\_predictions\] identified other spurious statistical correlations yielding similar prediction accuracies we propose the construction of a hybrid dataset. This hybrid dataset is constructed by randomly choosing half of the coupling strengths computed with the tight-binding formalism, and the other half with the Kohn-Sham orbital formalism. Note, that the differences in the ranges of the coupling strengths are accounted for by standardizing coupling strengths based on the characteristics of each individual dataset following
$$\begin{aligned}
j = \frac{J - \langle J \rangle_\text{train}}{\sqrt{ \langle \left( J - \langle J \rangle_\text{train} \right)^2 \rangle_\text{train} }},
\end{aligned}$$
where $\langle \cdot \rangle_\text{train}$ denotes the average over the training set.
We find that the proposed Bayesian CNN architecture yields prediction accuracies comparable to prior experiments with coupling strengths computed from one or the other formalism. We interpret this observation as an indicator showing that the model does not solely rely on spurious statistical correlations caused by inaccuracies of the physical models used to compute the coupling strengths.
Distribution of electronic coupling {#sec:si_coupling_strength_distributions}
===================================
![Distribution of electronic couplings in PEDOT:PEDOT complexes. The total number of complexes is 1,420. (A) is computed with the tight binding formalism model, and (B) with the orbital overlap method.[]{data-label="fig:couplings"}](J_g09_log_color.png "fig:"){width="0.7\linewidth"} ![Distribution of electronic couplings in PEDOT:PEDOT complexes. The total number of complexes is 1,420. (A) is computed with the tight binding formalism model, and (B) with the orbital overlap method.[]{data-label="fig:couplings"}](J_turbomole_log_color.png "fig:"){width="0.7\linewidth"}
[^1]: Note: average size of the different boxes for the duration of the NPT simulations
[^2]: RMSD: root-mean-square deviation, MAD: mean-absolute deviation, and $r^2$: coefficients of determination
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'This contribution argues that the notion of time used in the scientific modeling of reality deprives time of its real nature. Difficulties from logic paradoxes to mathematical incompleteness and numerical uncertainty ensue. How can the emergence of novelty in the Universe be explained? How can the creativity of the evolutionary process leading to ever more complex forms of life be captured in our models of reality? These questions are deeply related to our understanding of time. We argue here for a computational framework of modeling that seems to us the only currently known type of modeling available in Science able to capture aspects of the nature of time required to better model and understand real phenomena.'
author:
- Roger White
- Wolfgang Banzhaf
bibliography:
- 'NaturalTime.bib'
date: 'November 19, 2018'
title: Putting Natural Time into Science
---
[^1]
INTRODUCTION {#sec:Introduction}
============
Since its origin more than two millennia ago, epistemological thinking in the West has been driven by a desire to find a way to certify knowledge as certain. Mathematics and logic seemed to offer a template for certainty, and as a consequence, modern science as it emerged from the work of scholars like Galileo and Newton aimed as much as possible for mathematical formalization. But by the beginning of the twentieth century the certainty of logic and mathematics was no longer an unquestioned truth: it had become a research question. From paradoxes in the foundations of logic and mathematics to equations with solutions that are in a sense unknowable, certainty had begun to seem rather uncertain. By the 1930s, as mathematicians and logicians grappled with the problem of formalisation, they found themselves being forced to look beyond formal systems, and to contemplate the possibility that the solution was of a different nature than they had imagined, that the answers lay in the realm of the living, creative world — the world of uncertainty. Although they did not systematically pursue this line of thought, we believe that their intuition was essentially correct, and it serves as our starting point.
The basic reason for the failure of the formalisation programme, we contend, has to do with the inherent nature of logic and mathematics, the very quality that makes these fields so attractive as the source of certainty: their timelessness. Logic and mathematics, as formal systems, exist outside of time; hence their truths are timeless – they are eternal and certain. But because they exclude time, they are unable to represent one of the most fundamental characteristics of the world: its [*creativity*]{}, its ability to generate novel structures, processes, and entities.
Bergson [@bergson1911], by the beginning of the twentieth century, was already deeply bothered by the inability of mathematics and formal science to handle the creativity that is ubiquitous in the living world, and he believed that this was due to the use of “abstract” time — a representation of time — rather than “concrete” or real time. In spite of his perceptive analysis of the problem, however, he saw no real solution. In effect mathematics and the hard sciences would be unable to address the phenomenon of life because there was no way they could embody real or natural time [^2] in their formal and theoretical structures. Real time could only be intuited, and intuition fell outside the realm of hard science. This view had been anticipated by Goethe [@amrine1990; @miller2012], who took metamorphosis as the fundamental feature of nature. Implicitly recognising that formal systems could not encompass metamorphosis, he proposed a scientific methodology based on the development of an intuition of the phenomena by means of immersion in them. A recent echo of Bergson is found in the work of Latour [@latour2012], who argues that existence involves continuous re-creation in real (natural) time, while the representations of science short-circuit this process of constructive transformation by enabling direct jumps to conclusions. He cleverly refers to this time-eliminating aspect of science as “double clic”.
The solution to the problem of time, we believe, is to re-introduce real, natural time into formal systems. Bergson did not believe there was any way of doing this, because science was essentially tied to mathematics or other systems of abstraction which seemed always to eliminate real time. Today, however, we can introduce real or natural time into our formal systems by representing the systems as algorithms and executing them on a computer, which because it operates in natural time, introduces natural time into the algorithm. The answer, in other words, lies in computing. (In this chapter we will normally use the expression *natural time* rather than real time in order to avoid confusion with the common use of the latter expression to refer to something happening immediately, as in “a real time solution”.)
This contribution will develop the argument that various difficulties that arise in logic and formal scientific modelling point to the necessity of introducing natural time itself into our formal treatment of science, mathematics, and logic. We first discuss some of the difficulties in logic and scientific theory that arise either from a failure to include time in the formal system, or if it is included, from the way it is represented. We then provide examples of the explanatory power that becomes available with the introduction of natural time into formal systems. Finally, the last part of the contribution offers an outline of the way forward.
THE ROLE OF TIME IN MATHEMATICS AND SCIENCE
===========================================
By the end of the nineteenth century the question of the certainty of mathematics itself was being raised. This led initially to efforts to show that mathematics could be formalised as a logical system. Problems in the form of paradox soon emerged in this programme, and those difficulties in turn led to attempts to demonstrate that the programme was at least feasible in principle. However those attempts also failed when it was proven that mathematics must contain unprovable or undecidable propositions. The result was that mathematics came to resemble an archipelago of certainties surrounded by a sea of logically necessary uncertainty. Moreover, with the discovery of the phenomenon of deterministic chaos emerging in the classic three body problem, uncertainties emerged even within the islands of established mathematics. Meanwhile, in physics, while thermodynamics had long seemed to be somehow problematic because of the nature of entropy, at least it produced deterministic laws. During the second half of the 20th century, however, it was shown by Prigogine [@prigogine1981; @prigogine1984] and others [@nicolis1977] that these laws are special cases, and that there are no laws governing most phenomena arising in such systems, because the phenomena of interest arise when the systems are far from thermodynamic equilibrium, whereas the laws describe the equilibrium state. At the same time, in biology, there was a growing realisation that living systems exist in the energetic realm in which the traditional laws of thermodynamics are of only limited use. And with the discovery that DNA is the genetic material of life, much of biology was transformed into what might be termed the informatics of life.
All of these developments have in common that they introduced, irrevocably and in a radical way, uncertainty and unpredictability into mathematics and science. They also have in common that they arose from attempts to ensure certainty and predictability by keeping time, real time, out of the formal explanatory systems. To a large extent, in the practice of everyday science, these developments have been ignored. Science continues to focus on those areas where certainty seems to be attainable. However, many of the most interesting and important problems are ones that grow out of the uncertainties that have been uncovered over the past century: problems like the origin and nature of life, the nature of creative processes, and the origin of novel processes and entities. These tend to be treated discursively rather than scientifically. However, we believe that it is now possible to develop formal treatments of problems like these by including time — real, natural time rather than any formal representation of it — in the explanatory mechanism. This will also mean recognising that a certain degree of uncertainty and unpredictability is inherent in any scientific treatment of these problems. But that would be a strength rather than a failure, because uncertainty and indeterminacy are inherent characteristics of these systems, and so any explanatory mechanism that doesn’t generate an appropriate degree of unpredictability is a misrepresentation. Creativity itself cannot be timeless, but it relies on the timeless laws of physics and chemistry to produce new phenomena that transcend those laws without violating them. In the next four sections we discuss in more detail the role of time in the treatment of paradox, incompleteness, uncertainty, and emergence in order to justify the necessity of including natural time itself rather than formal time in our explanatory systems.
Paradox
-------
By the end of the 19th century the desire to show that mathematics is certain had led to Hilbert’s programme to show that mathematics could be recast as a syntactical system, one in which all operations were strictly “mechanical” and semantics played a minimal role. In this spirit, Frege, Russell and Whitehead reduced several branches of mathematics to logic, thus apparently justifying the belief that the programme was feasible. However, Russell’s work produced a paradox in set theory that ultimately raised doubts about it. Russell’s paradox asks if the set of all sets that do not contain themselves contains itself. Paradoxically, if it doesn’t, then it does; if it does, then it doesn’t. Several solutions have been proposed to rid logic of this paradox, but there is still some debate as to whether any of them is satisfactory [@irving2016].
The paradox seems to emerge because of the time-free nature of logic. If we treat the process described in the analysis as an algorithm and execute it, then the output is an endless oscillation. Does the set contain itself? First it doesn’t, then it does, then it doesn’t, …. There is no contradiction. This oscillation depends on our working in discontinuous time — in this case the time of the computer’s clock. We can in principle make the clock speed arbitrarily fast, but if we could go to the limit of continuous time in operating the computer we would see the paradox re-emerge as something analogous to a quantum superposition: since the superposition endures it doesn’t depend on time, and independent of time, the value is always both ’does’ and ’doesn’t’. However, if we were to observe the state of the system at a particular instant, the superposition would collapse to a definite but arbitrary value — “does” or “doesn’t” — just as Schroedinger’s famous cat is only definitively alive or dead when the box is opened.
Of course a computer must work with a finite clock speed and so the paradox cannot appear. This resolution of the paradox by casting it as an algorithm to be executed in natural time emphasizes the difference between the existence of a set and the process of constructing the set, a difference that echoes Prigogine’s [@prigogine1981], [@prigogine1984 p. 320] distinction between being and becoming. In the algorithmic construction of the set, the size of the set oscillates between $n$ and $n+1$, sizes that correspond to “doesn’t” and “does”. Most well known paradoxes are similar to Russell’s in that they involve self-reference. However, Yablo [@yanofsky2013 pp. 24-25] has demonstrated one that is non-self-referential; but it, too, yields an oscillation (true, false, true, false …) if executed as an algorithm.
Incompleteness
--------------
Whereas Russell discovered a paradox that cast doubt on the possibility of demonstrating that mathematics is a purely syntactical system, both G[ö]{}del and Turing came up with proofs that even to the extent that it *is* syntactical, it is impossible to demonstrate that it is. G[ö]{}del showed that even for a relatively limited deductive system there were true statements that couldn’t be proven true within the system. In order to prove those statements, the system would have to be enlarged in some way, for example with additional axioms. But in this enlarged system there would again appear true statements that were not provable within it. This result highlighted the degree to which mathematics as it existed depended not only on axioms and logical deduction — that is, syntax — but also on the products of what was called mathematical intuition — the source of new concepts and procedures introduced to deal with new problems or problems that were otherwise intractable. In other words, mathematics was progressively extended by constructions based on these intuitions. The intuitions came with semantic content, which G[ö]{}del’s result implicitly suggested could not be eliminated, even in principle.
But there was another problem. G[ö]{}del, like Hilbert, thought of deduction as a mechanical procedure; thus the idea of certainty was closely linked to the idea of a machine operating in a completely predictable manner. Of course at the time the idea was metaphorical; the deductions would actually be carried out by a person (the person was invariably referred to as a computer), but for the results to be reliable, the person would have to follow without error a precise sequence of operations that would yield the desired result; in other words, for each deduction the person would have to execute an algorithm. Thus the algorithm was implicitly part of the logical apparatus used to generate mathematics. It was clear to Turing and Church that a formal understanding of the nature of these algorithms was necessary. To that end Turing formulated what is now known as the Turing machine (at the time, of course — 1937 — it was purely conceptual), and at about the same time Church developed an equivalent approach, the lambda calculus. An algorithm was considered legitimate as a procedure if it could be described as a Turing machine or formulated in the lambda calculus. Each algorithm corresponded to a particular Turing machine. A Turing machine was required to have a finite number of states, and a proof calculated on the machine would need to finish in a finite number of steps. By proving that it could not in general be shown whether or not algorithms would execute in a finite number of steps (the halting problem), Turing demonstrated, in results analogous to G[ö]{}del’s, that some propositions were not provable. Church arrived at the same result using his lambda calculus. The results of G[ö]{}del, Turing, and Church showing that it is impossible to prove that a consistent mathematical system is also complete are complementary to Russell’s discovery of the paradox in set theory, which suggests that an attempt to make a formal system complete will introduce an inconsistency.
Turing presumably imposed the restrictions of finite machine states and finite steps in order to ensure that Turing machines would be close analogues of the recursive equations which were the standard at that time for computing proofs. These restrictions on the Turing machine were necessary conditions for proof, but as his results showed, they did not amount to sufficient conditions. In making these assumptions he effectively restricted the machines to producing time-free results. This was entirely reasonable given that his goal was to formalize a procedure for producing mathematical results which were themselves timeless. Nevertheless he found the restrictions to be somewhat problematic, as did Church and G[ö]{}del, because they meant that a Turing machine could not capture the generation of the new concepts and procedures that flowed from the exercise of mathematical intuition by mathematicians. There was much informal discussion of this issue, including possibilities for circumventing the limitations. Turing suggested that no single system of logic could include all methods of proof, and so a number of systems would be required for comprehensive results. Stating this thought in terms of Turing machines, he suggested a multiple machine theory of mind — mind because he was still thinking of the computer as a person, and the machine as the algorithm that the person would follow mechanically. The multiple machine idea was that Turing machines would be chained, so that different algorithms would be executed sequentially, thus overcoming some of the limitations of the simple Turing machine. How would the sequence be determined? Initially the idea was that it would be specified by the mathematician, but subsequently other methods were proposed, ranging from stochastic choice to a situation in which each machine would choose the subsequent machine. Eventually Turing proposed that learning could provide the basis of choice. He observed that mathematics advances by means of mathematicians exercising mathematical intuition, which they then use to create new mathematics, a process he thought of as learning. He then imagined a machine that could learn by experience, by means of altering its own algorithms. (Copeland and Shagrir, 2015)
Time had been kept out of mathematics by defining it as consisting only of the achieved corpus of results and ignoring the process by which those results were generated. Turing and Church took a first step toward including the process by formalizing the treatment of the algorithms by which proofs were derived. But this did not seem to them (or others) to be sufficient, because it did not capture the deeply creative nature of mathematics as seen in the continual introduction of new concepts and procedures that established whole new areas of mathematics. We might interpret the journey from Turing algorithm to learning as an implicit recognition of the necessity of natural time in mathematics. On the other hand, while Turing had formalized the proof process in terms of a machine which would have to operate in natural time, the machine was defined in such a way that the results that it produced would be time free. Thus in postulating strategies like chained Turing machines to represent learning, he probably assumed that the results would also be time free. However, if mathematics is understood to include the process by which it is created, it will have to involve natural time, even if the result of that creative process is time free. While Turing spoke of learning, a more appropriate term might be creativity, and creativity — the emergence of something new — necessarily involves time.
Uncertainty
-----------
But is mathematics, in the narrower sense of established results, really entirely timeless? Perhaps. But there is at least one small part of it — deterministic chaos — that seems to be trying to break free and take up residence in natural time. The phenomenon was discovered by Henri Poincar[é]{} at the beginning of the twentieth century as he attempted to solve the Newtonian three-body problem. An equation that exhibits deterministic chaotic dynamics can, up to a point, be treated as a timeless structure and its properties investigated using standard mathematical techniques (some invented by Poincar[é]{} for this purpose). It has been shown, for example, that the attractor is a fractal object, and therefore infinitely complicated. As a consequence the solution trajectory cannot be written explicitly. It can, however, be calculated numerically — but only up to a point: since the attractor, as a fractal, is infinitely complex, we can never know exactly where the system is on it, and hence cannot predict the future states of the system.
Take one of the simplest possible cases, the difference equation version of the logistic function: $$\begin{split}
X_{t+1} = r X_t (1 - X_t) \\ \mbox{ with } 0<X_t<1 \mbox{ and }0<r<4
\end{split}$$ Solving the equation for $X$ as a function of $t$ we have $$X^{*} = 1-1/r$$
This solution is stable for $r \le 3$; otherwise it is oscillatory. For $r > 3.57$ approximately, the oscillations are infinitely complex; i.e. the dynamics are chaotic. Since the solution cannot be written explicitly, to see what it looks like we calculate successive values of $X$, while recognizing that these values become increasingly approximate, and soon become entirely arbitrary — that is, they become unpredictable even though the equation generating them is deterministic. This can be dismissed as simply due to the rounding errors that result from the finite precision of the computer, but it is actually a consequence of the interaction of the rounding errors with the fractal nature of the attractor. The rate at which the values evolve from precise to arbitrary is described by the Lyapunov exponent. So while the system may look well defined and timeless from an analytical point of view because the attractor determines the complete behaviour of the system, in fact the attractor is unknowable analytically, and can only be known (and only in a very limited way) by iterating the equation, or by other iterative techniques such as the one developed by Poincar[é]{}. The iterations take place in time, and so it seems that at least some of our knowledge of the behaviour of the equation necessarily involves natural time. Note that a physical process characterised by chaotic dynamics must also be unpredictable in the long run, because as the process “executes” it will be following a fractal attractor, and the physical system that is executing the process, being finite, will be subject to “rounding errors”, which in effect act as a stochastic perturbation. In other words, the resolution that can be achieved by the physical system is less than that of the attractor. This is the case with the three body problem that Poincar[é]{} was working on, and the reason that the planetary trajectories of the solar system are unpredictable at timescales beyond a hundred million years or so. It is also, in Prigogine’s (1997) view, the fundamental reason for the unpredictability of thermodynamic systems at the microscopic level (Laplace’s demon cannot do the math well enough), and also a necessary factor in macroscopic self-organization in that the unpredictability permits symmetry breaking—Prigogine calls this order by fluctuations.
With chaotic systems, then, we lose the promise that mathematics has traditionally provided of a precise, God’s eye view over all time, and thus of certainty and predictability. We are left only with calculations in natural time that give us rapidly decreasing accuracy and hence increasing unpredictability. But from some points of view this is not necessarily a problem. As Turing speculated in his discussion of learning, learning requires trial and error, which would be pointless in a perfectly predictable world, and specifically it requires an element of stochasticity. Many others have made the same observation – that stochasticity seems to be a necessary element in any creative process. In physical systems stochastic perturbations are the basis of the symmetry breaking by which systems become increasingly complex and organized. Chaotic dynamics may thus play a positive role by providing a necessary source of stochasticity in physical, biological, and human systems.
Physics, with the major exception of thermodynamics (together with two very specific cases in particle physics: CPT and a case of a heavy-fermion superconductor [@schemm2014]), is characterized by laws that are “time reversible” in the sense that they remain valid if time runs backward. In other words, time can be treated as a variable, t, and the laws remain valid when $-t$ is substituted for $t$. This is referred to by some (e.g. [@smolin2013]) as spatialized time, because we can travel in both directions in it, as we can in space. Spatialized time is a conceptualization and representation of natural time, whereas natural time is the time in which we and the world exist, independently of any representation of it. Spatializing time is thus a way of eliminating natural time by substituting a model for the real thing. The physics of spatialized time is essentially a timeless physics, since we have access to the entire corpus of physical laws in the same sense that we have access to the entire body of timeless mathematics.
The fact that time can be treated as a variable permits the spectacularly accurate predictions that flow from physical theory: the equations can be solved to show the state of the system as a function of time, and thus the state at any particular time, whether past, present or future. This physics is in a deep sense deterministic. This is true even of quantum physics, where, as Prigogine [@prigogine1997] points out, the wave function evolves deterministically; uncertainty appears only when the wave function collapses as a result of observation. The determinism of spatialized time is the basis of Einstein’s famous remark that “for us convinced physicists, the distinction between past, present and future is an illusion, although a persistent one” (as quoted in [@prigogine1997 p. 165]). His point was that time as we experience it flowing inexorably and irreversibly is an illusion; in relativistic space-time, the reality that underlies our daily illusory existence, we have access to all times. However, Prigogine [@prigogine1997] points out that the space-time of relativity is not necessarily spatialized; that is just the conventional interpretation. In any case, because it is apparently timeless, the physics of quantum theory and relativity is understood to represent our closest approximation to certain knowledge of the world.
Thermodynamics represents a rude exception to this timelessly serene picture. Here time has a direction, and when it is reversed the physics doesn’t work quite the same way. In the forward direction of time, the entropy of an isolated system increases until it reaches the maximum possible value given local constraints. In this sense the system is predictable. But when time is reversed, so that entropy is progressively lowered, the system becomes unpredictable, because, as Prigogine showed, when the entropy of a system is lowered, an increasing number of possible states appears, states that are macroscopically quite distinct but have similar entropy levels. But only one of these can actually exist, and in general we have no certain way of knowing which one that will be. The same phenomenon appears in reversed computation. In other words the reversed time future is characterized by a bifurcation tree of possibilities. Its future is open; it is no longer deterministic or fully predictable, but rather path dependent. This discussion of reversed time futures applies to isolated systems, the ones for which thermodynamic theory was developed. However, we do not live in an isolated system. Our planet is bathed in solar energy, which keeps it far from equilibrium, and we supplement this with increasing amounts of energy from other sources. Thus our open-system world is equivalent to a reversed time, isolated-system world. It is a world of path dependency and open futures of a self-organizing system.
The open futures of these systems is a source of unpredictability or uncertainty, just as is the uncertainty arising from chaotic dynamics, and the two work together. In both of these situations in which unpredictability appears, time continues to be treated as a variable, but in the case of far from equilibrium systems the behaviour is time asymmetric: if we treat decreasing entropy as equivalent to time reversal, physical systems are deterministic in $+t$ but undetermined in $-t$. In the case of chaotic systems, while the *process* may be mathematically deterministic in both $+t$ and $-t$, the *outcome* is undetermined for both directions of time. In both cases, since a mathematical treatment of the phenomenon is of limited use, the preferred approach is computational. This is not just a pragmatic choice. It reflects the poverty of spatialized time compared to the possibilities offered by real time. Whereas physics, with the major exception of thermodynamics, is based on the assumption that spatialized time captures all the characteristics of time that are essential for a scientific understanding of the world, natural time involves no assumptions. It is simply itself. A computer can only implement an algorithm step by step, in natural time. As a consequence, algorithms as they are executed do not depend on any conceptualization or representation of time beyond a working assumption that time is discontinuous or quantized, rather than continuous, an assumption imposed by the computer’s clock.
Emergence
---------
Far from equilibrium, self-organizing systems are the ones that we live in; our planet is essentially a spherical bundle of such systems. The dynamics of plate tectonics is driven by energy generated by radioactive decay in the earth’s core; the complex behaviour of the oceans and atmosphere is driven by the flux of solar energy; and life itself, including human societies, also depends on the continuous input of energy from the sun. Self-organization is a kind of emergence — it is the process by which an organized structure or pattern emerges from the collective dynamics of the individual objects making up the system, whether these are molecules of nitrogen and oxygen organizing themselves into a cyclonic storm or individual people moving together to form an urban settlement. However, as the phrase self-organization suggests, there is no prior specification of the form that is to emerge, and because of the inherent indeterminacy of far-from equilibrium systems, there is always a degree of uncertainty as to exactly what form will appear, as well as where and when it will emerge. These forms are essentially just patterns in the collection of their constituent particles or objects. Unlike their constituent objects, they have no existence as independent entities, and they cannot, simply as patterns, act on their environment — in other words, they have no agency. For this reason the emergence of self-organized systems is called soft or weak emergence.
Strong emergence, on the other hand, refers to the appearance of new objects, or new types of objects, in the system. We can identify three levels of strong emergence:
1. In high energy physics, forces and particles emerge through symmetry breaking. Unlike the increasing energy input required to drive self-organization, this process occurs as free energy in the system decreases and entropy increases.
2. At relatively moderate energy levels, physical systems produce an increasing variety of chemical compounds. These molecules have an independent existence and distinctive properties, like a characteristic colour of solubility in water, that are not simply the sum of the characteristics of their constituent atoms. They also have a kind of passive agency: for example, they can interact with each other chemically to produce new molecules with new properties, like a new colour. Of course they can also interact physically, by means of collisions, to produce weak emergence, for example in the form of a convection cell or a cyclonic storm. But chemical reactions can result in the simultaneous occurrence of both strong and weak emergence, as when reacting molecules and their products generate the macroscopic self-organized spiral patterns of the Belosov-Zhabotinsky reaction. The production of a particular molecule may either use or produce free energy, i.e. it may be either entropy increasing or entropy decreasing.
3. Also at relatively moderate energy levels, living systems emerge through chemical processes, but also through self-assembly of larger structures (cells, organs, organisms). The key characteristic of this level of strong emergence is that the process is initiated and guided by an endogenous model of the system and its relationship with its environment. While in (1) and (2) emergence is determined by the laws of physics, in this case it is determined by the relevant models working together with the laws of physics and chemistry. We include in living systems the meta-systems of life such as ecological, social, political, technological, and economic systems.
It is this third kind of strong emergence, the kind that depends on and is guided by models, that is the focus of our interest. Nevertheless the weak emergence of self-organizing systems remains important in the context of strong emergence, because a process of strong emergence, as in the case of the development of a fertilised egg into a mature multi-cellular individual, often makes use of local self-organization. Furthermore, self-organized structures are often the precursors of individuals with agency, making the transition by means of a process of reification, as when a self-organized settlement is incorporated as a city, a process that endows it with independent agency. In general, while self-organized systems are forced to a state of lower entropy by an exogenously determined flux of energy, living systems create and maintain their organized structures in order to proactively import energy and thus maintain a state of lower entropy. The causal circularity is a characteristic of such systems.
It is this third kind of strong emergence, the kind that depends on and is guided by models, that is the focus of our interest. Nevertheless the weak emergence of self-organizing systems remains important in the context of strong emergence, because a process of strong emergence, as in the case of the development of a fertilised egg into a mature multi-cellular individual, often makes use of local self-organization. Furthermore, self-organized structures are often the precursors of individuals with agency, making the transition by means of a process of reification, as when a self-organized settlement is incorporated as a city, a process that endows it with independent agency. In general, while self-organized systems are forced to a state of lower entropy by an exogenously determined flux of energy, living systems create and maintain their organized structures in order to proactively import energy and thus maintain a state of lower entropy. The causal circularity is a characteristic of such systems.
Model based systems are a qualitatively new type. The models provide context dependent rules of behaviour that supplement the effects of the laws of physics and chemistry. Of course we can always reduce the structures that act as the models to their basic chemical components in order to understand, for example, the chemical structure of DNA or the chemistry of the synapses in a network of neurons, and there are good reasons for doing this: it allows us to understand the underlying physical mechanisms by which the model — and by extension the system of which it is a part — functions. But this reduction to chemistry and physics tells us nothing about how or why the system as a whole exists. These questions can only be answered at the level of the model considered as a model, because it is the model that guides the creation and functioning of the system of which it is a part. In other words, the reductionist programme reveals the syntax of the system, but tells us nothing of the semantics. It is the rules of behaviour of the system as a whole, rules provided by the model, that determine the actions of the system in its environment, and thus, ultimately, its success in terms of reproduction or survival. Part of the semantic content of the model is therefore the teleonomic goal of survival. The teleonomy is the result of the evolutionary process that produced the system. In this sense evolution is the ultimate source of semantics: as Dobzhansky said in the famous title of his paper, “Nothing in biology makes sense except in the light of evolution” [@dobzhansky1973]. The mathematical biologist Robert Rosen [@rosen1991][@rosen2000] speculated that life, rather than being a special case of physics and chemistry, in fact represents a generalization of those fields, in the sense that a scientific explanation of life would reveal new physics and chemistry. In other words the models inherent in living systems could be seen as new physics and chemistry: they introduce semantics as an emergent property of physico-chemical systems.
### Models
An interesting and useful definition of life, due to Rosen [@rosen1991], is that life consists of entities that contain models of themselves, that is, entities that exist and function by virtue of the models they contain. The most basic model is that coded in DNA. But neural systems also contain models, some of them, as we know, very elaborate. And of course some models are stored in external media such as books and computers. These three loci of models correspond to the three worlds of Karl Popper: World 1 is the world of physical existence, World 2 corresponds to mental phenomena or ideas; and World 3 consists of the externally stored and manipulated representations of the ideas. Worlds 2 and 3 are not generally considered by scientists to be constituents of the world that science seeks to explain. However, as Popper points out, they are in fact part of it, and exert causal powers on World 1 [@popper1982]. The implication is that a scientific understanding of biological and social phenomena requires not just an analysis at the level of physical and chemical causation, but also consideration of the causal role of meaning, or more specifically, meaning as embodied in models. Thus semantics re-enters the picture in a fundamental way.
A model that is a part of a living system must be a formal structure with semantics, not just syntax. It can function as a model only by virtue of its semantic content, since in order to be a model it must represent another system, a system of which, in the case of living organisms, it is itself usually a part. The modelled system thus provides the model with its semantic content. As Rosen points out, this contradicts the orthodox position of reductionist science (and in particular of Newtonian particle physics) that “every material behaviour \[can\] be...reduced to purely syntactical sequences of configurations in an underlying system of particles” [@rosen2000 p.68; see also p.46ff].
Non-living systems lack semantics; they might thus be characterised as identity models of themselves, or *zero-order models*. Models associated with living organisms (e.g. DNA or an idea of self) would then be *first order models*. And some scientific models, those that are models of models (e.g. a mathematical model of DNA), would be *second order models*. This chapter is concerned with first order models.\
We propose the following definition:\
***a*** is a first order model of ***A***, i.e. ***a*** is a *functional representation* of ***A*** (**[a ~~r~~ A]{}**), if
1. ***a*** is a structure (not just a collection) of appropriate elements in some medium, whether chemical (e.g. a DNA molecule composed of amino acids), cellular (e.g. a synaptic structure in a network of neurons), or symbolic (e.g. a program composed of legitimate statements in some programming language).
2. The structure ***a*** can act as an algorithm when executed on some suitable machine ***M***, where ***M*** may be either separate from ***A*** (e.g. a computer running a model of an economic system), or some part of ***A*** (e.g. a bacterial cell running the behavioural model coded in its DNA);
3. The output of the algorithm corresponds to or consists of some characteristics of ***A***. Specifically:
\(a) Given a suitable environment, ***a*** running on ***M*** can *create* a new instance of ***A*** (e.g. in the environment provided by a warm egg, the DNA being run by the egg cell containing the DNA can create a new instance of the kind of organism that produced the egg).
\(b) ***a*** can *guide the behaviour* of ***A*** in response to certain changes in the state of the environment (e.g. on the arrival of night, go to your nest; if inflation is greater than 3 percent, raise the interest rate).
4. 3(a) and 3(b) are *evolved* (or in human World 3 systems, *designed*) capabilities or functions that in general serve to maximise the chance of survival of ***A***.
5. If ***A*** is a living organism, **[ ~~r~~]{}** is an emergent property of the underlying physical and chemical systems.
First (and higher) order models are essentially predictive. Although the output of ***a*** when executed on ***M*** is literally a response to a current condition ***c*** which represents input to ***a***, because of the evolutionary history of ***a*** that brought it into existence, the behaviour of ***A*** in response to ***a*** is actually a response to a future, potentially detrimental, condition ***c***’ *predicted* by ***a***; in other words, on the basis of the current condition ***c***, ***a*** predicts that ***c***’ will occur, and as a consequence produces a response in ***A*** intended to prevent the occurrence of ***c***’ or mitigate its impact. Thus ***a*** acts as a predictive algorithm, and guides the behaviour of ***A*** on the basis of its predictions.
The model ***a*** is thus rich in time. It involves both past time (the time in which it evolved) and future time (the time of its prediction), as well as, during execution, the natural time of the present. This reminds us of Bergson’s [@bergson1911 p.20] observation regarding natural (“concrete”) time: “the whole of the past goes into the making of the living being’s present moment.” Only if we know already about evolution as a process can we see the three times present in ***a***. Only by virtue of being such a system ourselves do we have the ability to perceive its purpose.
In this three-time aspect, ***a*** as it represents ***A*** is fundamentally different from a purely chemical or physical phenomenon in the conventional sense. It has semantic content which would be eliminated by any possible reduction to the purely mechanical causation of chemical and physical events. From a reductionist standpoint we would see only chemical reactions, nothing of representation, purpose, past, or anticipation. On the other hand, since ***a*** does actually have this semantic content, that content must emerge in the chemical system itself. It does so by virtue of the relationship between the part of the system that constitutes ***a*** and the larger system that is being modelled, just as a molecular property like solubility in water emerges from interactions among the atoms making up the molecule. In this sense Rosen was correct that life represents a radical extension of chemistry and physics: at no point do we require a vital principle or a soul to breathe semantics, or even life, into chemistry.
In the specific case of DNA, as a molecule it is essentially fixed from the point of view of the organism: over that timescale, as a molecule, it is timeless. But natural time appears as the organism develops following conception, when various genes are turned on or off, and this behaviour continues in the fully developed organism as its interactions with the environment are guided by various contingently activated combinations of genes. In that sense DNA acts as a model that changes as a result of its interactions with the modelled system of which it is a part. This is reminiscent of the chained Turing machines proposed by Turing to permit creativity. Neural models, in contrast, lack a comprehensive fixed structure analogous to that of DNA; they are open ended and develop or change continually as a result of interactions with their host organism and the environment. But in both cases, as a computational system, life is essentially a case of open ended computation.
### Information
The model of a system represents information, and its role in the functioning of the system depends on its being treated as information by the system. Note that this is information in the sense of semantics, or meaningful information, rather than Shannon information, which is semantics-free and represents information capacity or potential information. In other words, we could say that while semantics represents the content or meaning of information, Shannon information represents its quantity, and syntax represents its structure. Shannon information is maximised when a system is in its maximum entropy state. In the case of a self-organized system, the macro-scale pattern constrains the behaviour of the constituent particles so that the system’s entropy and hence its Shannon information is less than it would be if its particles were unconstrained by the self-organized structures.
We do not know of a measure of semantic information; it seems unlikely that such a measure could even be defined. Nevertheless, it seems that the model is the means by which semantic content emerges from syntax. We speculate that it is ultimately the teleonomic nature of living systems that populates the vacant lands of Shannon information with the semantics of meaning-laden information. A model embedded in a living system does not simply represent some aspect of another system; it does so purposefully. In living systems, the function of the model is to guide the behaviour of the system of which it is a part, and it does this by predicting future states of both the system and its environment. System behaviour thus depends to some degree on the anticipated future state of the system and its environment — i.e. the behaviour is goal directed. In contrast, in the case of traditional feedback systems, behaviour depends on the current state of the system and its environment. We note the apparent irony that life, a system that depends for its origin and evolution on uncertainty, nevertheless depends for its survival on an ability to predict future states. In fact, it requires a balance of predictability and unpredictability. In Langton’s [@Langton1990] terms, it exists on the boundary between order and chaos.
### Agency
First order models emerged with life in an evolutionary process, one in which the model both depends on and facilitates the persistence of the system of which it is a part. The model thus necessarily has a teleonomic quality — its purpose is ultimately to enhance the likelihood of its own survival and that of the host system that implements it. To this end, the model endows its host system with agency — i.e. it transforms the system into an agent that can act independently. The relationship between model and evolutionary process, the basis of strong emergence, seems fundamental: each seems to be necessary for the other. This is in a sense the basic assumption of the theory of biological evolution. In contrast, a self-organized system, the result of weak emergence, does not act independently to ensure its own persistence. Living systems, by virtue of their agency, act to maintain themselves in a state of low entropy.
Creative Algorithms
-------------------
The models that guide the generation and behaviour of living systems are necessarily self-referential, since they are models of a system of which they are an essential part. This means that they cannot be represented purely as mathematical structures. However, if the mathematical structures are appropriately embedded in algorithms being executed in natural time, the problem disappears. Nevertheless, the definition of algorithm remains crucial. Rosen, with deep roots in mathematics, was never quite able to resolve the problems arising from self-reference because he worked with Turing’s definition of algorithm; this is clear when he claims, repeatedly, that life is not algorithmic. But as we have noted, the Turing machine was defined in such a way as to produce only results that are consistent with time-free mathematics. To generate that mathematics, the Turing machine must be supplemented by a source of learning or creativity. Learning and creativity are essential characteristics of living systems, as is the appearance of new entities with agency, which learning and creativity make possible. Consequently, a formal understanding of life must include a formal treatment of learning, creativity and strong emergence. That requires algorithms that transcend Turing’s definition. It requires algorithms that are able to model their own behaviour and alter themselves on the basis of their models of themselves. Using a computer operating in natural time to execute only Turing algorithms is like insisting on using three dimensional space to do only two dimensional geometry: it is a colossal waste of capacity as well as a refusal to consider the unfolding world of possibilities that emerge in natural time.
FURTHER EXPLORATIONS ON THE ROLE OF TIME IN SCIENCE
===================================================
We have gotten so used to the concept of creativity and completely new solutions to problems, or to inventions that make our life easier and are introduced the first time, that we tend to overlook the principle aspect of creating new things.
In the daily processes of synthetic chemistry, for example, new molecules are generated every day by combining existing molecules into new combinations. Given the enormous extent of the combinatorial space of chemistry, we have to presume that some of those are created the very first time. If some of these compounds are stable and created today in the Universe for the first time – note we speak of actual realization of material compounds, as opposed to the mere possibility of their existence being “discovered” – they come with a time stamp of today. Thus, every material substance or object has in some way attached to it a time stamp of when it or its earlier copies first appeared in the Universe. Time, therefore, is of absolute importance to everything that exists and is able to characterize it in some way.
Can we make use of that in the Sciences? Here, we want to look at the two sciences that provide modeling tools for others to use in their effort to model the material universe, mathematics and computer science.
Mathematics
-----------
We have already mentioned that mathematics uses the concept of time (if at all) in a spatial sense. This means, time can be considered as part of a space that can be traversed in all directions. Notably, it can be traversed backward in time! But mathematics is actually mostly concerned about the unchanging features of the objects and transformations it has conceptualized. Thus, it glosses over, or even ignores changes in features, as they could prevent truth from getting established. For instance, a mathematical proof is a set of transformations of a statement into the values “true” or “false”, values that are unchanging and not dynamic. This reliability is its strength. Once a statement is established to be true, it is accepted into the canon of mathematically proven statements, and can serve as an intermediate for other proof transformations.
But what about a mathematics of time? How would such a mathematics look like? We don’t know yet, perhaps because the notion of time is something many mathematicians look at with suspicion, and rather than asking how such a mathematics would look like, they ask themselves whether time exists at all and how they can prove that it does not exist — except as an illusion in our conciousness [@barbour2001]. Although Science has always worked like that — ignoring what it cannot explain and focusing on phenomena it can model and explain — we have reached a point now where we simply cannot ignore the nature of time any more as a concept that is key to our modeling of natural systems.
So let’s offer another speculation here. We said before that every object in the universe carries a property with it we can characterize as a time stamp, stating when it first appeared. This is one of its unalienable properties, whether we want to consider it or not. So how about imagining that every mathematical object and all the statements and transformations in mathematics would come with the feature of a time stamp? In other words, besides its other properties, an object, statement or transformation would carry a new property, the time when it was first created. This would help sort out some of the problems when trying to include the creation of mathematics into mathematics itself. It would actually give us a way of characterizing how mathematics is created by mathematicians. The rule would be that new objects, statements and transformations can only make use of what is already in existence at the time of their own creation.
Once we have achieved such a description, can we make a model of the process? Perhaps one of the natural things to ask is whether it would be possible to at least guess which objects, statements or transformations could be created next? The situation is a reminder of the “adjacent possible” of Kaufmann who proposed that ecological systems inhabit a state space that is constantly expanding through accessing “adjacent” states that increase its dimensionality. What this includes is a notion that only what interacts with the existing (which we can call “the adjacent”) could be realized next. Everything else would be a [*creatio ex nihilo*]{} and likely never be realized.
Here is an example: Suppose we have a set of differential rate equations that describe a system at the current state. For simplicity, let’s assume that all the variables of the system carry a time stamp of this moment. Suppose now that we want to introduce a new variable, for another quantity that develops according to a new differential rate equation. Would it make sense to do that even without any coupling of this new variable to the existing system? We don’t think it would. In fact, the very nature of our wish to introduce this variable has to do with its interaction with the system as it is currently described. Thus, introducing a variable that can describe the adjacent possible has at least to have [*some*]{} interaction with the current system.
Dynamic set theory [@liu1993] is an example of how this could work. Dynamic set theory was inspired by the need to deal with sets of changing elements in a software simulation. Mathematically, normal sets are static, in that membership in a set does not change over time. But dynamic sets allow just that: Sets can be defined over time intervals $T$, and might contain certain elements at certain times only. For example, if you have a set of elements $$X = \{a_1, a_2, a_3, b_1, b_2, b_3, c_1, c_2, c_3 \}$$ we can assign specific dynamic sets to a time interval $T$ as follows $$A^T = \{ (t_1, \{ a_1, b_1, c_1 \}), (t_2, \{a_2, b_1 \}),(t_T, \emptyset) \}$$ and $$B^T = \{ (t_1, \{ a_1, a_2 \}), (t_3, \{a_3, c_1, c_3 \}),(t_T, \emptyset) \}$$ We can then manipulate these sets using set operations, for instance: $$A^T \cap B^T = \{ (t_1, \{ a_1 \}),(t_T, \emptyset) \}$$ or $$\begin{split}
A^T \cup B^T = \{ (t_1, \{ a_1, a_2, b_1, c_1 \}), (t_2, \{a_2, b_1 \}), \\ (t_3, \{a_3, c_1, c_3 \}),
(t_T, \emptyset) \}
\end{split}$$ We can see here that each of these elements is tagged with a particular time at which they are part of the dynamic set, and can take part in set operations for that particular moment.[^3] A generalization of set theory is possible to this case. Our hope is that — ultimately — mathematics will be able to access the constructive, intuitional aspects of its own creation. Once we have assigned the additional property of time/age to mathematical objects, perhaps its generative process can be modeled.
Another example of mathematical attempts at capturing time in mathematics is real-time process algebra [@wang2002]. The idea of this approach is to try to describe formally what a computational system is able to do, in particular its dynamic behavior. This project of formalization was generalized under the heading of “denotational mathematics” [@wang2014; @wang2015].
These are all interesting attempts to capture the effect of time within the framework of Mathematics, but they fall short of the goal, because they are descriptive in nature, i.e. they are not generative and able to create novel structures, processes or variables themselves.
Computer Science
----------------
Computers allow the execution of mathematical models operationalized as algorithms. But as we have seen from the discussion in this chapter, mathematics currently deals with spatialized time, not real, natural time. Thus, if we were to only aim at simulating mathematical models, we do not need natural time. This is indeed Turing’s definition of an algorithm, restricted exactly in the way required to make sure that it cannot do anything that requires natural time, so that the computer executing a Turing algorithm is only doing what, in principle, timeless mathematics can do. Here, instead, we aim for algorithms to execute on machines that need to go beyond traditional mathematical models.
We need to provide operations within our algorithms that allow for modification of models. Let us briefly consider how variables (potential observables of the behavior of a \[simulated\] model) are realized in a computer: They are handled using the address of their memory location. Thus if we allow memory address manipulations in our algorithms, like allocating memory for new variables, or garbage collection (for variables/memory locations that have fallen out of use), we should be able to modify at least certain aspects of a model (the variables). Since the address space of a computer is limited, memory locations can be described by integer numbers in a certain range, so we are able to modify them during execution.
Of course, variables are but one class of entities that need to be modified from within the code. Reflective computer languages allow precisely this type of manipulation [@smith1982]. Reflection describes the ability of a computer language to modify its own structure and behavior. Mostly interpreted languages have been used for reflection, yet more modern approaches like [*SELF*]{} offer compiling capabilities, based on an object-oriented model. As Sobel and Friedman write: “Intuitively, reflective computational systems allow computations to observe and modify properties of their own behavior, especially properties that are typically observed only from some external, meta-level viewpoint” [@sobel1996]. What seems to make [*SELF*]{} particularly suitable is its ability to manipulate methods and variables in the same framework. In fact, there is no difference in [*SELF*]{} between them. Object classes are not based on an abstract collection of properties and their inheritance in instantiation, but on prototype objects, object copy and variation. We believe that [*SELF*]{} allows an easier implementation of an evolutionary system than other object-oriented languages.
Susan Stepney’s work [@stepney2011] in the context of the CoSMoS project provides a good discussion of the potential of reflective languages to allow to capture emergent phenomena through self-modification. In order for a self-modifying system not to sink into a chaotic mess, though, we probably shall need again to time stamp the generation of objects.
However, the open-ended power of those systems might only come into its own when one of the key aspects of natural time is respected as well — the fact that one cannot exit natural time. This calls for systems that are not terminated. Natural open-ended processes like scientific inquiry or economic activity or biological evolution [*do not allow*]{} termination and restart. While objects in those systems might have a limited lifetime, entire systems are not “rebooted”. Instead, new objects have to be created and integrated into the dynamics of the existing systems.
We return here to a theme already mentioned with Turing machines: The traditional idea of an algorithm, while having to make use of natural time during its execution as a step-by-step process, attempts to ignore time by requiring the algorithm to halt. Traditional algorithms are thus [*constructed*]{} to halt for their answer to be considered definitive. This, in fact, makes them closed system approaches to computation, as opposed to streaming processes, that analyze data continuously and provide transient answers at any time [@dodig2011]. We might want to ask: What are the requirements for systems that do not end, i.e. do not exit natural time? [@banzhaf2016]
OTHER SCIENCES
--------------
In this contribution we do not have enough space to discuss in detail how natural phenomena as encountered in simple and complex systems can inform the corresponding sciences — which attempt to model those phenomena (physics, chemistry, biology, ecology and economy) — about natural time. But we believe it is important to emphasize that a clear distinction should be made between our modeling attempts and the actual phenomena underlying them. In the past, there were times when the model and the reality were conceptually not separated. For instance, the universe was considered like clockwork, or later as a steam engine, and even later as a giant computer. All of these attempts to understand the universe mistook the underlying system for its metaphor.\
CONCLUSION
==========
Our argument here is not that the Universe is a giant computer [@fredkin2003], preferably running an irreducible computation [@wolfram2002]. This would interchange the actual system with the model of it. Rather, our argument is that time is so fundamental to the Universe that we need tools (computers) and formalisms (algorithms) that rely on natural time to be able to faithfully model its phenomena. We believe that there are many phenomena in the natural and artificially made world making use of novelty, innovation, emergence, or creativity, which have resisted modeling attempts with current techniques. We think those phenomena are worth the effort to change our concepts in order to accommodate them into our world view and allow us to develop models. As hard as it might be to do that, what would Science be without taking stock of what is out there in the world and attempting to incorporate it in our modelling efforts?
This essay was written on the occasion of the Festschrift for Susan Stepney’s 60th birthday. It is dedicated to Susan, whose work has been so inspiring and deep.
[^1]: To be published as chapter in ”From Astrophysics to Unconventional Computing”, Springer 2019
[^2]: While time as a parameter has been used in mathematical tools, this amounts to merely a representation of time.
[^3]: Note that we have skirted the issue of how to measure time, and how to precisely determine a particular moment and its synchronous counterparts in other regions of the Universe. For now, we’d stick to classical time and assume a naive ability to measure it precisely.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The entanglement dynamics of a pair of donor-based charge qubits is obtained in analytical form. The disentanglement is induced by off resonant scattering of acoustical phonons in the semiconductor host. According to our results a rather unusual recovery of entanglement occurs that depends on the geometrical configuration of the qubits. In addition, for large times a non-vanishing stationary entanglement is predicted. For the cases of one and two initial excitations a simple kinetic interpretation allows for an adequate analysis of the observed dynamics. Our results also reveal a direct relation between the disentanglement rate and the inter-donor decoherence rates.'
author:
- 'F. Lastra'
- 'S.A. Reyes'
- 'S. Wallentowitz'
date: 'August 11, 2011'
title: 'Phonon-induced entanglement dynamics of two donor-based charge quantum bits'
---
Introduction
============
Quantum information processing promises highly efficient solutions to cryptographic problems and exhaustive database search, that outperform the best known algorithms with classical computers [@niel]. Such quantum algorithms rely on the capability to process correlations among quantum subsystems. The quantum part of these correlations is denoted as entanglement. Most commonly, the subsystems are identified as two-level systems, representing a quantum counterpart of classical bits, called quantum bits (qubits). Among the well known applications of quantum information are quantum teleportation [@ben], superdense coding [@ben1], and secure distribution of cryptographic keys [@ekr].
Entanglement, being the key ingredient of such applications, at the same time is highly fragile and can be easily deteriorated during state preparation, addressing and control of individual qubits, and final readout. Hence, it is necessary to encode quantum information in physical systems, where entanglement is protected or can be preserved in a robust way from ambient effects. Prominent examples of physical implementations are trapped ions [@lei], nuclear magnetic resonance [@van], atoms in cavities [@rai], quantum dots [@han], semiconductor impurities [@kan], superconducting qubits [@dic], and impurities in diamond [@gae].
Among the solid-state implementations [@pro; @ami; @chi] impurities embedded in a semiconductor substrate offer the advantage of comparably easy scaling and production due to highly developed fabrication techniques. The encoding of qubits in the charge degrees of freedom of pairs of donor sites allows for the addressing of individual qubits by metallic gates [@hol; @bel]. Tunneling of the electron between donor sites together with Coulomb repulsion of electrons bound in neighboring qubits have been shown to allow for the realization of a CNOT gate [@hol; @tsu; @ste]. Such a gate is the elemental building block of any quantum algorithm. Typical coherence times of such systems are limited by phonon scattering to the order of $1\pico\second$ [@zhao; @non-m; @las2]. Therefore, the state preparation process must be faster and one may expect that entanglement is rapidly lost on this time scale.
In this work we show that to a large extent the entanglement survives beyond this time scale and therefore suffers further degradation only from other sources of decoherence, i.e. charge fluctuations on the control electrodes. The disentanglement dynamics is obtained in analytical form showing non-Markovian features, similar to the decoherence dynamics of a single qubit [@eck; @non-m; @las2]. Furthermore, it is shown that the disentanglement rate is directly related to inter-donor decoherence rates for the cases of one and two initial excitations. The structure of this rate can be explained by a simple kinetic interpretation that allows for the determination of the disentanglement rate from the geometry of the constituent donor sites.
The paper is organized as follows: In Sec. \[sec:Electronic-Dynamics\] the dynamics of a general $N$ qubit system subject to off-resonant scattering of acoustical phonons is derived. Using these results the dynamics of entanglement between two qubits is analytically obtained in Sec. III for the cases of one and two excitations. Finally, in Sec. IV we present a summary and conclusions.
Phonon-induced dephasing dynamics of qubits\[sec:Electronic-Dynamics\]
======================================================================
In this chapter we deduce the dynamics of the reduced density operator of the qubit system, which is induced by off-resonant scattering of acoustical phonons within the semiconductor material. It is assumed here that $2N$ donor sites are present that conform $N$ qubits, see Fig. \[fig:qubits\]. The results will later be specialized to the case of $N=2$ qubits. The dynamics of a single qubit conformed by two donor sites has been derived recently by us in Ref. [@non-m]. The present work goes beyond the case of two donors, which implies a more complex dynamics together with the possibility of studying the entanglement between pairs of qubits.
![\[fig:qubits\]Geometrical setup of $N$ qubits formed by $2N$ donor sites in a semiconductor material. Blue and red donor sites correspond to qubit states $m_{b}=-\frac{1}{2}$ and $m_{b}=+\frac{1}{2}$, respectively, where $b=1,\ldots,N$ denotes the qubit. Each qubit $b$ has an inter-site distance ${\mathbf}{d}_{b}$ pointing from site $m_{b}=-\frac{1}{2}$ to $m_{b}=+\frac{1}{2}$, and a position vector ${\mathbf}{r}_{b}$ pointing towards the center of the qubit. ](figure1){width="1\columnwidth"}
We consider the situation where a single electron is confined to each pair of donors, conforming the $N$ qubits being centered at the positions ${\mathbf}{r}_{b}$, where $b=1,\ldots,N$ labels the qubit under consideration. In addition, we assume that the distances between these qubits are much larger than the inter-donor vectors ${\mathbf}{d}_{b}$ within the qubits, i.e. $|{\mathbf}{r}_{b}-{\mathbf}{r}_{b^{\prime}}|\gg|{\mathbf}{d}_{b^{\prime\prime}}|$. Under these circumstances the tunneling of electrons between different qubits can be safely neglected. Considering the interaction of the qubit charges with acoustical phonons in the semiconductor substrate at room temperature or below, off-resonant phonon scattering is the main source of decoherence in the electron dynamics [@non-m]. This type of electron-phonon interaction is accurately described by the spin-boson Hamiltonian
$$\begin{aligned}
\hat{H} & = & \hbar\sum_{b}\left(\omega_{b}\hat{S}_{b,z}+\Delta_{b}\hat{S}_{b,x}\right)+\sum_{{\mathbf}{k}}\hbar\omega_{k}\hat{a}_{{\mathbf}{k}}^{\dagger}\hat{a}_{{\mathbf}{k}}\nonumber \\
& & +\hbar\sum_{b}\hat{S}_{b,z}\sum_{{\mathbf}{k}}\left(g_{b,{\mathbf}{k}}\hat{a}_{{\mathbf}{k}}^{\dagger}+g_{b,{\mathbf}{k}}^{\ast}\hat{a}_{{\mathbf}{k}}\right),\label{eq:hamiltonian}\end{aligned}$$
where the individual interaction rate of the $m_{b}$th donor site ($m_{b}=\pm1/2$) within qubit $b$ reads $$g_{b,{\mathbf}{k}}=\frac{D}{\hbar s}\sqrt{\frac{2\hbar\omega_{k}}{M_{0}}}\sum_{m_{b}=\pm1/2}\frac{m_{b}e^{-i{\mathbf}{k}\cdot({\mathbf}{r}_{b}+m_{b}{\mathbf}{d}_{b})}}{\left[1+\left(\frac{ka_{b,m_{b}}}{2}\right)^{2}\right]^{2}}.$$ Here $\hat{a}_{{\mathbf}{k}}$ are the annihilation operators of longitudinal acoustical phonon with dispersion relation $\omega_{k}=sk$, $s$ being the sound speed. Moreover, $M_{0}$ and $D$ are the mass within a unit cell of the semiconductor and the deformation constant, respectively. Each donor site is described by a $s$-wave ground-state with Bohr radius $a_{b,m_{b}}$ and the electronic transition of each qubit $b$ is generated by the pseudo spin-$\frac{1}{2}$ operator $\hat{{\mathbf}{S}}_{b}$, where $$\hat{S}_{b,z}=\frac{1}{2}\left(\biggl|{\mathbf}{r}_{b}+\frac{{\mathbf}{d}_{b}}{2}\biggr\rangle\biggl\langle{\mathbf}{r}_{b}+\frac{{\mathbf}{d}_{b}}{2}\biggr|-\biggl|{\mathbf}{r}_{b}-\frac{{\mathbf}{d}_{b}}{2}\biggr\rangle\biggl\langle{\mathbf}{r}_{b}-\frac{{\mathbf}{d}_{b}}{2}\biggr|\right),$$ with $|{\mathbf}{r}_{b}\pm{\mathbf}{d}_{b}/2\rangle$ being the states with the electron being localized at the corresponding donor site. In what follows we assume that during the free evolution of the system, tunneling is inhibited by either an applied potential barrier between the donor sites, or due to a strong bias between the qubit levels, $|\omega_{b}|\gg|\Delta_{b}|$, which can be provided for by the application of a DC electric field.
Following the same steps as in Ref. [@non-m], the Hamiltonian Eq.(\[eq:hamiltonian\]) can be diagonalized, to obtain the complete set of eigenstates as displaced number states $$|E_{\{m_{b}\},\{N_{{\mathbf}{k}}\}}\rangle=|\{m_{b}\}\rangle\otimes\hat{D}^{\dagger}\left(\left\{ \alpha_{\{m_{b}\},{\mathbf}{k}}\right\} \right)|\{N_{{\mathbf}{k}}\}\rangle,\label{eq:eigenstates}$$ with eigenenergies $$E_{\{m_{b}\},\{N_{{\mathbf}{k}}\}}=\hbar\sum_{b}\omega_{b}m_{b}+\sum_{{\mathbf}{k}}\hbar\omega_{k}N_{{\mathbf}{k}}.\label{eq:eigenenergies}$$ Here the multi-mode displacement operator reads $$\hat{D}\left(\left\{ \alpha_{{\mathbf}{k}}\right\} \right)=\exp\left[\sum_{{\mathbf}{k}}\left(\alpha_{{\mathbf}{k}}\hat{a}_{{\mathbf}{k}}^{\dagger}-\alpha_{{\mathbf}{k}}^{\ast}\hat{a}_{{\mathbf}{k}}\right)\right],$$ and $|\{N_{{\mathbf}{k}}\}\rangle=\prod_{{\mathbf}{k}}|N_{{\mathbf}{k}}\rangle$ are multi-mode number states of the acoustic phonons and the displacement amplitude reads $$\alpha_{\{m_{b}\},{\mathbf}{k}}=\sum_{b}m_{b}\alpha_{b,{\mathbf}{k}},$$ where $\alpha_{b,{\mathbf}{k}}=g_{b,{\mathbf}{k}}/\omega_{k}$. Furthermore, the quantum state of the qubits is encoded in the register state $$|\{m_{b}\}\rangle=|m_{1}\rangle\otimes|m_{2}\rangle\otimes\cdots\otimes|m_{N}\rangle,$$ where each of the qubits can be in states $|m_{b}=\pm\frac{1}{2}\rangle$ ($b=1,\ldots,N$).
Given the eigenstates (\[eq:eigenstates\]) and eigenenergies (\[eq:eigenenergies\]), the general solution of the reduced density operator of the qubits, i.e. traced over the phonons, reads $$\begin{aligned}
\hat{\varrho}(t) & = & \sum_{\{m_{b}\},\{s_{b}\}}|\{m_{b}\}\rangle\langle\{s_{b}\}|\sum_{\{N_{{\mathbf}{k}}\}}\sum_{\{N_{{\mathbf}{k}}^{\prime}\}}\varrho_{\{m_{b}\},\{N_{\lambda}\};\{s_{b}\},\{N_{\lambda}^{\prime}\}}\nonumber \\
& & \quad\times\langle\{N_{{\mathbf}{k}}^{\prime}\}|\hat{D}\left(\left\{ \alpha_{\{s_{b}\},{\mathbf}{k}}\right\} \right)\hat{D}^{\dagger}\left(\left\{ \alpha_{\{m_{b}\},{\mathbf}{k}}\right\} \right)|\{N_{{\mathbf}{k}}\}\rangle\nonumber \\
& & \quad\times\exp\left[-\frac{it}{\hbar}\left(E_{\{m_{b}\},\{N_{{\mathbf}{k}}\}}-E_{\{s_{b}\},\{N_{{\mathbf}{k}}^{\prime}\}}\right)\right].\label{eq:rho-reduced}\end{aligned}$$ Here the matrix elements of the initial density operator of the complete electron-phonon system in the basis of the energy eigenstates (\[eq:eigenstates\]) are $$\varrho_{\{m_{b}\},\{N_{{\mathbf}{k}}\};\{s_{b}\},\{N_{{\mathbf}{k}}^{\prime}\}}=\langle E_{\{m_{b}\},\{N_{{\mathbf}{k}}\}}|\hat{\varrho}(0)|E_{\{s_{b}\},\{N_{{\mathbf}{k}}^{\prime}\}}\rangle.\label{eq:initial-complete}$$ These density matrix elements are determined by the state-preparation process that is utilized to set up the initial entanglement between the qubits. A generic state-preparation process can be described as follows:
We assume that initially the system is at low enough temperature, $k_{{\rm B}}T\ll\hbar\omega_{b}$, such that it has relaxed completely into a state where all the qubits are in their lowest energy states,$m_{b}=-1/2$ ($b=1,\ldots,N$), and coexist in thermal equilibrium with the phonons in the substrate. Starting from this state the system undergoes a state preparation process in which the qubits can be coherently transferred into a superposition of ground and excited states, without affecting the quantum state of the phonons. The probability amplitudes to transfer from the initial qubit state $\{-\frac{1}{2},-\frac{1}{2},\ldots,-\frac{1}{2}\}$ to the qubit states $\{m_{b}\}$, shall be denoted by $\psi_{\{m_{b}\}}$. The corresponding transition can be generated by the application of the operator $$\hat{S}_{+,\{m_{b}\}}=\Pi_{b}\left[\delta_{m_{b},\frac{1}{2}}\hat{S}_{+,b}+\left(1-\delta_{m_{b},\frac{1}{2}}\right)\hat{I}_{b}\right],$$ where $\hat{I}_{b}$ is the identity operator in the Hilbert space of qubit $b$.Thus, the matrix elements of the prepared initial state (\[eq:initial-complete\]) become $$\begin{aligned}
\varrho_{\{m_{b}\},\{N_{{\mathbf}{k}}\};\{s_{b}\},\{N_{{\mathbf}{k}}^{\prime}\}}=\qquad\qquad\qquad\qquad\qquad\qquad\qquad\label{eq:psi-e-def-1-1}\\
\psi_{\{m_{b}\}}\psi_{\{s_{b}\}}^{\ast}\langle E_{\{m_{b}\},\{N_{{\mathbf}{k}}\}}|\hat{S}_{+,\{m_{b}\}}\hat{\varrho}_{T}\hat{S}_{+,\{s_{b}\}}^{\dagger}|E_{\{s_{b}\},\{N_{{\mathbf}{k}}^{\prime}\}}\rangle,\nonumber \end{aligned}$$ where $$\hat{\varrho}_{T}=\sum_{\{N_{{\mathbf}{k}}\}}P_{\{N_{{\mathbf}{k}}\}}|E_{\{-1/2,-1/2,...\},\{N_{{\mathbf}{k}}\}}\rangle\langle E_{\{-1/2,-1/2,...\},\{N_{{\mathbf}{k}}\}}|\label{eq:thermal-state}$$ is the initial thermal state with the phonon statistics $$P_{\{N_{{\mathbf}{k}}\}}=Z^{-1}\exp\left(-\sum_{{\mathbf}{k}}\beta_{k}N_{{\mathbf}{k}}\right),\label{eq:thermal-statistics}$$ with $Z$ satisfying $\sum_{\{N_{{\mathbf}{k}}\}}P_{\{N_{{\mathbf}{k}}\}}=1$, and $\beta_{k}=\hbar\omega_{k}/(k_{{\rm B}}T)$.
We note that the corresponding prepared initial reduced density operator of the qubits is pure, i.e. is of the form $|\psi(0)\rangle\langle\psi(0)|$ with the qubit state being the sought superposition $$|\psi(0)\rangle=\sum_{\{m_{b}\}}\psi_{\{m_{b}\}}|\{m_{b}\}\rangle.$$ Thus, the presence of the phonons does not prevent a coherent preparation of the initial qubit state. However, the initial state preparation takes the electron-phonon system out of thermal equilibrium so that the decoherence of the quantum state of the qubits cannot be treated by the usual method of spectral functions [@leggett].
The generic state preparation, as described above, can be implemented for example by switching the gate voltages so that the on-site energies within a qubit cross each otherinducing Landau-Zener transitions. At the end of the process, the system will be in a coherent superposition of all possible states of the system depending on how fast its two-level components where driven across the level crossing. Alternatively, it may be implemented by time controlled tunneling and employing the Coulomb repulsion between neighboring qubits to generate entangled qubit states, as proposed for implementing CNOT gates [@hol; @tsu; @ste]. Furthermore, it could also be implemented by $\tera\hertz$ Raman transitions between the donor sites [@aba].
We note that the matrix elements of the initial *reduced* electronic density operator are obtained from Eq. (\[eq:rho-reduced\]) as $$\begin{aligned}
\langle\{m_{b}\}|\hat{\varrho}_{S}(0)|\{s_{b}\}\rangle & = & \sum_{\{N_{{\mathbf}{k}}\}}\sum_{\{N_{{\mathbf}{k}}^{\prime}\}}\varrho_{\{m_{b}\},\{N_{{\mathbf}{k}}\};\{s_{b}\},\{N_{{\mathbf}{k}}^{\prime}\}}\nonumber \\
& & \qquad\times f_{\{s_{b}\},\{N_{{\mathbf}{k}}^{\prime}\};\{m_{b}\},\{N_{{\mathbf}{k}}\}},\label{eq:initial-reduced}\end{aligned}$$ with the non-diagonal elements containing the Franck–Condon type transition amplitudes $$\begin{aligned}
f_{\{s_{b}\},\{N_{{\mathbf}{k}}^{\prime}\};\{m_{b}\},\{N_{{\mathbf}{k}}\}}=\qquad\qquad\qquad\qquad\qquad\qquad\nonumber \\
\langle\{N_{{\mathbf}{k}}^{\prime}\}|\hat{D}\left(\left\{ \alpha_{\{s_{b}\},{\mathbf}{k}}\right\} \right)\hat{D}^{\dagger}\left(\left\{ \alpha_{\{m_{b}\},{\mathbf}{k}}\right\} \right)|\{N_{{\mathbf}{k}}\}\rangle.\end{aligned}$$ These factors are overlap integrals of two displaced phonon number states with displacements $\alpha_{\{s_{b}\},{\mathbf}{k}}$ and $\alpha_{\{m_{b}\},{\mathbf}{k}}$, respectively. Theirpresenceis due to the fact that the initial density matrix of the complete electron-phonon system (\[eq:initial-complete\]) is in the basis of theenergy eigenstates (\[eq:eigenstates\]), whereas the matrix elements of Eq. (\[eq:initial-reduced\]) are in the basis of the product states $|\{m_{b}\}\rangle\otimes|\{N_{{\mathbf}{k}}\}\rangle$ that differ by the displacement of the phonons.
Consistent with a dephasing model, the diagonal elements follow from Eq. (\[eq:rho-reduced\]) as invariants: $$\langle\{m_{b}\}|\hat{\varrho}_{S}(t)|\{m_{b}\}\rangle=\langle\{m_{b}\}|\hat{\varrho}_{S}(0)|\{m_{b}\}\rangle.$$ However, this dephasing — being induced by off-resonant scattering of acoustical phonons — modifies the time evolution of the off-diagonal density matrix elements as $$\begin{aligned}
\langle\{m_{b}\}|\hat{\varrho}_{S}(t)|\{s_{b}\}\rangle & = & e^{-i\sum_{b}(m_{b}-s_{b})\omega_{b}t}\label{eq:rho-ge}\\
& \times & \sum_{\{N_{{\mathbf}{k}}\}}\sum_{\{N_{{\mathbf}{k}}^{\prime}\}}\varrho_{\{m_{b}\},\{N_{{\mathbf}{k}}\};\{s_{b}\},\{N_{{\mathbf}{k}}^{\prime}\}}\nonumber \\
& \times & f_{\{s_{b}\},\{N_{{\mathbf}{k}}^{\prime}\};\{m_{b}\},\{N_{{\mathbf}{k}}\}}e^{-i\sum_{{\mathbf}{k}}\omega_{k}(N_{{\mathbf}{k}}-N_{{\mathbf}{k}}^{\prime})t}.\nonumber \end{aligned}$$ It can be observed in Eq. (\[eq:rho-ge\]) that apart from the free oscillation with angular frequency $\omega_{0}$, a dephasing is induced bydiffering phonon numbers in combination with the presence of the Franck–Condon factor.
To further evaluate the dephasing of the off-diagonal density matrix elements of the $N$ qubits, we insert the initial complete density matrix elements (\[eq:initial-complete\]) together with Eq. (\[eq:thermal-state\]) into Eq. (\[eq:rho-ge\]). From this we obtain
$$\begin{aligned}
\langle\{m_{b}\}|\hat{\varrho}_{S}(t)|\{s_{b}\}\rangle & = & \psi_{\{m_{b}\}}\psi_{\{s_{b}\}}^{\ast}e^{-i\sum_{b}(m_{b}-s_{b})\omega_{b}t}\sum_{\{M_{{\mathbf}{k}}\}}P_{\{M_{{\mathbf}{k}}\}}\sum_{\{N_{{\mathbf}{k}}\}}\sum_{\{N_{{\mathbf}{k}}^{\prime}\}}e^{-i\sum_{{\mathbf}{k}}\omega_{k}(N_{{\mathbf}{k}}-N_{{\mathbf}{k}}^{\prime})t}\nonumber \\
& \times & \times f_{\{s_{b}\},\{N_{{\mathbf}{k}}^{\prime}\};\{-\frac{1}{2},\ldots,-\frac{1}{2}\},\{M_{{\mathbf}{k}}\}}^{\ast}f_{\{s_{b}\},\{N_{{\mathbf}{k}}^{\prime}\};\{m_{b}\},\{N_{{\mathbf}{k}}\}}f_{\{m_{b}\},\{N_{{\mathbf}{k}}\};\{-\frac{1}{2},\ldots,-\frac{1}{2}\},\{M_{{\mathbf}{k}}\}}.\label{eq:rho-ge-1-2}\end{aligned}$$
Employing the thermal phonon statistics (\[eq:thermal-statistics\]) the sum over the phonon numbers in Eq. (\[eq:rho-ge-1-2\]) can be rewritten as a trace, which leaves us with $$\begin{aligned}
\langle\{m_{b}\}|\hat{\varrho}_{S}(t)|\{s_{b}\}\rangle & = & \psi_{\{m_{b}\}}\psi_{\{s_{b}\}}^{\ast}e^{-i\sum_{b}(m_{b}-s_{b})\omega_{b}t}Z^{-1}{\rm Tr}\left[\hat{D}(\{\alpha_{\{-\frac{1}{2},\ldots,-\frac{1}{2}\},{\mathbf}{k}}\})\hat{D}^{\dagger}(\{\alpha_{\{s_{b}\},{\mathbf}{k}}\})\right.\nonumber \\
& & \times\left.\hat{D}(\{\alpha_{\{s_{b}\},{\mathbf}{k}}(t)\})\hat{D}^{\dagger}(\{\alpha_{\{m_{b}\},{\mathbf}{k}}(t)\})\hat{D}(\{\alpha_{\{m_{b}\},{\mathbf}{k}}\})\hat{D}^{\dagger}(\{\alpha_{\{-\frac{1}{2},\ldots,-\frac{1}{2}\},{\mathbf}{k}}\})e^{-\sum_{{\mathbf}{k}}\beta_{k}\hat{N}_{{\mathbf}{k}}}\right],\label{eq:rho-ge-1-1}\end{aligned}$$
where we defined the time-dependent phonon displacement amplitude $\alpha_{\{m_{b}\},{\mathbf}{k}}(t)=\alpha_{\{m_{b}\},{\mathbf}{k}}\exp(-i\omega_{k}t)$. The displacement operators in Eq. (\[eq:rho-ge-1-1\]) can be combined to obtain $$\begin{aligned}
\langle\{m_{b}\}|\hat{\varrho}_{S}(t)|\{s_{b}\}\rangle=\psi_{\{m_{b}\}}\psi_{\{s_{b}\}}^{*}e^{-i\sum_{b}(m_{b}-s_{b})\omega_{b}t+i\Delta_{\{m_{b}\},\{s_{b}\}}(t)}\nonumber \\
\times Z^{-1}{\rm Tr}\left[\hat{D}\left(\delta_{\{m_{b}\},\{s_{b}\},{\mathbf}{k}}\left(1-e^{-i\omega_{k}t}\right)\right)e^{-\sum_{{\mathbf}{k}}\beta_{k}\hat{N}_{{\mathbf}{k}}}\right],\qquad\label{eq:rho-offdiag}\end{aligned}$$ with the time dependent phase being
$$\Delta_{\{m_{b}\},\{s_{b}\}}(t)=\Im\sum_{{\mathbf}{k}}\bar{\alpha}_{\{m_{b}\},\{s_{b}\},{\mathbf}{k}}\delta_{\{m_{b}\},\{s_{b}\},{\mathbf}{k}}^{\ast}\left(e^{i\omega_{k}t}-1\right)$$
Here we have defined sum and difference displacements: $$\begin{aligned}
\bar{\alpha}_{\{m_{b}\},\{s_{b}\},{\mathbf}{k}} & = & \alpha_{\{m_{b}\},{\mathbf}{k}}+\alpha_{\{s_{b}\},{\mathbf}{k}}-2\alpha_{\{-\frac{1}{2},\ldots,-\frac{1}{2}\},{\mathbf}{k}},\\
\delta_{\{m_{b}\},\{s_{b}\},{\mathbf}{k}} & = & \alpha_{\{m_{b}\},{\mathbf}{k}}-\alpha_{\{s_{b}\},{\mathbf}{k}}.\end{aligned}$$
The trace in Eq. (\[eq:rho-offdiag\]) represents a thermal average that can be evaluated in phase space to obtain $$\begin{aligned}
\langle\{m_{b}\}|\hat{\varrho}_{S}(t)|\{s_{b}\}\rangle & = & \psi_{\{m_{b}\}}\psi_{\{s_{b}\}}^{*}e^{-i\sum_{b}(m_{b}-s_{b})\omega_{b}t+i\Delta_{\{m_{b}\},\{s_{b}\}}(t)}\nonumber \\
& & \times\exp\left\{ -\int_{0}^{t}dt^{\prime}\Gamma_{\{m_{b}\},\{s_{b}\}}(t^{\prime})\right\} ,\label{eq:rho-offdiag2}\end{aligned}$$ where the decoherence rate of the qubit state is defined as $$\begin{aligned}
\Gamma_{\{m_{b}\},\{s_{b}\}}(t) & = & \sum_{b,b'}(m_{b}-s_{b})(m_{b'}-s_{b'})\gamma_{b,b'}(t).\label{eq:decrate}\end{aligned}$$ Whereas this rate depends on the quantum numbers of the density matrix element under consideration, the relation of pairs of bits is governed by the inter-bit decorrelation rate, given by $$\begin{aligned}
\gamma_{b,b'}(t) & =4 & \sum_{m_{b}}\sum_{s_{b^{\prime}}}m_{b}s_{b^{\prime}}\gamma\left(t;a_{b,m_{b}},a_{b',s_{b'}},l_{b,m_{b};b',s_{b'}}\right).\label{eq:inter-bit-rate}\end{aligned}$$ This rate in turn depends via the inter-donor decoherence rate $\gamma(t;a,a^{\prime},l)$ on the Bohr radii of the four donor sites of the two qubits and on the six possible distances between these four donors, $$l_{b,m_{b};b',s_{b'}}=\left|({\mathbf}{r}_{b}+m_{b}{\mathbf}{d}_{b})-({\mathbf}{r}_{b^{\prime}}+s_{b'}{\mathbf}{d}_{b'})\right|.\label{eq:inter-bit-distance}$$ Moreover, the inter-donor decoherence rate is obtained from Eqs. (\[eq:rho-offdiag\]) – (\[eq:inter-bit-distance\]) as
$$\gamma(t;a,a^{\prime},l)=\Gamma_{T}\left[\left(\frac{aa_{{\rm B}}}{a^{2}-a^{\prime2}}\right)^{2}\sum_{\sigma=\pm1}\sigma\left(\frac{|l-\sigma st|}{l}+\frac{a}{2l}\frac{a^{2}-5a^{\prime2}}{a^{2}-a^{\prime2}}\right)e^{-2|l-\sigma st|/a}+(a\leftrightarrow a^{\prime})\right].\label{eq:inter-donor-rate}$$
In this expression the temperature dependent rate reads $\Gamma_{T}=\omega_{{\rm B}}(T/T_{{\rm B}})$, where the convenient temperature scale is chosen as $$k_{{\rm B}}T_{{\rm B}}=N_{{\rm B}}M_{0}s^{2}\left(\frac{\hbar\omega_{{\rm B}}}{D}\right)^{2}.$$ Here $\omega_{{\rm B}}=2\pi s/a_{{\rm B}}$ with $a_{{\rm B}}$ being the average Bohr radius of all donor sites and $N_{{\rm B}}$ is the number of unit cells within the average Bohr volume $a_{{\rm B}}^{3}$.
In the limit of identical donor sites, $a_{b,m_{b}}\to a_{{\rm B}}$ ($b=1,\ldots,N$), the inter-donor decoherence rate (\[eq:inter-donor-rate\]) approaches the form $\gamma(t;a,a^{\prime},l)\to\gamma(t;l)$ with
$$\gamma(t;l)=\Gamma_{T}\frac{a_{B}}{l}\sum_{\sigma=\pm1}\sigma\left[\frac{1}{6}\left(\frac{|l-\sigma st|}{a_{{\rm B}}}\right)^{3}+\frac{1}{2}\left(\frac{|l-\sigma st|}{a_{{\rm B}}}\right)^{2}+\frac{5}{8}\left(\frac{|l-\sigma st|}{a_{{\rm B}}}\right)+\frac{5}{16}\right]e^{-2|l-\sigma st|/a_{{\rm B}}}.$$
We note that in the limit of vanishing distance between donor sites, $l\to0$, this function becomes $$\gamma(t;0)=\Gamma_{T}\left[\frac{2}{3}\left(\frac{st}{a_{{\rm B}}}\right)^{3}+\left(\frac{st}{a_{{\rm B}}}\right)^{2}+\frac{1}{2}\left(\frac{st}{a_{{\rm B}}}\right)\right]e^{-2st/a_{{\rm B}}}.$$ This is a function peaked at $t\sim a_{{\rm B}}/s$, i.e. at the time a phonon needs to travel the distance of one Bohr radius, see Fig. \[fig:Dimensionless-and-temperature\] (red curve). Different from this special case ($l\to0$) for $l>0$ the inter-donor decoherence rate is peaked at the time $t\sim l/s$ that is required for a phonon to travel the distance $l$, see Fig. \[fig:Dimensionless-and-temperature\] (green and blue curves).
![Dimensionless and temperature independent inter-site decoherence rate $\gamma(l;t)/\Gamma_{T}$ as a function of the traveled distance of a phonon in units of the Bohr radius, $st/a_{{\rm B}}$. \[fig:Dimensionless-and-temperature\]](figure2){width="1\columnwidth"}
Dynamics of entanglement between two qubits
===========================================
In the following we will discuss the special case of two qubits being present in the semiconductor system, i.e. $N=2$. In this case the state of the qubits $|\{m_{b}\}\rangle$ with $b=1,2$ lives in a four-dimensional Hilbert space and the entanglement of this bipartite system can be described by the concurrence $C$ [@woo]. This quantity is a measure of entanglement bounded in the range between zero and one, with the maximum entanglement corresponding to unit concurrence. On the other hand, for separable states, lacking any entanglement, the concurrence is zero.
Furthermore, due to the pure dephasing effect of the phonon scattering, no transitions of the qubits are induced. Therefore, the phonon scattering will transform an initial general superposition state $$|\psi(0)\rangle=\sum_{m_{1},m_{2}=\pm\frac{\text{1}}{2}}\psi_{\{m_{1},m_{2}\}}|\{m_{1},m_{2}\}\rangle,$$ into a non-pure statistical mixture of only those states that initially already existed. Thus, during the time evolution the density operator stays within the Hilbert sub space defined by the initial state. This feature allows us to separately treat the two prominent cases of having initially one or two “excitations”, respectively.
Case of one excitation
----------------------
Assume the initial state of the two qubits to be of the form $$|\psi(0)\rangle=\psi_{\{\frac{1}{2},-\frac{1}{2}\}}|\{{\textstyle \frac{1}{2}},-{\textstyle \frac{1}{2}}\}\rangle+\psi_{\{-\frac{1}{2},\frac{1}{2}\}}|\{-{\textstyle \frac{1}{2}},{\textstyle \frac{1}{2}}\}\rangle.$$ Since only one of the qubits is in its “excited” state, this superposition is usually denoted as the “one excitation” case. Choosing the basis vectors of the bipartite system as $\{|\{{\textstyle \frac{1}{2}},{\textstyle \frac{1}{2}}\}\rangle,|\{{\textstyle \frac{1}{2}},-{\textstyle \frac{1}{2}}\}\rangle,|\{-{\textstyle \frac{1}{2}},{\textstyle \frac{1}{2}}\}\rangle,|\{-{\textstyle \frac{1}{2}},-{\textstyle \frac{1}{2}}\}\rangle\}$, following Eq. (\[eq:rho-offdiag2\]) the time-dependent density matrix can be written as $$\begin{tabular}{l}
\ensuremath{\varrho_{{\rm S}}(t)=\left(\begin{array}{cccc}
0 & 0 & 0 & 0\\
0 & \varrho_{\{\frac{1}{2},-\frac{1}{2}\},\{\frac{1}{2},-\frac{1}{2}\}}(t) & \varrho_{\{\frac{1}{2},-\frac{1}{2}\},\{-\frac{1}{2},\frac{1}{2}\}}(t) & 0\\
0 & \varrho_{\{\frac{1}{2},-\frac{1}{2}\},\{-\frac{1}{2},\frac{1}{2}\}}^{\ast}(t) & \varrho_{\{-\frac{1}{2},\frac{1}{2}\},\{-\frac{1}{2},\frac{1}{2}\}}(t) & 0\\
0 & 0 & 0 & 0
\end{array}\right).}\end{tabular}\label{estado1}$$ For this particular form of the density matrix, the concurrence simplifies to $$C(t)=2\left|\varrho_{\{\frac{1}{2},-\frac{1}{2}\},\{-\frac{1}{2},\frac{1}{2}\}}(t)\right|.\label{eq:concurrence-1ex}$$ Inserting the corresponding density matrix element from Eq. (\[eq:rho-offdiag2\]) into Eq. (\[eq:concurrence-1ex\]), the concurrence results as $$C(t)=2\sqrt{p(1-p)}\exp\left[-\int_{0}^{t}dt^{\prime}\Gamma_{\{\frac{1}{2},-\frac{1}{2}\}\{-\frac{1}{2},\frac{1}{2}\}}(t^{\prime})\right],\label{eq:concurrence-1ex-2}$$ where $p=|\psi_{\{\frac{1}{2},-\frac{1}{2}\}}|^{2}$ is the initial probability for the two qubits being in state $|\{{\textstyle \frac{1}{2}},-{\textstyle \frac{1}{2}}\}\rangle$. The maximum initial concurrence is of course obtained for equal weights, $p=1/2$, of the two constituent states.
From Eq. (\[eq:concurrence-1ex-2\]) it becomes apparent that the decoherence rate of the state of the qubits acts as disentanglement rate. This rate is shown in Fig. \[fig:Dimensionless\] for qubits with $d=10a_{{\rm B}}$, an inter-qubit distance of $20a_{{\rm B}}$, and a relative angle of $45\degree$, see inset of Fig. \[fig:Dimensionless\]. It shows a series of alternating maxima and minima at increasing times. The principal positive maximum at the beginning occurs at $t\approx a_{{\rm B}}/s$, which is the time needed by the phonon to travel within a donor site. This is the main source of disentanglement. The times of the subsequent extrema can be identified as the travel times between pairs of donor sites, as indicated in the inset of Fig. \[fig:Dimensionless\]. Whereas the positive maxima destroy, the negative minima restore the entanglement between the qubits. The mapping of which phonon path between donor sites leads to positive or negative extrema in the disentanglement rate can be established as follows:
The concurrence is given by the modulus of the density matrix element $\varrho_{\{\frac{1}{2},-\frac{1}{2}\},\{-\frac{1}{2},\frac{1}{2}\}}(t)$ that describes the time-dependent correlation between states $|\{-\frac{1}{2},\frac{1}{2}\}\rangle$ and $|\{\frac{1}{2},-\frac{1}{2}\}\rangle$. Correlations between these states can only be created when a phonon travels between a donor site occupied by one state to another donor site that is occupied by the other state. The corresponding site occupations of each of the states involved are indicated in the inset of Fig. \[fig:Dimensionless\] by black and white colors, respectively. The sites of each qubit are marked by blue for $m_{b}=-\frac{1}{2}$ and red for $m_{b}=+\frac{1}{2}$. With this color scheme, the creation of correlations mediated by sound waves is produced via phonon travels between a black and a white donor site.
These phonon travels are: the passage within the individual qubits for the distance $10a_{{\rm B}}$, that produces in Fig. \[fig:Dimensionless\] the negative minimum 1, and the passages between blue and red sites at distances $\approx16.5a_{{\rm B}}$ and $\approx23.6a_{{\rm B}}$, that produce the minima 2 and 4. All the other phonon passages produce decorrelation and destroy the entanglement at distances $\approx18.6a_{{\rm B}}$ and $\approx25.0a_{{\rm B}}$, which generate the positive maxima 3 and 5 in Fig. \[fig:Dimensionless\].
![\[fig:Dimensionless\]Dimensionless disentanglement rate $\Gamma_{\{\frac{1}{2},-\frac{1}{2}\}\{-\frac{1}{2},\frac{1}{2}\}}(t)/\Gamma_{T}$ as a function of the dimensionless time $st/a_{{\rm B}}$ for the case of a single excitation. The inset shows the geometrical configuration of the donor sites: $d_{1}=d_{2}=10a_{{\rm B}}$, $|{\mathbf}{r}_{1}-{\mathbf}{r}_{2}|=20a_{{\rm B}}$, $45\degree$ angle between qubit axes. The rounded lengths of the inter-donor distances in units of $a_{{\rm B}}$ are: $l/a_{{\rm B}}=10$ (1), $16.53$ (2), $18.6$ (3), $23.6$ (4), $25.0$ (5).](figure3){width="1\columnwidth"}
We note that diminishing the $45\degree$ angle between the qubit axes results in lengths 2 and 4, and lengths 3 and 5, respectively, becoming progressively comparable. In the limiting case of two collinear qubits, i.e. $0\degree$ angle, these pairs of lengths are identical so that as a consequence the disentanglement rate shows only two positive and two negative peaks, as shown in Fig. \[fig:paralel\] (a). On the other hand, as one approaches the limiting angle of $90\degree$, i.e. the CNOT configuration [@hol; @tsu; @ste], the extrema first sparse and finally peaks 2 and 4 cancel peaks 3 and 5, respectively. As a result, the disentanglement rate shows only the principal positive peak at $t\approx a_{{\rm B}}/s$, see Fig. \[fig:paralel\] (b).
The evolution of the concurrence for the case of $45\degree$ between qubit axes is shown in Fig. \[fig:Concurrence-as-a\]. It can be seen that a stationary and non-vanishing value of the concurrence is reached for large times. Moreover, the temperature dependence indicates an only minor loss of entanglement at temperatures $T/T_{{\rm B}}<0.01$. Given that for P impurities embedded in a Si substrate the characteristic temperature is of the order of $T_{{\rm B}}\sim300\kelvin$, this case corresponds to liquid He temperatures.
![Dimensionless disentanglement rate $\Gamma_{\{\frac{1}{2},-\frac{1}{2}\}\{-\frac{1}{2},\frac{1}{2}\}}(t)/\Gamma_{T}$ as a function of the dimensionless time $st/a_{{\rm B}}$ for the case of a single excitation. Same parameters as in Fig. 3, but for collinear (a) and perpendicular (b) qubits.\[fig:paralel\]](figure4a){width="1\columnwidth"}
![Dimensionless disentanglement rate $\Gamma_{\{\frac{1}{2},-\frac{1}{2}\}\{-\frac{1}{2},\frac{1}{2}\}}(t)/\Gamma_{T}$ as a function of the dimensionless time $st/a_{{\rm B}}$ for the case of a single excitation. Same parameters as in Fig. 3, but for collinear (a) and perpendicular (b) qubits.\[fig:paralel\]](figure4b){width="1\columnwidth"}
![\[fig:Concurrence-as-a\]Concurrence as a function of the dimensionless time $st/a_{{\rm B}}$ for the case of a single excitation for temperatures $T/T_{{\rm B}}=0.01$ (solid curve), $0.05$ (dotted curve), $0.1$ (dashed curve). Other parameters same as in Fig. 3 with $p=1/2$.](figure5){width="1\columnwidth"}
Case of two excitations
-----------------------
The other prominent case is that of initial two excitations. This case is described by the initial state being of the form $$|\psi(0)\rangle=\psi_{\{\frac{1}{2},\frac{1}{2}\}}|\{{\textstyle \frac{1}{2}},{\textstyle \frac{1}{2}}\}\rangle+\psi_{\{-\frac{1}{2},-\frac{1}{2}\}}|\{-{\textstyle \frac{1}{2}},-{\textstyle \frac{1}{2}}\}\rangle.$$ In the same standard basis as before, the time-dependent density matrix results then as $$\begin{tabular}{l}
\ensuremath{\varrho(t)=\left(\begin{array}{cccc}
\varrho_{\{\frac{1}{2},\frac{1}{2}\},\{\frac{1}{2},\frac{1}{2}\}}(t) & 0 & 0 & \varrho_{\{\frac{1}{2},\frac{1}{2}\},\{-\frac{1}{2},-\frac{1}{2}\}}(t)\\
0 & 0 & 0 & 0\\
0 & 0 & 0 & 0\\
\varrho_{\{\frac{1}{2},\frac{1}{2}\},\{-\frac{1}{2},-\frac{1}{2}\}}^{\ast}(t) & 0 & 0 & \varrho_{\{-\frac{1}{2},-\frac{1}{2}\},\{-\frac{1}{2},-\frac{1}{2}\}}(t)
\end{array}\right).}\end{tabular}\label{estado2}$$ Also for this case the concurrence simplifies to a simple expression, given by $$C(t)=2\left|\varrho_{\{\frac{1}{2},\frac{1}{2}\},\{-\frac{1}{2},-\frac{1}{2}\}}(t)\right|,$$ which, after insertion of Eq. (), becomes
$$C(t)=2\sqrt{p(1-p)}\exp\left[-\int_{0}^{t}dt^{\prime}\Gamma_{\{\frac{1}{2},\frac{1}{2}\},\{-\frac{1}{2},-\frac{1}{2}\}}(t^{\prime})\right],$$
where now $p=|\psi_{\{\frac{1}{2},\frac{1}{2}\}}|^{2}$ is the probability for the two qubits being “excited”. From the change of indices it can be easily seen that, apart from the first positive and first negative peak, the signs of the peaks of the disentanglement rate are reversed as compared to the corresponding one excitation case, see Fig. \[fig:Dimensionless-disentanglement-ra-10\].
![\[fig:Dimensionless-disentanglement-ra-10\]Dimensionless disentanglement rate $\Gamma_{\{\frac{1}{2},\frac{1}{2}\}\{-\frac{1}{2},-\frac{1}{2}\}}(t)/\Gamma_{T}$ as a function of the dimensionless time $st/a_{{\rm B}}$ for the case of two excitations. The inset shows the geometrical configuration of the donor sites: $d_{1}=d_{2}=10a_{{\rm B}}$, $|{\mathbf}{r}_{1}-{\mathbf}{r}_{2}|=20a_{{\rm B}}$, $45\degree$ angle between qubit axes. The rounded lengths of the inter-donor distances in units of $a_{{\rm B}}$ are: $l/a_{{\rm B}}=10$ (1), $16.53$ (2), $18.6$ (3), $23.6$ (4), $25.0$ (5).](figure6){width="1\columnwidth"}
Summary and Outlook
===================
In this paper we considered the time evolution of entanglement in donor-based charge qubits that is induced by off-resonant scattering with acoustical phonons. We showed that this system can be solved analytically and that a non-Markovian behavior emerges with negative disentanglement rates, leading to non-monotonic disentanglement in time. Moreover, for the cases of one and two initial excitations the disentanglement rate is proportional to the decoherence rate of the two-qubit state. In both cases the concurrence attains a stationary and non-vanishing value at large times, which means that phonon scattering does not completely destroy the entanglement of the initially prepared two-qubit state.
The choice of the geometry of the donor sites determines the features of the concurrence as a time-dependent function. These features can be understood by a simple kinetic interpretation of phonon travels among the donor sites. In this work we focused on the cases of initially one and two excitations. However, we believe that also particular superpositions of both cases may be treated within this framework. Furthermore, our model includes already the case of $N>2$ qubits, where a trace over $N-2$ qubits would be required to obtain the entanglement between a selected pair of qubits. This will be subject of future work.
SW and FL acknowledge support by FONDECYT project no. 1095214. FL acknowledges support from Financiamiento Basal project no. 0807.
[References]{} M.A. Nielsen and I.L. Chuang, *Quantum Computation and Quantum Information* (Cambridge University Press, Cambridge, 2000)
C.H. Bennett, G. Brassard, C. Crépeau, R. Jozsa, A. Peres, and W.K. Wootters, Phys. Rev. Lett. **70**, 1895 (1993).
C.H. Bennett and S.J. Wiesner, Phys. Rev. Lett. **69**, 2881 (1992).
A.K. Ekert, Phys. Rev. Lett. **67**, 661 (1991).
D. Leibfried, R. Blatt, C. Monroe, and D. Wineland, Rev. Mod. Phys. **75**, 281 (2003).
L.M.K. Vandersypen and I.L. Chuang, Rev. Mod. Phys. **76**, 1037 (2004).
J.M. Raimond, M. Brune, and S. Haroche, Rev. Mod. Phys. **73**, 565 (2001).
R. Hanson, L.P. Kouwenhoven, J.R. Petta, S. Tarucha, and L.M.K. Vandersypen, e-print cond-mat/0610433.
B.E. Kane, Nature **393**, 133 (1998).
L. Dicarlo, J.M. Chow, J.M. Gambetta, L.S. Bishop, B.R. Johnson, D.I. Schuster, J. Majer, A. Blais, L. Frunzio, S.M. Girvin, and R.J. SchoelkopfNature **460**, 240 (2009).
T. Gaebel, M. Domhan, I. Popa, C. Wittmann, P. Neumann, F. Jelezko, J.R. Rabeau, N. Stavrias, A.D. Greentree, S. Prawer, J. Meijer, J. Twamley, P.R. Hemmer, and J. Wrachtrup, Nature Physics **2**, 408 (2006).
N.V. Prokof’ev and P.C.E. Stamp, Rep. Prog. Phys. **63**, 669 (2000).
L. Amico, R. Fazio, A. Osterloh, and V. Vedral, Rev. Mod. Phys. **80**, 517 (2008).
L. Chirolli and G. Burkard, Adv. in Physics **57**, 225 (2008).
L.C.L. Hollenberg, A.S. Dzurak, C. Wellard, A.R. Hamilton, D.J. Reilly, G.J. Milburn, and R.G. Clark, Phys. Rev. B **69**, 113301 (2004).
B. Koiller, X. Hu, and S. Das Sarma, Phys. Rev. B **73**, 045319 (2006).
A.V. Tsukanov and K.A. Valiev, Russian Microelectronics **36** (2), 6780 (2007).
D. Stepanenko and G. Burkard, Phys. Rev. B **75**, 085324 (2007).
H. Zhao and J.B. Freund, J. Appl. Phys. **104**, 033514 (2008).
F. Lastra, S.A. Reyes, and S. Wallentowitz, J. Phys. B **44**, 015504 (2010).
F. Lastra, S.A. Reyes, and S. Wallentowitz, Rev. Mex. Fís. **57**, 148 (2011).
J. Eckel, S. Weiss, and M. Thorwart, Eur. Phys. J. B **53**, 91 (2006).
A.J. Leggett, S. Chakravarty, A.T. Dorsey, M.P.A. Fisher, A. Garg, and W. Zwerger, Rev. Mod. Phys. **59**, 1 (1987).
M. Abanto, L. Davidovich, B. Koiller and R.L. de Matos Filho, Phys. Rev. B **81**, 085325 (2010).
W.K. Wooters, Phys. Rev. Lett. **80**, 2245 (1998).
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In a high-mass X-ray binary (HMXB) a massive star interacts with a neutron-star or black-hole companion in various ways. The gravitational interaction enables the measurement of fundamental parameters such as the mass of both binary components, providing important constraints on the evolutionary history of the system, the equation of state of matter at supra-nuclear density, and the supernova mechanism. The stellar wind of the massive star is intercepted by the strong gravitational field of the compact companion, giving rise to the production of X-rays. The X-rays increase the degree of ionization in a small to very extended region of the surrounding stellar wind, depending on the X-ray luminosity. This has observable consequences for the structure and dynamics of the accretion flow. In this paper we concentrate on the fundamental paramaters of the most massive HMXBs, i.e. those with an OB supergiant companion, including some systems exhibiting relativistic jets (“microquasars”).'
author:
- 'L. Kaper & A. Van der Meer'
title: 'Massive Stars and their Compact Remnants in High-mass X-ray Binaries'
---
High-mass X-ray binaries
========================
In a high-mass X-ray binary (HMXB) a massive OB-type star is in close orbit with a compact X-ray source, a neutron star or a black hole. The X-ray source is powered by accretion of material originating from the OB star, transported by the OB-star wind or by Roche-lobe overflow. The majority ($\ga 80$ %) of the HMXBs are Be/X-ray binaries with relatively wide ($P_{\rm orb}$ weeks to several years) and eccentric orbits. Most Be/X-ray binaries are transients, the X-ray flux being high when the compact star in its eccentric orbit passes through the dense equatorial disk around the Be star (Van den Heuvel & Rappaport 1987, Negueruela, these proceedings). About a dozen HMXBs host a massive OB-supergiant companion (about 10 to over 40 M$_{\odot}$), in a relatively tight orbit ($P_{\rm orb}$ several days) with an X-ray pulsar or black-hole companion (e.g. Kaper 2001). Some of these X-ray binaries include a dense accretion disk and produce relativistic jets (e.g. Fender 2005).
HMXBs mark an important though short phase (on the order of 10,000 year) in the evolution of the most massive binaries. The compact companion is the remnant of the initially most massive star in the system that exploded as a supernova (or as a gamma-ray burst). Due to a phase of mass transfer, the secondary became the most massive star in the system before the primary supernova, so that the system remained bound (Van den Heuvel & Heise 1972). A consequence, however, is that HMXBs are runaways due to the kick velocity exerted by the supernova (Blaauw 1961, Kaper et al. 1997, Van den Heuvel et al. 2000). When the secondary starts to become a supergiant, a second phase of mass transfer is initiated, first through an enhanced stellar wind, later by Roche-lobe overflow, which results in the production of X-rays by the compact companion. With increasing mass transfer rate, the system will enter a phase of common-envelope evolution causing the compact object to spiral into the OB companion. In the relatively wide Be/X-ray binaries the spiral-in likely results in the removal of the envelope of the Be companion, and after the supernova a bound (or disrupted) double neutron star remains, like the Hulse-Taylor binary pulsar PSR 1913+16 (or a neutron star – white dwarf system, if the mass of the Be companion is less than $\sim 8$ M$_{\odot}$). In close HMXBs (orbital period less than about a year) the compact object will enter the core of the OB companion which will become a so-called Thorne-Zytkow object (Thorne & Zytkow 1977), a red supergiant with a high mass-loss rate. These objects have been predicted on evolutionary grounds, but have so far not been recognized as such (for an extensive review on binary evolution, see Van den Heuvel 1994).
In case the system hosts an X-ray pulsar, its orbit can be very accurately determined through pulse-timing analysis. When also the radial-velocity curve of the OB supergiant is obtained, the mass of the neutron star and that of the massive star can be derived with precision, given an estimate of the system inclination (cf. Rappaport & Joss 1983). The neutron star mass carries information on the formation mechanism (i.e. the supernova), as well as on the equation of state (EOS) of matter at supra-nuclear density. Evolutionary calculations by Timmes et al. (1996) show that in single hydrogen-rich stars with a mass less than 19 M$_{\odot}$ the collapsing Fe core has a mass of about 1.3 M$_{\odot}$, while stars with a higher initial mass produce an Fe core of about 1.7 M$_{\odot}$. Therefore, one would predict a bimodal mass distribution of the neutron stars in these systems.
The EOS, i.e. the relation between pressure and density in the neutron-star interior, can so far only be studied on the basis of theoretical models. It is not yet possible to produce the required extremely high density (about an order of magnitude higher than an atomic nucleus) in accelerator experiments. Theoretical predictions of the EOS come in two flavors: the so-called “soft” and “hard” equations of state. The hardness of the EOS depends on the fraction of bosons formed in the neutron-star interior, which unlike the fermions (neutrons) do not contribute to the Fermi pressure that helps to sustain gravity. The harder the EOS, the higher the mass a neutron star can have. If a neutron star has a mass higher than the maximum mass allowed by a given EOS, this EOS is ruled out as only one EOS can be the right one. Obviously, this relatively simple measurement has major implications for our understanding of the properties of matter at supra-nuclear density.
The properties of the OB star in a HMXB give insight into the evolutionary history of the binary system: its current mass provides a constraint on the initial mass of the primary, its surface may show evidence of nuclearly processed material coming from the primary, and its rotation rate has been altered by the angular momentum content of the gained mass. Also, the space velocity of the system carries information on the amount of mass lost during the supernova explosion (Nelemans et al. 1999, Ankay et al. 2001).
Contrary to the Be/X-ray binaries, a few OB-supergiant systems host a black-hole companion (e.g. Cyg X-1, Gies & Bolton 1982), which suggests that black holes are formed by the most massive stars only. The majority of the about twenty known black-hole candidates have a low-mass companion (soft X-ray transient or X-ray nova, e.g. McClintock & Remillard 2005), but only the three black-hole candidates known in HMXBs are persistent X-ray sources. Some of the black holes in X-ray novae appear to be linked to hypernovae believed to power gamma-ray bursts, based on the detection of r-processed elements at the surface of the low-mass companions (e.g. Israelian et al. 1999).
OB supergiant systems
=====================
Table 1 lists the basic properties of the HMXBs with OB supergiant companions in the Milky Way and the Magellanic Clouds. Most sources contain an X-ray pulsar: the pulse period is short and the X-ray luminosity high (near the Eddington limit $L_{X} \sim
10^{38}$ erg s$^{-1}$) in systems undergoing Roche-lobe overflow due to the higher accretion rate. The latter systems also have circular orbits, while the wind-fed systems have eccentricities up to $e =
0.45$ (GX301-2) and an X-ray luminosity $L_{X} \sim
10^{35}-10^{36}$ erg s$^{-1}$.
The X-ray source drastically increases the degree of ionization of the surrounding stellar wind. Hatchett & McCray (1977) predicted that the ionizing power of the X-ray source would cause the orbital modulation of ultraviolet resonance lines formed in the stellar wind. The Hatchett-McCray effect has been detected in several systems (e.g. Van Loon et al. 2001), and recently in the wind-fed system 4U1700-37 for which the original prediction was made (see Iping et al., these proceedings). Van der Meer et al. (2005a) find evidence for the ionization zone in 4U1700-37 through X-ray spectroscopy carried out with XMM-[*Newton*]{}. A secondary effect of the presence of an X-ray ionization zone is the development of a strong shock (called a photo-ionization wake) at its trailing border, where fast wind collides with the stagnant flow inside the ionization zone (Blondin et al. 1990, Kaper et al. 1994). In Roche-lobe overflow systems the high X-ray luminosity makes that only in the X-ray shadow behind the OB supergiant a normal stellar wind can develop, a so-called shadow wind (e.g. Blondin 1994, Kaper et al. 2005).
[@llcccc]{} Name & Sp. Type & M$_{\rm OB}$ & M$_{\rm X}$ & P$_{\rm orb}$ & P$_{\rm pulse}$ & & (M$_{\odot}$) & (M$_{\odot}$) & (d) & (s) 2S0114+650 & B1 Ia & $\sim$16 & $\sim$1.7 & 11.6 & 860$^a$ SMC X-1 & B0 Ib & 15.5 & 1.1 & 3.89 & 0.71 LMC X-4 & O7 III-IV & 15.6 & 1.3 & 1.40 & 13.5 Vela X-1 & B0.5 Ib & 23.9 & 1.9 & 8.96 & 283 Cen X-3 & O6.5 II-III & 19.7 & 1.2 & 2.09 & 4.84 GX301-2 & B1.5 Ia$^{+}$ & $>$40 & $>$1.3 & 41.5 & 696 4U1538-52 & B0 Iab & 16.4 & 1.1 & 3.73 & 529 4U1700-37 & O6.5 Iaf+ & $\sim$58 & $\sim$2.4 & 3.41 & 4U1907+09 & early B I & $\sim$27 & $\sim$1.4 & 8.38 & 438 LMC X-3 & B3 Ve & $\sim$6 & 6–9 & 1.70 & LMC X-1 & O7-9 III & $\sim$20 & 4–10 & 4.22 & LS5039 & O6.5 V((f)) & 20–35 & 1.4 & 4.43 & SS433 & A3-7 I & 10.9 & 2.9 & 13.08 & Cyg X-1 & O9.7 Iab & 17.8 & 10 & 5.60 &
The mass of the OB supergiant primary and the compact companion in a HMXB can be accurately measured when the system hosts an X-ray pulsar. Knowledge of the orbital inclination is essential; in systems showing an X-ray eclipse the inclination must be larger than $i \sim
65^{\circ}$. For Roche-lobe overflow systems a valid assumption is that the OB supergiant is in corotation with the orbit, which provides a strong constraint on the inclination. In eclipsing systems the radius of the OB supergiant can be derived from the duration of the X-ray eclipse. The mass ratio sets the size of the Roche lobe (e.g.Eggleton 1983); it turns out that the measured radii of the OB supergiants are in very good agreement with the estimated size of the Roche lobe (Kaper 2001).
Earlier studies (e.g. Conti 1978, Rappaport & Joss 1983) suggested that the OB supergiants in HMXBs are too luminous for their masses. E.g. the O6.5 giant companion of Cen X-3 has a mass of 20 M$_{\odot}$, while its luminosity corresponds to that of a star of more than 50 M$_{\odot}$. Besides this, the radius corresponding to the luminosity and effective temperature is larger than its measured (Roche-lobe) radius (cf. Kaper 2001). Thus, apart from being undermassive, the OB supergiants in HMXBs also seem to be undersized for their luminosity and temperature. This inconsistency is probably related to the phenomenon of Roche-lobe overflow. The OB star tries to become a supergiant, but at some point it reaches its critical Roche lobe and starts to transfer mass to its companion. While the luminosity of the star is determined by the core (which does not notice much of what is happening to the outer mantle), the star would like to be bigger than allowed by its Roche lobe and is peeled off.
Neutron stars and black holes
=============================
In Figure 1 the mass distribution is shown of neutron stars and black holes, based on measurements collected from literature (Stairs 2004, McClintock & Remillard 2005). The neutron stars occupy a relatively narrow mass range near 1.4 M$_{\odot}$. The most accurate neutron-star masses have been derived for the binary radio pulsars, with an average mass of 1.35$\pm$0.04 M$_{\odot}$ (Thorsett & Chakrabarty 1999). The X-ray pulsars show a somewhat larger mass range, extending both below and above 1.35 M$_{\odot}$. The neutron star in Vela X-1 is significantly more massive: $1.86 \pm 0.16$ M$_{\odot}$ (Barziv et al. 2001, Quaintrell et al. 2003). Such a high neutron-star mass would rule out the soft equations of state. Also for 4U 1700-37 a high neutron-star mass is claimed ($2.4 \pm 0.3$ M$_{\odot}$, Clark et al. 2002), although the X-ray source is, contrary to Vela X-1, not an X-ray pulsar (and perhaps a low-mass black hole). Van der Meer et al. (these proceedings) are currently analyzing the radial-velocity curves of other OB-supergiant systems with an (eclipsing) X-ray pulsar to find out whether Vela X-1 is an exception or that the neutron stars in these systems systematically deviate from the “canonical” mass of 1.35 M$_{\odot}$. This would provide an important constraint on the neutron-star formation mechanism (i.e. the supernova).
Evolutionary calculations by Timmes et al. (1996) predict that massive stars in close binaries which explode as Type Ib supernova give rise to initial neutron star masses in a narrow mass range around 1.3 M$_{\odot}$. This value does not include subsequent mass accretion from a reverse shock or from a massive component in a binary system, and Timmes et al. expect that the final masses could be somewhat higher. Interestingly, for single stars, which explode as Type II supernovae, they find a bimodal neutron-star mass distribution, with narrow peaks near 1.27 and 1.76 M$_{\odot}$. As mentioned, they did not find a bimodal distribution for stars in close binaries, but at present it is not clear whether this result will hold. If stars in close binaries turn out to be more similar to single stars after all, one could assign most neutron stars to the first peak, and Vela X-1 (and 4U1700-37) to the second.
The estimated masses of black-hole candidates are substantially larger ($8.4 \pm 2.0$ M$_{\odot}$) than those measured for neutron stars. This suggests that neutron stars and black holes are formed in different ways. If, for example, black holes are the result of “failed” supernovae in which the stellar mantle is not blown away, but accreting on the compact remnant, one would expect a significant difference in mass between neutron stars and black holes. However, if the mass of the (proto) neutron star is increased by the fall back of material which was located outside the collapsing degenerate Fe core, one would predict that neutron stars would occupy a range in mass, up to the maximum neutron star mass allowed by the equation of state. Certainly in the binary radio pulsars such a mass distribution is not observed. With the recent evidence that a black hole may be formed during the collapse of a massive star during a gamma-ray burst (GRB980425, Galama et al. 1998, Iwamoto et al. 1998), the hypothesis would be that neutron stars are formed in “ordinary” supernovae, while black holes originate from gamma-ray bursts.
Microquasars
============
Some X-ray binaries, most notably those hosting a black hole, produce relativistic jets (e.g. Fender 2005). A famous example is SS433 which shows collimated, precessing jets with velocities of $v = 0.26c$ (Margon 1982). In some systems, e.g. GRS1915+105, superluminal motions have been measured, proving that the material in the jet is moving at relativistic velocities as is observed in quasars. In our sample (Tab. 1) three “microquasars” are included, i.e. LS5039, SS433, and Cyg X-1. LMC X-1 and LMC X-3 are also candidate members of this group, but for these systems only upper limits are obtained in observations searching for the radio synchrotron emission produced by the jets. This jet phenomenon is not unique to black-hole systems; also some X-ray binaries hosting a neutron star are known to produce jets (e.g. Sco X-1, Cir X-1). Besides the jets, these systems also include a (large) accretion disk. Apparently, a relatively large mass and angular momentum accretion rate results in the formation of a dense accretion disk and jets.
Cyg X-1 is one of the most famous stellar-mass black-hole candidates and one of the most intensively studied X-ray sources in the sky, at all wavelengths. Cyg X-1 probably represents a situation between pure, spherical wind accretion and Roche-lobe overflow. Every few years Cyg X-1 makes a transition from a low/hard state to a high/soft state in which the soft X-ray flux increases dramatically and the spectrum softens for a period of weeks to months. The radio flux also varies during state changes and is associated with jets (Stirling et al. 2001). The precise physical cause for the state transitions remains unclear, but may be triggered by episodes of decreased mass-loss rate in the supergiant donor star (Gies et al. 2003).
Hillwig et al. (2004) present spectroscopy of SS433 obtained near primary eclipse and disk precessional phase $\Phi=0.0$, when the accretion disk is expected to be most “face-on”. These conditions are the most favourable to have a change to detect the mass donor. The spectra show clear evidence of absorption features consistent with a classification of an A3-A7 supergiant. The observed radial velocity variations are in antiphase to the disk spectrum; the latter includes strong emission lines similar to those observed in Wolf-Rayet stars (see also Fuchs et al., these proceedings). Hillwig et al. derive masses of $10.9 \pm 3.1$ M$_{\odot}$ and $2.9 \pm 0.7$ M$_{\odot}$ for the mass donor and compact object plus disk, respectively.
LS5039 is an O6.5 V((f)) star (Clark et al. 2001) with a compact companion, most likely a neutron star. It has radio-emitting relativistic jets and is probably a high-energy gamma-ray source as well (Paredes et al. 2000). It is a 4.4-day binary with a high eccentricity ($e = 0.41$), which probably results from the huge mass loss that occured with the supernova producing the compact star. McSwain et al. (2004) present new optical and ultraviolet spectra of the O star and find evidence for nitrogen enhancement and carbon depletion in its atmosphere, indicative of the accretion of nuclearly processed material originating from the compact star’s massive progenitor. The observed eccentricity and runaway velocity can be reconciled only if the neutron star received a modest kick velocity due to a slight asymmetry in the supernova explosion (during which more than 5 M$_{\odot}$ was ejected).
Ankay, A., Kaper, L. De Bruijne, J.H.J., et al. 2001, A&A 370, 170 Barziv, O., Kaper, L., Van Kerkwijk, M.H., et al. 2001, A&A 377, 925 Blaauw, A. 1961, Bull. Astr. Inst. Neth. 15, 265 Blondin, J.M., Kallman, T.R., Fryxell, B.A., Taam, R.E. 1990, ApJ 356, 591 Blondin, J.M. 1994, ApJ 435, 756 Clark, J.S., Reig, P., Goodwin, S.P., et al. 2001, A&A 376, 476 Clark, J.S., Goodwin, S.P., Crowther, P.A., et al. 2002, A&A 392, 909 Conti, P.S. 1978, A&A 63, 225 Corbet, R.H.D., Finley, J.P., Peele, A.G. 1999, ApJ 511, 876 Cowley, A.P., Crampton, D., Hutchings, J.B., et al. 1983, ApJ 272, 118 Cox, N.L.J., Kaper, L., Mokiem, M.R. 2005, A&A in press Eggleton, P. 1983, ApJ 268, 368 Fender, R.P. 2005, in “Compact Stellar X-Ray Sources”, Eds. W.H.G. Lewin, M. van der Klis, Cambridge University Press, astro-ph/0303339 Galama, T.J., Vreeswijk, P.M., Van Paradijs, J., et al. 1998, Nat 395, 670 Gies, D.R., Bolton, C.T. 1982, ApJ 260, 240 Gies, D.R., Bolton, C.T., Thomson, J.R., et al. 2003, ApJ 583, 424 Hatchett, S., McCray, R. 1977, ApJ 211, 552 Herrero, A., Kudritzki, R.P., Gabler, R., et al. 1995, A&A 297, 556 Hillwig, T.C., Gies, D.R., Huang, W., et al. 2004, ApJ 615, 422 Hutchings, J.B., Crampton, D., Cowley, A.P., et al. 1987, AJ 94, 340 Israelian, G., Rebolo, R., Basri, G., et al. 1999, Nat 401, 142 Iwamoto, K., Mazzali, P.A., Nomoto, K. 1998, Nat 395, 672 Kaper, L., Hammerschlag-Hensberge, G., Zuiderwijk, E.J. 1994, A&A 289, 846 Kaper, L., Van Loon, Th., Augusteijn, T., et al. 1997, ApJ 475, L37 Kaper, L. 2001, in Proc. “The influence of binaries on stellar population studies”, Ed. D. Vanbeveren, Kluwer, p. 125 Kaper, L., Van der Meer, A., Najarro, P. 2005, to be submitted to A&A Kaper, L., Van Loon, J.Th, Hammerschlag-Hensberge, G., et al. 2005, to be submitted to A&A Margon, B. 1982, Science 215, 247 McClintock, J.E., Remillard, R.E. 2005, in “Compact Stellar X-Ray Sources”, Eds. W.H.G. Lewin, M. van der Klis, Cambridge University Press, astro-ph/0306213 McSwain, M.V., Gies, D.R., Huang, W., et al. 2004, ApJ 600, 927 Nelemans, G., Tauris, T.M., Van den Heuvel, E.P.J. 1999, A&A 352, L87 Paredes, J.M., Marti, J., Ribo, M., Massi, M. 2000, Science 288, 2340 Quaintrell, H., Norton, A.J., Ash, T.D.C., et al. 2003, A&A 401, 313 Rappaport, S.A., Joss, P.C. 1983, in “Accretion-driven stellar X-ray sources”, Eds. W.H.G. Lewin, Van den Heuvel, E.P.J., Cambridge University Press Reig, P., Chakrabarty, D., Coe, M.J., et al. 1996, A&A 311, 879 Stairs, I.H. 2004, Science 304, 547 Stirling, A.M., Spencer, R.E., de la Force, C.J., et al. 2001, MNRAS 327, 1273 Timmes, F.X., Woosley, S.E., Weaver, T.A. 1996, ApJ 457, 834 Thorne, K.S., Zytkow, A.N. 1977, ApJ 212, 832 Thorsett, S.E., Chakrabarty, D. 1999, ApJ 512, 288 Van den Heuvel, E.P.J., Heise, J. 1972, Nat. Phys. Sci. 239, 67 Van den Heuvel, E.P.J., Rappaport, S. 1987, in Proc. IAU Coll. 92, p. 291 Van den Heuvel, E.P.J. 1994, in “Saas-Fee Advanced Course on Interacting Binaries”, Springer-Verlag, p. 263 Van den Heuvel, E.P.J., Portegies-Zwart, S.P.Z., Bhattacharya, D., Kaper, L. 2000, A&A 364, 563 Van der Meer, A., Kaper, L. Di Salvo, T., et al. 2005a, A&A in press Van der Meer, A., Kaper, L., Van Kerkwijk, M.H., Van den Heuvel, E.P.J. 2005b, to be submitted to A&A Van Kerkwijk, M.H., Van Paradijs, J., Zuiderwijk, E.J. 1995, A&A 303, 497 Van Loon, J.Th., Kaper, L., Hammerschlag-Hensberge, G. 2001, A&A 375, 498
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The NSF’s Karl G. Jansky Very Large Array (VLA) is used to observe 122 magnetic cataclysmic variables (MCVs) during three observing semesters (13B, 15A, and 18A). We report radio detections of 33 stars with fluxes in the range 6–8031 $\mu$Jy. Twenty-eight stars are new radio sources, increasing the number of radio detected MCVs to more that 40. A surprising result is that about three-quarters (24 of 33 stars) of the detections show highly circularly polarized radio emission of short duration, which is characteristic of electron cyclotron maser emission. We argue that this emission originates from the lower corona of the donor star, and not from a region between the two stars. Maser emission enables a more direct estimate of the mean coronal magnetic field of the donor star, which we estimate to be 1–4 kG assuming a magnetic filling factor of 50%. A two-sample Kolmogorov-Smirnov test supports the conclusion that the distribution function of radio detected MCVs with orbital periods between 1.5–5 hours is similar to that of all MCVs. This result implies that rapidly-rotating (P$_{spin} < 10$ days), fully convective stars can sustain strong magnetic dynamos. These results support the model of @taam89 that the change in angular momentum loss across the fully convective boundary at P$_{orb} \approx 3$ hours is due to a change in the magnetic field structure of the donor star from a low-order to high-order multipolar field.'
address:
- 'George Washington University, Washington DC, 20052, USA'
- 'United States Naval Observatory, Washington DC 20392, USA'
- 'National Radio Astronomy Observatory, Charlottesville, VA 22903, USA'
- 'New Mexico State University, Las Cruces, NM, 88003, USA'
- 'Picture Rocks Observatory, 1025 S. Solano, Suite D, Las Cruces, NM 88001, USA'
- 'Indian Institute of Science Education and Research Mohali, Manauli, PO 140306, India'
author:
- Paul Barrett
- Christopher Dieck
- 'Anthony J. Beasley'
- 'Paul A. Mason'
- 'Kulinder P. Singh'
title: Radio Observations of Magnetic Cataclysmic Variables
---
cataclysmic variables – radio continuum: stars - stars: activity – stars: magnetic fields
Introduction
============
A project was initiated early in 2013 using the NRAO Very Long Baseline Array (VLBA) to perform astrometry of six radio-bright ($>1$ mJy) magnetic cataclysmic variables (MCVs) (namely, BG CMi, AM Her, DQ Her, ST LMi, GK Per, & AR UMa) in an attempt to accurately measure their distances. These observations were carried out in Semesters 13B and 14A. Astrometry requires at least four observations at roughly three month intervals with preferably two observations being at quadrature for best results. Unfortunately, source variability resulted in less than four successful observations per star. As a result, this project was only partially successful. The results of these observations are in preparation. At the same time, a survey was begun using the NSF’s Karl G. Jansky Very Large Array (VLA) to identify new radio-bright MCVs for follow-up VLBA astrometry. Previous to the current study, only eight MCVs were known radio sources; four polars (V834 Cen, AM Her, ST LMi, & AR UMa) and four Intermediate Polars (IPs; AE Aqr, BG CMi, DQ Her, & GK Per). We note that contemporaneous with this study, @copp15 [@copp16] observed nine nearby dwarf novae and nova-like CVs and detected eight of them at a flux level of a few tens of $\mu$Jy. The success of the first VLA survey in 2013 was motivation to expand this project to encompass all known MCVs.
The present paper is a status report of this project as of July 2018. Section 2 is a summary of the observations and data analysis of the surveys that have been performed since 2013, with particular emphasis on those during VLA Semesters 13B, 15A and 18A. Section 3 is a summary of the results of these three surveys, including the list of detected sources. The observations and results of the complete survey, including detections and non-detections, will be published in a separate paper. In Section 4, we primarily focus on the circular polarized emission from these sources. We discuss the most likely site of this emission and its implication for the donor star’s mean coronal magnetic field and CV evolution. Our conclusions are given in Section 5.
Observations and Data Analysis
==============================
For our initial set of VLA observations, we observed 111 MCVs during two observing semesters, 13B and 15A, at primarily three frequencies (C-, X-, and K-bands; 4–6, 8–10, and 20–22 GHz, respectively) at full polarization. A fourth frequency, Q-band (40–44 GHz), was also used for a few observations during Semester 13B. No detections were made at this frequency. For semester 13B, 40 hours of observing time was requested to observe 60 of the optically ([*B*]{} magnitude) brightest, and likely closest, MCVs from the @ritt03 *Catalog of cataclysmic binaries, low-mass X-ray binaries and related objects (Seventh edition);* (sp., ed. 7.19) that are north of declination $-40$[$^{\circ}$]{}. An equal number of targets were selected from the polar and IP subclasses in order to avoid biasing the sample toward the brighter IPs. The targets were also chosen to span a wide range of local sidereal time. The observing program allowed about three MCVs to be observed in each one hour scheduling block (SB). Exposures were approximately two minutes per frequency. Each SB was scheduled twice to increase the probability of detecting the MCVs. The VLA scheduled 35 of the 40 hour, resulting in 42 MCVs being observed. For semester 15A, 69 hour of observing time was requested to observe another 69 MCVs in the @ritt03 catalog. Except for one SB, each SB contained two MCVs with each exposure being approximately five minutes per frequency. The VLA scheduled all 69 hour resulting in an additional 69 MCVs being observed. See @barr17 for a detailed discussion of the data analysis and results of these observations.
We have also performed several additional surveys. For Semester 16B, we observed AM Her and AR UMa for 21 hours at C-, X-, & K-bands. An additional 56 hours of time were used to observe another set of 56 MCVs during semesters 17B and 18A; 14 hours during Semester 17B, and 42 hours during Semester 18A. The latter two surveys were only observed using the X-band, which enables longer ($\approx 20$ minutes) exposures on each source. In summary, for the surveys to date, we have obtained 195 out of 199 hours by performing 1 hour filler observations. The data presented in this paper are from Semesters 13B, 15A, and 18A. The data from Semesters 13B and 15A are presented in @barr17, whereas the data from Semester 18A are new. The remaining observations will appear in a series of forthcoming papers.
As discussed in @barr17, all data are calibrated using the VLA CASA automated calibration pipeline. The data from Semesters 13B and 15A use version 4.2.2, and Semester 18A uses version 5.1.2. The clean (ver. 4.2.2) and tclean (ver 5.1.2) algorithms are then used for source detection and flux measurements. Six polarization or Stokes images (sp., I, Q, U, V, RR, and LL, where LL and RR are the left and right polarization channels, resp.) are created for each observation and target. The cleaning is performed in two steps: one for the I, Q, U, and V planes; and one for the RR and LL planes. The Stokes I image is usually best for detecting weakly polarized sources; while the Stokes RR and LL images, for strongly polarized sources. (Henceforth, circular polarization is implied when referring to polarization. No sources show linear polarization.) The size of the image depends on the VLA configuration and observing band. A typical image size for the X band is $10 \times 10$ arcseconds. The clean algorithm uses natural weighting, because all sources are assumed to be point sources. The cleaned (model) and residual images are used to measure the source flux and the standard deviation of the noise, respectively. The polarization and its error are calculated from the cleaned and residual polarization images.
Results
=======
Radio Detections
----------------
Table 1 is an abbreviated list of detections of 33 MCVS as of July 2018, i.e., from observing semesters 13B, 15A, and 18A. A complete list of detections and upper limits will be given in a forthcoming paper. Columns 1-4 are respectively the GVCS name, MCV subclass, observing semester, and frequency band. The C, X, and K-band frequencies for Semesters 13B and 15A are 4.464–6.512, 7.964–10.012, and 20.060–22.108 GHz; and the X-band frequencies for Semesters 18A are 7.928–12.024 GHz. Columns 5–7 are the Stokes I, and the RR and LL polarization fluxes in $\mu$Jy. The fluxes of the remaining Stokes parameters (Q, U, & V) are omitted. Column 8 is the percentage of circular polarization. Column 9 is the total signal-to-noise (S/N) of the detection. It is the product of the S/N of the Stokes I flux, the circular polarization fluxes, and the PSF-normalized difference of the observed and expected source positions. For highly polarized sources, the S/N of the polarized flux can be much greater than the Stokes I flux and therefore be the major factor in the total S/N. The algorithm for the total S/N and the probability of source misidentification is described by @barr17. Except for a few sources, most of the observed positions are within a few tenths of an arcsecond of the Gaia position [@gaia16; @gaia18]. For sources having multiple detections at a particular frequency, only the greatest flux is listed. Because these radio sources are highly variable, it was not unusual for the source to be detected in only one of the two observations.
------------------------- ------ ------- ------ ------ ------------ ------ ---- ------ --------- ------ ----- ------ ----- ---------
Name Type Sem- Band I [Flux]{} S/N$^1$
ester
EQ Cet AM 15A K 0.78 0.30 96 30 111 55 0 55 +100 70 11.2
Cas 1 IP 13B C 1.87 2.50 21 10 0 33 117 32 -100 39 9.7
FL Cet AM 18A X 0.17 0.25 11 5 24 9 0 9 +100 53 6.6
BS Tri AM 15A X 0.07 1.00 57 9 49 16 0 15 +100 45 23.3
EF Eri AM 13B X 0.32 1.20 87 15 0 30 135 31 -100 30 25.8
UZ For AM 15A C 1.46 2.50 78 9 0 25 85 26 -100 42 39.6
Tau 4 AM? 15A X 0.22 1.00 105 32 0 72 150 74 -100 69 8.0
LW Cam AM 18A X 0.06 0.30 50 4 0 6 98 6 -100 9 $>$99.9
VV Pup AM 13B X 0.04 0.95 79 14 82 25 0 24 +100 42 20.4
13B K 0.36 0.45 49 29 103 60 47 61 +37 57 4.5
FR Lyn AM 18A X 0.13 0.22 28 4 12 6 38 6 -52 17 43.3
Hya 1 AM 18A X 0.04 0.23 6 6 0 8 29 8 -100 39 7.3
HS0922+1333$^{\dagger}$ AM 18A X 0.03 0.22 8 5 5 7 3 7 +25 100 4.5
WX LMi AM 15A C 0.06 0.38 73 12 49 17 26 19 +31 34 22.8
15A X 0.07 0.27 52 14 33 17 48 17 -19 30 13.4
ST LMi$^*$ AM 13B X 0.08 0.70 153 12 0 30 221 31 -100 20 95.2
AR UMa$^*$ AM 13B C 0.23 1.30 489 16 492 25 428 28 +7 4 $>$99.9
13B X 0.19 0.78 432 14 439 24 396 24 +5 4 $>$99.9
13B K 0.16 0.34 317 30 252 45 255 45 -1 13 79.4
EU UMa AM 18A X 0.06 0.22 39 5 34 6 33 7 +1 14 53.1
V1043 Cen AM 18A X 0.16 0.60 20 5 0 7 20 8 -100 53 11.2
J1503-2207 AM 18A X 0.03 0.45 29 5 66 7 0 6 +100 14 55.5
BM CrB AM 13B X 0.06 0.81 43 15 0 24 83 24 -100 41 10.3
MR Ser AM 13B C 0.12 1.35 239 17 371 27 0 32 +100 11 $>$99.9
13B X 0.06 0.78 116 15 221 24 0 24 +100 15 65.2
MQ Dra AM 18A X 0.20 0.25 17 4 25 6 0 6 +100 34 17.4
AP CrB AM 18A X 0.11 0.30 24 4 18 6 20 6 -5 22 27.0
Her 1 AM 15A K 1.27 0.42 48 17 106 33 0 34 +100 45 14.2
V1007 Her AM 15A X 2.72 1.20 38 9 75 15 0 16 +100 29 23.3
V1323 Her IP 15A C 2.16 1.90 43 9 80 12 0 11 +100 20 32.2
15A X 0.49 1.20 23 9 0 15 53 15 -100 40 10.0
AM Her$^*$ AM 13B C 0.27 1.60 88 5 57 9 84 9 -19 9 $>$99.9
13B X 0.13 0.92 192 12 176 17 171 17 +1 7 $>$99.9
13B K 0.22 0.61 476 83 172 122 440 109 -44 27 24.7
V603 Aql$^*$ SH 13B C 0.13 2.90 22 3 15 6 0 6 +100 29 35.4
13B X 0.16 1.70 32 7 35 12 0 12 +100 59 13.2
V1432 Aql AM 18A X 0.04 0.26 15 5 21 7 0 7 +100 47 9.3
J1955+0045 AM 18A X 0.09 0.28 79 5 70 6 73 7 -2 6 $>$99.9
QQ Vul AM 13B K 0.10 0.50 92 39 0 56 134 58 -100 60 36.9
AE Aqr$^*$ IP 15A C 0.01 1.39 5123 9 5048 20 5074 20 0 1 $>$99.9
15A X 0.01 0.88 5497 11 5462 24 5425 26 0 1 $>$99.9
15A K 0.03 0.40 8031 34 7924 54 7881 53 0 1 $>$99.9
HU Aqr AM 18A X 0.25 0.23 44 13 16 13 64 13 -60 23 18.4
V388 Peg AM 18A X 0.04 0.23 34 5 73 6 0 6 +100 12 86.8
------------------------- ------ ------- ------ ------ ------------ ------ ---- ------ --------- ------ ----- ------ ----- ---------
: List of radio detected MCVs. $^1$: The total S/N is the product of the S/N of the relative distance, flux, and polarized flux (see Section 3.1). $^*$: Previously known radio source. $^{\dagger}$: S/N is the product of two detections.
The detection rate from the shorter (2 and 5 minute) exposures in Semesters 13B and 15A is about 16% (19 of 122 stars). It increases to about 33% (14 of 42 stars) for the longer 20 minute exposures in Semester 18A. The detection rate at the C-, X-, and K-band frequencies are 8% (9 of 122 stars), 19% (30 of 142 stars), and 6% (7 of 122 stars), respectively. The fraction of stars showing high polarization in at least one observation or exposure is $\approx 73$% (24 of 33 stars). The number of stars showing little to no ($<10$%) polarization is about 24% (8 of 33) and one star (V1043 Cen) shows high polarization during the first detection and no polarization during the second.
Discussion
==========
Unpolarized Emission
--------------------
Fourteen of the radio detections show unpolarized continuum emission with a very high brightness temperature. Such emission is characteristic of an incoherent radiative process in an optically thick plasma. The radiation is believed to be mostly gyrosynchrotron emission by mildly relativistic ($1 < \gamma < 5$) electrons gyrating in a magnetic field. We believe that the source of the few tens of keV to few MeV electrons are magnetic reconnection events in the region between the two stars or near the surface of the late-type donor star.
{width="6cm"} {width="6cm"}
Polarized Emission
------------------
A surprising result of this survey is that approximately three quarters of the stars show highly polarized emission of short duration ($\sim$few minutes). An example of such emission is the left and right polarized X-band images of the polar LW Cam (see Figure 1). The left image shows an obvious point source, whereas the right does not. High circular polarization requires the presence of a strong magnetic field (B $> 100$ G) and, therefore, limits the possible radiation mechanisms to gyromagnetic and electron cyclotron maser emission. (see, e.g., @dulk83 [@barr85; @fuer86; @benz89]. Gyromagnetic emission is often divided into three energy regimes, non-relativistic ($\gamma = 1$) cyclotron, mild-relativistic ($\approx 1 < \gamma < \approx 5$) gyrosynchrotron, and relativistic ($\gamma > 5$) synchrotron emission, where $\gamma = (1-(v/c)^2)^{-1/2}$. Each of these mechanisms has problems reproducing the characteristics of the observed emission, i.e., high polarization and short duration (see below). Synchrotron radiation can be eliminated because its generates a broad continuous spectrum and polarization of $< 40$% [@bjor19]. Cyclotron and gyrosynchrotron emission are also unlikely because very specific conditions are needed to produce high polarization. Specifically, the radiation must be emitted along the magnetic field in a relatively homogeneous plasma [@barr85]. Any curvature of the magnetic field and changes in plasma density and temperature will significantly decrease the percentage of polarization. Therefore, the most likely cause of the polarized emission is an electron cyclotron maser. Theoretically, the characteristics of maser emission in a constant magnetic field are high polarization ($>80$%), short timescales ($<600$ s), and narrowband emission ($\Delta f/f < 0.01$, where f is frequency; see. e.g. @dulk83). The high polarization and short duration are both seen in the roughly ten minute observations of LW Cam and V603 Aql (see below). The absence of narrow band emission is probably due to emission from a broad range of magnetic fields and multiple emission regions.
One mechanism for producing the cyclotron maser is the loss-cone instability. In this scenario, a plasma of relativistic electrons with an isotropic velocity distribution is confined to a magnetic flux tube. The converging magnetic field creates a magnetic mirror causing most of the electrons to be reflected, while the highest energy electrons precipitate out. The loss cone anisotropy causes the index of refraction of the plasma to be negative. The magnetoionic modes of the plasma are then amplified with the extraordinary (x) mode typically growing much faster than the ordinary (o) mode. This mechanism results in highly polarized radiation. @dulk85 shows that the growth rate of the x-mode for a typical loss-cone distribution can be quite large ($\sim 10^7$ s$^{-1}$), which is equivalent to an amplification length of $\approx 10$ m.
{width="12cm"}
Figure 2 is a ten minute X-band (8–12 GHz) spectrum of the polar LW Cam. The blue and red points are the left and right polarization channels, respectively. Each flux value is the average of two adjacent 128 MHz spectral windows (256 MHz bandwidth total), and their flux centroids are within 0.3 arcseconds of the its Gaia position [@gaia18]. This averaging of the data reduces the noise, but smoothes the data, which may hide some narrow spectral features. However, the spectrum does suggest the presence of at least one narrowband ($<300$ MHz) emission feature at about 10.295 GHz, which we identify as a pronounced electron maser event. Assuming the maser emission is at the fundamental harmonic, the ambient magnetic field strength of the plasma is determined by the gyrofrequency: f$_{ce} = 2.8$ B MHz, where B is in Gauss. For this particular event, the magnetic field is $\approx$ 3.6 kG. We attribute the remaining polarized emission to be the sum of many smaller maser events with a range of magnetic field strengths.
Given the estimated strength of the magnetic field, there are two possible source locations: in the accretion column and near the surface of the donor star. First, suppose the emission is from the accretion column and the ambient magnetic field is due to the white dwarf. The magnetic field of the white dwarf is usually described by a dipole with $B = 2 \times 10^7 r_9^{-3}$ G, where $r_9$ is the radius of the WD in units of $10^9$ cm and the magnetic field of LW Cam is $\approx 20$ MG [@ferr15]. The emission then originates at a radius of about 18 WD radii, or about a third of the distance from the WD to the inner Lagrangian point at 46 WD radii. The latter radius assumes the WD and donor masses of 0.65 and 0.2 M$_{\odot}$, respectively, and an orbital period of 97.3 minutes. For the relevant equations, see equations 2.1b and 2.4c of @warn95. Within the accretion column, the electron number density varies as $n(r) = n_0 r_9^{-5/2}$ cm$^{-3}$, where $n_0 \sim 10^{16}$ cm$^{-3}$ at the base of the accretion column [@lamb73]. At 18 WD radii, the electron number density and plasma frequency are $7 n_{12}$ cm$^{-3}$ and 24 GHz, resp., where $n_{12} = 10^{12}$ cm$^{-3}$ and f$_{pe} = 9 n_{12}^{1/2}$ GHz. A cyclotron maser requires f$_{ce} >$ f$_{pe}$. A condition that is not met within the accretion stream for LW Cam at 8 GHz and is unlikely to be met for other polars at similar frequencies. However, this argument does not exclude maser emission from other low number density ($<10^{12}$ cm$^{-3}$) regions within the WD magnetosphere.
In addition, the non-magnetic CV V603 Aql (Nova Aquilae 1918) also shows highly polarized emission (see Table 1). The @ritt03 catalog classifies this CV as a possible IP due to a prominent photometric period, which explains its inclusion in our initial list of targets. However, the photometric period is now associated with a permanent superhump or precession of the accretion disk (see e.g., @muka05). If the WD has a magnetic field, it must be weak ($<1$ kG). Otherwise, the Alfvén radius would be at least a few WD radii (for a WD mass of 1.2 M$_{\odot}$ and an accretion rate of $1.3 \times 10^{17}$ g s$^{-1}$) and would noticeably truncate the inner edge of the accretion disk. The far ultraviolet study of V603 Aql by @sion15 shows that this is not the case. They do not use truncated accretion disk models to generate their synthetic spectra. The accretion disk is also unlikely to be the site of the polarized emission even when invoking a magneto-rotational instability in the disk in order to increase angular momentum transport out of the inner disk, because the magnetic field of the disk corona in such models is $<1$ kG [@scep19].
A more likely site of the polarized emission is in the lower corona of the donor star where the plasma density and frequency are low enough ($n_e <10^{12}$ cm$^{-3}$ and $f_{pe} < 320$ MHz, resp.; [@dulk83; @dulk85]) to allow cyclotron maser emission to escape. The emission site is probably near the footprints of one or more coronal loops with the source of the high energy electrons being from magnetic reconnection events in stellar flares. Although the mean coronal magnetic field of the Sun is low ($<$ few G), because its fractional surface coverage, or filling factor $f$, is very small ($<$1%), the magnetic fields in coronal loops can be $\sim$few kG [@livi06]. In the case of dMe stars, their mean coronal magnetic fields can be $\sim$few kG, because of their larger filling factors ($\sim$50%). For the dMe stars AU Mic, AD Leo, and EV Lac; @saar93 estimates mean fields of 2.3 kG, 2.6 kG, and 3.6 KG, respectively. For this survey, the product of the measured field strength in Table 2 and the estimated fillling factor, $fB$, can be used as a rough estimate of the donor star’s mean coronal magnetic field. The data imply mean coronal fields of 1–4 kG, assuming a filling factor of 50%. Higher field strengths are possible, since the current upper bound is constrained by the observing frequency.
Table 2 lists the measured luminosities and magnetic field strengths of the stars listed in Table 1. For stars with measurements at multiple frequencies, only the largest luminosity and highest field strength are listed. Columns 1-4 are the star’s name, the CV subclass, the orbital period in hours, and the star’s distance in parsecs from the Gaia Data Release 2. Column 5 is the flux value that is used to calculate the radio luminosity in column 6. Column 7 is the estimated magnetic field of the emitting region, which we attribute to maser emission. We only list the measured magnetic field strengths for stars having at least one measurement of high polarization. The field strength is calculated using the central frequency and bandwidth of the observation and therefore is only approximate. More accurate estimates of the ambient magnetic field are possible for the brighter stars via spectral analysis and may reveal the presence of multiple emission features and hence, multiple simultaneous emission regions. We intend to provide these results in a forthcoming paper. The range of luminosities is shown as a histogram in Figure 3. Except for AE Aqr, the highest luminosities are probably overestimates caused by inaccurate Gaia distances.
{width="12cm"}
------------- ------ -------- ---------- ----------- -------------------------- ------ -----
Name Type Period Distance Flux Luminosity$^*$
(h) (pc) ($\mu$Jy) ($10^{24}$ erg s$^{-1}$)
EQ Cet AM 1.5469 285 96 196 7530 365
Cas 1 IP 3.9396 1608 21 292 1960 365
FL Cet AM 1.4523 319 11 13 3563 730
BS Tri AM 1.6045 273 57 46 3210 365
EF Eri AM 1.3503 160 87 24 3210 365
UZ For AM 2.1087 240 78 24 1960 365
Tau 4 AM? 1.50 ? 210 105 50 3210 365
LW Cam AM 1.6211 542 50 175 3563 730
VV Pup AM 1.6739 137 79 16 7530 365
FR Lyn AM 1.8876 550 28 101
Hya 1 AM 2.3966 441 6 14 3563 730
HS0922+1333 AM 4.0394 163 8 3
WX LMi AM 2.7821 99 73 4
ST LMi AM 1.8981 113 153 21 3210 365
AR UMa AM 1.9320 101 489 27
EU UMa AM 1.5024 331 39 51
V1043 Cen AM 4.1902 173 20 7 3563 730
J1503-2207 AM 2.2228 392 29 53 3563 730
BM CrB AM 1.4040 419 43 81 3210 365
MR Ser AM 1.8911 132 239 22 3210 365
MQ Dra AM 4.3912 186 17 7 3563 730
AP CrB AM 2.5310 209 24 13
Her 1 AM 3.00 ? 1030 48 1281 7530 365
V1007 Her AM 1.9988 462 38 87 3210 365
V1323 Her IP 4.4016 2244 43 1163 3210 365
AM Her AM 3.0942 88 476 93
V603 Aql SH 3.3168 313 32 34 3210 365
V1432 Aql AM 3.3656 462 15 38 3563 730
J1955+0045 AM 1.3932 171 79 28
QQ Vul AM 3.7084 317 92 233 7530 365
AE Aqr IP 9.8797 91 8031 1673
HU Aqr AM 2.0836 192 44 19
V388 Peg AM 3.3751 689 34 193 3563 730
------------- ------ -------- ---------- ----------- -------------------------- ------ -----
: List of radio luminosity and donor magnetic field strength. $*$ Luminosity assumes a 256 MHz bandwidth. $\dagger$ The magnetic field assumes cyclotron maser emission from the corona of the donor star at the fundamental gyrofrequency, n =1.
Under the assumption that the highly polarized radio emission is from the donor star, these observations have important implications for stellar dynamos in fully convective stars and for CV evolution.
Cataclysmic Variable Evolution
------------------------------
{width="12cm"}
{width="12cm"}
The standard model of CV evolution is determined by angular moment loss (AML; @rapp82 [@spru83]). For P$_{orb} > 3$ hours, magnetic braking is believed to dominate the AML, assuming the magnetic field of the donor stars is $> 500$ G. The paucity of CVs with P$_{orb}$ between 2 and 3 hours was noted by @whyt80 and is call the “period gap”. @robi81 suggested that the upper edge of the gap coincides with the orbital period when the donor star becomes fully convective. It has therefore been assumed that magnetic braking is disrupted at P$_{orb} \approx 3$ hours. At this point, the donor star shrinks and mass transfer ceases. Gravitational radiation then becomes the dominating AML. At about two hours, the orbital radius decreases sufficiently that the donor star overflows its Roche lobe and mass transfer begins again.
The argument for fully convective stars having a weak magnetic field is as follows. Donor stars above the period gap have a tachocline, which is the interface between the radiative core and convective envelope. The tachocline provides a seed magnetic field for the magnetic dynamo in the convective envelope [@spie92]. Simulations show that strong shearing occurs in the tachocline, causing the local magnetic field to be amplified [@char14], generating an $\alpha-\Omega$ dynamo, named for the interplay between the cyclonic eddies (the $\alpha$ effect) and the shearing of the field (the $\Omega$ effect). Isolated main-sequence stars later than spectral type M3–M3.5 (M $<3.5$ M$_{\odot}$) are predicted to be fully convective and do not possess a tachocline. Therefore, it was conjectured that they cannot sustain a strong dynamo, resulting in a weak magnetic field (B $< 100$ G).
Observationally, fully convective stars do display all aspects of stellar activity including optical variability, strong emission lines, and ultraviolet, X-ray, and radio emission (see e.g., @giam86 [@ster94; @lins95; @hodg95; @flem95; @delf98]). @pevt03 has shown that the X-ray activity of a late-type star is a good proxy for the surface magnetic flux, and @wrig16 have shown that the level of X-ray emission as a function of stellar rotation period is essentially the same for fully convective and partially convective stars. Therefore, these stars must be capable of generating significant magnetic fields. @kuke99 studied dynamos in fully convective pre-main-sequence stars using mean field theory. They found that a second order effect, called the $\alpha^2$ dynamo, could be excited for moderate rotation rates (P$_{spin} < 10$ d), giving rise to steady, non-axisymmetric mean fields. Local fields as high as 10 kG are generated in some simulations by @brow08.
Unlike previous studies that have indirectly inferred a strong magnetic field for the donor star using stellar activity, our observations provide direct measurements of the magnetic field in coronal loops which are used to estimate the mean coronal magnetic fields of the donor stars. We conclude that all polars, and by implication all CVs, have strong magnetic fields throughout their evolution as a result of their rapid rotation. This can be seen in Figure 4, where there is no decrease in the distribution function of radio detected MCVs for P$_{orb} < 3$ hour. A two sample Kolmogorov-Smirnov (K-S) test comparing the two orbital period distributions functions of the radio detected MCVs and all MCVs gives a statistic of 0.129 and a P-value of 0.71. The K-S test result supports the conclusion that the orbital period distribution of radio detected MCVs is the same as that of all MCVs. Figure 5 shows the distribution of the approximate highest observed magnetic field strength with orbital period. Although the data have a strong observational bias because of the small number of detections and the frequency bandwidth of the observations, there is no evidence of a decrease in magnetic field strength with orbital period. This suggests that the magnetic fields of the donor stars for P$_{orb} < 3$ hours are just as strong as they are above 3 hours. This conclusion is supported by recent work on dynamos of fully convective stars.
If the fully convective donor stars in CVs have strong magnetic fields, then another mechanism must be the cause of the CV period gap. One possibility is that there is a change in the structure of the donor star’s magnetic field. Dynamo theory has shown that partially convective stars typically have dipolar or low-order multipolar fields, while fully convective stars have high-order multipoles. @taam89 have shown low-order multipolar fields are more efficient at AML than high-order multipolar fields. @garr18b model CV evolution using the magnetic braking prescription of @garr18a and show that it replicates the period gap. However, additional observations and analysis of the radio data are needed to confirm @taam89’s model. We believe that this explanation is consistent with the study of @knig11 who have shown that the optimal scaling factors for an empirical model of AML above and below the period gap are: f$_{MB} = 0.66 \pm 0.05$ and f$_{GR} = 2.47 \pm 0.22$, respectively, where f$_{MB} = $f$_{GR} = 1$ for the standard model of magnetic braking (MB) and gravitational radiation (GR). A less efficient version of magnetic braking satisfies the need for an additional AML mechanism to that of gravitational radiation for P$_{orb} < 3$ hours.
Conclusions
===========
Radio observations of magnetic cataclysmic variables provide a new window for studying the radiation mechanisms and dynamics of these interacting binaries. Under the assumption that the highly circularly polarized emission is due to electron cyclotron maser emission from coronal loops and the donor stars have a high magnetic filling factor, we are able to directly estimate the mean coronal magnetic field strengths to be about 1–4 kG. Although our sample size is limited (33 sources), a two sample K-S test supports our conclusion that the distribution function of donor star magnetic fields is similar to that all MCVs. This result implies that rapidly rotating (P$_{spin} < 10$ d), fully convective stars can sustain a strong magnetic dynamo, and hence, strong coronal magnetic fields. It also suggests that magnetic braking is important throughout the evolution of MCVs, and by implication all CVs, and the change in AML across the fully convective boundary at P$_{orb} \approx 3$ hours is the result of change in magnetic field structure from low-order to high-order multipolar fields as proposed by @taam89.
Acknowledgements
================
The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc.
This work has made use of data from the European Space Agency (ESA) mission [*Gaia*]{} (https://www.cosmos.esa.int/gaia), processed by the [*Gaia*]{} Data Processing and Analysis Consortium (DPAC, https://www.cosmos.esa.int/web/gaia/dpac/consortium). Funding for the DPAC has been provided by national institutions, in particular the institutions participating in the [*Gaia*]{} Multilateral Agreement.
The authors wish to thank the editor for her perseverance during this project and to the two anonymous referees for their constructive comments that greatly improved the final version of this paper.
Barrett, P. E. & Chanmugam, G. 1985 ’Cyclotron lines in accreting magnetic white dwarfs with an application to VV Puppis’, ApJ, 298, 743–751
Barrett, P. E., Dieck, C., Beasley, A. J., Singh, K. P., & Mason, P. A. 2017, ’A Jansky VLA Survey of Magnetic Cataclysmic Variables. I. The Data’, AJ, 154, 252–260
Benz, A. O. & Guedel, M. 1989, ’VLA detection of radio emission from a dwarf nova’, A&A, 218, 137–140
Björnsson, C. I. 2019, ’Circular Polarization in Compact Radio Sources: Constraints on Particle Acceleration and Electron-Positron Pairs’, ApJ, 873, 55–67
Browning, M. K. 2008, ’Simulations of Dynamo Action in Fully Convective Stars’, ApJ, 676, 1262–1280
Charbonneau, P. 2014, ’Solar Dynamo Theory’, ARA&A, 52, 251–290
Coppejans, D. L., Körding, E. G., Miller-Jones, J. C. A., Rupen, M. P., Knigge, C., Sivakoff, G. R., & Groot, P. J. 2015, ’Novalike cataclysmic variables are significant radio emitters’, MNRAS, 451, 3801–3813
Coppejans, D. L., Körding, E. G., Miller-Jones, J. C. A., Rupen, M. P., Sivakoff, G. R., Knigge, C., Groot, P. J., Woudt, P. A., Waagen, E. O., & Templeton, M. 2016, ’Dwarf nova-type cataclysmic variable stars are significant radio emitters’, MNRAS, 463, 2229–2241
Delfosse, X., Forveille, T., Perrier, C. & Mayor, M. 1998, ’Rotation and chromospheric activity in field M dwarfs’, A&A, 331, 581–595
Dulk, G. A., Bastian, T. S., & Chanmugam, G. 1983, ’Radio emission from AM Herculis: the quiescent component and an outburst’, ApJ, 273, 249–254
Dulk. G. A. 1985, ’Radio emission from the sun and stars’, ARA&A, 23, 169–224
Fuerst, E., Benz, A., Hirth, W., Kiplinger, A. & Geffert, M. 1986, ’Radio emission of cataclysmic variable stars’ A&A, 154, 377–378
Ferrario, L., de Martino, D., & Gänsicke, B. 2015, ’Magnetic White Dwarfs’, SSRv, 191, 111–169
Fleming, T. A., Schmitt, J. H. M. M. & Giampapa, M. S. 1995, ApJ, ’Correlations of Coronal X-Ray Emission with Activity, Mass, and Age of Nearby K and M Dwarfs’, 450, 401–410
Gaia Collaboration; Prusti, T. *et al.* 2016, ’The Gaia mission’ A&A, 595, 1–36
Gaia Collaboration; Brown, A. G. A. *et al.* 2018, ’Gaia Data Release 2. Summary of the contents and survey properties’, A&A, 616, 1–22
Garraffo, C., Drake, J. J., Dotter, A., Choi, J., Burke, D. J., Moschou, S. P., Alvarado-Gomez, J. D., , Kashyap, V. L., & Cohen, O. 2018, ’The Revolution Revolution: Magnetic Morphology Driven Spin-Down’, ApJ, 862, 90–96
Garraffo, C., Drake, J. J., Alvarado-Gomez, S. P., Moschou, S. P. & Cohen, O. 2018, ’The Magnetic Nature of the Cataclysmic Variable Period Gap’, ApJ, 868, 60–65
Giampapa, M. S. & Liebert, J. 1986, ’High-Resolution H alpha Observations of M Dwarf Stars: Implications for Stellar Dynamo Models and Stellar Kinematic Properties at Faint Magnitudes’, ApJ, 305, 784–794
Hodgkin, S. T., Jameson, R. F. & Steele, I. A. 1995, ’Chromospheric and coronal activity of low-mass stars in the Pleiades’, MNRAS, 274, 869–883
Knigge, C., Baraffe, L. & Patterson, J. 2011, ’The Evolution of Cataclysmic Variables as Revealed by Their Donor Stars’, ApJS, 194, 28–75
Küker, M. & Rüdiger, G. 1999, ’Dynamos in Fully Convective Stars’, ASP Conf Series, 178, 87–96
Lamb, F. K., Pethick, C. J. & Pines, D. 1973, ’A Model for Compact X-Ray Sources: Accretion by Rotating Magnetic Stars’, ApJ, 184, 271–290
Linsky, J. L., Wood, B. E., Brown, A., Giampapa, M. S. & Ambruster, C. 1995, ’Stellar Activity at the End of the Main Sequence: GHRS Observations of the M8 Ve Star VB 10’, ApJ, 455, 670–676
Livingston, W., Harvey, J. W., Malanushenko, O. V., & Webster, L. 2006, ’Sunspots with the strongest magnetic fields’, Sol Phys, 329, 41–68
Mukai, K. & Orio, M. 2005, ’X-Ray Observations of the Bright Old Nova V603 Aquilae’, ApJ, 622, 602–612
Pevtsov, A. A., Fisher, G. H., Acton, L. W., Longcope, D. W., Johns-Kull, C. M., Kankelborg, C. C. & Metcalf, T. R. 2003, ’The Relationship Between X-Ray Radiance and Magnetic Flkux’, ApJ, 598, 1387–1391
Ritter, H. & Kolb, U. 2003, ’Catalogue of cataclysmic binaries, low-mass X-ray binaries and related objects (Seventh edition)’, A&A, 404, 301–303
Rappaport, S., Joss, P. C & Webbink, R. F. 1982, ’The evolution of highly compact binary stellar systems’, ApJ, 254, 616–640
Robinson, E. L., Barker, E. S., Cochran, A. L., Cochran, W. D. & Nather, R. E. 1981, ’MV Lyr: spectrophotometric properties of minimum light: or on MV Lyr off’, ApJ, 251, 611–619
Saar, S. H. 1993, ’New Infrared Measurements of Magnetic Fields of Cool Stars’, IAU Symp. 154, ed. D. M. Rabin, J. T. Jefferies, & C. Lindsey, Kluwer Academic Publishers; Dordrecht, 493
Scepi, N., Dubus, G. & Lesur, G. 2019, ’Magnetic wind-driven accretion in dwarf novae’, A&A, 626, 116–123
Sion, E. M., Godon, P. & Bisol, A. 2015, ’Far-ultraviolet Spectroscopy of Old Novae, I. V603 Aql’, AJ, 150, 36–42
Spiegel, E. A. & Zahn, J. P. 1992, ’The solar tachocline’, A&A, 265, 106–114
Spruit, H. C. & Ritter, H. 1983, ’Stellar activity and the period gap in cataclysmic variables’, A&A, 124, 267–272
Stern, R. A., Schmitt, J. H. M. M., Pye, J. P., Hodgkin, S. T., Stauffer, J. R. & Simon, T. 1994, ’Coronal X-Ray Sources in the Hyades: A 40 Kilosecond ROSAT Pointing’, ApJ, 427, 808–821
Taam, R. E. & Spruit, H. C. 1989, ’The Disrupted Magnetic Braking Hypothesis and the Period Gap of Cataclysmic Variables’, ApJ, 345, 972–977
Warner, B. 1995, ’Cataclysmic Variable Stars’, Cambridge Univ. Press; Cambridge, 27
Whyte, C. A. & Eggleton, P. P. (1980), ’Comments on the evolution and origin of cataclysmic binaries’, MNRAS, 190, 801–823
Wright, N. J. & Drake, J. J. 2016, ’Solar-type dynamo behaviour in fully convective stars without a tachocline’, Nature, 535, 526–528
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'A system evolving under the driven Jaynes–Cummings model will undergo a phase transition at a critical driving field amplitude. This transition is foreshadowed by a collapse of the quasienergy level spectra of the system and remains present as the model is extended to include a counter-rotating interaction. We study this critical response and obtain the eigenvalues and eigenstates of the extended model by presenting a correspondence between the Jaynes–Cummings model and a charged Dirac particle subject to an external electromagnetic field. Under this correspondence, the field and two-level system that compose the Jaynes–Cummings model map to the external and internal degrees of freedom describing the Dirac particle, respectively. The phases of the system below (above) the critical drive are then characterized by discrete (continuous) solutions, with the manipulations required to obtain these solutions appearing naturally as Lorentz transformations.'
address: 'The Dodd-Walls Centre for Photonic and Quantum Technologies, Department of Physics, University of Auckland, Private Bag 92019, Auckland, New Zealand'
author:
- 'R. Gutiérrez-Jáuregui and H. J. Carmichael$^{\dagger}$'
title: |
Quasienergy collapse in the\
driven Jaynes–Cummings–Rabi model:\
correspondence with a charged Dirac particle\
in an electromagnetic field
---
[*Keywords* ]{}: driven Jaynes–Cummings-Rabi model, Dirac equation, quantum phase transition
Introduction
============
The interaction between a two-level system and a harmonic oscillator is ubiquitous in modern physics. This universality allows us to extrapolate ideas and techniques across different physical backgrounds which—appealing to the developed intuition—may appear more natural on one field over another. In a quantum optical context, the harmonic oscillator and the two-level system compose the quintessential model for the quantum theory of radiation and its interaction with matter, with the former modelling a single electromagnetic field mode and the latter describing two discrete energy levels of a material system. The dynamics of the coupled system are described up to a very good approximation by the Jaynes–Cummings (JC) model if the coupling is dipolar in character and the rotating wave approximation is valid [@Jaynes_1963]. While deceptively simple, these dynamics allow for deep physical insight and conceptual understanding of quantum concepts, as exemplified in the experimental demonstration of the discrete nature of the electromagnetic field [@Meekhof_1996; @Brune_1996; @Hofheinz_2008].
The operational simplicity of the model seems to translate to the driven Jaynes–Cummings model. The driven JC model describes a two-level system interacting with a field mode that is being driven out of the ground state by an external coherent field. When the resonance frequency of the two-level transition, field mode frequency, and drive frequency share the same value, it is possible to move to an interaction picture where the dynamics are simplified. In this interaction picture, the evolution of the system is ruled by the time-independent Hamiltonian $$\label{back_1}
\mathcal{H}_{0} = i\hbar \lambda \left( \hat{a} \hat{\sigma}_{+} - \hat{a}^{\dagger} \hat{\sigma}_{-} \right) + \hbar \epsilon \left( \hat{a} + \hat{a}^{\dagger} \right) \, ,$$ with $\hat{a}^{\dagger}$ and $\hat{a}$ the creation and annihilation operators of the field mode, $\hat{\sigma}_{+}$ and $\hat{\sigma}_{-}$ the raising and lowering operators of the two-level system, and parameters $\lambda$, $\epsilon$ to determine the dipole coupling strength and external field amplitude, respectively. This model describes an idealized scenario where a phase transition of light is found [@Carmichael_2015; @Alsing_1992_2; @Alsing_1992_1], caused by the competition between coupling strength and driving field amplitude, with the former advocating for a definite excitation number while the latter advocates for a definite phase. The phase transition is characterized by a change in the level structure of the Hamiltonian, equation (\[back\_1\]), at the critical drive amplitude $$\label{back_2}
\epsilon_{0} = \lambda / 2 \, .$$ For driving amplitudes below $\epsilon_{0}$ the Hamiltonian displays a discrete level structure, with level spacings that collapse to zero when this critical value is reached; the level structure becomes continuous above it [@Alsing_1992_2].
The critical behaviour persists if the interaction between the two-level system and field mode is extended to include counter-rotating coupling [@Gutierrez_2017b]. In this extended model the dynamics in the interaction picture are ruled by the generalized Jaynes–Cummings–Rabi Hamiltonian $$\begin{aligned}
\label{back_3}
\mathcal{H}_{\eta} = i\hbar \lambda^{\prime} \left[ \left( \hat{a} + \eta \hat{a}^{\dagger} \right) \hat{\sigma}_{+} - \left( \hat{a}^{\dagger} + \eta \hat{a} \right) \hat{\sigma}_{-} \right]
+ \hbar \epsilon^{\prime} \left( \hat{a} + \hat{a}^{\dagger} \right) \, ,\end{aligned}$$ with $\eta$ a continuous parameter ranging from $\left[ 0 ,1 \right]$. The expected dynamics under this Hamiltonian can be realized using a cavity QED architecture [@Dimer_2007]. In this realization the rotating and counter-rotating interactions are controlled independently by promoting externally driven Raman transitions between a pair of ground states via a manifold of far-detuned excited states; the time-independent Hamiltonian, equation (\[back\_3\]), is realized in an appropriate interaction picture (see Sections IIIC and D of Reference [@Gutierrez_2017b]). Considering this open system perspective clarifies the origin of the photons injected into the field mode through the counter-rotating coupling. In fact, equations (\[back\_1\]) and (\[back\_3\]) are connected through a squeezing transformation, $$\label{back_6}
\mathcal{S}(z) \mathcal{H}_{0} \mathcal{S}^{\dagger}(z) = \mathcal{H}_{\eta} \, ,$$ with squeeze parameter given by $$\begin{aligned}
\label{back_7_t}
\cosh z = \lambda^{\prime}/ \lambda \, , \label{back_7_1} \\ \sinh z = \eta \lambda^{\prime} / \lambda \, , \label{back_7_2}\end{aligned}$$ which introduces the scaled driving field amplitude $$\label{back_7_3}
\epsilon^{\prime} = \epsilon ( \lambda^{\prime} / \lambda ) \left( 1 + \eta \right) \, .$$ The addition of the counter-rotating interaction not only affects the mode photon number, but also displaces the critical drive towards the value $$\label{back_4}
\epsilon^{\prime}_{\eta} = \lambda^{\prime} (1 + \eta) / 2 \, .$$ These transformations establish a simple connection between the on-resonance JC and Jaynes–Cummings–Rabi models. We emphasize, in particular, that they allow us to target the treatment in the following sections to the JC interaction without loss of generality.
Previously the quasienergies and eigenstates of the system below the critical amplitude have been obtained through a Bogoliubov transformation [@Alsing_1992_2; @Gutierrez_2017b]; this method ceases to be applicable, however, for the continuous spectrum above the critical amplitude. We overcome this problem in the following, where we develop a new method for obtaining the quasienergies and eigenstates, both below and above critical drive, for the driven Jaynes–Cummings model. The solutions of the generalized model are then readily available through the squeezing transformation \[equations (\[back\_6\])-(\[back\_7\_3\])\]. The method is based on the correspondence between this model and the dynamics of a Dirac particle subject to an appropriate electromagnetic field configuration. Once the correspondence is established, the Lorentz covariance of the Dirac equation will be exploited to reach the desired solutions.
Section \[secc:Dirac\] is devoted to the correspondence between the relativistic and quantum optical models. We first introduce the Dirac equation for a charged particle subject to an external electromagnetic field and then present a particular field configuration where the evolution of the Dirac particle mimics the system under the driven Jaynes–Cummings Hamiltonian. In Section \[secc:Lorentz\] we obtain the eigenvalues and eigenstates of the Dirac equation; by considering the relativistic invariants of the electromagnetic field, they are shown to display two distinct behaviours which are connected to particularly simple physical scenarios through Lorentz transformations. In Section \[secc:Jaynes\_cummings\] we return to the quantum optical model and review the solutions in this context, where the relation between Lorentz and squeezing transformations is discussed. Section \[secc:conclusion\] is left for conclusions.
Correspondence to a relativistic system {#secc:Dirac}
=======================================
The dynamics of a spin-1/2 particle of mass $m$ and charge $e$ in the presence of an external electromagnetic field are accurately described by the Dirac equation $$\label{corr_1}
\left[ \gamma^{\mu}\left( \hat{p}_{\mu} - \frac{e}{c}A_{\mu} \right) - mc \right] \psi_{D} = 0 \, ,$$ where $\hat{p}_{\mu}= i \hbar \partial_{\mu}$ is the four-momentum operator, $A_{\mu}$ the electromagnetic potential, and the covariant $\gamma$ matrices satisfy the anticommutation relation $$\label{corr_2}
\left\lbrace \gamma^{\mu}, \gamma^{\nu} \right\rbrace_{+} = 2 g^{\mu \nu} \, ,$$ with $g^{\mu \nu}$ the metric tensor. We adopt the metric tensor with signature $\left( 1 , -1, -1, -1 \right)$ and consider the symmetric representation of the Dirac matrices: $$\begin{aligned}
\label{corr_3}
& \gamma^{0} = \left( \begin{array}{cc}
0 & \mathbf{1}_{2 \times 2}\\
\mathbf{1}_{2 \times 2} & 0 \\
\end{array} \right) \, , \hspace{.025\linewidth}
\gamma^{i} = \left( \begin{array}{cc}
0 & \sigma^{i}\\
-\sigma^{i} & 0 \\
\end{array} \right) \, ,\end{aligned}$$ written in terms of the Pauli matrices $\sigma^{i}$.
Consider the scenario of a massless Dirac particle $(m=0)$ subject to constant and homogeneous electric, $\mathbf{E} = E {\mathbf{e}}_{1}$, and magnetic, $\mathbf{B} = - B {\mathbf{e}}_{3}$, fields. This field configuration is commonly used for measuring magnetoresistance and transport properties of electron gases; relevant examples are found in [@MacDonald_1983; @Lukose_2007] where the electron gas is effectively described by equation (\[corr\_1\]). The electromagnetic potential generating this field configuration is determined up to a gauge transformation, with the gauge commonly selected to match the boundary conditions of the problem at hand [@Champel_2007]. We select $$\label{corr_4}
A_{\mu} = \left(E x_{1}; 0 , - B x_{1}, 0 \right) \, ,$$ for which equation (\[corr\_1\]) becomes invariant under translations in the ${\mathbf{e}}_{2}$ and ${\mathbf{e}}_{3}$ directions giving rise to the conserved quantities $\hat{p}_{2}$ and $\hat{p}_{3}$. Solutions to equation (\[corr\_1\]) can be written in terms of $\phi^{+}$ and $\phi^{-}$ spinors through $$\label{corr_5}
\psi_{D} = \left( \begin{array}{c}
\phi^{+}_{D}\\
\phi^{-}_{D} \\
\end{array} \right) \, ,$$ which, for $m=0$, satisfy the uncoupled equations $$\begin{aligned}
\left\lbrace i \hbar \partial_{t} - e E x_{1} + c \sigma_{1}\hat{p}_{1} + \sigma_{2}eBx_{1} + c \sigma_{j} \hat{p}_{j} \right\rbrace \phi^{+}_{D} =0 \, , \label{corr_6a} \\
\left\lbrace i \hbar \partial_{t} - e E x_{1} - c \sigma_{1}\hat{p}_{1} - \sigma_{2}eBx_{1} -c \sigma_{j} \hat{p}_{j} \right\rbrace \phi^{-}_{D} =0 \, , \label{corr_6b}\end{aligned}$$ with $j=2,3$ and the Einstein summation convention: $\sigma_{j} \hat{p}_ {j} = \sigma_{2} \hat{p}_{2} + \sigma_{3} \hat{p}_{3}$ is understood. The potential generated through the interaction with the magnetic field allows the position and momentum operators to be written in terms of bosonic operators, $$\label{corr_7}
\hat{a} = \frac{x_{1} + i(l^{2}_{\mathcal{B}}/\hbar)\hat{p}_{1}}{\sqrt{2} l_{\mathcal{B}}},\hspace{.025\linewidth} \hat{a}^{\dagger} = \frac{x_{1} - i(l^{2}_{\mathcal{B}}/\hbar)\hat{p}_{1}}{\sqrt{2} l_{\mathcal{B}}} \, ,$$ with the magnetic length $$\label{corr_8}
l_{\mathcal{B}} = \sqrt{\frac{\hbar c}{e B}} \,$$ providing a natural unit to measure the depth of the potential. Equations (\[corr\_6a\]) and (\[corr\_6b\]) take the form $$\label{corr_9}
\lbrace \mp i \hbar \partial_{t} - c \sigma_{j} \hat{p}_{j} \rbrace \phi^{\pm}_{D} = \left\lbrace i\sqrt{2}eB l_{\mathcal{B}} \left( \hat{a} \sigma_{+} - \hat{a}^{\dagger} \sigma_{-} \right) \mp \frac{eE l_{\mathcal{B}}}{ \sqrt{2}} \left( \hat{a} + \hat{a}^{\dagger} \right) \right\rbrace \phi^{\pm}_{D} \,$$ under these definitions. Since the eigenvalues of $\hat{p}_{2}$ and $\hat{p}_{3}$ remain constant through the evolution, we can consider the trivial scenario where they are zero, in which case the correspondence between equation (\[corr\_9\]) and the evolution under equation (\[back\_1\]) becomes transparent. The motion and spin state of the Dirac particle will map to the field mode and two-level system, respectively, in the quantum optical analogue. A similar correspondence between the undriven Jaynes–Cummings model and the Dirac oscillator was found in Ref. [@Rozmej_1999].
Eigenstates and eigenvalues: Quasienergy collapse {#secc:Lorentz}
=================================================
A method that has proven useful to determine the solutions of the Dirac equation in the presence of electromagnetic fields, equation (\[corr\_1\]), relies on considering the Dirac-Pauli equation as an auxiliary equation [@Landau_1971]. This allows us to extract a basis set for the spinors while giving a differential equation for the eigenfunctions to satisfy [@Gutierrez_2017]. For a massless particle ($m=0$), the Dirac-Pauli equation takes the simplified form $$\label{eige_1}
\left[ \left(\hat{p}^{\mu} - \frac{e}{c} A^{\mu} \right) \left(\hat{p}_{\mu} - \frac{e}{c} A_{\mu} \right) -\frac{i e \hbar}{4c} \left[ \gamma^{\mu}, \gamma^{\nu} \right]_{-} F_{\mu \nu} \right] \psi_{P} = 0 \, ,$$ with $$\label{eige_2}
F_{\mu \nu} = \partial_{\mu} A_{\nu} - \partial_{\nu} A_{\mu} \,$$ the electromagnetic tensor. Solutions $\psi_{D}$ to the Dirac equation are then built from solutions $\psi_{P}$ of this auxiliary second-order equation by applying the Dirac operator: $$\label{eige_0}
\psi_{D} = \gamma^{\mu}\left( \hat{p}_{\mu} - \frac{e}{c}A_{\mu} \right) \psi_{P} \, .$$ For the physical scenario presented above, the relativistic invariants $$\begin{aligned}
\label{eige_3}
F_{\mu \nu} F^{\mu \nu} = \mathbf{B} \cdot \mathbf{B} - \mathbf{E} \cdot \mathbf{E} \, , \\
\epsilon_{\mu \nu \rho \eta} F^{\mu \nu} F^{\rho \eta} = 2 \mathbf{E} \cdot \mathbf{B} \, ,\end{aligned}$$ will allow us to reduce the solutions of the system to only two cases, which correspond to the possible phases of light encountered in the quantum optical system. Given that when $F_{\mu \nu}F^{\mu \nu}$ is positive (negative) the existence of a reference frame where the electric (magnetic) field vanishes is guaranteed, the exact solution of the Dirac equation can be obtained in this privileged frame. Afterwards, by applying a Lorentz transformation, the solutions in the laboratory frame are found.
The boost required to reach the privileged frame in both scenarios must be taken in the $\mathbf{e}_{2}$ direction. The eigenvalues of $\hat{p}_{3}$ then keep the same constant value in both reference frames. Since we are interested in the solutions to the driven Jaynes–Cummings model on resonance, this operator can be considered zero throughout the following derivation without affecting the final result. Notice, however, that non-zero eigenvalues of $\hat{p}_{3}$ correspond to a non-zero detuning between the two-level system and field mode in the optical analogy, through equation (\[corr\_9\]). We then include this operator for completeness and adopt the notation $k_{\perp} x_{\perp} = k_{2} x_{2} + k_{3} x_{3}$ where required.
Below critical drive: Discrete spectrum
---------------------------------------
In the case where $\vert B \vert \geq \vert E \vert$ a boost in the $\mathbf{e}_{2}$ direction with parameter $$\label{eige_4}
\beta_{\mathcal{B}} = \frac{E}{B} \,$$ takes us to a reference frame where the electric field vanishes. Since the Dirac-Pauli equation is second-order in time, both Pauli spinors satisfy $$\label{eige_5}
\left[ \hat{p}_{0}^{\prime 2} - \hat{p}_{1}^{\prime 2} - \left( \hat{p}^{\prime}_{2} + \frac{eB'}{c} x^{\prime}_{1} \right)^{2} - \hat{p}^{\prime 2}_{3} - \frac{e\hbar B'}{c} \sigma_{3} \right] \phi_{P}^{\prime} = 0 \,$$ in this privileged frame. The basis set for the eigenspinors is then given by the eigenvectors of $\sigma_{3}$; the orientation of each is responsible for the addition of a constant in the differential equation, leaving the overall structure unaffected. Thus the differential equation is separable in all four components, with solutions given by the product of free particle states for $x^{\prime}_{2}$ and $x^{\prime}_{3}$ and eigenstates of the displaced harmonic oscillator, $H_{n}$, for $x^{\prime}_{1}$. The explicit expressions for the eigenspinors are $$\begin{aligned}
\phi_{P \uparrow}^{\prime} &= e^{i k_{\perp}' x_{\perp}' } \left( \begin{array}{c}
H_{\vert n \vert} \left( x_{1}'/l_{\mathcal{B}}'+ l_{\mathcal{B}}' k_{2}' \right) \\
0 \\
\end{array} \right) \, , \label{eige_6a}\\
\phi_{P \downarrow}^{\prime} &= e^{i k_{\perp}' x_{\perp}' } \left( \begin{array}{c}
0 \\
H_{\vert n \vert +1} \left( x_{1}'/l_{\mathcal{B}}'+ l_{\mathcal{B}}' k_{2}' \right) \\
\end{array} \right) \, , \label{eige_6b}\end{aligned}$$ with the corresponding Dirac particle energies given by $$\label{eige_7}
\upvarepsilon_{n,\pm}^{ \prime } = \pm\left( \frac{\hbar c} { l_{\mathcal{B}}'} \right) \sqrt{2 \vert n \vert + 2 + (l_{\mathcal{B}}' k^{\prime}_{3} )^{2}} \, ,$$ the eigenvalues of $\hat{p}_{0}^{\prime}$. The quantized motion described by these solutions is caused by the harmonic potential created through the interaction between particle spin and magnetic field. The Dirac spinors are obtained using equation (\[eige\_0\]) and will be given by a linear combination of Pauli spinors: $$\begin{aligned}
\phi_{D}^{+ \prime} & = e^{i k_{\perp}' x_{\perp}'} \left( \begin{array}{c}
i H_{\vert n \vert} \left( x_{1}'/l_{\mathcal{B}}'+ l_{\mathcal{B}}' k_{2}' \right) \\
c^{\prime}_{\pm} H_{\vert n \vert +1 } \left( x_{1}'/l_{\mathcal{B}}'+ l_{\mathcal{B}}' k_{2}' \right) \\
\end{array} \right) \, , \label{eige_8a}\\
\phi_{D}^{- \prime} & = e^{i k_{\perp}' x_{\perp}' } \left( \begin{array}{c}
c^{\prime}_{\pm} H_{\vert n \vert} \left( x_{1}'/l_{\mathcal{B}}'+ l_{\mathcal{B}}' k_{2}' \right) \\
i H_{\vert n \vert +1 } \left( x_{1}'/l_{\mathcal{B}}'+ l_{\mathcal{B}}' k_{2}' \right) \\
\end{array} \right) \, , \label{eige_8b}\end{aligned}$$ with the relative weights depending on the energy and momentum through $$\label{eige_9}
c^{\prime}_{\pm} = \pm \sqrt{\frac{\upvarepsilon_{n,\pm}^{\prime} + \hbar c k_{3}^{\prime}}{\upvarepsilon_{n,\pm}^{\prime} - \hbar c k_{3}^{\prime}}} \, .$$ Finally, the solutions are transformed back to the laboratory frame, where the energies are found to be $$\label{eige_10}
\upvarepsilon_{n, \pm}= \hbar c \left[ \beta_{\mathcal{B}} k_{2} \pm \sqrt{\frac{\left(1 - \beta_{\mathcal{B}}^{2} \right)^{3/2}}{l^{2}_{\mathcal{B}}} \left(2 \vert n \vert + 2 \right) + k_{3}^{2}} \right] ,$$ while the eigenstates are \[eige\_11t\] $$\begin{aligned}
\phi_{D}^{+} &= e^{i k_{\perp} x_{\perp}} \left( \begin{array}{c} i \Xi_{\mathcal{B}}^{+} H_{\vert n \vert} (z) - ic_{\pm} \Xi_{\mathcal{B}}^{-} H_{\vert n \vert +1} (z) \\
c_{\pm} \Xi_{\mathcal{B}}^{+} H_{\vert n \vert +1 } (z) - \Xi_{\mathcal{B}}^{+} H_{\vert n \vert} (z) \\
\end{array} \right) , \label{eige_11a}\\
\phi_{D}^{-} &= e^{i k_{\perp} x_{\perp}} \left( \begin{array}{c} \Xi_{\mathcal{B}}^{-} H_{\vert n \vert +1} (z) -
\Xi_{\mathcal{B}}^{+} c_{\pm} H_{\vert n \vert} (z) \\
i \Xi_{\mathcal{B}}^{-} c_{\pm} H_{\vert n \vert} (z) - i \Xi_{\mathcal{B}}^{+} H_{\vert n \vert +1 } (z) \\
\end{array} \right) , \label{eige_11b}\end{aligned}$$ with the displaced variable $$\begin{aligned}
\label{eige_12}
z = (1 & -\beta_{\mathcal{B}}^{2})^{1/4} \left(x_{1}/l_{\mathcal{B}} + l_{\mathcal{B}} k_{2} \right) \nonumber \\ &\pm \beta_{\mathcal{B}} \sqrt{2 \vert n \vert + 2 + \left(1-\beta^{2}_{\mathcal{B}} \right)^{-3/2} (l_{\mathcal{B}} k_{3})^{2}} ,\end{aligned}$$ and relative weights given by transforming equation (\[eige\_9\]) using equation (\[eige\_10\]) and defining the Lorentz parameter $$\label{eige_13}
\Xi^{\pm}_{\mathcal{B}} = \sqrt{1 \pm \left[1 - \beta_{\mathcal{B}}^{2} \right]^{1/2}} \, .$$ For a given $k_{2}$ and $k_{3}$, the energies of the particle display a $\sqrt{n}$-dependence that is maintained as we transform between reference frames; the separation between adjacent energies is, however, scaled by a factor accounting for dilation or contraction of the magnetic length.
Above critical drive: Continuous spectrum
-----------------------------------------
The same procedure can be used for $\vert E \vert \geq \vert B \vert$, where a boost in the $\mathbf{e}_{2}$ direction with parameter $$\label{eige_14}
\beta_{\mathcal{E}} = \frac{B}{E} \, ,$$ takes us to a reference frame where the magnetic field vanishes. The corresponding Dirac-Pauli equation then divides into two blocks that lead to the uncoupled equations $$\label{eige_15}
\left[\left(\hat{p}_{0}^{\prime} - \frac{e E^{\prime}}{c} x^{\prime}_{1} \right)^{2} - \hat{p}^{\prime 2}_{1} - \hat{p}^{\prime 2}_{2} - \hat{p}^{\prime 3}_{2} \mp i\frac{e \hbar E^{\prime}}{c} \sigma_{1} \right] \phi^{\pm \prime }_{P} = 0 \, ,$$ and the basis set for the eigenspinors is given by the eigenvectors of $\sigma_{1}$. While the general structure of the differential equations is similar to the previous case, there are key differences. Notice first that the energy has become a continuous parameter that displaces the origin of $x_{1}^{\prime}$. The eigenfunctions are still given by a product of free particle states for the $x_{2}^{\prime}$ and $x_{3}^{\prime}$ components, but the $x_{1}^{\prime}$ component satisfies Weber’s differential equation and, as such, the corresponding eigenfunctions are given by parabolic cylinder functions $D_{a}$ [@Abramowitz_1972]. The eigenspinors are $$\begin{aligned}
\phi_{P1}^{\pm \prime} &= e^{i k_{\perp}^{\prime} x_{\perp}^{\prime}} D_{-a-1}\left( \xi^{\prime} \right)\left( \begin{array}{c}
1 \\
\pm 1 \\
\end{array} \right) \, , \label{eige_16a} \\
\phi_{P2}^{\pm \prime} &= e^{i k_{\perp}^{\prime} x_{\perp}^{\prime}} D_{-a}\left( \xi^{\prime} \right)\left( \begin{array}{c}
1 \\
\mp 1 \\
\end{array} \right) \, , \label{eige_16b}\end{aligned}$$ where $$\begin{aligned}
l_{\mathcal{E}}^{\prime} & = \sqrt{\frac{\hbar c}{ e E^{\prime}}} \, , \\
\xi' &= \sqrt{2} e^{i \pi /4} \left( \upvarepsilon' l_{\mathcal{E}}^{\prime}/\hbar c - x_{1}^{\prime}/l_{\mathcal{E}}^{\prime} \right)\, , \label{eige_17}\\
a' &= -i l_{\mathcal{E}}^{\prime 2} \left( k_{2}^{\prime 2} + k_{3}^{\prime 2} \right) /2 \, . \label{eige_18}\end{aligned}$$ The Weber differential equation is related to the harmonic oscillator equation by flipping the sign of the potential, a change that leads to imaginary values in the definition of the Pauli spinors. The idea of an effective potential that is unbounded from below can be traced back to the classical dynamics, where a charged particle subject to a constant electric field would accelerate without boundary. The Dirac spinors corresponding to equations (\[eige\_16a\]) and (\[eige\_16b\]) are proportional to $$\begin{aligned}
\phi_{D}^{+ \prime} &= e^{i k^{\prime}_{\perp} x^{\prime}_{\perp}} \left( \begin{array}{c}
b_{+}^{\prime} D_{-a^{\prime}-1}(\xi^{\prime}) - D_{-a^{\prime}}(\xi^{\prime}) \\
b_{+}^{\prime} D_{-a^{\prime}-1}(\xi^{\prime}) + D_{-a^{\prime}}(\xi^{\prime}) \\
\end{array} \right) \, , \label{eige_19a} \\
\phi_{D}^{- \prime} &= e^{i k^{\prime}_{\perp} x^{\prime}_{\perp}} \left( \begin{array}{c}
D_{-a^{\prime}}(\xi^{\prime}) + b_{-}^{\prime} D_{-a^{\prime}-1}(\xi^{\prime}) \\
D_{-a^{\prime}}(\xi^{\prime}) - b_{-}^{\prime} D_{-a^{\prime}-1}(\xi^{\prime}) \\
\end{array} \right) \, , \label{eige_19b}\end{aligned}$$ with weights $$b_{\pm}^{'} = l^{\prime}_{\mathcal{E}} e^{i \pi /4} (k_{3}^{\prime} \pm i k^{\prime}_{2})/\sqrt{2} \, . \label{eige_20}$$
In the laboratory reference frame, the continuous spectrum is maintained and, up to a global phase factor, the eigenstates are given by $$\begin{aligned}
\label{eige_21t}
\phi_{D}^{+} &=e^{i k_{\perp} x_{\perp}} \left[ \Xi_{\mathcal{E}}^{+} + \Xi_{\mathcal{E}}^{-} \sigma_{2} \right] \left( \begin{array}{c}
b_{+} D_{-a-1}(\xi) - D_{-a}(\xi) \\
b_{+} D_{-a-1}(\xi) + D_{-a}(\xi) \\
\end{array} \right) \, , \label{eige_21a}\\
\phi_{D}^{-} &=e^{i k_{\perp} x_{\perp}} \left[ \Xi_{\mathcal{E}}^{+} - \Xi_{\mathcal{E}}^{-} \sigma_{2} \right] \left( \begin{array}{c}
D_{-a}(\xi) - b_{-}D_{-a-1}(\xi) \\
D_{-a}(\xi) + b_{-}D_{-a-1}(\xi) \\
\end{array} \right) \, , \label{eige_21b}\end{aligned}$$ with the transformed variables $$\begin{aligned}
& \xi = \sqrt{2} e^{i \pi /4} \left(1- \beta_{\mathcal{E}}^{2} \right)^{1/4} \left[ \frac{\left( \beta_{\mathcal{E}} \hbar c k_{2} - \upvarepsilon \right) l_{\mathcal{E}}}{ \hbar c \left(1 - \beta_{\mathcal{E}}^{2} \right) } - \frac{x_{1}}{l_{\mathcal{E}}} \right] \, , \label{eige_22} \\
& a = -\frac{i l_{\mathcal{E}}^{2}}{2} \left[ \frac{\left( k_{2} - \beta_{\mathcal{E}} \upvarepsilon / \hbar c \right)^{2}}{(1- \beta_{\mathcal{E}}^{2})^{3/2}} + \frac{k^{2}_{3}}{(1 - \beta_{\mathcal{E}}^{2})^{1/2}} \right] \, , \\
& \Xi^{\pm}_{\mathcal{E}} = \sqrt{1 \pm \left[1 - \beta_{\mathcal{E}}^{2} \right]^{1/2}} \, . \label{eige_23}\end{aligned}$$
Back to light {#secc:Jaynes_cummings}
=============
The quantized motion and discrete eigenvalues found when the magnetic field dominates \[equations (\[eige\_10\]) and (\[eige\_11a\])-(\[eige\_11b\])\] contrast the continuous behaviour encountered when the electric field dominates \[equations (\[eige\_21a\])-(\[eige\_21b\])\]. As mentioned in the introduction, in the simplified case where $k_{2}$ and $k_{3}$ both equal zero, the $\phi^{-}_{D}$ spinor of these two complementary cases corresponds to the eigenstates of the driven Jaynes–Cummings Hamiltonian below and above critical drive. A direct comparison between equation (\[back\_1\]) and (\[corr\_9\]) shows that the dipole coupling constant and driving amplitude relate to the electromagnetic fields through $$\begin{aligned}
\label{qpt_1}
\lambda &= \sqrt{2}e B l_{\mathcal{B}} \, , \\
\epsilon &= e E l_{\mathcal{B}}/\sqrt{2} \, ,\end{aligned}$$ such that the $F_{\mu \nu} F^{\mu \nu} =0$ condition refers to the transition point given in equation (\[back\_2\]). The inclusion of $l_{\mathcal{B}}$ in these definitions already hints at the importance of the magnetic field to the existence of discrete states. It is instructive to use the above relations to obtain the discrete quasi-energies of the driven JC model from equation (\[eige\_10\]): $$\upvarepsilon_{n, \pm}^{0} = \pm \hbar \lambda \left(1 - \frac{4 \epsilon^{2}}{\lambda^{2}} \right)^{3/4} \sqrt{n +1} \, .$$ As mentioned above, the same result has been obtained through an algebraic method by Alsing, Guo and Carmichael (see equation (50) of Ref. [@Alsing_1992_2]) where the effect of the external driving field is interpreted as a dynamic Stark shift of the JC quasienergies. This can be extended to include counter-rotating interactions through equation (\[back\_4\]), leading to $$\upvarepsilon_{n, \pm}^{\eta} = \pm \hbar \lambda \left(1 - \frac{4 \epsilon^{2}}{(1+\eta)^{2}\lambda^{2}} \right)^{3/4} \sqrt{n +1} \, ,$$ and a displaced transition point. In Figure \[fig:collapse\], the discrete quasienergies of the driven JC model are displayed as a function of the driving amplitude. The solutions were obtained by computing the eigenvalues of $\mathcal{H}_{0}$ in a truncated basis and show the characteristic $\sqrt{n}$-dependence and ensuing collapse when the critical point is reached.
![Quasi-energy level splitting and collapse at the critical point.[]{data-label="fig:collapse"}](tikz_collapse_2){width=".5\linewidth"}
Besides the collapse of the level structure of the driven JC Hamiltonian, the behaviour of the zero quasienergy state captures the nature of the transition at hand. Below the critical drive amplitude, the corresponding eigenstate in the position representation is $$\label{qpt_2}
\phi_{D}^{-} = \exp\left[- \frac{\sqrt{1 - \beta_{\mathcal{B}}^{2}} }{2 l_{\mathcal{B}}^{2}} x_{1}^{2} \right] \left( \begin{array}{c}
\Xi_{\mathcal{B}}^{-} \\
-i \Xi_{\mathcal{B}}^{+} \\
\end{array} \right) \, .$$ This separable product of field-mode and two-level system states is normalizable. The coherent drive determines the polarization of the two-level system through $$\label{qpt_3}
\langle \hat{\sigma}_{-} \rangle = i \beta_{\mathcal{B}} = i 2\epsilon/\lambda \, .$$ On the other hand, above critical drive, the eigenstate is $$\label{qpt_4}
\phi_{D}^{-} = \exp\left[- \frac{i \sqrt{1 - \beta_{\mathcal{E}}^{2}} }{2 l_{\mathcal{E}}^{2}} x_{1}^{2} \right] \left( \begin{array}{c}
\Xi_{\mathcal{E}}^{+} + i \Xi_{\mathcal{E}}^{-} \\
\Xi_{\mathcal{E}}^{+} - i \Xi_{\mathcal{E}}^{-} \\
\end{array} \right) \, ,$$ a separable product which can be delta-normalized [@Lo_1990]. Disregarding the normalization factor associated with the field mode, the polarization of the two-level system can be shown to settle in the equator of the Bloch sphere with its phase depending on the drive amplitude, $$\langle \hat{\sigma}_{-} \rangle = \sqrt{1- \beta_{\mathcal{E}}^{2} } + i \beta_{\mathcal{E}} = \sqrt{1- \frac{\lambda^{2}}{4\epsilon^{2}} } + i \frac{\lambda}{2 \epsilon} \, ,$$ as displayed in Figure \[fig:dipole\]. We have considered the zero quasi-energy states to underline the subtle differences between states above and below the critical point. The value of the polarization is, however, only dependent of the phase of the system and remains the same for each quasi-energy state. The same polarization values are found under the semiclassical approximation when dissipation and $N$ two-level systems are considered, but with finite photon number expectations [@Alsing_1992_1]. In fact, the divergences encountered above the critical point help to exemplify the role of dissipation in optical systems. Without dissipation, once the system is driven above the critical point, the field radiated by the two-level system is no longer strong enough to interfere destructively with the coherent drive \[see equation (\[qpt\_3\])\]. This leads to stationary states displaying an infinite photon number. Dissipation through the cavity walls would overcome the effect of the driving field, thus avoiding the divergences in photon number. While adding dissipation to the model goes beyond the scope of the current report, detailed studies of its effects in the driven Jaynes-Cummings model are presented in Refs. [@Carmichael_2015] and [@Alsing_1992_1]. The authors consider dissipation through two decay rates: one, $\kappa$, is related to damping of the field mode, and the other, $\gamma/2$, to spontaneous emission of the two-level system. It is found that even in the presence of these rates, the critical point identified in the Dirac-particle analogue \[equation (\[back\_2\])\] remains as the organizing center of the transition. For non-vanishing $\kappa$ the system reaches a steady-state in the long-time limit with zero photon number expectation below critical drive and a finite value above it. The interplay between $\kappa, \lambda$ and $\epsilon$ determine the maximum photon value, allowing for the definition of a thermodynamical limit in this system where threshold behaviour and its relation to a phase transition can be studied. For the generalized model ($\eta\neq0$), damping of the cavity mode is considered in Ref. [@Gutierrez_2017b].
![Bloch sphere representation of the polarization $\langle \hat{\mathbf{\sigma}} \rangle$ below (orange) and above (blue) the critical point $\epsilon_{0} = \lambda /2$. Two branches arise above the critical point since $\phi^{+}_{D}$ is also a solution of the JC Hamiltonian for $\varepsilon = 0$ under $E \rightarrow - E$ \[see equations (\[corr\_6a\]) and (\[corr\_6b\])\].[]{data-label="fig:dipole"}](tikz_bloch_5){width=".5\linewidth"}
Squeezing and Lorentz transformations
-------------------------------------
A word is in order regarding the apparent relationship between Lorentz transformations and squeezing, as found in the solutions below critical drive. The eigenstates found in the privileged frame, equation (\[eige\_8b\]), correspond to the dressed states of the Jaynes-Cummings model. When the Lorentz transformation was applied to reach the laboratory frame, the position and accompanying momentum operator appeared rescaled. This would translate to a squeezing transformation of these states. In reality, the natural length in the privileged frame is a relativistic invariant, $$l = \sqrt{ \frac{\hbar c }{ e ( F_{\mu \nu}F^{\mu \nu} )^{1/2}}} \, ,$$ which rules the behaviour of the system, and in this sense the transformation does not generate squeezing. The value of this quantity also helps to explain how the behaviour of the particle reacts towards the flipping of the potential introduced between magnetic and electric field-dominated cases. A real (complex) value determines discrete (continuous) solutions.
The creation and annihilation operators given in equation (\[corr\_7\]), however, are defined in terms of $l_{\mathcal{B}}$, not the relativistic invariant. They are affected by the Lorentz transformation. In the optical analogue these operators act upon the photon number of the field mode and the squeezing refers to photons injected into the mode through the coherent drive [@Alsing_1992_2]. In the mechanical analogue they act upon the Landau states. The expected particle number in these states corresponds to the number of magnetic flux quanta $(\Phi = \hbar c / \vert e \vert)$ that can be trapped in a circle of radius $l_{\mathcal{B}}$, a radius that changes between reference frames. It is important to stress out that these operators govern the quantized motion of the system, and are not responsible of creating and annihilating particles described by the Dirac equation. When working in a flat spacetime—as in the present report—there is a unique definition of vacuum regardless of the reference frame [@Ford_1998]; therefore, Lorentz transformations do not create particles and can not lead to squeezed states. This would not be the case in accelerated or curved spaces [@Davies_1975; @Unruh_1976; @Candelas_1978; @Jauregui_1991].
Conclusion {#secc:conclusion}
==========
We have obtained the eigenstates and eigenvalues of the on-resonance driven Jaynes-Cummings model, and, through equations (\[back\_6\])-(\[back\_7\_3\]), the driven Jaynes-Cummings-Rabi model. The solutions display two different phases of the system depending on the ratio between dipole coupling strength and driving field amplitude. Below a critical drive amplitude, the system exhibits a discrete level structure accompanied by a finite photon number expectation. This structure collapses at the critical drive, and becomes continuous with a divergent photon number expectation above it.
The exact solutions were obtained by mapping the driven Jaynes-Cummings model onto the evolution of a Dirac particle in an electromagnetic field. The relativistic covariance of the Dirac equation was then exploited and the solutions mapped back to the quantum optical system. Under this correspondence, dipole coupling and drive amplitude are related to magnetic and electric fields. When the former dominates, the interaction between spin and magnetic field creates a harmonic potential and discrete states naturally arise. When the latter dominates, the effective potential is unbounded from below and continuous solutions are encountered.
Aside from providing a theoretical method for finding solutions to the driven Jaynes-Cummings Rabi model, the introduced correspondence opens up a further connection between quantum phase transitions in quantum optical and condensed matter systems.
This research was supported by the Marsden fund of the Royal Society of New Zealand. R.G.J was supported on a scholarship funded by the New Zealand Tertiary Education Committee through the Dodd-Walls Centre for Photonic and Quantum Technologies. R.G.J. thanks R. Jáuregui for insightful discussions.
References {#references .unnumbered}
==========
[1]{}
Jaynes E T and Cummings F W 1963, *Proc. IEEE* **51**, 89 Meekhof D M, Monroe C, King B E, Itano W M and Wineland D J 1996 *Phys. Rev. Lett.* **76**, 1796-99 Brune M, Schmidt-Kaler F, Maali A, Dreyer J, Hagley E, Raimond J M and Haroche S 1996 *Phys. Rev. Lett.* **76**, 1800-03. Hofheinz M, Weig E M , Ansmann M, Bialczak R C , Lucero E, Neeley M, O’Connell A D, Wang H, Martinis J M and Cleland J M 2008, *Nature* **76**, 310-14 Carmichael H J 2015 *Phys. Rev. X* **5**, 031028 Alsing P, Guo D -S and Carmichael H J 1992 *Phys. Rev. A* **45**, 5135-43 Alsing P and Carmichael H J 1991 *Quantum Opt.* **3**, 13 Gutiérrez-Jáuregui R and Carmichael H J 2018 *Phys. Rev. A* **98** 023804 Dimer F, Estienne B, Parkins A S and Carmichael H J 2007 *Phys. Rev. A* **75**, 013804 MacDonald A H 1983 *Phys. Rev. B* **28**, 2235-36 Lukose V, Shankar R and Baskaran G 2007 *Phys. Rev. Lett.* **98**, 116802 Champel T and Florens S 2007 *Phys. Rev. B* **75**, 245326 Rozmej P and Arvieu R 1999 *J. Phys. A: Math Gen* **32**, 5367-82 Landau L D and Lifshitz E M 1971 *Relativistic Quantum Theory*, (Pergamon, New York), p 100. Gutiérrez-Jáuregui R Pérez-Pascual R, and Jáuregui R 2017 *Phys. Rev. A* **96**, 052109 M. Abramowitz and I. A. Stegun 1972 *Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables* (Dover, New York) Lo C F 1990 *Phys. Rev. A* **42**, 6752-55 Ford L H 1998 *Quantum Field Theory in Curved Spacetime*, in Proc. of the IXth Jorge André Swieca Summer School, eds. J. C. A. Barata, A. P. C. Malbouisson and S.F. Novaes (World Scientific, Singapore) Davies P C W 1975 *J. Phys. A* **8**, 609-16 Unruh W G 1976 *Phys. Rev. D* **14**, 870-92 Candelas P and Deutsch D 1978 *Proc. R. Soc. London* **A362**, 251-62 Jáuregui R, Torres M and Hacyan S 1991 *Phys. Rev. D* **43**, 3979-89
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The development of the radio remnant of SN 1987A has been followed using the Australia Telescope Compact Array since its first detection in 1990 August. The remnant has been observed at four frequencies, 1.4, 2.4, 4.8 and 8.6 GHz, at intervals of 4 – 6 weeks since the first detection. These data are combined with the 843 MHz data set of Ball et al. (2001) obtained at Molonglo Observatory to study the spectral and temporal variations of the emission. These observations show that the remnant continues to increase in brightness, with a larger rate of increase at recent times. They also show that the radio spectrum is becoming flatter, with the spectral index changing from $-0.97$ to $-0.88$ over the 11 years. In addition, at roughly yearly intervals since 1992, the remnant has been imaged at 9 GHz using super-resolution techniques to obtain an effective synthesised beamwidth of about 05. The imaging observations confirm the shell-morphology of the radio remnant and show that it continues to expand at $\sim3000$ . The bright regions of radio emission seen on the limb of the shell do not appear to be related to the optical hotspots which have subsequently appeared in surrounding circumstellar material.'
author:
- 'R. N. Manchester $^{1}$'
- 'B. M. Gaensler $^{2}$[^1]'
- 'V. C. Wheaton $^{1,3}$'
- 'L. Staveley-Smith $^{1}$'
- 'A. K. Tzioumis $^{1}$'
- 'N. S. Bizunok $^{2,4}$'
- 'M. J. Kesteven $^{1}$'
- 'J. E. Reynolds $^{1}$'
title: 'Evolution of the Radio Remnant of SN 1987A: 1990 – 2001'
---
=2em =17.5cm =24.6 cm =-2.5cm =-1.0cm =-1.0cm
[$^1$ Australia Telescope National Facility, CSIRO, PO Box 76, Epping NSW 1710\
[email protected]\
$^2$ Center for Space Research, Massachusetts Institute of Technology, Cambridge MA 02139, USA\
$^3$ School of Physics, University of Sydney, NSW 2006\
$^4$ Boston University, Boston MA 02215, USA\
]{}
[**Keywords:**]{} circumstellar matter — radio continuum: ISM — supernovae: individual (SN 1987A) — supernova remnants
Introduction
============
As the closest observed supernova in nearly 400 years, SN 1987A in the Large Magellanic Cloud offers unique opportunities for detailed study of the evolution of a supernova and the birth of a supernova remnant. In the 14 years since the explosion was detected, it has been extensively studied in all wavelength bands from radio to gamma-ray using both ground- and space-based observatories. These observations have revealed a complex evolution of both the supernova itself and the supernova remnant which is developing as the ejecta and their associated shocks interact with the circumstellar material.
Perhaps the most dramatic of these observations are the optical observations which show the beautiful triple-ringed structure surrounding and illuminated by the supernova (Burrows et al. 1995). The inner ring is believed to represent an equatorial density enhancement in the circumstellar gas at the interface between a dense wind emitted from an earlier red-giant phase of the supernova progenitor star, Sk$-69^{\circ}$202, and a faster wind emitted by this star in more recent times (Crotts, Kunkel & Heathcote 1995; Plait et al. 1995). In the past few years, there has been increasing evidence for interaction of the expanding ejecta or the associated shocks with the equatorial ring, with small regions of enhanced H$\alpha$ emission, or ‘hotspots’ appearing just inside the ring. The first of these was detected by Pun et al. (1997), but studies of ground-based and [*Hubble Space Telescope*]{} ([*HST*]{}) data by Lawrence et al. (2000) show that this spot was detectable as far back as March 1995 (day 2933 since the supernova[^2]). Lawrence et al. also present evidence for up to eight additional regions of enhanced emission from about day 4300 (December 1998). The most prominent are the original spot at position angle $29^{\circ}$ and a group of spots just south of east between position angles of $90^{\circ}$ and $140^{\circ}$.
At radio wavelengths, the initial outburst was very short-lived (Turtle et al. 1987) compared to other radio supernovae (e.g. Weiler et al. 1998). This prompt outburst is attributed to shock acceleration of synchrotron-emitting electrons in the stellar wind close to the star at the time of the explosion (Storey & Manchester 1987; Chevalier & Fransson 1987). After about three years, in mid-1990, radio emission was again detected from the supernova, with the Molonglo Observatory Synthesis Telescope (MOST) at 843 MHz (Ball et al. 1995) and with the Australia Telescope Compact Array (ATCA) at 1.4, 2.4, 4.8 and 8.6 GHz (Staveley-Smith et al. 1992; Gaensler et al. 1997). This emission has increased more-or-less monotonically since its first detection. The spectral index over the observed frequency range has remained close to $-0.9$, indicating optically thin synchrotron emission. X-ray emission was observed to turn on at about the same time as the second phase of radio emission and also has increased in intensity since then (Gorenstein, Hughes & Tucker 1994; Hasinger, Aschenbach & Trümper 1996). Recent observations with the [*Chandra X-ray Observatory*]{} have resolved the X-ray emission into an approximately circular shell (Burrows et al. 2000). This second phase of increasing emission is attributed to the interaction of shocks driven by the ejecta with circumstellar material and is distinct from the interaction with a radially decreasing stellar wind which characterises radio supernovae. It therefore signifies the birth of the [*remnant*]{} of SN 1987A – the first observation of the birth of a supernova remnant. We use the name SNR 1987A for the remnant.
By late-1992, SNR 1987A was sufficiently strong to image at 9 GHz using the ATCA (Staveley-Smith et al. 1993a). This image, which exploited super-resolution to resolve the source, showed that the remnant was roughly circular with a diameter of about 08, fitting inside the optical equatorial ring (Reynolds et al. 1995), and with bright lobes to the east and west, suggesting an annular structure with an axis similar to that of the optical emission. The eastern lobe was about 20% brighter than the western lobe. To follow the evolution of this structure, the remnant has been imaged at roughly yearly intervals since 1992.
Despite the increasing brightness, the size of SNR 1987A is increasing only slowly. Gaensler et al. (1997) fitted a model consisting of a thin spherical shell to the $uv$-plane data and showed that, between 1992 and 1995, the average expansion velocity of the remnant was only $2800 \pm 400$ km s$^{-1}$. In contrast, the average expansion speed between 1987 and 1991 was about 35,000 km s$^{-1}$ (Jauncey et al. 1988). Comparison of four images obtained between 1992 and 1995 showed that the brightness of the lobes increased relative to that of the spherical shell and that the asymmetry between the east and west lobes increased markedly over this period (Gaensler et al. 1997). By 1995, the peak brightness of the eastern lobe was 1.8 times that of the western lobe.
The only other young supernova which has been imaged with high resolution at radio wavelengths is SN 1993J in M81. Frequent observations using Very Long Baseline Interferometry (VLBI) techniques (e.g. van Dyk et al. 1994, Marcaide et al. 1997, Bartel et al. 2000) have shown that the radio emission is in the form of an expanding shell, the outline of which is nearly circular, but rather lumpy. Unlike SNR 1987A, SN 1993J remains in the radio supernova stage with decreasing flux density, making further imaging difficult. Recent observations of SN 1980K (Montes et al. 1998) and SN 1979C (Montes et al. 2000) have shown variations in the rate of decline of flux density or, in the case of SN 1979C, possibly small increases, indicating variations in the density of the circumstellar medium, but, like SN 1993J, these objects are best considered to be still in their radio supernova stage.
The slow expansion velocity of SNR 1987A suggests that the expanding shock and the leading ejecta have encountered a significant density enhancement, greatly reducing their velocity (Chevalier 1992; Duffy, Ball & Kirk 1995; Chevalier & Dwarkadas 1995). These models assume spherical symmetry and hence do not account for the increasing asymmetry of the source. Furthermore, they do not predict the continuing increase in brightness of the remnant as observed by Ball et al. (1995) and discussed further below. An alternative explanation for the slow apparent expansion velocity is that the shock excites slowly moving clumps of circumstellar material and then moves on. Ball & Kirk (1992) modelled the emission observed up to day 1800 by shock heating of two clumps and obtained a good fit to the data up to that time.
As discussed by Gaensler et al. (1997), it is not easy to account for the observed asymmetry of SNR 1987A. Models involving the annular structure of the circumstellar material (e.g. Chevalier & Dwarkadas 1995) have difficulty accounting for the degree of enhancement in the lobes and the east-west asymmetry. Similar difficulties are encountered in trying to explain the optical hotspots. Lawrence et al. (2000) conclude that “fingers or jets” in the distribution of ejecta from the supernova is the most plausible explanation.
In a recent publication, Ball et al. (2001) have analysed the 843 MHz flux densities observed with MOST to 2000 May (day 4820). They find a transition from a declining rate of increase observed from about mid-1991 (day $\sim 1600$) to early-1995 (day $\sim 2900$), to a larger and constant rate of increase, $62.7 \pm 0.5 \;\mu$Jy day$^{-1}$, since then. This change in slope occurred at about the same time as the appearance of the first optical hotspot (Lawrence et al. 2000), suggesting a possible connection.
In this paper, we extend the ATCA observational data base from 1995 to February 2001 (day 5100) and discuss these results in conjunction with those of Ball et al. (2001). The evolution of the radio flux densities is discussed in Section 2. The sequence of images is extended to late-2000 and discussed in relation to recent optical and X-ray imaging in Section 3. Future prospects are canvassed in Section 4.
Evolution of Radio Flux Densities and Spectral Index
====================================================
Flux density monitoring observations of SNR 1987A are made using the ATCA at 4 – 6 week intervals using one of the 6-km array configurations. Observations are made simultaneously at two frequencies, either 1380 and 2496 MHz (2368 MHz before mid-1997) or 4790 and 8640 MHz. A 128-MHz bandwidth is observed at all frequencies. The two frequency pairs are observed alternately, with 20 min on SNR 1987A and 3 min on phase calibrators before and after the SNR observation, with a total observation time typically of 12 hours. All observations are made with a J2000 pointing and phase centre of R.A. $05^{\rm h}\;35^{\rm m}\;27\fs90$, Dec. $-69^{\circ}\;16'\;21\farcs6$, approximately $10''$ south of the SN 1987A position (Reynolds et al. 1995). The phase calibrators are PKS B0530–727, PKS B0407–658 and (at 4790/8640 MHz) PKS B0454-810. Flux calibration is relative to PKS B1934-638, assumed to have flux densities of 14.95, 11.14, 5.83 and 2.84 Jy at the four frequencies, respectively.
Data are first checked for obvious interference or telescope problems, flagged if necessary and then are processed using [miriad]{}[^3] scripts to ensure consistency. Visibility data are calibrated in both phase and amplitude and images formed using baselines longer than 3 k$\lambda$. These images are cleaned and sources above a threshold identified. A table of source positions and integrated and peak flux densities is output for each frequency. At least at the two higher frequencies, SNR 1987A is resolved and so integrated flux densities are quoted. An unresolved source, J0536-6919, is present on all images with an observed flux density ranging between 80 mJy at 1.4 GHz and 6 mJy at 8.6 GHz and serves as a check on the system calibration. Its J2000 position determined from the 4.8 GHz observations since day 4000 is R.A. $05^{\rm h}\;36^{\rm m}\;04\fs789 \pm 0\fs005$, Dec. $-69^{\circ}\;18'\;44\farcs81 \pm 0\farcs02$. At 8.6 GHz it lies at about 1.5 primary beam radii from the pointing centre, so it’s flux density measurements at this frequency are not very reliable. Measured flux densities for SNR 1987A are given in Appendix 1.
Figure \[fg:fluxes\] shows the measured flux densities of SNR 1987A at the four ATCA frequencies. Estimated uncertainties represent a combination of random noise and scale errors resulting from errors in the calibration. Except at 8.6 GHz, the scale errors are estimated from the scatter in the measured flux densities of J0536-6918 since day 4000. At 8.6 GHz, J0536-6918 is outside the half-power radius of the primary beam, and scale errors are taken to be 1.25 times the 4790 MHz scale errors. At the lower frequencies and at later times, the errors in the flux density estimations are dominated by the scale errors. Since, except at 8.6 GHz, the flux density of J0536-6918 is comparable to that of SNR 1987A and since there is no evidence for systematic changes in the flux density of J0536-6918, these scale errors can be reduced by dividing the flux densities by the normalised flux density of J0536-6918 from the same observation. Flux densities from day 3000 scaled in this way for 1.4, 2.4 and 4.8 GHz are shown in Figure \[fg:scflux\].
These plots show that the continued increase in flux density observed at 843 MHz by Ball et al. (2001) is also observed at higher frequencies. The increase in slope observed at about day 2900 in the MOST data is also seen in the ATCA data (Fig. \[fg:fluxes\]). Ball et al. (2001) stated that the MOST data after day 3000 were well fitted by a linear trend. However, we believe that there is significant evidence for an long-term increase in slope after day 3000 in both the MOST data set and the ATCA data sets, i.e., the rate of increase in flux density is increasing. Evidence for this is shown in Fig. \[fg:fluxvar\] which shows residual flux densities after fitting a second-order polynomial to the MOST and scaled ATCA flux densities from day 3000 and subtracting the linear component. These plots all show a systematic trend with the residual flux densities being, on average, negative at central times and positive at both early and late times. Fig. \[fg:fluxvar\] also shows the second-order term of the fit which in all cases is positive and of about 3-$\sigma$ significance. These data sets are essentially independent so there can be little doubt about the significance of the effect.
To further quantify this effect, straight lines were fitted to the ATCA data sets from day 3000, both for the unscaled data and for the data (except at 8.6 GHz) scaled by the flux density of J0536-6918. Results of this fitting are given in Table \[tb:grad\]. The second and third columns give the gradient and $x$-intercept of the lines fitted to all data between day 3000 and day 5100 (cf. Fig. \[fg:fluxvar\]). The point of interest here is that there is a systematic increase of the date of intercept with frequency. Especially for the higher frequencies, the intercept is well after the date of first detection of the SNR. This shows that the higher frequencies have a higher relative rate of flux density increase and that the present rate of increase is much higher than that at early times. Note that this effect is seen in both the scaled and unscaled data, so it is not an artifact of the scaling procedure.
As a further check, the data sets were split into two halves, from day 3000 to day 4050 and from day 4050 to day 5100, and straight lines fitted to both halves. Results are tabulated in columns 4 – 7 of Table \[tb:grad\]. The values of reduced $\chi^2$ for the combined fit were computed by treating the two lines as a single model describing the complete data set from day 3000 onward. These show that, independently at all frequencies, the change in gradient is significant at about the 2-$\sigma$ level, and that this gradient increase is present in the MOST data and in both the scaled and the unscaled ATCA data.
[ccccccc]{} Frequency & Gradient & Intercept & Grad. ($<4050$) & Grad. ($>4050$) & Grad. change& Red. $\chi^2$\
(GHz) & ($\mu$Jy/day) & (day) & ($\mu$Jy/day) & ($\mu$Jy/day) & ($\mu$Jy/day)&\
\
0.843 & $63.1\pm 0.6$ & $1640\pm 20$ & $61.4\pm 1.0$ & $66.9\pm 2.1$ & $5.5\pm 2.3$ & 0.5\
1.4 & $40.9\pm 0.8$ & $1750\pm 50$ & $37.7\pm 1.7$ & $46.6\pm 2.4$ & $8.8\pm 2.9$ & 0.7\
2.4 & $25.2\pm 0.5$ & $1780\pm 50$ & $20.7\pm 1.2$ & $28.0\pm 1.6$ & $7.3\pm 1.9$ & 0.7\
4.8 & $15.2\pm 0.4$ & $2010\pm 55$ & $14.2\pm 0.6$ & $17.5\pm 1.2$ & $3.2\pm 1.4$ & 1.3\
8.6 & $10.6\pm 0.4$ & $2160\pm 65$ & $10.3\pm 0.8$ & $ 13.3\pm 1.7$ & $3.0\pm 1.9$ & 0.8\
\
1.4 & $41.2\pm 0.7$ & $1770\pm 45$ & $39.8\pm 2.0$ & $46.3\pm 2.0$ & $6.4\pm 2.8$ & 1.3\
2.4 & $26.0\pm 0.5$ & $1840\pm 45$ & $22.8\pm 1.5$ & $27.8\pm 1.4$ & $5.0\pm 2.0$ & 0.4\
4.8 & $15.3\pm 0.3$ & $2020\pm 45$ & $13.9\pm 0.8$ & $15.5\pm 1.0$ & $2.5\pm 1.3$ & 1.0\
Apparently significant short-term variations are seen, especially at 4.8 GHz, with a timescales of order 100 days. The clearest example is near the end of the 4.8 GHz data set and is best seen in Figure \[fg:fluxvar\]. This fluctuation is not obvious at lower frequencies, but could be masked by the low signal-to-noise ratio. If real, these short-term fluctuations imply significant interactions on a scale of 0.03 pc (01) or less, and furthermore, many such interactions.
Figure \[fg:spec\_ind\] shows MOST flux densities from Ball et al. (2001) and from the ATCA at 1.4, 2.4, 4.8 and 8.6 GHz at five epochs spread through the data set. Values of the spectral index $\alpha$, where $S=\nu^{\alpha}$ and $\nu$ is the frequency, found by linear regression, are given on each plot. Quoted errors are 1$\sigma$.
To test for systematic curvature in the spectrum, the spectral indices were calculated using the three lowest frequencies, $\alpha_1$ and separately using the three highest frequencies, $\alpha_2$. If there is systematic curvature in the spectrum, then the difference $\alpha_1 - \alpha_2$ should be significant, and roughly constant. In fact, the difference is typically small and of either sign, showing that there is no systematic curvature. The mean difference $\alpha_1 - \alpha_2$ across the entire data set is $-0.035$ compared to the rms fluctuation of $0.187$.
The observation that the relative flux density gradient is both greater and increasing more rapidly at the higher frequencies (Table \[tb:grad\]) implies a systematic change in the spectral index. Figure \[fg:si\_time\] shows the computed spectral indices from 843 MHz to 8.4 GHz as a function of time. Because of the uneven and non-simultaneous sampling at the different frequencies, spline curves were fitted to all data sets except that at 1.4 GHz. Spectral indices were then computed by interpolating values at 843 MHz, 2.4, 4.8 and 8.6 GHz to times when the 1.4 GHz flux density was measured, and fitting power-law spectra to flux densities. The interpolation was unreliable near the start of the data set for 2.4 and 8.6 GHz because of sparse data, so only three points were fitted there.
Apart from a few high points between days 1200 and 1400 near the start of the data set (which have large error bars), there is a more-or-less steady increase in spectral index, corresponding to a flattening of the radio spectrum, throughout the whole data set. It is possible that the variation consists of step changes, with the most obvious step times being around day 3000 and day 4700. At early times, the mean spectral index was about $-0.97$, but in the last year (2000) it was $-0.88$.
The radio spectra are remarkably close to power law and show no sign of either positive curvature, as would be expected if either free-free absorption or synchrotron self-absorption were important, or negative curvature as is predicted by diffusive shock acceleration theories (e.g. Reynolds & Ellison 1992). The radio spectral index of about $-0.9$ is considerably steeper than the canonical $-0.5$ expected for synchrotron emission from relativistic electrons accelerated in strong shocks, for which the spectral index $\alpha = (1-s)/2$ and the electron energy index $s = (r+2)/(r-2)$, where $r$ is the compression ratio in the shock (Jones & Ellison 1991). For high Mach-number shocks in a monatomic non-relativistic gas, $r=4.0$, giving $s=2.0$ and $\alpha =
-0.5$.
A steeper radio spectrum implies a steeper electron energy distribution and a smaller compression ratio in the shock; $s \sim 2.8$ and $r \sim 2.7$ for $\alpha = -0.9$. Monte-Carlo modelling of non-linear diffusive shock acceleration, including the dynamical effects of the accelerated particles (e.g. Baring et al. 1999), suggest that young and hence fast shocks have a [*larger*]{} compression ratio than older shocks, resulting in a flatter synchrotron spectrum. Duffy, Ball & Kirk (1995) included the effect of accelerated ions on the shock structure and were able to model the observed spectrum with the shock expanding into a terminated stellar wind structure. However, this model predicts a declining brightness and a steepening spectrum for the radio emission, both of which are contrary to observation.
The observed spectral index is also steeper than those for typical supernova remnants; no SNR in the Green (2000) catalogue has a spectral index definitely steeper than $-0.8$ and the mean spectral index for shell-type remnants is $-0.51$. It is interesting to speculate that the current flattening of the spectrum represents an evolution toward the typical index of $-0.5$. At the current rate, it will only take about 50 years to reach $-0.5$.
The clump interaction model of Ball & Kirk (1992) predicts a declining flux density at later times. However, this model was based on only two clumps. The model could be extended to invoke interaction with an increasing number of clumps, maybe tens or hundreds at the present time. A more accurate description would involve a statistical hierarchy of effective clump sizes. This could account for both the gradient increase and the short-term fluctuations in the observed flux density. Both radio and optical evidence (e.g. Spyromilio, Stathakis & Meurer 1993, van Dyk et al. 1994, Spyromilio 1994) point to a clumpy circumstellar medium around other supernovae.
9 GHz Imaging Observations of SNR 1987A
=======================================
Resolved Radio Images of SNR 1987A
----------------------------------
In previous papers we have shown that the ATCA’s diffraction-limited resolution at 9 GHz of 09 is sufficient to resolve the radio emission from SNR 1987A (Staveley-Smith et al. 1993b) and that super-resolution techniques can be used to improve the resolution to $\sim0\farcs5$. At this resolution, the radio emission forms a limb-brightened shell, with brightness enhancements on the eastern and western sides (Staveley-Smith et al. 1993a; Briggs 1994; Gaensler et al. 1997).
We have continued to make regular observations of SNR 1987A at 9 GHz; observing parameters for all imaging observations are given in Table \[tab\_9ghz\]. We have analysed these data in the same manner as described by Gaensler et al. (1997), forming an image at each epoch and deconvolving the resulting image using a maximum entropy algorithm. From 1996 onwards, the source has been of sufficient signal-to-noise that phase self-calibration can be successfully applied to the data, resulting in a significant improvement in the accuracy of the complex gains for each antenna.
------------ ------------- -------- ------- ------------- ------------- ---------- --
Mean Epoch Observing Day Array Frequencies Time on Radius
Date Number (MHz) Source (hr) ($''$)
1992.9 1992 Oct 21 2068 6C 8640,8900 15 0.66(2)
1993 Jan 04 2142 6A 8640,8900 13 0.66(2)
1993 Jan 05 2143 6A 8640,8900 5 0.62(2)
1993.6 1993 Jun 24 2314 6C 8640,8900 9 0.62(1)
1993 Jul 01 2321 6C 8640,8900 10 0.68(1)
1993 Oct 15 2426 6A 8640,9024 18 0.69(1)
1994.4 1994 Feb 16 2550 6B 8640,9024 9 0.69(2)
1994 Jun 27 2683 6C 8640,9024 21 0.659(7)
1994 Jul 01 2687 6A 8640,9024 10 0.659(9)
1995.7 1995 Jul 24 3074 6C 8640,9024 7 0.687(8)
1995 Aug 29 3111 6D 8896,9152 7 0.69(2)
1995 Nov 06 3178 6A 8640,9024 9 0.685(6)
1996.7 1996 Jul 21 3437 6C 8640,9024 14 0.688(4)
1996 Sep 08 3486 6B 8640,9024 13 0.684(4)
1996 Oct 05 3512 6A 8896,9152 8 0.692(6)
1998.0 1997 Nov 11 3914 6C 8512,8896 19 0.715(5)
1998 Feb 18 4013 6A 8896,9152 15 0.733(5)
1998 Feb 21 4016 6B 9024,8512 7 0.735(4)
1998.9 1998 Sep 13 4220 6A 8896,9152 12 0.721(3)
1998 Oct 31 4268 6D 9024,8512 11 0.729(5)
1999 Feb 12 4372 6C 8512,8896 10 0.737(3)
1999.7 1999 Sep 05 4578 6D 9152,8768 11 0.754(4)
1999 Sep 12 4585 6A 8512,8896 14 0.736(4)
2000.8 2000 Sep 28 4966 6A 8512,8896 10 0.756(2)
2000 Nov 12 5011 6C 8512,8896 11 0.764(3)
------------ ------------- -------- ------- ------------- ------------- ---------- --
: 9 GHz ATCA observations of SNR 1987A used for imaging. The radius listed is that obtained by fitting a thin spherical shell to each $u-v$ data-set (see text).[]{data-label="tab_9ghz"}
The resulting series of images are shown in Figures \[fig\_movie\_natural\] and \[fig\_movie\_super\] under the conditions of diffraction-limited resolution and super-resolution, respectively. In these figures and Figure \[fig\_models\], the R.A. and Dec. offsets are with respect to J2000 R.A. $05^{\rm h}\;35^{\rm m}\;28\fs00$, Dec. $-69^{\circ}\;16'\;11\farcs1$ The diffraction-limited images indicate that the source is clearly extended, primarily in the east-west direction, and continues to brighten. In the super-resolved sequence of images, it can be seen that the shell-like morphology reported by Staveley-Smith et al. (1993a) and by Gaensler et al. (1997) is maintained throughout, with two bright regions on the east and west sides of the rim.
Model Fits to the Radio Morphology
----------------------------------
Gaensler et al. (1997) showed that the size of SNR 1987A could be quantified at each epoch by approximating the morphology of SNR 1987A by a thin spherical shell of arbitrary position, flux and radius. The best-fit parameters are found by computing the Fourier-transform of this shell, subtracting this transform from the $u-v$ data, and then adjusting the properties of the model until the corresponding $\chi^2$ parameter is minimised (see also Staveley-Smith et al. 1993b). Using such an approach, Gaensler et al. (1997) were able to show that the radius of the supernova remnant increased from 065 at epoch 1992.9 to 068 at epoch 1995.7, corresponding to a (surprisingly low) mean expansion velocity of $2800\pm400$ (assuming a distance to the supernova of 50 kpc).
Here we extend and expand on these attempts to quantify the changes in the radio remnant. We first fit a spherical shell to all subsequent data-sets. The resulting radii are listed in Table \[tab\_9ghz\], where the uncertainty in the last quoted digit is given in parentheses. A fit to these radii gives a linear expansion rate of $3500\pm100$ . The mean radius at each observing epoch is listed in Table \[tab\_params\] and plotted in Figure \[fig\_expand\].
The signal-to-noise of the data presented by Gaensler et al. (1997) was too low to justify more complex fits to the data. However, we are now in a position to compare a thin spherical shell to other models. We first note that other simple spherically-symmetric models, such as face-on rings and gaussians, produce significant residuals, and are clearly inconsistent with the data. In Figure \[fig\_models\] we show three possible fits to the data from epoch 1998.0. In each case, we have generated an input model (shown in the first column). We have then simulated an observation of the sky-distribution corresponding to this model as follows: we Fourier-transform the model, then multiply this transform by the transfer function of the ATCA observations from this epoch to produce a set of $u-v$ tracks identical to those of the real data. We then image these visibilities in the same way as for the real data, and similarly deconvolve and super-resolve the image, to give the result shown in the second column. We also subtract the $u-v$ data corresponding to the model from the observed data, and then image the resulting visibilities to produce the residual image shown in the third column.
In the first row of Figure \[fig\_models\], we show the results of fitting a thin spherical shell to the 1998.0 data (cf. Gaensler et al. 1997) — the corresponding best-fit flux and radius are 18.4 mJy and 073. The resulting image successfully approximates the limb-brightened morphology seen in Figure \[fig\_movie\_super\], but lacks the enhancements in brightness seen on either side of the shell. The residual image, while having small amplitudes in an absolute sense, shows clear and systematic differences between this simple model and the data (maximum fractional residual of $\sim$20%).
Given that the main difference between a thin shell model and the data seems to be the presence of the bright lobes on either side, we next generate a model involving a thin shell (of arbitrary flux, position and radius), and two point-sources of emission (each of arbitrary flux and position). For the 1998.0 data set, the best-fit parameters for this model are a shell of flux 15.0 mJy and radius 079, and point-sources of fluxes 3.1 and 0.4 mJy, one projected near the edge of, but inside, the eastern half of the shell, and the other similarly positioned on the western half. Note that the radius of the shell in this fit is 8% larger than that obtained by fitting to a shell alone. The resulting model, image and residual are all shown in the second row of Figure \[fig\_models\]. The image corresponding to the model now reasonably approximates the morphology seen in the real data, and the residuals are greatly reduced (maximum fractional residual of $\sim$5%) when compared to the case of a shell alone.
We can further improve the fit between the model and the data by increasing the number of point sources in the model. In the bottom row of Figure \[fig\_models\], we show a fit to the 1998.0 data involving a spherical shell plus seven point sources. The large number of free parameters (25 in this case) makes finding an absolute minimum in $\chi^2$ very difficult — thus the “best fit” we have shown was found by trial-and-error. The shell has flux density 11.4 mJy and radius 072 (2% smaller than in the case of a shell alone); the point sources range in flux density between 0.3 and 3.5 mJy, and are distributed around the perimeter of the shell. The model image and residual image both indicate a very good match between the model and the data (maximum fractional residual $\sim$2%). We emphasise, though, that we make no claims to uniqueness for this solution.
-------- ------------ ----------------- ----------------- ----------------- ----------------- ------------- -------------------- --------------------- ------------- -------------------- ---------------------
Mean Total Flux
Epoch Density $S_{\rm shell}$ $r_{\rm shell}$ $S_{\rm shell}$ $r_{\rm shell}$ $S_{\rm 1}$ $\delta{\rm RA}_1$ $\delta{\rm Dec}_1$ $S_{\rm 2}$ $\delta{\rm RA}_2$ $\delta{\rm Dec}_2$
(mJy) (mJy) ($''$) (mJy) ($''$) (mJy) ($''$) ($''$) (mJy) ($''$) ($''$)
1992.9 5.6(2) 5.2 0.668(9) $\ldots$ $\ldots$ $\ldots$ $\ldots$ $\ldots$ $\ldots$ $\ldots$ $\ldots$
1993.6 6.9(2) 6.6 0.659(6) 5.1 0.73(2) 1.2 0.26(3) -0.05(2) 0.4 -0.73(9) 0.06(5)
1994.4 7.5(1) 7.8 0.671(4) 6.2 0.75(2) 1.3 0.27(2) -0.03(1) 0.4 -0.69(7) 0.06(4)
1995.7 11.8(1) 11.0 0.684(4) 8.6 0.71(2) 1.9 0.35(3) -0.03(1) 0.5 -0.76(8) 0.04(5)
1996.7 15.5(1) 15.1 0.688(2) 12.4 0.764(7) 2.5 0.32(1) 0.03(1) 0.4 -0.69(6) 0.15(3)
1998.0 18.3(1) 18.4 0.731(2) 15.0 0.786(8) 3.1 0.34(1) -0.02(1) 0.4 -0.75(7) -0.02(5)
1998.9 21.7(1) 21.5 0.730(2) 17.2 0.778(6) 3.4 0.40(1) -0.02(1) 1.0 -0.81(3) -0.04(2)
1999.7 24.7(1) 24.4 0.742(2) 19.7 0.796(6) 3.9 0.41(1) -0.02(1) 1.0 -0.81(3) -0.05(2)
2000.8 30.8(1) 30.8 0.761(2) 24.9 0.804(4) 4.8 0.46(1) -0.01(1) 1.3 -0.82(2) -0.03(2)
-------- ------------ ----------------- ----------------- ----------------- ----------------- ------------- -------------------- --------------------- ------------- -------------------- ---------------------
: Parameters from model fits to the 9 GHz images[]{data-label="tab_params"}
Given the difficulty in finding a multiple-component fit to the data by trial-and-error, and the non-uniqueness of this solution, we do not here present multi-parameter fits to other epochs such as those shown in the third row of Figure \[fig\_models\]. However, it seems clear that a shell alone is no longer the best simple model of the data, and that a shell with two point-sources is a significant improvement. We therefore have fitted all epochs with these two models, with all parameters free to vary in each case. The results of these fits are listed in Table \[tab\_params\] and plotted in Figure \[fig\_expand\]. Offsets of the two point sources in Table \[tab\_params\] are with respect to the reference position used for Figures 6, 7 & 9. The data from epoch 1992.9 is of low signal-to-noise, and no good fit using this more complex model cound be found.
It can be seen from Table \[tab\_params\] that the radius of the shell in models which include two point-sources is 5–10% larger than that obtained in the shell-only model used by Gaensler et al. (1997). It is also clear that the point sources are moving outward with the expansion of the shell, the eastern one apparently at a larger relative rate than the shell. As for the shell model, we can fit these larger radii by a linear increase, to obtain an expansion speed of $2300\pm300$ , about 35% slower than in the case for the shell alone. However, we note that if multi-parameter fits like those in the third row of Figure \[fig\_models\] are carried out for other epochs, the resulting radii are $\sim$5% [*smaller*]{} than for the fits to a shell alone, and the resulting inferred expansion speed is $\sim3700$ . While we have not carried out the corresponding analysis here, we note that Staveley-Smith et al. (1993a) fitted the radio emission from SNR 1987A with a thick spherical shell of outer-to-inner radius ratio 1.25. This results in a shell diameter about 10% greater than for the thin-shell fit, consistent with the range of possible radii considered here.
We therefore conclude that the radius of the remnant as determined from fitting to the $u-v$ data is uncertain by $\sim$10%, and that the resulting expansion velocity is uncertain by $\sim$30%. The main conclusions from earlier results — that the radio emission is originating from a region within the equatorial ring, and that the material producing this emission was initially moving rapidly but now has a very low rate of expansion — are unchanged, even when the assumption of spherical symmetry is relaxed.
Comparison With Other Wavelengths and Discussion
------------------------------------------------
In Figure \[fig\_overlay\] we compare our 9 GHz ATCA data to recent observations with [*HST*]{} and with [*Chandra*]{}. In the upper panel, the super-resolved ATCA image from epoch 2000.8 is compared to the difference-image of optical emission around the supernova produced by Lawrence et al. (2000). The HST image has been registered on the radio frame with an accuracy of better than $0\farcs1$ using the position of the central star from Reynolds et al. (1995). It shows both the fading equatorial ring, and and at least seven hotspots on this ring, all brightening as the supernova shock begins to interact with dense circumstellar material. It can be seen from this optical/radio comparison that the conclusions made by Gaensler et al. (1997) are maintained: the bright radio lobes align with the major axis of the optical ring, with the brighter eastern lobe clearly more distant from the central star than the fainter western lobe. Within the uncertainties, the eastern radio lobe is coincident with some of the optical hotspots (but not the brighest one), just inside the optical ring. In the west, the optical emission lies outside the radio lobe, close to the lowest radio contour in Figure \[fig\_overlay\]. In fact, there is a good correspondence of the radio emission with optical emission [*within*]{} the optical ring (not visible in Figure \[fig\_overlay\]) corresponding to the reverse-shock, as reported by Garnavich, Kirshner & Challis (1999). Just as in the radio data, two optical lobes are seen, one to the east and one to the west of the supernova site. Also similar to the case in the radio, the eastern optical lobe is brighter than and further from the supernova than the western lobe.
The lower panel of Figure \[fig\_overlay\] compares radio emission with a super-resolved image obtained by [*Chandra*]{} at epoch 1999.8 (Burrows et al. 2000). The X-ray and radio emission from the supernova remnant both take the form of limb-brightened shells; the images were aligned by placing the estimated centre of the X-ray remnant on the Reynolds et al. (1995) position (D. Burrows, private communication) and hence is less accurate than the optical–radio alignment. While there is a good correspondence between the brightest regions in each waveband, the X-ray maxima appear to lie outside the radio maxima. This is perhaps surprising since theoretical models (e.g. Borkowski, Blondin & McCray 1997) suggest that the radio emission is generated just inside the outer shock whereas the X-ray emission is generated by a reverse shock compressing and heating the denser ejecta as it propagates inwards. It is worth noting, however, that X-ray emission lying outside the radio emission and interpreted as coming from the outer shock has been detected in the SNR 1E 0102.2$-$7219 by Gaetz et al. (2000).
The two-lobed radio morphology seen for SNR 1987A was apparent at least as early as 1992 (Figure \[fig\_movie\_super\]; Staveley-Smith et al 1993a), while optical hotspots did not begin to appear on the eastern side of the optical ring until 1995 and on the western side until 1998 (Lawrence et al. 2000). Furthermore, the earliest-appearing and brightest optical hotspot does not coincide in position angle with either of the two main radio lobes, nor with the brightest X-ray emission seen by [*Chandra*]{}(Figure \[fig\_overlay\]; Burrows et al. 2000). We therefore argue that the more rapid rate in the increase of radio emission beginning around day 3000, as reported by Ball et al. (2001) and confirmed in Section 2 above, is not related to the appearance of optical hotspots seen at around the same time. While Ball et al. (2001) have pointed out that the various optical hotspots turned on at about the right time for them to have been produced by the arrival of the radio-producing shock(s) in these regions, it seems clear that this had little effect on the radio morphology of the remnant. The best explanation for deviations from spherical symmetry in the radio morphology of SNR 1987A still seems to be that they result simply from regions of enhanced emission in the equatorial plane of the progenitor system, where circumstellar gas is expected to be both densest and closest to the progenitor star (cf. Gaensler et al. 1997). The fact that at both radio and optical wavelengths the eastern lobe is both brighter and more distant from the explosion site than the western lobe may represent an asymmetry in the distribution of ejecta.
It is notable that the emission underlying the radio hotspots is well described by a spherical shell and that, in all of the models, the flux density of the shell dominates the total flux density. This is surprising given the strong equatorial enhancement evident in the optical data and implied for the circumstellar gas. One might expect eventually to see a faster expansion of the radio remnant in the polar directions (Blondin, Lundqvist & Chevalier 1996), but there is currently no evidence for this. Part of the radio emission from the lobes may be attributed to an equatorial enhancement but, since such an enhancement would be symmetric, it cannot account for all of the emission seen from the brighter eastern lobe. It is worth noting that the shell of SN 1993J also appears to be quite spherical (Marcaide et al. 1997, Bartel et al. 2000).
Future Prospects
================
Results at all wavelengths suggest that there is an increasing interaction between the expanding ejecta and the circumstellar material. Hotspots around or just inside the emission-line ring are certainly becoming more numerous and prominent in the optical band. Neither the radio imaging nor the X-ray imaging has sufficient resolution to separately identify hot spots. However, the modelling of the radio $u-v$ data and the possible short-term fluctuations in the rise of the high-frequency radio flux densities suggest that compact radio hotspots do exist. The overall morphology and the time-evolution of the radio emission suggest that there is no detailed correspondence of the radio hotspots with the hotspots seen in the optical data and, in fact, that the radio hotspots are much more numerous and widespread.
We have shown that the rate of increase of the radio emission from SNR 1987A has slowly increased over the past few years. It is possible that the rate of increase of the radio (and other) emission will dramatically increase when significant amounts of ejecta begin to interact with the dense circumstellar gas of the inner ring. Extrapolation of the radii and expansion speeds resulting model fits to the radio data suggest that this will happen in $2004\pm2$. The ATCA is presently being upgraded for observations in the 12 mm and 3 mm bands, with an expected commissioning date of mid-2002. While the radio remnant is unlikely to be detectable at 3 mm, we expect 12 mm observations to provide increased resolution. Extrapolating the present flux density increase and the currently observed power-law spectrum, we expect a flux density at 20 GHz in mid-2002 of about 17 mJy. The diffraction limited half-power beamwidth at 20 GHz will be about 04, somewhat less than the present super-resolved beamwidth at 9 GHz. Even in 2002, it is likely that super-resolution will be able to be applied to the 20 GHz data, giving a resolution of about 02, and this will certainly be true at later epochs if the flux density continues to rise. We hope and expect that this will reveal further detail in the radio images, allowing interesting comparisons with images obtained at other wavelengths and giving further insight into the physics of the radio emission process.
Acknowledgements {#acknowledgements .unnumbered}
================
We thank Lewis Ball for making available a pre-publication file of the MOST flux densities for SNR 1987A and for comments on an earlier version of the paper, Dave Burrows for providing the [*Chandra*]{} data, and Ben Sugerman and Peter Garnavich for providing [*HST*]{} data. We also thank the referees for helpful suggestions. The Australia Telescope is funded by the Commonwealth of Australia for operation as a National Facility managed by CSIRO. BMG acknowledges the support of NASA through Hubble Fellowship grant HF-01107.01-98A awarded by the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., for NASA under contract NAS 5–26555.
References {#references .unnumbered}
==========
Ball, L., Campbell-Wilson, D., Crawford, D. F., & Turtle, A. J. 1995, ApJ, 453, 864
Ball, L., Crawford, D. F., Hunstead, R. W., Klamer, I., & McIntyre, V. J. 2001, ApJ, 549, 599
Ball, L. & Kirk, J. G. 1992, ApJ Lett., 396, L39
, M. G., [Ellison]{}, D. C., [Reynolds]{}, S. P., [Grenier]{}, I. A., & [Goret]{}, P. 1999, ApJ, 513, 311
Bartel, N. et al. 2000, Science, 287, 112
Blondin, J. M., Lundqvist, P. & Chevalier, R. A., 1996, ApJ, 472, 257
Borkowski, K. J., Blondin, J. M. & McCray, R. 1997, ApJ, 476, L31
Briggs, D. S. 1994, in [ The Restoration of HST Images and Spectra II]{}, ed. R. J. Hanisch & R. L. White, (Baltimore: Space Telescope Science Institute), 250
Burrows, C. J. [et al.]{} 1995, ApJ, 452, 680
, D. N. [et al.]{} 2000, ApJ, 543, L149
Chevalier, R. A. 1992, Nature, 355, 617
Chevalier, R. A. & Dwarkadas, V. V. 1995, ApJ, 452, L45
Chevalier, R. A. & Fransson, C. 1987, Nature, 328, 44
Crotts, A. P. S., Kunkel, W. E., & Heathcote, S. R. 1995, ApJ, 438, 724
Duffy, P., Ball, L., & Kirk, J. G. 1995, ApJ, 447, 364
Gaensler, B. M., Manchester, R. N., Staveley-Smith, L., Tzioumis, A. K., Reynolds, J. E., & Kesteven, M. J. 1997, ApJ, 479, 845
, T. J., [Butt]{}, Y. M., [Edgar]{}, R. J., [Eriksen]{}, K. A., [Plucinsky]{}, P. P., [Schlegel]{}, E. M. & [Smith]{}, R. K. 2000, ApJ, 534, L47
, P., [Kirshner]{}, R., & [Challis]{}, P. 1999. IAU Circ. 7102
Gorenstein, P., Hughes, J. P., & Tucker, W. H. 1994, ApJ, 420, L25
Green, D. A. 2000, [ A Catalogue of Galactic Supernova Remnants]{}, (Cambridge: Mullard Radio Astronomy Observatory, Cavendish Laboratory). http://www.mrao.cam.ac.uk/surveys/snrs/.
Hasinger, G., Aschenbach, B., & Trümper, J. 1996, A&A, 312, L9
Jauncey, D. L., Kemball, A., Bartel, N., Shapiro, I. I., Whitney, A. R., Rogers, A. E. E., Preston, R. A., & Clark, T. 1988, Nature, 334, 412
Jones, F. C. & Ellison, D. C. 1991, Space Sci. Rev., 58, 259
, S. S., [Sugerman]{}, B. E., [Bouchet]{}, P., [Crotts]{}, A. P. S., [Uglesich]{}, R., & [Heathcote]{}, S. 2000, ApJ, 537, L123
Marcaide, J. E. et al. 1997, ApJ, 486, L31
Montes, M. J., van Dyk, S. D., Weiler, K. W., Sramek, R. A., & Panagia, N. 1998, ApJ, 506, 874
Montes, M. J., Weiler, K. W., Van Dyk, S. D., Panagia, N., Lacey, C. K., Sramek, R. A., & Park, R. 2000, ApJ, 532, 1124
Plait, P. C., Lundqvist, P., Chevalier, R. A., & Kirshner, R. P. 1995, ApJ, 439, 730
Pun, C. S. J., Sonneborn, G., Bowers, C., Gull, T., Heap, S., Kimble, R., Maran, S., & Woodgate, B. 1997. IAU Circ. 6665
Reynolds, J. E. [et al.]{} 1995, A&A, 304, 116
, S. P. & [Ellison]{}, D. C. 1992, ApJ, 399, L75
Spyromilio, J., Stathakis, R. A. & Meurer, G. R. 1993, MNRAS, 263, 530
Spyromilio, J. 1994, MNRAS, 266, 61
Staveley-Smith, L., Manchester, R. N., Kesteven, M. J., Campbell-Wilson, D., Crawford, D. F., Turtle, A. J., Reynolds, J. E., Tzioumis, A. K., Killeen, N. E. B. & Jauncey, D. L. 1992, Nature, 355, 147
Staveley-Smith, L., Briggs, D. S., Rowe, A. C. H., Manchester, R. N., Reynolds, J. E., Tzioumis, A. K., & Kesteven, M. J. 1993a, Nature, 366, 136
Staveley-Smith, L., Manchester, R. N., Kesteven, M. J., Tzioumis, A. K., & Reynolds, J. E. 1993b, PASA, 10, 331.
Storey, M. C. & Manchester, R. N. 1987, Nature, 329, 421
Turtle, A. J. [et al.]{} 1987, Nature, 327, 38
van Dyk, S. D., Weiler, K. W., Sramek, R. A., Rupen, M. P., & Panagia, N. 1994, ApJ, 432, L115
, K. W., [van Dyk]{}, S. D., [Montes]{}, M. J., [Panagia]{}, N., & [Sramek]{}, R. A. 1998, ApJ, 500, 51
Appendix {#appendix .unnumbered}
========
Table \[tb:fluxes\] gives flux densities measured at the four ATCA frequencies over the day range 2000 - 5100 (MJD 48449 to 51949). These are calibrated relative to the primary flux calibrator, PKS B1934-638, using standard techniques. This table, also including flux density measurements for the nearby unresolved source J0536-6918 and more recent measurements, is available at http://www.atnf.csiro.au/research/pulsar/snr/sn1987a/.
-------- ------------------ -------- ------------------ -------- ----------------- -------- ------------------
Day Flux Density Day Flux Density Day Flux Density Day Flux Density
Number (mJy) Number (mJy) Number (mJy) Number (mJy)
1243.9 $ 0.00\pm 1.20$ 1198.9 $ 0.00\pm 1.00$ 1243.6 $ 0.00\pm 0.25$ 1388.1 $ 1.33\pm 0.26$
1270.7 $ 2.92\pm 1.20$ 1314.5 $ 3.27\pm 1.00$ 1269.7 $ 0.66\pm 0.25$ 1432.2 $ 1.12\pm 0.26$
1386.4 $ 4.87\pm 1.21$ 1386.0 $ 3.45\pm 1.01$ 1305.5 $ 1.15\pm 0.25$ 1500.5 $ 1.74\pm 0.26$
1490.0 $ 7.61\pm 1.22$ 1517.9 $ 6.13\pm 1.02$ 1333.7 $ 1.49\pm 0.26$ 1515.0 $ 2.34\pm 0.28$
1517.9 $ 8.29\pm 1.23$ 1517.7 $ 6.20\pm 1.02$ 1366.3 $ 1.53\pm 0.26$ 1583.8 $ 2.62\pm 0.28$
1517.7 $ 9.55\pm 1.23$ 1595.8 $ 5.55\pm 1.01$ 1385.2 $ 1.96\pm 0.26$ 1593.8 $ 2.21\pm 0.27$
1595.8 $ 10.15\pm 1.24$ 1636.7 $ 8.71\pm 1.03$ 1401.4 $ 2.17\pm 0.26$ 1662.5 $ 2.92\pm 0.29$
1636.7 $ 15.52\pm 1.29$ 1660.5 $ 9.15\pm 1.04$ 1402.4 $ 2.15\pm 0.26$ 1661.5 $ 2.50\pm 0.28$
1660.5 $ 15.52\pm 1.29$ 1788.8 $ 10.67\pm 1.05$ 1403.4 $ 2.25\pm 0.27$ 1786.6 $ 3.92\pm 0.32$
1788.8 $ 20.59\pm 1.35$ 1850.1 $ 13.07\pm 1.07$ 1404.3 $ 2.12\pm 0.26$ 1852.1 $ 4.37\pm 0.33$
1850.1 $ 24.80\pm 1.41$ 1879.2 $ 15.03\pm 1.10$ 1405.3 $ 2.13\pm 0.26$ 1878.2 $ 4.14\pm 0.32$
1879.2 $ 24.04\pm 1.40$ 1969.7 $ 17.95\pm 1.14$ 1407.1 $ 2.06\pm 0.26$ 1947.8 $ 4.70\pm 0.34$
1969.7 $ 27.51\pm 1.46$ 2067.7 $ 19.00\pm 1.15$ 1408.3 $ 2.57\pm 0.27$ 1969.5 $ 4.93\pm 0.35$
2068.4 $ 33.15\pm 1.56$ 2313.7 $ 23.25\pm 1.22$ 1409.3 $ 2.53\pm 0.27$ 1986.4 $ 4.64\pm 0.34$
2313.7 $ 38.87\pm 1.67$ 2427.3 $ 23.37\pm 1.22$ 1410.3 $ 2.66\pm 0.27$ 2003.4 $ 5.09\pm 0.36$
2427.3 $ 39.96\pm 1.70$ 2505.2 $ 25.91\pm 1.27$ 1431.2 $ 2.29\pm 0.27$ 2067.5 $ 5.48\pm 0.37$
2505.2 $ 43.23\pm 1.77$ 2549.2 $ 26.56\pm 1.28$ 1446.2 $ 2.63\pm 0.27$ 2092.4 $ 5.92\pm 0.39$
2549.2 $ 43.83\pm 1.78$ 2680.7 $ 28.40\pm 1.31$ 1460.1 $ 3.03\pm 0.28$ 2142.2 $ 5.43\pm 0.37$
2680.7 $ 46.20\pm 1.83$ 2774.5 $ 29.92\pm 1.34$ 1516.9 $ 2.83\pm 0.27$ 2195.7 $ 4.20\pm 0.33$
2774.5 $ 49.26\pm 1.90$ 2773.7 $ 30.40\pm 1.35$ 1524.9 $ 2.74\pm 0.27$ 2261.8 $ 6.04\pm 0.39$
2773.7 $ 48.60\pm 1.89$ 2826.7 $ 28.40\pm 1.31$ 1586.8 $ 2.97\pm 0.28$ 2299.9 $ 7.14\pm 0.44$
2857.7 $ 50.90\pm 1.94$ 2857.7 $ 29.30\pm 1.33$ 1594.8 $ 4.50\pm 0.31$ 2299.7 $ 5.80\pm 0.38$
2874.3 $ 48.40\pm 1.88$ 2874.3 $ 25.20\pm 1.25$ 1635.6 $ 5.31\pm 0.33$ 2313.7 $ 6.92\pm 0.43$
2919.2 $ 53.77\pm 2.01$ 2919.2 $ 33.11\pm 1.41$ 1662.5 $ 5.02\pm 0.32$ 2320.7 $ 6.90\pm 0.43$
2919.3 $ 51.38\pm 1.95$ 2919.3 $ 34.20\pm 1.43$ 1746.8 $ 6.04\pm 0.35$ 2374.7 $ 5.38\pm 0.37$
2975.8 $ 54.97\pm 2.04$ 2975.8 $ 33.55\pm 1.42$ 1786.8 $ 7.44\pm 0.39$ 2403.5 $ 7.14\pm 0.44$
2975.7 $ 48.40\pm 1.88$ 2975.7 $ 32.50\pm 1.40$ 1852.1 $ 8.16\pm 0.41$ 2426.4 $ 7.68\pm 0.46$
3001.7 $ 48.80\pm 1.89$ 3001.7 $ 33.11\pm 1.41$ 1878.2 $ 7.06\pm 0.38$ 2462.7 $ 6.50\pm 0.41$
3000.7 $ 56.20\pm 2.07$ 3072.5 $ 34.60\pm 1.44$ 1947.8 $ 8.27\pm 0.41$ 2550.2 $ 7.47\pm 0.45$
3072.5 $ 61.70\pm 2.21$ 3139.7 $ 36.60\pm 1.49$ 1969.5 $ 9.11\pm 0.44$ 2572.0 $ 5.93\pm 0.39$
3072.5 $ 57.20\pm 2.09$ 3176.7 $ 35.90\pm 1.47$ 1986.4 $ 8.44\pm 0.42$ 2579.8 $ 6.92\pm 0.43$
3139.7 $ 61.00\pm 2.19$ 3202.7 $ 39.20\pm 1.54$ 2003.4 $ 8.72\pm 0.43$ 2579.7 $ 6.50\pm 0.41$
3176.7 $ 59.10\pm 2.14$ 3277.7 $ 41.60\pm 1.60$ 2067.2 $ 9.42\pm 0.45$ 2628.0 $ 9.00\pm 0.51$
-------- ------------------ -------- ------------------ -------- ----------------- -------- ------------------
: ATCA flux density measurements for SNR 1987A[]{data-label="tb:fluxes"}
-------- ------------------ -------- ------------------ -------- ----------------- -------- ------------------
Day Flux Density Day Flux Density Day Flux Density Day Flux Density
Number (mJy) Number (mJy) Number (mJy) Number (mJy)
3202.7 $ 64.20\pm 2.27$ 3325.7 $ 41.40\pm 1.59$ 2092.4 $ 9.94\pm 0.47$ 2627.7 $ 7.80\pm 0.46$
3277.7 $ 66.20\pm 2.32$ 3414.8 $ 44.10\pm 1.66$ 2142.5 $ 8.56\pm 0.42$ 2647.7 $ 6.10\pm 0.39$
3325.7 $ 67.10\pm 2.34$ 3455.7 $ 44.80\pm 1.68$ 2195.7 $ 9.10\pm 0.44$ 2754.8 $ 6.26\pm 0.40$
3414.8 $ 71.80\pm 2.47$ 3515.4 $ 45.80\pm 1.70$ 2261.8 $10.86\pm 0.50$ 2753.7 $ 5.40\pm 0.37$
3455.7 $ 71.80\pm 2.47$ 3579.2 $ 48.50\pm 1.77$ 2299.9 $11.50\pm 0.52$ 2774.5 $ 9.22\pm 0.52$
3515.4 $ 73.30\pm 2.51$ 3633.1 $ 48.00\pm 1.75$ 2299.7 $11.60\pm 0.53$ 2826.7 $ 9.80\pm 0.55$
3579.2 $ 77.50\pm 2.62$ 3679.0 $ 49.40\pm 1.79$ 2374.7 $10.82\pm 0.50$ 2919.2 $ 9.44\pm 0.53$
3633.1 $ 79.10\pm 2.66$ 3714.0 $ 50.10\pm 1.81$ 2403.5 $12.42\pm 0.56$ 2919.3 $ 11.30\pm 0.62$
3679.0 $ 79.60\pm 2.67$ 3744.7 $ 48.50\pm 1.77$ 2462.7 $12.80\pm 0.57$ 2975.8 $ 9.11\pm 0.52$
3714.0 $ 79.30\pm 2.66$ 3771.6 $ 49.90\pm 1.80$ 2511.0 $11.98\pm 0.54$ 2975.7 $ 8.30\pm 0.48$
3744.7 $ 84.40\pm 2.80$ 3833.6 $ 50.10\pm 1.81$ 2572.0 $13.99\pm 0.61$ 3000.7 $ 6.60\pm 0.41$
3771.6 $ 81.90\pm 2.73$ 3900.3 $ 52.60\pm 1.87$ 2579.8 $13.62\pm 0.60$ 3001.7 $ 9.23\pm 0.52$
3833.6 $ 84.20\pm 2.80$ 3945.4 $ 52.20\pm 1.86$ 2579.7 $13.10\pm 0.58$ 3072.5 $ 10.40\pm 0.58$
3900.3 $ 86.00\pm 2.85$ 3987.1 $ 52.40\pm 1.86$ 2628.0 $13.53\pm 0.60$ 3073.6 $ 11.40\pm 0.62$
3945.4 $ 88.10\pm 2.90$ 4015.1 $ 55.30\pm 1.94$ 2627.7 $14.10\pm 0.62$ 3110.6 $ 10.20\pm 0.57$
3987.1 $ 93.80\pm 3.06$ 4058.5 $ 58.30\pm 2.01$ 2647.7 $12.90\pm 0.57$ 3176.7 $ 9.80\pm 0.55$
4015.1 $ 89.30\pm 2.94$ 4100.8 $ 58.30\pm 2.01$ 2754.8 $14.17\pm 0.62$ 3202.7 $ 12.60\pm 0.68$
4058.5 $ 87.70\pm 2.89$ 4169.5 $ 61.40\pm 2.10$ 2753.7 $13.61\pm 0.60$ 3277.7 $ 12.10\pm 0.65$
4100.8 $ 96.70\pm 3.14$ 4222.5 $ 59.20\pm 2.04$ 2774.5 $14.63\pm 0.64$ 3325.7 $ 11.70\pm 0.64$
4169.5 $ 99.60\pm 3.22$ 4291.4 $ 64.70\pm 2.18$ 2773.7 $14.10\pm 0.62$ 3414.8 $ 11.50\pm 0.63$
4222.5 $ 97.50\pm 3.16$ 4373.1 $ 62.30\pm 2.12$ 2826.7 $14.50\pm 0.63$ 3436.7 $ 15.70\pm 0.82$
4291.4 $104.50\pm 3.36$ 4424.0 $ 67.10\pm 2.25$ 2919.2 $16.47\pm 0.70$ 3455.7 $ 14.20\pm 0.75$
4373.1 $108.30\pm 3.46$ 4460.9 $ 68.90\pm 2.30$ 2919.3 $16.10\pm 0.69$ 3485.5 $ 16.10\pm 0.84$
4424.0 $107.40\pm 3.44$ 4539.8 $ 75.50\pm 2.48$ 2975.8 $15.28\pm 0.66$ 3512.4 $ 16.50\pm 0.86$
4460.9 $113.20\pm 3.60$ 4571.7 $ 69.50\pm 2.31$ 2975.7 $16.50\pm 0.71$ 3515.4 $ 14.60\pm 0.77$
4539.8 $112.60\pm 3.58$ 4684.9 $ 75.20\pm 2.47$ 3000.7 $15.10\pm 0.65$ 3579.2 $ 15.50\pm 0.81$
4571.7 $116.10\pm 3.68$ 4728.9 $ 74.20\pm 2.44$ 3001.7 $16.84\pm 0.72$ 3633.1 $ 14.90\pm 0.79$
4684.9 $121.20\pm 3.83$ 4768.1 $ 77.10\pm 2.52$ 3072.5 $17.80\pm 0.75$ 3679.0 $ 16.30\pm 0.85$
4728.9 $121.40\pm 3.83$ 4799.9 $ 77.40\pm 2.53$ 3176.7 $16.90\pm 0.72$ 3714.0 $ 15.90\pm 0.83$
4768.1 $131.30\pm 4.12$ 4837.9 $ 83.80\pm 2.71$ 3202.7 $20.20\pm 0.85$ 3744.7 $ 16.30\pm 0.85$
4799.9 $124.90\pm 3.93$ 4850.8 $ 77.80\pm 2.54$ 3277.7 $21.60\pm 0.90$ 3771.6 $ 17.90\pm 0.93$
4837.9 $126.40\pm 3.98$ 4870.7 $ 77.90\pm 2.54$ 3325.7 $20.70\pm 0.86$ 3833.6 $ 17.70\pm 0.92$
4850.8 $128.70\pm 4.04$ 4936.5 $ 81.00\pm 2.63$ 3414.8 $21.30\pm 0.89$ 3900.3 $ 17.10\pm 0.89$
4870.7 $127.00\pm 3.99$ 4997.3 $ 83.60\pm 2.70$ 3455.7 $23.10\pm 0.96$ 3945.4 $ 17.90\pm 0.93$
4936.5 $133.40\pm 4.18$ 5024.8 $ 83.00\pm 2.68$ 3515.4 $24.80\pm 1.02$ 3987.1 $ 18.90\pm 0.98$
4997.3 $136.90\pm 4.28$ 5050.1 $ 86.20\pm 2.77$ 3579.2 $25.40\pm 1.05$ 4015.1 $ 19.40\pm 1.00$
5024.8 $135.00\pm 4.22$ 5093.0 $ 86.60\pm 2.78$ 3633.1 $25.10\pm 1.03$ 4100.8 $ 19.60\pm 1.01$
5050.1 $135.90\pm 4.25$ 3679.0 $24.50\pm 1.01$ 4222.5 $ 19.10\pm 0.99$
5093.0 $143.40\pm 4.47$ 3714.0 $24.50\pm 1.01$ 4291.4 $ 21.00\pm 1.08$
3744.7 $27.00\pm 1.11$ 4373.1 $ 21.20\pm 1.09$
3771.6 $27.10\pm 1.11$ 4424.0 $ 24.90\pm 1.27$
3833.6 $26.50\pm 1.09$ 4460.9 $ 25.50\pm 1.30$
-------- ------------------ -------- ------------------ -------- ----------------- -------- ------------------
: – [*continued*]{}
-------- -------------- -------- -------------- -------- ----------------- -------- ------------------
Day Flux Density Day Flux Density Day Flux Density Day Flux Density
Number (mJy) Number (mJy) Number (mJy) Number (mJy)
3900.3 $29.20\pm 1.19$ 4539.8 $ 25.00\pm 1.27$
3945.4 $29.80\pm 1.22$ 4571.7 $ 24.40\pm 1.25$
3987.1 $31.90\pm 1.30$ 4684.9 $ 26.70\pm 1.36$
4015.1 $30.20\pm 1.23$ 4728.9 $ 28.10\pm 1.43$
4100.8 $32.00\pm 1.30$ 4768.1 $ 25.70\pm 1.31$
4222.5 $32.00\pm 1.30$ 4799.9 $ 24.80\pm 1.26$
4291.4 $35.30\pm 1.43$ 4837.9 $ 31.50\pm 1.59$
4424.0 $34.10\pm 1.39$ 4850.8 $ 27.90\pm 1.42$
4460.9 $38.60\pm 1.56$ 4870.7 $ 29.90\pm 1.52$
4539.8 $37.60\pm 1.52$ 4936.5 $ 30.20\pm 1.53$
4571.7 $39.60\pm 1.60$ 4997.3 $ 27.50\pm 1.40$
4684.9 $40.10\pm 1.62$ 5024.8 $ 31.90\pm 1.61$
4728.9 $43.70\pm 1.77$ 5050.1 $ 31.90\pm 1.61$
4768.1 $42.70\pm 1.73$ 5093.0 $ 36.00\pm 1.82$
4799.9 $39.70\pm 1.61$
4837.9 $43.60\pm 1.76$
4850.8 $45.50\pm 1.84$
4870.7 $47.20\pm 1.90$
4936.5 $48.10\pm 1.94$
4997.3 $43.50\pm 1.76$
5024.8 $47.70\pm 1.92$
5050.1 $46.40\pm 1.87$
5093.0 $49.90\pm 2.01$
-------- -------------- -------- -------------- -------- ----------------- -------- ------------------
: – [*continued*]{}
[^1]: Hubble Fellow. Present address: Harvard-Smithsonian Centre for Astrophysics, Cambridge MA 02138, USA
[^2]: Day number = MJD $-46849.3$
[^3]: See http://www.atnf.csiro.au/computing/software/miriad/
|
{
"pile_set_name": "ArXiv"
}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.